]> git.pld-linux.org Git - packages/kernel.git/blob - kernel-imq.patch
- up to 4.2.1; builds on x8664; not tested
[packages/kernel.git] / kernel-imq.patch
1 diff --git a/drivers/net/Kconfig b/drivers/net/Kconfig
2 index c18f9e6..ec43bde 100644
3 --- a/drivers/net/Kconfig
4 +++ b/drivers/net/Kconfig
5 @@ -234,6 +234,125 @@ config RIONET_RX_SIZE
6         depends on RIONET
7         default "128"
8  
9 +config IMQ
10 +       tristate "IMQ (intermediate queueing device) support"
11 +       depends on NETDEVICES && NETFILTER
12 +       ---help---
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.
19 +
20 +         More information at: https://github.com/imq/linuximq
21 +
22 +         To compile this driver as a module, choose M here: the module
23 +         will be called imq.  If unsure, say N.
24 +
25 +choice
26 +       prompt "IMQ behavior (PRE/POSTROUTING)"
27 +       depends on IMQ
28 +       default IMQ_BEHAVIOR_AB
29 +       help
30 +         This setting defines how IMQ behaves in respect to its
31 +         hooking in PREROUTING and POSTROUTING.
32 +
33 +         IMQ can work in any of the following ways:
34 +
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
41 +
42 +         The default behavior is to hook before NAT on PREROUTING
43 +         and after NAT on POSTROUTING (#3).
44 +
45 +         This settings are specially usefull when trying to use IMQ
46 +         to shape NATed clients.
47 +
48 +         More information can be found at: https://github.com/imq/linuximq
49 +
50 +         If not sure leave the default settings alone.
51 +
52 +config IMQ_BEHAVIOR_AA
53 +       bool "IMQ AA"
54 +       help
55 +         This setting defines how IMQ behaves in respect to its
56 +         hooking in PREROUTING and POSTROUTING.
57 +
58 +         Choosing this option will make IMQ hook like this:
59 +
60 +         PREROUTING:   After NAT
61 +         POSTROUTING:  After NAT
62 +
63 +         More information can be found at: https://github.com/imq/linuximq
64 +
65 +         If not sure leave the default settings alone.
66 +
67 +config IMQ_BEHAVIOR_AB
68 +       bool "IMQ AB"
69 +       help
70 +         This setting defines how IMQ behaves in respect to its
71 +         hooking in PREROUTING and POSTROUTING.
72 +
73 +         Choosing this option will make IMQ hook like this:
74 +
75 +         PREROUTING:   After NAT
76 +         POSTROUTING:  Before NAT
77 +
78 +         More information can be found at: https://github.com/imq/linuximq
79 +
80 +         If not sure leave the default settings alone.
81 +
82 +config IMQ_BEHAVIOR_BA
83 +       bool "IMQ BA"
84 +       help
85 +         This setting defines how IMQ behaves in respect to its
86 +         hooking in PREROUTING and POSTROUTING.
87 +
88 +         Choosing this option will make IMQ hook like this:
89 +
90 +         PREROUTING:   Before NAT
91 +         POSTROUTING:  After NAT
92 +
93 +         More information can be found at: https://github.com/imq/linuximq
94 +
95 +         If not sure leave the default settings alone.
96 +
97 +config IMQ_BEHAVIOR_BB
98 +       bool "IMQ BB"
99 +       help
100 +         This setting defines how IMQ behaves in respect to its
101 +         hooking in PREROUTING and POSTROUTING.
102 +
103 +         Choosing this option will make IMQ hook like this:
104 +
105 +         PREROUTING:   Before NAT
106 +         POSTROUTING:  Before NAT
107 +
108 +         More information can be found at: https://github.com/imq/linuximq
109 +
110 +         If not sure leave the default settings alone.
111 +
112 +endchoice
113 +
114 +config IMQ_NUM_DEVS
115 +       int "Number of IMQ devices"
116 +       range 2 16
117 +       depends on IMQ
118 +       default "16"
119 +       help
120 +         This setting defines how many IMQ devices will be created.
121 +
122 +         The default value is 16.
123 +
124 +         More information can be found at: https://github.com/imq/linuximq
125 +
126 +         If not sure leave the default settings alone.
127 +
128  config TUN
129         tristate "Universal TUN/TAP device driver support"
130         depends on INET
131 diff --git a/drivers/net/Makefile b/drivers/net/Makefile
132 index c12cb22..03b82c6 100644
133 --- a/drivers/net/Makefile
134 +++ b/drivers/net/Makefile
135 @@ -10,6 +10,7 @@ obj-$(CONFIG_IPVLAN) += ipvlan/
136  obj-$(CONFIG_DUMMY) += dummy.o
137  obj-$(CONFIG_EQUALIZER) += eql.o
138  obj-$(CONFIG_IFB) += ifb.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
144 new file mode 100644
145 index 0000000..c60929b
146 --- /dev/null
147 +++ b/drivers/net/imq.c
148 @@ -0,0 +1,908 @@
149 +/*
150 + *             Pseudo-driver for the intermediate queue device.
151 + *
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.
156 + *
157 + * Authors:    Patrick McHardy, <kaber@trash.net>
158 + *
159 + *            The first version was written by Martin Devera, <devik@cdi.cz>
160 + *
161 + *                        See Creditis.txt
162 + */
163 +
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>
177 +#endif
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>
186 +#include <net/ip.h>
187 +#include <net/ipv6.h>
188 +
189 +static int imq_nf_queue(struct nf_queue_entry *entry, unsigned queue_num);
190 +
191 +static nf_hookfn imq_nf_hook;
192 +
193 +static struct nf_hook_ops imq_ops[] = {
194 +       {
195 +       /* imq_ingress_ipv4 */
196 +               .hook           = imq_nf_hook,
197 +               .owner          = THIS_MODULE,
198 +               .pf             = PF_INET,
199 +               .hooknum        = NF_INET_PRE_ROUTING,
200 +#if defined(CONFIG_IMQ_BEHAVIOR_BA) || defined(CONFIG_IMQ_BEHAVIOR_BB)
201 +               .priority       = NF_IP_PRI_MANGLE + 1,
202 +#else
203 +               .priority       = NF_IP_PRI_NAT_DST + 1,
204 +#endif
205 +       },
206 +       {
207 +       /* imq_egress_ipv4 */
208 +               .hook           = imq_nf_hook,
209 +               .owner          = THIS_MODULE,
210 +               .pf             = PF_INET,
211 +               .hooknum        = NF_INET_POST_ROUTING,
212 +#if defined(CONFIG_IMQ_BEHAVIOR_AA) || defined(CONFIG_IMQ_BEHAVIOR_BA)
213 +               .priority       = NF_IP_PRI_LAST,
214 +#else
215 +               .priority       = NF_IP_PRI_NAT_SRC - 1,
216 +#endif
217 +       },
218 +#if defined(CONFIG_IPV6) || defined(CONFIG_IPV6_MODULE)
219 +       {
220 +       /* imq_ingress_ipv6 */
221 +               .hook           = imq_nf_hook,
222 +               .owner          = THIS_MODULE,
223 +               .pf             = PF_INET6,
224 +               .hooknum        = NF_INET_PRE_ROUTING,
225 +#if defined(CONFIG_IMQ_BEHAVIOR_BA) || defined(CONFIG_IMQ_BEHAVIOR_BB)
226 +               .priority       = NF_IP6_PRI_MANGLE + 1,
227 +#else
228 +               .priority       = NF_IP6_PRI_NAT_DST + 1,
229 +#endif
230 +       },
231 +       {
232 +       /* imq_egress_ipv6 */
233 +               .hook           = imq_nf_hook,
234 +               .owner          = THIS_MODULE,
235 +               .pf             = PF_INET6,
236 +               .hooknum        = NF_INET_POST_ROUTING,
237 +#if defined(CONFIG_IMQ_BEHAVIOR_AA) || defined(CONFIG_IMQ_BEHAVIOR_BA)
238 +               .priority       = NF_IP6_PRI_LAST,
239 +#else
240 +               .priority       = NF_IP6_PRI_NAT_SRC - 1,
241 +#endif
242 +       },
243 +#endif
244 +};
245 +
246 +#if defined(CONFIG_IMQ_NUM_DEVS)
247 +static int numdevs = CONFIG_IMQ_NUM_DEVS;
248 +#else
249 +static int numdevs = IMQ_MAX_DEVS;
250 +#endif
251 +
252 +static struct net_device *imq_devs_cache[IMQ_MAX_DEVS];
253 +
254 +#define IMQ_MAX_QUEUES 32
255 +static int numqueues = 1;
256 +static u32 imq_hashrnd;
257 +static int imq_dev_accurate_stats = 1;
258 +
259 +static inline __be16 pppoe_proto(const struct sk_buff *skb)
260 +{
261 +       return *((__be16 *)(skb_mac_header(skb) + ETH_HLEN +
262 +                       sizeof(struct pppoe_hdr)));
263 +}
264 +
265 +static u16 imq_hash(struct net_device *dev, struct sk_buff *skb)
266 +{
267 +       unsigned int pull_len;
268 +       u16 protocol = skb->protocol;
269 +       u32 addr1, addr2;
270 +       u32 hash, ihl = 0;
271 +       union {
272 +               u16 in16[2];
273 +               u32 in32;
274 +       } ports;
275 +       u8 ip_proto;
276 +
277 +       pull_len = 0;
278 +
279 +recheck:
280 +       switch (protocol) {
281 +       case htons(ETH_P_8021Q): {
282 +               if (unlikely(skb_pull(skb, VLAN_HLEN) == NULL))
283 +                       goto other;
284 +
285 +               pull_len += VLAN_HLEN;
286 +               skb->network_header += VLAN_HLEN;
287 +
288 +               protocol = vlan_eth_hdr(skb)->h_vlan_encapsulated_proto;
289 +               goto recheck;
290 +       }
291 +
292 +       case htons(ETH_P_PPP_SES): {
293 +               if (unlikely(skb_pull(skb, PPPOE_SES_HLEN) == NULL))
294 +                       goto other;
295 +
296 +               pull_len += PPPOE_SES_HLEN;
297 +               skb->network_header += PPPOE_SES_HLEN;
298 +
299 +               protocol = pppoe_proto(skb);
300 +               goto recheck;
301 +       }
302 +
303 +       case htons(ETH_P_IP): {
304 +               const struct iphdr *iph = ip_hdr(skb);
305 +
306 +               if (unlikely(!pskb_may_pull(skb, sizeof(struct iphdr))))
307 +                       goto other;
308 +
309 +               addr1 = iph->daddr;
310 +               addr2 = iph->saddr;
311 +
312 +               ip_proto = !(ip_hdr(skb)->frag_off & htons(IP_MF | IP_OFFSET)) ?
