[NET] NETNS: Omit net_device->nd_net without CONFIG_NET_NS.
[linux-2.6] / net / ipv6 / netfilter / ip6_queue.c
1 /*
2  * This is a module which is used for queueing IPv6 packets and
3  * communicating with userspace via netlink.
4  *
5  * (C) 2001 Fernando Anton, this code is GPL.
6  *     IPv64 Project - Work based in IPv64 draft by Arturo Azcorra.
7  *     Universidad Carlos III de Madrid - Leganes (Madrid) - Spain
8  *     Universidad Politecnica de Alcala de Henares - Alcala de H. (Madrid) - Spain
9  *     email: fanton@it.uc3m.es
10  *
11  * This program is free software; you can redistribute it and/or modify
12  * it under the terms of the GNU General Public License version 2 as
13  * published by the Free Software Foundation.
14  */
15 #include <linux/module.h>
16 #include <linux/skbuff.h>
17 #include <linux/init.h>
18 #include <linux/ipv6.h>
19 #include <linux/notifier.h>
20 #include <linux/netdevice.h>
21 #include <linux/netfilter.h>
22 #include <linux/netlink.h>
23 #include <linux/spinlock.h>
24 #include <linux/sysctl.h>
25 #include <linux/proc_fs.h>
26 #include <linux/seq_file.h>
27 #include <linux/mutex.h>
28 #include <net/net_namespace.h>
29 #include <net/sock.h>
30 #include <net/ipv6.h>
31 #include <net/ip6_route.h>
32 #include <net/netfilter/nf_queue.h>
33 #include <linux/netfilter_ipv4/ip_queue.h>
34 #include <linux/netfilter_ipv4/ip_tables.h>
35 #include <linux/netfilter_ipv6/ip6_tables.h>
36
37 #define IPQ_QMAX_DEFAULT 1024
38 #define IPQ_PROC_FS_NAME "ip6_queue"
39 #define NET_IPQ_QMAX 2088
40 #define NET_IPQ_QMAX_NAME "ip6_queue_maxlen"
41
42 typedef int (*ipq_cmpfn)(struct nf_queue_entry *, unsigned long);
43
44 static unsigned char copy_mode __read_mostly = IPQ_COPY_NONE;
45 static unsigned int queue_maxlen __read_mostly = IPQ_QMAX_DEFAULT;
46 static DEFINE_RWLOCK(queue_lock);
47 static int peer_pid __read_mostly;
48 static unsigned int copy_range __read_mostly;
49 static unsigned int queue_total;
50 static unsigned int queue_dropped = 0;
51 static unsigned int queue_user_dropped = 0;
52 static struct sock *ipqnl __read_mostly;
53 static LIST_HEAD(queue_list);
54 static DEFINE_MUTEX(ipqnl_mutex);
55
56 static inline void
57 __ipq_enqueue_entry(struct nf_queue_entry *entry)
58 {
59        list_add_tail(&entry->list, &queue_list);
60        queue_total++;
61 }
62
63 static inline int
64 __ipq_set_mode(unsigned char mode, unsigned int range)
65 {
66         int status = 0;
67
68         switch(mode) {
69         case IPQ_COPY_NONE:
70         case IPQ_COPY_META:
71                 copy_mode = mode;
72                 copy_range = 0;
73                 break;
74
75         case IPQ_COPY_PACKET:
76                 copy_mode = mode;
77                 copy_range = range;
78                 if (copy_range > 0xFFFF)
79                         copy_range = 0xFFFF;
80                 break;
81
82         default:
83                 status = -EINVAL;
84
85         }
86         return status;
87 }
88
89 static void __ipq_flush(ipq_cmpfn cmpfn, unsigned long data);
90
91 static inline void
92 __ipq_reset(void)
93 {
94         peer_pid = 0;
95         net_disable_timestamp();
96         __ipq_set_mode(IPQ_COPY_NONE, 0);
97         __ipq_flush(NULL, 0);
98 }
99
100 static struct nf_queue_entry *
101 ipq_find_dequeue_entry(unsigned long id)
102 {
103         struct nf_queue_entry *entry = NULL, *i;
104
105         write_lock_bh(&queue_lock);
106
107         list_for_each_entry(i, &queue_list, list) {
108                 if ((unsigned long)i == id) {
109                         entry = i;
