net/netfilter/nfnetlink_queue.c

   1 /*
   2  * This is a module which is used for queueing packets and communicating with
   3  * userspace via nfetlink.
   4  *
   5  * (C) 2005 by Harald Welte <laforge@netfilter.org>
   6  *
   7  * Based on the old ipv4-only ip_queue.c:
   8  * (C) 2000-2002 James Morris <jmorris@intercode.com.au>
   9  * (C) 2003-2005 Netfilter Core Team <coreteam@netfilter.org>
  10  *
  11  * This program is free software; you can redistribute it and/or modify
  12  * it under the terms of the GNU General Public License version 2 as
  13  * published by the Free Software Foundation.
  14  *
  15  */
  16 #include <linux/module.h>
  17 #include <linux/skbuff.h>
  18 #include <linux/init.h>
  19 #include <linux/spinlock.h>
  20 #include <linux/notifier.h>
  21 #include <linux/netdevice.h>
  22 #include <linux/netfilter.h>
  23 #include <linux/proc_fs.h>
  24 #include <linux/netfilter_ipv4.h>
  25 #include <linux/netfilter_ipv6.h>
  26 #include <linux/netfilter/nfnetlink.h>
  27 #include <linux/netfilter/nfnetlink_queue.h>
  28 #include <linux/list.h>
  29 #include <net/sock.h>
  30
  31 #include <asm/atomic.h>
  32
  33 #define NFQNL_QMAX_DEFAULT 1024
  34
  35 #if 0
  36 #define QDEBUG(x, args ...)     printk(KERN_DEBUG "%s(%d):%s(): " x,       \
  37                                         __FILE__, __LINE__, __FUNCTION__,  \
  38                                         ## args)
  39 #else
  40 #define QDEBUG(x, ...)
  41 #endif
  42
  43 struct nfqnl_queue_entry {
  44         struct list_head list;
  45         struct nf_info *info;
  46         struct sk_buff *skb;
  47         unsigned int id;
  48 };
  49
  50 struct nfqnl_instance {
  51         struct hlist_node hlist;                /* global list of queues */
  52         atomic_t use;
  53
  54         int peer_pid;
  55         unsigned int queue_maxlen;
  56         unsigned int copy_range;
  57         unsigned int queue_total;
  58         unsigned int queue_dropped;
  59         unsigned int queue_user_dropped;
  60
  61         atomic_t id_sequence;                   /* 'sequence' of pkt ids */
  62
  63         u_int16_t queue_num;                    /* number of this queue */
  64         u_int8_t copy_mode;
  65
  66         spinlock_t lock;
  67
  68         struct list_head queue_list;            /* packets in queue */
  69 };
  70
  71 typedef int (*nfqnl_cmpfn)(struct nfqnl_queue_entry *, unsigned long);
  72
  73 static DEFINE_RWLOCK(instances_lock);
  74
  75 u_int64_t htonll(u_int64_t in)
  76 {
  77         u_int64_t out;
  78         int i;
  79
  80         for (i = 0; i < sizeof(u_int64_t); i++)
  81                 ((u_int8_t *)&out)[sizeof(u_int64_t)-1] = ((u_int8_t *)&in)[i];
  82
  83         return out;
  84 }
  85
  86 #define INSTANCE_BUCKETS        16
  87 static struct hlist_head instance_table[INSTANCE_BUCKETS];
  88
  89 static inline u_int8_t instance_hashfn(u_int16_t queue_num)
  90 {
  91         return ((queue_num >> 8) | queue_num) % INSTANCE_BUCKETS;
  92 }
  93
  94 static struct nfqnl_instance *
  95 __instance_lookup(u_int16_t queue_num)
  96 {
  97         struct hlist_head *head;
  98         struct hlist_node *pos;
  99         struct nfqnl_instance *inst;
 100
 101         head = &instance_table[instance_hashfn(queue_num)];
 102         hlist_for_each_entry(inst, pos, head, hlist) {
 103                 if (inst->queue_num == queue_num)
 104                         return inst;
 105         }
 106         return NULL;
 107 }
 108
 109 static struct nfqnl_instance *
 110 instance_lookup_get(u_int16_t queue_num)
 111 {
 112         struct nfqnl_instance *inst;
 113
 114         read_lock_bh(&instances_lock);
 115         inst = __instance_lookup(queue_num);
 116         if (inst)
 117                 atomic_inc(&inst->use);
 118         read_unlock_bh(&instances_lock);
 119
 120         return inst;
 121 }
 122
 123 static void
