]> pilppa.org Git - linux-2.6-omap-h63xx.git/blob - net/ipv6/netfilter/ip6_queue.c
/home/lenb/src/to-linus branch 'acpi-2.6.12'
[linux-2.6-omap-h63xx.git] / 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  * 2001-11-06: First try. Working with ip_queue.c for IPv4 and trying
16  *             to adapt it to IPv6
17  *             HEAVILY based in ipqueue.c by James Morris. It's just
18  *             a little modified version of it, so he's nearly the
19  *             real coder of this.
20  *             Few changes needed, mainly the hard_routing code and
21  *             the netlink socket protocol (we're NETLINK_IP6_FW).
22  * 2002-06-25: Code cleanup. [JM: ported cleanup over from ip_queue.c]
23  * 2005-02-04: Added /proc counter for dropped packets; fixed so
24  *             packets aren't delivered to user space if they're going
25  *             to be dropped.
26  */
27 #include <linux/module.h>
28 #include <linux/skbuff.h>
29 #include <linux/init.h>
30 #include <linux/ipv6.h>
31 #include <linux/notifier.h>
32 #include <linux/netdevice.h>
33 #include <linux/netfilter.h>
34 #include <linux/netlink.h>
35 #include <linux/spinlock.h>
36 #include <linux/sysctl.h>
37 #include <linux/proc_fs.h>
38 #include <net/sock.h>
39 #include <net/ipv6.h>
40 #include <net/ip6_route.h>
41 #include <linux/netfilter_ipv4/ip_queue.h>
42 #include <linux/netfilter_ipv4/ip_tables.h>
43 #include <linux/netfilter_ipv6/ip6_tables.h>
44
45 #define IPQ_QMAX_DEFAULT 1024
46 #define IPQ_PROC_FS_NAME "ip6_queue"
47 #define NET_IPQ_QMAX 2088
48 #define NET_IPQ_QMAX_NAME "ip6_queue_maxlen"
49
50 struct ipq_rt_info {
51         struct in6_addr daddr;
52         struct in6_addr saddr;
53 };
54
55 struct ipq_queue_entry {
56         struct list_head list;
57         struct nf_info *info;
58         struct sk_buff *skb;
59         struct ipq_rt_info rt_info;
60 };
61
62 typedef int (*ipq_cmpfn)(struct ipq_queue_entry *, unsigned long);
63
64 static unsigned char copy_mode = IPQ_COPY_NONE;
65 static unsigned int queue_maxlen = IPQ_QMAX_DEFAULT;
66 static DEFINE_RWLOCK(queue_lock);
67 static int peer_pid;
68 static unsigned int copy_range;
69 static unsigned int queue_total;
70 static unsigned int queue_dropped = 0;
71 static unsigned int queue_user_dropped = 0;
72 static struct sock *ipqnl;
73 static LIST_HEAD(queue_list);
74 static DECLARE_MUTEX(ipqnl_sem);
75
76 static void
77 ipq_issue_verdict(struct ipq_queue_entry *entry, int verdict)
78 {
79         local_bh_disable();
80         nf_reinject(entry->skb, entry->info, verdict);
81         local_bh_enable();
82         kfree(entry);
83 }
84
85 static inline void
86 __ipq_enqueue_entry(struct ipq_queue_entry *entry)
87 {
88        list_add(&entry->list, &queue_list);
89        queue_total++;
90 }
91
92 /*
93  * Find and return a queued entry matched by cmpfn, or return the last
94  * entry if cmpfn is NULL.
