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 <linux/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         unsigned int    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 void
 175 __dequeue_entry(struct nfqnl_instance *queue, struct nf_queue_entry *entry)
 176 {
 177         list_del(&entry->list);
 178         queue->queue_total--;
 179 }
 180
 181 static struct nf_queue_entry *
 182 find_dequeue_entry(struct nfqnl_instance *queue, unsigned int id)
 183 {
 184         struct nf_queue_entry *entry = NULL, *i;
 185
 186         spin_lock_bh(&queue->lock);
 187
 188         list_for_each_entry(i, &queue->queue_list, list) {
 189                 if (i->id == id) {
 190                         entry = i;
 191                         break;
 192                 }
 193         }
 194
 195         if (entry)
 196                 __dequeue_entry(queue, entry);
 197
 198         spin_unlock_bh(&queue->lock);
 199
 200         return entry;
 201 }
 202
 203 static void
 204 nfqnl_flush(struct nfqnl_instance *queue, nfqnl_cmpfn cmpfn, unsigned long data)
 205 {
 206         struct nf_queue_entry *entry, *next;
 207
 208         spin_lock_bh(&queue->lock);
 209         list_for_each_entry_safe(entry, next, &queue->queue_list, list) {
 210                 if (!cmpfn || cmpfn(entry, data)) {
 211                         list_del(&entry->list);
 212                         queue->queue_total--;
 213                         nf_reinject(entry, NF_DROP);
 214                 }
 215         }
 216         spin_unlock_bh(&queue->lock);
 217 }
 218
 219 static struct sk_buff *
 220 nfqnl_build_packet_message(struct nfqnl_instance *queue,
 221                            struct nf_queue_entry *entry,
 222                            __be32 **packet_id_ptr)
 223 {
 224         sk_buff_data_t old_tail;
 225         size_t size;
 226         size_t data_len = 0;
 227         struct sk_buff *skb;
 228         struct nlattr *nla;
 229         struct nfqnl_msg_packet_hdr *pmsg;
 230         struct nlmsghdr *nlh;
 231         struct nfgenmsg *nfmsg;
 232         struct sk_buff *entskb = entry->skb;
 233         struct net_device *indev;
 234         struct net_device *outdev;
 235
 236         size =    NLMSG_SPACE(sizeof(struct nfgenmsg))
 237                 + nla_total_size(sizeof(struct nfqnl_msg_packet_hdr))
 238                 + nla_total_size(sizeof(u_int32_t))     /* ifindex */
 239                 + nla_total_size(sizeof(u_int32_t))     /* ifindex */
 240 #ifdef CONFIG_BRIDGE_NETFILTER
 241                 + nla_total_size(sizeof(u_int32_t))     /* ifindex */
 242                 + nla_total_size(sizeof(u_int32_t))     /* ifindex */
 243 #endif
 244                 + nla_total_size(sizeof(u_int32_t))     /* mark */
 245                 + nla_total_size(sizeof(struct nfqnl_msg_packet_hw))
 246                 + nla_total_size(sizeof(struct nfqnl_msg_packet_timestamp));
 247
 248         outdev = entry->outdev;
 249
 250         switch ((enum nfqnl_config_mode)ACCESS_ONCE(queue->copy_mode)) {
 251         case NFQNL_COPY_META:
 252         case NFQNL_COPY_NONE:
 253                 break;
 254
 255         case NFQNL_COPY_PACKET:
 256                 if (entskb->ip_summed == CHECKSUM_PARTIAL &&
 257                     skb_checksum_help(entskb))
 258                         return NULL;
 259
 260                 data_len = ACCESS_ONCE(queue->copy_range);
 261                 if (data_len == 0 || data_len > entskb->len)
 262                         data_len = entskb->len;
 263
 264                 size += nla_total_size(data_len);
 265                 break;
 266         }
 267
 268
 269         skb = alloc_skb(size, GFP_ATOMIC);
 270         if (!skb)
 271                 goto nlmsg_failure;
 272
 273         old_tail = skb->tail;
 274         nlh = NLMSG_PUT(skb, 0, 0,
 275                         NFNL_SUBSYS_QUEUE << 8 | NFQNL_MSG_PACKET,
 276                         sizeof(struct nfgenmsg));
 277         nfmsg = NLMSG_DATA(nlh);
 278         nfmsg->nfgen_family = entry->pf;
 279         nfmsg->version = NFNETLINK_V0;
