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/mutex.h>
  28 #include <net/net_namespace.h>
  29 #include <net/sock.h>
  30 #include <net/route.h>
  31 #include <net/netfilter/nf_queue.h>
  32
  33 #define IPQ_QMAX_DEFAULT 1024
  34 #define IPQ_PROC_FS_NAME "ip_queue"
  35 #define NET_IPQ_QMAX 2088
  36 #define NET_IPQ_QMAX_NAME "ip_queue_maxlen"
  37
  38 typedef int (*ipq_cmpfn)(struct nf_queue_entry *, unsigned long);
  39
  40 static unsigned char copy_mode __read_mostly = IPQ_COPY_NONE;
  41 static unsigned int queue_maxlen __read_mostly = IPQ_QMAX_DEFAULT;
  42 static DEFINE_RWLOCK(queue_lock);
  43 static int peer_pid __read_mostly;
  44 static unsigned int copy_range __read_mostly;
  45 static unsigned int queue_total;
  46 static unsigned int queue_dropped = 0;
  47 static unsigned int queue_user_dropped = 0;
  48 static struct sock *ipqnl __read_mostly;
  49 static LIST_HEAD(queue_list);
  50 static DEFINE_MUTEX(ipqnl_mutex);
  51
  52 static inline void
  53 __ipq_enqueue_entry(struct nf_queue_entry *entry)
  54 {
  55        list_add_tail(&entry->list, &queue_list);
  56        queue_total++;
  57 }
  58
  59 static inline int
  60 __ipq_set_mode(unsigned char mode, unsigned int range)
  61 {
  62         int status = 0;
  63
  64         switch(mode) {
  65         case IPQ_COPY_NONE:
  66         case IPQ_COPY_META:
  67                 copy_mode = mode;
  68                 copy_range = 0;
  69                 break;
  70
  71         case IPQ_COPY_PACKET:
  72                 copy_mode = mode;
  73                 copy_range = range;
  74                 if (copy_range > 0xFFFF)
  75                         copy_range = 0xFFFF;
  76                 break;
  77
  78         default:
  79                 status = -EINVAL;
  80
  81         }
  82         return status;
  83 }
  84
  85 static void __ipq_flush(ipq_cmpfn cmpfn, unsigned long data);
  86
  87 static inline void
  88 __ipq_reset(void)
  89 {
  90         peer_pid = 0;
  91         net_disable_timestamp();
  92         __ipq_set_mode(IPQ_COPY_NONE, 0);
  93         __ipq_flush(NULL, 0);
  94 }
  95
  96 static struct nf_queue_entry *
  97 ipq_find_dequeue_entry(unsigned long id)
  98 {
  99         struct nf_queue_entry *entry = NULL, *i;
 100
 101         write_lock_bh(&queue_lock);
 102
 103         list_for_each_entry(i, &queue_list, list) {
 104                 if ((unsigned long)i == id) {
 105                         entry = i;
 106                         break;
 107                 }
 108         }
 109
 110         if (entry) {
 111                 list_del(&entry->list);
 112                 queue_total--;
 113         }
 114
 115         write_unlock_bh(&queue_lock);
 116         return entry;
 117 }
 118
 119 static void
 120 __ipq_flush(ipq_cmpfn cmpfn, unsigned long data)
 121 {
 122         struct nf_queue_entry *entry, *next;
 123
 124         list_for_each_entry_safe(entry, next, &queue_list, list) {
 125                 if (!cmpfn || cmpfn(entry, data)) {
 126                         list_del(&entry->list);
 127                         queue_total--;
 128                         nf_reinject(entry, NF_DROP);
 129                 }
 130         }
 131 }
 132
 133 static void
 134 ipq_flush(ipq_cmpfn cmpfn, unsigned long data)
 135 {
 136         write_lock_bh(&queue_lock);
 137         __ipq_flush(cmpfn, data);
 138         write_unlock_bh(&queue_lock);
 139 }
 140
 141 static struct sk_buff *
 142 ipq_build_packet_message(struct nf_queue_entry *entry, int *errp)
 143 {
 144         sk_buff_data_t old_tail;
 145         size_t size = 0;
 146         size_t data_len = 0;
 147         struct sk_buff *skb;
 148         struct ipq_packet_msg *pmsg;
 149         struct nlmsghdr *nlh;
 150         struct timeval tv;
 151
 152         read_lock_bh(&queue_lock);
 153
 154         switch (copy_mode) {
 155         case IPQ_COPY_META:
 156         case IPQ_COPY_NONE:
 157                 size = NLMSG_SPACE(sizeof(*pmsg));
 158                 data_len = 0;
 159                 break;
