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/kernel.h>
28 #include <linux/init.h>
29 #include <linux/signal.h>
30 #include <linux/sched.h>
31 #include <linux/errno.h>
32 #include <linux/string.h>
33 #include <linux/stat.h>
34 #include <linux/socket.h>
36 #include <linux/fcntl.h>
37 #include <linux/termios.h>
38 #include <linux/sockios.h>
39 #include <linux/net.h>
41 #include <linux/slab.h>
42 #include <asm/uaccess.h>
43 #include <linux/skbuff.h>
44 #include <linux/netdevice.h>
45 #include <linux/rtnetlink.h>
46 #include <linux/proc_fs.h>
47 #include <linux/seq_file.h>
48 #include <linux/smp_lock.h>
49 #include <linux/notifier.h>
50 #include <linux/security.h>
51 #include <linux/jhash.h>
52 #include <linux/jiffies.h>
53 #include <linux/random.h>
54 #include <linux/bitops.h>
56 #include <linux/types.h>
57 #include <linux/audit.h>
63 #define NLGRPSZ(x) (ALIGN(x, sizeof(unsigned long) * 8) / 8)
66 /* struct sock has to be the first member of netlink_sock */
74 unsigned long *groups
;
76 wait_queue_head_t wait
;
77 struct netlink_callback
*cb
;
79 void (*data_ready
)(struct sock
*sk
, int bytes
);
80 struct module
*module
;
83 #define NETLINK_KERNEL_SOCKET 0x1
84 #define NETLINK_RECV_PKTINFO 0x2
86 static inline struct netlink_sock
*nlk_sk(struct sock
*sk
)
88 return (struct netlink_sock
*)sk
;
92 struct hlist_head
*table
;
93 unsigned long rehash_time
;
99 unsigned int max_shift
;
104 struct netlink_table
{
105 struct nl_pid_hash hash
;
106 struct hlist_head mc_list
;
107 unsigned int nl_nonroot
;
109 struct module
*module
;
113 static struct netlink_table
*nl_table
;
115 static DECLARE_WAIT_QUEUE_HEAD(nl_table_wait
);
117 static int netlink_dump(struct sock
*sk
);
118 static void netlink_destroy_callback(struct netlink_callback
*cb
);
120 static DEFINE_RWLOCK(nl_table_lock
);
121 static atomic_t nl_table_users
= ATOMIC_INIT(0);
123 static struct notifier_block
*netlink_chain
;
125 static u32
netlink_group_mask(u32 group
)
127 return group
? 1 << (group
- 1) : 0;
130 static struct hlist_head
*nl_pid_hashfn(struct nl_pid_hash
*hash
, u32 pid
)
132 return &hash
->table
[jhash_1word(pid
, hash
->rnd
) & hash
->mask
];
135 static void netlink_sock_destruct(struct sock
*sk
)
137 skb_queue_purge(&sk
->sk_receive_queue
);
139 if (!sock_flag(sk
, SOCK_DEAD
)) {
140 printk("Freeing alive netlink socket %p\n", sk
);
143 BUG_TRAP(!atomic_read(&sk
->sk_rmem_alloc
));
144 BUG_TRAP(!atomic_read(&sk
->sk_wmem_alloc
));
145 BUG_TRAP(!nlk_sk(sk
)->cb
);
146 BUG_TRAP(!nlk_sk(sk
)->groups
);
149 /* This lock without WQ_FLAG_EXCLUSIVE is good on UP and it is _very_ bad on SMP.
150 * Look, when several writers sleep and reader wakes them up, all but one
151 * immediately hit write lock and grab all the cpus. Exclusive sleep solves
152 * this, _but_ remember, it adds useless work on UP machines.