110                         break;
111                 }
112         }
113
114         if (entry) {
115                 list_del(&entry->list);
116                 queue_total--;
117         }
118
119         write_unlock_bh(&queue_lock);
120         return entry;
121 }
122
123 static void
124 __ipq_flush(ipq_cmpfn cmpfn, unsigned long data)
125 {
126         struct nf_queue_entry *entry, *next;
127
128         list_for_each_entry_safe(entry, next, &queue_list, list) {
129                 if (!cmpfn || cmpfn(entry, data)) {
130                         list_del(&entry->list);
131                         queue_total--;
132                         nf_reinject(entry, NF_DROP);
133                 }
134         }
135 }
136
137 static void
138 ipq_flush(ipq_cmpfn cmpfn, unsigned long data)
139 {
140         write_lock_bh(&queue_lock);
141         __ipq_flush(cmpfn, data);
142         write_unlock_bh(&queue_lock);
143 }
144
145 static struct sk_buff *
146 ipq_build_packet_message(struct nf_queue_entry *entry, int *errp)
147 {
148         sk_buff_data_t old_tail;
149         size_t size = 0;
150         size_t data_len = 0;
151         struct sk_buff *skb;
152         struct ipq_packet_msg *pmsg;
153         struct nlmsghdr *nlh;
154         struct timeval tv;
155
156         read_lock_bh(&queue_lock);
157
158         switch (copy_mode) {
159         case IPQ_COPY_META:
160         case IPQ_COPY_NONE:
161                 size = NLMSG_SPACE(sizeof(*pmsg));
162                 data_len = 0;
163                 break;
164
165         case IPQ_COPY_PACKET:
166                 if ((entry->skb->ip_summed == CHECKSUM_PARTIAL ||
167                      entry->skb->ip_summed == CHECKSUM_COMPLETE) &&
168                     (*errp = skb_checksum_help(entry->skb))) {
169                         read_unlock_bh(&queue_lock);
170                         return NULL;
171                 }
172                 if (copy_range == 0 || copy_range > entry->skb->len)
173                         data_len = entry->skb->len;
174                 else
175                         data_len = copy_range;
176
177                 size = NLMSG_SPACE(sizeof(*pmsg) + data_len);
178                 break;
179
180         default:
181                 *errp = -EINVAL;
182                 read_unlock_bh(&queue_lock);
183                 return NULL;
184         }
185
186         read_unlock_bh(&queue_lock);
187
188         skb = alloc_skb(size, GFP_ATOMIC);
189         if (!skb)
190                 goto nlmsg_failure;
191
192         old_tail = skb->tail;
193         nlh = NLMSG_PUT(skb, 0, 0, IPQM_PACKET, size - sizeof(*nlh));
194         pmsg = NLMSG_DATA(nlh);
195         memset(pmsg, 0, sizeof(*pmsg));
196
197         pmsg->packet_id       = (unsigned long )entry;
198         pmsg->data_len        = data_len;
199         tv = ktime_to_timeval(entry->skb->tstamp);
200         pmsg->timestamp_sec   = tv.tv_sec;
201         pmsg->timestamp_usec  = tv.tv_usec;
202         pmsg->mark            = entry->skb->mark;
203         pmsg->hook            = entry->hook;
204         pmsg->hw_protocol     = entry->skb->protocol;
205
206         if (entry->indev)
207                 strcpy(pmsg->indev_name, entry->indev->name);
208         else
209                 pmsg->indev_name[0] = '\0';
210
211         if (entry->outdev)
212                 strcpy(pmsg->outdev_name, entry->outdev->name);
213         else
214                 pmsg->outdev_name[0] = '\0';
215
216         if (entry->indev && entry->skb->dev) {
217                 pmsg->hw_type = entry->skb->dev->type;
