Skip to contentSkip to conversation

nftables: ct: add byte/packet counter support

PATCH In review Florian Westphal 2 participants 6 messages Jan 7, 2016 – Jan 8, 2016

Subsystems (2)
lore
Share & tools
lore

Florian Westphal

author maintainer
#0 lore raw

PATCH 1/1 nftables: ct: add byte/packet counter support

If the accounting extension isn't present, we'll return a counter value of 0.

---
WARNING: It doesn't work, but AFAIU the bug is in nft_cmp which doesn't work with u64 and gt/lt test. Following is true after 1st packet is sent with enabled accounting:

nft add rule filter input ct original packets gt 10

Seems like it only works for mark, skuid etc because those are u32 and thus use the _fast_ops version.

include/uapi/linux/netfilter/nf_tables.h +2 -0
index 22043ce..535576e 100644
@@ -755,6 +755,8 @@ enum nft_ct_keys {
 	NFT_CT_PROTO_SRC,
 	NFT_CT_PROTO_DST,
 	NFT_CT_LABELS,
+	NFT_CT_PKTS,
+	NFT_CT_BYTES,
 };
 
 /**
net/netfilter/nft_ct.c +39 -0
index 8cbca34..dc7a3c6 100644
@@ -16,6 +16,7 @@
 #include <linux/netfilter/nf_tables.h>
 #include <net/netfilter/nf_tables.h>
 #include <net/netfilter/nf_conntrack.h>
+#include <net/netfilter/nf_conntrack_acct.h>
 #include <net/netfilter/nf_conntrack_tuple.h>
 #include <net/netfilter/nf_conntrack_helper.h>
 #include <net/netfilter/nf_conntrack_ecache.h>
@@ -30,6 +31,19 @@ struct nft_ct {
 	};
 };
 
+static u64 nft_ct_get_eval_counter(const struct nf_conn_counter *c,
+				   enum nft_ct_keys k,
+				   enum ip_conntrack_dir d)
+{
+	if (d == IP_CT_DIR_ORIGINAL ||
+	    d == IP_CT_DIR_REPLY)
+		return k == NFT_CT_BYTES ? atomic64_read(&c[d].bytes) :
+					   atomic64_read(&c[d].packets);
+
+	return nft_ct_get_eval_counter(c, k, IP_CT_DIR_ORIGINAL) +
+	       nft_ct_get_eval_counter(c, k, IP_CT_DIR_REPLY);
+}
+
 static void nft_ct_get_eval(const struct nft_expr *expr,
 			    struct nft_regs *regs,
 			    const struct nft_pktinfo *pkt)
@@ -114,6 +128,17 @@ static void nft_ct_get_eval(const struct nft_expr *expr,
 			       NF_CT_LABELS_MAX_SIZE - size);
 		return;
 	}
+	case NFT_CT_BYTES: /* fallthrough */
+	case NFT_CT_PKTS: {
+		const struct nf_conn_acct *acct = nf_conn_acct_find(ct);
+		u64 count = 0;
+
+		if (acct)
+			count = nft_ct_get_eval_counter(acct->counter,
+							priv->key, priv->dir);
+		memcpy(dest, &count, sizeof(count));
+		return;
+	}
 #endif
 	default:
 		break;
@@ -291,6 +316,13 @@ static int nft_ct_get_init(const struct nft_ctx *ctx,
 			return -EINVAL;
 		len = FIELD_SIZEOF(struct nf_conntrack_tuple, src.u.all);
 		break;
+	case NFT_CT_BYTES:
+	case NFT_CT_PKTS:
+		/* no direction? return sum of original + reply */
+		if (tb[NFTA_CT_DIRECTION] == NULL)
+			priv->dir = IP_CT_DIR_MAX;
+		len = sizeof(u64);
+		break;
 	default:
 		return -EOPNOTSUPP;
 	}
@@ -366,6 +398,13 @@ static int nft_ct_get_dump(struct sk_buff *skb, const struct nft_expr *expr)
 		goto nla_put_failure;
 
 	switch (priv->key) {
+	case NFT_CT_BYTES:
+	case NFT_CT_PKTS:
+		if ((priv->dir == IP_CT_DIR_ORIGINAL ||
+		     priv->dir == IP_CT_DIR_REPLY) &&
+		     nla_put_u8(skb, NFTA_CT_DIRECTION, priv->dir))
+			goto nla_put_failure;
+		break;
 	case NFT_CT_PROTOCOL:
 	case NFT_CT_SRC:
 	case NFT_CT_DST:
Signature · 1 line
2.4.10
author maintainer
#1 lore raw
Quoted 9 lines · Florian Westphal

If the accounting extension isn't present, we'll return a counter value of 0.

---
WARNING: It doesn't work, but AFAIU the bug is in nft_cmp which doesn't work with u64 and gt/lt test. Following is true after 1st packet is sent with enabled accounting:

Quoted 4 lines

nft add rule filter input ct original packets gt 10

Seems like it only works for mark, skuid etc because those are u32 and thus use the _fast_ops version.