155 static void netlink_table_grab(void)
157 write_lock_bh(&nl_table_lock
);
159 if (atomic_read(&nl_table_users
)) {
160 DECLARE_WAITQUEUE(wait
, current
);
162 add_wait_queue_exclusive(&nl_table_wait
, &wait
);
164 set_current_state(TASK_UNINTERRUPTIBLE
);
165 if (atomic_read(&nl_table_users
) == 0)
167 write_unlock_bh(&nl_table_lock
);
169 write_lock_bh(&nl_table_lock
);
172 __set_current_state(TASK_RUNNING
);
173 remove_wait_queue(&nl_table_wait
, &wait
);
177 static __inline__
void netlink_table_ungrab(void)
179 write_unlock_bh(&nl_table_lock
);
180 wake_up(&nl_table_wait
);
183 static __inline__
void
184 netlink_lock_table(void)
186 /* read_lock() synchronizes us to netlink_table_grab */
188 read_lock(&nl_table_lock
);
189 atomic_inc(&nl_table_users
);
190 read_unlock(&nl_table_lock
);
193 static __inline__
void
194 netlink_unlock_table(void)
196 if (atomic_dec_and_test(&nl_table_users
))
197 wake_up(&nl_table_wait
);
200 static __inline__
struct sock
*netlink_lookup(int protocol
, u32 pid
)
202 struct nl_pid_hash
*hash
= &nl_table
[protocol
].hash
;
203 struct hlist_head
*head
;
205 struct hlist_node
*node
;
207 read_lock(&nl_table_lock
);
208 head
= nl_pid_hashfn(hash
, pid
);
209 sk_for_each(sk
, node
, head
) {
210 if (nlk_sk(sk
)->pid
== pid
) {
217 read_unlock(&nl_table_lock
);
221 static inline struct hlist_head
*nl_pid_hash_alloc(size_t size
)
223 if (size
<= PAGE_SIZE
)
224 return kmalloc(size
, GFP_ATOMIC
);
226 return (struct hlist_head
*)
227 __get_free_pages(GFP_ATOMIC
, get_order(size
));
230 static inline void nl_pid_hash_free(struct hlist_head
*table
, size_t size
)
232 if (size
<= PAGE_SIZE
)
235 free_pages((unsigned long)table
, get_order(size
));
238 static int nl_pid_hash_rehash(struct nl_pid_hash
*hash
, int grow
)
240 unsigned int omask
, mask
, shift
;
242 struct hlist_head
*otable
, *table
;
245 omask
= mask
= hash
->mask
;
246 osize
= size
= (mask
+ 1) * sizeof(*table
);
250 if (++shift
> hash
->max_shift
)
256 table
= nl_pid_hash_alloc(size
);
260 memset(table
, 0, size
);
261 otable
= hash
->table
;
265 get_random_bytes(&hash
->rnd
, sizeof(hash
->rnd
));
267 for (i
= 0; i
<= omask
; i
++) {
269 struct hlist_node
*node
, *tmp
;
271 sk_for_each_safe(sk
, node
, tmp
, &otable
[i
])
272 __sk_add_node(sk
, nl_pid_hashfn(hash
, nlk_sk(sk
)->pid
));
275 nl_pid_hash_free(otable
, osize
);
276 hash
->rehash_time
= jiffies
+ 10 * 60 * HZ
;
280 static inline int nl_pid_hash_dilute(struct nl_pid_hash
*hash
, int len
)
282 int avg
= hash
->entries
>> hash
->shift
;
284 if (unlikely(avg
> 1) && nl_pid_hash_rehash(hash
, 1))
287 if (unlikely(len
> avg
) && time_after(jiffies
, hash
->rehash_time
)) {
288 nl_pid_hash_rehash(hash
, 0);
295 static struct proto_ops netlink_ops
;
297 static int netlink_insert(struct sock
*sk
, u32 pid
)
299 struct nl_pid_hash
*hash
= &nl_table
[sk
->sk_protocol
].hash
;
300 struct hlist_head
*head
;
301 int err
= -EADDRINUSE
;
303 struct hlist_node
*node
;
306 netlink_table_grab();
307 head
= nl_pid_hashfn(hash
, pid
);
309 sk_for_each(osk
, node
, head
) {
310 if (nlk_sk(osk