218                 pmsg->hw_addrlen = dev_parse_header(entry->skb, pmsg->hw_addr);
219         }
220
221         if (data_len)
222                 if (skb_copy_bits(entry->skb, 0, pmsg->payload, data_len))
223                         BUG();
224
225         nlh->nlmsg_len = skb->tail - old_tail;
226         return skb;
227
228 nlmsg_failure:
229         if (skb)
230                 kfree_skb(skb);
231         *errp = -EINVAL;
232         printk(KERN_ERR "ip6_queue: error creating packet message\n");
233         return NULL;
234 }
235
236 static int
237 ipq_enqueue_packet(struct nf_queue_entry *entry, unsigned int queuenum)
238 {
239         int status = -EINVAL;
240         struct sk_buff *nskb;
241
242         if (copy_mode == IPQ_COPY_NONE)
243                 return -EAGAIN;
244
245         nskb = ipq_build_packet_message(entry, &status);
246         if (nskb == NULL)
247                 return status;
248
249         write_lock_bh(&queue_lock);
250
251         if (!peer_pid)
252                 goto err_out_free_nskb;
253
254         if (queue_total >= queue_maxlen) {
255                 queue_dropped++;
256                 status = -ENOSPC;
257                 if (net_ratelimit())
258                         printk (KERN_WARNING "ip6_queue: fill at %d entries, "
259                                 "dropping packet(s).  Dropped: %d\n", queue_total,
260                                 queue_dropped);
261                 goto err_out_free_nskb;
262         }
263
264         /* netlink_unicast will either free the nskb or attach it to a socket */
265         status = netlink_unicast(ipqnl, nskb, peer_pid, MSG_DONTWAIT);
266         if (status < 0) {
267                 queue_user_dropped++;
268                 goto err_out_unlock;
269         }
270
271         __ipq_enqueue_entry(entry);
272
273         write_unlock_bh(&queue_lock);
274         return status;
275
276 err_out_free_nskb:
277         kfree_skb(nskb);
278
279 err_out_unlock:
280         write_unlock_bh(&queue_lock);
281         return status;
282 }
283
284 static int
285 ipq_mangle_ipv6(ipq_verdict_msg_t *v, struct nf_queue_entry *e)
286 {
287         int diff;
288         struct ipv6hdr *user_iph = (struct ipv6hdr *)v->payload;
289         struct sk_buff *nskb;
290
291         if (v->data_len < sizeof(*user_iph))
292                 return 0;
293         diff = v->data_len - e->skb->len;
294         if (diff < 0) {
295                 if (pskb_trim(e->skb, v->data_len))
296                         return -ENOMEM;
297         } else if (diff > 0) {
298                 if (v->data_len > 0xFFFF)
299                         return -EINVAL;
300                 if (diff > skb_tailroom(e->skb)) {
301                         nskb = skb_copy_expand(e->skb, 0,
302                                                diff - skb_tailroom(e->skb),
303                                                GFP_ATOMIC);
304                         if (!nskb) {
305                                 printk(KERN_WARNING "ip6_queue: OOM "
306                                       "in mangle, dropping packet\n");
307                                 return -ENOMEM;
308                         }
309                         kfree_skb(e->skb);
310                         e->skb = nskb;
311                 }
312                 skb_put(e->skb, diff);
313         }
314         if (!skb_make_writable(e->skb, v->data_len))
315                 return -ENOMEM;
316         skb_copy_to_linear_data(e->skb, v->payload, v->data_len);
317         e->skb->ip_summed = CHECKSUM_NONE;
318
319         return 0;
