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
 206
 207 static void
 208 issue_verdict(struct nf_queue_entry *entry, int verdict)
 209 {
 210         QDEBUG("entering for entry %p, verdict %u\n", entry, verdict);
 211
 212         /* TCP input path (and probably other bits) assume to be called
 213          * from softirq context, not from syscall, like issue_verdict is
 214          * called.  TCP input path deadlocks with locks taken from timer
 215          * softirq, e.g.  We therefore emulate this by local_bh_disable() */
 216
 217         local_bh_disable();
 218         nf_reinject(entry, verdict);
 219         local_bh_enable();
 220 }
 221
 222 static inline void
 223 __enqueue_entry(struct nfqnl_instance *queue, struct nf_queue_entry *entry)
 224 {
 225        list_add_tail(&entry->list, &queue->queue_list);
 226        queue->queue_total++;
 227 }
 228
 229 static inline int
 230 __nfqnl_set_mode(struct nfqnl_instance *queue,
 231                  unsigned char mode, unsigned int range)
 232 {
 233         int status = 0;
 234
 235         switch (mode) {
 236         case NFQNL_COPY_NONE:
 237         case NFQNL_COPY_META:
 238                 queue->copy_mode = mode;
 239                 queue->copy_range = 0;
 240                 break;
 241
 242         case NFQNL_COPY_PACKET:
 243                 queue->copy_mode = mode;
 244                 /* we're using struct nlattr which has 16bit nla_len */
 245                 if (range > 0xffff)
 246                         queue->copy_range = 0xffff;
 247                 else
 248                         queue->copy_range = range;
 249                 break;
 250
 251         default:
 252                 status = -EINVAL;
 253
 254         }
 255         return status;
 256 }
 257
 258 static struct nf_queue_entry *
 259 find_dequeue_entry(struct nfqnl_instance *queue, unsigned int id)
 260 {
 261         struct nf_queue_entry *entry = NULL, *i;
 262
 263         spin_lock_bh(&queue->lock);
 264
 265         list_for_each_entry(i, &queue->queue_list, list) {
 266                 if (i->id == id) {
 267                         entry = i;
 268                         break;
 269                 }
 270         }
 271
 272         if (entry) {
 273                 list_del(&entry->list);
 274                 queue->queue_total--;
 275         }
 276
 277         spin_unlock_bh(&queue->lock);
 278
 279         return entry;
 280 }
 281
 282 static void
 283 nfqnl_flush(struct nfqnl_instance *queue, nfqnl_cmpfn cmpfn, unsigned long data)
 284 {
 285         struct nf_queue_entry *entry, *next;
 286
 287         spin_lock_bh(&queue->lock);
 288         list_for_each_entry_safe(entry, next, &queue->queue_list, list) {
 289                 if (!cmpfn || cmpfn(entry, data)) {
 290                         list_del(&entry->list);
 291                         queue->queue_total--;
 292                         issue_verdict(entry, NF_DROP);
 293                 }
 294         }
 295         spin_unlock_bh(&queue->lock);
 296 }
 297
 298 static struct sk_buff *
 299 nfqnl_build_packet_message(struct nfqnl_instance *queue,
 300                            struct nf_queue_entry *entry, int *errp)
 301 {
 302         sk_buff_data_t old_tail;
 303         size_t size;
 304         size_t data_len = 0;
 305         struct sk_buff *skb;
 306         struct nfqnl_msg_packet_hdr pmsg;
 307         struct nlmsghdr *nlh;
 308         struct nfgenmsg *nfmsg;
 309         struct sk_buff *entskb = entry->skb;
 310         struct net_device *indev;
 311         struct net_device *outdev;
 312         __be32 tmp_uint;
 313
 314         QDEBUG("entered\n");
 315
 316         size =    NLMSG_ALIGN(sizeof(struct nfgenmsg))
 317                 + nla_total_size(sizeof(struct nfqnl_msg_packet_hdr))
 318                 + nla_total_size(sizeof(u_int32_t))     /* ifindex */
 319                 + nla_total_size(sizeof(u_int32_t))     /* ifindex */
