[openwrt.git] / target / linux / generic-2.6 / patches-2.6.28 / 150-netfilter_imq.patch

--- /dev/null
+++ b/drivers/net/imq.c
@@ -0,0 +1,567 @@
+/*
+ *             Pseudo-driver for the intermediate queue device.
+ *
+ *             This program is free software; you can redistribute it and/or
+ *             modify it under the terms of the GNU General Public License
+ *             as published by the Free Software Foundation; either version
+ *             2 of the License, or (at your option) any later version.
+ *
+ * Authors:    Patrick McHardy, <kaber@trash.net>
+ *
+ *            The first version was written by Martin Devera, <devik@cdi.cz>
+ *
+ * Credits:    Jan Rafaj <imq2t@cedric.vabo.cz>
+ *              - Update patch to 2.4.21
+ *             Sebastian Strollo <sstrollo@nortelnetworks.com>
+ *              - Fix "Dead-loop on netdevice imq"-issue
+ *             Marcel Sebek <sebek64@post.cz>
+ *              - Update to 2.6.2-rc1
+ *
+ *            After some time of inactivity there is a group taking care
+ *            of IMQ again: http://www.linuximq.net
+ *
+ *
+ *            2004/06/30 - New version of IMQ patch to kernels <=2.6.7
+ *             including the following changes:
+ *
+ *            - Correction of ipv6 support "+"s issue (Hasso Tepper)
+ *            - Correction of imq_init_devs() issue that resulted in
+ *            kernel OOPS unloading IMQ as module (Norbert Buchmuller)
+ *            - Addition of functionality to choose number of IMQ devices
+ *            during kernel config (Andre Correa)
+ *            - Addition of functionality to choose how IMQ hooks on
+ *            PRE and POSTROUTING (after or before NAT) (Andre Correa)
+ *            - Cosmetic corrections (Norbert Buchmuller) (Andre Correa)
+ *
+ *
+ *             2005/12/16 - IMQ versions between 2.6.7 and 2.6.13 were
+ *             released with almost no problems. 2.6.14-x was released
+ *             with some important changes: nfcache was removed; After
+ *             some weeks of trouble we figured out that some IMQ fields
+ *             in skb were missing in skbuff.c - skb_clone and copy_skb_header.
+ *             These functions are correctly patched by this new patch version.
+ *
+ *             Thanks for all who helped to figure out all the problems with
+ *             2.6.14.x: Patrick McHardy, Rune Kock, VeNoMouS, Max CtRiX,
+ *             Kevin Shanahan, Richard Lucassen, Valery Dachev (hopefully
+ *             I didn't forget anybody). I apologize again for my lack of time.
+ *
+ *
+ *             2008/06/17 - 2.6.25 - Changed imq.c to use qdisc_run() instead
+ *             of qdisc_restart() and moved qdisc_run() to tasklet to avoid
+ *             recursive locking. New initialization routines to fix 'rmmod' not
+ *             working anymore. Used code from ifb.c. (Jussi Kivilinna)
+ *
+ *             2008/08/06 - 2.6.26 - (JK)
+ *              - Replaced tasklet with 'netif_schedule()'.
+ *              - Cleaned up and added comments for imq_nf_queue().
+ *
+ *             2009/04/12
+ *              - Add skb_save_cb/skb_restore_cb helper functions for backuping
+ *                control buffer. This is needed because qdisc-layer on kernels
+ *                2.6.27 and newer overwrite control buffer. (Jussi Kivilinna)
+ *              - Add better locking for IMQ device. Hopefully this will solve
+ *                SMP issues. (Jussi Kivilinna)
+ *              - Port to 2.6.27
+ *              - Port to 2.6.28
+ *
+ *             2009/04/20 - (Jussi Kivilinna)
+ *              - Fix rmmod not working
+ *              - Use netdevice feature flags to avoid extra packet handling
+ *                by core networking layer and possibly increase performance.
+ *
+ *            Also, many thanks to pablo Sebastian Greco for making the initial
+ *            patch and to those who helped the testing.
+ *
+ *             More info at: http://www.linuximq.net/ (Andre Correa)
+ */
+
+#include <linux/module.h>
+#include <linux/kernel.h>
+#include <linux/moduleparam.h>
+#include <linux/list.h>
+#include <linux/skbuff.h>
+#include <linux/netdevice.h>
+#include <linux/etherdevice.h>
+#include <linux/rtnetlink.h>
+#include <linux/if_arp.h>
+#include <linux/netfilter.h>
+#include <linux/netfilter_ipv4.h>
+#if defined(CONFIG_IPV6) || defined(CONFIG_IPV6_MODULE)
+       #include <linux/netfilter_ipv6.h>
+#endif
+#include <linux/imq.h>
+#include <net/pkt_sched.h>
+#include <net/netfilter/nf_queue.h>
+
+static nf_hookfn imq_nf_hook;
+
+static struct nf_hook_ops imq_ingress_ipv4 = {
+       .hook           = imq_nf_hook,
+       .owner          = THIS_MODULE,
+       .pf             = PF_INET,
+       .hooknum        = NF_INET_PRE_ROUTING,
+#if defined(CONFIG_IMQ_BEHAVIOR_BA) || defined(CONFIG_IMQ_BEHAVIOR_BB)
+       .priority       = NF_IP_PRI_MANGLE + 1
+#else
+       .priority       = NF_IP_PRI_NAT_DST + 1
+#endif
+};
+
+static struct nf_hook_ops imq_egress_ipv4 = {
+       .hook           = imq_nf_hook,
+       .owner          = THIS_MODULE,
+       .pf             = PF_INET,
+       .hooknum        = NF_INET_POST_ROUTING,
+#if defined(CONFIG_IMQ_BEHAVIOR_AA) || defined(CONFIG_IMQ_BEHAVIOR_BA)
+       .priority       = NF_IP_PRI_LAST
+#else
+       .priority       = NF_IP_PRI_NAT_SRC - 1
+#endif
+};
+
+#if defined(CONFIG_IPV6) || defined(CONFIG_IPV6_MODULE)
+static struct nf_hook_ops imq_ingress_ipv6 = {
+       .hook           = imq_nf_hook,
+       .owner          = THIS_MODULE,
+       .pf             = PF_INET6,
+       .hooknum        = NF_INET_PRE_ROUTING,
+#if defined(CONFIG_IMQ_BEHAVIOR_BA) || defined(CONFIG_IMQ_BEHAVIOR_BB)
+       .priority       = NF_IP6_PRI_MANGLE + 1
+#else
+       .priority       = NF_IP6_PRI_NAT_DST + 1
+#endif
+};
+
+static struct nf_hook_ops imq_egress_ipv6 = {
+       .hook           = imq_nf_hook,
+       .owner          = THIS_MODULE,
+       .pf             = PF_INET6,
+       .hooknum        = NF_INET_POST_ROUTING,
+#if defined(CONFIG_IMQ_BEHAVIOR_AA) || defined(CONFIG_IMQ_BEHAVIOR_BA)
+       .priority       = NF_IP6_PRI_LAST
+#else
+       .priority       = NF_IP6_PRI_NAT_SRC - 1
+#endif
+};
+#endif
+
+#if defined(CONFIG_IMQ_NUM_DEVS)
+static unsigned int numdevs = CONFIG_IMQ_NUM_DEVS;
+#else
+static unsigned int numdevs = IMQ_MAX_DEVS;
+#endif
+
+static DEFINE_SPINLOCK(imq_nf_queue_lock);
+
+static struct net_device *imq_devs_cache[IMQ_MAX_DEVS];
+
+
+static struct net_device_stats *imq_get_stats(struct net_device *dev)
+{
+       return &dev->stats;
+}
+
+/* called for packets kfree'd in qdiscs at places other than enqueue */
+static void imq_skb_destructor(struct sk_buff *skb)
+{
+       struct nf_queue_entry *entry = skb->nf_queue_entry;
+
+       if (entry) {
+               nf_queue_entry_release_refs(entry);
+               kfree(entry);
+       }
+
+       skb_restore_cb(skb); /* kfree backup */
+}
+
+static void imq_nf_reinject(struct nf_queue_entry *entry, unsigned int verdict)
+{
+       int status;
+
+       if (!entry->next_outfn) {
+               spin_lock_bh(&imq_nf_queue_lock);
+               nf_reinject(entry, verdict);
+               spin_unlock_bh(&imq_nf_queue_lock);
+               return;
+       }
+
+       rcu_read_lock();
+       local_bh_disable();
+       status = entry->next_outfn(entry, entry->next_queuenum);
+       local_bh_enable();
+       if (status < 0) {
+               nf_queue_entry_release_refs(entry);
+               kfree_skb(entry->skb);
+               kfree(entry);
+       }
+
+       rcu_read_unlock();
+}
+
+static int imq_dev_xmit(struct sk_buff *skb, struct net_device *dev)
+{
+       dev->stats.tx_bytes += skb->len;
+       dev->stats.tx_packets++;
+
+       skb->imq_flags = 0;
+       skb->destructor = NULL;
+
+       skb_restore_cb(skb); /* restore skb->cb */
+
+       dev->trans_start = jiffies;
+       imq_nf_reinject(skb->nf_queue_entry, NF_ACCEPT);
+       return 0;
+}
+
+static int imq_nf_queue(struct nf_queue_entry *entry, unsigned queue_num)
+{
+       struct net_device *dev;
+       struct sk_buff *skb_orig, *skb, *skb_shared;
+       struct Qdisc *q;
+       struct netdev_queue *txq;
+       int users, index;
+       int retval = -EINVAL;
+
+       index = entry->skb->imq_flags & IMQ_F_IFMASK;
+       if (unlikely(index > numdevs - 1)) {
+               if (net_ratelimit())
+                       printk(KERN_WARNING
+                              "IMQ: invalid device specified, highest is %u\n",
+                              numdevs - 1);
+               retval = -EINVAL;
+               goto out;
+       }
+
+       /* check for imq device by index from cache */
+       dev = imq_devs_cache[index];
+       if (unlikely(!dev)) {
+               char buf[8];
+
+               /* get device by name and cache result */
+               snprintf(buf, sizeof(buf), "imq%d", index);
+               dev = dev_get_by_name(&init_net, buf);
+               if (!dev) {
+                       /* not found ?!*/
+                       BUG();
+                       retval = -ENODEV;
+                       goto out;
+               }
+
