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 (!(queue->flags & NFQA_CFG_F_GSO) &&
 331                     entskb->ip_summed == CHECKSUM_PARTIAL &&
 332                     skb_checksum_help(entskb))
 333                         return NULL;
 334
 335                 data_len = ACCESS_ONCE(queue->copy_range);
 336                 if (data_len == 0 || data_len > entskb->len)
 337                         data_len = entskb->len;
 338
 339
 340                 if (!entskb->head_frag ||
 341                     skb_headlen(entskb) < L1_CACHE_BYTES ||
 342                     skb_shinfo(entskb)->nr_frags >= MAX_SKB_FRAGS)
 343                         hlen = skb_headlen(entskb);
 344
 345                 if (skb_has_frag_list(entskb))
 346                         hlen = entskb->len;
 347                 hlen = min_t(int, data_len, hlen);
 348                 size += sizeof(struct nlattr) + hlen;
 349                 cap_len = entskb->len;
 350                 break;
 351         }
 352
 353         if (queue->flags & NFQA_CFG_F_CONNTRACK)
 354                 ct = nfqnl_ct_get(entskb, &size, &ctinfo);
 355
 356         skb = nfnetlink_alloc_skb(&init_net, size, queue->peer_portid,
 357                                   GFP_ATOMIC);
 358         if (!skb)
 359                 return NULL;
 360
 361         nlh = nlmsg_put(skb, 0, 0,
 362                         NFNL_SUBSYS_QUEUE << 8 | NFQNL_MSG_PACKET,
 363                         sizeof(struct nfgenmsg), 0);
 364         if (!nlh) {
 365                 kfree_skb(skb);
 366                 return NULL;
 367         }
 368         nfmsg = nlmsg_data(nlh);
 369         nfmsg->nfgen_family = entry->pf;
 370         nfmsg->version = NFNETLINK_V0;
 371         nfmsg->res_id = htons(queue->queue_num);
 372
 373         nla = __nla_reserve(skb, NFQA_PACKET_HDR, sizeof(*pmsg));
 374         pmsg = nla_data(nla);
 375         pmsg->hw_protocol       = entskb->protocol;
 376         pmsg->hook              = entry->hook;
 377         *packet_id_ptr          = &pmsg->packet_id;
 378
 379         indev = entry->indev;
 380         if (indev) {
 381 #ifndef CONFIG_BRIDGE_NETFILTER
 382                 if (nla_put_be32(skb, NFQA_IFINDEX_INDEV, htonl(indev->ifindex)))
 383                         goto nla_put_failure;
 384 #else
 385                 if (entry->pf == PF_BRIDGE) {
 386                         /* Case 1: indev is physical input device, we need to
 387                          * look for bridge group (when called from
 388                          * netfilter_bridge) */
 389                         if (nla_put_be32(skb, NFQA_IFINDEX_PHYSINDEV,
 390                                          htonl(indev->ifindex)) ||
 391                         /* this is the bridge group "brX" */
 392                         /* rcu_read_lock()ed by __nf_queue */
 393                             nla_put_be32(skb, NFQA_IFINDEX_INDEV,
 394                                          htonl(br_port_get_rcu(indev)->br->dev->ifindex)))
 395                                 goto nla_put_failure;
 396                 } else {
 397                         /* Case 2: indev is bridge group, we need to look for
 398                          * physical device (when called from ipv4) */
 399                         if (nla_put_be32(skb, NFQA_IFINDEX_INDEV,
 400                                          htonl(indev->ifindex)))
 401                                 goto nla_put_failure;
 402                         if (entskb->nf_bridge && entskb->nf_bridge->physindev &&
 403                             nla_put_be32(skb, NFQA_IFINDEX_PHYSINDEV,
 404                                          htonl(entskb->nf_bridge->physindev->ifindex)))
 405                                 goto nla_put_failure;
 406                 }
 407 #endif
 408         }
 409
 410         if (outdev) {
 411 #ifndef CONFIG_BRIDGE_NETFILTER
 412                 if (nla_put_be32(skb, NFQA_IFINDEX_OUTDEV, htonl(outdev->ifindex)))
 413                         goto nla_put_failure;
 414 #else
 415                 if (entry->pf == PF_BRIDGE) {
 416                         /* Case 1: outdev is physical output device, we need to
 417                          * look for bridge group (when called from
 418                          * netfilter_bridge) */
 419                         if (nla_put_be32(skb, NFQA_IFINDEX_PHYSOUTDEV,
 420                                          htonl(outdev->ifindex)) ||
 421                         /* this is the bridge group "brX" */
 422                         /* rcu_read_lock()ed by __nf_queue */
 423                             nla_put_be32(skb, NFQA_IFINDEX_OUTDEV,
 424                                          htonl(br_port_get_rcu(outdev)->br->dev->ifindex)))
 425                                 goto nla_put_failure;
 426                 } else {
 427                         /* Case 2: outdev is bridge group, we need to look for
 428                          * physical output device (when called from ipv4) */
 429                         if (nla_put_be32(skb, NFQA_IFINDEX_OUTDEV,
 430                                          htonl(outdev->ifindex)))