 320 #ifdef CONFIG_BRIDGE_NETFILTER
 321                 + nla_total_size(sizeof(u_int32_t))     /* ifindex */
 322                 + nla_total_size(sizeof(u_int32_t))     /* ifindex */
 323 #endif
 324                 + nla_total_size(sizeof(u_int32_t))     /* mark */
 325                 + nla_total_size(sizeof(struct nfqnl_msg_packet_hw))
 326                 + nla_total_size(sizeof(struct nfqnl_msg_packet_timestamp));
 327
 328         outdev = entry->outdev;
 329
 330         spin_lock_bh(&queue->lock);
 331
 332         switch (queue->copy_mode) {
 333         case NFQNL_COPY_META:
 334         case NFQNL_COPY_NONE:
 335                 data_len = 0;
 336                 break;
 337
 338         case NFQNL_COPY_PACKET:
 339                 if ((entskb->ip_summed == CHECKSUM_PARTIAL ||
 340                      entskb->ip_summed == CHECKSUM_COMPLETE) &&
 341                     (*errp = skb_checksum_help(entskb))) {
 342                         spin_unlock_bh(&queue->lock);
 343                         return NULL;
 344                 }
 345                 if (queue->copy_range == 0
 346                     || queue->copy_range > entskb->len)
 347                         data_len = entskb->len;
 348                 else
 349                         data_len = queue->copy_range;
 350
 351                 size += nla_total_size(data_len);
 352                 break;
 353
 354         default:
 355                 *errp = -EINVAL;
 356                 spin_unlock_bh(&queue->lock);
 357                 return NULL;
 358         }
 359
 360         spin_unlock_bh(&queue->lock);
 361
 362         skb = alloc_skb(size, GFP_ATOMIC);
 363         if (!skb)
 364                 goto nlmsg_failure;
 365
 366         old_tail = skb->tail;
 367         nlh = NLMSG_PUT(skb, 0, 0,
 368                         NFNL_SUBSYS_QUEUE << 8 | NFQNL_MSG_PACKET,
 369                         sizeof(struct nfgenmsg));
 370         nfmsg = NLMSG_DATA(nlh);
 371         nfmsg->nfgen_family = entry->pf;
 372         nfmsg->version = NFNETLINK_V0;
 373         nfmsg->res_id = htons(queue->queue_num);
 374
 375         pmsg.packet_id          = htonl(entry->id);
 376         pmsg.hw_protocol        = entskb->protocol;
 377         pmsg.hook               = entry->hook;
 378
 379         NLA_PUT(skb, NFQA_PACKET_HDR, sizeof(pmsg), &pmsg);
 380
 381         indev = entry->indev;
 382         if (indev) {
 383                 tmp_uint = htonl(indev->ifindex);
 384 #ifndef CONFIG_BRIDGE_NETFILTER
 385                 NLA_PUT(skb, NFQA_IFINDEX_INDEV, sizeof(tmp_uint), &tmp_uint);
 386 #else
 387                 if (entry->pf == PF_BRIDGE) {
 388                         /* Case 1: indev is physical input device, we need to
 389                          * look for bridge group (when called from
 390                          * netfilter_bridge) */
 391                         NLA_PUT(skb, NFQA_IFINDEX_PHYSINDEV, sizeof(tmp_uint),
 392                                 &tmp_uint);
 393                         /* this is the bridge group "brX" */
 394                         tmp_uint = htonl(indev->br_port->br->dev->ifindex);
 395                         NLA_PUT(skb, NFQA_IFINDEX_INDEV, sizeof(tmp_uint),
 396                                 &tmp_uint);
 397                 } else {
 398                         /* Case 2: indev is bridge group, we need to look for
 399                          * physical device (when called from ipv4) */
 400                         NLA_PUT(skb, NFQA_IFINDEX_INDEV, sizeof(tmp_uint),
 401                                 &tmp_uint);
 402                         if (entskb->nf_bridge
 403                             && entskb->nf_bridge->physindev) {
 404                                 tmp_uint = htonl(entskb->nf_bridge->physindev->ifindex);
 405                                 NLA_PUT(skb, NFQA_IFINDEX_PHYSINDEV,
 406                                         sizeof(tmp_uint), &tmp_uint);
