2 * NETLINK Kernel-user communication protocol.
4 * Authors: Alan Cox <alan@lxorguk.ukuu.org.uk>
5 * Alexey Kuznetsov <kuznet@ms2.inr.ac.ru>
6 * Patrick McHardy <kaber@trash.net>
8 * This program is free software; you can redistribute it and/or
9 * modify it under the terms of the GNU General Public License
10 * as published by the Free Software Foundation; either version
11 * 2 of the License, or (at your option) any later version.
13 * Tue Jun 26 14:36:48 MEST 2001 Herbert "herp" Rosmanith
14 * added netlink_proto_exit
15 * Tue Jan 22 18:32:44 BRST 2002 Arnaldo C. de Melo <acme@conectiva.com.br>
16 * use nlk_sk, as sk->protinfo is on a diet 8)
17 * Fri Jul 22 19:51:12 MEST 2005 Harald Welte <laforge@gnumonks.org>
18 * - inc module use count of module that owns
19 * the kernel socket in case userspace opens
20 * socket of same protocol
21 * - remove all module support, since netlink is
22 * mandatory if CONFIG_NET=y these days
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/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>
58 #include <linux/mutex.h>
59 #include <linux/vmalloc.h>
60 #include <linux/if_arp.h>
61 #include <linux/rhashtable.h>
62 #include <asm/cacheflush.h>
63 #include <linux/hash.h>
64 #include <linux/genetlink.h>
66 #include <net/net_namespace.h>
69 #include <net/netlink.h>
71 #include "af_netlink.h"
75 unsigned long masks
[0];
79 #define NETLINK_CONGESTED 0x0
82 #define NETLINK_KERNEL_SOCKET 0x1
83 #define NETLINK_RECV_PKTINFO 0x2
84 #define NETLINK_BROADCAST_SEND_ERROR 0x4
85 #define NETLINK_RECV_NO_ENOBUFS 0x8
87 static inline int netlink_is_kernel(struct sock
*sk
)
89 return nlk_sk(sk
)->flags
& NETLINK_KERNEL_SOCKET
;
92 struct netlink_table
*nl_table
;
93 EXPORT_SYMBOL_GPL(nl_table
);
95 static DECLARE_WAIT_QUEUE_HEAD(nl_table_wait
);
97 static int netlink_dump(struct sock
*sk
);
98 static void netlink_skb_destructor(struct sk_buff
*skb
);
100 /* nl_table locking explained:
101 * Lookup and traversal are protected with nl_sk_hash_lock or nl_table_lock
102 * combined with an RCU read-side lock. Insertion and removal are protected
103 * with nl_sk_hash_lock while using RCU list modification primitives and may
104 * run in parallel to nl_table_lock protected lookups. Destruction of the
105 * Netlink socket may only occur *after* nl_table_lock has been acquired
106 * either during or after the socket has been removed from the list.
108 DEFINE_RWLOCK(nl_table_lock
);
109 EXPORT_SYMBOL_GPL(nl_table_lock
);
110 static atomic_t nl_table_users
= ATOMIC_INIT(0);
112 #define nl_deref_protected(X) rcu_dereference_protected(X, lockdep_is_held(&nl_table_lock));
114 /* Protects netlink socket hash table mutations */
115 DEFINE_MUTEX(nl_sk_hash_lock
);
116 EXPORT_SYMBOL_GPL(nl_sk_hash_lock
);
118 #ifdef CONFIG_PROVE_LOCKING
119 static int lockdep_nl_sk_hash_is_held(void *parent
)
122 return lockdep_is_held(&nl_sk_hash_lock
) || lockdep_is_held(&nl_table_lock
);
127 static ATOMIC_NOTIFIER_HEAD(netlink_chain
);
129 static DEFINE_SPINLOCK(netlink_tap_lock
);
130 static struct list_head netlink_tap_all __read_mostly
;
132 static inline u32
netlink_group_mask(u32 group
)
134 return group
? 1 << (group
- 1) : 0;
137 int netlink_add_tap(struct netlink_tap
*nt
)
139 if (unlikely(nt
->dev
->type
!= ARPHRD_NETLINK
))
142 spin_lock(&netlink_tap_lock
);
143 list_add_rcu(&nt
->list
, &netlink_tap_all
);
144 spin_unlock(&netlink_tap_lock
);
146 __module_get(nt
->module
);
150 EXPORT_SYMBOL_GPL(netlink_add_tap
);
152 static int __netlink_remove_tap(struct netlink_tap
*nt
)
155 struct netlink_tap
*tmp
;
157 spin_lock(&netlink_tap_lock
);
159 list_for_each_entry(tmp
, &netlink_tap_all
, list
) {
161 list_del_rcu(&nt
->list
);
167 pr_warn("__netlink_remove_tap: %p not found\n", nt
);
169 spin_unlock(&netlink_tap_lock
);
171 if (found
&& nt
->module
)
172 module_put(nt
->module
);
174 return found
? 0 : -ENODEV
;
177 int netlink_remove_tap(struct netlink_tap
*nt
)
181 ret
= __netlink_remove_tap(nt
);
186 EXPORT_SYMBOL_GPL(netlink_remove_tap
);
188 static bool netlink_filter_tap(const struct sk_buff
*skb
)
190 struct sock
*sk
= skb
->sk
;
192 /* We take the more conservative approach and
193 * whitelist socket protocols that may pass.
195 switch (sk
->sk_protocol
) {
197 case NETLINK_USERSOCK
:
198 case NETLINK_SOCK_DIAG
:
201 case NETLINK_FIB_LOOKUP
:
202 case NETLINK_NETFILTER
:
203 case NETLINK_GENERIC
:
210 static int __netlink_deliver_tap_skb(struct sk_buff
*skb
,
211 struct net_device
*dev
)
213 struct sk_buff
*nskb
;
214 struct sock
*sk
= skb
->sk
;
218 nskb
= skb_clone(skb
, GFP_ATOMIC
);
221 nskb
->protocol
= htons((u16
) sk
->sk_protocol
);
222 nskb
->pkt_type
= netlink_is_kernel(sk
) ?
223 PACKET_KERNEL
: PACKET_USER
;
224 skb_reset_network_header(nskb
);
225 ret
= dev_queue_xmit(nskb
);
226 if (unlikely(ret
> 0))
227 ret
= net_xmit_errno(ret
);
234 static void __netlink_deliver_tap(struct sk_buff
*skb
)
237 struct netlink_tap
*tmp
;
239 if (!netlink_filter_tap(skb
))
242 list_for_each_entry_rcu(tmp
, &netlink_tap_all
, list
) {
243 ret
= __netlink_deliver_tap_skb(skb
, tmp
->dev
);
249 static void netlink_deliver_tap(struct sk_buff
*skb
)
253 if (unlikely(!list_empty(&netlink_tap_all
)))
254 __netlink_deliver_tap(skb
);
259 static void netlink_deliver_tap_kernel(struct sock
*dst
, struct sock
*src
,
262 if (!(netlink_is_kernel(dst
) && netlink_is_kernel(src
)))
263 netlink_deliver_tap(skb
);
266 static void netlink_overrun(struct sock
*sk
)
268 struct netlink_sock
*nlk
= nlk_sk(sk
);
270 if (!(nlk
->flags
& NETLINK_RECV_NO_ENOBUFS
)) {
271 if (!test_and_set_bit(NETLINK_CONGESTED
, &nlk_sk(sk
)->state
)) {
272 sk
->sk_err
= ENOBUFS
;
273 sk
->sk_error_report(sk
);
276 atomic_inc(&sk
->sk_drops
);
279 static void netlink_rcv_wake(struct sock
*sk
)
281 struct netlink_sock
*nlk
= nlk_sk(sk
);
283 if (skb_queue_empty(&sk
->sk_receive_queue
))
284 clear_bit(NETLINK_CONGESTED
, &nlk
->state
);
285 if (!test_bit(NETLINK_CONGESTED
, &nlk
->state
))
286 wake_up_interruptible(&nlk
->wait
);
289 #ifdef CONFIG_NETLINK_MMAP
290 static bool netlink_skb_is_mmaped(const struct sk_buff
*skb
)
292 return NETLINK_CB(skb
).flags
& NETLINK_SKB_MMAPED
;
295 static bool netlink_rx_is_mmaped(struct sock
*sk
)
297 return nlk_sk(sk
)->rx_ring
.pg_vec
!= NULL
;
300 static bool netlink_tx_is_mmaped(struct sock
*sk
)
302 return nlk_sk(sk
)->tx_ring
.pg_vec
!= NULL
;
305 static __pure
struct page
*pgvec_to_page(const void *addr
)
307 if (is_vmalloc_addr(addr
))
308 return vmalloc_to_page(addr
);
310 return virt_to_page(addr
);
313 static void free_pg_vec(void **pg_vec
, unsigned int order
, unsigned int len
)
317 for (i
= 0; i
< len
; i
++) {
318 if (pg_vec
[i
] != NULL
) {
319 if (is_vmalloc_addr(pg_vec
[i
]))
322 free_pages((unsigned long)pg_vec
[i
], order
);
328 static void *alloc_one_pg_vec_page(unsigned long order
)
331 gfp_t gfp_flags
= GFP_KERNEL
| __GFP_COMP
| __GFP_ZERO
|
332 __GFP_NOWARN
| __GFP_NORETRY
;
334 buffer
= (void *)__get_free_pages(gfp_flags
, order
);
338 buffer
= vzalloc((1 << order
) * PAGE_SIZE
);
342 gfp_flags
&= ~__GFP_NORETRY
;
343 return (void *)__get_free_pages(gfp_flags
, order
);
346 static void **alloc_pg_vec(struct netlink_sock
*nlk
,
347 struct nl_mmap_req
*req
, unsigned int order
)
349 unsigned int block_nr
= req
->nm_block_nr
;
353 pg_vec
= kcalloc(block_nr
, sizeof(void *), GFP_KERNEL
);
357 for (i
= 0; i
< block_nr
; i
++) {
358 pg_vec
[i
] = alloc_one_pg_vec_page(order
);
359 if (pg_vec
[i
] == NULL
)
365 free_pg_vec(pg_vec
, order
, block_nr
);
369 static int netlink_set_ring(struct sock
*sk
, struct nl_mmap_req
*req
,
370 bool closing
, bool tx_ring
)
372 struct netlink_sock
*nlk
= nlk_sk(sk
);
373 struct netlink_ring
*ring
;
374 struct sk_buff_head
*queue
;
375 void **pg_vec
= NULL
;
376 unsigned int order
= 0;
379 ring
= tx_ring
? &nlk
->tx_ring
: &nlk
->rx_ring
;
380 queue
= tx_ring
? &sk
->sk_write_queue
: &sk
->sk_receive_queue
;
383 if (atomic_read(&nlk
->mapped
))
385 if (atomic_read(&ring
->pending
))
389 if (req
->nm_block_nr
) {
390 if (ring
->pg_vec
!= NULL
)
393 if ((int)req
->nm_block_size
<= 0)
395 if (!PAGE_ALIGNED(req
->nm_block_size
))
397 if (req
->nm_frame_size
< NL_MMAP_HDRLEN
)
399 if (!IS_ALIGNED(req
->nm_frame_size
, NL_MMAP_MSG_ALIGNMENT
))
402 ring
->frames_per_block
= req
->nm_block_size
/
404 if (ring
->frames_per_block
== 0)
406 if (ring
->frames_per_block
* req
->nm_block_nr
!=
410 order
= get_order(req
->nm_block_size
);
411 pg_vec
= alloc_pg_vec(nlk
, req
, order
);
415 if (req
->nm_frame_nr
)
420 mutex_lock(&nlk
->pg_vec_lock
);
421 if (closing
|| atomic_read(&nlk
->mapped
) == 0) {
423 spin_lock_bh(&queue
->lock
);
425 ring
->frame_max
= req
->nm_frame_nr
- 1;
427 ring
->frame_size
= req
->nm_frame_size
;
428 ring
->pg_vec_pages
= req
->nm_block_size
/ PAGE_SIZE
;
430 swap(ring
->pg_vec_len
, req
->nm_block_nr
);
431 swap(ring
->pg_vec_order
, order
);
432 swap(ring
->pg_vec
, pg_vec
);
434 __skb_queue_purge(queue
);
435 spin_unlock_bh(&queue
->lock
);
437 WARN_ON(atomic_read(&nlk
->mapped
));
439 mutex_unlock(&nlk
->pg_vec_lock
);
442 free_pg_vec(pg_vec
, order
, req
->nm_block_nr
);
446 static void netlink_mm_open(struct vm_area_struct
*vma
)
448 struct file
*file
= vma
->vm_file
;
449 struct socket
*sock
= file
->private_data
;
450 struct sock
*sk
= sock
->sk
;
453 atomic_inc(&nlk_sk(sk
)->mapped
);
