Pileus Git - ~andy/linux/blob - 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, cap_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                 + nla_total_size(sizeof(u_int32_t)));   /* cap_len */
 252
 253         outdev = entry->outdev;
 254
 255         switch ((enum nfqnl_config_mode)ACCESS_ONCE(queue->copy_mode)) {
 256         case NFQNL_COPY_META:
 257         case NFQNL_COPY_NONE:
 258                 break;
 259
 260         case NFQNL_COPY_PACKET:
 261                 if (entskb->ip_summed == CHECKSUM_PARTIAL &&
 262                     skb_checksum_help(entskb))
 263                         return NULL;
 264
 265                 data_len = ACCESS_ONCE(queue->copy_range);
 266                 if (data_len == 0 || data_len > entskb->len)
 267                         data_len = entskb->len;
 268
 269                 size += nla_total_size(data_len);
 270                 cap_len = entskb->len;
 271                 break;
 272         }
 273
 274         if (queue->flags & NFQA_CFG_F_CONNTRACK)
 275                 ct = nfqnl_ct_get(entskb, &size, &ctinfo);
 276
 277         skb = alloc_skb(size, GFP_ATOMIC);
 278         if (!skb)
 279                 return NULL;
 280
 281         old_tail = skb->tail;
 282         nlh = nlmsg_put(skb, 0, 0,
 283                         NFNL_SUBSYS_QUEUE << 8 | NFQNL_MSG_PACKET,
 284                         sizeof(struct nfgenmsg), 0);
 285         if (!nlh) {
 286                 kfree_skb(skb);
 287                 return NULL;
 288         }
 289         nfmsg = nlmsg_data(nlh);
 290         nfmsg->nfgen_family = entry->pf;
 291         nfmsg->version = NFNETLINK_V0;
 292         nfmsg->res_id = htons(queue->queue_num);
 293
 294         nla = __nla_reserve(skb, NFQA_PACKET_HDR, sizeof(*pmsg));
 295         pmsg = nla_data(nla);
 296         pmsg->hw_protocol       = entskb->protocol;
 297         pmsg->hook              = entry->hook;
 298         *packet_id_ptr          = &pmsg->packet_id;
 299
 300         indev = entry->indev;
 301         if (indev) {
 302 #ifndef CONFIG_BRIDGE_NETFILTER
 303                 if (nla_put_be32(skb, NFQA_IFINDEX_INDEV, htonl(indev->ifindex)))
 304                         goto nla_put_failure;
 305 #else
 306                 if (entry->pf == PF_BRIDGE) {
 307                         /* Case 1: indev is physical input device, we need to
 308                          * look for bridge group (when called from
 309                          * netfilter_bridge) */
 310                         if (nla_put_be32(skb, NFQA_IFINDEX_PHYSINDEV,
 311                                          htonl(indev->ifindex)) ||
 312                         /* this is the bridge group "brX" */
 313                         /* rcu_read_lock()ed by __nf_queue */
 314                             nla_put_be32(skb, NFQA_IFINDEX_INDEV,
 315                                          htonl(br_port_get_rcu(indev)->br->dev->ifindex)))
 316                                 goto nla_put_failure;
 317                 } else {
 318                         /* Case 2: indev is bridge group, we need to look for
 319                          * physical device (when called from ipv4) */
 320                         if (nla_put_be32(skb, NFQA_IFINDEX_INDEV,
 321                                          htonl(indev->ifindex)))
 322                                 goto nla_put_failure;
 323                         if (entskb->nf_bridge && entskb->nf_bridge->physindev &&
 324                             nla_put_be32(skb, NFQA_IFINDEX_PHYSINDEV,
 325                                          htonl(entskb->nf_bridge->physindev->ifindex)))
 326                                 goto nla_put_failure;
 327                 }
 328 #endif
 329         }
 330
 331         if (outdev) {
 332 #ifndef CONFIG_BRIDGE_NETFILTER
 333                 if (nla_put_be32(skb, NFQA_IFINDEX_OUTDEV, htonl(outdev->ifindex)))
 334                         goto nla_put_failure;
 335 #else
 336                 if (entry->pf == PF_BRIDGE) {
 337                         /* Case 1: outdev is physical output device, we need to
