err.no Git - 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_queue_entry {
  49         struct list_head list;
  50         struct nf_info *info;
  51         struct sk_buff *skb;
  52         unsigned int id;
  53 };
  54
  55 struct nfqnl_instance {
  56         struct hlist_node hlist;                /* global list of queues */
  57         atomic_t use;
  58
  59         int peer_pid;
  60         unsigned int queue_maxlen;
  61         unsigned int copy_range;
  62         unsigned int queue_total;
  63         unsigned int queue_dropped;
  64         unsigned int queue_user_dropped;
  65
  66         atomic_t id_sequence;                   /* 'sequence' of pkt ids */
  67
  68         u_int16_t queue_num;                    /* number of this queue */
  69         u_int8_t copy_mode;
  70
  71         spinlock_t lock;
  72
  73         struct list_head queue_list;            /* packets in queue */
  74 };
  75
  76 typedef int (*nfqnl_cmpfn)(struct nfqnl_queue_entry *, unsigned long);
  77
  78 static DEFINE_RWLOCK(instances_lock);
  79
  80 #define INSTANCE_BUCKETS        16
  81 static struct hlist_head instance_table[INSTANCE_BUCKETS];
  82
  83 static inline u_int8_t instance_hashfn(u_int16_t queue_num)
  84 {
  85         return ((queue_num >> 8) | queue_num) % INSTANCE_BUCKETS;
  86 }
  87
  88 static struct nfqnl_instance *
  89 __instance_lookup(u_int16_t queue_num)
  90 {
  91         struct hlist_head *head;
  92         struct hlist_node *pos;
  93         struct nfqnl_instance *inst;
  94
  95         head = &instance_table[instance_hashfn(queue_num)];
  96         hlist_for_each_entry(inst, pos, head, hlist) {
  97                 if (inst->queue_num == queue_num)
  98                         return inst;
  99         }
 100         return NULL;
 101 }
 102
 103 static struct nfqnl_instance *
 104 instance_lookup_get(u_int16_t queue_num)
 105 {
 106         struct nfqnl_instance *inst;
 107
 108         read_lock_bh(&instances_lock);
 109         inst = __instance_lookup(queue_num);
 110         if (inst)
 111                 atomic_inc(&inst->use);
 112         read_unlock_bh(&instances_lock);
 113
 114         return inst;
 115 }
 116
 117 static void
 118 instance_put(struct nfqnl_instance *inst)
 119 {
 120         if (inst && atomic_dec_and_test(&inst->use)) {
 121                 QDEBUG("kfree(inst=%p)\n", inst);
 122                 kfree(inst);
 123         }
 124 }
 125
 126 static struct nfqnl_instance *
 127 instance_create(u_int16_t queue_num, int pid)
 128 {
 129         struct nfqnl_instance *inst;
 130
 131         QDEBUG("entering for queue_num=%u, pid=%d\n", queue_num, pid);
 132
 133         write_lock_bh(&instances_lock);
 134         if (__instance_lookup(queue_num)) {
 135                 inst = NULL;
 136                 QDEBUG("aborting, instance already exists\n");
 137                 goto out_unlock;
 138         }
 139
 140         inst = kzalloc(sizeof(*inst), GFP_ATOMIC);
 141         if (!inst)
 142                 goto out_unlock;
 143
 144         inst->queue_num = queue_num;
 145         inst->peer_pid = pid;
 146         inst->queue_maxlen = NFQNL_QMAX_DEFAULT;
 147         inst->copy_range = 0xfffff;
 148         inst->copy_mode = NFQNL_COPY_NONE;
 149         atomic_set(&inst->id_sequence, 0);
 150         /* needs to be two, since we _put() after creation */
 151         atomic_set(&inst->use, 2);
 152         spin_lock_init(&inst->lock);
 153         INIT_LIST_HEAD(&inst->queue_list);
 154
 155         if (!try_module_get(THIS_MODULE))
 156                 goto out_free;
 157
 158         hlist_add_head(&inst->hlist,
 159                        &instance_table[instance_hashfn(queue_num)]);
 160
 161         write_unlock_bh(&instances_lock);
 162
 163         QDEBUG("successfully created new instance\n");
 164
 165         return inst;
 166
 167 out_free:
 168         kfree(inst);
 169 out_unlock:
 170         write_unlock_bh(&instances_lock);
 171         return NULL;
 172 }
 173
 174 static void nfqnl_flush(struct nfqnl_instance *queue, int verdict);
 175
 176 static void
 177 _instance_destroy2(struct nfqnl_instance *inst, int lock)
 178 {
 179         /* first pull it out of the global list */
 180         if (lock)
 181                 write_lock_bh(&instances_lock);
 182
 183         QDEBUG("removing instance %p (queuenum=%u) from hash\n",
 184                 inst, inst->queue_num);
 185         hlist_del(&inst->hlist);
 186
 187         if (lock)
 188                 write_unlock_bh(&instances_lock);
 189
 190         /* then flush all pending skbs from the queue */
 191         nfqnl_flush(inst, NF_DROP);
 192
 193         /* and finally put the refcount */
 194         instance_put(inst);
 195
 196         module_put(THIS_MODULE);
 197 }
 198
 199 static inline void
 200 __instance_destroy(struct nfqnl_instance *inst)
 201 {
 202         _instance_destroy2(inst, 0);
 203 }
 204
 205 static inline void
 206 instance_destroy(struct nfqnl_instance *inst)
 207 {
 208         _instance_destroy2(inst, 1);
 209 }
 210
 211
 212
 213 static void
 214 issue_verdict(struct nfqnl_queue_entry *entry, int verdict)
 215 {
 216         QDEBUG("entering for entry %p, verdict %u\n", entry, verdict);
 217
 218         /* TCP input path (and probably other bits) assume to be called
 219          * from softirq context, not from syscall, like issue_verdict is
 220          * called.  TCP input path deadlocks with locks taken from timer
 221          * softirq, e.g.  We therefore emulate this by local_bh_disable() */
 222
 223         local_bh_disable();
 224         nf_reinject(entry->skb, entry->info, verdict);
 225         local_bh_enable();
 226
 227         kfree(entry);
 228 }
 229
 230 static inline void
 231 __enqueue_entry(struct nfqnl_instance *queue,
 232                       struct nfqnl_queue_entry *entry)
 233 {
 234        list_add_tail(&entry->list, &queue->queue_list);
 235        queue->queue_total++;
 236 }
 237
 238 /*
 239  * Find and return a queued entry matched by cmpfn, or return the last
 240  * entry if cmpfn is NULL.
