Add the rt linux 4.1.3-rt3 as base
[kvmfornfv.git] / kernel / net / ipv6 / netfilter / ip6_tables.c
diff --git a/kernel/net/ipv6/netfilter/ip6_tables.c b/kernel/net/ipv6/netfilter/ip6_tables.c
new file mode 100644 (file)
index 0000000..62f5b0d
--- /dev/null
@@ -0,0 +1,2296 @@
+/*
+ * Packet matching code.
+ *
+ * Copyright (C) 1999 Paul `Rusty' Russell & Michael J. Neuling
+ * Copyright (C) 2000-2005 Netfilter Core Team <coreteam@netfilter.org>
+ * Copyright (c) 2006-2010 Patrick McHardy <kaber@trash.net>
+ *
+ * This program is free software; you can redistribute it and/or modify
+ * it under the terms of the GNU General Public License version 2 as
+ * published by the Free Software Foundation.
+ */
+
+#define pr_fmt(fmt) KBUILD_MODNAME ": " fmt
+
+#include <linux/kernel.h>
+#include <linux/capability.h>
+#include <linux/in.h>
+#include <linux/skbuff.h>
+#include <linux/kmod.h>
+#include <linux/vmalloc.h>
+#include <linux/netdevice.h>
+#include <linux/module.h>
+#include <linux/poison.h>
+#include <linux/icmpv6.h>
+#include <net/ipv6.h>
+#include <net/compat.h>
+#include <asm/uaccess.h>
+#include <linux/mutex.h>
+#include <linux/proc_fs.h>
+#include <linux/err.h>
+#include <linux/cpumask.h>
+
+#include <linux/netfilter_ipv6/ip6_tables.h>
+#include <linux/netfilter/x_tables.h>
+#include <net/netfilter/nf_log.h>
+#include "../../netfilter/xt_repldata.h"
+
+MODULE_LICENSE("GPL");
+MODULE_AUTHOR("Netfilter Core Team <coreteam@netfilter.org>");
+MODULE_DESCRIPTION("IPv6 packet filter");
+
+/*#define DEBUG_IP_FIREWALL*/
+/*#define DEBUG_ALLOW_ALL*/ /* Useful for remote debugging */
+/*#define DEBUG_IP_FIREWALL_USER*/
+
+#ifdef DEBUG_IP_FIREWALL
+#define dprintf(format, args...) pr_info(format , ## args)
+#else
+#define dprintf(format, args...)
+#endif
+
+#ifdef DEBUG_IP_FIREWALL_USER
+#define duprintf(format, args...) pr_info(format , ## args)
+#else
+#define duprintf(format, args...)
+#endif
+
+#ifdef CONFIG_NETFILTER_DEBUG
+#define IP_NF_ASSERT(x)        WARN_ON(!(x))
+#else
+#define IP_NF_ASSERT(x)
+#endif
+
+#if 0
+/* All the better to debug you with... */
+#define static
+#define inline
+#endif
+
+void *ip6t_alloc_initial_table(const struct xt_table *info)
+{
+       return xt_alloc_initial_table(ip6t, IP6T);
+}
+EXPORT_SYMBOL_GPL(ip6t_alloc_initial_table);
+
+/*
+   We keep a set of rules for each CPU, so we can avoid write-locking
+   them in the softirq when updating the counters and therefore
+   only need to read-lock in the softirq; doing a write_lock_bh() in user
+   context stops packets coming through and allows user context to read
+   the counters or update the rules.
+
+   Hence the start of any table is given by get_table() below.  */
+
+/* Returns whether matches rule or not. */
+/* Performance critical - called for every packet */
+static inline bool
+ip6_packet_match(const struct sk_buff *skb,
+                const char *indev,
+                const char *outdev,
+                const struct ip6t_ip6 *ip6info,
+                unsigned int *protoff,
+                int *fragoff, bool *hotdrop)
+{
+       unsigned long ret;
+       const struct ipv6hdr *ipv6 = ipv6_hdr(skb);
+
+#define FWINV(bool, invflg) ((bool) ^ !!(ip6info->invflags & (invflg)))
+
+       if (FWINV(ipv6_masked_addr_cmp(&ipv6->saddr, &ip6info->smsk,
+                                      &ip6info->src), IP6T_INV_SRCIP) ||
+           FWINV(ipv6_masked_addr_cmp(&ipv6->daddr, &ip6info->dmsk,
+                                      &ip6info->dst), IP6T_INV_DSTIP)) {
+               dprintf("Source or dest mismatch.\n");
+/*
+               dprintf("SRC: %u. Mask: %u. Target: %u.%s\n", ip->saddr,
+                       ipinfo->smsk.s_addr, ipinfo->src.s_addr,
+                       ipinfo->invflags & IP6T_INV_SRCIP ? " (INV)" : "");
+               dprintf("DST: %u. Mask: %u. Target: %u.%s\n", ip->daddr,
+                       ipinfo->dmsk.s_addr, ipinfo->dst.s_addr,
+                       ipinfo->invflags & IP6T_INV_DSTIP ? " (INV)" : "");*/
+               return false;
+       }
+
+       ret = ifname_compare_aligned(indev, ip6info->iniface, ip6info->iniface_mask);
+
+       if (FWINV(ret != 0, IP6T_INV_VIA_IN)) {
+               dprintf("VIA in mismatch (%s vs %s).%s\n",
+                       indev, ip6info->iniface,
+                       ip6info->invflags&IP6T_INV_VIA_IN ?" (INV)":"");
+               return false;
+       }
+
+       ret = ifname_compare_aligned(outdev, ip6info->outiface, ip6info->outiface_mask);
+
+       if (FWINV(ret != 0, IP6T_INV_VIA_OUT)) {
+               dprintf("VIA out mismatch (%s vs %s).%s\n",
+                       outdev, ip6info->outiface,
+                       ip6info->invflags&IP6T_INV_VIA_OUT ?" (INV)":"");
+               return false;
+       }
+
+/* ... might want to do something with class and flowlabel here ... */
+
+       /* look for the desired protocol header */
+       if((ip6info->flags & IP6T_F_PROTO)) {
+               int protohdr;
+               unsigned short _frag_off;
+
+               protohdr = ipv6_find_hdr(skb, protoff, -1, &_frag_off, NULL);
+               if (protohdr < 0) {
+                       if (_frag_off == 0)
+                               *hotdrop = true;
+                       return false;
+               }
+               *fragoff = _frag_off;
+
+               dprintf("Packet protocol %hi ?= %s%hi.\n",
+                               protohdr,
+                               ip6info->invflags & IP6T_INV_PROTO ? "!":"",
+                               ip6info->proto);
+
+               if (ip6info->proto == protohdr) {
+                       if(ip6info->invflags & IP6T_INV_PROTO) {
+                               return false;
+                       }
+                       return true;
+               }
+
+               /* We need match for the '-p all', too! */
+               if ((ip6info->proto != 0) &&
+                       !(ip6info->invflags & IP6T_INV_PROTO))
+                       return false;
+       }
+       return true;
+}
+
+/* should be ip6 safe */
+static bool
+ip6_checkentry(const struct ip6t_ip6 *ipv6)
+{
+       if (ipv6->flags & ~IP6T_F_MASK) {
+               duprintf("Unknown flag bits set: %08X\n",
+                        ipv6->flags & ~IP6T_F_MASK);
+               return false;
+       }
+       if (ipv6->invflags & ~IP6T_INV_MASK) {
+               duprintf("Unknown invflag bits set: %08X\n",
+                        ipv6->invflags & ~IP6T_INV_MASK);
+               return false;
+       }
+       return true;
+}
+
+static unsigned int
+ip6t_error(struct sk_buff *skb, const struct xt_action_param *par)
+{
+       net_info_ratelimited("error: `%s'\n", (const char *)par->targinfo);
+
+       return NF_DROP;
+}
+
+static inline struct ip6t_entry *
+get_entry(const void *base, unsigned int offset)
+{
+       return (struct ip6t_entry *)(base + offset);
+}
+
+/* All zeroes == unconditional rule. */
+/* Mildly perf critical (only if packet tracing is on) */
+static inline bool unconditional(const struct ip6t_ip6 *ipv6)
+{
+       static const struct ip6t_ip6 uncond;
+
+       return memcmp(ipv6, &uncond, sizeof(uncond)) == 0;
+}
+
+static inline const struct xt_entry_target *
+ip6t_get_target_c(const struct ip6t_entry *e)
+{
+       return ip6t_get_target((struct ip6t_entry *)e);
+}
+
+#if IS_ENABLED(CONFIG_NETFILTER_XT_TARGET_TRACE)
+/* This cries for unification! */
+static const char *const hooknames[] = {
+       [NF_INET_PRE_ROUTING]           = "PREROUTING",
+       [NF_INET_LOCAL_IN]              = "INPUT",
+       [NF_INET_FORWARD]               = "FORWARD",
+       [NF_INET_LOCAL_OUT]             = "OUTPUT",
+       [NF_INET_POST_ROUTING]          = "POSTROUTING",
+};
+
+enum nf_ip_trace_comments {
+       NF_IP6_TRACE_COMMENT_RULE,
+       NF_IP6_TRACE_COMMENT_RETURN,
+       NF_IP6_TRACE_COMMENT_POLICY,
+};
+
+static const char *const comments[] = {
+       [NF_IP6_TRACE_COMMENT_RULE]     = "rule",
+       [NF_IP6_TRACE_COMMENT_RETURN]   = "return",
+       [NF_IP6_TRACE_COMMENT_POLICY]   = "policy",
+};
+
+static struct nf_loginfo trace_loginfo = {
+       .type = NF_LOG_TYPE_LOG,
+       .u = {
+               .log = {
+                       .level = LOGLEVEL_WARNING,
+                       .logflags = NF_LOG_MASK,
+               },
+       },
+};
+
+/* Mildly perf critical (only if packet tracing is on) */
+static inline int
+get_chainname_rulenum(const struct ip6t_entry *s, const struct ip6t_entry *e,
+                     const char *hookname, const char **chainname,
+                     const char **comment, unsigned int *rulenum)
+{
+       const struct xt_standard_target *t = (void *)ip6t_get_target_c(s);
+
+       if (strcmp(t->target.u.kernel.target->name, XT_ERROR_TARGET) == 0) {
+               /* Head of user chain: ERROR target with chainname */
+               *chainname = t->target.data;
+               (*rulenum) = 0;
+       } else if (s == e) {
+               (*rulenum)++;
+
+               if (s->target_offset == sizeof(struct ip6t_entry) &&
+                   strcmp(t->target.u.kernel.target->name,
+                          XT_STANDARD_TARGET) == 0 &&
+                   t->verdict < 0 &&
+                   unconditional(&s->ipv6)) {
+                       /* Tail of chains: STANDARD target (return/policy) */
+                       *comment = *chainname == hookname
+                               ? comments[NF_IP6_TRACE_COMMENT_POLICY]
+                               : comments[NF_IP6_TRACE_COMMENT_RETURN];
+               }
+               return 1;
+       } else
+               (*rulenum)++;
+
+       return 0;
+}
+
+static void trace_packet(const struct sk_buff *skb,
+                        unsigned int hook,
+                        const struct net_device *in,
+                        const struct net_device *out,
+                        const char *tablename,
+                        const struct xt_table_info *private,
+                        const struct ip6t_entry *e)
+{
+       const void *table_base;
+       const struct ip6t_entry *root;
+       const char *hookname, *chainname, *comment;
+       const struct ip6t_entry *iter;
+       unsigned int rulenum = 0;
+       struct net *net = dev_net(in ? in : out);
+
+       table_base = private->entries[smp_processor_id()];
+       root = get_entry(table_base, private->hook_entry[hook]);
+
+       hookname = chainname = hooknames[hook];
