net/netfilter/nfnetlink_queue.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
  34 #include <asm/atomic.h>
  35
  36 #ifdef CONFIG_BRIDGE_NETFILTER
  37 #include "../bridge/br_private.h"
  38 #endif
  39
  40 #define NFQNL_QMAX_DEFAULT 1024
  41
  42 struct nfqnl_instance {
  43         struct hlist_node hlist;                /* global list of queues */
  44         struct rcu_head rcu;
  45
  46         int peer_pid;
  47         unsigned int queue_maxlen;
  48         unsigned int copy_range;
  49         unsigned int queue_dropped;
  50         unsigned int queue_user_dropped;
  51
  52
  53         u_int16_t queue_num;                    /* number of this queue */
  54         u_int8_t copy_mode;
  55 /*
  56  * Following fields are dirtied for each queued packet,
  57  * keep them in same cache line if possible.
  58  */
  59         spinlock_t      lock;
  60         unsigned int    queue_total;
  61         atomic_t        id_sequence;            /* 'sequence' of pkt ids */
  62         struct list_head queue_list;            /* packets in queue */
  63 };
  64
  65 typedef int (*nfqnl_cmpfn)(struct nf_queue_entry *, unsigned long);
  66
  67 static DEFINE_SPINLOCK(instances_lock);
  68
  69 #define INSTANCE_BUCKETS        16
  70 static struct hlist_head instance_table[INSTANCE_BUCKETS] __read_mostly;
  71
  72 static inline u_int8_t instance_hashfn(u_int16_t queue_num)
  73 {
  74         return ((queue_num >> 8) | queue_num) % INSTANCE_BUCKETS;
  75 }
  76
  77 static struct nfqnl_instance *
  78 instance_lookup(u_int16_t queue_num)
  79 {
  80         struct hlist_head *head;
  81         struct hlist_node *pos;
  82         struct nfqnl_instance *inst;
  83
  84         head = &instance_table[instance_hashfn(queue_num)];
  85         hlist_for_each_entry_rcu(inst, pos, head, hlist) {
  86                 if (inst->queue_num == queue_num)
  87                         return inst;
  88         }
  89         return NULL;
  90 }
  91
  92 static struct nfqnl_instance *
  93 instance_create(u_int16_t queue_num, int pid)
  94 {
  95         struct nfqnl_instance *inst;
  96         unsigned int h;
  97         int err;
  98
  99         spin_lock(&instances_lock);
 100         if (instance_lookup(queue_num)) {
 101                 err = -EEXIST;
 102                 goto out_unlock;
 103         }
 104
 105         inst = kzalloc(sizeof(*inst), GFP_ATOMIC);
 106         if (!inst) {
 107                 err = -ENOMEM;
 108                 goto out_unlock;
 109         }
 110
 111         inst->queue_num = queue_num;
 112         inst->peer_pid = pid;
 113         inst->queue_maxlen = NFQNL_QMAX_DEFAULT;
 114         inst->copy_range = 0xfffff;
 115         inst->copy_mode = NFQNL_COPY_NONE;
 116         spin_lock_init(&inst->lock);
 117         INIT_LIST_HEAD(&inst->queue_list);
 118
 119         if (!try_module_get(THIS_MODULE)) {
 120                 err = -EAGAIN;
 121                 goto out_free;
 122         }
 123
 124         h = instance_hashfn(queue_num);
 125         hlist_add_head_rcu(&inst->hlist, &instance_table[h]);
 126
 127         spin_unlock(&instances_lock);
 128
 129         return inst;
 130
 131 out_free:
 132         kfree(inst);
 133 out_unlock:
 134         spin_unlock(&instances_lock);
 135         return ERR_PTR(err);
 136 }
 137
 138 static void nfqnl_flush(struct nfqnl_instance *queue, nfqnl_cmpfn cmpfn,
 139                         unsigned long data);
 140
 141 static void
 142 instance_destroy_rcu(struct rcu_head *head)
 143 {
 144         struct nfqnl_instance *inst = container_of(head, struct nfqnl_instance,
 145                                                    rcu);
 146
 147         nfqnl_flush(inst, NULL, 0);
 148         kfree(inst);
 149         module_put(THIS_MODULE);
 150 }
 151
 152 static void
 153 __instance_destroy(struct nfqnl_instance *inst)
 154 {
 155         hlist_del_rcu(&inst->hlist);
 156         call_rcu(&inst->rcu, instance_destroy_rcu);
 157 }
 158
 159 static void
 160 instance_destroy(struct nfqnl_instance *inst)
 161 {
 162         spin_lock(&instances_lock);
 163         __instance_destroy(inst);
 164         spin_unlock(&instances_lock);