 124 instance_put(struct nfqnl_instance *inst)
 125 {
 126         if (inst && atomic_dec_and_test(&inst->use)) {
 127                 QDEBUG("kfree(inst=%p)\n", inst);
 128                 kfree(inst);
 129         }
 130 }
 131
 132 static struct nfqnl_instance *
 133 instance_create(u_int16_t queue_num, int pid)
 134 {
 135         struct nfqnl_instance *inst;
 136
 137         QDEBUG("entering for queue_num=%u, pid=%d\n", queue_num, pid);
 138
 139         write_lock_bh(&instances_lock);
 140         if (__instance_lookup(queue_num)) {
 141                 inst = NULL;
 142                 QDEBUG("aborting, instance already exists\n");
 143                 goto out_unlock;
 144         }
 145
 146         inst = kmalloc(sizeof(*inst), GFP_ATOMIC);
 147         if (!inst)
 148                 goto out_unlock;
 149
 150         memset(inst, 0, sizeof(*inst));
 151         inst->queue_num = queue_num;
 152         inst->peer_pid = pid;
 153         inst->queue_maxlen = NFQNL_QMAX_DEFAULT;
 154         inst->copy_range = 0xfffff;
 155         inst->copy_mode = NFQNL_COPY_NONE;
 156         atomic_set(&inst->id_sequence, 0);
 157         /* needs to be two, since we _put() after creation */
 158         atomic_set(&inst->use, 2);
 159         inst->lock = SPIN_LOCK_UNLOCKED;
 160         INIT_LIST_HEAD(&inst->queue_list);
 161
 162         if (!try_module_get(THIS_MODULE))
 163                 goto out_free;
 164
 165         hlist_add_head(&inst->hlist,
 166                        &instance_table[instance_hashfn(queue_num)]);
 167
 168         write_unlock_bh(&instances_lock);
 169
 170         QDEBUG("successfully created new instance\n");
 171
 172         return inst;
 173
 174 out_free:
 175         kfree(inst);
 176 out_unlock:
 177         write_unlock_bh(&instances_lock);
 178         return NULL;
 179 }
 180
 181 static void nfqnl_flush(struct nfqnl_instance *queue, int verdict);
 182
 183 static void
 184 _instance_destroy2(struct nfqnl_instance *inst, int lock)
 185 {
 186         /* first pull it out of the global list */
 187         if (lock)
 188                 write_lock_bh(&instances_lock);
 189
 190         QDEBUG("removing instance %p (queuenum=%u) from hash\n",
 191                 inst, inst->queue_num);
 192         hlist_del(&inst->hlist);
 193
 194         if (lock)
 195                 write_unlock_bh(&instances_lock);
 196
 197         /* then flush all pending skbs from the queue */
 198         nfqnl_flush(inst, NF_DROP);
 199
 200         /* and finally put the refcount */
 201         instance_put(inst);
 202
 203         module_put(THIS_MODULE);
 204 }
 205
 206 static inline void
 207 __instance_destroy(struct nfqnl_instance *inst)
 208 {
 209         _instance_destroy2(inst, 0);
 210 }
 211
 212 static inline void
 213 instance_destroy(struct nfqnl_instance *inst)
 214 {
 215         _instance_destroy2(inst, 1);
 216 }
 217
 218
 219
 220 static void
 221 issue_verdict(struct nfqnl_queue_entry *entry, int verdict)
 222 {
 223         QDEBUG("entering for entry %p, verdict %u\n", entry, verdict);
 224
 225         /* TCP input path (and probably other bits) assume to be called
 226          * from softirq context, not from syscall, like issue_verdict is
 227          * called.  TCP input path deadlocks with locks taken from timer
 228          * softirq, e.g.  We therefore emulate this by local_bh_disable() */
 229
 230         local_bh_disable();
 231         nf_reinject(entry->skb, entry->info, verdict);
 232         local_bh_enable();
 233
 234         kfree(entry);
 235 }
 236
 237 static inline void
 238 __enqueue_entry(struct nfqnl_instance *queue,
 239                       struct nfqnl_queue_entry *entry)
 240 {
 241        list_add(&entry->list, &queue->queue_list);
 242        queue->queue_total++;
 243 }
 244
 245 /*
 246  * Find and return a queued entry matched by cmpfn, or return the last
 247  * entry if cmpfn is NULL.