+               imq_devs_cache[index] = dev;
+               dev_put(dev);
+       }
+
+       if (unlikely(!(dev->flags & IFF_UP))) {
+               entry->skb->imq_flags = 0;
+               imq_nf_reinject(entry, NF_ACCEPT);
+               retval = 0;
+               goto out;
+       }
+       dev->last_rx = jiffies;
+
+       skb = entry->skb;
+       skb_orig = NULL;
+
+       /* skb has owner? => make clone */
+       if (unlikely(skb->destructor)) {
+               skb_orig = skb;
+               skb = skb_clone(skb, GFP_ATOMIC);
+               if (!skb) {
+                       retval = -ENOMEM;
+                       goto out;
+               }
+               entry->skb = skb;
+       }
+
+       skb->nf_queue_entry = entry;
+
+       dev->stats.rx_bytes += skb->len;
+       dev->stats.rx_packets++;
+
+       txq = dev_pick_tx(dev, skb);
+
+       q = rcu_dereference(txq->qdisc);
+       if (unlikely(!q->enqueue))
+               goto packet_not_eaten_by_imq_dev;
+
+       spin_lock_bh(qdisc_lock(q));
+
+       users = atomic_read(&skb->users);
+
+       skb_shared = skb_get(skb); /* increase reference count by one */
+       skb_save_cb(skb_shared); /* backup skb->cb, as qdisc layer will
+                                       overwrite it */
+       qdisc_enqueue_root(skb_shared, q); /* might kfree_skb */
+
+       if (likely(atomic_read(&skb_shared->users) == users + 1)) {
+               kfree_skb(skb_shared); /* decrease reference count by one */
+
+               skb->destructor = &imq_skb_destructor;
+
+               /* cloned? */
+               if (skb_orig)
+                       kfree_skb(skb_orig); /* free original */
+
+               spin_unlock_bh(qdisc_lock(q));
+
+               /* schedule qdisc dequeue */
+               __netif_schedule(q);
+
+               retval = 0;
+               goto out;
+       } else {
+               skb_restore_cb(skb_shared); /* restore skb->cb */
+               /* qdisc dropped packet and decreased skb reference count of
+                * skb, so we don't really want to and try refree as that would
+                * actually destroy the skb. */
+               spin_unlock_bh(qdisc_lock(q));
+               goto packet_not_eaten_by_imq_dev;
+       }
+
+packet_not_eaten_by_imq_dev:
+       /* cloned? restore original */
+       if (skb_orig) {
+               kfree_skb(skb);
+               entry->skb = skb_orig;
+       }
+       retval = -1;
+out:
+       return retval;
+}
+
+static struct nf_queue_handler nfqh = {
+       .name  = "imq",
+       .outfn = imq_nf_queue,
+};
+
+static unsigned int imq_nf_hook(unsigned int hook, struct sk_buff *pskb,
+                               const struct net_device *indev,
+                               const struct net_device *outdev,
+                               int (*okfn)(struct sk_buff *))
+{
+       if (pskb->imq_flags & IMQ_F_ENQUEUE)
+               return NF_QUEUE;
+
+       return NF_ACCEPT;
+}
+
+static int imq_close(struct net_device *dev)
+{
+       netif_stop_queue(dev);
+       return 0;
+}
+
+static int imq_open(struct net_device *dev)
+{
+       netif_start_queue(dev);
+       return 0;
+}
+
+static void imq_setup(struct net_device *dev)
+{
+       dev->hard_start_xmit    = imq_dev_xmit;
+       dev->open               = imq_open;
+       dev->get_stats          = imq_get_stats;
+       dev->stop               = imq_close;
+       dev->type               = ARPHRD_VOID;
+       dev->mtu                = 16000;
+       dev->tx_queue_len       = 11000;
+       dev->flags              = IFF_NOARP;
+       dev->features           = NETIF_F_SG | NETIF_F_FRAGLIST |
+                                 NETIF_F_GSO | NETIF_F_HW_CSUM |
+                                 NETIF_F_HIGHDMA;
+}
+
+static int imq_validate(struct nlattr *tb[], struct nlattr *data[])
+{
+       int ret = 0;
+
+       if (tb[IFLA_ADDRESS]) {
+               if (nla_len(tb[IFLA_ADDRESS]) != ETH_ALEN) {
+                       ret = -EINVAL;
+                       goto end;
+               }
+               if (!is_valid_ether_addr(nla_data(tb[IFLA_ADDRESS]))) {
+                       ret = -EADDRNOTAVAIL;
+                       goto end;
+               }
+       }
+       return 0;
+end:
+       printk(KERN_WARNING "IMQ: imq_validate failed (%d)\n", ret);
+       return ret;
+}
+
+static struct rtnl_link_ops imq_link_ops __read_mostly = {
+       .kind           = "imq",
+       .priv_size      = 0,
+       .setup          = imq_setup,
+       .validate       = imq_validate,
+};
+
+static int __init imq_init_hooks(void)
+{
+       int err;
+
+       nf_register_queue_imq_handler(&nfqh);
+
+       err = nf_register_hook(&imq_ingress_ipv4);
+       if (err)
+               goto err1;
+
+       err = nf_register_hook(&imq_egress_ipv4);
+       if (err)
+               goto err2;
+
+#if defined(CONFIG_IPV6) || defined(CONFIG_IPV6_MODULE)
+       err = nf_register_hook(&imq_ingress_ipv6);
+       if (err)
+               goto err3;
+
+       err = nf_register_hook(&imq_egress_ipv6);
+       if (err)
+               goto err4;
+#endif
+
+       return 0;
+
+#if defined(CONFIG_IPV6) || defined(CONFIG_IPV6_MODULE)
+err4:
+       nf_unregister_hook(&imq_ingress_ipv6);
+err3:
+       nf_unregister_hook(&imq_egress_ipv4);
+#endif
+err2:
+       nf_unregister_hook(&imq_ingress_ipv4);
+err1:
+       nf_unregister_queue_imq_handler();
+       return err;
+}
+
+static int __init imq_init_one(int index)
+{
+       struct net_device *dev;
+       int ret;
+
+       dev = alloc_netdev(0, "imq%d", imq_setup);
+       if (!dev)
+               return -ENOMEM;
+
+       ret = dev_alloc_name(dev, dev->name);
+       if (ret < 0)
+               goto fail;
+
+       dev->rtnl_link_ops = &imq_link_ops;
+       ret = register_netdevice(dev);
+       if (ret < 0)
+               goto fail;
+
+       return 0;
+fail:
+       free_netdev(dev);
+       return ret;
+}
+
+static int __init imq_init_devs(void)
+{
+       int err, i;
+
+       if (numdevs < 1 || numdevs > IMQ_MAX_DEVS) {
+               printk(KERN_ERR "IMQ: numdevs has to be betweed 1 and %u\n",
+                      IMQ_MAX_DEVS);
+               return -EINVAL;
+       }
+
+       rtnl_lock();
+       err = __rtnl_link_register(&imq_link_ops);
+
+       for (i = 0; i < numdevs && !err; i++)
+               err = imq_init_one(i);
+
+       if (err) {
+               __rtnl_link_unregister(&imq_link_ops);
+               memset(imq_devs_cache, 0, sizeof(imq_devs_cache));
+       }
+       rtnl_unlock();
+
+       return err;
+}
+
+static int __init imq_init_module(void)
+{
+       int err;
+
+#if defined(CONFIG_IMQ_NUM_DEVS)
+       BUILD_BUG_ON(CONFIG_IMQ_NUM_DEVS > 16);
+       BUILD_BUG_ON(CONFIG_IMQ_NUM_DEVS < 2);
+       BUILD_BUG_ON(CONFIG_IMQ_NUM_DEVS - 1 > IMQ_F_IFMASK);
+#endif
+
+       err = imq_init_devs();
+       if (err) {
+               printk(KERN_ERR "IMQ: Error trying imq_init_devs(net)\n");
+               return err;
+       }
+
+       err = imq_init_hooks();
+       if (err) {
+               printk(KERN_ERR "IMQ: Error trying imq_init_hooks()\n");
+               rtnl_link_unregister(&imq_link_ops);
+               memset(imq_devs_cache, 0, sizeof(imq_devs_cache));
+               return err;
+       }
+
+       printk(KERN_INFO "IMQ driver loaded successfully.\n");
+
+#if defined(CONFIG_IMQ_BEHAVIOR_BA) || defined(CONFIG_IMQ_BEHAVIOR_BB)
+       printk(KERN_INFO "\tHooking IMQ before NAT on PREROUTING.\n");
+#else
+       printk(KERN_INFO "\tHooking IMQ after NAT on PREROUTING.\n");
+#endif
+#if defined(CONFIG_IMQ_BEHAVIOR_AB) || defined(CONFIG_IMQ_BEHAVIOR_BB)
+       printk(KERN_INFO "\tHooking IMQ before NAT on POSTROUTING.\n");
+#else
+       printk(KERN_INFO "\tHooking IMQ after NAT on POSTROUTING.\n");
+#endif
+
+       return 0;
+}
+
+static void __exit imq_unhook(void)
+{
+#if defined(CONFIG_IPV6) || defined(CONFIG_IPV6_MODULE)
+       nf_unregister_hook(&imq_ingress_ipv6);
+       nf_unregister_hook(&imq_egress_ipv6);
+#endif
+       nf_unregister_hook(&imq_ingress_ipv4);
+       nf_unregister_hook(&imq_egress_ipv4);
+
+       nf_unregister_queue_imq_handler();
+}
+
+static void __exit imq_cleanup_devs(void)
+{
+       rtnl_link_unregister(&imq_link_ops);
+       memset(imq_devs_cache, 0, sizeof(imq_devs_cache));
+}
+
+static void __exit imq_exit_module(void)
+{
+       imq_unhook();
+       imq_cleanup_devs();
+       printk(KERN_INFO "IMQ driver unloaded successfully.\n");
+}
+
+module_init(imq_init_module);
+module_exit(imq_exit_module);
+
+module_param(numdevs, int, 0);
+MODULE_PARM_DESC(numdevs, "number of IMQ devices (how many imq* devices will "
+                       "be created)");
+MODULE_AUTHOR("http://www.linuximq.net");
+MODULE_DESCRIPTION("Pseudo-driver for the intermediate queue device. See "
+                       "http://www.linuximq.net/ for more information.");
+MODULE_LICENSE("GPL");
+MODULE_ALIAS_RTNL_LINK("imq");
+
--- a/drivers/net/Kconfig
+++ b/drivers/net/Kconfig
@@ -109,6 +109,129 @@
          To compile this driver as a module, choose M here: the module
          will be called eql.  If unsure, say N.
 
