kernel-imq.patch

   1 diff -uNr linux-3.1/drivers/net/imq.c linux-3.1-imq/drivers/net/imq.c
   2 --- linux-3.1/drivers/net/imq.c 1970-01-01 02:00:00.000000000 +0200
   3 +++ linux-3.1-imq/drivers/net/imq.c     2011-11-04 12:16:10.454992642 +0200
   4 @@ -0,0 +1,850 @@
   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 + *            Also, many thanks to pablo Sebastian Greco for making the initial
 129 + *            patch and to those who helped the testing.
 130 + *
 131 + *             More info at: http://www.linuximq.net/ (Andre Correa)
 132 + */
 133 +
 134 +#include <linux/module.h>
 135 +#include <linux/kernel.h>
 136 +#include <linux/moduleparam.h>
 137 +#include <linux/list.h>
 138 +#include <linux/skbuff.h>
 139 +#include <linux/netdevice.h>
 140 +#include <linux/etherdevice.h>
 141 +#include <linux/rtnetlink.h>
 142 +#include <linux/if_arp.h>
 143 +#include <linux/netfilter.h>
 144 +#include <linux/netfilter_ipv4.h>
 145 +#if defined(CONFIG_IPV6) || defined(CONFIG_IPV6_MODULE)
 146 +       #include <linux/netfilter_ipv6.h>
 147 +#endif
 148 +#include <linux/imq.h>
 149 +#include <net/pkt_sched.h>
 150 +#include <net/netfilter/nf_queue.h>
 151 +#include <net/sock.h>
 152 +#include <linux/ip.h>
 153 +#include <linux/ipv6.h>
 154 +#include <linux/if_vlan.h>
 155 +#include <linux/if_pppox.h>
 156 +#include <net/ip.h>
 157 +#include <net/ipv6.h>
 158 +
 159 +static int imq_nf_queue(struct nf_queue_entry *entry, unsigned queue_num);
 160 +
 161 +static nf_hookfn imq_nf_hook;
 162 +
 163 +static struct nf_hook_ops imq_ops[] = {
 164 +       {
 165 +       /* imq_ingress_ipv4 */
 166 +               .hook           = imq_nf_hook,
 167 +               .owner          = THIS_MODULE,
 168 +               .pf             = PF_INET,
 169 +               .hooknum        = NF_INET_PRE_ROUTING,
 170 +#if defined(CONFIG_IMQ_BEHAVIOR_BA) || defined(CONFIG_IMQ_BEHAVIOR_BB)
 171 +               .priority       = NF_IP_PRI_MANGLE + 1,
 172 +#else
 173 +               .priority       = NF_IP_PRI_NAT_DST + 1,
 174 +#endif
 175 +       },
 176 +       {
 177 +       /* imq_egress_ipv4 */
 178 +               .hook           = imq_nf_hook,
 179 +               .owner          = THIS_MODULE,
 180 +               .pf             = PF_INET,
 181 +               .hooknum        = NF_INET_POST_ROUTING,
 182 +#if defined(CONFIG_IMQ_BEHAVIOR_AA) || defined(CONFIG_IMQ_BEHAVIOR_BA)
 183 +               .priority       = NF_IP_PRI_LAST,
 184 +#else
 185 +               .priority       = NF_IP_PRI_NAT_SRC - 1,
 186 +#endif
 187 +       },
 188 +#if defined(CONFIG_IPV6) || defined(CONFIG_IPV6_MODULE)
 189 +       {
 190 +       /* imq_ingress_ipv6 */
 191 +               .hook           = imq_nf_hook,
 192 +               .owner          = THIS_MODULE,
 193 +               .pf             = PF_INET6,
 194 +               .hooknum        = NF_INET_PRE_ROUTING,
 195 +#if defined(CONFIG_IMQ_BEHAVIOR_BA) || defined(CONFIG_IMQ_BEHAVIOR_BB)
 196 +               .priority       = NF_IP6_PRI_MANGLE + 1,
 197 +#else
 198 +               .priority       = NF_IP6_PRI_NAT_DST + 1,
 199 +#endif
 200 +       },
 201 +       {
 202 +       /* imq_egress_ipv6 */
 203 +               .hook           = imq_nf_hook,
 204 +               .owner          = THIS_MODULE,
 205 +               .pf             = PF_INET6,
 206 +               .hooknum        = NF_INET_POST_ROUTING,
 207 +#if defined(CONFIG_IMQ_BEHAVIOR_AA) || defined(CONFIG_IMQ_BEHAVIOR_BA)
 208 +               .priority       = NF_IP6_PRI_LAST,
 209 +#else
 210 +               .priority       = NF_IP6_PRI_NAT_SRC - 1,
 211 +#endif
 212 +       },
 213 +#endif
 214 +};
 215 +
 216 +#if defined(CONFIG_IMQ_NUM_DEVS)
 217 +static int numdevs = CONFIG_IMQ_NUM_DEVS;
 218 +#else
 219 +static int numdevs = IMQ_MAX_DEVS;
 220 +#endif
 221 +
 222 +static struct net_device *imq_devs_cache[IMQ_MAX_DEVS];
 223 +
 224 +#define IMQ_MAX_QUEUES 32
 225 +static int numqueues = 1;
 226 +static u32 imq_hashrnd;
 227 +
 228 +static inline __be16 pppoe_proto(const struct sk_buff *skb)
 229 +{
 230 +       return *((__be16 *)(skb_mac_header(skb) + ETH_HLEN +
 231 +                       sizeof(struct pppoe_hdr)));
 232 +}
 233 +
 234 +static u16 imq_hash(struct net_device *dev, struct sk_buff *skb)
 235 +{
 236 +       unsigned int pull_len;
 237 +       u16 protocol = skb->protocol;
 238 +       u32 addr1, addr2;
 239 +       u32 hash, ihl = 0;
 240 +       union {
 241 +               u16 in16[2];
 242 +               u32 in32;
 243 +       } ports;
 244 +       u8 ip_proto;
 245 +
 246 +       pull_len = 0;
 247 +
 248 +recheck:
 249 +       switch (protocol) {
 250 +       case htons(ETH_P_8021Q): {
 251 +               if (unlikely(skb_pull(skb, VLAN_HLEN) == NULL))
 252 +                       goto other;
 253 +
 254 +               pull_len += VLAN_HLEN;
 255 +               skb->network_header += VLAN_HLEN;
 256 +
 257 +               protocol = vlan_eth_hdr(skb)->h_vlan_encapsulated_proto;
 258 +               goto recheck;
 259 +       }
 260 +
 261 +       case htons(ETH_P_PPP_SES): {
 262 +               if (unlikely(skb_pull(skb, PPPOE_SES_HLEN) == NULL))
 263 +                       goto other;
 264 +
 265 +               pull_len += PPPOE_SES_HLEN;
 266 +               skb->network_header += PPPOE_SES_HLEN;
 267 +
 268 +               protocol = pppoe_proto(skb);
 269 +               goto recheck;
 270 +       }
 271 +
 272 +       case htons(ETH_P_IP): {
 273 +               const struct iphdr *iph = ip_hdr(skb);
 274 +
 275 +               if (unlikely(!pskb_may_pull(skb, sizeof(struct iphdr))))
 276 +                       goto other;
 277 +
 278 +               addr1 = iph->daddr;
 279 +               addr2 = iph->saddr;
 280 +
 281 +               ip_proto = !(ip_hdr(skb)->frag_off & htons(IP_MF | IP_OFFSET)) ?