+       comment = comments[NF_IP6_TRACE_COMMENT_RULE];
+
+       xt_entry_foreach(iter, root, private->size - private->hook_entry[hook])
+               if (get_chainname_rulenum(iter, e, hookname,
+                   &chainname, &comment, &rulenum) != 0)
+                       break;
+
+       nf_log_trace(net, AF_INET6, hook, skb, in, out, &trace_loginfo,
+                    "TRACE: %s:%s:%s:%u ",
+                    tablename, chainname, comment, rulenum);
+}
+#endif
+
+static inline __pure struct ip6t_entry *
+ip6t_next_entry(const struct ip6t_entry *entry)
+{
+       return (void *)entry + entry->next_offset;
+}
+
+/* Returns one of the generic firewall policies, like NF_ACCEPT. */
+unsigned int
+ip6t_do_table(struct sk_buff *skb,
+             unsigned int hook,
+             const struct nf_hook_state *state,
+             struct xt_table *table)
+{
+       static const char nulldevname[IFNAMSIZ] __attribute__((aligned(sizeof(long))));
+       /* Initializing verdict to NF_DROP keeps gcc happy. */
+       unsigned int verdict = NF_DROP;
+       const char *indev, *outdev;
+       const void *table_base;
+       struct ip6t_entry *e, **jumpstack;
+       unsigned int *stackptr, origptr, cpu;
+       const struct xt_table_info *private;
+       struct xt_action_param acpar;
+       unsigned int addend;
+
+       /* Initialization */
+       indev = state->in ? state->in->name : nulldevname;
+       outdev = state->out ? state->out->name : nulldevname;
+       /* We handle fragments by dealing with the first fragment as
+        * if it was a normal packet.  All other fragments are treated
+        * normally, except that they will NEVER match rules that ask
+        * things we don't know, ie. tcp syn flag or ports).  If the
+        * rule is also a fragment-specific rule, non-fragments won't
+        * match it. */
+       acpar.hotdrop = false;
+       acpar.in      = state->in;
+       acpar.out     = state->out;
+       acpar.family  = NFPROTO_IPV6;
+       acpar.hooknum = hook;
+
+       IP_NF_ASSERT(table->valid_hooks & (1 << hook));
+
+       local_bh_disable();
+       addend = xt_write_recseq_begin();
+       private = table->private;
+       /*
+        * Ensure we load private-> members after we've fetched the base
+        * pointer.
+        */
+       smp_read_barrier_depends();
+       cpu        = smp_processor_id();
+       table_base = private->entries[cpu];
+       jumpstack  = (struct ip6t_entry **)private->jumpstack[cpu];
+       stackptr   = per_cpu_ptr(private->stackptr, cpu);
+       origptr    = *stackptr;
+
+       e = get_entry(table_base, private->hook_entry[hook]);
+
+       do {
+               const struct xt_entry_target *t;
+               const struct xt_entry_match *ematch;
+
+               IP_NF_ASSERT(e);
+               acpar.thoff = 0;
+               if (!ip6_packet_match(skb, indev, outdev, &e->ipv6,
+                   &acpar.thoff, &acpar.fragoff, &acpar.hotdrop)) {
+ no_match:
+                       e = ip6t_next_entry(e);
+                       continue;
+               }
+
+               xt_ematch_foreach(ematch, e) {
+                       acpar.match     = ematch->u.kernel.match;
+                       acpar.matchinfo = ematch->data;
+                       if (!acpar.match->match(skb, &acpar))
+                               goto no_match;
+               }
+
+               ADD_COUNTER(e->counters, skb->len, 1);
+
+               t = ip6t_get_target_c(e);
+               IP_NF_ASSERT(t->u.kernel.target);
+
+#if IS_ENABLED(CONFIG_NETFILTER_XT_TARGET_TRACE)
+               /* The packet is traced: log it */
+               if (unlikely(skb->nf_trace))
+                       trace_packet(skb, hook, state->in, state->out,
+                                    table->name, private, e);
+#endif
+               /* Standard target? */
+               if (!t->u.kernel.target->target) {
+                       int v;
+
+                       v = ((struct xt_standard_target *)t)->verdict;
+                       if (v < 0) {
+                               /* Pop from stack? */
+                               if (v != XT_RETURN) {
+                                       verdict = (unsigned int)(-v) - 1;
+                                       break;
+                               }
+                               if (*stackptr <= origptr)
+                                       e = get_entry(table_base,
+                                           private->underflow[hook]);
+                               else
+                                       e = ip6t_next_entry(jumpstack[--*stackptr]);
+                               continue;
+                       }
+                       if (table_base + v != ip6t_next_entry(e) &&
+                           !(e->ipv6.flags & IP6T_F_GOTO)) {
+                               if (*stackptr >= private->stacksize) {
+                                       verdict = NF_DROP;
+                                       break;
+                               }
+                               jumpstack[(*stackptr)++] = e;
+                       }
+
+                       e = get_entry(table_base, v);
+                       continue;
+               }
+
+               acpar.target   = t->u.kernel.target;
+               acpar.targinfo = t->data;
+
+               verdict = t->u.kernel.target->target(skb, &acpar);
+               if (verdict == XT_CONTINUE)
+                       e = ip6t_next_entry(e);
+               else
+                       /* Verdict */
+                       break;
+       } while (!acpar.hotdrop);
+
+       *stackptr = origptr;
+
+       xt_write_recseq_end(addend);
+       local_bh_enable();
+
+#ifdef DEBUG_ALLOW_ALL
+       return NF_ACCEPT;
+#else
+       if (acpar.hotdrop)
+               return NF_DROP;
+       else return verdict;
+#endif
+}
+
+/* Figures out from what hook each rule can be called: returns 0 if
+   there are loops.  Puts hook bitmask in comefrom. */
+static int
+mark_source_chains(const struct xt_table_info *newinfo,
+                  unsigned int valid_hooks, void *entry0)
+{
+       unsigned int hook;
+
+       /* No recursion; use packet counter to save back ptrs (reset
+          to 0 as we leave), and comefrom to save source hook bitmask */
+       for (hook = 0; hook < NF_INET_NUMHOOKS; hook++) {
+               unsigned int pos = newinfo->hook_entry[hook];
+               struct ip6t_entry *e = (struct ip6t_entry *)(entry0 + pos);
+
+               if (!(valid_hooks & (1 << hook)))
+                       continue;
+
+               /* Set initial back pointer. */
+               e->counters.pcnt = pos;
+
+               for (;;) {
+                       const struct xt_standard_target *t
+                               = (void *)ip6t_get_target_c(e);
+                       int visited = e->comefrom & (1 << hook);
+
+                       if (e->comefrom & (1 << NF_INET_NUMHOOKS)) {
+                               pr_err("iptables: loop hook %u pos %u %08X.\n",
+                                      hook, pos, e->comefrom);
+                               return 0;
+                       }
+                       e->comefrom |= ((1 << hook) | (1 << NF_INET_NUMHOOKS));
+
+                       /* Unconditional return/END. */
+                       if ((e->target_offset == sizeof(struct ip6t_entry) &&
+                            (strcmp(t->target.u.user.name,
+                                    XT_STANDARD_TARGET) == 0) &&
+                            t->verdict < 0 &&
+                            unconditional(&e->ipv6)) || visited) {
+                               unsigned int oldpos, size;
+
+                               if ((strcmp(t->target.u.user.name,
+                                           XT_STANDARD_TARGET) == 0) &&
+                                   t->verdict < -NF_MAX_VERDICT - 1) {
+                                       duprintf("mark_source_chains: bad "
+                                               "negative verdict (%i)\n",
+                                                               t->verdict);
+                                       return 0;
+                               }
+
+                               /* Return: backtrack through the last
+                                  big jump. */
+                               do {
+                                       e->comefrom ^= (1<<NF_INET_NUMHOOKS);
+#ifdef DEBUG_IP_FIREWALL_USER
+                                       if (e->comefrom
+                                           & (1 << NF_INET_NUMHOOKS)) {
+                                               duprintf("Back unset "
+                                                        "on hook %u "
+                                                        "rule %u\n",
+                                                        hook, pos);
+                                       }
+#endif
+                                       oldpos = pos;
+                                       pos = e->counters.pcnt;
+                                       e->counters.pcnt = 0;
+
+                                       /* We're at the start. */
+                                       if (pos == oldpos)
+                                               goto next;
+
+                                       e = (struct ip6t_entry *)
+                                               (entry0 + pos);
+                               } while (oldpos == pos + e->next_offset);
+
+                               /* Move along one */
+                               size = e->next_offset;
+                               e = (struct ip6t_entry *)
+                                       (entry0 + pos + size);
+                               e->counters.pcnt = pos;
+                               pos += size;
+                       } else {
+                               int newpos = t->verdict;
+
+                               if (strcmp(t->target.u.user.name,
+                                          XT_STANDARD_TARGET) == 0 &&
+                                   newpos >= 0) {
+                                       if (newpos > newinfo->size -
+                                               sizeof(struct ip6t_entry)) {
+                                               duprintf("mark_source_chains: "
+                                                       "bad verdict (%i)\n",
