]> git.pld-linux.org Git - packages/kernel.git/blob - kernel-imq.patch
- 4.4.89
[packages/kernel.git] / kernel-imq.patch
1 diff --git a/drivers/net/Kconfig b/drivers/net/Kconfig
2 index f184fb5..0e08522 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 900b0c5..e093402 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..f80258f
146 --- /dev/null
147 +++ b/drivers/net/imq.c
148 @@ -0,0 +1,903 @@
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 +               .pf             = PF_INET,
198 +               .hooknum        = NF_INET_PRE_ROUTING,
199 +#if defined(CONFIG_IMQ_BEHAVIOR_BA) || defined(CONFIG_IMQ_BEHAVIOR_BB)
200 +               .priority       = NF_IP_PRI_MANGLE + 1,
201 +#else
202 +               .priority       = NF_IP_PRI_NAT_DST + 1,
203 +#endif
204 +       },
205 +       {
206 +       /* imq_egress_ipv4 */
207 +               .hook           = imq_nf_hook,
208 +               .pf             = PF_INET,
209 +               .hooknum        = NF_INET_POST_ROUTING,
210 +#if defined(CONFIG_IMQ_BEHAVIOR_AA) || defined(CONFIG_IMQ_BEHAVIOR_BA)
211 +               .priority       = NF_IP_PRI_LAST,
212 +#else
213 +               .priority       = NF_IP_PRI_NAT_SRC - 1,
214 +#endif
215 +       },
216 +#if defined(CONFIG_IPV6) || defined(CONFIG_IPV6_MODULE)
217 +       {
218 +       /* imq_ingress_ipv6 */
219 +               .hook           = imq_nf_hook,
220 +               .pf             = PF_INET6,
221 +               .hooknum        = NF_INET_PRE_ROUTING,
222 +#if defined(CONFIG_IMQ_BEHAVIOR_BA) || defined(CONFIG_IMQ_BEHAVIOR_BB)
223 +               .priority       = NF_IP6_PRI_MANGLE + 1,
224 +#else
225 +               .priority       = NF_IP6_PRI_NAT_DST + 1,
226 +#endif
227 +       },
228 +       {
229 +       /* imq_egress_ipv6 */
230 +               .hook           = imq_nf_hook,
231 +               .pf             = PF_INET6,
232 +               .hooknum        = NF_INET_POST_ROUTING,
233 +#if defined(CONFIG_IMQ_BEHAVIOR_AA) || defined(CONFIG_IMQ_BEHAVIOR_BA)
234 +               .priority       = NF_IP6_PRI_LAST,
235 +#else
236 +               .priority       = NF_IP6_PRI_NAT_SRC - 1,
237 +#endif
238 +       },
239 +#endif
240 +};
241 +
242 +#if defined(CONFIG_IMQ_NUM_DEVS)
243 +static int numdevs = CONFIG_IMQ_NUM_DEVS;
244 +#else
245 +static int numdevs = IMQ_MAX_DEVS;
246 +#endif
247 +
248 +static struct net_device *imq_devs_cache[IMQ_MAX_DEVS];
249 +
250 +#define IMQ_MAX_QUEUES 32
251 +static int numqueues = 1;
252 +static u32 imq_hashrnd;
253 +static int imq_dev_accurate_stats = 1;
254 +
255 +static inline __be16 pppoe_proto(const struct sk_buff *skb)
256 +{
257 +       return *((__be16 *)(skb_mac_header(skb) + ETH_HLEN +
258 +                       sizeof(struct pppoe_hdr)));
259 +}
260 +
261 +static u16 imq_hash(struct net_device *dev, struct sk_buff *skb)
262 +{
263 +       unsigned int pull_len;
264 +       u16 protocol = skb->protocol;
265 +       u32 addr1, addr2;
266 +       u32 hash, ihl = 0;
267 +       union {
268 +               u16 in16[2];
269 +               u32 in32;
270 +       } ports;
271 +       u8 ip_proto;
272 +
273 +       pull_len = 0;
274 +
275 +recheck:
276 +       switch (protocol) {
277 +       case htons(ETH_P_8021Q): {
278 +               if (unlikely(skb_pull(skb, VLAN_HLEN) == NULL))
279 +                       goto other;
280 +
281 +               pull_len += VLAN_HLEN;
282 +               skb->network_header += VLAN_HLEN;
283 +
284 +               protocol = vlan_eth_hdr(skb)->h_vlan_encapsulated_proto;
285 +               goto recheck;
286 +       }
287 +
288 +       case htons(ETH_P_PPP_SES): {
289 +               if (unlikely(skb_pull(skb, PPPOE_SES_HLEN) == NULL))
290 +                       goto other;
291 +
292 +               pull_len += PPPOE_SES_HLEN;
293 +               skb->network_header += PPPOE_SES_HLEN;
294 +
295 +               protocol = pppoe_proto(skb);
296 +               goto recheck;
297 +       }
298 +
299 +       case htons(ETH_P_IP): {
300 +               const struct iphdr *iph = ip_hdr(skb);
301 +
302 +               if (unlikely(!pskb_may_pull(skb, sizeof(struct iphdr))))
303 +                       goto other;
304 +
305 +               addr1 = iph->daddr;
306 +               addr2 = iph->saddr;
307 +
308 +               ip_proto = !(ip_hdr(skb)->frag_off & htons(IP_MF | IP_OFFSET)) ?
309 +                                iph->protocol : 0;
310 +               ihl = ip_hdrlen(skb);
311 +
312 +               break;
313 +       }
314 +#if defined(CONFIG_IPV6) || defined(CONFIG_IPV6_MODULE)
315 +       case htons(ETH_P_IPV6): {
316 +               const struct ipv6hdr *iph = ipv6_hdr(skb);
317 +               __be16 fo = 0;
318 +
319 +               if (unlikely(!pskb_may_pull(skb, sizeof(struct ipv6hdr))))
320 +                       goto other;
321 +
322 +               addr1 = iph->daddr.s6_addr32[3];
323 +               addr2 = iph->saddr.s6_addr32[3];
324 +               ihl = ipv6_skip_exthdr(skb, sizeof(struct ipv6hdr), &ip_proto,
325 +                                      &fo);
326 +               if (unlikely(ihl < 0))
327 +                       goto other;
328 +
329 +               break;
330 +       }
331 +#endif
332 +       default:
333 +other:
334 +               if (pull_len != 0) {
335 +                       skb_push(skb, pull_len);
336 +                       skb->network_header -= pull_len;
337 +               }
338 +
339 +               return (u16)(ntohs(protocol) % dev->real_num_tx_queues);
340 +       }
341 +
342 +       if (addr1 > addr2)
343 +               swap(addr1, addr2);
344 +
345 +       switch (ip_proto) {
346 +       case IPPROTO_TCP:
347 +       case IPPROTO_UDP:
348 +       case IPPROTO_DCCP:
349 +       case IPPROTO_ESP:
350 +       case IPPROTO_AH:
351 +       case IPPROTO_SCTP:
352 +       case IPPROTO_UDPLITE: {
353 +               if (likely(skb_copy_bits(skb, ihl, &ports.in32, 4) >= 0)) {
354 +                       if (ports.in16[0] > ports.in16[1])
355 +                               swap(ports.in16[0], ports.in16[1]);
356 +                       break;
357 +               }
358 +               /* fall-through */
359 +       }
360 +       default:
361 +               ports.in32 = 0;
362 +               break;
363 +       }
364 +
365 +       if (pull_len != 0) {
366 +               skb_push(skb, pull_len);
367 +               skb->network_header -= pull_len;
368 +       }
369 +
370 +       hash = jhash_3words(addr1, addr2, ports.in32, imq_hashrnd ^ ip_proto);
371 +
372 +       return (u16)(((u64)hash * dev->real_num_tx_queues) >> 32);
373 +}
374 +
375 +static inline bool sk_tx_queue_recorded(struct sock *sk)
376 +{
377 +       return (sk_tx_queue_get(sk) >= 0);
378 +}
379 +
380 +static struct netdev_queue *imq_select_queue(struct net_device *dev,
381 +                                               struct sk_buff *skb)
382 +{
383 +       u16 queue_index = 0;
384 +       u32 hash;
385 +
386 +       if (likely(dev->real_num_tx_queues == 1))
387 +               goto out;
388 +
389 +       /* IMQ can be receiving ingress or engress packets. */
390 +
391 +       /* Check first for if rx_queue is set */
392 +       if (skb_rx_queue_recorded(skb)) {
393 +               queue_index = skb_get_rx_queue(skb);
394 +               goto out;
395 +       }
396 +
397 +       /* Check if socket has tx_queue set */
398 +       if (sk_tx_queue_recorded(skb->sk)) {
399 +               queue_index = sk_tx_queue_get(skb->sk);
400 +               goto out;
401 +       }
402 +
403 +       /* Try use socket hash */
404 +       if (skb->sk && skb->sk->sk_hash) {
405 +               hash = skb->sk->sk_hash;
406 +               queue_index =
407 +                       (u16)(((u64)hash * dev->real_num_tx_queues) >> 32);
408 +               goto out;
409 +       }
410 +
411 +       /* Generate hash from packet data */
412 +       queue_index = imq_hash(dev, skb);
413 +
414 +out:
415 +       if (unlikely(queue_index >= dev->real_num_tx_queues))