456 static void netlink_mm_close(struct vm_area_struct
*vma
)
458 struct file
*file
= vma
->vm_file
;
459 struct socket
*sock
= file
->private_data
;
460 struct sock
*sk
= sock
->sk
;
463 atomic_dec(&nlk_sk(sk
)->mapped
);
466 static const struct vm_operations_struct netlink_mmap_ops
= {
467 .open
= netlink_mm_open
,
468 .close
= netlink_mm_close
,
471 static int netlink_mmap(struct file
*file
, struct socket
*sock
,
472 struct vm_area_struct
*vma
)
474 struct sock
*sk
= sock
->sk
;
475 struct netlink_sock
*nlk
= nlk_sk(sk
);
476 struct netlink_ring
*ring
;
477 unsigned long start
, size
, expected
;
484 mutex_lock(&nlk
->pg_vec_lock
);
487 for (ring
= &nlk
->rx_ring
; ring
<= &nlk
->tx_ring
; ring
++) {
488 if (ring
->pg_vec
== NULL
)
490 expected
+= ring
->pg_vec_len
* ring
->pg_vec_pages
* PAGE_SIZE
;
496 size
= vma
->vm_end
- vma
->vm_start
;
497 if (size
!= expected
)
500 start
= vma
->vm_start
;
501 for (ring
= &nlk
->rx_ring
; ring
<= &nlk
->tx_ring
; ring
++) {
502 if (ring
->pg_vec
== NULL
)
505 for (i
= 0; i
< ring
->pg_vec_len
; i
++) {
507 void *kaddr
= ring
->pg_vec
[i
];
510 for (pg_num
= 0; pg_num
< ring
->pg_vec_pages
; pg_num
++) {
511 page
= pgvec_to_page(kaddr
);
512 err
= vm_insert_page(vma
, start
, page
);
521 atomic_inc(&nlk
->mapped
);
522 vma
->vm_ops
= &netlink_mmap_ops
;
525 mutex_unlock(&nlk
->pg_vec_lock
);
529 static void netlink_frame_flush_dcache(const struct nl_mmap_hdr
*hdr
, unsigned int nm_len
)
531 #if ARCH_IMPLEMENTS_FLUSH_DCACHE_PAGE == 1
532 struct page
*p_start
, *p_end
;
534 /* First page is flushed through netlink_{get,set}_status */
535 p_start
= pgvec_to_page(hdr
+ PAGE_SIZE
);
536 p_end
= pgvec_to_page((void *)hdr
+ NL_MMAP_HDRLEN
+ nm_len
- 1);
537 while (p_start
<= p_end
) {
538 flush_dcache_page(p_start
);
544 static enum nl_mmap_status
netlink_get_status(const struct nl_mmap_hdr
*hdr
)
547 flush_dcache_page(pgvec_to_page(hdr
));
548 return hdr
->nm_status
;
551 static void netlink_set_status(struct nl_mmap_hdr
*hdr
,
552 enum nl_mmap_status status
)
555 hdr
->nm_status
= status
;
556 flush_dcache_page(pgvec_to_page(hdr
));
559 static struct nl_mmap_hdr
*
560 __netlink_lookup_frame(const struct netlink_ring
*ring
, unsigned int pos
)
562 unsigned int pg_vec_pos
, frame_off
;
564 pg_vec_pos
= pos
/ ring
->frames_per_block
;
565 frame_off
= pos
% ring
->frames_per_block
;
567 return ring
->pg_vec
[pg_vec_pos
] + (frame_off
* ring
->frame_size
);
570 static struct nl_mmap_hdr
*
571 netlink_lookup_frame(const struct netlink_ring
*ring
, unsigned int pos
,
572 enum nl_mmap_status status
)
574 struct nl_mmap_hdr
*hdr
;
576 hdr
= __netlink_lookup_frame(ring
, pos
);
577 if (netlink_get_status(hdr
) != status
)
583 static struct nl_mmap_hdr
*
584 netlink_current_frame(const struct netlink_ring
*ring
,
585 enum nl_mmap_status status
)
587 return netlink_lookup_frame(ring
, ring
->head
, status
);
590 static struct nl_mmap_hdr
*
591 netlink_previous_frame(const struct netlink_ring
*ring
,
592 enum nl_mmap_status status
)
596 prev
= ring
->head
? ring
->head
- 1 : ring
->frame_max
;
597 return netlink_lookup_frame(ring
, prev
, status
);
600 static void netlink_increment_head(struct netlink_ring
*ring
)
602 ring
->head
= ring
->head
!= ring
->frame_max
? ring
->head
+ 1 : 0;
605 static void netlink_forward_ring(struct netlink_ring
*ring
)
607 unsigned int head
= ring
->head
, pos
= head
;
608 const struct nl_mmap_hdr
*hdr
;
611 hdr
= __netlink_lookup_frame(ring
, pos
);
612 if (hdr
->nm_status
== NL_MMAP_STATUS_UNUSED
)
614 if (hdr
->nm_status
!= NL_MMAP_STATUS_SKIP
)
616 netlink_increment_head(ring
);
617 } while (ring
->head
!= head
);
620 static bool netlink_dump_space(struct netlink_sock
*nlk
)
622 struct netlink_ring
*ring
= &nlk
->rx_ring
;
623 struct nl_mmap_hdr
*hdr
;
626 hdr
= netlink_current_frame(ring
, NL_MMAP_STATUS_UNUSED
);
630 n
= ring
->head
+ ring
->frame_max
/ 2;
631 if (n
> ring
->frame_max
)
632 n
-= ring
->frame_max
;
634 hdr
= __netlink_lookup_frame(ring
, n
);
636 return hdr
->nm_status
== NL_MMAP_STATUS_UNUSED
;
639 static unsigned int netlink_poll(struct file
*file
, struct socket
*sock
,
642 struct sock
*sk
= sock
->sk
;
643 struct netlink_sock
*nlk
= nlk_sk(sk
);
647 if (nlk
->rx_ring
.pg_vec
!= NULL
) {
648 /* Memory mapped sockets don't call recvmsg(), so flow control
649 * for dumps is performed here. A dump is allowed to continue
650 * if at least half the ring is unused.
652 while (nlk
->cb_running
&& netlink_dump_space(nlk
)) {
653 err
= netlink_dump(sk
);
656 sk
->sk_error_report(sk
);
660 netlink_rcv_wake(sk
);
663 mask
= datagram_poll(file
, sock
, wait
);
665 spin_lock_bh(&sk
->sk_receive_queue
.lock
);
666 if (nlk
->rx_ring
.pg_vec
) {
667 netlink_forward_ring(&nlk
->rx_ring
);
668 if (!netlink_previous_frame(&nlk
->rx_ring
, NL_MMAP_STATUS_UNUSED
))
669 mask
|= POLLIN
| POLLRDNORM
;
671 spin_unlock_bh(&sk
->sk_receive_queue
.lock
);
673 spin_lock_bh(&sk
->sk_write_queue
.lock
);
674 if (nlk
->tx_ring
.pg_vec
) {
675 if (netlink_current_frame(&nlk
->tx_ring
, NL_MMAP_STATUS_UNUSED
))
676 mask
|= POLLOUT
| POLLWRNORM
;
678 spin_unlock_bh(&sk
->sk_write_queue
.lock
);
683 static struct nl_mmap_hdr
*netlink_mmap_hdr(struct sk_buff
*skb
)
685 return (struct nl_mmap_hdr
*)(skb
->head
- NL_MMAP_HDRLEN
);
688 static void netlink_ring_setup_skb(struct sk_buff
*skb
, struct sock
*sk
,
689 struct netlink_ring
*ring
,
690 struct nl_mmap_hdr
*hdr
)
695 size
= ring
->frame_size
- NL_MMAP_HDRLEN
;
696 data
= (void *)hdr
+ NL_MMAP_HDRLEN
;
700 skb_reset_tail_pointer(skb
);
701 skb
->end
= skb
->tail
+ size
;
704 skb
->destructor
= netlink_skb_destructor
;
705 NETLINK_CB(skb
).flags
|= NETLINK_SKB_MMAPED
;
706 NETLINK_CB(skb
).sk
= sk
;
709 static int netlink_mmap_sendmsg(struct sock
*sk
, struct msghdr
*msg
,
710 u32 dst_portid
, u32 dst_group
,
711 struct sock_iocb
*siocb
)
713 struct netlink_sock
*nlk
= nlk_sk(sk
);
714 struct netlink_ring
*ring
;
715 struct nl_mmap_hdr
*hdr
;
718 int err
= 0, len
= 0;
720 mutex_lock(&nlk
->pg_vec_lock
);
722 ring
= &nlk
->tx_ring
;
723 maxlen
= ring
->frame_size
- NL_MMAP_HDRLEN
;
728 hdr
= netlink_current_frame(ring
, NL_MMAP_STATUS_VALID
);
730 if (!(msg
->msg_flags
& MSG_DONTWAIT
) &&
731 atomic_read(&nlk
->tx_ring
.pending
))
736 nm_len
= ACCESS_ONCE(hdr
->nm_len
);
737 if (nm_len
> maxlen
) {
742 netlink_frame_flush_dcache(hdr
, nm_len
);
744 skb
= alloc_skb(nm_len
, GFP_KERNEL
);
749 __skb_put(skb
, nm_len
);
750 memcpy(skb
->data
, (void *)hdr
+ NL_MMAP_HDRLEN
, nm_len
);
751 netlink_set_status(hdr
, NL_MMAP_STATUS_UNUSED
);
753 netlink_increment_head(ring
);
755 NETLINK_CB(skb
).portid
= nlk
->portid
;
756 NETLINK_CB(skb
).dst_group
= dst_group
;
757 NETLINK_CB(skb
).creds
= siocb
->scm
->creds
;
759 err
= security_netlink_send(sk
, skb
);
765 if (unlikely(dst_group
)) {
766 atomic_inc(&skb
->users
);
767 netlink_broadcast(sk
, skb
, dst_portid
, dst_group
,
770 err
= netlink_unicast(sk
, skb
, dst_portid
,
771 msg
->msg_flags
& MSG_DONTWAIT
);
776 } while (hdr
!= NULL
||
777 (!(msg
->msg_flags
& MSG_DONTWAIT
) &&
778 atomic_read(&nlk
->tx_ring
.pending
)));
783 mutex_unlock(&nlk
->pg_vec_lock
);
787 static void netlink_queue_mmaped_skb(struct sock
*sk
, struct sk_buff
*skb
)
789 struct nl_mmap_hdr
*hdr
;
791 hdr
= netlink_mmap_hdr(skb
);
792 hdr
->nm_len
= skb
->len
;
793 hdr
->nm_group
= NETLINK_CB(skb
).dst_group
;
794 hdr
->nm_pid
= NETLINK_CB(skb
).creds
.pid
;
795 hdr
->nm_uid
= from_kuid(sk_user_ns(sk
), NETLINK_CB(skb
).creds
.uid
);
796 hdr
->nm_gid
= from_kgid(sk_user_ns(sk
), NETLINK_CB(skb
).creds
.gid
);
797 netlink_frame_flush_dcache(hdr
, hdr
->nm_len
);
798 netlink_set_status(hdr
, NL_MMAP_STATUS_VALID
);
800 NETLINK_CB(skb
).flags
|= NETLINK_SKB_DELIVERED
;
804 static void netlink_ring_set_copied(struct sock
*sk
, struct sk_buff
*skb
)
806 struct netlink_sock
*nlk
= nlk_sk(sk
);
807 struct netlink_ring
*ring
= &nlk
->rx_ring
;
808 struct nl_mmap_hdr
*hdr
;
810 spin_lock_bh(&sk
->sk_receive_queue
.lock
);
811 hdr
= netlink_current_frame(ring
, NL_MMAP_STATUS_UNUSED
);
813 spin_unlock_bh(&sk
->sk_receive_queue
.lock
);
818 netlink_increment_head(ring
);
819 __skb_queue_tail(&sk
->sk_receive_queue
, skb
);
820 spin_unlock_bh(&sk
->sk_receive_queue
.lock
);
822 hdr
->nm_len
= skb
->len
;
823 hdr
->nm_group
= NETLINK_CB(skb
).dst_group
;
824 hdr
->nm_pid
= NETLINK_CB(skb
).creds
.pid
;
825 hdr
->nm_uid
= from_kuid(sk_user_ns(sk
), NETLINK_CB(skb
).creds
.uid
);
826 hdr
->nm_gid
= from_kgid(sk_user_ns(sk
), NETLINK_CB(skb
).creds
.gid
);
827 netlink_set_status(hdr
, NL_MMAP_STATUS_COPY
);
830 #else /* CONFIG_NETLINK_MMAP */
831 #define netlink_skb_is_mmaped(skb) false
832 #define netlink_rx_is_mmaped(sk) false
833 #define netlink_tx_is_mmaped(sk) false
834 #define netlink_mmap sock_no_mmap
835 #define netlink_poll datagram_poll
836 #define netlink_mmap_sendmsg(sk, msg, dst_portid, dst_group, siocb) 0
837 #endif /* CONFIG_NETLINK_MMAP */
839 static void netlink_skb_destructor(struct sk_buff
*skb
)
841 #ifdef CONFIG_NETLINK_MMAP
842 struct nl_mmap_hdr
*hdr
;
843 struct netlink_ring
*ring
;
846 /* If a packet from the kernel to userspace was freed because of an
847 * error without being delivered to userspace, the kernel must reset
848 * the status. In the direction userspace to kernel, the status is
849 * always reset here after the packet was processed and freed.
