| 1 | --- /dev/null |
| 2 | +++ b/drivers/net/imq.c |
| 3 | @@ -0,0 +1,632 @@ |
| 4 | +/* |
| 5 | + * Pseudo-driver for the intermediate queue device. |
| 6 | + * |
| 7 | + * This program is free software; you can redistribute it and/or |
| 8 | + * modify it under the terms of the GNU General Public License |
| 9 | + * as published by the Free Software Foundation; either version |
| 10 | + * 2 of the License, or (at your option) any later version. |
| 11 | + * |
| 12 | + * Authors: Patrick McHardy, <kaber@trash.net> |
| 13 | + * |
| 14 | + * The first version was written by Martin Devera, <devik@cdi.cz> |
| 15 | + * |
| 16 | + * Credits: Jan Rafaj <imq2t@cedric.vabo.cz> |
| 17 | + * - Update patch to 2.4.21 |
| 18 | + * Sebastian Strollo <sstrollo@nortelnetworks.com> |
| 19 | + * - Fix "Dead-loop on netdevice imq"-issue |
| 20 | + * Marcel Sebek <sebek64@post.cz> |
| 21 | + * - Update to 2.6.2-rc1 |
| 22 | + * |
| 23 | + * After some time of inactivity there is a group taking care |
| 24 | + * of IMQ again: http://www.linuximq.net |
| 25 | + * |
| 26 | + * |
| 27 | + * 2004/06/30 - New version of IMQ patch to kernels <=2.6.7 |
| 28 | + * including the following changes: |
| 29 | + * |
| 30 | + * - Correction of ipv6 support "+"s issue (Hasso Tepper) |
| 31 | + * - Correction of imq_init_devs() issue that resulted in |
| 32 | + * kernel OOPS unloading IMQ as module (Norbert Buchmuller) |
| 33 | + * - Addition of functionality to choose number of IMQ devices |
| 34 | + * during kernel config (Andre Correa) |
| 35 | + * - Addition of functionality to choose how IMQ hooks on |
| 36 | + * PRE and POSTROUTING (after or before NAT) (Andre Correa) |
| 37 | + * - Cosmetic corrections (Norbert Buchmuller) (Andre Correa) |
| 38 | + * |
| 39 | + * |
| 40 | + * 2005/12/16 - IMQ versions between 2.6.7 and 2.6.13 were |
| 41 | + * released with almost no problems. 2.6.14-x was released |
| 42 | + * with some important changes: nfcache was removed; After |
| 43 | + * some weeks of trouble we figured out that some IMQ fields |
| 44 | + * in skb were missing in skbuff.c - skb_clone and copy_skb_header. |
| 45 | + * These functions are correctly patched by this new patch version. |
| 46 | + * |
| 47 | + * Thanks for all who helped to figure out all the problems with |
| 48 | + * 2.6.14.x: Patrick McHardy, Rune Kock, VeNoMouS, Max CtRiX, |
| 49 | + * Kevin Shanahan, Richard Lucassen, Valery Dachev (hopefully |
| 50 | + * I didn't forget anybody). I apologize again for my lack of time. |
| 51 | + * |
| 52 | + * |
| 53 | + * 2008/06/17 - 2.6.25 - Changed imq.c to use qdisc_run() instead |
| 54 | + * of qdisc_restart() and moved qdisc_run() to tasklet to avoid |
| 55 | + * recursive locking. New initialization routines to fix 'rmmod' not |
| 56 | + * working anymore. Used code from ifb.c. (Jussi Kivilinna) |
| 57 | + * |
| 58 | + * 2008/08/06 - 2.6.26 - (JK) |
| 59 | + * - Replaced tasklet with 'netif_schedule()'. |
| 60 | + * - Cleaned up and added comments for imq_nf_queue(). |
| 61 | + * |
| 62 | + * 2009/04/12 |
| 63 | + * - Add skb_save_cb/skb_restore_cb helper functions for backuping |
| 64 | + * control buffer. This is needed because qdisc-layer on kernels |
| 65 | + * 2.6.27 and newer overwrite control buffer. (Jussi Kivilinna) |
| 66 | + * - Add better locking for IMQ device. Hopefully this will solve |
| 67 | + * SMP issues. (Jussi Kivilinna) |
| 68 | + * - Port to 2.6.27 |
| 69 | + * - Port to 2.6.28 |
| 70 | + * - Port to 2.6.29 + fix rmmod not working |
| 71 | + * |
| 72 | + * 2009/04/20 - (Jussi Kivilinna) |
| 73 | + * - Use netdevice feature flags to avoid extra packet handling |
| 74 | + * by core networking layer and possibly increase performance. |
| 75 | + * |
| 76 | + * 2009/09/26 - (Jussi Kivilinna) |
| 77 | + * - Add imq_nf_reinject_lockless to fix deadlock with |
| 78 | + * imq_nf_queue/imq_nf_reinject. |
| 79 | + * |
| 80 | + * 2009/12/08 - (Jussi Kivilinna) |
| 81 | + * - Port to 2.6.32 |
| 82 | + * - Add check for skb->nf_queue_entry==NULL in imq_dev_xmit() |
| 83 | + * - Also add better error checking for skb->nf_queue_entry usage |
| 84 | + * |
| 85 | + * Also, many thanks to pablo Sebastian Greco for making the initial |
| 86 | + * patch and to those who helped the testing. |
| 87 | + * |
| 88 | + * More info at: http://www.linuximq.net/ (Andre Correa) |
| 89 | + */ |
| 90 | + |
| 91 | +#include <linux/module.h> |
| 92 | +#include <linux/kernel.h> |
| 93 | +#include <linux/moduleparam.h> |
| 94 | +#include <linux/list.h> |
| 95 | +#include <linux/skbuff.h> |
| 96 | +#include <linux/netdevice.h> |
| 97 | +#include <linux/etherdevice.h> |
| 98 | +#include <linux/rtnetlink.h> |
| 99 | +#include <linux/if_arp.h> |
| 100 | +#include <linux/netfilter.h> |
| 101 | +#include <linux/netfilter_ipv4.h> |
| 102 | +#if defined(CONFIG_IPV6) || defined(CONFIG_IPV6_MODULE) |
| 103 | + #include <linux/netfilter_ipv6.h> |
| 104 | +#endif |
| 105 | +#include <linux/imq.h> |
| 106 | +#include <net/pkt_sched.h> |
| 107 | +#include <net/netfilter/nf_queue.h> |
| 108 | + |
| 109 | +static nf_hookfn imq_nf_hook; |
| 110 | + |
| 111 | +static struct nf_hook_ops imq_ingress_ipv4 = { |
| 112 | + .hook = imq_nf_hook, |
| 113 | + .owner = THIS_MODULE, |
| 114 | + .pf = PF_INET, |
| 115 | + .hooknum = NF_INET_PRE_ROUTING, |
| 116 | +#if defined(CONFIG_IMQ_BEHAVIOR_BA) || defined(CONFIG_IMQ_BEHAVIOR_BB) |
| 117 | + .priority = NF_IP_PRI_MANGLE + 1 |
| 118 | +#else |
| 119 | + .priority = NF_IP_PRI_NAT_DST + 1 |
| 120 | +#endif |
| 121 | +}; |
| 122 | + |
| 123 | +static struct nf_hook_ops imq_egress_ipv4 = { |
| 124 | + .hook = imq_nf_hook, |
| 125 | + .owner = THIS_MODULE, |
| 126 | + .pf = PF_INET, |
| 127 | + .hooknum = NF_INET_POST_ROUTING, |
| 128 | +#if defined(CONFIG_IMQ_BEHAVIOR_AA) || defined(CONFIG_IMQ_BEHAVIOR_BA) |
| 129 | + .priority = NF_IP_PRI_LAST |
| 130 | +#else |
| 131 | + .priority = NF_IP_PRI_NAT_SRC - 1 |
| 132 | +#endif |
| 133 | +}; |
| 134 | + |
| 135 | +#if defined(CONFIG_IPV6) || defined(CONFIG_IPV6_MODULE) |
| 136 | +static struct nf_hook_ops imq_ingress_ipv6 = { |
