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>
65 #include <net/net_namespace.h>
68 #include <net/netlink.h>
70 #include "af_netlink.h"
74 unsigned long masks
[0];
78 #define NETLINK_CONGESTED 0x0
81 #define NETLINK_KERNEL_SOCKET 0x1
82 #define NETLINK_RECV_PKTINFO 0x2
83 #define NETLINK_BROADCAST_SEND_ERROR 0x4
84 #define NETLINK_RECV_NO_ENOBUFS 0x8
86 static inline int netlink_is_kernel(struct sock
*sk
)
88 return nlk_sk(sk
)->flags
& NETLINK_KERNEL_SOCKET
;
91 struct netlink_table
*nl_table
;
92 EXPORT_SYMBOL_GPL(nl_table
);
94 static DECLARE_WAIT_QUEUE_HEAD(nl_table_wait
);
96 static int netlink_dump(struct sock
*sk
);
97 static void netlink_skb_destructor(struct sk_buff
*skb
);
99 /* nl_table locking explained:
100 * Lookup and traversal are protected with nl_sk_hash_lock or nl_table_lock
101 * combined with an RCU read-side lock. Insertion and removal are protected
102 * with nl_sk_hash_lock while using RCU list modification primitives and may
103 * run in parallel to nl_table_lock protected lookups. Destruction of the
104 * Netlink socket may only occur *after* nl_table_lock has been acquired
105 * either during or after the socket has been removed from the list.
107 DEFINE_RWLOCK(nl_table_lock
);
108 EXPORT_SYMBOL_GPL(nl_table_lock
);
109 static atomic_t nl_table_users
= ATOMIC_INIT(0);
111 #define nl_deref_protected(X) rcu_dereference_protected(X, lockdep_is_held(&nl_table_lock));
113 /* Protects netlink socket hash table mutations */
114 DEFINE_MUTEX(nl_sk_hash_lock
);
115 EXPORT_SYMBOL_GPL(nl_sk_hash_lock
);
117 #ifdef CONFIG_PROVE_LOCKING
118 static int lockdep_nl_sk_hash_is_held(void *parent
)
121 return lockdep_is_held(&nl_sk_hash_lock
) || lockdep_is_held(&nl_table_lock
);
126 static ATOMIC_NOTIFIER_HEAD(netlink_chain
);
128 static DEFINE_SPINLOCK(netlink_tap_lock
);
129 static struct list_head netlink_tap_all __read_mostly
;
131 static inline u32
netlink_group_mask(u32 group
)
133 return group
? 1 << (group
- 1) : 0;
136 int netlink_add_tap(struct netlink_tap
*nt
)
138 if (unlikely(nt
->dev
->type
!= ARPHRD_NETLINK
))
141 spin_lock(&netlink_tap_lock
);
142 list_add_rcu(&nt
->list
, &netlink_tap_all
);
143 spin_unlock(&netlink_tap_lock
);
145 __module_get(nt
->module
);
149 EXPORT_SYMBOL_GPL(netlink_add_tap
);
151 static int __netlink_remove_tap(struct netlink_tap
*nt
)
154 struct netlink_tap
*tmp
;
156 spin_lock(&netlink_tap_lock
);
158 list_for_each_entry(tmp
, &netlink_tap_all
, list
) {
160 list_del_rcu(&nt
->list
);
166 pr_warn("__netlink_remove_tap: %p not found\n", nt
);
168 spin_unlock(&netlink_tap_lock
);
170 if (found
&& nt
->module
)
171 module_put(nt
->module
);
173 return found
? 0 : -ENODEV
;
176 int netlink_remove_tap(struct netlink_tap
*nt
)
180 ret
= __netlink_remove_tap(nt
);
185 EXPORT_SYMBOL_GPL(netlink_remove_tap
);
187 static bool netlink_filter_tap(const struct sk_buff
*skb
)
189 struct sock
*sk
= skb
->sk
;
191 /* We take the more conservative approach and
192 * whitelist socket protocols that may pass.
194 switch (sk
->sk_protocol
) {
196 case NETLINK_USERSOCK
:
197 case NETLINK_SOCK_DIAG
:
200 case NETLINK_FIB_LOOKUP
:
201 case NETLINK_NETFILTER
:
202 case NETLINK_GENERIC
:
209 static int __netlink_deliver_tap_skb(struct sk_buff
*skb
,
210 struct net_device
*dev
)
212 struct sk_buff
*nskb
;
213 struct sock
*sk
= skb
->sk
;
217 nskb
= skb_clone(skb
, GFP_ATOMIC
);
220 nskb
->protocol
= htons((u16
) sk
->sk_protocol
);
221 nskb
->pkt_type
= netlink_is_kernel(sk
) ?
222 PACKET_KERNEL
: PACKET_USER
;
223 skb_reset_network_header(nskb
);
224 ret
= dev_queue_xmit(nskb
);
225 if (unlikely(ret
> 0))
226 ret
= net_xmit_errno(ret
);
233 static void __netlink_deliver_tap(struct sk_buff
*skb
)
236 struct netlink_tap
*tmp
;
238 if (!netlink_filter_tap(skb
))
241 list_for_each_entry_rcu(tmp
, &netlink_tap_all
, list
) {
242 ret
= __netlink_deliver_tap_skb(skb
, tmp
->dev
);
248 static void netlink_deliver_tap(struct sk_buff
*skb
)
252 if (unlikely(!list_empty(&netlink_tap_all
)))
253 __netlink_deliver_tap(skb
);
258 static void netlink_deliver_tap_kernel(struct sock
*dst
, struct sock
*src
,
261 if (!(netlink_is_kernel(dst
) && netlink_is_kernel(src
)))
262 netlink_deliver_tap(skb
);
265 static void netlink_overrun(struct sock
*sk
)
267 struct netlink_sock
*nlk
= nlk_sk(sk
);
269 if (!(nlk
->flags
& NETLINK_RECV_NO_ENOBUFS
)) {
270 if (!test_and_set_bit(NETLINK_CONGESTED
, &nlk_sk(sk
)->state
)) {
271 sk
->sk_err
= ENOBUFS
;
272 sk
->sk_error_report(sk
);
275 atomic_inc(&sk
->sk_drops
);
278 static void netlink_rcv_wake(struct sock
*sk
)
280 struct netlink_sock
*nlk
= nlk_sk(sk
);
282 if (skb_queue_empty(&sk
->sk_receive_queue
))
283 clear_bit(NETLINK_CONGESTED
, &nlk
->state
);
284 if (!test_bit(NETLINK_CONGESTED
, &nlk
->state
))
285 wake_up_interruptible(&nlk
->wait
);
288 #ifdef CONFIG_NETLINK_MMAP
289 static bool netlink_skb_is_mmaped(const struct sk_buff
*skb
)
291 return NETLINK_CB(skb
).flags
& NETLINK_SKB_MMAPED
;
294 static bool netlink_rx_is_mmaped(struct sock
*sk
)
296 return nlk_sk(sk
)->rx_ring
.pg_vec
!= NULL
;
299 static bool netlink_tx_is_mmaped(struct sock
*sk
)
301 return nlk_sk(sk
)->tx_ring
.pg_vec
!= NULL
;
304 static __pure
struct page
*pgvec_to_page(const void *addr
)
306 if (is_vmalloc_addr(addr
))
307 return vmalloc_to_page(addr
);
309 return virt_to_page(addr
);
312 static void free_pg_vec(void **pg_vec
, unsigned int order
, unsigned int len
)
316 for (i
= 0; i
< len
; i
++) {
317 if (pg_vec
[i
] != NULL
) {
318 if (is_vmalloc_addr(pg_vec
[i
]))
321 free_pages((unsigned long)pg_vec
[i
], order
);
327 static void *alloc_one_pg_vec_page(unsigned long order
)
330 gfp_t gfp_flags
= GFP_KERNEL
| __GFP_COMP
| __GFP_ZERO
|
331 __GFP_NOWARN
| __GFP_NORETRY
;
333 buffer
= (void *)__get_free_pages(gfp_flags
, order
);
337 buffer
= vzalloc((1 << order
) * PAGE_SIZE
);
341 gfp_flags
&= ~__GFP_NORETRY
;
342 return (void *)__get_free_pages(gfp_flags
, order
);
345 static void **alloc_pg_vec(struct netlink_sock
*nlk
,
346 struct nl_mmap_req
*req
, unsigned int order
)
348 unsigned int block_nr
= req
->nm_block_nr
;
352 pg_vec
= kcalloc(block_nr
, sizeof(void *), GFP_KERNEL
);
356 for (i
= 0; i
< block_nr
; i
++) {
357 pg_vec
[i
] = alloc_one_pg_vec_page(order
);
358 if (pg_vec
[i
] == NULL
)
364 free_pg_vec(pg_vec
