kernel-imq.patch

   1 diff -U 5 -Nr linux-2.6.32/drivers/net/imq.c linux-2.6.32-imq/drivers/net/imq.c
   2 --- linux-2.6.32/drivers/net/imq.c      1970-01-01 02:00:00.000000000 +0200
   3 +++ linux-2.6.32-imq/drivers/net/imq.c  2009-12-11 15:08:01.958734740 +0200
   4 @@ -0,0 +1,632 @@
   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 + *            Also, many thanks to pablo Sebastian Greco for making the initial
  87 + *            patch and to those who helped the testing.
  88 + *
  89 + *             More info at: http://www.linuximq.net/ (Andre Correa)
  90 + */
  91 +
  92 +#include <linux/module.h>
  93 +#include <linux/kernel.h>
  94 +#include <linux/moduleparam.h>
  95 +#include <linux/list.h>
  96 +#include <linux/skbuff.h>
  97 +#include <linux/netdevice.h>
  98 +#include <linux/etherdevice.h>
  99 +#include <linux/rtnetlink.h>
 100 +#include <linux/if_arp.h>
 101 +#include <linux/netfilter.h>
 102 +#include <linux/netfilter_ipv4.h>
 103 +#if defined(CONFIG_IPV6) || defined(CONFIG_IPV6_MODULE)
 104 +       #include <linux/netfilter_ipv6.h>
 105 +#endif
 106 +#include <linux/imq.h>
 107 +#include <net/pkt_sched.h>
 108 +#include <net/netfilter/nf_queue.h>
 109 +
 110 +static nf_hookfn imq_nf_hook;
 111 +
 112 +static struct nf_hook_ops imq_ingress_ipv4 = {
 113 +       .hook           = imq_nf_hook,
 114 +       .owner          = THIS_MODULE,
 115 +       .pf             = PF_INET,
 116 +       .hooknum        = NF_INET_PRE_ROUTING,
 117 +#if defined(CONFIG_IMQ_BEHAVIOR_BA) || defined(CONFIG_IMQ_BEHAVIOR_BB)
 118 +       .priority       = NF_IP_PRI_MANGLE + 1
 119 +#else
 120 +       .priority       = NF_IP_PRI_NAT_DST + 1
 121 +#endif
 122 +};
 123 +
 124 +static struct nf_hook_ops imq_egress_ipv4 = {
 125 +       .hook           = imq_nf_hook,
 126 +       .owner          = THIS_MODULE,
 127 +       .pf             = PF_INET,
 128 +       .hooknum        = NF_INET_POST_ROUTING,
 129 +#if defined(CONFIG_IMQ_BEHAVIOR_AA) || defined(CONFIG_IMQ_BEHAVIOR_BA)
 130 +       .priority       = NF_IP_PRI_LAST
 131 +#else
 132 +       .priority       = NF_IP_PRI_NAT_SRC - 1
 133 +#endif
 134 +};
 135 +
 136 +#if defined(CONFIG_IPV6) || defined(CONFIG_IPV6_MODULE)
 137 +static struct nf_hook_ops imq_ingress_ipv6 = {
 138 +       .hook           = imq_nf_hook,
 139 +       .owner          = THIS_MODULE,
 140 +       .pf             = PF_INET6,
 141 +       .hooknum        = NF_INET_PRE_ROUTING,
 142 +#if defined(CONFIG_IMQ_BEHAVIOR_BA) || defined(CONFIG_IMQ_BEHAVIOR_BB)
 143 +       .priority       = NF_IP6_PRI_MANGLE + 1
 144 +#else
 145 +       .priority       = NF_IP6_PRI_NAT_DST + 1
 146 +#endif
 147 +};
 148 +
 149 +static struct nf_hook_ops imq_egress_ipv6 = {
 150 +       .hook           = imq_nf_hook,
 151 +       .owner          = THIS_MODULE,
 152 +       .pf             = PF_INET6,
 153 +       .hooknum        = NF_INET_POST_ROUTING,
 154 +#if defined(CONFIG_IMQ_BEHAVIOR_AA) || defined(CONFIG_IMQ_BEHAVIOR_BA)
 155 +       .priority       = NF_IP6_PRI_LAST
 156 +#else
 157 +       .priority       = NF_IP6_PRI_NAT_SRC - 1
 158 +#endif
 159 +};
 160 +#endif
 161 +
 162 +#if defined(CONFIG_IMQ_NUM_DEVS)
 163 +static unsigned int numdevs = CONFIG_IMQ_NUM_DEVS;
 164 +#else
 165 +static unsigned int numdevs = IMQ_MAX_DEVS;
 166 +#endif
 167 +
 168 +static DEFINE_SPINLOCK(imq_nf_queue_lock);
 169 +
 170 +static struct net_device *imq_devs_cache[IMQ_MAX_DEVS];
 171 +
 172 +
 173 +static struct net_device_stats *imq_get_stats(struct net_device *dev)
 174 +{
 175 +       return &dev->stats;
 176 +}
 177 +
 178 +/* called for packets kfree'd in qdiscs at places other than enqueue */
 179 +static void imq_skb_destructor(struct sk_buff *skb)
 180 +{
 181 +       struct nf_queue_entry *entry = skb->nf_queue_entry;
 182 +
 183 +       skb->nf_queue_entry = NULL;
 184 +
 185 +       if (entry) {
 186 +               nf_queue_entry_release_refs(entry);
 187 +               kfree(entry);
 188 +       }
 189 +
 190 +       skb_restore_cb(skb); /* kfree backup */
 191 +}
 192 +
 193 +/* locking not needed when called from imq_nf_queue */
 194 +static void imq_nf_reinject_lockless(struct nf_queue_entry *entry,
 195 +                                               unsigned int verdict)
 196 +{
 197 +       int status;
 198 +
 199 +       if (!entry->next_outfn) {
 200 +               nf_reinject(entry, verdict);
 201 +               return;
 202 +       }
 203 +
 204 +       status = entry->next_outfn(entry, entry->next_queuenum);
 205 +       if (status < 0) {
 206 +               nf_queue_entry_release_refs(entry);
 207 +               kfree_skb(entry->skb);
 208 +               kfree(entry);
 209 +       }
 210 +}
 211 +
 212 +static void imq_nf_reinject(struct nf_queue_entry *entry, unsigned int verdict)
 213 +{
 214 +       int status;
 215 +
 216 +       if (!entry->next_outfn) {
 217 +               spin_lock_bh(&imq_nf_queue_lock);
 218 +               nf_reinject(entry, verdict);
 219 +               spin_unlock_bh(&imq_nf_queue_lock);
 220 +               return;
 221 +       }
 222 +
 223 +       rcu_read_lock();
 224 +       local_bh_disable();
 225 +       status = entry->next_outfn(entry, entry->next_queuenum);
 226 +       local_bh_enable();
 227 +       if (status < 0) {
 228 +               nf_queue_entry_release_refs(entry);
 229 +               kfree_skb(entry->skb);
 230 +               kfree(entry);
 231 +       }
 232 +
 233 +       rcu_read_unlock();
 234 +}
 235 +
 236 +static netdev_tx_t imq_dev_xmit(struct sk_buff *skb, struct net_device *dev)
 237 +{
 238 +       struct nf_queue_entry *entry = skb->nf_queue_entry;
 239 +
 240 +       skb->nf_queue_entry = NULL;
 241 +       dev->trans_start = jiffies;
 242 +
 243 +       dev->stats.tx_bytes += skb->len;
 244 +       dev->stats.tx_packets++;
 245 +
 246 +       if (entry == NULL) {
 247 +               /* We don't know what is going on here.. packet is queued for
 248 +                * imq device, but (probably) not by us.
 249 +                *
 250 +                * If this packet was not send here by imq_nf_queue(), then
 251 +                * skb_save_cb() was not used and skb_free() should not show:
 252 +                *   WARNING: IMQ: kfree_skb: skb->cb_next:..
