source: npl/kernel/linux_src/linux-4.9-imq.diff @ 76ea60c

drivers/net/Kconfig

@@ -260 +260 @@
         depends on RIONET
         default "128"
 
+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: https://github.com/imq/linuximq
+
+          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 setting 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: https://github.com/imq/linuximq
+
+          If not sure leave the default settings alone.
+
+config IMQ_BEHAVIOR_AA
+        bool "IMQ AA"
+        help
+          This setting 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: https://github.com/imq/linuximq
+
+          If not sure leave the default settings alone.
+
+config IMQ_BEHAVIOR_AB
+        bool "IMQ AB"
+        help
+          This setting 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: https://github.com/imq/linuximq
+
+          If not sure leave the default settings alone.
+
+config IMQ_BEHAVIOR_BA
+        bool "IMQ BA"
+        help
+          This setting 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: https://github.com/imq/linuximq
+
+          If not sure leave the default settings alone.
+
+config IMQ_BEHAVIOR_BB
+        bool "IMQ BB"
+        help
+          This setting 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: https://github.com/imq/linuximq
+
+          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 setting defines how many IMQ devices will be created.
+
+          The default value is 16.
+
+          More information can be found at: https://github.com/imq/linuximq
+
+          If not sure leave the default settings alone.
+
 config TUN
         tristate "Universal TUN/TAP device driver support"
         depends on INET

drivers/net/Makefile

@@ -11 +11 @@
 obj-$(CONFIG_EQUALIZER) += eql.o
 obj-$(CONFIG_IFB) += ifb.o
 obj-$(CONFIG_MACSEC) += macsec.o
+obj-$(CONFIG_IMQ) += imq.o
 obj-$(CONFIG_MACVLAN) += macvlan.o
 obj-$(CONFIG_MACVTAP) += macvtap.o
 obj-$(CONFIG_MII) += mii.o

