Merge branch 'master' of /repos/git/net-next-2.6
[safe/jmp/linux-2.6] / net / ipv6 / reassembly.c
index aef5dd1..d93812d 100644 (file)
@@ -5,8 +5,6 @@
  *     Authors:
  *     Pedro Roque             <roque@di.fc.ul.pt>
  *
- *     $Id: reassembly.c,v 1.26 2001/03/07 22:00:57 davem Exp $
- *
  *     Based on: net/ipv4/ip_fragment.c
  *
  *     This program is free software; you can redistribute it and/or
 #include <net/addrconf.h>
 #include <net/inet_frag.h>
 
-int sysctl_ip6frag_high_thresh __read_mostly = 256*1024;
-int sysctl_ip6frag_low_thresh __read_mostly = 192*1024;
-
-int sysctl_ip6frag_time __read_mostly = IPV6_FRAG_TIMEOUT;
-
 struct ip6frag_skb_cb
 {
        struct inet6_skb_parm   h;
@@ -79,6 +72,7 @@ struct frag_queue
        struct inet_frag_queue  q;
 
        __be32                  id;             /* fragment id          */
+       u32                     user;
        struct in6_addr         saddr;
        struct in6_addr         daddr;
 
@@ -87,39 +81,27 @@ struct frag_queue
        __u16                   nhoffset;
 };
 
-/* Hash table. */
-
-#define IP6Q_HASHSZ    64
-
-static struct hlist_head ip6_frag_hash[IP6Q_HASHSZ];
-static DEFINE_RWLOCK(ip6_frag_lock);
-static u32 ip6_frag_hash_rnd;
-static LIST_HEAD(ip6_frag_lru_list);
-int ip6_frag_nqueues = 0;
+static struct inet_frags ip6_frags;
 
-static int ip6_frag_reasm(struct frag_queue *fq, struct sk_buff *prev,
-                         struct net_device *dev);
-
-static __inline__ void __fq_unlink(struct frag_queue *fq)
+int ip6_frag_nqueues(struct net *net)
 {
-       hlist_del(&fq->q.list);
-       list_del(&fq->q.lru_list);
-       ip6_frag_nqueues--;
+       return net->ipv6.frags.nqueues;
 }
 
-static __inline__ void fq_unlink(struct frag_queue *fq)
+int ip6_frag_mem(struct net *net)
 {
-       write_lock(&ip6_frag_lock);
-       __fq_unlink(fq);
-       write_unlock(&ip6_frag_lock);
+       return atomic_read(&net->ipv6.frags.mem);
 }
 
+static int ip6_frag_reasm(struct frag_queue *fq, struct sk_buff *prev,
+                         struct net_device *dev);
+
 /*
  * callers should be careful not to use the hash value outside the ipfrag_lock
  * as doing so could race with ipfrag_hash_rnd being recalculated.
  */
-static unsigned int ip6qhashfn(__be32 id, struct in6_addr *saddr,
-                              struct in6_addr *daddr)
+unsigned int inet6_hash_frag(__be32 id, const struct in6_addr *saddr,
+                            const struct in6_addr *daddr, u32 rnd)
 {
        u32 a, b, c;
 
@@ -129,7 +111,7 @@ static unsigned int ip6qhashfn(__be32 id, struct in6_addr *saddr,
 
        a += JHASH_GOLDEN_RATIO;
        b += JHASH_GOLDEN_RATIO;
-       c += ip6_frag_hash_rnd;
+       c += rnd;
        __jhash_mix(a, b, c);
 
        a += (__force u32)saddr->s6_addr32[3];
@@ -142,98 +124,57 @@ static unsigned int ip6qhashfn(__be32 id, struct in6_addr *saddr,
        c += (__force u32)id;
        __jhash_mix(a, b, c);
 
-       return c & (IP6Q_HASHSZ - 1);
+       return c & (INETFRAGS_HASHSZ - 1);
 }
+EXPORT_SYMBOL_GPL(inet6_hash_frag);
 
-static struct timer_list ip6_frag_secret_timer;
-int sysctl_ip6frag_secret_interval __read_mostly = 10 * 60 * HZ;
-
-static void ip6_frag_secret_rebuild(unsigned long dummy)
+static unsigned int ip6_hashfn(struct inet_frag_queue *q)
 {
-       unsigned long now = jiffies;
-       int i;
-
-       write_lock(&ip6_frag_lock);
-       get_random_bytes(&ip6_frag_hash_rnd, sizeof(u32));
-       for (i = 0; i < IP6Q_HASHSZ; i++) {
-               struct frag_queue *q;
-               struct hlist_node *p, *n;
-
-               hlist_for_each_entry_safe(q, p, n, &ip6_frag_hash[i], q.list) {
-                       unsigned int hval = ip6qhashfn(q->id,
-                                                      &q->saddr,
-                                                      &q->daddr);
-
-                       if (hval != i) {
-                               hlist_del(&q->q.list);
+       struct frag_queue *fq;
 
