9e40dffc204f3438e0dbabc1ed89be0031f13449
[linux-2.6.git] / 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  *
7  * This program is free software; you can redistribute it and/or modify
8  * it under the terms of the GNU General Public License version 2 as
9  * published by the Free Software Foundation.
10  *
11  * 2000-03-27: Simplified code (thanks to Andi Kleen for clues).
12  * 2000-05-20: Fixed notifier problems (following Miguel Freitas' report).
13  * 2000-06-19: Fixed so nfmark is copied to metadata (reported by Sebastian 
14  *             Zander).
15  * 2000-08-01: Added Nick Williams' MAC support.
16  * 2002-06-25: Code cleanup.
17  * 2005-01-10: Added /proc counter for dropped packets; fixed so
18  *             packets aren't delivered to user space if they're going 
19  *             to be dropped. 
20  *
21  */
22 #include <linux/module.h>
23 #include <linux/skbuff.h>
24 #include <linux/init.h>
25 #include <linux/ip.h>
26 #include <linux/notifier.h>
27 #include <linux/netdevice.h>
28 #include <linux/netfilter.h>
29 #include <linux/netfilter_ipv4/ip_queue.h>
30 #include <linux/netfilter_ipv4/ip_tables.h>
31 #include <linux/netlink.h>
32 #include <linux/spinlock.h>
33 #include <linux/sysctl.h>
34 #include <linux/proc_fs.h>
35 #include <linux/security.h>
36 #include <net/sock.h>
37 #include <net/route.h>
38
39 #define IPQ_QMAX_DEFAULT 1024
40 #define IPQ_PROC_FS_NAME "ip_queue"
41 #define NET_IPQ_QMAX 2088
42 #define NET_IPQ_QMAX_NAME "ip_queue_maxlen"
43
44 struct ipq_rt_info {
45         __u8 tos;
46         __u32 daddr;
47         __u32 saddr;
48 };
49
50 struct ipq_queue_entry {
51         struct list_head list;
52         struct nf_info *info;
53         struct sk_buff *skb;
54         struct ipq_rt_info rt_info;
55 };
56
57 typedef int (*ipq_cmpfn)(struct ipq_queue_entry *, unsigned long);
58
59 static unsigned char copy_mode = IPQ_COPY_NONE;
60 static unsigned int queue_maxlen = IPQ_QMAX_DEFAULT;
61 static DEFINE_RWLOCK(queue_lock);
62 static int peer_pid;
63 static unsigned int copy_range;
64 static unsigned int queue_total;
65 static unsigned int queue_dropped = 0;
66 static unsigned int queue_user_dropped = 0;
67 static struct sock *ipqnl;
68 static LIST_HEAD(queue_list);
69 static DECLARE_MUTEX(ipqnl_sem);
70
71 static void
72 ipq_issue_verdict(struct ipq_queue_entry *entry, int verdict)
73 {
74         nf_reinject(entry->skb, entry->info, verdict);
75         kfree(entry);
76 }
77
78 static inline void
79 __ipq_enqueue_entry(struct ipq_queue_entry *entry)
80 {
81        list_add(&entry->list, &queue_list);
82        queue_total++;
83 }
84
85 /*
86  * Find and return a queued entry matched by cmpfn, or return the last
87  * entry if cmpfn is NULL.
