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 void
  53 ipq_issue_verdict(struct nf_queue_entry *entry, int verdict)
  54 {
  55         /* TCP input path (and probably other bits) assume to be called
  56          * from softirq context, not from syscall, like ipq_issue_verdict is
  57          * called.  TCP input path deadlocks with locks taken from timer
  58          * softirq, e.g.  We therefore emulate this by local_bh_disable() */
  59
  60         local_bh_disable();
  61         nf_reinject(entry, verdict);
  62         local_bh_enable();
  63 }
  64
  65 static inline void
  66 __ipq_enqueue_entry(struct nf_queue_entry *entry)
  67 {
  68        list_add_tail(&entry->list, &queue_list);
  69        queue_total++;
  70 }
  71
  72 static inline int
  73 __ipq_set_mode(unsigned char mode, unsigned int range)
  74 {
  75         int status = 0;
  76
  77         switch(mode) {
  78         case IPQ_COPY_NONE:
  79         case IPQ_COPY_META:
  80                 copy_mode = mode;
  81                 copy_range = 0;
  82                 break;
  83
  84         case IPQ_COPY_PACKET:
  85                 copy_mode = mode;
  86                 copy_range = range;
  87                 if (copy_range > 0xFFFF)
  88                         copy_range = 0xFFFF;
  89                 break;
  90
  91         default:
  92                 status = -EINVAL;
  93
  94         }
  95         return status;
  96 }
  97
  98 static void __ipq_flush(ipq_cmpfn cmpfn, unsigned long data);
  99
 100 static inline void
 101 __ipq_reset(void)
 102 {
 103         peer_pid = 0;
 104         net_disable_timestamp();
 105         __ipq_set_mode(IPQ_COPY_NONE, 0);
 106         __ipq_flush(NULL, 0);
 107 }
 108
 109 static struct nf_queue_entry *
 110 ipq_find_dequeue_entry(unsigned long id)
 111 {
 112         struct nf_queue_entry *entry = NULL, *i;
 113
 114         write_lock_bh(&queue_lock);
 115
 116         list_for_each_entry(i, &queue_list, list) {
 117                 if ((unsigned long)i == id) {
 118                         entry = i;
 119                         break;
 120                 }
 121         }
 122
 123         if (entry) {
 124                 list_del(&entry->list);
 125                 queue_total--;
 126         }
 127
 128         write_unlock_bh(&queue_lock);
 129         return entry;
 130 }
 131
 132 static void
 133 __ipq_flush(ipq_cmpfn cmpfn, unsigned long data)
 134 {
 135         struct nf_queue_entry *entry, *next;
 136
 137         list_for_each_entry_safe(entry, next, &queue_list, list) {
 138                 if (!cmpfn || cmpfn(entry, data)) {
 139                         list_del(&entry->list);
 140                         queue_total--;
 141                         ipq_issue_verdict(entry, NF_DROP);
 142                 }
 143         }
 144 }
 145
 146 static void
 147 ipq_flush(ipq_cmpfn cmpfn, unsigned long data)
 148 {
 149         write_lock_bh(&queue_lock);
 150         __ipq_flush(cmpfn, data);
 151         write_unlock_bh(&queue_lock);
 152 }
 153
 154 static struct sk_buff *
 155 ipq_build_packet_message(struct nf_queue_entry *entry, int *errp)
 156 {
 157         sk_buff_data_t old_tail;
 158         size_t size = 0;
 159         size_t data_len = 0;
 160         struct sk_buff *skb;
 161         struct ipq_packet_msg *pmsg;
 162         struct nlmsghdr *nlh;
 163         struct timeval tv;
 164
 165         read_lock_bh(&queue_lock);
 166
 167         switch (copy_mode) {
 168         case IPQ_COPY_META:
 169         case IPQ_COPY_NONE:
 170                 size = NLMSG_SPACE(sizeof(*pmsg));
 171                 data_len = 0;
 172                 break;
 173
 174         case IPQ_COPY_PACKET:
 175                 if ((entry->skb->ip_summed == CHECKSUM_PARTIAL ||
 176                      entry->skb->ip_summed == CHECKSUM_COMPLETE) &&
 177                     (*errp = skb_checksum_help(entry->skb))) {
 178                         read_unlock_bh(&queue_lock);
 179                         return NULL;
 180                 }
 181                 if (copy_range == 0 || copy_range > entry->skb->len)
 182                         data_len = entry->skb->len;
