netfilter: replace list_head with single linked list
authorAaron Conole <aconole@bytheb.org>
Wed, 21 Sep 2016 15:35:07 +0000 (11:35 -0400)
committerPablo Neira Ayuso <pablo@netfilter.org>
Sun, 25 Sep 2016 12:38:48 +0000 (14:38 +0200)
The netfilter hook list never uses the prev pointer, and so can be trimmed to
be a simple singly-linked list.

In addition to having a more light weight structure for hook traversal,
struct net becomes 5568 bytes (down from 6400) and struct net_device becomes
2176 bytes (down from 2240).

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>
include/linux/netdevice.h
include/linux/netfilter.h
include/linux/netfilter_ingress.h
include/net/netfilter/nf_queue.h
include/net/netns/netfilter.h
net/bridge/br_netfilter_hooks.c
net/netfilter/core.c
net/netfilter/nf_internals.h
net/netfilter/nf_queue.c
net/netfilter/nfnetlink_queue.c

index 67bb978470dc25f47b28a003a768e6287b92a2ef..41f49f5ab62aeeafb9847b7fa4a6061bf5d4ccb8 100644 (file)
@@ -1783,7 +1783,7 @@ struct net_device {
 #endif
        struct netdev_queue __rcu *ingress_queue;
 #ifdef CONFIG_NETFILTER_INGRESS
-       struct list_head        nf_hooks_ingress;
+       struct nf_hook_entry __rcu *nf_hooks_ingress;
 #endif
 
        unsigned char           broadcast[MAX_ADDR_LEN];
index ad444f0b4ed0d30f51993348fd6f567befde9e87..44e20dac98a995e3c0f8ebb1ef7f464d69874847 100644 (file)
@@ -55,12 +55,34 @@ struct nf_hook_state {
        struct net_device *out;
        struct sock *sk;
        struct net *net;
-       struct list_head *hook_list;
+       struct nf_hook_entry __rcu *hook_entries;
        int (*okfn)(struct net *, struct sock *, struct sk_buff *);
 };
 
+typedef unsigned int nf_hookfn(void *priv,
+                              struct sk_buff *skb,
+                              const struct nf_hook_state *state);
+struct nf_hook_ops {
+       struct list_head        list;
+
+       /* User fills in from here down. */
+       nf_hookfn               *hook;
+       struct net_device       *dev;
+       void                    *priv;
+       u_int8_t                pf;
+       unsigned int            hooknum;
+       /* Hooks are ordered in ascending priority. */
+       int                     priority;
+};
+
+struct nf_hook_entry {
+       struct nf_hook_entry __rcu      *next;
+       struct nf_hook_ops              ops;
+       const struct nf_hook_ops        *orig_ops;
+};
+
 static inline void nf_hook_state_init(struct nf_hook_state *p,
-                                     struct list_head *hook_list,
+                                     struct nf_hook_entry *hook_entry,
                                      unsigned int hook,
                                      int thresh, u_int8_t pf,
                                      struct net_device *indev,
@@ -76,26 +98,11 @@ static inline void nf_hook_state_init(struct nf_hook_state *p,
        p->out = outdev;
        p->sk = sk;
        p->net = net;
-       p->hook_list = hook_list;
+       RCU_INIT_POINTER(p->hook_entries, hook_entry);
        p->okfn = okfn;
 }
 
-typedef unsigned int nf_hookfn(void *priv,
-                              struct sk_buff *skb,
-                              const struct nf_hook_state *state);
-
-struct nf_hook_ops {
-       struct list_head        list;
 
-       /* User fills in from here down. */
-       nf_hookfn               *hook;
-       struct net_device       *dev;
-       void                    *priv;
-       u_int8_t                pf;
-       unsigned int            hooknum;
-       /* Hooks are ordered in ascending priority. */
-       int                     priority;
-};
 
 struct nf_sockopt_ops {
        struct list_head list;
@@ -161,7 +168,8 @@ static inline int nf_hook_thresh(u_int8_t pf, unsigned int hook,
                                 int (*okfn)(struct net *, struct sock *, struct sk_buff *),
                                 int thresh)
 {
-       struct list_head *hook_list;
+       struct nf_hook_entry *hook_head;
+       int ret = 1;
 
 #ifdef HAVE_JUMP_LABEL
        if (__builtin_constant_p(pf) &&
@@ -170,22 +178,19 @@ static inline int nf_hook_thresh(u_int8_t pf, unsigned int hook,
                return 1;
 #endif
 