 253 +                * and/or
 254 +                *   WARNING: IMQ: kfree_skb: skb->nf_queue_entry...
 255 +                *
 256 +                * However if this message is shown, then IMQ is somehow broken
 257 +                * and you should report this to linuximq.net.
 258 +                */
 259 +
 260 +               /* imq_dev_xmit is black hole that eats all packets, report that
 261 +                * we eat this packet happily and increase dropped counters.
 262 +                */
 263 +
 264 +               dev->stats.tx_dropped++;
 265 +               dev_kfree_skb(skb);
 266 +
 267 +               return NETDEV_TX_OK;
 268 +       }
 269 +
 270 +       skb_restore_cb(skb); /* restore skb->cb */
 271 +
 272 +       skb->imq_flags = 0;
 273 +       skb->destructor = NULL;
 274 +
 275 +       imq_nf_reinject(entry, NF_ACCEPT);
 276 +
 277 +       return NETDEV_TX_OK;
 278 +}
 279 +
 280 +static int imq_nf_queue(struct nf_queue_entry *entry, unsigned queue_num)
 281 +{
 282 +       struct net_device *dev;
 283 +       struct sk_buff *skb_orig, *skb, *skb_shared;
 284 +       struct Qdisc *q;
 285 +       struct netdev_queue *txq;
 286 +       int users, index;
 287 +       int retval = -EINVAL;
 288 +
 289 +       index = entry->skb->imq_flags & IMQ_F_IFMASK;
 290 +       if (unlikely(index > numdevs - 1)) {
 291 +               if (net_ratelimit())
 292 +                       printk(KERN_WARNING
 293 +                              "IMQ: invalid device specified, highest is %u\n",
 294 +                              numdevs - 1);
 295 +               retval = -EINVAL;
 296 +               goto out;
 297 +       }
 298 +
 299 +       /* check for imq device by index from cache */
 300 +       dev = imq_devs_cache[index];
 301 +       if (unlikely(!dev)) {
 302 +               char buf[8];
 303 +
 304 +               /* get device by name and cache result */
 305 +               snprintf(buf, sizeof(buf), "imq%d", index);
 306 +               dev = dev_get_by_name(&init_net, buf);
 307 +               if (!dev) {
 308 +                       /* not found ?!*/
 309 +                       BUG();
 310 +                       retval = -ENODEV;
 311 +                       goto out;
 312 +               }
 313 +
 314 +               imq_devs_cache[index] = dev;
 315 +               dev_put(dev);
 316 +       }
 317 +
 318 +       if (unlikely(!(dev->flags & IFF_UP))) {
 319 +               entry->skb->imq_flags = 0;
 320 +               imq_nf_reinject_lockless(entry, NF_ACCEPT);
 321 +               retval = 0;
 322 +               goto out;
 323 +       }
 324 +       dev->last_rx = jiffies;
 325 +
 326 +       skb = entry->skb;
 327 +       skb_orig = NULL;
 328 +
 329 +       /* skb has owner? => make clone */
 330 +       if (unlikely(skb->destructor)) {
 331 +               skb_orig = skb;
 332 +               skb = skb_clone(skb, GFP_ATOMIC);
 333 +               if (!skb) {
 334 +                       retval = -ENOMEM;
 335 +                       goto out;
 336 +               }
 337 +               entry->skb = skb;
 338 +       }
 339 +
 340 +       skb->nf_queue_entry = entry;
 341 +
 342 +       dev->stats.rx_bytes += skb->len;
 343 +       dev->stats.rx_packets++;
 344 +
 345 +       txq = dev_pick_tx(dev, skb);
 346 +
 347 +       q = rcu_dereference(txq->qdisc);
 348 +       if (unlikely(!q->enqueue))
 349 +               goto packet_not_eaten_by_imq_dev;
 350 +
 351 +       spin_lock_bh(qdisc_lock(q));
 352 +
 353 +       users = atomic_read(&skb->users);
 354 +
 355 +       skb_shared = skb_get(skb); /* increase reference count by one */
 356 +       skb_save_cb(skb_shared); /* backup skb->cb, as qdisc layer will
 357 +                                       overwrite it */
 358 +       qdisc_enqueue_root(skb_shared, q); /* might kfree_skb */
 359 +
 360 +       if (likely(atomic_read(&skb_shared->users) == users + 1)) {
 361 +               kfree_skb(skb_shared); /* decrease reference count by one */
 362 +
 363 +               skb->destructor = &imq_skb_destructor;
 364 +
 365 +               /* cloned? */
 366 +               if (skb_orig)
 367 +                       kfree_skb(skb_orig); /* free original */
 368 +
 369 +               spin_unlock_bh(qdisc_lock(q));
 370 +
 371 +               /* schedule qdisc dequeue */
 372 +               __netif_schedule(q);
 373 +
 374 +               retval = 0;
 375 +               goto out;
 376 +       } else {
 377 +               skb_restore_cb(skb_shared); /* restore skb->cb */
 378 +               skb->nf_queue_entry = NULL;
 379 +               /* qdisc dropped packet and decreased skb reference count of
 380 +                * skb, so we don't really want to and try refree as that would
 381 +                * actually destroy the skb. */
 382 +               spin_unlock_bh(qdisc_lock(q));
 383 +               goto packet_not_eaten_by_imq_dev;
 384 +       }
 385 +
 386 +packet_not_eaten_by_imq_dev:
 387 +       /* cloned? restore original */
 388 +       if (skb_orig) {
 389 +               kfree_skb(skb);
 390 +               entry->skb = skb_orig;
 391 +       }
 392 +       retval = -1;
 393 +out:
 394 +       return retval;
 395 +}
 396 +
 397 +static struct nf_queue_handler nfqh = {
 398 +       .name  = "imq",
 399 +       .outfn = imq_nf_queue,
 400 +};
 401 +
 402 +static unsigned int imq_nf_hook(unsigned int hook, struct sk_buff *pskb,
 403 +                               const struct net_device *indev,
 404 +                               const struct net_device *outdev,
 405 +                               int (*okfn)(struct sk_buff *))
 406 +{
 407 +       if (pskb->imq_flags & IMQ_F_ENQUEUE)
 408 +               return NF_QUEUE;
 409 +
 410 +       return NF_ACCEPT;
 411 +}
 412 +
 413 +static int imq_close(struct net_device *dev)
 414 +{
 415 +       netif_stop_queue(dev);
 416 +       return 0;
 417 +}
 418 +
 419 +static int imq_open(struct net_device *dev)
 420 +{
 421 +       netif_start_queue(dev);
 422 +       return 0;
 423 +}
 424 +
 425 +static const struct net_device_ops imq_netdev_ops = {
 426 +       .ndo_open               = imq_open,
 427 +       .ndo_stop               = imq_close,
 428 +       .ndo_start_xmit         = imq_dev_xmit,
 429 +       .ndo_get_stats          = imq_get_stats,
 430 +};
 431 +
 432 +static void imq_setup(struct net_device *dev)
 433 +{
 434 +       dev->netdev_ops         = &imq_netdev_ops;
 435 +       dev->type               = ARPHRD_VOID;
 436 +       dev->mtu                = 16000;
 437 +       dev->tx_queue_len       = 11000;
 438 +       dev->flags              = IFF_NOARP;
 439 +       dev->features           = NETIF_F_SG | NETIF_F_FRAGLIST |
