net/netfilter/nfnetlink_queue_core.c

   1 /*
   2  * This is a module which is used for queueing packets and communicating with
   3  * userspace via nfnetlink.
   4  *
   5  * (C) 2005 by Harald Welte <laforge@netfilter.org>
   6  * (C) 2007 by Patrick McHardy <kaber@trash.net>
   7  *
   8  * Based on the old ipv4-only ip_queue.c:
   9  * (C) 2000-2002 James Morris <jmorris@intercode.com.au>
  10  * (C) 2003-2005 Netfilter Core Team <coreteam@netfilter.org>
  11  *
  12  * This program is free software; you can redistribute it and/or modify
  13  * it under the terms of the GNU General Public License version 2 as
  14  * published by the Free Software Foundation.
  15  *
  16  */
  17 #include <linux/module.h>
  18 #include <linux/skbuff.h>
  19 #include <linux/init.h>
  20 #include <linux/spinlock.h>
  21 #include <linux/slab.h>
  22 #include <linux/notifier.h>
  23 #include <linux/netdevice.h>
  24 #include <linux/netfilter.h>
  25 #include <linux/proc_fs.h>
  26 #include <linux/netfilter_ipv4.h>
  27 #include <linux/netfilter_ipv6.h>
  28 #include <linux/netfilter/nfnetlink.h>
  29 #include <linux/netfilter/nfnetlink_queue.h>
  30 #include <linux/list.h>
  31 #include <net/sock.h>
  32 #include <net/netfilter/nf_queue.h>
  33 #include <net/netfilter/nfnetlink_queue.h>
  34
  35 #include <linux/atomic.h>
  36
  37 #ifdef CONFIG_BRIDGE_NETFILTER
  38 #include "../bridge/br_private.h"
  39 #endif
  40
  41 #define NFQNL_QMAX_DEFAULT 1024
  42
  43 struct nfqnl_instance {
  44         struct hlist_node hlist;                /* global list of queues */
  45         struct rcu_head rcu;
  46
  47         int peer_portid;
  48         unsigned int queue_maxlen;
  49         unsigned int copy_range;
  50         unsigned int queue_dropped;
  51         unsigned int queue_user_dropped;
  52
  53
  54         u_int16_t queue_num;                    /* number of this queue */
  55         u_int8_t copy_mode;
  56         u_int32_t flags;                        /* Set using NFQA_CFG_FLAGS */
  57 /*
  58  * Following fields are dirtied for each queued packet,
  59  * keep them in same cache line if possible.
  60  */
  61         spinlock_t      lock;
  62         unsigned int    queue_total;
  63         unsigned int    id_sequence;            /* 'sequence' of pkt ids */
  64         struct list_head queue_list;            /* packets in queue */
  65 };
  66
  67 typedef int (*nfqnl_cmpfn)(struct nf_queue_entry *, unsigned long);
  68
  69 static DEFINE_SPINLOCK(instances_lock);
  70
  71 #define INSTANCE_BUCKETS        16
  72 static struct hlist_head instance_table[INSTANCE_BUCKETS] __read_mostly;
  73
  74 static inline u_int8_t instance_hashfn(u_int16_t queue_num)
  75 {
  76         return ((queue_num >> 8) | queue_num) % INSTANCE_BUCKETS;
  77 }
  78
  79 static struct nfqnl_instance *
  80 instance_lookup(u_int16_t queue_num)
  81 {
  82         struct hlist_head *head;
  83         struct nfqnl_instance *inst;
  84
  85         head = &instance_table[instance_hashfn(queue_num)];
  86         hlist_for_each_entry_rcu(inst, head, hlist) {
  87                 if (inst->queue_num == queue_num)
  88                         return inst;
  89         }
  90         return NULL;
  91 }
  92
  93 static struct nfqnl_instance *
  94 instance_create(u_int16_t queue_num, int portid)
  95 {
  96         struct nfqnl_instance *inst;
  97         unsigned int h;
  98         int err;
  99
 100         spin_lock(&instances_lock);
 101         if (instance_lookup(queue_num)) {
 102                 err = -EEXIST;
 103                 goto out_unlock;
 104         }
 105
 106         inst = kzalloc(sizeof(*inst), GFP_ATOMIC);
 107         if (!inst) {
 108                 err = -ENOMEM;
 109                 goto out_unlock;
 110         }
 111
 112         inst->queue_num = queue_num;
 113         inst->peer_portid = portid;
 114         inst->queue_maxlen = NFQNL_QMAX_DEFAULT;
 115         inst->copy_range = 0xfffff;
 116         inst->copy_mode = NFQNL_COPY_NONE;
 117         spin_lock_init(&inst->lock);
 118         INIT_LIST_HEAD(&inst->queue_list);
 119
 120         if (!try_module_get(THIS_MODULE)) {
 121                 err = -EAGAIN;
 122                 goto out_free;
 123         }
 124
 125         h = instance_hashfn(queue_num);
 126         hlist_add_head_rcu(&inst->hlist, &instance_table[h]);
 127
 128         spin_unlock(&instances_lock);
 129
 130         return inst;
 131
 132 out_free:
 133         kfree(inst);
 134 out_unlock:
 135         spin_unlock(&instances_lock);
 136         return ERR_PTR(err);
 137 }
 138
 139 static void nfqnl_flush(struct nfqnl_instance *queue, nfqnl_cmpfn cmpfn,
 140                         unsigned long data);
 141
 142 static void
 143 instance_destroy_rcu(struct rcu_head *head)
 144 {
 145         struct nfqnl_instance *inst = container_of(head, struct nfqnl_instance,
 146                                                    rcu);
 147
 148         nfqnl_flush(inst, NULL, 0);
 149         kfree(inst);
 150         module_put(THIS_MODULE);
 151 }
 152
 153 static void
 154 __instance_destroy(struct nfqnl_instance *inst)
 155 {
