2 * This is a module which is used for queueing IPv4 packets and
3 * communicating with userspace via netlink.
5 * (C) 2000-2002 James Morris <jmorris@intercode.com.au>
6 * (C) 2003-2005 Netfilter Core Team <coreteam@netfilter.org>
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.
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
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
21 * 2005-05-26: local_bh_{disable,enable} around nf_reinject (Harald Welte)
24 #include <linux/module.h>
25 #include <linux/skbuff.h>
26 #include <linux/init.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>
39 #include <net/route.h>
41 #define IPQ_QMAX_DEFAULT 1024
42 #define IPQ_PROC_FS_NAME "ip_queue"
43 #define NET_IPQ_QMAX 2088
44 #define NET_IPQ_QMAX_NAME "ip_queue_maxlen"
52 struct ipq_queue_entry
{
53 struct list_head list
;
56 struct ipq_rt_info rt_info
;
59 typedef int (*ipq_cmpfn
)(struct ipq_queue_entry
*, unsigned long);
61 static unsigned char copy_mode
= IPQ_COPY_NONE
;
62 static unsigned int queue_maxlen
= IPQ_QMAX_DEFAULT
;
63 static DEFINE_RWLOCK(queue_lock
);
65 static unsigned int copy_range
;
66 static unsigned int queue_total
;
67 static unsigned int queue_dropped
= 0;
68 static unsigned int queue_user_dropped
= 0;
69 static struct sock
*ipqnl
;
70 static LIST_HEAD(queue_list
);
71 static DECLARE_MUTEX(ipqnl_sem
);
74 ipq_issue_verdict(struct ipq_queue_entry
*entry
, int verdict
)
76 /* TCP input path (and probably other bits) assume to be called
77 * from softirq context, not from syscall, like ipq_issue_verdict is
78 * called. TCP input path deadlocks with locks taken from timer
79 * softirq, e.g. We therefore emulate this by local_bh_disable() */
82 nf_reinject(entry
->skb
, entry
->info
, verdict
);
89 __ipq_enqueue_entry(struct ipq_queue_entry
*entry
)
91 list_add(&entry
->list
, &queue_list
);
96 * Find and return a queued entry matched by cmpfn, or return the last
97 * entry if cmpfn is NULL.
99 static inline struct ipq_queue_entry
*
100 __ipq_find_entry(ipq_cmpfn cmpfn
, unsigned long data
)
104 list_for_each_prev(p
, &queue_list
) {
105 struct ipq_queue_entry
*entry
= (struct ipq_queue_entry
*)p
;
107 if (!cmpfn
|| cmpfn(entry
, data
))
114 __ipq_dequeue_entry(struct ipq_queue_entry
*entry
)
116 list_del(&entry
->list
);
120 static inline struct ipq_queue_entry
*
121 __ipq_find_dequeue_entry(ipq_cmpfn cmpfn
, unsigned long data
)
123 struct ipq_queue_entry
*entry
;
125 entry
= __ipq_find_entry(cmpfn
, data
);
129 __ipq_dequeue_entry(entry
);
135 __ipq_flush(int verdict
)
137 struct ipq_queue_entry
*entry
;
139 while ((entry
= __ipq_find_dequeue_entry(NULL
, 0)))
140 ipq_issue_verdict(entry
, verdict
);
144 __ipq_set_mode(unsigned char mode
, unsigned int range
)
155 case IPQ_COPY_PACKET
:
158 if (copy_range
> 0xFFFF)
173 net_disable_timestamp();
174 __ipq_set_mode(IPQ_COPY_NONE
, 0);
175 __ipq_flush(NF_DROP
);
178 static struct ipq_queue_entry
*
179 ipq_find_dequeue_entry(ipq_cmpfn cmpfn
, unsigned long data
)
181 struct ipq_queue_entry
*entry
;
183 write_lock_bh(&queue_lock
);
184 entry
= __ipq_find_dequeue_entry(cmpfn
, data
);
185 write_unlock_bh(&queue_lock
);
190 ipq_flush(int verdict
)
192 write_lock_bh(&queue_lock
);
193 __ipq_flush(verdict
);
194 write_unlock_bh(&queue_lock
);
197 static struct sk_buff
*
198 ipq_build_packet_message(struct ipq_queue_entry
*entry
, int *errp
)
200 unsigned char *old_tail
;
204 struct ipq_packet_msg
*pmsg
;
205 struct nlmsghdr
*nlh
;
207 read_lock_bh(&queue_lock
);
212 size
= NLMSG_SPACE(sizeof(*pmsg
));
216 case IPQ_COPY_PACKET
