net/sched: act_ct: fix skb leak on fragment check failure
Subsystems (3)
PATCH 1/1 net/sched: act_ct: fix skb leak on fragment check failure
When tcf_ct_handle_fragments() returns an error other than -EINPROGRESS (e.g. -EINVAL from malformed fragments), tcf_ct_act() jumps to out_frag which unconditionally returns TC_ACT_CONSUMED. This tells the caller the skb was consumed, but it was not freed, leaking one skb per malformed fragment.
TC_ACT_CONSUMED is only correct for -EINPROGRESS, where defragmentation is genuinely in progress and the skb has been queued. For all other errors the skb is still owned by the caller and must be freed via TC_ACT_SHOT.
--- net/sched/act_ct.c | 4 +++- 1 file changed, 3 insertions(+), 1 deletion(-)
net/sched/act_ct.c +3 -1
@@ -1107,8 +1107,10 @@ TC_INDIRECT_SCOPE int tcf_ct_act(struct sk_buff *skb, const struct tc_action *a,return retval; out_frag: -if (err != -EINPROGRESS ) +if (err != -EINPROGRESS ) {tcf_action_inc_drop_qstats (&c->common); +return TC_ACT_SHOT ; + }return TC_ACT_CONSUMED ; drop:
Signature · 1 line
2.39.3 (Apple Git-145)
Jamal Hadi Salim
maintainerQuoted 7 lines · Dudu Lu
When
tcf_ct_handle_fragments()returns an error other than-EINPROGRESS(e.g.-EINVALfrom malformed fragments),tcf_ct_act()jumps toout_fragwhich unconditionally returnsTC_ACT_CONSUMED. This tells the caller the skb was consumed, but it was not freed, leaking one skb per malformed fragment.
Quoted 7 lines
TC_ACT_CONSUMEDis only correct for-EINPROGRESS, where defragmentation is genuinely in progress and the skb has been queued. For all other errors the skb is still owned by the caller and must be freed viaTC_ACT_SHOT.Fixes 3f14b377d01d ("net/sched: act_ct: fix skb leak and crash on ooo frags")Signed-off-by Dudu Lu <[email protected]>
Do you have a reproducer? Always helps adding at least a tdc test.
Also: How did you find this issue? was it AI? If yes, please add the tag "Assisted-by:<AI name here>"
cheers,
jamal
Quoted 23 lines
--- net/sched/act_ct.c | 4 +++- 1 file changed, 3 insertions(+), 1 deletion(-)net/sched/act_ct.c +3
-2@@ -1107,8 +1107,10 @@ TC_INDIRECT_SCOPE int tcf_ct_act(struct sk_buff *skb, const struct tc_action *a,return retval; out_frag: -if (err != -EINPROGRESS ) +if (err != -EINPROGRESS ) {tcf_action_inc_drop_qstats (&c->common); +return TC_ACT_SHOT ; + }return TC_ACT_CONSUMED ; drop: --2.39.3 (Apple Git-145)
Found it through code review. Reproduced it on a 7.0-rc6 kernel using a veth pair with act_ct on ingress:
ip netns add ns_ct
ip link add veth0 type veth peer name veth1
ip link set veth1 netns ns_ct
ip link set veth0 up
ip netns exec ns_ct ip link set veth1 up
tc qdisc add dev veth0 clsact
tc filter add dev veth0 ingress protocol ip flower action ct zone 1
Then send truncated IP packets (10 bytes IP, need 20 minimum) from the namespace via raw AF_PACKET socket on veth1. This hits pskb_may_pull failure in tcf_ct_ipv4_is_fragment -> -EINVAL -> out_frag -> TC_ACT_CONSUMED. net/core/dev.c handles TC_ACT_CONSUMED by returning NULL without freeing the skb.
Result on unpatched kernel:
Sent: 222 packets
skbuff_head_cache: before=6439 after=6663 growth=224
FAIL: skb leak detected (224 objects leaked)
Attached a test script that automates this. With the fix applied (TC_ACT_SHOT for non-EINPROGRESS errors), the skbs get freed and the test passes. The test script is generated by AI.
