Skip to contentSkip to conversation

netfilter: nftables: add connlabel set support

PATCH Superseded Florian Westphal 2 participants 13 messages Apr 12, 2016 – Apr 18, 2016

Subsystems (3)

v5 of 7 Superseded Latest is v7 →

lore
Share & tools
lore

Florian Westphal

author maintainer
#0 lore raw

v5 cover netfilter: nftables: add connlabel set support

Hi.

This is round 5 of the connlabel set support set.
I'm only sending the kernel patches for now.

First 4 patches are preparation changes, patch #4 adds set support. I added a more generic CT_IMM nested attr that expects a nft_data struct.

Its up to the kernel to (using the key) to figure out how to interpret it. This approach is hopefully generic enough so it can be re-used for other set options that want to use an immediate value.

Florian Westphal (4):
netfilter: connlabels: move helpers to xt_connlabel
netfilter: labels: don't emit ct event if labels were not changed
netfilter: connlabels: change nf_connlabels_get bit arg to 'highest used'
netfilter: nftables: add connlabel set support

 include/net/netfilter/nf_conntrack_labels.h |    5 -
 include/uapi/linux/netfilter/nf_tables.h    |    2 
 net/netfilter/nf_conntrack_labels.c         |   44 +++++----------
 net/netfilter/nft_ct.c                      |   78 ++++++++++++++++++++++++++--
 net/netfilter/xt_connlabel.c                |   14 ++++-
 net/openvswitch/conntrack.c                 |    2 
 6 files changed, 108 insertions(+), 37 deletions(-)
author maintainer
#1 lore raw

PATCH v5 1/4 netfilter: connlabels: move helpers to xt_connlabel

Currently labels can only be set either by iptables connlabel match or via ctnetlink.

Before adding nftables set support, clean up the clabel core and move helpers that nft will not need after all to the xtables module.

---
No changes since last iteration.

 include/net/netfilter/nf_conntrack_labels.h |  1 -
 net/netfilter/nf_conntrack_labels.c         | 19 +------------------
 net/netfilter/xt_connlabel.c                | 12 +++++++++++-
 3 files changed, 12 insertions(+), 20 deletions(-)
include/net/netfilter/nf_conntrack_labels.h +0 -1
index 7e2b1d0..5167818 100644
@@ -45,7 +45,6 @@ static inline struct nf_conn_labels *nf_ct_labels_ext_add(struct nf_conn *ct)
 #endif
 }
 
-bool nf_connlabel_match(const struct nf_conn *ct, u16 bit);
 int nf_connlabel_set(struct nf_conn *ct, u16 bit);
 
 int nf_connlabels_replace(struct nf_conn *ct,
net/netfilter/nf_conntrack_labels.c +1 -18
index 3ce5c31..3a30900 100644
@@ -16,28 +16,11 @@
 
 static spinlock_t nf_connlabels_lock;
 
-static unsigned int label_bits(const struct nf_conn_labels *l)
-{
-	unsigned int longs = l->words;
-	return longs * BITS_PER_LONG;
-}
-
-bool nf_connlabel_match(const struct nf_conn *ct, u16 bit)
-{
-	struct nf_conn_labels *labels = nf_ct_labels_find(ct);
-
-	if (!labels)
-		return false;
-
-	return bit < label_bits(labels) && test_bit(bit, labels->bits);
-}
-EXPORT_SYMBOL_GPL(nf_connlabel_match);
-
 int nf_connlabel_set(struct nf_conn *ct, u16 bit)
 {
 	struct nf_conn_labels *labels = nf_ct_labels_find(ct);
 
-	if (!labels || bit >= label_bits(labels))
+	if (!labels || BIT_WORD(bit) >= labels->words)
 		return -ENOSPC;
 
 	if (test_bit(bit, labels->bits))
net/netfilter/xt_connlabel.c +11 -1
index bb9cbeb..d9b3e53 100644
@@ -18,6 +18,16 @@ MODULE_DESCRIPTION("Xtables: add/match connection trackling labels");
 MODULE_ALIAS("ipt_connlabel");
 MODULE_ALIAS("ip6t_connlabel");
 
+static bool connlabel_match(const struct nf_conn *ct, u16 bit)
+{
+	struct nf_conn_labels *labels = nf_ct_labels_find(ct);
+
+	if (!labels)
+		return false;
+
+	return BIT_WORD(bit) < labels->words && test_bit(bit, labels->bits);
+}
+
 static bool
 connlabel_mt(const struct sk_buff *skb, struct xt_action_param *par)
 {
@@ -33,7 +43,7 @@ connlabel_mt(const struct sk_buff *skb, struct xt_action_param *par)
 	if (info->options & XT_CONNLABEL_OP_SET)
 		return (nf_connlabel_set(ct, info->bit) == 0) ^ invert;
 
