[NET] NETNS: Omit net_device->nd_net without CONFIG_NET_NS.
[linux-3.10.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/mutex.h>
28 #include <net/net_namespace.h>
29 #include <net/sock.h>
30 #include <net/route.h>
31 #include <net/netfilter/nf_queue.h>
32 #include <net/ip.h>
33
34 #define IPQ_QMAX_DEFAULT 1024
35 #define IPQ_PROC_FS_NAME "ip_queue"
36 #define NET_IPQ_QMAX 2088
37 #define NET_IPQ_QMAX_NAME "ip_queue_maxlen"
38
39 typedef int (*ipq_cmpfn)(struct nf_queue_entry *, unsigned long);
40
41 static unsigned char copy_mode __read_mostly = IPQ_COPY_NONE;
42 static unsigned int queue_maxlen __read_mostly = IPQ_QMAX_DEFAULT;
43 static DEFINE_RWLOCK(queue_lock);
44 static int peer_pid __read_mostly;
45 static unsigned int copy_range __read_mostly;
46 static unsigned int queue_total;
47 static unsigned int queue_dropped = 0;
48 static unsigned int queue_user_dropped = 0;
49 static struct sock *ipqnl __read_mostly;
50 static LIST_HEAD(queue_list);
51 static DEFINE_MUTEX(ipqnl_mutex);
52
53 static inline void
54 __ipq_enqueue_entry(struct nf_queue_entry *entry)
55 {
56        list_add_tail(&entry->list, &queue_list);
57        queue_total++;
58 }
59
60 static inline int
61 __ipq_set_mode(unsigned char mode, unsigned int range)
62 {
63         int status = 0;
64
65         switch(mode) {
66         case IPQ_COPY_NONE:
67         case IPQ_COPY_META:
68                 copy_mode = mode;
69                 copy_range = 0;
70                 break;
71
72         case IPQ_COPY_PACKET:
73                 copy_mode = mode;
74                 copy_range = range;
75                 if (copy_range > 0xFFFF)
76                         copy_range = 0xFFFF;
77                 break;
78
79         default:
80                 status = -EINVAL;
81
82         }
83         return status;
84 }
85
86 static void __ipq_flush(ipq_cmpfn cmpfn, unsigned long data);
87
88 static inline void
89 __ipq_reset(void)
90 {
91         peer_pid = 0;
92         net_disable_timestamp();
93         __ipq_set_mode(IPQ_COPY_NONE, 0);
94         __ipq_flush(NULL, 0);
95 }
96
97 static struct nf_queue_entry *
98 ipq_find_dequeue_entry(unsigned long id)
99 {
100         struct nf_queue_entry *entry = NULL, *i;
101
102         write_lock_bh(&queue_lock);
103
104         list_for_each_entry(i, &queue_list, list) {
105                 if ((unsigned long)i == id) {
106                         entry = i;
107                         break;
108                 }
109         }
110
111         if (entry) {
112                 list_del(&entry->list);
113                 queue_total--;
114         }
115
116         write_unlock_bh(&queue_lock);
117         return entry;
118 }
119
120 static void
121 __ipq_flush(ipq_cmpfn cmpfn, unsigned long data)
122 {
123         struct nf_queue_entry *entry, *next;
124
125         list_for_each_entry_safe(entry, next, &queue_list, list) {
126                 if (!cmpfn || cmpfn(entry, data)) {
127                         list_del(&entry->list);
128                         queue_total--;
129                         nf_reinject(entry, NF_DROP);
130                 }
131         }
132 }
133
134 static void
135 ipq_flush(ipq_cmpfn cmpfn, unsigned long data)
136 {
137         write_lock_bh(&queue_lock);
138         __ipq_flush(cmpfn, data);
139         write_unlock_bh(&queue_lock);
140 }
141
142 static struct sk_buff *
143 ipq_build_packet_message(struct nf_queue_entry *entry, int *errp)
144 {
145         sk_buff_data_t old_tail;
146         size_t size = 0;
147         size_t data_len = 0;
148         struct sk_buff *skb;
149         struct ipq_packet_msg *pmsg;
150         struct nlmsghdr *nlh;
151         struct timeval tv;
152
153         read_lock_bh(&queue_lock);
154
155         switch (copy_mode) {
156         case IPQ_COPY_META:
157         case IPQ_COPY_NONE:
158                 size = NLMSG_SPACE(sizeof(*pmsg));
159                 data_len = 0;
160                 break;
161
162         case IPQ_COPY_PACKET:
163                 if ((entry->skb->ip_summed == CHECKSUM_PARTIAL ||