 248  */
 249 static inline struct nfqnl_queue_entry *
 250 __find_entry(struct nfqnl_instance *queue, nfqnl_cmpfn cmpfn,
 251                    unsigned long data)
 252 {
 253         struct list_head *p;
 254
 255         list_for_each_prev(p, &queue->queue_list) {
 256                 struct nfqnl_queue_entry *entry = (struct nfqnl_queue_entry *)p;
 257
 258                 if (!cmpfn || cmpfn(entry, data))
 259                         return entry;
 260         }
 261         return NULL;
 262 }
 263
 264 static inline void
 265 __dequeue_entry(struct nfqnl_instance *q, struct nfqnl_queue_entry *entry)
 266 {
 267         list_del(&entry->list);
 268         q->queue_total--;
 269 }
 270
 271 static inline struct nfqnl_queue_entry *
 272 __find_dequeue_entry(struct nfqnl_instance *queue,
 273                      nfqnl_cmpfn cmpfn, unsigned long data)
 274 {
 275         struct nfqnl_queue_entry *entry;
 276
 277         entry = __find_entry(queue, cmpfn, data);
 278         if (entry == NULL)
 279                 return NULL;
 280
 281         __dequeue_entry(queue, entry);
 282         return entry;
 283 }
 284
 285
 286 static inline void
 287 __nfqnl_flush(struct nfqnl_instance *queue, int verdict)
 288 {
 289         struct nfqnl_queue_entry *entry;
 290
 291         while ((entry = __find_dequeue_entry(queue, NULL, 0)))
 292                 issue_verdict(entry, verdict);
 293 }
 294
 295 static inline int
 296 __nfqnl_set_mode(struct nfqnl_instance *queue,
 297                  unsigned char mode, unsigned int range)
 298 {
 299         int status = 0;
 300
 301         switch (mode) {
 302         case NFQNL_COPY_NONE:
 303         case NFQNL_COPY_META:
 304                 queue->copy_mode = mode;
 305                 queue->copy_range = 0;
 306                 break;
 307
 308         case NFQNL_COPY_PACKET:
 309                 queue->copy_mode = mode;
 310                 /* we're using struct nfattr which has 16bit nfa_len */
 311                 if (range > 0xffff)
 312                         queue->copy_range = 0xffff;
 313                 else
 314                         queue->copy_range = range;
 315                 break;
 316
 317         default:
 318                 status = -EINVAL;
 319
 320         }
 321         return status;
 322 }
 323
 324 static struct nfqnl_queue_entry *
 325 find_dequeue_entry(struct nfqnl_instance *queue,
 326                          nfqnl_cmpfn cmpfn, unsigned long data)
 327 {
 328         struct nfqnl_queue_entry *entry;
 329
 330         spin_lock_bh(&queue->lock);
 331         entry = __find_dequeue_entry(queue, cmpfn, data);
 332         spin_unlock_bh(&queue->lock);
 333
 334         return entry;
 335 }
 336
 337 static void
 338 nfqnl_flush(struct nfqnl_instance *queue, int verdict)
 339 {
 340         spin_lock_bh(&queue->lock);
 341         __nfqnl_flush(queue, verdict);
 342         spin_unlock_bh(&queue->lock);
 343 }
 344
 345 static struct sk_buff *
 346 nfqnl_build_packet_message(struct nfqnl_instance *queue,
 347                            struct nfqnl_queue_entry *entry, int *errp)
 348 {
 349         unsigned char *old_tail;
 350         size_t size;
 351         size_t data_len = 0;
 352         struct sk_buff *skb;
 353         struct nfqnl_msg_packet_hdr pmsg;
 354         struct nlmsghdr *nlh;
 355         struct nfgenmsg *nfmsg;
 356         unsigned int tmp_uint;
 357
 358         QDEBUG("entered\n");
 359
 360         /* all macros expand to constant values at compile time */
 361         size =    NLMSG_SPACE(sizeof(struct nfqnl_msg_packet_hdr))
 362                 + NLMSG_SPACE(sizeof(u_int32_t))        /* ifindex */
 363                 + NLMSG_SPACE(sizeof(u_int32_t))        /* ifindex */
 364                 + NLMSG_SPACE(sizeof(u_int32_t))        /* mark */
 365                 + NLMSG_SPACE(sizeof(struct nfqnl_msg_packet_hw))
 366                 + NLMSG_SPACE(sizeof(struct nfqnl_msg_packet_timestamp));
 367
 368         spin_lock_bh(&queue->lock);
 369
 370         switch (queue->copy_mode) {
 371         case NFQNL_COPY_META:
 372         case NFQNL_COPY_NONE:
 373                 data_len = 0;
 374                 break;
 375
 376         case NFQNL_COPY_PACKET:
 377                 if (queue->copy_range == 0
 378                     || queue->copy_range > entry->skb->len)
 379                         data_len = entry->skb->len;
 380                 else
 381                         data_len = queue->copy_range;
 382
 383                 size += NLMSG_SPACE(data_len);
 384                 break;
 385
 386         default:
 387                 *errp = -EINVAL;
 388                 spin_unlock_bh(&queue->lock);
 389                 return NULL;
 390         }
 391
 392         spin_unlock_bh(&queue->lock);
 393
 394         skb = alloc_skb(size, GFP_ATOMIC);
