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
,
706 long timeo
, struct sock
*ssk
)
708 struct netlink_sock
*nlk
;
712 if (atomic_read(&sk
->sk_rmem_alloc
) > sk
->sk_rcvbuf
||
713 test_bit(0, &nlk
->state
)) {
714 DECLARE_WAITQUEUE(wait
, current
);
716 if (!ssk
|| nlk_sk(ssk
)->pid
== 0)
723 __set_current_state(TASK_INTERRUPTIBLE
);
724 add_wait_queue(&nlk
->wait
, &wait
);
726 if ((atomic_read(&sk
->sk_rmem_alloc
) > sk
->sk_rcvbuf
||
727 test_bit(0, &nlk
->state
)) &&
728 !sock_flag(sk
, SOCK_DEAD
))
729 timeo
= schedule_timeout(timeo
);
731 __set_current_state(TASK_RUNNING
);
732 remove_wait_queue(&nlk
->wait
, &wait
);
735 if (signal_pending(current
)) {
737 return sock_intr_errno(timeo
);
741 skb_set_owner_r(skb
, sk
);
745 int netlink_sendskb(struct sock
*sk
, struct sk_buff
*skb
, int protocol
)
749 skb_queue_tail(&sk
->sk_receive_queue
, skb
);
750 sk
->sk_data_ready(sk
, len
);
755 void netlink_detachskb(struct sock
*sk
, struct sk_buff
*skb
)
761 static inline struct sk_buff
*netlink_trim(struct sk_buff
*skb
,
768 delta
= skb
->end
- skb
->tail
;
769 if (delta
* 2 < skb
->truesize
)
772 if (skb_shared(skb
)) {
773 struct sk_buff
*nskb
= skb_clone(skb
, allocation
);
780 if (!pskb_expand_head(skb
, 0, -delta
, allocation
))
781 skb
->truesize
-= delta
;
786 int netlink_unicast(struct sock
*ssk
, struct sk_buff
*skb
, u32 pid
, int nonblock
)
792 skb
= netlink_trim(skb
, gfp_any());
794 timeo
= sock_sndtimeo(ssk
, nonblock
);
796 sk
= netlink_getsockbypid(ssk
, pid
);
801 err
= netlink_attachskb(sk
, skb
, nonblock
, timeo
, ssk
);
807 return netlink_sendskb(sk
, skb
, ssk
->sk_protocol
);
810 static __inline__
int netlink_broadcast_deliver(struct sock
*sk
, struct sk_buff
*skb
)
812 struct netlink_sock
*nlk
= nlk_sk(sk
);
814 if (atomic_read(&sk
->sk_rmem_alloc
) <= sk
->sk_rcvbuf
&&
815 !test_bit(0, &nlk
->state
)) {
816 skb_set_owner_r(skb
, sk
);
817 skb_queue_tail(&sk
->sk_receive_queue
, skb
);
818 sk
->sk_data_ready(sk
, skb
->len
);
819 return atomic_read(&sk
->sk_rmem_alloc
) > sk
->sk_rcvbuf
;
824 struct netlink_broadcast_data
{
825 struct sock
*exclude_sk
;
832 struct sk_buff
*skb
, *skb2
;
835 static inline int do_one_broadcast(struct sock
*sk
,
836 struct netlink_broadcast_data
*p
)
838 struct netlink_sock
*nlk
= nlk_sk(sk
);
841 if (p
->exclude_sk
== sk
)
844 if (nlk
->pid
== p
->pid
|| p
->group
- 1 >= nlk
->ngroups
||
845 !test_bit(p
->group
- 1, nlk
->groups
))
854 if (p
->skb2
== NULL
) {
855 if (skb_shared(p
->skb
)) {
856 p
->skb2
= skb_clone(p
->skb
, p
->allocation
);
858 p
->skb2
= skb_get(p
->skb
);
860 * skb ownership may have been set when
861 * delivered to a previous socket.
