diff --git a/gnu/llvm/llvm/lib/Target/X86/X86RetClean.cpp b/gnu/llvm/llvm/lib/Target/X86/X86RetClean.cpp
index 623bfede5..671f4a491 100644
--- a/gnu/llvm/llvm/lib/Target/X86/X86RetClean.cpp
+++ b/gnu/llvm/llvm/lib/Target/X86/X86RetClean.cpp
@@ -96,6 +96,10 @@ bool RetCleanPass::runOnMachineFunction(MachineFunction &MF) {
 
   bool modified = false;
 
+  // If a setjmp-like function is called by this function, we should not clean
+  if (MF.exposesReturnsTwice())
+    return false;
+
   for (auto &MBB : MF) {
     std::vector<MachineInstr*> fixups;
     bool foundcall = false;
diff --git a/lib/libcrypto/man/X509_STORE_CTX_set_verify.3 b/lib/libcrypto/man/X509_STORE_CTX_set_verify.3
index 2c0bd692a..8c27deea5 100644
--- a/lib/libcrypto/man/X509_STORE_CTX_set_verify.3
+++ b/lib/libcrypto/man/X509_STORE_CTX_set_verify.3
@@ -1,4 +1,4 @@
-.\" $OpenBSD: X509_STORE_CTX_set_verify.3,v 1.7 2023/08/10 16:15:42 schwarze Exp $
+.\" $OpenBSD: X509_STORE_CTX_set_verify.3,v 1.8 2024/06/07 05:51:39 tb Exp $
 .\"
 .\" Copyright (c) 2021, 2022 Ingo Schwarze <schwarze@openbsd.org>
 .\" Copyright (c) 2023 Job Snijders <job@openbsd.org>
@@ -15,7 +15,7 @@
 .\" ACTION OF CONTRACT, NEGLIGENCE OR OTHER TORTIOUS ACTION, ARISING OUT OF
 .\" OR IN CONNECTION WITH THE USE OR PERFORMANCE OF THIS SOFTWARE.
 .\"
-.Dd $Mdocdate: August 10 2023 $
+.Dd $Mdocdate: June 7 2024 $
 .Dt X509_STORE_CTX_SET_VERIFY 3
 .Os
 .Sh NAME
@@ -142,6 +142,28 @@ function provided by the user should check whether a given certificate
 was issued using the CA certificate
 .Fa issuer ,
 and must return 0 on failure and 1 on success.
+The default implementation ignores the
+.Fa ctx
+argument and returns success if and only if
+.Xr X509_check_issued 3
+returns
+.Dv X509_V_OK .
+It is important to pay close attention to the order of the
+.Fa issuer
+and
+.Fa subject
+arguments.
+In
+.Xr X509_check_issued 3
+the
+.Fa issuer
+precedes the
+.Fa subject
+while in
+.Fn check_issued
+the
+.Fa subject
+comes first.
 .Sh RETURN VALUES
 .Fn X509_STORE_CTX_verify_fn
 is supposed to return 1 to indicate that the chain is valid
@@ -221,3 +243,14 @@ and
 .Fn X509_STORE_CTX_get_check_issued
 first appeared in OpenSSL 1.1.0 and have been available since
 .Ox 7.3 .
+.Sh BUGS
+The reversal of order of
+.Fa subject
+and
+.Fa issuer
+between
+.Fn check_issued
+and
+.Xr X509_check_issued 3
+is very confusing.
+It has led to bugs and will cause many more.
diff --git a/lib/libcrypto/man/X509_cmp.3 b/lib/libcrypto/man/X509_cmp.3
index f90bc0e6d..b1cdec177 100644
--- a/lib/libcrypto/man/X509_cmp.3
+++ b/lib/libcrypto/man/X509_cmp.3
@@ -1,4 +1,4 @@
-.\" $OpenBSD: X509_cmp.3,v 1.3 2021/07/02 10:50:39 schwarze Exp $
+.\" $OpenBSD: X509_cmp.3,v 1.4 2024/06/07 14:00:09 job Exp $
 .\" full merge up to: OpenSSL ea5d4b89 Jun 6 11:42:02 2019 +0800
 .\"
 .\" This file is a derived work.
@@ -65,7 +65,7 @@
 .\" ARISING IN ANY WAY OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED
 .\" OF THE POSSIBILITY OF SUCH DAMAGE.
 .\"
-.Dd $Mdocdate: July 2 2021 $
+.Dd $Mdocdate: June 7 2024 $
 .Dt X509_CMP 3
 .Os
 .Sh NAME
@@ -123,8 +123,10 @@
 .Fn X509_cmp
 compares two X.509 certificates using
 .Xr memcmp 3
-on the SHA1 hashes of their canonical (DER) representations as generated with
+on the hashes of their canonical (DER) representations as generated with
 .Xr X509_digest 3 .
+The digest function is implementation-specific: LibreSSL uses SHA-512, other
+implementations use SHA-1.
 .Pp
 .Fn X509_NAME_cmp
 compares two X.501
@@ -172,8 +174,10 @@ structures using
 .Fn X509_CRL_match
 compares two certificate revocation lists using
 .Xr memcmp 3
-on the SHA1 hashes of their canonical (DER) representations as generated with
+on the hashes of their canonical (DER) representations as generated with
 .Xr X509_CRL_digest 3 .
+The digest function is implementation-specific: LibreSSL uses SHA-512, other
+implementations use SHA-1.
 .Sh RETURN VALUES
 All these functions return 0 to indicate a match or a non-zero value
 to indicate a mismatch.
diff --git a/lib/libcrypto/x509/x509_verify.c b/lib/libcrypto/x509/x509_verify.c
index c7b2219fa..d3534879b 100644
--- a/lib/libcrypto/x509/x509_verify.c
+++ b/lib/libcrypto/x509/x509_verify.c
@@ -1,4 +1,4 @@
-/* $OpenBSD: x509_verify.c,v 1.69 2024/04/08 23:46:21 beck Exp $ */
+/* $OpenBSD: x509_verify.c,v 1.70 2024/06/07 06:21:40 tb Exp $ */
 /*
  * Copyright (c) 2020-2021 Bob Beck <beck@openbsd.org>
  *
@@ -528,7 +528,7 @@ x509_verify_potential_parent(struct x509_verify_ctx *ctx, X509 *parent,
 		return (ctx->xsc->check_issued(ctx->xsc, child, parent));
 
 	/* XXX key usage */
-	return X509_check_issued(child, parent) != X509_V_OK;
+	return X509_check_issued(parent, child) == X509_V_OK;
 }
 
 static int
diff --git a/regress/sys/net/pf_divert/LICENSE b/regress/sys/net/pf_divert/LICENSE
index 52b251631..7c7565696 100644
--- a/regress/sys/net/pf_divert/LICENSE
+++ b/regress/sys/net/pf_divert/LICENSE
@@ -1,4 +1,4 @@
-# Copyright (c) 2010-2017 Alexander Bluhm <bluhm@openbsd.org>
+# Copyright (c) 2010-2024 Alexander Bluhm <bluhm@openbsd.org>
 #
 # Permission to use, copy, modify, and distribute this software for any
 # purpose with or without fee is hereby granted, provided that the above
diff --git a/regress/sys/net/pf_divert/remote.pl b/regress/sys/net/pf_divert/remote.pl
index a32330f53..7d0794315 100644
--- a/regress/sys/net/pf_divert/remote.pl
+++ b/regress/sys/net/pf_divert/remote.pl
@@ -1,7 +1,7 @@
 #!/usr/bin/perl
-#	$OpenBSD: remote.pl,v 1.9 2017/12/18 17:01:27 bluhm Exp $
+#	$OpenBSD: remote.pl,v 1.10 2024/06/08 22:50:40 bluhm Exp $
 
-# Copyright (c) 2010-2015 Alexander Bluhm <bluhm@openbsd.org>
+# Copyright (c) 2010-2024 Alexander Bluhm <bluhm@openbsd.org>
 #
 # Permission to use, copy, modify, and distribute this software for any
 # purpose with or without fee is hereby granted, provided that the above
@@ -272,4 +272,5 @@ sub copy_prefix {
 		chomp;
 		print $dst "$prefix: $_\n" if length;
 	}
+	$src->clearerr();
 }
diff --git a/regress/usr.bin/ssh/Makefile b/regress/usr.bin/ssh/Makefile
index 09870cd1a..27737c38a 100644
--- a/regress/usr.bin/ssh/Makefile
+++ b/regress/usr.bin/ssh/Makefile
@@ -1,4 +1,4 @@
-#	$OpenBSD: Makefile,v 1.133 2024/01/11 04:50:28 djm Exp $
+#	$OpenBSD: Makefile,v 1.134 2024/06/06 19:49:25 djm Exp $
 
 OPENSSL?=	yes
 
@@ -102,7 +102,8 @@ LTESTS= 	connect \
 		connection-timeout \
 		match-subsystem \
 		agent-pkcs11-restrict \
-		agent-pkcs11-cert
+		agent-pkcs11-cert \
+		penalty
 
 INTEROP_TESTS=	putty-transfer putty-ciphers putty-kex conch-ciphers
 INTEROP_TESTS+=	dropbear-ciphers dropbear-kex
diff --git a/regress/usr.bin/ssh/penalty.sh b/regress/usr.bin/ssh/penalty.sh
new file mode 100644
index 000000000..5ac7ef67e
--- /dev/null
+++ b/regress/usr.bin/ssh/penalty.sh
@@ -0,0 +1,51 @@
+#	$OpenBSD
+#	Placed in the Public Domain.
+
+tid="penalties"
+
+grep -vi PerSourcePenalties $OBJ/sshd_config > $OBJ/sshd_config.bak
+cp $OBJ/authorized_keys_${USER} $OBJ/authorized_keys_${USER}.bak
+
+conf() {
+	test -z "$PIDFILE" || stop_sshd
+	(cat $OBJ/sshd_config.bak ;
+	 echo "PerSourcePenalties $@") > $OBJ/sshd_config
+	cp $OBJ/authorized_keys_${USER}.bak $OBJ/authorized_keys_${USER}
+	start_sshd
+}
+
+conf "noauth:10s authfail:6s grace-exceeded:10s min:8s max:20s"
+
+verbose "test connect"
+${SSH} -F $OBJ/ssh_config somehost true || fatal "basic connect failed"
+
+verbose "penalty for authentication failure"
+
+# Fail authentication once
+cat /dev/null > $OBJ/authorized_keys_${USER}
+${SSH} -F $OBJ/ssh_config somehost true && fatal "noauth connect succeeded"
+cp $OBJ/authorized_keys_${USER}.bak $OBJ/authorized_keys_${USER}
+
+# Should be below penalty threshold
+${SSH} -F $OBJ/ssh_config somehost true || fatal "authfail not expired"
+
+# Fail authentication again; penalty should activate
+cat /dev/null > $OBJ/authorized_keys_${USER}
+${SSH} -F $OBJ/ssh_config somehost true && fatal "noauth connect succeeded"
+cp $OBJ/authorized_keys_${USER}.bak $OBJ/authorized_keys_${USER}
+
+# These should be refused by the active penalty
+${SSH} -F $OBJ/ssh_config somehost true && fail "authfail not rejected"
+sleep 5
+${SSH} -F $OBJ/ssh_config somehost true && fail "repeat authfail not rejected"
+
+# Penalty should have expired, this should succeed.
+sleep 8
+${SSH} -F $OBJ/ssh_config somehost true || fail "authfail not expired"
+
+verbose "penalty for no authentication"
+${SSHKEYSCAN} -p $PORT 127.0.0.1 >/dev/null 2>&1 || fatal "keyscan failed"
+
+# Repeat attempt should be penalised
+${SSHKEYSCAN} -p $PORT 127.0.0.1 >/dev/null 2>&1 && fail "keyscan not rejected"
+
diff --git a/regress/usr.bin/ssh/test-exec.sh b/regress/usr.bin/ssh/test-exec.sh
index fc9b2402b..fcf916414 100644
--- a/regress/usr.bin/ssh/test-exec.sh
+++ b/regress/usr.bin/ssh/test-exec.sh
@@ -1,4 +1,4 @@
-#	$OpenBSD: test-exec.sh,v 1.111 2024/05/17 01:45:22 djm Exp $
+#	$OpenBSD: test-exec.sh,v 1.114 2024/06/06 19:48:40 djm Exp $
 #	Placed in the Public Domain.
 
 #SUDO=sudo
@@ -265,33 +265,31 @@ export SSH_PKCS11_HELPER SSH_SK_HELPER
 
 stop_sshd ()
 {
-	if [ -f $PIDFILE ]; then
-		pid=`$SUDO cat $PIDFILE`
-		if [ "X$pid" = "X" ]; then
-			echo no sshd running
-		else
-			if [ $pid -lt 2 ]; then
-				echo bad pid for sshd: $pid
-			else
-				$SUDO kill $pid
-				trace "wait for sshd to exit"
-				i=0;
-				while [ -f $PIDFILE -a $i -lt 5 ]; do
-					i=`expr $i + 1`
-					sleep $i
-				done
-				if test -f $PIDFILE; then
-					if $SUDO kill -0 $pid; then
-						echo "sshd didn't exit " \
-						    "port $PORT pid $pid"
-					else
-						echo "sshd died without cleanup"
-					fi
-					exit 1
-				fi
-			fi
-		fi
+	[ -f $PIDFILE ] || return
+	pid=`$SUDO cat $PIDFILE`
+	if [ "X$pid" = "X" ]; then
+		echo "no sshd running" 1>&2
+		return
+	elif [ $pid -lt 2 ]; then
+		echo "bad pid for sshd: $pid" 1>&2
+		return
 	fi
+	$SUDO kill $pid
+	trace "wait for sshd to exit"
+	i=0;
+	while [ -f $PIDFILE -a $i -lt 5 ]; do
+		i=`expr $i + 1`
+		sleep $i
+	done
+	if test -f $PIDFILE; then
+		if $SUDO kill -0 $pid; then
+			echo "sshd didn't exit port $PORT pid $pid" 1>&2
+		else
+			echo "sshd died without cleanup" 1>&2
+		fi
+		exit 1
+	fi
+	PIDFILE=""
 }
 
 # helper
@@ -427,6 +425,7 @@ cat << EOF > $OBJ/sshd_config
 	AcceptEnv		_XXX_TEST
 	Subsystem	sftp	$SFTPSERVER
 	SshdSessionPath		$SSHD_SESSION
+	PerSourcePenalties	no
 EOF
 
 # This may be necessary if /usr/src and/or /usr/obj are group-writable,
@@ -689,6 +688,7 @@ start_sshd ()
 		i=`expr $i + 1`
 		sleep $i
 	done
+	ln -f -s ${logfile} $TEST_SSHD_LOGFILE
 
 	test -f $PIDFILE || fatal "no sshd running on port $PORT"
 }
diff --git a/sys/arch/amd64/amd64/cpu.c b/sys/arch/amd64/amd64/cpu.c
index 670beb278..6672773a9 100644
--- a/sys/arch/amd64/amd64/cpu.c
+++ b/sys/arch/amd64/amd64/cpu.c
@@ -1,4 +1,4 @@
-/*	$OpenBSD: cpu.c,v 1.189 2024/05/29 12:21:33 kettenis Exp $	*/
+/*	$OpenBSD: cpu.c,v 1.190 2024/06/07 16:53:35 kettenis Exp $	*/
 /* $NetBSD: cpu.c,v 1.1 2003/04/26 18:39:26 fvdl Exp $ */
 
 /*-
@@ -1469,7 +1469,10 @@ int cpu_suspended;
 void
 cpu_suspend_cycle(void)
 {
-	cpu_idle_cycle_fcn();
+	if (cpu_suspend_cycle_fcn)
+		cpu_suspend_cycle_fcn();
+	else
+		cpu_idle_cycle_fcn();
 }
 
 int
diff --git a/sys/arch/amd64/amd64/intr.c b/sys/arch/amd64/amd64/intr.c
index b85d5a006..efdfbc96c 100644
--- a/sys/arch/amd64/amd64/intr.c
+++ b/sys/arch/amd64/amd64/intr.c
@@ -1,4 +1,4 @@
-/*	$OpenBSD: intr.c,v 1.58 2024/05/29 12:21:33 kettenis Exp $	*/
+/*	$OpenBSD: intr.c,v 1.59 2024/06/07 06:26:23 jsg Exp $	*/
 /*	$NetBSD: intr.c,v 1.3 2003/03/03 22:16:20 fvdl Exp $	*/
 
 /*
@@ -559,8 +559,6 @@ intr_handler(struct intrframe *frame, struct intrhand *ih)
 	return rc;
 }
 
-#define CONCAT(x,y)	__CONCAT(x,y)
-
 /*
  * Fake interrupt handler structures for the benefit of symmetry with
  * other interrupt sources, and the benefit of intr_calculatemasks()
diff --git a/sys/arch/amd64/amd64/ipifuncs.c b/sys/arch/amd64/amd64/ipifuncs.c
index 5c330d7a1..482fc9eeb 100644
--- a/sys/arch/amd64/amd64/ipifuncs.c
+++ b/sys/arch/amd64/amd64/ipifuncs.c
@@ -1,4 +1,4 @@
-/*	$OpenBSD: ipifuncs.c,v 1.38 2023/10/30 12:50:59 mvs Exp $	*/
+/*	$OpenBSD: ipifuncs.c,v 1.39 2024/06/07 16:53:35 kettenis Exp $	*/
 /*	$NetBSD: ipifuncs.c,v 1.1 2003/04/26 18:39:28 fvdl Exp $ */
 
 /*-
@@ -128,7 +128,10 @@ x86_64_ipi_halt(struct cpu_info *ci)
 	wbinvd();
 
 	for(;;) {
-		__asm volatile("hlt");
+		if (cpu_suspend_cycle_fcn)
+			cpu_suspend_cycle_fcn();
+		else
+			__asm volatile("hlt");
 	}
 }
 
diff --git a/sys/arch/amd64/amd64/machdep.c b/sys/arch/amd64/amd64/machdep.c
index 343cc0f7e..35cea0b58 100644
--- a/sys/arch/amd64/amd64/machdep.c
+++ b/sys/arch/amd64/amd64/machdep.c
@@ -1,4 +1,4 @@
-/*	$OpenBSD: machdep.c,v 1.293 2024/04/29 00:29:48 jsg Exp $	*/
+/*	$OpenBSD: machdep.c,v 1.294 2024/06/07 16:53:35 kettenis Exp $	*/
 /*	$NetBSD: machdep.c,v 1.3 2003/05/07 22:58:18 fvdl Exp $	*/
 
 /*-
@@ -162,6 +162,7 @@ char machine[] = MACHINE;
  */
 void cpu_idle_cycle_hlt(void);
 void (*cpu_idle_cycle_fcn)(void) = &cpu_idle_cycle_hlt;
+void (*cpu_suspend_cycle_fcn)(void);
 
 /* the following is used externally for concurrent handlers */
 int setperf_prio = 0;
diff --git a/sys/arch/amd64/amd64/mpbios.c b/sys/arch/amd64/amd64/mpbios.c
index 4ecb31846..07df036ed 100644
--- a/sys/arch/amd64/amd64/mpbios.c
+++ b/sys/arch/amd64/amd64/mpbios.c
@@ -1,4 +1,4 @@
-/*	$OpenBSD: mpbios.c,v 1.31 2022/02/21 11:03:39 mpi Exp $	*/
+/*	$OpenBSD: mpbios.c,v 1.32 2024/06/07 23:19:18 jsg Exp $	*/
 /*	$NetBSD: mpbios.c,v 1.7 2003/05/15 16:32:50 fvdl Exp $	*/
 
 /*-
@@ -856,14 +856,6 @@ mp_print_eisa_intr(int intr)
 }
 #endif
 
-
-
-#define TAB_UNIT	4
-#define TAB_ROUND(a)	_TAB_ROUND(a, TAB_UNIT)
-
-#define _TAB_ROUND(a,u)	(((a) + (u - 1)) & ~(u - 1))
-#define EXTEND_TAB(a,u)	(!(_TAB_ROUND(a, u) == _TAB_ROUND((a + 1), u)))
-
 void
 mpbios_bus(const u_int8_t *ent, struct device *self)
 {
diff --git a/sys/arch/amd64/conf/Makefile.amd64 b/sys/arch/amd64/conf/Makefile.amd64
index 263e7fe8d..aefa751ee 100644
--- a/sys/arch/amd64/conf/Makefile.amd64
+++ b/sys/arch/amd64/conf/Makefile.amd64
@@ -1,4 +1,4 @@
-#	$OpenBSD: Makefile.amd64,v 1.136 2024/06/05 20:19:26 deraadt Exp $
+#	$OpenBSD: Makefile.amd64,v 1.137 2024/06/07 05:17:34 deraadt Exp $
 
 # For instructions on building kernels consult the config(8) and options(4)
 # manual pages.
@@ -73,7 +73,7 @@ CMACHFLAGS+=	-mno-retpoline -fcf-protection=none
 .endif
 .else
 CMACHFLAGS+=	-mretpoline-external-thunk -fcf-protection=branch
-#CMACHFLAGS+=	-fret-clean
+CMACHFLAGS+=	-fret-clean
 .endif
 .if ${COMPILER_VERSION:Mclang}
 NO_INTEGR_AS=	-no-integrated-as
diff --git a/sys/arch/amd64/include/cpu.h b/sys/arch/amd64/include/cpu.h
index fab34e5bd..90f5dd9e9 100644
--- a/sys/arch/amd64/include/cpu.h
+++ b/sys/arch/amd64/include/cpu.h
@@ -1,4 +1,4 @@
-/*	$OpenBSD: cpu.h,v 1.171 2024/05/29 12:21:33 kettenis Exp $	*/
+/*	$OpenBSD: cpu.h,v 1.172 2024/06/07 16:53:35 kettenis Exp $	*/
 /*	$NetBSD: cpu.h,v 1.1 2003/04/26 18:39:39 fvdl Exp $	*/
 
 /*-
@@ -420,6 +420,7 @@ void	cpu_proc_fork(struct proc *, struct proc *);
 int	amd64_pa_used(paddr_t);
 #define	cpu_idle_enter()	do { /* nothing */ } while (0)
 extern void (*cpu_idle_cycle_fcn)(void);
+extern void (*cpu_suspend_cycle_fcn)(void);
 #define	cpu_idle_cycle()	(*cpu_idle_cycle_fcn)()
 #define	cpu_idle_leave()	do { /* nothing */ } while (0)
 extern void (*initclock_func)(void);
diff --git a/sys/arch/amd64/include/i82489var.h b/sys/arch/amd64/include/i82489var.h
index 374c61474..61840dd38 100644
--- a/sys/arch/amd64/include/i82489var.h
+++ b/sys/arch/amd64/include/i82489var.h
@@ -1,4 +1,4 @@
-/*	$OpenBSD: i82489var.h,v 1.18 2018/10/04 05:00:40 guenther Exp $	*/
+/*	$OpenBSD: i82489var.h,v 1.19 2024/06/09 03:12:59 jsg Exp $	*/
 /*	$NetBSD: i82489var.h,v 1.1 2003/02/26 21:26:10 fvdl Exp $	*/
 
 /*-
@@ -87,21 +87,6 @@ extern void Xresume_lapic_ltimer(void);
 extern void Xrecurse_lapic_ltimer(void);
 #define LAPIC_TIMER_VECTOR		0xc0
 
-/*
- * 'pin numbers' for local APIC
- */
-#define LAPIC_PIN_TIMER		0
-#define LAPIC_PIN_PCINT		2
-#define LAPIC_PIN_LVINT0	3
-#define LAPIC_PIN_LVINT1	4
-#define LAPIC_PIN_LVERR		5
-
-extern void Xintr_lapic0(void);
-extern void Xintr_lapic2(void);
-extern void Xintr_lapic3(void);
-extern void Xintr_lapic4(void);
-extern void Xintr_lapic5(void);
-
 /*
  * Vector used for Xen HVM Event Channel Interrupts.
  */
diff --git a/sys/arch/i386/i386/db_trace.c b/sys/arch/i386/i386/db_trace.c
index 5f73dabee..4e83d064b 100644
--- a/sys/arch/i386/i386/db_trace.c
+++ b/sys/arch/i386/i386/db_trace.c
@@ -1,4 +1,4 @@
-/*	$OpenBSD: db_trace.c,v 1.44 2023/04/26 16:53:58 claudio Exp $	*/
+/*	$OpenBSD: db_trace.c,v 1.45 2024/06/07 10:14:29 jsg Exp $	*/
 /*	$NetBSD: db_trace.c,v 1.18 1996/05/03 19:42:01 christos Exp $	*/
 
