kernel-imq.patch

   1 net: add Intermediate Queueing Device (imq)
   2
   3 From: Jussi Kivilinna <jussi.kivilinna@iki.fi>
   4
   5 This patch is for kernel version 3.12.4+.
   6
   7 See: http://linuximq.net/
   8
   9 Signed-off-by: Jussi Kivilinna <jussi.kivilinna@iki.fi>
  10 ---
  11  drivers/net/Kconfig                     |  119 ++++
  12  drivers/net/Makefile                    |    1
  13  drivers/net/imq.c                       | 1007 +++++++++++++++++++++++++++++++
  14  include/linux/imq.h                     |   13
  15  include/linux/netfilter/xt_IMQ.h        |    9
  16  include/linux/netfilter_ipv4/ipt_IMQ.h  |   10
  17  include/linux/netfilter_ipv6/ip6t_IMQ.h |   10
  18  include/linux/skbuff.h                  |   22 +
  19  include/net/netfilter/nf_queue.h        |    6
  20  include/uapi/linux/netfilter.h          |    3
  21  net/core/dev.c                          |    8
  22  net/core/skbuff.c                       |  112 +++
  23  net/ipv6/ip6_output.c                   |   10
  24  net/netfilter/Kconfig                   |   12
  25  net/netfilter/Makefile                  |    1
  26  net/netfilter/core.c                    |    6
  27  net/netfilter/nf_internals.h            |    2
  28  net/netfilter/nf_queue.c                |   36 +
  29  net/netfilter/xt_IMQ.c                  |   72 ++
  30  19 files changed, 1449 insertions(+), 10 deletions(-)
  31  create mode 100644 drivers/net/imq.c
  32  create mode 100644 include/linux/imq.h
  33  create mode 100644 include/linux/netfilter/xt_IMQ.h
  34  create mode 100644 include/linux/netfilter_ipv4/ipt_IMQ.h
  35  create mode 100644 include/linux/netfilter_ipv6/ip6t_IMQ.h
  36  create mode 100644 net/netfilter/xt_IMQ.c
  37
  38 diff --git a/drivers/net/Kconfig b/drivers/net/Kconfig
  39 index b45b240..5a20da0 100644
  40 --- a/drivers/net/Kconfig
  41 +++ b/drivers/net/Kconfig
  42 @@ -203,6 +203,125 @@ config RIONET_RX_SIZE
  43         depends on RIONET
  44         default "128"
  45
  46 +config IMQ
  47 +       tristate "IMQ (intermediate queueing device) support"
  48 +       depends on NETDEVICES && NETFILTER
  49 +       ---help---
  50 +         The IMQ device(s) is used as placeholder for QoS queueing
  51 +         disciplines. Every packet entering/leaving the IP stack can be
  52 +         directed through the IMQ device where it's enqueued/dequeued to the
  53 +         attached qdisc. This allows you to treat network devices as classes
  54 +         and distribute bandwidth among them. Iptables is used to specify
  55 +         through which IMQ device, if any, packets travel.
  56 +
  57 +         More information at: http://www.linuximq.net/
  58 +
  59 +         To compile this driver as a module, choose M here: the module
  60 +         will be called imq.  If unsure, say N.
  61 +
  62 +choice
  63 +       prompt "IMQ behavior (PRE/POSTROUTING)"
  64 +       depends on IMQ
  65 +       default IMQ_BEHAVIOR_AB
  66 +       help
  67 +         This setting defines how IMQ behaves in respect to its
  68 +         hooking in PREROUTING and POSTROUTING.
  69 +
  70 +         IMQ can work in any of the following ways:
  71 +
  72 +             PREROUTING   |      POSTROUTING
  73 +         -----------------|-------------------
  74 +         #1  After NAT    |      After NAT
  75 +         #2  After NAT    |      Before NAT
  76 +         #3  Before NAT   |      After NAT
  77 +         #4  Before NAT   |      Before NAT
  78 +
  79 +         The default behavior is to hook before NAT on PREROUTING
  80 +         and after NAT on POSTROUTING (#3).
  81 +
  82 +         This settings are specially usefull when trying to use IMQ
  83 +         to shape NATed clients.
  84 +
  85 +         More information can be found at: www.linuximq.net
  86 +
  87 +         If not sure leave the default settings alone.
  88 +
  89 +config IMQ_BEHAVIOR_AA
  90 +       bool "IMQ AA"
  91 +       help
  92 +         This setting defines how IMQ behaves in respect to its
  93 +         hooking in PREROUTING and POSTROUTING.
  94 +
  95 +         Choosing this option will make IMQ hook like this:
  96 +
  97 +         PREROUTING:   After NAT
  98 +         POSTROUTING:  After NAT
  99 +
 100 +         More information can be found at: www.linuximq.net
 101 +
 102 +         If not sure leave the default settings alone.
 103 +
 104 +config IMQ_BEHAVIOR_AB
 105 +       bool "IMQ AB"
 106 +       help
 107 +         This setting defines how IMQ behaves in respect to its
 108 +         hooking in PREROUTING and POSTROUTING.
 109 +
 110 +         Choosing this option will make IMQ hook like this:
 111 +
 112 +         PREROUTING:   After NAT
 113 +         POSTROUTING:  Before NAT
 114 +
 115 +         More information can be found at: www.linuximq.net
 116 +
 117 +         If not sure leave the default settings alone.
 118 +
 119 +config IMQ_BEHAVIOR_BA
 120 +       bool "IMQ BA"
 121 +       help
 122 +         This setting defines how IMQ behaves in respect to its
 123 +         hooking in PREROUTING and POSTROUTING.
 124 +
 125 +         Choosing this option will make IMQ hook like this:
 126 +
 127 +         PREROUTING:   Before NAT
 128 +         POSTROUTING:  After NAT
 129 +
 130 +         More information can be found at: www.linuximq.net
 131 +
 132 +         If not sure leave the default settings alone.
 133 +
 134 +config IMQ_BEHAVIOR_BB
 135 +       bool "IMQ BB"
 136 +       help
 137 +         This setting defines how IMQ behaves in respect to its
 138 +         hooking in PREROUTING and POSTROUTING.
 139 +
 140 +         Choosing this option will make IMQ hook like this:
 141 +
 142 +         PREROUTING:   Before NAT
 143 +         POSTROUTING:  Before NAT
 144 +
 145 +         More information can be found at: www.linuximq.net
 146 +
 147 +         If not sure leave the default settings alone.
 148 +
 149 +endchoice
 150 +
 151 +config IMQ_NUM_DEVS
 152 +       int "Number of IMQ devices"
 153 +       range 2 16
 154 +       depends on IMQ
 155 +       default "16"
 156 +       help
 157 +         This setting defines how many IMQ devices will be created.
 158 +
 159 +         The default value is 16.
 160 +
 161 +         More information can be found at: www.linuximq.net
 162 +
 163 +         If not sure leave the default settings alone.
 164 +
 165  config TUN
 166         tristate "Universal TUN/TAP device driver support"
 167         select CRC32
 168 diff --git a/drivers/net/Makefile b/drivers/net/Makefile
 169 index 3fef8a8..12dafc0 100644
 170 --- a/drivers/net/Makefile
 171 +++ b/drivers/net/Makefile
 172 @@ -9,6 +9,7 @@ obj-$(CONFIG_BONDING) += bonding/
 173  obj-$(CONFIG_DUMMY) += dummy.o
 174  obj-$(CONFIG_EQUALIZER) += eql.o
 175  obj-$(CONFIG_IFB) += ifb.o
 176 +obj-$(CONFIG_IMQ) += imq.o
 177  obj-$(CONFIG_MACVLAN) += macvlan.o
 178  obj-$(CONFIG_MACVTAP) += macvtap.o
 179  obj-$(CONFIG_MII) += mii.o
 180 diff --git a/drivers/net/imq.c b/drivers/net/imq.c
 181 new file mode 100644
 182 index 0000000..801bc8c
 183 --- /dev/null
 184 +++ b/drivers/net/imq.c
 185 @@ -0,0 +1,1012 @@
 186 +/*
 187 + *             Pseudo-driver for the intermediate queue device.
 188 + *
 189 + *             This program is free software; you can redistribute it and/or
 190 + *             modify it under the terms of the GNU General Public License
 191 + *             as published by the Free Software Foundation; either version
 192 + *             2 of the License, or (at your option) any later version.
 193 + *
 194 + * Authors:    Patrick McHardy, <kaber@trash.net>
 195 + *
 196 + *            The first version was written by Martin Devera, <devik@cdi.cz>
 197 + *
 198 + * Credits:    Jan Rafaj <imq2t@cedric.vabo.cz>
 199 + *              - Update patch to 2.4.21
 200 + *             Sebastian Strollo <sstrollo@nortelnetworks.com>
 201 + *              - Fix "Dead-loop on netdevice imq"-issue
 202 + *             Marcel Sebek <sebek64@post.cz>
 203 + *              - Update to 2.6.2-rc1
 204 + *
 205 + *            After some time of inactivity there is a group taking care
 206 + *            of IMQ again: http://www.linuximq.net
 207 + *
 208 + *
 209 + *            2004/06/30 - New version of IMQ patch to kernels <=2.6.7
 210 + *             including the following changes:
 211 + *
 212 + *            - Correction of ipv6 support "+"s issue (Hasso Tepper)
 213 + *            - Correction of imq_init_devs() issue that resulted in
 214 + *            kernel OOPS unloading IMQ as module (Norbert Buchmuller)
 215 + *            - Addition of functionality to choose number of IMQ devices
 216 + *            during kernel config (Andre Correa)
 217 + *            - Addition of functionality to choose how IMQ hooks on
 218 + *            PRE and POSTROUTING (after or before NAT) (Andre Correa)
 219 + *            - Cosmetic corrections (Norbert Buchmuller) (Andre Correa)
 220 + *
 221 + *
 222 + *             2005/12/16 - IMQ versions between 2.6.7 and 2.6.13 were
 223 + *             released with almost no problems. 2.6.14-x was released
 224 + *             with some important changes: nfcache was removed; After
 225 + *             some weeks of trouble we figured out that some IMQ fields
 226 + *             in skb were missing in skbuff.c - skb_clone and copy_skb_header.