866 if (p
->skb2
== NULL
) {
868 /* Clone failed. Notify ALL listeners. */
870 } else if ((val
= netlink_broadcast_deliver(sk
, p
->skb2
)) < 0) {
883 int netlink_broadcast(struct sock
*ssk
, struct sk_buff
*skb
, u32 pid
,
884 u32 group
, gfp_t allocation
)
886 struct netlink_broadcast_data info
;
887 struct hlist_node
*node
;
890 skb
= netlink_trim(skb
, allocation
);
892 info
.exclude_sk
= ssk
;
898 info
.allocation
= allocation
;
902 /* While we sleep in clone, do not allow to change socket list */
904 netlink_lock_table();
906 sk_for_each_bound(sk
, node
, &nl_table
[ssk
->sk_protocol
].mc_list
)
907 do_one_broadcast(sk
, &info
);
911 netlink_unlock_table();
914 kfree_skb(info
.skb2
);
916 if (info
.delivered
) {
917 if (info
.congested
&& (allocation
& __GFP_WAIT
))
926 struct netlink_set_err_data
{
927 struct sock
*exclude_sk
;
933 static inline int do_one_set_err(struct sock
*sk
,
934 struct netlink_set_err_data
*p
)
936 struct netlink_sock
*nlk
= nlk_sk(sk
);
938 if (sk
== p
->exclude_sk
)
941 if (nlk
->pid
== p
->pid
|| p
->group
- 1 >= nlk
->ngroups
||
942 !test_bit(p
->group
- 1, nlk
->groups
))
945 sk
->sk_err
= p
->code
;
946 sk
->sk_error_report(sk
);
951 void netlink_set_err(struct sock
*ssk
, u32 pid
, u32 group
, int code
)
953 struct netlink_set_err_data info
;
954 struct hlist_node
*node
;
957 info
.exclude_sk
= ssk
;
962 read_lock(&nl_table_lock
);
964 sk_for_each_bound(sk
, node
, &nl_table
[ssk
->sk_protocol
].mc_list
)
965 do_one_set_err(sk
, &info
);
967 read_unlock(&nl_table_lock
);
970 static int netlink_setsockopt(struct socket
*sock
, int level
, int optname
,
971 char __user
*optval
, int optlen
)
973 struct sock
*sk
= sock
->sk
;
974 struct netlink_sock
*nlk
= nlk_sk(sk
);
977 if (level
!= SOL_NETLINK
)
980 if (optlen
>= sizeof(int) &&
981 get_user(val
, (int __user
*)optval
))
985 case NETLINK_PKTINFO
:
987 nlk
->flags
|= NETLINK_RECV_PKTINFO
;
989 nlk
->flags
&= ~NETLINK_RECV_PKTINFO
;
992 case NETLINK_ADD_MEMBERSHIP
:
993 case NETLINK_DROP_MEMBERSHIP
: {
994 unsigned int subscriptions
;
995 int old
, new = optname
== NETLINK_ADD_MEMBERSHIP
? 1 : 0;
997 if (!netlink_capable(sock
, NL_NONROOT_RECV
))
999 if (nlk
->groups
== NULL
) {
1000 err
= netlink_alloc_groups(sk
);
1004 if (!val
|| val
- 1 >= nlk
->ngroups
)
1006 netlink_table_grab();
1007 old
= test_bit(val
- 1, nlk
->groups
);
1008 subscriptions
= nlk
->subscriptions
- old
+ new;
1010 __set_bit(val
- 1, nlk
->groups
);
1012 __clear_bit(val
- 1, nlk
->groups
);
1013 netlink_update_subscriptions(sk
, subscriptions
);
1014 netlink_table_ungrab();
1024 static int netlink_getsockopt(struct socket
*sock
, int level
, int optname
,
1025 char __user
*optval
, int __user
*optlen
)
1027 struct sock
*sk
= sock
->sk
;
1028 struct netlink_sock
*nlk
= nlk_sk(sk
);
1031 if (level
!= SOL_NETLINK
)
1032 return -ENOPROTOOPT
;
1034 if (get_user(len
, optlen
))
1040 case NETLINK_PKTINFO
