kernel-imq.patch

   1 diff -uNr linux-2.6.39/drivers/net/imq.c linux-2.6.39-imqmq/drivers/net/imq.c
   2 --- linux-2.6.39/drivers/net/imq.c      1970-01-01 02:00:00.000000000 +0200
   3 +++ linux-2.6.39-imqmq/drivers/net/imq.c        2011-05-19 11:08:03.838522212 +0300
   4 @@ -0,0 +1,777 @@
   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 + *            Also, many thanks to pablo Sebastian Greco for making the initial
 109 + *            patch and to those who helped the testing.
 110 + *
 111 + *             More info at: http://www.linuximq.net/ (Andre Correa)
 112 + */
 113 +
 114 +#include <linux/module.h>
 115 +#include <linux/kernel.h>
 116 +#include <linux/moduleparam.h>
 117 +#include <linux/list.h>
 118 +#include <linux/skbuff.h>
 119 +#include <linux/netdevice.h>
 120 +#include <linux/etherdevice.h>
 121 +#include <linux/rtnetlink.h>
 122 +#include <linux/if_arp.h>
 123 +#include <linux/netfilter.h>
 124 +#include <linux/netfilter_ipv4.h>
 125 +#if defined(CONFIG_IPV6) || defined(CONFIG_IPV6_MODULE)
 126 +       #include <linux/netfilter_ipv6.h>
 127 +#endif
 128 +#include <linux/imq.h>
 129 +#include <net/pkt_sched.h>
 130 +#include <net/netfilter/nf_queue.h>
 131 +#include <net/sock.h>
 132 +#include <linux/ip.h>
 133 +#include <linux/ipv6.h>
 134 +#include <linux/if_vlan.h>
 135 +#include <linux/if_pppox.h>
 136 +#include <net/ip.h>
 137 +#include <net/ipv6.h>
 138 +
 139 +static int imq_nf_queue(struct nf_queue_entry *entry, unsigned queue_num);
 140 +
 141 +static nf_hookfn imq_nf_hook;
 142 +
 143 +static struct nf_hook_ops imq_ops[] = {
 144 +       {
 145 +       /* imq_ingress_ipv4 */
 146 +               .hook           = imq_nf_hook,
 147 +               .owner          = THIS_MODULE,
 148 +               .pf             = PF_INET,
 149 +               .hooknum        = NF_INET_PRE_ROUTING,
 150 +#if defined(CONFIG_IMQ_BEHAVIOR_BA) || defined(CONFIG_IMQ_BEHAVIOR_BB)
 151 +               .priority       = NF_IP_PRI_MANGLE + 1,
 152 +#else
 153 +               .priority       = NF_IP_PRI_NAT_DST + 1,
 154 +#endif
 155 +       },
 156 +       {
 157 +       /* imq_egress_ipv4 */
 158 +               .hook           = imq_nf_hook,
 159 +               .owner          = THIS_MODULE,
 160 +               .pf             = PF_INET,
 161 +               .hooknum        = NF_INET_POST_ROUTING,
 162 +#if defined(CONFIG_IMQ_BEHAVIOR_AA) || defined(CONFIG_IMQ_BEHAVIOR_BA)
 163 +               .priority       = NF_IP_PRI_LAST,
 164 +#else
 165 +               .priority       = NF_IP_PRI_NAT_SRC - 1,
 166 +#endif
 167 +       },
 168 +#if defined(CONFIG_IPV6) || defined(CONFIG_IPV6_MODULE)
 169 +       {
 170 +       /* imq_ingress_ipv6 */
 171 +               .hook           = imq_nf_hook,
 172 +               .owner          = THIS_MODULE,
 173 +               .pf             = PF_INET6,
 174 +               .hooknum        = NF_INET_PRE_ROUTING,
 175 +#if defined(CONFIG_IMQ_BEHAVIOR_BA) || defined(CONFIG_IMQ_BEHAVIOR_BB)
 176 +               .priority       = NF_IP6_PRI_MANGLE + 1,
 177 +#else
 178 +               .priority       = NF_IP6_PRI_NAT_DST + 1,
 179 +#endif
 180 +       },
 181 +       {
 182 +       /* imq_egress_ipv6 */
 183 +               .hook           = imq_nf_hook,
 184 +               .owner          = THIS_MODULE,
 185 +               .pf             = PF_INET6,
 186 +               .hooknum        = NF_INET_POST_ROUTING,
 187 +#if defined(CONFIG_IMQ_BEHAVIOR_AA) || defined(CONFIG_IMQ_BEHAVIOR_BA)
 188 +               .priority       = NF_IP6_PRI_LAST,
 189 +#else
 190 +               .priority       = NF_IP6_PRI_NAT_SRC - 1,
 191 +#endif
 192 +       },
 193 +#endif
 194 +};
 195 +
 196 +#if defined(CONFIG_IMQ_NUM_DEVS)
 197 +static int numdevs = CONFIG_IMQ_NUM_DEVS;
 198 +#else
 199 +static int numdevs = IMQ_MAX_DEVS;
 200 +#endif
 201 +
 202 +#define IMQ_MAX_QUEUES 32
 203 +static int numqueues = 1;
 204 +
 205 +/*static DEFINE_SPINLOCK(imq_nf_queue_lock);*/
 206 +
 207 +static struct net_device *imq_devs_cache[IMQ_MAX_DEVS];
 208 +
 209 +
 210 +static struct net_device_stats *imq_get_stats(struct net_device *dev)
 211 +{
 212 +       return &dev->stats;
 213 +}
 214 +
 215 +/* called for packets kfree'd in qdiscs at places other than enqueue */
 216 +static void imq_skb_destructor(struct sk_buff *skb)
 217 +{
 218 +       struct nf_queue_entry *entry = skb->nf_queue_entry;
 219 +
 220 +       skb->nf_queue_entry = NULL;
 221 +
 222 +       if (entry) {
 223 +               nf_queue_entry_release_refs(entry);
 224 +               kfree(entry);
 225 +       }
 226 +
 227 +       skb_restore_cb(skb); /* kfree backup */
 228 +}
 229 +
 230 +static netdev_tx_t imq_dev_xmit(struct sk_buff *skb, struct net_device *dev)
 231 +{
 232 +       struct nf_queue_entry *entry = skb->nf_queue_entry;
 233 +
 234 +       skb->nf_queue_entry = NULL;
 235 +       dev->trans_start = jiffies;
 236 +
 237 +       dev->stats.tx_bytes += skb->len;
 238 +       dev->stats.tx_packets++;
 239 +
 240 +       if (entry == NULL) {
 241 +               /* We don't know what is going on here.. packet is queued for
 242 +                * imq device, but (probably) not by us.
 243 +                *
 244 +                * If this packet was not send here by imq_nf_queue(), then
 245 +                * skb_save_cb() was not used and skb_free() should not show:
 246 +                *   WARNING: IMQ: kfree_skb: skb->cb_next:..
 247 +                * and/or
 248 +                *   WARNING: IMQ: kfree_skb: skb->nf_queue_entry...
 249 +                *
 250 +                * However if this message is shown, then IMQ is somehow broken
 251 +                * and you should report this to linuximq.net.
 252 +                */
 253 +
 254 +               /* imq_dev_xmit is black hole that eats all packets, report that
 255 +                * we eat this packet happily and increase dropped counters.
