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)
19 #include <linux/config.h>
20 #include <linux/module.h>
22 #include <linux/kernel.h>
23 #include <linux/init.h>
24 #include <linux/signal.h>
25 #include <linux/sched.h>
26 #include <linux/errno.h>
27 #include <linux/string.h>
28 #include <linux/stat.h>
29 #include <linux/socket.h>
31 #include <linux/fcntl.h>
32 #include <linux/termios.h>
33 #include <linux/sockios.h>
34 #include <linux/net.h>
36 #include <linux/slab.h>
37 #include <asm/uaccess.h>
38 #include <linux/skbuff.h>
39 #include <linux/netdevice.h>
40 #include <linux/rtnetlink.h>
41 #include <linux/proc_fs.h>
42 #include <linux/seq_file.h>
43 #include <linux/smp_lock.h>
44 #include <linux/notifier.h>
45 #include <linux/security.h>
46 #include <linux/jhash.h>
47 #include <linux/jiffies.h>
48 #include <linux/random.h>
49 #include <linux/bitops.h>
51 #include <linux/types.h>
58 /* struct sock has to be the first member of netlink_sock */
63 unsigned int dst_groups
;
65 wait_queue_head_t wait
;
66 struct netlink_callback
*cb
;
68 void (*data_ready
)(struct sock
*sk
, int bytes
);
71 static inline struct netlink_sock
*nlk_sk(struct sock
*sk
)
73 return (struct netlink_sock
*)sk
;
77 struct hlist_head
*table
;
78 unsigned long rehash_time
;
84 unsigned int max_shift
;
89 struct netlink_table
{
90 struct nl_pid_hash hash
;
91 struct hlist_head mc_list
;
92 unsigned int nl_nonroot
;
95 static struct netlink_table
*nl_table
;
97 static DECLARE_WAIT_QUEUE_HEAD(nl_table_wait
);
99 static int netlink_dump(struct sock
*sk
);
100 static void netlink_destroy_callback(struct netlink_callback
*cb
);
102 static DEFINE_RWLOCK(nl_table_lock
);
103 static atomic_t nl_table_users
= ATOMIC_INIT(0);
105 static struct notifier_block
*netlink_chain
;
107 static struct hlist_head
*nl_pid_hashfn(struct nl_pid_hash
*hash
, u32 pid
)
109 return &hash
->table
[jhash_1word(pid
, hash
->rnd
) & hash
->mask
];
112 static void netlink_sock_destruct(struct sock
*sk
)
114 skb_queue_purge(&sk
->sk_receive_queue
);
116 if (!sock_flag(sk
, SOCK_DEAD
)) {
117 printk("Freeing alive netlink socket %p\n", sk
);
120 BUG_TRAP(!atomic_read(&sk
->sk_rmem_alloc
));
121 BUG_TRAP(!atomic_read(&sk
->sk_wmem_alloc
));
122 BUG_TRAP(!nlk_sk(sk
)->cb
);
125 /* This lock without WQ_FLAG_EXCLUSIVE is good on UP and it is _very_ bad on SMP.
126 * Look, when several writers sleep and reader wakes them up, all but one
127 * immediately hit write lock and grab all the cpus. Exclusive sleep solves
128 * this, _but_ remember, it adds useless work on UP machines.
131 static void netlink_table_grab(void)
133 write_lock_bh(&nl_table_lock
);
135 if (atomic_read(&nl_table_users
)) {
136 DECLARE_WAITQUEUE(wait
, current
);
138 add_wait_queue_exclusive(&nl_table_wait
, &wait
);
140 set_current_state(TASK_UNINTERRUPTIBLE
);
141 if (atomic_read(&nl_table_users
) == 0)
143 write_unlock_bh(&nl_table_lock
);
145 write_lock_bh(&nl_table_lock
);
148 __set_current_state(TASK_RUNNING
);
149 remove_wait_queue(&nl_table_wait
, &wait
);
153 static __inline__
void netlink_table_ungrab(void)
155 write_unlock_bh(&nl_table_lock
);
156 wake_up(&nl_table_wait
);
159 static __inline__
void
160 netlink_lock_table(void)
162 /* read_lock() synchronizes us to netlink_table_grab */
164 read_lock(&nl_table_lock
);
165 atomic_inc(&nl_table_users
);
166 read_unlock(&nl_table_lock
);
169 static __inline__
void
170 netlink_unlock_table(void)
172 if (atomic_dec_and_test(&nl_table_users
))
173 wake_up(&nl_table_wait
);
176 static __inline__
struct sock
*netlink_lookup(int protocol
, u32 pid
)
178 struct nl_pid_hash
*hash
= &nl_table
[protocol
].hash
;
179 struct hlist_head
*head
;
181 struct hlist_node
*node
;
183 read_lock(&nl_table_lock