 /*
@@ -68,12 +68,6 @@ struct db_variable *db_eregs = db_regs + nitems(db_regs);
  */
 #define	INKERNEL(va)	(((vaddr_t)(va)) >= VM_MIN_KERNEL_ADDRESS)
 
-#define	NONE		0
-#define	TRAP		1
-#define	SYSCALL		2
-#define	INTERRUPT	3
-#define	AST		4
-
 int db_i386_numargs(struct callframe *);
 
 /*
diff --git a/sys/arch/i386/i386/machdep.c b/sys/arch/i386/i386/machdep.c
index 26e524330..abb46f435 100644
--- a/sys/arch/i386/i386/machdep.c
+++ b/sys/arch/i386/i386/machdep.c
@@ -1,4 +1,4 @@
-/*	$OpenBSD: machdep.c,v 1.671 2024/05/26 13:37:32 kettenis Exp $	*/
+/*	$OpenBSD: machdep.c,v 1.672 2024/06/07 16:53:35 kettenis Exp $	*/
 /*	$NetBSD: machdep.c,v 1.214 1996/11/10 03:16:17 thorpej Exp $	*/
 
 /*-
@@ -167,6 +167,7 @@ char machine[] = MACHINE;
 void (*cpu_idle_leave_fcn)(void) = NULL;
 void (*cpu_idle_cycle_fcn)(void) = NULL;
 void (*cpu_idle_enter_fcn)(void) = NULL;
+void (*cpu_suspend_cycle_fcn)(void);
 
 
 struct uvm_constraint_range  isa_constraint = { 0x0, 0x00ffffffUL };
diff --git a/sys/arch/i386/i386/mpbios.c b/sys/arch/i386/i386/mpbios.c
index fb1b70357..1775fb8ad 100644
--- a/sys/arch/i386/i386/mpbios.c
+++ b/sys/arch/i386/i386/mpbios.c
@@ -1,4 +1,4 @@
-/*	$OpenBSD: mpbios.c,v 1.46 2024/05/13 01:15:50 jsg Exp $	*/
+/*	$OpenBSD: mpbios.c,v 1.47 2024/06/07 23:19:18 jsg Exp $	*/
 /*	$NetBSD: mpbios.c,v 1.2 2002/10/01 12:56:57 fvdl Exp $	*/
 
 /*-
@@ -925,14 +925,6 @@ mp_print_eisa_intr(int intr)
 	printf(" EISA irq %d", intr);
 }
 
-
-
-#define TAB_UNIT	4
-#define TAB_ROUND(a)	_TAB_ROUND(a, TAB_UNIT)
-
-#define _TAB_ROUND(a,u)	(((a) + (u - 1)) & ~(u - 1))
-#define EXTEND_TAB(a,u)	(!(_TAB_ROUND(a, u) == _TAB_ROUND((a + 1), u)))
-
 void
 mpbios_bus(const u_int8_t *ent, struct device *self)
 {
diff --git a/sys/arch/i386/include/cpu.h b/sys/arch/i386/include/cpu.h
index 72e7d3381..4eb710735 100644
--- a/sys/arch/i386/include/cpu.h
+++ b/sys/arch/i386/include/cpu.h
@@ -1,4 +1,4 @@
-/*	$OpenBSD: cpu.h,v 1.189 2024/05/21 23:16:06 jsg Exp $	*/
+/*	$OpenBSD: cpu.h,v 1.190 2024/06/07 16:53:35 kettenis Exp $	*/
 /*	$NetBSD: cpu.h,v 1.35 1996/05/05 19:29:26 christos Exp $	*/
 
 /*-
@@ -375,6 +375,7 @@ extern const struct cpu_cpuid_nameclass i386_cpuid_cpus[];
 extern void (*cpu_idle_enter_fcn)(void);
 extern void (*cpu_idle_cycle_fcn)(void);
 extern void (*cpu_idle_leave_fcn)(void);
+extern void (*cpu_suspend_cycle_fcn)(void);
 
 extern int cpuspeed;
 
diff --git a/sys/arch/i386/isa/clock.c b/sys/arch/i386/isa/clock.c
index 3363ba15f..b9aad35cb 100644
--- a/sys/arch/i386/isa/clock.c
+++ b/sys/arch/i386/isa/clock.c
@@ -1,4 +1,4 @@
-/*	$OpenBSD: clock.c,v 1.69 2024/05/13 01:15:50 jsg Exp $	*/
+/*	$OpenBSD: clock.c,v 1.70 2024/06/08 00:24:00 jsg Exp $	*/
 /*	$NetBSD: clock.c,v 1.39 1996/05/12 23:11:54 mycroft Exp $	*/
 
 /*-
@@ -138,9 +138,6 @@ static struct timecounter i8254_timecounter = {
 struct mutex timer_mutex = MUTEX_INITIALIZER(IPL_HIGH);
 u_long rtclock_tval;
 
-#define	SECMIN	((unsigned)60)			/* seconds per minute */
-#define	SECHOUR	((unsigned)(60*SECMIN))		/* seconds per hour */
-
 u_int
 mc146818_read(void *sc, u_int reg)
 {
diff --git a/sys/dev/acpi/acpicpu.c b/sys/dev/acpi/acpicpu.c
index 0a5ee1fea..6a126fadd 100644
--- a/sys/dev/acpi/acpicpu.c
+++ b/sys/dev/acpi/acpicpu.c
@@ -1,4 +1,4 @@
-/* $OpenBSD: acpicpu.c,v 1.92 2022/04/06 18:59:27 naddy Exp $ */
+/* $OpenBSD: acpicpu.c,v 1.93 2024/06/07 16:53:35 kettenis Exp $ */
 /*
  * Copyright (c) 2005 Marco Peereboom <marco@openbsd.org>
  * Copyright (c) 2015 Philip Guenther <guenther@openbsd.org>
@@ -172,6 +172,7 @@ void	acpicpu_add_cstate(struct acpicpu_softc *_sc, int _state, int _method,
 	    int _flags, int _latency, int _power, uint64_t _address);
 void	acpicpu_set_pdc(struct acpicpu_softc *);
 void	acpicpu_idle(void);
+void	acpicpu_suspend(void);
 
 #if 0
 void    acpicpu_set_throttle(struct acpicpu_softc *, int);
@@ -747,6 +748,7 @@ acpicpu_attach(struct device *parent, struct device *self, void *aux)
 		extern uint32_t acpi_force_bm;
 
 		cpu_idle_cycle_fcn = &acpicpu_idle;
+		cpu_suspend_cycle_fcn = &acpicpu_suspend;
 
 		/*
 		 * C3 (and maybe C2?) needs BM_RLD to be set to
@@ -1277,3 +1279,66 @@ acpicpu_idle(void)
 	sc->sc_prev_sleep = (sc->sc_prev_sleep + (sc->sc_prev_sleep >> 1)
 	    + itime) >> 1;
 }
+
+void
+acpicpu_suspend(void)
+{
+	extern int cpu_suspended;
+	struct cpu_info *ci = curcpu();
+	struct acpicpu_softc *sc = (struct acpicpu_softc *)ci->ci_acpicpudev;
+	struct acpi_cstate *best, *cx;
+
+	if (sc == NULL) {
+		__asm volatile("sti");
+		panic("null acpicpu");
+	}
+
+	/*
+	 * Find the lowest usable state.
+	 */
+	best = cx = SLIST_FIRST(&sc->sc_cstates);
+	while ((cx->flags & CST_FLAG_SKIP)) {
+		if ((cx = SLIST_NEXT(cx, link)) == NULL)
+			break;
+		best = cx;
+	}
+
+	switch (best->method) {
+	default:
+	case CST_METH_HALT:
+		__asm volatile("sti; hlt");
+		break;
+
+	case CST_METH_IO_HALT:
+		inb((u_short)best->address);
+		__asm volatile("sti; hlt");
+		break;
+
+	case CST_METH_MWAIT:
+		{
+		unsigned int hints;
+
+		hints = (unsigned)best->address;
+		/* intel errata AAI65: cflush before monitor */
+		if (ci->ci_cflushsz != 0 &&
+		    strcmp(cpu_vendor, "GenuineIntel") == 0) {
+			membar_sync();
+			clflush((unsigned long)&cpu_suspended);
+			membar_sync();
+		}
+
+		monitor(&cpu_suspended, 0, 0);
+		if (cpu_suspended || !CPU_IS_PRIMARY(ci))
+			mwait(0, hints);
+
+		break;
+		}
+
+	case CST_METH_GAS_IO:
+		inb((u_short)best->address);
+		/* something harmless to give system time to change state */
+		acpi_read_pmreg(acpi_softc, ACPIREG_PM1_STS, 0);
+		break;
+
+	}
+}
diff --git a/sys/dev/audio.c b/sys/dev/audio.c
index 0d8bf45ab..7986ba477 100644
--- a/sys/dev/audio.c
+++ b/sys/dev/audio.c
@@ -1,4 +1,4 @@
-/*	$OpenBSD: audio.c,v 1.206 2023/02/10 14:34:16 visa Exp $	*/
+/*	$OpenBSD: audio.c,v 1.207 2024/06/07 08:48:10 jsg Exp $	*/
 /*
  * Copyright (c) 2015 Alexandre Ratchov <alex@caoua.org>
  *
@@ -48,8 +48,6 @@
 #define DPRINTFN(n, ...) do {} while(0)
 #endif
 
-#define IPL_SOFTAUDIO		IPL_SOFTNET
-
 #define DEVNAME(sc)		((sc)->dev.dv_xname)
 #define AUDIO_UNIT(n)		(minor(n) & 0x0f)
 #define AUDIO_DEV(n)		(minor(n) & 0xf0)
diff --git a/sys/dev/ic/ufshci.c b/sys/dev/ic/ufshci.c
index 8f3c5d120..1e56a3c06 100644
--- a/sys/dev/ic/ufshci.c
+++ b/sys/dev/ic/ufshci.c
@@ -1,4 +1,4 @@
-/*	$OpenBSD: ufshci.c,v 1.34 2024/06/05 04:58:05 mglocker Exp $ */
+/*	$OpenBSD: ufshci.c,v 1.35 2024/06/09 03:21:54 jsg Exp $ */
 
 /*
  * Copyright (c) 2022 Marcus Glocker <mglocker@openbsd.org>
@@ -89,8 +89,6 @@ int			 ufshci_utr_cmd_sync(struct ufshci_softc *,
 			     struct ufshci_ccb *, struct scsi_xfer *,
 			     uint32_t, uint16_t);
 int			 ufshci_xfer_complete(struct ufshci_softc *);
-int			 ufshci_powerdown(struct ufshci_softc *);
-int			 ufshci_resume(struct ufshci_softc *);
 
 /* SCSI */
 int			 ufshci_ccb_alloc(struct ufshci_softc *, int);
diff --git a/sys/dev/pci/fms.c b/sys/dev/pci/fms.c
index e8fa53bfd..eebd73e93 100644
--- a/sys/dev/pci/fms.c
+++ b/sys/dev/pci/fms.c
@@ -1,4 +1,4 @@
-/*	$OpenBSD: fms.c,v 1.38 2024/05/24 06:02:53 jsg Exp $ */
+/*	$OpenBSD: fms.c,v 1.39 2024/06/09 05:18:12 jsg Exp $ */
 /*	$NetBSD: fms.c,v 1.5.4.1 2000/06/30 16:27:50 simonb Exp $	*/
 
 /*-
@@ -118,9 +118,6 @@ int	fms_read_codec(void *, u_int8_t, u_int16_t *);
 int	fms_write_codec(void *, u_int8_t, u_int16_t);
 void	fms_reset_codec(void *);
 
-int	fms_allocmem(struct fms_softc *, size_t, size_t,
-			  struct fms_dma *);
-
 int
 fms_match(struct device *parent, void *match, void *aux)
 {
diff --git a/sys/dev/pci/if_em_hw.h b/sys/dev/pci/if_em_hw.h
index 61f2dde68..623e3feab 100644
--- a/sys/dev/pci/if_em_hw.h
+++ b/sys/dev/pci/if_em_hw.h
@@ -31,7 +31,7 @@
 
 *******************************************************************************/
 
-/* $OpenBSD: if_em_hw.h,v 1.94 2024/05/13 01:15:51 jsg Exp $ */
+/* $OpenBSD: if_em_hw.h,v 1.95 2024/06/09 05:18:12 jsg Exp $ */
 /* $FreeBSD: if_em_hw.h,v 1.15 2005/05/26 23:32:02 tackerman Exp $ */
 
 /* if_em_hw.h
@@ -408,8 +408,6 @@ struct em_host_mng_dhcp_cookie{
 };
 
 int32_t em_read_part_num(struct em_hw *hw, uint32_t *part_num);
-int32_t em_mng_write_dhcp_info(struct em_hw *hw, uint8_t *buffer,
-                                  uint16_t length);
 boolean_t em_check_mng_mode(struct em_hw *hw);
 boolean_t em_enable_tx_pkt_filtering(struct em_hw *hw);
 int32_t em_read_eeprom(struct em_hw *hw, uint16_t reg, uint16_t words, uint16_t *data);
diff --git a/sys/dev/pci/if_vmx.c b/sys/dev/pci/if_vmx.c
index 0c5883a1b..e70d9b50f 100644
--- a/sys/dev/pci/if_vmx.c
+++ b/sys/dev/pci/if_vmx.c
@@ -1,4 +1,4 @@
-/*	$OpenBSD: if_vmx.c,v 1.86 2024/05/21 19:49:06 jan Exp $	*/
+/*	$OpenBSD: if_vmx.c,v 1.87 2024/06/07 08:44:25 jan Exp $	*/
 
 /*
  * Copyright (c) 2013 Tsubai Masanari
@@ -114,6 +114,8 @@ struct vmxnet3_comp_ring {
 	};
 	u_int next;
 	u_int32_t gen;
+	struct mbuf *sendmp;
+	struct mbuf *lastmp;
 };
 
 struct vmxnet3_txqueue {
@@ -160,6 +162,7 @@ struct vmxnet3_softc {
 	struct vmxnet3_queue *sc_q;
 	struct intrmap *sc_intrmap;
 
+	u_int sc_vrrs;
 	struct vmxnet3_driver_shared *sc_ds;
 	u_int8_t *sc_mcast;
 	struct vmxnet3_upt1_rss_conf *sc_rss;
@@ -170,7 +173,7 @@ struct vmxnet3_softc {
 #endif
 };
 
-#define JUMBO_LEN (1024 * 9)
+#define JUMBO_LEN ((16 * 1024) - 1)
 #define DMAADDR(map) ((map)->dm_segs[0].ds_addr)
 
 #define READ_BAR0(sc, reg) bus_space_read_4((sc)->sc_iot0, (sc)->sc_ioh0, reg)
@@ -273,15 +276,21 @@ vmxnet3_attach(struct device *parent, struct device *self, void *aux)
 		return;
 	}
 
+	/* Vmxnet3 Revision Report and Selection */
 	ver = READ_BAR1(sc, VMXNET3_BAR1_VRRS);
-	if ((ver & 0x1) == 0) {
+	if (ISSET(ver, 0x2)) {
+		sc->sc_vrrs = 2;
+	} else if (ISSET(ver, 0x1)) {
+		sc->sc_vrrs = 1;
+	} else {
 		printf(": unsupported hardware version 0x%x\n", ver);
 		return;
 	}
-	WRITE_BAR1(sc, VMXNET3_BAR1_VRRS, 1);
+	WRITE_BAR1(sc, VMXNET3_BAR1_VRRS, sc->sc_vrrs);
 
+	/* UPT Version Report and Selection */
 	ver = READ_BAR1(sc, VMXNET3_BAR1_UVRS);
-	if ((ver & 0x1) == 0) {
+	if (!ISSET(ver, 0x1)) {
 		printf(": incompatible UPT version 0x%x\n", ver);
 		return;
 	}
@@ -410,6 +419,11 @@ vmxnet3_attach(struct device *parent, struct device *self, void *aux)
 
 	ifp->if_capabilities |= IFCAP_TSOv4 | IFCAP_TSOv6;
 
+	if (sc->sc_vrrs == 2) {
+		ifp->if_xflags |= IFXF_LRO;
+		ifp->if_capabilities |= IFCAP_LRO;
+	}
+
 #if NVLAN > 0
 	if (sc->sc_ds->upt_features & UPT1_F_VLAN)
 		ifp->if_capabilities |= IFCAP_VLAN_HWTAGGING;
@@ -704,6 +718,10 @@ vmxnet3_rxfill(struct vmxnet3_rxring *ring)
 	uint32_t rgen;
 	uint32_t type = htole32(VMXNET3_BTYPE_HEAD << VMXNET3_RX_BTYPE_S);
 
+	/* Second ring just contains packet bodies. */
+	if (ring->rid == 1)
+		type = htole32(VMXNET3_BTYPE_BODY << VMXNET3_RX_BTYPE_S);
+
 	MUTEX_ASSERT_LOCKED(&ring->mtx);
 
 	slots = if_rxr_get(&ring->rxr, NRXDESC);
@@ -781,17 +799,17 @@ vmxnet3_rxinit(struct vmxnet3_softc *sc, struct vmxnet3_rxqueue *rq)
 		    VMX_DMA_LEN(&ring->dmamem));
 		bus_dmamap_sync(sc->sc_dmat, VMX_DMA_MAP(&ring->dmamem),
 		    0, VMX_DMA_LEN(&ring->dmamem), BUS_DMASYNC_PREWRITE);
-	}
 
-	/* XXX only fill ring 0 */
-	ring = &rq->cmd_ring[0];
-	mtx_enter(&ring->mtx);
-	vmxnet3_rxfill(ring);
-	mtx_leave(&ring->mtx);
+		mtx_enter(&ring->mtx);
+		vmxnet3_rxfill(ring);
+		mtx_leave(&ring->mtx);
+	}
 
 	comp_ring = &rq->comp_ring;
 	comp_ring->next = 0;
 	comp_ring->gen = VMX_RXC_GEN;
+	comp_ring->sendmp = NULL;
+	comp_ring->lastmp = NULL;
 
 	memset(VMX_DMA_KVA(&comp_ring->dmamem), 0,
 	    VMX_DMA_LEN(&comp_ring->dmamem));
@@ -1074,9 +1092,9 @@ vmxnet3_rxintr(struct vmxnet3_softc *sc, struct vmxnet3_rxqueue *rq)
 	struct mbuf_list ml = MBUF_LIST_INITIALIZER();
 	struct mbuf *m;
 	bus_dmamap_t map;
-	unsigned int idx, len;
+	unsigned int idx;
 	unsigned int next, rgen;
-	unsigned int done = 0;
+	unsigned int rid, done[2] = {0, 0};
 
 	next = comp_ring->next;
 	rgen = comp_ring->gen;
@@ -1096,11 +1114,14 @@ vmxnet3_rxintr(struct vmxnet3_softc *sc, struct vmxnet3_rxqueue *rq)
 
 		idx = letoh32((rxcd->rxc_word0 >> VMXNET3_RXC_IDX_S) &
 		    VMXNET3_RXC_IDX_M);
+
 		if (letoh32((rxcd->rxc_word0 >> VMXNET3_RXC_QID_S) &
 		    VMXNET3_RXC_QID_M) < sc->sc_nqueues)
-			ring = &rq->cmd_ring[0];
+			rid = 0;
 		else
-			ring = &rq->cmd_ring[1];
+			rid = 1;
+
+		ring = &rq->cmd_ring[rid];
 
 		m = ring->m[idx];
 		KASSERT(m != NULL);
@@ -1111,31 +1132,62 @@ vmxnet3_rxintr(struct vmxnet3_softc *sc, struct vmxnet3_rxqueue *rq)
 		    BUS_DMASYNC_POSTREAD);
 		bus_dmamap_unload(sc->sc_dmat, map);
 
-		done++;
+		done[rid]++;
+
+		/*
+		 * A receive descriptor of type 4 which is flaged as start of
+		 * packet, contains the number of TCP segment of an LRO packet.
+		 */
+		if (letoh32((rxcd->rxc_word3 & VMXNET3_RXC_TYPE_M) >>
+		    VMXNET3_RXC_TYPE_S) == 4 &&
+		    ISSET(rxcd->rxc_word0, VMXNET3_RXC_SOP)) {
+			m->m_pkthdr.ph_mss = letoh32(rxcd->rxc_word1 &
+			    VMXNET3_RXC_SEG_CNT_M);
+		}
+
+		m->m_len = letoh32((rxcd->rxc_word2 >> VMXNET3_RXC_LEN_S) &
+		    VMXNET3_RXC_LEN_M);
+
+		if (comp_ring->sendmp == NULL) {
+			comp_ring->sendmp = comp_ring->lastmp = m;
+			comp_ring->sendmp->m_pkthdr.len = 0;
+		} else {
+			CLR(m->m_flags, M_PKTHDR);
+			comp_ring->lastmp->m_next = m;
+			comp_ring->lastmp = m;
+		}
+		comp_ring->sendmp->m_pkthdr.len += m->m_len;
+
+		if (!ISSET(rxcd->rxc_word0, VMXNET3_RXC_EOP))
+			continue;
+
+		/*
+		 * End of Packet
+		 */
 
 		if (letoh32(rxcd->rxc_word2 & VMXNET3_RXC_ERROR)) {
 			ifp->if_ierrors++;
-			m_freem(m);
+			m_freem(comp_ring->sendmp);
+			comp_ring->sendmp = comp_ring->lastmp = NULL;
 			continue;
 		}
 
-		len = letoh32((rxcd->rxc_word2 >> VMXNET3_RXC_LEN_S) &
-		    VMXNET3_RXC_LEN_M);
-		if (len < VMXNET3_MIN_MTU) {
-			m_freem(m);
+		if (comp_ring->sendmp->m_pkthdr.len < VMXNET3_MIN_MTU) {
+			m_freem(comp_ring->sendmp);
+			comp_ring->sendmp = comp_ring->lastmp = NULL;
 			continue;
 		}
-		m->m_pkthdr.len = m->m_len = len;
-
-		vmxnet3_rx_offload(rxcd, m);
 
 		if (((letoh32(rxcd->rxc_word0) >> VMXNET3_RXC_RSSTYPE_S) &
 		    VMXNET3_RXC_RSSTYPE_M) != VMXNET3_RXC_RSSTYPE_NONE) {
-			m->m_pkthdr.ph_flowid = letoh32(rxcd->rxc_word1);
-			SET(m->m_pkthdr.csum_flags, M_FLOWID);
+			comp_ring->sendmp->m_pkthdr.ph_flowid =
+			    letoh32(rxcd->rxc_word1);
+			SET(comp_ring->sendmp->m_pkthdr.csum_flags, M_FLOWID);
 		}
 
-		ml_enqueue(&ml, m);
+		vmxnet3_rx_offload(rxcd, comp_ring->sendmp);
+		ml_enqueue(&ml, comp_ring->sendmp);
+		comp_ring->sendmp = comp_ring->lastmp = NULL;
 	}
 
 	bus_dmamap_sync(sc->sc_dmat, VMX_DMA_MAP(&comp_ring->dmamem),
@@ -1144,19 +1196,20 @@ vmxnet3_rxintr(struct vmxnet3_softc *sc, struct vmxnet3_rxqueue *rq)
 	comp_ring->next = next;
 	comp_ring->gen = rgen;
 
-	if (done == 0)
-		return;
+	for (int i = 0; i < 2; i++) {
+		if (done[i] == 0)
+			continue;
 
-	ring = &rq->cmd_ring[0];
+		ring = &rq->cmd_ring[i];
 
-	if (ifiq_input(rq->ifiq, &ml))
-		if_rxr_livelocked(&ring->rxr);
+		if (ifiq_input(rq->ifiq, &ml))
+			if_rxr_livelocked(&ring->rxr);
 