+config IMQ
+       tristate "IMQ (intermediate queueing device) support"
+       depends on NETDEVICES && NETFILTER
+       ---help---
+         The IMQ device(s) is used as placeholder for QoS queueing
+         disciplines. Every packet entering/leaving the IP stack can be
+         directed through the IMQ device where it's enqueued/dequeued to the
+         attached qdisc. This allows you to treat network devices as classes
+         and distribute bandwidth among them. Iptables is used to specify
+         through which IMQ device, if any, packets travel.
+
+         More information at: http://www.linuximq.net/
+
+         To compile this driver as a module, choose M here: the module
+         will be called imq.  If unsure, say N.
+
+choice
+       prompt "IMQ behavior (PRE/POSTROUTING)"
+       depends on IMQ
+       default IMQ_BEHAVIOR_AB
+       help
+
+               This settings defines how IMQ behaves in respect to its
+               hooking in PREROUTING and POSTROUTING.
+
+               IMQ can work in any of the following ways:
+
+                   PREROUTING   |      POSTROUTING
+               -----------------|-------------------
+               #1  After NAT    |      After NAT
+               #2  After NAT    |      Before NAT
+               #3  Before NAT   |      After NAT
+               #4  Before NAT   |      Before NAT
+
+               The default behavior is to hook before NAT on PREROUTING
+               and after NAT on POSTROUTING (#3).
+
+               This settings are specially usefull when trying to use IMQ
+               to shape NATed clients.
+
+               More information can be found at: www.linuximq.net
+
+               If not sure leave the default settings alone.
+
+config IMQ_BEHAVIOR_AA
+       bool "IMQ AA"
+       help
+               This settings defines how IMQ behaves in respect to its
+               hooking in PREROUTING and POSTROUTING.
+
+               Choosing this option will make IMQ hook like this:
+
+               PREROUTING:   After NAT
+               POSTROUTING:  After NAT
+
+               More information can be found at: www.linuximq.net
+
+               If not sure leave the default settings alone.
+
+config IMQ_BEHAVIOR_AB
+       bool "IMQ AB"
+       help
+               This settings defines how IMQ behaves in respect to its
+               hooking in PREROUTING and POSTROUTING.
+
+               Choosing this option will make IMQ hook like this:
+
+               PREROUTING:   After NAT
+               POSTROUTING:  Before NAT
+
+               More information can be found at: www.linuximq.net
+
+               If not sure leave the default settings alone.
+
+config IMQ_BEHAVIOR_BA
+       bool "IMQ BA"
+       help
+               This settings defines how IMQ behaves in respect to its
+               hooking in PREROUTING and POSTROUTING.
+
+               Choosing this option will make IMQ hook like this:
+
+               PREROUTING:   Before NAT
+               POSTROUTING:  After NAT
+
+               More information can be found at: www.linuximq.net
+
+               If not sure leave the default settings alone.
+
+config IMQ_BEHAVIOR_BB
+       bool "IMQ BB"
+       help
+               This settings defines how IMQ behaves in respect to its
+               hooking in PREROUTING and POSTROUTING.
+
+               Choosing this option will make IMQ hook like this:
+
+               PREROUTING:   Before NAT
+               POSTROUTING:  Before NAT
+
+               More information can be found at: www.linuximq.net
+
+               If not sure leave the default settings alone.
+
+endchoice
+
+config IMQ_NUM_DEVS
+
+       int "Number of IMQ devices"
+       range 2 16
+       depends on IMQ
+       default "16"
+       help
+
+               This settings defines how many IMQ devices will be
+               created.
+
+               The default value is 16.
+
+               More information can be found at: www.linuximq.net
+
+               If not sure leave the default settings alone.
+
 config TUN
        tristate "Universal TUN/TAP device driver support"
        select CRC32
--- a/drivers/net/Makefile
+++ b/drivers/net/Makefile
@@ -148,6 +148,7 @@
 obj-$(CONFIG_XEN_NETDEV_FRONTEND) += xen-netfront.o
 
