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