 407                         }
 408                 }
 409 #endif
 410         }
 411
 412         if (outdev) {
 413                 tmp_uint = htonl(outdev->ifindex);
 414 #ifndef CONFIG_BRIDGE_NETFILTER
 415                 NLA_PUT(skb, NFQA_IFINDEX_OUTDEV, sizeof(tmp_uint), &tmp_uint);
 416 #else
 417                 if (entry->pf == PF_BRIDGE) {
 418                         /* Case 1: outdev is physical output device, we need to
 419                          * look for bridge group (when called from
 420                          * netfilter_bridge) */
 421                         NLA_PUT(skb, NFQA_IFINDEX_PHYSOUTDEV, sizeof(tmp_uint),
 422                                 &tmp_uint);
 423                         /* this is the bridge group "brX" */
 424                         tmp_uint = htonl(outdev->br_port->br->dev->ifindex);
 425                         NLA_PUT(skb, NFQA_IFINDEX_OUTDEV, sizeof(tmp_uint),
 426                                 &tmp_uint);
 427                 } else {
 428                         /* Case 2: outdev is bridge group, we need to look for
 429                          * physical output device (when called from ipv4) */
 430                         NLA_PUT(skb, NFQA_IFINDEX_OUTDEV, sizeof(tmp_uint),
 431                                 &tmp_uint);
 432                         if (entskb->nf_bridge
 433                             && entskb->nf_bridge->physoutdev) {
 434                                 tmp_uint = htonl(entskb->nf_bridge->physoutdev->ifindex);
 435                                 NLA_PUT(skb, NFQA_IFINDEX_PHYSOUTDEV,
 436                                         sizeof(tmp_uint), &tmp_uint);
 437                         }
 438                 }
 439 #endif
 440         }
 441
 442         if (entskb->mark) {
 443                 tmp_uint = htonl(entskb->mark);
 444                 NLA_PUT(skb, NFQA_MARK, sizeof(u_int32_t), &tmp_uint);
 445         }
 446
 447         if (indev && entskb->dev) {
 448                 struct nfqnl_msg_packet_hw phw;
 449                 int len = dev_parse_header(entskb, phw.hw_addr);
 450                 if (len) {
 451                         phw.hw_addrlen = htons(len);
 452                         NLA_PUT(skb, NFQA_HWADDR, sizeof(phw), &phw);
 453                 }
 454         }
 455
 456         if (entskb->tstamp.tv64) {
 457                 struct nfqnl_msg_packet_timestamp ts;
 458                 struct timeval tv = ktime_to_timeval(entskb->tstamp);
 459                 ts.sec = cpu_to_be64(tv.tv_sec);
 460                 ts.usec = cpu_to_be64(tv.tv_usec);
 461
 462                 NLA_PUT(skb, NFQA_TIMESTAMP, sizeof(ts), &ts);
 463         }
 464
 465         if (data_len) {
 466                 struct nlattr *nla;
 467                 int size = nla_attr_size(data_len);
 468
 469                 if (skb_tailroom(skb) < nla_total_size(data_len)) {
 470                         printk(KERN_WARNING "nf_queue: no tailroom!\n");
 471                         goto nlmsg_failure;
 472                 }
 473
 474                 nla = (struct nlattr *)skb_put(skb, nla_total_size(data_len));
 475                 nla->nla_type = NFQA_PAYLOAD;
 476                 nla->nla_len = size;
 477
 478                 if (skb_copy_bits(entskb, 0, nla_data(nla), data_len))
 479                         BUG();
 480         }
 481
 482         nlh->nlmsg_len = skb->tail - old_tail;
 483         return skb;
 484
 485 nlmsg_failure:
 486 nla_put_failure:
 487         if (skb)
 488                 kfree_skb(skb);
 489         *errp = -EINVAL;
 490         if (net_ratelimit())
 491                 printk(KERN_ERR "nf_queue: error creating packet message\n");
 492         return NULL;
 493 }
 494
 495 static int
 496 nfqnl_enqueue_packet(struct nf_queue_entry *entry, unsigned int queuenum)
 497 {
 498         int status = -EINVAL;
 499         struct sk_buff *nskb;