 227 + *             These functions are correctly patched by this new patch version.
 228 + *
 229 + *             Thanks for all who helped to figure out all the problems with
 230 + *             2.6.14.x: Patrick McHardy, Rune Kock, VeNoMouS, Max CtRiX,
 231 + *             Kevin Shanahan, Richard Lucassen, Valery Dachev (hopefully
 232 + *             I didn't forget anybody). I apologize again for my lack of time.
 233 + *
 234 + *
 235 + *             2008/06/17 - 2.6.25 - Changed imq.c to use qdisc_run() instead
 236 + *             of qdisc_restart() and moved qdisc_run() to tasklet to avoid
 237 + *             recursive locking. New initialization routines to fix 'rmmod' not
 238 + *             working anymore. Used code from ifb.c. (Jussi Kivilinna)
 239 + *
 240 + *             2008/08/06 - 2.6.26 - (JK)
 241 + *              - Replaced tasklet with 'netif_schedule()'.
 242 + *              - Cleaned up and added comments for imq_nf_queue().
 243 + *
 244 + *             2009/04/12
 245 + *              - Add skb_save_cb/skb_restore_cb helper functions for backuping
 246 + *                control buffer. This is needed because qdisc-layer on kernels
 247 + *                2.6.27 and newer overwrite control buffer. (Jussi Kivilinna)
 248 + *              - Add better locking for IMQ device. Hopefully this will solve
 249 + *                SMP issues. (Jussi Kivilinna)
 250 + *              - Port to 2.6.27
 251 + *              - Port to 2.6.28
 252 + *              - Port to 2.6.29 + fix rmmod not working
 253 + *
 254 + *             2009/04/20 - (Jussi Kivilinna)
 255 + *              - Use netdevice feature flags to avoid extra packet handling
 256 + *                by core networking layer and possibly increase performance.
 257 + *
 258 + *             2009/09/26 - (Jussi Kivilinna)
 259 + *              - Add imq_nf_reinject_lockless to fix deadlock with
 260 + *                imq_nf_queue/imq_nf_reinject.
 261 + *
 262 + *             2009/12/08 - (Jussi Kivilinna)
 263 + *              - Port to 2.6.32
 264 + *              - Add check for skb->nf_queue_entry==NULL in imq_dev_xmit()
 265 + *              - Also add better error checking for skb->nf_queue_entry usage
 266 + *
 267 + *             2010/02/25 - (Jussi Kivilinna)
 268 + *              - Port to 2.6.33
 269 + *
 270 + *             2010/08/15 - (Jussi Kivilinna)
 271 + *              - Port to 2.6.35
 272 + *              - Simplify hook registration by using nf_register_hooks.
 273 + *              - nf_reinject doesn't need spinlock around it, therefore remove
 274 + *                imq_nf_reinject function. Other nf_reinject users protect
 275 + *                their own data with spinlock. With IMQ however all data is
 276 + *                needed is stored per skbuff, so no locking is needed.
 277 + *              - Changed IMQ to use 'separate' NF_IMQ_QUEUE instead of
 278 + *                NF_QUEUE, this allows working coexistance of IMQ and other
 279 + *                NF_QUEUE users.
 280 + *              - Make IMQ multi-queue. Number of IMQ device queues can be
 281 + *                increased with 'numqueues' module parameters. Default number
 282 + *                of queues is 1, in other words by default IMQ works as
 283 + *                single-queue device. Multi-queue selection is based on
 284 + *                IFB multi-queue patch by Changli Gao <xiaosuo@gmail.com>.
 285 + *
 286 + *             2011/03/18 - (Jussi Kivilinna)
 287 + *              - Port to 2.6.38
 288 + *
 289 + *             2011/07/12 - (syoder89@gmail.com)
 290 + *              - Crash fix that happens when the receiving interface has more
 291 + *                than one queue (add missing skb_set_queue_mapping in
 292 + *                imq_select_queue).
 293 + *
 294 + *             2011/07/26 - (Jussi Kivilinna)
 295 + *              - Add queue mapping checks for packets exiting IMQ.
 296 + *              - Port to 3.0
 297 + *
 298 + *             2011/08/16 - (Jussi Kivilinna)
 299 + *              - Clear IFF_TX_SKB_SHARING flag that was added for linux 3.0.2
 300 + *
 301 + *             2011/11/03 - Germano Michel <germanomichel@gmail.com>
 302 + *              - Fix IMQ for net namespaces
 303 + *
 304 + *             2011/11/04 - Jussi Kivilinna <jussi.kivilinna@mbnet.fi>
 305 + *              - Port to 3.1
 306 + *              - Clean-up, move 'get imq device pointer by imqX name' to
 307 + *                separate function from imq_nf_queue().
 308 + *
 309 + *             2012/01/05 - Jussi Kivilinna <jussi.kivilinna@mbnet.fi>
 310 + *              - Port to 3.2
 311 + *
 312 + *             2012/03/19 - Jussi Kivilinna <jussi.kivilinna@mbnet.fi>
 313 + *              - Port to 3.3
 314 + *
 315 + *             2012/12/12 - Jussi Kivilinna <jussi.kivilinna@mbnet.fi>
 316 + *              - Port to 3.7
 317 + *              - Fix checkpatch.pl warnings
 318 + *
 319 + *             2013/09/10 - Jussi Kivilinna <jussi.kivilinna@iki.fi>
 320 + *              - Fixed GSO handling for 3.10, see imq_nf_queue() for comments.
 321 + *              - Don't copy skb->cb_next when copying or cloning skbuffs.
 322 + *
 323 + *             2013/09/16 - Jussi Kivilinna <jussi.kivilinna@iki.fi>
 324 + *              - Port to 3.11
 325 + *
 326 + *             2013/11/12 - Jussi Kivilinna <jussi.kivilinna@iki.fi>
 327 + *              - Port to 3.12
 328 + *
 329 + *            Also, many thanks to pablo Sebastian Greco for making the initial
 330 + *            patch and to those who helped the testing.
 331 + *
 332 + *             More info at: http://www.linuximq.net/ (Andre Correa)
 333 + */
 334 +
 335 +#include <linux/module.h>
 336 +#include <linux/kernel.h>
 337 +#include <linux/moduleparam.h>
 338 +#include <linux/list.h>
 339 +#include <linux/skbuff.h>
 340 +#include <linux/netdevice.h>
 341 +#include <linux/etherdevice.h>
 342 +#include <linux/rtnetlink.h>
 343 +#include <linux/if_arp.h>
 344 +#include <linux/netfilter.h>
 345 +#include <linux/netfilter_ipv4.h>
 346 +#if defined(CONFIG_IPV6) || defined(CONFIG_IPV6_MODULE)
 347 +       #include <linux/netfilter_ipv6.h>
 348 +#endif
 349 +#include <linux/imq.h>
 350 +#include <net/pkt_sched.h>
 351 +#include <net/netfilter/nf_queue.h>
 352 +#include <net/sock.h>
 353 +#include <linux/ip.h>
 354 +#include <linux/ipv6.h>
 355 +#include <linux/if_vlan.h>
 356 +#include <linux/if_pppox.h>
 357 +#include <net/ip.h>
 358 +#include <net/ipv6.h>
 359 +
 360 +static int imq_nf_queue(struct nf_queue_entry *entry, unsigned queue_num);
 361 +
 362 +static unsigned int imq_nf_hook(const struct nf_hook_ops *ops,
 363 +                               struct sk_buff *pskb,
 364 +                               const struct net_device *indev,
 365 +                               const struct net_device *outdev,
 366 +                               int (*okfn)(struct sk_buff *));
 367 +
 368 +static struct nf_hook_ops imq_ops[] = {
 369 +       {
 370 +       /* imq_ingress_ipv4 */
 371 +               .hook           = imq_nf_hook,
 372 +               .owner          = THIS_MODULE,
 373 +               .pf             = PF_INET,
 374 +               .hooknum        = NF_INET_PRE_ROUTING,
 375 +#if defined(CONFIG_IMQ_BEHAVIOR_BA) || defined(CONFIG_IMQ_BEHAVIOR_BB)
 376 +               .priority       = NF_IP_PRI_MANGLE + 1,
 377 +#else
 378 +               .priority       = NF_IP_PRI_NAT_DST + 1,
 379 +#endif
 380 +       },
 381 +       {
 382 +       /* imq_egress_ipv4 */
 383 +               .hook           = imq_nf_hook,
 384 +               .owner          = THIS_MODULE,
 385 +               .pf             = PF_INET,
 386 +               .hooknum        = NF_INET_POST_ROUTING,
 387 +#if defined(CONFIG_IMQ_BEHAVIOR_AA) || defined(CONFIG_IMQ_BEHAVIOR_BA)
 388 +               .priority       = NF_IP_PRI_LAST,
 389 +#else
 390 +               .priority       = NF_IP_PRI_NAT_SRC - 1,
 391 +#endif
 392 +       },
 393 +#if defined(CONFIG_IPV6) || defined(CONFIG_IPV6_MODULE)
 394 +       {
 395 +       /* imq_ingress_ipv6 */
 396 +               .hook           = imq_nf_hook,
 397 +               .owner          = THIS_MODULE,
 398 +               .pf             = PF_INET6,
 399 +               .hooknum        = NF_INET_PRE_ROUTING,
 400 +#if defined(CONFIG_IMQ_BEHAVIOR_BA) || defined(CONFIG_IMQ_BEHAVIOR_BB)
 401 +               .priority       = NF_IP6_PRI_MANGLE + 1,
 402 +#else
 403 +               .priority       = NF_IP6_PRI_NAT_DST + 1,
 404 +#endif
 405 +       },
 406 +       {
 407 +       /* imq_egress_ipv6 */
 408 +               .hook           = imq_nf_hook,