95  */
96 static inline struct ipq_queue_entry *
97 __ipq_find_entry(ipq_cmpfn cmpfn, unsigned long data)
98 {
99         struct list_head *p;
100
101         list_for_each_prev(p, &queue_list) {
102                 struct ipq_queue_entry *entry = (struct ipq_queue_entry *)p;
103                 
104                 if (!cmpfn || cmpfn(entry, data))
105                         return entry;
106         }
107         return NULL;
108 }
109
110 static inline void
111 __ipq_dequeue_entry(struct ipq_queue_entry *entry)
112 {
113         list_del(&entry->list);
114         queue_total--;
115 }
116
117 static inline struct ipq_queue_entry *
118 __ipq_find_dequeue_entry(ipq_cmpfn cmpfn, unsigned long data)
119 {
120         struct ipq_queue_entry *entry;
121
122         entry = __ipq_find_entry(cmpfn, data);
123         if (entry == NULL)
124                 return NULL;
125
126         __ipq_dequeue_entry(entry);
127         return entry;
128 }
129
130
131 static inline void
132 __ipq_flush(int verdict)
133 {
134         struct ipq_queue_entry *entry;
135         
136         while ((entry = __ipq_find_dequeue_entry(NULL, 0)))
137                 ipq_issue_verdict(entry, verdict);
138 }
139
140 static inline int
141 __ipq_set_mode(unsigned char mode, unsigned int range)
142 {
143         int status = 0;
144         
145         switch(mode) {
146         case IPQ_COPY_NONE:
147         case IPQ_COPY_META:
148                 copy_mode = mode;
149                 copy_range = 0;
150                 break;
151                 
152         case IPQ_COPY_PACKET:
153                 copy_mode = mode;
154                 copy_range = range;
155                 if (copy_range > 0xFFFF)
156                         copy_range = 0xFFFF;
157                 break;
158                 
159         default:
160                 status = -EINVAL;
161
162         }
163         return status;
164 }
165
166 static inline void
167 __ipq_reset(void)
168 {
169         peer_pid = 0;
170         net_disable_timestamp();
171         __ipq_set_mode(IPQ_COPY_NONE, 0);
172         __ipq_flush(NF_DROP);
173 }
174
175 static struct ipq_queue_entry *
176 ipq_find_dequeue_entry(ipq_cmpfn cmpfn, unsigned long data)
177 {
178         struct ipq_queue_entry *entry;
179         
180         write_lock_bh(&queue_lock);
181         entry = __ipq_find_dequeue_entry(cmpfn, data);
182         write_unlock_bh(&queue_lock);
183         return entry;
184 }
185
186 static void
187 ipq_flush(int verdict)
188 {
189         write_lock_bh(&queue_lock);
190         __ipq_flush(verdict);
191         write_unlock_bh(&queue_lock);
192 }
193
194 static struct sk_buff *
195 ipq_build_packet_message(struct ipq_queue_entry *entry, int *errp)
196 {
197         unsigned char *old_tail;
198         size_t size = 0;
199         size_t data_len = 0;
200         struct sk_buff *skb;
201         struct ipq_packet_msg *pmsg;
202         struct nlmsghdr *nlh;
203
204         read_lock_bh(&queue_lock);
205         
206         switch (copy_mode) {
207         case IPQ_COPY_META:
208         case IPQ_COPY_NONE:
209                 size = NLMSG_SPACE(sizeof(*pmsg));
210                 data_len = 0;
211                 break;
212         
213         case IPQ_COPY_PACKET:
214                 if (copy_range == 0 || copy_range > entry->skb->len)
215                         data_len = entry->skb->len;
216                 else
217                         data_len = copy_range;
218                 
219                 size = NLMSG_SPACE(sizeof(*pmsg) + data_len);
220                 break;
221         
222         default:
223                 *errp = -EINVAL;
224                 read_unlock_bh(&queue_lock);
225                 return NULL;
226         }
227
228         read_unlock_bh(&queue_lock);
229
230         skb = alloc_skb(size, GFP_ATOMIC);
231         if (!skb)
232                 goto nlmsg_failure;
233                 
234         old_tail= skb->tail;
235         nlh = NLMSG_PUT(skb, 0, 0, IPQM_PACKET, size - sizeof(*nlh));
236         pmsg = NLMSG_DATA(nlh);
237         memset(pmsg, 0, sizeof(*pmsg));
238
239         pmsg->packet_id       = (unsigned long )entry;
240         pmsg->data_len        = data_len;
241         pmsg->timestamp_sec   = entry->skb->stamp.tv_sec;
242         pmsg->timestamp_usec  = entry->skb->stamp.tv_usec;
243         pmsg->mark            = entry->skb->nfmark;
244         pmsg->hook            = entry->info->hook;
245         pmsg->hw_protocol     = entry->skb->protocol;
246         
247         if (entry->info->indev)
248                 strcpy(pmsg->indev_name, entry->info->indev->name);
249         else
250                 pmsg->indev_name[0] = '\0';
251         
252         if (entry->info->outdev)
253                 strcpy(pmsg->outdev_name, entry->info->outdev->name);
254         else
255                 pmsg->outdev_name[0] = '\0';
256         