313 +                                iph->protocol : 0;
314 +               ihl = ip_hdrlen(skb);
315 +
316 +               break;
317 +       }
318 +#if defined(CONFIG_IPV6) || defined(CONFIG_IPV6_MODULE)
319 +       case htons(ETH_P_IPV6): {
320 +               const struct ipv6hdr *iph = ipv6_hdr(skb);
321 +               __be16 fo = 0;
322 +
323 +               if (unlikely(!pskb_may_pull(skb, sizeof(struct ipv6hdr))))
324 +                       goto other;
325 +
326 +               addr1 = iph->daddr.s6_addr32[3];
327 +               addr2 = iph->saddr.s6_addr32[3];
328 +               ihl = ipv6_skip_exthdr(skb, sizeof(struct ipv6hdr), &ip_proto,
329 +                                      &fo);
330 +               if (unlikely(ihl < 0))
331 +                       goto other;
332 +
333 +               break;
334 +       }
335 +#endif
336 +       default:
337 +other:
338 +               if (pull_len != 0) {
339 +                       skb_push(skb, pull_len);
340 +                       skb->network_header -= pull_len;
341 +               }
342 +
343 +               return (u16)(ntohs(protocol) % dev->real_num_tx_queues);
344 +       }
345 +
346 +       if (addr1 > addr2)
347 +               swap(addr1, addr2);
348 +
349 +       switch (ip_proto) {
350 +       case IPPROTO_TCP:
351 +       case IPPROTO_UDP:
352 +       case IPPROTO_DCCP:
353 +       case IPPROTO_ESP:
354 +       case IPPROTO_AH:
355 +       case IPPROTO_SCTP:
356 +       case IPPROTO_UDPLITE: {
357 +               if (likely(skb_copy_bits(skb, ihl, &ports.in32, 4) >= 0)) {
358 +                       if (ports.in16[0] > ports.in16[1])
359 +                               swap(ports.in16[0], ports.in16[1]);
360 +                       break;
361 +               }
362 +               /* fall-through */
363 +       }
364 +       default:
365 +               ports.in32 = 0;
366 +               break;
367 +       }
368 +
369 +       if (pull_len != 0) {
370 +               skb_push(skb, pull_len);
371 +               skb->network_header -= pull_len;
372 +       }
373 +
374 +       hash = jhash_3words(addr1, addr2, ports.in32, imq_hashrnd ^ ip_proto);
375 +
376 +       return (u16)(((u64)hash * dev->real_num_tx_queues) >> 32);
377 +}
378 +
379 +static inline bool sk_tx_queue_recorded(struct sock *sk)
380 +{
381 +       return (sk_tx_queue_get(sk) >= 0);
382 +}
383 +
384 +static struct netdev_queue *imq_select_queue(struct net_device *dev,
385 +                                               struct sk_buff *skb)
386 +{
387 +       u16 queue_index = 0;
388 +       u32 hash;
389 +
390 +       if (likely(dev->real_num_tx_queues == 1))
391 +               goto out;
392 +
393 +       /* IMQ can be receiving ingress or engress packets. */
394 +
395 +       /* Check first for if rx_queue is set */
396 +       if (skb_rx_queue_recorded(skb)) {
397 +               queue_index = skb_get_rx_queue(skb);
398 +               goto out;
399 +       }
400 +
401 +       /* Check if socket has tx_queue set */
402 +       if (sk_tx_queue_recorded(skb->sk)) {
403 +               queue_index = sk_tx_queue_get(skb->sk);
404 +               goto out;
405 +       }
406 +
407 +       /* Try use socket hash */
408 +       if (skb->sk && skb->sk->sk_hash) {
409 +               hash = skb->sk->sk_hash;
410 +               queue_index =
411 +                       (u16)(((u64)hash * dev->real_num_tx_queues) >> 32);
412 +               goto out;
413 +       }
414 +
415 +       /* Generate hash from packet data */
416 +       queue_index = imq_hash(dev, skb);
417 +
418 +out:
419 +       if (unlikely(queue_index >= dev->real_num_tx_queues))
420 +               queue_index = (u16)((u32)queue_index % dev->real_num_tx_queues);
421 +
422 +       skb_set_queue_mapping(skb, queue_index);
423 +       return netdev_get_tx_queue(dev, queue_index);
424 +}
425 +
426 +static struct net_device_stats *imq_get_stats(struct net_device *dev)
427 +{
428 +       return &dev->stats;
429 +}
430 +
431 +/* called for packets kfree'd in qdiscs at places other than enqueue */
432 +static void imq_skb_destructor(struct sk_buff *skb)
433 +{
434 +       struct nf_queue_entry *entry = skb->nf_queue_entry;
435 +
436 +       skb->nf_queue_entry = NULL;
437 +
438 +       if (entry) {
439 +               nf_queue_entry_release_refs(entry);
440 +               kfree(entry);
441 +       }
442 +
443 +       skb_restore_cb(skb); /* kfree backup */
444 +}
445 +
446 +static void imq_done_check_queue_mapping(struct sk_buff *skb,
447 +                                        struct net_device *dev)
448 +{
449 +       unsigned int queue_index;
450 +
451 +       /* Don't let queue_mapping be left too large after exiting IMQ */
452 +       if (likely(skb->dev != dev && skb->dev != NULL)) {
453 +               queue_index = skb_get_queue_mapping(skb);
454 +               if (unlikely(queue_index >= skb->dev->real_num_tx_queues)) {
455 +                       queue_index = (u16)((u32)queue_index %
456 +                                               skb->dev->real_num_tx_queues);
457 +                       skb_set_queue_mapping(skb, queue_index);
458 +               }
459 +       } else {
460 +               /* skb->dev was IMQ device itself or NULL, be on safe side and
461 +                * just clear queue mapping.
462 +                */
463 +               skb_set_queue_mapping(skb, 0);
464 +       }
465 +}
466 +
467 +static netdev_tx_t imq_dev_xmit(struct sk_buff *skb, struct net_device *dev)
468 +{
469 +       struct nf_queue_entry *entry = skb->nf_queue_entry;
470 +
471 +       skb->nf_queue_entry = NULL;
472 +       dev->trans_start = jiffies;
473 +
474 +       dev->stats.tx_bytes += skb->len;
475 +       dev->stats.tx_packets++;
476 +
477 +       if (unlikely(entry == NULL)) {
478 +               /* We don't know what is going on here.. packet is queued for
479 +                * imq device, but (probably) not by us.
480 +                *
481 +                * If this packet was not send here by imq_nf_queue(), then
482 +                * skb_save_cb() was not used and skb_free() should not show:
483 +                *   WARNING: IMQ: kfree_skb: skb->cb_next:..
484 +                * and/or
485 +                *   WARNING: IMQ: kfree_skb: skb->nf_queue_entry...
486 +                *
487 +                * However if this message is shown, then IMQ is somehow broken
488 +                * and you should report this to linuximq.net.
489 +                */
490 +
491 +               /* imq_dev_xmit is black hole that eats all packets, report that
492 +                * we eat this packet happily and increase dropped counters.