-	/* XXX Should we (try to) allocate buffers for ring 2 too? */
-	mtx_enter(&ring->mtx);
-	if_rxr_put(&ring->rxr, done);
-	vmxnet3_rxfill(ring);
-	mtx_leave(&ring->mtx);
+		mtx_enter(&ring->mtx);
+		if_rxr_put(&ring->rxr, done[i]);
+		vmxnet3_rxfill(ring);
+		mtx_leave(&ring->mtx);
+	}
 }
 
 void
@@ -1211,6 +1264,8 @@ vmxnet3_iff(struct vmxnet3_softc *sc)
 void
 vmxnet3_rx_offload(struct vmxnet3_rxcompdesc *rxcd, struct mbuf *m)
 {
+	uint32_t pkts;
+
 	/*
 	 * VLAN Offload
 	 */
@@ -1243,6 +1298,45 @@ vmxnet3_rx_offload(struct vmxnet3_rxcompdesc *rxcd, struct mbuf *m)
 		else if (ISSET(rxcd->rxc_word3, VMXNET3_RXC_UDP))
 			SET(m->m_pkthdr.csum_flags, M_UDP_CSUM_IN_OK);
 	}
+
+	/*
+	 * TCP Large Receive Offload
+	 */
+
+	pkts = m->m_pkthdr.ph_mss;
+	m->m_pkthdr.ph_mss = 0;
+
+	if (pkts > 1) {
+		struct ether_extracted ext;
+		uint32_t paylen;
+
+		ether_extract_headers(m, &ext);
+
+		paylen = ext.iplen;
+		if (ext.ip4 || ext.ip6)
+			paylen -= ext.iphlen;
+
+		if (ext.tcp) {
+			paylen -= ext.tcphlen;
+			tcpstat_inc(tcps_inhwlro);
+			tcpstat_add(tcps_inpktlro, pkts);
+		} else {
+			tcpstat_inc(tcps_inbadlro);
+		}
+
+		/*
+		 * If we gonna forward this packet, we have to mark it as TSO,
+		 * set a correct mss, and recalculate the TCP checksum.
+		 */
+		if (ext.tcp && paylen >= pkts) {
+			SET(m->m_pkthdr.csum_flags, M_TCP_TSO);
+			m->m_pkthdr.ph_mss = paylen / pkts;
+		}
+		if (ext.tcp &&
+		    ISSET(m->m_pkthdr.csum_flags, M_TCP_CSUM_IN_OK)) {
+			SET(m->m_pkthdr.csum_flags, M_TCP_CSUM_OUT);
+		}
+	}
 }
 
 void
@@ -1309,6 +1403,13 @@ vmxnet3_init(struct vmxnet3_softc *sc)
 		return EIO;
 	}
 
+	/* TCP Large Receive Offload */
+	if (ISSET(ifp->if_xflags, IFXF_LRO))
+		SET(sc->sc_ds->upt_features, UPT1_F_LRO);
+	else
+		CLR(sc->sc_ds->upt_features, UPT1_F_LRO);
+	WRITE_CMD(sc, VMXNET3_CMD_SET_FEATURE);
+
 	/* Program promiscuous mode and multicast filters. */
 	vmxnet3_iff(sc);
 
diff --git a/sys/dev/pci/if_vmxreg.h b/sys/dev/pci/if_vmxreg.h
index c5338b720..55a5114c2 100644
--- a/sys/dev/pci/if_vmxreg.h
+++ b/sys/dev/pci/if_vmxreg.h
@@ -1,4 +1,4 @@
-/*	$OpenBSD: if_vmxreg.h,v 1.9 2020/07/07 01:36:49 dlg Exp $	*/
+/*	$OpenBSD: if_vmxreg.h,v 1.10 2024/06/07 08:44:25 jan Exp $	*/
 
 /*
  * Copyright (c) 2013 Tsubai Masanari
@@ -76,6 +76,7 @@ enum UPT1_RxStats {
 #define VMXNET3_CMD_RESET	0xcafe0002	/* reset device */
 #define VMXNET3_CMD_SET_RXMODE	0xcafe0003	/* set interface flags */
 #define VMXNET3_CMD_SET_FILTER	0xcafe0004	/* set address filter */
+#define VMXNET3_CMD_SET_FEATURE	0xcafe0009	/* set features */
 #define VMXNET3_CMD_GET_STATUS	0xf00d0000	/* get queue errors */
 #define VMXNET3_CMD_GET_STATS	0xf00d0001
 #define VMXNET3_CMD_GET_LINK	0xf00d0002	/* get link status */
@@ -189,6 +190,7 @@ struct vmxnet3_rxcompdesc {
 	u_int32_t		rxc_word1;
 #define VMXNET3_RXC_RSSHASH_M	0xffffffff	/* RSS hash value */
 #define VMXNET3_RXC_RSSHASH_S	0
+#define VMXNET3_RXC_SEG_CNT_M	0x000000ff	/* No. of seg. in LRO pkt */
 
 	u_int32_t		rxc_word2;
 #define VMXNET3_RXC_LEN_M	0x00003fff
@@ -210,6 +212,7 @@ struct vmxnet3_rxcompdesc {
 #define VMXNET3_RXC_FRAGMENT	0x00400000	/* IP fragment */
 #define VMXNET3_RXC_FCS		0x00800000	/* frame CRC correct */
 #define VMXNET3_RXC_TYPE_M	0x7f000000
+#define VMXNET3_RXC_TYPE_S	24
 #define VMXNET3_RXC_GEN_M	0x00000001U
 #define VMXNET3_RXC_GEN_S	31
 } __packed;
diff --git a/sys/dev/pci/igc_api.h b/sys/dev/pci/igc_api.h
index dee8557c7..ea56c5265 100644
--- a/sys/dev/pci/igc_api.h
+++ b/sys/dev/pci/igc_api.h
@@ -1,4 +1,4 @@
-/*	$OpenBSD: igc_api.h,v 1.2 2022/05/11 06:14:15 kevlo Exp $	*/
+/*	$OpenBSD: igc_api.h,v 1.3 2024/06/09 05:18:12 jsg Exp $	*/
 /*-
  * Copyright 2021 Intel Corp
  * Copyright 2021 Rubicon Communications, LLC (Netgate)
@@ -20,35 +20,17 @@ int		igc_setup_init_funcs(struct igc_hw *, bool);
 int		igc_init_mac_params(struct igc_hw *);
 int		igc_init_nvm_params(struct igc_hw *);
 int		igc_init_phy_params(struct igc_hw *);
-int		igc_get_bus_info(struct igc_hw *);
-void		igc_clear_vfta(struct igc_hw *);
-void		igc_write_vfta(struct igc_hw *, uint32_t, uint32_t);
-int		igc_force_mac_fc(struct igc_hw *);
 int		igc_check_for_link(struct igc_hw *);
 int		igc_reset_hw(struct igc_hw *);
 int		igc_init_hw(struct igc_hw *);
-int		igc_setup_link(struct igc_hw *);
 int		igc_get_speed_and_duplex(struct igc_hw *, uint16_t *,
 		    uint16_t *);
-int		igc_disable_pcie_master(struct igc_hw *);
-void		igc_config_collision_dist(struct igc_hw *);
 int		igc_rar_set(struct igc_hw *, uint8_t *, uint32_t);
-uint32_t	igc_hash_mc_addr(struct igc_hw *, uint8_t *);
 void		igc_update_mc_addr_list(struct igc_hw *, uint8_t *, uint32_t);
 int		igc_check_reset_block(struct igc_hw *);
-int		igc_get_cable_length(struct igc_hw *);
-int		igc_validate_mdi_setting(struct igc_hw *);
 int		igc_get_phy_info(struct igc_hw *);
 int		igc_phy_hw_reset(struct igc_hw *);
-void		igc_power_up_phy(struct igc_hw *);
-void		igc_power_down_phy(struct igc_hw *);
 int		igc_read_mac_addr(struct igc_hw *);
-int		igc_read_pba_string(struct igc_hw *, uint8_t *, uint32_t);
-void		igc_reload_nvm(struct igc_hw *);
 int		igc_validate_nvm_checksum(struct igc_hw *);
-int		igc_read_nvm(struct igc_hw *, uint16_t, uint16_t, uint16_t *);
-int		igc_write_nvm(struct igc_hw *, uint16_t, uint16_t, uint16_t *);
-int		igc_set_d3_lplu_state(struct igc_hw *, bool);
-int		igc_set_d0_lplu_state(struct igc_hw *, bool);
 
 #endif /* _IGC_API_H_ */
diff --git a/sys/dev/pci/igc_mac.h b/sys/dev/pci/igc_mac.h
index 0b800e9a1..524cd4caa 100644
--- a/sys/dev/pci/igc_mac.h
+++ b/sys/dev/pci/igc_mac.h
@@ -1,4 +1,4 @@
-/*	$OpenBSD: igc_mac.h,v 1.1 2021/10/31 14:52:57 patrick Exp $	*/
+/*	$OpenBSD: igc_mac.h,v 1.2 2024/06/09 05:18:12 jsg Exp $	*/
 /*-
  * Copyright 2021 Intel Corp
  * Copyright 2021 Rubicon Communications, LLC (Netgate)
@@ -12,34 +12,23 @@
 
 void	igc_init_mac_ops_generic(struct igc_hw *);
 int	igc_null_ops_generic(struct igc_hw *);
-int	igc_null_link_info(struct igc_hw *, uint16_t *, uint16_t *);
-bool	igc_null_mng_mode(struct igc_hw *);
-void	igc_null_update_mc(struct igc_hw *, uint8_t *h, uint32_t);
-void	igc_null_write_vfta(struct igc_hw *, uint32_t a, uint32_t);
-int	igc_check_for_copper_link_generic(struct igc_hw *);
 int	igc_config_fc_after_link_up_generic(struct igc_hw *);
 int	igc_disable_pcie_master_generic(struct igc_hw *);
 int	igc_force_mac_fc_generic(struct igc_hw *);
 int	igc_get_auto_rd_done_generic(struct igc_hw *);
-int	igc_get_bus_info_pcie_generic(struct igc_hw *);
-void	igc_set_lan_id_single_port(struct igc_hw *);
 int	igc_get_speed_and_duplex_copper_generic(struct igc_hw *, uint16_t *,
  	    uint16_t *);
 void	igc_update_mc_addr_list_generic(struct igc_hw *, uint8_t *, uint32_t);
 int	igc_rar_set_generic(struct igc_hw *, uint8_t *, uint32_t);
 int	igc_set_fc_watermarks_generic(struct igc_hw *);
 int	igc_setup_link_generic(struct igc_hw *);
-int	igc_validate_mdi_setting_crossover_generic(struct igc_hw *);
 
 int	igc_hash_mc_addr_generic(struct igc_hw *, uint8_t *);
 
 void	igc_clear_hw_cntrs_base_generic(struct igc_hw *);
-void	igc_clear_vfta_generic(struct igc_hw *);
 void	igc_init_rx_addrs_generic(struct igc_hw *, uint16_t);
-void	igc_pcix_mmrbc_workaround_generic(struct igc_hw *);
 void	igc_put_hw_semaphore_generic(struct igc_hw *);
 int	igc_check_alt_mac_addr_generic(struct igc_hw *);
-void	igc_set_pcie_no_snoop_generic(struct igc_hw *, uint32_t);
 void	igc_write_vfta_generic(struct igc_hw *, uint32_t, uint32_t);
 void	igc_config_collision_dist_generic(struct igc_hw *);
 
diff --git a/sys/dev/pci/igc_nvm.h b/sys/dev/pci/igc_nvm.h
index 39404cee0..cc1ec3d8e 100644
--- a/sys/dev/pci/igc_nvm.h
+++ b/sys/dev/pci/igc_nvm.h
@@ -1,4 +1,4 @@
-/*	$OpenBSD: igc_nvm.h,v 1.1 2021/10/31 14:52:57 patrick Exp $	*/
+/*	$OpenBSD: igc_nvm.h,v 1.2 2024/06/09 05:18:12 jsg Exp $	*/
 /*-
  * Copyright 2021 Intel Corp
  * Copyright 2021 Rubicon Communications, LLC (Netgate)
@@ -13,17 +13,11 @@
 void	igc_init_nvm_ops_generic(struct igc_hw *);
 int	igc_null_read_nvm(struct igc_hw *, uint16_t, uint16_t, uint16_t *);
 void	igc_null_nvm_generic(struct igc_hw *);
-int	igc_null_led_default(struct igc_hw *, uint16_t *);
 int	igc_null_write_nvm(struct igc_hw *, uint16_t, uint16_t, uint16_t *);
 int	igc_poll_eerd_eewr_done(struct igc_hw *, int);
 int	igc_read_mac_addr_generic(struct igc_hw *);
-int	igc_read_pba_string_generic(struct igc_hw *, uint8_t *, uint32_t);
 int	igc_read_nvm_eerd(struct igc_hw *, uint16_t, uint16_t, uint16_t *);
-int	igc_valid_led_default_generic(struct igc_hw *, uint16_t *);
 int	igc_validate_nvm_checksum_generic(struct igc_hw *);
-int	igc_write_nvm_spi(struct igc_hw *, uint16_t, uint16_t, uint16_t *);
-int	igc_update_nvm_checksum_generic(struct igc_hw *);
-void	igc_release_nvm_generic(struct igc_hw *);
 void	igc_reload_nvm_generic(struct igc_hw *);
 
 #endif	/* _IGC_NVM_H_ */
diff --git a/sys/dev/pci/igc_phy.h b/sys/dev/pci/igc_phy.h
index 9dafd927f..db9e41124 100644
--- a/sys/dev/pci/igc_phy.h
+++ b/sys/dev/pci/igc_phy.h
@@ -1,4 +1,4 @@
-/*	$OpenBSD: igc_phy.h,v 1.2 2022/05/11 06:14:15 kevlo Exp $	*/
+/*	$OpenBSD: igc_phy.h,v 1.3 2024/06/09 05:18:12 jsg Exp $	*/
 /*-
  * Copyright 2021 Intel Corp
  * Copyright 2021 Rubicon Communications, LLC (Netgate)
@@ -20,13 +20,8 @@ int	igc_check_downshift_generic(struct igc_hw *);
 int	igc_check_reset_block_generic(struct igc_hw *);
 int	igc_get_phy_id(struct igc_hw *);
 int	igc_phy_hw_reset_generic(struct igc_hw *);
-int	igc_phy_reset_dsp_generic(struct igc_hw *);
-int	igc_set_d3_lplu_state_generic(struct igc_hw *, bool);
 int	igc_setup_copper_link_generic(struct igc_hw *);
 int	igc_phy_has_link_generic(struct igc_hw *, uint32_t, uint32_t, bool *);
-int	igc_determine_phy_address(struct igc_hw *);
-int	igc_enable_phy_wakeup_reg_access_bm(struct igc_hw *, uint16_t *);
-int	igc_disable_phy_wakeup_reg_access_bm(struct igc_hw *, uint16_t *);
 void	igc_power_up_phy_copper(struct igc_hw *);
 void	igc_power_down_phy_copper(struct igc_hw *);
 int	igc_read_phy_reg_mdic(struct igc_hw *, uint32_t offset, uint16_t *);
diff --git a/sys/dev/pci/ixgbe.h b/sys/dev/pci/ixgbe.h
index 2851b2858..e0a03bd26 100644
--- a/sys/dev/pci/ixgbe.h
+++ b/sys/dev/pci/ixgbe.h
@@ -1,4 +1,4 @@
-/*	$OpenBSD: ixgbe.h,v 1.35 2024/04/12 19:27:43 jan Exp $	*/
+/*	$OpenBSD: ixgbe.h,v 1.36 2024/06/09 05:18:12 jsg Exp $	*/
 
 /******************************************************************************
 
@@ -268,10 +268,6 @@ int32_t ixgbe_init_hw(struct ixgbe_hw *hw);
 enum ixgbe_media_type ixgbe_get_media_type(struct ixgbe_hw *hw);
 int32_t ixgbe_identify_phy(struct ixgbe_hw *hw);
 void ixgbe_flap_tx_laser(struct ixgbe_hw *hw);
-int32_t ixgbe_setup_link(struct ixgbe_hw *hw, ixgbe_link_speed speed,
-			 bool autoneg_wait_to_complete);
-int32_t ixgbe_setup_mac_link(struct ixgbe_hw *hw, ixgbe_link_speed speed,
-			 bool autoneg_wait_to_complete);
 int32_t ixgbe_check_link(struct ixgbe_hw *hw, ixgbe_link_speed *speed,
 			 bool *link_up, bool link_up_wait_to_complete);
 
diff --git a/sys/net/pf.c b/sys/net/pf.c
index 8591b044e..ecc6bfef4 100644
--- a/sys/net/pf.c
+++ b/sys/net/pf.c
@@ -1,4 +1,4 @@
-/*	$OpenBSD: pf.c,v 1.1196 2024/05/14 08:26:13 jsg Exp $ */
+/*	$OpenBSD: pf.c,v 1.1197 2024/06/07 18:24:16 bluhm Exp $ */
 
 /*
  * Copyright (c) 2001 Daniel Hartmeier
@@ -7958,12 +7958,17 @@ done:
 		switch (pd.naf) {
 		case AF_INET:
 			if (pd.dir == PF_IN) {
-				if (ipforwarding == 0) {
+				int flags;
+
+				if (ip_forwarding == 0) {
 					ipstat_inc(ips_cantforward);
 					action = PF_DROP;
 					break;
 				}
-				ip_forward(pd.m, ifp, NULL, 1);
+				flags = IP_FORWARDING | IP_REDIRECT;
+				if (ip_directedbcast)
+					SET(flags, IP_ALLOWBROADCAST);
+				ip_forward(pd.m, ifp, NULL, flags);
 			} else
 				ip_output(pd.m, NULL, NULL, 0, NULL, NULL, 0);
 			break;
diff --git a/sys/net/pipex.c b/sys/net/pipex.c
index c2901741e..4aefbdfac 100644
--- a/sys/net/pipex.c
+++ b/sys/net/pipex.c
@@ -1,4 +1,4 @@
-/*	$OpenBSD: pipex.c,v 1.153 2024/01/23 17:57:21 mvs Exp $ */
+/*	$OpenBSD: pipex.c,v 1.154 2024/06/07 13:43:21 jsg Exp $ */
 
 /*-
  * Copyright (c) 2009 Internet Initiative Japan Inc.
@@ -104,9 +104,6 @@ int pipex_prune = 1;			/* [I] walk list every seconds */
 struct mbuf_queue pipexoutq = MBUF_QUEUE_INITIALIZER(
     IFQ_MAXLEN, IPL_SOFTNET);
 
-/* borrow an mbuf pkthdr field */
-#define ph_ppp_proto ether_vtag
-
 #ifdef PIPEX_DEBUG
 int pipex_debug = 0;		/* [A] systcl net.inet.ip.pipex_debug */
 #endif
diff --git a/sys/netinet/ip_esp.c b/sys/netinet/ip_esp.c
index ac4d6292b..a951063b8 100644
--- a/sys/netinet/ip_esp.c
+++ b/sys/netinet/ip_esp.c
@@ -1,4 +1,4 @@
-/*	$OpenBSD: ip_esp.c,v 1.195 2022/05/03 09:18:11 claudio Exp $ */
+/*	$OpenBSD: ip_esp.c,v 1.196 2024/06/07 13:15:25 jsg Exp $ */
 /*
  * The authors of this code are John Ioannidis (ji@tla.org),
  * Angelos D. Keromytis (kermit@csd.uch.gr) and
@@ -334,8 +334,6 @@ esp_zeroize(struct tdb *tdbp)
 	return error;
 }
 
-#define MAXBUFSIZ (AH_ALEN_MAX > ESP_MAX_IVS ? AH_ALEN_MAX : ESP_MAX_IVS)
-
 /*
  * ESP input processing, called (eventually) through the protocol switch.
  */
diff --git a/sys/netinet/ip_icmp.c b/sys/netinet/ip_icmp.c
index 30bec63a5..2d42796dc 100644
--- a/sys/netinet/ip_icmp.c
+++ b/sys/netinet/ip_icmp.c
@@ -1,4 +1,4 @@
-/*	$OpenBSD: ip_icmp.c,v 1.192 2023/09/16 09:33:27 mpi Exp $	*/
+/*	$OpenBSD: ip_icmp.c,v 1.193 2024/06/07 18:24:16 bluhm Exp $	*/
 /*	$NetBSD: ip_icmp.c,v 1.19 1996/02/13 23:42:22 christos Exp $	*/
 
 /*
@@ -589,7 +589,7 @@ reflect:
 		struct sockaddr_in ssrc;
 		struct rtentry *newrt = NULL;
 
-		if (icmp_rediraccept == 0 || ipforwarding == 1)
+		if (icmp_rediraccept == 0 || ip_forwarding == 1)
 			goto freeit;
 		if (code > 3)
 			goto badcode;
diff --git a/sys/netinet/ip_input.c b/sys/netinet/ip_input.c
index ad0455b50..38057fceb 100644
--- a/sys/netinet/ip_input.c
+++ b/sys/netinet/ip_input.c
@@ -1,4 +1,4 @@
-/*	$OpenBSD: ip_input.c,v 1.394 2024/05/08 13:01:30 bluhm Exp $	*/
+/*	$OpenBSD: ip_input.c,v 1.395 2024/06/07 18:24:16 bluhm Exp $	*/
 /*	$NetBSD: ip_input.c,v 1.30 1996/03/16 23:53:58 christos Exp $	*/
 
 /*
@@ -84,10 +84,10 @@
 #endif
 
 /* values controllable via sysctl */
-int	ipforwarding = 0;
+int	ip_forwarding = 0;
 int	ipmforwarding = 0;
 int	ipmultipath = 0;
-int	ipsendredirects = 1;
+int	ip_sendredirects = 1;
 int	ip_dosourceroute = 0;
 int	ip_defttl = IPDEFTTL;
 int	ip_mtudisc = 1;
@@ -108,8 +108,8 @@ const struct sysctl_bounded_args ipctl_vars[] = {
 #ifdef MROUTING
 	{ IPCTL_MRTPROTO, &ip_mrtproto, SYSCTL_INT_READONLY },
 #endif
-	{ IPCTL_FORWARDING, &ipforwarding, 0, 2 },
-	{ IPCTL_SENDREDIRECTS, &ipsendredirects, 0, 1 },
+	{ IPCTL_FORWARDING, &ip_forwarding, 0, 2 },
+	{ IPCTL_SENDREDIRECTS, &ip_sendredirects, 0, 1 },
 	{ IPCTL_DEFTTL, &ip_defttl, 0, 255 },
 	{ IPCTL_DIRECTEDBCAST, &ip_directedbcast, 0, 1 },
 	{ IPCTL_IPPORT_FIRSTAUTO, &ipport_firstauto, 0, 65535 },
@@ -137,8 +137,8 @@ static struct mbuf_queue	ipsendraw_mq;
 extern struct niqueue		arpinq;
 
 int	ip_ours(struct mbuf **, int *, int, int);
-int	ip_dooptions(struct mbuf *, struct ifnet *);
-int	in_ouraddr(struct mbuf *, struct ifnet *, struct route *);
+int	ip_dooptions(struct mbuf *, struct ifnet *, int);
+int	in_ouraddr(struct mbuf *, struct ifnet *, struct route *, int);
 
 int		ip_fragcheck(struct mbuf **, int *);
 struct mbuf *	ip_reass(struct ipqent *, struct ipq *);
@@ -431,7 +431,7 @@ ip_input_if(struct mbuf **mp, int *offp, int nxt, int af, struct ifnet *ifp)
 #if NPF > 0
 	struct in_addr odst;
 #endif
-	int pfrdr = 0;
+	int flags = 0;
 
 	KASSERT(*offp == 0);
 
@@ -461,9 +461,15 @@ ip_input_if(struct mbuf **mp, int *offp, int nxt, int af, struct ifnet *ifp)
 		goto bad;
 
 	ip = mtod(m, struct ip *);
-	pfrdr = odst.s_addr != ip->ip_dst.s_addr;
+	if (odst.s_addr != ip->ip_dst.s_addr)
+		SET(flags, IP_REDIRECT);
 #endif
 
+	if (ip_forwarding != 0)
+		SET(flags, IP_FORWARDING);
+	if (ip_directedbcast)
+		SET(flags, IP_ALLOWBROADCAST);
+
 	hlen = ip->ip_hl << 2;
 
 	/*
@@ -472,7 +478,7 @@ ip_input_if(struct mbuf **mp, int *offp, int nxt, int af, struct ifnet *ifp)
 	 * error was detected (causing an icmp message
 	 * to be sent and the original packet to be freed).
 	 */
-	if (hlen > sizeof (struct ip) && ip_dooptions(m, ifp)) {
+	if (hlen > sizeof (struct ip) && ip_dooptions(m, ifp, flags)) {
 		m = *mp = NULL;
 		goto bad;
 	}
@@ -483,7 +489,7 @@ ip_input_if(struct mbuf **mp, int *offp, int nxt, int af, struct ifnet *ifp)
 		goto out;
 	}
 
