net/netfilter/nfnetlink_queue_core.c

   1 /*
   2  * This is a module which is used for queueing packets and communicating with
   3  * userspace via nfnetlink.
   4  *
   5  * (C) 2005 by Harald Welte <laforge@netfilter.org>
   6  * (C) 2007 by Patrick McHardy <kaber@trash.net>
   7  *
   8  * Based on the old ipv4-only ip_queue.c:
   9  * (C) 2000-2002 James Morris <jmorris@intercode.com.au>
  10  * (C) 2003-2005 Netfilter Core Team <coreteam@netfilter.org>
  11  *
  12  * This program is free software; you can redistribute it and/or modify
  13  * it under the terms of the GNU General Public License version 2 as
  14  * published by the Free Software Foundation.
  15  *
  16  */
  17 #include <linux/module.h>
  18 #include <linux/skbuff.h>
  19 #include <linux/init.h>
  20 #include <linux/spinlock.h>
  21 #include <linux/slab.h>
  22 #include <linux/notifier.h>
  23 #include <linux/netdevice.h>
  24 #include <linux/netfilter.h>
  25 #include <linux/proc_fs.h>
  26 #include <linux/netfilter_ipv4.h>
  27 #include <linux/netfilter_ipv6.h>
  28 #include <linux/netfilter/nfnetlink.h>
  29 #include <linux/netfilter/nfnetlink_queue.h>
  30 #include <linux/list.h>
  31 #include <net/sock.h>
  32 #include <net/netfilter/nf_queue.h>
  33 #include <net/netns/generic.h>
  34 #include <net/netfilter/nfnetlink_queue.h>
  35
  36 #include <linux/atomic.h>
  37
  38 #ifdef CONFIG_BRIDGE_NETFILTER
  39 #include "../bridge/br_private.h"
  40 #endif
  41
  42 #define NFQNL_QMAX_DEFAULT 1024
  43
  44 struct nfqnl_instance {
  45         struct hlist_node hlist;                /* global list of queues */
  46         struct rcu_head rcu;
  47
  48         int peer_portid;
  49         unsigned int queue_maxlen;
  50         unsigned int copy_range;
  51         unsigned int queue_dropped;
  52         unsigned int queue_user_dropped;
  53
  54
  55         u_int16_t queue_num;                    /* number of this queue */
  56         u_int8_t copy_mode;
  57         u_int32_t flags;                        /* Set using NFQA_CFG_FLAGS */
  58 /*
  59  * Following fields are dirtied for each queued packet,
  60  * keep them in same cache line if possible.
  61  */
  62         spinlock_t      lock;
  63         unsigned int    queue_total;
  64         unsigned int    id_sequence;            /* 'sequence' of pkt ids */
  65         struct list_head queue_list;            /* packets in queue */
  66 };
  67
  68 typedef int (*nfqnl_cmpfn)(struct nf_queue_entry *, unsigned long);
  69
  70 static int nfnl_queue_net_id __read_mostly;
  71
  72 #define INSTANCE_BUCKETS        16
  73 struct nfnl_queue_net {
  74         spinlock_t instances_lock;
  75         struct hlist_head instance_table[INSTANCE_BUCKETS];
  76 };
  77
  78 static struct nfnl_queue_net *nfnl_queue_pernet(struct net *net)
  79 {
  80         return net_generic(net, nfnl_queue_net_id);
  81 }
  82
  83 static inline u_int8_t instance_hashfn(u_int16_t queue_num)
  84 {
  85         return ((queue_num >> 8) ^ queue_num) % INSTANCE_BUCKETS;
  86 }
  87
  88 static struct nfqnl_instance *
  89 instance_lookup(struct nfnl_queue_net *q, u_int16_t queue_num)
  90 {
  91         struct hlist_head *head;
  92         struct nfqnl_instance *inst;
  93
  94         head = &q->instance_table[instance_hashfn(queue_num)];
  95         hlist_for_each_entry_rcu(inst, head, hlist) {
  96                 if (inst->queue_num == queue_num)
  97                         return inst;
  98         }
  99         return NULL;
 100 }
 101
 102 static struct nfqnl_instance *
 103 instance_create(struct nfnl_queue_net *q, u_int16_t queue_num,
 104                 int portid)
 105 {
 106         struct nfqnl_instance *inst;
 107         unsigned int h;
 108         int err;
 109
 110         spin_lock(&q->instances_lock);
 111         if (instance_lookup(q, queue_num)) {
 112                 err = -EEXIST;
 113                 goto out_unlock;
 114         }
 115
 116         inst = kzalloc(sizeof(*inst), GFP_ATOMIC);
 117         if (!inst) {
 118                 err = -ENOMEM;
 119                 goto out_unlock;
 120         }
 121
 122         inst->queue_num = queue_num;
 123         inst->peer_portid = portid;
 124         inst->queue_maxlen = NFQNL_QMAX_DEFAULT;
 125         inst->copy_range = 0xffff;
 126         inst->copy_mode = NFQNL_COPY_NONE;
 127         spin_lock_init(&inst->lock);
 128         INIT_LIST_HEAD(&inst->queue_list);
 129
 130         if (!try_module_get(THIS_MODULE)) {
 131                 err = -EAGAIN;
 132                 goto out_free;
 133         }
 134
 135         h = instance_hashfn(queue_num);
 136         hlist_add_head_rcu(&inst->hlist, &q->instance_table[h]);
 137
 138         spin_unlock(&q->instances_lock);
 139
 140         return inst;
 141
 142 out_free:
 143         kfree(inst);
 144 out_unlock:
 145         spin_unlock(&q->instances_lock);
 146         return ERR_PTR(err);
 147 }
 148
 149 static void nfqnl_flush(struct nfqnl_instance *queue, nfqnl_cmpfn cmpfn,
 150                         unsigned long data);
 151
 152 static void
 153 instance_destroy_rcu(struct rcu_head *head)
 154 {
 155         struct nfqnl_instance *inst = container_of(head, struct nfqnl_instance,
 156                                                    rcu);
 157
 158         nfqnl_flush(inst, NULL, 0);
 159         kfree(inst);
 160         module_put(THIS_MODULE);
 161 }
 162
 163 static void
 164 __instance_destroy(struct nfqnl_instance *inst)
 165 {
 166         hlist_del_rcu(&inst->hlist);
 167         call_rcu(&inst->rcu, instance_destroy_rcu);
 168 }
 169
 170 static void
 171 instance_destroy(struct nfnl_queue_net *q, struct nfqnl_instance *inst)
 172 {
 173         spin_lock(&q->instances_lock);
 174         __instance_destroy(inst);
 175         spin_unlock(&q->instances_lock);
 176 }
 177
 178 static inline void
 179 __enqueue_entry(struct nfqnl_instance *queue, struct nf_queue_entry *entry)
 180 {
 181        list_add_tail(&entry->list, &queue->queue_list);
 182        queue->queue_total++;
 183 }
 184
 185 static void
 186 __dequeue_entry(struct nfqnl_instance *queue, struct nf_queue_entry *entry)
 187 {
 188         list_del(&entry->list);
 189         queue->queue_total--;
 190 }
 191
