kernel-imq.patch

   1 diff -uNr linux-3.0/drivers/net/imq.c linux-3.0-imq/drivers/net/imq.c
   2 --- linux-3.0/drivers/net/imq.c 1970-01-01 02:00:00.000000000 +0200
   3 +++ linux-3.0-imq/drivers/net/imq.c     2011-07-26 07:24:09.843279145 +0300
   4 @@ -0,0 +1,820 @@
   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 + *            Also, many thanks to pablo Sebastian Greco for making the initial
 118 + *            patch and to those who helped the testing.
 119 + *
 120 + *             More info at: http://www.linuximq.net/ (Andre Correa)
 121 + */
 122 +
 123 +#include <linux/module.h>
 124 +#include <linux/kernel.h>
 125 +#include <linux/moduleparam.h>
 126 +#include <linux/list.h>
 127 +#include <linux/skbuff.h>
 128 +#include <linux/netdevice.h>
 129 +#include <linux/etherdevice.h>
 130 +#include <linux/rtnetlink.h>
 131 +#include <linux/if_arp.h>
 132 +#include <linux/netfilter.h>
 133 +#include <linux/netfilter_ipv4.h>
 134 +#if defined(CONFIG_IPV6) || defined(CONFIG_IPV6_MODULE)
 135 +       #include <linux/netfilter_ipv6.h>
 136 +#endif
 137 +#include <linux/imq.h>
 138 +#include <net/pkt_sched.h>
 139 +#include <net/netfilter/nf_queue.h>
 140 +#include <net/sock.h>
 141 +#include <linux/ip.h>
 142 +#include <linux/ipv6.h>
 143 +#include <linux/if_vlan.h>
 144 +#include <linux/if_pppox.h>
 145 +#include <net/ip.h>
 146 +#include <net/ipv6.h>
 147 +
 148 +static int imq_nf_queue(struct nf_queue_entry *entry, unsigned queue_num);
 149 +
 150 +static nf_hookfn imq_nf_hook;
 151 +
 152 +static struct nf_hook_ops imq_ops[] = {
 153 +       {
 154 +       /* imq_ingress_ipv4 */
 155 +               .hook           = imq_nf_hook,
 156 +               .owner          = THIS_MODULE,
 157 +               .pf             = PF_INET,
 158 +               .hooknum        = NF_INET_PRE_ROUTING,
 159 +#if defined(CONFIG_IMQ_BEHAVIOR_BA) || defined(CONFIG_IMQ_BEHAVIOR_BB)
 160 +               .priority       = NF_IP_PRI_MANGLE + 1,
 161 +#else
 162 +               .priority       = NF_IP_PRI_NAT_DST + 1,
 163 +#endif
 164 +       },
 165 +       {
 166 +       /* imq_egress_ipv4 */
 167 +               .hook           = imq_nf_hook,
 168 +               .owner          = THIS_MODULE,
 169 +               .pf             = PF_INET,
 170 +               .hooknum        = NF_INET_POST_ROUTING,
 171 +#if defined(CONFIG_IMQ_BEHAVIOR_AA) || defined(CONFIG_IMQ_BEHAVIOR_BA)
 172 +               .priority       = NF_IP_PRI_LAST,
 173 +#else
 174 +               .priority       = NF_IP_PRI_NAT_SRC - 1,
 175 +#endif
 176 +       },
 177 +#if defined(CONFIG_IPV6) || defined(CONFIG_IPV6_MODULE)
 178 +       {
 179 +       /* imq_ingress_ipv6 */
 180 +               .hook           = imq_nf_hook,
 181 +               .owner          = THIS_MODULE,
 182 +               .pf             = PF_INET6,
 183 +               .hooknum        = NF_INET_PRE_ROUTING,
 184 +#if defined(CONFIG_IMQ_BEHAVIOR_BA) || defined(CONFIG_IMQ_BEHAVIOR_BB)
 185 +               .priority       = NF_IP6_PRI_MANGLE + 1,
 186 +#else
 187 +               .priority       = NF_IP6_PRI_NAT_DST + 1,
 188 +#endif
 189 +       },
 190 +       {
 191 +       /* imq_egress_ipv6 */
 192 +               .hook           = imq_nf_hook,
 193 +               .owner          = THIS_MODULE,
 194 +               .pf             = PF_INET6,
 195 +               .hooknum        = NF_INET_POST_ROUTING,
 196 +#if defined(CONFIG_IMQ_BEHAVIOR_AA) || defined(CONFIG_IMQ_BEHAVIOR_BA)
 197 +               .priority       = NF_IP6_PRI_LAST,
 198 +#else
 199 +               .priority       = NF_IP6_PRI_NAT_SRC - 1,
 200 +#endif
 201 +       },
 202 +#endif
 203 +};
 204 +
 205 +#if defined(CONFIG_IMQ_NUM_DEVS)
 206 +static int numdevs = CONFIG_IMQ_NUM_DEVS;
 207 +#else
 208 +static int numdevs = IMQ_MAX_DEVS;
 209 +#endif
 210 +
 211 +static struct net_device *imq_devs_cache[IMQ_MAX_DEVS];
 212 +
 213 +#define IMQ_MAX_QUEUES 32
 214 +static int numqueues = 1;
 215 +static u32 imq_hashrnd;
 216 +
 217 +static inline __be16 pppoe_proto(const struct sk_buff *skb)
 218 +{
 219 +       return *((__be16 *)(skb_mac_header(skb) + ETH_HLEN +
 220 +                       sizeof(struct pppoe_hdr)));
 221 +}
 222 +
 223 +static u16 imq_hash(struct net_device *dev, struct sk_buff *skb)
 224 +{
 225 +       unsigned int pull_len;
 226 +       u16 protocol = skb->protocol;
 227 +       u32 addr1, addr2;
 228 +       u32 hash, ihl = 0;
 229 +       union {
 230 +               u16 in16[2];
 231 +               u32 in32;
 232 +       } ports;
 233 +       u8 ip_proto;
 234 +
 235 +       pull_len = 0;
 236 +
 237 +recheck:
 238 +       switch (protocol) {
 239 +       case htons(ETH_P_8021Q): {
 240 +               if (unlikely(skb_pull(skb, VLAN_HLEN) == NULL))
 241 +                       goto other;
 242 +
 243 +               pull_len += VLAN_HLEN;
 244 +               skb->network_header += VLAN_HLEN;
 245 +
 246 +               protocol = vlan_eth_hdr(skb)->h_vlan_encapsulated_proto;
 247 +               goto recheck;
 248 +       }
 249 +
 250 +       case htons(ETH_P_PPP_SES): {
 251 +               if (unlikely(skb_pull(skb, PPPOE_SES_HLEN) == NULL))
 252 +                       goto other;
 253 +
 254 +               pull_len += PPPOE_SES_HLEN;
 255 +               skb->network_header += PPPOE_SES_HLEN;
 256 +
 257 +               protocol = pppoe_proto(skb);
 258 +               goto recheck;
 259 +       }
 260 +
 261 +       case htons(ETH_P_IP): {
 262 +               const struct iphdr *iph = ip_hdr(skb);
 263 +
 264 +               if (unlikely(!pskb_may_pull(skb, sizeof(struct iphdr))))
 265 +                       goto other;
 266 +
 267 +               addr1 = iph->daddr;
 268 +               addr2 = iph->saddr;
 269 +
 270 +               ip_proto = !(ip_hdr(skb)->frag_off & htons(IP_MF | IP_OFFSET)) ?
