netfilter: convert hook list to an array

This converts the storage and layout of netfilter hook entries from a
linked list to an array.  After this commit, hook entries will be
stored adjacent in memory.  The next pointer is no longer required.

The ops pointers are stored at the end of the array as they are only
used in the register/unregister path and in the legacy br_netfilter code.

nf_unregister_net_hooks() is slower than needed as it just calls
nf_unregister_net_hook in a loop (i.e. at least n synchronize_net()
calls), this will be addressed in followup patch.

Test setup:
 - ixgbe 10gbit
 - netperf UDP_STREAM, 64 byte packets
 - 5 hooks: (raw + mangle prerouting, mangle+filter input, inet filter):
empty mangle and raw prerouting, mangle and filter input hooks:
353.9
this patch:
364.2

Signed-off-by: Aaron Conole <aconole@bytheb.org>
Signed-off-by: Florian Westphal <fw@strlen.de>
Signed-off-by: Pablo Neira Ayuso <pablo@netfilter.org>

4 files changed, 279 insertions, 107 deletions

diff --git a/net/bridge/br_netfilter_hooks.c b/net/bridge/br_netfilter_hooks.c
index 626f4b2cef16..c2eea1b8737a 100644
--- a/net/bridge/br_netfilter_hooks.c
+++ b/net/bridge/br_netfilter_hooks.c
@@ -985,22 +985,25 @@ int br_nf_hook_thresh(unsigned int hook, struct net *net,
 		      int (*okfn)(struct net *, struct sock *,
 				  struct sk_buff *))
 {
-	struct nf_hook_entry *elem;
+	const struct nf_hook_entries *e;
 	struct nf_hook_state state;
+	struct nf_hook_ops **ops;
+	unsigned int i;
 	int ret;
 
-	for (elem = rcu_dereference(net->nf.hooks[NFPROTO_BRIDGE][hook]);
-	     elem && nf_hook_entry_priority(elem) <= NF_BR_PRI_BRNF;
-	     elem = rcu_dereference(elem->next))
-		;
-
-	if (!elem)
+	e = rcu_dereference(net->nf.hooks[NFPROTO_BRIDGE][hook]);
+	if (!e)
 		return okfn(net, sk, skb);
 
+	ops = nf_hook_entries_get_hook_ops(e);
+	for (i = 0; i < e->num_hook_entries &&
+	      ops[i]->priority <= NF_BR_PRI_BRNF; i++)
+		;
+
 	nf_hook_state_init(&state, hook, NFPROTO_BRIDGE, indev, outdev,
 			   sk, net, okfn);
 
-	ret = nf_hook_slow(skb, &state, elem);
+	ret = nf_hook_slow(skb, &state, e, i);
 	if (ret == 1)
 		ret = okfn(net, sk, skb);
 
diff --git a/net/netfilter/core.c b/net/netfilter/core.c
index 974cf2a3795a..1a9e23c9ab98 100644
--- a/net/netfilter/core.c
+++ b/net/netfilter/core.c
@@ -21,7 +21,7 @@
 #include <linux/inetdevice.h>
 #include <linux/proc_fs.h>
 #include <linux/mutex.h>
-#include <linux/slab.h>
+#include <linux/mm.h>
 #include <linux/rcupdate.h>
 #include <net/net_namespace.h>
 #include <net/sock.h>
@@ -62,10 +62,160 @@ EXPORT_SYMBOL(nf_hooks_needed);
 #endif
 
 static DEFINE_MUTEX(nf_hook_mutex);
+
+/* max hooks per family/hooknum */
+#define MAX_HOOK_COUNT		1024
+
 #define nf_entry_dereference(e) \
 	rcu_dereference_protected(e, lockdep_is_held(&nf_hook_mutex))
 