320 }
321
322 static int
323 ipq_set_verdict(struct ipq_verdict_msg *vmsg, unsigned int len)
324 {
325         struct nf_queue_entry *entry;
326
327         if (vmsg->value > NF_MAX_VERDICT)
328                 return -EINVAL;
329
330         entry = ipq_find_dequeue_entry(vmsg->id);
331         if (entry == NULL)
332                 return -ENOENT;
333         else {
334                 int verdict = vmsg->value;
335
336                 if (vmsg->data_len && vmsg->data_len == len)
337                         if (ipq_mangle_ipv6(vmsg, entry) < 0)
338                                 verdict = NF_DROP;
339
340                 nf_reinject(entry, verdict);
341                 return 0;
342         }
343 }
344
345 static int
346 ipq_set_mode(unsigned char mode, unsigned int range)
347 {
348         int status;
349
350         write_lock_bh(&queue_lock);
351         status = __ipq_set_mode(mode, range);
352         write_unlock_bh(&queue_lock);
353         return status;
354 }
355
356 static int
357 ipq_receive_peer(struct ipq_peer_msg *pmsg,
358                  unsigned char type, unsigned int len)
359 {
360         int status = 0;
361
362         if (len < sizeof(*pmsg))
363                 return -EINVAL;
364
365         switch (type) {
366         case IPQM_MODE:
367                 status = ipq_set_mode(pmsg->msg.mode.value,
368                                       pmsg->msg.mode.range);
369                 break;
370
371         case IPQM_VERDICT:
372                 if (pmsg->msg.verdict.value > NF_MAX_VERDICT)
373                         status = -EINVAL;
374                 else
375                         status = ipq_set_verdict(&pmsg->msg.verdict,
376                                                  len - sizeof(*pmsg));
377                         break;
378         default:
379                 status = -EINVAL;
380         }
381         return status;
382 }
383
384 static int
385 dev_cmp(struct nf_queue_entry *entry, unsigned long ifindex)
386 {
387         if (entry->indev)
388                 if (entry->indev->ifindex == ifindex)
389                         return 1;
390
391         if (entry->outdev)
392                 if (entry->outdev->ifindex == ifindex)
393                         return 1;
394 #ifdef CONFIG_BRIDGE_NETFILTER
395         if (entry->skb->nf_bridge) {
396                 if (entry->skb->nf_bridge->physindev &&
397                     entry->skb->nf_bridge->physindev->ifindex == ifindex)
398                         return 1;
399                 if (entry->skb->nf_bridge->physoutdev &&
400                     entry->skb->nf_bridge->physoutdev->ifindex == ifindex)
401                         return 1;
402         }
403 #endif
404         return 0;
405 }
406
407 static void
408 ipq_dev_drop(int ifindex)
409 {
410         ipq_flush(dev_cmp, ifindex);
411 }
412
413 #define RCV_SKB_FAIL(err) do { netlink_ack(skb, nlh, (err)); return; } while (0)
414
415 static inline void
416 __ipq_rcv_skb(struct sk_buff *skb)
417 {
418         int status, type, pid, flags, nlmsglen, skblen;
419         struct nlmsghdr *nlh;
420
421         skblen = skb->len;
422         if (skblen < sizeof(*nlh))
423                 return;
424
425         nlh = nlmsg_hdr(skb);
426         nlmsglen = nlh->nlmsg_len;
427         if (nlmsglen < sizeof(*nlh) || skblen < nlmsglen)
428                 return;
429
430         pid = nlh->nlmsg_pid;
431         flags = nlh->nlmsg_flags;
432
433         if(pid <= 0 || !(flags & NLM_F_REQUEST) || flags & NLM_F_MULTI)
434                 RCV_SKB_FAIL(-EINVAL);
435
