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 /* We're using struct nlattr which has 16bit nla_len. Note that nla_len
  45  * includes the header length. Thus, the maximum packet length that we
  46  * support is 65531 bytes. We send truncated packets if the specified length
  47  * is larger than that.  Userspace can check for presence of NFQA_CAP_LEN
  48  * attribute to detect truncation.
  49  */
  50 #define NFQNL_MAX_COPY_RANGE (0xffff - NLA_HDRLEN)
  51
  52 struct nfqnl_instance {
  53         struct hlist_node hlist;                /* global list of queues */
  54         struct rcu_head rcu;
  55
  56         int peer_portid;
  57         unsigned int queue_maxlen;
  58         unsigned int copy_range;
  59         unsigned int queue_dropped;
  60         unsigned int queue_user_dropped;
  61
  62
  63         u_int16_t queue_num;                    /* number of this queue */
  64         u_int8_t copy_mode;
  65         u_int32_t flags;                        /* Set using NFQA_CFG_FLAGS */
  66 /*
  67  * Following fields are dirtied for each queued packet,
  68  * keep them in same cache line if possible.
  69  */
  70         spinlock_t      lock;
  71         unsigned int    queue_total;
  72         unsigned int    id_sequence;            /* 'sequence' of pkt ids */
  73         struct list_head queue_list;            /* packets in queue */
  74 };
  75
  76 typedef int (*nfqnl_cmpfn)(struct nf_queue_entry *, unsigned long);
  77
  78 static int nfnl_queue_net_id __read_mostly;
  79
  80 #define INSTANCE_BUCKETS        16
  81 struct nfnl_queue_net {
  82         spinlock_t instances_lock;
  83         struct hlist_head instance_table[INSTANCE_BUCKETS];
  84 };
  85
  86 static struct nfnl_queue_net *nfnl_queue_pernet(struct net *net)
  87 {
  88         return net_generic(net, nfnl_queue_net_id);
  89 }
  90
  91 static inline u_int8_t instance_hashfn(u_int16_t queue_num)
  92 {
  93         return ((queue_num >> 8) ^ queue_num) % INSTANCE_BUCKETS;
  94 }
  95
  96 static struct nfqnl_instance *
  97 instance_lookup(struct nfnl_queue_net *q, u_int16_t queue_num)
  98 {
  99         struct hlist_head *head;
 100         struct nfqnl_instance *inst;
 101
 102         head = &q->instance_table[instance_hashfn(queue_num)];
 103         hlist_for_each_entry_rcu(inst, head, hlist) {
 104                 if (inst->queue_num == queue_num)
 105                         return inst;
 106         }
 107         return NULL;
 108 }
 109
 110 static struct nfqnl_instance *
 111 instance_create(struct nfnl_queue_net *q, u_int16_t queue_num,
 112                 int portid)
 113 {
 114         struct nfqnl_instance *inst;
 115         unsigned int h;
 116         int err;
 117
 118         spin_lock(&q->instances_lock);
 119         if (instance_lookup(q, queue_num)) {
 120                 err = -EEXIST;
 121                 goto out_unlock;
 122         }
 123
 124         inst = kzalloc(sizeof(*inst), GFP_ATOMIC);
 125         if (!inst) {
 126                 err = -ENOMEM;
 127                 goto out_unlock;
 128         }
 129
 130         inst->queue_num = queue_num;
 131         inst->peer_portid = portid;
 132         inst->queue_maxlen = NFQNL_QMAX_DEFAULT;
 133         inst->copy_range = NFQNL_MAX_COPY_RANGE;
 134         inst->copy_mode = NFQNL_COPY_NONE;
 135         spin_lock_init(&inst->lock);
 136         INIT_LIST_HEAD(&inst->queue_list);
 137
 138         if (!try_module_get(THIS_MODULE)) {
 139                 err = -EAGAIN;
 140                 goto out_free;
 141         }
 142
 143         h = instance_hashfn(queue_num);
 144         hlist_add_head_rcu(&inst->hlist, &q->instance_table[h]);
 145
 146         spin_unlock(&q->instances_lock);
 147
 148         return inst;
 149
 150 out_free:
 151         kfree(inst);
 152 out_unlock:
 153         spin_unlock(&q->instances_lock);
 154         return ERR_PTR(err);
 155 }
 156
 157 static void nfqnl_flush(struct nfqnl_instance *queue, nfqnl_cmpfn cmpfn,
 158                         unsigned long data);
 159
 160 static void
 161 instance_destroy_rcu(struct rcu_head *head)
 162 {
 163         struct nfqnl_instance *inst = container_of(head, struct nfqnl_instance,
 164                                                    rcu);
 165
 166         nfqnl_flush(inst, NULL, 0);
 167         kfree(inst);
 168         module_put(THIS_MODULE);
 169 }
 170
 171 static void
 172 __instance_destroy(struct nfqnl_instance *inst)
 173 {
 174         hlist_del_rcu(&inst->hlist);
 175         call_rcu(&inst->rcu, instance_destroy_rcu);
 176 }
 177
 178 static void
 179 instance_destroy(struct nfnl_queue_net *q, struct nfqnl_instance *inst)
 180 {
 181         spin_lock(&q->instances_lock);
 182         __instance_destroy(inst);
 183         spin_unlock(&q->instances_lock);
 184 }
 185
 186 static inline void
 187 __enqueue_entry(struct nfqnl_instance *queue, struct nf_queue_entry *entry)
 188 {
 189        list_add_tail(&entry->list, &queue->queue_list);
 190        queue->queue_total++;
 191 }
 192
 193 static void
 194 __dequeue_entry(struct nfqnl_instance *queue, struct nf_queue_entry *entry)
 195 {
 196         list_del(&entry->list);
 197         queue->queue_total--;
 198 }
 199
 200 static struct nf_queue_entry *
 201 find_dequeue_entry(struct nfqnl_instance *queue, unsigned int id)
 202 {
 203         struct nf_queue_entry *entry = NULL, *i;
 204
 205         spin_lock_bh(&queue->lock);
 206
 207         list_for_each_entry(i, &queue->queue_list, list) {
 208                 if (i->id == id) {
 209                         entry = i;
 210                         break;
 211                 }
 212         }
 213
 214         if (entry)
 215                 __dequeue_entry(queue, entry);
 216
 217         spin_unlock_bh(&queue->lock);
 218
 219         return entry;
 220 }
 221
 222 static void
 223 nfqnl_flush(struct nfqnl_instance *queue, nfqnl_cmpfn cmpfn, unsigned long data)
 224 {
 225         struct nf_queue_entry *entry, *next;
 226
 227         spin_lock_bh(&queue->lock);
 228         list_for_each_entry_safe(entry, next, &queue->queue_list, list) {