-       hook_list = &net->nf.hooks[pf][hook];
-
-       if (!list_empty(hook_list)) {
+       rcu_read_lock();
+       hook_head = rcu_dereference(net->nf.hooks[pf][hook]);
+       if (hook_head) {
                struct nf_hook_state state;
-               int ret;
 
-               /* We may already have this, but read-locks nest anyway */
-               rcu_read_lock();
-               nf_hook_state_init(&state, hook_list, hook, thresh,
+               nf_hook_state_init(&state, hook_head, hook, thresh,
                                   pf, indev, outdev, sk, net, okfn);
 
                ret = nf_hook_slow(skb, &state);
-               rcu_read_unlock();
-               return ret;
        }
-       return 1;
+       rcu_read_unlock();
+
+       return ret;
 }
 
 static inline int nf_hook(u_int8_t pf, unsigned int hook, struct net *net,
index 6965ba09eba70ac48c6e2a2dc0a08e9a9b46a042..33e37fb41d5d40df1779860e6a4f17ab6a47ef6f 100644 (file)
@@ -11,23 +11,30 @@ static inline bool nf_hook_ingress_active(const struct sk_buff *skb)
        if (!static_key_false(&nf_hooks_needed[NFPROTO_NETDEV][NF_NETDEV_INGRESS]))
                return false;
 #endif
-       return !list_empty(&skb->dev->nf_hooks_ingress);
+       return rcu_access_pointer(skb->dev->nf_hooks_ingress);
 }
 
 /* caller must hold rcu_read_lock */
 static inline int nf_hook_ingress(struct sk_buff *skb)
 {
+       struct nf_hook_entry *e = rcu_dereference(skb->dev->nf_hooks_ingress);
        struct nf_hook_state state;
 
-       nf_hook_state_init(&state, &skb->dev->nf_hooks_ingress,
-                          NF_NETDEV_INGRESS, INT_MIN, NFPROTO_NETDEV,
-                          skb->dev, NULL, NULL, dev_net(skb->dev), NULL);
+       /* Must recheck the ingress hook head, in the event it became NULL
+        * after the check in nf_hook_ingress_active evaluated to true.
+        */
+       if (unlikely(!e))
+               return 0;
+
+       nf_hook_state_init(&state, e, NF_NETDEV_INGRESS, INT_MIN,
+                          NFPROTO_NETDEV, skb->dev, NULL, NULL,
+                          dev_net(skb->dev), NULL);
        return nf_hook_slow(skb, &state);
 }
 
 static inline void nf_hook_ingress_init(struct net_device *dev)
 {
-       INIT_LIST_HEAD(&dev->nf_hooks_ingress);
+       RCU_INIT_POINTER(dev->nf_hooks_ingress, NULL);
 }
 #else /* CONFIG_NETFILTER_INGRESS */
 static inline int nf_hook_ingress_active(struct sk_buff *skb)
index 8fe85b98b5c8c7f5a537af1ba325e2939bab7863..2280cfe86c56157f3ca1165baf6f4ad0e73a18be 100644 (file)
@@ -11,7 +11,6 @@ struct nf_queue_entry {
        struct sk_buff          *skb;
        unsigned int            id;
 
-       struct nf_hook_ops      *elem;
        struct nf_hook_state    state;
        u16                     size; /* sizeof(entry) + saved route keys */
 
@@ -25,7 +24,7 @@ struct nf_queue_handler {
        int             (*outfn)(struct nf_queue_entry *entry,
                                 unsigned int queuenum);
        void            (*nf_hook_drop)(struct net *net,
-                                       struct nf_hook_ops *ops);
+                                       const struct nf_hook_entry *hooks);
 };
 
 void nf_register_queue_handler(struct net *net, const struct nf_queue_handler *qh);
index 36d723579af21f78e2da780c4840762ce88a4770..58487b1cc99ab09e9d5f00e1cc9608ac9db606e9 100644 (file)
@@ -16,6 +16,6 @@ struct netns_nf {
 #ifdef CONFIG_SYSCTL
        struct ctl_table_header *nf_log_dir_header;
 #endif
-       struct list_head hooks[NFPROTO_NUMPROTO][NF_MAX_HOOKS];
+       struct nf_hook_entry __rcu *hooks[NFPROTO_NUMPROTO][NF_MAX_HOOKS];
 };
 #endif
index 6029af47377d679b81307276edc5c539be8eb015..2fe9345c1407108bf5a0802671fa419aa9b24e7e 100644 (file)
@@ -1002,28 +1002,21 @@ int br_nf_hook_thresh(unsigned int hook, struct net *net,
                      int (*okfn)(struct net *, struct sock *,
                                  struct sk_buff *))
 {
-       struct nf_hook_ops *elem;
+       struct nf_hook_entry *elem;
        struct nf_hook_state state;
-       struct list_head *head;
        int ret;
 
