]> err.no Git - linux-2.6/commitdiff
[NETFILTER]: Add "nfnetlink_queue" netfilter queue handler over nfnetlink
authorHarald Welte <laforge@netfilter.org>
Wed, 10 Aug 2005 02:44:15 +0000 (19:44 -0700)
committerDavid S. Miller <davem@sunset.davemloft.net>
Mon, 29 Aug 2005 22:36:56 +0000 (15:36 -0700)
- Add new nfnetlink_queue module
- Add new ipt_NFQUEUE and ip6t_NFQUEUE modules to access queue numbers 1-65535
- Mark ip_queue and ip6_queue Kconfig options as OBSOLETE
- Update feature-removal-schedule to remove ip[6]_queue in December

Signed-off-by: Harald Welte <laforge@netfilter.org>
Signed-off-by: David S. Miller <davem@davemloft.net>
12 files changed:
Documentation/feature-removal-schedule.txt
include/linux/netfilter/nfnetlink_queue.h [new file with mode: 0644]
include/linux/netfilter_ipv4/ipt_NFQUEUE.h [new file with mode: 0644]
net/ipv4/netfilter/Kconfig
net/ipv4/netfilter/Makefile
net/ipv4/netfilter/ipt_NFQUEUE.c [new file with mode: 0644]
net/ipv6/netfilter/Kconfig
net/ipv6/netfilter/Makefile
net/ipv6/netfilter/ip6t_NFQUEUE.c [new file with mode: 0644]
net/netfilter/Kconfig
net/netfilter/Makefile
net/netfilter/nfnetlink_queue.c [new file with mode: 0644]

index 8b1430b4665571d613dd1f85c6a4e4e22d387755..0665cb12bd6650f65f692ab726212deed76422c9 100644 (file)
@@ -135,3 +135,15 @@ Why:       With the 16-bit PCMCIA subsystem now behaving (almost) like a
        pcmciautils package available at
        http://kernel.org/pub/linux/utils/kernel/pcmcia/
 Who:   Dominik Brodowski <linux@brodo.de>
+
+---------------------------
+
+What:  ip_queue and ip6_queue (old ipv4-only and ipv6-only netfilter queue)
+When:  December 2005
+Why:   This interface has been obsoleted by the new layer3-independent
+       "nfnetlink_queue".  The Kernel interface is compatible, so the old
+       ip[6]tables "QUEUE" targets still work and will transparently handle
+       all packets into nfnetlink queue number 0.  Userspace users will have
+       to link against API-compatible library on top of libnfnetlink_queue 
+       instead of the current 'libipq'.
+Who:   Harald Welte <laforge@netfilter.org>
diff --git a/include/linux/netfilter/nfnetlink_queue.h b/include/linux/netfilter/nfnetlink_queue.h
new file mode 100644 (file)
index 0000000..edb463a
--- /dev/null
@@ -0,0 +1,85 @@
+#ifndef _NFNETLINK_QUEUE_H
+#define _NFNETLINK_QUEUE_H
+
+#include <linux/netfilter/nfnetlink.h>
+
+enum nfqnl_msg_types {
+       NFQNL_MSG_PACKET,               /* packet from kernel to userspace */
+       NFQNL_MSG_VERDICT,              /* verdict from userspace to kernel */
+       NFQNL_MSG_CONFIG,               /* connect to a particular queue */
+
+       NFQNL_MSG_MAX
+};
+
+struct nfqnl_msg_packet_hdr {
+       u_int32_t       packet_id;      /* unique ID of packet in queue */
+       u_int16_t       hw_protocol;    /* hw protocol (network order) */
+       u_int8_t        hook;           /* netfilter hook */
+} __attribute__ ((packed));
+
+struct nfqnl_msg_packet_hw {
+       u_int16_t       hw_addrlen;
+       u_int16_t       _pad;
+       u_int8_t        hw_addr[8];
+} __attribute__ ((packed));
+
+struct nfqnl_msg_packet_timestamp {
+       u_int64_t       sec;
+       u_int64_t       usec;
+} __attribute__ ((packed));
+
+enum nfqnl_attr_type {
+       NFQA_UNSPEC,
+       NFQA_PACKET_HDR,
+       NFQA_VERDICT_HDR,               /* nfqnl_msg_verdict_hrd */
+       NFQA_MARK,                      /* u_int32_t nfmark */
+       NFQA_TIMESTAMP,                 /* nfqnl_msg_packet_timestamp */
+       NFQA_IFINDEX_INDEV,             /* u_int32_t ifindex */
+       NFQA_IFINDEX_OUTDEV,            /* u_int32_t ifindex */
+       NFQA_HWADDR,                    /* nfqnl_msg_packet_hw */
+       NFQA_PAYLOAD,                   /* opaque data payload */
+
+       __NFQA_MAX
+};
+#define NFQA_MAX (__NFQA_MAX - 1)
+
+struct nfqnl_msg_verdict_hdr {
+       u_int32_t verdict;
+       u_int32_t id;
+} __attribute__ ((packed));
+
+
+enum nfqnl_msg_config_cmds {
+       NFQNL_CFG_CMD_NONE,
+       NFQNL_CFG_CMD_BIND,
+       NFQNL_CFG_CMD_UNBIND,
+       NFQNL_CFG_CMD_PF_BIND,
+       NFQNL_CFG_CMD_PF_UNBIND,
+};
+
+struct nfqnl_msg_config_cmd {
+       u_int8_t        command;        /* nfqnl_msg_config_cmds */
+       u_int8_t        _pad;
+       u_int16_t       pf;             /* AF_xxx for PF_[UN]BIND */
+} __attribute__ ((packed));
+
+enum nfqnl_config_mode {
+       NFQNL_COPY_NONE,
+       NFQNL_COPY_META,
+       NFQNL_COPY_PACKET,
+};
+
+struct nfqnl_msg_config_params {
+       u_int32_t       copy_range;
+       u_int8_t        copy_mode;      /* enum nfqnl_config_mode */
+} __attribute__ ((packed));
+
+
+enum nfqnl_attr_config {
+       NFQA_CFG_UNSPEC,
+       NFQA_CFG_CMD,                   /* nfqnl_msg_config_cmd */
+       NFQA_CFG_PARAMS,                /* nfqnl_msg_config_params */
+       __NFQA_CFG_MAX
+};
+
+#endif /* _NFNETLINK_QUEUE_H */
diff --git a/include/linux/netfilter_ipv4/ipt_NFQUEUE.h b/include/linux/netfilter_ipv4/ipt_NFQUEUE.h
new file mode 100644 (file)
index 0000000..b5b2943
--- /dev/null
@@ -0,0 +1,16 @@
+/* iptables module for using NFQUEUE mechanism
+ *
+ * (C) 2005 Harald Welte <laforge@netfilter.org>
+ *
+ * This software is distributed under GNU GPL v2, 1991
+ * 
+*/
+#ifndef _IPT_NFQ_TARGET_H
+#define _IPT_NFQ_TARGET_H
+
+/* target info */
+struct ipt_NFQ_info {
+       u_int16_t queuenum;
+};
+
+#endif /* _IPT_DSCP_TARGET_H */
index e47ba39eb6571a7a1e738feb567a0572e40078c9..2fa26a41fa47e3fb04061b057b9a74d60ef63937 100644 (file)
@@ -110,11 +110,15 @@ config IP_NF_AMANDA
          To compile it as a module, choose M here.  If unsure, say Y.
