mountkernfs: fix mount of /sys and /proc without initrd.
[people/teissler/ipfire-2.x.git] / src / patches / imq_kernel3.10.patch
1 diff -uNr linux-3.9.1/drivers/net/imq.c linux-3.9.1-imqmq/drivers/net/imq.c
2 --- linux-3.9.1/drivers/net/imq.c 1970-01-01 02:00:00.000000000 +0200
3 +++ linux-3.9.1-imqmq/drivers/net/imq.c 2013-05-08 17:30:41.715552053 +0300
4 @@ -0,0 +1,861 @@
5 +/*
6 + * Pseudo-driver for the intermediate queue device.
7 + *
8 + * This program is free software; you can redistribute it and/or
9 + * modify it under the terms of the GNU General Public License
10 + * as published by the Free Software Foundation; either version
11 + * 2 of the License, or (at your option) any later version.
12 + *
13 + * Authors: Patrick McHardy, <kaber@trash.net>
14 + *
15 + * The first version was written by Martin Devera, <devik@cdi.cz>
16 + *
17 + * Credits: Jan Rafaj <imq2t@cedric.vabo.cz>
18 + * - Update patch to 2.4.21
19 + * Sebastian Strollo <sstrollo@nortelnetworks.com>
20 + * - Fix "Dead-loop on netdevice imq"-issue
21 + * Marcel Sebek <sebek64@post.cz>
22 + * - Update to 2.6.2-rc1
23 + *
24 + * After some time of inactivity there is a group taking care
25 + * of IMQ again: http://www.linuximq.net
26 + *
27 + *
28 + * 2004/06/30 - New version of IMQ patch to kernels <=2.6.7
29 + * including the following changes:
30 + *
31 + * - Correction of ipv6 support "+"s issue (Hasso Tepper)
32 + * - Correction of imq_init_devs() issue that resulted in
33 + * kernel OOPS unloading IMQ as module (Norbert Buchmuller)
34 + * - Addition of functionality to choose number of IMQ devices
35 + * during kernel config (Andre Correa)
36 + * - Addition of functionality to choose how IMQ hooks on
37 + * PRE and POSTROUTING (after or before NAT) (Andre Correa)
38 + * - Cosmetic corrections (Norbert Buchmuller) (Andre Correa)
39 + *
40 + *
41 + * 2005/12/16 - IMQ versions between 2.6.7 and 2.6.13 were
42 + * released with almost no problems. 2.6.14-x was released
43 + * with some important changes: nfcache was removed; After
44 + * some weeks of trouble we figured out that some IMQ fields
45 + * in skb were missing in skbuff.c - skb_clone and copy_skb_header.
46 + * These functions are correctly patched by this new patch version.
47 + *
48 + * Thanks for all who helped to figure out all the problems with
49 + * 2.6.14.x: Patrick McHardy, Rune Kock, VeNoMouS, Max CtRiX,
50 + * Kevin Shanahan, Richard Lucassen, Valery Dachev (hopefully
51 + * I didn't forget anybody). I apologize again for my lack of time.
52 + *
53 + *
54 + * 2008/06/17 - 2.6.25 - Changed imq.c to use qdisc_run() instead
55 + * of qdisc_restart() and moved qdisc_run() to tasklet to avoid
56 + * recursive locking. New initialization routines to fix 'rmmod' not
57 + * working anymore. Used code from ifb.c. (Jussi Kivilinna)
58 + *
59 + * 2008/08/06 - 2.6.26 - (JK)
60 + * - Replaced tasklet with 'netif_schedule()'.
61 + * - Cleaned up and added comments for imq_nf_queue().
62 + *
63 + * 2009/04/12
64 + * - Add skb_save_cb/skb_restore_cb helper functions for backuping
65 + * control buffer. This is needed because qdisc-layer on kernels
66 + * 2.6.27 and newer overwrite control buffer. (Jussi Kivilinna)
67 + * - Add better locking for IMQ device. Hopefully this will solve
68 + * SMP issues. (Jussi Kivilinna)
69 + * - Port to 2.6.27
70 + * - Port to 2.6.28
71 + * - Port to 2.6.29 + fix rmmod not working
72 + *
73 + * 2009/04/20 - (Jussi Kivilinna)
74 + * - Use netdevice feature flags to avoid extra packet handling
75 + * by core networking layer and possibly increase performance.
76 + *
77 + * 2009/09/26 - (Jussi Kivilinna)
78 + * - Add imq_nf_reinject_lockless to fix deadlock with
79 + * imq_nf_queue/imq_nf_reinject.
80 + *
81 + * 2009/12/08 - (Jussi Kivilinna)
82 + * - Port to 2.6.32
83 + * - Add check for skb->nf_queue_entry==NULL in imq_dev_xmit()
84 + * - Also add better error checking for skb->nf_queue_entry usage
85 + *
86 + * 2010/02/25 - (Jussi Kivilinna)
87 + * - Port to 2.6.33
88 + *
89 + * 2010/08/15 - (Jussi Kivilinna)
90 + * - Port to 2.6.35
91 + * - Simplify hook registration by using nf_register_hooks.
92 + * - nf_reinject doesn't need spinlock around it, therefore remove
93 + * imq_nf_reinject function. Other nf_reinject users protect
94 + * their own data with spinlock. With IMQ however all data is
95 + * needed is stored per skbuff, so no locking is needed.
96 + * - Changed IMQ to use 'separate' NF_IMQ_QUEUE instead of
97 + * NF_QUEUE, this allows working coexistance of IMQ and other
98 + * NF_QUEUE users.
99 + * - Make IMQ multi-queue. Number of IMQ device queues can be
100 + * increased with 'numqueues' module parameters. Default number
101 + * of queues is 1, in other words by default IMQ works as
102 + * single-queue device. Multi-queue selection is based on
103 + * IFB multi-queue patch by Changli Gao <xiaosuo@gmail.com>.
104 + *
105 + * 2011/03/18 - (Jussi Kivilinna)
106 + * - Port to 2.6.38
107 + *
108 + * 2011/07/12 - (syoder89@gmail.com)
109 + * - Crash fix that happens when the receiving interface has more
110 + * than one queue (add missing skb_set_queue_mapping in
111 + * imq_select_queue).
112 + *
113 + * 2011/07/26 - (Jussi Kivilinna)
114 + * - Add queue mapping checks for packets exiting IMQ.
115 + * - Port to 3.0
116 + *
117 + * 2011/08/16 - (Jussi Kivilinna)
118 + * - Clear IFF_TX_SKB_SHARING flag that was added for linux 3.0.2
119 + *
120 + * 2011/11/03 - Germano Michel <germanomichel@gmail.com>
121 + * - Fix IMQ for net namespaces
122 + *
123 + * 2011/11/04 - Jussi Kivilinna <jussi.kivilinna@mbnet.fi>
124 + * - Port to 3.1
125 + * - Clean-up, move 'get imq device pointer by imqX name' to
126 + * separate function from imq_nf_queue().
127 + *
128 + * 2012/01/05 - Jussi Kivilinna <jussi.kivilinna@mbnet.fi>
129 + * - Port to 3.2
130 + *
131 + * 2012/03/19 - Jussi Kivilinna <jussi.kivilinna@mbnet.fi>
132 + * - Port to 3.3
133 + *
134 + * 2012/12/12 - Jussi Kivilinna <jussi.kivilinna@mbnet.fi>
135 + * - Port to 3.7
136 + * - Fix checkpatch.pl warnings
137 + *
138 + * Also, many thanks to pablo Sebastian Greco for making the initial
139 + * patch and to those who helped the testing.