| 137 | + .hook = imq_nf_hook, |
| 138 | + .owner = THIS_MODULE, |
| 139 | + .pf = PF_INET6, |
| 140 | + .hooknum = NF_INET_PRE_ROUTING, |
| 141 | +#if defined(CONFIG_IMQ_BEHAVIOR_BA) || defined(CONFIG_IMQ_BEHAVIOR_BB) |
| 142 | + .priority = NF_IP6_PRI_MANGLE + 1 |
| 143 | +#else |
| 144 | + .priority = NF_IP6_PRI_NAT_DST + 1 |
| 145 | +#endif |
| 146 | +}; |
| 147 | + |
| 148 | +static struct nf_hook_ops imq_egress_ipv6 = { |
| 149 | + .hook = imq_nf_hook, |
| 150 | + .owner = THIS_MODULE, |
| 151 | + .pf = PF_INET6, |
| 152 | + .hooknum = NF_INET_POST_ROUTING, |
| 153 | +#if defined(CONFIG_IMQ_BEHAVIOR_AA) || defined(CONFIG_IMQ_BEHAVIOR_BA) |
| 154 | + .priority = NF_IP6_PRI_LAST |
| 155 | +#else |
| 156 | + .priority = NF_IP6_PRI_NAT_SRC - 1 |
| 157 | +#endif |
| 158 | +}; |
| 159 | +#endif |
| 160 | + |
| 161 | +#if defined(CONFIG_IMQ_NUM_DEVS) |
| 162 | +static unsigned int numdevs = CONFIG_IMQ_NUM_DEVS; |
| 163 | +#else |
| 164 | +static unsigned int numdevs = IMQ_MAX_DEVS; |
| 165 | +#endif |
| 166 | + |
| 167 | +static DEFINE_SPINLOCK(imq_nf_queue_lock); |
| 168 | + |
| 169 | +static struct net_device *imq_devs_cache[IMQ_MAX_DEVS]; |
| 170 | + |
| 171 | + |
| 172 | +static struct net_device_stats *imq_get_stats(struct net_device *dev) |
| 173 | +{ |
| 174 | + return &dev->stats; |
| 175 | +} |
| 176 | + |
| 177 | +/* called for packets kfree'd in qdiscs at places other than enqueue */ |
| 178 | +static void imq_skb_destructor(struct sk_buff *skb) |
| 179 | +{ |
| 180 | + struct nf_queue_entry *entry = skb->nf_queue_entry; |
| 181 | + |
| 182 | + skb->nf_queue_entry = NULL; |
| 183 | + |
| 184 | + if (entry) { |
| 185 | + nf_queue_entry_release_refs(entry); |
| 186 | + kfree(entry); |
| 187 | + } |
| 188 | + |
| 189 | + skb_restore_cb(skb); /* kfree backup */ |
| 190 | +} |
| 191 | + |
| 192 | +/* locking not needed when called from imq_nf_queue */ |
| 193 | +static void imq_nf_reinject_lockless(struct nf_queue_entry *entry, |
| 194 | + unsigned int verdict) |
| 195 | +{ |
| 196 | + int status; |
| 197 | + |
| 198 | + if (!entry->next_outfn) { |
| 199 | + nf_reinject(entry, verdict); |
| 200 | + return; |
| 201 | + } |
| 202 | + |
| 203 | + status = entry->next_outfn(entry, entry->next_queuenum); |
| 204 | + if (status < 0) { |
| 205 | + nf_queue_entry_release_refs(entry); |
| 206 | + kfree_skb(entry->skb); |
| 207 | + kfree(entry); |
| 208 | + } |
| 209 | +} |
| 210 | + |
| 211 | +static void imq_nf_reinject(struct nf_queue_entry *entry, unsigned int verdict) |
| 212 | +{ |
| 213 | + int status; |
| 214 | + |
| 215 | + if (!entry->next_outfn) { |
| 216 | + spin_lock_bh(&imq_nf_queue_lock); |
| 217 | + nf_reinject(entry, verdict); |
| 218 | + spin_unlock_bh(&imq_nf_queue_lock); |
| 219 | + return; |
| 220 | + } |
| 221 | + |
| 222 | + rcu_read_lock(); |
| 223 | + local_bh_disable(); |
| 224 | + status = entry->next_outfn(entry, entry->next_queuenum); |
| 225 | + local_bh_enable(); |
| 226 | + if (status < 0) { |
| 227 | + nf_queue_entry_release_refs(entry); |
| 228 | + kfree_skb(entry->skb); |
| 229 | + kfree(entry); |
| 230 | + } |
| 231 | + |
| 232 | + rcu_read_unlock(); |
| 233 | +} |
| 234 | + |
| 235 | +static netdev_tx_t imq_dev_xmit(struct sk_buff *skb, struct net_device *dev) |
| 236 | +{ |
| 237 | + struct nf_queue_entry *entry = skb->nf_queue_entry; |
| 238 | + |
| 239 | + skb->nf_queue_entry = NULL; |
| 240 | + dev->trans_start = jiffies; |
| 241 | + |
| 242 | + dev->stats.tx_bytes += skb->len; |
| 243 | + dev->stats.tx_packets++; |
| 244 | + |
| 245 | + if (entry == NULL) { |
| 246 | + /* We don't know what is going on here.. packet is queued for |
| 247 | + * imq device, but (probably) not by us. |
| 248 | + * |
| 249 | + * If this packet was not send here by imq_nf_queue(), then |
| 250 | + * skb_save_cb() was not used and skb_free() should not show: |
| 251 | + * WARNING: IMQ: kfree_skb: skb->cb_next:.. |
| 252 | + * and/or |
| 253 | + * WARNING: IMQ: kfree_skb: skb->nf_queue_entry... |
| 254 | + * |
| 255 | + * However if this message is shown, then IMQ is somehow broken |
| 256 | + * and you should report this to linuximq.net. |
| 257 | + */ |
| 258 | + |
| 259 | + /* imq_dev_xmit is black hole that eats all packets, report that |
| 260 | + * we eat this packet happily and increase dropped counters. |
| 261 | + */ |
| 262 | + |
| 263 | + dev->stats.tx_dropped++; |
| 264 | + dev_kfree_skb(skb); |
| 265 | + |
| 266 | + return NETDEV_TX_OK; |
| 267 | + } |
| 268 | + |
| 269 | + skb_restore_cb(skb); /* restore skb->cb */ |
| 270 | + |
| 271 | + skb->imq_flags = 0; |
| 272 | + skb->destructor = NULL; |
| 273 | + |
| 274 | + imq_nf_reinject(entry, NF_ACCEPT); |
| 275 | + |
| 276 | + return NETDEV_TX_OK; |
| 277 | +} |
| 278 | + |
| 279 | +static int imq_nf_queue(struct nf_queue_entry *entry, unsigned queue_num) |
| 280 | +{ |
| 281 | + struct net_device *dev; |
| 282 | + struct sk_buff *skb_orig, *skb, *skb_shared; |
| 283 | + struct Qdisc *q; |
| 284 | + struct netdev_queue *txq; |
| 285 | + int users, index; |
| 286 | + int retval = -EINVAL; |
| 287 | + |
| 288 | + index = entry->skb->imq_flags & IMQ_F_IFMASK; |
| 289 | + if (unlikely(index > numdevs - 1)) { |
| 290 | + if (net_ratelimit()) |
| 291 | + printk(KERN_WARNING |
| 292 | + "IMQ: invalid device specified, highest is %u\n", |
| 293 | + numdevs - 1); |
| 294 | + retval = -EINVAL; |
| 295 | + goto out; |
| 296 | + } |
| 297 | + |
| 298 | + /* check for imq device by index from cache */ |
| 299 | + dev = imq_devs_cache[index]; |
| 300 | + if (unlikely(!dev)) { |
| 301 | + char buf[8]; |
| 302 | + |
| 303 | + /* get device by name and cache result */ |
| 304 | + snprintf(buf, sizeof(buf), "imq%d", index); |
| 305 | + dev = dev_get_by_name(&init_net, buf); |
| 306 | + if (!dev) { |
| 307 | + /* not found ?!*/ |
| 308 | + BUG(); |
| 309 | + retval = -ENODEV; |
| 310 | + goto out; |
| 311 | + } |
| 312 | + |
| 313 | + imq_devs_cache[index] = dev; |
| 314 | + dev_put(dev); |
| 315 | + } |
| 316 | + |
| 317 | + if (unlikely(!(dev->flags & IFF_UP))) { |
| 318 | + entry->skb->imq_flags = 0; |