);
184 head
= nl_pid_hashfn(hash
, pid
);
185 sk_for_each(sk
, node
, head
) {
186 if (nlk_sk(sk
)->pid
== pid
) {
193 read_unlock(&nl_table_lock
);
197 static inline struct hlist_head
*nl_pid_hash_alloc(size_t size
)
199 if (size
<= PAGE_SIZE
)
200 return kmalloc(size
, GFP_ATOMIC
);
202 return (struct hlist_head
*)
203 __get_free_pages(GFP_ATOMIC
, get_order(size
));
206 static inline void nl_pid_hash_free(struct hlist_head
*table
, size_t size
)
208 if (size
<= PAGE_SIZE
)
211 free_pages((unsigned long)table
, get_order(size
));
214 static int nl_pid_hash_rehash(struct nl_pid_hash
*hash
, int grow
)
216 unsigned int omask
, mask
, shift
;
218 struct hlist_head
*otable
, *table
;
221 omask
= mask
= hash
->mask
;
222 osize
= size
= (mask
+ 1) * sizeof(*table
);
226 if (++shift
> hash
->max_shift
)
232 table
= nl_pid_hash_alloc(size
);
236 memset(table
, 0, size
);
237 otable
= hash
->table
;
241 get_random_bytes(&hash
->rnd
, sizeof(hash
->rnd
));
243 for (i
= 0; i
<= omask
; i
++) {
245 struct hlist_node
*node
, *tmp
;
247 sk_for_each_safe(sk
, node
, tmp
, &otable
[i
])
248 __sk_add_node(sk
, nl_pid_hashfn(hash
, nlk_sk(sk
)->pid
));
251 nl_pid_hash_free(otable
, osize
);
252 hash
->rehash_time
= jiffies
+ 10 * 60 * HZ
;
256 static inline int nl_pid_hash_dilute(struct nl_pid_hash
*hash
, int len
)
258 int avg
= hash
->entries
>> hash
->shift
;
260 if (unlikely(avg
> 1) && nl_pid_hash_rehash(hash
, 1))
263 if (unlikely(len
> avg
) && time_after(jiffies
, hash
->rehash_time
)) {
264 nl_pid_hash_rehash(hash
, 0);
271 static struct proto_ops netlink_ops
;
273 static int netlink_insert(struct sock
*sk
, u32 pid
)
275 struct nl_pid_hash
*hash
= &nl_table
[sk
->sk_protocol
].hash
;
276 struct hlist_head
*head
;
277 int err
= -EADDRINUSE
;
279 struct hlist_node
*node
;
282 netlink_table_grab();
283 head
= nl_pid_hashfn(hash
, pid
);
285 sk_for_each(osk
, node
, head
) {
286 if (nlk_sk(osk
)->pid
== pid
)
298 if (BITS_PER_LONG
> 32 && unlikely(hash
->entries
>= UINT_MAX
))
301 if (len
&& nl_pid_hash_dilute(hash
, len
))
302 head
= nl_pid_hashfn(hash
, pid
);
304 nlk_sk(sk
)->pid
= pid
;
305 sk_add_node(sk
, head
);
309 netlink_table_ungrab();
313 static void netlink_remove(struct sock
*sk
)
315 netlink_table_grab();
316 nl_table
[sk
->sk_protocol
].hash
.entries
--;
317 sk_del_node_init(sk
);
318 if (nlk_sk(sk
)->groups
)
319 __sk_del_bind_node(sk
);
320 netlink_table_ungrab();
323 static struct proto netlink_proto
= {
325 .owner
= THIS_MODULE
,
326 .obj_size
= sizeof(struct netlink_sock
),
329 static int netlink_create(struct socket
*sock
, int protocol
)
332 struct netlink_sock
*nlk
;
334 sock
->state
= SS_UNCONNECTED
;
336 if (sock
->type
!= SOCK_RAW
&& sock
->type
!= SOCK_DGRAM
)
337 return -ESOCKTNOSUPPORT
;
339 if (protocol
<0 || protocol
>= MAX_LINKS
)
340 return -EPROTONOSUPPORT
;
342 sock
->ops
= &netlink_ops
;
344 sk
= sk_alloc(PF_NETLINK
, GFP_KERNEL
, &netlink_proto
, 1);
348 sock_init_data(sock
, sk
);
352 spin_lock_init(&nlk
->cb_lock
);
353 init_waitqueue_head(&nlk
->wait
);
354 sk
->sk_destruct
= netlink_sock_destruct
;
356 sk
->sk_protocol
= protocol
;
360 static int netlink_release(struct socket
*sock
)
362 struct sock
*sk
= sock
->sk
;
363 struct netlink_sock
*nlk
;
371 spin_lock(&nlk
->cb_lock
);
373 nlk
->cb
->done(nlk
->cb
);
374 netlink_destroy_callback(nlk
->cb
);
378 spin_unlock(&nlk
->cb_lock
);
380 /* OK. Socket is unlinked, and, therefore,
381 no new packets will arrive */
385 wake_up_interruptible_all(&nlk
->wait
);
387 skb_queue_purge(&sk
->sk_write_queue
);
389 if (nlk
->pid
&& !nlk
->groups
) {
390 struct netlink_notify n
= {
391 .protocol
= sk
->sk_protocol
,
394 notifier_call_chain(&netlink_chain
, NETLINK_URELEASE