140 + *
141 + * More info at: http://www.linuximq.net/ (Andre Correa)
142 + */
143 +
144 +#include <linux/module.h>
145 +#include <linux/kernel.h>
146 +#include <linux/moduleparam.h>
147 +#include <linux/list.h>
148 +#include <linux/skbuff.h>
149 +#include <linux/netdevice.h>
150 +#include <linux/etherdevice.h>
151 +#include <linux/rtnetlink.h>
152 +#include <linux/if_arp.h>
153 +#include <linux/netfilter.h>
154 +#include <linux/netfilter_ipv4.h>
155 +#if defined(CONFIG_IPV6) || defined(CONFIG_IPV6_MODULE)
156 + #include <linux/netfilter_ipv6.h>
157 +#endif
158 +#include <linux/imq.h>
159 +#include <net/pkt_sched.h>
160 +#include <net/netfilter/nf_queue.h>
161 +#include <net/sock.h>
162 +#include <linux/ip.h>
163 +#include <linux/ipv6.h>
164 +#include <linux/if_vlan.h>
165 +#include <linux/if_pppox.h>
166 +#include <net/ip.h>
167 +#include <net/ipv6.h>
168 +
169 +static int imq_nf_queue(struct nf_queue_entry *entry, unsigned int queue_num);
170 +
171 +static nf_hookfn imq_nf_hook;
172 +
173 +static struct nf_hook_ops imq_ops[] = {
174 + {
175 + /* imq_ingress_ipv4 */
176 + .hook = imq_nf_hook,
177 + .owner = THIS_MODULE,
178 + .pf = PF_INET,
179 + .hooknum = NF_INET_PRE_ROUTING,
180 +#if defined(CONFIG_IMQ_BEHAVIOR_BA) || defined(CONFIG_IMQ_BEHAVIOR_BB)
181 + .priority = NF_IP_PRI_MANGLE + 1,
182 +#else
183 + .priority = NF_IP_PRI_NAT_DST + 1,
184 +#endif
185 + },
186 + {
187 + /* imq_egress_ipv4 */
188 + .hook = imq_nf_hook,
189 + .owner = THIS_MODULE,
190 + .pf = PF_INET,
191 + .hooknum = NF_INET_POST_ROUTING,
192 +#if defined(CONFIG_IMQ_BEHAVIOR_AA) || defined(CONFIG_IMQ_BEHAVIOR_BA)
193 + .priority = NF_IP_PRI_LAST,
194 +#else
195 + .priority = NF_IP_PRI_NAT_SRC - 1,
196 +#endif
197 + },
198 +#if defined(CONFIG_IPV6) || defined(CONFIG_IPV6_MODULE)
199 + {
200 + /* imq_ingress_ipv6 */
201 + .hook = imq_nf_hook,
202 + .owner = THIS_MODULE,
203 + .pf = PF_INET6,
204 + .hooknum = NF_INET_PRE_ROUTING,
205 +#if defined(CONFIG_IMQ_BEHAVIOR_BA) || defined(CONFIG_IMQ_BEHAVIOR_BB)
206 + .priority = NF_IP6_PRI_MANGLE + 1,
207 +#else
208 + .priority = NF_IP6_PRI_NAT_DST + 1,
209 +#endif
210 + },
211 + {
212 + /* imq_egress_ipv6 */
213 + .hook = imq_nf_hook,
214 + .owner = THIS_MODULE,
215 + .pf = PF_INET6,
216 + .hooknum = NF_INET_POST_ROUTING,
217 +#if defined(CONFIG_IMQ_BEHAVIOR_AA) || defined(CONFIG_IMQ_BEHAVIOR_BA)
218 + .priority = NF_IP6_PRI_LAST,
219 +#else
220 + .priority = NF_IP6_PRI_NAT_SRC - 1,
221 +#endif
222 + },
223 +#endif
224 +};
225 +
226 +#if defined(CONFIG_IMQ_NUM_DEVS)
227 +static int numdevs = CONFIG_IMQ_NUM_DEVS;
228 +#else
229 +static int numdevs = IMQ_MAX_DEVS;
230 +#endif
231 +
232 +static struct net_device *imq_devs_cache[IMQ_MAX_DEVS];
233 +
234 +#define IMQ_MAX_QUEUES 32
235 +static int numqueues = 1;
236 +static u32 imq_hashrnd;
237 +
238 +static inline __be16 pppoe_proto(const struct sk_buff *skb)
239 +{
240 + return *((__be16 *)(skb_mac_header(skb) + ETH_HLEN +
241 + sizeof(struct pppoe_hdr)));
242 +}
243 +
244 +static u16 imq_hash(struct net_device *dev, struct sk_buff *skb)
245 +{
246 + unsigned int pull_len;
247 + u16 protocol = skb->protocol;
248 + u32 addr1, addr2;
249 + u32 hash, ihl = 0;
250 + union {
251 + u16 in16[2];
252 + u32 in32;
253 + } ports;
254 + u8 ip_proto;
255 +
256 + pull_len = 0;
257 +
258 +recheck:
259 + switch (protocol) {
260 + case htons(ETH_P_8021Q): {
261 + if (unlikely(skb_pull(skb, VLAN_HLEN) == NULL))
262 + goto other;
263 +
264 + pull_len += VLAN_HLEN;
265 + skb->network_header += VLAN_HLEN;
266 +
267 + protocol = vlan_eth_hdr(skb)->h_vlan_encapsulated_proto;
268 + goto recheck;
269 + }
270 +
271 + case htons(ETH_P_PPP_SES): {
272 + if (unlikely(skb_pull(skb, PPPOE_SES_HLEN) == NULL))
273 + goto other;
274 +
275 + pull_len += PPPOE_SES_HLEN;
276 + skb->network_header += PPPOE_SES_HLEN;
277 +
278 + protocol = pppoe_proto(skb);
279 + goto recheck;
280 + }
281 +
282 + case htons(ETH_P_IP): {
283 + const struct iphdr *iph = ip_hdr(skb);
284 +
285 + if (unlikely(!pskb_may_pull(skb, sizeof(struct iphdr))))
286 + goto other;
287 +
288 + addr1 = iph->daddr;
289 + addr2 = iph->saddr;
290 +
291 + ip_proto = !(ip_hdr(skb)->frag_off & htons(IP_MF | IP_OFFSET)) ?
292 + iph->protocol : 0;
293 + ihl = ip_hdrlen(skb);
294 +
295 + break;
296 + }
297 +#if defined(CONFIG_IPV6) || defined(CONFIG_IPV6_MODULE)
298 + case htons(ETH_P_IPV6): {
299 + const struct ipv6hdr *iph = ipv6_hdr(skb);
300 + __be16 fo = 0;
301 +
302 + if (unlikely(!pskb_may_pull(skb, sizeof(struct ipv6hdr))))
303 + goto other;
304 +
305 + addr1 = iph->daddr.s6_addr32[3];
306 + addr2 = iph->saddr.s6_addr32[3];
307 + ihl = ipv6_skip_exthdr(skb, sizeof(struct ipv6hdr), &ip_proto,
308 + &fo);
309 + if (unlikely(ihl < 0))
310 + goto other;
311 +
312 + break;
313 + }
314 +#endif
315 + default:
316 +other:
317 + if (pull_len != 0) {
318 + skb_push(skb, pull_len);
319 + skb->network_header -= pull_len;
320 + }
321 +
322 + return (u16)(ntohs(protocol) % dev->real_num_tx_queues);
323 + }
324 +
325 + if (addr1 > addr2)
326 + swap(addr1, addr2);
327 +
328 + switch (ip_proto) {
329 + case IPPROTO_TCP:
330 + case IPPROTO_UDP:
331 + case IPPROTO_DCCP:
332 + case IPPROTO_ESP:
333 + case IPPROTO_AH:
334 + case IPPROTO_SCTP:
335 + case IPPROTO_UDPLITE: {
336 + if (likely(skb_copy_bits(skb, ihl, &ports.in32, 4) >= 0)) {
337 + if (ports.in16[0] > ports.in16[1])
338 + swap(ports.in16[0], ports.in16[1]);
339 + break;
340 + }
341 + /* fall-through */
342 + }
343 + default:
344 + ports.in32 = 0;
345 + break;
346 + }
347 +
348 + if (pull_len != 0) {
349 + skb_push(skb, pull_len);
350 + skb->network_header -= pull_len;
351 + }
352 +
353 + hash = jhash_3words(addr1, addr2, ports.in32, imq_hashrnd ^ ip_proto);
354 +
355 + return (u16)(((u64)hash * dev->real_num_tx_queues) >> 32);
356 +}
357 +
358 +static inline bool sk_tx_queue_recorded(struct sock *sk)
359 +{
360 + return (sk_tx_queue_get(sk) >= 0);
361 +}
362 +
363 +static struct netdev_queue *imq_select_queue(struct net_device *dev,
364 + struct sk_buff *skb)
365 +{
366 + u16 queue_index = 0;
367 + u32 hash;
368 +
369 + if (likely(dev->real_num_tx_queues == 1))
370 + goto out;
371 +
372 + /* IMQ can be receiving ingress or engress packets. */
373 +
374 + /* Check first for if rx_queue is set */
375 + if (skb_rx_queue_recorded(skb)) {
376 + queue_index = skb_get_rx_queue(skb);
377 + goto out;
378 + }
379 +
380 + /* Check if socket has tx_queue set */
381 + if (sk_tx_queue_recorded(skb->sk)) {
382 + queue_index = sk_tx_queue_get(skb->sk);
383 + goto out;
384 + }
385 +
386 + /* Try use socket hash */
387 + if (skb->sk && skb->sk->sk_hash) {
388 + hash = skb->sk->sk_hash;
389 + queue_index =
390 + (u16)(((u64)hash * dev->real_num_tx_queues) >> 32);
391 + goto out;
392 + }
393 +
394 + /* Generate hash from packet data */
395 + queue_index = imq_hash(dev, skb);
396 +
397 +out:
398 + if (unlikely(queue_index >= dev->real_num_tx_queues))
399 + queue_index = (u16)((u32)queue_index % dev->real_num_tx_queues);
400 +
401 + skb_set_queue_mapping(skb, queue_index);
402 + return netdev_get_tx_queue(dev, queue_index);
403 +}
404 +
405 +static struct net_device_stats *imq_get_stats(struct net_device *dev)
406 +{
407 + return &dev->stats;
408 +}
409 +
410 +/* called for packets kfree'd in qdiscs at places other than enqueue */
411 +static void imq_skb_destructor(struct sk_buff *skb)
412 +{
413 + struct nf_queue_entry *entry = skb->nf_queue_entry;
414 +
415 + skb->nf_queue_entry = NULL;
416 +
417 + if (entry) {
418 + nf_queue_entry_release_refs(entry);
419 + kfree(entry);
420 + }
421 +
422 + skb_restore_cb(skb); /* kfree backup */
423 +}
424 +
425 +static void imq_done_check_queue_mapping(struct sk_buff *skb,
426 + struct net_device *dev)
427 +{
428 + unsigned int queue_index;
429 +
430 + /* Don't let queue_mapping be left too large after exiting IMQ */
431 + if (likely(skb->dev != dev && skb->dev != NULL)) {
432 + queue_index = skb_get_queue_mapping(skb);
433 + if (unlikely(queue_index >= skb->dev->real_num_tx_queues)) {
434 + queue_index = (u16)((u32)queue_index %
435 + skb->dev->real_num_tx_queues);
436 + skb_set_queue_mapping(skb, queue_index);
437 + }
438 + } else {
439 + /* skb->dev was IMQ device itself or NULL, be on safe side and
440 + * just clear queue mapping.
