2 * NETLINK Kernel-user communication protocol.
4 * Authors: Alan Cox <alan@redhat.com>
5 * Alexey Kuznetsov <kuznet@ms2.inr.ac.ru>
7 * This program is free software; you can redistribute it and/or
8 * modify it under the terms of the GNU General Public License
9 * as published by the Free Software Foundation; either version
10 * 2 of the License, or (at your option) any later version.
12 * Tue Jun 26 14:36:48 MEST 2001 Herbert "herp" Rosmanith
13 * added netlink_proto_exit
14 * Tue Jan 22 18:32:44 BRST 2002 Arnaldo C. de Melo <acme@conectiva.com.br>
15 * use nlk_sk, as sk->protinfo is on a diet 8)
16 * Fri Jul 22 19:51:12 MEST 2005 Harald Welte <laforge@gnumonks.org>
17 * - inc module use count of module that owns
18 * the kernel socket in case userspace opens
19 * socket of same protocol
20 * - remove all module support, since netlink is
21 * mandatory if CONFIG_NET=y these days
24 #include <linux/config.h>
25 #include <linux/module.h>
27 #include <linux/capability.h>
28 #include <linux/kernel.h>
29 #include <linux/init.h>
30 #include <linux/signal.h>
31 #include <linux/sched.h>
32 #include <linux/errno.h>
33 #include <linux/string.h>
34 #include <linux/stat.h>
35 #include <linux/socket.h>
37 #include <linux/fcntl.h>
38 #include <linux/termios.h>
39 #include <linux/sockios.h>
40 #include <linux/net.h>
42 #include <linux/slab.h>
43 #include <asm/uaccess.h>
44 #include <linux/skbuff.h>
45 #include <linux/netdevice.h>
46 #include <linux/rtnetlink.h>
47 #include <linux/proc_fs.h>
48 #include <linux/seq_file.h>
49 #include <linux/smp_lock.h>
50 #include <linux/notifier.h>
51 #include <linux/security.h>
52 #include <linux/jhash.h>
53 #include <linux/jiffies.h>
54 #include <linux/random.h>
55 #include <linux/bitops.h>
57 #include <linux/types.h>
58 #include <linux/audit.h>
62 #include <net/netlink.h>
65 #define NLGRPSZ(x) (ALIGN(x, sizeof(unsigned long) * 8) / 8)
68 /* struct sock has to be the first member of netlink_sock */
76 unsigned long *groups
;
78 wait_queue_head_t wait
;
79 struct netlink_callback
*cb
;
81 void (*data_ready
)(struct sock
*sk
, int bytes
);
82 struct module
*module
;
85 #define NETLINK_KERNEL_SOCKET 0x1
86 #define NETLINK_RECV_PKTINFO 0x2
88 static inline struct netlink_sock
*nlk_sk(struct sock
*sk
)
90 return (struct netlink_sock
*)sk
;
94 struct hlist_head
*table
;
95 unsigned long rehash_time
;
100 unsigned int entries
;
101 unsigned int max_shift
;
106 struct netlink_table
{
107 struct nl_pid_hash hash
;
108 struct hlist_head mc_list
;
109 unsigned int nl_nonroot
;
111 struct module
*module
;
115 static struct netlink_table
*nl_table
;
117 static DECLARE_WAIT_QUEUE_HEAD(nl_table_wait
);
119 static int netlink_dump(struct sock
*sk
);
120 static void netlink_destroy_callback(struct netlink_callback
*cb
);
122 static DEFINE_RWLOCK(nl_table_lock
);
123 static atomic_t nl_table_users
= ATOMIC_INIT(0);
125 static struct notifier_block
*netlink_chain
;
127 static u32
netlink_group_mask(u32 group
)
129 return group
? 1 << (group
- 1) : 0;
132 static struct hlist_head
*nl_pid_hashfn(struct nl_pid_hash
*hash
, u32 pid
)
134 return &hash
->table
[jhash_1word(pid
, hash
->rnd
) & hash
->mask
];
137 static void netlink_sock_destruct(struct sock
*sk
)
139 skb_queue_purge(&sk
->sk_receive_queue
);
141 if (!sock_flag(sk
, SOCK_DEAD
)) {
142 printk("Freeing alive netlink socket %p\n", sk
);
145 BUG_TRAP(!atomic_read(&sk
->sk_rmem_alloc
));
146 BUG_TRAP(!atomic_read(&sk
->sk_wmem_alloc
));
147 BUG_TRAP(!nlk_sk(sk
)->cb
);
148 BUG_TRAP(!nlk_sk(sk
)->groups
);
151 /* This lock without WQ_FLAG_EXCLUSIVE is good on UP and it is _very_ bad on SMP.
152 * Look, when several writers sleep and reader wakes them up, all but one
153 * immediately hit write lock and grab all the cpus. Exclusive sleep solves
154 * this, _but_ remember, it adds useless work on UP machines.