 500         struct nfqnl_instance *queue;
 501
 502         QDEBUG("entered\n");
 503
 504         queue = instance_lookup_get(queuenum);
 505         if (!queue) {
 506                 QDEBUG("no queue instance matching\n");
 507                 return -EINVAL;
 508         }
 509
 510         if (queue->copy_mode == NFQNL_COPY_NONE) {
 511                 QDEBUG("mode COPY_NONE, aborting\n");
 512                 status = -EAGAIN;
 513                 goto err_out_put;
 514         }
 515
 516         entry->id = atomic_inc_return(&queue->id_sequence);
 517
 518         nskb = nfqnl_build_packet_message(queue, entry, &status);
 519         if (nskb == NULL)
 520                 goto err_out_put;
 521
 522         spin_lock_bh(&queue->lock);
 523
 524         if (!queue->peer_pid)
 525                 goto err_out_free_nskb;
 526
 527         if (queue->queue_total >= queue->queue_maxlen) {
 528                 queue->queue_dropped++;
 529                 status = -ENOSPC;
 530                 if (net_ratelimit())
 531                           printk(KERN_WARNING "nf_queue: full at %d entries, "
 532                                  "dropping packets(s). Dropped: %d\n",
 533                                  queue->queue_total, queue->queue_dropped);
 534                 goto err_out_free_nskb;
 535         }
 536
 537         /* nfnetlink_unicast will either free the nskb or add it to a socket */
 538         status = nfnetlink_unicast(nskb, queue->peer_pid, MSG_DONTWAIT);
 539         if (status < 0) {
 540                 queue->queue_user_dropped++;
 541                 goto err_out_unlock;
 542         }
 543
 544         __enqueue_entry(queue, entry);
 545
 546         spin_unlock_bh(&queue->lock);
 547         instance_put(queue);
 548         return status;
 549
 550 err_out_free_nskb:
 551         kfree_skb(nskb);
 552
 553 err_out_unlock:
 554         spin_unlock_bh(&queue->lock);
 555
 556 err_out_put:
 557         instance_put(queue);
 558         return status;
 559 }
 560
 561 static int
 562 nfqnl_mangle(void *data, int data_len, struct nf_queue_entry *e)
 563 {
 564         int diff;
 565         int err;
 566
 567         diff = data_len - e->skb->len;
 568         if (diff < 0) {
 569                 if (pskb_trim(e->skb, data_len))
 570                         return -ENOMEM;
 571         } else if (diff > 0) {
 572                 if (data_len > 0xFFFF)
 573                         return -EINVAL;
 574                 if (diff > skb_tailroom(e->skb)) {
 575                         err = pskb_expand_head(e->skb, 0,
 576                                                diff - skb_tailroom(e->skb),
 577                                                GFP_ATOMIC);
 578                         if (err) {
 579                                 printk(KERN_WARNING "nf_queue: OOM "
 580                                       "in mangle, dropping packet\n");
 581                                 return err;
 582                         }
 583                 }
 584                 skb_put(e->skb, diff);
 585         }
 586         if (!skb_make_writable(e->skb, data_len))
 587                 return -ENOMEM;
 588         skb_copy_to_linear_data(e->skb, data, data_len);
 589         e->skb->ip_summed = CHECKSUM_NONE;
 590         return 0;
 591 }
 592
 593 static int
 594 nfqnl_set_mode(struct nfqnl_instance *queue,
 595                unsigned char mode, unsigned int range)
 596 {
 597         int status;
 598
 599         spin_lock_bh(&queue->lock);
 600         status = __nfqnl_set_mode(queue, mode, range);
 601         spin_unlock_bh(&queue->lock);
 602
 603         return status;
 604 }
 605
 606 static int
 607 dev_cmp(struct nf_queue_entry *entry, unsigned long ifindex)
 608 {
 609         if (entry->indev)
 610                 if (entry->indev->ifindex == ifindex)
 611                         return 1;
 612         if (entry->outdev)
