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 int
 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         int ret;
 236         struct page *page;
 237         unsigned int offset;
 238
 239         /* dont bother with small payloads */
 240         if (len <= skb_tailroom(to))
 241                 return skb_copy_bits(from, 0, skb_put(to, len), len);
 242
 243         if (hlen) {
 244                 ret = skb_copy_bits(from, 0, skb_put(to, hlen), hlen);
 245                 if (unlikely(ret))
 246                         return ret;
 247                 len -= hlen;
 248         } else {
 249                 plen = min_t(int, skb_headlen(from), len);
 250                 if (plen) {
 251                         page = virt_to_head_page(from->head);
 252                         offset = from->data - (unsigned char *)page_address(page);
 253                         __skb_fill_page_desc(to, 0, page, offset, plen);
 254                         get_page(page);
 255                         j = 1;
 256                         len -= plen;
 257                 }
 258         }
 259
 260         to->truesize += len + plen;
 261         to->len += len + plen;
 262         to->data_len += len + plen;
 263
 264         if (unlikely(skb_orphan_frags(from, GFP_ATOMIC))) {
 265                 skb_tx_error(from);
 266                 return -ENOMEM;
 267         }
 268
 269         for (i = 0; i < skb_shinfo(from)->nr_frags; i++) {
 270                 if (!len)
 271                         break;
 272                 skb_shinfo(to)->frags[j] = skb_shinfo(from)->frags[i];
 273                 skb_shinfo(to)->frags[j].size = min_t(int, skb_shinfo(to)->frags[j].size, len);
 274                 len -= skb_shinfo(to)->frags[j].size;
 275                 skb_frag_ref(to, j);
 276                 j++;
 277         }
 278         skb_shinfo(to)->nr_frags = j;
 279
 280         return 0;
 281 }
 282
 283 static int nfqnl_put_packet_info(struct sk_buff *nlskb, struct sk_buff *packet)
 284 {
 285         __u32 flags = 0;
 286
 287         if (packet->ip_summed == CHECKSUM_PARTIAL)
 288                 flags = NFQA_SKB_CSUMNOTREADY;
 289         if (skb_is_gso(packet))
 290                 flags |= NFQA_SKB_GSO;
 291
 292         return flags ? nla_put_be32(nlskb, NFQA_SKB_INFO, htonl(flags)) : 0;
 293 }
 294
 295 static struct sk_buff *
 296 nfqnl_build_packet_message(struct nfqnl_instance *queue,
 297                            struct nf_queue_entry *entry,
 298                            __be32 **packet_id_ptr)
 299 {
 300         size_t size;
 301         size_t data_len = 0, cap_len = 0;
 302         int hlen = 0;
 303         struct sk_buff *skb;
 304         struct nlattr *nla;
 305         struct nfqnl_msg_packet_hdr *pmsg;
 306         struct nlmsghdr *nlh;
 307         struct nfgenmsg *nfmsg;
 308         struct sk_buff *entskb = entry->skb;
 309         struct net_device *indev;
 310         struct net_device *outdev;
 311         struct nf_conn *ct = NULL;
 312         enum ip_conntrack_info uninitialized_var(ctinfo);
 313
 314         size =    nlmsg_total_size(sizeof(struct nfgenmsg))
 315                 + nla_total_size(sizeof(struct nfqnl_msg_packet_hdr))
 316                 + nla_total_size(sizeof(u_int32_t))     /* ifindex */
 317                 + nla_total_size(sizeof(u_int32_t))     /* ifindex */
 318 #ifdef CONFIG_BRIDGE_NETFILTER
 319                 + nla_total_size(sizeof(u_int32_t))     /* ifindex */
 320                 + nla_total_size(sizeof(u_int32_t))     /* ifindex */
 321 #endif
 322                 + nla_total_size(sizeof(u_int32_t))     /* mark */
 323                 + nla_total_size(sizeof(struct nfqnl_msg_packet_hw))
 324                 + nla_total_size(sizeof(u_int32_t))     /* skbinfo */
 325                 + nla_total_size(sizeof(u_int32_t));    /* cap_len */
 326
 327         if (entskb->tstamp.tv64)
 328                 size += nla_total_size(sizeof(struct nfqnl_msg_packet_timestamp));
 329
 330         outdev = entry->outdev;
 331
 332         switch ((enum nfqnl_config_mode)ACCESS_ONCE(queue->copy_mode)) {
 333         case NFQNL_COPY_META:
 334         case NFQNL_COPY_NONE:
 335                 break;
 336
 337         case NFQNL_COPY_PACKET:
 338                 if (!(queue->flags & NFQA_CFG_F_GSO) &&
 339                     entskb->ip_summed == CHECKSUM_PARTIAL &&
 340                     skb_checksum_help(entskb))
 341                         return NULL;
 342
 343                 data_len = ACCESS_ONCE(queue->copy_range);
 344                 if (data_len == 0 || data_len > entskb->len)
 345                         data_len = entskb->len;
 346
 347
 348                 if (!entskb->head_frag ||
 349                     skb_headlen(entskb) < L1_CACHE_BYTES ||
 350                     skb_shinfo(entskb)->nr_frags >= MAX_SKB_FRAGS)
 351                         hlen = skb_headlen(entskb);
 352
 353                 if (skb_has_frag_list(entskb))
 354                         hlen = entskb->len;
 355                 hlen = min_t(int, data_len, hlen);
 356                 size += sizeof(struct nlattr) + hlen;
 357                 cap_len = entskb->len;
 358                 break;
 359         }
 360
 361         if (queue->flags & NFQA_CFG_F_CONNTRACK)
 362                 ct = nfqnl_ct_get(entskb, &size, &ctinfo);
 363
 364         skb = nfnetlink_alloc_skb(&init_net, size, queue->peer_portid,
 365                                   GFP_ATOMIC);
 366         if (!skb) {
 367                 skb_tx_error(entskb);
 368                 return NULL;
 369         }
 370
