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_SPINLOCK(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                 if (range > 0xFFFF)
  76                         range = 0xFFFF;
  77                 copy_range = range;
  78                 copy_mode = mode;
  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         spin_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         spin_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         spin_lock_bh(&queue_lock);
 140         __ipq_flush(cmpfn, data);
 141         spin_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         switch (ACCESS_ONCE(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                     (*errp = skb_checksum_help(entry->skb)))
 164                         return NULL;
 165
 166                 data_len = ACCESS_ONCE(copy_range);
 167                 if (data_len == 0 || data_len > entry->skb->len)
 168                         data_len = entry->skb->len;
 169
 170                 size = NLMSG_SPACE(sizeof(*pmsg) + data_len);
 171                 break;
 172
 173         default:
 174                 *errp = -EINVAL;
 175                 return NULL;
 176         }
 177
 178         skb = alloc_skb(size, GFP_ATOMIC);
 179         if (!skb)
 180                 goto nlmsg_failure;
 181
 182         old_tail = skb->tail;
 183         nlh = NLMSG_PUT(skb, 0, 0, IPQM_PACKET, size - sizeof(*nlh));
 184         pmsg = NLMSG_DATA(nlh);
 185         memset(pmsg, 0, sizeof(*pmsg));
 186
 187         pmsg->packet_id       = (unsigned long )entry;
 188         pmsg->data_len        = data_len;
 189         tv = ktime_to_timeval(entry->skb->tstamp);
 190         pmsg->timestamp_sec   = tv.tv_sec;
 191         pmsg->timestamp_usec  = tv.tv_usec;
 192         pmsg->mark            = entry->skb->mark;
 193         pmsg->hook            = entry->hook;
 194         pmsg->hw_protocol     = entry->skb->protocol;
 195
 196         if (entry->indev)
 197                 strcpy(pmsg->indev_name, entry->indev->name);
 198         else
 199                 pmsg->indev_name[0] = '\0';
 200
 201         if (entry->outdev)
 202                 strcpy(pmsg->outdev_name, entry->outdev->name);
 203         else
 204                 pmsg->outdev_name[0] = '\0';
 205
 206         if (entry->indev && entry->skb->dev) {
 207                 pmsg->hw_type = entry->skb->dev->type;
 208                 pmsg->hw_addrlen = dev_parse_header(entry->skb,
 209                                                     pmsg->hw_addr);
 210         }
 211
 212         if (data_len)
 213                 if (skb_copy_bits(entry->skb, 0, pmsg->payload, data_len))
 214                         BUG();
 215
 216         nlh->nlmsg_len = skb->tail - old_tail;
 217         return skb;
 218
 219 nlmsg_failure:
 220         *errp = -EINVAL;
 221         printk(KERN_ERR "ip_queue: error creating packet message\n");
 222         return NULL;
 223 }
 224
 225 static int
 226 ipq_enqueue_packet(struct nf_queue_entry *entry, unsigned int queuenum)
 227 {
 228         int status = -EINVAL;
 229         struct sk_buff *nskb;
 230
 231         if (copy_mode == IPQ_COPY_NONE)
 232                 return -EAGAIN;
 233
 234         nskb = ipq_build_packet_message(entry, &status);
 235         if (nskb == NULL)
 236                 return status;
 237
 238         spin_lock_bh(&queue_lock);
 239
 240         if (!peer_pid)
 241                 goto err_out_free_nskb;
 242
 243         if (queue_total >= queue_maxlen) {
 244                 queue_dropped++;
 245                 status = -ENOSPC;
 246                 if (net_ratelimit())
 247                           printk (KERN_WARNING "ip_queue: full at %d entries, "
 248                                   "dropping packets(s). Dropped: %d\n", queue_total,
 249                                   queue_dropped);
 250                 goto err_out_free_nskb;
 251         }
 252
 253         /* netlink_unicast will either free the nskb or attach it to a socket */
 254         status = netlink_unicast(ipqnl, nskb, peer_pid, MSG_DONTWAIT);
 255         if (status < 0) {
 256                 queue_user_dropped++;
 257                 goto err_out_unlock;
 258         }
 259