 
 config IP_NF_QUEUE
-       tristate "Userspace queueing via NETLINK"
+       tristate "IP Userspace queueing via NETLINK (OBSOLETE)"
        help
          Netfilter has the ability to queue packets to user space: the
          netlink device can be used to access them using this driver.
 
+         This option enables the old IPv4-only "ip_queue" implementation
+         which has been obsoleted by the new "nfnetlink_queue" code (see
+         CONFIG_NETFILTER_NETLINK_QUEUE).
+
          To compile it as a module, choose M here.  If unsure, say N.
 
 config IP_NF_IPTABLES
index abf2a7d1a584ae89a3d81ee08f67b362fec7321f..c2ae663b723f58fc023f28f55b306501bb63bc35 100644 (file)
@@ -91,3 +91,4 @@ obj-$(CONFIG_IP_NF_ARP_MANGLE) += arpt_mangle.o
 obj-$(CONFIG_IP_NF_ARPFILTER) += arptable_filter.o
 
 obj-$(CONFIG_IP_NF_QUEUE) += ip_queue.o
+obj-$(CONFIG_NETFILTER_NETLINK_QUEUE) += ipt_NFQUEUE.o
diff --git a/net/ipv4/netfilter/ipt_NFQUEUE.c b/net/ipv4/netfilter/ipt_NFQUEUE.c
new file mode 100644 (file)
index 0000000..3cedc9b
--- /dev/null
@@ -0,0 +1,70 @@
+/* iptables module for using new netfilter netlink queue
+ *
+ * (C) 2005 by Harald Welte <laforge@netfilter.org>
+ *
+ * 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.
+ * 
+ */
+
+#include <linux/module.h>
+#include <linux/skbuff.h>
+
+#include <linux/netfilter.h>
+#include <linux/netfilter_ipv4/ip_tables.h>
+#include <linux/netfilter_ipv4/ipt_NFQUEUE.h>
+
+MODULE_AUTHOR("Harald Welte <laforge@netfilter.org>");
+MODULE_DESCRIPTION("iptables NFQUEUE target");
+MODULE_LICENSE("GPL");
+
+static unsigned int
+target(struct sk_buff **pskb,
+       const struct net_device *in,
+       const struct net_device *out,
+       unsigned int hooknum,
+       const void *targinfo,
+       void *userinfo)
+{
+       const struct ipt_NFQ_info *tinfo = targinfo;
+
+       return NF_QUEUE_NR(tinfo->queuenum);
+}
+
+static int
+checkentry(const char *tablename,
+          const struct ipt_entry *e,
+           void *targinfo,
+           unsigned int targinfosize,
+           unsigned int hook_mask)
+{
+       if (targinfosize != IPT_ALIGN(sizeof(struct ipt_NFQ_info))) {
+               printk(KERN_WARNING "NFQUEUE: targinfosize %u != %Zu\n",
+                      targinfosize,
+                      IPT_ALIGN(sizeof(struct ipt_NFQ_info)));
+               return 0;
+       }
+
+       return 1;
+}
+
+static struct ipt_target ipt_NFQ_reg = {
+       .name           = "NFQUEUE",
+       .target         = target,
+       .checkentry     = checkentry,
+       .me             = THIS_MODULE,
+};
+
+static int __init init(void)
+{
+       return ipt_register_target(&ipt_NFQ_reg);
+}
+
+static void __exit fini(void)
+{
+       ipt_unregister_target(&ipt_NFQ_reg);
+}
+
+module_init(init);
+module_exit(fini);
index 77ec704c9ee34c6315c07f917bc5b17da5e5203e..cd1551983c636d3428080bf68b397e9deca26c90 100644 (file)
@@ -10,13 +10,16 @@ menu "IPv6: Netfilter Configuration (EXPERIMENTAL)"
 #  dep_tristate '  FTP protocol support' CONFIG_IP6_NF_FTP $CONFIG_IP6_NF_CONNTRACK
 #fi
 config IP6_NF_QUEUE
-       tristate "Userspace queueing via NETLINK"
+       tristate "IP6 Userspace queueing via NETLINK (OBSOLETE)"
        ---help---
 
          This option adds a queue handler to the kernel for IPv6
-         packets which lets us to receive the filtered packets
-         with QUEUE target using libiptc as we can do with
-         the IPv4 now.