 256 +                */
 257 +
 258 +               dev->stats.tx_dropped++;
 259 +               dev_kfree_skb(skb);
 260 +
 261 +               return NETDEV_TX_OK;
 262 +       }
 263 +
 264 +       skb_restore_cb(skb); /* restore skb->cb */
 265 +
 266 +       skb->imq_flags = 0;
 267 +       skb->destructor = NULL;
 268 +
 269 +       nf_reinject(entry, NF_ACCEPT);
 270 +
 271 +       return NETDEV_TX_OK;
 272 +}
 273 +
 274 +static u32 imq_hashrnd;
 275 +
 276 +static inline __be16 pppoe_proto(const struct sk_buff *skb)
 277 +{
 278 +       return *((__be16 *)(skb_mac_header(skb) + ETH_HLEN +
 279 +                       sizeof(struct pppoe_hdr)));
 280 +}
 281 +
 282 +static u16 imq_hash(struct net_device *dev, struct sk_buff *skb)
 283 +{
 284 +       unsigned int pull_len;
 285 +       u16 protocol = skb->protocol;
 286 +       u32 addr1, addr2;
 287 +       u32 hash, ihl = 0;
 288 +       union {
 289 +               u16 in16[2];
 290 +               u32 in32;
 291 +       } ports;
 292 +       u8 ip_proto;
 293 +
 294 +       pull_len = 0;
 295 +
 296 +recheck:
 297 +       switch (protocol) {
 298 +       case htons(ETH_P_8021Q): {
 299 +               if (unlikely(skb_pull(skb, VLAN_HLEN) == NULL))
 300 +                       goto other;
 301 +
 302 +               pull_len += VLAN_HLEN;
 303 +               skb->network_header += VLAN_HLEN;
 304 +
 305 +               protocol = vlan_eth_hdr(skb)->h_vlan_encapsulated_proto;
 306 +               goto recheck;
 307 +       }
 308 +
 309 +       case htons(ETH_P_PPP_SES): {
 310 +               if (unlikely(skb_pull(skb, PPPOE_SES_HLEN) == NULL))
 311 +                       goto other;
 312 +
 313 +               pull_len += PPPOE_SES_HLEN;
 314 +               skb->network_header += PPPOE_SES_HLEN;
 315 +
 316 +               protocol = pppoe_proto(skb);
 317 +               goto recheck;
 318 +       }
 319 +
 320 +       case htons(ETH_P_IP): {
 321 +               const struct iphdr *iph = ip_hdr(skb);
 322 +
 323 +               if (unlikely(!pskb_may_pull(skb, sizeof(struct iphdr))))
 324 +                       goto other;
 325 +
 326 +               addr1 = iph->daddr;
 327 +               addr2 = iph->saddr;
 328 +
 329 +               ip_proto = !(ip_hdr(skb)->frag_off & htons(IP_MF | IP_OFFSET)) ?
 330 +                                iph->protocol : 0;
 331 +               ihl = ip_hdrlen(skb);
 332 +
 333 +               break;
 334 +       }
 335 +#if defined(CONFIG_IPV6) || defined(CONFIG_IPV6_MODULE)
 336 +       case htons(ETH_P_IPV6): {
 337 +               const struct ipv6hdr *iph = ipv6_hdr(skb);
 338 +
 339 +               if (unlikely(!pskb_may_pull(skb, sizeof(struct ipv6hdr))))
 340 +                       goto other;
 341 +
 342 +               addr1 = iph->daddr.s6_addr32[3];
 343 +               addr2 = iph->saddr.s6_addr32[3];
 344 +               ihl = ipv6_skip_exthdr(skb, sizeof(struct ipv6hdr), &ip_proto);
 345 +               if (unlikely(ihl < 0))
 346 +                       goto other;
 347 +
 348 +               break;
 349 +       }
 350 +#endif
 351 +       default:
 352 +other:
 353 +               if (pull_len != 0) {
 354 +                       skb_push(skb, pull_len);
 355 +                       skb->network_header -= pull_len;
 356 +               }
 357 +
 358 +               return (u16)(ntohs(protocol) % dev->real_num_tx_queues);
 359 +       }
 360 +
 361 +       if (addr1 > addr2)
 362 +               swap(addr1, addr2);
 363 +
 364 +       switch (ip_proto) {
 365 +       case IPPROTO_TCP:
 366 +       case IPPROTO_UDP:
 367 +       case IPPROTO_DCCP:
 368 +       case IPPROTO_ESP:
 369 +       case IPPROTO_AH:
 370 +       case IPPROTO_SCTP:
 371 +       case IPPROTO_UDPLITE: {
 372 +               if (likely(skb_copy_bits(skb, ihl, &ports.in32, 4) >= 0)) {
 373 +                       if (ports.in16[0] > ports.in16[1])
 374 +                               swap(ports.in16[0], ports.in16[1]);
 375 +                       break;
 376 +               }
 377 +               /* fall-through */
 378 +       }
 379 +       default:
 380 +               ports.in32 = 0;
 381 +               break;
 382 +       }
 383 +
 384 +       if (pull_len != 0) {
 385 +               skb_push(skb, pull_len);
 386 +               skb->network_header -= pull_len;
 387 +       }
 388 +
 389 +       hash = jhash_3words(addr1, addr2, ports.in32, imq_hashrnd ^ ip_proto);
 390 +
 391 +       return (u16)(((u64)hash * dev->real_num_tx_queues) >> 32);
 392 +}
 393 +
 394 +static inline bool sk_tx_queue_recorded(struct sock *sk)
 395 +{
 396 +       return (sk_tx_queue_get(sk) >= 0);
 397 +}
 398 +
 399 +static struct netdev_queue *imq_select_queue(struct net_device *dev,
 400 +                                               struct sk_buff *skb)
 401 +{
 402 +       u16 queue_index = 0;
 403 +       u32 hash;
 404 +
 405 +       if (likely(dev->real_num_tx_queues == 1))
 406 +               goto out;
 407 +
 408 +       /* IMQ can be receiving ingress or engress packets. */
 409 +
 410 +       /* Check first for if rx_queue is set */
 411 +       if (skb_rx_queue_recorded(skb)) {
 412 +               queue_index = skb_get_rx_queue(skb);
 413 +               goto out;
 414 +       }
 415 +
 416 +       /* Check if socket has tx_queue set */
 417 +       if (sk_tx_queue_recorded(skb->sk)) {
 418 +               queue_index = sk_tx_queue_get(skb->sk);
 419 +               goto out;
 420 +       }
 421 +
 422 +       /* Try use socket hash */
 423 +       if (skb->sk && skb->sk->sk_hash) {
 424 +               hash = skb->sk->sk_hash;
 425 +               queue_index =
 426 +                       (u16)(((u64)hash * dev->real_num_tx_queues) >> 32);
 427 +               goto out;
 428 +       }
 429 +
 430 +       /* Generate hash from packet data */
 431 +       queue_index = imq_hash(dev, skb);
 432 +
 433 +out:
 434 +       if (unlikely(queue_index >= dev->real_num_tx_queues))
 435 +               queue_index = (u16)((u32)queue_index % dev->real_num_tx_queues);
 436 +
 437 +       return netdev_get_tx_queue(dev, queue_index);
 438 +}
 439 +
 440 +static int imq_nf_queue(struct nf_queue_entry *entry, unsigned queue_num)
 441 +{
 442 +       struct net_device *dev;
 443 +       struct sk_buff *skb_orig, *skb, *skb_shared;
 444 +       struct Qdisc *q;
 445 +       struct netdev_queue *txq;
 446 +       spinlock_t *root_lock;
 447 +       int users, index;
 448 +       int retval = -EINVAL;
 449 +
 450 +       index = entry->skb->imq_flags & IMQ_F_IFMASK;
 451 +       if (unlikely(index > numdevs - 1)) {
 452 +               if (net_ratelimit())
 453 +                       printk(KERN_WARNING
 454 +                              "IMQ: invalid device specified, highest is %u\n",
 455 +                              numdevs - 1);
 456 +               retval = -EINVAL;
 457 +               goto out;
 458 +       }
 459 +
 460 +       /* check for imq device by index from cache */
 461 +       dev = imq_devs_cache[index];
 462 +       if (unlikely(!dev)) {
 463 +               char buf[8];
 464 +
 465 +               /* get device by name and cache result */
 466 +               snprintf(buf, sizeof(buf), "imq%d", index);
 467 +               dev = dev_get_by_name(&init_net, buf);
 468 +               if (unlikely(!dev)) {
 469 +                       /* not found ?!*/
 470 +                       BUG();
 471 +                       retval = -ENODEV;
 472 +                       goto out;
 473 +               }
 474 +
 475 +               imq_devs_cache[index] = dev;
 476 +               dev_put(dev);
 477 +       }
 478 +
 479 +       if (unlikely(!(dev->flags & IFF_UP))) {
 480 +               entry->skb->imq_flags = 0;
