net/ipv4/netfilter/ip_queue.c

   1 /*
   2  * This is a module which is used for queueing IPv4 packets and
   3  * communicating with userspace via netlink.
   4  *
   5  * (C) 2000-2002 James Morris <jmorris@intercode.com.au>
   6  * (C) 2003-2005 Netfilter Core Team <coreteam@netfilter.org>
   7  *
   8  * This program is free software; you can redistribute it and/or modify
   9  * it under the terms of the GNU General Public License version 2 as
  10  * published by the Free Software Foundation.
  11  */
  12 #include <linux/module.h>
  13 #include <linux/skbuff.h>
  14 #include <linux/init.h>
  15 #include <linux/ip.h>
  16 #include <linux/notifier.h>
  17 #include <linux/netdevice.h>
  18 #include <linux/netfilter.h>
  19 #include <linux/netfilter_ipv4/ip_queue.h>
  20 #include <linux/netfilter_ipv4/ip_tables.h>
  21 #include <linux/netlink.h>
  22 #include <linux/spinlock.h>
  23 #include <linux/sysctl.h>
  24 #include <linux/proc_fs.h>
  25 #include <linux/seq_file.h>
  26 #include <linux/security.h>
  27 #include <linux/net.h>
  28 #include <linux/mutex.h>
  29 #include <linux/slab.h>
  30 #include <net/net_namespace.h>
  31 #include <net/sock.h>
  32 #include <net/route.h>
  33 #include <net/netfilter/nf_queue.h>
  34 #include <net/ip.h>
  35
  36 #define IPQ_QMAX_DEFAULT 1024
  37 #define IPQ_PROC_FS_NAME "ip_queue"
  38 #define NET_IPQ_QMAX 2088
  39 #define NET_IPQ_QMAX_NAME "ip_queue_maxlen"
  40
  41 typedef int (*ipq_cmpfn)(struct nf_queue_entry *, unsigned long);
  42
  43 static unsigned char copy_mode __read_mostly = IPQ_COPY_NONE;
  44 static unsigned int queue_maxlen __read_mostly = IPQ_QMAX_DEFAULT;
  45 static DEFINE_RWLOCK(queue_lock);
  46 static int peer_pid __read_mostly;
  47 static unsigned int copy_range __read_mostly;
  48 static unsigned int queue_total;
  49 static unsigned int queue_dropped = 0;
  50 static unsigned int queue_user_dropped = 0;
  51 static struct sock *ipqnl __read_mostly;
  52 static LIST_HEAD(queue_list);
  53 static DEFINE_MUTEX(ipqnl_mutex);
  54
  55 static inline void
  56 __ipq_enqueue_entry(struct nf_queue_entry *entry)
  57 {
  58        list_add_tail(&entry->list, &queue_list);
  59        queue_total++;
  60 }
  61
  62 static inline int
  63 __ipq_set_mode(unsigned char mode, unsigned int range)
  64 {
  65         int status = 0;
  66
  67         switch(mode) {
  68         case IPQ_COPY_NONE:
  69         case IPQ_COPY_META:
  70                 copy_mode = mode;
  71                 copy_range = 0;
  72                 break;
  73
  74         case IPQ_COPY_PACKET:
  75                 copy_mode = mode;
  76                 copy_range = range;
  77                 if (copy_range > 0xFFFF)
  78                         copy_range = 0xFFFF;
  79                 break;
  80
  81         default:
  82                 status = -EINVAL;
  83
  84         }
  85         return status;
  86 }
  87
  88 static void __ipq_flush(ipq_cmpfn cmpfn, unsigned long data);
  89
  90 static inline void
  91 __ipq_reset(void)
  92 {
  93         peer_pid = 0;
  94         net_disable_timestamp();
  95         __ipq_set_mode(IPQ_COPY_NONE, 0);
  96         __ipq_flush(NULL, 0);
  97 }
  98
  99 static struct nf_queue_entry *
 100 ipq_find_dequeue_entry(unsigned long id)
 101 {
 102         struct nf_queue_entry *entry = NULL, *i;
 103
 104         write_lock_bh(&queue_lock);
 105
 106         list_for_each_entry(i, &queue_list, list) {
 107                 if ((unsigned long)i == id) {
 108                         entry = i;
 109                         break;
 110                 }
 111         }
 112