88  */
89 static inline struct ipq_queue_entry *
90 __ipq_find_entry(ipq_cmpfn cmpfn, unsigned long data)
91 {
92         struct list_head *p;
93
94         list_for_each_prev(p, &queue_list) {
95                 struct ipq_queue_entry *entry = (struct ipq_queue_entry *)p;
96                 
97                 if (!cmpfn || cmpfn(entry, data))
98                         return entry;
99         }
100         return NULL;
101 }
102
103 static inline void
104 __ipq_dequeue_entry(struct ipq_queue_entry *entry)
105 {
106         list_del(&entry->list);
107         queue_total--;
108 }
109
110 static inline struct ipq_queue_entry *
111 __ipq_find_dequeue_entry(ipq_cmpfn cmpfn, unsigned long data)
112 {
113         struct ipq_queue_entry *entry;
114
115         entry = __ipq_find_entry(cmpfn, data);
116         if (entry == NULL)
117                 return NULL;
118
119         __ipq_dequeue_entry(entry);
120         return entry;
121 }
122
123
124 static inline void
125 __ipq_flush(int verdict)
126 {
127         struct ipq_queue_entry *entry;
128         
129         while ((entry = __ipq_find_dequeue_entry(NULL, 0)))
130                 ipq_issue_verdict(entry, verdict);
131 }
132
133 static inline int
134 __ipq_set_mode(unsigned char mode, unsigned int range)
135 {
136         int status = 0;
137         
138         switch(mode) {
139         case IPQ_COPY_NONE:
140         case IPQ_COPY_META:
141                 copy_mode = mode;
142                 copy_range = 0;
143                 break;
144                 
145         case IPQ_COPY_PACKET:
146                 copy_mode = mode;
147                 copy_range = range;
148                 if (copy_range > 0xFFFF)
149                         copy_range = 0xFFFF;
150                 break;
151                 
152         default:
153                 status = -EINVAL;
154
155         }
156         return status;
157 }
158
159 static inline void
160 __ipq_reset(void)
161 {
162         peer_pid = 0;
163         net_disable_timestamp();
164         __ipq_set_mode(IPQ_COPY_NONE, 0);
165         __ipq_flush(NF_DROP);
166 }
167
168 static struct ipq_queue_entry *
169 ipq_find_dequeue_entry(ipq_cmpfn cmpfn, unsigned long data)
170 {
171         struct ipq_queue_entry *entry;
172         
173         write_lock_bh(&queue_lock);
174         entry = __ipq_find_dequeue_entry(cmpfn, data);
175         write_unlock_bh(&queue_lock);
176         return entry;
177 }
178
179 static void
180 ipq_flush(int verdict)
181 {
182         write_lock_bh(&queue_lock);
183         __ipq_flush(verdict);
184         write_unlock_bh(&queue_lock);
185 }
186
187 static struct sk_buff *
188 ipq_build_packet_message(struct ipq_queue_entry *entry, int *errp)
189 {
190         unsigned char *old_tail;
191         size_t size = 0;
192         size_t data_len = 0;
193         struct sk_buff *skb;
194         struct ipq_packet_msg *pmsg;
195         struct nlmsghdr *nlh;
196
197         read_lock_bh(&queue_lock);
198         
199         switch (copy_mode) {
200         case IPQ_COPY_META:
201         case IPQ_COPY_NONE:
202                 size = NLMSG_SPACE(sizeof(*pmsg));
203                 data_len = 0;
204                 break;
205         
206         case IPQ_COPY_PACKET:
207                 if (copy_range == 0 || copy_range > entry->skb->len)
208                         data_len = entry->skb->len;
209                 else
210                         data_len = copy_range;
211                 
212                 size = NLMSG_SPACE(sizeof(*pmsg) + data_len);
213                 break;
214         
215         default:
216                 *errp = -EINVAL;
217                 read_unlock_bh(&queue_lock);
218                 return NULL;
219         }
220
221         read_unlock_bh(&queue_lock);
222
223         skb = alloc_skb(size, GFP_ATOMIC);
224         if (!skb)
225                 goto nlmsg_failure;
226                 
227         old_tail= skb->tail;
228         nlh = NLMSG_PUT(skb, 0, 0, IPQM_PACKET, size - sizeof(*nlh));
229         pmsg = NLMSG_DATA(nlh);
230         memset(pmsg, 0, sizeof(*pmsg));
231
232         pmsg->packet_id       = (unsigned long )entry;
233         pmsg->data_len        = data_len;
234         pmsg->timestamp_sec   = entry->skb->stamp.tv_sec;
235         pmsg->timestamp_usec  = entry->skb->stamp.tv_usec;
236         pmsg->mark            = entry->skb->nfmark;
237         pmsg->hook            = entry->info->hook;
238         pmsg->hw_protocol     = entry->skb->protocol;
239         
240         if (entry->info->indev)
241                 strcpy(pmsg->indev_name, entry->info->indev->name);
242         else
243                 pmsg->indev_name[0] = '\0';
244         
245         if (entry->info->outdev)
246                 strcpy(pmsg->outdev_name, entry->info->outdev->name);
247         else
248                 pmsg->outdev_name[0] = '\0';
249         