 338                          * look for bridge group (when called from
 339                          * netfilter_bridge) */
 340                         if (nla_put_be32(skb, NFQA_IFINDEX_PHYSOUTDEV,
 341                                          htonl(outdev->ifindex)) ||
 342                         /* this is the bridge group "brX" */
 343                         /* rcu_read_lock()ed by __nf_queue */
 344                             nla_put_be32(skb, NFQA_IFINDEX_OUTDEV,
 345                                          htonl(br_port_get_rcu(outdev)->br->dev->ifindex)))
 346                                 goto nla_put_failure;
 347                 } else {
 348                         /* Case 2: outdev is bridge group, we need to look for
 349                          * physical output device (when called from ipv4) */
 350                         if (nla_put_be32(skb, NFQA_IFINDEX_OUTDEV,
 351                                          htonl(outdev->ifindex)))
 352                                 goto nla_put_failure;
 353                         if (entskb->nf_bridge && entskb->nf_bridge->physoutdev &&
 354                             nla_put_be32(skb, NFQA_IFINDEX_PHYSOUTDEV,
 355                                          htonl(entskb->nf_bridge->physoutdev->ifindex)))
 356                                 goto nla_put_failure;
 357                 }
 358 #endif
 359         }
 360
 361         if (entskb->mark &&
 362             nla_put_be32(skb, NFQA_MARK, htonl(entskb->mark)))
 363                 goto nla_put_failure;
 364
 365         if (indev && entskb->dev &&
 366             entskb->mac_header != entskb->network_header) {
 367                 struct nfqnl_msg_packet_hw phw;
 368                 int len = dev_parse_header(entskb, phw.hw_addr);
 369                 if (len) {
 370                         phw.hw_addrlen = htons(len);
 371                         if (nla_put(skb, NFQA_HWADDR, sizeof(phw), &phw))
 372                                 goto nla_put_failure;
 373                 }
 374         }
 375
 376         if (entskb->tstamp.tv64) {
 377                 struct nfqnl_msg_packet_timestamp ts;
 378                 struct timeval tv = ktime_to_timeval(entskb->tstamp);
 379                 ts.sec = cpu_to_be64(tv.tv_sec);
 380                 ts.usec = cpu_to_be64(tv.tv_usec);
 381
 382                 if (nla_put(skb, NFQA_TIMESTAMP, sizeof(ts), &ts))
 383                         goto nla_put_failure;
 384         }
 385
 386         if (data_len) {
 387                 struct nlattr *nla;
 388                 int sz = nla_attr_size(data_len);
 389
 390                 if (skb_tailroom(skb) < nla_total_size(data_len)) {
 391                         printk(KERN_WARNING "nf_queue: no tailroom!\n");
 392                         kfree_skb(skb);
 393                         return NULL;
 394                 }
 395
 396                 nla = (struct nlattr *)skb_put(skb, nla_total_size(data_len));
 397                 nla->nla_type = NFQA_PAYLOAD;
 398                 nla->nla_len = sz;
 399
 400                 if (skb_copy_bits(entskb, 0, nla_data(nla), data_len))
 401                         BUG();
 402         }
 403
 404         if (ct && nfqnl_ct_put(skb, ct, ctinfo) < 0)
 405                 goto nla_put_failure;
 406
 407         if (cap_len > 0 && nla_put_be32(skb, NFQA_CAP_LEN, htonl(cap_len)))
 408                 goto nla_put_failure;
 409
 410         nlh->nlmsg_len = skb->tail - old_tail;
 411         return skb;
 412
 413 nla_put_failure:
 414         kfree_skb(skb);
 415         net_err_ratelimited("nf_queue: error creating packet message\n");
 416         return NULL;
 417 }
 418
 419 static int
 420 nfqnl_enqueue_packet(struct nf_queue_entry *entry, unsigned int queuenum)
 421 {
 422         struct sk_buff *nskb;
 423         struct nfqnl_instance *queue;
 424         int err = -ENOBUFS;
 425         __be32 *packet_id_ptr;
 426         int failopen = 0;
 427
 428         /* rcu_read_lock()ed by nf_hook_slow() */
 429         queue = instance_lookup(queuenum);