 409 +               .owner          = THIS_MODULE,
 410 +               .pf             = PF_INET6,
 411 +               .hooknum        = NF_INET_POST_ROUTING,
 412 +#if defined(CONFIG_IMQ_BEHAVIOR_AA) || defined(CONFIG_IMQ_BEHAVIOR_BA)
 413 +               .priority       = NF_IP6_PRI_LAST,
 414 +#else
 415 +               .priority       = NF_IP6_PRI_NAT_SRC - 1,
 416 +#endif
 417 +       },
 418 +#endif
 419 +};
 420 +
 421 +#if defined(CONFIG_IMQ_NUM_DEVS)
 422 +static int numdevs = CONFIG_IMQ_NUM_DEVS;
 423 +#else
 424 +static int numdevs = IMQ_MAX_DEVS;
 425 +#endif
 426 +
 427 +static struct net_device *imq_devs_cache[IMQ_MAX_DEVS];
 428 +
 429 +#define IMQ_MAX_QUEUES 32
 430 +static int numqueues = 1;
 431 +static u32 imq_hashrnd;
 432 +
 433 +static inline __be16 pppoe_proto(const struct sk_buff *skb)
 434 +{
 435 +       return *((__be16 *)(skb_mac_header(skb) + ETH_HLEN +
 436 +                       sizeof(struct pppoe_hdr)));
 437 +}
 438 +
 439 +static u16 imq_hash(struct net_device *dev, struct sk_buff *skb)
 440 +{
 441 +       unsigned int pull_len;
 442 +       u16 protocol = skb->protocol;
 443 +       u32 addr1, addr2;
 444 +       u32 hash, ihl = 0;
 445 +       union {
 446 +               u16 in16[2];
 447 +               u32 in32;
 448 +       } ports;
 449 +       u8 ip_proto;
 450 +
 451 +       pull_len = 0;
 452 +
 453 +recheck:
 454 +       switch (protocol) {
 455 +       case htons(ETH_P_8021Q): {
 456 +               if (unlikely(skb_pull(skb, VLAN_HLEN) == NULL))
 457 +                       goto other;
 458 +
 459 +               pull_len += VLAN_HLEN;
 460 +               skb->network_header += VLAN_HLEN;
 461 +
 462 +               protocol = vlan_eth_hdr(skb)->h_vlan_encapsulated_proto;
 463 +               goto recheck;
 464 +       }
 465 +
 466 +       case htons(ETH_P_PPP_SES): {
 467 +               if (unlikely(skb_pull(skb, PPPOE_SES_HLEN) == NULL))
 468 +                       goto other;
 469 +
 470 +               pull_len += PPPOE_SES_HLEN;
 471 +               skb->network_header += PPPOE_SES_HLEN;
 472 +
 473 +               protocol = pppoe_proto(skb);
 474 +               goto recheck;
 475 +       }
 476 +
 477 +       case htons(ETH_P_IP): {
 478 +               const struct iphdr *iph = ip_hdr(skb);
 479 +
 480 +               if (unlikely(!pskb_may_pull(skb, sizeof(struct iphdr))))
 481 +                       goto other;
 482 +
 483 +               addr1 = iph->daddr;
 484 +               addr2 = iph->saddr;
 485 +
 486 +               ip_proto = !(ip_hdr(skb)->frag_off & htons(IP_MF | IP_OFFSET)) ?
 487 +                                iph->protocol : 0;
 488 +               ihl = ip_hdrlen(skb);
 489 +
 490 +               break;
 491 +       }
 492 +#if defined(CONFIG_IPV6) || defined(CONFIG_IPV6_MODULE)
 493 +       case htons(ETH_P_IPV6): {
 494 +               const struct ipv6hdr *iph = ipv6_hdr(skb);
 495 +               __be16 fo = 0;
 496 +
 497 +               if (unlikely(!pskb_may_pull(skb, sizeof(struct ipv6hdr))))
 498 +                       goto other;
 499 +
 500 +               addr1 = iph->daddr.s6_addr32[3];
 501 +               addr2 = iph->saddr.s6_addr32[3];
 502 +               ihl = ipv6_skip_exthdr(skb, sizeof(struct ipv6hdr), &ip_proto,
 503 +                                      &fo);
 504 +               if (unlikely(ihl < 0))
 505 +                       goto other;
 506 +
 507 +               break;
 508 +       }
 509 +#endif
 510 +       default:
 511 +other:
 512 +               if (pull_len != 0) {
 513 +                       skb_push(skb, pull_len);
 514 +                       skb->network_header -= pull_len;
 515 +               }
 516 +
 517 +               return (u16)(ntohs(protocol) % dev->real_num_tx_queues);
 518 +       }
 519 +
 520 +       if (addr1 > addr2)
 521 +               swap(addr1, addr2);
 522 +
 523 +       switch (ip_proto) {
 524 +       case IPPROTO_TCP:
 525 +       case IPPROTO_UDP:
 526 +       case IPPROTO_DCCP:
 527 +       case IPPROTO_ESP:
 528 +       case IPPROTO_AH:
 529 +       case IPPROTO_SCTP:
 530 +       case IPPROTO_UDPLITE: {
 531 +               if (likely(skb_copy_bits(skb, ihl, &ports.in32, 4) >= 0)) {
 532 +                       if (ports.in16[0] > ports.in16[1])
 533 +                               swap(ports.in16[0], ports.in16[1]);
 534 +                       break;
 535 +               }
 536 +               /* fall-through */
 537 +       }
 538 +       default:
 539 +               ports.in32 = 0;
 540 +               break;
 541 +       }
 542 +
 543 +       if (pull_len != 0) {
 544 +               skb_push(skb, pull_len);
 545 +               skb->network_header -= pull_len;
 546 +       }
 547 +
 548 +       hash = jhash_3words(addr1, addr2, ports.in32, imq_hashrnd ^ ip_proto);
 549 +
 550 +       return (u16)(((u64)hash * dev->real_num_tx_queues) >> 32);
 551 +}
 552 +
 553 +static inline bool sk_tx_queue_recorded(struct sock *sk)
 554 +{
 555 +       return (sk_tx_queue_get(sk) >= 0);
 556 +}
 557 +
 558 +static struct netdev_queue *imq_select_queue(struct net_device *dev,
 559 +                                               struct sk_buff *skb)
 560 +{
 561 +       u16 queue_index = 0;
 562 +       u32 hash;
 563 +
 564 +       if (likely(dev->real_num_tx_queues == 1))
 565 +               goto out;
 566 +
 567 +       /* IMQ can be receiving ingress or engress packets. */
 568 +
 569 +       /* Check first for if rx_queue is set */
 570 +       if (skb_rx_queue_recorded(skb)) {
 571 +               queue_index = skb_get_rx_queue(skb);
 572 +               goto out;
 573 +       }
 574 +
 575 +       /* Check if socket has tx_queue set */
 576 +       if (sk_tx_queue_recorded(skb->sk)) {
 577 +               queue_index = sk_tx_queue_get(skb->sk);
 578 +               goto out;
 579 +       }
 580 +
 581 +       /* Try use socket hash */
 582 +       if (skb->sk && skb->sk->sk_hash) {
 583 +               hash = skb->sk->sk_hash;
 584 +               queue_index =
 585 +                       (u16)(((u64)hash * dev->real_num_tx_queues) >> 32);
 586 +               goto out;
 587 +       }
 588 +
 589 +       /* Generate hash from packet data */
 590 +       queue_index = imq_hash(dev, skb);
 591 +
 592 +out:
 593 +       if (unlikely(queue_index >= dev->real_num_tx_queues))
 594 +               queue_index = (u16)((u32)queue_index % dev->real_num_tx_queues);
 595 +
 596 +       skb_set_queue_mapping(skb, queue_index);
 597 +       return netdev_get_tx_queue(dev, queue_index);
 598 +}
 599 +
 600 +static struct net_device_stats *imq_get_stats(struct net_device *dev)
 601 +{
 602 +       return &dev->stats;
 603 +}
 604 +
 605 +/* called for packets kfree'd in qdiscs at places other than enqueue */
 606 +static void imq_skb_destructor(struct sk_buff *skb)
 607 +{
 608 +       struct nf_queue_entry *entry = skb->nf_queue_entry;
 609 +
 610 +       skb->nf_queue_entry = NULL;
 611 +
 612 +       if (entry) {
 613 +               nf_queue_entry_release_refs(entry);
 614 +               kfree(entry);
 615 +       }
 616 +
 617 +       skb_restore_cb(skb); /* kfree backup */
 618 +}
 619 +
 620 +static void imq_done_check_queue_mapping(struct sk_buff *skb,
 621 +                                        struct net_device *dev)
 622 +{
 623 +       unsigned int queue_index;
 624 +
 625 +       /* Don't let queue_mapping be left too large after exiting IMQ */
 626 +       if (likely(skb->dev != dev && skb->dev != NULL)) {
 627 +               queue_index = skb_get_queue_mapping(skb);
 628 +               if (unlikely(queue_index >= skb->dev->real_num_tx_queues)) {
 629 +                       queue_index = (u16)((u32)queue_index %
 630 +                                               skb->dev->real_num_tx_queues);
 631 +                       skb_set_queue_mapping(skb, queue_index);
 632 +               }
 633 +       } else {
 634 +               /* skb->dev was IMQ device itself or NULL, be on safe side and
 635 +                * just clear queue mapping.
 636 +                */
 637 +               skb_set_queue_mapping(skb, 0);
 638 +       }
 639 +}
 640 +
 641 +static netdev_tx_t imq_dev_xmit(struct sk_buff *skb, struct net_device *dev)
 642 +{
 643 +       struct nf_queue_entry *entry = skb->nf_queue_entry;
 644 +
 645 +       skb->nf_queue_entry = NULL;
 646 +       dev->trans_start = jiffies;
 647 +
 648 +       dev->stats.tx_bytes += skb->len;
 649 +       dev->stats.tx_packets++;
 650 +
 651 +       if (unlikely(entry == NULL)) {
 652 +               /* We don't know what is going on here.. packet is queued for
 653 +                * imq device, but (probably) not by us.