 165 }
 166
 167 static inline void
 168 __enqueue_entry(struct nfqnl_instance *queue, struct nf_queue_entry *entry)
 169 {
 170        list_add_tail(&entry->list, &queue->queue_list);
 171        queue->queue_total++;
 172 }
 173
 174 static struct nf_queue_entry *
 175 find_dequeue_entry(struct nfqnl_instance *queue, unsigned int id)
 176 {
 177         struct nf_queue_entry *entry = NULL, *i;
 178
 179         spin_lock_bh(&queue->lock);
 180
 181         list_for_each_entry(i, &queue->queue_list, list) {
 182                 if (i->id == id) {
 183                         entry = i;
 184                         break;
 185                 }
 186         }
 187
 188         if (entry) {
 189                 list_del(&entry->list);
 190                 queue->queue_total--;
 191         }
 192
 193         spin_unlock_bh(&queue->lock);
 194
 195         return entry;
 196 }
 197
 198 static void
 199 nfqnl_flush(struct nfqnl_instance *queue, nfqnl_cmpfn cmpfn, unsigned long data)
 200 {
 201         struct nf_queue_entry *entry, *next;
 202
 203         spin_lock_bh(&queue->lock);
 204         list_for_each_entry_safe(entry, next, &queue->queue_list, list) {
 205                 if (!cmpfn || cmpfn(entry, data)) {
 206                         list_del(&entry->list);
 207                         queue->queue_total--;
 208                         nf_reinject(entry, NF_DROP);
 209                 }
 210         }
 211         spin_unlock_bh(&queue->lock);
 212 }
 213
 214 static struct sk_buff *
 215 nfqnl_build_packet_message(struct nfqnl_instance *queue,
 216                            struct nf_queue_entry *entry)
 217 {
 218         sk_buff_data_t old_tail;
 219         size_t size;
 220         size_t data_len = 0;
 221         struct sk_buff *skb;
 222         struct nfqnl_msg_packet_hdr pmsg;
 223         struct nlmsghdr *nlh;
 224         struct nfgenmsg *nfmsg;
 225         struct sk_buff *entskb = entry->skb;
 226         struct net_device *indev;
 227         struct net_device *outdev;
 228
 229         size =    NLMSG_SPACE(sizeof(struct nfgenmsg))
 230                 + nla_total_size(sizeof(struct nfqnl_msg_packet_hdr))
 231                 + nla_total_size(sizeof(u_int32_t))     /* ifindex */
 232                 + nla_total_size(sizeof(u_int32_t))     /* ifindex */
 233 #ifdef CONFIG_BRIDGE_NETFILTER
 234                 + nla_total_size(sizeof(u_int32_t))     /* ifindex */
 235                 + nla_total_size(sizeof(u_int32_t))     /* ifindex */
 236 #endif
 237                 + nla_total_size(sizeof(u_int32_t))     /* mark */
 238                 + nla_total_size(sizeof(struct nfqnl_msg_packet_hw))
 239                 + nla_total_size(sizeof(struct nfqnl_msg_packet_timestamp));
 240
 241         outdev = entry->outdev;
 242
 243         switch ((enum nfqnl_config_mode)ACCESS_ONCE(queue->copy_mode)) {
 244         case NFQNL_COPY_META:
 245         case NFQNL_COPY_NONE:
 246                 break;
 247
 248         case NFQNL_COPY_PACKET:
 249                 if (entskb->ip_summed == CHECKSUM_PARTIAL &&
 250                     skb_checksum_help(entskb))
 251                         return NULL;
 252
 253                 data_len = ACCESS_ONCE(queue->copy_range);
 254                 if (data_len == 0 || data_len > entskb->len)
 255                         data_len = entskb->len;
 256
 257                 size += nla_total_size(data_len);
 258                 break;
 259         }
 260
 261
 262         skb = alloc_skb(size, GFP_ATOMIC);
 263         if (!skb)
 264                 goto nlmsg_failure;
 265
 266         old_tail = skb->tail;
 267         nlh = NLMSG_PUT(skb, 0, 0,
 268                         NFNL_SUBSYS_QUEUE << 8 | NFQNL_MSG_PACKET,
 269                         sizeof(struct nfgenmsg));
 270         nfmsg = NLMSG_DATA(nlh);
 271         nfmsg->nfgen_family = entry->pf;
 272         nfmsg->version = NFNETLINK_V0;
 273         nfmsg->res_id = htons(queue->queue_num);
 274
 275         entry->id = atomic_inc_return(&queue->id_sequence);
 276         pmsg.packet_id          = htonl(entry->id);
 277         pmsg.hw_protocol        = entskb->protocol;
 278         pmsg.hook               = entry->hook;
