tcp: fix ICMP-RTO war
[safe/jmp/linux-2.6] / net / ipv4 / ip_fragment.c
index 91e3214..86964b3 100644 (file)
@@ -6,7 +6,7 @@
  *             The IP fragmentation functionality.
  *
  * Authors:    Fred N. van Kempen <waltje@uWalt.NL.Mugnet.ORG>
- *             Alan Cox <Alan.Cox@linux.org>
+ *             Alan Cox <alan@lxorguk.ukuu.org.uk>
  *
  * Fixes:
  *             Alan Cox        :       Split from ip.c , see ip_input.c for history.
@@ -56,7 +56,7 @@ struct ipfrag_skb_cb
        int                     offset;
 };
 
-#define FRAG_CB(skb)   ((struct ipfrag_skb_cb*)((skb)->cb))
+#define FRAG_CB(skb)   ((struct ipfrag_skb_cb *)((skb)->cb))
 
 /* Describe an entry in the "incomplete datagrams" queue. */
 struct ipq {
@@ -178,7 +178,7 @@ static void ip_evictor(struct net *net)
 
        evicted = inet_frag_evictor(&net->ipv4.frags, &ip4_frags);
        if (evicted)
-               IP_ADD_STATS_BH(IPSTATS_MIB_REASMFAILS, evicted);
+               IP_ADD_STATS_BH(net, IPSTATS_MIB_REASMFAILS, evicted);
 }
 
 /*
@@ -187,8 +187,10 @@ static void ip_evictor(struct net *net)
 static void ip_expire(unsigned long arg)
 {
        struct ipq *qp;
+       struct net *net;
 
        qp = container_of((struct inet_frag_queue *) arg, struct ipq, q);
+       net = container_of(qp->q.net, struct net, ipv4.frags);
 
        spin_lock(&qp->q.lock);
 
@@ -197,19 +199,18 @@ static void ip_expire(unsigned long arg)
 
        ipq_kill(qp);
 
-       IP_INC_STATS_BH(IPSTATS_MIB_REASMTIMEOUT);
-       IP_INC_STATS_BH(IPSTATS_MIB_REASMFAILS);
+       IP_INC_STATS_BH(net, IPSTATS_MIB_REASMTIMEOUT);
+       IP_INC_STATS_BH(net, IPSTATS_MIB_REASMFAILS);
 
        if ((qp->q.last_in & INET_FRAG_FIRST_IN) && qp->q.fragments != NULL) {
                struct sk_buff *head = qp->q.fragments;
-               struct net *net;
 
-               net = container_of(qp->q.net, struct net, ipv4.frags);
                /* Send an ICMP "Fragment Reassembly Timeout" message. */
-               if ((head->dev = dev_get_by_index(net, qp->iif)) != NULL) {
+               rcu_read_lock();
+               head->dev = dev_get_by_index_rcu(net, qp->iif);
+               if (head->dev)
                        icmp_send(head, ICMP_TIME_EXCEEDED, ICMP_EXC_FRAGTIME, 0);
-                       dev_put(head->dev);
-               }
+               rcu_read_unlock();
        }
 out:
        spin_unlock(&qp->q.lock);
@@ -227,6 +228,8 @@ static inline struct ipq *ip_find(struct net *net, struct iphdr *iph, u32 user)
 
        arg.iph = iph;
        arg.user = user;
+
+       read_lock(&ip4_frags.lock);
        hash = ipqhashfn(iph->id, iph->saddr, iph->daddr, iph->protocol);
 
        q = inet_frag_find(&net->ipv4.frags, &ip4_frags, &arg, hash);
@@ -259,7 +262,10 @@ static inline int ip_frag_too_far(struct ipq *qp)
        rc = qp->q.fragments && (end - start) > max;
 
        if (rc) {
-               IP_INC_STATS_BH(IPSTATS_MIB_REASMFAILS);
+               struct net *net;
+
+               net = container_of(qp->q.net, struct net, ipv4.frags);
+               IP_INC_STATS_BH(net, IPSTATS_MIB_REASMFAILS);
        }
 
        return rc;
@@ -458,6 +464,7 @@ err:
 static int ip_frag_reasm(struct ipq *qp, struct sk_buff *prev,
                         struct net_device *dev)
 {
+       struct net *net = container_of(qp->q.net, struct net, ipv4.frags);
        struct iphdr *iph;
        struct sk_buff *fp, *head = qp->q.fragments;
        int len;
@@ -483,8 +490,8 @@ static int ip_frag_reasm(struct ipq *qp, struct sk_buff *prev,
                qp->q.fragments = head;
        }
 
-       BUG_TRAP(head != NULL);
-       BUG_TRAP(FRAG_CB(head)->offset == 0);
+       WARN_ON(head == NULL);
+       WARN_ON(FRAG_CB(head)->offset != 0);
 
        /* Allocate a new buffer for the datagram. */
        ihlen = ip_hdrlen(head);
@@ -501,7 +508,7 @@ static int ip_frag_reasm(struct ipq *qp, 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;
 