157 static void netlink_table_grab(void)
159 write_lock_bh(&nl_table_lock
);
161 if (atomic_read(&nl_table_users
)) {
162 DECLARE_WAITQUEUE(wait
, current
);
164 add_wait_queue_exclusive(&nl_table_wait
, &wait
);
166 set_current_state(TASK_UNINTERRUPTIBLE
);
167 if (atomic_read(&nl_table_users
) == 0)
169 write_unlock_bh(&nl_table_lock
);
171 write_lock_bh(&nl_table_lock
);
174 __set_current_state(TASK_RUNNING
);
175 remove_wait_queue(&nl_table_wait
, &wait
);
179 static __inline__
void netlink_table_ungrab(void)
181 write_unlock_bh(&nl_table_lock
);
182 wake_up(&nl_table_wait
);
185 static __inline__
void
186 netlink_lock_table(void)
188 /* read_lock() synchronizes us to netlink_table_grab */
190 read_lock(&nl_table_lock
);
191 atomic_inc(&nl_table_users
);
192 read_unlock(&nl_table_lock
);
195 static __inline__
void
196 netlink_unlock_table(void)
198 if (atomic_dec_and_test(&nl_table_users
))
199 wake_up(&nl_table_wait
);
202 static __inline__
struct sock
*netlink_lookup(int protocol
, u32 pid
)
204 struct nl_pid_hash
*hash
= &nl_table
[protocol
].hash
;
205 struct hlist_head
*head
;
207 struct hlist_node
*node
;
209 read_lock(&nl_table_lock
);
210 head
= nl_pid_hashfn(hash
, pid
);
211 sk_for_each(sk
, node
, head
) {
212 if (nlk_sk(sk
)->pid
== pid
) {
219 read_unlock(&nl_table_lock
);
223 static inline struct hlist_head
*nl_pid_hash_alloc(size_t size
)
225 if (size
<= PAGE_SIZE
)
226 return kmalloc(size
, GFP_ATOMIC
);
228 return (struct hlist_head
*)
229 __get_free_pages(GFP_ATOMIC
, get_order(size
));
232 static inline void nl_pid_hash_free(struct hlist_head
*table
, size_t size
)
234 if (size
<= PAGE_SIZE
)
237 free_pages((unsigned long)table
, get_order(size
));
240 static int nl_pid_hash_rehash(struct nl_pid_hash
*hash
, int grow
)
242 unsigned int omask
, mask
, shift
;
244 struct hlist_head
*otable
, *table
;
247 omask
= mask
= hash
->mask
;
248 osize
= size
= (mask
+ 1) * sizeof(*table
);
252 if (++shift
> hash
->max_shift
)
258 table
= nl_pid_hash_alloc(size
);
262 memset(table
, 0, size
);
263 otable
= hash
->table
;
267 get_random_bytes(&hash
->rnd
, sizeof(hash
->rnd
));
269 for (i
= 0; i
<= omask
; i
++) {
271 struct hlist_node
*node
, *tmp
;
273 sk_for_each_safe(sk
, node
, tmp
, &otable
[i
])
274 __sk_add_node(sk
, nl_pid_hashfn(hash
, nlk_sk(sk
)->pid
));
277 nl_pid_hash_free(otable
, osize
);
278 hash
->rehash_time
= jiffies
+ 10 * 60 * HZ
;
282 static inline int nl_pid_hash_dilute(struct nl_pid_hash
*hash
, int len
)
284 int avg
= hash
->entries
>> hash
->shift
;
286 if (unlikely(avg
> 1) && nl_pid_hash_rehash(hash
, 1))
289 if (unlikely(len
> avg
) && time_after(jiffies
, hash
->rehash_time
)) {
290 nl_pid_hash_rehash(hash
, 0);
297 static const struct proto_ops netlink_ops
;
299 static int netlink_insert(struct sock
*sk
, u32 pid
)
301 struct nl_pid_hash
*hash
= &nl_table
[sk
->sk_protocol
].hash
;
302 struct hlist_head
*head
;
303 int err
= -EADDRINUSE
;
305 struct hlist_node
*node
;
308 netlink_table_grab();
309 head
= nl_pid_hashfn(hash
, pid
);
311 sk_for_each(osk
, node
, head
) {
312 if (nlk_sk(osk
)->pid
== pid
)
324 if (BITS_PER_LONG
> 32 && unlikely(hash
->entries
>= UINT_MAX
))
327 if (len
&& nl_pid_hash_dilute(hash
, len
))
328 head
= nl_pid_hashfn(hash
, pid
);
330 nlk_sk(sk
)->pid
= pid
;
331 sk_add_node(sk
, head
);
335 netlink_table_ungrab();
339 static void netlink_remove(struct sock
*sk
)
341 netlink_table_grab();
342 if (sk_del_node_init(sk
))
343 nl_table
[sk
->sk_protocol
].hash
.entries
--;
344 if (nlk_sk(sk
)->subscriptions
)
345 __sk_del_bind_node(sk
);
346 netlink_table_ungrab();
349 static struct proto netlink_proto
= {
351 .owner
= THIS_MODULE
,
352 .obj_size
= sizeof(struct netlink_sock
),
355 static int __netlink_create(struct socket
*sock
, int protocol
)
358 struct netlink_sock
*nlk
;
360 sock
->ops
= &netlink_ops
;
362 sk
= sk_alloc(PF_NETLINK
, GFP_KERNEL
, &netlink_proto
, 1);
366 sock_init_data(sock
, sk
);
369 spin_lock_init(&nlk
->cb_lock
);
370 init_waitqueue_head(&nlk
->wait
);
372 sk
->sk_destruct
= netlink_sock_destruct
;
373 sk
->sk_protocol
= protocol
;
377 static int netlink_create(struct socket
*sock
, int protocol
)
379 struct module
*module
= NULL
;
380 struct netlink_sock
*nlk
;
384 sock
->state
= SS_UNCONNECTED
;
386 if (sock
->type
!= SOCK_RAW
&& sock
->type
!= SOCK_DGRAM
)
387 return -ESOCKTNOSUPPORT
;
389 if (protocol
<0 || protocol
>= MAX_LINKS
)
390 return -EPROTONOSUPPORT
;
392 netlink_lock_table();
394 if (!nl_table
[protocol
].registered
) {
395 netlink_unlock_table();
396 request_module("net-pf-%d-proto-%d", PF_NETLINK
, protocol
);
397 netlink_lock_table();
400 if (nl_table
[protocol
].registered
&&
401 try_module_get(nl_table
[protocol
].module
))
402 module
= nl_table
[protocol
].module
;
403 groups
= nl_table
[protocol
].groups
;
404 netlink_unlock_table();
406 if ((err
= __netlink_create(sock
, protocol
)) < 0)
409 nlk
= nlk_sk(sock
->sk
);
410 nlk
->module
= module
;
419 static int netlink_release(struct socket
*sock
)
421 struct sock
*sk
= sock
->sk
;
422 struct netlink_sock
*nlk
;
430 spin_lock(&nlk
->cb_lock
);
433 nlk
->cb
->done(nlk
->cb
);
434 netlink_destroy_callback(nlk
->cb
);
437 spin_unlock(&nlk
->cb_lock
);
439 /* OK. Socket is unlinked, and, therefore,