+                                                               newpos);
+                                               return 0;
+                                       }
+                                       /* This a jump; chase it. */
+                                       duprintf("Jump rule %u -> %u\n",
+                                                pos, newpos);
+                               } else {
+                                       /* ... this is a fallthru */
+                                       newpos = pos + e->next_offset;
+                               }
+                               e = (struct ip6t_entry *)
+                                       (entry0 + newpos);
+                               e->counters.pcnt = pos;
+                               pos = newpos;
+                       }
+               }
+               next:
+               duprintf("Finished chain %u\n", hook);
+       }
+       return 1;
+}
+
+static void cleanup_match(struct xt_entry_match *m, struct net *net)
+{
+       struct xt_mtdtor_param par;
+
+       par.net       = net;
+       par.match     = m->u.kernel.match;
+       par.matchinfo = m->data;
+       par.family    = NFPROTO_IPV6;
+       if (par.match->destroy != NULL)
+               par.match->destroy(&par);
+       module_put(par.match->me);
+}
+
+static int
+check_entry(const struct ip6t_entry *e, const char *name)
+{
+       const struct xt_entry_target *t;
+
+       if (!ip6_checkentry(&e->ipv6)) {
+               duprintf("ip_tables: ip check failed %p %s.\n", e, name);
+               return -EINVAL;
+       }
+
+       if (e->target_offset + sizeof(struct xt_entry_target) >
+           e->next_offset)
+               return -EINVAL;
+
+       t = ip6t_get_target_c(e);
+       if (e->target_offset + t->u.target_size > e->next_offset)
+               return -EINVAL;
+
+       return 0;
+}
+
+static int check_match(struct xt_entry_match *m, struct xt_mtchk_param *par)
+{
+       const struct ip6t_ip6 *ipv6 = par->entryinfo;
+       int ret;
+
+       par->match     = m->u.kernel.match;
+       par->matchinfo = m->data;
+
+       ret = xt_check_match(par, m->u.match_size - sizeof(*m),
+                            ipv6->proto, ipv6->invflags & IP6T_INV_PROTO);
+       if (ret < 0) {
+               duprintf("ip_tables: check failed for `%s'.\n",
+                        par.match->name);
+               return ret;
+       }
+       return 0;
+}
+
+static int
+find_check_match(struct xt_entry_match *m, struct xt_mtchk_param *par)
+{
+       struct xt_match *match;
+       int ret;
+
+       match = xt_request_find_match(NFPROTO_IPV6, m->u.user.name,
+                                     m->u.user.revision);
+       if (IS_ERR(match)) {
+               duprintf("find_check_match: `%s' not found\n", m->u.user.name);
+               return PTR_ERR(match);
+       }
+       m->u.kernel.match = match;
+
+       ret = check_match(m, par);
+       if (ret)
+               goto err;
+
+       return 0;
+err:
+       module_put(m->u.kernel.match->me);
+       return ret;
+}
+
+static int check_target(struct ip6t_entry *e, struct net *net, const char *name)
+{
+       struct xt_entry_target *t = ip6t_get_target(e);
+       struct xt_tgchk_param par = {
+               .net       = net,
+               .table     = name,
+               .entryinfo = e,
+               .target    = t->u.kernel.target,
+               .targinfo  = t->data,
+               .hook_mask = e->comefrom,
+               .family    = NFPROTO_IPV6,
+       };
+       int ret;
+
+       t = ip6t_get_target(e);
+       ret = xt_check_target(&par, t->u.target_size - sizeof(*t),
+             e->ipv6.proto, e->ipv6.invflags & IP6T_INV_PROTO);
+       if (ret < 0) {
+               duprintf("ip_tables: check failed for `%s'.\n",
+                        t->u.kernel.target->name);
+               return ret;
+       }
+       return 0;
+}
+
+static int
+find_check_entry(struct ip6t_entry *e, struct net *net, const char *name,
+                unsigned int size)
+{
+       struct xt_entry_target *t;
+       struct xt_target *target;
+       int ret;
+       unsigned int j;
+       struct xt_mtchk_param mtpar;
+       struct xt_entry_match *ematch;
+
+       ret = check_entry(e, name);
+       if (ret)
+               return ret;
+
+       j = 0;
+       mtpar.net       = net;
+       mtpar.table     = name;
+       mtpar.entryinfo = &e->ipv6;
+       mtpar.hook_mask = e->comefrom;
+       mtpar.family    = NFPROTO_IPV6;
+       xt_ematch_foreach(ematch, e) {
+               ret = find_check_match(ematch, &mtpar);
+               if (ret != 0)
+                       goto cleanup_matches;
+               ++j;
+       }
+
+       t = ip6t_get_target(e);
+       target = xt_request_find_target(NFPROTO_IPV6, t->u.user.name,
+                                       t->u.user.revision);
+       if (IS_ERR(target)) {
+               duprintf("find_check_entry: `%s' not found\n", t->u.user.name);
+               ret = PTR_ERR(target);
+               goto cleanup_matches;
+       }
+       t->u.kernel.target = target;
+
+       ret = check_target(e, net, name);
+       if (ret)
+               goto err;
+       return 0;
+ err:
+       module_put(t->u.kernel.target->me);
+ cleanup_matches:
+       xt_ematch_foreach(ematch, e) {
+               if (j-- == 0)
+                       break;
+               cleanup_match(ematch, net);
+       }
+       return ret;
+}
+
+static bool check_underflow(const struct ip6t_entry *e)
+{
+       const struct xt_entry_target *t;
+       unsigned int verdict;
+
+       if (!unconditional(&e->ipv6))
+               return false;
+       t = ip6t_get_target_c(e);
+       if (strcmp(t->u.user.name, XT_STANDARD_TARGET) != 0)
+               return false;
+       verdict = ((struct xt_standard_target *)t)->verdict;
+       verdict = -verdict - 1;
+       return verdict == NF_DROP || verdict == NF_ACCEPT;
+}
+
+static int
+check_entry_size_and_hooks(struct ip6t_entry *e,
+                          struct xt_table_info *newinfo,
+                          const unsigned char *base,
+                          const unsigned char *limit,
+                          const unsigned int *hook_entries,
+                          const unsigned int *underflows,
+                          unsigned int valid_hooks)
+{
+       unsigned int h;
+
+       if ((unsigned long)e % __alignof__(struct ip6t_entry) != 0 ||
+           (unsigned char *)e + sizeof(struct ip6t_entry) >= limit) {
+               duprintf("Bad offset %p\n", e);
+               return -EINVAL;
+       }
+
+       if (e->next_offset
+           < sizeof(struct ip6t_entry) + sizeof(struct xt_entry_target)) {
+               duprintf("checking: element %p size %u\n",
+                        e, e->next_offset);
+               return -EINVAL;
+       }
+
+       /* Check hooks & underflows */
+       for (h = 0; h < NF_INET_NUMHOOKS; h++) {
+               if (!(valid_hooks & (1 << h)))
+                       continue;
+               if ((unsigned char *)e - base == hook_entries[h])
+                       newinfo->hook_entry[h] = hook_entries[h];
+               if ((unsigned char *)e - base == underflows[h]) {
+                       if (!check_underflow(e)) {
+                               pr_err("Underflows must be unconditional and "
+                                      "use the STANDARD target with "
+                                      "ACCEPT/DROP\n");
+                               return -EINVAL;
+                       }
+                       newinfo->underflow[h] = underflows[h];
+               }
+       }
+
+       /* Clear counters and comefrom */
+       e->counters = ((struct xt_counters) { 0, 0 });
+       e->comefrom = 0;
+       return 0;
+}
+
+static void cleanup_entry(struct ip6t_entry *e, struct net *net)
+{
+       struct xt_tgdtor_param par;
+       struct xt_entry_target *t;
+       struct xt_entry_match *ematch;
+
+       /* Cleanup all matches */
+       xt_ematch_foreach(ematch, e)
+               cleanup_match(ematch, net);
+       t = ip6t_get_target(e);
+
+       par.net      = net;
+       par.target   = t->u.kernel.target;
+       par.targinfo = t->data;
+       par.family   = NFPROTO_IPV6;
+       if (par.target->destroy != NULL)
+               par.target->destroy(&par);
+       module_put(par.target->me);
+}
+
+/* Checks and translates the user-supplied table segment (held in
+   newinfo) */
+static int
+translate_table(struct net *net, struct xt_table_info *newinfo, void *entry0,
+                const struct ip6t_replace *repl)
+{
+       struct ip6t_entry *iter;
+       unsigned int i;
+       int ret = 0;
+
+       newinfo->size = repl->size;
+       newinfo->number = repl->num_entries;
+
+       /* Init all hooks to impossible value. */
+       for (i = 0; i < NF_INET_NUMHOOKS; i++) {
+               newinfo->hook_entry[i] = 0xFFFFFFFF;
+               newinfo->underflow[i] = 0xFFFFFFFF;
+       }
+
+       duprintf("translate_table: size %u\n", newinfo->size);
+       i = 0;
+       /* Walk through entries, checking offsets. */
+       xt_entry_foreach(iter, entry0, newinfo->size) {
+               ret = check_entry_size_and_hooks(iter, newinfo, entry0,
+                                                entry0 + repl->size,
+                                                repl->hook_entry,
+                                                repl->underflow,
+                                                repl->valid_hooks);
+               if (ret != 0)
+                       return ret;
+               ++i;
+               if (strcmp(ip6t_get_target(iter)->u.user.name,
+                   XT_ERROR_TARGET) == 0)
+                       ++newinfo->stacksize;
+       }
+
+       if (i != repl->num_entries) {
+               duprintf("translate_table: %u not %u entries\n",
+                        i, repl->num_entries);
+               return -EINVAL;
+       }
+
+       /* Check hooks all assigned */
+       for (i = 0; i < NF_INET_NUMHOOKS; i++) {
+               /* Only hooks which are valid */
+               if (!(repl->valid_hooks & (1 << i)))
+                       continue;
+               if (newinfo->hook_entry[i] == 0xFFFFFFFF) {
+                       duprintf("Invalid hook entry %u %u\n",
+                                i, repl->hook_entry[i]);
+                       return -EINVAL;
