net/netfilter/nfnetlink_queue.c

   1 /*
   2  * This is a module which is used for queueing packets and communicating with
   3  * userspace via nfetlink.
   4  *
   5  * (C) 2005 by Harald Welte <laforge@netfilter.org>
   6  *
   7  * Based on the old ipv4-only ip_queue.c:
   8  * (C) 2000-2002 James Morris <jmorris@intercode.com.au>
   9  * (C) 2003-2005 Netfilter Core Team <coreteam@netfilter.org>
  10  *
  11  * This program is free software; you can redistribute it and/or modify
  12  * it under the terms of the GNU General Public License version 2 as
  13  * published by the Free Software Foundation.
  14  *
  15  */
  16 #include <linux/module.h>
  17 #include <linux/skbuff.h>
  18 #include <linux/init.h>
  19 #include <linux/spinlock.h>
  20 #include <linux/notifier.h>
  21 #include <linux/netdevice.h>
  22 #include <linux/netfilter.h>
  23 #include <linux/proc_fs.h>
  24 #include <linux/netfilter_ipv4.h>
  25 #include <linux/netfilter_ipv6.h>
  26 #include <linux/netfilter/nfnetlink.h>
  27 #include <linux/netfilter/nfnetlink_queue.h>
  28 #include <linux/list.h>
  29 #include <net/sock.h>
  30
  31 #include <asm/atomic.h>
  32
  33 #ifdef CONFIG_BRIDGE_NETFILTER
  34 #include "../bridge/br_private.h"
  35 #endif
  36
  37 #define NFQNL_QMAX_DEFAULT 1024
  38
  39 #if 0
  40 #define QDEBUG(x, args ...)     printk(KERN_DEBUG "%s(%d):%s(): " x,       \
  41                                         __FILE__, __LINE__, __FUNCTION__,  \
  42                                         ## args)
  43 #else
  44 #define QDEBUG(x, ...)
  45 #endif
  46
  47 struct nfqnl_queue_entry {
  48         struct list_head list;
  49         struct nf_info *info;
  50         struct sk_buff *skb;
  51         unsigned int id;
  52 };
  53
  54 struct nfqnl_instance {
  55         struct hlist_node hlist;                /* global list of queues */
  56         atomic_t use;
  57
  58         int peer_pid;
  59         unsigned int queue_maxlen;
  60         unsigned int copy_range;
  61         unsigned int queue_total;
  62         unsigned int queue_dropped;
  63         unsigned int queue_user_dropped;
  64
  65         atomic_t id_sequence;                   /* 'sequence' of pkt ids */
  66
  67         u_int16_t queue_num;                    /* number of this queue */
  68         u_int8_t copy_mode;
  69
  70         spinlock_t lock;
  71
  72         struct list_head queue_list;            /* packets in queue */
  73 };
  74
  75 typedef int (*nfqnl_cmpfn)(struct nfqnl_queue_entry *, unsigned long);
  76
  77 static DEFINE_RWLOCK(instances_lock);
  78
  79 #define INSTANCE_BUCKETS        16
  80 static struct hlist_head instance_table[INSTANCE_BUCKETS];
  81
  82 static inline u_int8_t instance_hashfn(u_int16_t queue_num)
  83 {
  84         return ((queue_num >> 8) | queue_num) % INSTANCE_BUCKETS;
  85 }
  86
  87 static struct nfqnl_instance *
  88 __instance_lookup(u_int16_t queue_num)
  89 {
  90         struct hlist_head *head;
  91         struct hlist_node *pos;
  92         struct nfqnl_instance *inst;
  93
  94         head = &instance_table[instance_hashfn(queue_num)];
  95         hlist_for_each_entry(inst, pos, head, hlist) {
  96                 if (inst->queue_num == queue_num)
  97                         return inst;
  98         }
  99         return NULL;
 100 }
 101
 102 static struct nfqnl_instance *
 103 instance_lookup_get(u_int16_t queue_num)
 104 {
 105         struct nfqnl_instance *inst;
 106
 107         read_lock_bh(&instances_lock);
 108         inst = __instance_lookup(queue_num);
 109         if (inst)
 110                 atomic_inc(&inst->use);
 111         read_unlock_bh(&instances_lock);
 112
 113         return inst;
 114 }
 115
 116 static void
 117 instance_put(struct nfqnl_instance *inst)
 118 {
 119         if (inst && atomic_dec_and_test(&inst->use)) {
 120                 QDEBUG("kfree(inst=%p)\n", inst);
 121                 kfree(inst);
 122         }
 123 }
 124
 125 static struct nfqnl_instance *
 126 instance_create(u_int16_t queue_num, int pid)
 127 {
 128         struct nfqnl_instance *inst;
 129
 130         QDEBUG("entering for queue_num=%u, pid=%d\n", queue_num, pid);
 131
 132         write_lock_bh(&instances_lock);
 133         if (__instance_lookup(queue_num)) {
 134                 inst = NULL;
 135                 QDEBUG("aborting, instance already exists\n");
 136                 goto out_unlock;
 137         }
 138
 139         inst = kzalloc(sizeof(*inst), GFP_ATOMIC);
 140         if (!inst)
 141                 goto out_unlock;
 142
 143         inst->queue_num = queue_num;
 144         inst->peer_pid = pid;
 145         inst->queue_maxlen = NFQNL_QMAX_DEFAULT;
 146         inst->copy_range = 0xfffff;
 147         inst->copy_mode = NFQNL_COPY_NONE;
 148         atomic_set(&inst->id_sequence, 0);
 149         /* needs to be two, since we _put() after creation */
 150         atomic_set(&inst->use, 2);
 151         spin_lock_init(&inst->lock);
 152         INIT_LIST_HEAD(&inst->queue_list);
 153