 440 +                                 NETIF_F_GSO | NETIF_F_HW_CSUM |
 441 +                                 NETIF_F_HIGHDMA;
 442 +       dev->priv_flags         &= ~IFF_XMIT_DST_RELEASE;
 443 +}
 444 +
 445 +static int imq_validate(struct nlattr *tb[], struct nlattr *data[])
 446 +{
 447 +       int ret = 0;
 448 +
 449 +       if (tb[IFLA_ADDRESS]) {
 450 +               if (nla_len(tb[IFLA_ADDRESS]) != ETH_ALEN) {
 451 +                       ret = -EINVAL;
 452 +                       goto end;
 453 +               }
 454 +               if (!is_valid_ether_addr(nla_data(tb[IFLA_ADDRESS]))) {
 455 +                       ret = -EADDRNOTAVAIL;
 456 +                       goto end;
 457 +               }
 458 +       }
 459 +       return 0;
 460 +end:
 461 +       printk(KERN_WARNING "IMQ: imq_validate failed (%d)\n", ret);
 462 +       return ret;
 463 +}
 464 +
 465 +static struct rtnl_link_ops imq_link_ops __read_mostly = {
 466 +       .kind           = "imq",
 467 +       .priv_size      = 0,
 468 +       .setup          = imq_setup,
 469 +       .validate       = imq_validate,
 470 +};
 471 +
 472 +static int __init imq_init_hooks(void)
 473 +{
 474 +       int err;
 475 +
 476 +       nf_register_queue_imq_handler(&nfqh);
 477 +
 478 +       err = nf_register_hook(&imq_ingress_ipv4);
 479 +       if (err)
 480 +               goto err1;
 481 +
 482 +       err = nf_register_hook(&imq_egress_ipv4);
 483 +       if (err)
 484 +               goto err2;
 485 +
 486 +#if defined(CONFIG_IPV6) || defined(CONFIG_IPV6_MODULE)
 487 +       err = nf_register_hook(&imq_ingress_ipv6);
 488 +       if (err)
 489 +               goto err3;
 490 +
 491 +       err = nf_register_hook(&imq_egress_ipv6);
 492 +       if (err)
 493 +               goto err4;
 494 +#endif
 495 +
 496 +       return 0;
 497 +
 498 +#if defined(CONFIG_IPV6) || defined(CONFIG_IPV6_MODULE)
 499 +err4:
 500 +       nf_unregister_hook(&imq_ingress_ipv6);
 501 +err3:
 502 +       nf_unregister_hook(&imq_egress_ipv4);
 503 +#endif
 504 +err2:
 505 +       nf_unregister_hook(&imq_ingress_ipv4);
 506 +err1:
 507 +       nf_unregister_queue_imq_handler();
 508 +       return err;
 509 +}
 510 +
 511 +static int __init imq_init_one(int index)
 512 +{
 513 +       struct net_device *dev;
 514 +       int ret;
 515 +
 516 +       dev = alloc_netdev(0, "imq%d", imq_setup);
 517 +       if (!dev)
 518 +               return -ENOMEM;
 519 +
 520 +       ret = dev_alloc_name(dev, dev->name);
 521 +       if (ret < 0)
 522 +               goto fail;
 523 +
 524 +       dev->rtnl_link_ops = &imq_link_ops;
 525 +       ret = register_netdevice(dev);
 526 +       if (ret < 0)
 527 +               goto fail;
 528 +
 529 +       return 0;
 530 +fail:
 531 +       free_netdev(dev);
 532 +       return ret;
 533 +}
 534 +
 535 +static int __init imq_init_devs(void)
 536 +{
 537 +       int err, i;
 538 +
 539 +       if (numdevs < 1 || numdevs > IMQ_MAX_DEVS) {
 540 +               printk(KERN_ERR "IMQ: numdevs has to be betweed 1 and %u\n",
 541 +                      IMQ_MAX_DEVS);
 542 +               return -EINVAL;
 543 +       }
 544 +
 545 +       rtnl_lock();
 546 +       err = __rtnl_link_register(&imq_link_ops);
 547 +
 548 +       for (i = 0; i < numdevs && !err; i++)
 549 +               err = imq_init_one(i);
 550 +
 551 +       if (err) {
 552 +               __rtnl_link_unregister(&imq_link_ops);
 553 +               memset(imq_devs_cache, 0, sizeof(imq_devs_cache));
 554 +       }
 555 +       rtnl_unlock();
 556 +
 557 +       return err;
 558 +}
 559 +
 560 +static int __init imq_init_module(void)
 561 +{
 562 +       int err;
 563 +
 564 +#if defined(CONFIG_IMQ_NUM_DEVS)
 565 +       BUILD_BUG_ON(CONFIG_IMQ_NUM_DEVS > 16);
 566 +       BUILD_BUG_ON(CONFIG_IMQ_NUM_DEVS < 2);
 567 +       BUILD_BUG_ON(CONFIG_IMQ_NUM_DEVS - 1 > IMQ_F_IFMASK);
 568 +#endif
 569 +
 570 +       err = imq_init_devs();
 571 +       if (err) {
 572 +               printk(KERN_ERR "IMQ: Error trying imq_init_devs(net)\n");
 573 +               return err;
 574 +       }
 575 +
 576 +       err = imq_init_hooks();
 577 +       if (err) {
 578 +               printk(KERN_ERR "IMQ: Error trying imq_init_hooks()\n");
 579 +               rtnl_link_unregister(&imq_link_ops);
 580 +               memset(imq_devs_cache, 0, sizeof(imq_devs_cache));
 581 +               return err;
 582 +       }
 583 +
 584 +       printk(KERN_INFO "IMQ driver loaded successfully.\n");
 585 +
 586 +#if defined(CONFIG_IMQ_BEHAVIOR_BA) || defined(CONFIG_IMQ_BEHAVIOR_BB)
 587 +       printk(KERN_INFO "\tHooking IMQ before NAT on PREROUTING.\n");
 588 +#else
 589 +       printk(KERN_INFO "\tHooking IMQ after NAT on PREROUTING.\n");
 590 +#endif
 591 +#if defined(CONFIG_IMQ_BEHAVIOR_AB) || defined(CONFIG_IMQ_BEHAVIOR_BB)
 592 +       printk(KERN_INFO "\tHooking IMQ before NAT on POSTROUTING.\n");
 593 +#else
 594 +       printk(KERN_INFO "\tHooking IMQ after NAT on POSTROUTING.\n");
 595 +#endif
 596 +
 597 +       return 0;
 598 +}
 599 +
 600 +static void __exit imq_unhook(void)
 601 +{
 602 +#if defined(CONFIG_IPV6) || defined(CONFIG_IPV6_MODULE)
 603 +       nf_unregister_hook(&imq_ingress_ipv6);
 604 +       nf_unregister_hook(&imq_egress_ipv6);
 605 +#endif
 606 +       nf_unregister_hook(&imq_ingress_ipv4);
 607 +       nf_unregister_hook(&imq_egress_ipv4);
 608 +
 609 +       nf_unregister_queue_imq_handler();
 610 +}
 611 +
 612 +static void __exit imq_cleanup_devs(void)
 613 +{
 614 +       rtnl_link_unregister(&imq_link_ops);
 615 +       memset(imq_devs_cache, 0, sizeof(imq_devs_cache));
 616 +}
 617 +
 618 +static void __exit imq_exit_module(void)
 619 +{
 620 +       imq_unhook();
 621 +       imq_cleanup_devs();
 622 +       printk(KERN_INFO "IMQ driver unloaded successfully.\n");
 623 +}
 624 +
 625 +module_init(imq_init_module);
 626 +module_exit(imq_exit_module);
 627 +
 628 +module_param(numdevs, int, 0);
 629 +MODULE_PARM_DESC(numdevs, "number of IMQ devices (how many imq* devices will "
 630 +                       "be created)");
 631 +MODULE_AUTHOR("http://www.linuximq.net");
 632 +MODULE_DESCRIPTION("Pseudo-driver for the intermediate queue device. See "
 633 +                       "http://www.linuximq.net/ for more information.");
 634 +MODULE_LICENSE("GPL");
 635 +MODULE_ALIAS_RTNL_LINK("imq");
 636 +
 637 diff -U 5 -Nr linux-2.6.32/drivers/net/Kconfig linux-2.6.32-imq/drivers/net/Kconfig
 638 --- linux-2.6.32/drivers/net/Kconfig    2009-12-03 05:51:21.000000000 +0200
 639 +++ linux-2.6.32-imq/drivers/net/Kconfig        2009-12-11 14:16:42.678730699 +0200
 640 @@ -107,10 +107,133 @@
 641           <http://www.tldp.org/docs.html#howto>.
