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 #include <net/ip.h>
  33
  34 #define IPQ_QMAX_DEFAULT 1024
  35 #define IPQ_PROC_FS_NAME "ip_queue"
  36 #define NET_IPQ_QMAX 2088
  37 #define NET_IPQ_QMAX_NAME "ip_queue_maxlen"
  38
  39 typedef int (*ipq_cmpfn)(struct nf_queue_entry *, unsigned long);
  40
  41 static unsigned char copy_mode __read_mostly = IPQ_COPY_NONE;
  42 static unsigned int queue_maxlen __read_mostly = IPQ_QMAX_DEFAULT;
  43 static DEFINE_RWLOCK(queue_lock);
  44 static int peer_pid __read_mostly;
  45 static unsigned int copy_range __read_mostly;
  46 static unsigned int queue_total;
  47 static unsigned int queue_dropped = 0;
  48 static unsigned int queue_user_dropped = 0;
  49 static struct sock *ipqnl __read_mostly;
  50 static LIST_HEAD(queue_list);
  51 static DEFINE_MUTEX(ipqnl_mutex);
  52
  53 static inline void
  54 __ipq_enqueue_entry(struct nf_queue_entry *entry)
  55 {
  56        list_add_tail(&entry->list, &queue_list);
  57        queue_total++;
  58 }
  59
  60 static inline int
  61 __ipq_set_mode(unsigned char mode, unsigned int range)
  62 {
  63         int status = 0;
  64
  65         switch(mode) {
  66         case IPQ_COPY_NONE:
  67         case IPQ_COPY_META:
  68                 copy_mode = mode;
  69                 copy_range = 0;
  70                 break;
  71
  72         case IPQ_COPY_PACKET:
  73                 copy_mode = mode;
  74                 copy_range = range;
  75                 if (copy_range > 0xFFFF)
  76                         copy_range = 0xFFFF;
  77                 break;
  78
  79         default:
  80                 status = -EINVAL;
  81
  82         }
  83         return status;
  84 }
  85
  86 static void __ipq_flush(ipq_cmpfn cmpfn, unsigned long data);
  87
  88 static inline void
  89 __ipq_reset(void)
  90 {
  91         peer_pid = 0;
  92         net_disable_timestamp();
  93         __ipq_set_mode(IPQ_COPY_NONE, 0);
  94         __ipq_flush(NULL, 0);
  95 }
  96
  97 static struct nf_queue_entry *
  98 ipq_find_dequeue_entry(unsigned long id)
  99 {
 100         struct nf_queue_entry *entry = NULL, *i;
 101
 102         write_lock_bh(&queue_lock);
 103
 104         list_for_each_entry(i, &queue_list, list) {
 105                 if ((unsigned long)i == id) {
 106                         entry = i;
 107                         break;
 108                 }
 109         }
 110
 111         if (entry) {
 112                 list_del(&entry->list);
 113                 queue_total--;
 114         }
 115
 116         write_unlock_bh(&queue_lock);
 117         return entry;
 118 }
 119
 120 static void
 121 __ipq_flush(ipq_cmpfn cmpfn, unsigned long data)
 122 {
 123         struct nf_queue_entry *entry, *next;
 124
 125         list_for_each_entry_safe(entry, next, &queue_list, list) {
 126                 if (!cmpfn || cmpfn(entry, data)) {
 127                         list_del(&entry->list);
 128                         queue_total--;
 129                         nf_reinject(entry, NF_DROP);
 130                 }
 131         }
 132 }
 133
 134 static void
 135 ipq_flush(ipq_cmpfn cmpfn, unsigned long data)
 136 {
 137         write_lock_bh(&queue_lock);
 138         __ipq_flush(cmpfn, data);
 139         write_unlock_bh(&queue_lock);
 140 }
 141
 142 static struct sk_buff *
 143 ipq_build_packet_message(struct nf_queue_entry *entry, int *errp)
 144 {
 145         sk_buff_data_t old_tail;
 146         size_t size = 0;
 147         size_t data_len = 0;
 148         struct sk_buff *skb;
 149         struct ipq_packet_msg *pmsg;
 150         struct nlmsghdr *nlh;
 151         struct timeval tv;
 152
 153         read_lock_bh(&queue_lock);
 154
 155         switch (copy_mode) {
 156         case IPQ_COPY_META:
 157         case IPQ_COPY_NONE:
 158                 size = NLMSG_SPACE(sizeof(*pmsg));
 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         *errp = -EINVAL;
 227         printk(KERN_ERR "ip_queue: error creating packet message\n");
 228         return NULL;
 229 }
 230
 231 static int
 232 ipq_enqueue_packet(struct nf_queue_entry *entry, unsigned int queuenum)
 233 {
 234         int status = -EINVAL;
