SAFE public projects git trees. - safe/jmp/linux-2.6/blob - 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 #include <net/netfilter/nf_queue.h>
  31
  32 #include <asm/atomic.h>
  33
  34 #ifdef CONFIG_BRIDGE_NETFILTER
  35 #include "../bridge/br_private.h"
  36 #endif
  37
  38 #define NFQNL_QMAX_DEFAULT 1024
  39
  40 #if 0
  41 #define QDEBUG(x, args ...)     printk(KERN_DEBUG "%s(%d):%s(): " x,       \
  42                                         __FILE__, __LINE__, __FUNCTION__,  \
  43                                         ## args)
  44 #else
  45 #define QDEBUG(x, ...)
  46 #endif
  47
  48 struct nfqnl_instance {
  49         struct hlist_node hlist;                /* global list of queues */
  50         atomic_t use;
  51
  52         int peer_pid;
  53         unsigned int queue_maxlen;
  54         unsigned int copy_range;
  55         unsigned int queue_total;
  56         unsigned int queue_dropped;
  57         unsigned int queue_user_dropped;
  58
  59         atomic_t id_sequence;                   /* 'sequence' of pkt ids */
  60
  61         u_int16_t queue_num;                    /* number of this queue */
  62         u_int8_t copy_mode;
  63
  64         spinlock_t lock;
  65
  66         struct list_head queue_list;            /* packets in queue */
  67 };
  68
  69 typedef int (*nfqnl_cmpfn)(struct nf_queue_entry *, unsigned long);
  70
  71 static DEFINE_RWLOCK(instances_lock);
  72
  73 #define INSTANCE_BUCKETS        16
  74 static struct hlist_head instance_table[INSTANCE_BUCKETS];
  75
  76 static inline u_int8_t instance_hashfn(u_int16_t queue_num)
  77 {
  78         return ((queue_num >> 8) | queue_num) % INSTANCE_BUCKETS;
  79 }
  80
  81 static struct nfqnl_instance *
  82 __instance_lookup(u_int16_t queue_num)
  83 {
  84         struct hlist_head *head;
  85         struct hlist_node *pos;
  86         struct nfqnl_instance *inst;
  87
  88         head = &instance_table[instance_hashfn(queue_num)];
  89         hlist_for_each_entry(inst, pos, head, hlist) {
  90                 if (inst->queue_num == queue_num)
  91                         return inst;
  92         }
  93         return NULL;
  94 }
  95
  96 static struct nfqnl_instance *
  97 instance_lookup_get(u_int16_t queue_num)
  98 {
  99         struct nfqnl_instance *inst;
 100
 101         read_lock_bh(&instances_lock);
 102         inst = __instance_lookup(queue_num);
 103         if (inst)
 104                 atomic_inc(&inst->use);
 105         read_unlock_bh(&instances_lock);
 106
 107         return inst;
 108 }
 109
 110 static void
 111 instance_put(struct nfqnl_instance *inst)
 112 {
 113         if (inst && atomic_dec_and_test(&inst->use)) {
 114                 QDEBUG("kfree(inst=%p)\n", inst);
 115                 kfree(inst);
 116         }
 117 }
 118
 119 static struct nfqnl_instance *
 120 instance_create(u_int16_t queue_num, int pid)
 121 {
 122         struct nfqnl_instance *inst;
 123
 124         QDEBUG("entering for queue_num=%u, pid=%d\n", queue_num, pid);
 125
 126         write_lock_bh(&instances_lock);
 127         if (__instance_lookup(queue_num)) {
 128                 inst = NULL;
 129                 QDEBUG("aborting, instance already exists\n");
 130                 goto out_unlock;
 131         }
 132
 133         inst = kzalloc(sizeof(*inst), GFP_ATOMIC);
 134         if (!inst)
 135                 goto out_unlock;
 136
 137         inst->queue_num = queue_num;
 138         inst->peer_pid = pid;
 139         inst->queue_maxlen = NFQNL_QMAX_DEFAULT;
 140         inst->copy_range = 0xfffff;
 141         inst->copy_mode = NFQNL_COPY_NONE;
 142         atomic_set(&inst->id_sequence, 0);
 143         /* needs to be two, since we _put() after creation */
 144         atomic_set(&inst->use, 2);
 145         spin_lock_init(&inst->lock);
 146         INIT_LIST_HEAD(&inst->queue_list);
 147
 148         if (!try_module_get(THIS_MODULE))
 149                 goto out_free;
 150
 151         hlist_add_head(&inst->hlist,
 152                        &instance_table[instance_hashfn(queue_num)]);
 153
 154         write_unlock_bh(&instances_lock);
 155
 156         QDEBUG("successfully created new instance\n");
 157
 158         return inst;
 159
 160 out_free:
 161         kfree(inst);
 162 out_unlock:
 163         write_unlock_bh(&instances_lock);
 164         return NULL;
 165 }
 166
 167 static void nfqnl_flush(struct nfqnl_instance *queue, nfqnl_cmpfn cmpfn,
 168                         unsigned long data);
 169
 170 static void