 183                 else
 184                         data_len = copy_range;
 185
 186                 size = NLMSG_SPACE(sizeof(*pmsg) + data_len);
 187                 break;
 188
 189         default:
 190                 *errp = -EINVAL;
 191                 read_unlock_bh(&queue_lock);
 192                 return NULL;
 193         }
 194
 195         read_unlock_bh(&queue_lock);
 196
 197         skb = alloc_skb(size, GFP_ATOMIC);
 198         if (!skb)
 199                 goto nlmsg_failure;
 200
 201         old_tail = skb->tail;
 202         nlh = NLMSG_PUT(skb, 0, 0, IPQM_PACKET, size - sizeof(*nlh));
 203         pmsg = NLMSG_DATA(nlh);
 204         memset(pmsg, 0, sizeof(*pmsg));
 205
 206         pmsg->packet_id       = (unsigned long )entry;
 207         pmsg->data_len        = data_len;
 208         tv = ktime_to_timeval(entry->skb->tstamp);
 209         pmsg->timestamp_sec   = tv.tv_sec;
 210         pmsg->timestamp_usec  = tv.tv_usec;
 211         pmsg->mark            = entry->skb->mark;
 212         pmsg->hook            = entry->hook;
 213         pmsg->hw_protocol     = entry->skb->protocol;
 214
 215         if (entry->indev)
 216                 strcpy(pmsg->indev_name, entry->indev->name);
 217         else
 218                 pmsg->indev_name[0] = '\0';
 219
 220         if (entry->outdev)
 221                 strcpy(pmsg->outdev_name, entry->outdev->name);
 222         else
 223                 pmsg->outdev_name[0] = '\0';
 224
 225         if (entry->indev && entry->skb->dev) {
 226                 pmsg->hw_type = entry->skb->dev->type;
 227                 pmsg->hw_addrlen = dev_parse_header(entry->skb,
 228                                                     pmsg->hw_addr);
 229         }
 230
 231         if (data_len)
 232                 if (skb_copy_bits(entry->skb, 0, pmsg->payload, data_len))
 233                         BUG();
 234
 235         nlh->nlmsg_len = skb->tail - old_tail;
 236         return skb;
 237
 238 nlmsg_failure:
 239         if (skb)
 240                 kfree_skb(skb);
 241         *errp = -EINVAL;
 242         printk(KERN_ERR "ip_queue: error creating packet message\n");
 243         return NULL;
 244 }
 245
 246 static int
 247 ipq_enqueue_packet(struct nf_queue_entry *entry, unsigned int queuenum)
 248 {
 249         int status = -EINVAL;
 250         struct sk_buff *nskb;
 251
 252         if (copy_mode == IPQ_COPY_NONE)
 253                 return -EAGAIN;
 254
 255         nskb = ipq_build_packet_message(entry, &status);
 256         if (nskb == NULL)
 257                 return status;
 258
 259         write_lock_bh(&queue_lock);
 260
 261         if (!peer_pid)
 262                 goto err_out_free_nskb;
 263
 264         if (queue_total >= queue_maxlen) {
 265                 queue_dropped++;
 266                 status = -ENOSPC;
 267                 if (net_ratelimit())
 268                           printk (KERN_WARNING "ip_queue: full at %d entries, "
 269                                   "dropping packets(s). Dropped: %d\n", queue_total,
 270                                   queue_dropped);
 271                 goto err_out_free_nskb;
 272         }
 273
 274         /* netlink_unicast will either free the nskb or attach it to a socket */
 275         status = netlink_unicast(ipqnl, nskb, peer_pid, MSG_DONTWAIT);
 276         if (status < 0) {
 277                 queue_user_dropped++;
 278                 goto err_out_unlock;
 279         }
 280
 281         __ipq_enqueue_entry(entry);
 282
 283         write_unlock_bh(&queue_lock);
 284         return status;
 285
 286 err_out_free_nskb:
 287         kfree_skb(nskb);
 288
 289 err_out_unlock:
 290         write_unlock_bh(&queue_lock);
 291         return status;
 292 }
 293
 294 static int
 295 ipq_mangle_ipv4(ipq_verdict_msg_t *v, struct nf_queue_entry *e)
 296 {
 297         int diff;
 298         int err;
 299         struct iphdr *user_iph = (struct iphdr *)v->payload;
 300
 301         if (v->data_len < sizeof(*user_iph))
 302                 return 0;
 303         diff = v->data_len - e->skb->len;
 304         if (diff < 0) {
 305                 if (pskb_trim(e->skb, v->data_len))