 229                 if (!cmpfn || cmpfn(entry, data)) {
 230                         list_del(&entry->list);
 231                         queue->queue_total--;
 232                         nf_reinject(entry, NF_DROP);
 233                 }
 234         }
 235         spin_unlock_bh(&queue->lock);
 236 }
 237
 238 static void
 239 nfqnl_zcopy(struct sk_buff *to, const struct sk_buff *from, int len, int hlen)
 240 {
 241         int i, j = 0;
 242         int plen = 0; /* length of skb->head fragment */
 243         struct page *page;
 244         unsigned int offset;
 245
 246         /* dont bother with small payloads */
 247         if (len <= skb_tailroom(to)) {
 248                 skb_copy_bits(from, 0, skb_put(to, len), len);
 249                 return;
 250         }
 251
 252         if (hlen) {
 253                 skb_copy_bits(from, 0, skb_put(to, hlen), hlen);
 254                 len -= hlen;
 255         } else {
 256                 plen = min_t(int, skb_headlen(from), len);
 257                 if (plen) {
 258                         page = virt_to_head_page(from->head);
 259                         offset = from->data - (unsigned char *)page_address(page);
 260                         __skb_fill_page_desc(to, 0, page, offset, plen);
 261                         get_page(page);
 262                         j = 1;
 263                         len -= plen;
 264                 }
 265         }
 266
 267         to->truesize += len + plen;
 268         to->len += len + plen;
 269         to->data_len += len + plen;
 270
 271         for (i = 0; i < skb_shinfo(from)->nr_frags; i++) {
 272                 if (!len)
 273                         break;
 274                 skb_shinfo(to)->frags[j] = skb_shinfo(from)->frags[i];
 275                 skb_shinfo(to)->frags[j].size = min_t(int, skb_shinfo(to)->frags[j].size, len);
 276                 len -= skb_shinfo(to)->frags[j].size;
 277                 skb_frag_ref(to, j);
 278                 j++;
 279         }
 280         skb_shinfo(to)->nr_frags = j;
 281 }
 282
 283 static int
 284 nfqnl_put_packet_info(struct sk_buff *nlskb, struct sk_buff *packet,
 285                       bool csum_verify)
 286 {
 287         __u32 flags = 0;
 288
 289         if (packet->ip_summed == CHECKSUM_PARTIAL)
 290                 flags = NFQA_SKB_CSUMNOTREADY;
 291         else if (csum_verify)
 292                 flags = NFQA_SKB_CSUM_NOTVERIFIED;
 293
 294         if (skb_is_gso(packet))
 295                 flags |= NFQA_SKB_GSO;
 296
 297         return flags ? nla_put_be32(nlskb, NFQA_SKB_INFO, htonl(flags)) : 0;
 298 }
 299
 300 static struct sk_buff *
 301 nfqnl_build_packet_message(struct nfqnl_instance *queue,
 302                            struct nf_queue_entry *entry,
 303                            __be32 **packet_id_ptr)
 304 {
 305         size_t size;
 306         size_t data_len = 0, cap_len = 0;
 307         int hlen = 0;
 308         struct sk_buff *skb;
 309         struct nlattr *nla;
 310         struct nfqnl_msg_packet_hdr *pmsg;
 311         struct nlmsghdr *nlh;
 312         struct nfgenmsg *nfmsg;
 313         struct sk_buff *entskb = entry->skb;
 314         struct net_device *indev;
 315         struct net_device *outdev;
 316         struct nf_conn *ct = NULL;
 317         enum ip_conntrack_info uninitialized_var(ctinfo);
 318         bool csum_verify;
 319
 320         size =    nlmsg_total_size(sizeof(struct nfgenmsg))
 321                 + nla_total_size(sizeof(struct nfqnl_msg_packet_hdr))
 322                 + nla_total_size(sizeof(u_int32_t))     /* ifindex */
 323                 + nla_total_size(sizeof(u_int32_t))     /* ifindex */
 324 #ifdef CONFIG_BRIDGE_NETFILTER
 325                 + nla_total_size(sizeof(u_int32_t))     /* ifindex */
 326                 + nla_total_size(sizeof(u_int32_t))     /* ifindex */
 327 #endif
 328                 + nla_total_size(sizeof(u_int32_t))     /* mark */
 329                 + nla_total_size(sizeof(struct nfqnl_msg_packet_hw))
 330                 + nla_total_size(sizeof(u_int32_t))     /* skbinfo */
 331                 + nla_total_size(sizeof(u_int32_t));    /* cap_len */
 332
 333         if (entskb->tstamp.tv64)
 334                 size += nla_total_size(sizeof(struct nfqnl_msg_packet_timestamp));
 335
 336         if (entry->hook <= NF_INET_FORWARD ||
 337            (entry->hook == NF_INET_POST_ROUTING && entskb->sk == NULL))
 338                 csum_verify = !skb_csum_unnecessary(entskb);
 339         else
 340                 csum_verify = false;
 341
 342         outdev = entry->outdev;
 343
 344         switch ((enum nfqnl_config_mode)ACCESS_ONCE(queue->copy_mode)) {
 345         case NFQNL_COPY_META:
 346         case NFQNL_COPY_NONE:
 347                 break;
 348
 349         case NFQNL_COPY_PACKET:
 350                 if (!(queue->flags & NFQA_CFG_F_GSO) &&
 351                     entskb->ip_summed == CHECKSUM_PARTIAL &&
 352                     skb_checksum_help(entskb))
 353                         return NULL;
 354
 355                 data_len = ACCESS_ONCE(queue->copy_range);
 356                 if (data_len > entskb->len)
 357                         data_len = entskb->len;
 358
 359                 if (!entskb->head_frag ||
 360                     skb_headlen(entskb) < L1_CACHE_BYTES ||
 361                     skb_shinfo(entskb)->nr_frags >= MAX_SKB_FRAGS)
 362                         hlen = skb_headlen(entskb);
 363
 364                 if (skb_has_frag_list(entskb))
 365                         hlen = entskb->len;
 366                 hlen = min_t(int, data_len, hlen);
 367                 size += sizeof(struct nlattr) + hlen;
 368                 cap_len = entskb->len;
 369                 break;
 370         }
 371
 372         if (queue->flags & NFQA_CFG_F_CONNTRACK)
 373                 ct = nfqnl_ct_get(entskb, &size, &ctinfo);
 374
 375         skb = nfnetlink_alloc_skb(&init_net, size, queue->peer_portid,
 376                                   GFP_ATOMIC);
 377         if (!skb)