416 +               queue_index = (u16)((u32)queue_index % dev->real_num_tx_queues);
417 +
418 +       skb_set_queue_mapping(skb, queue_index);
419 +       return netdev_get_tx_queue(dev, queue_index);
420 +}
421 +
422 +static struct net_device_stats *imq_get_stats(struct net_device *dev)
423 +{
424 +       return &dev->stats;
425 +}
426 +
427 +/* called for packets kfree'd in qdiscs at places other than enqueue */
428 +static void imq_skb_destructor(struct sk_buff *skb)
429 +{
430 +       struct nf_queue_entry *entry = skb->nf_queue_entry;
431 +
432 +       skb->nf_queue_entry = NULL;
433 +
434 +       if (entry) {
435 +               nf_queue_entry_release_refs(entry);
436 +               kfree(entry);
437 +       }
438 +
439 +       skb_restore_cb(skb); /* kfree backup */
440 +}
441 +
442 +static void imq_done_check_queue_mapping(struct sk_buff *skb,
443 +                                        struct net_device *dev)
444 +{
445 +       unsigned int queue_index;
446 +
447 +       /* Don't let queue_mapping be left too large after exiting IMQ */
448 +       if (likely(skb->dev != dev && skb->dev != NULL)) {
449 +               queue_index = skb_get_queue_mapping(skb);
450 +               if (unlikely(queue_index >= skb->dev->real_num_tx_queues)) {
451 +                       queue_index = (u16)((u32)queue_index %
452 +                                               skb->dev->real_num_tx_queues);
453 +                       skb_set_queue_mapping(skb, queue_index);
454 +               }
455 +       } else {
456 +               /* skb->dev was IMQ device itself or NULL, be on safe side and
457 +                * just clear queue mapping.
458 +                */
459 +               skb_set_queue_mapping(skb, 0);
460 +       }
461 +}
462 +
463 +static netdev_tx_t imq_dev_xmit(struct sk_buff *skb, struct net_device *dev)
464 +{
465 +       struct nf_queue_entry *entry = skb->nf_queue_entry;
466 +
467 +       skb->nf_queue_entry = NULL;
468 +       dev->trans_start = jiffies;
469 +
470 +       dev->stats.tx_bytes += skb->len;
471 +       dev->stats.tx_packets++;
472 +
473 +       if (unlikely(entry == NULL)) {
474 +               /* We don't know what is going on here.. packet is queued for
475 +                * imq device, but (probably) not by us.
476 +                *
477 +                * If this packet was not send here by imq_nf_queue(), then
478 +                * skb_save_cb() was not used and skb_free() should not show:
479 +                *   WARNING: IMQ: kfree_skb: skb->cb_next:..
480 +                * and/or
481 +                *   WARNING: IMQ: kfree_skb: skb->nf_queue_entry...
482 +                *
483 +                * However if this message is shown, then IMQ is somehow broken
484 +                * and you should report this to linuximq.net.
485 +                */
486 +
487 +               /* imq_dev_xmit is black hole that eats all packets, report that
488 +                * we eat this packet happily and increase dropped counters.
489 +                */
490 +
491 +               dev->stats.tx_dropped++;
492 +               dev_kfree_skb(skb);
493 +
494 +               return NETDEV_TX_OK;
495 +       }
496 +
497 +       skb_restore_cb(skb); /* restore skb->cb */
498 +
499 +       skb->imq_flags = 0;
500 +       skb->destructor = NULL;
501 +
502 +       imq_done_check_queue_mapping(skb, dev);
503 +
504 +       nf_reinject(entry, NF_ACCEPT);
505 +
506 +       return NETDEV_TX_OK;
507 +}
508 +
509 +static struct net_device *get_imq_device_by_index(int index)
510 +{
511 +       struct net_device *dev = NULL;
512 +       struct net *net;
513 +       char buf[8];
514 +
515 +       /* get device by name and cache result */
516 +       snprintf(buf, sizeof(buf), "imq%d", index);
517 +
518 +       /* Search device from all namespaces. */
519 +       for_each_net(net) {
520 +               dev = dev_get_by_name(net, buf);
521 +               if (dev)
522 +                       break;
523 +       }
524 +
525 +       if (WARN_ON_ONCE(dev == NULL)) {
526 +               /* IMQ device not found. Exotic config? */
527 +               return ERR_PTR(-ENODEV);
528 +       }
529 +
530 +       imq_devs_cache[index] = dev;
531 +       dev_put(dev);
532 +
533 +       return dev;
534 +}
535 +
536 +static struct nf_queue_entry *nf_queue_entry_dup(struct nf_queue_entry *e)
537 +{
538 +       struct nf_queue_entry *entry = kmemdup(e, e->size, GFP_ATOMIC);
539 +       if (entry) {
540 +               nf_queue_entry_get_refs(entry);
541 +                       return entry;
542 +       }
543 +       return NULL;
544 +}
545 +
546 +#ifdef CONFIG_BRIDGE_NETFILTER
547 +/* When called from bridge netfilter, skb->data must point to MAC header
548 + * before calling skb_gso_segment(). Else, original MAC header is lost
549 + * and segmented skbs will be sent to wrong destination.
550 + */
551 +static void nf_bridge_adjust_skb_data(struct sk_buff *skb)
552 +{
553 +       if (skb->nf_bridge)
554 +               __skb_push(skb, skb->network_header - skb->mac_header);
555 +}
556 +
557 +static void nf_bridge_adjust_segmented_data(struct sk_buff *skb)
558 +{
559 +       if (skb->nf_bridge)
560 +               __skb_pull(skb, skb->network_header - skb->mac_header);
561 +}
562 +#else
563 +#define nf_bridge_adjust_skb_data(s) do {} while (0)
564 +#define nf_bridge_adjust_segmented_data(s) do {} while (0)
565 +#endif
566 +
567 +static void free_entry(struct nf_queue_entry *entry)
568 +{
569 +       nf_queue_entry_release_refs(entry);
570 +       kfree(entry);
571 +}
572 +
573 +static int __imq_nf_queue(struct nf_queue_entry *entry, struct net_device *dev);
574 +
575 +static int __imq_nf_queue_gso(struct nf_queue_entry *entry,
576 +                             struct net_device *dev, struct sk_buff *skb)
577 +{
578 +       int ret = -ENOMEM;
579 +       struct nf_queue_entry *entry_seg;
580 +
581 +       nf_bridge_adjust_segmented_data(skb);
582 +
583 +       if (skb->next == NULL) { /* last packet, no need to copy entry */
584 +               struct sk_buff *gso_skb = entry->skb;
585 +               entry->skb = skb;
586 +               ret = __imq_nf_queue(entry, dev);
587 +               if (ret)
588 +                       entry->skb = gso_skb;
589 +               return ret;
590 +       }
591 +
592 +       skb->next = NULL;
593 +
594 +       entry_seg = nf_queue_entry_dup(entry);
595 +       if (entry_seg) {
596 +               entry_seg->skb = skb;
597 +               ret = __imq_nf_queue(entry_seg, dev);
598 +               if (ret)
599 +                       free_entry(entry_seg);
600 +       }
601 +       return ret;
602 +}
603 +
604 +static int imq_nf_queue(struct nf_queue_entry *entry, unsigned queue_num)
605 +{
606 +       struct sk_buff *skb, *segs;
607 +       struct net_device *dev;
608 +       unsigned int queued;
609 +       int index, retval, err;
610 +
611 +       index = entry->skb->imq_flags & IMQ_F_IFMASK;
612 +       if (unlikely(index > numdevs - 1)) {
613 +               if (net_ratelimit())
614 +                       pr_warn("IMQ: invalid device specified, highest is %u\n",
615 +                               numdevs - 1);
616 +               retval = -EINVAL;
617 +               goto out_no_dev;
618 +       }
619 +
620 +       /* check for imq device by index from cache */
621 +       dev = imq_devs_cache[index];
622 +       if (unlikely(!dev)) {
623 +               dev = get_imq_device_by_index(index);
624 +               if (IS_ERR(dev)) {
625 +                       retval = PTR_ERR(dev);
626 +                       goto out_no_dev;
627 +               }
628 +       }
629 +
630 +       if (unlikely(!(dev->flags & IFF_UP))) {
631 +               entry->skb->imq_flags = 0;
632 +               retval = -ECANCELED;
633 +               goto out_no_dev;
634 +       }
635 +
636 +       /* Since 3.10.x, GSO handling moved here as result of upstream commit
637 +        * a5fedd43d5f6c94c71053a66e4c3d2e35f1731a2 (netfilter: move
638 +        * skb_gso_segment into nfnetlink_queue module).