250         if (entry->info->indev && entry->skb->dev) {
251                 pmsg->hw_type = entry->skb->dev->type;
252                 if (entry->skb->dev->hard_header_parse)
253                         pmsg->hw_addrlen =
254                                 entry->skb->dev->hard_header_parse(entry->skb,
255                                                                    pmsg->hw_addr);
256         }
257         
258         if (data_len)
259                 if (skb_copy_bits(entry->skb, 0, pmsg->payload, data_len))
260                         BUG();
261                 
262         nlh->nlmsg_len = skb->tail - old_tail;
263         return skb;
264
265 nlmsg_failure:
266         if (skb)
267                 kfree_skb(skb);
268         *errp = -EINVAL;
269         printk(KERN_ERR "ip_queue: error creating packet message\n");
270         return NULL;
271 }
272
273 static int
274 ipq_enqueue_packet(struct sk_buff *skb, struct nf_info *info, void *data)
275 {
276         int status = -EINVAL;
277         struct sk_buff *nskb;
278         struct ipq_queue_entry *entry;
279
280         if (copy_mode == IPQ_COPY_NONE)
281                 return -EAGAIN;
282
283         entry = kmalloc(sizeof(*entry), GFP_ATOMIC);
284         if (entry == NULL) {
285                 printk(KERN_ERR "ip_queue: OOM in ipq_enqueue_packet()\n");
286                 return -ENOMEM;
287         }
288
289         entry->info = info;
290         entry->skb = skb;
291
292         if (entry->info->hook == NF_IP_LOCAL_OUT) {
293                 struct iphdr *iph = skb->nh.iph;
294
295                 entry->rt_info.tos = iph->tos;
296                 entry->rt_info.daddr = iph->daddr;
297                 entry->rt_info.saddr = iph->saddr;
298         }
299
300         nskb = ipq_build_packet_message(entry, &status);
301         if (nskb == NULL)
302                 goto err_out_free;
303                 
304         write_lock_bh(&queue_lock);
305         
306         if (!peer_pid)
307                 goto err_out_free_nskb; 
308
309         if (queue_total >= queue_maxlen) {
310                 queue_dropped++;
311                 status = -ENOSPC;
312                 if (net_ratelimit())
313                           printk (KERN_WARNING "ip_queue: full at %d entries, "
314                                   "dropping packets(s). Dropped: %d\n", queue_total,
315                                   queue_dropped);
316                 goto err_out_free_nskb;
317         }
318
319         /* netlink_unicast will either free the nskb or attach it to a socket */ 
320         status = netlink_unicast(ipqnl, nskb, peer_pid, MSG_DONTWAIT);
321         if (status < 0) {
322                 queue_user_dropped++;
323                 goto err_out_unlock;
324         }
325
326         __ipq_enqueue_entry(entry);
327
328         write_unlock_bh(&queue_lock);
329         return status;
330
331 err_out_free_nskb:
332         kfree_skb(nskb); 
333         
334 err_out_unlock:
335         write_unlock_bh(&queue_lock);
336
337 err_out_free:
338         kfree(entry);
339         return status;
340 }
341
342 static int
343 ipq_mangle_ipv4(ipq_verdict_msg_t *v, struct ipq_queue_entry *e)
344 {
345         int diff;
346         struct iphdr *user_iph = (struct iphdr *)v->payload;
347
348         if (v->data_len < sizeof(*user_iph))
349                 return 0;
350         diff = v->data_len - e->skb->len;
351         if (diff < 0)
352                 skb_trim(e->skb, v->data_len);
353         else if (diff > 0) {
354                 if (v->data_len > 0xFFFF)
355                         return -EINVAL;
356                 if (diff > skb_tailroom(e->skb)) {
357                         struct sk_buff *newskb;
358                         
359                         newskb = skb_copy_expand(e->skb,
360                                                  skb_headroom(e->skb),
361                                                  diff,
362                                                  GFP_ATOMIC);
363                         if (newskb == NULL) {
364                                 printk(KERN_WARNING "ip_queue: OOM "
365                                       "in mangle, dropping packet\n");
366                                 return -ENOMEM;
367                         }
368                         if (e->skb->sk)
369                                 skb_set_owner_w(newskb, e->skb->sk);
370                         kfree_skb(e->skb);
371                         e->skb = newskb;
372                 }
373                 skb_put(e->skb, diff);
374         }
375         if (!skb_ip_make_writable(&e->skb, v->data_len))
376                 return -ENOMEM;
377         memcpy(e->skb->data, v->payload, v->data_len);
378         e->skb->nfcache |= NFC_ALTERED;
379
380         /*
381          * Extra routing may needed on local out, as the QUEUE target never
382          * returns control to the table.
