1 diff -ruN linux-3.10.27/drivers/net/imq.c linux-3.10.27-imq/drivers/net/imq.c
2 --- linux-3.10.27/drivers/net/imq.c 1970-01-01 01:00:00.000000000 +0100
3 +++ linux-3.10.27-imq/drivers/net/imq.c 2014-01-18 10:19:59.342342913 +0100
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>
17 + * Credits: Jan Rafaj <imq2t@cedric.vabo.cz>
18 + * - Update patch to 2.4.21
19 + * Sebastian Strollo <sstrollo@nortelnetworks.com>
20 + * - Fix "Dead-loop on netdevice imq"-issue
21 + * Marcel Sebek <sebek64@post.cz>
22 + * - Update to 2.6.2-rc1
24 + * After some time of inactivity there is a group taking care
25 + * of IMQ again: http://www.linuximq.net
28 + * 2004/06/30 - New version of IMQ patch to kernels <=2.6.7
29 + * including the following changes:
31 + * - Correction of ipv6 support "+"s issue (Hasso Tepper)
32 + * - Correction of imq_init_devs() issue that resulted in
33 + * kernel OOPS unloading IMQ as module (Norbert Buchmuller)
34 + * - Addition of functionality to choose number of IMQ devices
35 + * during kernel config (Andre Correa)
36 + * - Addition of functionality to choose how IMQ hooks on
37 + * PRE and POSTROUTING (after or before NAT) (Andre Correa)
38 + * - Cosmetic corrections (Norbert Buchmuller) (Andre Correa)
41 + * 2005/12/16 - IMQ versions between 2.6.7 and 2.6.13 were
42 + * released with almost no problems. 2.6.14-x was released
43 + * with some important changes: nfcache was removed; After
44 + * some weeks of trouble we figured out that some IMQ fields
45 + * in skb were missing in skbuff.c - skb_clone and copy_skb_header.
46 + * These functions are correctly patched by this new patch version.
48 + * Thanks for all who helped to figure out all the problems with
49 + * 2.6.14.x: Patrick McHardy, Rune Kock, VeNoMouS, Max CtRiX,
50 + * Kevin Shanahan, Richard Lucassen, Valery Dachev (hopefully
51 + * I didn't forget anybody). I apologize again for my lack of time.
54 + * 2008/06/17 - 2.6.25 - Changed imq.c to use qdisc_run() instead
55 + * of qdisc_restart() and moved qdisc_run() to tasklet to avoid
56 + * recursive locking. New initialization routines to fix 'rmmod' not
57 + * working anymore. Used code from ifb.c. (Jussi Kivilinna)
59 + * 2008/08/06 - 2.6.26 - (JK)
60 + * - Replaced tasklet with 'netif_schedule()'.
61 + * - Cleaned up and added comments for imq_nf_queue().
64 + * - Add skb_save_cb/skb_restore_cb helper functions for backuping
65 + * control buffer. This is needed because qdisc-layer on kernels
66 + * 2.6.27 and newer overwrite control buffer. (Jussi Kivilinna)
67 + * - Add better locking for IMQ device. Hopefully this will solve
68 + * SMP issues. (Jussi Kivilinna)
71 + * - Port to 2.6.29 + fix rmmod not working
73 + * 2009/04/20 - (Jussi Kivilinna)
74 + * - Use netdevice feature flags to avoid extra packet handling
75 + * by core networking layer and possibly increase performance.
77 + * 2009/09/26 - (Jussi Kivilinna)
78 + * - Add imq_nf_reinject_lockless to fix deadlock with
79 + * imq_nf_queue/imq_nf_reinject.
81 + * 2009/12/08 - (Jussi Kivilinna)
83 + * - Add check for skb->nf_queue_entry==NULL in imq_dev_xmit()
84 + * - Also add better error checking for skb->nf_queue_entry usage
86 + * 2010/02/25 - (Jussi Kivilinna)
89 + * 2010/08/15 - (Jussi Kivilinna)
91 + * - Simplify hook registration by using nf_register_hooks.
92 + * - nf_reinject doesn't need spinlock around it, therefore remove
93 + * imq_nf_reinject function. Other nf_reinject users protect
94 + * their own data with spinlock. With IMQ however all data is
95 + * needed is stored per skbuff, so no locking is needed.
96 + * - Changed IMQ to use 'separate' NF_IMQ_QUEUE instead of
97 + * NF_QUEUE, this allows working coexistance of IMQ and other
99 + * - Make IMQ multi-queue. Number of IMQ device queues can be
100 + * increased with 'numqueues' module parameters. Default number
101 + * of queues is 1, in other words by default IMQ works as
102 + * single-queue device. Multi-queue selection is based on
103 + * IFB multi-queue patch by Changli Gao <xiaosuo@gmail.com>.
105 + * 2011/03/18 - (Jussi Kivilinna)
108 + * 2011/07/12 - (syoder89@gmail.com)
109 + * - Crash fix that happens when the receiving interface has more
110 + * than one queue (add missing skb_set_queue_mapping in
111 + * imq_select_queue).
113 + * 2011/07/26 - (Jussi Kivilinna)
114 + * - Add queue mapping checks for packets exiting IMQ.
117 + * 2011/08/16 - (Jussi Kivilinna)
118 + * - Clear IFF_TX_SKB_SHARING flag that was added for linux 3.0.2
120 + * 2011/11/03 - Germano Michel <germanomichel@gmail.com>
121 + * - Fix IMQ for net namespaces
123 + * 2011/11/04 - Jussi Kivilinna <jussi.kivilinna@mbnet.fi>
125 + * - Clean-up, move 'get imq device pointer by imqX name' to
126 + * separate function from imq_nf_queue().
128 + * 2012/01/05 - Jussi Kivilinna <jussi.kivilinna@mbnet.fi>
131 + * 2012/03/19 - Jussi Kivilinna <jussi.kivilinna@mbnet.fi>
134 + * 2012/12/12 - Jussi Kivilinna <jussi.kivilinna@mbnet.fi>
136 + * - Fix checkpatch.pl warnings
138 + * 2013/09/10 - Jussi Kivilinna <jussi.kivilinna@iki.fi>
139 + * - Fixed GSO handling for 3.10, see imq_nf_queue() for comments.
140 + * - Don't copy skb->cb_next when copying or cloning skbuffs.
142 + * Also, many thanks to pablo Sebastian Greco for making the initial
143 + * patch and to those who helped the testing.