 280         nfmsg->res_id = htons(queue->queue_num);
 281
 282         nla = __nla_reserve(skb, NFQA_PACKET_HDR, sizeof(*pmsg));
 283         pmsg = nla_data(nla);
 284         pmsg->hw_protocol       = entskb->protocol;
 285         pmsg->hook              = entry->hook;
 286         *packet_id_ptr          = &pmsg->packet_id;
 287
 288         indev = entry->indev;
 289         if (indev) {
 290 #ifndef CONFIG_BRIDGE_NETFILTER
 291                 if (nla_put_be32(skb, NFQA_IFINDEX_INDEV, htonl(indev->ifindex)))
 292                         goto nla_put_failure;
 293 #else
 294                 if (entry->pf == PF_BRIDGE) {
 295                         /* Case 1: indev is physical input device, we need to
 296                          * look for bridge group (when called from
 297                          * netfilter_bridge) */
 298                         if (nla_put_be32(skb, NFQA_IFINDEX_PHYSINDEV,
 299                                          htonl(indev->ifindex)) ||
 300                         /* this is the bridge group "brX" */
 301                         /* rcu_read_lock()ed by __nf_queue */
 302                             nla_put_be32(skb, NFQA_IFINDEX_INDEV,
 303                                          htonl(br_port_get_rcu(indev)->br->dev->ifindex)))
 304                                 goto nla_put_failure;
 305                 } else {
 306                         /* Case 2: indev is bridge group, we need to look for
 307                          * physical device (when called from ipv4) */
 308                         if (nla_put_be32(skb, NFQA_IFINDEX_INDEV,
 309                                          htonl(indev->ifindex)))
 310                                 goto nla_put_failure;
 311                         if (entskb->nf_bridge && entskb->nf_bridge->physindev &&
 312                             nla_put_be32(skb, NFQA_IFINDEX_PHYSINDEV,
 313                                          htonl(entskb->nf_bridge->physindev->ifindex)))
 314                                 goto nla_put_failure;
 315                 }
 316 #endif
 317         }
 318
 319         if (outdev) {
 320 #ifndef CONFIG_BRIDGE_NETFILTER
 321                 if (nla_put_be32(skb, NFQA_IFINDEX_OUTDEV, htonl(outdev->ifindex)))
 322                         goto nla_put_failure;
 323 #else
 324                 if (entry->pf == PF_BRIDGE) {
 325                         /* Case 1: outdev is physical output device, we need to
 326                          * look for bridge group (when called from
 327                          * netfilter_bridge) */
 328                         if (nla_put_be32(skb, NFQA_IFINDEX_PHYSOUTDEV,
 329                                          htonl(outdev->ifindex)) ||
 330                         /* this is the bridge group "brX" */
 331                         /* rcu_read_lock()ed by __nf_queue */
 332                             nla_put_be32(skb, NFQA_IFINDEX_OUTDEV,
 333                                          htonl(br_port_get_rcu(outdev)->br->dev->ifindex)))
 334                                 goto nla_put_failure;
 335                 } else {
 336                         /* Case 2: outdev is bridge group, we need to look for
 337                          * physical output device (when called from ipv4) */
 338                         if (nla_put_be32(skb, NFQA_IFINDEX_OUTDEV,
 339                                          htonl(outdev->ifindex)))
 340                                 goto nla_put_failure;
 341                         if (entskb->nf_bridge && entskb->nf_bridge->physoutdev &&
 342                             nla_put_be32(skb, NFQA_IFINDEX_PHYSOUTDEV,
 343                                          htonl(entskb->nf_bridge->physoutdev->ifindex)))
 344                                 goto nla_put_failure;
 345                 }
 346 #endif
 347         }
 348
 349         if (entskb->mark &&
 350             nla_put_be32(skb, NFQA_MARK, htonl(entskb->mark)))
 351                 goto nla_put_failure;
 352
 353         if (indev && entskb->dev &&
 354             entskb->mac_header != entskb->network_header) {