 obj-$(CONFIG_DUMMY) += dummy.o
+obj-$(CONFIG_IMQ) += imq.o
 obj-$(CONFIG_IFB) += ifb.o
 obj-$(CONFIG_MACVLAN) += macvlan.o
 obj-$(CONFIG_DE600) += de600.o
--- /dev/null
+++ b/include/linux/imq.h
@@ -0,0 +1,13 @@
+#ifndef _IMQ_H
+#define _IMQ_H
+
+/* IFMASK (16 device indexes, 0 to 15) and flag(s) fit in 5 bits */
+#define IMQ_F_BITS     5
+
+#define IMQ_F_IFMASK   0x0f
+#define IMQ_F_ENQUEUE  0x10
+
+#define IMQ_MAX_DEVS   (IMQ_F_IFMASK + 1)
+
+#endif /* _IMQ_H */
+
--- /dev/null
+++ b/include/linux/netfilter_ipv4/ipt_IMQ.h
@@ -0,0 +1,10 @@
+#ifndef _IPT_IMQ_H
+#define _IPT_IMQ_H
+
+/* Backwards compatibility for old userspace */
+#include <linux/netfilter/xt_IMQ.h>
+
+#define ipt_imq_info xt_imq_info
+
+#endif /* _IPT_IMQ_H */
+
--- /dev/null
+++ b/include/linux/netfilter_ipv6/ip6t_IMQ.h
@@ -0,0 +1,10 @@
+#ifndef _IP6T_IMQ_H
+#define _IP6T_IMQ_H
+
+/* Backwards compatibility for old userspace */
+#include <linux/netfilter/xt_IMQ.h>
+
+#define ip6t_imq_info xt_imq_info
+
+#endif /* _IP6T_IMQ_H */
+
--- a/include/linux/skbuff.h
+++ b/include/linux/skbuff.h
@@ -28,6 +28,9 @@
 #include <linux/rcupdate.h>
 #include <linux/dmaengine.h>
 #include <linux/hrtimer.h>
+#if defined(CONFIG_IMQ) || defined(CONFIG_IMQ_MODULE)
+#include <linux/imq.h>
+#endif
 