| 319 | + imq_nf_reinject_lockless(entry, NF_ACCEPT); |
| 320 | + retval = 0; |
| 321 | + goto out; |
| 322 | + } |
| 323 | + dev->last_rx = jiffies; |
| 324 | + |
| 325 | + skb = entry->skb; |
| 326 | + skb_orig = NULL; |
| 327 | + |
| 328 | + /* skb has owner? => make clone */ |
| 329 | + if (unlikely(skb->destructor)) { |
| 330 | + skb_orig = skb; |
| 331 | + skb = skb_clone(skb, GFP_ATOMIC); |
| 332 | + if (!skb) { |
| 333 | + retval = -ENOMEM; |
| 334 | + goto out; |
| 335 | + } |
| 336 | + entry->skb = skb; |
| 337 | + } |
| 338 | + |
| 339 | + skb->nf_queue_entry = entry; |
| 340 | + |
| 341 | + dev->stats.rx_bytes += skb->len; |
| 342 | + dev->stats.rx_packets++; |
| 343 | + |
| 344 | + txq = dev_pick_tx(dev, skb); |
| 345 | + |
| 346 | + q = rcu_dereference(txq->qdisc); |
| 347 | + if (unlikely(!q->enqueue)) |
| 348 | + goto packet_not_eaten_by_imq_dev; |
| 349 | + |
| 350 | + spin_lock_bh(qdisc_lock(q)); |
| 351 | + |
| 352 | + users = atomic_read(&skb->users); |
| 353 | + |
| 354 | + skb_shared = skb_get(skb); /* increase reference count by one */ |
| 355 | + skb_save_cb(skb_shared); /* backup skb->cb, as qdisc layer will |
| 356 | + overwrite it */ |
| 357 | + qdisc_enqueue_root(skb_shared, q); /* might kfree_skb */ |
| 358 | + |
| 359 | + if (likely(atomic_read(&skb_shared->users) == users + 1)) { |
| 360 | + kfree_skb(skb_shared); /* decrease reference count by one */ |
| 361 | + |
| 362 | + skb->destructor = &imq_skb_destructor; |
| 363 | + |
| 364 | + /* cloned? */ |
| 365 | + if (skb_orig) |
| 366 | + kfree_skb(skb_orig); /* free original */ |
| 367 | + |
| 368 | + spin_unlock_bh(qdisc_lock(q)); |
| 369 | + |
| 370 | + /* schedule qdisc dequeue */ |
| 371 | + __netif_schedule(q); |
| 372 | + |
| 373 | + retval = 0; |
| 374 | + goto out; |
| 375 | + } else { |
| 376 | + skb_restore_cb(skb_shared); /* restore skb->cb */ |
| 377 | + skb->nf_queue_entry = NULL; |
| 378 | + /* qdisc dropped packet and decreased skb reference count of |
| 379 | + * skb, so we don't really want to and try refree as that would |
| 380 | + * actually destroy the skb. */ |
| 381 | + spin_unlock_bh(qdisc_lock(q)); |
| 382 | + goto packet_not_eaten_by_imq_dev; |
| 383 | + } |
| 384 | + |
| 385 | +packet_not_eaten_by_imq_dev: |
| 386 | + /* cloned? restore original */ |
| 387 | + if (skb_orig) { |
| 388 | + kfree_skb(skb); |
| 389 | + entry->skb = skb_orig; |
| 390 | + } |
| 391 | + retval = -1; |
| 392 | +out: |
| 393 | + return retval; |
| 394 | +} |
| 395 | + |
| 396 | +static struct nf_queue_handler nfqh = { |
| 397 | + .name = "imq", |
| 398 | + .outfn = imq_nf_queue, |
| 399 | +}; |
| 400 | + |
| 401 | +static unsigned int imq_nf_hook(unsigned int hook, struct sk_buff *pskb, |
| 402 | + const struct net_device *indev, |
| 403 | + const struct net_device *outdev, |
| 404 | + int (*okfn)(struct sk_buff *)) |
| 405 | +{ |
| 406 | + if (pskb->imq_flags & IMQ_F_ENQUEUE) |
| 407 | + return NF_QUEUE; |
| 408 | + |
| 409 | + return NF_ACCEPT; |
| 410 | +} |
| 411 | + |
| 412 | +static int imq_close(struct net_device *dev) |
| 413 | +{ |
| 414 | + netif_stop_queue(dev); |
| 415 | + return 0; |
| 416 | +} |
| 417 | + |
| 418 | +static int imq_open(struct net_device *dev) |
| 419 | +{ |
| 420 | + netif_start_queue(dev); |
| 421 | + return 0; |
| 422 | +} |
| 423 | + |
| 424 | +static const struct net_device_ops imq_netdev_ops = { |
| 425 | + .ndo_open = imq_open, |
| 426 | + .ndo_stop = imq_close, |
| 427 | + .ndo_start_xmit = imq_dev_xmit, |
| 428 | + .ndo_get_stats = imq_get_stats, |
| 429 | +}; |
| 430 | + |
| 431 | +static void imq_setup(struct net_device *dev) |
| 432 | +{ |
| 433 | + dev->netdev_ops = &imq_netdev_ops; |
| 434 | + dev->type = ARPHRD_VOID; |
| 435 | + dev->mtu = 16000; |
| 436 | + dev->tx_queue_len = 11000; |
| 437 | + dev->flags = IFF_NOARP; |
| 438 | + dev->features = NETIF_F_SG | NETIF_F_FRAGLIST | |
| 439 | + NETIF_F_GSO | NETIF_F_HW_CSUM | |
| 440 | + NETIF_F_HIGHDMA; |
| 441 | + dev->priv_flags &= ~IFF_XMIT_DST_RELEASE; |
| 442 | +} |
| 443 | + |
| 444 | +static int imq_validate(struct nlattr *tb[], struct nlattr *data[]) |
| 445 | +{ |
| 446 | + int ret = 0; |
| 447 | + |
| 448 | + if (tb[IFLA_ADDRESS]) { |
| 449 | + if (nla_len(tb[IFLA_ADDRESS]) != ETH_ALEN) { |
| 450 | + ret = -EINVAL; |
| 451 | + goto end; |
| 452 | + } |
| 453 | + if (!is_valid_ether_addr(nla_data(tb[IFLA_ADDRESS]))) { |
| 454 | + ret = -EADDRNOTAVAIL; |
| 455 | + goto end; |
| 456 | + } |
| 457 | + } |
| 458 | + return 0; |
| 459 | +end: |
| 460 | + printk(KERN_WARNING "IMQ: imq_validate failed (%d)\n", ret); |
| 461 | + return ret; |
| 462 | +} |
| 463 | + |
| 464 | +static struct rtnl_link_ops imq_link_ops __read_mostly = { |
| 465 | + .kind = "imq", |
| 466 | + .priv_size = 0, |
| 467 | + .setup = imq_setup, |
| 468 | + .validate = imq_validate, |
| 469 | +}; |
| 470 | + |
| 471 | +static int __init imq_init_hooks(void) |
| 472 | +{ |
| 473 | + int err; |
| 474 | + |
| 475 | + nf_register_queue_imq_handler(&nfqh); |
| 476 | + |
| 477 | + err = nf_register_hook(&imq_ingress_ipv4); |
| 478 | + if (err) |
| 479 | + goto err1; |
| 480 | + |
| 481 | + err = nf_register_hook(&imq_egress_ipv4); |
| 482 | + if (err) |
| 483 | + goto err2; |
| 484 | + |
| 485 | +#if defined(CONFIG_IPV6) || defined(CONFIG_IPV6_MODULE) |
| 486 | + err = nf_register_hook(&imq_ingress_ipv6); |
| 487 | + if (err) |
| 488 | + goto err3; |
| 489 | + |
| 490 | + err = nf_register_hook(&imq_egress_ipv6); |
| 491 | + if (err) |
| 492 | + goto err4; |
| 493 | +#endif |
| 494 | + |
| 495 | + return 0; |
| 496 | + |
| 497 | +#if defined(CONFIG_IPV6) || defined(CONFIG_IPV6_MODULE) |
| 498 | +err4: |
| 499 | + nf_unregister_hook(&imq_ingress_ipv6); |
| 500 | +err3: |
| 501 | + nf_unregister_hook(&imq_egress_ipv4); |
| 502 | +#endif |
| 503 | +err2: |
| 504 | + nf_unregister_hook(&imq_ingress_ipv4); |
| 505 | +err1: |
| 506 | + nf_unregister_queue_imq_handler(); |
| 507 | + return err; |
| 508 | +} |
| 509 | + |
| 510 | +static int __init imq_init_one(int index) |
| 511 | +{ |
| 512 | + struct net_device *dev; |
| 513 | + int ret; |
| 514 | + |