 306                         return -ENOMEM;
 307         } else if (diff > 0) {
 308                 if (v->data_len > 0xFFFF)
 309                         return -EINVAL;
 310                 if (diff > skb_tailroom(e->skb)) {
 311                         err = pskb_expand_head(e->skb, 0,
 312                                                diff - skb_tailroom(e->skb),
 313                                                GFP_ATOMIC);
 314                         if (err) {
 315                                 printk(KERN_WARNING "ip_queue: error "
 316                                       "in mangle, dropping packet: %d\n", -err);
 317                                 return err;
 318                         }
 319                 }
 320                 skb_put(e->skb, diff);
 321         }
 322         if (!skb_make_writable(e->skb, v->data_len))
 323                 return -ENOMEM;
 324         skb_copy_to_linear_data(e->skb, v->payload, v->data_len);
 325         e->skb->ip_summed = CHECKSUM_NONE;
 326
 327         return 0;
 328 }
 329
 330 static int
 331 ipq_set_verdict(struct ipq_verdict_msg *vmsg, unsigned int len)
 332 {
 333         struct nf_queue_entry *entry;
 334
 335         if (vmsg->value > NF_MAX_VERDICT)
 336                 return -EINVAL;
 337
 338         entry = ipq_find_dequeue_entry(vmsg->id);
 339         if (entry == NULL)
 340                 return -ENOENT;
 341         else {
 342                 int verdict = vmsg->value;
 343
 344                 if (vmsg->data_len && vmsg->data_len == len)
 345                         if (ipq_mangle_ipv4(vmsg, entry) < 0)
 346                                 verdict = NF_DROP;
 347
 348                 ipq_issue_verdict(entry, verdict);
 349                 return 0;
 350         }
 351 }
 352
 353 static int
 354 ipq_set_mode(unsigned char mode, unsigned int range)
 355 {
 356         int status;
 357
 358         write_lock_bh(&queue_lock);
 359         status = __ipq_set_mode(mode, range);
 360         write_unlock_bh(&queue_lock);
 361         return status;
 362 }
 363
 364 static int
 365 ipq_receive_peer(struct ipq_peer_msg *pmsg,
 366                  unsigned char type, unsigned int len)
 367 {
 368         int status = 0;
 369
 370         if (len < sizeof(*pmsg))
 371                 return -EINVAL;
 372
 373         switch (type) {
 374         case IPQM_MODE:
 375                 status = ipq_set_mode(pmsg->msg.mode.value,
 376                                       pmsg->msg.mode.range);
 377                 break;
 378
 379         case IPQM_VERDICT:
 380                 if (pmsg->msg.verdict.value > NF_MAX_VERDICT)
 381                         status = -EINVAL;
 382                 else
 383                         status = ipq_set_verdict(&pmsg->msg.verdict,
 384                                                  len - sizeof(*pmsg));
 385                         break;
 386         default:
 387                 status = -EINVAL;
 388         }
 389         return status;
 390 }
 391
 392 static int
 393 dev_cmp(struct nf_queue_entry *entry, unsigned long ifindex)
 394 {
 395         if (entry->indev)
 396                 if (entry->indev->ifindex == ifindex)
 397                         return 1;
 398         if (entry->outdev)
 399                 if (entry->outdev->ifindex == ifindex)
 400                         return 1;
 401 #ifdef CONFIG_BRIDGE_NETFILTER
 402         if (entry->skb->nf_bridge) {
 403                 if (entry->skb->nf_bridge->physindev &&
 404                     entry->skb->nf_bridge->physindev->ifindex == ifindex)
 405                         return 1;
 406                 if (entry->skb->nf_bridge->physoutdev &&
 407                     entry->skb->nf_bridge->physoutdev->ifindex == ifindex)
 408                         return 1;
 409         }
 410 #endif
 411         return 0;
 412 }
 413
 414 static void
 415 ipq_dev_drop(int ifindex)
 416 {
 417         ipq_flush(dev_cmp, ifindex);
 418 }
 419
 420 #define RCV_SKB_FAIL(err) do { netlink_ack(skb, nlh, (err)); return; } while (0)
 421
 422 static inline void
 423 __ipq_rcv_skb(struct sk_buff *skb)
 424 {