-                               /* Relink to new hash chain. */
-                               hlist_add_head(&q->q.list,
-                                              &ip6_frag_hash[hval]);
+       fq = container_of(q, struct frag_queue, q);
+       return inet6_hash_frag(fq->id, &fq->saddr, &fq->daddr, ip6_frags.rnd);
+}
 
-                       }
-               }
-       }
-       write_unlock(&ip6_frag_lock);
+int ip6_frag_match(struct inet_frag_queue *q, void *a)
+{
+       struct frag_queue *fq;
+       struct ip6_create_arg *arg = a;
 
-       mod_timer(&ip6_frag_secret_timer, now + sysctl_ip6frag_secret_interval);
+       fq = container_of(q, struct frag_queue, q);
+       return (fq->id == arg->id && fq->user == arg->user &&
+                       ipv6_addr_equal(&fq->saddr, arg->src) &&
+                       ipv6_addr_equal(&fq->daddr, arg->dst));
 }
-
-atomic_t ip6_frag_mem = ATOMIC_INIT(0);
+EXPORT_SYMBOL(ip6_frag_match);
 
 /* Memory Tracking Functions. */
-static inline void frag_kfree_skb(struct sk_buff *skb, int *work)
+static inline void frag_kfree_skb(struct netns_frags *nf,
+               struct sk_buff *skb, int *work)
 {
        if (work)
                *work -= skb->truesize;
-       atomic_sub(skb->truesize, &ip6_frag_mem);
+       atomic_sub(skb->truesize, &nf->mem);
        kfree_skb(skb);
 }
 
-static inline void frag_free_queue(struct frag_queue *fq, int *work)
-{
-       if (work)
-               *work -= sizeof(struct frag_queue);
-       atomic_sub(sizeof(struct frag_queue), &ip6_frag_mem);
-       kfree(fq);
-}
-
-static inline struct frag_queue *frag_alloc_queue(void)
+void ip6_frag_init(struct inet_frag_queue *q, void *a)
 {
-       struct frag_queue *fq = kzalloc(sizeof(struct frag_queue), GFP_ATOMIC);
+       struct frag_queue *fq = container_of(q, struct frag_queue, q);
+       struct ip6_create_arg *arg = a;
 
-       if(!fq)
-               return NULL;
-       atomic_add(sizeof(struct frag_queue), &ip6_frag_mem);
-       return fq;
+       fq->id = arg->id;
+       fq->user = arg->user;
+       ipv6_addr_copy(&fq->saddr, arg->src);
+       ipv6_addr_copy(&fq->daddr, arg->dst);
 }
+EXPORT_SYMBOL(ip6_frag_init);
 
 /* Destruction primitives. */
 
-/* Complete destruction of fq. */
-static void ip6_frag_destroy(struct frag_queue *fq, int *work)
-{
-       struct sk_buff *fp;
-
-       BUG_TRAP(fq->q.last_in&COMPLETE);
-       BUG_TRAP(del_timer(&fq->q.timer) == 0);
-
-       /* Release all fragment data. */
-       fp = fq->q.fragments;
-       while (fp) {
-               struct sk_buff *xp = fp->next;
-
-               frag_kfree_skb(fp, work);
-               fp = xp;
-       }
-
-       frag_free_queue(fq, work);
-}
-
-static __inline__ void fq_put(struct frag_queue *fq, int *work)
+static __inline__ void fq_put(struct frag_queue *fq)
 {
-       if (atomic_dec_and_test(&fq->q.refcnt))
-               ip6_frag_destroy(fq, work);
+       inet_frag_put(&fq->q, &ip6_frags);
 }
 
 /* Kill fq entry. It is not destroyed immediately,
@@ -241,71 +182,45 @@ static __inline__ void fq_put(struct frag_queue *fq, int *work)
  */
 static __inline__ void fq_kill(struct frag_queue *fq)
 {
-       if (del_timer(&fq->q.timer))
-               atomic_dec(&fq->q.refcnt);
-
-       if (!(fq->q.last_in & COMPLETE)) {
-               fq_unlink(fq);
-               atomic_dec(&fq->q.refcnt);
-               fq->q.last_in |= COMPLETE;
-       }
+       inet_frag_kill(&fq->q, &ip6_frags);
 }
 