-	switch(in_ouraddr(m, ifp, &ro)) {
+	switch(in_ouraddr(m, ifp, &ro, flags)) {
 	case 2:
 		goto bad;
 	case 1:
@@ -565,7 +571,7 @@ ip_input_if(struct mbuf **mp, int *offp, int nxt, int af, struct ifnet *ifp)
 	/*
 	 * Not for us; forward if possible and desirable.
 	 */
-	if (ipforwarding == 0) {
+	if (!ISSET(flags, IP_FORWARDING)) {
 		ipstat_inc(ips_cantforward);
 		goto bad;
 	}
@@ -585,7 +591,7 @@ ip_input_if(struct mbuf **mp, int *offp, int nxt, int af, struct ifnet *ifp)
 	}
 #endif /* IPSEC */
 
-	ip_forward(m, ifp, &ro, pfrdr);
+	ip_forward(m, ifp, &ro, flags);
 	*mp = NULL;
 	rtfree(ro.ro_rt);
 	return IPPROTO_DONE;
@@ -807,7 +813,7 @@ ip_deliver(struct mbuf **mp, int *offp, int nxt, int af, int shared)
 #undef IPSTAT_INC
 
 int
-in_ouraddr(struct mbuf *m, struct ifnet *ifp, struct route *ro)
+in_ouraddr(struct mbuf *m, struct ifnet *ifp, struct route *ro, int flags)
 {
 	struct rtentry		*rt;
 	struct ip		*ip;
@@ -837,7 +843,8 @@ in_ouraddr(struct mbuf *m, struct ifnet *ifp, struct route *ro)
 		 * if it is received on the interface with that address.
 		 */
 		if (ISSET(rt->rt_flags, RTF_BROADCAST) &&
-		    (!ip_directedbcast || rt->rt_ifidx == ifp->if_index)) {
+		    (!ISSET(flags, IP_ALLOWBROADCAST) ||
+		    rt->rt_ifidx == ifp->if_index)) {
 			match = 1;
 
 			/* Make sure M_BCAST is set */
@@ -876,7 +883,8 @@ in_ouraddr(struct mbuf *m, struct ifnet *ifp, struct route *ro)
 				break;
 			}
 		}
-	} else if (ipforwarding == 0 && rt->rt_ifidx != ifp->if_index &&
+	} else if (!ISSET(flags, IP_FORWARDING) &&
+	    rt->rt_ifidx != ifp->if_index &&
 	    !((ifp->if_flags & IFF_LOOPBACK) || (ifp->if_type == IFT_ENC) ||
 	    (m->m_pkthdr.pf.flags & PF_TAG_TRANSLATE_LOCALHOST))) {
 		/* received on wrong interface. */
@@ -1150,7 +1158,7 @@ ip_flush(void)
  * 0 if the packet should be processed further.
  */
 int
-ip_dooptions(struct mbuf *m, struct ifnet *ifp)
+ip_dooptions(struct mbuf *m, struct ifnet *ifp, int flags)
 {
 	struct ip *ip = mtod(m, struct ip *);
 	unsigned int rtableid = m->m_pkthdr.ph_rtableid;
@@ -1371,8 +1379,8 @@ ip_dooptions(struct mbuf *m, struct ifnet *ifp)
 		}
 	}
 	KERNEL_UNLOCK();
-	if (forward && ipforwarding > 0) {
-		ip_forward(m, ifp, NULL, 1);
+	if (forward && ISSET(flags, IP_FORWARDING)) {
+		ip_forward(m, ifp, NULL, flags | IP_REDIRECT);
 		return (1);
 	}
 	return (0);
@@ -1514,7 +1522,7 @@ const u_char inetctlerrmap[PRC_NCMDS] = {
  * of codes and types.
  *
  * If not forwarding, just drop the packet.  This could be confusing
- * if ipforwarding was zero but some routing protocol was advancing
+ * if ip_forwarding was zero but some routing protocol was advancing
  * us as a gateway to somewhere.  However, we must let the routing
  * protocol deal with that.
  *
@@ -1522,7 +1530,7 @@ const u_char inetctlerrmap[PRC_NCMDS] = {
  * via a source route.
  */
 void
-ip_forward(struct mbuf *m, struct ifnet *ifp, struct route *ro, int srcrt)
+ip_forward(struct mbuf *m, struct ifnet *ifp, struct route *ro, int flags)
 {
 	struct mbuf mfake, *mcopy;
 	struct ip *ip = mtod(m, struct ip *);
@@ -1588,7 +1596,7 @@ ip_forward(struct mbuf *m, struct ifnet *ifp, struct route *ro, int srcrt)
 	if ((rt->rt_ifidx == ifp->if_index) &&
 	    (rt->rt_flags & (RTF_DYNAMIC|RTF_MODIFIED)) == 0 &&
 	    satosin(rt_key(rt))->sin_addr.s_addr != 0 &&
-	    ipsendredirects && !srcrt &&
+	    ip_sendredirects && !ISSET(flags, IP_REDIRECT) &&
 	    !arpproxy(satosin(rt_key(rt))->sin_addr, m->m_pkthdr.ph_rtableid)) {
 		if ((ip->ip_src.s_addr & ifatoia(rt->rt_ifa)->ia_netmask) ==
 		    ifatoia(rt->rt_ifa)->ia_net) {
@@ -1602,9 +1610,7 @@ ip_forward(struct mbuf *m, struct ifnet *ifp, struct route *ro, int srcrt)
 		}
 	}
 
-	error = ip_output(m, NULL, ro,
-	    (IP_FORWARDING | (ip_directedbcast ? IP_ALLOWBROADCAST : 0)),
-	    NULL, NULL, 0);
+	error = ip_output(m, NULL, ro, flags | IP_FORWARDING, NULL, NULL, 0);
 	rt = ro->ro_rt;
 	if (error)
 		ipstat_inc(ips_cantforward);
diff --git a/sys/netinet/ip_output.c b/sys/netinet/ip_output.c
index c0aeb7693..b21683a11 100644
--- a/sys/netinet/ip_output.c
+++ b/sys/netinet/ip_output.c
@@ -1,4 +1,4 @@
-/*	$OpenBSD: ip_output.c,v 1.399 2024/05/16 13:01:04 bluhm Exp $	*/
+/*	$OpenBSD: ip_output.c,v 1.400 2024/06/07 18:24:16 bluhm Exp $	*/
 /*	$NetBSD: ip_output.c,v 1.28 1996/02/13 23:43:07 christos Exp $	*/
 
 /*
@@ -428,7 +428,7 @@ sendit:
 #endif
 
 #ifdef IPSEC
-	if ((flags & IP_FORWARDING) && ipforwarding == 2 &&
+	if ((flags & IP_FORWARDING) && ip_forwarding == 2 &&
 	    (!ipsec_in_use ||
 	    m_tag_find(m, PACKET_TAG_IPSEC_IN_DONE, NULL) == NULL)) {
 		error = EHOSTUNREACH;
diff --git a/sys/netinet/ip_var.h b/sys/netinet/ip_var.h
index a963798db..d4e216344 100644
--- a/sys/netinet/ip_var.h
+++ b/sys/netinet/ip_var.h
@@ -1,4 +1,4 @@
-/*	$OpenBSD: ip_var.h,v 1.117 2024/04/17 20:48:51 bluhm Exp $	*/
+/*	$OpenBSD: ip_var.h,v 1.118 2024/06/07 18:24:16 bluhm Exp $	*/
 /*	$NetBSD: ip_var.h,v 1.16 1996/02/13 23:43:20 christos Exp $	*/
 
 /*
@@ -204,10 +204,11 @@ struct ipoffnxt {
 };
 
 /* flags passed to ip_output */
-#define	IP_FORWARDING		0x1		/* most of ip header exists */
-#define	IP_RAWOUTPUT		0x2		/* raw ip header exists */
-#define	IP_ALLOWBROADCAST	SO_BROADCAST	/* can send broadcast packets */
-#define	IP_MTUDISC		0x0800		/* pmtu discovery, set DF */
+#define IP_FORWARDING		0x0001	/* most of ip header exists */
+#define IP_RAWOUTPUT		0x0002	/* raw ip header exists */
+#define IP_REDIRECT		0x0004	/* redirected by pf or source route */
+#define IP_ALLOWBROADCAST	SO_BROADCAST	/* can send broadcast packets */
+#define IP_MTUDISC		0x0800	/* pmtu discovery, set DF */
 
 extern struct ipstat ipstat;
 extern int ip_defttl;			/* default IP ttl */
@@ -221,11 +222,12 @@ extern int ipport_firstauto;		/* min port for port allocation */
 extern int ipport_lastauto;		/* max port for port allocation */
 extern int ipport_hifirstauto;		/* min dynamic/private port number */
 extern int ipport_hilastauto;		/* max dynamic/private port number */
-extern int ipforwarding;		/* enable IP forwarding */
+extern int ip_forwarding;		/* enable IP forwarding */
 #ifdef MROUTING
 extern int ipmforwarding;		/* enable multicast forwarding */
 #endif
 extern int ipmultipath;			/* enable multipath routing */
+extern int ip_directedbcast;		/* accept all broadcast packets */
 extern unsigned int la_hold_total;
 
 extern const struct pr_usrreqs rip_usrreqs;
diff --git a/sys/netinet/tcp_input.c b/sys/netinet/tcp_input.c
index 328521d6f..9ba66ae6d 100644
--- a/sys/netinet/tcp_input.c
+++ b/sys/netinet/tcp_input.c
@@ -1,4 +1,4 @@
-/*	$OpenBSD: tcp_input.c,v 1.405 2024/04/17 20:48:51 bluhm Exp $	*/
+/*	$OpenBSD: tcp_input.c,v 1.406 2024/06/07 08:02:17 jsg Exp $	*/
 /*	$NetBSD: tcp_input.c,v 1.23 1996/02/13 23:43:44 christos Exp $	*/
 
 /*
@@ -106,11 +106,6 @@ int tcp_flush_queue(struct tcpcb *);
 #ifdef INET6
 #include <netinet6/in6_var.h>
 #include <netinet6/nd6.h>
-
-/* for the packet header length in the mbuf */
-#define M_PH_LEN(m)      (((struct mbuf *)(m))->m_pkthdr.len)
-#define M_V6_LEN(m)      (M_PH_LEN(m) - sizeof(struct ip6_hdr))
-#define M_V4_LEN(m)      (M_PH_LEN(m) - sizeof(struct ip))
 #endif /* INET6 */
 
 int	tcprexmtthresh = 3;
diff --git a/sys/netinet6/in6.c b/sys/netinet6/in6.c
index d5e8e2c25..9202ddacb 100644
--- a/sys/netinet6/in6.c
+++ b/sys/netinet6/in6.c
@@ -1,4 +1,4 @@
-/*	$OpenBSD: in6.c,v 1.266 2024/05/21 15:12:25 florian Exp $	*/
+/*	$OpenBSD: in6.c,v 1.267 2024/06/07 09:48:19 florian Exp $	*/
 /*	$KAME: in6.c,v 1.372 2004/06/14 08:14:21 itojun Exp $	*/
 
 /*
@@ -549,7 +549,7 @@ in6_update_ifa(struct ifnet *ifp, struct in6_aliasreq *ifra,
     struct in6_ifaddr *ia6)
 {
 	int error = 0, hostIsNew = 0, plen = -1;
-	struct sockaddr_in6 dst6;
+	struct sockaddr_in6 dst6, gw6;
 	struct in6_addrlifetime *lt;
 	struct in6_multi_mship *imm;
 	struct rtentry *rt;
@@ -604,7 +604,13 @@ in6_update_ifa(struct ifnet *ifp, struct in6_aliasreq *ifra,
 		plen = in6_mask2len(&ia6->ia_prefixmask.sin6_addr, NULL);
 	}
 
-	dst6 = ifra->ifra_dstaddr;
+	if (ifra->ifra_flags & IN6_IFF_AUTOCONF) {
+		gw6 = ifra->ifra_dstaddr;
+		memset(&dst6, 0, sizeof(dst6));
+	} else {
+		dst6 = ifra->ifra_dstaddr;
+		memset(&gw6, 0, sizeof(gw6));
+	}
 	if (dst6.sin6_family == AF_INET6) {
 		error = in6_check_embed_scope(&dst6, ifp->if_index);
 		if (error)
@@ -614,6 +620,11 @@ in6_update_ifa(struct ifnet *ifp, struct in6_aliasreq *ifra,
 		    (ifp->if_flags & IFF_LOOPBACK)) && plen != 128)
 			return (EINVAL);
 	}
+	if (gw6.sin6_family == AF_INET6) {
+		error = in6_check_embed_scope(&gw6, ifp->if_index);
+		if (error)
+			return error;
+	}
 	/* lifetime consistency check */
 	lt = &ifra->ifra_lifetime;
 	if (lt->ia6t_pltime > lt->ia6t_vltime)
@@ -702,10 +713,10 @@ in6_update_ifa(struct ifnet *ifp, struct in6_aliasreq *ifra,
 	}
 
 	if ((ifra->ifra_flags & IN6_IFF_AUTOCONF) &&
-	    dst6.sin6_family == AF_INET6 &&
+	    gw6.sin6_family == AF_INET6 &&
 	    !IN6_ARE_ADDR_EQUAL(&dst6.sin6_addr, &ia6->ia_gwaddr.sin6_addr)) {
 		/* Set or update announcing router */
-		ia6->ia_gwaddr = dst6;
+		ia6->ia_gwaddr = gw6;
 	}
 
 	/*
diff --git a/sys/netinet6/ip6_forward.c b/sys/netinet6/ip6_forward.c
index 913fb6430..13e603754 100644
--- a/sys/netinet6/ip6_forward.c
+++ b/sys/netinet6/ip6_forward.c
@@ -1,4 +1,4 @@
-/*	$OpenBSD: ip6_forward.c,v 1.117 2024/04/16 12:56:39 bluhm Exp $	*/
+/*	$OpenBSD: ip6_forward.c,v 1.118 2024/06/07 18:24:16 bluhm Exp $	*/
 /*	$KAME: ip6_forward.c,v 1.75 2001/06/29 12:42:13 jinmei Exp $	*/
 
 /*
@@ -75,7 +75,7 @@
  * of codes and types.
  *
  * If not forwarding, just drop the packet.  This could be confusing
- * if ipforwarding was zero but some routing protocol was advancing
+ * if ip6_forwarding was zero but some routing protocol was advancing
  * us as a gateway to somewhere.  However, we must let the routing
  * protocol deal with that.
  *
diff --git a/sys/netinet6/ip6_mroute.c b/sys/netinet6/ip6_mroute.c
index bf700ec32..f2ab244ff 100644
--- a/sys/netinet6/ip6_mroute.c
+++ b/sys/netinet6/ip6_mroute.c
@@ -1,4 +1,4 @@
-/*	$OpenBSD: ip6_mroute.c,v 1.141 2024/04/06 14:23:27 bluhm Exp $	*/
+/*	$OpenBSD: ip6_mroute.c,v 1.142 2024/06/07 08:37:59 jsg Exp $	*/
 /*	$NetBSD: ip6_mroute.c,v 1.59 2003/12/10 09:28:38 itojun Exp $	*/
 /*	$KAME: ip6_mroute.c,v 1.45 2001/03/25 08:38:51 itojun Exp $	*/
 
@@ -135,34 +135,6 @@ int		ip6_mrouter_ver = 0;
 int		ip6_mrtproto;    /* for netstat only */
 struct mrt6stat	mrt6stat;
 
-#define NO_RTE_FOUND	0x1
-#define RTE_FOUND	0x2
-
-/*
- * Macros to compute elapsed time efficiently
- * Borrowed from Van Jacobson's scheduling code
- */
-#define TV_DELTA(a, b, delta) do { \
-	    int xxs; \
-		\
-	    delta = (a).tv_usec - (b).tv_usec; \
-	    if ((xxs = (a).tv_sec - (b).tv_sec)) { \
-	       switch (xxs) { \
-		      case 2: \
-			  delta += 1000000; \
-			      /* FALLTHROUGH */ \
-		      case 1: \
-			  delta += 1000000; \
-			  break; \
-		      default: \
-			  delta += (1000000 * xxs); \
-	       } \
-	    } \
-} while (0)
-
-#define TV_LT(a, b) (((a).tv_usec < (b).tv_usec && \
-	      (a).tv_sec <= (b).tv_sec) || (a).tv_sec < (b).tv_sec)
-
 int get_sg6_cnt(struct sioc_sg_req6 *, unsigned int);
 int get_mif6_cnt(struct sioc_mif_req6 *, unsigned int);
 int ip6_mrouter_init(struct socket *, int, int);
diff --git a/sys/ufs/ext2fs/ext2fs_vfsops.c b/sys/ufs/ext2fs/ext2fs_vfsops.c
index f74a722a2..ce929251c 100644
--- a/sys/ufs/ext2fs/ext2fs_vfsops.c
+++ b/sys/ufs/ext2fs/ext2fs_vfsops.c
@@ -1,4 +1,4 @@
-/*	$OpenBSD: ext2fs_vfsops.c,v 1.119 2024/05/12 09:19:54 jsg Exp $	*/
+/*	$OpenBSD: ext2fs_vfsops.c,v 1.120 2024/06/07 09:26:37 jsg Exp $	*/
 /*	$NetBSD: ext2fs_vfsops.c,v 1.1 1997/06/11 09:34:07 bouyer Exp $	*/
 
 /*
@@ -101,10 +101,7 @@ ext2fs_init(struct vfsconf *vfsp)
 
 /*
  * Called by main() when ext2fs is going to be mounted as root.
- *
- * Name is updated by mount(8) after booting.
  */
-#define ROOTNAME	"root_device"
 
 int
 ext2fs_mountroot(void)
diff --git a/sys/uvm/uvm_addr.c b/sys/uvm/uvm_addr.c
index 69aff636a..904218034 100644
--- a/sys/uvm/uvm_addr.c
+++ b/sys/uvm/uvm_addr.c
@@ -1,4 +1,4 @@
-/*	$OpenBSD: uvm_addr.c,v 1.34 2024/05/13 01:15:53 jsg Exp $	*/
+/*	$OpenBSD: uvm_addr.c,v 1.35 2024/06/07 06:04:43 jsg Exp $	*/
 
 /*
  * Copyright (c) 2011 Ariane van der Steldt <ariane@stack.nl>
@@ -24,8 +24,6 @@
 #include <uvm/uvm_addr.h>
 #include <sys/pool.h>
 
-/* Max gap between hint allocations. */
-#define UADDR_HINT_MAXGAP	(4 * PAGE_SIZE)
 /* Number of pivots in pivot allocator. */
 #define NUM_PIVOTS		16
 /*
diff --git a/usr.bin/ssh/misc.c b/usr.bin/ssh/misc.c
index 38f987549..33327ad5c 100644
--- a/usr.bin/ssh/misc.c
+++ b/usr.bin/ssh/misc.c
@@ -1,4 +1,4 @@
-/* $OpenBSD: misc.c,v 1.195 2024/05/17 06:11:17 deraadt Exp $ */
+/* $OpenBSD: misc.c,v 1.196 2024/06/06 17:15:25 djm Exp $ */
 /*
  * Copyright (c) 2000 Markus Friedl.  All rights reserved.
  * Copyright (c) 2005-2020 Damien Miller.  All rights reserved.
@@ -2947,3 +2947,19 @@ lib_contains_symbol(const char *path, const char *s)
 	free(nl[0].n_name);
 	return ret;
 }
+
+int
+signal_is_crash(int sig)
+{
+	switch (sig) {
+	case SIGSEGV:
+	case SIGBUS:
+	case SIGTRAP:
+	case SIGSYS:
+	case SIGFPE:
+	case SIGILL:
+	case SIGABRT:
+		return 1;
+	}
+	return 0;
+}
diff --git a/usr.bin/ssh/misc.h b/usr.bin/ssh/misc.h
index 00b8bc5d0..7589d28e8 100644
--- a/usr.bin/ssh/misc.h
+++ b/usr.bin/ssh/misc.h
@@ -1,4 +1,4 @@
-/* $OpenBSD: misc.h,v 1.108 2024/05/17 00:30:24 djm Exp $ */
+/* $OpenBSD: misc.h,v 1.109 2024/06/06 17:15:25 djm Exp $ */
 
 /*
  * Author: Tatu Ylonen <ylo@cs.hut.fi>
@@ -250,6 +250,7 @@ void	notify_complete(struct notifier_ctx *, const char *, ...)
 
 typedef void (*sshsig_t)(int);
 sshsig_t ssh_signal(int, sshsig_t);
+int signal_is_crash(int);
 
 /* On OpenBSD time_t is int64_t which is long long. */
 #define SSH_TIME_T_MAX LLONG_MAX
diff --git a/usr.bin/ssh/monitor.c b/usr.bin/ssh/monitor.c
index b5bbdf305..0ada9a5b9 100644
--- a/usr.bin/ssh/monitor.c
+++ b/usr.bin/ssh/monitor.c
@@ -1,4 +1,4 @@
-/* $OpenBSD: monitor.c,v 1.239 2024/05/17 06:42:04 jsg Exp $ */
+/* $OpenBSD: monitor.c,v 1.240 2024/06/06 17:15:25 djm Exp $ */
 /*
  * Copyright 2002 Niels Provos <provos@citi.umich.edu>
  * Copyright 2002 Markus Friedl <markus@openbsd.org>
@@ -132,6 +132,7 @@ static char *auth_submethod = NULL;
 static u_int session_id2_len = 0;
 static u_char *session_id2 = NULL;
 static pid_t monitor_child_pid;
+int auth_attempted = 0;
 
 struct mon_table {
 	enum monitor_reqtype type;
@@ -248,6 +249,10 @@ monitor_child_preauth(struct ssh *ssh, struct monitor *pmonitor)
 		authenticated = (monitor_read(ssh, pmonitor,
 		    mon_dispatch, &ent) == 1);
 
+		/* Record that auth was attempted to set exit status later */
+		if ((ent->flags & MON_AUTH) != 0)
+			auth_attempted = 1;
+
 		/* Special handling for multiple required authentications */
 		if (options.num_auth_methods != 0) {
 			if (authenticated &&
@@ -290,6 +295,7 @@ monitor_child_preauth(struct ssh *ssh, struct monitor *pmonitor)
 		fatal_f("authentication method name unknown");
 
 	debug_f("user %s authenticated by privileged process", authctxt->user);
+	auth_attempted = 0;
 	ssh->authctxt = NULL;
 	ssh_packet_set_log_preamble(ssh, "user %s", authctxt->user);
 
diff --git a/usr.bin/ssh/monitor_wrap.c b/usr.bin/ssh/monitor_wrap.c
index 6287a8c71..ae254bc26 100644
--- a/usr.bin/ssh/monitor_wrap.c
+++ b/usr.bin/ssh/monitor_wrap.c
@@ -1,4 +1,4 @@
-/* $OpenBSD: monitor_wrap.c,v 1.130 2024/05/17 00:30:24 djm Exp $ */
+/* $OpenBSD: monitor_wrap.c,v 1.131 2024/06/06 17:15:25 djm Exp $ */
 /*
  * Copyright 2002 Niels Provos <provos@citi.umich.edu>
  * Copyright 2002 Markus Friedl <markus@openbsd.org>
@@ -28,6 +28,7 @@
 #include <sys/types.h>
 #include <sys/uio.h>
 #include <sys/queue.h>
+#include <sys/wait.h>
 
 #include <errno.h>
 #include <pwd.h>
@@ -69,6 +70,7 @@
 #include "session.h"
 #include "servconf.h"
 #include "monitor_wrap.h"
+#include "srclimit.h"
 
 #include "ssherr.h"
 
@@ -133,6 +135,36 @@ mm_request_send(int sock, enum monitor_reqtype type, struct sshbuf *m)
 		fatal_f("write: %s", strerror(errno));
 }
 