-       head = &net->nf.hooks[NFPROTO_BRIDGE][hook];
+       elem = rcu_dereference(net->nf.hooks[NFPROTO_BRIDGE][hook]);
 
-       list_for_each_entry_rcu(elem, head, list) {
-               struct nf_hook_ops *next;
+       while (elem && (elem->ops.priority <= NF_BR_PRI_BRNF))
+               elem = rcu_dereference(elem->next);
 
-               next = list_entry_rcu(list_next_rcu(&elem->list),
-                                     struct nf_hook_ops, list);
-               if (next->priority <= NF_BR_PRI_BRNF)
-                       continue;
-       }
-
-       if (&elem->list == head)
+       if (!elem)
                return okfn(net, sk, skb);
 
        /* We may already have this, but read-locks nest anyway */
        rcu_read_lock();
-       nf_hook_state_init(&state, head, hook, NF_BR_PRI_BRNF + 1,
+       nf_hook_state_init(&state, elem, hook, NF_BR_PRI_BRNF + 1,
                           NFPROTO_BRIDGE, indev, outdev, sk, net, okfn);
 
        ret = nf_hook_slow(skb, &state);
index 67b74287535dfbb7059b04a9e8d4b7b180b070c5..72fc514ec676813af54e2360f144f544d0673e99 100644 (file)
@@ -22,6 +22,7 @@
 #include <linux/proc_fs.h>
 #include <linux/mutex.h>
 #include <linux/slab.h>
+#include <linux/rcupdate.h>
 #include <net/net_namespace.h>
 #include <net/sock.h>
 
@@ -61,33 +62,50 @@ EXPORT_SYMBOL(nf_hooks_needed);
 #endif
 
 static DEFINE_MUTEX(nf_hook_mutex);
+#define nf_entry_dereference(e) \
+       rcu_dereference_protected(e, lockdep_is_held(&nf_hook_mutex))
 
-static struct list_head *nf_find_hook_list(struct net *net,
-                                          const struct nf_hook_ops *reg)
+static struct nf_hook_entry *nf_hook_entry_head(struct net *net,
+                                               const struct nf_hook_ops *reg)
 {
-       struct list_head *hook_list = NULL;
+       struct nf_hook_entry *hook_head = NULL;
 
        if (reg->pf != NFPROTO_NETDEV)
-               hook_list = &net->nf.hooks[reg->pf][reg->hooknum];
+               hook_head = nf_entry_dereference(net->nf.hooks[reg->pf]
+                                                [reg->hooknum]);
        else if (reg->hooknum == NF_NETDEV_INGRESS) {
 #ifdef CONFIG_NETFILTER_INGRESS
                if (reg->dev && dev_net(reg->dev) == net)
-                       hook_list = &reg->dev->nf_hooks_ingress;
+                       hook_head =
+                               nf_entry_dereference(
+                                       reg->dev->nf_hooks_ingress);
 #endif
        }
-       return hook_list;
+       return hook_head;
 }
 
-struct nf_hook_entry {
-       const struct nf_hook_ops        *orig_ops;
-       struct nf_hook_ops              ops;
-};
+/* must hold nf_hook_mutex */
+static void nf_set_hooks_head(struct net *net, const struct nf_hook_ops *reg,
+                             struct nf_hook_entry *entry)
+{
+       switch (reg->pf) {
+       case NFPROTO_NETDEV:
+               /* We already checked in nf_register_net_hook() that this is
+                * used from ingress.
+                */
+               rcu_assign_pointer(reg->dev->nf_hooks_ingress, entry);
+               break;
+       default:
+               rcu_assign_pointer(net->nf.hooks[reg->pf][reg->hooknum],
+                                  entry);
+               break;
+       }
+}
 
 int nf_register_net_hook(struct net *net, const struct nf_hook_ops *reg)
 {
-       struct list_head *hook_list;
+       struct nf_hook_entry *hooks_entry;
        struct nf_hook_entry *entry;
-       struct nf_hook_ops *elem;
 
        if (reg->pf == NFPROTO_NETDEV &&
            (reg->hooknum != NF_NETDEV_INGRESS ||
@@ -100,19 +118,30 @@ int nf_register_net_hook(struct net *net, const struct nf_hook_ops *reg)
 
        entry->orig_ops = reg;
        entry->ops      = *reg;
+       entry->next     = NULL;
+
+       mutex_lock(&nf_hook_mutex);
+       hooks_entry = nf_hook_entry_head(net, reg);
 
