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 u_int64_t htonll(u_int64_t in)
  80 {
  81         u_int64_t out;
  82         int i;
  83
  84         for (i = 0; i < sizeof(u_int64_t); i++)
  85                 ((u_int8_t *)&out)[sizeof(u_int64_t)-1] = ((u_int8_t *)&in)[i];
  86
  87         return out;
  88 }
  89
  90 #define INSTANCE_BUCKETS        16
  91 static struct hlist_head instance_table[INSTANCE_BUCKETS];
  92
  93 static inline u_int8_t instance_hashfn(u_int16_t queue_num)
  94 {
  95         return ((queue_num >> 8) | queue_num) % INSTANCE_BUCKETS;
  96 }
  97
  98 static struct nfqnl_instance *
  99 __instance_lookup(u_int16_t queue_num)
 100 {
 101         struct hlist_head *head;
 102         struct hlist_node *pos;
 103         struct nfqnl_instance *inst;
 104
 105         head = &instance_table[instance_hashfn(queue_num)];
 106         hlist_for_each_entry(inst, pos, head, hlist) {
 107                 if (inst->queue_num == queue_num)
 108                         return inst;
 109         }
 110         return NULL;
 111 }
 112
 113 static struct nfqnl_instance *
 114 instance_lookup_get(u_int16_t queue_num)
 115 {
 116         struct nfqnl_instance *inst;
 117
 118         read_lock_bh(&instances_lock);
 119         inst = __instance_lookup(queue_num);
 120         if (inst)
 121                 atomic_inc(&inst->use);
 122         read_unlock_bh(&instances_lock);
 123
 124         return inst;
 125 }
 126
 127 static void
 128 instance_put(struct nfqnl_instance *inst)
 129 {
 130         if (inst && atomic_dec_and_test(&inst->use)) {
 131                 QDEBUG("kfree(inst=%p)\n", inst);
 132                 kfree(inst);
 133         }
 134 }
 135
 136 static struct nfqnl_instance *
 137 instance_create(u_int16_t queue_num, int pid)
 138 {
 139         struct nfqnl_instance *inst;
 140
 141         QDEBUG("entering for queue_num=%u, pid=%d\n", queue_num, pid);
 142
 143         write_lock_bh(&instances_lock);
 144         if (__instance_lookup(queue_num)) {
 145                 inst = NULL;
 146                 QDEBUG("aborting, instance already exists\n");
 147                 goto out_unlock;
 148         }
 149
 150         inst = kmalloc(sizeof(*inst), GFP_ATOMIC);
 151         if (!inst)
 152                 goto out_unlock;
 153
 154         memset(inst, 0, sizeof(*inst));
 155         inst->queue_num = queue_num;
 156         inst->peer_pid = pid;
 157         inst->queue_maxlen = NFQNL_QMAX_DEFAULT;
 158         inst->copy_range = 0xfffff;
 159         inst->copy_mode = NFQNL_COPY_NONE;
 160         atomic_set(&inst->id_sequence, 0);
 161         /* needs to be two, since we _put() after creation */
 162         atomic_set(&inst->use, 2);
 163         inst->lock = SPIN_LOCK_UNLOCKED;
 164         INIT_LIST_HEAD(&inst->queue_list);
 165
 166         if (!try_module_get(THIS_MODULE))
 167                 goto out_free;
 168
 169         hlist_add_head(&inst->hlist,
 170                        &instance_table[instance_hashfn(queue_num)]);
 171
 172         write_unlock_bh(&instances_lock);
 173
 174         QDEBUG("successfully created new instance\n");
 175
 176         return inst;
 177
 178 out_free:
 179         kfree(inst);
 180 out_unlock:
 181         write_unlock_bh(&instances_lock);
 182         return NULL;
 183 }
 184
 185 static void nfqnl_flush(struct nfqnl_instance *queue, int verdict);
 186
 187 static void
 188 _instance_destroy2(struct nfqnl_instance *inst, int lock)
 189 {
 190         /* first pull it out of the global list */
 191         if (lock)
 192                 write_lock_bh(&instances_lock);
 193
 194         QDEBUG("removing instance %p (queuenum=%u) from hash\n",
 195                 inst, inst->queue_num);
 196         hlist_del(&inst->hlist);
 197
 198         if (lock)
 199                 write_unlock_bh(&instances_lock);
 200
 201         /* then flush all pending skbs from the queue */
 202         nfqnl_flush(inst, NF_DROP);
 203
 204         /* and finally put the refcount */
 205         instance_put(inst);
 206
 207         module_put(THIS_MODULE);
 208 }
 209
 210 static inline void
 211 __instance_destroy(struct nfqnl_instance *inst)
 212 {
 213         _instance_destroy2(inst, 0);
 214 }
 215
 216 static inline void
 217 instance_destroy(struct nfqnl_instance *inst)
 218 {
 219         _instance_destroy2(inst, 1);
 220 }
 221
 222
 223
 224 static void
 225 issue_verdict(struct nfqnl_queue_entry *entry, int verdict)
 226 {
 227         QDEBUG("entering for entry %p, verdict %u\n", entry, verdict);
 228
 229         /* TCP input path (and probably other bits) assume to be called
 230          * from softirq context, not from syscall, like issue_verdict is
 231          * called.  TCP input path deadlocks with locks taken from timer
 232          * softirq, e.g.  We therefore emulate this by local_bh_disable() */
 233
 234         local_bh_disable();
 235         nf_reinject(entry->skb, entry->info, verdict);
 236         local_bh_enable();
 237
 238         kfree(entry);
 239 }
 240
 241 static inline void
 242 __enqueue_entry(struct nfqnl_instance *queue,
 243                       struct nfqnl_queue_entry *entry)
 244 {
 245        list_add(&entry->list, &queue->queue_list);
 246        queue->queue_total++;
 247 }
 248
 249 /*
 250  * Find and return a queued entry matched by cmpfn, or return the last
 251  * entry if cmpfn is NULL.