 271 +                                iph->protocol : 0;
 272 +               ihl = ip_hdrlen(skb);
 273 +
 274 +               break;
 275 +       }
 276 +#if defined(CONFIG_IPV6) || defined(CONFIG_IPV6_MODULE)
 277 +       case htons(ETH_P_IPV6): {
 278 +               const struct ipv6hdr *iph = ipv6_hdr(skb);
 279 +
 280 +               if (unlikely(!pskb_may_pull(skb, sizeof(struct ipv6hdr))))
 281 +                       goto other;
 282 +
 283 +               addr1 = iph->daddr.s6_addr32[3];
 284 +               addr2 = iph->saddr.s6_addr32[3];
 285 +               ihl = ipv6_skip_exthdr(skb, sizeof(struct ipv6hdr), &ip_proto);
 286 +               if (unlikely(ihl < 0))
 287 +                       goto other;
 288 +
 289 +               break;
 290 +       }
 291 +#endif
 292 +       default:
 293 +other:
 294 +               if (pull_len != 0) {
 295 +                       skb_push(skb, pull_len);
 296 +                       skb->network_header -= pull_len;
 297 +               }
 298 +
 299 +               return (u16)(ntohs(protocol) % dev->real_num_tx_queues);
 300 +       }
 301 +
 302 +       if (addr1 > addr2)
 303 +               swap(addr1, addr2);
 304 +
 305 +       switch (ip_proto) {
 306 +       case IPPROTO_TCP:
 307 +       case IPPROTO_UDP:
 308 +       case IPPROTO_DCCP:
 309 +       case IPPROTO_ESP:
 310 +       case IPPROTO_AH:
 311 +       case IPPROTO_SCTP:
 312 +       case IPPROTO_UDPLITE: {
 313 +               if (likely(skb_copy_bits(skb, ihl, &ports.in32, 4) >= 0)) {
 314 +                       if (ports.in16[0] > ports.in16[1])
 315 +                               swap(ports.in16[0], ports.in16[1]);
 316 +                       break;
 317 +               }
 318 +               /* fall-through */
 319 +       }
 320 +       default:
 321 +               ports.in32 = 0;
 322 +               break;
 323 +       }
 324 +
 325 +       if (pull_len != 0) {
 326 +               skb_push(skb, pull_len);
 327 +               skb->network_header -= pull_len;
 328 +       }
 329 +
 330 +       hash = jhash_3words(addr1, addr2, ports.in32, imq_hashrnd ^ ip_proto);
 331 +
 332 +       return (u16)(((u64)hash * dev->real_num_tx_queues) >> 32);
 333 +}
 334 +
 335 +static inline bool sk_tx_queue_recorded(struct sock *sk)
 336 +{
 337 +       return (sk_tx_queue_get(sk) >= 0);
 338 +}
 339 +
 340 +static struct netdev_queue *imq_select_queue(struct net_device *dev,
 341 +                                               struct sk_buff *skb)
 342 +{
 343 +       u16 queue_index = 0;
 344 +       u32 hash;
 345 +
 346 +       if (likely(dev->real_num_tx_queues == 1))
 347 +               goto out;
 348 +
 349 +       /* IMQ can be receiving ingress or engress packets. */
 350 +
 351 +       /* Check first for if rx_queue is set */
 352 +       if (skb_rx_queue_recorded(skb)) {
 353 +               queue_index = skb_get_rx_queue(skb);
 354 +               goto out;
 355 +       }
 356 +
 357 +       /* Check if socket has tx_queue set */
 358 +       if (sk_tx_queue_recorded(skb->sk)) {
 359 +               queue_index = sk_tx_queue_get(skb->sk);
 360 +               goto out;
 361 +       }
 362 +
 363 +       /* Try use socket hash */
 364 +       if (skb->sk && skb->sk->sk_hash) {
 365 +               hash = skb->sk->sk_hash;
 366 +               queue_index =
 367 +                       (u16)(((u64)hash * dev->real_num_tx_queues) >> 32);
 368 +               goto out;
 369 +       }
 370 +
 371 +       /* Generate hash from packet data */
 372 +       queue_index = imq_hash(dev, skb);
 373 +
 374 +out:
 375 +       if (unlikely(queue_index >= dev->real_num_tx_queues))
 376 +               queue_index = (u16)((u32)queue_index % dev->real_num_tx_queues);
 377 +
 378 +       skb_set_queue_mapping(skb, queue_index);
 379 +       return netdev_get_tx_queue(dev, queue_index);
 380 +}
 381 +
 382 +static struct net_device_stats *imq_get_stats(struct net_device *dev)
 383 +{
 384 +       return &dev->stats;
 385 +}
 386 +
 387 +/* called for packets kfree'd in qdiscs at places other than enqueue */
 388 +static void imq_skb_destructor(struct sk_buff *skb)
 389 +{
 390 +       struct nf_queue_entry *entry = skb->nf_queue_entry;
 391 +
 392 +       skb->nf_queue_entry = NULL;
 393 +
 394 +       if (entry) {
 395 +               nf_queue_entry_release_refs(entry);
 396 +               kfree(entry);
 397 +       }
 398 +
 399 +       skb_restore_cb(skb); /* kfree backup */
 400 +}
 401 +
 402 +static void imq_done_check_queue_mapping(struct sk_buff *skb,
 403 +                                        struct net_device *dev)
 404 +{
 405 +       unsigned int queue_index;
 406 +
 407 +       /* Don't let queue_mapping be left too large after exiting IMQ */
 408 +       if (likely(skb->dev != dev && skb->dev != NULL)) {
 409 +               queue_index = skb_get_queue_mapping(skb);
 410 +               if (unlikely(queue_index >= skb->dev->real_num_tx_queues)) {
 411 +                       queue_index = (u16)((u32)queue_index %
 412 +                                               skb->dev->real_num_tx_queues);
 413 +                       skb_set_queue_mapping(skb, queue_index);
 414 +               }
 415 +       } else {
 416 +               /* skb->dev was IMQ device itself or NULL, be on safe side and
 417 +                * just clear queue mapping.
 418 +                */
 419 +               skb_set_queue_mapping(skb, 0);
 420 +       }
 421 +}
 422 +
 423 +static netdev_tx_t imq_dev_xmit(struct sk_buff *skb, struct net_device *dev)
 424 +{
 425 +       struct nf_queue_entry *entry = skb->nf_queue_entry;
 426 +
 427 +       skb->nf_queue_entry = NULL;
 428 +       dev->trans_start = jiffies;
 429 +
 430 +       dev->stats.tx_bytes += skb->len;
 431 +       dev->stats.tx_packets++;
 432 +
 433 +       if (unlikely(entry == NULL)) {
 434 +               /* We don't know what is going on here.. packet is queued for
 435 +                * imq device, but (probably) not by us.
 436 +                *
 437 +                * If this packet was not send here by imq_nf_queue(), then
 438 +                * skb_save_cb() was not used and skb_free() should not show:
 439 +                *   WARNING: IMQ: kfree_skb: skb->cb_next:..
 440 +                * and/or
 441 +                *   WARNING: IMQ: kfree_skb: skb->nf_queue_entry...
 442 +                *
 443 +                * However if this message is shown, then IMQ is somehow broken
 444 +                * and you should report this to linuximq.net.
 445 +                */
 446 +
 447 +               /* imq_dev_xmit is black hole that eats all packets, report that
 448 +                * we eat this packet happily and increase dropped counters.