 430         if (!queue) {
 431                 err = -ESRCH;
 432                 goto err_out;
 433         }
 434
 435         if (queue->copy_mode == NFQNL_COPY_NONE) {
 436                 err = -EINVAL;
 437                 goto err_out;
 438         }
 439
 440         nskb = nfqnl_build_packet_message(queue, entry, &packet_id_ptr);
 441         if (nskb == NULL) {
 442                 err = -ENOMEM;
 443                 goto err_out;
 444         }
 445         spin_lock_bh(&queue->lock);
 446
 447         if (!queue->peer_portid) {
 448                 err = -EINVAL;
 449                 goto err_out_free_nskb;
 450         }
 451         if (queue->queue_total >= queue->queue_maxlen) {
 452                 if (queue->flags & NFQA_CFG_F_FAIL_OPEN) {
 453                         failopen = 1;
 454                         err = 0;
 455                 } else {
 456                         queue->queue_dropped++;
 457                         net_warn_ratelimited("nf_queue: full at %d entries, dropping packets(s)\n",
 458                                              queue->queue_total);
 459                 }
 460                 goto err_out_free_nskb;
 461         }
 462         entry->id = ++queue->id_sequence;
 463         *packet_id_ptr = htonl(entry->id);
 464
 465         /* nfnetlink_unicast will either free the nskb or add it to a socket */
 466         err = nfnetlink_unicast(nskb, &init_net, queue->peer_portid, MSG_DONTWAIT);
 467         if (err < 0) {
 468                 queue->queue_user_dropped++;
 469                 goto err_out_unlock;
 470         }
 471
 472         __enqueue_entry(queue, entry);
 473
 474         spin_unlock_bh(&queue->lock);
 475         return 0;
 476
 477 err_out_free_nskb:
 478         kfree_skb(nskb);
 479 err_out_unlock:
 480         spin_unlock_bh(&queue->lock);
 481         if (failopen)
 482                 nf_reinject(entry, NF_ACCEPT);
 483 err_out:
 484         return err;
 485 }
 486
 487 static int
 488 nfqnl_mangle(void *data, int data_len, struct nf_queue_entry *e, int diff)
 489 {
 490         struct sk_buff *nskb;
 491
 492         if (diff < 0) {
 493                 if (pskb_trim(e->skb, data_len))
 494                         return -ENOMEM;
 495         } else if (diff > 0) {
 496                 if (data_len > 0xFFFF)
 497                         return -EINVAL;
 498                 if (diff > skb_tailroom(e->skb)) {
 499                         nskb = skb_copy_expand(e->skb, skb_headroom(e->skb),
 500                                                diff, GFP_ATOMIC);
 501                         if (!nskb) {
 502                                 printk(KERN_WARNING "nf_queue: OOM "
 503                                       "in mangle, dropping packet\n");
 504                                 return -ENOMEM;
 505                         }
 506                         kfree_skb(e->skb);
 507                         e->skb = nskb;
 508                 }
 509                 skb_put(e->skb, diff);
 510         }
 511         if (!skb_make_writable(e->skb, data_len))
 512                 return -ENOMEM;
 513         skb_copy_to_linear_data(e->skb, data, data_len);
 514         e->skb->ip_summed = CHECKSUM_NONE;
 515         return 0;
 516 }
 517
 518 static int
 519 nfqnl_set_mode(struct nfqnl_instance *queue,
 520                unsigned char mode, unsigned int range)
 521 {
 522         int status = 0;
 523
 524         spin_lock_bh(&queue->lock);
 525         switch (mode) {
 526         case NFQNL_COPY_NONE:
 527         case NFQNL_COPY_META:
 528                 queue->copy_mode = mode;
 529                 queue->copy_range = 0;
 530                 break;
 531
 532         case NFQNL_COPY_PACKET:
 533                 queue->copy_mode = mode;
 534                 /* We're using struct nlattr which has 16bit nla_len. Note that
 535                  * nla_len includes the header length. Thus, the maximum packet
 536                  * length that we support is 65531 bytes. We send truncated
 537                  * packets if the specified length is larger than that.