 260         __ipq_enqueue_entry(entry);
 261
 262         spin_unlock_bh(&queue_lock);
 263         return status;
 264
 265 err_out_free_nskb:
 266         kfree_skb(nskb);
 267
 268 err_out_unlock:
 269         spin_unlock_bh(&queue_lock);
 270         return status;
 271 }
 272
 273 static int
 274 ipq_mangle_ipv4(ipq_verdict_msg_t *v, struct nf_queue_entry *e)
 275 {
 276         int diff;
 277         struct iphdr *user_iph = (struct iphdr *)v->payload;
 278         struct sk_buff *nskb;
 279
 280         if (v->data_len < sizeof(*user_iph))
 281                 return 0;
 282         diff = v->data_len - e->skb->len;
 283         if (diff < 0) {
 284                 if (pskb_trim(e->skb, v->data_len))
 285                         return -ENOMEM;
 286         } else if (diff > 0) {
 287                 if (v->data_len > 0xFFFF)
 288                         return -EINVAL;
 289                 if (diff > skb_tailroom(e->skb)) {
 290                         nskb = skb_copy_expand(e->skb, skb_headroom(e->skb),
 291                                                diff, GFP_ATOMIC);
 292                         if (!nskb) {
 293                                 printk(KERN_WARNING "ip_queue: error "
 294                                       "in mangle, dropping packet\n");
 295                                 return -ENOMEM;
 296                         }
 297                         kfree_skb(e->skb);
 298                         e->skb = nskb;
 299                 }
 300                 skb_put(e->skb, diff);
 301         }
 302         if (!skb_make_writable(e->skb, v->data_len))
 303                 return -ENOMEM;
 304         skb_copy_to_linear_data(e->skb, v->payload, v->data_len);
 305         e->skb->ip_summed = CHECKSUM_NONE;
 306
 307         return 0;
 308 }
 309
 310 static int
 311 ipq_set_verdict(struct ipq_verdict_msg *vmsg, unsigned int len)
 312 {
 313         struct nf_queue_entry *entry;
 314
 315         if (vmsg->value > NF_MAX_VERDICT)
 316                 return -EINVAL;
 317
 318         entry = ipq_find_dequeue_entry(vmsg->id);
 319         if (entry == NULL)
 320                 return -ENOENT;
 321         else {
 322                 int verdict = vmsg->value;
 323
 324                 if (vmsg->data_len && vmsg->data_len == len)
 325                         if (ipq_mangle_ipv4(vmsg, entry) < 0)
 326                                 verdict = NF_DROP;
 327
 328                 nf_reinject(entry, verdict);
 329                 return 0;
 330         }
 331 }
 332
 333 static int
 334 ipq_set_mode(unsigned char mode, unsigned int range)
 335 {
 336         int status;
 337
 338         spin_lock_bh(&queue_lock);
 339         status = __ipq_set_mode(mode, range);
 340         spin_unlock_bh(&queue_lock);
 341         return status;
 342 }
 343
 344 static int
 345 ipq_receive_peer(struct ipq_peer_msg *pmsg,
 346                  unsigned char type, unsigned int len)
 347 {
 348         int status = 0;
 349
 350         if (len < sizeof(*pmsg))
 351                 return -EINVAL;
 352
 353         switch (type) {
 354         case IPQM_MODE:
 355                 status = ipq_set_mode(pmsg->msg.mode.value,
 356                                       pmsg->msg.mode.range);
 357                 break;
 358
 359         case IPQM_VERDICT:
 360                 if (pmsg->msg.verdict.value > NF_MAX_VERDICT)
 361                         status = -EINVAL;
 362                 else
 363                         status = ipq_set_verdict(&pmsg->msg.verdict,
 364                                                  len - sizeof(*pmsg));
 365                         break;
 366         default:
 367                 status = -EINVAL;
 368         }
 369         return status;
 370 }
 371
 372 static int
 373 dev_cmp(struct nf_queue_entry *entry, unsigned long ifindex)
 374 {
 375         if (entry->indev)
 376                 if (entry->indev->ifindex == ifindex)
 377                         return 1;
 378         if (entry->outdev)
 379                 if (entry->outdev->ifindex == ifindex)
 380                         return 1;
 381 #ifdef CONFIG_BRIDGE_NETFILTER
 382         if (entry->skb->nf_bridge) {
 383                 if (entry->skb->nf_bridge->physindev &&
