git: d80b60653fb8 - releng/14.5 - esp: Fix synchronization of stream cipher counter updates
- Go to: [ bottom of page ] [ top of archives ] [ this month ]
Date: Tue, 17 Nov 2026 17:05:10 UTC
The branch releng/14.5 has been updated by markj:
URL: https://cgit.FreeBSD.org/src/commit/?id=d80b60653fb8094ec5916f643265aad153f27024
commit d80b60653fb8094ec5916f643265aad153f27024
Author: R. Christian McDonald <rcm@FreeBSD.org>
AuthorDate: 2026-11-12 18:10:09 +0000
Commit: Mark Johnston <markj@FreeBSD.org>
CommitDate: 2026-11-16 21:17:12 +0000
esp: Fix synchronization of stream cipher counter updates
After commit 0361f165f219, the counter embedded in the IV for AES-CTR,
AES-GCM and ChaCha20 transforms is incremented under a shared lock,
which creates the potential for nonce reuse. This can be abused by an
on-path attacker to alter the ciphertext of intercepted packets, and
potentially reveal the plaintext of packets whose counter values
collide.
Approved by: so
Security: FreeBSD-SA-26:82.ipsec
Security: CVE-2026-101318
Sponsored by: Rubicon Communications, LLC ("Netgate")
Fixes: 0361f165f219 ("ipsec: replace SECASVAR mtx by rmlock")
---
sys/netipsec/xform_esp.c | 6 ++-
tests/sys/netipsec/tunnel/Makefile | 3 +-
tests/sys/netipsec/tunnel/esp_iv.sh | 91 +++++++++++++++++++++++++++++++++++++
3 files changed, 98 insertions(+), 2 deletions(-)
diff --git a/sys/netipsec/xform_esp.c b/sys/netipsec/xform_esp.c
index 7ad6085db87f..9c3f97355e7f 100644
--- a/sys/netipsec/xform_esp.c
+++ b/sys/netipsec/xform_esp.c
@@ -824,8 +824,12 @@ esp_output(struct mbuf *m, struct secpolicy *sp, struct secasvar *sav,
SECREPLAY_UNLOCK(sav->replay);
}
cryptoid = sav->tdb_cryptoid;
+ /*
+ * The SA lock is shared by every CPU sending on the SA; the IV
+ * counter must still advance once per packet.
+ */
if (SAV_ISCTRORGCM(sav) || SAV_ISCHACHA(sav))
- cntr = sav->cntr++;
+ cntr = atomic_fetchadd_64(&sav->cntr, 1);
SECASVAR_RUNLOCK(sav);
/*
diff --git a/tests/sys/netipsec/tunnel/Makefile b/tests/sys/netipsec/tunnel/Makefile
index c28826711cb8..babce374d59a 100644
--- a/tests/sys/netipsec/tunnel/Makefile
+++ b/tests/sys/netipsec/tunnel/Makefile
@@ -12,7 +12,8 @@ ATF_TESTS_SH+= empty \
aesni_aes_cbc_256_hmac_sha2_256 \
aesni_aes_gcm_128 \
aesni_aes_gcm_256 \
- chacha20_poly1305
+ chacha20_poly1305 \
+ esp_iv
# Each test uses the same names for its jails, so they must be run serially.
TEST_METADATA+= is_exclusive=true
diff --git a/tests/sys/netipsec/tunnel/esp_iv.sh b/tests/sys/netipsec/tunnel/esp_iv.sh
new file mode 100644
index 000000000000..f218113a5c95
--- /dev/null
+++ b/tests/sys/netipsec/tunnel/esp_iv.sh
@@ -0,0 +1,91 @@
+. $(atf_get_srcdir)/utils.subr
+
+# The explicit IV of AES-GCM, AES-CTR and ChaCha20-Poly1305 must never repeat
+# under one key. esp_output() takes it from a per-SA counter that every CPU
+# sending on the SA advances, so drive one tunnel from several processes at
+# once, capture the ESP on the wire and look for a repeated IV per SPI.