 #define HAVE_ALLOC_SKB         /* For the drivers to know */
 #define HAVE_ALIGNABLE_SKB     /* Ditto 8)                */
@@ -278,6 +281,9 @@
         * first. This is owned by whoever has the skb queued ATM.
         */
        char                    cb[48];
+#if defined(CONFIG_IMQ) || defined(CONFIG_IMQ_MODULE)
+       void                    *cb_next;
+#endif
 
        unsigned int            len,
                                data_len;
@@ -308,6 +314,9 @@
        struct nf_conntrack     *nfct;
        struct sk_buff          *nfct_reasm;
 #endif
+#if defined(CONFIG_IMQ) || defined(CONFIG_IMQ_MODULE)
+       struct nf_queue_entry   *nf_queue_entry;
+#endif
 #ifdef CONFIG_BRIDGE_NETFILTER
        struct nf_bridge_info   *nf_bridge;
 #endif
@@ -327,6 +336,9 @@
        __u8                    do_not_encrypt:1;
 #endif
        /* 0/13/14 bit hole */
+#if defined(CONFIG_IMQ) || defined(CONFIG_IMQ_MODULE)
+       __u8                    imq_flags:IMQ_F_BITS;
+#endif
 
 #ifdef CONFIG_NET_DMA
        dma_cookie_t            dma_cookie;
@@ -367,6 +379,12 @@
                          enum dma_data_direction dir);
 #endif
 