:
217 if (entry
->skb
->ip_summed
== CHECKSUM_HW
&&
218 (*errp
= skb_checksum_help(entry
->skb
,
219 entry
->info
->outdev
== NULL
))) {
220 read_unlock_bh(&queue_lock
);
223 if (copy_range
== 0 || copy_range
> entry
->skb
->len
)
224 data_len
= entry
->skb
->len
;
226 data_len
= copy_range
;
228 size
= NLMSG_SPACE(sizeof(*pmsg
) + data_len
);
233 read_unlock_bh(&queue_lock
);
237 read_unlock_bh(&queue_lock
);
239 skb
= alloc_skb(size
, GFP_ATOMIC
);
244 nlh
= NLMSG_PUT(skb
, 0, 0, IPQM_PACKET
, size
- sizeof(*nlh
));
245 pmsg
= NLMSG_DATA(nlh
);
246 memset(pmsg
, 0, sizeof(*pmsg
));
248 pmsg
->packet_id
= (unsigned long )entry
;
249 pmsg
->data_len
= data_len
;
250 pmsg
->timestamp_sec
= entry
->skb
->stamp
.tv_sec
;
251 pmsg
->timestamp_usec
= entry
->skb
->stamp
.tv_usec
;
252 pmsg
->mark
= entry
->skb
->nfmark
;
253 pmsg
->hook
= entry
->info
->hook
;
254 pmsg
->hw_protocol
= entry
->skb
->protocol
;
256 if (entry
->info
->indev
)
257 strcpy(pmsg
->indev_name
, entry
->info
->indev
->name
);
259 pmsg
->indev_name
[0] = '\0';
261 if (entry
->info
->outdev
)
262 strcpy(pmsg
->outdev_name
, entry
->info
->outdev
->name
);
264 pmsg
->outdev_name
[0] = '\0';
266 if (entry
->info
->indev
&& entry
->skb
->dev
) {
267 pmsg
->hw_type
= entry
->skb
->dev
->type
;
268 if (entry
->skb
->dev
->hard_header_parse
)
270 entry
->skb
->dev
->hard_header_parse(entry
->skb
,
275 if (skb_copy_bits(entry
->skb
, 0, pmsg
->payload
, data_len
))
278 nlh
->nlmsg_len
= skb
->tail
- old_tail
;
285 printk(KERN_ERR
"ip_queue: error creating packet message\n");
290 ipq_enqueue_packet(struct sk_buff
*skb
, struct nf_info
*info
, void *data
)
292 int status
= -EINVAL
;
293 struct sk_buff
*nskb
;
294 struct ipq_queue_entry
*entry
;
296 if (copy_mode
== IPQ_COPY_NONE
)
299 entry
= kmalloc(sizeof(*entry
), GFP_ATOMIC
);
301 printk(KERN_ERR
"ip_queue: OOM in ipq_enqueue_packet()\n");
308 if (entry
->info
->hook
== NF_IP_LOCAL_OUT
) {
309 struct iphdr
*iph
= skb
->nh
.iph
;
311 entry
->rt_info
.tos
= iph
->tos
;
312 entry
->rt_info
.daddr
= iph
->daddr
;
313 entry
->rt_info
.saddr
= iph
->saddr
;
316 nskb
= ipq_build_packet_message(entry
, &status
);
320 write_lock_bh(&queue_lock
);
323 goto err_out_free_nskb
;
325 if (queue_total
>= queue_maxlen
) {
329 printk (KERN_WARNING
"ip_queue: full at %d entries, "
330 "dropping packets(s). Dropped: %d\n", queue_total
,
332 goto err_out_free_nskb
;
335 /* netlink_unicast will either free the nskb or attach it to a socket */
336 status
= netlink_unicast(ipqnl
, nskb
, peer_pid
, MSG_DONTWAIT
);
338 queue_user_dropped
++;
342 __ipq_enqueue_entry(entry
);
344 write_unlock_bh(&queue_lock
);
351 write_unlock_bh(&queue_lock
);
359 ipq_mangle_ipv4(ipq_verdict_msg_t
*v
, struct ipq_queue_entry
*e
)
362 struct iphdr
*user_iph
= (struct iphdr
*)v
->payload
;
364 if (v
->data_len
< sizeof(*user_iph
))
366 diff
= v
->data_len
- e
->skb
->len
;
368 skb_trim(e
->skb
, v
->data_len
);
370 if (v
->data_len
> 0xFFFF)
372 if (diff
> skb_tailroom(e
->skb
)) {
373 struct sk_buff
*newskb
;
375 newskb
= skb_copy_expand(e
->skb
,
376 skb_headroom(e
->skb
),
379 if (newskb
== NULL
) {
380 printk(KERN_WARNING
"ip_queue: OOM "
381 "in mangle, dropping packet\n");
385 skb_set_owner_w(newskb
, e
->skb
->sk
);
389 skb_put(e
->skb
, diff
);
391 if (!skb_make_writable(&e
->skb
, v
->data_len
))
393 memcpy(e
->skb
->data
, v
->payload
, v
->data_len
);
394 e
->skb
->ip_summed
= CHECKSUM_NONE
;
397 * Extra routing may needed on local out, as the QUEUE target never
398 * returns control to the table.