+         packets which enables users to receive the filtered packets
+         with QUEUE target using libipq.
+
+         THis option enables the old IPv6-only "ip6_queue" implementation
+         which has been obsoleted by the new "nfnetlink_queue" code (see
+         CONFIG_NETFILTER_NETLINK_QUEUE).
 
          (C) Fernando Anton 2001
          IPv64 Project - Work based in IPv64 draft by Arturo Azcorra.
index 2e51714953b6576d74f328dc9c7c397131692f1f..847651dbcd2a8445ca48b3454f07fbaa8d4126c2 100644 (file)
@@ -24,3 +24,4 @@ obj-$(CONFIG_IP6_NF_QUEUE) += ip6_queue.o
 obj-$(CONFIG_IP6_NF_TARGET_LOG) += ip6t_LOG.o
 obj-$(CONFIG_IP6_NF_RAW) += ip6table_raw.o
 obj-$(CONFIG_IP6_NF_MATCH_HL) += ip6t_hl.o
+obj-$(CONFIG_NETFILTER_NETLINK_QUEUE) += ip6t_NFQUEUE.o
diff --git a/net/ipv6/netfilter/ip6t_NFQUEUE.c b/net/ipv6/netfilter/ip6t_NFQUEUE.c
new file mode 100644 (file)
index 0000000..c6e3730
--- /dev/null
@@ -0,0 +1,70 @@
+/* ip6tables module for using new netfilter netlink queue
+ *
+ * (C) 2005 by Harald Welte <laforge@netfilter.org>
+ *
+ * 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.
+ * 
+ */
+
+#include <linux/module.h>
+#include <linux/skbuff.h>
+
+#include <linux/netfilter.h>
+#include <linux/netfilter_ipv6/ip6_tables.h>
+#include <linux/netfilter_ipv4/ipt_NFQUEUE.h>
+
+MODULE_AUTHOR("Harald Welte <laforge@netfilter.org>");
+MODULE_DESCRIPTION("ip6tables NFQUEUE target");
+MODULE_LICENSE("GPL");
+
+static unsigned int
+target(struct sk_buff **pskb,
+       const struct net_device *in,
+       const struct net_device *out,
+       unsigned int hooknum,
+       const void *targinfo,
+       void *userinfo)
+{
+       const struct ipt_NFQ_info *tinfo = targinfo;
+
+       return NF_QUEUE_NR(tinfo->queuenum);
+}
+
+static int
+checkentry(const char *tablename,
+          const struct ip6t_entry *e,
+           void *targinfo,
+           unsigned int targinfosize,
+           unsigned int hook_mask)
+{
+       if (targinfosize != IP6T_ALIGN(sizeof(struct ipt_NFQ_info))) {
+               printk(KERN_WARNING "NFQUEUE: targinfosize %u != %Zu\n",
+                      targinfosize,
+                      IP6T_ALIGN(sizeof(struct ipt_NFQ_info)));
+               return 0;
+       }
+
+       return 1;
+}
+
+static struct ip6t_target ipt_NFQ_reg = {
+       .name           = "NFQUEUE",
+       .target         = target,
+       .checkentry     = checkentry,
+       .me             = THIS_MODULE,
+};
+
+static int __init init(void)
+{
+       return ip6t_register_target(&ipt_NFQ_reg);
+}
+
+static void __exit fini(void)
+{
+       ip6t_unregister_target(&ipt_NFQ_reg);
+}
+
+module_init(init);
+module_exit(fini);
index 3629d3d1776d174dcc8ea11bbf36839619628d41..f0eb23e5c5f170059cdf7001e231c6595e5fb4db 100644 (file)
@@ -3,3 +3,11 @@ config NETFILTER_NETLINK
        help
          If this option is enabled, the kernel will include support
          for the new netfilter netlink interface.
+
+config NETFILTER_NETLINK_QUEUE
+       tristate "Netfilter NFQUEUE over NFNETLINK interface"
+       depends on NETFILTER_NETLINK
+       help
+         If this option isenabled, the kernel will include support
+         for queueing packets via NFNETLINK.
+         
index 02e67d37194168abd84107fe5f417b8f937babdb..14a0b187e75e4b6eb969ac1eafd18a43908446df 100644 (file)
@@ -1 +1,2 @@
 obj-$(CONFIG_NETFILTER_NETLINK) += nfnetlink.o
+obj-$(CONFIG_NETFILTER_NETLINK_QUEUE) += nfnetlink_queue.o
diff --git a/net/netfilter/nfnetlink_queue.c b/net/netfilter/nfnetlink_queue.c
new file mode 100644 (file)
index 0000000..2403261
--- /dev/null
@@ -0,0 +1,877 @@
+/*
+ * This is a module which is used for queueing packets and communicating with
+ * userspace via nfetlink.
+ *
+ * (C) 2005 by Harald Welte <laforge@netfilter.org>
+ *
+ * Based on the old ipv4-only ip_queue.c:
+ * (C) 2000-2002 James Morris <jmorris@intercode.com.au>
+ * (C) 2003-2005 Netfilter Core Team <coreteam@netfilter.org>
+ *
+ * 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.
+ *
+ */
+#include <linux/module.h>
+#include <linux/skbuff.h>
+#include <linux/init.h>
+#include <linux/spinlock.h>
+#include <linux/notifier.h>
+#include <linux/netdevice.h>
+#include <linux/netfilter.h>
+#include <linux/netfilter_ipv4.h>
+#include <linux/netfilter_ipv6.h>
+#include <linux/netfilter/nfnetlink.h>
+#include <linux/netfilter/nfnetlink_queue.h>
+#include <linux/list.h>
+#include <net/sock.h>
+
+#include <asm/atomic.h>
+
+#define NFQNL_QMAX_DEFAULT 1024
+
+#if 0
+#define QDEBUG(x, args ...)    printk(KERN_DEBUG "%s(%d):%s(): " x,       \
+                                       __FILE__, __LINE__, __FUNCTION__,  \
+                                       ## args)
+#else
+#define QDEBUG(x, ...)