 154         if (!try_module_get(THIS_MODULE))
 155                 goto out_free;
 156
 157         hlist_add_head(&inst->hlist,
 158                        &instance_table[instance_hashfn(queue_num)]);
 159
 160         write_unlock_bh(&instances_lock);
 161
 162         QDEBUG("successfully created new instance\n");
 163
 164         return inst;
 165
 166 out_free:
 167         kfree(inst);
 168 out_unlock:
 169         write_unlock_bh(&instances_lock);
 170         return NULL;
 171 }
 172
 173 static void nfqnl_flush(struct nfqnl_instance *queue, int verdict);
 174
 175 static void
 176 _instance_destroy2(struct nfqnl_instance *inst, int lock)
 177 {
 178         /* first pull it out of the global list */
 179         if (lock)
 180                 write_lock_bh(&instances_lock);
 181
 182         QDEBUG("removing instance %p (queuenum=%u) from hash\n",
 183                 inst, inst->queue_num);
 184         hlist_del(&inst->hlist);
 185
 186         if (lock)
 187                 write_unlock_bh(&instances_lock);
 188
 189         /* then flush all pending skbs from the queue */
 190         nfqnl_flush(inst, NF_DROP);
 191
 192         /* and finally put the refcount */
 193         instance_put(inst);
 194
 195         module_put(THIS_MODULE);
 196 }
 197
 198 static inline void
 199 __instance_destroy(struct nfqnl_instance *inst)
 200 {
 201         _instance_destroy2(inst, 0);
 202 }
 203
 204 static inline void
 205 instance_destroy(struct nfqnl_instance *inst)
 206 {
 207         _instance_destroy2(inst, 1);
 208 }
 209
 210
 211
 212 static void
 213 issue_verdict(struct nfqnl_queue_entry *entry, int verdict)
 214 {
 215         QDEBUG("entering for entry %p, verdict %u\n", entry, verdict);
 216
 217         /* TCP input path (and probably other bits) assume to be called
 218          * from softirq context, not from syscall, like issue_verdict is
 219          * called.  TCP input path deadlocks with locks taken from timer
 220          * softirq, e.g.  We therefore emulate this by local_bh_disable() */
 221
 222         local_bh_disable();
 223         nf_reinject(entry->skb, entry->info, verdict);
 224         local_bh_enable();
 225
 226         kfree(entry);
 227 }
 228
 229 static inline void
 230 __enqueue_entry(struct nfqnl_instance *queue,
 231                       struct nfqnl_queue_entry *entry)
 232 {
 233        list_add(&entry->list, &queue->queue_list);
 234        queue->queue_total++;
 235 }
 236
 237 /*
 238  * Find and return a queued entry matched by cmpfn, or return the last
 239  * entry if cmpfn is NULL.
 240  */
 241 static inline struct nfqnl_queue_entry *
 242 __find_entry(struct nfqnl_instance *queue, nfqnl_cmpfn cmpfn,
 243                    unsigned long data)
 244 {
 245         struct list_head *p;
 246
 247         list_for_each_prev(p, &queue->queue_list) {
 248                 struct nfqnl_queue_entry *entry = (struct nfqnl_queue_entry *)p;
 249
 250                 if (!cmpfn || cmpfn(entry, data))
 251                         return entry;
 252         }
 253         return NULL;
 254 }
 255
 256 static inline void
 257 __dequeue_entry(struct nfqnl_instance *q, struct nfqnl_queue_entry *entry)
 258 {
 259         list_del(&entry->list);
 260         q->queue_total--;
 261 }
 262
 263 static inline struct nfqnl_queue_entry *
 264 __find_dequeue_entry(struct nfqnl_instance *queue,
 265                      nfqnl_cmpfn cmpfn, unsigned long data)
 266 {
 267         struct nfqnl_queue_entry *entry;
 268
 269         entry = __find_entry(queue, cmpfn, data);
 270         if (entry == NULL)
 271                 return NULL;
 272
 273         __dequeue_entry(queue, entry);
 274         return entry;
 275 }
 276
 277
 278 static inline void
 279 __nfqnl_flush(struct nfqnl_instance *queue, int verdict)
 280 {
 281         struct nfqnl_queue_entry *entry;
 282
 283         while ((entry = __find_dequeue_entry(queue, NULL, 0)))
 284                 issue_verdict(entry, verdict);
 285 }
 286
 287 static inline int
 288 __nfqnl_set_mode(struct nfqnl_instance *queue,
 289                  unsigned char mode, unsigned int range)
 290 {
 291         int status = 0;
 292
 293         switch (mode) {
 294         case NFQNL_COPY_NONE:
 295         case NFQNL_COPY_META:
 296                 queue->copy_mode = mode;
 297                 queue->copy_range = 0;
 298                 break;
 299
 300         case NFQNL_COPY_PACKET:
 301                 queue->copy_mode = mode;
 302                 /* we're using struct nfattr which has 16bit nfa_len */
 303                 if (range > 0xffff)
 304                         queue->copy_range = 0xffff;
 305                 else
 306                         queue->copy_range = range;
 307                 break;
 308
 309         default:
 310                 status = -EINVAL;
 311
 312         }
 313         return status;
 314 }
 315
 316 static struct nfqnl_queue_entry *
 317 find_dequeue_entry(struct nfqnl_instance *queue,
 318                          nfqnl_cmpfn cmpfn, unsigned long data)
