Merge branch 'master' of /repos/git/net-next-2.6
[safe/jmp/linux-2.6] / net / ipv6 / reassembly.c
index f936d04..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
@@ -74,6 +72,7 @@ struct frag_queue
        struct inet_frag_queue  q;
 
        __be32                  id;             /* fragment id          */
+       u32                     user;
        struct in6_addr         saddr;
        struct in6_addr         daddr;
 
@@ -101,8 +100,8 @@ static int ip6_frag_reasm(struct frag_queue *fq, struct sk_buff *prev,
  * 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;
 
@@ -112,7 +111,7 @@ static unsigned int ip6qhashfn(__be32 id, struct in6_addr *saddr,
 
        a += JHASH_GOLDEN_RATIO;
        b += JHASH_GOLDEN_RATIO;
-       c += ip6_frags.rnd;
+       c += rnd;
        __jhash_mix(a, b, c);
 
        a += (__force u32)saddr->s6_addr32[3];
@@ -127,13 +126,14 @@ static unsigned int ip6qhashfn(__be32 id, struct in6_addr *saddr,
 
        return c & (INETFRAGS_HASHSZ - 1);
 }
+EXPORT_SYMBOL_GPL(inet6_hash_frag);
 
 static unsigned int ip6_hashfn(struct inet_frag_queue *q)
 {
        struct frag_queue *fq;
 
        fq = container_of(q, struct frag_queue, q);
-       return ip6qhashfn(fq->id, &fq->saddr, &fq->daddr);
+       return inet6_hash_frag(fq->id, &fq->saddr, &fq->daddr, ip6_frags.rnd);
 }
 
 int ip6_frag_match(struct inet_frag_queue *q, void *a)
@@ -142,7 +142,7 @@ int ip6_frag_match(struct inet_frag_queue *q, void *a)
        struct ip6_create_arg *arg = a;
 
        fq = container_of(q, struct frag_queue, q);
-       return (fq->id == arg->id &&
+       return (fq->id == arg->id && fq->user == arg->user &&
                        ipv6_addr_equal(&fq->saddr, arg->src) &&
                        ipv6_addr_equal(&fq->daddr, arg->dst));
 }
@@ -164,6 +164,7 @@ void ip6_frag_init(struct inet_frag_queue *q, void *a)
        struct ip6_create_arg *arg = a;
 
        fq->id = arg->id;
+       fq->user = arg->user;
        ipv6_addr_copy(&fq->saddr, arg->src);
        ipv6_addr_copy(&fq->daddr, arg->dst);
 }
@@ -190,35 +191,36 @@ static void ip6_evictor(struct net *net, struct inet6_dev *idev)
 
        evicted = inet_frag_evictor(&net->ipv6.frags, &ip6_frags);
        if (evicted)
-               IP6_ADD_STATS_BH(idev, IPSTATS_MIB_REASMFAILS, evicted);
+               IP6_ADD_STATS_BH(net, idev, IPSTATS_MIB_REASMFAILS, evicted);
 }
 
 static void ip6_frag_expire(unsigned long 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
@@ -227,9 +229,9 @@ 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);
 }
@@ -243,9 +245,12 @@ fq_find(struct net *net, __be32 id, struct in6_addr *src, struct in6_addr *dst,
        unsigned int hash;
 
        arg.id = id;
+       arg.user = IP6_DEFRAG_LOCAL_DELIVER;
        arg.src = src;
        arg.dst = dst;
-       hash = ip6qhashfn(id, src, dst);
+
+       read_lock(&ip6_frags.lock);
+       hash = inet6_hash_frag(id, src, dst, ip6_frags.rnd);
 
        q = inet_frag_find(&net->ipv6.frags, &ip6_frags, &arg, hash);
        if (q == NULL)
@@ -254,7 +259,7 @@ fq_find(struct net *net, __be32 id, struct in6_addr *src, struct in6_addr *dst,
        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;
 }
 
@@ -264,8 +269,9 @@ static int ip6_frag_queue(struct frag_queue *fq, struct sk_buff *skb,
        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;
@@ -273,7 +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 -
@@ -294,9 +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.
@@ -306,7 +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));
@@ -314,7 +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;
                }
@@ -417,10 +423,11 @@ static int ip6_frag_queue(struct frag_queue *fq, struct sk_buff *skb,
         */
        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_frags.lock);
@@ -429,7 +436,8 @@ static int ip6_frag_queue(struct frag_queue *fq, struct sk_buff *skb,
        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;
 }
@@ -446,6 +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;
@@ -470,8 +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)) -
@@ -487,7 +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;
 
@@ -496,7 +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;
@@ -545,7 +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;
@@ -559,7 +568,7 @@ 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;
 }
@@ -569,24 +578,17 @@ static int ipv6_frag_rcv(struct sk_buff *skb)
        struct frag_hdr *fhdr;
        struct frag_queue *fq;
        struct ipv6hdr *hdr = ipv6_hdr(skb);
-       struct net *net;
+       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);
@@ -594,18 +596,18 @@ static int ipv6_frag_rcv(struct sk_buff *skb)
        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;
        }
 