 371         nlh = nlmsg_put(skb, 0, 0,
 372                         NFNL_SUBSYS_QUEUE << 8 | NFQNL_MSG_PACKET,
 373                         sizeof(struct nfgenmsg), 0);
 374         if (!nlh) {
 375                 skb_tx_error(entskb);
 376                 kfree_skb(skb);
 377                 return NULL;
 378         }
 379         nfmsg = nlmsg_data(nlh);
 380         nfmsg->nfgen_family = entry->pf;
 381         nfmsg->version = NFNETLINK_V0;
 382         nfmsg->res_id = htons(queue->queue_num);
 383
 384         nla = __nla_reserve(skb, NFQA_PACKET_HDR, sizeof(*pmsg));
 385         pmsg = nla_data(nla);
 386         pmsg->hw_protocol       = entskb->protocol;
 387         pmsg->hook              = entry->hook;
 388         *packet_id_ptr          = &pmsg->packet_id;
 389
 390         indev = entry->indev;
 391         if (indev) {
 392 #ifndef CONFIG_BRIDGE_NETFILTER
 393                 if (nla_put_be32(skb, NFQA_IFINDEX_INDEV, htonl(indev->ifindex)))
 394                         goto nla_put_failure;
 395 #else
 396                 if (entry->pf == PF_BRIDGE) {
 397                         /* Case 1: indev is physical input device, we need to
 398                          * look for bridge group (when called from
 399                          * netfilter_bridge) */
 400                         if (nla_put_be32(skb, NFQA_IFINDEX_PHYSINDEV,
 401                                          htonl(indev->ifindex)) ||
 402                         /* this is the bridge group "brX" */
 403                         /* rcu_read_lock()ed by __nf_queue */
 404                             nla_put_be32(skb, NFQA_IFINDEX_INDEV,
 405                                          htonl(br_port_get_rcu(indev)->br->dev->ifindex)))
 406                                 goto nla_put_failure;
 407                 } else {
 408                         /* Case 2: indev is bridge group, we need to look for
 409                          * physical device (when called from ipv4) */
 410                         if (nla_put_be32(skb, NFQA_IFINDEX_INDEV,
 411                                          htonl(indev->ifindex)))
 412                                 goto nla_put_failure;
 413                         if (entskb->nf_bridge && entskb->nf_bridge->physindev &&
 414                             nla_put_be32(skb, NFQA_IFINDEX_PHYSINDEV,
 415                                          htonl(entskb->nf_bridge->physindev->ifindex)))
 416                                 goto nla_put_failure;
 417                 }
 418 #endif
 419         }
 420
 421         if (outdev) {
 422 #ifndef CONFIG_BRIDGE_NETFILTER
 423                 if (nla_put_be32(skb, NFQA_IFINDEX_OUTDEV, htonl(outdev->ifindex)))
 424                         goto nla_put_failure;
 425 #else
 426                 if (entry->pf == PF_BRIDGE) {
 427                         /* Case 1: outdev is physical output device, we need to
 428                          * look for bridge group (when called from
 429                          * netfilter_bridge) */
 430                         if (nla_put_be32(skb, NFQA_IFINDEX_PHYSOUTDEV,
 431                                          htonl(outdev->ifindex)) ||
 432                         /* this is the bridge group "brX" */
 433                         /* rcu_read_lock()ed by __nf_queue */
 434                             nla_put_be32(skb, NFQA_IFINDEX_OUTDEV,
 435                                          htonl(br_port_get_rcu(outdev)->br->dev->ifindex)))
 436                                 goto nla_put_failure;
 437                 } else {
 438                         /* Case 2: outdev is bridge group, we need to look for
 439                          * physical output device (when called from ipv4) */
 440                         if (nla_put_be32(skb, NFQA_IFINDEX_OUTDEV,
 441                                          htonl(outdev->ifindex)))
 442                                 goto nla_put_failure;
 443                         if (entskb->nf_bridge && entskb->nf_bridge->physoutdev &&
 444                             nla_put_be32(skb, NFQA_IFINDEX_PHYSOUTDEV,
 445                                          htonl(entskb->nf_bridge->physoutdev->ifindex)))
 446                                 goto nla_put_failure;
 447                 }
 448 #endif
 449         }
 450
 451         if (entskb->mark &&
 452             nla_put_be32(skb, NFQA_MARK, htonl(entskb->mark)))
 453                 goto nla_put_failure;
 454
 455         if (indev && entskb->dev &&
 456             entskb->mac_header != entskb->network_header) {
 457                 struct nfqnl_msg_packet_hw phw;
 458                 int len = dev_parse_header(entskb, phw.hw_addr);
 459                 if (len) {
 460                         phw.hw_addrlen = htons(len);
 461                         if (nla_put(skb, NFQA_HWADDR, sizeof(phw), &phw))
 462                                 goto nla_put_failure;
 463                 }
 464         }
 465
 466         if (entskb->tstamp.tv64) {
 467                 struct nfqnl_msg_packet_timestamp ts;
 468                 struct timeval tv = ktime_to_timeval(entskb->tstamp);
 469                 ts.sec = cpu_to_be64(tv.tv_sec);
 470                 ts.usec = cpu_to_be64(tv.tv_usec);
 471
 472                 if (nla_put(skb, NFQA_TIMESTAMP, sizeof(ts), &ts))
 473                         goto nla_put_failure;
 474         }
 475
 476         if (ct && nfqnl_ct_put(skb, ct, ctinfo) < 0)
 477                 goto nla_put_failure;
 478
 479         if (cap_len > 0 && nla_put_be32(skb, NFQA_CAP_LEN, htonl(cap_len)))
 480                 goto nla_put_failure;
 481