164                      entry->skb->ip_summed == CHECKSUM_COMPLETE) &&
165                     (*errp = skb_checksum_help(entry->skb))) {
166                         read_unlock_bh(&queue_lock);
167                         return NULL;
168                 }
169                 if (copy_range == 0 || copy_range > entry->skb->len)
170                         data_len = entry->skb->len;
171                 else
172                         data_len = copy_range;
173
174                 size = NLMSG_SPACE(sizeof(*pmsg) + data_len);
175                 break;
176
177         default:
178                 *errp = -EINVAL;
179                 read_unlock_bh(&queue_lock);
180                 return NULL;
181         }
182
183         read_unlock_bh(&queue_lock);
184
185         skb = alloc_skb(size, GFP_ATOMIC);
186         if (!skb)
187                 goto nlmsg_failure;
188
189         old_tail = skb->tail;
190         nlh = NLMSG_PUT(skb, 0, 0, IPQM_PACKET, size - sizeof(*nlh));
191         pmsg = NLMSG_DATA(nlh);
192         memset(pmsg, 0, sizeof(*pmsg));
193
194         pmsg->packet_id       = (unsigned long )entry;
195         pmsg->data_len        = data_len;
196         tv = ktime_to_timeval(entry->skb->tstamp);
197         pmsg->timestamp_sec   = tv.tv_sec;
198         pmsg->timestamp_usec  = tv.tv_usec;
199         pmsg->mark            = entry->skb->mark;
200         pmsg->hook            = entry->hook;
201         pmsg->hw_protocol     = entry->skb->protocol;
202
203         if (entry->indev)
204                 strcpy(pmsg->indev_name, entry->indev->name);
205         else
206                 pmsg->indev_name[0] = '\0';
207
208         if (entry->outdev)
209                 strcpy(pmsg->outdev_name, entry->outdev->name);
210         else
211                 pmsg->outdev_name[0] = '\0';
212
213         if (entry->indev && entry->skb->dev) {
214                 pmsg->hw_type = entry->skb->dev->type;
215                 pmsg->hw_addrlen = dev_parse_header(entry->skb,
216                                                     pmsg->hw_addr);
217         }
218
219         if (data_len)
220                 if (skb_copy_bits(entry->skb, 0, pmsg->payload, data_len))
221                         BUG();
222
223         nlh->nlmsg_len = skb->tail - old_tail;
224         return skb;
225
226 nlmsg_failure:
227         if (skb)
228                 kfree_skb(skb);
229         *errp = -EINVAL;
230         printk(KERN_ERR "ip_queue: error creating packet message\n");
231         return NULL;
232 }
233
234 static int
235 ipq_enqueue_packet(struct nf_queue_entry *entry, unsigned int queuenum)
236 {
237         int status = -EINVAL;
238         struct sk_buff *nskb;
239
240         if (copy_mode == IPQ_COPY_NONE)
241                 return -EAGAIN;
242
243         nskb = ipq_build_packet_message(entry, &status);
244         if (nskb == NULL)
245                 return status;
246
247         write_lock_bh(&queue_lock);
248
249         if (!peer_pid)
250                 goto err_out_free_nskb;
251
252         if (queue_total >= queue_maxlen) {
253                 queue_dropped++;
254                 status = -ENOSPC;
255                 if (net_ratelimit())
256                           printk (KERN_WARNING "ip_queue: full at %d entries, "
257                                   "dropping packets(s). Dropped: %d\n", queue_total,
258                                   queue_dropped);
259                 goto err_out_free_nskb;
260         }
261
262         /* netlink_unicast will either free the nskb or attach it to a socket */
263         status = netlink_unicast(ipqnl, nskb, peer_pid, MSG_DONTWAIT);
264         if (status < 0) {
265                 queue_user_dropped++;
266                 goto err_out_unlock;
267         }
268
269         __ipq_enqueue_entry(entry);
270
271         write_unlock_bh(&queue_lock);
272         return status;
273
274 err_out_free_nskb:
275         kfree_skb(nskb);
276
277 err_out_unlock:
278         write_unlock_bh(&queue_lock);
279         return status;
280 }
281
282 static int
283 ipq_mangle_ipv4(ipq_verdict_msg_t *v, struct nf_queue_entry *e)