 241  */
 242 static inline struct nfqnl_queue_entry *
 243 __find_entry(struct nfqnl_instance *queue, nfqnl_cmpfn cmpfn,
 244                    unsigned long data)
 245 {
 246         struct nfqnl_queue_entry *entry;
 247
 248         list_for_each_entry(entry, &queue->queue_list, list) {
 249                 if (!cmpfn || cmpfn(entry, data))
 250                         return entry;
 251         }
 252         return NULL;
 253 }
 254
 255 static inline void
 256 __dequeue_entry(struct nfqnl_instance *q, struct nfqnl_queue_entry *entry)
 257 {
 258         list_del(&entry->list);
 259         q->queue_total--;
 260 }
 261
 262 static inline struct nfqnl_queue_entry *
 263 __find_dequeue_entry(struct nfqnl_instance *queue,
 264                      nfqnl_cmpfn cmpfn, unsigned long data)
 265 {
 266         struct nfqnl_queue_entry *entry;
 267
 268         entry = __find_entry(queue, cmpfn, data);
 269         if (entry == NULL)
 270                 return NULL;
 271
 272         __dequeue_entry(queue, entry);
 273         return entry;
 274 }
 275
 276
 277 static inline void
 278 __nfqnl_flush(struct nfqnl_instance *queue, int verdict)
 279 {
 280         struct nfqnl_queue_entry *entry;
 281
 282         while ((entry = __find_dequeue_entry(queue, NULL, 0)))
 283                 issue_verdict(entry, verdict);
 284 }
 285
 286 static inline int
 287 __nfqnl_set_mode(struct nfqnl_instance *queue,
 288                  unsigned char mode, unsigned int range)
 289 {
 290         int status = 0;
 291
 292         switch (mode) {
 293         case NFQNL_COPY_NONE:
 294         case NFQNL_COPY_META:
 295                 queue->copy_mode = mode;
 296                 queue->copy_range = 0;
 297                 break;
 298
 299         case NFQNL_COPY_PACKET:
 300                 queue->copy_mode = mode;
 301                 /* we're using struct nlattr which has 16bit nla_len */
 302                 if (range > 0xffff)
 303                         queue->copy_range = 0xffff;
 304                 else
 305                         queue->copy_range = range;
 306                 break;
 307
 308         default:
 309                 status = -EINVAL;
 310
 311         }
 312         return status;
 313 }
 314
 315 static struct nfqnl_queue_entry *
 316 find_dequeue_entry(struct nfqnl_instance *queue,
 317                          nfqnl_cmpfn cmpfn, unsigned long data)
 318 {
 319         struct nfqnl_queue_entry *entry;
 320
 321         spin_lock_bh(&queue->lock);
 322         entry = __find_dequeue_entry(queue, cmpfn, data);
 323         spin_unlock_bh(&queue->lock);
 324
 325         return entry;
 326 }
 327
 328 static void
 329 nfqnl_flush(struct nfqnl_instance *queue, int verdict)
 330 {
 331         spin_lock_bh(&queue->lock);
 332         __nfqnl_flush(queue, verdict);
 333         spin_unlock_bh(&queue->lock);
 334 }
 335
 336 static struct sk_buff *
 337 nfqnl_build_packet_message(struct nfqnl_instance *queue,
 338                            struct nfqnl_queue_entry *entry, int *errp)
 339 {
 340         sk_buff_data_t old_tail;
 341         size_t size;
 342         size_t data_len = 0;
 343         struct sk_buff *skb;
 344         struct nfqnl_msg_packet_hdr pmsg;
 345         struct nlmsghdr *nlh;
 346         struct nfgenmsg *nfmsg;
 347         struct nf_info *entinf = entry->info;
 348         struct sk_buff *entskb = entry->skb;
 349         struct net_device *indev;
 350         struct net_device *outdev;
 351         __be32 tmp_uint;
 352
 353         QDEBUG("entered\n");
 354
 355         size =    NLMSG_ALIGN(sizeof(struct nfgenmsg))
 356                 + nla_total_size(sizeof(struct nfqnl_msg_packet_hdr))
 357                 + nla_total_size(sizeof(u_int32_t))     /* ifindex */
 358                 + nla_total_size(sizeof(u_int32_t))     /* ifindex */
 359 #ifdef CONFIG_BRIDGE_NETFILTER
 360                 + nla_total_size(sizeof(u_int32_t))     /* ifindex */
 361                 + nla_total_size(sizeof(u_int32_t))     /* ifindex */
 362 #endif
 363                 + nla_total_size(sizeof(u_int32_t))     /* mark */
 364                 + nla_total_size(sizeof(struct nfqnl_msg_packet_hw))
 365                 + nla_total_size(sizeof(struct nfqnl_msg_packet_timestamp));
 366
 367         outdev = entinf->outdev;
 368
 369         spin_lock_bh(&queue->lock);
 370
 371         switch (queue->copy_mode) {
 372         case NFQNL_COPY_META:
 373         case NFQNL_COPY_NONE:
 374                 data_len = 0;
 375                 break;
 376
 377         case NFQNL_COPY_PACKET:
 378                 if ((entskb->ip_summed == CHECKSUM_PARTIAL ||