 449 +                */
 450 +
 451 +               dev->stats.tx_dropped++;
 452 +               dev_kfree_skb(skb);
 453 +
 454 +               return NETDEV_TX_OK;
 455 +       }
 456 +
 457 +       skb_restore_cb(skb); /* restore skb->cb */
 458 +
 459 +       skb->imq_flags = 0;
 460 +       skb->destructor = NULL;
 461 +
 462 +       imq_done_check_queue_mapping(skb, dev);
 463 +
 464 +       nf_reinject(entry, NF_ACCEPT);
 465 +
 466 +       return NETDEV_TX_OK;
 467 +}
 468 +
 469 +static int imq_nf_queue(struct nf_queue_entry *entry, unsigned queue_num)
 470 +{
 471 +       struct net_device *dev;
 472 +       struct sk_buff *skb_orig, *skb, *skb_shared;
 473 +       struct Qdisc *q;
 474 +       struct netdev_queue *txq;
 475 +       spinlock_t *root_lock;
 476 +       int users, index;
 477 +       int retval = -EINVAL;
 478 +       unsigned int orig_queue_index;
 479 +
 480 +       index = entry->skb->imq_flags & IMQ_F_IFMASK;
 481 +       if (unlikely(index > numdevs - 1)) {
 482 +               if (net_ratelimit())
 483 +                       printk(KERN_WARNING
 484 +                              "IMQ: invalid device specified, highest is %u\n",
 485 +                              numdevs - 1);
 486 +               retval = -EINVAL;
 487 +               goto out;
 488 +       }
 489 +
 490 +       /* check for imq device by index from cache */
 491 +       dev = imq_devs_cache[index];
 492 +       if (unlikely(!dev)) {
 493 +               char buf[8];
 494 +
 495 +               /* get device by name and cache result */
 496 +               snprintf(buf, sizeof(buf), "imq%d", index);
 497 +               dev = dev_get_by_name(&init_net, buf);
 498 +               if (unlikely(!dev)) {
 499 +                       /* not found ?!*/
 500 +                       BUG();
 501 +                       retval = -ENODEV;
 502 +                       goto out;
 503 +               }
 504 +
 505 +               imq_devs_cache[index] = dev;
 506 +               dev_put(dev);
 507 +       }
 508 +
 509 +       if (unlikely(!(dev->flags & IFF_UP))) {
 510 +               entry->skb->imq_flags = 0;
 511 +               nf_reinject(entry, NF_ACCEPT);
 512 +               retval = 0;
 513 +               goto out;
 514 +       }
 515 +       dev->last_rx = jiffies;
 516 +
 517 +       skb = entry->skb;
 518 +       skb_orig = NULL;
 519 +
 520 +       /* skb has owner? => make clone */
 521 +       if (unlikely(skb->destructor)) {
 522 +               skb_orig = skb;
 523 +               skb = skb_clone(skb, GFP_ATOMIC);
 524 +               if (unlikely(!skb)) {
 525 +                       retval = -ENOMEM;
 526 +                       goto out;
 527 +               }
 528 +               entry->skb = skb;
 529 +       }
 530 +
 531 +       skb->nf_queue_entry = entry;
 532 +
 533 +       dev->stats.rx_bytes += skb->len;
 534 +       dev->stats.rx_packets++;
 535 +
 536 +       if (!skb->dev) {
 537 +               /* skb->dev == NULL causes problems, try the find cause. */
 538 +               if (net_ratelimit()) {
 539 +                       dev_warn(&dev->dev,
 540 +                                "received packet with skb->dev == NULL\n");
 541 +                       dump_stack();
 542 +               }
 543 +
 544 +               skb->dev = dev;
 545 +       }
 546 +
 547 +       /* Disables softirqs for lock below */
 548 +       rcu_read_lock_bh();
 549 +
 550 +       /* Multi-queue selection */
 551 +       orig_queue_index = skb_get_queue_mapping(skb);
 552 +       txq = imq_select_queue(dev, skb);
 553 +
 554 +       q = rcu_dereference(txq->qdisc);
 555 +       if (unlikely(!q->enqueue))
 556 +               goto packet_not_eaten_by_imq_dev;
 557 +
 558 +       root_lock = qdisc_lock(q);
 559 +       spin_lock(root_lock);
 560 +
 561 +       users = atomic_read(&skb->users);
 562 +
 563 +       skb_shared = skb_get(skb); /* increase reference count by one */
 564 +       skb_save_cb(skb_shared); /* backup skb->cb, as qdisc layer will
 565 +                                       overwrite it */
 566 +       qdisc_enqueue_root(skb_shared, q); /* might kfree_skb */
 567 +
 568 +       if (likely(atomic_read(&skb_shared->users) == users + 1)) {
 569 +               kfree_skb(skb_shared); /* decrease reference count by one */
 570 +
 571 +               skb->destructor = &imq_skb_destructor;
 572 +
 573 +               /* cloned? */
 574 +               if (unlikely(skb_orig))
 575 +                       kfree_skb(skb_orig); /* free original */
 576 +
 577 +               spin_unlock(root_lock);
 578 +               rcu_read_unlock_bh();
 579 +
 580 +               /* schedule qdisc dequeue */
 581 +               __netif_schedule(q);
 582 +
 583 +               retval = 0;
 584 +               goto out;
 585 +       } else {
 586 +               skb_restore_cb(skb_shared); /* restore skb->cb */
 587 +               skb->nf_queue_entry = NULL;
 588 +               /* qdisc dropped packet and decreased skb reference count of
 589 +                * skb, so we don't really want to and try refree as that would
 590 +                * actually destroy the skb. */
 591 +               spin_unlock(root_lock);
 592 +               goto packet_not_eaten_by_imq_dev;
 593 +       }
 594 +
 595 +packet_not_eaten_by_imq_dev:
 596 +       skb_set_queue_mapping(skb, orig_queue_index);
 597 +       rcu_read_unlock_bh();
 598 +
 599 +       /* cloned? restore original */
 600 +       if (unlikely(skb_orig)) {
 601 +               kfree_skb(skb);
 602 +               entry->skb = skb_orig;
 603 +       }
 604 +       retval = -1;
 605 +out:
 606 +       return retval;
 607 +}
 608 +
 609 +static unsigned int imq_nf_hook(unsigned int hook, struct sk_buff *pskb,
 610 +                               const struct net_device *indev,
 611 +                               const struct net_device *outdev,
 612 +                               int (*okfn)(struct sk_buff *))
 613 +{
 614 +       return (pskb->imq_flags & IMQ_F_ENQUEUE) ? NF_IMQ_QUEUE : NF_ACCEPT;
 615 +}
 616 +
 617 +static int imq_close(struct net_device *dev)
 618 +{
 619 +       netif_stop_queue(dev);
 620 +       return 0;
 621 +}
 622 +
 623 +static int imq_open(struct net_device *dev)
 624 +{
 625 +       netif_start_queue(dev);
 626 +       return 0;
 627 +}
 628 +
 629 +static const struct net_device_ops imq_netdev_ops = {
 630 +       .ndo_open               = imq_open,
 631 +       .ndo_stop               = imq_close,
 632 +       .ndo_start_xmit         = imq_dev_xmit,
 633 +       .ndo_get_stats          = imq_get_stats,
 634 +};
 635 +
 636 +static void imq_setup(struct net_device *dev)
 637 +{
 638 +       dev->netdev_ops         = &imq_netdev_ops;
 639 +       dev->type               = ARPHRD_VOID;
 640 +       dev->mtu                = 16000;
 641 +       dev->tx_queue_len       = 11000;
 642 +       dev->flags              = IFF_NOARP;
 643 +       dev->features           = NETIF_F_SG | NETIF_F_FRAGLIST |
 644 +                                 NETIF_F_GSO | NETIF_F_HW_CSUM |
 645 +                                 NETIF_F_HIGHDMA;
 646 +       dev->priv_flags         &= ~IFF_XMIT_DST_RELEASE;
 647 +}
 648 +
 649 +static int imq_validate(struct nlattr *tb[], struct nlattr *data[])
 650 +{
 651 +       int ret = 0;
 652 +
 653 +       if (tb[IFLA_ADDRESS]) {
 654 +               if (nla_len(tb[IFLA_ADDRESS]) != ETH_ALEN) {
 655 +                       ret = -EINVAL;
 656 +                       goto end;
 657 +               }
 658 +               if (!is_valid_ether_addr(nla_data(tb[IFLA_ADDRESS]))) {
 659 +                       ret = -EADDRNOTAVAIL;
 660 +                       goto end;
 661 +               }
 662 +       }
 663 +       return 0;
 664 +end:
 665 +       printk(KERN_WARNING "IMQ: imq_validate failed (%d)\n", ret);
 666 +       return ret;
 667 +}
 668 +
 669 +static struct rtnl_link_ops imq_link_ops __read_mostly = {
 670 +       .kind           = "imq",
 671 +       .priv_size      = 0,