 282 +                                iph->protocol : 0;
 283 +               ihl = ip_hdrlen(skb);
 284 +
 285 +               break;
 286 +       }
 287 +#if defined(CONFIG_IPV6) || defined(CONFIG_IPV6_MODULE)
 288 +       case htons(ETH_P_IPV6): {
 289 +               const struct ipv6hdr *iph = ipv6_hdr(skb);
 290 +               __be16 frag_off;
 291 +               if (unlikely(!pskb_may_pull(skb, sizeof(struct ipv6hdr))))
 292 +                       goto other;
 293 +
 294 +               addr1 = iph->daddr.s6_addr32[3];
 295 +               addr2 = iph->saddr.s6_addr32[3];
 296 +               ihl = ipv6_skip_exthdr(skb, sizeof(struct ipv6hdr), &ip_proto, &frag_off);
 297 +               if (unlikely(ihl < 0))
 298 +                       goto other;
 299 +
 300 +               break;
 301 +       }
 302 +#endif
 303 +       default:
 304 +other:
 305 +               if (pull_len != 0) {
 306 +                       skb_push(skb, pull_len);
 307 +                       skb->network_header -= pull_len;
 308 +               }
 309 +
 310 +               return (u16)(ntohs(protocol) % dev->real_num_tx_queues);
 311 +       }
 312 +
 313 +       if (addr1 > addr2)
 314 +               swap(addr1, addr2);
 315 +
 316 +       switch (ip_proto) {
 317 +       case IPPROTO_TCP:
 318 +       case IPPROTO_UDP:
 319 +       case IPPROTO_DCCP:
 320 +       case IPPROTO_ESP:
 321 +       case IPPROTO_AH:
 322 +       case IPPROTO_SCTP:
 323 +       case IPPROTO_UDPLITE: {
 324 +               if (likely(skb_copy_bits(skb, ihl, &ports.in32, 4) >= 0)) {
 325 +                       if (ports.in16[0] > ports.in16[1])
 326 +                               swap(ports.in16[0], ports.in16[1]);
 327 +                       break;
 328 +               }
 329 +               /* fall-through */
 330 +       }
 331 +       default:
 332 +               ports.in32 = 0;
 333 +               break;
 334 +       }
 335 +
 336 +       if (pull_len != 0) {
 337 +               skb_push(skb, pull_len);
 338 +               skb->network_header -= pull_len;
 339 +       }
 340 +
 341 +       hash = jhash_3words(addr1, addr2, ports.in32, imq_hashrnd ^ ip_proto);
 342 +
 343 +       return (u16)(((u64)hash * dev->real_num_tx_queues) >> 32);
 344 +}
 345 +
 346 +static inline bool sk_tx_queue_recorded(struct sock *sk)
 347 +{
 348 +       return (sk_tx_queue_get(sk) >= 0);
 349 +}
 350 +
 351 +static struct netdev_queue *imq_select_queue(struct net_device *dev,
 352 +                                               struct sk_buff *skb)
 353 +{
 354 +       u16 queue_index = 0;
 355 +       u32 hash;
 356 +
 357 +       if (likely(dev->real_num_tx_queues == 1))
 358 +               goto out;
 359 +
 360 +       /* IMQ can be receiving ingress or engress packets. */
 361 +
 362 +       /* Check first for if rx_queue is set */
 363 +       if (skb_rx_queue_recorded(skb)) {
 364 +               queue_index = skb_get_rx_queue(skb);
 365 +               goto out;
 366 +       }
 367 +
 368 +       /* Check if socket has tx_queue set */
 369 +       if (sk_tx_queue_recorded(skb->sk)) {
 370 +               queue_index = sk_tx_queue_get(skb->sk);
 371 +               goto out;
 372 +       }
 373 +
 374 +       /* Try use socket hash */
 375 +       if (skb->sk && skb->sk->sk_hash) {
 376 +               hash = skb->sk->sk_hash;
 377 +               queue_index =
 378 +                       (u16)(((u64)hash * dev->real_num_tx_queues) >> 32);
 379 +               goto out;
 380 +       }
 381 +
 382 +       /* Generate hash from packet data */
 383 +       queue_index = imq_hash(dev, skb);
 384 +
 385 +out:
 386 +       if (unlikely(queue_index >= dev->real_num_tx_queues))
 387 +               queue_index = (u16)((u32)queue_index % dev->real_num_tx_queues);
 388 +
 389 +       skb_set_queue_mapping(skb, queue_index);
 390 +       return netdev_get_tx_queue(dev, queue_index);
 391 +}
 392 +
 393 +static struct net_device_stats *imq_get_stats(struct net_device *dev)
 394 +{
 395 +       return &dev->stats;
 396 +}
 397 +
 398 +/* called for packets kfree'd in qdiscs at places other than enqueue */
 399 +static void imq_skb_destructor(struct sk_buff *skb)
 400 +{
 401 +       struct nf_queue_entry *entry = skb->nf_queue_entry;
 402 +
 403 +       skb->nf_queue_entry = NULL;
 404 +
 405 +       if (entry) {
 406 +               nf_queue_entry_release_refs(entry);
 407 +               kfree(entry);
 408 +       }
 409 +
 410 +       skb_restore_cb(skb); /* kfree backup */
 411 +}
 412 +
 413 +static void imq_done_check_queue_mapping(struct sk_buff *skb,
 414 +                                        struct net_device *dev)
 415 +{
 416 +       unsigned int queue_index;
 417 +
 418 +       /* Don't let queue_mapping be left too large after exiting IMQ */
 419 +       if (likely(skb->dev != dev && skb->dev != NULL)) {
 420 +               queue_index = skb_get_queue_mapping(skb);
 421 +               if (unlikely(queue_index >= skb->dev->real_num_tx_queues)) {
 422 +                       queue_index = (u16)((u32)queue_index %
 423 +                                               skb->dev->real_num_tx_queues);
 424 +                       skb_set_queue_mapping(skb, queue_index);
 425 +               }
 426 +       } else {
 427 +               /* skb->dev was IMQ device itself or NULL, be on safe side and
 428 +                * just clear queue mapping.
 429 +                */
 430 +               skb_set_queue_mapping(skb, 0);
 431 +       }
 432 +}
 433 +
 434 +static netdev_tx_t imq_dev_xmit(struct sk_buff *skb, struct net_device *dev)
 435 +{
 436 +       struct nf_queue_entry *entry = skb->nf_queue_entry;
 437 +
 438 +       skb->nf_queue_entry = NULL;
 439 +       dev->trans_start = jiffies;
 440 +
 441 +       dev->stats.tx_bytes += skb->len;
 442 +       dev->stats.tx_packets++;
 443 +
 444 +       if (unlikely(entry == NULL)) {
 445 +               /* We don't know what is going on here.. packet is queued for
 446 +                * imq device, but (probably) not by us.
 447 +                *
 448 +                * If this packet was not send here by imq_nf_queue(), then
 449 +                * skb_save_cb() was not used and skb_free() should not show:
 450 +                *   WARNING: IMQ: kfree_skb: skb->cb_next:..
 451 +                * and/or
 452 +                *   WARNING: IMQ: kfree_skb: skb->nf_queue_entry...
 453 +                *
 454 +                * However if this message is shown, then IMQ is somehow broken
 455 +                * and you should report this to linuximq.net.
 456 +                */
 457 +
 458 +               /* imq_dev_xmit is black hole that eats all packets, report that
 459 +                * we eat this packet happily and increase dropped counters.