 613                 if (entry->outdev->ifindex == ifindex)
 614                         return 1;
 615 #ifdef CONFIG_BRIDGE_NETFILTER
 616         if (entry->skb->nf_bridge) {
 617                 if (entry->skb->nf_bridge->physindev &&
 618                     entry->skb->nf_bridge->physindev->ifindex == ifindex)
 619                         return 1;
 620                 if (entry->skb->nf_bridge->physoutdev &&
 621                     entry->skb->nf_bridge->physoutdev->ifindex == ifindex)
 622                         return 1;
 623         }
 624 #endif
 625         return 0;
 626 }
 627
 628 /* drop all packets with either indev or outdev == ifindex from all queue
 629  * instances */
 630 static void
 631 nfqnl_dev_drop(int ifindex)
 632 {
 633         int i;
 634
 635         QDEBUG("entering for ifindex %u\n", ifindex);
 636
 637         /* this only looks like we have to hold the readlock for a way too long
 638          * time, issue_verdict(),  nf_reinject(), ... - but we always only
 639          * issue NF_DROP, which is processed directly in nf_reinject() */
 640         read_lock_bh(&instances_lock);
 641
 642         for  (i = 0; i < INSTANCE_BUCKETS; i++) {
 643                 struct hlist_node *tmp;
 644                 struct nfqnl_instance *inst;
 645                 struct hlist_head *head = &instance_table[i];
 646
 647                 hlist_for_each_entry(inst, tmp, head, hlist)
 648                         nfqnl_flush(inst, dev_cmp, ifindex);
 649         }
 650
 651         read_unlock_bh(&instances_lock);
 652 }
 653
 654 #define RCV_SKB_FAIL(err) do { netlink_ack(skb, nlh, (err)); return; } while (0)
 655
 656 static int
 657 nfqnl_rcv_dev_event(struct notifier_block *this,
 658                     unsigned long event, void *ptr)
 659 {
 660         struct net_device *dev = ptr;
 661
 662         if (dev->nd_net != &init_net)
 663                 return NOTIFY_DONE;
 664
 665         /* Drop any packets associated with the downed device */
 666         if (event == NETDEV_DOWN)
 667                 nfqnl_dev_drop(dev->ifindex);
 668         return NOTIFY_DONE;
 669 }
 670
 671 static struct notifier_block nfqnl_dev_notifier = {
 672         .notifier_call  = nfqnl_rcv_dev_event,
 673 };
 674
 675 static int
 676 nfqnl_rcv_nl_event(struct notifier_block *this,
 677                    unsigned long event, void *ptr)
 678 {
 679         struct netlink_notify *n = ptr;
 680
 681         if (event == NETLINK_URELEASE &&
 682             n->protocol == NETLINK_NETFILTER && n->pid) {
 683                 int i;
 684
 685                 /* destroy all instances for this pid */
 686                 write_lock_bh(&instances_lock);
 687                 for  (i = 0; i < INSTANCE_BUCKETS; i++) {
 688                         struct hlist_node *tmp, *t2;
 689                         struct nfqnl_instance *inst;
 690                         struct hlist_head *head = &instance_table[i];
 691
 692                         hlist_for_each_entry_safe(inst, tmp, t2, head, hlist) {
 693                                 if ((n->net == &init_net) &&
 694                                     (n->pid == inst->peer_pid))
 695                                         __instance_destroy(inst);
 696                         }
 697                 }
 698                 write_unlock_bh(&instances_lock);
 699         }
 700         return NOTIFY_DONE;
 701 }
 702
 703 static struct notifier_block nfqnl_rtnl_notifier = {
 704         .notifier_call  = nfqnl_rcv_nl_event,
 705 };
 706
 707 static const struct nla_policy nfqa_verdict_policy[NFQA_MAX+1] = {
 708         [NFQA_VERDICT_HDR]      = { .len = sizeof(struct nfqnl_msg_verdict_hdr) },
 709         [NFQA_MARK]             = { .type = NLA_U32 },
 710         [NFQA_PAYLOAD]          = { .type = NLA_UNSPEC },
 711 };
 712
 713 static int
 714 nfqnl_recv_verdict(struct sock *ctnl, struct sk_buff *skb,
 715                    struct nlmsghdr *nlh, struct nlattr *nfqa[])
