[PATCH net v3 1/2] macsec: check the resolved SCI for duplicates

[PATCH net v3 1/2] macsec: check the resolved SCI for duplicates

[Haseeb Malik via B4 Relay]

From: Haseeb Malik <haseebulhaq55@gmail.com>

An all-ones IFLA_MACSEC_SCI selects the default SCI derived from the
MACsec device's MAC address and port 1. macsec_init_secy() resolves this
value and stores the result in secy.sci, but macsec_newlink() checks for
duplicates using the unchanged local sci argument.

Consequently, an all-ones request can create a second MACsec device with
the same transmit SCI on the same lower device, while requesting that
SCI explicitly returns -EBUSY.

Resolve an undefined SCI in macsec_newlink() before initializing the
SecY, so initialization and duplicate detection use the same value.
Preserve the all-ones fallback when the resulting SCI is available.

The duplicate acceptance was flagged by Sashiko, an LLM review bot,
during review of an earlier MACsec initialization fix. Source and history
inspection, followed by raw-netlink checks and selftests on the unfixed
kernel, confirmed the issue.

Fixes: c09440f7dcb3 ("macsec: introduce IEEE 802.1AE driver")
Link: https://lists.openwall.net/netdev/2026/09/16/11
Assisted-by: LLM
Signed-off-by: Haseeb Malik <haseebulhaq55@gmail.com>
---
Changes in v3:
- Split the driver fix and selftests into separate patches, as Jakub requested.
- Explain how the issue was discovered, as the submission-info bot requested.
- No code changes from v2.

Changes in v2:
- Resolve undefined SCI in macsec_newlink(), as suggested by Sabrina.

Link to v2: https://lore.kernel.org/netdev/20261001-fix-macsec-duplicate-sci-v2-1-330311abe5ec@gmail.com/

Tested on arm64/virtme with KASAN and lockdep: the regression tests
in patch 2 go from 5 pass/2 fail to 7 pass/0 fail. The full MACsec
selftest passes all 15 cases without skips.

 drivers/net/macsec.c | 8 +++-----
 1 file changed, 3 insertions(+), 5 deletions(-)

diff --git a/drivers/net/macsec.c b/drivers/net/macsec.c
index 78a19b134632..69686f22eda8 100644
--- a/drivers/net/macsec.c
+++ b/drivers/net/macsec.c
@@ -4143,9 +4143,6 @@ static void macsec_init_secy(struct net_device *dev, sci_t sci, u8 icv_len)
 	struct macsec_dev *macsec = macsec_priv(dev);
 	struct macsec_secy *secy = &macsec->secy;
 
-	if (sci == MACSEC_UNDEF_SCI)
-		sci = dev_to_sci(dev, MACSEC_PORT_ES);
-
 	secy->netdev = dev;
 	secy->operational = true;
 	secy->key_len = DEFAULT_SAK_LEN;
@@ -4178,7 +4175,7 @@ static int macsec_newlink(struct net_device *dev,
 	u8 icv_len = MACSEC_DEFAULT_ICV_LEN;
 	struct net_device *real_dev;
 	int err, mtu;
-	sci_t sci;
+	sci_t sci = MACSEC_UNDEF_SCI;
 
 	if (!tb[IFLA_LINK])
 		return -EINVAL;
@@ -4231,7 +4228,8 @@ static int macsec_newlink(struct net_device *dev,
 		sci = nla_get_sci(data[IFLA_MACSEC_SCI]);
 	else if (data && data[IFLA_MACSEC_PORT])
 		sci = dev_to_sci(dev, nla_get_be16(data[IFLA_MACSEC_PORT]));
-	else
+
+	if (sci == MACSEC_UNDEF_SCI)
 		sci = dev_to_sci(dev, MACSEC_PORT_ES);
 
 	/* Registration can notify listeners before returning. */


base-commit: 11536ee3d3e0b1bd35b6f3f8df55a6053eb0c71d
change-id: 20260926-fix-macsec-duplicate-sci-ed7635ac1c35

-- 
2.43.0

[PATCH net v3 2/2] selftests: net: test MACsec undefined SCI uniqueness

[Haseeb Malik via B4 Relay]

From: Haseeb Malik <haseebulhaq55@gmail.com>

Exercise duplicate transmit SCI rejection using default, explicit and
all-ones SCI requests on the same lower device. Cover an all-ones request
for both the first and second device, and clean up the second device if
a kernel incorrectly accepts the duplicate.

Also verify that an all-ones SCI still resolves to the default when a
different port is already in use, and that deleting the device makes
its SCI available for reuse.

These cases cover the duplicate acceptance flagged by Sashiko during
review of an earlier MACsec initialization fix. The same tests produce
five passes and two failures before the fix, and seven passes after it.