-       hook_list = nf_find_hook_list(net, reg);
-       if (!hook_list) {
-               kfree(entry);
-               return -ENOENT;
+       if (hooks_entry && hooks_entry->orig_ops->priority > reg->priority) {
+               /* This is the case where we need to insert at the head */
+               entry->next = hooks_entry;
+               hooks_entry = NULL;
        }
 
-       mutex_lock(&nf_hook_mutex);
-       list_for_each_entry(elem, hook_list, list) {
-               if (reg->priority < elem->priority)
-                       break;
+       while (hooks_entry &&
+               reg->priority >= hooks_entry->orig_ops->priority &&
+               nf_entry_dereference(hooks_entry->next)) {
+               hooks_entry = nf_entry_dereference(hooks_entry->next);
+       }
+
+       if (hooks_entry) {
+               entry->next = nf_entry_dereference(hooks_entry->next);
+               rcu_assign_pointer(hooks_entry->next, entry);
+       } else {
+               nf_set_hooks_head(net, reg, entry);
        }
-       list_add_rcu(&entry->ops.list, elem->list.prev);
+
        mutex_unlock(&nf_hook_mutex);
 #ifdef CONFIG_NETFILTER_INGRESS
        if (reg->pf == NFPROTO_NETDEV && reg->hooknum == NF_NETDEV_INGRESS)
@@ -127,24 +156,33 @@ EXPORT_SYMBOL(nf_register_net_hook);
 
 void nf_unregister_net_hook(struct net *net, const struct nf_hook_ops *reg)
 {
-       struct list_head *hook_list;
-       struct nf_hook_entry *entry;
-       struct nf_hook_ops *elem;
-
-       hook_list = nf_find_hook_list(net, reg);
-       if (!hook_list)
-               return;
+       struct nf_hook_entry *hooks_entry;
 
        mutex_lock(&nf_hook_mutex);
-       list_for_each_entry(elem, hook_list, list) {
-               entry = container_of(elem, struct nf_hook_entry, ops);
-               if (entry->orig_ops == reg) {
-                       list_del_rcu(&entry->ops.list);
-                       break;
+       hooks_entry = nf_hook_entry_head(net, reg);
+       if (hooks_entry->orig_ops == reg) {
+               nf_set_hooks_head(net, reg,
+                                 nf_entry_dereference(hooks_entry->next));
+               goto unlock;
+       }
+       while (hooks_entry && nf_entry_dereference(hooks_entry->next)) {
+               struct nf_hook_entry *next =
+                       nf_entry_dereference(hooks_entry->next);
+               struct nf_hook_entry *nnext;
+
+               if (next->orig_ops != reg) {
+                       hooks_entry = next;
+                       continue;
                }
+               nnext = nf_entry_dereference(next->next);
+               rcu_assign_pointer(hooks_entry->next, nnext);
+               hooks_entry = next;
+               break;
        }
+
+unlock:
        mutex_unlock(&nf_hook_mutex);
-       if (&elem->list == hook_list) {
+       if (!hooks_entry) {
                WARN(1, "nf_unregister_net_hook: hook not found!\n");
                return;
        }
@@ -156,10 +194,10 @@ void nf_unregister_net_hook(struct net *net, const struct nf_hook_ops *reg)
        static_key_slow_dec(&nf_hooks_needed[reg->pf][reg->hooknum]);
 #endif
        synchronize_net();
-       nf_queue_nf_hook_drop(net, &entry->ops);
+       nf_queue_nf_hook_drop(net, hooks_entry);
        /* other cpu might still process nfqueue verdict that used reg */
        synchronize_net();
-       kfree(entry);
+       kfree(hooks_entry);
 }
 EXPORT_SYMBOL(nf_unregister_net_hook);
 
@@ -258,10 +296,9 @@ void nf_unregister_hooks(struct nf_hook_ops *reg, unsigned int n)
 }
 EXPORT_SYMBOL(nf_unregister_hooks);
 