851 if (netlink_skb_is_mmaped(skb
)) {
852 hdr
= netlink_mmap_hdr(skb
);
853 sk
= NETLINK_CB(skb
).sk
;
855 if (NETLINK_CB(skb
).flags
& NETLINK_SKB_TX
) {
856 netlink_set_status(hdr
, NL_MMAP_STATUS_UNUSED
);
857 ring
= &nlk_sk(sk
)->tx_ring
;
859 if (!(NETLINK_CB(skb
).flags
& NETLINK_SKB_DELIVERED
)) {
861 netlink_set_status(hdr
, NL_MMAP_STATUS_VALID
);
863 ring
= &nlk_sk(sk
)->rx_ring
;
866 WARN_ON(atomic_read(&ring
->pending
) == 0);
867 atomic_dec(&ring
->pending
);
873 if (is_vmalloc_addr(skb
->head
)) {
875 !atomic_dec_return(&(skb_shinfo(skb
)->dataref
)))
884 static void netlink_skb_set_owner_r(struct sk_buff
*skb
, struct sock
*sk
)
886 WARN_ON(skb
->sk
!= NULL
);
888 skb
->destructor
= netlink_skb_destructor
;
889 atomic_add(skb
->truesize
, &sk
->sk_rmem_alloc
);
890 sk_mem_charge(sk
, skb
->truesize
);
893 static void netlink_sock_destruct(struct sock
*sk
)
895 struct netlink_sock
*nlk
= nlk_sk(sk
);
897 if (nlk
->cb_running
) {
899 nlk
->cb
.done(&nlk
->cb
);
901 module_put(nlk
->cb
.module
);
902 kfree_skb(nlk
->cb
.skb
);
905 skb_queue_purge(&sk
->sk_receive_queue
);
906 #ifdef CONFIG_NETLINK_MMAP
908 struct nl_mmap_req req
;
910 memset(&req
, 0, sizeof(req
));
911 if (nlk
->rx_ring
.pg_vec
)
912 netlink_set_ring(sk
, &req
, true, false);
913 memset(&req
, 0, sizeof(req
));
914 if (nlk
->tx_ring
.pg_vec
)
915 netlink_set_ring(sk
, &req
, true, true);
917 #endif /* CONFIG_NETLINK_MMAP */
919 if (!sock_flag(sk
, SOCK_DEAD
)) {
920 printk(KERN_ERR
"Freeing alive netlink socket %p\n", sk
);
924 WARN_ON(atomic_read(&sk
->sk_rmem_alloc
));
925 WARN_ON(atomic_read(&sk
->sk_wmem_alloc
));
926 WARN_ON(nlk_sk(sk
)->groups
);
929 /* This lock without WQ_FLAG_EXCLUSIVE is good on UP and it is _very_ bad on
930 * SMP. Look, when several writers sleep and reader wakes them up, all but one
931 * immediately hit write lock and grab all the cpus. Exclusive sleep solves
932 * this, _but_ remember, it adds useless work on UP machines.
935 void netlink_table_grab(void)
936 __acquires(nl_table_lock
)
940 write_lock_irq(&nl_table_lock
);
942 if (atomic_read(&nl_table_users
)) {
943 DECLARE_WAITQUEUE(wait
, current
);
945 add_wait_queue_exclusive(&nl_table_wait
, &wait
);
947 set_current_state(TASK_UNINTERRUPTIBLE
);
948 if (atomic_read(&nl_table_users
) == 0)
950 write_unlock_irq(&nl_table_lock
);
952 write_lock_irq(&nl_table_lock
);
955 __set_current_state(TASK_RUNNING
);
956 remove_wait_queue(&nl_table_wait
, &wait
);
960 void netlink_table_ungrab(void)
961 __releases(nl_table_lock
)
963 write_unlock_irq(&nl_table_lock
);
964 wake_up(&nl_table_wait
);
968 netlink_lock_table(void)
970 /* read_lock() synchronizes us to netlink_table_grab */
972 read_lock(&nl_table_lock
);
973 atomic_inc(&nl_table_users
);
974 read_unlock(&nl_table_lock
);
978 netlink_unlock_table(void)
980 if (atomic_dec_and_test(&nl_table_users
))
981 wake_up(&nl_table_wait
);
984 struct netlink_compare_arg
990 static bool netlink_compare(void *ptr
, void *arg
)
992 struct netlink_compare_arg
*x
= arg
;
993 struct sock
*sk
= ptr
;
995 return nlk_sk(sk
)->portid
== x
->portid
&&
996 net_eq(sock_net(sk
), x
->net
);
999 static struct sock
*__netlink_lookup(struct netlink_table
*table
, u32 portid
,
1002 struct netlink_compare_arg arg
= {
1008 hash
= rhashtable_hashfn(&table
->hash
, &portid
, sizeof(portid
));
1010 return rhashtable_lookup_compare(&table
->hash
, hash
,
1011 &netlink_compare
, &arg
);
1014 static struct sock
*netlink_lookup(struct net
*net
, int protocol
, u32 portid
)
1016 struct netlink_table
*table
= &nl_table
[protocol
];
1019 read_lock(&nl_table_lock
);
1021 sk
= __netlink_lookup(table
, portid
, net
);
1025 read_unlock(&nl_table_lock
);
1030 static const struct proto_ops netlink_ops
;
1033 netlink_update_listeners(struct sock
*sk
)
1035 struct netlink_table
*tbl
= &nl_table
[sk
->sk_protocol
];
1038 struct listeners
*listeners
;
1040 listeners
= nl_deref_protected(tbl
->listeners
);
1044 for (i
= 0; i
< NLGRPLONGS(tbl
->groups
); i
++) {
1046 sk_for_each_bound(sk
, &tbl
->mc_list
) {
1047 if (i
< NLGRPLONGS(nlk_sk(sk
)->ngroups
))
1048 mask
|= nlk_sk(sk
)->groups
[i
];
1050 listeners
->masks
[i
] = mask
;
1052 /* this function is only called with the netlink table "grabbed", which
1053 * makes sure updates are visible before bind or setsockopt return. */
1056 static int netlink_insert(struct sock
*sk
, struct net
*net
, u32 portid
)
1058 struct netlink_table
*table
= &nl_table
[sk
->sk_protocol
];
1059 int err
= -EADDRINUSE
;
1061 mutex_lock(&nl_sk_hash_lock
);
1062 if (__netlink_lookup(table
, portid
, net
))
1066 if (nlk_sk(sk
)->portid
)
1070 if (BITS_PER_LONG
> 32 && unlikely(table
->hash
.nelems
>= UINT_MAX
))
1073 nlk_sk(sk
)->portid
= portid
;
1075 rhashtable_insert(&table
->hash
, &nlk_sk(sk
)->node
);
1078 mutex_unlock(&nl_sk_hash_lock
);
1082 static void netlink_remove(struct sock
*sk
)
1084 struct netlink_table
*table
;
1086 mutex_lock(&nl_sk_hash_lock
);
1087 table
= &nl_table
[sk
->sk_protocol
];
1088 if (rhashtable_remove(&table
->hash
, &nlk_sk(sk
)->node
)) {
1089 WARN_ON(atomic_read(&sk
->sk_refcnt
) == 1);
1092 mutex_unlock(&nl_sk_hash_lock
);
1094 netlink_table_grab();
1095 if (nlk_sk(sk
)->subscriptions
) {
1096 __sk_del_bind_node(sk
);
1097 netlink_update_listeners(sk
);
1099 if (sk
->sk_protocol
== NETLINK_GENERIC
)
1100 atomic_inc(&genl_sk_destructing_cnt
);
1101 netlink_table_ungrab();
1104 static struct proto netlink_proto
= {
1106 .owner
= THIS_MODULE
,
1107 .obj_size
= sizeof(struct netlink_sock
),
1110 static int __netlink_create(struct net
*net
, struct socket
*sock
,
1111 struct mutex
*cb_mutex
, int protocol
)
1114 struct netlink_sock
*nlk
;
1116 sock
->ops
= &netlink_ops
;
1118 sk
= sk_alloc(net
, PF_NETLINK
, GFP_KERNEL
, &netlink_proto
);
1122 sock_init_data(sock
, sk
);
1126 nlk
->cb_mutex
= cb_mutex
;
1128 nlk
->cb_mutex
= &nlk
->cb_def_mutex
;
1129 mutex_init(nlk
->cb_mutex
);
1131 init_waitqueue_head(&nlk
->wait
);
1132 #ifdef CONFIG_NETLINK_MMAP
1133 mutex_init(&nlk
->pg_vec_lock
);
1136 sk
->sk_destruct
= netlink_sock_destruct
;
1137 sk
->sk_protocol
= protocol
;
1141 static int netlink_create(struct net
*net
, struct socket
*sock
, int protocol
,
1144 struct module
*module
= NULL
;
1145 struct mutex
*cb_mutex
;
1146 struct netlink_sock
*nlk
;
1147 int (*bind
)(struct net
*net
, int group
);
1148 void (*unbind
)(struct net
*net
, int group
);
1151 sock
->state
= SS_UNCONNECTED
;
1153 if (sock
->type
!= SOCK_RAW
&& sock
->type
!= SOCK_DGRAM
)
1154 return -ESOCKTNOSUPPORT
;
1156 if (protocol
< 0 || protocol
>= MAX_LINKS
)
1157 return -EPROTONOSUPPORT
;
1159 netlink_lock_table();
1160 #ifdef CONFIG_MODULES
1161 if (!nl_table
[protocol
].registered
) {
1162 netlink_unlock_table();
1163 request_module("net-pf-%d-proto-%d", PF_NETLINK
, protocol
);
1164 netlink_lock_table();
1167 if (nl_table
[protocol
].registered
&&
1168 try_module_get(nl_table
[protocol
].module
))
1169 module
= nl_table
[protocol
].module
;
1171 err
= -EPROTONOSUPPORT
;
1172 cb_mutex
= nl_table
[protocol
].cb_mutex
;
1173 bind
= nl_table
[protocol
].bind
;
1174 unbind
= nl_table
[protocol
].unbind
;
1175 netlink_unlock_table();
1180 err
= __netlink_create(net
, sock
, cb_mutex
, protocol
);
1185 sock_prot_inuse_add(net
, &netlink_proto
, 1);
1188 nlk
= nlk_sk(sock
->sk
);
1189 nlk
->module
= module
;
1190 nlk
->netlink_bind
= bind
;
1191 nlk
->netlink_unbind
= unbind
;
1200 static int netlink_release(struct socket
*sock
)
1202 struct sock
*sk
= sock
->sk
;
1203 struct netlink_sock
*nlk
;
1213 * OK. Socket is unlinked, any packets that arrive now
1217 /* must not acquire netlink_table_lock in any way again before unbind
1218 * and notifying genetlink is done as otherwise it might deadlock
1220 if (nlk
->netlink_unbind
) {
1223 for (i
= 0; i
< nlk
->ngroups
; i
++)
1224 if (test_bit(i
, nlk
->groups
))
1225 nlk
->netlink_unbind(sock_net(sk
), i
+ 1);
1227 if (sk
->sk_protocol
== NETLINK_GENERIC
&&
1228 atomic_dec_return(&genl_sk_destructing_cnt
) == 0)
1229 wake_up(&genl_sk_destructing_waitq
);
1232 wake_up_interruptible_all(&nlk
->wait
);
1234 skb_queue_purge(&sk
->sk_write_queue
);
1237 struct netlink_notify n
= {
1238 .net
= sock_net(sk
),
1239 .protocol
= sk
->sk_protocol
,
1240 .portid
= nlk
->portid
,
1242 atomic_notifier_call_chain(&netlink_chain
,
1243 NETLINK_URELEASE
, &n
);
1246 module_put(nlk
->module
);
1248 if (netlink_is_kernel(sk
)) {
1249 netlink_table_grab();
1250 BUG_ON(nl_table
[sk
->sk_protocol
].registered
== 0);
1251 if (--nl_table
[sk
->sk_protocol
].registered
== 0) {
1252 struct listeners
*old
;
1254 old
= nl_deref_protected(nl_table
[sk
->sk_protocol
].listeners
);
1255 RCU_INIT_POINTER(nl_table
[sk
->sk_protocol
].listeners
, NULL
);
1256 kfree_rcu(old
, rcu
);
1257 nl_table
[sk
->sk_protocol