493 +                */
494 +
495 +               dev->stats.tx_dropped++;
496 +               dev_kfree_skb(skb);
497 +
498 +               return NETDEV_TX_OK;
499 +       }
500 +
501 +       skb_restore_cb(skb); /* restore skb->cb */
502 +
503 +       skb->imq_flags = 0;
504 +       skb->destructor = NULL;
505 +
506 +       imq_done_check_queue_mapping(skb, dev);
507 +
508 +       nf_reinject(entry, NF_ACCEPT);
509 +
510 +       return NETDEV_TX_OK;
511 +}
512 +
513 +static struct net_device *get_imq_device_by_index(int index)
514 +{
515 +       struct net_device *dev = NULL;
516 +       struct net *net;
517 +       char buf[8];
518 +
519 +       /* get device by name and cache result */
520 +       snprintf(buf, sizeof(buf), "imq%d", index);
521 +
522 +       /* Search device from all namespaces. */
523 +       for_each_net(net) {
524 +               dev = dev_get_by_name(net, buf);
525 +               if (dev)
526 +                       break;
527 +       }
528 +
529 +       if (WARN_ON_ONCE(dev == NULL)) {
530 +               /* IMQ device not found. Exotic config? */
531 +               return ERR_PTR(-ENODEV);
532 +       }
533 +
534 +       imq_devs_cache[index] = dev;
535 +       dev_put(dev);
536 +
537 +       return dev;
538 +}
539 +
540 +static struct nf_queue_entry *nf_queue_entry_dup(struct nf_queue_entry *e)
541 +{
542 +       struct nf_queue_entry *entry = kmemdup(e, e->size, GFP_ATOMIC);
543 +       if (entry) {
544 +               if (nf_queue_entry_get_refs(entry))
545 +                       return entry;
546 +               kfree(entry);
547 +       }
548 +       return NULL;
549 +}
550 +
551 +#ifdef CONFIG_BRIDGE_NETFILTER
552 +/* When called from bridge netfilter, skb->data must point to MAC header
553 + * before calling skb_gso_segment(). Else, original MAC header is lost
554 + * and segmented skbs will be sent to wrong destination.
555 + */
556 +static void nf_bridge_adjust_skb_data(struct sk_buff *skb)
557 +{
558 +       if (skb->nf_bridge)
559 +               __skb_push(skb, skb->network_header - skb->mac_header);
560 +}
561 +
562 +static void nf_bridge_adjust_segmented_data(struct sk_buff *skb)
563 +{
564 +       if (skb->nf_bridge)
565 +               __skb_pull(skb, skb->network_header - skb->mac_header);
566 +}
567 +#else
568 +#define nf_bridge_adjust_skb_data(s) do {} while (0)
569 +#define nf_bridge_adjust_segmented_data(s) do {} while (0)
570 +#endif
571 +
572 +static void free_entry(struct nf_queue_entry *entry)
573 +{
574 +       nf_queue_entry_release_refs(entry);
575 +       kfree(entry);
576 +}
577 +
578 +static int __imq_nf_queue(struct nf_queue_entry *entry, struct net_device *dev);
579 +
580 +static int __imq_nf_queue_gso(struct nf_queue_entry *entry,
581 +                             struct net_device *dev, struct sk_buff *skb)
582 +{
583 +       int ret = -ENOMEM;
584 +       struct nf_queue_entry *entry_seg;
585 +
586 +       nf_bridge_adjust_segmented_data(skb);
587 +
588 +       if (skb->next == NULL) { /* last packet, no need to copy entry */
589 +               struct sk_buff *gso_skb = entry->skb;
590 +               entry->skb = skb;
591 +               ret = __imq_nf_queue(entry, dev);
592 +               if (ret)
593 +                       entry->skb = gso_skb;
594 +               return ret;
595 +       }
596 +
597 +       skb->next = NULL;
598 +
599 +       entry_seg = nf_queue_entry_dup(entry);
600 +       if (entry_seg) {
601 +               entry_seg->skb = skb;
602 +               ret = __imq_nf_queue(entry_seg, dev);
603 +               if (ret)
604 +                       free_entry(entry_seg);
605 +       }
606 +       return ret;
607 +}
608 +
609 +static int imq_nf_queue(struct nf_queue_entry *entry, unsigned queue_num)
610 +{
611 +       struct sk_buff *skb, *segs;
612 +       struct net_device *dev;
613 +       unsigned int queued;
614 +       int index, retval, err;
615 +
616 +       index = entry->skb->imq_flags & IMQ_F_IFMASK;
617 +       if (unlikely(index > numdevs - 1)) {
618 +               if (net_ratelimit())
619 +                       pr_warn("IMQ: invalid device specified, highest is %u\n",
620 +                               numdevs - 1);
621 +               retval = -EINVAL;
622 +               goto out_no_dev;
623 +       }
624 +
625 +       /* check for imq device by index from cache */
626 +       dev = imq_devs_cache[index];
627 +       if (unlikely(!dev)) {
628 +               dev = get_imq_device_by_index(index);
629 +               if (IS_ERR(dev)) {
630 +                       retval = PTR_ERR(dev);
631 +                       goto out_no_dev;
632 +               }
633 +       }
634 +
635 +       if (unlikely(!(dev->flags & IFF_UP))) {
636 +               entry->skb->imq_flags = 0;
637 +               retval = -ECANCELED;
638 +               goto out_no_dev;
639 +       }
640 +
641 +       /* Since 3.10.x, GSO handling moved here as result of upstream commit
642 +        * a5fedd43d5f6c94c71053a66e4c3d2e35f1731a2 (netfilter: move
643 +        * skb_gso_segment into nfnetlink_queue module).
644 +        *
645 +        * Following code replicates the gso handling from
646 +        * 'net/netfilter/nfnetlink_queue_core.c':nfqnl_enqueue_packet().
647 +        */
648 +
649 +       skb = entry->skb;
650 +
651 +       switch (entry->state.pf) {
652 +       case NFPROTO_IPV4:
653 +               skb->protocol = htons(ETH_P_IP);
654 +               break;
655 +       case NFPROTO_IPV6:
656 +               skb->protocol = htons(ETH_P_IPV6);
657 +               break;
658 +       }
659 +
660 +       if (!skb_is_gso(entry->skb))
661 +               return __imq_nf_queue(entry, dev);
662 +
663 +       nf_bridge_adjust_skb_data(skb);
664 +       segs = skb_gso_segment(skb, 0);
665 +       /* Does not use PTR_ERR to limit the number of error codes that can be
666 +        * returned by nf_queue.  For instance, callers rely on -ECANCELED to
667 +        * mean 'ignore this hook'.