-	return nf_connlabel_match(ct, info->bit) ^ invert;
+	return connlabel_match(ct, info->bit) ^ invert;
 }
 
 static int connlabel_mt_check(const struct xt_mtchk_param *par)
Signature · 1 line
2.7.3
author maintainer
#2 lore raw

PATCH v5 2/4 netfilter: labels: don't emit ct event if labels were not changed

In reply to Florian Westphal

make the replace function only send a ctnetlink event if the contents of the new set is different.

Otherwise 'ct label set ct label | bar'

will cause netlink event storm since we "replace" labels for each packet.

---
No changes since last version.

 net/netfilter/nf_conntrack_labels.c | 16 ++++++++++------
 1 file changed, 10 insertions(+), 6 deletions(-)
net/netfilter/nf_conntrack_labels.c +10 -6
index 3a30900..bd7f26b 100644
@@ -33,14 +33,18 @@ int nf_connlabel_set(struct nf_conn *ct, u16 bit)
 }
 EXPORT_SYMBOL_GPL(nf_connlabel_set);
 
-static void replace_u32(u32 *address, u32 mask, u32 new)
+static int replace_u32(u32 *address, u32 mask, u32 new)
 {
 	u32 old, tmp;
 
 	do {
 		old = *address;
 		tmp = (old & mask) ^ new;
+		if (old == tmp)
+			return 0;
 	} while (cmpxchg(address, old, tmp) != old);
+
+	return 1;
 }
 
 int nf_connlabels_replace(struct nf_conn *ct,
@@ -49,6 +53,7 @@ int nf_connlabels_replace(struct nf_conn *ct,
 {
 	struct nf_conn_labels *labels;
 	unsigned int size, i;
+	int changed = 0;
 	u32 *dst;
 
 	labels = nf_ct_labels_find(ct);
@@ -60,16 +65,15 @@ int nf_connlabels_replace(struct nf_conn *ct,
 		words32 = size / sizeof(u32);
 
 	dst = (u32 *) labels->bits;
-	if (words32) {
-		for (i = 0; i < words32; i++)
-			replace_u32(&dst[i], mask ? ~mask[i] : 0, data[i]);
-	}
+	for (i = 0; i < words32; i++)
+		changed |= replace_u32(&dst[i], mask ? ~mask[i] : 0, data[i]);
 
 	size /= sizeof(u32);
 	for (i = words32; i < size; i++) /* pad */
 		replace_u32(&dst[i], 0, 0);
 
-	nf_conntrack_event_cache(IPCT_LABEL, ct);
+	if (changed)
+		nf_conntrack_event_cache(IPCT_LABEL, ct);
 	return 0;
 }
 EXPORT_SYMBOL_GPL(nf_connlabels_replace);
Signature · 1 line
2.7.3
author maintainer
#3 lore raw

PATCH v5 3/4 netfilter: connlabels: change nf_connlabels_get bit arg to 'highest used'

In reply to Florian Westphal

nf_connlabel_set() takes the bit number that we would like to set. nf_connlabels_get() however took the number of bits that we want to support.

So e.g. nf_connlabels_get(32) support bits 0 to 31, but not 32. This changes nf_connlabels_get() to take the highest bit that we want to set.

Callers then don't have to cope with a potential integer wrap when using nf_connlabels_get(bit + 1) anymore.

Current callers are fine, this change is only to make folloup nft ct label set support simpler.

---
No changes since last version.

 include/net/netfilter/nf_conntrack_labels.h | 4 ++--
 net/netfilter/nf_conntrack_labels.c         | 9 +++++----
 net/netfilter/nft_ct.c                      | 2 ++
 net/netfilter/xt_connlabel.c                | 2 +-
 net/openvswitch/conntrack.c                 | 2 +-
 5 files changed, 11 insertions(+), 8 deletions(-)
include/net/netfilter/nf_conntrack_labels.h +2 -2
index 5167818..c5f8fc73 100644
@@ -53,11 +53,11 @@ int nf_connlabels_replace(struct nf_conn *ct,
 #ifdef CONFIG_NF_CONNTRACK_LABELS
 int nf_conntrack_labels_init(void);
 void nf_conntrack_labels_fini(void);
-int nf_connlabels_get(struct net *net, unsigned int n_bits);
+int nf_connlabels_get(struct net *net, unsigned int bit);
 void nf_connlabels_put(struct net *net);
 #else
 static inline int nf_conntrack_labels_init(void) { return 0; }
 static inline void nf_conntrack_labels_fini(void) {}
-static inline int nf_connlabels_get(struct net *net, unsigned int n_bits) { return 0; }
+static inline int nf_connlabels_get(struct net *net, unsigned int bit) { return 0; }
 static inline void nf_connlabels_put(struct net *net) {}
 #endif
net/netfilter/nf_conntrack_labels.c +5 -4
index bd7f26b..252e6a7 100644
@@ -78,15 +78,14 @@ int nf_connlabels_replace(struct nf_conn *ct,
 }
 EXPORT_SYMBOL_GPL(nf_connlabels_replace);
 
