Skip to contentSkip to conversation

netfilter: nftables_offload: set address type in control dissector

PATCH In review Pablo Neira Ayuso 2 participants 5 messages Nov 25, 2020 – Nov 26, 2020

Subsystems (2)
lore
Share & tools
lore

Pablo Neira Ayuso

author maintainer
#0 lore raw

PATCH v2 1/2 netfilter: nftables_offload: set address type in control dissector

This patch adds nft_flow_rule_set_addr_type() to set the address type from the nft_payload expression accordingly.

If the address type is not set in the control dissector then a rule that matches either source or destination IP address does not work.

After this patch, nft hardware offload generates the flow dissector configuration as tc-flower to match on an IP address.

This patch has been also tested functionally to make sure packets are filtered out by the NIC.

This is also getting the code aligned with the existing netfilter flow offload infrastructure which is also setting the control dissector.

Fixes c9626a2cbdb2 ("netfilter: nf_tables: add hardware offload support")

---
v2: add a nice patch description and remove unnecessary EXPORT_SYMBOL()
per Jakub Kicinski.

 include/net/netfilter/nf_tables_offload.h |  4 ++++
 net/netfilter/nf_tables_offload.c         | 17 +++++++++++++++++
 net/netfilter/nft_payload.c               |  4 ++++
 3 files changed, 25 insertions(+)
include/net/netfilter/nf_tables_offload.h +4 -0
index ea7d1d78b92d..bddd34c5bd79 100644
@@ -37,6 +37,7 @@ void nft_offload_update_dependency(struct nft_offload_ctx *ctx,
 
 struct nft_flow_key {
 	struct flow_dissector_key_basic			basic;
+	struct flow_dissector_key_control		control;
 	union {
 		struct flow_dissector_key_ipv4_addrs	ipv4;
 		struct flow_dissector_key_ipv6_addrs	ipv6;
@@ -62,6 +63,9 @@ struct nft_flow_rule {
 
 #define NFT_OFFLOAD_F_ACTION	(1 << 0)
 
+void nft_flow_rule_set_addr_type(struct nft_flow_rule *flow,
+				 enum flow_dissector_key_id addr_type);
+
 struct nft_rule;
 struct nft_flow_rule *nft_flow_rule_create(struct net *net, const struct nft_rule *rule);
 void nft_flow_rule_destroy(struct nft_flow_rule *flow);
net/netfilter/nf_tables_offload.c +17 -0
index 9f625724a20f..9ae14270c543 100644
@@ -28,6 +28,23 @@ static struct nft_flow_rule *nft_flow_rule_alloc(int num_actions)
 	return flow;
 }
 
+void nft_flow_rule_set_addr_type(struct nft_flow_rule *flow,
+				 enum flow_dissector_key_id addr_type)
+{
+	struct nft_flow_match *match = &flow->match;
+	struct nft_flow_key *mask = &match->mask;
+	struct nft_flow_key *key = &match->key;
+
+	if (match->dissector.used_keys & BIT(FLOW_DISSECTOR_KEY_CONTROL))
+		return;
+
+	key->control.addr_type = addr_type;
+	mask->control.addr_type = 0xffff;
+	match->dissector.used_keys |= BIT(FLOW_DISSECTOR_KEY_CONTROL);
+	match->dissector.offset[FLOW_DISSECTOR_KEY_CONTROL] =
+		offsetof(struct nft_flow_key, control);
+}
+
 struct nft_flow_rule *nft_flow_rule_create(struct net *net,
 					   const struct nft_rule *rule)
 {
net/netfilter/nft_payload.c +4 -0
index dcd3c7b8a367..bbf811d030d5 100644
@@ -244,6 +244,7 @@ static int nft_payload_offload_ip(struct nft_offload_ctx *ctx,
 
 		NFT_OFFLOAD_MATCH(FLOW_DISSECTOR_KEY_IPV4_ADDRS, ipv4, src,
 				  sizeof(struct in_addr), reg);
+		nft_flow_rule_set_addr_type(flow, FLOW_DISSECTOR_KEY_IPV4_ADDRS);
 		break;
 	case offsetof(struct iphdr, daddr):
 		if (priv->len != sizeof(struct in_addr))
@@ -251,6 +252,7 @@ static int nft_payload_offload_ip(struct nft_offload_ctx *ctx,
 