 279
 280         NLA_PUT(skb, NFQA_PACKET_HDR, sizeof(pmsg), &pmsg);
 281
 282         indev = entry->indev;
 283         if (indev) {
 284 #ifndef CONFIG_BRIDGE_NETFILTER
 285                 NLA_PUT_BE32(skb, NFQA_IFINDEX_INDEV, htonl(indev->ifindex));
 286 #else
 287                 if (entry->pf == PF_BRIDGE) {
 288                         /* Case 1: indev is physical input device, we need to
 289                          * look for bridge group (when called from
 290                          * netfilter_bridge) */
 291                         NLA_PUT_BE32(skb, NFQA_IFINDEX_PHYSINDEV,
 292                                      htonl(indev->ifindex));
 293                         /* this is the bridge group "brX" */
 294                         /* rcu_read_lock()ed by __nf_queue */
 295                         NLA_PUT_BE32(skb, NFQA_IFINDEX_INDEV,
 296                                      htonl(br_port_get_rcu(indev)->br->dev->ifindex));
 297                 } else {
 298                         /* Case 2: indev is bridge group, we need to look for
 299                          * physical device (when called from ipv4) */
 300                         NLA_PUT_BE32(skb, NFQA_IFINDEX_INDEV,
 301                                      htonl(indev->ifindex));
 302                         if (entskb->nf_bridge && entskb->nf_bridge->physindev)
 303                                 NLA_PUT_BE32(skb, NFQA_IFINDEX_PHYSINDEV,
 304                                              htonl(entskb->nf_bridge->physindev->ifindex));
 305                 }
 306 #endif
 307         }
 308
 309         if (outdev) {
 310 #ifndef CONFIG_BRIDGE_NETFILTER
 311                 NLA_PUT_BE32(skb, NFQA_IFINDEX_OUTDEV, htonl(outdev->ifindex));
 312 #else
 313                 if (entry->pf == PF_BRIDGE) {
 314                         /* Case 1: outdev is physical output device, we need to
 315                          * look for bridge group (when called from
 316                          * netfilter_bridge) */
 317                         NLA_PUT_BE32(skb, NFQA_IFINDEX_PHYSOUTDEV,
 318                                      htonl(outdev->ifindex));
 319                         /* this is the bridge group "brX" */
 320                         /* rcu_read_lock()ed by __nf_queue */
 321                         NLA_PUT_BE32(skb, NFQA_IFINDEX_OUTDEV,
 322                                      htonl(br_port_get_rcu(outdev)->br->dev->ifindex));
 323                 } else {
 324                         /* Case 2: outdev is bridge group, we need to look for
 325                          * physical output device (when called from ipv4) */
 326                         NLA_PUT_BE32(skb, NFQA_IFINDEX_OUTDEV,
 327                                      htonl(outdev->ifindex));
 328                         if (entskb->nf_bridge && entskb->nf_bridge->physoutdev)
 329                                 NLA_PUT_BE32(skb, NFQA_IFINDEX_PHYSOUTDEV,
 330                                              htonl(entskb->nf_bridge->physoutdev->ifindex));
 331                 }
 332 #endif
 333         }
 334
 335         if (entskb->mark)
 336                 NLA_PUT_BE32(skb, NFQA_MARK, htonl(entskb->mark));
 337
 338         if (indev && entskb->dev &&
 339             entskb->mac_header != entskb->network_header) {
 340                 struct nfqnl_msg_packet_hw phw;
 341                 int len = dev_parse_header(entskb, phw.hw_addr);
 342                 if (len) {
 343                         phw.hw_addrlen = htons(len);
 344                         NLA_PUT(skb, NFQA_HWADDR, sizeof(phw), &phw);
 345                 }
 346         }
 347
 348         if (entskb->tstamp.tv64) {
 349                 struct nfqnl_msg_packet_timestamp ts;
 350                 struct timeval tv = ktime_to_timeval(entskb->tstamp);
 351                 ts.sec = cpu_to_be64(tv.tv_sec);
 352                 ts.usec = cpu_to_be64(tv.tv_usec);
 353
 354                 NLA_PUT(skb, NFQA_TIMESTAMP, sizeof(ts), &ts);
 355         }
 356
 357         if (data_len) {
 358                 struct nlattr *nla;
 359                 int sz = nla_attr_size(data_len);
 360