-int nf_connlabels_get(struct net *net, unsigned int n_bits)
+int nf_connlabels_get(struct net *net, unsigned int bits)
 {
 	size_t words;
 
-	if (n_bits > (NF_CT_LABELS_MAX_SIZE * BITS_PER_BYTE))
+	words = BIT_WORD(bits) + 1;
+	if (words > NF_CT_LABELS_MAX_SIZE / sizeof(long))
 		return -ERANGE;
 
-	words = BITS_TO_LONGS(n_bits);
-
 	spin_lock(&nf_connlabels_lock);
 	net->ct.labels_used++;
 	if (words > net->ct.label_words)
@@ -115,6 +114,8 @@ static struct nf_ct_ext_type labels_extend __read_mostly = {
 
 int nf_conntrack_labels_init(void)
 {
+	BUILD_BUG_ON(NF_CT_LABELS_MAX_SIZE / sizeof(long) >= U8_MAX);
+
 	spin_lock_init(&nf_connlabels_lock);
 	return nf_ct_extend_register(&labels_extend);
 }
net/netfilter/nft_ct.c +2 -0
index d4a4619..25998fa 100644
@@ -484,6 +484,8 @@ static struct nft_expr_type nft_ct_type __read_mostly = {
 
 static int __init nft_ct_module_init(void)
 {
+	BUILD_BUG_ON(NF_CT_LABELS_MAX_SIZE > NFT_REG_SIZE);
+
 	return nft_register_expr(&nft_ct_type);
 }
 
net/netfilter/xt_connlabel.c +1 -1
index d9b3e53..a79af25 100644
@@ -65,7 +65,7 @@ static int connlabel_mt_check(const struct xt_mtchk_param *par)
 		return ret;
 	}
 
-	ret = nf_connlabels_get(par->net, info->bit + 1);
+	ret = nf_connlabels_get(par->net, info->bit);
 	if (ret < 0)
 		nf_ct_l3proto_module_put(par->family);
 	return ret;
net/openvswitch/conntrack.c +1 -1
index dc5eb29..3da6107 100644
@@ -1342,7 +1342,7 @@ void ovs_ct_init(struct net *net)
 	unsigned int n_bits = sizeof(struct ovs_key_ct_labels) * BITS_PER_BYTE;
 	struct ovs_net *ovs_net = net_generic(net, ovs_net_id);
 
-	if (nf_connlabels_get(net, n_bits)) {
+	if (nf_connlabels_get(net, n_bits - 1)) {
 		ovs_net->xt_label = false;
 		OVS_NLERR(true, "Failed to set connlabel length");
 	} else {
Signature · 1 line
2.7.3
author maintainer
#4 lore raw

PATCH v5 4/4 netfilter: nftables: add connlabel set support

In reply to Florian Westphal

Instead of taking the value to set from a source register, userspace passes the bit that we should set as an immediate netlink value.

This follows a similar approach that xtables 'connlabel' match uses, so when user inputs

ct label set bar

then we will set the bit used by the 'bar' label and leave the rest alone. Pablo suggested to re-use the immediate attributes already used by nft_immediate, nft_bitwise and nft_cmp to re-use as much code as possible.

Just add new NFTA_CT_IMM that contains nested data attributes. We can then use nft_data_init and nft_data_dump for this as well.

