1 diff --git a/drivers/net/Kconfig b/drivers/net/Kconfig
2 index 95c32f2..93fada5 100644
3 --- a/drivers/net/Kconfig
4 +++ b/drivers/net/Kconfig
5 @@ -260,6 +260,125 @@ config RIONET_RX_SIZE
10 + tristate "IMQ (intermediate queueing device) support"
11 + depends on NETDEVICES && NETFILTER
13 + The IMQ device(s) is used as placeholder for QoS queueing
14 + disciplines. Every packet entering/leaving the IP stack can be
15 + directed through the IMQ device where it's enqueued/dequeued to the
16 + attached qdisc. This allows you to treat network devices as classes
17 + and distribute bandwidth among them. Iptables is used to specify
18 + through which IMQ device, if any, packets travel.
20 + More information at: https://github.com/imq/linuximq
22 + To compile this driver as a module, choose M here: the module
23 + will be called imq. If unsure, say N.
26 + prompt "IMQ behavior (PRE/POSTROUTING)"
28 + default IMQ_BEHAVIOR_AB
30 + This setting defines how IMQ behaves in respect to its
31 + hooking in PREROUTING and POSTROUTING.
33 + IMQ can work in any of the following ways:
35 + PREROUTING | POSTROUTING
36 + -----------------|-------------------
37 + #1 After NAT | After NAT
38 + #2 After NAT | Before NAT
39 + #3 Before NAT | After NAT
40 + #4 Before NAT | Before NAT
42 + The default behavior is to hook before NAT on PREROUTING
43 + and after NAT on POSTROUTING (#3).
45 + This settings are specially usefull when trying to use IMQ
46 + to shape NATed clients.
48 + More information can be found at: https://github.com/imq/linuximq
50 + If not sure leave the default settings alone.
52 +config IMQ_BEHAVIOR_AA
55 + This setting defines how IMQ behaves in respect to its
56 + hooking in PREROUTING and POSTROUTING.
58 + Choosing this option will make IMQ hook like this:
60 + PREROUTING: After NAT
61 + POSTROUTING: After NAT
63 + More information can be found at: https://github.com/imq/linuximq
65 + If not sure leave the default settings alone.
67 +config IMQ_BEHAVIOR_AB
70 + This setting defines how IMQ behaves in respect to its
71 + hooking in PREROUTING and POSTROUTING.
73 + Choosing this option will make IMQ hook like this:
75 + PREROUTING: After NAT
76 + POSTROUTING: Before NAT
78 + More information can be found at: https://github.com/imq/linuximq
80 + If not sure leave the default settings alone.
82 +config IMQ_BEHAVIOR_BA
85 + This setting defines how IMQ behaves in respect to its
86 + hooking in PREROUTING and POSTROUTING.
88 + Choosing this option will make IMQ hook like this:
90 + PREROUTING: Before NAT
91 + POSTROUTING: After NAT
93 + More information can be found at: https://github.com/imq/linuximq
95 + If not sure leave the default settings alone.
97 +config IMQ_BEHAVIOR_BB
100 + This setting defines how IMQ behaves in respect to its
101 + hooking in PREROUTING and POSTROUTING.
103 + Choosing this option will make IMQ hook like this:
105 + PREROUTING: Before NAT
106 + POSTROUTING: Before NAT
108 + More information can be found at: https://github.com/imq/linuximq
110 + If not sure leave the default settings alone.
115 + int "Number of IMQ devices"
120 + This setting defines how many IMQ devices will be created.
122 + The default value is 16.
124 + More information can be found at: https://github.com/imq/linuximq
126 + If not sure leave the default settings alone.
129 tristate "Universal TUN/TAP device driver support"
131 diff --git a/drivers/net/Makefile b/drivers/net/Makefile
132 index 7336cbd..d6d7ad4 100644
133 --- a/drivers/net/Makefile
134 +++ b/drivers/net/Makefile
135 @@ -11,6 +11,7 @@ obj-$(CONFIG_DUMMY) += dummy.o
136 obj-$(CONFIG_EQUALIZER) += eql.o
137 obj-$(CONFIG_IFB) += ifb.o
138 obj-$(CONFIG_MACSEC) += macsec.o
139 +obj-$(CONFIG_IMQ) += imq.o
140 obj-$(CONFIG_MACVLAN) += macvlan.o
141 obj-$(CONFIG_MACVTAP) += macvtap.o
142 obj-$(CONFIG_MII) += mii.o
143 diff --git a/drivers/net/imq.c b/drivers/net/imq.c
145 index 0000000..bc3b997
147 +++ b/drivers/net/imq.c
150 + * Pseudo-driver for the intermediate queue device.
152 + * This program is free software; you can redistribute it and/or
153 + * modify it under the terms of the GNU General Public License
154 + * as published by the Free Software Foundation; either version
155 + * 2 of the License, or (at your option) any later version.
157 + * Authors: Patrick McHardy, <kaber@trash.net>
159 + * The first version was written by Martin Devera, <devik@cdi.cz>
164 +#include <linux/module.h>
165 +#include <linux/kernel.h>
166 +#include <linux/moduleparam.h>
167 +#include <linux/list.h>
168 +#include <linux/skbuff.h>
169 +#include <linux/netdevice.h>
170 +#include <linux/etherdevice.h>
171 +#include <linux/rtnetlink.h>
172 +#include <linux/if_arp.h>
173 +#include <linux/netfilter.h>
174 +#include <linux/netfilter_ipv4.h>
175 +#if defined(CONFIG_IPV6) || defined(CONFIG_IPV6_MODULE)
176 + #include <linux/netfilter_ipv6.h>
178 +#include <linux/imq.h>
179 +#include <net/pkt_sched.h>
180 +#include <net/netfilter/nf_queue.h>
181 +#include <net/sock.h>
182 +#include <linux/ip.h>
183 +#include <linux/ipv6.h>
184 +#include <linux/if_vlan.h>
185 +#include <linux/if_pppox.h>
187 +#include <net/ipv6.h>
189 +static int imq_nf_queue(struct nf_queue_entry *entry, unsigned queue_num);
191 +static nf_hookfn imq_nf_hook;
193 +static struct nf_hook_ops imq_ops[] = {
195 + /* imq_ingress_ipv4 */
196 + .hook = imq_nf_hook,
198 + .hooknum = NF_INET_PRE_ROUTING,
199 +#if defined(CONFIG_IMQ_BEHAVIOR_BA) || defined(CONFIG_IMQ_BEHAVIOR_BB)
200 + .priority = NF_IP_PRI_MANGLE + 1,
202 + .priority = NF_IP_PRI_NAT_DST + 1,
206 + /* imq_egress_ipv4 */
207 + .hook = imq_nf_hook,
209 + .hooknum = NF_INET_POST_ROUTING,
