net: dont call jump_label_dec from irq context
[linux-2.6.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 #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 <linux/slab.h>
29 #include <net/net_namespace.h>
30 #include <net/sock.h>
31 #include <net/ipv6.h>
32 #include <net/ip6_route.h>
33 #include <net/netfilter/nf_queue.h>
34 #include <linux/netfilter_ipv4/ip_queue.h>
35 #include <linux/netfilter_ipv4/ip_tables.h>
36 #include <linux/netfilter_ipv6/ip6_tables.h>
37
38 #define IPQ_QMAX_DEFAULT 1024
39 #define IPQ_PROC_FS_NAME "ip6_queue"
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_SPINLOCK(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                 if (range > 0xFFFF)
77                         range = 0xFFFF;
78                 copy_range = range;
79                 copy_mode = mode;
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         spin_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         spin_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         spin_lock_bh(&queue_lock);
141         __ipq_flush(cmpfn, data);
142         spin_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         switch (ACCESS_ONCE(copy_mode)) {
157         case IPQ_COPY_META:
158         case IPQ_COPY_NONE:
159                 size = NLMSG_SPACE(sizeof(*pmsg));
160                 break;
161
162         case IPQ_COPY_PACKET:
163                 if (entry->skb->ip_summed == CHECKSUM_PARTIAL &&
164                     (*errp = skb_checksum_help(entry->skb)))
165                         return NULL;
166
167                 data_len = ACCESS_ONCE(copy_range);
168                 if (data_len == 0 || data_len > entry->skb->len)
169                         data_len = entry->skb->len;
170
171                 size = NLMSG_SPACE(sizeof(*pmsg) + data_len);
172                 break;
173
174         default:
175                 *errp = -EINVAL;
176                 return NULL;
177         }
178
179         skb = alloc_skb(size, GFP_ATOMIC);
180         if (!skb)
181                 goto nlmsg_failure;
182
183         old_tail = skb->tail;
184         nlh = NLMSG_PUT(skb, 0, 0, IPQM_PACKET, size - sizeof(*nlh));
185         pmsg = NLMSG_DATA(nlh);
186         memset(pmsg, 0, sizeof(*pmsg));
187
188         pmsg->packet_id       = (unsigned long )entry;
189         pmsg->data_len        = data_len;
190         tv = ktime_to_timeval(entry->skb->tstamp);
191         pmsg->timestamp_sec   = tv.tv_sec;
192         pmsg->timestamp_usec  = tv.tv_usec;
193         pmsg->mark            = entry->skb->mark;
194         pmsg->hook            = entry->hook;
195         pmsg->hw_protocol     = entry->skb->protocol;
196
197         if (entry->indev)
198                 strcpy(pmsg->indev_name, entry->indev->name);
199         else
200                 pmsg->indev_name[0] = '\0';
201
202         if (entry->outdev)
203                 strcpy(pmsg->outdev_name, entry->outdev->name);
204         else
205                 pmsg->outdev_name[0] = '\0';
206
207         if (entry->indev && entry->skb->dev &&
208             entry->skb->mac_header != entry->skb->network_header) {
209                 pmsg->hw_type = entry->skb->dev->type;
210                 pmsg->hw_addrlen = dev_parse_header(entry->skb, pmsg->hw_addr);
211         }
212
213         if (data_len)
214                 if (skb_copy_bits(entry->skb, 0, pmsg->payload, data_len))
215                         BUG();
216
217         nlh->nlmsg_len = skb->tail - old_tail;
218         return skb;
219
220 nlmsg_failure:
221         kfree_skb(skb);
222         *errp = -EINVAL;
223         printk(KERN_ERR "ip6_queue: error creating packet message\n");
224         return NULL;
225 }
226
227 static int
228 ipq_enqueue_packet(struct nf_queue_entry *entry, unsigned int queuenum)
229 {
230         int status = -EINVAL;
231         struct sk_buff *nskb;
232
233         if (copy_mode == IPQ_COPY_NONE)
234                 return -EAGAIN;
235
236         nskb = ipq_build_packet_message(entry, &status);
237         if (nskb == NULL)
238                 return status;
239
240         spin_lock_bh(&queue_lock);
241
242         if (!peer_pid)
243                 goto err_out_free_nskb;
244
245         if (queue_total >= queue_maxlen) {
246                 queue_dropped++;
247                 status = -ENOSPC;