, &n
);
401 static int netlink_autobind(struct socket
*sock
)
403 struct sock
*sk
= sock
->sk
;
404 struct nl_pid_hash
*hash
= &nl_table
[sk
->sk_protocol
].hash
;
405 struct hlist_head
*head
;
407 struct hlist_node
*node
;
408 s32 pid
= current
->pid
;
410 static s32 rover
= -4097;
414 netlink_table_grab();
415 head
= nl_pid_hashfn(hash
, pid
);
416 sk_for_each(osk
, node
, head
) {
417 if (nlk_sk(osk
)->pid
== pid
) {
418 /* Bind collision, search negative pid values. */
422 netlink_table_ungrab();
426 netlink_table_ungrab();
428 err
= netlink_insert(sk
, pid
);
429 if (err
== -EADDRINUSE
)
434 static inline int netlink_capable(struct socket
*sock
, unsigned int flag
)
436 return (nl_table
[sock
->sk
->sk_protocol
].nl_nonroot
& flag
) ||
437 capable(CAP_NET_ADMIN
);
440 static int netlink_bind(struct socket
*sock
, struct sockaddr
*addr
, int addr_len
)
442 struct sock
*sk
= sock
->sk
;
443 struct netlink_sock
*nlk
= nlk_sk(sk
);
444 struct sockaddr_nl
*nladdr
= (struct sockaddr_nl
*)addr
;
447 if (nladdr
->nl_family
!= AF_NETLINK
)
450 /* Only superuser is allowed to listen multicasts */
451 if (nladdr
->nl_groups
&& !netlink_capable(sock
, NL_NONROOT_RECV
))
455 if (nladdr
->nl_pid
!= nlk
->pid
)
458 err
= nladdr
->nl_pid
?
459 netlink_insert(sk
, nladdr
->nl_pid
) :
460 netlink_autobind(sock
);
465 if (!nladdr
->nl_groups
&& !nlk
->groups
)
468 netlink_table_grab();
469 if (nlk
->groups
&& !nladdr
->nl_groups
)
470 __sk_del_bind_node(sk
);
471 else if (!nlk
->groups
&& nladdr
->nl_groups
)
472 sk_add_bind_node(sk
, &nl_table
[sk
->sk_protocol
].mc_list
);
473 nlk
->groups
= nladdr
->nl_groups
;
474 netlink_table_ungrab();
479 static int netlink_connect(struct socket
*sock
, struct sockaddr
*addr
,
483 struct sock
*sk
= sock
->sk
;
484 struct netlink_sock
*nlk
= nlk_sk(sk
);
485 struct sockaddr_nl
*nladdr
=(struct sockaddr_nl
*)addr
;
487 if (addr
->sa_family
== AF_UNSPEC
) {
488 sk
->sk_state
= NETLINK_UNCONNECTED
;
493 if (addr
->sa_family
!= AF_NETLINK
)
496 /* Only superuser is allowed to send multicasts */
497 if (nladdr
->nl_groups
&& !netlink_capable(sock
, NL_NONROOT_SEND
))
501 err
= netlink_autobind(sock
);
504 sk
->sk_state
= NETLINK_CONNECTED
;
505 nlk
->dst_pid
= nladdr
->nl_pid
;
506 nlk
->dst_groups
= nladdr
->nl_groups
;
512 static int netlink_getname(struct socket
*sock
, struct sockaddr
*addr
, int *addr_len
, int peer
)
514 struct sock
*sk
= sock
->sk
;
515 struct netlink_sock
*nlk
= nlk_sk(sk
);
516 struct sockaddr_nl
*nladdr
=(struct sockaddr_nl
*)addr
;
518 nladdr
->nl_family
= AF_NETLINK
;
520 *addr_len
= sizeof(*nladdr
);
523 nladdr
->nl_pid
= nlk
->dst_pid
;
524 nladdr
->nl_groups
= nlk
->dst_groups
;
526 nladdr
->nl_pid
= nlk
->pid
;
527 nladdr
->nl_groups
= nlk
->groups
;
532 static void netlink_overrun(struct sock
*sk
)
534 if (!test_and_set_bit(0, &nlk_sk(sk
)->state
)) {
535 sk
->sk_err
= ENOBUFS
;
536 sk
->sk_error_report(sk
);
540 static struct sock
*netlink_getsockbypid(struct sock
*ssk
, u32 pid
)
542 int protocol
= ssk
->sk_protocol
;
544 struct netlink_sock
*nlk
;
546 sock
= netlink_lookup(protocol
, pid
);
548 return ERR_PTR(-ECONNREFUSED
);
550 /* Don't bother queuing skb if kernel socket has no input function */
552 if ((nlk
->pid
== 0 && !nlk
->data_ready
) ||
553 (sock
->sk_state
== NETLINK_CONNECTED
&&
554 nlk
->dst_pid
!= nlk_sk(ssk
)->pid
)) {
556 return ERR_PTR(-ECONNREFUSED
);
561 struct sock
*netlink_getsockbyfilp(struct file
*filp
)
563 struct inode
*inode
= filp
->f_dentry
->d_inode
;
566 if (!S_ISSOCK(inode
->i_mode
))
567 return ERR_PTR(-ENOTSOCK
);
569 sock
= SOCKET_I(inode
)->sk
;
570 if (sock
->sk_family
!= AF_NETLINK
)
571 return ERR_PTR(-EINVAL
);
578 * Attach a skb to a netlink socket.