 379                      entskb->ip_summed == CHECKSUM_COMPLETE) &&
 380                     (*errp = skb_checksum_help(entskb))) {
 381                         spin_unlock_bh(&queue->lock);
 382                         return NULL;
 383                 }
 384                 if (queue->copy_range == 0
 385                     || queue->copy_range > entskb->len)
 386                         data_len = entskb->len;
 387                 else
 388                         data_len = queue->copy_range;
 389
 390                 size += nla_total_size(data_len);
 391                 break;
 392
 393         default:
 394                 *errp = -EINVAL;
 395                 spin_unlock_bh(&queue->lock);
 396                 return NULL;
 397         }
 398
 399         spin_unlock_bh(&queue->lock);
 400
 401         skb = alloc_skb(size, GFP_ATOMIC);
 402         if (!skb)
 403                 goto nlmsg_failure;
 404
 405         old_tail = skb->tail;
 406         nlh = NLMSG_PUT(skb, 0, 0,
 407                         NFNL_SUBSYS_QUEUE << 8 | NFQNL_MSG_PACKET,
 408                         sizeof(struct nfgenmsg));
 409         nfmsg = NLMSG_DATA(nlh);
 410         nfmsg->nfgen_family = entinf->pf;
 411         nfmsg->version = NFNETLINK_V0;
 412         nfmsg->res_id = htons(queue->queue_num);
 413
 414         pmsg.packet_id          = htonl(entry->id);
 415         pmsg.hw_protocol        = entskb->protocol;
 416         pmsg.hook               = entinf->hook;
 417
 418         NLA_PUT(skb, NFQA_PACKET_HDR, sizeof(pmsg), &pmsg);
 419
 420         indev = entinf->indev;
 421         if (indev) {
 422                 tmp_uint = htonl(indev->ifindex);
 423 #ifndef CONFIG_BRIDGE_NETFILTER
 424                 NLA_PUT(skb, NFQA_IFINDEX_INDEV, sizeof(tmp_uint), &tmp_uint);
 425 #else
 426                 if (entinf->pf == PF_BRIDGE) {
 427                         /* Case 1: indev is physical input device, we need to
 428                          * look for bridge group (when called from
 429                          * netfilter_bridge) */
 430                         NLA_PUT(skb, NFQA_IFINDEX_PHYSINDEV, sizeof(tmp_uint),
 431                                 &tmp_uint);
 432                         /* this is the bridge group "brX" */
 433                         tmp_uint = htonl(indev->br_port->br->dev->ifindex);
 434                         NLA_PUT(skb, NFQA_IFINDEX_INDEV, sizeof(tmp_uint),
 435                                 &tmp_uint);
 436                 } else {
 437                         /* Case 2: indev is bridge group, we need to look for
 438                          * physical device (when called from ipv4) */
 439                         NLA_PUT(skb, NFQA_IFINDEX_INDEV, sizeof(tmp_uint),
 440                                 &tmp_uint);
 441                         if (entskb->nf_bridge
 442                             && entskb->nf_bridge->physindev) {
 443                                 tmp_uint = htonl(entskb->nf_bridge->physindev->ifindex);
 444                                 NLA_PUT(skb, NFQA_IFINDEX_PHYSINDEV,
 445                                         sizeof(tmp_uint), &tmp_uint);
 446                         }
 447                 }
 448 #endif
 449         }
 450
 451         if (outdev) {
 452                 tmp_uint = htonl(outdev->ifindex);
 453 #ifndef CONFIG_BRIDGE_NETFILTER
 454                 NLA_PUT(skb, NFQA_IFINDEX_OUTDEV, sizeof(tmp_uint), &tmp_uint);
 455 #else
 456                 if (entinf->pf == PF_BRIDGE) {
 457                         /* Case 1: outdev is physical output device, we need to
 458                          * look for bridge group (when called from
 459                          * netfilter_bridge) */
 460                         NLA_PUT(skb, NFQA_IFINDEX_PHYSOUTDEV, sizeof(tmp_uint),
 461                                 &tmp_uint);
 462                         /* this is the bridge group "brX" */
 463                         tmp_uint = htonl(outdev->br_port->br->dev->ifindex);
 464                         NLA_PUT(skb, NFQA_IFINDEX_OUTDEV, sizeof(tmp_uint),
 465                                 &tmp_uint);
 466                 } else {
 467                         /* Case 2: outdev is bridge group, we need to look for
 468                          * physical output device (when called from ipv4) */
 469                         NLA_PUT(skb, NFQA_IFINDEX_OUTDEV, sizeof(tmp_uint),
 470                                 &tmp_uint);
 471                         if (entskb->nf_bridge
 472                             && entskb->nf_bridge->physoutdev) {
 473                                 tmp_uint = htonl(entskb->nf_bridge->physoutdev->ifindex);