 355                 struct nfqnl_msg_packet_hw phw;
 356                 int len = dev_parse_header(entskb, phw.hw_addr);
 357                 if (len) {
 358                         phw.hw_addrlen = htons(len);
 359                         if (nla_put(skb, NFQA_HWADDR, sizeof(phw), &phw))
 360                                 goto nla_put_failure;
 361                 }
 362         }
 363
 364         if (entskb->tstamp.tv64) {
 365                 struct nfqnl_msg_packet_timestamp ts;
 366                 struct timeval tv = ktime_to_timeval(entskb->tstamp);
 367                 ts.sec = cpu_to_be64(tv.tv_sec);
 368                 ts.usec = cpu_to_be64(tv.tv_usec);
 369
 370                 if (nla_put(skb, NFQA_TIMESTAMP, sizeof(ts), &ts))
 371                         goto nla_put_failure;
 372         }
 373
 374         if (data_len) {
 375                 struct nlattr *nla;
 376                 int sz = nla_attr_size(data_len);
 377
 378                 if (skb_tailroom(skb) < nla_total_size(data_len)) {
 379                         printk(KERN_WARNING "nf_queue: no tailroom!\n");
 380                         goto nlmsg_failure;
 381                 }
 382
 383                 nla = (struct nlattr *)skb_put(skb, nla_total_size(data_len));
 384                 nla->nla_type = NFQA_PAYLOAD;
 385                 nla->nla_len = sz;
 386
 387                 if (skb_copy_bits(entskb, 0, nla_data(nla), data_len))
 388                         BUG();
 389         }
 390
 391         nlh->nlmsg_len = skb->tail - old_tail;
 392         return skb;
 393
 394 nlmsg_failure:
 395 nla_put_failure:
 396         if (skb)
 397                 kfree_skb(skb);
 398         net_err_ratelimited("nf_queue: error creating packet message\n");
 399         return NULL;
 400 }
 401
 402 static int
 403 nfqnl_enqueue_packet(struct nf_queue_entry *entry, unsigned int queuenum)
 404 {
 405         struct sk_buff *nskb;
 406         struct nfqnl_instance *queue;
 407         int err = -ENOBUFS;
 408         __be32 *packet_id_ptr;
 409
 410         /* rcu_read_lock()ed by nf_hook_slow() */
 411         queue = instance_lookup(queuenum);
 412         if (!queue) {
 413                 err = -ESRCH;
 414                 goto err_out;
 415         }
 416
 417         if (queue->copy_mode == NFQNL_COPY_NONE) {
 418                 err = -EINVAL;
 419                 goto err_out;
 420         }
 421
 422         nskb = nfqnl_build_packet_message(queue, entry, &packet_id_ptr);
 423         if (nskb == NULL) {
 424                 err = -ENOMEM;
 425                 goto err_out;
 426         }
 427         spin_lock_bh(&queue->lock);
 428
 429         if (!queue->peer_pid) {
 430                 err = -EINVAL;
 431                 goto err_out_free_nskb;
 432         }
 433         if (queue->queue_total >= queue->queue_maxlen) {
 434                 queue->queue_dropped++;
 435                 net_warn_ratelimited("nf_queue: full at %d entries, dropping packets(s)\n",
 436                                      queue->queue_total);
 437                 goto err_out_free_nskb;
 438         }
 439         entry->id = ++queue->id_sequence;
 440         *packet_id_ptr = htonl(entry->id);
 441
 442         /* nfnetlink_unicast will either free the nskb or add it to a socket */
 443         err = nfnetlink_unicast(nskb, &init_net, queue->peer_pid, MSG_DONTWAIT);
 444         if (err < 0) {
 445                 queue->queue_user_dropped++;
 446                 goto err_out_unlock;
 447         }
 448
 449         __enqueue_entry(queue, entry);
 450
 451         spin_unlock_bh(&queue->lock);
 452         return 0;
 453
 454 err_out_free_nskb:
 455         kfree_skb(nskb);
 456 err_out_unlock:
 457         spin_unlock_bh(&queue->lock);
 458 err_out:
 459         return err;
 460 }
 461
 462 static int
 463 nfqnl_mangle(void *data, int data_len, struct nf_queue_entry *e)
 464 {
 465         struct sk_buff *nskb;
 466         int diff;
 467
 468         diff = data_len - e->skb->len;
 469         if (diff < 0) {
 470                 if (pskb_trim(e->skb, data_len))