 171 _instance_destroy2(struct nfqnl_instance *inst, int lock)
 172 {
 173         /* first pull it out of the global list */
 174         if (lock)
 175                 write_lock_bh(&instances_lock);
 176
 177         QDEBUG("removing instance %p (queuenum=%u) from hash\n",
 178                 inst, inst->queue_num);
 179         hlist_del(&inst->hlist);
 180
 181         if (lock)
 182                 write_unlock_bh(&instances_lock);
 183
 184         /* then flush all pending skbs from the queue */
 185         nfqnl_flush(inst, NULL, 0);
 186
 187         /* and finally put the refcount */
 188         instance_put(inst);
 189
 190         module_put(THIS_MODULE);
 191 }
 192
 193 static inline void
 194 __instance_destroy(struct nfqnl_instance *inst)
 195 {
 196         _instance_destroy2(inst, 0);
 197 }
 198
 199 static inline void
 200 instance_destroy(struct nfqnl_instance *inst)
 201 {
 202         _instance_destroy2(inst, 1);
 203 }
 204
 205 static inline void
 206 __enqueue_entry(struct nfqnl_instance *queue, struct nf_queue_entry *entry)
 207 {
 208        list_add_tail(&entry->list, &queue->queue_list);
 209        queue->queue_total++;
 210 }
 211
 212 static inline int
 213 __nfqnl_set_mode(struct nfqnl_instance *queue,
 214                  unsigned char mode, unsigned int range)
 215 {
 216         int status = 0;
 217
 218         switch (mode) {
 219         case NFQNL_COPY_NONE:
 220         case NFQNL_COPY_META:
 221                 queue->copy_mode = mode;
 222                 queue->copy_range = 0;
 223                 break;
 224
 225         case NFQNL_COPY_PACKET:
 226                 queue->copy_mode = mode;
 227                 /* we're using struct nlattr which has 16bit nla_len */
 228                 if (range > 0xffff)
 229                         queue->copy_range = 0xffff;
 230                 else
 231                         queue->copy_range = range;
 232                 break;
 233
 234         default:
 235                 status = -EINVAL;
 236
 237         }
 238         return status;
 239 }
 240
 241 static struct nf_queue_entry *
 242 find_dequeue_entry(struct nfqnl_instance *queue, unsigned int id)
 243 {
 244         struct nf_queue_entry *entry = NULL, *i;
 245
 246         spin_lock_bh(&queue->lock);
 247
 248         list_for_each_entry(i, &queue->queue_list, list) {
 249                 if (i->id == id) {
 250                         entry = i;
 251                         break;
 252                 }
 253         }
 254
 255         if (entry) {
 256                 list_del(&entry->list);
 257                 queue->queue_total--;
 258         }
 259
 260         spin_unlock_bh(&queue->lock);
 261
 262         return entry;
 263 }
 264
 265 static void
 266 nfqnl_flush(struct nfqnl_instance *queue, nfqnl_cmpfn cmpfn, unsigned long data)
 267 {
 268         struct nf_queue_entry *entry, *next;
 269
 270         spin_lock_bh(&queue->lock);
 271         list_for_each_entry_safe(entry, next, &queue->queue_list, list) {
 272                 if (!cmpfn || cmpfn(entry, data)) {
 273                         list_del(&entry->list);
 274                         queue->queue_total--;
 275                         nf_reinject(entry, NF_DROP);
 276                 }
 277         }
 278         spin_unlock_bh(&queue->lock);
 279 }
 280
 281 static struct sk_buff *
 282 nfqnl_build_packet_message(struct nfqnl_instance *queue,
 283                            struct nf_queue_entry *entry, int *errp)
 284 {
 285         sk_buff_data_t old_tail;
 286         size_t size;
 287         size_t data_len = 0;
 288         struct sk_buff *skb;
 289         struct nfqnl_msg_packet_hdr pmsg;
 290         struct nlmsghdr *nlh;
 291         struct nfgenmsg *nfmsg;
 292         struct sk_buff *entskb = entry->skb;
 293         struct net_device *indev;
 294         struct net_device *outdev;
 295         __be32 tmp_uint;
 296
 297         QDEBUG("entered\n");
 298
 299         size =    NLMSG_ALIGN(sizeof(struct nfgenmsg))
 300                 + nla_total_size(sizeof(struct nfqnl_msg_packet_hdr))
 301                 + nla_total_size(sizeof(u_int32_t))     /* ifindex */
 302                 + nla_total_size(sizeof(u_int32_t))     /* ifindex */
 303 #ifdef CONFIG_BRIDGE_NETFILTER
 304                 + nla_total_size(sizeof(u_int32_t))     /* ifindex */
 305                 + nla_total_size(sizeof(u_int32_t))     /* ifindex */
 306 #endif
 307                 + nla_total_size(sizeof(u_int32_t))     /* mark */
