[GRE]: Add net/gre_net argument to some functions.
[safe/jmp/linux-2.6] / net / ipv4 / ip_fragment.c
index 4f01334..cd6ce6a 100644 (file)
@@ -74,25 +74,6 @@ struct ipq {
        struct inet_peer *peer;
 };
 
-static struct inet_frags_ctl ip4_frags_ctl __read_mostly = {
-       /*
-        * Fragment cache limits. We will commit 256K at one time. Should we
-        * cross that limit we will prune down to 192K. This should cope with
-        * even the most extreme cases without allowing an attacker to
-        * measurably harm machine performance.
-        */
-       .high_thresh     = 256 * 1024,
-       .low_thresh      = 192 * 1024,
-
-       /*
-        * Important NOTE! Fragment queue must be destroyed before MSL expires.
-        * RFC791 is wrong proposing to prolongate timer each fragment arrival
-        * by TTL.
-        */
-       .timeout         = IP_FRAG_TIME,
-       .secret_interval = 10 * 60 * HZ,
-};
-
 static struct inet_frags ip4_frags;
 
 int ip_frag_nqueues(struct net *net)
@@ -213,7 +194,7 @@ static void ip_expire(unsigned long arg)
 
        spin_lock(&qp->q.lock);
 
-       if (qp->q.last_in & COMPLETE)
+       if (qp->q.last_in & INET_FRAG_COMPLETE)
                goto out;
 
        ipq_kill(qp);
@@ -221,10 +202,13 @@ static void ip_expire(unsigned long arg)
        IP_INC_STATS_BH(IPSTATS_MIB_REASMTIMEOUT);
        IP_INC_STATS_BH(IPSTATS_MIB_REASMFAILS);
 
-       if ((qp->q.last_in&FIRST_IN) && qp->q.fragments != NULL) {
+       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(&init_net, qp->iif)) != NULL) {
+               if ((head->dev = dev_get_by_index(net, qp->iif)) != NULL) {
                        icmp_send(head, ICMP_TIME_EXCEEDED, ICMP_EXC_FRAGTIME, 0);
                        dev_put(head->dev);
                }
@@ -287,7 +271,7 @@ static int ip_frag_reinit(struct ipq *qp)
 {
        struct sk_buff *fp;
 
-       if (!mod_timer(&qp->q.timer, jiffies + ip4_frags_ctl.timeout)) {
+       if (!mod_timer(&qp->q.timer, jiffies + qp->q.net->timeout)) {
                atomic_inc(&qp->q.refcnt);
                return -ETIMEDOUT;
        }
@@ -317,7 +301,7 @@ static int ip_frag_queue(struct ipq *qp, struct sk_buff *skb)
        int ihl, end;
        int err = -ENOENT;
 
-       if (qp->q.last_in & COMPLETE)
+       if (qp->q.last_in & INET_FRAG_COMPLETE)
                goto err;
 
        if (!(IPCB(skb)->flags & IPSKB_FRAG_COMPLETE) &&
@@ -343,9 +327,9 @@ static int ip_frag_queue(struct ipq *qp, struct sk_buff *skb)
                 * or have different end, the segment is corrrupted.
                 */
                if (end < qp->q.len ||
-                   ((qp->q.last_in & LAST_IN) && end != qp->q.len))
+                   ((qp->q.last_in & INET_FRAG_LAST_IN) && end != qp->q.len))
                        goto err;
-               qp->q.last_in |= LAST_IN;
+               qp->q.last_in |= INET_FRAG_LAST_IN;
                qp->q.len = end;
        } else {
                if (end&7) {
@@ -355,7 +339,7 @@ static int ip_frag_queue(struct ipq *qp, struct sk_buff *skb)
                }
                if (end > qp->q.len) {
                        /* Some bits beyond end -> corruption. */
-                       if (qp->q.last_in & LAST_IN)
+                       if (qp->q.last_in & INET_FRAG_LAST_IN)
                                goto err;
                        qp->q.len = end;
                }
@@ -454,13 +438,14 @@ static int ip_frag_queue(struct ipq *qp, struct sk_buff *skb)
        qp->q.meat += skb->len;
        atomic_add(skb->truesize, &qp->q.net->mem);
        if (offset == 0)
-               qp->q.last_in |= FIRST_IN;
+               qp->q.last_in |= INET_FRAG_FIRST_IN;
 