 642
 643           To compile this driver as a module, choose M here: the module
 644           will be called eql.  If unsure, say N.
 645
 646 +config IMQ
 647 +       tristate "IMQ (intermediate queueing device) support"
 648 +       depends on NETDEVICES && NETFILTER
 649 +       ---help---
 650 +         The IMQ device(s) is used as placeholder for QoS queueing
 651 +         disciplines. Every packet entering/leaving the IP stack can be
 652 +         directed through the IMQ device where it's enqueued/dequeued to the
 653 +         attached qdisc. This allows you to treat network devices as classes
 654 +         and distribute bandwidth among them. Iptables is used to specify
 655 +         through which IMQ device, if any, packets travel.
 656 +
 657 +         More information at: http://www.linuximq.net/
 658 +
 659 +         To compile this driver as a module, choose M here: the module
 660 +         will be called imq.  If unsure, say N.
 661 +
 662 +choice
 663 +       prompt "IMQ behavior (PRE/POSTROUTING)"
 664 +       depends on IMQ
 665 +       default IMQ_BEHAVIOR_AB
 666 +       help
 667 +
 668 +               This settings defines how IMQ behaves in respect to its
 669 +               hooking in PREROUTING and POSTROUTING.
 670 +
 671 +               IMQ can work in any of the following ways:
 672 +
 673 +                   PREROUTING   |      POSTROUTING
 674 +               -----------------|-------------------
 675 +               #1  After NAT    |      After NAT
 676 +               #2  After NAT    |      Before NAT
 677 +               #3  Before NAT   |      After NAT
 678 +               #4  Before NAT   |      Before NAT
 679 +
 680 +               The default behavior is to hook before NAT on PREROUTING
 681 +               and after NAT on POSTROUTING (#3).
 682 +
 683 +               This settings are specially usefull when trying to use IMQ
 684 +               to shape NATed clients.
 685 +
 686 +               More information can be found at: www.linuximq.net
 687 +
 688 +               If not sure leave the default settings alone.
 689 +
 690 +config IMQ_BEHAVIOR_AA
 691 +       bool "IMQ AA"
 692 +       help
 693 +               This settings defines how IMQ behaves in respect to its
 694 +               hooking in PREROUTING and POSTROUTING.
 695 +
 696 +               Choosing this option will make IMQ hook like this:
 697 +
 698 +               PREROUTING:   After NAT
 699 +               POSTROUTING:  After NAT
 700 +
 701 +               More information can be found at: www.linuximq.net
 702 +
 703 +               If not sure leave the default settings alone.
 704 +
 705 +config IMQ_BEHAVIOR_AB
 706 +       bool "IMQ AB"
 707 +       help
 708 +               This settings defines how IMQ behaves in respect to its
 709 +               hooking in PREROUTING and POSTROUTING.
 710 +
 711 +               Choosing this option will make IMQ hook like this:
 712 +
 713 +               PREROUTING:   After NAT
 714 +               POSTROUTING:  Before NAT
 715 +
 716 +               More information can be found at: www.linuximq.net
 717 +
 718 +               If not sure leave the default settings alone.
 719 +
 720 +config IMQ_BEHAVIOR_BA
 721 +       bool "IMQ BA"
 722 +       help
 723 +               This settings defines how IMQ behaves in respect to its
 724 +               hooking in PREROUTING and POSTROUTING.
 725 +
 726 +               Choosing this option will make IMQ hook like this:
 727 +
 728 +               PREROUTING:   Before NAT
 729 +               POSTROUTING:  After NAT
 730 +
 731 +               More information can be found at: www.linuximq.net
 732 +
 733 +               If not sure leave the default settings alone.
 734 +
 735 +config IMQ_BEHAVIOR_BB
 736 +       bool "IMQ BB"
 737 +       help
 738 +               This settings defines how IMQ behaves in respect to its
 739 +               hooking in PREROUTING and POSTROUTING.
 740 +
 741 +               Choosing this option will make IMQ hook like this:
 742 +
 743 +               PREROUTING:   Before NAT
 744 +               POSTROUTING:  Before NAT
 745 +
 746 +               More information can be found at: www.linuximq.net
 747 +
 748 +               If not sure leave the default settings alone.
 749 +
 750 +endchoice
 751 +
 752 +config IMQ_NUM_DEVS
 753 +
 754 +       int "Number of IMQ devices"
 755 +       range 2 16
 756 +       depends on IMQ
 757 +       default "16"
 758 +       help
 759 +
 760 +               This settings defines how many IMQ devices will be
 761 +               created.
 762 +
 763 +               The default value is 16.
 764 +
 765 +               More information can be found at: www.linuximq.net
 766 +
 767 +               If not sure leave the default settings alone.
 768 +
 769  config TUN
 770         tristate "Universal TUN/TAP device driver support"
 771         select CRC32
 772         ---help---
 773           TUN/TAP provides packet reception and transmission for user space
 774 diff -U 5 -Nr linux-2.6.32/drivers/net/Makefile linux-2.6.32-imq/drivers/net/Makefile
 775 --- linux-2.6.32/drivers/net/Makefile   2009-12-03 05:51:21.000000000 +0200
 776 +++ linux-2.6.32-imq/drivers/net/Makefile       2009-12-11 14:16:42.678730699 +0200
 777 @@ -163,10 +163,11 @@
 778  obj-$(CONFIG_SLHC) += slhc.o
 779
 780  obj-$(CONFIG_XEN_NETDEV_FRONTEND) += xen-netfront.o
 781
 782  obj-$(CONFIG_DUMMY) += dummy.o
 783 +obj-$(CONFIG_IMQ) += imq.o
 784  obj-$(CONFIG_IFB) += ifb.o
 785  obj-$(CONFIG_MACVLAN) += macvlan.o
 786  obj-$(CONFIG_DE600) += de600.o
 787  obj-$(CONFIG_DE620) += de620.o
 788  obj-$(CONFIG_LANCE) += lance.o
 789 diff -U 5 -Nr linux-2.6.32/include/linux/imq.h linux-2.6.32-imq/include/linux/imq.h
 790 --- linux-2.6.32/include/linux/imq.h    1970-01-01 02:00:00.000000000 +0200
 791 +++ linux-2.6.32-imq/include/linux/imq.h        2009-12-11 14:16:42.678730699 +0200
 792 @@ -0,0 +1,13 @@
 793 +#ifndef _IMQ_H
 794 +#define _IMQ_H
 795 +
 796 +/* IFMASK (16 device indexes, 0 to 15) and flag(s) fit in 5 bits */
 797 +#define IMQ_F_BITS     5
 798 +
 799 +#define IMQ_F_IFMASK   0x0f
 800 +#define IMQ_F_ENQUEUE  0x10
 801 +
 802 +#define IMQ_MAX_DEVS   (IMQ_F_IFMASK + 1)
 803 +
 804 +#endif /* _IMQ_H */
 805 +
 806 diff -U 5 -Nr linux-2.6.32/include/linux/netdevice.h linux-2.6.32-imq/include/linux/netdevice.h
 807 --- linux-2.6.32/include/linux/netdevice.h      2009-12-03 05:51:21.000000000 +0200
 808 +++ linux-2.6.32-imq/include/linux/netdevice.h  2009-12-11 14:16:42.679730960 +0200
 809 @@ -1112,10 +1112,11 @@
 810  extern struct net_device       *__dev_get_by_name(struct net *net, const char *name);
 811  extern int             dev_alloc_name(struct net_device *dev, const char *name);
 812  extern int             dev_open(struct net_device *dev);
 813  extern int             dev_close(struct net_device *dev);
 814  extern void            dev_disable_lro(struct net_device *dev);