I take that back -- nft_cmp is fine, the culprit is lack of cpu_to_be64 in nft. This works fine with a local patch to nft + nft_byteorder.c.

Problem was nft generated two htonl() calls for upper and lower half of the counter which then makes nft_cmp behave a bit random ;)

I'll submit fixes soon.

author maintainer
#2 lore raw
In reply to Florian Westphal

net/netfilter/nft_ct.c static void nft_ct_get_eval(const struct nft_expr *expr, · line 139

Quoted 9 lines · Florian Westphal
net/netfilter/nft_ct.c +9 -0
+	case NFT_CT_BYTES: /* fallthrough */
+	case NFT_CT_PKTS: {
+		const struct nf_conn_acct *acct = nf_conn_acct_find(ct);
+		u64 count = 0;
+
+		if (acct)
+			count = nft_ct_get_eval_counter(acct->counter,
+							priv->key, priv->dir);
+		memcpy(dest, &count, sizeof(count));

Seems this needs to be a put_unaligned(). V2 coming soon.

author maintainer
#3 lore raw
Quoted 12 lines · Florian Westphal

net/netfilter/nft_ct.c static void nft_ct_get_eval(const struct nft_expr *expr, · line 139

Quoted 9 lines · Florian Westphal
net/netfilter/nft_ct.c +9 -0
+	case NFT_CT_BYTES: /* fallthrough */
+	case NFT_CT_PKTS: {
+		const struct nf_conn_acct *acct = nf_conn_acct_find(ct);
+		u64 count = 0;
+
+		if (acct)
+			count = nft_ct_get_eval_counter(acct->counter,
+							priv->key, priv->dir);
+		memcpy(dest, &count, sizeof(count));

Seems this needs to be a put_unaligned(). V2 coming soon.

Ahem. No. This is fine.
Seems I need to rest, I'll send the nft_byteorder patch tomorrow.

Sorry.

#4 lore raw
In reply to Florian Westphal

net/netfilter/nft_ct.c static int nft_ct_get_dump(struct sk_buff *skb, const struct nft_expr *expr) · line 408

Quoted 12 lines · Florian Westphal
net/netfilter/nft_ct.c +7 -0
@@ -366,6 +398,13 @@ static int nft_ct_get_dump(struct sk_buff *skb, const struct nft_expr *expr)
 		goto nla_put_failure;
 
 	switch (priv->key) {
+	case NFT_CT_BYTES:
+	case NFT_CT_PKTS:
+		if ((priv->dir == IP_CT_DIR_ORIGINAL ||
+		     priv->dir == IP_CT_DIR_REPLY) &&
+		     nla_put_u8(skb, NFTA_CT_DIRECTION, priv->dir))
+			goto nla_put_failure;
+		break;
 	case NFT_CT_PROTOCOL:

Any concern if I fold this change into your patch? It just simplifies this check, see below:

        switch (priv->key) {
        case NFT_CT_BYTES:
        case NFT_CT_PKTS:

- if ((priv->dir == IP_CT_DIR_ORIGINAL ||
- priv->dir == IP_CT_DIR_REPLY) &&
- nla_put_u8(skb, NFTA_CT_DIRECTION, priv->dir))
+ if (priv->dir < IP_CT_DIR_MAX &&
+ nla_put_u8(skb, NFTA_CT_DIRECTION, priv->dir))

                        goto nla_put_failure;
                break;
        case NFT_CT_PROTOCOL:

Thanks.

author maintainer
#5 lore raw
Quoted 20 lines · Pablo Neira Ayuso

net/netfilter/nft_ct.c static int nft_ct_get_dump(struct sk_buff *skb, const struct nft_expr *expr) · line 408

Quoted 12 lines · Florian Westphal
net/netfilter/nft_ct.c +7 -0
@@ -366,6 +398,13 @@ static int nft_ct_get_dump(struct sk_buff *skb, const struct nft_expr *expr)
 		goto nla_put_failure;
 
 	switch (priv->key) {
+	case NFT_CT_BYTES:
+	case NFT_CT_PKTS:
+		if ((priv->dir == IP_CT_DIR_ORIGINAL ||
+		     priv->dir == IP_CT_DIR_REPLY) &&
+		     nla_put_u8(skb, NFTA_CT_DIRECTION, priv->dir))
+			goto nla_put_failure;
+		break;
 	case NFT_CT_PROTOCOL:

Any concern if I fold this change into your patch? It just simplifies this check, see below:

        switch (priv->key) {
        case NFT_CT_BYTES:
        case NFT_CT_PKTS:

Review

Quoted 6 lines
(quoted diff) +2 -3
-               if ((priv->dir == IP_CT_DIR_ORIGINAL ||
-                    priv->dir == IP_CT_DIR_REPLY) &&
-                    nla_put_u8(skb, NFTA_CT_DIRECTION, priv->dir))
+               if (priv->dir < IP_CT_DIR_MAX &&
+                   nla_put_u8(skb, NFTA_CT_DIRECTION, priv->dir))
                        goto nla_put_failure;

No, go ahead.