 252  */
 253 static inline struct nfqnl_queue_entry *
 254 __find_entry(struct nfqnl_instance *queue, nfqnl_cmpfn cmpfn,
 255                    unsigned long data)
 256 {
 257         struct list_head *p;
 258
 259         list_for_each_prev(p, &queue->queue_list) {
 260                 struct nfqnl_queue_entry *entry = (struct nfqnl_queue_entry *)p;
 261
 262                 if (!cmpfn || cmpfn(entry, data))
 263                         return entry;
 264         }
 265         return NULL;
 266 }
 267
 268 static inline void
 269 __dequeue_entry(struct nfqnl_instance *q, struct nfqnl_queue_entry *entry)
 270 {
 271         list_del(&entry->list);
 272         q->queue_total--;
 273 }
 274
 275 static inline struct nfqnl_queue_entry *
 276 __find_dequeue_entry(struct nfqnl_instance *queue,
 277                      nfqnl_cmpfn cmpfn, unsigned long data)
 278 {
 279         struct nfqnl_queue_entry *entry;
 280
 281         entry = __find_entry(queue, cmpfn, data);
 282         if (entry == NULL)
 283                 return NULL;
 284
 285         __dequeue_entry(queue, entry);
 286         return entry;
 287 }
 288
 289
 290 static inline void
 291 __nfqnl_flush(struct nfqnl_instance *queue, int verdict)
 292 {
 293         struct nfqnl_queue_entry *entry;
 294
 295         while ((entry = __find_dequeue_entry(queue, NULL, 0)))
 296                 issue_verdict(entry, verdict);
 297 }
 298
 299 static inline int
 300 __nfqnl_set_mode(struct nfqnl_instance *queue,
 301                  unsigned char mode, unsigned int range)
 302 {
 303         int status = 0;
 304
 305         switch (mode) {
 306         case NFQNL_COPY_NONE:
 307         case NFQNL_COPY_META:
 308                 queue->copy_mode = mode;
 309                 queue->copy_range = 0;
 310                 break;
 311
 312         case NFQNL_COPY_PACKET:
 313                 queue->copy_mode = mode;
 314                 /* we're using struct nfattr which has 16bit nfa_len */
 315                 if (range > 0xffff)
 316                         queue->copy_range = 0xffff;
 317                 else
 318                         queue->copy_range = range;
 319                 break;
 320
 321         default:
 322                 status = -EINVAL;
 323
 324         }
 325         return status;
 326 }
 327
 328 static struct nfqnl_queue_entry *
 329 find_dequeue_entry(struct nfqnl_instance *queue,
 330                          nfqnl_cmpfn cmpfn, unsigned long data)
 331 {
 332         struct nfqnl_queue_entry *entry;
 333
 334         spin_lock_bh(&queue->lock);
 335         entry = __find_dequeue_entry(queue, cmpfn, data);
 336         spin_unlock_bh(&queue->lock);
 337
 338         return entry;
 339 }
 340
 341 static void
 342 nfqnl_flush(struct nfqnl_instance *queue, int verdict)
 343 {
 344         spin_lock_bh(&queue->lock);
 345         __nfqnl_flush(queue, verdict);
 346         spin_unlock_bh(&queue->lock);
 347 }
 348
 349 static struct sk_buff *
 350 nfqnl_build_packet_message(struct nfqnl_instance *queue,
 351                            struct nfqnl_queue_entry *entry, int *errp)
 352 {
 353         unsigned char *old_tail;
 354         size_t size;
 355         size_t data_len = 0;
 356         struct sk_buff *skb;
 357         struct nfqnl_msg_packet_hdr pmsg;
 358         struct nlmsghdr *nlh;
 359         struct nfgenmsg *nfmsg;
 360         unsigned int tmp_uint;
 361
 362         QDEBUG("entered\n");
 363
 364         /* all macros expand to constant values at compile time */
 365         size =    NLMSG_SPACE(sizeof(struct nfqnl_msg_packet_hdr))
 366                 + NLMSG_SPACE(sizeof(u_int32_t))        /* ifindex */
 367                 + NLMSG_SPACE(sizeof(u_int32_t))        /* ifindex */
 368 #ifdef CONFIG_BRIDGE_NETFILTER
 369                 + NLMSG_SPACE(sizeof(u_int32_t))        /* ifindex */
 370                 + NLMSG_SPACE(sizeof(u_int32_t))        /* ifindex */
 371 #endif
 372                 + NLMSG_SPACE(sizeof(u_int32_t))        /* mark */
 373                 + NLMSG_SPACE(sizeof(struct nfqnl_msg_packet_hw))
 374                 + NLMSG_SPACE(sizeof(struct nfqnl_msg_packet_timestamp));
 375
 376         spin_lock_bh(&queue->lock);
 377
 378         switch (queue->copy_mode) {
 379         case NFQNL_COPY_META:
 380         case NFQNL_COPY_NONE:
 381                 data_len = 0;
 382                 break;
 383
 384         case NFQNL_COPY_PACKET:
 385                 if (queue->copy_range == 0