| 515 | + dev = alloc_netdev(0, "imq%d", imq_setup); |
| 516 | + if (!dev) |
| 517 | + return -ENOMEM; |
| 518 | + |
| 519 | + ret = dev_alloc_name(dev, dev->name); |
| 520 | + if (ret < 0) |
| 521 | + goto fail; |
| 522 | + |
| 523 | + dev->rtnl_link_ops = &imq_link_ops; |
| 524 | + ret = register_netdevice(dev); |
| 525 | + if (ret < 0) |
| 526 | + goto fail; |
| 527 | + |
| 528 | + return 0; |
| 529 | +fail: |
| 530 | + free_netdev(dev); |
| 531 | + return ret; |
| 532 | +} |
| 533 | + |
| 534 | +static int __init imq_init_devs(void) |
| 535 | +{ |
| 536 | + int err, i; |
| 537 | + |
| 538 | + if (numdevs < 1 || numdevs > IMQ_MAX_DEVS) { |
| 539 | + printk(KERN_ERR "IMQ: numdevs has to be betweed 1 and %u\n", |
| 540 | + IMQ_MAX_DEVS); |
| 541 | + return -EINVAL; |
| 542 | + } |
| 543 | + |
| 544 | + rtnl_lock(); |
| 545 | + err = __rtnl_link_register(&imq_link_ops); |
| 546 | + |
| 547 | + for (i = 0; i < numdevs && !err; i++) |
| 548 | + err = imq_init_one(i); |
| 549 | + |
| 550 | + if (err) { |
| 551 | + __rtnl_link_unregister(&imq_link_ops); |
| 552 | + memset(imq_devs_cache, 0, sizeof(imq_devs_cache)); |
| 553 | + } |
| 554 | + rtnl_unlock(); |
| 555 | + |
| 556 | + return err; |
| 557 | +} |
| 558 | + |
| 559 | +static int __init imq_init_module(void) |
| 560 | +{ |
| 561 | + int err; |
| 562 | + |
| 563 | +#if defined(CONFIG_IMQ_NUM_DEVS) |
| 564 | + BUILD_BUG_ON(CONFIG_IMQ_NUM_DEVS > 16); |
| 565 | + BUILD_BUG_ON(CONFIG_IMQ_NUM_DEVS < 2); |
| 566 | + BUILD_BUG_ON(CONFIG_IMQ_NUM_DEVS - 1 > IMQ_F_IFMASK); |
| 567 | +#endif |
| 568 | + |
| 569 | + err = imq_init_devs(); |
| 570 | + if (err) { |
| 571 | + printk(KERN_ERR "IMQ: Error trying imq_init_devs(net)\n"); |
| 572 | + return err; |
| 573 | + } |
| 574 | + |
| 575 | + err = imq_init_hooks(); |
| 576 | + if (err) { |
| 577 | + printk(KERN_ERR "IMQ: Error trying imq_init_hooks()\n"); |
| 578 | + rtnl_link_unregister(&imq_link_ops); |
| 579 | + memset(imq_devs_cache, 0, sizeof(imq_devs_cache)); |
| 580 | + return err; |
| 581 | + } |
| 582 | + |
| 583 | + printk(KERN_INFO "IMQ driver loaded successfully.\n"); |
| 584 | + |
| 585 | +#if defined(CONFIG_IMQ_BEHAVIOR_BA) || defined(CONFIG_IMQ_BEHAVIOR_BB) |
| 586 | + printk(KERN_INFO "\tHooking IMQ before NAT on PREROUTING.\n"); |
| 587 | +#else |
| 588 | + printk(KERN_INFO "\tHooking IMQ after NAT on PREROUTING.\n"); |
| 589 | +#endif |
| 590 | +#if defined(CONFIG_IMQ_BEHAVIOR_AB) || defined(CONFIG_IMQ_BEHAVIOR_BB) |
| 591 | + printk(KERN_INFO "\tHooking IMQ before NAT on POSTROUTING.\n"); |
| 592 | +#else |
| 593 | + printk(KERN_INFO "\tHooking IMQ after NAT on POSTROUTING.\n"); |
| 594 | +#endif |
| 595 | + |
| 596 | + return 0; |
| 597 | +} |
| 598 | + |
| 599 | +static void __exit imq_unhook(void) |
| 600 | +{ |
| 601 | +#if defined(CONFIG_IPV6) || defined(CONFIG_IPV6_MODULE) |
| 602 | + nf_unregister_hook(&imq_ingress_ipv6); |
| 603 | + nf_unregister_hook(&imq_egress_ipv6); |
| 604 | +#endif |
| 605 | + nf_unregister_hook(&imq_ingress_ipv4); |
| 606 | + nf_unregister_hook(&imq_egress_ipv4); |
| 607 | + |
| 608 | + nf_unregister_queue_imq_handler(); |
| 609 | +} |
| 610 | + |
| 611 | +static void __exit imq_cleanup_devs(void) |
| 612 | +{ |
| 613 | + rtnl_link_unregister(&imq_link_ops); |
| 614 | + memset(imq_devs_cache, 0, sizeof(imq_devs_cache)); |
| 615 | +} |
| 616 | + |
| 617 | +static void __exit imq_exit_module(void) |
| 618 | +{ |
| 619 | + imq_unhook(); |
| 620 | + imq_cleanup_devs(); |
| 621 | + printk(KERN_INFO "IMQ driver unloaded successfully.\n"); |
| 622 | +} |
| 623 | + |
| 624 | +module_init(imq_init_module); |
| 625 | +module_exit(imq_exit_module); |
| 626 | + |
| 627 | +module_param(numdevs, int, 0); |
| 628 | +MODULE_PARM_DESC(numdevs, "number of IMQ devices (how many imq* devices will " |
| 629 | + "be created)"); |
| 630 | +MODULE_AUTHOR("http://www.linuximq.net"); |
| 631 | +MODULE_DESCRIPTION("Pseudo-driver for the intermediate queue device. See " |
| 632 | + "http://www.linuximq.net/ for more information."); |
| 633 | +MODULE_LICENSE("GPL"); |
| 634 | +MODULE_ALIAS_RTNL_LINK("imq"); |
| 635 | + |
| 636 | --- a/drivers/net/Kconfig |
| 637 | +++ b/drivers/net/Kconfig |
| 638 | @@ -109,6 +109,129 @@ config EQUALIZER |
| 639 | To compile this driver as a module, choose M here: the module |
| 640 | will be called eql. If unsure, say N. |
| 641 | |
| 642 | +config IMQ |
| 643 | + tristate "IMQ (intermediate queueing device) support" |
| 644 | + depends on NETDEVICES && NETFILTER |
| 645 | + ---help--- |
| 646 | + The IMQ device(s) is used as placeholder for QoS queueing |
| 647 | + disciplines. Every packet entering/leaving the IP stack can be |
| 648 | + directed through the IMQ device where it's enqueued/dequeued to the |
| 649 | + attached qdisc. This allows you to treat network devices as classes |
| 650 | + and distribute bandwidth among them. Iptables is used to specify |
| 651 | + through which IMQ device, if any, packets travel. |
| 652 | + |
| 653 | + More information at: http://www.linuximq.net/ |
| 654 | + |
| 655 | + To compile this driver as a module, choose M here: the module |
| 656 | + will be called imq. If unsure, say N. |
| 657 | + |
| 658 | +choice |
| 659 | + prompt "IMQ behavior (PRE/POSTROUTING)" |
| 660 | + depends on IMQ |
| 661 | + default IMQ_BEHAVIOR_AB |
| 662 | + help |
| 663 | + |
| 664 | + This settings defines how IMQ behaves in respect to its |
| 665 | + hooking in PREROUTING and POSTROUTING. |
| 666 | + |
| 667 | + IMQ can work in any of the following ways: |
| 668 | + |
| 669 | + PREROUTING | POSTROUTING |
| 670 | + -----------------|------------------- |
| 671 | + #1 After NAT | After NAT |
| 672 | + #2 After NAT | Before NAT |
| 673 | + #3 Before NAT | After NAT |
| 674 | + #4 Before NAT | Before NAT |
| 675 | + |
| 676 | + The default behavior is to hook before NAT on PREROUTING |
| 677 | + and after NAT on POSTROUTING (#3). |
| 678 | + |
| 679 | + This settings are specially usefull when trying to use IMQ |
| 680 | + to shape NATed clients. |
| 681 | + |
| 682 | + More information can be found at: www.linuximq.net |
