Skip to contentSkip to conversation

chain rename and rule replacement

PATCH Superseded Tomasz Bursztyka 2 participants 6 messages Oct 31, 2012

Subsystems (2)

v1 of 2 Superseded Latest is v2 →

lore
Share & tools
lore
#0 lore raw

cover chain rename and rule replacement

Hi,

Here is two patches, there title is self-explaining.

Renaming chain's name applies only on user's chain. Not on builtin or base ones.

Please review,

Tomasz Bursztyka (2):
nf_tables: Add support for changing users chain's name
nf_tables: Add support for replacing a rule by another one.

 include/linux/netfilter/nf_tables.h |  1 +
 net/netfilter/nf_tables_api.c       | 82 ++++++++++++++++++++++++++++++++++---
 2 files changed, 77 insertions(+), 6 deletions(-)
Signature · 1 line
1.7.12.4
#1 lore raw

PATCH 1/2 nf_tables: Add support for changing users chain's name

---
 include/linux/netfilter/nf_tables.h |  1 +
 net/netfilter/nf_tables_api.c       | 58 ++++++++++++++++++++++++++++++++++++-
 2 files changed, 58 insertions(+), 1 deletion(-)
include/linux/netfilter/nf_tables.h +1 -0
index 0115a2f..542b654 100644
@@ -67,6 +67,7 @@ enum nft_chain_attributes {
 	NFTA_CHAIN_HOOK,
 	NFTA_CHAIN_POLICY,
 	NFTA_CHAIN_USE,
+	NFTA_CHAIN_NEW_NAME,
 	__NFTA_CHAIN_MAX
 };
 #define NFTA_CHAIN_MAX		(__NFTA_CHAIN_MAX - 1)
net/netfilter/nf_tables_api.c +57 -1

70 lines not on this page · raw message

Signature · 1 line
1.7.12.4
#2 lore raw

Re: [nf-next/nf_tables-experiments - PATCH 1/2] nf_tables: Add support for changing users chain's name

Hi Tomasz,

The patchset looks good, I just have a comment on this one.

Quoted 1 line

From net/netfilter/nf_tables_core.c:

        case NFT_GOTO:
                chain = data[NFT_REG_VERDICT].chain;

In the GOTO case, we already point to the chain object. That chain object is loaded in nft_immediate.c _eval(). However, the private data of immediate is initialized in the _init() path. That means we would need to refresh the entire rule-set to point to the correct new chain object, otherwise we would crash.

I'd follow a simpler solution to avoid issues. Use a fixed chain name length (the same length as iptables does, or just 32 bytes).

We can revisit this later to see if we can support renaming and dynamically allocated chain objects at the same time.

Would you resend a new version of this patch?

Thanks.

Quoted 101 lines · Tomasz Bursztyka
---
 include/linux/netfilter/nf_tables.h |  1 +
 net/netfilter/nf_tables_api.c       | 58 ++++++++++++++++++++++++++++++++++++-
 2 files changed, 58 insertions(+), 1 deletion(-)
include/linux/netfilter/nf_tables.h +1 -0
index 0115a2f..542b654 100644
@@ -67,6 +67,7 @@ enum nft_chain_attributes {
 	NFTA_CHAIN_HOOK,
 	NFTA_CHAIN_POLICY,
 	NFTA_CHAIN_USE,
+	NFTA_CHAIN_NEW_NAME,
 	__NFTA_CHAIN_MAX
 };
 #define NFTA_CHAIN_MAX		(__NFTA_CHAIN_MAX - 1)
net/netfilter/nf_tables_api.c +57 -1
index e0e4616..fd1b624 100644
@@ -676,6 +676,62 @@ nf_tables_chain_policy(struct nft_chain *chain, const struct nlattr *attr)
 	return 0;
 }
 