 460 +                */
 461 +
 462 +               dev->stats.tx_dropped++;
 463 +               dev_kfree_skb(skb);
 464 +
 465 +               return NETDEV_TX_OK;
 466 +       }
 467 +
 468 +       skb_restore_cb(skb); /* restore skb->cb */
 469 +
 470 +       skb->imq_flags = 0;
 471 +       skb->destructor = NULL;
 472 +
 473 +       imq_done_check_queue_mapping(skb, dev);
 474 +
 475 +       nf_reinject(entry, NF_ACCEPT);
 476 +
 477 +       return NETDEV_TX_OK;
 478 +}
 479 +
 480 +static struct net_device *get_imq_device_by_index(int index)
 481 +{
 482 +       struct net_device *dev = NULL;
 483 +       struct net *net;
 484 +       char buf[8];
 485 +
 486 +       /* get device by name and cache result */
 487 +       snprintf(buf, sizeof(buf), "imq%d", index);
 488 +
 489 +       /* Search device from all namespaces. */
 490 +       for_each_net(net) {
 491 +               dev = dev_get_by_name(net, buf);
 492 +               if (dev)
 493 +                       break;
 494 +       }
 495 +
 496 +       if (WARN_ON_ONCE(dev == NULL)) {
 497 +               /* IMQ device not found. Exotic config? */
 498 +               return ERR_PTR(-ENODEV);
 499 +       }
 500 +
 501 +       imq_devs_cache[index] = dev;
 502 +       dev_put(dev);
 503 +
 504 +       return dev;
 505 +}
 506 +
 507 +static int imq_nf_queue(struct nf_queue_entry *entry, unsigned queue_num)
 508 +{
 509 +       struct net_device *dev;
 510 +       struct sk_buff *skb_orig, *skb, *skb_shared;
 511 +       struct Qdisc *q;
 512 +       struct netdev_queue *txq;
 513 +       spinlock_t *root_lock;
 514 +       int users, index;
 515 +       int retval = -EINVAL;
 516 +       unsigned int orig_queue_index;
 517 +
 518 +       index = entry->skb->imq_flags & IMQ_F_IFMASK;
 519 +       if (unlikely(index > numdevs - 1)) {
 520 +               if (net_ratelimit())
 521 +                       printk(KERN_WARNING
 522 +                              "IMQ: invalid device specified, highest is %u\n",
 523 +                              numdevs - 1);
 524 +               retval = -EINVAL;
 525 +               goto out;
 526 +       }
 527 +
 528 +       /* check for imq device by index from cache */
 529 +       dev = imq_devs_cache[index];
 530 +       if (unlikely(!dev)) {
 531 +               dev = get_imq_device_by_index(index);
 532 +               if (IS_ERR(dev)) {
 533 +                       retval = PTR_ERR(dev);
 534 +                       goto out;
 535 +               }
 536 +       }
 537 +
 538 +       if (unlikely(!(dev->flags & IFF_UP))) {
 539 +               entry->skb->imq_flags = 0;
 540 +               nf_reinject(entry, NF_ACCEPT);
 541 +               retval = 0;
 542 +               goto out;
 543 +       }
 544 +       dev->last_rx = jiffies;
 545 +
 546 +       skb = entry->skb;
 547 +       skb_orig = NULL;
 548 +
 549 +       /* skb has owner? => make clone */
 550 +       if (unlikely(skb->destructor)) {
 551 +               skb_orig = skb;
 552 +               skb = skb_clone(skb, GFP_ATOMIC);
 553 +               if (unlikely(!skb)) {
 554 +                       retval = -ENOMEM;
 555 +                       goto out;
 556 +               }
 557 +               entry->skb = skb;
 558 +       }
 559 +
 560 +       skb->nf_queue_entry = entry;
 561 +
 562 +       dev->stats.rx_bytes += skb->len;
 563 +       dev->stats.rx_packets++;
 564 +
 565 +       if (!skb->dev) {
 566 +               /* skb->dev == NULL causes problems, try the find cause. */
 567 +               if (net_ratelimit()) {
 568 +                       dev_warn(&dev->dev,
 569 +                                "received packet with skb->dev == NULL\n");
 570 +                       dump_stack();
 571 +               }
 572 +
 573 +               skb->dev = dev;
 574 +       }
 575 +
 576 +       /* Disables softirqs for lock below */
 577 +       rcu_read_lock_bh();
 578 +
 579 +       /* Multi-queue selection */
 580 +       orig_queue_index = skb_get_queue_mapping(skb);
 581 +       txq = imq_select_queue(dev, skb);
 582 +
 583 +       q = rcu_dereference(txq->qdisc);
 584 +       if (unlikely(!q->enqueue))
 585 +               goto packet_not_eaten_by_imq_dev;
 586 +
 587 +       root_lock = qdisc_lock(q);
 588 +       spin_lock(root_lock);
 589 +
 590 +       users = atomic_read(&skb->users);
 591 +
 592 +       skb_shared = skb_get(skb); /* increase reference count by one */
 593 +       skb_save_cb(skb_shared); /* backup skb->cb, as qdisc layer will
 594 +                                       overwrite it */
 595 +       qdisc_enqueue_root(skb_shared, q); /* might kfree_skb */
 596 +
 597 +       if (likely(atomic_read(&skb_shared->users) == users + 1)) {
 598 +               kfree_skb(skb_shared); /* decrease reference count by one */
 599 +
 600 +               skb->destructor = &imq_skb_destructor;
 601 +
 602 +               /* cloned? */
 603 +               if (unlikely(skb_orig))
 604 +                       kfree_skb(skb_orig); /* free original */
 605 +
 606 +               spin_unlock(root_lock);
 607 +               rcu_read_unlock_bh();
 608 +
 609 +               /* schedule qdisc dequeue */
 610 +               __netif_schedule(q);
 611 +
 612 +               retval = 0;
 613 +               goto out;
 614 +       } else {
 615 +               skb_restore_cb(skb_shared); /* restore skb->cb */
 616 +               skb->nf_queue_entry = NULL;
 617 +               /* qdisc dropped packet and decreased skb reference count of
 618 +                * skb, so we don't really want to and try refree as that would
 619 +                * actually destroy the skb. */
 620 +               spin_unlock(root_lock);
 621 +               goto packet_not_eaten_by_imq_dev;
 622 +       }
 623 +
 624 +packet_not_eaten_by_imq_dev:
 625 +       skb_set_queue_mapping(skb, orig_queue_index);
 626 +       rcu_read_unlock_bh();
 627 +
 628 +       /* cloned? restore original */
 629 +       if (unlikely(skb_orig)) {
 630 +               kfree_skb(skb);
 631 +               entry->skb = skb_orig;
 632 +       }
 633 +       retval = -1;
 634 +out:
 635 +       return retval;
 636 +}
 637 +
 638 +static unsigned int imq_nf_hook(unsigned int hook, struct sk_buff *pskb,
 639 +                               const struct net_device *indev,
 640 +                               const struct net_device *outdev,
 641 +                               int (*okfn)(struct sk_buff *))
 642 +{
 643 +       return (pskb->imq_flags & IMQ_F_ENQUEUE) ? NF_IMQ_QUEUE : NF_ACCEPT;
 644 +}
 645 +
 646 +static int imq_close(struct net_device *dev)
 647 +{
 648 +       netif_stop_queue(dev);
 649 +       return 0;
 650 +}
 651 +
 652 +static int imq_open(struct net_device *dev)
 653 +{
 654 +       netif_start_queue(dev);
 655 +       return 0;
 656 +}
 657 +
 658 +static const struct net_device_ops imq_netdev_ops = {
 659 +       .ndo_open               = imq_open,
 660 +       .ndo_stop               = imq_close,
 661 +       .ndo_start_xmit         = imq_dev_xmit,
 662 +       .ndo_get_stats          = imq_get_stats,
 663 +};
 664 +
 665 +static void imq_setup(struct net_device *dev)
 666 +{
 667 +       dev->netdev_ops         = &imq_netdev_ops;
 668 +       dev->type               = ARPHRD_VOID;
 669 +       dev->mtu                = 16000; /* too small? */
 670 +       dev->tx_queue_len       = 11000; /* too big? */
 671 +       dev->flags              = IFF_NOARP;
 672 +       dev->features           = NETIF_F_SG | NETIF_F_FRAGLIST |
 673 +                                 NETIF_F_GSO | NETIF_F_HW_CSUM |
 674 +                                 NETIF_F_HIGHDMA;
 675 +       dev->priv_flags         &= ~(IFF_XMIT_DST_RELEASE |
 676 +                                    IFF_TX_SKB_SHARING);
 677 +}
 678 +
 679 +static int imq_validate(struct nlattr *tb[], struct nlattr *data[])
 680 +{
 681 +       int ret = 0;
 682 +
 683 +       if (tb[IFLA_ADDRESS]) {
 684 +               if (nla_len(tb[IFLA_ADDRESS]) != ETH_ALEN) {
 685 +                       ret = -EINVAL;
 686 +                       goto end;
 687 +               }
 688 +               if (!is_valid_ether_addr(nla_data(tb[IFLA_ADDRESS]))) {
 689 +                       ret = -EADDRNOTAVAIL;
 690 +                       goto end;
 691 +               }
 692 +       }