 474                                 NLA_PUT(skb, NFQA_IFINDEX_PHYSOUTDEV,
 475                                         sizeof(tmp_uint), &tmp_uint);
 476                         }
 477                 }
 478 #endif
 479         }
 480
 481         if (entskb->mark) {
 482                 tmp_uint = htonl(entskb->mark);
 483                 NLA_PUT(skb, NFQA_MARK, sizeof(u_int32_t), &tmp_uint);
 484         }
 485
 486         if (indev && entskb->dev) {
 487                 struct nfqnl_msg_packet_hw phw;
 488                 int len = dev_parse_header(entskb, phw.hw_addr);
 489                 if (len) {
 490                         phw.hw_addrlen = htons(len);
 491                         NLA_PUT(skb, NFQA_HWADDR, sizeof(phw), &phw);
 492                 }
 493         }
 494
 495         if (entskb->tstamp.tv64) {
 496                 struct nfqnl_msg_packet_timestamp ts;
 497                 struct timeval tv = ktime_to_timeval(entskb->tstamp);
 498                 ts.sec = cpu_to_be64(tv.tv_sec);
 499                 ts.usec = cpu_to_be64(tv.tv_usec);
 500
 501                 NLA_PUT(skb, NFQA_TIMESTAMP, sizeof(ts), &ts);
 502         }
 503
 504         if (data_len) {
 505                 struct nlattr *nla;
 506                 int size = nla_attr_size(data_len);
 507
 508                 if (skb_tailroom(skb) < nla_total_size(data_len)) {
 509                         printk(KERN_WARNING "nf_queue: no tailroom!\n");
 510                         goto nlmsg_failure;
 511                 }
 512
 513                 nla = (struct nlattr *)skb_put(skb, nla_total_size(data_len));
 514                 nla->nla_type = NFQA_PAYLOAD;
 515                 nla->nla_len = size;
 516
 517                 if (skb_copy_bits(entskb, 0, nla_data(nla), data_len))
 518                         BUG();
 519         }
 520
 521         nlh->nlmsg_len = skb->tail - old_tail;
 522         return skb;
 523
 524 nlmsg_failure:
 525 nla_put_failure:
 526         if (skb)
 527                 kfree_skb(skb);
 528         *errp = -EINVAL;
 529         if (net_ratelimit())
 530                 printk(KERN_ERR "nf_queue: error creating packet message\n");
 531         return NULL;
 532 }
 533
 534 static int
 535 nfqnl_enqueue_packet(struct sk_buff *skb, struct nf_info *info,
 536                      unsigned int queuenum)
 537 {
 538         int status = -EINVAL;
 539         struct sk_buff *nskb;
 540         struct nfqnl_instance *queue;
 541         struct nfqnl_queue_entry *entry;
 542
 543         QDEBUG("entered\n");
 544
 545         queue = instance_lookup_get(queuenum);
 546         if (!queue) {
 547                 QDEBUG("no queue instance matching\n");
 548                 return -EINVAL;
 549         }
 550
 551         if (queue->copy_mode == NFQNL_COPY_NONE) {
 552                 QDEBUG("mode COPY_NONE, aborting\n");
 553                 status = -EAGAIN;
 554                 goto err_out_put;
 555         }
 556
 557         entry = kmalloc(sizeof(*entry), GFP_ATOMIC);
 558         if (entry == NULL) {
 559                 if (net_ratelimit())
 560                         printk(KERN_ERR
 561                                 "nf_queue: OOM in nfqnl_enqueue_packet()\n");
 562                 status = -ENOMEM;
 563                 goto err_out_put;
 564         }
 565
 566         entry->info = info;
 567         entry->skb = skb;
 568         entry->id = atomic_inc_return(&queue->id_sequence);
 569
 570         nskb = nfqnl_build_packet_message(queue, entry, &status);
 571         if (nskb == NULL)
 572                 goto err_out_free;
 573
 574         spin_lock_bh(&queue->lock);
 575
 576         if (!queue->peer_pid)
 577                 goto err_out_free_nskb;
 578
 579         if (queue->queue_total >= queue->queue_maxlen) {
 580                 queue->queue_dropped++;
 581                 status = -ENOSPC;
 582                 if (net_ratelimit())
 583                           printk(KERN_WARNING "nf_queue: full at %d entries, "
 584                                  "dropping packets(s). Dropped: %d\n",
 585                                  queue->queue_total, queue->queue_dropped);
 586                 goto err_out_free_nskb;
 587         }
 588
 589         /* nfnetlink_unicast will either free the nskb or add it to a socket */
 590         status = nfnetlink_unicast(nskb, queue->peer_pid, MSG_DONTWAIT);
 591         if (status < 0) {
 592                 queue->queue_user_dropped++;
 593                 goto err_out_unlock;
 594         }
 595
 596         __enqueue_entry(queue, entry);
 597
 598         spin_unlock_bh(&queue->lock);
 599         instance_put(queue);