 113         if (entry) {
 114                 list_del(&entry->list);
 115                 queue_total--;
 116         }
 117
 118         write_unlock_bh(&queue_lock);
 119         return entry;
 120 }
 121
 122 static void
 123 __ipq_flush(ipq_cmpfn cmpfn, unsigned long data)
 124 {
 125         struct nf_queue_entry *entry, *next;
 126
 127         list_for_each_entry_safe(entry, next, &queue_list, list) {
 128                 if (!cmpfn || cmpfn(entry, data)) {
 129                         list_del(&entry->list);
 130                         queue_total--;
 131                         nf_reinject(entry, NF_DROP);
 132                 }
 133         }
 134 }
 135
 136 static void
 137 ipq_flush(ipq_cmpfn cmpfn, unsigned long data)
 138 {
 139         write_lock_bh(&queue_lock);
 140         __ipq_flush(cmpfn, data);
 141         write_unlock_bh(&queue_lock);
 142 }
 143
 144 static struct sk_buff *
 145 ipq_build_packet_message(struct nf_queue_entry *entry, int *errp)
 146 {
 147         sk_buff_data_t old_tail;
 148         size_t size = 0;
 149         size_t data_len = 0;
 150         struct sk_buff *skb;
 151         struct ipq_packet_msg *pmsg;
 152         struct nlmsghdr *nlh;
 153         struct timeval tv;
 154
 155         read_lock_bh(&queue_lock);
 156
 157         switch (copy_mode) {
 158         case IPQ_COPY_META:
 159         case IPQ_COPY_NONE:
 160                 size = NLMSG_SPACE(sizeof(*pmsg));
 161                 break;
 162
 163         case IPQ_COPY_PACKET:
 164                 if (entry->skb->ip_summed == CHECKSUM_PARTIAL &&
 165                     (*errp = skb_checksum_help(entry->skb))) {
 166                         read_unlock_bh(&queue_lock);
 167                         return NULL;
 168                 }
 169                 if (copy_range == 0 || copy_range > entry->skb->len)
 170                         data_len = entry->skb->len;
 171                 else
 172                         data_len = copy_range;
 173
 174                 size = NLMSG_SPACE(sizeof(*pmsg) + data_len);
 175                 break;
 176
 177         default:
 178                 *errp = -EINVAL;
 179                 read_unlock_bh(&queue_lock);
 180                 return NULL;
 181         }
 182
 183         read_unlock_bh(&queue_lock);
 184
 185         skb = alloc_skb(size, GFP_ATOMIC);
 186         if (!skb)
 187                 goto nlmsg_failure;
 188
 189         old_tail = skb->tail;
 190         nlh = NLMSG_PUT(skb, 0, 0, IPQM_PACKET, size - sizeof(*nlh));
 191         pmsg = NLMSG_DATA(nlh);
 192         memset(pmsg, 0, sizeof(*pmsg));
 193
 194         pmsg->packet_id       = (unsigned long )entry;
 195         pmsg->data_len        = data_len;
 196         tv = ktime_to_timeval(entry->skb->tstamp);
 197         pmsg->timestamp_sec   = tv.tv_sec;
 198         pmsg->timestamp_usec  = tv.tv_usec;
 199         pmsg->mark            = entry->skb->mark;
 200         pmsg->hook            = entry->hook;
 201         pmsg->hw_protocol     = entry->skb->protocol;
 202
 203         if (entry->indev)
 204                 strcpy(pmsg->indev_name, entry->indev->name);
 205         else
 206                 pmsg->indev_name[0] = '\0';
 207
 208         if (entry->outdev)
 209                 strcpy(pmsg->outdev_name, entry->outdev->name);
 210         else
 211                 pmsg->outdev_name[0] = '\0';
 212
 213         if (entry->indev && entry->skb->dev) {
 214                 pmsg->hw_type = entry->skb->dev->type;
 215                 pmsg->hw_addrlen = dev_parse_header(entry->skb,
 216                                                     pmsg->hw_addr);