 693 +       return 0;
 694 +end:
 695 +       printk(KERN_WARNING "IMQ: imq_validate failed (%d)\n", ret);
 696 +       return ret;
 697 +}
 698 +
 699 +static struct rtnl_link_ops imq_link_ops __read_mostly = {
 700 +       .kind           = "imq",
 701 +       .priv_size      = 0,
 702 +       .setup          = imq_setup,
 703 +       .validate       = imq_validate,
 704 +};
 705 +
 706 +static const struct nf_queue_handler imq_nfqh = {
 707 +       .name  = "imq",
 708 +       .outfn = imq_nf_queue,
 709 +};
 710 +
 711 +static int __init imq_init_hooks(void)
 712 +{
 713 +       int ret;
 714 +
 715 +       nf_register_queue_imq_handler(&imq_nfqh);
 716 +
 717 +       ret = nf_register_hooks(imq_ops, ARRAY_SIZE(imq_ops));
 718 +       if (ret < 0)
 719 +               nf_unregister_queue_imq_handler();
 720 +
 721 +       return ret;
 722 +}
 723 +
 724 +static int __init imq_init_one(int index)
 725 +{
 726 +       struct net_device *dev;
 727 +       int ret;
 728 +
 729 +       dev = alloc_netdev_mq(0, "imq%d", imq_setup, numqueues);
 730 +       if (!dev)
 731 +               return -ENOMEM;
 732 +
 733 +       ret = dev_alloc_name(dev, dev->name);
 734 +       if (ret < 0)
 735 +               goto fail;
 736 +
 737 +       dev->rtnl_link_ops = &imq_link_ops;
 738 +       ret = register_netdevice(dev);
 739 +       if (ret < 0)
 740 +               goto fail;
 741 +
 742 +       return 0;
 743 +fail:
 744 +       free_netdev(dev);
 745 +       return ret;
 746 +}
 747 +
 748 +static int __init imq_init_devs(void)
 749 +{
 750 +       int err, i;
 751 +
 752 +       if (numdevs < 1 || numdevs > IMQ_MAX_DEVS) {
 753 +               printk(KERN_ERR "IMQ: numdevs has to be betweed 1 and %u\n",
 754 +                      IMQ_MAX_DEVS);
 755 +               return -EINVAL;
 756 +       }
 757 +
 758 +       if (numqueues < 1 || numqueues > IMQ_MAX_QUEUES) {
 759 +               printk(KERN_ERR "IMQ: numqueues has to be betweed 1 and %u\n",
 760 +                      IMQ_MAX_QUEUES);
 761 +               return -EINVAL;
 762 +       }
 763 +
 764 +       get_random_bytes(&imq_hashrnd, sizeof(imq_hashrnd));
 765 +
 766 +       rtnl_lock();
 767 +       err = __rtnl_link_register(&imq_link_ops);
 768 +
 769 +       for (i = 0; i < numdevs && !err; i++)
 770 +               err = imq_init_one(i);
 771 +
 772 +       if (err) {
 773 +               __rtnl_link_unregister(&imq_link_ops);
 774 +               memset(imq_devs_cache, 0, sizeof(imq_devs_cache));
 775 +       }
 776 +       rtnl_unlock();
 777 +
 778 +       return err;
 779 +}
 780 +
 781 +static int __init imq_init_module(void)
 782 +{
 783 +       int err;
 784 +
 785 +#if defined(CONFIG_IMQ_NUM_DEVS)
 786 +       BUILD_BUG_ON(CONFIG_IMQ_NUM_DEVS > 16);
 787 +       BUILD_BUG_ON(CONFIG_IMQ_NUM_DEVS < 2);
 788 +       BUILD_BUG_ON(CONFIG_IMQ_NUM_DEVS - 1 > IMQ_F_IFMASK);
 789 +#endif
 790 +
 791 +       err = imq_init_devs();
 792 +       if (err) {
 793 +               printk(KERN_ERR "IMQ: Error trying imq_init_devs(net)\n");
 794 +               return err;
 795 +       }
 796 +
 797 +       err = imq_init_hooks();
 798 +       if (err) {
 799 +               printk(KERN_ERR "IMQ: Error trying imq_init_hooks()\n");
 800 +               rtnl_link_unregister(&imq_link_ops);
 801 +               memset(imq_devs_cache, 0, sizeof(imq_devs_cache));
 802 +               return err;
 803 +       }
 804 +
 805 +       printk(KERN_INFO "IMQ driver loaded successfully. "
 806 +               "(numdevs = %d, numqueues = %d)\n", numdevs, numqueues);
 807 +
 808 +#if defined(CONFIG_IMQ_BEHAVIOR_BA) || defined(CONFIG_IMQ_BEHAVIOR_BB)
 809 +       printk(KERN_INFO "\tHooking IMQ before NAT on PREROUTING.\n");
 810 +#else
 811 +       printk(KERN_INFO "\tHooking IMQ after NAT on PREROUTING.\n");
 812 +#endif
 813 +#if defined(CONFIG_IMQ_BEHAVIOR_AB) || defined(CONFIG_IMQ_BEHAVIOR_BB)
 814 +       printk(KERN_INFO "\tHooking IMQ before NAT on POSTROUTING.\n");
 815 +#else
 816 +       printk(KERN_INFO "\tHooking IMQ after NAT on POSTROUTING.\n");
 817 +#endif
 818 +
 819 +       return 0;
 820 +}
 821 +
 822 +static void __exit imq_unhook(void)
 823 +{
 824 +       nf_unregister_hooks(imq_ops, ARRAY_SIZE(imq_ops));
 825 +       nf_unregister_queue_imq_handler();
 826 +}
 827 +
 828 +static void __exit imq_cleanup_devs(void)
 829 +{
 830 +       rtnl_link_unregister(&imq_link_ops);
 831 +       memset(imq_devs_cache, 0, sizeof(imq_devs_cache));
 832 +}
 833 +
 834 +static void __exit imq_exit_module(void)
 835 +{
 836 +       imq_unhook();
 837 +       imq_cleanup_devs();
 838 +       printk(KERN_INFO "IMQ driver unloaded successfully.\n");
 839 +}
 840 +
 841 +module_init(imq_init_module);
 842 +module_exit(imq_exit_module);
 843 +
 844 +module_param(numdevs, int, 0);
 845 +module_param(numqueues, int, 0);
 846 +MODULE_PARM_DESC(numdevs, "number of IMQ devices (how many imq* devices will "
 847 +                       "be created)");
 848 +MODULE_PARM_DESC(numqueues, "number of queues per IMQ device");
 849 +MODULE_AUTHOR("http://www.linuximq.net");
 850 +MODULE_DESCRIPTION("Pseudo-driver for the intermediate queue device. See "
 851 +                       "http://www.linuximq.net/ for more information.");
 852 +MODULE_LICENSE("GPL");
 853 +MODULE_ALIAS_RTNL_LINK("imq");
 854 +
 855 diff -uNr linux-3.1/drivers/net/Kconfig linux-3.1-imq/drivers/net/Kconfig
 856 --- linux-3.1/drivers/net/Kconfig       2011-10-24 10:10:05.000000000 +0300
 857 +++ linux-3.1-imq/drivers/net/Kconfig   2011-11-04 11:12:52.106390309 +0200
 858 @@ -124,6 +124,125 @@
 859           To compile this driver as a module, choose M here: the module
 860           will be called eql.  If unsure, say N.
 861
 862 +config IMQ
 863 +       tristate "IMQ (intermediate queueing device) support"
 864 +       depends on NETDEVICES && NETFILTER
 865 +       ---help---
 866 +         The IMQ device(s) is used as placeholder for QoS queueing
 867 +         disciplines. Every packet entering/leaving the IP stack can be
 868 +         directed through the IMQ device where it's enqueued/dequeued to the
 869 +         attached qdisc. This allows you to treat network devices as classes
 870 +         and distribute bandwidth among them. Iptables is used to specify
 871 +         through which IMQ device, if any, packets travel.
 872 +
 873 +         More information at: http://www.linuximq.net/
 874 +
 875 +         To compile this driver as a module, choose M here: the module
 876 +         will be called imq.  If unsure, say N.
 877 +
 878 +choice
 879 +       prompt "IMQ behavior (PRE/POSTROUTING)"
 880 +       depends on IMQ
 881 +       default IMQ_BEHAVIOR_AB
 882 +       help
 883 +         This setting defines how IMQ behaves in respect to its
 884 +         hooking in PREROUTING and POSTROUTING.
 885 +
 886 +         IMQ can work in any of the following ways:
 887 +
 888 +             PREROUTING   |      POSTROUTING
 889 +         -----------------|-------------------
 890 +         #1  After NAT    |      After NAT
 891 +         #2  After NAT    |      Before NAT
 892 +         #3  Before NAT   |      After NAT
 893 +         #4  Before NAT   |      Before NAT
 894 +
 895 +         The default behavior is to hook before NAT on PREROUTING
 896 +         and after NAT on POSTROUTING (#3).
 897 +
 898 +         This settings are specially usefull when trying to use IMQ
 899 +         to shape NATed clients.
 900 +
 901 +         More information can be found at: www.linuximq.net
 902 +
 903 +         If not sure leave the default settings alone.
 904 +
 905 +config IMQ_BEHAVIOR_AA
 906 +       bool "IMQ AA"
 907 +       help
 908 +         This setting defines how IMQ behaves in respect to its
 909 +         hooking in PREROUTING and POSTROUTING.
 910 +
 911 +         Choosing this option will make IMQ hook like this:
 912 +
 913 +         PREROUTING:   After NAT
 914 +         POSTROUTING:  After NAT
 915 +
 916 +         More information can be found at: www.linuximq.net
 917 +
 918 +         If not sure leave the default settings alone.
 919 +
 920 +config IMQ_BEHAVIOR_AB
 921 +       bool "IMQ AB"
 922 +       help
 923 +         This setting defines how IMQ behaves in respect to its
 924 +         hooking in PREROUTING and POSTROUTING.