 672 +       .setup          = imq_setup,
 673 +       .validate       = imq_validate,
 674 +};
 675 +
 676 +static const struct nf_queue_handler imq_nfqh = {
 677 +       .name  = "imq",
 678 +       .outfn = imq_nf_queue,
 679 +};
 680 +
 681 +static int __init imq_init_hooks(void)
 682 +{
 683 +       int ret;
 684 +
 685 +       nf_register_queue_imq_handler(&imq_nfqh);
 686 +
 687 +       ret = nf_register_hooks(imq_ops, ARRAY_SIZE(imq_ops));
 688 +       if (ret < 0)
 689 +               nf_unregister_queue_imq_handler();
 690 +
 691 +       return ret;
 692 +}
 693 +
 694 +static int __init imq_init_one(int index)
 695 +{
 696 +       struct net_device *dev;
 697 +       int ret;
 698 +
 699 +       dev = alloc_netdev_mq(0, "imq%d", imq_setup, numqueues);
 700 +       if (!dev)
 701 +               return -ENOMEM;
 702 +
 703 +       ret = dev_alloc_name(dev, dev->name);
 704 +       if (ret < 0)
 705 +               goto fail;
 706 +
 707 +       dev->rtnl_link_ops = &imq_link_ops;
 708 +       ret = register_netdevice(dev);
 709 +       if (ret < 0)
 710 +               goto fail;
 711 +
 712 +       return 0;
 713 +fail:
 714 +       free_netdev(dev);
 715 +       return ret;
 716 +}
 717 +
 718 +static int __init imq_init_devs(void)
 719 +{
 720 +       int err, i;
 721 +
 722 +       if (numdevs < 1 || numdevs > IMQ_MAX_DEVS) {
 723 +               printk(KERN_ERR "IMQ: numdevs has to be betweed 1 and %u\n",
 724 +                      IMQ_MAX_DEVS);
 725 +               return -EINVAL;
 726 +       }
 727 +
 728 +       if (numqueues < 1 || numqueues > IMQ_MAX_QUEUES) {
 729 +               printk(KERN_ERR "IMQ: numqueues has to be betweed 1 and %u\n",
 730 +                      IMQ_MAX_QUEUES);
 731 +               return -EINVAL;
 732 +       }
 733 +
 734 +       get_random_bytes(&imq_hashrnd, sizeof(imq_hashrnd));
 735 +
 736 +       rtnl_lock();
 737 +       err = __rtnl_link_register(&imq_link_ops);
 738 +
 739 +       for (i = 0; i < numdevs && !err; i++)
 740 +               err = imq_init_one(i);
 741 +
 742 +       if (err) {
 743 +               __rtnl_link_unregister(&imq_link_ops);
 744 +               memset(imq_devs_cache, 0, sizeof(imq_devs_cache));
 745 +       }
 746 +       rtnl_unlock();
 747 +
 748 +       return err;
 749 +}
 750 +
 751 +static int __init imq_init_module(void)
 752 +{
 753 +       int err;
 754 +
 755 +#if defined(CONFIG_IMQ_NUM_DEVS)
 756 +       BUILD_BUG_ON(CONFIG_IMQ_NUM_DEVS > 16);
 757 +       BUILD_BUG_ON(CONFIG_IMQ_NUM_DEVS < 2);
 758 +       BUILD_BUG_ON(CONFIG_IMQ_NUM_DEVS - 1 > IMQ_F_IFMASK);
 759 +#endif
 760 +
 761 +       err = imq_init_devs();
 762 +       if (err) {
 763 +               printk(KERN_ERR "IMQ: Error trying imq_init_devs(net)\n");
 764 +               return err;
 765 +       }
 766 +
 767 +       err = imq_init_hooks();
 768 +       if (err) {
 769 +               printk(KERN_ERR "IMQ: Error trying imq_init_hooks()\n");
 770 +               rtnl_link_unregister(&imq_link_ops);
 771 +               memset(imq_devs_cache, 0, sizeof(imq_devs_cache));
 772 +               return err;
 773 +       }
 774 +
 775 +       printk(KERN_INFO "IMQ driver loaded successfully. "
 776 +               "(numdevs = %d, numqueues = %d)\n", numdevs, numqueues);
 777 +
 778 +#if defined(CONFIG_IMQ_BEHAVIOR_BA) || defined(CONFIG_IMQ_BEHAVIOR_BB)
 779 +       printk(KERN_INFO "\tHooking IMQ before NAT on PREROUTING.\n");
 780 +#else
 781 +       printk(KERN_INFO "\tHooking IMQ after NAT on PREROUTING.\n");
 782 +#endif
 783 +#if defined(CONFIG_IMQ_BEHAVIOR_AB) || defined(CONFIG_IMQ_BEHAVIOR_BB)
 784 +       printk(KERN_INFO "\tHooking IMQ before NAT on POSTROUTING.\n");
 785 +#else
 786 +       printk(KERN_INFO "\tHooking IMQ after NAT on POSTROUTING.\n");
 787 +#endif
 788 +
 789 +       return 0;
 790 +}
 791 +
 792 +static void __exit imq_unhook(void)
 793 +{
 794 +       nf_unregister_hooks(imq_ops, ARRAY_SIZE(imq_ops));
 795 +       nf_unregister_queue_imq_handler();
 796 +}
 797 +
 798 +static void __exit imq_cleanup_devs(void)
 799 +{
 800 +       rtnl_link_unregister(&imq_link_ops);
 801 +       memset(imq_devs_cache, 0, sizeof(imq_devs_cache));
 802 +}
 803 +
 804 +static void __exit imq_exit_module(void)
 805 +{
 806 +       imq_unhook();
 807 +       imq_cleanup_devs();
 808 +       printk(KERN_INFO "IMQ driver unloaded successfully.\n");
 809 +}
 810 +
 811 +module_init(imq_init_module);
 812 +module_exit(imq_exit_module);
 813 +
 814 +module_param(numdevs, int, 0);
 815 +module_param(numqueues, int, 0);
 816 +MODULE_PARM_DESC(numdevs, "number of IMQ devices (how many imq* devices will "
 817 +                       "be created)");
 818 +MODULE_PARM_DESC(numqueues, "number of queues per IMQ device");
 819 +MODULE_AUTHOR("http://www.linuximq.net");
 820 +MODULE_DESCRIPTION("Pseudo-driver for the intermediate queue device. See "
 821 +                       "http://www.linuximq.net/ for more information.");
 822 +MODULE_LICENSE("GPL");
 823 +MODULE_ALIAS_RTNL_LINK("imq");
 824 +
 825 diff -uNr linux-3.0/drivers/net/Kconfig linux-3.0-imq/drivers/net/Kconfig
 826 --- linux-3.0/drivers/net/Kconfig       2011-07-22 05:17:23.000000000 +0300
 827 +++ linux-3.0-imq/drivers/net/Kconfig   2011-07-26 06:31:36.176747906 +0300
 828 @@ -124,6 +124,125 @@
 829           To compile this driver as a module, choose M here: the module
 830           will be called eql.  If unsure, say N.
 831
 832 +config IMQ
 833 +       tristate "IMQ (intermediate queueing device) support"
 834 +       depends on NETDEVICES && NETFILTER
 835 +       ---help---
 836 +         The IMQ device(s) is used as placeholder for QoS queueing
 837 +         disciplines. Every packet entering/leaving the IP stack can be
 838 +         directed through the IMQ device where it's enqueued/dequeued to the
 839 +         attached qdisc. This allows you to treat network devices as classes
 840 +         and distribute bandwidth among them. Iptables is used to specify
 841 +         through which IMQ device, if any, packets travel.
 842 +
 843 +         More information at: http://www.linuximq.net/
 844 +
 845 +         To compile this driver as a module, choose M here: the module
 846 +         will be called imq.  If unsure, say N.
 847 +
 848 +choice
 849 +       prompt "IMQ behavior (PRE/POSTROUTING)"
 850 +       depends on IMQ
 851 +       default IMQ_BEHAVIOR_AB
 852 +       help
 853 +         This setting defines how IMQ behaves in respect to its
 854 +         hooking in PREROUTING and POSTROUTING.
 855 +
 856 +         IMQ can work in any of the following ways:
 857 +
 858 +             PREROUTING   |      POSTROUTING
 859 +         -----------------|-------------------
 860 +         #1  After NAT    |      After NAT
 861 +         #2  After NAT    |      Before NAT
 862 +         #3  Before NAT   |      After NAT
 863 +         #4  Before NAT   |      Before NAT
 864 +
 865 +         The default behavior is to hook before NAT on PREROUTING
 866 +         and after NAT on POSTROUTING (#3).
 867 +
 868 +         This settings are specially usefull when trying to use IMQ
 869 +         to shape NATed clients.
 870 +
 871 +         More information can be found at: www.linuximq.net
 872 +
 873 +         If not sure leave the default settings alone.
 874 +
 875 +config IMQ_BEHAVIOR_AA
 876 +       bool "IMQ AA"
 877 +       help
 878 +         This setting defines how IMQ behaves in respect to its
 879 +         hooking in PREROUTING and POSTROUTING.
 880 +
 881 +         Choosing this option will make IMQ hook like this:
 882 +
 883 +         PREROUTING:   After NAT
 884 +         POSTROUTING:  After NAT
 885 +
 886 +         More information can be found at: www.linuximq.net
 887 +
 888 +         If not sure leave the default settings alone.
 889 +
 890 +config IMQ_BEHAVIOR_AB
 891 +       bool "IMQ AB"
 892 +       help
 893 +         This setting defines how IMQ behaves in respect to its
 894 +         hooking in PREROUTING and POSTROUTING.