 217         }
 218
 219         if (data_len)
 220                 if (skb_copy_bits(entry->skb, 0, pmsg->payload, data_len))
 221                         BUG();
 222
 223         nlh->nlmsg_len = skb->tail - old_tail;
 224         return skb;
 225
 226 nlmsg_failure:
 227         *errp = -EINVAL;
 228         printk(KERN_ERR "ip_queue: error creating packet message\n");
 229         return NULL;
 230 }
 231
 232 static int
 233 ipq_enqueue_packet(struct nf_queue_entry *entry, unsigned int queuenum)
 234 {
 235         int status = -EINVAL;
 236         struct sk_buff *nskb;
 237
 238         if (copy_mode == IPQ_COPY_NONE)
 239                 return -EAGAIN;
 240
 241         nskb = ipq_build_packet_message(entry, &status);
 242         if (nskb == NULL)
 243                 return status;
 244
 245         write_lock_bh(&queue_lock);
 246
 247         if (!peer_pid)
 248                 goto err_out_free_nskb;
 249
 250         if (queue_total >= queue_maxlen) {
 251                 queue_dropped++;
 252                 status = -ENOSPC;
 253                 if (net_ratelimit())
 254                           printk (KERN_WARNING "ip_queue: full at %d entries, "
 255                                   "dropping packets(s). Dropped: %d\n", queue_total,
 256                                   queue_dropped);
 257                 goto err_out_free_nskb;
 258         }
 259
 260         /* netlink_unicast will either free the nskb or attach it to a socket */
 261         status = netlink_unicast(ipqnl, nskb, peer_pid, MSG_DONTWAIT);
 262         if (status < 0) {
 263                 queue_user_dropped++;
 264                 goto err_out_unlock;
 265         }
 266
 267         __ipq_enqueue_entry(entry);
 268
 269         write_unlock_bh(&queue_lock);
 270         return status;
 271
 272 err_out_free_nskb:
 273         kfree_skb(nskb);
 274
 275 err_out_unlock:
 276         write_unlock_bh(&queue_lock);
 277         return status;
 278 }
 279
 280 static int
 281 ipq_mangle_ipv4(ipq_verdict_msg_t *v, struct nf_queue_entry *e)
 282 {
 283         int diff;
 284         struct iphdr *user_iph = (struct iphdr *)v->payload;
 285         struct sk_buff *nskb;
 286
 287         if (v->data_len < sizeof(*user_iph))
 288                 return 0;
 289         diff = v->data_len - e->skb->len;
 290         if (diff < 0) {
 291                 if (pskb_trim(e->skb, v->data_len))
 292                         return -ENOMEM;
 293         } else if (diff > 0) {
 294                 if (v->data_len > 0xFFFF)
 295                         return -EINVAL;
 296                 if (diff > skb_tailroom(e->skb)) {
 297                         nskb = skb_copy_expand(e->skb, skb_headroom(e->skb),
 298                                                diff, GFP_ATOMIC);
 299                         if (!nskb) {
 300                                 printk(KERN_WARNING "ip_queue: error "
 301                                       "in mangle, dropping packet\n");
 302                                 return -ENOMEM;
 303                         }
 304                         kfree_skb(e->skb);
 305                         e->skb = nskb;
 306                 }
 307                 skb_put(e->skb, diff);
 308         }
 309         if (!skb_make_writable(e->skb, v->data_len))
 310                 return -ENOMEM;
 311         skb_copy_to_linear_data(e->skb, v->payload, v->data_len);
 312         e->skb->ip_summed = CHECKSUM_NONE;
 313
 314         return 0;
 315 }
 316
 317 static int
 318 ipq_set_verdict(struct ipq_verdict_msg *vmsg, unsigned int len)
