e59aabd0eae4f9a63559672914db90d5a54c97b6
[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  * (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 #include <linux/module.h>
13 #include <linux/skbuff.h>
14 #include <linux/init.h>
15 #include <linux/ip.h>
16 #include <linux/notifier.h>
17 #include <linux/netdevice.h>
18 #include <linux/netfilter.h>
19 #include <linux/netfilter_ipv4/ip_queue.h>
20 #include <linux/netfilter_ipv4/ip_tables.h>
21 #include <linux/netlink.h>
22 #include <linux/spinlock.h>
23 #include <linux/sysctl.h>
24 #include <linux/proc_fs.h>
25 #include <linux/seq_file.h>
26 #include <linux/security.h>
27 #include <linux/net.h>
28 #include <linux/mutex.h>
29 #include <linux/slab.h>
30 #include <net/net_namespace.h>
31 #include <net/sock.h>
32 #include <net/route.h>
33 #include <net/netfilter/nf_queue.h>
34 #include <net/ip.h>
35
36 #define IPQ_QMAX_DEFAULT 1024
37 #define IPQ_PROC_FS_NAME "ip_queue"
38 #define NET_IPQ_QMAX 2088
39 #define NET_IPQ_QMAX_NAME "ip_queue_maxlen"
40
41 typedef int (*ipq_cmpfn)(struct nf_queue_entry *, unsigned long);
42
43 static unsigned char copy_mode __read_mostly = IPQ_COPY_NONE;
44 static unsigned int queue_maxlen __read_mostly = IPQ_QMAX_DEFAULT;
45 static DEFINE_SPINLOCK(queue_lock);
46 static int peer_pid __read_mostly;
47 static unsigned int copy_range __read_mostly;
48 static unsigned int queue_total;
49 static unsigned int queue_dropped = 0;
50 static unsigned int queue_user_dropped = 0;
51 static struct sock *ipqnl __read_mostly;
52 static LIST_HEAD(queue_list);
53 static DEFINE_MUTEX(ipqnl_mutex);
54
55 static inline void
56 __ipq_enqueue_entry(struct nf_queue_entry *entry)
57 {
58        list_add_tail(&entry->list, &queue_list);
59        queue_total++;
60 }
61
62 static inline int
63 __ipq_set_mode(unsigned char mode, unsigned int range)
64 {
65         int status = 0;
66
67         switch(mode) {
68         case IPQ_COPY_NONE:
69         case IPQ_COPY_META:
70                 copy_mode = mode;
71                 copy_range = 0;
72                 break;
73
74         case IPQ_COPY_PACKET:
75                 if (range > 0xFFFF)
76                         range = 0xFFFF;
77                 copy_range = range;
78                 copy_mode = mode;
79                 break;
80
81         default:
82                 status = -EINVAL;
83
84         }
85         return status;
86 }
87
88 static void __ipq_flush(ipq_cmpfn cmpfn, unsigned long data);
89
90 static inline void
91 __ipq_reset(void)
92 {
93         peer_pid = 0;
94         net_disable_timestamp();
95         __ipq_set_mode(IPQ_COPY_NONE, 0);
96         __ipq_flush(NULL, 0);
97 }
98
99 static struct nf_queue_entry *
100 ipq_find_dequeue_entry(unsigned long id)
101 {
102         struct nf_queue_entry *entry = NULL, *i;
103
104         spin_lock_bh(&queue_lock);
105
106         list_for_each_entry(i, &queue_list, list) {
107                 if ((unsigned long)i == id) {
108                         entry = i;
109                         break;
110                 }
111         }
112
113         if (entry) {
114                 list_del(&entry->list);
115                 queue_total--;
116         }
117
118         spin_unlock_bh(&queue_lock);
119         return entry;
120 }
121
122 static void
123 __ipq_flush(ipq_cmpfn cmpfn, unsigned long data)
124 {
125         struct nf_queue_entry *entry, *next;
126
127         list_for_each_entry_safe(entry, next, &queue_list, list) {
128                 if (!cmpfn || cmpfn(entry, data)) {
129                         list_del(&entry->list);
130                         queue_total--;
131                         nf_reinject(entry, NF_DROP);
132                 }
133         }
134 }
135
136 static void
137 ipq_flush(ipq_cmpfn cmpfn, unsigned long data)
138 {
139         spin_lock_bh(&queue_lock);
140         __ipq_flush(cmpfn, data);
141         spin_unlock_bh(&queue_lock);
