1 diff -Naupr linux-4.14_orig/drivers/net/imq.c linux-4.14/drivers/net/imq.c
2 --- linux-4.14_orig/drivers/net/imq.c 1970-01-01 07:00:00.000000000 +0700
3 +++ linux-4.14/drivers/net/imq.c 2017-11-13 11:46:45.844089945 +0700
6 + * Pseudo-driver for the intermediate queue device.
8 + * This program is free software; you can redistribute it and/or
9 + * modify it under the terms of the GNU General Public License
10 + * as published by the Free Software Foundation; either version
11 + * 2 of the License, or (at your option) any later version.
13 + * Authors: Patrick McHardy, <kaber@trash.net>
15 + * The first version was written by Martin Devera, <devik@cdi.cz>
20 +#include <linux/module.h>
21 +#include <linux/kernel.h>
22 +#include <linux/moduleparam.h>
23 +#include <linux/list.h>
24 +#include <linux/skbuff.h>
25 +#include <linux/netdevice.h>
26 +#include <linux/etherdevice.h>
27 +#include <linux/rtnetlink.h>
28 +#include <linux/if_arp.h>
29 +#include <linux/netfilter.h>
30 +#include <linux/netfilter_ipv4.h>
31 +#if defined(CONFIG_IPV6) || defined(CONFIG_IPV6_MODULE)
32 +#include <linux/netfilter_ipv6.h>
34 +#include <linux/imq.h>
35 +#include <net/pkt_sched.h>
36 +#include <net/netfilter/nf_queue.h>
37 +#include <net/sock.h>
38 +#include <linux/ip.h>
39 +#include <linux/ipv6.h>
40 +#include <linux/if_vlan.h>
41 +#include <linux/if_pppox.h>
43 +#include <net/ipv6.h>
45 +static int imq_nf_queue(struct nf_queue_entry *entry, unsigned queue_num);
47 +static nf_hookfn imq_nf_hook;
49 +static struct nf_hook_ops imq_ops[] = {
51 + /* imq_ingress_ipv4 */
52 + .hook = imq_nf_hook,
54 + .hooknum = NF_INET_PRE_ROUTING,
55 +#if defined(CONFIG_IMQ_BEHAVIOR_BA) || defined(CONFIG_IMQ_BEHAVIOR_BB)
56 + .priority = NF_IP_PRI_MANGLE + 1,
58 + .priority = NF_IP_PRI_NAT_DST + 1,
62 + /* imq_egress_ipv4 */
63 + .hook = imq_nf_hook,
65 + .hooknum = NF_INET_POST_ROUTING,
66 +#if defined(CONFIG_IMQ_BEHAVIOR_AA) || defined(CONFIG_IMQ_BEHAVIOR_BA)
67 + .priority = NF_IP_PRI_LAST,
69 + .priority = NF_IP_PRI_NAT_SRC - 1,
72 +#if defined(CONFIG_IPV6) || defined(CONFIG_IPV6_MODULE)
74 + /* imq_ingress_ipv6 */
75 + .hook = imq_nf_hook,
77 + .hooknum = NF_INET_PRE_ROUTING,
78 +#if defined(CONFIG_IMQ_BEHAVIOR_BA) || defined(CONFIG_IMQ_BEHAVIOR_BB)
79 + .priority = NF_IP6_PRI_MANGLE + 1,
81 + .priority = NF_IP6_PRI_NAT_DST + 1,
85 + /* imq_egress_ipv6 */
86 + .hook = imq_nf_hook,
88 + .hooknum = NF_INET_POST_ROUTING,
89 +#if defined(CONFIG_IMQ_BEHAVIOR_AA) || defined(CONFIG_IMQ_BEHAVIOR_BA)
90 + .priority = NF_IP6_PRI_LAST,
92 + .priority = NF_IP6_PRI_NAT_SRC - 1,
98 +#if defined(CONFIG_IMQ_NUM_DEVS)
99 +static int numdevs = CONFIG_IMQ_NUM_DEVS;
101 +static int numdevs = IMQ_MAX_DEVS;
104 +static struct net_device *imq_devs_cache[IMQ_MAX_DEVS];
106 +#define IMQ_MAX_QUEUES 32
107 +static int numqueues = 1;
108 +static u32 imq_hashrnd;
109 +static int imq_dev_accurate_stats = 1;
111 +static inline __be16 pppoe_proto(const struct sk_buff *skb)
113 + return *((__be16 *)(skb_mac_header(skb) + ETH_HLEN +
114 + sizeof(struct pppoe_hdr)));
117 +static u16 imq_hash(struct net_device *dev, struct sk_buff *skb)
119 + unsigned int pull_len;
120 + u16 protocol = skb->protocol;
132 + switch (protocol) {
133 + case htons(ETH_P_8021Q): {
134 + if (unlikely(skb_pull(skb, VLAN_HLEN) == NULL))
137 + pull_len += VLAN_HLEN;
138 + skb->network_header += VLAN_HLEN;
140 + protocol = vlan_eth_hdr(skb)->h_vlan_encapsulated_proto;
144 + case htons(ETH_P_PPP_SES): {
145 + if (unlikely(skb_pull(skb, PPPOE_SES_HLEN) == NULL))
148 + pull_len += PPPOE_SES_HLEN;
149 + skb->network_header += PPPOE_SES_HLEN;
151 + protocol = pppoe_proto(skb);
155 + case htons(ETH_P_IP): {
156 + const struct iphdr *iph = ip_hdr(skb);
158 + if (unlikely(!pskb_may_pull(skb, sizeof(struct iphdr))))
161 + addr1 = iph->daddr;
162 + addr2 = iph->saddr;
164 + ip_proto = !(ip_hdr(skb)->frag_off & htons(IP_MF | IP_OFFSET)) ?
166 + ihl = ip_hdrlen(skb);
170 +#if defined(CONFIG_IPV6) || defined(CONFIG_IPV6_MODULE)
171 + case htons(ETH_P_IPV6): {
172 + const struct ipv6hdr *iph = ipv6_hdr(skb);
175 + if (unlikely(!pskb_may_pull(skb, sizeof(struct ipv6hdr))))
178 + addr1 = iph->daddr.s6_addr32[3];
179 + addr2 = iph->saddr.s6_addr32[3];
180 + ihl = ipv6_skip_exthdr(skb, sizeof(struct ipv6hdr), &ip_proto,
182 + if (unlikely(ihl < 0))
190 + if (pull_len != 0) {
191 + skb_push(skb, pull_len);
192 + skb->network_header -= pull_len;
195 + return (u16)(ntohs(protocol) % dev->real_num_tx_queues);
199 + swap(addr1, addr2);
201 + switch (ip_proto) {
208 + case IPPROTO_UDPLITE: {
209 + if (likely(skb_copy_bits(skb, ihl, &ports.in32, 4) >= 0)) {
210 + if (ports.in16[0] > ports.in16[1])
211 + swap(ports.in16[0], ports.in16[1]);
221 + if (pull_len != 0) {
222 + skb_push(skb, pull_len);
223 + skb->network_header -= pull_len;
226 + hash = jhash_3words(addr1, addr2, ports.in32, imq_hashrnd ^ ip_proto);
228 + return (u16)(((u64)hash * dev->real_num_tx_queues) >> 32);
231 +static inline bool sk_tx_queue_recorded(struct sock *sk)
233 + return (sk_tx_queue_get(sk) >= 0);
236 +static struct netdev_queue *imq_select_queue(struct net_device *dev,
237 + struct sk_buff *skb)
239 + u16 queue_index = 0;
242 + if (likely(dev->real_num_tx_queues == 1))
245 + /* IMQ can be receiving ingress or engress packets. */
247 + /* Check first for if rx_queue is set */
248 + if (skb_rx_queue_recorded(skb)) {
249 + queue_index = skb_get_rx_queue(skb);
253 + /* Check if socket has tx_queue set */
254 + if (sk_tx_queue_recorded(skb->sk)) {
255 + queue_index = sk_tx_queue_get(skb->sk);
259 + /* Try use socket hash */
260 + if (skb->sk && skb->sk->sk_hash) {
261 + hash = skb->sk->sk_hash;
263 + (u16)(((u64)hash * dev->real_num_tx_queues) >> 32);
267 + /* Generate hash from packet data */
268 + queue_index = imq_hash(dev, skb);
271 + if (unlikely(queue_index >= dev->real_num_tx_queues))
272 + queue_index = (u16)((u32)queue_index % dev->real_num_tx_queues);
274 + skb_set_queue_mapping(skb, queue_index);
275 + return netdev_get_tx_queue(dev, queue_index);
278 +static struct net_device_stats *imq_get_stats(struct net_device *dev)
280 + return &dev->stats;
283 +/* called for packets kfree'd in qdiscs at places other than enqueue */
284 +static void imq_skb_destructor(struct sk_buff *skb)
286 + struct nf_queue_entry *entry = skb->nf_queue_entry;
288 + skb->nf_queue_entry = NULL;
291 + nf_queue_entry_release_refs(entry);
295 + skb_restore_cb(skb); /* kfree backup */
298 +static void imq_done_check_queue_mapping(struct sk_buff *skb,
299 + struct net_device *dev)
301 + unsigned int queue_index;
303 + /* Don't let queue_mapping be left too large after exiting IMQ */
304 + if (likely(skb->dev != dev && skb->dev != NULL)) {
305 + queue_index = skb_get_queue_mapping(skb);
306 + if (unlikely(queue_index >= skb->dev->real_num_tx_queues)) {
307 + queue_index = (u16)((u32)queue_index %
308 + skb->dev->real_num_tx_queues);
309 + skb_set_queue_mapping(skb, queue_index);
312 + /* skb->dev was IMQ device itself or NULL, be on safe side and
313 + * just clear queue mapping.