 378                 return NULL;
 379
 380         nlh = nlmsg_put(skb, 0, 0,
 381                         NFNL_SUBSYS_QUEUE << 8 | NFQNL_MSG_PACKET,
 382                         sizeof(struct nfgenmsg), 0);
 383         if (!nlh) {
 384                 kfree_skb(skb);
 385                 return NULL;
 386         }
 387         nfmsg = nlmsg_data(nlh);
 388         nfmsg->nfgen_family = entry->pf;
 389         nfmsg->version = NFNETLINK_V0;
 390         nfmsg->res_id = htons(queue->queue_num);
 391
 392         nla = __nla_reserve(skb, NFQA_PACKET_HDR, sizeof(*pmsg));
 393         pmsg = nla_data(nla);
 394         pmsg->hw_protocol       = entskb->protocol;
 395         pmsg->hook              = entry->hook;
 396         *packet_id_ptr          = &pmsg->packet_id;
 397
 398         indev = entry->indev;
 399         if (indev) {
 400 #ifndef CONFIG_BRIDGE_NETFILTER
 401                 if (nla_put_be32(skb, NFQA_IFINDEX_INDEV, htonl(indev->ifindex)))
 402                         goto nla_put_failure;
 403 #else
 404                 if (entry->pf == PF_BRIDGE) {
 405                         /* Case 1: indev is physical input device, we need to
 406                          * look for bridge group (when called from
 407                          * netfilter_bridge) */
 408                         if (nla_put_be32(skb, NFQA_IFINDEX_PHYSINDEV,
 409                                          htonl(indev->ifindex)) ||
 410                         /* this is the bridge group "brX" */
 411                         /* rcu_read_lock()ed by __nf_queue */
 412                             nla_put_be32(skb, NFQA_IFINDEX_INDEV,
 413                                          htonl(br_port_get_rcu(indev)->br->dev->ifindex)))
 414                                 goto nla_put_failure;
 415                 } else {
 416                         /* Case 2: indev is bridge group, we need to look for
 417                          * physical device (when called from ipv4) */
 418                         if (nla_put_be32(skb, NFQA_IFINDEX_INDEV,
 419                                          htonl(indev->ifindex)))
 420                                 goto nla_put_failure;
 421                         if (entskb->nf_bridge && entskb->nf_bridge->physindev &&
 422                             nla_put_be32(skb, NFQA_IFINDEX_PHYSINDEV,
 423                                          htonl(entskb->nf_bridge->physindev->ifindex)))
 424                                 goto nla_put_failure;
 425                 }
 426 #endif
 427         }
 428
 429         if (outdev) {
 430 #ifndef CONFIG_BRIDGE_NETFILTER
 431                 if (nla_put_be32(skb, NFQA_IFINDEX_OUTDEV, htonl(outdev->ifindex)))
 432                         goto nla_put_failure;
 433 #else
 434                 if (entry->pf == PF_BRIDGE) {
 435                         /* Case 1: outdev is physical output device, we need to
 436                          * look for bridge group (when called from
 437                          * netfilter_bridge) */
 438                         if (nla_put_be32(skb, NFQA_IFINDEX_PHYSOUTDEV,
 439                                          htonl(outdev->ifindex)) ||
 440                         /* this is the bridge group "brX" */
 441                         /* rcu_read_lock()ed by __nf_queue */
 442                             nla_put_be32(skb, NFQA_IFINDEX_OUTDEV,
 443                                          htonl(br_port_get_rcu(outdev)->br->dev->ifindex)))
 444                                 goto nla_put_failure;
 445                 } else {
 446                         /* Case 2: outdev is bridge group, we need to look for
 447                          * physical output device (when called from ipv4) */
 448                         if (nla_put_be32(skb, NFQA_IFINDEX_OUTDEV,
 449                                          htonl(outdev->ifindex)))
 450                                 goto nla_put_failure;
 451                         if (entskb->nf_bridge && entskb->nf_bridge->physoutdev &&
 452                             nla_put_be32(skb, NFQA_IFINDEX_PHYSOUTDEV,
 453                                          htonl(entskb->nf_bridge->physoutdev->ifindex)))
 454                                 goto nla_put_failure;
 455                 }
 456 #endif
 457         }
 458
 459         if (entskb->mark &&
 460             nla_put_be32(skb, NFQA_MARK, htonl(entskb->mark)))
 461                 goto nla_put_failure;
 462
 463         if (indev && entskb->dev &&
 464             entskb->mac_header != entskb->network_header) {
 465                 struct nfqnl_msg_packet_hw phw;
 466                 int len = dev_parse_header(entskb, phw.hw_addr);
 467                 if (len) {
 468                         phw.hw_addrlen = htons(len);
 469                         if (nla_put(skb, NFQA_HWADDR, sizeof(phw), &phw))
 470                                 goto nla_put_failure;
 471                 }
 472         }
 473
 474         if (entskb->tstamp.tv64) {
 475                 struct nfqnl_msg_packet_timestamp ts;
 476                 struct timeval tv = ktime_to_timeval(entskb->tstamp);
 477                 ts.sec = cpu_to_be64(tv.tv_sec);
 478                 ts.usec = cpu_to_be64(tv.tv_usec);
 479
 480                 if (nla_put(skb, NFQA_TIMESTAMP, sizeof(ts), &ts))
 481                         goto nla_put_failure;
 482         }
 483
 484         if (ct && nfqnl_ct_put(skb, ct, ctinfo) < 0)
 485                 goto nla_put_failure;
 486
 487         if (cap_len > data_len &&
 488             nla_put_be32(skb, NFQA_CAP_LEN, htonl(cap_len)))
 489                 goto nla_put_failure;
 490
 491         if (nfqnl_put_packet_info(skb, entskb, csum_verify))
 492                 goto nla_put_failure;
 493
 494         if (data_len) {
 495                 struct nlattr *nla;
 496
 497                 if (skb_tailroom(skb) < sizeof(*nla) + hlen)
 498                         goto nla_put_failure;
 499
 500                 nla = (struct nlattr *)skb_put(skb, sizeof(*nla));
 501                 nla->nla_type = NFQA_PAYLOAD;
 502                 nla->nla_len = nla_attr_size(data_len);