 		NFT_OFFLOAD_MATCH(FLOW_DISSECTOR_KEY_IPV4_ADDRS, ipv4, dst,
 				  sizeof(struct in_addr), reg);
+		nft_flow_rule_set_addr_type(flow, FLOW_DISSECTOR_KEY_IPV4_ADDRS);
 		break;
 	case offsetof(struct iphdr, protocol):
 		if (priv->len != sizeof(__u8))
@@ -280,6 +282,7 @@ static int nft_payload_offload_ip6(struct nft_offload_ctx *ctx,
 
 		NFT_OFFLOAD_MATCH(FLOW_DISSECTOR_KEY_IPV6_ADDRS, ipv6, src,
 				  sizeof(struct in6_addr), reg);
+		nft_flow_rule_set_addr_type(flow, FLOW_DISSECTOR_KEY_IPV6_ADDRS);
 		break;
 	case offsetof(struct ipv6hdr, daddr):
 		if (priv->len != sizeof(struct in6_addr))
@@ -287,6 +290,7 @@ static int nft_payload_offload_ip6(struct nft_offload_ctx *ctx,
 
 		NFT_OFFLOAD_MATCH(FLOW_DISSECTOR_KEY_IPV6_ADDRS, ipv6, dst,
 				  sizeof(struct in6_addr), reg);
+		nft_flow_rule_set_addr_type(flow, FLOW_DISSECTOR_KEY_IPV6_ADDRS);
 		break;
 	case offsetof(struct ipv6hdr, nexthdr):
 		if (priv->len != sizeof(__u8))
Signature · 1 line
2.20.1
author maintainer
#1 lore raw

PATCH v2 2/2 netfilter: nftables_offload: build mask based from the matching bytes

Userspace might match on prefix bytes of header fields if they are on the byte boundary, this requires that the mask is adjusted accordingly. Use NFT_OFFLOAD_MATCH_EXACT() for meta since prefix byte matching is not allowed for this type of selector.

The bitwise expression might be optimized out by userspace, hence the kernel need to infer the prefix from the number of payload bytes to match on. This patch adds nft_payload_offload_mask() to calculate the bitmask to match on the prefix.

Fixes c9626a2cbdb2 ("netfilter: nf_tables: add hardware offload support")

---
v2: make a nice commit description as well for this.

 include/net/netfilter/nf_tables_offload.h |  3 ++
 net/netfilter/nft_cmp.c                   |  8 +--
 net/netfilter/nft_meta.c                  | 16 +++---
 net/netfilter/nft_payload.c               | 66 +++++++++++++++++------
 4 files changed, 64 insertions(+), 29 deletions(-)
include/net/netfilter/nf_tables_offload.h +3 -0
index bddd34c5bd79..1d34fe154fe0 100644
@@ -78,6 +78,9 @@ int nft_flow_rule_offload_commit(struct net *net);
 		offsetof(struct nft_flow_key, __base.__field);		\
 	(__reg)->len		= __len;				\
 	(__reg)->key		= __key;				\
+
+#define NFT_OFFLOAD_MATCH_EXACT(__key, __base, __field, __len, __reg)	\
+	NFT_OFFLOAD_MATCH(__key, __base, __field, __len, __reg)		\
 	memset(&(__reg)->mask, 0xff, (__reg)->len);
 
 int nft_chain_offload_priority(struct nft_base_chain *basechain);
net/netfilter/nft_cmp.c +4 -4
index bc079d68a536..00e563a72d3d 100644
@@ -123,11 +123,11 @@ static int __nft_cmp_offload(struct nft_offload_ctx *ctx,
 	u8 *mask = (u8 *)&flow->match.mask;
 	u8 *key = (u8 *)&flow->match.key;
 
-	if (priv->op != NFT_CMP_EQ || reg->len != priv->len)
+	if (priv->op != NFT_CMP_EQ || priv->len > reg->len)
 		return -EOPNOTSUPP;
 