 319 {
 320         struct nfqnl_queue_entry *entry;
 321
 322         spin_lock_bh(&queue->lock);
 323         entry = __find_dequeue_entry(queue, cmpfn, data);
 324         spin_unlock_bh(&queue->lock);
 325
 326         return entry;
 327 }
 328
 329 static void
 330 nfqnl_flush(struct nfqnl_instance *queue, int verdict)
 331 {
 332         spin_lock_bh(&queue->lock);
 333         __nfqnl_flush(queue, verdict);
 334         spin_unlock_bh(&queue->lock);
 335 }
 336
 337 static struct sk_buff *
 338 nfqnl_build_packet_message(struct nfqnl_instance *queue,
 339                            struct nfqnl_queue_entry *entry, int *errp)
 340 {
 341         unsigned char *old_tail;
 342         size_t size;
 343         size_t data_len = 0;
 344         struct sk_buff *skb;
 345         struct nfqnl_msg_packet_hdr pmsg;
 346         struct nlmsghdr *nlh;
 347         struct nfgenmsg *nfmsg;
 348         struct nf_info *entinf = entry->info;
 349         struct sk_buff *entskb = entry->skb;
 350         struct net_device *indev;
 351         struct net_device *outdev;
 352         unsigned int tmp_uint;
 353
 354         QDEBUG("entered\n");
 355
 356         /* all macros expand to constant values at compile time */
 357         size =    NLMSG_SPACE(sizeof(struct nfgenmsg)) +
 358                 + NFA_SPACE(sizeof(struct nfqnl_msg_packet_hdr))
 359                 + NFA_SPACE(sizeof(u_int32_t))  /* ifindex */
 360                 + NFA_SPACE(sizeof(u_int32_t))  /* ifindex */
 361 #ifdef CONFIG_BRIDGE_NETFILTER
 362                 + NFA_SPACE(sizeof(u_int32_t))  /* ifindex */
 363                 + NFA_SPACE(sizeof(u_int32_t))  /* ifindex */
 364 #endif
 365                 + NFA_SPACE(sizeof(u_int32_t))  /* mark */
 366                 + NFA_SPACE(sizeof(struct nfqnl_msg_packet_hw))
 367                 + NFA_SPACE(sizeof(struct nfqnl_msg_packet_timestamp));
 368
 369         outdev = entinf->outdev;
 370
 371         spin_lock_bh(&queue->lock);
 372
 373         switch (queue->copy_mode) {
 374         case NFQNL_COPY_META:
 375         case NFQNL_COPY_NONE:
 376                 data_len = 0;
 377                 break;
 378
 379         case NFQNL_COPY_PACKET:
 380                 if (entskb->ip_summed == CHECKSUM_HW &&
 381                     (*errp = skb_checksum_help(entskb,
 382                                                outdev == NULL))) {
 383                         spin_unlock_bh(&queue->lock);
 384                         return NULL;
 385                 }
 386                 if (queue->copy_range == 0
 387                     || queue->copy_range > entskb->len)
 388                         data_len = entskb->len;
 389                 else
 390                         data_len = queue->copy_range;
 391
 392                 size += NFA_SPACE(data_len);
 393                 break;
 394
 395         default:
 396                 *errp = -EINVAL;
 397                 spin_unlock_bh(&queue->lock);
 398                 return NULL;
 399         }
 400
 401         spin_unlock_bh(&queue->lock);
 402
 403         skb = alloc_skb(size, GFP_ATOMIC);
 404         if (!skb)
 405                 goto nlmsg_failure;
 406
 407         old_tail= skb->tail;
 408         nlh = NLMSG_PUT(skb, 0, 0,
 409                         NFNL_SUBSYS_QUEUE << 8 | NFQNL_MSG_PACKET,
 410                         sizeof(struct nfgenmsg));
 411         nfmsg = NLMSG_DATA(nlh);
 412         nfmsg->nfgen_family = entinf->pf;
 413         nfmsg->version = NFNETLINK_V0;
 414         nfmsg->res_id = htons(queue->queue_num);
 415
 416         pmsg.packet_id          = htonl(entry->id);
 417         pmsg.hw_protocol        = htons(entskb->protocol);
 418         pmsg.hook               = entinf->hook;
 419
 420         NFA_PUT(skb, NFQA_PACKET_HDR, sizeof(pmsg), &pmsg);
 421
 422         indev = entinf->indev;
 423         if (indev) {
 424                 tmp_uint = htonl(indev->ifindex);
 425 #ifndef CONFIG_BRIDGE_NETFILTER
 426                 NFA_PUT(skb, NFQA_IFINDEX_INDEV, sizeof(tmp_uint), &tmp_uint);
 427 #else
 428                 if (entinf->pf == PF_BRIDGE) {
 429                         /* Case 1: indev is physical input device, we need to
 430                          * look for bridge group (when called from
 431                          * netfilter_bridge) */
 432                         NFA_PUT(skb, NFQA_IFINDEX_PHYSINDEV, sizeof(tmp_uint),
 433                                 &tmp_uint);
 434                         /* this is the bridge group "brX" */
 435                         tmp_uint = htonl(indev->br_port->br->dev->ifindex);
 436                         NFA_PUT(skb, NFQA_IFINDEX_INDEV, sizeof(tmp_uint),
 437                                 &tmp_uint);
 438                 } else {
 439                         /* Case 2: indev is bridge group, we need to look for
 440                          * physical device (when called from ipv4) */
 441                         NFA_PUT(skb, NFQA_IFINDEX_INDEV, sizeof(tmp_uint),
 442                                 &tmp_uint);