 716 {
 717         struct nfgenmsg *nfmsg = NLMSG_DATA(nlh);
 718         u_int16_t queue_num = ntohs(nfmsg->res_id);
 719
 720         struct nfqnl_msg_verdict_hdr *vhdr;
 721         struct nfqnl_instance *queue;
 722         unsigned int verdict;
 723         struct nf_queue_entry *entry;
 724         int err;
 725
 726         queue = instance_lookup_get(queue_num);
 727         if (!queue)
 728                 return -ENODEV;
 729
 730         if (queue->peer_pid != NETLINK_CB(skb).pid) {
 731                 err = -EPERM;
 732                 goto err_out_put;
 733         }
 734
 735         if (!nfqa[NFQA_VERDICT_HDR]) {
 736                 err = -EINVAL;
 737                 goto err_out_put;
 738         }
 739
 740         vhdr = nla_data(nfqa[NFQA_VERDICT_HDR]);
 741         verdict = ntohl(vhdr->verdict);
 742
 743         if ((verdict & NF_VERDICT_MASK) > NF_MAX_VERDICT) {
 744                 err = -EINVAL;
 745                 goto err_out_put;
 746         }
 747
 748         entry = find_dequeue_entry(queue, ntohl(vhdr->id));
 749         if (entry == NULL) {
 750                 err = -ENOENT;
 751                 goto err_out_put;
 752         }
 753
 754         if (nfqa[NFQA_PAYLOAD]) {
 755                 if (nfqnl_mangle(nla_data(nfqa[NFQA_PAYLOAD]),
 756                                  nla_len(nfqa[NFQA_PAYLOAD]), entry) < 0)
 757                         verdict = NF_DROP;
 758         }
 759
 760         if (nfqa[NFQA_MARK])
 761                 entry->skb->mark = ntohl(*(__be32 *)
 762                                          nla_data(nfqa[NFQA_MARK]));
 763
 764         issue_verdict(entry, verdict);
 765         instance_put(queue);
 766         return 0;
 767
 768 err_out_put:
 769         instance_put(queue);
 770         return err;
 771 }
 772
 773 static int
 774 nfqnl_recv_unsupp(struct sock *ctnl, struct sk_buff *skb,
 775                   struct nlmsghdr *nlh, struct nlattr *nfqa[])
 776 {
 777         return -ENOTSUPP;
 778 }
 779
 780 static const struct nla_policy nfqa_cfg_policy[NFQA_CFG_MAX+1] = {
 781         [NFQA_CFG_CMD]          = { .len = sizeof(struct nfqnl_msg_config_cmd) },
 782         [NFQA_CFG_PARAMS]       = { .len = sizeof(struct nfqnl_msg_config_params) },
 783 };
 784
 785 static const struct nf_queue_handler nfqh = {
 786         .name   = "nf_queue",
 787         .outfn  = &nfqnl_enqueue_packet,
 788 };
 789
 790 static int
 791 nfqnl_recv_config(struct sock *ctnl, struct sk_buff *skb,
 792                   struct nlmsghdr *nlh, struct nlattr *nfqa[])
 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         queue = instance_lookup_get(queue_num);
 802         if (nfqa[NFQA_CFG_CMD]) {
 803                 struct nfqnl_msg_config_cmd *cmd;
 804                 cmd = nla_data(nfqa[NFQA_CFG_CMD]);
 805                 QDEBUG("found CFG_CMD\n");
 806
 807                 switch (cmd->command) {
 808                 case NFQNL_CFG_CMD_BIND:
 809                         if (queue)
 810                                 return -EBUSY;
 811
 812                         queue = instance_create(queue_num, NETLINK_CB(skb).pid);
 813                         if (!queue)
 814                                 return -EINVAL;
 815                         break;
 816                 case NFQNL_CFG_CMD_UNBIND:
 817                         if (!queue)
 818                                 return -ENODEV;
 819
 820                         if (queue->peer_pid != NETLINK_CB(skb).pid) {
 821                                 ret = -EPERM;
 822                                 goto out_put;
 823                         }
 824
 825                         instance_destroy(queue);
 826                         break;
 827                 case NFQNL_CFG_CMD_PF_BIND:
 828                         QDEBUG("registering queue handler for pf=%u\n",
 829                                 ntohs(cmd->pf));