+
+atf_test_case "unique" "cleanup"
+unique_head()
+{
+ atf_set descr 'The explicit ESP IV is unique per SA under concurrent senders'
+ atf_set require.user root
+ atf_set require.progs tcpdump
+ atf_set timeout 600
+}
+
+unique_body()
+{
+ if [ "$(sysctl -n hw.ncpu)" -lt 2 ]; then
+ atf_skip "This test needs more than one CPU"
+ fi
+
+ ist_init
+ ist_labsetup
+ ist_v4_setup
+ ist_setkey ipsecA out 4 aes-gcm-16 "12345678901234567890"
+ atf_check -s exit:0 -o ignore \
+ jexec ipsecA setkey -f ${TMPDIR}/ipsec.ipsecA.conf
+ ist_setkey ipsecB in 4 aes-gcm-16 "12345678901234567890"
+ atf_check -s exit:0 -o ignore \
+ jexec ipsecB setkey -f ${TMPDIR}/ipsec.ipsecB.conf
+ atf_check -s exit:0 -o ignore jexec hostA ping -c 1 203.0.113.1
+
+ # ESP in both directions crosses the router.
+ jexec router tcpdump -ni ${epair_PUB_A}b -s 64 -w ${TMPDIR}/esp.pcap esp \
+ > /dev/null 2>&1 &
+ tcpdump_pid=$!
+ sleep 1
+
+ # Flood pings from the gateway itself, so that each one runs
+ # esp_output() in its own context on whatever CPU the scheduler gives
+ # it: many concurrent senders on the same SA each way. (Sent from
+ # hostA they would all arrive through one epair queue and be
+ # serialised before they reach IPsec.)
+ pids=""
+ for i in 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16; do
+ jexec ipsecA ping -q -f -c 12000 -S 192.0.2.2 203.0.113.1 \
+ > /dev/null 2>&1 &
+ pids="${pids} $!"
+ done
+ wait ${pids}
+ sleep 1
+ kill -INT ${tcpdump_pid}
+ wait ${tcpdump_pid} 2>/dev/null
+
+ # Per packet: hex of the first 48 bytes after the link header; the ESP
+ # header follows the IPv4 header, the explicit IV follows the 8-byte
+ # ESP header. Count a (SPI, IV) pair seen more than once.
+ tcpdump -nnr ${TMPDIR}/esp.pcap -x esp 2>/dev/null | awk '
+ /^[0-9][0-9]:/ { flush(); next }
+ /^[ \t]*0x00[012]0:/ { line = $0; sub(/^[ \t]*0x00[0-9a-f]+:[ \t]*/, "", line); gsub(/ /, "", line); hex = hex line; next }
+ function flush() {
+ if (length(hex) >= 72) {
+ ihl = strtonum_hex(substr(hex, 2, 1)) * 8;
+ spi = substr(hex, ihl + 1, 8); iv = substr(hex, ihl + 17, 16);
+ n++; if (++seen[spi " " iv] == 2) dup++;
+ }
+ hex = ""
+ }
+ function strtonum_hex(c) { return index("0123456789abcdef", c) - 1 }
+ END { flush(); printf "%d %d\n", n, dup }' > ${TMPDIR}/esp.count
+ read packets dups < ${TMPDIR}/esp.count
+ echo "ESP packets captured: ${packets}, repeated IVs: ${dups}"
+
+ if [ "${packets}" -lt 100000 ]; then
+ atf_fail "Captured only ${packets} ESP packets"
+ fi
+ if [ "${dups}" -ne 0 ]; then
+ atf_fail "${dups} ESP packets reused an explicit IV"
+ fi
+}
+
+unique_cleanup()
+{
+ ist_cleanup
+}
+
+atf_init_test_cases()
+{
+ atf_add_test_case "unique"
+}