+#endif
+
+struct nfqnl_queue_entry {
+       struct list_head list;
+       struct nf_info *info;
+       struct sk_buff *skb;
+       unsigned int id;
+};
+
+struct nfqnl_instance {
+       struct hlist_node hlist;                /* global list of queues */
+
+       int peer_pid;
+       unsigned int queue_maxlen;
+       unsigned int copy_range;
+       unsigned int queue_total;
+       unsigned int queue_dropped;
+       unsigned int queue_user_dropped;
+
+       atomic_t id_sequence;                   /* 'sequence' of pkt ids */
+
+       u_int16_t queue_num;                    /* number of this queue */
+       u_int8_t copy_mode;
+
+       spinlock_t lock;
+
+       struct list_head queue_list;            /* packets in queue */
+};
+
+typedef int (*nfqnl_cmpfn)(struct nfqnl_queue_entry *, unsigned long);
+
+static DEFINE_RWLOCK(instances_lock);
+
+u_int64_t htonll(u_int64_t in)
+{
+       u_int64_t out;
+       int i;
+
+       for (i = 0; i < sizeof(u_int64_t); i++)
+               ((u_int8_t *)&out)[sizeof(u_int64_t)-1] = ((u_int8_t *)&in)[i];
+
+       return out;
+}
+
+#define INSTANCE_BUCKETS       16
+static struct hlist_head instance_table[INSTANCE_BUCKETS];
+
+static inline u_int8_t instance_hashfn(u_int16_t queue_num)
+{
+       return ((queue_num >> 8) | queue_num) % INSTANCE_BUCKETS;
+}
+
+static struct nfqnl_instance *
+__instance_lookup(u_int16_t queue_num)
+{
+       struct hlist_head *head;
+       struct hlist_node *pos;
+       struct nfqnl_instance *inst;
+
+       head = &instance_table[instance_hashfn(queue_num)];
+       hlist_for_each_entry(inst, pos, head, hlist) {
+               if (inst->queue_num == queue_num)
+                       return inst;
+       }
+       return NULL;
+}
+
+static struct nfqnl_instance *
+instance_lookup(u_int16_t queue_num)
+{
+       struct nfqnl_instance *inst;
+
+       read_lock_bh(&instances_lock);
+       inst = __instance_lookup(queue_num);
+       read_unlock_bh(&instances_lock);
+
+       return inst;
+}
+
+static struct nfqnl_instance *
+instance_create(u_int16_t queue_num, int pid)
+{
+       struct nfqnl_instance *inst;
+
+       QDEBUG("entering for queue_num=%u, pid=%d\n", queue_num, pid);
+
+       write_lock_bh(&instances_lock); 
+       if (__instance_lookup(queue_num)) {
+               inst = NULL;
+               QDEBUG("aborting, instance already exists\n");
+               goto out_unlock;
+       }
+
+       inst = kmalloc(sizeof(*inst), GFP_ATOMIC);
+       if (!inst)
+               goto out_unlock;
+
+       memset(inst, 0, sizeof(*inst));
+       inst->queue_num = queue_num;
+       inst->peer_pid = pid;
+       inst->queue_maxlen = NFQNL_QMAX_DEFAULT;
+       inst->copy_range = 0xfffff;
+       inst->copy_mode = NFQNL_COPY_NONE;
+       atomic_set(&inst->id_sequence, 0);
+       inst->lock = SPIN_LOCK_UNLOCKED;
+       INIT_LIST_HEAD(&inst->queue_list);
+
+       if (!try_module_get(THIS_MODULE))
+               goto out_free;
+
+       hlist_add_head(&inst->hlist, 
+                      &instance_table[instance_hashfn(queue_num)]);
+
+       write_unlock_bh(&instances_lock);
+
+       QDEBUG("successfully created new instance\n");
+
+       return inst;
+
+out_free:
+       kfree(inst);
+out_unlock:
+       write_unlock_bh(&instances_lock);
+       return NULL;
+}
+
+static void nfqnl_flush(struct nfqnl_instance *queue, int verdict);
+
+static void
+_instance_destroy2(struct nfqnl_instance *inst, int lock)
+{
+       /* first pull it out of the global list */
+       if (lock)
+               write_lock_bh(&instances_lock);
+
+       QDEBUG("removing instance %p (queuenum=%u) from hash\n",
+               inst, inst->queue_num);
+       hlist_del(&inst->hlist);
+
+       if (lock)
+               write_unlock_bh(&instances_lock);
+
+       /* then flush all pending skbs from the queue */
+       nfqnl_flush(inst, NF_DROP);
+
+       /* and finally free the data structure */
+       kfree(inst);
+
+       module_put(THIS_MODULE);
+}
+
+static inline void
+__instance_destroy(struct nfqnl_instance *inst)
+{
+       _instance_destroy2(inst, 0);
+}
+
+static inline void
+instance_destroy(struct nfqnl_instance *inst)
+{
+       _instance_destroy2(inst, 1);
+}
+
+
+
+static void
+issue_verdict(struct nfqnl_queue_entry *entry, int verdict)
+{
+       QDEBUG("entering for entry %p, verdict %u\n", entry, verdict);
+
+       /* TCP input path (and probably other bits) assume to be called
+        * from softirq context, not from syscall, like issue_verdict is
+        * called.  TCP input path deadlocks with locks taken from timer
+        * softirq, e.g.  We therefore emulate this by local_bh_disable() */
+
+       local_bh_disable();
+       nf_reinject(entry->skb, entry->info, verdict);
+       local_bh_enable();
+
+       kfree(entry);
+}
+
+static inline void
+__enqueue_entry(struct nfqnl_instance *queue,
+                     struct nfqnl_queue_entry *entry)
+{
+       list_add(&entry->list, &queue->queue_list);
+       queue->queue_total++;
+}
+
+/*
+ * Find and return a queued entry matched by cmpfn, or return the last
+ * entry if cmpfn is NULL.