 830                         ret = nf_register_queue_handler(ntohs(cmd->pf), &nfqh);
 831                         break;
 832                 case NFQNL_CFG_CMD_PF_UNBIND:
 833                         QDEBUG("unregistering queue handler for pf=%u\n",
 834                                 ntohs(cmd->pf));
 835                         ret = nf_unregister_queue_handler(ntohs(cmd->pf), &nfqh);
 836                         break;
 837                 default:
 838                         ret = -EINVAL;
 839                         break;
 840                 }
 841         } else {
 842                 if (!queue) {
 843                         QDEBUG("no config command, and no instance ENOENT\n");
 844                         ret = -ENOENT;
 845                         goto out_put;
 846                 }
 847
 848                 if (queue->peer_pid != NETLINK_CB(skb).pid) {
 849                         QDEBUG("no config command, and wrong pid\n");
 850                         ret = -EPERM;
 851                         goto out_put;
 852                 }
 853         }
 854
 855         if (nfqa[NFQA_CFG_PARAMS]) {
 856                 struct nfqnl_msg_config_params *params;
 857
 858                 if (!queue) {
 859                         ret = -ENOENT;
 860                         goto out_put;
 861                 }
 862                 params = nla_data(nfqa[NFQA_CFG_PARAMS]);
 863                 nfqnl_set_mode(queue, params->copy_mode,
 864                                 ntohl(params->copy_range));
 865         }
 866
 867         if (nfqa[NFQA_CFG_QUEUE_MAXLEN]) {
 868                 __be32 *queue_maxlen;
 869                 queue_maxlen = nla_data(nfqa[NFQA_CFG_QUEUE_MAXLEN]);
 870                 spin_lock_bh(&queue->lock);
 871                 queue->queue_maxlen = ntohl(*queue_maxlen);
 872                 spin_unlock_bh(&queue->lock);
 873         }
 874
 875 out_put:
 876         instance_put(queue);
 877         return ret;
 878 }
 879
 880 static const struct nfnl_callback nfqnl_cb[NFQNL_MSG_MAX] = {
 881         [NFQNL_MSG_PACKET]      = { .call = nfqnl_recv_unsupp,
 882                                     .attr_count = NFQA_MAX, },
 883         [NFQNL_MSG_VERDICT]     = { .call = nfqnl_recv_verdict,
 884                                     .attr_count = NFQA_MAX,
 885                                     .policy = nfqa_verdict_policy },
 886         [NFQNL_MSG_CONFIG]      = { .call = nfqnl_recv_config,
 887                                     .attr_count = NFQA_CFG_MAX,
 888                                     .policy = nfqa_cfg_policy },
 889 };
 890
 891 static const struct nfnetlink_subsystem nfqnl_subsys = {
 892         .name           = "nf_queue",
 893         .subsys_id      = NFNL_SUBSYS_QUEUE,
 894         .cb_count       = NFQNL_MSG_MAX,
 895         .cb             = nfqnl_cb,
 896 };
 897
 898 #ifdef CONFIG_PROC_FS
 899 struct iter_state {
 900         unsigned int bucket;
 901 };
 902
 903 static struct hlist_node *get_first(struct seq_file *seq)
 904 {
 905         struct iter_state *st = seq->private;
 906
 907         if (!st)
 908                 return NULL;
 909
 910         for (st->bucket = 0; st->bucket < INSTANCE_BUCKETS; st->bucket++) {
 911                 if (!hlist_empty(&instance_table[st->bucket]))
 912                         return instance_table[st->bucket].first;
 913         }
 914         return NULL;
 915 }
 916
 917 static struct hlist_node *get_next(struct seq_file *seq, struct hlist_node *h)
 918 {
 919         struct iter_state *st = seq->private;
 920
 921         h = h->next;
 922         while (!h) {
 923                 if (++st->bucket >= INSTANCE_BUCKETS)
 924                         return NULL;
 925
 926                 h = instance_table[st->bucket].first;
 927         }
 928         return h;
 929 }
 930
 931 static struct hlist_node *get_idx(struct seq_file *seq, loff_t pos)
 932 {