440 no new packets will arrive */
444 wake_up_interruptible_all(&nlk
->wait
);
446 skb_queue_purge(&sk
->sk_write_queue
);
448 if (nlk
->pid
&& !nlk
->subscriptions
) {
449 struct netlink_notify n
= {
450 .protocol
= sk
->sk_protocol
,
453 notifier_call_chain(&netlink_chain
, NETLINK_URELEASE
, &n
);
457 module_put(nlk
->module
);
459 if (nlk
->flags
& NETLINK_KERNEL_SOCKET
) {
460 netlink_table_grab();
461 nl_table
[sk
->sk_protocol
].module
= NULL
;
462 nl_table
[sk
->sk_protocol
].registered
= 0;
463 netlink_table_ungrab();
473 static int netlink_autobind(struct socket
*sock
)
475 struct sock
*sk
= sock
->sk
;
476 struct nl_pid_hash
*hash
= &nl_table
[sk
->sk_protocol
].hash
;
477 struct hlist_head
*head
;
479 struct hlist_node
*node
;
480 s32 pid
= current
->tgid
;
482 static s32 rover
= -4097;
486 netlink_table_grab();
487 head
= nl_pid_hashfn(hash
, pid
);
488 sk_for_each(osk
, node
, head
) {
489 if (nlk_sk(osk
)->pid
== pid
) {
490 /* Bind collision, search negative pid values. */
494 netlink_table_ungrab();
498 netlink_table_ungrab();
500 err
= netlink_insert(sk
, pid
);
501 if (err
== -EADDRINUSE
)
504 /* If 2 threads race to autobind, that is fine. */
511 static inline int netlink_capable(struct socket
*sock
, unsigned int flag
)
513 return (nl_table
[sock
->sk
->sk_protocol
].nl_nonroot
& flag
) ||
514 capable(CAP_NET_ADMIN
);
518 netlink_update_subscriptions(struct sock
*sk
, unsigned int subscriptions
)
520 struct netlink_sock
*nlk
= nlk_sk(sk
);
522 if (nlk
->subscriptions
&& !subscriptions
)
523 __sk_del_bind_node(sk
);
524 else if (!nlk
->subscriptions
&& subscriptions
)
525 sk_add_bind_node(sk
, &nl_table
[sk
->sk_protocol
].mc_list
);
526 nlk
->subscriptions
= subscriptions
;
529 static int netlink_alloc_groups(struct sock
*sk
)
531 struct netlink_sock
*nlk
= nlk_sk(sk
);
535 netlink_lock_table();
536 groups
= nl_table
[sk
->sk_protocol
].groups
;
537 if (!nl_table
[sk
->sk_protocol
].registered
)
539 netlink_unlock_table();
544 nlk
->groups
= kmalloc(NLGRPSZ(groups
), GFP_KERNEL
);
545 if (nlk
->groups
== NULL
)
547 memset(nlk
->groups
, 0, NLGRPSZ(groups
));
548 nlk
->ngroups
= groups
;
552 static int netlink_bind(struct socket
*sock
, struct sockaddr
*addr
, int addr_len
)
554 struct sock
*sk
= sock
->sk
;
555 struct netlink_sock
*nlk
= nlk_sk(sk
);
556 struct sockaddr_nl
*nladdr
= (struct sockaddr_nl
*)addr
;
559 if (nladdr
->nl_family
!= AF_NETLINK
)
562 /* Only superuser is allowed to listen multicasts */
563 if (nladdr
->nl_groups
) {
564 if (!netlink_capable(sock
, NL_NONROOT_RECV
))
566 if (nlk
->groups
== NULL
) {
567 err
= netlink_alloc_groups(sk
);
574 if (nladdr
->nl_pid
!= nlk
->pid
)
577 err
= nladdr
->nl_pid
?
578 netlink_insert(sk
, nladdr
->nl_pid
) :
579 netlink_autobind(sock
);
584 if (!nladdr
->nl_groups
&& (nlk
->groups
== NULL
|| !(u32
)nlk
->groups
[0]))
587 netlink_table_grab();
588 netlink_update_subscriptions(sk
, nlk
->subscriptions
+
589 hweight32(nladdr
->nl_groups
) -
590 hweight32(nlk
->groups
[0]));
591 nlk
->groups
[0] = (nlk
->groups
[0] & ~0xffffffffUL
) | nladdr
->nl_groups
;
592 netlink_table_ungrab();
597 static int netlink_connect(struct socket
*sock
, struct sockaddr
*addr
,
601 struct sock
*sk
= sock
->sk
;
602 struct netlink_sock
*nlk
= nlk_sk(sk
);
603 struct sockaddr_nl
*nladdr
=(struct sockaddr_nl
*)addr
;
605 if (addr
->sa_family
== AF_UNSPEC
) {
606 sk
->sk_state
= NETLINK_UNCONNECTED
;
611 if (addr
->sa_family
!= AF_NETLINK
)
614 /* Only superuser is allowed to send multicasts */
615 if (nladdr
->nl_groups
&& !netlink_capable(sock
, NL_NONROOT_SEND
))
619 err
= netlink_autobind(sock
);
622 sk
->sk_state
= NETLINK_CONNECTED
;
623 nlk
->dst_pid
= nladdr
->nl_pid
;
624 nlk
->dst_group
= ffs(nladdr
->nl_groups
);
630 static int netlink_getname(struct socket
*sock
, struct sockaddr
*addr
, int *addr_len
, int peer
)
632 struct sock
*sk
= sock
->sk
;
633 struct netlink_sock
*nlk
= nlk_sk(sk
);
634 struct sockaddr_nl
*nladdr
=(struct sockaddr_nl
*)addr
;
636 nladdr
->nl_family
= AF_NETLINK
;
638 *addr_len
= sizeof(*nladdr
);
641 nladdr
->nl_pid
= nlk
->dst_pid
;
642 nladdr
->nl_groups
= netlink_group_mask(nlk
->dst_group
);
644 nladdr
->nl_pid
= nlk
->pid
;
645 nladdr
->nl_groups
= nlk
->groups
? nlk
->groups
[0] : 0;
650 static void netlink_overrun(struct sock
*sk
)
652 if (!test_and_set_bit(0, &nlk_sk(sk
)->state
)) {
653 sk
->sk_err
= ENOBUFS
;
654 sk
->sk_error_report(sk
);
658 static struct sock
*netlink_getsockbypid(struct sock
*ssk
, u32 pid
)
660 int protocol
= ssk
->sk_protocol
;
662 struct netlink_sock
*nlk
;
664 sock
= netlink_lookup(protocol
, pid
);
666 return ERR_PTR(-ECONNREFUSED
);
668 /* Don't bother queuing skb if kernel socket has no input function */
670 if ((nlk
->pid
== 0 && !nlk
->data_ready
) ||
671 (sock
->sk_state
== NETLINK_CONNECTED
&&
672 nlk
->dst_pid
!= nlk_sk(ssk
)->pid
)) {
674 return ERR_PTR(-ECONNREFUSED
);
679 struct sock
*netlink_getsockbyfilp(struct file
*filp
)
681 struct inode
*inode
= filp
->f_dentry
->d_inode
;
684 if (!S_ISSOCK(inode
->i_mode
))
685 return ERR_PTR(-ENOTSOCK
);
687 sock
= SOCKET_I(inode
)->sk
;
688 if (sock
->sk_family
!= AF_NETLINK
)
689 return ERR_PTR(-EINVAL
);
696 * Attach a skb to a netlink socket.