)->pid
== pid
)
322 if (BITS_PER_LONG
> 32 && unlikely(hash
->entries
>= UINT_MAX
))
325 if (len
&& nl_pid_hash_dilute(hash
, len
))
326 head
= nl_pid_hashfn(hash
, pid
);
328 nlk_sk(sk
)->pid
= pid
;
329 sk_add_node(sk
, head
);
333 netlink_table_ungrab();
337 static void netlink_remove(struct sock
*sk
)
339 netlink_table_grab();
340 if (sk_del_node_init(sk
))
341 nl_table
[sk
->sk_protocol
].hash
.entries
--;
342 if (nlk_sk(sk
)->subscriptions
)
343 __sk_del_bind_node(sk
);
344 netlink_table_ungrab();
347 static struct proto netlink_proto
= {
349 .owner
= THIS_MODULE
,
350 .obj_size
= sizeof(struct netlink_sock
),
353 static int __netlink_create(struct socket
*sock
, int protocol
)
356 struct netlink_sock
*nlk
;
358 sock
->ops
= &netlink_ops
;
360 sk
= sk_alloc(PF_NETLINK
, GFP_KERNEL
, &netlink_proto
, 1);
364 sock_init_data(sock
, sk
);
367 spin_lock_init(&nlk
->cb_lock
);
368 init_waitqueue_head(&nlk
->wait
);
370 sk
->sk_destruct
= netlink_sock_destruct
;
371 sk
->sk_protocol
= protocol
;
375 static int netlink_create(struct socket
*sock
, int protocol
)
377 struct module
*module
= NULL
;
378 struct netlink_sock
*nlk
;
382 sock
->state
= SS_UNCONNECTED
;
384 if (sock
->type
!= SOCK_RAW
&& sock
->type
!= SOCK_DGRAM
)
385 return -ESOCKTNOSUPPORT
;
387 if (protocol
<0 || protocol
>= MAX_LINKS
)
388 return -EPROTONOSUPPORT
;
390 netlink_lock_table();
392 if (!nl_table
[protocol
].registered
) {
393 netlink_unlock_table();
394 request_module("net-pf-%d-proto-%d", PF_NETLINK
, protocol
);
395 netlink_lock_table();
398 if (nl_table
[protocol
].registered
&&
399 try_module_get(nl_table
[protocol
].module
))
400 module
= nl_table
[protocol
].module
;
401 groups
= nl_table
[protocol
].groups
;
402 netlink_unlock_table();
404 if ((err
= __netlink_create(sock
, protocol
) < 0))
407 nlk
= nlk_sk(sock
->sk
);
408 nlk
->module
= module
;
417 static int netlink_release(struct socket
*sock
)
419 struct sock
*sk
= sock
->sk
;
420 struct netlink_sock
*nlk
;
428 spin_lock(&nlk
->cb_lock
);
430 nlk
->cb
->done(nlk
->cb
);
431 netlink_destroy_callback(nlk
->cb
);
434 spin_unlock(&nlk
->cb_lock
);
436 /* OK. Socket is unlinked, and, therefore,
437 no new packets will arrive */
441 wake_up_interruptible_all(&nlk
->wait
);
443 skb_queue_purge(&sk
->sk_write_queue
);
445 if (nlk
->pid
&& !nlk
->subscriptions
) {
446 struct netlink_notify n
= {
447 .protocol
= sk
->sk_protocol
,
450 notifier_call_chain(&netlink_chain
, NETLINK_URELEASE
, &n
);
454 module_put(nlk
->module
);
456 if (nlk
->flags
& NETLINK_KERNEL_SOCKET
) {
457 netlink_table_grab();
458 nl_table
[sk
->sk_protocol
].module
= NULL
;
459 nl_table
[sk
->sk_protocol
].registered
= 0;
460 netlink_table_ungrab();
470 static int netlink_autobind(struct socket
*sock
)
472 struct sock
*sk
= sock
->sk
;
473 struct nl_pid_hash
*hash
= &nl_table
[sk
->sk_protocol
].hash
;
474 struct hlist_head
*head
;
476 struct hlist_node
*node
;
477 s32 pid
= current
->pid
;
479 static s32 rover
= -4097;
483 netlink_table_grab();
484 head
= nl_pid_hashfn(hash
, pid
);
485 sk_for_each(osk
, node