+static int nf_tables_mvchain(struct sk_buff *skb, const struct nlmsghdr *nlh,
+			     struct nft_table *table,
+			     struct nft_chain *old_chain,
+			     const struct nlattr * const nla[])
+{
+	const struct nfgenmsg *nfmsg = nlmsg_data(nlh);
+	int family = nfmsg->nfgen_family;
+	struct nft_chain *new_chain;
+	const struct nlattr *name;
+	unsigned int size;
+
+	if (!nla[NFTA_CHAIN_NEW_NAME])
+		return -EINVAL;
+
+	if (old_chain->flags & NFT_CHAIN_BUILTIN ||
+	    old_chain->flags & NFT_BASE_CHAIN)
+		return -EOPNOTSUPP;
+
+	if (old_chain->use > 0)
+		return -EBUSY;
+
+	name = nla[NFTA_CHAIN_NEW_NAME];
+	new_chain = nf_tables_chain_lookup(table, name);
+	if (IS_ERR(new_chain)) {
+		if (PTR_ERR(new_chain) != -ENOENT)
+			return PTR_ERR(new_chain);
+		new_chain = NULL;
+	}
+
+	if (new_chain != NULL)
+		return -EEXIST;
+
+	size = nla_len(name);
+	new_chain = kzalloc(sizeof(*new_chain) + size, GFP_KERNEL);
+	if (new_chain == NULL)
+		return -ENOMEM;
+
+	list_del(&old_chain->list);
+
+	INIT_LIST_HEAD(&new_chain->rules);
+	nla_strlcpy(new_chain->name, name, size);
+
+	/* Copying content from old chain */
+	new_chain->flags = old_chain->flags;
+	list_replace_init(&old_chain->rules, &new_chain->rules);
+
+	list_add_tail(&new_chain->list, &table->chains);
+
+	nf_tables_chain_notify(skb, nlh, table, old_chain, NFT_MSG_DELCHAIN,
+			       family);
+	kfree(old_chain);
+	nf_tables_chain_notify(skb, nlh, table, new_chain, NFT_MSG_NEWCHAIN,
+			       family);
+	return 0;
+}
+
 static int nf_tables_newchain(struct sock *nlsk, struct sk_buff *skb,
 			      const struct nlmsghdr *nlh,
 			      const struct nlattr * const nla[])
@@ -714,7 +770,7 @@ static int nf_tables_newchain(struct sock *nlsk, struct sk_buff *skb,
 		if (nlh->nlmsg_flags & NLM_F_EXCL)
 			return -EEXIST;
 		if (nlh->nlmsg_flags & NLM_F_REPLACE)
-			return -EOPNOTSUPP;
+			return nf_tables_mvchain(skb, nlh, table, chain, nla);
 
 		if ((chain->flags & NFT_BASE_CHAIN) && nla[NFTA_CHAIN_POLICY]) {
 			return nf_tables_chain_policy(chain,
Signature · 6 lines
1.7.12.4

--
To unsubscribe from this list: send the line "unsubscribe netfilter-devel" in
the body of a message to [email protected]
More majordomo info at  http://vger.kernel.org/majordomo-info.html
#3 lore raw

Re: [nf-next/nf_tables-experiments - PATCH 1/2] nf_tables: Add support for changing users chain's name

Hi,

Quoted 10 lines
Quoted 1 line

From net/netfilter/nf_tables_core.c:

         case NFT_GOTO:
                 chain = data[NFT_REG_VERDICT].chain;

In the GOTO case, we already point to the chain object. That chain object is loaded in nft_immediate.c _eval(). However, the private data of immediate is initialized in the _init() path. That means we would need to refresh the entire rule-set to point to the correct new chain object, otherwise we would crash.

Missed that one completely...

Quoted 5 lines

I'd follow a simpler solution to avoid issues. Use a fixed chain name length (the same length as iptables does, or just 32 bytes).

We can revisit this later to see if we can support renaming and dynamically allocated chain objects at the same time.

First solution is easy to implement, and do not imply any performance drop, only a bit of memory loss for short names.

Second one, I cannot see anything but a 2 pass system (a hash table handling unique handle related to a chain pointer). But then the gain of bytes - with variable name size - versus the size of the hash table etc... might be pointless. Maybe I am just wrong. Is there any other solution?

Quoted 1 line

Would you resend a new version of this patch?

Ok, I will go with first solution, unless you or someone else comes up with a better proposition.