:
1041 if (len
< sizeof(int))
1044 val
= nlk
->flags
& NETLINK_RECV_PKTINFO
? 1 : 0;
1045 put_user(len
, optlen
);
1046 put_user(val
, optval
);
1055 static void netlink_cmsg_recv_pktinfo(struct msghdr
*msg
, struct sk_buff
*skb
)
1057 struct nl_pktinfo info
;
1059 info
.group
= NETLINK_CB(skb
).dst_group
;
1060 put_cmsg(msg
, SOL_NETLINK
, NETLINK_PKTINFO
, sizeof(info
), &info
);
1063 static inline void netlink_rcv_wake(struct sock
*sk
)
1065 struct netlink_sock
*nlk
= nlk_sk(sk
);
1067 if (skb_queue_empty(&sk
->sk_receive_queue
))
1068 clear_bit(0, &nlk
->state
);
1069 if (!test_bit(0, &nlk
->state
))
1070 wake_up_interruptible(&nlk
->wait
);
1073 static int netlink_sendmsg(struct kiocb
*kiocb
, struct socket
*sock
,
1074 struct msghdr
*msg
, size_t len
)
1076 struct sock_iocb
*siocb
= kiocb_to_siocb(kiocb
);
1077 struct sock
*sk
= sock
->sk
;
1078 struct netlink_sock
*nlk
= nlk_sk(sk
);
1079 struct sockaddr_nl
*addr
=msg
->msg_name
;
1082 struct sk_buff
*skb
;
1084 struct scm_cookie scm
;
1086 if (msg
->msg_flags
&MSG_OOB
)
1089 if (NULL
== siocb
->scm
)
1091 err
= scm_send(sock
, msg
, siocb
->scm
);
1095 if (msg
->msg_namelen
) {
1096 if (addr
->nl_family
!= AF_NETLINK
)
1098 dst_pid
= addr
->nl_pid
;
1099 dst_group
= ffs(addr
->nl_groups
);
1100 if (dst_group
&& !netlink_capable(sock
, NL_NONROOT_SEND
))
1103 dst_pid
= nlk
->dst_pid
;
1104 dst_group
= nlk
->dst_group
;
1108 err
= netlink_autobind(sock
);
1114 if (len
> sk
->sk_sndbuf
- 32)
1117 skb
= alloc_skb(len
, GFP_KERNEL
);
1121 NETLINK_CB(skb
).pid
= nlk
->pid
;
1122 NETLINK_CB(skb
).dst_pid
= dst_pid
;
1123 NETLINK_CB(skb
).dst_group
= dst_group
;
1124 NETLINK_CB(skb
).loginuid
= audit_get_loginuid(current
->audit_context
);
1125 memcpy(NETLINK_CREDS(skb
), &siocb
->scm
->creds
, sizeof(struct ucred
));
1127 /* What can I do? Netlink is asynchronous, so that
1128 we will have to save current capabilities to
1129 check them, when this message will be delivered
1130 to corresponding kernel module. --ANK (980802)
1134 if (memcpy_fromiovec(skb_put(skb
,len
), msg
->msg_iov
, len
)) {
1139 err
= security_netlink_send(sk
, skb
);
1146 atomic_inc(&skb
->users
);
1147 netlink_broadcast(sk
, skb
, dst_pid
, dst_group
, GFP_KERNEL
);
1149 err
= netlink_unicast(sk
, skb
, dst_pid
, msg
->msg_flags
&MSG_DONTWAIT
);
1155 static int netlink_recvmsg(struct kiocb
*kiocb
, struct socket
*sock
,
1156 struct msghdr
*msg
, size_t len
,
1159 struct sock_iocb
*siocb
= kiocb_to_siocb(kiocb
);
1160 struct scm_cookie scm
;
1161 struct sock
*sk
= sock
->sk
;
1162 struct netlink_sock
*nlk
= nlk_sk(sk
);
1163 int noblock
= flags
&MSG_DONTWAIT
;
1165 struct sk_buff
*skb
;
1173 skb
= skb_recv_datagram(sk
,flags
,noblock
,&err
);
1177 msg
->msg_namelen
= 0;
1181 msg
->msg_flags
|= MSG_TRUNC
;
1185 skb
->h
.raw
= skb
->data
;
1186 err
= skb_copy_datagram_iovec(skb
, 0, msg
->msg_iov
, copied
);
1188 if (msg
->msg_name
) {