 431                                 goto nla_put_failure;
 432                         if (entskb->nf_bridge && entskb->nf_bridge->physoutdev &&
 433                             nla_put_be32(skb, NFQA_IFINDEX_PHYSOUTDEV,
 434                                          htonl(entskb->nf_bridge->physoutdev->ifindex)))
 435                                 goto nla_put_failure;
 436                 }
 437 #endif
 438         }
 439
 440         if (entskb->mark &&
 441             nla_put_be32(skb, NFQA_MARK, htonl(entskb->mark)))
 442                 goto nla_put_failure;
 443
 444         if (indev && entskb->dev &&
 445             entskb->mac_header != entskb->network_header) {
 446                 struct nfqnl_msg_packet_hw phw;
 447                 int len = dev_parse_header(entskb, phw.hw_addr);
 448                 if (len) {
 449                         phw.hw_addrlen = htons(len);
 450                         if (nla_put(skb, NFQA_HWADDR, sizeof(phw), &phw))
 451                                 goto nla_put_failure;
 452                 }
 453         }
 454
 455         if (entskb->tstamp.tv64) {
 456                 struct nfqnl_msg_packet_timestamp ts;
 457                 struct timeval tv = ktime_to_timeval(entskb->tstamp);
 458                 ts.sec = cpu_to_be64(tv.tv_sec);
 459                 ts.usec = cpu_to_be64(tv.tv_usec);
 460
 461                 if (nla_put(skb, NFQA_TIMESTAMP, sizeof(ts), &ts))
 462                         goto nla_put_failure;
 463         }
 464
 465         if (ct && nfqnl_ct_put(skb, ct, ctinfo) < 0)
 466                 goto nla_put_failure;
 467
 468         if (cap_len > 0 && nla_put_be32(skb, NFQA_CAP_LEN, htonl(cap_len)))
 469                 goto nla_put_failure;
 470
 471         if (nfqnl_put_packet_info(skb, entskb))
 472                 goto nla_put_failure;
 473
 474         if (data_len) {
 475                 struct nlattr *nla;
 476
 477                 if (skb_tailroom(skb) < sizeof(*nla) + hlen)
 478                         goto nla_put_failure;
 479
 480                 nla = (struct nlattr *)skb_put(skb, sizeof(*nla));
 481                 nla->nla_type = NFQA_PAYLOAD;
 482                 nla->nla_len = nla_attr_size(data_len);
 483
 484                 nfqnl_zcopy(skb, entskb, data_len, hlen);
 485         }
 486
 487         nlh->nlmsg_len = skb->len;
 488         return skb;
 489
 490 nla_put_failure:
 491         kfree_skb(skb);
 492         net_err_ratelimited("nf_queue: error creating packet message\n");
 493         return NULL;
 494 }
 495
 496 static int
 497 __nfqnl_enqueue_packet(struct net *net, struct nfqnl_instance *queue,
 498                         struct nf_queue_entry *entry)
 499 {
 500         struct sk_buff *nskb;
 501         int err = -ENOBUFS;
 502         __be32 *packet_id_ptr;
 503         int failopen = 0;
 504
 505         nskb = nfqnl_build_packet_message(queue, entry, &packet_id_ptr);
 506         if (nskb == NULL) {
 507                 err = -ENOMEM;
 508                 goto err_out;
 509         }
 510         spin_lock_bh(&queue->lock);
 511
 512         if (!queue->peer_portid) {
 513                 err = -EINVAL;
 514                 goto err_out_free_nskb;
 515         }
 516         if (queue->queue_total >= queue->queue_maxlen) {
 517                 if (queue->flags & NFQA_CFG_F_FAIL_OPEN) {
 518                         failopen = 1;
 519                         err = 0;
 520                 } else {
 521                         queue->queue_dropped++;
 522                         net_warn_ratelimited("nf_queue: full at %d entries, dropping packets(s)\n",
 523                                              queue->queue_total);
 524                 }
 525                 goto err_out_free_nskb;
 526         }
 527         entry->id = ++queue->id_sequence;
 528         *packet_id_ptr = htonl(entry->id);
 529
 530         /* nfnetlink_unicast will either free the nskb or add it to a socket */
 531         err = nfnetlink_unicast(nskb, net, queue->peer_portid, MSG_DONTWAIT);
 532         if (err < 0) {
 533                 queue->queue_user_dropped++;
 534                 goto err_out_unlock;
 535         }
 536
 537         __enqueue_entry(queue, entry);
 538
 539         spin_unlock_bh(&queue->lock);
 540         return 0;
 541
 542 err_out_free_nskb:
 543         kfree_skb(nskb);
 544 err_out_unlock:
 545         spin_unlock_bh(&queue->lock);
 546         if (failopen)
 547                 nf_reinject(entry, NF_ACCEPT);
 548 err_out:
 549         return err;
 550 }
 551
 552 static struct nf_queue_entry *
 553 nf_queue_entry_dup(struct nf_queue_entry *e)
 554 {
 555         struct nf_queue_entry *entry = kmemdup(e, e->size, GFP_ATOMIC);
 556         if (entry) {
 557                 if (nf_queue_entry_get_refs(entry))
 558                         return entry;
 559                 kfree(entry);
 560         }
 561         return NULL;
 562 }
 563
 564 #ifdef CONFIG_BRIDGE_NETFILTER
 565 /* When called from bridge netfilter, skb->data must point to MAC header
 566  * before calling skb_gso_segment(). Else, original MAC header is lost
 567  * and segmented skbs will be sent to wrong destination.