 600         return status;
 601
 602 err_out_free_nskb:
 603         kfree_skb(nskb);
 604
 605 err_out_unlock:
 606         spin_unlock_bh(&queue->lock);
 607
 608 err_out_free:
 609         kfree(entry);
 610 err_out_put:
 611         instance_put(queue);
 612         return status;
 613 }
 614
 615 static int
 616 nfqnl_mangle(void *data, int data_len, struct nfqnl_queue_entry *e)
 617 {
 618         int diff;
 619         int err;
 620
 621         diff = data_len - e->skb->len;
 622         if (diff < 0) {
 623                 if (pskb_trim(e->skb, data_len))
 624                         return -ENOMEM;
 625         } else if (diff > 0) {
 626                 if (data_len > 0xFFFF)
 627                         return -EINVAL;
 628                 if (diff > skb_tailroom(e->skb)) {
 629                         err = pskb_expand_head(e->skb, 0,
 630                                                diff - skb_tailroom(e->skb),
 631                                                GFP_ATOMIC);
 632                         if (err) {
 633                                 printk(KERN_WARNING "nf_queue: OOM "
 634                                       "in mangle, dropping packet\n");
 635                                 return err;
 636                         }
 637                 }
 638                 skb_put(e->skb, diff);
 639         }
 640         if (!skb_make_writable(e->skb, data_len))
 641                 return -ENOMEM;
 642         skb_copy_to_linear_data(e->skb, data, data_len);
 643         e->skb->ip_summed = CHECKSUM_NONE;
 644         return 0;
 645 }
 646
 647 static inline int
 648 id_cmp(struct nfqnl_queue_entry *e, unsigned long id)
 649 {
 650         return (id == e->id);
 651 }
 652
 653 static int
 654 nfqnl_set_mode(struct nfqnl_instance *queue,
 655                unsigned char mode, unsigned int range)
 656 {
 657         int status;
 658
 659         spin_lock_bh(&queue->lock);
 660         status = __nfqnl_set_mode(queue, mode, range);
 661         spin_unlock_bh(&queue->lock);
 662
 663         return status;
 664 }
 665
 666 static int
 667 dev_cmp(struct nfqnl_queue_entry *entry, unsigned long ifindex)
 668 {
 669         struct nf_info *entinf = entry->info;
 670
 671         if (entinf->indev)
 672                 if (entinf->indev->ifindex == ifindex)
 673                         return 1;
 674         if (entinf->outdev)
 675                 if (entinf->outdev->ifindex == ifindex)
 676                         return 1;
 677 #ifdef CONFIG_BRIDGE_NETFILTER
 678         if (entry->skb->nf_bridge) {
 679                 if (entry->skb->nf_bridge->physindev &&
 680                     entry->skb->nf_bridge->physindev->ifindex == ifindex)
 681                         return 1;
 682                 if (entry->skb->nf_bridge->physoutdev &&
 683                     entry->skb->nf_bridge->physoutdev->ifindex == ifindex)
 684                         return 1;
 685         }
 686 #endif
 687         return 0;
 688 }
 689
 690 /* drop all packets with either indev or outdev == ifindex from all queue
 691  * instances */
 692 static void
 693 nfqnl_dev_drop(int ifindex)
 694 {
 695         int i;
 696
 697         QDEBUG("entering for ifindex %u\n", ifindex);
 698
 699         /* this only looks like we have to hold the readlock for a way too long
 700          * time, issue_verdict(),  nf_reinject(), ... - but we always only
 701          * issue NF_DROP, which is processed directly in nf_reinject() */
 702         read_lock_bh(&instances_lock);
 703
 704         for  (i = 0; i < INSTANCE_BUCKETS; i++) {
 705                 struct hlist_node *tmp;
 706                 struct nfqnl_instance *inst;
 707                 struct hlist_head *head = &instance_table[i];
 708
 709                 hlist_for_each_entry(inst, tmp, head, hlist) {
 710                         struct nfqnl_queue_entry *entry;
 711                         while ((entry = find_dequeue_entry(inst, dev_cmp,
 712                                                            ifindex)) != NULL)
 713                                 issue_verdict(entry, NF_DROP);
 714                 }
 715         }
 716
 717         read_unlock_bh(&instances_lock);
 718 }
 719
 720 #define RCV_SKB_FAIL(err) do { netlink_ack(skb, nlh, (err)); return; } while (0)
 721
 722 static int
 723 nfqnl_rcv_dev_event(struct notifier_block *this,
 724                     unsigned long event, void *ptr)
 725 {
 726         struct net_device *dev = ptr;
 727
 728         if (dev->nd_net != &init_net)
 729                 return NOTIFY_DONE;
 730
 731         /* Drop any packets associated with the downed device */