1189 struct sockaddr_nl
*addr
= (struct sockaddr_nl
*)msg
->msg_name
;
1190 addr
->nl_family
= AF_NETLINK
;
1192 addr
->nl_pid
= NETLINK_CB(skb
).pid
;
1193 addr
->nl_groups
= netlink_group_mask(NETLINK_CB(skb
).dst_group
);
1194 msg
->msg_namelen
= sizeof(*addr
);
1197 if (NULL
== siocb
->scm
) {
1198 memset(&scm
, 0, sizeof(scm
));
1201 siocb
->scm
->creds
= *NETLINK_CREDS(skb
);
1202 skb_free_datagram(sk
, skb
);
1204 if (nlk
->cb
&& atomic_read(&sk
->sk_rmem_alloc
) <= sk
->sk_rcvbuf
/ 2)
1207 scm_recv(sock
, msg
, siocb
->scm
, flags
);
1208 if (nlk
->flags
& NETLINK_RECV_PKTINFO
)
1209 netlink_cmsg_recv_pktinfo(msg
, skb
);
1212 netlink_rcv_wake(sk
);
1213 return err
? : copied
;
1216 static void netlink_data_ready(struct sock
*sk
, int len
)
1218 struct netlink_sock
*nlk
= nlk_sk(sk
);
1220 if (nlk
->data_ready
)
1221 nlk
->data_ready(sk
, len
);
1222 netlink_rcv_wake(sk
);
1226 * We export these functions to other modules. They provide a
1227 * complete set of kernel non-blocking support for message
1232 netlink_kernel_create(int unit
, unsigned int groups
,
1233 void (*input
)(struct sock
*sk
, int len
),
1234 struct module
*module
)
1236 struct socket
*sock
;
1238 struct netlink_sock
*nlk
;
1243 if (unit
<0 || unit
>=MAX_LINKS
)
1246 if (sock_create_lite(PF_NETLINK
, SOCK_DGRAM
, unit
, &sock
))
1249 if (__netlink_create(sock
, unit
) < 0)
1250 goto out_sock_release
;
1253 sk
->sk_data_ready
= netlink_data_ready
;
1255 nlk_sk(sk
)->data_ready
= input
;
1257 if (netlink_insert(sk
, 0))
1258 goto out_sock_release
;
1261 nlk
->flags
|= NETLINK_KERNEL_SOCKET
;
1263 netlink_table_grab();
1264 nl_table
[unit
].groups
= groups
< 32 ? 32 : groups
;
1265 nl_table
[unit
].module
= module
;
1266 nl_table
[unit
].registered
= 1;
1267 netlink_table_ungrab();
1276 void netlink_set_nonroot(int protocol
, unsigned int flags
)
1278 if ((unsigned int)protocol
< MAX_LINKS
)
1279 nl_table
[protocol
].nl_nonroot
= flags
;
1282 static void netlink_destroy_callback(struct netlink_callback
*cb
)
1290 * It looks a bit ugly.
1291 * It would be better to create kernel thread.
1294 static int netlink_dump(struct sock
*sk
)
1296 struct netlink_sock
*nlk
= nlk_sk(sk
);
1297 struct netlink_callback
*cb
;
1298 struct sk_buff
*skb
;
1299 struct nlmsghdr
*nlh
;
1302 skb
= sock_rmalloc(sk
, NLMSG_GOODSIZE
, 0, GFP_KERNEL
);
1306 spin_lock(&nlk
->cb_lock
);
1310 spin_unlock(&nlk
->cb_lock
);
1315 len
= cb
->dump(skb
, cb
);
1318 spin_unlock(&nlk
->cb_lock
);
1319 skb_queue_tail(&sk
->sk_receive_queue
, skb
);
1320 sk
->sk_data_ready(sk
, len
);
1324 nlh
= NLMSG_NEW_ANSWER(skb
, cb
, NLMSG_DONE
, sizeof(len
), NLM_F_MULTI
);
1325 memcpy(NLMSG_DATA(nlh
), &len
, sizeof(len
));
1326 skb_queue_tail(&sk
->sk_receive_queue
, skb
);
1327 sk
->sk_data_ready(sk
, skb
->len
);
1332 spin_unlock(&nlk
->cb_lock
);
1334 netlink_destroy_callback(cb
);
1341 int netlink_dump_start(struct sock
*ssk