 192 static struct nf_queue_entry *
 193 find_dequeue_entry(struct nfqnl_instance *queue, unsigned int id)
 194 {
 195         struct nf_queue_entry *entry = NULL, *i;
 196
 197         spin_lock_bh(&queue->lock);
 198
 199         list_for_each_entry(i, &queue->queue_list, list) {
 200                 if (i->id == id) {
 201                         entry = i;
 202                         break;
 203                 }
 204         }
 205
 206         if (entry)
 207                 __dequeue_entry(queue, entry);
 208
 209         spin_unlock_bh(&queue->lock);
 210
 211         return entry;
 212 }
 213
 214 static void
 215 nfqnl_flush(struct nfqnl_instance *queue, nfqnl_cmpfn cmpfn, unsigned long data)
 216 {
 217         struct nf_queue_entry *entry, *next;
 218
 219         spin_lock_bh(&queue->lock);
 220         list_for_each_entry_safe(entry, next, &queue->queue_list, list) {
 221                 if (!cmpfn || cmpfn(entry, data)) {
 222                         list_del(&entry->list);
 223                         queue->queue_total--;
 224                         nf_reinject(entry, NF_DROP);
 225                 }
 226         }
 227         spin_unlock_bh(&queue->lock);
 228 }
 229
 230 static void
 231 nfqnl_zcopy(struct sk_buff *to, const struct sk_buff *from, int len, int hlen)
 232 {
 233         int i, j = 0;
 234         int plen = 0; /* length of skb->head fragment */
 235         struct page *page;
 236         unsigned int offset;
 237
 238         /* dont bother with small payloads */
 239         if (len <= skb_tailroom(to)) {
 240                 skb_copy_bits(from, 0, skb_put(to, len), len);
 241                 return;
 242         }
 243
 244         if (hlen) {
 245                 skb_copy_bits(from, 0, skb_put(to, hlen), hlen);
 246                 len -= hlen;
 247         } else {
 248                 plen = min_t(int, skb_headlen(from), len);
 249                 if (plen) {
 250                         page = virt_to_head_page(from->head);
 251                         offset = from->data - (unsigned char *)page_address(page);
 252                         __skb_fill_page_desc(to, 0, page, offset, plen);
 253                         get_page(page);
 254                         j = 1;
 255                         len -= plen;
 256                 }
 257         }
 258
 259         to->truesize += len + plen;
 260         to->len += len + plen;
 261         to->data_len += len + plen;
 262
 263         for (i = 0; i < skb_shinfo(from)->nr_frags; i++) {
 264                 if (!len)
 265                         break;
 266                 skb_shinfo(to)->frags[j] = skb_shinfo(from)->frags[i];
 267                 skb_shinfo(to)->frags[j].size = min_t(int, skb_shinfo(to)->frags[j].size, len);
 268                 len -= skb_shinfo(to)->frags[j].size;
 269                 skb_frag_ref(to, j);
 270                 j++;
 271         }
 272         skb_shinfo(to)->nr_frags = j;
 273 }
 274
 275 static int nfqnl_put_packet_info(struct sk_buff *nlskb, struct sk_buff *packet)
 276 {
 277         __u32 flags = 0;
 278
 279         if (packet->ip_summed == CHECKSUM_PARTIAL)
 280                 flags = NFQA_SKB_CSUMNOTREADY;
 281         if (skb_is_gso(packet))
 282                 flags |= NFQA_SKB_GSO;
 283
 284         return flags ? nla_put_be32(nlskb, NFQA_SKB_INFO, htonl(flags)) : 0;
 285 }
 286
 287 static struct sk_buff *
 288 nfqnl_build_packet_message(struct nfqnl_instance *queue,
 289                            struct nf_queue_entry *entry,
 290                            __be32 **packet_id_ptr)
 291 {
 292         size_t size;
 293         size_t data_len = 0, cap_len = 0;
 294         int hlen = 0;
 295         struct sk_buff *skb;
 296         struct nlattr *nla;
 297         struct nfqnl_msg_packet_hdr *pmsg;
 298         struct nlmsghdr *nlh;
 299         struct nfgenmsg *nfmsg;
 300         struct sk_buff *entskb = entry->skb;
 301         struct net_device *indev;
 302         struct net_device *outdev;
 303         struct nf_conn *ct = NULL;
 304         enum ip_conntrack_info uninitialized_var(ctinfo);
 305
 306         size =    nlmsg_total_size(sizeof(struct nfgenmsg))
 307                 + nla_total_size(sizeof(struct nfqnl_msg_packet_hdr))
 308                 + nla_total_size(sizeof(u_int32_t))     /* ifindex */
 309                 + nla_total_size(sizeof(u_int32_t))     /* ifindex */
 310 #ifdef CONFIG_BRIDGE_NETFILTER
 311                 + nla_total_size(sizeof(u_int32_t))     /* ifindex */
 312                 + nla_total_size(sizeof(u_int32_t))     /* ifindex */
 313 #endif
 314                 + nla_total_size(sizeof(u_int32_t))     /* mark */
 315                 + nla_total_size(sizeof(struct nfqnl_msg_packet_hw))
 316                 + nla_total_size(sizeof(u_int32_t))     /* skbinfo */
 317                 + nla_total_size(sizeof(u_int32_t));    /* cap_len */
 318
 319         if (entskb->tstamp.tv64)
 320                 size += nla_total_size(sizeof(struct nfqnl_msg_packet_timestamp));
 321
 322         outdev = entry->outdev;
 323
 324         switch ((enum nfqnl_config_mode)ACCESS_ONCE(queue->copy_mode)) {
 325         case NFQNL_COPY_META:
 326         case NFQNL_COPY_NONE:
 327                 break;
 328
 329         case NFQNL_COPY_PACKET:
 330                 if (entskb->ip_summed == CHECKSUM_PARTIAL &&
 331                     skb_checksum_help(entskb))
 332                         return NULL;
 333
 334                 data_len = ACCESS_ONCE(queue->copy_range);
 335                 if (data_len == 0 || data_len > entskb->len)
 336                         data_len = entskb->len;
 337
 338
 339                 if (!entskb->head_frag ||
 340                     skb_headlen(entskb) < L1_CACHE_BYTES ||
 341                     skb_shinfo(entskb)->nr_frags >= MAX_SKB_FRAGS)
 342                         hlen = skb_headlen(entskb);
 343
 344                 if (skb_has_frag_list(entskb))
 345                         hlen = entskb->len;
 346                 hlen = min_t(int, data_len, hlen);
 347                 size += sizeof(struct nlattr) + hlen;
 348                 cap_len = entskb->len;
 349                 break;
 350         }
 351
 352         if (queue->flags & NFQA_CFG_F_CONNTRACK)