 443                         if (entskb->nf_bridge
 444                             && entskb->nf_bridge->physindev) {
 445                                 tmp_uint = htonl(entskb->nf_bridge->physindev->ifindex);
 446                                 NFA_PUT(skb, NFQA_IFINDEX_PHYSINDEV,
 447                                         sizeof(tmp_uint), &tmp_uint);
 448                         }
 449                 }
 450 #endif
 451         }
 452
 453         if (outdev) {
 454                 tmp_uint = htonl(outdev->ifindex);
 455 #ifndef CONFIG_BRIDGE_NETFILTER
 456                 NFA_PUT(skb, NFQA_IFINDEX_OUTDEV, sizeof(tmp_uint), &tmp_uint);
 457 #else
 458                 if (entinf->pf == PF_BRIDGE) {
 459                         /* Case 1: outdev is physical output device, we need to
 460                          * look for bridge group (when called from
 461                          * netfilter_bridge) */
 462                         NFA_PUT(skb, NFQA_IFINDEX_PHYSOUTDEV, sizeof(tmp_uint),
 463                                 &tmp_uint);
 464                         /* this is the bridge group "brX" */
 465                         tmp_uint = htonl(outdev->br_port->br->dev->ifindex);
 466                         NFA_PUT(skb, NFQA_IFINDEX_OUTDEV, sizeof(tmp_uint),
 467                                 &tmp_uint);
 468                 } else {
 469                         /* Case 2: outdev is bridge group, we need to look for
 470                          * physical output device (when called from ipv4) */
 471                         NFA_PUT(skb, NFQA_IFINDEX_OUTDEV, sizeof(tmp_uint),
 472                                 &tmp_uint);
 473                         if (entskb->nf_bridge
 474                             && entskb->nf_bridge->physoutdev) {
 475                                 tmp_uint = htonl(entskb->nf_bridge->physoutdev->ifindex);
 476                                 NFA_PUT(skb, NFQA_IFINDEX_PHYSOUTDEV,
 477                                         sizeof(tmp_uint), &tmp_uint);
 478                         }
 479                 }
 480 #endif
 481         }
 482
 483         if (entskb->nfmark) {
 484                 tmp_uint = htonl(entskb->nfmark);
 485                 NFA_PUT(skb, NFQA_MARK, sizeof(u_int32_t), &tmp_uint);
 486         }
 487
 488         if (indev && entskb->dev
 489             && entskb->dev->hard_header_parse) {
 490                 struct nfqnl_msg_packet_hw phw;
 491
 492                 phw.hw_addrlen =
 493                         entskb->dev->hard_header_parse(entskb,
 494                                                            phw.hw_addr);
 495                 phw.hw_addrlen = htons(phw.hw_addrlen);
 496                 NFA_PUT(skb, NFQA_HWADDR, sizeof(phw), &phw);
 497         }
 498
 499         if (entskb->tstamp.off_sec) {
 500                 struct nfqnl_msg_packet_timestamp ts;
 501
 502                 ts.sec = cpu_to_be64(entskb->tstamp.off_sec);
 503                 ts.usec = cpu_to_be64(entskb->tstamp.off_usec);
 504
 505                 NFA_PUT(skb, NFQA_TIMESTAMP, sizeof(ts), &ts);
 506         }
 507
 508         if (data_len) {
 509                 struct nfattr *nfa;
 510                 int size = NFA_LENGTH(data_len);
 511
 512                 if (skb_tailroom(skb) < (int)NFA_SPACE(data_len)) {
 513                         printk(KERN_WARNING "nf_queue: no tailroom!\n");
 514                         goto nlmsg_failure;
 515                 }
 516
 517                 nfa = (struct nfattr *)skb_put(skb, NFA_ALIGN(size));
 518                 nfa->nfa_type = NFQA_PAYLOAD;
 519                 nfa->nfa_len = size;
 520
 521                 if (skb_copy_bits(entskb, 0, NFA_DATA(nfa), data_len))
 522                         BUG();
 523         }
 524
 525         nlh->nlmsg_len = skb->tail - old_tail;
 526         return skb;
 527
 528 nlmsg_failure:
 529 nfattr_failure:
 530         if (skb)
 531                 kfree_skb(skb);
 532         *errp = -EINVAL;
 533         if (net_ratelimit())
 534                 printk(KERN_ERR "nf_queue: error creating packet message\n");
 535         return NULL;
 536 }
 537
 538 static int
 539 nfqnl_enqueue_packet(struct sk_buff *skb, struct nf_info *info,
 540                      unsigned int queuenum, void *data)
 541 {
 542         int status = -EINVAL;
 543         struct sk_buff *nskb;
 544         struct nfqnl_instance *queue;
 545         struct nfqnl_queue_entry *entry;
 546
 547         QDEBUG("entered\n");
 548
 549         queue = instance_lookup_get(queuenum);
 550         if (!queue) {
 551                 QDEBUG("no queue instance matching\n");
 552                 return -EINVAL;
 553         }
 554
 555         if (queue->copy_mode == NFQNL_COPY_NONE) {
 556                 QDEBUG("mode COPY_NONE, aborting\n");
 557                 status = -EAGAIN;
 558                 goto err_out_put;
 559         }
 560
 561         entry = kmalloc(sizeof(*entry), GFP_ATOMIC);
 562         if (entry == NULL) {