-static void ip6_evictor(struct inet6_dev *idev)
+static void ip6_evictor(struct net *net, struct inet6_dev *idev)
 {
-       struct frag_queue *fq;
-       struct list_head *tmp;
-       int work;
-
-       work = atomic_read(&ip6_frag_mem) - sysctl_ip6frag_low_thresh;
-       if (work <= 0)
-               return;
-
-       while(work > 0) {
-               read_lock(&ip6_frag_lock);
-               if (list_empty(&ip6_frag_lru_list)) {
-                       read_unlock(&ip6_frag_lock);
-                       return;
-               }
-               tmp = ip6_frag_lru_list.next;
-               fq = list_entry(tmp, struct frag_queue, q.lru_list);
-               atomic_inc(&fq->q.refcnt);
-               read_unlock(&ip6_frag_lock);
+       int evicted;
 
-               spin_lock(&fq->q.lock);
-               if (!(fq->q.last_in&COMPLETE))
-                       fq_kill(fq);
-               spin_unlock(&fq->q.lock);
-
-               fq_put(fq, &work);
-               IP6_INC_STATS_BH(idev, IPSTATS_MIB_REASMFAILS);
-       }
+       evicted = inet_frag_evictor(&net->ipv6.frags, &ip6_frags);
+       if (evicted)
+               IP6_ADD_STATS_BH(net, idev, IPSTATS_MIB_REASMFAILS, evicted);
 }
 
 static void ip6_frag_expire(unsigned long data)
 {
-       struct frag_queue *fq = (struct frag_queue *) data;
+       struct frag_queue *fq;
        struct net_device *dev = NULL;
+       struct net *net;
+
+       fq = container_of((struct inet_frag_queue *)data, struct frag_queue, q);
 
        spin_lock(&fq->q.lock);
 
-       if (fq->q.last_in & COMPLETE)
+       if (fq->q.last_in & INET_FRAG_COMPLETE)
                goto out;
 
        fq_kill(fq);
 
-       dev = dev_get_by_index(&init_net, fq->iif);
+       net = container_of(fq->q.net, struct net, ipv6.frags);
+       rcu_read_lock();
+       dev = dev_get_by_index_rcu(net, fq->iif);
        if (!dev)
-               goto out;
+               goto out_rcu_unlock;
 
-       rcu_read_lock();
-       IP6_INC_STATS_BH(__in6_dev_get(dev), IPSTATS_MIB_REASMTIMEOUT);
-       IP6_INC_STATS_BH(__in6_dev_get(dev), IPSTATS_MIB_REASMFAILS);
-       rcu_read_unlock();
+       IP6_INC_STATS_BH(net, __in6_dev_get(dev), IPSTATS_MIB_REASMTIMEOUT);
+       IP6_INC_STATS_BH(net, __in6_dev_get(dev), IPSTATS_MIB_REASMFAILS);
 
        /* Don't send error if the first segment did not arrive. */
-       if (!(fq->q.last_in&FIRST_IN) || !fq->q.fragments)
-               goto out;
+       if (!(fq->q.last_in & INET_FRAG_FIRST_IN) || !fq->q.fragments)
+               goto out_rcu_unlock;
 
        /*
           But use as source device on which LAST ARRIVED
@@ -314,113 +229,49 @@ static void ip6_frag_expire(unsigned long data)
         */
        fq->q.fragments->dev = dev;
        icmpv6_send(fq->q.fragments, ICMPV6_TIME_EXCEED, ICMPV6_EXC_FRAGTIME, 0, dev);
+out_rcu_unlock:
+       rcu_read_unlock();
 out:
-       if (dev)
-               dev_put(dev);
        spin_unlock(&fq->q.lock);
-       fq_put(fq, NULL);
+       fq_put(fq);
 }
 
-/* Creation primitives. */
-
-
-static struct frag_queue *ip6_frag_intern(struct frag_queue *fq_in)
+static __inline__ struct frag_queue *
+fq_find(struct net *net, __be32 id, struct in6_addr *src, struct in6_addr *dst,
+       struct inet6_dev *idev)
 {
-       struct frag_queue *fq;
+       struct inet_frag_queue *q;
+       struct ip6_create_arg arg;
        unsigned int hash;
-#ifdef CONFIG_SMP
-       struct hlist_node *n;
-#endif
 