 386                     || queue->copy_range > entry->skb->len)
 387                         data_len = entry->skb->len;
 388                 else
 389                         data_len = queue->copy_range;
 390
 391                 size += NLMSG_SPACE(data_len);
 392                 break;
 393
 394         default:
 395                 *errp = -EINVAL;
 396                 spin_unlock_bh(&queue->lock);
 397                 return NULL;
 398         }
 399
 400         spin_unlock_bh(&queue->lock);
 401
 402         skb = alloc_skb(size, GFP_ATOMIC);
 403         if (!skb)
 404                 goto nlmsg_failure;
 405
 406         old_tail= skb->tail;
 407         nlh = NLMSG_PUT(skb, 0, 0,
 408                         NFNL_SUBSYS_QUEUE << 8 | NFQNL_MSG_PACKET,
 409                         sizeof(struct nfgenmsg));
 410         nfmsg = NLMSG_DATA(nlh);
 411         nfmsg->nfgen_family = entry->info->pf;
 412         nfmsg->version = NFNETLINK_V0;
 413         nfmsg->res_id = htons(queue->queue_num);
 414
 415         pmsg.packet_id          = htonl(entry->id);
 416         pmsg.hw_protocol        = htons(entry->skb->protocol);
 417         pmsg.hook               = entry->info->hook;
 418
 419         NFA_PUT(skb, NFQA_PACKET_HDR, sizeof(pmsg), &pmsg);
 420
 421         if (entry->info->indev) {
 422                 tmp_uint = htonl(entry->info->indev->ifindex);
 423 #ifndef CONFIG_BRIDGE_NETFILTER
 424                 NFA_PUT(skb, NFQA_IFINDEX_INDEV, sizeof(tmp_uint), &tmp_uint);
 425 #else
 426                 if (entry->info->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                         NFA_PUT(skb, NFQA_IFINDEX_PHYSINDEV, sizeof(tmp_uint),
 431                                 &tmp_uint);
 432                         /* this is the bridge group "brX" */
 433                         tmp_uint = htonl(entry->info->indev->br_port->br->dev->ifindex);
 434                         NFA_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                         NFA_PUT(skb, NFQA_IFINDEX_INDEV, sizeof(tmp_uint),
 440                                 &tmp_uint);
 441                         if (entry->skb->nf_bridge
 442                             && entry->skb->nf_bridge->physindev) {
 443                                 tmp_uint = htonl(entry->skb->nf_bridge->physindev->ifindex);
 444                                 NFA_PUT(skb, NFQA_IFINDEX_PHYSINDEV,
 445                                         sizeof(tmp_uint), &tmp_uint);
 446                         }
 447                 }
 448 #endif
 449         }
 450
 451         if (entry->info->outdev) {
 452                 tmp_uint = htonl(entry->info->outdev->ifindex);
 453 #ifndef CONFIG_BRIDGE_NETFILTER
 454                 NFA_PUT(skb, NFQA_IFINDEX_OUTDEV, sizeof(tmp_uint), &tmp_uint);
 455 #else
 456                 if (entry->info->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                         NFA_PUT(skb, NFQA_IFINDEX_PHYSOUTDEV, sizeof(tmp_uint),
 461                                 &tmp_uint);
 462                         /* this is the bridge group "brX" */
 463                         tmp_uint = htonl(entry->info->outdev->br_port->br->dev->ifindex);
 464                         NFA_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                         NFA_PUT(skb, NFQA_IFINDEX_OUTDEV, sizeof(tmp_uint),
 470                                 &tmp_uint);
 471                         if (entry->skb->nf_bridge
 472                             && entry->skb->nf_bridge->physoutdev) {
 473                                 tmp_uint = htonl(entry->skb->nf_bridge->physoutdev->ifindex);
 474                                 NFA_PUT(skb, NFQA_IFINDEX_PHYSOUTDEV,
 475                                         sizeof(tmp_uint), &tmp_uint);
 476                         }
 477                 }
 478 #endif
 479         }
 480
 481         if (entry->skb->nfmark) {
 482                 tmp_uint = htonl(entry->skb->nfmark);
 483                 NFA_PUT(skb, NFQA_MARK, sizeof(u_int32_t), &tmp_uint);
 484         }
 485
 486         if (entry->info->indev && entry->skb->dev
 487             && entry->skb->dev->hard_header_parse) {
 488                 struct nfqnl_msg_packet_hw phw;
 489
 490                 phw.hw_addrlen =
 491                         entry->skb->dev->hard_header_parse(entry->skb,
 492                                                            phw.hw_addr);
 493                 phw.hw_addrlen = htons(phw.hw_addrlen);