, order
, block_nr
);
368 static int netlink_set_ring(struct sock
*sk
, struct nl_mmap_req
*req
,
369 bool closing
, bool tx_ring
)
371 struct netlink_sock
*nlk
= nlk_sk(sk
);
372 struct netlink_ring
*ring
;
373 struct sk_buff_head
*queue
;
374 void **pg_vec
= NULL
;
375 unsigned int order
= 0;
378 ring
= tx_ring
? &nlk
->tx_ring
: &nlk
->rx_ring
;
379 queue
= tx_ring
? &sk
->sk_write_queue
: &sk
->sk_receive_queue
;
382 if (atomic_read(&nlk
->mapped
))
384 if (atomic_read(&ring
->pending
))
388 if (req
->nm_block_nr
) {
389 if (ring
->pg_vec
!= NULL
)
392 if ((int)req
->nm_block_size
<= 0)
394 if (!PAGE_ALIGNED(req
->nm_block_size
))
396 if (req
->nm_frame_size
< NL_MMAP_HDRLEN
)
398 if (!IS_ALIGNED(req
->nm_frame_size
, NL_MMAP_MSG_ALIGNMENT
))
401 ring
->frames_per_block
= req
->nm_block_size
/
403 if (ring
->frames_per_block
== 0)
405 if (ring
->frames_per_block
* req
->nm_block_nr
!=
409 order
= get_order(req
->nm_block_size
);
410 pg_vec
= alloc_pg_vec(nlk
, req
, order
);
414 if (req
->nm_frame_nr
)
419 mutex_lock(&nlk
->pg_vec_lock
);
420 if (closing
|| atomic_read(&nlk
->mapped
) == 0) {
422 spin_lock_bh(&queue
->lock
);
424 ring
->frame_max
= req
->nm_frame_nr
- 1;
426 ring
->frame_size
= req
->nm_frame_size
;
427 ring
->pg_vec_pages
= req
->nm_block_size
/ PAGE_SIZE
;
429 swap(ring
->pg_vec_len
, req
->nm_block_nr
);
430 swap(ring
->pg_vec_order
, order
);
431 swap(ring
->pg_vec
, pg_vec
);
433 __skb_queue_purge(queue
);
434 spin_unlock_bh(&queue
->lock
);
436 WARN_ON(atomic_read(&nlk
->mapped
));
438 mutex_unlock(&nlk
->pg_vec_lock
);
441 free_pg_vec(pg_vec
, order
, req
->nm_block_nr
);
445 static void netlink_mm_open(struct vm_area_struct
*vma
)
447 struct file
*file
= vma
->vm_file
;
448 struct socket
*sock
= file
->private_data
;
449 struct sock
*sk
= sock
->sk
;
452 atomic_inc(&nlk_sk(sk
)->mapped
);
455 static void netlink_mm_close(struct vm_area_struct
*vma
)
457 struct file
*file
= vma
->vm_file
;
458 struct socket
*sock
= file
->private_data
;
459 struct sock
*sk
= sock
->sk
;
462 atomic_dec(&nlk_sk(sk
)->mapped
);
465 static const struct vm_operations_struct netlink_mmap_ops
= {
466 .open
= netlink_mm_open
,
467 .close
= netlink_mm_close
,
470 static int netlink_mmap(struct file
*file
, struct socket
*sock
,
471 struct vm_area_struct
*vma
)
473 struct sock
*sk
= sock
->sk
;
474 struct netlink_sock
*nlk
= nlk_sk(sk
);
475 struct netlink_ring
*ring
;
476 unsigned long start
, size
, expected
;
483 mutex_lock(&nlk
->pg_vec_lock
);
486 for (ring
= &nlk
->rx_ring
; ring
<= &nlk
->tx_ring
; ring
++) {
487 if (ring
->pg_vec
== NULL
)
489 expected
+= ring
->pg_vec_len
* ring
->pg_vec_pages
* PAGE_SIZE
;
495 size
= vma
->vm_end
- vma
->vm_start
;
496 if (size
!= expected
)
499 start
= vma
->vm_start
;
500 for (ring
= &nlk
->rx_ring
; ring
<= &nlk
->tx_ring
; ring
++) {
501 if (ring
->pg_vec
== NULL
)
504 for (i
= 0; i
< ring
->pg_vec_len
; i
++) {
506 void *kaddr
= ring
->pg_vec
[i
];
509 for (pg_num
= 0; pg_num
< ring
->pg_vec_pages
; pg_num
++) {
510 page
= pgvec_to_page(kaddr
);
511 err
= vm_insert_page(vma
, start
, page
);
520 atomic_inc(&nlk
->mapped
);
521 vma
->vm_ops
= &netlink_mmap_ops
;
524 mutex_unlock(&nlk
->pg_vec_lock
);
528 static void netlink_frame_flush_dcache(const struct nl_mmap_hdr
*hdr
, unsigned int nm_len
)
530 #if ARCH_IMPLEMENTS_FLUSH_DCACHE_PAGE == 1
531 struct page
*p_start
, *p_end
;
533 /* First page is flushed through netlink_{get,set}_status */
534 p_start
= pgvec_to_page(hdr
+ PAGE_SIZE
);
535 p_end
= pgvec_to_page((void *)hdr
+ NL_MMAP_HDRLEN
+ nm_len
- 1);
536 while (p_start
<= p_end
) {
537 flush_dcache_page(p_start
);
543 static enum nl_mmap_status
netlink_get_status(const struct nl_mmap_hdr
*hdr
)
546 flush_dcache_page(pgvec_to_page(hdr
));
547 return hdr
->nm_status
;
550 static void netlink_set_status(struct nl_mmap_hdr
*hdr
,
551 enum nl_mmap_status status
)
554 hdr
->nm_status
= status
;
555 flush_dcache_page(pgvec_to_page(hdr
));
558 static struct nl_mmap_hdr
*
559 __netlink_lookup_frame(const struct netlink_ring
*ring
, unsigned int pos
)
561 unsigned int pg_vec_pos
, frame_off
;
563 pg_vec_pos
= pos
/ ring
->frames_per_block
;
564 frame_off
= pos
% ring
->frames_per_block
;
566 return ring
->pg_vec
[pg_vec_pos
] + (frame_off
* ring
->frame_size
);
569 static struct nl_mmap_hdr
*
570 netlink_lookup_frame(const struct netlink_ring
*ring
, unsigned int pos
,
571 enum nl_mmap_status status
)
573 struct nl_mmap_hdr
*hdr
;
575 hdr
= __netlink_lookup_frame(ring
, pos
);
576 if (netlink_get_status(hdr
) != status
)
582 static struct nl_mmap_hdr
*
583 netlink_current_frame(const struct netlink_ring
*ring
,
584 enum nl_mmap_status status
)
586 return netlink_lookup_frame(ring
, ring
->head
, status
);
589 static struct nl_mmap_hdr
*
590 netlink_previous_frame(const struct netlink_ring
*ring
,
591 enum nl_mmap_status status
)
595 prev
= ring
->head
? ring
->head
- 1 : ring
->frame_max
;
596 return netlink_lookup_frame(ring
, prev
, status
);
599 static void netlink_increment_head(struct netlink_ring
*ring
)
601 ring
->head
= ring
->head
!= ring
->frame_max
? ring
->head
+ 1 : 0;
604 static void netlink_forward_ring(struct netlink_ring
*ring
)
606 unsigned int head
= ring
->head
, pos
= head
;
607 const struct nl_mmap_hdr
*hdr
;
610 hdr
= __netlink_lookup_frame(ring
, pos
);
611 if (hdr
->nm_status
== NL_MMAP_STATUS_UNUSED
)
613 if (hdr
->nm_status
!= NL_MMAP_STATUS_SKIP
)
615 netlink_increment_head(ring
);
616 } while (ring
->head
!= head
);
619 static bool netlink_dump_space(struct netlink_sock
*nlk
)
621 struct netlink_ring
*ring
= &nlk
->rx_ring
;
622 struct nl_mmap_hdr
*hdr
;
625 hdr
= netlink_current_frame(ring
, NL_MMAP_STATUS_UNUSED
);
629 n
= ring
->head
+ ring
->frame_max
/ 2;
630 if (n
> ring
->frame_max
)
631 n
-= ring
->frame_max
;
633 hdr
= __netlink_lookup_frame(ring
, n
);
635 return hdr
->nm_status
== NL_MMAP_STATUS_UNUSED
;
638 static unsigned int netlink_poll(struct file
*file
, struct socket
*sock
,
641 struct sock
*sk
= sock
->sk
;
642 struct netlink_sock
*nlk
= nlk_sk(sk
);
646 if (nlk
->rx_ring
.pg_vec
!= NULL
) {
647 /* Memory mapped sockets don't call recvmsg(), so flow control
648 * for dumps is performed here. A dump is allowed to continue
649 * if at least half the ring is unused.