-static struct nf_hook_entry __rcu **nf_hook_entry_head(struct net *net, const struct nf_hook_ops *reg)
+static struct nf_hook_entries *allocate_hook_entries_size(u16 num)
+{
+	struct nf_hook_entries *e;
+	size_t alloc = sizeof(*e) +
+		       sizeof(struct nf_hook_entry) * num +
+		       sizeof(struct nf_hook_ops *) * num;
+
+	if (num == 0)
+		return NULL;
+
+	e = kvzalloc(alloc, GFP_KERNEL);
+	if (e)
+		e->num_hook_entries = num;
+	return e;
+}
+
+static unsigned int accept_all(void *priv,
+			       struct sk_buff *skb,
+			       const struct nf_hook_state *state)
+{
+	return NF_ACCEPT; /* ACCEPT makes nf_hook_slow call next hook */
+}
+
+static const struct nf_hook_ops dummy_ops = {
+	.hook = accept_all,
+	.priority = INT_MIN,
+};
+
+static struct nf_hook_entries *
+nf_hook_entries_grow(const struct nf_hook_entries *old,
+		     const struct nf_hook_ops *reg)
+{
+	unsigned int i, alloc_entries, nhooks, old_entries;
+	struct nf_hook_ops **orig_ops = NULL;
+	struct nf_hook_ops **new_ops;
+	struct nf_hook_entries *new;
+	bool inserted = false;
+
+	alloc_entries = 1;
+	old_entries = old ? old->num_hook_entries : 0;
+
+	if (old) {
+		orig_ops = nf_hook_entries_get_hook_ops(old);
+
+		for (i = 0; i < old_entries; i++) {
+			if (orig_ops[i] != &dummy_ops)
+				alloc_entries++;
+		}
+	}
+
+	if (alloc_entries > MAX_HOOK_COUNT)
+		return ERR_PTR(-E2BIG);
+
+	new = allocate_hook_entries_size(alloc_entries);
+	if (!new)
+		return ERR_PTR(-ENOMEM);
+
+	new_ops = nf_hook_entries_get_hook_ops(new);
+
+	i = 0;
+	nhooks = 0;
+	while (i < old_entries) {
+		if (orig_ops[i] == &dummy_ops) {
+			++i;
+			continue;
+		}
+		if (inserted || reg->priority > orig_ops[i]->priority) {
+			new_ops[nhooks] = (void *)orig_ops[i];
+			new->hooks[nhooks] = old->hooks[i];
+			i++;
+		} else {
+			new_ops[nhooks] = (void *)reg;
+			new->hooks[nhooks].hook = reg->hook;
+			new->hooks[nhooks].priv = reg->priv;
+			inserted = true;
+		}
+		nhooks++;
+	}
+
+	if (!inserted) {
+		new_ops[nhooks] = (void *)reg;
+		new->hooks[nhooks].hook = reg->hook;
+		new->hooks[nhooks].priv = reg->priv;
+	}
+
+	return new;
+}
+
+/*
+ * __nf_hook_entries_try_shrink - try to shrink hook array
+ *
+ * @pp -- location of hook blob
+ *
+ * Hook unregistration must always succeed, so to-be-removed hooks
+ * are replaced by a dummy one that will just move to next hook.
+ *
+ * This counts the current dummy hooks, attempts to allocate new blob,
+ * copies the live hooks, then replaces and discards old one.
+ *
+ * return values:
+ *
+ * Returns address to free, or NULL.
+ */
+static void *__nf_hook_entries_try_shrink(struct nf_hook_entries __rcu **pp)
+{
+	struct nf_hook_entries *old, *new = NULL;
+	unsigned int i, j, skip = 0, hook_entries;
+	struct nf_hook_ops **orig_ops;
+	struct nf_hook_ops **new_ops;
+
+	old = nf_entry_dereference(*pp);
+	if (WARN_ON_ONCE(!old))
+		return NULL;
+
+	orig_ops = nf_hook_entries_get_hook_ops(old);
+	for (i = 0; i < old->num_hook_entries; i++) {
+		if (orig_ops[i] == &dummy_ops)
+			skip++;
+	}
+
+	/* if skip == hook_entries all hooks have been removed */
+	hook_entries = old->num_hook_entries;
+	if (skip == hook_entries)
+		goto out_assign;
+
+	if (WARN_ON(skip == 0))
+		return NULL;
+
+	hook_entries -= skip;
+	new = allocate_hook_entries_size(hook_entries);
+	if (!new)
+		return NULL;
+
+	new_ops = nf_hook_entries_get_hook_ops(new);
+	for (i = 0, j = 0; i < old->num_hook_entries; i++) {
+		if (orig_ops[i] == &dummy_ops)
+			continue;
+		new->hooks[j] = old->hooks[i];
+		new_ops[j] = (void *)orig_ops[i];
+		j++;
+	}
+out_assign:
+	rcu_assign_pointer(*pp, new);
+	return old;
+}
+
+static struct nf_hook_entries __rcu **nf_hook_entry_head(struct net *net, const struct nf_hook_ops *reg)
 {
 	if (reg->pf != NFPROTO_NETDEV)
 		return net->nf.hooks[reg->pf]+reg->hooknum;
@@ -76,13 +226,14 @@ static struct nf_hook_entry __rcu **nf_hook_entry_head(struct net *net, const st
 			return &reg->dev->nf_hooks_ingress;
 	}
 #endif
+	WARN_ON_ONCE(1);
 	return NULL;
 }
 