 568  */
 569 static void nf_bridge_adjust_skb_data(struct sk_buff *skb)
 570 {
 571         if (skb->nf_bridge)
 572                 __skb_push(skb, skb->network_header - skb->mac_header);
 573 }
 574
 575 static void nf_bridge_adjust_segmented_data(struct sk_buff *skb)
 576 {
 577         if (skb->nf_bridge)
 578                 __skb_pull(skb, skb->network_header - skb->mac_header);
 579 }
 580 #else
 581 #define nf_bridge_adjust_skb_data(s) do {} while (0)
 582 #define nf_bridge_adjust_segmented_data(s) do {} while (0)
 583 #endif
 584
 585 static void free_entry(struct nf_queue_entry *entry)
 586 {
 587         nf_queue_entry_release_refs(entry);
 588         kfree(entry);
 589 }
 590
 591 static int
 592 __nfqnl_enqueue_packet_gso(struct net *net, struct nfqnl_instance *queue,
 593                            struct sk_buff *skb, struct nf_queue_entry *entry)
 594 {
 595         int ret = -ENOMEM;
 596         struct nf_queue_entry *entry_seg;
 597
 598         nf_bridge_adjust_segmented_data(skb);
 599
 600         if (skb->next == NULL) { /* last packet, no need to copy entry */
 601                 struct sk_buff *gso_skb = entry->skb;
 602                 entry->skb = skb;
 603                 ret = __nfqnl_enqueue_packet(net, queue, entry);
 604                 if (ret)
 605                         entry->skb = gso_skb;
 606                 return ret;
 607         }
 608
 609         skb->next = NULL;
 610
 611         entry_seg = nf_queue_entry_dup(entry);
 612         if (entry_seg) {
 613                 entry_seg->skb = skb;
 614                 ret = __nfqnl_enqueue_packet(net, queue, entry_seg);
 615                 if (ret)
 616                         free_entry(entry_seg);
 617         }
 618         return ret;
 619 }
 620
 621 static int
 622 nfqnl_enqueue_packet(struct nf_queue_entry *entry, unsigned int queuenum)
 623 {
 624         unsigned int queued;
 625         struct nfqnl_instance *queue;
 626         struct sk_buff *skb, *segs;
 627         int err = -ENOBUFS;
 628         struct net *net = dev_net(entry->indev ?
 629                                   entry->indev : entry->outdev);
 630         struct nfnl_queue_net *q = nfnl_queue_pernet(net);
 631
 632         /* rcu_read_lock()ed by nf_hook_slow() */
 633         queue = instance_lookup(q, queuenum);
 634         if (!queue)
 635                 return -ESRCH;
 636
 637         if (queue->copy_mode == NFQNL_COPY_NONE)
 638                 return -EINVAL;
 639
 640         skb = entry->skb;
 641
 642         switch (entry->pf) {
 643         case NFPROTO_IPV4:
 644                 skb->protocol = htons(ETH_P_IP);
 645                 break;
 646         case NFPROTO_IPV6:
 647                 skb->protocol = htons(ETH_P_IPV6);
 648                 break;
 649         }
 650
 651         if ((queue->flags & NFQA_CFG_F_GSO) || !skb_is_gso(skb))
 652                 return __nfqnl_enqueue_packet(net, queue, entry);
 653
 654         nf_bridge_adjust_skb_data(skb);
 655         segs = skb_gso_segment(skb, 0);
 656         /* Does not use PTR_ERR to limit the number of error codes that can be
 657          * returned by nf_queue.  For instance, callers rely on -ECANCELED to
 658          * mean 'ignore this hook'.