697 * The caller must hold a reference to the destination socket. On error, the
698 * reference is dropped. The skb is not send to the destination, just all
699 * all error checks are performed and memory in the queue is reserved.
701 * < 0: error. skb freed, reference to sock dropped.
703 * 1: repeat lookup - reference dropped while waiting for socket memory.
705 int netlink_attachskb(struct sock
*sk
, struct sk_buff
*skb
, int nonblock
, long timeo
)
707 struct netlink_sock
*nlk
;
711 if (atomic_read(&sk
->sk_rmem_alloc
) > sk
->sk_rcvbuf
||
712 test_bit(0, &nlk
->state
)) {
713 DECLARE_WAITQUEUE(wait
, current
);
722 __set_current_state(TASK_INTERRUPTIBLE
);
723 add_wait_queue(&nlk
->wait
, &wait
);
725 if ((atomic_read(&sk
->sk_rmem_alloc
) > sk
->sk_rcvbuf
||
726 test_bit(0, &nlk
->state
)) &&
727 !sock_flag(sk
, SOCK_DEAD
))
728 timeo
= schedule_timeout(timeo
);
730 __set_current_state(TASK_RUNNING
);
731 remove_wait_queue(&nlk
->wait
, &wait
);
734 if (signal_pending(current
)) {
736 return sock_intr_errno(timeo
);
740 skb_set_owner_r(skb
, sk
);
744 int netlink_sendskb(struct sock
*sk
, struct sk_buff
*skb
, int protocol
)
748 skb_queue_tail(&sk
->sk_receive_queue
, skb
);
749 sk
->sk_data_ready(sk
, len
);
754 void netlink_detachskb(struct sock
*sk
, struct sk_buff
*skb
)
760 static inline struct sk_buff
*netlink_trim(struct sk_buff
*skb
,
767 delta
= skb
->end
- skb
->tail
;
768 if (delta
* 2 < skb
->truesize
)
771 if (skb_shared(skb
)) {
772 struct sk_buff
*nskb
= skb_clone(skb
, allocation
);
779 if (!pskb_expand_head(skb
, 0, -delta
, allocation
))
780 skb
->truesize
-= delta
;
785 int netlink_unicast(struct sock
*ssk
, struct sk_buff
*skb
, u32 pid
, int nonblock
)
791 skb
= netlink_trim(skb
, gfp_any());
793 timeo
= sock_sndtimeo(ssk
, nonblock
);
795 sk
= netlink_getsockbypid(ssk
, pid
);
800 err
= netlink_attachskb(sk
, skb
, nonblock
, timeo
);
806 return netlink_sendskb(sk
, skb
, ssk
->sk_protocol
);
809 static __inline__
int netlink_broadcast_deliver(struct sock
*sk
, struct sk_buff
*skb
)
811 struct netlink_sock
*nlk
= nlk_sk(sk
);
813 if (atomic_read(&sk
->sk_rmem_alloc
) <= sk
->sk_rcvbuf
&&
814 !test_bit(0, &nlk
->state
)) {
815 skb_set_owner_r(skb
, sk
);
816 skb_queue_tail(&sk
->sk_receive_queue
, skb
);
817 sk
->sk_data_ready(sk
, skb
->len
);
818 return atomic_read(&sk
->sk_rmem_alloc
) > sk
->sk_rcvbuf
;
823 struct netlink_broadcast_data
{
824 struct sock
*exclude_sk
;
831 struct sk_buff
*skb
, *skb2
;
834 static inline int do_one_broadcast(struct sock
*sk
,
835 struct netlink_broadcast_data
*p
)
837 struct netlink_sock
*nlk
= nlk_sk(sk
);
840 if (p
->exclude_sk
== sk
)
843 if (nlk
->pid
== p
->pid
|| p
->group
- 1 >= nlk
->ngroups
||
844 !test_bit(p
->group
- 1, nlk
->groups
))
853 if (p
->skb2
== NULL
) {
854 if (skb_shared(p
->skb
)) {
855 p
->skb2
= skb_clone(p
->skb
, p
->allocation
);
857 p
->skb2
= skb_get(p
->skb
);
859 * skb ownership may have been set when
860 * delivered to a previous socket.