579 * The caller must hold a reference to the destination socket. On error, the
580 * reference is dropped. The skb is not send to the destination, just all
581 * all error checks are performed and memory in the queue is reserved.
583 * < 0: error. skb freed, reference to sock dropped.
585 * 1: repeat lookup - reference dropped while waiting for socket memory.
587 int netlink_attachskb(struct sock
*sk
, struct sk_buff
*skb
, int nonblock
, long timeo
)
589 struct netlink_sock
*nlk
;
593 if (atomic_read(&sk
->sk_rmem_alloc
) > sk
->sk_rcvbuf
||
594 test_bit(0, &nlk
->state
)) {
595 DECLARE_WAITQUEUE(wait
, current
);
604 __set_current_state(TASK_INTERRUPTIBLE
);
605 add_wait_queue(&nlk
->wait
, &wait
);
607 if ((atomic_read(&sk
->sk_rmem_alloc
) > sk
->sk_rcvbuf
||
608 test_bit(0, &nlk
->state
)) &&
609 !sock_flag(sk
, SOCK_DEAD
))
610 timeo
= schedule_timeout(timeo
);
612 __set_current_state(TASK_RUNNING
);
613 remove_wait_queue(&nlk
->wait
, &wait
);
616 if (signal_pending(current
)) {
618 return sock_intr_errno(timeo
);
622 skb_set_owner_r(skb
, sk
);
626 int netlink_sendskb(struct sock
*sk
, struct sk_buff
*skb
, int protocol
)
628 struct netlink_sock
*nlk
;
633 skb_queue_tail(&sk
->sk_receive_queue
, skb
);
634 sk
->sk_data_ready(sk
, len
);
639 void netlink_detachskb(struct sock
*sk
, struct sk_buff
*skb
)
645 static inline struct sk_buff
*netlink_trim(struct sk_buff
*skb
, int allocation
)
651 delta
= skb
->end
- skb
->tail
;
652 if (delta
* 2 < skb
->truesize
)
655 if (skb_shared(skb
)) {
656 struct sk_buff
*nskb
= skb_clone(skb
, allocation
);
663 if (!pskb_expand_head(skb
, 0, -delta
, allocation
))
664 skb
->truesize
-= delta
;
669 int netlink_unicast(struct sock
*ssk
, struct sk_buff
*skb
, u32 pid
, int nonblock
)
675 skb
= netlink_trim(skb
, gfp_any());
677 timeo
= sock_sndtimeo(ssk
, nonblock
);
679 sk
= netlink_getsockbypid(ssk
, pid
);
684 err
= netlink_attachskb(sk
, skb
, nonblock
, timeo
);
690 return netlink_sendskb(sk
, skb
, ssk
->sk_protocol
);
693 static __inline__
int netlink_broadcast_deliver(struct sock
*sk
, struct sk_buff
*skb
)
695 struct netlink_sock
*nlk
= nlk_sk(sk
);
697 if (atomic_read(&sk
->sk_rmem_alloc
) <= sk
->sk_rcvbuf
&&
698 !test_bit(0, &nlk
->state
)) {
699 skb_set_owner_r(skb
, sk
);
700 skb_queue_tail(&sk
->sk_receive_queue
, skb
);
701 sk
->sk_data_ready(sk
, skb
->len
);
702 return atomic_read(&sk
->sk_rmem_alloc
) > sk
->sk_rcvbuf
;
707 struct netlink_broadcast_data
{
708 struct sock
*exclude_sk
;
715 struct sk_buff
*skb
, *skb2
;
718 static inline int do_one_broadcast(struct sock
*sk
,
719 struct netlink_broadcast_data
*p
)
721 struct netlink_sock
*nlk
= nlk_sk(sk
);
724 if (p
->exclude_sk
== sk
)
727 if (nlk
->pid
== p
->pid
|| !(nlk
->groups
& p
->group
))
736 if (p
->skb2
== NULL
) {
737 if (atomic_read(&p
->skb
->users
) != 1) {
738 p
->skb2
= skb_clone(p
->skb
, p
->allocation
);
741 atomic_inc(&p
->skb
->users
);
744 if (p
->skb2
== NULL
) {
746 /* Clone failed. Notify ALL listeners. */
748 } else if ((val
= netlink_broadcast_deliver(sk
, p
->skb2
)) < 0) {
761 int netlink_broadcast(struct sock
*ssk
, struct sk_buff
*skb
, u32 pid
,
762 u32 group
, int allocation
)