668 +        */
669 +       err = -ENOBUFS;
670 +       if (IS_ERR(segs))
671 +               goto out_err;
672 +       queued = 0;
673 +       err = 0;
674 +       do {
675 +               struct sk_buff *nskb = segs->next;
676 +               if (nskb && nskb->next)
677 +                       nskb->cb_next = NULL;
678 +               if (err == 0)
679 +                       err = __imq_nf_queue_gso(entry, dev, segs);
680 +               if (err == 0)
681 +                       queued++;
682 +               else
683 +                       kfree_skb(segs);
684 +               segs = nskb;
685 +       } while (segs);
686 +
687 +       if (queued) {
688 +               if (err) /* some segments are already queued */
689 +                       free_entry(entry);
690 +               kfree_skb(skb);
691 +               return 0;
692 +       }
693 +
694 +out_err:
695 +       nf_bridge_adjust_segmented_data(skb);
696 +       retval = err;
697 +out_no_dev:
698 +       return retval;
699 +}
700 +
701 +static int __imq_nf_queue(struct nf_queue_entry *entry, struct net_device *dev)
702 +{
703 +       struct sk_buff *skb_orig, *skb, *skb_shared, *skb_popd;
704 +       struct Qdisc *q;
705 +       struct netdev_queue *txq;
706 +       spinlock_t *root_lock;
707 +       int users;
708 +       int retval = -EINVAL;
709 +       unsigned int orig_queue_index;
710 +
711 +       dev->last_rx = jiffies;
712 +
713 +       skb = entry->skb;
714 +       skb_orig = NULL;
715 +
716 +       /* skb has owner? => make clone */
717 +       if (unlikely(skb->destructor)) {
718 +               skb_orig = skb;
719 +               skb = skb_clone(skb, GFP_ATOMIC);
720 +               if (unlikely(!skb)) {
721 +                       retval = -ENOMEM;
722 +                       goto out;
723 +               }
724 +               skb->cb_next = NULL;
725 +               entry->skb = skb;
726 +       }
727 +
728 +       dev->stats.rx_bytes += skb->len;
729 +       dev->stats.rx_packets++;
730 +
731 +       if (!skb->dev) {
732 +               /* skb->dev == NULL causes problems, try the find cause. */
733 +               if (net_ratelimit()) {
734 +                       dev_warn(&dev->dev,
735 +                                "received packet with skb->dev == NULL\n");
736 +                       dump_stack();
737 +               }
738 +
739 +               skb->dev = dev;
740 +       }
741 +
742 +       /* Disables softirqs for lock below */
743 +       rcu_read_lock_bh();
744 +
745 +       /* Multi-queue selection */
746 +       orig_queue_index = skb_get_queue_mapping(skb);
747 +       txq = imq_select_queue(dev, skb);
748 +
749 +       q = rcu_dereference(txq->qdisc);
750 +       if (unlikely(!q->enqueue))
751 +               goto packet_not_eaten_by_imq_dev;
752 +
753 +       skb->nf_queue_entry = entry;
754 +       root_lock = qdisc_lock(q);
755 +       spin_lock(root_lock);
756 +
757 +       users = atomic_read(&skb->users);
758 +
759 +       skb_shared = skb_get(skb); /* increase reference count by one */
760 +
761 +       /* backup skb->cb, as qdisc layer will overwrite it */
762 +       skb_save_cb(skb_shared);
763 +       qdisc_enqueue_root(skb_shared, q); /* might kfree_skb */
764 +       if (likely(atomic_read(&skb_shared->users) == users + 1)) {
765 +               bool validate;
766 +
767 +               kfree_skb(skb_shared); /* decrease reference count by one */
768 +
769 +               skb->destructor = &imq_skb_destructor;
770 +
771 +               skb_popd = qdisc_dequeue_skb(q, &validate);
772 +
773 +               /* cloned? */
774 +               if (unlikely(skb_orig))
775 +                       kfree_skb(skb_orig); /* free original */
776 +
777 +               spin_unlock(root_lock);
778 +
779 +#if 0
780 +               /* schedule qdisc dequeue */
781 +               __netif_schedule(q);
782 +#else
783 +               if (likely(skb_popd)) {
784 +                       /* Note that we validate skb (GSO, checksum, ...) outside of locks */
785 +                       if (validate)
786 +                       skb_popd = validate_xmit_skb_list(skb_popd, dev);
787 +                       
788 +                       if (skb_popd) {
789 +                               int dummy_ret;
790 +                               int cpu = smp_processor_id(); /* ok because BHs are off */
791 +
792 +                               txq = skb_get_tx_queue(dev, skb_popd);
793 +                               /* 
794 +                               IMQ device will not be frozen or stoped, and it always be successful.
795 +                               So we need not check its status and return value to accelerate.
796 +                               */
797 +                               if (imq_dev_accurate_stats && txq->xmit_lock_owner != cpu) {
798 +                                       HARD_TX_LOCK(dev, txq, cpu);
799 +                                       if (!netif_xmit_frozen_or_stopped(txq)) {
800 +                                               dev_hard_start_xmit(skb_popd, dev, txq, &dummy_ret);
801 +                                       }
802 +                                       HARD_TX_UNLOCK(dev, txq);
803 +                               } else {
804 +                                       if (!netif_xmit_frozen_or_stopped(txq)) {
805 +                                               dev_hard_start_xmit(skb_popd, dev, txq, &dummy_ret);
806 +                                       }
807 +                               }
808 +                       }
809 +               } else {
810 +                       /* No ready skb, then schedule it */
811 +                       __netif_schedule(q);
812 +               }
813 +#endif
814 +               rcu_read_unlock_bh();
815 +               retval = 0;
816 +               goto out;
817 +       } else {
818 +               skb_restore_cb(skb_shared); /* restore skb->cb */
819 +               skb->nf_queue_entry = NULL;
820 +               /*
821 +                * qdisc dropped packet and decreased skb reference count of
822 +                * skb, so we don't really want to and try refree as that would
823 +                * actually destroy the skb.
824 +                */
825 +               spin_unlock(root_lock);
826 +               goto packet_not_eaten_by_imq_dev;
827 +       }
828 +
829 +packet_not_eaten_by_imq_dev:
830 +       skb_set_queue_mapping(skb, orig_queue_index);
831 +       rcu_read_unlock_bh();
832 +
833 +       /* cloned? restore original */
834 +       if (unlikely(skb_orig)) {
835 +               kfree_skb(skb);
836 +               entry->skb = skb_orig;
837 +       }
838 +       retval = -1;
839 +out:
840 +       return retval;
841 +}
842 +static unsigned int imq_nf_hook(const struct nf_hook_ops *hook_ops,
843 +                               struct sk_buff *skb,
844 +                               const struct nf_hook_state *state)
845 +{
846 +       return (skb->imq_flags & IMQ_F_ENQUEUE) ? NF_IMQ_QUEUE : NF_ACCEPT;
847 +}
848 +
849 +static int imq_close(struct net_device *dev)
850 +{
851 +       netif_stop_queue(dev);
852 +       return 0;
853 +}
854 +
855 +static int imq_open(struct net_device *dev)
856 +{
857 +       netif_start_queue(dev);
858 +       return 0;
859 +}
860 +
861 +static const struct net_device_ops imq_netdev_ops = {
862 +       .ndo_open               = imq_open,
863 +       .ndo_stop               = imq_close,
864 +       .ndo_start_xmit         = imq_dev_xmit,
865 +       .ndo_get_stats          = imq_get_stats,
866 +};
867 +
868 +static void imq_setup(struct net_device *dev)
869 +{
870 +       dev->netdev_ops         = &imq_netdev_ops;
871 +       dev->type               = ARPHRD_VOID;
872 +       dev->mtu                = 16000; /* too small? */
873 +       dev->tx_queue_len       = 11000; /* too big? */
874 +       dev->flags              = IFF_NOARP;
875 +       dev->features           = NETIF_F_SG | NETIF_F_FRAGLIST |
876 +                                 NETIF_F_GSO | NETIF_F_HW_CSUM |
877 +                                 NETIF_F_HIGHDMA;
878 +       dev->priv_flags         &= ~(IFF_XMIT_DST_RELEASE |
879 +                                    IFF_TX_SKB_SHARING);
880 +}
881 +
882 +static int imq_validate(struct nlattr *tb[], struct nlattr *data[])
883 +{
884 +       int ret = 0;
885 +
886 +       if (tb[IFLA_ADDRESS]) {
887 +               if (nla_len(tb[IFLA_ADDRESS]) != ETH_ALEN) {
888 +                       ret = -EINVAL;
889 +                       goto end;
890 +               }
891 +               if (!is_valid_ether_addr(nla_data(tb[IFLA_ADDRESS]))) {