 361                 if (skb_tailroom(skb) < nla_total_size(data_len)) {
 362                         printk(KERN_WARNING "nf_queue: no tailroom!\n");
 363                         goto nlmsg_failure;
 364                 }
 365
 366                 nla = (struct nlattr *)skb_put(skb, nla_total_size(data_len));
 367                 nla->nla_type = NFQA_PAYLOAD;
 368                 nla->nla_len = sz;
 369
 370                 if (skb_copy_bits(entskb, 0, nla_data(nla), data_len))
 371                         BUG();
 372         }
 373
 374         nlh->nlmsg_len = skb->tail - old_tail;
 375         return skb;
 376
 377 nlmsg_failure:
 378 nla_put_failure:
 379         if (skb)
 380                 kfree_skb(skb);
 381         if (net_ratelimit())
 382                 printk(KERN_ERR "nf_queue: error creating packet message\n");
 383         return NULL;
 384 }
 385
 386 static int
 387 nfqnl_enqueue_packet(struct nf_queue_entry *entry, unsigned int queuenum)
 388 {
 389         struct sk_buff *nskb;
 390         struct nfqnl_instance *queue;
 391         int err = -ENOBUFS;
 392
 393         /* rcu_read_lock()ed by nf_hook_slow() */
 394         queue = instance_lookup(queuenum);
 395         if (!queue) {
 396                 err = -ESRCH;
 397                 goto err_out;
 398         }
 399
 400         if (queue->copy_mode == NFQNL_COPY_NONE) {
 401                 err = -EINVAL;
 402                 goto err_out;
 403         }
 404
 405         nskb = nfqnl_build_packet_message(queue, entry);
 406         if (nskb == NULL) {
 407                 err = -ENOMEM;
 408                 goto err_out;
 409         }
 410         spin_lock_bh(&queue->lock);
 411
 412         if (!queue->peer_pid) {
 413                 err = -EINVAL;
 414                 goto err_out_free_nskb;
 415         }
 416         if (queue->queue_total >= queue->queue_maxlen) {
 417                 queue->queue_dropped++;
 418                 if (net_ratelimit())
 419                           printk(KERN_WARNING "nf_queue: full at %d entries, "
 420                                  "dropping packets(s).\n",
 421                                  queue->queue_total);
 422                 goto err_out_free_nskb;
 423         }
 424
 425         /* nfnetlink_unicast will either free the nskb or add it to a socket */
 426         err = nfnetlink_unicast(nskb, &init_net, queue->peer_pid, MSG_DONTWAIT);
 427         if (err < 0) {
 428                 queue->queue_user_dropped++;
 429                 goto err_out_unlock;
 430         }
 431
 432         __enqueue_entry(queue, entry);
 433
 434         spin_unlock_bh(&queue->lock);
 435         return 0;
 436
 437 err_out_free_nskb:
 438         kfree_skb(nskb);
 439 err_out_unlock:
 440         spin_unlock_bh(&queue->lock);
 441 err_out:
 442         return err;
 443 }
 444
 445 static int
 446 nfqnl_mangle(void *data, int data_len, struct nf_queue_entry *e)
 447 {
 448         struct sk_buff *nskb;
 449         int diff;
 450
 451         diff = data_len - e->skb->len;
 452         if (diff < 0) {
 453                 if (pskb_trim(e->skb, data_len))
 454                         return -ENOMEM;
 455         } else if (diff > 0) {
 456                 if (data_len > 0xFFFF)
 457                         return -EINVAL;
 458                 if (diff > skb_tailroom(e->skb)) {
 459                         nskb = skb_copy_expand(e->skb, skb_headroom(e->skb),
 460                                                diff, GFP_ATOMIC);
 461                         if (!nskb) {
 462                                 printk(KERN_WARNING "nf_queue: OOM "
 463                                       "in mangle, dropping packet\n");
 464                                 return -ENOMEM;
 465                         }
 466                         kfree_skb(e->skb);
 467                         e->skb = nskb;
 468                 }
 469                 skb_put(e->skb, diff);
 470         }
 471         if (!skb_make_writable(e->skb, data_len))
 472                 return -ENOMEM;
 473         skb_copy_to_linear_data(e->skb, data, data_len);
 474         e->skb->ip_summed = CHECKSUM_NONE;
 475         return 0;
 476 }
 477
 478 static int