 395         if (!skb)
 396                 goto nlmsg_failure;
 397
 398         old_tail= skb->tail;
 399         nlh = NLMSG_PUT(skb, 0, 0,
 400                         NFNL_SUBSYS_QUEUE << 8 | NFQNL_MSG_PACKET,
 401                         sizeof(struct nfgenmsg));
 402         nfmsg = NLMSG_DATA(nlh);
 403         nfmsg->nfgen_family = entry->info->pf;
 404         nfmsg->version = NFNETLINK_V0;
 405         nfmsg->res_id = htons(queue->queue_num);
 406
 407         pmsg.packet_id          = htonl(entry->id);
 408         pmsg.hw_protocol        = htons(entry->skb->protocol);
 409         pmsg.hook               = entry->info->hook;
 410
 411         NFA_PUT(skb, NFQA_PACKET_HDR, sizeof(pmsg), &pmsg);
 412
 413         if (entry->info->indev) {
 414                 tmp_uint = htonl(entry->info->indev->ifindex);
 415                 NFA_PUT(skb, NFQA_IFINDEX_INDEV, sizeof(tmp_uint), &tmp_uint);
 416         }
 417
 418         if (entry->info->outdev) {
 419                 tmp_uint = htonl(entry->info->outdev->ifindex);
 420                 NFA_PUT(skb, NFQA_IFINDEX_OUTDEV, sizeof(tmp_uint), &tmp_uint);
 421         }
 422
 423         if (entry->skb->nfmark) {
 424                 tmp_uint = htonl(entry->skb->nfmark);
 425                 NFA_PUT(skb, NFQA_MARK, sizeof(u_int32_t), &tmp_uint);
 426         }
 427
 428         if (entry->info->indev && entry->skb->dev
 429             && entry->skb->dev->hard_header_parse) {
 430                 struct nfqnl_msg_packet_hw phw;
 431
 432                 phw.hw_addrlen =
 433                         entry->skb->dev->hard_header_parse(entry->skb,
 434                                                            phw.hw_addr);
 435                 phw.hw_addrlen = htons(phw.hw_addrlen);
 436                 NFA_PUT(skb, NFQA_HWADDR, sizeof(phw), &phw);
 437         }
 438
 439         if (entry->skb->stamp.tv_sec) {
 440                 struct nfqnl_msg_packet_timestamp ts;
 441
 442                 ts.sec = htonll(entry->skb->stamp.tv_sec);
 443                 ts.usec = htonll(entry->skb->stamp.tv_usec);
 444
 445                 NFA_PUT(skb, NFQA_TIMESTAMP, sizeof(ts), &ts);
 446         }
 447
 448         if (data_len) {
 449                 struct nfattr *nfa;
 450                 int size = NFA_LENGTH(data_len);
 451
 452                 if (skb_tailroom(skb) < (int)NFA_SPACE(data_len)) {
 453                         printk(KERN_WARNING "nf_queue: no tailroom!\n");
 454                         goto nlmsg_failure;
 455                 }
 456
 457                 nfa = (struct nfattr *)skb_put(skb, NFA_ALIGN(size));
 458                 nfa->nfa_type = NFQA_PAYLOAD;
 459                 nfa->nfa_len = size;
 460
 461                 if (skb_copy_bits(entry->skb, 0, NFA_DATA(nfa), data_len))
 462                         BUG();
 463         }
 464
 465         nlh->nlmsg_len = skb->tail - old_tail;
 466         return skb;
 467
 468 nlmsg_failure:
 469 nfattr_failure:
 470         if (skb)
 471                 kfree_skb(skb);
 472         *errp = -EINVAL;
 473         if (net_ratelimit())
 474                 printk(KERN_ERR "nf_queue: error creating packet message\n");
 475         return NULL;
 476 }
 477
 478 static int
 479 nfqnl_enqueue_packet(struct sk_buff *skb, struct nf_info *info,
 480                      unsigned int queuenum, void *data)
 481 {
 482         int status = -EINVAL;
 483         struct sk_buff *nskb;
 484         struct nfqnl_instance *queue;
 485         struct nfqnl_queue_entry *entry;
 486
 487         QDEBUG("entered\n");
 488
 489         queue = instance_lookup_get(queuenum);
 490         if (!queue) {
 491                 QDEBUG("no queue instance matching\n");
 492                 return -EINVAL;
 493         }
 494
 495         if (queue->copy_mode == NFQNL_COPY_NONE) {
 496                 QDEBUG("mode COPY_NONE, aborting\n");
 497                 status = -EAGAIN;
 498                 goto err_out_put;
 499         }
 500
 501         entry = kmalloc(sizeof(*entry), GFP_ATOMIC);
 502         if (entry == NULL) {
 503                 if (net_ratelimit())
 504                         printk(KERN_ERR
 505                                 "nf_queue: OOM in nfqnl_enqueue_packet()\n");
 506                 status = -ENOMEM;
 507                 goto err_out_put;
 508         }
 509
 510         entry->info = info;
 511         entry->skb = skb;
 512         entry->id = atomic_inc_return(&queue->id_sequence);
 513
 514         nskb = nfqnl_build_packet_message(queue, entry, &status);
 515         if (nskb == NULL)
 516                 goto err_out_free;
 517
 518         spin_lock_bh(&queue->lock);
 519
 520         if (!queue->peer_pid)
 521                 goto err_out_free_nskb;
 522
 523         if (queue->queue_total >= queue->queue_maxlen) {