 494                 NFA_PUT(skb, NFQA_HWADDR, sizeof(phw), &phw);
 495         }
 496
 497         if (entry->skb->tstamp.off_sec) {
 498                 struct nfqnl_msg_packet_timestamp ts;
 499
 500                 ts.sec = htonll(skb_tv_base.tv_sec + entry->skb->tstamp.off_sec);
 501                 ts.usec = htonll(skb_tv_base.tv_usec + entry->skb->tstamp.off_usec);
 502
 503                 NFA_PUT(skb, NFQA_TIMESTAMP, sizeof(ts), &ts);
 504         }
 505
 506         if (data_len) {
 507                 struct nfattr *nfa;
 508                 int size = NFA_LENGTH(data_len);
 509
 510                 if (skb_tailroom(skb) < (int)NFA_SPACE(data_len)) {
 511                         printk(KERN_WARNING "nf_queue: no tailroom!\n");
 512                         goto nlmsg_failure;
 513                 }
 514
 515                 nfa = (struct nfattr *)skb_put(skb, NFA_ALIGN(size));
 516                 nfa->nfa_type = NFQA_PAYLOAD;
 517                 nfa->nfa_len = size;
 518
 519                 if (skb_copy_bits(entry->skb, 0, NFA_DATA(nfa), data_len))
 520                         BUG();
 521         }
 522
 523         nlh->nlmsg_len = skb->tail - old_tail;
 524         return skb;
 525
 526 nlmsg_failure:
 527 nfattr_failure:
 528         if (skb)
 529                 kfree_skb(skb);
 530         *errp = -EINVAL;
 531         if (net_ratelimit())
 532                 printk(KERN_ERR "nf_queue: error creating packet message\n");
 533         return NULL;
 534 }
 535
 536 static int
 537 nfqnl_enqueue_packet(struct sk_buff *skb, struct nf_info *info,
 538                      unsigned int queuenum, void *data)
 539 {
 540         int status = -EINVAL;
 541         struct sk_buff *nskb;
 542         struct nfqnl_instance *queue;
 543         struct nfqnl_queue_entry *entry;
 544
 545         QDEBUG("entered\n");
 546
 547         queue = instance_lookup_get(queuenum);
 548         if (!queue) {
 549                 QDEBUG("no queue instance matching\n");
 550                 return -EINVAL;
 551         }
 552
 553         if (queue->copy_mode == NFQNL_COPY_NONE) {
 554                 QDEBUG("mode COPY_NONE, aborting\n");
 555                 status = -EAGAIN;
 556                 goto err_out_put;
 557         }
 558
 559         entry = kmalloc(sizeof(*entry), GFP_ATOMIC);
 560         if (entry == NULL) {
 561                 if (net_ratelimit())
 562                         printk(KERN_ERR
 563                                 "nf_queue: OOM in nfqnl_enqueue_packet()\n");
 564                 status = -ENOMEM;
 565                 goto err_out_put;
 566         }
 567
 568         entry->info = info;
 569         entry->skb = skb;
 570         entry->id = atomic_inc_return(&queue->id_sequence);
 571
 572         nskb = nfqnl_build_packet_message(queue, entry, &status);
 573         if (nskb == NULL)
 574                 goto err_out_free;
 575
 576         spin_lock_bh(&queue->lock);
 577
 578         if (!queue->peer_pid)
 579                 goto err_out_free_nskb;
 580
 581         if (queue->queue_total >= queue->queue_maxlen) {
 582                 queue->queue_dropped++;
 583                 status = -ENOSPC;
 584                 if (net_ratelimit())
 585                           printk(KERN_WARNING "ip_queue: full at %d entries, "
 586                                  "dropping packets(s). Dropped: %d\n",
 587                                  queue->queue_total, queue->queue_dropped);
 588                 goto err_out_free_nskb;
 589         }
 590
 591         /* nfnetlink_unicast will either free the nskb or add it to a socket */
 592         status = nfnetlink_unicast(nskb, queue->peer_pid, MSG_DONTWAIT);
 593         if (status < 0) {
 594                 queue->queue_user_dropped++;
 595                 goto err_out_unlock;
 596         }
 597
 598         __enqueue_entry(queue, entry);
 599
 600         spin_unlock_bh(&queue->lock);
 601         instance_put(queue);
 602         return status;
 603
 604 err_out_free_nskb:
 605         kfree_skb(nskb);
 606
 607 err_out_unlock:
 608         spin_unlock_bh(&queue->lock);
 609
 610 err_out_free:
 611         kfree(entry);
 612 err_out_put:
 613         instance_put(queue);
 614         return status;
 615 }
 616
 617 static int
 618 nfqnl_mangle(void *data, int data_len, struct nfqnl_queue_entry *e)
 619 {
 620         int diff;
 621
 622         diff = data_len - e->skb->len;
 623         if (diff < 0)
 624                 skb_trim(e->skb, data_len);
 625         else if (diff > 0) {