 659          */
 660         if (IS_ERR(segs))
 661                 goto out_err;
 662         queued = 0;
 663         err = 0;
 664         do {
 665                 struct sk_buff *nskb = segs->next;
 666                 if (err == 0)
 667                         err = __nfqnl_enqueue_packet_gso(net, queue,
 668                                                         segs, entry);
 669                 if (err == 0)
 670                         queued++;
 671                 else
 672                         kfree_skb(segs);
 673                 segs = nskb;
 674         } while (segs);
 675
 676         if (queued) {
 677                 if (err) /* some segments are already queued */
 678                         free_entry(entry);
 679                 kfree_skb(skb);
 680                 return 0;
 681         }
 682  out_err:
 683         nf_bridge_adjust_segmented_data(skb);
 684         return err;
 685 }
 686
 687 static int
 688 nfqnl_mangle(void *data, int data_len, struct nf_queue_entry *e, int diff)
 689 {
 690         struct sk_buff *nskb;
 691
 692         if (diff < 0) {
 693                 if (pskb_trim(e->skb, data_len))
 694                         return -ENOMEM;
 695         } else if (diff > 0) {
 696                 if (data_len > 0xFFFF)
 697                         return -EINVAL;
 698                 if (diff > skb_tailroom(e->skb)) {
 699                         nskb = skb_copy_expand(e->skb, skb_headroom(e->skb),
 700                                                diff, GFP_ATOMIC);
 701                         if (!nskb) {
 702                                 printk(KERN_WARNING "nf_queue: OOM "
 703                                       "in mangle, dropping packet\n");
 704                                 return -ENOMEM;
 705                         }
 706                         kfree_skb(e->skb);
 707                         e->skb = nskb;
 708                 }
 709                 skb_put(e->skb, diff);
 710         }
 711         if (!skb_make_writable(e->skb, data_len))
 712                 return -ENOMEM;
 713         skb_copy_to_linear_data(e->skb, data, data_len);
 714         e->skb->ip_summed = CHECKSUM_NONE;
 715         return 0;
 716 }
 717
 718 static int
 719 nfqnl_set_mode(struct nfqnl_instance *queue,
 720                unsigned char mode, unsigned int range)
 721 {
 722         int status = 0;
 723
 724         spin_lock_bh(&queue->lock);
 725         switch (mode) {
 726         case NFQNL_COPY_NONE:
 727         case NFQNL_COPY_META:
 728                 queue->copy_mode = mode;
 729                 queue->copy_range = 0;
 730                 break;
 731
 732         case NFQNL_COPY_PACKET:
 733                 queue->copy_mode = mode;
 734                 /* We're using struct nlattr which has 16bit nla_len. Note that
 735                  * nla_len includes the header length. Thus, the maximum packet
 736                  * length that we support is 65531 bytes. We send truncated
 737                  * packets if the specified length is larger than that.
 738                  */
 739                 if (range > 0xffff - NLA_HDRLEN)
 740                         queue->copy_range = 0xffff - NLA_HDRLEN;
 741                 else
 742                         queue->copy_range = range;
 743                 break;
 744
 745         default:
 746                 status = -EINVAL;
 747
 748         }
 749         spin_unlock_bh(&queue->lock);
 750
 751         return status;
 752 }
 753
 754 static int
 755 dev_cmp(struct nf_queue_entry *entry, unsigned long ifindex)
 756 {
 757         if (entry->indev)
 758                 if (entry->indev->ifindex == ifindex)
 759                         return 1;
 760         if (entry->outdev)
 761                 if (entry->outdev->ifindex == ifindex)
 762                         return 1;
 763 #ifdef CONFIG_BRIDGE_NETFILTER
 764         if (entry->skb->nf_bridge) {
 765                 if (entry->skb->nf_bridge->physindev &&
 766                     entry->skb->nf_bridge->physindev->ifindex == ifindex)
 767                         return 1;
 768                 if (entry->skb->nf_bridge->physoutdev &&
 769                     entry->skb->nf_bridge->physoutdev->ifindex == ifindex)
 770                         return 1;
 771         }
 772 #endif
 773         return 0;
 774 }
 775
 776 /* drop all packets with either indev or outdev == ifindex from all queue
 777  * instances */
 778 static void
 779 nfqnl_dev_drop(struct net *net, int ifindex)
 780 {
 781         int i;
 782         struct nfnl_queue_net *q = nfnl_queue_pernet(net);
 783
 784         rcu_read_lock();
 785
 786         for (i = 0; i < INSTANCE_BUCKETS; i++) {
 787                 struct nfqnl_instance *inst;
 788                 struct hlist_head *head = &q->instance_table[i];
 789
 790                 hlist_for_each_entry_rcu(inst, head, hlist)
 791                         nfqnl_flush(inst, dev_cmp, ifindex);
 792         }
 793
 794         rcu_read_unlock();
 795 }
 796
 797 #define RCV_SKB_FAIL(err) do { netlink_ack(skb, nlh, (err)); return; } while (0)
 798
 799 static int
 800 nfqnl_rcv_dev_event(struct notifier_block *this,
 801                     unsigned long event, void *ptr)
 802 {
 803         struct net_device *dev = ptr;
 804
 805         /* Drop any packets associated with the downed device */
 806         if (event == NETDEV_DOWN)
 807                 nfqnl_dev_drop(dev_net(dev), dev->ifindex);
 808         return NOTIFY_DONE;
 809 }
 810
 811 static struct notifier_block nfqnl_dev_notifier = {
 812         .notifier_call  = nfqnl_rcv_dev_event,
 813 };
 814
 815 static int
 816 nfqnl_rcv_nl_event(struct notifier_block *this,