 538                  */
 539                 if (range > 0xffff - NLA_HDRLEN)
 540                         queue->copy_range = 0xffff - NLA_HDRLEN;
 541                 else
 542                         queue->copy_range = range;
 543                 break;
 544
 545         default:
 546                 status = -EINVAL;
 547
 548         }
 549         spin_unlock_bh(&queue->lock);
 550
 551         return status;
 552 }
 553
 554 static int
 555 dev_cmp(struct nf_queue_entry *entry, unsigned long ifindex)
 556 {
 557         if (entry->indev)
 558                 if (entry->indev->ifindex == ifindex)
 559                         return 1;
 560         if (entry->outdev)
 561                 if (entry->outdev->ifindex == ifindex)
 562                         return 1;
 563 #ifdef CONFIG_BRIDGE_NETFILTER
 564         if (entry->skb->nf_bridge) {
 565                 if (entry->skb->nf_bridge->physindev &&
 566                     entry->skb->nf_bridge->physindev->ifindex == ifindex)
 567                         return 1;
 568                 if (entry->skb->nf_bridge->physoutdev &&
 569                     entry->skb->nf_bridge->physoutdev->ifindex == ifindex)
 570                         return 1;
 571         }
 572 #endif
 573         return 0;
 574 }
 575
 576 /* drop all packets with either indev or outdev == ifindex from all queue
 577  * instances */
 578 static void
 579 nfqnl_dev_drop(int ifindex)
 580 {
 581         int i;
 582
 583         rcu_read_lock();
 584
 585         for (i = 0; i < INSTANCE_BUCKETS; i++) {
 586                 struct hlist_node *tmp;
 587                 struct nfqnl_instance *inst;
 588                 struct hlist_head *head = &instance_table[i];
 589
 590                 hlist_for_each_entry_rcu(inst, tmp, head, hlist)
 591                         nfqnl_flush(inst, dev_cmp, ifindex);
 592         }
 593
 594         rcu_read_unlock();
 595 }
 596
 597 #define RCV_SKB_FAIL(err) do { netlink_ack(skb, nlh, (err)); return; } while (0)
 598
 599 static int
 600 nfqnl_rcv_dev_event(struct notifier_block *this,
 601                     unsigned long event, void *ptr)
 602 {
 603         struct net_device *dev = ptr;
 604
 605         if (!net_eq(dev_net(dev), &init_net))
 606                 return NOTIFY_DONE;
 607
 608         /* Drop any packets associated with the downed device */
 609         if (event == NETDEV_DOWN)
 610                 nfqnl_dev_drop(dev->ifindex);
 611         return NOTIFY_DONE;
 612 }
 613
 614 static struct notifier_block nfqnl_dev_notifier = {
 615         .notifier_call  = nfqnl_rcv_dev_event,
 616 };
 617
 618 static int
 619 nfqnl_rcv_nl_event(struct notifier_block *this,
 620                    unsigned long event, void *ptr)
 621 {
 622         struct netlink_notify *n = ptr;
 623
 624         if (event == NETLINK_URELEASE && n->protocol == NETLINK_NETFILTER) {
 625                 int i;
 626
 627                 /* destroy all instances for this portid */
 628                 spin_lock(&instances_lock);
 629                 for (i = 0; i < INSTANCE_BUCKETS; i++) {
 630                         struct hlist_node *tmp, *t2;
 631                         struct nfqnl_instance *inst;
 632                         struct hlist_head *head = &instance_table[i];
 633
 634                         hlist_for_each_entry_safe(inst, tmp, t2, head, hlist) {
 635                                 if ((n->net == &init_net) &&
 636                                     (n->portid == inst->peer_portid))
 637                                         __instance_destroy(inst);
 638                         }
 639                 }
 640                 spin_unlock(&instances_lock);
 641         }
 642         return NOTIFY_DONE;
 643 }
 644
 645 static struct notifier_block nfqnl_rtnl_notifier = {
 646         .notifier_call  = nfqnl_rcv_nl_event,
 647 };
 648
 649 static const struct nla_policy nfqa_verdict_policy[NFQA_MAX+1] = {
 650         [NFQA_VERDICT_HDR]      = { .len = sizeof(struct nfqnl_msg_verdict_hdr) },
 651         [NFQA_MARK]             = { .type = NLA_U32 },
 652         [NFQA_PAYLOAD]          = { .type = NLA_UNSPEC },
 653         [NFQA_CT]               = { .type = NLA_UNSPEC },
 654 };
 655
 656 static const struct nla_policy nfqa_verdict_batch_policy[NFQA_MAX+1] = {
 657         [NFQA_VERDICT_HDR]      = { .len = sizeof(struct nfqnl_msg_verdict_hdr) },
 658         [NFQA_MARK]             = { .type = NLA_U32 },
 659 };
 660
 661 static struct nfqnl_instance *verdict_instance_lookup(u16 queue_num, int nlportid)