248                 if (net_ratelimit())
249                         printk (KERN_WARNING "ip6_queue: fill at %d entries, "
250                                 "dropping packet(s).  Dropped: %d\n", queue_total,
251                                 queue_dropped);
252                 goto err_out_free_nskb;
253         }
254
255         /* netlink_unicast will either free the nskb or attach it to a socket */
256         status = netlink_unicast(ipqnl, nskb, peer_pid, MSG_DONTWAIT);
257         if (status < 0) {
258                 queue_user_dropped++;
259                 goto err_out_unlock;
260         }
261
262         __ipq_enqueue_entry(entry);
263
264         spin_unlock_bh(&queue_lock);
265         return status;
266
267 err_out_free_nskb:
268         kfree_skb(nskb);
269
270 err_out_unlock:
271         spin_unlock_bh(&queue_lock);
272         return status;
273 }
274
275 static int
276 ipq_mangle_ipv6(ipq_verdict_msg_t *v, struct nf_queue_entry *e)
277 {
278         int diff;
279         struct ipv6hdr *user_iph = (struct ipv6hdr *)v->payload;
280         struct sk_buff *nskb;
281
282         if (v->data_len < sizeof(*user_iph))
283                 return 0;
284         diff = v->data_len - e->skb->len;
285         if (diff < 0) {
286                 if (pskb_trim(e->skb, v->data_len))
287                         return -ENOMEM;
288         } else if (diff > 0) {
289                 if (v->data_len > 0xFFFF)
290                         return -EINVAL;
291                 if (diff > skb_tailroom(e->skb)) {
292                         nskb = skb_copy_expand(e->skb, skb_headroom(e->skb),
293                                                diff, GFP_ATOMIC);
294                         if (!nskb) {
295                                 printk(KERN_WARNING "ip6_queue: OOM "
296                                       "in mangle, dropping packet\n");
297                                 return -ENOMEM;
298                         }
299                         kfree_skb(e->skb);
300                         e->skb = nskb;
301                 }
302                 skb_put(e->skb, diff);
303         }
304         if (!skb_make_writable(e->skb, v->data_len))
305                 return -ENOMEM;
306         skb_copy_to_linear_data(e->skb, v->payload, v->data_len);
307         e->skb->ip_summed = CHECKSUM_NONE;
308
309         return 0;
310 }
311
312 static int
313 ipq_set_verdict(struct ipq_verdict_msg *vmsg, unsigned int len)
314 {
315         struct nf_queue_entry *entry;
316
317         if (vmsg->value > NF_MAX_VERDICT || vmsg->value == NF_STOLEN)
318                 return -EINVAL;
319
320         entry = ipq_find_dequeue_entry(vmsg->id);
321         if (entry == NULL)
322                 return -ENOENT;
323         else {
324                 int verdict = vmsg->value;
325
326                 if (vmsg->data_len && vmsg->data_len == len)
327                         if (ipq_mangle_ipv6(vmsg, entry) < 0)
328                                 verdict = NF_DROP;
329
330                 nf_reinject(entry, verdict);
331                 return 0;
332         }
333 }
334
335 static int
336 ipq_set_mode(unsigned char mode, unsigned int range)
337 {
338         int status;
339
340         spin_lock_bh(&queue_lock);
341         status = __ipq_set_mode(mode, range);
342         spin_unlock_bh(&queue_lock);
343         return status;
344 }
345
346 static int
347 ipq_receive_peer(struct ipq_peer_msg *pmsg,
348                  unsigned char type, unsigned int len)
349 {
350         int status = 0;
351
352         if (len < sizeof(*pmsg))
353                 return -EINVAL;
354
355         switch (type) {
356         case IPQM_MODE:
357                 status = ipq_set_mode(pmsg->msg.mode.value,
358                                       pmsg->msg.mode.range);
359                 break;
360
361         case IPQM_VERDICT:
362                 status = ipq_set_verdict(&pmsg->msg.verdict,
363                                          len - sizeof(*pmsg));
364                 break;
365         default:
366                 status = -EINVAL;
367         }
368         return status;
369 }
370
371 static int
372 dev_cmp(struct nf_queue_entry *entry, unsigned long ifindex)
373 {
374         if (entry->indev)
375                 if (entry->indev->ifindex == ifindex)