 815 +extern struct netdev_queue *dev_pick_tx(struct net_device *dev, struct sk_buff *skb);
 816  extern int             dev_queue_xmit(struct sk_buff *skb);
 817  extern int             register_netdevice(struct net_device *dev);
 818  extern void            unregister_netdevice_queue(struct net_device *dev,
 819                                                    struct list_head *head);
 820  extern void            unregister_netdevice_many(struct list_head *head);
 821 diff -U 5 -Nr linux-2.6.32/include/linux/netfilter/xt_IMQ.h linux-2.6.32-imq/include/linux/netfilter/xt_IMQ.h
 822 --- linux-2.6.32/include/linux/netfilter/xt_IMQ.h       1970-01-01 02:00:00.000000000 +0200
 823 +++ linux-2.6.32-imq/include/linux/netfilter/xt_IMQ.h   2009-12-11 14:16:42.679730960 +0200
 824 @@ -0,0 +1,9 @@
 825 +#ifndef _XT_IMQ_H
 826 +#define _XT_IMQ_H
 827 +
 828 +struct xt_imq_info {
 829 +       unsigned int todev;     /* target imq device */
 830 +};
 831 +
 832 +#endif /* _XT_IMQ_H */
 833 +
 834 diff -U 5 -Nr linux-2.6.32/include/linux/netfilter_ipv4/ipt_IMQ.h linux-2.6.32-imq/include/linux/netfilter_ipv4/ipt_IMQ.h
 835 --- linux-2.6.32/include/linux/netfilter_ipv4/ipt_IMQ.h 1970-01-01 02:00:00.000000000 +0200
 836 +++ linux-2.6.32-imq/include/linux/netfilter_ipv4/ipt_IMQ.h     2009-12-11 14:16:42.679730960 +0200
 837 @@ -0,0 +1,10 @@
 838 +#ifndef _IPT_IMQ_H
 839 +#define _IPT_IMQ_H
 840 +
 841 +/* Backwards compatibility for old userspace */
 842 +#include <linux/netfilter/xt_IMQ.h>
 843 +
 844 +#define ipt_imq_info xt_imq_info
 845 +
 846 +#endif /* _IPT_IMQ_H */
 847 +
 848 diff -U 5 -Nr linux-2.6.32/include/linux/netfilter_ipv6/ip6t_IMQ.h linux-2.6.32-imq/include/linux/netfilter_ipv6/ip6t_IMQ.h
 849 --- linux-2.6.32/include/linux/netfilter_ipv6/ip6t_IMQ.h        1970-01-01 02:00:00.000000000 +0200
 850 +++ linux-2.6.32-imq/include/linux/netfilter_ipv6/ip6t_IMQ.h    2009-12-11 14:16:42.679730960 +0200
 851 @@ -0,0 +1,10 @@
 852 +#ifndef _IP6T_IMQ_H
 853 +#define _IP6T_IMQ_H
 854 +
 855 +/* Backwards compatibility for old userspace */
 856 +#include <linux/netfilter/xt_IMQ.h>
 857 +
 858 +#define ip6t_imq_info xt_imq_info
 859 +
 860 +#endif /* _IP6T_IMQ_H */
 861 +
 862 diff -U 5 -Nr linux-2.6.32/include/linux/skbuff.h linux-2.6.32-imq/include/linux/skbuff.h
 863 --- linux-2.6.32/include/linux/skbuff.h 2009-12-03 05:51:21.000000000 +0200
 864 +++ linux-2.6.32-imq/include/linux/skbuff.h     2009-12-11 14:16:42.680730834 +0200
 865 @@ -27,10 +27,13 @@
 866  #include <linux/textsearch.h>
 867  #include <net/checksum.h>
 868  #include <linux/rcupdate.h>
 869  #include <linux/dmaengine.h>
 870  #include <linux/hrtimer.h>
 871 +#if defined(CONFIG_IMQ) || defined(CONFIG_IMQ_MODULE)
 872 +#include <linux/imq.h>
 873 +#endif
 874
 875  /* Don't change this without changing skb_csum_unnecessary! */
 876  #define CHECKSUM_NONE 0
 877  #define CHECKSUM_UNNECESSARY 1
 878  #define CHECKSUM_COMPLETE 2
 879 @@ -328,10 +331,13 @@
 880          * layer. Please put your private variables there. If you
 881          * want to keep them across layers you have to do a skb_clone()
 882          * first. This is owned by whoever has the skb queued ATM.
 883          */
 884         char                    cb[48];
 885 +#if defined(CONFIG_IMQ) || defined(CONFIG_IMQ_MODULE)
 886 +       void                    *cb_next;
 887 +#endif
 888
 889         unsigned int            len,
 890                                 data_len;
 891         __u16                   mac_len,
 892                                 hdr_len;
 893 @@ -360,10 +366,13 @@
 894         void                    (*destructor)(struct sk_buff *skb);
 895  #if defined(CONFIG_NF_CONNTRACK) || defined(CONFIG_NF_CONNTRACK_MODULE)
 896         struct nf_conntrack     *nfct;
 897         struct sk_buff          *nfct_reasm;
 898  #endif
 899 +#if defined(CONFIG_IMQ) || defined(CONFIG_IMQ_MODULE)
 900 +       struct nf_queue_entry   *nf_queue_entry;
 901 +#endif
 902  #ifdef CONFIG_BRIDGE_NETFILTER
 903         struct nf_bridge_info   *nf_bridge;
 904  #endif
 905
 906         int                     iif;
 907 @@ -381,10 +390,14 @@
 908  #endif
 909         kmemcheck_bitfield_end(flags2);
 910
 911         /* 0/14 bit hole */
 912
 913 +#if defined(CONFIG_IMQ) || defined(CONFIG_IMQ_MODULE)
 914 +       __u8                    imq_flags:IMQ_F_BITS;
 915 +#endif
 916 +
 917  #ifdef CONFIG_NET_DMA
 918         dma_cookie_t            dma_cookie;
 919  #endif
 920  #ifdef CONFIG_NETWORK_SECMARK
 921         __u32                   secmark;
 922 @@ -435,10 +448,16 @@
 923  static inline struct rtable *skb_rtable(const struct sk_buff *skb)
 924  {
 925         return (struct rtable *)skb_dst(skb);
 926  }
 927
 928 +
 929 +#if defined(CONFIG_IMQ) || defined(CONFIG_IMQ_MODULE)
 930 +extern int skb_save_cb(struct sk_buff *skb);
 931 +extern int skb_restore_cb(struct sk_buff *skb);
 932 +#endif
 933 +
 934  extern void kfree_skb(struct sk_buff *skb);
 935  extern void consume_skb(struct sk_buff *skb);
 936  extern void           __kfree_skb(struct sk_buff *skb);
 937  extern struct sk_buff *__alloc_skb(unsigned int size,
 938                                    gfp_t priority, int fclone, int node);
 939 @@ -1970,10 +1989,14 @@
 940         nf_conntrack_get(src->nfct);
 941         dst->nfctinfo = src->nfctinfo;
 942         dst->nfct_reasm = src->nfct_reasm;
 943         nf_conntrack_get_reasm(src->nfct_reasm);
 944  #endif
 945 +#if defined(CONFIG_IMQ) || defined(CONFIG_IMQ_MODULE)
 946 +       dst->imq_flags = src->imq_flags;
 947 +       dst->nf_queue_entry = src->nf_queue_entry;
 948 +#endif
 949  #ifdef CONFIG_BRIDGE_NETFILTER
 950         dst->nf_bridge  = src->nf_bridge;
 951         nf_bridge_get(src->nf_bridge);
 952  #endif
 953  }
 954 diff -U 5 -Nr linux-2.6.32/include/net/netfilter/nf_queue.h linux-2.6.32-imq/include/net/netfilter/nf_queue.h
 955 --- linux-2.6.32/include/net/netfilter/nf_queue.h       2009-12-03 05:51:21.000000000 +0200
 956 +++ linux-2.6.32-imq/include/net/netfilter/nf_queue.h   2009-12-11 14:16:42.680730834 +0200
 957 @@ -11,10 +11,16 @@
 958         u_int8_t                pf;
 959         unsigned int            hook;
 960         struct net_device       *indev;
 961         struct net_device       *outdev;
 962         int                     (*okfn)(struct sk_buff *);
 963 +
 964 +#if defined(CONFIG_IMQ) || defined(CONFIG_IMQ_MODULE)
 965 +       int                     (*next_outfn)(struct nf_queue_entry *entry,
 966 +                                             unsigned int queuenum);
 967 +       unsigned int            next_queuenum;
 968 +#endif
 969  };
 970
 971  #define nf_queue_entry_reroute(x) ((void *)x + sizeof(struct nf_queue_entry))
 972
 973  /* Packet queuing */
 974 @@ -28,7 +34,13 @@
 975                                      const struct nf_queue_handler *qh);
 976  extern int nf_unregister_queue_handler(u_int8_t pf,
 977                                        const struct nf_queue_handler *qh);