+
+#if defined(CONFIG_IMQ) || defined(CONFIG_IMQ_MODULE)
+extern int skb_save_cb(struct sk_buff *skb);
+extern int skb_restore_cb(struct sk_buff *skb);
+#endif
+
 extern void kfree_skb(struct sk_buff *skb);
 extern void           __kfree_skb(struct sk_buff *skb);
 extern struct sk_buff *__alloc_skb(unsigned int size,
@@ -1804,6 +1822,10 @@
        dst->nfct_reasm = src->nfct_reasm;
        nf_conntrack_get_reasm(src->nfct_reasm);
 #endif
+#if defined(CONFIG_IMQ) || defined(CONFIG_IMQ_MODULE)
+       dst->imq_flags = src->imq_flags;
+       dst->nf_queue_entry = src->nf_queue_entry;
+#endif
 #ifdef CONFIG_BRIDGE_NETFILTER
        dst->nf_bridge  = src->nf_bridge;
        nf_bridge_get(src->nf_bridge);
--- a/net/core/dev.c
+++ b/net/core/dev.c
@@ -96,6 +96,9 @@
 #include <net/net_namespace.h>
 #include <net/sock.h>
 #include <linux/rtnetlink.h>
+#if defined(CONFIG_IMQ) || defined(CONFIG_IMQ_MODULE)
+#include <linux/imq.h>
+#endif
 #include <linux/proc_fs.h>
 #include <linux/seq_file.h>
 #include <linux/stat.h>
@@ -1655,7 +1658,11 @@
                        struct netdev_queue *txq)
 {
        if (likely(!skb->next)) {
-               if (!list_empty(&ptype_all))
+               if (!list_empty(&ptype_all)
+#if defined(CONFIG_IMQ) || defined(CONFIG_IMQ_MODULE)
+                   && !(skb->imq_flags & IMQ_F_ENQUEUE)
+#endif
+                   )
                        dev_queue_xmit_nit(skb, dev);
 
                if (netif_needs_gso(dev, skb)) {
@@ -1746,8 +1753,7 @@
        return (u16) (((u64) hash * dev->real_num_tx_queues) >> 32);
 }
 
-static struct netdev_queue *dev_pick_tx(struct net_device *dev,
-                                       struct sk_buff *skb)
+struct netdev_queue *dev_pick_tx(struct net_device *dev, struct sk_buff *skb)
 {
        u16 queue_index = 0;
 
@@ -1759,6 +1765,7 @@
        skb_set_queue_mapping(skb, queue_index);
        return netdev_get_tx_queue(dev, queue_index);
 }
+EXPORT_SYMBOL(dev_pick_tx);
 
 /**
  *     dev_queue_xmit - transmit a buffer
--- a/include/linux/netdevice.h
+++ b/include/linux/netdevice.h
@@ -950,6 +950,7 @@
 extern int             dev_open(struct net_device *dev);
 extern int             dev_close(struct net_device *dev);
 extern void            dev_disable_lro(struct net_device *dev);
+extern struct netdev_queue *dev_pick_tx(struct net_device *dev, struct sk_buff *skb);
 extern int             dev_queue_xmit(struct sk_buff *skb);
 extern int             register_netdevice(struct net_device *dev);
 extern void            unregister_netdevice(struct net_device *dev);
--- /dev/null
+++ b/include/linux/netfilter/xt_IMQ.h
@@ -0,0 +1,9 @@
+#ifndef _XT_IMQ_H
+#define _XT_IMQ_H
+
+struct xt_imq_info {
+       unsigned int todev;     /* target imq device */
+};
+
+#endif /* _XT_IMQ_H */
+
--- a/include/net/netfilter/nf_queue.h
+++ b/include/net/netfilter/nf_queue.h
@@ -13,6 +13,12 @@
        struct net_device       *indev;
        struct net_device       *outdev;
        int                     (*okfn)(struct sk_buff *);
+
+#if defined(CONFIG_IMQ) || defined(CONFIG_IMQ_MODULE)
+       int                     (*next_outfn)(struct nf_queue_entry *entry,
+                                             unsigned int queuenum);
+       unsigned int            next_queuenum;
+#endif
 };
 
 #define nf_queue_entry_reroute(x) ((void *)x + sizeof(struct nf_queue_entry))
@@ -30,5 +36,11 @@
                                       const struct nf_queue_handler *qh);
 extern void nf_unregister_queue_handlers(const struct nf_queue_handler *qh);
 extern void nf_reinject(struct nf_queue_entry *entry, unsigned int verdict);
+extern void nf_queue_entry_release_refs(struct nf_queue_entry *entry);
+
+#if defined(CONFIG_IMQ) || defined(CONFIG_IMQ_MODULE)
+extern void nf_register_queue_imq_handler(const struct nf_queue_handler *qh);
+extern void nf_unregister_queue_imq_handler(void);
+#endif
 
 #endif /* _NF_QUEUE_H */
--- a/net/core/skbuff.c
+++ b/net/core/skbuff.c
@@ -69,6 +69,9 @@
 
 static struct kmem_cache *skbuff_head_cache __read_mostly;
 static struct kmem_cache *skbuff_fclone_cache __read_mostly;
+#if defined(CONFIG_IMQ) || defined(CONFIG_IMQ_MODULE)
+static struct kmem_cache *skbuff_cb_store_cache __read_mostly;
+#endif
 
 static void sock_pipe_buf_release(struct pipe_inode_info *pipe,
                                  struct pipe_buffer *buf)
@@ -88,6 +91,80 @@
        return 1;
 }
 