 308                 + nla_total_size(sizeof(struct nfqnl_msg_packet_hw))
 309                 + nla_total_size(sizeof(struct nfqnl_msg_packet_timestamp));
 310
 311         outdev = entry->outdev;
 312
 313         spin_lock_bh(&queue->lock);
 314
 315         switch (queue->copy_mode) {
 316         case NFQNL_COPY_META:
 317         case NFQNL_COPY_NONE:
 318                 data_len = 0;
 319                 break;
 320
 321         case NFQNL_COPY_PACKET:
 322                 if ((entskb->ip_summed == CHECKSUM_PARTIAL ||
 323                      entskb->ip_summed == CHECKSUM_COMPLETE) &&
 324                     (*errp = skb_checksum_help(entskb))) {
 325                         spin_unlock_bh(&queue->lock);
 326                         return NULL;
 327                 }
 328                 if (queue->copy_range == 0
 329                     || queue->copy_range > entskb->len)
 330                         data_len = entskb->len;
 331                 else
 332                         data_len = queue->copy_range;
 333
 334                 size += nla_total_size(data_len);
 335                 break;
 336
 337         default:
 338                 *errp = -EINVAL;
 339                 spin_unlock_bh(&queue->lock);
 340                 return NULL;
 341         }
 342
 343         spin_unlock_bh(&queue->lock);
 344
 345         skb = alloc_skb(size, GFP_ATOMIC);
 346         if (!skb)
 347                 goto nlmsg_failure;
 348
 349         old_tail = skb->tail;
 350         nlh = NLMSG_PUT(skb, 0, 0,
 351                         NFNL_SUBSYS_QUEUE << 8 | NFQNL_MSG_PACKET,
 352                         sizeof(struct nfgenmsg));
 353         nfmsg = NLMSG_DATA(nlh);
 354         nfmsg->nfgen_family = entry->pf;
 355         nfmsg->version = NFNETLINK_V0;
 356         nfmsg->res_id = htons(queue->queue_num);
 357
 358         pmsg.packet_id          = htonl(entry->id);
 359         pmsg.hw_protocol        = entskb->protocol;
 360         pmsg.hook               = entry->hook;
 361
 362         NLA_PUT(skb, NFQA_PACKET_HDR, sizeof(pmsg), &pmsg);
 363
 364         indev = entry->indev;
 365         if (indev) {
 366                 tmp_uint = htonl(indev->ifindex);
 367 #ifndef CONFIG_BRIDGE_NETFILTER
 368                 NLA_PUT(skb, NFQA_IFINDEX_INDEV, sizeof(tmp_uint), &tmp_uint);
 369 #else
 370                 if (entry->pf == PF_BRIDGE) {
 371                         /* Case 1: indev is physical input device, we need to
 372                          * look for bridge group (when called from
 373                          * netfilter_bridge) */
 374                         NLA_PUT(skb, NFQA_IFINDEX_PHYSINDEV, sizeof(tmp_uint),
 375                                 &tmp_uint);
 376                         /* this is the bridge group "brX" */
 377                         tmp_uint = htonl(indev->br_port->br->dev->ifindex);
 378                         NLA_PUT(skb, NFQA_IFINDEX_INDEV, sizeof(tmp_uint),
 379                                 &tmp_uint);
 380                 } else {
 381                         /* Case 2: indev is bridge group, we need to look for
 382                          * physical device (when called from ipv4) */
 383                         NLA_PUT(skb, NFQA_IFINDEX_INDEV, sizeof(tmp_uint),
 384                                 &tmp_uint);
 385                         if (entskb->nf_bridge
 386                             && entskb->nf_bridge->physindev) {
 387                                 tmp_uint = htonl(entskb->nf_bridge->physindev->ifindex);
 388                                 NLA_PUT(skb, NFQA_IFINDEX_PHYSINDEV,
 389                                         sizeof(tmp_uint), &tmp_uint);
 390                         }
 391                 }
 392 #endif
 393         }
 394
 395         if (outdev) {
 396                 tmp_uint = htonl(outdev->ifindex);
 397 #ifndef CONFIG_BRIDGE_NETFILTER
 398                 NLA_PUT(skb, NFQA_IFINDEX_OUTDEV, sizeof(tmp_uint), &tmp_uint);
 399 #else
 400                 if (entry->pf == PF_BRIDGE) {
 401                         /* Case 1: outdev is physical output device, we need to
 402                          * look for bridge group (when called from
 403                          * netfilter_bridge) */
 404                         NLA_PUT(skb, NFQA_IFINDEX_PHYSOUTDEV, sizeof(tmp_uint),
 405                                 &tmp_uint);
 406                         /* this is the bridge group "brX" */
 407                         tmp_uint = htonl(outdev->br_port->br->dev->ifindex);