 978  extern void nf_unregister_queue_handlers(const struct nf_queue_handler *qh);
 979  extern void nf_reinject(struct nf_queue_entry *entry, unsigned int verdict);
 980 +extern void nf_queue_entry_release_refs(struct nf_queue_entry *entry);
 981 +
 982 +#if defined(CONFIG_IMQ) || defined(CONFIG_IMQ_MODULE)
 983 +extern void nf_register_queue_imq_handler(const struct nf_queue_handler *qh);
 984 +extern void nf_unregister_queue_imq_handler(void);
 985 +#endif
 986
 987  #endif /* _NF_QUEUE_H */
 988 diff -U 5 -Nr linux-2.6.32/net/core/dev.c linux-2.6.32-imq/net/core/dev.c
 989 --- linux-2.6.32/net/core/dev.c 2009-12-03 05:51:21.000000000 +0200
 990 +++ linux-2.6.32-imq/net/core/dev.c     2009-12-11 14:16:42.681731014 +0200
 991 @@ -94,10 +94,13 @@
 992  #include <linux/notifier.h>
 993  #include <linux/skbuff.h>
 994  #include <net/net_namespace.h>
 995  #include <net/sock.h>
 996  #include <linux/rtnetlink.h>
 997 +#if defined(CONFIG_IMQ) || defined(CONFIG_IMQ_MODULE)
 998 +#include <linux/imq.h>
 999 +#endif
1000  #include <linux/proc_fs.h>
1001  #include <linux/seq_file.h>
1002  #include <linux/stat.h>
1003  #include <linux/if_bridge.h>
1004  #include <linux/if_macvlan.h>
1005 @@ -1702,11 +1705,15 @@
1006  {
1007         const struct net_device_ops *ops = dev->netdev_ops;
1008         int rc = NETDEV_TX_OK;
1009
1010         if (likely(!skb->next)) {
1011 -               if (!list_empty(&ptype_all))
1012 +               if (!list_empty(&ptype_all)
1013 +#if defined(CONFIG_IMQ) || defined(CONFIG_IMQ_MODULE)
1014 +                   && !(skb->imq_flags & IMQ_F_ENQUEUE)
1015 +#endif
1016 +                   )
1017                         dev_queue_xmit_nit(skb, dev);
1018
1019                 if (netif_needs_gso(dev, skb)) {
1020                         if (unlikely(dev_gso_segment(skb)))
1021                                 goto out_kfree_skb;
1022 @@ -1787,7 +1794,7 @@
1023         return queue_index;
1024  }
1025
1026 -static struct netdev_queue *dev_pick_tx(struct net_device *dev,
1027 +struct netdev_queue *dev_pick_tx(struct net_device *dev,
1028                                         struct sk_buff *skb)
1029  {
1030         u16 queue_index;
1031 @@ -1801,10 +1807,11 @@
1032                 queue_index = skb_tx_hash(dev, skb);
1033
1034         skb_set_queue_mapping(skb, queue_index);
1035         return netdev_get_tx_queue(dev, queue_index);
1036  }
1037 +EXPORT_SYMBOL(dev_pick_tx);
1038
1039  static inline int __dev_xmit_skb(struct sk_buff *skb, struct Qdisc *q,
1040                                  struct net_device *dev,
1041                                  struct netdev_queue *txq)
1042  {
1043 diff -U 5 -Nr linux-2.6.32/net/core/skbuff.c linux-2.6.32-imq/net/core/skbuff.c
1044 --- linux-2.6.32/net/core/skbuff.c      2009-12-03 05:51:21.000000000 +0200
1045 +++ linux-2.6.32-imq/net/core/skbuff.c  2009-12-11 15:12:39.294981618 +0200
1046 @@ -70,10 +70,13 @@
1047
1048  #include "kmap_skb.h"
1049
1050  static struct kmem_cache *skbuff_head_cache __read_mostly;
1051  static struct kmem_cache *skbuff_fclone_cache __read_mostly;
1052 +#if defined(CONFIG_IMQ) || defined(CONFIG_IMQ_MODULE)
1053 +static struct kmem_cache *skbuff_cb_store_cache __read_mostly;
1054 +#endif
1055
1056  static void sock_pipe_buf_release(struct pipe_inode_info *pipe,
1057                                   struct pipe_buffer *buf)
1058  {
1059         put_page(buf->page);
1060 @@ -89,10 +92,87 @@
1061                                struct pipe_buffer *buf)
1062  {
1063         return 1;
1064  }
1065
1066 +#if defined(CONFIG_IMQ) || defined(CONFIG_IMQ_MODULE)
1067 +/* Control buffer save/restore for IMQ devices */
1068 +struct skb_cb_table {
1069 +       void                    *cb_next;
1070 +       atomic_t                refcnt;
1071 +       char                    cb[48];
1072 +};
1073 +
1074 +static DEFINE_SPINLOCK(skb_cb_store_lock);
1075 +
1076 +int skb_save_cb(struct sk_buff *skb)
1077 +{
1078 +       struct skb_cb_table *next;
1079 +
1080 +       next = kmem_cache_alloc(skbuff_cb_store_cache, GFP_ATOMIC);
1081 +       if (!next)
1082 +               return -ENOMEM;
1083 +
1084 +       BUILD_BUG_ON(sizeof(skb->cb) != sizeof(next->cb));
1085 +
1086 +       memcpy(next->cb, skb->cb, sizeof(skb->cb));
1087 +       next->cb_next = skb->cb_next;
1088 +
1089 +       atomic_set(&next->refcnt, 1);
1090 +
1091 +       skb->cb_next = next;
1092 +       return 0;
1093 +}
1094 +EXPORT_SYMBOL(skb_save_cb);
1095 +
1096 +int skb_restore_cb(struct sk_buff *skb)
1097 +{
1098 +       struct skb_cb_table *next;
1099 +
1100 +       if (!skb->cb_next)
1101 +               return 0;
1102 +
1103 +       next = skb->cb_next;
1104 +
1105 +       BUILD_BUG_ON(sizeof(skb->cb) != sizeof(next->cb));
1106 +
1107 +       memcpy(skb->cb, next->cb, sizeof(skb->cb));
1108 +       skb->cb_next = next->cb_next;
1109 +
1110 +       spin_lock(&skb_cb_store_lock);
1111 +
1112 +       if (atomic_dec_and_test(&next->refcnt)) {
1113 +               kmem_cache_free(skbuff_cb_store_cache, next);
1114 +       }
1115 +
1116 +       spin_unlock(&skb_cb_store_lock);
1117 +
1118 +       return 0;
1119 +}
1120 +EXPORT_SYMBOL(skb_restore_cb);
1121 +
1122 +static void skb_copy_stored_cb(struct sk_buff *new, const struct sk_buff *__old)
1123 +{
1124 +       struct skb_cb_table *next;
1125 +       struct sk_buff *old;
1126 +
1127 +       if (!__old->cb_next) {
1128 +               new->cb_next = NULL;
1129 +               return;
1130 +       }
1131 +
1132 +       spin_lock(&skb_cb_store_lock);
1133 +
1134 +       old = (struct sk_buff *)__old;
1135 +
1136 +       next = old->cb_next;
1137 +       atomic_inc(&next->refcnt);
1138 +       new->cb_next = next;
1139 +
1140 +       spin_unlock(&skb_cb_store_lock);
1141 +}
1142 +#endif
1143
1144  /* Pipe buffer operations for a socket. */
1145  static const struct pipe_buf_operations sock_pipe_buf_ops = {
1146         .can_merge = 0,
1147         .map = generic_pipe_buf_map,
1148 @@ -396,10 +476,30 @@
1149  #endif
1150         if (skb->destructor) {
1151                 WARN_ON(in_irq());
1152                 skb->destructor(skb);
1153         }
1154 +#if defined(CONFIG_IMQ) || defined(CONFIG_IMQ_MODULE)
1155 +       /* This should not happen. When it does, avoid memleak by restoring
1156 +       the chain of cb-backups. */
1157 +       while(skb->cb_next != NULL) {
1158 +               if (net_ratelimit())
1159 +                       printk(KERN_WARNING "IMQ: kfree_skb: skb->cb_next: "
1160 +                               "%08x\n", (unsigned int)skb->cb_next);
1161 +
1162 +               skb_restore_cb(skb);
1163 +       }
1164 +       /* This should not happen either, nf_queue_entry is nullified in
1165 +        * imq_dev_xmit(). If we have non-NULL nf_queue_entry then we are
1166 +        * leaking entry pointers, maybe memory. We don't know if this is
1167 +        * pointer to already freed memory, or should this be freed.