 471                         return -ENOMEM;
 472         } else if (diff > 0) {
 473                 if (data_len > 0xFFFF)
 474                         return -EINVAL;
 475                 if (diff > skb_tailroom(e->skb)) {
 476                         nskb = skb_copy_expand(e->skb, skb_headroom(e->skb),
 477                                                diff, GFP_ATOMIC);
 478                         if (!nskb) {
 479                                 printk(KERN_WARNING "nf_queue: OOM "
 480                                       "in mangle, dropping packet\n");
 481                                 return -ENOMEM;
 482                         }
 483                         kfree_skb(e->skb);
 484                         e->skb = nskb;
 485                 }
 486                 skb_put(e->skb, diff);
 487         }
 488         if (!skb_make_writable(e->skb, data_len))
 489                 return -ENOMEM;
 490         skb_copy_to_linear_data(e->skb, data, data_len);
 491         e->skb->ip_summed = CHECKSUM_NONE;
 492         return 0;
 493 }
 494
 495 static int
 496 nfqnl_set_mode(struct nfqnl_instance *queue,
 497                unsigned char mode, unsigned int range)
 498 {
 499         int status = 0;
 500
 501         spin_lock_bh(&queue->lock);
 502         switch (mode) {
 503         case NFQNL_COPY_NONE:
 504         case NFQNL_COPY_META:
 505                 queue->copy_mode = mode;
 506                 queue->copy_range = 0;
 507                 break;
 508
 509         case NFQNL_COPY_PACKET:
 510                 queue->copy_mode = mode;
 511                 /* we're using struct nlattr which has 16bit nla_len */
 512                 if (range > 0xffff)
 513                         queue->copy_range = 0xffff;
 514                 else
 515                         queue->copy_range = range;
 516                 break;
 517
 518         default:
 519                 status = -EINVAL;
 520
 521         }
 522         spin_unlock_bh(&queue->lock);
 523
 524         return status;
 525 }
 526
 527 static int
 528 dev_cmp(struct nf_queue_entry *entry, unsigned long ifindex)
 529 {
 530         if (entry->indev)
 531                 if (entry->indev->ifindex == ifindex)
 532                         return 1;
 533         if (entry->outdev)
 534                 if (entry->outdev->ifindex == ifindex)
 535                         return 1;
 536 #ifdef CONFIG_BRIDGE_NETFILTER
 537         if (entry->skb->nf_bridge) {
 538                 if (entry->skb->nf_bridge->physindev &&
 539                     entry->skb->nf_bridge->physindev->ifindex == ifindex)
 540                         return 1;
 541                 if (entry->skb->nf_bridge->physoutdev &&
 542                     entry->skb->nf_bridge->physoutdev->ifindex == ifindex)
 543                         return 1;
 544         }
 545 #endif
 546         return 0;
 547 }
 548
 549 /* drop all packets with either indev or outdev == ifindex from all queue
 550  * instances */
 551 static void
 552 nfqnl_dev_drop(int ifindex)
 553 {
 554         int i;
 555
 556         rcu_read_lock();
 557
 558         for (i = 0; i < INSTANCE_BUCKETS; i++) {
 559                 struct hlist_node *tmp;
 560                 struct nfqnl_instance *inst;
 561                 struct hlist_head *head = &instance_table[i];
 562
 563                 hlist_for_each_entry_rcu(inst, tmp, head, hlist)
 564                         nfqnl_flush(inst, dev_cmp, ifindex);
 565         }
 566
 567         rcu_read_unlock();
 568 }
 569
 570 #define RCV_SKB_FAIL(err) do { netlink_ack(skb, nlh, (err)); return; } while (0)
 571
 572 static int
 573 nfqnl_rcv_dev_event(struct notifier_block *this,
 574                     unsigned long event, void *ptr)
 575 {
 576         struct net_device *dev = ptr;
 577
 578         if (!net_eq(dev_net(dev), &init_net))
 579                 return NOTIFY_DONE;
 580
 581         /* Drop any packets associated with the downed device */
 582         if (event == NETDEV_DOWN)
 583                 nfqnl_dev_drop(dev->ifindex);
 584         return NOTIFY_DONE;