Tomasz

#4 lore raw

Re: [nf-next/nf_tables-experiments - PATCH 1/2] nf_tables: Add support for changing users chain's name

Quoted 20 lines · Tomasz Bursztyka

Hi,

Quoted 10 lines
Quoted 1 line

From net/netfilter/nf_tables_core.c:

        case NFT_GOTO:
                chain = data[NFT_REG_VERDICT].chain;

In the GOTO case, we already point to the chain object. That chain object is loaded in nft_immediate.c _eval(). However, the private data of immediate is initialized in the _init() path. That means we would need to refresh the entire rule-set to point to the correct new chain object, otherwise we would crash.

Missed that one completely...

Quoted 5 lines

I'd follow a simpler solution to avoid issues. Use a fixed chain name length (the same length as iptables does, or just 32 bytes).

We can revisit this later to see if we can support renaming and dynamically allocated chain objects at the same time.

Quoted 8 lines

First solution is easy to implement, and do not imply any performance drop, only a bit of memory loss for short names.

Second one, I cannot see anything but a 2 pass system (a hash table handling unique handle related to a chain pointer).
But then the gain of bytes - with variable name size - versus the size of the hash table etc... might be pointless.
Maybe I am just wrong. Is there any other solution?

I don't see any at this moment.

Quoted 4 lines
Quoted 1 line

Would you resend a new version of this patch?

Ok, I will go with first solution, unless you or someone else comes up with a better proposition.

Agreed. It's simple and we can still revisit this later.

#5 lore raw

PATCH 2/2 nf_tables: Add support for replacing a rule by another one.

In reply to Tomasz Bursztyka
---
 net/netfilter/nf_tables_api.c | 24 +++++++++++++++++++-----
 1 file changed, 19 insertions(+), 5 deletions(-)
net/netfilter/nf_tables_api.c +19 -5
index fd1b624..3ce8aa1 100644
@@ -1321,7 +1321,7 @@ static int nf_tables_newrule(struct sock *nlsk, struct sk_buff *skb,
 	const struct nft_af_info *afi;
 	const struct nft_table *table;
 	struct nft_chain *chain;
-	struct nft_rule *rule;
+	struct nft_rule *rule, *old_rule = NULL;
 	struct nft_expr_info info[NFT_RULE_MAXEXPRS];
 	struct nft_expr *expr;
 	struct nft_ctx ctx;
@@ -1357,9 +1357,11 @@ static int nf_tables_newrule(struct sock *nlsk, struct sk_buff *skb,
 		if (rule != NULL) {
 			if (nlh->nlmsg_flags & NLM_F_EXCL)
 				return -EEXIST;
-			if (nlh->nlmsg_flags & NLM_F_REPLACE)
-				return -EOPNOTSUPP;
-			return 0;
+			if (nlh->nlmsg_flags & NLM_F_REPLACE) {
+				old_rule = rule;
+				rule = NULL;
+			} else
+				return 0;
 		}
 	} else
 		handle = nf_tables_rule_alloc_handle(chain);
@@ -1402,7 +1404,19 @@ static int nf_tables_newrule(struct sock *nlsk, struct sk_buff *skb,
 		expr = nft_expr_next(expr);
 	}
 
-	if (nlh->nlmsg_flags & NLM_F_APPEND)
+	if (nlh->nlmsg_flags & NLM_F_REPLACE) {
+		if (old_rule == NULL)
+			goto err2;
+
+		list_replace_rcu(&old_rule->list, &rule->list);
+
+		// FIXME: this makes deletion performance *really* suck
+		synchronize_rcu();
+
+		nf_tables_rule_notify(skb, nlh, table, chain, old_rule,
+				      NFT_MSG_DELRULE, nfmsg->nfgen_family);
+		nf_tables_rule_destroy(&ctx, old_rule);
+	} else if (nlh->nlmsg_flags & NLM_F_APPEND)
 		list_add_tail_rcu(&rule->list, &chain->rules);
 	else
 		list_add_rcu(&rule->list, &chain->rules);
Signature · 1 line
1.7.12.4