+static void
+mm_reap(void)
+{
+	int status = -1;
+
+	if (!mm_is_monitor())
+		return;
+	while (waitpid(pmonitor->m_pid, &status, 0) == -1) {
+		if (errno == EINTR)
+			continue;
+		pmonitor->m_pid = -1;
+		fatal_f("waitpid: %s", strerror(errno));
+	}
+	if (WIFEXITED(status)) {
+		if (WEXITSTATUS(status) != 0) {
+			debug_f("preauth child exited with status %d",
+			    WEXITSTATUS(status));
+			cleanup_exit(255);
+		}
+	} else if (WIFSIGNALED(status)) {
+		error_f("preauth child terminated by signal %d",
+		    WTERMSIG(status));
+		cleanup_exit(signal_is_crash(WTERMSIG(status)) ?
+		    EXIT_CHILD_CRASH : 255);
+	} else {
+		error_f("preauth child terminated abnormally");
+		cleanup_exit(EXIT_CHILD_CRASH);
+	}
+}
+
 void
 mm_request_receive(int sock, struct sshbuf *m)
 {
@@ -145,6 +177,7 @@ mm_request_receive(int sock, struct sshbuf *m)
 	if (atomicio(read, sock, buf, sizeof(buf)) != sizeof(buf)) {
 		if (errno == EPIPE) {
 			debug3_f("monitor fd closed");
+			mm_reap();
 			cleanup_exit(255);
 		}
 		fatal_f("read: %s", strerror(errno));
diff --git a/usr.bin/ssh/servconf.c b/usr.bin/ssh/servconf.c
index 02ec30553..89909819c 100644
--- a/usr.bin/ssh/servconf.c
+++ b/usr.bin/ssh/servconf.c
@@ -1,4 +1,4 @@
-/* $OpenBSD: servconf.c,v 1.407 2024/05/17 01:17:40 djm Exp $ */
+/* $OpenBSD: servconf.c,v 1.409 2024/06/06 20:25:48 djm Exp $ */
 /*
  * Copyright (c) 1995 Tatu Ylonen <ylo@cs.hut.fi>, Espoo, Finland
  *                    All rights reserved
@@ -145,6 +145,16 @@ initialize_server_options(ServerOptions *options)
 	options->per_source_max_startups = -1;
 	options->per_source_masklen_ipv4 = -1;
 	options->per_source_masklen_ipv6 = -1;
+	options->per_source_penalty_exempt = NULL;
+	options->per_source_penalty.enabled = -1;
+	options->per_source_penalty.max_sources = -1;
+	options->per_source_penalty.overflow_mode = -1;
+	options->per_source_penalty.penalty_crash = -1;
+	options->per_source_penalty.penalty_authfail = -1;
+	options->per_source_penalty.penalty_noauth = -1;
+	options->per_source_penalty.penalty_grace = -1;
+	options->per_source_penalty.penalty_max = -1;
+	options->per_source_penalty.penalty_min = -1;
 	options->max_authtries = -1;
 	options->max_sessions = -1;
 	options->banner = NULL;
@@ -377,6 +387,24 @@ fill_default_server_options(ServerOptions *options)
 		options->per_source_masklen_ipv4 = 32;
 	if (options->per_source_masklen_ipv6 == -1)
 		options->per_source_masklen_ipv6 = 128;
+	if (options->per_source_penalty.enabled == -1)
+		options->per_source_penalty.enabled = 1;
+	if (options->per_source_penalty.max_sources == -1)
+		options->per_source_penalty.max_sources = 65536;
+	if (options->per_source_penalty.overflow_mode == -1)
+		options->per_source_penalty.overflow_mode = PER_SOURCE_PENALTY_OVERFLOW_PERMISSIVE;
+	if (options->per_source_penalty.penalty_crash == -1)
+		options->per_source_penalty.penalty_crash = 90;
+	if (options->per_source_penalty.penalty_grace == -1)
+		options->per_source_penalty.penalty_grace = 20;
+	if (options->per_source_penalty.penalty_authfail == -1)
+		options->per_source_penalty.penalty_authfail = 5;
+	if (options->per_source_penalty.penalty_noauth == -1)
+		options->per_source_penalty.penalty_noauth = 1;
+	if (options->per_source_penalty.penalty_min == -1)
+		options->per_source_penalty.penalty_min = 15;
+	if (options->per_source_penalty.penalty_max == -1)
+		options->per_source_penalty.penalty_max = 600;
 	if (options->max_authtries == -1)
 		options->max_authtries = DEFAULT_AUTH_FAIL_MAX;
 	if (options->max_sessions == -1)
@@ -454,6 +482,7 @@ fill_default_server_options(ServerOptions *options)
 	CLEAR_ON_NONE(options->chroot_directory);
 	CLEAR_ON_NONE(options->routing_domain);
 	CLEAR_ON_NONE(options->host_key_agent);
+	CLEAR_ON_NONE(options->per_source_penalty_exempt);
 
 	for (i = 0; i < options->num_host_key_files; i++)
 		CLEAR_ON_NONE(options->host_key_files[i]);
@@ -485,6 +514,7 @@ typedef enum {
 	sBanner, sUseDNS, sHostbasedAuthentication,
 	sHostbasedUsesNameFromPacketOnly, sHostbasedAcceptedAlgorithms,
 	sHostKeyAlgorithms, sPerSourceMaxStartups, sPerSourceNetBlockSize,
+	sPerSourcePenalties, sPerSourcePenaltyExemptList,
 	sClientAliveInterval, sClientAliveCountMax, sAuthorizedKeysFile,
 	sGssAuthentication, sGssCleanupCreds, sGssStrictAcceptor,
 	sAcceptEnv, sSetEnv, sPermitTunnel,
@@ -601,6 +631,8 @@ static struct {
 	{ "maxstartups", sMaxStartups, SSHCFG_GLOBAL },
 	{ "persourcemaxstartups", sPerSourceMaxStartups, SSHCFG_GLOBAL },
 	{ "persourcenetblocksize", sPerSourceNetBlockSize, SSHCFG_GLOBAL },
+	{ "persourcepenalties", sPerSourcePenalties, SSHCFG_GLOBAL },
+	{ "persourcepenaltyexemptlist", sPerSourcePenaltyExemptList, SSHCFG_GLOBAL },
 	{ "maxauthtries", sMaxAuthTries, SSHCFG_ALL },
 	{ "maxsessions", sMaxSessions, SSHCFG_ALL },
 	{ "banner", sBanner, SSHCFG_ALL },
@@ -1888,6 +1920,89 @@ process_server_config_line_depth(ServerOptions *options, char *line,
 			options->per_source_max_startups = value;
 		break;
 
+	case sPerSourcePenaltyExemptList:
+		charptr = &options->per_source_penalty_exempt;
+		arg = argv_next(&ac, &av);
+		if (!arg || *arg == '\0')
+			fatal("%s line %d: missing file name.",
+			    filename, linenum);
+		if (addr_match_list(NULL, arg) != 0) {
+			fatal("%s line %d: keyword %s "
+			    "invalid address argument.",
+			    filename, linenum, keyword);
+		}
+		if (*activep && *charptr == NULL)
+			*charptr = xstrdup(arg);
+		break;
+
+	case sPerSourcePenalties:
+		while ((arg = argv_next(&ac, &av)) != NULL) {
+			found = 1;
+			value = -1;
+			value2 = 0;
+			p = NULL;
+			/* Allow no/yes only in first position */
+			if (strcasecmp(arg, "no") == 0 ||
+			    (value2 = (strcasecmp(arg, "yes") == 0))) {
+				if (ac > 0) {
+					fatal("%s line %d: keyword %s \"%s\" "
+					    "argument must appear alone.",
+					    filename, linenum, keyword, arg);
+				}
+				if (*activep &&
+				    options->per_source_penalty.enabled == -1)
+					options->per_source_penalty.enabled = value2;
+				continue;
+			} else if (strncmp(arg, "crash:", 6) == 0) {
+				p = arg + 6;
+				intptr = &options->per_source_penalty.penalty_crash;
+			} else if (strncmp(arg, "authfail:", 9) == 0) {
+				p = arg + 9;
+				intptr = &options->per_source_penalty.penalty_authfail;
+			} else if (strncmp(arg, "noauth:", 7) == 0) {
+				p = arg + 7;
+				intptr = &options->per_source_penalty.penalty_noauth;
+			} else if (strncmp(arg, "grace-exceeded:", 15) == 0) {
+				p = arg + 15;
+				intptr = &options->per_source_penalty.penalty_grace;
+			} else if (strncmp(arg, "max:", 4) == 0) {
+				p = arg + 4;
+				intptr = &options->per_source_penalty.penalty_max;
+			} else if (strncmp(arg, "min:", 4) == 0) {
+				p = arg + 4;
+				intptr = &options->per_source_penalty.penalty_min;
+			} else if (strncmp(arg, "max-sources:", 12) == 0) {
+				intptr = &options->per_source_penalty.max_sources;
+				if ((errstr = atoi_err(arg+12, &value)) != NULL)
+					fatal("%s line %d: %s value %s.",
+					    filename, linenum, keyword, errstr);
+			} else if (strcmp(arg, "overflow:deny-all") == 0) {
+				intptr = &options->per_source_penalty.overflow_mode;
+				value = PER_SOURCE_PENALTY_OVERFLOW_DENY_ALL;
+			} else if (strcmp(arg, "overflow:permissive") == 0) {
+				intptr = &options->per_source_penalty.overflow_mode;
+				value = PER_SOURCE_PENALTY_OVERFLOW_PERMISSIVE;
+			} else {
+				fatal("%s line %d: unsupported %s keyword %s",
+				    filename, linenum, keyword, arg);
+			}
+			/* If no value was parsed above, assume it's a time */
+			if (value == -1 && (value = convtime(p)) == -1) {
+				fatal("%s line %d: invalid %s time value.",
+				    filename, linenum, keyword);
+			}
+			if (*activep && *intptr == -1) {
+				*intptr = value;
+				/* any option implicitly enables penalties */
+				options->per_source_penalty.enabled = 1;
+			}
+		}
+		if (!found) {
+			fatal("%s line %d: no %s specified",
+			    filename, linenum, keyword);
+		}
+		break;
+
 	case sMaxAuthTries:
 		intptr = &options->max_authtries;
 		goto parse_int;
@@ -3012,6 +3127,7 @@ dump_config(ServerOptions *o)
 	dump_cfg_string(sPubkeyAcceptedAlgorithms, o->pubkey_accepted_algos);
 	dump_cfg_string(sRDomain, o->routing_domain);
 	dump_cfg_string(sSshdSessionPath, o->sshd_session_path);
+	dump_cfg_string(sPerSourcePenaltyExemptList, o->per_source_penalty_exempt);
 
 	/* string arguments requiring a lookup */
 	dump_cfg_string(sLogLevel, log_level_name(o->log_level));
@@ -3099,4 +3215,20 @@ dump_config(ServerOptions *o)
 	if (o->pubkey_auth_options & PUBKEYAUTH_VERIFY_REQUIRED)
 		printf(" verify-required");
 	printf("\n");
+
+	if (o->per_source_penalty.enabled) {
+		printf("persourcepenalties crash:%d authfail:%d noauth:%d "
+		    "grace-exceeded:%d max:%d min:%d max-sources:%d "
+		    "overflow:%s\n", o->per_source_penalty.penalty_crash,
+		    o->per_source_penalty.penalty_authfail,
+		    o->per_source_penalty.penalty_noauth,
+		    o->per_source_penalty.penalty_grace,
+		    o->per_source_penalty.penalty_max,
+		    o->per_source_penalty.penalty_min,
+		    o->per_source_penalty.max_sources,
+		    o->per_source_penalty.overflow_mode ==
+		    PER_SOURCE_PENALTY_OVERFLOW_DENY_ALL ?
+		    "deny-all" : "permissive");
+	} else
+		printf("persourcepenalties no\n");
 }
diff --git a/usr.bin/ssh/servconf.h b/usr.bin/ssh/servconf.h
index 8ebdca5e2..4a4ac1cdb 100644
--- a/usr.bin/ssh/servconf.h
+++ b/usr.bin/ssh/servconf.h
@@ -1,4 +1,4 @@
-/* $OpenBSD: servconf.h,v 1.163 2024/05/23 23:47:16 jsg Exp $ */
+/* $OpenBSD: servconf.h,v 1.164 2024/06/06 17:15:25 djm Exp $ */
 
 /*
  * Author: Tatu Ylonen <ylo@cs.hut.fi>
@@ -65,6 +65,20 @@ struct listenaddr {
 	struct addrinfo *addrs;
 };
 
+#define PER_SOURCE_PENALTY_OVERFLOW_DENY_ALL	1
+#define PER_SOURCE_PENALTY_OVERFLOW_PERMISSIVE	2
+struct per_source_penalty {
+	int	enabled;
+	int	max_sources;
+	int	overflow_mode;
+	int	penalty_crash;
+	int	penalty_grace;
+	int	penalty_authfail;
+	int	penalty_noauth;
+	int	penalty_max;
+	int	penalty_min;
+};
+
 typedef struct {
 	u_int	num_ports;
 	u_int	ports_from_cmdline;
@@ -172,6 +186,8 @@ typedef struct {
 	int	per_source_max_startups;
 	int	per_source_masklen_ipv4;
 	int	per_source_masklen_ipv6;
+	char	*per_source_penalty_exempt;
+	struct per_source_penalty per_source_penalty;
 	int	max_authtries;
 	int	max_sessions;
 	char   *banner;			/* SSH-2 banner message */
diff --git a/usr.bin/ssh/srclimit.c b/usr.bin/ssh/srclimit.c
index 853a0eef9..ffea6a397 100644
--- a/usr.bin/ssh/srclimit.c
+++ b/usr.bin/ssh/srclimit.c
@@ -1,5 +1,6 @@
 /*
  * Copyright (c) 2020 Darren Tucker <dtucker@openbsd.org>
+ * Copyright (c) 2024 Damien Miller <djm@mindrot.org>
  *
  * Permission to use, copy, modify, and distribute this software for any
  * purpose with or without fee is hereby granted, provided that the above
@@ -16,11 +17,13 @@
 
 #include <sys/socket.h>
 #include <sys/types.h>
+#include <sys/tree.h>
 
 #include <limits.h>
 #include <netdb.h>
 #include <stdio.h>
 #include <string.h>
+#include <stdlib.h>
 
 #include "addr.h"
 #include "canohost.h"
@@ -28,17 +31,71 @@
 #include "misc.h"
 #include "srclimit.h"
 #include "xmalloc.h"
+#include "servconf.h"
+#include "match.h"
 
 static int max_children, max_persource, ipv4_masklen, ipv6_masklen;
+static struct per_source_penalty penalty_cfg;
+static char *penalty_exempt;
 
 /* Per connection state, used to enforce unauthenticated connection limit. */
 static struct child_info {
 	int id;
 	struct xaddr addr;
-} *child;
+} *children;
+
+/*
+ * Penalised addresses, active entries here prohibit connections until expired.
+ * Entries become active when more than penalty_min seconds of penalty are
+ * outstanding.
+ */
+struct penalty {
+	struct xaddr addr;
+	time_t expiry;
+	int active;
+	const char *reason;
+	RB_ENTRY(penalty) by_addr;
+	RB_ENTRY(penalty) by_expiry;
+};
+static int penalty_addr_cmp(struct penalty *a, struct penalty *b);
+static int penalty_expiry_cmp(struct penalty *a, struct penalty *b);
+RB_HEAD(penalties_by_addr, penalty) penalties_by_addr;
+RB_HEAD(penalties_by_expiry, penalty) penalties_by_expiry;
+RB_GENERATE_STATIC(penalties_by_addr, penalty, by_addr, penalty_addr_cmp)
+RB_GENERATE_STATIC(penalties_by_expiry, penalty, by_expiry, penalty_expiry_cmp)
+static size_t npenalties;
+
+static int
+srclimit_mask_addr(const struct xaddr *addr, int bits, struct xaddr *masked)
+{
+	struct xaddr xmask;
+
+	/* Mask address off address to desired size. */
+	if (addr_netmask(addr->af, bits, &xmask) != 0 ||
+	    addr_and(masked, addr, &xmask) != 0) {
+		debug3_f("%s: invalid mask %d bits", __func__, bits);
+		return -1;
+	}
+	return 0;
+}
+
+static int
+srclimit_peer_addr(int sock, struct xaddr *addr)
+{
+	struct sockaddr_storage storage;
+	socklen_t addrlen = sizeof(storage);
+	struct sockaddr *sa = (struct sockaddr *)&storage;
+
+	if (getpeername(sock, sa, &addrlen) != 0)
+		return 1;	/* not remote socket? */
+	if (addr_sa_to_xaddr(sa, addrlen, addr) != 0)
+		return 1;	/* unknown address family? */
+	return 0;
+}
 
 void
-srclimit_init(int max, int persource, int ipv4len, int ipv6len)
+srclimit_init(int max, int persource, int ipv4len, int ipv6len,
+    struct per_source_penalty *penalty_conf, const char *penalty_exempt_conf)
 {
 	int i;
 
@@ -46,25 +103,27 @@ srclimit_init(int max, int persource, int ipv4len, int ipv6len)
 	ipv4_masklen = ipv4len;
 	ipv6_masklen = ipv6len;
 	max_persource = persource;
+	penalty_cfg = *penalty_conf;
+	penalty_exempt = penalty_exempt_conf == NULL ?
+	    NULL : xstrdup(penalty_exempt_conf);
 	if (max_persource == INT_MAX)	/* no limit */
 		return;
 	debug("%s: max connections %d, per source %d, masks %d,%d", __func__,
 	    max, persource, ipv4len, ipv6len);
 	if (max <= 0)
 		fatal("%s: invalid number of sockets: %d", __func__, max);
-	child = xcalloc(max_children, sizeof(*child));
+	children = xcalloc(max_children, sizeof(*children));
 	for (i = 0; i < max_children; i++)
-		child[i].id = -1;
+		children[i].id = -1;
+	RB_INIT(&penalties_by_addr);
+	RB_INIT(&penalties_by_expiry);
 }
 
 /* returns 1 if connection allowed, 0 if not allowed. */
 int
 srclimit_check_allow(int sock, int id)
 {
-	struct xaddr xa, xb, xmask;
-	struct sockaddr_storage addr;
-	socklen_t addrlen = sizeof(addr);
-	struct sockaddr *sa = (struct sockaddr *)&addr;
+	struct xaddr xa, xb;
 	int i, bits, first_unused, count = 0;
 	char xas[NI_MAXHOST];
 
@@ -72,26 +131,19 @@ srclimit_check_allow(int sock, int id)
 		return 1;
 
 	debug("%s: sock %d id %d limit %d", __func__, sock, id, max_persource);
-	if (getpeername(sock, sa, &addrlen) != 0)
-		return 1;	/* not remote socket? */
-	if (addr_sa_to_xaddr(sa, addrlen, &xa) != 0)
-		return 1;	/* unknown address family? */
-
-	/* Mask address off address to desired size. */
-	bits = xa.af == AF_INET ? ipv4_masklen : ipv6_masklen;
-	if (addr_netmask(xa.af, bits, &xmask) != 0 ||
-	    addr_and(&xb, &xa, &xmask) != 0) {
-		debug3("%s: invalid mask %d bits", __func__, bits);
+	if (srclimit_peer_addr(sock, &xa) != 0)
+		return 1;
+	bits = xa.af == AF_INET ? ipv4_masklen : ipv6_masklen;
+	if (srclimit_mask_addr(&xa, bits, &xb) != 0)
 		return 1;
-	}
 
 	first_unused = max_children;
 	/* Count matching entries and find first unused one. */
 	for (i = 0; i < max_children; i++) {
-		if (child[i].id == -1) {
+		if (children[i].id == -1) {
 			if (i < first_unused)
 				first_unused = i;
-		} else if (addr_cmp(&child[i].addr, &xb) == 0) {
+		} else if (addr_cmp(&children[i].addr, &xb) == 0) {
 			count++;
 		}
 	}
@@ -114,8 +166,8 @@ srclimit_check_allow(int sock, int id)
 		return 0;
 
 	/* Connection allowed, store masked address. */
-	child[first_unused].id = id;
-	memcpy(&child[first_unused].addr, &xb, sizeof(xb));
+	children[first_unused].id = id;
+	memcpy(&children[first_unused].addr, &xb, sizeof(xb));
 	return 1;
 }
 