 817                    unsigned long event, void *ptr)
 818 {
 819         struct netlink_notify *n = ptr;
 820         struct nfnl_queue_net *q = nfnl_queue_pernet(n->net);
 821
 822         if (event == NETLINK_URELEASE && n->protocol == NETLINK_NETFILTER) {
 823                 int i;
 824
 825                 /* destroy all instances for this portid */
 826                 spin_lock(&q->instances_lock);
 827                 for (i = 0; i < INSTANCE_BUCKETS; i++) {
 828                         struct hlist_node *t2;
 829                         struct nfqnl_instance *inst;
 830                         struct hlist_head *head = &q->instance_table[i];
 831
 832                         hlist_for_each_entry_safe(inst, t2, head, hlist) {
 833                                 if (n->portid == inst->peer_portid)
 834                                         __instance_destroy(inst);
 835                         }
 836                 }
 837                 spin_unlock(&q->instances_lock);
 838         }
 839         return NOTIFY_DONE;
 840 }
 841
 842 static struct notifier_block nfqnl_rtnl_notifier = {
 843         .notifier_call  = nfqnl_rcv_nl_event,
 844 };
 845
 846 static const struct nla_policy nfqa_verdict_policy[NFQA_MAX+1] = {
 847         [NFQA_VERDICT_HDR]      = { .len = sizeof(struct nfqnl_msg_verdict_hdr) },
 848         [NFQA_MARK]             = { .type = NLA_U32 },
 849         [NFQA_PAYLOAD]          = { .type = NLA_UNSPEC },
 850         [NFQA_CT]               = { .type = NLA_UNSPEC },
 851 };
 852
 853 static const struct nla_policy nfqa_verdict_batch_policy[NFQA_MAX+1] = {
 854         [NFQA_VERDICT_HDR]      = { .len = sizeof(struct nfqnl_msg_verdict_hdr) },
 855         [NFQA_MARK]             = { .type = NLA_U32 },
 856 };
 857
 858 static struct nfqnl_instance *
 859 verdict_instance_lookup(struct nfnl_queue_net *q, u16 queue_num, int nlportid)
 860 {
 861         struct nfqnl_instance *queue;
 862
 863         queue = instance_lookup(q, queue_num);
 864         if (!queue)
 865                 return ERR_PTR(-ENODEV);
 866
 867         if (queue->peer_portid != nlportid)
 868                 return ERR_PTR(-EPERM);
 869
 870         return queue;
 871 }
 872
 873 static struct nfqnl_msg_verdict_hdr*
 874 verdicthdr_get(const struct nlattr * const nfqa[])
 875 {
 876         struct nfqnl_msg_verdict_hdr *vhdr;
 877         unsigned int verdict;
 878
 879         if (!nfqa[NFQA_VERDICT_HDR])
 880                 return NULL;
 881
 882         vhdr = nla_data(nfqa[NFQA_VERDICT_HDR]);
 883         verdict = ntohl(vhdr->verdict) & NF_VERDICT_MASK;
 884         if (verdict > NF_MAX_VERDICT || verdict == NF_STOLEN)
 885                 return NULL;
 886         return vhdr;
 887 }
 888
 889 static int nfq_id_after(unsigned int id, unsigned int max)
 890 {
 891         return (int)(id - max) > 0;
 892 }
 893
 894 static int
 895 nfqnl_recv_verdict_batch(struct sock *ctnl, struct sk_buff *skb,
 896                    const struct nlmsghdr *nlh,
 897                    const struct nlattr * const nfqa[])
 898 {
 899         struct nfgenmsg *nfmsg = nlmsg_data(nlh);
 900         struct nf_queue_entry *entry, *tmp;
 901         unsigned int verdict, maxid;
 902         struct nfqnl_msg_verdict_hdr *vhdr;
 903         struct nfqnl_instance *queue;
 904         LIST_HEAD(batch_list);
 905         u16 queue_num = ntohs(nfmsg->res_id);
 906
 907         struct net *net = sock_net(ctnl);
 908         struct nfnl_queue_net *q = nfnl_queue_pernet(net);
 909
 910         queue = verdict_instance_lookup(q, queue_num,
 911                                         NETLINK_CB(skb).portid);
 912         if (IS_ERR(queue))
 913                 return PTR_ERR(queue);
 914
 915         vhdr = verdicthdr_get(nfqa);
 916         if (!vhdr)
 917                 return -EINVAL;
 918
 919         verdict = ntohl(vhdr->verdict);
 920         maxid = ntohl(vhdr->id);
 921
 922         spin_lock_bh(&queue->lock);
 923
 924         list_for_each_entry_safe(entry, tmp, &queue->queue_list, list) {
 925                 if (nfq_id_after(entry->id, maxid))
 926                         break;
 927                 __dequeue_entry(queue, entry);
 928                 list_add_tail(&entry->list, &batch_list);
 929         }
 930
 931         spin_unlock_bh(&queue->lock);
 932
 933         if (list_empty(&batch_list))
 934                 return -ENOENT;
 935
 936         list_for_each_entry_safe(entry, tmp, &batch_list, list) {
 937                 if (nfqa[NFQA_MARK])
 938                         entry->skb->mark = ntohl(nla_get_be32(nfqa[NFQA_MARK]));
 939                 nf_reinject(entry, verdict);
 940         }
 941         return 0;
 942 }
 943
 944 static int
 945 nfqnl_recv_verdict(struct sock *ctnl, struct sk_buff *skb,
 946                    const struct nlmsghdr *nlh,
 947                    const struct nlattr * const nfqa[])
 948 {
 949         struct nfgenmsg *nfmsg = nlmsg_data(nlh);
 950         u_int16_t queue_num = ntohs(nfmsg->res_id);
 951
 952         struct nfqnl_msg_verdict_hdr *vhdr;
 953         struct nfqnl_instance *queue;
 954         unsigned int verdict;
 955         struct nf_queue_entry *entry;
 956         enum ip_conntrack_info uninitialized_var(ctinfo);
 957         struct nf_conn *ct = NULL;
 958
 959         struct net *net = sock_net(ctnl);
 960         struct nfnl_queue_net *q = nfnl_queue_pernet(net);
 961
 962         queue = instance_lookup(q, queue_num);
 963         if (!queue)
 964                 queue = verdict_instance_lookup(q, queue_num,
 965                                                 NETLINK_CB(skb).portid);
 966         if (IS_ERR(queue))
 967                 return PTR_ERR(queue);