 353                 ct = nfqnl_ct_get(entskb, &size, &ctinfo);
 354
 355         skb = nfnetlink_alloc_skb(&init_net, size, queue->peer_portid,
 356                                   GFP_ATOMIC);
 357         if (!skb)
 358                 return NULL;
 359
 360         nlh = nlmsg_put(skb, 0, 0,
 361                         NFNL_SUBSYS_QUEUE << 8 | NFQNL_MSG_PACKET,
 362                         sizeof(struct nfgenmsg), 0);
 363         if (!nlh) {
 364                 kfree_skb(skb);
 365                 return NULL;
 366         }
 367         nfmsg = nlmsg_data(nlh);
 368         nfmsg->nfgen_family = entry->pf;
 369         nfmsg->version = NFNETLINK_V0;
 370         nfmsg->res_id = htons(queue->queue_num);
 371
 372         nla = __nla_reserve(skb, NFQA_PACKET_HDR, sizeof(*pmsg));
 373         pmsg = nla_data(nla);
 374         pmsg->hw_protocol       = entskb->protocol;
 375         pmsg->hook              = entry->hook;
 376         *packet_id_ptr          = &pmsg->packet_id;
 377
 378         indev = entry->indev;
 379         if (indev) {
 380 #ifndef CONFIG_BRIDGE_NETFILTER
 381                 if (nla_put_be32(skb, NFQA_IFINDEX_INDEV, htonl(indev->ifindex)))
 382                         goto nla_put_failure;
 383 #else
 384                 if (entry->pf == PF_BRIDGE) {
 385                         /* Case 1: indev is physical input device, we need to
 386                          * look for bridge group (when called from
 387                          * netfilter_bridge) */
 388                         if (nla_put_be32(skb, NFQA_IFINDEX_PHYSINDEV,
 389                                          htonl(indev->ifindex)) ||
 390                         /* this is the bridge group "brX" */
 391                         /* rcu_read_lock()ed by __nf_queue */
 392                             nla_put_be32(skb, NFQA_IFINDEX_INDEV,
 393                                          htonl(br_port_get_rcu(indev)->br->dev->ifindex)))
 394                                 goto nla_put_failure;
 395                 } else {
 396                         /* Case 2: indev is bridge group, we need to look for
 397                          * physical device (when called from ipv4) */
 398                         if (nla_put_be32(skb, NFQA_IFINDEX_INDEV,
 399                                          htonl(indev->ifindex)))
 400                                 goto nla_put_failure;
 401                         if (entskb->nf_bridge && entskb->nf_bridge->physindev &&
 402                             nla_put_be32(skb, NFQA_IFINDEX_PHYSINDEV,
 403                                          htonl(entskb->nf_bridge->physindev->ifindex)))
 404                                 goto nla_put_failure;
 405                 }
 406 #endif
 407         }
 408
 409         if (outdev) {
 410 #ifndef CONFIG_BRIDGE_NETFILTER
 411                 if (nla_put_be32(skb, NFQA_IFINDEX_OUTDEV, htonl(outdev->ifindex)))
 412                         goto nla_put_failure;
 413 #else
 414                 if (entry->pf == PF_BRIDGE) {
 415                         /* Case 1: outdev is physical output device, we need to
 416                          * look for bridge group (when called from
 417                          * netfilter_bridge) */
 418                         if (nla_put_be32(skb, NFQA_IFINDEX_PHYSOUTDEV,
 419                                          htonl(outdev->ifindex)) ||
 420                         /* this is the bridge group "brX" */
 421                         /* rcu_read_lock()ed by __nf_queue */
 422                             nla_put_be32(skb, NFQA_IFINDEX_OUTDEV,
 423                                          htonl(br_port_get_rcu(outdev)->br->dev->ifindex)))
 424                                 goto nla_put_failure;
 425                 } else {
 426                         /* Case 2: outdev is bridge group, we need to look for
 427                          * physical output device (when called from ipv4) */
 428                         if (nla_put_be32(skb, NFQA_IFINDEX_OUTDEV,
 429                                          htonl(outdev->ifindex)))
 430                                 goto nla_put_failure;
 431                         if (entskb->nf_bridge && entskb->nf_bridge->physoutdev &&
 432                             nla_put_be32(skb, NFQA_IFINDEX_PHYSOUTDEV,
 433                                          htonl(entskb->nf_bridge->physoutdev->ifindex)))
 434                                 goto nla_put_failure;
 435                 }
 436 #endif
 437         }
 438
 439         if (entskb->mark &&
 440             nla_put_be32(skb, NFQA_MARK, htonl(entskb->mark)))
 441                 goto nla_put_failure;
 442
 443         if (indev && entskb->dev &&
 444             entskb->mac_header != entskb->network_header) {
 445                 struct nfqnl_msg_packet_hw phw;
 446                 int len = dev_parse_header(entskb, phw.hw_addr);
 447                 if (len) {
 448                         phw.hw_addrlen = htons(len);
 449                         if (nla_put(skb, NFQA_HWADDR, sizeof(phw), &phw))
 450                                 goto nla_put_failure;
 451                 }
 452         }
 453
 454         if (entskb->tstamp.tv64) {
 455                 struct nfqnl_msg_packet_timestamp ts;
 456                 struct timeval tv = ktime_to_timeval(entskb->tstamp);
 457                 ts.sec = cpu_to_be64(tv.tv_sec);
 458                 ts.usec = cpu_to_be64(tv.tv_usec);
 459
 460                 if (nla_put(skb, NFQA_TIMESTAMP, sizeof(ts), &ts))
 461                         goto nla_put_failure;
 462         }
 463
 464         if (ct && nfqnl_ct_put(skb, ct, ctinfo) < 0)
 465                 goto nla_put_failure;
 466
 467         if (cap_len > 0 && nla_put_be32(skb, NFQA_CAP_LEN, htonl(cap_len)))
 468                 goto nla_put_failure;
 469
 470         if (nfqnl_put_packet_info(skb, entskb))
 471                 goto nla_put_failure;
 472
 473         if (data_len) {
 474                 struct nlattr *nla;
 475
 476                 if (skb_tailroom(skb) < sizeof(*nla) + hlen)
 477                         goto nla_put_failure;
 478
 479                 nla = (struct nlattr *)skb_put(skb, sizeof(*nla));
 480                 nla->nla_type = NFQA_PAYLOAD;