+ */
+static inline struct nfqnl_queue_entry *
+__find_entry(struct nfqnl_instance *queue, nfqnl_cmpfn cmpfn, 
+                  unsigned long data)
+{
+       struct list_head *p;
+
+       list_for_each_prev(p, &queue->queue_list) {
+               struct nfqnl_queue_entry *entry = (struct nfqnl_queue_entry *)p;
+               
+               if (!cmpfn || cmpfn(entry, data))
+                       return entry;
+       }
+       return NULL;
+}
+
+static inline void
+__dequeue_entry(struct nfqnl_instance *q, struct nfqnl_queue_entry *entry)
+{
+       list_del(&entry->list);
+       q->queue_total--;
+}
+
+static inline struct nfqnl_queue_entry *
+__find_dequeue_entry(struct nfqnl_instance *queue,
+                    nfqnl_cmpfn cmpfn, unsigned long data)
+{
+       struct nfqnl_queue_entry *entry;
+
+       entry = __find_entry(queue, cmpfn, data);
+       if (entry == NULL)
+               return NULL;
+
+       __dequeue_entry(queue, entry);
+       return entry;
+}
+
+
+static inline void
+__nfqnl_flush(struct nfqnl_instance *queue, int verdict)
+{
+       struct nfqnl_queue_entry *entry;
+       
+       while ((entry = __find_dequeue_entry(queue, NULL, 0)))
+               issue_verdict(entry, verdict);
+}
+
+static inline int
+__nfqnl_set_mode(struct nfqnl_instance *queue,
+                unsigned char mode, unsigned int range)
+{
+       int status = 0;
+       
+       switch (mode) {
+       case NFQNL_COPY_NONE:
+       case NFQNL_COPY_META:
+               queue->copy_mode = mode;
+               queue->copy_range = 0;
+               break;
+               
+       case NFQNL_COPY_PACKET:
+               queue->copy_mode = mode;
+               /* we're using struct nfattr which has 16bit nfa_len */
+               if (range > 0xffff)
+                       queue->copy_range = 0xffff;
+               else
+                       queue->copy_range = range;
+               break;
+               
+       default:
+               status = -EINVAL;
+
+       }
+       return status;
+}
+
+static struct nfqnl_queue_entry *
+find_dequeue_entry(struct nfqnl_instance *queue,
+                        nfqnl_cmpfn cmpfn, unsigned long data)
+{
+       struct nfqnl_queue_entry *entry;
+       
+       spin_lock_bh(&queue->lock);
+       entry = __find_dequeue_entry(queue, cmpfn, data);
+       spin_unlock_bh(&queue->lock);
+
+       return entry;
+}
+
+static void
+nfqnl_flush(struct nfqnl_instance *queue, int verdict)
+{
+       spin_lock_bh(&queue->lock);
+       __nfqnl_flush(queue, verdict);
+       spin_unlock_bh(&queue->lock);
+}
+
+static struct sk_buff *
+nfqnl_build_packet_message(struct nfqnl_instance *queue,
+                          struct nfqnl_queue_entry *entry, int *errp)
+{
+       unsigned char *old_tail;
+       size_t size;
+       size_t data_len = 0;
+       struct sk_buff *skb;
+       struct nfqnl_msg_packet_hdr pmsg;
+       struct nlmsghdr *nlh;
+       struct nfgenmsg *nfmsg;
+       unsigned int tmp_uint;
+
+       QDEBUG("entered\n");
+
+       /* all macros expand to constant values at compile time */
+       size =    NLMSG_SPACE(sizeof(struct nfqnl_msg_packet_hdr))
+               + NLMSG_SPACE(sizeof(u_int32_t))        /* ifindex */
+               + NLMSG_SPACE(sizeof(u_int32_t))        /* ifindex */
+               + NLMSG_SPACE(sizeof(u_int32_t))        /* mark */
+               + NLMSG_SPACE(sizeof(struct nfqnl_msg_packet_hw))
+               + NLMSG_SPACE(sizeof(struct nfqnl_msg_packet_timestamp));
+
+       spin_lock_bh(&queue->lock);
+       
+       switch (queue->copy_mode) {
+       case NFQNL_COPY_META:
+       case NFQNL_COPY_NONE:
+               data_len = 0;
+               break;
+       
+       case NFQNL_COPY_PACKET:
+               if (queue->copy_range == 0 
+                   || queue->copy_range > entry->skb->len)
+                       data_len = entry->skb->len;
+               else
+                       data_len = queue->copy_range;
+               
+               size += NLMSG_SPACE(data_len);
+               break;
+       
+       default:
+               *errp = -EINVAL;
+               spin_unlock_bh(&queue->lock);
+               return NULL;
+       }
+
+       spin_unlock_bh(&queue->lock);
+
+       skb = alloc_skb(size, GFP_ATOMIC);
+       if (!skb)
+               goto nlmsg_failure;
+               
+       old_tail= skb->tail;
+       nlh = NLMSG_PUT(skb, 0, 0, 
+                       NFNL_SUBSYS_QUEUE << 8 | NFQNL_MSG_PACKET,
+                       sizeof(struct nfgenmsg));
+       nfmsg = NLMSG_DATA(nlh);
+       nfmsg->nfgen_family = entry->info->pf;
+       nfmsg->version = NFNETLINK_V0;
+       nfmsg->res_id = htons(queue->queue_num);
+
+       pmsg.packet_id          = htonl(entry->id);
+       pmsg.hw_protocol        = htons(entry->skb->protocol);
+       pmsg.hook               = entry->info->hook;
+
+       NFA_PUT(skb, NFQA_PACKET_HDR, sizeof(pmsg), &pmsg);
+
+       if (entry->info->indev) {
+               tmp_uint = htonl(entry->info->indev->ifindex);
+               NFA_PUT(skb, NFQA_IFINDEX_INDEV, sizeof(tmp_uint), &tmp_uint);
+       }
+
+       if (entry->info->outdev) {