257         if (entry->info->indev && entry->skb->dev) {
258                 pmsg->hw_type = entry->skb->dev->type;
259                 if (entry->skb->dev->hard_header_parse)
260                         pmsg->hw_addrlen =
261                                 entry->skb->dev->hard_header_parse(entry->skb,
262                                                                    pmsg->hw_addr);
263         }
264         
265         if (data_len)
266                 if (skb_copy_bits(entry->skb, 0, pmsg->payload, data_len))
267                         BUG();
268                 
269         nlh->nlmsg_len = skb->tail - old_tail;
270         return skb;
271
272 nlmsg_failure:
273         if (skb)
274                 kfree_skb(skb);
275         *errp = -EINVAL;
276         printk(KERN_ERR "ip6_queue: error creating packet message\n");
277         return NULL;
278 }
279
280 static int
281 ipq_enqueue_packet(struct sk_buff *skb, struct nf_info *info, void *data)
282 {
283         int status = -EINVAL;
284         struct sk_buff *nskb;
285         struct ipq_queue_entry *entry;
286
287         if (copy_mode == IPQ_COPY_NONE)
288                 return -EAGAIN;
289
290         entry = kmalloc(sizeof(*entry), GFP_ATOMIC);
291         if (entry == NULL) {
292                 printk(KERN_ERR "ip6_queue: OOM in ipq_enqueue_packet()\n");
293                 return -ENOMEM;
294         }
295
296         entry->info = info;
297         entry->skb = skb;
298
299         if (entry->info->hook == NF_IP_LOCAL_OUT) {
300                 struct ipv6hdr *iph = skb->nh.ipv6h;
301
302                 entry->rt_info.daddr = iph->daddr;
303                 entry->rt_info.saddr = iph->saddr;
304         }
305
306         nskb = ipq_build_packet_message(entry, &status);
307         if (nskb == NULL)
308                 goto err_out_free;
309                 
310         write_lock_bh(&queue_lock);
311         
312         if (!peer_pid)
313                 goto err_out_free_nskb; 
314
315         if (queue_total >= queue_maxlen) {
316                 queue_dropped++;
317                 status = -ENOSPC;
318                 if (net_ratelimit())
319                         printk (KERN_WARNING "ip6_queue: fill at %d entries, "
320                                 "dropping packet(s).  Dropped: %d\n", queue_total,
321                                 queue_dropped);
322                 goto err_out_free_nskb;
323         }
324
325         /* netlink_unicast will either free the nskb or attach it to a socket */ 
326         status = netlink_unicast(ipqnl, nskb, peer_pid, MSG_DONTWAIT);
327         if (status < 0) {
328                 queue_user_dropped++;
329                 goto err_out_unlock;
330         }
331         
332         __ipq_enqueue_entry(entry);
333
334         write_unlock_bh(&queue_lock);
335         return status;
336         
337 err_out_free_nskb:
338         kfree_skb(nskb); 
339         
340 err_out_unlock:
341         write_unlock_bh(&queue_lock);
342
343 err_out_free:
344         kfree(entry);
345         return status;
346 }
347
348 static int
349 ipq_mangle_ipv6(ipq_verdict_msg_t *v, struct ipq_queue_entry *e)
350 {
351         int diff;
352         struct ipv6hdr *user_iph = (struct ipv6hdr *)v->payload;
353
354         if (v->data_len < sizeof(*user_iph))
355                 return 0;
356         diff = v->data_len - e->skb->len;
357         if (diff < 0)
358                 skb_trim(e->skb, v->data_len);
359         else if (diff > 0) {
360                 if (v->data_len > 0xFFFF)
361                         return -EINVAL;
362                 if (diff > skb_tailroom(e->skb)) {
363                         struct sk_buff *newskb;
364                         
365                         newskb = skb_copy_expand(e->skb,
366                                                  skb_headroom(e->skb),
367                                                  diff,
368                                                  GFP_ATOMIC);
369                         if (newskb == NULL) {
370                                 printk(KERN_WARNING "ip6_queue: OOM "
371                                       "in mangle, dropping packet\n");
372                                 return -ENOMEM;
373                         }
374                         if (e->skb->sk)
375                                 skb_set_owner_w(newskb, e->skb->sk);
376                         kfree_skb(e->skb);
377                         e->skb = newskb;
378                 }
379                 skb_put(e->skb, diff);
380         }
381         if (!skb_ip_make_writable(&e->skb, v->data_len))
382                 return -ENOMEM;
383         memcpy(e->skb->data, v->payload, v->data_len);
384         e->skb->nfcache |= NFC_ALTERED;
385
386         /*
387          * Extra routing may needed on local out, as the QUEUE target never
388          * returns control to the table.