764 struct netlink_broadcast_data info
;
765 struct hlist_node
*node
;
768 skb
= netlink_trim(skb
, allocation
);
770 info
.exclude_sk
= ssk
;
776 info
.allocation
= allocation
;
780 /* While we sleep in clone, do not allow to change socket list */
782 netlink_lock_table();
784 sk_for_each_bound(sk
, node
, &nl_table
[ssk
->sk_protocol
].mc_list
)
785 do_one_broadcast(sk
, &info
);
787 netlink_unlock_table();
790 kfree_skb(info
.skb2
);
793 if (info
.delivered
) {
794 if (info
.congested
&& (allocation
& __GFP_WAIT
))
803 struct netlink_set_err_data
{
804 struct sock
*exclude_sk
;
810 static inline int do_one_set_err(struct sock
*sk
,
811 struct netlink_set_err_data
*p
)
813 struct netlink_sock
*nlk
= nlk_sk(sk
);
815 if (sk
== p
->exclude_sk
)
818 if (nlk
->pid
== p
->pid
|| !(nlk
->groups
& p
->group
))
821 sk
->sk_err
= p
->code
;
822 sk
->sk_error_report(sk
);
827 void netlink_set_err(struct sock
*ssk
, u32 pid
, u32 group
, int code
)
829 struct netlink_set_err_data info
;
830 struct hlist_node
*node
;
833 info
.exclude_sk
= ssk
;
838 read_lock(&nl_table_lock
);
840 sk_for_each_bound(sk
, node
, &nl_table
[ssk
->sk_protocol
].mc_list
)
841 do_one_set_err(sk
, &info
);
843 read_unlock(&nl_table_lock
);
846 static inline void netlink_rcv_wake(struct sock
*sk
)
848 struct netlink_sock
*nlk
= nlk_sk(sk
);
850 if (!skb_queue_len(&sk
->sk_receive_queue
))
851 clear_bit(0, &nlk
->state
);
852 if (!test_bit(0, &nlk
->state
))
853 wake_up_interruptible(&nlk
->wait
);
856 static int netlink_sendmsg(struct kiocb
*kiocb
, struct socket
*sock
,
857 struct msghdr
*msg
, size_t len
)
859 struct sock_iocb
*siocb
= kiocb_to_siocb(kiocb
);
860 struct sock
*sk
= sock
->sk
;
861 struct netlink_sock
*nlk
= nlk_sk(sk
);
862 struct sockaddr_nl
*addr
=msg
->msg_name
;
867 struct scm_cookie scm
;
869 if (msg
->msg_flags
&MSG_OOB
)
872 if (NULL
== siocb
->scm
)
874 err
= scm_send(sock
, msg
, siocb
->scm
);
878 if (msg
->msg_namelen
) {
879 if (addr
->nl_family
!= AF_NETLINK
)
881 dst_pid
= addr
->nl_pid
;
882 dst_groups
= addr
->nl_groups
;
883 if (dst_groups
&& !netlink_capable(sock
, NL_NONROOT_SEND
))
886 dst_pid
= nlk
->dst_pid
;
887 dst_groups
= nlk
->dst_groups
;
891 err
= netlink_autobind(sock
);
897 if (len
> sk
->sk_sndbuf
- 32)
900 skb
= alloc_skb(len
, GFP_KERNEL
);
904 NETLINK_CB(skb
).pid
= nlk
->pid
;
905 NETLINK_CB(skb
).groups
= nlk
->groups
;
906 NETLINK_CB(skb
).dst_pid
= dst_pid
;
907 NETLINK_CB(skb
).dst_groups
= dst_groups
;
908 memcpy(NETLINK_CREDS(skb
), &siocb
->scm
->creds
, sizeof(struct ucred
));
910 /* What can I do? Netlink is asynchronous, so that
911 we will have to save current capabilities to
912 check them, when this message will be delivered
913 to corresponding kernel module. --ANK (980802)
917 if (memcpy_fromiovec(skb_put(skb
,len
), msg
->msg_iov
, len
)) {
922 err
= security_netlink_send(sk
, skb
);
929 atomic_inc(&skb
->users
);
930 netlink_broadcast(sk
, skb
, dst_pid
, dst_groups
, GFP_KERNEL
);
932 err
= netlink_unicast(sk
, skb
, dst_pid
, msg
->msg_flags
&MSG_DONTWAIT
);
938 static int netlink_recvmsg(struct kiocb
*kiocb
, struct socket
*sock
,
939 struct msghdr
*msg
, size_t len
,
942 struct sock_iocb
*siocb
= kiocb_to_siocb(kiocb