+               tmp_uint = htonl(entry->info->outdev->ifindex);
+               NFA_PUT(skb, NFQA_IFINDEX_OUTDEV, sizeof(tmp_uint), &tmp_uint);
+       }
+
+       if (entry->skb->nfmark) {
+               tmp_uint = htonl(entry->skb->nfmark);
+               NFA_PUT(skb, NFQA_MARK, sizeof(u_int32_t), &tmp_uint);
+       }
+
+       if (entry->info->indev && entry->skb->dev
+           && entry->skb->dev->hard_header_parse) {
+               struct nfqnl_msg_packet_hw phw;
+
+               phw.hw_addrlen =
+                       entry->skb->dev->hard_header_parse(entry->skb,
+                                                          phw.hw_addr);
+               phw.hw_addrlen = htons(phw.hw_addrlen);
+               NFA_PUT(skb, NFQA_HWADDR, sizeof(phw), &phw);
+       }
+
+       if (entry->skb->stamp.tv_sec) {
+               struct nfqnl_msg_packet_timestamp ts;
+
+               ts.sec = htonll(entry->skb->stamp.tv_sec);
+               ts.usec = htonll(entry->skb->stamp.tv_usec);
+
+               NFA_PUT(skb, NFQA_TIMESTAMP, sizeof(ts), &ts);
+       }
+
+       if (data_len) {
+               struct nfattr *nfa;
+               int size = NFA_LENGTH(data_len);
+
+               if (skb_tailroom(skb) < (int)NFA_SPACE(data_len)) {
+                       printk(KERN_WARNING "nf_queue: no tailroom!\n");
+                       goto nlmsg_failure;
+               }
+
+               nfa = (struct nfattr *)skb_put(skb, NFA_ALIGN(size));
+               nfa->nfa_type = NFQA_PAYLOAD;
+               nfa->nfa_len = size;
+
+               if (skb_copy_bits(entry->skb, 0, NFA_DATA(nfa), data_len))
+                       BUG();
+       }
+               
+       nlh->nlmsg_len = skb->tail - old_tail;
+       return skb;
+
+nlmsg_failure:
+nfattr_failure:
+       if (skb)
+               kfree_skb(skb);
+       *errp = -EINVAL;
+       if (net_ratelimit())
+               printk(KERN_ERR "nf_queue: error creating packet message\n");
+       return NULL;
+}
+
+static int
+nfqnl_enqueue_packet(struct sk_buff *skb, struct nf_info *info, 
+                    unsigned int queuenum, void *data)
+{
+       int status = -EINVAL;
+       struct sk_buff *nskb;
+       struct nfqnl_instance *queue;
+       struct nfqnl_queue_entry *entry;
+
+       QDEBUG("entered\n");
+
+       queue = instance_lookup(queuenum);
+       if (!queue) {
+               QDEBUG("no queue instance matching\n");
+               return -EINVAL;
+       }
+
+       if (queue->copy_mode == NFQNL_COPY_NONE) {
+               QDEBUG("mode COPY_NONE, aborting\n");
+               return -EAGAIN;
+       }
+
+       entry = kmalloc(sizeof(*entry), GFP_ATOMIC);
+       if (entry == NULL) {
+               if (net_ratelimit())
+                       printk(KERN_ERR 
+                               "nf_queue: OOM in nfqnl_enqueue_packet()\n");
+               return -ENOMEM;
+       }
+
+       entry->info = info;
+       entry->skb = skb;
+       entry->id = atomic_inc_return(&queue->id_sequence);
+
+       nskb = nfqnl_build_packet_message(queue, entry, &status);
+       if (nskb == NULL)
+               goto err_out_free;
+               
+       spin_lock_bh(&queue->lock);
+       
+       if (!queue->peer_pid)
+               goto err_out_free_nskb; 
+
+       if (queue->queue_total >= queue->queue_maxlen) {
+                queue->queue_dropped++;
+               status = -ENOSPC;
+               if (net_ratelimit())
+                         printk(KERN_WARNING "ip_queue: full at %d entries, "
+                                "dropping packets(s). Dropped: %d\n", 
+                                queue->queue_total, queue->queue_dropped);
+               goto err_out_free_nskb;
+       }
+
+       /* nfnetlink_unicast will either free the nskb or add it to a socket */
+       status = nfnetlink_unicast(nskb, queue->peer_pid, MSG_DONTWAIT);
+       if (status < 0) {
+               queue->queue_user_dropped++;
+               goto err_out_unlock;
+       }
+
+       __enqueue_entry(queue, entry);
+
+       spin_unlock_bh(&queue->lock);
+       return status;
+
+err_out_free_nskb:
+       kfree_skb(nskb); 
+       
+err_out_unlock:
+       spin_unlock_bh(&queue->lock);
+
+err_out_free:
+       kfree(entry);
+       return status;
+}
+
+static int
+nfqnl_mangle(void *data, int data_len, struct nfqnl_queue_entry *e)
+{
+       int diff;
+
+       diff = data_len - e->skb->len;
+       if (diff < 0)
+               skb_trim(e->skb, data_len);
+       else if (diff > 0) {
+               if (data_len > 0xFFFF)
+                       return -EINVAL;
+               if (diff > skb_tailroom(e->skb)) {
+                       struct sk_buff *newskb;
+                       
+                       newskb = skb_copy_expand(e->skb,
+                                                skb_headroom(e->skb),
+                                                diff,
+                                                GFP_ATOMIC);
+                       if (newskb == NULL) {
+                               printk(KERN_WARNING "ip_queue: OOM "
+                                     "in mangle, dropping packet\n");