 425         int status, type, pid, flags, nlmsglen, skblen;
 426         struct nlmsghdr *nlh;
 427
 428         skblen = skb->len;
 429         if (skblen < sizeof(*nlh))
 430                 return;
 431
 432         nlh = nlmsg_hdr(skb);
 433         nlmsglen = nlh->nlmsg_len;
 434         if (nlmsglen < sizeof(*nlh) || skblen < nlmsglen)
 435                 return;
 436
 437         pid = nlh->nlmsg_pid;
 438         flags = nlh->nlmsg_flags;
 439
 440         if(pid <= 0 || !(flags & NLM_F_REQUEST) || flags & NLM_F_MULTI)
 441                 RCV_SKB_FAIL(-EINVAL);
 442
 443         if (flags & MSG_TRUNC)
 444                 RCV_SKB_FAIL(-ECOMM);
 445
 446         type = nlh->nlmsg_type;
 447         if (type < NLMSG_NOOP || type >= IPQM_MAX)
 448                 RCV_SKB_FAIL(-EINVAL);
 449
 450         if (type <= IPQM_BASE)
 451                 return;
 452
 453         if (security_netlink_recv(skb, CAP_NET_ADMIN))
 454                 RCV_SKB_FAIL(-EPERM);
 455
 456         write_lock_bh(&queue_lock);
 457
 458         if (peer_pid) {
 459                 if (peer_pid != pid) {
 460                         write_unlock_bh(&queue_lock);
 461                         RCV_SKB_FAIL(-EBUSY);
 462                 }
 463         } else {
 464                 net_enable_timestamp();
 465                 peer_pid = pid;
 466         }
 467
 468         write_unlock_bh(&queue_lock);
 469
 470         status = ipq_receive_peer(NLMSG_DATA(nlh), type,
 471                                   nlmsglen - NLMSG_LENGTH(0));
 472         if (status < 0)
 473                 RCV_SKB_FAIL(status);
 474
 475         if (flags & NLM_F_ACK)
 476                 netlink_ack(skb, nlh, 0);
 477         return;
 478 }
 479
 480 static void
 481 ipq_rcv_skb(struct sk_buff *skb)
 482 {
 483         mutex_lock(&ipqnl_mutex);
 484         __ipq_rcv_skb(skb);
 485         mutex_unlock(&ipqnl_mutex);
 486 }
 487
 488 static int
 489 ipq_rcv_dev_event(struct notifier_block *this,
 490                   unsigned long event, void *ptr)
 491 {
 492         struct net_device *dev = ptr;
 493
 494         if (dev->nd_net != &init_net)
 495                 return NOTIFY_DONE;
 496
 497         /* Drop any packets associated with the downed device */
 498         if (event == NETDEV_DOWN)
 499                 ipq_dev_drop(dev->ifindex);
 500         return NOTIFY_DONE;
 501 }
 502
 503 static struct notifier_block ipq_dev_notifier = {
 504         .notifier_call  = ipq_rcv_dev_event,
 505 };
 506
 507 static int
 508 ipq_rcv_nl_event(struct notifier_block *this,
 509                  unsigned long event, void *ptr)
 510 {
 511         struct netlink_notify *n = ptr;
 512
 513         if (event == NETLINK_URELEASE &&
 514             n->protocol == NETLINK_FIREWALL && n->pid) {
 515                 write_lock_bh(&queue_lock);
 516                 if ((n->net == &init_net) && (n->pid == peer_pid))
 517                         __ipq_reset();
 518                 write_unlock_bh(&queue_lock);
 519         }
 520         return NOTIFY_DONE;
 521 }
 522
 523 static struct notifier_block ipq_nl_notifier = {
 524         .notifier_call  = ipq_rcv_nl_event,
 525 };
 526
 527 static struct ctl_table_header *ipq_sysctl_header;
 528
 529 static ctl_table ipq_table[] = {
 530         {
 531                 .ctl_name       = NET_IPQ_QMAX,
 532                 .procname       = NET_IPQ_QMAX_NAME,
 533                 .data           = &queue_maxlen,
 534                 .maxlen         = sizeof(queue_maxlen),
 535                 .mode           = 0644,
 536                 .proc_handler   = proc_dointvec
 537         },
 538         { .ctl_name = 0 }
 539 };
 540
 541 static ctl_table ipq_dir_table[] = {
 542         {
 543                 .ctl_name       = NET_IPV4,
 544                 .procname       = "ipv4",
 545                 .mode           = 0555,
 546                 .child          = ipq_table
 547         },
 548         { .ctl_name = 0 }
 549 };
 550
 551 static ctl_table ipq_root_table[] = {
 552         {
 553                 .ctl_name       = CTL_NET,
 554                 .procname       = "net",
 555                 .mode           = 0555,