-       net = skb->dev->nd_net;
        if (atomic_read(&net->ipv6.frags.mem) > net->ipv6.frags.high_thresh)
-               ip6_evictor(net, ip6_dst_idev(skb->dst));
+               ip6_evictor(net, ip6_dst_idev(skb_dst(skb)));
 
        if ((fq = fq_find(net, fhdr->identification, &hdr->saddr, &hdr->daddr,
-                         ip6_dst_idev(skb->dst))) != NULL) {
+                         ip6_dst_idev(skb_dst(skb)))) != NULL) {
                int ret;
 
                spin_lock(&fq->q.lock);
@@ -617,71 +619,73 @@ static int ipv6_frag_rcv(struct sk_buff *skb)
                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,
 };
 
 #ifdef CONFIG_SYSCTL
-static struct ctl_table ip6_frags_ctl_table[] = {
+static struct ctl_table ip6_frags_ns_ctl_table[] = {
        {
-               .ctl_name       = NET_IPV6_IP6FRAG_HIGH_THRESH,
                .procname       = "ip6frag_high_thresh",
                .data           = &init_net.ipv6.frags.high_thresh,
                .maxlen         = sizeof(int),
                .mode           = 0644,
-               .proc_handler   = &proc_dointvec
+               .proc_handler   = proc_dointvec
        },
        {
-               .ctl_name       = NET_IPV6_IP6FRAG_LOW_THRESH,
                .procname       = "ip6frag_low_thresh",
                .data           = &init_net.ipv6.frags.low_thresh,
                .maxlen         = sizeof(int),
                .mode           = 0644,
-               .proc_handler   = &proc_dointvec
+               .proc_handler   = proc_dointvec
        },
        {
-               .ctl_name       = NET_IPV6_IP6FRAG_TIME,
                .procname       = "ip6frag_time",
                .data           = &init_net.ipv6.frags.timeout,
                .maxlen         = sizeof(int),
                .mode           = 0644,
-               .proc_handler   = &proc_dointvec_jiffies,
-               .strategy       = &sysctl_jiffies,
+               .proc_handler   = proc_dointvec_jiffies,
        },
+       { }
+};
+
+static struct ctl_table ip6_frags_ctl_table[] = {
        {
-               .ctl_name       = NET_IPV6_IP6FRAG_SECRET_INTERVAL,
                .procname       = "ip6frag_secret_interval",
                .data           = &ip6_frags.secret_interval,
                .maxlen         = sizeof(int),
                .mode           = 0644,
-               .proc_handler   = &proc_dointvec_jiffies,
-               .strategy       = &sysctl_jiffies
+               .proc_handler   = proc_dointvec_jiffies,
        },
        { }
 };
 
-static int ip6_frags_sysctl_register(struct net *net)
+static int __net_init ip6_frags_ns_sysctl_register(struct net *net)
 {
        struct ctl_table *table;
        struct ctl_table_header *hdr;
 
-       table = ip6_frags_ctl_table;
-       if (net != &init_net) {
-               table = kmemdup(table, sizeof(ip6_frags_ctl_table), GFP_KERNEL);
+       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;
-               table[3].mode &= ~0222;
        }
 
        hdr = register_net_sysctl_table(net, net_ipv6_ctl_path, table);
@@ -692,45 +696,69 @@ static int ip6_frags_sysctl_register(struct net *net)
        return 0;
 
 err_reg:
-       if (net != &init_net)
+       if (!net_eq(net, &init_net))
                kfree(table);
 err_alloc:
        return -ENOMEM;
 }
 
-static void ip6_frags_sysctl_unregister(struct net *net)
+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);
-       kfree(table);
+       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_sysctl_register(struct net *net)
+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(struct net *net)
+static inline void ip6_frags_sysctl_unregister(void)
 {
 }
 #endif
 
-static int ipv6_frags_init_net(struct net *net)
+static int __net_init ipv6_frags_init_net(struct net *net)
 {
-       net->ipv6.frags.high_thresh = 256 * 1024;
-       net->ipv6.frags.low_thresh = 192 * 1024;
+       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_sysctl_register(net);
+       return ip6_frags_ns_sysctl_register(net);
 }
 
-static void ipv6_frags_exit_net(struct net *net)
+static void __net_exit ipv6_frags_exit_net(struct net *net)
 {
-       ip6_frags_sysctl_unregister(net);
+       ip6_frags_ns_sysctl_unregister(net);
        inet_frags_exit_net(&net->ipv6.frags, &ip6_frags);
 }
 
@@ -747,7 +775,13 @@ int __init ipv6_frag_init(void)
        if (ret)
                goto out;
 
-       register_pernet_subsys(&ip6_frags_ops);
+       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;
@@ -760,11 +794,18 @@ int __init ipv6_frag_init(void)
        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;
 }
 
 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);
 }