+                               return -ENOMEM;
+                       }
+                       if (e->skb->sk)
+                               skb_set_owner_w(newskb, e->skb->sk);
+                       kfree_skb(e->skb);
+                       e->skb = newskb;
+               }
+               skb_put(e->skb, diff);
+       }
+       if (!skb_make_writable(&e->skb, data_len))
+               return -ENOMEM;
+       memcpy(e->skb->data, data, data_len);
+
+       return 0;
+}
+
+static inline int
+id_cmp(struct nfqnl_queue_entry *e, unsigned long id)
+{
+       return (id == e->id);
+}
+
+static int
+nfqnl_set_mode(struct nfqnl_instance *queue,
+              unsigned char mode, unsigned int range)
+{
+       int status;
+
+       spin_lock_bh(&queue->lock);
+       status = __nfqnl_set_mode(queue, mode, range);
+       spin_unlock_bh(&queue->lock);
+
+       return status;
+}
+
+static int
+dev_cmp(struct nfqnl_queue_entry *entry, unsigned long ifindex)
+{
+       if (entry->info->indev)
+               if (entry->info->indev->ifindex == ifindex)
+                       return 1;
+                       
+       if (entry->info->outdev)
+               if (entry->info->outdev->ifindex == ifindex)
+                       return 1;
+
+       return 0;
+}
+
+/* drop all packets with either indev or outdev == ifindex from all queue
+ * instances */
+static void
+nfqnl_dev_drop(int ifindex)
+{
+       int i;
+       
+       QDEBUG("entering for ifindex %u\n", ifindex);
+
+       /* this only looks like we have to hold the readlock for a way too long
+        * time, issue_verdict(),  nf_reinject(), ... - but we always only
+        * issue NF_DROP, which is processed directly in nf_reinject() */
+       read_lock_bh(&instances_lock);
+
+       for  (i = 0; i < INSTANCE_BUCKETS; i++) {
+               struct hlist_node *tmp;
+               struct nfqnl_instance *inst;
+               struct hlist_head *head = &instance_table[i];
+
+               hlist_for_each_entry(inst, tmp, head, hlist) {
+                       struct nfqnl_queue_entry *entry;
+                       while ((entry = find_dequeue_entry(inst, dev_cmp, 
+                                                          ifindex)) != NULL)
+                               issue_verdict(entry, NF_DROP);
+               }
+       }
+
+       read_unlock_bh(&instances_lock);
+}
+
+#define RCV_SKB_FAIL(err) do { netlink_ack(skb, nlh, (err)); return; } while (0)
+
+static int
+nfqnl_rcv_dev_event(struct notifier_block *this,
+                   unsigned long event, void *ptr)
+{
+       struct net_device *dev = ptr;
+
+       /* Drop any packets associated with the downed device */
+       if (event == NETDEV_DOWN)
+               nfqnl_dev_drop(dev->ifindex);
+       return NOTIFY_DONE;
+}
+
+static struct notifier_block nfqnl_dev_notifier = {
+       .notifier_call  = nfqnl_rcv_dev_event,
+};
+
+static int
+nfqnl_rcv_nl_event(struct notifier_block *this,
+                  unsigned long event, void *ptr)
+{
+       struct netlink_notify *n = ptr;
+
+       if (event == NETLINK_URELEASE &&
+           n->protocol == NETLINK_NETFILTER && n->pid) {
+               int i;
+
+               /* destroy all instances for this pid */
+               write_lock_bh(&instances_lock);
+               for  (i = 0; i < INSTANCE_BUCKETS; i++) {
+                       struct hlist_node *tmp, *t2;
+                       struct nfqnl_instance *inst;
+                       struct hlist_head *head = &instance_table[i];
+
+                       hlist_for_each_entry_safe(inst, tmp, t2, head, hlist) {
+                               if (n->pid == inst->peer_pid)
+                                       __instance_destroy(inst);
+                       }
+               }
+               write_unlock_bh(&instances_lock);
+       }
+       return NOTIFY_DONE;
+}
+
+static struct notifier_block nfqnl_rtnl_notifier = {
+       .notifier_call  = nfqnl_rcv_nl_event,
+};
+
+static int
+nfqnl_recv_verdict(struct sock *ctnl, struct sk_buff *skb,
+                  struct nlmsghdr *nlh, struct nfattr *nfqa[], int *errp)
+{
+       struct nfgenmsg *nfmsg = NLMSG_DATA(nlh);
+       u_int16_t queue_num = ntohs(nfmsg->res_id);
+
+       struct nfqnl_msg_verdict_hdr *vhdr;
+       struct nfqnl_instance *queue;
+       unsigned int verdict;
+       struct nfqnl_queue_entry *entry;
+
+       queue = instance_lookup(queue_num);
+       if (!queue)
+               return -ENODEV;
+
+       if (queue->peer_pid != NETLINK_CB(skb).pid)
+               return -EPERM;
+
+       if (!nfqa[NFQA_VERDICT_HDR-1])
+               return -EINVAL;
+
+       vhdr = NFA_DATA(nfqa[NFQA_VERDICT_HDR-1]);
+       verdict = ntohl(vhdr->verdict);
+
+       if ((verdict & NF_VERDICT_MASK) > NF_MAX_VERDICT)
+               return -EINVAL;
+
+       entry = find_dequeue_entry(queue, id_cmp, ntohl(vhdr->id));
+       if (entry == NULL)
+               return -ENOENT;
+
+       if (nfqa[NFQA_PAYLOAD-1]) {
+               if (nfqnl_mangle(NFA_DATA(nfqa[NFQA_PAYLOAD-1]),
+                                NFA_PAYLOAD(nfqa[NFQA_PAYLOAD-1]), entry) < 0)
+                       verdict = NF_DROP;