 int nf_register_net_hook(struct net *net, const struct nf_hook_ops *reg)
 {
-	struct nf_hook_entry __rcu **pp;
-	struct nf_hook_entry *entry, *p;
+	struct nf_hook_entries *p, *new_hooks;
+	struct nf_hook_entries __rcu **pp;
 
 	if (reg->pf == NFPROTO_NETDEV) {
 #ifndef CONFIG_NETFILTER_INGRESS
@@ -98,23 +249,18 @@ int nf_register_net_hook(struct net *net, const struct nf_hook_ops *reg)
 	if (!pp)
 		return -EINVAL;
 
-	entry = kmalloc(sizeof(*entry), GFP_KERNEL);
-	if (!entry)
-		return -ENOMEM;
-
-	nf_hook_entry_init(entry, reg);
-
 	mutex_lock(&nf_hook_mutex);
 
-	/* Find the spot in the list */
-	for (; (p = nf_entry_dereference(*pp)) != NULL; pp = &p->next) {
-		if (reg->priority < nf_hook_entry_priority(p))
-			break;
-	}
-	rcu_assign_pointer(entry->next, p);
-	rcu_assign_pointer(*pp, entry);
+	p = nf_entry_dereference(*pp);
+	new_hooks = nf_hook_entries_grow(p, reg);
+
+	if (!IS_ERR(new_hooks))
+		rcu_assign_pointer(*pp, new_hooks);
 
 	mutex_unlock(&nf_hook_mutex);
+	if (IS_ERR(new_hooks))
+		return PTR_ERR(new_hooks);
+
 #ifdef CONFIG_NETFILTER_INGRESS
 	if (reg->pf == NFPROTO_NETDEV && reg->hooknum == NF_NETDEV_INGRESS)
 		net_inc_ingress_queue();
@@ -122,48 +268,74 @@ int nf_register_net_hook(struct net *net, const struct nf_hook_ops *reg)
 #ifdef HAVE_JUMP_LABEL
 	static_key_slow_inc(&nf_hooks_needed[reg->pf][reg->hooknum]);
 #endif
+	synchronize_net();
+	BUG_ON(p == new_hooks);
+	kvfree(p);
 	return 0;
 }
 EXPORT_SYMBOL(nf_register_net_hook);
 
-static struct nf_hook_entry *
-__nf_unregister_net_hook(struct net *net, const struct nf_hook_ops *reg)
+/*
+ * __nf_unregister_net_hook - remove a hook from blob
+ *
+ * @oldp: current address of hook blob
+ * @unreg: hook to unregister
+ *
+ * This cannot fail, hook unregistration must always succeed.
+ * Therefore replace the to-be-removed hook with a dummy hook.
+ */
+static void __nf_unregister_net_hook(struct nf_hook_entries *old,
+				     const struct nf_hook_ops *unreg)
 {
-	struct nf_hook_entry __rcu **pp;
-	struct nf_hook_entry *p;
-
-	pp = nf_hook_entry_head(net, reg);
-	if (WARN_ON_ONCE(!pp))
-		return NULL;
+	struct nf_hook_ops **orig_ops;
+	bool found = false;
+	unsigned int i;
 