new file drivers/net/imq.c

@@ +1 @@
+/*
+ *             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>
+ *
+ *                         See Creditis.txt
+ */
+
+#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>
+#include <net/sock.h>
+#include <linux/ip.h>
+#include <linux/ipv6.h>
+#include <linux/if_vlan.h>
+#include <linux/if_pppox.h>
+#include <net/ip.h>
+#include <net/ipv6.h>
+
+static int imq_nf_queue(struct nf_queue_entry *entry, unsigned queue_num);
+
+static nf_hookfn imq_nf_hook;
+
+static struct nf_hook_ops imq_ops[] = {
+        {
+        /* imq_ingress_ipv4 */
+                .hook           = imq_nf_hook,
+                .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
+        },
+        {
+        /* imq_egress_ipv4 */
+                .hook           = imq_nf_hook,
+                .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)
+        {
+        /* imq_ingress_ipv6 */
+                .hook           = imq_nf_hook,
+                .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
+        },
+        {
+        /* imq_egress_ipv6 */
+                .hook           = imq_nf_hook,
+                .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 int numdevs = CONFIG_IMQ_NUM_DEVS;
+#else
+static int numdevs = IMQ_MAX_DEVS;
+#endif
+
+static struct net_device *imq_devs_cache[IMQ_MAX_DEVS];
+
+#define IMQ_MAX_QUEUES 32
+static int numqueues = 1;
+static u32 imq_hashrnd;
+static int imq_dev_accurate_stats = 1;
+
+static inline __be16 pppoe_proto(const struct sk_buff *skb)
+{
+        return *((__be16 *)(skb_mac_header(skb) + ETH_HLEN +
+                        sizeof(struct pppoe_hdr)));
+}
+
+static u16 imq_hash(struct net_device *dev, struct sk_buff *skb)
+{
+        unsigned int pull_len;
+        u16 protocol = skb->protocol;
+        u32 addr1, addr2;
+        u32 hash, ihl = 0;
+        union {
+                u16 in16[2];
+                u32 in32;
+        } ports;
+        u8 ip_proto;
+
+        pull_len = 0;
+
+recheck:
+        switch (protocol) {
+        case htons(ETH_P_8021Q): {
+                if (unlikely(skb_pull(skb, VLAN_HLEN) == NULL))
+                        goto other;
+
+                pull_len += VLAN_HLEN;
+                skb->network_header += VLAN_HLEN;
+
+                protocol = vlan_eth_hdr(skb)->h_vlan_encapsulated_proto;
+                goto recheck;
+        }
+
+        case htons(ETH_P_PPP_SES): {
+                if (unlikely(skb_pull(skb, PPPOE_SES_HLEN) == NULL))
+                        goto other;
+
+                pull_len += PPPOE_SES_HLEN;
+                skb->network_header += PPPOE_SES_HLEN;
+
+                protocol = pppoe_proto(skb);
+                goto recheck;
+        }
+
+        case htons(ETH_P_IP): {
+                const struct iphdr *iph = ip_hdr(skb);
+
+                if (unlikely(!pskb_may_pull(skb, sizeof(struct iphdr))))
+                        goto other;
+
+                addr1 = iph->daddr;
+                addr2 = iph->saddr;
+
+                ip_proto = !(ip_hdr(skb)->frag_off & htons(IP_MF | IP_OFFSET)) ?
+                                 iph->protocol : 0;
+                ihl = ip_hdrlen(skb);
+
+                break;
+        }
+#if defined(CONFIG_IPV6) || defined(CONFIG_IPV6_MODULE)
+        case htons(ETH_P_IPV6): {
+                const struct ipv6hdr *iph = ipv6_hdr(skb);
+                __be16 fo = 0;
+
+                if (unlikely(!pskb_may_pull(skb, sizeof(struct ipv6hdr))))
+                        goto other;
+
+                addr1 = iph->daddr.s6_addr32[3];
+                addr2 = iph->saddr.s6_addr32[3];
+                ihl = ipv6_skip_exthdr(skb, sizeof(struct ipv6hdr), &ip_proto,
+                                       &fo);
+                if (unlikely(ihl < 0))
+                        goto other;
+
+                break;
+        }
+#endif
+        default:
+other:
+                if (pull_len != 0) {
+                        skb_push(skb, pull_len);
+                        skb->network_header -= pull_len;
+                }
+
+                return (u16)(ntohs(protocol) % dev->real_num_tx_queues);
+        }
+
+        if (addr1 > addr2)
+                swap(addr1, addr2);
+
+        switch (ip_proto) {
+        case IPPROTO_TCP:
+        case IPPROTO_UDP:
+        case IPPROTO_DCCP:
+        case IPPROTO_ESP:
+        case IPPROTO_AH:
+        case IPPROTO_SCTP:
+        case IPPROTO_UDPLITE: {
+                if (likely(skb_copy_bits(skb, ihl, &ports.in32, 4) >= 0)) {
+                        if (ports.in16[0] > ports.in16[1])
+                                swap(ports.in16[0], ports.in16[1]);
+                        break;
+                }
+                /* fall-through */
+        }
+        default:
+                ports.in32 = 0;
+                break;
+        }
+
+        if (pull_len != 0) {
+                skb_push(skb, pull_len);
+                skb->network_header -= pull_len;
+        }
+
+        hash = jhash_3words(addr1, addr2, ports.in32, imq_hashrnd ^ ip_proto);
+
+        return (u16)(((u64)hash * dev->real_num_tx_queues) >> 32);
+}
+
+static inline bool sk_tx_queue_recorded(struct sock *sk)
+{
+        return (sk_tx_queue_get(sk) >= 0);
+}
+
+static struct netdev_queue *imq_select_queue(struct net_device *dev,
+                                                struct sk_buff *skb)
+{
+        u16 queue_index = 0;
+        u32 hash;
+
+        if (likely(dev->real_num_tx_queues == 1))
+                goto out;
+
+        /* IMQ can be receiving ingress or engress packets. */
+
+        /* Check first for if rx_queue is set */
+        if (skb_rx_queue_recorded(skb)) {
+                queue_index = skb_get_rx_queue(skb);
+                goto out;
+        }
+
+        /* Check if socket has tx_queue set */
+        if (sk_tx_queue_recorded(skb->sk)) {
+                queue_index = sk_tx_queue_get(skb->sk);
+                goto out;
+        }
+
+        /* Try use socket hash */
+        if (skb->sk && skb->sk->sk_hash) {
+                hash = skb->sk->sk_hash;