+               }
+               if (newinfo->underflow[i] == 0xFFFFFFFF) {
+                       duprintf("Invalid underflow %u %u\n",
+                                i, repl->underflow[i]);
+                       return -EINVAL;
+               }
+       }
+
+       if (!mark_source_chains(newinfo, repl->valid_hooks, entry0))
+               return -ELOOP;
+
+       /* Finally, each sanity check must pass */
+       i = 0;
+       xt_entry_foreach(iter, entry0, newinfo->size) {
+               ret = find_check_entry(iter, net, repl->name, repl->size);
+               if (ret != 0)
+                       break;
+               ++i;
+       }
+
+       if (ret != 0) {
+               xt_entry_foreach(iter, entry0, newinfo->size) {
+                       if (i-- == 0)
+                               break;
+                       cleanup_entry(iter, net);
+               }
+               return ret;
+       }
+
+       /* And one copy for every other CPU */
+       for_each_possible_cpu(i) {
+               if (newinfo->entries[i] && newinfo->entries[i] != entry0)
+                       memcpy(newinfo->entries[i], entry0, newinfo->size);
+       }
+
+       return ret;
+}
+
+static void
+get_counters(const struct xt_table_info *t,
+            struct xt_counters counters[])
+{
+       struct ip6t_entry *iter;
+       unsigned int cpu;
+       unsigned int i;
+
+       for_each_possible_cpu(cpu) {
+               seqcount_t *s = &per_cpu(xt_recseq, cpu);
+
+               i = 0;
+               xt_entry_foreach(iter, t->entries[cpu], t->size) {
+                       u64 bcnt, pcnt;
+                       unsigned int start;
+
+                       do {
+                               start = read_seqcount_begin(s);
+                               bcnt = iter->counters.bcnt;
+                               pcnt = iter->counters.pcnt;
+                       } while (read_seqcount_retry(s, start));
+
+                       ADD_COUNTER(counters[i], bcnt, pcnt);
+                       ++i;
+               }
+       }
+}
+
+static struct xt_counters *alloc_counters(const struct xt_table *table)
+{
+       unsigned int countersize;
+       struct xt_counters *counters;
+       const struct xt_table_info *private = table->private;
+
+       /* We need atomic snapshot of counters: rest doesn't change
+          (other than comefrom, which userspace doesn't care
+          about). */
+       countersize = sizeof(struct xt_counters) * private->number;
+       counters = vzalloc(countersize);
+
+       if (counters == NULL)
+               return ERR_PTR(-ENOMEM);
+
+       get_counters(private, counters);
+
+       return counters;
+}
+
+static int
+copy_entries_to_user(unsigned int total_size,
+                    const struct xt_table *table,
+                    void __user *userptr)
+{
+       unsigned int off, num;
+       const struct ip6t_entry *e;
+       struct xt_counters *counters;
+       const struct xt_table_info *private = table->private;
+       int ret = 0;
+       const void *loc_cpu_entry;
+
+       counters = alloc_counters(table);
+       if (IS_ERR(counters))
+               return PTR_ERR(counters);
+
+       /* choose the copy that is on our node/cpu, ...
+        * This choice is lazy (because current thread is
+        * allowed to migrate to another cpu)
+        */
+       loc_cpu_entry = private->entries[raw_smp_processor_id()];
+       if (copy_to_user(userptr, loc_cpu_entry, total_size) != 0) {
+               ret = -EFAULT;
+               goto free_counters;
+       }
+
+       /* FIXME: use iterator macros --RR */
+       /* ... then go back and fix counters and names */
+       for (off = 0, num = 0; off < total_size; off += e->next_offset, num++){
+               unsigned int i;
+               const struct xt_entry_match *m;
+               const struct xt_entry_target *t;
+
+               e = (struct ip6t_entry *)(loc_cpu_entry + off);
+               if (copy_to_user(userptr + off
+                                + offsetof(struct ip6t_entry, counters),
+                                &counters[num],
+                                sizeof(counters[num])) != 0) {
+                       ret = -EFAULT;
+                       goto free_counters;
+               }
+
+               for (i = sizeof(struct ip6t_entry);
+                    i < e->target_offset;
+                    i += m->u.match_size) {
+                       m = (void *)e + i;
+
+                       if (copy_to_user(userptr + off + i
+                                        + offsetof(struct xt_entry_match,
+                                                   u.user.name),
+                                        m->u.kernel.match->name,
+                                        strlen(m->u.kernel.match->name)+1)
+                           != 0) {
+                               ret = -EFAULT;
+                               goto free_counters;
+                       }
+               }
+
+               t = ip6t_get_target_c(e);
+               if (copy_to_user(userptr + off + e->target_offset
+                                + offsetof(struct xt_entry_target,
+                                           u.user.name),
+                                t->u.kernel.target->name,
+                                strlen(t->u.kernel.target->name)+1) != 0) {
+                       ret = -EFAULT;
+                       goto free_counters;
+               }
+       }
+
+ free_counters:
+       vfree(counters);
+       return ret;
+}
+
+#ifdef CONFIG_COMPAT
+static void compat_standard_from_user(void *dst, const void *src)
+{
+       int v = *(compat_int_t *)src;
+
+       if (v > 0)
+               v += xt_compat_calc_jump(AF_INET6, v);
+       memcpy(dst, &v, sizeof(v));
+}
+
+static int compat_standard_to_user(void __user *dst, const void *src)
+{
+       compat_int_t cv = *(int *)src;
+
+       if (cv > 0)
+               cv -= xt_compat_calc_jump(AF_INET6, cv);
+       return copy_to_user(dst, &cv, sizeof(cv)) ? -EFAULT : 0;
+}
+
+static int compat_calc_entry(const struct ip6t_entry *e,
+                            const struct xt_table_info *info,
+                            const void *base, struct xt_table_info *newinfo)
+{
+       const struct xt_entry_match *ematch;
+       const struct xt_entry_target *t;
+       unsigned int entry_offset;
+       int off, i, ret;
+
+       off = sizeof(struct ip6t_entry) - sizeof(struct compat_ip6t_entry);
+       entry_offset = (void *)e - base;
+       xt_ematch_foreach(ematch, e)
+               off += xt_compat_match_offset(ematch->u.kernel.match);
+       t = ip6t_get_target_c(e);
+       off += xt_compat_target_offset(t->u.kernel.target);
+       newinfo->size -= off;
+       ret = xt_compat_add_offset(AF_INET6, entry_offset, off);
+       if (ret)
+               return ret;
+
+       for (i = 0; i < NF_INET_NUMHOOKS; i++) {
+               if (info->hook_entry[i] &&
+                   (e < (struct ip6t_entry *)(base + info->hook_entry[i])))
+                       newinfo->hook_entry[i] -= off;
+               if (info->underflow[i] &&
+                   (e < (struct ip6t_entry *)(base + info->underflow[i])))
+                       newinfo->underflow[i] -= off;
+       }
+       return 0;
+}
+
+static int compat_table_info(const struct xt_table_info *info,
+                            struct xt_table_info *newinfo)
+{
+       struct ip6t_entry *iter;
+       void *loc_cpu_entry;
+       int ret;
+
+       if (!newinfo || !info)
+               return -EINVAL;
+
+       /* we dont care about newinfo->entries[] */
+       memcpy(newinfo, info, offsetof(struct xt_table_info, entries));
+       newinfo->initial_entries = 0;
+       loc_cpu_entry = info->entries[raw_smp_processor_id()];
+       xt_compat_init_offsets(AF_INET6, info->number);
+       xt_entry_foreach(iter, loc_cpu_entry, info->size) {
+               ret = compat_calc_entry(iter, info, loc_cpu_entry, newinfo);
+               if (ret != 0)
+                       return ret;
+       }
+       return 0;
+}
+#endif
+
+static int get_info(struct net *net, void __user *user,
+                    const int *len, int compat)
+{
+       char name[XT_TABLE_MAXNAMELEN];
+       struct xt_table *t;
+       int ret;
+
+       if (*len != sizeof(struct ip6t_getinfo)) {
+               duprintf("length %u != %zu\n", *len,
+                        sizeof(struct ip6t_getinfo));
+               return -EINVAL;
+       }
+
+       if (copy_from_user(name, user, sizeof(name)) != 0)
+               return -EFAULT;
+
+       name[XT_TABLE_MAXNAMELEN-1] = '\0';
+#ifdef CONFIG_COMPAT
+       if (compat)
+               xt_compat_lock(AF_INET6);
+#endif
+       t = try_then_request_module(xt_find_table_lock(net, AF_INET6, name),
+                                   "ip6table_%s", name);
+       if (!IS_ERR_OR_NULL(t)) {
+               struct ip6t_getinfo info;
+               const struct xt_table_info *private = t->private;
+#ifdef CONFIG_COMPAT
+               struct xt_table_info tmp;
+
+               if (compat) {
+                       ret = compat_table_info(private, &tmp);
+                       xt_compat_flush_offsets(AF_INET6);
+                       private = &tmp;
+               }
+#endif
+               memset(&info, 0, sizeof(info));
+               info.valid_hooks = t->valid_hooks;
+               memcpy(info.hook_entry, private->hook_entry,
+                      sizeof(info.hook_entry));
+               memcpy(info.underflow, private->underflow,
+                      sizeof(info.underflow));
+               info.num_entries = private->number;
+               info.size = private->size;
+               strcpy(info.name, name);
+
+               if (copy_to_user(user, &info, *len) != 0)
+                       ret = -EFAULT;
+               else
+                       ret = 0;
+
+               xt_table_unlock(t);
+               module_put(t->me);
+       } else
+               ret = t ? PTR_ERR(t) : -ENOENT;
+#ifdef CONFIG_COMPAT
+       if (compat)
+               xt_compat_unlock(AF_INET6);
+#endif
+       return ret;
+}
+
+static int
+get_entries(struct net *net, struct ip6t_get_entries __user *uptr,
+            const int *len)
+{
+       int ret;
+       struct ip6t_get_entries get;
+       struct xt_table *t;
+
+       if (*len < sizeof(get)) {
+               duprintf("get_entries: %u < %zu\n", *len, sizeof(get));