, struct sk_buff
*skb
,
1342 struct nlmsghdr
*nlh
,
1343 int (*dump
)(struct sk_buff
*skb
, struct netlink_callback
*),
1344 int (*done
)(struct netlink_callback
*))
1346 struct netlink_callback
*cb
;
1348 struct netlink_sock
*nlk
;
1350 cb
= kmalloc(sizeof(*cb
), GFP_KERNEL
);
1354 memset(cb
, 0, sizeof(*cb
));
1358 atomic_inc(&skb
->users
);
1361 sk
= netlink_lookup(ssk
->sk_protocol
, NETLINK_CB(skb
).pid
);
1363 netlink_destroy_callback(cb
);
1364 return -ECONNREFUSED
;
1367 /* A dump is in progress... */
1368 spin_lock(&nlk
->cb_lock
);
1370 spin_unlock(&nlk
->cb_lock
);
1371 netlink_destroy_callback(cb
);
1376 spin_unlock(&nlk
->cb_lock
);
1383 void netlink_ack(struct sk_buff
*in_skb
, struct nlmsghdr
*nlh
, int err
)
1385 struct sk_buff
*skb
;
1386 struct nlmsghdr
*rep
;
1387 struct nlmsgerr
*errmsg
;
1391 size
= NLMSG_SPACE(sizeof(struct nlmsgerr
));
1393 size
= NLMSG_SPACE(4 + NLMSG_ALIGN(nlh
->nlmsg_len
));
1395 skb
= alloc_skb(size
, GFP_KERNEL
);
1399 sk
= netlink_lookup(in_skb
->sk
->sk_protocol
,
1400 NETLINK_CB(in_skb
).pid
);
1402 sk
->sk_err
= ENOBUFS
;
1403 sk
->sk_error_report(sk
);
1409 rep
= __nlmsg_put(skb
, NETLINK_CB(in_skb
).pid
, nlh
->nlmsg_seq
,
1410 NLMSG_ERROR
, sizeof(struct nlmsgerr
), 0);
1411 errmsg
= NLMSG_DATA(rep
);
1412 errmsg
->error
= err
;
1413 memcpy(&errmsg
->msg
, nlh
, err
? nlh
->nlmsg_len
: sizeof(struct nlmsghdr
));
1414 netlink_unicast(in_skb
->sk
, skb
, NETLINK_CB(in_skb
).pid
, MSG_DONTWAIT
);
1417 static int netlink_rcv_skb(struct sk_buff
*skb
, int (*cb
)(struct sk_buff
*,
1418 struct nlmsghdr
*, int *))
1420 unsigned int total_len
;
1421 struct nlmsghdr
*nlh
;
1424 while (skb
->len
>= nlmsg_total_size(0)) {
1425 nlh
= (struct nlmsghdr
*) skb
->data
;
1427 if (nlh
->nlmsg_len
< NLMSG_HDRLEN
|| skb
->len
< nlh
->nlmsg_len
)
1430 total_len
= min(NLMSG_ALIGN(nlh
->nlmsg_len
), skb
->len
);
1432 if (cb(skb
, nlh
, &err
) < 0) {
1433 /* Not an error, but we have to interrupt processing
1434 * here. Note: that in this case we do not pull
1435 * message from skb, it will be processed later.
1439 netlink_ack(skb
, nlh
, err
);
1440 } else if (nlh
->nlmsg_flags
& NLM_F_ACK
)
1441 netlink_ack(skb
, nlh
, 0);
1443 skb_pull(skb
, total_len
);
1450 * nelink_run_queue - Process netlink receive queue.
1451 * @sk: Netlink socket containing the queue
1452 * @qlen: Place to store queue length upon entry
1453 * @cb: Callback function invoked for each netlink message found
1455 * Processes as much as there was in the queue upon entry and invokes
1456 * a callback function for each netlink message found. The callback
1457 * function may refuse a message by returning a negative error code
1458 * but setting the error pointer to 0 in which case this function
1459 * returns with a qlen != 0.
1461 * qlen must be initialized to 0 before the initial entry, afterwards
1462 * the function may be called repeatedly until qlen reaches 0.