 481                 nla->nla_len = nla_attr_size(data_len);
 482
 483                 nfqnl_zcopy(skb, entskb, data_len, hlen);
 484         }
 485
 486         nlh->nlmsg_len = skb->len;
 487         return skb;
 488
 489 nla_put_failure:
 490         kfree_skb(skb);
 491         net_err_ratelimited("nf_queue: error creating packet message\n");
 492         return NULL;
 493 }
 494
 495 static int
 496 __nfqnl_enqueue_packet(struct net *net, struct nfqnl_instance *queue,
 497                         struct nf_queue_entry *entry)
 498 {
 499         struct sk_buff *nskb;
 500         int err = -ENOBUFS;
 501         __be32 *packet_id_ptr;
 502         int failopen = 0;
 503
 504         nskb = nfqnl_build_packet_message(queue, entry, &packet_id_ptr);
 505         if (nskb == NULL) {
 506                 err = -ENOMEM;
 507                 goto err_out;
 508         }
 509         spin_lock_bh(&queue->lock);
 510
 511         if (!queue->peer_portid) {
 512                 err = -EINVAL;
 513                 goto err_out_free_nskb;
 514         }
 515         if (queue->queue_total >= queue->queue_maxlen) {
 516                 if (queue->flags & NFQA_CFG_F_FAIL_OPEN) {
 517                         failopen = 1;
 518                         err = 0;
 519                 } else {
 520                         queue->queue_dropped++;
 521                         net_warn_ratelimited("nf_queue: full at %d entries, dropping packets(s)\n",
 522                                              queue->queue_total);
 523                 }
 524                 goto err_out_free_nskb;
 525         }
 526         entry->id = ++queue->id_sequence;
 527         *packet_id_ptr = htonl(entry->id);
 528
 529         /* nfnetlink_unicast will either free the nskb or add it to a socket */
 530         err = nfnetlink_unicast(nskb, net, queue->peer_portid, MSG_DONTWAIT);
 531         if (err < 0) {
 532                 queue->queue_user_dropped++;
 533                 goto err_out_unlock;
 534         }
 535
 536         __enqueue_entry(queue, entry);
 537
 538         spin_unlock_bh(&queue->lock);
 539         return 0;
 540
 541 err_out_free_nskb:
 542         kfree_skb(nskb);
 543 err_out_unlock:
 544         spin_unlock_bh(&queue->lock);
 545         if (failopen)
 546                 nf_reinject(entry, NF_ACCEPT);
 547 err_out:
 548         return err;
 549 }
 550
 551 static struct nf_queue_entry *
 552 nf_queue_entry_dup(struct nf_queue_entry *e)
 553 {
 554         struct nf_queue_entry *entry = kmemdup(e, e->size, GFP_ATOMIC);
 555         if (entry) {
 556                 if (nf_queue_entry_get_refs(entry))
 557                         return entry;
 558                 kfree(entry);
 559         }
 560         return NULL;
 561 }
 562
 563 #ifdef CONFIG_BRIDGE_NETFILTER
 564 /* When called from bridge netfilter, skb->data must point to MAC header
 565  * before calling skb_gso_segment(). Else, original MAC header is lost
 566  * and segmented skbs will be sent to wrong destination.
 567  */
 568 static void nf_bridge_adjust_skb_data(struct sk_buff *skb)
 569 {
 570         if (skb->nf_bridge)
 571                 __skb_push(skb, skb->network_header - skb->mac_header);
 572 }
 573
 574 static void nf_bridge_adjust_segmented_data(struct sk_buff *skb)
 575 {
 576         if (skb->nf_bridge)
 577                 __skb_pull(skb, skb->network_header - skb->mac_header);
 578 }
 579 #else
 580 #define nf_bridge_adjust_skb_data(s) do {} while (0)
 581 #define nf_bridge_adjust_segmented_data(s) do {} while (0)
 582 #endif
 583
 584 static void free_entry(struct nf_queue_entry *entry)
 585 {
 586         nf_queue_entry_release_refs(entry);
 587         kfree(entry);
 588 }
 589
 590 static int
 591 __nfqnl_enqueue_packet_gso(struct net *net, struct nfqnl_instance *queue,
 592                            struct sk_buff *skb, struct nf_queue_entry *entry)
 593 {
 594         int ret = -ENOMEM;
 595         struct nf_queue_entry *entry_seg;
 596
 597         nf_bridge_adjust_segmented_data(skb);
 598
 599         if (skb->next == NULL) { /* last packet, no need to copy entry */
 600                 struct sk_buff *gso_skb = entry->skb;
 601                 entry->skb = skb;
 602                 ret = __nfqnl_enqueue_packet(net, queue, entry);
 603                 if (ret)
 604                         entry->skb = gso_skb;
 605                 return ret;
 606         }
 607
 608         skb->next = NULL;
 609
 610         entry_seg = nf_queue_entry_dup(entry);
 611         if (entry_seg) {
 612                 entry_seg->skb = skb;
 613                 ret = __nfqnl_enqueue_packet(net, queue, entry_seg);
 614                 if (ret)
 615                         free_entry(entry_seg);
 616         }
 617         return ret;
 618 }
 619
 620 static int
 621 nfqnl_enqueue_packet(struct nf_queue_entry *entry, unsigned int queuenum)
 622 {
 623         unsigned int queued;
 624         struct nfqnl_instance *queue;
 625         struct sk_buff *skb, *segs;
 626         int err = -ENOBUFS;
 627         struct net *net = dev_net(entry->indev ?
 628                                   entry->indev : entry->outdev);
 629         struct nfnl_queue_net *q = nfnl_queue_pernet(net);
 630
 631         /* rcu_read_lock()ed by nf_hook_slow() */
 632         queue = instance_lookup(q, queuenum);
 633         if (!queue)
 634                 return -ESRCH;
 635
 636         if (queue->copy_mode == NFQNL_COPY_NONE)
 637                 return -EINVAL;
 638
 639         if (!skb_is_gso(entry->skb))
 640                 return __nfqnl_enqueue_packet(net, queue, entry);
 641
 642         skb = entry->skb;
 643
 644         switch (entry->pf) {
 645         case NFPROTO_IPV4:
 646                 skb->protocol = htons(ETH_P_IP);
 647                 break;
 648         case NFPROTO_IPV6:
 649                 skb->protocol = htons(ETH_P_IPV6);
 650                 break;
 651         }
 652
 653         nf_bridge_adjust_skb_data(skb);
 654         segs = skb_gso_segment(skb, 0);
 655         /* Does not use PTR_ERR to limit the number of error codes that can be
 656          * returned by nf_queue.  For instance, callers rely on -ECANCELED to
 657          * mean 'ignore this hook'.