+               return -EINVAL;
+       }
+       if (copy_from_user(&get, uptr, sizeof(get)) != 0)
+               return -EFAULT;
+       if (*len != sizeof(struct ip6t_get_entries) + get.size) {
+               duprintf("get_entries: %u != %zu\n",
+                        *len, sizeof(get) + get.size);
+               return -EINVAL;
+       }
+
+       t = xt_find_table_lock(net, AF_INET6, get.name);
+       if (!IS_ERR_OR_NULL(t)) {
+               struct xt_table_info *private = t->private;
+               duprintf("t->private->number = %u\n", private->number);
+               if (get.size == private->size)
+                       ret = copy_entries_to_user(private->size,
+                                                  t, uptr->entrytable);
+               else {
+                       duprintf("get_entries: I've got %u not %u!\n",
+                                private->size, get.size);
+                       ret = -EAGAIN;
+               }
+               module_put(t->me);
+               xt_table_unlock(t);
+       } else
+               ret = t ? PTR_ERR(t) : -ENOENT;
+
+       return ret;
+}
+
+static int
+__do_replace(struct net *net, const char *name, unsigned int valid_hooks,
+            struct xt_table_info *newinfo, unsigned int num_counters,
+            void __user *counters_ptr)
+{
+       int ret;
+       struct xt_table *t;
+       struct xt_table_info *oldinfo;
+       struct xt_counters *counters;
+       const void *loc_cpu_old_entry;
+       struct ip6t_entry *iter;
+
+       ret = 0;
+       counters = vzalloc(num_counters * sizeof(struct xt_counters));
+       if (!counters) {
+               ret = -ENOMEM;
+               goto out;
+       }
+
+       t = try_then_request_module(xt_find_table_lock(net, AF_INET6, name),
+                                   "ip6table_%s", name);
+       if (IS_ERR_OR_NULL(t)) {
+               ret = t ? PTR_ERR(t) : -ENOENT;
+               goto free_newinfo_counters_untrans;
+       }
+
+       /* You lied! */
+       if (valid_hooks != t->valid_hooks) {
+               duprintf("Valid hook crap: %08X vs %08X\n",
+                        valid_hooks, t->valid_hooks);
+               ret = -EINVAL;
+               goto put_module;
+       }
+
+       oldinfo = xt_replace_table(t, num_counters, newinfo, &ret);
+       if (!oldinfo)
+               goto put_module;
+
+       /* Update module usage count based on number of rules */
+       duprintf("do_replace: oldnum=%u, initnum=%u, newnum=%u\n",
+               oldinfo->number, oldinfo->initial_entries, newinfo->number);
+       if ((oldinfo->number > oldinfo->initial_entries) ||
+           (newinfo->number <= oldinfo->initial_entries))
+               module_put(t->me);
+       if ((oldinfo->number > oldinfo->initial_entries) &&
+           (newinfo->number <= oldinfo->initial_entries))
+               module_put(t->me);
+
+       /* Get the old counters, and synchronize with replace */
+       get_counters(oldinfo, counters);
+
+       /* Decrease module usage counts and free resource */
+       loc_cpu_old_entry = oldinfo->entries[raw_smp_processor_id()];
+       xt_entry_foreach(iter, loc_cpu_old_entry, oldinfo->size)
+               cleanup_entry(iter, net);
+
+       xt_free_table_info(oldinfo);
+       if (copy_to_user(counters_ptr, counters,
+                        sizeof(struct xt_counters) * num_counters) != 0) {
+               /* Silent error, can't fail, new table is already in place */
+               net_warn_ratelimited("ip6tables: counters copy to user failed while replacing table\n");
+       }
+       vfree(counters);
+       xt_table_unlock(t);
+       return ret;
+
+ put_module:
+       module_put(t->me);
+       xt_table_unlock(t);
+ free_newinfo_counters_untrans:
+       vfree(counters);
+ out:
+       return ret;
+}
+
+static int
+do_replace(struct net *net, const void __user *user, unsigned int len)
+{
+       int ret;
+       struct ip6t_replace tmp;
+       struct xt_table_info *newinfo;
+       void *loc_cpu_entry;
+       struct ip6t_entry *iter;
+
+       if (copy_from_user(&tmp, user, sizeof(tmp)) != 0)
+               return -EFAULT;
+
+       /* overflow check */
+       if (tmp.num_counters >= INT_MAX / sizeof(struct xt_counters))
+               return -ENOMEM;
+       if (tmp.num_counters == 0)
+               return -EINVAL;
+
+       tmp.name[sizeof(tmp.name)-1] = 0;
+
+       newinfo = xt_alloc_table_info(tmp.size);
+       if (!newinfo)
+               return -ENOMEM;
+
+       /* choose the copy that is on our node/cpu */
+       loc_cpu_entry = newinfo->entries[raw_smp_processor_id()];
+       if (copy_from_user(loc_cpu_entry, user + sizeof(tmp),
+                          tmp.size) != 0) {
+               ret = -EFAULT;
+               goto free_newinfo;
+       }
+
+       ret = translate_table(net, newinfo, loc_cpu_entry, &tmp);
+       if (ret != 0)
+               goto free_newinfo;
+
+       duprintf("ip_tables: Translated table\n");
+
+       ret = __do_replace(net, tmp.name, tmp.valid_hooks, newinfo,
+                          tmp.num_counters, tmp.counters);
+       if (ret)
+               goto free_newinfo_untrans;
+       return 0;
+
+ free_newinfo_untrans:
+       xt_entry_foreach(iter, loc_cpu_entry, newinfo->size)
+               cleanup_entry(iter, net);
+ free_newinfo:
+       xt_free_table_info(newinfo);
+       return ret;
+}
+
+static int
+do_add_counters(struct net *net, const void __user *user, unsigned int len,
+               int compat)
+{
+       unsigned int i, curcpu;
+       struct xt_counters_info tmp;
+       struct xt_counters *paddc;
+       unsigned int num_counters;
+       char *name;
+       int size;
+       void *ptmp;
+       struct xt_table *t;
+       const struct xt_table_info *private;
+       int ret = 0;
+       const void *loc_cpu_entry;
+       struct ip6t_entry *iter;
+       unsigned int addend;
+#ifdef CONFIG_COMPAT
+       struct compat_xt_counters_info compat_tmp;
+
+       if (compat) {
+               ptmp = &compat_tmp;
+               size = sizeof(struct compat_xt_counters_info);
+       } else
+#endif
+       {
+               ptmp = &tmp;
+               size = sizeof(struct xt_counters_info);
+       }
+
+       if (copy_from_user(ptmp, user, size) != 0)
+               return -EFAULT;
+
+#ifdef CONFIG_COMPAT
+       if (compat) {
+               num_counters = compat_tmp.num_counters;
+               name = compat_tmp.name;
+       } else
+#endif
+       {
+               num_counters = tmp.num_counters;
+               name = tmp.name;
+       }
+
+       if (len != size + num_counters * sizeof(struct xt_counters))
+               return -EINVAL;
+
+       paddc = vmalloc(len - size);
+       if (!paddc)
+               return -ENOMEM;
+
+       if (copy_from_user(paddc, user + size, len - size) != 0) {
+               ret = -EFAULT;
+               goto free;
+       }
+
+       t = xt_find_table_lock(net, AF_INET6, name);
+       if (IS_ERR_OR_NULL(t)) {
+               ret = t ? PTR_ERR(t) : -ENOENT;
+               goto free;
+       }
+
+
+       local_bh_disable();
+       private = t->private;
+       if (private->number != num_counters) {
+               ret = -EINVAL;
+               goto unlock_up_free;
+       }
+
+       i = 0;
+       /* Choose the copy that is on our node */
+       curcpu = smp_processor_id();
+       addend = xt_write_recseq_begin();
+       loc_cpu_entry = private->entries[curcpu];
+       xt_entry_foreach(iter, loc_cpu_entry, private->size) {
+               ADD_COUNTER(iter->counters, paddc[i].bcnt, paddc[i].pcnt);
+               ++i;
+       }
+       xt_write_recseq_end(addend);
+
+ unlock_up_free:
+       local_bh_enable();
+       xt_table_unlock(t);
+       module_put(t->me);
+ free:
+       vfree(paddc);
+
+       return ret;
+}
+
+#ifdef CONFIG_COMPAT
+struct compat_ip6t_replace {
+       char                    name[XT_TABLE_MAXNAMELEN];
+       u32                     valid_hooks;
+       u32                     num_entries;
+       u32                     size;
+       u32                     hook_entry[NF_INET_NUMHOOKS];
+       u32                     underflow[NF_INET_NUMHOOKS];
+       u32                     num_counters;
+       compat_uptr_t           counters;       /* struct xt_counters * */
+       struct compat_ip6t_entry entries[0];
+};
+
+static int
+compat_copy_entry_to_user(struct ip6t_entry *e, void __user **dstptr,
+                         unsigned int *size, struct xt_counters *counters,
+                         unsigned int i)
+{
+       struct xt_entry_target *t;
+       struct compat_ip6t_entry __user *ce;
+       u_int16_t target_offset, next_offset;
+       compat_uint_t origsize;
+       const struct xt_entry_match *ematch;
+       int ret = 0;
+
+       origsize = *size;
+       ce = (struct compat_ip6t_entry __user *)*dstptr;
+       if (copy_to_user(ce, e, sizeof(struct ip6t_entry)) != 0 ||
+           copy_to_user(&ce->counters, &counters[i],
+           sizeof(counters[i])) != 0)
+               return -EFAULT;
+
+       *dstptr += sizeof(struct compat_ip6t_entry);
+       *size -= sizeof(struct ip6t_entry) - sizeof(struct compat_ip6t_entry);
+
+       xt_ematch_foreach(ematch, e) {
+               ret = xt_compat_match_to_user(ematch, dstptr, size);
+               if (ret != 0)
+                       return ret;
+       }
+       target_offset = e->target_offset - (origsize - *size);
+       t = ip6t_get_target(e);
+       ret = xt_compat_target_to_user(t, dstptr, size);
+       if (ret)
+               return ret;
+       next_offset = e->next_offset - (origsize - *size);
+       if (put_user(target_offset, &ce->target_offset) != 0 ||
+           put_user(next_offset, &ce->next_offset) != 0)
+               return -EFAULT;
+       return 0;
+}
+
+static int
+compat_find_calc_match(struct xt_entry_match *m,
+                      const char *name,
+                      const struct ip6t_ip6 *ipv6,
+                      unsigned int hookmask,
+                      int *size)
+{
+       struct xt_match *match;
+
+       match = xt_request_find_match(NFPROTO_IPV6, m->u.user.name,
+                                     m->u.user.revision);
+       if (IS_ERR(match)) {
+               duprintf("compat_check_calc_match: `%s' not found\n",
+                        m->u.user.name);
+               return PTR_ERR(match);
+       }
+       m->u.kernel.match = match;