383          */
384         if (e->info->hook == NF_IP_LOCAL_OUT) {
385                 struct iphdr *iph = e->skb->nh.iph;
386
387                 if (!(iph->tos == e->rt_info.tos
388                       && iph->daddr == e->rt_info.daddr
389                       && iph->saddr == e->rt_info.saddr))
390                         return ip_route_me_harder(&e->skb);
391         }
392         return 0;
393 }
394
395 static inline int
396 id_cmp(struct ipq_queue_entry *e, unsigned long id)
397 {
398         return (id == (unsigned long )e);
399 }
400
401 static int
402 ipq_set_verdict(struct ipq_verdict_msg *vmsg, unsigned int len)
403 {
404         struct ipq_queue_entry *entry;
405
406         if (vmsg->value > NF_MAX_VERDICT)
407                 return -EINVAL;
408
409         entry = ipq_find_dequeue_entry(id_cmp, vmsg->id);
410         if (entry == NULL)
411                 return -ENOENT;
412         else {
413                 int verdict = vmsg->value;
414                 
415                 if (vmsg->data_len && vmsg->data_len == len)
416                         if (ipq_mangle_ipv4(vmsg, entry) < 0)
417                                 verdict = NF_DROP;
418                 
419                 ipq_issue_verdict(entry, verdict);
420                 return 0;
421         }
422 }
423
424 static int
425 ipq_set_mode(unsigned char mode, unsigned int range)
426 {
427         int status;
428
429         write_lock_bh(&queue_lock);
430         status = __ipq_set_mode(mode, range);
431         write_unlock_bh(&queue_lock);
432         return status;
433 }
434
435 static int
436 ipq_receive_peer(struct ipq_peer_msg *pmsg,
437                  unsigned char type, unsigned int len)
438 {
439         int status = 0;
440
441         if (len < sizeof(*pmsg))
442                 return -EINVAL;
443
444         switch (type) {
445         case IPQM_MODE:
446                 status = ipq_set_mode(pmsg->msg.mode.value,
447                                       pmsg->msg.mode.range);
448                 break;
449                 
450         case IPQM_VERDICT:
451                 if (pmsg->msg.verdict.value > NF_MAX_VERDICT)
452                         status = -EINVAL;
453                 else
454                         status = ipq_set_verdict(&pmsg->msg.verdict,
455                                                  len - sizeof(*pmsg));
456                         break;
457         default:
458                 status = -EINVAL;
459         }
460         return status;
461 }
462
463 static int
464 dev_cmp(struct ipq_queue_entry *entry, unsigned long ifindex)
465 {
466         if (entry->info->indev)
467                 if (entry->info->indev->ifindex == ifindex)
468                         return 1;
469                         
470         if (entry->info->outdev)
471                 if (entry->info->outdev->ifindex == ifindex)
472                         return 1;
473
474         return 0;
475 }
476
477 static void
478 ipq_dev_drop(int ifindex)
479 {
480         struct ipq_queue_entry *entry;
481         
482         while ((entry = ipq_find_dequeue_entry(dev_cmp, ifindex)) != NULL)
483                 ipq_issue_verdict(entry, NF_DROP);
484 }
485
486 #define RCV_SKB_FAIL(err) do { netlink_ack(skb, nlh, (err)); return; } while (0)
487
488 static inline void
489 ipq_rcv_skb(struct sk_buff *skb)
490 {
491         int status, type, pid, flags, nlmsglen, skblen;
492         struct nlmsghdr *nlh;
493
494         skblen = skb->len;
495         if (skblen < sizeof(*nlh))
496                 return;
497
498         nlh = (struct nlmsghdr *)skb->data;
499         nlmsglen = nlh->nlmsg_len;
500         if (nlmsglen < sizeof(*nlh) || skblen < nlmsglen)
501                 return;
502
503         pid = nlh->nlmsg_pid;