 925 +
 926 +         Choosing this option will make IMQ hook like this:
 927 +
 928 +         PREROUTING:   After NAT
 929 +         POSTROUTING:  Before NAT
 930 +
 931 +         More information can be found at: www.linuximq.net
 932 +
 933 +         If not sure leave the default settings alone.
 934 +
 935 +config IMQ_BEHAVIOR_BA
 936 +       bool "IMQ BA"
 937 +       help
 938 +         This setting defines how IMQ behaves in respect to its
 939 +         hooking in PREROUTING and POSTROUTING.
 940 +
 941 +         Choosing this option will make IMQ hook like this:
 942 +
 943 +         PREROUTING:   Before NAT
 944 +         POSTROUTING:  After NAT
 945 +
 946 +         More information can be found at: www.linuximq.net
 947 +
 948 +         If not sure leave the default settings alone.
 949 +
 950 +config IMQ_BEHAVIOR_BB
 951 +       bool "IMQ BB"
 952 +       help
 953 +         This setting defines how IMQ behaves in respect to its
 954 +         hooking in PREROUTING and POSTROUTING.
 955 +
 956 +         Choosing this option will make IMQ hook like this:
 957 +
 958 +         PREROUTING:   Before NAT
 959 +         POSTROUTING:  Before NAT
 960 +
 961 +         More information can be found at: www.linuximq.net
 962 +
 963 +         If not sure leave the default settings alone.
 964 +
 965 +endchoice
 966 +
 967 +config IMQ_NUM_DEVS
 968 +       int "Number of IMQ devices"
 969 +       range 2 16
 970 +       depends on IMQ
 971 +       default "16"
 972 +       help
 973 +         This setting defines how many IMQ devices will be created.
 974 +
 975 +         The default value is 16.
 976 +
 977 +         More information can be found at: www.linuximq.net
 978 +
 979 +         If not sure leave the default settings alone.
 980 +
 981  config TUN
 982         tristate "Universal TUN/TAP device driver support"
 983         select CRC32
 984 diff -uNr linux-3.1/drivers/net/Makefile linux-3.1-imq/drivers/net/Makefile
 985 --- linux-3.1/drivers/net/Makefile      2011-10-24 10:10:05.000000000 +0300
 986 +++ linux-3.1-imq/drivers/net/Makefile  2011-11-04 11:12:52.106390309 +0200
 987 @@ -7,6 +7,7 @@
 988  #
 989  obj-$(CONFIG_BONDING) += bonding/
 990  obj-$(CONFIG_DUMMY) += dummy.o
 991 +obj-$(CONFIG_IMQ) += imq.o
 992  obj-$(CONFIG_EQUALIZER) += eql.o
 993  obj-$(CONFIG_IFB) += ifb.o
 994  obj-$(CONFIG_MACVLAN) += macvlan.o
 995 diff -uNr linux-3.1/include/linux/imq.h linux-3.1-imq/include/linux/imq.h
 996 --- linux-3.1/include/linux/imq.h       1970-01-01 02:00:00.000000000 +0200
 997 +++ linux-3.1-imq/include/linux/imq.h   2011-11-04 11:12:52.109723710 +0200
 998 @@ -0,0 +1,13 @@
 999 +#ifndef _IMQ_H
1000 +#define _IMQ_H
1001 +
1002 +/* IFMASK (16 device indexes, 0 to 15) and flag(s) fit in 5 bits */
1003 +#define IMQ_F_BITS     5
1004 +
1005 +#define IMQ_F_IFMASK   0x0f
1006 +#define IMQ_F_ENQUEUE  0x10
1007 +
1008 +#define IMQ_MAX_DEVS   (IMQ_F_IFMASK + 1)
1009 +
1010 +#endif /* _IMQ_H */
1011 +
1012 diff -uNr linux-3.1/include/linux/netfilter/xt_IMQ.h linux-3.1-imq/include/linux/netfilter/xt_IMQ.h
1013 --- linux-3.1/include/linux/netfilter/xt_IMQ.h  1970-01-01 02:00:00.000000000 +0200
1014 +++ linux-3.1-imq/include/linux/netfilter/xt_IMQ.h      2011-11-04 11:12:52.109723710 +0200
1015 @@ -0,0 +1,9 @@
1016 +#ifndef _XT_IMQ_H
1017 +#define _XT_IMQ_H
1018 +
1019 +struct xt_imq_info {
1020 +       unsigned int todev;     /* target imq device */
1021 +};
1022 +
1023 +#endif /* _XT_IMQ_H */
1024 +
1025 diff -uNr linux-3.1/include/linux/netfilter.h linux-3.1-imq/include/linux/netfilter.h
1026 --- linux-3.1/include/linux/netfilter.h 2011-10-24 10:10:05.000000000 +0300
1027 +++ linux-3.1-imq/include/linux/netfilter.h     2011-11-04 11:12:52.109723710 +0200
1028 @@ -22,7 +22,8 @@
1029  #define NF_QUEUE 3
1030  #define NF_REPEAT 4
1031  #define NF_STOP 5
1032 -#define NF_MAX_VERDICT NF_STOP
1033 +#define NF_IMQ_QUEUE 6
1034 +#define NF_MAX_VERDICT NF_IMQ_QUEUE
1035
1036  /* we overload the higher bits for encoding auxiliary data such as the queue
1037   * number or errno values. Not nice, but better than additional function
1038 diff -uNr linux-3.1/include/linux/netfilter_ipv4/ipt_IMQ.h linux-3.1-imq/include/linux/netfilter_ipv4/ipt_IMQ.h
1039 --- linux-3.1/include/linux/netfilter_ipv4/ipt_IMQ.h    1970-01-01 02:00:00.000000000 +0200
1040 +++ linux-3.1-imq/include/linux/netfilter_ipv4/ipt_IMQ.h        2011-11-04 11:12:52.109723710 +0200
1041 @@ -0,0 +1,10 @@
1042 +#ifndef _IPT_IMQ_H
1043 +#define _IPT_IMQ_H
1044 +
1045 +/* Backwards compatibility for old userspace */
1046 +#include <linux/netfilter/xt_IMQ.h>
1047 +
1048 +#define ipt_imq_info xt_imq_info
1049 +
1050 +#endif /* _IPT_IMQ_H */
1051 +
1052 diff -uNr linux-3.1/include/linux/netfilter_ipv6/ip6t_IMQ.h linux-3.1-imq/include/linux/netfilter_ipv6/ip6t_IMQ.h
1053 --- linux-3.1/include/linux/netfilter_ipv6/ip6t_IMQ.h   1970-01-01 02:00:00.000000000 +0200
1054 +++ linux-3.1-imq/include/linux/netfilter_ipv6/ip6t_IMQ.h       2011-11-04 11:12:52.113057113 +0200
1055 @@ -0,0 +1,10 @@
1056 +#ifndef _IP6T_IMQ_H
1057 +#define _IP6T_IMQ_H
1058 +
1059 +/* Backwards compatibility for old userspace */
1060 +#include <linux/netfilter/xt_IMQ.h>
1061 +
1062 +#define ip6t_imq_info xt_imq_info
1063 +
1064 +#endif /* _IP6T_IMQ_H */
1065 +
1066 diff -uNr linux-3.1/include/linux/skbuff.h linux-3.1-imq/include/linux/skbuff.h
1067 --- linux-3.1/include/linux/skbuff.h    2011-10-24 10:10:05.000000000 +0300
1068 +++ linux-3.1-imq/include/linux/skbuff.h        2011-11-04 11:12:52.116390515 +0200
1069 @@ -29,6 +29,9 @@
1070  #include <linux/hrtimer.h>
1071  #include <linux/dma-mapping.h>
1072  #include <linux/netdev_features.h>
1073 +#if defined(CONFIG_IMQ) || defined(CONFIG_IMQ_MODULE)
1074 +#include <linux/imq.h>
1075 +#endif
1076
1077  /* Don't change this without changing skb_csum_unnecessary! */
1078  #define CHECKSUM_NONE 0
1079 @@ -356,6 +359,9 @@
1080          * first. This is owned by whoever has the skb queued ATM.