436         if (flags & MSG_TRUNC)
437                 RCV_SKB_FAIL(-ECOMM);
438
439         type = nlh->nlmsg_type;
440         if (type < NLMSG_NOOP || type >= IPQM_MAX)
441                 RCV_SKB_FAIL(-EINVAL);
442
443         if (type <= IPQM_BASE)
444                 return;
445
446         if (security_netlink_recv(skb, CAP_NET_ADMIN))
447                 RCV_SKB_FAIL(-EPERM);
448
449         write_lock_bh(&queue_lock);
450
451         if (peer_pid) {
452                 if (peer_pid != pid) {
453                         write_unlock_bh(&queue_lock);
454                         RCV_SKB_FAIL(-EBUSY);
455                 }
456         } else {
457                 net_enable_timestamp();
458                 peer_pid = pid;
459         }
460
461         write_unlock_bh(&queue_lock);
462
463         status = ipq_receive_peer(NLMSG_DATA(nlh), type,
464                                   nlmsglen - NLMSG_LENGTH(0));
465         if (status < 0)
466                 RCV_SKB_FAIL(status);
467
468         if (flags & NLM_F_ACK)
469                 netlink_ack(skb, nlh, 0);
470         return;
471 }
472
473 static void
474 ipq_rcv_skb(struct sk_buff *skb)
475 {
476         mutex_lock(&ipqnl_mutex);
477         __ipq_rcv_skb(skb);
478         mutex_unlock(&ipqnl_mutex);
479 }
480
481 static int
482 ipq_rcv_dev_event(struct notifier_block *this,
483                   unsigned long event, void *ptr)
484 {
485         struct net_device *dev = ptr;
486
487         if (dev_net(dev) != &init_net)
488                 return NOTIFY_DONE;
489
490         /* Drop any packets associated with the downed device */
491         if (event == NETDEV_DOWN)
492                 ipq_dev_drop(dev->ifindex);
493         return NOTIFY_DONE;
494 }
495
496 static struct notifier_block ipq_dev_notifier = {
497         .notifier_call  = ipq_rcv_dev_event,
498 };
499
500 static int
501 ipq_rcv_nl_event(struct notifier_block *this,
502                  unsigned long event, void *ptr)
503 {
504         struct netlink_notify *n = ptr;
505
506         if (event == NETLINK_URELEASE &&
507             n->protocol == NETLINK_IP6_FW && n->pid) {
508                 write_lock_bh(&queue_lock);
509                 if ((n->net == &init_net) && (n->pid == peer_pid))
510                         __ipq_reset();
511                 write_unlock_bh(&queue_lock);
512         }
513         return NOTIFY_DONE;
514 }
515
516 static struct notifier_block ipq_nl_notifier = {
517         .notifier_call  = ipq_rcv_nl_event,
518 };
519
520 #ifdef CONFIG_SYSCTL
521 static struct ctl_table_header *ipq_sysctl_header;
522
523 static ctl_table ipq_table[] = {
524         {
525                 .ctl_name       = NET_IPQ_QMAX,
526                 .procname       = NET_IPQ_QMAX_NAME,
527                 .data           = &queue_maxlen,
528                 .maxlen         = sizeof(queue_maxlen),
529                 .mode           = 0644,
530                 .proc_handler   = proc_dointvec
531         },
532         { .ctl_name = 0 }
533 };
534 #endif
535
536 #ifdef CONFIG_PROC_FS
537 static int ip6_queue_show(struct seq_file *m, void *v)
538 {
539         read_lock_bh(&queue_lock);
540
541         seq_printf(m,
542                       "Peer PID          : %d\n"
543                       "Copy mode         : %hu\n"
544                       "Copy range        : %u\n"
545                       "Queue length      : %u\n"
546                       "Queue max. length : %u\n"
547                       "Queue dropped     : %u\n"
548                       "Netfilter dropped : %u\n",