284 {
285         int diff;
286         struct iphdr *user_iph = (struct iphdr *)v->payload;
287         struct sk_buff *nskb;
288
289         if (v->data_len < sizeof(*user_iph))
290                 return 0;
291         diff = v->data_len - e->skb->len;
292         if (diff < 0) {
293                 if (pskb_trim(e->skb, v->data_len))
294                         return -ENOMEM;
295         } else if (diff > 0) {
296                 if (v->data_len > 0xFFFF)
297                         return -EINVAL;
298                 if (diff > skb_tailroom(e->skb)) {
299                         nskb = skb_copy_expand(e->skb, 0,
300                                                diff - skb_tailroom(e->skb),
301                                                GFP_ATOMIC);
302                         if (!nskb) {
303                                 printk(KERN_WARNING "ip_queue: error "
304                                       "in mangle, dropping packet\n");
305                                 return -ENOMEM;
306                         }
307                         kfree_skb(e->skb);
308                         e->skb = nskb;
309                 }
310                 skb_put(e->skb, diff);
311         }
312         if (!skb_make_writable(e->skb, v->data_len))
313                 return -ENOMEM;
314         skb_copy_to_linear_data(e->skb, v->payload, v->data_len);
315         e->skb->ip_summed = CHECKSUM_NONE;
316
317         return 0;
318 }
319
320 static int
321 ipq_set_verdict(struct ipq_verdict_msg *vmsg, unsigned int len)
322 {
323         struct nf_queue_entry *entry;
324
325         if (vmsg->value > NF_MAX_VERDICT)
326                 return -EINVAL;
327
328         entry = ipq_find_dequeue_entry(vmsg->id);
329         if (entry == NULL)
330                 return -ENOENT;
331         else {
332                 int verdict = vmsg->value;
333
334                 if (vmsg->data_len && vmsg->data_len == len)
335                         if (ipq_mangle_ipv4(vmsg, entry) < 0)
336                                 verdict = NF_DROP;
337
338                 nf_reinject(entry, verdict);
339                 return 0;
340         }
341 }
342
343 static int
344 ipq_set_mode(unsigned char mode, unsigned int range)
345 {
346         int status;
347
348         write_lock_bh(&queue_lock);
349         status = __ipq_set_mode(mode, range);
350         write_unlock_bh(&queue_lock);
351         return status;
352 }
353
354 static int
355 ipq_receive_peer(struct ipq_peer_msg *pmsg,
356                  unsigned char type, unsigned int len)
357 {
358         int status = 0;
359
360         if (len < sizeof(*pmsg))
361                 return -EINVAL;
362
363         switch (type) {
364         case IPQM_MODE:
365                 status = ipq_set_mode(pmsg->msg.mode.value,
366                                       pmsg->msg.mode.range);
367                 break;
368
369         case IPQM_VERDICT:
370                 if (pmsg->msg.verdict.value > NF_MAX_VERDICT)
371                         status = -EINVAL;
372                 else
373                         status = ipq_set_verdict(&pmsg->msg.verdict,
374                                                  len - sizeof(*pmsg));
375                         break;
376         default:
377                 status = -EINVAL;
378         }
379         return status;
380 }
381
382 static int
383 dev_cmp(struct nf_queue_entry *entry, unsigned long ifindex)
384 {
385         if (entry->indev)
386                 if (entry->indev->ifindex == ifindex)
387                         return 1;
388         if (entry->outdev)
389                 if (entry->outdev->ifindex == ifindex)
390                         return 1;
391 #ifdef CONFIG_BRIDGE_NETFILTER
392         if (entry->skb->nf_bridge) {
393                 if (entry->skb->nf_bridge->physindev &&
394                     entry->skb->nf_bridge->physindev->ifindex == ifindex)
395                         return 1;
396                 if (entry->skb->nf_bridge->physoutdev &&
397                     entry->skb->nf_bridge->physoutdev->ifindex == ifindex)
