Add the rt linux 4.1.3-rt3 as base
[kvmfornfv.git] / kernel / net / atm / clip.c
diff --git a/kernel/net/atm/clip.c b/kernel/net/atm/clip.c
new file mode 100644 (file)
index 0000000..17e55df
--- /dev/null
@@ -0,0 +1,939 @@
+/* net/atm/clip.c - RFC1577 Classical IP over ATM */
+
+/* Written 1995-2000 by Werner Almesberger, EPFL LRC/ICA */
+
+#define pr_fmt(fmt) KBUILD_MODNAME ":%s: " fmt, __func__
+
+#include <linux/string.h>
+#include <linux/errno.h>
+#include <linux/kernel.h> /* for UINT_MAX */
+#include <linux/module.h>
+#include <linux/init.h>
+#include <linux/netdevice.h>
+#include <linux/skbuff.h>
+#include <linux/wait.h>
+#include <linux/timer.h>
+#include <linux/if_arp.h> /* for some manifest constants */
+#include <linux/notifier.h>
+#include <linux/atm.h>
+#include <linux/atmdev.h>
+#include <linux/atmclip.h>
+#include <linux/atmarp.h>
+#include <linux/capability.h>
+#include <linux/ip.h> /* for net/route.h */
+#include <linux/in.h> /* for struct sockaddr_in */
+#include <linux/if.h> /* for IFF_UP */
+#include <linux/inetdevice.h>
+#include <linux/bitops.h>
+#include <linux/poison.h>
+#include <linux/proc_fs.h>
+#include <linux/seq_file.h>
+#include <linux/rcupdate.h>
+#include <linux/jhash.h>
+#include <linux/slab.h>
+#include <net/route.h> /* for struct rtable and routing */
+#include <net/icmp.h> /* icmp_send */
+#include <net/arp.h>
+#include <linux/param.h> /* for HZ */
+#include <linux/uaccess.h>
+#include <asm/byteorder.h> /* for htons etc. */
+#include <linux/atomic.h>
+
+#include "common.h"
+#include "resources.h"
+#include <net/atmclip.h>
+
+static struct net_device *clip_devs;
+static struct atm_vcc *atmarpd;
+static struct timer_list idle_timer;
+static const struct neigh_ops clip_neigh_ops;
+
+static int to_atmarpd(enum atmarp_ctrl_type type, int itf, __be32 ip)
+{
+       struct sock *sk;
+       struct atmarp_ctrl *ctrl;
+       struct sk_buff *skb;
+
+       pr_debug("(%d)\n", type);
+       if (!atmarpd)
+               return -EUNATCH;
+       skb = alloc_skb(sizeof(struct atmarp_ctrl), GFP_ATOMIC);
+       if (!skb)
+               return -ENOMEM;
+       ctrl = (struct atmarp_ctrl *)skb_put(skb, sizeof(struct atmarp_ctrl));
+       ctrl->type = type;
+       ctrl->itf_num = itf;
+       ctrl->ip = ip;
+       atm_force_charge(atmarpd, skb->truesize);
+
+       sk = sk_atm(atmarpd);
+       skb_queue_tail(&sk->sk_receive_queue, skb);
+       sk->sk_data_ready(sk);
+       return 0;
+}
+
+static void link_vcc(struct clip_vcc *clip_vcc, struct atmarp_entry *entry)
+{
+       pr_debug("%p to entry %p (neigh %p)\n", clip_vcc, entry, entry->neigh);
+       clip_vcc->entry = entry;
+       clip_vcc->xoff = 0;     /* @@@ may overrun buffer by one packet */
+       clip_vcc->next = entry->vccs;
+       entry->vccs = clip_vcc;
+       entry->neigh->used = jiffies;
+}
+
+static void unlink_clip_vcc(struct clip_vcc *clip_vcc)
+{
+       struct atmarp_entry *entry = clip_vcc->entry;
+       struct clip_vcc **walk;
+
+       if (!entry) {
+               pr_crit("!clip_vcc->entry (clip_vcc %p)\n", clip_vcc);
+               return;
+       }
+       netif_tx_lock_bh(entry->neigh->dev);    /* block clip_start_xmit() */
+       entry->neigh->used = jiffies;
+       for (walk = &entry->vccs; *walk; walk = &(*walk)->next)
+               if (*walk == clip_vcc) {
+                       int error;
+
+                       *walk = clip_vcc->next; /* atomic */
+                       clip_vcc->entry = NULL;
+                       if (clip_vcc->xoff)
+                               netif_wake_queue(entry->neigh->dev);
+                       if (entry->vccs)
+                               goto out;
+                       entry->expires = jiffies - 1;
+                       /* force resolution or expiration */
+                       error = neigh_update(entry->neigh, NULL, NUD_NONE,
+                                            NEIGH_UPDATE_F_ADMIN);
+                       if (error)
+                               pr_crit("neigh_update failed with %d\n", error);
+                       goto out;
+               }