-	mutex_lock(&nf_hook_mutex);
-	for (; (p = nf_entry_dereference(*pp)) != NULL; pp = &p->next) {
-		if (nf_hook_entry_ops(p) == reg) {
-			rcu_assign_pointer(*pp, p->next);
-			break;
-		}
-	}
-	mutex_unlock(&nf_hook_mutex);
-	if (!p) {
-		WARN(1, "nf_unregister_net_hook: hook not found!\n");
-		return NULL;
+	orig_ops = nf_hook_entries_get_hook_ops(old);
+	for (i = 0; i < old->num_hook_entries; i++) {
+		if (orig_ops[i] != unreg)
+			continue;
+		WRITE_ONCE(old->hooks[i].hook, accept_all);
+		WRITE_ONCE(orig_ops[i], &dummy_ops);
+		found = true;
+		break;
 	}
+
+	if (found) {
 #ifdef CONFIG_NETFILTER_INGRESS
-	if (reg->pf == NFPROTO_NETDEV && reg->hooknum == NF_NETDEV_INGRESS)
-		net_dec_ingress_queue();
+		if (unreg->pf == NFPROTO_NETDEV && unreg->hooknum == NF_NETDEV_INGRESS)
+			net_dec_ingress_queue();
 #endif
 #ifdef HAVE_JUMP_LABEL
-	static_key_slow_dec(&nf_hooks_needed[reg->pf][reg->hooknum]);
+		static_key_slow_dec(&nf_hooks_needed[unreg->pf][unreg->hooknum]);
 #endif
-
-	return p;
+	} else {
+		WARN_ONCE(1, "hook not found, pf %d num %d", unreg->pf, unreg->hooknum);
+	}
 }
 
 void nf_unregister_net_hook(struct net *net, const struct nf_hook_ops *reg)
 {
-	struct nf_hook_entry *p = __nf_unregister_net_hook(net, reg);
+	struct nf_hook_entries __rcu **pp;
+	struct nf_hook_entries *p;
 	unsigned int nfq;
 
+	pp = nf_hook_entry_head(net, reg);
+	if (!pp)
+		return;
+
+	mutex_lock(&nf_hook_mutex);
+
+	p = nf_entry_dereference(*pp);
+	if (WARN_ON_ONCE(!p)) {
+		mutex_unlock(&nf_hook_mutex);
+		return;
+	}
+
+	__nf_unregister_net_hook(p, reg);
+
+	p = __nf_hook_entries_try_shrink(pp);
+	mutex_unlock(&nf_hook_mutex);
 	if (!p)
 		return;
 
@@ -173,7 +345,7 @@ void nf_unregister_net_hook(struct net *net, const struct nf_hook_ops *reg)
 	nfq = nf_queue_nf_hook_drop(net);
 	if (nfq)
 		synchronize_net();
-	kfree(p);
+	kvfree(p);
 }
 EXPORT_SYMBOL(nf_unregister_net_hook);
 
@@ -200,46 +372,25 @@ EXPORT_SYMBOL(nf_register_net_hooks);
 void nf_unregister_net_hooks(struct net *net, const struct nf_hook_ops *reg,
 			     unsigned int hookcount)
 {
-	struct nf_hook_entry *to_free[16];
-	unsigned int i, n, nfq;
-
-	do {
-		n = min_t(unsigned int, hookcount, ARRAY_SIZE(to_free));
-
-		for (i = 0; i < n; i++)
-			to_free[i] = __nf_unregister_net_hook(net, &reg[i]);
-
-		synchronize_net();
-
-		/* need 2nd synchronize_net() if nfqueue is used, skb
-		 * can get reinjected right before nf_queue_hook_drop()
-		 */
-		nfq = nf_queue_nf_hook_drop(net);
-		if (nfq)
-			synchronize_net();
-
-		for (i = 0; i < n; i++)
-			kfree(to_free[i]);
+	unsigned int i;
 