 503
 504                 nfqnl_zcopy(skb, entskb, data_len, hlen);
 505         }
 506
 507         nlh->nlmsg_len = skb->len;
 508         return skb;
 509
 510 nla_put_failure:
 511         kfree_skb(skb);
 512         net_err_ratelimited("nf_queue: error creating packet message\n");
 513         return NULL;
 514 }
 515
 516 static int
 517 __nfqnl_enqueue_packet(struct net *net, struct nfqnl_instance *queue,
 518                         struct nf_queue_entry *entry)
 519 {
 520         struct sk_buff *nskb;
 521         int err = -ENOBUFS;
 522         __be32 *packet_id_ptr;
 523         int failopen = 0;
 524
 525         nskb = nfqnl_build_packet_message(queue, entry, &packet_id_ptr);
 526         if (nskb == NULL) {
 527                 err = -ENOMEM;
 528                 goto err_out;
 529         }
 530         spin_lock_bh(&queue->lock);
 531
 532         if (queue->queue_total >= queue->queue_maxlen) {
 533                 if (queue->flags & NFQA_CFG_F_FAIL_OPEN) {
 534                         failopen = 1;
 535                         err = 0;
 536                 } else {
 537                         queue->queue_dropped++;
 538                         net_warn_ratelimited("nf_queue: full at %d entries, dropping packets(s)\n",
 539                                              queue->queue_total);
 540                 }
 541                 goto err_out_free_nskb;
 542         }
 543         entry->id = ++queue->id_sequence;
 544         *packet_id_ptr = htonl(entry->id);
 545
 546         /* nfnetlink_unicast will either free the nskb or add it to a socket */
 547         err = nfnetlink_unicast(nskb, net, queue->peer_portid, MSG_DONTWAIT);
 548         if (err < 0) {
 549                 queue->queue_user_dropped++;
 550                 goto err_out_unlock;
 551         }
 552
 553         __enqueue_entry(queue, entry);
 554
 555         spin_unlock_bh(&queue->lock);
 556         return 0;
 557
 558 err_out_free_nskb:
 559         kfree_skb(nskb);
 560 err_out_unlock:
 561         spin_unlock_bh(&queue->lock);
 562         if (failopen)
 563                 nf_reinject(entry, NF_ACCEPT);
 564 err_out:
 565         return err;
 566 }
 567
 568 static struct nf_queue_entry *
 569 nf_queue_entry_dup(struct nf_queue_entry *e)
 570 {
 571         struct nf_queue_entry *entry = kmemdup(e, e->size, GFP_ATOMIC);
 572         if (entry) {
 573                 if (nf_queue_entry_get_refs(entry))
 574                         return entry;
 575                 kfree(entry);
 576         }
 577         return NULL;
 578 }
 579
 580 #ifdef CONFIG_BRIDGE_NETFILTER
 581 /* When called from bridge netfilter, skb->data must point to MAC header
 582  * before calling skb_gso_segment(). Else, original MAC header is lost
 583  * and segmented skbs will be sent to wrong destination.
 584  */
 585 static void nf_bridge_adjust_skb_data(struct sk_buff *skb)
 586 {
 587         if (skb->nf_bridge)
 588                 __skb_push(skb, skb->network_header - skb->mac_header);
 589 }
 590
 591 static void nf_bridge_adjust_segmented_data(struct sk_buff *skb)
 592 {
 593         if (skb->nf_bridge)
 594                 __skb_pull(skb, skb->network_header - skb->mac_header);
 595 }
 596 #else
 597 #define nf_bridge_adjust_skb_data(s) do {} while (0)
 598 #define nf_bridge_adjust_segmented_data(s) do {} while (0)
 599 #endif
 600
 601 static void free_entry(struct nf_queue_entry *entry)
 602 {
 603         nf_queue_entry_release_refs(entry);
 604         kfree(entry);
 605 }
 606
 607 static int
 608 __nfqnl_enqueue_packet_gso(struct net *net, struct nfqnl_instance *queue,
 609                            struct sk_buff *skb, struct nf_queue_entry *entry)
 610 {
 611         int ret = -ENOMEM;
 612         struct nf_queue_entry *entry_seg;
 613
 614         nf_bridge_adjust_segmented_data(skb);
 615
 616         if (skb->next == NULL) { /* last packet, no need to copy entry */
 617                 struct sk_buff *gso_skb = entry->skb;
 618                 entry->skb = skb;
 619                 ret = __nfqnl_enqueue_packet(net, queue, entry);
 620                 if (ret)
 621                         entry->skb = gso_skb;
 622                 return ret;
 623         }
 624
 625         skb->next = NULL;
 626
 627         entry_seg = nf_queue_entry_dup(entry);
 628         if (entry_seg) {
 629                 entry_seg->skb = skb;
 630                 ret = __nfqnl_enqueue_packet(net, queue, entry_seg);
 631                 if (ret)
 632                         free_entry(entry_seg);
 633         }
 634         return ret;
 635 }
 636
 637 static int
 638 nfqnl_enqueue_packet(struct nf_queue_entry *entry, unsigned int queuenum)
 639 {
 640         unsigned int queued;
 641         struct nfqnl_instance *queue;
 642         struct sk_buff *skb, *segs;
 643         int err = -ENOBUFS;
 644         struct net *net = dev_net(entry->indev ?
 645                                   entry->indev : entry->outdev);
 646         struct nfnl_queue_net *q = nfnl_queue_pernet(net);
 647
 648         /* rcu_read_lock()ed by nf_hook_slow() */
 649         queue = instance_lookup(q, queuenum);
 650         if (!queue)
 651                 return -ESRCH;
 652
 653         if (queue->copy_mode == NFQNL_COPY_NONE)
 654                 return -EINVAL;
 655
 656         skb = entry->skb;
 657
 658         switch (entry->pf) {
 659         case NFPROTO_IPV4:
 660                 skb->protocol = htons(ETH_P_IP);
 661                 break;
 662         case NFPROTO_IPV6:
 663                 skb->protocol = htons(ETH_P_IPV6);
 664                 break;
 665         }
 666
 667         if ((queue->flags & NFQA_CFG_F_GSO) || !skb_is_gso(skb))
 668                 return __nfqnl_enqueue_packet(net, queue, entry);
 669
 670         nf_bridge_adjust_skb_data(skb);
 671         segs = skb_gso_segment(skb, 0);
 672         /* Does not use PTR_ERR to limit the number of error codes that can be
 673          * returned by nf_queue.  For instance, callers rely on -ECANCELED to
 674          * mean 'ignore this hook'.