-       write_lock(&ip6_frag_lock);
-       hash = ip6qhashfn(fq_in->id, &fq_in->saddr, &fq_in->daddr);
-#ifdef CONFIG_SMP
-       hlist_for_each_entry(fq, n, &ip6_frag_hash[hash], q.list) {
-               if (fq->id == fq_in->id &&
-                   ipv6_addr_equal(&fq_in->saddr, &fq->saddr) &&
-                   ipv6_addr_equal(&fq_in->daddr, &fq->daddr)) {
-                       atomic_inc(&fq->q.refcnt);
-                       write_unlock(&ip6_frag_lock);
-                       fq_in->q.last_in |= COMPLETE;
-                       fq_put(fq_in, NULL);
-                       return fq;
-               }
-       }
-#endif
-       fq = fq_in;
-
-       if (!mod_timer(&fq->q.timer, jiffies + sysctl_ip6frag_time))
-               atomic_inc(&fq->q.refcnt);
-
-       atomic_inc(&fq->q.refcnt);
-       hlist_add_head(&fq->q.list, &ip6_frag_hash[hash]);
-       INIT_LIST_HEAD(&fq->q.lru_list);
-       list_add_tail(&fq->q.lru_list, &ip6_frag_lru_list);
-       ip6_frag_nqueues++;
-       write_unlock(&ip6_frag_lock);
-       return fq;
-}
+       arg.id = id;
+       arg.user = IP6_DEFRAG_LOCAL_DELIVER;
+       arg.src = src;
+       arg.dst = dst;
 
+       read_lock(&ip6_frags.lock);
+       hash = inet6_hash_frag(id, src, dst, ip6_frags.rnd);
 
-static struct frag_queue *
-ip6_frag_create(__be32 id, struct in6_addr *src, struct in6_addr *dst,
-               struct inet6_dev *idev)
-{
-       struct frag_queue *fq;
-
-       if ((fq = frag_alloc_queue()) == NULL)
+       q = inet_frag_find(&net->ipv6.frags, &ip6_frags, &arg, hash);
+       if (q == NULL)
                goto oom;
 
-       fq->id = id;
-       ipv6_addr_copy(&fq->saddr, src);
-       ipv6_addr_copy(&fq->daddr, dst);
-
-       init_timer(&fq->q.timer);
-       fq->q.timer.function = ip6_frag_expire;
-       fq->q.timer.data = (long) fq;
-       spin_lock_init(&fq->q.lock);
-       atomic_set(&fq->q.refcnt, 1);
-
-       return ip6_frag_intern(fq);
+       return container_of(q, struct frag_queue, q);
 
 oom:
-       IP6_INC_STATS_BH(idev, IPSTATS_MIB_REASMFAILS);
+       IP6_INC_STATS_BH(net, idev, IPSTATS_MIB_REASMFAILS);
        return NULL;
 }
 
-static __inline__ struct frag_queue *
-fq_find(__be32 id, struct in6_addr *src, struct in6_addr *dst,
-       struct inet6_dev *idev)
-{
-       struct frag_queue *fq;
-       struct hlist_node *n;
-       unsigned int hash;
-
-       read_lock(&ip6_frag_lock);
-       hash = ip6qhashfn(id, src, dst);
-       hlist_for_each_entry(fq, n, &ip6_frag_hash[hash], q.list) {
-               if (fq->id == id &&
-                   ipv6_addr_equal(src, &fq->saddr) &&
-                   ipv6_addr_equal(dst, &fq->daddr)) {
-                       atomic_inc(&fq->q.refcnt);
-                       read_unlock(&ip6_frag_lock);
-                       return fq;
-               }
-       }
-       read_unlock(&ip6_frag_lock);
-
-       return ip6_frag_create(id, src, dst, idev);
-}
-
-
 static int ip6_frag_queue(struct frag_queue *fq, struct sk_buff *skb,
                           struct frag_hdr *fhdr, int nhoff)
 {
        struct sk_buff *prev, *next;
        struct net_device *dev;
        int offset, end;
+       struct net *net = dev_net(skb_dst(skb)->dev);
 
-       if (fq->q.last_in & COMPLETE)
+       if (fq->q.last_in & INET_FRAG_COMPLETE)
                goto err;
 
        offset = ntohs(fhdr->frag_off) & ~0x7;
@@ -428,7 +279,7 @@ static int ip6_frag_queue(struct frag_queue *fq, struct sk_buff *skb,
                        ((u8 *)(fhdr + 1) - (u8 *)(ipv6_hdr(skb) + 1)));
 