 156         hlist_del_rcu(&inst->hlist);
 157         call_rcu(&inst->rcu, instance_destroy_rcu);
 158 }
 159
 160 static void
 161 instance_destroy(struct nfqnl_instance *inst)
 162 {
 163         spin_lock(&instances_lock);
 164         __instance_destroy(inst);
 165         spin_unlock(&instances_lock);
 166 }
 167
 168 static inline void
 169 __enqueue_entry(struct nfqnl_instance *queue, struct nf_queue_entry *entry)
 170 {
 171        list_add_tail(&entry->list, &queue->queue_list);
 172        queue->queue_total++;
 173 }
 174
 175 static void
 176 __dequeue_entry(struct nfqnl_instance *queue, struct nf_queue_entry *entry)
 177 {
 178         list_del(&entry->list);
 179         queue->queue_total--;
 180 }
 181
 182 static struct nf_queue_entry *
 183 find_dequeue_entry(struct nfqnl_instance *queue, unsigned int id)
 184 {
 185         struct nf_queue_entry *entry = NULL, *i;
 186
 187         spin_lock_bh(&queue->lock);
 188
 189         list_for_each_entry(i, &queue->queue_list, list) {
 190                 if (i->id == id) {
 191                         entry = i;
 192                         break;
 193                 }
 194         }
 195
 196         if (entry)
 197                 __dequeue_entry(queue, entry);
 198
 199         spin_unlock_bh(&queue->lock);
 200
 201         return entry;
 202 }
 203
 204 static void
 205 nfqnl_flush(struct nfqnl_instance *queue, nfqnl_cmpfn cmpfn, unsigned long data)
 206 {
 207         struct nf_queue_entry *entry, *next;
 208
 209         spin_lock_bh(&queue->lock);
 210         list_for_each_entry_safe(entry, next, &queue->queue_list, list) {
 211                 if (!cmpfn || cmpfn(entry, data)) {
 212                         list_del(&entry->list);
 213                         queue->queue_total--;
 214                         nf_reinject(entry, NF_DROP);
 215                 }
 216         }
 217         spin_unlock_bh(&queue->lock);
 218 }
 219
 220 static struct sk_buff *
 221 nfqnl_build_packet_message(struct nfqnl_instance *queue,
 222                            struct nf_queue_entry *entry,
 223                            __be32 **packet_id_ptr)
 224 {
 225         sk_buff_data_t old_tail;
 226         size_t size;
 227         size_t data_len = 0, cap_len = 0;
 228         struct sk_buff *skb;
 229         struct nlattr *nla;
 230         struct nfqnl_msg_packet_hdr *pmsg;
 231         struct nlmsghdr *nlh;
 232         struct nfgenmsg *nfmsg;
 233         struct sk_buff *entskb = entry->skb;
 234         struct net_device *indev;
 235         struct net_device *outdev;
 236         struct nf_conn *ct = NULL;
 237         enum ip_conntrack_info uninitialized_var(ctinfo);
 238
 239         size =    NLMSG_SPACE(sizeof(struct nfgenmsg))
 240                 + nla_total_size(sizeof(struct nfqnl_msg_packet_hdr))
 241                 + nla_total_size(sizeof(u_int32_t))     /* ifindex */
 242                 + nla_total_size(sizeof(u_int32_t))     /* ifindex */
 243 #ifdef CONFIG_BRIDGE_NETFILTER
 244                 + nla_total_size(sizeof(u_int32_t))     /* ifindex */
 245                 + nla_total_size(sizeof(u_int32_t))     /* ifindex */
 246 #endif
 247                 + nla_total_size(sizeof(u_int32_t))     /* mark */
 248                 + nla_total_size(sizeof(struct nfqnl_msg_packet_hw))
 249                 + nla_total_size(sizeof(struct nfqnl_msg_packet_timestamp)
 250                 + nla_total_size(sizeof(u_int32_t)));   /* cap_len */
 251
 252         outdev = entry->outdev;
 253
 254         switch ((enum nfqnl_config_mode)ACCESS_ONCE(queue->copy_mode)) {
 255         case NFQNL_COPY_META:
 256         case NFQNL_COPY_NONE:
 257                 break;
 258
 259         case NFQNL_COPY_PACKET:
 260                 if (entskb->ip_summed == CHECKSUM_PARTIAL &&
 261                     skb_checksum_help(entskb))
 262                         return NULL;
 263
 264                 data_len = ACCESS_ONCE(queue->copy_range);
 265                 if (data_len == 0 || data_len > entskb->len)
 266                         data_len = entskb->len;
 267
 268                 size += nla_total_size(data_len);
 269                 cap_len = entskb->len;
 270                 break;
 271         }
 272
 273         if (queue->flags & NFQA_CFG_F_CONNTRACK)
 274                 ct = nfqnl_ct_get(entskb, &size, &ctinfo);
 275
 276         skb = alloc_skb(size, GFP_ATOMIC);
 277         if (!skb)
 278                 return NULL;
 279
 280         old_tail = skb->tail;
 281         nlh = nlmsg_put(skb, 0, 0,
 282                         NFNL_SUBSYS_QUEUE << 8 | NFQNL_MSG_PACKET,
 283                         sizeof(struct nfgenmsg), 0);
 284         if (!nlh) {
 285                 kfree_skb(skb);
 286                 return NULL;
 287         }
 288         nfmsg = nlmsg_data(nlh);
 289         nfmsg->nfgen_family = entry->pf;
 290         nfmsg->version = NFNETLINK_V0;
 291         nfmsg->res_id = htons(queue->queue_num);
 292
 293         nla = __nla_reserve(skb, NFQA_PACKET_HDR, sizeof(*pmsg));
 294         pmsg = nla_data(nla);
 295         pmsg->hw_protocol       = entskb->protocol;
 296         pmsg->hook              = entry->hook;