+                queue_index =
+                        (u16)(((u64)hash * dev->real_num_tx_queues) >> 32);
+                goto out;
+        }
+
+        /* Generate hash from packet data */
+        queue_index = imq_hash(dev, skb);
+
+out:
+        if (unlikely(queue_index >= dev->real_num_tx_queues))
+                queue_index = (u16)((u32)queue_index % dev->real_num_tx_queues);
+
+        skb_set_queue_mapping(skb, queue_index);
+        return netdev_get_tx_queue(dev, queue_index);
+}
+
+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;
+
+        skb->nf_queue_entry = NULL;
+
+        if (entry) {
+                nf_queue_entry_release_refs(entry);
+                kfree(entry);
+        }
+
+        skb_restore_cb(skb); /* kfree backup */
+}
+
+static void imq_done_check_queue_mapping(struct sk_buff *skb,
+                                         struct net_device *dev)
+{
+        unsigned int queue_index;
+
+        /* Don't let queue_mapping be left too large after exiting IMQ */
+        if (likely(skb->dev != dev && skb->dev != NULL)) {
+                queue_index = skb_get_queue_mapping(skb);
+                if (unlikely(queue_index >= skb->dev->real_num_tx_queues)) {
+                        queue_index = (u16)((u32)queue_index %
+                                                skb->dev->real_num_tx_queues);
+                        skb_set_queue_mapping(skb, queue_index);
+                }
+        } else {
+                /* skb->dev was IMQ device itself or NULL, be on safe side and
+                 * just clear queue mapping.
+                 */
+                skb_set_queue_mapping(skb, 0);
+        }
+}
+
+static netdev_tx_t imq_dev_xmit(struct sk_buff *skb, struct net_device *dev)
+{
+        struct nf_queue_entry *entry = skb->nf_queue_entry;
+
+        skb->nf_queue_entry = NULL;
+        netif_trans_update(dev);
+
+        dev->stats.tx_bytes += skb->len;
+        dev->stats.tx_packets++;
+
+        if (unlikely(entry == NULL)) {
+                /* We don't know what is going on here.. packet is queued for
+                 * imq device, but (probably) not by us.
+                 *
+                 * If this packet was not send here by imq_nf_queue(), then
+                 * skb_save_cb() was not used and skb_free() should not show:
+                 *   WARNING: IMQ: kfree_skb: skb->cb_next:..
+                 * and/or
+                 *   WARNING: IMQ: kfree_skb: skb->nf_queue_entry...
+                 *
+                 * However if this message is shown, then IMQ is somehow broken
+                 * and you should report this to linuximq.net.
+                 */
+
+                /* imq_dev_xmit is black hole that eats all packets, report that
+                 * we eat this packet happily and increase dropped counters.
+                 */
+
+                dev->stats.tx_dropped++;
+                dev_kfree_skb(skb);
+
+                return NETDEV_TX_OK;
+        }
+
+        skb_restore_cb(skb); /* restore skb->cb */
+
+        skb->imq_flags = 0;
+        skb->destructor = NULL;
+
+        imq_done_check_queue_mapping(skb, dev);
+
+        nf_reinject(entry, NF_ACCEPT);
+
+        return NETDEV_TX_OK;
+}
+
+static struct net_device *get_imq_device_by_index(int index)
+{
+        struct net_device *dev = NULL;
+        struct net *net;
+        char buf[8];
+
+        /* get device by name and cache result */
+        snprintf(buf, sizeof(buf), "imq%d", index);
+
+        /* Search device from all namespaces. */
+        for_each_net(net) {
+                dev = dev_get_by_name(net, buf);
+                if (dev)
+                        break;
+        }
+
+        if (WARN_ON_ONCE(dev == NULL)) {
+                /* IMQ device not found. Exotic config? */
+                return ERR_PTR(-ENODEV);
+        }
+
+        imq_devs_cache[index] = dev;
+        dev_put(dev);
+
+        return dev;
+}
+
+static struct nf_queue_entry *nf_queue_entry_dup(struct nf_queue_entry *e)
+{
+        struct nf_queue_entry *entry = kmemdup(e, e->size, GFP_ATOMIC);
+        if (entry) {
+                nf_queue_entry_get_refs(entry);
+                        return entry;
+        }
+        return NULL;
+}
+
+#ifdef CONFIG_BRIDGE_NETFILTER
+/* When called from bridge netfilter, skb->data must point to MAC header
+ * before calling skb_gso_segment(). Else, original MAC header is lost
+ * and segmented skbs will be sent to wrong destination.
+ */
+static void nf_bridge_adjust_skb_data(struct sk_buff *skb)
+{
+        if (skb->nf_bridge)
+                __skb_push(skb, skb->network_header - skb->mac_header);
+}
+
+static void nf_bridge_adjust_segmented_data(struct sk_buff *skb)
+{
+        if (skb->nf_bridge)
+                __skb_pull(skb, skb->network_header - skb->mac_header);
+}
+#else
+#define nf_bridge_adjust_skb_data(s) do {} while (0)
+#define nf_bridge_adjust_segmented_data(s) do {} while (0)
+#endif
+
+static void free_entry(struct nf_queue_entry *entry)
+{
+        nf_queue_entry_release_refs(entry);
+        kfree(entry);
+}
+
+static int __imq_nf_queue(struct nf_queue_entry *entry, struct net_device *dev);
+
+static int __imq_nf_queue_gso(struct nf_queue_entry *entry,
+                              struct net_device *dev, struct sk_buff *skb)
+{
+        int ret = -ENOMEM;
+        struct nf_queue_entry *entry_seg;
+
+        nf_bridge_adjust_segmented_data(skb);
+
+        if (skb->next == NULL) { /* last packet, no need to copy entry */
+                struct sk_buff *gso_skb = entry->skb;
+                entry->skb = skb;
+                ret = __imq_nf_queue(entry, dev);
+                if (ret)
+                        entry->skb = gso_skb;
+                return ret;
+        }
+
+        skb->next = NULL;
+
+        entry_seg = nf_queue_entry_dup(entry);
+        if (entry_seg) {
+                entry_seg->skb = skb;
+                ret = __imq_nf_queue(entry_seg, dev);
+                if (ret)
+                        free_entry(entry_seg);
+        }
+        return ret;
+}
+