-unsigned int nf_iterate(struct list_head *head,
-                       struct sk_buff *skb,
+unsigned int nf_iterate(struct sk_buff *skb,
                        struct nf_hook_state *state,
-                       struct nf_hook_ops **elemp)
+                       struct nf_hook_entry **entryp)
 {
        unsigned int verdict;
 
@@ -269,20 +306,23 @@ unsigned int nf_iterate(struct list_head *head,
         * The caller must not block between calls to this
         * function because of risk of continuing from deleted element.
         */
-       list_for_each_entry_continue_rcu((*elemp), head, list) {
-               if (state->thresh > (*elemp)->priority)
+       while (*entryp) {
+               if (state->thresh > (*entryp)->ops.priority) {
+                       *entryp = rcu_dereference((*entryp)->next);
                        continue;
+               }
 
                /* Optimization: we don't need to hold module
                   reference here, since function can't sleep. --RR */
 repeat:
-               verdict = (*elemp)->hook((*elemp)->priv, skb, state);
+               verdict = (*entryp)->ops.hook((*entryp)->ops.priv, skb, state);
                if (verdict != NF_ACCEPT) {
 #ifdef CONFIG_NETFILTER_DEBUG
                        if (unlikely((verdict & NF_VERDICT_MASK)
                                                        > NF_MAX_VERDICT)) {
                                NFDEBUG("Evil return from %p(%u).\n",
-                                       (*elemp)->hook, state->hook);
+                                       (*entryp)->ops.hook, state->hook);
+                               *entryp = rcu_dereference((*entryp)->next);
                                continue;
                        }
 #endif
@@ -290,6 +330,7 @@ repeat:
                                return verdict;
                        goto repeat;
                }
+               *entryp = rcu_dereference((*entryp)->next);
        }
        return NF_ACCEPT;
 }
@@ -299,13 +340,13 @@ repeat:
  * -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_ops *elem;
+       struct nf_hook_entry *entry;
        unsigned int verdict;
        int ret = 0;
 
-       elem = list_entry_rcu(state->hook_list, struct nf_hook_ops, list);
+       entry = rcu_dereference(state->hook_entries);
 next_hook:
-       verdict = nf_iterate(state->hook_list, skb, state, &elem);
+       verdict = nf_iterate(skb, state, &entry);
        if (verdict == NF_ACCEPT || verdict == NF_STOP) {
                ret = 1;
        } else if ((verdict & NF_VERDICT_MASK) == NF_DROP) {
@@ -314,8 +355,10 @@ next_hook:
                if (ret == 0)
                        ret = -EPERM;
        } else if ((verdict & NF_VERDICT_MASK) == NF_QUEUE) {
-               int err = nf_queue(skb, elem, state,
-                                  verdict >> NF_VERDICT_QBITS);
+               int err;
+
+               RCU_INIT_POINTER(state->hook_entries, entry);
+               err = nf_queue(skb, state, verdict >> NF_VERDICT_QBITS);
                if (err < 0) {
                        if (err == -ESRCH &&
                           (verdict & NF_VERDICT_FLAG_QUEUE_BYPASS))
@@ -442,7 +485,7 @@ static int __net_init netfilter_net_init(struct net *net)
 
        for (i = 0; i < ARRAY_SIZE(net->nf.hooks); i++) {
                for (h = 0; h < NF_MAX_HOOKS; h++)
-                       INIT_LIST_HEAD(&net->nf.hooks[i][h]);
+                       RCU_INIT_POINTER(net->nf.hooks[i][h], NULL);
        }
 
 #ifdef CONFIG_PROC_FS
index 065522564ac6a032ce1ece83be62438f7d8b0636..e0adb5959342148d9501a48f6bb92b90d2566c00 100644 (file)
 
 
 /* core.c */
-unsigned int nf_iterate(struct list_head *head, struct sk_buff *skb,
-                       struct nf_hook_state *state, struct nf_hook_ops **elemp);
+unsigned int nf_iterate(struct sk_buff *skb, struct nf_hook_state *state,
+                       struct nf_hook_entry **entryp);
 