145 + * More info at: http://www.linuximq.net/ (Andre Correa)
148 +#include <linux/module.h>
149 +#include <linux/kernel.h>
150 +#include <linux/moduleparam.h>
151 +#include <linux/list.h>
152 +#include <linux/skbuff.h>
153 +#include <linux/netdevice.h>
154 +#include <linux/etherdevice.h>
155 +#include <linux/rtnetlink.h>
156 +#include <linux/if_arp.h>
157 +#include <linux/netfilter.h>
158 +#include <linux/netfilter_ipv4.h>
159 +#if defined(CONFIG_IPV6) || defined(CONFIG_IPV6_MODULE)
160 + #include <linux/netfilter_ipv6.h>
162 +#include <linux/imq.h>
163 +#include <net/pkt_sched.h>
164 +#include <net/netfilter/nf_queue.h>
165 +#include <net/sock.h>
166 +#include <linux/ip.h>
167 +#include <linux/ipv6.h>
168 +#include <linux/if_vlan.h>
169 +#include <linux/if_pppox.h>
171 +#include <net/ipv6.h>
173 +static int imq_nf_queue(struct nf_queue_entry *entry, unsigned queue_num);
175 +static nf_hookfn imq_nf_hook;
177 +static struct nf_hook_ops imq_ops[] = {
179 + /* imq_ingress_ipv4 */
180 + .hook = imq_nf_hook,
181 + .owner = THIS_MODULE,
183 + .hooknum = NF_INET_PRE_ROUTING,
184 +#if defined(CONFIG_IMQ_BEHAVIOR_BA) || defined(CONFIG_IMQ_BEHAVIOR_BB)
185 + .priority = NF_IP_PRI_MANGLE + 1,
187 + .priority = NF_IP_PRI_NAT_DST + 1,
191 + /* imq_egress_ipv4 */
192 + .hook = imq_nf_hook,
193 + .owner = THIS_MODULE,
195 + .hooknum = NF_INET_POST_ROUTING,
196 +#if defined(CONFIG_IMQ_BEHAVIOR_AA) || defined(CONFIG_IMQ_BEHAVIOR_BA)
197 + .priority = NF_IP_PRI_LAST,
199 + .priority = NF_IP_PRI_NAT_SRC - 1,
202 +#if defined(CONFIG_IPV6) || defined(CONFIG_IPV6_MODULE)
204 + /* imq_ingress_ipv6 */
205 + .hook = imq_nf_hook,
206 + .owner = THIS_MODULE,
208 + .hooknum = NF_INET_PRE_ROUTING,
209 +#if defined(CONFIG_IMQ_BEHAVIOR_BA) || defined(CONFIG_IMQ_BEHAVIOR_BB)
210 + .priority = NF_IP6_PRI_MANGLE + 1,
212 + .priority = NF_IP6_PRI_NAT_DST + 1,
216 + /* imq_egress_ipv6 */
217 + .hook = imq_nf_hook,
218 + .owner = THIS_MODULE,
220 + .hooknum = NF_INET_POST_ROUTING,
221 +#if defined(CONFIG_IMQ_BEHAVIOR_AA) || defined(CONFIG_IMQ_BEHAVIOR_BA)
222 + .priority = NF_IP6_PRI_LAST,
224 + .priority = NF_IP6_PRI_NAT_SRC - 1,
230 +#if defined(CONFIG_IMQ_NUM_DEVS)
231 +static int numdevs = CONFIG_IMQ_NUM_DEVS;
233 +static int numdevs = IMQ_MAX_DEVS;
236 +static struct net_device *imq_devs_cache[IMQ_MAX_DEVS];
238 +#define IMQ_MAX_QUEUES 32
239 +static int numqueues = 1;
240 +static u32 imq_hashrnd;
242 +static inline __be16 pppoe_proto(const struct sk_buff *skb)
244 + return *((__be16 *)(skb_mac_header(skb) + ETH_HLEN +
245 + sizeof(struct pppoe_hdr)));
248 +static u16 imq_hash(struct net_device *dev, struct sk_buff *skb)
250 + unsigned int pull_len;
251 + u16 protocol = skb->protocol;
263 + switch (protocol) {
264 + case htons(ETH_P_8021Q): {
265 + if (unlikely(skb_pull(skb, VLAN_HLEN) == NULL))
268 + pull_len += VLAN_HLEN;
269 + skb->network_header += VLAN_HLEN;
271 + protocol = vlan_eth_hdr(skb)->h_vlan_encapsulated_proto;
275 + case htons(ETH_P_PPP_SES): {
276 + if (unlikely(skb_pull(skb, PPPOE_SES_HLEN) == NULL))
279 + pull_len += PPPOE_SES_HLEN;
280 + skb->network_header += PPPOE_SES_HLEN;
282 + protocol = pppoe_proto(skb);
286 + case htons(ETH_P_IP): {
287 + const struct iphdr *iph = ip_hdr(skb);
289 + if (unlikely(!pskb_may_pull(skb, sizeof(struct iphdr))))
292 + addr1 = iph->daddr;
293 + addr2 = iph->saddr;
295 + ip_proto = !(ip_hdr(skb)->frag_off & htons(IP_MF | IP_OFFSET)) ?
297 + ihl = ip_hdrlen(skb);
301 +#if defined(CONFIG_IPV6) || defined(CONFIG_IPV6_MODULE)
302 + case htons(ETH_P_IPV6): {
303 + const struct ipv6hdr *iph = ipv6_hdr(skb);
306 + if (unlikely(!pskb_may_pull(skb, sizeof(struct ipv6hdr))))
309 + addr1 = iph->daddr.s6_addr32[3];
310 + addr2 = iph->saddr.s6_addr32[3];
311 + ihl = ipv6_skip_exthdr(skb, sizeof(struct ipv6hdr), &ip_proto,
313 + if (unlikely(ihl < 0))
321 + if (pull_len != 0) {
322 + skb_push(skb, pull_len);
323 + skb->network_header -= pull_len;
326 + return (u16)(ntohs(protocol) % dev->real_num_tx_queues);
330 + swap(addr1, addr2);
332 + switch (ip_proto) {
339 + case IPPROTO_UDPLITE: {
340 + if (likely(skb_copy_bits(skb, ihl, &ports.in32, 4) >= 0)) {
341 + if (ports.in16[0] > ports.in16[1])
342 + swap(ports.in16[0], ports.in16[1]);
352 + if (pull_len != 0) {
353 + skb_push(skb, pull_len);
354 + skb->network_header -= pull_len;
357 + hash = jhash_3words(addr1, addr2, ports.in32, imq_hashrnd ^ ip_proto);
359 + return (u16)(((u64)hash * dev->real_num_tx_queues) >> 32);
362 +static inline bool sk_tx_queue_recorded(struct sock *sk)
364 + return (sk_tx_queue_get(sk) >= 0);
367 +static struct netdev_queue *imq_select_queue(struct net_device *dev,
368 + struct sk_buff *skb)
370 + u16 queue_index = 0;
373 + if (likely(dev->real_num_tx_queues == 1))
376 + /* IMQ can be receiving ingress or engress packets. */
378 + /* Check first for if rx_queue is set */
379 + if (skb_rx_queue_recorded(skb)) {
380 + queue_index = skb_get_rx_queue(skb);
384 + /* Check if socket has tx_queue set */
385 + if (sk_tx_queue_recorded(skb->sk)) {
386 + queue_index = sk_tx_queue_get(skb->sk);
390 + /* Try use socket hash */
391 + if (skb->sk && skb->sk->sk_hash) {
392 + hash = skb->sk->sk_hash;
394 + (u16)(((u64)hash * dev->real_num_tx_queues) >> 32);
398 + /* Generate hash from packet data */
399 + queue_index = imq_hash(dev, skb);
402 + if (unlikely(queue_index >= dev->real_num_tx_queues))
403 + queue_index = (u16)((u32)queue_index % dev->real_num_tx_queues);
405 + skb_set_queue_mapping(skb, queue_index);
406 + return netdev_get_tx_queue(dev, queue_index);
409 +static struct net_device_stats *imq_get_stats(struct net_device *dev)
411 + return &dev->stats;
414 +/* called for packets kfree'd in qdiscs at places other than enqueue */
415 +static void imq_skb_destructor(struct sk_buff *skb)
417 + struct nf_queue_entry *entry = skb->nf_queue_entry;
419 + skb->nf_queue_entry = NULL;
422 + nf_queue_entry_release_refs(entry);
426 + skb_restore_cb(skb); /* kfree backup */
429 +static void imq_done_check_queue_mapping(struct sk_buff *skb,
430 + struct net_device *dev)
432 + unsigned int queue_index;
434 + /* Don't let queue_mapping be left too large after exiting IMQ */
435 + if (likely(skb->dev != dev && skb->dev != NULL)) {
436 + queue_index = skb_get_queue_mapping(skb);
437 + if (unlikely(queue_index >= skb->dev->real_num_tx_queues)) {
438 + queue_index = (u16)((u32)queue_index %
439 + skb->dev->real_num_tx_queues);
440 + skb_set_queue_mapping(skb, queue_index);
443 + /* skb->dev was IMQ device itself or NULL, be on safe side and
444 + * just clear queue mapping.