 585 }
 586
 587 static struct notifier_block nfqnl_dev_notifier = {
 588         .notifier_call  = nfqnl_rcv_dev_event,
 589 };
 590
 591 static int
 592 nfqnl_rcv_nl_event(struct notifier_block *this,
 593                    unsigned long event, void *ptr)
 594 {
 595         struct netlink_notify *n = ptr;
 596
 597         if (event == NETLINK_URELEASE && n->protocol == NETLINK_NETFILTER) {
 598                 int i;
 599
 600                 /* destroy all instances for this pid */
 601                 spin_lock(&instances_lock);
 602                 for (i = 0; i < INSTANCE_BUCKETS; i++) {
 603                         struct hlist_node *tmp, *t2;
 604                         struct nfqnl_instance *inst;
 605                         struct hlist_head *head = &instance_table[i];
 606
 607                         hlist_for_each_entry_safe(inst, tmp, t2, head, hlist) {
 608                                 if ((n->net == &init_net) &&
 609                                     (n->pid == inst->peer_pid))
 610                                         __instance_destroy(inst);
 611                         }
 612                 }
 613                 spin_unlock(&instances_lock);
 614         }
 615         return NOTIFY_DONE;
 616 }
 617
 618 static struct notifier_block nfqnl_rtnl_notifier = {
 619         .notifier_call  = nfqnl_rcv_nl_event,
 620 };
 621
 622 static const struct nla_policy nfqa_verdict_policy[NFQA_MAX+1] = {
 623         [NFQA_VERDICT_HDR]      = { .len = sizeof(struct nfqnl_msg_verdict_hdr) },
 624         [NFQA_MARK]             = { .type = NLA_U32 },
 625         [NFQA_PAYLOAD]          = { .type = NLA_UNSPEC },
 626 };
 627
 628 static const struct nla_policy nfqa_verdict_batch_policy[NFQA_MAX+1] = {
 629         [NFQA_VERDICT_HDR]      = { .len = sizeof(struct nfqnl_msg_verdict_hdr) },
 630         [NFQA_MARK]             = { .type = NLA_U32 },
 631 };
 632
 633 static struct nfqnl_instance *verdict_instance_lookup(u16 queue_num, int nlpid)
 634 {
 635         struct nfqnl_instance *queue;
 636
 637         queue = instance_lookup(queue_num);
 638         if (!queue)
 639                 return ERR_PTR(-ENODEV);
 640
 641         if (queue->peer_pid != nlpid)
 642                 return ERR_PTR(-EPERM);
 643
 644         return queue;
 645 }
 646
 647 static struct nfqnl_msg_verdict_hdr*
 648 verdicthdr_get(const struct nlattr * const nfqa[])
 649 {
 650         struct nfqnl_msg_verdict_hdr *vhdr;
 651         unsigned int verdict;
 652
 653         if (!nfqa[NFQA_VERDICT_HDR])
 654                 return NULL;
 655
 656         vhdr = nla_data(nfqa[NFQA_VERDICT_HDR]);
 657         verdict = ntohl(vhdr->verdict) & NF_VERDICT_MASK;
 658         if (verdict > NF_MAX_VERDICT || verdict == NF_STOLEN)
 659                 return NULL;
 660         return vhdr;
 661 }
 662
 663 static int nfq_id_after(unsigned int id, unsigned int max)
 664 {
 665         return (int)(id - max) > 0;
 666 }
 667
 668 static int
 669 nfqnl_recv_verdict_batch(struct sock *ctnl, struct sk_buff *skb,
 670                    const struct nlmsghdr *nlh,
 671                    const struct nlattr * const nfqa[])
 672 {
 673         struct nfgenmsg *nfmsg = NLMSG_DATA(nlh);
 674         struct nf_queue_entry *entry, *tmp;
 675         unsigned int verdict, maxid;
 676         struct nfqnl_msg_verdict_hdr *vhdr;
 677         struct nfqnl_instance *queue;
 678         LIST_HEAD(batch_list);
 679         u16 queue_num = ntohs(nfmsg->res_id);
 680
 681         queue = verdict_instance_lookup(queue_num, NETLINK_CB(skb).pid);
 682         if (IS_ERR(queue))
 683                 return PTR_ERR(queue);
 684
 685         vhdr = verdicthdr_get(nfqa);
 686         if (!vhdr)
 687                 return -EINVAL;
 688
 689         verdict = ntohl(vhdr->verdict);
 690         maxid = ntohl(vhdr->id);
 691
 692         spin_lock_bh(&queue->lock);
 693
 694         list_for_each_entry_safe(entry, tmp, &queue->queue_list, list) {
 695                 if (nfq_id_after(entry->id, maxid))
 696                         break;
 697                 __dequeue_entry(queue, entry);