 /* nf_queue.c */
-int nf_queue(struct sk_buff *skb, struct nf_hook_ops *elem,
-            struct nf_hook_state *state, unsigned int queuenum);
-void nf_queue_nf_hook_drop(struct net *net, struct nf_hook_ops *ops);
+int nf_queue(struct sk_buff *skb, struct nf_hook_state *state,
+            unsigned int queuenum);
+void nf_queue_nf_hook_drop(struct net *net, const struct nf_hook_entry *entry);
 int __init netfilter_queue_init(void);
 
 /* nf_log.c */
index b19ad20a705ca7845ad21a5d272accba190f3f38..96964a0070e11da46aa97b3fe94fb4f778e40418 100644 (file)
@@ -96,14 +96,14 @@ void nf_queue_entry_get_refs(struct nf_queue_entry *entry)
 }
 EXPORT_SYMBOL_GPL(nf_queue_entry_get_refs);
 
-void nf_queue_nf_hook_drop(struct net *net, struct nf_hook_ops *ops)
+void nf_queue_nf_hook_drop(struct net *net, const struct nf_hook_entry *entry)
 {
        const struct nf_queue_handler *qh;
 
        rcu_read_lock();
        qh = rcu_dereference(net->nf.queue_handler);
        if (qh)
-               qh->nf_hook_drop(net, ops);
+               qh->nf_hook_drop(net, entry);
        rcu_read_unlock();
 }
 
@@ -112,7 +112,6 @@ void nf_queue_nf_hook_drop(struct net *net, struct nf_hook_ops *ops)
  * through nf_reinject().
  */
 int nf_queue(struct sk_buff *skb,
-            struct nf_hook_ops *elem,
             struct nf_hook_state *state,
             unsigned int queuenum)
 {
@@ -141,7 +140,6 @@ int nf_queue(struct sk_buff *skb,
 
        *entry = (struct nf_queue_entry) {
                .skb    = skb,
-               .elem   = elem,
                .state  = *state,
                .size   = sizeof(*entry) + afinfo->route_key_size,
        };
@@ -165,11 +163,15 @@ err:
 
 void nf_reinject(struct nf_queue_entry *entry, unsigned int verdict)
 {
+       struct nf_hook_entry *hook_entry;
        struct sk_buff *skb = entry->skb;
-       struct nf_hook_ops *elem = entry->elem;
        const struct nf_afinfo *afinfo;
+       struct nf_hook_ops *elem;
        int err;
 
+       hook_entry = rcu_dereference(entry->state.hook_entries);
+       elem = &hook_entry->ops;
+
        nf_queue_entry_release_refs(entry);
 
        /* Continue traversal iff userspace said ok... */
@@ -186,8 +188,7 @@ void nf_reinject(struct nf_queue_entry *entry, unsigned int verdict)
 
        if (verdict == NF_ACCEPT) {
        next_hook:
-               verdict = nf_iterate(entry->state.hook_list,
-                                    skb, &entry->state, &elem);
+               verdict = nf_iterate(skb, &entry->state, &hook_entry);
        }
 
        switch (verdict & NF_VERDICT_MASK) {
@@ -198,7 +199,8 @@ void nf_reinject(struct nf_queue_entry *entry, unsigned int verdict)
                local_bh_enable();
                break;
        case NF_QUEUE:
-               err = nf_queue(skb, elem, &entry->state,
+               RCU_INIT_POINTER(entry->state.hook_entries, hook_entry);
+               err = nf_queue(skb, &entry->state,
                               verdict >> NF_VERDICT_QBITS);
                if (err < 0) {
                        if (err == -ESRCH &&
index 7caa8b082c41feedbf62fec563891a0ae7539244..af832c5260485f3f99a3dae8fdf0019f896350ef 100644 (file)
@@ -917,12 +917,14 @@ static struct notifier_block nfqnl_dev_notifier = {
        .notifier_call  = nfqnl_rcv_dev_event,
 };
 
-static int nf_hook_cmp(struct nf_queue_entry *entry, unsigned long ops_ptr)
+static int nf_hook_cmp(struct nf_queue_entry *entry, unsigned long entry_ptr)
 {
-       return entry->elem == (struct nf_hook_ops *)ops_ptr;
+       return rcu_access_pointer(entry->state.hook_entries) ==
+               (struct nf_hook_entry *)entry_ptr;
 }
 
-static void nfqnl_nf_hook_drop(struct net *net, struct nf_hook_ops *hook)
+static void nfqnl_nf_hook_drop(struct net *net,
+                              const struct nf_hook_entry *hook)
 {
        struct nfnl_queue_net *q = nfnl_queue_pernet(net);
        int i;