---
Changes since last version:
- add generic NFTA_CT_IMM and pass it a struct nft_data.

 include/uapi/linux/netfilter/nf_tables.h |  2 +
 net/netfilter/nft_ct.c                   | 76 ++++++++++++++++++++++++++++++--
 2 files changed, 75 insertions(+), 3 deletions(-)
include/uapi/linux/netfilter/nf_tables.h +2 -0
index eeffde1..608b647 100644
@@ -770,6 +770,7 @@ enum nft_ct_keys {
  * @NFTA_CT_KEY: conntrack data item to load (NLA_U32: nft_ct_keys)
  * @NFTA_CT_DIRECTION: direction in case of directional keys (NLA_U8)
  * @NFTA_CT_SREG: source register (NLA_U32)
+ * @NFTA_CT_IMM: immediate value (NLA_NESTED)
  */
 enum nft_ct_attributes {
 	NFTA_CT_UNSPEC,
@@ -777,6 +778,7 @@ enum nft_ct_attributes {
 	NFTA_CT_KEY,
 	NFTA_CT_DIRECTION,
 	NFTA_CT_SREG,
+	NFTA_CT_IMM,
 	__NFTA_CT_MAX
 };
 #define NFTA_CT_MAX		(__NFTA_CT_MAX - 1)
net/netfilter/nft_ct.c +73 -3

136 lines not on this page · raw message

Signature · 1 line
2.7.3
#5 lore raw

net/netfilter/nft_ct.c struct nft_ct { · line 36

Quoted 13 lines · Florian Westphal
net/netfilter/nft_ct.c +5 -0
index 25998fa..4ec1cea 100644
@@ -29,6 +29,11 @@ struct nft_ct {
 		enum nft_registers	dreg:8;
 		enum nft_registers	sreg:8;
 	};
+	union {
+		u8		set_bit;
+	} imm;
+	unsigned int		imm_len:8;
+	struct nft_data		immediate;

Could you use select_ops() so we don't increase the size of nft_ct for other users?

author maintainer
#6 lore raw
Quoted 17 lines · Pablo Neira Ayuso

net/netfilter/nft_ct.c struct nft_ct { · line 36

Quoted 13 lines · Florian Westphal
net/netfilter/nft_ct.c +5 -0
index 25998fa..4ec1cea 100644
@@ -29,6 +29,11 @@ struct nft_ct {
 		enum nft_registers	dreg:8;
 		enum nft_registers	sreg:8;
 	};
+	union {
+		u8		set_bit;
+	} imm;
+	unsigned int		imm_len:8;
+	struct nft_data		immediate;

Could you use select_ops() so we don't increase the size of nft_ct for other users?

Sure.

I'd split this into nft_ct (sreg/dreg)
and nft_ct_set_imm (set from immediate).

Does that sound okay?

#7 lore raw
Quoted 13 lines · Florian Westphal
Quoted 12 lines · Pablo Neira Ayuso

net/netfilter/nft_ct.c struct nft_ct { · line 34

Quoted 11 lines · Florian Westphal
net/netfilter/nft_ct.c +3 -0
index 25998fa..4ec1cea 100644
@@ -29,6 +29,11 @@ struct nft_ct {
 		enum nft_registers	dreg:8;
 		enum nft_registers	sreg:8;
 	};
+	union {
+		u8		set_bit;
+	} imm;

BTW, do you really need this set_bit? I think we can just take the data from the nft_data structure.

Quoted 1 line
Quoted 1 line

net/netfilter/nft_ct.c struct nft_ct { · line 35

Quoted 1 line
net/netfilter/nft_ct.c +1 -0
+	unsigned int		imm_len:8;

This length, you will not need anymore with select_ops(), right=

Quoted 9 lines
Quoted 4 lines

net/netfilter/nft_ct.c struct nft_ct { · line 36

Quoted 1 line
net/netfilter/nft_ct.c +1 -0
+	struct nft_data		immediate;

Could you use select_ops() so we don't increase the size of nft_ct for other users?

Sure.

I'd split this into nft_ct (sreg/dreg)
and nft_ct_set_imm (set from immediate).

I'd suggest "struct nft_ct_reg" and "struct nft_ct_imm", so we can reuse the immediate from the get part if we can get rid of the imm_len and set_bit fields.

Thanks.

author maintainer
#8 lore raw
Quoted 17 lines · Pablo Neira Ayuso
Quoted 13 lines · Florian Westphal
Quoted 12 lines · Pablo Neira Ayuso

net/netfilter/nft_ct.c struct nft_ct { · line 34

Quoted 11 lines · Florian Westphal
net/netfilter/nft_ct.c +3 -0
index 25998fa..4ec1cea 100644
@@ -29,6 +29,11 @@ struct nft_ct {
 		enum nft_registers	dreg:8;
 		enum nft_registers	sreg:8;
 	};
+	union {
+		u8		set_bit;
+	} imm;

BTW, do you really need this set_bit? I think we can just take the data from the nft_data structure.

An earlier patch did that (did not submit it); I found it ugly (set_bit("ntohl(priv->data.data[0])").