@@ -130,9 +182,249 @@ srclimit_done(int id)
 	debug("%s: id %d", __func__, id);
 	/* Clear corresponding state entry. */
 	for (i = 0; i < max_children; i++) {
-		if (child[i].id == id) {
-			child[i].id = -1;
+		if (children[i].id == id) {
+			children[i].id = -1;
 			return;
 		}
 	}
 }
+
+static int
+penalty_addr_cmp(struct penalty *a, struct penalty *b)
+{
+	return addr_cmp(&a->addr, &b->addr);
+	/* Addresses must be unique in by_addr, so no need to tiebreak */
+}
+
+static int
+penalty_expiry_cmp(struct penalty *a, struct penalty *b)
+{
+	if (a->expiry != b->expiry)
+		return a->expiry < b->expiry ? -1 : 1;
+	/* Tiebreak on addresses */
+	return addr_cmp(&a->addr, &b->addr);
+}
+
+static void
+expire_penalties(time_t now)
+{
+	struct penalty *penalty, *tmp;
+
+	/* XXX avoid full scan of tree, e.g. min-heap */
+	RB_FOREACH_SAFE(penalty, penalties_by_expiry,
+	    &penalties_by_expiry, tmp) {
+		if (penalty->expiry >= now)
+			break;
+		if (RB_REMOVE(penalties_by_expiry, &penalties_by_expiry,
+		    penalty) != penalty ||
+		    RB_REMOVE(penalties_by_addr, &penalties_by_addr,
+		    penalty) != penalty)
+			fatal_f("internal error: penalty tables corrupt");
+		free(penalty);
+		if (npenalties-- == 0)
+			fatal_f("internal error: npenalties underflow");
+	}
+}
+
+static void
+addr_masklen_ntop(struct xaddr *addr, int masklen, char *s, size_t slen)
+{
+	size_t o;
+
+	if (addr_ntop(addr, s, slen) != 0) {
+		strlcpy(s, "UNKNOWN", slen);
+		return;
+	}
+	if ((o = strlen(s)) < slen)
+		snprintf(s + o, slen - o, "/%d", masklen);
+}
+
+int
+srclimit_penalty_check_allow(int sock, const char **reason)
+{
+	struct xaddr addr;
+	struct penalty find, *penalty;
+	time_t now;
+	int bits;
+	char addr_s[NI_MAXHOST];
+
+	if (!penalty_cfg.enabled)
+		return 1;
+	if (srclimit_peer_addr(sock, &addr) != 0)
+		return 1;
+	if (penalty_exempt != NULL) {
+		if (addr_ntop(&addr, addr_s, sizeof(addr_s)) != 0)
+			return 1; /* shouldn't happen */
+		if (addr_match_list(addr_s, penalty_exempt) == 1) {
+			return 1;
+		}
+	}
+	if (npenalties > (size_t)penalty_cfg.max_sources &&
+	    penalty_cfg.overflow_mode == PER_SOURCE_PENALTY_OVERFLOW_DENY_ALL) {
+		*reason = "too many penalised addresses";
+		return 0;
+	}
+	bits = addr.af == AF_INET ? ipv4_masklen : ipv6_masklen;
+	memset(&find, 0, sizeof(find));
+	if (srclimit_mask_addr(&addr, bits, &find.addr) != 0)
+		return 1;
+	now = monotime();
+	if ((penalty = RB_FIND(penalties_by_addr,
+	    &penalties_by_addr, &find)) == NULL)
+		return 1; /* no penalty */
+	if (penalty->expiry < now) {
+		expire_penalties(now);
+		return 1; /* expired penalty */
+	}
+	if (!penalty->active)
+		return 1; /* Penalty hasn't hit activation threshold yet */
+	*reason = penalty->reason;
+	return 0;
+}
+
+static void
+srclimit_remove_expired_penalties(void)
+{
+	struct penalty *p = NULL;
+	int bits;
+	char s[NI_MAXHOST + 4];
+
+	/* Delete the soonest-to-expire penalties. */
+	while (npenalties > (size_t)penalty_cfg.max_sources) {
+		if ((p = RB_MIN(penalties_by_expiry,
+		    &penalties_by_expiry)) == NULL)
+			break; /* shouldn't happen */
+		bits = p->addr.af == AF_INET ? ipv4_masklen : ipv6_masklen;
+		addr_masklen_ntop(&p->addr, bits, s, sizeof(s));
+		debug3_f("overflow, remove %s", s);
+		if (RB_REMOVE(penalties_by_expiry,
+		    &penalties_by_expiry, p) != p ||
+		    RB_REMOVE(penalties_by_addr, &penalties_by_addr, p) != p)
+			fatal_f("internal error: penalty tables corrupt");
+		free(p);
+		npenalties--;
+	}
+}
+
+void
+srclimit_penalise(struct xaddr *addr, int penalty_type)
+{
+	struct xaddr masked;
+	struct penalty *penalty, *existing;
+	time_t now;
+	int bits, penalty_secs;
+	char addrnetmask[NI_MAXHOST + 4];
+	const char *reason = NULL;
+
+	if (!penalty_cfg.enabled)
+		return;
+	if (penalty_exempt != NULL) {
+		if (addr_ntop(addr, addrnetmask, sizeof(addrnetmask)) != 0)
+			return; /* shouldn't happen */
+		if (addr_match_list(addrnetmask, penalty_exempt) == 1) {
+			debug3_f("address %s is exempt", addrnetmask);
+			return;
+		}
+	}
+
+	switch (penalty_type) {
+	case SRCLIMIT_PENALTY_NONE:
+		return;
+	case SRCLIMIT_PENALTY_CRASH:
+		penalty_secs = penalty_cfg.penalty_crash;
+		reason = "penalty: caused crash";
+		break;
+	case SRCLIMIT_PENALTY_AUTHFAIL:
+		penalty_secs = penalty_cfg.penalty_authfail;
+		reason = "penalty: failed authentication";
+		break;
+	case SRCLIMIT_PENALTY_NOAUTH:
+		penalty_secs = penalty_cfg.penalty_noauth;
+		reason = "penalty: connections without attempting authentication";
+		break;
+	case SRCLIMIT_PENALTY_GRACE_EXCEEDED:
+		penalty_secs = penalty_cfg.penalty_crash;
+		reason = "penalty: exceeded LoginGraceTime";
+		break;
+	default:
+		fatal_f("internal error: unknown penalty %d", penalty_type);
+	}
+	bits = addr->af == AF_INET ? ipv4_masklen : ipv6_masklen;
+	if (srclimit_mask_addr(addr, bits, &masked) != 0)
+		return;
+	addr_masklen_ntop(addr, bits, addrnetmask, sizeof(addrnetmask));
+
+	now = monotime();
+	expire_penalties(now);
+	if (npenalties > (size_t)penalty_cfg.max_sources &&
+	    penalty_cfg.overflow_mode == PER_SOURCE_PENALTY_OVERFLOW_DENY_ALL) {
+		verbose_f("penalty table full, cannot penalise %s for %s",
+		    addrnetmask, reason);
+		return;
+	}
+
+	penalty = xcalloc(1, sizeof(*penalty));
+	penalty->addr = masked;
+	penalty->expiry = now + penalty_secs;
+	penalty->reason = reason;
+	if ((existing = RB_INSERT(penalties_by_addr, &penalties_by_addr,
+	    penalty)) == NULL) {
+		/* penalty didn't previously exist */
+		if (penalty_secs > penalty_cfg.penalty_min)
+			penalty->active = 1;
+		if (RB_INSERT(penalties_by_expiry, &penalties_by_expiry,
+		    penalty) != NULL)
+			fatal_f("internal error: penalty tables corrupt");
+		verbose_f("%s: new %s penalty of %d seconds for %s",
+		    addrnetmask, penalty->active ? "active" : "deferred",
+		    penalty_secs, reason);
+		if (++npenalties > (size_t)penalty_cfg.max_sources)
+			srclimit_remove_expired_penalties(); /* permissive */
+		return;
+	}
+	debug_f("%s penalty for %s already exists, %lld seconds remaining",
+	    existing->active ? "active" : "inactive",
+	    addrnetmask, (long long)(existing->expiry - now));
+	/* Expiry information is about to change, remove from tree */
+	if (RB_REMOVE(penalties_by_expiry, &penalties_by_expiry,
+	    existing) != existing)
+		fatal_f("internal error: penalty tables corrupt (remove)");
+	/* An entry already existed. Accumulate penalty up to maximum */
+	existing->expiry += penalty_secs;
+	if (existing->expiry - now > penalty_cfg.penalty_max)
+		existing->expiry = now + penalty_cfg.penalty_max;
+	if (existing->expiry - now > penalty_cfg.penalty_min &&
+	    !existing->active) {
+		verbose_f("%s: activating penalty of %lld seconds for %s",
+		    addrnetmask, (long long)(existing->expiry - now), reason);
+		existing->active = 1;
+	}
+	existing->reason = penalty->reason;
+	free(penalty);
+	/* Re-insert into expiry tree */
+	if (RB_INSERT(penalties_by_expiry, &penalties_by_expiry,
+	    existing) != NULL)
+		fatal_f("internal error: penalty tables corrupt (insert)");
+}
+
+void
+srclimit_penalty_info(void)
+{
+	struct penalty *p = NULL;
+	int bits;
+	char s[NI_MAXHOST + 4];
+	time_t now;
+
+	now = monotime();
+	logit("%zu active penalties", npenalties);
+	RB_FOREACH(p, penalties_by_expiry, &penalties_by_expiry) {
+		bits = p->addr.af == AF_INET ? ipv4_masklen : ipv6_masklen;
+		addr_masklen_ntop(&p->addr, bits, s, sizeof(s));
+		if (p->expiry < now)
+			logit("client %s %s (expired)", s, p->reason);
+		else {
+			logit("client %s %s (%llu secs left)", s, p->reason,
+			   (long long)(p->expiry - now));
+		}
+	}
+}
diff --git a/usr.bin/ssh/srclimit.h b/usr.bin/ssh/srclimit.h
index 6e04f32b3..74a6f2b83 100644
--- a/usr.bin/ssh/srclimit.h
+++ b/usr.bin/ssh/srclimit.h
@@ -13,6 +13,26 @@
  * ACTION OF CONTRACT, NEGLIGENCE OR OTHER TORTIOUS ACTION, ARISING OUT OF
  * OR IN CONNECTION WITH THE USE OR PERFORMANCE OF THIS SOFTWARE.
  */
-void	srclimit_init(int, int, int, int);
+struct xaddr;
+
+struct per_source_penalty;
+
+void	srclimit_init(int, int, int, int,
+    struct per_source_penalty *, const char *);
 int	srclimit_check_allow(int, int);
 void	srclimit_done(int);
+
+#define SRCLIMIT_PENALTY_NONE		0
+#define SRCLIMIT_PENALTY_CRASH		1
+#define SRCLIMIT_PENALTY_AUTHFAIL	2
+#define SRCLIMIT_PENALTY_GRACE_EXCEEDED	3
+#define SRCLIMIT_PENALTY_NOAUTH		4
+
+/* meaningful exit values, used by sshd listener for penalties */
+#define EXIT_LOGIN_GRACE	3	/* login grace period exceeded */
+#define EXIT_CHILD_CRASH	4	/* preauth child crashed */
+#define EXIT_AUTH_ATTEMPTED	5	/* at least one auth attempt made */
+
+void	srclimit_penalise(struct xaddr *, int);
+int	srclimit_penalty_check_allow(int, const char **);
+void	srclimit_penalty_info(void);
diff --git a/usr.bin/ssh/sshd-session.c b/usr.bin/ssh/sshd-session.c
index b800f2b5a..e75b4f80b 100644
--- a/usr.bin/ssh/sshd-session.c
+++ b/usr.bin/ssh/sshd-session.c
@@ -1,4 +1,4 @@
-/* $OpenBSD: sshd-session.c,v 1.2 2024/05/17 02:39:11 jsg Exp $ */
+/* $OpenBSD: sshd-session.c,v 1.3 2024/06/06 17:15:25 djm Exp $ */
 /*
  * SSH2 implementation:
  * Privilege Separation:
@@ -188,11 +188,7 @@ grace_alarm_handler(int sig)
 		ssh_signal(SIGTERM, SIG_IGN);
 		kill(0, SIGTERM);
 	}
-
-	/* Log error and exit. */
-	sigdie("Timeout before authentication for %s port %d",
-	    ssh_remote_ipaddr(the_active_state),
-	    ssh_remote_port(the_active_state));
+	_exit(EXIT_LOGIN_GRACE);
 }
 
 /* Destroy the host and server keys.  They will no longer be needed. */
@@ -1220,6 +1216,8 @@ main(int ac, char **av)
 	ssh_signal(SIGALRM, SIG_DFL);
 	authctxt->authenticated = 1;
 	if (startup_pipe != -1) {
+		/* signal listener that authentication completed successfully */
+		(void)atomicio(vwrite, startup_pipe, "\001", 1);
 		close(startup_pipe);
 		startup_pipe = -1;
 	}
@@ -1338,6 +1336,8 @@ do_ssh2_kex(struct ssh *ssh)
 void
 cleanup_exit(int i)
 {
+	extern int auth_attempted; /* monitor.c */
+
 	if (the_active_state != NULL && the_authctxt != NULL) {
 		do_cleanup(the_active_state, the_authctxt);
 		if (privsep_is_preauth &&
@@ -1350,5 +1350,8 @@ cleanup_exit(int i)
 			}
 		}
 	}
+	/* Override default fatal exit value when auth was attempted */
+	if (i == 255 && auth_attempted)
+		_exit(EXIT_AUTH_ATTEMPTED);
 	_exit(i);
 }
diff --git a/usr.bin/ssh/sshd.c b/usr.bin/ssh/sshd.c
index d310779be..adc71cefe 100644
--- a/usr.bin/ssh/sshd.c
+++ b/usr.bin/ssh/sshd.c
@@ -1,4 +1,4 @@
-/* $OpenBSD: sshd.c,v 1.605 2024/06/01 07:03:37 djm Exp $ */
+/* $OpenBSD: sshd.c,v 1.607 2024/06/06 19:50:01 djm Exp $ */
 /*
  * Copyright (c) 2000, 2001, 2002 Markus Friedl.  All rights reserved.
  * Copyright (c) 2002 Niels Provos.  All rights reserved.
@@ -72,6 +72,7 @@
 #include "version.h"
 #include "ssherr.h"
 #include "sk-api.h"
+#include "addr.h"
 #include "srclimit.h"
 
 /* Re-exec fds */
@@ -120,6 +121,8 @@ struct {
 } sensitive_data;
 
 /* This is set to true when a signal is received. */
+static volatile sig_atomic_t received_siginfo = 0;
+static volatile sig_atomic_t received_sigchld = 0;
 static volatile sig_atomic_t received_sighup = 0;
 static volatile sig_atomic_t received_sigterm = 0;
 
@@ -127,8 +130,9 @@ static volatile sig_atomic_t received_sigterm = 0;
 u_int utmp_len = HOST_NAME_MAX+1;
 
 /*
- * startup_pipes/flags are used for tracking children of the listening sshd
- * process early in their lifespans. This tracking is needed for three things:
+ * The early_child/children array below is used for tracking children of the
+ * listening sshd process early in their lifespans, before they have
+ * completed authentication. This tracking is needed for four things:
  *
  * 1) Implementing the MaxStartups limit of concurrent unauthenticated
  *    connections.
@@ -137,14 +141,31 @@ u_int utmp_len = HOST_NAME_MAX+1;
  *    after it restarts.
  * 3) Ensuring that rexec'd sshd processes have received their initial state
  *    from the parent listen process before handling SIGHUP.
+ * 4) Tracking and logging unsuccessful exits from the preauth sshd monitor,
+ *    including and especially those for LoginGraceTime timeouts.
  *
  * Child processes signal that they have completed closure of the listen_socks
  * and (if applicable) received their rexec state by sending a char over their
- * sock. Child processes signal that authentication has completed by closing
- * the sock (or by exiting).
+ * sock.
+ *
+ * Child processes signal that authentication has completed by sending a
+ * second char over the socket before closing it, otherwise the listener will
+ * continue tracking the child (and using up a MaxStartups slot) until the
+ * preauth subprocess exits, whereupon the listener will log its exit status.
+ * preauth processes will exit with a status of EXIT_LOGIN_GRACE to indicate
+ * they did not authenticate before the LoginGraceTime alarm fired.
  */
-static int *startup_pipes = NULL;
-static int *startup_flags = NULL;	/* Indicates child closed listener */
+struct early_child {
+	int pipefd;
+	int early;		/* Indicates child closed listener */
+	char *id;		/* human readable connection identifier */
+	pid_t pid;
+	struct xaddr addr;
+	int have_addr;
+	int status, have_status;
+};
+static struct early_child *children;
+static int children_active;
 static int startup_pipe = -1;		/* in child */
 
 /* sshd_config buffer */
@@ -171,15 +192,257 @@ close_listen_socks(void)
 	num_listen_socks = 0;
 }
 
+/* Allocate and initialise the children array */
+static void
+child_alloc(void)
+{
+	int i;
+
+	children = xcalloc(options.max_startups, sizeof(*children));
+	for (i = 0; i < options.max_startups; i++) {
+		children[i].pipefd = -1;
+		children[i].pid = -1;
+	}
+}
+
+/* Register a new connection in the children array; child pid comes later */
+static struct early_child *
+child_register(int pipefd, int sockfd)
+{
+	int i, lport, rport;
+	char *laddr = NULL, *raddr = NULL;
+	struct early_child *child = NULL;
+	struct sockaddr_storage addr;
+	socklen_t addrlen = sizeof(addr);
+	struct sockaddr *sa = (struct sockaddr *)&addr;
+
+	for (i = 0; i < options.max_startups; i++) {
+		if (children[i].pipefd != -1 || children[i].pid > 0)
+			continue;
+		child = &(children[i]);
+		break;
+	}
+	if (child == NULL) {
+		fatal_f("error: accepted connection when all %d child "
+		    " slots full", options.max_startups);
+	}
+	child->pipefd = pipefd;
+	child->early = 1;
+	/* record peer address, if available */
+	if (getpeername(sockfd, sa, &addrlen) == 0 &&
+	   addr_sa_to_xaddr(sa, addrlen, &child->addr) == 0)
+		child->have_addr = 1;
+	/* format peer address string for logs */
+	if ((lport = get_local_port(sockfd)) == 0 ||
+	    (rport = get_peer_port(sockfd)) == 0) {
+		/* Not a TCP socket */
+		raddr = get_peer_ipaddr(sockfd);
+		xasprintf(&child->id, "connection from %s", raddr);
+	} else {
+		laddr = get_local_ipaddr(sockfd);
+		raddr = get_peer_ipaddr(sockfd);
+		xasprintf(&child->id, "connection from %s to %s", laddr, raddr);
+	}
+	free(laddr);
+	free(raddr);
+	if (++children_active > options.max_startups)
+		fatal_f("internal error: more children than max_startups");
+
+	return child;
+}
+
+/*
+ * Finally free a child entry. Don't call this directly.
+ */
+static void
+child_finish(struct early_child *child)
+{
+	if (children_active == 0)
+		fatal_f("internal error: children_active underflow");
+	if (child->pipefd != -1)
+		close(child->pipefd);
+	free(child->id);
+	memset(child, '\0', sizeof(*child));
+	child->pipefd = -1;
+	child->pid = -1;
+	children_active--;
+}
+
+/*
+ * Close a child's pipe. This will not stop tracking the child immediately
+ * (it will still be tracked for waitpid()) unless force_final is set, or
+ * child has already exited.
+ */
+static void
+child_close(struct early_child *child, int force_final, int quiet)
+{
+	if (!quiet)
+		debug_f("enter%s", force_final ? " (forcing)" : "");
+	if (child->pipefd != -1) {
+		close(child->pipefd);
+		child->pipefd = -1;
+	}
+	if (child->pid == -1 || force_final)
+		child_finish(child);
+}
+
+/* Record a child exit. Safe to call from signal handlers */
+static void
+child_exit(pid_t pid, int status)
+{
+	int i;
+
+	if (children == NULL || pid <= 0)
+		return;
+	for (i = 0; i < options.max_startups; i++) {
+		if (children[i].pid == pid) {
+			children[i].have_status = 1;
+			children[i].status = status;
+			break;
+		}
+	}
+}
+
+/*
+ * Reap a child entry that has exited, as previously flagged
+ * using child_exit().
+ * Handles logging of exit condition and will finalise the child if its pipe
+ * had already been closed.
+ */
+static void
+child_reap(struct early_child *child)
+{
+	LogLevel level = SYSLOG_LEVEL_DEBUG1;
+	int was_crash, penalty_type = SRCLIMIT_PENALTY_NONE;
+
+	/* Log exit information */
+	if (WIFSIGNALED(child->status)) {
+		/*
+		 * Increase logging for signals potentially associated
+		 * with serious conditions.
+		 */
+		if ((was_crash = signal_is_crash(WTERMSIG(child->status))))
+			level = SYSLOG_LEVEL_ERROR;
+		do_log2(level, "session process %ld for %s killed by "
+		    "signal %d%s", (long)child->pid, child->id,
+		    WTERMSIG(child->status), child->early ? " (early)" : "");
+		if (was_crash)
+			penalty_type = SRCLIMIT_PENALTY_CRASH;
+	} else if (!WIFEXITED(child->status)) {
+		penalty_type = SRCLIMIT_PENALTY_CRASH;
+		error("session process %ld for %s terminated abnormally, "
+		    "status=0x%x%s", (long)child->pid, child->id, child->status,
+		    child->early ? " (early)" : "");
+	} else {
+		/* Normal exit. We care about the status */
+		switch (WEXITSTATUS(child->status)) {
+		case 0:
+			debug3_f("preauth child %ld for %s completed "
+			    "normally %s", (long)child->pid, child->id,
+			    child->early ? " (early)" : "");
+			break;
+		case EXIT_LOGIN_GRACE:
+			penalty_type = SRCLIMIT_PENALTY_GRACE_EXCEEDED;
+			logit("Timeout before authentication for %s, "
+			    "pid = %ld%s", child->id, (long)child->pid,
+			    child->early ? " (early)" : "");
+			break;
+		case EXIT_CHILD_CRASH:
+			penalty_type = SRCLIMIT_PENALTY_CRASH;
+			logit("Session process %ld unpriv child crash for %s%s",
+			    (long)child->pid, child->id,
+			    child->early ? " (early)" : "");
+			break;
+		case EXIT_AUTH_ATTEMPTED:
+			penalty_type = SRCLIMIT_PENALTY_AUTHFAIL;
+			debug_f("preauth child %ld for %s exited "
+			    "after unsuccessful auth attempt %s",
+			    (long)child->pid, child->id,
+			    child->early ? " (early)" : "");
+			break;
+		default:
+			penalty_type = SRCLIMIT_PENALTY_NOAUTH;
+			debug_f("preauth child %ld for %s exited "
+			    "with status %d%s", (long)child->pid, child->id,
+			    WEXITSTATUS(child->status),
+			    child->early ? " (early)" : "");
+			break;
+		}
+	}
+	/*
+	 * XXX would be nice to have more subtlety here.
+	 *  - Different penalties
+	 *      a) authentication failures without success (e.g. brute force)
+	 *      b) login grace exceeded (penalise DoS)
+	 *      c) monitor crash (penalise exploit attempt)
+	 *      d) unpriv preauth crash (penalise exploit attempt)
+	 *  - Unpriv auth exit status/WIFSIGNALLED is not available because
+	 *    the "mm_request_receive: monitor fd closed" fatal kills the
+	 *    monitor before waitpid() can occur. It would be good to use the
+	 *    unpriv exit status to detect crashes.
+	 *
+	 * For now, just penalise (a), (b) and (c), since that is what we have
+	 * readily available. The authentication failures detection cannot
+	 * discern between failed authentication and other connection problems
+	 * until we have the unpriv exist status plumbed through (and the unpriv
+	 * child modified to use a different exit status when auth has been
+	 * attempted), but it's a start.
+	 */
+	if (child->have_addr)
+		srclimit_penalise(&child->addr, penalty_type);
+
+	child->pid = -1;
+	child->have_status = 0;
+	if (child->pipefd == -1)
+		child_finish(child);
+}
+
+/* Reap all children that have exited; called after SIGCHLD */
+static void
+child_reap_all_exited(void)
+{
+	int i;
+
+	if (children == NULL)
+		return;
+	for (i = 0; i < options.max_startups; i++) {
+		if (!children[i].have_status)
+			continue;
+		child_reap(&(children[i]));
+	}
+}
+
 static void
 close_startup_pipes(void)
 {
 	int i;
 
-	if (startup_pipes)
-		for (i = 0; i < options.max_startups; i++)
-			if (startup_pipes[i] != -1)
-				close(startup_pipes[i]);
+	if (children == NULL)
+		return;
+	for (i = 0; i < options.max_startups; i++) {
+		if (children[i].pipefd != -1)
+			child_close(&(children[i]), 1, 1);
+	}
+}
+
+/* Called after SIGINFO */
+static void
+show_info(void)
+{
+	int i;
+
+	/* XXX print listening sockets here too */
+	if (children == NULL)
+		return;
+	logit("%d active startups", children_active);
+	for (i = 0; i < options.max_startups; i++) {
+		if (children[i].pipefd == -1 && children[i].pid <= 0)
+			continue;
+		logit("child %d: fd=%d pid=%ld %s%s", i, children[i].pipefd,
+		    (long)children[i].pid, children[i].id,
+		    children[i].early ? " (early)" : "");
+	}
+	srclimit_penalty_info();
 }
 
 /*
@@ -222,6 +485,12 @@ sigterm_handler(int sig)
 	received_sigterm = sig;
 }
 
+static void
+siginfo_handler(int sig)
+{
+	received_siginfo = 1;
+}
+
 /*
  * SIGCHLD handler.  This is called whenever a child dies.  This will then
  * reap any zombies left by exited children.
@@ -233,9 +502,17 @@ main_sigchld_handler(int sig)
 	pid_t pid;
 	int status;
 
-	while ((pid = waitpid(-1, &status, WNOHANG)) > 0 ||
-	    (pid == -1 && errno == EINTR))
-		;
+	for (;;) {
+		if ((pid = waitpid(-1, &status, WNOHANG)) == 0)
+			break;
+		else if (pid == -1) {
+			if (errno == EINTR)
+				continue;
+			break;
+		}
+		child_exit(pid, status);
+		received_sigchld = 1;
+	}
 	errno = save_errno;
 }
 
@@ -268,7 +545,7 @@ should_drop_connection(int startups)
 }
 
 /*
- * Check whether connection should be accepted by MaxStartups.
+ * Check whether connection should be accepted by MaxStartups or for penalty.
  * Returns 0 if the connection is accepted. If the connection is refused,
  * returns 1 and attempts to send notification to client.
  * Logs when the MaxStartups condition is entered or exited, and periodically
@@ -278,12 +555,17 @@ static int
 drop_connection(int sock, int startups, int notify_pipe)
 {
 	char *laddr, *raddr;
-	const char msg[] = "Exceeded MaxStartups\r\n";
+	const char *reason = NULL, msg[] = "Not allowed at this time\r\n";
 	static time_t last_drop, first_drop;
 	static u_int ndropped;
 	LogLevel drop_level = SYSLOG_LEVEL_VERBOSE;
 	time_t now;
 
+	if (!srclimit_penalty_check_allow(sock, &reason)) {
+		drop_level = SYSLOG_LEVEL_INFO;
+		goto handle;
+	}
+
 	now = monotime();
 	if (!should_drop_connection(startups) &&
 	    srclimit_check_allow(sock, notify_pipe) == 1) {
@@ -313,12 +595,16 @@ drop_connection(int sock, int startups, int notify_pipe)
 	}
 	last_drop = now;
 	ndropped++;
+	reason = "past Maxstartups";
 
+ handle:
 	laddr = get_local_ipaddr(sock);
 	raddr = get_peer_ipaddr(sock);
-	do_log2(drop_level, "drop connection #%d from [%s]:%d on [%s]:%d "
-	    "past MaxStartups", startups, raddr, get_peer_port(sock),
-	    laddr, get_local_port(sock));
+	do_log2(drop_level, "drop connection #%d from [%s]:%d on [%s]:%d %s",
+	    startups,
+	    raddr, get_peer_port(sock),
+	    laddr, get_local_port(sock),
+	    reason);
 	free(laddr);
 	free(raddr);
 	/* best-effort notification to client */
@@ -521,8 +807,12 @@ server_listen(void)
 	u_int i;
 
 	/* Initialise per-source limit tracking. */
-	srclimit_init(options.max_startups, options.per_source_max_startups,
-	    options.per_source_masklen_ipv4, options.per_source_masklen_ipv6);
+	srclimit_init(options.max_startups,
+	    options.per_source_max_startups,
+	    options.per_source_masklen_ipv4,
+	    options.per_source_masklen_ipv6,
+	    &options.per_source_penalty,
+	    options.per_source_penalty_exempt);
 
 	for (i = 0; i < options.num_listen_addrs; i++) {
 		listen_on_addrs(&options.listen_addrs[i]);
@@ -548,32 +838,30 @@ server_accept_loop(int *sock_in, int *sock_out, int *newsock, int *config_s,
     int log_stderr)
 {
 	struct pollfd *pfd = NULL;
-	int i, j, ret, npfd;
-	int ostartups = -1, startups = 0, listening = 0, lameduck = 0;
+	int i, ret, npfd;
+	int oactive = -1, listening = 0, lameduck = 0;
 	int startup_p[2] = { -1 , -1 }, *startup_pollfd;
 	char c = 0;
 	struct sockaddr_storage from;
+	struct early_child *child;
 	socklen_t fromlen;
-	pid_t pid;
 	sigset_t nsigset, osigset;
 