389          * Not a nice way to cmp, but works
390          */
391         if (e->info->hook == NF_IP_LOCAL_OUT) {
392                 struct ipv6hdr *iph = e->skb->nh.ipv6h;
393                 if (!ipv6_addr_equal(&iph->daddr, &e->rt_info.daddr) ||
394                     !ipv6_addr_equal(&iph->saddr, &e->rt_info.saddr))
395                         return ip6_route_me_harder(e->skb);
396         }
397         return 0;
398 }
399
400 static inline int
401 id_cmp(struct ipq_queue_entry *e, unsigned long id)
402 {
403         return (id == (unsigned long )e);
404 }
405
406 static int
407 ipq_set_verdict(struct ipq_verdict_msg *vmsg, unsigned int len)
408 {
409         struct ipq_queue_entry *entry;
410
411         if (vmsg->value > NF_MAX_VERDICT)
412                 return -EINVAL;
413
414         entry = ipq_find_dequeue_entry(id_cmp, vmsg->id);
415         if (entry == NULL)
416                 return -ENOENT;
417         else {
418                 int verdict = vmsg->value;
419                 
420                 if (vmsg->data_len && vmsg->data_len == len)
421                         if (ipq_mangle_ipv6(vmsg, entry) < 0)
422                                 verdict = NF_DROP;
423                 
424                 ipq_issue_verdict(entry, verdict);
425                 return 0;
426         }
427 }
428
429 static int
430 ipq_set_mode(unsigned char mode, unsigned int range)
431 {
432         int status;
433
434         write_lock_bh(&queue_lock);
435         status = __ipq_set_mode(mode, range);
436         write_unlock_bh(&queue_lock);
437         return status;
438 }
439
440 static int
441 ipq_receive_peer(struct ipq_peer_msg *pmsg,
442                  unsigned char type, unsigned int len)
443 {
444         int status = 0;
445
446         if (len < sizeof(*pmsg))
447                 return -EINVAL;
448
449         switch (type) {
450         case IPQM_MODE:
451                 status = ipq_set_mode(pmsg->msg.mode.value,
452                                       pmsg->msg.mode.range);
453                 break;
454                 
455         case IPQM_VERDICT:
456                 if (pmsg->msg.verdict.value > NF_MAX_VERDICT)
457                         status = -EINVAL;
458                 else
459                         status = ipq_set_verdict(&pmsg->msg.verdict,
460                                                  len - sizeof(*pmsg));
461                         break;
462         default:
463                 status = -EINVAL;
464         }
465         return status;
466 }
467
468 static int
469 dev_cmp(struct ipq_queue_entry *entry, unsigned long ifindex)
470 {
471         if (entry->info->indev)
472                 if (entry->info->indev->ifindex == ifindex)
473                         return 1;
474                         
475         if (entry->info->outdev)
476                 if (entry->info->outdev->ifindex == ifindex)
477                         return 1;
478
479         return 0;
480 }
481
482 static void
483 ipq_dev_drop(int ifindex)
484 {
485         struct ipq_queue_entry *entry;
486         
487         while ((entry = ipq_find_dequeue_entry(dev_cmp, ifindex)) != NULL)
488                 ipq_issue_verdict(entry, NF_DROP);
489 }
490
491 #define RCV_SKB_FAIL(err) do { netlink_ack(skb, nlh, (err)); return; } while (0)
492
493 static inline void
494 ipq_rcv_skb(struct sk_buff *skb)
495 {
496         int status, type, pid, flags, nlmsglen, skblen;
497         struct nlmsghdr *nlh;
498
499         skblen = skb->len;
500         if (skblen < sizeof(*nlh))
501                 return;
502
503         nlh = (struct nlmsghdr *)skb->data;
504         nlmsglen = nlh->nlmsg_len;
505         if (nlmsglen < sizeof(*nlh) || skblen < nlmsglen)