 895 +
 896 +         Choosing this option will make IMQ hook like this:
 897 +
 898 +         PREROUTING:   After NAT
 899 +         POSTROUTING:  Before NAT
 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_BA
 906 +       bool "IMQ BA"
 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:   Before 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_BB
 921 +       bool "IMQ BB"
 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:   Before 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 +endchoice
 936 +
 937 +config IMQ_NUM_DEVS
 938 +       int "Number of IMQ devices"
 939 +       range 2 16
 940 +       depends on IMQ
 941 +       default "16"
 942 +       help
 943 +         This setting defines how many IMQ devices will be created.
 944 +
 945 +         The default value is 16.
 946 +
 947 +         More information can be found at: www.linuximq.net
 948 +
 949 +         If not sure leave the default settings alone.
 950 +
 951  config TUN
 952         tristate "Universal TUN/TAP device driver support"
 953         select CRC32
 954 diff -uNr linux-3.0/drivers/net/Makefile linux-3.0-imq/drivers/net/Makefile
 955 --- linux-3.0/drivers/net/Makefile      2011-07-22 05:17:23.000000000 +0300
 956 +++ linux-3.0-imq/drivers/net/Makefile  2011-07-24 12:06:25.922003276 +0300
 957 @@ -176,6 +176,7 @@
 958  obj-$(CONFIG_XEN_NETDEV_BACKEND) += xen-netback/
 959
 960  obj-$(CONFIG_DUMMY) += dummy.o
 961 +obj-$(CONFIG_IMQ) += imq.o
 962  obj-$(CONFIG_IFB) += ifb.o
 963  obj-$(CONFIG_MACVLAN) += macvlan.o
 964  obj-$(CONFIG_MACVTAP) += macvtap.o
 965 diff -uNr linux-3.0/include/linux/imq.h linux-3.0-imq/include/linux/imq.h
 966 --- linux-3.0/include/linux/imq.h       1970-01-01 02:00:00.000000000 +0200
 967 +++ linux-3.0-imq/include/linux/imq.h   2011-07-24 12:06:25.932003270 +0300
 968 @@ -0,0 +1,13 @@
 969 +#ifndef _IMQ_H
 970 +#define _IMQ_H
 971 +
 972 +/* IFMASK (16 device indexes, 0 to 15) and flag(s) fit in 5 bits */
 973 +#define IMQ_F_BITS     5
 974 +
 975 +#define IMQ_F_IFMASK   0x0f
 976 +#define IMQ_F_ENQUEUE  0x10
 977 +
 978 +#define IMQ_MAX_DEVS   (IMQ_F_IFMASK + 1)
 979 +
 980 +#endif /* _IMQ_H */
 981 +
 982 diff -uNr linux-3.0/include/linux/netfilter/xt_IMQ.h linux-3.0-imq/include/linux/netfilter/xt_IMQ.h
 983 --- linux-3.0/include/linux/netfilter/xt_IMQ.h  1970-01-01 02:00:00.000000000 +0200
 984 +++ linux-3.0-imq/include/linux/netfilter/xt_IMQ.h      2011-07-24 12:06:25.932003270 +0300
 985 @@ -0,0 +1,9 @@
 986 +#ifndef _XT_IMQ_H
 987 +#define _XT_IMQ_H
 988 +
 989 +struct xt_imq_info {
 990 +       unsigned int todev;     /* target imq device */
 991 +};
 992 +
 993 +#endif /* _XT_IMQ_H */
 994 +
 995 diff -uNr linux-3.0/include/linux/netfilter.h linux-3.0-imq/include/linux/netfilter.h
 996 --- linux-3.0/include/linux/netfilter.h 2011-07-22 05:17:23.000000000 +0300
 997 +++ linux-3.0-imq/include/linux/netfilter.h     2011-07-24 12:06:25.955336605 +0300
 998 @@ -22,7 +22,8 @@
 999  #define NF_QUEUE 3
1000  #define NF_REPEAT 4
1001  #define NF_STOP 5
1002 -#define NF_MAX_VERDICT NF_STOP
1003 +#define NF_IMQ_QUEUE 6
1004 +#define NF_MAX_VERDICT NF_IMQ_QUEUE
1005
1006  /* we overload the higher bits for encoding auxiliary data such as the queue
1007   * number or errno values. Not nice, but better than additional function
1008 diff -uNr linux-3.0/include/linux/netfilter_ipv4/ipt_IMQ.h linux-3.0-imq/include/linux/netfilter_ipv4/ipt_IMQ.h
1009 --- linux-3.0/include/linux/netfilter_ipv4/ipt_IMQ.h    1970-01-01 02:00:00.000000000 +0200
1010 +++ linux-3.0-imq/include/linux/netfilter_ipv4/ipt_IMQ.h        2011-07-24 12:06:25.955336605 +0300
1011 @@ -0,0 +1,10 @@
1012 +#ifndef _IPT_IMQ_H
1013 +#define _IPT_IMQ_H
1014 +
1015 +/* Backwards compatibility for old userspace */
1016 +#include <linux/netfilter/xt_IMQ.h>
1017 +
1018 +#define ipt_imq_info xt_imq_info
1019 +
1020 +#endif /* _IPT_IMQ_H */
1021 +
1022 diff -uNr linux-3.0/include/linux/netfilter_ipv6/ip6t_IMQ.h linux-3.0-imq/include/linux/netfilter_ipv6/ip6t_IMQ.h
1023 --- linux-3.0/include/linux/netfilter_ipv6/ip6t_IMQ.h   1970-01-01 02:00:00.000000000 +0200
1024 +++ linux-3.0-imq/include/linux/netfilter_ipv6/ip6t_IMQ.h       2011-07-24 12:06:25.955336605 +0300
1025 @@ -0,0 +1,10 @@
1026 +#ifndef _IP6T_IMQ_H
1027 +#define _IP6T_IMQ_H
1028 +
1029 +/* Backwards compatibility for old userspace */
1030 +#include <linux/netfilter/xt_IMQ.h>
1031 +
1032 +#define ip6t_imq_info xt_imq_info
1033 +
1034 +#endif /* _IP6T_IMQ_H */
1035 +
1036 diff -uNr linux-3.0/include/linux/skbuff.h linux-3.0-imq/include/linux/skbuff.h
1037 --- linux-3.0/include/linux/skbuff.h    2011-07-22 05:17:23.000000000 +0300
1038 +++ linux-3.0-imq/include/linux/skbuff.h        2011-07-24 12:06:25.968669945 +0300
1039 @@ -29,6 +29,9 @@
1040  #include <linux/rcupdate.h>
1041  #include <linux/dmaengine.h>
1042  #include <linux/hrtimer.h>
1043 +#if defined(CONFIG_IMQ) || defined(CONFIG_IMQ_MODULE)
1044 +#include <linux/imq.h>
1045 +#endif
1046
1047  /* Don't change this without changing skb_csum_unnecessary! */
1048  #define CHECKSUM_NONE 0
1049 @@ -339,6 +342,9 @@
1050          * first. This is owned by whoever has the skb queued ATM.