 235         struct sk_buff *nskb;
 236
 237         if (copy_mode == IPQ_COPY_NONE)
 238                 return -EAGAIN;
 239
 240         nskb = ipq_build_packet_message(entry, &status);
 241         if (nskb == NULL)
 242                 return status;
 243
 244         write_lock_bh(&queue_lock);
 245
 246         if (!peer_pid)
 247                 goto err_out_free_nskb;
 248
 249         if (queue_total >= queue_maxlen) {
 250                 queue_dropped++;
 251                 status = -ENOSPC;
 252                 if (net_ratelimit())
 253                           printk (KERN_WARNING "ip_queue: full at %d entries, "
 254                                   "dropping packets(s). Dropped: %d\n", queue_total,
 255                                   queue_dropped);
 256                 goto err_out_free_nskb;
 257         }
 258
 259         /* netlink_unicast will either free the nskb or attach it to a socket */
 260         status = netlink_unicast(ipqnl, nskb, peer_pid, MSG_DONTWAIT);
 261         if (status < 0) {
 262                 queue_user_dropped++;
 263                 goto err_out_unlock;
 264         }
 265
 266         __ipq_enqueue_entry(entry);
 267
 268         write_unlock_bh(&queue_lock);
 269         return status;
 270
 271 err_out_free_nskb:
 272         kfree_skb(nskb);
 273
 274 err_out_unlock:
 275         write_unlock_bh(&queue_lock);
 276         return status;
 277 }
 278
 279 static int
 280 ipq_mangle_ipv4(ipq_verdict_msg_t *v, struct nf_queue_entry *e)
 281 {
 282         int diff;
 283         struct iphdr *user_iph = (struct iphdr *)v->payload;
 284         struct sk_buff *nskb;
 285
 286         if (v->data_len < sizeof(*user_iph))
 287                 return 0;
 288         diff = v->data_len - e->skb->len;
 289         if (diff < 0) {
 290                 if (pskb_trim(e->skb, v->data_len))
 291                         return -ENOMEM;
 292         } else if (diff > 0) {
 293                 if (v->data_len > 0xFFFF)
 294                         return -EINVAL;
 295                 if (diff > skb_tailroom(e->skb)) {
 296                         nskb = skb_copy_expand(e->skb, skb_headroom(e->skb),
 297                                                diff, GFP_ATOMIC);
 298                         if (!nskb) {
 299                                 printk(KERN_WARNING "ip_queue: error "
 300                                       "in mangle, dropping packet\n");
 301                                 return -ENOMEM;
 302                         }
 303                         kfree_skb(e->skb);
 304                         e->skb = nskb;
 305                 }
 306                 skb_put(e->skb, diff);
 307         }
 308         if (!skb_make_writable(e->skb, v->data_len))
 309                 return -ENOMEM;
 310         skb_copy_to_linear_data(e->skb, v->payload, v->data_len);
 311         e->skb->ip_summed = CHECKSUM_NONE;
 312
 313         return 0;
 314 }
 315
 316 static int
 317 ipq_set_verdict(struct ipq_verdict_msg *vmsg, unsigned int len)
 318 {
 319         struct nf_queue_entry *entry;
 320
 321         if (vmsg->value > NF_MAX_VERDICT)
 322                 return -EINVAL;
 323
 324         entry = ipq_find_dequeue_entry(vmsg->id);
 325         if (entry == NULL)
 326                 return -ENOENT;
 327         else {
 328                 int verdict = vmsg->value;
 329
 330                 if (vmsg->data_len && vmsg->data_len == len)
 331                         if (ipq_mangle_ipv4(vmsg, entry) < 0)
 332                                 verdict = NF_DROP;
 333
 334                 nf_reinject(entry, verdict);
 335                 return 0;
 336         }
 337 }
 338
 339 static int
 340 ipq_set_mode(unsigned char mode, unsigned int range)
 341 {
 342         int status;
 343
 344         write_lock_bh(&queue_lock);
 345         status = __ipq_set_mode(mode, range);
 346         write_unlock_bh(&queue_lock);
 347         return status;
 348 }
 349
 350 static int
 351 ipq_receive_peer(struct ipq_peer_msg *pmsg,
 352                  unsigned char type, unsigned int len)
 353 {
 354         int status = 0;
 355
 356         if (len < sizeof(*pmsg))