376                         return 1;
377
378         if (entry->outdev)
379                 if (entry->outdev->ifindex == ifindex)
380                         return 1;
381 #ifdef CONFIG_BRIDGE_NETFILTER
382         if (entry->skb->nf_bridge) {
383                 if (entry->skb->nf_bridge->physindev &&
384                     entry->skb->nf_bridge->physindev->ifindex == ifindex)
385                         return 1;
386                 if (entry->skb->nf_bridge->physoutdev &&
387                     entry->skb->nf_bridge->physoutdev->ifindex == ifindex)
388                         return 1;
389         }
390 #endif
391         return 0;
392 }
393
394 static void
395 ipq_dev_drop(int ifindex)
396 {
397         ipq_flush(dev_cmp, ifindex);
398 }
399
400 #define RCV_SKB_FAIL(err) do { netlink_ack(skb, nlh, (err)); return; } while (0)
401
402 static inline void
403 __ipq_rcv_skb(struct sk_buff *skb)
404 {
405         int status, type, pid, flags;
406         unsigned int nlmsglen, skblen;
407         struct nlmsghdr *nlh;
408         bool enable_timestamp = false;
409
410         skblen = skb->len;
411         if (skblen < sizeof(*nlh))
412                 return;
413
414         nlh = nlmsg_hdr(skb);
415         nlmsglen = nlh->nlmsg_len;
416         if (nlmsglen < sizeof(*nlh) || skblen < nlmsglen)
417                 return;
418
419         pid = nlh->nlmsg_pid;
420         flags = nlh->nlmsg_flags;
421
422         if(pid <= 0 || !(flags & NLM_F_REQUEST) || flags & NLM_F_MULTI)
423                 RCV_SKB_FAIL(-EINVAL);
424
425         if (flags & MSG_TRUNC)
426                 RCV_SKB_FAIL(-ECOMM);
427
428         type = nlh->nlmsg_type;
429         if (type < NLMSG_NOOP || type >= IPQM_MAX)
430                 RCV_SKB_FAIL(-EINVAL);
431
432         if (type <= IPQM_BASE)
433                 return;
434
435         if (security_netlink_recv(skb, CAP_NET_ADMIN))
436                 RCV_SKB_FAIL(-EPERM);
437
438         spin_lock_bh(&queue_lock);
439
440         if (peer_pid) {
441                 if (peer_pid != pid) {
442                         spin_unlock_bh(&queue_lock);
443                         RCV_SKB_FAIL(-EBUSY);
444                 }
445         } else {
446                 enable_timestamp = true;
447                 peer_pid = pid;
448         }
449
450         spin_unlock_bh(&queue_lock);
451         if (enable_timestamp)
452                 net_enable_timestamp();
453
454         status = ipq_receive_peer(NLMSG_DATA(nlh), type,
455                                   nlmsglen - NLMSG_LENGTH(0));
456         if (status < 0)
457                 RCV_SKB_FAIL(status);
458
459         if (flags & NLM_F_ACK)
460                 netlink_ack(skb, nlh, 0);
461 }
462
463 static void
464 ipq_rcv_skb(struct sk_buff *skb)
465 {
466         mutex_lock(&ipqnl_mutex);
467         __ipq_rcv_skb(skb);
468         mutex_unlock(&ipqnl_mutex);
469 }
470
471 static int
472 ipq_rcv_dev_event(struct notifier_block *this,
473                   unsigned long event, void *ptr)
474 {
475         struct net_device *dev = ptr;
476
477         if (!net_eq(dev_net(dev), &init_net))
478                 return NOTIFY_DONE;
479
480         /* Drop any packets associated with the downed device */
481         if (event == NETDEV_DOWN)
482                 ipq_dev_drop(dev->ifindex);
483         return NOTIFY_DONE;
484 }
485
486 static struct notifier_block ipq_dev_notifier = {
487         .notifier_call  = ipq_rcv_dev_event,
488 };
489
490 static int
491 ipq_rcv_nl_event(struct notifier_block *this,
492                  unsigned long event, void *ptr)
493 {
494         struct netlink_notify *n = ptr;
495
496         if (event == NETLINK_URELEASE && n->protocol == NETLINK_IP6_FW) {
497                 spin_lock_bh(&queue_lock);
498                 if ((net_eq(n->net, &init_net)) && (n->pid == peer_pid))
499                         __ipq_reset();
500                 spin_unlock_bh(&queue_lock);
501         }
502         return NOTIFY_DONE;
503 }
504
505 static struct notifier_block ipq_nl_notifier = {
506         .notifier_call  = ipq_rcv_nl_event,
507 };
508
509 #ifdef CONFIG_SYSCTL
510 static struct ctl_table_header *ipq_sysctl_header;
511
512 static ctl_table ipq_table[] = {
513         {
514                 .procname       = NET_IPQ_QMAX_NAME,