315 + skb_set_queue_mapping(skb, 0);
319 +static netdev_tx_t imq_dev_xmit(struct sk_buff *skb, struct net_device *dev)
321 + struct nf_queue_entry *entry = skb->nf_queue_entry;
325 + skb->nf_queue_entry = NULL;
326 + netif_trans_update(dev);
328 + dev->stats.tx_bytes += skb->len;
329 + dev->stats.tx_packets++;
331 + if (unlikely(entry == NULL)) {
332 + /* We don't know what is going on here.. packet is queued for
333 + * imq device, but (probably) not by us.
335 + * If this packet was not send here by imq_nf_queue(), then
336 + * skb_save_cb() was not used and skb_free() should not show:
337 + * WARNING: IMQ: kfree_skb: skb->cb_next:..
339 + * WARNING: IMQ: kfree_skb: skb->nf_queue_entry...
341 + * However if this message is shown, then IMQ is somehow broken
342 + * and you should report this to linuximq.net.
345 + /* imq_dev_xmit is black hole that eats all packets, report that
346 + * we eat this packet happily and increase dropped counters.
349 + dev->stats.tx_dropped++;
350 + dev_kfree_skb(skb);
353 + return NETDEV_TX_OK;
356 + skb_restore_cb(skb); /* restore skb->cb */
358 + skb->imq_flags = 0;
359 + skb->destructor = NULL;
361 + imq_done_check_queue_mapping(skb, dev);
363 + nf_reinject(entry, NF_ACCEPT);
366 + return NETDEV_TX_OK;
369 +static struct net_device *get_imq_device_by_index(int index)
371 + struct net_device *dev = NULL;
375 + /* get device by name and cache result */
376 + snprintf(buf, sizeof(buf), "imq%d", index);
378 + /* Search device from all namespaces. */
379 + for_each_net(net) {
380 + dev = dev_get_by_name(net, buf);
385 + if (WARN_ON_ONCE(dev == NULL)) {
386 + /* IMQ device not found. Exotic config? */
387 + return ERR_PTR(-ENODEV);
390 + imq_devs_cache[index] = dev;
396 +static struct nf_queue_entry *nf_queue_entry_dup(struct nf_queue_entry *e)
398 + struct nf_queue_entry *entry = kmemdup(e, e->size, GFP_ATOMIC);
400 + nf_queue_entry_get_refs(entry);
406 +#ifdef CONFIG_BRIDGE_NETFILTER
407 +/* When called from bridge netfilter, skb->data must point to MAC header
408 + * before calling skb_gso_segment(). Else, original MAC header is lost
409 + * and segmented skbs will be sent to wrong destination.
411 +static void nf_bridge_adjust_skb_data(struct sk_buff *skb)
413 + if (skb->nf_bridge)
414 + __skb_push(skb, skb->network_header - skb->mac_header);
417 +static void nf_bridge_adjust_segmented_data(struct sk_buff *skb)
419 + if (skb->nf_bridge)
420 + __skb_pull(skb, skb->network_header - skb->mac_header);
423 +#define nf_bridge_adjust_skb_data(s) do {} while (0)
424 +#define nf_bridge_adjust_segmented_data(s) do {} while (0)
427 +static void free_entry(struct nf_queue_entry *entry)
429 + nf_queue_entry_release_refs(entry);
433 +static int __imq_nf_queue(struct nf_queue_entry *entry, struct net_device *dev);
435 +static int __imq_nf_queue_gso(struct nf_queue_entry *entry,
436 + struct net_device *dev, struct sk_buff *skb)
439 + struct nf_queue_entry *entry_seg;
441 + nf_bridge_adjust_segmented_data(skb);
443 + if (skb->next == NULL) { /* last packet, no need to copy entry */
444 + struct sk_buff *gso_skb = entry->skb;
446 + ret = __imq_nf_queue(entry, dev);
448 + entry->skb = gso_skb;
454 + entry_seg = nf_queue_entry_dup(entry);
456 + entry_seg->skb = skb;
457 + ret = __imq_nf_queue(entry_seg, dev);
459 + free_entry(entry_seg);
464 +static int imq_nf_queue(struct nf_queue_entry *entry, unsigned queue_num)
466 + struct sk_buff *skb, *segs;
467 + struct net_device *dev;
468 + unsigned int queued;
469 + int index, retval, err;
471 + index = entry->skb->imq_flags & IMQ_F_IFMASK;
472 + if (unlikely(index > numdevs - 1)) {
473 + if (net_ratelimit())
474 + pr_warn("IMQ: invalid device specified, highest is %u\n",
480 + /* check for imq device by index from cache */
481 + dev = imq_devs_cache[index];
482 + if (unlikely(!dev)) {
483 + dev = get_imq_device_by_index(index);
485 + retval = PTR_ERR(dev);
490 + if (unlikely(!(dev->flags & IFF_UP))) {
491 + entry->skb->imq_flags = 0;
492 + retval = -ECANCELED;
496 + /* Since 3.10.x, GSO handling moved here as result of upstream commit
497 + * a5fedd43d5f6c94c71053a66e4c3d2e35f1731a2 (netfilter: move
498 + * skb_gso_segment into nfnetlink_queue module).
500 + * Following code replicates the gso handling from
501 + * 'net/netfilter/nfnetlink_queue_core.c':nfqnl_enqueue_packet().
506 + switch (entry->state.pf) {
508 + skb->protocol = htons(ETH_P_IP);
511 + skb->protocol = htons(ETH_P_IPV6);
515 + if (!skb_is_gso(entry->skb))
516 + return __imq_nf_queue(entry, dev);
518 + nf_bridge_adjust_skb_data(skb);
519 + segs = skb_gso_segment(skb, 0);
520 + /* Does not use PTR_ERR to limit the number of error codes that can be
521 + * returned by nf_queue. For instance, callers rely on -ECANCELED to
522 + * mean 'ignore this hook'.