1168 +        * If this happens we need to add refcounting, etc for nf_queue_entry.
1169 +        */
1170 +       if (skb->nf_queue_entry && net_ratelimit())
1171 +               printk(KERN_WARNING
1172 +                               "IMQ: kfree_skb: skb->nf_queue_entry != NULL");
1173 +#endif
1174  #if defined(CONFIG_NF_CONNTRACK) || defined(CONFIG_NF_CONNTRACK_MODULE)
1175         nf_conntrack_put(skb->nfct);
1176         nf_conntrack_put_reasm(skb->nfct_reasm);
1177  #endif
1178  #ifdef CONFIG_BRIDGE_NETFILTER
1179 @@ -533,10 +633,13 @@
1180         skb_dst_set(new, dst_clone(skb_dst(old)));
1181  #ifdef CONFIG_XFRM
1182         new->sp                 = secpath_get(old->sp);
1183  #endif
1184         memcpy(new->cb, old->cb, sizeof(old->cb));
1185 +#if defined(CONFIG_IMQ) || defined(CONFIG_IMQ_MODULE)
1186 +       skb_copy_stored_cb(new, old);
1187 +#endif
1188         new->csum               = old->csum;
1189         new->local_df           = old->local_df;
1190         new->pkt_type           = old->pkt_type;
1191         new->ip_summed          = old->ip_summed;
1192         skb_copy_queue_mapping(new, old);
1193 @@ -2774,10 +2877,17 @@
1194                                                 (2*sizeof(struct sk_buff)) +
1195                                                 sizeof(atomic_t),
1196                                                 0,
1197                                                 SLAB_HWCACHE_ALIGN|SLAB_PANIC,
1198                                                 NULL);
1199 +#if defined(CONFIG_IMQ) || defined(CONFIG_IMQ_MODULE)
1200 +       skbuff_cb_store_cache = kmem_cache_create("skbuff_cb_store_cache",
1201 +                                                 sizeof(struct skb_cb_table),
1202 +                                                 0,
1203 +                                                 SLAB_HWCACHE_ALIGN|SLAB_PANIC,
1204 +                                                 NULL);
1205 +#endif
1206  }
1207
1208  /**
1209   *     skb_to_sgvec - Fill a scatter-gather list from a socket buffer
1210   *     @skb: Socket buffer containing the buffers to be mapped
1211 diff -U 5 -Nr linux-2.6.32/net/netfilter/Kconfig linux-2.6.32-imq/net/netfilter/Kconfig
1212 --- linux-2.6.32/net/netfilter/Kconfig  2009-12-03 05:51:21.000000000 +0200
1213 +++ linux-2.6.32-imq/net/netfilter/Kconfig      2009-12-11 14:16:42.681731014 +0200
1214 @@ -394,10 +394,22 @@
1215             echo netfilter-ssh > /sys/class/leds/<ledname>/trigger
1216
1217           For more information on the LEDs available on your system, see
1218           Documentation/leds-class.txt
1219
1220 +config NETFILTER_XT_TARGET_IMQ
1221 +        tristate '"IMQ" target support'
1222 +       depends on NETFILTER_XTABLES
1223 +       depends on IP_NF_MANGLE || IP6_NF_MANGLE
1224 +       select IMQ
1225 +       default m if NETFILTER_ADVANCED=n
1226 +        help
1227 +          This option adds a `IMQ' target which is used to specify if and
1228 +          to which imq device packets should get enqueued/dequeued.
1229 +
1230 +          To compile it as a module, choose M here.  If unsure, say N.
1231 +
1232  config NETFILTER_XT_TARGET_MARK
1233         tristate '"MARK" target support'
1234         default m if NETFILTER_ADVANCED=n
1235         help
1236           This option adds a `MARK' target, which allows you to create rules
1237 diff -U 5 -Nr linux-2.6.32/net/netfilter/Makefile linux-2.6.32-imq/net/netfilter/Makefile
1238 --- linux-2.6.32/net/netfilter/Makefile 2009-12-03 05:51:21.000000000 +0200
1239 +++ linux-2.6.32-imq/net/netfilter/Makefile     2009-12-11 14:16:42.681731014 +0200
1240 @@ -44,10 +44,11 @@
1241  obj-$(CONFIG_NETFILTER_XT_TARGET_CLASSIFY) += xt_CLASSIFY.o
1242  obj-$(CONFIG_NETFILTER_XT_TARGET_CONNMARK) += xt_CONNMARK.o
1243  obj-$(CONFIG_NETFILTER_XT_TARGET_CONNSECMARK) += xt_CONNSECMARK.o
1244  obj-$(CONFIG_NETFILTER_XT_TARGET_DSCP) += xt_DSCP.o
1245  obj-$(CONFIG_NETFILTER_XT_TARGET_HL) += xt_HL.o
1246 +obj-$(CONFIG_NETFILTER_XT_TARGET_IMQ) += xt_IMQ.o
1247  obj-$(CONFIG_NETFILTER_XT_TARGET_LED) += xt_LED.o
1248  obj-$(CONFIG_NETFILTER_XT_TARGET_MARK) += xt_MARK.o
1249  obj-$(CONFIG_NETFILTER_XT_TARGET_NFLOG) += xt_NFLOG.o
1250  obj-$(CONFIG_NETFILTER_XT_TARGET_NFQUEUE) += xt_NFQUEUE.o
1251  obj-$(CONFIG_NETFILTER_XT_TARGET_NOTRACK) += xt_NOTRACK.o
1252 diff -U 5 -Nr linux-2.6.32/net/netfilter/nf_queue.c linux-2.6.32-imq/net/netfilter/nf_queue.c
1253 --- linux-2.6.32/net/netfilter/nf_queue.c       2009-12-03 05:51:21.000000000 +0200
1254 +++ linux-2.6.32-imq/net/netfilter/nf_queue.c   2009-12-11 14:16:42.681731014 +0200
1255 @@ -18,10 +18,30 @@
1256   */
1257  static const struct nf_queue_handler *queue_handler[NFPROTO_NUMPROTO] __read_mostly;
1258
1259  static DEFINE_MUTEX(queue_handler_mutex);
1260
1261 +#if defined(CONFIG_IMQ) || defined(CONFIG_IMQ_MODULE)
1262 +static const struct nf_queue_handler *queue_imq_handler;
1263 +
1264 +void nf_register_queue_imq_handler(const struct nf_queue_handler *qh)
1265 +{
1266 +       mutex_lock(&queue_handler_mutex);
1267 +       rcu_assign_pointer(queue_imq_handler, qh);
1268 +       mutex_unlock(&queue_handler_mutex);
1269 +}
1270 +EXPORT_SYMBOL(nf_register_queue_imq_handler);
1271 +
1272 +void nf_unregister_queue_imq_handler(void)
1273 +{
1274 +       mutex_lock(&queue_handler_mutex);
1275 +       rcu_assign_pointer(queue_imq_handler, NULL);
1276 +       mutex_unlock(&queue_handler_mutex);
1277 +}
1278 +EXPORT_SYMBOL(nf_unregister_queue_imq_handler);
1279 +#endif
1280 +
1281  /* return EBUSY when somebody else is registered, return EEXIST if the
1282   * same handler is registered, return 0 in case of success. */
1283  int nf_register_queue_handler(u_int8_t pf, const struct nf_queue_handler *qh)
1284  {
1285         int ret;
1286 @@ -78,11 +98,11 @@
1287
1288         synchronize_rcu();
1289  }
1290  EXPORT_SYMBOL_GPL(nf_unregister_queue_handlers);
1291
1292 -static void nf_queue_entry_release_refs(struct nf_queue_entry *entry)
1293 +void nf_queue_entry_release_refs(struct nf_queue_entry *entry)
1294  {
1295         /* Release those devices we held, or Alexey will kill me. */
1296         if (entry->indev)
1297                 dev_put(entry->indev);
1298         if (entry->outdev)
1299 @@ -98,10 +118,11 @@
1300         }
1301  #endif
1302         /* Drop reference to owner of hook which queued us. */
1303         module_put(entry->elem->owner);
1304  }
1305 +EXPORT_SYMBOL_GPL(nf_queue_entry_release_refs);
1306
1307  /*
1308   * Any packet that leaves via this function must come back
1309   * through nf_reinject().