 675          */
 676         if (IS_ERR(segs))
 677                 goto out_err;
 678         queued = 0;
 679         err = 0;
 680         do {
 681                 struct sk_buff *nskb = segs->next;
 682                 if (err == 0)
 683                         err = __nfqnl_enqueue_packet_gso(net, queue,
 684                                                         segs, entry);
 685                 if (err == 0)
 686                         queued++;
 687                 else
 688                         kfree_skb(segs);
 689                 segs = nskb;
 690         } while (segs);
 691
 692         if (queued) {
 693                 if (err) /* some segments are already queued */
 694                         free_entry(entry);
 695                 kfree_skb(skb);
 696                 return 0;
 697         }
 698  out_err:
 699         nf_bridge_adjust_segmented_data(skb);
 700         return err;
 701 }
 702
 703 static int
 704 nfqnl_mangle(void *data, int data_len, struct nf_queue_entry *e, int diff)
 705 {
 706         struct sk_buff *nskb;
 707
 708         if (diff < 0) {
 709                 if (pskb_trim(e->skb, data_len))
 710                         return -ENOMEM;
 711         } else if (diff > 0) {
 712                 if (data_len > 0xFFFF)
 713                         return -EINVAL;
 714                 if (diff > skb_tailroom(e->skb)) {
 715                         nskb = skb_copy_expand(e->skb, skb_headroom(e->skb),
 716                                                diff, GFP_ATOMIC);
 717                         if (!nskb) {
 718                                 printk(KERN_WARNING "nf_queue: OOM "
 719                                       "in mangle, dropping packet\n");
 720                                 return -ENOMEM;
 721                         }
 722                         kfree_skb(e->skb);
 723                         e->skb = nskb;
 724                 }
 725                 skb_put(e->skb, diff);
 726         }
 727         if (!skb_make_writable(e->skb, data_len))
 728                 return -ENOMEM;
 729         skb_copy_to_linear_data(e->skb, data, data_len);
 730         e->skb->ip_summed = CHECKSUM_NONE;
 731         return 0;
 732 }
 733
 734 static int
 735 nfqnl_set_mode(struct nfqnl_instance *queue,
 736                unsigned char mode, unsigned int range)
 737 {
 738         int status = 0;
 739
 740         spin_lock_bh(&queue->lock);
 741         switch (mode) {
 742         case NFQNL_COPY_NONE:
 743         case NFQNL_COPY_META:
 744                 queue->copy_mode = mode;
 745                 queue->copy_range = 0;
 746                 break;
 747
 748         case NFQNL_COPY_PACKET:
 749                 queue->copy_mode = mode;
 750                 if (range == 0 || range > NFQNL_MAX_COPY_RANGE)
 751                         queue->copy_range = NFQNL_MAX_COPY_RANGE;
 752                 else
 753                         queue->copy_range = range;
 754                 break;
 755
 756         default:
 757                 status = -EINVAL;
 758
 759         }
 760         spin_unlock_bh(&queue->lock);
 761
 762         return status;
 763 }
 764
 765 static int
 766 dev_cmp(struct nf_queue_entry *entry, unsigned long ifindex)
 767 {
 768         if (entry->indev)
 769                 if (entry->indev->ifindex == ifindex)
 770                         return 1;
 771         if (entry->outdev)
 772                 if (entry->outdev->ifindex == ifindex)
 773                         return 1;
 774 #ifdef CONFIG_BRIDGE_NETFILTER
 775         if (entry->skb->nf_bridge) {
 776                 if (entry->skb->nf_bridge->physindev &&
 777                     entry->skb->nf_bridge->physindev->ifindex == ifindex)
 778                         return 1;
 779                 if (entry->skb->nf_bridge->physoutdev &&
 780                     entry->skb->nf_bridge->physoutdev->ifindex == ifindex)
 781                         return 1;
 782         }
 783 #endif
 784         return 0;
 785 }
 786
 787 /* drop all packets with either indev or outdev == ifindex from all queue
 788  * instances */
 789 static void
 790 nfqnl_dev_drop(struct net *net, int ifindex)
 791 {
 792         int i;
 793         struct nfnl_queue_net *q = nfnl_queue_pernet(net);
 794
 795         rcu_read_lock();
 796
 797         for (i = 0; i < INSTANCE_BUCKETS; i++) {
 798                 struct nfqnl_instance *inst;
 799                 struct hlist_head *head = &q->instance_table[i];
 800
 801                 hlist_for_each_entry_rcu(inst, head, hlist)
 802                         nfqnl_flush(inst, dev_cmp, ifindex);
 803         }
 804
 805         rcu_read_unlock();
 806 }
 807
 808 #define RCV_SKB_FAIL(err) do { netlink_ack(skb, nlh, (err)); return; } while (0)
 809
 810 static int
 811 nfqnl_rcv_dev_event(struct notifier_block *this,
 812                     unsigned long event, void *ptr)
 813 {
 814         struct net_device *dev = netdev_notifier_info_to_dev(ptr);
 815
 816         /* Drop any packets associated with the downed device */
 817         if (event == NETDEV_DOWN)
 818                 nfqnl_dev_drop(dev_net(dev), dev->ifindex);
 819         return NOTIFY_DONE;
 820 }
 821
 822 static struct notifier_block nfqnl_dev_notifier = {
 823         .notifier_call  = nfqnl_rcv_dev_event,
 824 };
 825
 826 static int
 827 nfqnl_rcv_nl_event(struct notifier_block *this,
 828                    unsigned long event, void *ptr)
 829 {
 830         struct netlink_notify *n = ptr;
 831         struct nfnl_queue_net *q = nfnl_queue_pernet(n->net);
 832
 833         if (event == NETLINK_URELEASE && n->protocol == NETLINK_NETFILTER) {
 834                 int i;
 835
 836                 /* destroy all instances for this portid */
 837                 spin_lock(&q->instances_lock);
 838                 for (i = 0; i < INSTANCE_BUCKETS; i++) {
 839                         struct hlist_node *t2;
 840                         struct nfqnl_instance *inst;
 841                         struct hlist_head *head = &q->instance_table[i];
 842
 843                         hlist_for_each_entry_safe(inst, t2, head, hlist) {
 844                                 if (n->portid == inst->peer_portid)