 563                 if (net_ratelimit())
 564                         printk(KERN_ERR
 565                                 "nf_queue: OOM in nfqnl_enqueue_packet()\n");
 566                 status = -ENOMEM;
 567                 goto err_out_put;
 568         }
 569
 570         entry->info = info;
 571         entry->skb = skb;
 572         entry->id = atomic_inc_return(&queue->id_sequence);
 573
 574         nskb = nfqnl_build_packet_message(queue, entry, &status);
 575         if (nskb == NULL)
 576                 goto err_out_free;
 577
 578         spin_lock_bh(&queue->lock);
 579
 580         if (!queue->peer_pid)
 581                 goto err_out_free_nskb;
 582
 583         if (queue->queue_total >= queue->queue_maxlen) {
 584                 queue->queue_dropped++;
 585                 status = -ENOSPC;
 586                 if (net_ratelimit())
 587                           printk(KERN_WARNING "ip_queue: full at %d entries, "
 588                                  "dropping packets(s). Dropped: %d\n",
 589                                  queue->queue_total, queue->queue_dropped);
 590                 goto err_out_free_nskb;
 591         }
 592
 593         /* nfnetlink_unicast will either free the nskb or add it to a socket */
 594         status = nfnetlink_unicast(nskb, queue->peer_pid, MSG_DONTWAIT);
 595         if (status < 0) {
 596                 queue->queue_user_dropped++;
 597                 goto err_out_unlock;
 598         }
 599
 600         __enqueue_entry(queue, entry);
 601
 602         spin_unlock_bh(&queue->lock);
 603         instance_put(queue);
 604         return status;
 605
 606 err_out_free_nskb:
 607         kfree_skb(nskb);
 608
 609 err_out_unlock:
 610         spin_unlock_bh(&queue->lock);
 611
 612 err_out_free:
 613         kfree(entry);
 614 err_out_put:
 615         instance_put(queue);
 616         return status;
 617 }
 618
 619 static int
 620 nfqnl_mangle(void *data, int data_len, struct nfqnl_queue_entry *e)
 621 {
 622         int diff;
 623
 624         diff = data_len - e->skb->len;
 625         if (diff < 0)
 626                 skb_trim(e->skb, data_len);
 627         else if (diff > 0) {
 628                 if (data_len > 0xFFFF)
 629                         return -EINVAL;
 630                 if (diff > skb_tailroom(e->skb)) {
 631                         struct sk_buff *newskb;
 632
 633                         newskb = skb_copy_expand(e->skb,
 634                                                  skb_headroom(e->skb),
 635                                                  diff,
 636                                                  GFP_ATOMIC);
 637                         if (newskb == NULL) {
 638                                 printk(KERN_WARNING "ip_queue: OOM "
 639                                       "in mangle, dropping packet\n");
 640                                 return -ENOMEM;
 641                         }
 642                         if (e->skb->sk)
 643                                 skb_set_owner_w(newskb, e->skb->sk);
 644                         kfree_skb(e->skb);
 645                         e->skb = newskb;
 646                 }
 647                 skb_put(e->skb, diff);
 648         }
 649         if (!skb_make_writable(&e->skb, data_len))
 650                 return -ENOMEM;
 651         memcpy(e->skb->data, data, data_len);
 652         e->skb->ip_summed = CHECKSUM_NONE;
 653         return 0;
 654 }
 655
 656 static inline int
 657 id_cmp(struct nfqnl_queue_entry *e, unsigned long id)
 658 {
 659         return (id == e->id);
 660 }
 661
 662 static int
 663 nfqnl_set_mode(struct nfqnl_instance *queue,
 664                unsigned char mode, unsigned int range)
 665 {
 666         int status;
 667
 668         spin_lock_bh(&queue->lock);
 669         status = __nfqnl_set_mode(queue, mode, range);
 670         spin_unlock_bh(&queue->lock);
 671
 672         return status;
 673 }
 674
 675 static int
 676 dev_cmp(struct nfqnl_queue_entry *entry, unsigned long ifindex)
 677 {
 678         struct nf_info *entinf = entry->info;
 679
 680         if (entinf->indev)
 681                 if (entinf->indev->ifindex == ifindex)
 682                         return 1;
 683
 684         if (entinf->outdev)
 685                 if (entinf->outdev->ifindex == ifindex)
 686                         return 1;
 687
 688         return 0;
 689 }
 690
 691 /* drop all packets with either indev or outdev == ifindex from all queue
 692  * instances */
 693 static void
 694 nfqnl_dev_drop(int ifindex)
 695 {
 696         int i;
 697
 698         QDEBUG("entering for ifindex %u\n", ifindex);
 699
 700         /* this only looks like we have to hold the readlock for a way too long
 701          * time, issue_verdict(),  nf_reinject(), ... - but we always only
 702          * issue NF_DROP, which is processed directly in nf_reinject() */
 703         read_lock_bh(&instances_lock);
 704
 705         for  (i = 0; i < INSTANCE_BUCKETS; i++) {
 706                 struct hlist_node *tmp;