].module
= NULL
;
1258 nl_table
[sk
->sk_protocol
].bind
= NULL
;
1259 nl_table
[sk
->sk_protocol
].unbind
= NULL
;
1260 nl_table
[sk
->sk_protocol
].flags
= 0;
1261 nl_table
[sk
->sk_protocol
].registered
= 0;
1263 netlink_table_ungrab();
1270 sock_prot_inuse_add(sock_net(sk
), &netlink_proto
, -1);
1276 static int netlink_autobind(struct socket
*sock
)
1278 struct sock
*sk
= sock
->sk
;
1279 struct net
*net
= sock_net(sk
);
1280 struct netlink_table
*table
= &nl_table
[sk
->sk_protocol
];
1281 s32 portid
= task_tgid_vnr(current
);
1283 static s32 rover
= -4097;
1287 netlink_table_grab();
1289 if (__netlink_lookup(table
, portid
, net
)) {
1290 /* Bind collision, search negative portid values. */
1295 netlink_table_ungrab();
1299 netlink_table_ungrab();
1301 err
= netlink_insert(sk
, net
, portid
);
1302 if (err
== -EADDRINUSE
)
1305 /* If 2 threads race to autobind, that is fine. */
1313 * __netlink_ns_capable - General netlink message capability test
1314 * @nsp: NETLINK_CB of the socket buffer holding a netlink command from userspace.
1315 * @user_ns: The user namespace of the capability to use
1316 * @cap: The capability to use
1318 * Test to see if the opener of the socket we received the message
1319 * from had when the netlink socket was created and the sender of the
1320 * message has has the capability @cap in the user namespace @user_ns.
1322 bool __netlink_ns_capable(const struct netlink_skb_parms
*nsp
,
1323 struct user_namespace
*user_ns
, int cap
)
1325 return ((nsp
->flags
& NETLINK_SKB_DST
) ||
1326 file_ns_capable(nsp
->sk
->sk_socket
->file
, user_ns
, cap
)) &&
1327 ns_capable(user_ns
, cap
);
1329 EXPORT_SYMBOL(__netlink_ns_capable
);
1332 * netlink_ns_capable - General netlink message capability test
1333 * @skb: socket buffer holding a netlink command from userspace
1334 * @user_ns: The user namespace of the capability to use
1335 * @cap: The capability to use
1337 * Test to see if the opener of the socket we received the message
1338 * from had when the netlink socket was created and the sender of the
1339 * message has has the capability @cap in the user namespace @user_ns.
1341 bool netlink_ns_capable(const struct sk_buff
*skb
,
1342 struct user_namespace
*user_ns
, int cap
)
1344 return __netlink_ns_capable(&NETLINK_CB(skb
), user_ns
, cap
);
1346 EXPORT_SYMBOL(netlink_ns_capable
);
1349 * netlink_capable - Netlink global message capability test
1350 * @skb: socket buffer holding a netlink command from userspace
1351 * @cap: The capability to use
1353 * Test to see if the opener of the socket we received the message
1354 * from had when the netlink socket was created and the sender of the
1355 * message has has the capability @cap in all user namespaces.
1357 bool netlink_capable(const struct sk_buff
*skb
, int cap
)
1359 return netlink_ns_capable(skb
, &init_user_ns
, cap
);
1361 EXPORT_SYMBOL(netlink_capable
);
1364 * netlink_net_capable - Netlink network namespace message capability test
1365 * @skb: socket buffer holding a netlink command from userspace
1366 * @cap: The capability to use
1368 * Test to see if the opener of the socket we received the message
1369 * from had when the netlink socket was created and the sender of the
1370 * message has has the capability @cap over the network namespace of
1371 * the socket we received the message from.
1373 bool netlink_net_capable(const struct sk_buff
*skb
, int cap
)
1375 return netlink_ns_capable(skb
, sock_net(skb
->sk
)->user_ns
, cap
);
1377 EXPORT_SYMBOL(netlink_net_capable
);
1379 static inline int netlink_allowed(const struct socket
*sock
, unsigned int flag
)
1381 return (nl_table
[sock
->sk
->sk_protocol
].flags
& flag
) ||
1382 ns_capable(sock_net(sock
->sk
)->user_ns
, CAP_NET_ADMIN
);
1386 netlink_update_subscriptions(struct sock
*sk
, unsigned int subscriptions
)
1388 struct netlink_sock
*nlk
= nlk_sk(sk
);
1390 if (nlk
->subscriptions
&& !subscriptions
)
1391 __sk_del_bind_node(sk
);
1392 else if (!nlk
->subscriptions
&& subscriptions
)
1393 sk_add_bind_node(sk
, &nl_table
[sk
->sk_protocol
].mc_list
);
1394 nlk
->subscriptions
= subscriptions
;
1397 static int netlink_realloc_groups(struct sock
*sk
)
1399 struct netlink_sock
*nlk
= nlk_sk(sk
);
1400 unsigned int groups
;
1401 unsigned long *new_groups
;
1404 netlink_table_grab();
1406 groups
= nl_table
[sk
->sk_protocol
].groups
;
1407 if (!nl_table
[sk
->sk_protocol
].registered
) {
1412 if (nlk
->ngroups
>= groups
)
1415 new_groups
= krealloc(nlk
->groups
, NLGRPSZ(groups
), GFP_ATOMIC
);
1416 if (new_groups
== NULL
) {
1420 memset((char *)new_groups
+ NLGRPSZ(nlk
->ngroups
), 0,
1421 NLGRPSZ(groups
) - NLGRPSZ(nlk
->ngroups
));
1423 nlk
->groups
= new_groups
;
1424 nlk
->ngroups
= groups
;
1426 netlink_table_ungrab();
1430 static void netlink_undo_bind(int group
, long unsigned int groups
,
1433 struct netlink_sock
*nlk
= nlk_sk(sk
);
1436 if (!nlk
->netlink_unbind
)
1439 for (undo
= 0; undo
< group
; undo
++)
1440 if (test_bit(undo
, &groups
))
1441 nlk
->netlink_unbind(sock_net(sk
), undo
);
1444 static int netlink_bind(struct socket
*sock
, struct sockaddr
*addr
,
1447 struct sock
*sk
= sock
->sk
;
1448 struct net
*net
= sock_net(sk
);
1449 struct netlink_sock
*nlk
= nlk_sk(sk
);
1450 struct sockaddr_nl
*nladdr
= (struct sockaddr_nl
*)addr
;
1452 long unsigned int groups
= nladdr
->nl_groups
;
1454 if (addr_len
< sizeof(struct sockaddr_nl
))
1457 if (nladdr
->nl_family
!= AF_NETLINK
)
1460 /* Only superuser is allowed to listen multicasts */
1462 if (!netlink_allowed(sock
, NL_CFG_F_NONROOT_RECV
))
1464 err
= netlink_realloc_groups(sk
);
1470 if (nladdr
->nl_pid
!= nlk
->portid
)
1473 if (nlk
->netlink_bind
&& groups
) {
1476 for (group
= 0; group
< nlk
->ngroups
; group
++) {
1477 if (!test_bit(group
, &groups
))
1479 err
= nlk
->netlink_bind(net
, group
);
1482 netlink_undo_bind(group
, groups
, sk
);
1488 err
= nladdr
->nl_pid
?
1489 netlink_insert(sk
, net
, nladdr
->nl_pid
) :
1490 netlink_autobind(sock
);
1492 netlink_undo_bind(nlk
->ngroups
, groups
, sk
);
1497 if (!groups
&& (nlk
->groups
== NULL
|| !(u32
)nlk
->groups
[0]))
1500 netlink_table_grab();
1501 netlink_update_subscriptions(sk
, nlk
->subscriptions
+
1503 hweight32(nlk
->groups
[0]));
1504 nlk
->groups
[0] = (nlk
->groups
[0] & ~0xffffffffUL
) | groups
;
1505 netlink_update_listeners(sk
);
1506 netlink_table_ungrab();
1511 static int netlink_connect(struct socket
*sock
, struct sockaddr
*addr
,
1512 int alen
, int flags
)
1515 struct sock
*sk
= sock
->sk
;
1516 struct netlink_sock
*nlk
= nlk_sk(sk
);
1517 struct sockaddr_nl
*nladdr
= (struct sockaddr_nl
*)addr
;
1519 if (alen
< sizeof(addr
->sa_family
))
1522 if (addr
->sa_family
== AF_UNSPEC
) {
1523 sk
->sk_state
= NETLINK_UNCONNECTED
;
1524 nlk
->dst_portid
= 0;
1528 if (addr
->sa_family
!= AF_NETLINK
)
1531 if ((nladdr
->nl_groups
|| nladdr
->nl_pid
) &&
1532 !netlink_allowed(sock
, NL_CFG_F_NONROOT_SEND
))
1536 err
= netlink_autobind(sock
);
1539 sk
->sk_state
= NETLINK_CONNECTED
;
1540 nlk
->dst_portid
= nladdr
->nl_pid
;
1541 nlk
->dst_group
= ffs(nladdr
->nl_groups
);
1547 static int netlink_getname(struct socket
*sock
, struct sockaddr
*addr
,
1548 int *addr_len
, int peer
)
1550 struct sock
*sk
= sock
->sk
;
1551 struct netlink_sock
*nlk
= nlk_sk(sk
);
1552 DECLARE_SOCKADDR(struct sockaddr_nl
*, nladdr
, addr
);
1554 nladdr
->nl_family
= AF_NETLINK
;
1556 *addr_len
= sizeof(*nladdr
);
1559 nladdr
->nl_pid
= nlk
->dst_portid
;
1560 nladdr
->nl_groups
= netlink_group_mask(nlk
->dst_group
);
1562 nladdr
->nl_pid
= nlk
->portid
;
1563 nladdr
->nl_groups
= nlk
->groups
? nlk
->groups
[0] : 0;
1568 static struct sock
*netlink_getsockbyportid(struct sock
*ssk
, u32 portid
)
1571 struct netlink_sock
*nlk
;
1573 sock
= netlink_lookup(sock_net(ssk
), ssk
->sk_protocol
, portid
);
1575 return ERR_PTR(-ECONNREFUSED
);
1577 /* Don't bother queuing skb if kernel socket has no input function */
1579 if (sock
->sk_state
== NETLINK_CONNECTED
&&
1580 nlk
->dst_portid
!= nlk_sk(ssk
)->portid
) {
1582 return ERR_PTR(-ECONNREFUSED
);
1587 struct sock
*netlink_getsockbyfilp(struct file
*filp
)
1589 struct inode
*inode
= file_inode(filp
);
1592 if (!S_ISSOCK(inode
->i_mode
))
1593 return ERR_PTR(-ENOTSOCK
);
1595 sock
= SOCKET_I(inode
)->sk
;
1596 if (sock
->sk_family
!= AF_NETLINK
)
1597 return ERR_PTR(-EINVAL
);
1603 static struct sk_buff
*netlink_alloc_large_skb(unsigned int size
,
1606 struct sk_buff
*skb
;
1609 if (size
<= NLMSG_GOODSIZE
|| broadcast
)
1610 return alloc_skb(size
, GFP_KERNEL
);
1612 size
= SKB_DATA_ALIGN(size
) +
1613 SKB_DATA_ALIGN(sizeof(struct skb_shared_info
));
1615 data
= vmalloc(size
);
1619 skb
= build_skb(data
, size
);
1624 skb
->destructor
= netlink_skb_destructor
;
1631 * Attach a skb to a netlink socket.