 319 {
 320         struct nf_queue_entry *entry;
 321
 322         if (vmsg->value > NF_MAX_VERDICT)
 323                 return -EINVAL;
 324
 325         entry = ipq_find_dequeue_entry(vmsg->id);
 326         if (entry == NULL)
 327                 return -ENOENT;
 328         else {
 329                 int verdict = vmsg->value;
 330
 331                 if (vmsg->data_len && vmsg->data_len == len)
 332                         if (ipq_mangle_ipv4(vmsg, entry) < 0)
 333                                 verdict = NF_DROP;
 334
 335                 nf_reinject(entry, verdict);
 336                 return 0;
 337         }
 338 }
 339
 340 static int
 341 ipq_set_mode(unsigned char mode, unsigned int range)
 342 {
 343         int status;
 344
 345         write_lock_bh(&queue_lock);
 346         status = __ipq_set_mode(mode, range);
 347         write_unlock_bh(&queue_lock);
 348         return status;
 349 }
 350
 351 static int
 352 ipq_receive_peer(struct ipq_peer_msg *pmsg,
 353                  unsigned char type, unsigned int len)
 354 {
 355         int status = 0;
 356
 357         if (len < sizeof(*pmsg))
 358                 return -EINVAL;
 359
 360         switch (type) {
 361         case IPQM_MODE:
 362                 status = ipq_set_mode(pmsg->msg.mode.value,
 363                                       pmsg->msg.mode.range);
 364                 break;
 365
 366         case IPQM_VERDICT:
 367                 if (pmsg->msg.verdict.value > NF_MAX_VERDICT)
 368                         status = -EINVAL;
 369                 else
 370                         status = ipq_set_verdict(&pmsg->msg.verdict,
 371                                                  len - sizeof(*pmsg));
 372                         break;
 373         default:
 374                 status = -EINVAL;
 375         }
 376         return status;
 377 }
 378
 379 static int
 380 dev_cmp(struct nf_queue_entry *entry, unsigned long ifindex)
 381 {
 382         if (entry->indev)
 383                 if (entry->indev->ifindex == ifindex)
 384                         return 1;
 385         if (entry->outdev)
 386                 if (entry->outdev->ifindex == ifindex)
 387                         return 1;
 388 #ifdef CONFIG_BRIDGE_NETFILTER
 389         if (entry->skb->nf_bridge) {
 390                 if (entry->skb->nf_bridge->physindev &&
 391                     entry->skb->nf_bridge->physindev->ifindex == ifindex)
 392                         return 1;
 393                 if (entry->skb->nf_bridge->physoutdev &&
 394                     entry->skb->nf_bridge->physoutdev->ifindex == ifindex)
 395                         return 1;
 396         }
 397 #endif
 398         return 0;
 399 }
 400
 401 static void
 402 ipq_dev_drop(int ifindex)
 403 {
 404         ipq_flush(dev_cmp, ifindex);
 405 }
 406
 407 #define RCV_SKB_FAIL(err) do { netlink_ack(skb, nlh, (err)); return; } while (0)
 408
 409 static inline void
 410 __ipq_rcv_skb(struct sk_buff *skb)
 411 {
 412         int status, type, pid, flags, nlmsglen, skblen;
 413         struct nlmsghdr *nlh;
 414
 415         skblen = skb->len;
 416         if (skblen < sizeof(*nlh))
 417                 return;
 418
 419         nlh = nlmsg_hdr(skb);
 420         nlmsglen = nlh->nlmsg_len;
 421         if (nlmsglen < sizeof(*nlh) || skblen < nlmsglen)
 422                 return;
 423
 424         pid = nlh->nlmsg_pid;
 425         flags = nlh->nlmsg_flags;
 426
 427         if(pid <= 0 || !(flags & NLM_F_REQUEST) || flags & NLM_F_MULTI)
 428                 RCV_SKB_FAIL(-EINVAL);
 429
 430         if (flags & MSG_TRUNC)