-	memcpy(key + reg->offset, &priv->data, priv->len);
-	memcpy(mask + reg->offset, &reg->mask, priv->len);
+	memcpy(key + reg->offset, &priv->data, reg->len);
+	memcpy(mask + reg->offset, &reg->mask, reg->len);
 
 	flow->match.dissector.used_keys |= BIT(reg->key);
 	flow->match.dissector.offset[reg->key] = reg->base_offset;
@@ -137,7 +137,7 @@ static int __nft_cmp_offload(struct nft_offload_ctx *ctx,
 	    nft_reg_load16(priv->data.data) != ARPHRD_ETHER)
 		return -EOPNOTSUPP;
 
-	nft_offload_update_dependency(ctx, &priv->data, priv->len);
+	nft_offload_update_dependency(ctx, &priv->data, reg->len);
 
 	return 0;
 }
net/netfilter/nft_meta.c +8 -8
index b37bd02448d8..bf4b3ad5314c 100644
@@ -724,22 +724,22 @@ static int nft_meta_get_offload(struct nft_offload_ctx *ctx,
 
 	switch (priv->key) {
 	case NFT_META_PROTOCOL:
-		NFT_OFFLOAD_MATCH(FLOW_DISSECTOR_KEY_BASIC, basic, n_proto,
-				  sizeof(__u16), reg);
+		NFT_OFFLOAD_MATCH_EXACT(FLOW_DISSECTOR_KEY_BASIC, basic, n_proto,
+					sizeof(__u16), reg);
 		nft_offload_set_dependency(ctx, NFT_OFFLOAD_DEP_NETWORK);
 		break;
 	case NFT_META_L4PROTO:
-		NFT_OFFLOAD_MATCH(FLOW_DISSECTOR_KEY_BASIC, basic, ip_proto,
-				  sizeof(__u8), reg);
+		NFT_OFFLOAD_MATCH_EXACT(FLOW_DISSECTOR_KEY_BASIC, basic, ip_proto,
+					sizeof(__u8), reg);
 		nft_offload_set_dependency(ctx, NFT_OFFLOAD_DEP_TRANSPORT);
 		break;
 	case NFT_META_IIF:
-		NFT_OFFLOAD_MATCH(FLOW_DISSECTOR_KEY_META, meta,
-				  ingress_ifindex, sizeof(__u32), reg);
+		NFT_OFFLOAD_MATCH_EXACT(FLOW_DISSECTOR_KEY_META, meta,
+					ingress_ifindex, sizeof(__u32), reg);
 		break;
 	case NFT_META_IIFTYPE:
-		NFT_OFFLOAD_MATCH(FLOW_DISSECTOR_KEY_META, meta,
-				  ingress_iftype, sizeof(__u16), reg);
+		NFT_OFFLOAD_MATCH_EXACT(FLOW_DISSECTOR_KEY_META, meta,
+					ingress_iftype, sizeof(__u16), reg);
 		break;
 	default:
 		return -EOPNOTSUPP;
net/netfilter/nft_payload.c +49 -17

174 lines not on this page · raw message

Signature · 1 line
2.20.1
maintainer
#2 lore raw
In reply to Pablo Neira Ayuso
Quoted 105 lines · On Thu, 26 Nov 2020 00:04:45 +0100 Pablo Neira Ayuso wrote:

This patch adds nft_flow_rule_set_addr_type() to set the address type from the nft_payload expression accordingly.

If the address type is not set in the control dissector then a rule that matches either source or destination IP address does not work.

After this patch, nft hardware offload generates the flow dissector configuration as tc-flower to match on an IP address.

This patch has been also tested functionally to make sure packets are filtered out by the NIC.

This is also getting the code aligned with the existing netfilter flow offload infrastructure which is also setting the control dissector.