 482         if (nfqnl_put_packet_info(skb, entskb))
 483                 goto nla_put_failure;
 484
 485         if (data_len) {
 486                 struct nlattr *nla;
 487
 488                 if (skb_tailroom(skb) < sizeof(*nla) + hlen)
 489                         goto nla_put_failure;
 490
 491                 nla = (struct nlattr *)skb_put(skb, sizeof(*nla));
 492                 nla->nla_type = NFQA_PAYLOAD;
 493                 nla->nla_len = nla_attr_size(data_len);
 494
 495                 if (nfqnl_zcopy(skb, entskb, data_len, hlen))
 496                         goto nla_put_failure;
 497         }
 498
 499         nlh->nlmsg_len = skb->len;
 500         return skb;
 501
 502 nla_put_failure:
 503         skb_tx_error(entskb);
 504         kfree_skb(skb);
 505         net_err_ratelimited("nf_queue: error creating packet message\n");
 506         return NULL;
 507 }
 508
 509 static int
 510 __nfqnl_enqueue_packet(struct net *net, struct nfqnl_instance *queue,
 511                         struct nf_queue_entry *entry)
 512 {
 513         struct sk_buff *nskb;
 514         int err = -ENOBUFS;
 515         __be32 *packet_id_ptr;
 516         int failopen = 0;
 517
 518         nskb = nfqnl_build_packet_message(queue, entry, &packet_id_ptr);
 519         if (nskb == NULL) {
 520                 err = -ENOMEM;
 521                 goto err_out;
 522         }
 523         spin_lock_bh(&queue->lock);
 524
 525         if (!queue->peer_portid) {
 526                 err = -EINVAL;
 527                 goto err_out_free_nskb;
 528         }
 529         if (queue->queue_total >= queue->queue_maxlen) {
 530                 if (queue->flags & NFQA_CFG_F_FAIL_OPEN) {
 531                         failopen = 1;
 532                         err = 0;
 533                 } else {
 534                         queue->queue_dropped++;
 535                         net_warn_ratelimited("nf_queue: full at %d entries, dropping packets(s)\n",
 536                                              queue->queue_total);
 537                 }
 538                 goto err_out_free_nskb;
 539         }
 540         entry->id = ++queue->id_sequence;
 541         *packet_id_ptr = htonl(entry->id);
 542
 543         /* nfnetlink_unicast will either free the nskb or add it to a socket */
 544         err = nfnetlink_unicast(nskb, net, queue->peer_portid, MSG_DONTWAIT);
 545         if (err < 0) {
 546                 queue->queue_user_dropped++;
 547                 goto err_out_unlock;
 548         }
 549
 550         __enqueue_entry(queue, entry);
 551
 552         spin_unlock_bh(&queue->lock);
 553         return 0;
 554
 555 err_out_free_nskb:
 556         kfree_skb(nskb);
 557 err_out_unlock:
 558         spin_unlock_bh(&queue->lock);
 559         if (failopen)
 560                 nf_reinject(entry, NF_ACCEPT);
 561 err_out:
 562         return err;
 563 }
 564
 565 static struct nf_queue_entry *
 566 nf_queue_entry_dup(struct nf_queue_entry *e)
 567 {
 568         struct nf_queue_entry *entry = kmemdup(e, e->size, GFP_ATOMIC);
 569         if (entry) {
 570                 if (nf_queue_entry_get_refs(entry))
 571                         return entry;
 572                 kfree(entry);
 573         }
 574         return NULL;
 575 }
 576
 577 #ifdef CONFIG_BRIDGE_NETFILTER
 578 /* When called from bridge netfilter, skb->data must point to MAC header
 579  * before calling skb_gso_segment(). Else, original MAC header is lost
 580  * and segmented skbs will be sent to wrong destination.
 581  */
 582 static void nf_bridge_adjust_skb_data(struct sk_buff *skb)
 583 {
 584         if (skb->nf_bridge)
 585                 __skb_push(skb, skb->network_header - skb->mac_header);
 586 }
 587
 588 static void nf_bridge_adjust_segmented_data(struct sk_buff *skb)
 589 {
 590         if (skb->nf_bridge)
 591                 __skb_pull(skb, skb->network_header - skb->mac_header);
 592 }
 593 #else
 594 #define nf_bridge_adjust_skb_data(s) do {} while (0)
 595 #define nf_bridge_adjust_segmented_data(s) do {} while (0)
 596 #endif
 597
 598 static void free_entry(struct nf_queue_entry *entry)
 599 {
 600         nf_queue_entry_release_refs(entry);
 601         kfree(entry);
 602 }
 603
 604 static int
 605 __nfqnl_enqueue_packet_gso(struct net *net, struct nfqnl_instance *queue,
 606                            struct sk_buff *skb, struct nf_queue_entry *entry)
 607 {
 608         int ret = -ENOMEM;
 609         struct nf_queue_entry *entry_seg;
 610
 611         nf_bridge_adjust_segmented_data(skb);
 612
 613         if (skb->next == NULL) { /* last packet, no need to copy entry */
 614                 struct sk_buff *gso_skb = entry->skb;
 615                 entry->skb = skb;
 616                 ret = __nfqnl_enqueue_packet(net, queue, entry);
 617                 if (ret)
 618                         entry->skb = gso_skb;
 619                 return ret;
 620         }
 621
 622         skb->next = NULL;
 623
 624         entry_seg = nf_queue_entry_dup(entry);
 625         if (entry_seg) {
 626                 entry_seg->skb = skb;
 627                 ret = __nfqnl_enqueue_packet(net, queue, entry_seg);
 628                 if (ret)
 629                         free_entry(entry_seg);
 630         }
 631         return ret;
 632 }
 633
 634 static int
 635 nfqnl_enqueue_packet(struct nf_queue_entry *entry, unsigned int queuenum)
 636 {
 637         unsigned int queued;
 638         struct nfqnl_instance *queue;
 639         struct sk_buff *skb, *segs;
 640         int err = -ENOBUFS;
 641         struct net *net = dev_net(entry->indev ?