        if ((unsigned int)end > IPV6_MAXPLEN) {
-               IP6_INC_STATS_BH(ip6_dst_idev(skb->dst),
+               IP6_INC_STATS_BH(net, ip6_dst_idev(skb_dst(skb)),
                                 IPSTATS_MIB_INHDRERRORS);
                icmpv6_param_prob(skb, ICMPV6_HDR_FIELD,
                                  ((u8 *)&fhdr->frag_off -
@@ -449,9 +300,9 @@ static int ip6_frag_queue(struct frag_queue *fq, struct sk_buff *skb,
                 * or have different end, the segment is corrupted.
                 */
                if (end < fq->q.len ||
-                   ((fq->q.last_in & LAST_IN) && end != fq->q.len))
+                   ((fq->q.last_in & INET_FRAG_LAST_IN) && end != fq->q.len))
                        goto err;
-               fq->q.last_in |= LAST_IN;
+               fq->q.last_in |= INET_FRAG_LAST_IN;
                fq->q.len = end;
        } else {
                /* Check if the fragment is rounded to 8 bytes.
@@ -461,7 +312,7 @@ static int ip6_frag_queue(struct frag_queue *fq, struct sk_buff *skb,
                        /* RFC2460 says always send parameter problem in
                         * this case. -DaveM
                         */
-                       IP6_INC_STATS_BH(ip6_dst_idev(skb->dst),
+                       IP6_INC_STATS_BH(net, ip6_dst_idev(skb_dst(skb)),
                                         IPSTATS_MIB_INHDRERRORS);
                        icmpv6_param_prob(skb, ICMPV6_HDR_FIELD,
                                          offsetof(struct ipv6hdr, payload_len));
@@ -469,7 +320,7 @@ static int ip6_frag_queue(struct frag_queue *fq, struct sk_buff *skb,
                }
                if (end > fq->q.len) {
                        /* Some bits beyond end -> corruption. */
-                       if (fq->q.last_in & LAST_IN)
+                       if (fq->q.last_in & INET_FRAG_LAST_IN)
                                goto err;
                        fq->q.len = end;
                }
@@ -545,7 +396,7 @@ static int ip6_frag_queue(struct frag_queue *fq, struct sk_buff *skb,
                                fq->q.fragments = next;
 
                        fq->q.meat -= free_it->len;
-                       frag_kfree_skb(free_it, NULL);
+                       frag_kfree_skb(fq->q.net, free_it, NULL);
                }
        }
 
@@ -565,26 +416,28 @@ static int ip6_frag_queue(struct frag_queue *fq, struct sk_buff *skb,
        }
        fq->q.stamp = skb->tstamp;
        fq->q.meat += skb->len;
-       atomic_add(skb->truesize, &ip6_frag_mem);
+       atomic_add(skb->truesize, &fq->q.net->mem);
 
        /* The first fragment.
         * nhoffset is obtained from the first fragment, of course.
         */
        if (offset == 0) {
                fq->nhoffset = nhoff;
-               fq->q.last_in |= FIRST_IN;
+               fq->q.last_in |= INET_FRAG_FIRST_IN;
        }
 
-       if (fq->q.last_in == (FIRST_IN | LAST_IN) && fq->q.meat == fq->q.len)
+       if (fq->q.last_in == (INET_FRAG_FIRST_IN | INET_FRAG_LAST_IN) &&
+           fq->q.meat == fq->q.len)
                return ip6_frag_reasm(fq, prev, dev);
 
-       write_lock(&ip6_frag_lock);
-       list_move_tail(&fq->q.lru_list, &ip6_frag_lru_list);
-       write_unlock(&ip6_frag_lock);
+       write_lock(&ip6_frags.lock);
+       list_move_tail(&fq->q.lru_list, &fq->q.net->lru_list);
+       write_unlock(&ip6_frags.lock);
        return -1;
 
 err:
-       IP6_INC_STATS(ip6_dst_idev(skb->dst), IPSTATS_MIB_REASMFAILS);
+       IP6_INC_STATS(net, ip6_dst_idev(skb_dst(skb)),
+                     IPSTATS_MIB_REASMFAILS);
        kfree_skb(skb);
        return -1;
 }
@@ -601,6 +454,7 @@ err:
 static int ip6_frag_reasm(struct frag_queue *fq, struct sk_buff *prev,
                          struct net_device *dev)
 {
+       struct net *net = container_of(fq->q.net, struct net, ipv6.frags);
        struct sk_buff *fp, *head = fq->q.fragments;
        int    payload_len;
        unsigned int nhoff;
@@ -625,8 +479,8 @@ static int ip6_frag_reasm(struct frag_queue *fq, struct sk_buff *prev,
                fq->q.fragments = head;
        }
 
-       BUG_TRAP(head != NULL);
-       BUG_TRAP(FRAG6_CB(head)->offset == 0);
+       WARN_ON(head == NULL);
+       WARN_ON(FRAG6_CB(head)->offset != 0);
 
        /* Unfragmented part is taken from the first segment. */
        payload_len = ((head->data - skb_network_header(head)) -
@@ -642,7 +496,7 @@ static int ip6_frag_reasm(struct frag_queue *fq, struct sk_buff *prev,
        /* If the first fragment is fragmented itself, we split
         * it to two chunks: the first with data and paged part
         * and the second, holding only fragments. */
-       if (skb_shinfo(head)->frag_list) {
+       if (skb_has_frags(head)) {
                struct sk_buff *clone;
                int i, plen = 0;
 