I can cahnge it back if you want.

Quoted 3 lines
Quoted 1 line
Quoted 1 line

net/netfilter/nft_ct.c struct nft_ct { · line 35

Quoted 1 line
net/netfilter/nft_ct.c +1 -0
+	unsigned int		imm_len:8;

This length, you will not need anymore with select_ops(), right=

Hmm, I followed nft_cmp_fast implementation.
We expect a 32bit data type for the "label" key.

The alternative to storing length is to have a hard-coded size dispatch based on the key (e.g. use sizeof(u32) for labels).

But this might not work for all future cases.

I found saving the length from the desc struct to be much simpler (because dump function is shorter and doesn't have to guess the right
size when dumping).

Quoted 3 lines

I'd suggest "struct nft_ct_reg" and "struct nft_ct_imm", so we can reuse the immediate from the get part if we can get rid of the imm_len and set_bit fields.

Oh. I did not expect use of IMM for get operations.

Do we need a distinct attribute (IMM_GET vs IMM_SET)?

At the moment the assumption is that presence of IMM requests a set operation (presence of both IMM and sreg is an error).

#9 lore raw
Quoted 28 lines · Florian Westphal
Quoted 17 lines · Pablo Neira Ayuso
Quoted 13 lines · Florian Westphal
Quoted 12 lines · Pablo Neira Ayuso

net/netfilter/nft_ct.c struct nft_ct { · line 34

Quoted 11 lines · Florian Westphal
net/netfilter/nft_ct.c +3 -0
index 25998fa..4ec1cea 100644
@@ -29,6 +29,11 @@ struct nft_ct {
 		enum nft_registers	dreg:8;
 		enum nft_registers	sreg:8;
 	};
+	union {
+		u8		set_bit;
+	} imm;

BTW, do you really need this set_bit? I think we can just take the data from the nft_data structure.

An earlier patch did that (did not submit it); I found it ugly (set_bit("ntohl(priv->data.data[0])").

I can cahnge it back if you want.

Quoted 3 lines
Quoted 1 line
Quoted 1 line

net/netfilter/nft_ct.c struct nft_ct { · line 35

Quoted 1 line
net/netfilter/nft_ct.c +1 -0
+	unsigned int		imm_len:8;

This length, you will not need anymore with select_ops(), right=

Quoted 7 lines

Hmm, I followed nft_cmp_fast implementation.
We expect a 32bit data type for the "label" key.

The alternative to storing length is to have a hard-coded size dispatch based on the key (e.g. use sizeof(u32) for labels).

But this might not work for all future cases.

No problem, we can revisit this. If you believe this is the right structure layout I'm ok with it, so just focus on the reworking this to use select_ops().

Quoted 3 lines

I found saving the length from the desc struct to be much simpler (because dump function is shorter and doesn't have to guess the right
size when dumping).

Dumping is not performance critical path, so I wouldn't bother on this.

Quoted 5 lines
Quoted 3 lines

I'd suggest "struct nft_ct_reg" and "struct nft_ct_imm", so we can reuse the immediate from the get part if we can get rid of the imm_len and set_bit fields.

Oh. I did not expect use of IMM for get operations.

Right.

Quoted 4 lines

Do we need a distinct attribute (IMM_GET vs IMM_SET)?

At the moment the assumption is that presence of IMM requests a set operation (presence of both IMM and sreg is an error).

If you use select_ops(), we can just reject whenever the register attribute is set via -EINVAL.

#10 lore raw
In reply to Florian Westphal
Quoted 6 lines · Florian Westphal

Hi.

This is round 5 of the connlabel set support set.
I'm only sending the kernel patches for now.

First 4 patches are preparation changes, patch #4 adds set support.

Florian, I can push into the tree the 3 preparation patches, I don't see many changes on them. You can follow up with the nft_ct changes later on.

author maintainer
#11 lore raw
Quoted 5 lines · Pablo Neira Ayuso
Quoted 1 line

First 4 patches are preparation changes, patch #4 adds set support.

Florian, I can push into the tree the 3 preparation patches, I don't see many changes on them. You can follow up with the nft_ct changes later on.

That would be great, thanks!

#12 lore raw
Quoted 8 lines · Florian Westphal
Quoted 5 lines · Pablo Neira Ayuso
Quoted 1 line

First 4 patches are preparation changes, patch #4 adds set support.

Florian, I can push into the tree the 3 preparation patches, I don't see many changes on them. You can follow up with the nft_ct changes later on.

That would be great, thanks!

Done, thanks!