 658          */
 659         if (IS_ERR(segs))
 660                 goto out_err;
 661         queued = 0;
 662         err = 0;
 663         do {
 664                 struct sk_buff *nskb = segs->next;
 665                 if (err == 0)
 666                         err = __nfqnl_enqueue_packet_gso(net, queue,
 667                                                         segs, entry);
 668                 if (err == 0)
 669                         queued++;
 670                 else
 671                         kfree_skb(segs);
 672                 segs = nskb;
 673         } while (segs);
 674
 675         if (queued) {
 676                 if (err) /* some segments are already queued */
 677                         free_entry(entry);
 678                 kfree_skb(skb);
 679                 return 0;
 680         }
 681  out_err:
 682         nf_bridge_adjust_segmented_data(skb);
 683         return err;
 684 }
 685
 686 static int
 687 nfqnl_mangle(void *data, int data_len, struct nf_queue_entry *e, int diff)
 688 {
 689         struct sk_buff *nskb;
 690
 691         if (diff < 0) {
 692                 if (pskb_trim(e->skb, data_len))
 693                         return -ENOMEM;
 694         } else if (diff > 0) {
 695                 if (data_len > 0xFFFF)
 696                         return -EINVAL;
 697                 if (diff > skb_tailroom(e->skb)) {
 698                         nskb = skb_copy_expand(e->skb, skb_headroom(e->skb),
 699                                                diff, GFP_ATOMIC);
 700                         if (!nskb) {
 701                                 printk(KERN_WARNING "nf_queue: OOM "
 702                                       "in mangle, dropping packet\n");
 703                                 return -ENOMEM;
 704                         }
 705                         kfree_skb(e->skb);
 706                         e->skb = nskb;
 707                 }
 708                 skb_put(e->skb, diff);
 709         }
 710         if (!skb_make_writable(e->skb, data_len))
 711                 return -ENOMEM;
 712         skb_copy_to_linear_data(e->skb, data, data_len);
 713         e->skb->ip_summed = CHECKSUM_NONE;
 714         return 0;
 715 }
 716
 717 static int
 718 nfqnl_set_mode(struct nfqnl_instance *queue,
 719                unsigned char mode, unsigned int range)
 720 {
 721         int status = 0;
 722
 723         spin_lock_bh(&queue->lock);
 724         switch (mode) {
 725         case NFQNL_COPY_NONE:
 726         case NFQNL_COPY_META:
 727                 queue->copy_mode = mode;
 728                 queue->copy_range = 0;
 729                 break;
 730
 731         case NFQNL_COPY_PACKET:
 732                 queue->copy_mode = mode;
 733                 /* We're using struct nlattr which has 16bit nla_len. Note that
 734                  * nla_len includes the header length. Thus, the maximum packet
 735                  * length that we support is 65531 bytes. We send truncated
 736                  * packets if the specified length is larger than that.
 737                  */
 738                 if (range > 0xffff - NLA_HDRLEN)
 739                         queue->copy_range = 0xffff - NLA_HDRLEN;
 740                 else
 741                         queue->copy_range = range;
 742                 break;
 743
 744         default:
 745                 status = -EINVAL;
 746
 747         }
 748         spin_unlock_bh(&queue->lock);
 749
 750         return status;
 751 }
 752
 753 static int
 754 dev_cmp(struct nf_queue_entry *entry, unsigned long ifindex)
 755 {
 756         if (entry->indev)
 757                 if (entry->indev->ifindex == ifindex)
 758                         return 1;
 759         if (entry->outdev)
 760                 if (entry->outdev->ifindex == ifindex)
 761                         return 1;
 762 #ifdef CONFIG_BRIDGE_NETFILTER
 763         if (entry->skb->nf_bridge) {
 764                 if (entry->skb->nf_bridge->physindev &&
 765                     entry->skb->nf_bridge->physindev->ifindex == ifindex)
 766                         return 1;
 767                 if (entry->skb->nf_bridge->physoutdev &&
 768                     entry->skb->nf_bridge->physoutdev->ifindex == ifindex)
 769                         return 1;
 770         }
 771 #endif
 772         return 0;
 773 }
 774
 775 /* drop all packets with either indev or outdev == ifindex from all queue
 776  * instances */
 777 static void
 778 nfqnl_dev_drop(struct net *net, int ifindex)
 779 {
 780         int i;
 781         struct nfnl_queue_net *q = nfnl_queue_pernet(net);
 782
 783         rcu_read_lock();
 784
 785         for (i = 0; i < INSTANCE_BUCKETS; i++) {
 786                 struct nfqnl_instance *inst;
 787                 struct hlist_head *head = &q->instance_table[i];
 788
 789                 hlist_for_each_entry_rcu(inst, head, hlist)
 790                         nfqnl_flush(inst, dev_cmp, ifindex);
 791         }
 792
 793         rcu_read_unlock();
 794 }
 795
 796 #define RCV_SKB_FAIL(err) do { netlink_ack(skb, nlh, (err)); return; } while (0)
 797
 798 static int
 799 nfqnl_rcv_dev_event(struct notifier_block *this,
 800                     unsigned long event, void *ptr)
 801 {
 802         struct net_device *dev = ptr;
 803
 804         /* Drop any packets associated with the downed device */
 805         if (event == NETDEV_DOWN)
 806                 nfqnl_dev_drop(dev_net(dev), dev->ifindex);
 807         return NOTIFY_DONE;
 808 }
 809
 810 static struct notifier_block nfqnl_dev_notifier = {
 811         .notifier_call  = nfqnl_rcv_dev_event,
 812 };
 813
 814 static int
 815 nfqnl_rcv_nl_event(struct notifier_block *this,
 816                    unsigned long event, void *ptr)
 817 {
 818         struct netlink_notify *n = ptr;
 819         struct nfnl_queue_net *q = nfnl_queue_pernet(n->net);
 820
 821         if (event == NETLINK_URELEASE && n->protocol == NETLINK_NETFILTER) {
 822                 int i;
 823
 824                 /* destroy all instances for this portid */
 825                 spin_lock(&q->instances_lock);
 826                 for (i = 0; i < INSTANCE_BUCKETS; i++) {
 827                         struct hlist_node *t2;
 828                         struct nfqnl_instance *inst;