+       pr_crit("ATMARP: failed (entry %p, vcc 0x%p)\n", entry, clip_vcc);
+out:
+       netif_tx_unlock_bh(entry->neigh->dev);
+}
+
+/* The neighbour entry n->lock is held. */
+static int neigh_check_cb(struct neighbour *n)
+{
+       struct atmarp_entry *entry = neighbour_priv(n);
+       struct clip_vcc *cv;
+
+       if (n->ops != &clip_neigh_ops)
+               return 0;
+       for (cv = entry->vccs; cv; cv = cv->next) {
+               unsigned long exp = cv->last_use + cv->idle_timeout;
+
+               if (cv->idle_timeout && time_after(jiffies, exp)) {
+                       pr_debug("releasing vcc %p->%p of entry %p\n",
+                                cv, cv->vcc, entry);
+                       vcc_release_async(cv->vcc, -ETIMEDOUT);
+               }
+       }
+
+       if (entry->vccs || time_before(jiffies, entry->expires))
+               return 0;
+
+       if (atomic_read(&n->refcnt) > 1) {
+               struct sk_buff *skb;
+
+               pr_debug("destruction postponed with ref %d\n",
+                        atomic_read(&n->refcnt));
+
+               while ((skb = skb_dequeue(&n->arp_queue)) != NULL)
+                       dev_kfree_skb(skb);
+
+               return 0;
+       }
+
+       pr_debug("expired neigh %p\n", n);
+       return 1;
+}
+
+static void idle_timer_check(unsigned long dummy)
+{
+       write_lock(&arp_tbl.lock);
+       __neigh_for_each_release(&arp_tbl, neigh_check_cb);
+       mod_timer(&idle_timer, jiffies + CLIP_CHECK_INTERVAL * HZ);
+       write_unlock(&arp_tbl.lock);
+}
+
+static int clip_arp_rcv(struct sk_buff *skb)
+{
+       struct atm_vcc *vcc;
+
+       pr_debug("\n");
+       vcc = ATM_SKB(skb)->vcc;
+       if (!vcc || !atm_charge(vcc, skb->truesize)) {
+               dev_kfree_skb_any(skb);
+               return 0;
+       }
+       pr_debug("pushing to %p\n", vcc);
+       pr_debug("using %p\n", CLIP_VCC(vcc)->old_push);
+       CLIP_VCC(vcc)->old_push(vcc, skb);
+       return 0;
+}
+
+static const unsigned char llc_oui[] = {
+       0xaa,   /* DSAP: non-ISO */
+       0xaa,   /* SSAP: non-ISO */
+       0x03,   /* Ctrl: Unnumbered Information Command PDU */
+       0x00,   /* OUI: EtherType */
+       0x00,
+       0x00
+};
+
+static void clip_push(struct atm_vcc *vcc, struct sk_buff *skb)
+{
+       struct clip_vcc *clip_vcc = CLIP_VCC(vcc);
+
+       pr_debug("\n");
+
+       if (!clip_devs) {
+               atm_return(vcc, skb->truesize);
+               kfree_skb(skb);
+               return;
+       }
+
+       if (!skb) {
+               pr_debug("removing VCC %p\n", clip_vcc);
+               if (clip_vcc->entry)
+                       unlink_clip_vcc(clip_vcc);
+               clip_vcc->old_push(vcc, NULL);  /* pass on the bad news */
+               kfree(clip_vcc);
+               return;
+       }
+       atm_return(vcc, skb->truesize);
+       skb->dev = clip_vcc->entry ? clip_vcc->entry->neigh->dev : clip_devs;
+       /* clip_vcc->entry == NULL if we don't have an IP address yet */
+       if (!skb->dev) {
+               dev_kfree_skb_any(skb);
+               return;
+       }
+       ATM_SKB(skb)->vcc = vcc;
+       skb_reset_mac_header(skb);
+       if (!clip_vcc->encap ||
+           skb->len < RFC1483LLC_LEN ||
+           memcmp(skb->data, llc_oui, sizeof(llc_oui)))
+               skb->protocol = htons(ETH_P_IP);
+       else {
+               skb->protocol = ((__be16 *)skb->data)[3];
+               skb_pull(skb, RFC1483LLC_LEN);
+               if (skb->protocol == htons(ETH_P_ARP)) {
+                       skb->dev->stats.rx_packets++;
+                       skb->dev->stats.rx_bytes += skb->len;
+                       clip_arp_rcv(skb);
+                       return;
+               }
+       }
+       clip_vcc->last_use = jiffies;
+       skb->dev->stats.rx_packets++;
+       skb->dev->stats.rx_bytes += skb->len;
+       memset(ATM_SKB(skb), 0, sizeof(struct atm_skb_data));
+       netif_rx(skb);
+}
+
+/*
+ * Note: these spinlocks _must_not_ block on non-SMP. The only goal is that
+ * clip_pop is atomic with respect to the critical section in clip_start_xmit.