 	/* setup fd set for accept */
 	/* pipes connected to unauthenticated child sshd processes */
-	startup_pipes = xcalloc(options.max_startups, sizeof(int));
-	startup_flags = xcalloc(options.max_startups, sizeof(int));
+	child_alloc();
 	startup_pollfd = xcalloc(options.max_startups, sizeof(int));
-	for (i = 0; i < options.max_startups; i++)
-		startup_pipes[i] = -1;
 
 	/*
 	 * Prepare signal mask that we use to block signals that might set
-	 * received_sigterm or received_sighup, so that we are guaranteed
+	 * received_sigterm/hup/chld/info, so that we are guaranteed
 	 * to immediately wake up the ppoll if a signal is received after
 	 * the flag is checked.
 	 */
 	sigemptyset(&nsigset);
 	sigaddset(&nsigset, SIGHUP);
 	sigaddset(&nsigset, SIGCHLD);
+	sigaddset(&nsigset, SIGINFO);
 	sigaddset(&nsigset, SIGTERM);
 	sigaddset(&nsigset, SIGQUIT);
 
@@ -595,11 +883,19 @@ server_accept_loop(int *sock_in, int *sock_out, int *newsock, int *config_s,
 				unlink(options.pid_file);
 			exit(received_sigterm == SIGTERM ? 0 : 255);
 		}
-		if (ostartups != startups) {
+		if (received_sigchld) {
+			child_reap_all_exited();
+			received_sigchld = 0;
+		}
+		if (received_siginfo) {
+			show_info();
+			received_siginfo = 0;
+		}
+		if (oactive != children_active) {
 			setproctitle("%s [listener] %d of %d-%d startups",
-			    listener_proctitle, startups,
+			    listener_proctitle, children_active,
 			    options.max_startups_begin, options.max_startups);
-			ostartups = startups;
+			oactive = children_active;
 		}
 		if (received_sighup) {
 			if (!lameduck) {
@@ -620,8 +916,8 @@ server_accept_loop(int *sock_in, int *sock_out, int *newsock, int *config_s,
 		npfd = num_listen_socks;
 		for (i = 0; i < options.max_startups; i++) {
 			startup_pollfd[i] = -1;
-			if (startup_pipes[i] != -1) {
-				pfd[npfd].fd = startup_pipes[i];
+			if (children[i].pipefd != -1) {
+				pfd[npfd].fd = children[i].pipefd;
 				pfd[npfd].events = POLLIN;
 				startup_pollfd[i] = npfd++;
 			}
@@ -639,34 +935,46 @@ server_accept_loop(int *sock_in, int *sock_out, int *newsock, int *config_s,
 			continue;
 
 		for (i = 0; i < options.max_startups; i++) {
-			if (startup_pipes[i] == -1 ||
+			if (children[i].pipefd == -1 ||
 			    startup_pollfd[i] == -1 ||
 			    !(pfd[startup_pollfd[i]].revents & (POLLIN|POLLHUP)))
 				continue;
-			switch (read(startup_pipes[i], &c, sizeof(c))) {
+			switch (read(children[i].pipefd, &c, sizeof(c))) {
 			case -1:
 				if (errno == EINTR || errno == EAGAIN)
 					continue;
 				if (errno != EPIPE) {
 					error_f("startup pipe %d (fd=%d): "
-					    "read %s", i, startup_pipes[i],
+					    "read %s", i, children[i].pipefd,
 					    strerror(errno));
 				}
 				/* FALLTHROUGH */
 			case 0:
-				/* child exited or completed auth */
-				close(startup_pipes[i]);
-				srclimit_done(startup_pipes[i]);
-				startup_pipes[i] = -1;
-				startups--;
-				if (startup_flags[i])
+				/* child exited preauth */
+				if (children[i].early)
 					listening--;
+				srclimit_done(children[i].pipefd);
+				child_close(&(children[i]), 0, 0);
 				break;
 			case 1:
-				/* child has finished preliminaries */
-				if (startup_flags[i]) {
+				if (children[i].early && c == '\0') {
+					/* child has finished preliminaries */
 					listening--;
-					startup_flags[i] = 0;
+					children[i].early = 0;
+					debug2_f("child %lu for %s received "
+					    "config", (long)children[i].pid,
+					    children[i].id);
+				} else if (!children[i].early && c == '\001') {
+					/* child has completed auth */
+					debug2_f("child %lu for %s auth done",
+					    (long)children[i].pid,
+					    children[i].id);
+					child_close(&(children[i]), 1, 0);
+				} else {
+					error_f("unexpected message 0x%02x "
+					    "child %ld for %s in state %d",
+					    (int)c, (long)children[i].pid,
+					    children[i].id, children[i].early);
 				}
 				break;
 			}
@@ -695,7 +1003,8 @@ server_accept_loop(int *sock_in, int *sock_out, int *newsock, int *config_s,
 				close(*newsock);
 				continue;
 			}
-			if (drop_connection(*newsock, startups, startup_p[0])) {
+			if (drop_connection(*newsock,
+			    children_active, startup_p[0])) {
 				close(*newsock);
 				close(startup_p[0]);
 				close(startup_p[1]);
@@ -712,14 +1021,6 @@ server_accept_loop(int *sock_in, int *sock_out, int *newsock, int *config_s,
 				continue;
 			}
 
-			for (j = 0; j < options.max_startups; j++)
-				if (startup_pipes[j] == -1) {
-					startup_pipes[j] = startup_p[0];
-					startups++;
-					startup_flags[j] = 1;
-					break;
-				}
-
 			/*
 			 * Got connection.  Fork a child to handle it, unless
 			 * we are in debugging mode.
@@ -737,7 +1038,6 @@ server_accept_loop(int *sock_in, int *sock_out, int *newsock, int *config_s,
 				close(startup_p[0]);
 				close(startup_p[1]);
 				startup_pipe = -1;
-				pid = getpid();
 				send_rexec_state(config_s[0], cfg);
 				close(config_s[0]);
 				free(pfd);
@@ -750,7 +1050,8 @@ server_accept_loop(int *sock_in, int *sock_out, int *newsock, int *config_s,
 			 * parent continues listening.
 			 */
 			listening++;
-			if ((pid = fork()) == 0) {
+			child = child_register(startup_p[0], *newsock);
+			if ((child->pid = fork()) == 0) {
 				/*
 				 * Child.  Close the listening and
 				 * max_startup sockets.  Start using
@@ -774,10 +1075,10 @@ server_accept_loop(int *sock_in, int *sock_out, int *newsock, int *config_s,
 			}
 
 			/* Parent.  Stay in the loop. */
-			if (pid == -1)
+			if (child->pid == -1)
 				error("fork: %.100s", strerror(errno));
 			else
-				debug("Forked child %ld.", (long)pid);
+				debug("Forked child %ld.", (long)child->pid);
 
 			close(startup_p[1]);
 
@@ -1340,6 +1641,7 @@ main(int ac, char **av)
 		ssh_signal(SIGCHLD, main_sigchld_handler);
 		ssh_signal(SIGTERM, sigterm_handler);
 		ssh_signal(SIGQUIT, sigterm_handler);
+		ssh_signal(SIGINFO, siginfo_handler);
 
 		/*
 		 * Write out the pid file after the sigterm handler
@@ -1395,6 +1697,7 @@ main(int ac, char **av)
 			fatal("dup2 startup_p: %s", strerror(errno));
 		close(startup_pipe);
 	}
+	log_redirect_stderr_to(NULL);
 	closefrom(REEXEC_MIN_FREE_FD);
 
 	ssh_signal(SIGHUP, SIG_IGN); /* avoid reset to SIG_DFL */
diff --git a/usr.bin/ssh/sshd_config.5 b/usr.bin/ssh/sshd_config.5
index 93afc3eeb..92c183466 100644
--- a/usr.bin/ssh/sshd_config.5
+++ b/usr.bin/ssh/sshd_config.5
@@ -33,8 +33,8 @@
 .\" (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE OF
 .\" THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
 .\"
-.\" $OpenBSD: sshd_config.5,v 1.355 2024/02/21 06:17:29 djm Exp $
-.Dd $Mdocdate: February 21 2024 $
+.\" $OpenBSD: sshd_config.5,v 1.358 2024/06/06 21:14:49 jmc Exp $
+.Dd $Mdocdate: June 6 2024 $
 .Dt SSHD_CONFIG 5
 .Os
 .Sh NAME
@@ -1558,6 +1558,69 @@ Values for IPv4 and optionally IPv6 may be specified, separated by a colon.
 The default is
 .Cm 32:128 ,
 which means each address is considered individually.
+.It Cm PerSourcePenalties
+Controls penalties for various conditions that may represent attacks on
+.Xr sshd 8 .
+If a penalty is enforced against a client then its source address and any
+others in the
+.Cm PerSourceNetBlockSize
+will be refused connection for a period.
+A penalty doesn't affect concurrent connections in progress, but multiple
+penalties from the same source from concurrent connections will accumulate
+up to a maximum.
+Conversely, penalties are not applied until a minimum threshold time has been
+accumulated.
+Penalties are off by default but may be enabled using default settings using the
+.Cm yes
+keyword or by specifying one or more of the keywords below.
+.Pp
+Penalties are controlled using the following keywords, all of which accept
+arguments, e.g.\&
+.Qq crash:2m .
+.Bl -tag -width Ds
+.It Cm crash:duration
+Specifies how long to refuse clients that cause a crash of
+.Xr sshd 8 .
+.It Cm authfail:duration
+Specifies how long to refuse clients that disconnect after making one or more
+unsuccessful authentication attempts.
+.It Cm noauth:duration
+Specifies how long to refuse clients that disconnect without attempting
+authentication.
+This timeout should be used cautiously otherwise it may penalise legitimate
+scanning tools such as
+.Xr ssh-keyscan 1 .
+.It Cm grace-exceeded:duration
+Specifies how long to refuse clients that fail to authenticate after
+.Cm LoginGraceTime .
+.It Cm max:duration
+Specifies the maximum time a particular source address range will be refused
+access for.
+Repeated penalties will accumulate up to this maximum.
+.It Cm min:duration
+Specifies the minimum penalty that must accrue before enforcement begins.
+.It Cm max-sources:number
+Specifies the maximum number of penalise client address ranges to track.
+.It Cm overflow:mode
+Controls how the server behaves when
+.Cm max-sources
+is exceeded.
+There are two operating modes:
+.Cm deny-all ,
+which denies all incoming connections other than those exempted via
+.Cm PerSourcePenaltyExemptList
+until a penalty expires, and
+.Cm permissive ,
+which allows new connections by removing existing penalties early.
+.El
+.It Cm PerSourcePenaltyExemptList
+Specifies a comma-separated list of addresses to exempt from penalties.
+This list may contain wildcards and CIDR address/masklen ranges.
+Note that the mask length provided must be consistent with the address -
+it is an error to specify a mask length that is too long for the address
+or one with bits set in this host portion of the address.
+For example, 192.0.2.0/33 and 192.0.2.0/8, respectively.
+The default is not to exempt any addresses.
 .It Cm PidFile
 Specifies the file that contains the process ID of the
 SSH daemon, or
diff --git a/usr.sbin/dhcp6leasectl/dhcp6leasectl.8 b/usr.sbin/dhcp6leasectl/dhcp6leasectl.8
index f5ebd9a08..bb07409bc 100644
--- a/usr.sbin/dhcp6leasectl/dhcp6leasectl.8
+++ b/usr.sbin/dhcp6leasectl/dhcp6leasectl.8
@@ -1,4 +1,4 @@
-.\"	$OpenBSD: dhcp6leasectl.8,v 1.1 2024/06/06 15:16:57 florian Exp $
+.\"	$OpenBSD: dhcp6leasectl.8,v 1.2 2024/06/07 17:38:22 jmc Exp $
 .\"
 .\" Copyright (c) 2021 Florian Obser <florian@openbsd.org>
 .\" Copyright (c) 2016 Kenneth R Westerback <kwesterback@gmail.com>
@@ -16,7 +16,7 @@
 .\" ACTION OF CONTRACT, NEGLIGENCE OR OTHER TORTIOUS ACTION, ARISING OUT OF
 .\" OR IN CONNECTION WITH THE USE OR PERFORMANCE OF THIS SOFTWARE.
 .\"
-.Dd $Mdocdate: June 6 2024 $
+.Dd $Mdocdate: June 7 2024 $
 .Dt DHCP6LEASECTL 8
 .Os
 .Sh NAME
@@ -52,9 +52,7 @@ to communicate with
 Specify the maximum number of seconds to wait for
 .Ar interface
 to be configured.
-The default is to wait 10 seconds unless
-.Fl l
-is specified.
+The default is 10 seconds.
 .El
 .Sh FILES
 .Bl -tag -width "/dev/dhcp6leased.sockXX" -compact
diff --git a/usr.sbin/dhcpleasectl/dhcpleasectl.8 b/usr.sbin/dhcpleasectl/dhcpleasectl.8
index adf87157b..e346dd06b 100644
--- a/usr.sbin/dhcpleasectl/dhcpleasectl.8
+++ b/usr.sbin/dhcpleasectl/dhcpleasectl.8
@@ -1,4 +1,4 @@
-.\"	$OpenBSD: dhcpleasectl.8,v 1.5 2023/03/02 17:09:54 jmc Exp $
+.\"	$OpenBSD: dhcpleasectl.8,v 1.6 2024/06/07 17:38:22 jmc Exp $
 .\"
 .\" Copyright (c) 2021 Florian Obser <florian@openbsd.org>
 .\" Copyright (c) 2016 Kenneth R Westerback <kwesterback@gmail.com>
@@ -16,7 +16,7 @@
 .\" ACTION OF CONTRACT, NEGLIGENCE OR OTHER TORTIOUS ACTION, ARISING OUT OF
 .\" OR IN CONNECTION WITH THE USE OR PERFORMANCE OF THIS SOFTWARE.
 .\"
-.Dd $Mdocdate: March 2 2023 $
+.Dd $Mdocdate: June 7 2024 $
 .Dt DHCPLEASECTL 8
 .Os
 .Sh NAME
@@ -52,9 +52,7 @@ to communicate with
 Specify the maximum number of seconds to wait for
 .Ar interface
 to be configured.
-The default is to wait 10 seconds unless
-.Fl l
-is specified.
+The default is 10 seconds.
 .El
 .Sh FILES
 .Bl -tag -width "/dev/dhcpleased.sockXX" -compact
diff --git a/usr.sbin/rpki-client/cert.c b/usr.sbin/rpki-client/cert.c
index 0b0794412..ba29b7e48 100644
--- a/usr.sbin/rpki-client/cert.c
+++ b/usr.sbin/rpki-client/cert.c
@@ -1,4 +1,4 @@
-/*	$OpenBSD: cert.c,v 1.140 2024/06/06 12:38:02 tb Exp $ */
+/*	$OpenBSD: cert.c,v 1.144 2024/06/08 13:33:49 tb Exp $ */
 /*
  * Copyright (c) 2022 Theo Buehler <tb@openbsd.org>
  * Copyright (c) 2021 Job Snijders <job@openbsd.org>
@@ -744,6 +744,15 @@ cert_parse_ee_cert(const char *fn, int talid, X509 *x)
 	if (!cert_check_subject_and_issuer(fn, x))
 		goto out;
 
+	if (!x509_cache_extensions(x, fn))
+		goto out;
+
+	if ((cert->purpose = x509_get_purpose(x, fn)) != CERT_PURPOSE_EE) {
+		warnx("%s: expected EE cert, got %s", fn,
+		    purpose2str(cert->purpose));
+		goto out;
+	}
+
 	if (X509_get_key_usage(x) != KU_DIGITAL_SIGNATURE) {
 		warnx("%s: RFC 6487 section 4.8.4: KU must be digitalSignature",
 		    fn);
@@ -827,11 +836,8 @@ cert_parse_pre(const char *fn, const unsigned char *der, size_t len)
 		goto out;
 	}
 
-	/* Cache X509v3 extensions, see X509_check_ca(3). */
-	if (X509_check_purpose(x, -1, -1) <= 0) {
-		warnx("%s: could not cache X509v3 extensions", fn);
+	if (!x509_cache_extensions(x, fn))
 		goto out;
-	}
 
 	if (X509_get_version(x) != 2) {
 		warnx("%s: RFC 6487 4.1: X.509 version must be v3", fn);
@@ -957,11 +963,12 @@ cert_parse_pre(const char *fn, const unsigned char *der, size_t len)
 		goto out;
 	if (!x509_get_notafter(x, fn, &cert->notafter))
 		goto out;
-	cert->purpose = x509_get_purpose(x, fn);
 
 	/* Validation on required fields. */
-
+	cert->purpose = x509_get_purpose(x, fn);
 	switch (cert->purpose) {
+	case CERT_PURPOSE_TA:
+		/* XXX - caller should indicate if it expects TA or CA cert */
 	case CERT_PURPOSE_CA:
 		if ((pkey = X509_get0_pubkey(x)) == NULL) {
 			warnx("%s: X509_get0_pubkey failed", fn);
@@ -1015,6 +1022,9 @@ cert_parse_pre(const char *fn, const unsigned char *der, size_t len)
 			goto out;
 		}
 		break;
+	case CERT_PURPOSE_EE:
+		warn("%s: unexpected EE cert", fn);
+		goto out;
 	default:
 		warnx("%s: x509_get_purpose failed in %s", fn, __func__);
 		goto out;
@@ -1117,8 +1127,9 @@ ta_parse(const char *fn, struct cert *p, const unsigned char *pkey,
 		    "trust anchor may not specify CRL resource", fn);
 		goto badcert;
 	}
-	if (p->purpose == CERT_PURPOSE_BGPSEC_ROUTER) {
-		warnx("%s: BGPsec cert cannot be a trust anchor", fn);
+	if (p->purpose != CERT_PURPOSE_TA) {
+		warnx("%s: expected trust anchor purpose, got %s", fn,
+		    purpose2str(p->purpose));
 		goto badcert;
 	}
 	/*
diff --git a/usr.sbin/rpki-client/cms.c b/usr.sbin/rpki-client/cms.c
index 5ef647e15..0394680e5 100644
--- a/usr.sbin/rpki-client/cms.c
+++ b/usr.sbin/rpki-client/cms.c
@@ -1,4 +1,4 @@
-/*	$OpenBSD: cms.c,v 1.45 2024/05/24 12:57:20 tb Exp $ */
+/*	$OpenBSD: cms.c,v 1.46 2024/06/08 13:28:35 tb Exp $ */
 /*
  * Copyright (c) 2019 Kristaps Dzonsons <kristaps@bsd.lv>
  *
@@ -324,11 +324,8 @@ cms_parse_validate_internal(X509 **xp, const char *fn, const unsigned char *der,
 		goto out;
 	}
 
-	/* Cache X509v3 extensions, see X509_check_ca(3). */
-	if (X509_check_purpose(*xp, -1, -1) <= 0) {
-		warnx("%s: could not cache X509v3 extensions", fn);
+	if (!x509_cache_extensions(*xp, fn))
 		goto out;
-	}
 
 	if (!x509_get_notafter(*xp, fn, &notafter))
 		goto out;
diff --git a/usr.sbin/rpki-client/extern.h b/usr.sbin/rpki-client/extern.h
index 601d7d289..b811b485e 100644
--- a/usr.sbin/rpki-client/extern.h
+++ b/usr.sbin/rpki-client/extern.h
@@ -1,4 +1,4 @@
-/*	$OpenBSD: extern.h,v 1.221 2024/06/04 04:17:18 tb Exp $ */
+/*	$OpenBSD: extern.h,v 1.224 2024/06/08 13:30:35 tb Exp $ */
 /*
  * Copyright (c) 2019 Kristaps Dzonsons <kristaps@bsd.lv>
  *
@@ -107,8 +107,10 @@ struct cert_ip {
 
 enum cert_purpose {
 	CERT_PURPOSE_INVALID,
+	CERT_PURPOSE_TA,
 	CERT_PURPOSE_CA,
-	CERT_PURPOSE_BGPSEC_ROUTER
+	CERT_PURPOSE_EE,
+	CERT_PURPOSE_BGPSEC_ROUTER,
 };
 
 /*
@@ -901,6 +903,7 @@ struct ibuf	*io_buf_recvfd(int, struct ibuf **);
 /* X509 helpers. */
 
 void		 x509_init_oid(void);
+int		 x509_cache_extensions(X509 *, const char *);
 int		 x509_get_aia(X509 *, const char *, char **);
 int		 x509_get_aki(X509 *, const char *, char **);
 int		 x509_get_sia(X509 *, const char *, char **);
@@ -922,6 +925,7 @@ time_t		 x509_find_expires(time_t, struct auth *, struct crl_tree *);
 
 /* printers */
 char		*nid2str(int);
+const char	*purpose2str(enum cert_purpose);
 char		*time2str(time_t);
 void		 x509_print(const X509 *);
 void		 tal_print(const struct tal *);
diff --git a/usr.sbin/rpki-client/filemode.c b/usr.sbin/rpki-client/filemode.c
index 630b633b2..689817c7c 100644
--- a/usr.sbin/rpki-client/filemode.c
+++ b/usr.sbin/rpki-client/filemode.c
@@ -1,4 +1,4 @@
-/*	$OpenBSD: filemode.c,v 1.43 2024/06/06 07:19:10 tb Exp $ */
+/*	$OpenBSD: filemode.c,v 1.45 2024/06/08 13:34:59 tb Exp $ */
 /*
  * Copyright (c) 2019 Claudio Jeker <claudio@openbsd.org>
  * Copyright (c) 2019 Kristaps Dzonsons <kristaps@bsd.lv>
@@ -157,7 +157,8 @@ parse_load_cert(char *uri)
 	if (cert == NULL)
 		goto done;
 	if (cert->purpose != CERT_PURPOSE_CA) {
-		warnx("AIA reference to bgpsec cert %s", uri);
+		warnx("AIA reference to %s in %s",
+		    purpose2str(cert->purpose), uri);
 		goto done;
 	}
 	/* try to load the CRL of this cert */
@@ -403,7 +404,7 @@ proc_parser_file(char *file, unsigned char *buf, size_t len)
 		cert = cert_parse_pre(file, buf, len);
 		if (cert == NULL)
 			break;
-		is_ta = X509_get_extension_flags(cert->x509) & EXFLAG_SS;
+		is_ta = (cert->purpose == CERT_PURPOSE_TA);
 		if (!is_ta)
 			cert = cert_parse(file, cert);
 		if (cert == NULL)
diff --git a/usr.sbin/rpki-client/main.c b/usr.sbin/rpki-client/main.c
index ffd010013..342548ce3 100644
--- a/usr.sbin/rpki-client/main.c
+++ b/usr.sbin/rpki-client/main.c
@@ -1,4 +1,4 @@
-/*	$OpenBSD: main.c,v 1.258 2024/05/20 15:51:43 claudio Exp $ */
+/*	$OpenBSD: main.c,v 1.260 2024/06/08 13:31:38 tb Exp $ */
 /*
  * Copyright (c) 2021 Claudio Jeker <claudio@openbsd.org>
  * Copyright (c) 2019 Kristaps Dzonsons <kristaps@bsd.lv>
@@ -481,7 +481,7 @@ queue_add_from_tal(struct tal *tal)
 	/* steal the pkey from the tal structure */
 	data = tal->pkey;
 	tal->pkey = NULL;
-	entityq_add(NULL, nfile, RTYPE_CER, DIR_VALID, repo, data,
+	entityq_add(NULL, nfile, RTYPE_CER, DIR_UNKNOWN, repo, data,
 	    tal->pkeysz, tal->id, tal->id, NULL);
 }
 