 524                 queue->queue_dropped++;
 525                 status = -ENOSPC;
 526                 if (net_ratelimit())
 527                           printk(KERN_WARNING "ip_queue: full at %d entries, "
 528                                  "dropping packets(s). Dropped: %d\n",
 529                                  queue->queue_total, queue->queue_dropped);
 530                 goto err_out_free_nskb;
 531         }
 532
 533         /* nfnetlink_unicast will either free the nskb or add it to a socket */
 534         status = nfnetlink_unicast(nskb, queue->peer_pid, MSG_DONTWAIT);
 535         if (status < 0) {
 536                 queue->queue_user_dropped++;
 537                 goto err_out_unlock;
 538         }
 539
 540         __enqueue_entry(queue, entry);
 541
 542         spin_unlock_bh(&queue->lock);
 543         instance_put(queue);
 544         return status;
 545
 546 err_out_free_nskb:
 547         kfree_skb(nskb);
 548
 549 err_out_unlock:
 550         spin_unlock_bh(&queue->lock);
 551
 552 err_out_free:
 553         kfree(entry);
 554 err_out_put:
 555         instance_put(queue);
 556         return status;
 557 }
 558
 559 static int
 560 nfqnl_mangle(void *data, int data_len, struct nfqnl_queue_entry *e)
 561 {
 562         int diff;
 563
 564         diff = data_len - e->skb->len;
 565         if (diff < 0)
 566                 skb_trim(e->skb, data_len);
 567         else if (diff > 0) {
 568                 if (data_len > 0xFFFF)
 569                         return -EINVAL;
 570                 if (diff > skb_tailroom(e->skb)) {
 571                         struct sk_buff *newskb;
 572
 573                         newskb = skb_copy_expand(e->skb,
 574                                                  skb_headroom(e->skb),
 575                                                  diff,
 576                                                  GFP_ATOMIC);
 577                         if (newskb == NULL) {
 578                                 printk(KERN_WARNING "ip_queue: OOM "
 579                                       "in mangle, dropping packet\n");
 580                                 return -ENOMEM;
 581                         }
 582                         if (e->skb->sk)
 583                                 skb_set_owner_w(newskb, e->skb->sk);
 584                         kfree_skb(e->skb);
 585                         e->skb = newskb;
 586                 }
 587                 skb_put(e->skb, diff);
 588         }
 589         if (!skb_make_writable(&e->skb, data_len))
 590                 return -ENOMEM;
 591         memcpy(e->skb->data, data, data_len);
 592
 593         return 0;
 594 }
 595
 596 static inline int
 597 id_cmp(struct nfqnl_queue_entry *e, unsigned long id)
 598 {
 599         return (id == e->id);
 600 }
 601
 602 static int
 603 nfqnl_set_mode(struct nfqnl_instance *queue,
 604                unsigned char mode, unsigned int range)
 605 {
 606         int status;
 607
 608         spin_lock_bh(&queue->lock);
 609         status = __nfqnl_set_mode(queue, mode, range);
 610         spin_unlock_bh(&queue->lock);
 611
 612         return status;
 613 }
 614
 615 static int
 616 dev_cmp(struct nfqnl_queue_entry *entry, unsigned long ifindex)
 617 {
 618         if (entry->info->indev)
 619                 if (entry->info->indev->ifindex == ifindex)
 620                         return 1;
 621
 622         if (entry->info->outdev)
 623                 if (entry->info->outdev->ifindex == ifindex)
 624                         return 1;
 625
 626         return 0;
 627 }
 628
 629 /* drop all packets with either indev or outdev == ifindex from all queue
 630  * instances */
 631 static void
 632 nfqnl_dev_drop(int ifindex)
 633 {
 634         int i;
 635
 636         QDEBUG("entering for ifindex %u\n", ifindex);
 637
 638         /* this only looks like we have to hold the readlock for a way too long
 639          * time, issue_verdict(),  nf_reinject(), ... - but we always only
 640          * issue NF_DROP, which is processed directly in nf_reinject() */
 641         read_lock_bh(&instances_lock);
 642
 643         for  (i = 0; i < INSTANCE_BUCKETS; i++) {
 644                 struct hlist_node *tmp;
 645                 struct nfqnl_instance *inst;
 646                 struct hlist_head *head = &instance_table[i];
 647
 648                 hlist_for_each_entry(inst, tmp, head, hlist) {
 649                         struct nfqnl_queue_entry *entry;
 650                         while ((entry = find_dequeue_entry(inst, dev_cmp,
 651                                                            ifindex)) != NULL)
 652                                 issue_verdict(entry, NF_DROP);
 653                 }
 654         }
 655
 656         read_unlock_bh(&instances_lock);
 657 }
 658