@@ -651,7 +505,7 @@ static int ip6_frag_reasm(struct frag_queue *fq, struct sk_buff *prev,
                clone->next = head->next;
                head->next = clone;
                skb_shinfo(clone)->frag_list = skb_shinfo(head)->frag_list;
-               skb_shinfo(head)->frag_list = NULL;
+               skb_frag_list_init(head);
                for (i=0; i<skb_shinfo(head)->nr_frags; i++)
                        plen += skb_shinfo(head)->frags[i].size;
                clone->len = clone->data_len = head->data_len - plen;
@@ -659,7 +513,7 @@ static int ip6_frag_reasm(struct frag_queue *fq, struct sk_buff *prev,
                head->len -= clone->len;
                clone->csum = 0;
                clone->ip_summed = head->ip_summed;
-               atomic_add(clone->truesize, &ip6_frag_mem);
+               atomic_add(clone->truesize, &fq->q.net->mem);
        }
 
        /* We have to remove fragment header from datagram and to relocate
@@ -674,7 +528,7 @@ static int ip6_frag_reasm(struct frag_queue *fq, struct sk_buff *prev,
        skb_shinfo(head)->frag_list = head->next;
        skb_reset_transport_header(head);
        skb_push(head, head->data - skb_network_header(head));
-       atomic_sub(head->truesize, &ip6_frag_mem);
+       atomic_sub(head->truesize, &fq->q.net->mem);
 
        for (fp=head->next; fp; fp = fp->next) {
                head->data_len += fp->len;
@@ -684,7 +538,7 @@ static int ip6_frag_reasm(struct frag_queue *fq, struct sk_buff *prev,
                else if (head->ip_summed == CHECKSUM_COMPLETE)
                        head->csum = csum_add(head->csum, fp->csum);
                head->truesize += fp->truesize;
-               atomic_sub(fp->truesize, &ip6_frag_mem);
+               atomic_sub(fp->truesize, &fq->q.net->mem);
        }
 
        head->next = NULL;
@@ -700,7 +554,7 @@ static int ip6_frag_reasm(struct frag_queue *fq, struct sk_buff *prev,
                                          head->csum);
 
        rcu_read_lock();
-       IP6_INC_STATS_BH(__in6_dev_get(dev), IPSTATS_MIB_REASMOKS);
+       IP6_INC_STATS_BH(net, __in6_dev_get(dev), IPSTATS_MIB_REASMOKS);
        rcu_read_unlock();
        fq->q.fragments = NULL;
        return 1;
@@ -714,34 +568,27 @@ out_oom:
                printk(KERN_DEBUG "ip6_frag_reasm: no memory for reassembly\n");
 out_fail:
        rcu_read_lock();
-       IP6_INC_STATS_BH(__in6_dev_get(dev), IPSTATS_MIB_REASMFAILS);
+       IP6_INC_STATS_BH(net, __in6_dev_get(dev), IPSTATS_MIB_REASMFAILS);
        rcu_read_unlock();
        return -1;
 }
 
-static int ipv6_frag_rcv(struct sk_buff **skbp)
+static int ipv6_frag_rcv(struct sk_buff *skb)
 {
-       struct sk_buff *skb = *skbp;
        struct frag_hdr *fhdr;
        struct frag_queue *fq;
        struct ipv6hdr *hdr = ipv6_hdr(skb);
+       struct net *net = dev_net(skb_dst(skb)->dev);
 
-       IP6_INC_STATS_BH(ip6_dst_idev(skb->dst), IPSTATS_MIB_REASMREQDS);
+       IP6_INC_STATS_BH(net, ip6_dst_idev(skb_dst(skb)), IPSTATS_MIB_REASMREQDS);
 
        /* Jumbo payload inhibits frag. header */
-       if (hdr->payload_len==0) {
-               IP6_INC_STATS(ip6_dst_idev(skb->dst), IPSTATS_MIB_INHDRERRORS);
-               icmpv6_param_prob(skb, ICMPV6_HDR_FIELD,
-                                 skb_network_header_len(skb));
-               return -1;
-       }
+       if (hdr->payload_len==0)
+               goto fail_hdr;
+
        if (!pskb_may_pull(skb, (skb_transport_offset(skb) +
-                                sizeof(struct frag_hdr)))) {
-               IP6_INC_STATS(ip6_dst_idev(skb->dst), IPSTATS_MIB_INHDRERRORS);
-               icmpv6_param_prob(skb, ICMPV6_HDR_FIELD,
-                                 skb_network_header_len(skb));
-               return -1;
-       }
+                                sizeof(struct frag_hdr))))
+               goto fail_hdr;
 
        hdr = ipv6_hdr(skb);
        fhdr = (struct frag_hdr *)skb_transport_header(skb);
@@ -749,17 +596,18 @@ static int ipv6_frag_rcv(struct sk_buff **skbp)
        if (!(fhdr->frag_off & htons(0xFFF9))) {
                /* It is not a fragmented frame */
                skb->transport_header += sizeof(struct frag_hdr);
-               IP6_INC_STATS_BH(ip6_dst_idev(skb->dst), IPSTATS_MIB_REASMOKS);
+               IP6_INC_STATS_BH(net,
+                                ip6_dst_idev(skb_dst(skb)), IPSTATS_MIB_REASMOKS);
 