Check local iproute2 MACsec support before running the new tests. Keep
the existing local and remote checks for callers supplying a configuration,
without requiring remote or offload support for the new local tests.

Link: https://lists.openwall.net/netdev/2026/09/16/11
Assisted-by: LLM
Signed-off-by: Haseeb Malik <haseebulhaq55@gmail.com>
---
Changes in v3:
- Split the selftests from the driver fix, as Jakub requested.
- No code changes from v2.

Changes in v2:
- Check local iproute2 MACsec support in the new tests, as Sashiko noted.

Link to v2: https://lore.kernel.org/netdev/20261001-fix-macsec-duplicate-sci-v2-1-330311abe5ec@gmail.com/

 tools/testing/selftests/drivers/net/macsec.py | 57 +++++++++++++++++++++++++--
 1 file changed, 54 insertions(+), 3 deletions(-)

diff --git a/tools/testing/selftests/drivers/net/macsec.py b/tools/testing/selftests/drivers/net/macsec.py
index 9a83d9542e04..81003bda9b0d 100755
--- a/tools/testing/selftests/drivers/net/macsec.py
+++ b/tools/testing/selftests/drivers/net/macsec.py
@@ -34,9 +34,10 @@ def _get_features(dev):
     return ethtool(f"-k {dev}", json=True)[0]
 
 
-def _require_ip_macsec(cfg):
-    """SKIP if iproute2 on local or remote lacks 'ip macsec' support."""
-    for host in [None, cfg.remote]:
+def _require_ip_macsec(cfg=None):
+    """SKIP if iproute2 lacks MACsec locally or on cfg.remote, when given."""
+    hosts = [None, cfg.remote] if cfg is not None else [None]
+    for host in hosts:
         out = cmd("ip macsec help", fail=False, host=host)
         if "Usage" not in out.stdout + out.stderr:
             where = "remote" if host else "local"
@@ -263,6 +264,54 @@ def test_offload_state(cfg) -> None:
             "features should match first offload-on snapshot")
 
 
+@ksft_variants([
+    KsftNamedVariant("default", "", ""),
+    KsftNamedVariant("explicit", "", "sci {sci}"),
+    KsftNamedVariant("undefined", "", "sci ffffffffffffffff"),
+    KsftNamedVariant("undefined_default", "sci ffffffffffffffff", ""),
+    KsftNamedVariant("undefined_explicit", "sci ffffffffffffffff", "sci {sci}"),
+    KsftNamedVariant("undefined_twice", "sci ffffffffffffffff",
+                     "sci ffffffffffffffff"),
+])
+def test_duplicate_sci(cfg, first, second) -> None:
+    """Reject duplicate transmit SCIs, including the undefined-SCI fallback."""
+
+    _require_ip_macsec()
+    ms0 = _macsec_name(0)
+    ms1 = _macsec_name(1)
+    sci = _get_mac(cfg.ifname).replace(":", "") + "0001"
+
+    ip(f"link add link {cfg.ifname} {ms0} type macsec {first}")
+    defer(ip, f"link del {ms0}")
+    with ksft_raises(CmdExitFailure):
+        ip(f"link add link {cfg.ifname} {ms1} type macsec "
+           f"{second.format(sci=sci)}")
+        # Clean up if the kernel incorrectly accepted the duplicate.
+        defer(ip, f"link del {ms1}")
+
+
+def test_undefined_sci(cfg) -> None:
+    """An undefined SCI still selects the default when it is available."""
+
+    _require_ip_macsec()
+    ms0 = _macsec_name(0)
+    ms1 = _macsec_name(1)
+    sci = _get_mac(cfg.ifname).replace(":", "") + "0001"
+
+    # A different port on the same lower device must not block the fallback.
+    ip(f"link add link {cfg.ifname} {ms0} type macsec port 2")
+    defer(ip, f"link del {ms0}")
+    ip(f"link add link {cfg.ifname} {ms1} type macsec sci ffffffffffffffff")
+    cleanup = defer(ip, f"link del {ms1}")
+    info = ip(f"-d link show dev {ms1}", json=True)[0]
+    ksft_eq(info["linkinfo"]["info_data"]["sci"], sci)
+
+    # Deleting a device must make its SCI available again.
+    cleanup.exec()
+    ip(f"link add link {cfg.ifname} {ms1} type macsec sci ffffffffffffffff")
+    defer(ip, f"link del {ms1}")
+
+
 def _check_nsim_vid(cfg, vid, expected) -> None:
     """Checks if a VLAN is present. Only works on netdevsim."""
 
@@ -333,6 +382,8 @@ def main() -> None:
                   test_max_secy,
                   test_max_sc,
                   test_offload_state,
+                  test_duplicate_sci,
+                  test_undefined_sci,
                   test_vlan,
                   test_vlan_toggle,
                   ], args=(cfg,))

-- 
2.43.0