 160
 161         case IPQ_COPY_PACKET:
 162                 if ((entry->skb->ip_summed == CHECKSUM_PARTIAL ||
 163                      entry->skb->ip_summed == CHECKSUM_COMPLETE) &&
 164                     (*errp = skb_checksum_help(entry->skb))) {
 165                         read_unlock_bh(&queue_lock);
 166                         return NULL;
 167                 }
 168                 if (copy_range == 0 || copy_range > entry->skb->len)
 169                         data_len = entry->skb->len;
 170                 else
 171                         data_len = copy_range;
 172
 173                 size = NLMSG_SPACE(sizeof(*pmsg) + data_len);
 174                 break;
 175
 176         default:
 177                 *errp = -EINVAL;
 178                 read_unlock_bh(&queue_lock);
 179                 return NULL;
 180         }
 181
 182         read_unlock_bh(&queue_lock);
 183
 184         skb = alloc_skb(size, GFP_ATOMIC);
 185         if (!skb)
 186                 goto nlmsg_failure;
 187
 188         old_tail = skb->tail;
 189         nlh = NLMSG_PUT(skb, 0, 0, IPQM_PACKET, size - sizeof(*nlh));
 190         pmsg = NLMSG_DATA(nlh);
 191         memset(pmsg, 0, sizeof(*pmsg));
 192
 193         pmsg->packet_id       = (unsigned long )entry;
 194         pmsg->data_len        = data_len;
 195         tv = ktime_to_timeval(entry->skb->tstamp);
 196         pmsg->timestamp_sec   = tv.tv_sec;
 197         pmsg->timestamp_usec  = tv.tv_usec;
 198         pmsg->mark            = entry->skb->mark;
 199         pmsg->hook            = entry->hook;
 200         pmsg->hw_protocol     = entry->skb->protocol;
 201
 202         if (entry->indev)
 203                 strcpy(pmsg->indev_name, entry->indev->name);
 204         else
 205                 pmsg->indev_name[0] = '\0';
 206
 207         if (entry->outdev)
 208                 strcpy(pmsg->outdev_name, entry->outdev->name);
 209         else
 210                 pmsg->outdev_name[0] = '\0';
 211
 212         if (entry->indev && entry->skb->dev) {
 213                 pmsg->hw_type = entry->skb->dev->type;
 214                 pmsg->hw_addrlen = dev_parse_header(entry->skb,
 215                                                     pmsg->hw_addr);
 216         }
 217
 218         if (data_len)
 219                 if (skb_copy_bits(entry->skb, 0, pmsg->payload, data_len))
 220                         BUG();
 221
 222         nlh->nlmsg_len = skb->tail - old_tail;
 223         return skb;
 224
 225 nlmsg_failure:
 226         if (skb)
 227                 kfree_skb(skb);
 228         *errp = -EINVAL;
 229         printk(KERN_ERR "ip_queue: error creating packet message\n");
 230         return NULL;
 231 }
 232
 233 static int
 234 ipq_enqueue_packet(struct nf_queue_entry *entry, unsigned int queuenum)
 235 {
 236         int status = -EINVAL;
 237         struct sk_buff *nskb;
 238
 239         if (copy_mode == IPQ_COPY_NONE)
 240                 return -EAGAIN;
 241
 242         nskb = ipq_build_packet_message(entry, &status);
 243         if (nskb == NULL)
 244                 return status;
 245
 246         write_lock_bh(&queue_lock);
 247
 248         if (!peer_pid)
 249                 goto err_out_free_nskb;
 250
 251         if (queue_total >= queue_maxlen) {
 252                 queue_dropped++;
 253                 status = -ENOSPC;
 254                 if (net_ratelimit())
 255                           printk (KERN_WARNING "ip_queue: full at %d entries, "
 256                                   "dropping packets(s). Dropped: %d\n", queue_total,
 257                                   queue_dropped);
 258                 goto err_out_free_nskb;
 259         }
 260
 261         /* netlink_unicast will either free the nskb or attach it to a socket */
 262         status = netlink_unicast(ipqnl, nskb, peer_pid, MSG_DONTWAIT);
 263         if (status < 0) {
 264                 queue_user_dropped++;
 265                 goto err_out_unlock;
 266         }
 267
 268         __ipq_enqueue_entry(entry);