 408                         NLA_PUT(skb, NFQA_IFINDEX_OUTDEV, sizeof(tmp_uint),
 409                                 &tmp_uint);
 410                 } else {
 411                         /* Case 2: outdev is bridge group, we need to look for
 412                          * physical output device (when called from ipv4) */
 413                         NLA_PUT(skb, NFQA_IFINDEX_OUTDEV, sizeof(tmp_uint),
 414                                 &tmp_uint);
 415                         if (entskb->nf_bridge
 416                             && entskb->nf_bridge->physoutdev) {
 417                                 tmp_uint = htonl(entskb->nf_bridge->physoutdev->ifindex);
 418                                 NLA_PUT(skb, NFQA_IFINDEX_PHYSOUTDEV,
 419                                         sizeof(tmp_uint), &tmp_uint);
 420                         }
 421                 }
 422 #endif
 423         }
 424
 425         if (entskb->mark) {
 426                 tmp_uint = htonl(entskb->mark);
 427                 NLA_PUT(skb, NFQA_MARK, sizeof(u_int32_t), &tmp_uint);
 428         }
 429
 430         if (indev && entskb->dev) {
 431                 struct nfqnl_msg_packet_hw phw;
 432                 int len = dev_parse_header(entskb, phw.hw_addr);
 433                 if (len) {
 434                         phw.hw_addrlen = htons(len);
 435                         NLA_PUT(skb, NFQA_HWADDR, sizeof(phw), &phw);
 436                 }
 437         }
 438
 439         if (entskb->tstamp.tv64) {
 440                 struct nfqnl_msg_packet_timestamp ts;
 441                 struct timeval tv = ktime_to_timeval(entskb->tstamp);
 442                 ts.sec = cpu_to_be64(tv.tv_sec);
 443                 ts.usec = cpu_to_be64(tv.tv_usec);
 444
 445                 NLA_PUT(skb, NFQA_TIMESTAMP, sizeof(ts), &ts);
 446         }
 447
 448         if (data_len) {
 449                 struct nlattr *nla;
 450                 int size = nla_attr_size(data_len);
 451
 452                 if (skb_tailroom(skb) < nla_total_size(data_len)) {
 453                         printk(KERN_WARNING "nf_queue: no tailroom!\n");
 454                         goto nlmsg_failure;
 455                 }
 456
 457                 nla = (struct nlattr *)skb_put(skb, nla_total_size(data_len));
 458                 nla->nla_type = NFQA_PAYLOAD;
 459                 nla->nla_len = size;
 460
 461                 if (skb_copy_bits(entskb, 0, nla_data(nla), data_len))
 462                         BUG();
 463         }
 464
 465         nlh->nlmsg_len = skb->tail - old_tail;
 466         return skb;
 467
 468 nlmsg_failure:
 469 nla_put_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 nf_queue_entry *entry, unsigned int queuenum)
 480 {
 481         int status = -EINVAL;
 482         struct sk_buff *nskb;
 483         struct nfqnl_instance *queue;
 484
 485         QDEBUG("entered\n");
 486
 487         queue = instance_lookup_get(queuenum);
 488         if (!queue) {
 489                 QDEBUG("no queue instance matching\n");
 490                 return -EINVAL;
 491         }
 492
 493         if (queue->copy_mode == NFQNL_COPY_NONE) {
 494                 QDEBUG("mode COPY_NONE, aborting\n");
 495                 status = -EAGAIN;
 496                 goto err_out_put;
 497         }
 498
 499         entry->id = atomic_inc_return(&queue->id_sequence);
 500
 501         nskb = nfqnl_build_packet_message(queue, entry, &status);
 502         if (nskb == NULL)
 503                 goto err_out_put;
 504
 505         spin_lock_bh(&queue->lock);
 506
 507         if (!queue->peer_pid)
 508                 goto err_out_free_nskb;
 509
 510         if (queue->queue_total >= queue->queue_maxlen) {
 511                 queue->queue_dropped++;
 512                 status = -ENOSPC;
 513                 if (net_ratelimit())
 514                           printk(KERN_WARNING "nf_queue: full at %d entries, "
 515                                  "dropping packets(s). Dropped: %d\n",
 516                                  queue->queue_total, queue->queue_dropped);
 517                 goto err_out_free_nskb;
 518         }
 519
 520         /* nfnetlink_unicast will either free the nskb or add it to a socket */
 521         status = nfnetlink_unicast(nskb, queue->peer_pid, MSG_DONTWAIT);
 522         if (status < 0) {
 523                 queue->queue_user_dropped++;
 524                 goto err_out_unlock;
 525         }
 526
 527         __enqueue_entry(queue, entry);
 528
 529         spin_unlock_bh(&queue->lock);