 662 {
 663         struct nfqnl_instance *queue;
 664
 665         queue = instance_lookup(queue_num);
 666         if (!queue)
 667                 return ERR_PTR(-ENODEV);
 668
 669         if (queue->peer_portid != nlportid)
 670                 return ERR_PTR(-EPERM);
 671
 672         return queue;
 673 }
 674
 675 static struct nfqnl_msg_verdict_hdr*
 676 verdicthdr_get(const struct nlattr * const nfqa[])
 677 {
 678         struct nfqnl_msg_verdict_hdr *vhdr;
 679         unsigned int verdict;
 680
 681         if (!nfqa[NFQA_VERDICT_HDR])
 682                 return NULL;
 683
 684         vhdr = nla_data(nfqa[NFQA_VERDICT_HDR]);
 685         verdict = ntohl(vhdr->verdict) & NF_VERDICT_MASK;
 686         if (verdict > NF_MAX_VERDICT || verdict == NF_STOLEN)
 687                 return NULL;
 688         return vhdr;
 689 }
 690
 691 static int nfq_id_after(unsigned int id, unsigned int max)
 692 {
 693         return (int)(id - max) > 0;
 694 }
 695
 696 static int
 697 nfqnl_recv_verdict_batch(struct sock *ctnl, struct sk_buff *skb,
 698                    const struct nlmsghdr *nlh,
 699                    const struct nlattr * const nfqa[])
 700 {
 701         struct nfgenmsg *nfmsg = nlmsg_data(nlh);
 702         struct nf_queue_entry *entry, *tmp;
 703         unsigned int verdict, maxid;
 704         struct nfqnl_msg_verdict_hdr *vhdr;
 705         struct nfqnl_instance *queue;
 706         LIST_HEAD(batch_list);
 707         u16 queue_num = ntohs(nfmsg->res_id);
 708
 709         queue = verdict_instance_lookup(queue_num, NETLINK_CB(skb).portid);
 710         if (IS_ERR(queue))
 711                 return PTR_ERR(queue);
 712
 713         vhdr = verdicthdr_get(nfqa);
 714         if (!vhdr)
 715                 return -EINVAL;
 716
 717         verdict = ntohl(vhdr->verdict);
 718         maxid = ntohl(vhdr->id);
 719
 720         spin_lock_bh(&queue->lock);
 721
 722         list_for_each_entry_safe(entry, tmp, &queue->queue_list, list) {
 723                 if (nfq_id_after(entry->id, maxid))
 724                         break;
 725                 __dequeue_entry(queue, entry);
 726                 list_add_tail(&entry->list, &batch_list);
 727         }
 728
 729         spin_unlock_bh(&queue->lock);
 730
 731         if (list_empty(&batch_list))
 732                 return -ENOENT;
 733
 734         list_for_each_entry_safe(entry, tmp, &batch_list, list) {
 735                 if (nfqa[NFQA_MARK])
 736                         entry->skb->mark = ntohl(nla_get_be32(nfqa[NFQA_MARK]));
 737                 nf_reinject(entry, verdict);
 738         }
 739         return 0;
 740 }
 741
 742 static int
 743 nfqnl_recv_verdict(struct sock *ctnl, struct sk_buff *skb,
 744                    const struct nlmsghdr *nlh,
 745                    const struct nlattr * const nfqa[])
 746 {
 747         struct nfgenmsg *nfmsg = nlmsg_data(nlh);
 748         u_int16_t queue_num = ntohs(nfmsg->res_id);
 749
 750         struct nfqnl_msg_verdict_hdr *vhdr;
 751         struct nfqnl_instance *queue;
 752         unsigned int verdict;
 753         struct nf_queue_entry *entry;
 754         enum ip_conntrack_info uninitialized_var(ctinfo);
 755         struct nf_conn *ct = NULL;
 756
 757         queue = instance_lookup(queue_num);
 758         if (!queue)
 759
 760         queue = verdict_instance_lookup(queue_num, NETLINK_CB(skb).portid);
 761         if (IS_ERR(queue))
 762                 return PTR_ERR(queue);
 763
 764         vhdr = verdicthdr_get(nfqa);
 765         if (!vhdr)
 766                 return -EINVAL;
 767
 768         verdict = ntohl(vhdr->verdict);
 769
 770         entry = find_dequeue_entry(queue, ntohl(vhdr->id));
 771         if (entry == NULL)
 772                 return -ENOENT;
 773
 774         rcu_read_lock();
 775         if (nfqa[NFQA_CT] && (queue->flags & NFQA_CFG_F_CONNTRACK))
 776                 ct = nfqnl_ct_parse(entry->skb, nfqa[NFQA_CT], &ctinfo);
 777
 778         if (nfqa[NFQA_PAYLOAD]) {
 779                 u16 payload_len = nla_len(nfqa[NFQA_PAYLOAD]);
 780                 int diff = payload_len - entry->skb->len;
 781
 782                 if (nfqnl_mangle(nla_data(nfqa[NFQA_PAYLOAD]),
 783                                  payload_len, entry, diff) < 0)
 784                         verdict = NF_DROP;
 785
 786                 if (ct)
 787                         nfqnl_ct_seq_adjust(skb, ct, ctinfo, diff);
 788         }
 789         rcu_read_unlock();