 642                                   entry->indev : entry->outdev);
 643         struct nfnl_queue_net *q = nfnl_queue_pernet(net);
 644
 645         /* rcu_read_lock()ed by nf_hook_slow() */
 646         queue = instance_lookup(q, queuenum);
 647         if (!queue)
 648                 return -ESRCH;
 649
 650         if (queue->copy_mode == NFQNL_COPY_NONE)
 651                 return -EINVAL;
 652
 653         skb = entry->skb;
 654
 655         switch (entry->pf) {
 656         case NFPROTO_IPV4:
 657                 skb->protocol = htons(ETH_P_IP);
 658                 break;
 659         case NFPROTO_IPV6:
 660                 skb->protocol = htons(ETH_P_IPV6);
 661                 break;
 662         }
 663
 664         if ((queue->flags & NFQA_CFG_F_GSO) || !skb_is_gso(skb))
 665                 return __nfqnl_enqueue_packet(net, queue, entry);
 666
 667         nf_bridge_adjust_skb_data(skb);
 668         segs = skb_gso_segment(skb, 0);
 669         /* Does not use PTR_ERR to limit the number of error codes that can be
 670          * returned by nf_queue.  For instance, callers rely on -ECANCELED to
 671          * mean 'ignore this hook'.
 672          */
 673         if (IS_ERR(segs))
 674                 goto out_err;
 675         queued = 0;
 676         err = 0;
 677         do {
 678                 struct sk_buff *nskb = segs->next;
 679                 if (err == 0)
 680                         err = __nfqnl_enqueue_packet_gso(net, queue,
 681                                                         segs, entry);
 682                 if (err == 0)
 683                         queued++;
 684                 else
 685                         kfree_skb(segs);
 686                 segs = nskb;
 687         } while (segs);
 688
 689         if (queued) {
 690                 if (err) /* some segments are already queued */
 691                         free_entry(entry);
 692                 kfree_skb(skb);
 693                 return 0;
 694         }
 695  out_err:
 696         nf_bridge_adjust_segmented_data(skb);
 697         return err;
 698 }
 699
 700 static int
 701 nfqnl_mangle(void *data, int data_len, struct nf_queue_entry *e, int diff)
 702 {
 703         struct sk_buff *nskb;
 704
 705         if (diff < 0) {
 706                 if (pskb_trim(e->skb, data_len))
 707                         return -ENOMEM;
 708         } else if (diff > 0) {
 709                 if (data_len > 0xFFFF)
 710                         return -EINVAL;
 711                 if (diff > skb_tailroom(e->skb)) {
 712                         nskb = skb_copy_expand(e->skb, skb_headroom(e->skb),
 713                                                diff, GFP_ATOMIC);
 714                         if (!nskb) {
 715                                 printk(KERN_WARNING "nf_queue: OOM "
 716                                       "in mangle, dropping packet\n");
 717                                 return -ENOMEM;
 718                         }
 719                         kfree_skb(e->skb);
 720                         e->skb = nskb;
 721                 }
 722                 skb_put(e->skb, diff);
 723         }
 724         if (!skb_make_writable(e->skb, data_len))
 725                 return -ENOMEM;
 726         skb_copy_to_linear_data(e->skb, data, data_len);
 727         e->skb->ip_summed = CHECKSUM_NONE;
 728         return 0;
 729 }
 730
 731 static int
 732 nfqnl_set_mode(struct nfqnl_instance *queue,
 733                unsigned char mode, unsigned int range)
 734 {
 735         int status = 0;
 736
 737         spin_lock_bh(&queue->lock);
 738         switch (mode) {
 739         case NFQNL_COPY_NONE:
 740         case NFQNL_COPY_META:
 741                 queue->copy_mode = mode;
 742                 queue->copy_range = 0;
 743                 break;
 744
 745         case NFQNL_COPY_PACKET:
 746                 queue->copy_mode = mode;
 747                 /* We're using struct nlattr which has 16bit nla_len. Note that
 748                  * nla_len includes the header length. Thus, the maximum packet
 749                  * length that we support is 65531 bytes. We send truncated
 750                  * packets if the specified length is larger than that.