446 + skb_set_queue_mapping(skb, 0);
450 +static netdev_tx_t imq_dev_xmit(struct sk_buff *skb, struct net_device *dev)
452 + struct nf_queue_entry *entry = skb->nf_queue_entry;
454 + skb->nf_queue_entry = NULL;
455 + dev->trans_start = jiffies;
457 + dev->stats.tx_bytes += skb->len;
458 + dev->stats.tx_packets++;
460 + if (unlikely(entry == NULL)) {
461 + /* We don't know what is going on here.. packet is queued for
462 + * imq device, but (probably) not by us.
464 + * If this packet was not send here by imq_nf_queue(), then
465 + * skb_save_cb() was not used and skb_free() should not show:
466 + * WARNING: IMQ: kfree_skb: skb->cb_next:..
468 + * WARNING: IMQ: kfree_skb: skb->nf_queue_entry...
470 + * However if this message is shown, then IMQ is somehow broken
471 + * and you should report this to linuximq.net.
474 + /* imq_dev_xmit is black hole that eats all packets, report that
475 + * we eat this packet happily and increase dropped counters.
478 + dev->stats.tx_dropped++;
479 + dev_kfree_skb(skb);
481 + return NETDEV_TX_OK;
484 + skb_restore_cb(skb); /* restore skb->cb */
486 + skb->imq_flags = 0;
487 + skb->destructor = NULL;
489 + imq_done_check_queue_mapping(skb, dev);
491 + nf_reinject(entry, NF_ACCEPT);
493 + return NETDEV_TX_OK;
496 +static struct net_device *get_imq_device_by_index(int index)
498 + struct net_device *dev = NULL;
502 + /* get device by name and cache result */
503 + snprintf(buf, sizeof(buf), "imq%d", index);
505 + /* Search device from all namespaces. */
506 + for_each_net(net) {
507 + dev = dev_get_by_name(net, buf);
512 + if (WARN_ON_ONCE(dev == NULL)) {
513 + /* IMQ device not found. Exotic config? */
514 + return ERR_PTR(-ENODEV);
517 + imq_devs_cache[index] = dev;
523 +static struct nf_queue_entry *nf_queue_entry_dup(struct nf_queue_entry *e)
525 + struct nf_queue_entry *entry = kmemdup(e, e->size, GFP_ATOMIC);
527 + if (nf_queue_entry_get_refs(entry))
534 +#ifdef CONFIG_BRIDGE_NETFILTER
535 +/* When called from bridge netfilter, skb->data must point to MAC header
536 + * before calling skb_gso_segment(). Else, original MAC header is lost
537 + * and segmented skbs will be sent to wrong destination.
539 +static void nf_bridge_adjust_skb_data(struct sk_buff *skb)
541 + if (skb->nf_bridge)
542 + __skb_push(skb, skb->network_header - skb->mac_header);
545 +static void nf_bridge_adjust_segmented_data(struct sk_buff *skb)
547 + if (skb->nf_bridge)
548 + __skb_pull(skb, skb->network_header - skb->mac_header);
551 +#define nf_bridge_adjust_skb_data(s) do {} while (0)
552 +#define nf_bridge_adjust_segmented_data(s) do {} while (0)
555 +static void free_entry(struct nf_queue_entry *entry)
557 + nf_queue_entry_release_refs(entry);
561 +static int __imq_nf_queue(struct nf_queue_entry *entry, struct net_device *dev);
563 +static int __imq_nf_queue_gso(struct nf_queue_entry *entry,
564 + struct net_device *dev, struct sk_buff *skb)
567 + struct nf_queue_entry *entry_seg;
569 + nf_bridge_adjust_segmented_data(skb);
571 + if (skb->next == NULL) { /* last packet, no need to copy entry */
572 + struct sk_buff *gso_skb = entry->skb;
574 + ret = __imq_nf_queue(entry, dev);
576 + entry->skb = gso_skb;
582 + entry_seg = nf_queue_entry_dup(entry);
584 + entry_seg->skb = skb;
585 + ret = __imq_nf_queue(entry_seg, dev);
587 + free_entry(entry_seg);
592 +static int imq_nf_queue(struct nf_queue_entry *entry, unsigned queue_num)
594 + struct sk_buff *skb, *segs;
595 + struct net_device *dev;
596 + unsigned int queued;
597 + int index, retval, err;
599 + index = entry->skb->imq_flags & IMQ_F_IFMASK;
600 + if (unlikely(index > numdevs - 1)) {
601 + if (net_ratelimit())
602 + pr_warn("IMQ: invalid device specified, highest is %u\n",
608 + /* check for imq device by index from cache */
609 + dev = imq_devs_cache[index];
610 + if (unlikely(!dev)) {
611 + dev = get_imq_device_by_index(index);
613 + retval = PTR_ERR(dev);
618 + if (unlikely(!(dev->flags & IFF_UP))) {
619 + entry->skb->imq_flags = 0;
620 + retval = -ECANCELED;
624 + if (!skb_is_gso(entry->skb))
625 + return __imq_nf_queue(entry, dev);
627 + /* Since 3.10.x, GSO handling moved here as result of upstream commit
628 + * a5fedd43d5f6c94c71053a66e4c3d2e35f1731a2 (netfilter: move
629 + * skb_gso_segment into nfnetlink_queue module).
631 + * Following code replicates the gso handling from
632 + * 'net/netfilter/nfnetlink_queue_core.c':nfqnl_enqueue_packet().
637 + switch (entry->pf) {
639 + skb->protocol = htons(ETH_P_IP);
642 + skb->protocol = htons(ETH_P_IPV6);
646 + nf_bridge_adjust_skb_data(skb);
647 + segs = skb_gso_segment(skb, 0);
648 + /* Does not use PTR_ERR to limit the number of error codes that can be
649 + * returned by nf_queue. For instance, callers rely on -ECANCELED to
650 + * mean 'ignore this hook'.