-       if (qp->q.last_in == (FIRST_IN | LAST_IN) && qp->q.meat == qp->q.len)
+       if (qp->q.last_in == (INET_FRAG_FIRST_IN | INET_FRAG_LAST_IN) &&
+           qp->q.meat == qp->q.len)
                return ip_frag_reasm(qp, prev, dev);
 
        write_lock(&ip4_frags.lock);
-       list_move_tail(&qp->q.lru_list, &ip4_frags.lru_list);
+       list_move_tail(&qp->q.lru_list, &qp->q.net->lru_list);
        write_unlock(&ip4_frags.lock);
        return -EINPROGRESS;
 
@@ -572,7 +557,7 @@ out_nomem:
 out_oversize:
        if (net_ratelimit())
                printk(KERN_INFO
-                       "Oversized IP packet from %d.%d.%d.%d.\n",
+                       "Oversized IP packet from " NIPQUAD_FMT ".\n",
                        NIPQUAD(qp->saddr));
 out_fail:
        IP_INC_STATS_BH(IPSTATS_MIB_REASMFAILS);
@@ -587,9 +572,9 @@ int ip_defrag(struct sk_buff *skb, u32 user)
 
        IP_INC_STATS_BH(IPSTATS_MIB_REASMREQDS);
 
-       net = skb->dev->nd_net;
+       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) > ip4_frags_ctl.high_thresh)
+       if (atomic_read(&net->ipv4.frags.mem) > net->ipv4.frags.high_thresh)
                ip_evictor(net);
 
        /* Lookup (or create) queue header */
@@ -617,7 +602,7 @@ static struct ctl_table ip4_frags_ctl_table[] = {
        {
                .ctl_name       = NET_IPV4_IPFRAG_HIGH_THRESH,
                .procname       = "ipfrag_high_thresh",
-               .data           = &ip4_frags_ctl.high_thresh,
+               .data           = &init_net.ipv4.frags.high_thresh,
                .maxlen         = sizeof(int),
                .mode           = 0644,
                .proc_handler   = &proc_dointvec
@@ -625,7 +610,7 @@ static struct ctl_table ip4_frags_ctl_table[] = {
        {
                .ctl_name       = NET_IPV4_IPFRAG_LOW_THRESH,
                .procname       = "ipfrag_low_thresh",
-               .data           = &ip4_frags_ctl.low_thresh,
+               .data           = &init_net.ipv4.frags.low_thresh,
                .maxlen         = sizeof(int),
                .mode           = 0644,
                .proc_handler   = &proc_dointvec
@@ -633,7 +618,7 @@ static struct ctl_table ip4_frags_ctl_table[] = {
        {
                .ctl_name       = NET_IPV4_IPFRAG_TIME,
                .procname       = "ipfrag_time",
-               .data           = &ip4_frags_ctl.timeout,
+               .data           = &init_net.ipv4.frags.timeout,
                .maxlen         = sizeof(int),
                .mode           = 0644,
                .proc_handler   = &proc_dointvec_jiffies,
@@ -642,7 +627,7 @@ static struct ctl_table ip4_frags_ctl_table[] = {
        {
                .ctl_name       = NET_IPV4_IPFRAG_SECRET_INTERVAL,
                .procname       = "ipfrag_secret_interval",
-               .data           = &ip4_frags_ctl.secret_interval,
+               .data           = &ip4_frags.secret_interval,
                .maxlen         = sizeof(int),
                .mode           = 0644,
                .proc_handler   = &proc_dointvec_jiffies,
@@ -661,30 +646,91 @@ static struct ctl_table ip4_frags_ctl_table[] = {
 
 static int ip4_frags_ctl_register(struct net *net)
 {
+       struct ctl_table *table;
        struct ctl_table_header *hdr;
 