 654 +                *
 655 +                * If this packet was not send here by imq_nf_queue(), then
 656 +                * skb_save_cb() was not used and skb_free() should not show:
 657 +                *   WARNING: IMQ: kfree_skb: skb->cb_next:..
 658 +                * and/or
 659 +                *   WARNING: IMQ: kfree_skb: skb->nf_queue_entry...
 660 +                *
 661 +                * However if this message is shown, then IMQ is somehow broken
 662 +                * and you should report this to linuximq.net.
 663 +                */
 664 +
 665 +               /* imq_dev_xmit is black hole that eats all packets, report that
 666 +                * we eat this packet happily and increase dropped counters.
 667 +                */
 668 +
 669 +               dev->stats.tx_dropped++;
 670 +               dev_kfree_skb(skb);
 671 +
 672 +               return NETDEV_TX_OK;
 673 +       }
 674 +
 675 +       skb_restore_cb(skb); /* restore skb->cb */
 676 +
 677 +       skb->imq_flags = 0;
 678 +       skb->destructor = NULL;
 679 +
 680 +       imq_done_check_queue_mapping(skb, dev);
 681 +
 682 +       nf_reinject(entry, NF_ACCEPT);
 683 +
 684 +       return NETDEV_TX_OK;
 685 +}
 686 +
 687 +static struct net_device *get_imq_device_by_index(int index)
 688 +{
 689 +       struct net_device *dev = NULL;
 690 +       struct net *net;
 691 +       char buf[8];
 692 +
 693 +       /* get device by name and cache result */
 694 +       snprintf(buf, sizeof(buf), "imq%d", index);
 695 +
 696 +       /* Search device from all namespaces. */
 697 +       for_each_net(net) {
 698 +               dev = dev_get_by_name(net, buf);
 699 +               if (dev)
 700 +                       break;
 701 +       }
 702 +
 703 +       if (WARN_ON_ONCE(dev == NULL)) {
 704 +               /* IMQ device not found. Exotic config? */
 705 +               return ERR_PTR(-ENODEV);
 706 +       }
 707 +
 708 +       imq_devs_cache[index] = dev;
 709 +       dev_put(dev);
 710 +
 711 +       return dev;
 712 +}
 713 +
 714 +static struct nf_queue_entry *nf_queue_entry_dup(struct nf_queue_entry *e)
 715 +{
 716 +       struct nf_queue_entry *entry = kmemdup(e, e->size, GFP_ATOMIC);
 717 +       if (entry) {
 718 +               if (nf_queue_entry_get_refs(entry))
 719 +                       return entry;
 720 +               kfree(entry);
 721 +       }
 722 +       return NULL;
 723 +}
 724 +
 725 +#ifdef CONFIG_BRIDGE_NETFILTER
 726 +/* When called from bridge netfilter, skb->data must point to MAC header
 727 + * before calling skb_gso_segment(). Else, original MAC header is lost
 728 + * and segmented skbs will be sent to wrong destination.
 729 + */
 730 +static void nf_bridge_adjust_skb_data(struct sk_buff *skb)
 731 +{
 732 +       if (skb->nf_bridge)
 733 +               __skb_push(skb, skb->network_header - skb->mac_header);
 734 +}
 735 +
 736 +static void nf_bridge_adjust_segmented_data(struct sk_buff *skb)
 737 +{
 738 +       if (skb->nf_bridge)
 739 +               __skb_pull(skb, skb->network_header - skb->mac_header);
 740 +}
 741 +#else
 742 +#define nf_bridge_adjust_skb_data(s) do {} while (0)
 743 +#define nf_bridge_adjust_segmented_data(s) do {} while (0)
 744 +#endif
 745 +
 746 +static void free_entry(struct nf_queue_entry *entry)
 747 +{
 748 +       nf_queue_entry_release_refs(entry);
 749 +       kfree(entry);
 750 +}
 751 +
 752 +static int __imq_nf_queue(struct nf_queue_entry *entry, struct net_device *dev);
 753 +
 754 +static int __imq_nf_queue_gso(struct nf_queue_entry *entry,
 755 +                             struct net_device *dev, struct sk_buff *skb)
 756 +{
 757 +       int ret = -ENOMEM;
 758 +       struct nf_queue_entry *entry_seg;
 759 +
 760 +       nf_bridge_adjust_segmented_data(skb);
 761 +
 762 +       if (skb->next == NULL) { /* last packet, no need to copy entry */
 763 +               struct sk_buff *gso_skb = entry->skb;
 764 +               entry->skb = skb;
 765 +               ret = __imq_nf_queue(entry, dev);
 766 +               if (ret)
 767 +                       entry->skb = gso_skb;
 768 +               return ret;
 769 +       }
 770 +
 771 +       skb->next = NULL;
 772 +
 773 +       entry_seg = nf_queue_entry_dup(entry);
 774 +       if (entry_seg) {
 775 +               entry_seg->skb = skb;
 776 +               ret = __imq_nf_queue(entry_seg, dev);
 777 +               if (ret)
 778 +                       free_entry(entry_seg);
 779 +       }
 780 +       return ret;
 781 +}
 782 +
 783 +static int imq_nf_queue(struct nf_queue_entry *entry, unsigned queue_num)
 784 +{
 785 +       struct sk_buff *skb, *segs;
 786 +       struct net_device *dev;
 787 +       unsigned int queued;
 788 +       int index, retval, err;
 789 +
 790 +       index = entry->skb->imq_flags & IMQ_F_IFMASK;
 791 +       if (unlikely(index > numdevs - 1)) {
 792 +               if (net_ratelimit())
 793 +                       pr_warn("IMQ: invalid device specified, highest is %u\n",
 794 +                               numdevs - 1);
 795 +               retval = -EINVAL;
 796 +               goto out_no_dev;
 797 +       }
 798 +
 799 +       /* check for imq device by index from cache */
 800 +       dev = imq_devs_cache[index];
 801 +       if (unlikely(!dev)) {
 802 +               dev = get_imq_device_by_index(index);
 803 +               if (IS_ERR(dev)) {
 804 +                       retval = PTR_ERR(dev);
 805 +                       goto out_no_dev;
 806 +               }
 807 +       }
 808 +
 809 +       if (unlikely(!(dev->flags & IFF_UP))) {
 810 +               entry->skb->imq_flags = 0;
 811 +               retval = -ECANCELED;
 812 +               goto out_no_dev;
 813 +       }
 814 +
 815 +       if (!skb_is_gso(entry->skb))
 816 +               return __imq_nf_queue(entry, dev);
 817 +
 818 +       /* Since 3.10.x, GSO handling moved here as result of upstream commit
 819 +        * a5fedd43d5f6c94c71053a66e4c3d2e35f1731a2 (netfilter: move
 820 +        * skb_gso_segment into nfnetlink_queue module).
 821 +        *
 822 +        * Following code replicates the gso handling from
 823 +        * 'net/netfilter/nfnetlink_queue_core.c':nfqnl_enqueue_packet().
 824 +        */
 825 +
 826 +       skb = entry->skb;
 827 +
 828 +       switch (entry->pf) {
 829 +       case NFPROTO_IPV4:
 830 +               skb->protocol = htons(ETH_P_IP);
 831 +               break;
 832 +       case NFPROTO_IPV6:
 833 +               skb->protocol = htons(ETH_P_IPV6);
 834 +               break;
 835 +       }
 836 +
 837 +       nf_bridge_adjust_skb_data(skb);
 838 +       segs = skb_gso_segment(skb, 0);
 839 +       /* Does not use PTR_ERR to limit the number of error codes that can be
 840 +        * returned by nf_queue.  For instance, callers rely on -ECANCELED to
 841 +        * mean 'ignore this hook'.
 842 +        */
 843 +       err = -ENOBUFS;
 844 +       if (IS_ERR(segs))
 845 +               goto out_err;
 846 +       queued = 0;
 847 +       err = 0;
 848 +       do {
 849 +               struct sk_buff *nskb = segs->next;
 850 +               if (nskb && nskb->next)
 851 +                       nskb->cb_next = NULL;
 852 +               if (err == 0)
 853 +                       err = __imq_nf_queue_gso(entry, dev, segs);
 854 +               if (err == 0)
 855 +                       queued++;
 856 +               else
 857 +                       kfree_skb(segs);
 858 +               segs = nskb;
 859 +       } while (segs);
 860 +
 861 +       if (queued) {
 862 +               if (err) /* some segments are already queued */
 863 +                       free_entry(entry);
 864 +               kfree_skb(skb);
 865 +               return 0;
 866 +       }
 867 +
 868 +out_err:
 869 +       nf_bridge_adjust_segmented_data(skb);
 870 +       retval = err;
 871 +out_no_dev:
 872 +       return retval;
 873 +}
 874 +
 875 +static int __imq_nf_queue(struct nf_queue_entry *entry, struct net_device *dev)
 876 +{
 877 +       struct sk_buff *skb_orig, *skb, *skb_shared;
 878 +       struct Qdisc *q;
 879 +       struct netdev_queue *txq;
 880 +       spinlock_t *root_lock;
 881 +       int users;
 882 +       int retval = -EINVAL;
 883 +       unsigned int orig_queue_index;
 884 +
 885 +       dev->last_rx = jiffies;
 886 +
 887 +       skb = entry->skb;
 888 +       skb_orig = NULL;
 889 +
 890 +       /* skb has owner? => make clone */
 891 +       if (unlikely(skb->destructor)) {
 892 +               skb_orig = skb;
 893 +               skb = skb_clone(skb, GFP_ATOMIC);
 894 +               if (unlikely(!skb)) {
 895 +                       retval = -ENOMEM;
 896 +                       goto out;
 897 +               }
 898 +               skb->cb_next = NULL;
 899 +               entry->skb = skb;
 900 +       }
 901 +
 902 +       skb->nf_queue_entry = entry;
 903 +
 904 +       dev->stats.rx_bytes += skb->len;
 905 +       dev->stats.rx_packets++;
 906 +
 907 +       if (!skb->dev) {
 908 +               /* skb->dev == NULL causes problems, try the find cause. */
 909 +               if (net_ratelimit()) {
 910 +                       dev_warn(&dev->dev,
 911 +                                "received packet with skb->dev == NULL\n");
 912 +                       dump_stack();
 913 +               }
 914 +
 915 +               skb->dev = dev;
 916 +       }
 917 +
 918 +       /* Disables softirqs for lock below */
 919 +       rcu_read_lock_bh();
 920 +
 921 +       /* Multi-queue selection */
 922 +       orig_queue_index = skb_get_queue_mapping(skb);
 923 +       txq = imq_select_queue(dev, skb);
 924 +
 925 +       q = rcu_dereference(txq->qdisc);
 926 +       if (unlikely(!q->enqueue))
 927 +               goto packet_not_eaten_by_imq_dev;
 928 +
 929 +       root_lock = qdisc_lock(q);
 930 +       spin_lock(root_lock);
 931 +
 932 +       users = atomic_read(&skb->users);
 933 +
 934 +       skb_shared = skb_get(skb); /* increase reference count by one */
 935 +
 936 +       /* backup skb->cb, as qdisc layer will overwrite it */
 937 +       skb_save_cb(skb_shared);
 938 +       qdisc_enqueue_root(skb_shared, q); /* might kfree_skb */
 939 +
 940 +       if (likely(atomic_read(&skb_shared->users) == users + 1)) {
 941 +               kfree_skb(skb_shared); /* decrease reference count by one */
 942 +
 943 +               skb->destructor = &imq_skb_destructor;
 944 +
 945 +               /* cloned? */
 946 +               if (unlikely(skb_orig))
 947 +                       kfree_skb(skb_orig); /* free original */
 948 +
 949 +               spin_unlock(root_lock);
 950 +               rcu_read_unlock_bh();
 951 +
 952 +               /* schedule qdisc dequeue */
 953 +               __netif_schedule(q);
 954 +
 955 +               retval = 0;
 956 +               goto out;
 957 +       } else {
 958 +               skb_restore_cb(skb_shared); /* restore skb->cb */
 959 +               skb->nf_queue_entry = NULL;
 960 +               /*
 961 +                * qdisc dropped packet and decreased skb reference count of
 962 +                * skb, so we don't really want to and try refree as that would
 963 +                * actually destroy the skb.