651 while (nlk
->cb_running
&& netlink_dump_space(nlk
)) {
652 err
= netlink_dump(sk
);
655 sk
->sk_error_report(sk
);
659 netlink_rcv_wake(sk
);
662 mask
= datagram_poll(file
, sock
, wait
);
664 spin_lock_bh(&sk
->sk_receive_queue
.lock
);
665 if (nlk
->rx_ring
.pg_vec
) {
666 netlink_forward_ring(&nlk
->rx_ring
);
667 if (!netlink_previous_frame(&nlk
->rx_ring
, NL_MMAP_STATUS_UNUSED
))
668 mask
|= POLLIN
| POLLRDNORM
;
670 spin_unlock_bh(&sk
->sk_receive_queue
.lock
);
672 spin_lock_bh(&sk
->sk_write_queue
.lock
);
673 if (nlk
->tx_ring
.pg_vec
) {
674 if (netlink_current_frame(&nlk
->tx_ring
, NL_MMAP_STATUS_UNUSED
))
675 mask
|= POLLOUT
| POLLWRNORM
;
677 spin_unlock_bh(&sk
->sk_write_queue
.lock
);
682 static struct nl_mmap_hdr
*netlink_mmap_hdr(struct sk_buff
*skb
)
684 return (struct nl_mmap_hdr
*)(skb
->head
- NL_MMAP_HDRLEN
);
687 static void netlink_ring_setup_skb(struct sk_buff
*skb
, struct sock
*sk
,
688 struct netlink_ring
*ring
,
689 struct nl_mmap_hdr
*hdr
)
694 size
= ring
->frame_size
- NL_MMAP_HDRLEN
;
695 data
= (void *)hdr
+ NL_MMAP_HDRLEN
;
699 skb_reset_tail_pointer(skb
);
700 skb
->end
= skb
->tail
+ size
;
703 skb
->destructor
= netlink_skb_destructor
;
704 NETLINK_CB(skb
).flags
|= NETLINK_SKB_MMAPED
;
705 NETLINK_CB(skb
).sk
= sk
;
708 static int netlink_mmap_sendmsg(struct sock
*sk
, struct msghdr
*msg
,
709 u32 dst_portid
, u32 dst_group
,
710 struct sock_iocb
*siocb
)
712 struct netlink_sock
*nlk
= nlk_sk(sk
);
713 struct netlink_ring
*ring
;
714 struct nl_mmap_hdr
*hdr
;
717 int err
= 0, len
= 0;
719 mutex_lock(&nlk
->pg_vec_lock
);
721 ring
= &nlk
->tx_ring
;
722 maxlen
= ring
->frame_size
- NL_MMAP_HDRLEN
;
727 hdr
= netlink_current_frame(ring
, NL_MMAP_STATUS_VALID
);
729 if (!(msg
->msg_flags
& MSG_DONTWAIT
) &&
730 atomic_read(&nlk
->tx_ring
.pending
))
735 nm_len
= ACCESS_ONCE(hdr
->nm_len
);
736 if (nm_len
> maxlen
) {
741 netlink_frame_flush_dcache(hdr
, nm_len
);
743 skb
= alloc_skb(nm_len
, GFP_KERNEL
);
748 __skb_put(skb
, nm_len
);
749 memcpy(skb
->data
, (void *)hdr
+ NL_MMAP_HDRLEN
, nm_len
);
750 netlink_set_status(hdr
, NL_MMAP_STATUS_UNUSED
);
752 netlink_increment_head(ring
);
754 NETLINK_CB(skb
).portid
= nlk
->portid
;
755 NETLINK_CB(skb
).dst_group
= dst_group
;
756 NETLINK_CB(skb
).creds
= siocb
->scm
->creds
;
758 err
= security_netlink_send(sk
, skb
);
764 if (unlikely(dst_group
)) {
765 atomic_inc(&skb
->users
);
766 netlink_broadcast(sk
, skb
, dst_portid
, dst_group
,
769 err
= netlink_unicast(sk
, skb
, dst_portid
,
770 msg
->msg_flags
& MSG_DONTWAIT
);
775 } while (hdr
!= NULL
||
776 (!(msg
->msg_flags
& MSG_DONTWAIT
) &&
777 atomic_read(&nlk
->tx_ring
.pending
)));
782 mutex_unlock(&nlk
->pg_vec_lock
);
786 static void netlink_queue_mmaped_skb(struct sock
*sk
, struct sk_buff
*skb
)
788 struct nl_mmap_hdr
*hdr
;
790 hdr
= netlink_mmap_hdr(skb
);
791 hdr
->nm_len
= skb
->len
;
792 hdr
->nm_group
= NETLINK_CB(skb
).dst_group
;
793 hdr
->nm_pid
= NETLINK_CB(skb
).creds
.pid
;
794 hdr
->nm_uid
= from_kuid(sk_user_ns(sk
), NETLINK_CB(skb
).creds
.uid
);
795 hdr
->nm_gid
= from_kgid(sk_user_ns(sk
), NETLINK_CB(skb
).creds
.gid
);
796 netlink_frame_flush_dcache(hdr
, hdr
->nm_len
);
797 netlink_set_status(hdr
, NL_MMAP_STATUS_VALID
);
799 NETLINK_CB(skb
).flags
|= NETLINK_SKB_DELIVERED
;
803 static void netlink_ring_set_copied(struct sock
*sk
, struct sk_buff
*skb
)
805 struct netlink_sock
*nlk
= nlk_sk(sk
);
806 struct netlink_ring
*ring
= &nlk
->rx_ring
;
807 struct nl_mmap_hdr
*hdr
;
809 spin_lock_bh(&sk
->sk_receive_queue
.lock
);
810 hdr
= netlink_current_frame(ring
, NL_MMAP_STATUS_UNUSED
);
812 spin_unlock_bh(&sk
->sk_receive_queue
.lock
);
817 netlink_increment_head(ring
);
818 __skb_queue_tail(&sk
->sk_receive_queue
, skb
);
819 spin_unlock_bh(&sk
->sk_receive_queue
.lock
);
821 hdr
->nm_len
= skb
->len
;
822 hdr
->nm_group
= NETLINK_CB(skb
).dst_group
;
823 hdr
->nm_pid
= NETLINK_CB(skb
).creds
.pid
;
824 hdr
->nm_uid
= from_kuid(sk_user_ns(sk
), NETLINK_CB(skb
).creds
.uid
);
825 hdr
->nm_gid
= from_kgid(sk_user_ns(sk
), NETLINK_CB(skb
).creds
.gid
);
826 netlink_set_status(hdr
, NL_MMAP_STATUS_COPY
);
829 #else /* CONFIG_NETLINK_MMAP */
830 #define netlink_skb_is_mmaped(skb) false
831 #define netlink_rx_is_mmaped(sk) false
832 #define netlink_tx_is_mmaped(sk) false
833 #define netlink_mmap sock_no_mmap
834 #define netlink_poll datagram_poll
835 #define netlink_mmap_sendmsg(sk, msg, dst_portid, dst_group, siocb) 0
836 #endif /* CONFIG_NETLINK_MMAP */
838 static void netlink_skb_destructor(struct sk_buff
*skb
)
840 #ifdef CONFIG_NETLINK_MMAP
841 struct nl_mmap_hdr
*hdr
;
842 struct netlink_ring
*ring
;
845 /* If a packet from the kernel to userspace was freed because of an
846 * error without being delivered to userspace, the kernel must reset
847 * the status. In the direction userspace to kernel, the status is
848 * always reset here after the packet was processed and freed.
850 if (netlink_skb_is_mmaped(skb
)) {
851 hdr
= netlink_mmap_hdr(skb
);
852 sk
= NETLINK_CB(skb
).sk
;
854 if (NETLINK_CB(skb
).flags
& NETLINK_SKB_TX
) {
855 netlink_set_status(hdr
, NL_MMAP_STATUS_UNUSED
);
856 ring
= &nlk_sk(sk
)->tx_ring
;
858 if (!(NETLINK_CB(skb
).flags
& NETLINK_SKB_DELIVERED
)) {
860 netlink_set_status(hdr
, NL_MMAP_STATUS_VALID
);
862 ring
= &nlk_sk(sk
)->rx_ring
;
865 WARN_ON(atomic_read(&ring
->pending
) == 0);
866 atomic_dec(&ring
->pending
);
872 if (is_vmalloc_addr(skb
->head
)) {
874 !atomic_dec_return(&(skb_shinfo(skb
)->dataref
)))
883 static void netlink_skb_set_owner_r(struct sk_buff
*skb
, struct sock
*sk
)
885 WARN_ON(skb
->sk
!= NULL
);
887 skb
->destructor
= netlink_skb_destructor
;
888 atomic_add(skb
->truesize
, &sk
->sk_rmem_alloc
);
889 sk_mem_charge(sk
, skb
->truesize
);
892 static void netlink_sock_destruct(struct sock
*sk
)
894 struct netlink_sock
*nlk
= nlk_sk(sk
);
896 if (nlk
->cb_running
) {
898 nlk
->cb
.done(&nlk
->cb
);
900 module_put(nlk
->cb
.module
);
901 kfree_skb(nlk
->cb
.skb
);
904 skb_queue_purge(&sk
->sk_receive_queue
);
905 #ifdef CONFIG_NETLINK_MMAP
907 struct nl_mmap_req req
;
909 memset(&req
, 0, sizeof(req
));
910 if (nlk
->rx_ring
.pg_vec
)
911 netlink_set_ring(sk
, &req
, true, false);
912 memset(&req
, 0, sizeof(req
));
913 if (nlk
->tx_ring
.pg_vec
)
914 netlink_set_ring(sk
, &req
, true, true);
916 #endif /* CONFIG_NETLINK_MMAP */
918 if (!sock_flag(sk
, SOCK_DEAD
)) {
919 printk(KERN_ERR
"Freeing alive netlink socket %p\n", sk
);
923 WARN_ON(atomic_read(&sk
->sk_rmem_alloc
));
924 WARN_ON(atomic_read(&sk
->sk_wmem_alloc
));
925 WARN_ON(nlk_sk(sk
)->groups
);
928 /* This lock without WQ_FLAG_EXCLUSIVE is good on UP and it is _very_ bad on
929 * SMP. Look, when several writers sleep and reader wakes them up, all but one
930 * immediately hit write lock and grab all the cpus. Exclusive sleep solves
931 * this, _but_ remember, it adds useless work on UP machines.