Fixes c9626a2cbdb2 ("netfilter: nf_tables: add hardware offload support")

---
v2: add a nice patch description and remove unnecessary EXPORT_SYMBOL()
per Jakub Kicinski.

 include/net/netfilter/nf_tables_offload.h |  4 ++++
 net/netfilter/nf_tables_offload.c         | 17 +++++++++++++++++
 net/netfilter/nft_payload.c               |  4 ++++
 3 files changed, 25 insertions(+)
include/net/netfilter/nf_tables_offload.h +4 -0
index ea7d1d78b92d..bddd34c5bd79 100644
@@ -37,6 +37,7 @@ void nft_offload_update_dependency(struct nft_offload_ctx *ctx,
 
 struct nft_flow_key {
 	struct flow_dissector_key_basic			basic;
+	struct flow_dissector_key_control		control;
 	union {
 		struct flow_dissector_key_ipv4_addrs	ipv4;
 		struct flow_dissector_key_ipv6_addrs	ipv6;
@@ -62,6 +63,9 @@ struct nft_flow_rule {
 
 #define NFT_OFFLOAD_F_ACTION	(1 << 0)
 
+void nft_flow_rule_set_addr_type(struct nft_flow_rule *flow,
+				 enum flow_dissector_key_id addr_type);
+
 struct nft_rule;
 struct nft_flow_rule *nft_flow_rule_create(struct net *net, const struct nft_rule *rule);
 void nft_flow_rule_destroy(struct nft_flow_rule *flow);
net/netfilter/nf_tables_offload.c +17 -0
index 9f625724a20f..9ae14270c543 100644
@@ -28,6 +28,23 @@ static struct nft_flow_rule *nft_flow_rule_alloc(int num_actions)
 	return flow;
 }
 
+void nft_flow_rule_set_addr_type(struct nft_flow_rule *flow,
+				 enum flow_dissector_key_id addr_type)
+{
+	struct nft_flow_match *match = &flow->match;
+	struct nft_flow_key *mask = &match->mask;
+	struct nft_flow_key *key = &match->key;
+
+	if (match->dissector.used_keys & BIT(FLOW_DISSECTOR_KEY_CONTROL))
+		return;
+
+	key->control.addr_type = addr_type;
+	mask->control.addr_type = 0xffff;
+	match->dissector.used_keys |= BIT(FLOW_DISSECTOR_KEY_CONTROL);
+	match->dissector.offset[FLOW_DISSECTOR_KEY_CONTROL] =
+		offsetof(struct nft_flow_key, control);
+}
+
 struct nft_flow_rule *nft_flow_rule_create(struct net *net,
 					   const struct nft_rule *rule)
 {
net/netfilter/nft_payload.c +3 -0
index dcd3c7b8a367..bbf811d030d5 100644
@@ -244,6 +244,7 @@ static int nft_payload_offload_ip(struct nft_offload_ctx *ctx,
 
 		NFT_OFFLOAD_MATCH(FLOW_DISSECTOR_KEY_IPV4_ADDRS, ipv4, src,
 				  sizeof(struct in_addr), reg);
+		nft_flow_rule_set_addr_type(flow, FLOW_DISSECTOR_KEY_IPV4_ADDRS);
 		break;
 	case offsetof(struct iphdr, daddr):
 		if (priv->len != sizeof(struct in_addr))
@@ -251,6 +252,7 @@ static int nft_payload_offload_ip(struct nft_offload_ctx *ctx,
 
 		NFT_OFFLOAD_MATCH(FLOW_DISSECTOR_KEY_IPV4_ADDRS, ipv4, dst,
 				  sizeof(struct in_addr), reg);
+		nft_flow_rule_set_addr_type(flow, FLOW_DISSECTOR_KEY_IPV4_ADDRS);
 		break;
 	case offsetof(struct iphdr, protocol):
 		if (priv->len != sizeof(__u8))
@@ -280,6 +282,7 @@ static int nft_payload_offload_ip6(struct nft_offload_ctx *ctx,
 
 		NFT_OFFLOAD_MATCH(FLOW_DISSECTOR_KEY_IPV6_ADDRS, ipv6, src,
 				  sizeof(struct in6_addr), reg);
+		nft_flow_rule_set_addr_type(flow, FLOW_DISSECTOR_KEY_IPV6_ADDRS);
 		break;
 	case offsetof(struct ipv6hdr, daddr):
 		if (priv->len != sizeof(struct in6_addr))

net/netfilter/nft_payload.c static int nft_payload_offload_ip6(struct nft_offload_ctx *ctx, · line 296

Quoted 7 lines
net/netfilter/nft_payload.c +1 -0
@@ -287,6 +290,7 @@ static int nft_payload_offload_ip6(struct nft_offload_ctx *ctx,
 