@@ -618,6 +618,7 @@ entity_process(struct ibuf *b, struct stats *st, struct vrp_tree *tree,
 		}
 		cert = cert_read(b);
 		switch (cert->purpose) {
+		case CERT_PURPOSE_TA:
 		case CERT_PURPOSE_CA:
 			queue_add_from_cert(cert);
 			break;
@@ -626,7 +627,7 @@ entity_process(struct ibuf *b, struct stats *st, struct vrp_tree *tree,
 			repo_stat_inc(rp, talid, type, STYPE_BGPSEC);
 			break;
 		default:
-			errx(1, "unexpected cert purpose received");
+			errx(1, "unexpected %s", purpose2str(cert->purpose));
 			break;
 		}
 		cert_free(cert);
diff --git a/usr.sbin/rpki-client/parser.c b/usr.sbin/rpki-client/parser.c
index 98b1eac70..10f7975a9 100644
--- a/usr.sbin/rpki-client/parser.c
+++ b/usr.sbin/rpki-client/parser.c
@@ -1,4 +1,4 @@
-/*	$OpenBSD: parser.c,v 1.136 2024/05/20 15:51:43 claudio Exp $ */
+/*	$OpenBSD: parser.c,v 1.139 2024/06/07 13:24:35 tb Exp $ */
 /*
  * Copyright (c) 2019 Claudio Jeker <claudio@openbsd.org>
  * Copyright (c) 2019 Kristaps Dzonsons <kristaps@bsd.lv>
@@ -570,35 +570,101 @@ proc_parser_cert(char *file, const unsigned char *der, size_t len,
 	return cert;
 }
 
-/*
- * Root certificates come from TALs (has a pkey and is self-signed).
- * Parse the certificate, ensure that its public key matches the
- * known public key from the TAL, and then validate the RPKI
- * content.
- *
- * This returns a certificate (which must not be freed) or NULL on
- * parse failure.
- */
-static struct cert *
-proc_parser_root_cert(char *file, const unsigned char *der, size_t len,
-    unsigned char *pkey, size_t pkeysz, int talid)
+static int
+proc_parser_ta_cmp(const struct cert *cert1, const struct cert *cert2)
 {
-	struct cert		*cert;
-
-	/* Extract certificate data. */
-
-	cert = cert_parse_pre(file, der, len);
-	cert = ta_parse(file, cert, pkey, pkeysz);
-	if (cert == NULL)
-		return NULL;
-	cert->talid = talid;
+	if (cert1 == NULL)
+		return -1;
+	if (cert2 == NULL)
+		return 1;
 
 	/*
-	 * Add valid roots to the RPKI auth tree.
+	 * The standards don't specify tiebreakers. While RFC 6487 and other
+	 * sources advise against backdating, it's explicitly allowed and some
+	 * TAs do. Some TAs have also re-issued with new dates and old
+	 * serialNumber.
+	 * Our tiebreaker logic: a more recent notBefore is taken to mean a
+	 * more recent issuance, and thus preferable. Given equal notBefore
+	 * values, prefer the TA cert with the narrower validity window. This
+	 * hopefully encourages TA operators to reduce egregiously long TA
+	 * validity periods.
 	 */
-	auth_insert(file, &auths, cert, NULL);
 
-	return cert;
+	if (cert1->notbefore < cert2->notbefore)
+		return -1;
+	if (cert1->notbefore > cert2->notbefore)
+		return 1;
+
+	if (cert1->notafter > cert2->notafter)
+		return -1;
+	if (cert1->notafter < cert2->notafter)
+		return 1;
+
+	/*
+	 * Both certs are valid from our perspective. If anything changed,
+	 * prefer the freshly-fetched one. We rely on cert_parse_pre() having
+	 * cached the extensions and thus libcrypto has already computed the
+	 * certs' hashes (SHA-1 for OpenSSL, SHA-512 for LibreSSL). The below
+	 * compares them.
+	 */
+
+	return X509_cmp(cert1->x509, cert2->x509) != 0;
+}
+
+/*
+ * Root certificates come from TALs. Inspect and validate both options and
+ * compare the two. The cert in out_cert must not be freed. Returns the file
+ * name of the chosen TA.
+ */
+static char *
+proc_parser_root_cert(struct entity *entp, struct cert **out_cert)
+{
+	struct cert		*cert1 = NULL, *cert2 = NULL;
+	char			*file1 = NULL, *file2 = NULL;
+	unsigned char		*der = NULL, *pkey = entp->data;
+	size_t			 der_len = 0, pkeysz = entp->datasz;
+	int			 cmp;
+
+	*out_cert = NULL;
+
+	file2 = parse_filepath(entp->repoid, entp->path, entp->file, DIR_VALID);
+	der = load_file(file2, &der_len);
+	cert2 = cert_parse_pre(file2, der, der_len);
+	free(der);
+	cert2 = ta_parse(file2, cert2, pkey, pkeysz);
+
+	if (!noop) {
+		file1 = parse_filepath(entp->repoid, entp->path, entp->file,
+		    DIR_TEMP);
+		der = load_file(file1, &der_len);
+		cert1 = cert_parse_pre(file1, der, der_len);
+		free(der);
+		cert1 = ta_parse(file1, cert1, pkey, pkeysz);
+	}
+
+	if ((cmp = proc_parser_ta_cmp(cert1, cert2)) > 0) {
+		cert_free(cert2);
+		free(file2);
+
+		cert1->talid = entp->talid;
+		auth_insert(file1, &auths, cert1, NULL);
+
+		*out_cert = cert1;
+		return file1;
+	} else {
+		if (cmp < 0 && cert1 != NULL && cert2 != NULL)
+			warnx("%s: cached TA is newer", entp->file);
+		cert_free(cert1);
+		free(file1);
+
+		if (cert2 != 0) {
+			cert2->talid = entp->talid;
+			auth_insert(file2, &auths, cert2, NULL);
+		}
+
+		*out_cert = cert2;
+		return file2;
+	}
 }
 
 /*
@@ -784,14 +850,13 @@ parse_entity(struct entityq *q, struct msgbuf *msgq)
 			tal_free(tal);
 			break;
 		case RTYPE_CER:
-			file = parse_load_file(entp, &f, &flen);
-			io_str_buffer(b, file);
-			if (entp->data != NULL)
-				cert = proc_parser_root_cert(file,
-				    f, flen, entp->data, entp->datasz,
-				    entp->talid);
-			else
+			if (entp->data != NULL) {
+				file = proc_parser_root_cert(entp, &cert);
+			} else {
+				file = parse_load_file(entp, &f, &flen);
 				cert = proc_parser_cert(file, f, flen, entp);
+			}
+			io_str_buffer(b, file);
 			if (cert != NULL)
 				mtime = cert->notbefore;
 			io_simple_buffer(b, &mtime, sizeof(mtime));
diff --git a/usr.sbin/rpki-client/print.c b/usr.sbin/rpki-client/print.c
index 3fbf35537..6738cbc7a 100644
--- a/usr.sbin/rpki-client/print.c
+++ b/usr.sbin/rpki-client/print.c
@@ -1,4 +1,4 @@
-/*	$OpenBSD: print.c,v 1.54 2024/06/06 05:57:36 tb Exp $ */
+/*	$OpenBSD: print.c,v 1.55 2024/06/08 13:30:35 tb Exp $ */
 /*
  * Copyright (c) 2021 Claudio Jeker <claudio@openbsd.org>
  * Copyright (c) 2019 Kristaps Dzonsons <kristaps@bsd.lv>
@@ -65,6 +65,25 @@ nid2str(int nid)
 	return buf;
 }
 
+const char *
+purpose2str(enum cert_purpose purpose)
+{
+	switch (purpose) {
+	case CERT_PURPOSE_INVALID:
+		return "invalid cert";
+	case CERT_PURPOSE_TA:
+		return "TA cert";
+	case CERT_PURPOSE_CA:
+		return "CA cert";
+	case CERT_PURPOSE_EE:
+		return "EE cert";
+	case CERT_PURPOSE_BGPSEC_ROUTER:
+		return "BGPsec Router cert";
+	default:
+		return "unknown certificate purpose";
+	}
+}
+
 char *
 time2str(time_t t)
 {
diff --git a/usr.sbin/rpki-client/repo.c b/usr.sbin/rpki-client/repo.c
index 15ff513f0..0dd47cd0b 100644
--- a/usr.sbin/rpki-client/repo.c
+++ b/usr.sbin/rpki-client/repo.c
@@ -1,4 +1,4 @@
-/*	$OpenBSD: repo.c,v 1.59 2024/05/30 12:33:15 claudio Exp $ */
+/*	$OpenBSD: repo.c,v 1.60 2024/06/07 08:22:53 claudio Exp $ */
 /*
  * Copyright (c) 2021 Claudio Jeker <claudio@openbsd.org>
  * Copyright (c) 2019 Kristaps Dzonsons <kristaps@bsd.lv>
@@ -78,7 +78,6 @@ struct tarepo {
 	SLIST_ENTRY(tarepo)	 entry;
 	char			*descr;
 	char			*basedir;
-	char			*temp;
 	char			**uri;
 	size_t			 urisz;
 	size_t			 uriidx;
@@ -322,7 +321,7 @@ repo_done(const void *vp, int ok)
  * If temp is set add Xs for mkostemp.
  */
 static char *
-ta_filename(const struct tarepo *tr, int temp)
+ta_filename(const struct tarepo *tr)
 {
 	const char *file;
 	char *nfile;
@@ -331,8 +330,7 @@ ta_filename(const struct tarepo *tr, int temp)
 	file = strrchr(tr->uri[0], '/');
 	assert(file);
 
-	if (asprintf(&nfile, "%s%s%s", tr->basedir, file,
-	    temp ? ".XXXXXXXX" : "") == -1)
+	if (asprintf(&nfile, "%s%s", tr->basedir, file) == -1)
 		err(1, NULL);
 
 	return nfile;
@@ -367,18 +365,21 @@ ta_fetch(struct tarepo *tr)
 		 */
 		rsync_fetch(tr->id, tr->uri[tr->uriidx], tr->basedir, NULL);
 	} else {
+		char *temp;
 		int fd;
 
-		tr->temp = ta_filename(tr, 1);
-		fd = mkostemp(tr->temp, O_CLOEXEC);
+		temp = ta_filename(tr);
+		fd = open(temp,
+		    O_WRONLY | O_CREAT | O_TRUNC | O_NOFOLLOW | O_CLOEXEC,
+		    S_IRUSR | S_IWUSR | S_IRGRP | S_IROTH);
 		if (fd == -1) {
-			warn("mkostemp: %s", tr->temp);
+			warn("open: %s", temp);
+			free(temp);
 			http_finish(tr->id, HTTP_FAILED, NULL);
 			return;
 		}
-		if (fchmod(fd, S_IRUSR | S_IWUSR | S_IRGRP | S_IROTH) == -1)
-			warn("fchmod: %s", tr->temp);
 
+		free(temp);
 		http_fetch(tr->id, tr->uri[tr->uriidx], NULL, fd);
 	}
 }
@@ -392,12 +393,21 @@ ta_get(struct tal *tal)
 
 	if ((tr = calloc(1, sizeof(*tr))) == NULL)
 		err(1, NULL);
+
 	tr->id = ++repoid;
 	SLIST_INSERT_HEAD(&tarepos, tr, entry);
 
 	if ((tr->descr = strdup(tal->descr)) == NULL)
 		err(1, NULL);
-	tr->basedir = repo_dir(tal->descr, "ta", 0);
+	tr->basedir = repo_dir(tal->descr, ".ta", 0);
+
+	/* create base directory */
+	if (mkpath(tr->basedir) == -1) {
+		warn("mkpath %s", tr->basedir);
+		tr->state = REPO_FAILED;
+		repo_done(tr, 0);
+		return tr;
+	}
 
 	/* steal URI information from TAL */
 	tr->urisz = tal->urisz;
@@ -430,7 +440,6 @@ ta_free(void)
 		SLIST_REMOVE_HEAD(&tarepos, entry);
 		free(tr->descr);
 		free(tr->basedir);
-		free(tr->temp);
 		free(tr->uri);
 		free(tr);
 	}
@@ -1070,20 +1079,12 @@ http_finish(unsigned int id, enum http_result res, const char *last_mod)
 
 	/* Move downloaded TA file into place, or unlink on failure. */
 	if (res == HTTP_OK) {
-		char *file;
-
-		file = ta_filename(tr, 0);
-		if (rename(tr->temp, file) == -1)
-			warn("rename to %s", file);
-		free(file);
-
 		logx("ta/%s: loaded from network", tr->descr);
 		tr->state = REPO_DONE;
 		stats.http_repos++;
 		repo_done(tr, 1);
 	} else {
-		if (unlink(tr->temp) == -1 && errno != ENOENT)
-			warn("unlink %s", tr->temp);
+		remove_contents(tr->basedir);
 
 		tr->uriidx++;
 		warnx("ta/%s: load from network failed", tr->descr);
@@ -1632,15 +1633,19 @@ repo_move_valid(struct filepath_tree *tree)
 	struct filepath *fp, *nfp;
 	size_t rsyncsz = strlen(".rsync/");
 	size_t rrdpsz = strlen(".rrdp/");
+	size_t tasz = strlen(".ta/");
 	char *fn, *base;
 
 	RB_FOREACH_SAFE(fp, filepath_tree, tree, nfp) {
 		if (strncmp(fp->file, ".rsync/", rsyncsz) != 0 &&
-		    strncmp(fp->file, ".rrdp/", rrdpsz) != 0)
+		    strncmp(fp->file, ".rrdp/", rrdpsz) != 0 &&
+		    strncmp(fp->file, ".ta/", tasz) != 0)
 			continue; /* not a temporary file path */
 
 		if (strncmp(fp->file, ".rsync/", rsyncsz) == 0) {
 			fn = fp->file + rsyncsz;
+		} else if (strncmp(fp->file, ".ta/", tasz) == 0) {
+			fn = fp->file + 1; /* just skip the '.' */
 		} else {
 			base = strchr(fp->file + rrdpsz, '/');
 			assert(base != NULL);
@@ -1690,8 +1695,8 @@ repo_move_valid(struct filepath_tree *tree)
 }
 
 struct fts_state {
-	enum { BASE_DIR, RSYNC_DIR, RRDP_DIR }	type;
-	struct repo				*rp;
+	enum { BASE_DIR, RSYNC_DIR, TA_DIR, RRDP_DIR }	type;
+	struct repo					*rp;
 } fts_state;
 
 static const struct rrdprepo *
@@ -1768,8 +1773,9 @@ repo_cleanup_entry(FTSENT *e, struct filepath_tree *tree, int cachefd)
  unlink:
 			if (unlink(e->fts_accpath) == -1) {
 				warn("unlink %s", path);
-			} else if (fts_state.type == RSYNC_DIR) {
-				/* no need to keep rsync files */
+			} else if (fts_state.type == RSYNC_DIR ||
+			     fts_state.type == TA_DIR) {
+				/* no need to keep rsync or ta files */
 				if (verbose > 1)
 					logx("deleted superfluous %s", path);
 				if (fts_state.rp != NULL)
@@ -1792,9 +1798,11 @@ repo_cleanup_entry(FTSENT *e, struct filepath_tree *tree, int cachefd)
 			fts_state.rp = NULL;
 		}
 		if (e->fts_level == 1) {
-			/* rpki.example.org or .rrdp / .rsync */
+			/* rpki.example.org or .rrdp / .rsync / .ta */
 			if (strcmp(".rsync", e->fts_name) == 0)
 				fts_state.type = RSYNC_DIR;
+			else if (strcmp(".ta", e->fts_name) == 0)
+				fts_state.type = TA_DIR;
 			else if (strcmp(".rrdp", e->fts_name) == 0)
 				fts_state.type = RRDP_DIR;
 			else
@@ -1805,6 +1813,8 @@ repo_cleanup_entry(FTSENT *e, struct filepath_tree *tree, int cachefd)
 			/* rpki.example.org/repository or .rrdp/hashdir */
 			if (fts_state.type == BASE_DIR)
 				fts_state.rp = repo_bypath(path);
+			if (fts_state.type == TA_DIR)
+				fts_state.rp = repo_bypath(path + 1);
 			/*
 			 * special handling for rrdp directories,
 			 * clear them if they are not used anymore but
@@ -1826,7 +1836,8 @@ repo_cleanup_entry(FTSENT *e, struct filepath_tree *tree, int cachefd)
 			/* do not remove .rsync and .rrdp */
 			fts_state.rp = NULL;
 			if (fts_state.type == RRDP_DIR ||
-			    fts_state.type == RSYNC_DIR)
+			    fts_state.type == RSYNC_DIR ||
+			    fts_state.type == TA_DIR)
 				break;
 		}
 
diff --git a/usr.sbin/rpki-client/x509.c b/usr.sbin/rpki-client/x509.c
index 4082242e5..c264b7327 100644
--- a/usr.sbin/rpki-client/x509.c
+++ b/usr.sbin/rpki-client/x509.c
@@ -1,4 +1,4 @@
-/*	$OpenBSD: x509.c,v 1.93 2024/06/04 14:17:24 tb Exp $ */
+/*	$OpenBSD: x509.c,v 1.97 2024/06/08 13:32:30 tb Exp $ */
 /*
  * Copyright (c) 2022 Theo Buehler <tb@openbsd.org>
  * Copyright (c) 2021 Claudio Jeker <claudio@openbsd.org>
@@ -132,6 +132,26 @@ x509_init_oid(void)
 	}
 }
 
+/*
+ * A number of critical OpenSSL API functions can't properly indicate failure
+ * and are unreliable if the extensions aren't already cached. An old trick is
+ * to cache the extensions using an error-checked call to X509_check_purpose()
+ * with a purpose of -1. This way functions such as X509_check_ca(), X509_cmp(),
+ * X509_get_key_usage(), X509_get_extended_key_usage() won't lie.
+ *
+ * Should be called right after deserialization and is essentially free to call
+ * multiple times.
+ */
+int
+x509_cache_extensions(X509 *x509, const char *fn)
+{
+	if (X509_check_purpose(x509, -1, 0) <= 0) {
+		warnx("%s: could not cache X509v3 extensions", fn);
+		return 0;
+	}
+	return 1;
+}
+
 /*
  * Parse X509v3 authority key identifier (AKI), RFC 6487 sec. 4.8.3.
  * Returns the AKI or NULL if it could not be parsed.
@@ -246,18 +266,34 @@ x509_get_ski(X509 *x, const char *fn, char **ski)
 }
 
 /*
- * Check the certificate's purpose: CA or BGPsec Router.
- * Return a member of enum cert_purpose.
+ * Check the cert's purpose: the cA bit in basic constraints distinguishes
+ * between TA/CA and EE/BGPsec router. TAs are self-signed, CAs not self-issued,
+ * EEs have no extended key usage, BGPsec router have id-kp-bgpsec-router OID.
  */
 enum cert_purpose
 x509_get_purpose(X509 *x, const char *fn)
 {
 	BASIC_CONSTRAINTS		*bc = NULL;
 	EXTENDED_KEY_USAGE		*eku = NULL;
-	int				 crit;
+	int				 crit, ext_flags, is_ca;
 	enum cert_purpose		 purpose = CERT_PURPOSE_INVALID;
 
-	if (X509_check_ca(x) == 1) {
+	if (!x509_cache_extensions(x, fn))
+		goto out;
+
+	ext_flags = X509_get_extension_flags(x);
+
+	/* This weird API can return 0, 1, 2, 4, 5 but can't error... */
+	if ((is_ca = X509_check_ca(x)) > 1) {
+		if (is_ca == 4)
+			warnx("%s: RFC 6487: sections 4.8.1 and 4.8.4: "
+			    "no basic constraints, but keyCertSign set", fn);
+		else
+			warnx("%s: unexpected legacy certificate", fn);
+		goto out;
+	}
+
+	if (is_ca) {
 		bc = X509_get_ext_d2i(x, NID_basic_constraints, &crit, NULL);
 		if (bc == NULL) {
 			if (crit != -1)
@@ -278,27 +314,50 @@ x509_get_purpose(X509 *x, const char *fn)
 			    "Constraint must be absent", fn);
 			goto out;
 		}
-		purpose = CERT_PURPOSE_CA;
+		/*
+		 * EXFLAG_SI means that issuer and subject are identical.
+		 * EXFLAG_SS is SI plus the AKI is absent or matches the SKI.
+		 * Thus, exactly the trust anchors should have EXFLAG_SS set
+		 * and we should never see EXFLAG_SI without EXFLAG_SS.
+		 */
+		if ((ext_flags & EXFLAG_SS) != 0)
+			purpose = CERT_PURPOSE_TA;
+		else if ((ext_flags & EXFLAG_SI) == 0)
+			purpose = CERT_PURPOSE_CA;
+		else
+			warnx("%s: RFC 6487, section 4.8.3: "
+			    "self-issued cert with AKI-SKI mismatch", fn);
 		goto out;
 	}
 
-	if (X509_get_extension_flags(x) & EXFLAG_BCONS) {
+	if ((ext_flags & EXFLAG_BCONS) != 0) {
 		warnx("%s: Basic Constraints ext in non-CA cert", fn);
 		goto out;
 	}
 
+	/*
+	 * EKU is only defined for BGPsec Router certs and must be absent from
+	 * EE certs.
+	 */
 	eku = X509_get_ext_d2i(x, NID_ext_key_usage, &crit, NULL);
 	if (eku == NULL) {
 		if (crit != -1)
 			warnx("%s: error parsing EKU", fn);
 		else
-			warnx("%s: EKU: extension missing", fn);
+			purpose = CERT_PURPOSE_EE; /* EKU absent */
 		goto out;
 	}
 	if (crit != 0) {
 		warnx("%s: EKU: extension must not be marked critical", fn);
 		goto out;
 	}
+
+	/*
+	 * XXX - this isn't quite correct: other EKU OIDs are allowed per
+	 * RFC 8209, section 3.1.3.2, e.g., anyEKU could potentially help
+	 * avoid tripping up validators that don't know about the BGPsec
+	 * router purpose. Drop check or downgrade from error to warning?
+	 */
 	if (sk_ASN1_OBJECT_num(eku) != 1) {
 		warnx("%s: EKU: expected 1 purpose, have %d", fn,
 		    sk_ASN1_OBJECT_num(eku));
diff --git a/usr.sbin/sysupgrade/sysupgrade.sh b/usr.sbin/sysupgrade/sysupgrade.sh
index 588c48693..4b9fb053c 100644
--- a/usr.sbin/sysupgrade/sysupgrade.sh
+++ b/usr.sbin/sysupgrade/sysupgrade.sh
@@ -1,6 +1,6 @@
 #!/bin/ksh
 #
-# $OpenBSD: sysupgrade.sh,v 1.49 2023/10/12 12:31:15 kn Exp $
+# $OpenBSD: sysupgrade.sh,v 1.50 2024/06/08 06:05:40 florian Exp $
 #
 # Copyright (c) 1997-2015 Todd Miller, Theo de Raadt, Ken Westerback
 # Copyright (c) 2015 Robert Peichaer <rpe@openbsd.org>
@@ -139,16 +139,21 @@ unpriv -f SHA256.sig ftp -N sysupgrade -Vmo SHA256.sig ${URL}SHA256.sig
 _KEY=secbsd-${_KERNV[0]%.*}${_KERNV[0]#*.}-base.pub
 _NEXTKEY=secbsd-${NEXT_VERSION%.*}${NEXT_VERSION#*.}-base.pub
 
-read _LINE <SHA256.sig
-case ${_LINE} in
-*\ ${_KEY})	SIGNIFY_KEY=/etc/signify/${_KEY} ;;
-*\ ${_NEXTKEY})	SIGNIFY_KEY=/etc/signify/${_NEXTKEY} ;;
-*)		err "invalid signing key" ;;
-esac
+if $SNAP; then
+	unpriv -f SHA256 signify -Ve -x SHA256.sig -m SHA256
+else
+	read _LINE <SHA256.sig
+	case ${_LINE} in
+	*\ ${_KEY})	SIGNIFY_KEY=/etc/signify/${_KEY} ;;
+	*\ ${_NEXTKEY})	SIGNIFY_KEY=/etc/signify/${_NEXTKEY} ;;
+	*)		err "invalid signing key" ;;
+	esac
 
-[[ -f ${SIGNIFY_KEY} ]] || err "cannot find ${SIGNIFY_KEY}"
+	[[ -f ${SIGNIFY_KEY} ]] || err "cannot find ${SIGNIFY_KEY}"
+
+	unpriv -f SHA256 signify -Ve -p "${SIGNIFY_KEY}" -x SHA256.sig -m SHA256
+fi
 
-unpriv -f SHA256 signify -Ve -p "${SIGNIFY_KEY}" -x SHA256.sig -m SHA256
 rm SHA256.sig
 
 if cmp -s /var/db/installed.SHA256 SHA256 && ! $FORCE; then