934 void netlink_table_grab(void)
935 __acquires(nl_table_lock
)
939 write_lock_irq(&nl_table_lock
);
941 if (atomic_read(&nl_table_users
)) {
942 DECLARE_WAITQUEUE(wait
, current
);
944 add_wait_queue_exclusive(&nl_table_wait
, &wait
);
946 set_current_state(TASK_UNINTERRUPTIBLE
);
947 if (atomic_read(&nl_table_users
) == 0)
949 write_unlock_irq(&nl_table_lock
);
951 write_lock_irq(&nl_table_lock
);
954 __set_current_state(TASK_RUNNING
);
955 remove_wait_queue(&nl_table_wait
, &wait
);
959 void netlink_table_ungrab(void)
960 __releases(nl_table_lock
)
962 write_unlock_irq(&nl_table_lock
);
963 wake_up(&nl_table_wait
);
967 netlink_lock_table(void)
969 /* read_lock() synchronizes us to netlink_table_grab */
971 read_lock(&nl_table_lock
);
972 atomic_inc(&nl_table_users
);
973 read_unlock(&nl_table_lock
);
977 netlink_unlock_table(void)
979 if (atomic_dec_and_test(&nl_table_users
))
980 wake_up(&nl_table_wait
);
983 struct netlink_compare_arg
989 static bool netlink_compare(void *ptr
, void *arg
)
991 struct netlink_compare_arg
*x
= arg
;
992 struct sock
*sk
= ptr
;
994 return nlk_sk(sk
)->portid
== x
->portid
&&
995 net_eq(sock_net(sk
), x
->net
);
998 static struct sock
*__netlink_lookup(struct netlink_table
*table
, u32 portid
,
1001 struct netlink_compare_arg arg
= {
1006 return rhashtable_lookup_compare(&table
->hash
, &portid
,
1007 &netlink_compare
, &arg
);
1010 static struct sock
*netlink_lookup(struct net
*net
, int protocol
, u32 portid
)
1012 struct netlink_table
*table
= &nl_table
[protocol
];
1015 read_lock(&nl_table_lock
);
1017 sk
= __netlink_lookup(table
, portid
, net
);
1021 read_unlock(&nl_table_lock
);
1026 static const struct proto_ops netlink_ops
;
1029 netlink_update_listeners(struct sock
*sk
)
1031 struct netlink_table
*tbl
= &nl_table
[sk
->sk_protocol
];
1034 struct listeners
*listeners
;
1036 listeners
= nl_deref_protected(tbl
->listeners
);
1040 for (i
= 0; i
< NLGRPLONGS(tbl
->groups
); i
++) {
1042 sk_for_each_bound(sk
, &tbl
->mc_list
) {
1043 if (i
< NLGRPLONGS(nlk_sk(sk
)->ngroups
))
1044 mask
|= nlk_sk(sk
)->groups
[i
];
1046 listeners
->masks
[i
] = mask
;
1048 /* this function is only called with the netlink table "grabbed", which
1049 * makes sure updates are visible before bind or setsockopt return. */
1052 static int netlink_insert(struct sock
*sk
, struct net
*net
, u32 portid
)
1054 struct netlink_table
*table
= &nl_table
[sk
->sk_protocol
];
1055 int err
= -EADDRINUSE
;
1057 mutex_lock(&nl_sk_hash_lock
);
1058 if (__netlink_lookup(table
, portid
, net
))
1062 if (nlk_sk(sk
)->portid
)
1066 if (BITS_PER_LONG
> 32 && unlikely(table
->hash
.nelems
>= UINT_MAX
))
1069 nlk_sk(sk
)->portid
= portid
;
1071 rhashtable_insert(&table
->hash
, &nlk_sk(sk
)->node
);
1074 mutex_unlock(&nl_sk_hash_lock
);
1078 static void netlink_remove(struct sock
*sk
)
1080 struct netlink_table
*table
;
1082 mutex_lock(&nl_sk_hash_lock
);
1083 table
= &nl_table
[sk
->sk_protocol
];
1084 if (rhashtable_remove(&table
->hash
, &nlk_sk(sk
)->node
)) {
1085 WARN_ON(atomic_read(&sk
->sk_refcnt
) == 1);
1088 mutex_unlock(&nl_sk_hash_lock
);
1090 netlink_table_grab();
1091 if (nlk_sk(sk
)->subscriptions
) {
1092 __sk_del_bind_node(sk
);
1093 netlink_update_listeners(sk
);
1095 netlink_table_ungrab();
1098 static struct proto netlink_proto
= {
1100 .owner
= THIS_MODULE
,
1101 .obj_size
= sizeof(struct netlink_sock
),
1104 static int __netlink_create(struct net
*net
, struct socket
*sock
,
1105 struct mutex
*cb_mutex
, int protocol
)
1108 struct netlink_sock
*nlk
;
1110 sock
->ops
= &netlink_ops
;
1112 sk
= sk_alloc(net
, PF_NETLINK
, GFP_KERNEL
, &netlink_proto
);
1116 sock_init_data(sock
, sk
);
1120 nlk
->cb_mutex
= cb_mutex
;
1122 nlk
->cb_mutex
= &nlk
->cb_def_mutex
;
1123 mutex_init(nlk
->cb_mutex
);
1125 init_waitqueue_head(&nlk
->wait
);
1126 #ifdef CONFIG_NETLINK_MMAP
1127 mutex_init(&nlk
->pg_vec_lock
);
1130 sk
->sk_destruct
= netlink_sock_destruct
;
1131 sk
->sk_protocol
= protocol
;
1135 static int netlink_create(struct net
*net
, struct socket
*sock
, int protocol
,
1138 struct module
*module
= NULL
;
1139 struct mutex
*cb_mutex
;
1140 struct netlink_sock
*nlk
;
1141 int (*bind
)(struct net
*net
, int group
);
1142 void (*unbind
)(struct net
*net
, int group
);
1145 sock
->state
= SS_UNCONNECTED
;
1147 if (sock
->type
!= SOCK_RAW
&& sock
->type
!= SOCK_DGRAM
)
1148 return -ESOCKTNOSUPPORT
;
1150 if (protocol
< 0 || protocol
>= MAX_LINKS
)
1151 return -EPROTONOSUPPORT
;
1153 netlink_lock_table();
1154 #ifdef CONFIG_MODULES
1155 if (!nl_table
[protocol
].registered
) {
1156 netlink_unlock_table();
1157 request_module("net-pf-%d-proto-%d", PF_NETLINK
, protocol
);
1158 netlink_lock_table();
1161 if (nl_table
[protocol
].registered
&&
1162 try_module_get(nl_table
[protocol
].module
))
1163 module
= nl_table
[protocol
].module
;
1165 err
= -EPROTONOSUPPORT
;
1166 cb_mutex
= nl_table
[protocol
].cb_mutex
;
1167 bind
= nl_table
[protocol
].bind
;
1168 unbind
= nl_table
[protocol
].unbind
;
1169 netlink_unlock_table();
1174 err
= __netlink_create(net
, sock
, cb_mutex
, protocol
);
1179 sock_prot_inuse_add(net
, &netlink_proto
, 1);
1182 nlk
= nlk_sk(sock
->sk
);
1183 nlk
->module
= module
;
1184 nlk
->netlink_bind
= bind
;
1185 nlk
->netlink_unbind
= unbind
;
1194 static int netlink_release(struct socket
*sock
)
1196 struct sock
*sk
= sock
->sk
;
1197 struct netlink_sock
*nlk
;
1207 * OK. Socket is unlinked, any packets that arrive now
1212 wake_up_interruptible_all(&nlk
->wait
);
1214 skb_queue_purge(&sk
->sk_write_queue
);
1217 struct netlink_notify n
= {
1218 .net
= sock_net(sk
),
1219 .protocol
= sk
->sk_protocol
,
1220 .portid
= nlk
->portid
,
1222 atomic_notifier_call_chain(&netlink_chain
,
1223 NETLINK_URELEASE
, &n
);
1226 module_put(nlk
->module
);
1228 if (netlink_is_kernel(sk
)) {
1229 netlink_table_grab();
1230 BUG_ON(nl_table
[sk
->sk_protocol
].registered
== 0);
1231 if (--nl_table
[sk
->sk_protocol
].registered
== 0) {
1232 struct listeners
*old
;
1234 old
= nl_deref_protected(nl_table
[sk
->sk_protocol
].listeners
);
1235 RCU_INIT_POINTER(nl_table
[sk
->sk_protocol
].listeners
, NULL
);
1236 kfree_rcu(old
, rcu
);
1237 nl_table
[sk
->sk_protocol
].module
= NULL
;
1238 nl_table
[sk
->sk_protocol
].bind
= NULL
;
1239 nl_table
[sk
->sk_protocol
].unbind
= NULL
;
1240 nl_table
[sk
->sk_protocol
].flags
= 0;
1241 nl_table
[sk
->sk_protocol
].registered
= 0;
1243 netlink_table_ungrab();
1246 if (nlk
->netlink_unbind
) {
1249 for (i
= 0; i
< nlk
->ngroups
; i
++)
1250 if (test_bit(i
, nlk
->groups
))
1251 nlk
->netlink_unbind(sock_net(sk
), i
+ 1);
1257 sock_prot_inuse_add(sock_net(sk
), &netlink_proto
, -1);
1263 static int netlink_autobind(struct socket
*sock
)
1265 struct sock
*sk
= sock
->sk
;
1266 struct net
*net
= sock_net(sk
);
1267 struct netlink_table
*table
= &nl_table
[sk
->sk_protocol
];
1268 s32 portid
= task_tgid_vnr(current
);
1270 static s32 rover
= -4097;
1274 netlink_table_grab();
1276 if (__netlink_lookup(table
, portid
, net
)) {
1277 /* Bind collision, search negative portid values. */
1282 netlink_table_ungrab();
1286 netlink_table_ungrab();
1288 err
= netlink_insert(sk
, net
, portid
);
1289 if (err
== -EADDRINUSE
)
1292 /* If 2 threads race to autobind, that is fine. */
1300 * __netlink_ns_capable - General netlink message capability test
1301 * @nsp: NETLINK_CB of the socket buffer holding a netlink command from userspace.