 732         if (event == NETDEV_DOWN)
 733                 nfqnl_dev_drop(dev->ifindex);
 734         return NOTIFY_DONE;
 735 }
 736
 737 static struct notifier_block nfqnl_dev_notifier = {
 738         .notifier_call  = nfqnl_rcv_dev_event,
 739 };
 740
 741 static int
 742 nfqnl_rcv_nl_event(struct notifier_block *this,
 743                    unsigned long event, void *ptr)
 744 {
 745         struct netlink_notify *n = ptr;
 746
 747         if (event == NETLINK_URELEASE &&
 748             n->protocol == NETLINK_NETFILTER && n->pid) {
 749                 int i;
 750
 751                 /* destroy all instances for this pid */
 752                 write_lock_bh(&instances_lock);
 753                 for  (i = 0; i < INSTANCE_BUCKETS; i++) {
 754                         struct hlist_node *tmp, *t2;
 755                         struct nfqnl_instance *inst;
 756                         struct hlist_head *head = &instance_table[i];
 757
 758                         hlist_for_each_entry_safe(inst, tmp, t2, head, hlist) {
 759                                 if ((n->net == &init_net) &&
 760                                     (n->pid == inst->peer_pid))
 761                                         __instance_destroy(inst);
 762                         }
 763                 }
 764                 write_unlock_bh(&instances_lock);
 765         }
 766         return NOTIFY_DONE;
 767 }
 768
 769 static struct notifier_block nfqnl_rtnl_notifier = {
 770         .notifier_call  = nfqnl_rcv_nl_event,
 771 };
 772
 773 static const struct nla_policy nfqa_verdict_policy[NFQA_MAX+1] = {
 774         [NFQA_VERDICT_HDR]      = { .len = sizeof(struct nfqnl_msg_verdict_hdr) },
 775         [NFQA_MARK]             = { .type = NLA_U32 },
 776         [NFQA_PAYLOAD]          = { .type = NLA_UNSPEC },
 777 };
 778
 779 static int
 780 nfqnl_recv_verdict(struct sock *ctnl, struct sk_buff *skb,
 781                    struct nlmsghdr *nlh, struct nlattr *nfqa[])
 782 {
 783         struct nfgenmsg *nfmsg = NLMSG_DATA(nlh);
 784         u_int16_t queue_num = ntohs(nfmsg->res_id);
 785
 786         struct nfqnl_msg_verdict_hdr *vhdr;
 787         struct nfqnl_instance *queue;
 788         unsigned int verdict;
 789         struct nfqnl_queue_entry *entry;
 790         int err;
 791
 792         queue = instance_lookup_get(queue_num);
 793         if (!queue)
 794                 return -ENODEV;
 795
 796         if (queue->peer_pid != NETLINK_CB(skb).pid) {
 797                 err = -EPERM;
 798                 goto err_out_put;
 799         }
 800
 801         if (!nfqa[NFQA_VERDICT_HDR]) {
 802                 err = -EINVAL;
 803                 goto err_out_put;
 804         }
 805
 806         vhdr = nla_data(nfqa[NFQA_VERDICT_HDR]);
 807         verdict = ntohl(vhdr->verdict);
 808
 809         if ((verdict & NF_VERDICT_MASK) > NF_MAX_VERDICT) {
 810                 err = -EINVAL;
 811                 goto err_out_put;
 812         }
 813
 814         entry = find_dequeue_entry(queue, id_cmp, ntohl(vhdr->id));
 815         if (entry == NULL) {
 816                 err = -ENOENT;
 817                 goto err_out_put;
 818         }
 819
 820         if (nfqa[NFQA_PAYLOAD]) {
 821                 if (nfqnl_mangle(nla_data(nfqa[NFQA_PAYLOAD]),
 822                                  nla_len(nfqa[NFQA_PAYLOAD]), entry) < 0)
 823                         verdict = NF_DROP;
 824         }
 825
 826         if (nfqa[NFQA_MARK])
 827                 entry->skb->mark = ntohl(*(__be32 *)
 828                                          nla_data(nfqa[NFQA_MARK]));
 829
 830         issue_verdict(entry, verdict);
 831         instance_put(queue);
 832         return 0;
 833
 834 err_out_put:
 835         instance_put(queue);
 836         return err;
 837 }
 838
 839 static int
 840 nfqnl_recv_unsupp(struct sock *ctnl, struct sk_buff *skb,
 841                   struct nlmsghdr *nlh, struct nlattr *nfqa[])
 842 {
 843         return -ENOTSUPP;
 844 }
 845
 846 static const struct nla_policy nfqa_cfg_policy[NFQA_CFG_MAX+1] = {
 847         [NFQA_CFG_CMD]          = { .len = sizeof(struct nfqnl_msg_config_cmd) },
 848         [NFQA_CFG_PARAMS]       = { .len = sizeof(struct nfqnl_msg_config_params) },
 849 };
 850
 851 static const struct nf_queue_handler nfqh = {
 852         .name   = "nf_queue",
 853         .outfn  = &nfqnl_enqueue_packet,
 854 };
 855
 856 static int
 857 nfqnl_recv_config(struct sock *ctnl, struct sk_buff *skb,
 858                   struct nlmsghdr *nlh, struct nlattr *nfqa[])
 859 {
 860         struct nfgenmsg *nfmsg = NLMSG_DATA(nlh);
 861         u_int16_t queue_num = ntohs(nfmsg->res_id);
