Skip to contentSkip to conversation

macsec: offload-related fixes

PATCH Superseded Sabrina Dubroca 5 participants 15 messages Oct 26, 2022 – Oct 30, 2022

Subsystems (3)

v2 of 3 Superseded Latest is v3 →

lore
Share & tools
lore

Sabrina Dubroca

author maintainer
#0 lore raw

v2 cover macsec: offload-related fixes

I'm working on a dummy offload for macsec on netdevsim. It just has a small SecY and RXSC table so I can trigger failures easily on the ndo_* side. It has exposed a couple of issues.

The first patch is a revert of commit c850240b6c41 ("net: macsec:
report real_dev features when HW offloading is enabled"). That commit tried to improve the performance of macsec offload by taking advantage of some of the NIC's features, but in doing so, broke macsec offload when the lower device supports both macsec and ipsec offload, as the ipsec offload feature flags were copied from the real device. Since the macsec device doesn't provide xdo_* ops, the XFRM core rejects the registration of the new macsec device in xfrm_api_check.

I'm working on re-adding those feature flags when offload is available, but I haven't fully solved that yet. I think it would be safer to do that second part in net-next considering how complex feature interactions tend to be.

v2:
- better describe the issue introduced by commit c850240b6c41 (Leon Romanovsky)
- drop unnecessary !! (Leon Romanovsky)

Sabrina Dubroca (5):
Revert "net: macsec: report real_dev features when HW offloading is
enabled"
macsec: delete new rxsc when offload fails
macsec: fix secy->n_rx_sc accounting
macsec: fix detection of RXSCs when toggling offloading
macsec: clear encryption keys from the stack after setting up offload

 drivers/net/macsec.c | 51 ++++++++++++++++----------------------------
 1 file changed, 18 insertions(+), 33 deletions(-)
Signature · 1 line
2.38.0
author maintainer
#1 lore raw

PATCH v2 1/5 Revert "net: macsec: report real_dev features when HW offloading is enabled"

This reverts commit c850240b6c4132574a00f2da439277ab94265b66.

That commit tried to improve the performance of macsec offload by taking advantage of some of the NIC's features, but in doing so, broke macsec offload when the lower device supports both macsec and ipsec offload, as the ipsec offload feature flags (mainly NETIF_F_HW_ESP) were copied from the real device. Since the macsec device doesn't provide xdo_* ops, the XFRM core rejects the registration of the new macsec device in xfrm_api_check.

Example perf trace when running
ip link add link eni1np1 type macsec port 4 offload mac

    ip   737 [003]   795.477676: probe:xfrm_dev_event__REGISTER      name="macsec0" features=0x1c000080014869
              xfrm_dev_event+0x3a
              notifier_call_chain+0x47
              register_netdevice+0x846
              macsec_newlink+0x25a

    ip   737 [003]   795.477687:   probe:xfrm_dev_event__return      ret=0x8002 (NOTIFY_BAD)
             notifier_call_chain+0x47
             register_netdevice+0x846
             macsec_newlink+0x25a

dev->features includes NETIF_F_HW_ESP (0x04000000000000), so xfrm_api_check returns NOTIFY_BAD because we don't have dev->xfrmdev_ops on the macsec device.

We could probably propagate GSO and a few other features from the lower device, similar to macvlan. This will be done in a future patch.

---
 drivers/net/macsec.c | 27 ++++-----------------------
 1 file changed, 4 insertions(+), 23 deletions(-)
drivers/net/macsec.c +4 -23
index c891b60937a7..b3f76e8071f2 100644
@@ -2654,11 +2654,6 @@ static int macsec_upd_offload(struct sk_buff *skb, struct genl_info *info)
 	if (ret)
 		goto rollback;
 
-	/* Force features update, since they are different for SW MACSec and
-	 * HW offloading cases.
-	 */
-	netdev_update_features(dev);
-
 	rtnl_unlock();
 	return 0;
 
@@ -3432,16 +3427,9 @@ static netdev_tx_t macsec_start_xmit(struct sk_buff *skb,
 	return ret;
 }
 
-#define SW_MACSEC_FEATURES \
+#define MACSEC_FEATURES \
 	(NETIF_F_SG | NETIF_F_HIGHDMA | NETIF_F_FRAGLIST)
 