142 }
143
144 static struct sk_buff *
145 ipq_build_packet_message(struct nf_queue_entry *entry, int *errp)
146 {
147         sk_buff_data_t old_tail;
148         size_t size = 0;
149         size_t data_len = 0;
150         struct sk_buff *skb;
151         struct ipq_packet_msg *pmsg;
152         struct nlmsghdr *nlh;
153         struct timeval tv;
154
155         switch (ACCESS_ONCE(copy_mode)) {
156         case IPQ_COPY_META:
157         case IPQ_COPY_NONE:
158                 size = NLMSG_SPACE(sizeof(*pmsg));
159                 break;
160
161         case IPQ_COPY_PACKET:
162                 if (entry->skb->ip_summed == CHECKSUM_PARTIAL &&
163                     (*errp = skb_checksum_help(entry->skb)))
164                         return NULL;
165
166                 data_len = ACCESS_ONCE(copy_range);
167                 if (data_len == 0 || data_len > entry->skb->len)
168                         data_len = entry->skb->len;
169
170                 size = NLMSG_SPACE(sizeof(*pmsg) + data_len);
171                 break;
172
173         default:
174                 *errp = -EINVAL;
175                 return NULL;
176         }
177
178         skb = alloc_skb(size, GFP_ATOMIC);
179         if (!skb)
180                 goto nlmsg_failure;
181
182         old_tail = skb->tail;
183         nlh = NLMSG_PUT(skb, 0, 0, IPQM_PACKET, size - sizeof(*nlh));
184         pmsg = NLMSG_DATA(nlh);
185         memset(pmsg, 0, sizeof(*pmsg));
186
187         pmsg->packet_id       = (unsigned long )entry;
188         pmsg->data_len        = data_len;
189         tv = ktime_to_timeval(entry->skb->tstamp);
190         pmsg->timestamp_sec   = tv.tv_sec;
191         pmsg->timestamp_usec  = tv.tv_usec;
192         pmsg->mark            = entry->skb->mark;
193         pmsg->hook            = entry->hook;
194         pmsg->hw_protocol     = entry->skb->protocol;
195
196         if (entry->indev)
197                 strcpy(pmsg->indev_name, entry->indev->name);
198         else
199                 pmsg->indev_name[0] = '\0';
200
201         if (entry->outdev)
202                 strcpy(pmsg->outdev_name, entry->outdev->name);
203         else
204                 pmsg->outdev_name[0] = '\0';
205
206         if (entry->indev && entry->skb->dev &&
207             entry->skb->mac_header != entry->skb->network_header) {
208                 pmsg->hw_type = entry->skb->dev->type;
209                 pmsg->hw_addrlen = dev_parse_header(entry->skb,
210                                                     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 "ip_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 "ip_queue: full at %d entries, "
250                                   "dropping packets(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_ipv4(ipq_verdict_msg_t *v, struct nf_queue_entry *e)
277 {
278         int diff;
279         struct iphdr *user_iph = (struct iphdr *)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 "ip_queue: error "
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_ipv4(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         if (entry->outdev)
378                 if (entry->outdev->ifindex == ifindex)
379                         return 1;
380 #ifdef CONFIG_BRIDGE_NETFILTER
381         if (entry->skb->nf_bridge) {
382                 if (entry->skb->nf_bridge->physindev &&
383                     entry->skb->nf_bridge->physindev->ifindex == ifindex)
384                         return 1;
385                 if (entry->skb->nf_bridge->physoutdev &&
386                     entry->skb->nf_bridge->physoutdev->ifindex == ifindex)
387                         return 1;
388         }
389 #endif
390         return 0;
391 }
392
393 static void
394 ipq_dev_drop(int ifindex)
395 {
396         ipq_flush(dev_cmp, ifindex);
397 }
398
399 #define RCV_SKB_FAIL(err) do { netlink_ack(skb, nlh, (err)); return; } while (0)