639 +        *
640 +        * Following code replicates the gso handling from
641 +        * 'net/netfilter/nfnetlink_queue_core.c':nfqnl_enqueue_packet().
642 +        */
643 +
644 +       skb = entry->skb;
645 +
646 +       switch (entry->state.pf) {
647 +       case NFPROTO_IPV4:
648 +               skb->protocol = htons(ETH_P_IP);
649 +               break;
650 +       case NFPROTO_IPV6:
651 +               skb->protocol = htons(ETH_P_IPV6);
652 +               break;
653 +       }
654 +
655 +       if (!skb_is_gso(entry->skb))
656 +               return __imq_nf_queue(entry, dev);
657 +
658 +       nf_bridge_adjust_skb_data(skb);
659 +       segs = skb_gso_segment(skb, 0);
660 +       /* Does not use PTR_ERR to limit the number of error codes that can be
661 +        * returned by nf_queue.  For instance, callers rely on -ECANCELED to
662 +        * mean 'ignore this hook'.
663 +        */
664 +       err = -ENOBUFS;
665 +       if (IS_ERR(segs))
666 +               goto out_err;
667 +       queued = 0;
668 +       err = 0;
669 +       do {
670 +               struct sk_buff *nskb = segs->next;
671 +               if (nskb && nskb->next)
672 +                       nskb->cb_next = NULL;
673 +               if (err == 0)
674 +                       err = __imq_nf_queue_gso(entry, dev, segs);
675 +               if (err == 0)
676 +                       queued++;
677 +               else
678 +                       kfree_skb(segs);
679 +               segs = nskb;
680 +       } while (segs);
681 +
682 +       if (queued) {
683 +               if (err) /* some segments are already queued */
684 +                       free_entry(entry);
685 +               kfree_skb(skb);
686 +               return 0;
687 +       }
688 +
689 +out_err:
690 +       nf_bridge_adjust_segmented_data(skb);
691 +       retval = err;
692 +out_no_dev:
693 +       return retval;
694 +}
695 +
696 +static int __imq_nf_queue(struct nf_queue_entry *entry, struct net_device *dev)
697 +{
698 +       struct sk_buff *skb_orig, *skb, *skb_shared, *skb_popd;
699 +       struct Qdisc *q;
700 +       struct netdev_queue *txq;
701 +       spinlock_t *root_lock;
702 +       int users;
703 +       int retval = -EINVAL;
704 +       unsigned int orig_queue_index;
705 +
706 +       dev->last_rx = jiffies;
707 +
708 +       skb = entry->skb;
709 +       skb_orig = NULL;
710 +
711 +       /* skb has owner? => make clone */
712 +       if (unlikely(skb->destructor)) {
713 +               skb_orig = skb;
714 +               skb = skb_clone(skb, GFP_ATOMIC);
715 +               if (unlikely(!skb)) {
716 +                       retval = -ENOMEM;
717 +                       goto out;
718 +               }
719 +               skb->cb_next = NULL;
720 +               entry->skb = skb;
721 +       }
722 +
723 +       dev->stats.rx_bytes += skb->len;
724 +       dev->stats.rx_packets++;
725 +
726 +       if (!skb->dev) {
727 +               /* skb->dev == NULL causes problems, try the find cause. */
728 +               if (net_ratelimit()) {
729 +                       dev_warn(&dev->dev,
730 +                                "received packet with skb->dev == NULL\n");
731 +                       dump_stack();
732 +               }
733 +
734 +               skb->dev = dev;
735 +       }
736 +
737 +       /* Disables softirqs for lock below */
738 +       rcu_read_lock_bh();
739 +
740 +       /* Multi-queue selection */
741 +       orig_queue_index = skb_get_queue_mapping(skb);
742 +       txq = imq_select_queue(dev, skb);
743 +
744 +       q = rcu_dereference(txq->qdisc);
745 +       if (unlikely(!q->enqueue))
746 +               goto packet_not_eaten_by_imq_dev;
747 +
748 +       skb->nf_queue_entry = entry;
749 +       root_lock = qdisc_lock(q);
750 +       spin_lock(root_lock);
751 +
752 +       users = atomic_read(&skb->users);
753 +
754 +       skb_shared = skb_get(skb); /* increase reference count by one */
755 +
756 +       /* backup skb->cb, as qdisc layer will overwrite it */
757 +       skb_save_cb(skb_shared);
758 +       qdisc_enqueue_root(skb_shared, q); /* might kfree_skb */
759 +       if (likely(atomic_read(&skb_shared->users) == users + 1)) {
760 +               bool validate;
761 +
762 +               kfree_skb(skb_shared); /* decrease reference count by one */
763 +
764 +               skb->destructor = &imq_skb_destructor;
765 +
766 +               skb_popd = qdisc_dequeue_skb(q, &validate);
767 +
768 +               /* cloned? */
769 +               if (unlikely(skb_orig))
770 +                       kfree_skb(skb_orig); /* free original */
771 +
772 +               spin_unlock(root_lock);
773 +
774 +#if 0
775 +               /* schedule qdisc dequeue */
776 +               __netif_schedule(q);
777 +#else
778 +               if (likely(skb_popd)) {
779 +                       /* Note that we validate skb (GSO, checksum, ...) outside of locks */
780 +                       if (validate)
781 +                       skb_popd = validate_xmit_skb_list(skb_popd, dev);
782 +                       
783 +                       if (skb_popd) {
784 +                               int dummy_ret;
785 +                               int cpu = smp_processor_id(); /* ok because BHs are off */
786 +
787 +                               txq = skb_get_tx_queue(dev, skb_popd);
788 +                               /* 
789 +                               IMQ device will not be frozen or stoped, and it always be successful.
790 +                               So we need not check its status and return value to accelerate.
791 +                               */
792 +                               if (imq_dev_accurate_stats && txq->xmit_lock_owner != cpu) {
793 +                                       HARD_TX_LOCK(dev, txq, cpu);
794 +                                       if (!netif_xmit_frozen_or_stopped(txq)) {
795 +                                               dev_hard_start_xmit(skb_popd, dev, txq, &dummy_ret);
796 +                                       }
797 +                                       HARD_TX_UNLOCK(dev, txq);
798 +                               } else {
799 +                                       if (!netif_xmit_frozen_or_stopped(txq)) {
800 +                                               dev_hard_start_xmit(skb_popd, dev, txq, &dummy_ret);
801 +                                       }
802 +                               }
803 +                       }
804 +               } else {
805 +                       /* No ready skb, then schedule it */
806 +                       __netif_schedule(q);
807 +               }
808 +#endif
809 +               rcu_read_unlock_bh();
810 +               retval = 0;
811 +               goto out;
812 +       } else {
813 +               skb_restore_cb(skb_shared); /* restore skb->cb */
814 +               skb->nf_queue_entry = NULL;
815 +               /*
816 +                * qdisc dropped packet and decreased skb reference count of
817 +                * skb, so we don't really want to and try refree as that would
818 +                * actually destroy the skb.