865 if (p
->skb2
== NULL
) {
867 /* Clone failed. Notify ALL listeners. */
869 } else if ((val
= netlink_broadcast_deliver(sk
, p
->skb2
)) < 0) {
882 int netlink_broadcast(struct sock
*ssk
, struct sk_buff
*skb
, u32 pid
,
883 u32 group
, gfp_t allocation
)
885 struct netlink_broadcast_data info
;
886 struct hlist_node
*node
;
889 skb
= netlink_trim(skb
, allocation
);
891 info
.exclude_sk
= ssk
;
897 info
.allocation
= allocation
;
901 /* While we sleep in clone, do not allow to change socket list */
903 netlink_lock_table();
905 sk_for_each_bound(sk
, node
, &nl_table
[ssk
->sk_protocol
].mc_list
)
906 do_one_broadcast(sk
, &info
);
910 netlink_unlock_table();
913 kfree_skb(info
.skb2
);
915 if (info
.delivered
) {
916 if (info
.congested
&& (allocation
& __GFP_WAIT
))
925 struct netlink_set_err_data
{
926 struct sock
*exclude_sk
;
932 static inline int do_one_set_err(struct sock
*sk
,
933 struct netlink_set_err_data
*p
)
935 struct netlink_sock
*nlk
= nlk_sk(sk
);
937 if (sk
== p
->exclude_sk
)
940 if (nlk
->pid
== p
->pid
|| p
->group
- 1 >= nlk
->ngroups
||
941 !test_bit(p
->group
- 1, nlk
->groups
))
944 sk
->sk_err
= p
->code
;
945 sk
->sk_error_report(sk
);
950 void netlink_set_err(struct sock
*ssk
, u32 pid
, u32 group
, int code
)
952 struct netlink_set_err_data info
;
953 struct hlist_node
*node
;
956 info
.exclude_sk
= ssk
;
961 read_lock(&nl_table_lock
);
963 sk_for_each_bound(sk
, node
, &nl_table
[ssk
->sk_protocol
].mc_list
)
964 do_one_set_err(sk
, &info
);
966 read_unlock(&nl_table_lock
);
969 static int netlink_setsockopt(struct socket
*sock
, int level
, int optname
,
970 char __user
*optval
, int optlen
)
972 struct sock
*sk
= sock
->sk
;
973 struct netlink_sock
*nlk
= nlk_sk(sk
);
976 if (level
!= SOL_NETLINK
)
979 if (optlen
>= sizeof(int) &&
980 get_user(val
, (int __user
*)optval
))
984 case NETLINK_PKTINFO
:
986 nlk
->flags
|= NETLINK_RECV_PKTINFO
;
988 nlk
->flags
&= ~NETLINK_RECV_PKTINFO
;
991 case NETLINK_ADD_MEMBERSHIP
:
992 case NETLINK_DROP_MEMBERSHIP
: {
993 unsigned int subscriptions
;
994 int old
, new = optname
== NETLINK_ADD_MEMBERSHIP
? 1 : 0;
996 if (!netlink_capable(sock
, NL_NONROOT_RECV
))
998 if (nlk
->groups
== NULL
) {
999 err
= netlink_alloc_groups(sk
);
1003 if (!val
|| val
- 1 >= nlk
->ngroups
)
1005 netlink_table_grab();
1006 old
= test_bit(val
- 1, nlk
->groups
);
1007 subscriptions
= nlk
->subscriptions
- old
+ new;
1009 __set_bit(val
- 1, nlk
->groups
);
1011 __clear_bit(val
- 1, nlk
->groups
);
1012 netlink_update_subscriptions(sk
, subscriptions
);
1013 netlink_table_ungrab();
1023 static int netlink_getsockopt(struct socket
*sock
, int level
, int optname
,
1024 char __user
*optval
, int __user
*optlen
)
1026 struct sock
*sk
= sock
->sk
;
1027 struct netlink_sock
*nlk
= nlk_sk(sk
);
1030 if (level
!= SOL_NETLINK
)
1031 return -ENOPROTOOPT
;
1033 if (get_user(len
, optlen
))
1039 case NETLINK_PKTINFO
:
1040 if (len
< sizeof(int))
1043 val
= nlk
->flags
& NETLINK_RECV_PKTINFO
? 1 : 0;
1044 put_user(len
, optlen
);
1045 put_user(val
, optval
);
1054 static void netlink_cmsg_recv_pktinfo(struct msghdr
*msg
, struct sk_buff
*skb
)
1056 struct nl_pktinfo info
;
1058 info
.group
= NETLINK_CB(skb
).dst_group
;
1059 put_cmsg(msg
, SOL_NETLINK
, NETLINK_PKTINFO
, sizeof(info
), &info
);
1062 static inline void netlink_rcv_wake(struct sock
*sk
)
1064 struct netlink_sock
*nlk
= nlk_sk(sk
);
1066 if (skb_queue_empty(&sk
->sk_receive_queue
))
1067 clear_bit(0, &nlk
->state
);
1068 if (!test_bit(0, &nlk
->state
))
1069 wake_up_interruptible(&nlk
->wait
);
1072 static int netlink_sendmsg(struct kiocb
*kiocb
, struct socket
*sock
,
1073 struct msghdr
*msg
, size_t len
)
1075 struct sock_iocb
*siocb
= kiocb_to_siocb(kiocb
);
1076 struct sock
*sk
= sock
->sk
;
1077 struct netlink_sock
*nlk
= nlk_sk(sk
);
1078 struct sockaddr_nl
*addr
=msg
->msg_name
;
1081 struct sk_buff
*skb
;
1083 struct scm_cookie scm
;
1085 if (msg
->msg_flags
&MSG_OOB
)