892 +                       ret = -EADDRNOTAVAIL;
893 +                       goto end;
894 +               }
895 +       }
896 +       return 0;
897 +end:
898 +       pr_warn("IMQ: imq_validate failed (%d)\n", ret);
899 +       return ret;
900 +}
901 +
902 +static struct rtnl_link_ops imq_link_ops __read_mostly = {
903 +       .kind           = "imq",
904 +       .priv_size      = 0,
905 +       .setup          = imq_setup,
906 +       .validate       = imq_validate,
907 +};
908 +
909 +static const struct nf_queue_handler imq_nfqh = {
910 +       .outfn = imq_nf_queue,
911 +};
912 +
913 +static int __init imq_init_hooks(void)
914 +{
915 +       int ret;
916 +
917 +       nf_register_queue_imq_handler(&imq_nfqh);
918 +
919 +       ret = nf_register_hooks(imq_ops, ARRAY_SIZE(imq_ops));
920 +       if (ret < 0)
921 +               nf_unregister_queue_imq_handler();
922 +
923 +       return ret;
924 +}
925 +
926 +static int __init imq_init_one(int index)
927 +{
928 +       struct net_device *dev;
929 +       int ret;
930 +
931 +       dev = alloc_netdev_mq(0, "imq%d", NET_NAME_UNKNOWN, imq_setup, numqueues);
932 +       if (!dev)
933 +               return -ENOMEM;
934 +
935 +       ret = dev_alloc_name(dev, dev->name);
936 +       if (ret < 0)
937 +               goto fail;
938 +
939 +       dev->rtnl_link_ops = &imq_link_ops;
940 +       ret = register_netdevice(dev);
941 +       if (ret < 0)
942 +               goto fail;
943 +
944 +       return 0;
945 +fail:
946 +       free_netdev(dev);
947 +       return ret;
948 +}
949 +
950 +static int __init imq_init_devs(void)
951 +{
952 +       int err, i;
953 +
954 +       if (numdevs < 1 || numdevs > IMQ_MAX_DEVS) {
955 +               pr_err("IMQ: numdevs has to be betweed 1 and %u\n",
956 +                      IMQ_MAX_DEVS);
957 +               return -EINVAL;
958 +       }
959 +
960 +       if (numqueues < 1 || numqueues > IMQ_MAX_QUEUES) {
961 +               pr_err("IMQ: numqueues has to be betweed 1 and %u\n",
962 +                      IMQ_MAX_QUEUES);
963 +               return -EINVAL;
964 +       }
965 +
966 +       get_random_bytes(&imq_hashrnd, sizeof(imq_hashrnd));
967 +
968 +       rtnl_lock();
969 +       err = __rtnl_link_register(&imq_link_ops);
970 +
971 +       for (i = 0; i < numdevs && !err; i++)
972 +               err = imq_init_one(i);
973 +
974 +       if (err) {
975 +               __rtnl_link_unregister(&imq_link_ops);
976 +               memset(imq_devs_cache, 0, sizeof(imq_devs_cache));
977 +       }
978 +       rtnl_unlock();
979 +
980 +       return err;
981 +}
982 +
983 +static int __init imq_init_module(void)
984 +{
985 +       int err;
986 +
987 +#if defined(CONFIG_IMQ_NUM_DEVS)
988 +       BUILD_BUG_ON(CONFIG_IMQ_NUM_DEVS > 16);
989 +       BUILD_BUG_ON(CONFIG_IMQ_NUM_DEVS < 2);
990 +       BUILD_BUG_ON(CONFIG_IMQ_NUM_DEVS - 1 > IMQ_F_IFMASK);
991 +#endif
992 +
993 +       err = imq_init_devs();
994 +       if (err) {
995 +               pr_err("IMQ: Error trying imq_init_devs(net)\n");
996 +               return err;
997 +       }
998 +
999 +       err = imq_init_hooks();
1000 +       if (err) {
1001 +               pr_err(KERN_ERR "IMQ: Error trying imq_init_hooks()\n");
1002 +               rtnl_link_unregister(&imq_link_ops);
1003 +               memset(imq_devs_cache, 0, sizeof(imq_devs_cache));
1004 +               return err;
1005 +       }
1006 +
1007 +       pr_info("IMQ driver loaded successfully. (numdevs = %d, numqueues = %d, imq_dev_accurate_stats = %d)\n",
1008 +               numdevs, numqueues, imq_dev_accurate_stats);
1009 +
1010 +#if defined(CONFIG_IMQ_BEHAVIOR_BA) || defined(CONFIG_IMQ_BEHAVIOR_BB)
1011 +       pr_info("\tHooking IMQ before NAT on PREROUTING.\n");
1012 +#else
1013 +       pr_info("\tHooking IMQ after NAT on PREROUTING.\n");
1014 +#endif
1015 +#if defined(CONFIG_IMQ_BEHAVIOR_AB) || defined(CONFIG_IMQ_BEHAVIOR_BB)
1016 +       pr_info("\tHooking IMQ before NAT on POSTROUTING.\n");
1017 +#else
1018 +       pr_info("\tHooking IMQ after NAT on POSTROUTING.\n");
1019 +#endif
1020 +
1021 +       return 0;
1022 +}
1023 +
1024 +static void __exit imq_unhook(void)
1025 +{
1026 +       nf_unregister_hooks(imq_ops, ARRAY_SIZE(imq_ops));
1027 +       nf_unregister_queue_imq_handler();
1028 +}
1029 +
1030 +static void __exit imq_cleanup_devs(void)
1031 +{
1032 +       rtnl_link_unregister(&imq_link_ops);
1033 +       memset(imq_devs_cache, 0, sizeof(imq_devs_cache));
1034 +}
1035 +
1036 +static void __exit imq_exit_module(void)
1037 +{
1038 +       imq_unhook();
1039 +       imq_cleanup_devs();
1040 +       pr_info("IMQ driver unloaded successfully.\n");
1041 +}
1042 +
1043 +module_init(imq_init_module);
1044 +module_exit(imq_exit_module);
1045 +
1046 +module_param(numdevs, int, 0);
1047 +module_param(numqueues, int, 0);
1048 +module_param(imq_dev_accurate_stats, int, 0);
1049 +MODULE_PARM_DESC(numdevs, "number of IMQ devices (how many imq* devices will be created)");
1050 +MODULE_PARM_DESC(numqueues, "number of queues per IMQ device");
1051 +MODULE_PARM_DESC(imq_dev_accurate_stats, "Notify if need the accurate imq device stats");
1052 +
1053 +MODULE_AUTHOR("http://https://github.com/imq/linuximq");
1054 +MODULE_DESCRIPTION("Pseudo-driver for the intermediate queue device. See https://github.com/imq/linuximq/wiki for more information.");
1055 +MODULE_LICENSE("GPL");
1056 +MODULE_ALIAS_RTNL_LINK("imq");
1057 diff --git a/include/linux/imq.h b/include/linux/imq.h
1058 new file mode 100644
1059 index 0000000..1babb09
1060 --- /dev/null
1061 +++ b/include/linux/imq.h
1062 @@ -0,0 +1,13 @@
1063 +#ifndef _IMQ_H
1064 +#define _IMQ_H
1065 +
1066 +/* IFMASK (16 device indexes, 0 to 15) and flag(s) fit in 5 bits */
1067 +#define IMQ_F_BITS     5
1068 +
1069 +#define IMQ_F_IFMASK   0x0f
1070 +#define IMQ_F_ENQUEUE  0x10
1071 +
1072 +#define IMQ_MAX_DEVS   (IMQ_F_IFMASK + 1)
1073 +
1074 +#endif /* _IMQ_H */
1075 +
1076 diff --git a/include/linux/netdevice.h b/include/linux/netdevice.h
1077 index e20979d..9c8f9a1 100644
1078 --- a/include/linux/netdevice.h
1079 +++ b/include/linux/netdevice.h
1080 @@ -3279,6 +3279,19 @@ static inline void netif_tx_unlock_bh(struct net_device *dev)
1081         }                                               \
1082  }
1083  
1084 +#define HARD_TX_LOCK_BH(dev, txq) {           \
1085 +    if ((dev->features & NETIF_F_LLTX) == 0) {  \
1086 +        __netif_tx_lock_bh(txq);      \
1087 +    }                       \
1088 +}
1089 +
1090 +#define HARD_TX_UNLOCK_BH(dev, txq) {          \
1091 +    if ((dev->features & NETIF_F_LLTX) == 0) {  \
1092 +        __netif_tx_unlock_bh(txq);         \
1093 +    }                       \
1094 +}
1095 +
1096 +
1097  static inline void netif_tx_disable(struct net_device *dev)
1098  {
1099         unsigned int i;
1100 diff --git a/include/linux/netfilter/xt_IMQ.h b/include/linux/netfilter/xt_IMQ.h
1101 new file mode 100644
1102 index 0000000..9b07230
1103 --- /dev/null
1104 +++ b/include/linux/netfilter/xt_IMQ.h
1105 @@ -0,0 +1,9 @@
1106 +#ifndef _XT_IMQ_H
1107 +#define _XT_IMQ_H
1108 +
1109 +struct xt_imq_info {
1110 +       unsigned int todev;     /* target imq device */
1111 +};
1112 +
1113 +#endif /* _XT_IMQ_H */
1114 +
1115 diff --git a/include/linux/netfilter_ipv4/ipt_IMQ.h b/include/linux/netfilter_ipv4/ipt_IMQ.h
1116 new file mode 100644
1117 index 0000000..7af320f
1118 --- /dev/null
1119 +++ b/include/linux/netfilter_ipv4/ipt_IMQ.h
1120 @@ -0,0 +1,10 @@
1121 +#ifndef _IPT_IMQ_H
1122 +#define _IPT_IMQ_H
1123 +
1124 +/* Backwards compatibility for old userspace */
1125 +#include <linux/netfilter/xt_IMQ.h>
1126 +
1127 +#define ipt_imq_info xt_imq_info
1128 +
1129 +#endif /* _IPT_IMQ_H */
1130 +
1131 diff --git a/include/linux/netfilter_ipv6/ip6t_IMQ.h b/include/linux/netfilter_ipv6/ip6t_IMQ.h
1132 new file mode 100644
1133 index 0000000..198ac01
1134 --- /dev/null
1135 +++ b/include/linux/netfilter_ipv6/ip6t_IMQ.h
1136 @@ -0,0 +1,10 @@
1137 +#ifndef _IP6T_IMQ_H
1138 +#define _IP6T_IMQ_H
1139 +
1140 +/* Backwards compatibility for old userspace */
1141 +#include <linux/netfilter/xt_IMQ.h>
1142 +
1143 +#define ip6t_imq_info xt_imq_info
1144 +
1145 +#endif /* _IP6T_IMQ_H */
1146 +
1147 diff --git a/include/linux/skbuff.h b/include/linux/skbuff.h
1148 index 9b88536..61686b0 100644
1149 --- a/include/linux/skbuff.h
1150 +++ b/include/linux/skbuff.h
1151 @@ -37,6 +37,9 @@
1152  #include <net/flow_dissector.h>
1153  #include <linux/splice.h>
1154  #include <linux/in6.h>
1155 +#if defined(CONFIG_IMQ) || defined(CONFIG_IMQ_MODULE)
1156 +#include <linux/imq.h>
1157 +#endif
1158  
1159  /* A. Checksumming of received packets by device.