+#if defined(CONFIG_IMQ) || defined(CONFIG_IMQ_MODULE)
+/* Control buffer save/restore for IMQ devices */
+struct skb_cb_table {
+       void                    *cb_next;
+       atomic_t                refcnt;
+       char                    cb[48];
+};
+
+static DEFINE_SPINLOCK(skb_cb_store_lock);
+
+int skb_save_cb(struct sk_buff *skb)
+{
+       struct skb_cb_table *next;
+
+       next = kmem_cache_alloc(skbuff_cb_store_cache, GFP_ATOMIC);
+       if (!next)
+               return -ENOMEM;
+
+       BUILD_BUG_ON(sizeof(skb->cb) != sizeof(next->cb));
+
+       memcpy(next->cb, skb->cb, sizeof(skb->cb));
+       next->cb_next = skb->cb_next;
+
+       atomic_set(&next->refcnt, 1);
+
+       skb->cb_next = next;
+       return 0;
+}
+EXPORT_SYMBOL(skb_save_cb);
+
+int skb_restore_cb(struct sk_buff *skb)
+{
+       struct skb_cb_table *next;
+
+       if (!skb->cb_next)
+               return 0;
+
+       next = skb->cb_next;
+
+       BUILD_BUG_ON(sizeof(skb->cb) != sizeof(next->cb));
+
+       memcpy(skb->cb, next->cb, sizeof(skb->cb));
+       skb->cb_next = next->cb_next;
+
+       spin_lock(&skb_cb_store_lock);
+
+       if (atomic_dec_and_test(&next->refcnt)) {
+               kmem_cache_free(skbuff_cb_store_cache, next);
+       }
+
+       spin_unlock(&skb_cb_store_lock);
+
+       return 0;
+}
+EXPORT_SYMBOL(skb_restore_cb);
+
+static void skb_copy_stored_cb(struct sk_buff *new, struct sk_buff *old)
+{
+       struct skb_cb_table *next;
+
+       if (!old->cb_next) {
+               new->cb_next = 0;
+               return;
+       }
+
+       spin_lock(&skb_cb_store_lock);
+
+       next = old->cb_next;
+       atomic_inc(&next->refcnt);
+       new->cb_next = next;
+
+       spin_unlock(&skb_cb_store_lock);
+}
+#endif
 
 /* Pipe buffer operations for a socket. */
 static struct pipe_buf_operations sock_pipe_buf_ops = {
@@ -381,6 +458,15 @@
                WARN_ON(in_irq());
                skb->destructor(skb);
        }
+#if defined(CONFIG_IMQ) || defined(CONFIG_IMQ_MODULE)
+       /* This should not happen. When it does, avoid memleak by restoring
+       the chain of cb-backups. */
+       while(skb->cb_next != NULL) {
+               printk(KERN_WARNING "kfree_skb: skb->cb_next: %08x\n",
+                       skb->cb_next);
+               skb_restore_cb(skb);
+       }
+#endif
 #if defined(CONFIG_NF_CONNTRACK) || defined(CONFIG_NF_CONNTRACK_MODULE)
        nf_conntrack_put(skb->nfct);
        nf_conntrack_put_reasm(skb->nfct_reasm);
@@ -493,6 +579,9 @@
        new->sp                 = secpath_get(old->sp);
 #endif
        memcpy(new->cb, old->cb, sizeof(old->cb));
+#if defined(CONFIG_IMQ) || defined(CONFIG_IMQ_MODULE)
+       skb_copy_stored_cb(new, old);
+#endif
        new->csum_start         = old->csum_start;
        new->csum_offset        = old->csum_offset;
        new->local_df           = old->local_df;
@@ -2397,6 +2486,13 @@
                                                0,
                                                SLAB_HWCACHE_ALIGN|SLAB_PANIC,
                                                NULL);
+#if defined(CONFIG_IMQ) || defined(CONFIG_IMQ_MODULE)
+       skbuff_cb_store_cache = kmem_cache_create("skbuff_cb_store_cache",
+                                                 sizeof(struct skb_cb_table),
+                                                 0,
+                                                 SLAB_HWCACHE_ALIGN|SLAB_PANIC,
+                                                 NULL);
+#endif
 }
 
 /**
--- a/net/netfilter/Kconfig
+++ b/net/netfilter/Kconfig
@@ -357,6 +357,18 @@
 
          To compile it as a module, choose M here.  If unsure, say N.
 
+config NETFILTER_XT_TARGET_IMQ
+        tristate '"IMQ" target support'
+       depends on NETFILTER_XTABLES
+       depends on IP_NF_MANGLE || IP6_NF_MANGLE
+       select IMQ
+       default m if NETFILTER_ADVANCED=n
+        help
+          This option adds a `IMQ' target which is used to specify if and
+          to which imq device packets should get enqueued/dequeued.
+
+          To compile it as a module, choose M here.  If unsure, say N.
+
 config NETFILTER_XT_TARGET_MARK
        tristate '"MARK" target support'
        default m if NETFILTER_ADVANCED=n
--- a/net/netfilter/Makefile
+++ b/net/netfilter/Makefile
@@ -45,6 +45,7 @@
 obj-$(CONFIG_NETFILTER_XT_TARGET_CONNMARK) += xt_CONNMARK.o
 obj-$(CONFIG_NETFILTER_XT_TARGET_CONNSECMARK) += xt_CONNSECMARK.o
 obj-$(CONFIG_NETFILTER_XT_TARGET_DSCP) += xt_DSCP.o
+obj-$(CONFIG_NETFILTER_XT_TARGET_IMQ) += xt_IMQ.o
 obj-$(CONFIG_NETFILTER_XT_TARGET_MARK) += xt_MARK.o
 obj-$(CONFIG_NETFILTER_XT_TARGET_NFLOG) += xt_NFLOG.o
 obj-$(CONFIG_NETFILTER_XT_TARGET_NFQUEUE) += xt_NFQUEUE.o
--- a/net/netfilter/nf_queue.c
+++ b/net/netfilter/nf_queue.c
@@ -20,6 +20,26 @@
 
 static DEFINE_MUTEX(queue_handler_mutex);
 