-       hdr = register_net_sysctl_table(net, net_ipv4_ctl_path,
-                       ip4_frags_ctl_table);
-       return hdr == NULL ? -ENOMEM : 0;
+       table = ip4_frags_ctl_table;
+       if (net != &init_net) {
+               table = kmemdup(table, sizeof(ip4_frags_ctl_table), GFP_KERNEL);
+               if (table == NULL)
+                       goto err_alloc;
+
+               table[0].data = &net->ipv4.frags.high_thresh;
+               table[1].data = &net->ipv4.frags.low_thresh;
+               table[2].data = &net->ipv4.frags.timeout;
+               table[3].mode &= ~0222;
+               table[4].mode &= ~0222;
+       }
+
+       hdr = register_net_sysctl_table(net, net_ipv4_ctl_path, table);
+       if (hdr == NULL)
+               goto err_reg;
+
+       net->ipv4.frags_hdr = hdr;
+       return 0;
+
+err_reg:
+       if (net != &init_net)
+               kfree(table);
+err_alloc:
+       return -ENOMEM;
+}
+
+static void ip4_frags_ctl_unregister(struct net *net)
+{
+       struct ctl_table *table;
+
+       table = net->ipv4.frags_hdr->ctl_table_arg;
+       unregister_net_sysctl_table(net->ipv4.frags_hdr);
+       kfree(table);
 }
 #else
 static inline int ip4_frags_ctl_register(struct net *net)
 {
        return 0;
 }
+
+static inline void ip4_frags_ctl_unregister(struct net *net)
+{
+}
 #endif
 
 static int ipv4_frags_init_net(struct net *net)
 {
+       /*
+        * Fragment cache limits. We will commit 256K at one time. Should we
+        * cross that limit we will prune down to 192K. This should cope with
+        * even the most extreme cases without allowing an attacker to
+        * measurably harm machine performance.
+        */
+       net->ipv4.frags.high_thresh = 256 * 1024;
+       net->ipv4.frags.low_thresh = 192 * 1024;
+       /*
+        * Important NOTE! Fragment queue must be destroyed before MSL expires.
+        * RFC791 is wrong proposing to prolongate timer each fragment arrival
+        * by TTL.
+        */
+       net->ipv4.frags.timeout = IP_FRAG_TIME;
+
        inet_frags_init_net(&net->ipv4.frags);
 
        return ip4_frags_ctl_register(net);
 }
 
+static void ipv4_frags_exit_net(struct net *net)
+{
+       ip4_frags_ctl_unregister(net);
+       inet_frags_exit_net(&net->ipv4.frags, &ip4_frags);
+}
+
+static struct pernet_operations ip4_frags_ops = {
+       .init = ipv4_frags_init_net,
+       .exit = ipv4_frags_exit_net,
+};
+
 void __init ipfrag_init(void)
 {
-       ipv4_frags_init_net(&init_net);
-       ip4_frags.ctl = &ip4_frags_ctl;
+       register_pernet_subsys(&ip4_frags_ops);
        ip4_frags.hashfn = ip4_hashfn;
        ip4_frags.constructor = ip4_frag_init;
        ip4_frags.destructor = ip4_frag_free;
@@ -692,6 +738,7 @@ void __init ipfrag_init(void)
        ip4_frags.qsize = sizeof(struct ipq);
        ip4_frags.match = ip4_frag_match;
        ip4_frags.frag_expire = ip_expire;
+       ip4_frags.secret_interval = 10 * 60 * HZ;
        inet_frags_init(&ip4_frags);
 }