 269
 270         write_unlock_bh(&queue_lock);
 271         return status;
 272
 273 err_out_free_nskb:
 274         kfree_skb(nskb);
 275
 276 err_out_unlock:
 277         write_unlock_bh(&queue_lock);
 278         return status;
 279 }
 280
 281 static int
 282 ipq_mangle_ipv4(ipq_verdict_msg_t *v, struct nf_queue_entry *e)
 283 {
 284         int diff;
 285         int err;
 286         struct iphdr *user_iph = (struct iphdr *)v->payload;
 287
 288         if (v->data_len < sizeof(*user_iph))
 289                 return 0;
 290         diff = v->data_len - e->skb->len;
 291         if (diff < 0) {
 292                 if (pskb_trim(e->skb, v->data_len))
 293                         return -ENOMEM;
 294         } else if (diff > 0) {
 295                 if (v->data_len > 0xFFFF)
 296                         return -EINVAL;
 297                 if (diff > skb_tailroom(e->skb)) {
 298                         err = pskb_expand_head(e->skb, 0,
 299                                                diff - skb_tailroom(e->skb),
 300                                                GFP_ATOMIC);
 301                         if (err) {
 302                                 printk(KERN_WARNING "ip_queue: error "
 303                                       "in mangle, dropping packet: %d\n", -err);
 304                                 return err;
 305                         }
 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         return;
 465 }
 466
 467 static void
 468 ipq_rcv_skb(struct sk_buff *skb)
 469 {
 470         mutex_lock(&ipqnl_mutex);
 471         __ipq_rcv_skb(skb);
 472         mutex_unlock(&ipqnl_mutex);
 473 }
 474
 475 static int
 476 ipq_rcv_dev_event(struct notifier_block *this,
 477                   unsigned long event, void *ptr)
 478 {
 479         struct net_device *dev = ptr;
 480
 481         if (dev->nd_net != &init_net)
 482                 return NOTIFY_DONE;
 483
 484         /* Drop any packets associated with the downed device */
 485         if (event == NETDEV_DOWN)
 486                 ipq_dev_drop(dev->ifindex);
 487         return NOTIFY_DONE;
 488 }
 489
 490 static struct notifier_block ipq_dev_notifier = {
 491         .notifier_call  = ipq_rcv_dev_event,
 492 };
 493
 494 static int
 495 ipq_rcv_nl_event(struct notifier_block *this,
 496                  unsigned long event, void *ptr)
 497 {
 498         struct netlink_notify *n = ptr;
 499
 500         if (event == NETLINK_URELEASE &&
 501             n->protocol == NETLINK_FIREWALL && n->pid) {
 502                 write_lock_bh(&queue_lock);
 503                 if ((n->net == &init_net) && (n->pid == peer_pid))
 504                         __ipq_reset();
 505                 write_unlock_bh(&queue_lock);
 506         }
 507         return NOTIFY_DONE;
 508 }
 509
 510 static struct notifier_block ipq_nl_notifier = {
 511         .notifier_call  = ipq_rcv_nl_event,
 512 };
 513
 514 static struct ctl_table_header *ipq_sysctl_header;
 515
 516 static ctl_table ipq_table[] = {
 517         {
 518                 .ctl_name       = NET_IPQ_QMAX,
 519                 .procname       = NET_IPQ_QMAX_NAME,
 520                 .data           = &queue_maxlen,
 521                 .maxlen         = sizeof(queue_maxlen),
 522                 .mode           = 0644,
 523                 .proc_handler   = proc_dointvec
 524         },
 525         { .ctl_name = 0 }
 526 };
 527
 528 static ctl_table ipq_dir_table[] = {
 529         {
 530                 .ctl_name       = NET_IPV4,
 531                 .procname       = "ipv4",
 532                 .mode           = 0555,
 533                 .child          = ipq_table
 534         },
 535         { .ctl_name = 0 }
 536 };
 537
 538 static ctl_table ipq_root_table[] = {
 539         {
 540                 .ctl_name       = CTL_NET,
 541                 .procname       = "net",
 542                 .mode           = 0555,
 543                 .child          = ipq_dir_table
 544         },
 545         { .ctl_name = 0 }
 546 };
 547