819 +                */
820 +               spin_unlock(root_lock);
821 +               goto packet_not_eaten_by_imq_dev;
822 +       }
823 +
824 +packet_not_eaten_by_imq_dev:
825 +       skb_set_queue_mapping(skb, orig_queue_index);
826 +       rcu_read_unlock_bh();
827 +
828 +       /* cloned? restore original */
829 +       if (unlikely(skb_orig)) {
830 +               kfree_skb(skb);
831 +               entry->skb = skb_orig;
832 +       }
833 +       retval = -1;
834 +out:
835 +       return retval;
836 +}
837 +static unsigned int imq_nf_hook(void *priv,
838 +                               struct sk_buff *skb,
839 +                               const struct nf_hook_state *state)
840 +{
841 +       return (skb->imq_flags & IMQ_F_ENQUEUE) ? NF_IMQ_QUEUE : NF_ACCEPT;
842 +}
843 +
844 +static int imq_close(struct net_device *dev)
845 +{
846 +       netif_stop_queue(dev);
847 +       return 0;
848 +}
849 +
850 +static int imq_open(struct net_device *dev)
851 +{
852 +       netif_start_queue(dev);
853 +       return 0;
854 +}
855 +
856 +static const struct net_device_ops imq_netdev_ops = {
857 +       .ndo_open               = imq_open,
858 +       .ndo_stop               = imq_close,
859 +       .ndo_start_xmit         = imq_dev_xmit,
860 +       .ndo_get_stats          = imq_get_stats,
861 +};
862 +
863 +static void imq_setup(struct net_device *dev)
864 +{
865 +       dev->netdev_ops         = &imq_netdev_ops;
866 +       dev->type               = ARPHRD_VOID;
867 +       dev->mtu                = 16000; /* too small? */
868 +       dev->tx_queue_len       = 11000; /* too big? */
869 +       dev->flags              = IFF_NOARP;
870 +       dev->features           = NETIF_F_SG | NETIF_F_FRAGLIST |
871 +                                 NETIF_F_GSO | NETIF_F_HW_CSUM |
872 +                                 NETIF_F_HIGHDMA;
873 +       dev->priv_flags         &= ~(IFF_XMIT_DST_RELEASE |
874 +                                    IFF_TX_SKB_SHARING);
875 +}
876 +
877 +static int imq_validate(struct nlattr *tb[], struct nlattr *data[])
878 +{
879 +       int ret = 0;
880 +
881 +       if (tb[IFLA_ADDRESS]) {
882 +               if (nla_len(tb[IFLA_ADDRESS]) != ETH_ALEN) {
883 +                       ret = -EINVAL;
884 +                       goto end;
885 +               }
886 +               if (!is_valid_ether_addr(nla_data(tb[IFLA_ADDRESS]))) {
887 +                       ret = -EADDRNOTAVAIL;
888 +                       goto end;
889 +               }
890 +       }
891 +       return 0;
892 +end:
893 +       pr_warn("IMQ: imq_validate failed (%d)\n", ret);
894 +       return ret;
895 +}
896 +
897 +static struct rtnl_link_ops imq_link_ops __read_mostly = {
898 +       .kind           = "imq",
899 +       .priv_size      = 0,
900 +       .setup          = imq_setup,
901 +       .validate       = imq_validate,
902 +};
903 +
904 +static const struct nf_queue_handler imq_nfqh = {
905 +       .outfn = imq_nf_queue,
906 +};
907 +
908 +static int __init imq_init_hooks(void)
909 +{
910 +       int ret;
911 +
912 +       nf_register_queue_imq_handler(&imq_nfqh);
913 +
914 +       ret = nf_register_hooks(imq_ops, ARRAY_SIZE(imq_ops));
915 +       if (ret < 0)
916 +               nf_unregister_queue_imq_handler();
917 +
918 +       return ret;
919 +}
920 +
921 +static int __init imq_init_one(int index)
922 +{
923 +       struct net_device *dev;
924 +       int ret;
925 +
926 +       dev = alloc_netdev_mq(0, "imq%d", NET_NAME_UNKNOWN, imq_setup, numqueues);
927 +       if (!dev)
928 +               return -ENOMEM;
929 +
930 +       ret = dev_alloc_name(dev, dev->name);
931 +       if (ret < 0)
932 +               goto fail;
933 +
934 +       dev->rtnl_link_ops = &imq_link_ops;
935 +       ret = register_netdevice(dev);
936 +       if (ret < 0)
937 +               goto fail;
938 +
939 +       return 0;
940 +fail:
941 +       free_netdev(dev);
942 +       return ret;
943 +}
944 +
945 +static int __init imq_init_devs(void)
946 +{
947 +       int err, i;
948 +
949 +       if (numdevs < 1 || numdevs > IMQ_MAX_DEVS) {
950 +               pr_err("IMQ: numdevs has to be betweed 1 and %u\n",
951 +                      IMQ_MAX_DEVS);
952 +               return -EINVAL;
953 +       }
954 +
955 +       if (numqueues < 1 || numqueues > IMQ_MAX_QUEUES) {
956 +               pr_err("IMQ: numqueues has to be betweed 1 and %u\n",
957 +                      IMQ_MAX_QUEUES);
958 +               return -EINVAL;
959 +       }
960 +
961 +       get_random_bytes(&imq_hashrnd, sizeof(imq_hashrnd));
962 +
963 +       rtnl_lock();
964 +       err = __rtnl_link_register(&imq_link_ops);
965 +
966 +       for (i = 0; i < numdevs && !err; i++)
967 +               err = imq_init_one(i);
968 +
969 +       if (err) {
970 +               __rtnl_link_unregister(&imq_link_ops);
971 +               memset(imq_devs_cache, 0, sizeof(imq_devs_cache));
972 +       }
973 +       rtnl_unlock();
974 +
975 +       return err;
976 +}
977 +
978 +static int __init imq_init_module(void)
979 +{
980 +       int err;
981 +
982 +#if defined(CONFIG_IMQ_NUM_DEVS)
983 +       BUILD_BUG_ON(CONFIG_IMQ_NUM_DEVS > 16);
984 +       BUILD_BUG_ON(CONFIG_IMQ_NUM_DEVS < 2);
985 +       BUILD_BUG_ON(CONFIG_IMQ_NUM_DEVS - 1 > IMQ_F_IFMASK);
986 +#endif
987 +
988 +       err = imq_init_devs();
989 +       if (err) {
990 +               pr_err("IMQ: Error trying imq_init_devs(net)\n");
991 +               return err;
992 +       }
993 +
994 +       err = imq_init_hooks();
995 +       if (err) {
996 +               pr_err(KERN_ERR "IMQ: Error trying imq_init_hooks()\n");
997 +               rtnl_link_unregister(&imq_link_ops);
998 +               memset(imq_devs_cache, 0, sizeof(imq_devs_cache));
999 +               return err;
1000 +       }
1001 +
1002 +       pr_info("IMQ driver loaded successfully. (numdevs = %d, numqueues = %d, imq_dev_accurate_stats = %d)\n",
1003 +               numdevs, numqueues, imq_dev_accurate_stats);
1004 +
1005 +#if defined(CONFIG_IMQ_BEHAVIOR_BA) || defined(CONFIG_IMQ_BEHAVIOR_BB)
1006 +       pr_info("\tHooking IMQ before NAT on PREROUTING.\n");
1007 +#else
1008 +       pr_info("\tHooking IMQ after NAT on PREROUTING.\n");
1009 +#endif
1010 +#if defined(CONFIG_IMQ_BEHAVIOR_AB) || defined(CONFIG_IMQ_BEHAVIOR_BB)
1011 +       pr_info("\tHooking IMQ before NAT on POSTROUTING.\n");
1012 +#else
1013 +       pr_info("\tHooking IMQ after NAT on POSTROUTING.\n");
1014 +#endif
1015 +
1016 +       return 0;
1017 +}
1018 +
1019 +static void __exit imq_unhook(void)
1020 +{
1021 +       nf_unregister_hooks(imq_ops, ARRAY_SIZE(imq_ops));
1022 +       nf_unregister_queue_imq_handler();
1023 +}
1024 +
1025 +static void __exit imq_cleanup_devs(void)
1026 +{
1027 +       rtnl_link_unregister(&imq_link_ops);
1028 +       memset(imq_devs_cache, 0, sizeof(imq_devs_cache));