Quoted 28 lines · Jamal Hadi Salim
Quoted 7 lines · Dudu Lu
When
tcf_ct_handle_fragments()returns an error other than-EINPROGRESS(e.g.-EINVALfrom malformed fragments),tcf_ct_act()jumps toout_fragwhich unconditionally returnsTC_ACT_CONSUMED. This tells the caller the skb was consumed, but it was not freed, leaking one skb per malformed fragment.Quoted 6 lines
TC_ACT_CONSUMEDis only correct for-EINPROGRESS, where defragmentation is genuinely in progress and the skb has been queued. For all other errors the skb is still owned by the caller and must be freed viaTC_ACT_SHOT.Fixes 3f14b377d01d ("net/sched: act_ct: fix skb leak and crash on ooofrags")
Quoted 1 line
Signed-off-by Dudu Lu <[email protected]>Do you have a reproducer? Always helps adding at least a tdc test.
Also: How did you find this issue? was it AI? If yes, please add the tag "Assisted-by:<AI name here>"cheers,
jamalQuoted 4 lines
--- net/sched/act_ct.c | 4 +++- 1 file changed, 3 insertions(+), 1 deletion(-)
Quoted 21 lines
net/sched/act_ct.c TC_INDIRECT_SCOPE int tcf_ct_act(struct sk_buff
Quoted 5 lines
net/sched/act_ct.c +0
-0@@ -1107,8 +1107,10 @@ TC_INDIRECT_SCOPE int tcf_ct_act(struct sk_buff*skb, const struct
tc_action*a,Quoted 14 lines
net/sched/act_ct.c +3
-2return retval; out_frag: -if (err != -EINPROGRESS ) +if (err != -EINPROGRESS ) {tcf_action_inc_drop_qstats (&c->common); +return TC_ACT_SHOT ; + }return TC_ACT_CONSUMED ; drop: --2.39.3 (Apple Git-145)
#!/bin/bash
# SPDX-License-Identifier: GPL-2.0
#
# Test for skb leak in act_ct when tcf_ct_handle_fragments() fails. #
# When a truncated IP packet hits act_ct, tcf_ct_ipv4_is_fragment()
# returns -EINVAL. The out_frag path returns TC_ACT_CONSUMED, but the # skb was never actually consumed — it leaks. # # This test sends truncated IP packets through a veth pair with act_ct # on ingress and checks skbuff_head_cache slab growth.
set -e
readonly NS="ns-act-ct-leak-$(mktemp -u XXXXXX)" readonly DEV="veth-ct0"
readonly DEV_PEER="veth-ct1"
readonly NUM_PKTS=500
# Minimum slab growth to consider a leak (allow some noise) readonly LEAK_THRESHOLD=200
cleanup() {
ip link del $DEV 2>/dev/null || true
ip netns del $NS 2>/dev/null || true
}trap cleanup EXIT
get_skb_active() {
awk '/skbuff_head_cache/ {print $2}' /proc/slabinfo
}
# Build the packet sender
build_sender() {
local prog="$1"
local src="$prog.c"
cat > "$src" << 'EOF'#include <stdio.h>
#include <string.h>
#include <stdlib.h>
#include <unistd.h>
#include <sys/socket.h>
#include <linux/if_packet.h>
#include <net/if.h>
#include <net/ethernet.h>
#include <arpa/inet.h>
#include <fcntl.h>
int main(int argc, char **argv) {
const char *ifname = argv[1];
int count = atoi(argv[2]);
int fd, i, sent = 0;
fd = socket(AF_PACKET, SOCK_RAW, htons(ETH_P_ALL));
if (fd < 0) { perror("socket"); return 1; }
fcntl(fd, F_SETFL, O_NONBLOCK);
struct sockaddr_ll sa = {};
sa.sll_family = AF_PACKET;
sa.sll_protocol = htons(ETH_P_IP);
sa.sll_ifindex = if_nametoindex(ifname);
sa.sll_halen = 6;
memset(sa.sll_addr, 0xff, 6);
/* Ethernet(14) + truncated IP(10) = 24 bytes.
* IP header needs 20 bytes minimum, so pskb_may_pull fails
* in tcf_ct_ipv4_is_fragment() -> -EINVAL.