| 683 | + |
| 684 | + If not sure leave the default settings alone. |
| 685 | + |
| 686 | +config IMQ_BEHAVIOR_AA |
| 687 | + bool "IMQ AA" |
| 688 | + help |
| 689 | + This settings defines how IMQ behaves in respect to its |
| 690 | + hooking in PREROUTING and POSTROUTING. |
| 691 | + |
| 692 | + Choosing this option will make IMQ hook like this: |
| 693 | + |
| 694 | + PREROUTING: After NAT |
| 695 | + POSTROUTING: After NAT |
| 696 | + |
| 697 | + More information can be found at: www.linuximq.net |
| 698 | + |
| 699 | + If not sure leave the default settings alone. |
| 700 | + |
| 701 | +config IMQ_BEHAVIOR_AB |
| 702 | + bool "IMQ AB" |
| 703 | + help |
| 704 | + This settings defines how IMQ behaves in respect to its |
| 705 | + hooking in PREROUTING and POSTROUTING. |
| 706 | + |
| 707 | + Choosing this option will make IMQ hook like this: |
| 708 | + |
| 709 | + PREROUTING: After NAT |
| 710 | + POSTROUTING: Before NAT |
| 711 | + |
| 712 | + More information can be found at: www.linuximq.net |
| 713 | + |
| 714 | + If not sure leave the default settings alone. |
| 715 | + |
| 716 | +config IMQ_BEHAVIOR_BA |
| 717 | + bool "IMQ BA" |
| 718 | + help |
| 719 | + This settings defines how IMQ behaves in respect to its |
| 720 | + hooking in PREROUTING and POSTROUTING. |
| 721 | + |
| 722 | + Choosing this option will make IMQ hook like this: |
| 723 | + |
| 724 | + PREROUTING: Before NAT |
| 725 | + POSTROUTING: After NAT |
| 726 | + |
| 727 | + More information can be found at: www.linuximq.net |
| 728 | + |
| 729 | + If not sure leave the default settings alone. |
| 730 | + |
| 731 | +config IMQ_BEHAVIOR_BB |
| 732 | + bool "IMQ BB" |
| 733 | + help |
| 734 | + This settings defines how IMQ behaves in respect to its |
| 735 | + hooking in PREROUTING and POSTROUTING. |
| 736 | + |
| 737 | + Choosing this option will make IMQ hook like this: |
| 738 | + |
| 739 | + PREROUTING: Before NAT |
| 740 | + POSTROUTING: Before NAT |
| 741 | + |
| 742 | + More information can be found at: www.linuximq.net |
| 743 | + |
| 744 | + If not sure leave the default settings alone. |
| 745 | + |
| 746 | +endchoice |
| 747 | + |
| 748 | +config IMQ_NUM_DEVS |
| 749 | + |
| 750 | + int "Number of IMQ devices" |
| 751 | + range 2 16 |
| 752 | + depends on IMQ |
| 753 | + default "16" |
| 754 | + help |
| 755 | + |
| 756 | + This settings defines how many IMQ devices will be |
| 757 | + created. |
| 758 | + |
| 759 | + The default value is 16. |
| 760 | + |
| 761 | + More information can be found at: www.linuximq.net |
| 762 | + |
| 763 | + If not sure leave the default settings alone. |
| 764 | + |
| 765 | config TUN |
| 766 | tristate "Universal TUN/TAP device driver support" |
| 767 | select CRC32 |
| 768 | --- a/drivers/net/Makefile |
| 769 | +++ b/drivers/net/Makefile |
| 770 | @@ -165,6 +165,7 @@ obj-$(CONFIG_SLHC) += slhc.o |
| 771 | obj-$(CONFIG_XEN_NETDEV_FRONTEND) += xen-netfront.o |
| 772 | |
| 773 | obj-$(CONFIG_DUMMY) += dummy.o |
| 774 | +obj-$(CONFIG_IMQ) += imq.o |
| 775 | obj-$(CONFIG_IFB) += ifb.o |
| 776 | obj-$(CONFIG_MACVLAN) += macvlan.o |
| 777 | obj-$(CONFIG_DE600) += de600.o |
| 778 | --- /dev/null |
| 779 | +++ b/include/linux/imq.h |
| 780 | @@ -0,0 +1,13 @@ |
| 781 | +#ifndef _IMQ_H |
| 782 | +#define _IMQ_H |
| 783 | + |
| 784 | +/* IFMASK (16 device indexes, 0 to 15) and flag(s) fit in 5 bits */ |
| 785 | +#define IMQ_F_BITS 5 |
| 786 | + |
| 787 | +#define IMQ_F_IFMASK 0x0f |
| 788 | +#define IMQ_F_ENQUEUE 0x10 |
| 789 | + |
| 790 | +#define IMQ_MAX_DEVS (IMQ_F_IFMASK + 1) |
| 791 | + |
| 792 | +#endif /* _IMQ_H */ |
| 793 | + |
| 794 | --- a/include/linux/netdevice.h |
| 795 | +++ b/include/linux/netdevice.h |
| 796 | @@ -1114,6 +1114,7 @@ extern int dev_alloc_name(struct net_de |
| 797 | extern int dev_open(struct net_device *dev); |
| 798 | extern int dev_close(struct net_device *dev); |
| 799 | extern void dev_disable_lro(struct net_device *dev); |
| 800 | +extern struct netdev_queue *dev_pick_tx(struct net_device *dev, struct sk_buff *skb); |
| 801 | extern int dev_queue_xmit(struct sk_buff *skb); |
| 802 | extern int register_netdevice(struct net_device *dev); |
| 803 | extern void unregister_netdevice(struct net_device *dev); |
| 804 | --- /dev/null |
| 805 | +++ b/include/linux/netfilter/xt_IMQ.h |
| 806 | @@ -0,0 +1,9 @@ |
| 807 | +#ifndef _XT_IMQ_H |
| 808 | +#define _XT_IMQ_H |
| 809 | + |
| 810 | +struct xt_imq_info { |
| 811 | + unsigned int todev; /* target imq device */ |
| 812 | +}; |
| 813 | + |
| 814 | +#endif /* _XT_IMQ_H */ |
| 815 | + |
| 816 | --- /dev/null |
| 817 | +++ b/include/linux/netfilter_ipv4/ipt_IMQ.h |
| 818 | @@ -0,0 +1,10 @@ |
| 819 | +#ifndef _IPT_IMQ_H |
| 820 | +#define _IPT_IMQ_H |
| 821 | + |
| 822 | +/* Backwards compatibility for old userspace */ |
| 823 | +#include <linux/netfilter/xt_IMQ.h> |
| 824 | + |
| 825 | +#define ipt_imq_info xt_imq_info |
| 826 | + |
| 827 | +#endif /* _IPT_IMQ_H */ |
| 828 | + |
| 829 | --- /dev/null |
| 830 | +++ b/include/linux/netfilter_ipv6/ip6t_IMQ.h |
| 831 | @@ -0,0 +1,10 @@ |
| 832 | +#ifndef _IP6T_IMQ_H |
| 833 | +#define _IP6T_IMQ_H |
| 834 | + |
| 835 | +/* Backwards compatibility for old userspace */ |
| 836 | +#include <linux/netfilter/xt_IMQ.h> |
| 837 | + |
| 838 | +#define ip6t_imq_info xt_imq_info |
| 839 | + |
| 840 | +#endif /* _IP6T_IMQ_H */ |
| 841 | + |
| 842 | --- a/include/linux/skbuff.h |
| 843 | +++ b/include/linux/skbuff.h |
| 844 | @@ -29,6 +29,9 @@ |
| 845 | #include <linux/rcupdate.h> |
| 846 | #include <linux/dmaengine.h> |
| 847 | #include <linux/hrtimer.h> |
| 848 | +#if defined(CONFIG_IMQ) || defined(CONFIG_IMQ_MODULE) |
| 849 | +#include <linux/imq.h> |
| 850 | +#endif |
| 851 | |
| 852 | /* Don't change this without changing skb_csum_unnecessary! */ |
| 853 | #define CHECKSUM_NONE 0 |
| 854 | @@ -330,6 +333,9 @@ struct sk_buff { |
| 855 | * first. This is owned by whoever has the skb queued ATM. |
| 856 | */ |
| 857 | char cb[48]; |
| 858 | +#if defined(CONFIG_IMQ) || defined(CONFIG_IMQ_MODULE) |
| 859 | + void *cb_next; |
| 860 | +#endif |