+ */
+
+static void clip_pop(struct atm_vcc *vcc, struct sk_buff *skb)
+{
+       struct clip_vcc *clip_vcc = CLIP_VCC(vcc);
+       struct net_device *dev = skb->dev;
+       int old;
+       unsigned long flags;
+
+       pr_debug("(vcc %p)\n", vcc);
+       clip_vcc->old_pop(vcc, skb);
+       /* skb->dev == NULL in outbound ARP packets */
+       if (!dev)
+               return;
+       spin_lock_irqsave(&PRIV(dev)->xoff_lock, flags);
+       if (atm_may_send(vcc, 0)) {
+               old = xchg(&clip_vcc->xoff, 0);
+               if (old)
+                       netif_wake_queue(dev);
+       }
+       spin_unlock_irqrestore(&PRIV(dev)->xoff_lock, flags);
+}
+
+static void clip_neigh_solicit(struct neighbour *neigh, struct sk_buff *skb)
+{
+       __be32 *ip = (__be32 *) neigh->primary_key;
+
+       pr_debug("(neigh %p, skb %p)\n", neigh, skb);
+       to_atmarpd(act_need, PRIV(neigh->dev)->number, *ip);
+}
+
+static void clip_neigh_error(struct neighbour *neigh, struct sk_buff *skb)
+{
+#ifndef CONFIG_ATM_CLIP_NO_ICMP
+       icmp_send(skb, ICMP_DEST_UNREACH, ICMP_HOST_UNREACH, 0);
+#endif
+       kfree_skb(skb);
+}
+
+static const struct neigh_ops clip_neigh_ops = {
+       .family =               AF_INET,
+       .solicit =              clip_neigh_solicit,
+       .error_report =         clip_neigh_error,
+       .output =               neigh_direct_output,
+       .connected_output =     neigh_direct_output,
+};
+
+static int clip_constructor(struct neighbour *neigh)
+{
+       struct atmarp_entry *entry = neighbour_priv(neigh);
+
+       if (neigh->tbl->family != AF_INET)
+               return -EINVAL;
+
+       if (neigh->type != RTN_UNICAST)
+               return -EINVAL;
+
+       neigh->nud_state = NUD_NONE;
+       neigh->ops = &clip_neigh_ops;
+       neigh->output = neigh->ops->output;
+       entry->neigh = neigh;
+       entry->vccs = NULL;
+       entry->expires = jiffies - 1;
+
+       return 0;
+}
+
+/* @@@ copy bh locking from arp.c -- need to bh-enable atm code before */
+
+/*
+ * We play with the resolve flag: 0 and 1 have the usual meaning, but -1 means
+ * to allocate the neighbour entry but not to ask atmarpd for resolution. Also,
+ * don't increment the usage count. This is used to create entries in
+ * clip_setentry.