 357                 return -EINVAL;
 358
 359         switch (type) {
 360         case IPQM_MODE:
 361                 status = ipq_set_mode(pmsg->msg.mode.value,
 362                                       pmsg->msg.mode.range);
 363                 break;
 364
 365         case IPQM_VERDICT:
 366                 if (pmsg->msg.verdict.value > NF_MAX_VERDICT)
 367                         status = -EINVAL;
 368                 else
 369                         status = ipq_set_verdict(&pmsg->msg.verdict,
 370                                                  len - sizeof(*pmsg));
 371                         break;
 372         default:
 373                 status = -EINVAL;
 374         }
 375         return status;
 376 }
 377
 378 static int
 379 dev_cmp(struct nf_queue_entry *entry, unsigned long ifindex)
 380 {
 381         if (entry->indev)
 382                 if (entry->indev->ifindex == ifindex)
 383                         return 1;
 384         if (entry->outdev)
 385                 if (entry->outdev->ifindex == ifindex)
 386                         return 1;
 387 #ifdef CONFIG_BRIDGE_NETFILTER
 388         if (entry->skb->nf_bridge) {
 389                 if (entry->skb->nf_bridge->physindev &&
 390                     entry->skb->nf_bridge->physindev->ifindex == ifindex)
 391                         return 1;
 392                 if (entry->skb->nf_bridge->physoutdev &&
 393                     entry->skb->nf_bridge->physoutdev->ifindex == ifindex)
 394                         return 1;
 395         }
 396 #endif
 397         return 0;
 398 }
 399
 400 static void
 401 ipq_dev_drop(int ifindex)
 402 {
 403         ipq_flush(dev_cmp, ifindex);
 404 }
 405
 406 #define RCV_SKB_FAIL(err) do { netlink_ack(skb, nlh, (err)); return; } while (0)
 407
 408 static inline void
 409 __ipq_rcv_skb(struct sk_buff *skb)
 410 {
 411         int status, type, pid, flags, nlmsglen, skblen;
 412         struct nlmsghdr *nlh;
 413
 414         skblen = skb->len;
 415         if (skblen < sizeof(*nlh))
 416                 return;
 417
 418         nlh = nlmsg_hdr(skb);
 419         nlmsglen = nlh->nlmsg_len;
 420         if (nlmsglen < sizeof(*nlh) || skblen < nlmsglen)
 421                 return;
 422
 423         pid = nlh->nlmsg_pid;
 424         flags = nlh->nlmsg_flags;
 425
 426         if(pid <= 0 || !(flags & NLM_F_REQUEST) || flags & NLM_F_MULTI)
 427                 RCV_SKB_FAIL(-EINVAL);
 428
 429         if (flags & MSG_TRUNC)
 430                 RCV_SKB_FAIL(-ECOMM);
 431
 432         type = nlh->nlmsg_type;
 433         if (type < NLMSG_NOOP || type >= IPQM_MAX)
 434                 RCV_SKB_FAIL(-EINVAL);
 435
 436         if (type <= IPQM_BASE)
 437                 return;
 438
 439         if (security_netlink_recv(skb, CAP_NET_ADMIN))
 440                 RCV_SKB_FAIL(-EPERM);
 441
 442         write_lock_bh(&queue_lock);
 443
 444         if (peer_pid) {
 445                 if (peer_pid != pid) {
 446                         write_unlock_bh(&queue_lock);
 447                         RCV_SKB_FAIL(-EBUSY);
 448                 }
 449         } else {
 450                 net_enable_timestamp();
 451                 peer_pid = pid;
 452         }
 453
 454         write_unlock_bh(&queue_lock);
 455
 456         status = ipq_receive_peer(NLMSG_DATA(nlh), type,
 457                                   nlmsglen - NLMSG_LENGTH(0));
 458         if (status < 0)
 459                 RCV_SKB_FAIL(status);
 460
 461         if (flags & NLM_F_ACK)
 462                 netlink_ack(skb, nlh, 0);
 463         return;
 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 &&
 500             n->protocol == NETLINK_FIREWALL && n->pid) {
 501                 write_lock_bh(&queue_lock);
 502                 if ((n->net == &init_net) && (n->pid == peer_pid))
 503                         __ipq_reset();
 504                 write_unlock_bh(&queue_lock);
 505         }
 506         return NOTIFY_DONE;
 507 }
 508
 509 static struct notifier_block ipq_nl_notifier = {
 510         .notifier_call  = ipq_rcv_nl_event,
 511 };
 512
 513 #ifdef CONFIG_SYSCTL
 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 #endif
 528