+#if defined(CONFIG_IMQ) || defined(CONFIG_IMQ_MODULE)
+static const struct nf_queue_handler *queue_imq_handler;
+
+void nf_register_queue_imq_handler(const struct nf_queue_handler *qh)
+{
+       mutex_lock(&queue_handler_mutex);
+       rcu_assign_pointer(queue_imq_handler, qh);
+       mutex_unlock(&queue_handler_mutex);
+}
+EXPORT_SYMBOL(nf_register_queue_imq_handler);
+
+void nf_unregister_queue_imq_handler(void)
+{
+       mutex_lock(&queue_handler_mutex);
+       rcu_assign_pointer(queue_imq_handler, NULL);
+       mutex_unlock(&queue_handler_mutex);
+}
+EXPORT_SYMBOL(nf_unregister_queue_imq_handler);
+#endif
+
 /* return EBUSY when somebody else is registered, return EEXIST if the
  * same handler is registered, return 0 in case of success. */
 int nf_register_queue_handler(u_int8_t pf, const struct nf_queue_handler *qh)
@@ -80,7 +100,7 @@
 }
 EXPORT_SYMBOL_GPL(nf_unregister_queue_handlers);
 
-static void nf_queue_entry_release_refs(struct nf_queue_entry *entry)
+void nf_queue_entry_release_refs(struct nf_queue_entry *entry)
 {
        /* Release those devices we held, or Alexey will kill me. */
        if (entry->indev)
@@ -100,6 +120,7 @@
        /* Drop reference to owner of hook which queued us. */
        module_put(entry->elem->owner);
 }
+EXPORT_SYMBOL_GPL(nf_queue_entry_release_refs);
 
 /*
  * Any packet that leaves via this function must come back
@@ -121,12 +142,26 @@
 #endif
        const struct nf_afinfo *afinfo;
        const struct nf_queue_handler *qh;
+#if defined(CONFIG_IMQ) || defined(CONFIG_IMQ_MODULE)
+       const struct nf_queue_handler *qih = NULL;
+#endif
 
        /* QUEUE == DROP if noone is waiting, to be safe. */
        rcu_read_lock();
 
        qh = rcu_dereference(queue_handler[pf]);
+#if defined(CONFIG_IMQ) || defined(CONFIG_IMQ_MODULE)
+#if defined(CONFIG_IPV6) || defined(CONFIG_IPV6_MODULE)
+       if (pf == PF_INET || pf == PF_INET6)
+#else
+       if (pf == PF_INET)
+#endif
+               qih = rcu_dereference(queue_imq_handler);
+
+       if (!qh && !qih)
+#else /* !IMQ */
        if (!qh)
+#endif
                goto err_unlock;
 
        afinfo = nf_get_afinfo(pf);
@@ -145,6 +180,10 @@
                .indev  = indev,
                .outdev = outdev,
                .okfn   = okfn,
+#if defined(CONFIG_IMQ) || defined(CONFIG_IMQ_MODULE)
+               .next_outfn = qh ? qh->outfn : NULL,
+               .next_queuenum = queuenum,
+#endif
        };
 
        /* If it's going away, ignore hook. */
@@ -170,8 +209,19 @@
        }
 #endif
        afinfo->saveroute(skb, entry);
+
+#if defined(CONFIG_IMQ) || defined(CONFIG_IMQ_MODULE)
+       if (qih) {
+               status = qih->outfn(entry, queuenum);
+               goto imq_skip_queue;
+       }
+#endif
+
        status = qh->outfn(entry, queuenum);
 
+#if defined(CONFIG_IMQ) || defined(CONFIG_IMQ_MODULE)
+imq_skip_queue:
+#endif
        rcu_read_unlock();
 
        if (status < 0) {
--- /dev/null
+++ b/net/netfilter/xt_IMQ.c
@@ -0,0 +1,73 @@
+/*
+ * This target marks packets to be enqueued to an imq device
+ */
+#include <linux/module.h>
+#include <linux/skbuff.h>
+#include <linux/netfilter/x_tables.h>
+#include <linux/netfilter/xt_IMQ.h>
+#include <linux/imq.h>
+
+static unsigned int imq_target(struct sk_buff *pskb,
+                               const struct xt_target_param *par)
+{
+       const struct xt_imq_info *mr = par->targinfo;
+
+       pskb->imq_flags = (mr->todev & IMQ_F_IFMASK) | IMQ_F_ENQUEUE;
+
+       return XT_CONTINUE;
+}
+
+static bool imq_checkentry(const struct xt_tgchk_param *par)
+{
+       struct xt_imq_info *mr = par->targinfo;
+
+       if (mr->todev > IMQ_MAX_DEVS - 1) {
+               printk(KERN_WARNING
+                      "IMQ: invalid device specified, highest is %u\n",
+                      IMQ_MAX_DEVS - 1);
+               return 0;
+       }
+
+       return 1;
+}
+
+static struct xt_target xt_imq_reg[] __read_mostly = {
+       {
+               .name           = "IMQ",
+               .family         = AF_INET,
+               .checkentry     = imq_checkentry,
+               .target         = imq_target,
+               .targetsize     = sizeof(struct xt_imq_info),
+               .table          = "mangle",
+               .me             = THIS_MODULE
+       },
+       {
+               .name           = "IMQ",
+               .family         = AF_INET6,
+               .checkentry     = imq_checkentry,
+               .target         = imq_target,
+               .targetsize     = sizeof(struct xt_imq_info),
+               .table          = "mangle",
+               .me             = THIS_MODULE
+       },
+};
+
+static int __init imq_init(void)
+{
+       return xt_register_targets(xt_imq_reg, ARRAY_SIZE(xt_imq_reg));
+}
+
+static void __exit imq_fini(void)
+{
+       xt_unregister_targets(xt_imq_reg, ARRAY_SIZE(xt_imq_reg));
+}
+
+module_init(imq_init);
+module_exit(imq_fini);
+
+MODULE_AUTHOR("http://www.linuximq.net");
+MODULE_DESCRIPTION("Pseudo-driver for the intermediate queue device. See http://www.linuximq.net/ for more information.");
+MODULE_LICENSE("GPL");
+MODULE_ALIAS("ipt_IMQ");
+MODULE_ALIAS("ip6t_IMQ");
+