 626                 if (data_len > 0xFFFF)
 627                         return -EINVAL;
 628                 if (diff > skb_tailroom(e->skb)) {
 629                         struct sk_buff *newskb;
 630
 631                         newskb = skb_copy_expand(e->skb,
 632                                                  skb_headroom(e->skb),
 633                                                  diff,
 634                                                  GFP_ATOMIC);
 635                         if (newskb == NULL) {
 636                                 printk(KERN_WARNING "ip_queue: OOM "
 637                                       "in mangle, dropping packet\n");
 638                                 return -ENOMEM;
 639                         }
 640                         if (e->skb->sk)
 641                                 skb_set_owner_w(newskb, e->skb->sk);
 642                         kfree_skb(e->skb);
 643                         e->skb = newskb;
 644                 }
 645                 skb_put(e->skb, diff);
 646         }
 647         if (!skb_make_writable(&e->skb, data_len))
 648                 return -ENOMEM;
 649         memcpy(e->skb->data, data, data_len);
 650
 651         return 0;
 652 }
 653
 654 static inline int
 655 id_cmp(struct nfqnl_queue_entry *e, unsigned long id)
 656 {
 657         return (id == e->id);
 658 }
 659
 660 static int
 661 nfqnl_set_mode(struct nfqnl_instance *queue,
 662                unsigned char mode, unsigned int range)
 663 {
 664         int status;
 665
 666         spin_lock_bh(&queue->lock);
 667         status = __nfqnl_set_mode(queue, mode, range);
 668         spin_unlock_bh(&queue->lock);
 669
 670         return status;
 671 }
 672
 673 static int
 674 dev_cmp(struct nfqnl_queue_entry *entry, unsigned long ifindex)
 675 {
 676         if (entry->info->indev)
 677                 if (entry->info->indev->ifindex == ifindex)
 678                         return 1;
 679
 680         if (entry->info->outdev)
 681                 if (entry->info->outdev->ifindex == ifindex)
 682                         return 1;
 683
 684         return 0;
 685 }
 686
 687 /* drop all packets with either indev or outdev == ifindex from all queue
 688  * instances */
 689 static void
 690 nfqnl_dev_drop(int ifindex)
 691 {
 692         int i;
 693
 694         QDEBUG("entering for ifindex %u\n", ifindex);
 695
 696         /* this only looks like we have to hold the readlock for a way too long
 697          * time, issue_verdict(),  nf_reinject(), ... - but we always only
 698          * issue NF_DROP, which is processed directly in nf_reinject() */
 699         read_lock_bh(&instances_lock);
 700
 701         for  (i = 0; i < INSTANCE_BUCKETS; i++) {
 702                 struct hlist_node *tmp;
 703                 struct nfqnl_instance *inst;
 704                 struct hlist_head *head = &instance_table[i];
 705
 706                 hlist_for_each_entry(inst, tmp, head, hlist) {
 707                         struct nfqnl_queue_entry *entry;
 708                         while ((entry = find_dequeue_entry(inst, dev_cmp,
 709                                                            ifindex)) != NULL)
 710                                 issue_verdict(entry, NF_DROP);
 711                 }
 712         }
 713
 714         read_unlock_bh(&instances_lock);
 715 }
 716
 717 #define RCV_SKB_FAIL(err) do { netlink_ack(skb, nlh, (err)); return; } while (0)
 718
 719 static int
 720 nfqnl_rcv_dev_event(struct notifier_block *this,
 721                     unsigned long event, void *ptr)
 722 {
 723         struct net_device *dev = ptr;
 724
 725         /* Drop any packets associated with the downed device */
 726         if (event == NETDEV_DOWN)
 727                 nfqnl_dev_drop(dev->ifindex);
 728         return NOTIFY_DONE;
 729 }
 730
 731 static struct notifier_block nfqnl_dev_notifier = {
 732         .notifier_call  = nfqnl_rcv_dev_event,
 733 };
 734
 735 static int
 736 nfqnl_rcv_nl_event(struct notifier_block *this,
 737                    unsigned long event, void *ptr)
 738 {
 739         struct netlink_notify *n = ptr;
 740
 741         if (event == NETLINK_URELEASE &&
 742             n->protocol == NETLINK_NETFILTER && n->pid) {
 743                 int i;
 744
 745                 /* destroy all instances for this pid */
 746                 write_lock_bh(&instances_lock);
 747                 for  (i = 0; i < INSTANCE_BUCKETS; i++) {