1310   */
1311 @@ -119,16 +140,30 @@
1312         struct net_device *physindev;
1313         struct net_device *physoutdev;
1314  #endif
1315         const struct nf_afinfo *afinfo;
1316         const struct nf_queue_handler *qh;
1317 +#if defined(CONFIG_IMQ) || defined(CONFIG_IMQ_MODULE)
1318 +       const struct nf_queue_handler *qih = NULL;
1319 +#endif
1320
1321         /* QUEUE == DROP if noone is waiting, to be safe. */
1322         rcu_read_lock();
1323
1324         qh = rcu_dereference(queue_handler[pf]);
1325 +#if defined(CONFIG_IMQ) || defined(CONFIG_IMQ_MODULE)
1326 +#if defined(CONFIG_IPV6) || defined(CONFIG_IPV6_MODULE)
1327 +       if (pf == PF_INET || pf == PF_INET6)
1328 +#else
1329 +       if (pf == PF_INET)
1330 +#endif
1331 +               qih = rcu_dereference(queue_imq_handler);
1332 +
1333 +       if (!qh && !qih)
1334 +#else /* !IMQ */
1335         if (!qh)
1336 +#endif
1337                 goto err_unlock;
1338
1339         afinfo = nf_get_afinfo(pf);
1340         if (!afinfo)
1341                 goto err_unlock;
1342 @@ -143,10 +178,14 @@
1343                 .pf     = pf,
1344                 .hook   = hook,
1345                 .indev  = indev,
1346                 .outdev = outdev,
1347                 .okfn   = okfn,
1348 +#if defined(CONFIG_IMQ) || defined(CONFIG_IMQ_MODULE)
1349 +               .next_outfn = qh ? qh->outfn : NULL,
1350 +               .next_queuenum = queuenum,
1351 +#endif
1352         };
1353
1354         /* If it's going away, ignore hook. */
1355         if (!try_module_get(entry->elem->owner)) {
1356                 rcu_read_unlock();
1357 @@ -168,12 +207,23 @@
1358                 if (physoutdev)
1359                         dev_hold(physoutdev);
1360         }
1361  #endif
1362         afinfo->saveroute(skb, entry);
1363 +
1364 +#if defined(CONFIG_IMQ) || defined(CONFIG_IMQ_MODULE)
1365 +       if (qih) {
1366 +               status = qih->outfn(entry, queuenum);
1367 +               goto imq_skip_queue;
1368 +       }
1369 +#endif
1370 +
1371         status = qh->outfn(entry, queuenum);
1372
1373 +#if defined(CONFIG_IMQ) || defined(CONFIG_IMQ_MODULE)
1374 +imq_skip_queue:
1375 +#endif
1376         rcu_read_unlock();
1377
1378         if (status < 0) {
1379                 nf_queue_entry_release_refs(entry);
1380                 goto err;
1381 diff -U 5 -Nr linux-2.6.32/net/netfilter/xt_IMQ.c linux-2.6.32-imq/net/netfilter/xt_IMQ.c
1382 --- linux-2.6.32/net/netfilter/xt_IMQ.c 1970-01-01 02:00:00.000000000 +0200
1383 +++ linux-2.6.32-imq/net/netfilter/xt_IMQ.c     2009-12-11 14:16:42.681731014 +0200
1384 @@ -0,0 +1,73 @@
1385 +/*
1386 + * This target marks packets to be enqueued to an imq device
1387 + */
1388 +#include <linux/module.h>
1389 +#include <linux/skbuff.h>
1390 +#include <linux/netfilter/x_tables.h>
1391 +#include <linux/netfilter/xt_IMQ.h>
1392 +#include <linux/imq.h>
1393 +
1394 +static unsigned int imq_target(struct sk_buff *pskb,
1395 +                               const struct xt_target_param *par)
1396 +{
1397 +       const struct xt_imq_info *mr = par->targinfo;
1398 +
1399 +       pskb->imq_flags = (mr->todev & IMQ_F_IFMASK) | IMQ_F_ENQUEUE;
1400 +
1401 +       return XT_CONTINUE;
1402 +}
1403 +
1404 +static bool imq_checkentry(const struct xt_tgchk_param *par)
1405 +{
1406 +       struct xt_imq_info *mr = par->targinfo;
1407 +
1408 +       if (mr->todev > IMQ_MAX_DEVS - 1) {
1409 +               printk(KERN_WARNING
1410 +                      "IMQ: invalid device specified, highest is %u\n",
1411 +                      IMQ_MAX_DEVS - 1);
1412 +               return 0;
1413 +       }
1414 +
1415 +       return 1;
1416 +}
1417 +
1418 +static struct xt_target xt_imq_reg[] __read_mostly = {
1419 +       {
1420 +               .name           = "IMQ",
1421 +               .family         = AF_INET,
1422 +               .checkentry     = imq_checkentry,
1423 +               .target         = imq_target,
1424 +               .targetsize     = sizeof(struct xt_imq_info),
1425 +               .table          = "mangle",
1426 +               .me             = THIS_MODULE
1427 +       },
1428 +       {
1429 +               .name           = "IMQ",
1430 +               .family         = AF_INET6,
1431 +               .checkentry     = imq_checkentry,
1432 +               .target         = imq_target,
1433 +               .targetsize     = sizeof(struct xt_imq_info),
1434 +               .table          = "mangle",
1435 +               .me             = THIS_MODULE
1436 +       },
1437 +};
1438 +
1439 +static int __init imq_init(void)
1440 +{
1441 +       return xt_register_targets(xt_imq_reg, ARRAY_SIZE(xt_imq_reg));
1442 +}
1443 +
1444 +static void __exit imq_fini(void)
1445 +{
1446 +       xt_unregister_targets(xt_imq_reg, ARRAY_SIZE(xt_imq_reg));
1447 +}
1448 +
1449 +module_init(imq_init);
1450 +module_exit(imq_fini);
1451 +
1452 +MODULE_AUTHOR("http://www.linuximq.net");
1453 +MODULE_DESCRIPTION("Pseudo-driver for the intermediate queue device. See http://www.linuximq.net/ for more information.");
1454 +MODULE_LICENSE("GPL");
1455 +MODULE_ALIAS("ipt_IMQ");
1456 +MODULE_ALIAS("ip6t_IMQ");
1457 +