1160   *
1161 @@ -548,6 +551,9 @@ struct sk_buff {
1162          * first. This is owned by whoever has the skb queued ATM.
1163          */
1164         char                    cb[48] __aligned(8);
1165 +#if defined(CONFIG_IMQ) || defined(CONFIG_IMQ_MODULE)
1166 +       void                    *cb_next;
1167 +#endif
1168  
1169         unsigned long           _skb_refdst;
1170         void                    (*destructor)(struct sk_buff *skb);
1171 @@ -557,6 +563,9 @@ struct sk_buff {
1172  #if defined(CONFIG_NF_CONNTRACK) || defined(CONFIG_NF_CONNTRACK_MODULE)
1173         struct nf_conntrack     *nfct;
1174  #endif
1175 +#if defined(CONFIG_IMQ) || defined(CONFIG_IMQ_MODULE)
1176 +       struct nf_queue_entry   *nf_queue_entry;
1177 +#endif
1178  #if IS_ENABLED(CONFIG_BRIDGE_NETFILTER)
1179         struct nf_bridge_info   *nf_bridge;
1180  #endif
1181 @@ -624,6 +633,9 @@ struct sk_buff {
1182         __u8                    inner_protocol_type:1;
1183         __u8                    remcsum_offload:1;
1184         /* 3 or 5 bit hole */
1185 +       #if defined(CONFIG_IMQ) || defined(CONFIG_IMQ_MODULE)
1186 +       __u8                    imq_flags:IMQ_F_BITS;
1187 +       #endif
1188  
1189  #ifdef CONFIG_NET_SCHED
1190         __u16                   tc_index;       /* traffic control index */
1191 @@ -774,6 +786,12 @@ 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);
1195 +
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);
1199 +#endif
1200 +
1201  extern struct kmem_cache *skbuff_head_cache;
1202  
1203  void kfree_skb_partial(struct sk_buff *skb, bool head_stolen);
1204 @@ -3232,6 +3250,10 @@ static inline void __nf_copy(struct sk_buff *dst, const struct sk_buff *src,
1205         if (copy)
1206                 dst->nfctinfo = src->nfctinfo;
1207  #endif
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;
1211 +#endif
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 e863585..40904cb 100644
1217 --- a/include/net/netfilter/nf_queue.h
1218 +++ b/include/net/netfilter/nf_queue.h
1219 @@ -31,6 +31,12 @@ struct nf_queue_handler {
1220  void nf_register_queue_handler(const struct nf_queue_handler *qh);
1221  void nf_unregister_queue_handler(void);
1222  void nf_reinject(struct nf_queue_entry *entry, unsigned int verdict);
1223 +void nf_queue_entry_release_refs(struct nf_queue_entry *entry);
1224 +
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);
1228 +#endif
1229  
1230  bool 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 2342bf1..149dec9 100644
1234 --- a/include/net/pkt_sched.h
1235 +++ b/include/net/pkt_sched.h
1236 @@ -104,6 +104,8 @@ int sch_direct_xmit(struct sk_buff *skb, struct Qdisc *q,
1237  
1238  void __qdisc_run(struct Qdisc *q);
1239  
1240 +struct sk_buff *qdisc_dequeue_skb(struct Qdisc *q, bool *validate);
1241 +
1242  static inline void qdisc_run(struct Qdisc *q)
1243  {
1244         if (qdisc_run_begin(q))
1245 diff --git a/include/net/sch_generic.h b/include/net/sch_generic.h
1246 index 2738f6f..cc0af3e 100644
1247 --- a/include/net/sch_generic.h
1248 +++ b/include/net/sch_generic.h
1249 @@ -501,6 +501,12 @@ static inline int qdisc_enqueue(struct sk_buff *skb, struct Qdisc *sch)
1250         return sch->enqueue(skb, sch);
1251  }
1252  
1253 +static inline int qdisc_enqueue_root(struct sk_buff *skb, struct Qdisc *sch)
1254 +{
1255 +    qdisc_skb_cb(skb)->pkt_len = skb->len;
1256 +    return qdisc_enqueue(skb, sch) & NET_XMIT_MASK;
1257 +}
1258 +
1259  static inline bool qdisc_is_percpu_stats(const struct Qdisc *q)
1260  {
1261         return q->flags & TCQ_F_CPUSTATS;
1262 diff --git a/include/uapi/linux/netfilter.h b/include/uapi/linux/netfilter.h
1263 index d93f949..23fb6d1 100644
1264 --- a/include/uapi/linux/netfilter.h
1265 +++ b/include/uapi/linux/netfilter.h
1266 @@ -14,7 +14,8 @@
1267  #define NF_QUEUE 3
1268  #define NF_REPEAT 4
1269  #define NF_STOP 5
1270 -#define NF_MAX_VERDICT NF_STOP
1271 +#define NF_IMQ_QUEUE 6
1272 +#define NF_MAX_VERDICT NF_IMQ_QUEUE
1273  
1274  /* we overload the higher bits for encoding auxiliary data such as the queue
1275   * number or errno values. Not nice, but better than additional function
1276 diff --git a/net/core/dev.c b/net/core/dev.c
1277 index a8e4dd4..f84cd5a 100644
1278 --- a/net/core/dev.c
1279 +++ b/net/core/dev.c
1280 @@ -136,6 +136,9 @@
1281  #include <linux/errqueue.h>
1282  #include <linux/hrtimer.h>
1283  #include <linux/netfilter_ingress.h>
1284 +#if defined(CONFIG_IMQ) || defined(CONFIG_IMQ_MODULE)
1285 +#include <linux/imq.h>
1286 +#endif
1287  
1288  #include "net-sysfs.h"
1289  
1290 @@ -2675,7 +2678,12 @@ static int xmit_one(struct sk_buff *skb, struct net_device *dev,
1291         unsigned int len;
1292         int rc;
1293  
1294 +#if defined(CONFIG_IMQ) || defined(CONFIG_IMQ_MODULE)
1295 +       if ((!list_empty(&ptype_all) || !list_empty(&dev->ptype_all)) &&
1296 +               !(skb->imq_flags & IMQ_F_ENQUEUE))
1297 +#else
1298         if (!list_empty(&ptype_all) || !list_empty(&dev->ptype_all))
1299 +#endif
1300                 dev_queue_xmit_nit(skb, dev);
1301  
1302         len = skb->len;
1303 @@ -2713,6 +2721,7 @@ out:
1304         *ret = rc;
1305         return skb;
1306  }
1307 +EXPORT_SYMBOL(dev_hard_start_xmit);
1308  
1309  static struct sk_buff *validate_xmit_vlan(struct sk_buff *skb,
1310                                           netdev_features_t features)
1311 @@ -2801,6 +2810,7 @@ struct sk_buff *validate_xmit_skb_list(struct sk_buff *skb, struct net_device *d
1312         }
1313         return head;
1314  }
1315 +EXPORT_SYMBOL(validate_xmit_skb_list);
1316  
1317  static void qdisc_pkt_len_init(struct sk_buff *skb)
1318  {
1319 diff --git a/net/core/skbuff.c b/net/core/skbuff.c
1320 index 7b84330..a313d22 100644
1321 --- a/net/core/skbuff.c
1322 +++ b/net/core/skbuff.c
1323 @@ -79,6 +79,87 @@
1324  
1325  struct kmem_cache *skbuff_head_cache __read_mostly;
1326  static struct kmem_cache *skbuff_fclone_cache __read_mostly;
1327 +#if defined(CONFIG_IMQ) || defined(CONFIG_IMQ_MODULE)
1328 +static struct kmem_cache *skbuff_cb_store_cache __read_mostly;
1329 +#endif
1330 +
1331 +#if defined(CONFIG_IMQ) || defined(CONFIG_IMQ_MODULE)
1332 +/* Control buffer save/restore for IMQ devices */
1333 +struct skb_cb_table {
1334 +       char                    cb[48] __aligned(8);
1335 +       void                    *cb_next;
1336 +       atomic_t                refcnt;
1337 +};
1338 +
1339 +static DEFINE_SPINLOCK(skb_cb_store_lock);
1340 +
1341 +int skb_save_cb(struct sk_buff *skb)
1342 +{
1343 +       struct skb_cb_table *next;
1344 +
1345 +       next = kmem_cache_alloc(skbuff_cb_store_cache, GFP_ATOMIC);
1346 +       if (!next)
1347 +               return -ENOMEM;
1348 +
1349 +       BUILD_BUG_ON(sizeof(skb->cb) != sizeof(next->cb));
1350 +
1351 +       memcpy(next->cb, skb->cb, sizeof(skb->cb));
1352 +       next->cb_next = skb->cb_next;
1353 +
1354 +       atomic_set(&next->refcnt, 1);
1355 +
1356 +       skb->cb_next = next;
1357 +       return 0;
1358 +}
1359 +EXPORT_SYMBOL(skb_save_cb);
1360 +
1361 +int skb_restore_cb(struct sk_buff *skb)
1362 +{
1363 +       struct skb_cb_table *next;
1364 +
1365 +       if (!skb->cb_next)
1366 +               return 0;
1367 +
1368 +       next = skb->cb_next;
1369 +
1370 +       BUILD_BUG_ON(sizeof(skb->cb) != sizeof(next->cb));
1371 +
1372 +       memcpy(skb->cb, next->cb, sizeof(skb->cb));
1373 +       skb->cb_next = next->cb_next;
1374 +
1375 +       spin_lock(&skb_cb_store_lock);
1376 +
1377 +       if (atomic_dec_and_test(&next->refcnt))
1378 +               kmem_cache_free(skbuff_cb_store_cache, next);
1379 +
1380 +       spin_unlock(&skb_cb_store_lock);
1381 +
1382 +       return 0;
1383 +}
1384 +EXPORT_SYMBOL(skb_restore_cb);
1385 +
1386 +static void skb_copy_stored_cb(struct sk_buff *   , const struct sk_buff *     ) __attribute__ ((unused));
1387 +static void skb_copy_stored_cb(struct sk_buff *new, const struct sk_buff *__old)
1388 +{
1389 +       struct skb_cb_table *next;
1390 +       struct sk_buff *old;
1391 +
1392 +       if (!__old->cb_next) {
1393 +               new->cb_next = NULL;
1394 +               return;
1395 +       }
1396 +
1397 +       spin_lock(&skb_cb_store_lock);
1398 +
1399 +       old = (struct sk_buff *)__old;
1400 +
1401 +       next = old->cb_next;
1402 +       atomic_inc(&next->refcnt);
1403 +       new->cb_next = next;
1404 +
1405 +       spin_unlock(&skb_cb_store_lock);
1406 +}
1407 +#endif
1408  
1409  /**
1410   *     skb_panic - private function for out-of-line support
1411 @@ -643,6 +724,28 @@ static void skb_release_head_state(struct sk_buff *skb)
1412                 WARN_ON(in_irq());
1413                 skb->destructor(skb);
1414         }
1415 +#if defined(CONFIG_IMQ) || defined(CONFIG_IMQ_MODULE)
1416 +       /*
1417 +        * This should not happen. When it does, avoid memleak by restoring
1418 +        * the chain of cb-backups.