, head
) {
486 if (nlk_sk(osk
)->pid
== pid
) {
487 /* Bind collision, search negative pid values. */
491 netlink_table_ungrab();
495 netlink_table_ungrab();
497 err
= netlink_insert(sk
, pid
);
498 if (err
== -EADDRINUSE
)
501 /* If 2 threads race to autobind, that is fine. */
508 static inline int netlink_capable(struct socket
*sock
, unsigned int flag
)
510 return (nl_table
[sock
->sk
->sk_protocol
].nl_nonroot
& flag
) ||
511 capable(CAP_NET_ADMIN
);
515 netlink_update_subscriptions(struct sock
*sk
, unsigned int subscriptions
)
517 struct netlink_sock
*nlk
= nlk_sk(sk
);
519 if (nlk
->subscriptions
&& !subscriptions
)
520 __sk_del_bind_node(sk
);
521 else if (!nlk
->subscriptions
&& subscriptions
)
522 sk_add_bind_node(sk
, &nl_table
[sk
->sk_protocol
].mc_list
);
523 nlk
->subscriptions
= subscriptions
;
526 static int netlink_alloc_groups(struct sock
*sk
)
528 struct netlink_sock
*nlk
= nlk_sk(sk
);
532 netlink_lock_table();
533 groups
= nl_table
[sk
->sk_protocol
].groups
;
534 if (!nl_table
[sk
->sk_protocol
].registered
)
536 netlink_unlock_table();
541 nlk
->groups
= kmalloc(NLGRPSZ(groups
), GFP_KERNEL
);
542 if (nlk
->groups
== NULL
)
544 memset(nlk
->groups
, 0, NLGRPSZ(groups
));
545 nlk
->ngroups
= groups
;
549 static int netlink_bind(struct socket
*sock
, struct sockaddr
*addr
, int addr_len
)
551 struct sock
*sk
= sock
->sk
;
552 struct netlink_sock
*nlk
= nlk_sk(sk
);
553 struct sockaddr_nl
*nladdr
= (struct sockaddr_nl
*)addr
;
556 if (nladdr
->nl_family
!= AF_NETLINK
)
559 /* Only superuser is allowed to listen multicasts */
560 if (nladdr
->nl_groups
) {
561 if (!netlink_capable(sock
, NL_NONROOT_RECV
))
563 if (nlk
->groups
== NULL
) {
564 err
= netlink_alloc_groups(sk
);
571 if (nladdr
->nl_pid
!= nlk
->pid
)
574 err
= nladdr
->nl_pid
?
575 netlink_insert(sk
, nladdr
->nl_pid
) :
576 netlink_autobind(sock
);
581 if (!nladdr
->nl_groups
&& (nlk
->groups
== NULL
|| !(u32
)nlk
->groups
[0]))
584 netlink_table_grab();
585 netlink_update_subscriptions(sk
, nlk
->subscriptions
+
586 hweight32(nladdr
->nl_groups
) -
587 hweight32(nlk
->groups
[0]));
588 nlk
->groups
[0] = (nlk
->groups
[0] & ~0xffffffffUL
) | nladdr
->nl_groups
;
589 netlink_table_ungrab();
594 static int netlink_connect(struct socket
*sock
, struct sockaddr
*addr
,
598 struct sock
*sk
= sock
->sk
;
599 struct netlink_sock
*nlk
= nlk_sk(sk
);
600 struct sockaddr_nl
*nladdr
=(struct sockaddr_nl
*)addr
;
602 if (addr
->sa_family
== AF_UNSPEC
) {
603 sk
->sk_state
= NETLINK_UNCONNECTED
;
608 if (addr
->sa_family
!= AF_NETLINK
)
611 /* Only superuser is allowed to send multicasts */
612 if (nladdr
->nl_groups
&& !netlink_capable(sock
, NL_NONROOT_SEND
))
616 err
= netlink_autobind(sock
);
619 sk
->sk_state
= NETLINK_CONNECTED
;
620 nlk
->dst_pid
= nladdr
->nl_pid
;
621 nlk
->dst_group
= ffs(nladdr
->nl_groups
);
627 static int netlink_getname(struct socket
*sock
, struct sockaddr
*addr
, int *addr_len
, int peer
)
629 struct sock
*sk
= sock
->sk
;
630 struct netlink_sock
*nlk
= nlk_sk(sk
);
631 struct sockaddr_nl
*nladdr
=(struct sockaddr_nl