| 861 | |
| 862 | unsigned int len, |
| 863 | data_len; |
| 864 | @@ -362,6 +368,9 @@ struct sk_buff { |
| 865 | struct nf_conntrack *nfct; |
| 866 | struct sk_buff *nfct_reasm; |
| 867 | #endif |
| 868 | +#if defined(CONFIG_IMQ) || defined(CONFIG_IMQ_MODULE) |
| 869 | + struct nf_queue_entry *nf_queue_entry; |
| 870 | +#endif |
| 871 | #ifdef CONFIG_BRIDGE_NETFILTER |
| 872 | struct nf_bridge_info *nf_bridge; |
| 873 | #endif |
| 874 | @@ -383,6 +392,10 @@ struct sk_buff { |
| 875 | |
| 876 | /* 0/14 bit hole */ |
| 877 | |
| 878 | +#if defined(CONFIG_IMQ) || defined(CONFIG_IMQ_MODULE) |
| 879 | + __u8 imq_flags:IMQ_F_BITS; |
| 880 | +#endif |
| 881 | + |
| 882 | #ifdef CONFIG_NET_DMA |
| 883 | dma_cookie_t dma_cookie; |
| 884 | #endif |
| 885 | @@ -437,6 +450,12 @@ static inline struct rtable *skb_rtable( |
| 886 | return (struct rtable *)skb_dst(skb); |
| 887 | } |
| 888 | |
| 889 | + |
| 890 | +#if defined(CONFIG_IMQ) || defined(CONFIG_IMQ_MODULE) |
| 891 | +extern int skb_save_cb(struct sk_buff *skb); |
| 892 | +extern int skb_restore_cb(struct sk_buff *skb); |
| 893 | +#endif |
| 894 | + |
| 895 | extern void kfree_skb(struct sk_buff *skb); |
| 896 | extern void consume_skb(struct sk_buff *skb); |
| 897 | extern void __kfree_skb(struct sk_buff *skb); |
| 898 | @@ -1972,6 +1991,10 @@ static inline void __nf_copy(struct sk_b |
| 899 | dst->nfct_reasm = src->nfct_reasm; |
| 900 | nf_conntrack_get_reasm(src->nfct_reasm); |
| 901 | #endif |
| 902 | +#if defined(CONFIG_IMQ) || defined(CONFIG_IMQ_MODULE) |
| 903 | + dst->imq_flags = src->imq_flags; |
| 904 | + dst->nf_queue_entry = src->nf_queue_entry; |
| 905 | +#endif |
| 906 | #ifdef CONFIG_BRIDGE_NETFILTER |
| 907 | dst->nf_bridge = src->nf_bridge; |
| 908 | nf_bridge_get(src->nf_bridge); |
| 909 | --- a/include/net/netfilter/nf_queue.h |
| 910 | +++ b/include/net/netfilter/nf_queue.h |
| 911 | @@ -13,6 +13,12 @@ struct nf_queue_entry { |
| 912 | struct net_device *indev; |
| 913 | struct net_device *outdev; |
| 914 | int (*okfn)(struct sk_buff *); |
| 915 | + |
| 916 | +#if defined(CONFIG_IMQ) || defined(CONFIG_IMQ_MODULE) |
| 917 | + int (*next_outfn)(struct nf_queue_entry *entry, |
| 918 | + unsigned int queuenum); |
| 919 | + unsigned int next_queuenum; |
| 920 | +#endif |
| 921 | }; |
| 922 | |
| 923 | #define nf_queue_entry_reroute(x) ((void *)x + sizeof(struct nf_queue_entry)) |
| 924 | @@ -30,5 +36,11 @@ extern int nf_unregister_queue_handler(u |
| 925 | const struct nf_queue_handler *qh); |
| 926 | extern void nf_unregister_queue_handlers(const struct nf_queue_handler *qh); |
| 927 | extern void nf_reinject(struct nf_queue_entry *entry, unsigned int verdict); |
| 928 | +extern void nf_queue_entry_release_refs(struct nf_queue_entry *entry); |
| 929 | + |
| 930 | +#if defined(CONFIG_IMQ) || defined(CONFIG_IMQ_MODULE) |
| 931 | +extern void nf_register_queue_imq_handler(const struct nf_queue_handler *qh); |
| 932 | +extern void nf_unregister_queue_imq_handler(void); |
| 933 | +#endif |
| 934 | |
| 935 | #endif /* _NF_QUEUE_H */ |
| 936 | --- a/net/core/dev.c |
| 937 | +++ b/net/core/dev.c |
| 938 | @@ -96,6 +96,9 @@ |
| 939 | #include <net/net_namespace.h> |
| 940 | #include <net/sock.h> |
| 941 | #include <linux/rtnetlink.h> |
| 942 | +#if defined(CONFIG_IMQ) || defined(CONFIG_IMQ_MODULE) |
| 943 | +#include <linux/imq.h> |
| 944 | +#endif |
| 945 | #include <linux/proc_fs.h> |
| 946 | #include <linux/seq_file.h> |
| 947 | #include <linux/stat.h> |
| 948 | @@ -1704,7 +1707,11 @@ int dev_hard_start_xmit(struct sk_buff * |
| 949 | int rc; |
| 950 | |
| 951 | if (likely(!skb->next)) { |
| 952 | - if (!list_empty(&ptype_all)) |
| 953 | + if (!list_empty(&ptype_all) |
| 954 | +#if defined(CONFIG_IMQ) || defined(CONFIG_IMQ_MODULE) |
| 955 | + && !(skb->imq_flags & IMQ_F_ENQUEUE) |
| 956 | +#endif |
| 957 | + ) |
| 958 | dev_queue_xmit_nit(skb, dev); |
| 959 | |
| 960 | if (netif_needs_gso(dev, skb)) { |
| 961 | @@ -1789,8 +1796,7 @@ u16 skb_tx_hash(const struct net_device |
| 962 | } |
| 963 | EXPORT_SYMBOL(skb_tx_hash); |
| 964 | |
| 965 | -static struct netdev_queue *dev_pick_tx(struct net_device *dev, |
| 966 | - struct sk_buff *skb) |
| 967 | +struct netdev_queue *dev_pick_tx(struct net_device *dev, struct sk_buff *skb) |
| 968 | { |
| 969 | const struct net_device_ops *ops = dev->netdev_ops; |
| 970 | u16 queue_index = 0; |
| 971 | @@ -1803,6 +1809,7 @@ static struct netdev_queue *dev_pick_tx( |
| 972 | skb_set_queue_mapping(skb, queue_index); |
| 973 | return netdev_get_tx_queue(dev, queue_index); |
| 974 | } |
| 975 | +EXPORT_SYMBOL(dev_pick_tx); |
| 976 | |
| 977 | static inline int __dev_xmit_skb(struct sk_buff *skb, struct Qdisc *q, |
| 978 | struct net_device *dev, |
| 979 | --- a/net/core/skbuff.c |
| 980 | +++ b/net/core/skbuff.c |
| 981 | @@ -72,6 +72,9 @@ |
| 982 | |
| 983 | static struct kmem_cache *skbuff_head_cache __read_mostly; |
| 984 | static struct kmem_cache *skbuff_fclone_cache __read_mostly; |
| 985 | +#if defined(CONFIG_IMQ) || defined(CONFIG_IMQ_MODULE) |
| 986 | +static struct kmem_cache *skbuff_cb_store_cache __read_mostly; |
| 987 | +#endif |
| 988 | |
| 989 | static void sock_pipe_buf_release(struct pipe_inode_info *pipe, |
| 990 | struct pipe_buffer *buf) |
| 991 | @@ -91,6 +94,83 @@ static int sock_pipe_buf_steal(struct pi |
| 992 | return 1; |
| 993 | } |
| 994 | |
| 995 | +#if defined(CONFIG_IMQ) || defined(CONFIG_IMQ_MODULE) |
| 996 | +/* Control buffer save/restore for IMQ devices */ |
| 997 | +struct skb_cb_table { |
| 998 | + void *cb_next; |
| 999 | + atomic_t refcnt; |
| 1000 | + char cb[48]; |
| 1001 | +}; |
| 1002 | + |
| 1003 | +static DEFINE_SPINLOCK(skb_cb_store_lock); |
| 1004 | + |
| 1005 | +int skb_save_cb(struct sk_buff *skb) |
| 1006 | +{ |
| 1007 | + struct skb_cb_table *next; |
| 1008 | + |
| 1009 | + next = kmem_cache_alloc(skbuff_cb_store_cache, GFP_ATOMIC); |
| 1010 | + if (!next) |
| 1011 | + return -ENOMEM; |
| 1012 | + |
| 1013 | + BUILD_BUG_ON(sizeof(skb->cb) != sizeof(next->cb)); |
| 1014 | + |
| 1015 | + memcpy(next->cb, skb->cb, sizeof(skb->cb)); |