 297         *packet_id_ptr          = &pmsg->packet_id;
 298
 299         indev = entry->indev;
 300         if (indev) {
 301 #ifndef CONFIG_BRIDGE_NETFILTER
 302                 if (nla_put_be32(skb, NFQA_IFINDEX_INDEV, htonl(indev->ifindex)))
 303                         goto nla_put_failure;
 304 #else
 305                 if (entry->pf == PF_BRIDGE) {
 306                         /* Case 1: indev is physical input device, we need to
 307                          * look for bridge group (when called from
 308                          * netfilter_bridge) */
 309                         if (nla_put_be32(skb, NFQA_IFINDEX_PHYSINDEV,
 310                                          htonl(indev->ifindex)) ||
 311                         /* this is the bridge group "brX" */
 312                         /* rcu_read_lock()ed by __nf_queue */
 313                             nla_put_be32(skb, NFQA_IFINDEX_INDEV,
 314                                          htonl(br_port_get_rcu(indev)->br->dev->ifindex)))
 315                                 goto nla_put_failure;
 316                 } else {
 317                         /* Case 2: indev is bridge group, we need to look for
 318                          * physical device (when called from ipv4) */
 319                         if (nla_put_be32(skb, NFQA_IFINDEX_INDEV,
 320                                          htonl(indev->ifindex)))
 321                                 goto nla_put_failure;
 322                         if (entskb->nf_bridge && entskb->nf_bridge->physindev &&
 323                             nla_put_be32(skb, NFQA_IFINDEX_PHYSINDEV,
 324                                          htonl(entskb->nf_bridge->physindev->ifindex)))
 325                                 goto nla_put_failure;
 326                 }
 327 #endif
 328         }
 329
 330         if (outdev) {
 331 #ifndef CONFIG_BRIDGE_NETFILTER
 332                 if (nla_put_be32(skb, NFQA_IFINDEX_OUTDEV, htonl(outdev->ifindex)))
 333                         goto nla_put_failure;
 334 #else
 335                 if (entry->pf == PF_BRIDGE) {
 336                         /* Case 1: outdev is physical output device, we need to
 337                          * look for bridge group (when called from
 338                          * netfilter_bridge) */
 339                         if (nla_put_be32(skb, NFQA_IFINDEX_PHYSOUTDEV,
 340                                          htonl(outdev->ifindex)) ||
 341                         /* this is the bridge group "brX" */
 342                         /* rcu_read_lock()ed by __nf_queue */
 343                             nla_put_be32(skb, NFQA_IFINDEX_OUTDEV,
 344                                          htonl(br_port_get_rcu(outdev)->br->dev->ifindex)))
 345                                 goto nla_put_failure;
 346                 } else {
 347                         /* Case 2: outdev is bridge group, we need to look for
 348                          * physical output device (when called from ipv4) */
 349                         if (nla_put_be32(skb, NFQA_IFINDEX_OUTDEV,
 350                                          htonl(outdev->ifindex)))
 351                                 goto nla_put_failure;
 352                         if (entskb->nf_bridge && entskb->nf_bridge->physoutdev &&
 353                             nla_put_be32(skb, NFQA_IFINDEX_PHYSOUTDEV,
 354                                          htonl(entskb->nf_bridge->physoutdev->ifindex)))
 355                                 goto nla_put_failure;
 356                 }
 357 #endif
 358         }
 359
 360         if (entskb->mark &&
 361             nla_put_be32(skb, NFQA_MARK, htonl(entskb->mark)))
 362                 goto nla_put_failure;
 363
 364         if (indev && entskb->dev &&
 365             entskb->mac_header != entskb->network_header) {
 366                 struct nfqnl_msg_packet_hw phw;
 367                 int len = dev_parse_header(entskb, phw.hw_addr);
 368                 if (len) {
 369                         phw.hw_addrlen = htons(len);
 370                         if (nla_put(skb, NFQA_HWADDR, sizeof(phw), &phw))
 371                                 goto nla_put_failure;
 372                 }
 373         }
 374
 375         if (entskb->tstamp.tv64) {
 376                 struct nfqnl_msg_packet_timestamp ts;
 377                 struct timeval tv = ktime_to_timeval(entskb->tstamp);
 378                 ts.sec = cpu_to_be64(tv.tv_sec);
 379                 ts.usec = cpu_to_be64(tv.tv_usec);
 380
 381                 if (nla_put(skb, NFQA_TIMESTAMP, sizeof(ts), &ts))
 382                         goto nla_put_failure;
 383         }
 384
 385         if (data_len) {
 386                 struct nlattr *nla;
 387                 int sz = nla_attr_size(data_len);
 388
 389                 if (skb_tailroom(skb) < nla_total_size(data_len)) {
 390                         printk(KERN_WARNING "nf_queue: no tailroom!\n");
 391                         kfree_skb(skb);
 392                         return NULL;
 393                 }
 394
 395                 nla = (struct nlattr *)skb_put(skb, nla_total_size(data_len));
 396                 nla->nla_type = NFQA_PAYLOAD;