506                 return;
507
508         pid = nlh->nlmsg_pid;
509         flags = nlh->nlmsg_flags;
510         
511         if(pid <= 0 || !(flags & NLM_F_REQUEST) || flags & NLM_F_MULTI)
512                 RCV_SKB_FAIL(-EINVAL);
513                 
514         if (flags & MSG_TRUNC)
515                 RCV_SKB_FAIL(-ECOMM);
516                 
517         type = nlh->nlmsg_type;
518         if (type < NLMSG_NOOP || type >= IPQM_MAX)
519                 RCV_SKB_FAIL(-EINVAL);
520                 
521         if (type <= IPQM_BASE)
522                 return;
523         
524         if (security_netlink_recv(skb))
525                 RCV_SKB_FAIL(-EPERM);   
526
527         write_lock_bh(&queue_lock);
528         
529         if (peer_pid) {
530                 if (peer_pid != pid) {
531                         write_unlock_bh(&queue_lock);
532                         RCV_SKB_FAIL(-EBUSY);
533                 }
534         } else {
535                 net_enable_timestamp();
536                 peer_pid = pid;
537         }
538                 
539         write_unlock_bh(&queue_lock);
540         
541         status = ipq_receive_peer(NLMSG_DATA(nlh), type,
542                                   skblen - NLMSG_LENGTH(0));
543         if (status < 0)
544                 RCV_SKB_FAIL(status);
545                 
546         if (flags & NLM_F_ACK)
547                 netlink_ack(skb, nlh, 0);
548         return;
549 }
550
551 static void
552 ipq_rcv_sk(struct sock *sk, int len)
553 {
554         struct sk_buff *skb;
555         unsigned int qlen;
556
557         down(&ipqnl_sem);
558                         
559         for (qlen = skb_queue_len(&sk->sk_receive_queue); qlen; qlen--) {
560                 skb = skb_dequeue(&sk->sk_receive_queue);
561                 ipq_rcv_skb(skb);
562                 kfree_skb(skb);
563         }
564                 
565         up(&ipqnl_sem);
566 }
567
568 static int
569 ipq_rcv_dev_event(struct notifier_block *this,
570                   unsigned long event, void *ptr)
571 {
572         struct net_device *dev = ptr;
573
574         /* Drop any packets associated with the downed device */
575         if (event == NETDEV_DOWN)
576                 ipq_dev_drop(dev->ifindex);
577         return NOTIFY_DONE;
578 }
579
580 static struct notifier_block ipq_dev_notifier = {
581         .notifier_call  = ipq_rcv_dev_event,
582 };
583
584 static int
585 ipq_rcv_nl_event(struct notifier_block *this,
586                  unsigned long event, void *ptr)
587 {
588         struct netlink_notify *n = ptr;
589
590         if (event == NETLINK_URELEASE &&
591             n->protocol == NETLINK_IP6_FW && n->pid) {
592                 write_lock_bh(&queue_lock);
593                 if (n->pid == peer_pid)
594                         __ipq_reset();
595                 write_unlock_bh(&queue_lock);
596         }
597         return NOTIFY_DONE;
598 }
599
600 static struct notifier_block ipq_nl_notifier = {
601         .notifier_call  = ipq_rcv_nl_event,
602 };
603
604 static struct ctl_table_header *ipq_sysctl_header;
605
606 static ctl_table ipq_table[] = {
607         {
608                 .ctl_name       = NET_IPQ_QMAX,
609                 .procname       = NET_IPQ_QMAX_NAME,
610                 .data           = &queue_maxlen,
611                 .maxlen         = sizeof(queue_maxlen),
612                 .mode           = 0644,
613                 .proc_handler   = proc_dointvec
614         },
615         { .ctl_name = 0 }
616 };
617
618 static ctl_table ipq_dir_table[] = {
619         {
620                 .ctl_name       = NET_IPV6,
621                 .procname       = "ipv6",
622                 .mode           = 0555,