1081          */
1082         char                    cb[48] __aligned(8);
1083 +#if defined(CONFIG_IMQ) || defined(CONFIG_IMQ_MODULE)
1084 +       void                    *cb_next;
1085 +#endif
1086
1087         unsigned long           _skb_refdst;
1088  #ifdef CONFIG_XFRM
1089 @@ -394,6 +400,9 @@
1090  #ifdef NET_SKBUFF_NF_DEFRAG_NEEDED
1091         struct sk_buff          *nfct_reasm;
1092  #endif
1093 +#if defined(CONFIG_IMQ) || defined(CONFIG_IMQ_MODULE)
1094 +       struct nf_queue_entry   *nf_queue_entry;
1095 +#endif
1096  #ifdef CONFIG_BRIDGE_NETFILTER
1097         struct nf_bridge_info   *nf_bridge;
1098  #endif
1099 @@ -418,6 +427,10 @@
1100
1101         /* 0/13 bit hole */
1102
1103 +#if defined(CONFIG_IMQ) || defined(CONFIG_IMQ_MODULE)
1104 +       __u8                    imq_flags:IMQ_F_BITS;
1105 +#endif
1106 +
1107  #ifdef CONFIG_NET_DMA
1108         dma_cookie_t            dma_cookie;
1109  #endif
1110 @@ -504,6 +517,12 @@
1111         return (struct rtable *)skb_dst(skb);
1112  }
1113
1114 +
1115 +#if defined(CONFIG_IMQ) || defined(CONFIG_IMQ_MODULE)
1116 +extern int skb_save_cb(struct sk_buff *skb);
1117 +extern int skb_restore_cb(struct sk_buff *skb);
1118 +#endif
1119 +
1120  extern void kfree_skb(struct sk_buff *skb);
1121  extern void consume_skb(struct sk_buff *skb);
1122  extern void           __kfree_skb(struct sk_buff *skb);
1123 @@ -2157,6 +2176,10 @@
1124         dst->nfct_reasm = src->nfct_reasm;
1125         nf_conntrack_get_reasm(src->nfct_reasm);
1126  #endif
1127 +#if defined(CONFIG_IMQ) || defined(CONFIG_IMQ_MODULE)
1128 +       dst->imq_flags = src->imq_flags;
1129 +       dst->nf_queue_entry = src->nf_queue_entry;
1130 +#endif
1131  #ifdef CONFIG_BRIDGE_NETFILTER
1132         dst->nf_bridge  = src->nf_bridge;
1133         nf_bridge_get(src->nf_bridge);
1134 diff -uNr linux-3.1/include/net/netfilter/nf_queue.h linux-3.1-imq/include/net/netfilter/nf_queue.h
1135 --- linux-3.1/include/net/netfilter/nf_queue.h  2011-10-24 10:10:05.000000000 +0300
1136 +++ linux-3.1-imq/include/net/netfilter/nf_queue.h      2011-11-04 11:12:52.116390515 +0200
1137 @@ -30,5 +30,11 @@
1138                                        const struct nf_queue_handler *qh);
1139  extern void nf_unregister_queue_handlers(const struct nf_queue_handler *qh);
1140  extern void nf_reinject(struct nf_queue_entry *entry, unsigned int verdict);
1141 +extern void nf_queue_entry_release_refs(struct nf_queue_entry *entry);
1142 +
1143 +#if defined(CONFIG_IMQ) || defined(CONFIG_IMQ_MODULE)
1144 +extern void nf_register_queue_imq_handler(const struct nf_queue_handler *qh);
1145 +extern void nf_unregister_queue_imq_handler(void);
1146 +#endif
1147
1148  #endif /* _NF_QUEUE_H */
1149 diff -uNr linux-3.1/net/core/dev.c linux-3.1-imq/net/core/dev.c
1150 --- linux-3.1/net/core/dev.c    2011-10-24 10:10:05.000000000 +0300
1151 +++ linux-3.1-imq/net/core/dev.c        2011-11-04 11:12:52.119723915 +0200
1152 @@ -98,6 +98,9 @@
1153  #include <net/net_namespace.h>
1154  #include <net/sock.h>
1155  #include <linux/rtnetlink.h>
1156 +#if defined(CONFIG_IMQ) || defined(CONFIG_IMQ_MODULE)
1157 +#include <linux/imq.h>
1158 +#endif
1159  #include <linux/proc_fs.h>
1160  #include <linux/seq_file.h>
1161  #include <linux/stat.h>
1162 @@ -2126,7 +2129,12 @@
1163                 if (dev->priv_flags & IFF_XMIT_DST_RELEASE)
1164                         skb_dst_drop(skb);
1165
1166 +#if defined(CONFIG_IMQ) || defined(CONFIG_IMQ_MODULE)
1167 +               if (!list_empty(&ptype_all) &&
1168 +                                       !(skb->imq_flags & IMQ_F_ENQUEUE))
1169 +#else
1170                 if (!list_empty(&ptype_all))
1171 +#endif
1172                         dev_queue_xmit_nit(skb, dev);
1173
1174                 skb_orphan_try(skb);
1175 diff -uNr linux-3.1/net/core/skbuff.c linux-3.1-imq/net/core/skbuff.c
1176 --- linux-3.1/net/core/skbuff.c 2011-10-24 10:10:05.000000000 +0300
1177 +++ linux-3.1-imq/net/core/skbuff.c     2011-11-04 11:12:52.123057315 +0200
1178 @@ -73,6 +73,9 @@
1179
1180  static struct kmem_cache *skbuff_head_cache __read_mostly;
1181  static struct kmem_cache *skbuff_fclone_cache __read_mostly;
1182 +#if defined(CONFIG_IMQ) || defined(CONFIG_IMQ_MODULE)
1183 +static struct kmem_cache *skbuff_cb_store_cache __read_mostly;
1184 +#endif
1185
1186  static void sock_pipe_buf_release(struct pipe_inode_info *pipe,
1187                                   struct pipe_buffer *buf)
1188 @@ -92,6 +95,82 @@
1189         return 1;
1190  }
1191
1192 +#if defined(CONFIG_IMQ) || defined(CONFIG_IMQ_MODULE)
1193 +/* Control buffer save/restore for IMQ devices */
1194 +struct skb_cb_table {
1195 +       char                    cb[48] __aligned(8);
1196 +       void                    *cb_next;
1197 +       atomic_t                refcnt;
1198 +};
1199 +
1200 +static DEFINE_SPINLOCK(skb_cb_store_lock);
1201 +
1202 +int skb_save_cb(struct sk_buff *skb)
1203 +{
1204 +       struct skb_cb_table *next;
1205 +
1206 +       next = kmem_cache_alloc(skbuff_cb_store_cache, GFP_ATOMIC);
1207 +       if (!next)
1208 +               return -ENOMEM;
1209 +
1210 +       BUILD_BUG_ON(sizeof(skb->cb) != sizeof(next->cb));
1211 +
1212 +       memcpy(next->cb, skb->cb, sizeof(skb->cb));
1213 +       next->cb_next = skb->cb_next;
1214 +
1215 +       atomic_set(&next->refcnt, 1);
1216 +
1217 +       skb->cb_next = next;
1218 +       return 0;
1219 +}
1220 +EXPORT_SYMBOL(skb_save_cb);
1221 +
1222 +int skb_restore_cb(struct sk_buff *skb)
1223 +{
1224 +       struct skb_cb_table *next;
1225 +
1226 +       if (!skb->cb_next)
1227 +               return 0;
1228 +
1229 +       next = skb->cb_next;
1230 +
1231 +       BUILD_BUG_ON(sizeof(skb->cb) != sizeof(next->cb));
1232 +
1233 +       memcpy(skb->cb, next->cb, sizeof(skb->cb));
1234 +       skb->cb_next = next->cb_next;
1235 +
1236 +       spin_lock(&skb_cb_store_lock);
1237 +
1238 +       if (atomic_dec_and_test(&next->refcnt))
1239 +               kmem_cache_free(skbuff_cb_store_cache, next);
1240 +
1241 +       spin_unlock(&skb_cb_store_lock);
1242 +
1243 +       return 0;
1244 +}
1245 +EXPORT_SYMBOL(skb_restore_cb);
1246 +
1247 +static void skb_copy_stored_cb(struct sk_buff *new, const struct sk_buff *__old)
1248 +{
1249 +       struct skb_cb_table *next;
1250 +       struct sk_buff *old;
1251 +
1252 +       if (!__old->cb_next) {
1253 +               new->cb_next = NULL;
1254 +               return;
1255 +       }
1256 +
1257 +       spin_lock(&skb_cb_store_lock);
1258 +
1259 +       old = (struct sk_buff *)__old;
1260 +
1261 +       next = old->cb_next;
1262 +       atomic_inc(&next->refcnt);
1263 +       new->cb_next = next;
1264 +
1265 +       spin_unlock(&skb_cb_store_lock);
1266 +}
1267 +#endif
1268
1269  /* Pipe buffer operations for a socket. */
1270  static const struct pipe_buf_operations sock_pipe_buf_ops = {
1271 @@ -392,6 +471,26 @@
1272                 WARN_ON(in_irq());
1273                 skb->destructor(skb);
1274         }
1275 +#if defined(CONFIG_IMQ) || defined(CONFIG_IMQ_MODULE)
1276 +       /* This should not happen. When it does, avoid memleak by restoring
1277 +       the chain of cb-backups. */
1278 +       while (skb->cb_next != NULL) {
1279 +               if (net_ratelimit())
1280 +                       printk(KERN_WARNING "IMQ: kfree_skb: skb->cb_next: "
1281 +                               "%08x\n", (unsigned int)skb->cb_next);
1282 +
1283 +               skb_restore_cb(skb);
1284 +       }
1285 +       /* This should not happen either, nf_queue_entry is nullified in
1286 +        * imq_dev_xmit(). If we have non-NULL nf_queue_entry then we are
1287 +        * leaking entry pointers, maybe memory. We don't know if this is
1288 +        * pointer to already freed memory, or should this be freed.