+ */
+
+static int clip_encap(struct atm_vcc *vcc, int mode)
+{
+       CLIP_VCC(vcc)->encap = mode;
+       return 0;
+}
+
+static netdev_tx_t clip_start_xmit(struct sk_buff *skb,
+                                  struct net_device *dev)
+{
+       struct clip_priv *clip_priv = PRIV(dev);
+       struct dst_entry *dst = skb_dst(skb);
+       struct atmarp_entry *entry;
+       struct neighbour *n;
+       struct atm_vcc *vcc;
+       struct rtable *rt;
+       __be32 *daddr;
+       int old;
+       unsigned long flags;
+
+       pr_debug("(skb %p)\n", skb);
+       if (!dst) {
+               pr_err("skb_dst(skb) == NULL\n");
+               dev_kfree_skb(skb);
+               dev->stats.tx_dropped++;
+               return NETDEV_TX_OK;
+       }
+       rt = (struct rtable *) dst;
+       if (rt->rt_gateway)
+               daddr = &rt->rt_gateway;
+       else
+               daddr = &ip_hdr(skb)->daddr;
+       n = dst_neigh_lookup(dst, daddr);
+       if (!n) {
+               pr_err("NO NEIGHBOUR !\n");
+               dev_kfree_skb(skb);
+               dev->stats.tx_dropped++;
+               return NETDEV_TX_OK;
+       }
+       entry = neighbour_priv(n);
+       if (!entry->vccs) {
+               if (time_after(jiffies, entry->expires)) {
+                       /* should be resolved */
+                       entry->expires = jiffies + ATMARP_RETRY_DELAY * HZ;
+                       to_atmarpd(act_need, PRIV(dev)->number, *((__be32 *)n->primary_key));
+               }
+               if (entry->neigh->arp_queue.qlen < ATMARP_MAX_UNRES_PACKETS)
+                       skb_queue_tail(&entry->neigh->arp_queue, skb);
+               else {
+                       dev_kfree_skb(skb);
+                       dev->stats.tx_dropped++;
+               }
+               goto out_release_neigh;
+       }
+       pr_debug("neigh %p, vccs %p\n", entry, entry->vccs);
+       ATM_SKB(skb)->vcc = vcc = entry->vccs->vcc;
+       pr_debug("using neighbour %p, vcc %p\n", n, vcc);
+       if (entry->vccs->encap) {
+               void *here;
+
+               here = skb_push(skb, RFC1483LLC_LEN);
+               memcpy(here, llc_oui, sizeof(llc_oui));
+               ((__be16 *) here)[3] = skb->protocol;
+       }
+       atomic_add(skb->truesize, &sk_atm(vcc)->sk_wmem_alloc);
+       ATM_SKB(skb)->atm_options = vcc->atm_options;
+       entry->vccs->last_use = jiffies;
+       pr_debug("atm_skb(%p)->vcc(%p)->dev(%p)\n", skb, vcc, vcc->dev);
+       old = xchg(&entry->vccs->xoff, 1);      /* assume XOFF ... */
+       if (old) {
+               pr_warn("XOFF->XOFF transition\n");
+               goto out_release_neigh;
+       }
+       dev->stats.tx_packets++;
+       dev->stats.tx_bytes += skb->len;
+       vcc->send(vcc, skb);
+       if (atm_may_send(vcc, 0)) {
+               entry->vccs->xoff = 0;
+               goto out_release_neigh;
+       }
+       spin_lock_irqsave(&clip_priv->xoff_lock, flags);
+       netif_stop_queue(dev);  /* XOFF -> throttle immediately */
+       barrier();
+       if (!entry->vccs->xoff)
+               netif_start_queue(dev);
+       /* Oh, we just raced with clip_pop. netif_start_queue should be
+          good enough, because nothing should really be asleep because
+          of the brief netif_stop_queue. If this isn't true or if it
+          changes, use netif_wake_queue instead. */
+       spin_unlock_irqrestore(&clip_priv->xoff_lock, flags);
+out_release_neigh:
+       neigh_release(n);
+       return NETDEV_TX_OK;
+}
+
+static int clip_mkip(struct atm_vcc *vcc, int timeout)
+{
+       struct clip_vcc *clip_vcc;
+
+       if (!vcc->push)
+               return -EBADFD;
+       clip_vcc = kmalloc(sizeof(struct clip_vcc), GFP_KERNEL);
+       if (!clip_vcc)
+               return -ENOMEM;
+       pr_debug("%p vcc %p\n", clip_vcc, vcc);
+       clip_vcc->vcc = vcc;
+       vcc->user_back = clip_vcc;