 481 +               nf_reinject(entry, NF_ACCEPT);
 482 +               retval = 0;
 483 +               goto out;
 484 +       }
 485 +       dev->last_rx = jiffies;
 486 +
 487 +       skb = entry->skb;
 488 +       skb_orig = NULL;
 489 +
 490 +       /* skb has owner? => make clone */
 491 +       if (unlikely(skb->destructor)) {
 492 +               skb_orig = skb;
 493 +               skb = skb_clone(skb, GFP_ATOMIC);
 494 +               if (unlikely(!skb)) {
 495 +                       retval = -ENOMEM;
 496 +                       goto out;
 497 +               }
 498 +               entry->skb = skb;
 499 +       }
 500 +
 501 +       skb->nf_queue_entry = entry;
 502 +
 503 +       dev->stats.rx_bytes += skb->len;
 504 +       dev->stats.rx_packets++;
 505 +
 506 +       /* Disables softirqs for lock below */
 507 +       rcu_read_lock_bh();
 508 +
 509 +       /* Multi-queue selection */
 510 +       txq = imq_select_queue(dev, skb);
 511 +
 512 +       q = rcu_dereference(txq->qdisc);
 513 +       if (unlikely(!q->enqueue))
 514 +               goto packet_not_eaten_by_imq_dev;
 515 +
 516 +       root_lock = qdisc_lock(q);
 517 +       spin_lock(root_lock);
 518 +
 519 +       users = atomic_read(&skb->users);
 520 +
 521 +       skb_shared = skb_get(skb); /* increase reference count by one */
 522 +       skb_save_cb(skb_shared); /* backup skb->cb, as qdisc layer will
 523 +                                       overwrite it */
 524 +       qdisc_enqueue_root(skb_shared, q); /* might kfree_skb */
 525 +
 526 +       if (likely(atomic_read(&skb_shared->users) == users + 1)) {
 527 +               kfree_skb(skb_shared); /* decrease reference count by one */
 528 +
 529 +               skb->destructor = &imq_skb_destructor;
 530 +
 531 +               /* cloned? */
 532 +               if (unlikely(skb_orig))
 533 +                       kfree_skb(skb_orig); /* free original */
 534 +
 535 +               spin_unlock(root_lock);
 536 +               rcu_read_unlock_bh();
 537 +
 538 +               /* schedule qdisc dequeue */
 539 +               __netif_schedule(q);
 540 +
 541 +               retval = 0;
 542 +               goto out;
 543 +       } else {
 544 +               skb_restore_cb(skb_shared); /* restore skb->cb */
 545 +               skb->nf_queue_entry = NULL;
 546 +               /* qdisc dropped packet and decreased skb reference count of
 547 +                * skb, so we don't really want to and try refree as that would
 548 +                * actually destroy the skb. */
 549 +               spin_unlock(root_lock);
 550 +               goto packet_not_eaten_by_imq_dev;
 551 +       }
 552 +
 553 +packet_not_eaten_by_imq_dev:
 554 +       rcu_read_unlock_bh();
 555 +
 556 +       /* cloned? restore original */
 557 +       if (unlikely(skb_orig)) {
 558 +               kfree_skb(skb);
 559 +               entry->skb = skb_orig;
 560 +       }
 561 +       retval = -1;
 562 +out:
 563 +       return retval;
 564 +}
 565 +
 566 +static unsigned int imq_nf_hook(unsigned int hook, struct sk_buff *pskb,
 567 +                               const struct net_device *indev,
 568 +                               const struct net_device *outdev,
 569 +                               int (*okfn)(struct sk_buff *))
 570 +{
 571 +       return (pskb->imq_flags & IMQ_F_ENQUEUE) ? NF_IMQ_QUEUE : NF_ACCEPT;
 572 +}
 573 +
 574 +static int imq_close(struct net_device *dev)
 575 +{
 576 +       netif_stop_queue(dev);
 577 +       return 0;
 578 +}
 579 +
 580 +static int imq_open(struct net_device *dev)
 581 +{
 582 +       netif_start_queue(dev);
 583 +       return 0;
 584 +}
 585 +
 586 +static const struct net_device_ops imq_netdev_ops = {
 587 +       .ndo_open               = imq_open,
 588 +       .ndo_stop               = imq_close,
 589 +       .ndo_start_xmit         = imq_dev_xmit,
 590 +       .ndo_get_stats          = imq_get_stats,
 591 +};
 592 +
 593 +static void imq_setup(struct net_device *dev)
 594 +{
 595 +       dev->netdev_ops         = &imq_netdev_ops;
 596 +       dev->type               = ARPHRD_VOID;
 597 +       dev->mtu                = 16000;
 598 +       dev->tx_queue_len       = 11000;
 599 +       dev->flags              = IFF_NOARP;
 600 +       dev->features           = NETIF_F_SG | NETIF_F_FRAGLIST |
 601 +                                 NETIF_F_GSO | NETIF_F_HW_CSUM |
 602 +                                 NETIF_F_HIGHDMA;
 603 +       dev->priv_flags         &= ~IFF_XMIT_DST_RELEASE;
 604 +}
 605 +
 606 +static int imq_validate(struct nlattr *tb[], struct nlattr *data[])
 607 +{
 608 +       int ret = 0;
 609 +
 610 +       if (tb[IFLA_ADDRESS]) {
 611 +               if (nla_len(tb[IFLA_ADDRESS]) != ETH_ALEN) {
 612 +                       ret = -EINVAL;
 613 +                       goto end;
 614 +               }
 615 +               if (!is_valid_ether_addr(nla_data(tb[IFLA_ADDRESS]))) {
 616 +                       ret = -EADDRNOTAVAIL;
 617 +                       goto end;
 618 +               }
 619 +       }
 620 +       return 0;
 621 +end:
 622 +       printk(KERN_WARNING "IMQ: imq_validate failed (%d)\n", ret);
 623 +       return ret;
 624 +}
 625 +
 626 +static struct rtnl_link_ops imq_link_ops __read_mostly = {
 627 +       .kind           = "imq",
 628 +       .priv_size      = 0,
 629 +       .setup          = imq_setup,
 630 +       .validate       = imq_validate,
 631 +};
 632 +
 633 +static const struct nf_queue_handler imq_nfqh = {
 634 +       .name  = "imq",
 635 +       .outfn = imq_nf_queue,
 636 +};
 637 +
 638 +static int __init imq_init_hooks(void)
 639 +{
 640 +       int ret;
 641 +
 642 +       nf_register_queue_imq_handler(&imq_nfqh);
 643 +
 644 +       ret = nf_register_hooks(imq_ops, ARRAY_SIZE(imq_ops));
 645 +       if (ret < 0)
 646 +               nf_unregister_queue_imq_handler();
 647 +
 648 +       return ret;
 649 +}
 650 +
 651 +static int __init imq_init_one(int index)
 652 +{
 653 +       struct net_device *dev;
 654 +       int ret;
 655 +
 656 +       dev = alloc_netdev_mq(0, "imq%d", imq_setup, numqueues);
 657 +       if (!dev)
 658 +               return -ENOMEM;
 659 +
 660 +       ret = dev_alloc_name(dev, dev->name);
 661 +       if (ret < 0)
 662 +               goto fail;
 663 +
 664 +       dev->rtnl_link_ops = &imq_link_ops;
 665 +       ret = register_netdevice(dev);
 666 +       if (ret < 0)
 667 +               goto fail;
 668 +
 669 +       return 0;
 670 +fail:
 671 +       free_netdev(dev);
 672 +       return ret;
 673 +}
 674 +
 675 +static int __init imq_init_devs(void)
 676 +{
 677 +       int err, i;
 678 +
 679 +       if (numdevs < 1 || numdevs > IMQ_MAX_DEVS) {
 680 +               printk(KERN_ERR "IMQ: numdevs has to be betweed 1 and %u\n",
 681 +                      IMQ_MAX_DEVS);
 682 +               return -EINVAL;
 683 +       }
 684 +
 685 +       if (numqueues < 1 || numqueues > IMQ_MAX_QUEUES) {
 686 +               printk(KERN_ERR "IMQ: numqueues has to be betweed 1 and %u\n",
 687 +                      IMQ_MAX_QUEUES);
 688 +               return -EINVAL;
 689 +       }
 690 +
 691 +       get_random_bytes(&imq_hashrnd, sizeof(imq_hashrnd));
 692 +
 693 +       rtnl_lock();
 694 +       err = __rtnl_link_register(&imq_link_ops);
 695 +
 696 +       for (i = 0; i < numdevs && !err; i++)
 697 +               err = imq_init_one(i);
 698 +
 699 +       if (err) {
 700 +               __rtnl_link_unregister(&imq_link_ops);
 701 +               memset(imq_devs_cache, 0, sizeof(imq_devs_cache));
 702 +       }
 703 +       rtnl_unlock();
 704 +
 705 +       return err;
 706 +}
 707 +
 708 +static int __init imq_init_module(void)
 709 +{
 710 +       int err;
 711 +