+       *size += xt_compat_match_offset(match);
+       return 0;
+}
+
+static void compat_release_entry(struct compat_ip6t_entry *e)
+{
+       struct xt_entry_target *t;
+       struct xt_entry_match *ematch;
+
+       /* Cleanup all matches */
+       xt_ematch_foreach(ematch, e)
+               module_put(ematch->u.kernel.match->me);
+       t = compat_ip6t_get_target(e);
+       module_put(t->u.kernel.target->me);
+}
+
+static int
+check_compat_entry_size_and_hooks(struct compat_ip6t_entry *e,
+                                 struct xt_table_info *newinfo,
+                                 unsigned int *size,
+                                 const unsigned char *base,
+                                 const unsigned char *limit,
+                                 const unsigned int *hook_entries,
+                                 const unsigned int *underflows,
+                                 const char *name)
+{
+       struct xt_entry_match *ematch;
+       struct xt_entry_target *t;
+       struct xt_target *target;
+       unsigned int entry_offset;
+       unsigned int j;
+       int ret, off, h;
+
+       duprintf("check_compat_entry_size_and_hooks %p\n", e);
+       if ((unsigned long)e % __alignof__(struct compat_ip6t_entry) != 0 ||
+           (unsigned char *)e + sizeof(struct compat_ip6t_entry) >= limit) {
+               duprintf("Bad offset %p, limit = %p\n", e, limit);
+               return -EINVAL;
+       }
+
+       if (e->next_offset < sizeof(struct compat_ip6t_entry) +
+                            sizeof(struct compat_xt_entry_target)) {
+               duprintf("checking: element %p size %u\n",
+                        e, e->next_offset);
+               return -EINVAL;
+       }
+
+       /* For purposes of check_entry casting the compat entry is fine */
+       ret = check_entry((struct ip6t_entry *)e, name);
+       if (ret)
+               return ret;
+
+       off = sizeof(struct ip6t_entry) - sizeof(struct compat_ip6t_entry);
+       entry_offset = (void *)e - (void *)base;
+       j = 0;
+       xt_ematch_foreach(ematch, e) {
+               ret = compat_find_calc_match(ematch, name,
+                                            &e->ipv6, e->comefrom, &off);
+               if (ret != 0)
+                       goto release_matches;
+               ++j;
+       }
+
+       t = compat_ip6t_get_target(e);
+       target = xt_request_find_target(NFPROTO_IPV6, t->u.user.name,
+                                       t->u.user.revision);
+       if (IS_ERR(target)) {
+               duprintf("check_compat_entry_size_and_hooks: `%s' not found\n",
+                        t->u.user.name);
+               ret = PTR_ERR(target);
+               goto release_matches;
+       }
+       t->u.kernel.target = target;
+
+       off += xt_compat_target_offset(target);
+       *size += off;
+       ret = xt_compat_add_offset(AF_INET6, entry_offset, off);
+       if (ret)
+               goto out;
+
+       /* Check hooks & underflows */
+       for (h = 0; h < NF_INET_NUMHOOKS; h++) {
+               if ((unsigned char *)e - base == hook_entries[h])
+                       newinfo->hook_entry[h] = hook_entries[h];
+               if ((unsigned char *)e - base == underflows[h])
+                       newinfo->underflow[h] = underflows[h];
+       }
+
+       /* Clear counters and comefrom */
+       memset(&e->counters, 0, sizeof(e->counters));
+       e->comefrom = 0;
+       return 0;
+
+out:
+       module_put(t->u.kernel.target->me);
+release_matches:
+       xt_ematch_foreach(ematch, e) {
+               if (j-- == 0)
+                       break;
+               module_put(ematch->u.kernel.match->me);
+       }
+       return ret;
+}
+
+static int
+compat_copy_entry_from_user(struct compat_ip6t_entry *e, void **dstptr,
+                           unsigned int *size, const char *name,
+                           struct xt_table_info *newinfo, unsigned char *base)
+{
+       struct xt_entry_target *t;
+       struct ip6t_entry *de;
+       unsigned int origsize;
+       int ret, h;
+       struct xt_entry_match *ematch;
+
+       ret = 0;
+       origsize = *size;
+       de = (struct ip6t_entry *)*dstptr;
+       memcpy(de, e, sizeof(struct ip6t_entry));
+       memcpy(&de->counters, &e->counters, sizeof(e->counters));
+
+       *dstptr += sizeof(struct ip6t_entry);
+       *size += sizeof(struct ip6t_entry) - sizeof(struct compat_ip6t_entry);
+
+       xt_ematch_foreach(ematch, e) {
+               ret = xt_compat_match_from_user(ematch, dstptr, size);
+               if (ret != 0)
+                       return ret;
+       }
+       de->target_offset = e->target_offset - (origsize - *size);
+       t = compat_ip6t_get_target(e);
+       xt_compat_target_from_user(t, dstptr, size);
+
+       de->next_offset = e->next_offset - (origsize - *size);
+       for (h = 0; h < NF_INET_NUMHOOKS; h++) {
+               if ((unsigned char *)de - base < newinfo->hook_entry[h])
+                       newinfo->hook_entry[h] -= origsize - *size;
+               if ((unsigned char *)de - base < newinfo->underflow[h])
+                       newinfo->underflow[h] -= origsize - *size;
+       }
+       return ret;
+}
+
+static int compat_check_entry(struct ip6t_entry *e, struct net *net,
+                             const char *name)
+{
+       unsigned int j;
+       int ret = 0;
+       struct xt_mtchk_param mtpar;
+       struct xt_entry_match *ematch;
+
+       j = 0;
+       mtpar.net       = net;
+       mtpar.table     = name;
+       mtpar.entryinfo = &e->ipv6;
+       mtpar.hook_mask = e->comefrom;
+       mtpar.family    = NFPROTO_IPV6;
+       xt_ematch_foreach(ematch, e) {
+               ret = check_match(ematch, &mtpar);
+               if (ret != 0)
+                       goto cleanup_matches;
+               ++j;
+       }
+
+       ret = check_target(e, net, name);
+       if (ret)
+               goto cleanup_matches;
+       return 0;
+
+ cleanup_matches:
+       xt_ematch_foreach(ematch, e) {
+               if (j-- == 0)
+                       break;
+               cleanup_match(ematch, net);
+       }
+       return ret;
+}
+
+static int
+translate_compat_table(struct net *net,
+                      const char *name,
+                      unsigned int valid_hooks,
+                      struct xt_table_info **pinfo,
+                      void **pentry0,
+                      unsigned int total_size,
+                      unsigned int number,
+                      unsigned int *hook_entries,
+                      unsigned int *underflows)
+{
+       unsigned int i, j;
+       struct xt_table_info *newinfo, *info;
+       void *pos, *entry0, *entry1;
+       struct compat_ip6t_entry *iter0;
+       struct ip6t_entry *iter1;
+       unsigned int size;
+       int ret = 0;
+
+       info = *pinfo;
+       entry0 = *pentry0;
+       size = total_size;
+       info->number = number;
+
+       /* Init all hooks to impossible value. */
+       for (i = 0; i < NF_INET_NUMHOOKS; i++) {
+               info->hook_entry[i] = 0xFFFFFFFF;
+               info->underflow[i] = 0xFFFFFFFF;
+       }
+
+       duprintf("translate_compat_table: size %u\n", info->size);
+       j = 0;
+       xt_compat_lock(AF_INET6);
+       xt_compat_init_offsets(AF_INET6, number);
+       /* Walk through entries, checking offsets. */
+       xt_entry_foreach(iter0, entry0, total_size) {
+               ret = check_compat_entry_size_and_hooks(iter0, info, &size,
+                                                       entry0,
+                                                       entry0 + total_size,
+                                                       hook_entries,
+                                                       underflows,
+                                                       name);
+               if (ret != 0)
+                       goto out_unlock;
+               ++j;
+       }
+
+       ret = -EINVAL;
+       if (j != number) {
+               duprintf("translate_compat_table: %u not %u entries\n",
+                        j, number);
+               goto out_unlock;
+       }
+
+       /* Check hooks all assigned */
+       for (i = 0; i < NF_INET_NUMHOOKS; i++) {
+               /* Only hooks which are valid */
+               if (!(valid_hooks & (1 << i)))
+                       continue;
+               if (info->hook_entry[i] == 0xFFFFFFFF) {
+                       duprintf("Invalid hook entry %u %u\n",
+                                i, hook_entries[i]);
+                       goto out_unlock;
+               }
+               if (info->underflow[i] == 0xFFFFFFFF) {
+                       duprintf("Invalid underflow %u %u\n",
+                                i, underflows[i]);
+                       goto out_unlock;
+               }
+       }
+
+       ret = -ENOMEM;
+       newinfo = xt_alloc_table_info(size);
+       if (!newinfo)
+               goto out_unlock;
+
+       newinfo->number = number;
+       for (i = 0; i < NF_INET_NUMHOOKS; i++) {
+               newinfo->hook_entry[i] = info->hook_entry[i];
+               newinfo->underflow[i] = info->underflow[i];
+       }
+       entry1 = newinfo->entries[raw_smp_processor_id()];
+       pos = entry1;
+       size = total_size;
+       xt_entry_foreach(iter0, entry0, total_size) {
+               ret = compat_copy_entry_from_user(iter0, &pos, &size,
+                                                 name, newinfo, entry1);
+               if (ret != 0)
+                       break;
+       }
+       xt_compat_flush_offsets(AF_INET6);
+       xt_compat_unlock(AF_INET6);
+       if (ret)
+               goto free_newinfo;
+
+       ret = -ELOOP;
+       if (!mark_source_chains(newinfo, valid_hooks, entry1))
+               goto free_newinfo;
+
+       i = 0;
+       xt_entry_foreach(iter1, entry1, newinfo->size) {
+               ret = compat_check_entry(iter1, net, name);
+               if (ret != 0)
+                       break;
+               ++i;
+               if (strcmp(ip6t_get_target(iter1)->u.user.name,
+                   XT_ERROR_TARGET) == 0)
+                       ++newinfo->stacksize;
+       }
+       if (ret) {
+               /*
+                * The first i matches need cleanup_entry (calls ->destroy)
+                * because they had called ->check already. The other j-i
+                * entries need only release.