 964 +                */
 965 +               spin_unlock(root_lock);
 966 +               goto packet_not_eaten_by_imq_dev;
 967 +       }
 968 +
 969 +packet_not_eaten_by_imq_dev:
 970 +       skb_set_queue_mapping(skb, orig_queue_index);
 971 +       rcu_read_unlock_bh();
 972 +
 973 +       /* cloned? restore original */
 974 +       if (unlikely(skb_orig)) {
 975 +               kfree_skb(skb);
 976 +               entry->skb = skb_orig;
 977 +       }
 978 +       retval = -1;
 979 +out:
 980 +       return retval;
 981 +}
 982 +
 983 +static unsigned int imq_nf_hook(const struct nf_hook_ops *ops,
 984 +                               struct sk_buff *pskb,
 985 +                               const struct net_device *indev,
 986 +                               const struct net_device *outdev,
 987 +                               int (*okfn)(struct sk_buff *))
 988 +{
 989 +       return (pskb->imq_flags & IMQ_F_ENQUEUE) ? NF_IMQ_QUEUE : NF_ACCEPT;
 990 +}
 991 +
 992 +static int imq_close(struct net_device *dev)
 993 +{
 994 +       netif_stop_queue(dev);
 995 +       return 0;
 996 +}
 997 +
 998 +static int imq_open(struct net_device *dev)
 999 +{
1000 +       netif_start_queue(dev);
1001 +       return 0;
1002 +}
1003 +
1004 +static const struct net_device_ops imq_netdev_ops = {
1005 +       .ndo_open               = imq_open,
1006 +       .ndo_stop               = imq_close,
1007 +       .ndo_start_xmit         = imq_dev_xmit,
1008 +       .ndo_get_stats          = imq_get_stats,
1009 +};
1010 +
1011 +static void imq_setup(struct net_device *dev)
1012 +{
1013 +       dev->netdev_ops         = &imq_netdev_ops;
1014 +       dev->type               = ARPHRD_VOID;
1015 +       dev->mtu                = 16000; /* too small? */
1016 +       dev->tx_queue_len       = 11000; /* too big? */
1017 +       dev->flags              = IFF_NOARP;
1018 +       dev->features           = NETIF_F_SG | NETIF_F_FRAGLIST |
1019 +                                 NETIF_F_GSO | NETIF_F_HW_CSUM |
1020 +                                 NETIF_F_HIGHDMA;
1021 +       dev->priv_flags         &= ~(IFF_XMIT_DST_RELEASE |
1022 +                                    IFF_TX_SKB_SHARING);
1023 +}
1024 +
1025 +static int imq_validate(struct nlattr *tb[], struct nlattr *data[])
1026 +{
1027 +       int ret = 0;
1028 +
1029 +       if (tb[IFLA_ADDRESS]) {
1030 +               if (nla_len(tb[IFLA_ADDRESS]) != ETH_ALEN) {
1031 +                       ret = -EINVAL;
1032 +                       goto end;
1033 +               }
1034 +               if (!is_valid_ether_addr(nla_data(tb[IFLA_ADDRESS]))) {
1035 +                       ret = -EADDRNOTAVAIL;
1036 +                       goto end;
1037 +               }
1038 +       }
1039 +       return 0;
1040 +end:
1041 +       pr_warn("IMQ: imq_validate failed (%d)\n", ret);
1042 +       return ret;
1043 +}
1044 +
1045 +static struct rtnl_link_ops imq_link_ops __read_mostly = {
1046 +       .kind           = "imq",
1047 +       .priv_size      = 0,
1048 +       .setup          = imq_setup,
1049 +       .validate       = imq_validate,
1050 +};
1051 +
1052 +static const struct nf_queue_handler imq_nfqh = {
1053 +       .outfn = imq_nf_queue,
1054 +};
1055 +
1056 +static int __init imq_init_hooks(void)
1057 +{
1058 +       int ret;
1059 +
1060 +       nf_register_queue_imq_handler(&imq_nfqh);
1061 +
1062 +       ret = nf_register_hooks(imq_ops, ARRAY_SIZE(imq_ops));
1063 +       if (ret < 0)
1064 +               nf_unregister_queue_imq_handler();
1065 +
1066 +       return ret;
1067 +}
1068 +
1069 +static int __init imq_init_one(int index)
1070 +{
1071 +       struct net_device *dev;
1072 +       int ret;
1073 +
1074 +       dev = alloc_netdev_mq(0, "imq%d", imq_setup, numqueues);
1075 +       if (!dev)
1076 +               return -ENOMEM;
1077 +
1078 +       ret = dev_alloc_name(dev, dev->name);
1079 +       if (ret < 0)
1080 +               goto fail;
1081 +
1082 +       dev->rtnl_link_ops = &imq_link_ops;
1083 +       ret = register_netdevice(dev);
1084 +       if (ret < 0)
1085 +               goto fail;
1086 +
1087 +       return 0;
1088 +fail:
1089 +       free_netdev(dev);
1090 +       return ret;
1091 +}
1092 +
1093 +static int __init imq_init_devs(void)
1094 +{
1095 +       int err, i;
1096 +
1097 +       if (numdevs < 1 || numdevs > IMQ_MAX_DEVS) {
1098 +               pr_err("IMQ: numdevs has to be betweed 1 and %u\n",
1099 +                      IMQ_MAX_DEVS);
1100 +               return -EINVAL;
1101 +       }
1102 +
1103 +       if (numqueues < 1 || numqueues > IMQ_MAX_QUEUES) {
1104 +               pr_err("IMQ: numqueues has to be betweed 1 and %u\n",
1105 +                      IMQ_MAX_QUEUES);
1106 +               return -EINVAL;
1107 +       }
1108 +
1109 +       get_random_bytes(&imq_hashrnd, sizeof(imq_hashrnd));
1110 +
1111 +       rtnl_lock();
1112 +       err = __rtnl_link_register(&imq_link_ops);
1113 +
1114 +       for (i = 0; i < numdevs && !err; i++)
1115 +               err = imq_init_one(i);
1116 +
1117 +       if (err) {
1118 +               __rtnl_link_unregister(&imq_link_ops);
1119 +               memset(imq_devs_cache, 0, sizeof(imq_devs_cache));
1120 +       }
1121 +       rtnl_unlock();
1122 +
1123 +       return err;
1124 +}
1125 +
1126 +static int __init imq_init_module(void)
1127 +{
1128 +       int err;
1129 +
1130 +#if defined(CONFIG_IMQ_NUM_DEVS)
1131 +       BUILD_BUG_ON(CONFIG_IMQ_NUM_DEVS > 16);
1132 +       BUILD_BUG_ON(CONFIG_IMQ_NUM_DEVS < 2);
1133 +       BUILD_BUG_ON(CONFIG_IMQ_NUM_DEVS - 1 > IMQ_F_IFMASK);
1134 +#endif
1135 +
1136 +       err = imq_init_devs();
1137 +       if (err) {
1138 +               pr_err("IMQ: Error trying imq_init_devs(net)\n");
1139 +               return err;
1140 +       }
1141 +
1142 +       err = imq_init_hooks();
1143 +       if (err) {
1144 +               pr_err(KERN_ERR "IMQ: Error trying imq_init_hooks()\n");
1145 +               rtnl_link_unregister(&imq_link_ops);
1146 +               memset(imq_devs_cache, 0, sizeof(imq_devs_cache));
1147 +               return err;
1148 +       }
1149 +
1150 +       pr_info("IMQ driver loaded successfully. (numdevs = %d, numqueues = %d)\n",
1151 +               numdevs, numqueues);
1152 +
1153 +#if defined(CONFIG_IMQ_BEHAVIOR_BA) || defined(CONFIG_IMQ_BEHAVIOR_BB)
1154 +       pr_info("\tHooking IMQ before NAT on PREROUTING.\n");
1155 +#else
1156 +       pr_info("\tHooking IMQ after NAT on PREROUTING.\n");
1157 +#endif
1158 +#if defined(CONFIG_IMQ_BEHAVIOR_AB) || defined(CONFIG_IMQ_BEHAVIOR_BB)
1159 +       pr_info("\tHooking IMQ before NAT on POSTROUTING.\n");
1160 +#else
1161 +       pr_info("\tHooking IMQ after NAT on POSTROUTING.\n");
1162 +#endif
1163 +
1164 +       return 0;
1165 +}
1166 +
1167 +static void __exit imq_unhook(void)
1168 +{
1169 +       nf_unregister_hooks(imq_ops, ARRAY_SIZE(imq_ops));
1170 +       nf_unregister_queue_imq_handler();
1171 +}
1172 +
1173 +static void __exit imq_cleanup_devs(void)
1174 +{
1175 +       rtnl_link_unregister(&imq_link_ops);
1176 +       memset(imq_devs_cache, 0, sizeof(imq_devs_cache));
1177 +}
1178 +
1179 +static void __exit imq_exit_module(void)
1180 +{
1181 +       imq_unhook();
1182 +       imq_cleanup_devs();
1183 +       pr_info("IMQ driver unloaded successfully.\n");
1184 +}
1185 +
1186 +module_init(imq_init_module);
1187 +module_exit(imq_exit_module);