 968
 969         vhdr = verdicthdr_get(nfqa);
 970         if (!vhdr)
 971                 return -EINVAL;
 972
 973         verdict = ntohl(vhdr->verdict);
 974
 975         entry = find_dequeue_entry(queue, ntohl(vhdr->id));
 976         if (entry == NULL)
 977                 return -ENOENT;
 978
 979         rcu_read_lock();
 980         if (nfqa[NFQA_CT] && (queue->flags & NFQA_CFG_F_CONNTRACK))
 981                 ct = nfqnl_ct_parse(entry->skb, nfqa[NFQA_CT], &ctinfo);
 982
 983         if (nfqa[NFQA_PAYLOAD]) {
 984                 u16 payload_len = nla_len(nfqa[NFQA_PAYLOAD]);
 985                 int diff = payload_len - entry->skb->len;
 986
 987                 if (nfqnl_mangle(nla_data(nfqa[NFQA_PAYLOAD]),
 988                                  payload_len, entry, diff) < 0)
 989                         verdict = NF_DROP;
 990
 991                 if (ct)
 992                         nfqnl_ct_seq_adjust(skb, ct, ctinfo, diff);
 993         }
 994         rcu_read_unlock();
 995
 996         if (nfqa[NFQA_MARK])
 997                 entry->skb->mark = ntohl(nla_get_be32(nfqa[NFQA_MARK]));
 998
 999         nf_reinject(entry, verdict);
1000         return 0;
1001 }
1002
1003 static int
1004 nfqnl_recv_unsupp(struct sock *ctnl, struct sk_buff *skb,
1005                   const struct nlmsghdr *nlh,
1006                   const struct nlattr * const nfqa[])
1007 {
1008         return -ENOTSUPP;
1009 }
1010
1011 static const struct nla_policy nfqa_cfg_policy[NFQA_CFG_MAX+1] = {
1012         [NFQA_CFG_CMD]          = { .len = sizeof(struct nfqnl_msg_config_cmd) },
1013         [NFQA_CFG_PARAMS]       = { .len = sizeof(struct nfqnl_msg_config_params) },
1014 };
1015
1016 static const struct nf_queue_handler nfqh = {
1017         .outfn  = &nfqnl_enqueue_packet,
1018 };
1019
1020 static int
1021 nfqnl_recv_config(struct sock *ctnl, struct sk_buff *skb,
1022                   const struct nlmsghdr *nlh,
1023                   const struct nlattr * const nfqa[])
1024 {
1025         struct nfgenmsg *nfmsg = nlmsg_data(nlh);
1026         u_int16_t queue_num = ntohs(nfmsg->res_id);
1027         struct nfqnl_instance *queue;
1028         struct nfqnl_msg_config_cmd *cmd = NULL;
1029         struct net *net = sock_net(ctnl);
1030         struct nfnl_queue_net *q = nfnl_queue_pernet(net);
1031         int ret = 0;
1032
1033         if (nfqa[NFQA_CFG_CMD]) {
1034                 cmd = nla_data(nfqa[NFQA_CFG_CMD]);
1035
1036                 /* Obsolete commands without queue context */
1037                 switch (cmd->command) {
1038                 case NFQNL_CFG_CMD_PF_BIND: return 0;
1039                 case NFQNL_CFG_CMD_PF_UNBIND: return 0;
1040                 }
1041         }
1042
1043         rcu_read_lock();
1044         queue = instance_lookup(q, queue_num);
1045         if (queue && queue->peer_portid != NETLINK_CB(skb).portid) {
1046                 ret = -EPERM;
1047                 goto err_out_unlock;
1048         }
1049
1050         if (cmd != NULL) {
1051                 switch (cmd->command) {
1052                 case NFQNL_CFG_CMD_BIND:
1053                         if (queue) {
1054                                 ret = -EBUSY;
1055                                 goto err_out_unlock;
1056                         }
1057                         queue = instance_create(q, queue_num,
1058                                                 NETLINK_CB(skb).portid);
1059                         if (IS_ERR(queue)) {
1060                                 ret = PTR_ERR(queue);
1061                                 goto err_out_unlock;
1062                         }
1063                         break;
1064                 case NFQNL_CFG_CMD_UNBIND:
1065                         if (!queue) {
1066                                 ret = -ENODEV;
1067                                 goto err_out_unlock;
1068                         }
1069                         instance_destroy(q, queue);
1070                         break;
1071                 case NFQNL_CFG_CMD_PF_BIND:
1072                 case NFQNL_CFG_CMD_PF_UNBIND:
1073                         break;
1074                 default:
1075                         ret = -ENOTSUPP;
1076                         break;
1077                 }
1078         }
1079
1080         if (nfqa[NFQA_CFG_PARAMS]) {
1081                 struct nfqnl_msg_config_params *params;
1082
1083                 if (!queue) {
1084                         ret = -ENODEV;
1085                         goto err_out_unlock;
1086                 }
1087                 params = nla_data(nfqa[NFQA_CFG_PARAMS]);
1088                 nfqnl_set_mode(queue, params->copy_mode,
1089                                 ntohl(params->copy_range));
1090         }
1091
1092         if (nfqa[NFQA_CFG_QUEUE_MAXLEN]) {
1093                 __be32 *queue_maxlen;
1094
1095                 if (!queue) {
1096                         ret = -ENODEV;
1097                         goto err_out_unlock;
1098                 }
1099                 queue_maxlen = nla_data(nfqa[NFQA_CFG_QUEUE_MAXLEN]);
1100                 spin_lock_bh(&queue->lock);
1101                 queue->queue_maxlen = ntohl(*queue_maxlen);
1102                 spin_unlock_bh(&queue->lock);
1103         }
1104
1105         if (nfqa[NFQA_CFG_FLAGS]) {
1106                 __u32 flags, mask;
1107
1108                 if (!queue) {
1109                         ret = -ENODEV;
1110                         goto err_out_unlock;
1111                 }
1112
1113                 if (!nfqa[NFQA_CFG_MASK]) {
1114                         /* A mask is needed to specify which flags are being
1115                          * changed.