 845                                         __instance_destroy(inst);
 846                         }
 847                 }
 848                 spin_unlock(&q->instances_lock);
 849         }
 850         return NOTIFY_DONE;
 851 }
 852
 853 static struct notifier_block nfqnl_rtnl_notifier = {
 854         .notifier_call  = nfqnl_rcv_nl_event,
 855 };
 856
 857 static const struct nla_policy nfqa_verdict_policy[NFQA_MAX+1] = {
 858         [NFQA_VERDICT_HDR]      = { .len = sizeof(struct nfqnl_msg_verdict_hdr) },
 859         [NFQA_MARK]             = { .type = NLA_U32 },
 860         [NFQA_PAYLOAD]          = { .type = NLA_UNSPEC },
 861         [NFQA_CT]               = { .type = NLA_UNSPEC },
 862 };
 863
 864 static const struct nla_policy nfqa_verdict_batch_policy[NFQA_MAX+1] = {
 865         [NFQA_VERDICT_HDR]      = { .len = sizeof(struct nfqnl_msg_verdict_hdr) },
 866         [NFQA_MARK]             = { .type = NLA_U32 },
 867 };
 868
 869 static struct nfqnl_instance *
 870 verdict_instance_lookup(struct nfnl_queue_net *q, u16 queue_num, int nlportid)
 871 {
 872         struct nfqnl_instance *queue;
 873
 874         queue = instance_lookup(q, queue_num);
 875         if (!queue)
 876                 return ERR_PTR(-ENODEV);
 877
 878         if (queue->peer_portid != nlportid)
 879                 return ERR_PTR(-EPERM);
 880
 881         return queue;
 882 }
 883
 884 static struct nfqnl_msg_verdict_hdr*
 885 verdicthdr_get(const struct nlattr * const nfqa[])
 886 {
 887         struct nfqnl_msg_verdict_hdr *vhdr;
 888         unsigned int verdict;
 889
 890         if (!nfqa[NFQA_VERDICT_HDR])
 891                 return NULL;
 892
 893         vhdr = nla_data(nfqa[NFQA_VERDICT_HDR]);
 894         verdict = ntohl(vhdr->verdict) & NF_VERDICT_MASK;
 895         if (verdict > NF_MAX_VERDICT || verdict == NF_STOLEN)
 896                 return NULL;
 897         return vhdr;
 898 }
 899
 900 static int nfq_id_after(unsigned int id, unsigned int max)
 901 {
 902         return (int)(id - max) > 0;
 903 }
 904
 905 static int
 906 nfqnl_recv_verdict_batch(struct sock *ctnl, struct sk_buff *skb,
 907                    const struct nlmsghdr *nlh,
 908                    const struct nlattr * const nfqa[])
 909 {
 910         struct nfgenmsg *nfmsg = nlmsg_data(nlh);
 911         struct nf_queue_entry *entry, *tmp;
 912         unsigned int verdict, maxid;
 913         struct nfqnl_msg_verdict_hdr *vhdr;
 914         struct nfqnl_instance *queue;
 915         LIST_HEAD(batch_list);
 916         u16 queue_num = ntohs(nfmsg->res_id);
 917
 918         struct net *net = sock_net(ctnl);
 919         struct nfnl_queue_net *q = nfnl_queue_pernet(net);
 920
 921         queue = verdict_instance_lookup(q, queue_num,
 922                                         NETLINK_CB(skb).portid);
 923         if (IS_ERR(queue))
 924                 return PTR_ERR(queue);
 925
 926         vhdr = verdicthdr_get(nfqa);
 927         if (!vhdr)
 928                 return -EINVAL;
 929
 930         verdict = ntohl(vhdr->verdict);
 931         maxid = ntohl(vhdr->id);
 932
 933         spin_lock_bh(&queue->lock);
 934
 935         list_for_each_entry_safe(entry, tmp, &queue->queue_list, list) {
 936                 if (nfq_id_after(entry->id, maxid))
 937                         break;
 938                 __dequeue_entry(queue, entry);
 939                 list_add_tail(&entry->list, &batch_list);
 940         }
 941
 942         spin_unlock_bh(&queue->lock);
 943
 944         if (list_empty(&batch_list))
 945                 return -ENOENT;
 946
 947         list_for_each_entry_safe(entry, tmp, &batch_list, list) {
 948                 if (nfqa[NFQA_MARK])
 949                         entry->skb->mark = ntohl(nla_get_be32(nfqa[NFQA_MARK]));
 950                 nf_reinject(entry, verdict);
 951         }
 952         return 0;
 953 }
 954
 955 static int
 956 nfqnl_recv_verdict(struct sock *ctnl, struct sk_buff *skb,
 957                    const struct nlmsghdr *nlh,
 958                    const struct nlattr * const nfqa[])
 959 {
 960         struct nfgenmsg *nfmsg = nlmsg_data(nlh);
 961         u_int16_t queue_num = ntohs(nfmsg->res_id);
 962
 963         struct nfqnl_msg_verdict_hdr *vhdr;
 964         struct nfqnl_instance *queue;
 965         unsigned int verdict;
 966         struct nf_queue_entry *entry;
 967         enum ip_conntrack_info uninitialized_var(ctinfo);
 968         struct nf_conn *ct = NULL;
 969
 970         struct net *net = sock_net(ctnl);
 971         struct nfnl_queue_net *q = nfnl_queue_pernet(net);
 972
 973         queue = instance_lookup(q, queue_num);
 974         if (!queue)
 975                 queue = verdict_instance_lookup(q, queue_num,
 976                                                 NETLINK_CB(skb).portid);
 977         if (IS_ERR(queue))
 978                 return PTR_ERR(queue);
 979
 980         vhdr = verdicthdr_get(nfqa);
 981         if (!vhdr)
 982                 return -EINVAL;
 983
 984         verdict = ntohl(vhdr->verdict);
 985
 986         entry = find_dequeue_entry(queue, ntohl(vhdr->id));
 987         if (entry == NULL)
 988                 return -ENOENT;
 989
 990         rcu_read_lock();
 991         if (nfqa[NFQA_CT] && (queue->flags & NFQA_CFG_F_CONNTRACK))
 992                 ct = nfqnl_ct_parse(entry->skb, nfqa[NFQA_CT], &ctinfo);
 993
 994         if (nfqa[NFQA_PAYLOAD]) {
 995                 u16 payload_len = nla_len(nfqa[NFQA_PAYLOAD]);
 996                 int diff = payload_len - entry->skb->len;
 997
 998                 if (nfqnl_mangle(nla_data(nfqa[NFQA_PAYLOAD]),
 999                                  payload_len, entry, diff) < 0)
1000                         verdict = NF_DROP;
1001
1002                 if (ct)
1003                         nfqnl_ct_seq_adjust(skb, ct, ctinfo, diff);
1004         }
1005         rcu_read_unlock();
1006
1007         if (nfqa[NFQA_MARK])
1008                 entry->skb->mark = ntohl(nla_get_be32(nfqa[NFQA_MARK]));
1009
1010         nf_reinject(entry, verdict);
1011         return 0;
1012 }
1013
1014 static int
1015 nfqnl_recv_unsupp(struct sock *ctnl, struct sk_buff *skb,
1016                   const struct nlmsghdr *nlh,