1464 void netlink_run_queue(struct sock
*sk
, unsigned int *qlen
,
1465 int (*cb
)(struct sk_buff
*, struct nlmsghdr
*, int *))
1467 struct sk_buff
*skb
;
1469 if (!*qlen
|| *qlen
> skb_queue_len(&sk
->sk_receive_queue
))
1470 *qlen
= skb_queue_len(&sk
->sk_receive_queue
);
1472 for (; *qlen
; (*qlen
)--) {
1473 skb
= skb_dequeue(&sk
->sk_receive_queue
);
1474 if (netlink_rcv_skb(skb
, cb
)) {
1476 skb_queue_head(&sk
->sk_receive_queue
, skb
);
1489 * netlink_queue_skip - Skip netlink message while processing queue.
1490 * @nlh: Netlink message to be skipped
1491 * @skb: Socket buffer containing the netlink messages.
1493 * Pulls the given netlink message off the socket buffer so the next
1494 * call to netlink_queue_run() will not reconsider the message.
1496 void netlink_queue_skip(struct nlmsghdr
*nlh
, struct sk_buff
*skb
)
1498 int msglen
= NLMSG_ALIGN(nlh
->nlmsg_len
);
1500 if (msglen
> skb
->len
)
1503 skb_pull(skb
, msglen
);
1506 #ifdef CONFIG_PROC_FS
1507 struct nl_seq_iter
{
1512 static struct sock
*netlink_seq_socket_idx(struct seq_file
*seq
, loff_t pos
)
1514 struct nl_seq_iter
*iter
= seq
->private;
1517 struct hlist_node
*node
;
1520 for (i
=0; i
<MAX_LINKS
; i
++) {
1521 struct nl_pid_hash
*hash
= &nl_table
[i
].hash
;
1523 for (j
= 0; j
<= hash
->mask
; j
++) {
1524 sk_for_each(s
, node
, &hash
->table
[j
]) {
1537 static void *netlink_seq_start(struct seq_file
*seq
, loff_t
*pos
)
1539 read_lock(&nl_table_lock
);
1540 return *pos
? netlink_seq_socket_idx(seq
, *pos
- 1) : SEQ_START_TOKEN
;
1543 static void *netlink_seq_next(struct seq_file
*seq
, void *v
, loff_t
*pos
)
1546 struct nl_seq_iter
*iter
;
1551 if (v
== SEQ_START_TOKEN
)
1552 return netlink_seq_socket_idx(seq
, 0);
1558 iter
= seq
->private;
1560 j
= iter
->hash_idx
+ 1;
1563 struct nl_pid_hash
*hash
= &nl_table
[i
].hash
;
1565 for (; j
<= hash
->mask
; j
++) {
1566 s
= sk_head(&hash
->table
[j
]);
1575 } while (++i
< MAX_LINKS
);
1580 static void netlink_seq_stop(struct seq_file
*seq
, void *v
)
1582 read_unlock(&nl_table_lock
);
1586 static int netlink_seq_show(struct seq_file
*seq
, void *v
)
1588 if (v
== SEQ_START_TOKEN
)
1590 "sk Eth Pid Groups "
1591 "Rmem Wmem Dump Locks\n");
1594 struct netlink_sock
*nlk
= nlk_sk(s
);
1596 seq_printf(seq
, "%p %-3d %-6d %08x %-8d %-8d %p %d\n",
1600 nlk
->groups
? (u32
)nlk
->groups
[0] : 0,
1601 atomic_read(&s
->sk_rmem_alloc
),
1602 atomic_read(&s
->sk_wmem_alloc
),
1604 atomic_read(&s
->sk_refcnt
)
1611 static struct seq_operations netlink_seq_ops
= {
1612 .start
= netlink_seq_start
,
1613 .next
= netlink_seq_next
,
1614 .stop
= netlink_seq_stop
,
1615 .show
= netlink_seq_show
,
1619 static int netlink_seq_open(struct inode
*inode
, struct file
*file
)
1621 struct seq_file
*seq
;
1622 struct nl_seq_iter
*iter
;
1625 iter
= kmalloc(sizeof(*iter
), GFP_KERNEL
);
1629 err
= seq_open(file
, &netlink_seq_ops
);
1635 memset(iter
, 0, sizeof(*iter
));
1636 seq
= file
->private_data
;
1637 seq
->private = iter
;
1641 static struct file_operations netlink_seq_fops
= {
1642 .owner