 712 +#if defined(CONFIG_IMQ_NUM_DEVS)
 713 +       BUILD_BUG_ON(CONFIG_IMQ_NUM_DEVS > 16);
 714 +       BUILD_BUG_ON(CONFIG_IMQ_NUM_DEVS < 2);
 715 +       BUILD_BUG_ON(CONFIG_IMQ_NUM_DEVS - 1 > IMQ_F_IFMASK);
 716 +#endif
 717 +
 718 +       err = imq_init_devs();
 719 +       if (err) {
 720 +               printk(KERN_ERR "IMQ: Error trying imq_init_devs(net)\n");
 721 +               return err;
 722 +       }
 723 +
 724 +       err = imq_init_hooks();
 725 +       if (err) {
 726 +               printk(KERN_ERR "IMQ: Error trying imq_init_hooks()\n");
 727 +               rtnl_link_unregister(&imq_link_ops);
 728 +               memset(imq_devs_cache, 0, sizeof(imq_devs_cache));
 729 +               return err;
 730 +       }
 731 +
 732 +       printk(KERN_INFO "IMQ driver loaded successfully. "
 733 +               "(numdevs = %d, numqueues = %d)\n", numdevs, numqueues);
 734 +
 735 +#if defined(CONFIG_IMQ_BEHAVIOR_BA) || defined(CONFIG_IMQ_BEHAVIOR_BB)
 736 +       printk(KERN_INFO "\tHooking IMQ before NAT on PREROUTING.\n");
 737 +#else
 738 +       printk(KERN_INFO "\tHooking IMQ after NAT on PREROUTING.\n");
 739 +#endif
 740 +#if defined(CONFIG_IMQ_BEHAVIOR_AB) || defined(CONFIG_IMQ_BEHAVIOR_BB)
 741 +       printk(KERN_INFO "\tHooking IMQ before NAT on POSTROUTING.\n");
 742 +#else
 743 +       printk(KERN_INFO "\tHooking IMQ after NAT on POSTROUTING.\n");
 744 +#endif
 745 +
 746 +       return 0;
 747 +}
 748 +
 749 +static void __exit imq_unhook(void)
 750 +{
 751 +       nf_unregister_hooks(imq_ops, ARRAY_SIZE(imq_ops));
 752 +       nf_unregister_queue_imq_handler();
 753 +}
 754 +
 755 +static void __exit imq_cleanup_devs(void)
 756 +{
 757 +       rtnl_link_unregister(&imq_link_ops);
 758 +       memset(imq_devs_cache, 0, sizeof(imq_devs_cache));
 759 +}
 760 +
 761 +static void __exit imq_exit_module(void)
 762 +{
 763 +       imq_unhook();
 764 +       imq_cleanup_devs();
 765 +       printk(KERN_INFO "IMQ driver unloaded successfully.\n");
 766 +}
 767 +
 768 +module_init(imq_init_module);
 769 +module_exit(imq_exit_module);
 770 +
 771 +module_param(numdevs, int, 0);
 772 +module_param(numqueues, int, 0);
 773 +MODULE_PARM_DESC(numdevs, "number of IMQ devices (how many imq* devices will "
 774 +                       "be created)");
 775 +MODULE_PARM_DESC(numqueues, "number of queues per IMQ device");
 776 +MODULE_AUTHOR("http://www.linuximq.net");
 777 +MODULE_DESCRIPTION("Pseudo-driver for the intermediate queue device. See "
 778 +                       "http://www.linuximq.net/ for more information.");
 779 +MODULE_LICENSE("GPL");
 780 +MODULE_ALIAS_RTNL_LINK("imq");
 781 +
 782 diff -uNr linux-2.6.39/drivers/net/Kconfig linux-2.6.39-imqmq/drivers/net/Kconfig
 783 --- linux-2.6.39/drivers/net/Kconfig    2011-05-19 07:06:34.000000000 +0300
 784 +++ linux-2.6.39-imqmq/drivers/net/Kconfig      2011-05-19 11:08:04.281864473 +0300
 785 @@ -124,6 +124,129 @@
 786           To compile this driver as a module, choose M here: the module
 787           will be called eql.  If unsure, say N.
 788
 789 +config IMQ
 790 +       tristate "IMQ (intermediate queueing device) support"
 791 +       depends on NETDEVICES && NETFILTER
 792 +       ---help---
 793 +         The IMQ device(s) is used as placeholder for QoS queueing
 794 +         disciplines. Every packet entering/leaving the IP stack can be
 795 +         directed through the IMQ device where it's enqueued/dequeued to the
 796 +         attached qdisc. This allows you to treat network devices as classes
 797 +         and distribute bandwidth among them. Iptables is used to specify
 798 +         through which IMQ device, if any, packets travel.
 799 +
 800 +         More information at: http://www.linuximq.net/
 801 +
 802 +         To compile this driver as a module, choose M here: the module
 803 +         will be called imq.  If unsure, say N.
 804 +
 805 +choice
 806 +       prompt "IMQ behavior (PRE/POSTROUTING)"
 807 +       depends on IMQ
 808 +       default IMQ_BEHAVIOR_AB
 809 +       help
 810 +
 811 +               This settings defines how IMQ behaves in respect to its
 812 +               hooking in PREROUTING and POSTROUTING.
 813 +
 814 +               IMQ can work in any of the following ways:
 815 +
 816 +                   PREROUTING   |      POSTROUTING
 817 +               -----------------|-------------------
 818 +               #1  After NAT    |      After NAT
 819 +               #2  After NAT    |      Before NAT
 820 +               #3  Before NAT   |      After NAT
 821 +               #4  Before NAT   |      Before NAT
 822 +
 823 +               The default behavior is to hook before NAT on PREROUTING
 824 +               and after NAT on POSTROUTING (#3).
 825 +
 826 +               This settings are specially usefull when trying to use IMQ
 827 +               to shape NATed clients.
 828 +
 829 +               More information can be found at: www.linuximq.net
 830 +
 831 +               If not sure leave the default settings alone.
 832 +
 833 +config IMQ_BEHAVIOR_AA
 834 +       bool "IMQ AA"
 835 +       help
 836 +               This settings defines how IMQ behaves in respect to its
 837 +               hooking in PREROUTING and POSTROUTING.
 838 +
 839 +               Choosing this option will make IMQ hook like this:
 840 +
 841 +               PREROUTING:   After NAT
 842 +               POSTROUTING:  After NAT
 843 +
 844 +               More information can be found at: www.linuximq.net
 845 +
 846 +               If not sure leave the default settings alone.
 847 +
 848 +config IMQ_BEHAVIOR_AB
 849 +       bool "IMQ AB"
 850 +       help
 851 +               This settings defines how IMQ behaves in respect to its
 852 +               hooking in PREROUTING and POSTROUTING.
 853 +
 854 +               Choosing this option will make IMQ hook like this:
 855 +
 856 +               PREROUTING:   After NAT
 857 +               POSTROUTING:  Before NAT
 858 +
 859 +               More information can be found at: www.linuximq.net
 860 +
 861 +               If not sure leave the default settings alone.
 862 +
 863 +config IMQ_BEHAVIOR_BA
 864 +       bool "IMQ BA"
 865 +       help
 866 +               This settings defines how IMQ behaves in respect to its
 867 +               hooking in PREROUTING and POSTROUTING.
 868 +
 869 +               Choosing this option will make IMQ hook like this:
 870 +
 871 +               PREROUTING:   Before NAT
 872 +               POSTROUTING:  After NAT
 873 +
 874 +               More information can be found at: www.linuximq.net
 875 +
 876 +               If not sure leave the default settings alone.
 877 +
 878 +config IMQ_BEHAVIOR_BB
 879 +       bool "IMQ BB"
 880 +       help
 881 +               This settings defines how IMQ behaves in respect to its
 882 +               hooking in PREROUTING and POSTROUTING.
 883 +
 884 +               Choosing this option will make IMQ hook like this:
 885 +
 886 +               PREROUTING:   Before NAT
 887 +               POSTROUTING:  Before NAT
 888 +
 889 +               More information can be found at: www.linuximq.net
 890 +
 891 +               If not sure leave the default settings alone.
 892 +
 893 +endchoice
 894 +
 895 +config IMQ_NUM_DEVS
 896 +
 897 +       int "Number of IMQ devices"
 898 +       range 2 16
 899 +       depends on IMQ
 900 +       default "16"
 901 +       help
 902 +
 903 +               This settings defines how many IMQ devices will be
 904 +               created.
 905 +
 906 +               The default value is 16.
 907 +
 908 +               More information can be found at: www.linuximq.net
 909 +
 910 +               If not sure leave the default settings alone.