530 + struct sk_buff *nskb = segs->next;
531 + if (nskb && nskb->next)
532 + nskb->cb_next = NULL;
534 + err = __imq_nf_queue_gso(entry, dev, segs);
543 + if (err) /* some segments are already queued */
550 + nf_bridge_adjust_segmented_data(skb);
556 +static int __imq_nf_queue(struct nf_queue_entry *entry, struct net_device *dev)
558 + struct sk_buff *skb_orig, *skb, *skb_shared, *skb_popd;
560 + struct sk_buff *to_free = NULL;
561 + struct netdev_queue *txq;
562 + spinlock_t *root_lock;
564 + int retval = -EINVAL;
565 + unsigned int orig_queue_index;
567 + dev->last_rx = jiffies;
572 + /* skb has owner? => make clone */
573 + if (unlikely(skb->destructor)) {
575 + skb = skb_clone(skb, GFP_ATOMIC);
576 + if (unlikely(!skb)) {
580 + skb->cb_next = NULL;
584 + dev->stats.rx_bytes += skb->len;
585 + dev->stats.rx_packets++;
588 + /* skb->dev == NULL causes problems, try the find cause. */
589 + if (net_ratelimit()) {
590 + dev_warn(&dev->dev,
591 + "received packet with skb->dev == NULL\n");
598 + /* Disables softirqs for lock below */
599 + rcu_read_lock_bh();
601 + /* Multi-queue selection */
602 + orig_queue_index = skb_get_queue_mapping(skb);
603 + txq = imq_select_queue(dev, skb);
605 + q = rcu_dereference(txq->qdisc);
606 + if (unlikely(!q->enqueue))
607 + goto packet_not_eaten_by_imq_dev;
609 + skb->nf_queue_entry = entry;
610 + root_lock = qdisc_lock(q);
611 + spin_lock(root_lock);
613 + users = refcount_read(&skb->users);
615 + skb_shared = skb_get(skb); /* increase reference count by one */
617 + /* backup skb->cb, as qdisc layer will overwrite it */
618 + skb_save_cb(skb_shared);
619 + qdisc_enqueue_root(skb_shared, q, &to_free); /* might kfree_skb */
620 + if (likely(refcount_read(&skb_shared->users) == users + 1)) {
623 + kfree_skb(skb_shared); /* decrease reference count by one */
625 + skb->destructor = &imq_skb_destructor;
627 + skb_popd = qdisc_dequeue_skb(q, &validate);
630 + if (unlikely(skb_orig))
631 + kfree_skb(skb_orig); /* free original */
633 + spin_unlock(root_lock);
636 + /* schedule qdisc dequeue */
637 + __netif_schedule(q);
639 + if (likely(skb_popd)) {
640 + /* Note that we validate skb (GSO, checksum, ...) outside of locks */
642 + skb_popd = validate_xmit_skb_list(skb_popd, dev);
646 + int cpu = smp_processor_id(); /* ok because BHs are off */
648 + txq = skb_get_tx_queue(dev, skb_popd);
650 + IMQ device will not be frozen or stoped, and it always be successful.
651 + So we need not check its status and return value to accelerate.
653 + if (imq_dev_accurate_stats && txq->xmit_lock_owner != cpu) {
654 + HARD_TX_LOCK(dev, txq, cpu);
655 + if (!netif_xmit_frozen_or_stopped(txq)) {
656 + dev_hard_start_xmit(skb_popd, dev, txq, &dummy_ret);
658 + HARD_TX_UNLOCK(dev, txq);
660 + if (!netif_xmit_frozen_or_stopped(txq)) {
661 + dev_hard_start_xmit(skb_popd, dev, txq, &dummy_ret);
666 + /* No ready skb, then schedule it */
667 + __netif_schedule(q);
670 + rcu_read_unlock_bh();
674 + skb_restore_cb(skb_shared); /* restore skb->cb */
675 + skb->nf_queue_entry = NULL;
677 + * qdisc dropped packet and decreased skb reference count of
678 + * skb, so we don't really want to and try refree as that would
679 + * actually destroy the skb.
681 + spin_unlock(root_lock);
682 + goto packet_not_eaten_by_imq_dev;
685 +packet_not_eaten_by_imq_dev:
686 + skb_set_queue_mapping(skb, orig_queue_index);
687 + rcu_read_unlock_bh();
689 + /* cloned? restore original */
690 + if (unlikely(skb_orig)) {
692 + entry->skb = skb_orig;
696 + if (unlikely(to_free)) {
697 + kfree_skb_list(to_free);
701 +static unsigned int imq_nf_hook(void *priv,
702 + struct sk_buff *skb,
703 + const struct nf_hook_state *state)
705 + return (skb->imq_flags & IMQ_F_ENQUEUE) ? NF_IMQ_QUEUE : NF_ACCEPT;
708 +static int imq_close(struct net_device *dev)
710 + netif_stop_queue(dev);
714 +static int imq_open(struct net_device *dev)
716 + netif_start_queue(dev);
720 +static struct device_type imq_device_type = {
724 +static const struct net_device_ops imq_netdev_ops = {
725 + .ndo_open = imq_open,
726 + .ndo_stop = imq_close,
727 + .ndo_start_xmit = imq_dev_xmit,
728 + .ndo_get_stats = imq_get_stats,
731 +static void imq_setup(struct net_device *dev)
733 + dev->netdev_ops = &imq_netdev_ops;
734 + dev->type = ARPHRD_VOID;
735 + dev->mtu = 16000; /* too small? */
736 + dev->tx_queue_len = 11000; /* too big? */
737 + dev->flags = IFF_NOARP;
738 + dev->features = NETIF_F_SG | NETIF_F_FRAGLIST |
739 + NETIF_F_GSO | NETIF_F_HW_CSUM |
741 + dev->priv_flags &= ~(IFF_XMIT_DST_RELEASE |
742 + IFF_TX_SKB_SHARING);
745 +static int imq_validate(struct nlattr *tb[], struct nlattr *data[],
746 + struct netlink_ext_ack *extack)
750 + if (tb[IFLA_ADDRESS]) {
751 + if (nla_len(tb[IFLA_ADDRESS]) != ETH_ALEN) {
755 + if (!is_valid_ether_addr(nla_data(tb[IFLA_ADDRESS]))) {
756 + ret = -EADDRNOTAVAIL;
762 + pr_warn("IMQ: imq_validate failed (%d)\n", ret);
766 +static struct rtnl_link_ops imq_link_ops __read_mostly = {
769 + .setup = imq_setup,
770 + .validate = imq_validate,
773 +static const struct nf_queue_handler imq_nfqh = {
774 + .outfn = imq_nf_queue,
777 +static int __net_init imq_nf_register(struct net *net)
779 + return nf_register_net_hooks(net, imq_ops,
780 + ARRAY_SIZE(imq_ops));
783 +static void __net_exit imq_nf_unregister(struct net *net)
785 + nf_unregister_net_hooks(net, imq_ops,
786 + ARRAY_SIZE(imq_ops));
789 +static struct pernet_operations imq_net_ops = {
790 + .init = imq_nf_register,
791 + .exit = imq_nf_unregister,
794 +static int __net_init imq_init_hooks(void)
797 + nf_register_queue_imq_handler(&imq_nfqh);
799 + ret = register_pernet_subsys(&imq_net_ops);
801 + nf_unregister_queue_imq_handler();
806 +#ifdef CONFIG_LOCKDEP
807 + static struct lock_class_key imq_netdev_addr_lock_key;
809 + static void __init imq_dev_set_lockdep_one(struct net_device *dev,
810 + struct netdev_queue *txq, void *arg)
813 + * the IMQ transmit locks can be taken recursively,
814 + * for example with one IMQ rule for input- and one for
815 + * output network devices in iptables!