 659 #define RCV_SKB_FAIL(err) do { netlink_ack(skb, nlh, (err)); return; } while (0)
 660
 661 static int
 662 nfqnl_rcv_dev_event(struct notifier_block *this,
 663                     unsigned long event, void *ptr)
 664 {
 665         struct net_device *dev = ptr;
 666
 667         /* Drop any packets associated with the downed device */
 668         if (event == NETDEV_DOWN)
 669                 nfqnl_dev_drop(dev->ifindex);
 670         return NOTIFY_DONE;
 671 }
 672
 673 static struct notifier_block nfqnl_dev_notifier = {
 674         .notifier_call  = nfqnl_rcv_dev_event,
 675 };
 676
 677 static int
 678 nfqnl_rcv_nl_event(struct notifier_block *this,
 679                    unsigned long event, void *ptr)
 680 {
 681         struct netlink_notify *n = ptr;
 682
 683         if (event == NETLINK_URELEASE &&
 684             n->protocol == NETLINK_NETFILTER && n->pid) {
 685                 int i;
 686
 687                 /* destroy all instances for this pid */
 688                 write_lock_bh(&instances_lock);
 689                 for  (i = 0; i < INSTANCE_BUCKETS; i++) {
 690                         struct hlist_node *tmp, *t2;
 691                         struct nfqnl_instance *inst;
 692                         struct hlist_head *head = &instance_table[i];
 693
 694                         hlist_for_each_entry_safe(inst, tmp, t2, head, hlist) {
 695                                 if (n->pid == inst->peer_pid)
 696                                         __instance_destroy(inst);
 697                         }
 698                 }
 699                 write_unlock_bh(&instances_lock);
 700         }
 701         return NOTIFY_DONE;
 702 }
 703
 704 static struct notifier_block nfqnl_rtnl_notifier = {
 705         .notifier_call  = nfqnl_rcv_nl_event,
 706 };
 707
 708 static const int nfqa_verdict_min[NFQA_MAX] = {
 709         [NFQA_VERDICT_HDR-1]    = sizeof(struct nfqnl_msg_verdict_hdr),
 710         [NFQA_MARK-1]           = sizeof(u_int32_t),
 711         [NFQA_PAYLOAD-1]        = 0,
 712 };
 713
 714 static int
 715 nfqnl_recv_verdict(struct sock *ctnl, struct sk_buff *skb,
 716                    struct nlmsghdr *nlh, struct nfattr *nfqa[], int *errp)
 717 {
 718         struct nfgenmsg *nfmsg = NLMSG_DATA(nlh);
 719         u_int16_t queue_num = ntohs(nfmsg->res_id);
 720
 721         struct nfqnl_msg_verdict_hdr *vhdr;
 722         struct nfqnl_instance *queue;
 723         unsigned int verdict;
 724         struct nfqnl_queue_entry *entry;
 725         int err;
 726
 727         if (nfattr_bad_size(nfqa, NFQA_MAX, nfqa_verdict_min)) {
 728                 QDEBUG("bad attribute size\n");
 729                 return -EINVAL;
 730         }
 731
 732         queue = instance_lookup_get(queue_num);
 733         if (!queue)
 734                 return -ENODEV;
 735
 736         if (queue->peer_pid != NETLINK_CB(skb).pid) {
 737                 err = -EPERM;
 738                 goto err_out_put;
 739         }
 740
 741         if (!nfqa[NFQA_VERDICT_HDR-1]) {
 742                 err = -EINVAL;
 743                 goto err_out_put;
 744         }
 745
 746         vhdr = NFA_DATA(nfqa[NFQA_VERDICT_HDR-1]);
 747         verdict = ntohl(vhdr->verdict);
 748
 749         if ((verdict & NF_VERDICT_MASK) > NF_MAX_VERDICT) {
 750                 err = -EINVAL;
 751                 goto err_out_put;
 752         }
 753
 754         entry = find_dequeue_entry(queue, id_cmp, ntohl(vhdr->id));
 755         if (entry == NULL) {
 756                 err = -ENOENT;
 757                 goto err_out_put;
 758         }
 759
 760         if (nfqa[NFQA_PAYLOAD-1]) {
 761                 if (nfqnl_mangle(NFA_DATA(nfqa[NFQA_PAYLOAD-1]),
 762                                  NFA_PAYLOAD(nfqa[NFQA_PAYLOAD-1]), entry) < 0)
 763                         verdict = NF_DROP;
 764         }
 765
 766         if (nfqa[NFQA_MARK-1])
 767                 skb->nfmark = ntohl(*(u_int32_t *)NFA_DATA(nfqa[NFQA_MARK-1]));
 768
 769         issue_verdict(entry, verdict);
 770         instance_put(queue);
 771         return 0;
 772
 773 err_out_put:
 774         instance_put(queue);
 775         return err;
 776 }
 777
 778 static int
 779 nfqnl_recv_unsupp(struct sock *ctnl, struct sk_buff *skb,
 780                   struct nlmsghdr *nlh, struct nfattr *nfqa[], int *errp)
 781 {
 782         return -ENOTSUPP;
 783 }
 784
 785 static const int nfqa_cfg_min[NFQA_CFG_MAX] = {
 786         [NFQA_CFG_CMD-1]        = sizeof(struct nfqnl_msg_config_cmd),
 787         [NFQA_CFG_PARAMS-1]     = sizeof(struct nfqnl_msg_config_params),
 788 };
 789
 790 static int
 791 nfqnl_recv_config(struct sock *ctnl, struct sk_buff *skb,
 792                   struct nlmsghdr *nlh, struct nfattr *nfqa[], int *errp)
 793 {
 794         struct nfgenmsg *nfmsg = NLMSG_DATA(nlh);