210 +#if defined(CONFIG_IMQ_BEHAVIOR_AA) || defined(CONFIG_IMQ_BEHAVIOR_BA)
211 + .priority = NF_IP_PRI_LAST,
213 + .priority = NF_IP_PRI_NAT_SRC - 1,
216 +#if defined(CONFIG_IPV6) || defined(CONFIG_IPV6_MODULE)
218 + /* imq_ingress_ipv6 */
219 + .hook = imq_nf_hook,
221 + .hooknum = NF_INET_PRE_ROUTING,
222 +#if defined(CONFIG_IMQ_BEHAVIOR_BA) || defined(CONFIG_IMQ_BEHAVIOR_BB)
223 + .priority = NF_IP6_PRI_MANGLE + 1,
225 + .priority = NF_IP6_PRI_NAT_DST + 1,
229 + /* imq_egress_ipv6 */
230 + .hook = imq_nf_hook,
232 + .hooknum = NF_INET_POST_ROUTING,
233 +#if defined(CONFIG_IMQ_BEHAVIOR_AA) || defined(CONFIG_IMQ_BEHAVIOR_BA)
234 + .priority = NF_IP6_PRI_LAST,
236 + .priority = NF_IP6_PRI_NAT_SRC - 1,
242 +#if defined(CONFIG_IMQ_NUM_DEVS)
243 +static int numdevs = CONFIG_IMQ_NUM_DEVS;
245 +static int numdevs = IMQ_MAX_DEVS;
248 +static struct net_device *imq_devs_cache[IMQ_MAX_DEVS];
250 +#define IMQ_MAX_QUEUES 32
251 +static int numqueues = 1;
252 +static u32 imq_hashrnd;
253 +static int imq_dev_accurate_stats = 1;
255 +static inline __be16 pppoe_proto(const struct sk_buff *skb)
257 + return *((__be16 *)(skb_mac_header(skb) + ETH_HLEN +
258 + sizeof(struct pppoe_hdr)));
261 +static u16 imq_hash(struct net_device *dev, struct sk_buff *skb)
263 + unsigned int pull_len;
264 + u16 protocol = skb->protocol;
276 + switch (protocol) {
277 + case htons(ETH_P_8021Q): {
278 + if (unlikely(skb_pull(skb, VLAN_HLEN) == NULL))
281 + pull_len += VLAN_HLEN;
282 + skb->network_header += VLAN_HLEN;
284 + protocol = vlan_eth_hdr(skb)->h_vlan_encapsulated_proto;
288 + case htons(ETH_P_PPP_SES): {
289 + if (unlikely(skb_pull(skb, PPPOE_SES_HLEN) == NULL))
292 + pull_len += PPPOE_SES_HLEN;
293 + skb->network_header += PPPOE_SES_HLEN;
295 + protocol = pppoe_proto(skb);
299 + case htons(ETH_P_IP): {
300 + const struct iphdr *iph = ip_hdr(skb);
302 + if (unlikely(!pskb_may_pull(skb, sizeof(struct iphdr))))
305 + addr1 = iph->daddr;
306 + addr2 = iph->saddr;
308 + ip_proto = !(ip_hdr(skb)->frag_off & htons(IP_MF | IP_OFFSET)) ?
310 + ihl = ip_hdrlen(skb);
314 +#if defined(CONFIG_IPV6) || defined(CONFIG_IPV6_MODULE)
315 + case htons(ETH_P_IPV6): {
316 + const struct ipv6hdr *iph = ipv6_hdr(skb);
319 + if (unlikely(!pskb_may_pull(skb, sizeof(struct ipv6hdr))))
322 + addr1 = iph->daddr.s6_addr32[3];
323 + addr2 = iph->saddr.s6_addr32[3];
324 + ihl = ipv6_skip_exthdr(skb, sizeof(struct ipv6hdr), &ip_proto,
326 + if (unlikely(ihl < 0))
334 + if (pull_len != 0) {
335 + skb_push(skb, pull_len);
336 + skb->network_header -= pull_len;
339 + return (u16)(ntohs(protocol) % dev->real_num_tx_queues);
343 + swap(addr1, addr2);
345 + switch (ip_proto) {
352 + case IPPROTO_UDPLITE: {
353 + if (likely(skb_copy_bits(skb, ihl, &ports.in32, 4) >= 0)) {
354 + if (ports.in16[0] > ports.in16[1])
355 + swap(ports.in16[0], ports.in16[1]);
365 + if (pull_len != 0) {
366 + skb_push(skb, pull_len);
367 + skb->network_header -= pull_len;
370 + hash = jhash_3words(addr1, addr2, ports.in32, imq_hashrnd ^ ip_proto);
372 + return (u16)(((u64)hash * dev->real_num_tx_queues) >> 32);
375 +static inline bool sk_tx_queue_recorded(struct sock *sk)
377 + return (sk_tx_queue_get(sk) >= 0);
380 +static struct netdev_queue *imq_select_queue(struct net_device *dev,
381 + struct sk_buff *skb)
383 + u16 queue_index = 0;
386 + if (likely(dev->real_num_tx_queues == 1))
389 + /* IMQ can be receiving ingress or engress packets. */
391 + /* Check first for if rx_queue is set */
392 + if (skb_rx_queue_recorded(skb)) {
393 + queue_index = skb_get_rx_queue(skb);
397 + /* Check if socket has tx_queue set */
398 + if (sk_tx_queue_recorded(skb->sk)) {
399 + queue_index = sk_tx_queue_get(skb->sk);
403 + /* Try use socket hash */
404 + if (skb->sk && skb->sk->sk_hash) {
405 + hash = skb->sk->sk_hash;
407 + (u16)(((u64)hash * dev->real_num_tx_queues) >> 32);
411 + /* Generate hash from packet data */
412 + queue_index = imq_hash(dev, skb);
415 + if (unlikely(queue_index >= dev->real_num_tx_queues))
416 + queue_index = (u16)((u32)queue_index % dev->real_num_tx_queues);
418 + skb_set_queue_mapping(skb, queue_index);
419 + return netdev_get_tx_queue(dev, queue_index);
422 +static struct net_device_stats *imq_get_stats(struct net_device *dev)
424 + return &dev->stats;
427 +/* called for packets kfree'd in qdiscs at places other than enqueue */
428 +static void imq_skb_destructor(struct sk_buff *skb)
430 + struct nf_queue_entry *entry = skb->nf_queue_entry;
432 + skb->nf_queue_entry = NULL;
435 + nf_queue_entry_release_refs(entry);
439 + skb_restore_cb(skb); /* kfree backup */
442 +static void imq_done_check_queue_mapping(struct sk_buff *skb,
443 + struct net_device *dev)
445 + unsigned int queue_index;
447 + /* Don't let queue_mapping be left too large after exiting IMQ */
448 + if (likely(skb->dev != dev && skb->dev != NULL)) {
449 + queue_index = skb_get_queue_mapping(skb);
450 + if (unlikely(queue_index >= skb->dev->real_num_tx_queues)) {
451 + queue_index = (u16)((u32)queue_index %
452 + skb->dev->real_num_tx_queues);
453 + skb_set_queue_mapping(skb, queue_index);
456 + /* skb->dev was IMQ device itself or NULL, be on safe side and
457 + * just clear queue mapping.
459 + skb_set_queue_mapping(skb, 0);
463 +static netdev_tx_t imq_dev_xmit(struct sk_buff *skb, struct net_device *dev)
465 + struct nf_queue_entry *entry = skb->nf_queue_entry;
467 + skb->nf_queue_entry = NULL;
468 + netif_trans_update(dev);
470 + dev->stats.tx_bytes += skb->len;
471 + dev->stats.tx_packets++;
473 + if (unlikely(entry == NULL)) {
474 + /* We don't know what is going on here.. packet is queued for
475 + * imq device, but (probably) not by us.