-/* If h/w offloading is enabled, use real device features save for
- *   VLAN_FEATURES - they require additional ops
- *   HW_MACSEC - no reason to report it
- */
-#define REAL_DEV_FEATURES(dev) \
-	((dev)->features & ~(NETIF_F_VLAN_FEATURES | NETIF_F_HW_MACSEC))
-
 static int macsec_dev_init(struct net_device *dev)
 {
 	struct macsec_dev *macsec = macsec_priv(dev);
@@ -3458,12 +3446,8 @@ static int macsec_dev_init(struct net_device *dev)
 		return err;
 	}
 
-	if (macsec_is_offloaded(macsec)) {
-		dev->features = REAL_DEV_FEATURES(real_dev);
-	} else {
-		dev->features = real_dev->features & SW_MACSEC_FEATURES;
-		dev->features |= NETIF_F_LLTX | NETIF_F_GSO_SOFTWARE;
-	}
+	dev->features = real_dev->features & MACSEC_FEATURES;
+	dev->features |= NETIF_F_LLTX | NETIF_F_GSO_SOFTWARE;
 
 	dev->needed_headroom = real_dev->needed_headroom +
 			       MACSEC_NEEDED_HEADROOM;
@@ -3495,10 +3479,7 @@ static netdev_features_t macsec_fix_features(struct net_device *dev,
 	struct macsec_dev *macsec = macsec_priv(dev);
 	struct net_device *real_dev = macsec->real_dev;
 
-	if (macsec_is_offloaded(macsec))
-		return REAL_DEV_FEATURES(real_dev);
-
-	features &= (real_dev->features & SW_MACSEC_FEATURES) |
+	features &= (real_dev->features & MACSEC_FEATURES) |
 		    NETIF_F_GSO_SOFTWARE | NETIF_F_SOFT_FEATURES;
 	features |= NETIF_F_LLTX;
 
Signature · 1 line
2.38.0
Reviewed-by on v2 1/5
#2 lore raw

Re: [PATCH net v2 1/5] Revert "net: macsec: report real_dev features when HW offloading is enabled"

Quoted 28 lines · Sabrina Dubroca

This reverts commit c850240b6c4132574a00f2da439277ab94265b66.

That commit tried to improve the performance of macsec offload by taking advantage of some of the NIC's features, but in doing so, broke macsec offload when the lower device supports both macsec and ipsec offload, as the ipsec offload feature flags (mainly NETIF_F_HW_ESP) were copied from the real device. Since the macsec device doesn't provide xdo_* ops, the XFRM core rejects the registration of the new macsec device in xfrm_api_check.

Example perf trace when running
ip link add link eni1np1 type macsec port 4 offload mac

    ip   737 [003]   795.477676: probe:xfrm_dev_event__REGISTER      name="macsec0" features=0x1c000080014869
              xfrm_dev_event+0x3a
              notifier_call_chain+0x47
              register_netdevice+0x846
              macsec_newlink+0x25a

    ip   737 [003]   795.477687:   probe:xfrm_dev_event__return      ret=0x8002 (NOTIFY_BAD)
             notifier_call_chain+0x47
             register_netdevice+0x846
             macsec_newlink+0x25a

dev->features includes NETIF_F_HW_ESP (0x04000000000000), so xfrm_api_check returns NOTIFY_BAD because we don't have dev->xfrmdev_ops on the macsec device.

Quoted 8 lines

We could probably propagate GSO and a few other features from the lower device, similar to macvlan. This will be done in a future patch.

---
 drivers/net/macsec.c | 27 ++++-----------------------
 1 file changed, 4 insertions(+), 23 deletions(-)

It is still mystery for me why mlx5 works.

Thanks,

#3 lore raw

RE: [PATCH net v2 1/5] Revert "net: macsec: report real_dev features when HW offloading is enabled"

Quoted 34 lines · Sabrina Dubroca
Quoted 10 lines

This reverts commit c850240b6c4132574a00f2da439277ab94265b66.

That commit tried to improve the performance of macsec offload by taking advantage of some of the NIC's features, but in doing so, broke macsec offload when the lower device supports both macsec and ipsec offload, as the ipsec offload feature flags (mainly NETIF_F_HW_ESP) were copied from the real device. Since the macsec device doesn't provide xdo_* ops, the XFRM core rejects the registration of the new macsec device in xfrm_api_check.