 751                  */
 752                 if (range > 0xffff - NLA_HDRLEN)
 753                         queue->copy_range = 0xffff - NLA_HDRLEN;
 754                 else
 755                         queue->copy_range = range;
 756                 break;
 757
 758         default:
 759                 status = -EINVAL;
 760
 761         }
 762         spin_unlock_bh(&queue->lock);
 763
 764         return status;
 765 }
 766
 767 static int
 768 dev_cmp(struct nf_queue_entry *entry, unsigned long ifindex)
 769 {
 770         if (entry->indev)
 771                 if (entry->indev->ifindex == ifindex)
 772                         return 1;
 773         if (entry->outdev)
 774                 if (entry->outdev->ifindex == ifindex)
 775                         return 1;
 776 #ifdef CONFIG_BRIDGE_NETFILTER
 777         if (entry->skb->nf_bridge) {
 778                 if (entry->skb->nf_bridge->physindev &&
 779                     entry->skb->nf_bridge->physindev->ifindex == ifindex)
 780                         return 1;
 781                 if (entry->skb->nf_bridge->physoutdev &&
 782                     entry->skb->nf_bridge->physoutdev->ifindex == ifindex)
 783                         return 1;
 784         }
 785 #endif
 786         return 0;
 787 }
 788
 789 /* drop all packets with either indev or outdev == ifindex from all queue
 790  * instances */
 791 static void
 792 nfqnl_dev_drop(struct net *net, int ifindex)
 793 {
 794         int i;
 795         struct nfnl_queue_net *q = nfnl_queue_pernet(net);
 796
 797         rcu_read_lock();
 798
 799         for (i = 0; i < INSTANCE_BUCKETS; i++) {
 800                 struct nfqnl_instance *inst;
 801                 struct hlist_head *head = &q->instance_table[i];
 802
 803                 hlist_for_each_entry_rcu(inst, head, hlist)
 804                         nfqnl_flush(inst, dev_cmp, ifindex);
 805         }
 806
 807         rcu_read_unlock();
 808 }
 809
 810 #define RCV_SKB_FAIL(err) do { netlink_ack(skb, nlh, (err)); return; } while (0)
 811
 812 static int
 813 nfqnl_rcv_dev_event(struct notifier_block *this,
 814                     unsigned long event, void *ptr)
 815 {
 816         struct net_device *dev = ptr;
 817
 818         /* Drop any packets associated with the downed device */
 819         if (event == NETDEV_DOWN)
 820                 nfqnl_dev_drop(dev_net(dev), dev->ifindex);
 821         return NOTIFY_DONE;
 822 }
 823
 824 static struct notifier_block nfqnl_dev_notifier = {
 825         .notifier_call  = nfqnl_rcv_dev_event,
 826 };
 827
 828 static int
 829 nfqnl_rcv_nl_event(struct notifier_block *this,
 830                    unsigned long event, void *ptr)
 831 {
 832         struct netlink_notify *n = ptr;
 833         struct nfnl_queue_net *q = nfnl_queue_pernet(n->net);
 834
 835         if (event == NETLINK_URELEASE && n->protocol == NETLINK_NETFILTER) {
 836                 int i;
 837
 838                 /* destroy all instances for this portid */
 839                 spin_lock(&q->instances_lock);
 840                 for (i = 0; i < INSTANCE_BUCKETS; i++) {
 841                         struct hlist_node *t2;
 842                         struct nfqnl_instance *inst;
 843                         struct hlist_head *head = &q->instance_table[i];
 844
 845                         hlist_for_each_entry_safe(inst, t2, head, hlist) {
 846                                 if (n->portid == inst->peer_portid)
 847                                         __instance_destroy(inst);
 848                         }
 849                 }
 850                 spin_unlock(&q->instances_lock);
 851         }
 852         return NOTIFY_DONE;
 853 }
 854
 855 static struct notifier_block nfqnl_rtnl_notifier = {
 856         .notifier_call  = nfqnl_rcv_nl_event,
 857 };
 858
 859 static const struct nla_policy nfqa_verdict_policy[NFQA_MAX+1] = {
 860         [NFQA_VERDICT_HDR]      = { .len = sizeof(struct nfqnl_msg_verdict_hdr) },
 861         [NFQA_MARK]             = { .type = NLA_U32 },
 862         [NFQA_PAYLOAD]          = { .type = NLA_UNSPEC },
 863         [NFQA_CT]               = { .type = NLA_UNSPEC },
 864         [NFQA_EXP]              = { .type = NLA_UNSPEC },
 865 };
 866
 867 static const struct nla_policy nfqa_verdict_batch_policy[NFQA_MAX+1] = {
 868         [NFQA_VERDICT_HDR]      = { .len = sizeof(struct nfqnl_msg_verdict_hdr) },
 869         [NFQA_MARK]             = { .type = NLA_U32 },
 870 };
 871
 872 static struct nfqnl_instance *
 873 verdict_instance_lookup(struct nfnl_queue_net *q, u16 queue_num, int nlportid)
 874 {
 875         struct nfqnl_instance *queue;
 876
 877         queue = instance_lookup(q, queue_num);
 878         if (!queue)
 879                 return ERR_PTR(-ENODEV);
 880
 881         if (queue->peer_portid != nlportid)
 882                 return ERR_PTR(-EPERM);
 883
 884         return queue;
 885 }
 886
 887 static struct nfqnl_msg_verdict_hdr*
 888 verdicthdr_get(const struct nlattr * const nfqa[])
 889 {
 890         struct nfqnl_msg_verdict_hdr *vhdr;
 891         unsigned int verdict;
 892
 893         if (!nfqa[NFQA_VERDICT_HDR])
 894                 return NULL;
 895
 896         vhdr = nla_data(nfqa[NFQA_VERDICT_HDR]);
 897         verdict = ntohl(vhdr->verdict) & NF_VERDICT_MASK;
 898         if (verdict > NF_MAX_VERDICT || verdict == NF_STOLEN)
 899                 return NULL;
 900         return vhdr;
 901 }
 902
 903 static int nfq_id_after(unsigned int id, unsigned int max)
 904 {
 905         return (int)(id - max) > 0;
 906 }
 907
 908 static int