549                       peer_pid,
550                       copy_mode,
551                       copy_range,
552                       queue_total,
553                       queue_maxlen,
554                       queue_dropped,
555                       queue_user_dropped);
556
557         read_unlock_bh(&queue_lock);
558         return 0;
559 }
560
561 static int ip6_queue_open(struct inode *inode, struct file *file)
562 {
563         return single_open(file, ip6_queue_show, NULL);
564 }
565
566 static const struct file_operations ip6_queue_proc_fops = {
567         .open           = ip6_queue_open,
568         .read           = seq_read,
569         .llseek         = seq_lseek,
570         .release        = single_release,
571         .owner          = THIS_MODULE,
572 };
573 #endif
574
575 static const struct nf_queue_handler nfqh = {
576         .name   = "ip6_queue",
577         .outfn  = &ipq_enqueue_packet,
578 };
579
580 static int __init ip6_queue_init(void)
581 {
582         int status = -ENOMEM;
583         struct proc_dir_entry *proc __maybe_unused;
584
585         netlink_register_notifier(&ipq_nl_notifier);
586         ipqnl = netlink_kernel_create(&init_net, NETLINK_IP6_FW, 0,
587                                       ipq_rcv_skb, NULL, THIS_MODULE);
588         if (ipqnl == NULL) {
589                 printk(KERN_ERR "ip6_queue: failed to create netlink socket\n");
590                 goto cleanup_netlink_notifier;
591         }
592
593 #ifdef CONFIG_PROC_FS
594         proc = create_proc_entry(IPQ_PROC_FS_NAME, 0, init_net.proc_net);
595         if (proc) {
596                 proc->owner = THIS_MODULE;
597                 proc->proc_fops = &ip6_queue_proc_fops;
598         } else {
599                 printk(KERN_ERR "ip6_queue: failed to create proc entry\n");
600                 goto cleanup_ipqnl;
601         }
602 #endif
603         register_netdevice_notifier(&ipq_dev_notifier);
604 #ifdef CONFIG_SYSCTL
605         ipq_sysctl_header = register_sysctl_paths(net_ipv6_ctl_path, ipq_table);
606 #endif
607         status = nf_register_queue_handler(PF_INET6, &nfqh);
608         if (status < 0) {
609                 printk(KERN_ERR "ip6_queue: failed to register queue handler\n");
610                 goto cleanup_sysctl;
611         }
612         return status;
613
614 cleanup_sysctl:
615 #ifdef CONFIG_SYSCTL
616         unregister_sysctl_table(ipq_sysctl_header);
617 #endif
618         unregister_netdevice_notifier(&ipq_dev_notifier);
619         proc_net_remove(&init_net, IPQ_PROC_FS_NAME);
620
621 cleanup_ipqnl: __maybe_unused
622         netlink_kernel_release(ipqnl);
623         mutex_lock(&ipqnl_mutex);
624         mutex_unlock(&ipqnl_mutex);
625
626 cleanup_netlink_notifier:
627         netlink_unregister_notifier(&ipq_nl_notifier);
628         return status;
629 }
630
631 static void __exit ip6_queue_fini(void)
632 {
633         nf_unregister_queue_handlers(&nfqh);
634         synchronize_net();
635         ipq_flush(NULL, 0);
636
637 #ifdef CONFIG_SYSCTL
638         unregister_sysctl_table(ipq_sysctl_header);
639 #endif
640         unregister_netdevice_notifier(&ipq_dev_notifier);
641         proc_net_remove(&init_net, IPQ_PROC_FS_NAME);
642
643         netlink_kernel_release(ipqnl);
644         mutex_lock(&ipqnl_mutex);
645         mutex_unlock(&ipqnl_mutex);
646
647         netlink_unregister_notifier(&ipq_nl_notifier);
648 }
649
650 MODULE_DESCRIPTION("IPv6 packet queue handler");
651 MODULE_LICENSE("GPL");
652
653 module_init(ip6_queue_init);
654 module_exit(ip6_queue_fini);