1188 +
1189 +module_param(numdevs, int, 0);
1190 +module_param(numqueues, int, 0);
1191 +MODULE_PARM_DESC(numdevs, "number of IMQ devices (how many imq* devices will be created)");
1192 +MODULE_PARM_DESC(numqueues, "number of queues per IMQ device");
1193 +MODULE_AUTHOR("http://www.linuximq.net");
1194 +MODULE_DESCRIPTION("Pseudo-driver for the intermediate queue device. See http://www.linuximq.net/ for more information.");
1195 +MODULE_LICENSE("GPL");
1196 +MODULE_ALIAS_RTNL_LINK("imq");
1197 +
1198 diff --git a/include/linux/imq.h b/include/linux/imq.h
1199 new file mode 100644
1200 index 0000000..1babb09
1201 --- /dev/null
1202 +++ b/include/linux/imq.h
1203 @@ -0,0 +1,13 @@
1204 +#ifndef _IMQ_H
1205 +#define _IMQ_H
1206 +
1207 +/* IFMASK (16 device indexes, 0 to 15) and flag(s) fit in 5 bits */
1208 +#define IMQ_F_BITS     5
1209 +
1210 +#define IMQ_F_IFMASK   0x0f
1211 +#define IMQ_F_ENQUEUE  0x10
1212 +
1213 +#define IMQ_MAX_DEVS   (IMQ_F_IFMASK + 1)
1214 +
1215 +#endif /* _IMQ_H */
1216 +
1217 diff --git a/include/linux/netfilter/xt_IMQ.h b/include/linux/netfilter/xt_IMQ.h
1218 new file mode 100644
1219 index 0000000..9b07230
1220 --- /dev/null
1221 +++ b/include/linux/netfilter/xt_IMQ.h
1222 @@ -0,0 +1,9 @@
1223 +#ifndef _XT_IMQ_H
1224 +#define _XT_IMQ_H
1225 +
1226 +struct xt_imq_info {
1227 +       unsigned int todev;     /* target imq device */
1228 +};
1229 +
1230 +#endif /* _XT_IMQ_H */
1231 +
1232 diff --git a/include/linux/netfilter_ipv4/ipt_IMQ.h b/include/linux/netfilter_ipv4/ipt_IMQ.h
1233 new file mode 100644
1234 index 0000000..7af320f
1235 --- /dev/null
1236 +++ b/include/linux/netfilter_ipv4/ipt_IMQ.h
1237 @@ -0,0 +1,10 @@
1238 +#ifndef _IPT_IMQ_H
1239 +#define _IPT_IMQ_H
1240 +
1241 +/* Backwards compatibility for old userspace */
1242 +#include <linux/netfilter/xt_IMQ.h>
1243 +
1244 +#define ipt_imq_info xt_imq_info
1245 +
1246 +#endif /* _IPT_IMQ_H */
1247 +
1248 diff --git a/include/linux/netfilter_ipv6/ip6t_IMQ.h b/include/linux/netfilter_ipv6/ip6t_IMQ.h
1249 new file mode 100644
1250 index 0000000..198ac01
1251 --- /dev/null
1252 +++ b/include/linux/netfilter_ipv6/ip6t_IMQ.h
1253 @@ -0,0 +1,10 @@
1254 +#ifndef _IP6T_IMQ_H
1255 +#define _IP6T_IMQ_H
1256 +
1257 +/* Backwards compatibility for old userspace */
1258 +#include <linux/netfilter/xt_IMQ.h>
1259 +
1260 +#define ip6t_imq_info xt_imq_info
1261 +
1262 +#endif /* _IP6T_IMQ_H */
1263 +
1264 diff --git a/include/linux/skbuff.h b/include/linux/skbuff.h
1265 index f66f346..d699b19 100644
1266 --- a/include/linux/skbuff.h
1267 +++ b/include/linux/skbuff.h
1268 @@ -33,6 +33,9 @@
1269  #include <linux/dma-mapping.h>
1270  #include <linux/netdev_features.h>
1271  #include <net/flow_keys.h>
1272 +#if defined(CONFIG_IMQ) || defined(CONFIG_IMQ_MODULE)
1273 +#include <linux/imq.h>
1274 +#endif
1275
1276  /* Don't change this without changing skb_csum_unnecessary! */
1277  #define CHECKSUM_NONE 0
1278 @@ -418,6 +421,9 @@ struct sk_buff {
1279          * first. This is owned by whoever has the skb queued ATM.
1280          */
1281         char                    cb[48] __aligned(8);
1282 +#if defined(CONFIG_IMQ) || defined(CONFIG_IMQ_MODULE)
1283 +       void                    *cb_next;
1284 +#endif
1285
1286         unsigned long           _skb_refdst;
1287  #ifdef CONFIG_XFRM
1288 @@ -453,6 +459,9 @@ struct sk_buff {
1289  #if defined(CONFIG_NF_CONNTRACK) || defined(CONFIG_NF_CONNTRACK_MODULE)
1290         struct nf_conntrack     *nfct;
1291  #endif
1292 +#if defined(CONFIG_IMQ) || defined(CONFIG_IMQ_MODULE)
1293 +       struct nf_queue_entry   *nf_queue_entry;
1294 +#endif
1295  #ifdef CONFIG_BRIDGE_NETFILTER
1296         struct nf_bridge_info   *nf_bridge;
1297  #endif
1298 @@ -490,6 +499,9 @@ struct sk_buff {
1299          */
1300         __u8                    encapsulation:1;
1301         /* 6/8 bit hole (depending on ndisc_nodetype presence) */
1302 +#if defined(CONFIG_IMQ) || defined(CONFIG_IMQ_MODULE)
1303 +       __u8                    imq_flags:IMQ_F_BITS;
1304 +#endif
1305         kmemcheck_bitfield_end(flags2);
1306
1307  #if defined CONFIG_NET_DMA || defined CONFIG_NET_RX_BUSY_POLL
1308 @@ -625,6 +637,12 @@ static inline struct rtable *skb_rtable(const struct sk_buff *skb)
1309         return (struct rtable *)skb_dst(skb);
1310  }
1311
1312 +
1313 +#if defined(CONFIG_IMQ) || defined(CONFIG_IMQ_MODULE)
1314 +extern int skb_save_cb(struct sk_buff *skb);
1315 +extern int skb_restore_cb(struct sk_buff *skb);
1316 +#endif
1317 +
1318  void kfree_skb(struct sk_buff *skb);
1319  void kfree_skb_list(struct sk_buff *segs);
1320  void skb_tx_error(struct sk_buff *skb);
1321 @@ -2635,6 +2653,10 @@ static inline void __nf_copy(struct sk_buff *dst, const struct sk_buff *src)
1322         nf_conntrack_get(src->nfct);
1323         dst->nfctinfo = src->nfctinfo;
1324  #endif
1325 +#if defined(CONFIG_IMQ) || defined(CONFIG_IMQ_MODULE)
1326 +       dst->imq_flags = src->imq_flags;
1327 +       dst->nf_queue_entry = src->nf_queue_entry;
1328 +#endif
1329  #ifdef CONFIG_BRIDGE_NETFILTER
1330         dst->nf_bridge  = src->nf_bridge;
1331         nf_bridge_get(src->nf_bridge);
1332 diff --git a/include/net/netfilter/nf_queue.h b/include/net/netfilter/nf_queue.h
1333 index aaba4bb..f6e92a4 100644
1334 --- a/include/net/netfilter/nf_queue.h
1335 +++ b/include/net/netfilter/nf_queue.h
1336 @@ -29,6 +29,12 @@ struct nf_queue_handler {
1337  void nf_register_queue_handler(const struct nf_queue_handler *qh);
1338  void nf_unregister_queue_handler(void);
1339  void nf_reinject(struct nf_queue_entry *entry, unsigned int verdict);
1340 +extern void nf_queue_entry_release_refs(struct nf_queue_entry *entry);
1341 +
1342 +#if defined(CONFIG_IMQ) || defined(CONFIG_IMQ_MODULE)
1343 +extern void nf_register_queue_imq_handler(const struct nf_queue_handler *qh);
1344 +extern void nf_unregister_queue_imq_handler(void);
1345 +#endif
1346
1347  bool nf_queue_entry_get_refs(struct nf_queue_entry *entry);
1348  void nf_queue_entry_release_refs(struct nf_queue_entry *entry);
1349 diff --git a/include/uapi/linux/netfilter.h b/include/uapi/linux/netfilter.h
1350 index f7dc0eb..58c46a9 100644
1351 --- a/include/uapi/linux/netfilter.h
1352 +++ b/include/uapi/linux/netfilter.h
1353 @@ -13,7 +13,8 @@
1354  #define NF_QUEUE 3
1355  #define NF_REPEAT 4
1356  #define NF_STOP 5
1357 -#define NF_MAX_VERDICT NF_STOP
1358 +#define NF_IMQ_QUEUE 6
1359 +#define NF_MAX_VERDICT NF_IMQ_QUEUE
1360
1361  /* we overload the higher bits for encoding auxiliary data such as the queue
1362   * number or errno values. Not nice, but better than additional function
1363 diff --git a/net/core/dev.c b/net/core/dev.c
1364 index 3d13874..9842f21 100644
1365 --- a/net/core/dev.c
1366 +++ b/net/core/dev.c
1367 @@ -131,6 +131,9 @@
1368  #include <linux/hashtable.h>
1369  #include <linux/vmalloc.h>
1370  #include <linux/if_macvlan.h>
1371 +#if defined(CONFIG_IMQ) || defined(CONFIG_IMQ_MODULE)
1372 +#include <linux/imq.h>
1373 +#endif
1374
1375  #include "net-sysfs.h"
1376
1377 @@ -2595,7 +2598,12 @@ int dev_hard_start_xmit(struct sk_buff *skb, struct net_device *dev,
1378                         }
1379                 }
1380
1381 +#if defined(CONFIG_IMQ) || defined(CONFIG_IMQ_MODULE)
1382 +               if (!list_empty(&ptype_all) &&
1383 +                                       !(skb->imq_flags & IMQ_F_ENQUEUE))