 529 #ifdef CONFIG_PROC_FS
 530 static int ip_queue_show(struct seq_file *m, void *v)
 531 {
 532         read_lock_bh(&queue_lock);
 533
 534         seq_printf(m,
 535                       "Peer PID          : %d\n"
 536                       "Copy mode         : %hu\n"
 537                       "Copy range        : %u\n"
 538                       "Queue length      : %u\n"
 539                       "Queue max. length : %u\n"
 540                       "Queue dropped     : %u\n"
 541                       "Netlink dropped   : %u\n",
 542                       peer_pid,
 543                       copy_mode,
 544                       copy_range,
 545                       queue_total,
 546                       queue_maxlen,
 547                       queue_dropped,
 548                       queue_user_dropped);
 549
 550         read_unlock_bh(&queue_lock);
 551         return 0;
 552 }
 553
 554 static int ip_queue_open(struct inode *inode, struct file *file)
 555 {
 556         return single_open(file, ip_queue_show, NULL);
 557 }
 558
 559 static const struct file_operations ip_queue_proc_fops = {
 560         .open           = ip_queue_open,
 561         .read           = seq_read,
 562         .llseek         = seq_lseek,
 563         .release        = single_release,
 564         .owner          = THIS_MODULE,
 565 };
 566 #endif
 567
 568 static const struct nf_queue_handler nfqh = {
 569         .name   = "ip_queue",
 570         .outfn  = &ipq_enqueue_packet,
 571 };
 572
 573 static int __init ip_queue_init(void)
 574 {
 575         int status = -ENOMEM;
 576         struct proc_dir_entry *proc __maybe_unused;
 577
 578         netlink_register_notifier(&ipq_nl_notifier);
 579         ipqnl = netlink_kernel_create(&init_net, NETLINK_FIREWALL, 0,
 580                                       ipq_rcv_skb, NULL, THIS_MODULE);
 581         if (ipqnl == NULL) {
 582                 printk(KERN_ERR "ip_queue: failed to create netlink socket\n");
 583                 goto cleanup_netlink_notifier;
 584         }
 585
 586 #ifdef CONFIG_PROC_FS
 587         proc = proc_create(IPQ_PROC_FS_NAME, 0, init_net.proc_net,
 588                            &ip_queue_proc_fops);
 589         if (!proc) {
 590                 printk(KERN_ERR "ip_queue: failed to create proc entry\n");
 591                 goto cleanup_ipqnl;
 592         }
 593 #endif
 594         register_netdevice_notifier(&ipq_dev_notifier);
 595 #ifdef CONFIG_SYSCTL
 596         ipq_sysctl_header = register_sysctl_paths(net_ipv4_ctl_path, ipq_table);
 597 #endif
 598         status = nf_register_queue_handler(PF_INET, &nfqh);
 599         if (status < 0) {
 600                 printk(KERN_ERR "ip_queue: failed to register queue handler\n");
 601                 goto cleanup_sysctl;
 602         }
 603         return status;
 604
 605 cleanup_sysctl:
 606 #ifdef CONFIG_SYSCTL
 607         unregister_sysctl_table(ipq_sysctl_header);
 608 #endif
 609         unregister_netdevice_notifier(&ipq_dev_notifier);
 610         proc_net_remove(&init_net, IPQ_PROC_FS_NAME);
 611 cleanup_ipqnl: __maybe_unused
 612         netlink_kernel_release(ipqnl);
 613         mutex_lock(&ipqnl_mutex);
 614         mutex_unlock(&ipqnl_mutex);
 615
 616 cleanup_netlink_notifier:
 617         netlink_unregister_notifier(&ipq_nl_notifier);
 618         return status;
 619 }
 620
 621 static void __exit ip_queue_fini(void)
 622 {
 623         nf_unregister_queue_handlers(&nfqh);
 624         synchronize_net();
 625         ipq_flush(NULL, 0);
 626
 627 #ifdef CONFIG_SYSCTL
 628         unregister_sysctl_table(ipq_sysctl_header);
 629 #endif
 630         unregister_netdevice_notifier(&ipq_dev_notifier);
 631         proc_net_remove(&init_net, IPQ_PROC_FS_NAME);
 632
 633         netlink_kernel_release(ipqnl);
 634         mutex_lock(&ipqnl_mutex);
 635         mutex_unlock(&ipqnl_mutex);
 636
 637         netlink_unregister_notifier(&ipq_nl_notifier);
 638 }
 639
 640 MODULE_DESCRIPTION("IPv4 packet queue handler");
 641 MODULE_AUTHOR("James Morris <jmorris@intercode.com.au>");
 642 MODULE_LICENSE("GPL");
 643
 644 module_init(ip_queue_init);
 645 module_exit(ip_queue_fini);