 530         instance_put(queue);
 531         return status;
 532
 533 err_out_free_nskb:
 534         kfree_skb(nskb);
 535
 536 err_out_unlock:
 537         spin_unlock_bh(&queue->lock);
 538
 539 err_out_put:
 540         instance_put(queue);
 541         return status;
 542 }
 543
 544 static int
 545 nfqnl_mangle(void *data, int data_len, struct nf_queue_entry *e)
 546 {
 547         int diff;
 548         int err;
 549
 550         diff = data_len - e->skb->len;
 551         if (diff < 0) {
 552                 if (pskb_trim(e->skb, data_len))
 553                         return -ENOMEM;
 554         } else if (diff > 0) {
 555                 if (data_len > 0xFFFF)
 556                         return -EINVAL;
 557                 if (diff > skb_tailroom(e->skb)) {
 558                         err = pskb_expand_head(e->skb, 0,
 559                                                diff - skb_tailroom(e->skb),
 560                                                GFP_ATOMIC);
 561                         if (err) {
 562                                 printk(KERN_WARNING "nf_queue: OOM "
 563                                       "in mangle, dropping packet\n");
 564                                 return err;
 565                         }
 566                 }
 567                 skb_put(e->skb, diff);
 568         }
 569         if (!skb_make_writable(e->skb, data_len))
 570                 return -ENOMEM;
 571         skb_copy_to_linear_data(e->skb, data, data_len);
 572         e->skb->ip_summed = CHECKSUM_NONE;
 573         return 0;
 574 }
 575
 576 static int
 577 nfqnl_set_mode(struct nfqnl_instance *queue,
 578                unsigned char mode, unsigned int range)
 579 {
 580         int status;
 581
 582         spin_lock_bh(&queue->lock);
 583         status = __nfqnl_set_mode(queue, mode, range);
 584         spin_unlock_bh(&queue->lock);
 585
 586         return status;
 587 }
 588
 589 static int
 590 dev_cmp(struct nf_queue_entry *entry, unsigned long ifindex)
 591 {
 592         if (entry->indev)
 593                 if (entry->indev->ifindex == ifindex)
 594                         return 1;
 595         if (entry->outdev)
 596                 if (entry->outdev->ifindex == ifindex)
 597                         return 1;
 598 #ifdef CONFIG_BRIDGE_NETFILTER
 599         if (entry->skb->nf_bridge) {
 600                 if (entry->skb->nf_bridge->physindev &&
 601                     entry->skb->nf_bridge->physindev->ifindex == ifindex)
 602                         return 1;
 603                 if (entry->skb->nf_bridge->physoutdev &&
 604                     entry->skb->nf_bridge->physoutdev->ifindex == ifindex)
 605                         return 1;
 606         }
 607 #endif
 608         return 0;
 609 }
 610
 611 /* drop all packets with either indev or outdev == ifindex from all queue
 612  * instances */
 613 static void
 614 nfqnl_dev_drop(int ifindex)
 615 {
 616         int i;
 617
 618         QDEBUG("entering for ifindex %u\n", ifindex);
 619
 620         /* this only looks like we have to hold the readlock for a way too long
 621          * time, issue_verdict(),  nf_reinject(), ... - but we always only
 622          * issue NF_DROP, which is processed directly in nf_reinject() */
 623         read_lock_bh(&instances_lock);
 624
 625         for  (i = 0; i < INSTANCE_BUCKETS; i++) {
 626                 struct hlist_node *tmp;
 627                 struct nfqnl_instance *inst;
 628                 struct hlist_head *head = &instance_table[i];
 629
 630                 hlist_for_each_entry(inst, tmp, head, hlist)
 631                         nfqnl_flush(inst, dev_cmp, ifindex);
 632         }
 633
 634         read_unlock_bh(&instances_lock);
 635 }
 636
 637 #define RCV_SKB_FAIL(err) do { netlink_ack(skb, nlh, (err)); return; } while (0)
 638
 639 static int
 640 nfqnl_rcv_dev_event(struct notifier_block *this,
 641                     unsigned long event, void *ptr)
 642 {
 643         struct net_device *dev = ptr;
 644
 645         if (dev->nd_net != &init_net)
 646                 return NOTIFY_DONE;
 647
 648         /* Drop any packets associated with the downed device */
 649         if (event == NETDEV_DOWN)
 650                 nfqnl_dev_drop(dev->ifindex);
 651         return NOTIFY_DONE;
 652 }
 653
 654 static struct notifier_block nfqnl_dev_notifier = {
 655         .notifier_call  = nfqnl_rcv_dev_event,
 656 };
 657
 658 static int
 659 nfqnl_rcv_nl_event(struct notifier_block *this,
 660                    unsigned long event, void *ptr)