 748                         struct hlist_node *tmp, *t2;
 749                         struct nfqnl_instance *inst;
 750                         struct hlist_head *head = &instance_table[i];
 751
 752                         hlist_for_each_entry_safe(inst, tmp, t2, head, hlist) {
 753                                 if (n->pid == inst->peer_pid)
 754                                         __instance_destroy(inst);
 755                         }
 756                 }
 757                 write_unlock_bh(&instances_lock);
 758         }
 759         return NOTIFY_DONE;
 760 }
 761
 762 static struct notifier_block nfqnl_rtnl_notifier = {
 763         .notifier_call  = nfqnl_rcv_nl_event,
 764 };
 765
 766 static const int nfqa_verdict_min[NFQA_MAX] = {
 767         [NFQA_VERDICT_HDR-1]    = sizeof(struct nfqnl_msg_verdict_hdr),
 768         [NFQA_MARK-1]           = sizeof(u_int32_t),
 769         [NFQA_PAYLOAD-1]        = 0,
 770 };
 771
 772 static int
 773 nfqnl_recv_verdict(struct sock *ctnl, struct sk_buff *skb,
 774                    struct nlmsghdr *nlh, struct nfattr *nfqa[], int *errp)
 775 {
 776         struct nfgenmsg *nfmsg = NLMSG_DATA(nlh);
 777         u_int16_t queue_num = ntohs(nfmsg->res_id);
 778
 779         struct nfqnl_msg_verdict_hdr *vhdr;
 780         struct nfqnl_instance *queue;
 781         unsigned int verdict;
 782         struct nfqnl_queue_entry *entry;
 783         int err;
 784
 785         if (nfattr_bad_size(nfqa, NFQA_MAX, nfqa_verdict_min)) {
 786                 QDEBUG("bad attribute size\n");
 787                 return -EINVAL;
 788         }
 789
 790         queue = instance_lookup_get(queue_num);
 791         if (!queue)
 792                 return -ENODEV;
 793
 794         if (queue->peer_pid != NETLINK_CB(skb).pid) {
 795                 err = -EPERM;
 796                 goto err_out_put;
 797         }
 798
 799         if (!nfqa[NFQA_VERDICT_HDR-1]) {
 800                 err = -EINVAL;
 801                 goto err_out_put;
 802         }
 803
 804         vhdr = NFA_DATA(nfqa[NFQA_VERDICT_HDR-1]);
 805         verdict = ntohl(vhdr->verdict);
 806
 807         if ((verdict & NF_VERDICT_MASK) > NF_MAX_VERDICT) {
 808                 err = -EINVAL;
 809                 goto err_out_put;
 810         }
 811
 812         entry = find_dequeue_entry(queue, id_cmp, ntohl(vhdr->id));
 813         if (entry == NULL) {
 814                 err = -ENOENT;
 815                 goto err_out_put;
 816         }
 817
 818         if (nfqa[NFQA_PAYLOAD-1]) {
 819                 if (nfqnl_mangle(NFA_DATA(nfqa[NFQA_PAYLOAD-1]),
 820                                  NFA_PAYLOAD(nfqa[NFQA_PAYLOAD-1]), entry) < 0)
 821                         verdict = NF_DROP;
 822         }
 823
 824         if (nfqa[NFQA_MARK-1])
 825                 skb->nfmark = ntohl(*(u_int32_t *)NFA_DATA(nfqa[NFQA_MARK-1]));
 826
 827         issue_verdict(entry, verdict);
 828         instance_put(queue);
 829         return 0;
 830
 831 err_out_put:
 832         instance_put(queue);
 833         return err;
 834 }
 835
 836 static int
 837 nfqnl_recv_unsupp(struct sock *ctnl, struct sk_buff *skb,
 838                   struct nlmsghdr *nlh, struct nfattr *nfqa[], int *errp)
 839 {
 840         return -ENOTSUPP;
 841 }
 842
 843 static const int nfqa_cfg_min[NFQA_CFG_MAX] = {
 844         [NFQA_CFG_CMD-1]        = sizeof(struct nfqnl_msg_config_cmd),
 845         [NFQA_CFG_PARAMS-1]     = sizeof(struct nfqnl_msg_config_params),
 846 };
 847
 848 static struct nf_queue_handler nfqh = {
 849         .name   = "nf_queue",
 850         .outfn  = &nfqnl_enqueue_packet,
 851 };
 852
 853 static int
 854 nfqnl_recv_config(struct sock *ctnl, struct sk_buff *skb,
 855                   struct nlmsghdr *nlh, struct nfattr *nfqa[], int *errp)
 856 {
 857         struct nfgenmsg *nfmsg = NLMSG_DATA(nlh);
 858         u_int16_t queue_num = ntohs(nfmsg->res_id);
 859         struct nfqnl_instance *queue;
 860         int ret = 0;
 861
 862         QDEBUG("entering for msg %u\n", NFNL_MSG_TYPE(nlh->nlmsg_type));
 863
 864         if (nfattr_bad_size(nfqa, NFQA_CFG_MAX, nfqa_cfg_min)) {
 865                 QDEBUG("bad attribute size\n");
 866                 return -EINVAL;
 867         }
 868
 869         queue = instance_lookup_get(queue_num);
 870         if (nfqa[NFQA_CFG_CMD-1]) {
 871                 struct nfqnl_msg_config_cmd *cmd;
 872                 cmd = NFA_DATA(nfqa[NFQA_CFG_CMD-1]);
 873                 QDEBUG("found CFG_CMD\n");
 874