 698                 list_add_tail(&entry->list, &batch_list);
 699         }
 700
 701         spin_unlock_bh(&queue->lock);
 702
 703         if (list_empty(&batch_list))
 704                 return -ENOENT;
 705
 706         list_for_each_entry_safe(entry, tmp, &batch_list, list) {
 707                 if (nfqa[NFQA_MARK])
 708                         entry->skb->mark = ntohl(nla_get_be32(nfqa[NFQA_MARK]));
 709                 nf_reinject(entry, verdict);
 710         }
 711         return 0;
 712 }
 713
 714 static int
 715 nfqnl_recv_verdict(struct sock *ctnl, struct sk_buff *skb,
 716                    const struct nlmsghdr *nlh,
 717                    const struct nlattr * const nfqa[])
 718 {
 719         struct nfgenmsg *nfmsg = NLMSG_DATA(nlh);
 720         u_int16_t queue_num = ntohs(nfmsg->res_id);
 721
 722         struct nfqnl_msg_verdict_hdr *vhdr;
 723         struct nfqnl_instance *queue;
 724         unsigned int verdict;
 725         struct nf_queue_entry *entry;
 726
 727         queue = instance_lookup(queue_num);
 728         if (!queue)
 729
 730         queue = verdict_instance_lookup(queue_num, NETLINK_CB(skb).pid);
 731         if (IS_ERR(queue))
 732                 return PTR_ERR(queue);
 733
 734         vhdr = verdicthdr_get(nfqa);
 735         if (!vhdr)
 736                 return -EINVAL;
 737
 738         verdict = ntohl(vhdr->verdict);
 739
 740         entry = find_dequeue_entry(queue, ntohl(vhdr->id));
 741         if (entry == NULL)
 742                 return -ENOENT;
 743
 744         if (nfqa[NFQA_PAYLOAD]) {
 745                 if (nfqnl_mangle(nla_data(nfqa[NFQA_PAYLOAD]),
 746                                  nla_len(nfqa[NFQA_PAYLOAD]), entry) < 0)
 747                         verdict = NF_DROP;
 748         }
 749
 750         if (nfqa[NFQA_MARK])
 751                 entry->skb->mark = ntohl(nla_get_be32(nfqa[NFQA_MARK]));
 752
 753         nf_reinject(entry, verdict);
 754         return 0;
 755 }
 756
 757 static int
 758 nfqnl_recv_unsupp(struct sock *ctnl, struct sk_buff *skb,
 759                   const struct nlmsghdr *nlh,
 760                   const struct nlattr * const nfqa[])
 761 {
 762         return -ENOTSUPP;
 763 }
 764
 765 static const struct nla_policy nfqa_cfg_policy[NFQA_CFG_MAX+1] = {
 766         [NFQA_CFG_CMD]          = { .len = sizeof(struct nfqnl_msg_config_cmd) },
 767         [NFQA_CFG_PARAMS]       = { .len = sizeof(struct nfqnl_msg_config_params) },
 768 };
 769
 770 static const struct nf_queue_handler nfqh = {
 771         .name   = "nf_queue",
 772         .outfn  = &nfqnl_enqueue_packet,
 773 };
 774
 775 static int
 776 nfqnl_recv_config(struct sock *ctnl, struct sk_buff *skb,
 777                   const struct nlmsghdr *nlh,
 778                   const struct nlattr * const nfqa[])
 779 {
 780         struct nfgenmsg *nfmsg = NLMSG_DATA(nlh);
 781         u_int16_t queue_num = ntohs(nfmsg->res_id);
 782         struct nfqnl_instance *queue;
 783         struct nfqnl_msg_config_cmd *cmd = NULL;
 784         int ret = 0;
 785
 786         if (nfqa[NFQA_CFG_CMD]) {
 787                 cmd = nla_data(nfqa[NFQA_CFG_CMD]);
 788
 789                 /* Commands without queue context - might sleep */
 790                 switch (cmd->command) {
 791                 case NFQNL_CFG_CMD_PF_BIND:
 792                         return nf_register_queue_handler(ntohs(cmd->pf),
 793                                                          &nfqh);
 794                 case NFQNL_CFG_CMD_PF_UNBIND:
 795                         return nf_unregister_queue_handler(ntohs(cmd->pf),
 796                                                            &nfqh);
 797                 }
 798         }
 799
 800         rcu_read_lock();
 801         queue = instance_lookup(queue_num);
 802         if (queue && queue->peer_pid != NETLINK_CB(skb).pid) {
 803                 ret = -EPERM;
 804                 goto err_out_unlock;
 805         }
 806
 807         if (cmd != NULL) {
 808                 switch (cmd->command) {
 809                 case NFQNL_CFG_CMD_BIND:
 810                         if (queue) {
 811                                 ret = -EBUSY;
 812                                 goto err_out_unlock;
 813                         }
 814                         queue = instance_create(queue_num, NETLINK_CB(skb).pid);
 815                         if (IS_ERR(queue)) {
 816                                 ret = PTR_ERR(queue);
 817                                 goto err_out_unlock;
 818                         }
 819                         break;
 820                 case NFQNL_CFG_CMD_UNBIND:
 821                         if (!queue) {
 822                                 ret = -ENODEV;
 823                                 goto err_out_unlock;
 824                         }
 825                         instance_destroy(queue);
 826                         break;
 827                 case NFQNL_CFG_CMD_PF_BIND:
 828                 case NFQNL_CFG_CMD_PF_UNBIND:
 829                         break;
 830                 default:
 831                         ret = -ENOTSUPP;
 832                         break;
 833                 }
 834         }
 835
 836         if (nfqa[NFQA_CFG_PARAMS]) {
 837                 struct nfqnl_msg_config_params *params;
 838
 839                 if (!queue) {
 840                         ret = -ENODEV;
 841                         goto err_out_unlock;
 842                 }
 843                 params = nla_data(nfqa[NFQA_CFG_PARAMS]);
 844                 nfqnl_set_mode(queue, params->copy_mode,
 845                                 ntohl(params->copy_range));
 846         }
 847
 848         if (nfqa[NFQA_CFG_QUEUE_MAXLEN]) {
 849                 __be32 *queue_maxlen;
 850
 851                 if (!queue) {
 852                         ret = -ENODEV;
 853                         goto err_out_unlock;
 854                 }
 855                 queue_maxlen = nla_data(nfqa[NFQA_CFG_QUEUE_MAXLEN]);
 856                 spin_lock_bh(&queue->lock);
 857                 queue->queue_maxlen = ntohl(*queue_maxlen);
 858                 spin_unlock_bh(&queue->lock);
 859         }
 860
 861 err_out_unlock:
 862         rcu_read_unlock();
 863         return ret;
 864 }
 865
 866 static const struct nfnl_callback nfqnl_cb[NFQNL_MSG_MAX] = {
 867         [NFQNL_MSG_PACKET]      = { .call_rcu = nfqnl_recv_unsupp,
 868                                     .attr_count = NFQA_MAX, },
 869         [NFQNL_MSG_VERDICT]     = { .call_rcu = nfqnl_recv_verdict,
 870                                     .attr_count = NFQA_MAX,
 871                                     .policy = nfqa_verdict_policy },
 872         [NFQNL_MSG_CONFIG]      = { .call = nfqnl_recv_config,
 873                                     .attr_count = NFQA_CFG_MAX,
 874                                     .policy = nfqa_cfg_policy },
 875         [NFQNL_MSG_VERDICT_BATCH]={ .call_rcu = nfqnl_recv_verdict_batch,
 876                                     .attr_count = NFQA_MAX,
 877                                     .policy = nfqa_verdict_batch_policy },
 878 };
 879
 880 static const struct nfnetlink_subsystem nfqnl_subsys = {
 881         .name           = "nf_queue",
 882         .subsys_id      = NFNL_SUBSYS_QUEUE,
 883         .cb_count       = NFQNL_MSG_MAX,
 884         .cb             = nfqnl_cb,
 885 };
 886
 887 #ifdef CONFIG_PROC_FS
 888 struct iter_state {
 889         unsigned int bucket;
 890 };
 891
 892 static struct hlist_node *get_first(struct seq_file *seq)
 893 {
 894         struct iter_state *st = seq->private;
 895
 896         if (!st)
 897                 return NULL;
 898
 899         for (st->bucket = 0; st->bucket < INSTANCE_BUCKETS; st->bucket++) {
 900                 if (!hlist_empty(&instance_table[st->bucket]))
 901                         return instance_table[st->bucket].first;
 902         }
 903         return NULL;
 904 }
 905
 906 static struct hlist_node *get_next(struct seq_file *seq, struct hlist_node *h)
 907 {
 908         struct iter_state *st = seq->private;
 909
 910         h = h->next;
 911         while (!h) {