 479 nfqnl_set_mode(struct nfqnl_instance *queue,
 480                unsigned char mode, unsigned int range)
 481 {
 482         int status = 0;
 483
 484         spin_lock_bh(&queue->lock);
 485         switch (mode) {
 486         case NFQNL_COPY_NONE:
 487         case NFQNL_COPY_META:
 488                 queue->copy_mode = mode;
 489                 queue->copy_range = 0;
 490                 break;
 491
 492         case NFQNL_COPY_PACKET:
 493                 queue->copy_mode = mode;
 494                 /* we're using struct nlattr which has 16bit nla_len */
 495                 if (range > 0xffff)
 496                         queue->copy_range = 0xffff;
 497                 else
 498                         queue->copy_range = range;
 499                 break;
 500
 501         default:
 502                 status = -EINVAL;
 503
 504         }
 505         spin_unlock_bh(&queue->lock);
 506
 507         return status;
 508 }
 509
 510 static int
 511 dev_cmp(struct nf_queue_entry *entry, unsigned long ifindex)
 512 {
 513         if (entry->indev)
 514                 if (entry->indev->ifindex == ifindex)
 515                         return 1;
 516         if (entry->outdev)
 517                 if (entry->outdev->ifindex == ifindex)
 518                         return 1;
 519 #ifdef CONFIG_BRIDGE_NETFILTER
 520         if (entry->skb->nf_bridge) {
 521                 if (entry->skb->nf_bridge->physindev &&
 522                     entry->skb->nf_bridge->physindev->ifindex == ifindex)
 523                         return 1;
 524                 if (entry->skb->nf_bridge->physoutdev &&
 525                     entry->skb->nf_bridge->physoutdev->ifindex == ifindex)
 526                         return 1;
 527         }
 528 #endif
 529         return 0;
 530 }
 531
 532 /* drop all packets with either indev or outdev == ifindex from all queue
 533  * instances */
 534 static void
 535 nfqnl_dev_drop(int ifindex)
 536 {
 537         int i;
 538
 539         rcu_read_lock();
 540
 541         for (i = 0; i < INSTANCE_BUCKETS; i++) {
 542                 struct hlist_node *tmp;
 543                 struct nfqnl_instance *inst;
 544                 struct hlist_head *head = &instance_table[i];
 545
 546                 hlist_for_each_entry_rcu(inst, tmp, head, hlist)
 547                         nfqnl_flush(inst, dev_cmp, ifindex);
 548         }
 549
 550         rcu_read_unlock();
 551 }
 552
 553 #define RCV_SKB_FAIL(err) do { netlink_ack(skb, nlh, (err)); return; } while (0)
 554
 555 static int
 556 nfqnl_rcv_dev_event(struct notifier_block *this,
 557                     unsigned long event, void *ptr)
 558 {
 559         struct net_device *dev = ptr;
 560
 561         if (!net_eq(dev_net(dev), &init_net))
 562                 return NOTIFY_DONE;
 563
 564         /* Drop any packets associated with the downed device */
 565         if (event == NETDEV_DOWN)
 566                 nfqnl_dev_drop(dev->ifindex);
 567         return NOTIFY_DONE;
 568 }
 569
 570 static struct notifier_block nfqnl_dev_notifier = {
 571         .notifier_call  = nfqnl_rcv_dev_event,
 572 };
 573
 574 static int
 575 nfqnl_rcv_nl_event(struct notifier_block *this,
 576                    unsigned long event, void *ptr)
 577 {
 578         struct netlink_notify *n = ptr;
 579
 580         if (event == NETLINK_URELEASE && n->protocol == NETLINK_NETFILTER) {
 581                 int i;
 582
 583                 /* destroy all instances for this pid */
 584                 spin_lock(&instances_lock);
 585                 for (i = 0; i < INSTANCE_BUCKETS; i++) {
 586                         struct hlist_node *tmp, *t2;
 587                         struct nfqnl_instance *inst;
 588                         struct hlist_head *head = &instance_table[i];
 589
 590                         hlist_for_each_entry_safe(inst, tmp, t2, head, hlist) {
 591                                 if ((n->net == &init_net) &&
 592                                     (n->pid == inst->peer_pid))
 593                                         __instance_destroy(inst);
 594                         }
 595                 }
 596                 spin_unlock(&instances_lock);
 597         }