1384 +#else
1385                 if (!list_empty(&ptype_all))
1386 +#endif
1387                         dev_queue_xmit_nit(skb, dev);
1388
1389                 skb_len = skb->len;
1390 diff --git a/net/core/skbuff.c b/net/core/skbuff.c
1391 index c28c7fe..a5f1888 100644
1392 --- a/net/core/skbuff.c
1393 +++ b/net/core/skbuff.c
1394 @@ -73,6 +73,84 @@
1395
1396  struct kmem_cache *skbuff_head_cache __read_mostly;
1397  static struct kmem_cache *skbuff_fclone_cache __read_mostly;
1398 +#if defined(CONFIG_IMQ) || defined(CONFIG_IMQ_MODULE)
1399 +static struct kmem_cache *skbuff_cb_store_cache __read_mostly;
1400 +
1401 +/* Control buffer save/restore for IMQ devices */
1402 +struct skb_cb_table {
1403 +       char                    cb[48] __aligned(8);
1404 +       void                    *cb_next;
1405 +       atomic_t                refcnt;
1406 +};
1407 +
1408 +static DEFINE_SPINLOCK(skb_cb_store_lock);
1409 +
1410 +int skb_save_cb(struct sk_buff *skb)
1411 +{
1412 +       struct skb_cb_table *next;
1413 +
1414 +       next = kmem_cache_alloc(skbuff_cb_store_cache, GFP_ATOMIC);
1415 +       if (!next)
1416 +               return -ENOMEM;
1417 +
1418 +       BUILD_BUG_ON(sizeof(skb->cb) != sizeof(next->cb));
1419 +
1420 +       memcpy(next->cb, skb->cb, sizeof(skb->cb));
1421 +       next->cb_next = skb->cb_next;
1422 +
1423 +       atomic_set(&next->refcnt, 1);
1424 +
1425 +       skb->cb_next = next;
1426 +       return 0;
1427 +}
1428 +EXPORT_SYMBOL(skb_save_cb);
1429 +
1430 +int skb_restore_cb(struct sk_buff *skb)
1431 +{
1432 +       struct skb_cb_table *next;
1433 +
1434 +       if (!skb->cb_next)
1435 +               return 0;
1436 +
1437 +       next = skb->cb_next;
1438 +
1439 +       BUILD_BUG_ON(sizeof(skb->cb) != sizeof(next->cb));
1440 +
1441 +       memcpy(skb->cb, next->cb, sizeof(skb->cb));
1442 +       skb->cb_next = next->cb_next;
1443 +
1444 +       spin_lock(&skb_cb_store_lock);
1445 +
1446 +       if (atomic_dec_and_test(&next->refcnt))
1447 +               kmem_cache_free(skbuff_cb_store_cache, next);
1448 +
1449 +       spin_unlock(&skb_cb_store_lock);
1450 +
1451 +       return 0;
1452 +}
1453 +EXPORT_SYMBOL(skb_restore_cb);
1454 +
1455 +static void skb_copy_stored_cb(struct sk_buff *new, const struct sk_buff *__old)
1456 +{
1457 +       struct skb_cb_table *next;
1458 +       struct sk_buff *old;
1459 +
1460 +       if (!__old->cb_next) {
1461 +               new->cb_next = NULL;
1462 +               return;
1463 +       }
1464 +
1465 +       spin_lock(&skb_cb_store_lock);
1466 +
1467 +       old = (struct sk_buff *)__old;
1468 +
1469 +       next = old->cb_next;
1470 +       atomic_inc(&next->refcnt);
1471 +       new->cb_next = next;
1472 +
1473 +       spin_unlock(&skb_cb_store_lock);
1474 +}
1475 +#endif
1476
1477  /**
1478   *     skb_panic - private function for out-of-line support
1479 @@ -577,6 +656,28 @@ static void skb_release_head_state(struct sk_buff *skb)
1480                 WARN_ON(in_irq());
1481                 skb->destructor(skb);
1482         }
1483 +#if defined(CONFIG_IMQ) || defined(CONFIG_IMQ_MODULE)
1484 +       /*
1485 +        * This should not happen. When it does, avoid memleak by restoring
1486 +        * the chain of cb-backups.
1487 +        */
1488 +       while (skb->cb_next != NULL) {
1489 +               if (net_ratelimit())
1490 +                       pr_warn("IMQ: kfree_skb: skb->cb_next: %08x\n",
1491 +                               (unsigned int)skb->cb_next);
1492 +
1493 +               skb_restore_cb(skb);
1494 +       }
1495 +       /*
1496 +        * This should not happen either, nf_queue_entry is nullified in
1497 +        * imq_dev_xmit(). If we have non-NULL nf_queue_entry then we are
1498 +        * leaking entry pointers, maybe memory. We don't know if this is
1499 +        * pointer to already freed memory, or should this be freed.
1500 +        * If this happens we need to add refcounting, etc for nf_queue_entry.
1501 +        */
1502 +       if (skb->nf_queue_entry && net_ratelimit())
1503 +               pr_warn("%s\n", "IMQ: kfree_skb: skb->nf_queue_entry != NULL");
1504 +#endif
1505  #if IS_ENABLED(CONFIG_NF_CONNTRACK)
1506         nf_conntrack_put(skb->nfct);
1507  #endif
1508 @@ -709,6 +810,10 @@ static void __copy_skb_header(struct sk_buff *new, const struct sk_buff *old)
1509         new->sp                 = secpath_get(old->sp);
1510  #endif
1511         memcpy(new->cb, old->cb, sizeof(old->cb));
1512 +#if defined(CONFIG_IMQ) || defined(CONFIG_IMQ_MODULE)
1513 +       new->cb_next = NULL;
1514 +       /*skb_copy_stored_cb(new, old);*/
1515 +#endif
1516         new->csum               = old->csum;
1517         new->local_df           = old->local_df;
1518         new->pkt_type           = old->pkt_type;
1519 @@ -3112,6 +3217,13 @@ void __init skb_init(void)
1520                                                 0,
1521                                                 SLAB_HWCACHE_ALIGN|SLAB_PANIC,
1522                                                 NULL);
1523 +#if defined(CONFIG_IMQ) || defined(CONFIG_IMQ_MODULE)
1524 +       skbuff_cb_store_cache = kmem_cache_create("skbuff_cb_store_cache",
1525 +                                                 sizeof(struct skb_cb_table),
1526 +                                                 0,
1527 +                                                 SLAB_HWCACHE_ALIGN|SLAB_PANIC,
1528 +                                                 NULL);
1529 +#endif
1530  }
1531
1532  /**
1533 diff --git a/net/ipv6/ip6_output.c b/net/ipv6/ip6_output.c
1534 index b6fa35e..08dcfef 100644
1535 --- a/net/ipv6/ip6_output.c
1536 +++ b/net/ipv6/ip6_output.c
1537 @@ -64,9 +64,6 @@ static int ip6_finish_output2(struct sk_buff *skb)
1538         struct in6_addr *nexthop;
1539         int ret;
1540
1541 -       skb->protocol = htons(ETH_P_IPV6);
1542 -       skb->dev = dev;
1543 -
1544         if (ipv6_addr_is_multicast(&ipv6_hdr(skb)->daddr)) {
1545                 struct inet6_dev *idev = ip6_dst_idev(skb_dst(skb));
1546
1547 @@ -143,6 +140,13 @@ int ip6_output(struct sk_buff *skb)
1548                 return 0;
1549         }
1550
1551 +       /*
1552 +        * IMQ-patch: moved setting skb->dev and skb->protocol from
1553 +        * ip6_finish_output2 to fix crashing at netif_skb_features().
1554 +        */
1555 +       skb->protocol = htons(ETH_P_IPV6);
1556 +       skb->dev = dev;
1557 +
1558         return NF_HOOK_COND(NFPROTO_IPV6, NF_INET_POST_ROUTING, skb, NULL, dev,
1559                             ip6_finish_output,
1560                             !(IP6CB(skb)->flags & IP6SKB_REROUTED));
1561 diff --git a/net/netfilter/Kconfig b/net/netfilter/Kconfig
1562 index 6e839b6..45ac31c 100644
1563 --- a/net/netfilter/Kconfig
1564 +++ b/net/netfilter/Kconfig
1565 @@ -630,6 +630,18 @@ config NETFILTER_XT_TARGET_LOG
1566
1567           To compile it as a module, choose M here.  If unsure, say N.
1568
1569 +config NETFILTER_XT_TARGET_IMQ
1570 +        tristate '"IMQ" target support'
1571 +       depends on NETFILTER_XTABLES
1572 +       depends on IP_NF_MANGLE || IP6_NF_MANGLE
1573 +       select IMQ
1574 +       default m if NETFILTER_ADVANCED=n
1575 +        help
1576 +          This option adds a `IMQ' target which is used to specify if and
1577 +          to which imq device packets should get enqueued/dequeued.