477 + * If this packet was not send here by imq_nf_queue(), then
478 + * skb_save_cb() was not used and skb_free() should not show:
479 + * WARNING: IMQ: kfree_skb: skb->cb_next:..
481 + * WARNING: IMQ: kfree_skb: skb->nf_queue_entry...
483 + * However if this message is shown, then IMQ is somehow broken
484 + * and you should report this to linuximq.net.
487 + /* imq_dev_xmit is black hole that eats all packets, report that
488 + * we eat this packet happily and increase dropped counters.
491 + dev->stats.tx_dropped++;
492 + dev_kfree_skb(skb);
494 + return NETDEV_TX_OK;
497 + skb_restore_cb(skb); /* restore skb->cb */
499 + skb->imq_flags = 0;
500 + skb->destructor = NULL;
502 + imq_done_check_queue_mapping(skb, dev);
504 + nf_reinject(entry, NF_ACCEPT);
506 + return NETDEV_TX_OK;
509 +static struct net_device *get_imq_device_by_index(int index)
511 + struct net_device *dev = NULL;
515 + /* get device by name and cache result */
516 + snprintf(buf, sizeof(buf), "imq%d", index);
518 + /* Search device from all namespaces. */
519 + for_each_net(net) {
520 + dev = dev_get_by_name(net, buf);
525 + if (WARN_ON_ONCE(dev == NULL)) {
526 + /* IMQ device not found. Exotic config? */
527 + return ERR_PTR(-ENODEV);
530 + imq_devs_cache[index] = dev;
536 +static struct nf_queue_entry *nf_queue_entry_dup(struct nf_queue_entry *e)
538 + struct nf_queue_entry *entry = kmemdup(e, e->size, GFP_ATOMIC);
540 + nf_queue_entry_get_refs(entry);
546 +#ifdef CONFIG_BRIDGE_NETFILTER
547 +/* When called from bridge netfilter, skb->data must point to MAC header
548 + * before calling skb_gso_segment(). Else, original MAC header is lost
549 + * and segmented skbs will be sent to wrong destination.
551 +static void nf_bridge_adjust_skb_data(struct sk_buff *skb)
553 + if (skb->nf_bridge)
554 + __skb_push(skb, skb->network_header - skb->mac_header);
557 +static void nf_bridge_adjust_segmented_data(struct sk_buff *skb)
559 + if (skb->nf_bridge)
560 + __skb_pull(skb, skb->network_header - skb->mac_header);
563 +#define nf_bridge_adjust_skb_data(s) do {} while (0)
564 +#define nf_bridge_adjust_segmented_data(s) do {} while (0)
567 +static void free_entry(struct nf_queue_entry *entry)
569 + nf_queue_entry_release_refs(entry);
573 +static int __imq_nf_queue(struct nf_queue_entry *entry, struct net_device *dev);
575 +static int __imq_nf_queue_gso(struct nf_queue_entry *entry,
576 + struct net_device *dev, struct sk_buff *skb)
579 + struct nf_queue_entry *entry_seg;
581 + nf_bridge_adjust_segmented_data(skb);
583 + if (skb->next == NULL) { /* last packet, no need to copy entry */
584 + struct sk_buff *gso_skb = entry->skb;
586 + ret = __imq_nf_queue(entry, dev);
588 + entry->skb = gso_skb;
594 + entry_seg = nf_queue_entry_dup(entry);
596 + entry_seg->skb = skb;
597 + ret = __imq_nf_queue(entry_seg, dev);
599 + free_entry(entry_seg);
604 +static int imq_nf_queue(struct nf_queue_entry *entry, unsigned queue_num)
606 + struct sk_buff *skb, *segs;
607 + struct net_device *dev;
608 + unsigned int queued;
609 + int index, retval, err;
611 + index = entry->skb->imq_flags & IMQ_F_IFMASK;
612 + if (unlikely(index > numdevs - 1)) {
613 + if (net_ratelimit())
614 + pr_warn("IMQ: invalid device specified, highest is %u\n",
620 + /* check for imq device by index from cache */
621 + dev = imq_devs_cache[index];
622 + if (unlikely(!dev)) {
623 + dev = get_imq_device_by_index(index);
625 + retval = PTR_ERR(dev);
630 + if (unlikely(!(dev->flags & IFF_UP))) {
631 + entry->skb->imq_flags = 0;
632 + retval = -ECANCELED;
636 + /* Since 3.10.x, GSO handling moved here as result of upstream commit
637 + * a5fedd43d5f6c94c71053a66e4c3d2e35f1731a2 (netfilter: move
638 + * skb_gso_segment into nfnetlink_queue module).
640 + * Following code replicates the gso handling from
641 + * 'net/netfilter/nfnetlink_queue_core.c':nfqnl_enqueue_packet().
646 + switch (entry->state.pf) {
648 + skb->protocol = htons(ETH_P_IP);
651 + skb->protocol = htons(ETH_P_IPV6);
655 + if (!skb_is_gso(entry->skb))
656 + return __imq_nf_queue(entry, dev);
658 + nf_bridge_adjust_skb_data(skb);
659 + segs = skb_gso_segment(skb, 0);
660 + /* Does not use PTR_ERR to limit the number of error codes that can be
661 + * returned by nf_queue. For instance, callers rely on -ECANCELED to
662 + * mean 'ignore this hook'.
670 + struct sk_buff *nskb = segs->next;
671 + if (nskb && nskb->next)
672 + nskb->cb_next = NULL;
674 + err = __imq_nf_queue_gso(entry, dev, segs);
683 + if (err) /* some segments are already queued */
690 + nf_bridge_adjust_segmented_data(skb);
696 +static int __imq_nf_queue(struct nf_queue_entry *entry, struct net_device *dev)
698 + struct sk_buff *skb_orig, *skb, *skb_shared, *skb_popd;
700 + struct sk_buff *to_free = NULL;
701 + struct netdev_queue *txq;
702 + spinlock_t *root_lock;
704 + int retval = -EINVAL;
705 + unsigned int orig_queue_index;
707 + dev->last_rx = jiffies;
712 + /* skb has owner? => make clone */
713 + if (unlikely(skb->destructor)) {
715 + skb = skb_clone(skb, GFP_ATOMIC);
716 + if (unlikely(!skb)) {
720 + skb->cb_next = NULL;
724 + dev->stats.rx_bytes += skb->len;
725 + dev->stats.rx_packets++;
728 + /* skb->dev == NULL causes problems, try the find cause. */
729 + if (net_ratelimit()) {
730 + dev_warn(&dev->dev,
731 + "received packet with skb->dev == NULL\n");
738 + /* Disables softirqs for lock below */
739 + rcu_read_lock_bh();
741 + /* Multi-queue selection */
742 + orig_queue_index = skb_get_queue_mapping(skb);
743 + txq = imq_select_queue(dev, skb);
745 + q = rcu_dereference(txq->qdisc);
746 + if (unlikely(!q->enqueue))
747 + goto packet_not_eaten_by_imq_dev;
749 + skb->nf_queue_entry = entry;
750 + root_lock = qdisc_lock(q);
751 + spin_lock(root_lock);
753 + users = atomic_read(&skb->users);
755 + skb_shared = skb_get(skb); /* increase reference count by one */
757 + /* backup skb->cb, as qdisc layer will overwrite it */
758 + skb_save_cb(skb_shared);
759 + qdisc_enqueue_root(skb_shared, q, &to_free); /* might kfree_skb */
760 + if (likely(atomic_read(&skb_shared->users) == users + 1)) {
763 + kfree_skb(skb_shared); /* decrease reference count by one */
765 + skb->destructor = &imq_skb_destructor;
767 + skb_popd = qdisc_dequeue_skb(q, &validate);
770 + if (unlikely(skb_orig))
771 + kfree_skb(skb_orig); /* free original */
773 + spin_unlock(root_lock);
776 + /* schedule qdisc dequeue */
777 + __netif_schedule(q);
779 + if (likely(skb_popd)) {
780 + /* Note that we validate skb (GSO, checksum, ...) outside of locks */
782 + skb_popd = validate_xmit_skb_list(skb_popd, dev);
786 + int cpu = smp_processor_id(); /* ok because BHs are off */
788 + txq = skb_get_tx_queue(dev, skb_popd);
790 + IMQ device will not be frozen or stoped, and it always be successful.