400
401 static inline void
402 __ipq_rcv_skb(struct sk_buff *skb)
403 {
404         int status, type, pid, flags;
405         unsigned int nlmsglen, skblen;
406         struct nlmsghdr *nlh;
407
408         skblen = skb->len;
409         if (skblen < sizeof(*nlh))
410                 return;
411
412         nlh = nlmsg_hdr(skb);
413         nlmsglen = nlh->nlmsg_len;
414         if (nlmsglen < sizeof(*nlh) || skblen < nlmsglen)
415                 return;
416
417         pid = nlh->nlmsg_pid;
418         flags = nlh->nlmsg_flags;
419
420         if(pid <= 0 || !(flags & NLM_F_REQUEST) || flags & NLM_F_MULTI)
421                 RCV_SKB_FAIL(-EINVAL);
422
423         if (flags & MSG_TRUNC)
424                 RCV_SKB_FAIL(-ECOMM);
425
426         type = nlh->nlmsg_type;
427         if (type < NLMSG_NOOP || type >= IPQM_MAX)
428                 RCV_SKB_FAIL(-EINVAL);
429
430         if (type <= IPQM_BASE)
431                 return;
432
433         if (security_netlink_recv(skb, CAP_NET_ADMIN))
434                 RCV_SKB_FAIL(-EPERM);
435
436         spin_lock_bh(&queue_lock);
437
438         if (peer_pid) {
439                 if (peer_pid != pid) {
440                         spin_unlock_bh(&queue_lock);
441                         RCV_SKB_FAIL(-EBUSY);
442                 }
443         } else {
444                 net_enable_timestamp();
445                 peer_pid = pid;
446         }
447
448         spin_unlock_bh(&queue_lock);
449
450         status = ipq_receive_peer(NLMSG_DATA(nlh), type,
451                                   nlmsglen - NLMSG_LENGTH(0));
452         if (status < 0)
453                 RCV_SKB_FAIL(status);
454
455         if (flags & NLM_F_ACK)
456                 netlink_ack(skb, nlh, 0);
457 }
458
459 static void
460 ipq_rcv_skb(struct sk_buff *skb)
461 {
462         mutex_lock(&ipqnl_mutex);
463         __ipq_rcv_skb(skb);
464         mutex_unlock(&ipqnl_mutex);
465 }
466
467 static int
468 ipq_rcv_dev_event(struct notifier_block *this,
469                   unsigned long event, void *ptr)
470 {
471         struct net_device *dev = ptr;
472
473         if (!net_eq(dev_net(dev), &init_net))
474                 return NOTIFY_DONE;
475
476         /* Drop any packets associated with the downed device */
477         if (event == NETDEV_DOWN)
478                 ipq_dev_drop(dev->ifindex);
479         return NOTIFY_DONE;
480 }
481
482 static struct notifier_block ipq_dev_notifier = {
483         .notifier_call  = ipq_rcv_dev_event,
484 };
485
486 static int
487 ipq_rcv_nl_event(struct notifier_block *this,
488                  unsigned long event, void *ptr)
489 {
490         struct netlink_notify *n = ptr;
491
492         if (event == NETLINK_URELEASE && n->protocol == NETLINK_FIREWALL) {
493                 spin_lock_bh(&queue_lock);
494                 if ((net_eq(n->net, &init_net)) && (n->pid == peer_pid))
495                         __ipq_reset();
496                 spin_unlock_bh(&queue_lock);
497         }
498         return NOTIFY_DONE;
499 }
500
501 static struct notifier_block ipq_nl_notifier = {
502         .notifier_call  = ipq_rcv_nl_event,
503 };
504
505 #ifdef CONFIG_SYSCTL
506 static struct ctl_table_header *ipq_sysctl_header;
507
508 static ctl_table ipq_table[] = {
509         {
510                 .procname       = NET_IPQ_QMAX_NAME,
511                 .data           = &queue_maxlen,
512                 .maxlen         = sizeof(queue_maxlen),
513                 .mode           = 0644,
514                 .proc_handler   = proc_dointvec
515         },
516         { }
517 };
518 #endif
519
520 #ifdef CONFIG_PROC_FS
521 static int ip_queue_show(struct seq_file *m, void *v)
522 {
523         spin_lock_bh(&queue_lock);
524
525         seq_printf(m,
526                       "Peer PID          : %d\n"