515                 .data           = &queue_maxlen,
516                 .maxlen         = sizeof(queue_maxlen),
517                 .mode           = 0644,
518                 .proc_handler   = proc_dointvec
519         },
520         { }
521 };
522 #endif
523
524 #ifdef CONFIG_PROC_FS
525 static int ip6_queue_show(struct seq_file *m, void *v)
526 {
527         spin_lock_bh(&queue_lock);
528
529         seq_printf(m,
530                       "Peer PID          : %d\n"
531                       "Copy mode         : %hu\n"
532                       "Copy range        : %u\n"
533                       "Queue length      : %u\n"
534                       "Queue max. length : %u\n"
535                       "Queue dropped     : %u\n"
536                       "Netfilter dropped : %u\n",
537                       peer_pid,
538                       copy_mode,
539                       copy_range,
540                       queue_total,
541                       queue_maxlen,
542                       queue_dropped,
543                       queue_user_dropped);
544
545         spin_unlock_bh(&queue_lock);
546         return 0;
547 }
548
549 static int ip6_queue_open(struct inode *inode, struct file *file)
550 {
551         return single_open(file, ip6_queue_show, NULL);
552 }
553
554 static const struct file_operations ip6_queue_proc_fops = {
555         .open           = ip6_queue_open,
556         .read           = seq_read,
557         .llseek         = seq_lseek,
558         .release        = single_release,
559         .owner          = THIS_MODULE,
560 };
561 #endif
562
563 static const struct nf_queue_handler nfqh = {
564         .name   = "ip6_queue",
565         .outfn  = &ipq_enqueue_packet,
566 };
567
568 static int __init ip6_queue_init(void)
569 {
570         int status = -ENOMEM;
571         struct proc_dir_entry *proc __maybe_unused;
572
573         netlink_register_notifier(&ipq_nl_notifier);
574         ipqnl = netlink_kernel_create(&init_net, NETLINK_IP6_FW, 0,
575                                       ipq_rcv_skb, NULL, THIS_MODULE);
576         if (ipqnl == NULL) {
577                 printk(KERN_ERR "ip6_queue: failed to create netlink socket\n");
578                 goto cleanup_netlink_notifier;
579         }
580
581 #ifdef CONFIG_PROC_FS
582         proc = proc_create(IPQ_PROC_FS_NAME, 0, init_net.proc_net,
583                            &ip6_queue_proc_fops);
584         if (!proc) {
585                 printk(KERN_ERR "ip6_queue: failed to create proc entry\n");
586                 goto cleanup_ipqnl;
587         }
588 #endif
589         register_netdevice_notifier(&ipq_dev_notifier);
590 #ifdef CONFIG_SYSCTL
591         ipq_sysctl_header = register_sysctl_paths(net_ipv6_ctl_path, ipq_table);
592 #endif
593         status = nf_register_queue_handler(NFPROTO_IPV6, &nfqh);
594         if (status < 0) {
595                 printk(KERN_ERR "ip6_queue: failed to register queue handler\n");
596                 goto cleanup_sysctl;
597         }
598         return status;
599
600 cleanup_sysctl:
601 #ifdef CONFIG_SYSCTL
602         unregister_sysctl_table(ipq_sysctl_header);
603 #endif
604         unregister_netdevice_notifier(&ipq_dev_notifier);
605         proc_net_remove(&init_net, IPQ_PROC_FS_NAME);
606
607 cleanup_ipqnl: __maybe_unused
608         netlink_kernel_release(ipqnl);
609         mutex_lock(&ipqnl_mutex);
610         mutex_unlock(&ipqnl_mutex);
611
612 cleanup_netlink_notifier:
613         netlink_unregister_notifier(&ipq_nl_notifier);
614         return status;
615 }
616
617 static void __exit ip6_queue_fini(void)
618 {
619         nf_unregister_queue_handlers(&nfqh);
620
621         ipq_flush(NULL, 0);
622
623 #ifdef CONFIG_SYSCTL
624         unregister_sysctl_table(ipq_sysctl_header);
625 #endif
626         unregister_netdevice_notifier(&ipq_dev_notifier);
627         proc_net_remove(&init_net, IPQ_PROC_FS_NAME);
628
629         netlink_kernel_release(ipqnl);
630         mutex_lock(&ipqnl_mutex);
631         mutex_unlock(&ipqnl_mutex);
632
633         netlink_unregister_notifier(&ipq_nl_notifier);
634 }
635
636 MODULE_DESCRIPTION("IPv6 packet queue handler");
637 MODULE_LICENSE("GPL");
638 MODULE_ALIAS_NET_PF_PROTO(PF_NETLINK, NETLINK_IP6_FW);
639
640 module_init(ip6_queue_init);
641 module_exit(ip6_queue_fini);