791 + So we need not check its status and return value to accelerate.
793 + if (imq_dev_accurate_stats && txq->xmit_lock_owner != cpu) {
794 + HARD_TX_LOCK(dev, txq, cpu);
795 + if (!netif_xmit_frozen_or_stopped(txq)) {
796 + dev_hard_start_xmit(skb_popd, dev, txq, &dummy_ret);
798 + HARD_TX_UNLOCK(dev, txq);
800 + if (!netif_xmit_frozen_or_stopped(txq)) {
801 + dev_hard_start_xmit(skb_popd, dev, txq, &dummy_ret);
806 + /* No ready skb, then schedule it */
807 + __netif_schedule(q);
810 + rcu_read_unlock_bh();
814 + skb_restore_cb(skb_shared); /* restore skb->cb */
815 + skb->nf_queue_entry = NULL;
817 + * qdisc dropped packet and decreased skb reference count of
818 + * skb, so we don't really want to and try refree as that would
819 + * actually destroy the skb.
821 + spin_unlock(root_lock);
822 + goto packet_not_eaten_by_imq_dev;
825 +packet_not_eaten_by_imq_dev:
826 + skb_set_queue_mapping(skb, orig_queue_index);
827 + rcu_read_unlock_bh();
829 + /* cloned? restore original */
830 + if (unlikely(skb_orig)) {
832 + entry->skb = skb_orig;
836 + if (unlikely(to_free)) {
837 + kfree_skb_list(to_free);
841 +static unsigned int imq_nf_hook(void *priv,
842 + struct sk_buff *skb,
843 + const struct nf_hook_state *state)
845 + return (skb->imq_flags & IMQ_F_ENQUEUE) ? NF_IMQ_QUEUE : NF_ACCEPT;
848 +static int imq_close(struct net_device *dev)
850 + netif_stop_queue(dev);
854 +static int imq_open(struct net_device *dev)
856 + netif_start_queue(dev);
860 +static const struct net_device_ops imq_netdev_ops = {
861 + .ndo_open = imq_open,
862 + .ndo_stop = imq_close,
863 + .ndo_start_xmit = imq_dev_xmit,
864 + .ndo_get_stats = imq_get_stats,
867 +static void imq_setup(struct net_device *dev)
869 + dev->netdev_ops = &imq_netdev_ops;
870 + dev->type = ARPHRD_VOID;
871 + dev->mtu = 16000; /* too small? */
872 + dev->tx_queue_len = 11000; /* too big? */
873 + dev->flags = IFF_NOARP;
874 + dev->features = NETIF_F_SG | NETIF_F_FRAGLIST |
875 + NETIF_F_GSO | NETIF_F_HW_CSUM |
877 + dev->priv_flags &= ~(IFF_XMIT_DST_RELEASE |
878 + IFF_TX_SKB_SHARING);
881 +static int imq_validate(struct nlattr *tb[], struct nlattr *data[])
885 + if (tb[IFLA_ADDRESS]) {
886 + if (nla_len(tb[IFLA_ADDRESS]) != ETH_ALEN) {
890 + if (!is_valid_ether_addr(nla_data(tb[IFLA_ADDRESS]))) {
891 + ret = -EADDRNOTAVAIL;
897 + pr_warn("IMQ: imq_validate failed (%d)\n", ret);
901 +static struct rtnl_link_ops imq_link_ops __read_mostly = {
904 + .setup = imq_setup,
905 + .validate = imq_validate,
908 +static const struct nf_queue_handler imq_nfqh = {
909 + .outfn = imq_nf_queue,
912 +static int __init imq_init_hooks(void)
916 + nf_register_queue_imq_handler(&imq_nfqh);
918 + ret = nf_register_hooks(imq_ops, ARRAY_SIZE(imq_ops));
920 + nf_unregister_queue_imq_handler();
925 +static int __init imq_init_one(int index)
927 + struct net_device *dev;
930 + dev = alloc_netdev_mq(0, "imq%d", NET_NAME_UNKNOWN, imq_setup, numqueues);
934 + ret = dev_alloc_name(dev, dev->name);
938 + dev->rtnl_link_ops = &imq_link_ops;
939 + ret = register_netdevice(dev);
949 +static int __init imq_init_devs(void)
953 + if (numdevs < 1 || numdevs > IMQ_MAX_DEVS) {
954 + pr_err("IMQ: numdevs has to be betweed 1 and %u\n",
959 + if (numqueues < 1 || numqueues > IMQ_MAX_QUEUES) {
960 + pr_err("IMQ: numqueues has to be betweed 1 and %u\n",
965 + get_random_bytes(&imq_hashrnd, sizeof(imq_hashrnd));
968 + err = __rtnl_link_register(&imq_link_ops);
970 + for (i = 0; i < numdevs && !err; i++)
971 + err = imq_init_one(i);
974 + __rtnl_link_unregister(&imq_link_ops);
975 + memset(imq_devs_cache, 0, sizeof(imq_devs_cache));
982 +static int __init imq_init_module(void)
986 +#if defined(CONFIG_IMQ_NUM_DEVS)
987 + BUILD_BUG_ON(CONFIG_IMQ_NUM_DEVS > 16);
988 + BUILD_BUG_ON(CONFIG_IMQ_NUM_DEVS < 2);
989 + BUILD_BUG_ON(CONFIG_IMQ_NUM_DEVS - 1 > IMQ_F_IFMASK);
992 + err = imq_init_devs();
994 + pr_err("IMQ: Error trying imq_init_devs(net)\n");
998 + err = imq_init_hooks();
1000 + pr_err(KERN_ERR "IMQ: Error trying imq_init_hooks()\n");
1001 + rtnl_link_unregister(&imq_link_ops);
1002 + memset(imq_devs_cache, 0, sizeof(imq_devs_cache));
1006 + pr_info("IMQ driver loaded successfully. (numdevs = %d, numqueues = %d, imq_dev_accurate_stats = %d)\n",