 829                         struct hlist_head *head = &q->instance_table[i];
 830
 831                         hlist_for_each_entry_safe(inst, t2, head, hlist) {
 832                                 if (n->portid == inst->peer_portid)
 833                                         __instance_destroy(inst);
 834                         }
 835                 }
 836                 spin_unlock(&q->instances_lock);
 837         }
 838         return NOTIFY_DONE;
 839 }
 840
 841 static struct notifier_block nfqnl_rtnl_notifier = {
 842         .notifier_call  = nfqnl_rcv_nl_event,
 843 };
 844
 845 static const struct nla_policy nfqa_verdict_policy[NFQA_MAX+1] = {
 846         [NFQA_VERDICT_HDR]      = { .len = sizeof(struct nfqnl_msg_verdict_hdr) },
 847         [NFQA_MARK]             = { .type = NLA_U32 },
 848         [NFQA_PAYLOAD]          = { .type = NLA_UNSPEC },
 849         [NFQA_CT]               = { .type = NLA_UNSPEC },
 850 };
 851
 852 static const struct nla_policy nfqa_verdict_batch_policy[NFQA_MAX+1] = {
 853         [NFQA_VERDICT_HDR]      = { .len = sizeof(struct nfqnl_msg_verdict_hdr) },
 854         [NFQA_MARK]             = { .type = NLA_U32 },
 855 };
 856
 857 static struct nfqnl_instance *
 858 verdict_instance_lookup(struct nfnl_queue_net *q, u16 queue_num, int nlportid)
 859 {
 860         struct nfqnl_instance *queue;
 861
 862         queue = instance_lookup(q, queue_num);
 863         if (!queue)
 864                 return ERR_PTR(-ENODEV);
 865
 866         if (queue->peer_portid != nlportid)
 867                 return ERR_PTR(-EPERM);
 868
 869         return queue;
 870 }
 871
 872 static struct nfqnl_msg_verdict_hdr*
 873 verdicthdr_get(const struct nlattr * const nfqa[])
 874 {
 875         struct nfqnl_msg_verdict_hdr *vhdr;
 876         unsigned int verdict;
 877
 878         if (!nfqa[NFQA_VERDICT_HDR])
 879                 return NULL;
 880
 881         vhdr = nla_data(nfqa[NFQA_VERDICT_HDR]);
 882         verdict = ntohl(vhdr->verdict) & NF_VERDICT_MASK;
 883         if (verdict > NF_MAX_VERDICT || verdict == NF_STOLEN)
 884                 return NULL;
 885         return vhdr;
 886 }
 887
 888 static int nfq_id_after(unsigned int id, unsigned int max)
 889 {
 890         return (int)(id - max) > 0;
 891 }
 892
 893 static int
 894 nfqnl_recv_verdict_batch(struct sock *ctnl, struct sk_buff *skb,
 895                    const struct nlmsghdr *nlh,
 896                    const struct nlattr * const nfqa[])
 897 {
 898         struct nfgenmsg *nfmsg = nlmsg_data(nlh);
 899         struct nf_queue_entry *entry, *tmp;
 900         unsigned int verdict, maxid;
 901         struct nfqnl_msg_verdict_hdr *vhdr;
 902         struct nfqnl_instance *queue;
 903         LIST_HEAD(batch_list);
 904         u16 queue_num = ntohs(nfmsg->res_id);
 905
 906         struct net *net = sock_net(ctnl);
 907         struct nfnl_queue_net *q = nfnl_queue_pernet(net);
 908
 909         queue = verdict_instance_lookup(q, queue_num,
 910                                         NETLINK_CB(skb).portid);
 911         if (IS_ERR(queue))
 912                 return PTR_ERR(queue);
 913
 914         vhdr = verdicthdr_get(nfqa);
 915         if (!vhdr)
 916                 return -EINVAL;
 917
 918         verdict = ntohl(vhdr->verdict);
 919         maxid = ntohl(vhdr->id);
 920
 921         spin_lock_bh(&queue->lock);
 922
 923         list_for_each_entry_safe(entry, tmp, &queue->queue_list, list) {
 924                 if (nfq_id_after(entry->id, maxid))
 925                         break;
 926                 __dequeue_entry(queue, entry);
 927                 list_add_tail(&entry->list, &batch_list);
 928         }
 929
 930         spin_unlock_bh(&queue->lock);
 931
 932         if (list_empty(&batch_list))
 933                 return -ENOENT;
 934
 935         list_for_each_entry_safe(entry, tmp, &batch_list, list) {
 936                 if (nfqa[NFQA_MARK])
 937                         entry->skb->mark = ntohl(nla_get_be32(nfqa[NFQA_MARK]));
 938                 nf_reinject(entry, verdict);
 939         }
 940         return 0;
 941 }
 942
 943 static int
 944 nfqnl_recv_verdict(struct sock *ctnl, struct sk_buff *skb,
 945                    const struct nlmsghdr *nlh,
 946                    const struct nlattr * const nfqa[])
 947 {
 948         struct nfgenmsg *nfmsg = nlmsg_data(nlh);
 949         u_int16_t queue_num = ntohs(nfmsg->res_id);
 950
 951         struct nfqnl_msg_verdict_hdr *vhdr;
 952         struct nfqnl_instance *queue;
 953         unsigned int verdict;
 954         struct nf_queue_entry *entry;
 955         enum ip_conntrack_info uninitialized_var(ctinfo);
 956         struct nf_conn *ct = NULL;
 957
 958         struct net *net = sock_net(ctnl);
 959         struct nfnl_queue_net *q = nfnl_queue_pernet(net);
 960
 961         queue = instance_lookup(q, queue_num);
 962         if (!queue)
 963                 queue = verdict_instance_lookup(q, queue_num,
 964                                                 NETLINK_CB(skb).portid);
 965         if (IS_ERR(queue))
 966                 return PTR_ERR(queue);
 967
 968         vhdr = verdicthdr_get(nfqa);
 969         if (!vhdr)
 970                 return -EINVAL;
 971
 972         verdict = ntohl(vhdr->verdict);
 973
 974         entry = find_dequeue_entry(queue, ntohl(vhdr->id));
 975         if (entry == NULL)
 976                 return -ENOENT;
 977
 978         rcu_read_lock();
 979         if (nfqa[NFQA_CT] && (queue->flags & NFQA_CFG_F_CONNTRACK))
 980                 ct = nfqnl_ct_parse(entry->skb, nfqa[NFQA_CT], &ctinfo);
 981
 982         if (nfqa[NFQA_PAYLOAD]) {
 983                 u16 payload_len = nla_len(nfqa[NFQA_PAYLOAD]);
 984                 int diff = payload_len - entry->skb->len;
 985
 986                 if (nfqnl_mangle(nla_data(nfqa[NFQA_PAYLOAD]),
 987                                  payload_len, entry, diff) < 0)
 988                         verdict = NF_DROP;
 989
 990                 if (ct)
 991                         nfqnl_ct_seq_adjust(skb, ct, ctinfo, diff);
 992         }
 993         rcu_read_unlock();
 994
 995         if (nfqa[NFQA_MARK])
 996                 entry->skb->mark = ntohl(nla_get_be32(nfqa[NFQA_MARK]));
 997
 998         nf_reinject(entry, verdict);
 999         return 0;