1302 * @user_ns: The user namespace of the capability to use
1303 * @cap: The capability to use
1305 * Test to see if the opener of the socket we received the message
1306 * from had when the netlink socket was created and the sender of the
1307 * message has has the capability @cap in the user namespace @user_ns.
1309 bool __netlink_ns_capable(const struct netlink_skb_parms
*nsp
,
1310 struct user_namespace
*user_ns
, int cap
)
1312 return ((nsp
->flags
& NETLINK_SKB_DST
) ||
1313 file_ns_capable(nsp
->sk
->sk_socket
->file
, user_ns
, cap
)) &&
1314 ns_capable(user_ns
, cap
);
1316 EXPORT_SYMBOL(__netlink_ns_capable
);
1319 * netlink_ns_capable - General netlink message capability test
1320 * @skb: socket buffer holding a netlink command from userspace
1321 * @user_ns: The user namespace of the capability to use
1322 * @cap: The capability to use
1324 * Test to see if the opener of the socket we received the message
1325 * from had when the netlink socket was created and the sender of the
1326 * message has has the capability @cap in the user namespace @user_ns.
1328 bool netlink_ns_capable(const struct sk_buff
*skb
,
1329 struct user_namespace
*user_ns
, int cap
)
1331 return __netlink_ns_capable(&NETLINK_CB(skb
), user_ns
, cap
);
1333 EXPORT_SYMBOL(netlink_ns_capable
);
1336 * netlink_capable - Netlink global message capability test
1337 * @skb: socket buffer holding a netlink command from userspace
1338 * @cap: The capability to use
1340 * Test to see if the opener of the socket we received the message
1341 * from had when the netlink socket was created and the sender of the
1342 * message has has the capability @cap in all user namespaces.
1344 bool netlink_capable(const struct sk_buff
*skb
, int cap
)
1346 return netlink_ns_capable(skb
, &init_user_ns
, cap
);
1348 EXPORT_SYMBOL(netlink_capable
);
1351 * netlink_net_capable - Netlink network namespace message capability test
1352 * @skb: socket buffer holding a netlink command from userspace
1353 * @cap: The capability to use
1355 * Test to see if the opener of the socket we received the message
1356 * from had when the netlink socket was created and the sender of the
1357 * message has has the capability @cap over the network namespace of
1358 * the socket we received the message from.
1360 bool netlink_net_capable(const struct sk_buff
*skb
, int cap
)
1362 return netlink_ns_capable(skb
, sock_net(skb
->sk
)->user_ns
, cap
);
1364 EXPORT_SYMBOL(netlink_net_capable
);
1366 static inline int netlink_allowed(const struct socket
*sock
, unsigned int flag
)
1368 return (nl_table
[sock
->sk
->sk_protocol
].flags
& flag
) ||
1369 ns_capable(sock_net(sock
->sk
)->user_ns
, CAP_NET_ADMIN
);
1373 netlink_update_subscriptions(struct sock
*sk
, unsigned int subscriptions
)
1375 struct netlink_sock
*nlk
= nlk_sk(sk
);
1377 if (nlk
->subscriptions
&& !subscriptions
)
1378 __sk_del_bind_node(sk
);
1379 else if (!nlk
->subscriptions
&& subscriptions
)
1380 sk_add_bind_node(sk
, &nl_table
[sk
->sk_protocol
].mc_list
);
1381 nlk
->subscriptions
= subscriptions
;
1384 static int netlink_realloc_groups(struct sock
*sk
)
1386 struct netlink_sock
*nlk
= nlk_sk(sk
);
1387 unsigned int groups
;
1388 unsigned long *new_groups
;
1391 netlink_table_grab();
1393 groups
= nl_table
[sk
->sk_protocol
].groups
;
1394 if (!nl_table
[sk
->sk_protocol
].registered
) {
1399 if (nlk
->ngroups
>= groups
)
1402 new_groups
= krealloc(nlk
->groups
, NLGRPSZ(groups
), GFP_ATOMIC
);
1403 if (new_groups
== NULL
) {
1407 memset((char *)new_groups
+ NLGRPSZ(nlk
->ngroups
), 0,
1408 NLGRPSZ(groups
) - NLGRPSZ(nlk
->ngroups
));
1410 nlk
->groups
= new_groups
;
1411 nlk
->ngroups
= groups
;
1413 netlink_table_ungrab();
1417 static void netlink_undo_bind(int group
, long unsigned int groups
,
1420 struct netlink_sock
*nlk
= nlk_sk(sk
);
1423 if (!nlk
->netlink_unbind
)
1426 for (undo
= 0; undo
< group
; undo
++)
1427 if (test_bit(undo
, &groups
))
1428 nlk
->netlink_unbind(sock_net(sk
), undo
);
1431 static int netlink_bind(struct socket
*sock
, struct sockaddr
*addr
,
1434 struct sock
*sk
= sock
->sk
;
1435 struct net
*net
= sock_net(sk
);
1436 struct netlink_sock
*nlk
= nlk_sk(sk
);
1437 struct sockaddr_nl
*nladdr
= (struct sockaddr_nl
*)addr
;
1439 long unsigned int groups
= nladdr
->nl_groups
;
1441 if (addr_len
< sizeof(struct sockaddr_nl
))
1444 if (nladdr
->nl_family
!= AF_NETLINK
)
1447 /* Only superuser is allowed to listen multicasts */
1449 if (!netlink_allowed(sock
, NL_CFG_F_NONROOT_RECV
))
1451 err
= netlink_realloc_groups(sk
);
1457 if (nladdr
->nl_pid
!= nlk
->portid
)
1460 if (nlk
->netlink_bind
&& groups
) {
1463 for (group
= 0; group
< nlk
->ngroups
; group
++) {
1464 if (!test_bit(group
, &groups
))
1466 err
= nlk
->netlink_bind(net
, group
);
1469 netlink_undo_bind(group
, groups
, sk
);
1475 err
= nladdr
->nl_pid
?
1476 netlink_insert(sk
, net
, nladdr
->nl_pid
) :
1477 netlink_autobind(sock
);
1479 netlink_undo_bind(nlk
->ngroups
, groups
, sk
);
1484 if (!groups
&& (nlk
->groups
== NULL
|| !(u32
)nlk
->groups
[0]))
1487 netlink_table_grab();
1488 netlink_update_subscriptions(sk
, nlk
->subscriptions
+
1490 hweight32(nlk
->groups
[0]));
1491 nlk
->groups
[0] = (nlk
->groups
[0] & ~0xffffffffUL
) | groups
;
1492 netlink_update_listeners(sk
);
1493 netlink_table_ungrab();
1498 static int netlink_connect(struct socket
*sock
, struct sockaddr
*addr
,
1499 int alen
, int flags
)
1502 struct sock
*sk
= sock
->sk
;
1503 struct netlink_sock
*nlk
= nlk_sk(sk
);
1504 struct sockaddr_nl
*nladdr
= (struct sockaddr_nl
*)addr
;
1506 if (alen
< sizeof(addr
->sa_family
))
1509 if (addr
->sa_family
== AF_UNSPEC
) {
1510 sk
->sk_state
= NETLINK_UNCONNECTED
;
1511 nlk
->dst_portid
= 0;
1515 if (addr
->sa_family
!= AF_NETLINK
)
1518 if ((nladdr
->nl_groups
|| nladdr
->nl_pid
) &&
1519 !netlink_allowed(sock
, NL_CFG_F_NONROOT_SEND
))
1523 err
= netlink_autobind(sock
);
1526 sk
->sk_state
= NETLINK_CONNECTED
;
1527 nlk
->dst_portid
= nladdr
->nl_pid
;
1528 nlk
->dst_group
= ffs(nladdr
->nl_groups
);
1534 static int netlink_getname(struct socket
*sock
, struct sockaddr
*addr
,
1535 int *addr_len
, int peer
)
1537 struct sock
*sk
= sock
->sk
;
1538 struct netlink_sock
*nlk
= nlk_sk(sk
);
1539 DECLARE_SOCKADDR(struct sockaddr_nl
*, nladdr
, addr
);
1541 nladdr
->nl_family
= AF_NETLINK
;
1543 *addr_len
= sizeof(*nladdr
);
1546 nladdr
->nl_pid
= nlk
->dst_portid
;
1547 nladdr
->nl_groups
= netlink_group_mask(nlk
->dst_group
);
1549 nladdr
->nl_pid
= nlk
->portid
;
1550 nladdr
->nl_groups
= nlk
->groups
? nlk
->groups
[0] : 0;
1555 static struct sock
*netlink_getsockbyportid(struct sock
*ssk
, u32 portid
)
1558 struct netlink_sock
*nlk
;
1560 sock
= netlink_lookup(sock_net(ssk
), ssk
->sk_protocol
, portid
);
1562 return ERR_PTR(-ECONNREFUSED
);
1564 /* Don't bother queuing skb if kernel socket has no input function */
1566 if (sock
->sk_state
== NETLINK_CONNECTED
&&
1567 nlk
->dst_portid
!= nlk_sk(ssk
)->portid
) {
1569 return ERR_PTR(-ECONNREFUSED
);
1574 struct sock
*netlink_getsockbyfilp(struct file
*filp
)
1576 struct inode
*inode
= file_inode(filp
);
1579 if (!S_ISSOCK(inode
->i_mode
))
1580 return ERR_PTR(-ENOTSOCK
);
1582 sock
= SOCKET_I(inode
)->sk
;
1583 if (sock
->sk_family
!= AF_NETLINK
)
1584 return ERR_PTR(-EINVAL
);
1590 static struct sk_buff
*netlink_alloc_large_skb(unsigned int size
,
1593 struct sk_buff
*skb
;
1596 if (size
<= NLMSG_GOODSIZE
|| broadcast
)
1597 return alloc_skb(size
, GFP_KERNEL
);
1599 size
= SKB_DATA_ALIGN(size
) +
1600 SKB_DATA_ALIGN(sizeof(struct skb_shared_info
));
1602 data
= vmalloc(size
);
1606 skb
= build_skb(data
, size
);
1611 skb
->destructor
= netlink_skb_destructor
;
1618 * Attach a skb to a netlink socket.