 384                     entry->skb->nf_bridge->physindev->ifindex == ifindex)
 385                         return 1;
 386                 if (entry->skb->nf_bridge->physoutdev &&
 387                     entry->skb->nf_bridge->physoutdev->ifindex == ifindex)
 388                         return 1;
 389         }
 390 #endif
 391         return 0;
 392 }
 393
 394 static void
 395 ipq_dev_drop(int ifindex)
 396 {
 397         ipq_flush(dev_cmp, ifindex);
 398 }
 399
 400 #define RCV_SKB_FAIL(err) do { netlink_ack(skb, nlh, (err)); return; } while (0)
 401
 402 static inline void
 403 __ipq_rcv_skb(struct sk_buff *skb)
 404 {
 405         int status, type, pid, flags;
 406         unsigned int nlmsglen, skblen;
 407         struct nlmsghdr *nlh;
 408
 409         skblen = skb->len;
 410         if (skblen < sizeof(*nlh))
 411                 return;
 412
 413         nlh = nlmsg_hdr(skb);
 414         nlmsglen = nlh->nlmsg_len;
 415         if (nlmsglen < sizeof(*nlh) || skblen < nlmsglen)
 416                 return;
 417
 418         pid = nlh->nlmsg_pid;
 419         flags = nlh->nlmsg_flags;
 420
 421         if(pid <= 0 || !(flags & NLM_F_REQUEST) || flags & NLM_F_MULTI)
 422                 RCV_SKB_FAIL(-EINVAL);
 423
 424         if (flags & MSG_TRUNC)
 425                 RCV_SKB_FAIL(-ECOMM);
 426
 427         type = nlh->nlmsg_type;
 428         if (type < NLMSG_NOOP || type >= IPQM_MAX)
 429                 RCV_SKB_FAIL(-EINVAL);
 430
 431         if (type <= IPQM_BASE)
 432                 return;
 433
 434         if (security_netlink_recv(skb, CAP_NET_ADMIN))
 435                 RCV_SKB_FAIL(-EPERM);
 436
 437         spin_lock_bh(&queue_lock);
 438
 439         if (peer_pid) {
 440                 if (peer_pid != pid) {
 441                         spin_unlock_bh(&queue_lock);
 442                         RCV_SKB_FAIL(-EBUSY);
 443                 }
 444         } else {
 445                 net_enable_timestamp();
 446                 peer_pid = pid;
 447         }
 448
 449         spin_unlock_bh(&queue_lock);
 450
 451         status = ipq_receive_peer(NLMSG_DATA(nlh), type,
 452                                   nlmsglen - NLMSG_LENGTH(0));
 453         if (status < 0)
 454                 RCV_SKB_FAIL(status);
 455
 456         if (flags & NLM_F_ACK)
 457                 netlink_ack(skb, nlh, 0);
 458 }
 459
 460 static void
 461 ipq_rcv_skb(struct sk_buff *skb)
 462 {
 463         mutex_lock(&ipqnl_mutex);
 464         __ipq_rcv_skb(skb);
 465         mutex_unlock(&ipqnl_mutex);
 466 }
 467
 468 static int
 469 ipq_rcv_dev_event(struct notifier_block *this,
 470                   unsigned long event, void *ptr)
 471 {
 472         struct net_device *dev = ptr;
 473
 474         if (!net_eq(dev_net(dev), &init_net))
 475                 return NOTIFY_DONE;
 476
 477         /* Drop any packets associated with the downed device */
 478         if (event == NETDEV_DOWN)
 479                 ipq_dev_drop(dev->ifindex);
 480         return NOTIFY_DONE;
 481 }
 482
 483 static struct notifier_block ipq_dev_notifier = {
 484         .notifier_call  = ipq_rcv_dev_event,
 485 };
 486
 487 static int
 488 ipq_rcv_nl_event(struct notifier_block *this,
 489                  unsigned long event, void *ptr)
 490 {
 491         struct netlink_notify *n = ptr;
 492
 493         if (event == NETLINK_URELEASE && n->protocol == NETLINK_FIREWALL) {
 494                 spin_lock_bh(&queue_lock);
 495                 if ((net_eq(n->net, &init_net)) && (n->pid == peer_pid))
 496                         __ipq_reset();
 497                 spin_unlock_bh(&queue_lock);
 498         }
 499         return NOTIFY_DONE;
 500 }
 501
 502 static struct notifier_block ipq_nl_notifier = {
 503         .notifier_call  = ipq_rcv_nl_event,
 504 };
 505
 506 #ifdef CONFIG_SYSCTL
 507 static struct ctl_table_header *ipq_sysctl_header;
 508
 509 static ctl_table ipq_table[] = {
 510         {
 511                 .procname       = NET_IPQ_QMAX_NAME,
 512                 .data           = &queue_maxlen,
 513                 .maxlen         = sizeof(queue_maxlen),
 514                 .mode           = 0644,
 515                 .proc_handler   = proc_dointvec
 516         },
 517         { }
 518 };