Quoted 4 lines

Example perf trace when running
ip link add link eni1np1 type macsec port 4 offload mac

    ip   737 [003]   795.477676: probe:xfrm_dev_event__REGISTER

name="macsec0" features=0x1c000080014869

Quoted 6 lines
              xfrm_dev_event+0x3a
              notifier_call_chain+0x47
              register_netdevice+0x846
              macsec_newlink+0x25a

    ip   737 [003]   795.477687:   probe:xfrm_dev_event__return      ret=0x8002

(NOTIFY_BAD)

Quoted 11 lines
             notifier_call_chain+0x47
             register_netdevice+0x846
             macsec_newlink+0x25a

dev->features includes NETIF_F_HW_ESP (0x04000000000000), so xfrm_api_check returns NOTIFY_BAD because we don't have dev->xfrmdev_ops on the macsec device.

We could probably propagate GSO and a few other features from the lower device, similar to macvlan. This will be done in a future patch.

Quoted 7 lines
Quoted 5 lines
---
 drivers/net/macsec.c | 27 ++++-----------------------
 1 file changed, 4 insertions(+), 23 deletions(-)

It is still mystery for me why mlx5 works.

I think it works when the offload enablement on the macsec device is done after the macsec netdev creation stage i.e.:
Ip link add link eni1np1 type macsec port 4
Then
ip macsec offload $MACSEC_IF mac
I think this path skips xfrm core rejection

Quoted 3 lines

Thanks,

#4 lore raw

RE: [PATCH net v2 1/5] Revert "net: macsec: report real_dev features when HW offloading is enabled"

Quoted 41 lines
Quoted 11 lines · Sabrina Dubroca
Quoted 10 lines

This reverts commit c850240b6c4132574a00f2da439277ab94265b66.

That commit tried to improve the performance of macsec offload by taking advantage of some of the NIC's features, but in doing so, broke macsec offload when the lower device supports both macsec and ipsec offload, as the ipsec offload feature flags (mainly NETIF_F_HW_ESP) were copied from the real device. Since the macsec device doesn't provide xdo_* ops, the XFRM core rejects the registration of the new macsec device in xfrm_api_check.

Quoted 11 lines
Quoted 4 lines

Example perf trace when running
ip link add link eni1np1 type macsec port 4 offload mac

    ip   737 [003]   795.477676: probe:xfrm_dev_event__REGISTER

name="macsec0" features=0x1c000080014869

Quoted 6 lines
              xfrm_dev_event+0x3a
              notifier_call_chain+0x47
              register_netdevice+0x846
              macsec_newlink+0x25a

    ip   737 [003]   795.477687:   probe:xfrm_dev_event__return

ret=0x8002

Quoted 18 lines

(NOTIFY_BAD)

Quoted 16 lines
             notifier_call_chain+0x47
             register_netdevice+0x846
             macsec_newlink+0x25a

dev->features includes NETIF_F_HW_ESP (0x04000000000000), so xfrm_api_check returns NOTIFY_BAD because we don't have dev->xfrmdev_ops on the macsec device.

We could probably propagate GSO and a few other features from the lower device, similar to macvlan. This will be done in a future patch.

---
 drivers/net/macsec.c | 27 ++++-----------------------
 1 file changed, 4 insertions(+), 23 deletions(-)
Quoted 5 lines
Quoted 1 line

It is still mystery for me why mlx5 works.

I think it works when the offload enablement on the macsec device is done after the macsec netdev creation stage i.e.:
Ip link add link eni1np1 type macsec port 4 Then ip macsec offload $MACSEC_IF mac I think this path skips xfrm core rejection

Further check it shows that xfrm_api_check fails in both configuration flavors, in first method because of register_netdevice call the second method because of netdev_update_features call, however, using the second method (that "works") masks this error so user space is not aware of it, hence why the second method works.

Quoted 3 lines
Quoted 3 lines

Thanks,

author maintainer
#5 lore raw

PATCH v2 2/5 macsec: delete new rxsc when offload fails

In reply to Sabrina Dubroca

Currently we get an inconsistent state:
- netlink returns the error to userspace
- the RXSC is installed but not offloaded

Then the device could get confused when we try to add an RXSA, because the RXSC isn't supposed to exist.