658 + struct sk_buff *nskb = segs->next;
659 + if (nskb && nskb->next)
660 + nskb->cb_next = NULL;
662 + err = __imq_nf_queue_gso(entry, dev, segs);
671 + if (err) /* some segments are already queued */
678 + nf_bridge_adjust_segmented_data(skb);
684 +static int __imq_nf_queue(struct nf_queue_entry *entry, struct net_device *dev)
686 + struct sk_buff *skb_orig, *skb, *skb_shared;
688 + struct netdev_queue *txq;
689 + spinlock_t *root_lock;
691 + int retval = -EINVAL;
692 + unsigned int orig_queue_index;
694 + dev->last_rx = jiffies;
699 + /* skb has owner? => make clone */
700 + if (unlikely(skb->destructor)) {
702 + skb = skb_clone(skb, GFP_ATOMIC);
703 + if (unlikely(!skb)) {
707 + skb->cb_next = NULL;
711 + skb->nf_queue_entry = entry;
713 + dev->stats.rx_bytes += skb->len;
714 + dev->stats.rx_packets++;
717 + /* skb->dev == NULL causes problems, try the find cause. */
718 + if (net_ratelimit()) {
719 + dev_warn(&dev->dev,
720 + "received packet with skb->dev == NULL\n");
727 + /* Disables softirqs for lock below */
728 + rcu_read_lock_bh();
730 + /* Multi-queue selection */
731 + orig_queue_index = skb_get_queue_mapping(skb);
732 + txq = imq_select_queue(dev, skb);
734 + q = rcu_dereference(txq->qdisc);
735 + if (unlikely(!q->enqueue))
736 + goto packet_not_eaten_by_imq_dev;
738 + root_lock = qdisc_lock(q);
739 + spin_lock(root_lock);
741 + users = atomic_read(&skb->users);
743 + skb_shared = skb_get(skb); /* increase reference count by one */
745 + /* backup skb->cb, as qdisc layer will overwrite it */
746 + skb_save_cb(skb_shared);
747 + qdisc_enqueue_root(skb_shared, q); /* might kfree_skb */
749 + if (likely(atomic_read(&skb_shared->users) == users + 1)) {
750 + kfree_skb(skb_shared); /* decrease reference count by one */
752 + skb->destructor = &imq_skb_destructor;
755 + if (unlikely(skb_orig))
756 + kfree_skb(skb_orig); /* free original */
758 + spin_unlock(root_lock);
759 + rcu_read_unlock_bh();
761 + /* schedule qdisc dequeue */
762 + __netif_schedule(q);
767 + skb_restore_cb(skb_shared); /* restore skb->cb */
768 + skb->nf_queue_entry = NULL;
770 + * qdisc dropped packet and decreased skb reference count of
771 + * skb, so we don't really want to and try refree as that would
772 + * actually destroy the skb.
774 + spin_unlock(root_lock);
775 + goto packet_not_eaten_by_imq_dev;
778 +packet_not_eaten_by_imq_dev:
779 + skb_set_queue_mapping(skb, orig_queue_index);
780 + rcu_read_unlock_bh();
782 + /* cloned? restore original */
783 + if (unlikely(skb_orig)) {
785 + entry->skb = skb_orig;
792 +static unsigned int imq_nf_hook(unsigned int hook, struct sk_buff *pskb,
793 + const struct net_device *indev,
794 + const struct net_device *outdev,
795 + int (*okfn)(struct sk_buff *))
797 + return (pskb->imq_flags & IMQ_F_ENQUEUE) ? NF_IMQ_QUEUE : NF_ACCEPT;
800 +static int imq_close(struct net_device *dev)
802 + netif_stop_queue(dev);
806 +static int imq_open(struct net_device *dev)
808 + netif_start_queue(dev);
812 +static const struct net_device_ops imq_netdev_ops = {
813 + .ndo_open = imq_open,
814 + .ndo_stop = imq_close,
815 + .ndo_start_xmit = imq_dev_xmit,
816 + .ndo_get_stats = imq_get_stats,
819 +static void imq_setup(struct net_device *dev)
821 + dev->netdev_ops = &imq_netdev_ops;
822 + dev->type = ARPHRD_VOID;
823 + dev->mtu = 16000; /* too small? */
824 + dev->tx_queue_len = 11000; /* too big? */
825 + dev->flags = IFF_NOARP;
826 + dev->features = NETIF_F_SG | NETIF_F_FRAGLIST |
827 + NETIF_F_GSO | NETIF_F_HW_CSUM |
829 + dev->priv_flags &= ~(IFF_XMIT_DST_RELEASE |
830 + IFF_TX_SKB_SHARING);
833 +static int imq_validate(struct nlattr *tb[], struct nlattr *data[])
837 + if (tb[IFLA_ADDRESS]) {
838 + if (nla_len(tb[IFLA_ADDRESS]) != ETH_ALEN) {
842 + if (!is_valid_ether_addr(nla_data(tb[IFLA_ADDRESS]))) {
843 + ret = -EADDRNOTAVAIL;
849 + pr_warn("IMQ: imq_validate failed (%d)\n", ret);
853 +static struct rtnl_link_ops imq_link_ops __read_mostly = {
856 + .setup = imq_setup,
857 + .validate = imq_validate,
860 +static const struct nf_queue_handler imq_nfqh = {
861 + .outfn = imq_nf_queue,
864 +static int __init imq_init_hooks(void)
868 + nf_register_queue_imq_handler(&imq_nfqh);
870 + ret = nf_register_hooks(imq_ops, ARRAY_SIZE(imq_ops));
872 + nf_unregister_queue_imq_handler();
877 +static int __init imq_init_one(int index)
879 + struct net_device *dev;
882 + dev = alloc_netdev_mq(0, "imq%d", imq_setup, numqueues);
886 + ret = dev_alloc_name(dev, dev->name);
890 + dev->rtnl_link_ops = &imq_link_ops;
891 + ret = register_netdevice(dev);
901 +static int __init imq_init_devs(void)
905 + if (numdevs < 1 || numdevs > IMQ_MAX_DEVS) {
906 + pr_err("IMQ: numdevs has to be betweed 1 and %u\n",
911 + if (numqueues < 1 || numqueues > IMQ_MAX_QUEUES) {
912 + pr_err("IMQ: numqueues has to be betweed 1 and %u\n",
917 + get_random_bytes(&imq_hashrnd, sizeof(imq_hashrnd));
920 + err = __rtnl_link_register(&imq_link_ops);
922 + for (i = 0; i < numdevs && !err; i++)
923 + err = imq_init_one(i);
926 + __rtnl_link_unregister(&imq_link_ops);
927 + memset(imq_devs_cache, 0, sizeof(imq_devs_cache));
934 +static int __init imq_init_module(void)
938 +#if defined(CONFIG_IMQ_NUM_DEVS)
939 + BUILD_BUG_ON(CONFIG_IMQ_NUM_DEVS > 16);
940 + BUILD_BUG_ON(CONFIG_IMQ_NUM_DEVS < 2);
941 + BUILD_BUG_ON(CONFIG_IMQ_NUM_DEVS - 1 > IMQ_F_IFMASK);
944 + err = imq_init_devs();
946 + pr_err("IMQ: Error trying imq_init_devs(net)\n");
950 + err = imq_init_hooks();
952 + pr_err(KERN_ERR "IMQ: Error trying imq_init_hooks()\n");
953 + rtnl_link_unregister(&imq_link_ops);
954 + memset(imq_devs_cache, 0, sizeof(imq_devs_cache));
958 + pr_info("IMQ driver loaded successfully. (numdevs = %d, numqueues = %d)\n",
959 + numdevs, numqueues);
961 +#if defined(CONFIG_IMQ_BEHAVIOR_BA) || defined(CONFIG_IMQ_BEHAVIOR_BB)
962 + pr_info("\tHooking IMQ before NAT on PREROUTING.\n");
964 + pr_info("\tHooking IMQ after NAT on PREROUTING.\n");
966 +#if defined(CONFIG_IMQ_BEHAVIOR_AB) || defined(CONFIG_IMQ_BEHAVIOR_BB)
967 + pr_info("\tHooking IMQ before NAT on POSTROUTING.\n");
969 + pr_info("\tHooking IMQ after NAT on POSTROUTING.\n");
975 +static void __exit imq_unhook(void)
977 + nf_unregister_hooks(imq_ops, ARRAY_SIZE(imq_ops));
978 + nf_unregister_queue_imq_handler();
981 +static void __exit imq_cleanup_devs(void)
983 + rtnl_link_unregister(&imq_link_ops);
984 + memset(imq_devs_cache, 0, sizeof(imq_devs_cache));
987 +static void __exit imq_exit_module(void)
990 + imq_cleanup_devs();
991 + pr_info("IMQ driver unloaded successfully.\n");
994 +module_init(imq_init_module);
995 +module_exit(imq_exit_module);
997 +module_param(numdevs, int, 0);
998 +module_param(numqueues, int, 0);
999 +MODULE_PARM_DESC(numdevs, "number of IMQ devices (how many imq* devices will be created)");
1000 +MODULE_PARM_DESC(numqueues, "number of queues per IMQ device");
1001 +MODULE_AUTHOR("http://www.linuximq.net");
1002 +MODULE_DESCRIPTION("Pseudo-driver for the intermediate queue device. See http://www.linuximq.net/ for more information.");
1003 +MODULE_LICENSE("GPL");
1004 +MODULE_ALIAS_RTNL_LINK("imq");
1006 diff -ruN linux-3.10.27/drivers/net/Kconfig linux-3.10.27-imq/drivers/net/Kconfig
1007 --- linux-3.10.27/drivers/net/Kconfig 2014-01-16 00:29:14.000000000 +0100
1008 +++ linux-3.10.27-imq/drivers/net/Kconfig 2014-01-18 10:19:59.341342885 +0100
1009 @@ -207,6 +207,125 @@
1014 + tristate "IMQ (intermediate queueing device) support"
1015 + depends on NETDEVICES && NETFILTER
1017 + The IMQ device(s) is used as placeholder for QoS queueing
1018 + disciplines. Every packet entering/leaving the IP stack can be
1019 + directed through the IMQ device where it's enqueued/dequeued to the
1020 + attached qdisc. This allows you to treat network devices as classes
1021 + and distribute bandwidth among them. Iptables is used to specify
1022 + through which IMQ device, if any, packets travel.