1007 + numdevs, numqueues, imq_dev_accurate_stats);
1009 +#if defined(CONFIG_IMQ_BEHAVIOR_BA) || defined(CONFIG_IMQ_BEHAVIOR_BB)
1010 + pr_info("\tHooking IMQ before NAT on PREROUTING.\n");
1012 + pr_info("\tHooking IMQ after NAT on PREROUTING.\n");
1014 +#if defined(CONFIG_IMQ_BEHAVIOR_AB) || defined(CONFIG_IMQ_BEHAVIOR_BB)
1015 + pr_info("\tHooking IMQ before NAT on POSTROUTING.\n");
1017 + pr_info("\tHooking IMQ after NAT on POSTROUTING.\n");
1023 +static void __exit imq_unhook(void)
1025 + nf_unregister_hooks(imq_ops, ARRAY_SIZE(imq_ops));
1026 + nf_unregister_queue_imq_handler();
1029 +static void __exit imq_cleanup_devs(void)
1031 + rtnl_link_unregister(&imq_link_ops);
1032 + memset(imq_devs_cache, 0, sizeof(imq_devs_cache));
1035 +static void __exit imq_exit_module(void)
1038 + imq_cleanup_devs();
1039 + pr_info("IMQ driver unloaded successfully.\n");
1042 +module_init(imq_init_module);
1043 +module_exit(imq_exit_module);
1045 +module_param(numdevs, int, 0);
1046 +module_param(numqueues, int, 0);
1047 +module_param(imq_dev_accurate_stats, int, 0);
1048 +MODULE_PARM_DESC(numdevs, "number of IMQ devices (how many imq* devices will be created)");
1049 +MODULE_PARM_DESC(numqueues, "number of queues per IMQ device");
1050 +MODULE_PARM_DESC(imq_dev_accurate_stats, "Notify if need the accurate imq device stats");
1052 +MODULE_AUTHOR("https://github.com/imq/linuximq");
1053 +MODULE_DESCRIPTION("Pseudo-driver for the intermediate queue device. See https://github.com/imq/linuximq/wiki for more information.");
1054 +MODULE_LICENSE("GPL");
1055 +MODULE_ALIAS_RTNL_LINK("imq");
1056 diff --git a/include/linux/imq.h b/include/linux/imq.h
1057 new file mode 100644
1058 index 0000000..1babb09
1060 +++ b/include/linux/imq.h
1065 +/* IFMASK (16 device indexes, 0 to 15) and flag(s) fit in 5 bits */
1066 +#define IMQ_F_BITS 5
1068 +#define IMQ_F_IFMASK 0x0f
1069 +#define IMQ_F_ENQUEUE 0x10
1071 +#define IMQ_MAX_DEVS (IMQ_F_IFMASK + 1)
1073 +#endif /* _IMQ_H */
1075 diff --git a/include/linux/netdevice.h b/include/linux/netdevice.h
1076 index e16a2a9..4a1090a 100644
1077 --- a/include/linux/netdevice.h
1078 +++ b/include/linux/netdevice.h
1079 @@ -3669,6 +3669,19 @@ static inline void netif_tx_unlock_bh(struct net_device *dev)
1083 +#define HARD_TX_LOCK_BH(dev, txq) { \
1084 + if ((dev->features & NETIF_F_LLTX) == 0) { \
1085 + __netif_tx_lock_bh(txq); \
1089 +#define HARD_TX_UNLOCK_BH(dev, txq) { \
1090 + if ((dev->features & NETIF_F_LLTX) == 0) { \
1091 + __netif_tx_unlock_bh(txq); \
1096 static inline void netif_tx_disable(struct net_device *dev)
1099 diff --git a/include/linux/netfilter/xt_IMQ.h b/include/linux/netfilter/xt_IMQ.h
1100 new file mode 100644
1101 index 0000000..9b07230
1103 +++ b/include/linux/netfilter/xt_IMQ.h
1108 +struct xt_imq_info {
1109 + unsigned int todev; /* target imq device */
1112 +#endif /* _XT_IMQ_H */
1114 diff --git a/include/linux/netfilter_ipv4/ipt_IMQ.h b/include/linux/netfilter_ipv4/ipt_IMQ.h
1115 new file mode 100644
1116 index 0000000..7af320f
1118 +++ b/include/linux/netfilter_ipv4/ipt_IMQ.h
1123 +/* Backwards compatibility for old userspace */
1124 +#include <linux/netfilter/xt_IMQ.h>
1126 +#define ipt_imq_info xt_imq_info
1128 +#endif /* _IPT_IMQ_H */
1130 diff --git a/include/linux/netfilter_ipv6/ip6t_IMQ.h b/include/linux/netfilter_ipv6/ip6t_IMQ.h
1131 new file mode 100644
1132 index 0000000..198ac01
1134 +++ b/include/linux/netfilter_ipv6/ip6t_IMQ.h
1136 +#ifndef _IP6T_IMQ_H
1137 +#define _IP6T_IMQ_H
1139 +/* Backwards compatibility for old userspace */
1140 +#include <linux/netfilter/xt_IMQ.h>
1142 +#define ip6t_imq_info xt_imq_info
1144 +#endif /* _IP6T_IMQ_H */
1146 diff --git a/include/linux/skbuff.h b/include/linux/skbuff.h
1147 index 32810f2..4ce1d0a 100644
1148 --- a/include/linux/skbuff.h
1149 +++ b/include/linux/skbuff.h
1151 #include <linux/in6.h>
1152 #include <linux/if_packet.h>
1153 #include <net/flow.h>
1154 +#if defined(CONFIG_IMQ) || defined(CONFIG_IMQ_MODULE)
1155 +#include <linux/imq.h>
1159 /* The interface for checksum offload between the stack and networking drivers
1161 @@ -654,6 +658,9 @@ struct sk_buff {
1162 * first. This is owned by whoever has the skb queued ATM.