                IP6CB(skb)->nhoff = (u8 *)fhdr - skb_network_header(skb);
                return 1;
        }
 
-       if (atomic_read(&ip6_frag_mem) > sysctl_ip6frag_high_thresh)
-               ip6_evictor(ip6_dst_idev(skb->dst));
+       if (atomic_read(&net->ipv6.frags.mem) > net->ipv6.frags.high_thresh)
+               ip6_evictor(net, ip6_dst_idev(skb_dst(skb)));
 
-       if ((fq = fq_find(fhdr->identification, &hdr->saddr, &hdr->daddr,
-                         ip6_dst_idev(skb->dst))) != NULL) {
+       if ((fq = fq_find(net, fhdr->identification, &hdr->saddr, &hdr->daddr,
+                         ip6_dst_idev(skb_dst(skb)))) != NULL) {
                int ret;
 
                spin_lock(&fq->q.lock);
@@ -767,31 +615,197 @@ static int ipv6_frag_rcv(struct sk_buff **skbp)
                ret = ip6_frag_queue(fq, skb, fhdr, IP6CB(skb)->nhoff);
 
                spin_unlock(&fq->q.lock);
-               fq_put(fq, NULL);
+               fq_put(fq);
                return ret;
        }
 
-       IP6_INC_STATS_BH(ip6_dst_idev(skb->dst), IPSTATS_MIB_REASMFAILS);
+       IP6_INC_STATS_BH(net, ip6_dst_idev(skb_dst(skb)), IPSTATS_MIB_REASMFAILS);
        kfree_skb(skb);
        return -1;
+
+fail_hdr:
+       IP6_INC_STATS(net, ip6_dst_idev(skb_dst(skb)), IPSTATS_MIB_INHDRERRORS);
+       icmpv6_param_prob(skb, ICMPV6_HDR_FIELD, skb_network_header_len(skb));
+       return -1;
 }
 
-static struct inet6_protocol frag_protocol =
+static const struct inet6_protocol frag_protocol =
 {
        .handler        =       ipv6_frag_rcv,
        .flags          =       INET6_PROTO_NOPOLICY,
 };
 