 431                 RCV_SKB_FAIL(-ECOMM);
 432
 433         type = nlh->nlmsg_type;
 434         if (type < NLMSG_NOOP || type >= IPQM_MAX)
 435                 RCV_SKB_FAIL(-EINVAL);
 436
 437         if (type <= IPQM_BASE)
 438                 return;
 439
 440         if (security_netlink_recv(skb, CAP_NET_ADMIN))
 441                 RCV_SKB_FAIL(-EPERM);
 442
 443         write_lock_bh(&queue_lock);
 444
 445         if (peer_pid) {
 446                 if (peer_pid != pid) {
 447                         write_unlock_bh(&queue_lock);
 448                         RCV_SKB_FAIL(-EBUSY);
 449                 }
 450         } else {
 451                 net_enable_timestamp();
 452                 peer_pid = pid;
 453         }
 454
 455         write_unlock_bh(&queue_lock);
 456
 457         status = ipq_receive_peer(NLMSG_DATA(nlh), type,
 458                                   nlmsglen - NLMSG_LENGTH(0));
 459         if (status < 0)
 460                 RCV_SKB_FAIL(status);
 461
 462         if (flags & NLM_F_ACK)
 463                 netlink_ack(skb, nlh, 0);
 464 }
 465
 466 static void
 467 ipq_rcv_skb(struct sk_buff *skb)
 468 {
 469         mutex_lock(&ipqnl_mutex);
 470         __ipq_rcv_skb(skb);
 471         mutex_unlock(&ipqnl_mutex);
 472 }
 473
 474 static int
 475 ipq_rcv_dev_event(struct notifier_block *this,
 476                   unsigned long event, void *ptr)
 477 {
 478         struct net_device *dev = ptr;
 479
 480         if (!net_eq(dev_net(dev), &init_net))
 481                 return NOTIFY_DONE;
 482
 483         /* Drop any packets associated with the downed device */
 484         if (event == NETDEV_DOWN)
 485                 ipq_dev_drop(dev->ifindex);
 486         return NOTIFY_DONE;
 487 }
 488
 489 static struct notifier_block ipq_dev_notifier = {
 490         .notifier_call  = ipq_rcv_dev_event,
 491 };
 492
 493 static int
 494 ipq_rcv_nl_event(struct notifier_block *this,
 495                  unsigned long event, void *ptr)
 496 {
 497         struct netlink_notify *n = ptr;
 498
 499         if (event == NETLINK_URELEASE && n->protocol == NETLINK_FIREWALL) {
 500                 write_lock_bh(&queue_lock);
 501                 if ((net_eq(n->net, &init_net)) && (n->pid == peer_pid))
 502                         __ipq_reset();
 503                 write_unlock_bh(&queue_lock);
 504         }
 505         return NOTIFY_DONE;
 506 }
 507
 508 static struct notifier_block ipq_nl_notifier = {
 509         .notifier_call  = ipq_rcv_nl_event,
 510 };
 511
 512 #ifdef CONFIG_SYSCTL
 513 static struct ctl_table_header *ipq_sysctl_header;
 514
 515 static ctl_table ipq_table[] = {
 516         {
 517                 .procname       = NET_IPQ_QMAX_NAME,
 518                 .data           = &queue_maxlen,
 519                 .maxlen         = sizeof(queue_maxlen),
 520                 .mode           = 0644,
 521                 .proc_handler   = proc_dointvec
 522         },
 523         { }
 524 };
 525 #endif
 526
 527 #ifdef CONFIG_PROC_FS
 528 static int ip_queue_show(struct seq_file *m, void *v)
 529 {
 530         read_lock_bh(&queue_lock);
 531
 532         seq_printf(m,
 533                       "Peer PID          : %d\n"
 534                       "Copy mode         : %hu\n"
 535                       "Copy range        : %u\n"
 536                       "Queue length      : %u\n"
 537                       "Queue max. length : %u\n"
 538                       "Queue dropped     : %u\n"
 539                       "Netlink dropped   : %u\n",
 540                       peer_pid,
 541                       copy_mode,