1578 +
1579 +          To compile it as a module, choose M here.  If unsure, say N.
1580 +
1581  config NETFILTER_XT_TARGET_MARK
1582         tristate '"MARK" target support'
1583         depends on NETFILTER_ADVANCED
1584 diff --git a/net/netfilter/Makefile b/net/netfilter/Makefile
1585 index c3a0a12..9647f06 100644
1586 --- a/net/netfilter/Makefile
1587 +++ b/net/netfilter/Makefile
1588 @@ -82,6 +82,7 @@ obj-$(CONFIG_NETFILTER_XT_TARGET_CT) += xt_CT.o
1589  obj-$(CONFIG_NETFILTER_XT_TARGET_DSCP) += xt_DSCP.o
1590  obj-$(CONFIG_NETFILTER_XT_TARGET_HL) += xt_HL.o
1591  obj-$(CONFIG_NETFILTER_XT_TARGET_HMARK) += xt_HMARK.o
1592 +obj-$(CONFIG_NETFILTER_XT_TARGET_IMQ) += xt_IMQ.o
1593  obj-$(CONFIG_NETFILTER_XT_TARGET_LED) += xt_LED.o
1594  obj-$(CONFIG_NETFILTER_XT_TARGET_LOG) += xt_LOG.o
1595  obj-$(CONFIG_NETFILTER_XT_TARGET_NETMAP) += xt_NETMAP.o
1596 diff --git a/net/netfilter/core.c b/net/netfilter/core.c
1597 index 593b16e..740cd69 100644
1598 --- a/net/netfilter/core.c
1599 +++ b/net/netfilter/core.c
1600 @@ -191,9 +191,11 @@ next_hook:
1601                 ret = NF_DROP_GETERR(verdict);
1602                 if (ret == 0)
1603                         ret = -EPERM;
1604 -       } else if ((verdict & NF_VERDICT_MASK) == NF_QUEUE) {
1605 +       } else if ((verdict & NF_VERDICT_MASK) == NF_QUEUE ||
1606 +                  (verdict & NF_VERDICT_MASK) == NF_IMQ_QUEUE) {
1607                 int err = nf_queue(skb, elem, pf, hook, indev, outdev, okfn,
1608 -                                               verdict >> NF_VERDICT_QBITS);
1609 +                                               verdict >> NF_VERDICT_QBITS,
1610 +                                               verdict & NF_VERDICT_MASK);
1611                 if (err < 0) {
1612                         if (err == -ECANCELED)
1613                                 goto next_hook;
1614 diff --git a/net/netfilter/nf_internals.h b/net/netfilter/nf_internals.h
1615 index 3deec99..c1a1397 100644
1616 --- a/net/netfilter/nf_internals.h
1617 +++ b/net/netfilter/nf_internals.h
1618 @@ -29,7 +29,7 @@ extern int nf_queue(struct sk_buff *skb,
1619  int nf_queue(struct sk_buff *skb, struct nf_hook_ops *elem, u_int8_t pf,
1620              unsigned int hook, struct net_device *indev,
1621              struct net_device *outdev, int (*okfn)(struct sk_buff *),
1622 -            unsigned int queuenum);
1623 +            unsigned int queuenum, unsigned int queuetype);
1624  int __init netfilter_queue_init(void);
1625
1626  /* nf_log.c */
1627 diff --git a/net/netfilter/nf_queue.c b/net/netfilter/nf_queue.c
1628 index 5d24b1f..28317dc 100644
1629 --- a/net/netfilter/nf_queue.c
1630 +++ b/net/netfilter/nf_queue.c
1631 @@ -27,6 +27,23 @@
1632   */
1633  static const struct nf_queue_handler __rcu *queue_handler __read_mostly;
1634
1635 +#if defined(CONFIG_IMQ) || defined(CONFIG_IMQ_MODULE)
1636 +static const struct nf_queue_handler __rcu *queue_imq_handler __read_mostly;
1637 +
1638 +void nf_register_queue_imq_handler(const struct nf_queue_handler *qh)
1639 +{
1640 +       rcu_assign_pointer(queue_imq_handler, qh);
1641 +}
1642 +EXPORT_SYMBOL_GPL(nf_register_queue_imq_handler);
1643 +
1644 +void nf_unregister_queue_imq_handler(void)
1645 +{
1646 +       RCU_INIT_POINTER(queue_imq_handler, NULL);
1647 +       synchronize_rcu();
1648 +}
1649 +EXPORT_SYMBOL_GPL(nf_unregister_queue_imq_handler);
1650 +#endif
1651 +
1652  /* return EBUSY when somebody else is registered, return EEXIST if the
1653   * same handler is registered, return 0 in case of success. */
1654  void nf_register_queue_handler(const struct nf_queue_handler *qh)
1655 @@ -105,7 +122,8 @@ int nf_queue(struct sk_buff *skb,
1656                       struct net_device *indev,
1657                       struct net_device *outdev,
1658                       int (*okfn)(struct sk_buff *),
1659 -                     unsigned int queuenum)
1660 +                     unsigned int queuenum,
1661 +                     unsigned int queuetype)
1662  {
1663         int status = -ENOENT;
1664         struct nf_queue_entry *entry = NULL;
1665 @@ -115,7 +133,17 @@ int nf_queue(struct sk_buff *skb,
1666         /* QUEUE == DROP if no one is waiting, to be safe. */
1667         rcu_read_lock();
1668
1669 -       qh = rcu_dereference(queue_handler);
1670 +       if (queuetype == NF_IMQ_QUEUE) {
1671 +#if defined(CONFIG_IMQ) || defined(CONFIG_IMQ_MODULE)
1672 +               qh = rcu_dereference(queue_imq_handler);
1673 +#else
1674 +               BUG();
1675 +               goto err_unlock;
1676 +#endif
1677 +       } else {
1678 +               qh = rcu_dereference(queue_handler);
1679 +       }
1680 +
1681         if (!qh) {
1682                 status = -ESRCH;
1683                 goto err_unlock;
1684 @@ -205,9 +233,11 @@ void nf_reinject(struct nf_queue_entry *entry, unsigned int verdict)
1685                 local_bh_enable();
1686                 break;
1687         case NF_QUEUE:
1688 +       case NF_IMQ_QUEUE:
1689                 err = nf_queue(skb, elem, entry->pf, entry->hook,
1690                                 entry->indev, entry->outdev, entry->okfn,
1691 -                               verdict >> NF_VERDICT_QBITS);
1692 +                               verdict >> NF_VERDICT_QBITS,
1693 +                               verdict & NF_VERDICT_MASK);
1694                 if (err < 0) {
1695                         if (err == -ECANCELED)
1696                                 goto next_hook;
1697 diff --git a/net/netfilter/xt_IMQ.c b/net/netfilter/xt_IMQ.c
1698 new file mode 100644
1699 index 0000000..1c3cd66
1700 --- /dev/null
1701 +++ b/net/netfilter/xt_IMQ.c
1702 @@ -0,0 +1,72 @@
1703 +/*
1704 + * This target marks packets to be enqueued to an imq device
1705 + */
1706 +#include <linux/module.h>
1707 +#include <linux/skbuff.h>
1708 +#include <linux/netfilter/x_tables.h>
1709 +#include <linux/netfilter/xt_IMQ.h>
1710 +#include <linux/imq.h>
1711 +
1712 +static unsigned int imq_target(struct sk_buff *pskb,
1713 +                               const struct xt_action_param *par)
1714 +{
1715 +       const struct xt_imq_info *mr = par->targinfo;
1716 +
1717 +       pskb->imq_flags = (mr->todev & IMQ_F_IFMASK) | IMQ_F_ENQUEUE;
1718 +
1719 +       return XT_CONTINUE;
1720 +}
1721 +
1722 +static int imq_checkentry(const struct xt_tgchk_param *par)
1723 +{
1724 +       struct xt_imq_info *mr = par->targinfo;
1725 +
1726 +       if (mr->todev > IMQ_MAX_DEVS - 1) {
1727 +               pr_warn("IMQ: invalid device specified, highest is %u\n",
1728 +                       IMQ_MAX_DEVS - 1);
1729 +               return -EINVAL;
1730 +       }
1731 +
1732 +       return 0;
1733 +}
1734 +
1735 +static struct xt_target xt_imq_reg[] __read_mostly = {
1736 +       {
1737 +               .name           = "IMQ",
1738 +               .family         = AF_INET,
1739 +               .checkentry     = imq_checkentry,
1740 +               .target         = imq_target,
1741 +               .targetsize     = sizeof(struct xt_imq_info),
1742 +               .table          = "mangle",
1743 +               .me             = THIS_MODULE
1744 +       },
1745 +       {
1746 +               .name           = "IMQ",
1747 +               .family         = AF_INET6,
1748 +               .checkentry     = imq_checkentry,
1749 +               .target         = imq_target,
1750 +               .targetsize     = sizeof(struct xt_imq_info),
1751 +               .table          = "mangle",
1752 +               .me             = THIS_MODULE
1753 +       },
1754 +};
1755 +
1756 +static int __init imq_init(void)
1757 +{
1758 +       return xt_register_targets(xt_imq_reg, ARRAY_SIZE(xt_imq_reg));
1759 +}
1760 +
1761 +static void __exit imq_fini(void)
1762 +{
1763 +       xt_unregister_targets(xt_imq_reg, ARRAY_SIZE(xt_imq_reg));
1764 +}
1765 +
1766 +module_init(imq_init);
1767 +module_exit(imq_fini);
1768 +
1769 +MODULE_AUTHOR("http://www.linuximq.net");
1770 +MODULE_DESCRIPTION("Pseudo-driver for the intermediate queue device. See http://www.linuximq.net/ for more information.");
1771 +MODULE_LICENSE("GPL");
1772 +MODULE_ALIAS("ipt_IMQ");
1773 +MODULE_ALIAS("ip6t_IMQ");
1774 +