| 1016 | + next->cb_next = skb->cb_next; |
| 1017 | + |
| 1018 | + atomic_set(&next->refcnt, 1); |
| 1019 | + |
| 1020 | + skb->cb_next = next; |
| 1021 | + return 0; |
| 1022 | +} |
| 1023 | +EXPORT_SYMBOL(skb_save_cb); |
| 1024 | + |
| 1025 | +int skb_restore_cb(struct sk_buff *skb) |
| 1026 | +{ |
| 1027 | + struct skb_cb_table *next; |
| 1028 | + |
| 1029 | + if (!skb->cb_next) |
| 1030 | + return 0; |
| 1031 | + |
| 1032 | + next = skb->cb_next; |
| 1033 | + |
| 1034 | + BUILD_BUG_ON(sizeof(skb->cb) != sizeof(next->cb)); |
| 1035 | + |
| 1036 | + memcpy(skb->cb, next->cb, sizeof(skb->cb)); |
| 1037 | + skb->cb_next = next->cb_next; |
| 1038 | + |
| 1039 | + spin_lock(&skb_cb_store_lock); |
| 1040 | + |
| 1041 | + if (atomic_dec_and_test(&next->refcnt)) { |
| 1042 | + kmem_cache_free(skbuff_cb_store_cache, next); |
| 1043 | + } |
| 1044 | + |
| 1045 | + spin_unlock(&skb_cb_store_lock); |
| 1046 | + |
| 1047 | + return 0; |
| 1048 | +} |
| 1049 | +EXPORT_SYMBOL(skb_restore_cb); |
| 1050 | + |
| 1051 | +static void skb_copy_stored_cb(struct sk_buff *new, const struct sk_buff *__old) |
| 1052 | +{ |
| 1053 | + struct skb_cb_table *next; |
| 1054 | + struct sk_buff *old; |
| 1055 | + |
| 1056 | + if (!__old->cb_next) { |
| 1057 | + new->cb_next = NULL; |
| 1058 | + return; |
| 1059 | + } |
| 1060 | + |
| 1061 | + spin_lock(&skb_cb_store_lock); |
| 1062 | + |
| 1063 | + old = (struct sk_buff *)__old; |
| 1064 | + |
| 1065 | + next = old->cb_next; |
| 1066 | + atomic_inc(&next->refcnt); |
| 1067 | + new->cb_next = next; |
| 1068 | + |
| 1069 | + spin_unlock(&skb_cb_store_lock); |
| 1070 | +} |
| 1071 | +#endif |
| 1072 | |
| 1073 | /* Pipe buffer operations for a socket. */ |
| 1074 | static struct pipe_buf_operations sock_pipe_buf_ops = { |
| 1075 | @@ -398,6 +478,26 @@ static void skb_release_head_state(struc |
| 1076 | WARN_ON(in_irq()); |
| 1077 | skb->destructor(skb); |
| 1078 | } |
| 1079 | +#if defined(CONFIG_IMQ) || defined(CONFIG_IMQ_MODULE) |
| 1080 | + /* This should not happen. When it does, avoid memleak by restoring |
| 1081 | + the chain of cb-backups. */ |
| 1082 | + while(skb->cb_next != NULL) { |
| 1083 | + if (net_ratelimit()) |
| 1084 | + printk(KERN_WARNING "IMQ: kfree_skb: skb->cb_next: " |
| 1085 | + "%08x\n", (unsigned int)skb->cb_next); |
| 1086 | + |
| 1087 | + skb_restore_cb(skb); |
| 1088 | + } |
| 1089 | + /* This should not happen either, nf_queue_entry is nullified in |
| 1090 | + * imq_dev_xmit(). If we have non-NULL nf_queue_entry then we are |
| 1091 | + * leaking entry pointers, maybe memory. We don't know if this is |
| 1092 | + * pointer to already freed memory, or should this be freed. |
| 1093 | + * If this happens we need to add refcounting, etc for nf_queue_entry. |
| 1094 | + */ |
| 1095 | + if (skb->nf_queue_entry && net_ratelimit()) |
| 1096 | + printk(KERN_WARNING |
| 1097 | + "IMQ: kfree_skb: skb->nf_queue_entry != NULL"); |
| 1098 | +#endif |
| 1099 | #if defined(CONFIG_NF_CONNTRACK) || defined(CONFIG_NF_CONNTRACK_MODULE) |
| 1100 | nf_conntrack_put(skb->nfct); |
| 1101 | nf_conntrack_put_reasm(skb->nfct_reasm); |
| 1102 | @@ -535,6 +635,9 @@ static void __copy_skb_header(struct sk_ |
| 1103 | new->sp = secpath_get(old->sp); |
| 1104 | #endif |
| 1105 | memcpy(new->cb, old->cb, sizeof(old->cb)); |
| 1106 | +#if defined(CONFIG_IMQ) || defined(CONFIG_IMQ_MODULE) |
| 1107 | + skb_copy_stored_cb(new, old); |
| 1108 | +#endif |
| 1109 | new->csum = old->csum; |
| 1110 | new->local_df = old->local_df; |
| 1111 | new->pkt_type = old->pkt_type; |
| 1112 | @@ -2781,6 +2884,13 @@ void __init skb_init(void) |
| 1113 | 0, |
| 1114 | SLAB_HWCACHE_ALIGN|SLAB_PANIC, |
| 1115 | NULL); |
| 1116 | +#if defined(CONFIG_IMQ) || defined(CONFIG_IMQ_MODULE) |
| 1117 | + skbuff_cb_store_cache = kmem_cache_create("skbuff_cb_store_cache", |
| 1118 | + sizeof(struct skb_cb_table), |
| 1119 | + 0, |
| 1120 | + SLAB_HWCACHE_ALIGN|SLAB_PANIC, |
| 1121 | + NULL); |
| 1122 | +#endif |
| 1123 | } |
| 1124 | |
| 1125 | /** |
| 1126 | --- a/net/netfilter/Kconfig |
| 1127 | +++ b/net/netfilter/Kconfig |
| 1128 | @@ -396,6 +396,18 @@ config NETFILTER_XT_TARGET_LED |
| 1129 | For more information on the LEDs available on your system, see |
| 1130 | Documentation/leds-class.txt |
| 1131 | |
| 1132 | +config NETFILTER_XT_TARGET_IMQ |
| 1133 | + tristate '"IMQ" target support' |
| 1134 | + depends on NETFILTER_XTABLES |
| 1135 | + depends on IP_NF_MANGLE || IP6_NF_MANGLE |
| 1136 | + select IMQ |
| 1137 | + default m if NETFILTER_ADVANCED=n |
| 1138 | + help |
| 1139 | + This option adds a `IMQ' target which is used to specify if and |
| 1140 | + to which imq device packets should get enqueued/dequeued. |
| 1141 | + |
| 1142 | + To compile it as a module, choose M here. If unsure, say N. |
| 1143 | + |
| 1144 | config NETFILTER_XT_TARGET_MARK |
| 1145 | tristate '"MARK" target support' |
| 1146 | default m if NETFILTER_ADVANCED=n |
| 1147 | --- a/net/netfilter/Makefile |
| 1148 | +++ b/net/netfilter/Makefile |
| 1149 | @@ -46,6 +46,7 @@ obj-$(CONFIG_NETFILTER_XT_TARGET_CONNMAR |
| 1150 | obj-$(CONFIG_NETFILTER_XT_TARGET_CONNSECMARK) += xt_CONNSECMARK.o |
| 1151 | obj-$(CONFIG_NETFILTER_XT_TARGET_DSCP) += xt_DSCP.o |
| 1152 | obj-$(CONFIG_NETFILTER_XT_TARGET_HL) += xt_HL.o |
| 1153 | +obj-$(CONFIG_NETFILTER_XT_TARGET_IMQ) += xt_IMQ.o |
| 1154 | obj-$(CONFIG_NETFILTER_XT_TARGET_LED) += xt_LED.o |
| 1155 | obj-$(CONFIG_NETFILTER_XT_TARGET_MARK) += xt_MARK.o |
| 1156 | obj-$(CONFIG_NETFILTER_XT_TARGET_NFLOG) += xt_NFLOG.o |
| 1157 | --- a/net/netfilter/nf_queue.c |
| 1158 | +++ b/net/netfilter/nf_queue.c |
| 1159 | @@ -20,6 +20,26 @@ static const struct nf_queue_handler *qu |
| 1160 | |
| 1161 | static DEFINE_MUTEX(queue_handler_mutex); |
| 1162 | |
| 1163 | +#if defined(CONFIG_IMQ) || defined(CONFIG_IMQ_MODULE) |
| 1164 | +static const struct nf_queue_handler *queue_imq_handler; |
| 1165 | + |
| 1166 | +void nf_register_queue_imq_handler(const struct nf_queue_handler *qh) |
| 1167 | +{ |
| 1168 | + mutex_lock(&queue_handler_mutex); |
| 1169 | + rcu_assign_pointer(queue_imq_handler, qh); |