 911 +
 912  config TUN
 913         tristate "Universal TUN/TAP device driver support"
 914         select CRC32
 915 diff -uNr linux-2.6.39/drivers/net/Makefile linux-2.6.39-imqmq/drivers/net/Makefile
 916 --- linux-2.6.39/drivers/net/Makefile   2011-05-19 07:06:34.000000000 +0300
 917 +++ linux-2.6.39-imqmq/drivers/net/Makefile     2011-05-19 11:08:04.281864473 +0300
 918 @@ -175,6 +175,7 @@
 919  obj-$(CONFIG_XEN_NETDEV_BACKEND) += xen-netback/
 920
 921  obj-$(CONFIG_DUMMY) += dummy.o
 922 +obj-$(CONFIG_IMQ) += imq.o
 923  obj-$(CONFIG_IFB) += ifb.o
 924  obj-$(CONFIG_MACVLAN) += macvlan.o
 925  obj-$(CONFIG_MACVTAP) += macvtap.o
 926 diff -uNr linux-2.6.39/include/linux/imq.h linux-2.6.39-imqmq/include/linux/imq.h
 927 --- linux-2.6.39/include/linux/imq.h    1970-01-01 02:00:00.000000000 +0200
 928 +++ linux-2.6.39-imqmq/include/linux/imq.h      2011-05-19 11:08:04.281864473 +0300
 929 @@ -0,0 +1,13 @@
 930 +#ifndef _IMQ_H
 931 +#define _IMQ_H
 932 +
 933 +/* IFMASK (16 device indexes, 0 to 15) and flag(s) fit in 5 bits */
 934 +#define IMQ_F_BITS     5
 935 +
 936 +#define IMQ_F_IFMASK   0x0f
 937 +#define IMQ_F_ENQUEUE  0x10
 938 +
 939 +#define IMQ_MAX_DEVS   (IMQ_F_IFMASK + 1)
 940 +
 941 +#endif /* _IMQ_H */
 942 +
 943 diff -uNr linux-2.6.39/include/linux/netfilter/xt_IMQ.h linux-2.6.39-imqmq/include/linux/netfilter/xt_IMQ.h
 944 --- linux-2.6.39/include/linux/netfilter/xt_IMQ.h       1970-01-01 02:00:00.000000000 +0200
 945 +++ linux-2.6.39-imqmq/include/linux/netfilter/xt_IMQ.h 2011-05-19 11:08:04.281864473 +0300
 946 @@ -0,0 +1,9 @@
 947 +#ifndef _XT_IMQ_H
 948 +#define _XT_IMQ_H
 949 +
 950 +struct xt_imq_info {
 951 +       unsigned int todev;     /* target imq device */
 952 +};
 953 +
 954 +#endif /* _XT_IMQ_H */
 955 +
 956 diff -uNr linux-2.6.39/include/linux/netfilter.h linux-2.6.39-imqmq/include/linux/netfilter.h
 957 --- linux-2.6.39/include/linux/netfilter.h      2011-05-19 07:06:34.000000000 +0300
 958 +++ linux-2.6.39-imqmq/include/linux/netfilter.h        2011-05-19 11:08:04.285197874 +0300
 959 @@ -21,7 +21,8 @@
 960  #define NF_QUEUE 3
 961  #define NF_REPEAT 4
 962  #define NF_STOP 5
 963 -#define NF_MAX_VERDICT NF_STOP
 964 +#define NF_IMQ_QUEUE 6
 965 +#define NF_MAX_VERDICT NF_IMQ_QUEUE
 966
 967  /* we overload the higher bits for encoding auxiliary data such as the queue
 968   * number or errno values. Not nice, but better than additional function
 969 diff -uNr linux-2.6.39/include/linux/netfilter_ipv4/ipt_IMQ.h linux-2.6.39-imqmq/include/linux/netfilter_ipv4/ipt_IMQ.h
 970 --- linux-2.6.39/include/linux/netfilter_ipv4/ipt_IMQ.h 1970-01-01 02:00:00.000000000 +0200
 971 +++ linux-2.6.39-imqmq/include/linux/netfilter_ipv4/ipt_IMQ.h   2011-05-19 11:08:04.285197874 +0300
 972 @@ -0,0 +1,10 @@
 973 +#ifndef _IPT_IMQ_H
 974 +#define _IPT_IMQ_H
 975 +
 976 +/* Backwards compatibility for old userspace */
 977 +#include <linux/netfilter/xt_IMQ.h>
 978 +
 979 +#define ipt_imq_info xt_imq_info
 980 +
 981 +#endif /* _IPT_IMQ_H */
 982 +
 983 diff -uNr linux-2.6.39/include/linux/netfilter_ipv6/ip6t_IMQ.h linux-2.6.39-imqmq/include/linux/netfilter_ipv6/ip6t_IMQ.h
 984 --- linux-2.6.39/include/linux/netfilter_ipv6/ip6t_IMQ.h        1970-01-01 02:00:00.000000000 +0200
 985 +++ linux-2.6.39-imqmq/include/linux/netfilter_ipv6/ip6t_IMQ.h  2011-05-19 11:08:04.285197874 +0300
 986 @@ -0,0 +1,10 @@
 987 +#ifndef _IP6T_IMQ_H
 988 +#define _IP6T_IMQ_H
 989 +
 990 +/* Backwards compatibility for old userspace */
 991 +#include <linux/netfilter/xt_IMQ.h>
 992 +
 993 +#define ip6t_imq_info xt_imq_info
 994 +
 995 +#endif /* _IP6T_IMQ_H */
 996 +
 997 diff -uNr linux-2.6.39/include/linux/skbuff.h linux-2.6.39-imqmq/include/linux/skbuff.h
 998 --- linux-2.6.39/include/linux/skbuff.h 2011-05-19 07:06:34.000000000 +0300
 999 +++ linux-2.6.39-imqmq/include/linux/skbuff.h   2011-05-19 11:08:04.288531274 +0300
1000 @@ -29,6 +29,9 @@
1001  #include <linux/rcupdate.h>
1002  #include <linux/dmaengine.h>
1003  #include <linux/hrtimer.h>
1004 +#if defined(CONFIG_IMQ) || defined(CONFIG_IMQ_MODULE)
1005 +#include <linux/imq.h>
1006 +#endif
1007
1008  /* Don't change this without changing skb_csum_unnecessary! */
1009  #define CHECKSUM_NONE 0
1010 @@ -339,6 +342,9 @@
1011          * first. This is owned by whoever has the skb queued ATM.
1012          */
1013         char                    cb[48] __aligned(8);
1014 +#if defined(CONFIG_IMQ) || defined(CONFIG_IMQ_MODULE)
1015 +       void                    *cb_next;
1016 +#endif
1017
1018         unsigned long           _skb_refdst;
1019  #ifdef CONFIG_XFRM
1020 @@ -377,6 +383,9 @@
1021  #ifdef NET_SKBUFF_NF_DEFRAG_NEEDED
1022         struct sk_buff          *nfct_reasm;
1023  #endif
1024 +#if defined(CONFIG_IMQ) || defined(CONFIG_IMQ_MODULE)
1025 +       struct nf_queue_entry   *nf_queue_entry;
1026 +#endif
1027  #ifdef CONFIG_BRIDGE_NETFILTER
1028         struct nf_bridge_info   *nf_bridge;
1029  #endif
1030 @@ -401,6 +410,10 @@
1031
1032         /* 0/13 bit hole */
1033
1034 +#if defined(CONFIG_IMQ) || defined(CONFIG_IMQ_MODULE)
1035 +       __u8                    imq_flags:IMQ_F_BITS;
1036 +#endif
1037 +
1038  #ifdef CONFIG_NET_DMA
1039         dma_cookie_t            dma_cookie;
1040  #endif
1041 @@ -487,6 +500,12 @@
1042         return (struct rtable *)skb_dst(skb);
1043  }
1044
1045 +
1046 +#if defined(CONFIG_IMQ) || defined(CONFIG_IMQ_MODULE)
1047 +extern int skb_save_cb(struct sk_buff *skb);
1048 +extern int skb_restore_cb(struct sk_buff *skb);
1049 +#endif
1050 +
1051  extern void kfree_skb(struct sk_buff *skb);
1052  extern void consume_skb(struct sk_buff *skb);
1053  extern void           __kfree_skb(struct sk_buff *skb);
1054 @@ -2129,6 +2148,10 @@
1055         dst->nfct_reasm = src->nfct_reasm;
1056         nf_conntrack_get_reasm(src->nfct_reasm);
1057  #endif
1058 +#if defined(CONFIG_IMQ) || defined(CONFIG_IMQ_MODULE)
1059 +       dst->imq_flags = src->imq_flags;
1060 +       dst->nf_queue_entry = src->nf_queue_entry;
1061 +#endif
1062  #ifdef CONFIG_BRIDGE_NETFILTER
1063         dst->nf_bridge  = src->nf_bridge;
1064         nf_bridge_get(src->nf_bridge);
1065 diff -uNr linux-2.6.39/include/net/netfilter/nf_queue.h linux-2.6.39-imqmq/include/net/netfilter/nf_queue.h
1066 --- linux-2.6.39/include/net/netfilter/nf_queue.h       2011-05-19 07:06:34.000000000 +0300
1067 +++ linux-2.6.39-imqmq/include/net/netfilter/nf_queue.h 2011-05-19 11:08:04.288531274 +0300
1068 @@ -30,5 +30,11 @@