1000 }
1001
1002 static int
1003 nfqnl_recv_unsupp(struct sock *ctnl, struct sk_buff *skb,
1004                   const struct nlmsghdr *nlh,
1005                   const struct nlattr * const nfqa[])
1006 {
1007         return -ENOTSUPP;
1008 }
1009
1010 static const struct nla_policy nfqa_cfg_policy[NFQA_CFG_MAX+1] = {
1011         [NFQA_CFG_CMD]          = { .len = sizeof(struct nfqnl_msg_config_cmd) },
1012         [NFQA_CFG_PARAMS]       = { .len = sizeof(struct nfqnl_msg_config_params) },
1013 };
1014
1015 static const struct nf_queue_handler nfqh = {
1016         .outfn  = &nfqnl_enqueue_packet,
1017 };
1018
1019 static int
1020 nfqnl_recv_config(struct sock *ctnl, struct sk_buff *skb,
1021                   const struct nlmsghdr *nlh,
1022                   const struct nlattr * const nfqa[])
1023 {
1024         struct nfgenmsg *nfmsg = nlmsg_data(nlh);
1025         u_int16_t queue_num = ntohs(nfmsg->res_id);
1026         struct nfqnl_instance *queue;
1027         struct nfqnl_msg_config_cmd *cmd = NULL;
1028         struct net *net = sock_net(ctnl);
1029         struct nfnl_queue_net *q = nfnl_queue_pernet(net);
1030         int ret = 0;
1031
1032         if (nfqa[NFQA_CFG_CMD]) {
1033                 cmd = nla_data(nfqa[NFQA_CFG_CMD]);
1034
1035                 /* Obsolete commands without queue context */
1036                 switch (cmd->command) {
1037                 case NFQNL_CFG_CMD_PF_BIND: return 0;
1038                 case NFQNL_CFG_CMD_PF_UNBIND: return 0;
1039                 }
1040         }
1041
1042         rcu_read_lock();
1043         queue = instance_lookup(q, queue_num);
1044         if (queue && queue->peer_portid != NETLINK_CB(skb).portid) {
1045                 ret = -EPERM;
1046                 goto err_out_unlock;
1047         }
1048
1049         if (cmd != NULL) {
1050                 switch (cmd->command) {
1051                 case NFQNL_CFG_CMD_BIND:
1052                         if (queue) {
1053                                 ret = -EBUSY;
1054                                 goto err_out_unlock;
1055                         }
1056                         queue = instance_create(q, queue_num,
1057                                                 NETLINK_CB(skb).portid);
1058                         if (IS_ERR(queue)) {
1059                                 ret = PTR_ERR(queue);
1060                                 goto err_out_unlock;
1061                         }
1062                         break;
1063                 case NFQNL_CFG_CMD_UNBIND:
1064                         if (!queue) {
1065                                 ret = -ENODEV;
1066                                 goto err_out_unlock;
1067                         }
1068                         instance_destroy(q, queue);
1069                         break;
1070                 case NFQNL_CFG_CMD_PF_BIND:
1071                 case NFQNL_CFG_CMD_PF_UNBIND:
1072                         break;
1073                 default:
1074                         ret = -ENOTSUPP;
1075                         break;
1076                 }
1077         }
1078
1079         if (nfqa[NFQA_CFG_PARAMS]) {
1080                 struct nfqnl_msg_config_params *params;
1081
1082                 if (!queue) {
1083                         ret = -ENODEV;
1084                         goto err_out_unlock;
1085                 }
1086                 params = nla_data(nfqa[NFQA_CFG_PARAMS]);
1087                 nfqnl_set_mode(queue, params->copy_mode,
1088                                 ntohl(params->copy_range));
1089         }
1090
1091         if (nfqa[NFQA_CFG_QUEUE_MAXLEN]) {
1092                 __be32 *queue_maxlen;
1093
1094                 if (!queue) {
1095                         ret = -ENODEV;
1096                         goto err_out_unlock;
1097                 }
1098                 queue_maxlen = nla_data(nfqa[NFQA_CFG_QUEUE_MAXLEN]);
1099                 spin_lock_bh(&queue->lock);
1100                 queue->queue_maxlen = ntohl(*queue_maxlen);
1101                 spin_unlock_bh(&queue->lock);
1102         }
1103
1104         if (nfqa[NFQA_CFG_FLAGS]) {
1105                 __u32 flags, mask;
1106
1107                 if (!queue) {
1108                         ret = -ENODEV;
1109                         goto err_out_unlock;
1110                 }
1111
1112                 if (!nfqa[NFQA_CFG_MASK]) {
1113                         /* A mask is needed to specify which flags are being
1114                          * changed.
1115                          */
1116                         ret = -EINVAL;
1117                         goto err_out_unlock;
1118                 }
1119
1120                 flags = ntohl(nla_get_be32(nfqa[NFQA_CFG_FLAGS]));
1121                 mask = ntohl(nla_get_be32(nfqa[NFQA_CFG_MASK]));
1122
1123                 if (flags >= NFQA_CFG_F_MAX) {
1124                         ret = -EOPNOTSUPP;
1125                         goto err_out_unlock;
1126                 }
1127
1128                 spin_lock_bh(&queue->lock);
1129                 queue->flags &= ~mask;
1130                 queue->flags |= flags & mask;
1131                 spin_unlock_bh(&queue->lock);
1132         }
1133
1134 err_out_unlock:
1135         rcu_read_unlock();
1136         return ret;
1137 }
1138
1139 static const struct nfnl_callback nfqnl_cb[NFQNL_MSG_MAX] = {
1140         [NFQNL_MSG_PACKET]      = { .call_rcu = nfqnl_recv_unsupp,
1141                                     .attr_count = NFQA_MAX, },
1142         [NFQNL_MSG_VERDICT]     = { .call_rcu = nfqnl_recv_verdict,
1143                                     .attr_count = NFQA_MAX,
1144                                     .policy = nfqa_verdict_policy },
1145         [NFQNL_MSG_CONFIG]      = { .call = nfqnl_recv_config,
1146                                     .attr_count = NFQA_CFG_MAX,
1147                                     .policy = nfqa_cfg_policy },
1148         [NFQNL_MSG_VERDICT_BATCH]={ .call_rcu = nfqnl_recv_verdict_batch,
1149                                     .attr_count = NFQA_MAX,
1150                                     .policy = nfqa_verdict_batch_policy },
1151 };
1152
1153 static const struct nfnetlink_subsystem nfqnl_subsys = {
1154         .name           = "nf_queue",
1155         .subsys_id      = NFNL_SUBSYS_QUEUE,