 397                 nla->nla_len = sz;
 398
 399                 if (skb_copy_bits(entskb, 0, nla_data(nla), data_len))
 400                         BUG();
 401         }
 402
 403         if (ct && nfqnl_ct_put(skb, ct, ctinfo) < 0)
 404                 goto nla_put_failure;
 405
 406         if (cap_len > 0 && nla_put_be32(skb, NFQA_CAP_LEN, htonl(cap_len)))
 407                 goto nla_put_failure;
 408
 409         nlh->nlmsg_len = skb->tail - old_tail;
 410         return skb;
 411
 412 nla_put_failure:
 413         kfree_skb(skb);
 414         net_err_ratelimited("nf_queue: error creating packet message\n");
 415         return NULL;
 416 }
 417
 418 static int
 419 nfqnl_enqueue_packet(struct nf_queue_entry *entry, unsigned int queuenum)
 420 {
 421         struct sk_buff *nskb;
 422         struct nfqnl_instance *queue;
 423         int err = -ENOBUFS;
 424         __be32 *packet_id_ptr;
 425         int failopen = 0;
 426
 427         /* rcu_read_lock()ed by nf_hook_slow() */
 428         queue = instance_lookup(queuenum);
 429         if (!queue) {
 430                 err = -ESRCH;
 431                 goto err_out;
 432         }
 433
 434         if (queue->copy_mode == NFQNL_COPY_NONE) {
 435                 err = -EINVAL;
 436                 goto err_out;
 437         }
 438
 439         nskb = nfqnl_build_packet_message(queue, entry, &packet_id_ptr);
 440         if (nskb == NULL) {
 441                 err = -ENOMEM;
 442                 goto err_out;
 443         }
 444         spin_lock_bh(&queue->lock);
 445
 446         if (!queue->peer_portid) {
 447                 err = -EINVAL;
 448                 goto err_out_free_nskb;
 449         }
 450         if (queue->queue_total >= queue->queue_maxlen) {
 451                 if (queue->flags & NFQA_CFG_F_FAIL_OPEN) {
 452                         failopen = 1;
 453                         err = 0;
 454                 } else {
 455                         queue->queue_dropped++;
 456                         net_warn_ratelimited("nf_queue: full at %d entries, dropping packets(s)\n",
 457                                              queue->queue_total);
 458                 }
 459                 goto err_out_free_nskb;
 460         }
 461         entry->id = ++queue->id_sequence;
 462         *packet_id_ptr = htonl(entry->id);
 463
 464         /* nfnetlink_unicast will either free the nskb or add it to a socket */
 465         err = nfnetlink_unicast(nskb, &init_net, queue->peer_portid, MSG_DONTWAIT);
 466         if (err < 0) {
 467                 queue->queue_user_dropped++;
 468                 goto err_out_unlock;
 469         }
 470
 471         __enqueue_entry(queue, entry);
 472
 473         spin_unlock_bh(&queue->lock);
 474         return 0;
 475
 476 err_out_free_nskb:
 477         kfree_skb(nskb);
 478 err_out_unlock:
 479         spin_unlock_bh(&queue->lock);
 480         if (failopen)
 481                 nf_reinject(entry, NF_ACCEPT);
 482 err_out:
 483         return err;
 484 }
 485
 486 static int
 487 nfqnl_mangle(void *data, int data_len, struct nf_queue_entry *e, int diff)
 488 {
 489         struct sk_buff *nskb;
 490
 491         if (diff < 0) {
 492                 if (pskb_trim(e->skb, data_len))
 493                         return -ENOMEM;
 494         } else if (diff > 0) {
 495                 if (data_len > 0xFFFF)
 496                         return -EINVAL;
 497                 if (diff > skb_tailroom(e->skb)) {
 498                         nskb = skb_copy_expand(e->skb, skb_headroom(e->skb),
 499                                                diff, GFP_ATOMIC);
 500                         if (!nskb) {
 501                                 printk(KERN_WARNING "nf_queue: OOM "
 502                                       "in mangle, dropping packet\n");
 503                                 return -ENOMEM;
 504                         }
 505                         kfree_skb(e->skb);
 506                         e->skb = nskb;
 507                 }
 508                 skb_put(e->skb, diff);
 509         }
 510         if (!skb_make_writable(e->skb, data_len))
 511                 return -ENOMEM;
 512         skb_copy_to_linear_data(e->skb, data, data_len);
 513         e->skb->ip_summed = CHECKSUM_NONE;
 514         return 0;
 515 }
 516
 517 static int
 518 nfqnl_set_mode(struct nfqnl_instance *queue,
 519                unsigned char mode, unsigned int range)
 520 {
 521         int status = 0;
 522
 523         spin_lock_bh(&queue->lock);
 524         switch (mode) {
 525         case NFQNL_COPY_NONE:
 526         case NFQNL_COPY_META:
 527                 queue->copy_mode = mode;
 528                 queue->copy_range = 0;
 529                 break;
 530
 531         case NFQNL_COPY_PACKET:
 532                 queue->copy_mode = mode;
 533                 /* We're using struct nlattr which has 16bit nla_len. Note that
 534                  * nla_len includes the header length. Thus, the maximum packet
 535                  * length that we support is 65531 bytes. We send truncated
 536                  * packets if the specified length is larger than that.