@@ -510,7 +517,7 @@ static int ip_frag_reasm(struct ipq *qp, 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;
@@ -543,7 +550,7 @@ static int ip_frag_reasm(struct ipq *qp, struct sk_buff *prev,
        iph = ip_hdr(head);
        iph->frag_off = 0;
        iph->tot_len = htons(len);
-       IP_INC_STATS_BH(IPSTATS_MIB_REASMOKS);
+       IP_INC_STATS_BH(net, IPSTATS_MIB_REASMOKS);
        qp->q.fragments = NULL;
        return 0;
 
@@ -554,11 +561,10 @@ out_nomem:
        goto out_fail;
 out_oversize:
        if (net_ratelimit())
-               printk(KERN_INFO
-                       "Oversized IP packet from " NIPQUAD_FMT ".\n",
-                       NIPQUAD(qp->saddr));
+               printk(KERN_INFO "Oversized IP packet from %pI4.\n",
+                       &qp->saddr);
 out_fail:
-       IP_INC_STATS_BH(IPSTATS_MIB_REASMFAILS);
+       IP_INC_STATS_BH(net, IPSTATS_MIB_REASMFAILS);
        return err;
 }
 
@@ -568,9 +574,9 @@ int ip_defrag(struct sk_buff *skb, u32 user)
        struct ipq *qp;
        struct net *net;
 
-       IP_INC_STATS_BH(IPSTATS_MIB_REASMREQDS);
+       net = skb->dev ? dev_net(skb->dev) : dev_net(skb_dst(skb)->dev);
+       IP_INC_STATS_BH(net, IPSTATS_MIB_REASMREQDS);
 
-       net = skb->dev ? dev_net(skb->dev) : dev_net(skb->dst->dev);
        /* Start by cleaning up the memory. */
        if (atomic_read(&net->ipv4.frags.mem) > net->ipv4.frags.high_thresh)
                ip_evictor(net);
@@ -588,7 +594,7 @@ int ip_defrag(struct sk_buff *skb, u32 user)
                return ret;
        }
 
-       IP_INC_STATS_BH(IPSTATS_MIB_REASMFAILS);
+       IP_INC_STATS_BH(net, IPSTATS_MIB_REASMFAILS);
        kfree_skb(skb);
        return -ENOMEM;
 }
@@ -598,49 +604,43 @@ static int zero;
 
 static struct ctl_table ip4_frags_ns_ctl_table[] = {
        {
-               .ctl_name       = NET_IPV4_IPFRAG_HIGH_THRESH,
                .procname       = "ipfrag_high_thresh",
                .data           = &init_net.ipv4.frags.high_thresh,
                .maxlen         = sizeof(int),
                .mode           = 0644,
-               .proc_handler   = &proc_dointvec
+               .proc_handler   = proc_dointvec
        },
        {
-               .ctl_name       = NET_IPV4_IPFRAG_LOW_THRESH,
                .procname       = "ipfrag_low_thresh",
                .data           = &init_net.ipv4.frags.low_thresh,
                .maxlen         = sizeof(int),
                .mode           = 0644,
-               .proc_handler   = &proc_dointvec
+               .proc_handler   = proc_dointvec
        },
        {
-               .ctl_name       = NET_IPV4_IPFRAG_TIME,
                .procname       = "ipfrag_time",
                .data           = &init_net.ipv4.frags.timeout,
                .maxlen         = sizeof(int),
                .mode           = 0644,
-               .proc_handler   = &proc_dointvec_jiffies,
-               .strategy       = &sysctl_jiffies
+               .proc_handler   = proc_dointvec_jiffies,
        },
        { }
 };
 
 static struct ctl_table ip4_frags_ctl_table[] = {
        {
-               .ctl_name       = NET_IPV4_IPFRAG_SECRET_INTERVAL,
                .procname       = "ipfrag_secret_interval",
                .data           = &ip4_frags.secret_interval,
                .maxlen         = sizeof(int),
                .mode           = 0644,
-               .proc_handler   = &proc_dointvec_jiffies,
-               .strategy       = &sysctl_jiffies
+               .proc_handler   = proc_dointvec_jiffies,
        },
        {
                .procname       = "ipfrag_max_dist",
                .data           = &sysctl_ipfrag_max_dist,
                .maxlen         = sizeof(int),
                .mode           = 0644,
-               .proc_handler   = &proc_dointvec_minmax,
+               .proc_handler   = proc_dointvec_minmax,
                .extra1         = &zero
        },
        { }
@@ -652,7 +652,7 @@ static int ip4_frags_ns_ctl_register(struct net *net)
        struct ctl_table_header *hdr;
 
        table = ip4_frags_ns_ctl_table;
-       if (net != &init_net) {
+       if (!net_eq(net, &init_net)) {
                table = kmemdup(table, sizeof(ip4_frags_ns_ctl_table), GFP_KERNEL);
                if (table == NULL)
                        goto err_alloc;
@@ -670,7 +670,7 @@ static int ip4_frags_ns_ctl_register(struct net *net)
        return 0;
 
 err_reg:
-       if (net != &init_net)
+       if (!net_eq(net, &init_net))
                kfree(table);
 err_alloc:
        return -ENOMEM;