1164 char cb[48] __aligned(8);
1165 +#if defined(CONFIG_IMQ) || defined(CONFIG_IMQ_MODULE)
1169 unsigned long _skb_refdst;
1170 void (*destructor)(struct sk_buff *skb);
1171 @@ -663,6 +670,9 @@ struct sk_buff {
1172 #if defined(CONFIG_NF_CONNTRACK) || defined(CONFIG_NF_CONNTRACK_MODULE)
1173 struct nf_conntrack *nfct;
1175 +#if defined(CONFIG_IMQ) || defined(CONFIG_IMQ_MODULE)
1176 + struct nf_queue_entry *nf_queue_entry;
1178 #if IS_ENABLED(CONFIG_BRIDGE_NETFILTER)
1179 struct nf_bridge_info *nf_bridge;
1181 @@ -743,6 +753,9 @@ struct sk_buff {
1182 __u8 offload_fwd_mark:1;
1184 /* 2, 4 or 5 bit hole */
1185 +#if defined(CONFIG_IMQ) || defined(CONFIG_IMQ_MODULE)
1186 + __u8 imq_flags:IMQ_F_BITS;
1189 #ifdef CONFIG_NET_SCHED
1190 __u16 tc_index; /* traffic control index */
1191 @@ -903,6 +916,12 @@ void kfree_skb_list(struct sk_buff *segs);
1192 void skb_tx_error(struct sk_buff *skb);
1193 void consume_skb(struct sk_buff *skb);
1194 void __kfree_skb(struct sk_buff *skb);
1196 +#if defined(CONFIG_IMQ) || defined(CONFIG_IMQ_MODULE)
1197 +int skb_save_cb(struct sk_buff *skb);
1198 +int skb_restore_cb(struct sk_buff *skb);
1201 extern struct kmem_cache *skbuff_head_cache;
1203 void kfree_skb_partial(struct sk_buff *skb, bool head_stolen);
1204 @@ -3594,6 +3613,10 @@ static inline void __nf_copy(struct sk_buff *dst, const struct sk_buff *src,
1206 dst->nfctinfo = src->nfctinfo;
1208 +#if defined(CONFIG_IMQ) || defined(CONFIG_IMQ_MODULE)
1209 + dst->imq_flags = src->imq_flags;
1210 + dst->nf_queue_entry = src->nf_queue_entry;
1212 #if IS_ENABLED(CONFIG_BRIDGE_NETFILTER)
1213 dst->nf_bridge = src->nf_bridge;
1214 nf_bridge_get(src->nf_bridge);
1215 diff --git a/include/net/netfilter/nf_queue.h b/include/net/netfilter/nf_queue.h
1216 index 2280cfe..ec8fa51 100644
1217 --- a/include/net/netfilter/nf_queue.h
1218 +++ b/include/net/netfilter/nf_queue.h
1219 @@ -30,6 +30,12 @@ struct nf_queue_handler {
1220 void nf_register_queue_handler(struct net *net, const struct nf_queue_handler *qh);
1221 void nf_unregister_queue_handler(struct net *net);
1222 void nf_reinject(struct nf_queue_entry *entry, unsigned int verdict);
1223 +void nf_queue_entry_release_refs(struct nf_queue_entry *entry);
1225 +#if defined(CONFIG_IMQ) || defined(CONFIG_IMQ_MODULE)
1226 +void nf_register_queue_imq_handler(const struct nf_queue_handler *qh);
1227 +void nf_unregister_queue_imq_handler(void);
1230 void nf_queue_entry_get_refs(struct nf_queue_entry *entry);
1231 void nf_queue_entry_release_refs(struct nf_queue_entry *entry);
1232 diff --git a/include/net/pkt_sched.h b/include/net/pkt_sched.h
1233 index cd334c9..6757228 100644
1234 --- a/include/net/pkt_sched.h
1235 +++ b/include/net/pkt_sched.h
1236 @@ -105,6 +105,8 @@ int sch_direct_xmit(struct sk_buff *skb, struct Qdisc *q,
1238 void __qdisc_run(struct Qdisc *q);
1240 +struct sk_buff *qdisc_dequeue_skb(struct Qdisc *q, bool *validate);
1242 static inline void qdisc_run(struct Qdisc *q)
1244 if (qdisc_run_begin(q))
1245 diff --git a/include/net/sch_generic.h b/include/net/sch_generic.h
1246 index e6aa0a2..08b37dc 100644
1247 --- a/include/net/sch_generic.h
1248 +++ b/include/net/sch_generic.h
1249 @@ -518,6 +518,13 @@ static inline int qdisc_enqueue(struct sk_buff *skb, struct Qdisc *sch,
1250 return sch->enqueue(skb, sch, to_free);
1253 +static inline int qdisc_enqueue_root(struct sk_buff *skb, struct Qdisc *sch,
1254 + struct sk_buff **to_free)
1256 + qdisc_skb_cb(skb)->pkt_len = skb->len;
1257 + return qdisc_enqueue(skb, sch, to_free) & NET_XMIT_MASK;
1260 static inline bool qdisc_is_percpu_stats(const struct Qdisc *q)
1262 return q->flags & TCQ_F_CPUSTATS;
1263 diff --git a/include/uapi/linux/netfilter.h b/include/uapi/linux/netfilter.h
1264 index d93f949..23fb6d1 100644
1265 --- a/include/uapi/linux/netfilter.h
1266 +++ b/include/uapi/linux/netfilter.h
1271 -#define NF_MAX_VERDICT NF_STOP
1272 +#define NF_IMQ_QUEUE 6
1273 +#define NF_MAX_VERDICT NF_IMQ_QUEUE
1275 /* we overload the higher bits for encoding auxiliary data such as the queue
1276 * number or errno values. Not nice, but better than additional function
1277 diff --git a/net/core/dev.c b/net/core/dev.c
1278 index 6666b28..3e12add 100644
1279 --- a/net/core/dev.c
1280 +++ b/net/core/dev.c
1282 #include <linux/netfilter_ingress.h>
1283 #include <linux/sctp.h>
1284 #include <linux/crash_dump.h>
1285 +#if defined(CONFIG_IMQ) || defined(CONFIG_IMQ_MODULE)
1286 +#include <linux/imq.h>
1289 #include "net-sysfs.h"
1291 @@ -2906,7 +2909,12 @@ static int xmit_one(struct sk_buff *skb, struct net_device *dev,
1295 +#if defined(CONFIG_IMQ) || defined(CONFIG_IMQ_MODULE)
1296 + if ((!list_empty(&ptype_all) || !list_empty(&dev->ptype_all)) &&
1297 + !(skb->imq_flags & IMQ_F_ENQUEUE))
1299 if (!list_empty(&ptype_all) || !list_empty(&dev->ptype_all))
1301 dev_queue_xmit_nit(skb, dev);
1304 @@ -2945,6 +2953,8 @@ struct sk_buff *dev_hard_start_xmit(struct sk_buff *first, struct net_device *de
1308 +EXPORT_SYMBOL_GPL(dev_hard_start_xmit);
1310 static struct sk_buff *validate_xmit_vlan(struct sk_buff *skb,
1311 netdev_features_t features)
1313 diff --git a/net/core/skbuff.c b/net/core/skbuff.c
1314 index 1e3e008..379236e 100644
1315 --- a/net/core/skbuff.c
1316 +++ b/net/core/skbuff.c
1317 @@ -82,6 +82,87 @@ struct kmem_cache *skbuff_head_cache __read_mostly;