441 + */
442 + skb_set_queue_mapping(skb, 0);
443 + }
444 +}
445 +
446 +static netdev_tx_t imq_dev_xmit(struct sk_buff *skb, struct net_device *dev)
447 +{
448 + struct nf_queue_entry *entry = skb->nf_queue_entry;
449 +
450 + skb->nf_queue_entry = NULL;
451 + dev->trans_start = jiffies;
452 +
453 + dev->stats.tx_bytes += skb->len;
454 + dev->stats.tx_packets++;
455 +
456 + if (unlikely(entry == NULL)) {
457 + /* We don't know what is going on here.. packet is queued for
458 + * imq device, but (probably) not by us.
459 + *
460 + * If this packet was not send here by imq_nf_queue(), then
461 + * skb_save_cb() was not used and skb_free() should not show:
462 + * WARNING: IMQ: kfree_skb: skb->cb_next:..
463 + * and/or
464 + * WARNING: IMQ: kfree_skb: skb->nf_queue_entry...
465 + *
466 + * However if this message is shown, then IMQ is somehow broken
467 + * and you should report this to linuximq.net.
468 + */
469 +
470 + /* imq_dev_xmit is black hole that eats all packets, report that
471 + * we eat this packet happily and increase dropped counters.
472 + */
473 +
474 + dev->stats.tx_dropped++;
475 + dev_kfree_skb(skb);
476 +
477 + return NETDEV_TX_OK;
478 + }
479 +
480 + skb_restore_cb(skb); /* restore skb->cb */
481 +
482 + skb->imq_flags = 0;
483 + skb->destructor = NULL;
484 +
485 + imq_done_check_queue_mapping(skb, dev);
486 +
487 + nf_reinject(entry, NF_ACCEPT);
488 +
489 + return NETDEV_TX_OK;
490 +}
491 +
492 +static struct net_device *get_imq_device_by_index(int index)
493 +{
494 + struct net_device *dev = NULL;
495 + struct net *net;
496 + char buf[8];
497 +
498 + /* get device by name and cache result */
499 + snprintf(buf, sizeof(buf), "imq%d", index);
500 +
501 + /* Search device from all namespaces. */
502 + for_each_net(net) {
503 + dev = dev_get_by_name(net, buf);
504 + if (dev)
505 + break;
506 + }
507 +
508 + if (WARN_ON_ONCE(dev == NULL)) {
509 + /* IMQ device not found. Exotic config? */
510 + return ERR_PTR(-ENODEV);
511 + }
512 +
513 + imq_devs_cache[index] = dev;
514 + dev_put(dev);
515 +
516 + return dev;
517 +}
518 +
519 +static int imq_nf_queue(struct nf_queue_entry *entry, unsigned int queue_num)
520 +{
521 + struct net_device *dev;
522 + struct sk_buff *skb_orig, *skb, *skb_shared;
523 + struct Qdisc *q;
524 + struct netdev_queue *txq;
525 + spinlock_t *root_lock;
526 + int users, index;
527 + int retval = -EINVAL;
528 + unsigned int orig_queue_index;
529 +
530 + index = entry->skb->imq_flags & IMQ_F_IFMASK;
531 + if (unlikely(index > numdevs - 1)) {
532 + if (net_ratelimit())
533 + pr_warn("IMQ: invalid device specified, highest is %u\n",
534 + numdevs - 1);
535 + retval = -EINVAL;
536 + goto out;
537 + }
538 +
539 + /* check for imq device by index from cache */
540 + dev = imq_devs_cache[index];
541 + if (unlikely(!dev)) {
542 + dev = get_imq_device_by_index(index);
543 + if (IS_ERR(dev)) {
544 + retval = PTR_ERR(dev);
545 + goto out;
546 + }
547 + }
548 +
549 + if (unlikely(!(dev->flags & IFF_UP))) {
550 + entry->skb->imq_flags = 0;
551 + nf_reinject(entry, NF_ACCEPT);
552 + retval = 0;
553 + goto out;
554 + }
555 + dev->last_rx = jiffies;
556 +
557 + skb = entry->skb;
558 + skb_orig = NULL;
559 +
560 + /* skb has owner? => make clone */
561 + if (unlikely(skb->destructor)) {
562 + skb_orig = skb;
563 + skb = skb_clone(skb, GFP_ATOMIC);
564 + if (unlikely(!skb)) {
565 + retval = -ENOMEM;
566 + goto out;
567 + }
568 + entry->skb = skb;
569 + }
570 +
571 + skb->nf_queue_entry = entry;
572 +
573 + dev->stats.rx_bytes += skb->len;
574 + dev->stats.rx_packets++;
575 +
576 + if (!skb->dev) {
577 + /* skb->dev == NULL causes problems, try the find cause. */
578 + if (net_ratelimit()) {
579 + dev_warn(&dev->dev,
580 + "received packet with skb->dev == NULL\n");
581 + dump_stack();
582 + }
583 +
584 + skb->dev = dev;
585 + }
586 +
587 + /* Disables softirqs for lock below */
588 + rcu_read_lock_bh();
589 +
590 + /* Multi-queue selection */
591 + orig_queue_index = skb_get_queue_mapping(skb);
592 + txq = imq_select_queue(dev, skb);
593 +
594 + q = rcu_dereference(txq->qdisc);
595 + if (unlikely(!q->enqueue))
596 + goto packet_not_eaten_by_imq_dev;
597 +
598 + root_lock = qdisc_lock(q);
599 + spin_lock(root_lock);
600 +
601 + users = atomic_read(&skb->users);
602 +
603 + skb_shared = skb_get(skb); /* increase reference count by one */
604 +
605 + /* backup skb->cb, as qdisc layer will overwrite it */
606 + skb_save_cb(skb_shared);
607 + qdisc_enqueue_root(skb_shared, q); /* might kfree_skb */
608 +
609 + if (likely(atomic_read(&skb_shared->users) == users + 1)) {
610 + kfree_skb(skb_shared); /* decrease reference count by one */
611 +
612 + skb->destructor = &imq_skb_destructor;
613 +
614 + /* cloned? */
615 + if (unlikely(skb_orig))
616 + kfree_skb(skb_orig); /* free original */
617 +
618 + spin_unlock(root_lock);
619 + rcu_read_unlock_bh();
620 +
621 + /* schedule qdisc dequeue */
622 + __netif_schedule(q);
623 +
624 + retval = 0;
625 + goto out;
626 + } else {
627 + skb_restore_cb(skb_shared); /* restore skb->cb */
628 + skb->nf_queue_entry = NULL;
629 + /*
630 + * qdisc dropped packet and decreased skb reference count of
631 + * skb, so we don't really want to and try refree as that would
632 + * actually destroy the skb.