 790
 791         if (nfqa[NFQA_MARK])
 792                 entry->skb->mark = ntohl(nla_get_be32(nfqa[NFQA_MARK]));
 793
 794         nf_reinject(entry, verdict);
 795         return 0;
 796 }
 797
 798 static int
 799 nfqnl_recv_unsupp(struct sock *ctnl, struct sk_buff *skb,
 800                   const struct nlmsghdr *nlh,
 801                   const struct nlattr * const nfqa[])
 802 {
 803         return -ENOTSUPP;
 804 }
 805
 806 static const struct nla_policy nfqa_cfg_policy[NFQA_CFG_MAX+1] = {
 807         [NFQA_CFG_CMD]          = { .len = sizeof(struct nfqnl_msg_config_cmd) },
 808         [NFQA_CFG_PARAMS]       = { .len = sizeof(struct nfqnl_msg_config_params) },
 809 };
 810
 811 static const struct nf_queue_handler nfqh = {
 812         .outfn  = &nfqnl_enqueue_packet,
 813 };
 814
 815 static int
 816 nfqnl_recv_config(struct sock *ctnl, struct sk_buff *skb,
 817                   const struct nlmsghdr *nlh,
 818                   const struct nlattr * const nfqa[])
 819 {
 820         struct nfgenmsg *nfmsg = nlmsg_data(nlh);
 821         u_int16_t queue_num = ntohs(nfmsg->res_id);
 822         struct nfqnl_instance *queue;
 823         struct nfqnl_msg_config_cmd *cmd = NULL;
 824         int ret = 0;
 825
 826         if (nfqa[NFQA_CFG_CMD]) {
 827                 cmd = nla_data(nfqa[NFQA_CFG_CMD]);
 828
 829                 /* Obsolete commands without queue context */
 830                 switch (cmd->command) {
 831                 case NFQNL_CFG_CMD_PF_BIND: return 0;
 832                 case NFQNL_CFG_CMD_PF_UNBIND: return 0;
 833                 }
 834         }
 835
 836         rcu_read_lock();
 837         queue = instance_lookup(queue_num);
 838         if (queue && queue->peer_portid != NETLINK_CB(skb).portid) {
 839                 ret = -EPERM;
 840                 goto err_out_unlock;
 841         }
 842
 843         if (cmd != NULL) {
 844                 switch (cmd->command) {
 845                 case NFQNL_CFG_CMD_BIND:
 846                         if (queue) {
 847                                 ret = -EBUSY;
 848                                 goto err_out_unlock;
 849                         }
 850                         queue = instance_create(queue_num, NETLINK_CB(skb).portid);
 851                         if (IS_ERR(queue)) {
 852                                 ret = PTR_ERR(queue);
 853                                 goto err_out_unlock;
 854                         }
 855                         break;
 856                 case NFQNL_CFG_CMD_UNBIND:
 857                         if (!queue) {
 858                                 ret = -ENODEV;
 859                                 goto err_out_unlock;
 860                         }
 861                         instance_destroy(queue);
 862                         break;
 863                 case NFQNL_CFG_CMD_PF_BIND:
 864                 case NFQNL_CFG_CMD_PF_UNBIND:
 865                         break;
 866                 default:
 867                         ret = -ENOTSUPP;
 868                         break;
 869                 }
 870         }
 871
 872         if (nfqa[NFQA_CFG_PARAMS]) {
 873                 struct nfqnl_msg_config_params *params;
 874
 875                 if (!queue) {
 876                         ret = -ENODEV;
 877                         goto err_out_unlock;
 878                 }
 879                 params = nla_data(nfqa[NFQA_CFG_PARAMS]);
 880                 nfqnl_set_mode(queue, params->copy_mode,
 881                                 ntohl(params->copy_range));
 882         }
 883
 884         if (nfqa[NFQA_CFG_QUEUE_MAXLEN]) {
 885                 __be32 *queue_maxlen;
 886
 887                 if (!queue) {
 888                         ret = -ENODEV;
 889                         goto err_out_unlock;
 890                 }
 891                 queue_maxlen = nla_data(nfqa[NFQA_CFG_QUEUE_MAXLEN]);
 892                 spin_lock_bh(&queue->lock);
 893                 queue->queue_maxlen = ntohl(*queue_maxlen);
 894                 spin_unlock_bh(&queue->lock);
 895         }
 896
 897         if (nfqa[NFQA_CFG_FLAGS]) {
 898                 __u32 flags, mask;
 899
 900                 if (!queue) {
 901                         ret = -ENODEV;
 902                         goto err_out_unlock;
 903                 }
 904
 905                 if (!nfqa[NFQA_CFG_MASK]) {
 906                         /* A mask is needed to specify which flags are being
 907                          * changed.