 862         struct nfqnl_instance *queue;
 863         int ret = 0;
 864
 865         QDEBUG("entering for msg %u\n", NFNL_MSG_TYPE(nlh->nlmsg_type));
 866
 867         queue = instance_lookup_get(queue_num);
 868         if (nfqa[NFQA_CFG_CMD]) {
 869                 struct nfqnl_msg_config_cmd *cmd;
 870                 cmd = nla_data(nfqa[NFQA_CFG_CMD]);
 871                 QDEBUG("found CFG_CMD\n");
 872
 873                 switch (cmd->command) {
 874                 case NFQNL_CFG_CMD_BIND:
 875                         if (queue)
 876                                 return -EBUSY;
 877
 878                         queue = instance_create(queue_num, NETLINK_CB(skb).pid);
 879                         if (!queue)
 880                                 return -EINVAL;
 881                         break;
 882                 case NFQNL_CFG_CMD_UNBIND:
 883                         if (!queue)
 884                                 return -ENODEV;
 885
 886                         if (queue->peer_pid != NETLINK_CB(skb).pid) {
 887                                 ret = -EPERM;
 888                                 goto out_put;
 889                         }
 890
 891                         instance_destroy(queue);
 892                         break;
 893                 case NFQNL_CFG_CMD_PF_BIND:
 894                         QDEBUG("registering queue handler for pf=%u\n",
 895                                 ntohs(cmd->pf));
 896                         ret = nf_register_queue_handler(ntohs(cmd->pf), &nfqh);
 897                         break;
 898                 case NFQNL_CFG_CMD_PF_UNBIND:
 899                         QDEBUG("unregistering queue handler for pf=%u\n",
 900                                 ntohs(cmd->pf));
 901                         ret = nf_unregister_queue_handler(ntohs(cmd->pf), &nfqh);
 902                         break;
 903                 default:
 904                         ret = -EINVAL;
 905                         break;
 906                 }
 907         } else {
 908                 if (!queue) {
 909                         QDEBUG("no config command, and no instance ENOENT\n");
 910                         ret = -ENOENT;
 911                         goto out_put;
 912                 }
 913
 914                 if (queue->peer_pid != NETLINK_CB(skb).pid) {
 915                         QDEBUG("no config command, and wrong pid\n");
 916                         ret = -EPERM;
 917                         goto out_put;
 918                 }
 919         }
 920
 921         if (nfqa[NFQA_CFG_PARAMS]) {
 922                 struct nfqnl_msg_config_params *params;
 923
 924                 if (!queue) {
 925                         ret = -ENOENT;
 926                         goto out_put;
 927                 }
 928                 params = nla_data(nfqa[NFQA_CFG_PARAMS]);
 929                 nfqnl_set_mode(queue, params->copy_mode,
 930                                 ntohl(params->copy_range));
 931         }
 932
 933         if (nfqa[NFQA_CFG_QUEUE_MAXLEN]) {
 934                 __be32 *queue_maxlen;
 935                 queue_maxlen = nla_data(nfqa[NFQA_CFG_QUEUE_MAXLEN]);
 936                 spin_lock_bh(&queue->lock);
 937                 queue->queue_maxlen = ntohl(*queue_maxlen);
 938                 spin_unlock_bh(&queue->lock);
 939         }
 940
 941 out_put:
 942         instance_put(queue);
 943         return ret;
 944 }
 945
 946 static const struct nfnl_callback nfqnl_cb[NFQNL_MSG_MAX] = {
 947         [NFQNL_MSG_PACKET]      = { .call = nfqnl_recv_unsupp,
 948                                     .attr_count = NFQA_MAX, },
 949         [NFQNL_MSG_VERDICT]     = { .call = nfqnl_recv_verdict,
 950                                     .attr_count = NFQA_MAX,
 951                                     .policy = nfqa_verdict_policy },
 952         [NFQNL_MSG_CONFIG]      = { .call = nfqnl_recv_config,
 953                                     .attr_count = NFQA_CFG_MAX,
 954                                     .policy = nfqa_cfg_policy },
 955 };
 956
 957 static const struct nfnetlink_subsystem nfqnl_subsys = {
 958         .name           = "nf_queue",
 959         .subsys_id      = NFNL_SUBSYS_QUEUE,
 960         .cb_count       = NFQNL_MSG_MAX,
 961         .cb             = nfqnl_cb,
 962 };
 963
 964 #ifdef CONFIG_PROC_FS
 965 struct iter_state {
 966         unsigned int bucket;
 967 };
 968
 969 static struct hlist_node *get_first(struct seq_file *seq)
 970 {
 971         struct iter_state *st = seq->private;
 972
 973         if (!st)
 974                 return NULL;
 975
 976         for (st->bucket = 0; st->bucket < INSTANCE_BUCKETS; st->bucket++) {