1029 +}
1030 +
1031 +static void __exit imq_exit_module(void)
1032 +{
1033 +       imq_unhook();
1034 +       imq_cleanup_devs();
1035 +       pr_info("IMQ driver unloaded successfully.\n");
1036 +}
1037 +
1038 +module_init(imq_init_module);
1039 +module_exit(imq_exit_module);
1040 +
1041 +module_param(numdevs, int, 0);
1042 +module_param(numqueues, int, 0);
1043 +module_param(imq_dev_accurate_stats, int, 0);
1044 +MODULE_PARM_DESC(numdevs, "number of IMQ devices (how many imq* devices will be created)");
1045 +MODULE_PARM_DESC(numqueues, "number of queues per IMQ device");
1046 +MODULE_PARM_DESC(imq_dev_accurate_stats, "Notify if need the accurate imq device stats");
1047 +
1048 +MODULE_AUTHOR("http://https://github.com/imq/linuximq");
1049 +MODULE_DESCRIPTION("Pseudo-driver for the intermediate queue device. See https://github.com/imq/linuximq/wiki for more information.");
1050 +MODULE_LICENSE("GPL");
1051 +MODULE_ALIAS_RTNL_LINK("imq");
1052 diff --git a/include/linux/imq.h b/include/linux/imq.h
1053 new file mode 100644
1054 index 0000000..1babb09
1055 --- /dev/null
1056 +++ b/include/linux/imq.h
1057 @@ -0,0 +1,13 @@
1058 +#ifndef _IMQ_H
1059 +#define _IMQ_H
1060 +
1061 +/* IFMASK (16 device indexes, 0 to 15) and flag(s) fit in 5 bits */
1062 +#define IMQ_F_BITS     5
1063 +
1064 +#define IMQ_F_IFMASK   0x0f
1065 +#define IMQ_F_ENQUEUE  0x10
1066 +
1067 +#define IMQ_MAX_DEVS   (IMQ_F_IFMASK + 1)
1068 +
1069 +#endif /* _IMQ_H */
1070 +
1071 diff --git a/include/linux/netdevice.h b/include/linux/netdevice.h
1072 index 3143c84..e213b31 100644
1073 --- a/include/linux/netdevice.h
1074 +++ b/include/linux/netdevice.h
1075 @@ -3341,6 +3341,19 @@ static inline void netif_tx_unlock_bh(struct net_device *dev)
1076         }                                               \
1077  }
1078  
1079 +#define HARD_TX_LOCK_BH(dev, txq) {           \
1080 +    if ((dev->features & NETIF_F_LLTX) == 0) {  \
1081 +        __netif_tx_lock_bh(txq);      \
1082 +    }                       \
1083 +}
1084 +
1085 +#define HARD_TX_UNLOCK_BH(dev, txq) {          \
1086 +    if ((dev->features & NETIF_F_LLTX) == 0) {  \
1087 +        __netif_tx_unlock_bh(txq);         \
1088 +    }                       \
1089 +}
1090 +
1091 +
1092  static inline void netif_tx_disable(struct net_device *dev)
1093  {
1094         unsigned int i;
1095 diff --git a/include/linux/netfilter/xt_IMQ.h b/include/linux/netfilter/xt_IMQ.h
1096 new file mode 100644
1097 index 0000000..9b07230
1098 --- /dev/null
1099 +++ b/include/linux/netfilter/xt_IMQ.h
1100 @@ -0,0 +1,9 @@
1101 +#ifndef _XT_IMQ_H
1102 +#define _XT_IMQ_H
1103 +
1104 +struct xt_imq_info {
1105 +       unsigned int todev;     /* target imq device */
1106 +};
1107 +
1108 +#endif /* _XT_IMQ_H */
1109 +
1110 diff --git a/include/linux/netfilter_ipv4/ipt_IMQ.h b/include/linux/netfilter_ipv4/ipt_IMQ.h
1111 new file mode 100644
1112 index 0000000..7af320f
1113 --- /dev/null
1114 +++ b/include/linux/netfilter_ipv4/ipt_IMQ.h
1115 @@ -0,0 +1,10 @@
1116 +#ifndef _IPT_IMQ_H
1117 +#define _IPT_IMQ_H
1118 +
1119 +/* Backwards compatibility for old userspace */
1120 +#include <linux/netfilter/xt_IMQ.h>
1121 +
1122 +#define ipt_imq_info xt_imq_info
1123 +
1124 +#endif /* _IPT_IMQ_H */
1125 +
1126 diff --git a/include/linux/netfilter_ipv6/ip6t_IMQ.h b/include/linux/netfilter_ipv6/ip6t_IMQ.h
1127 new file mode 100644
1128 index 0000000..198ac01
1129 --- /dev/null
1130 +++ b/include/linux/netfilter_ipv6/ip6t_IMQ.h
1131 @@ -0,0 +1,10 @@
1132 +#ifndef _IP6T_IMQ_H
1133 +#define _IP6T_IMQ_H
1134 +
1135 +/* Backwards compatibility for old userspace */
1136 +#include <linux/netfilter/xt_IMQ.h>
1137 +
1138 +#define ip6t_imq_info xt_imq_info
1139 +
1140 +#endif /* _IP6T_IMQ_H */
1141 +
1142 diff --git a/include/linux/skbuff.h b/include/linux/skbuff.h
1143 index 4355129..47914c0 100644
1144 --- a/include/linux/skbuff.h
1145 +++ b/include/linux/skbuff.h
1146 @@ -38,6 +38,10 @@
1147  #include <linux/splice.h>
1148  #include <linux/in6.h>
1149  #include <net/flow.h>
1150 +#if defined(CONFIG_IMQ) || defined(CONFIG_IMQ_MODULE)
1151 +#include <linux/imq.h>
1152 +#endif
1153 +
1154  
1155  /* A. Checksumming of received packets by device.
1156   *
1157 @@ -566,6 +570,9 @@ struct sk_buff {
1158          * first. This is owned by whoever has the skb queued ATM.
1159          */
1160         char                    cb[48] __aligned(8);
1161 +#if defined(CONFIG_IMQ) || defined(CONFIG_IMQ_MODULE)
1162 +       void                    *cb_next;
1163 +#endif
1164  
1165         unsigned long           _skb_refdst;
1166         void                    (*destructor)(struct sk_buff *skb);
1167 @@ -575,6 +582,9 @@ struct sk_buff {
1168  #if defined(CONFIG_NF_CONNTRACK) || defined(CONFIG_NF_CONNTRACK_MODULE)
1169         struct nf_conntrack     *nfct;
1170  #endif
1171 +#if defined(CONFIG_IMQ) || defined(CONFIG_IMQ_MODULE)
1172 +       struct nf_queue_entry   *nf_queue_entry;
1173 +#endif
1174  #if IS_ENABLED(CONFIG_BRIDGE_NETFILTER)
1175         struct nf_bridge_info   *nf_bridge;
1176  #endif
1177 @@ -642,6 +652,9 @@ struct sk_buff {
1178         __u8                    inner_protocol_type:1;
1179         __u8                    remcsum_offload:1;
1180         /* 3 or 5 bit hole */
1181 +       #if defined(CONFIG_IMQ) || defined(CONFIG_IMQ_MODULE)
1182 +       __u8                    imq_flags:IMQ_F_BITS;
1183 +       #endif
1184  
1185  #ifdef CONFIG_NET_SCHED
1186         __u16                   tc_index;       /* traffic control index */
1187 @@ -798,6 +811,12 @@ void kfree_skb_list(struct sk_buff *segs);
1188  void skb_tx_error(struct sk_buff *skb);
1189  void consume_skb(struct sk_buff *skb);
1190  void  __kfree_skb(struct sk_buff *skb);
1191 +
1192 +#if defined(CONFIG_IMQ) || defined(CONFIG_IMQ_MODULE)
1193 +int skb_save_cb(struct sk_buff *skb);
1194 +int skb_restore_cb(struct sk_buff *skb);
1195 +#endif
1196 +
1197  extern struct kmem_cache *skbuff_head_cache;
1198  
1199  void kfree_skb_partial(struct sk_buff *skb, bool head_stolen);
1200 @@ -3344,6 +3363,10 @@ static inline void __nf_copy(struct sk_buff *dst, const struct sk_buff *src,
1201         if (copy)
1202                 dst->nfctinfo = src->nfctinfo;
1203  #endif
1204 +#if defined(CONFIG_IMQ) || defined(CONFIG_IMQ_MODULE)
1205 +       dst->imq_flags = src->imq_flags;
1206 +       dst->nf_queue_entry = src->nf_queue_entry;
1207 +#endif
1208  #if IS_ENABLED(CONFIG_BRIDGE_NETFILTER)
1209         dst->nf_bridge  = src->nf_bridge;
1210         nf_bridge_get(src->nf_bridge);
1211 diff --git a/include/net/netfilter/nf_queue.h b/include/net/netfilter/nf_queue.h
1212 index 9c5638a..b173aa7 100644
1213 --- a/include/net/netfilter/nf_queue.h
1214 +++ b/include/net/netfilter/nf_queue.h
1215 @@ -31,6 +31,12 @@ struct nf_queue_handler {
1216  void nf_register_queue_handler(const struct nf_queue_handler *qh);