= THIS_MODULE
,
1643 .open
= netlink_seq_open
,
1645 .llseek
= seq_lseek
,
1646 .release
= seq_release_private
,
1651 int netlink_register_notifier(struct notifier_block
*nb
)
1653 return notifier_chain_register(&netlink_chain
, nb
);
1656 int netlink_unregister_notifier(struct notifier_block
*nb
)
1658 return notifier_chain_unregister(&netlink_chain
, nb
);
1661 static const struct proto_ops netlink_ops
= {
1662 .family
= PF_NETLINK
,
1663 .owner
= THIS_MODULE
,
1664 .release
= netlink_release
,
1665 .bind
= netlink_bind
,
1666 .connect
= netlink_connect
,
1667 .socketpair
= sock_no_socketpair
,
1668 .accept
= sock_no_accept
,
1669 .getname
= netlink_getname
,
1670 .poll
= datagram_poll
,
1671 .ioctl
= sock_no_ioctl
,
1672 .listen
= sock_no_listen
,
1673 .shutdown
= sock_no_shutdown
,
1674 .setsockopt
= netlink_setsockopt
,
1675 .getsockopt
= netlink_getsockopt
,
1676 .sendmsg
= netlink_sendmsg
,
1677 .recvmsg
= netlink_recvmsg
,
1678 .mmap
= sock_no_mmap
,
1679 .sendpage
= sock_no_sendpage
,
1682 static struct net_proto_family netlink_family_ops
= {
1683 .family
= PF_NETLINK
,
1684 .create
= netlink_create
,
1685 .owner
= THIS_MODULE
, /* for consistency 8) */
1688 extern void netlink_skb_parms_too_large(void);
1690 static int __init
netlink_proto_init(void)
1692 struct sk_buff
*dummy_skb
;
1696 int err
= proto_register(&netlink_proto
, 0);
1701 if (sizeof(struct netlink_skb_parms
) > sizeof(dummy_skb
->cb
))
1702 netlink_skb_parms_too_large();
1704 nl_table
= kmalloc(sizeof(*nl_table
) * MAX_LINKS
, GFP_KERNEL
);
1707 printk(KERN_CRIT
"netlink_init: Cannot allocate nl_table\n");
1711 memset(nl_table
, 0, sizeof(*nl_table
) * MAX_LINKS
);
1713 if (num_physpages
>= (128 * 1024))
1714 max
= num_physpages
>> (21 - PAGE_SHIFT
);
1716 max
= num_physpages
>> (23 - PAGE_SHIFT
);
1718 order
= get_bitmask_order(max
) - 1 + PAGE_SHIFT
;
1719 max
= (1UL << order
) / sizeof(struct hlist_head
);
1720 order
= get_bitmask_order(max
> UINT_MAX
? UINT_MAX
: max
) - 1;
1722 for (i
= 0; i
< MAX_LINKS
; i
++) {
1723 struct nl_pid_hash
*hash
= &nl_table
[i
].hash
;
1725 hash
->table
= nl_pid_hash_alloc(1 * sizeof(*hash
->table
));
1728 nl_pid_hash_free(nl_table
[i
].hash
.table
,
1729 1 * sizeof(*hash
->table
));
1733 memset(hash
->table
, 0, 1 * sizeof(*hash
->table
));
1734 hash
->max_shift
= order
;
1737 hash
->rehash_time
= jiffies
;
1740 sock_register(&netlink_family_ops
);
1741 #ifdef CONFIG_PROC_FS
1742 proc_net_fops_create("netlink", 0, &netlink_seq_fops
);
1744 /* The netlink device handler may be needed early. */
1750 core_initcall(netlink_proto_init
);
1752 EXPORT_SYMBOL(netlink_ack
);
1753 EXPORT_SYMBOL(netlink_run_queue
);
1754 EXPORT_SYMBOL(netlink_queue_skip
);
1755 EXPORT_SYMBOL(netlink_broadcast
);
1756 EXPORT_SYMBOL(netlink_dump_start
);
1757 EXPORT_SYMBOL(netlink_kernel_create
);
1758 EXPORT_SYMBOL(netlink_register_notifier
);
1759 EXPORT_SYMBOL(netlink_set_err
);
1760 EXPORT_SYMBOL(netlink_set_nonroot
);
1761 EXPORT_SYMBOL(netlink_unicast
);
1762 EXPORT_SYMBOL(netlink_unregister_notifier
);