633 + */
634 + spin_unlock(root_lock);
635 + goto packet_not_eaten_by_imq_dev;
636 + }
637 +
638 +packet_not_eaten_by_imq_dev:
639 + skb_set_queue_mapping(skb, orig_queue_index);
640 + rcu_read_unlock_bh();
641 +
642 + /* cloned? restore original */
643 + if (unlikely(skb_orig)) {
644 + kfree_skb(skb);
645 + entry->skb = skb_orig;
646 + }
647 + retval = -1;
648 +out:
649 + return retval;
650 +}
651 +
652 +static unsigned int imq_nf_hook(unsigned int hook, struct sk_buff *pskb,
653 + const struct net_device *indev,
654 + const struct net_device *outdev,
655 + int (*okfn)(struct sk_buff *))
656 +{
657 + return (pskb->imq_flags & IMQ_F_ENQUEUE) ? NF_IMQ_QUEUE : NF_ACCEPT;
658 +}
659 +
660 +static int imq_close(struct net_device *dev)
661 +{
662 + netif_stop_queue(dev);
663 + return 0;
664 +}
665 +
666 +static int imq_open(struct net_device *dev)
667 +{
668 + netif_start_queue(dev);
669 + return 0;
670 +}
671 +
672 +static const struct net_device_ops imq_netdev_ops = {
673 + .ndo_open = imq_open,
674 + .ndo_stop = imq_close,
675 + .ndo_start_xmit = imq_dev_xmit,
676 + .ndo_get_stats = imq_get_stats,
677 +};
678 +
679 +static void imq_setup(struct net_device *dev)
680 +{
681 + dev->netdev_ops = &imq_netdev_ops;
682 + dev->type = ARPHRD_VOID;
683 + dev->mtu = 16000; /* too small? */
684 + dev->tx_queue_len = 11000; /* too big? */
685 + dev->flags = IFF_NOARP;
686 + dev->features = NETIF_F_SG | NETIF_F_FRAGLIST |
687 + NETIF_F_GSO | NETIF_F_HW_CSUM |
688 + NETIF_F_HIGHDMA;
689 + dev->priv_flags &= ~(IFF_XMIT_DST_RELEASE |
690 + IFF_TX_SKB_SHARING);
691 +}
692 +
693 +static int imq_validate(struct nlattr *tb[], struct nlattr *data[])
694 +{
695 + int ret = 0;
696 +
697 + if (tb[IFLA_ADDRESS]) {
698 + if (nla_len(tb[IFLA_ADDRESS]) != ETH_ALEN) {
699 + ret = -EINVAL;
700 + goto end;
701 + }
702 + if (!is_valid_ether_addr(nla_data(tb[IFLA_ADDRESS]))) {
703 + ret = -EADDRNOTAVAIL;
704 + goto end;
705 + }
706 + }
707 + return 0;
708 +end:
709 + pr_warn("IMQ: imq_validate failed (%d)\n", ret);
710 + return ret;
711 +}
712 +
713 +static struct rtnl_link_ops imq_link_ops __read_mostly = {
714 + .kind = "imq",
715 + .priv_size = 0,
716 + .setup = imq_setup,
717 + .validate = imq_validate,
718 +};
719 +
720 +static const struct nf_queue_handler imq_nfqh = {
721 + .outfn = imq_nf_queue,
722 +};
723 +
724 +static int __init imq_init_hooks(void)
725 +{
726 + int ret;
727 +
728 + nf_register_queue_imq_handler(&imq_nfqh);
729 +
730 + ret = nf_register_hooks(imq_ops, ARRAY_SIZE(imq_ops));
731 + if (ret < 0)
732 + nf_unregister_queue_imq_handler();
733 +
734 + return ret;
735 +}
736 +
737 +static int __init imq_init_one(int index)
738 +{
739 + struct net_device *dev;
740 + int ret;
741 +
742 + dev = alloc_netdev_mq(0, "imq%d", imq_setup, numqueues);
743 + if (!dev)
744 + return -ENOMEM;
745 +
746 + ret = dev_alloc_name(dev, dev->name);
747 + if (ret < 0)
748 + goto fail;
749 +
750 + dev->rtnl_link_ops = &imq_link_ops;
751 + ret = register_netdevice(dev);
752 + if (ret < 0)
753 + goto fail;
754 +
755 + return 0;
756 +fail:
757 + free_netdev(dev);
758 + return ret;
759 +}
760 +
761 +static int __init imq_init_devs(void)
762 +{
763 + int err, i;
764 +
765 + if (numdevs < 1 || numdevs > IMQ_MAX_DEVS) {
766 + pr_err("IMQ: numdevs has to be betweed 1 and %u\n",
767 + IMQ_MAX_DEVS);
768 + return -EINVAL;
769 + }
770 +
771 + if (numqueues < 1 || numqueues > IMQ_MAX_QUEUES) {
772 + pr_err("IMQ: numqueues has to be betweed 1 and %u\n",
773 + IMQ_MAX_QUEUES);
774 + return -EINVAL;
775 + }
776 +
777 + get_random_bytes(&imq_hashrnd, sizeof(imq_hashrnd));
778 +
779 + rtnl_lock();
780 + err = __rtnl_link_register(&imq_link_ops);
781 +
782 + for (i = 0; i < numdevs && !err; i++)
783 + err = imq_init_one(i);
784 +
785 + if (err) {
786 + __rtnl_link_unregister(&imq_link_ops);
787 + memset(imq_devs_cache, 0, sizeof(imq_devs_cache));
788 + }
789 + rtnl_unlock();
790 +
791 + return err;
792 +}
793 +
794 +static int __init imq_init_module(void)
795 +{
796 + int err;
797 +
798 +#if defined(CONFIG_IMQ_NUM_DEVS)
799 + BUILD_BUG_ON(CONFIG_IMQ_NUM_DEVS > 16);
800 + BUILD_BUG_ON(CONFIG_IMQ_NUM_DEVS < 2);
801 + BUILD_BUG_ON(CONFIG_IMQ_NUM_DEVS - 1 > IMQ_F_IFMASK);
802 +#endif
803 +
804 + err = imq_init_devs();
805 + if (err) {
806 + pr_err("IMQ: Error trying imq_init_devs(net)\n");
807 + return err;
808 + }
809 +
810 + err = imq_init_hooks();
811 + if (err) {
812 + pr_err(KERN_ERR "IMQ: Error trying imq_init_hooks()\n");
813 + rtnl_link_unregister(&imq_link_ops);
814 + memset(imq_devs_cache, 0, sizeof(imq_devs_cache));
815 + return err;
816 + }
817 +
818 + pr_info("IMQ driver loaded successfully. (numdevs = %d, numqueues = %d)\n",
819 + numdevs, numqueues);
820 +
821 +#if defined(CONFIG_IMQ_BEHAVIOR_BA) || defined(CONFIG_IMQ_BEHAVIOR_BB)
822 + pr_info("\tHooking IMQ before NAT on PREROUTING.\n");
823 +#else
824 + pr_info("\tHooking IMQ after NAT on PREROUTING.\n");
825 +#endif
826 +#if defined(CONFIG_IMQ_BEHAVIOR_AB) || defined(CONFIG_IMQ_BEHAVIOR_BB)
827 + pr_info("\tHooking IMQ before NAT on POSTROUTING.\n");
828 +#else
829 + pr_info("\tHooking IMQ after NAT on POSTROUTING.\n");
830 +#endif
831 +
832 + return 0;
833 +}
834 +
835 +static void __exit imq_unhook(void)
836 +{
837 + nf_unregister_hooks(imq_ops, ARRAY_SIZE(imq_ops));
838 + nf_unregister_queue_imq_handler();
839 +}
840 +
841 +static void __exit imq_cleanup_devs(void)
842 +{
843 + rtnl_link_unregister(&imq_link_ops);
844 + memset(imq_devs_cache, 0, sizeof(imq_devs_cache));
845 +}
846 +
847 +static void __exit imq_exit_module(void)
848 +{
849 + imq_unhook();
850 + imq_cleanup_devs();
851 + pr_info("IMQ driver unloaded successfully.\n");
852 +}
853 +
854 +module_init(imq_init_module);
855 +module_exit(imq_exit_module);
856 +
857 +module_param(numdevs, int, 0);
858 +module_param(numqueues, int, 0);
859 +MODULE_PARM_DESC(numdevs, "number of IMQ devices (how many imq* devices will be created)");
860 +MODULE_PARM_DESC(numqueues, "number of queues per IMQ device");
861 +MODULE_AUTHOR("http://www.linuximq.net");
862 +MODULE_DESCRIPTION("Pseudo-driver for the intermediate queue device. See http://www.linuximq.net/ for more information.");
863 +MODULE_LICENSE("GPL");
864 +MODULE_ALIAS_RTNL_LINK("imq");
865 +
866 diff -uNr linux-3.9.1/drivers/net/Kconfig linux-3.9.1-imqmq/drivers/net/Kconfig
867 --- linux-3.9.1/drivers/net/Kconfig 2013-05-08 06:58:03.000000000 +0300
868 +++ linux-3.9.1-imqmq/drivers/net/Kconfig 2013-05-08 17:30:29.011952562 +0300
869 @@ -206,6 +206,125 @@
870 depends on RIONET
871 default "128"
872
873 +config IMQ
874 + tristate "IMQ (intermediate queueing device) support"
875 + depends on NETDEVICES && NETFILTER
876 + ---help---
877 + The IMQ device(s) is used as placeholder for QoS queueing
878 + disciplines. Every packet entering/leaving the IP stack can be
879 + directed through the IMQ device where it's enqueued/dequeued to the
880 + attached qdisc. This allows you to treat network devices as classes
881 + and distribute bandwidth among them. Iptables is used to specify
882 + through which IMQ device, if any, packets travel.