 598         return NOTIFY_DONE;
 599 }
 600
 601 static struct notifier_block nfqnl_rtnl_notifier = {
 602         .notifier_call  = nfqnl_rcv_nl_event,
 603 };
 604
 605 static const struct nla_policy nfqa_verdict_policy[NFQA_MAX+1] = {
 606         [NFQA_VERDICT_HDR]      = { .len = sizeof(struct nfqnl_msg_verdict_hdr) },
 607         [NFQA_MARK]             = { .type = NLA_U32 },
 608         [NFQA_PAYLOAD]          = { .type = NLA_UNSPEC },
 609 };
 610
 611 static int
 612 nfqnl_recv_verdict(struct sock *ctnl, struct sk_buff *skb,
 613                    const struct nlmsghdr *nlh,
 614                    const struct nlattr * const nfqa[])
 615 {
 616         struct nfgenmsg *nfmsg = NLMSG_DATA(nlh);
 617         u_int16_t queue_num = ntohs(nfmsg->res_id);
 618
 619         struct nfqnl_msg_verdict_hdr *vhdr;
 620         struct nfqnl_instance *queue;
 621         unsigned int verdict;
 622         struct nf_queue_entry *entry;
 623         int err;
 624
 625         rcu_read_lock();
 626         queue = instance_lookup(queue_num);
 627         if (!queue) {
 628                 err = -ENODEV;
 629                 goto err_out_unlock;
 630         }
 631
 632         if (queue->peer_pid != NETLINK_CB(skb).pid) {
 633                 err = -EPERM;
 634                 goto err_out_unlock;
 635         }
 636
 637         if (!nfqa[NFQA_VERDICT_HDR]) {
 638                 err = -EINVAL;
 639                 goto err_out_unlock;
 640         }
 641
 642         vhdr = nla_data(nfqa[NFQA_VERDICT_HDR]);
 643         verdict = ntohl(vhdr->verdict);
 644
 645         if ((verdict & NF_VERDICT_MASK) > NF_MAX_VERDICT) {
 646                 err = -EINVAL;
 647                 goto err_out_unlock;
 648         }
 649
 650         entry = find_dequeue_entry(queue, ntohl(vhdr->id));
 651         if (entry == NULL) {
 652                 err = -ENOENT;
 653                 goto err_out_unlock;
 654         }
 655         rcu_read_unlock();
 656
 657         if (nfqa[NFQA_PAYLOAD]) {
 658                 if (nfqnl_mangle(nla_data(nfqa[NFQA_PAYLOAD]),
 659                                  nla_len(nfqa[NFQA_PAYLOAD]), entry) < 0)
 660                         verdict = NF_DROP;
 661         }
 662
 663         if (nfqa[NFQA_MARK])
 664                 entry->skb->mark = ntohl(nla_get_be32(nfqa[NFQA_MARK]));
 665
 666         nf_reinject(entry, verdict);
 667         return 0;
 668
 669 err_out_unlock:
 670         rcu_read_unlock();
 671         return err;
 672 }
 673
 674 static int
 675 nfqnl_recv_unsupp(struct sock *ctnl, struct sk_buff *skb,
 676                   const struct nlmsghdr *nlh,
 677                   const struct nlattr * const nfqa[])
 678 {
 679         return -ENOTSUPP;
 680 }
 681
 682 static const struct nla_policy nfqa_cfg_policy[NFQA_CFG_MAX+1] = {
 683         [NFQA_CFG_CMD]          = { .len = sizeof(struct nfqnl_msg_config_cmd) },
 684         [NFQA_CFG_PARAMS]       = { .len = sizeof(struct nfqnl_msg_config_params) },
 685 };
 686
 687 static const struct nf_queue_handler nfqh = {
 688         .name   = "nf_queue",
 689         .outfn  = &nfqnl_enqueue_packet,
 690 };
 691
 692 static int
 693 nfqnl_recv_config(struct sock *ctnl, struct sk_buff *skb,
 694                   const struct nlmsghdr *nlh,
 695                   const struct nlattr * const nfqa[])
 696 {
 697         struct nfgenmsg *nfmsg = NLMSG_DATA(nlh);
 698         u_int16_t queue_num = ntohs(nfmsg->res_id);
 699         struct nfqnl_instance *queue;
 700         struct nfqnl_msg_config_cmd *cmd = NULL;
 701         int ret = 0;
 702
 703         if (nfqa[NFQA_CFG_CMD]) {
 704                 cmd = nla_data(nfqa[NFQA_CFG_CMD]);
 705
 706                 /* Commands without queue context - might sleep */
 707                 switch (cmd->command) {
 708                 case NFQNL_CFG_CMD_PF_BIND:
 709                         return nf_register_queue_handler(ntohs(cmd->pf),
 710                                                          &nfqh);
 711                 case NFQNL_CFG_CMD_PF_UNBIND:
 712                         return nf_unregister_queue_handler(ntohs(cmd->pf),
 713                                                            &nfqh);
