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