+       set_bit(ATM_VF_IS_CLIP, &vcc->flags);
+       clip_vcc->entry = NULL;
+       clip_vcc->xoff = 0;
+       clip_vcc->encap = 1;
+       clip_vcc->last_use = jiffies;
+       clip_vcc->idle_timeout = timeout * HZ;
+       clip_vcc->old_push = vcc->push;
+       clip_vcc->old_pop = vcc->pop;
+       vcc->push = clip_push;
+       vcc->pop = clip_pop;
+
+       /* re-process everything received between connection setup and MKIP */
+       vcc_process_recv_queue(vcc);
+
+       return 0;
+}
+
+static int clip_setentry(struct atm_vcc *vcc, __be32 ip)
+{
+       struct neighbour *neigh;
+       struct atmarp_entry *entry;
+       int error;
+       struct clip_vcc *clip_vcc;
+       struct rtable *rt;
+
+       if (vcc->push != clip_push) {
+               pr_warn("non-CLIP VCC\n");
+               return -EBADF;
+       }
+       clip_vcc = CLIP_VCC(vcc);
+       if (!ip) {
+               if (!clip_vcc->entry) {
+                       pr_err("hiding hidden ATMARP entry\n");
+                       return 0;
+               }
+               pr_debug("remove\n");
+               unlink_clip_vcc(clip_vcc);
+               return 0;
+       }
+       rt = ip_route_output(&init_net, ip, 0, 1, 0);
+       if (IS_ERR(rt))
+               return PTR_ERR(rt);
+       neigh = __neigh_lookup(&arp_tbl, &ip, rt->dst.dev, 1);
+       ip_rt_put(rt);
+       if (!neigh)
+               return -ENOMEM;
+       entry = neighbour_priv(neigh);
+       if (entry != clip_vcc->entry) {
+               if (!clip_vcc->entry)
+                       pr_debug("add\n");
+               else {
+                       pr_debug("update\n");
+                       unlink_clip_vcc(clip_vcc);
+               }
+               link_vcc(clip_vcc, entry);
+       }
+       error = neigh_update(neigh, llc_oui, NUD_PERMANENT,
+                            NEIGH_UPDATE_F_OVERRIDE | NEIGH_UPDATE_F_ADMIN);
+       neigh_release(neigh);
+       return error;
+}
+
+static const struct net_device_ops clip_netdev_ops = {
+       .ndo_start_xmit         = clip_start_xmit,
+       .ndo_neigh_construct    = clip_constructor,
+};
+
+static void clip_setup(struct net_device *dev)
+{
+       dev->netdev_ops = &clip_netdev_ops;
+       dev->type = ARPHRD_ATM;
+       dev->neigh_priv_len = sizeof(struct atmarp_entry);
+       dev->hard_header_len = RFC1483LLC_LEN;
+       dev->mtu = RFC1626_MTU;
+       dev->tx_queue_len = 100;        /* "normal" queue (packets) */
+       /* When using a "real" qdisc, the qdisc determines the queue */
+       /* length. tx_queue_len is only used for the default case, */
+       /* without any more elaborate queuing. 100 is a reasonable */
+       /* compromise between decent burst-tolerance and protection */
+       /* against memory hogs. */
+       netif_keep_dst(dev);
+}
+
+static int clip_create(int number)
+{
+       struct net_device *dev;
+       struct clip_priv *clip_priv;
+       int error;
+
+       if (number != -1) {
+               for (dev = clip_devs; dev; dev = PRIV(dev)->next)
+                       if (PRIV(dev)->number == number)
+                               return -EEXIST;
+       } else {
+               number = 0;
+               for (dev = clip_devs; dev; dev = PRIV(dev)->next)
+                       if (PRIV(dev)->number >= number)
+                               number = PRIV(dev)->number + 1;
+       }
+       dev = alloc_netdev(sizeof(struct clip_priv), "", NET_NAME_UNKNOWN,
+                          clip_setup);
+       if (!dev)
+               return -ENOMEM;
+       clip_priv = PRIV(dev);
+       sprintf(dev->name, "atm%d", number);
+       spin_lock_init(&clip_priv->xoff_lock);
+       clip_priv->number = number;
+       error = register_netdev(dev);
+       if (error) {
+               free_netdev(dev);
+               return error;
+       }
+       clip_priv->next = clip_devs;
+       clip_devs = dev;