623                 .child          = ipq_table
624         },
625         { .ctl_name = 0 }
626 };
627
628 static ctl_table ipq_root_table[] = {
629         {
630                 .ctl_name       = CTL_NET,
631                 .procname       = "net",
632                 .mode           = 0555,
633                 .child          = ipq_dir_table
634         },
635         { .ctl_name = 0 }
636 };
637
638 static int
639 ipq_get_info(char *buffer, char **start, off_t offset, int length)
640 {
641         int len;
642
643         read_lock_bh(&queue_lock);
644         
645         len = sprintf(buffer,
646                       "Peer PID          : %d\n"
647                       "Copy mode         : %hu\n"
648                       "Copy range        : %u\n"
649                       "Queue length      : %u\n"
650                       "Queue max. length : %u\n"
651                       "Queue dropped     : %u\n"
652                       "Netfilter dropped : %u\n",
653                       peer_pid,
654                       copy_mode,
655                       copy_range,
656                       queue_total,
657                       queue_maxlen,
658                       queue_dropped,
659                       queue_user_dropped);
660
661         read_unlock_bh(&queue_lock);
662         
663         *start = buffer + offset;
664         len -= offset;
665         if (len > length)
666                 len = length;
667         else if (len < 0)
668                 len = 0;
669         return len;
670 }
671
672 static int
673 init_or_cleanup(int init)
674 {
675         int status = -ENOMEM;
676         struct proc_dir_entry *proc;
677         
678         if (!init)
679                 goto cleanup;
680
681         netlink_register_notifier(&ipq_nl_notifier);
682         ipqnl = netlink_kernel_create(NETLINK_IP6_FW, ipq_rcv_sk);
683         if (ipqnl == NULL) {
684                 printk(KERN_ERR "ip6_queue: failed to create netlink socket\n");
685                 goto cleanup_netlink_notifier;
686         }
687
688         proc = proc_net_create(IPQ_PROC_FS_NAME, 0, ipq_get_info);
689         if (proc)
690                 proc->owner = THIS_MODULE;
691         else {
692                 printk(KERN_ERR "ip6_queue: failed to create proc entry\n");
693                 goto cleanup_ipqnl;
694         }
695         
696         register_netdevice_notifier(&ipq_dev_notifier);
697         ipq_sysctl_header = register_sysctl_table(ipq_root_table, 0);
698         
699         status = nf_register_queue_handler(PF_INET6, ipq_enqueue_packet, NULL);
700         if (status < 0) {
701                 printk(KERN_ERR "ip6_queue: failed to register queue handler\n");
702                 goto cleanup_sysctl;
703         }
704         return status;
705
706 cleanup:
707         nf_unregister_queue_handler(PF_INET6);
708         synchronize_net();
709         ipq_flush(NF_DROP);
710         
711 cleanup_sysctl:
712         unregister_sysctl_table(ipq_sysctl_header);
713         unregister_netdevice_notifier(&ipq_dev_notifier);
714         proc_net_remove(IPQ_PROC_FS_NAME);
715         
716 cleanup_ipqnl:
717         sock_release(ipqnl->sk_socket);
718         down(&ipqnl_sem);
719         up(&ipqnl_sem);
720         
721 cleanup_netlink_notifier:
722         netlink_unregister_notifier(&ipq_nl_notifier);
723         return status;
724 }
725
726 static int __init init(void)
727 {
728         
729         return init_or_cleanup(1);
730 }
731
732 static void __exit fini(void)
733 {
734         init_or_cleanup(0);
735 }
736
737 MODULE_DESCRIPTION("IPv6 packet queue handler");
738 MODULE_LICENSE("GPL");
739
740 module_init(init);
741 module_exit(fini);