1619 * The caller must hold a reference to the destination socket. On error, the
1620 * reference is dropped. The skb is not send to the destination, just all
1621 * all error checks are performed and memory in the queue is reserved.
1623 * < 0: error. skb freed, reference to sock dropped.
1625 * 1: repeat lookup - reference dropped while waiting for socket memory.
1627 int netlink_attachskb(struct sock
*sk
, struct sk_buff
*skb
,
1628 long *timeo
, struct sock
*ssk
)
1630 struct netlink_sock
*nlk
;
1634 if ((atomic_read(&sk
->sk_rmem_alloc
) > sk
->sk_rcvbuf
||
1635 test_bit(NETLINK_CONGESTED
, &nlk
->state
)) &&
1636 !netlink_skb_is_mmaped(skb
)) {
1637 DECLARE_WAITQUEUE(wait
, current
);
1639 if (!ssk
|| netlink_is_kernel(ssk
))
1640 netlink_overrun(sk
);
1646 __set_current_state(TASK_INTERRUPTIBLE
);
1647 add_wait_queue(&nlk
->wait
, &wait
);
1649 if ((atomic_read(&sk
->sk_rmem_alloc
) > sk
->sk_rcvbuf
||
1650 test_bit(NETLINK_CONGESTED
, &nlk
->state
)) &&
1651 !sock_flag(sk
, SOCK_DEAD
))
1652 *timeo
= schedule_timeout(*timeo
);
1654 __set_current_state(TASK_RUNNING
);
1655 remove_wait_queue(&nlk
->wait
, &wait
);
1658 if (signal_pending(current
)) {
1660 return sock_intr_errno(*timeo
);
1664 netlink_skb_set_owner_r(skb
, sk
);
1668 static int __netlink_sendskb(struct sock
*sk
, struct sk_buff
*skb
)
1672 netlink_deliver_tap(skb
);
1674 #ifdef CONFIG_NETLINK_MMAP
1675 if (netlink_skb_is_mmaped(skb
))
1676 netlink_queue_mmaped_skb(sk
, skb
);
1677 else if (netlink_rx_is_mmaped(sk
))
1678 netlink_ring_set_copied(sk
, skb
);
1680 #endif /* CONFIG_NETLINK_MMAP */
1681 skb_queue_tail(&sk
->sk_receive_queue
, skb
);
1682 sk
->sk_data_ready(sk
);
1686 int netlink_sendskb(struct sock
*sk
, struct sk_buff
*skb
)
1688 int len
= __netlink_sendskb(sk
, skb
);
1694 void netlink_detachskb(struct sock
*sk
, struct sk_buff
*skb
)
1700 static struct sk_buff
*netlink_trim(struct sk_buff
*skb
, gfp_t allocation
)
1704 WARN_ON(skb
->sk
!= NULL
);
1705 if (netlink_skb_is_mmaped(skb
))
1708 delta
= skb
->end
- skb
->tail
;
1709 if (is_vmalloc_addr(skb
->head
) || delta
* 2 < skb
->truesize
)
1712 if (skb_shared(skb
)) {
1713 struct sk_buff
*nskb
= skb_clone(skb
, allocation
);
1720 if (!pskb_expand_head(skb
, 0, -delta
, allocation
))
1721 skb
->truesize
-= delta
;
1726 static int netlink_unicast_kernel(struct sock
*sk
, struct sk_buff
*skb
,
1730 struct netlink_sock
*nlk
= nlk_sk(sk
);
1732 ret
= -ECONNREFUSED
;
1733 if (nlk
->netlink_rcv
!= NULL
) {
1735 netlink_skb_set_owner_r(skb
, sk
);
1736 NETLINK_CB(skb
).sk
= ssk
;
1737 netlink_deliver_tap_kernel(sk
, ssk
, skb
);
1738 nlk
->netlink_rcv(skb
);
1747 int netlink_unicast(struct sock
*ssk
, struct sk_buff
*skb
,
1748 u32 portid
, int nonblock
)
1754 skb
= netlink_trim(skb
, gfp_any());
1756 timeo
= sock_sndtimeo(ssk
, nonblock
);
1758 sk
= netlink_getsockbyportid(ssk
, portid
);
1763 if (netlink_is_kernel(sk
))
1764 return netlink_unicast_kernel(sk
, skb
, ssk
);
1766 if (sk_filter(sk
, skb
)) {
1773 err
= netlink_attachskb(sk
, skb
, &timeo
, ssk
);
1779 return netlink_sendskb(sk
, skb
);
1781 EXPORT_SYMBOL(netlink_unicast
);
1783 struct sk_buff
*netlink_alloc_skb(struct sock
*ssk
, unsigned int size
,
1784 u32 dst_portid
, gfp_t gfp_mask
)
1786 #ifdef CONFIG_NETLINK_MMAP
1787 struct sock
*sk
= NULL
;
1788 struct sk_buff
*skb
;
1789 struct netlink_ring
*ring
;
1790 struct nl_mmap_hdr
*hdr
;
1791 unsigned int maxlen
;
1793 sk
= netlink_getsockbyportid(ssk
, dst_portid
);
1797 ring
= &nlk_sk(sk
)->rx_ring
;
1798 /* fast-path without atomic ops for common case: non-mmaped receiver */
1799 if (ring
->pg_vec
== NULL
)
1802 if (ring
->frame_size
- NL_MMAP_HDRLEN
< size
)
1805 skb
= alloc_skb_head(gfp_mask
);
1809 spin_lock_bh(&sk
->sk_receive_queue
.lock
);
1810 /* check again under lock */
1811 if (ring
->pg_vec
== NULL
)
1814 /* check again under lock */
1815 maxlen
= ring
->frame_size
- NL_MMAP_HDRLEN
;
1819 netlink_forward_ring(ring
);
1820 hdr
= netlink_current_frame(ring
, NL_MMAP_STATUS_UNUSED
);
1823 netlink_ring_setup_skb(skb
, sk
, ring
, hdr
);
1824 netlink_set_status(hdr
, NL_MMAP_STATUS_RESERVED
);
1825 atomic_inc(&ring
->pending
);
1826 netlink_increment_head(ring
);
1828 spin_unlock_bh(&sk
->sk_receive_queue
.lock
);
1833 spin_unlock_bh(&sk
->sk_receive_queue
.lock
);
1834 netlink_overrun(sk
);
1841 spin_unlock_bh(&sk
->sk_receive_queue
.lock
);
1846 return alloc_skb(size
, gfp_mask
);
1848 EXPORT_SYMBOL_GPL(netlink_alloc_skb
);
1850 int netlink_has_listeners(struct sock
*sk
, unsigned int group
)
1853 struct listeners
*listeners
;
1855 BUG_ON(!netlink_is_kernel(sk
));
1858 listeners
= rcu_dereference(nl_table
[sk
->sk_protocol
].listeners
);
1860 if (listeners
&& group
- 1 < nl_table
[sk
->sk_protocol
].groups
)
1861 res
= test_bit(group
- 1, listeners
->masks
);
1867 EXPORT_SYMBOL_GPL(netlink_has_listeners
);
1869 static int netlink_broadcast_deliver(struct sock
*sk
, struct sk_buff
*skb
)
1871 struct netlink_sock
*nlk
= nlk_sk(sk
);
1873 if (atomic_read(&sk
->sk_rmem_alloc
) <= sk
->sk_rcvbuf
&&
1874 !test_bit(NETLINK_CONGESTED
, &nlk
->state
)) {
1875 netlink_skb_set_owner_r(skb
, sk
);
1876 __netlink_sendskb(sk
, skb
);
1877 return atomic_read(&sk
->sk_rmem_alloc
) > (sk
->sk_rcvbuf
>> 1);
1882 struct netlink_broadcast_data
{
1883 struct sock
*exclude_sk
;
1888 int delivery_failure
;
1892 struct sk_buff
*skb
, *skb2
;
1893 int (*tx_filter
)(struct sock
*dsk
, struct sk_buff
*skb
, void *data
);
1897 static void do_one_broadcast(struct sock
*sk
,
1898 struct netlink_broadcast_data
*p
)
1900 struct netlink_sock
*nlk
= nlk_sk(sk
);
1903 if (p
->exclude_sk
== sk
)
1906 if (nlk
->portid
== p
->portid
|| p
->group
- 1 >= nlk
->ngroups
||
1907 !test_bit(p
->group
- 1, nlk
->groups
))
1910 if (!net_eq(sock_net(sk
), p
->net
))
1914 netlink_overrun(sk
);
1919 if (p
->skb2
== NULL
) {
1920 if (skb_shared(p
->skb
)) {
1921 p
->skb2
= skb_clone(p
->skb
, p
->allocation
);
1923 p
->skb2
= skb_get(p
->skb
);
1925 * skb ownership may have been set when
1926 * delivered to a previous socket.