-void __init ipv6_frag_init(void)
+#ifdef CONFIG_SYSCTL
+static struct ctl_table ip6_frags_ns_ctl_table[] = {
+       {
+               .procname       = "ip6frag_high_thresh",
+               .data           = &init_net.ipv6.frags.high_thresh,
+               .maxlen         = sizeof(int),
+               .mode           = 0644,
+               .proc_handler   = proc_dointvec
+       },
+       {
+               .procname       = "ip6frag_low_thresh",
+               .data           = &init_net.ipv6.frags.low_thresh,
+               .maxlen         = sizeof(int),
+               .mode           = 0644,
+               .proc_handler   = proc_dointvec
+       },
+       {
+               .procname       = "ip6frag_time",
+               .data           = &init_net.ipv6.frags.timeout,
+               .maxlen         = sizeof(int),
+               .mode           = 0644,
+               .proc_handler   = proc_dointvec_jiffies,
+       },
+       { }
+};
+
+static struct ctl_table ip6_frags_ctl_table[] = {
+       {
+               .procname       = "ip6frag_secret_interval",
+               .data           = &ip6_frags.secret_interval,
+               .maxlen         = sizeof(int),
+               .mode           = 0644,
+               .proc_handler   = proc_dointvec_jiffies,
+       },
+       { }
+};
+
+static int __net_init ip6_frags_ns_sysctl_register(struct net *net)
+{
+       struct ctl_table *table;
+       struct ctl_table_header *hdr;
+
+       table = ip6_frags_ns_ctl_table;
+       if (!net_eq(net, &init_net)) {
+               table = kmemdup(table, sizeof(ip6_frags_ns_ctl_table), GFP_KERNEL);
+               if (table == NULL)
+                       goto err_alloc;
+
+               table[0].data = &net->ipv6.frags.high_thresh;
+               table[1].data = &net->ipv6.frags.low_thresh;
+               table[2].data = &net->ipv6.frags.timeout;
+       }
+
+       hdr = register_net_sysctl_table(net, net_ipv6_ctl_path, table);
+       if (hdr == NULL)
+               goto err_reg;
+
+       net->ipv6.sysctl.frags_hdr = hdr;
+       return 0;
+
+err_reg:
+       if (!net_eq(net, &init_net))
+               kfree(table);
+err_alloc:
+       return -ENOMEM;
+}
+
+static void __net_exit ip6_frags_ns_sysctl_unregister(struct net *net)
+{
+       struct ctl_table *table;
+
+       table = net->ipv6.sysctl.frags_hdr->ctl_table_arg;
+       unregister_net_sysctl_table(net->ipv6.sysctl.frags_hdr);
+       if (!net_eq(net, &init_net))
+               kfree(table);
+}
+
+static struct ctl_table_header *ip6_ctl_header;
+
+static int ip6_frags_sysctl_register(void)
+{
+       ip6_ctl_header = register_net_sysctl_rotable(net_ipv6_ctl_path,
+                       ip6_frags_ctl_table);
+       return ip6_ctl_header == NULL ? -ENOMEM : 0;
+}
+
+static void ip6_frags_sysctl_unregister(void)
+{
+       unregister_net_sysctl_table(ip6_ctl_header);
+}
+#else
+static inline int ip6_frags_ns_sysctl_register(struct net *net)
+{
+       return 0;
+}
+
+static inline void ip6_frags_ns_sysctl_unregister(struct net *net)
+{
+}
+
+static inline int ip6_frags_sysctl_register(void)
+{
+       return 0;
+}
+
+static inline void ip6_frags_sysctl_unregister(void)
+{
+}
+#endif
+
+static int __net_init ipv6_frags_init_net(struct net *net)
+{
+       net->ipv6.frags.high_thresh = IPV6_FRAG_HIGH_THRESH;
+       net->ipv6.frags.low_thresh = IPV6_FRAG_LOW_THRESH;
+       net->ipv6.frags.timeout = IPV6_FRAG_TIMEOUT;
+
+       inet_frags_init_net(&net->ipv6.frags);
+
+       return ip6_frags_ns_sysctl_register(net);
+}
+
+static void __net_exit ipv6_frags_exit_net(struct net *net)
+{
+       ip6_frags_ns_sysctl_unregister(net);
+       inet_frags_exit_net(&net->ipv6.frags, &ip6_frags);
+}
+
+static struct pernet_operations ip6_frags_ops = {
+       .init = ipv6_frags_init_net,
+       .exit = ipv6_frags_exit_net,
+};
+
+int __init ipv6_frag_init(void)
 {
-       if (inet6_add_protocol(&frag_protocol, IPPROTO_FRAGMENT) < 0)
-               printk(KERN_ERR "ipv6_frag_init: Could not register protocol\n");
+       int ret;
+
+       ret = inet6_add_protocol(&frag_protocol, IPPROTO_FRAGMENT);
+       if (ret)
+               goto out;
 
-       ip6_frag_hash_rnd = (u32) ((num_physpages ^ (num_physpages>>7)) ^
-                                  (jiffies ^ (jiffies >> 6)));
+       ret = ip6_frags_sysctl_register();
+       if (ret)
+               goto err_sysctl;
+
+       ret = register_pernet_subsys(&ip6_frags_ops);
+       if (ret)
+               goto err_pernet;
+
+       ip6_frags.hashfn = ip6_hashfn;
+       ip6_frags.constructor = ip6_frag_init;
+       ip6_frags.destructor = NULL;
+       ip6_frags.skb_free = NULL;
+       ip6_frags.qsize = sizeof(struct frag_queue);
+       ip6_frags.match = ip6_frag_match;
+       ip6_frags.frag_expire = ip6_frag_expire;
+       ip6_frags.secret_interval = 10 * 60 * HZ;
+       inet_frags_init(&ip6_frags);
+out:
+       return ret;
+
+err_pernet:
+       ip6_frags_sysctl_unregister();
+err_sysctl:
+       inet6_del_protocol(&frag_protocol, IPPROTO_FRAGMENT);
+       goto out;
+}
 
-       init_timer(&ip6_frag_secret_timer);
-       ip6_frag_secret_timer.function = ip6_frag_secret_rebuild;
-       ip6_frag_secret_timer.expires = jiffies + sysctl_ip6frag_secret_interval;
-       add_timer(&ip6_frag_secret_timer);
+void ipv6_frag_exit(void)
+{
+       inet_frags_fini(&ip6_frags);
+       ip6_frags_sysctl_unregister();
+       unregister_pernet_subsys(&ip6_frags_ops);
+       inet6_del_protocol(&frag_protocol, IPPROTO_FRAGMENT);
 }