1217  void nf_unregister_queue_handler(void);
1218  void nf_reinject(struct nf_queue_entry *entry, unsigned int verdict);
1219 +void nf_queue_entry_release_refs(struct nf_queue_entry *entry);
1220 +
1221 +#if defined(CONFIG_IMQ) || defined(CONFIG_IMQ_MODULE)
1222 +void nf_register_queue_imq_handler(const struct nf_queue_handler *qh);
1223 +void nf_unregister_queue_imq_handler(void);
1224 +#endif
1225  
1226  void nf_queue_entry_get_refs(struct nf_queue_entry *entry);
1227  void nf_queue_entry_release_refs(struct nf_queue_entry *entry);
1228 diff --git a/include/net/pkt_sched.h b/include/net/pkt_sched.h
1229 index 401038d..4668849 100644
1230 --- a/include/net/pkt_sched.h
1231 +++ b/include/net/pkt_sched.h
1232 @@ -104,6 +104,8 @@ int sch_direct_xmit(struct sk_buff *skb, struct Qdisc *q,
1233  
1234  void __qdisc_run(struct Qdisc *q);
1235  
1236 +struct sk_buff *qdisc_dequeue_skb(struct Qdisc *q, bool *validate);
1237 +
1238  static inline void qdisc_run(struct Qdisc *q)
1239  {
1240         if (qdisc_run_begin(q))
1241 diff --git a/include/net/sch_generic.h b/include/net/sch_generic.h
1242 index b2a8e63..d9feaa3 100644
1243 --- a/include/net/sch_generic.h
1244 +++ b/include/net/sch_generic.h
1245 @@ -506,6 +506,12 @@ static inline int qdisc_enqueue(struct sk_buff *skb, struct Qdisc *sch)
1246         return sch->enqueue(skb, sch);
1247  }
1248  
1249 +static inline int qdisc_enqueue_root(struct sk_buff *skb, struct Qdisc *sch)
1250 +{
1251 +    qdisc_skb_cb(skb)->pkt_len = skb->len;
1252 +    return qdisc_enqueue(skb, sch) & NET_XMIT_MASK;
1253 +}
1254 +
1255  static inline bool qdisc_is_percpu_stats(const struct Qdisc *q)
1256  {
1257         return q->flags & TCQ_F_CPUSTATS;
1258 diff --git a/include/uapi/linux/netfilter.h b/include/uapi/linux/netfilter.h
1259 index d93f949..23fb6d1 100644
1260 --- a/include/uapi/linux/netfilter.h
1261 +++ b/include/uapi/linux/netfilter.h
1262 @@ -14,7 +14,8 @@
1263  #define NF_QUEUE 3
1264  #define NF_REPEAT 4
1265  #define NF_STOP 5
1266 -#define NF_MAX_VERDICT NF_STOP
1267 +#define NF_IMQ_QUEUE 6
1268 +#define NF_MAX_VERDICT NF_IMQ_QUEUE
1269  
1270  /* we overload the higher bits for encoding auxiliary data such as the queue
1271   * number or errno values. Not nice, but better than additional function
1272 diff --git a/net/core/dev.c b/net/core/dev.c
1273 index ae00b89..1cdcd02 100644
1274 --- a/net/core/dev.c
1275 +++ b/net/core/dev.c
1276 @@ -137,6 +137,9 @@
1277  #include <linux/errqueue.h>
1278  #include <linux/hrtimer.h>
1279  #include <linux/netfilter_ingress.h>
1280 +#if defined(CONFIG_IMQ) || defined(CONFIG_IMQ_MODULE)
1281 +#include <linux/imq.h>
1282 +#endif
1283  
1284  #include "net-sysfs.h"
1285  
1286 @@ -2705,7 +2708,12 @@ static int xmit_one(struct sk_buff *skb, struct net_device *dev,
1287         unsigned int len;
1288         int rc;
1289  
1290 +#if defined(CONFIG_IMQ) || defined(CONFIG_IMQ_MODULE)
1291 +       if ((!list_empty(&ptype_all) || !list_empty(&dev->ptype_all)) &&
1292 +               !(skb->imq_flags & IMQ_F_ENQUEUE))
1293 +#else
1294         if (!list_empty(&ptype_all) || !list_empty(&dev->ptype_all))
1295 +#endif
1296                 dev_queue_xmit_nit(skb, dev);
1297  
1298         len = skb->len;
1299 @@ -2743,6 +2751,7 @@ out:
1300         *ret = rc;
1301         return skb;
1302  }
1303 +EXPORT_SYMBOL(dev_hard_start_xmit);
1304  
1305  static struct sk_buff *validate_xmit_vlan(struct sk_buff *skb,
1306                                           netdev_features_t features)
1307 diff --git a/net/core/skbuff.c b/net/core/skbuff.c
1308 index b2df375..bc3c51e 100644
1309 --- a/net/core/skbuff.c
1310 +++ b/net/core/skbuff.c
1311 @@ -79,6 +79,87 @@
1312  static struct kmem_cache *skbuff_fclone_cache __read_mostly;
1313  int sysctl_max_skb_frags __read_mostly = MAX_SKB_FRAGS;
1314  EXPORT_SYMBOL(sysctl_max_skb_frags);
1315 +#if defined(CONFIG_IMQ) || defined(CONFIG_IMQ_MODULE)
1316 +static struct kmem_cache *skbuff_cb_store_cache __read_mostly;
1317 +#endif
1318 +
1319 +#if defined(CONFIG_IMQ) || defined(CONFIG_IMQ_MODULE)
1320 +/* Control buffer save/restore for IMQ devices */
1321 +struct skb_cb_table {
1322 +       char                    cb[48] __aligned(8);
1323 +       void                    *cb_next;
1324 +       atomic_t                refcnt;
1325 +};
1326 +
1327 +static DEFINE_SPINLOCK(skb_cb_store_lock);
1328 +
1329 +int skb_save_cb(struct sk_buff *skb)
1330 +{
1331 +       struct skb_cb_table *next;
1332 +
1333 +       next = kmem_cache_alloc(skbuff_cb_store_cache, GFP_ATOMIC);
1334 +       if (!next)
1335 +               return -ENOMEM;
1336 +
1337 +       BUILD_BUG_ON(sizeof(skb->cb) != sizeof(next->cb));
1338 +
1339 +       memcpy(next->cb, skb->cb, sizeof(skb->cb));
1340 +       next->cb_next = skb->cb_next;
1341 +
1342 +       atomic_set(&next->refcnt, 1);
1343 +
1344 +       skb->cb_next = next;
1345 +       return 0;
1346 +}
1347 +EXPORT_SYMBOL(skb_save_cb);
1348 +
1349 +int skb_restore_cb(struct sk_buff *skb)
1350 +{
1351 +       struct skb_cb_table *next;
1352 +
1353 +       if (!skb->cb_next)
1354 +               return 0;
1355 +
1356 +       next = skb->cb_next;
1357 +
1358 +       BUILD_BUG_ON(sizeof(skb->cb) != sizeof(next->cb));
1359 +
1360 +       memcpy(skb->cb, next->cb, sizeof(skb->cb));
1361 +       skb->cb_next = next->cb_next;
1362 +
1363 +       spin_lock(&skb_cb_store_lock);
1364 +
1365 +       if (atomic_dec_and_test(&next->refcnt))
1366 +               kmem_cache_free(skbuff_cb_store_cache, next);
1367 +
1368 +       spin_unlock(&skb_cb_store_lock);
1369 +
1370 +       return 0;
1371 +}
1372 +EXPORT_SYMBOL(skb_restore_cb);
1373 +
1374 +static void skb_copy_stored_cb(struct sk_buff *   , const struct sk_buff *     ) __attribute__ ((unused));
1375 +static void skb_copy_stored_cb(struct sk_buff *new, const struct sk_buff *__old)
1376 +{
1377 +       struct skb_cb_table *next;
1378 +       struct sk_buff *old;
1379 +
1380 +       if (!__old->cb_next) {
1381 +               new->cb_next = NULL;
1382 +               return;
1383 +       }
1384 +
1385 +       spin_lock(&skb_cb_store_lock);
1386 +
1387 +       old = (struct sk_buff *)__old;
1388 +
1389 +       next = old->cb_next;
1390 +       atomic_inc(&next->refcnt);
1391 +       new->cb_next = next;
1392 +
1393 +       spin_unlock(&skb_cb_store_lock);
1394 +}
1395 +#endif
1396  
1397  /**
1398   *     skb_panic - private function for out-of-line support
1399 @@ -643,6 +724,28 @@ static void skb_release_head_state(struct sk_buff *skb)
1400                 WARN_ON(in_irq());
1401                 skb->destructor(skb);
1402         }
1403 +#if defined(CONFIG_IMQ) || defined(CONFIG_IMQ_MODULE)
1404 +       /*
1405 +        * This should not happen. When it does, avoid memleak by restoring
1406 +        * the chain of cb-backups.