527                       "Copy mode         : %hu\n"
528                       "Copy range        : %u\n"
529                       "Queue length      : %u\n"
530                       "Queue max. length : %u\n"
531                       "Queue dropped     : %u\n"
532                       "Netlink dropped   : %u\n",
533                       peer_pid,
534                       copy_mode,
535                       copy_range,
536                       queue_total,
537                       queue_maxlen,
538                       queue_dropped,
539                       queue_user_dropped);
540
541         spin_unlock_bh(&queue_lock);
542         return 0;
543 }
544
545 static int ip_queue_open(struct inode *inode, struct file *file)
546 {
547         return single_open(file, ip_queue_show, NULL);
548 }
549
550 static const struct file_operations ip_queue_proc_fops = {
551         .open           = ip_queue_open,
552         .read           = seq_read,
553         .llseek         = seq_lseek,
554         .release        = single_release,
555         .owner          = THIS_MODULE,
556 };
557 #endif
558
559 static const struct nf_queue_handler nfqh = {
560         .name   = "ip_queue",
561         .outfn  = &ipq_enqueue_packet,
562 };
563
564 static int __init ip_queue_init(void)
565 {
566         int status = -ENOMEM;
567         struct proc_dir_entry *proc __maybe_unused;
568
569         netlink_register_notifier(&ipq_nl_notifier);
570         ipqnl = netlink_kernel_create(&init_net, NETLINK_FIREWALL, 0,
571                                       ipq_rcv_skb, NULL, THIS_MODULE);
572         if (ipqnl == NULL) {
573                 printk(KERN_ERR "ip_queue: failed to create netlink socket\n");
574                 goto cleanup_netlink_notifier;
575         }
576
577 #ifdef CONFIG_PROC_FS
578         proc = proc_create(IPQ_PROC_FS_NAME, 0, init_net.proc_net,
579                            &ip_queue_proc_fops);
580         if (!proc) {
581                 printk(KERN_ERR "ip_queue: failed to create proc entry\n");
582                 goto cleanup_ipqnl;
583         }
584 #endif
585         register_netdevice_notifier(&ipq_dev_notifier);
586 #ifdef CONFIG_SYSCTL
587         ipq_sysctl_header = register_sysctl_paths(net_ipv4_ctl_path, ipq_table);
588 #endif
589         status = nf_register_queue_handler(NFPROTO_IPV4, &nfqh);
590         if (status < 0) {
591                 printk(KERN_ERR "ip_queue: failed to register queue handler\n");
592                 goto cleanup_sysctl;
593         }
594         return status;
595
596 cleanup_sysctl:
597 #ifdef CONFIG_SYSCTL
598         unregister_sysctl_table(ipq_sysctl_header);
599 #endif
600         unregister_netdevice_notifier(&ipq_dev_notifier);
601         proc_net_remove(&init_net, IPQ_PROC_FS_NAME);
602 cleanup_ipqnl: __maybe_unused
603         netlink_kernel_release(ipqnl);
604         mutex_lock(&ipqnl_mutex);
605         mutex_unlock(&ipqnl_mutex);
606
607 cleanup_netlink_notifier:
608         netlink_unregister_notifier(&ipq_nl_notifier);
609         return status;
610 }
611
612 static void __exit ip_queue_fini(void)
613 {
614         nf_unregister_queue_handlers(&nfqh);
615
616         ipq_flush(NULL, 0);
617
618 #ifdef CONFIG_SYSCTL
619         unregister_sysctl_table(ipq_sysctl_header);
620 #endif
621         unregister_netdevice_notifier(&ipq_dev_notifier);
622         proc_net_remove(&init_net, IPQ_PROC_FS_NAME);
623
624         netlink_kernel_release(ipqnl);
625         mutex_lock(&ipqnl_mutex);
626         mutex_unlock(&ipqnl_mutex);
627
628         netlink_unregister_notifier(&ipq_nl_notifier);
629 }
630
631 MODULE_DESCRIPTION("IPv4 packet queue handler");
632 MODULE_AUTHOR("James Morris <jmorris@intercode.com.au>");
633 MODULE_LICENSE("GPL");
634 MODULE_ALIAS_NET_PF_PROTO(PF_NETLINK, NETLINK_FIREWALL);
635
636 module_init(ip_queue_init);
637 module_exit(ip_queue_fini);