 714                 }
 715         }
 716
 717         rcu_read_lock();
 718         queue = instance_lookup(queue_num);
 719         if (queue && queue->peer_pid != NETLINK_CB(skb).pid) {
 720                 ret = -EPERM;
 721                 goto err_out_unlock;
 722         }
 723
 724         if (cmd != NULL) {
 725                 switch (cmd->command) {
 726                 case NFQNL_CFG_CMD_BIND:
 727                         if (queue) {
 728                                 ret = -EBUSY;
 729                                 goto err_out_unlock;
 730                         }
 731                         queue = instance_create(queue_num, NETLINK_CB(skb).pid);
 732                         if (IS_ERR(queue)) {
 733                                 ret = PTR_ERR(queue);
 734                                 goto err_out_unlock;
 735                         }
 736                         break;
 737                 case NFQNL_CFG_CMD_UNBIND:
 738                         if (!queue) {
 739                                 ret = -ENODEV;
 740                                 goto err_out_unlock;
 741                         }
 742                         instance_destroy(queue);
 743                         break;
 744                 case NFQNL_CFG_CMD_PF_BIND:
 745                 case NFQNL_CFG_CMD_PF_UNBIND:
 746                         break;
 747                 default:
 748                         ret = -ENOTSUPP;
 749                         break;
 750                 }
 751         }
 752
 753         if (nfqa[NFQA_CFG_PARAMS]) {
 754                 struct nfqnl_msg_config_params *params;
 755
 756                 if (!queue) {
 757                         ret = -ENODEV;
 758                         goto err_out_unlock;
 759                 }
 760                 params = nla_data(nfqa[NFQA_CFG_PARAMS]);
 761                 nfqnl_set_mode(queue, params->copy_mode,
 762                                 ntohl(params->copy_range));
 763         }
 764
 765         if (nfqa[NFQA_CFG_QUEUE_MAXLEN]) {
 766                 __be32 *queue_maxlen;
 767
 768                 if (!queue) {
 769                         ret = -ENODEV;
 770                         goto err_out_unlock;
 771                 }
 772                 queue_maxlen = nla_data(nfqa[NFQA_CFG_QUEUE_MAXLEN]);
 773                 spin_lock_bh(&queue->lock);
 774                 queue->queue_maxlen = ntohl(*queue_maxlen);
 775                 spin_unlock_bh(&queue->lock);
 776         }
 777
 778 err_out_unlock:
 779         rcu_read_unlock();
 780         return ret;
 781 }
 782
 783 static const struct nfnl_callback nfqnl_cb[NFQNL_MSG_MAX] = {
 784         [NFQNL_MSG_PACKET]      = { .call = nfqnl_recv_unsupp,
 785                                     .attr_count = NFQA_MAX, },
 786         [NFQNL_MSG_VERDICT]     = { .call = nfqnl_recv_verdict,
 787                                     .attr_count = NFQA_MAX,
 788                                     .policy = nfqa_verdict_policy },
 789         [NFQNL_MSG_CONFIG]      = { .call = nfqnl_recv_config,
 790                                     .attr_count = NFQA_CFG_MAX,
 791                                     .policy = nfqa_cfg_policy },
 792 };
 793
 794 static const struct nfnetlink_subsystem nfqnl_subsys = {
 795         .name           = "nf_queue",
 796         .subsys_id      = NFNL_SUBSYS_QUEUE,
 797         .cb_count       = NFQNL_MSG_MAX,
 798         .cb             = nfqnl_cb,
 799 };
 800
 801 #ifdef CONFIG_PROC_FS
 802 struct iter_state {
 803         unsigned int bucket;
 804 };
 805
 806 static struct hlist_node *get_first(struct seq_file *seq)
 807 {
 808         struct iter_state *st = seq->private;
 809
 810         if (!st)
 811                 return NULL;
 812
 813         for (st->bucket = 0; st->bucket < INSTANCE_BUCKETS; st->bucket++) {
 814                 if (!hlist_empty(&instance_table[st->bucket]))
 815                         return instance_table[st->bucket].first;
 816         }
 817         return NULL;
 818 }
 819
 820 static struct hlist_node *get_next(struct seq_file *seq, struct hlist_node *h)
 821 {
 822         struct iter_state *st = seq->private;
 823