 519 #endif
 520
 521 #ifdef CONFIG_PROC_FS
 522 static int ip_queue_show(struct seq_file *m, void *v)
 523 {
 524         spin_lock_bh(&queue_lock);
 525
 526         seq_printf(m,
 527                       "Peer PID          : %d\n"
 528                       "Copy mode         : %hu\n"
 529                       "Copy range        : %u\n"
 530                       "Queue length      : %u\n"
 531                       "Queue max. length : %u\n"
 532                       "Queue dropped     : %u\n"
 533                       "Netlink dropped   : %u\n",
 534                       peer_pid,
 535                       copy_mode,
 536                       copy_range,
 537                       queue_total,
 538                       queue_maxlen,
 539                       queue_dropped,
 540                       queue_user_dropped);
 541
 542         spin_unlock_bh(&queue_lock);
 543         return 0;
 544 }
 545
 546 static int ip_queue_open(struct inode *inode, struct file *file)
 547 {
 548         return single_open(file, ip_queue_show, NULL);
 549 }
 550
 551 static const struct file_operations ip_queue_proc_fops = {
 552         .open           = ip_queue_open,
 553         .read           = seq_read,
 554         .llseek         = seq_lseek,
 555         .release        = single_release,
 556         .owner          = THIS_MODULE,
 557 };
 558 #endif
 559
 560 static const struct nf_queue_handler nfqh = {
 561         .name   = "ip_queue",
 562         .outfn  = &ipq_enqueue_packet,
 563 };
 564
 565 static int __init ip_queue_init(void)
 566 {
 567         int status = -ENOMEM;
 568         struct proc_dir_entry *proc __maybe_unused;
 569
 570         netlink_register_notifier(&ipq_nl_notifier);
 571         ipqnl = netlink_kernel_create(&init_net, NETLINK_FIREWALL, 0,
 572                                       ipq_rcv_skb, NULL, THIS_MODULE);
 573         if (ipqnl == NULL) {
 574                 printk(KERN_ERR "ip_queue: failed to create netlink socket\n");
 575                 goto cleanup_netlink_notifier;
 576         }
 577
 578 #ifdef CONFIG_PROC_FS
 579         proc = proc_create(IPQ_PROC_FS_NAME, 0, init_net.proc_net,
 580                            &ip_queue_proc_fops);
 581         if (!proc) {
 582                 printk(KERN_ERR "ip_queue: failed to create proc entry\n");
 583                 goto cleanup_ipqnl;
 584         }
 585 #endif
 586         register_netdevice_notifier(&ipq_dev_notifier);
 587 #ifdef CONFIG_SYSCTL
 588         ipq_sysctl_header = register_sysctl_paths(net_ipv4_ctl_path, ipq_table);
 589 #endif
 590         status = nf_register_queue_handler(NFPROTO_IPV4, &nfqh);
 591         if (status < 0) {
 592                 printk(KERN_ERR "ip_queue: failed to register queue handler\n");
 593                 goto cleanup_sysctl;
 594         }
 595         return status;
 596
 597 cleanup_sysctl:
 598 #ifdef CONFIG_SYSCTL
 599         unregister_sysctl_table(ipq_sysctl_header);
 600 #endif
 601         unregister_netdevice_notifier(&ipq_dev_notifier);
 602         proc_net_remove(&init_net, IPQ_PROC_FS_NAME);
 603 cleanup_ipqnl: __maybe_unused
 604         netlink_kernel_release(ipqnl);
 605         mutex_lock(&ipqnl_mutex);
 606         mutex_unlock(&ipqnl_mutex);
 607
 608 cleanup_netlink_notifier:
 609         netlink_unregister_notifier(&ipq_nl_notifier);
 610         return status;
 611 }
 612
 613 static void __exit ip_queue_fini(void)
 614 {
 615         nf_unregister_queue_handlers(&nfqh);
 616
 617         ipq_flush(NULL, 0);
 618
 619 #ifdef CONFIG_SYSCTL
 620         unregister_sysctl_table(ipq_sysctl_header);
 621 #endif
 622         unregister_netdevice_notifier(&ipq_dev_notifier);
 623         proc_net_remove(&init_net, IPQ_PROC_FS_NAME);
 624
 625         netlink_kernel_release(ipqnl);
 626         mutex_lock(&ipqnl_mutex);
 627         mutex_unlock(&ipqnl_mutex);
 628
 629         netlink_unregister_notifier(&ipq_nl_notifier);
 630 }
 631
 632 MODULE_DESCRIPTION("IPv4 packet queue handler");
 633 MODULE_AUTHOR("James Morris <jmorris@intercode.com.au>");
 634 MODULE_LICENSE("GPL");
 635 MODULE_ALIAS_NET_PF_PROTO(PF_NETLINK, NETLINK_FIREWALL);
 636
 637 module_init(ip_queue_init);
 638 module_exit(ip_queue_fini);