-		reg += n;
-		hookcount -= n;
-	} while (hookcount > 0);
+	for (i = 0; i < hookcount; i++)
+		nf_unregister_net_hook(net, &reg[i]);
 }
 EXPORT_SYMBOL(nf_unregister_net_hooks);
 
 /* Returns 1 if okfn() needs to be executed by the caller,
  * -EPERM for NF_DROP, 0 otherwise.  Caller must hold rcu_read_lock. */
 int nf_hook_slow(struct sk_buff *skb, struct nf_hook_state *state,
-		 struct nf_hook_entry *entry)
+		 const struct nf_hook_entries *e, unsigned int s)
 {
 	unsigned int verdict;
 	int ret;
 
-	do {
-		verdict = nf_hook_entry_hookfn(entry, skb, state);
+	for (; s < e->num_hook_entries; s++) {
+		verdict = nf_hook_entry_hookfn(&e->hooks[s], skb, state);
 		switch (verdict & NF_VERDICT_MASK) {
 		case NF_ACCEPT:
-			entry = rcu_dereference(entry->next);
 			break;
 		case NF_DROP:
 			kfree_skb(skb);
@@ -248,8 +399,8 @@ int nf_hook_slow(struct sk_buff *skb, struct nf_hook_state *state,
 				ret = -EPERM;
 			return ret;
 		case NF_QUEUE:
-			ret = nf_queue(skb, state, &entry, verdict);
-			if (ret == 1 && entry)
+			ret = nf_queue(skb, state, e, s, verdict);
+			if (ret == 1)
 				continue;
 			return ret;
 		default:
@@ -258,7 +409,7 @@ int nf_hook_slow(struct sk_buff *skb, struct nf_hook_state *state,
 			 */
 			return 0;
 		}
-	} while (entry);
+	}
 
 	return 1;
 }
diff --git a/net/netfilter/nf_internals.h b/net/netfilter/nf_internals.h
index 19f00a47a710..bacd6363946e 100644
--- a/net/netfilter/nf_internals.h
+++ b/net/netfilter/nf_internals.h
@@ -13,7 +13,8 @@
 
 /* nf_queue.c */
 int nf_queue(struct sk_buff *skb, struct nf_hook_state *state,
-	     struct nf_hook_entry **entryp, unsigned int verdict);
+	     const struct nf_hook_entries *entries, unsigned int index,
+	     unsigned int verdict);
 unsigned int nf_queue_nf_hook_drop(struct net *net);
 
 /* nf_log.c */
diff --git a/net/netfilter/nf_queue.c b/net/netfilter/nf_queue.c
index 4f4d80a58fb5..f7e21953b1de 100644
--- a/net/netfilter/nf_queue.c
+++ b/net/netfilter/nf_queue.c
@@ -112,7 +112,8 @@ unsigned int nf_queue_nf_hook_drop(struct net *net)
 EXPORT_SYMBOL_GPL(nf_queue_nf_hook_drop);
 
 static int __nf_queue(struct sk_buff *skb, const struct nf_hook_state *state,
-		      struct nf_hook_entry *hook_entry, unsigned int queuenum)
+		      const struct nf_hook_entries *entries,
+		      unsigned int index, unsigned int queuenum)
 {
 	int status = -ENOENT;
 	struct nf_queue_entry *entry = NULL;
@@ -140,7 +141,7 @@ static int __nf_queue(struct sk_buff *skb, const struct nf_hook_state *state,
 	*entry = (struct nf_queue_entry) {
 		.skb	= skb,
 		.state	= *state,
-		.hook	= hook_entry,
+		.hook_index = index,
 		.size	= sizeof(*entry) + afinfo->route_key_size,
 	};
 
@@ -163,18 +164,16 @@ err:
 
 /* Packets leaving via this function must come back through nf_reinject(). */
 int nf_queue(struct sk_buff *skb, struct nf_hook_state *state,
-	     struct nf_hook_entry **entryp, unsigned int verdict)
+	     const struct nf_hook_entries *entries, unsigned int index,
+	     unsigned int verdict)
 {
-	struct nf_hook_entry *entry = *entryp;
 	int ret;
 