 537                  */
 538                 if (range > 0xffff - NLA_HDRLEN)
 539                         queue->copy_range = 0xffff - NLA_HDRLEN;
 540                 else
 541                         queue->copy_range = range;
 542                 break;
 543
 544         default:
 545                 status = -EINVAL;
 546
 547         }
 548         spin_unlock_bh(&queue->lock);
 549
 550         return status;
 551 }
 552
 553 static int
 554 dev_cmp(struct nf_queue_entry *entry, unsigned long ifindex)
 555 {
 556         if (entry->indev)
 557                 if (entry->indev->ifindex == ifindex)
 558                         return 1;
 559         if (entry->outdev)
 560                 if (entry->outdev->ifindex == ifindex)
 561                         return 1;
 562 #ifdef CONFIG_BRIDGE_NETFILTER
 563         if (entry->skb->nf_bridge) {
 564                 if (entry->skb->nf_bridge->physindev &&
 565                     entry->skb->nf_bridge->physindev->ifindex == ifindex)
 566                         return 1;
 567                 if (entry->skb->nf_bridge->physoutdev &&
 568                     entry->skb->nf_bridge->physoutdev->ifindex == ifindex)
 569                         return 1;
 570         }
 571 #endif
 572         return 0;
 573 }
 574
 575 /* drop all packets with either indev or outdev == ifindex from all queue
 576  * instances */
 577 static void
 578 nfqnl_dev_drop(int ifindex)
 579 {
 580         int i;
 581
 582         rcu_read_lock();
 583
 584         for (i = 0; i < INSTANCE_BUCKETS; i++) {
 585                 struct nfqnl_instance *inst;
 586                 struct hlist_head *head = &instance_table[i];
 587
 588                 hlist_for_each_entry_rcu(inst, head, hlist)
 589                         nfqnl_flush(inst, dev_cmp, ifindex);
 590         }
 591
 592         rcu_read_unlock();
 593 }
 594
 595 #define RCV_SKB_FAIL(err) do { netlink_ack(skb, nlh, (err)); return; } while (0)
 596
 597 static int
 598 nfqnl_rcv_dev_event(struct notifier_block *this,
 599                     unsigned long event, void *ptr)
 600 {
 601         struct net_device *dev = ptr;
 602
 603         if (!net_eq(dev_net(dev), &init_net))
 604                 return NOTIFY_DONE;
 605
 606         /* Drop any packets associated with the downed device */
 607         if (event == NETDEV_DOWN)
 608                 nfqnl_dev_drop(dev->ifindex);
 609         return NOTIFY_DONE;
 610 }
 611
 612 static struct notifier_block nfqnl_dev_notifier = {
 613         .notifier_call  = nfqnl_rcv_dev_event,
 614 };
 615
 616 static int
 617 nfqnl_rcv_nl_event(struct notifier_block *this,
 618                    unsigned long event, void *ptr)
 619 {
 620         struct netlink_notify *n = ptr;
 621
 622         if (event == NETLINK_URELEASE && n->protocol == NETLINK_NETFILTER) {
 623                 int i;
 624
 625                 /* destroy all instances for this portid */
 626                 spin_lock(&instances_lock);
 627                 for (i = 0; i < INSTANCE_BUCKETS; i++) {
 628                         struct hlist_node *t2;
 629                         struct nfqnl_instance *inst;
 630                         struct hlist_head *head = &instance_table[i];
 631
 632                         hlist_for_each_entry_safe(inst, t2, head, hlist) {
 633                                 if ((n->net == &init_net) &&
 634                                     (n->portid == inst->peer_portid))
 635                                         __instance_destroy(inst);
 636                         }
 637                 }
 638                 spin_unlock(&instances_lock);
 639         }
 640         return NOTIFY_DONE;
 641 }
 642
 643 static struct notifier_block nfqnl_rtnl_notifier = {
 644         .notifier_call  = nfqnl_rcv_nl_event,
 645 };
 646
 647 static const struct nla_policy nfqa_verdict_policy[NFQA_MAX+1] = {
 648         [NFQA_VERDICT_HDR]      = { .len = sizeof(struct nfqnl_msg_verdict_hdr) },
 649         [NFQA_MARK]             = { .type = NLA_U32 },
 650         [NFQA_PAYLOAD]          = { .type = NLA_UNSPEC },
 651         [NFQA_CT]               = { .type = NLA_UNSPEC },
 652 };
 653
 654 static const struct nla_policy nfqa_verdict_batch_policy[NFQA_MAX+1] = {
 655         [NFQA_VERDICT_HDR]      = { .len = sizeof(struct nfqnl_msg_verdict_hdr) },
 656         [NFQA_MARK]             = { .type = NLA_U32 },
 657 };
 658
 659 static struct nfqnl_instance *verdict_instance_lookup(u16 queue_num, int nlportid)
 660 {
 661         struct nfqnl_instance *queue;
 662
 663         queue = instance_lookup(queue_num);
 664         if (!queue)
 665                 return ERR_PTR(-ENODEV);
 666
 667         if (queue->peer_portid != nlportid)
 668                 return ERR_PTR(-EPERM);
 669
 670         return queue;
 671 }
 672
 673 static struct nfqnl_msg_verdict_hdr*
 674 verdicthdr_get(const struct nlattr * const nfqa[])
 675 {
 676         struct nfqnl_msg_verdict_hdr *vhdr;
 677         unsigned int verdict;
 678
 679         if (!nfqa[NFQA_VERDICT_HDR])
 680                 return NULL;