883 +
884 + More information at: http://www.linuximq.net/
885 +
886 + To compile this driver as a module, choose M here: the module
887 + will be called imq. If unsure, say N.
888 +
889 +choice
890 + prompt "IMQ behavior (PRE/POSTROUTING)"
891 + depends on IMQ
892 + default IMQ_BEHAVIOR_AB
893 + help
894 + This setting defines how IMQ behaves in respect to its
895 + hooking in PREROUTING and POSTROUTING.
896 +
897 + IMQ can work in any of the following ways:
898 +
899 + PREROUTING | POSTROUTING
900 + -----------------|-------------------
901 + #1 After NAT | After NAT
902 + #2 After NAT | Before NAT
903 + #3 Before NAT | After NAT
904 + #4 Before NAT | Before NAT
905 +
906 + The default behavior is to hook before NAT on PREROUTING
907 + and after NAT on POSTROUTING (#3).
908 +
909 + This settings are specially usefull when trying to use IMQ
910 + to shape NATed clients.
911 +
912 + More information can be found at: www.linuximq.net
913 +
914 + If not sure leave the default settings alone.
915 +
916 +config IMQ_BEHAVIOR_AA
917 + bool "IMQ AA"
918 + help
919 + This setting defines how IMQ behaves in respect to its
920 + hooking in PREROUTING and POSTROUTING.
921 +
922 + Choosing this option will make IMQ hook like this:
923 +
924 + PREROUTING: After NAT
925 + POSTROUTING: After NAT
926 +
927 + More information can be found at: www.linuximq.net
928 +
929 + If not sure leave the default settings alone.
930 +
931 +config IMQ_BEHAVIOR_AB
932 + bool "IMQ AB"
933 + help
934 + This setting defines how IMQ behaves in respect to its
935 + hooking in PREROUTING and POSTROUTING.
936 +
937 + Choosing this option will make IMQ hook like this:
938 +
939 + PREROUTING: After NAT
940 + POSTROUTING: Before NAT
941 +
942 + More information can be found at: www.linuximq.net
943 +
944 + If not sure leave the default settings alone.
945 +
946 +config IMQ_BEHAVIOR_BA
947 + bool "IMQ BA"
948 + help
949 + This setting defines how IMQ behaves in respect to its
950 + hooking in PREROUTING and POSTROUTING.
951 +
952 + Choosing this option will make IMQ hook like this:
953 +
954 + PREROUTING: Before NAT
955 + POSTROUTING: After NAT
956 +
957 + More information can be found at: www.linuximq.net
958 +
959 + If not sure leave the default settings alone.
960 +
961 +config IMQ_BEHAVIOR_BB
962 + bool "IMQ BB"
963 + help
964 + This setting defines how IMQ behaves in respect to its
965 + hooking in PREROUTING and POSTROUTING.
966 +
967 + Choosing this option will make IMQ hook like this:
968 +
969 + PREROUTING: Before NAT
970 + POSTROUTING: Before NAT
971 +
972 + More information can be found at: www.linuximq.net
973 +
974 + If not sure leave the default settings alone.
975 +
976 +endchoice
977 +
978 +config IMQ_NUM_DEVS
979 + int "Number of IMQ devices"
980 + range 2 16
981 + depends on IMQ
982 + default "16"
983 + help
984 + This setting defines how many IMQ devices will be created.
985 +
986 + The default value is 16.
987 +
988 + More information can be found at: www.linuximq.net
989 +
990 + If not sure leave the default settings alone.
991 +
992 config TUN
993 tristate "Universal TUN/TAP device driver support"
994 select CRC32
995 diff -uNr linux-3.9.1/drivers/net/Makefile linux-3.9.1-imqmq/drivers/net/Makefile
996 --- linux-3.9.1/drivers/net/Makefile 2013-05-08 06:58:03.000000000 +0300
997 +++ linux-3.9.1-imqmq/drivers/net/Makefile 2013-05-08 17:30:29.011952562 +0300
998 @@ -9,6 +9,7 @@
999 obj-$(CONFIG_DUMMY) += dummy.o
1000 obj-$(CONFIG_EQUALIZER) += eql.o
1001 obj-$(CONFIG_IFB) += ifb.o
1002 +obj-$(CONFIG_IMQ) += imq.o
1003 obj-$(CONFIG_MACVLAN) += macvlan.o
1004 obj-$(CONFIG_MACVTAP) += macvtap.o
1005 obj-$(CONFIG_MII) += mii.o
1006 diff -uNr linux-3.9.1/include/linux/imq.h linux-3.9.1-imqmq/include/linux/imq.h
1007 --- linux-3.9.1/include/linux/imq.h 1970-01-01 02:00:00.000000000 +0200
1008 +++ linux-3.9.1-imqmq/include/linux/imq.h 2013-05-08 17:30:29.011952562 +0300
1009 @@ -0,0 +1,13 @@
1010 +#ifndef _IMQ_H
1011 +#define _IMQ_H
1012 +
1013 +/* IFMASK (16 device indexes, 0 to 15) and flag(s) fit in 5 bits */
1014 +#define IMQ_F_BITS 5
1015 +
1016 +#define IMQ_F_IFMASK 0x0f
1017 +#define IMQ_F_ENQUEUE 0x10
1018 +
1019 +#define IMQ_MAX_DEVS (IMQ_F_IFMASK + 1)
1020 +
1021 +#endif /* _IMQ_H */
1022 +
1023 diff -uNr linux-3.9.1/include/linux/netfilter/xt_IMQ.h linux-3.9.1-imqmq/include/linux/netfilter/xt_IMQ.h
1024 --- linux-3.9.1/include/linux/netfilter/xt_IMQ.h 1970-01-01 02:00:00.000000000 +0200
1025 +++ linux-3.9.1-imqmq/include/linux/netfilter/xt_IMQ.h 2013-05-08 17:30:29.011952562 +0300
1026 @@ -0,0 +1,9 @@
1027 +#ifndef _XT_IMQ_H
1028 +#define _XT_IMQ_H
1029 +
1030 +struct xt_imq_info {
1031 + unsigned int todev; /* target imq device */
1032 +};
1033 +
1034 +#endif /* _XT_IMQ_H */
1035 +
1036 diff -uNr linux-3.9.1/include/linux/netfilter_ipv4/ipt_IMQ.h linux-3.9.1-imqmq/include/linux/netfilter_ipv4/ipt_IMQ.h
1037 --- linux-3.9.1/include/linux/netfilter_ipv4/ipt_IMQ.h 1970-01-01 02:00:00.000000000 +0200
1038 +++ linux-3.9.1-imqmq/include/linux/netfilter_ipv4/ipt_IMQ.h 2013-05-08 17:30:29.011952562 +0300
1039 @@ -0,0 +1,10 @@
1040 +#ifndef _IPT_IMQ_H
1041 +#define _IPT_IMQ_H
1042 +
1043 +/* Backwards compatibility for old userspace */
1044 +#include <linux/netfilter/xt_IMQ.h>
1045 +
1046 +#define ipt_imq_info xt_imq_info
1047 +
1048 +#endif /* _IPT_IMQ_H */
1049 +
1050 diff -uNr linux-3.9.1/include/linux/netfilter_ipv6/ip6t_IMQ.h linux-3.9.1-imqmq/include/linux/netfilter_ipv6/ip6t_IMQ.h
1051 --- linux-3.9.1/include/linux/netfilter_ipv6/ip6t_IMQ.h 1970-01-01 02:00:00.000000000 +0200
1052 +++ linux-3.9.1-imqmq/include/linux/netfilter_ipv6/ip6t_IMQ.h 2013-05-08 17:30:29.011952562 +0300
1053 @@ -0,0 +1,10 @@
1054 +#ifndef _IP6T_IMQ_H
1055 +#define _IP6T_IMQ_H
1056 +
1057 +/* Backwards compatibility for old userspace */
1058 +#include <linux/netfilter/xt_IMQ.h>
1059 +
1060 +#define ip6t_imq_info xt_imq_info
1061 +
1062 +#endif /* _IP6T_IMQ_H */
1063 +
1064 diff -uNr linux-3.9.1/include/linux/skbuff.h linux-3.9.1-imqmq/include/linux/skbuff.h
1065 --- linux-3.9.1/include/linux/skbuff.h 2013-05-08 06:58:03.000000000 +0300
1066 +++ linux-3.9.1-imqmq/include/linux/skbuff.h 2013-05-08 17:30:29.015285965 +0300
1067 @@ -32,6 +32,9 @@
1068 #include <linux/dma-mapping.h>
1069 #include <linux/netdev_features.h>
1070 #include <net/flow_keys.h>
1071 +#if defined(CONFIG_IMQ) || defined(CONFIG_IMQ_MODULE)
1072 +#include <linux/imq.h>
1073 +#endif
1074
1075 /* Don't change this without changing skb_csum_unnecessary! */
1076 #define CHECKSUM_NONE 0
1077 @@ -415,6 +418,9 @@
1078 * first. This is owned by whoever has the skb queued ATM.