 		NFT_OFFLOAD_MATCH(FLOW_DISSECTOR_KEY_IPV6_ADDRS, ipv6, dst,
 				  sizeof(struct in6_addr), reg);
+		nft_flow_rule_set_addr_type(flow, FLOW_DISSECTOR_KEY_IPV6_ADDRS);
 		break;
 	case offsetof(struct ipv6hdr, nexthdr):
 		if (priv->len != sizeof(__u8))

Still worries me this is done in a response to a match.

skb_flow_dissector_init() has a straight up BUG_ON() if the dissector did not set the CONTROL or BASIC. It says in the comment that both must be initialized. But nft does not call skb_flow_dissector_init()?

Are you 100% sure all cases will set CONTROL and BASIC now?

author maintainer
#3 lore raw

Hi Jakub,

Quoted 116 lines · Jakub Kicinski
Quoted 105 lines · On Thu, 26 Nov 2020 00:04:45 +0100 Pablo Neira Ayuso wrote:

This patch adds nft_flow_rule_set_addr_type() to set the address type from the nft_payload expression accordingly.

If the address type is not set in the control dissector then a rule that matches either source or destination IP address does not work.

After this patch, nft hardware offload generates the flow dissector configuration as tc-flower to match on an IP address.

This patch has been also tested functionally to make sure packets are filtered out by the NIC.

This is also getting the code aligned with the existing netfilter flow offload infrastructure which is also setting the control dissector.

Fixes c9626a2cbdb2 ("netfilter: nf_tables: add hardware offload support")

---
v2: add a nice patch description and remove unnecessary EXPORT_SYMBOL()
per Jakub Kicinski.

 include/net/netfilter/nf_tables_offload.h |  4 ++++
 net/netfilter/nf_tables_offload.c         | 17 +++++++++++++++++
 net/netfilter/nft_payload.c               |  4 ++++
 3 files changed, 25 insertions(+)
include/net/netfilter/nf_tables_offload.h +4 -0
index ea7d1d78b92d..bddd34c5bd79 100644
@@ -37,6 +37,7 @@ void nft_offload_update_dependency(struct nft_offload_ctx *ctx,
 
 struct nft_flow_key {
 	struct flow_dissector_key_basic			basic;
+	struct flow_dissector_key_control		control;
 	union {
 		struct flow_dissector_key_ipv4_addrs	ipv4;
 		struct flow_dissector_key_ipv6_addrs	ipv6;
@@ -62,6 +63,9 @@ struct nft_flow_rule {
 
 #define NFT_OFFLOAD_F_ACTION	(1 << 0)
 
+void nft_flow_rule_set_addr_type(struct nft_flow_rule *flow,
+				 enum flow_dissector_key_id addr_type);
+
 struct nft_rule;
 struct nft_flow_rule *nft_flow_rule_create(struct net *net, const struct nft_rule *rule);
 void nft_flow_rule_destroy(struct nft_flow_rule *flow);
net/netfilter/nf_tables_offload.c +17 -0
index 9f625724a20f..9ae14270c543 100644
@@ -28,6 +28,23 @@ static struct nft_flow_rule *nft_flow_rule_alloc(int num_actions)
 	return flow;
 }
 
+void nft_flow_rule_set_addr_type(struct nft_flow_rule *flow,
+				 enum flow_dissector_key_id addr_type)
+{
+	struct nft_flow_match *match = &flow->match;
+	struct nft_flow_key *mask = &match->mask;
+	struct nft_flow_key *key = &match->key;
+
+	if (match->dissector.used_keys & BIT(FLOW_DISSECTOR_KEY_CONTROL))
+		return;
+
+	key->control.addr_type = addr_type;
+	mask->control.addr_type = 0xffff;
+	match->dissector.used_keys |= BIT(FLOW_DISSECTOR_KEY_CONTROL);
+	match->dissector.offset[FLOW_DISSECTOR_KEY_CONTROL] =
+		offsetof(struct nft_flow_key, control);
+}
+
 struct nft_flow_rule *nft_flow_rule_create(struct net *net,
 					   const struct nft_rule *rule)
 {
net/netfilter/nft_payload.c +3 -0
index dcd3c7b8a367..bbf811d030d5 100644
@@ -244,6 +244,7 @@ static int nft_payload_offload_ip(struct nft_offload_ctx *ctx,
 