1632 * The caller must hold a reference to the destination socket. On error, the
1633 * reference is dropped. The skb is not send to the destination, just all
1634 * all error checks are performed and memory in the queue is reserved.
1636 * < 0: error. skb freed, reference to sock dropped.
1638 * 1: repeat lookup - reference dropped while waiting for socket memory.
1640 int netlink_attachskb(struct sock
*sk
, struct sk_buff
*skb
,
1641 long *timeo
, struct sock
*ssk
)
1643 struct netlink_sock
*nlk
;
1647 if ((atomic_read(&sk
->sk_rmem_alloc
) > sk
->sk_rcvbuf
||
1648 test_bit(NETLINK_CONGESTED
, &nlk
->state
)) &&
1649 !netlink_skb_is_mmaped(skb
)) {
1650 DECLARE_WAITQUEUE(wait
, current
);
1652 if (!ssk
|| netlink_is_kernel(ssk
))
1653 netlink_overrun(sk
);
1659 __set_current_state(TASK_INTERRUPTIBLE
);
1660 add_wait_queue(&nlk
->wait
, &wait
);
1662 if ((atomic_read(&sk
->sk_rmem_alloc
) > sk
->sk_rcvbuf
||
1663 test_bit(NETLINK_CONGESTED
, &nlk
->state
)) &&
1664 !sock_flag(sk
, SOCK_DEAD
))
1665 *timeo
= schedule_timeout(*timeo
);
1667 __set_current_state(TASK_RUNNING
);
1668 remove_wait_queue(&nlk
->wait
, &wait
);
1671 if (signal_pending(current
)) {
1673 return sock_intr_errno(*timeo
);
1677 netlink_skb_set_owner_r(skb
, sk
);
1681 static int __netlink_sendskb(struct sock
*sk
, struct sk_buff
*skb
)
1685 netlink_deliver_tap(skb
);
1687 #ifdef CONFIG_NETLINK_MMAP
1688 if (netlink_skb_is_mmaped(skb
))
1689 netlink_queue_mmaped_skb(sk
, skb
);
1690 else if (netlink_rx_is_mmaped(sk
))
1691 netlink_ring_set_copied(sk
, skb
);
1693 #endif /* CONFIG_NETLINK_MMAP */
1694 skb_queue_tail(&sk
->sk_receive_queue
, skb
);
1695 sk
->sk_data_ready(sk
);
1699 int netlink_sendskb(struct sock
*sk
, struct sk_buff
*skb
)
1701 int len
= __netlink_sendskb(sk
, skb
);
1707 void netlink_detachskb(struct sock
*sk
, struct sk_buff
*skb
)
1713 static struct sk_buff
*netlink_trim(struct sk_buff
*skb
, gfp_t allocation
)
1717 WARN_ON(skb
->sk
!= NULL
);
1718 if (netlink_skb_is_mmaped(skb
))
1721 delta
= skb
->end
- skb
->tail
;
1722 if (is_vmalloc_addr(skb
->head
) || delta
* 2 < skb
->truesize
)
1725 if (skb_shared(skb
)) {
1726 struct sk_buff
*nskb
= skb_clone(skb
, allocation
);
1733 if (!pskb_expand_head(skb
, 0, -delta
, allocation
))
1734 skb
->truesize
-= delta
;
1739 static int netlink_unicast_kernel(struct sock
*sk
, struct sk_buff
*skb
,
1743 struct netlink_sock
*nlk
= nlk_sk(sk
);
1745 ret
= -ECONNREFUSED
;
1746 if (nlk
->netlink_rcv
!= NULL
) {
1748 netlink_skb_set_owner_r(skb
, sk
);
1749 NETLINK_CB(skb
).sk
= ssk
;
1750 netlink_deliver_tap_kernel(sk
, ssk
, skb
);
1751 nlk
->netlink_rcv(skb
);
1760 int netlink_unicast(struct sock
*ssk
, struct sk_buff
*skb
,
1761 u32 portid
, int nonblock
)
1767 skb
= netlink_trim(skb
, gfp_any());
1769 timeo
= sock_sndtimeo(ssk
, nonblock
);
1771 sk
= netlink_getsockbyportid(ssk
, portid
);
1776 if (netlink_is_kernel(sk
))
1777 return netlink_unicast_kernel(sk
, skb
, ssk
);
1779 if (sk_filter(sk
, skb
)) {
1786 err
= netlink_attachskb(sk
, skb
, &timeo
, ssk
);
1792 return netlink_sendskb(sk
, skb
);
1794 EXPORT_SYMBOL(netlink_unicast
);
1796 struct sk_buff
*netlink_alloc_skb(struct sock
*ssk
, unsigned int size
,
1797 u32 dst_portid
, gfp_t gfp_mask
)
1799 #ifdef CONFIG_NETLINK_MMAP
1800 struct sock
*sk
= NULL
;
1801 struct sk_buff
*skb
;
1802 struct netlink_ring
*ring
;
1803 struct nl_mmap_hdr
*hdr
;
1804 unsigned int maxlen
;
1806 sk
= netlink_getsockbyportid(ssk
, dst_portid
);
1810 ring
= &nlk_sk(sk
)->rx_ring
;
1811 /* fast-path without atomic ops for common case: non-mmaped receiver */
1812 if (ring
->pg_vec
== NULL
)
1815 if (ring
->frame_size
- NL_MMAP_HDRLEN
< size
)
1818 skb
= alloc_skb_head(gfp_mask
);
1822 spin_lock_bh(&sk
->sk_receive_queue
.lock
);
1823 /* check again under lock */
1824 if (ring
->pg_vec
== NULL
)
1827 /* check again under lock */
1828 maxlen
= ring
->frame_size
- NL_MMAP_HDRLEN
;
1832 netlink_forward_ring(ring
);
1833 hdr
= netlink_current_frame(ring
, NL_MMAP_STATUS_UNUSED
);
1836 netlink_ring_setup_skb(skb
, sk
, ring
, hdr
);
1837 netlink_set_status(hdr
, NL_MMAP_STATUS_RESERVED
);
1838 atomic_inc(&ring
->pending
);
1839 netlink_increment_head(ring
);
1841 spin_unlock_bh(&sk
->sk_receive_queue
.lock
);
1846 spin_unlock_bh(&sk
->sk_receive_queue
.lock
);
1847 netlink_overrun(sk
);
1854 spin_unlock_bh(&sk
->sk_receive_queue
.lock
);
1859 return alloc_skb(size
, gfp_mask
);
1861 EXPORT_SYMBOL_GPL(netlink_alloc_skb
);
1863 int netlink_has_listeners(struct sock
*sk
, unsigned int group
)
1866 struct listeners
*listeners
;
1868 BUG_ON(!netlink_is_kernel(sk
));
1871 listeners
= rcu_dereference(nl_table
[sk
->sk_protocol
].listeners
);
1873 if (listeners
&& group
- 1 < nl_table
[sk
->sk_protocol
].groups
)
1874 res
= test_bit(group
- 1, listeners
->masks
);
1880 EXPORT_SYMBOL_GPL(netlink_has_listeners
);
1882 static int netlink_broadcast_deliver(struct sock
*sk
, struct sk_buff
*skb
)
1884 struct netlink_sock
*nlk
= nlk_sk(sk
);
1886 if (atomic_read(&sk
->sk_rmem_alloc
) <= sk
->sk_rcvbuf
&&
1887 !test_bit(NETLINK_CONGESTED
, &nlk
->state
)) {
1888 netlink_skb_set_owner_r(skb
, sk
);
1889 __netlink_sendskb(sk
, skb
);
1890 return atomic_read(&sk
->sk_rmem_alloc
) > (sk
->sk_rcvbuf
>> 1);
1895 struct netlink_broadcast_data
{
1896 struct sock
*exclude_sk
;
1901 int delivery_failure
;
1905 struct sk_buff
*skb
, *skb2
;
1906 int (*tx_filter
)(struct sock
*dsk
, struct sk_buff
*skb
, void *data
);
1910 static void do_one_broadcast(struct sock
*sk
,
1911 struct netlink_broadcast_data
*p
)
1913 struct netlink_sock
*nlk
= nlk_sk(sk
);
1916 if (p
->exclude_sk
== sk
)
1919 if (nlk
->portid
== p
->portid
|| p
->group
- 1 >= nlk
->ngroups
||
1920 !test_bit(p
->group
- 1, nlk
->groups
))
1923 if (!net_eq(sock_net(sk
), p
->net
))
1927 netlink_overrun(sk
);
1932 if (p
->skb2
== NULL
) {
1933 if (skb_shared(p
->skb
)) {
1934 p
->skb2
= skb_clone(p
->skb
, p
->allocation
);
1936 p
->skb2
= skb_get(p
->skb
);
1938 * skb ownership may have been set when
1939 * delivered to a previous socket.