1069                                        const struct nf_queue_handler *qh);
1070  extern void nf_unregister_queue_handlers(const struct nf_queue_handler *qh);
1071  extern void nf_reinject(struct nf_queue_entry *entry, unsigned int verdict);
1072 +extern void nf_queue_entry_release_refs(struct nf_queue_entry *entry);
1073 +
1074 +#if defined(CONFIG_IMQ) || defined(CONFIG_IMQ_MODULE)
1075 +extern void nf_register_queue_imq_handler(const struct nf_queue_handler *qh);
1076 +extern void nf_unregister_queue_imq_handler(void);
1077 +#endif
1078
1079  #endif /* _NF_QUEUE_H */
1080 diff -uNr linux-2.6.39/net/core/dev.c linux-2.6.39-imqmq/net/core/dev.c
1081 --- linux-2.6.39/net/core/dev.c 2011-05-19 07:06:34.000000000 +0300
1082 +++ linux-2.6.39-imqmq/net/core/dev.c   2011-05-19 11:08:04.288531274 +0300
1083 @@ -98,6 +98,9 @@
1084  #include <net/net_namespace.h>
1085  #include <net/sock.h>
1086  #include <linux/rtnetlink.h>
1087 +#if defined(CONFIG_IMQ) || defined(CONFIG_IMQ_MODULE)
1088 +#include <linux/imq.h>
1089 +#endif
1090  #include <linux/proc_fs.h>
1091  #include <linux/seq_file.h>
1092  #include <linux/stat.h>
1093 @@ -2099,12 +2102,21 @@
1094                 if (dev->priv_flags & IFF_XMIT_DST_RELEASE)
1095                         skb_dst_drop(skb);
1096
1097 +#if defined(CONFIG_IMQ) || defined(CONFIG_IMQ_MODULE)
1098 +               if (!list_empty(&ptype_all) &&
1099 +                                       !(skb->imq_flags & IMQ_F_ENQUEUE))
1100 +#else
1101                 if (!list_empty(&ptype_all))
1102 +#endif
1103                         dev_queue_xmit_nit(skb, dev);
1104
1105                 skb_orphan_try(skb);
1106
1107 +#if defined(CONFIG_IMQ) || defined(CONFIG_IMQ_MODULE)
1108 +               features = skb->dev ? netif_skb_features(skb) : dev->features;
1109 +#else
1110                 features = netif_skb_features(skb);
1111 +#endif
1112
1113                 if (vlan_tx_tag_present(skb) &&
1114                     !(features & NETIF_F_HW_VLAN_TX)) {
1115 @@ -2269,8 +2281,7 @@
1116  #endif
1117  }
1118
1119 -static struct netdev_queue *dev_pick_tx(struct net_device *dev,
1120 -                                       struct sk_buff *skb)
1121 +static struct netdev_queue *dev_pick_tx(struct net_device *dev, struct sk_buff *skb)
1122  {
1123         int queue_index;
1124         const struct net_device_ops *ops = dev->netdev_ops;
1125 diff -uNr linux-2.6.39/net/core/skbuff.c linux-2.6.39-imqmq/net/core/skbuff.c
1126 --- linux-2.6.39/net/core/skbuff.c      2011-05-19 07:06:34.000000000 +0300
1127 +++ linux-2.6.39-imqmq/net/core/skbuff.c        2011-05-19 11:08:04.288531274 +0300
1128 @@ -72,6 +72,9 @@
1129
1130  static struct kmem_cache *skbuff_head_cache __read_mostly;
1131  static struct kmem_cache *skbuff_fclone_cache __read_mostly;
1132 +#if defined(CONFIG_IMQ) || defined(CONFIG_IMQ_MODULE)
1133 +static struct kmem_cache *skbuff_cb_store_cache __read_mostly;
1134 +#endif
1135
1136  static void sock_pipe_buf_release(struct pipe_inode_info *pipe,
1137                                   struct pipe_buffer *buf)
1138 @@ -91,6 +94,82 @@
1139         return 1;
1140  }
1141
1142 +#if defined(CONFIG_IMQ) || defined(CONFIG_IMQ_MODULE)
1143 +/* Control buffer save/restore for IMQ devices */
1144 +struct skb_cb_table {
1145 +       char                    cb[48] __aligned(8);
1146 +       void                    *cb_next;
1147 +       atomic_t                refcnt;
1148 +};
1149 +
1150 +static DEFINE_SPINLOCK(skb_cb_store_lock);
1151 +
1152 +int skb_save_cb(struct sk_buff *skb)
1153 +{
1154 +       struct skb_cb_table *next;
1155 +
1156 +       next = kmem_cache_alloc(skbuff_cb_store_cache, GFP_ATOMIC);
1157 +       if (!next)
1158 +               return -ENOMEM;
1159 +
1160 +       BUILD_BUG_ON(sizeof(skb->cb) != sizeof(next->cb));
1161 +
1162 +       memcpy(next->cb, skb->cb, sizeof(skb->cb));
1163 +       next->cb_next = skb->cb_next;
1164 +
1165 +       atomic_set(&next->refcnt, 1);
1166 +
1167 +       skb->cb_next = next;
1168 +       return 0;
1169 +}
1170 +EXPORT_SYMBOL(skb_save_cb);
1171 +
1172 +int skb_restore_cb(struct sk_buff *skb)
1173 +{
1174 +       struct skb_cb_table *next;
1175 +
1176 +       if (!skb->cb_next)
1177 +               return 0;
1178 +
1179 +       next = skb->cb_next;
1180 +
1181 +       BUILD_BUG_ON(sizeof(skb->cb) != sizeof(next->cb));
1182 +
1183 +       memcpy(skb->cb, next->cb, sizeof(skb->cb));
1184 +       skb->cb_next = next->cb_next;
1185 +
1186 +       spin_lock(&skb_cb_store_lock);
1187 +
1188 +       if (atomic_dec_and_test(&next->refcnt))
1189 +               kmem_cache_free(skbuff_cb_store_cache, next);
1190 +
1191 +       spin_unlock(&skb_cb_store_lock);
1192 +
1193 +       return 0;
1194 +}
1195 +EXPORT_SYMBOL(skb_restore_cb);
1196 +
1197 +static void skb_copy_stored_cb(struct sk_buff *new, const struct sk_buff *__old)
1198 +{
1199 +       struct skb_cb_table *next;
1200 +       struct sk_buff *old;
1201 +
1202 +       if (!__old->cb_next) {
1203 +               new->cb_next = NULL;
1204 +               return;
1205 +       }
1206 +
1207 +       spin_lock(&skb_cb_store_lock);
1208 +
1209 +       old = (struct sk_buff *)__old;
1210 +
1211 +       next = old->cb_next;
1212 +       atomic_inc(&next->refcnt);
1213 +       new->cb_next = next;
1214 +
1215 +       spin_unlock(&skb_cb_store_lock);
1216 +}
1217 +#endif
1218
1219  /* Pipe buffer operations for a socket. */
1220  static const struct pipe_buf_operations sock_pipe_buf_ops = {
1221 @@ -379,6 +458,26 @@
1222                 WARN_ON(in_irq());
1223                 skb->destructor(skb);
1224         }
1225 +#if defined(CONFIG_IMQ) || defined(CONFIG_IMQ_MODULE)
1226 +       /* This should not happen. When it does, avoid memleak by restoring
1227 +       the chain of cb-backups. */
1228 +       while (skb->cb_next != NULL) {
1229 +               if (net_ratelimit())
1230 +                       printk(KERN_WARNING "IMQ: kfree_skb: skb->cb_next: "
1231 +                               "%08x\n", (unsigned int)skb->cb_next);
1232 +
1233 +               skb_restore_cb(skb);
1234 +       }
1235 +       /* This should not happen either, nf_queue_entry is nullified in
1236 +        * imq_dev_xmit(). If we have non-NULL nf_queue_entry then we are
1237 +        * leaking entry pointers, maybe memory. We don't know if this is
1238 +        * pointer to already freed memory, or should this be freed.
1239 +        * If this happens we need to add refcounting, etc for nf_queue_entry.