1079 */
1080 char cb[48] __aligned(8);
1081 +#if defined(CONFIG_IMQ) || defined(CONFIG_IMQ_MODULE)
1082 + void *cb_next;
1083 +#endif
1084
1085 unsigned long _skb_refdst;
1086 #ifdef CONFIG_XFRM
1087 @@ -453,6 +459,9 @@
1088 #ifdef NET_SKBUFF_NF_DEFRAG_NEEDED
1089 struct sk_buff *nfct_reasm;
1090 #endif
1091 +#if defined(CONFIG_IMQ) || defined(CONFIG_IMQ_MODULE)
1092 + struct nf_queue_entry *nf_queue_entry;
1093 +#endif
1094 #ifdef CONFIG_BRIDGE_NETFILTER
1095 struct nf_bridge_info *nf_bridge;
1096 #endif
1097 @@ -491,6 +500,10 @@
1098 /* 7/9 bit hole (depending on ndisc_nodetype presence) */
1099 kmemcheck_bitfield_end(flags2);
1100
1101 +#if defined(CONFIG_IMQ) || defined(CONFIG_IMQ_MODULE)
1102 + __u8 imq_flags:IMQ_F_BITS;
1103 +#endif
1104 +
1105 #ifdef CONFIG_NET_DMA
1106 dma_cookie_t dma_cookie;
1107 #endif
1108 @@ -586,6 +599,12 @@
1109 return (struct rtable *)skb_dst(skb);
1110 }
1111
1112 +
1113 +#if defined(CONFIG_IMQ) || defined(CONFIG_IMQ_MODULE)
1114 +extern int skb_save_cb(struct sk_buff *skb);
1115 +extern int skb_restore_cb(struct sk_buff *skb);
1116 +#endif
1117 +
1118 extern void kfree_skb(struct sk_buff *skb);
1119 extern void skb_tx_error(struct sk_buff *skb);
1120 extern void consume_skb(struct sk_buff *skb);
1121 @@ -2662,6 +2681,10 @@
1122 dst->nfct_reasm = src->nfct_reasm;
1123 nf_conntrack_get_reasm(src->nfct_reasm);
1124 #endif
1125 +#if defined(CONFIG_IMQ) || defined(CONFIG_IMQ_MODULE)
1126 + dst->imq_flags = src->imq_flags;
1127 + dst->nf_queue_entry = src->nf_queue_entry;
1128 +#endif
1129 #ifdef CONFIG_BRIDGE_NETFILTER
1130 dst->nf_bridge = src->nf_bridge;
1131 nf_bridge_get(src->nf_bridge);
1132 diff -uNr linux-3.9.1/include/net/netfilter/nf_queue.h linux-3.9.1-imqmq/include/net/netfilter/nf_queue.h
1133 --- linux-3.9.1/include/net/netfilter/nf_queue.h 2013-05-08 06:58:03.000000000 +0300
1134 +++ linux-3.9.1-imqmq/include/net/netfilter/nf_queue.h 2013-05-08 17:30:29.015285965 +0300
1135 @@ -26,5 +26,11 @@
1136 void nf_register_queue_handler(const struct nf_queue_handler *qh);
1137 void nf_unregister_queue_handler(void);
1138 extern void nf_reinject(struct nf_queue_entry *entry, unsigned int verdict);
1139 +extern void nf_queue_entry_release_refs(struct nf_queue_entry *entry);
1140 +
1141 +#if defined(CONFIG_IMQ) || defined(CONFIG_IMQ_MODULE)
1142 +extern void nf_register_queue_imq_handler(const struct nf_queue_handler *qh);
1143 +extern void nf_unregister_queue_imq_handler(void);
1144 +#endif
1145
1146 #endif /* _NF_QUEUE_H */
1147 diff -uNr linux-3.9.1/include/uapi/linux/netfilter.h linux-3.9.1-imqmq/include/uapi/linux/netfilter.h
1148 --- linux-3.9.1/include/uapi/linux/netfilter.h 2013-05-08 06:58:03.000000000 +0300
1149 +++ linux-3.9.1-imqmq/include/uapi/linux/netfilter.h 2013-05-08 17:30:29.015285965 +0300
1150 @@ -13,7 +13,8 @@
1151 #define NF_QUEUE 3
1152 #define NF_REPEAT 4
1153 #define NF_STOP 5
1154 -#define NF_MAX_VERDICT NF_STOP
1155 +#define NF_IMQ_QUEUE 6
1156 +#define NF_MAX_VERDICT NF_IMQ_QUEUE
1157
1158 /* we overload the higher bits for encoding auxiliary data such as the queue
1159 * number or errno values. Not nice, but better than additional function
1160 diff -uNr linux-3.9.1/net/core/dev.c linux-3.9.1-imqmq/net/core/dev.c
1161 --- linux-3.9.1/net/core/dev.c 2013-05-08 06:58:03.000000000 +0300
1162 +++ linux-3.9.1-imqmq/net/core/dev.c 2013-05-08 17:30:29.018619368 +0300
1163 @@ -129,6 +129,9 @@
1164 #include <linux/inetdevice.h>
1165 #include <linux/cpu_rmap.h>
1166 #include <linux/static_key.h>
1167 +#if defined(CONFIG_IMQ) || defined(CONFIG_IMQ_MODULE)
1168 +#include <linux/imq.h>
1169 +#endif
1170
1171 #include "net-sysfs.h"
1172
1173 @@ -2529,7 +2532,12 @@
1174 }
1175 }
1176
1177 +#if defined(CONFIG_IMQ) || defined(CONFIG_IMQ_MODULE)
1178 + if (!list_empty(&ptype_all) &&
1179 + !(skb->imq_flags & IMQ_F_ENQUEUE))
1180 +#else
1181 if (!list_empty(&ptype_all))
1182 +#endif
1183 dev_queue_xmit_nit(skb, dev);
1184
1185 skb_len = skb->len;
1186 diff -uNr linux-3.9.1/net/core/skbuff.c linux-3.9.1-imqmq/net/core/skbuff.c
1187 --- linux-3.9.1/net/core/skbuff.c 2013-05-08 06:58:03.000000000 +0300
1188 +++ linux-3.9.1-imqmq/net/core/skbuff.c 2013-05-08 17:30:29.021952772 +0300
1189 @@ -73,6 +73,9 @@
1190
1191 struct kmem_cache *skbuff_head_cache __read_mostly;
1192 static struct kmem_cache *skbuff_fclone_cache __read_mostly;
1193 +#if defined(CONFIG_IMQ) || defined(CONFIG_IMQ_MODULE)
1194 +static struct kmem_cache *skbuff_cb_store_cache __read_mostly;
1195 +#endif
1196
1197 static void sock_pipe_buf_release(struct pipe_inode_info *pipe,
1198 struct pipe_buffer *buf)
1199 @@ -92,6 +95,82 @@
1200 return 1;
1201 }
1202
1203 +#if defined(CONFIG_IMQ) || defined(CONFIG_IMQ_MODULE)
1204 +/* Control buffer save/restore for IMQ devices */
1205 +struct skb_cb_table {
1206 + char cb[48] __aligned(8);
1207 + void *cb_next;
1208 + atomic_t refcnt;
1209 +};
1210 +
1211 +static DEFINE_SPINLOCK(skb_cb_store_lock);
1212 +
1213 +int skb_save_cb(struct sk_buff *skb)
1214 +{
1215 + struct skb_cb_table *next;
1216 +
1217 + next = kmem_cache_alloc(skbuff_cb_store_cache, GFP_ATOMIC);
1218 + if (!next)
1219 + return -ENOMEM;
1220 +
1221 + BUILD_BUG_ON(sizeof(skb->cb) != sizeof(next->cb));
1222 +
1223 + memcpy(next->cb, skb->cb, sizeof(skb->cb));
1224 + next->cb_next = skb->cb_next;
1225 +
1226 + atomic_set(&next->refcnt, 1);
1227 +
1228 + skb->cb_next = next;
1229 + return 0;
1230 +}
1231 +EXPORT_SYMBOL(skb_save_cb);
1232 +
1233 +int skb_restore_cb(struct sk_buff *skb)