1024 + More information at: http://www.linuximq.net/
1026 + To compile this driver as a module, choose M here: the module
1027 + will be called imq. If unsure, say N.
1030 + prompt "IMQ behavior (PRE/POSTROUTING)"
1032 + default IMQ_BEHAVIOR_AB
1034 + This setting defines how IMQ behaves in respect to its
1035 + hooking in PREROUTING and POSTROUTING.
1037 + IMQ can work in any of the following ways:
1039 + PREROUTING | POSTROUTING
1040 + -----------------|-------------------
1041 + #1 After NAT | After NAT
1042 + #2 After NAT | Before NAT
1043 + #3 Before NAT | After NAT
1044 + #4 Before NAT | Before NAT
1046 + The default behavior is to hook before NAT on PREROUTING
1047 + and after NAT on POSTROUTING (#3).
1049 + This settings are specially usefull when trying to use IMQ
1050 + to shape NATed clients.
1052 + More information can be found at: www.linuximq.net
1054 + If not sure leave the default settings alone.
1056 +config IMQ_BEHAVIOR_AA
1059 + This setting defines how IMQ behaves in respect to its
1060 + hooking in PREROUTING and POSTROUTING.
1062 + Choosing this option will make IMQ hook like this:
1064 + PREROUTING: After NAT
1065 + POSTROUTING: After NAT
1067 + More information can be found at: www.linuximq.net
1069 + If not sure leave the default settings alone.
1071 +config IMQ_BEHAVIOR_AB
1074 + This setting defines how IMQ behaves in respect to its
1075 + hooking in PREROUTING and POSTROUTING.
1077 + Choosing this option will make IMQ hook like this:
1079 + PREROUTING: After NAT
1080 + POSTROUTING: Before NAT
1082 + More information can be found at: www.linuximq.net
1084 + If not sure leave the default settings alone.
1086 +config IMQ_BEHAVIOR_BA
1089 + This setting defines how IMQ behaves in respect to its
1090 + hooking in PREROUTING and POSTROUTING.
1092 + Choosing this option will make IMQ hook like this:
1094 + PREROUTING: Before NAT
1095 + POSTROUTING: After NAT
1097 + More information can be found at: www.linuximq.net
1099 + If not sure leave the default settings alone.
1101 +config IMQ_BEHAVIOR_BB
1104 + This setting defines how IMQ behaves in respect to its
1105 + hooking in PREROUTING and POSTROUTING.
1107 + Choosing this option will make IMQ hook like this:
1109 + PREROUTING: Before NAT
1110 + POSTROUTING: Before NAT
1112 + More information can be found at: www.linuximq.net
1114 + If not sure leave the default settings alone.
1118 +config IMQ_NUM_DEVS
1119 + int "Number of IMQ devices"
1124 + This setting defines how many IMQ devices will be created.
1126 + The default value is 16.
1128 + More information can be found at: www.linuximq.net
1130 + If not sure leave the default settings alone.
1133 tristate "Universal TUN/TAP device driver support"
1135 diff -ruN linux-3.10.27/drivers/net/Makefile linux-3.10.27-imq/drivers/net/Makefile
1136 --- linux-3.10.27/drivers/net/Makefile 2014-01-16 00:29:14.000000000 +0100
1137 +++ linux-3.10.27-imq/drivers/net/Makefile 2014-01-18 10:19:59.341342885 +0100
1139 obj-$(CONFIG_DUMMY) += dummy.o
1140 obj-$(CONFIG_EQUALIZER) += eql.o
1141 obj-$(CONFIG_IFB) += ifb.o
1142 +obj-$(CONFIG_IMQ) += imq.o
1143 obj-$(CONFIG_MACVLAN) += macvlan.o
1144 obj-$(CONFIG_MACVTAP) += macvtap.o
1145 obj-$(CONFIG_MII) += mii.o
1146 diff -ruN linux-3.10.27/include/linux/imq.h linux-3.10.27-imq/include/linux/imq.h
1147 --- linux-3.10.27/include/linux/imq.h 1970-01-01 01:00:00.000000000 +0100
1148 +++ linux-3.10.27-imq/include/linux/imq.h 2014-01-18 10:19:59.342342913 +0100
1153 +/* IFMASK (16 device indexes, 0 to 15) and flag(s) fit in 5 bits */
1154 +#define IMQ_F_BITS 5
1156 +#define IMQ_F_IFMASK 0x0f
1157 +#define IMQ_F_ENQUEUE 0x10
1159 +#define IMQ_MAX_DEVS (IMQ_F_IFMASK + 1)
1161 +#endif /* _IMQ_H */
1163 diff -ruN linux-3.10.27/include/linux/netfilter/xt_IMQ.h linux-3.10.27-imq/include/linux/netfilter/xt_IMQ.h
1164 --- linux-3.10.27/include/linux/netfilter/xt_IMQ.h 1970-01-01 01:00:00.000000000 +0100
1165 +++ linux-3.10.27-imq/include/linux/netfilter/xt_IMQ.h 2014-01-18 10:19:59.342342913 +0100
1170 +struct xt_imq_info {
1171 + unsigned int todev; /* target imq device */
1174 +#endif /* _XT_IMQ_H */
1176 diff -ruN linux-3.10.27/include/linux/netfilter_ipv4/ipt_IMQ.h linux-3.10.27-imq/include/linux/netfilter_ipv4/ipt_IMQ.h
1177 --- linux-3.10.27/include/linux/netfilter_ipv4/ipt_IMQ.h 1970-01-01 01:00:00.000000000 +0100
1178 +++ linux-3.10.27-imq/include/linux/netfilter_ipv4/ipt_IMQ.h 2014-01-18 10:19:59.343342933 +0100
1183 +/* Backwards compatibility for old userspace */
1184 +#include <linux/netfilter/xt_IMQ.h>
1186 +#define ipt_imq_info xt_imq_info
1188 +#endif /* _IPT_IMQ_H */
1190 diff -ruN linux-3.10.27/include/linux/netfilter_ipv6/ip6t_IMQ.h linux-3.10.27-imq/include/linux/netfilter_ipv6/ip6t_IMQ.h
1191 --- linux-3.10.27/include/linux/netfilter_ipv6/ip6t_IMQ.h 1970-01-01 01:00:00.000000000 +0100
1192 +++ linux-3.10.27-imq/include/linux/netfilter_ipv6/ip6t_IMQ.h 2014-01-18 10:19:59.343342933 +0100
1194 +#ifndef _IP6T_IMQ_H
1195 +#define _IP6T_IMQ_H
1197 +/* Backwards compatibility for old userspace */
1198 +#include <linux/netfilter/xt_IMQ.h>
1200 +#define ip6t_imq_info xt_imq_info
1202 +#endif /* _IP6T_IMQ_H */
1204 diff -ruN linux-3.10.27/include/linux/skbuff.h linux-3.10.27-imq/include/linux/skbuff.h
1205 --- linux-3.10.27/include/linux/skbuff.h 2014-01-16 00:29:14.000000000 +0100
1206 +++ linux-3.10.27-imq/include/linux/skbuff.h 2014-01-18 10:18:22.220271201 +0100
1208 #include <linux/dma-mapping.h>
1209 #include <linux/netdev_features.h>
1210 #include <net/flow_keys.h>
1211 +#if defined(CONFIG_IMQ) || defined(CONFIG_IMQ_MODULE)
1212 +#include <linux/imq.h>
1215 /* Don't change this without changing skb_csum_unnecessary! */
1216 #define CHECKSUM_NONE 0
1218 * first. This is owned by whoever has the skb queued ATM.