+       pr_debug("registered (net:%s)\n", dev->name);
+       return number;
+}
+
+static int clip_device_event(struct notifier_block *this, unsigned long event,
+                            void *ptr)
+{
+       struct net_device *dev = netdev_notifier_info_to_dev(ptr);
+
+       if (!net_eq(dev_net(dev), &init_net))
+               return NOTIFY_DONE;
+
+       if (event == NETDEV_UNREGISTER)
+               return NOTIFY_DONE;
+
+       /* ignore non-CLIP devices */
+       if (dev->type != ARPHRD_ATM || dev->netdev_ops != &clip_netdev_ops)
+               return NOTIFY_DONE;
+
+       switch (event) {
+       case NETDEV_UP:
+               pr_debug("NETDEV_UP\n");
+               to_atmarpd(act_up, PRIV(dev)->number, 0);
+               break;
+       case NETDEV_GOING_DOWN:
+               pr_debug("NETDEV_DOWN\n");
+               to_atmarpd(act_down, PRIV(dev)->number, 0);
+               break;
+       case NETDEV_CHANGE:
+       case NETDEV_CHANGEMTU:
+               pr_debug("NETDEV_CHANGE*\n");
+               to_atmarpd(act_change, PRIV(dev)->number, 0);
+               break;
+       }
+       return NOTIFY_DONE;
+}
+
+static int clip_inet_event(struct notifier_block *this, unsigned long event,
+                          void *ifa)
+{
+       struct in_device *in_dev;
+       struct netdev_notifier_info info;
+
+       in_dev = ((struct in_ifaddr *)ifa)->ifa_dev;
+       /*
+        * Transitions are of the down-change-up type, so it's sufficient to
+        * handle the change on up.
+        */
+       if (event != NETDEV_UP)
+               return NOTIFY_DONE;
+       netdev_notifier_info_init(&info, in_dev->dev);
+       return clip_device_event(this, NETDEV_CHANGE, &info);
+}
+
+static struct notifier_block clip_dev_notifier = {
+       .notifier_call = clip_device_event,
+};
+
+
+
+static struct notifier_block clip_inet_notifier = {
+       .notifier_call = clip_inet_event,
+};
+
+
+
+static void atmarpd_close(struct atm_vcc *vcc)
+{
+       pr_debug("\n");
+
+       rtnl_lock();
+       atmarpd = NULL;
+       skb_queue_purge(&sk_atm(vcc)->sk_receive_queue);
+       rtnl_unlock();
+
+       pr_debug("(done)\n");
+       module_put(THIS_MODULE);
+}
+
+static struct atmdev_ops atmarpd_dev_ops = {
+       .close = atmarpd_close
+};
+
+
+static struct atm_dev atmarpd_dev = {
+       .ops =                  &atmarpd_dev_ops,
+       .type =                 "arpd",
+       .number =               999,
+       .lock =                 __SPIN_LOCK_UNLOCKED(atmarpd_dev.lock)
+};
+
+
+static int atm_init_atmarp(struct atm_vcc *vcc)
+{
+       rtnl_lock();
+       if (atmarpd) {
+               rtnl_unlock();
+               return -EADDRINUSE;
+       }
+
+       mod_timer(&idle_timer, jiffies + CLIP_CHECK_INTERVAL * HZ);
+
+       atmarpd = vcc;
+       set_bit(ATM_VF_META, &vcc->flags);
+       set_bit(ATM_VF_READY, &vcc->flags);
+           /* allow replies and avoid getting closed if signaling dies */
+       vcc->dev = &atmarpd_dev;
+       vcc_insert_socket(sk_atm(vcc));
+       vcc->push = NULL;
+       vcc->pop = NULL; /* crash */
+       vcc->push_oam = NULL; /* crash */
+       rtnl_unlock();
+       return 0;
+}
+
+static int clip_ioctl(struct socket *sock, unsigned int cmd, unsigned long arg)
+{
+       struct atm_vcc *vcc = ATM_SD(sock);
+       int err = 0;
+
+       switch (cmd) {
+       case SIOCMKCLIP:
+       case ATMARPD_CTRL:
+       case ATMARP_MKIP:
+       case ATMARP_SETENTRY:
+       case ATMARP_ENCAP:
+               if (!capable(CAP_NET_ADMIN))
+                       return -EPERM;
+               break;
+       default:
+               return -ENOIOCTLCMD;
+       }
+
+       switch (cmd) {
+       case SIOCMKCLIP:
+               err = clip_create(arg);
+               break;
+       case ATMARPD_CTRL:
+               err = atm_init_atmarp(vcc);
+               if (!err) {