398                         return 1;
399         }
400 #endif
401         return 0;
402 }
403
404 static void
405 ipq_dev_drop(int ifindex)
406 {
407         ipq_flush(dev_cmp, ifindex);
408 }
409
410 #define RCV_SKB_FAIL(err) do { netlink_ack(skb, nlh, (err)); return; } while (0)
411
412 static inline void
413 __ipq_rcv_skb(struct sk_buff *skb)
414 {
415         int status, type, pid, flags, nlmsglen, skblen;
416         struct nlmsghdr *nlh;
417
418         skblen = skb->len;
419         if (skblen < sizeof(*nlh))
420                 return;
421
422         nlh = nlmsg_hdr(skb);
423         nlmsglen = nlh->nlmsg_len;
424         if (nlmsglen < sizeof(*nlh) || skblen < nlmsglen)
425                 return;
426
427         pid = nlh->nlmsg_pid;
428         flags = nlh->nlmsg_flags;
429
430         if(pid <= 0 || !(flags & NLM_F_REQUEST) || flags & NLM_F_MULTI)
431                 RCV_SKB_FAIL(-EINVAL);
432
433         if (flags & MSG_TRUNC)
434                 RCV_SKB_FAIL(-ECOMM);
435
436         type = nlh->nlmsg_type;
437         if (type < NLMSG_NOOP || type >= IPQM_MAX)
438                 RCV_SKB_FAIL(-EINVAL);
439
440         if (type <= IPQM_BASE)
441                 return;
442
443         if (security_netlink_recv(skb, CAP_NET_ADMIN))
444                 RCV_SKB_FAIL(-EPERM);
445
446         write_lock_bh(&queue_lock);
447
448         if (peer_pid) {
449                 if (peer_pid != pid) {
450                         write_unlock_bh(&queue_lock);
451                         RCV_SKB_FAIL(-EBUSY);
452                 }
453         } else {
454                 net_enable_timestamp();
455                 peer_pid = pid;
456         }
457
458         write_unlock_bh(&queue_lock);
459
460         status = ipq_receive_peer(NLMSG_DATA(nlh), type,
461                                   nlmsglen - NLMSG_LENGTH(0));
462         if (status < 0)
463                 RCV_SKB_FAIL(status);
464
465         if (flags & NLM_F_ACK)
466                 netlink_ack(skb, nlh, 0);
467         return;
468 }
469
470 static void
471 ipq_rcv_skb(struct sk_buff *skb)
472 {
473         mutex_lock(&ipqnl_mutex);
474         __ipq_rcv_skb(skb);
475         mutex_unlock(&ipqnl_mutex);
476 }
477
478 static int
479 ipq_rcv_dev_event(struct notifier_block *this,
480                   unsigned long event, void *ptr)
481 {
482         struct net_device *dev = ptr;
483
484         if (dev_net(dev) != &init_net)
485                 return NOTIFY_DONE;
486
487         /* Drop any packets associated with the downed device */
488         if (event == NETDEV_DOWN)
489                 ipq_dev_drop(dev->ifindex);
490         return NOTIFY_DONE;
491 }
492
493 static struct notifier_block ipq_dev_notifier = {
494         .notifier_call  = ipq_rcv_dev_event,
495 };
496
497 static int
498 ipq_rcv_nl_event(struct notifier_block *this,
499                  unsigned long event, void *ptr)
500 {
501         struct netlink_notify *n = ptr;
502
503         if (event == NETLINK_URELEASE &&
504             n->protocol == NETLINK_FIREWALL && n->pid) {
505                 write_lock_bh(&queue_lock);
506                 if ((n->net == &init_net) && (n->pid == peer_pid))
507                         __ipq_reset();
508                 write_unlock_bh(&queue_lock);
509         }
510         return NOTIFY_DONE;
511 }
512
513 static struct notifier_block ipq_nl_notifier = {
514         .notifier_call  = ipq_rcv_nl_event,
515 };
516
517 #ifdef CONFIG_SYSCTL
518 static struct ctl_table_header *ipq_sysctl_header;
519
520 static ctl_table ipq_table[] = {
521         {
522                 .ctl_name       = NET_IPQ_QMAX,
523                 .procname       = NET_IPQ_QMAX_NAME,
524                 .data           = &queue_maxlen,
525                 .maxlen         = sizeof(queue_maxlen),
526                 .mode           = 0644,
527                 .proc_handler   = proc_dointvec
528         },
529         { .ctl_name = 0 }
530 };
531 #endif
532
533 #ifdef CONFIG_PROC_FS
534 static int ip_queue_show(struct seq_file *m, void *v)