1017                   const struct nlattr * const nfqa[])
1018 {
1019         return -ENOTSUPP;
1020 }
1021
1022 static const struct nla_policy nfqa_cfg_policy[NFQA_CFG_MAX+1] = {
1023         [NFQA_CFG_CMD]          = { .len = sizeof(struct nfqnl_msg_config_cmd) },
1024         [NFQA_CFG_PARAMS]       = { .len = sizeof(struct nfqnl_msg_config_params) },
1025 };
1026
1027 static const struct nf_queue_handler nfqh = {
1028         .outfn  = &nfqnl_enqueue_packet,
1029 };
1030
1031 static int
1032 nfqnl_recv_config(struct sock *ctnl, struct sk_buff *skb,
1033                   const struct nlmsghdr *nlh,
1034                   const struct nlattr * const nfqa[])
1035 {
1036         struct nfgenmsg *nfmsg = nlmsg_data(nlh);
1037         u_int16_t queue_num = ntohs(nfmsg->res_id);
1038         struct nfqnl_instance *queue;
1039         struct nfqnl_msg_config_cmd *cmd = NULL;
1040         struct net *net = sock_net(ctnl);
1041         struct nfnl_queue_net *q = nfnl_queue_pernet(net);
1042         int ret = 0;
1043
1044         if (nfqa[NFQA_CFG_CMD]) {
1045                 cmd = nla_data(nfqa[NFQA_CFG_CMD]);
1046
1047                 /* Obsolete commands without queue context */
1048                 switch (cmd->command) {
1049                 case NFQNL_CFG_CMD_PF_BIND: return 0;
1050                 case NFQNL_CFG_CMD_PF_UNBIND: return 0;
1051                 }
1052         }
1053
1054         rcu_read_lock();
1055         queue = instance_lookup(q, queue_num);
1056         if (queue && queue->peer_portid != NETLINK_CB(skb).portid) {
1057                 ret = -EPERM;
1058                 goto err_out_unlock;
1059         }
1060
1061         if (cmd != NULL) {
1062                 switch (cmd->command) {
1063                 case NFQNL_CFG_CMD_BIND:
1064                         if (queue) {
1065                                 ret = -EBUSY;
1066                                 goto err_out_unlock;
1067                         }
1068                         queue = instance_create(q, queue_num,
1069                                                 NETLINK_CB(skb).portid);
1070                         if (IS_ERR(queue)) {
1071                                 ret = PTR_ERR(queue);
1072                                 goto err_out_unlock;
1073                         }
1074                         break;
1075                 case NFQNL_CFG_CMD_UNBIND:
1076                         if (!queue) {
1077                                 ret = -ENODEV;
1078                                 goto err_out_unlock;
1079                         }
1080                         instance_destroy(q, queue);
1081                         break;
1082                 case NFQNL_CFG_CMD_PF_BIND:
1083                 case NFQNL_CFG_CMD_PF_UNBIND:
1084                         break;
1085                 default:
1086                         ret = -ENOTSUPP;
1087                         break;
1088                 }
1089         }
1090
1091         if (nfqa[NFQA_CFG_PARAMS]) {
1092                 struct nfqnl_msg_config_params *params;
1093
1094                 if (!queue) {
1095                         ret = -ENODEV;
1096                         goto err_out_unlock;
1097                 }
1098                 params = nla_data(nfqa[NFQA_CFG_PARAMS]);
1099                 nfqnl_set_mode(queue, params->copy_mode,
1100                                 ntohl(params->copy_range));
1101         }
1102
1103         if (nfqa[NFQA_CFG_QUEUE_MAXLEN]) {
1104                 __be32 *queue_maxlen;
1105
1106                 if (!queue) {
1107                         ret = -ENODEV;
1108                         goto err_out_unlock;
1109                 }
1110                 queue_maxlen = nla_data(nfqa[NFQA_CFG_QUEUE_MAXLEN]);
1111                 spin_lock_bh(&queue->lock);
1112                 queue->queue_maxlen = ntohl(*queue_maxlen);
1113                 spin_unlock_bh(&queue->lock);
1114         }
1115
1116         if (nfqa[NFQA_CFG_FLAGS]) {
1117                 __u32 flags, mask;
1118
1119                 if (!queue) {
1120                         ret = -ENODEV;
1121                         goto err_out_unlock;
1122                 }
1123
1124                 if (!nfqa[NFQA_CFG_MASK]) {
1125                         /* A mask is needed to specify which flags are being
1126                          * changed.
1127                          */
1128                         ret = -EINVAL;
1129                         goto err_out_unlock;
1130                 }
1131
1132                 flags = ntohl(nla_get_be32(nfqa[NFQA_CFG_FLAGS]));
1133                 mask = ntohl(nla_get_be32(nfqa[NFQA_CFG_MASK]));
1134
1135                 if (flags >= NFQA_CFG_F_MAX) {
1136                         ret = -EOPNOTSUPP;
1137                         goto err_out_unlock;
1138                 }
1139
1140                 spin_lock_bh(&queue->lock);
1141                 queue->flags &= ~mask;
1142                 queue->flags |= flags & mask;
1143                 spin_unlock_bh(&queue->lock);
1144         }
1145
1146 err_out_unlock:
1147         rcu_read_unlock();
1148         return ret;
1149 }
1150
1151 static const struct nfnl_callback nfqnl_cb[NFQNL_MSG_MAX] = {
1152         [NFQNL_MSG_PACKET]      = { .call_rcu = nfqnl_recv_unsupp,
1153                                     .attr_count = NFQA_MAX, },
1154         [NFQNL_MSG_VERDICT]     = { .call_rcu = nfqnl_recv_verdict,
1155                                     .attr_count = NFQA_MAX,
1156                                     .policy = nfqa_verdict_policy },
1157         [NFQNL_MSG_CONFIG]      = { .call = nfqnl_recv_config,
1158                                     .attr_count = NFQA_CFG_MAX,
1159                                     .policy = nfqa_cfg_policy },
1160         [NFQNL_MSG_VERDICT_BATCH]={ .call_rcu = nfqnl_recv_verdict_batch,
1161                                     .attr_count = NFQA_MAX,
1162                                     .policy = nfqa_verdict_batch_policy },
1163 };
1164
1165 static const struct nfnetlink_subsystem nfqnl_subsys = {
1166         .name           = "nf_queue",
1167         .subsys_id      = NFNL_SUBSYS_QUEUE,
1168         .cb_count       = NFQNL_MSG_MAX,
1169         .cb             = nfqnl_cb,
1170 };
1171
1172 #ifdef CONFIG_PROC_FS