 542                       copy_range,
 543                       queue_total,
 544                       queue_maxlen,
 545                       queue_dropped,
 546                       queue_user_dropped);
 547
 548         read_unlock_bh(&queue_lock);
 549         return 0;
 550 }
 551
 552 static int ip_queue_open(struct inode *inode, struct file *file)
 553 {
 554         return single_open(file, ip_queue_show, NULL);
 555 }
 556
 557 static const struct file_operations ip_queue_proc_fops = {
 558         .open           = ip_queue_open,
 559         .read           = seq_read,
 560         .llseek         = seq_lseek,
 561         .release        = single_release,
 562         .owner          = THIS_MODULE,
 563 };
 564 #endif
 565
 566 static const struct nf_queue_handler nfqh = {
 567         .name   = "ip_queue",
 568         .outfn  = &ipq_enqueue_packet,
 569 };
 570
 571 static int __init ip_queue_init(void)
 572 {
 573         int status = -ENOMEM;
 574         struct proc_dir_entry *proc __maybe_unused;
 575
 576         netlink_register_notifier(&ipq_nl_notifier);
 577         ipqnl = netlink_kernel_create(&init_net, NETLINK_FIREWALL, 0,
 578                                       ipq_rcv_skb, NULL, THIS_MODULE);
 579         if (ipqnl == NULL) {
 580                 printk(KERN_ERR "ip_queue: failed to create netlink socket\n");
 581                 goto cleanup_netlink_notifier;
 582         }
 583
 584 #ifdef CONFIG_PROC_FS
 585         proc = proc_create(IPQ_PROC_FS_NAME, 0, init_net.proc_net,
 586                            &ip_queue_proc_fops);
 587         if (!proc) {
 588                 printk(KERN_ERR "ip_queue: failed to create proc entry\n");
 589                 goto cleanup_ipqnl;
 590         }
 591 #endif
 592         register_netdevice_notifier(&ipq_dev_notifier);
 593 #ifdef CONFIG_SYSCTL
 594         ipq_sysctl_header = register_sysctl_paths(net_ipv4_ctl_path, ipq_table);
 595 #endif
 596         status = nf_register_queue_handler(NFPROTO_IPV4, &nfqh);
 597         if (status < 0) {
 598                 printk(KERN_ERR "ip_queue: failed to register queue handler\n");
 599                 goto cleanup_sysctl;
 600         }
 601         return status;
 602
 603 cleanup_sysctl:
 604 #ifdef CONFIG_SYSCTL
 605         unregister_sysctl_table(ipq_sysctl_header);
 606 #endif
 607         unregister_netdevice_notifier(&ipq_dev_notifier);
 608         proc_net_remove(&init_net, IPQ_PROC_FS_NAME);
 609 cleanup_ipqnl: __maybe_unused
 610         netlink_kernel_release(ipqnl);
 611         mutex_lock(&ipqnl_mutex);
 612         mutex_unlock(&ipqnl_mutex);
 613
 614 cleanup_netlink_notifier:
 615         netlink_unregister_notifier(&ipq_nl_notifier);
 616         return status;
 617 }
 618
 619 static void __exit ip_queue_fini(void)
 620 {
 621         nf_unregister_queue_handlers(&nfqh);
 622
 623         ipq_flush(NULL, 0);
 624
 625 #ifdef CONFIG_SYSCTL
 626         unregister_sysctl_table(ipq_sysctl_header);
 627 #endif
 628         unregister_netdevice_notifier(&ipq_dev_notifier);
 629         proc_net_remove(&init_net, IPQ_PROC_FS_NAME);
 630
 631         netlink_kernel_release(ipqnl);
 632         mutex_lock(&ipqnl_mutex);
 633         mutex_unlock(&ipqnl_mutex);
 634
 635         netlink_unregister_notifier(&ipq_nl_notifier);
 636 }
 637
 638 MODULE_DESCRIPTION("IPv4 packet queue handler");
 639 MODULE_AUTHOR("James Morris <jmorris@intercode.com.au>");
 640 MODULE_LICENSE("GPL");
 641 MODULE_ALIAS_NET_PF_PROTO(PF_NETLINK, NETLINK_FIREWALL);
 642
 643 module_init(ip_queue_init);
 644 module_exit(ip_queue_fini);