*)addr
;
633 nladdr
->nl_family
= AF_NETLINK
;
635 *addr_len
= sizeof(*nladdr
);
638 nladdr
->nl_pid
= nlk
->dst_pid
;
639 nladdr
->nl_groups
= netlink_group_mask(nlk
->dst_group
);
641 nladdr
->nl_pid
= nlk
->pid
;
642 nladdr
->nl_groups
= nlk
->groups
? nlk
->groups
[0] : 0;
647 static void netlink_overrun(struct sock
*sk
)
649 if (!test_and_set_bit(0, &nlk_sk(sk
)->state
)) {
650 sk
->sk_err
= ENOBUFS
;
651 sk
->sk_error_report(sk
);
655 static struct sock
*netlink_getsockbypid(struct sock
*ssk
, u32 pid
)
657 int protocol
= ssk
->sk_protocol
;
659 struct netlink_sock
*nlk
;
661 sock
= netlink_lookup(protocol
, pid
);
663 return ERR_PTR(-ECONNREFUSED
);
665 /* Don't bother queuing skb if kernel socket has no input function */
667 if ((nlk
->pid
== 0 && !nlk
->data_ready
) ||
668 (sock
->sk_state
== NETLINK_CONNECTED
&&
669 nlk
->dst_pid
!= nlk_sk(ssk
)->pid
)) {
671 return ERR_PTR(-ECONNREFUSED
);
676 struct sock
*netlink_getsockbyfilp(struct file
*filp
)
678 struct inode
*inode
= filp
->f_dentry
->d_inode
;
681 if (!S_ISSOCK(inode
->i_mode
))
682 return ERR_PTR(-ENOTSOCK
);
684 sock
= SOCKET_I(inode
)->sk
;
685 if (sock
->sk_family
!= AF_NETLINK
)
686 return ERR_PTR(-EINVAL
);
693 * Attach a skb to a netlink socket.
694 * The caller must hold a reference to the destination socket. On error, the
695 * reference is dropped. The skb is not send to the destination, just all
696 * all error checks are performed and memory in the queue is reserved.
698 * < 0: error. skb freed, reference to sock dropped.
700 * 1: repeat lookup - reference dropped while waiting for socket memory.
702 int netlink_attachskb(struct sock
*sk
, struct sk_buff
*skb
, int nonblock
, long timeo
)
704 struct netlink_sock
*nlk
;
708 if (atomic_read(&sk
->sk_rmem_alloc
) > sk
->sk_rcvbuf
||
709 test_bit(0, &nlk
->state
)) {
710 DECLARE_WAITQUEUE(wait
, current
);
719 __set_current_state(TASK_INTERRUPTIBLE
);
720 add_wait_queue(&nlk
->wait
, &wait
);
722 if ((atomic_read(&sk
->sk_rmem_alloc
) > sk
->sk_rcvbuf
||
723 test_bit(0, &nlk
->state
)) &&
724 !sock_flag(sk
, SOCK_DEAD
))
725 timeo
= schedule_timeout(timeo
);
727 __set_current_state(TASK_RUNNING
);
728 remove_wait_queue(&nlk
->wait
, &wait
);
731 if (signal_pending(current
)) {
733 return sock_intr_errno(timeo
);
737 skb_set_owner_r(skb
, sk
);
741 int netlink_sendskb(struct sock
*sk
, struct sk_buff
*skb
, int protocol
)
743 struct netlink_sock
*nlk
;
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
;
830 unsigned int allocation
;
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
);
1330 spin_unlock(&nlk
->cb_lock
);
1332 netlink_destroy_callback(cb
);
1339 int netlink_dump_start(struct sock
*ssk
, struct sk_buff
*skb
,
1340 struct nlmsghdr
*nlh
,
1341 int (*dump
)(struct sk_buff
*skb
, struct netlink_callback
*),
1342 int (*done
)(struct netlink_callback
*))
1344 struct netlink_callback
*cb
;
1346 struct netlink_sock
*nlk
;
1348 cb
= kmalloc(sizeof(*cb
), GFP_KERNEL
);
1352 memset(cb
, 0, sizeof(*cb
));
1356 atomic_inc(&skb
->users
);
1359 sk
= netlink_lookup(ssk
->sk_protocol
