Skip to contentSkip to conversation

nft: Optimize class-based IP prefix matches

PATCH Superseded Phil Sutter 3 participants 7 messages Oct 2, 2020 – Oct 6, 2020 netfilter

v1 of 2 Superseded Latest is v2 →

lore
Share & tools
lore

Phil Sutter

author reviewer
#0 lore raw

PATCH 1/1 nft: Optimize class-based IP prefix matches

Payload expression works on byte-boundaries, leverage this with suitable prefix lengths.

---
 iptables/nft-ipv4.c   |  6 ++++--
 iptables/nft-ipv6.c   |  6 ++++--
 iptables/nft-shared.c | 14 ++++++++++----
 iptables/nft-shared.h |  4 ++++
 4 files changed, 22 insertions(+), 8 deletions(-)
iptables/nft-ipv4.c +4 -2
index afdecf9711e64..ce702041af0f4 100644
@@ -199,7 +199,8 @@ static void nft_ipv4_parse_payload(struct nft_xt_ctx *ctx,
 			parse_mask_ipv4(ctx, &cs->fw.ip.smsk);
 			ctx->flags &= ~NFT_XT_CTX_BITWISE;
 		} else {
-			cs->fw.ip.smsk.s_addr = 0xffffffff;
+			memset(&cs->fw.ip.smsk, 0xff,
+			       min(ctx->payload.len, sizeof(struct in_addr)));
 		}
 
 		if (inv)
@@ -212,7 +213,8 @@ static void nft_ipv4_parse_payload(struct nft_xt_ctx *ctx,
 			parse_mask_ipv4(ctx, &cs->fw.ip.dmsk);
 			ctx->flags &= ~NFT_XT_CTX_BITWISE;
 		} else {
-			cs->fw.ip.dmsk.s_addr = 0xffffffff;
+			memset(&cs->fw.ip.dmsk, 0xff,
+			       min(ctx->payload.len, sizeof(struct in_addr)));
 		}
 
 		if (inv)
iptables/nft-ipv6.c +4 -2
index 4008b7eab4f2a..c877ec6d10887 100644
@@ -146,7 +146,8 @@ static void nft_ipv6_parse_payload(struct nft_xt_ctx *ctx,
 			parse_mask_ipv6(ctx, &cs->fw6.ipv6.smsk);
 			ctx->flags &= ~NFT_XT_CTX_BITWISE;
 		} else {
-			memset(&cs->fw6.ipv6.smsk, 0xff, sizeof(struct in6_addr));
+			memset(&cs->fw6.ipv6.smsk, 0xff,
+			       min(ctx->payload.len, sizeof(struct in6_addr)));
 		}
 
 		if (inv)
@@ -159,7 +160,8 @@ static void nft_ipv6_parse_payload(struct nft_xt_ctx *ctx,
 			parse_mask_ipv6(ctx, &cs->fw6.ipv6.dmsk);
 			ctx->flags &= ~NFT_XT_CTX_BITWISE;
 		} else {
-			memset(&cs->fw6.ipv6.dmsk, 0xff, sizeof(struct in6_addr));
+			memset(&cs->fw6.ipv6.dmsk, 0xff,
+			       min(ctx->payload.len, sizeof(struct in6_addr)));
 		}
 
 		if (inv)
iptables/nft-shared.c +10 -4
index 7741d23befc5a..545e9c60fa015 100644
@@ -166,16 +166,22 @@ void add_addr(struct nftnl_rule *r, int offset,
 	      void *data, void *mask, size_t len, uint32_t op)
 {
 	const unsigned char *m = mask;
+	bool bitwise = false;
 	int i;
 
-	add_payload(r, offset, len, NFT_PAYLOAD_NETWORK_HEADER);
-
 	for (i = 0; i < len; i++) {
-		if (m[i] != 0xff)
+		if (m[i] != 0xff) {
+			bitwise = m[i] != 0;
 			break;
+		}
 	}
 
-	if (i != len)
+	if (!bitwise)
+		len = i;
+
+	add_payload(r, offset, len, NFT_PAYLOAD_NETWORK_HEADER);
+
+	if (bitwise)
 		add_bitwise(r, mask, len);
 
 	add_cmp_ptr(r, op, data, len);
iptables/nft-shared.h +4 -0
index 4440fd17bfeac..a52463342b30a 100644
@@ -247,4 +247,8 @@ void xtables_restore_parse(struct nft_handle *h,
 			   const struct nft_xt_restore_parse *p);
 
 void nft_check_xt_legacy(int family, bool is_ipt_save);
+
+#define min(x, y) ((x) < (y) ? (x) : (y))
+#define max(x, y) ((x) > (y) ? (x) : (y))
+
 #endif
Signature · 1 line
2.28.0
#1 lore raw
Quoted 1 line · Phil Sutter

+#define max(x, y) ((x) > (y) ? (x) : (y))

This is unused, right? not sure if on purpose.

author reviewer
#2 lore raw
Quoted 4 lines · Arturo Borrero Gonzalez
Quoted 1 line · Phil Sutter

+#define max(x, y) ((x) > (y) ? (x) : (y))

This is unused, right? not sure if on purpose.