1088 if (NULL
== siocb
->scm
)
1090 err
= scm_send(sock
, msg
, siocb
->scm
);
1094 if (msg
->msg_namelen
) {
1095 if (addr
->nl_family
!= AF_NETLINK
)
1097 dst_pid
= addr
->nl_pid
;
1098 dst_group
= ffs(addr
->nl_groups
);
1099 if (dst_group
&& !netlink_capable(sock
, NL_NONROOT_SEND
))
1102 dst_pid
= nlk
->dst_pid
;
1103 dst_group
= nlk
->dst_group
;
1107 err
= netlink_autobind(sock
);
1113 if (len
> sk
->sk_sndbuf
- 32)
1116 skb
= alloc_skb(len
, GFP_KERNEL
);
1120 NETLINK_CB(skb
).pid
= nlk
->pid
;
1121 NETLINK_CB(skb
).dst_pid
= dst_pid
;
1122 NETLINK_CB(skb
).dst_group
= dst_group
;
1123 NETLINK_CB(skb
).loginuid
= audit_get_loginuid(current
->audit_context
);
1124 memcpy(NETLINK_CREDS(skb
), &siocb
->scm
->creds
, sizeof(struct ucred
));
1126 /* What can I do? Netlink is asynchronous, so that
1127 we will have to save current capabilities to
1128 check them, when this message will be delivered
1129 to corresponding kernel module. --ANK (980802)
1133 if (memcpy_fromiovec(skb_put(skb
,len
), msg
->msg_iov
, len
)) {
1138 err
= security_netlink_send(sk
, skb
);
1145 atomic_inc(&skb
->users
);
1146 netlink_broadcast(sk
, skb
, dst_pid
, dst_group
, GFP_KERNEL
);
1148 err
= netlink_unicast(sk
, skb
, dst_pid
, msg
->msg_flags
&MSG_DONTWAIT
);
1154 static int netlink_recvmsg(struct kiocb
*kiocb
, struct socket
*sock
,
1155 struct msghdr
*msg
, size_t len
,
1158 struct sock_iocb
*siocb
= kiocb_to_siocb(kiocb
);
1159 struct scm_cookie scm
;
1160 struct sock
*sk
= sock
->sk
;
1161 struct netlink_sock
*nlk
= nlk_sk(sk
);
1162 int noblock
= flags
&MSG_DONTWAIT
;
1164 struct sk_buff
*skb
;
1172 skb
= skb_recv_datagram(sk
,flags
,noblock
,&err
);
1176 msg
->msg_namelen
= 0;
1180 msg
->msg_flags
|= MSG_TRUNC
;
1184 skb
->h
.raw
= skb
->data
;
1185 err
= skb_copy_datagram_iovec(skb
, 0, msg
->msg_iov
, copied
);
1187 if (msg
->msg_name
) {
1188 struct sockaddr_nl
*addr
= (struct sockaddr_nl
*)msg
->msg_name
;
1189 addr
->nl_family
= AF_NETLINK
;
1191 addr
->nl_pid
= NETLINK_CB(skb
).pid
;
1192 addr
->nl_groups
= netlink_group_mask(NETLINK_CB(skb
).dst_group
);
1193 msg
->msg_namelen
= sizeof(*addr
);
1196 if (NULL
== siocb
->scm
) {
1197 memset(&scm
, 0, sizeof(scm
));
1200 siocb
->scm
->creds
= *NETLINK_CREDS(skb
);
1201 skb_free_datagram(sk
, skb
);
1203 if (nlk
->cb
&& atomic_read(&sk
->sk_rmem_alloc
) <= sk
->sk_rcvbuf
/ 2)
1206 scm_recv(sock
, msg
, siocb
->scm
, flags
);
1207 if (nlk
->flags
& NETLINK_RECV_PKTINFO
)
1208 netlink_cmsg_recv_pktinfo(msg
, skb
);
1211 netlink_rcv_wake(sk
);
1212 return err
? : copied
;
1215 static void netlink_data_ready(struct sock
*sk
, int len
)
1217 struct netlink_sock
*nlk
= nlk_sk(sk
);
1219 if (nlk
->data_ready
)
1220 nlk
->data_ready(sk
, len
);
1221 netlink_rcv_wake(sk
);
1225 * We export these functions to other modules. They provide a
1226 * complete set of kernel non-blocking support for message
1231 netlink_kernel_create(int unit
, unsigned int groups
,
1232 void (*input
)(struct sock
*sk
, int len
),
1233 struct module
*module
)
1235 struct socket
*sock
;
1237 struct netlink_sock
*nlk
;
1242 if (unit
<0 || unit
>=MAX_LINKS
)
1245 if (sock_create_lite(PF_NETLINK
, SOCK_DGRAM
, unit
, &sock
))
1248 if (__netlink_create(sock
, unit
) < 0)
1249 goto out_sock_release
;
1252 sk
->sk_data_ready
= netlink_data_ready
;
1254 nlk_sk(sk
)->data_ready
= input
;
1256 if (netlink_insert(sk
, 0))
1257 goto out_sock_release
;
1260 nlk
->flags
|= NETLINK_KERNEL_SOCKET
;
1262 netlink_table_grab();
1263 nl_table
[unit
].groups
= groups
< 32 ? 32 : groups
;
1264 nl_table
[unit
].module
= module
;
1265 nl_table
[unit
].registered
= 1;
1266 netlink_table_ungrab();
1275 void netlink_set_nonroot(int protocol
, unsigned int flags
)
1277 if ((unsigned int)protocol
< MAX_LINKS
)
1278 nl_table
[protocol
].nl_nonroot
= flags
;
1281 static void netlink_destroy_callback(struct netlink_callback
*cb
)
1289 * It looks a bit ugly.
1290 * It would be better to create kernel thread.