, NETLINK_CB(skb
).pid
);
1361 netlink_destroy_callback(cb
);
1362 return -ECONNREFUSED
;
1365 /* A dump is in progress... */
1366 spin_lock(&nlk
->cb_lock
);
1368 spin_unlock(&nlk
->cb_lock
);
1369 netlink_destroy_callback(cb
);
1374 spin_unlock(&nlk
->cb_lock
);
1381 void netlink_ack(struct sk_buff
*in_skb
, struct nlmsghdr
*nlh
, int err
)
1383 struct sk_buff
*skb
;
1384 struct nlmsghdr
*rep
;
1385 struct nlmsgerr
*errmsg
;
1389 size
= NLMSG_SPACE(sizeof(struct nlmsgerr
));
1391 size
= NLMSG_SPACE(4 + NLMSG_ALIGN(nlh
->nlmsg_len
));
1393 skb
= alloc_skb(size
, GFP_KERNEL
);
1397 sk
= netlink_lookup(in_skb
->sk
->sk_protocol
,
1398 NETLINK_CB(in_skb
).pid
);
1400 sk
->sk_err
= ENOBUFS
;
1401 sk
->sk_error_report(sk
);
1407 rep
= __nlmsg_put(skb
, NETLINK_CB(in_skb
).pid
, nlh
->nlmsg_seq
,
1408 NLMSG_ERROR
, sizeof(struct nlmsgerr
), 0);
1409 errmsg
= NLMSG_DATA(rep
);
1410 errmsg
->error
= err
;
1411 memcpy(&errmsg
->msg
, nlh
, err
? nlh
->nlmsg_len
: sizeof(struct nlmsghdr
));
1412 netlink_unicast(in_skb
->sk
, skb
, NETLINK_CB(in_skb
).pid
, MSG_DONTWAIT
);
1416 #ifdef CONFIG_PROC_FS
1417 struct nl_seq_iter
{
1422 static struct sock
*netlink_seq_socket_idx(struct seq_file
*seq
, loff_t pos
)
1424 struct nl_seq_iter
*iter
= seq
->private;
1427 struct hlist_node
*node
;
1430 for (i
=0; i
<MAX_LINKS
; i
++) {
1431 struct nl_pid_hash
*hash
= &nl_table
[i
].hash
;
1433 for (j
= 0; j
<= hash
->mask
; j
++) {
1434 sk_for_each(s
, node
, &hash
->table
[j
]) {
1447 static void *netlink_seq_start(struct seq_file
*seq
, loff_t
*pos
)
1449 read_lock(&nl_table_lock
);
1450 return *pos
? netlink_seq_socket_idx(seq
, *pos
- 1) : SEQ_START_TOKEN
;
1453 static void *netlink_seq_next(struct seq_file
*seq
, void *v
, loff_t
*pos
)
1456 struct nl_seq_iter
*iter
;
1461 if (v
== SEQ_START_TOKEN
)
1462 return netlink_seq_socket_idx(seq
, 0);
1468 iter
= seq
->private;
1470 j
= iter
->hash_idx
+ 1;
1473 struct nl_pid_hash
*hash
= &nl_table
[i
].hash
;
1475 for (; j
<= hash
->mask
; j
++) {
1476 s
= sk_head(&hash
->table
[j
]);
1485 } while (++i
< MAX_LINKS
);
1490 static void netlink_seq_stop(struct seq_file
*seq
, void *v
)
1492 read_unlock(&nl_table_lock
);
1496 static int netlink_seq_show(struct seq_file
*seq
, void *v
)
1498 if (v
== SEQ_START_TOKEN
)
1500 "sk Eth Pid Groups "
1501 "Rmem Wmem Dump Locks\n");
1504 struct netlink_sock
*nlk
= nlk_sk(s
);
1506 seq_printf(seq
, "%p %-3d %-6d %08x %-8d %-8d %p %d\n",
1510 nlk
->groups
? (u32
)nlk
->groups
[0] : 0,
1511 atomic_read(&s
->sk_rmem_alloc
),
1512 atomic_read(&s
->sk_wmem_alloc
),
1514 atomic_read(&s
->sk_refcnt
)
1521 static struct seq_operations netlink_seq_ops
= {
1522 .start
= netlink_seq_start
,
1523 .next
= netlink_seq_next
,
1524 .stop
= netlink_seq_stop
,
1525 .show
= netlink_seq_show
,
1529 static int netlink_seq_open(struct inode
*inode
, struct file
*file
)
1531 struct seq_file
*seq
;
1532 struct nl_seq_iter
*iter
;
1535 iter
= kmalloc(sizeof(*iter
), GFP_KERNEL
);
1539 err
= seq_open(file
, &netlink_seq_ops
);
1545 memset(iter
, 0, sizeof(*iter
));
1546 seq
= file