Ah yes, I felt like adding it while being at it. Fine with you or should I drop it?

Thanks for reviewing!

Phil

#3 lore raw
In reply to Phil Sutter
Quoted 2 lines · Phil Sutter

Payload expression works on byte-boundaries, leverage this with suitable prefix lengths.

Interesing. But it kicks in the raw payload expression in nftables.

# nft list ruleset
table ip filter {
        chain INPUT {
                type filter hook input priority filter; policy accept;
                @nh,96,24 8323072 counter packets 0 bytes 0
        }

Would you send a patch for nftables too? There is already approximate offset matching in the tree, it should not be too hard to amend.

Thanks.

author reviewer
#4 lore raw
Quoted 13 lines · Pablo Neira Ayuso
Quoted 2 lines · Phil Sutter

Payload expression works on byte-boundaries, leverage this with suitable prefix lengths.

Interesing. But it kicks in the raw payload expression in nftables.

# nft list ruleset
table ip filter {
        chain INPUT {
                type filter hook input priority filter; policy accept;
                @nh,96,24 8323072 counter packets 0 bytes 0
        }
Quoted 2 lines

Would you send a patch for nftables too? There is already approximate offset matching in the tree, it should not be too hard to amend.

I had a quick look but it didn't seem trivial to me. It is in payload_expr_complete() where a template lookup happens based on expression offset and length which fails due to the unexpected length. Is this the right place to adjust or am I wrong?

Strictly speaking, this is just a lack of feature in nftables and nothing breaks due to it. Do you still want to block the iptables change for it?

Cheers, Phil

#5 lore raw
Quoted 17 lines · Phil Sutter
Quoted 13 lines · Pablo Neira Ayuso
Quoted 2 lines · Phil Sutter

Payload expression works on byte-boundaries, leverage this with suitable prefix lengths.

Interesing. But it kicks in the raw payload expression in nftables.

# nft list ruleset
table ip filter {
        chain INPUT {
                type filter hook input priority filter; policy accept;
                @nh,96,24 8323072 counter packets 0 bytes 0
        }
Quoted 2 lines

Would you send a patch for nftables too? There is already approximate offset matching in the tree, it should not be too hard to amend.

Quoted 8 lines

I had a quick look but it didn't seem trivial to me. It is in payload_expr_complete() where a template lookup happens based on expression offset and length which fails due to the unexpected length. Is this the right place to adjust or am I wrong?

Strictly speaking, this is just a lack of feature in nftables and nothing breaks due to it. Do you still want to block the iptables change for it?

Not block. Just get things aligned. This is a bit of a step back in the integration between iptables-nft and nft IMO.

I will have a look.

author reviewer
#6 lore raw
Quoted 23 lines · Pablo Neira Ayuso
Quoted 22 lines · Phil Sutter
Quoted 13 lines · Pablo Neira Ayuso
Quoted 2 lines · Phil Sutter

Payload expression works on byte-boundaries, leverage this with suitable prefix lengths.

Interesing. But it kicks in the raw payload expression in nftables.

# nft list ruleset
table ip filter {
        chain INPUT {
                type filter hook input priority filter; policy accept;
                @nh,96,24 8323072 counter packets 0 bytes 0
        }
Quoted 2 lines

Would you send a patch for nftables too? There is already approximate offset matching in the tree, it should not be too hard to amend.

I had a quick look but it didn't seem trivial to me. It is in payload_expr_complete() where a template lookup happens based on expression offset and length which fails due to the unexpected length. Is this the right place to adjust or am I wrong?

Quoted 6 lines
Quoted 3 lines

Strictly speaking, this is just a lack of feature in nftables and nothing breaks due to it. Do you still want to block the iptables change for it?

Not block. Just get things aligned. This is a bit of a step back in the integration between iptables-nft and nft IMO.

Well yes, it takes iptables-nft ahead in that regard. We should implement the same flexible payload size matching in nftables, too.

Quoted 1 line

I will have a look.

Thanks!