 681
 682         vhdr = nla_data(nfqa[NFQA_VERDICT_HDR]);
 683         verdict = ntohl(vhdr->verdict) & NF_VERDICT_MASK;
 684         if (verdict > NF_MAX_VERDICT || verdict == NF_STOLEN)
 685                 return NULL;
 686         return vhdr;
 687 }
 688
 689 static int nfq_id_after(unsigned int id, unsigned int max)
 690 {
 691         return (int)(id - max) > 0;
 692 }
 693
 694 static int
 695 nfqnl_recv_verdict_batch(struct sock *ctnl, struct sk_buff *skb,
 696                    const struct nlmsghdr *nlh,
 697                    const struct nlattr * const nfqa[])
 698 {
 699         struct nfgenmsg *nfmsg = nlmsg_data(nlh);
 700         struct nf_queue_entry *entry, *tmp;
 701         unsigned int verdict, maxid;
 702         struct nfqnl_msg_verdict_hdr *vhdr;
 703         struct nfqnl_instance *queue;
 704         LIST_HEAD(batch_list);
 705         u16 queue_num = ntohs(nfmsg->res_id);
 706
 707         queue = verdict_instance_lookup(queue_num, NETLINK_CB(skb).portid);
 708         if (IS_ERR(queue))
 709                 return PTR_ERR(queue);
 710
 711         vhdr = verdicthdr_get(nfqa);
 712         if (!vhdr)
 713                 return -EINVAL;
 714
 715         verdict = ntohl(vhdr->verdict);
 716         maxid = ntohl(vhdr->id);
 717
 718         spin_lock_bh(&queue->lock);
 719
 720         list_for_each_entry_safe(entry, tmp, &queue->queue_list, list) {
 721                 if (nfq_id_after(entry->id, maxid))
 722                         break;
 723                 __dequeue_entry(queue, entry);
 724                 list_add_tail(&entry->list, &batch_list);
 725         }
 726
 727         spin_unlock_bh(&queue->lock);
 728
 729         if (list_empty(&batch_list))
 730                 return -ENOENT;
 731
 732         list_for_each_entry_safe(entry, tmp, &batch_list, list) {
 733                 if (nfqa[NFQA_MARK])
 734                         entry->skb->mark = ntohl(nla_get_be32(nfqa[NFQA_MARK]));
 735                 nf_reinject(entry, verdict);
 736         }
 737         return 0;
 738 }
 739
 740 static int
 741 nfqnl_recv_verdict(struct sock *ctnl, struct sk_buff *skb,
 742                    const struct nlmsghdr *nlh,
 743                    const struct nlattr * const nfqa[])
 744 {
 745         struct nfgenmsg *nfmsg = nlmsg_data(nlh);
 746         u_int16_t queue_num = ntohs(nfmsg->res_id);
 747
 748         struct nfqnl_msg_verdict_hdr *vhdr;
 749         struct nfqnl_instance *queue;
 750         unsigned int verdict;
 751         struct nf_queue_entry *entry;
 752         enum ip_conntrack_info uninitialized_var(ctinfo);
 753         struct nf_conn *ct = NULL;
 754
 755         queue = instance_lookup(queue_num);
 756         if (!queue)
 757
 758         queue = verdict_instance_lookup(queue_num, NETLINK_CB(skb).portid);
 759         if (IS_ERR(queue))
 760                 return PTR_ERR(queue);
 761
 762         vhdr = verdicthdr_get(nfqa);
 763         if (!vhdr)
 764                 return -EINVAL;
 765
 766         verdict = ntohl(vhdr->verdict);
 767
 768         entry = find_dequeue_entry(queue, ntohl(vhdr->id));
 769         if (entry == NULL)
 770                 return -ENOENT;
 771
 772         rcu_read_lock();
 773         if (nfqa[NFQA_CT] && (queue->flags & NFQA_CFG_F_CONNTRACK))
 774                 ct = nfqnl_ct_parse(entry->skb, nfqa[NFQA_CT], &ctinfo);
 775
 776         if (nfqa[NFQA_PAYLOAD]) {
 777                 u16 payload_len = nla_len(nfqa[NFQA_PAYLOAD]);
 778                 int diff = payload_len - entry->skb->len;
 779
 780                 if (nfqnl_mangle(nla_data(nfqa[NFQA_PAYLOAD]),
 781                                  payload_len, entry, diff) < 0)
 782                         verdict = NF_DROP;
 783
 784                 if (ct)
 785                         nfqnl_ct_seq_adjust(skb, ct, ctinfo, diff);
 786         }
 787         rcu_read_unlock();
 788
 789         if (nfqa[NFQA_MARK])
 790                 entry->skb->mark = ntohl(nla_get_be32(nfqa[NFQA_MARK]));
 791
 792         nf_reinject(entry, verdict);
 793         return 0;
 794 }
 795
 796 static int
 797 nfqnl_recv_unsupp(struct sock *ctnl, struct sk_buff *skb,
 798                   const struct nlmsghdr *nlh,
 799                   const struct nlattr * const nfqa[])
 800 {
 801         return -ENOTSUPP;
 802 }
 803
 804 static const struct nla_policy nfqa_cfg_policy[NFQA_CFG_MAX+1] = {
 805         [NFQA_CFG_CMD]          = { .len = sizeof(struct nfqnl_msg_config_cmd) },
 806         [NFQA_CFG_PARAMS]       = { .len = sizeof(struct nfqnl_msg_config_params) },
 807 };
 808
 809 static const struct nf_queue_handler nfqh = {
 810         .outfn  = &nfqnl_enqueue_packet,
 811 };
 812
 813 static int
 814 nfqnl_recv_config(struct sock *ctnl, struct sk_buff *skb,
 815                   const struct nlmsghdr *nlh,
 816                   const struct nlattr * const nfqa[])
 817 {
 818         struct nfgenmsg *nfmsg = nlmsg_data(nlh);
 819         u_int16_t queue_num = ntohs(nfmsg->res_id);
 820         struct nfqnl_instance *queue;