->private_data
;
1547 seq
->private = iter
;
1551 static struct file_operations netlink_seq_fops
= {
1552 .owner
= THIS_MODULE
,
1553 .open
= netlink_seq_open
,
1555 .llseek
= seq_lseek
,
1556 .release
= seq_release_private
,
1561 int netlink_register_notifier(struct notifier_block
*nb
)
1563 return notifier_chain_register(&netlink_chain
, nb
);
1566 int netlink_unregister_notifier(struct notifier_block
*nb
)
1568 return notifier_chain_unregister(&netlink_chain
, nb
);
1571 static struct proto_ops netlink_ops
= {
1572 .family
= PF_NETLINK
,
1573 .owner
= THIS_MODULE
,
1574 .release
= netlink_release
,
1575 .bind
= netlink_bind
,
1576 .connect
= netlink_connect
,
1577 .socketpair
= sock_no_socketpair
,
1578 .accept
= sock_no_accept
,
1579 .getname
= netlink_getname
,
1580 .poll
= datagram_poll
,
1581 .ioctl
= sock_no_ioctl
,
1582 .listen
= sock_no_listen
,
1583 .shutdown
= sock_no_shutdown
,
1584 .setsockopt
= netlink_setsockopt
,
1585 .getsockopt
= netlink_getsockopt
,
1586 .sendmsg
= netlink_sendmsg
,
1587 .recvmsg
= netlink_recvmsg
,
1588 .mmap
= sock_no_mmap
,
1589 .sendpage
= sock_no_sendpage
,
1592 static struct net_proto_family netlink_family_ops
= {
1593 .family
= PF_NETLINK
,
1594 .create
= netlink_create
,
1595 .owner
= THIS_MODULE
, /* for consistency 8) */
1598 extern void netlink_skb_parms_too_large(void);
1600 static int __init
netlink_proto_init(void)
1602 struct sk_buff
*dummy_skb
;
1606 int err
= proto_register(&netlink_proto
, 0);
1611 if (sizeof(struct netlink_skb_parms
) > sizeof(dummy_skb
->cb
))
1612 netlink_skb_parms_too_large();
1614 nl_table
= kmalloc(sizeof(*nl_table
) * MAX_LINKS
, GFP_KERNEL
);
1617 printk(KERN_CRIT
"netlink_init: Cannot allocate nl_table\n");
1621 memset(nl_table
, 0, sizeof(*nl_table
) * MAX_LINKS
);
1623 if (num_physpages
>= (128 * 1024))
1624 max
= num_physpages
>> (21 - PAGE_SHIFT
);
1626 max
= num_physpages
>> (23 - PAGE_SHIFT
);
1628 order
= get_bitmask_order(max
) - 1 + PAGE_SHIFT
;
1629 max
= (1UL << order
) / sizeof(struct hlist_head
);
1630 order
= get_bitmask_order(max
> UINT_MAX
? UINT_MAX
: max
) - 1;
1632 for (i
= 0; i
< MAX_LINKS
; i
++) {
1633 struct nl_pid_hash
*hash
= &nl_table
[i
].hash
;
1635 hash
->table
= nl_pid_hash_alloc(1 * sizeof(*hash
->table
));
1638 nl_pid_hash_free(nl_table
[i
].hash
.table
,
1639 1 * sizeof(*hash
->table
));
1643 memset(hash
->table
, 0, 1 * sizeof(*hash
->table
));
1644 hash
->max_shift
= order
;
1647 hash
->rehash_time
= jiffies
;
1650 sock_register(&netlink_family_ops
);
1651 #ifdef CONFIG_PROC_FS
1652 proc_net_fops_create("netlink", 0, &netlink_seq_fops
);
1654 /* The netlink device handler may be needed early. */
1660 core_initcall(netlink_proto_init
);
1662 EXPORT_SYMBOL(netlink_ack
);
1663 EXPORT_SYMBOL(netlink_broadcast
);
1664 EXPORT_SYMBOL(netlink_dump_start
);
1665 EXPORT_SYMBOL(netlink_kernel_create
);
1666 EXPORT_SYMBOL(netlink_register_notifier
);
1667 EXPORT_SYMBOL(netlink_set_err
);
1668 EXPORT_SYMBOL(netlink_set_nonroot
);
1669 EXPORT_SYMBOL(netlink_unicast
);
1670 EXPORT_SYMBOL(netlink_unregister_notifier
);