Fixes 3cf3227a21d1 ("net: macsec: hardware offloading infrastructure")
---
 drivers/net/macsec.c | 5 ++---
 1 file changed, 2 insertions(+), 3 deletions(-)
drivers/net/macsec.c +2 -3
index b3f76e8071f2..0d6fe34b91ae 100644
@@ -1876,7 +1876,6 @@ static int macsec_add_rxsc(struct sk_buff *skb, struct genl_info *info)
 	struct macsec_rx_sc *rx_sc;
 	struct nlattr *tb_rxsc[MACSEC_RXSC_ATTR_MAX + 1];
 	struct macsec_secy *secy;
-	bool was_active;
 	int ret;
 
 	if (!attrs[MACSEC_ATTR_IFINDEX])
@@ -1904,7 +1903,6 @@ static int macsec_add_rxsc(struct sk_buff *skb, struct genl_info *info)
 		return PTR_ERR(rx_sc);
 	}
 
-	was_active = rx_sc->active;
 	if (tb_rxsc[MACSEC_RXSC_ATTR_ACTIVE])
 		rx_sc->active = !!nla_get_u8(tb_rxsc[MACSEC_RXSC_ATTR_ACTIVE]);
 
@@ -1931,7 +1929,8 @@ static int macsec_add_rxsc(struct sk_buff *skb, struct genl_info *info)
 	return 0;
 
 cleanup:
-	rx_sc->active = was_active;
+	del_rx_sc(secy, sci);
+	free_rx_sc(rx_sc);
 	rtnl_unlock();
 	return ret;
 }
Signature · 1 line
2.38.0
Reviewed-by on v2 2/5
#6 lore raw

Re: [PATCH net v2 2/5] macsec: delete new rxsc when offload fails

Quoted 4 lines · Sabrina Dubroca

Currently we get an inconsistent state:
- netlink returns the error to userspace
- the RXSC is installed but not offloaded

Quoted 9 lines

Then the device could get confused when we try to add an RXSA, because the RXSC isn't supposed to exist.

Fixes 3cf3227a21d1 ("net: macsec: hardware offloading infrastructure")
---
 drivers/net/macsec.c | 5 ++---
 1 file changed, 2 insertions(+), 3 deletions(-)

Thanks,

author maintainer
#7 lore raw

PATCH v2 3/5 macsec: fix secy->n_rx_sc accounting

In reply to Sabrina Dubroca

secy->n_rx_sc is supposed to be the number of _active_ rxsc's within a secy. This is then used by macsec_send_sci to help decide if we should add the SCI to the header or not.

This logic is currently broken when we create a new RXSC and turn it off at creation, as create_rx_sc always sets ->active to true (and immediately uses that to increment n_rx_sc), and only later macsec_add_rxsc sets rx_sc->active.

Fixes c09440f7dcb3 ("macsec: introduce IEEE 802.1AE driver")

---
v2: drop unnecessary !! (Leon Romanovsky)

 drivers/net/macsec.c | 15 +++++++++------
 1 file changed, 9 insertions(+), 6 deletions(-)
drivers/net/macsec.c +9 -6
index 0d6fe34b91ae..48e2dd2d5778 100644
@@ -1413,7 +1413,8 @@ static struct macsec_rx_sc *del_rx_sc(struct macsec_secy *secy, sci_t sci)
 	return NULL;
 }
 
-static struct macsec_rx_sc *create_rx_sc(struct net_device *dev, sci_t sci)
+static struct macsec_rx_sc *create_rx_sc(struct net_device *dev, sci_t sci,
+					 bool active)
 {
 	struct macsec_rx_sc *rx_sc;
 	struct macsec_dev *macsec;
@@ -1437,7 +1438,7 @@ static struct macsec_rx_sc *create_rx_sc(struct net_device *dev, sci_t sci)
 	}
 
 	rx_sc->sci = sci;
-	rx_sc->active = true;
+	rx_sc->active = active;
 	refcount_set(&rx_sc->refcnt, 1);
 
 	secy = &macsec_priv(dev)->secy;
@@ -1876,6 +1877,7 @@ static int macsec_add_rxsc(struct sk_buff *skb, struct genl_info *info)
 	struct macsec_rx_sc *rx_sc;
 	struct nlattr *tb_rxsc[MACSEC_RXSC_ATTR_MAX + 1];
 	struct macsec_secy *secy;
+	bool active = true;
 	int ret;
 
 	if (!attrs[MACSEC_ATTR_IFINDEX])
@@ -1897,15 +1899,16 @@ static int macsec_add_rxsc(struct sk_buff *skb, struct genl_info *info)
 	secy = &macsec_priv(dev)->secy;
 	sci = nla_get_sci(tb_rxsc[MACSEC_RXSC_ATTR_SCI]);
 