1173 struct iter_state {
1174         struct seq_net_private p;
1175         unsigned int bucket;
1176 };
1177
1178 static struct hlist_node *get_first(struct seq_file *seq)
1179 {
1180         struct iter_state *st = seq->private;
1181         struct net *net;
1182         struct nfnl_queue_net *q;
1183
1184         if (!st)
1185                 return NULL;
1186
1187         net = seq_file_net(seq);
1188         q = nfnl_queue_pernet(net);
1189         for (st->bucket = 0; st->bucket < INSTANCE_BUCKETS; st->bucket++) {
1190                 if (!hlist_empty(&q->instance_table[st->bucket]))
1191                         return q->instance_table[st->bucket].first;
1192         }
1193         return NULL;
1194 }
1195
1196 static struct hlist_node *get_next(struct seq_file *seq, struct hlist_node *h)
1197 {
1198         struct iter_state *st = seq->private;
1199         struct net *net = seq_file_net(seq);
1200
1201         h = h->next;
1202         while (!h) {
1203                 struct nfnl_queue_net *q;
1204
1205                 if (++st->bucket >= INSTANCE_BUCKETS)
1206                         return NULL;
1207
1208                 q = nfnl_queue_pernet(net);
1209                 h = q->instance_table[st->bucket].first;
1210         }
1211         return h;
1212 }
1213
1214 static struct hlist_node *get_idx(struct seq_file *seq, loff_t pos)
1215 {
1216         struct hlist_node *head;
1217         head = get_first(seq);
1218
1219         if (head)
1220                 while (pos && (head = get_next(seq, head)))
1221                         pos--;
1222         return pos ? NULL : head;
1223 }
1224
1225 static void *seq_start(struct seq_file *s, loff_t *pos)
1226         __acquires(nfnl_queue_pernet(seq_file_net(s))->instances_lock)
1227 {
1228         spin_lock(&nfnl_queue_pernet(seq_file_net(s))->instances_lock);
1229         return get_idx(s, *pos);
1230 }
1231
1232 static void *seq_next(struct seq_file *s, void *v, loff_t *pos)
1233 {
1234         (*pos)++;
1235         return get_next(s, v);
1236 }
1237
1238 static void seq_stop(struct seq_file *s, void *v)
1239         __releases(nfnl_queue_pernet(seq_file_net(s))->instances_lock)
1240 {
1241         spin_unlock(&nfnl_queue_pernet(seq_file_net(s))->instances_lock);
1242 }
1243
1244 static int seq_show(struct seq_file *s, void *v)
1245 {
1246         const struct nfqnl_instance *inst = v;
1247
1248         return seq_printf(s, "%5d %6d %5d %1d %5d %5d %5d %8d %2d\n",
1249                           inst->queue_num,
1250                           inst->peer_portid, inst->queue_total,
1251                           inst->copy_mode, inst->copy_range,
1252                           inst->queue_dropped, inst->queue_user_dropped,
1253                           inst->id_sequence, 1);
1254 }
1255
1256 static const struct seq_operations nfqnl_seq_ops = {
1257         .start  = seq_start,
1258         .next   = seq_next,
1259         .stop   = seq_stop,
1260         .show   = seq_show,
1261 };
1262
1263 static int nfqnl_open(struct inode *inode, struct file *file)
1264 {
1265         return seq_open_net(inode, file, &nfqnl_seq_ops,
1266                         sizeof(struct iter_state));
1267 }
1268
1269 static const struct file_operations nfqnl_file_ops = {
1270         .owner   = THIS_MODULE,
1271         .open    = nfqnl_open,
1272         .read    = seq_read,
1273         .llseek  = seq_lseek,
1274         .release = seq_release_net,
1275 };
1276
1277 #endif /* PROC_FS */
1278
1279 static int __net_init nfnl_queue_net_init(struct net *net)
1280 {
1281         unsigned int i;
1282         struct nfnl_queue_net *q = nfnl_queue_pernet(net);
1283
1284         for (i = 0; i < INSTANCE_BUCKETS; i++)
1285                 INIT_HLIST_HEAD(&q->instance_table[i]);
1286
1287         spin_lock_init(&q->instances_lock);
1288
1289 #ifdef CONFIG_PROC_FS
1290         if (!proc_create("nfnetlink_queue", 0440,
1291                          net->nf.proc_netfilter, &nfqnl_file_ops))
1292                 return -ENOMEM;
1293 #endif
1294         return 0;
1295 }
1296
1297 static void __net_exit nfnl_queue_net_exit(struct net *net)
1298 {
1299 #ifdef CONFIG_PROC_FS
1300         remove_proc_entry("nfnetlink_queue", net->nf.proc_netfilter);
1301 #endif
1302 }
1303
1304 static struct pernet_operations nfnl_queue_net_ops = {
1305         .init   = nfnl_queue_net_init,
1306         .exit   = nfnl_queue_net_exit,
1307         .id     = &nfnl_queue_net_id,
1308         .size   = sizeof(struct nfnl_queue_net),
1309 };
1310
1311 static int __init nfnetlink_queue_init(void)
1312 {
1313         int status = -ENOMEM;
1314
1315         netlink_register_notifier(&nfqnl_rtnl_notifier);
1316         status = nfnetlink_subsys_register(&nfqnl_subsys);
1317         if (status < 0) {
1318                 pr_err("nf_queue: failed to create netlink socket\n");
1319                 goto cleanup_netlink_notifier;
1320         }
1321
1322         status = register_pernet_subsys(&nfnl_queue_net_ops);
1323         if (status < 0) {
1324                 pr_err("nf_queue: failed to register pernet ops\n");
1325                 goto cleanup_subsys;
1326         }
1327         register_netdevice_notifier(&nfqnl_dev_notifier);
1328         nf_register_queue_handler(&nfqh);
1329         return status;
1330
1331 cleanup_subsys:
1332         nfnetlink_subsys_unregister(&nfqnl_subsys);
1333 cleanup_netlink_notifier:
1334         netlink_unregister_notifier(&nfqnl_rtnl_notifier);
1335         return status;
1336 }
1337
1338 static void __exit nfnetlink_queue_fini(void)
1339 {
1340         nf_unregister_queue_handler();
1341         unregister_netdevice_notifier(&nfqnl_dev_notifier);
1342         unregister_pernet_subsys(&nfnl_queue_net_ops);
1343         nfnetlink_subsys_unregister(&nfqnl_subsys);
1344         netlink_unregister_notifier(&nfqnl_rtnl_notifier);
1345
1346         rcu_barrier(); /* Wait for completion of call_rcu()'s */
1347 }
1348
1349 MODULE_DESCRIPTION("netfilter packet queue handler");
1350 MODULE_AUTHOR("Harald Welte <laforge@netfilter.org>");
1351 MODULE_LICENSE("GPL");
1352 MODULE_ALIAS_NFNL_SUBSYS(NFNL_SUBSYS_QUEUE);
1353
1354 module_init(nfnetlink_queue_init);
1355 module_exit(nfnetlink_queue_fini);