 548 static int ip_queue_show(struct seq_file *m, void *v)
 549 {
 550         read_lock_bh(&queue_lock);
 551
 552         seq_printf(m,
 553                       "Peer PID          : %d\n"
 554                       "Copy mode         : %hu\n"
 555                       "Copy range        : %u\n"
 556                       "Queue length      : %u\n"
 557                       "Queue max. length : %u\n"
 558                       "Queue dropped     : %u\n"
 559                       "Netlink dropped   : %u\n",
 560                       peer_pid,
 561                       copy_mode,
 562                       copy_range,
 563                       queue_total,
 564                       queue_maxlen,
 565                       queue_dropped,
 566                       queue_user_dropped);
 567
 568         read_unlock_bh(&queue_lock);
 569         return 0;
 570 }
 571
 572 static int ip_queue_open(struct inode *inode, struct file *file)
 573 {
 574         return single_open(file, ip_queue_show, NULL);
 575 }
 576
 577 static const struct file_operations ip_queue_proc_fops = {
 578         .open           = ip_queue_open,
 579         .read           = seq_read,
 580         .llseek         = seq_lseek,
 581         .release        = single_release,
 582         .owner          = THIS_MODULE,
 583 };
 584
 585 static const struct nf_queue_handler nfqh = {
 586         .name   = "ip_queue",
 587         .outfn  = &ipq_enqueue_packet,
 588 };
 589
 590 static int __init ip_queue_init(void)
 591 {
 592         int status = -ENOMEM;
 593         struct proc_dir_entry *proc;
 594
 595         netlink_register_notifier(&ipq_nl_notifier);
 596         ipqnl = netlink_kernel_create(&init_net, NETLINK_FIREWALL, 0,
 597                                       ipq_rcv_skb, NULL, THIS_MODULE);
 598         if (ipqnl == NULL) {
 599                 printk(KERN_ERR "ip_queue: failed to create netlink socket\n");
 600                 goto cleanup_netlink_notifier;
 601         }
 602
 603         proc = create_proc_entry(IPQ_PROC_FS_NAME, 0, init_net.proc_net);
 604         if (proc) {
 605                 proc->owner = THIS_MODULE;
 606                 proc->proc_fops = &ip_queue_proc_fops;
 607         } else {
 608                 printk(KERN_ERR "ip_queue: failed to create proc entry\n");
 609                 goto cleanup_ipqnl;
 610         }
 611
 612         register_netdevice_notifier(&ipq_dev_notifier);
 613         ipq_sysctl_header = register_sysctl_table(ipq_root_table);
 614
 615         status = nf_register_queue_handler(PF_INET, &nfqh);
 616         if (status < 0) {
 617                 printk(KERN_ERR "ip_queue: failed to register queue handler\n");
 618                 goto cleanup_sysctl;
 619         }
 620         return status;
 621
 622 cleanup_sysctl:
 623         unregister_sysctl_table(ipq_sysctl_header);
 624         unregister_netdevice_notifier(&ipq_dev_notifier);
 625         proc_net_remove(&init_net, IPQ_PROC_FS_NAME);
 626 cleanup_ipqnl:
 627         sock_release(ipqnl->sk_socket);
 628         mutex_lock(&ipqnl_mutex);
 629         mutex_unlock(&ipqnl_mutex);
 630
 631 cleanup_netlink_notifier:
 632         netlink_unregister_notifier(&ipq_nl_notifier);
 633         return status;
 634 }
 635
 636 static void __exit ip_queue_fini(void)
 637 {
 638         nf_unregister_queue_handlers(&nfqh);
 639         synchronize_net();
 640         ipq_flush(NULL, 0);
 641
 642         unregister_sysctl_table(ipq_sysctl_header);
 643         unregister_netdevice_notifier(&ipq_dev_notifier);
 644         proc_net_remove(&init_net, IPQ_PROC_FS_NAME);
 645
 646         sock_release(ipqnl->sk_socket);
 647         mutex_lock(&ipqnl_mutex);
 648         mutex_unlock(&ipqnl_mutex);
 649
 650         netlink_unregister_notifier(&ipq_nl_notifier);
 651 }
 652
 653 MODULE_DESCRIPTION("IPv4 packet queue handler");
 654 MODULE_AUTHOR("James Morris <jmorris@intercode.com.au>");
 655 MODULE_LICENSE("GPL");
 656
 657 module_init(ip_queue_init);
 658 module_exit(ip_queue_fini);