+                       sock->state = SS_CONNECTED;
+                       __module_get(THIS_MODULE);
+               }
+               break;
+       case ATMARP_MKIP:
+               err = clip_mkip(vcc, arg);
+               break;
+       case ATMARP_SETENTRY:
+               err = clip_setentry(vcc, (__force __be32)arg);
+               break;
+       case ATMARP_ENCAP:
+               err = clip_encap(vcc, arg);
+               break;
+       }
+       return err;
+}
+
+static struct atm_ioctl clip_ioctl_ops = {
+       .owner = THIS_MODULE,
+       .ioctl = clip_ioctl,
+};
+
+#ifdef CONFIG_PROC_FS
+
+static void svc_addr(struct seq_file *seq, struct sockaddr_atmsvc *addr)
+{
+       static int code[] = { 1, 2, 10, 6, 1, 0 };
+       static int e164[] = { 1, 8, 4, 6, 1, 0 };
+
+       if (*addr->sas_addr.pub) {
+               seq_printf(seq, "%s", addr->sas_addr.pub);
+               if (*addr->sas_addr.prv)
+                       seq_putc(seq, '+');
+       } else if (!*addr->sas_addr.prv) {
+               seq_printf(seq, "%s", "(none)");
+               return;
+       }
+       if (*addr->sas_addr.prv) {
+               unsigned char *prv = addr->sas_addr.prv;
+               int *fields;
+               int i, j;
+
+               fields = *prv == ATM_AFI_E164 ? e164 : code;
+               for (i = 0; fields[i]; i++) {
+                       for (j = fields[i]; j; j--)
+                               seq_printf(seq, "%02X", *prv++);
+                       if (fields[i + 1])
+                               seq_putc(seq, '.');
+               }
+       }
+}
+
+/* This means the neighbour entry has no attached VCC objects. */
+#define SEQ_NO_VCC_TOKEN       ((void *) 2)
+
+static void atmarp_info(struct seq_file *seq, struct neighbour *n,
+                       struct atmarp_entry *entry, struct clip_vcc *clip_vcc)
+{
+       struct net_device *dev = n->dev;
+       unsigned long exp;
+       char buf[17];
+       int svc, llc, off;
+
+       svc = ((clip_vcc == SEQ_NO_VCC_TOKEN) ||
+              (sk_atm(clip_vcc->vcc)->sk_family == AF_ATMSVC));
+
+       llc = ((clip_vcc == SEQ_NO_VCC_TOKEN) || clip_vcc->encap);
+
+       if (clip_vcc == SEQ_NO_VCC_TOKEN)
+               exp = entry->neigh->used;
+       else
+               exp = clip_vcc->last_use;
+
+       exp = (jiffies - exp) / HZ;
+
+       seq_printf(seq, "%-6s%-4s%-4s%5ld ",
+                  dev->name, svc ? "SVC" : "PVC", llc ? "LLC" : "NULL", exp);
+
+       off = scnprintf(buf, sizeof(buf) - 1, "%pI4", n->primary_key);
+       while (off < 16)
+               buf[off++] = ' ';
+       buf[off] = '\0';
+       seq_printf(seq, "%s", buf);
+
+       if (clip_vcc == SEQ_NO_VCC_TOKEN) {
+               if (time_before(jiffies, entry->expires))
+                       seq_printf(seq, "(resolving)\n");
+               else
+                       seq_printf(seq, "(expired, ref %d)\n",
+                                  atomic_read(&entry->neigh->refcnt));
+       } else if (!svc) {
+               seq_printf(seq, "%d.%d.%d\n",
+                          clip_vcc->vcc->dev->number,
+                          clip_vcc->vcc->vpi, clip_vcc->vcc->vci);
+       } else {
+               svc_addr(seq, &clip_vcc->vcc->remote);
+               seq_putc(seq, '\n');
+       }
+}
+
+struct clip_seq_state {
+       /* This member must be first. */
+       struct neigh_seq_state ns;
+
+       /* Local to clip specific iteration. */
+       struct clip_vcc *vcc;
+};
+
+static struct clip_vcc *clip_seq_next_vcc(struct atmarp_entry *e,
+                                         struct clip_vcc *curr)
+{
+       if (!curr) {
+               curr = e->vccs;
+               if (!curr)
+                       return SEQ_NO_VCC_TOKEN;
+               return curr;
+       }
+       if (curr == SEQ_NO_VCC_TOKEN)
+               return NULL;
+
+       curr = curr->next;
+
+       return curr;