 821         struct nfqnl_msg_config_cmd *cmd = NULL;
 822         int ret = 0;
 823
 824         if (nfqa[NFQA_CFG_CMD]) {
 825                 cmd = nla_data(nfqa[NFQA_CFG_CMD]);
 826
 827                 /* Obsolete commands without queue context */
 828                 switch (cmd->command) {
 829                 case NFQNL_CFG_CMD_PF_BIND: return 0;
 830                 case NFQNL_CFG_CMD_PF_UNBIND: return 0;
 831                 }
 832         }
 833
 834         rcu_read_lock();
 835         queue = instance_lookup(queue_num);
 836         if (queue && queue->peer_portid != NETLINK_CB(skb).portid) {
 837                 ret = -EPERM;
 838                 goto err_out_unlock;
 839         }
 840
 841         if (cmd != NULL) {
 842                 switch (cmd->command) {
 843                 case NFQNL_CFG_CMD_BIND:
 844                         if (queue) {
 845                                 ret = -EBUSY;
 846                                 goto err_out_unlock;
 847                         }
 848                         queue = instance_create(queue_num, NETLINK_CB(skb).portid);
 849                         if (IS_ERR(queue)) {
 850                                 ret = PTR_ERR(queue);
 851                                 goto err_out_unlock;
 852                         }
 853                         break;
 854                 case NFQNL_CFG_CMD_UNBIND:
 855                         if (!queue) {
 856                                 ret = -ENODEV;
 857                                 goto err_out_unlock;
 858                         }
 859                         instance_destroy(queue);
 860                         break;
 861                 case NFQNL_CFG_CMD_PF_BIND:
 862                 case NFQNL_CFG_CMD_PF_UNBIND:
 863                         break;
 864                 default:
 865                         ret = -ENOTSUPP;
 866                         break;
 867                 }
 868         }
 869
 870         if (nfqa[NFQA_CFG_PARAMS]) {
 871                 struct nfqnl_msg_config_params *params;
 872
 873                 if (!queue) {
 874                         ret = -ENODEV;
 875                         goto err_out_unlock;
 876                 }
 877                 params = nla_data(nfqa[NFQA_CFG_PARAMS]);
 878                 nfqnl_set_mode(queue, params->copy_mode,
 879                                 ntohl(params->copy_range));
 880         }
 881
 882         if (nfqa[NFQA_CFG_QUEUE_MAXLEN]) {
 883                 __be32 *queue_maxlen;
 884
 885                 if (!queue) {
 886                         ret = -ENODEV;
 887                         goto err_out_unlock;
 888                 }
 889                 queue_maxlen = nla_data(nfqa[NFQA_CFG_QUEUE_MAXLEN]);
 890                 spin_lock_bh(&queue->lock);
 891                 queue->queue_maxlen = ntohl(*queue_maxlen);
 892                 spin_unlock_bh(&queue->lock);
 893         }
 894
 895         if (nfqa[NFQA_CFG_FLAGS]) {
 896                 __u32 flags, mask;
 897
 898                 if (!queue) {
 899                         ret = -ENODEV;
 900                         goto err_out_unlock;
 901                 }
 902
 903                 if (!nfqa[NFQA_CFG_MASK]) {
 904                         /* A mask is needed to specify which flags are being
 905                          * changed.
 906                          */
 907                         ret = -EINVAL;
 908                         goto err_out_unlock;
 909                 }
 910
 911                 flags = ntohl(nla_get_be32(nfqa[NFQA_CFG_FLAGS]));
 912                 mask = ntohl(nla_get_be32(nfqa[NFQA_CFG_MASK]));
 913
 914                 if (flags >= NFQA_CFG_F_MAX) {
 915                         ret = -EOPNOTSUPP;
 916                         goto err_out_unlock;
 917                 }
 918
 919                 spin_lock_bh(&queue->lock);
 920                 queue->flags &= ~mask;
 921                 queue->flags |= flags & mask;
 922                 spin_unlock_bh(&queue->lock);
 923         }
 924
 925 err_out_unlock:
 926         rcu_read_unlock();
 927         return ret;
 928 }
 929
 930 static const struct nfnl_callback nfqnl_cb[NFQNL_MSG_MAX] = {
 931         [NFQNL_MSG_PACKET]      = { .call_rcu = nfqnl_recv_unsupp,
 932                                     .attr_count = NFQA_MAX, },
 933         [NFQNL_MSG_VERDICT]     = { .call_rcu = nfqnl_recv_verdict,
 934                                     .attr_count = NFQA_MAX,
 935                                     .policy = nfqa_verdict_policy },
 936         [NFQNL_MSG_CONFIG]      = { .call = nfqnl_recv_config,
 937                                     .attr_count = NFQA_CFG_MAX,
 938                                     .policy = nfqa_cfg_policy },
 939         [NFQNL_MSG_VERDICT_BATCH]={ .call_rcu = nfqnl_recv_verdict_batch,
 940                                     .attr_count = NFQA_MAX,
 941                                     .policy = nfqa_verdict_batch_policy },
 942 };
 943
 944 static const struct nfnetlink_subsystem nfqnl_subsys = {
 945         .name           = "nf_queue",