816 + * until we find a better solution ignore them.
818 + lockdep_set_novalidate_class(&txq->_xmit_lock);
821 + static void imq_dev_set_lockdep_class(struct net_device *dev)
823 + lockdep_set_class_and_name(&dev->addr_list_lock,
824 + &imq_netdev_addr_lock_key, "_xmit_addr_IMQ");
825 + netdev_for_each_tx_queue(dev, imq_dev_set_lockdep_one, NULL);
828 + static inline void imq_dev_set_lockdep_class(struct net_device *dev)
833 +static int __init imq_init_one(int index)
835 + struct net_device *dev;
838 + dev = alloc_netdev_mq(0, "imq%d", NET_NAME_UNKNOWN, imq_setup, numqueues);
842 + ret = dev_alloc_name(dev, dev->name);
846 + dev->rtnl_link_ops = &imq_link_ops;
847 + SET_NETDEV_DEVTYPE(dev, &imq_device_type);
848 + ret = register_netdevice(dev);
852 + imq_dev_set_lockdep_class(dev);
860 +static int __init imq_init_devs(void)
864 + if (numdevs < 1 || numdevs > IMQ_MAX_DEVS) {
865 + pr_err("IMQ: numdevs has to be betweed 1 and %u\n",
870 + if (numqueues < 1 || numqueues > IMQ_MAX_QUEUES) {
871 + pr_err("IMQ: numqueues has to be betweed 1 and %u\n",
876 + get_random_bytes(&imq_hashrnd, sizeof(imq_hashrnd));
879 + err = __rtnl_link_register(&imq_link_ops);
881 + for (i = 0; i < numdevs && !err; i++)
882 + err = imq_init_one(i);
885 + __rtnl_link_unregister(&imq_link_ops);
886 + memset(imq_devs_cache, 0, sizeof(imq_devs_cache));
893 +static int __init imq_init_module(void)
897 +#if defined(CONFIG_IMQ_NUM_DEVS)
898 + BUILD_BUG_ON(CONFIG_IMQ_NUM_DEVS > 16);
899 + BUILD_BUG_ON(CONFIG_IMQ_NUM_DEVS < 2);
900 + BUILD_BUG_ON(CONFIG_IMQ_NUM_DEVS - 1 > IMQ_F_IFMASK);
903 + err = imq_init_devs();
905 + pr_err("IMQ: Error trying imq_init_devs(net)\n");
909 + err = imq_init_hooks();
911 + pr_err(KERN_ERR "IMQ: Error trying imq_init_hooks()\n");
912 + rtnl_link_unregister(&imq_link_ops);
913 + memset(imq_devs_cache, 0, sizeof(imq_devs_cache));
917 + pr_info("IMQ driver loaded successfully. (numdevs = %d, numqueues = %d, imq_dev_accurate_stats = %d)\n",
918 + numdevs, numqueues, imq_dev_accurate_stats);
920 +#if defined(CONFIG_IMQ_BEHAVIOR_BA) || defined(CONFIG_IMQ_BEHAVIOR_BB)
921 + pr_info("\tHooking IMQ before NAT on PREROUTING.\n");
923 + pr_info("\tHooking IMQ after NAT on PREROUTING.\n");
925 +#if defined(CONFIG_IMQ_BEHAVIOR_AB) || defined(CONFIG_IMQ_BEHAVIOR_BB)
926 + pr_info("\tHooking IMQ before NAT on POSTROUTING.\n");
928 + pr_info("\tHooking IMQ after NAT on POSTROUTING.\n");
934 +static void __exit imq_unhook(void)
936 + unregister_pernet_subsys(&imq_net_ops);
937 + nf_unregister_queue_imq_handler();
940 +static void __exit imq_cleanup_devs(void)
942 + rtnl_link_unregister(&imq_link_ops);
943 + memset(imq_devs_cache, 0, sizeof(imq_devs_cache));
946 +static void __exit imq_exit_module(void)
949 + imq_cleanup_devs();
950 + pr_info("IMQ driver unloaded successfully.\n");
953 +module_init(imq_init_module);
954 +module_exit(imq_exit_module);
956 +module_param(numdevs, int, 0);
957 +module_param(numqueues, int, 0);
958 +module_param(imq_dev_accurate_stats, int, 0);
959 +MODULE_PARM_DESC(numdevs, "number of IMQ devices (how many imq* devices will be created)");
960 +MODULE_PARM_DESC(numqueues, "number of queues per IMQ device");
961 +MODULE_PARM_DESC(imq_dev_accurate_stats, "Notify if need the accurate imq device stats");
963 +MODULE_AUTHOR("https://github.com/imq/linuximq");
964 +MODULE_DESCRIPTION("Pseudo-driver for the intermediate queue device. See https://github.com/imq/linuximq/wiki for more information.");
965 +MODULE_LICENSE("GPL");
966 +MODULE_ALIAS_RTNL_LINK("imq");
967 diff -Naupr linux-4.14_orig/drivers/net/Kconfig linux-4.14/drivers/net/Kconfig
968 --- linux-4.14_orig/drivers/net/Kconfig 2017-11-13 01:46:13.000000000 +0700
969 +++ linux-4.14/drivers/net/Kconfig 2017-11-13 11:46:45.844089945 +0700
970 @@ -277,6 +277,125 @@ config RIONET_RX_SIZE
975 + tristate "IMQ (intermediate queueing device) support"
976 + depends on NETDEVICES && NETFILTER
978 + The IMQ device(s) is used as placeholder for QoS queueing
979 + disciplines. Every packet entering/leaving the IP stack can be
980 + directed through the IMQ device where it's enqueued/dequeued to the
981 + attached qdisc. This allows you to treat network devices as classes
982 + and distribute bandwidth among them. Iptables is used to specify
983 + through which IMQ device, if any, packets travel.
985 + More information at: https://github.com/imq/linuximq
987 + To compile this driver as a module, choose M here: the module
988 + will be called imq. If unsure, say N.
991 + prompt "IMQ behavior (PRE/POSTROUTING)"
993 + default IMQ_BEHAVIOR_AB
995 + This setting defines how IMQ behaves in respect to its
996 + hooking in PREROUTING and POSTROUTING.
998 + IMQ can work in any of the following ways:
1000 + PREROUTING | POSTROUTING
1001 + -----------------|-------------------
1002 + #1 After NAT | After NAT
1003 + #2 After NAT | Before NAT
1004 + #3 Before NAT | After NAT
1005 + #4 Before NAT | Before NAT
1007 + The default behavior is to hook before NAT on PREROUTING
1008 + and after NAT on POSTROUTING (#3).
1010 + This settings are specially usefull when trying to use IMQ
1011 + to shape NATed clients.
1013 + More information can be found at: https://github.com/imq/linuximq
1015 + If not sure leave the default settings alone.
1017 +config IMQ_BEHAVIOR_AA
1020 + This setting defines how IMQ behaves in respect to its
1021 + hooking in PREROUTING and POSTROUTING.
1023 + Choosing this option will make IMQ hook like this:
1025 + PREROUTING: After NAT
1026 + POSTROUTING: After NAT
1028 + More information can be found at: https://github.com/imq/linuximq
1030 + If not sure leave the default settings alone.
1032 +config IMQ_BEHAVIOR_AB
1035 + This setting defines how IMQ behaves in respect to its
1036 + hooking in PREROUTING and POSTROUTING.
1038 + Choosing this option will make IMQ hook like this:
1040 + PREROUTING: After NAT
1041 + POSTROUTING: Before NAT
1043 + More information can be found at: https://github.com/imq/linuximq
1045 + If not sure leave the default settings alone.
1047 +config IMQ_BEHAVIOR_BA
1050 + This setting defines how IMQ behaves in respect to its
1051 + hooking in PREROUTING and POSTROUTING.