+}
+
+static void *clip_seq_vcc_walk(struct clip_seq_state *state,
+                              struct atmarp_entry *e, loff_t * pos)
+{
+       struct clip_vcc *vcc = state->vcc;
+
+       vcc = clip_seq_next_vcc(e, vcc);
+       if (vcc && pos != NULL) {
+               while (*pos) {
+                       vcc = clip_seq_next_vcc(e, vcc);
+                       if (!vcc)
+                               break;
+                       --(*pos);
+               }
+       }
+       state->vcc = vcc;
+
+       return vcc;
+}
+
+static void *clip_seq_sub_iter(struct neigh_seq_state *_state,
+                              struct neighbour *n, loff_t * pos)
+{
+       struct clip_seq_state *state = (struct clip_seq_state *)_state;
+
+       if (n->dev->type != ARPHRD_ATM)
+               return NULL;
+
+       return clip_seq_vcc_walk(state, neighbour_priv(n), pos);
+}
+
+static void *clip_seq_start(struct seq_file *seq, loff_t * pos)
+{
+       struct clip_seq_state *state = seq->private;
+       state->ns.neigh_sub_iter = clip_seq_sub_iter;
+       return neigh_seq_start(seq, pos, &arp_tbl, NEIGH_SEQ_NEIGH_ONLY);
+}
+
+static int clip_seq_show(struct seq_file *seq, void *v)
+{
+       static char atm_arp_banner[] =
+           "IPitf TypeEncp Idle IP address      ATM address\n";
+
+       if (v == SEQ_START_TOKEN) {
+               seq_puts(seq, atm_arp_banner);
+       } else {
+               struct clip_seq_state *state = seq->private;
+               struct clip_vcc *vcc = state->vcc;
+               struct neighbour *n = v;
+
+               atmarp_info(seq, n, neighbour_priv(n), vcc);
+       }
+       return 0;
+}
+
+static const struct seq_operations arp_seq_ops = {
+       .start  = clip_seq_start,
+       .next   = neigh_seq_next,
+       .stop   = neigh_seq_stop,
+       .show   = clip_seq_show,
+};
+
+static int arp_seq_open(struct inode *inode, struct file *file)
+{
+       return seq_open_net(inode, file, &arp_seq_ops,
+                           sizeof(struct clip_seq_state));
+}
+
+static const struct file_operations arp_seq_fops = {
+       .open           = arp_seq_open,
+       .read           = seq_read,
+       .llseek         = seq_lseek,
+       .release        = seq_release_net,
+       .owner          = THIS_MODULE
+};
+#endif
+
+static void atm_clip_exit_noproc(void);
+
+static int __init atm_clip_init(void)
+{
+       register_atm_ioctl(&clip_ioctl_ops);
+       register_netdevice_notifier(&clip_dev_notifier);
+       register_inetaddr_notifier(&clip_inet_notifier);
+
+       setup_timer(&idle_timer, idle_timer_check, 0);
+
+#ifdef CONFIG_PROC_FS
+       {
+               struct proc_dir_entry *p;
+
+               p = proc_create("arp", S_IRUGO, atm_proc_root, &arp_seq_fops);
+               if (!p) {
+                       pr_err("Unable to initialize /proc/net/atm/arp\n");
+                       atm_clip_exit_noproc();
+                       return -ENOMEM;
+               }
+       }
+#endif
+
+       return 0;
+}
+
+static void atm_clip_exit_noproc(void)
+{
+       struct net_device *dev, *next;
+
+       unregister_inetaddr_notifier(&clip_inet_notifier);
+       unregister_netdevice_notifier(&clip_dev_notifier);
+
+       deregister_atm_ioctl(&clip_ioctl_ops);
+
+       /* First, stop the idle timer, so it stops banging
+        * on the table.
+        */
+       del_timer_sync(&idle_timer);
+
+       dev = clip_devs;
+       while (dev) {
+               next = PRIV(dev)->next;
+               unregister_netdev(dev);
+               free_netdev(dev);
+               dev = next;
+       }
+}
+
+static void __exit atm_clip_exit(void)
+{
+       remove_proc_entry("arp", atm_proc_root);
+
+       atm_clip_exit_noproc();
+}
+
+module_init(atm_clip_init);
+module_exit(atm_clip_exit);
+MODULE_AUTHOR("Werner Almesberger");
+MODULE_DESCRIPTION("Classical/IP over ATM interface");
+MODULE_LICENSE("GPL");