*/
unsigned char pkt[24] = {};
memset(pkt, 0xff, 6); /* dst = broadcast */
pkt[12] = 0x08; pkt[13] = 0x00;/* ethertype = IPv4 */
pkt[14] = 0x45; /* ver=4, ihl=5 */
pkt[16] = 0x00; pkt[17] = 0x0a;/* total_len=10 */
pkt[23] = 0x06; /* proto=TCP */
for (i = 0; i < count; i++) {
if (sendto(fd, pkt, sizeof(pkt), 0,
(struct sockaddr *)&sa, sizeof(sa)) >= 0)
sent++;
}
printf("%d\n", sent);
close(fd);
return 0;
}
EOF
gcc -Wall -o "$prog" "$src"
}echo "=== act_ct skb leak test ==="
# Setup veth pair with namespace
ip netns add $NS
ip link add $DEV type veth peer name $DEV_PEER
ip link set $DEV_PEER netns $NS
ip link set $DEV up
ip netns exec $NS ip link set $DEV_PEER up
# Add act_ct filter on ingress
tc qdisc add dev $DEV clsact
tc filter add dev $DEV ingress protocol ip flower action ct zone 1
# Build sender
SENDER=$(mktemp)build_sender "$SENDER"
# Record slab state
BEFORE=$(get_skb_active)
# Send truncated packets from namespace SENT=$(ip netns exec $NS "$SENDER" $DEV_PEER $NUM_PKTS) sleep 1
# Record slab state again
AFTER=$(get_skb_active)
# Check tc stats
DROPPED=$(tc -s filter show dev $DEV ingress | \
awk '/dropped/ {for(i=1;i<=NF;i++) if($i=="dropped") print $(i+1)}' | \
tr -d ',')GROWTH=$((AFTER - BEFORE))
echo "Sent: $SENT packets"
echo "TC dropped: $DROPPED"
echo "skbuff_head_cache: before=$BEFORE after=$AFTER growth=$GROWTH"
rm -f "$SENDER" "${SENDER}.c"
if [ "$GROWTH" -ge "$LEAK_THRESHOLD" ]; then
echo "FAIL: skb leak detected ($GROWTH objects leaked)" exit 1
else
echo "PASS: no significant skb leak"
exit 0
fi
Jamal Hadi Salim
maintainerQuoted 4 lines · phx
Found it through code review. Reproduced it on a 7.0-rc6 kernel using a veth pair with
act_cton ingress:
Quoted 8 lines
ip netns add
ns_ct
ip link add veth0 type veth peer name veth1
ip link set veth1 netnsns_ct
ip link set veth0 up
ip netns execns_ctip link set veth1 up
tc qdisc add dev veth0 clsact
tc filter add dev veth0 ingress protocol ip flower action ct zone 1
Please create a test case using tdc as a second patch. And look at what Sashiko is complaining about and address it if it makes sense...
cheers,
jamal
Quoted 63 lines
Then send truncated IP packets (10 bytes IP, need 20 minimum) from the namespace via raw
AF_PACKETsocket on veth1. This hitspskb_may_pullfailure intcf_ct_ipv4_is_fragment->-EINVAL->out_frag->TC_ACT_CONSUMED.net/core/dev.chandlesTC_ACT_CONSUMEDby returning NULL without freeing the skb.Result on unpatched kernel:
Sent: 222 packets
skbuff_head_cache: before=6439 after=6663 growth=224
FAIL: skb leak detected (224 objects leaked)Attached a test script that automates this. With the fix applied (
TC_ACT_SHOTfor non-EINPROGRESS errors), the skbs get freed and the test passes. The test script is generated by AI.Quoted 47 lines · Jamal Hadi Salim
Quoted 7 lines · Dudu Lu
When
tcf_ct_handle_fragments()returns an error other than-EINPROGRESS(e.g.-EINVALfrom malformed fragments),tcf_ct_act()jumps toout_fragwhich unconditionally returnsTC_ACT_CONSUMED. This tells the caller the skb was consumed, but it was not freed, leaking one skb per malformed fragment.Quoted 7 lines
TC_ACT_CONSUMEDis only correct for-EINPROGRESS, where defragmentation is genuinely in progress and the skb has been queued. For all other errors the skb is still owned by the caller and must be freed viaTC_ACT_SHOT.Fixes 3f14b377d01d ("net/sched: act_ct: fix skb leak and crash on ooo frags")Signed-off-by Dudu Lu <[email protected]>Do you have a reproducer? Always helps adding at least a tdc test.
Also: How did you find this issue? was it AI? If yes, please add the tag "Assisted-by:<AI name here>"cheers,
jamalQuoted 23 lines
--- net/sched/act_ct.c | 4 +++- 1 file changed, 3 insertions(+), 1 deletion(-)net/sched/act_ct.c +3
-2@@ -1107,8 +1107,10 @@ TC_INDIRECT_SCOPE int tcf_ct_act(struct sk_buff *skb, const struct tc_action *a,return retval; out_frag: -if (err != -EINPROGRESS ) +if (err != -EINPROGRESS ) {tcf_action_inc_drop_qstats (&c->common); +return TC_ACT_SHOT ; + }return TC_ACT_CONSUMED ; drop: --2.39.3 (Apple Git-145)