 909 nfqnl_recv_verdict_batch(struct sock *ctnl, struct sk_buff *skb,
 910                    const struct nlmsghdr *nlh,
 911                    const struct nlattr * const nfqa[])
 912 {
 913         struct nfgenmsg *nfmsg = nlmsg_data(nlh);
 914         struct nf_queue_entry *entry, *tmp;
 915         unsigned int verdict, maxid;
 916         struct nfqnl_msg_verdict_hdr *vhdr;
 917         struct nfqnl_instance *queue;
 918         LIST_HEAD(batch_list);
 919         u16 queue_num = ntohs(nfmsg->res_id);
 920
 921         struct net *net = sock_net(ctnl);
 922         struct nfnl_queue_net *q = nfnl_queue_pernet(net);
 923
 924         queue = verdict_instance_lookup(q, queue_num,
 925                                         NETLINK_CB(skb).portid);
 926         if (IS_ERR(queue))
 927                 return PTR_ERR(queue);
 928
 929         vhdr = verdicthdr_get(nfqa);
 930         if (!vhdr)
 931                 return -EINVAL;
 932
 933         verdict = ntohl(vhdr->verdict);
 934         maxid = ntohl(vhdr->id);
 935
 936         spin_lock_bh(&queue->lock);
 937
 938         list_for_each_entry_safe(entry, tmp, &queue->queue_list, list) {
 939                 if (nfq_id_after(entry->id, maxid))
 940                         break;
 941                 __dequeue_entry(queue, entry);
 942                 list_add_tail(&entry->list, &batch_list);
 943         }
 944
 945         spin_unlock_bh(&queue->lock);
 946
 947         if (list_empty(&batch_list))
 948                 return -ENOENT;
 949
 950         list_for_each_entry_safe(entry, tmp, &batch_list, list) {
 951                 if (nfqa[NFQA_MARK])
 952                         entry->skb->mark = ntohl(nla_get_be32(nfqa[NFQA_MARK]));
 953                 nf_reinject(entry, verdict);
 954         }
 955         return 0;
 956 }
 957
 958 static int
 959 nfqnl_recv_verdict(struct sock *ctnl, struct sk_buff *skb,
 960                    const struct nlmsghdr *nlh,
 961                    const struct nlattr * const nfqa[])
 962 {
 963         struct nfgenmsg *nfmsg = nlmsg_data(nlh);
 964         u_int16_t queue_num = ntohs(nfmsg->res_id);
 965
 966         struct nfqnl_msg_verdict_hdr *vhdr;
 967         struct nfqnl_instance *queue;
 968         unsigned int verdict;
 969         struct nf_queue_entry *entry;
 970         enum ip_conntrack_info uninitialized_var(ctinfo);
 971         struct nf_conn *ct = NULL;
 972
 973         struct net *net = sock_net(ctnl);
 974         struct nfnl_queue_net *q = nfnl_queue_pernet(net);
 975
 976         queue = instance_lookup(q, queue_num);
 977         if (!queue)
 978                 queue = verdict_instance_lookup(q, queue_num,
 979                                                 NETLINK_CB(skb).portid);
 980         if (IS_ERR(queue))
 981                 return PTR_ERR(queue);
 982
 983         vhdr = verdicthdr_get(nfqa);
 984         if (!vhdr)
 985                 return -EINVAL;
 986
 987         verdict = ntohl(vhdr->verdict);
 988
 989         entry = find_dequeue_entry(queue, ntohl(vhdr->id));
 990         if (entry == NULL)
 991                 return -ENOENT;
 992
 993         if (nfqa[NFQA_CT]) {
 994                 ct = nfqnl_ct_parse(entry->skb, nfqa[NFQA_CT], &ctinfo);
 995                 if (ct && nfqa[NFQA_EXP]) {
 996                         nfqnl_attach_expect(ct, nfqa[NFQA_EXP],
 997                                             NETLINK_CB(skb).portid,
 998                                             nlmsg_report(nlh));
 999                 }
1000         }
1001
1002         if (nfqa[NFQA_PAYLOAD]) {
1003                 u16 payload_len = nla_len(nfqa[NFQA_PAYLOAD]);
1004                 int diff = payload_len - entry->skb->len;
1005
1006                 if (nfqnl_mangle(nla_data(nfqa[NFQA_PAYLOAD]),
1007                                  payload_len, entry, diff) < 0)
1008                         verdict = NF_DROP;
1009
1010                 if (ct)
1011                         nfqnl_ct_seq_adjust(skb, ct, ctinfo, diff);
1012         }
1013         rcu_read_unlock();
1014
1015         if (nfqa[NFQA_MARK])
1016                 entry->skb->mark = ntohl(nla_get_be32(nfqa[NFQA_MARK]));
1017
1018         nf_reinject(entry, verdict);
1019         return 0;
1020 }
1021
1022 static int
1023 nfqnl_recv_unsupp(struct sock *ctnl, struct sk_buff *skb,
1024                   const struct nlmsghdr *nlh,
1025                   const struct nlattr * const nfqa[])
1026 {
1027         return -ENOTSUPP;
1028 }
1029
1030 static const struct nla_policy nfqa_cfg_policy[NFQA_CFG_MAX+1] = {
1031         [NFQA_CFG_CMD]          = { .len = sizeof(struct nfqnl_msg_config_cmd) },
1032         [NFQA_CFG_PARAMS]       = { .len = sizeof(struct nfqnl_msg_config_params) },
1033 };
1034
1035 static const struct nf_queue_handler nfqh = {
1036         .outfn  = &nfqnl_enqueue_packet,
1037 };
1038
1039 static int
1040 nfqnl_recv_config(struct sock *ctnl, struct sk_buff *skb,
1041                   const struct nlmsghdr *nlh,
1042                   const struct nlattr * const nfqa[])
1043 {
1044         struct nfgenmsg *nfmsg = nlmsg_data(nlh);
1045         u_int16_t queue_num = ntohs(nfmsg->res_id);
1046         struct nfqnl_instance *queue;
1047         struct nfqnl_msg_config_cmd *cmd = NULL;
1048         struct net *net = sock_net(ctnl);
1049         struct nfnl_queue_net *q = nfnl_queue_pernet(net);
1050         int ret = 0;
1051
1052         if (nfqa[NFQA_CFG_CMD]) {
1053                 cmd = nla_data(nfqa[NFQA_CFG_CMD]);
1054
1055                 /* Obsolete commands without queue context */
1056                 switch (cmd->command) {