 661 {
 662         struct netlink_notify *n = ptr;
 663
 664         if (event == NETLINK_URELEASE &&
 665             n->protocol == NETLINK_NETFILTER && n->pid) {
 666                 int i;
 667
 668                 /* destroy all instances for this pid */
 669                 write_lock_bh(&instances_lock);
 670                 for  (i = 0; i < INSTANCE_BUCKETS; i++) {
 671                         struct hlist_node *tmp, *t2;
 672                         struct nfqnl_instance *inst;
 673                         struct hlist_head *head = &instance_table[i];
 674
 675                         hlist_for_each_entry_safe(inst, tmp, t2, head, hlist) {
 676                                 if ((n->net == &init_net) &&
 677                                     (n->pid == inst->peer_pid))
 678                                         __instance_destroy(inst);
 679                         }
 680                 }
 681                 write_unlock_bh(&instances_lock);
 682         }
 683         return NOTIFY_DONE;
 684 }
 685
 686 static struct notifier_block nfqnl_rtnl_notifier = {
 687         .notifier_call  = nfqnl_rcv_nl_event,
 688 };
 689
 690 static const struct nla_policy nfqa_verdict_policy[NFQA_MAX+1] = {
 691         [NFQA_VERDICT_HDR]      = { .len = sizeof(struct nfqnl_msg_verdict_hdr) },
 692         [NFQA_MARK]             = { .type = NLA_U32 },
 693         [NFQA_PAYLOAD]          = { .type = NLA_UNSPEC },
 694 };
 695
 696 static int
 697 nfqnl_recv_verdict(struct sock *ctnl, struct sk_buff *skb,
 698                    struct nlmsghdr *nlh, struct nlattr *nfqa[])
 699 {
 700         struct nfgenmsg *nfmsg = NLMSG_DATA(nlh);
 701         u_int16_t queue_num = ntohs(nfmsg->res_id);
 702
 703         struct nfqnl_msg_verdict_hdr *vhdr;
 704         struct nfqnl_instance *queue;
 705         unsigned int verdict;
 706         struct nf_queue_entry *entry;
 707         int err;
 708
 709         queue = instance_lookup_get(queue_num);
 710         if (!queue)
 711                 return -ENODEV;
 712
 713         if (queue->peer_pid != NETLINK_CB(skb).pid) {
 714                 err = -EPERM;
 715                 goto err_out_put;
 716         }
 717
 718         if (!nfqa[NFQA_VERDICT_HDR]) {
 719                 err = -EINVAL;
 720                 goto err_out_put;
 721         }
 722
 723         vhdr = nla_data(nfqa[NFQA_VERDICT_HDR]);
 724         verdict = ntohl(vhdr->verdict);
 725
 726         if ((verdict & NF_VERDICT_MASK) > NF_MAX_VERDICT) {
 727                 err = -EINVAL;
 728                 goto err_out_put;
 729         }
 730
 731         entry = find_dequeue_entry(queue, ntohl(vhdr->id));
 732         if (entry == NULL) {
 733                 err = -ENOENT;
 734                 goto err_out_put;
 735         }
 736
 737         if (nfqa[NFQA_PAYLOAD]) {
 738                 if (nfqnl_mangle(nla_data(nfqa[NFQA_PAYLOAD]),
 739                                  nla_len(nfqa[NFQA_PAYLOAD]), entry) < 0)
 740                         verdict = NF_DROP;
 741         }
 742
 743         if (nfqa[NFQA_MARK])
 744                 entry->skb->mark = ntohl(*(__be32 *)
 745                                          nla_data(nfqa[NFQA_MARK]));
 746
 747         nf_reinject(entry, verdict);
 748         instance_put(queue);
 749         return 0;
 750
 751 err_out_put:
 752         instance_put(queue);
 753         return err;
 754 }
 755
 756 static int
 757 nfqnl_recv_unsupp(struct sock *ctnl, struct sk_buff *skb,
 758                   struct nlmsghdr *nlh, struct nlattr *nfqa[])
 759 {
 760         return -ENOTSUPP;
 761 }
 762
 763 static const struct nla_policy nfqa_cfg_policy[NFQA_CFG_MAX+1] = {
 764         [NFQA_CFG_CMD]          = { .len = sizeof(struct nfqnl_msg_config_cmd) },
 765         [NFQA_CFG_PARAMS]       = { .len = sizeof(struct nfqnl_msg_config_params) },
 766 };
 767
 768 static const struct nf_queue_handler nfqh = {
 769         .name   = "nf_queue",
 770         .outfn  = &nfqnl_enqueue_packet,
 771 };
 772
 773 static int
 774 nfqnl_recv_config(struct sock *ctnl, struct sk_buff *skb,
 775                   struct nlmsghdr *nlh, struct nlattr *nfqa[])
 776 {
 777         struct nfgenmsg *nfmsg = NLMSG_DATA(nlh);
 778         u_int16_t queue_num = ntohs(nfmsg->res_id);
 779         struct nfqnl_instance *queue;
 780         int ret = 0;
 781
 782         QDEBUG("entering for msg %u\n", NFNL_MSG_TYPE(nlh->nlmsg_type));
 783
 784         queue = instance_lookup_get(queue_num);
 785         if (nfqa[NFQA_CFG_CMD]) {