1928 skb_orphan(p
->skb2
);
1931 if (p
->skb2
== NULL
) {
1932 netlink_overrun(sk
);
1933 /* Clone failed. Notify ALL listeners. */
1935 if (nlk
->flags
& NETLINK_BROADCAST_SEND_ERROR
)
1936 p
->delivery_failure
= 1;
1937 } else if (p
->tx_filter
&& p
->tx_filter(sk
, p
->skb2
, p
->tx_data
)) {
1940 } else if (sk_filter(sk
, p
->skb2
)) {
1943 } else if ((val
= netlink_broadcast_deliver(sk
, p
->skb2
)) < 0) {
1944 netlink_overrun(sk
);
1945 if (nlk
->flags
& NETLINK_BROADCAST_SEND_ERROR
)
1946 p
->delivery_failure
= 1;
1948 p
->congested
|= val
;
1955 int netlink_broadcast_filtered(struct sock
*ssk
, struct sk_buff
*skb
, u32 portid
,
1956 u32 group
, gfp_t allocation
,
1957 int (*filter
)(struct sock
*dsk
, struct sk_buff
*skb
, void *data
),
1960 struct net
*net
= sock_net(ssk
);
1961 struct netlink_broadcast_data info
;
1964 skb
= netlink_trim(skb
, allocation
);
1966 info
.exclude_sk
= ssk
;
1968 info
.portid
= portid
;
1971 info
.delivery_failure
= 0;
1974 info
.allocation
= allocation
;
1977 info
.tx_filter
= filter
;
1978 info
.tx_data
= filter_data
;
1980 /* While we sleep in clone, do not allow to change socket list */
1982 netlink_lock_table();
1984 sk_for_each_bound(sk
, &nl_table
[ssk
->sk_protocol
].mc_list
)
1985 do_one_broadcast(sk
, &info
);
1989 netlink_unlock_table();
1991 if (info
.delivery_failure
) {
1992 kfree_skb(info
.skb2
);
1995 consume_skb(info
.skb2
);
1997 if (info
.delivered
) {
1998 if (info
.congested
&& (allocation
& __GFP_WAIT
))
2004 EXPORT_SYMBOL(netlink_broadcast_filtered
);
2006 int netlink_broadcast(struct sock
*ssk
, struct sk_buff
*skb
, u32 portid
,
2007 u32 group
, gfp_t allocation
)
2009 return netlink_broadcast_filtered(ssk
, skb
, portid
, group
, allocation
,
2012 EXPORT_SYMBOL(netlink_broadcast
);
2014 struct netlink_set_err_data
{
2015 struct sock
*exclude_sk
;
2021 static int do_one_set_err(struct sock
*sk
, struct netlink_set_err_data
*p
)
2023 struct netlink_sock
*nlk
= nlk_sk(sk
);
2026 if (sk
== p
->exclude_sk
)
2029 if (!net_eq(sock_net(sk
), sock_net(p
->exclude_sk
)))
2032 if (nlk
->portid
== p
->portid
|| p
->group
- 1 >= nlk
->ngroups
||
2033 !test_bit(p
->group
- 1, nlk
->groups
))
2036 if (p
->code
== ENOBUFS
&& nlk
->flags
& NETLINK_RECV_NO_ENOBUFS
) {
2041 sk
->sk_err
= p
->code
;
2042 sk
->sk_error_report(sk
);
2048 * netlink_set_err - report error to broadcast listeners
2049 * @ssk: the kernel netlink socket, as returned by netlink_kernel_create()
2050 * @portid: the PORTID of a process that we want to skip (if any)
2051 * @group: the broadcast group that will notice the error
2052 * @code: error code, must be negative (as usual in kernelspace)
2054 * This function returns the number of broadcast listeners that have set the
2055 * NETLINK_RECV_NO_ENOBUFS socket option.
2057 int netlink_set_err(struct sock
*ssk
, u32 portid
, u32 group
, int code
)
2059 struct netlink_set_err_data info
;
2063 info
.exclude_sk
= ssk
;
2064 info
.portid
= portid
;
2066 /* sk->sk_err wants a positive error value */
2069 read_lock(&nl_table_lock
);
2071 sk_for_each_bound(sk
, &nl_table
[ssk
->sk_protocol
].mc_list
)
2072 ret
+= do_one_set_err(sk
, &info
);
2074 read_unlock(&nl_table_lock
);
2077 EXPORT_SYMBOL(netlink_set_err
);
2079 /* must be called with netlink table grabbed */
2080 static void netlink_update_socket_mc(struct netlink_sock
*nlk
,
2084 int old
, new = !!is_new
, subscriptions
;
2086 old
= test_bit(group
- 1, nlk
->groups
);
2087 subscriptions
= nlk
->subscriptions
- old
+ new;
2089 __set_bit(group
- 1, nlk
->groups
);
2091 __clear_bit(group
- 1, nlk
->groups
);
2092 netlink_update_subscriptions(&nlk
->sk
, subscriptions
);
2093 netlink_update_listeners(&nlk
->sk
);
2096 static int netlink_setsockopt(struct socket
*sock
, int level
, int optname
,
2097 char __user
*optval
, unsigned int optlen
)
2099 struct sock
*sk
= sock
->sk
;
2100 struct netlink_sock
*nlk
= nlk_sk(sk
);
2101 unsigned int val
= 0;
2104 if (level
!= SOL_NETLINK
)
2105 return -ENOPROTOOPT
;
2107 if (optname
!= NETLINK_RX_RING
&& optname
!= NETLINK_TX_RING
&&
2108 optlen
>= sizeof(int) &&
2109 get_user(val
, (unsigned int __user
*)optval
))
2113 case NETLINK_PKTINFO
:
2115 nlk
->flags
|= NETLINK_RECV_PKTINFO
;
2117 nlk
->flags
&= ~NETLINK_RECV_PKTINFO
;
2120 case NETLINK_ADD_MEMBERSHIP
:
2121 case NETLINK_DROP_MEMBERSHIP
: {
2122 if (!netlink_allowed(sock
, NL_CFG_F_NONROOT_RECV
))
2124 err
= netlink_realloc_groups(sk
);
2127 if (!val
|| val
- 1 >= nlk
->ngroups
)
2129 if (optname
== NETLINK_ADD_MEMBERSHIP
&& nlk
->netlink_bind
) {
2130 err
= nlk
->netlink_bind(sock_net(sk
), val
);
2134 netlink_table_grab();
2135 netlink_update_socket_mc(nlk
, val
,
2136 optname
== NETLINK_ADD_MEMBERSHIP
);
2137 netlink_table_ungrab();
2138 if (optname
== NETLINK_DROP_MEMBERSHIP
&& nlk
->netlink_unbind
)
2139 nlk
->netlink_unbind(sock_net(sk
), val
);
2144 case NETLINK_BROADCAST_ERROR
:
2146 nlk
->flags
|= NETLINK_BROADCAST_SEND_ERROR
;
2148 nlk
->flags
&= ~NETLINK_BROADCAST_SEND_ERROR
;
2151 case NETLINK_NO_ENOBUFS
:
2153 nlk
->flags
|= NETLINK_RECV_NO_ENOBUFS
;
2154 clear_bit(NETLINK_CONGESTED
, &nlk
->state
);
2155 wake_up_interruptible(&nlk
->wait
);
2157 nlk
->flags
&= ~NETLINK_RECV_NO_ENOBUFS
;
2161 #ifdef CONFIG_NETLINK_MMAP
2162 case NETLINK_RX_RING
:
2163 case NETLINK_TX_RING
: {
2164 struct nl_mmap_req req
;
2166 /* Rings might consume more memory than queue limits, require
2169 if (!capable(CAP_NET_ADMIN
))
2171 if (optlen
< sizeof(req
))
2173 if (copy_from_user(&req
, optval
, sizeof(req
)))
2175 err
= netlink_set_ring(sk
, &req
, false,
2176 optname
== NETLINK_TX_RING
);
2179 #endif /* CONFIG_NETLINK_MMAP */
2186 static int netlink_getsockopt(struct socket
*sock
, int level
, int optname
,
2187 char __user
*optval
, int __user
*optlen
)
2189 struct sock
*sk
= sock
->sk
;
2190 struct netlink_sock
*nlk
= nlk_sk(sk
);
2193 if (level
!= SOL_NETLINK
)
2194 return -ENOPROTOOPT
;
2196 if (get_user(len
, optlen
))
2202 case NETLINK_PKTINFO
:
2203 if (len
< sizeof(int))
2206 val
= nlk
->flags
& NETLINK_RECV_PKTINFO
? 1 : 0;
2207 if (put_user(len
, optlen
) ||
2208 put_user(val
, optval
))
2212 case NETLINK_BROADCAST_ERROR
:
2213 if (len
< sizeof(int))
2216 val
= nlk
->flags
& NETLINK_BROADCAST_SEND_ERROR
? 1 : 0;
2217 if (put_user(len
, optlen
) ||
2218 put_user(val
, optval
))
2222 case NETLINK_NO_ENOBUFS
:
2223 if (len
< sizeof(int))
2226 val
= nlk
->flags
& NETLINK_RECV_NO_ENOBUFS
? 1 : 0;
2227 if (put_user(len
, optlen
) ||
2228 put_user(val
, optval
))
2238 static void netlink_cmsg_recv_pktinfo(struct msghdr
*msg
, struct sk_buff
*skb
)
2240 struct nl_pktinfo info
;
2242 info
.group
= NETLINK_CB(skb
).dst_group
;
2243 put_cmsg(msg
, SOL_NETLINK
, NETLINK_PKTINFO
, sizeof(info
), &info
);
2246 static int netlink_sendmsg(struct kiocb
*kiocb
, struct socket
*sock
,
2247 struct msghdr
*msg
, size_t len
)
2249 struct sock_iocb