 912                 if (++st->bucket >= INSTANCE_BUCKETS)
 913                         return NULL;
 914
 915                 h = instance_table[st->bucket].first;
 916         }
 917         return h;
 918 }
 919
 920 static struct hlist_node *get_idx(struct seq_file *seq, loff_t pos)
 921 {
 922         struct hlist_node *head;
 923         head = get_first(seq);
 924
 925         if (head)
 926                 while (pos && (head = get_next(seq, head)))
 927                         pos--;
 928         return pos ? NULL : head;
 929 }
 930
 931 static void *seq_start(struct seq_file *seq, loff_t *pos)
 932         __acquires(instances_lock)
 933 {
 934         spin_lock(&instances_lock);
 935         return get_idx(seq, *pos);
 936 }
 937
 938 static void *seq_next(struct seq_file *s, void *v, loff_t *pos)
 939 {
 940         (*pos)++;
 941         return get_next(s, v);
 942 }
 943
 944 static void seq_stop(struct seq_file *s, void *v)
 945         __releases(instances_lock)
 946 {
 947         spin_unlock(&instances_lock);
 948 }
 949
 950 static int seq_show(struct seq_file *s, void *v)
 951 {
 952         const struct nfqnl_instance *inst = v;
 953
 954         return seq_printf(s, "%5d %6d %5d %1d %5d %5d %5d %8d %2d\n",
 955                           inst->queue_num,
 956                           inst->peer_pid, inst->queue_total,
 957                           inst->copy_mode, inst->copy_range,
 958                           inst->queue_dropped, inst->queue_user_dropped,
 959                           inst->id_sequence, 1);
 960 }
 961
 962 static const struct seq_operations nfqnl_seq_ops = {
 963         .start  = seq_start,
 964         .next   = seq_next,
 965         .stop   = seq_stop,
 966         .show   = seq_show,
 967 };
 968
 969 static int nfqnl_open(struct inode *inode, struct file *file)
 970 {
 971         return seq_open_private(file, &nfqnl_seq_ops,
 972                         sizeof(struct iter_state));
 973 }
 974
 975 static const struct file_operations nfqnl_file_ops = {
 976         .owner   = THIS_MODULE,
 977         .open    = nfqnl_open,
 978         .read    = seq_read,
 979         .llseek  = seq_lseek,
 980         .release = seq_release_private,
 981 };
 982
 983 #endif /* PROC_FS */
 984
 985 static int __init nfnetlink_queue_init(void)
 986 {
 987         int i, status = -ENOMEM;
 988
 989         for (i = 0; i < INSTANCE_BUCKETS; i++)
 990                 INIT_HLIST_HEAD(&instance_table[i]);
 991
 992         netlink_register_notifier(&nfqnl_rtnl_notifier);
 993         status = nfnetlink_subsys_register(&nfqnl_subsys);
 994         if (status < 0) {
 995                 printk(KERN_ERR "nf_queue: failed to create netlink socket\n");
 996                 goto cleanup_netlink_notifier;
 997         }
 998
 999 #ifdef CONFIG_PROC_FS
1000         if (!proc_create("nfnetlink_queue", 0440,
1001                          proc_net_netfilter, &nfqnl_file_ops))
1002                 goto cleanup_subsys;
1003 #endif
1004
1005         register_netdevice_notifier(&nfqnl_dev_notifier);
1006         return status;
1007
1008 #ifdef CONFIG_PROC_FS
1009 cleanup_subsys:
1010         nfnetlink_subsys_unregister(&nfqnl_subsys);
1011 #endif
1012 cleanup_netlink_notifier:
1013         netlink_unregister_notifier(&nfqnl_rtnl_notifier);
1014         return status;
1015 }
1016
1017 static void __exit nfnetlink_queue_fini(void)
1018 {
1019         nf_unregister_queue_handlers(&nfqh);
1020         unregister_netdevice_notifier(&nfqnl_dev_notifier);
1021 #ifdef CONFIG_PROC_FS
1022         remove_proc_entry("nfnetlink_queue", proc_net_netfilter);
1023 #endif
1024         nfnetlink_subsys_unregister(&nfqnl_subsys);
1025         netlink_unregister_notifier(&nfqnl_rtnl_notifier);
1026
1027         rcu_barrier(); /* Wait for completion of call_rcu()'s */
1028 }
1029
1030 MODULE_DESCRIPTION("netfilter packet queue handler");
1031 MODULE_AUTHOR("Harald Welte <laforge@netfilter.org>");
1032 MODULE_LICENSE("GPL");
1033 MODULE_ALIAS_NFNL_SUBSYS(NFNL_SUBSYS_QUEUE);
1034
1035 module_init(nfnetlink_queue_init);
1036 module_exit(nfnetlink_queue_fini);