1407 +        */
1408 +       while (skb->cb_next != NULL) {
1409 +               if (net_ratelimit())
1410 +                       pr_warn("IMQ: kfree_skb: skb->cb_next: %08x\n",
1411 +                               (unsigned int)(uintptr_t)skb->cb_next);
1412 +
1413 +               skb_restore_cb(skb);
1414 +       }
1415 +       /*
1416 +        * This should not happen either, nf_queue_entry is nullified in
1417 +        * imq_dev_xmit(). If we have non-NULL nf_queue_entry then we are
1418 +        * leaking entry pointers, maybe memory. We don't know if this is
1419 +        * pointer to already freed memory, or should this be freed.
1420 +        * If this happens we need to add refcounting, etc for nf_queue_entry.
1421 +        */
1422 +       if (skb->nf_queue_entry && net_ratelimit())
1423 +               pr_warn("%s\n", "IMQ: kfree_skb: skb->nf_queue_entry != NULL");
1424 +#endif
1425  #if IS_ENABLED(CONFIG_NF_CONNTRACK)
1426         nf_conntrack_put(skb->nfct);
1427  #endif
1428 @@ -765,6 +868,10 @@ static void __copy_skb_header(struct sk_buff *new, const struct sk_buff *old)
1429         new->sp                 = secpath_get(old->sp);
1430  #endif
1431         __nf_copy(new, old, false);
1432 +#if defined(CONFIG_IMQ) || defined(CONFIG_IMQ_MODULE)
1433 +       new->cb_next = NULL;
1434 +       /*skb_copy_stored_cb(new, old);*/
1435 +#endif
1436  
1437         /* Note : this field could be in headers_start/headers_end section
1438          * It is not yet because we do not want to have a 16 bit hole
1439 @@ -3325,6 +3432,13 @@ void __init skb_init(void)
1440                                                 0,
1441                                                 SLAB_HWCACHE_ALIGN|SLAB_PANIC,
1442                                                 NULL);
1443 +#if defined(CONFIG_IMQ) || defined(CONFIG_IMQ_MODULE)
1444 +       skbuff_cb_store_cache = kmem_cache_create("skbuff_cb_store_cache",
1445 +                                                 sizeof(struct skb_cb_table),
1446 +                                                 0,
1447 +                                                 SLAB_HWCACHE_ALIGN|SLAB_PANIC,
1448 +                                                 NULL);
1449 +#endif
1450  }
1451  
1452  /**
1453 diff --git a/net/ipv6/ip6_output.c b/net/ipv6/ip6_output.c
1454 index e6a7bd15..c81496e 100644
1455 --- a/net/ipv6/ip6_output.c
1456 +++ b/net/ipv6/ip6_output.c
1457 @@ -65,9 +65,6 @@ static int ip6_finish_output2(struct net *net, struct sock *sk, struct sk_buff *
1458         struct in6_addr *nexthop;
1459         int ret;
1460  
1461 -       skb->protocol = htons(ETH_P_IPV6);
1462 -       skb->dev = dev;
1463 -
1464         if (ipv6_addr_is_multicast(&ipv6_hdr(skb)->daddr)) {
1465                 struct inet6_dev *idev = ip6_dst_idev(skb_dst(skb));
1466  
1467 @@ -142,6 +139,13 @@ int ip6_output(struct net *net, struct sock *sk, struct sk_buff *skb)
1468                 return 0;
1469         }
1470  
1471 +       /*
1472 +       * IMQ-patch: moved setting skb->dev and skb->protocol from
1473 +       * ip6_finish_output2 to fix crashing at netif_skb_features().
1474 +       */
1475 +       skb->protocol = htons(ETH_P_IPV6);
1476 +       skb->dev = dev;
1477 +
1478         return NF_HOOK_COND(NFPROTO_IPV6, NF_INET_POST_ROUTING,
1479                             net, sk, skb, NULL, dev,
1480                             ip6_finish_output,
1481 diff --git a/net/netfilter/Kconfig b/net/netfilter/Kconfig
1482 index 4692782..dce47db 100644
1483 --- a/net/netfilter/Kconfig
1484 +++ b/net/netfilter/Kconfig
1485 @@ -785,6 +785,18 @@ config NETFILTER_XT_TARGET_LOG
1486  
1487           To compile it as a module, choose M here.  If unsure, say N.
1488  
1489 +config NETFILTER_XT_TARGET_IMQ
1490 +        tristate '"IMQ" target support'
1491 +       depends on NETFILTER_XTABLES
1492 +       depends on IP_NF_MANGLE || IP6_NF_MANGLE
1493 +       select IMQ
1494 +       default m if NETFILTER_ADVANCED=n
1495 +        help
1496 +          This option adds a `IMQ' target which is used to specify if and
1497 +          to which imq device packets should get enqueued/dequeued.
1498 +
1499 +          To compile it as a module, choose M here.  If unsure, say N.