);
943 struct scm_cookie scm
;
944 struct sock
*sk
= sock
->sk
;
945 struct netlink_sock
*nlk
= nlk_sk(sk
);
946 int noblock
= flags
&MSG_DONTWAIT
;
956 skb
= skb_recv_datagram(sk
,flags
,noblock
,&err
);
960 msg
->msg_namelen
= 0;
964 msg
->msg_flags
|= MSG_TRUNC
;
968 skb
->h
.raw
= skb
->data
;
969 err
= skb_copy_datagram_iovec(skb
, 0, msg
->msg_iov
, copied
);
972 struct sockaddr_nl
*addr
= (struct sockaddr_nl
*)msg
->msg_name
;
973 addr
->nl_family
= AF_NETLINK
;
975 addr
->nl_pid
= NETLINK_CB(skb
).pid
;
976 addr
->nl_groups
= NETLINK_CB(skb
).dst_groups
;
977 msg
->msg_namelen
= sizeof(*addr
);
980 if (NULL
== siocb
->scm
) {
981 memset(&scm
, 0, sizeof(scm
));
984 siocb
->scm
->creds
= *NETLINK_CREDS(skb
);
985 skb_free_datagram(sk
, skb
);
987 if (nlk
->cb
&& atomic_read(&sk
->sk_rmem_alloc
) <= sk
->sk_rcvbuf
/ 2)
990 scm_recv(sock
, msg
, siocb
->scm
, flags
);
993 netlink_rcv_wake(sk
);
994 return err
? : copied
;
997 static void netlink_data_ready(struct sock
*sk
, int len
)
999 struct netlink_sock
*nlk
= nlk_sk(sk
);
1001 if (nlk
->data_ready
)
1002 nlk
->data_ready(sk
, len
);
1003 netlink_rcv_wake(sk
);
1007 * We export these functions to other modules. They provide a
1008 * complete set of kernel non-blocking support for message
1013 netlink_kernel_create(int unit
, void (*input
)(struct sock
*sk
, int len
))
1015 struct socket
*sock
;
1021 if (unit
<0 || unit
>=MAX_LINKS
)
1024 if (sock_create_lite(PF_NETLINK
, SOCK_DGRAM
, unit
, &sock
))
1027 if (netlink_create(sock
, unit
) < 0) {
1032 sk
->sk_data_ready
= netlink_data_ready
;
1034 nlk_sk(sk
)->data_ready
= input
;
1036 if (netlink_insert(sk
, 0)) {
1043 void netlink_set_nonroot(int protocol
, unsigned int flags
)
1045 if ((unsigned int)protocol
< MAX_LINKS
)
1046 nl_table
[protocol
].nl_nonroot
= flags
;
1049 static void netlink_destroy_callback(struct netlink_callback
*cb
)
1057 * It looks a bit ugly.
1058 * It would be better to create kernel thread.
1061 static int netlink_dump(struct sock
*sk
)
1063 struct netlink_sock
*nlk
= nlk_sk(sk
);
1064 struct netlink_callback
*cb
;
1065 struct sk_buff
*skb
;
1066 struct nlmsghdr
*nlh
;
1069 skb
= sock_rmalloc(sk
, NLMSG_GOODSIZE
, 0, GFP_KERNEL
);
1073 spin_lock(&nlk
->cb_lock
);
1077 spin_unlock(&nlk
->cb_lock
);
1082 len
= cb
->dump(skb
, cb
);
1085 spin_unlock(&nlk
->cb_lock
);
1086 skb_queue_tail(&sk
->sk_receive_queue
, skb
);
1087 sk
->sk_data_ready(sk
, len
);
1091 nlh
= __nlmsg_put(skb
, NETLINK_CB(cb
->skb
).pid
, cb
->nlh
->nlmsg_seq
, NLMSG_DONE
, sizeof(int));
1092 nlh
->nlmsg_flags
|= NLM_F_MULTI
;
1093 memcpy(NLMSG_DATA(nlh
), &len
, sizeof(len
));
1094 skb_queue_tail(&sk
->sk_receive_queue
, skb
);
1095 sk
->sk_data_ready(sk
, skb
->len
);
1099 spin_unlock(&nlk
->cb_lock
);
1101 netlink_destroy_callback(cb
);
1106 int netlink_dump_start(struct sock
*ssk
, struct sk_buff
*skb
,
1107 struct nlmsghdr
*nlh
,
1108 int (*dump
)(struct sk_buff
*skb
, struct netlink_callback
*),
1109 int (*done
)(struct netlink_callback
*))
1111 struct netlink_callback
*cb
;
1113 struct netlink_sock
*nlk
;
1115 cb
= kmalloc(sizeof(*cb
), GFP_KERNEL
);
1119 memset(cb
, 0, sizeof(*cb
));
1123 atomic_inc(&skb
->users
);
1126 sk