+                */
+               int skip = i;
+               j -= i;
+               xt_entry_foreach(iter0, entry0, newinfo->size) {
+                       if (skip-- > 0)
+                               continue;
+                       if (j-- == 0)
+                               break;
+                       compat_release_entry(iter0);
+               }
+               xt_entry_foreach(iter1, entry1, newinfo->size) {
+                       if (i-- == 0)
+                               break;
+                       cleanup_entry(iter1, net);
+               }
+               xt_free_table_info(newinfo);
+               return ret;
+       }
+
+       /* And one copy for every other CPU */
+       for_each_possible_cpu(i)
+               if (newinfo->entries[i] && newinfo->entries[i] != entry1)
+                       memcpy(newinfo->entries[i], entry1, newinfo->size);
+
+       *pinfo = newinfo;
+       *pentry0 = entry1;
+       xt_free_table_info(info);
+       return 0;
+
+free_newinfo:
+       xt_free_table_info(newinfo);
+out:
+       xt_entry_foreach(iter0, entry0, total_size) {
+               if (j-- == 0)
+                       break;
+               compat_release_entry(iter0);
+       }
+       return ret;
+out_unlock:
+       xt_compat_flush_offsets(AF_INET6);
+       xt_compat_unlock(AF_INET6);
+       goto out;
+}
+
+static int
+compat_do_replace(struct net *net, void __user *user, unsigned int len)
+{
+       int ret;
+       struct compat_ip6t_replace tmp;
+       struct xt_table_info *newinfo;
+       void *loc_cpu_entry;
+       struct ip6t_entry *iter;
+
+       if (copy_from_user(&tmp, user, sizeof(tmp)) != 0)
+               return -EFAULT;
+
+       /* overflow check */
+       if (tmp.size >= INT_MAX / num_possible_cpus())
+               return -ENOMEM;
+       if (tmp.num_counters >= INT_MAX / sizeof(struct xt_counters))
+               return -ENOMEM;
+       if (tmp.num_counters == 0)
+               return -EINVAL;
+
+       tmp.name[sizeof(tmp.name)-1] = 0;
+
+       newinfo = xt_alloc_table_info(tmp.size);
+       if (!newinfo)
+               return -ENOMEM;
+
+       /* choose the copy that is on our node/cpu */
+       loc_cpu_entry = newinfo->entries[raw_smp_processor_id()];
+       if (copy_from_user(loc_cpu_entry, user + sizeof(tmp),
+                          tmp.size) != 0) {
+               ret = -EFAULT;
+               goto free_newinfo;
+       }
+
+       ret = translate_compat_table(net, tmp.name, tmp.valid_hooks,
+                                    &newinfo, &loc_cpu_entry, tmp.size,
+                                    tmp.num_entries, tmp.hook_entry,
+                                    tmp.underflow);
+       if (ret != 0)
+               goto free_newinfo;
+
+       duprintf("compat_do_replace: Translated table\n");
+
+       ret = __do_replace(net, tmp.name, tmp.valid_hooks, newinfo,
+                          tmp.num_counters, compat_ptr(tmp.counters));
+       if (ret)
+               goto free_newinfo_untrans;
+       return 0;
+
+ free_newinfo_untrans:
+       xt_entry_foreach(iter, loc_cpu_entry, newinfo->size)
+               cleanup_entry(iter, net);
+ free_newinfo:
+       xt_free_table_info(newinfo);
+       return ret;
+}
+
+static int
+compat_do_ip6t_set_ctl(struct sock *sk, int cmd, void __user *user,
+                      unsigned int len)
+{
+       int ret;
+
+       if (!ns_capable(sock_net(sk)->user_ns, CAP_NET_ADMIN))
+               return -EPERM;
+
+       switch (cmd) {
+       case IP6T_SO_SET_REPLACE:
+               ret = compat_do_replace(sock_net(sk), user, len);
+               break;
+
+       case IP6T_SO_SET_ADD_COUNTERS:
+               ret = do_add_counters(sock_net(sk), user, len, 1);
+               break;
+
+       default:
+               duprintf("do_ip6t_set_ctl:  unknown request %i\n", cmd);
+               ret = -EINVAL;
+       }
+
+       return ret;
+}
+
+struct compat_ip6t_get_entries {
+       char name[XT_TABLE_MAXNAMELEN];
+       compat_uint_t size;
+       struct compat_ip6t_entry entrytable[0];
+};
+
+static int
+compat_copy_entries_to_user(unsigned int total_size, struct xt_table *table,
+                           void __user *userptr)
+{
+       struct xt_counters *counters;
+       const struct xt_table_info *private = table->private;
+       void __user *pos;
+       unsigned int size;
+       int ret = 0;
+       const void *loc_cpu_entry;
+       unsigned int i = 0;
+       struct ip6t_entry *iter;
+
+       counters = alloc_counters(table);
+       if (IS_ERR(counters))
+               return PTR_ERR(counters);
+
+       /* choose the copy that is on our node/cpu, ...
+        * This choice is lazy (because current thread is
+        * allowed to migrate to another cpu)
+        */
+       loc_cpu_entry = private->entries[raw_smp_processor_id()];
+       pos = userptr;
+       size = total_size;
+       xt_entry_foreach(iter, loc_cpu_entry, total_size) {
+               ret = compat_copy_entry_to_user(iter, &pos,
+                                               &size, counters, i++);
+               if (ret != 0)
+                       break;
+       }
+
+       vfree(counters);
+       return ret;
+}
+
+static int
+compat_get_entries(struct net *net, struct compat_ip6t_get_entries __user *uptr,
+                  int *len)
+{
+       int ret;
+       struct compat_ip6t_get_entries get;
+       struct xt_table *t;
+
+       if (*len < sizeof(get)) {
+               duprintf("compat_get_entries: %u < %zu\n", *len, sizeof(get));
+               return -EINVAL;
+       }
+
+       if (copy_from_user(&get, uptr, sizeof(get)) != 0)
+               return -EFAULT;
+
+       if (*len != sizeof(struct compat_ip6t_get_entries) + get.size) {
+               duprintf("compat_get_entries: %u != %zu\n",
+                        *len, sizeof(get) + get.size);
+               return -EINVAL;
+       }
+
+       xt_compat_lock(AF_INET6);
+       t = xt_find_table_lock(net, AF_INET6, get.name);
+       if (!IS_ERR_OR_NULL(t)) {
+               const struct xt_table_info *private = t->private;
+               struct xt_table_info info;
+               duprintf("t->private->number = %u\n", private->number);
+               ret = compat_table_info(private, &info);
+               if (!ret && get.size == info.size) {
+                       ret = compat_copy_entries_to_user(private->size,
+                                                         t, uptr->entrytable);
+               } else if (!ret) {
+                       duprintf("compat_get_entries: I've got %u not %u!\n",
+                                private->size, get.size);
+                       ret = -EAGAIN;
+               }
+               xt_compat_flush_offsets(AF_INET6);
+               module_put(t->me);
+               xt_table_unlock(t);
+       } else
+               ret = t ? PTR_ERR(t) : -ENOENT;
+
+       xt_compat_unlock(AF_INET6);
+       return ret;
+}
+
+static int do_ip6t_get_ctl(struct sock *, int, void __user *, int *);
+
+static int
+compat_do_ip6t_get_ctl(struct sock *sk, int cmd, void __user *user, int *len)
+{
+       int ret;
+
+       if (!ns_capable(sock_net(sk)->user_ns, CAP_NET_ADMIN))
+               return -EPERM;
+
+       switch (cmd) {
+       case IP6T_SO_GET_INFO:
+               ret = get_info(sock_net(sk), user, len, 1);
+               break;
+       case IP6T_SO_GET_ENTRIES:
+               ret = compat_get_entries(sock_net(sk), user, len);
+               break;
+       default:
+               ret = do_ip6t_get_ctl(sk, cmd, user, len);
+       }
+       return ret;
+}
+#endif
+
+static int
+do_ip6t_set_ctl(struct sock *sk, int cmd, void __user *user, unsigned int len)
+{
+       int ret;
+
+       if (!ns_capable(sock_net(sk)->user_ns, CAP_NET_ADMIN))
+               return -EPERM;
+
+       switch (cmd) {
+       case IP6T_SO_SET_REPLACE:
+               ret = do_replace(sock_net(sk), user, len);
+               break;
+
+       case IP6T_SO_SET_ADD_COUNTERS:
+               ret = do_add_counters(sock_net(sk), user, len, 0);
+               break;
+
+       default:
+               duprintf("do_ip6t_set_ctl:  unknown request %i\n", cmd);
+               ret = -EINVAL;
+       }
+
+       return ret;
+}
+
+static int
+do_ip6t_get_ctl(struct sock *sk, int cmd, void __user *user, int *len)
+{
+       int ret;
+