1234 +{
1235 + struct skb_cb_table *next;
1236 +
1237 + if (!skb->cb_next)
1238 + return 0;
1239 +
1240 + next = skb->cb_next;
1241 +
1242 + BUILD_BUG_ON(sizeof(skb->cb) != sizeof(next->cb));
1243 +
1244 + memcpy(skb->cb, next->cb, sizeof(skb->cb));
1245 + skb->cb_next = next->cb_next;
1246 +
1247 + spin_lock(&skb_cb_store_lock);
1248 +
1249 + if (atomic_dec_and_test(&next->refcnt))
1250 + kmem_cache_free(skbuff_cb_store_cache, next);
1251 +
1252 + spin_unlock(&skb_cb_store_lock);
1253 +
1254 + return 0;
1255 +}
1256 +EXPORT_SYMBOL(skb_restore_cb);
1257 +
1258 +static void skb_copy_stored_cb(struct sk_buff *new, const struct sk_buff *__old)
1259 +{
1260 + struct skb_cb_table *next;
1261 + struct sk_buff *old;
1262 +
1263 + if (!__old->cb_next) {
1264 + new->cb_next = NULL;
1265 + return;
1266 + }
1267 +
1268 + spin_lock(&skb_cb_store_lock);
1269 +
1270 + old = (struct sk_buff *)__old;
1271 +
1272 + next = old->cb_next;
1273 + atomic_inc(&next->refcnt);
1274 + new->cb_next = next;
1275 +
1276 + spin_unlock(&skb_cb_store_lock);
1277 +}
1278 +#endif
1279
1280 /* Pipe buffer operations for a socket. */
1281 static const struct pipe_buf_operations sock_pipe_buf_ops = {
1282 @@ -562,6 +641,28 @@
1283 WARN_ON(in_irq());
1284 skb->destructor(skb);
1285 }
1286 +#if defined(CONFIG_IMQ) || defined(CONFIG_IMQ_MODULE)
1287 + /*
1288 + * This should not happen. When it does, avoid memleak by restoring
1289 + * the chain of cb-backups.
1290 + */
1291 + while (skb->cb_next != NULL) {
1292 + if (net_ratelimit())
1293 + pr_warn("IMQ: kfree_skb: skb->cb_next: %08x\n",
1294 + (unsigned int)skb->cb_next);
1295 +
1296 + skb_restore_cb(skb);
1297 + }
1298 + /*
1299 + * This should not happen either, nf_queue_entry is nullified in
1300 + * imq_dev_xmit(). If we have non-NULL nf_queue_entry then we are
1301 + * leaking entry pointers, maybe memory. We don't know if this is
1302 + * pointer to already freed memory, or should this be freed.
1303 + * If this happens we need to add refcounting, etc for nf_queue_entry.
1304 + */
1305 + if (skb->nf_queue_entry && net_ratelimit())
1306 + pr_warn("%s\n", "IMQ: kfree_skb: skb->nf_queue_entry != NULL");
1307 +#endif
1308 #if IS_ENABLED(CONFIG_NF_CONNTRACK)
1309 nf_conntrack_put(skb->nfct);
1310 #endif
1311 @@ -683,6 +784,9 @@
1312 new->sp = secpath_get(old->sp);
1313 #endif
1314 memcpy(new->cb, old->cb, sizeof(old->cb));
1315 +#if defined(CONFIG_IMQ) || defined(CONFIG_IMQ_MODULE)
1316 + skb_copy_stored_cb(new, old);
1317 +#endif
1318 new->csum = old->csum;
1319 new->local_df = old->local_df;
1320 new->pkt_type = old->pkt_type;
1321 @@ -3053,6 +3157,13 @@
1322 0,
1323 SLAB_HWCACHE_ALIGN|SLAB_PANIC,
1324 NULL);
1325 +#if defined(CONFIG_IMQ) || defined(CONFIG_IMQ_MODULE)
1326 + skbuff_cb_store_cache = kmem_cache_create("skbuff_cb_store_cache",
1327 + sizeof(struct skb_cb_table),
1328 + 0,
1329 + SLAB_HWCACHE_ALIGN|SLAB_PANIC,
1330 + NULL);
1331 +#endif
1332 }
1333
1334 /**
1335 diff -uNr linux-3.9.1/net/ipv6/ip6_output.c linux-3.9.1-imqmq/net/ipv6/ip6_output.c
1336 --- linux-3.9.1/net/ipv6/ip6_output.c 2013-05-08 06:58:03.000000000 +0300
1337 +++ linux-3.9.1-imqmq/net/ipv6/ip6_output.c 2013-05-08 17:30:29.021952772 +0300
1338 @@ -89,9 +89,6 @@
1339 struct in6_addr *nexthop;
1340 int ret;
1341
1342 - skb->protocol = htons(ETH_P_IPV6);
1343 - skb->dev = dev;
1344 -
1345 if (ipv6_addr_is_multicast(&ipv6_hdr(skb)->daddr)) {
1346 struct inet6_dev *idev = ip6_dst_idev(skb_dst(skb));
1347
1348 @@ -167,6 +164,13 @@
1349 return 0;
1350 }
1351
1352 + /*
1353 + * IMQ-patch: moved setting skb->dev and skb->protocol from
1354 + * ip6_finish_output2 to fix crashing at netif_skb_features().
1355 + */
1356 + skb->protocol = htons(ETH_P_IPV6);
1357 + skb->dev = dev;
1358 +
1359 return NF_HOOK_COND(NFPROTO_IPV6, NF_INET_POST_ROUTING, skb, NULL, dev,
1360 ip6_finish_output,
1361 !(IP6CB(skb)->flags & IP6SKB_REROUTED));
1362 diff -uNr linux-3.9.1/net/netfilter/core.c linux-3.9.1-imqmq/net/netfilter/core.c
1363 --- linux-3.9.1/net/netfilter/core.c 2013-05-08 06:58:03.000000000 +0300
1364 +++ linux-3.9.1-imqmq/net/netfilter/core.c 2013-05-08 17:30:29.025286174 +0300
1365 @@ -188,9 +188,11 @@
1366 ret = NF_DROP_GETERR(verdict);
1367 if (ret == 0)
1368 ret = -EPERM;
1369 - } else if ((verdict & NF_VERDICT_MASK) == NF_QUEUE) {
1370 + } else if ((verdict & NF_VERDICT_MASK) == NF_QUEUE ||
1371 + (verdict & NF_VERDICT_MASK) == NF_IMQ_QUEUE) {
1372 int err = nf_queue(skb, elem, pf, hook, indev, outdev, okfn,
1373 - verdict >> NF_VERDICT_QBITS);
1374 + verdict >> NF_VERDICT_QBITS,
1375 + verdict & NF_VERDICT_MASK);
1376 if (err < 0) {
1377 if (err == -ECANCELED)
1378 goto next_hook;
1379 diff -uNr linux-3.9.1/net/netfilter/Kconfig linux-3.9.1-imqmq/net/netfilter/Kconfig
1380 --- linux-3.9.1/net/netfilter/Kconfig 2013-05-08 06:58:03.000000000 +0300
1381 +++ linux-3.9.1-imqmq/net/netfilter/Kconfig 2013-05-08 17:30:29.025286174 +0300
1382 @@ -641,6 +641,18 @@
1383
1384 To compile it as a module, choose M here. If unsure, say N.
1385
1386 +config NETFILTER_XT_TARGET_IMQ
1387 + tristate '"IMQ" target support'
1388 + depends on NETFILTER_XTABLES
1389 + depends on IP_NF_MANGLE || IP6_NF_MANGLE
1390 + select IMQ
1391 + default m if NETFILTER_ADVANCED=n
1392 + help
1393 + This option adds a `IMQ' target which is used to specify if and
1394 + to which imq device packets should get enqueued/dequeued.
1395 +
1396 + To compile it as a module, choose M here. If unsure, say N.