+static int imq_nf_queue(struct nf_queue_entry *entry, unsigned queue_num)
+{
+        struct sk_buff *skb, *segs;
+        struct net_device *dev;
+        unsigned int queued;
+        int index, retval, err;
+
+        index = entry->skb->imq_flags & IMQ_F_IFMASK;
+        if (unlikely(index > numdevs - 1)) {
+                if (net_ratelimit())
+                        pr_warn("IMQ: invalid device specified, highest is %u\n",
+                                numdevs - 1);
+                retval = -EINVAL;
+                goto out_no_dev;
+        }
+
+        /* check for imq device by index from cache */
+        dev = imq_devs_cache[index];
+        if (unlikely(!dev)) {
+                dev = get_imq_device_by_index(index);
+                if (IS_ERR(dev)) {
+                        retval = PTR_ERR(dev);
+                        goto out_no_dev;
+                }
+        }
+
+        if (unlikely(!(dev->flags & IFF_UP))) {
+                entry->skb->imq_flags = 0;
+                retval = -ECANCELED;
+                goto out_no_dev;
+        }
+
+        /* Since 3.10.x, GSO handling moved here as result of upstream commit
+         * a5fedd43d5f6c94c71053a66e4c3d2e35f1731a2 (netfilter: move
+         * skb_gso_segment into nfnetlink_queue module).
+         *
+         * Following code replicates the gso handling from
+         * 'net/netfilter/nfnetlink_queue_core.c':nfqnl_enqueue_packet().
+         */
+
+        skb = entry->skb;
+
+        switch (entry->state.pf) {
+        case NFPROTO_IPV4:
+                skb->protocol = htons(ETH_P_IP);
+                break;
+        case NFPROTO_IPV6:
+                skb->protocol = htons(ETH_P_IPV6);
+                break;
+        }
+
+        if (!skb_is_gso(entry->skb))
+                return __imq_nf_queue(entry, dev);
+
+        nf_bridge_adjust_skb_data(skb);
+        segs = skb_gso_segment(skb, 0);
+        /* Does not use PTR_ERR to limit the number of error codes that can be
+         * returned by nf_queue.  For instance, callers rely on -ECANCELED to
+         * mean 'ignore this hook'.
+         */
+        err = -ENOBUFS;
+        if (IS_ERR(segs))
+                goto out_err;
+        queued = 0;
+        err = 0;
+        do {
+                struct sk_buff *nskb = segs->next;
+                if (nskb && nskb->next)
+                        nskb->cb_next = NULL;
+                if (err == 0)
+                        err = __imq_nf_queue_gso(entry, dev, segs);
+                if (err == 0)
+                        queued++;
+                else
+                        kfree_skb(segs);
+                segs = nskb;
+        } while (segs);
+
+        if (queued) {
+                if (err) /* some segments are already queued */
+                        free_entry(entry);
+                kfree_skb(skb);
+                return 0;
+        }
+
+out_err:
+        nf_bridge_adjust_segmented_data(skb);
+        retval = err;
+out_no_dev:
+        return retval;
+}
+
+static int __imq_nf_queue(struct nf_queue_entry *entry, struct net_device *dev)
+{
+        struct sk_buff *skb_orig, *skb, *skb_shared, *skb_popd;
+        struct Qdisc *q;
+        struct sk_buff *to_free = NULL;
+        struct netdev_queue *txq;
+        spinlock_t *root_lock;
+        int users;
+        int retval = -EINVAL;
+        unsigned int orig_queue_index;
+
+        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 (unlikely(!skb)) {
+                        retval = -ENOMEM;
+                        goto out;
+                }
+                skb->cb_next = NULL;
+                entry->skb = skb;
+        }
+
+        dev->stats.rx_bytes += skb->len;
+        dev->stats.rx_packets++;
+
+        if (!skb->dev) {
+                /* skb->dev == NULL causes problems, try the find cause. */
+                if (net_ratelimit()) {
+                        dev_warn(&dev->dev,
+                                 "received packet with skb->dev == NULL\n");
+                        dump_stack();
+                }
+
+                skb->dev = dev;
+        }
+
+        /* Disables softirqs for lock below */
+        rcu_read_lock_bh();
+
+        /* Multi-queue selection */
+        orig_queue_index = skb_get_queue_mapping(skb);
+        txq = imq_select_queue(dev, skb);
+
+        q = rcu_dereference(txq->qdisc);
+        if (unlikely(!q->enqueue))
+                goto packet_not_eaten_by_imq_dev;
+
+        skb->nf_queue_entry = entry;
+        root_lock = qdisc_lock(q);
+        spin_lock(root_lock);
+
+        users = atomic_read(&skb->users);
+
+        skb_shared = skb_get(skb); /* increase reference count by one */
+
+        /* backup skb->cb, as qdisc layer will overwrite it */
+        skb_save_cb(skb_shared);
+        qdisc_enqueue_root(skb_shared, q, &to_free); /* might kfree_skb */
+        if (likely(atomic_read(&skb_shared->users) == users + 1)) {
+                bool validate;
+
+                kfree_skb(skb_shared); /* decrease reference count by one */
+
+                skb->destructor = &imq_skb_destructor;
+
+                skb_popd = qdisc_dequeue_skb(q, &validate);
+
+                /* cloned? */
+                if (unlikely(skb_orig))
+                        kfree_skb(skb_orig); /* free original */
+
+                spin_unlock(root_lock);
+
+#if 0
+                /* schedule qdisc dequeue */
+                __netif_schedule(q);
+#else
+                if (likely(skb_popd)) {
+                        /* Note that we validate skb (GSO, checksum, ...) outside of locks */
+                        if (validate)
+                        skb_popd = validate_xmit_skb_list(skb_popd, dev);
+
+                        if (skb_popd) {
+                                int dummy_ret;
+                                int cpu = smp_processor_id(); /* ok because BHs are off */
+
+                                txq = skb_get_tx_queue(dev, skb_popd);
+                                /*
+                                IMQ device will not be frozen or stoped, and it always be successful.