1053 + Choosing this option will make IMQ hook like this:
1055 + PREROUTING: Before NAT
1056 + POSTROUTING: After NAT
1058 + More information can be found at: https://github.com/imq/linuximq
1060 + If not sure leave the default settings alone.
1062 +config IMQ_BEHAVIOR_BB
1065 + This setting defines how IMQ behaves in respect to its
1066 + hooking in PREROUTING and POSTROUTING.
1068 + Choosing this option will make IMQ hook like this:
1070 + PREROUTING: Before NAT
1071 + POSTROUTING: Before NAT
1073 + More information can be found at: https://github.com/imq/linuximq
1075 + If not sure leave the default settings alone.
1079 +config IMQ_NUM_DEVS
1080 + int "Number of IMQ devices"
1085 + This setting defines how many IMQ devices will be created.
1087 + The default value is 16.
1089 + More information can be found at: https://github.com/imq/linuximq
1091 + If not sure leave the default settings alone.
1094 tristate "Universal TUN/TAP device driver support"
1096 diff -Naupr linux-4.14_orig/drivers/net/Makefile linux-4.14/drivers/net/Makefile
1097 --- linux-4.14_orig/drivers/net/Makefile 2017-11-13 01:46:13.000000000 +0700
1098 +++ linux-4.14/drivers/net/Makefile 2017-11-13 11:46:45.844089945 +0700
1099 @@ -13,6 +13,7 @@ obj-$(CONFIG_DUMMY) += dummy.o
1100 obj-$(CONFIG_EQUALIZER) += eql.o
1101 obj-$(CONFIG_IFB) += ifb.o
1102 obj-$(CONFIG_MACSEC) += macsec.o
1103 +obj-$(CONFIG_IMQ) += imq.o
1104 obj-$(CONFIG_MACVLAN) += macvlan.o
1105 obj-$(CONFIG_MACVTAP) += macvtap.o
1106 obj-$(CONFIG_MII) += mii.o
1107 diff -Naupr linux-4.14_orig/include/linux/imq.h linux-4.14/include/linux/imq.h
1108 --- linux-4.14_orig/include/linux/imq.h 1970-01-01 07:00:00.000000000 +0700
1109 +++ linux-4.14/include/linux/imq.h 2017-11-13 11:46:45.844089945 +0700
1114 +/* IFMASK (16 device indexes, 0 to 15) and flag(s) fit in 5 bits */
1115 +#define IMQ_F_BITS 5
1117 +#define IMQ_F_IFMASK 0x0f
1118 +#define IMQ_F_ENQUEUE 0x10
1120 +#define IMQ_MAX_DEVS (IMQ_F_IFMASK + 1)
1122 +#endif /* _IMQ_H */
1124 diff -Naupr linux-4.14_orig/include/linux/netdevice.h linux-4.14/include/linux/netdevice.h
1125 --- linux-4.14_orig/include/linux/netdevice.h 2017-11-13 01:46:13.000000000 +0700
1126 +++ linux-4.14/include/linux/netdevice.h 2017-11-13 11:46:45.844089945 +0700
1127 @@ -1771,6 +1771,11 @@ struct net_device {
1129 * Cache lines mostly used on receive path (including eth_type_trans())
1132 +#if defined(CONFIG_IMQ) || defined(CONFIG_IMQ_MODULE)
1133 + unsigned long last_rx;
1136 /* Interface address info used in eth_type_trans() */
1137 unsigned char *dev_addr;
1139 @@ -3631,6 +3636,19 @@ static inline void netif_tx_unlock_bh(st
1143 +#define HARD_TX_LOCK_BH(dev, txq) { \
1144 + if ((dev->features & NETIF_F_LLTX) == 0) { \
1145 + __netif_tx_lock_bh(txq); \
1149 +#define HARD_TX_UNLOCK_BH(dev, txq) { \
1150 + if ((dev->features & NETIF_F_LLTX) == 0) { \
1151 + __netif_tx_unlock_bh(txq); \
1156 static inline void netif_tx_disable(struct net_device *dev)
1159 diff -Naupr linux-4.14_orig/include/linux/netfilter/xt_IMQ.h linux-4.14/include/linux/netfilter/xt_IMQ.h
1160 --- linux-4.14_orig/include/linux/netfilter/xt_IMQ.h 1970-01-01 07:00:00.000000000 +0700
1161 +++ linux-4.14/include/linux/netfilter/xt_IMQ.h 2017-11-13 11:46:45.847423298 +0700
1166 +struct xt_imq_info {
1167 + unsigned int todev; /* target imq device */
1170 +#endif /* _XT_IMQ_H */
1172 diff -Naupr linux-4.14_orig/include/linux/netfilter_ipv4/ipt_IMQ.h linux-4.14/include/linux/netfilter_ipv4/ipt_IMQ.h
1173 --- linux-4.14_orig/include/linux/netfilter_ipv4/ipt_IMQ.h 1970-01-01 07:00:00.000000000 +0700
1174 +++ linux-4.14/include/linux/netfilter_ipv4/ipt_IMQ.h 2017-11-13 11:46:45.847423298 +0700
1179 +/* Backwards compatibility for old userspace */
1180 +#include <linux/netfilter/xt_IMQ.h>
1182 +#define ipt_imq_info xt_imq_info
1184 +#endif /* _IPT_IMQ_H */
1186 diff -Naupr linux-4.14_orig/include/linux/netfilter_ipv6/ip6t_IMQ.h linux-4.14/include/linux/netfilter_ipv6/ip6t_IMQ.h
1187 --- linux-4.14_orig/include/linux/netfilter_ipv6/ip6t_IMQ.h 1970-01-01 07:00:00.000000000 +0700
1188 +++ linux-4.14/include/linux/netfilter_ipv6/ip6t_IMQ.h 2017-11-13 11:46:45.847423298 +0700
1190 +#ifndef _IP6T_IMQ_H
1191 +#define _IP6T_IMQ_H
1193 +/* Backwards compatibility for old userspace */
1194 +#include <linux/netfilter/xt_IMQ.h>
1196 +#define ip6t_imq_info xt_imq_info
1198 +#endif /* _IP6T_IMQ_H */
1200 diff -Naupr linux-4.14_orig/include/linux/skbuff.h linux-4.14/include/linux/skbuff.h
1201 --- linux-4.14_orig/include/linux/skbuff.h 2017-11-13 01:46:13.000000000 +0700
1202 +++ linux-4.14/include/linux/skbuff.h 2017-11-13 11:46:45.847423298 +0700
1204 #include <linux/in6.h>
1205 #include <linux/if_packet.h>
1206 #include <net/flow.h>
1207 +#if defined(CONFIG_IMQ) || defined(CONFIG_IMQ_MODULE)
1208 +#include <linux/imq.h>
1212 /* The interface for checksum offload between the stack and networking drivers
1214 @@ -581,7 +585,7 @@ typedef unsigned int sk_buff_data_t;
1215 typedef unsigned char *sk_buff_data_t;
1220 * struct sk_buff - socket buffer
1221 * @next: Next buffer in list
1222 * @prev: Previous buffer in list
1223 @@ -684,6 +688,9 @@ struct sk_buff {
1224 * first. This is owned by whoever has the skb queued ATM.