1318 static struct kmem_cache *skbuff_fclone_cache __read_mostly;
1319 int sysctl_max_skb_frags __read_mostly = MAX_SKB_FRAGS;
1320 EXPORT_SYMBOL(sysctl_max_skb_frags);
1321 +#if defined(CONFIG_IMQ) || defined(CONFIG_IMQ_MODULE)
1322 +static struct kmem_cache *skbuff_cb_store_cache __read_mostly;
1325 +#if defined(CONFIG_IMQ) || defined(CONFIG_IMQ_MODULE)
1326 +/* Control buffer save/restore for IMQ devices */
1327 +struct skb_cb_table {
1328 + char cb[48] __aligned(8);
1333 +static DEFINE_SPINLOCK(skb_cb_store_lock);
1335 +int skb_save_cb(struct sk_buff *skb)
1337 + struct skb_cb_table *next;
1339 + next = kmem_cache_alloc(skbuff_cb_store_cache, GFP_ATOMIC);
1343 + BUILD_BUG_ON(sizeof(skb->cb) != sizeof(next->cb));
1345 + memcpy(next->cb, skb->cb, sizeof(skb->cb));
1346 + next->cb_next = skb->cb_next;
1348 + atomic_set(&next->refcnt, 1);
1350 + skb->cb_next = next;
1353 +EXPORT_SYMBOL(skb_save_cb);
1355 +int skb_restore_cb(struct sk_buff *skb)
1357 + struct skb_cb_table *next;
1359 + if (!skb->cb_next)
1362 + next = skb->cb_next;
1364 + BUILD_BUG_ON(sizeof(skb->cb) != sizeof(next->cb));
1366 + memcpy(skb->cb, next->cb, sizeof(skb->cb));
1367 + skb->cb_next = next->cb_next;
1369 + spin_lock(&skb_cb_store_lock);
1371 + if (atomic_dec_and_test(&next->refcnt))
1372 + kmem_cache_free(skbuff_cb_store_cache, next);
1374 + spin_unlock(&skb_cb_store_lock);
1378 +EXPORT_SYMBOL(skb_restore_cb);
1380 +static void skb_copy_stored_cb(struct sk_buff * , const struct sk_buff * ) __attribute__ ((unused));
1381 +static void skb_copy_stored_cb(struct sk_buff *new, const struct sk_buff *__old)
1383 + struct skb_cb_table *next;
1384 + struct sk_buff *old;
1386 + if (!__old->cb_next) {
1387 + new->cb_next = NULL;
1391 + spin_lock(&skb_cb_store_lock);
1393 + old = (struct sk_buff *)__old;
1395 + next = old->cb_next;
1396 + atomic_inc(&next->refcnt);
1397 + new->cb_next = next;
1399 + spin_unlock(&skb_cb_store_lock);
1404 * skb_panic - private function for out-of-line support
1405 @@ -654,6 +735,28 @@ static void skb_release_head_state(struct sk_buff *skb)
1407 skb->destructor(skb);
1409 +#if defined(CONFIG_IMQ) || defined(CONFIG_IMQ_MODULE)
1411 + * This should not happen. When it does, avoid memleak by restoring
1412 + * the chain of cb-backups.
1414 + while (skb->cb_next != NULL) {
1415 + if (net_ratelimit())
1416 + pr_warn("IMQ: kfree_skb: skb->cb_next: %08x\n",
1417 + (unsigned int)(uintptr_t)skb->cb_next);
1419 + skb_restore_cb(skb);
1422 + * This should not happen either, nf_queue_entry is nullified in
1423 + * imq_dev_xmit(). If we have non-NULL nf_queue_entry then we are
1424 + * leaking entry pointers, maybe memory. We don't know if this is
1425 + * pointer to already freed memory, or should this be freed.
1426 + * If this happens we need to add refcounting, etc for nf_queue_entry.
1428 + if (skb->nf_queue_entry && net_ratelimit())
1429 + pr_warn("%s\n", "IMQ: kfree_skb: skb->nf_queue_entry != NULL");
1431 #if IS_ENABLED(CONFIG_NF_CONNTRACK)
1432 nf_conntrack_put(skb->nfct);
1434 @@ -843,6 +946,10 @@ static void __copy_skb_header(struct sk_buff *new, const struct sk_buff *old)
1435 new->sp = secpath_get(old->sp);
1437 __nf_copy(new, old, false);
1438 +#if defined(CONFIG_IMQ) || defined(CONFIG_IMQ_MODULE)
1439 + new->cb_next = NULL;
1440 + /*skb_copy_stored_cb(new, old);*/
1443 /* Note : this field could be in headers_start/headers_end section
1444 * It is not yet because we do not want to have a 16 bit hole
1445 @@ -3463,6 +3570,13 @@ void __init skb_init(void)
1447 SLAB_HWCACHE_ALIGN|SLAB_PANIC,
1449 +#if defined(CONFIG_IMQ) || defined(CONFIG_IMQ_MODULE)
1450 + skbuff_cb_store_cache = kmem_cache_create("skbuff_cb_store_cache",
1451 + sizeof(struct skb_cb_table),
1453 + SLAB_HWCACHE_ALIGN|SLAB_PANIC,
1459 diff --git a/net/ipv6/ip6_output.c b/net/ipv6/ip6_output.c
1460 index 59eb4ed..8020b07 100644
1461 --- a/net/ipv6/ip6_output.c
1462 +++ b/net/ipv6/ip6_output.c
1463 @@ -66,9 +66,6 @@ static int ip6_finish_output2(struct net *net, struct sock *sk, struct sk_buff *
1464 struct in6_addr *nexthop;
1467 - skb->protocol = htons(ETH_P_IPV6);
1470 if (ipv6_addr_is_multicast(&ipv6_hdr(skb)->daddr)) {
1471 struct inet6_dev *idev = ip6_dst_idev(skb_dst(skb));
1473 @@ -150,6 +147,13 @@ int ip6_output(struct net *net, struct sock *sk, struct sk_buff *skb)
1478 + * IMQ-patch: moved setting skb->dev and skb->protocol from
1479 + * ip6_finish_output2 to fix crashing at netif_skb_features().
1481 + skb->protocol = htons(ETH_P_IPV6);
1484 return NF_HOOK_COND(NFPROTO_IPV6, NF_INET_POST_ROUTING,
1485 net, sk, skb, NULL, dev,
1487 diff --git a/net/netfilter/Kconfig b/net/netfilter/Kconfig
1488 index e8d56d9..1ed3468 100644
1489 --- a/net/netfilter/Kconfig
1490 +++ b/net/netfilter/Kconfig
1491 @@ -823,6 +823,18 @@ config NETFILTER_XT_TARGET_LOG
1493 To compile it as a module, choose M here. If unsure, say N.
1495 +config NETFILTER_XT_TARGET_IMQ
1496 + tristate '"IMQ" target support'
1497 + depends on NETFILTER_XTABLES
1498 + depends on IP_NF_MANGLE || IP6_NF_MANGLE
1500 + default m if NETFILTER_ADVANCED=n
1502 + This option adds a `IMQ' target which is used to specify if and
1503 + to which imq device packets should get enqueued/dequeued.
1505 + To compile it as a module, choose M here. If unsure, say N.