+                                So we need not check its status and return value to accelerate.
+                                */
+                                if (imq_dev_accurate_stats && txq->xmit_lock_owner != cpu) {
+                                        HARD_TX_LOCK(dev, txq, cpu);
+                                        if (!netif_xmit_frozen_or_stopped(txq)) {
+                                                dev_hard_start_xmit(skb_popd, dev, txq, &dummy_ret);
+                                        }
+                                        HARD_TX_UNLOCK(dev, txq);
+                                } else {
+                                        if (!netif_xmit_frozen_or_stopped(txq)) {
+                                                dev_hard_start_xmit(skb_popd, dev, txq, &dummy_ret);
+                                        }
+                                }
+                        }
+                } else {
+                        /* No ready skb, then schedule it */
+                        __netif_schedule(q);
+                }
+#endif
+                rcu_read_unlock_bh();
+                retval = 0;
+                goto out;
+        } else {
+                skb_restore_cb(skb_shared); /* restore skb->cb */
+                skb->nf_queue_entry = NULL;
+                /*
+                 * 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(root_lock);
+                goto packet_not_eaten_by_imq_dev;
+        }
+
+packet_not_eaten_by_imq_dev:
+        skb_set_queue_mapping(skb, orig_queue_index);
+        rcu_read_unlock_bh();
+
+        /* cloned? restore original */
+        if (unlikely(skb_orig)) {
+                kfree_skb(skb);
+                entry->skb = skb_orig;
+        }
+        retval = -1;
+out:
+        if (unlikely(to_free)) {
+                kfree_skb_list(to_free);
+        }
+        return retval;
+}
+static unsigned int imq_nf_hook(void *priv,
+                                struct sk_buff *skb,
+                                const struct nf_hook_state *state)
+{
+        return (skb->imq_flags & IMQ_F_ENQUEUE) ? NF_IMQ_QUEUE : 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 const struct net_device_ops imq_netdev_ops = {
+        .ndo_open               = imq_open,
+        .ndo_stop               = imq_close,
+        .ndo_start_xmit         = imq_dev_xmit,
+        .ndo_get_stats          = imq_get_stats,
+};
+
+static void imq_setup(struct net_device *dev)
+{
+        dev->netdev_ops         = &imq_netdev_ops;
+        dev->type               = ARPHRD_VOID;
+        dev->mtu                = 16000; /* too small? */
+        dev->tx_queue_len       = 11000; /* too big? */
+        dev->flags              = IFF_NOARP;
+        dev->features           = NETIF_F_SG | NETIF_F_FRAGLIST |
+                                  NETIF_F_GSO | NETIF_F_HW_CSUM |
+                                  NETIF_F_HIGHDMA;
+        dev->priv_flags         &= ~(IFF_XMIT_DST_RELEASE |
+                                     IFF_TX_SKB_SHARING);
+}
+
+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:
+        pr_warn("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 const struct nf_queue_handler imq_nfqh = {
+        .outfn = imq_nf_queue,
+};
+
+static int __init imq_init_hooks(void)
+{
+        int ret;
+
+        nf_register_queue_imq_handler(&imq_nfqh);
+
+        ret = nf_register_hooks(imq_ops, ARRAY_SIZE(imq_ops));
+        if (ret < 0)
+                nf_unregister_queue_imq_handler();
+
+        return ret;
+}
+
+static int __init imq_init_one(int index)
+{
+        struct net_device *dev;
+        int ret;
+
+        dev = alloc_netdev_mq(0, "imq%d", NET_NAME_UNKNOWN, imq_setup, numqueues);
+        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) {
+                pr_err("IMQ: numdevs has to be betweed 1 and %u\n",
+                       IMQ_MAX_DEVS);
+                return -EINVAL;
+        }
+
+        if (numqueues < 1 || numqueues > IMQ_MAX_QUEUES) {
+                pr_err("IMQ: numqueues has to be betweed 1 and %u\n",
+                       IMQ_MAX_QUEUES);
+                return -EINVAL;
+        }
+
+        get_random_bytes(&imq_hashrnd, sizeof(imq_hashrnd));
+
+        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) {
+                pr_err("IMQ: Error trying imq_init_devs(net)\n");
+                return err;
+        }
+
+        err = imq_init_hooks();
+        if (err) {
+                pr_err(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;
+        }
+
+        pr_info("IMQ driver loaded successfully. (numdevs = %d, numqueues = %d, imq_dev_accurate_stats = %d)\n",
+                numdevs, numqueues, imq_dev_accurate_stats);
+
+#if defined(CONFIG_IMQ_BEHAVIOR_BA) || defined(CONFIG_IMQ_BEHAVIOR_BB)
+        pr_info("\tHooking IMQ before NAT on PREROUTING.\n");
+#else
+        pr_info("\tHooking IMQ after NAT on PREROUTING.\n");
+#endif
+#if defined(CONFIG_IMQ_BEHAVIOR_AB) || defined(CONFIG_IMQ_BEHAVIOR_BB)
+        pr_info("\tHooking IMQ before NAT on POSTROUTING.\n");
+#else
+        pr_info("\tHooking IMQ after NAT on POSTROUTING.\n");
+#endif
+
+        return 0;
+}
+
+static void __exit imq_unhook(void)
+{
+        nf_unregister_hooks(imq_ops, ARRAY_SIZE(imq_ops));
+        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();
+        pr_info("IMQ driver unloaded successfully.\n");
+}
+
+module_init(imq_init_module);
+module_exit(imq_exit_module);
+
+module_param(numdevs, int, 0);
+module_param(numqueues, int, 0);
+module_param(imq_dev_accurate_stats, int, 0);
+MODULE_PARM_DESC(numdevs, "number of IMQ devices (how many imq* devices will be created)");
+MODULE_PARM_DESC(numqueues, "number of queues per IMQ device");
+MODULE_PARM_DESC(imq_dev_accurate_stats, "Notify if need the accurate imq device stats");
+
+MODULE_AUTHOR("https://github.com/imq/linuximq");
+MODULE_DESCRIPTION("Pseudo-driver for the intermediate queue device. See https://github.com/imq/linuximq/wiki for more information.");
+MODULE_LICENSE("GPL");
+MODULE_ALIAS_RTNL_LINK("imq");