504         flags = nlh->nlmsg_flags;
505         
506         if(pid <= 0 || !(flags & NLM_F_REQUEST) || flags & NLM_F_MULTI)
507                 RCV_SKB_FAIL(-EINVAL);
508                 
509         if (flags & MSG_TRUNC)
510                 RCV_SKB_FAIL(-ECOMM);
511                 
512         type = nlh->nlmsg_type;
513         if (type < NLMSG_NOOP || type >= IPQM_MAX)
514                 RCV_SKB_FAIL(-EINVAL);
515                 
516         if (type <= IPQM_BASE)
517                 return;
518                 
519         if (security_netlink_recv(skb))
520                 RCV_SKB_FAIL(-EPERM);
521         
522         write_lock_bh(&queue_lock);
523         
524         if (peer_pid) {
525                 if (peer_pid != pid) {
526                         write_unlock_bh(&queue_lock);
527                         RCV_SKB_FAIL(-EBUSY);
528                 }
529         } else {
530                 net_enable_timestamp();
531                 peer_pid = pid;
532         }
533                 
534         write_unlock_bh(&queue_lock);
535         
536         status = ipq_receive_peer(NLMSG_DATA(nlh), type,
537                                   skblen - NLMSG_LENGTH(0));
538         if (status < 0)
539                 RCV_SKB_FAIL(status);
540                 
541         if (flags & NLM_F_ACK)
542                 netlink_ack(skb, nlh, 0);
543         return;
544 }
545
546 static void
547 ipq_rcv_sk(struct sock *sk, int len)
548 {
549         do {
550                 struct sk_buff *skb;
551
552                 if (down_trylock(&ipqnl_sem))
553                         return;
554                         
555                 while ((skb = skb_dequeue(&sk->sk_receive_queue)) != NULL) {
556                         ipq_rcv_skb(skb);
557                         kfree_skb(skb);
558                 }
559                 
560                 up(&ipqnl_sem);
561
562         } while (ipqnl && ipqnl->sk_receive_queue.qlen);
563 }
564
565 static int
566 ipq_rcv_dev_event(struct notifier_block *this,
567                   unsigned long event, void *ptr)
568 {
569         struct net_device *dev = ptr;
570
571         /* Drop any packets associated with the downed device */
572         if (event == NETDEV_DOWN)
573                 ipq_dev_drop(dev->ifindex);
574         return NOTIFY_DONE;
575 }
576
577 static struct notifier_block ipq_dev_notifier = {
578         .notifier_call  = ipq_rcv_dev_event,
579 };
580
581 static int
582 ipq_rcv_nl_event(struct notifier_block *this,
583                  unsigned long event, void *ptr)
584 {
585         struct netlink_notify *n = ptr;
586
587         if (event == NETLINK_URELEASE &&
588             n->protocol == NETLINK_FIREWALL && n->pid) {
589                 write_lock_bh(&queue_lock);
590                 if (n->pid == peer_pid)
591                         __ipq_reset();
592                 write_unlock_bh(&queue_lock);
593         }
594         return NOTIFY_DONE;
595 }
596
597 static struct notifier_block ipq_nl_notifier = {
598         .notifier_call  = ipq_rcv_nl_event,
599 };
600
601 static struct ctl_table_header *ipq_sysctl_header;
602
603 static ctl_table ipq_table[] = {
604         {
605                 .ctl_name       = NET_IPQ_QMAX,
606                 .procname       = NET_IPQ_QMAX_NAME,
607                 .data           = &queue_maxlen,
608                 .maxlen         = sizeof(queue_maxlen),
609                 .mode           = 0644,
610                 .proc_handler   = proc_dointvec
611         },
612         { .ctl_name = 0 }
613 };
614
615 static ctl_table ipq_dir_table[] = {
616         {
617                 .ctl_name       = NET_IPV4,
618                 .procname       = "ipv4",
619                 .mode           = 0555,
620                 .child          = ipq_table