1240 +        */
1241 +       if (skb->nf_queue_entry && net_ratelimit())
1242 +               printk(KERN_WARNING
1243 +                               "IMQ: kfree_skb: skb->nf_queue_entry != NULL");
1244 +#endif
1245  #if defined(CONFIG_NF_CONNTRACK) || defined(CONFIG_NF_CONNTRACK_MODULE)
1246         nf_conntrack_put(skb->nfct);
1247  #endif
1248 @@ -517,6 +616,9 @@
1249         new->sp                 = secpath_get(old->sp);
1250  #endif
1251         memcpy(new->cb, old->cb, sizeof(old->cb));
1252 +#if defined(CONFIG_IMQ) || defined(CONFIG_IMQ_MODULE)
1253 +       skb_copy_stored_cb(new, old);
1254 +#endif
1255         new->csum               = old->csum;
1256         new->local_df           = old->local_df;
1257         new->pkt_type           = old->pkt_type;
1258 @@ -2780,6 +2882,13 @@
1259                                                 0,
1260                                                 SLAB_HWCACHE_ALIGN|SLAB_PANIC,
1261                                                 NULL);
1262 +#if defined(CONFIG_IMQ) || defined(CONFIG_IMQ_MODULE)
1263 +       skbuff_cb_store_cache = kmem_cache_create("skbuff_cb_store_cache",
1264 +                                                 sizeof(struct skb_cb_table),
1265 +                                                 0,
1266 +                                                 SLAB_HWCACHE_ALIGN|SLAB_PANIC,
1267 +                                                 NULL);
1268 +#endif
1269  }
1270
1271  /**
1272 diff -uNr linux-2.6.39/net/netfilter/core.c linux-2.6.39-imqmq/net/netfilter/core.c
1273 --- linux-2.6.39/net/netfilter/core.c   2011-05-19 07:06:34.000000000 +0300
1274 +++ linux-2.6.39-imqmq/net/netfilter/core.c     2011-05-19 11:13:19.891558119 +0300
1275 @@ -191,6 +191,20 @@
1276                         kfree_skb(skb);
1277                 }
1278                 ret = 0;
1279 +#if defined(CONFIG_IMQ) || defined(CONFIG_IMQ_MODULE)
1280 +       } else if ((verdict & NF_VERDICT_MASK) == NF_IMQ_QUEUE) {
1281 +               ret = nf_imq_queue(skb, elem, pf, hook, indev, outdev, okfn,
1282 +                              verdict >> NF_VERDICT_QBITS);
1283 +               if (ret < 0) {
1284 +                       if (ret == -ECANCELED)
1285 +                               goto next_hook;
1286 +                       if (ret == -ESRCH &&
1287 +                          (verdict & NF_VERDICT_FLAG_QUEUE_BYPASS))
1288 +                               goto next_hook;
1289 +                       kfree_skb(skb);
1290 +               }
1291 +               ret = 0;
1292 +#endif
1293         }
1294         rcu_read_unlock();
1295         return ret;
1296 diff -uNr linux-2.6.39/net/netfilter/Kconfig linux-2.6.39-imqmq/net/netfilter/Kconfig
1297 --- linux-2.6.39/net/netfilter/Kconfig  2011-05-19 07:06:34.000000000 +0300
1298 +++ linux-2.6.39-imqmq/net/netfilter/Kconfig    2011-05-19 11:08:04.288531274 +0300
1299 @@ -507,6 +507,18 @@
1300           For more information on the LEDs available on your system, see
1301           Documentation/leds-class.txt
1302
1303 +config NETFILTER_XT_TARGET_IMQ
1304 +        tristate '"IMQ" target support'
1305 +       depends on NETFILTER_XTABLES
1306 +       depends on IP_NF_MANGLE || IP6_NF_MANGLE
1307 +       select IMQ
1308 +       default m if NETFILTER_ADVANCED=n
1309 +        help
1310 +          This option adds a `IMQ' target which is used to specify if and
1311 +          to which imq device packets should get enqueued/dequeued.
1312 +
1313 +          To compile it as a module, choose M here.  If unsure, say N.
1314 +
1315  config NETFILTER_XT_TARGET_MARK
1316         tristate '"MARK" target support'
1317         depends on NETFILTER_ADVANCED
1318 diff -uNr linux-2.6.39/net/netfilter/Makefile linux-2.6.39-imqmq/net/netfilter/Makefile
1319 --- linux-2.6.39/net/netfilter/Makefile 2011-05-19 07:06:34.000000000 +0300
1320 +++ linux-2.6.39-imqmq/net/netfilter/Makefile   2011-05-19 11:08:04.291864674 +0300
1321 @@ -56,6 +56,7 @@
1322  obj-$(CONFIG_NETFILTER_XT_TARGET_CT) += xt_CT.o
1323  obj-$(CONFIG_NETFILTER_XT_TARGET_DSCP) += xt_DSCP.o
1324  obj-$(CONFIG_NETFILTER_XT_TARGET_HL) += xt_HL.o
1325 +obj-$(CONFIG_NETFILTER_XT_TARGET_IMQ) += xt_IMQ.o
1326  obj-$(CONFIG_NETFILTER_XT_TARGET_LED) += xt_LED.o
1327  obj-$(CONFIG_NETFILTER_XT_TARGET_NFLOG) += xt_NFLOG.o
1328  obj-$(CONFIG_NETFILTER_XT_TARGET_NFQUEUE) += xt_NFQUEUE.o
1329 diff -uNr linux-2.6.39/net/netfilter/nf_internals.h linux-2.6.39-imqmq/net/netfilter/nf_internals.h
1330 --- linux-2.6.39/net/netfilter/nf_internals.h   2011-05-19 07:06:34.000000000 +0300
1331 +++ linux-2.6.39-imqmq/net/netfilter/nf_internals.h     2011-05-19 11:08:04.291864674 +0300
1332 @@ -30,6 +30,15 @@
1333                     struct net_device *outdev,
1334                     int (*okfn)(struct sk_buff *),
1335                     unsigned int queuenum);
1336 +#if defined(CONFIG_IMQ) || defined(CONFIG_IMQ_MODULE)
1337 +extern int nf_imq_queue(struct sk_buff *skb,
1338 +                   struct list_head *elem,
1339 +                   u_int8_t pf, unsigned int hook,
1340 +                   struct net_device *indev,
1341 +                   struct net_device *outdev,
1342 +                   int (*okfn)(struct sk_buff *),
1343 +                   unsigned int queuenum);
1344 +#endif
1345  extern int __init netfilter_queue_init(void);
1346
1347  /* nf_log.c */
1348 diff -uNr linux-2.6.39/net/netfilter/nf_queue.c linux-2.6.39-imqmq/net/netfilter/nf_queue.c
1349 --- linux-2.6.39/net/netfilter/nf_queue.c       2011-05-19 07:06:34.000000000 +0300
1350 +++ linux-2.6.39-imqmq/net/netfilter/nf_queue.c 2011-05-19 11:22:38.189467462 +0300
1351 @@ -22,6 +22,26 @@
1352
1353  static DEFINE_MUTEX(queue_handler_mutex);
1354
1355 +#if defined(CONFIG_IMQ) || defined(CONFIG_IMQ_MODULE)
1356 +static const struct nf_queue_handler *queue_imq_handler;
1357 +
1358 +void nf_register_queue_imq_handler(const struct nf_queue_handler *qh)
1359 +{
1360 +       mutex_lock(&queue_handler_mutex);
1361 +       rcu_assign_pointer(queue_imq_handler, qh);
1362 +       mutex_unlock(&queue_handler_mutex);
1363 +}
1364 +EXPORT_SYMBOL_GPL(nf_register_queue_imq_handler);
1365 +
1366 +void nf_unregister_queue_imq_handler(void)
1367 +{
1368 +       mutex_lock(&queue_handler_mutex);
1369 +       rcu_assign_pointer(queue_imq_handler, NULL);
1370 +       mutex_unlock(&queue_handler_mutex);
1371 +}
1372 +EXPORT_SYMBOL_GPL(nf_unregister_queue_imq_handler);
1373 +#endif
1374 +
1375  /* return EBUSY when somebody else is registered, return EEXIST if the
1376   * same handler is registered, return 0 in case of success. */
1377  int nf_register_queue_handler(u_int8_t pf, const struct nf_queue_handler *qh)
1378 @@ -92,7 +112,7 @@
1379  }
1380  EXPORT_SYMBOL_GPL(nf_unregister_queue_handlers);
1381
1382 -static void nf_queue_entry_release_refs(struct nf_queue_entry *entry)
1383 +void nf_queue_entry_release_refs(struct nf_queue_entry *entry)
1384  {
1385         /* Release those devices we held, or Alexey will kill me. */
1386         if (entry->indev)
1387 @@ -112,6 +132,7 @@
1388         /* Drop reference to owner of hook which queued us. */
1389         module_put(entry->elem->owner);
1390  }
1391 +EXPORT_SYMBOL_GPL(nf_queue_entry_release_refs);
1392
1393  /*
1394   * Any packet that leaves via this function must come back
1395 @@ -123,7 +144,8 @@
1396                       struct net_device *indev,
1397                       struct net_device *outdev,
1398                       int (*okfn)(struct sk_buff *),
1399 -                     unsigned int queuenum)
1400 +                     unsigned int queuenum,
1401 +                     bool imq_queue)
1402  {
1403         int status = -ENOENT;