new file include/linux/imq.h

@@ +1 @@
+#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 */
+

include/linux/netdevice.h

@@ -3669 +3669 @@
         }                                               \
 }
 
+#define HARD_TX_LOCK_BH(dev, txq) {           \
+    if ((dev->features & NETIF_F_LLTX) == 0) {  \
+        __netif_tx_lock_bh(txq);      \
+    }                       \
+}
+
+#define HARD_TX_UNLOCK_BH(dev, txq) {          \
+    if ((dev->features & NETIF_F_LLTX) == 0) {  \
+        __netif_tx_unlock_bh(txq);         \
+    }                       \
+}
+
+
 static inline void netif_tx_disable(struct net_device *dev)
 {
         unsigned int i;

new file include/linux/netfilter/xt_IMQ.h

@@ +1 @@
+#ifndef _XT_IMQ_H
+#define _XT_IMQ_H
+
+struct xt_imq_info {
+        unsigned int todev;     /* target imq device */
+};
+
+#endif /* _XT_IMQ_H */
+

new file include/linux/netfilter_ipv4/ipt_IMQ.h

@@ +1 @@
+#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 */
+

new file include/linux/netfilter_ipv6/ip6t_IMQ.h

@@ +1 @@
+#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 */
+

include/linux/skbuff.h

@@ -39 +39 @@
 #include <linux/in6.h>
 #include <linux/if_packet.h>
 #include <net/flow.h>
+#if defined(CONFIG_IMQ) || defined(CONFIG_IMQ_MODULE)
+#include <linux/imq.h>
+#endif
+
 
 /* The interface for checksum offload between the stack and networking drivers
  * is as follows...
@@ -654 +658 @@
          * first. This is owned by whoever has the skb queued ATM.
          */
         char                    cb[48] __aligned(8);
+#if defined(CONFIG_IMQ) || defined(CONFIG_IMQ_MODULE)
+        void                    *cb_next;
+#endif
 
         unsigned long           _skb_refdst;
         void                    (*destructor)(struct sk_buff *skb);
@@ -663 +670 @@
 #if defined(CONFIG_NF_CONNTRACK) || defined(CONFIG_NF_CONNTRACK_MODULE)
         struct nf_conntrack     *nfct;
 #endif
+#if defined(CONFIG_IMQ) || defined(CONFIG_IMQ_MODULE)
+       struct nf_queue_entry   *nf_queue_entry;
+#endif
 #if IS_ENABLED(CONFIG_BRIDGE_NETFILTER)
         struct nf_bridge_info   *nf_bridge;
 #endif
@@ -743 +753 @@
         __u8                    offload_fwd_mark:1;
 #endif
         /* 2, 4 or 5 bit hole */
+#if defined(CONFIG_IMQ) || defined(CONFIG_IMQ_MODULE)
+        __u8                    imq_flags:IMQ_F_BITS;
+#endif
 
 #ifdef CONFIG_NET_SCHED
         __u16                   tc_index;       /* traffic control index */
@@ -903 +916 @@
 void skb_tx_error(struct sk_buff *skb);
 void consume_skb(struct sk_buff *skb);
 void  __kfree_skb(struct sk_buff *skb);
+
+#if defined(CONFIG_IMQ) || defined(CONFIG_IMQ_MODULE)
+int skb_save_cb(struct sk_buff *skb);
+int skb_restore_cb(struct sk_buff *skb);
+#endif
+
 extern struct kmem_cache *skbuff_head_cache;
 
 void kfree_skb_partial(struct sk_buff *skb, bool head_stolen);
@@ -3594 +3613 @@
         if (copy)
                 dst->nfctinfo = src->nfctinfo;
 #endif
+#if defined(CONFIG_IMQ) || defined(CONFIG_IMQ_MODULE)
+       dst->imq_flags = src->imq_flags;
+       dst->nf_queue_entry = src->nf_queue_entry;
+#endif
 #if IS_ENABLED(CONFIG_BRIDGE_NETFILTER)
         dst->nf_bridge  = src->nf_bridge;
         nf_bridge_get(src->nf_bridge);

include/net/netfilter/nf_queue.h

@@ -30 +30 @@
 void nf_register_queue_handler(struct net *net, const struct nf_queue_handler *qh);
 void nf_unregister_queue_handler(struct net *net);
 void nf_reinject(struct nf_queue_entry *entry, unsigned int verdict);
+void nf_queue_entry_release_refs(struct nf_queue_entry *entry);
+
+#if defined(CONFIG_IMQ) || defined(CONFIG_IMQ_MODULE)
+void nf_register_queue_imq_handler(const struct nf_queue_handler *qh);
+void nf_unregister_queue_imq_handler(void);
+#endif
 
 void nf_queue_entry_get_refs(struct nf_queue_entry *entry);
 void nf_queue_entry_release_refs(struct nf_queue_entry *entry);

include/net/pkt_sched.h

@@ -105 +105 @@
 
 void __qdisc_run(struct Qdisc *q);
 
+struct sk_buff *qdisc_dequeue_skb(struct Qdisc *q, bool *validate);
+
 static inline void qdisc_run(struct Qdisc *q)
 {
         if (qdisc_run_begin(q))

include/net/sch_generic.h

@@ -518 +518 @@
         return sch->enqueue(skb, sch, to_free);
 }
 
+static inline int qdisc_enqueue_root(struct sk_buff *skb, struct Qdisc *sch,
+                                      struct sk_buff **to_free)
+{
+    qdisc_skb_cb(skb)->pkt_len = skb->len;
+    return qdisc_enqueue(skb, sch, to_free) & NET_XMIT_MASK;
+}
+
 static inline bool qdisc_is_percpu_stats(const struct Qdisc *q)
 {
         return q->flags & TCQ_F_CPUSTATS;

include/uapi/linux/netfilter.h

@@ -14 +14 @@
 #define NF_QUEUE 3
 #define NF_REPEAT 4
 #define NF_STOP 5
-#define NF_MAX_VERDICT NF_STOP
+#define NF_IMQ_QUEUE 6
+#define NF_MAX_VERDICT NF_IMQ_QUEUE
 