 875                 switch (cmd->command) {
 876                 case NFQNL_CFG_CMD_BIND:
 877                         if (queue)
 878                                 return -EBUSY;
 879
 880                         queue = instance_create(queue_num, NETLINK_CB(skb).pid);
 881                         if (!queue)
 882                                 return -EINVAL;
 883                         break;
 884                 case NFQNL_CFG_CMD_UNBIND:
 885                         if (!queue)
 886                                 return -ENODEV;
 887
 888                         if (queue->peer_pid != NETLINK_CB(skb).pid) {
 889                                 ret = -EPERM;
 890                                 goto out_put;
 891                         }
 892
 893                         instance_destroy(queue);
 894                         break;
 895                 case NFQNL_CFG_CMD_PF_BIND:
 896                         QDEBUG("registering queue handler for pf=%u\n",
 897                                 ntohs(cmd->pf));
 898                         ret = nf_register_queue_handler(ntohs(cmd->pf), &nfqh);
 899                         break;
 900                 case NFQNL_CFG_CMD_PF_UNBIND:
 901                         QDEBUG("unregistering queue handler for pf=%u\n",
 902                                 ntohs(cmd->pf));
 903                         /* This is a bug and a feature.  We can unregister
 904                          * other handlers(!) */
 905                         ret = nf_unregister_queue_handler(ntohs(cmd->pf));
 906                         break;
 907                 default:
 908                         ret = -EINVAL;
 909                         break;
 910                 }
 911         } else {
 912                 if (!queue) {
 913                         QDEBUG("no config command, and no instance ENOENT\n");
 914                         ret = -ENOENT;
 915                         goto out_put;
 916                 }
 917
 918                 if (queue->peer_pid != NETLINK_CB(skb).pid) {
 919                         QDEBUG("no config command, and wrong pid\n");
 920                         ret = -EPERM;
 921                         goto out_put;
 922                 }
 923         }
 924
 925         if (nfqa[NFQA_CFG_PARAMS-1]) {
 926                 struct nfqnl_msg_config_params *params;
 927                 params = NFA_DATA(nfqa[NFQA_CFG_PARAMS-1]);
 928
 929                 nfqnl_set_mode(queue, params->copy_mode,
 930                                 ntohl(params->copy_range));
 931         }
 932
 933 out_put:
 934         instance_put(queue);
 935         return ret;
 936 }
 937
 938 static struct nfnl_callback nfqnl_cb[NFQNL_MSG_MAX] = {
 939         [NFQNL_MSG_PACKET]      = { .call = nfqnl_recv_unsupp,
 940                                     .attr_count = NFQA_MAX,
 941                                     .cap_required = CAP_NET_ADMIN },
 942         [NFQNL_MSG_VERDICT]     = { .call = nfqnl_recv_verdict,
 943                                     .attr_count = NFQA_MAX,
 944                                     .cap_required = CAP_NET_ADMIN },
 945         [NFQNL_MSG_CONFIG]      = { .call = nfqnl_recv_config,
 946                                     .attr_count = NFQA_CFG_MAX,
 947                                     .cap_required = CAP_NET_ADMIN },
 948 };
 949
 950 static struct nfnetlink_subsystem nfqnl_subsys = {
 951         .name           = "nf_queue",
 952         .subsys_id      = NFNL_SUBSYS_QUEUE,
 953         .cb_count       = NFQNL_MSG_MAX,
 954         .cb             = nfqnl_cb,
 955 };
 956
 957 #ifdef CONFIG_PROC_FS
 958 struct iter_state {
 959         unsigned int bucket;
 960 };
 961
 962 static struct hlist_node *get_first(struct seq_file *seq)
 963 {
 964         struct iter_state *st = seq->private;
 965
 966         if (!st)
 967                 return NULL;
 968
 969         for (st->bucket = 0; st->bucket < INSTANCE_BUCKETS; st->bucket++) {
 970                 if (!hlist_empty(&instance_table[st->bucket]))
 971                         return instance_table[st->bucket].first;
 972         }
 973         return NULL;
 974 }
 975
 976 static struct hlist_node *get_next(struct seq_file *seq, struct hlist_node *h)
 977 {
 978         struct iter_state *st = seq->private;
 979
 980         h = h->next;
 981         while (!h) {
 982                 if (++st->bucket >= INSTANCE_BUCKETS)
 983                         return NULL;
 984
 985                 h = instance_table[st->bucket].first;