-	rx_sc = create_rx_sc(dev, sci);
+
+	if (tb_rxsc[MACSEC_RXSC_ATTR_ACTIVE])
+		active = nla_get_u8(tb_rxsc[MACSEC_RXSC_ATTR_ACTIVE]);
+
+	rx_sc = create_rx_sc(dev, sci, active);
 	if (IS_ERR(rx_sc)) {
 		rtnl_unlock();
 		return PTR_ERR(rx_sc);
 	}
 
-	if (tb_rxsc[MACSEC_RXSC_ATTR_ACTIVE])
-		rx_sc->active = !!nla_get_u8(tb_rxsc[MACSEC_RXSC_ATTR_ACTIVE]);
-
 	if (macsec_is_offloaded(netdev_priv(dev))) {
 		const struct macsec_ops *ops;
 		struct macsec_context ctx;
Signature · 1 line
2.38.0
maintainer
#8 lore raw

Re: [PATCH net v2 3/5] macsec: fix secy->n_rx_sc accounting

drivers/net/macsec.c static int macsec_add_rxsc(struct sk_buff *skb, struct genl_info *info) · line 1903

Quoted 4 lines · On Wed, 26 Oct 2022 23:56:25 +0200 Sabrina Dubroca wrote:
drivers/net/macsec.c +2 -1
-	rx_sc = create_rx_sc(dev, sci);
+
+	if (tb_rxsc[MACSEC_RXSC_ATTR_ACTIVE])

nit: double new line :(

Reviewed-by on v2 3/5
#9 lore raw

Re: [PATCH net v2 3/5] macsec: fix secy->n_rx_sc accounting

In reply to Sabrina Dubroca
Quoted 9 lines · Sabrina Dubroca

secy->n_rx_sc is supposed to be the number of _active_ rxsc's within a secy. This is then used by macsec_send_sci to help decide if we should add the SCI to the header or not.

This logic is currently broken when we create a new RXSC and turn it off at creation, as create_rx_sc always sets ->active to true (and immediately uses that to increment n_rx_sc), and only later macsec_add_rxsc sets rx_sc->active.

Quoted 5 lines
Fixes c09440f7dcb3 ("macsec: introduce IEEE 802.1AE driver")

---
v2: drop unnecessary !! (Leon Romanovsky)

With comment from Jakub.

Thanks,

author maintainer
#10 lore raw

PATCH v2 4/5 macsec: fix detection of RXSCs when toggling offloading

In reply to Sabrina Dubroca

macsec_is_configured incorrectly uses secy->n_rx_sc to check if some RXSCs exist. secy->n_rx_sc only counts the number of active RXSCs, but there can also be inactive SCs as well, which may be stored in the driver (in case we're disabling offloading), or would have to be pushed to the device (in case we're trying to enable offloading).

As long as RXSCs active on creation and never turned off, the issue is not visible.

Fixes dcb780fb2795 ("net: macsec: add nla support for changing the offloading selection")
---
 drivers/net/macsec.c | 2 +-
 1 file changed, 1 insertion(+), 1 deletion(-)
drivers/net/macsec.c +1 -1
index 48e2dd2d5778..5a2c1bd65d89 100644
@@ -2572,7 +2572,7 @@ static bool macsec_is_configured(struct macsec_dev *macsec)
 	struct macsec_tx_sc *tx_sc = &secy->tx_sc;
 	int i;
 
-	if (secy->n_rx_sc > 0)
+	if (secy->rx_sc)
 		return true;
 
 	for (i = 0; i < MACSEC_NUM_AN; i++)
Signature · 1 line
2.38.0
Reviewed-by on v2 4/5
#11 lore raw

Re: [PATCH net v2 4/5] macsec: fix detection of RXSCs when toggling offloading

Quoted 6 lines · Sabrina Dubroca

macsec_is_configured incorrectly uses secy->n_rx_sc to check if some RXSCs exist. secy->n_rx_sc only counts the number of active RXSCs, but there can also be inactive SCs as well, which may be stored in the driver (in case we're disabling offloading), or would have to be pushed to the device (in case we're trying to enable offloading).