 908                          */
 909                         ret = -EINVAL;
 910                         goto err_out_unlock;
 911                 }
 912
 913                 flags = ntohl(nla_get_be32(nfqa[NFQA_CFG_FLAGS]));
 914                 mask = ntohl(nla_get_be32(nfqa[NFQA_CFG_MASK]));
 915
 916                 if (flags >= NFQA_CFG_F_MAX) {
 917                         ret = -EOPNOTSUPP;
 918                         goto err_out_unlock;
 919                 }
 920
 921                 spin_lock_bh(&queue->lock);
 922                 queue->flags &= ~mask;
 923                 queue->flags |= flags & mask;
 924                 spin_unlock_bh(&queue->lock);
 925         }
 926
 927 err_out_unlock:
 928         rcu_read_unlock();
 929         return ret;
 930 }
 931
 932 static const struct nfnl_callback nfqnl_cb[NFQNL_MSG_MAX] = {
 933         [NFQNL_MSG_PACKET]      = { .call_rcu = nfqnl_recv_unsupp,
 934                                     .attr_count = NFQA_MAX, },
 935         [NFQNL_MSG_VERDICT]     = { .call_rcu = nfqnl_recv_verdict,
 936                                     .attr_count = NFQA_MAX,
 937                                     .policy = nfqa_verdict_policy },
 938         [NFQNL_MSG_CONFIG]      = { .call = nfqnl_recv_config,
 939                                     .attr_count = NFQA_CFG_MAX,
 940                                     .policy = nfqa_cfg_policy },
 941         [NFQNL_MSG_VERDICT_BATCH]={ .call_rcu = nfqnl_recv_verdict_batch,
 942                                     .attr_count = NFQA_MAX,
 943                                     .policy = nfqa_verdict_batch_policy },
 944 };
 945
 946 static const struct nfnetlink_subsystem nfqnl_subsys = {
 947         .name           = "nf_queue",
 948         .subsys_id      = NFNL_SUBSYS_QUEUE,
 949         .cb_count       = NFQNL_MSG_MAX,
 950         .cb             = nfqnl_cb,
 951 };
 952
 953 #ifdef CONFIG_PROC_FS
 954 struct iter_state {
 955         unsigned int bucket;
 956 };
 957
 958 static struct hlist_node *get_first(struct seq_file *seq)
 959 {
 960         struct iter_state *st = seq->private;
 961
 962         if (!st)
 963                 return NULL;
 964
 965         for (st->bucket = 0; st->bucket < INSTANCE_BUCKETS; st->bucket++) {
 966                 if (!hlist_empty(&instance_table[st->bucket]))
 967                         return instance_table[st->bucket].first;
 968         }
 969         return NULL;
 970 }
 971
 972 static struct hlist_node *get_next(struct seq_file *seq, struct hlist_node *h)
 973 {
 974         struct iter_state *st = seq->private;
 975
 976         h = h->next;
 977         while (!h) {
 978                 if (++st->bucket >= INSTANCE_BUCKETS)
 979                         return NULL;
 980
 981                 h = instance_table[st->bucket].first;
 982         }
 983         return h;
 984 }
 985
 986 static struct hlist_node *get_idx(struct seq_file *seq, loff_t pos)
 987 {
 988         struct hlist_node *head;
 989         head = get_first(seq);
 990
 991         if (head)
 992                 while (pos && (head = get_next(seq, head)))
 993                         pos--;
 994         return pos ? NULL : head;
 995 }
 996
 997 static void *seq_start(struct seq_file *seq, loff_t *pos)
 998         __acquires(instances_lock)
 999 {
1000         spin_lock(&instances_lock);
1001         return get_idx(seq, *pos);
1002 }
1003