1220 char cb[48] __aligned(8);
1221 +#if defined(CONFIG_IMQ) || defined(CONFIG_IMQ_MODULE)
1225 unsigned long _skb_refdst;
1228 #if defined(CONFIG_NF_CONNTRACK) || defined(CONFIG_NF_CONNTRACK_MODULE)
1229 struct nf_conntrack *nfct;
1231 +#if defined(CONFIG_IMQ) || defined(CONFIG_IMQ_MODULE)
1232 + struct nf_queue_entry *nf_queue_entry;
1234 #ifdef CONFIG_BRIDGE_NETFILTER
1235 struct nf_bridge_info *nf_bridge;
1238 __u8 encapsulation:1;
1239 /* 7/9 bit hole (depending on ndisc_nodetype presence) */
1240 kmemcheck_bitfield_end(flags2);
1242 +#if defined(CONFIG_IMQ) || defined(CONFIG_IMQ_MODULE)
1243 + __u8 imq_flags:IMQ_F_BITS;
1245 #ifdef CONFIG_NET_DMA
1246 dma_cookie_t dma_cookie;
1248 @@ -616,7 +627,10 @@
1250 return (struct rtable *)skb_dst(skb);
1253 +#if defined(CONFIG_IMQ) || defined(CONFIG_IMQ_MODULE)
1254 +extern int skb_save_cb(struct sk_buff *skb);
1255 +extern int skb_restore_cb(struct sk_buff *skb);
1257 extern void kfree_skb(struct sk_buff *skb);
1258 extern void kfree_skb_list(struct sk_buff *segs);
1259 extern void skb_tx_error(struct sk_buff *skb);
1260 @@ -2735,6 +2749,10 @@
1261 nf_conntrack_get(src->nfct);
1262 dst->nfctinfo = src->nfctinfo;
1264 +#if defined(CONFIG_IMQ) || defined(CONFIG_IMQ_MODULE)
1265 + dst->imq_flags = src->imq_flags;
1266 + dst->nf_queue_entry = src->nf_queue_entry;
1268 #ifdef CONFIG_BRIDGE_NETFILTER
1269 dst->nf_bridge = src->nf_bridge;
1270 nf_bridge_get(src->nf_bridge);
1271 diff -ruN linux-3.10.27/include/net/netfilter/nf_queue.h linux-3.10.27-imq/include/net/netfilter/nf_queue.h
1272 --- linux-3.10.27/include/net/netfilter/nf_queue.h 2014-01-16 00:29:14.000000000 +0100
1273 +++ linux-3.10.27-imq/include/net/netfilter/nf_queue.h 2014-01-18 10:19:59.345342949 +0100
1275 void nf_register_queue_handler(const struct nf_queue_handler *qh);
1276 void nf_unregister_queue_handler(void);
1277 extern void nf_reinject(struct nf_queue_entry *entry, unsigned int verdict);
1278 +extern void nf_queue_entry_release_refs(struct nf_queue_entry *entry);
1280 +#if defined(CONFIG_IMQ) || defined(CONFIG_IMQ_MODULE)
1281 +extern void nf_register_queue_imq_handler(const struct nf_queue_handler *qh);
1282 +extern void nf_unregister_queue_imq_handler(void);
1285 bool nf_queue_entry_get_refs(struct nf_queue_entry *entry);
1286 void nf_queue_entry_release_refs(struct nf_queue_entry *entry);
1287 diff -ruN linux-3.10.27/include/uapi/linux/netfilter.h linux-3.10.27-imq/include/uapi/linux/netfilter.h
1288 --- linux-3.10.27/include/uapi/linux/netfilter.h 2014-01-16 00:29:14.000000000 +0100
1289 +++ linux-3.10.27-imq/include/uapi/linux/netfilter.h 2014-01-18 10:19:59.345342949 +0100
1294 -#define NF_MAX_VERDICT NF_STOP
1295 +#define NF_IMQ_QUEUE 6
1296 +#define NF_MAX_VERDICT NF_IMQ_QUEUE
1298 /* we overload the higher bits for encoding auxiliary data such as the queue
1299 * number or errno values. Not nice, but better than additional function
1300 diff -ruN linux-3.10.27/net/core/dev.c linux-3.10.27-imq/net/core/dev.c
1301 --- linux-3.10.27/net/core/dev.c 2014-01-16 00:29:14.000000000 +0100
1302 +++ linux-3.10.27-imq/net/core/dev.c 2014-01-18 10:19:59.347342963 +0100
1304 #include <linux/inetdevice.h>
1305 #include <linux/cpu_rmap.h>
1306 #include <linux/static_key.h>
1307 +#if defined(CONFIG_IMQ) || defined(CONFIG_IMQ_MODULE)
1308 +#include <linux/imq.h>
1311 #include "net-sysfs.h"
1313 @@ -2573,7 +2576,12 @@
1317 +#if defined(CONFIG_IMQ) || defined(CONFIG_IMQ_MODULE)
1318 + if (!list_empty(&ptype_all) &&
1319 + !(skb->imq_flags & IMQ_F_ENQUEUE))
1321 if (!list_empty(&ptype_all))
1323 dev_queue_xmit_nit(skb, dev);
1326 diff -ruN linux-3.10.27/net/core/skbuff.c linux-3.10.27-imq/net/core/skbuff.c
1327 --- linux-3.10.27/net/core/skbuff.c 2014-01-16 00:29:14.000000000 +0100
1328 +++ linux-3.10.27-imq/net/core/skbuff.c 2014-01-18 10:19:59.348342972 +0100
1331 struct kmem_cache *skbuff_head_cache __read_mostly;
1332 static struct kmem_cache *skbuff_fclone_cache __read_mostly;
1333 +#if defined(CONFIG_IMQ) || defined(CONFIG_IMQ_MODULE)
1334 +static struct kmem_cache *skbuff_cb_store_cache __read_mostly;
1337 +#if defined(CONFIG_IMQ) || defined(CONFIG_IMQ_MODULE)
1338 +/* Control buffer save/restore for IMQ devices */
1339 +struct skb_cb_table {
1340 + char cb[48] __aligned(8);
1345 +static DEFINE_SPINLOCK(skb_cb_store_lock);
1347 +int skb_save_cb(struct sk_buff *skb)
1349 + struct skb_cb_table *next;
1351 + next = kmem_cache_alloc(skbuff_cb_store_cache, GFP_ATOMIC);
1355 + BUILD_BUG_ON(sizeof(skb->cb) != sizeof(next->cb));
1357 + memcpy(next->cb, skb->cb, sizeof(skb->cb));
1358 + next->cb_next = skb->cb_next;
1360 + atomic_set(&next->refcnt, 1);
1362 + skb->cb_next = next;
1365 +EXPORT_SYMBOL(skb_save_cb);