1057                 case NFQNL_CFG_CMD_PF_BIND: return 0;
1058                 case NFQNL_CFG_CMD_PF_UNBIND: return 0;
1059                 }
1060         }
1061
1062         rcu_read_lock();
1063         queue = instance_lookup(q, queue_num);
1064         if (queue && queue->peer_portid != NETLINK_CB(skb).portid) {
1065                 ret = -EPERM;
1066                 goto err_out_unlock;
1067         }
1068
1069         if (cmd != NULL) {
1070                 switch (cmd->command) {
1071                 case NFQNL_CFG_CMD_BIND:
1072                         if (queue) {
1073                                 ret = -EBUSY;
1074                                 goto err_out_unlock;
1075                         }
1076                         queue = instance_create(q, queue_num,
1077                                                 NETLINK_CB(skb).portid);
1078                         if (IS_ERR(queue)) {
1079                                 ret = PTR_ERR(queue);
1080                                 goto err_out_unlock;
1081                         }
1082                         break;
1083                 case NFQNL_CFG_CMD_UNBIND:
1084                         if (!queue) {
1085                                 ret = -ENODEV;
1086                                 goto err_out_unlock;
1087                         }
1088                         instance_destroy(q, queue);
1089                         break;
1090                 case NFQNL_CFG_CMD_PF_BIND:
1091                 case NFQNL_CFG_CMD_PF_UNBIND:
1092                         break;
1093                 default:
1094                         ret = -ENOTSUPP;
1095                         break;
1096                 }
1097         }
1098
1099         if (nfqa[NFQA_CFG_PARAMS]) {
1100                 struct nfqnl_msg_config_params *params;
1101
1102                 if (!queue) {
1103                         ret = -ENODEV;
1104                         goto err_out_unlock;
1105                 }
1106                 params = nla_data(nfqa[NFQA_CFG_PARAMS]);
1107                 nfqnl_set_mode(queue, params->copy_mode,
1108                                 ntohl(params->copy_range));
1109         }
1110
1111         if (nfqa[NFQA_CFG_QUEUE_MAXLEN]) {
1112                 __be32 *queue_maxlen;
1113
1114                 if (!queue) {
1115                         ret = -ENODEV;
1116                         goto err_out_unlock;
1117                 }
1118                 queue_maxlen = nla_data(nfqa[NFQA_CFG_QUEUE_MAXLEN]);
1119                 spin_lock_bh(&queue->lock);
1120                 queue->queue_maxlen = ntohl(*queue_maxlen);
1121                 spin_unlock_bh(&queue->lock);
1122         }
1123
1124         if (nfqa[NFQA_CFG_FLAGS]) {
1125                 __u32 flags, mask;
1126
1127                 if (!queue) {
1128                         ret = -ENODEV;
1129                         goto err_out_unlock;
1130                 }
1131
1132                 if (!nfqa[NFQA_CFG_MASK]) {
1133                         /* A mask is needed to specify which flags are being
1134                          * changed.
1135                          */
1136                         ret = -EINVAL;
1137                         goto err_out_unlock;
1138                 }
1139
1140                 flags = ntohl(nla_get_be32(nfqa[NFQA_CFG_FLAGS]));
1141                 mask = ntohl(nla_get_be32(nfqa[NFQA_CFG_MASK]));
1142
1143                 if (flags >= NFQA_CFG_F_MAX) {
1144                         ret = -EOPNOTSUPP;
1145                         goto err_out_unlock;
1146                 }
1147
1148                 spin_lock_bh(&queue->lock);
1149                 queue->flags &= ~mask;
1150                 queue->flags |= flags & mask;
1151                 spin_unlock_bh(&queue->lock);
1152         }
1153
1154 err_out_unlock:
1155         rcu_read_unlock();
1156         return ret;
1157 }
1158
1159 static const struct nfnl_callback nfqnl_cb[NFQNL_MSG_MAX] = {
1160         [NFQNL_MSG_PACKET]      = { .call_rcu = nfqnl_recv_unsupp,
1161                                     .attr_count = NFQA_MAX, },
1162         [NFQNL_MSG_VERDICT]     = { .call_rcu = nfqnl_recv_verdict,
1163                                     .attr_count = NFQA_MAX,
1164                                     .policy = nfqa_verdict_policy },
1165         [NFQNL_MSG_CONFIG]      = { .call = nfqnl_recv_config,
1166                                     .attr_count = NFQA_CFG_MAX,
1167                                     .policy = nfqa_cfg_policy },
1168         [NFQNL_MSG_VERDICT_BATCH]={ .call_rcu = nfqnl_recv_verdict_batch,
1169                                     .attr_count = NFQA_MAX,
1170                                     .policy = nfqa_verdict_batch_policy },
1171 };
1172
1173 static const struct nfnetlink_subsystem nfqnl_subsys = {
1174         .name           = "nf_queue",
1175         .subsys_id      = NFNL_SUBSYS_QUEUE,
1176         .cb_count       = NFQNL_MSG_MAX,
1177         .cb             = nfqnl_cb,
1178 };
1179
1180 #ifdef CONFIG_PROC_FS
1181 struct iter_state {
1182         struct seq_net_private p;
1183         unsigned int bucket;
1184 };
1185
1186 static struct hlist_node *get_first(struct seq_file *seq)
1187 {
1188         struct iter_state *st = seq->private;
1189         struct net *net;
1190         struct nfnl_queue_net *q;
1191
1192         if (!st)
1193                 return NULL;
1194
1195         net = seq_file_net(seq);
1196         q = nfnl_queue_pernet(net);
1197         for (st->bucket = 0; st->bucket < INSTANCE_BUCKETS; st->bucket++) {
1198                 if (!hlist_empty(&q->instance_table[st->bucket]))