 824         h = h->next;
 825         while (!h) {
 826                 if (++st->bucket >= INSTANCE_BUCKETS)
 827                         return NULL;
 828
 829                 h = instance_table[st->bucket].first;
 830         }
 831         return h;
 832 }
 833
 834 static struct hlist_node *get_idx(struct seq_file *seq, loff_t pos)
 835 {
 836         struct hlist_node *head;
 837         head = get_first(seq);
 838
 839         if (head)
 840                 while (pos && (head = get_next(seq, head)))
 841                         pos--;
 842         return pos ? NULL : head;
 843 }
 844
 845 static void *seq_start(struct seq_file *seq, loff_t *pos)
 846         __acquires(instances_lock)
 847 {
 848         spin_lock(&instances_lock);
 849         return get_idx(seq, *pos);
 850 }
 851
 852 static void *seq_next(struct seq_file *s, void *v, loff_t *pos)
 853 {
 854         (*pos)++;
 855         return get_next(s, v);
 856 }
 857
 858 static void seq_stop(struct seq_file *s, void *v)
 859         __releases(instances_lock)
 860 {
 861         spin_unlock(&instances_lock);
 862 }
 863
 864 static int seq_show(struct seq_file *s, void *v)
 865 {
 866         const struct nfqnl_instance *inst = v;
 867
 868         return seq_printf(s, "%5d %6d %5d %1d %5d %5d %5d %8d %2d\n",
 869                           inst->queue_num,
 870                           inst->peer_pid, inst->queue_total,
 871                           inst->copy_mode, inst->copy_range,
 872                           inst->queue_dropped, inst->queue_user_dropped,
 873                           atomic_read(&inst->id_sequence), 1);
 874 }
 875
 876 static const struct seq_operations nfqnl_seq_ops = {
 877         .start  = seq_start,
 878         .next   = seq_next,
 879         .stop   = seq_stop,
 880         .show   = seq_show,
 881 };
 882
 883 static int nfqnl_open(struct inode *inode, struct file *file)
 884 {
 885         return seq_open_private(file, &nfqnl_seq_ops,
 886                         sizeof(struct iter_state));
 887 }
 888
 889 static const struct file_operations nfqnl_file_ops = {
 890         .owner   = THIS_MODULE,
 891         .open    = nfqnl_open,
 892         .read    = seq_read,
 893         .llseek  = seq_lseek,
 894         .release = seq_release_private,
 895 };
 896
 897 #endif /* PROC_FS */
 898
 899 static int __init nfnetlink_queue_init(void)
 900 {
 901         int i, status = -ENOMEM;
 902
 903         for (i = 0; i < INSTANCE_BUCKETS; i++)
 904                 INIT_HLIST_HEAD(&instance_table[i]);
 905
 906         netlink_register_notifier(&nfqnl_rtnl_notifier);
 907         status = nfnetlink_subsys_register(&nfqnl_subsys);
 908         if (status < 0) {
 909                 printk(KERN_ERR "nf_queue: failed to create netlink socket\n");
 910                 goto cleanup_netlink_notifier;
 911         }
 912
 913 #ifdef CONFIG_PROC_FS
 914         if (!proc_create("nfnetlink_queue", 0440,
 915                          proc_net_netfilter, &nfqnl_file_ops))
 916                 goto cleanup_subsys;
 917 #endif
 918
 919         register_netdevice_notifier(&nfqnl_dev_notifier);
 920         return status;
 921
 922 #ifdef CONFIG_PROC_FS
 923 cleanup_subsys:
 924         nfnetlink_subsys_unregister(&nfqnl_subsys);
 925 #endif
 926 cleanup_netlink_notifier:
 927         netlink_unregister_notifier(&nfqnl_rtnl_notifier);
 928         return status;
 929 }
 930
 931 static void __exit nfnetlink_queue_fini(void)
 932 {
 933         nf_unregister_queue_handlers(&nfqh);
 934         unregister_netdevice_notifier(&nfqnl_dev_notifier);
 935 #ifdef CONFIG_PROC_FS
 936         remove_proc_entry("nfnetlink_queue", proc_net_netfilter);
 937 #endif
 938         nfnetlink_subsys_unregister(&nfqnl_subsys);
 939         netlink_unregister_notifier(&nfqnl_rtnl_notifier);
 940
 941         rcu_barrier(); /* Wait for completion of call_rcu()'s */
 942 }
 943
 944 MODULE_DESCRIPTION("netfilter packet queue handler");
 945 MODULE_AUTHOR("Harald Welte <laforge@netfilter.org>");
 946 MODULE_LICENSE("GPL");
 947 MODULE_ALIAS_NFNL_SUBSYS(NFNL_SUBSYS_QUEUE);
 948
 949 module_init(nfnetlink_queue_init);
 950 module_exit(nfnetlink_queue_fini);