1226 char cb[48] __aligned(8);
1227 +#if defined(CONFIG_IMQ) || defined(CONFIG_IMQ_MODULE)
1231 unsigned long _skb_refdst;
1232 void (*destructor)(struct sk_buff *skb);
1233 @@ -693,6 +700,9 @@ struct sk_buff {
1234 #if defined(CONFIG_NF_CONNTRACK) || defined(CONFIG_NF_CONNTRACK_MODULE)
1235 unsigned long _nfct;
1237 +#if defined(CONFIG_IMQ) || defined(CONFIG_IMQ_MODULE)
1238 + struct nf_queue_entry *nf_queue_entry;
1240 #if IS_ENABLED(CONFIG_BRIDGE_NETFILTER)
1241 struct nf_bridge_info *nf_bridge;
1243 @@ -772,6 +782,9 @@ struct sk_buff {
1244 #ifdef CONFIG_NET_SWITCHDEV
1245 __u8 offload_fwd_mark:1;
1247 +#if defined(CONFIG_IMQ) || defined(CONFIG_IMQ_MODULE)
1248 + __u8 imq_flags:IMQ_F_BITS;
1250 #ifdef CONFIG_NET_CLS_ACT
1251 __u8 tc_skip_classify:1;
1252 __u8 tc_at_ingress:1;
1253 @@ -870,7 +883,7 @@ static inline bool skb_pfmemalloc(const
1255 static inline struct dst_entry *skb_dst(const struct sk_buff *skb)
1257 - /* If refdst was not refcounted, check we still are in a
1258 + /* If refdst was not refcounted, check we still are in a
1259 * rcu_read_lock section
1261 WARN_ON((skb->_skb_refdst & SKB_DST_NOREF) &&
1262 @@ -960,6 +973,12 @@ void skb_tx_error(struct sk_buff *skb);
1263 void consume_skb(struct sk_buff *skb);
1264 void __consume_stateless_skb(struct sk_buff *skb);
1265 void __kfree_skb(struct sk_buff *skb);
1267 +#if defined(CONFIG_IMQ) || defined(CONFIG_IMQ_MODULE)
1268 +int skb_save_cb(struct sk_buff *skb);
1269 +int skb_restore_cb(struct sk_buff *skb);
1272 extern struct kmem_cache *skbuff_head_cache;
1274 void kfree_skb_partial(struct sk_buff *skb, bool head_stolen);
1275 @@ -3785,8 +3804,12 @@ static inline void __nf_copy(struct sk_b
1276 dst->_nfct = src->_nfct;
1277 nf_conntrack_get(skb_nfct(src));
1279 +#if defined(CONFIG_IMQ) || defined(CONFIG_IMQ_MODULE)
1280 + dst->imq_flags = src->imq_flags;
1281 + dst->nf_queue_entry = src->nf_queue_entry;
1283 #if IS_ENABLED(CONFIG_BRIDGE_NETFILTER)
1284 - dst->nf_bridge = src->nf_bridge;
1285 + dst->nf_bridge = src->nf_bridge;
1286 nf_bridge_get(src->nf_bridge);
1288 #if IS_ENABLED(CONFIG_NETFILTER_XT_TARGET_TRACE) || defined(CONFIG_NF_TABLES)
1289 diff -Naupr linux-4.14_orig/include/net/netfilter/nf_queue.h linux-4.14/include/net/netfilter/nf_queue.h
1290 --- linux-4.14_orig/include/net/netfilter/nf_queue.h 2017-11-13 01:46:13.000000000 +0700
1291 +++ linux-4.14/include/net/netfilter/nf_queue.h 2017-11-13 11:46:45.847423298 +0700
1292 @@ -31,6 +31,12 @@ struct nf_queue_handler {
1293 void nf_register_queue_handler(struct net *net, const struct nf_queue_handler *qh);
1294 void nf_unregister_queue_handler(struct net *net);
1295 void nf_reinject(struct nf_queue_entry *entry, unsigned int verdict);
1296 +void nf_queue_entry_release_refs(struct nf_queue_entry *entry);
1298 +#if defined(CONFIG_IMQ) || defined(CONFIG_IMQ_MODULE)
1299 +void nf_register_queue_imq_handler(const struct nf_queue_handler *qh);
1300 +void nf_unregister_queue_imq_handler(void);
1303 void nf_queue_entry_get_refs(struct nf_queue_entry *entry);
1304 void nf_queue_entry_release_refs(struct nf_queue_entry *entry);
1305 diff -Naupr linux-4.14_orig/include/net/pkt_sched.h linux-4.14/include/net/pkt_sched.h
1306 --- linux-4.14_orig/include/net/pkt_sched.h 2017-11-13 01:46:13.000000000 +0700
1307 +++ linux-4.14/include/net/pkt_sched.h 2017-11-13 11:46:45.850756651 +0700
1308 @@ -109,6 +109,8 @@ int sch_direct_xmit(struct sk_buff *skb,
1310 void __qdisc_run(struct Qdisc *q);
1312 +struct sk_buff *qdisc_dequeue_skb(struct Qdisc *q, bool *validate);
1314 static inline void qdisc_run(struct Qdisc *q)
1316 if (qdisc_run_begin(q))
1317 diff -Naupr linux-4.14_orig/include/net/sch_generic.h linux-4.14/include/net/sch_generic.h
1318 --- linux-4.14_orig/include/net/sch_generic.h 2017-11-13 01:46:13.000000000 +0700
1319 +++ linux-4.14/include/net/sch_generic.h 2017-11-13 11:46:45.850756651 +0700
1320 @@ -567,6 +567,13 @@ static inline int qdisc_enqueue(struct s
1321 return sch->enqueue(skb, sch, to_free);
1324 +static inline int qdisc_enqueue_root(struct sk_buff *skb, struct Qdisc *sch,
1325 + struct sk_buff **to_free)
1327 + qdisc_skb_cb(skb)->pkt_len = skb->len;
1328 + return qdisc_enqueue(skb, sch, to_free) & NET_XMIT_MASK;
1331 static inline bool qdisc_is_percpu_stats(const struct Qdisc *q)
1333 return q->flags & TCQ_F_CPUSTATS;
1334 diff -Naupr linux-4.14_orig/include/uapi/linux/netfilter.h linux-4.14/include/uapi/linux/netfilter.h
1335 --- linux-4.14_orig/include/uapi/linux/netfilter.h 2017-11-13 01:46:13.000000000 +0700
1336 +++ linux-4.14/include/uapi/linux/netfilter.h 2017-11-13 11:46:45.850756651 +0700
1340 #define NF_STOP 5 /* Deprecated, for userspace nf_queue compatibility. */
1341 -#define NF_MAX_VERDICT NF_STOP
1342 +#define NF_IMQ_QUEUE 6
1343 +#define NF_MAX_VERDICT NF_IMQ_QUEUE
1345 /* we overload the higher bits for encoding auxiliary data such as the queue
1346 * number or errno values. Not nice, but better than additional function
1347 diff -Naupr linux-4.14_orig/net/core/dev.c linux-4.14/net/core/dev.c
1348 --- linux-4.14_orig/net/core/dev.c 2017-11-13 01:46:13.000000000 +0700
1349 +++ linux-4.14/net/core/dev.c 2017-11-13 11:46:45.854090004 +0700
1351 #include <linux/hrtimer.h>
1352 #include <linux/netfilter_ingress.h>
1353 #include <linux/crash_dump.h>
1354 +#if defined(CONFIG_IMQ) || defined(CONFIG_IMQ_MODULE)
1355 +#include <linux/imq.h>
1357 #include <linux/sctp.h>
1358 #include <net/udp_tunnel.h>
1360 @@ -2971,7 +2974,12 @@ static int xmit_one(struct sk_buff *skb,
1364 +#if defined(CONFIG_IMQ) || defined(CONFIG_IMQ_MODULE)
1365 + if ((!list_empty(&ptype_all) || !list_empty(&dev->ptype_all)) &&