= netlink_lookup(ssk
->sk_protocol
, NETLINK_CB(skb
).pid
);
1128 netlink_destroy_callback(cb
);
1129 return -ECONNREFUSED
;
1132 /* A dump is in progress... */
1133 spin_lock(&nlk
->cb_lock
);
1135 spin_unlock(&nlk
->cb_lock
);
1136 netlink_destroy_callback(cb
);
1142 spin_unlock(&nlk
->cb_lock
);
1149 void netlink_ack(struct sk_buff
*in_skb
, struct nlmsghdr
*nlh
, int err
)
1151 struct sk_buff
*skb
;
1152 struct nlmsghdr
*rep
;
1153 struct nlmsgerr
*errmsg
;
1157 size
= NLMSG_SPACE(sizeof(struct nlmsgerr
));
1159 size
= NLMSG_SPACE(4 + NLMSG_ALIGN(nlh
->nlmsg_len
));
1161 skb
= alloc_skb(size
, GFP_KERNEL
);
1165 sk
= netlink_lookup(in_skb
->sk
->sk_protocol
,
1166 NETLINK_CB(in_skb
).pid
);
1168 sk
->sk_err
= ENOBUFS
;
1169 sk
->sk_error_report(sk
);
1175 rep
= __nlmsg_put(skb
, NETLINK_CB(in_skb
).pid
, nlh
->nlmsg_seq
,
1176 NLMSG_ERROR
, sizeof(struct nlmsgerr
));
1177 errmsg
= NLMSG_DATA(rep
);
1178 errmsg
->error
= err
;
1179 memcpy(&errmsg
->msg
, nlh
, err
? nlh
->nlmsg_len
: sizeof(struct nlmsghdr
));
1180 netlink_unicast(in_skb
->sk
, skb
, NETLINK_CB(in_skb
).pid
, MSG_DONTWAIT
);
1184 #ifdef CONFIG_PROC_FS
1185 struct nl_seq_iter
{
1190 static struct sock
*netlink_seq_socket_idx(struct seq_file
*seq
, loff_t pos
)
1192 struct nl_seq_iter
*iter
= seq
->private;
1195 struct hlist_node
*node
;
1198 for (i
=0; i
<MAX_LINKS
; i
++) {
1199 struct nl_pid_hash
*hash
= &nl_table
[i
].hash
;
1201 for (j
= 0; j
<= hash
->mask
; j
++) {
1202 sk_for_each(s
, node
, &hash
->table
[j
]) {
1215 static void *netlink_seq_start(struct seq_file
*seq
, loff_t
*pos
)
1217 read_lock(&nl_table_lock
);
1218 return *pos
? netlink_seq_socket_idx(seq
, *pos
- 1) : SEQ_START_TOKEN
;
1221 static void *netlink_seq_next(struct seq_file
*seq
, void *v
, loff_t
*pos
)
1224 struct nl_seq_iter
*iter
;
1229 if (v
== SEQ_START_TOKEN
)
1230 return netlink_seq_socket_idx(seq
, 0);
1236 iter
= seq
->private;
1238 j
= iter
->hash_idx
+ 1;
1241 struct nl_pid_hash
*hash
= &nl_table
[i
].hash
;
1243 for (; j
<= hash
->mask
; j
++) {
1244 s
= sk_head(&hash
->table
[j
]);
1253 } while (++i
< MAX_LINKS
);
1258 static void netlink_seq_stop(struct seq_file
*seq
, void *v
)
1260 read_unlock(&nl_table_lock
);
1264 static int netlink_seq_show(struct seq_file
*seq
, void *v
)
1266 if (v
== SEQ_START_TOKEN
)
1268 "sk Eth Pid Groups "
1269 "Rmem Wmem Dump Locks\n");
1272 struct netlink_sock
*nlk
= nlk_sk(s
);
1274 seq_printf(seq
, "%p %-3d %-6d %08x %-8d %-8d %p %d\n",
1279 atomic_read(&s
->sk_rmem_alloc
),
1280 atomic_read(&s
->sk_wmem_alloc
),
1282 atomic_read(&s
->sk_refcnt
)
1289 static struct seq_operations netlink_seq_ops
= {
1290 .start
= netlink_seq_start
,
1291 .next
= netlink_seq_next
,
1292 .stop
= netlink_seq_stop
,
1293 .show
= netlink_seq_show
,
1297 static int netlink_seq_open(struct inode
*inode
, struct file
*file
)
1299 struct seq_file
*seq
;
1300 struct nl_seq_iter
*iter
;
1303 iter
= kmalloc(sizeof(*iter
), GFP_KERNEL
);
1307 err
= seq_open(file
, &netlink_seq_ops
);
1313 memset(iter
, 0, sizeof(*iter
));
1314 seq
= file
->private_data
;
1315 seq
->private = iter
;
1319 static struct file_operations netlink_seq_fops
= {
1320 .owner
= THIS_MODULE
,
1321 .open