 556                 .child          = ipq_dir_table
 557         },
 558         { .ctl_name = 0 }
 559 };
 560
 561 static int ip_queue_show(struct seq_file *m, void *v)
 562 {
 563         read_lock_bh(&queue_lock);
 564
 565         seq_printf(m,
 566                       "Peer PID          : %d\n"
 567                       "Copy mode         : %hu\n"
 568                       "Copy range        : %u\n"
 569                       "Queue length      : %u\n"
 570                       "Queue max. length : %u\n"
 571                       "Queue dropped     : %u\n"
 572                       "Netlink dropped   : %u\n",
 573                       peer_pid,
 574                       copy_mode,
 575                       copy_range,
 576                       queue_total,
 577                       queue_maxlen,
 578                       queue_dropped,
 579                       queue_user_dropped);
 580
 581         read_unlock_bh(&queue_lock);
 582         return 0;
 583 }
 584
 585 static int ip_queue_open(struct inode *inode, struct file *file)
 586 {
 587         return single_open(file, ip_queue_show, NULL);
 588 }
 589
 590 static const struct file_operations ip_queue_proc_fops = {
 591         .open           = ip_queue_open,
 592         .read           = seq_read,
 593         .llseek         = seq_lseek,
 594         .release        = single_release,
 595         .owner          = THIS_MODULE,
 596 };
 597
 598 static const struct nf_queue_handler nfqh = {
 599         .name   = "ip_queue",
 600         .outfn  = &ipq_enqueue_packet,
 601 };
 602
 603 static int __init ip_queue_init(void)
 604 {
 605         int status = -ENOMEM;
 606         struct proc_dir_entry *proc;
 607
 608         netlink_register_notifier(&ipq_nl_notifier);
 609         ipqnl = netlink_kernel_create(&init_net, NETLINK_FIREWALL, 0,
 610                                       ipq_rcv_skb, NULL, THIS_MODULE);
 611         if (ipqnl == NULL) {
 612                 printk(KERN_ERR "ip_queue: failed to create netlink socket\n");
 613                 goto cleanup_netlink_notifier;
 614         }
 615
 616         proc = create_proc_entry(IPQ_PROC_FS_NAME, 0, init_net.proc_net);
 617         if (proc) {
 618                 proc->owner = THIS_MODULE;
 619                 proc->proc_fops = &ip_queue_proc_fops;
 620         } else {
 621                 printk(KERN_ERR "ip_queue: failed to create proc entry\n");
 622                 goto cleanup_ipqnl;
 623         }
 624
 625         register_netdevice_notifier(&ipq_dev_notifier);
 626         ipq_sysctl_header = register_sysctl_table(ipq_root_table);
 627
 628         status = nf_register_queue_handler(PF_INET, &nfqh);
 629         if (status < 0) {
 630                 printk(KERN_ERR "ip_queue: failed to register queue handler\n");
 631                 goto cleanup_sysctl;
 632         }
 633         return status;
 634
 635 cleanup_sysctl:
 636         unregister_sysctl_table(ipq_sysctl_header);
 637         unregister_netdevice_notifier(&ipq_dev_notifier);
 638         proc_net_remove(&init_net, IPQ_PROC_FS_NAME);
 639 cleanup_ipqnl:
 640         sock_release(ipqnl->sk_socket);
 641         mutex_lock(&ipqnl_mutex);
 642         mutex_unlock(&ipqnl_mutex);
 643
 644 cleanup_netlink_notifier:
 645         netlink_unregister_notifier(&ipq_nl_notifier);
 646         return status;
 647 }
 648
 649 static void __exit ip_queue_fini(void)
 650 {
 651         nf_unregister_queue_handlers(&nfqh);
 652         synchronize_net();
 653         ipq_flush(NULL, 0);
 654
 655         unregister_sysctl_table(ipq_sysctl_header);
 656         unregister_netdevice_notifier(&ipq_dev_notifier);
 657         proc_net_remove(&init_net, IPQ_PROC_FS_NAME);
 658
 659         sock_release(ipqnl->sk_socket);
 660         mutex_lock(&ipqnl_mutex);
 661         mutex_unlock(&ipqnl_mutex);
 662
 663         netlink_unregister_notifier(&ipq_nl_notifier);
 664 }
 665
 666 MODULE_DESCRIPTION("IPv4 packet queue handler");
 667 MODULE_AUTHOR("James Morris <jmorris@intercode.com.au>");
 668 MODULE_LICENSE("GPL");
 669
 670 module_init(ip_queue_init);
 671 module_exit(ip_queue_fini);