1404         struct nf_queue_entry *entry = NULL;
1405 @@ -137,7 +159,14 @@
1406         /* QUEUE == DROP if no one is waiting, to be safe. */
1407         rcu_read_lock();
1408
1409 -       qh = rcu_dereference(queue_handler[pf]);
1410 +#if defined(CONFIG_IMQ) || defined(CONFIG_IMQ_MODULE)
1411 +       if (imq_queue)
1412 +               qh = rcu_dereference(queue_imq_handler);
1413 +       else
1414 +               qh = rcu_dereference(queue_handler[pf]);
1415 +#else
1416 +       qh = rcu_dereference(queue_handler[pf]);
1417 +#endif
1418         if (!qh) {
1419                 status = -ESRCH;
1420                 goto err_unlock;
1421 @@ -203,13 +232,14 @@
1422         return status;
1423  }
1424
1425 -int nf_queue(struct sk_buff *skb,
1426 -            struct list_head *elem,
1427 -            u_int8_t pf, unsigned int hook,
1428 -            struct net_device *indev,
1429 -            struct net_device *outdev,
1430 -            int (*okfn)(struct sk_buff *),
1431 -            unsigned int queuenum)
1432 +static int _nf_queue(struct sk_buff *skb,
1433 +                    struct list_head *elem,
1434 +                    u_int8_t pf, unsigned int hook,
1435 +                    struct net_device *indev,
1436 +                    struct net_device *outdev,
1437 +                    int (*okfn)(struct sk_buff *),
1438 +                    unsigned int queuenum,
1439 +                    bool imq_queue)
1440  {
1441         struct sk_buff *segs;
1442         int err;
1443 @@ -217,7 +247,7 @@
1444
1445         if (!skb_is_gso(skb))
1446                 return __nf_queue(skb, elem, pf, hook, indev, outdev, okfn,
1447 -                                 queuenum);
1448 +                                 queuenum, imq_queue);
1449
1450         switch (pf) {
1451         case NFPROTO_IPV4:
1452 @@ -244,7 +274,7 @@
1453                 segs->next = NULL;
1454                 if (err == 0)
1455                         err = __nf_queue(segs, elem, pf, hook, indev,
1456 -                                          outdev, okfn, queuenum);
1457 +                                        outdev, okfn, queuenum, imq_queue);
1458                 if (err == 0)
1459                         queued++;
1460                 else
1461 @@ -260,6 +290,32 @@
1462         return err;
1463  }
1464
1465 +int nf_queue(struct sk_buff *skb,
1466 +            struct list_head *elem,
1467 +            u_int8_t pf, unsigned int hook,
1468 +            struct net_device *indev,
1469 +            struct net_device *outdev,
1470 +            int (*okfn)(struct sk_buff *),
1471 +            unsigned int queuenum)
1472 +{
1473 +       return _nf_queue(skb, elem, pf, hook, indev, outdev, okfn, queuenum,
1474 +                        false);
1475 +}
1476 +
1477 +#if defined(CONFIG_IMQ) || defined(CONFIG_IMQ_MODULE)
1478 +int nf_imq_queue(struct sk_buff *skb,
1479 +            struct list_head *elem,
1480 +            u_int8_t pf, unsigned int hook,
1481 +            struct net_device *indev,
1482 +            struct net_device *outdev,
1483 +            int (*okfn)(struct sk_buff *),
1484 +            unsigned int queuenum)
1485 +{
1486 +       return _nf_queue(skb, elem, pf, hook, indev, outdev, okfn, queuenum,
1487 +                        true);
1488 +}
1489 +#endif
1490 +
1491  void nf_reinject(struct nf_queue_entry *entry, unsigned int verdict)
1492  {
1493         struct sk_buff *skb = entry->skb;
1494 @@ -301,7 +357,7 @@
1495         case NF_QUEUE:
1496                 err = __nf_queue(skb, elem, entry->pf, entry->hook,
1497                                  entry->indev, entry->outdev, entry->okfn,
1498 -                                verdict >> NF_VERDICT_QBITS);
1499 +                                verdict >> NF_VERDICT_QBITS, false);
1500                 if (err < 0) {
1501                         if (err == -ECANCELED)
1502                                 goto next_hook;
1503 @@ -311,6 +367,21 @@
1504                         kfree_skb(skb);
1505                 }
1506                 break;
1507 +#if defined(CONFIG_IMQ) || defined(CONFIG_IMQ_MODULE)
1508 +       case NF_IMQ_QUEUE:
1509 +               err = __nf_queue(skb, elem, entry->pf, entry->hook,
1510 +                                entry->indev, entry->outdev, entry->okfn,
1511 +                                verdict >> NF_VERDICT_QBITS, true);
1512 +               if (err < 0) {
1513 +                       if (err == -ECANCELED)
1514 +                               goto next_hook;
1515 +                       if (err == -ESRCH &&
1516 +                          (verdict & NF_VERDICT_FLAG_QUEUE_BYPASS))
1517 +                               goto next_hook;
1518 +                       kfree_skb(skb);
1519 +               }
1520 +               break;
1521 +#endif
1522         case NF_STOLEN:
1523         default:
1524                 kfree_skb(skb);
1525 diff -uNr linux-2.6.39/net/netfilter/xt_IMQ.c linux-2.6.39-imqmq/net/netfilter/xt_IMQ.c
1526 --- linux-2.6.39/net/netfilter/xt_IMQ.c 1970-01-01 02:00:00.000000000 +0200
1527 +++ linux-2.6.39-imqmq/net/netfilter/xt_IMQ.c   2011-05-19 11:08:04.308531677 +0300
1528 @@ -0,0 +1,74 @@
1529 +/*
1530 + * This target marks packets to be enqueued to an imq device
1531 + */
1532 +#include <linux/module.h>
1533 +#include <linux/skbuff.h>
1534 +#include <linux/netfilter/x_tables.h>
1535 +#include <linux/netfilter/xt_IMQ.h>
1536 +#include <linux/imq.h>
1537 +
1538 +static unsigned int imq_target(struct sk_buff *pskb,
1539 +                               const struct xt_action_param *par)
1540 +{
1541 +       const struct xt_imq_info *mr = par->targinfo;
1542 +
1543 +       pskb->imq_flags = (mr->todev & IMQ_F_IFMASK) | IMQ_F_ENQUEUE;
1544 +
1545 +       return XT_CONTINUE;
1546 +}
1547 +
1548 +static int imq_checkentry(const struct xt_tgchk_param *par)
1549 +{
1550 +       struct xt_imq_info *mr = par->targinfo;
1551 +
1552 +       if (mr->todev > IMQ_MAX_DEVS - 1) {
1553 +               printk(KERN_WARNING
1554 +                      "IMQ: invalid device specified, highest is %u\n",
1555 +                      IMQ_MAX_DEVS - 1);
1556 +               return -EINVAL;
1557 +       }
1558 +
1559 +       return 0;
1560 +}
1561 +
1562 +static struct xt_target xt_imq_reg[] __read_mostly = {
1563 +       {
1564 +               .name           = "IMQ",
1565 +               .family         = AF_INET,
1566 +               .checkentry     = imq_checkentry,
1567 +               .target         = imq_target,
1568 +               .targetsize     = sizeof(struct xt_imq_info),
1569 +               .table          = "mangle",
1570 +               .me             = THIS_MODULE
1571 +       },
1572 +       {
1573 +               .name           = "IMQ",
1574 +               .family         = AF_INET6,
1575 +               .checkentry     = imq_checkentry,
1576 +               .target         = imq_target,
1577 +               .targetsize     = sizeof(struct xt_imq_info),
1578 +               .table          = "mangle",
1579 +               .me             = THIS_MODULE
1580 +       },
1581 +};
1582 +
1583 +static int __init imq_init(void)
1584 +{
1585 +       return xt_register_targets(xt_imq_reg, ARRAY_SIZE(xt_imq_reg));
1586 +}
1587 +
1588 +static void __exit imq_fini(void)
1589 +{
1590 +       xt_unregister_targets(xt_imq_reg, ARRAY_SIZE(xt_imq_reg));
1591 +}
1592 +
1593 +module_init(imq_init);
1594 +module_exit(imq_fini);
1595 +
1596 +MODULE_AUTHOR("http://www.linuximq.net");
1597 +MODULE_DESCRIPTION("Pseudo-driver for the intermediate queue device. "
1598 +                  "See http://www.linuximq.net/ for more information.");
1599 +MODULE_LICENSE("GPL");
1600 +MODULE_ALIAS("ipt_IMQ");
1601 +MODULE_ALIAS("ip6t_IMQ");
1602 +