1051          */
1052         char                    cb[48] __aligned(8);
1053 +#if defined(CONFIG_IMQ) || defined(CONFIG_IMQ_MODULE)
1054 +       void                    *cb_next;
1055 +#endif
1056
1057         unsigned long           _skb_refdst;
1058  #ifdef CONFIG_XFRM
1059 @@ -377,6 +383,9 @@
1060  #ifdef NET_SKBUFF_NF_DEFRAG_NEEDED
1061         struct sk_buff          *nfct_reasm;
1062  #endif
1063 +#if defined(CONFIG_IMQ) || defined(CONFIG_IMQ_MODULE)
1064 +       struct nf_queue_entry   *nf_queue_entry;
1065 +#endif
1066  #ifdef CONFIG_BRIDGE_NETFILTER
1067         struct nf_bridge_info   *nf_bridge;
1068  #endif
1069 @@ -401,6 +410,10 @@
1070
1071         /* 0/13 bit hole */
1072
1073 +#if defined(CONFIG_IMQ) || defined(CONFIG_IMQ_MODULE)
1074 +       __u8                    imq_flags:IMQ_F_BITS;
1075 +#endif
1076 +
1077  #ifdef CONFIG_NET_DMA
1078         dma_cookie_t            dma_cookie;
1079  #endif
1080 @@ -487,6 +500,12 @@
1081         return (struct rtable *)skb_dst(skb);
1082  }
1083
1084 +
1085 +#if defined(CONFIG_IMQ) || defined(CONFIG_IMQ_MODULE)
1086 +extern int skb_save_cb(struct sk_buff *skb);
1087 +extern int skb_restore_cb(struct sk_buff *skb);
1088 +#endif
1089 +
1090  extern void kfree_skb(struct sk_buff *skb);
1091  extern void consume_skb(struct sk_buff *skb);
1092  extern void           __kfree_skb(struct sk_buff *skb);
1093 @@ -2134,6 +2153,10 @@
1094         dst->nfct_reasm = src->nfct_reasm;
1095         nf_conntrack_get_reasm(src->nfct_reasm);
1096  #endif
1097 +#if defined(CONFIG_IMQ) || defined(CONFIG_IMQ_MODULE)
1098 +       dst->imq_flags = src->imq_flags;
1099 +       dst->nf_queue_entry = src->nf_queue_entry;
1100 +#endif
1101  #ifdef CONFIG_BRIDGE_NETFILTER
1102         dst->nf_bridge  = src->nf_bridge;
1103         nf_bridge_get(src->nf_bridge);
1104 diff -uNr linux-3.0/include/net/netfilter/nf_queue.h linux-3.0-imq/include/net/netfilter/nf_queue.h
1105 --- linux-3.0/include/net/netfilter/nf_queue.h  2011-07-22 05:17:23.000000000 +0300
1106 +++ linux-3.0-imq/include/net/netfilter/nf_queue.h      2011-07-24 12:06:25.975336612 +0300
1107 @@ -30,5 +30,11 @@
1108                                        const struct nf_queue_handler *qh);
1109  extern void nf_unregister_queue_handlers(const struct nf_queue_handler *qh);
1110  extern void nf_reinject(struct nf_queue_entry *entry, unsigned int verdict);
1111 +extern void nf_queue_entry_release_refs(struct nf_queue_entry *entry);
1112 +
1113 +#if defined(CONFIG_IMQ) || defined(CONFIG_IMQ_MODULE)
1114 +extern void nf_register_queue_imq_handler(const struct nf_queue_handler *qh);
1115 +extern void nf_unregister_queue_imq_handler(void);
1116 +#endif
1117
1118  #endif /* _NF_QUEUE_H */
1119 diff -uNr linux-3.0/net/core/dev.c linux-3.0-imq/net/core/dev.c
1120 --- linux-3.0/net/core/dev.c    2011-07-22 05:17:23.000000000 +0300
1121 +++ linux-3.0-imq/net/core/dev.c        2011-07-26 07:52:00.513207402 +0300
1122 @@ -98,6 +98,9 @@
1123  #include <net/net_namespace.h>
1124  #include <net/sock.h>
1125  #include <linux/rtnetlink.h>
1126 +#if defined(CONFIG_IMQ) || defined(CONFIG_IMQ_MODULE)
1127 +#include <linux/imq.h>
1128 +#endif
1129  #include <linux/proc_fs.h>
1130  #include <linux/seq_file.h>
1131  #include <linux/stat.h>
1132 @@ -2108,7 +2111,12 @@
1133                 if (dev->priv_flags & IFF_XMIT_DST_RELEASE)
1134                         skb_dst_drop(skb);
1135
1136 +#if defined(CONFIG_IMQ) || defined(CONFIG_IMQ_MODULE)
1137 +               if (!list_empty(&ptype_all) &&
1138 +                                       !(skb->imq_flags & IMQ_F_ENQUEUE))
1139 +#else
1140                 if (!list_empty(&ptype_all))
1141 +#endif
1142                         dev_queue_xmit_nit(skb, dev);
1143
1144                 skb_orphan_try(skb);
1145 diff -uNr linux-3.0/net/core/skbuff.c linux-3.0-imq/net/core/skbuff.c
1146 --- linux-3.0/net/core/skbuff.c 2011-07-22 05:17:23.000000000 +0300
1147 +++ linux-3.0-imq/net/core/skbuff.c     2011-07-24 12:06:26.008669943 +0300
1148 @@ -73,6 +73,9 @@
1149
1150  static struct kmem_cache *skbuff_head_cache __read_mostly;
1151  static struct kmem_cache *skbuff_fclone_cache __read_mostly;
1152 +#if defined(CONFIG_IMQ) || defined(CONFIG_IMQ_MODULE)
1153 +static struct kmem_cache *skbuff_cb_store_cache __read_mostly;
1154 +#endif
1155
1156  static void sock_pipe_buf_release(struct pipe_inode_info *pipe,
1157                                   struct pipe_buffer *buf)
1158 @@ -92,6 +95,82 @@
1159         return 1;
1160  }
1161
1162 +#if defined(CONFIG_IMQ) || defined(CONFIG_IMQ_MODULE)
1163 +/* Control buffer save/restore for IMQ devices */
1164 +struct skb_cb_table {
1165 +       char                    cb[48] __aligned(8);
1166 +       void                    *cb_next;
1167 +       atomic_t                refcnt;
1168 +};
1169 +
1170 +static DEFINE_SPINLOCK(skb_cb_store_lock);
1171 +
1172 +int skb_save_cb(struct sk_buff *skb)
1173 +{
1174 +       struct skb_cb_table *next;
1175 +
1176 +       next = kmem_cache_alloc(skbuff_cb_store_cache, GFP_ATOMIC);
1177 +       if (!next)
1178 +               return -ENOMEM;
1179 +
1180 +       BUILD_BUG_ON(sizeof(skb->cb) != sizeof(next->cb));
1181 +
1182 +       memcpy(next->cb, skb->cb, sizeof(skb->cb));
1183 +       next->cb_next = skb->cb_next;
1184 +
1185 +       atomic_set(&next->refcnt, 1);
1186 +
1187 +       skb->cb_next = next;
1188 +       return 0;
1189 +}
1190 +EXPORT_SYMBOL(skb_save_cb);
1191 +
1192 +int skb_restore_cb(struct sk_buff *skb)
1193 +{
1194 +       struct skb_cb_table *next;
1195 +
1196 +       if (!skb->cb_next)
1197 +               return 0;
1198 +
1199 +       next = skb->cb_next;
1200 +
1201 +       BUILD_BUG_ON(sizeof(skb->cb) != sizeof(next->cb));
1202 +
1203 +       memcpy(skb->cb, next->cb, sizeof(skb->cb));
1204 +       skb->cb_next = next->cb_next;
1205 +
1206 +       spin_lock(&skb_cb_store_lock);
1207 +
1208 +       if (atomic_dec_and_test(&next->refcnt))
1209 +               kmem_cache_free(skbuff_cb_store_cache, next);
1210 +
1211 +       spin_unlock(&skb_cb_store_lock);
1212 +
1213 +       return 0;
1214 +}
1215 +EXPORT_SYMBOL(skb_restore_cb);
1216 +
1217 +static void skb_copy_stored_cb(struct sk_buff *new, const struct sk_buff *__old)
1218 +{
1219 +       struct skb_cb_table *next;
1220 +       struct sk_buff *old;
1221 +
1222 +       if (!__old->cb_next) {
1223 +               new->cb_next = NULL;
1224 +               return;
1225 +       }
1226 +
1227 +       spin_lock(&skb_cb_store_lock);
1228 +
1229 +       old = (struct sk_buff *)__old;
1230 +
1231 +       next = old->cb_next;
1232 +       atomic_inc(&next->refcnt);
1233 +       new->cb_next = next;
1234 +
1235 +       spin_unlock(&skb_cb_store_lock);
1236 +}
1237 +#endif
1238
1239  /* Pipe buffer operations for a socket. */
1240  static const struct pipe_buf_operations sock_pipe_buf_ops = {
1241 @@ -380,6 +459,26 @@
1242                 WARN_ON(in_irq());
1243                 skb->destructor(skb);
1244         }
1245 +#if defined(CONFIG_IMQ) || defined(CONFIG_IMQ_MODULE)
1246 +       /* This should not happen. When it does, avoid memleak by restoring
1247 +       the chain of cb-backups. */
1248 +       while (skb->cb_next != NULL) {
1249 +               if (net_ratelimit())
1250 +                       printk(KERN_WARNING "IMQ: kfree_skb: skb->cb_next: "
1251 +                               "%08x\n", (unsigned int)skb->cb_next);
1252 +
1253 +               skb_restore_cb(skb);
1254 +       }
1255 +       /* This should not happen either, nf_queue_entry is nullified in
1256 +        * imq_dev_xmit(). If we have non-NULL nf_queue_entry then we are
1257 +        * leaking entry pointers, maybe memory. We don't know if this is
1258 +        * pointer to already freed memory, or should this be freed.