621         },
622         { .ctl_name = 0 }
623 };
624
625 static ctl_table ipq_root_table[] = {
626         {
627                 .ctl_name       = CTL_NET,
628                 .procname       = "net",
629                 .mode           = 0555,
630                 .child          = ipq_dir_table
631         },
632         { .ctl_name = 0 }
633 };
634
635 #ifdef CONFIG_PROC_FS
636 static int
637 ipq_get_info(char *buffer, char **start, off_t offset, int length)
638 {
639         int len;
640
641         read_lock_bh(&queue_lock);
642         
643         len = sprintf(buffer,
644                       "Peer PID          : %d\n"
645                       "Copy mode         : %hu\n"
646                       "Copy range        : %u\n"
647                       "Queue length      : %u\n"
648                       "Queue max. length : %u\n"
649                       "Queue dropped     : %u\n"
650                       "Netlink dropped   : %u\n",
651                       peer_pid,
652                       copy_mode,
653                       copy_range,
654                       queue_total,
655                       queue_maxlen,
656                       queue_dropped,
657                       queue_user_dropped);
658
659         read_unlock_bh(&queue_lock);
660         
661         *start = buffer + offset;
662         len -= offset;
663         if (len > length)
664                 len = length;
665         else if (len < 0)
666                 len = 0;
667         return len;
668 }
669 #endif /* CONFIG_PROC_FS */
670
671 static int
672 init_or_cleanup(int init)
673 {
674         int status = -ENOMEM;
675         struct proc_dir_entry *proc;
676         
677         if (!init)
678                 goto cleanup;
679
680         netlink_register_notifier(&ipq_nl_notifier);
681         ipqnl = netlink_kernel_create(NETLINK_FIREWALL, ipq_rcv_sk);
682         if (ipqnl == NULL) {
683                 printk(KERN_ERR "ip_queue: failed to create netlink socket\n");
684                 goto cleanup_netlink_notifier;
685         }
686
687         proc = proc_net_create(IPQ_PROC_FS_NAME, 0, ipq_get_info);
688         if (proc)
689                 proc->owner = THIS_MODULE;
690         else {
691                 printk(KERN_ERR "ip_queue: failed to create proc entry\n");
692                 goto cleanup_ipqnl;
693         }
694         
695         register_netdevice_notifier(&ipq_dev_notifier);
696         ipq_sysctl_header = register_sysctl_table(ipq_root_table, 0);
697         
698         status = nf_register_queue_handler(PF_INET, ipq_enqueue_packet, NULL);
699         if (status < 0) {
700                 printk(KERN_ERR "ip_queue: failed to register queue handler\n");
701                 goto cleanup_sysctl;
702         }
703         return status;
704
705 cleanup:
706         nf_unregister_queue_handler(PF_INET);
707         synchronize_net();
708         ipq_flush(NF_DROP);
709         
710 cleanup_sysctl:
711         unregister_sysctl_table(ipq_sysctl_header);
712         unregister_netdevice_notifier(&ipq_dev_notifier);
713         proc_net_remove(IPQ_PROC_FS_NAME);
714         
715 cleanup_ipqnl:
716         sock_release(ipqnl->sk_socket);
717         down(&ipqnl_sem);
718         up(&ipqnl_sem);
719         
720 cleanup_netlink_notifier:
721         netlink_unregister_notifier(&ipq_nl_notifier);
722         return status;
723 }
724
725 static int __init init(void)
726 {
727         
728         return init_or_cleanup(1);
729 }
730
731 static void __exit fini(void)
732 {
733         init_or_cleanup(0);
734 }
735
736 MODULE_DESCRIPTION("IPv4 packet queue handler");
737 MODULE_AUTHOR("James Morris <jmorris@intercode.com.au>");
738 MODULE_LICENSE("GPL");
739
740 module_init(init);
741 module_exit(fini);