 795         u_int16_t queue_num = ntohs(nfmsg->res_id);
 796         struct nfqnl_instance *queue;
 797         int ret = 0;
 798
 799         QDEBUG("entering for msg %u\n", NFNL_MSG_TYPE(nlh->nlmsg_type));
 800
 801         if (nfattr_bad_size(nfqa, NFQA_CFG_MAX, nfqa_cfg_min)) {
 802                 QDEBUG("bad attribute size\n");
 803                 return -EINVAL;
 804         }
 805
 806         queue = instance_lookup_get(queue_num);
 807         if (nfqa[NFQA_CFG_CMD-1]) {
 808                 struct nfqnl_msg_config_cmd *cmd;
 809                 cmd = NFA_DATA(nfqa[NFQA_CFG_CMD-1]);
 810                 QDEBUG("found CFG_CMD\n");
 811
 812                 switch (cmd->command) {
 813                 case NFQNL_CFG_CMD_BIND:
 814                         if (queue)
 815                                 return -EBUSY;
 816
 817                         queue = instance_create(queue_num, NETLINK_CB(skb).pid);
 818                         if (!queue)
 819                                 return -EINVAL;
 820                         break;
 821                 case NFQNL_CFG_CMD_UNBIND:
 822                         if (!queue)
 823                                 return -ENODEV;
 824
 825                         if (queue->peer_pid != NETLINK_CB(skb).pid) {
 826                                 ret = -EPERM;
 827                                 goto out_put;
 828                         }
 829
 830                         instance_destroy(queue);
 831                         break;
 832                 case NFQNL_CFG_CMD_PF_BIND:
 833                         QDEBUG("registering queue handler for pf=%u\n",
 834                                 ntohs(cmd->pf));
 835                         ret = nf_register_queue_handler(ntohs(cmd->pf),
 836                                                         nfqnl_enqueue_packet,
 837                                                         NULL);
 838
 839                         break;
 840                 case NFQNL_CFG_CMD_PF_UNBIND:
 841                         QDEBUG("unregistering queue handler for pf=%u\n",
 842                                 ntohs(cmd->pf));
 843                         /* This is a bug and a feature.  We can unregister
 844                          * other handlers(!) */
 845                         ret = nf_unregister_queue_handler(ntohs(cmd->pf));
 846                         break;
 847                 default:
 848                         ret = -EINVAL;
 849                         break;
 850                 }
 851         } else {
 852                 if (!queue) {
 853                         QDEBUG("no config command, and no instance ENOENT\n");
 854                         ret = -ENOENT;
 855                         goto out_put;
 856                 }
 857
 858                 if (queue->peer_pid != NETLINK_CB(skb).pid) {
 859                         QDEBUG("no config command, and wrong pid\n");
 860                         ret = -EPERM;
 861                         goto out_put;
 862                 }
 863         }
 864
 865         if (nfqa[NFQA_CFG_PARAMS-1]) {
 866                 struct nfqnl_msg_config_params *params;
 867                 params = NFA_DATA(nfqa[NFQA_CFG_PARAMS-1]);
 868
 869                 nfqnl_set_mode(queue, params->copy_mode,
 870                                 ntohl(params->copy_range));
 871         }
 872
 873 out_put:
 874         instance_put(queue);
 875         return ret;
 876 }
 877
 878 static struct nfnl_callback nfqnl_cb[NFQNL_MSG_MAX] = {
 879         [NFQNL_MSG_PACKET]      = { .call = nfqnl_recv_unsupp,
 880                                     .cap_required = CAP_NET_ADMIN },
 881         [NFQNL_MSG_VERDICT]     = { .call = nfqnl_recv_verdict,
 882                                     .cap_required = CAP_NET_ADMIN },
 883         [NFQNL_MSG_CONFIG]      = { .call = nfqnl_recv_config,
 884                                     .cap_required = CAP_NET_ADMIN },
 885 };
 886
 887 static struct nfnetlink_subsystem nfqnl_subsys = {
 888         .name           = "nf_queue",
 889         .subsys_id      = NFNL_SUBSYS_QUEUE,
 890         .cb_count       = NFQNL_MSG_MAX,
 891         .attr_count     = NFQA_MAX,
 892         .cb             = nfqnl_cb,
 893 };
 894
 895 #ifdef CONFIG_PROC_FS
 896 struct iter_state {
 897         unsigned int bucket;
 898 };
 899
 900 static struct hlist_node *get_first(struct seq_file *seq)
 901 {
 902         struct iter_state *st = seq->private;
 903
 904         if (!st)
 905                 return NULL;
 906
 907         for (st->bucket = 0; st->bucket < INSTANCE_BUCKETS; st->bucket++) {
 908                 if (!hlist_empty(&instance_table[st->bucket]))
 909                         return instance_table[st->bucket].first;
 910         }
 911         return NULL;
 912 }
 913