 707                 struct nfqnl_instance *inst;
 708                 struct hlist_head *head = &instance_table[i];
 709
 710                 hlist_for_each_entry(inst, tmp, head, hlist) {
 711                         struct nfqnl_queue_entry *entry;
 712                         while ((entry = find_dequeue_entry(inst, dev_cmp,
 713                                                            ifindex)) != NULL)
 714                                 issue_verdict(entry, NF_DROP);
 715                 }
 716         }
 717
 718         read_unlock_bh(&instances_lock);
 719 }
 720
 721 #define RCV_SKB_FAIL(err) do { netlink_ack(skb, nlh, (err)); return; } while (0)
 722
 723 static int
 724 nfqnl_rcv_dev_event(struct notifier_block *this,
 725                     unsigned long event, void *ptr)
 726 {
 727         struct net_device *dev = ptr;
 728
 729         /* Drop any packets associated with the downed device */
 730         if (event == NETDEV_DOWN)
 731                 nfqnl_dev_drop(dev->ifindex);
 732         return NOTIFY_DONE;
 733 }
 734
 735 static struct notifier_block nfqnl_dev_notifier = {
 736         .notifier_call  = nfqnl_rcv_dev_event,
 737 };
 738
 739 static int
 740 nfqnl_rcv_nl_event(struct notifier_block *this,
 741                    unsigned long event, void *ptr)
 742 {
 743         struct netlink_notify *n = ptr;
 744
 745         if (event == NETLINK_URELEASE &&
 746             n->protocol == NETLINK_NETFILTER && n->pid) {
 747                 int i;
 748
 749                 /* destroy all instances for this pid */
 750                 write_lock_bh(&instances_lock);
 751                 for  (i = 0; i < INSTANCE_BUCKETS; i++) {
 752                         struct hlist_node *tmp, *t2;
 753                         struct nfqnl_instance *inst;
 754                         struct hlist_head *head = &instance_table[i];
 755
 756                         hlist_for_each_entry_safe(inst, tmp, t2, head, hlist) {
 757                                 if (n->pid == inst->peer_pid)
 758                                         __instance_destroy(inst);
 759                         }
 760                 }
 761                 write_unlock_bh(&instances_lock);
 762         }
 763         return NOTIFY_DONE;
 764 }
 765
 766 static struct notifier_block nfqnl_rtnl_notifier = {
 767         .notifier_call  = nfqnl_rcv_nl_event,
 768 };
 769
 770 static const int nfqa_verdict_min[NFQA_MAX] = {
 771         [NFQA_VERDICT_HDR-1]    = sizeof(struct nfqnl_msg_verdict_hdr),
 772         [NFQA_MARK-1]           = sizeof(u_int32_t),
 773         [NFQA_PAYLOAD-1]        = 0,
 774 };
 775
 776 static int
 777 nfqnl_recv_verdict(struct sock *ctnl, struct sk_buff *skb,
 778                    struct nlmsghdr *nlh, struct nfattr *nfqa[], int *errp)
 779 {
 780         struct nfgenmsg *nfmsg = NLMSG_DATA(nlh);
 781         u_int16_t queue_num = ntohs(nfmsg->res_id);
 782
 783         struct nfqnl_msg_verdict_hdr *vhdr;
 784         struct nfqnl_instance *queue;
 785         unsigned int verdict;
 786         struct nfqnl_queue_entry *entry;
 787         int err;
 788
 789         if (nfattr_bad_size(nfqa, NFQA_MAX, nfqa_verdict_min)) {
 790                 QDEBUG("bad attribute size\n");
 791                 return -EINVAL;
 792         }
 793
 794         queue = instance_lookup_get(queue_num);
 795         if (!queue)
 796                 return -ENODEV;
 797
 798         if (queue->peer_pid != NETLINK_CB(skb).pid) {
 799                 err = -EPERM;
 800                 goto err_out_put;
 801         }
 802
 803         if (!nfqa[NFQA_VERDICT_HDR-1]) {
 804                 err = -EINVAL;
 805                 goto err_out_put;
 806         }
 807
 808         vhdr = NFA_DATA(nfqa[NFQA_VERDICT_HDR-1]);
 809         verdict = ntohl(vhdr->verdict);
 810
 811         if ((verdict & NF_VERDICT_MASK) > NF_MAX_VERDICT) {
 812                 err = -EINVAL;
 813                 goto err_out_put;
 814         }
 815
 816         entry = find_dequeue_entry(queue, id_cmp, ntohl(vhdr->id));
 817         if (entry == NULL) {
 818                 err = -ENOENT;
 819                 goto err_out_put;
 820         }
 821
 822         if (nfqa[NFQA_PAYLOAD-1]) {
 823                 if (nfqnl_mangle(NFA_DATA(nfqa[NFQA_PAYLOAD-1]),
 824                                  NFA_PAYLOAD(nfqa[NFQA_PAYLOAD-1]), entry) < 0)
 825                         verdict = NF_DROP;
 826         }
 827
 828         if (nfqa[NFQA_MARK-1])
 829                 entry->skb->nfmark = ntohl(*(u_int32_t *)
 830                                            NFA_DATA(nfqa[NFQA_MARK-1]));
 831
 832         issue_verdict(entry, verdict);
 833         instance_put(queue);
 834         return 0;
 835
 836 err_out_put:
 837         instance_put(queue);
 838         return err;
 839 }
 840
 841 static int
 842 nfqnl_recv_unsupp(struct sock *ctnl, struct sk_buff *skb,
 843                   struct nlmsghdr *nlh, struct nfattr *nfqa[], int *errp)
 844 {
 845         return -ENOTSUPP;
 846 }
 847
 848 static const int nfqa_cfg_min[NFQA_CFG_MAX] = {