1289 +        * If this happens we need to add refcounting, etc for nf_queue_entry.
1290 +        */
1291 +       if (skb->nf_queue_entry && net_ratelimit())
1292 +               printk(KERN_WARNING
1293 +                               "IMQ: kfree_skb: skb->nf_queue_entry != NULL");
1294 +#endif
1295  #if IS_ENABLED(CONFIG_NF_CONNTRACK)
1296         nf_conntrack_put(skb->nfct);
1297  #endif
1298 @@ -533,6 +632,9 @@
1299         new->sp                 = secpath_get(old->sp);
1300  #endif
1301         memcpy(new->cb, old->cb, sizeof(old->cb));
1302 +#if defined(CONFIG_IMQ) || defined(CONFIG_IMQ_MODULE)
1303 +       skb_copy_stored_cb(new, old);
1304 +#endif
1305         new->csum               = old->csum;
1306         new->local_df           = old->local_df;
1307         new->pkt_type           = old->pkt_type;
1308 @@ -2888,6 +2990,13 @@
1309                                                 0,
1310                                                 SLAB_HWCACHE_ALIGN|SLAB_PANIC,
1311                                                 NULL);
1312 +#if defined(CONFIG_IMQ) || defined(CONFIG_IMQ_MODULE)
1313 +       skbuff_cb_store_cache = kmem_cache_create("skbuff_cb_store_cache",
1314 +                                                 sizeof(struct skb_cb_table),
1315 +                                                 0,
1316 +                                                 SLAB_HWCACHE_ALIGN|SLAB_PANIC,
1317 +                                                 NULL);
1318 +#endif
1319  }
1320
1321  /**
1322 diff -uNr linux-3.1/net/ipv6/ip6_output.c linux-3.1-imq/net/ipv6/ip6_output.c
1323 --- linux-3.1/net/ipv6/ip6_output.c     2011-10-24 10:10:05.000000000 +0300
1324 +++ linux-3.1-imq/net/ipv6/ip6_output.c 2011-11-04 11:12:52.123057315 +0200
1325 @@ -102,9 +102,6 @@
1326         struct net_device *dev = dst->dev;
1327         struct neighbour *neigh;
1328
1329 -       skb->protocol = htons(ETH_P_IPV6);
1330 -       skb->dev = dev;
1331 -
1332         if (ipv6_addr_is_multicast(&ipv6_hdr(skb)->daddr)) {
1333                 struct inet6_dev *idev = ip6_dst_idev(skb_dst(skb));
1334
1335 @@ -170,6 +167,11 @@
1336                 return 0;
1337         }
1338
1339 +       /* IMQ-patch: moved setting skb->dev and skb->protocol from
1340 +        * ip6_finish_output2 to fix crashing at netif_skb_features(). */
1341 +       skb->protocol = htons(ETH_P_IPV6);
1342 +       skb->dev = dev;
1343 +
1344         return NF_HOOK_COND(NFPROTO_IPV6, NF_INET_POST_ROUTING, skb, NULL, dev,
1345                             ip6_finish_output,
1346                             !(IP6CB(skb)->flags & IP6SKB_REROUTED));
1347 diff -uNr linux-3.1/net/netfilter/core.c linux-3.1-imq/net/netfilter/core.c
1348 --- linux-3.1/net/netfilter/core.c      2011-10-24 10:10:05.000000000 +0300
1349 +++ linux-3.1-imq/net/netfilter/core.c  2011-11-04 11:12:52.123057315 +0200
1350 @@ -179,9 +179,11 @@
1351                 ret = NF_DROP_GETERR(verdict);
1352                 if (ret == 0)
1353                         ret = -EPERM;
1354 -       } else if ((verdict & NF_VERDICT_MASK) == NF_QUEUE) {
1355 +       } else if ((verdict & NF_VERDICT_MASK) == NF_QUEUE ||
1356 +                  (verdict & NF_VERDICT_MASK) == NF_IMQ_QUEUE) {
1357                 int err = nf_queue(skb, elem, pf, hook, indev, outdev, okfn,
1358 -                                               verdict >> NF_VERDICT_QBITS);
1359 +                                               verdict >> NF_VERDICT_QBITS,
1360 +                                               verdict & NF_VERDICT_MASK);
1361                 if (err < 0) {
1362                         if (err == -ECANCELED)
1363                                 goto next_hook;
1364 diff -uNr linux-3.1/net/netfilter/Kconfig linux-3.1-imq/net/netfilter/Kconfig
1365 --- linux-3.1/net/netfilter/Kconfig     2011-10-24 10:10:05.000000000 +0300
1366 +++ linux-3.1-imq/net/netfilter/Kconfig 2011-11-04 11:12:52.123057315 +0200
1367 @@ -507,6 +507,18 @@
1368           For more information on the LEDs available on your system, see
1369           Documentation/leds-class.txt
1370
1371 +config NETFILTER_XT_TARGET_IMQ
1372 +        tristate '"IMQ" target support'
1373 +       depends on NETFILTER_XTABLES
1374 +       depends on IP_NF_MANGLE || IP6_NF_MANGLE
1375 +       select IMQ
1376 +       default m if NETFILTER_ADVANCED=n
1377 +        help
1378 +          This option adds a `IMQ' target which is used to specify if and
1379 +          to which imq device packets should get enqueued/dequeued.
1380 +
1381 +          To compile it as a module, choose M here.  If unsure, say N.