1366 + !(skb->imq_flags & IMQ_F_ENQUEUE))
1368 if (!list_empty(&ptype_all) || !list_empty(&dev->ptype_all))
1370 dev_queue_xmit_nit(skb, dev);
1373 @@ -3010,6 +3018,8 @@ out:
1377 +EXPORT_SYMBOL_GPL(dev_hard_start_xmit);
1379 static struct sk_buff *validate_xmit_vlan(struct sk_buff *skb,
1380 netdev_features_t features)
1382 diff -Naupr linux-4.14_orig/net/core/skbuff.c linux-4.14/net/core/skbuff.c
1383 --- linux-4.14_orig/net/core/skbuff.c 2017-11-13 01:46:13.000000000 +0700
1384 +++ linux-4.14/net/core/skbuff.c 2017-11-13 11:46:45.854090004 +0700
1385 @@ -82,6 +82,87 @@ struct kmem_cache *skbuff_head_cache __r
1386 static struct kmem_cache *skbuff_fclone_cache __read_mostly;
1387 int sysctl_max_skb_frags __read_mostly = MAX_SKB_FRAGS;
1388 EXPORT_SYMBOL(sysctl_max_skb_frags);
1389 +#if defined(CONFIG_IMQ) || defined(CONFIG_IMQ_MODULE)
1390 +static struct kmem_cache *skbuff_cb_store_cache __read_mostly;
1393 +#if defined(CONFIG_IMQ) || defined(CONFIG_IMQ_MODULE)
1394 +/* Control buffer save/restore for IMQ devices */
1395 +struct skb_cb_table {
1396 + char cb[48] __aligned(8);
1401 +static DEFINE_SPINLOCK(skb_cb_store_lock);
1403 +int skb_save_cb(struct sk_buff *skb)
1405 + struct skb_cb_table *next;
1407 + next = kmem_cache_alloc(skbuff_cb_store_cache, GFP_ATOMIC);
1411 + BUILD_BUG_ON(sizeof(skb->cb) != sizeof(next->cb));
1413 + memcpy(next->cb, skb->cb, sizeof(skb->cb));
1414 + next->cb_next = skb->cb_next;
1416 + atomic_set(&next->refcnt, 1);
1418 + skb->cb_next = next;
1421 +EXPORT_SYMBOL(skb_save_cb);
1423 +int skb_restore_cb(struct sk_buff *skb)
1425 + struct skb_cb_table *next;
1427 + if (!skb->cb_next)
1430 + next = skb->cb_next;
1432 + BUILD_BUG_ON(sizeof(skb->cb) != sizeof(next->cb));
1434 + memcpy(skb->cb, next->cb, sizeof(skb->cb));
1435 + skb->cb_next = next->cb_next;
1437 + spin_lock(&skb_cb_store_lock);
1439 + if (atomic_dec_and_test(&next->refcnt))
1440 + kmem_cache_free(skbuff_cb_store_cache, next);
1442 + spin_unlock(&skb_cb_store_lock);
1446 +EXPORT_SYMBOL(skb_restore_cb);
1448 +static void skb_copy_stored_cb(struct sk_buff * , const struct sk_buff * ) __attribute__ ((unused));
1449 +static void skb_copy_stored_cb(struct sk_buff *new, const struct sk_buff *__old)
1451 + struct skb_cb_table *next;
1452 + struct sk_buff *old;
1454 + if (!__old->cb_next) {
1455 + new->cb_next = NULL;
1459 + spin_lock(&skb_cb_store_lock);
1461 + old = (struct sk_buff *)__old;
1463 + next = old->cb_next;
1464 + atomic_inc(&next->refcnt);
1465 + new->cb_next = next;
1467 + spin_unlock(&skb_cb_store_lock);
1472 * skb_panic - private function for out-of-line support
1473 @@ -615,6 +696,28 @@ void skb_release_head_state(struct sk_bu
1475 skb->destructor(skb);
1477 +#if defined(CONFIG_IMQ) || defined(CONFIG_IMQ_MODULE)
1479 + * This should not happen. When it does, avoid memleak by restoring
1480 + * the chain of cb-backups.
1482 + while (skb->cb_next != NULL) {
1483 + if (net_ratelimit())
1484 + pr_warn("IMQ: kfree_skb: skb->cb_next: %08x\n",
1485 + (unsigned int)(uintptr_t)skb->cb_next);
1487 + skb_restore_cb(skb);
1490 + * This should not happen either, nf_queue_entry is nullified in
1491 + * imq_dev_xmit(). If we have non-NULL nf_queue_entry then we are
1492 + * leaking entry pointers, maybe memory. We don't know if this is
1493 + * pointer to already freed memory, or should this be freed.
1494 + * If this happens we need to add refcounting, etc for nf_queue_entry.
1496 + if (skb->nf_queue_entry && net_ratelimit())
1497 + pr_warn("%s\n", "IMQ: kfree_skb: skb->nf_queue_entry != NULL");
1499 #if IS_ENABLED(CONFIG_NF_CONNTRACK)
1500 nf_conntrack_put(skb_nfct(skb));
1502 @@ -804,6 +907,10 @@ static void __copy_skb_header(struct sk_
1503 new->sp = secpath_get(old->sp);
1505 __nf_copy(new, old, false);
1506 +#if defined(CONFIG_IMQ) || defined(CONFIG_IMQ_MODULE)
1507 + new->cb_next = NULL;
1508 + /*skb_copy_stored_cb(new, old);*/
1511 /* Note : this field could be in headers_start/headers_end section
1512 * It is not yet because we do not want to have a 16 bit hole
1513 @@ -3902,6 +4009,13 @@ void __init skb_init(void)
1515 SLAB_HWCACHE_ALIGN|SLAB_PANIC,
1517 +#if defined(CONFIG_IMQ) || defined(CONFIG_IMQ_MODULE)
1518 + skbuff_cb_store_cache = kmem_cache_create("skbuff_cb_store_cache",
1519 + sizeof(struct skb_cb_table),
1521 + SLAB_HWCACHE_ALIGN|SLAB_PANIC,
1527 diff -Naupr linux-4.14_orig/net/netfilter/core.c linux-4.14/net/netfilter/core.c
1528 --- linux-4.14_orig/net/netfilter/core.c 2017-11-13 01:46:13.000000000 +0700
1529 +++ linux-4.14/net/netfilter/core.c 2017-11-13 14:16:05.896850774 +0700
1530 @@ -474,6 +474,11 @@ int nf_hook_slow(struct sk_buff *skb, st
1534 + case NF_IMQ_QUEUE:
1535 + ret = nf_queue(skb, state, e, s, verdict);
1536 + if (ret == -ECANCELED)
1540 ret = nf_queue(skb, state, e, s, verdict);
1542 diff -Naupr linux-4.14_orig/net/netfilter/Kconfig linux-4.14/net/netfilter/Kconfig
1543 --- linux-4.14_orig/net/netfilter/Kconfig 2017-11-13 01:46:13.000000000 +0700
1544 +++ linux-4.14/net/netfilter/Kconfig 2017-11-13 11:46:45.857423358 +0700
1545 @@ -867,6 +867,18 @@ config NETFILTER_XT_TARGET_LOG
1547 To compile it as a module, choose M here. If unsure, say N.
1549 +config NETFILTER_XT_TARGET_IMQ
1550 + tristate '"IMQ" target support'
1551 + depends on NETFILTER_XTABLES
1552 + depends on IP_NF_MANGLE || IP6_NF_MANGLE
1554 + default m if NETFILTER_ADVANCED=n
1556 + This option adds a `IMQ' target which is used to specify if and
1557 + to which imq device packets should get enqueued/dequeued.
1559 + To compile it as a module, choose M here. If unsure, say N.