-	ret = __nf_queue(skb, state, entry, verdict >> NF_VERDICT_QBITS);
+	ret = __nf_queue(skb, state, entries, index, verdict >> NF_VERDICT_QBITS);
 	if (ret < 0) {
 		if (ret == -ESRCH &&
-		    (verdict & NF_VERDICT_FLAG_QUEUE_BYPASS)) {
-			*entryp = rcu_dereference(entry->next);
+		    (verdict & NF_VERDICT_FLAG_QUEUE_BYPASS))
 			return 1;
-		}
 		kfree_skb(skb);
 	}
 
@@ -183,33 +182,56 @@ int nf_queue(struct sk_buff *skb, struct nf_hook_state *state,
 
 static unsigned int nf_iterate(struct sk_buff *skb,
 			       struct nf_hook_state *state,
-			       struct nf_hook_entry **entryp)
+			       const struct nf_hook_entries *hooks,
+			       unsigned int *index)
 {
-	unsigned int verdict;
+	const struct nf_hook_entry *hook;
+	unsigned int verdict, i = *index;
 
-	do {
+	while (i < hooks->num_hook_entries) {
+		hook = &hooks->hooks[i];
 repeat:
-		verdict = nf_hook_entry_hookfn((*entryp), skb, state);
+		verdict = nf_hook_entry_hookfn(hook, skb, state);
 		if (verdict != NF_ACCEPT) {
 			if (verdict != NF_REPEAT)
 				return verdict;
 			goto repeat;
 		}
-		*entryp = rcu_dereference((*entryp)->next);
-	} while (*entryp);
+		i++;
+	}
 
+	*index = i;
 	return NF_ACCEPT;
 }
 
+/* Caller must hold rcu read-side lock */
 void nf_reinject(struct nf_queue_entry *entry, unsigned int verdict)
 {
-	struct nf_hook_entry *hook_entry = entry->hook;
+	const struct nf_hook_entry *hook_entry;
+	const struct nf_hook_entries *hooks;
 	struct sk_buff *skb = entry->skb;
 	const struct nf_afinfo *afinfo;
+	const struct net *net;
+	unsigned int i;
 	int err;
+	u8 pf;
+
+	net = entry->state.net;
+	pf = entry->state.pf;
+
+	hooks = rcu_dereference(net->nf.hooks[pf][entry->state.hook]);
 
 	nf_queue_entry_release_refs(entry);
 
+	i = entry->hook_index;
+	if (WARN_ON_ONCE(i >= hooks->num_hook_entries)) {
+		kfree_skb(skb);
+		kfree(entry);
+		return;
+	}
+
+	hook_entry = &hooks->hooks[i];
+
 	/* Continue traversal iff userspace said ok... */
 	if (verdict == NF_REPEAT)
 		verdict = nf_hook_entry_hookfn(hook_entry, skb, &entry->state);
@@ -221,27 +243,22 @@ void nf_reinject(struct nf_queue_entry *entry, unsigned int verdict)
 	}
 
 	if (verdict == NF_ACCEPT) {
-		hook_entry = rcu_dereference(hook_entry->next);
-		if (hook_entry)
 next_hook:
-			verdict = nf_iterate(skb, &entry->state, &hook_entry);
+		++i;
+		verdict = nf_iterate(skb, &entry->state, hooks, &i);
 	}
 
 	switch (verdict & NF_VERDICT_MASK) {
 	case NF_ACCEPT:
 	case NF_STOP:
-okfn:
 		local_bh_disable();
 		entry->state.okfn(entry->state.net, entry->state.sk, skb);
 		local_bh_enable();
 		break;
 	case NF_QUEUE:
-		err = nf_queue(skb, &entry->state, &hook_entry, verdict);
-		if (err == 1) {
-			if (hook_entry)
-				goto next_hook;
-			goto okfn;
-		}
+		err = nf_queue(skb, &entry->state, hooks, i, verdict);
+		if (err == 1)
+			goto next_hook;
 		break;
 	case NF_STOLEN:
 		break;