 933         struct hlist_node *head;
 934         head = get_first(seq);
 935
 936         if (head)
 937                 while (pos && (head = get_next(seq, head)))
 938                         pos--;
 939         return pos ? NULL : head;
 940 }
 941
 942 static void *seq_start(struct seq_file *seq, loff_t *pos)
 943 {
 944         read_lock_bh(&instances_lock);
 945         return get_idx(seq, *pos);
 946 }
 947
 948 static void *seq_next(struct seq_file *s, void *v, loff_t *pos)
 949 {
 950         (*pos)++;
 951         return get_next(s, v);
 952 }
 953
 954 static void seq_stop(struct seq_file *s, void *v)
 955 {
 956         read_unlock_bh(&instances_lock);
 957 }
 958
 959 static int seq_show(struct seq_file *s, void *v)
 960 {
 961         const struct nfqnl_instance *inst = v;
 962
 963         return seq_printf(s, "%5d %6d %5d %1d %5d %5d %5d %8d %2d\n",
 964                           inst->queue_num,
 965                           inst->peer_pid, inst->queue_total,
 966                           inst->copy_mode, inst->copy_range,
 967                           inst->queue_dropped, inst->queue_user_dropped,
 968                           atomic_read(&inst->id_sequence),
 969                           atomic_read(&inst->use));
 970 }
 971
 972 static const struct seq_operations nfqnl_seq_ops = {
 973         .start  = seq_start,
 974         .next   = seq_next,
 975         .stop   = seq_stop,
 976         .show   = seq_show,
 977 };
 978
 979 static int nfqnl_open(struct inode *inode, struct file *file)
 980 {
 981         return seq_open_private(file, &nfqnl_seq_ops,
 982                         sizeof(struct iter_state));
 983 }
 984
 985 static const struct file_operations nfqnl_file_ops = {
 986         .owner   = THIS_MODULE,
 987         .open    = nfqnl_open,
 988         .read    = seq_read,
 989         .llseek  = seq_lseek,
 990         .release = seq_release_private,
 991 };
 992
 993 #endif /* PROC_FS */
 994
 995 static int __init nfnetlink_queue_init(void)
 996 {
 997         int i, status = -ENOMEM;
 998 #ifdef CONFIG_PROC_FS
 999         struct proc_dir_entry *proc_nfqueue;
1000 #endif
1001
1002         for (i = 0; i < INSTANCE_BUCKETS; i++)
1003                 INIT_HLIST_HEAD(&instance_table[i]);
1004
1005         netlink_register_notifier(&nfqnl_rtnl_notifier);
1006         status = nfnetlink_subsys_register(&nfqnl_subsys);
1007         if (status < 0) {
1008                 printk(KERN_ERR "nf_queue: failed to create netlink socket\n");
1009                 goto cleanup_netlink_notifier;
1010         }
1011
1012 #ifdef CONFIG_PROC_FS
1013         proc_nfqueue = create_proc_entry("nfnetlink_queue", 0440,
1014                                          proc_net_netfilter);
1015         if (!proc_nfqueue)
1016                 goto cleanup_subsys;
1017         proc_nfqueue->proc_fops = &nfqnl_file_ops;
1018 #endif
1019
1020         register_netdevice_notifier(&nfqnl_dev_notifier);
1021         return status;
1022
1023 #ifdef CONFIG_PROC_FS
1024 cleanup_subsys:
1025         nfnetlink_subsys_unregister(&nfqnl_subsys);
1026 #endif
1027 cleanup_netlink_notifier:
1028         netlink_unregister_notifier(&nfqnl_rtnl_notifier);
1029         return status;
1030 }
1031
1032 static void __exit nfnetlink_queue_fini(void)
1033 {
1034         nf_unregister_queue_handlers(&nfqh);
1035         unregister_netdevice_notifier(&nfqnl_dev_notifier);
1036 #ifdef CONFIG_PROC_FS
1037         remove_proc_entry("nfnetlink_queue", proc_net_netfilter);
1038 #endif
1039         nfnetlink_subsys_unregister(&nfqnl_subsys);
1040         netlink_unregister_notifier(&nfqnl_rtnl_notifier);
1041 }
1042
1043 MODULE_DESCRIPTION("netfilter packet queue handler");
1044 MODULE_AUTHOR("Harald Welte <laforge@netfilter.org>");
1045 MODULE_LICENSE("GPL");
1046 MODULE_ALIAS_NFNL_SUBSYS(NFNL_SUBSYS_QUEUE);
1047
1048 module_init(nfnetlink_queue_init);
1049 module_exit(nfnetlink_queue_fini);