1367 +int skb_restore_cb(struct sk_buff *skb)
1369 + struct skb_cb_table *next;
1371 + if (!skb->cb_next)
1374 + next = skb->cb_next;
1376 + BUILD_BUG_ON(sizeof(skb->cb) != sizeof(next->cb));
1378 + memcpy(skb->cb, next->cb, sizeof(skb->cb));
1379 + skb->cb_next = next->cb_next;
1381 + spin_lock(&skb_cb_store_lock);
1383 + if (atomic_dec_and_test(&next->refcnt))
1384 + kmem_cache_free(skbuff_cb_store_cache, next);
1386 + spin_unlock(&skb_cb_store_lock);
1390 +EXPORT_SYMBOL(skb_restore_cb);
1392 +static void skb_copy_stored_cb(struct sk_buff *new, const struct sk_buff *__old)
1394 + struct skb_cb_table *next;
1395 + struct sk_buff *old;
1397 + if (!__old->cb_next) {
1398 + new->cb_next = NULL;
1402 + spin_lock(&skb_cb_store_lock);
1404 + old = (struct sk_buff *)__old;
1406 + next = old->cb_next;
1407 + atomic_inc(&next->refcnt);
1408 + new->cb_next = next;
1410 + spin_unlock(&skb_cb_store_lock);
1415 * skb_panic - private function for out-of-line support
1416 @@ -582,6 +661,28 @@
1418 skb->destructor(skb);
1420 +#if defined(CONFIG_IMQ) || defined(CONFIG_IMQ_MODULE)
1422 + * This should not happen. When it does, avoid memleak by restoring
1423 + * the chain of cb-backups.
1425 + while (skb->cb_next != NULL) {
1426 + if (net_ratelimit())
1427 + pr_warn("IMQ: kfree_skb: skb->cb_next: %08x\n",
1428 + (unsigned int)skb->cb_next);
1430 + skb_restore_cb(skb);
1433 + * This should not happen either, nf_queue_entry is nullified in
1434 + * imq_dev_xmit(). If we have non-NULL nf_queue_entry then we are
1435 + * leaking entry pointers, maybe memory. We don't know if this is
1436 + * pointer to already freed memory, or should this be freed.
1437 + * If this happens we need to add refcounting, etc for nf_queue_entry.
1439 + if (skb->nf_queue_entry && net_ratelimit())
1440 + pr_warn("%s\n", "IMQ: kfree_skb: skb->nf_queue_entry != NULL");
1442 #if IS_ENABLED(CONFIG_NF_CONNTRACK)
1443 nf_conntrack_put(skb->nfct);
1445 @@ -713,6 +814,10 @@
1446 new->sp = secpath_get(old->sp);
1448 memcpy(new->cb, old->cb, sizeof(old->cb));
1449 +#if defined(CONFIG_IMQ) || defined(CONFIG_IMQ_MODULE)
1450 + new->cb_next = NULL;
1451 + /*skb_copy_stored_cb(new, old);*/
1453 new->csum = old->csum;
1454 new->local_df = old->local_df;
1455 new->pkt_type = old->pkt_type;
1456 @@ -3093,6 +3198,13 @@
1458 SLAB_HWCACHE_ALIGN|SLAB_PANIC,
1460 +#if defined(CONFIG_IMQ) || defined(CONFIG_IMQ_MODULE)
1461 + skbuff_cb_store_cache = kmem_cache_create("skbuff_cb_store_cache",
1462 + sizeof(struct skb_cb_table),
1464 + SLAB_HWCACHE_ALIGN|SLAB_PANIC,
1470 diff -ruN linux-3.10.27/net/ipv6/ip6_output.c linux-3.10.27-imq/net/ipv6/ip6_output.c
1471 --- linux-3.10.27/net/ipv6/ip6_output.c 2014-01-16 00:29:14.000000000 +0100
1472 +++ linux-3.10.27-imq/net/ipv6/ip6_output.c 2014-01-18 10:19:59.348342972 +0100
1474 struct in6_addr *nexthop;
1477 - skb->protocol = htons(ETH_P_IPV6);
1480 if (ipv6_addr_is_multicast(&ipv6_hdr(skb)->daddr)) {
1481 struct inet6_dev *idev = ip6_dst_idev(skb_dst(skb));
1483 @@ -168,6 +165,13 @@
1488 + * IMQ-patch: moved setting skb->dev and skb->protocol from
1489 + * ip6_finish_output2 to fix crashing at netif_skb_features().
1491 + skb->protocol = htons(ETH_P_IPV6);
1494 return NF_HOOK_COND(NFPROTO_IPV6, NF_INET_POST_ROUTING, skb, NULL, dev,
1496 !(IP6CB(skb)->flags & IP6SKB_REROUTED));
1497 diff -ruN linux-3.10.27/net/netfilter/core.c linux-3.10.27-imq/net/netfilter/core.c
1498 --- linux-3.10.27/net/netfilter/core.c 2014-01-16 00:29:14.000000000 +0100
1499 +++ linux-3.10.27-imq/net/netfilter/core.c 2014-01-18 10:19:59.349342984 +0100
1500 @@ -191,9 +191,11 @@
1501 ret = NF_DROP_GETERR(verdict);
1504 - } else if ((verdict & NF_VERDICT_MASK) == NF_QUEUE) {
1505 + } else if ((verdict & NF_VERDICT_MASK) == NF_QUEUE ||
1506 + (verdict & NF_VERDICT_MASK) == NF_IMQ_QUEUE) {
1507 int err = nf_queue(skb, elem, pf, hook, indev, outdev, okfn,
1508 - verdict >> NF_VERDICT_QBITS);
1509 + verdict >> NF_VERDICT_QBITS,
1510 + verdict & NF_VERDICT_MASK);
1512 if (err == -ECANCELED)
1514 diff -ruN linux-3.10.27/net/netfilter/Kconfig linux-3.10.27-imq/net/netfilter/Kconfig
1515 --- linux-3.10.27/net/netfilter/Kconfig 2014-01-16 00:29:14.000000000 +0100
1516 +++ linux-3.10.27-imq/net/netfilter/Kconfig 2014-01-18 10:19:59.349342984 +0100
1517 @@ -641,6 +641,18 @@
1519 To compile it as a module, choose M here. If unsure, say N.
1521 +config NETFILTER_XT_TARGET_IMQ
1522 + tristate '"IMQ" target support'
1523 + depends on NETFILTER_XTABLES
1524 + depends on IP_NF_MANGLE || IP6_NF_MANGLE
1526 + default m if NETFILTER_ADVANCED=n
1528 + This option adds a `IMQ' target which is used to specify if and
1529 + to which imq device packets should get enqueued/dequeued.