1941 skb_orphan(p
->skb2
);
1944 if (p
->skb2
== NULL
) {
1945 netlink_overrun(sk
);
1946 /* Clone failed. Notify ALL listeners. */
1948 if (nlk
->flags
& NETLINK_BROADCAST_SEND_ERROR
)
1949 p
->delivery_failure
= 1;
1950 } else if (p
->tx_filter
&& p
->tx_filter(sk
, p
->skb2
, p
->tx_data
)) {
1953 } else if (sk_filter(sk
, p
->skb2
)) {
1956 } else if ((val
= netlink_broadcast_deliver(sk
, p
->skb2
)) < 0) {
1957 netlink_overrun(sk
);
1958 if (nlk
->flags
& NETLINK_BROADCAST_SEND_ERROR
)
1959 p
->delivery_failure
= 1;
1961 p
->congested
|= val
;
1968 int netlink_broadcast_filtered(struct sock
*ssk
, struct sk_buff
*skb
, u32 portid
,
1969 u32 group
, gfp_t allocation
,
1970 int (*filter
)(struct sock
*dsk
, struct sk_buff
*skb
, void *data
),
1973 struct net
*net
= sock_net(ssk
);
1974 struct netlink_broadcast_data info
;
1977 skb
= netlink_trim(skb
, allocation
);
1979 info
.exclude_sk
= ssk
;
1981 info
.portid
= portid
;
1984 info
.delivery_failure
= 0;
1987 info
.allocation
= allocation
;
1990 info
.tx_filter
= filter
;
1991 info
.tx_data
= filter_data
;
1993 /* While we sleep in clone, do not allow to change socket list */
1995 netlink_lock_table();
1997 sk_for_each_bound(sk
, &nl_table
[ssk
->sk_protocol
].mc_list
)
1998 do_one_broadcast(sk
, &info
);
2002 netlink_unlock_table();
2004 if (info
.delivery_failure
) {
2005 kfree_skb(info
.skb2
);
2008 consume_skb(info
.skb2
);
2010 if (info
.delivered
) {
2011 if (info
.congested
&& (allocation
& __GFP_WAIT
))
2017 EXPORT_SYMBOL(netlink_broadcast_filtered
);
2019 int netlink_broadcast(struct sock
*ssk
, struct sk_buff
*skb
, u32 portid
,
2020 u32 group
, gfp_t allocation
)
2022 return netlink_broadcast_filtered(ssk
, skb
, portid
, group
, allocation
,
2025 EXPORT_SYMBOL(netlink_broadcast
);
2027 struct netlink_set_err_data
{
2028 struct sock
*exclude_sk
;
2034 static int do_one_set_err(struct sock
*sk
, struct netlink_set_err_data
*p
)
2036 struct netlink_sock
*nlk
= nlk_sk(sk
);
2039 if (sk
== p
->exclude_sk
)
2042 if (!net_eq(sock_net(sk
), sock_net(p
->exclude_sk
)))
2045 if (nlk
->portid
== p
->portid
|| p
->group
- 1 >= nlk
->ngroups
||
2046 !test_bit(p
->group
- 1, nlk
->groups
))
2049 if (p
->code
== ENOBUFS
&& nlk
->flags
& NETLINK_RECV_NO_ENOBUFS
) {
2054 sk
->sk_err
= p
->code
;
2055 sk
->sk_error_report(sk
);
2061 * netlink_set_err - report error to broadcast listeners
2062 * @ssk: the kernel netlink socket, as returned by netlink_kernel_create()
2063 * @portid: the PORTID of a process that we want to skip (if any)
2064 * @group: the broadcast group that will notice the error
2065 * @code: error code, must be negative (as usual in kernelspace)
2067 * This function returns the number of broadcast listeners that have set the
2068 * NETLINK_RECV_NO_ENOBUFS socket option.
2070 int netlink_set_err(struct sock
*ssk
, u32 portid
, u32 group
, int code
)
2072 struct netlink_set_err_data info
;
2076 info
.exclude_sk
= ssk
;
2077 info
.portid
= portid
;
2079 /* sk->sk_err wants a positive error value */
2082 read_lock(&nl_table_lock
);
2084 sk_for_each_bound(sk
, &nl_table
[ssk
->sk_protocol
].mc_list
)
2085 ret
+= do_one_set_err(sk
, &info
);
2087 read_unlock(&nl_table_lock
);
2090 EXPORT_SYMBOL(netlink_set_err
);
2092 /* must be called with netlink table grabbed */
2093 static void netlink_update_socket_mc(struct netlink_sock
*nlk
,
2097 int old
, new = !!is_new
, subscriptions
;
2099 old
= test_bit(group
- 1, nlk
->groups
);
2100 subscriptions
= nlk
->subscriptions
- old
+ new;
2102 __set_bit(group
- 1, nlk
->groups
);
2104 __clear_bit(group
- 1, nlk
->groups
);
2105 netlink_update_subscriptions(&nlk
->sk
, subscriptions
);
2106 netlink_update_listeners(&nlk
->sk
);
2109 static int netlink_setsockopt(struct socket
*sock
, int level
, int optname
,
2110 char __user
*optval
, unsigned int optlen
)
2112 struct sock
*sk
= sock
->sk
;
2113 struct netlink_sock
*nlk
= nlk_sk(sk
);
2114 unsigned int val
= 0;
2117 if (level
!= SOL_NETLINK
)
2118 return -ENOPROTOOPT
;
2120 if (optname
!= NETLINK_RX_RING
&& optname
!= NETLINK_TX_RING
&&
2121 optlen
>= sizeof(int) &&
2122 get_user(val
, (unsigned int __user
*)optval
))
2126 case NETLINK_PKTINFO
:
2128 nlk
->flags
|= NETLINK_RECV_PKTINFO
;
2130 nlk
->flags
&= ~NETLINK_RECV_PKTINFO
;
2133 case NETLINK_ADD_MEMBERSHIP
:
2134 case NETLINK_DROP_MEMBERSHIP
: {
2135 if (!netlink_allowed(sock
, NL_CFG_F_NONROOT_RECV
))
2137 err
= netlink_realloc_groups(sk
);
2140 if (!val
|| val
- 1 >= nlk
->ngroups
)
2142 if (optname
== NETLINK_ADD_MEMBERSHIP
&& nlk
->netlink_bind
) {
2143 err
= nlk
->netlink_bind(sock_net(sk
), val
);
2147 netlink_table_grab();
2148 netlink_update_socket_mc(nlk
, val
,
2149 optname
== NETLINK_ADD_MEMBERSHIP
);
2150 netlink_table_ungrab();
2151 if (optname
== NETLINK_DROP_MEMBERSHIP
&& nlk
->netlink_unbind
)
2152 nlk
->netlink_unbind(sock_net(sk
), val
);
2157 case NETLINK_BROADCAST_ERROR
:
2159 nlk
->flags
|= NETLINK_BROADCAST_SEND_ERROR
;
2161 nlk
->flags
&= ~NETLINK_BROADCAST_SEND_ERROR
;
2164 case NETLINK_NO_ENOBUFS
:
2166 nlk
->flags
|= NETLINK_RECV_NO_ENOBUFS
;
2167 clear_bit(NETLINK_CONGESTED
, &nlk
->state
);
2168 wake_up_interruptible(&nlk
->wait
);
2170 nlk
->flags
&= ~NETLINK_RECV_NO_ENOBUFS
;
2174 #ifdef CONFIG_NETLINK_MMAP
2175 case NETLINK_RX_RING
:
2176 case NETLINK_TX_RING
: {
2177 struct nl_mmap_req req
;
2179 /* Rings might consume more memory than queue limits, require
2182 if (!capable(CAP_NET_ADMIN
))
2184 if (optlen
< sizeof(req
))
2186 if (copy_from_user(&req
, optval
, sizeof(req
)))
2188 err
= netlink_set_ring(sk
, &req
, false,
2189 optname
== NETLINK_TX_RING
);
2192 #endif /* CONFIG_NETLINK_MMAP */
2199 static int netlink_getsockopt(struct socket
*sock
, int level
, int optname
,
2200 char __user
*optval
, int __user
*optlen
)
2202 struct sock
*sk
= sock
->sk
;
2203 struct netlink_sock
*nlk
= nlk_sk(sk
);
2206 if (level
!= SOL_NETLINK
)
2207 return -ENOPROTOOPT
;
2209 if (get_user(len
, optlen
))
2215 case NETLINK_PKTINFO
:
2216 if (len
< sizeof(int))
2219 val
= nlk
->flags
& NETLINK_RECV_PKTINFO
? 1 : 0;
2220 if (put_user(len
, optlen
) ||
2221 put_user(val
, optval
))
2225 case NETLINK_BROADCAST_ERROR
:
2226 if (len
< sizeof(int))
2229 val
= nlk
->flags
& NETLINK_BROADCAST_SEND_ERROR
? 1 : 0;
2230 if (put_user(len
, optlen
) ||
2231 put_user(val
, optval
))
2235 case NETLINK_NO_ENOBUFS
:
2236 if (len
< sizeof(int))
2239 val
= nlk
->flags
& NETLINK_RECV_NO_ENOBUFS
? 1 : 0;
2240 if (put_user(len
, optlen
) ||
2241 put_user(val
, optval
))
2251 static void netlink_cmsg_recv_pktinfo(struct msghdr
*msg
, struct sk_buff
*skb
)
2253 struct nl_pktinfo info
;
2255 info
.group
= NETLINK_CB(skb
).dst_group
;
2256 put_cmsg(msg
, SOL_NETLINK
, NETLINK_PKTINFO
, sizeof(info
), &info
);
2259 static int netlink_sendmsg(struct kiocb
*kiocb
, struct socket
*sock
,
2260 struct msghdr
*msg
, size_t len
)
2262 struct sock_iocb
*siocb
= kiocb_to_siocb(kiocb
);
2263 struct sock
*sk
= sock
->sk
;
2264 struct netlink_sock
*nlk
= nlk_sk(sk
);
2265 DECLARE_SOCKADDR(struct sockaddr_nl
*, addr
, msg
->msg_name
);
2268 struct sk_buff
*skb
;
2270 struct scm_cookie scm
;
2271 u32 netlink_skb_flags
= 0;
2273 if (msg
->msg_flags
&MSG_OOB
)
2276 if (NULL
== siocb
->scm
)
2279 err
= scm_send(sock
, msg
, siocb
->scm
, true);
2283 if (msg
->msg_namelen
) {
2285 if (addr
->nl_family
!= AF_NETLINK
)
2287 dst_portid
= addr
->nl_pid
;
2288 dst_group
= ffs(addr
->nl_groups
);
2290 if ((dst_group
|| dst_portid
) &&
2291 !netlink_allowed(sock
, NL_CFG_F_NONROOT_SEND
))
2293 netlink_skb_flags
|= NETLINK_SKB_DST
;
2295 dst_portid
= nlk
->dst_portid
;
2296 dst_group
= nlk
->dst_group
;
2300 err
= netlink_autobind(sock
);
2305 if (netlink_tx_is_mmaped(sk
) &&
2306 msg
->msg_iter
.iov
->iov_base
== NULL
) {
2307 err
= netlink_mmap_sendmsg(sk
, msg
, dst_portid
, dst_group
,
2313 if (len
> sk
->sk_sndbuf
- 32)
2316 skb
= netlink_alloc_large_skb(len
, dst_group
);
2320 NETLINK_CB(skb
).portid
= nlk
->portid
;
2321 NETLINK_CB(skb
).dst_group
= dst_group
;
2322 NETLINK_CB(skb
).creds
= siocb
->scm
->creds
;
2323 NETLINK_CB(skb
).flags
= netlink_skb_flags
;
2326 if (memcpy_from_msg(skb_put(skb
, len
), msg
, len
)) {
2331 err
= security_netlink_send(sk
, skb
);
2338 atomic_inc(&skb
->users
);
2339 netlink_broadcast(sk
, skb
, dst_portid
, dst_group
, GFP_KERNEL
);
2341 err
= netlink_unicast(sk
, skb
, dst_portid
, msg
->msg_flags
&MSG_DONTWAIT
);
2344 scm_destroy(siocb
->scm
);
2348 static int netlink_recvmsg(struct kiocb
*kiocb
, struct socket
*sock
,
2349 struct msghdr
*msg
, size_t len
,
2352 struct sock_iocb
*siocb
= kiocb_to_siocb(kiocb
);
2353 struct scm_cookie scm
;
2354 struct sock
*sk
= sock
->sk
;
2355 struct netlink_sock
*nlk
= nlk_sk(sk
);
2356 int noblock
= flags
&MSG_DONTWAIT
;
2358 struct sk_buff
*skb
, *data_skb
;
2366 skb
= skb_recv_datagram(sk
, flags
, noblock
, &err
);
2372 #ifdef CONFIG_COMPAT_NETLINK_MESSAGES
2373 if (unlikely(skb_shinfo(skb
)->frag_list
)) {
2375 * If this skb has a frag_list, then here that means that we
2376 * will have to use the frag_list skb's data for compat tasks
2377 * and the regular skb's data for normal (non-compat) tasks.