400 if (e
->info
->hook
== NF_IP_LOCAL_OUT
) {
401 struct iphdr
*iph
= e
->skb
->nh
.iph
;
403 if (!(iph
->tos
== e
->rt_info
.tos
404 && iph
->daddr
== e
->rt_info
.daddr
405 && iph
->saddr
== e
->rt_info
.saddr
))
406 return ip_route_me_harder(&e
->skb
);
412 id_cmp(struct ipq_queue_entry
*e
, unsigned long id
)
414 return (id
== (unsigned long )e
);
418 ipq_set_verdict(struct ipq_verdict_msg
*vmsg
, unsigned int len
)
420 struct ipq_queue_entry
*entry
;
422 if (vmsg
->value
> NF_MAX_VERDICT
)
425 entry
= ipq_find_dequeue_entry(id_cmp
, vmsg
->id
);
429 int verdict
= vmsg
->value
;
431 if (vmsg
->data_len
&& vmsg
->data_len
== len
)
432 if (ipq_mangle_ipv4(vmsg
, entry
) < 0)
435 ipq_issue_verdict(entry
, verdict
);
441 ipq_set_mode(unsigned char mode
, unsigned int range
)
445 write_lock_bh(&queue_lock
);
446 status
= __ipq_set_mode(mode
, range
);
447 write_unlock_bh(&queue_lock
);
452 ipq_receive_peer(struct ipq_peer_msg
*pmsg
,
453 unsigned char type
, unsigned int len
)
457 if (len
< sizeof(*pmsg
))
462 status
= ipq_set_mode(pmsg
->msg
.mode
.value
,
463 pmsg
->msg
.mode
.range
);
467 if (pmsg
->msg
.verdict
.value
> NF_MAX_VERDICT
)
470 status
= ipq_set_verdict(&pmsg
->msg
.verdict
,
471 len
- sizeof(*pmsg
));
480 dev_cmp(struct ipq_queue_entry
*entry
, unsigned long ifindex
)
482 if (entry
->info
->indev
)
483 if (entry
->info
->indev
->ifindex
== ifindex
)
486 if (entry
->info
->outdev
)
487 if (entry
->info
->outdev
->ifindex
== ifindex
)
494 ipq_dev_drop(int ifindex
)
496 struct ipq_queue_entry
*entry
;
498 while ((entry
= ipq_find_dequeue_entry(dev_cmp
, ifindex
)) != NULL
)
499 ipq_issue_verdict(entry
, NF_DROP
);
502 #define RCV_SKB_FAIL(err) do { netlink_ack(skb, nlh, (err)); return; } while (0)
505 ipq_rcv_skb(struct sk_buff
*skb
)
507 int status
, type
, pid
, flags
, nlmsglen
, skblen
;
508 struct nlmsghdr
*nlh
;
511 if (skblen
< sizeof(*nlh
))
514 nlh
= (struct nlmsghdr
*)skb
->data
;
515 nlmsglen
= nlh
->nlmsg_len
;
516 if (nlmsglen
< sizeof(*nlh
) || skblen
< nlmsglen
)
519 pid
= nlh
->nlmsg_pid
;
520 flags
= nlh
->nlmsg_flags
;
522 if(pid
<= 0 || !(flags
& NLM_F_REQUEST
) || flags
& NLM_F_MULTI
)
523 RCV_SKB_FAIL(-EINVAL
);
525 if (flags
& MSG_TRUNC
)
526 RCV_SKB_FAIL(-ECOMM
);
528 type
= nlh
->nlmsg_type
;
529 if (type
< NLMSG_NOOP
|| type
>= IPQM_MAX
)
530 RCV_SKB_FAIL(-EINVAL
);
532 if (type
<= IPQM_BASE
)
535 if (security_netlink_recv(skb
))
536 RCV_SKB_FAIL(-EPERM
);
538 write_lock_bh(&queue_lock
);
541 if (peer_pid
!= pid
) {
542 write_unlock_bh(&queue_lock
);
543 RCV_SKB_FAIL(-EBUSY
);
546 net_enable_timestamp();
550 write_unlock_bh(&queue_lock
);
552 status
= ipq_receive_peer(NLMSG_DATA(nlh
), type
,
553 skblen
- NLMSG_LENGTH(0));
555 RCV_SKB_FAIL(status
);
557 if (flags
& NLM_F_ACK
)
558 netlink_ack(skb
, nlh
, 0);
563 ipq_rcv_sk(struct sock
*sk
, int len
)
570 for (qlen
= skb_queue_len(&sk
->sk_receive_queue