 986         }
 987         return h;
 988 }
 989
 990 static struct hlist_node *get_idx(struct seq_file *seq, loff_t pos)
 991 {
 992         struct hlist_node *head;
 993         head = get_first(seq);
 994
 995         if (head)
 996                 while (pos && (head = get_next(seq, head)))
 997                         pos--;
 998         return pos ? NULL : head;
 999 }
1000
1001 static void *seq_start(struct seq_file *seq, loff_t *pos)
1002 {
1003         read_lock_bh(&instances_lock);
1004         return get_idx(seq, *pos);
1005 }
1006
1007 static void *seq_next(struct seq_file *s, void *v, loff_t *pos)
1008 {
1009         (*pos)++;
1010         return get_next(s, v);
1011 }
1012
1013 static void seq_stop(struct seq_file *s, void *v)
1014 {
1015         read_unlock_bh(&instances_lock);
1016 }
1017
1018 static int seq_show(struct seq_file *s, void *v)
1019 {
1020         const struct nfqnl_instance *inst = v;
1021
1022         return seq_printf(s, "%5d %6d %5d %1d %5d %5d %5d %8d %2d\n",
1023                           inst->queue_num,
1024                           inst->peer_pid, inst->queue_total,
1025                           inst->copy_mode, inst->copy_range,
1026                           inst->queue_dropped, inst->queue_user_dropped,
1027                           atomic_read(&inst->id_sequence),
1028                           atomic_read(&inst->use));
1029 }
1030
1031 static struct seq_operations nfqnl_seq_ops = {
1032         .start  = seq_start,
1033         .next   = seq_next,
1034         .stop   = seq_stop,
1035         .show   = seq_show,
1036 };
1037
1038 static int nfqnl_open(struct inode *inode, struct file *file)
1039 {
1040         struct seq_file *seq;
1041         struct iter_state *is;
1042         int ret;
1043
1044         is = kmalloc(sizeof(*is), GFP_KERNEL);
1045         if (!is)
1046                 return -ENOMEM;
1047         memset(is, 0, sizeof(*is));
1048         ret = seq_open(file, &nfqnl_seq_ops);
1049         if (ret < 0)
1050                 goto out_free;
1051         seq = file->private_data;
1052         seq->private = is;
1053         return ret;
1054 out_free:
1055         kfree(is);
1056         return ret;
1057 }
1058
1059 static struct file_operations nfqnl_file_ops = {
1060         .owner   = THIS_MODULE,
1061         .open    = nfqnl_open,
1062         .read    = seq_read,
1063         .llseek  = seq_lseek,
1064         .release = seq_release_private,
1065 };
1066
1067 #endif /* PROC_FS */
1068
1069 static int
1070 init_or_cleanup(int init)
1071 {
1072         int i, status = -ENOMEM;
1073 #ifdef CONFIG_PROC_FS
1074         struct proc_dir_entry *proc_nfqueue;
1075 #endif
1076
1077         if (!init)
1078                 goto cleanup;
1079
1080         for (i = 0; i < INSTANCE_BUCKETS; i++)
1081                 INIT_HLIST_HEAD(&instance_table[i]);
1082
1083         netlink_register_notifier(&nfqnl_rtnl_notifier);
1084         status = nfnetlink_subsys_register(&nfqnl_subsys);
1085         if (status < 0) {
1086                 printk(KERN_ERR "nf_queue: failed to create netlink socket\n");
1087                 goto cleanup_netlink_notifier;
1088         }
1089
1090 #ifdef CONFIG_PROC_FS
1091         proc_nfqueue = create_proc_entry("nfnetlink_queue", 0440,
1092                                          proc_net_netfilter);
1093         if (!proc_nfqueue)
1094                 goto cleanup_subsys;
1095         proc_nfqueue->proc_fops = &nfqnl_file_ops;
1096 #endif
1097
1098         register_netdevice_notifier(&nfqnl_dev_notifier);
1099
1100         return status;
1101
1102 cleanup:
1103         nf_unregister_queue_handlers(&nfqh);
1104         unregister_netdevice_notifier(&nfqnl_dev_notifier);
1105 #ifdef CONFIG_PROC_FS
1106         remove_proc_entry("nfnetlink_queue", proc_net_netfilter);
1107 cleanup_subsys:
1108 #endif
1109         nfnetlink_subsys_unregister(&nfqnl_subsys);
1110 cleanup_netlink_notifier:
1111         netlink_unregister_notifier(&nfqnl_rtnl_notifier);
1112         return status;
1113 }
1114
1115 static int __init init(void)
1116 {
1117
1118         return init_or_cleanup(1);
1119 }
1120
1121 static void __exit fini(void)
1122 {
1123         init_or_cleanup(0);
1124 }
1125
1126 MODULE_DESCRIPTION("netfilter packet queue handler");
1127 MODULE_AUTHOR("Harald Welte <laforge@netfilter.org>");
1128 MODULE_LICENSE("GPL");
1129 MODULE_ALIAS_NFNL_SUBSYS(NFNL_SUBSYS_QUEUE);
1130
1131 module_init(init);
1132 module_exit(fini);