1116                          */
1117                         ret = -EINVAL;
1118                         goto err_out_unlock;
1119                 }
1120
1121                 flags = ntohl(nla_get_be32(nfqa[NFQA_CFG_FLAGS]));
1122                 mask = ntohl(nla_get_be32(nfqa[NFQA_CFG_MASK]));
1123
1124                 if (flags >= NFQA_CFG_F_MAX) {
1125                         ret = -EOPNOTSUPP;
1126                         goto err_out_unlock;
1127                 }
1128
1129                 spin_lock_bh(&queue->lock);
1130                 queue->flags &= ~mask;
1131                 queue->flags |= flags & mask;
1132                 spin_unlock_bh(&queue->lock);
1133         }
1134
1135 err_out_unlock:
1136         rcu_read_unlock();
1137         return ret;
1138 }
1139
1140 static const struct nfnl_callback nfqnl_cb[NFQNL_MSG_MAX] = {
1141         [NFQNL_MSG_PACKET]      = { .call_rcu = nfqnl_recv_unsupp,
1142                                     .attr_count = NFQA_MAX, },
1143         [NFQNL_MSG_VERDICT]     = { .call_rcu = nfqnl_recv_verdict,
1144                                     .attr_count = NFQA_MAX,
1145                                     .policy = nfqa_verdict_policy },
1146         [NFQNL_MSG_CONFIG]      = { .call = nfqnl_recv_config,
1147                                     .attr_count = NFQA_CFG_MAX,
1148                                     .policy = nfqa_cfg_policy },
1149         [NFQNL_MSG_VERDICT_BATCH]={ .call_rcu = nfqnl_recv_verdict_batch,
1150                                     .attr_count = NFQA_MAX,
1151                                     .policy = nfqa_verdict_batch_policy },
1152 };
1153
1154 static const struct nfnetlink_subsystem nfqnl_subsys = {
1155         .name           = "nf_queue",
1156         .subsys_id      = NFNL_SUBSYS_QUEUE,
1157         .cb_count       = NFQNL_MSG_MAX,
1158         .cb             = nfqnl_cb,
1159 };
1160
1161 #ifdef CONFIG_PROC_FS
1162 struct iter_state {
1163         struct seq_net_private p;
1164         unsigned int bucket;
1165 };
1166
1167 static struct hlist_node *get_first(struct seq_file *seq)
1168 {
1169         struct iter_state *st = seq->private;
1170         struct net *net;
1171         struct nfnl_queue_net *q;
1172
1173         if (!st)
1174                 return NULL;
1175
1176         net = seq_file_net(seq);
1177         q = nfnl_queue_pernet(net);
1178         for (st->bucket = 0; st->bucket < INSTANCE_BUCKETS; st->bucket++) {
1179                 if (!hlist_empty(&q->instance_table[st->bucket]))
1180                         return q->instance_table[st->bucket].first;
1181         }
1182         return NULL;
1183 }
1184
1185 static struct hlist_node *get_next(struct seq_file *seq, struct hlist_node *h)
1186 {
1187         struct iter_state *st = seq->private;
1188         struct net *net = seq_file_net(seq);
1189
1190         h = h->next;
1191         while (!h) {
1192                 struct nfnl_queue_net *q;
1193
1194                 if (++st->bucket >= INSTANCE_BUCKETS)
1195                         return NULL;
1196
1197                 q = nfnl_queue_pernet(net);
1198                 h = q->instance_table[st->bucket].first;
1199         }
1200         return h;
1201 }
1202
1203 static struct hlist_node *get_idx(struct seq_file *seq, loff_t pos)
1204 {
1205         struct hlist_node *head;
1206         head = get_first(seq);
1207
1208         if (head)
1209                 while (pos && (head = get_next(seq, head)))
1210                         pos--;
1211         return pos ? NULL : head;
1212 }
1213
1214 static void *seq_start(struct seq_file *s, loff_t *pos)
1215         __acquires(nfnl_queue_pernet(seq_file_net(s))->instances_lock)
1216 {
1217         spin_lock(&nfnl_queue_pernet(seq_file_net(s))->instances_lock);
1218         return get_idx(s, *pos);
1219 }
1220
1221 static void *seq_next(struct seq_file *s, void *v, loff_t *pos)
1222 {
1223         (*pos)++;
1224         return get_next(s, v);
1225 }
1226
1227 static void seq_stop(struct seq_file *s, void *v)
1228         __releases(nfnl_queue_pernet(seq_file_net(s))->instances_lock)