1259 +        * If this happens we need to add refcounting, etc for nf_queue_entry.
1260 +        */
1261 +       if (skb->nf_queue_entry && net_ratelimit())
1262 +               printk(KERN_WARNING
1263 +                               "IMQ: kfree_skb: skb->nf_queue_entry != NULL");
1264 +#endif
1265  #if defined(CONFIG_NF_CONNTRACK) || defined(CONFIG_NF_CONNTRACK_MODULE)
1266         nf_conntrack_put(skb->nfct);
1267  #endif
1268 @@ -518,6 +617,9 @@
1269         new->sp                 = secpath_get(old->sp);
1270  #endif
1271         memcpy(new->cb, old->cb, sizeof(old->cb));
1272 +#if defined(CONFIG_IMQ) || defined(CONFIG_IMQ_MODULE)
1273 +       skb_copy_stored_cb(new, old);
1274 +#endif
1275         new->csum               = old->csum;
1276         new->local_df           = old->local_df;
1277         new->pkt_type           = old->pkt_type;
1278 @@ -2781,6 +2883,13 @@
1279                                                 0,
1280                                                 SLAB_HWCACHE_ALIGN|SLAB_PANIC,
1281                                                 NULL);
1282 +#if defined(CONFIG_IMQ) || defined(CONFIG_IMQ_MODULE)
1283 +       skbuff_cb_store_cache = kmem_cache_create("skbuff_cb_store_cache",
1284 +                                                 sizeof(struct skb_cb_table),
1285 +                                                 0,
1286 +                                                 SLAB_HWCACHE_ALIGN|SLAB_PANIC,
1287 +                                                 NULL);
1288 +#endif
1289  }
1290
1291  /**
1292 diff -uNr linux-3.0/net/ipv6/ip6_output.c linux-3.0-imq/net/ipv6/ip6_output.c
1293 --- linux-3.0/net/ipv6/ip6_output.c     2011-07-22 05:17:23.000000000 +0300
1294 +++ linux-3.0-imq/net/ipv6/ip6_output.c 2011-07-24 16:46:04.789482257 +0300
1295 @@ -101,9 +101,6 @@
1296         struct dst_entry *dst = skb_dst(skb);
1297         struct net_device *dev = dst->dev;
1298
1299 -       skb->protocol = htons(ETH_P_IPV6);
1300 -       skb->dev = dev;
1301 -
1302         if (ipv6_addr_is_multicast(&ipv6_hdr(skb)->daddr)) {
1303                 struct inet6_dev *idev = ip6_dst_idev(skb_dst(skb));
1304
1305 @@ -165,6 +162,11 @@
1306                 return 0;
1307         }
1308
1309 +       /* IMQ-patch: moved setting skb->dev and skb->protocol from
1310 +        * ip6_finish_output2 to fix crashing at netif_skb_features(). */
1311 +       skb->protocol = htons(ETH_P_IPV6);
1312 +       skb->dev = dev;
1313 +
1314         return NF_HOOK_COND(NFPROTO_IPV6, NF_INET_POST_ROUTING, skb, NULL, dev,
1315                             ip6_finish_output,
1316                             !(IP6CB(skb)->flags & IP6SKB_REROUTED));
1317 diff -uNr linux-3.0/net/netfilter/core.c linux-3.0-imq/net/netfilter/core.c
1318 --- linux-3.0/net/netfilter/core.c      2011-07-22 05:17:23.000000000 +0300
1319 +++ linux-3.0-imq/net/netfilter/core.c  2011-07-24 12:53:52.972141108 +0300
1320 @@ -179,9 +179,11 @@
1321                 ret = NF_DROP_GETERR(verdict);
1322                 if (ret == 0)
1323                         ret = -EPERM;
1324 -       } else if ((verdict & NF_VERDICT_MASK) == NF_QUEUE) {
1325 +       } else if ((verdict & NF_VERDICT_MASK) == NF_QUEUE ||
1326 +                  (verdict & NF_VERDICT_MASK) == NF_IMQ_QUEUE) {
1327                 ret = nf_queue(skb, elem, pf, hook, indev, outdev, okfn,
1328 -                              verdict >> NF_VERDICT_QBITS);
1329 +                              verdict >> NF_VERDICT_QBITS,
1330 +                              verdict & NF_VERDICT_MASK);
1331                 if (ret < 0) {
1332                         if (ret == -ECANCELED)
1333                                 goto next_hook;
1334 diff -uNr linux-3.0/net/netfilter/Kconfig linux-3.0-imq/net/netfilter/Kconfig
1335 --- linux-3.0/net/netfilter/Kconfig     2011-07-22 05:17:23.000000000 +0300
1336 +++ linux-3.0-imq/net/netfilter/Kconfig 2011-07-24 12:06:26.035336611 +0300
1337 @@ -507,6 +507,18 @@
1338           For more information on the LEDs available on your system, see
1339           Documentation/leds-class.txt
1340
1341 +config NETFILTER_XT_TARGET_IMQ
1342 +        tristate '"IMQ" target support'
1343 +       depends on NETFILTER_XTABLES
1344 +       depends on IP_NF_MANGLE || IP6_NF_MANGLE
1345 +       select IMQ
1346 +       default m if NETFILTER_ADVANCED=n
1347 +        help
1348 +          This option adds a `IMQ' target which is used to specify if and
1349 +          to which imq device packets should get enqueued/dequeued.
1350 +
1351 +          To compile it as a module, choose M here.  If unsure, say N.
1352 +
1353  config NETFILTER_XT_TARGET_MARK
1354         tristate '"MARK" target support'
1355         depends on NETFILTER_ADVANCED
1356 diff -uNr linux-3.0/net/netfilter/Makefile linux-3.0-imq/net/netfilter/Makefile
1357 --- linux-3.0/net/netfilter/Makefile    2011-07-22 05:17:23.000000000 +0300
1358 +++ linux-3.0-imq/net/netfilter/Makefile        2011-07-24 12:06:26.042003277 +0300
1359 @@ -56,6 +56,7 @@
1360  obj-$(CONFIG_NETFILTER_XT_TARGET_CT) += xt_CT.o
1361  obj-$(CONFIG_NETFILTER_XT_TARGET_DSCP) += xt_DSCP.o
1362  obj-$(CONFIG_NETFILTER_XT_TARGET_HL) += xt_HL.o
1363 +obj-$(CONFIG_NETFILTER_XT_TARGET_IMQ) += xt_IMQ.o
1364  obj-$(CONFIG_NETFILTER_XT_TARGET_LED) += xt_LED.o
1365  obj-$(CONFIG_NETFILTER_XT_TARGET_NFLOG) += xt_NFLOG.o
1366  obj-$(CONFIG_NETFILTER_XT_TARGET_NFQUEUE) += xt_NFQUEUE.o
1367 diff -uNr linux-3.0/net/netfilter/nf_internals.h linux-3.0-imq/net/netfilter/nf_internals.h
1368 --- linux-3.0/net/netfilter/nf_internals.h      2011-07-22 05:17:23.000000000 +0300
1369 +++ linux-3.0-imq/net/netfilter/nf_internals.h  2011-07-24 12:54:17.615475634 +0300
1370 @@ -29,7 +29,7 @@
1371                     struct net_device *indev,
1372                     struct net_device *outdev,
1373                     int (*okfn)(struct sk_buff *),
1374 -                   unsigned int queuenum);
1375 +                   unsigned int queuenum, unsigned int queuetype);
1376  extern int __init netfilter_queue_init(void);
1377
1378  /* nf_log.c */
1379 diff -uNr linux-3.0/net/netfilter/nf_queue.c linux-3.0-imq/net/netfilter/nf_queue.c
1380 --- linux-3.0/net/netfilter/nf_queue.c  2011-07-22 05:17:23.000000000 +0300
1381 +++ linux-3.0-imq/net/netfilter/nf_queue.c      2011-07-24 13:05:00.682173434 +0300
1382 @@ -22,6 +22,26 @@
1383
1384  static DEFINE_MUTEX(queue_handler_mutex);
1385
1386 +#if defined(CONFIG_IMQ) || defined(CONFIG_IMQ_MODULE)
1387 +static const struct nf_queue_handler *queue_imq_handler;
1388 +
1389 +void nf_register_queue_imq_handler(const struct nf_queue_handler *qh)
1390 +{
1391 +       mutex_lock(&queue_handler_mutex);
1392 +       rcu_assign_pointer(queue_imq_handler, qh);
1393 +       mutex_unlock(&queue_handler_mutex);
1394 +}
1395 +EXPORT_SYMBOL_GPL(nf_register_queue_imq_handler);