1382 +
1383  config NETFILTER_XT_TARGET_MARK
1384         tristate '"MARK" target support'
1385         depends on NETFILTER_ADVANCED
1386 diff -uNr linux-3.1/net/netfilter/Makefile linux-3.1-imq/net/netfilter/Makefile
1387 --- linux-3.1/net/netfilter/Makefile    2011-10-24 10:10:05.000000000 +0300
1388 +++ linux-3.1-imq/net/netfilter/Makefile        2011-11-04 11:12:52.123057315 +0200
1389 @@ -61,6 +61,7 @@
1390  obj-$(CONFIG_NETFILTER_XT_TARGET_CT) += xt_CT.o
1391  obj-$(CONFIG_NETFILTER_XT_TARGET_DSCP) += xt_DSCP.o
1392  obj-$(CONFIG_NETFILTER_XT_TARGET_HL) += xt_HL.o
1393 +obj-$(CONFIG_NETFILTER_XT_TARGET_IMQ) += xt_IMQ.o
1394  obj-$(CONFIG_NETFILTER_XT_TARGET_HMARK) += xt_HMARK.o
1395  obj-$(CONFIG_NETFILTER_XT_TARGET_LED) += xt_LED.o
1396  obj-$(CONFIG_NETFILTER_XT_TARGET_LOG) += xt_LOG.o
1397 diff -uNr linux-3.1/net/netfilter/nf_internals.h linux-3.1-imq/net/netfilter/nf_internals.h
1398 --- linux-3.1/net/netfilter/nf_internals.h      2011-10-24 10:10:05.000000000 +0300
1399 +++ linux-3.1-imq/net/netfilter/nf_internals.h  2011-11-04 11:12:52.123057315 +0200
1400 @@ -29,7 +29,7 @@
1401                     struct net_device *indev,
1402                     struct net_device *outdev,
1403                     int (*okfn)(struct sk_buff *),
1404 -                   unsigned int queuenum);
1405 +                   unsigned int queuenum, unsigned int queuetype);
1406  extern int __init netfilter_queue_init(void);
1407
1408  /* nf_log.c */
1409 diff -uNr linux-3.1/net/netfilter/nf_queue.c linux-3.1-imq/net/netfilter/nf_queue.c
1410 --- linux-3.1/net/netfilter/nf_queue.c  2011-10-24 10:10:05.000000000 +0300
1411 +++ linux-3.1-imq/net/netfilter/nf_queue.c      2011-11-04 11:12:52.123057315 +0200
1412 @@ -22,6 +22,26 @@
1413
1414  static DEFINE_MUTEX(queue_handler_mutex);
1415
1416 +#if defined(CONFIG_IMQ) || defined(CONFIG_IMQ_MODULE)
1417 +static const struct nf_queue_handler *queue_imq_handler;
1418 +
1419 +void nf_register_queue_imq_handler(const struct nf_queue_handler *qh)
1420 +{
1421 +       mutex_lock(&queue_handler_mutex);
1422 +       rcu_assign_pointer(queue_imq_handler, qh);
1423 +       mutex_unlock(&queue_handler_mutex);
1424 +}
1425 +EXPORT_SYMBOL_GPL(nf_register_queue_imq_handler);
1426 +
1427 +void nf_unregister_queue_imq_handler(void)
1428 +{
1429 +       mutex_lock(&queue_handler_mutex);
1430 +       rcu_assign_pointer(queue_imq_handler, NULL);
1431 +       mutex_unlock(&queue_handler_mutex);
1432 +}
1433 +EXPORT_SYMBOL_GPL(nf_unregister_queue_imq_handler);
1434 +#endif
1435 +
1436  /* return EBUSY when somebody else is registered, return EEXIST if the
1437   * same handler is registered, return 0 in case of success. */
1438  int nf_register_queue_handler(u_int8_t pf, const struct nf_queue_handler *qh)
1439 @@ -92,7 +112,7 @@
1440  }
1441  EXPORT_SYMBOL_GPL(nf_unregister_queue_handlers);
1442
1443 -static void nf_queue_entry_release_refs(struct nf_queue_entry *entry)
1444 +void nf_queue_entry_release_refs(struct nf_queue_entry *entry)
1445  {
1446         /* Release those devices we held, or Alexey will kill me. */
1447         if (entry->indev)
1448 @@ -112,6 +132,7 @@
1449         /* Drop reference to owner of hook which queued us. */
1450         module_put(entry->elem->owner);
1451  }
1452 +EXPORT_SYMBOL_GPL(nf_queue_entry_release_refs);
1453
1454  /*
1455   * Any packet that leaves via this function must come back
1456 @@ -123,7 +144,8 @@
1457                       struct net_device *indev,
1458                       struct net_device *outdev,
1459                       int (*okfn)(struct sk_buff *),
1460 -                     unsigned int queuenum)
1461 +                     unsigned int queuenum,
1462 +                     unsigned int queuetype)
1463  {
1464         int status = -ENOENT;
1465         struct nf_queue_entry *entry = NULL;
1466 @@ -137,7 +159,17 @@
1467         /* QUEUE == DROP if no one is waiting, to be safe. */
1468         rcu_read_lock();
1469
1470 -       qh = rcu_dereference(queue_handler[pf]);
1471 +       if (queuetype == NF_IMQ_QUEUE) {
1472 +#if defined(CONFIG_IMQ) || defined(CONFIG_IMQ_MODULE)
1473 +               qh = rcu_dereference(queue_imq_handler);
1474 +#else
1475 +               BUG();
1476 +               goto err_unlock;
1477 +#endif
1478 +       } else {
1479 +               qh = rcu_dereference(queue_handler[pf]);
1480 +       }
1481 +
1482         if (!qh) {
1483                 status = -ESRCH;
1484                 goto err_unlock;
1485 @@ -209,7 +241,8 @@
1486              struct net_device *indev,
1487              struct net_device *outdev,
1488              int (*okfn)(struct sk_buff *),
1489 -            unsigned int queuenum)
1490 +            unsigned int queuenum,
1491 +            unsigned int queuetype)
1492  {
1493         struct sk_buff *segs;
1494         int err;
1495 @@ -217,7 +250,7 @@
1496
1497         if (!skb_is_gso(skb))
1498                 return __nf_queue(skb, elem, pf, hook, indev, outdev, okfn,
1499 -                                 queuenum);
1500 +                                 queuenum, queuetype);
1501
1502         switch (pf) {
1503         case NFPROTO_IPV4:
1504 @@ -299,7 +299,7 @@ int nf_queue(struct sk_buff *skb,
1505                 if (err == 0) {
1506                         nf_bridge_adjust_segmented_data(segs);
1507                         err = __nf_queue(segs, elem, pf, hook, indev,
1508 -                                          outdev, okfn, queuenum);
1509 +                                          outdev, okfn, queuenum, queuetype);
1510                 }
1511                 if (err == 0)
1512                         queued++;
1513 @@ -299,9 +332,11 @@
1514                 local_bh_enable();
1515                 break;
1516         case NF_QUEUE:
1517 +       case NF_IMQ_QUEUE:
1518                 err = __nf_queue(skb, elem, entry->pf, entry->hook,
1519                                  entry->indev, entry->outdev, entry->okfn,
1520 -                                verdict >> NF_VERDICT_QBITS);
1521 +                                verdict >> NF_VERDICT_QBITS,
1522 +                                verdict & NF_VERDICT_MASK);
1523                 if (err < 0) {
1524                         if (err == -ECANCELED)
1525                                 goto next_hook;
1526 diff -uNr linux-3.1/net/netfilter/xt_IMQ.c linux-3.1-imq/net/netfilter/xt_IMQ.c
1527 --- linux-3.1/net/netfilter/xt_IMQ.c    1970-01-01 02:00:00.000000000 +0200
1528 +++ linux-3.1-imq/net/netfilter/xt_IMQ.c        2011-11-04 11:12:52.123057315 +0200
1529 @@ -0,0 +1,74 @@
1530 +/*
1531 + * This target marks packets to be enqueued to an imq device
1532 + */
1533 +#include <linux/module.h>
1534 +#include <linux/skbuff.h>
1535 +#include <linux/netfilter/x_tables.h>
1536 +#include <linux/netfilter/xt_IMQ.h>
1537 +#include <linux/imq.h>
1538 +
1539 +static unsigned int imq_target(struct sk_buff *pskb,
1540 +                               const struct xt_action_param *par)
1541 +{
1542 +       const struct xt_imq_info *mr = par->targinfo;
1543 +
1544 +       pskb->imq_flags = (mr->todev & IMQ_F_IFMASK) | IMQ_F_ENQUEUE;
1545 +
1546 +       return XT_CONTINUE;
1547 +}
1548 +
1549 +static int imq_checkentry(const struct xt_tgchk_param *par)
1550 +{
1551 +       struct xt_imq_info *mr = par->targinfo;
1552 +
1553 +       if (mr->todev > IMQ_MAX_DEVS - 1) {
1554 +               printk(KERN_WARNING
1555 +                      "IMQ: invalid device specified, highest is %u\n",
1556 +                      IMQ_MAX_DEVS - 1);
1557 +               return -EINVAL;
1558 +       }
1559 +
1560 +       return 0;
1561 +}
1562 +
1563 +static struct xt_target xt_imq_reg[] __read_mostly = {
1564 +       {
1565 +               .name           = "IMQ",
1566 +               .family         = AF_INET,
1567 +               .checkentry     = imq_checkentry,
1568 +               .target         = imq_target,
1569 +               .targetsize     = sizeof(struct xt_imq_info),
1570 +               .table          = "mangle",
1571 +               .me             = THIS_MODULE
1572 +       },
1573 +       {
1574 +               .name           = "IMQ",
1575 +               .family         = AF_INET6,
1576 +               .checkentry     = imq_checkentry,
1577 +               .target         = imq_target,
1578 +               .targetsize     = sizeof(struct xt_imq_info),
1579 +               .table          = "mangle",
1580 +               .me             = THIS_MODULE
1581 +       },
1582 +};
1583 +
1584 +static int __init imq_init(void)
1585 +{
1586 +       return xt_register_targets(xt_imq_reg, ARRAY_SIZE(xt_imq_reg));
1587 +}
1588 +
1589 +static void __exit imq_fini(void)
1590 +{
1591 +       xt_unregister_targets(xt_imq_reg, ARRAY_SIZE(xt_imq_reg));
1592 +}
1593 +
1594 +module_init(imq_init);
1595 +module_exit(imq_fini);
1596 +
1597 +MODULE_AUTHOR("http://www.linuximq.net");
1598 +MODULE_DESCRIPTION("Pseudo-driver for the intermediate queue device. "
1599 +                  "See http://www.linuximq.net/ for more information.");
1600 +MODULE_LICENSE("GPL");
1601 +MODULE_ALIAS("ipt_IMQ");
1602 +MODULE_ALIAS("ip6t_IMQ");
1603 +