535 {
536         read_lock_bh(&queue_lock);
537
538         seq_printf(m,
539                       "Peer PID          : %d\n"
540                       "Copy mode         : %hu\n"
541                       "Copy range        : %u\n"
542                       "Queue length      : %u\n"
543                       "Queue max. length : %u\n"
544                       "Queue dropped     : %u\n"
545                       "Netlink dropped   : %u\n",
546                       peer_pid,
547                       copy_mode,
548                       copy_range,
549                       queue_total,
550                       queue_maxlen,
551                       queue_dropped,
552                       queue_user_dropped);
553
554         read_unlock_bh(&queue_lock);
555         return 0;
556 }
557
558 static int ip_queue_open(struct inode *inode, struct file *file)
559 {
560         return single_open(file, ip_queue_show, NULL);
561 }
562
563 static const struct file_operations ip_queue_proc_fops = {
564         .open           = ip_queue_open,
565         .read           = seq_read,
566         .llseek         = seq_lseek,
567         .release        = single_release,
568         .owner          = THIS_MODULE,
569 };
570 #endif
571
572 static const struct nf_queue_handler nfqh = {
573         .name   = "ip_queue",
574         .outfn  = &ipq_enqueue_packet,
575 };
576
577 static int __init ip_queue_init(void)
578 {
579         int status = -ENOMEM;
580         struct proc_dir_entry *proc __maybe_unused;
581
582         netlink_register_notifier(&ipq_nl_notifier);
583         ipqnl = netlink_kernel_create(&init_net, NETLINK_FIREWALL, 0,
584                                       ipq_rcv_skb, NULL, THIS_MODULE);
585         if (ipqnl == NULL) {
586                 printk(KERN_ERR "ip_queue: failed to create netlink socket\n");
587                 goto cleanup_netlink_notifier;
588         }
589
590 #ifdef CONFIG_PROC_FS
591         proc = create_proc_entry(IPQ_PROC_FS_NAME, 0, init_net.proc_net);
592         if (proc) {
593                 proc->owner = THIS_MODULE;
594                 proc->proc_fops = &ip_queue_proc_fops;
595         } else {
596                 printk(KERN_ERR "ip_queue: failed to create proc entry\n");
597                 goto cleanup_ipqnl;
598         }
599 #endif
600         register_netdevice_notifier(&ipq_dev_notifier);
601 #ifdef CONFIG_SYSCTL
602         ipq_sysctl_header = register_sysctl_paths(net_ipv4_ctl_path, ipq_table);
603 #endif
604         status = nf_register_queue_handler(PF_INET, &nfqh);
605         if (status < 0) {
606                 printk(KERN_ERR "ip_queue: failed to register queue handler\n");
607                 goto cleanup_sysctl;
608         }
609         return status;
610
611 cleanup_sysctl:
612 #ifdef CONFIG_SYSCTL
613         unregister_sysctl_table(ipq_sysctl_header);
614 #endif
615         unregister_netdevice_notifier(&ipq_dev_notifier);
616         proc_net_remove(&init_net, IPQ_PROC_FS_NAME);
617 cleanup_ipqnl: __maybe_unused
618         netlink_kernel_release(ipqnl);
619         mutex_lock(&ipqnl_mutex);
620         mutex_unlock(&ipqnl_mutex);
621
622 cleanup_netlink_notifier:
623         netlink_unregister_notifier(&ipq_nl_notifier);
624         return status;
625 }
626
627 static void __exit ip_queue_fini(void)
628 {
629         nf_unregister_queue_handlers(&nfqh);
630         synchronize_net();
631         ipq_flush(NULL, 0);
632
633 #ifdef CONFIG_SYSCTL
634         unregister_sysctl_table(ipq_sysctl_header);
635 #endif
636         unregister_netdevice_notifier(&ipq_dev_notifier);
637         proc_net_remove(&init_net, IPQ_PROC_FS_NAME);
638
639         netlink_kernel_release(ipqnl);
640         mutex_lock(&ipqnl_mutex);
641         mutex_unlock(&ipqnl_mutex);
642
643         netlink_unregister_notifier(&ipq_nl_notifier);
644 }
645
646 MODULE_DESCRIPTION("IPv4 packet queue handler");
647 MODULE_AUTHOR("James Morris <jmorris@intercode.com.au>");
648 MODULE_LICENSE("GPL");
649
650 module_init(ip_queue_init);
651 module_exit(ip_queue_fini);