 849         [NFQA_CFG_CMD-1]        = sizeof(struct nfqnl_msg_config_cmd),
 850         [NFQA_CFG_PARAMS-1]     = sizeof(struct nfqnl_msg_config_params),
 851 };
 852
 853 static struct nf_queue_handler nfqh = {
 854         .name   = "nf_queue",
 855         .outfn  = &nfqnl_enqueue_packet,
 856 };
 857
 858 static int
 859 nfqnl_recv_config(struct sock *ctnl, struct sk_buff *skb,
 860                   struct nlmsghdr *nlh, struct nfattr *nfqa[], int *errp)
 861 {
 862         struct nfgenmsg *nfmsg = NLMSG_DATA(nlh);
 863         u_int16_t queue_num = ntohs(nfmsg->res_id);
 864         struct nfqnl_instance *queue;
 865         int ret = 0;
 866
 867         QDEBUG("entering for msg %u\n", NFNL_MSG_TYPE(nlh->nlmsg_type));
 868
 869         if (nfattr_bad_size(nfqa, NFQA_CFG_MAX, nfqa_cfg_min)) {
 870                 QDEBUG("bad attribute size\n");
 871                 return -EINVAL;
 872         }
 873
 874         queue = instance_lookup_get(queue_num);
 875         if (nfqa[NFQA_CFG_CMD-1]) {
 876                 struct nfqnl_msg_config_cmd *cmd;
 877                 cmd = NFA_DATA(nfqa[NFQA_CFG_CMD-1]);
 878                 QDEBUG("found CFG_CMD\n");
 879
 880                 switch (cmd->command) {
 881                 case NFQNL_CFG_CMD_BIND:
 882                         if (queue)
 883                                 return -EBUSY;
 884
 885                         queue = instance_create(queue_num, NETLINK_CB(skb).pid);
 886                         if (!queue)
 887                                 return -EINVAL;
 888                         break;
 889                 case NFQNL_CFG_CMD_UNBIND:
 890                         if (!queue)
 891                                 return -ENODEV;
 892
 893                         if (queue->peer_pid != NETLINK_CB(skb).pid) {
 894                                 ret = -EPERM;
 895                                 goto out_put;
 896                         }
 897
 898                         instance_destroy(queue);
 899                         break;
 900                 case NFQNL_CFG_CMD_PF_BIND:
 901                         QDEBUG("registering queue handler for pf=%u\n",
 902                                 ntohs(cmd->pf));
 903                         ret = nf_register_queue_handler(ntohs(cmd->pf), &nfqh);
 904                         break;
 905                 case NFQNL_CFG_CMD_PF_UNBIND:
 906                         QDEBUG("unregistering queue handler for pf=%u\n",
 907                                 ntohs(cmd->pf));
 908                         /* This is a bug and a feature.  We can unregister
 909                          * other handlers(!) */
 910                         ret = nf_unregister_queue_handler(ntohs(cmd->pf));
 911                         break;
 912                 default:
 913                         ret = -EINVAL;
 914                         break;
 915                 }
 916         } else {
 917                 if (!queue) {
 918                         QDEBUG("no config command, and no instance ENOENT\n");
 919                         ret = -ENOENT;
 920                         goto out_put;
 921                 }
 922
 923                 if (queue->peer_pid != NETLINK_CB(skb).pid) {
 924                         QDEBUG("no config command, and wrong pid\n");
 925                         ret = -EPERM;
 926                         goto out_put;
 927                 }
 928         }
 929
 930         if (nfqa[NFQA_CFG_PARAMS-1]) {
 931                 struct nfqnl_msg_config_params *params;
 932
 933                 if (!queue) {
 934                         ret = -ENOENT;
 935                         goto out_put;
 936                 }
 937                 params = NFA_DATA(nfqa[NFQA_CFG_PARAMS-1]);
 938                 nfqnl_set_mode(queue, params->copy_mode,
 939                                 ntohl(params->copy_range));
 940         }
 941
 942 out_put:
 943         instance_put(queue);
 944         return ret;
 945 }
 946
 947 static struct nfnl_callback nfqnl_cb[NFQNL_MSG_MAX] = {
 948         [NFQNL_MSG_PACKET]      = { .call = nfqnl_recv_unsupp,
 949                                     .attr_count = NFQA_MAX, },
 950         [NFQNL_MSG_VERDICT]     = { .call = nfqnl_recv_verdict,
 951                                     .attr_count = NFQA_MAX, },
 952         [NFQNL_MSG_CONFIG]      = { .call = nfqnl_recv_config,
 953                                     .attr_count = NFQA_CFG_MAX, },
 954 };
 955
 956 static struct nfnetlink_subsystem nfqnl_subsys = {
 957         .name           = "nf_queue",
 958         .subsys_id      = NFNL_SUBSYS_QUEUE,
 959         .cb_count       = NFQNL_MSG_MAX,
 960         .cb             = nfqnl_cb,
 961 };
 962
 963 #ifdef CONFIG_PROC_FS
 964 struct iter_state {
 965         unsigned int bucket;
 966 };
 967
 968 static struct hlist_node *get_first(struct seq_file *seq)
 969 {
 970         struct iter_state *st = seq->private;
 971
 972         if (!st)
 973                 return NULL;
 974