| 1170 | + mutex_unlock(&queue_handler_mutex); |
| 1171 | +} |
| 1172 | +EXPORT_SYMBOL(nf_register_queue_imq_handler); |
| 1173 | + |
| 1174 | +void nf_unregister_queue_imq_handler(void) |
| 1175 | +{ |
| 1176 | + mutex_lock(&queue_handler_mutex); |
| 1177 | + rcu_assign_pointer(queue_imq_handler, NULL); |
| 1178 | + mutex_unlock(&queue_handler_mutex); |
| 1179 | +} |
| 1180 | +EXPORT_SYMBOL(nf_unregister_queue_imq_handler); |
| 1181 | +#endif |
| 1182 | + |
| 1183 | /* return EBUSY when somebody else is registered, return EEXIST if the |
| 1184 | * same handler is registered, return 0 in case of success. */ |
| 1185 | int nf_register_queue_handler(u_int8_t pf, const struct nf_queue_handler *qh) |
| 1186 | @@ -80,7 +100,7 @@ void nf_unregister_queue_handlers(const |
| 1187 | } |
| 1188 | EXPORT_SYMBOL_GPL(nf_unregister_queue_handlers); |
| 1189 | |
| 1190 | -static void nf_queue_entry_release_refs(struct nf_queue_entry *entry) |
| 1191 | +void nf_queue_entry_release_refs(struct nf_queue_entry *entry) |
| 1192 | { |
| 1193 | /* Release those devices we held, or Alexey will kill me. */ |
| 1194 | if (entry->indev) |
| 1195 | @@ -100,6 +120,7 @@ static void nf_queue_entry_release_refs( |
| 1196 | /* Drop reference to owner of hook which queued us. */ |
| 1197 | module_put(entry->elem->owner); |
| 1198 | } |
| 1199 | +EXPORT_SYMBOL_GPL(nf_queue_entry_release_refs); |
| 1200 | |
| 1201 | /* |
| 1202 | * Any packet that leaves via this function must come back |
| 1203 | @@ -121,12 +142,26 @@ static int __nf_queue(struct sk_buff *sk |
| 1204 | #endif |
| 1205 | const struct nf_afinfo *afinfo; |
| 1206 | const struct nf_queue_handler *qh; |
| 1207 | +#if defined(CONFIG_IMQ) || defined(CONFIG_IMQ_MODULE) |
| 1208 | + const struct nf_queue_handler *qih = NULL; |
| 1209 | +#endif |
| 1210 | |
| 1211 | /* QUEUE == DROP if noone is waiting, to be safe. */ |
| 1212 | rcu_read_lock(); |
| 1213 | |
| 1214 | qh = rcu_dereference(queue_handler[pf]); |
| 1215 | +#if defined(CONFIG_IMQ) || defined(CONFIG_IMQ_MODULE) |
| 1216 | +#if defined(CONFIG_IPV6) || defined(CONFIG_IPV6_MODULE) |
| 1217 | + if (pf == PF_INET || pf == PF_INET6) |
| 1218 | +#else |
| 1219 | + if (pf == PF_INET) |
| 1220 | +#endif |
| 1221 | + qih = rcu_dereference(queue_imq_handler); |
| 1222 | + |
| 1223 | + if (!qh && !qih) |
| 1224 | +#else /* !IMQ */ |
| 1225 | if (!qh) |
| 1226 | +#endif |
| 1227 | goto err_unlock; |
| 1228 | |
| 1229 | afinfo = nf_get_afinfo(pf); |
| 1230 | @@ -145,6 +180,10 @@ static int __nf_queue(struct sk_buff *sk |
| 1231 | .indev = indev, |
| 1232 | .outdev = outdev, |
| 1233 | .okfn = okfn, |
| 1234 | +#if defined(CONFIG_IMQ) || defined(CONFIG_IMQ_MODULE) |
| 1235 | + .next_outfn = qh ? qh->outfn : NULL, |
| 1236 | + .next_queuenum = queuenum, |
| 1237 | +#endif |
| 1238 | }; |
| 1239 | |
| 1240 | /* If it's going away, ignore hook. */ |
| 1241 | @@ -170,8 +209,19 @@ static int __nf_queue(struct sk_buff *sk |
| 1242 | } |
| 1243 | #endif |
| 1244 | afinfo->saveroute(skb, entry); |
| 1245 | + |
| 1246 | +#if defined(CONFIG_IMQ) || defined(CONFIG_IMQ_MODULE) |
| 1247 | + if (qih) { |
| 1248 | + status = qih->outfn(entry, queuenum); |
| 1249 | + goto imq_skip_queue; |
| 1250 | + } |
| 1251 | +#endif |
| 1252 | + |
| 1253 | status = qh->outfn(entry, queuenum); |
| 1254 | |
| 1255 | +#if defined(CONFIG_IMQ) || defined(CONFIG_IMQ_MODULE) |
| 1256 | +imq_skip_queue: |
| 1257 | +#endif |
| 1258 | rcu_read_unlock(); |
| 1259 | |
| 1260 | if (status < 0) { |
| 1261 | --- /dev/null |
| 1262 | +++ b/net/netfilter/xt_IMQ.c |
| 1263 | @@ -0,0 +1,73 @@ |
| 1264 | +/* |
| 1265 | + * This target marks packets to be enqueued to an imq device |
| 1266 | + */ |
| 1267 | +#include <linux/module.h> |
| 1268 | +#include <linux/skbuff.h> |
| 1269 | +#include <linux/netfilter/x_tables.h> |
| 1270 | +#include <linux/netfilter/xt_IMQ.h> |
| 1271 | +#include <linux/imq.h> |
| 1272 | + |
| 1273 | +static unsigned int imq_target(struct sk_buff *pskb, |
| 1274 | + const struct xt_target_param *par) |
| 1275 | +{ |
| 1276 | + const struct xt_imq_info *mr = par->targinfo; |
| 1277 | + |
| 1278 | + pskb->imq_flags = (mr->todev & IMQ_F_IFMASK) | IMQ_F_ENQUEUE; |
| 1279 | + |
| 1280 | + return XT_CONTINUE; |
| 1281 | +} |
| 1282 | + |
| 1283 | +static bool imq_checkentry(const struct xt_tgchk_param *par) |
| 1284 | +{ |
| 1285 | + struct xt_imq_info *mr = par->targinfo; |
| 1286 | + |
| 1287 | + if (mr->todev > IMQ_MAX_DEVS - 1) { |
| 1288 | + printk(KERN_WARNING |
| 1289 | + "IMQ: invalid device specified, highest is %u\n", |
| 1290 | + IMQ_MAX_DEVS - 1); |
| 1291 | + return 0; |
| 1292 | + } |
| 1293 | + |
| 1294 | + return 1; |
| 1295 | +} |
| 1296 | + |
| 1297 | +static struct xt_target xt_imq_reg[] __read_mostly = { |
| 1298 | + { |
| 1299 | + .name = "IMQ", |
| 1300 | + .family = AF_INET, |
| 1301 | + .checkentry = imq_checkentry, |
| 1302 | + .target = imq_target, |
| 1303 | + .targetsize = sizeof(struct xt_imq_info), |
| 1304 | + .table = "mangle", |
| 1305 | + .me = THIS_MODULE |
| 1306 | + }, |
| 1307 | + { |
| 1308 | + .name = "IMQ", |
| 1309 | + .family = AF_INET6, |
| 1310 | + .checkentry = imq_checkentry, |
| 1311 | + .target = imq_target, |
| 1312 | + .targetsize = sizeof(struct xt_imq_info), |
| 1313 | + .table = "mangle", |
| 1314 | + .me = THIS_MODULE |
| 1315 | + }, |
| 1316 | +}; |
| 1317 | + |
| 1318 | +static int __init imq_init(void) |
| 1319 | +{ |
| 1320 | + return xt_register_targets(xt_imq_reg, ARRAY_SIZE(xt_imq_reg)); |
| 1321 | +} |
| 1322 | + |
| 1323 | +static void __exit imq_fini(void) |
| 1324 | +{ |
| 1325 | + xt_unregister_targets(xt_imq_reg, ARRAY_SIZE(xt_imq_reg)); |
| 1326 | +} |
| 1327 | + |
| 1328 | +module_init(imq_init); |
| 1329 | +module_exit(imq_fini); |
| 1330 | + |
| 1331 | +MODULE_AUTHOR("http://www.linuximq.net"); |
| 1332 | +MODULE_DESCRIPTION("Pseudo-driver for the intermediate queue device. See http://www.linuximq.net/ for more information."); |
| 1333 | +MODULE_LICENSE("GPL"); |
| 1334 | +MODULE_ALIAS("ipt_IMQ"); |
| 1335 | +MODULE_ALIAS("ip6t_IMQ"); |
| 1336 | + |
| 1337 | |