+       }
+
+       if (nfqa[NFQA_MARK-1])
+               skb->nfmark = ntohl(*(u_int32_t *)NFA_DATA(nfqa[NFQA_MARK-1]));
+               
+       issue_verdict(entry, verdict);
+       return 0;
+}
+
+static int
+nfqnl_recv_unsupp(struct sock *ctnl, struct sk_buff *skb,
+                 struct nlmsghdr *nlh, struct nfattr *nfqa[], int *errp)
+{
+       return -ENOTSUPP;
+}
+
+static int
+nfqnl_recv_config(struct sock *ctnl, struct sk_buff *skb,
+                 struct nlmsghdr *nlh, struct nfattr *nfqa[], int *errp)
+{
+       struct nfgenmsg *nfmsg = NLMSG_DATA(nlh);
+       u_int16_t queue_num = ntohs(nfmsg->res_id);
+       struct nfqnl_instance *queue;
+
+       QDEBUG("entering for msg %u\n", NFNL_MSG_TYPE(nlh->nlmsg_type));
+
+       queue = instance_lookup(queue_num);
+       if (nfqa[NFQA_CFG_CMD-1]) {
+               struct nfqnl_msg_config_cmd *cmd;
+               cmd = NFA_DATA(nfqa[NFQA_CFG_CMD-1]);
+               QDEBUG("found CFG_CMD\n");
+
+               switch (cmd->command) {
+               case NFQNL_CFG_CMD_BIND:
+                       if (queue)
+                               return -EBUSY;
+
+                       queue = instance_create(queue_num, NETLINK_CB(skb).pid);
+                       if (!queue)
+                               return -EINVAL;
+                       break;
+               case NFQNL_CFG_CMD_UNBIND:
+                       if (!queue)
+                               return -ENODEV;
+
+                       if (queue->peer_pid != NETLINK_CB(skb).pid)
+                               return -EPERM;
+
+                       instance_destroy(queue);
+                       break;
+               case NFQNL_CFG_CMD_PF_BIND:
+                       QDEBUG("registering queue handler for pf=%u\n",
+                               ntohs(cmd->pf));
+                       return nf_register_queue_handler(ntohs(cmd->pf),
+                                                        nfqnl_enqueue_packet,
+                                                        NULL);
+
+                       break;
+               case NFQNL_CFG_CMD_PF_UNBIND:
+                       QDEBUG("unregistering queue handler for pf=%u\n",
+                               ntohs(cmd->pf));
+                       /* This is a bug and a feature.  We can unregister
+                        * other handlers(!) */
+                       return nf_unregister_queue_handler(ntohs(cmd->pf));
+                       break;
+               default:
+                       return -EINVAL;
+               }
+       } else {
+               if (!queue) {
+                       QDEBUG("no config command, and no instance ENOENT\n");
+                       return -ENOENT;
+               }
+
+               if (queue->peer_pid != NETLINK_CB(skb).pid) {
+                       QDEBUG("no config command, and wrong pid\n");
+                       return -EPERM;
+               }
+       }
+
+       if (nfqa[NFQA_CFG_PARAMS-1]) {
+               struct nfqnl_msg_config_params *params;
+               params = NFA_DATA(nfqa[NFQA_CFG_PARAMS-1]);
+
+               nfqnl_set_mode(queue, params->copy_mode,
+                               ntohl(params->copy_range));
+       }
+
+       return 0;
+}
+
+static struct nfnl_callback nfqnl_cb[NFQNL_MSG_MAX] = {
+       [NFQNL_MSG_PACKET]      = { .call = nfqnl_recv_unsupp,
+                                   .cap_required = CAP_NET_ADMIN },
+       [NFQNL_MSG_VERDICT]     = { .call = nfqnl_recv_verdict,
+                                   .cap_required = CAP_NET_ADMIN },
+       [NFQNL_MSG_CONFIG]      = { .call = nfqnl_recv_config,
+                                   .cap_required = CAP_NET_ADMIN },
+};
+
+static struct nfnetlink_subsystem nfqnl_subsys = {
+       .name           = "nf_queue",
+       .subsys_id      = NFNL_SUBSYS_QUEUE,
+       .cb_count       = NFQNL_MSG_MAX,
+       .attr_count     = NFQA_MAX,
+       .cb             = nfqnl_cb,
+};
+
+static int
+init_or_cleanup(int init)
+{
+       int status = -ENOMEM;
+       
+       if (!init)
+               goto cleanup;
+
+       netlink_register_notifier(&nfqnl_rtnl_notifier);
+       status = nfnetlink_subsys_register(&nfqnl_subsys);
+       if (status < 0) {
+               printk(KERN_ERR "nf_queue: failed to create netlink socket\n");
+               goto cleanup_netlink_notifier;
+       }
+
+       register_netdevice_notifier(&nfqnl_dev_notifier);
+       return status;
+
+cleanup:
+       nf_unregister_queue_handlers(nfqnl_enqueue_packet);
+       unregister_netdevice_notifier(&nfqnl_dev_notifier);
+       nfnetlink_subsys_unregister(&nfqnl_subsys);
+       
+cleanup_netlink_notifier:
+       netlink_unregister_notifier(&nfqnl_rtnl_notifier);
+       return status;
+}
+
+static int __init init(void)
+{
+       
+       return init_or_cleanup(1);
+}
+
+static void __exit fini(void)
+{
+       init_or_cleanup(0);
+}
+
+MODULE_DESCRIPTION("netfilter packet queue handler");
+MODULE_AUTHOR("Harald Welte <laforge@netfilter.org>");
+MODULE_LICENSE("GPL");
+MODULE_ALIAS_NFNL_SUBSYS(NFNL_SUBSYS_QUEUE);
+
+module_init(init);
+module_exit(fini);