2379 * If we need to send the compat skb, assign it to the
2380 * 'data_skb' variable so that it will be used below for data
2381 * copying. We keep 'skb' for everything else, including
2382 * freeing both later.
2384 if (flags
& MSG_CMSG_COMPAT
)
2385 data_skb
= skb_shinfo(skb
)->frag_list
;
2389 /* Record the max length of recvmsg() calls for future allocations */
2390 nlk
->max_recvmsg_len
= max(nlk
->max_recvmsg_len
, len
);
2391 nlk
->max_recvmsg_len
= min_t(size_t, nlk
->max_recvmsg_len
,
2394 copied
= data_skb
->len
;
2396 msg
->msg_flags
|= MSG_TRUNC
;
2400 skb_reset_transport_header(data_skb
);
2401 err
= skb_copy_datagram_msg(data_skb
, 0, msg
, copied
);
2403 if (msg
->msg_name
) {
2404 DECLARE_SOCKADDR(struct sockaddr_nl
*, addr
, msg
->msg_name
);
2405 addr
->nl_family
= AF_NETLINK
;
2407 addr
->nl_pid
= NETLINK_CB(skb
).portid
;
2408 addr
->nl_groups
= netlink_group_mask(NETLINK_CB(skb
).dst_group
);
2409 msg
->msg_namelen
= sizeof(*addr
);
2412 if (nlk
->flags
& NETLINK_RECV_PKTINFO
)
2413 netlink_cmsg_recv_pktinfo(msg
, skb
);
2415 if (NULL
== siocb
->scm
) {
2416 memset(&scm
, 0, sizeof(scm
));
2419 siocb
->scm
->creds
= *NETLINK_CREDS(skb
);
2420 if (flags
& MSG_TRUNC
)
2421 copied
= data_skb
->len
;
2423 skb_free_datagram(sk
, skb
);
2425 if (nlk
->cb_running
&&
2426 atomic_read(&sk
->sk_rmem_alloc
) <= sk
->sk_rcvbuf
/ 2) {
2427 ret
= netlink_dump(sk
);
2430 sk
->sk_error_report(sk
);
2434 scm_recv(sock
, msg
, siocb
->scm
, flags
);
2436 netlink_rcv_wake(sk
);
2437 return err
? : copied
;
2440 static void netlink_data_ready(struct sock
*sk
)
2446 * We export these functions to other modules. They provide a
2447 * complete set of kernel non-blocking support for message
2452 __netlink_kernel_create(struct net
*net
, int unit
, struct module
*module
,
2453 struct netlink_kernel_cfg
*cfg
)
2455 struct socket
*sock
;
2457 struct netlink_sock
*nlk
;
2458 struct listeners
*listeners
= NULL
;
2459 struct mutex
*cb_mutex
= cfg
? cfg
->cb_mutex
: NULL
;
2460 unsigned int groups
;
2464 if (unit
< 0 || unit
>= MAX_LINKS
)
2467 if (sock_create_lite(PF_NETLINK
, SOCK_DGRAM
, unit
, &sock
))
2471 * We have to just have a reference on the net from sk, but don't
2472 * get_net it. Besides, we cannot get and then put the net here.
2473 * So we create one inside init_net and the move it to net.
2476 if (__netlink_create(&init_net
, sock
, cb_mutex
, unit
) < 0)
2477 goto out_sock_release_nosk
;
2480 sk_change_net(sk
, net
);
2482 if (!cfg
|| cfg
->groups
< 32)
2485 groups
= cfg
->groups
;
2487 listeners
= kzalloc(sizeof(*listeners
) + NLGRPSZ(groups
), GFP_KERNEL
);
2489 goto out_sock_release
;
2491 sk
->sk_data_ready
= netlink_data_ready
;
2492 if (cfg
&& cfg
->input
)
2493 nlk_sk(sk
)->netlink_rcv
= cfg
->input
;
2495 if (netlink_insert(sk
, net
, 0))
2496 goto out_sock_release
;
2499 nlk
->flags
|= NETLINK_KERNEL_SOCKET
;
2501 netlink_table_grab();
2502 if (!nl_table
[unit
].registered
) {
2503 nl_table
[unit
].groups
= groups
;
2504 rcu_assign_pointer(nl_table
[unit
].listeners
, listeners
);
2505 nl_table
[unit
].cb_mutex
= cb_mutex
;
2506 nl_table
[unit
].module
= module
;
2508 nl_table
[unit
].bind
= cfg
->bind
;
2509 nl_table
[unit
].unbind
= cfg
->unbind
;
2510 nl_table
[unit
].flags
= cfg
->flags
;
2512 nl_table
[unit
].compare
= cfg
->compare
;
2514 nl_table
[unit
].registered
= 1;
2517 nl_table
[unit
].registered
++;
2519 netlink_table_ungrab();
2524 netlink_kernel_release(sk
);
2527 out_sock_release_nosk
:
2531 EXPORT_SYMBOL(__netlink_kernel_create
);
2534 netlink_kernel_release(struct sock
*sk
)
2536 sk_release_kernel(sk
);
2538 EXPORT_SYMBOL(netlink_kernel_release
);
2540 int __netlink_change_ngroups(struct sock
*sk
, unsigned int groups
)
2542 struct listeners
*new, *old
;
2543 struct netlink_table
*tbl
= &nl_table
[sk
->sk_protocol
];
2548 if (NLGRPSZ(tbl
->groups
) < NLGRPSZ(groups
)) {
2549 new = kzalloc(sizeof(*new) + NLGRPSZ(groups
), GFP_ATOMIC
);
2552 old
= nl_deref_protected(tbl
->listeners
);
2553 memcpy(new->masks
, old
->masks
, NLGRPSZ(tbl
->groups
));
2554 rcu_assign_pointer(tbl
->listeners
, new);
2556 kfree_rcu(old
, rcu
);
2558 tbl
->groups
= groups
;
2564 * netlink_change_ngroups - change number of multicast groups
2566 * This changes the number of multicast groups that are available
2567 * on a certain netlink family. Note that it is not possible to
2568 * change the number of groups to below 32. Also note that it does
2569 * not implicitly call netlink_clear_multicast_users() when the
2570 * number of groups is reduced.
2572 * @sk: The kernel netlink socket, as returned by netlink_kernel_create().
2573 * @groups: The new number of groups.
2575 int netlink_change_ngroups(struct sock
*sk
, unsigned int groups
)
2579 netlink_table_grab();
2580 err
= __netlink_change_ngroups(sk
, groups
);
2581 netlink_table_ungrab();
2586 void __netlink_clear_multicast_users(struct sock
*ksk
, unsigned int group
)
2589 struct netlink_table
*tbl
= &nl_table
[ksk
->sk_protocol
];
2591 sk_for_each_bound(sk
, &tbl
->mc_list
)
2592 netlink_update_socket_mc(nlk_sk(sk
), group
, 0);
2596 __nlmsg_put(struct sk_buff
*skb
, u32 portid
, u32 seq
, int type
, int len
, int flags
)
2598 struct nlmsghdr
*nlh
;
2599 int size
= nlmsg_msg_size(len
);
2601 nlh
= (struct nlmsghdr
*)skb_put(skb
, NLMSG_ALIGN(size
));
2602 nlh
->nlmsg_type
= type
;
2603 nlh
->nlmsg_len
= size
;
2604 nlh
->nlmsg_flags
= flags
;
2605 nlh
->nlmsg_pid
= portid
;
2606 nlh
->nlmsg_seq
= seq
;
2607 if (!__builtin_constant_p(size
) || NLMSG_ALIGN(size
) - size
!= 0)
2608 memset(nlmsg_data(nlh
) + len
, 0, NLMSG_ALIGN(size
) - size
);
2611 EXPORT_SYMBOL(__nlmsg_put
);
2614 * It looks a bit ugly.
2615 * It would be better to create kernel thread.
2618 static int netlink_dump(struct sock
*sk
)
2620 struct netlink_sock
*nlk
= nlk_sk(sk
);
2621 struct netlink_callback
*cb
;
2622 struct sk_buff
*skb
= NULL
;
2623 struct nlmsghdr
*nlh
;
2624 int len
, err
= -ENOBUFS
;
2627 mutex_lock(nlk
->cb_mutex
);
2628 if (!nlk
->cb_running
) {
2634 alloc_size
= max_t(int, cb
->min_dump_alloc
, NLMSG_GOODSIZE
);
2636 if (!netlink_rx_is_mmaped(sk
) &&
2637 atomic_read(&sk
->sk_rmem_alloc
) >= sk
->sk_rcvbuf
)
2640 /* NLMSG_GOODSIZE is small to avoid high order allocations being
2641 * required, but it makes sense to _attempt_ a 16K bytes allocation
2642 * to reduce number of system calls on dump operations, if user
2643 * ever provided a big enough buffer.
2645 if (alloc_size
< nlk
->max_recvmsg_len
) {
2646 skb
= netlink_alloc_skb(sk
,
2647 nlk
->max_recvmsg_len
,
2652 /* available room should be exact amount to avoid MSG_TRUNC */
2654 skb_reserve(skb
, skb_tailroom(skb
) -
2655 nlk
->max_recvmsg_len
);
2658 skb
= netlink_alloc_skb(sk
, alloc_size
, nlk
->portid
,
2662 netlink_skb_set_owner_r(skb
, sk
);
2664 len
= cb
->dump(skb
, cb
);
2667 mutex_unlock(nlk
->cb_mutex
);
2669 if (sk_filter(sk
, skb
))
2672 __netlink_sendskb(sk
, skb
);
2676 nlh
= nlmsg_put_answer(skb
, cb
, NLMSG_DONE
, sizeof(len
), NLM_F_MULTI
);
2680 nl_dump_check_consistent(cb
, nlh
);
2682 memcpy(nlmsg_data(nlh
), &len
, sizeof(len
));
2684 if (sk_filter(sk
, skb
))
2687 __netlink_sendskb(sk
, skb
);
2692 nlk
->cb_running
= false;
2693 mutex_unlock(nlk
->cb_mutex
);
2694 module_put(cb
->module
);
2695 consume_skb(cb
->skb
);
2699 mutex_unlock(nlk
->cb_mutex
);
2704 int __netlink_dump_start(struct sock
*ssk
, struct sk_buff
*skb
,
2705 const struct nlmsghdr
*nlh
,
2706 struct netlink_dump_control
*control
)
2708 struct netlink_callback
*cb
;
2710 struct netlink_sock
*nlk
;
2713 /* Memory mapped dump requests need to be copied to avoid looping
2714 * on the pending state in netlink_mmap_sendmsg() while the CB hold
2715 * a reference to the skb.
2717 if (netlink_skb_is_mmaped(skb
)) {
2718 skb
= skb_copy(skb
, GFP_KERNEL
);
2722 atomic_inc(&skb
->users
);
2724 sk
= netlink_lookup(sock_net(ssk
), ssk
->sk_protocol
, NETLINK_CB(skb
).portid
);
2726 ret
= -ECONNREFUSED
;
2731 mutex_lock(nlk
->cb_mutex
);
2732 /* A dump is in progress... */
2733 if (nlk
->cb_running
) {
2737 /* add reference of module which cb->dump belongs to */
2738 if (!try_module_get(control
->module
)) {
2739 ret
= -EPROTONOSUPPORT
;
2744 memset(cb
, 0, sizeof(*cb
));
2745 cb
->dump
= control
->dump
;
2746 cb
->done
= control
->done
;
2748 cb
->data
= control
->data
;
2749 cb
->module
= control
->module
;
2750 cb
->min_dump_alloc
= control
->min_dump_alloc
;
2753 nlk
->cb_running
= true;
2755 mutex_unlock(nlk
->cb_mutex
);
2757 ret
= netlink_dump(sk
);
2763 /* We successfully started a dump, by returning -EINTR we
2764 * signal not to send ACK even if it was requested.