1199                         return q->instance_table[st->bucket].first;
1200         }
1201         return NULL;
1202 }
1203
1204 static struct hlist_node *get_next(struct seq_file *seq, struct hlist_node *h)
1205 {
1206         struct iter_state *st = seq->private;
1207         struct net *net = seq_file_net(seq);
1208
1209         h = h->next;
1210         while (!h) {
1211                 struct nfnl_queue_net *q;
1212
1213                 if (++st->bucket >= INSTANCE_BUCKETS)
1214                         return NULL;
1215
1216                 q = nfnl_queue_pernet(net);
1217                 h = q->instance_table[st->bucket].first;
1218         }
1219         return h;
1220 }
1221
1222 static struct hlist_node *get_idx(struct seq_file *seq, loff_t pos)
1223 {
1224         struct hlist_node *head;
1225         head = get_first(seq);
1226
1227         if (head)
1228                 while (pos && (head = get_next(seq, head)))
1229                         pos--;
1230         return pos ? NULL : head;
1231 }
1232
1233 static void *seq_start(struct seq_file *s, loff_t *pos)
1234         __acquires(nfnl_queue_pernet(seq_file_net(s))->instances_lock)
1235 {
1236         spin_lock(&nfnl_queue_pernet(seq_file_net(s))->instances_lock);
1237         return get_idx(s, *pos);
1238 }
1239
1240 static void *seq_next(struct seq_file *s, void *v, loff_t *pos)
1241 {
1242         (*pos)++;
1243         return get_next(s, v);
1244 }
1245
1246 static void seq_stop(struct seq_file *s, void *v)
1247         __releases(nfnl_queue_pernet(seq_file_net(s))->instances_lock)
1248 {
1249         spin_unlock(&nfnl_queue_pernet(seq_file_net(s))->instances_lock);
1250 }
1251
1252 static int seq_show(struct seq_file *s, void *v)
1253 {
1254         const struct nfqnl_instance *inst = v;
1255
1256         return seq_printf(s, "%5d %6d %5d %1d %5d %5d %5d %8d %2d\n",
1257                           inst->queue_num,
1258                           inst->peer_portid, inst->queue_total,
1259                           inst->copy_mode, inst->copy_range,
1260                           inst->queue_dropped, inst->queue_user_dropped,
1261                           inst->id_sequence, 1);
1262 }
1263
1264 static const struct seq_operations nfqnl_seq_ops = {
1265         .start  = seq_start,
1266         .next   = seq_next,
1267         .stop   = seq_stop,
1268         .show   = seq_show,
1269 };
1270
1271 static int nfqnl_open(struct inode *inode, struct file *file)
1272 {
1273         return seq_open_net(inode, file, &nfqnl_seq_ops,
1274                         sizeof(struct iter_state));
1275 }
1276
1277 static const struct file_operations nfqnl_file_ops = {
1278         .owner   = THIS_MODULE,
1279         .open    = nfqnl_open,
1280         .read    = seq_read,
1281         .llseek  = seq_lseek,
1282         .release = seq_release_net,
1283 };
1284
1285 #endif /* PROC_FS */
1286
1287 static int __net_init nfnl_queue_net_init(struct net *net)
1288 {
1289         unsigned int i;
1290         struct nfnl_queue_net *q = nfnl_queue_pernet(net);
1291
1292         for (i = 0; i < INSTANCE_BUCKETS; i++)
1293                 INIT_HLIST_HEAD(&q->instance_table[i]);
1294
1295         spin_lock_init(&q->instances_lock);
1296
1297 #ifdef CONFIG_PROC_FS
1298         if (!proc_create("nfnetlink_queue", 0440,
1299                          net->nf.proc_netfilter, &nfqnl_file_ops))
1300                 return -ENOMEM;
1301 #endif
1302         return 0;
1303 }
1304
1305 static void __net_exit nfnl_queue_net_exit(struct net *net)
1306 {
1307 #ifdef CONFIG_PROC_FS
1308         remove_proc_entry("nfnetlink_queue", net->nf.proc_netfilter);
1309 #endif
1310 }
1311
1312 static struct pernet_operations nfnl_queue_net_ops = {
1313         .init   = nfnl_queue_net_init,
1314         .exit   = nfnl_queue_net_exit,
1315         .id     = &nfnl_queue_net_id,
1316         .size   = sizeof(struct nfnl_queue_net),
1317 };
1318
1319 static int __init nfnetlink_queue_init(void)
1320 {
1321         int status = -ENOMEM;
1322
1323         netlink_register_notifier(&nfqnl_rtnl_notifier);
1324         status = nfnetlink_subsys_register(&nfqnl_subsys);
1325         if (status < 0) {
1326                 pr_err("nf_queue: failed to create netlink socket\n");
1327                 goto cleanup_netlink_notifier;
1328         }
1329
1330         status = register_pernet_subsys(&nfnl_queue_net_ops);
1331         if (status < 0) {
1332                 pr_err("nf_queue: failed to register pernet ops\n");
1333                 goto cleanup_subsys;
1334         }
1335         register_netdevice_notifier(&nfqnl_dev_notifier);
1336         nf_register_queue_handler(&nfqh);
1337         return status;
1338
1339 cleanup_subsys:
1340         nfnetlink_subsys_unregister(&nfqnl_subsys);
1341 cleanup_netlink_notifier:
1342         netlink_unregister_notifier(&nfqnl_rtnl_notifier);
1343         return status;
1344 }
1345
1346 static void __exit nfnetlink_queue_fini(void)
1347 {
1348         nf_unregister_queue_handler();
1349         unregister_netdevice_notifier(&nfqnl_dev_notifier);
1350         unregister_pernet_subsys(&nfnl_queue_net_ops);
1351         nfnetlink_subsys_unregister(&nfqnl_subsys);
1352         netlink_unregister_notifier(&nfqnl_rtnl_notifier);
1353
1354         rcu_barrier(); /* Wait for completion of call_rcu()'s */
1355 }
1356
1357 MODULE_DESCRIPTION("netfilter packet queue handler");
1358 MODULE_AUTHOR("Harald Welte <laforge@netfilter.org>");
1359 MODULE_LICENSE("GPL");
1360 MODULE_ALIAS_NFNL_SUBSYS(NFNL_SUBSYS_QUEUE);
1361
1362 module_init(nfnetlink_queue_init);
1363 module_exit(nfnetlink_queue_fini);