 /* we overload the higher bits for encoding auxiliary data such as the queue
  * number or errno values. Not nice, but better than additional function

net/core/dev.c

@@ -141 +141 @@
 #include <linux/netfilter_ingress.h>
 #include <linux/sctp.h>
 #include <linux/crash_dump.h>
+#if defined(CONFIG_IMQ) || defined(CONFIG_IMQ_MODULE)
+#include <linux/imq.h>
+#endif
 
 #include "net-sysfs.h"
 
@@ -2906 +2909 @@
         unsigned int len;
         int rc;
 
+#if defined(CONFIG_IMQ) || defined(CONFIG_IMQ_MODULE)
+        if ((!list_empty(&ptype_all) || !list_empty(&dev->ptype_all)) &&
+                !(skb->imq_flags & IMQ_F_ENQUEUE))
+#else
         if (!list_empty(&ptype_all) || !list_empty(&dev->ptype_all))
+#endif
                 dev_queue_xmit_nit(skb, dev);
 
         len = skb->len;
@@ -2945 +2953 @@
         return skb;
 }
 
+EXPORT_SYMBOL_GPL(dev_hard_start_xmit);
+
 static struct sk_buff *validate_xmit_vlan(struct sk_buff *skb,
                                           netdev_features_t features)
 {

net/core/skbuff.c

@@ -82 +82 @@
 static struct kmem_cache *skbuff_fclone_cache __read_mostly;
 int sysctl_max_skb_frags __read_mostly = MAX_SKB_FRAGS;
 EXPORT_SYMBOL(sysctl_max_skb_frags);
+#if defined(CONFIG_IMQ) || defined(CONFIG_IMQ_MODULE)
+static struct kmem_cache *skbuff_cb_store_cache __read_mostly;
+#endif
+
+#if defined(CONFIG_IMQ) || defined(CONFIG_IMQ_MODULE)
+/* Control buffer save/restore for IMQ devices */
+struct skb_cb_table {
+        char                    cb[48] __aligned(8);
+        void                    *cb_next;
+        atomic_t                refcnt;
+};
+
+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 *   , const struct sk_buff *     ) __attribute__ ((unused));
+static void skb_copy_stored_cb(struct sk_buff *new, const struct sk_buff *__old)
+{
+        struct skb_cb_table *next;
+        struct sk_buff *old;
+
+        if (!__old->cb_next) {
+                new->cb_next = NULL;
+                return;
+        }
+
+        spin_lock(&skb_cb_store_lock);
+
+        old = (struct sk_buff *)__old;
+
+        next = old->cb_next;
+        atomic_inc(&next->refcnt);
+        new->cb_next = next;
+
+        spin_unlock(&skb_cb_store_lock);
+}
+#endif
 
 /**
  *      skb_panic - private function for out-of-line support
@@ -654 +735 @@
                 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) {
+                if (net_ratelimit())
+                        pr_warn("IMQ: kfree_skb: skb->cb_next: %08x\n",
+                                (unsigned int)(uintptr_t)skb->cb_next);
+
+                skb_restore_cb(skb);
+        }
+        /*
+         * This should not happen either, nf_queue_entry is nullified in
+         * imq_dev_xmit(). If we have non-NULL nf_queue_entry then we are
+         * leaking entry pointers, maybe memory. We don't know if this is
+         * pointer to already freed memory, or should this be freed.
+         * If this happens we need to add refcounting, etc for nf_queue_entry.
+         */
+        if (skb->nf_queue_entry && net_ratelimit())
+                pr_warn("%s\n", "IMQ: kfree_skb: skb->nf_queue_entry != NULL");
+#endif
 #if IS_ENABLED(CONFIG_NF_CONNTRACK)
         nf_conntrack_put(skb->nfct);
 #endif
@@ -843 +946 @@
         new->sp                 = secpath_get(old->sp);
 #endif
         __nf_copy(new, old, false);
+#if defined(CONFIG_IMQ) || defined(CONFIG_IMQ_MODULE)
+        new->cb_next = NULL;
+        /*skb_copy_stored_cb(new, old);*/
+#endif
 
         /* Note : this field could be in headers_start/headers_end section
          * It is not yet because we do not want to have a 16 bit hole
@@ -3464 +3571 @@
                                                 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
 }
 
 /**

net/ipv6/ip6_output.c

@@ -66 +66 @@
         struct in6_addr *nexthop;
         int ret;
 
-        skb->protocol = htons(ETH_P_IPV6);
-        skb->dev = dev;
-
         if (ipv6_addr_is_multicast(&ipv6_hdr(skb)->daddr)) {
                 struct inet6_dev *idev = ip6_dst_idev(skb_dst(skb));
 
@@ -150 +147 @@
                 return 0;
         }
 
+        /*
+        * IMQ-patch: moved setting skb->dev and skb->protocol from
+        * ip6_finish_output2 to fix crashing at netif_skb_features().
+        */
+        skb->protocol = htons(ETH_P_IPV6);
+        skb->dev = dev;
+
         return NF_HOOK_COND(NFPROTO_IPV6, NF_INET_POST_ROUTING,
                             net, sk, skb, NULL, dev,
                             ip6_finish_output,

net/netfilter/Kconfig

@@ -833 +833 @@
 
           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'
         depends on NETFILTER_ADVANCED