2770 mutex_unlock(nlk
->cb_mutex
);
2775 EXPORT_SYMBOL(__netlink_dump_start
);
2777 void netlink_ack(struct sk_buff
*in_skb
, struct nlmsghdr
*nlh
, int err
)
2779 struct sk_buff
*skb
;
2780 struct nlmsghdr
*rep
;
2781 struct nlmsgerr
*errmsg
;
2782 size_t payload
= sizeof(*errmsg
);
2784 /* error messages get the original request appened */
2786 payload
+= nlmsg_len(nlh
);
2788 skb
= netlink_alloc_skb(in_skb
->sk
, nlmsg_total_size(payload
),
2789 NETLINK_CB(in_skb
).portid
, GFP_KERNEL
);
2793 sk
= netlink_lookup(sock_net(in_skb
->sk
),
2794 in_skb
->sk
->sk_protocol
,
2795 NETLINK_CB(in_skb
).portid
);
2797 sk
->sk_err
= ENOBUFS
;
2798 sk
->sk_error_report(sk
);
2804 rep
= __nlmsg_put(skb
, NETLINK_CB(in_skb
).portid
, nlh
->nlmsg_seq
,
2805 NLMSG_ERROR
, payload
, 0);
2806 errmsg
= nlmsg_data(rep
);
2807 errmsg
->error
= err
;
2808 memcpy(&errmsg
->msg
, nlh
, err
? nlh
->nlmsg_len
: sizeof(*nlh
));
2809 netlink_unicast(in_skb
->sk
, skb
, NETLINK_CB(in_skb
).portid
, MSG_DONTWAIT
);
2811 EXPORT_SYMBOL(netlink_ack
);
2813 int netlink_rcv_skb(struct sk_buff
*skb
, int (*cb
)(struct sk_buff
*,
2816 struct nlmsghdr
*nlh
;
2819 while (skb
->len
>= nlmsg_total_size(0)) {
2822 nlh
= nlmsg_hdr(skb
);
2825 if (nlh
->nlmsg_len
< NLMSG_HDRLEN
|| skb
->len
< nlh
->nlmsg_len
)
2828 /* Only requests are handled by the kernel */
2829 if (!(nlh
->nlmsg_flags
& NLM_F_REQUEST
))
2832 /* Skip control messages */
2833 if (nlh
->nlmsg_type
< NLMSG_MIN_TYPE
)
2841 if (nlh
->nlmsg_flags
& NLM_F_ACK
|| err
)
2842 netlink_ack(skb
, nlh
, err
);
2845 msglen
= NLMSG_ALIGN(nlh
->nlmsg_len
);
2846 if (msglen
> skb
->len
)
2848 skb_pull(skb
, msglen
);
2853 EXPORT_SYMBOL(netlink_rcv_skb
);
2856 * nlmsg_notify - send a notification netlink message
2857 * @sk: netlink socket to use
2858 * @skb: notification message
2859 * @portid: destination netlink portid for reports or 0
2860 * @group: destination multicast group or 0
2861 * @report: 1 to report back, 0 to disable
2862 * @flags: allocation flags
2864 int nlmsg_notify(struct sock
*sk
, struct sk_buff
*skb
, u32 portid
,
2865 unsigned int group
, int report
, gfp_t flags
)
2870 int exclude_portid
= 0;
2873 atomic_inc(&skb
->users
);
2874 exclude_portid
= portid
;
2877 /* errors reported via destination sk->sk_err, but propagate
2878 * delivery errors if NETLINK_BROADCAST_ERROR flag is set */
2879 err
= nlmsg_multicast(sk
, skb
, exclude_portid
, group
, flags
);
2885 err2
= nlmsg_unicast(sk
, skb
, portid
);
2886 if (!err
|| err
== -ESRCH
)
2892 EXPORT_SYMBOL(nlmsg_notify
);
2894 #ifdef CONFIG_PROC_FS
2895 struct nl_seq_iter
{
2896 struct seq_net_private p
;
2901 static struct sock
*netlink_seq_socket_idx(struct seq_file
*seq
, loff_t pos
)
2903 struct nl_seq_iter
*iter
= seq
->private;
2905 struct netlink_sock
*nlk
;
2909 for (i
= 0; i
< MAX_LINKS
; i
++) {
2910 struct rhashtable
*ht
= &nl_table
[i
].hash
;
2911 const struct bucket_table
*tbl
= rht_dereference_rcu(ht
->tbl
, ht
);
2913 for (j
= 0; j
< tbl
->size
; j
++) {
2914 rht_for_each_entry_rcu(nlk
, tbl
->buckets
[j
], node
) {
2915 s
= (struct sock
*)nlk
;
2917 if (sock_net(s
) != seq_file_net(seq
))
2931 static void *netlink_seq_start(struct seq_file
*seq
, loff_t
*pos
)
2932 __acquires(nl_table_lock
) __acquires(RCU
)
2934 read_lock(&nl_table_lock
);
2936 return *pos
? netlink_seq_socket_idx(seq
, *pos
- 1) : SEQ_START_TOKEN
;
2939 static void *netlink_seq_next(struct seq_file
*seq
, void *v
, loff_t
*pos
)
2941 struct rhashtable
*ht
;
2942 struct netlink_sock
*nlk
;
2943 struct nl_seq_iter
*iter
;
2949 if (v
== SEQ_START_TOKEN
)
2950 return netlink_seq_socket_idx(seq
, 0);
2952 net
= seq_file_net(seq
);
2953 iter
= seq
->private;
2957 ht
= &nl_table
[i
].hash
;
2958 rht_for_each_entry(nlk
, nlk
->node
.next
, ht
, node
)
2959 if (net_eq(sock_net((struct sock
*)nlk
), net
))
2962 j
= iter
->hash_idx
+ 1;
2965 const struct bucket_table
*tbl
= rht_dereference_rcu(ht
->tbl
, ht
);
2967 for (; j
< tbl
->size
; j
++) {
2968 rht_for_each_entry(nlk
, tbl
->buckets
[j
], ht
, node
) {
2969 if (net_eq(sock_net((struct sock
*)nlk
), net
)) {
2978 } while (++i
< MAX_LINKS
);
2983 static void netlink_seq_stop(struct seq_file
*seq
, void *v
)
2984 __releases(RCU
) __releases(nl_table_lock
)
2987 read_unlock(&nl_table_lock
);
2991 static int netlink_seq_show(struct seq_file
*seq
, void *v
)
2993 if (v
== SEQ_START_TOKEN
) {
2995 "sk Eth Pid Groups "
2996 "Rmem Wmem Dump Locks Drops Inode\n");
2999 struct netlink_sock
*nlk
= nlk_sk(s
);
3001 seq_printf(seq
, "%pK %-3d %-6u %08x %-8d %-8d %d %-8d %-8d %-8lu\n",
3005 nlk
->groups
? (u32
)nlk
->groups
[0] : 0,
3006 sk_rmem_alloc_get(s
),
3007 sk_wmem_alloc_get(s
),
3009 atomic_read(&s
->sk_refcnt
),
3010 atomic_read(&s
->sk_drops
),
3018 static const struct seq_operations netlink_seq_ops
= {
3019 .start
= netlink_seq_start
,
3020 .next
= netlink_seq_next
,
3021 .stop
= netlink_seq_stop
,
3022 .show
= netlink_seq_show
,
3026 static int netlink_seq_open(struct inode
*inode
, struct file
*file
)
3028 return seq_open_net(inode
, file
, &netlink_seq_ops
,
3029 sizeof(struct nl_seq_iter
));
3032 static const struct file_operations netlink_seq_fops
= {
3033 .owner
= THIS_MODULE
,
3034 .open
= netlink_seq_open
,
3036 .llseek
= seq_lseek
,
3037 .release
= seq_release_net
,
3042 int netlink_register_notifier(struct notifier_block
*nb
)
3044 return atomic_notifier_chain_register(&netlink_chain
, nb
);
3046 EXPORT_SYMBOL(netlink_register_notifier
);
3048 int netlink_unregister_notifier(struct notifier_block
*nb
)
3050 return atomic_notifier_chain_unregister(&netlink_chain
, nb
);
3052 EXPORT_SYMBOL(netlink_unregister_notifier
);
3054 static const struct proto_ops netlink_ops
= {
3055 .family
= PF_NETLINK
,
3056 .owner
= THIS_MODULE
,
3057 .release
= netlink_release
,
3058 .bind
= netlink_bind
,
3059 .connect
= netlink_connect
,
3060 .socketpair
= sock_no_socketpair
,
3061 .accept
= sock_no_accept
,
3062 .getname
= netlink_getname
,
3063 .poll
= netlink_poll
,
3064 .ioctl
= sock_no_ioctl
,
3065 .listen
= sock_no_listen
,
3066 .shutdown
= sock_no_shutdown
,
3067 .setsockopt
= netlink_setsockopt
,
3068 .getsockopt
= netlink_getsockopt
,
3069 .sendmsg
= netlink_sendmsg
,
3070 .recvmsg
= netlink_recvmsg
,
3071 .mmap
= netlink_mmap
,
3072 .sendpage
= sock_no_sendpage
,
3075 static const struct net_proto_family netlink_family_ops
= {
3076 .family
= PF_NETLINK
,
3077 .create
= netlink_create
,
3078 .owner
= THIS_MODULE
, /* for consistency 8) */
3081 static int __net_init
netlink_net_init(struct net
*net
)
3083 #ifdef CONFIG_PROC_FS
3084 if (!proc_create("netlink", 0, net
->proc_net
, &netlink_seq_fops
))
3090 static void __net_exit
netlink_net_exit(struct net
*net
)
3092 #ifdef CONFIG_PROC_FS
3093 remove_proc_entry("netlink", net
->proc_net
);
3097 static void __init
netlink_add_usersock_entry(void)
3099 struct listeners
*listeners
;
3102 listeners
= kzalloc(sizeof(*listeners
) + NLGRPSZ(groups
), GFP_KERNEL
);
3104 panic("netlink_add_usersock_entry: Cannot allocate listeners\n");
3106 netlink_table_grab();
3108 nl_table
[NETLINK_USERSOCK
].groups
= groups
;
3109 rcu_assign_pointer(nl_table
[NETLINK_USERSOCK
].listeners
, listeners
);
3110 nl_table
[NETLINK_USERSOCK
].module
= THIS_MODULE
;
3111 nl_table
[NETLINK_USERSOCK
].registered
= 1;
3112 nl_table
[NETLINK_USERSOCK
].flags
= NL_CFG_F_NONROOT_SEND
;
3114 netlink_table_ungrab();
3117 static struct pernet_operations __net_initdata netlink_net_ops
= {
3118 .init
= netlink_net_init
,
3119 .exit
= netlink_net_exit
,
3122 static int __init
netlink_proto_init(void)
3125 int err
= proto_register(&netlink_proto
, 0);
3126 struct rhashtable_params ht_params
= {
3127 .head_offset
= offsetof(struct netlink_sock
, node
),
3128 .key_offset
= offsetof(struct netlink_sock
, portid
),
3129 .key_len
= sizeof(u32
), /* portid */
3131 .max_shift
= 16, /* 64K */
3132 .grow_decision
= rht_grow_above_75
,
3133 .shrink_decision
= rht_shrink_below_30
,
3134 #ifdef CONFIG_PROVE_LOCKING
3135 .mutex_is_held
= lockdep_nl_sk_hash_is_held
,
3142 BUILD_BUG_ON(sizeof(struct netlink_skb_parms
) > FIELD_SIZEOF(struct sk_buff
, cb
));
3144 nl_table
= kcalloc(MAX_LINKS
, sizeof(*nl_table
), GFP_KERNEL
);
3148 for (i
= 0; i
< MAX_LINKS
; i
++) {
3149 if (rhashtable_init(&nl_table
[i
].hash
, &ht_params
) < 0) {
3151 rhashtable_destroy(&nl_table
[i
].hash
);
3157 INIT_LIST_HEAD(&netlink_tap_all
);
3159 netlink_add_usersock_entry();
3161 sock_register(&netlink_family_ops
);
3162 register_pernet_subsys(&netlink_net_ops
);
3163 /* The netlink device handler may be needed early. */
3168 panic("netlink_init: Cannot allocate nl_table\n");
3171 core_initcall(netlink_proto_init
);