1500 +
1501  config NETFILTER_XT_TARGET_MARK
1502         tristate '"MARK" target support'
1503         depends on NETFILTER_ADVANCED
1504 diff --git a/net/netfilter/Makefile b/net/netfilter/Makefile
1505 index 7638c36..614ad8a 100644
1506 --- a/net/netfilter/Makefile
1507 +++ b/net/netfilter/Makefile
1508 @@ -108,6 +108,7 @@ obj-$(CONFIG_NETFILTER_XT_TARGET_CT) += xt_CT.o
1509  obj-$(CONFIG_NETFILTER_XT_TARGET_DSCP) += xt_DSCP.o
1510  obj-$(CONFIG_NETFILTER_XT_TARGET_HL) += xt_HL.o
1511  obj-$(CONFIG_NETFILTER_XT_TARGET_HMARK) += xt_HMARK.o
1512 +obj-$(CONFIG_NETFILTER_XT_TARGET_IMQ) += xt_IMQ.o
1513  obj-$(CONFIG_NETFILTER_XT_TARGET_LED) += xt_LED.o
1514  obj-$(CONFIG_NETFILTER_XT_TARGET_LOG) += xt_LOG.o
1515  obj-$(CONFIG_NETFILTER_XT_TARGET_NETMAP) += xt_NETMAP.o
1516 diff --git a/net/netfilter/core.c b/net/netfilter/core.c
1517 index f39276d..9877a27 100644
1518 --- a/net/netfilter/core.c
1519 +++ b/net/netfilter/core.c
1520 @@ -311,9 +311,11 @@ next_hook:
1521                 ret = NF_DROP_GETERR(verdict);
1522                 if (ret == 0)
1523                         ret = -EPERM;
1524 -       } else if ((verdict & NF_VERDICT_MASK) == NF_QUEUE) {
1525 +       } else if ((verdict & NF_VERDICT_MASK) == NF_QUEUE ||
1526 +               (verdict & NF_VERDICT_MASK) == NF_IMQ_QUEUE) {
1527                 int err = nf_queue(skb, elem, state,
1528 -                                  verdict >> NF_VERDICT_QBITS);
1529 +                                  verdict >> NF_VERDICT_QBITS,
1530 +                                 verdict & NF_VERDICT_MASK);
1531                 if (err < 0) {
1532                         if (err == -ESRCH &&
1533                            (verdict & NF_VERDICT_FLAG_QUEUE_BYPASS))
1534 diff --git a/net/netfilter/nf_internals.h b/net/netfilter/nf_internals.h
1535 index 0655225..25d4141 100644
1536 --- a/net/netfilter/nf_internals.h
1537 +++ b/net/netfilter/nf_internals.h
1538 @@ -18,7 +18,7 @@ unsigned int nf_iterate(struct list_head *head, struct sk_buff *skb,
1539  
1540  /* nf_queue.c */
1541  int nf_queue(struct sk_buff *skb, struct nf_hook_ops *elem,
1542 -            struct nf_hook_state *state, unsigned int queuenum);
1543 +            struct nf_hook_state *state, unsigned int queuenum, unsigned int queuetype);
1544  void nf_queue_nf_hook_drop(struct net *net, struct nf_hook_ops *ops);
1545  int __init netfilter_queue_init(void);
1546  
1547 diff --git a/net/netfilter/nf_queue.c b/net/netfilter/nf_queue.c
1548 index 5baa8e2..9740e8c 100644
1549 --- a/net/netfilter/nf_queue.c
1550 +++ b/net/netfilter/nf_queue.c
1551 @@ -28,6 +28,23 @@
1552   */
1553  static const struct nf_queue_handler __rcu *queue_handler __read_mostly;
1554  
1555 +#if defined(CONFIG_IMQ) || defined(CONFIG_IMQ_MODULE)
1556 +static const struct nf_queue_handler __rcu *queue_imq_handler __read_mostly;
1557 +
1558 +void nf_register_queue_imq_handler(const struct nf_queue_handler *qh)
1559 +{
1560 +       rcu_assign_pointer(queue_imq_handler, qh);
1561 +}
1562 +EXPORT_SYMBOL_GPL(nf_register_queue_imq_handler);
1563 +
1564 +void nf_unregister_queue_imq_handler(void)
1565 +{
1566 +       RCU_INIT_POINTER(queue_imq_handler, NULL);
1567 +       synchronize_rcu();
1568 +}
1569 +EXPORT_SYMBOL_GPL(nf_unregister_queue_imq_handler);
1570 +#endif
1571 +
1572  /* return EBUSY when somebody else is registered, return EEXIST if the
1573   * same handler is registered, return 0 in case of success. */
1574  void nf_register_queue_handler(const struct nf_queue_handler *qh)
1575 @@ -116,7 +133,8 @@ void nf_queue_nf_hook_drop(struct net *net, struct nf_hook_ops *ops)
1576  int nf_queue(struct sk_buff *skb,
1577              struct nf_hook_ops *elem,
1578              struct nf_hook_state *state,
1579 -            unsigned int queuenum)
1580 +            unsigned int queuenum,
1581 +                unsigned int queuetype)
1582  {
1583         int status = -ENOENT;
1584         struct nf_queue_entry *entry = NULL;
1585 @@ -124,7 +142,17 @@ int nf_queue(struct sk_buff *skb,
1586         const struct nf_queue_handler *qh;
1587  
1588         /* QUEUE == DROP if no one is waiting, to be safe. */
1589 -       qh = rcu_dereference(queue_handler);
1590 +       if (queuetype == NF_IMQ_QUEUE) {
1591 +#if defined(CONFIG_IMQ) || defined(CONFIG_IMQ_MODULE)
1592 +               qh = rcu_dereference(queue_imq_handler);
1593 +#else
1594 +               BUG();
1595 +               goto err_unlock;
1596 +#endif
1597 +       } else {
1598 +               qh = rcu_dereference(queue_handler);
1599 +       }
1600 +
1601         if (!qh) {
1602                 status = -ESRCH;
1603                 goto err;
1604 @@ -199,8 +227,10 @@ void nf_reinject(struct nf_queue_entry *entry, unsigned int verdict)
1605                 local_bh_enable();
1606                 break;
1607         case NF_QUEUE:
1608 +       case NF_IMQ_QUEUE:
1609                 err = nf_queue(skb, elem, &entry->state,
1610 -                              verdict >> NF_VERDICT_QBITS);
1611 +                              verdict >> NF_VERDICT_QBITS,
1612 +                                  verdict & NF_VERDICT_MASK);
1613                 if (err < 0) {
1614                         if (err == -ESRCH &&
1615                            (verdict & NF_VERDICT_FLAG_QUEUE_BYPASS))
1616 diff --git a/net/netfilter/xt_IMQ.c b/net/netfilter/xt_IMQ.c
1617 new file mode 100644
1618 index 0000000..86d7b84
1619 --- /dev/null
1620 +++ b/net/netfilter/xt_IMQ.c
1621 @@ -0,0 +1,72 @@
1622 +/*
1623 + * This target marks packets to be enqueued to an imq device
1624 + */
1625 +#include <linux/module.h>
1626 +#include <linux/skbuff.h>
1627 +#include <linux/netfilter/x_tables.h>
1628 +#include <linux/netfilter/xt_IMQ.h>
1629 +#include <linux/imq.h>
1630 +
1631 +static unsigned int imq_target(struct sk_buff *pskb,
1632 +                               const struct xt_action_param *par)
1633 +{
1634 +       const struct xt_imq_info *mr = par->targinfo;
1635 +
1636 +       pskb->imq_flags = (mr->todev & IMQ_F_IFMASK) | IMQ_F_ENQUEUE;
1637 +
1638 +       return XT_CONTINUE;
1639 +}
1640 +
1641 +static int imq_checkentry(const struct xt_tgchk_param *par)
1642 +{
1643 +       struct xt_imq_info *mr = par->targinfo;
1644 +
1645 +       if (mr->todev > IMQ_MAX_DEVS - 1) {
1646 +               pr_warn("IMQ: invalid device specified, highest is %u\n",
1647 +                       IMQ_MAX_DEVS - 1);
1648 +               return -EINVAL;
1649 +       }
1650 +
1651 +       return 0;
1652 +}
1653 +
1654 +static struct xt_target xt_imq_reg[] __read_mostly = {
1655 +       {
1656 +               .name           = "IMQ",
1657 +               .family         = AF_INET,
1658 +               .checkentry     = imq_checkentry,
1659 +               .target         = imq_target,
1660 +               .targetsize     = sizeof(struct xt_imq_info),
1661 +               .table          = "mangle",
1662 +               .me             = THIS_MODULE
1663 +       },
1664 +       {
1665 +               .name           = "IMQ",
1666 +               .family         = AF_INET6,
1667 +               .checkentry     = imq_checkentry,
1668 +               .target         = imq_target,
1669 +               .targetsize     = sizeof(struct xt_imq_info),
1670 +               .table          = "mangle",
1671 +               .me             = THIS_MODULE
1672 +       },
1673 +};
1674 +
1675 +static int __init imq_init(void)
1676 +{
1677 +       return xt_register_targets(xt_imq_reg, ARRAY_SIZE(xt_imq_reg));
1678 +}
1679 +
1680 +static void __exit imq_fini(void)
1681 +{
1682 +       xt_unregister_targets(xt_imq_reg, ARRAY_SIZE(xt_imq_reg));
1683 +}
1684 +
1685 +module_init(imq_init);
1686 +module_exit(imq_fini);
1687 +
1688 +MODULE_AUTHOR("http://https://github.com/imq/linuximq");
1689 +MODULE_DESCRIPTION("Pseudo-driver for the intermediate queue device. See https://github.com/imq/linuximq/wiki for more information.");
1690 +MODULE_LICENSE("GPL");
1691 +MODULE_ALIAS("ipt_IMQ");
1692 +MODULE_ALIAS("ip6t_IMQ");
1693 +
1694 diff --git a/net/sched/sch_generic.c b/net/sched/sch_generic.c
1695 index e82a1ad..b55331d 100644
1696 --- a/net/sched/sch_generic.c
1697 +++ b/net/sched/sch_generic.c
1698 @@ -108,6 +108,14 @@ static struct sk_buff *dequeue_skb(struct Qdisc *q, bool *validate,
1699         return skb;
1700  }
1701  
1702 +struct sk_buff *qdisc_dequeue_skb(struct Qdisc *q, bool *validate)
1703 +{
1704 +       int packets;
1705 +
1706 +       return dequeue_skb(q, validate, &packets);
1707 +}
1708 +EXPORT_SYMBOL(qdisc_dequeue_skb);
1709 +
1710  static inline int handle_dev_cpu_collision(struct sk_buff *skb,
1711                                            struct netdev_queue *dev_queue,
1712                                            struct Qdisc *q)
This page took 0.393601 seconds and 3 git commands to generate.