); qlen
; qlen
--) {
571 skb
= skb_dequeue(&sk
->sk_receive_queue
);
580 ipq_rcv_dev_event(struct notifier_block
*this,
581 unsigned long event
, void *ptr
)
583 struct net_device
*dev
= ptr
;
585 /* Drop any packets associated with the downed device */
586 if (event
== NETDEV_DOWN
)
587 ipq_dev_drop(dev
->ifindex
);
591 static struct notifier_block ipq_dev_notifier
= {
592 .notifier_call
= ipq_rcv_dev_event
,
596 ipq_rcv_nl_event(struct notifier_block
*this,
597 unsigned long event
, void *ptr
)
599 struct netlink_notify
*n
= ptr
;
601 if (event
== NETLINK_URELEASE
&&
602 n
->protocol
== NETLINK_FIREWALL
&& n
->pid
) {
603 write_lock_bh(&queue_lock
);
604 if (n
->pid
== peer_pid
)
606 write_unlock_bh(&queue_lock
);
611 static struct notifier_block ipq_nl_notifier
= {
612 .notifier_call
= ipq_rcv_nl_event
,
615 static struct ctl_table_header
*ipq_sysctl_header
;
617 static ctl_table ipq_table
[] = {
619 .ctl_name
= NET_IPQ_QMAX
,
620 .procname
= NET_IPQ_QMAX_NAME
,
621 .data
= &queue_maxlen
,
622 .maxlen
= sizeof(queue_maxlen
),
624 .proc_handler
= proc_dointvec
629 static ctl_table ipq_dir_table
[] = {
631 .ctl_name
= NET_IPV4
,
639 static ctl_table ipq_root_table
[] = {
644 .child
= ipq_dir_table
649 #ifdef CONFIG_PROC_FS
651 ipq_get_info(char *buffer
, char **start
, off_t offset
, int length
)
655 read_lock_bh(&queue_lock
);
657 len
= sprintf(buffer
,
661 "Queue length : %u\n"
662 "Queue max. length : %u\n"
663 "Queue dropped : %u\n"
664 "Netlink dropped : %u\n",
673 read_unlock_bh(&queue_lock
);
675 *start
= buffer
+ offset
;
683 #endif /* CONFIG_PROC_FS */
686 init_or_cleanup(int init
)
688 int status
= -ENOMEM
;
689 struct proc_dir_entry
*proc
;
694 netlink_register_notifier(&ipq_nl_notifier
);
695 ipqnl
= netlink_kernel_create(NETLINK_FIREWALL
, ipq_rcv_sk
);
697 printk(KERN_ERR
"ip_queue: failed to create netlink socket\n");
698 goto cleanup_netlink_notifier
;
701 proc
= proc_net_create(IPQ_PROC_FS_NAME
, 0, ipq_get_info
);
703 proc
->owner
= THIS_MODULE
;
705 printk(KERN_ERR
"ip_queue: failed to create proc entry\n");
709 register_netdevice_notifier(&ipq_dev_notifier
);
710 ipq_sysctl_header
= register_sysctl_table(ipq_root_table
, 0);
712 status
= nf_register_queue_handler(PF_INET
, ipq_enqueue_packet
, NULL
);
714 printk(KERN_ERR
"ip_queue: failed to register queue handler\n");
720 nf_unregister_queue_handler(PF_INET
);
725 unregister_sysctl_table(ipq_sysctl_header
);
726 unregister_netdevice_notifier(&ipq_dev_notifier
);
727 proc_net_remove(IPQ_PROC_FS_NAME
);
730 sock_release(ipqnl
->sk_socket
);
734 cleanup_netlink_notifier
:
735 netlink_unregister_notifier(&ipq_nl_notifier
);
739 static int __init
init(void)
742 return init_or_cleanup(1);
745 static void __exit
fini(void)
750 MODULE_DESCRIPTION("IPv4 packet queue handler");
751 MODULE_AUTHOR("James Morris <jmorris@intercode.com.au>");
752 MODULE_LICENSE("GPL");