1531 + To compile it as a module, choose M here. If unsure, say N.
1533 config NETFILTER_XT_TARGET_MARK
1534 tristate '"MARK" target support'
1535 depends on NETFILTER_ADVANCED
1536 diff -ruN linux-3.10.27/net/netfilter/Makefile linux-3.10.27-imq/net/netfilter/Makefile
1537 --- linux-3.10.27/net/netfilter/Makefile 2014-01-16 00:29:14.000000000 +0100
1538 +++ linux-3.10.27-imq/net/netfilter/Makefile 2014-01-18 10:19:59.349342984 +0100
1540 obj-$(CONFIG_NETFILTER_XT_TARGET_DSCP) += xt_DSCP.o
1541 obj-$(CONFIG_NETFILTER_XT_TARGET_HL) += xt_HL.o
1542 obj-$(CONFIG_NETFILTER_XT_TARGET_HMARK) += xt_HMARK.o
1543 +obj-$(CONFIG_NETFILTER_XT_TARGET_IMQ) += xt_IMQ.o
1544 obj-$(CONFIG_NETFILTER_XT_TARGET_LED) += xt_LED.o
1545 obj-$(CONFIG_NETFILTER_XT_TARGET_LOG) += xt_LOG.o
1546 obj-$(CONFIG_NETFILTER_XT_TARGET_NETMAP) += xt_NETMAP.o
1547 diff -ruN linux-3.10.27/net/netfilter/nf_internals.h linux-3.10.27-imq/net/netfilter/nf_internals.h
1548 --- linux-3.10.27/net/netfilter/nf_internals.h 2014-01-16 00:29:14.000000000 +0100
1549 +++ linux-3.10.27-imq/net/netfilter/nf_internals.h 2014-01-18 10:19:59.349342984 +0100
1551 struct net_device *indev,
1552 struct net_device *outdev,
1553 int (*okfn)(struct sk_buff *),
1554 - unsigned int queuenum);
1555 + unsigned int queuenum, unsigned int queuetype);
1556 extern int __init netfilter_queue_init(void);
1559 diff -ruN linux-3.10.27/net/netfilter/nf_queue.c linux-3.10.27-imq/net/netfilter/nf_queue.c
1560 --- linux-3.10.27/net/netfilter/nf_queue.c 2014-01-16 00:29:14.000000000 +0100
1561 +++ linux-3.10.27-imq/net/netfilter/nf_queue.c 2014-01-18 10:19:59.350342998 +0100
1564 static const struct nf_queue_handler __rcu *queue_handler __read_mostly;
1566 +#if defined(CONFIG_IMQ) || defined(CONFIG_IMQ_MODULE)
1567 +static const struct nf_queue_handler __rcu *queue_imq_handler __read_mostly;
1569 +void nf_register_queue_imq_handler(const struct nf_queue_handler *qh)
1571 + rcu_assign_pointer(queue_imq_handler, qh);
1573 +EXPORT_SYMBOL_GPL(nf_register_queue_imq_handler);
1575 +void nf_unregister_queue_imq_handler(void)
1577 + RCU_INIT_POINTER(queue_imq_handler, NULL);
1578 + synchronize_rcu();
1580 +EXPORT_SYMBOL_GPL(nf_unregister_queue_imq_handler);
1583 /* return EBUSY when somebody else is registered, return EEXIST if the
1584 * same handler is registered, return 0 in case of success. */
1585 void nf_register_queue_handler(const struct nf_queue_handler *qh)
1587 struct net_device *indev,
1588 struct net_device *outdev,
1589 int (*okfn)(struct sk_buff *),
1590 - unsigned int queuenum)
1591 + unsigned int queuenum,
1592 + unsigned int queuetype)
1594 int status = -ENOENT;
1595 struct nf_queue_entry *entry = NULL;
1596 @@ -115,7 +133,17 @@
1597 /* QUEUE == DROP if no one is waiting, to be safe. */
1600 - qh = rcu_dereference(queue_handler);
1601 + if (queuetype == NF_IMQ_QUEUE) {
1602 +#if defined(CONFIG_IMQ) || defined(CONFIG_IMQ_MODULE)
1603 + qh = rcu_dereference(queue_imq_handler);
1609 + qh = rcu_dereference(queue_handler);
1615 @@ -205,9 +233,11 @@
1619 + case NF_IMQ_QUEUE:
1620 err = nf_queue(skb, elem, entry->pf, entry->hook,
1621 entry->indev, entry->outdev, entry->okfn,
1622 - verdict >> NF_VERDICT_QBITS);
1623 + verdict >> NF_VERDICT_QBITS,
1624 + verdict & NF_VERDICT_MASK);
1626 if (err == -ECANCELED)
1628 diff -ruN linux-3.10.27/net/netfilter/xt_IMQ.c linux-3.10.27-imq/net/netfilter/xt_IMQ.c
1629 --- linux-3.10.27/net/netfilter/xt_IMQ.c 1970-01-01 01:00:00.000000000 +0100
1630 +++ linux-3.10.27-imq/net/netfilter/xt_IMQ.c 2014-01-18 10:19:59.350342998 +0100
1633 + * This target marks packets to be enqueued to an imq device
1635 +#include <linux/module.h>
1636 +#include <linux/skbuff.h>
1637 +#include <linux/netfilter/x_tables.h>
1638 +#include <linux/netfilter/xt_IMQ.h>
1639 +#include <linux/imq.h>
1641 +static unsigned int imq_target(struct sk_buff *pskb,
1642 + const struct xt_action_param *par)
1644 + const struct xt_imq_info *mr = par->targinfo;
1646 + pskb->imq_flags = (mr->todev & IMQ_F_IFMASK) | IMQ_F_ENQUEUE;
1648 + return XT_CONTINUE;
1651 +static int imq_checkentry(const struct xt_tgchk_param *par)
1653 + struct xt_imq_info *mr = par->targinfo;
1655 + if (mr->todev > IMQ_MAX_DEVS - 1) {
1656 + pr_warn("IMQ: invalid device specified, highest is %u\n",
1657 + IMQ_MAX_DEVS - 1);
1664 +static struct xt_target xt_imq_reg[] __read_mostly = {
1667 + .family = AF_INET,
1668 + .checkentry = imq_checkentry,
1669 + .target = imq_target,
1670 + .targetsize = sizeof(struct xt_imq_info),
1671 + .table = "mangle",
1676 + .family = AF_INET6,
1677 + .checkentry = imq_checkentry,
1678 + .target = imq_target,
1679 + .targetsize = sizeof(struct xt_imq_info),
1680 + .table = "mangle",
1685 +static int __init imq_init(void)
1687 + return xt_register_targets(xt_imq_reg, ARRAY_SIZE(xt_imq_reg));
1690 +static void __exit imq_fini(void)
1692 + xt_unregister_targets(xt_imq_reg, ARRAY_SIZE(xt_imq_reg));
1695 +module_init(imq_init);
1696 +module_exit(imq_fini);
1698 +MODULE_AUTHOR("http://www.linuximq.net");
1699 +MODULE_DESCRIPTION("Pseudo-driver for the intermediate queue device. See http://www.linuximq.net/ for more information.");
1700 +MODULE_LICENSE("GPL");
1701 +MODULE_ALIAS("ipt_IMQ");
1702 +MODULE_ALIAS("ip6t_IMQ");