+       if (!ns_capable(sock_net(sk)->user_ns, CAP_NET_ADMIN))
+               return -EPERM;
+
+       switch (cmd) {
+       case IP6T_SO_GET_INFO:
+               ret = get_info(sock_net(sk), user, len, 0);
+               break;
+
+       case IP6T_SO_GET_ENTRIES:
+               ret = get_entries(sock_net(sk), user, len);
+               break;
+
+       case IP6T_SO_GET_REVISION_MATCH:
+       case IP6T_SO_GET_REVISION_TARGET: {
+               struct xt_get_revision rev;
+               int target;
+
+               if (*len != sizeof(rev)) {
+                       ret = -EINVAL;
+                       break;
+               }
+               if (copy_from_user(&rev, user, sizeof(rev)) != 0) {
+                       ret = -EFAULT;
+                       break;
+               }
+               rev.name[sizeof(rev.name)-1] = 0;
+
+               if (cmd == IP6T_SO_GET_REVISION_TARGET)
+                       target = 1;
+               else
+                       target = 0;
+
+               try_then_request_module(xt_find_revision(AF_INET6, rev.name,
+                                                        rev.revision,
+                                                        target, &ret),
+                                       "ip6t_%s", rev.name);
+               break;
+       }
+
+       default:
+               duprintf("do_ip6t_get_ctl: unknown request %i\n", cmd);
+               ret = -EINVAL;
+       }
+
+       return ret;
+}
+
+struct xt_table *ip6t_register_table(struct net *net,
+                                    const struct xt_table *table,
+                                    const struct ip6t_replace *repl)
+{
+       int ret;
+       struct xt_table_info *newinfo;
+       struct xt_table_info bootstrap = {0};
+       void *loc_cpu_entry;
+       struct xt_table *new_table;
+
+       newinfo = xt_alloc_table_info(repl->size);
+       if (!newinfo) {
+               ret = -ENOMEM;
+               goto out;
+       }
+
+       /* choose the copy on our node/cpu, but dont care about preemption */
+       loc_cpu_entry = newinfo->entries[raw_smp_processor_id()];
+       memcpy(loc_cpu_entry, repl->entries, repl->size);
+
+       ret = translate_table(net, newinfo, loc_cpu_entry, repl);
+       if (ret != 0)
+               goto out_free;
+
+       new_table = xt_register_table(net, table, &bootstrap, newinfo);
+       if (IS_ERR(new_table)) {
+               ret = PTR_ERR(new_table);
+               goto out_free;
+       }
+       return new_table;
+
+out_free:
+       xt_free_table_info(newinfo);
+out:
+       return ERR_PTR(ret);
+}
+
+void ip6t_unregister_table(struct net *net, struct xt_table *table)
+{
+       struct xt_table_info *private;
+       void *loc_cpu_entry;
+       struct module *table_owner = table->me;
+       struct ip6t_entry *iter;
+
+       private = xt_unregister_table(table);
+
+       /* Decrease module usage counts and free resources */
+       loc_cpu_entry = private->entries[raw_smp_processor_id()];
+       xt_entry_foreach(iter, loc_cpu_entry, private->size)
+               cleanup_entry(iter, net);
+       if (private->number > private->initial_entries)
+               module_put(table_owner);
+       xt_free_table_info(private);
+}
+
+/* Returns 1 if the type and code is matched by the range, 0 otherwise */
+static inline bool
+icmp6_type_code_match(u_int8_t test_type, u_int8_t min_code, u_int8_t max_code,
+                    u_int8_t type, u_int8_t code,
+                    bool invert)
+{
+       return (type == test_type && code >= min_code && code <= max_code)
+               ^ invert;
+}
+
+static bool
+icmp6_match(const struct sk_buff *skb, struct xt_action_param *par)
+{
+       const struct icmp6hdr *ic;
+       struct icmp6hdr _icmph;
+       const struct ip6t_icmp *icmpinfo = par->matchinfo;
+
+       /* Must not be a fragment. */
+       if (par->fragoff != 0)
+               return false;
+
+       ic = skb_header_pointer(skb, par->thoff, sizeof(_icmph), &_icmph);
+       if (ic == NULL) {
+               /* We've been asked to examine this packet, and we
+                * can't.  Hence, no choice but to drop.
+                */
+               duprintf("Dropping evil ICMP tinygram.\n");
+               par->hotdrop = true;
+               return false;
+       }
+
+       return icmp6_type_code_match(icmpinfo->type,
+                                    icmpinfo->code[0],
+                                    icmpinfo->code[1],
+                                    ic->icmp6_type, ic->icmp6_code,
+                                    !!(icmpinfo->invflags&IP6T_ICMP_INV));
+}
+
+/* Called when user tries to insert an entry of this type. */
+static int icmp6_checkentry(const struct xt_mtchk_param *par)
+{
+       const struct ip6t_icmp *icmpinfo = par->matchinfo;
+
+       /* Must specify no unknown invflags */
+       return (icmpinfo->invflags & ~IP6T_ICMP_INV) ? -EINVAL : 0;
+}
+
+/* The built-in targets: standard (NULL) and error. */
+static struct xt_target ip6t_builtin_tg[] __read_mostly = {
+       {
+               .name             = XT_STANDARD_TARGET,
+               .targetsize       = sizeof(int),
+               .family           = NFPROTO_IPV6,
+#ifdef CONFIG_COMPAT
+               .compatsize       = sizeof(compat_int_t),
+               .compat_from_user = compat_standard_from_user,
+               .compat_to_user   = compat_standard_to_user,
+#endif
+       },
+       {
+               .name             = XT_ERROR_TARGET,
+               .target           = ip6t_error,
+               .targetsize       = XT_FUNCTION_MAXNAMELEN,
+               .family           = NFPROTO_IPV6,
+       },
+};
+
+static struct nf_sockopt_ops ip6t_sockopts = {
+       .pf             = PF_INET6,
+       .set_optmin     = IP6T_BASE_CTL,
+       .set_optmax     = IP6T_SO_SET_MAX+1,
+       .set            = do_ip6t_set_ctl,
+#ifdef CONFIG_COMPAT
+       .compat_set     = compat_do_ip6t_set_ctl,
+#endif
+       .get_optmin     = IP6T_BASE_CTL,
+       .get_optmax     = IP6T_SO_GET_MAX+1,
+       .get            = do_ip6t_get_ctl,
+#ifdef CONFIG_COMPAT
+       .compat_get     = compat_do_ip6t_get_ctl,
+#endif
+       .owner          = THIS_MODULE,
+};
+
+static struct xt_match ip6t_builtin_mt[] __read_mostly = {
+       {
+               .name       = "icmp6",
+               .match      = icmp6_match,
+               .matchsize  = sizeof(struct ip6t_icmp),
+               .checkentry = icmp6_checkentry,
+               .proto      = IPPROTO_ICMPV6,
+               .family     = NFPROTO_IPV6,
+       },
+};
+
+static int __net_init ip6_tables_net_init(struct net *net)
+{
+       return xt_proto_init(net, NFPROTO_IPV6);
+}
+
+static void __net_exit ip6_tables_net_exit(struct net *net)
+{
+       xt_proto_fini(net, NFPROTO_IPV6);
+}
+
+static struct pernet_operations ip6_tables_net_ops = {
+       .init = ip6_tables_net_init,
+       .exit = ip6_tables_net_exit,
+};
+
+static int __init ip6_tables_init(void)
+{
+       int ret;
+
+       ret = register_pernet_subsys(&ip6_tables_net_ops);
+       if (ret < 0)
+               goto err1;
+
+       /* No one else will be downing sem now, so we won't sleep */
+       ret = xt_register_targets(ip6t_builtin_tg, ARRAY_SIZE(ip6t_builtin_tg));
+       if (ret < 0)
+               goto err2;
+       ret = xt_register_matches(ip6t_builtin_mt, ARRAY_SIZE(ip6t_builtin_mt));
+       if (ret < 0)
+               goto err4;
+
+       /* Register setsockopt */
+       ret = nf_register_sockopt(&ip6t_sockopts);
+       if (ret < 0)
+               goto err5;
+
+       pr_info("(C) 2000-2006 Netfilter Core Team\n");
+       return 0;
+
+err5:
+       xt_unregister_matches(ip6t_builtin_mt, ARRAY_SIZE(ip6t_builtin_mt));
+err4:
+       xt_unregister_targets(ip6t_builtin_tg, ARRAY_SIZE(ip6t_builtin_tg));
+err2:
+       unregister_pernet_subsys(&ip6_tables_net_ops);
+err1:
+       return ret;
+}
+
+static void __exit ip6_tables_fini(void)
+{
+       nf_unregister_sockopt(&ip6t_sockopts);
+
+       xt_unregister_matches(ip6t_builtin_mt, ARRAY_SIZE(ip6t_builtin_mt));
+       xt_unregister_targets(ip6t_builtin_tg, ARRAY_SIZE(ip6t_builtin_tg));
+       unregister_pernet_subsys(&ip6_tables_net_ops);
+}
+
+EXPORT_SYMBOL(ip6t_register_table);
+EXPORT_SYMBOL(ip6t_unregister_table);
+EXPORT_SYMBOL(ip6t_do_table);
+
+module_init(ip6_tables_init);
+module_exit(ip6_tables_fini);