1156         .cb_count       = NFQNL_MSG_MAX,
1157         .cb             = nfqnl_cb,
1158 };
1159
1160 #ifdef CONFIG_PROC_FS
1161 struct iter_state {
1162         struct seq_net_private p;
1163         unsigned int bucket;
1164 };
1165
1166 static struct hlist_node *get_first(struct seq_file *seq)
1167 {
1168         struct iter_state *st = seq->private;
1169         struct net *net;
1170         struct nfnl_queue_net *q;
1171
1172         if (!st)
1173                 return NULL;
1174
1175         net = seq_file_net(seq);
1176         q = nfnl_queue_pernet(net);
1177         for (st->bucket = 0; st->bucket < INSTANCE_BUCKETS; st->bucket++) {
1178                 if (!hlist_empty(&q->instance_table[st->bucket]))
1179                         return q->instance_table[st->bucket].first;
1180         }
1181         return NULL;
1182 }
1183
1184 static struct hlist_node *get_next(struct seq_file *seq, struct hlist_node *h)
1185 {
1186         struct iter_state *st = seq->private;
1187         struct net *net = seq_file_net(seq);
1188
1189         h = h->next;
1190         while (!h) {
1191                 struct nfnl_queue_net *q;
1192
1193                 if (++st->bucket >= INSTANCE_BUCKETS)
1194                         return NULL;
1195
1196                 q = nfnl_queue_pernet(net);
1197                 h = q->instance_table[st->bucket].first;
1198         }
1199         return h;
1200 }
1201
1202 static struct hlist_node *get_idx(struct seq_file *seq, loff_t pos)
1203 {
1204         struct hlist_node *head;
1205         head = get_first(seq);
1206
1207         if (head)
1208                 while (pos && (head = get_next(seq, head)))
1209                         pos--;
1210         return pos ? NULL : head;
1211 }
1212
1213 static void *seq_start(struct seq_file *s, loff_t *pos)
1214         __acquires(nfnl_queue_pernet(seq_file_net(s))->instances_lock)
1215 {
1216         spin_lock(&nfnl_queue_pernet(seq_file_net(s))->instances_lock);
1217         return get_idx(s, *pos);
1218 }
1219
1220 static void *seq_next(struct seq_file *s, void *v, loff_t *pos)
1221 {
1222         (*pos)++;
1223         return get_next(s, v);
1224 }
1225
1226 static void seq_stop(struct seq_file *s, void *v)
1227         __releases(nfnl_queue_pernet(seq_file_net(s))->instances_lock)
1228 {
1229         spin_unlock(&nfnl_queue_pernet(seq_file_net(s))->instances_lock);
1230 }
1231
1232 static int seq_show(struct seq_file *s, void *v)
1233 {
1234         const struct nfqnl_instance *inst = v;
1235
1236         return seq_printf(s, "%5d %6d %5d %1d %5d %5d %5d %8d %2d\n",
1237                           inst->queue_num,
1238                           inst->peer_portid, inst->queue_total,
1239                           inst->copy_mode, inst->copy_range,
1240                           inst->queue_dropped, inst->queue_user_dropped,
1241                           inst->id_sequence, 1);
1242 }
1243
1244 static const struct seq_operations nfqnl_seq_ops = {
1245         .start  = seq_start,
1246         .next   = seq_next,
1247         .stop   = seq_stop,
1248         .show   = seq_show,
1249 };
1250
1251 static int nfqnl_open(struct inode *inode, struct file *file)
1252 {
1253         return seq_open_net(inode, file, &nfqnl_seq_ops,
1254                         sizeof(struct iter_state));
1255 }
1256
1257 static const struct file_operations nfqnl_file_ops = {
1258         .owner   = THIS_MODULE,
1259         .open    = nfqnl_open,
1260         .read    = seq_read,
1261         .llseek  = seq_lseek,
1262         .release = seq_release_net,
1263 };
1264
1265 #endif /* PROC_FS */
1266
1267 static int __net_init nfnl_queue_net_init(struct net *net)
1268 {
1269         unsigned int i;
1270         struct nfnl_queue_net *q = nfnl_queue_pernet(net);
1271
1272         for (i = 0; i < INSTANCE_BUCKETS; i++)
1273                 INIT_HLIST_HEAD(&q->instance_table[i]);
1274
1275         spin_lock_init(&q->instances_lock);
1276
1277 #ifdef CONFIG_PROC_FS
1278         if (!proc_create("nfnetlink_queue", 0440,
1279                          net->nf.proc_netfilter, &nfqnl_file_ops))
1280                 return -ENOMEM;
1281 #endif
1282         return 0;
1283 }
1284
1285 static void __net_exit nfnl_queue_net_exit(struct net *net)
1286 {
1287         remove_proc_entry("nfnetlink_queue", net->nf.proc_netfilter);
1288 }
1289
1290 static struct pernet_operations nfnl_queue_net_ops = {
1291         .init   = nfnl_queue_net_init,
1292         .exit   = nfnl_queue_net_exit,
1293         .id     = &nfnl_queue_net_id,
1294         .size   = sizeof(struct nfnl_queue_net),
1295 };
1296
1297 static int __init nfnetlink_queue_init(void)
1298 {
1299         int status = -ENOMEM;
1300
1301         netlink_register_notifier(&nfqnl_rtnl_notifier);
1302         status = nfnetlink_subsys_register(&nfqnl_subsys);
1303         if (status < 0) {
1304                 pr_err("nf_queue: failed to create netlink socket\n");
1305                 goto cleanup_netlink_notifier;
1306         }
1307
1308         status = register_pernet_subsys(&nfnl_queue_net_ops);
1309         if (status < 0) {
1310                 pr_err("nf_queue: failed to register pernet ops\n");
1311                 goto cleanup_subsys;
1312         }
1313         register_netdevice_notifier(&nfqnl_dev_notifier);
1314         nf_register_queue_handler(&nfqh);
1315         return status;
1316
1317 cleanup_subsys:
1318         nfnetlink_subsys_unregister(&nfqnl_subsys);
1319 cleanup_netlink_notifier:
1320         netlink_unregister_notifier(&nfqnl_rtnl_notifier);
1321         return status;
1322 }
1323
1324 static void __exit nfnetlink_queue_fini(void)
1325 {
1326         nf_unregister_queue_handler();
1327         unregister_netdevice_notifier(&nfqnl_dev_notifier);
1328         unregister_pernet_subsys(&nfnl_queue_net_ops);
1329         nfnetlink_subsys_unregister(&nfqnl_subsys);
1330         netlink_unregister_notifier(&nfqnl_rtnl_notifier);
1331
1332         rcu_barrier(); /* Wait for completion of call_rcu()'s */
1333 }
1334
1335 MODULE_DESCRIPTION("netfilter packet queue handler");
1336 MODULE_AUTHOR("Harald Welte <laforge@netfilter.org>");
1337 MODULE_LICENSE("GPL");
1338 MODULE_ALIAS_NFNL_SUBSYS(NFNL_SUBSYS_QUEUE);
1339
1340 module_init(nfnetlink_queue_init);
1341 module_exit(nfnetlink_queue_fini);