 977                 if (!hlist_empty(&instance_table[st->bucket]))
 978                         return instance_table[st->bucket].first;
 979         }
 980         return NULL;
 981 }
 982
 983 static struct hlist_node *get_next(struct seq_file *seq, struct hlist_node *h)
 984 {
 985         struct iter_state *st = seq->private;
 986
 987         h = h->next;
 988         while (!h) {
 989                 if (++st->bucket >= INSTANCE_BUCKETS)
 990                         return NULL;
 991
 992                 h = instance_table[st->bucket].first;
 993         }
 994         return h;
 995 }
 996
 997 static struct hlist_node *get_idx(struct seq_file *seq, loff_t pos)
 998 {
 999         struct hlist_node *head;
1000         head = get_first(seq);
1001
1002         if (head)
1003                 while (pos && (head = get_next(seq, head)))
1004                         pos--;
1005         return pos ? NULL : head;
1006 }
1007
1008 static void *seq_start(struct seq_file *seq, loff_t *pos)
1009 {
1010         read_lock_bh(&instances_lock);
1011         return get_idx(seq, *pos);
1012 }
1013
1014 static void *seq_next(struct seq_file *s, void *v, loff_t *pos)
1015 {
1016         (*pos)++;
1017         return get_next(s, v);
1018 }
1019
1020 static void seq_stop(struct seq_file *s, void *v)
1021 {
1022         read_unlock_bh(&instances_lock);
1023 }
1024
1025 static int seq_show(struct seq_file *s, void *v)
1026 {
1027         const struct nfqnl_instance *inst = v;
1028
1029         return seq_printf(s, "%5d %6d %5d %1d %5d %5d %5d %8d %2d\n",
1030                           inst->queue_num,
1031                           inst->peer_pid, inst->queue_total,
1032                           inst->copy_mode, inst->copy_range,
1033                           inst->queue_dropped, inst->queue_user_dropped,
1034                           atomic_read(&inst->id_sequence),
1035                           atomic_read(&inst->use));
1036 }
1037
1038 static const struct seq_operations nfqnl_seq_ops = {
1039         .start  = seq_start,
1040         .next   = seq_next,
1041         .stop   = seq_stop,
1042         .show   = seq_show,
1043 };
1044
1045 static int nfqnl_open(struct inode *inode, struct file *file)
1046 {
1047         return seq_open_private(file, &nfqnl_seq_ops,
1048                         sizeof(struct iter_state));
1049 }
1050
1051 static const struct file_operations nfqnl_file_ops = {
1052         .owner   = THIS_MODULE,
1053         .open    = nfqnl_open,
1054         .read    = seq_read,
1055         .llseek  = seq_lseek,
1056         .release = seq_release_private,
1057 };
1058
1059 #endif /* PROC_FS */
1060
1061 static int __init nfnetlink_queue_init(void)
1062 {
1063         int i, status = -ENOMEM;
1064 #ifdef CONFIG_PROC_FS
1065         struct proc_dir_entry *proc_nfqueue;
1066 #endif
1067
1068         for (i = 0; i < INSTANCE_BUCKETS; i++)
1069                 INIT_HLIST_HEAD(&instance_table[i]);
1070
1071         netlink_register_notifier(&nfqnl_rtnl_notifier);
1072         status = nfnetlink_subsys_register(&nfqnl_subsys);
1073         if (status < 0) {
1074                 printk(KERN_ERR "nf_queue: failed to create netlink socket\n");
1075                 goto cleanup_netlink_notifier;
1076         }
1077
1078 #ifdef CONFIG_PROC_FS
1079         proc_nfqueue = create_proc_entry("nfnetlink_queue", 0440,
1080                                          proc_net_netfilter);
1081         if (!proc_nfqueue)
1082                 goto cleanup_subsys;
1083         proc_nfqueue->proc_fops = &nfqnl_file_ops;
1084 #endif
1085
1086         register_netdevice_notifier(&nfqnl_dev_notifier);
1087         return status;
1088
1089 #ifdef CONFIG_PROC_FS
1090 cleanup_subsys:
1091         nfnetlink_subsys_unregister(&nfqnl_subsys);
1092 #endif
1093 cleanup_netlink_notifier:
1094         netlink_unregister_notifier(&nfqnl_rtnl_notifier);
1095         return status;
1096 }
1097
1098 static void __exit nfnetlink_queue_fini(void)
1099 {
1100         nf_unregister_queue_handlers(&nfqh);
1101         unregister_netdevice_notifier(&nfqnl_dev_notifier);
1102 #ifdef CONFIG_PROC_FS
1103         remove_proc_entry("nfnetlink_queue", proc_net_netfilter);
1104 #endif
1105         nfnetlink_subsys_unregister(&nfqnl_subsys);
1106         netlink_unregister_notifier(&nfqnl_rtnl_notifier);
1107 }
1108
1109 MODULE_DESCRIPTION("netfilter packet queue handler");
1110 MODULE_AUTHOR("Harald Welte <laforge@netfilter.org>");
1111 MODULE_LICENSE("GPL");
1112 MODULE_ALIAS_NFNL_SUBSYS(NFNL_SUBSYS_QUEUE);
1113
1114 module_init(nfnetlink_queue_init);
1115 module_exit(nfnetlink_queue_fini);