net/netfilter/Makefile

@@ -119 +119 @@
 obj-$(CONFIG_NETFILTER_XT_TARGET_DSCP) += xt_DSCP.o
 obj-$(CONFIG_NETFILTER_XT_TARGET_HL) += xt_HL.o
 obj-$(CONFIG_NETFILTER_XT_TARGET_HMARK) += xt_HMARK.o
+obj-$(CONFIG_NETFILTER_XT_TARGET_IMQ) += xt_IMQ.o
 obj-$(CONFIG_NETFILTER_XT_TARGET_LED) += xt_LED.o
 obj-$(CONFIG_NETFILTER_XT_TARGET_LOG) += xt_LOG.o
 obj-$(CONFIG_NETFILTER_XT_TARGET_NETMAP) += xt_NETMAP.o

net/netfilter/core.c

@@ -360 +360 @@
                 ret = NF_DROP_GETERR(verdict);
                 if (ret == 0)
                         ret = -EPERM;
-        } else if ((verdict & NF_VERDICT_MASK) == NF_QUEUE) {
+        } else if ((verdict & NF_VERDICT_MASK) == NF_QUEUE ||
+                   (verdict & NF_VERDICT_MASK) == NF_IMQ_QUEUE) {
                 ret = nf_queue(skb, state, &entry, verdict);
+                if (ret == -ECANCELED)
+                        goto next_hook;
                 if (ret == 1 && entry)
                         goto next_hook;
         }

net/netfilter/nf_queue.c

@@ -27 +27 @@
  * receives, no matter what.
  */
 
+#if defined(CONFIG_IMQ) || defined(CONFIG_IMQ_MODULE)
+static const struct nf_queue_handler __rcu *queue_imq_handler __read_mostly;
+
+void nf_register_queue_imq_handler(const struct nf_queue_handler *qh)
+{
+        rcu_assign_pointer(queue_imq_handler, qh);
+}
+EXPORT_SYMBOL_GPL(nf_register_queue_imq_handler);
+
+void nf_unregister_queue_imq_handler(void)
+{
+        RCU_INIT_POINTER(queue_imq_handler, NULL);
+        synchronize_rcu();
+}
+EXPORT_SYMBOL_GPL(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. */
 void nf_register_queue_handler(struct net *net, const struct nf_queue_handler *qh)
@@ -108 +125 @@
 }
 
 static int __nf_queue(struct sk_buff *skb, const struct nf_hook_state *state,
-                      unsigned int queuenum)
+                      unsigned int verdict)
 {
         int status = -ENOENT;
         struct nf_queue_entry *entry = NULL;
         const struct nf_afinfo *afinfo;
         const struct nf_queue_handler *qh;
         struct net *net = state->net;
+        unsigned int queuetype = verdict & NF_VERDICT_MASK;
+        unsigned int queuenum  = verdict >> NF_VERDICT_QBITS;
 
         /* QUEUE == DROP if no one is waiting, to be safe. */
-        qh = rcu_dereference(net->nf.queue_handler);
+        if (queuetype == NF_IMQ_QUEUE) {
+#if defined(CONFIG_IMQ) || defined(CONFIG_IMQ_MODULE)
+                qh = rcu_dereference(queue_imq_handler);
+#else
+                BUG();
+                goto err_unlock;
+#endif
+        } else {
+                qh = rcu_dereference(net->nf.queue_handler);
+        }
+
         if (!qh) {
                 status = -ESRCH;
                 goto err;
@@ -164 +193 @@
         int ret;
 
         RCU_INIT_POINTER(state->hook_entries, entry);
-        ret = __nf_queue(skb, state, verdict >> NF_VERDICT_QBITS);
+        ret = __nf_queue(skb, state, verdict);
         if (ret < 0) {
+#if defined(CONFIG_IMQ) || defined(CONFIG_IMQ_MODULE)
+                if (ret == -ECANCELED && skb->imq_flags == 0) { // down interface
+                        *entryp = rcu_dereference(entry->next);
+                        return 1;
+                }
+#endif
                 if (ret == -ESRCH &&
                     (verdict & NF_VERDICT_FLAG_QUEUE_BYPASS)) {
                         *entryp = rcu_dereference(entry->next);
@@ -218 +253 @@
                 local_bh_enable();
                 break;
         case NF_QUEUE:
+        case NF_IMQ_QUEUE:
                 err = nf_queue(skb, &entry->state, &hook_entry, verdict);
                 if (err == 1) {
                         if (hook_entry)

new file net/netfilter/xt_IMQ.c

@@ +1 @@
+/*
+ * 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_action_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 int imq_checkentry(const struct xt_tgchk_param *par)
+{
+        struct xt_imq_info *mr = par->targinfo;
+
+        if (mr->todev > IMQ_MAX_DEVS - 1) {
+                pr_warn("IMQ: invalid device specified, highest is %u\n",
+                        IMQ_MAX_DEVS - 1);
+                return -EINVAL;
+        }
+
+        return 0;
+}
+
+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("https://github.com/imq/linuximq");
+MODULE_DESCRIPTION("Pseudo-driver for the intermediate queue device. See https://github.com/imq/linuximq/wiki for more information.");
+MODULE_LICENSE("GPL");
+MODULE_ALIAS("ipt_IMQ");
+MODULE_ALIAS("ip6t_IMQ");
+

net/sched/sch_generic.c

@@ -154 +154 @@
         return skb;
 }
 
+struct sk_buff *qdisc_dequeue_skb(struct Qdisc *q, bool *validate)
+{
+        int packets;
+
+        return dequeue_skb(q, validate, &packets);
+}
+EXPORT_SYMBOL(qdisc_dequeue_skb);
+
 /*
  * Transmit possibly several skbs, and handle the return status as
  * required. Owning running seqcount bit guarantees that