 946         .subsys_id      = NFNL_SUBSYS_QUEUE,
 947         .cb_count       = NFQNL_MSG_MAX,
 948         .cb             = nfqnl_cb,
 949 };
 950
 951 #ifdef CONFIG_PROC_FS
 952 struct iter_state {
 953         unsigned int bucket;
 954 };
 955
 956 static struct hlist_node *get_first(struct seq_file *seq)
 957 {
 958         struct iter_state *st = seq->private;
 959
 960         if (!st)
 961                 return NULL;
 962
 963         for (st->bucket = 0; st->bucket < INSTANCE_BUCKETS; st->bucket++) {
 964                 if (!hlist_empty(&instance_table[st->bucket]))
 965                         return instance_table[st->bucket].first;
 966         }
 967         return NULL;
 968 }
 969
 970 static struct hlist_node *get_next(struct seq_file *seq, struct hlist_node *h)
 971 {
 972         struct iter_state *st = seq->private;
 973
 974         h = h->next;
 975         while (!h) {
 976                 if (++st->bucket >= INSTANCE_BUCKETS)
 977                         return NULL;
 978
 979                 h = instance_table[st->bucket].first;
 980         }
 981         return h;
 982 }
 983
 984 static struct hlist_node *get_idx(struct seq_file *seq, loff_t pos)
 985 {
 986         struct hlist_node *head;
 987         head = get_first(seq);
 988
 989         if (head)
 990                 while (pos && (head = get_next(seq, head)))
 991                         pos--;
 992         return pos ? NULL : head;
 993 }
 994
 995 static void *seq_start(struct seq_file *seq, loff_t *pos)
 996         __acquires(instances_lock)
 997 {
 998         spin_lock(&instances_lock);
 999         return get_idx(seq, *pos);
1000 }
1001
1002 static void *seq_next(struct seq_file *s, void *v, loff_t *pos)
1003 {
1004         (*pos)++;
1005         return get_next(s, v);
1006 }
1007
1008 static void seq_stop(struct seq_file *s, void *v)
1009         __releases(instances_lock)
1010 {
1011         spin_unlock(&instances_lock);
1012 }
1013
1014 static int seq_show(struct seq_file *s, void *v)
1015 {
1016         const struct nfqnl_instance *inst = v;
1017
1018         return seq_printf(s, "%5d %6d %5d %1d %5d %5d %5d %8d %2d\n",
1019                           inst->queue_num,
1020                           inst->peer_portid, inst->queue_total,
1021                           inst->copy_mode, inst->copy_range,
1022                           inst->queue_dropped, inst->queue_user_dropped,
1023                           inst->id_sequence, 1);
1024 }
1025
1026 static const struct seq_operations nfqnl_seq_ops = {
1027         .start  = seq_start,
1028         .next   = seq_next,
1029         .stop   = seq_stop,
1030         .show   = seq_show,
1031 };
1032
1033 static int nfqnl_open(struct inode *inode, struct file *file)
1034 {
1035         return seq_open_private(file, &nfqnl_seq_ops,
1036                         sizeof(struct iter_state));
1037 }
1038
1039 static const struct file_operations nfqnl_file_ops = {
1040         .owner   = THIS_MODULE,
1041         .open    = nfqnl_open,
1042         .read    = seq_read,
1043         .llseek  = seq_lseek,
1044         .release = seq_release_private,
1045 };
1046
1047 #endif /* PROC_FS */
1048
1049 static int __init nfnetlink_queue_init(void)
1050 {
1051         int i, status = -ENOMEM;
1052
1053         for (i = 0; i < INSTANCE_BUCKETS; i++)
1054                 INIT_HLIST_HEAD(&instance_table[i]);
1055
1056         netlink_register_notifier(&nfqnl_rtnl_notifier);
1057         status = nfnetlink_subsys_register(&nfqnl_subsys);
1058         if (status < 0) {
1059                 printk(KERN_ERR "nf_queue: failed to create netlink socket\n");
1060                 goto cleanup_netlink_notifier;
1061         }
1062
1063 #ifdef CONFIG_PROC_FS
1064         if (!proc_create("nfnetlink_queue", 0440,
1065                          proc_net_netfilter, &nfqnl_file_ops))
1066                 goto cleanup_subsys;
1067 #endif
1068
1069         register_netdevice_notifier(&nfqnl_dev_notifier);
1070         nf_register_queue_handler(&nfqh);
1071         return status;
1072
1073 #ifdef CONFIG_PROC_FS
1074 cleanup_subsys:
1075         nfnetlink_subsys_unregister(&nfqnl_subsys);
1076 #endif
1077 cleanup_netlink_notifier:
1078         netlink_unregister_notifier(&nfqnl_rtnl_notifier);
1079         return status;
1080 }
1081
1082 static void __exit nfnetlink_queue_fini(void)
1083 {
1084         nf_unregister_queue_handler();
1085         unregister_netdevice_notifier(&nfqnl_dev_notifier);
1086 #ifdef CONFIG_PROC_FS
1087         remove_proc_entry("nfnetlink_queue", proc_net_netfilter);
1088 #endif
1089         nfnetlink_subsys_unregister(&nfqnl_subsys);
1090         netlink_unregister_notifier(&nfqnl_rtnl_notifier);
1091
1092         rcu_barrier(); /* Wait for completion of call_rcu()'s */
1093 }
1094
1095 MODULE_DESCRIPTION("netfilter packet queue handler");
1096 MODULE_AUTHOR("Harald Welte <laforge@netfilter.org>");
1097 MODULE_LICENSE("GPL");
1098 MODULE_ALIAS_NFNL_SUBSYS(NFNL_SUBSYS_QUEUE);
1099
1100 module_init(nfnetlink_queue_init);
1101 module_exit(nfnetlink_queue_fini);