1419 +        */
1420 +       while (skb->cb_next != NULL) {
1421 +               if (net_ratelimit())
1422 +                       pr_warn("IMQ: kfree_skb: skb->cb_next: %08x\n",
1423 +                               (unsigned int)(uintptr_t)skb->cb_next);
1424 +
1425 +               skb_restore_cb(skb);
1426 +       }
1427 +       /*
1428 +        * This should not happen either, nf_queue_entry is nullified in
1429 +        * imq_dev_xmit(). If we have non-NULL nf_queue_entry then we are
1430 +        * leaking entry pointers, maybe memory. We don't know if this is
1431 +        * pointer to already freed memory, or should this be freed.
1432 +        * If this happens we need to add refcounting, etc for nf_queue_entry.
1433 +        */
1434 +       if (skb->nf_queue_entry && net_ratelimit())
1435 +               pr_warn("%s\n", "IMQ: kfree_skb: skb->nf_queue_entry != NULL");
1436 +#endif
1437  #if IS_ENABLED(CONFIG_NF_CONNTRACK)
1438         nf_conntrack_put(skb->nfct);
1439  #endif
1440 @@ -765,6 +868,10 @@ static void __copy_skb_header(struct sk_buff *new, const struct sk_buff *old)
1441         new->sp                 = secpath_get(old->sp);
1442  #endif
1443         __nf_copy(new, old, false);
1444 +#if defined(CONFIG_IMQ) || defined(CONFIG_IMQ_MODULE)
1445 +       new->cb_next = NULL;
1446 +       /*skb_copy_stored_cb(new, old);*/
1447 +#endif
1448  
1449         /* Note : this field could be in headers_start/headers_end section
1450          * It is not yet because we do not want to have a 16 bit hole
1451 @@ -3324,6 +3431,13 @@ void __init skb_init(void)
1452                                                 0,
1453                                                 SLAB_HWCACHE_ALIGN|SLAB_PANIC,
1454                                                 NULL);
1455 +#if defined(CONFIG_IMQ) || defined(CONFIG_IMQ_MODULE)
1456 +       skbuff_cb_store_cache = kmem_cache_create("skbuff_cb_store_cache",
1457 +                                                 sizeof(struct skb_cb_table),
1458 +                                                 0,
1459 +                                                 SLAB_HWCACHE_ALIGN|SLAB_PANIC,
1460 +                                                 NULL);
1461 +#endif
1462  }
1463  
1464  /**
1465 diff --git a/net/ipv6/ip6_output.c b/net/ipv6/ip6_output.c
1466 index d5f7716..dd12857 100644
1467 --- a/net/ipv6/ip6_output.c
1468 +++ b/net/ipv6/ip6_output.c
1469 @@ -64,9 +64,6 @@ static int ip6_finish_output2(struct sock *sk, struct sk_buff *skb)
1470         struct in6_addr *nexthop;
1471         int ret;
1472  
1473 -       skb->protocol = htons(ETH_P_IPV6);
1474 -       skb->dev = dev;
1475 -
1476         if (ipv6_addr_is_multicast(&ipv6_hdr(skb)->daddr)) {
1477                 struct inet6_dev *idev = ip6_dst_idev(skb_dst(skb));
1478  
1479 @@ -143,6 +140,13 @@ int ip6_output(struct sock *sk, struct sk_buff *skb)
1480                 return 0;
1481         }
1482  
1483 +       /*
1484 +       * IMQ-patch: moved setting skb->dev and skb->protocol from
1485 +       * ip6_finish_output2 to fix crashing at netif_skb_features().
1486 +       */
1487 +       skb->protocol = htons(ETH_P_IPV6);
1488 +       skb->dev = dev;
1489 +
1490         return NF_HOOK_COND(NFPROTO_IPV6, NF_INET_POST_ROUTING, sk, skb,
1491                             NULL, dev,
1492                             ip6_finish_output,
1493 diff --git a/net/netfilter/Kconfig b/net/netfilter/Kconfig
1494 index 6eae69a..ca3b763 100644
1495 --- a/net/netfilter/Kconfig
1496 +++ b/net/netfilter/Kconfig
1497 @@ -784,6 +784,18 @@ config NETFILTER_XT_TARGET_LOG
1498  
1499           To compile it as a module, choose M here.  If unsure, say N.
1500  
1501 +config NETFILTER_XT_TARGET_IMQ
1502 +        tristate '"IMQ" target support'
1503 +       depends on NETFILTER_XTABLES
1504 +       depends on IP_NF_MANGLE || IP6_NF_MANGLE
1505 +       select IMQ
1506 +       default m if NETFILTER_ADVANCED=n
1507 +        help
1508 +          This option adds a `IMQ' target which is used to specify if and
1509 +          to which imq device packets should get enqueued/dequeued.
1510 +
1511 +          To compile it as a module, choose M here.  If unsure, say N.