 786                 struct nfqnl_msg_config_cmd *cmd;
 787                 cmd = nla_data(nfqa[NFQA_CFG_CMD]);
 788                 QDEBUG("found CFG_CMD\n");
 789
 790                 switch (cmd->command) {
 791                 case NFQNL_CFG_CMD_BIND:
 792                         if (queue)
 793                                 return -EBUSY;
 794
 795                         queue = instance_create(queue_num, NETLINK_CB(skb).pid);
 796                         if (!queue)
 797                                 return -EINVAL;
 798                         break;
 799                 case NFQNL_CFG_CMD_UNBIND:
 800                         if (!queue)
 801                                 return -ENODEV;
 802
 803                         if (queue->peer_pid != NETLINK_CB(skb).pid) {
 804                                 ret = -EPERM;
 805                                 goto out_put;
 806                         }
 807
 808                         instance_destroy(queue);
 809                         break;
 810                 case NFQNL_CFG_CMD_PF_BIND:
 811                         QDEBUG("registering queue handler for pf=%u\n",
 812                                 ntohs(cmd->pf));
 813                         ret = nf_register_queue_handler(ntohs(cmd->pf), &nfqh);
 814                         break;
 815                 case NFQNL_CFG_CMD_PF_UNBIND:
 816                         QDEBUG("unregistering queue handler for pf=%u\n",
 817                                 ntohs(cmd->pf));
 818                         ret = nf_unregister_queue_handler(ntohs(cmd->pf), &nfqh);
 819                         break;
 820                 default:
 821                         ret = -EINVAL;
 822                         break;
 823                 }
 824         } else {
 825                 if (!queue) {
 826                         QDEBUG("no config command, and no instance ENOENT\n");
 827                         ret = -ENOENT;
 828                         goto out_put;
 829                 }
 830
 831                 if (queue->peer_pid != NETLINK_CB(skb).pid) {
 832                         QDEBUG("no config command, and wrong pid\n");
 833                         ret = -EPERM;
 834                         goto out_put;
 835                 }
 836         }
 837
 838         if (nfqa[NFQA_CFG_PARAMS]) {
 839                 struct nfqnl_msg_config_params *params;
 840
 841                 if (!queue) {
 842                         ret = -ENOENT;
 843                         goto out_put;
 844                 }
 845                 params = nla_data(nfqa[NFQA_CFG_PARAMS]);
 846                 nfqnl_set_mode(queue, params->copy_mode,
 847                                 ntohl(params->copy_range));
 848         }
 849
 850         if (nfqa[NFQA_CFG_QUEUE_MAXLEN]) {
 851                 __be32 *queue_maxlen;
 852                 queue_maxlen = nla_data(nfqa[NFQA_CFG_QUEUE_MAXLEN]);
 853                 spin_lock_bh(&queue->lock);
 854                 queue->queue_maxlen = ntohl(*queue_maxlen);
 855                 spin_unlock_bh(&queue->lock);
 856         }
 857
 858 out_put:
 859         instance_put(queue);
 860         return ret;
 861 }
 862
 863 static const struct nfnl_callback nfqnl_cb[NFQNL_MSG_MAX] = {
 864         [NFQNL_MSG_PACKET]      = { .call = nfqnl_recv_unsupp,
 865                                     .attr_count = NFQA_MAX, },
 866         [NFQNL_MSG_VERDICT]     = { .call = nfqnl_recv_verdict,
 867                                     .attr_count = NFQA_MAX,
 868                                     .policy = nfqa_verdict_policy },
 869         [NFQNL_MSG_CONFIG]      = { .call = nfqnl_recv_config,
 870                                     .attr_count = NFQA_CFG_MAX,
 871                                     .policy = nfqa_cfg_policy },
 872 };
 873
 874 static const struct nfnetlink_subsystem nfqnl_subsys = {
 875         .name           = "nf_queue",
 876         .subsys_id      = NFNL_SUBSYS_QUEUE,
 877         .cb_count       = NFQNL_MSG_MAX,
 878         .cb             = nfqnl_cb,
 879 };
 880
 881 #ifdef CONFIG_PROC_FS
 882 struct iter_state {
 883         unsigned int bucket;
 884 };
 885
 886 static struct hlist_node *get_first(struct seq_file *seq)
 887 {
 888         struct iter_state *st = seq->private;
 889
 890         if (!st)
 891                 return NULL;
 892
 893         for (st->bucket = 0; st->bucket < INSTANCE_BUCKETS; st->bucket++) {
 894                 if (!hlist_empty(&instance_table[st->bucket]))