1293 static int netlink_dump(struct sock
*sk
)
1295 struct netlink_sock
*nlk
= nlk_sk(sk
);
1296 struct netlink_callback
*cb
;
1297 struct sk_buff
*skb
;
1298 struct nlmsghdr
*nlh
;
1301 skb
= sock_rmalloc(sk
, NLMSG_GOODSIZE
, 0, GFP_KERNEL
);
1305 spin_lock(&nlk
->cb_lock
);
1309 spin_unlock(&nlk
->cb_lock
);
1314 len
= cb
->dump(skb
, cb
);
1317 spin_unlock(&nlk
->cb_lock
);
1318 skb_queue_tail(&sk
->sk_receive_queue
, skb
);
1319 sk
->sk_data_ready(sk
, len
);
1323 nlh
= NLMSG_NEW_ANSWER(skb
, cb
, NLMSG_DONE
, sizeof(len
), NLM_F_MULTI
);
1324 memcpy(NLMSG_DATA(nlh
), &len
, sizeof(len
));
1325 skb_queue_tail(&sk
->sk_receive_queue
, skb
);
1326 sk
->sk_data_ready(sk
, skb
->len
);
1331 spin_unlock(&nlk
->cb_lock
);
1333 netlink_destroy_callback(cb
);
1340 int netlink_dump_start(struct sock
*ssk
, struct sk_buff
*skb
,
1341 struct nlmsghdr
*nlh
,
1342 int (*dump
)(struct sk_buff
*skb
, struct netlink_callback
*),
1343 int (*done
)(struct netlink_callback
*))
1345 struct netlink_callback
*cb
;
1347 struct netlink_sock
*nlk
;
1349 cb
= kmalloc(sizeof(*cb
), GFP_KERNEL
);
1353 memset(cb
, 0, sizeof(*cb
));
1357 atomic_inc(&skb
->users
);
1360 sk
= netlink_lookup(ssk
->sk_protocol
, NETLINK_CB(skb
).pid
);
1362 netlink_destroy_callback(cb
);
1363 return -ECONNREFUSED
;
1366 /* A dump is in progress... */
1367 spin_lock(&nlk
->cb_lock
);
1369 spin_unlock(&nlk
->cb_lock
);
1370 netlink_destroy_callback(cb
);
1375 spin_unlock(&nlk
->cb_lock
);
1382 void netlink_ack(struct sk_buff
*in_skb
, struct nlmsghdr
*nlh
, int err
)
1384 struct sk_buff
*skb
;
1385 struct nlmsghdr
*rep
;
1386 struct nlmsgerr
*errmsg
;
1390 size
= NLMSG_SPACE(sizeof(struct nlmsgerr
));
1392 size
= NLMSG_SPACE(4 + NLMSG_ALIGN(nlh
->nlmsg_len
));
1394 skb
= alloc_skb(size
, GFP_KERNEL
);
1398 sk
= netlink_lookup(in_skb
->sk
->sk_protocol
,
1399 NETLINK_CB(in_skb
).pid
);
1401 sk
->sk_err
= ENOBUFS
;
1402 sk
->sk_error_report(sk
);
1408 rep
= __nlmsg_put(skb
, NETLINK_CB(in_skb
).pid
, nlh
->nlmsg_seq
,
1409 NLMSG_ERROR
, sizeof(struct nlmsgerr
), 0);
1410 errmsg
= NLMSG_DATA(rep
);
1411 errmsg
->error
= err
;
1412 memcpy(&errmsg
->msg
, nlh
, err
? nlh
->nlmsg_len
: sizeof(struct nlmsghdr
));
1413 netlink_unicast(in_skb
->sk
, skb
, NETLINK_CB(in_skb
).pid
, MSG_DONTWAIT
);
1416 static int netlink_rcv_skb(struct sk_buff
*skb
, int (*cb
)(struct sk_buff
*,
1417 struct nlmsghdr
*, int *))
1419 unsigned int total_len
;
1420 struct nlmsghdr
*nlh
;
1423 while (skb
->len
>= nlmsg_total_size(0)) {
1424 nlh
= (struct nlmsghdr
*) skb
->data
;
1426 if (nlh
->nlmsg_len
< NLMSG_HDRLEN
|| skb
->len
< nlh
->nlmsg_len
)
1429 total_len
= min(NLMSG_ALIGN(nlh
->nlmsg_len
), skb
->len
);
1431 if (cb(skb
, nlh
, &err
) < 0) {
1432 /* Not an error, but we have to interrupt processing
1433 * here. Note: that in this case we do not pull
1434 * message from skb, it will be processed later.
1438 netlink_ack(skb
, nlh
, err
);
1439 } else if (nlh
->nlmsg_flags
& NLM_F_ACK
)
1440 netlink_ack(skb
, nlh
, 0);
1442 skb_pull(skb
, total_len
);
1449 * nelink_run_queue - Process netlink receive queue.
1450 * @sk: Netlink socket containing the queue
1451 * @qlen: Place to store queue length upon entry
1452 * @cb: Callback function invoked for each netlink message found
1454 * Processes as much as there was in the queue upon entry and invokes
1455 * a callback function for each netlink message found. The callback
1456 * function may refuse a message by returning a negative error code
1457 * but setting the error pointer to 0 in which case this function
1458 * returns with a qlen != 0.
1460 * qlen must be initialized to 0 before the initial entry, afterwards
1461 * the function may be called repeatedly until qlen reaches 0.
1463 void netlink_run_queue(struct sock
*sk
, unsigned int *qlen
,
1464 int (*cb
)(struct sk_buff
*, struct nlmsghdr
*, int *))
1466 struct sk_buff
*skb
;
1468 if (!*qlen
|| *qlen
> skb_queue_len(&sk
->sk_receive_queue
))
1469 *qlen
= skb_queue_len(&sk
->sk_receive_queue
);
1471 for (; *qlen
; (*qlen
)--) {
1472 skb
= skb_dequeue(&sk
->sk_receive_queue
);
1473 if (netlink_rcv_skb(skb
, cb
)) {
1475 skb_queue_head(&sk
->sk_receive_queue
, skb
);
1488 * netlink_queue_skip - Skip netlink message while processing queue.
1489 * @nlh: Netlink message to be skipped
1490 * @skb: Socket buffer containing the netlink messages.
1492 * Pulls the given netlink message off the socket buffer so the next
1493 * call to netlink_queue_run() will not reconsider the message.
