net/netfilter/nfnetlink_queue_core.c

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