 895                         return instance_table[st->bucket].first;
 896         }
 897         return NULL;
 898 }
 899
 900 static struct hlist_node *get_next(struct seq_file *seq, struct hlist_node *h)
 901 {
 902         struct iter_state *st = seq->private;
 903
 904         h = h->next;
 905         while (!h) {
 906                 if (++st->bucket >= INSTANCE_BUCKETS)
 907                         return NULL;
 908
 909                 h = instance_table[st->bucket].first;
 910         }
 911         return h;
 912 }
 913
 914 static struct hlist_node *get_idx(struct seq_file *seq, loff_t pos)
 915 {
 916         struct hlist_node *head;
 917         head = get_first(seq);
 918
 919         if (head)
 920                 while (pos && (head = get_next(seq, head)))
 921                         pos--;
 922         return pos ? NULL : head;
 923 }
 924
 925 static void *seq_start(struct seq_file *seq, loff_t *pos)
 926 {
 927         read_lock_bh(&instances_lock);
 928         return get_idx(seq, *pos);
 929 }
 930
 931 static void *seq_next(struct seq_file *s, void *v, loff_t *pos)
 932 {
 933         (*pos)++;
 934         return get_next(s, v);
 935 }
 936
 937 static void seq_stop(struct seq_file *s, void *v)
 938 {
 939         read_unlock_bh(&instances_lock);
 940 }
 941
 942 static int seq_show(struct seq_file *s, void *v)
 943 {
 944         const struct nfqnl_instance *inst = v;
 945
 946         return seq_printf(s, "%5d %6d %5d %1d %5d %5d %5d %8d %2d\n",
 947                           inst->queue_num,
 948                           inst->peer_pid, inst->queue_total,
 949                           inst->copy_mode, inst->copy_range,
 950                           inst->queue_dropped, inst->queue_user_dropped,
 951                           atomic_read(&inst->id_sequence),
 952                           atomic_read(&inst->use));
 953 }
 954
 955 static const struct seq_operations nfqnl_seq_ops = {
 956         .start  = seq_start,
 957         .next   = seq_next,
 958         .stop   = seq_stop,
 959         .show   = seq_show,
 960 };
 961
 962 static int nfqnl_open(struct inode *inode, struct file *file)
 963 {
 964         return seq_open_private(file, &nfqnl_seq_ops,
 965                         sizeof(struct iter_state));
 966 }
 967
 968 static const struct file_operations nfqnl_file_ops = {
 969         .owner   = THIS_MODULE,
 970         .open    = nfqnl_open,
 971         .read    = seq_read,
 972         .llseek  = seq_lseek,
 973         .release = seq_release_private,
 974 };
 975
 976 #endif /* PROC_FS */
 977
 978 static int __init nfnetlink_queue_init(void)
 979 {
 980         int i, status = -ENOMEM;
 981 #ifdef CONFIG_PROC_FS
 982         struct proc_dir_entry *proc_nfqueue;
 983 #endif
 984
 985         for (i = 0; i < INSTANCE_BUCKETS; i++)
 986                 INIT_HLIST_HEAD(&instance_table[i]);
 987
 988         netlink_register_notifier(&nfqnl_rtnl_notifier);
 989         status = nfnetlink_subsys_register(&nfqnl_subsys);
 990         if (status < 0) {
 991                 printk(KERN_ERR "nf_queue: failed to create netlink socket\n");
 992                 goto cleanup_netlink_notifier;
 993         }
 994
 995 #ifdef CONFIG_PROC_FS
 996         proc_nfqueue = create_proc_entry("nfnetlink_queue", 0440,
 997                                          proc_net_netfilter);
 998         if (!proc_nfqueue)
 999                 goto cleanup_subsys;
1000         proc_nfqueue->proc_fops = &nfqnl_file_ops;
1001 #endif
1002
1003         register_netdevice_notifier(&nfqnl_dev_notifier);
1004         return status;
1005
1006 #ifdef CONFIG_PROC_FS
1007 cleanup_subsys:
1008         nfnetlink_subsys_unregister(&nfqnl_subsys);
1009 #endif
1010 cleanup_netlink_notifier:
1011         netlink_unregister_notifier(&nfqnl_rtnl_notifier);
1012         return status;
1013 }
1014
1015 static void __exit nfnetlink_queue_fini(void)
1016 {
1017         nf_unregister_queue_handlers(&nfqh);
1018         unregister_netdevice_notifier(&nfqnl_dev_notifier);
1019 #ifdef CONFIG_PROC_FS
1020         remove_proc_entry("nfnetlink_queue", proc_net_netfilter);
1021 #endif
1022         nfnetlink_subsys_unregister(&nfqnl_subsys);
1023         netlink_unregister_notifier(&nfqnl_rtnl_notifier);
1024 }
1025
1026 MODULE_DESCRIPTION("netfilter packet queue handler");
1027 MODULE_AUTHOR("Harald Welte <laforge@netfilter.org>");
1028 MODULE_LICENSE("GPL");
1029 MODULE_ALIAS_NFNL_SUBSYS(NFNL_SUBSYS_QUEUE);
1030
1031 module_init(nfnetlink_queue_init);
1032 module_exit(nfnetlink_queue_fini);