1495 void netlink_queue_skip(struct nlmsghdr
*nlh
, struct sk_buff
*skb
)
1497 int msglen
= NLMSG_ALIGN(nlh
->nlmsg_len
);
1499 if (msglen
> skb
->len
)
1502 skb_pull(skb
, msglen
);
1505 #ifdef CONFIG_PROC_FS
1506 struct nl_seq_iter
{
1511 static struct sock
*netlink_seq_socket_idx(struct seq_file
*seq
, loff_t pos
)
1513 struct nl_seq_iter
*iter
= seq
->private;
1516 struct hlist_node
*node
;
1519 for (i
=0; i
<MAX_LINKS
; i
++) {
1520 struct nl_pid_hash
*hash
= &nl_table
[i
].hash
;
1522 for (j
= 0; j
<= hash
->mask
; j
++) {
1523 sk_for_each(s
, node
, &hash
->table
[j
]) {
1536 static void *netlink_seq_start(struct seq_file
*seq
, loff_t
*pos
)
1538 read_lock(&nl_table_lock
);
1539 return *pos
? netlink_seq_socket_idx(seq
, *pos
- 1) : SEQ_START_TOKEN
;
1542 static void *netlink_seq_next(struct seq_file
*seq
, void *v
, loff_t
*pos
)
1545 struct nl_seq_iter
*iter
;
1550 if (v
== SEQ_START_TOKEN
)
1551 return netlink_seq_socket_idx(seq
, 0);
1557 iter
= seq
->private;
1559 j
= iter
->hash_idx
+ 1;
1562 struct nl_pid_hash
*hash
= &nl_table
[i
].hash
;
1564 for (; j
<= hash
->mask
; j
++) {
1565 s
= sk_head(&hash
->table
[j
]);
1574 } while (++i
< MAX_LINKS
);
1579 static void netlink_seq_stop(struct seq_file
*seq
, void *v
)
1581 read_unlock(&nl_table_lock
);
1585 static int netlink_seq_show(struct seq_file
*seq
, void *v
)
1587 if (v
== SEQ_START_TOKEN
)
1589 "sk Eth Pid Groups "
1590 "Rmem Wmem Dump Locks\n");
1593 struct netlink_sock
*nlk
= nlk_sk(s
);
1595 seq_printf(seq
, "%p %-3d %-6d %08x %-8d %-8d %p %d\n",
1599 nlk
->groups
? (u32
)nlk
->groups
[0] : 0,
1600 atomic_read(&s
->sk_rmem_alloc
),
1601 atomic_read(&s
->sk_wmem_alloc
),
1603 atomic_read(&s
->sk_refcnt
)
1610 static struct seq_operations netlink_seq_ops
= {
1611 .start
= netlink_seq_start
,
1612 .next
= netlink_seq_next
,
1613 .stop
= netlink_seq_stop
,
1614 .show
= netlink_seq_show
,
1618 static int netlink_seq_open(struct inode
*inode
, struct file
*file
)
1620 struct seq_file
*seq
;
1621 struct nl_seq_iter
*iter
;
1624 iter
= kmalloc(sizeof(*iter
), GFP_KERNEL
);
1628 err
= seq_open(file
, &netlink_seq_ops
);
1634 memset(iter
, 0, sizeof(*iter
));
1635 seq
= file
->private_data
;
1636 seq
->private = iter
;
1640 static struct file_operations netlink_seq_fops
= {
1641 .owner
= THIS_MODULE
,
1642 .open
= netlink_seq_open
,
1644 .llseek
= seq_lseek
,
1645 .release
= seq_release_private
,
1650 int netlink_register_notifier(struct notifier_block
*nb
)
1652 return notifier_chain_register(&netlink_chain
, nb
);
1655 int netlink_unregister_notifier(struct notifier_block
*nb
)
1657 return notifier_chain_unregister(&netlink_chain
, nb
);
1660 static const struct proto_ops netlink_ops
= {
1661 .family
= PF_NETLINK
,
1662 .owner
= THIS_MODULE
,
1663 .release
= netlink_release
,
1664 .bind
= netlink_bind
,
1665 .connect
= netlink_connect
,
1666 .socketpair
= sock_no_socketpair
,
1667 .accept
= sock_no_accept
,
1668 .getname
= netlink_getname
,
1669 .poll
= datagram_poll
,
1670 .ioctl
= sock_no_ioctl
,
1671 .listen
= sock_no_listen
,
1672 .shutdown
= sock_no_shutdown
,
1673 .setsockopt
= netlink_setsockopt
,
1674 .getsockopt
= netlink_getsockopt
,
1675 .sendmsg
= netlink_sendmsg
,
1676 .recvmsg
= netlink_recvmsg
,
1677 .mmap
= sock_no_mmap
,
1678 .sendpage
= sock_no_sendpage
,
1681 static struct net_proto_family netlink_family_ops
= {
1682 .family
= PF_NETLINK
,
1683 .create
= netlink_create
,
1684 .owner
= THIS_MODULE
, /* for consistency 8) */
1687 extern void netlink_skb_parms_too_large(void);
1689 static int __init
netlink_proto_init(void)
1691 struct sk_buff
*dummy_skb
;
1695 int err
= proto_register(&netlink_proto
, 0);
1700 if (sizeof(struct netlink_skb_parms
) > sizeof(dummy_skb
->cb
))
1701 netlink_skb_parms_too_large();
1703 nl_table
= kmalloc(sizeof(*nl_table
) * MAX_LINKS
, GFP_KERNEL
);
1706 printk(KERN_CRIT
"netlink_init: Cannot allocate nl_table\n");
1710 memset(nl_table
, 0, sizeof(*nl_table
) * MAX_LINKS
);
1712 if (num_physpages
>= (128 * 1024))
1713 max
= num_physpages
>> (21 - PAGE_SHIFT
);
1715 max
= num_physpages
>> (23 - PAGE_SHIFT
);
1717 order
= get_bitmask_order(max
) - 1 + PAGE_SHIFT
;
1718 max
= (1UL << order
) / sizeof(struct hlist_head
);
1719 order
= get_bitmask_order(max
> UINT_MAX
? UINT_MAX
: max
) - 1;
1721 for (i
= 0; i
< MAX_LINKS
; i
++) {
1722 struct nl_pid_hash
*hash
= &nl_table
[i
].hash
;
1724 hash
->table
= nl_pid_hash_alloc(1 * sizeof(*hash
->table
));
1727 nl_pid_hash_free(nl_table
[i
].hash
.table
,
1728 1 * sizeof(*hash
->table
));
1732 memset(hash
->table
, 0, 1 * sizeof(*hash
->table
));
1733 hash
->max_shift
= order
;
1736 hash
->rehash_time
= jiffies
;
1739 sock_register(&netlink_family_ops
);
1740 #ifdef CONFIG_PROC_FS
1741 proc_net_fops_create("netlink", 0, &netlink_seq_fops
);
1743 /* The netlink device handler may be needed early. */
1749 core_initcall(netlink_proto_init
);
1751 EXPORT_SYMBOL(netlink_ack
);
1752 EXPORT_SYMBOL(netlink_run_queue
);
1753 EXPORT_SYMBOL(netlink_queue_skip
);
1754 EXPORT_SYMBOL(netlink_broadcast
);
1755 EXPORT_SYMBOL(netlink_dump_start
);
1756 EXPORT_SYMBOL(netlink_kernel_create
);
1757 EXPORT_SYMBOL(netlink_register_notifier
);
1758 EXPORT_SYMBOL(netlink_set_err
);
1759 EXPORT_SYMBOL(netlink_set_nonroot
);
1760 EXPORT_SYMBOL(netlink_unicast
);
1761 EXPORT_SYMBOL(netlink_unregister_notifier
);