1512 +
1513  config NETFILTER_XT_TARGET_MARK
1514         tristate '"MARK" target support'
1515         depends on NETFILTER_ADVANCED
1516 diff --git a/net/netfilter/Makefile b/net/netfilter/Makefile
1517 index 70d026d..5469b14 100644
1518 --- a/net/netfilter/Makefile
1519 +++ b/net/netfilter/Makefile
1520 @@ -110,6 +110,7 @@ obj-$(CONFIG_NETFILTER_XT_TARGET_CT) += xt_CT.o
1521  obj-$(CONFIG_NETFILTER_XT_TARGET_DSCP) += xt_DSCP.o
1522  obj-$(CONFIG_NETFILTER_XT_TARGET_HL) += xt_HL.o
1523  obj-$(CONFIG_NETFILTER_XT_TARGET_HMARK) += xt_HMARK.o
1524 +obj-$(CONFIG_NETFILTER_XT_TARGET_IMQ) += xt_IMQ.o
1525  obj-$(CONFIG_NETFILTER_XT_TARGET_LED) += xt_LED.o
1526  obj-$(CONFIG_NETFILTER_XT_TARGET_LOG) += xt_LOG.o
1527  obj-$(CONFIG_NETFILTER_XT_TARGET_NETMAP) += xt_NETMAP.o
1528 diff --git a/net/netfilter/core.c b/net/netfilter/core.c
1529 index a0e5497..a24276c 100644
1530 --- a/net/netfilter/core.c
1531 +++ b/net/netfilter/core.c
1532 @@ -206,9 +206,11 @@ next_hook:
1533                 ret = NF_DROP_GETERR(verdict);
1534                 if (ret == 0)
1535                         ret = -EPERM;
1536 -       } else if ((verdict & NF_VERDICT_MASK) == NF_QUEUE) {
1537 +       } else if ((verdict & NF_VERDICT_MASK) == NF_QUEUE ||
1538 +               (verdict & NF_VERDICT_MASK) == NF_IMQ_QUEUE) {
1539                 int err = nf_queue(skb, elem, state,
1540 -                                  verdict >> NF_VERDICT_QBITS);
1541 +                                  verdict >> NF_VERDICT_QBITS,
1542 +                                 verdict & NF_VERDICT_MASK);
1543                 if (err < 0) {
1544                         if (err == -ECANCELED)
1545                                 goto next_hook;
1546 diff --git a/net/netfilter/nf_internals.h b/net/netfilter/nf_internals.h
1547 index 3992106..35cbc7b 100644
1548 --- a/net/netfilter/nf_internals.h
1549 +++ b/net/netfilter/nf_internals.h
1550 @@ -18,7 +18,7 @@ unsigned int nf_iterate(struct list_head *head, struct sk_buff *skb,
1551  
1552  /* nf_queue.c */
1553  int nf_queue(struct sk_buff *skb, struct nf_hook_ops *elem,
1554 -            struct nf_hook_state *state, unsigned int queuenum);
1555 +            struct nf_hook_state *state, unsigned int queuenum, unsigned int queuetype);
1556  void nf_queue_nf_hook_drop(struct nf_hook_ops *ops);
1557  int __init netfilter_queue_init(void);
1558  
1559 diff --git a/net/netfilter/nf_queue.c b/net/netfilter/nf_queue.c
1560 index 8a8b2ab..91ba768 100644
1561 --- a/net/netfilter/nf_queue.c
1562 +++ b/net/netfilter/nf_queue.c
1563 @@ -28,6 +28,23 @@
1564   */
1565  static const struct nf_queue_handler __rcu *queue_handler __read_mostly;
1566  
1567 +#if defined(CONFIG_IMQ) || defined(CONFIG_IMQ_MODULE)
1568 +static const struct nf_queue_handler __rcu *queue_imq_handler __read_mostly;
1569 +
1570 +void nf_register_queue_imq_handler(const struct nf_queue_handler *qh)
1571 +{
1572 +       rcu_assign_pointer(queue_imq_handler, qh);
1573 +}
1574 +EXPORT_SYMBOL_GPL(nf_register_queue_imq_handler);
1575 +
1576 +void nf_unregister_queue_imq_handler(void)
1577 +{
1578 +       RCU_INIT_POINTER(queue_imq_handler, NULL);
1579 +       synchronize_rcu();
1580 +}
1581 +EXPORT_SYMBOL_GPL(nf_unregister_queue_imq_handler);
1582 +#endif
1583 +
1584  /* return EBUSY when somebody else is registered, return EEXIST if the
1585   * same handler is registered, return 0 in case of success. */
1586  void nf_register_queue_handler(const struct nf_queue_handler *qh)
1587 @@ -129,7 +146,8 @@ void nf_queue_nf_hook_drop(struct nf_hook_ops *ops)
1588  int nf_queue(struct sk_buff *skb,
1589              struct nf_hook_ops *elem,
1590              struct nf_hook_state *state,
1591 -            unsigned int queuenum)
1592 +            unsigned int queuenum,
1593 +                unsigned int queuetype)
1594  {
1595         int status = -ENOENT;
1596         struct nf_queue_entry *entry = NULL;
1597 @@ -139,7 +157,17 @@ int nf_queue(struct sk_buff *skb,
1598         /* QUEUE == DROP if no one is waiting, to be safe. */
1599         rcu_read_lock();
1600  
1601 -       qh = rcu_dereference(queue_handler);
1602 +       if (queuetype == NF_IMQ_QUEUE) {
1603 +#if defined(CONFIG_IMQ) || defined(CONFIG_IMQ_MODULE)
1604 +               qh = rcu_dereference(queue_imq_handler);
1605 +#else
1606 +               BUG();
1607 +               goto err_unlock;
1608 +#endif
1609 +       } else {
1610 +               qh = rcu_dereference(queue_handler);
1611 +       }
1612 +
1613         if (!qh) {
1614                 status = -ESRCH;
1615                 goto err_unlock;
1616 @@ -225,8 +253,10 @@ void nf_reinject(struct nf_queue_entry *entry, unsigned int verdict)
1617                 local_bh_enable();
1618                 break;
1619         case NF_QUEUE:
1620 +       case NF_IMQ_QUEUE:
1621                 err = nf_queue(skb, elem, &entry->state,
1622 -                              verdict >> NF_VERDICT_QBITS);
1623 +                              verdict >> NF_VERDICT_QBITS,
1624 +                                  verdict & NF_VERDICT_MASK);
1625                 if (err < 0) {
1626                         if (err == -ECANCELED)
1627                                 goto next_hook;
1628 diff --git a/net/netfilter/xt_IMQ.c b/net/netfilter/xt_IMQ.c
1629 new file mode 100644
1630 index 0000000..86d7b84
1631 --- /dev/null
1632 +++ b/net/netfilter/xt_IMQ.c
1633 @@ -0,0 +1,72 @@
1634 +/*
1635 + * This target marks packets to be enqueued to an imq device
1636 + */
1637 +#include <linux/module.h>
1638 +#include <linux/skbuff.h>
1639 +#include <linux/netfilter/x_tables.h>
1640 +#include <linux/netfilter/xt_IMQ.h>
1641 +#include <linux/imq.h>
1642 +
1643 +static unsigned int imq_target(struct sk_buff *pskb,
1644 +                               const struct xt_action_param *par)
1645 +{
1646 +       const struct xt_imq_info *mr = par->targinfo;
1647 +
1648 +       pskb->imq_flags = (mr->todev & IMQ_F_IFMASK) | IMQ_F_ENQUEUE;
1649 +
1650 +       return XT_CONTINUE;
1651 +}
1652 +
1653 +static int imq_checkentry(const struct xt_tgchk_param *par)
1654 +{
1655 +       struct xt_imq_info *mr = par->targinfo;
1656 +
1657 +       if (mr->todev > IMQ_MAX_DEVS - 1) {
1658 +               pr_warn("IMQ: invalid device specified, highest is %u\n",
1659 +                       IMQ_MAX_DEVS - 1);
1660 +               return -EINVAL;
1661 +       }
1662 +
1663 +       return 0;
1664 +}
1665 +
1666 +static struct xt_target xt_imq_reg[] __read_mostly = {
1667 +       {
1668 +               .name           = "IMQ",
1669 +               .family         = AF_INET,
1670 +               .checkentry     = imq_checkentry,
1671 +               .target         = imq_target,
1672 +               .targetsize     = sizeof(struct xt_imq_info),
1673 +               .table          = "mangle",
1674 +               .me             = THIS_MODULE
1675 +       },
1676 +       {
1677 +               .name           = "IMQ",
1678 +               .family         = AF_INET6,
1679 +               .checkentry     = imq_checkentry,
1680 +               .target         = imq_target,
1681 +               .targetsize     = sizeof(struct xt_imq_info),
1682 +               .table          = "mangle",
1683 +               .me             = THIS_MODULE
1684 +       },
1685 +};
1686 +
1687 +static int __init imq_init(void)
1688 +{
1689 +       return xt_register_targets(xt_imq_reg, ARRAY_SIZE(xt_imq_reg));
1690 +}
1691 +
1692 +static void __exit imq_fini(void)
1693 +{
1694 +       xt_unregister_targets(xt_imq_reg, ARRAY_SIZE(xt_imq_reg));
1695 +}
1696 +
1697 +module_init(imq_init);
1698 +module_exit(imq_fini);
1699 +
1700 +MODULE_AUTHOR("http://https://github.com/imq/linuximq");
1701 +MODULE_DESCRIPTION("Pseudo-driver for the intermediate queue device. See https://github.com/imq/linuximq/wiki for more information.");
1702 +MODULE_LICENSE("GPL");
1703 +MODULE_ALIAS("ipt_IMQ");
1704 +MODULE_ALIAS("ip6t_IMQ");
1705 +
1706 diff --git a/net/sched/sch_generic.c b/net/sched/sch_generic.c
1707 index 6efca30..a4e448f 100644
1708 --- a/net/sched/sch_generic.c
1709 +++ b/net/sched/sch_generic.c
1710 @@ -108,6 +108,14 @@ static struct sk_buff *dequeue_skb(struct Qdisc *q, bool *validate,
1711         return skb;
1712  }
1713  
1714 +struct sk_buff *qdisc_dequeue_skb(struct Qdisc *q, bool *validate)
1715 +{
1716 +       int packets;
1717 +
1718 +       return dequeue_skb(q, validate, &packets);
1719 +}
1720 +EXPORT_SYMBOL(qdisc_dequeue_skb);
1721 +
1722  static inline int handle_dev_cpu_collision(struct sk_buff *skb,
1723                                            struct netdev_queue *dev_queue,
1724                                            struct Qdisc *q)
This page took 0.284781 seconds and 4 git commands to generate.