1397 +
1398 config NETFILTER_XT_TARGET_MARK
1399 tristate '"MARK" target support'
1400 depends on NETFILTER_ADVANCED
1401 diff -uNr linux-3.9.1/net/netfilter/Makefile linux-3.9.1-imqmq/net/netfilter/Makefile
1402 --- linux-3.9.1/net/netfilter/Makefile 2013-05-08 06:58:03.000000000 +0300
1403 +++ linux-3.9.1-imqmq/net/netfilter/Makefile 2013-05-08 17:30:29.025286174 +0300
1404 @@ -82,6 +82,7 @@
1405 obj-$(CONFIG_NETFILTER_XT_TARGET_DSCP) += xt_DSCP.o
1406 obj-$(CONFIG_NETFILTER_XT_TARGET_HL) += xt_HL.o
1407 obj-$(CONFIG_NETFILTER_XT_TARGET_HMARK) += xt_HMARK.o
1408 +obj-$(CONFIG_NETFILTER_XT_TARGET_IMQ) += xt_IMQ.o
1409 obj-$(CONFIG_NETFILTER_XT_TARGET_LED) += xt_LED.o
1410 obj-$(CONFIG_NETFILTER_XT_TARGET_LOG) += xt_LOG.o
1411 obj-$(CONFIG_NETFILTER_XT_TARGET_NETMAP) += xt_NETMAP.o
1412 diff -uNr linux-3.9.1/net/netfilter/nf_internals.h linux-3.9.1-imqmq/net/netfilter/nf_internals.h
1413 --- linux-3.9.1/net/netfilter/nf_internals.h 2013-05-08 06:58:03.000000000 +0300
1414 +++ linux-3.9.1-imqmq/net/netfilter/nf_internals.h 2013-05-08 17:30:29.025286174 +0300
1415 @@ -29,7 +29,7 @@
1416 struct net_device *indev,
1417 struct net_device *outdev,
1418 int (*okfn)(struct sk_buff *),
1419 - unsigned int queuenum);
1420 + unsigned int queuenum, unsigned int queuetype);
1421 extern int __init netfilter_queue_init(void);
1422
1423 /* nf_log.c */
1424 diff -uNr linux-3.9.1/net/netfilter/nf_queue.c linux-3.9.1-imqmq/net/netfilter/nf_queue.c
1425 --- linux-3.9.1/net/netfilter/nf_queue.c 2013-05-08 06:58:03.000000000 +0300
1426 +++ linux-3.9.1-imqmq/net/netfilter/nf_queue.c 2013-05-08 17:30:29.025286174 +0300
1427 @@ -22,6 +22,23 @@
1428 */
1429 static const struct nf_queue_handler __rcu *queue_handler __read_mostly;
1430
1431 +#if defined(CONFIG_IMQ) || defined(CONFIG_IMQ_MODULE)
1432 +static const struct nf_queue_handler __rcu *queue_imq_handler __read_mostly;
1433 +
1434 +void nf_register_queue_imq_handler(const struct nf_queue_handler *qh)
1435 +{
1436 + rcu_assign_pointer(queue_imq_handler, qh);
1437 +}
1438 +EXPORT_SYMBOL_GPL(nf_register_queue_imq_handler);
1439 +
1440 +void nf_unregister_queue_imq_handler(void)
1441 +{
1442 + RCU_INIT_POINTER(queue_imq_handler, NULL);
1443 + synchronize_rcu();
1444 +}
1445 +EXPORT_SYMBOL_GPL(nf_unregister_queue_imq_handler);
1446 +#endif
1447 +
1448 /* return EBUSY when somebody else is registered, return EEXIST if the
1449 * same handler is registered, return 0 in case of success. */
1450 void nf_register_queue_handler(const struct nf_queue_handler *qh)
1451 @@ -71,7 +89,8 @@
1452 struct net_device *indev,
1453 struct net_device *outdev,
1454 int (*okfn)(struct sk_buff *),
1455 - unsigned int queuenum)
1456 + unsigned int queuenum,
1457 + unsigned int queuetype)
1458 {
1459 int status = -ENOENT;
1460 struct nf_queue_entry *entry = NULL;
1461 @@ -85,7 +104,17 @@
1462 /* QUEUE == DROP if no one is waiting, to be safe. */
1463 rcu_read_lock();
1464
1465 - qh = rcu_dereference(queue_handler);
1466 + if (queuetype == NF_IMQ_QUEUE) {
1467 +#if defined(CONFIG_IMQ) || defined(CONFIG_IMQ_MODULE)
1468 + qh = rcu_dereference(queue_imq_handler);
1469 +#else
1470 + BUG();
1471 + goto err_unlock;
1472 +#endif
1473 + } else {
1474 + qh = rcu_dereference(queue_handler);
1475 + }
1476 +
1477 if (!qh) {
1478 status = -ESRCH;
1479 goto err_unlock;
1480 @@ -233,9 +261,11 @@
1481 local_bh_enable();
1482 break;
1483 case NF_QUEUE:
1484 + case NF_IMQ_QUEUE:
1485 err = nf_queue(skb, elem, entry->pf, entry->hook,
1486 entry->indev, entry->outdev, entry->okfn,
1487 - verdict >> NF_VERDICT_QBITS);
1488 + verdict >> NF_VERDICT_QBITS,
1489 + verdict & NF_VERDICT_MASK);
1490 if (err < 0) {
1491 if (err == -ECANCELED)
1492 goto next_hook;
1493 diff -uNr linux-3.9.1/net/netfilter/xt_IMQ.c linux-3.9.1-imqmq/net/netfilter/xt_IMQ.c
1494 --- linux-3.9.1/net/netfilter/xt_IMQ.c 1970-01-01 02:00:00.000000000 +0200
1495 +++ linux-3.9.1-imqmq/net/netfilter/xt_IMQ.c 2013-05-08 17:30:29.025286174 +0300
1496 @@ -0,0 +1,72 @@
1497 +/*
1498 + * This target marks packets to be enqueued to an imq device
1499 + */
1500 +#include <linux/module.h>
1501 +#include <linux/skbuff.h>
1502 +#include <linux/netfilter/x_tables.h>
1503 +#include <linux/netfilter/xt_IMQ.h>
1504 +#include <linux/imq.h>
1505 +
1506 +static unsigned int imq_target(struct sk_buff *pskb,
1507 + const struct xt_action_param *par)
1508 +{
1509 + const struct xt_imq_info *mr = par->targinfo;
1510 +
1511 + pskb->imq_flags = (mr->todev & IMQ_F_IFMASK) | IMQ_F_ENQUEUE;
1512 +
1513 + return XT_CONTINUE;
1514 +}
1515 +
1516 +static int imq_checkentry(const struct xt_tgchk_param *par)
1517 +{
1518 + struct xt_imq_info *mr = par->targinfo;
1519 +
1520 + if (mr->todev > IMQ_MAX_DEVS - 1) {
1521 + pr_warn("IMQ: invalid device specified, highest is %u\n",
1522 + IMQ_MAX_DEVS - 1);
1523 + return -EINVAL;
1524 + }
1525 +
1526 + return 0;
1527 +}
1528 +
1529 +static struct xt_target xt_imq_reg[] __read_mostly = {
1530 + {
1531 + .name = "IMQ",
1532 + .family = AF_INET,
1533 + .checkentry = imq_checkentry,
1534 + .target = imq_target,
1535 + .targetsize = sizeof(struct xt_imq_info),
1536 + .table = "mangle",
1537 + .me = THIS_MODULE
1538 + },
1539 + {
1540 + .name = "IMQ",
1541 + .family = AF_INET6,
1542 + .checkentry = imq_checkentry,
1543 + .target = imq_target,
1544 + .targetsize = sizeof(struct xt_imq_info),
1545 + .table = "mangle",
1546 + .me = THIS_MODULE
1547 + },
1548 +};
1549 +
1550 +static int __init imq_init(void)
1551 +{
1552 + return xt_register_targets(xt_imq_reg, ARRAY_SIZE(xt_imq_reg));
1553 +}
1554 +
1555 +static void __exit imq_fini(void)
1556 +{
1557 + xt_unregister_targets(xt_imq_reg, ARRAY_SIZE(xt_imq_reg));
1558 +}
1559 +
1560 +module_init(imq_init);
1561 +module_exit(imq_fini);
1562 +
1563 +MODULE_AUTHOR("http://www.linuximq.net");
1564 +MODULE_DESCRIPTION("Pseudo-driver for the intermediate queue device. See http://www.linuximq.net/ for more information.");
1565 +MODULE_LICENSE("GPL");
1566 +MODULE_ALIAS("ipt_IMQ");
1567 +MODULE_ALIAS("ip6t_IMQ");
1568 +