1396 +
1397 +void nf_unregister_queue_imq_handler(void)
1398 +{
1399 +       mutex_lock(&queue_handler_mutex);
1400 +       rcu_assign_pointer(queue_imq_handler, NULL);
1401 +       mutex_unlock(&queue_handler_mutex);
1402 +}
1403 +EXPORT_SYMBOL_GPL(nf_unregister_queue_imq_handler);
1404 +#endif
1405 +
1406  /* return EBUSY when somebody else is registered, return EEXIST if the
1407   * same handler is registered, return 0 in case of success. */
1408  int nf_register_queue_handler(u_int8_t pf, const struct nf_queue_handler *qh)
1409 @@ -92,7 +112,7 @@
1410  }
1411  EXPORT_SYMBOL_GPL(nf_unregister_queue_handlers);
1412
1413 -static void nf_queue_entry_release_refs(struct nf_queue_entry *entry)
1414 +void nf_queue_entry_release_refs(struct nf_queue_entry *entry)
1415  {
1416         /* Release those devices we held, or Alexey will kill me. */
1417         if (entry->indev)
1418 @@ -112,6 +132,7 @@
1419         /* Drop reference to owner of hook which queued us. */
1420         module_put(entry->elem->owner);
1421  }
1422 +EXPORT_SYMBOL_GPL(nf_queue_entry_release_refs);
1423
1424  /*
1425   * Any packet that leaves via this function must come back
1426 @@ -123,7 +144,8 @@
1427                       struct net_device *indev,
1428                       struct net_device *outdev,
1429                       int (*okfn)(struct sk_buff *),
1430 -                     unsigned int queuenum)
1431 +                     unsigned int queuenum,
1432 +                     unsigned int queuetype)
1433  {
1434         int status = -ENOENT;
1435         struct nf_queue_entry *entry = NULL;
1436 @@ -137,7 +159,17 @@
1437         /* QUEUE == DROP if no one is waiting, to be safe. */
1438         rcu_read_lock();
1439
1440 -       qh = rcu_dereference(queue_handler[pf]);
1441 +       if (queuetype == NF_IMQ_QUEUE) {
1442 +#if defined(CONFIG_IMQ) || defined(CONFIG_IMQ_MODULE)
1443 +               qh = rcu_dereference(queue_imq_handler);
1444 +#else
1445 +               BUG();
1446 +               goto err_unlock;
1447 +#endif
1448 +       } else {
1449 +               qh = rcu_dereference(queue_handler[pf]);
1450 +       }
1451 +
1452         if (!qh) {
1453                 status = -ESRCH;
1454                 goto err_unlock;
1455 @@ -209,7 +241,8 @@
1456              struct net_device *indev,
1457              struct net_device *outdev,
1458              int (*okfn)(struct sk_buff *),
1459 -            unsigned int queuenum)
1460 +            unsigned int queuenum,
1461 +            unsigned int queuetype)
1462  {
1463         struct sk_buff *segs;
1464         int err;
1465 @@ -217,7 +250,7 @@
1466
1467         if (!skb_is_gso(skb))
1468                 return __nf_queue(skb, elem, pf, hook, indev, outdev, okfn,
1469 -                                 queuenum);
1470 +                                 queuenum, queuetype);
1471
1472         switch (pf) {
1473         case NFPROTO_IPV4:
1474 @@ -244,7 +277,7 @@
1475                 segs->next = NULL;
1476                 if (err == 0)
1477                         err = __nf_queue(segs, elem, pf, hook, indev,
1478 -                                          outdev, okfn, queuenum);
1479 +                                        outdev, okfn, queuenum, queuetype);
1480                 if (err == 0)
1481                         queued++;
1482                 else
1483 @@ -299,9 +332,11 @@
1484                 local_bh_enable();
1485                 break;
1486         case NF_QUEUE:
1487 +       case NF_IMQ_QUEUE:
1488                 err = __nf_queue(skb, elem, entry->pf, entry->hook,
1489                                  entry->indev, entry->outdev, entry->okfn,
1490 -                                verdict >> NF_VERDICT_QBITS);
1491 +                                verdict >> NF_VERDICT_QBITS,
1492 +                                verdict & NF_VERDICT_MASK);
1493                 if (err < 0) {
1494                         if (err == -ECANCELED)
1495                                 goto next_hook;
1496 diff -uNr linux-3.0/net/netfilter/xt_IMQ.c linux-3.0-imq/net/netfilter/xt_IMQ.c
1497 --- linux-3.0/net/netfilter/xt_IMQ.c    1970-01-01 02:00:00.000000000 +0200
1498 +++ linux-3.0-imq/net/netfilter/xt_IMQ.c        2011-07-24 12:06:26.062003279 +0300
1499 @@ -0,0 +1,74 @@
1500 +/*
1501 + * This target marks packets to be enqueued to an imq device
1502 + */
1503 +#include <linux/module.h>
1504 +#include <linux/skbuff.h>
1505 +#include <linux/netfilter/x_tables.h>
1506 +#include <linux/netfilter/xt_IMQ.h>
1507 +#include <linux/imq.h>
1508 +
1509 +static unsigned int imq_target(struct sk_buff *pskb,
1510 +                               const struct xt_action_param *par)
1511 +{
1512 +       const struct xt_imq_info *mr = par->targinfo;
1513 +
1514 +       pskb->imq_flags = (mr->todev & IMQ_F_IFMASK) | IMQ_F_ENQUEUE;
1515 +
1516 +       return XT_CONTINUE;
1517 +}
1518 +
1519 +static int imq_checkentry(const struct xt_tgchk_param *par)
1520 +{
1521 +       struct xt_imq_info *mr = par->targinfo;
1522 +
1523 +       if (mr->todev > IMQ_MAX_DEVS - 1) {
1524 +               printk(KERN_WARNING
1525 +                      "IMQ: invalid device specified, highest is %u\n",
1526 +                      IMQ_MAX_DEVS - 1);
1527 +               return -EINVAL;
1528 +       }
1529 +
1530 +       return 0;
1531 +}
1532 +
1533 +static struct xt_target xt_imq_reg[] __read_mostly = {
1534 +       {
1535 +               .name           = "IMQ",
1536 +               .family         = AF_INET,
1537 +               .checkentry     = imq_checkentry,
1538 +               .target         = imq_target,
1539 +               .targetsize     = sizeof(struct xt_imq_info),
1540 +               .table          = "mangle",
1541 +               .me             = THIS_MODULE
1542 +       },
1543 +       {
1544 +               .name           = "IMQ",
1545 +               .family         = AF_INET6,
1546 +               .checkentry     = imq_checkentry,
1547 +               .target         = imq_target,
1548 +               .targetsize     = sizeof(struct xt_imq_info),
1549 +               .table          = "mangle",
1550 +               .me             = THIS_MODULE
1551 +       },
1552 +};
1553 +
1554 +static int __init imq_init(void)
1555 +{
1556 +       return xt_register_targets(xt_imq_reg, ARRAY_SIZE(xt_imq_reg));
1557 +}
1558 +
1559 +static void __exit imq_fini(void)
1560 +{
1561 +       xt_unregister_targets(xt_imq_reg, ARRAY_SIZE(xt_imq_reg));
1562 +}
1563 +
1564 +module_init(imq_init);
1565 +module_exit(imq_fini);
1566 +
1567 +MODULE_AUTHOR("http://www.linuximq.net");
1568 +MODULE_DESCRIPTION("Pseudo-driver for the intermediate queue device. "
1569 +                  "See http://www.linuximq.net/ for more information.");
1570 +MODULE_LICENSE("GPL");
1571 +MODULE_ALIAS("ipt_IMQ");
1572 +MODULE_ALIAS("ip6t_IMQ");
1573 +