 914 static struct hlist_node *get_next(struct seq_file *seq, struct hlist_node *h)
 915 {
 916         struct iter_state *st = seq->private;
 917
 918         h = h->next;
 919         while (!h) {
 920                 if (++st->bucket >= INSTANCE_BUCKETS)
 921                         return NULL;
 922
 923                 h = instance_table[st->bucket].first;
 924         }
 925         return h;
 926 }
 927
 928 static struct hlist_node *get_idx(struct seq_file *seq, loff_t pos)
 929 {
 930         struct hlist_node *head;
 931         head = get_first(seq);
 932
 933         if (head)
 934                 while (pos && (head = get_next(seq, head)))
 935                         pos--;
 936         return pos ? NULL : head;
 937 }
 938
 939 static void *seq_start(struct seq_file *seq, loff_t *pos)
 940 {
 941         read_lock_bh(&instances_lock);
 942         return get_idx(seq, *pos);
 943 }
 944
 945 static void *seq_next(struct seq_file *s, void *v, loff_t *pos)
 946 {
 947         (*pos)++;
 948         return get_next(s, v);
 949 }
 950
 951 static void seq_stop(struct seq_file *s, void *v)
 952 {
 953         read_unlock_bh(&instances_lock);
 954 }
 955
 956 static int seq_show(struct seq_file *s, void *v)
 957 {
 958         const struct nfqnl_instance *inst = v;
 959
 960         return seq_printf(s, "%5d %6d %5d %1d %5d %5d %5d %8d %2d\n",
 961                           inst->queue_num,
 962                           inst->peer_pid, inst->queue_total,
 963                           inst->copy_mode, inst->copy_range,
 964                           inst->queue_dropped, inst->queue_user_dropped,
 965                           atomic_read(&inst->id_sequence),
 966                           atomic_read(&inst->use));
 967 }
 968
 969 static struct seq_operations nfqnl_seq_ops = {
 970         .start  = seq_start,
 971         .next   = seq_next,
 972         .stop   = seq_stop,
 973         .show   = seq_show,
 974 };
 975
 976 static int nfqnl_open(struct inode *inode, struct file *file)
 977 {
 978         struct seq_file *seq;
 979         struct iter_state *is;
 980         int ret;
 981
 982         is = kmalloc(sizeof(*is), GFP_KERNEL);
 983         if (!is)
 984                 return -ENOMEM;
 985         memset(is, 0, sizeof(*is));
 986         ret = seq_open(file, &nfqnl_seq_ops);
 987         if (ret < 0)
 988                 goto out_free;
 989         seq = file->private_data;
 990         seq->private = is;
 991         return ret;
 992 out_free:
 993         kfree(is);
 994         return ret;
 995 }
 996
 997 static struct file_operations nfqnl_file_ops = {
 998         .owner   = THIS_MODULE,
 999         .open    = nfqnl_open,
1000         .read    = seq_read,
1001         .llseek  = seq_lseek,
1002         .release = seq_release_private,
1003 };
1004
1005 #endif /* PROC_FS */
1006
1007 static int
1008 init_or_cleanup(int init)
1009 {
1010         int i, status = -ENOMEM;
1011 #ifdef CONFIG_PROC_FS
1012         struct proc_dir_entry *proc_nfqueue;
1013 #endif
1014
1015         if (!init)
1016                 goto cleanup;
1017
1018         for (i = 0; i < INSTANCE_BUCKETS; i++)
1019                 INIT_HLIST_HEAD(&instance_table[i]);
1020
1021         netlink_register_notifier(&nfqnl_rtnl_notifier);
1022         status = nfnetlink_subsys_register(&nfqnl_subsys);
1023         if (status < 0) {
1024                 printk(KERN_ERR "nf_queue: failed to create netlink socket\n");
1025                 goto cleanup_netlink_notifier;
1026         }
1027
1028 #ifdef CONFIG_PROC_FS
1029         proc_nfqueue = create_proc_entry("nfnetlink_queue", 0440,
1030                                          proc_net_netfilter);
1031         if (!proc_nfqueue)
1032                 goto cleanup_subsys;
1033         proc_nfqueue->proc_fops = &nfqnl_file_ops;
1034 #endif
1035
1036         register_netdevice_notifier(&nfqnl_dev_notifier);
1037
1038         return status;
1039
1040 cleanup:
1041         nf_unregister_queue_handlers(nfqnl_enqueue_packet);
1042         unregister_netdevice_notifier(&nfqnl_dev_notifier);
1043 #ifdef CONFIG_PROC_FS
1044         remove_proc_entry("nfnetlink_queue", proc_net_netfilter);
1045 cleanup_subsys:
1046 #endif
1047         nfnetlink_subsys_unregister(&nfqnl_subsys);
1048 cleanup_netlink_notifier:
1049         netlink_unregister_notifier(&nfqnl_rtnl_notifier);
1050         return status;
1051 }
1052
1053 static int __init init(void)
1054 {
1055
1056         return init_or_cleanup(1);
1057 }
1058
1059 static void __exit fini(void)
1060 {
1061         init_or_cleanup(0);
1062 }
1063
1064 MODULE_DESCRIPTION("netfilter packet queue handler");
1065 MODULE_AUTHOR("Harald Welte <laforge@netfilter.org>");
1066 MODULE_LICENSE("GPL");
1067 MODULE_ALIAS_NFNL_SUBSYS(NFNL_SUBSYS_QUEUE);
1068
1069 module_init(init);
1070 module_exit(fini);