 975         for (st->bucket = 0; st->bucket < INSTANCE_BUCKETS; st->bucket++) {
 976                 if (!hlist_empty(&instance_table[st->bucket]))
 977                         return instance_table[st->bucket].first;
 978         }
 979         return NULL;
 980 }
 981
 982 static struct hlist_node *get_next(struct seq_file *seq, struct hlist_node *h)
 983 {
 984         struct iter_state *st = seq->private;
 985
 986         h = h->next;
 987         while (!h) {
 988                 if (++st->bucket >= INSTANCE_BUCKETS)
 989                         return NULL;
 990
 991                 h = instance_table[st->bucket].first;
 992         }
 993         return h;
 994 }
 995
 996 static struct hlist_node *get_idx(struct seq_file *seq, loff_t pos)
 997 {
 998         struct hlist_node *head;
 999         head = get_first(seq);
1000
1001         if (head)
1002                 while (pos && (head = get_next(seq, head)))
1003                         pos--;
1004         return pos ? NULL : head;
1005 }
1006
1007 static void *seq_start(struct seq_file *seq, loff_t *pos)
1008 {
1009         read_lock_bh(&instances_lock);
1010         return get_idx(seq, *pos);
1011 }
1012
1013 static void *seq_next(struct seq_file *s, void *v, loff_t *pos)
1014 {
1015         (*pos)++;
1016         return get_next(s, v);
1017 }
1018
1019 static void seq_stop(struct seq_file *s, void *v)
1020 {
1021         read_unlock_bh(&instances_lock);
1022 }
1023
1024 static int seq_show(struct seq_file *s, void *v)
1025 {
1026         const struct nfqnl_instance *inst = v;
1027
1028         return seq_printf(s, "%5d %6d %5d %1d %5d %5d %5d %8d %2d\n",
1029                           inst->queue_num,
1030                           inst->peer_pid, inst->queue_total,
1031                           inst->copy_mode, inst->copy_range,
1032                           inst->queue_dropped, inst->queue_user_dropped,
1033                           atomic_read(&inst->id_sequence),
1034                           atomic_read(&inst->use));
1035 }
1036
1037 static struct seq_operations nfqnl_seq_ops = {
1038         .start  = seq_start,
1039         .next   = seq_next,
1040         .stop   = seq_stop,
1041         .show   = seq_show,
1042 };
1043
1044 static int nfqnl_open(struct inode *inode, struct file *file)
1045 {
1046         struct seq_file *seq;
1047         struct iter_state *is;
1048         int ret;
1049
1050         is = kzalloc(sizeof(*is), GFP_KERNEL);
1051         if (!is)
1052                 return -ENOMEM;
1053         ret = seq_open(file, &nfqnl_seq_ops);
1054         if (ret < 0)
1055                 goto out_free;
1056         seq = file->private_data;
1057         seq->private = is;
1058         return ret;
1059 out_free:
1060         kfree(is);
1061         return ret;
1062 }
1063
1064 static struct file_operations nfqnl_file_ops = {
1065         .owner   = THIS_MODULE,
1066         .open    = nfqnl_open,
1067         .read    = seq_read,
1068         .llseek  = seq_lseek,
1069         .release = seq_release_private,
1070 };
1071
1072 #endif /* PROC_FS */
1073
1074 static int __init nfnetlink_queue_init(void)
1075 {
1076         int i, status = -ENOMEM;
1077 #ifdef CONFIG_PROC_FS
1078         struct proc_dir_entry *proc_nfqueue;
1079 #endif
1080
1081         for (i = 0; i < INSTANCE_BUCKETS; i++)
1082                 INIT_HLIST_HEAD(&instance_table[i]);
1083
1084         netlink_register_notifier(&nfqnl_rtnl_notifier);
1085         status = nfnetlink_subsys_register(&nfqnl_subsys);
1086         if (status < 0) {
1087                 printk(KERN_ERR "nf_queue: failed to create netlink socket\n");
1088                 goto cleanup_netlink_notifier;
1089         }
1090
1091 #ifdef CONFIG_PROC_FS
1092         proc_nfqueue = create_proc_entry("nfnetlink_queue", 0440,
1093                                          proc_net_netfilter);
1094         if (!proc_nfqueue)
1095                 goto cleanup_subsys;
1096         proc_nfqueue->proc_fops = &nfqnl_file_ops;
1097 #endif
1098
1099         register_netdevice_notifier(&nfqnl_dev_notifier);
1100         return status;
1101
1102 #ifdef CONFIG_PROC_FS
1103 cleanup_subsys:
1104         nfnetlink_subsys_unregister(&nfqnl_subsys);
1105 #endif
1106 cleanup_netlink_notifier:
1107         netlink_unregister_notifier(&nfqnl_rtnl_notifier);
1108         return status;
1109 }
1110
1111 static void __exit nfnetlink_queue_fini(void)
1112 {
1113         nf_unregister_queue_handlers(&nfqh);
1114         unregister_netdevice_notifier(&nfqnl_dev_notifier);
1115 #ifdef CONFIG_PROC_FS
1116         remove_proc_entry("nfnetlink_queue", proc_net_netfilter);
1117 #endif
1118         nfnetlink_subsys_unregister(&nfqnl_subsys);
1119         netlink_unregister_notifier(&nfqnl_rtnl_notifier);
1120 }
1121
1122 MODULE_DESCRIPTION("netfilter packet queue handler");
1123 MODULE_AUTHOR("Harald Welte <laforge@netfilter.org>");
1124 MODULE_LICENSE("GPL");
1125 MODULE_ALIAS_NFNL_SUBSYS(NFNL_SUBSYS_QUEUE);
1126
1127 module_init(nfnetlink_queue_init);
1128 module_exit(nfnetlink_queue_fini);