1507 config NETFILTER_XT_TARGET_MARK
1508 tristate '"MARK" target support'
1509 depends on NETFILTER_ADVANCED
1510 diff --git a/net/netfilter/Makefile b/net/netfilter/Makefile
1511 index c23c3c8..99911ef 100644
1512 --- a/net/netfilter/Makefile
1513 +++ b/net/netfilter/Makefile
1514 @@ -119,6 +119,7 @@ obj-$(CONFIG_NETFILTER_XT_TARGET_CT) += xt_CT.o
1515 obj-$(CONFIG_NETFILTER_XT_TARGET_DSCP) += xt_DSCP.o
1516 obj-$(CONFIG_NETFILTER_XT_TARGET_HL) += xt_HL.o
1517 obj-$(CONFIG_NETFILTER_XT_TARGET_HMARK) += xt_HMARK.o
1518 +obj-$(CONFIG_NETFILTER_XT_TARGET_IMQ) += xt_IMQ.o
1519 obj-$(CONFIG_NETFILTER_XT_TARGET_LED) += xt_LED.o
1520 obj-$(CONFIG_NETFILTER_XT_TARGET_LOG) += xt_LOG.o
1521 obj-$(CONFIG_NETFILTER_XT_TARGET_NETMAP) += xt_NETMAP.o
1522 diff --git a/net/netfilter/core.c b/net/netfilter/core.c
1523 index 004af03..768a08b 100644
1524 --- a/net/netfilter/core.c
1525 +++ b/net/netfilter/core.c
1526 @@ -360,8 +360,11 @@ int nf_hook_slow(struct sk_buff *skb, struct nf_hook_state *state)
1527 ret = NF_DROP_GETERR(verdict);
1530 - } else if ((verdict & NF_VERDICT_MASK) == NF_QUEUE) {
1531 + } else if ((verdict & NF_VERDICT_MASK) == NF_QUEUE ||
1532 + (verdict & NF_VERDICT_MASK) == NF_IMQ_QUEUE) {
1533 ret = nf_queue(skb, state, &entry, verdict);
1534 + if (ret == -ECANCELED)
1536 if (ret == 1 && entry)
1539 diff --git a/net/netfilter/nf_queue.c b/net/netfilter/nf_queue.c
1540 index 8f08d75..8d362c0 100644
1541 --- a/net/netfilter/nf_queue.c
1542 +++ b/net/netfilter/nf_queue.c
1544 * receives, no matter what.
1547 +#if defined(CONFIG_IMQ) || defined(CONFIG_IMQ_MODULE)
1548 +static const struct nf_queue_handler __rcu *queue_imq_handler __read_mostly;
1550 +void nf_register_queue_imq_handler(const struct nf_queue_handler *qh)
1552 + rcu_assign_pointer(queue_imq_handler, qh);
1554 +EXPORT_SYMBOL_GPL(nf_register_queue_imq_handler);
1556 +void nf_unregister_queue_imq_handler(void)
1558 + RCU_INIT_POINTER(queue_imq_handler, NULL);
1559 + synchronize_rcu();
1561 +EXPORT_SYMBOL_GPL(nf_unregister_queue_imq_handler);
1564 /* return EBUSY when somebody else is registered, return EEXIST if the
1565 * same handler is registered, return 0 in case of success. */
1566 void nf_register_queue_handler(struct net *net, const struct nf_queue_handler *qh)
1567 @@ -108,16 +125,28 @@ void nf_queue_nf_hook_drop(struct net *net, const struct nf_hook_entry *entry)
1570 static int __nf_queue(struct sk_buff *skb, const struct nf_hook_state *state,
1571 - unsigned int queuenum)
1572 + unsigned int verdict)
1574 int status = -ENOENT;
1575 struct nf_queue_entry *entry = NULL;
1576 const struct nf_afinfo *afinfo;
1577 const struct nf_queue_handler *qh;
1578 struct net *net = state->net;
1579 + unsigned int queuetype = verdict & NF_VERDICT_MASK;
1580 + unsigned int queuenum = verdict >> NF_VERDICT_QBITS;
1582 /* QUEUE == DROP if no one is waiting, to be safe. */
1583 - qh = rcu_dereference(net->nf.queue_handler);
1584 + if (queuetype == NF_IMQ_QUEUE) {
1585 +#if defined(CONFIG_IMQ) || defined(CONFIG_IMQ_MODULE)
1586 + qh = rcu_dereference(queue_imq_handler);
1592 + qh = rcu_dereference(net->nf.queue_handler);
1598 @@ -218,6 +247,7 @@ void nf_reinject(struct nf_queue_entry *entry, unsigned int verdict)
1602 + case NF_IMQ_QUEUE:
1603 err = nf_queue(skb, &entry->state, &hook_entry, verdict);
1606 diff --git a/net/netfilter/xt_IMQ.c b/net/netfilter/xt_IMQ.c
1607 new file mode 100644
1608 index 0000000..f9c5817
1610 +++ b/net/netfilter/xt_IMQ.c
1613 + * This target marks packets to be enqueued to an imq device
1615 +#include <linux/module.h>
1616 +#include <linux/skbuff.h>
1617 +#include <linux/netfilter/x_tables.h>
1618 +#include <linux/netfilter/xt_IMQ.h>
1619 +#include <linux/imq.h>
1621 +static unsigned int imq_target(struct sk_buff *pskb,
1622 + const struct xt_action_param *par)
1624 + const struct xt_imq_info *mr = par->targinfo;
1626 + pskb->imq_flags = (mr->todev & IMQ_F_IFMASK) | IMQ_F_ENQUEUE;
1628 + return XT_CONTINUE;
1631 +static int imq_checkentry(const struct xt_tgchk_param *par)
1633 + struct xt_imq_info *mr = par->targinfo;
1635 + if (mr->todev > IMQ_MAX_DEVS - 1) {
1636 + pr_warn("IMQ: invalid device specified, highest is %u\n",
1637 + IMQ_MAX_DEVS - 1);
1644 +static struct xt_target xt_imq_reg[] __read_mostly = {
1647 + .family = AF_INET,
1648 + .checkentry = imq_checkentry,
1649 + .target = imq_target,
1650 + .targetsize = sizeof(struct xt_imq_info),
1651 + .table = "mangle",
1656 + .family = AF_INET6,
1657 + .checkentry = imq_checkentry,
1658 + .target = imq_target,
1659 + .targetsize = sizeof(struct xt_imq_info),
1660 + .table = "mangle",
1665 +static int __init imq_init(void)
1667 + return xt_register_targets(xt_imq_reg, ARRAY_SIZE(xt_imq_reg));
1670 +static void __exit imq_fini(void)
1672 + xt_unregister_targets(xt_imq_reg, ARRAY_SIZE(xt_imq_reg));
1675 +module_init(imq_init);
1676 +module_exit(imq_fini);
1678 +MODULE_AUTHOR("https://github.com/imq/linuximq");
1679 +MODULE_DESCRIPTION("Pseudo-driver for the intermediate queue device. See https://github.com/imq/linuximq/wiki for more information.");
1680 +MODULE_LICENSE("GPL");
1681 +MODULE_ALIAS("ipt_IMQ");
1682 +MODULE_ALIAS("ip6t_IMQ");
1684 diff --git a/net/sched/sch_generic.c b/net/sched/sch_generic.c
1685 index 6cfb6e9..4c675e9 100644
1686 --- a/net/sched/sch_generic.c
1687 +++ b/net/sched/sch_generic.c
1688 @@ -154,6 +154,14 @@ static struct sk_buff *dequeue_skb(struct Qdisc *q, bool *validate,
1692 +struct sk_buff *qdisc_dequeue_skb(struct Qdisc *q, bool *validate)
1696 + return dequeue_skb(q, validate, &packets);
1698 +EXPORT_SYMBOL(qdisc_dequeue_skb);
1701 * Transmit possibly several skbs, and handle the return status as
1702 * required. Owning running seqcount bit guarantees that