1561 config NETFILTER_XT_TARGET_MARK
1562 tristate '"MARK" target support'
1563 depends on NETFILTER_ADVANCED
1564 diff -Naupr linux-4.14_orig/net/netfilter/Makefile linux-4.14/net/netfilter/Makefile
1565 --- linux-4.14_orig/net/netfilter/Makefile 2017-11-13 01:46:13.000000000 +0700
1566 +++ linux-4.14/net/netfilter/Makefile 2017-11-13 11:46:45.857423358 +0700
1567 @@ -125,6 +125,7 @@ obj-$(CONFIG_NETFILTER_XT_TARGET_CT) +=
1568 obj-$(CONFIG_NETFILTER_XT_TARGET_DSCP) += xt_DSCP.o
1569 obj-$(CONFIG_NETFILTER_XT_TARGET_HL) += xt_HL.o
1570 obj-$(CONFIG_NETFILTER_XT_TARGET_HMARK) += xt_HMARK.o
1571 +obj-$(CONFIG_NETFILTER_XT_TARGET_IMQ) += xt_IMQ.o
1572 obj-$(CONFIG_NETFILTER_XT_TARGET_LED) += xt_LED.o
1573 obj-$(CONFIG_NETFILTER_XT_TARGET_LOG) += xt_LOG.o
1574 obj-$(CONFIG_NETFILTER_XT_TARGET_NETMAP) += xt_NETMAP.o
1575 diff -Naupr linux-4.14_orig/net/netfilter/nf_queue.c linux-4.14/net/netfilter/nf_queue.c
1576 --- linux-4.14_orig/net/netfilter/nf_queue.c 2017-11-13 01:46:13.000000000 +0700
1577 +++ linux-4.14/net/netfilter/nf_queue.c 2017-11-13 14:25:21.436864671 +0700
1581 * Rusty Russell (C)2000 -- This code is GPL.
1582 * Patrick McHardy (c) 2006-2012
1585 * receives, no matter what.
1588 +#if defined(CONFIG_IMQ) || defined(CONFIG_IMQ_MODULE)
1589 +static const struct nf_queue_handler __rcu *queue_imq_handler __read_mostly;
1591 +void nf_register_queue_imq_handler(const struct nf_queue_handler *qh)
1593 + rcu_assign_pointer(queue_imq_handler, qh);
1595 +EXPORT_SYMBOL_GPL(nf_register_queue_imq_handler);
1597 +void nf_unregister_queue_imq_handler(void)
1599 + RCU_INIT_POINTER(queue_imq_handler, NULL);
1600 + synchronize_rcu();
1602 +EXPORT_SYMBOL_GPL(nf_unregister_queue_imq_handler);
1605 /* return EBUSY when somebody else is registered, return EEXIST if the
1606 * same handler is registered, return 0 in case of success. */
1607 void nf_register_queue_handler(struct net *net, const struct nf_queue_handler *qh)
1608 @@ -113,16 +130,29 @@ EXPORT_SYMBOL_GPL(nf_queue_nf_hook_drop)
1610 static int __nf_queue(struct sk_buff *skb, const struct nf_hook_state *state,
1611 const struct nf_hook_entries *entries,
1612 - unsigned int index, unsigned int queuenum)
1613 + unsigned int index, unsigned int verdict)
1615 int status = -ENOENT;
1616 struct nf_queue_entry *entry = NULL;
1617 const struct nf_afinfo *afinfo;
1618 const struct nf_queue_handler *qh;
1619 struct net *net = state->net;
1620 + unsigned int queuetype = verdict & NF_VERDICT_MASK;
1621 + unsigned int queuenum = verdict >> NF_VERDICT_QBITS;
1623 /* QUEUE == DROP if no one is waiting, to be safe. */
1624 - qh = rcu_dereference(net->nf.queue_handler);
1626 + if (queuetype == NF_IMQ_QUEUE) {
1627 +#if defined(CONFIG_IMQ) || defined(CONFIG_IMQ_MODULE)
1628 + qh = rcu_dereference(queue_imq_handler);
1634 + qh = rcu_dereference(net->nf.queue_handler);
1640 @@ -169,8 +199,16 @@ int nf_queue(struct sk_buff *skb, struct
1644 - ret = __nf_queue(skb, state, entries, index, verdict >> NF_VERDICT_QBITS);
1645 + ret = __nf_queue(skb, state, entries, index, verdict);
1648 +#if defined(CONFIG_IMQ) || defined(CONFIG_IMQ_MODULE)
1650 + if (ret == -ECANCELED && skb->imq_flags == 0) {
1655 if (ret == -ESRCH &&
1656 (verdict & NF_VERDICT_FLAG_QUEUE_BYPASS))
1658 @@ -256,6 +294,7 @@ next_hook:
1662 + case NF_IMQ_QUEUE:
1663 err = nf_queue(skb, &entry->state, hooks, i, verdict);
1666 diff -Naupr linux-4.14_orig/net/netfilter/xt_IMQ.c linux-4.14/net/netfilter/xt_IMQ.c
1667 --- linux-4.14_orig/net/netfilter/xt_IMQ.c 1970-01-01 07:00:00.000000000 +0700
1668 +++ linux-4.14/net/netfilter/xt_IMQ.c 2017-11-13 11:46:45.857423358 +0700
1671 + * This target marks packets to be enqueued to an imq device
1673 +#include <linux/module.h>
1674 +#include <linux/skbuff.h>
1675 +#include <linux/netfilter/x_tables.h>
1676 +#include <linux/netfilter/xt_IMQ.h>
1677 +#include <linux/imq.h>
1679 +static unsigned int imq_target(struct sk_buff *pskb,
1680 + const struct xt_action_param *par)
1682 + const struct xt_imq_info *mr = par->targinfo;
1684 + pskb->imq_flags = (mr->todev & IMQ_F_IFMASK) | IMQ_F_ENQUEUE;
1686 + return XT_CONTINUE;
1689 +static int imq_checkentry(const struct xt_tgchk_param *par)
1691 + struct xt_imq_info *mr = par->targinfo;
1693 + if (mr->todev > IMQ_MAX_DEVS - 1) {
1694 + pr_warn("IMQ: invalid device specified, highest is %u\n",
1695 + IMQ_MAX_DEVS - 1);
1702 +static struct xt_target xt_imq_reg[] __read_mostly = {
1705 + .family = AF_INET,
1706 + .checkentry = imq_checkentry,
1707 + .target = imq_target,
1708 + .targetsize = sizeof(struct xt_imq_info),
1709 + .table = "mangle",
1714 + .family = AF_INET6,
1715 + .checkentry = imq_checkentry,
1716 + .target = imq_target,
1717 + .targetsize = sizeof(struct xt_imq_info),
1718 + .table = "mangle",
1723 +static int __init imq_init(void)
1725 + return xt_register_targets(xt_imq_reg, ARRAY_SIZE(xt_imq_reg));
1728 +static void __exit imq_fini(void)
1730 + xt_unregister_targets(xt_imq_reg, ARRAY_SIZE(xt_imq_reg));
1733 +module_init(imq_init);
1734 +module_exit(imq_fini);
1736 +MODULE_AUTHOR("https://github.com/imq/linuximq");
1737 +MODULE_DESCRIPTION("Pseudo-driver for the intermediate queue device. See https://github.com/imq/linuximq/wiki for more information.");
1738 +MODULE_LICENSE("GPL");
1739 +MODULE_ALIAS("ipt_IMQ");
1740 +MODULE_ALIAS("ip6t_IMQ");
1742 diff -Naupr linux-4.14_orig/net/sched/sch_generic.c linux-4.14/net/sched/sch_generic.c
1743 --- linux-4.14_orig/net/sched/sch_generic.c 2017-11-13 01:46:13.000000000 +0700
1744 +++ linux-4.14/net/sched/sch_generic.c 2017-11-13 11:46:45.857423358 +0700
1745 @@ -158,6 +158,14 @@ trace:
1749 +struct sk_buff *qdisc_dequeue_skb(struct Qdisc *q, bool *validate)
1753 + return dequeue_skb(q, validate, &packets);
1755 +EXPORT_SYMBOL(qdisc_dequeue_skb);
1758 * Transmit possibly several skbs, and handle the return status as
1759 * required. Owning running seqcount bit guarantees that