= netlink_seq_open
,
1323 .llseek
= seq_lseek
,
1324 .release
= seq_release_private
,
1329 int netlink_register_notifier(struct notifier_block
*nb
)
1331 return notifier_chain_register(&netlink_chain
, nb
);
1334 int netlink_unregister_notifier(struct notifier_block
*nb
)
1336 return notifier_chain_unregister(&netlink_chain
, nb
);
1339 static struct proto_ops netlink_ops
= {
1340 .family
= PF_NETLINK
,
1341 .owner
= THIS_MODULE
,
1342 .release
= netlink_release
,
1343 .bind
= netlink_bind
,
1344 .connect
= netlink_connect
,
1345 .socketpair
= sock_no_socketpair
,
1346 .accept
= sock_no_accept
,
1347 .getname
= netlink_getname
,
1348 .poll
= datagram_poll
,
1349 .ioctl
= sock_no_ioctl
,
1350 .listen
= sock_no_listen
,
1351 .shutdown
= sock_no_shutdown
,
1352 .setsockopt
= sock_no_setsockopt
,
1353 .getsockopt
= sock_no_getsockopt
,
1354 .sendmsg
= netlink_sendmsg
,
1355 .recvmsg
= netlink_recvmsg
,
1356 .mmap
= sock_no_mmap
,
1357 .sendpage
= sock_no_sendpage
,
1360 static struct net_proto_family netlink_family_ops
= {
1361 .family
= PF_NETLINK
,
1362 .create
= netlink_create
,
1363 .owner
= THIS_MODULE
, /* for consistency 8) */
1366 extern void netlink_skb_parms_too_large(void);
1368 static int __init
netlink_proto_init(void)
1370 struct sk_buff
*dummy_skb
;
1374 int err
= proto_register(&netlink_proto
, 0);
1379 if (sizeof(struct netlink_skb_parms
) > sizeof(dummy_skb
->cb
))
1380 netlink_skb_parms_too_large();
1382 nl_table
= kmalloc(sizeof(*nl_table
) * MAX_LINKS
, GFP_KERNEL
);
1385 printk(KERN_CRIT
"netlink_init: Cannot allocate nl_table\n");
1389 memset(nl_table
, 0, sizeof(*nl_table
) * MAX_LINKS
);
1391 if (num_physpages
>= (128 * 1024))
1392 max
= num_physpages
>> (21 - PAGE_SHIFT
);
1394 max
= num_physpages
>> (23 - PAGE_SHIFT
);
1396 order
= get_bitmask_order(max
) - 1 + PAGE_SHIFT
;
1397 max
= (1UL << order
) / sizeof(struct hlist_head
);
1398 order
= get_bitmask_order(max
> UINT_MAX
? UINT_MAX
: max
) - 1;
1400 for (i
= 0; i
< MAX_LINKS
; i
++) {
1401 struct nl_pid_hash
*hash
= &nl_table
[i
].hash
;
1403 hash
->table
= nl_pid_hash_alloc(1 * sizeof(*hash
->table
));
1406 nl_pid_hash_free(nl_table
[i
].hash
.table
,
1407 1 * sizeof(*hash
->table
));
1411 memset(hash
->table
, 0, 1 * sizeof(*hash
->table
));
1412 hash
->max_shift
= order
;
1415 hash
->rehash_time
= jiffies
;
1418 sock_register(&netlink_family_ops
);
1419 #ifdef CONFIG_PROC_FS
1420 proc_net_fops_create("netlink", 0, &netlink_seq_fops
);
1422 /* The netlink device handler may be needed early. */
1428 static void __exit
netlink_proto_exit(void)
1430 sock_unregister(PF_NETLINK
);
1431 proc_net_remove("netlink");
1434 proto_unregister(&netlink_proto
);
1437 core_initcall(netlink_proto_init
);
1438 module_exit(netlink_proto_exit
);
1440 MODULE_LICENSE("GPL");
1442 MODULE_ALIAS_NETPROTO(PF_NETLINK
);
1444 EXPORT_SYMBOL(netlink_ack
);
1445 EXPORT_SYMBOL(netlink_broadcast
);
1446 EXPORT_SYMBOL(netlink_dump_start
);
1447 EXPORT_SYMBOL(netlink_kernel_create
);
1448 EXPORT_SYMBOL(netlink_register_notifier
);
1449 EXPORT_SYMBOL(netlink_set_err
);
1450 EXPORT_SYMBOL(netlink_set_nonroot
);
1451 EXPORT_SYMBOL(netlink_unicast
);
1452 EXPORT_SYMBOL(netlink_unregister_notifier
);