 		NFT_OFFLOAD_MATCH(FLOW_DISSECTOR_KEY_IPV4_ADDRS, ipv4, src,
 				  sizeof(struct in_addr), reg);
+		nft_flow_rule_set_addr_type(flow, FLOW_DISSECTOR_KEY_IPV4_ADDRS);
 		break;
 	case offsetof(struct iphdr, daddr):
 		if (priv->len != sizeof(struct in_addr))
@@ -251,6 +252,7 @@ static int nft_payload_offload_ip(struct nft_offload_ctx *ctx,
 
 		NFT_OFFLOAD_MATCH(FLOW_DISSECTOR_KEY_IPV4_ADDRS, ipv4, dst,
 				  sizeof(struct in_addr), reg);
+		nft_flow_rule_set_addr_type(flow, FLOW_DISSECTOR_KEY_IPV4_ADDRS);
 		break;
 	case offsetof(struct iphdr, protocol):
 		if (priv->len != sizeof(__u8))
@@ -280,6 +282,7 @@ static int nft_payload_offload_ip6(struct nft_offload_ctx *ctx,
 
 		NFT_OFFLOAD_MATCH(FLOW_DISSECTOR_KEY_IPV6_ADDRS, ipv6, src,
 				  sizeof(struct in6_addr), reg);
+		nft_flow_rule_set_addr_type(flow, FLOW_DISSECTOR_KEY_IPV6_ADDRS);
 		break;
 	case offsetof(struct ipv6hdr, daddr):
 		if (priv->len != sizeof(struct in6_addr))

net/netfilter/nft_payload.c static int nft_payload_offload_ip6(struct nft_offload_ctx *ctx, · line 296

Quoted 7 lines
net/netfilter/nft_payload.c +1 -0
@@ -287,6 +290,7 @@ static int nft_payload_offload_ip6(struct nft_offload_ctx *ctx,
 
 		NFT_OFFLOAD_MATCH(FLOW_DISSECTOR_KEY_IPV6_ADDRS, ipv6, dst,
 				  sizeof(struct in6_addr), reg);
+		nft_flow_rule_set_addr_type(flow, FLOW_DISSECTOR_KEY_IPV6_ADDRS);
 		break;
 	case offsetof(struct ipv6hdr, nexthdr):
 		if (priv->len != sizeof(__u8))

Still worries me this is done in a response to a match.

Quoted 5 lines

skb_flow_dissector_init() has a straight up BUG_ON() if the dissector did not set the CONTROL or BASIC. It says in the comment that both must be initialized. But nft does not call skb_flow_dissector_init()?

Are you 100% sure all cases will set CONTROL and BASIC now?

Enforcing skb_flow_dissector_init() for software make sense, but in Netfilter this is used for hardware offload only.

All drivers in the tree check for:

if (flow_rule_match_key(rule, FLOW_DISSECTOR_KEY_CONTROL))

before accessing struct flow_match_control.

I can set it on inconditionally, but the driver will get a value 0x0 and mask 0x0, which is the same as leaving it unset.

maintainer
#4 lore raw
Quoted 11 lines · On Thu, 26 Nov 2020 01:44:37 +0100 Pablo Neira Ayuso wrote:
Quoted 7 lines

Still worries me this is done in a response to a match.

skb_flow_dissector_init() has a straight up BUG_ON() if the dissector did not set the CONTROL or BASIC. It says in the comment that both must be initialized. But nft does not call skb_flow_dissector_init()?

Are you 100% sure all cases will set CONTROL and BASIC now?

Enforcing skb_flow_dissector_init() for software make sense, but in Netfilter this is used for hardware offload only.

Quoted 8 lines

All drivers in the tree check for:

if (flow_rule_match_key(rule, FLOW_DISSECTOR_KEY_CONTROL))

before accessing struct flow_match_control.

I can set it on inconditionally, but the driver will get a value 0x0 and mask 0x0, which is the same as leaving it unset.

Ack, I didn't realize you don't actually ever use the dissector other than for offload.