1004 static void *seq_next(struct seq_file *s, void *v, loff_t *pos)
1005 {
1006         (*pos)++;
1007         return get_next(s, v);
1008 }
1009
1010 static void seq_stop(struct seq_file *s, void *v)
1011         __releases(instances_lock)
1012 {
1013         spin_unlock(&instances_lock);
1014 }
1015
1016 static int seq_show(struct seq_file *s, void *v)
1017 {
1018         const struct nfqnl_instance *inst = v;
1019
1020         return seq_printf(s, "%5d %6d %5d %1d %5d %5d %5d %8d %2d\n",
1021                           inst->queue_num,
1022                           inst->peer_portid, inst->queue_total,
1023                           inst->copy_mode, inst->copy_range,
1024                           inst->queue_dropped, inst->queue_user_dropped,
1025                           inst->id_sequence, 1);
1026 }
1027
1028 static const struct seq_operations nfqnl_seq_ops = {
1029         .start  = seq_start,
1030         .next   = seq_next,
1031         .stop   = seq_stop,
1032         .show   = seq_show,
1033 };
1034
1035 static int nfqnl_open(struct inode *inode, struct file *file)
1036 {
1037         return seq_open_private(file, &nfqnl_seq_ops,
1038                         sizeof(struct iter_state));
1039 }
1040
1041 static const struct file_operations nfqnl_file_ops = {
1042         .owner   = THIS_MODULE,
1043         .open    = nfqnl_open,
1044         .read    = seq_read,
1045         .llseek  = seq_lseek,
1046         .release = seq_release_private,
1047 };
1048
1049 #endif /* PROC_FS */
1050
1051 static int __init nfnetlink_queue_init(void)
1052 {
1053         int i, status = -ENOMEM;
1054
1055         for (i = 0; i < INSTANCE_BUCKETS; i++)
1056                 INIT_HLIST_HEAD(&instance_table[i]);
1057
1058         netlink_register_notifier(&nfqnl_rtnl_notifier);
1059         status = nfnetlink_subsys_register(&nfqnl_subsys);
1060         if (status < 0) {
1061                 printk(KERN_ERR "nf_queue: failed to create netlink socket\n");
1062                 goto cleanup_netlink_notifier;
1063         }
1064
1065 #ifdef CONFIG_PROC_FS
1066         if (!proc_create("nfnetlink_queue", 0440,
1067                          proc_net_netfilter, &nfqnl_file_ops))
1068                 goto cleanup_subsys;
1069 #endif
1070
1071         register_netdevice_notifier(&nfqnl_dev_notifier);
1072         nf_register_queue_handler(&nfqh);
1073         return status;
1074
1075 #ifdef CONFIG_PROC_FS
1076 cleanup_subsys:
1077         nfnetlink_subsys_unregister(&nfqnl_subsys);
1078 #endif
1079 cleanup_netlink_notifier:
1080         netlink_unregister_notifier(&nfqnl_rtnl_notifier);
1081         return status;
1082 }
1083
1084 static void __exit nfnetlink_queue_fini(void)
1085 {
1086         nf_unregister_queue_handler();
1087         unregister_netdevice_notifier(&nfqnl_dev_notifier);
1088 #ifdef CONFIG_PROC_FS
1089         remove_proc_entry("nfnetlink_queue", proc_net_netfilter);
1090 #endif
1091         nfnetlink_subsys_unregister(&nfqnl_subsys);
1092         netlink_unregister_notifier(&nfqnl_rtnl_notifier);
1093
1094         rcu_barrier(); /* Wait for completion of call_rcu()'s */
1095 }
1096
1097 MODULE_DESCRIPTION("netfilter packet queue handler");
1098 MODULE_AUTHOR("Harald Welte <laforge@netfilter.org>");
1099 MODULE_LICENSE("GPL");
1100 MODULE_ALIAS_NFNL_SUBSYS(NFNL_SUBSYS_QUEUE);
1101
1102 module_init(nfnetlink_queue_init);
1103 module_exit(nfnetlink_queue_fini);