Quoted 9 lines

As long as RXSCs active on creation and never turned off, the issue is not visible.

Fixes dcb780fb2795 ("net: macsec: add nla support for changing the offloading selection")
---
 drivers/net/macsec.c | 2 +-
 1 file changed, 1 insertion(+), 1 deletion(-)

Thanks,

author maintainer
#12 lore raw

PATCH v2 5/5 macsec: clear encryption keys from the stack after setting up offload

In reply to Sabrina Dubroca

macsec_add_rxsa and macsec_add_txsa copy the key to an on-stack offloading context to pass it to the drivers, but leaves it there when it's done. Clear it with memzero_explicit as soon as it's not needed anymore.

Fixes 3cf3227a21d1 ("net: macsec: hardware offloading infrastructure")
---
 drivers/net/macsec.c | 2 ++
 1 file changed, 2 insertions(+)
drivers/net/macsec.c +2 -0
index 5a2c1bd65d89..7710d549012f 100644
@@ -1839,6 +1839,7 @@ static int macsec_add_rxsa(struct sk_buff *skb, struct genl_info *info)
 		       secy->key_len);
 
 		err = macsec_offload(ops->mdo_add_rxsa, &ctx);
+		memzero_explicit(ctx.sa.key, secy->key_len);
 		if (err)
 			goto cleanup;
 	}
@@ -2082,6 +2083,7 @@ static int macsec_add_txsa(struct sk_buff *skb, struct genl_info *info)
 		       secy->key_len);
 
 		err = macsec_offload(ops->mdo_add_txsa, &ctx);
+		memzero_explicit(ctx.sa.key, secy->key_len);
 		if (err)
 			goto cleanup;
 	}
Signature · 1 line
2.38.0
Reviewed-by on v2 5/5
#13 lore raw

Re: [PATCH net v2 5/5] macsec: clear encryption keys from the stack after setting up offload

Quoted 11 lines · Sabrina Dubroca

macsec_add_rxsa and macsec_add_txsa copy the key to an on-stack offloading context to pass it to the drivers, but leaves it there when it's done. Clear it with memzero_explicit as soon as it's not needed anymore.

Fixes 3cf3227a21d1 ("net: macsec: hardware offloading infrastructure")
---
 drivers/net/macsec.c | 2 ++
 1 file changed, 2 insertions(+)

Thanks,

Reviewed-by on the v2 series
#14 lore raw
In reply to Sabrina Dubroca

Quoting Sabrina Dubroca (2022-10-26 23:56:22)

Quoted 23 lines

I'm working on a dummy offload for macsec on netdevsim. It just has a small SecY and RXSC table so I can trigger failures easily on the ndo_* side. It has exposed a couple of issues.

The first patch is a revert of commit c850240b6c41 ("net: macsec:
report real_dev features when HW offloading is enabled"). That commit tried to improve the performance of macsec offload by taking advantage of some of the NIC's features, but in doing so, broke macsec offload when the lower device supports both macsec and ipsec offload, as the ipsec offload feature flags were copied from the real device. Since the macsec device doesn't provide xdo_* ops, the XFRM core rejects the registration of the new macsec device in xfrm_api_check.

I'm working on re-adding those feature flags when offload is available, but I haven't fully solved that yet. I think it would be safer to do that second part in net-next considering how complex feature interactions tend to be.

v2:
- better describe the issue introduced by commit c850240b6c41 (Leon Romanovsky)
- drop unnecessary !! (Leon Romanovsky)

Quoted 7 lines

Sabrina Dubroca (5):
Revert "net: macsec: report real_dev features when HW offloading is
enabled"
macsec: delete new rxsc when offload fails
macsec: fix secy->n_rx_sc accounting
macsec: fix detection of RXSCs when toggling offloading
macsec: clear encryption keys from the stack after setting up offload

Series,

Thanks!