1229 {
1230         spin_unlock(&nfnl_queue_pernet(seq_file_net(s))->instances_lock);
1231 }
1232
1233 static int seq_show(struct seq_file *s, void *v)
1234 {
1235         const struct nfqnl_instance *inst = v;
1236
1237         return seq_printf(s, "%5d %6d %5d %1d %5d %5d %5d %8d %2d\n",
1238                           inst->queue_num,
1239                           inst->peer_portid, inst->queue_total,
1240                           inst->copy_mode, inst->copy_range,
1241                           inst->queue_dropped, inst->queue_user_dropped,
1242                           inst->id_sequence, 1);
1243 }
1244
1245 static const struct seq_operations nfqnl_seq_ops = {
1246         .start  = seq_start,
1247         .next   = seq_next,
1248         .stop   = seq_stop,
1249         .show   = seq_show,
1250 };
1251
1252 static int nfqnl_open(struct inode *inode, struct file *file)
1253 {
1254         return seq_open_net(inode, file, &nfqnl_seq_ops,
1255                         sizeof(struct iter_state));
1256 }
1257
1258 static const struct file_operations nfqnl_file_ops = {
1259         .owner   = THIS_MODULE,
1260         .open    = nfqnl_open,
1261         .read    = seq_read,
1262         .llseek  = seq_lseek,
1263         .release = seq_release_net,
1264 };
1265
1266 #endif /* PROC_FS */
1267
1268 static int __net_init nfnl_queue_net_init(struct net *net)
1269 {
1270         unsigned int i;
1271         struct nfnl_queue_net *q = nfnl_queue_pernet(net);
1272
1273         for (i = 0; i < INSTANCE_BUCKETS; i++)
1274                 INIT_HLIST_HEAD(&q->instance_table[i]);
1275
1276         spin_lock_init(&q->instances_lock);
1277
1278 #ifdef CONFIG_PROC_FS
1279         if (!proc_create("nfnetlink_queue", 0440,
1280                          net->nf.proc_netfilter, &nfqnl_file_ops))
1281                 return -ENOMEM;
1282 #endif
1283         return 0;
1284 }
1285
1286 static void __net_exit nfnl_queue_net_exit(struct net *net)
1287 {
1288 #ifdef CONFIG_PROC_FS
1289         remove_proc_entry("nfnetlink_queue", net->nf.proc_netfilter);
1290 #endif
1291 }
1292
1293 static struct pernet_operations nfnl_queue_net_ops = {
1294         .init   = nfnl_queue_net_init,
1295         .exit   = nfnl_queue_net_exit,
1296         .id     = &nfnl_queue_net_id,
1297         .size   = sizeof(struct nfnl_queue_net),
1298 };
1299
1300 static int __init nfnetlink_queue_init(void)
1301 {
1302         int status = -ENOMEM;
1303
1304         netlink_register_notifier(&nfqnl_rtnl_notifier);
1305         status = nfnetlink_subsys_register(&nfqnl_subsys);
1306         if (status < 0) {
1307                 pr_err("nf_queue: failed to create netlink socket\n");
1308                 goto cleanup_netlink_notifier;
1309         }
1310
1311         status = register_pernet_subsys(&nfnl_queue_net_ops);
1312         if (status < 0) {
1313                 pr_err("nf_queue: failed to register pernet ops\n");
1314                 goto cleanup_subsys;
1315         }
1316         register_netdevice_notifier(&nfqnl_dev_notifier);
1317         nf_register_queue_handler(&nfqh);
1318         return status;
1319
1320 cleanup_subsys:
1321         nfnetlink_subsys_unregister(&nfqnl_subsys);
1322 cleanup_netlink_notifier:
1323         netlink_unregister_notifier(&nfqnl_rtnl_notifier);
1324         return status;
1325 }
1326
1327 static void __exit nfnetlink_queue_fini(void)
1328 {
1329         nf_unregister_queue_handler();
1330         unregister_netdevice_notifier(&nfqnl_dev_notifier);
1331         unregister_pernet_subsys(&nfnl_queue_net_ops);
1332         nfnetlink_subsys_unregister(&nfqnl_subsys);
1333         netlink_unregister_notifier(&nfqnl_rtnl_notifier);
1334
1335         rcu_barrier(); /* Wait for completion of call_rcu()'s */
1336 }
1337
1338 MODULE_DESCRIPTION("netfilter packet queue handler");
1339 MODULE_AUTHOR("Harald Welte <laforge@netfilter.org>");
1340 MODULE_LICENSE("GPL");
1341 MODULE_ALIAS_NFNL_SUBSYS(NFNL_SUBSYS_QUEUE);
1342
1343 module_init(nfnetlink_queue_init);
1344 module_exit(nfnetlink_queue_fini);