*siocb
= kiocb_to_siocb(kiocb
);
2250 struct sock
*sk
= sock
->sk
;
2251 struct netlink_sock
*nlk
= nlk_sk(sk
);
2252 DECLARE_SOCKADDR(struct sockaddr_nl
*, addr
, msg
->msg_name
);
2255 struct sk_buff
*skb
;
2257 struct scm_cookie scm
;
2258 u32 netlink_skb_flags
= 0;
2260 if (msg
->msg_flags
&MSG_OOB
)
2263 if (NULL
== siocb
->scm
)
2266 err
= scm_send(sock
, msg
, siocb
->scm
, true);
2270 if (msg
->msg_namelen
) {
2272 if (addr
->nl_family
!= AF_NETLINK
)
2274 dst_portid
= addr
->nl_pid
;
2275 dst_group
= ffs(addr
->nl_groups
);
2277 if ((dst_group
|| dst_portid
) &&
2278 !netlink_allowed(sock
, NL_CFG_F_NONROOT_SEND
))
2280 netlink_skb_flags
|= NETLINK_SKB_DST
;
2282 dst_portid
= nlk
->dst_portid
;
2283 dst_group
= nlk
->dst_group
;
2287 err
= netlink_autobind(sock
);
2292 if (netlink_tx_is_mmaped(sk
) &&
2293 msg
->msg_iter
.iov
->iov_base
== NULL
) {
2294 err
= netlink_mmap_sendmsg(sk
, msg
, dst_portid
, dst_group
,
2300 if (len
> sk
->sk_sndbuf
- 32)
2303 skb
= netlink_alloc_large_skb(len
, dst_group
);
2307 NETLINK_CB(skb
).portid
= nlk
->portid
;
2308 NETLINK_CB(skb
).dst_group
= dst_group
;
2309 NETLINK_CB(skb
).creds
= siocb
->scm
->creds
;
2310 NETLINK_CB(skb
).flags
= netlink_skb_flags
;
2313 if (memcpy_from_msg(skb_put(skb
, len
), msg
, len
)) {
2318 err
= security_netlink_send(sk
, skb
);
2325 atomic_inc(&skb
->users
);
2326 netlink_broadcast(sk
, skb
, dst_portid
, dst_group
, GFP_KERNEL
);
2328 err
= netlink_unicast(sk
, skb
, dst_portid
, msg
->msg_flags
&MSG_DONTWAIT
);
2331 scm_destroy(siocb
->scm
);
2335 static int netlink_recvmsg(struct kiocb
*kiocb
, struct socket
*sock
,
2336 struct msghdr
*msg
, size_t len
,
2339 struct sock_iocb
*siocb
= kiocb_to_siocb(kiocb
);
2340 struct scm_cookie scm
;
2341 struct sock
*sk
= sock
->sk
;
2342 struct netlink_sock
*nlk
= nlk_sk(sk
);
2343 int noblock
= flags
&MSG_DONTWAIT
;
2345 struct sk_buff
*skb
, *data_skb
;
2353 skb
= skb_recv_datagram(sk
, flags
, noblock
, &err
);
2359 #ifdef CONFIG_COMPAT_NETLINK_MESSAGES
2360 if (unlikely(skb_shinfo(skb
)->frag_list
)) {
2362 * If this skb has a frag_list, then here that means that we
2363 * will have to use the frag_list skb's data for compat tasks
2364 * and the regular skb's data for normal (non-compat) tasks.
2366 * If we need to send the compat skb, assign it to the
2367 * 'data_skb' variable so that it will be used below for data
2368 * copying. We keep 'skb' for everything else, including
2369 * freeing both later.
2371 if (flags
& MSG_CMSG_COMPAT
)
2372 data_skb
= skb_shinfo(skb
)->frag_list
;
2376 /* Record the max length of recvmsg() calls for future allocations */
2377 nlk
->max_recvmsg_len
= max(nlk
->max_recvmsg_len
, len
);
2378 nlk
->max_recvmsg_len
= min_t(size_t, nlk
->max_recvmsg_len
,
2381 copied
= data_skb
->len
;
2383 msg
->msg_flags
|= MSG_TRUNC
;
2387 skb_reset_transport_header(data_skb
);
2388 err
= skb_copy_datagram_msg(data_skb
, 0, msg
, copied
);
2390 if (msg
->msg_name
) {
2391 DECLARE_SOCKADDR(struct sockaddr_nl
*, addr
, msg
->msg_name
);
2392 addr
->nl_family
= AF_NETLINK
;
2394 addr
->nl_pid
= NETLINK_CB(skb
).portid
;
2395 addr
->nl_groups
= netlink_group_mask(NETLINK_CB(skb
).dst_group
);
2396 msg
->msg_namelen
= sizeof(*addr
);
2399 if (nlk
->flags
& NETLINK_RECV_PKTINFO
)
2400 netlink_cmsg_recv_pktinfo(msg
, skb
);
2402 if (NULL
== siocb
->scm
) {
2403 memset(&scm
, 0, sizeof(scm
));
2406 siocb
->scm
->creds
= *NETLINK_CREDS(skb
);
2407 if (flags
& MSG_TRUNC
)
2408 copied
= data_skb
->len
;
2410 skb_free_datagram(sk
, skb
);
2412 if (nlk
->cb_running
&&
2413 atomic_read(&sk
->sk_rmem_alloc
) <= sk
->sk_rcvbuf
/ 2) {
2414 ret
= netlink_dump(sk
);
2417 sk
->sk_error_report(sk
);
2421 scm_recv(sock
, msg
, siocb
->scm
, flags
);
2423 netlink_rcv_wake(sk
);
2424 return err
? : copied
;
2427 static void netlink_data_ready(struct sock
*sk
)
2433 * We export these functions to other modules. They provide a
2434 * complete set of kernel non-blocking support for message
2439 __netlink_kernel_create(struct net
*net
, int unit
, struct module
*module
,
2440 struct netlink_kernel_cfg
*cfg
)
2442 struct socket
*sock
;
2444 struct netlink_sock
*nlk
;
2445 struct listeners
*listeners
= NULL
;
2446 struct mutex
*cb_mutex
= cfg
? cfg
->cb_mutex
: NULL
;
2447 unsigned int groups
;
2451 if (unit
< 0 || unit
>= MAX_LINKS
)
2454 if (sock_create_lite(PF_NETLINK
, SOCK_DGRAM
, unit
, &sock
))
2458 * We have to just have a reference on the net from sk, but don't
2459 * get_net it. Besides, we cannot get and then put the net here.
2460 * So we create one inside init_net and the move it to net.
2463 if (__netlink_create(&init_net
, sock
, cb_mutex
, unit
) < 0)
2464 goto out_sock_release_nosk
;
2467 sk_change_net(sk
, net
);
2469 if (!cfg
|| cfg
->groups
< 32)
2472 groups
= cfg
->groups
;
2474 listeners
= kzalloc(sizeof(*listeners
) + NLGRPSZ(groups
), GFP_KERNEL
);
2476 goto out_sock_release
;
2478 sk
->sk_data_ready
= netlink_data_ready
;
2479 if (cfg
&& cfg
->input
)
2480 nlk_sk(sk
)->netlink_rcv
= cfg
->input
;
2482 if (netlink_insert(sk
, net
, 0))
2483 goto out_sock_release
;
2486 nlk
->flags
|= NETLINK_KERNEL_SOCKET
;
2488 netlink_table_grab();
2489 if (!nl_table
[unit
].registered
) {
2490 nl_table
[unit
].groups
= groups
;
2491 rcu_assign_pointer(nl_table
[unit
].listeners
, listeners
);
2492 nl_table
[unit
].cb_mutex
= cb_mutex
;
2493 nl_table
[unit
].module
= module
;
2495 nl_table
[unit
].bind
= cfg
->bind
;
2496 nl_table
[unit
].unbind
= cfg
->unbind
;
2497 nl_table
[unit
].flags
= cfg
->flags
;
2499 nl_table
[unit
].compare
= cfg
->compare
;
2501 nl_table
[unit
].registered
= 1;
2504 nl_table
[unit
].registered
++;
2506 netlink_table_ungrab();
2511 netlink_kernel_release(sk
);
2514 out_sock_release_nosk
:
2518 EXPORT_SYMBOL(__netlink_kernel_create
);
2521 netlink_kernel_release(struct sock
*sk
)
2523 sk_release_kernel(sk
);
2525 EXPORT_SYMBOL(netlink_kernel_release
);
2527 int __netlink_change_ngroups(struct sock
*sk
, unsigned int groups
)
2529 struct listeners
*new, *old
;
2530 struct netlink_table
*tbl
= &nl_table
[sk
->sk_protocol
];
2535 if (NLGRPSZ(tbl
->groups
) < NLGRPSZ(groups
)) {
2536 new = kzalloc(sizeof(*new) + NLGRPSZ(groups
), GFP_ATOMIC
);
2539 old
= nl_deref_protected(tbl
->listeners
);
2540 memcpy(new->masks
, old
->masks
, NLGRPSZ(tbl
->groups
));
2541 rcu_assign_pointer(tbl
->listeners
, new);
2543 kfree_rcu(old
, rcu
);
2545 tbl
->groups
= groups
;
2551 * netlink_change_ngroups - change number of multicast groups
2553 * This changes the number of multicast groups that are available
2554 * on a certain netlink family. Note that it is not possible to
2555 * change the number of groups to below 32. Also note that it does
2556 * not implicitly call netlink_clear_multicast_users() when the
2557 * number of groups is reduced.
2559 * @sk: The kernel netlink socket, as returned by netlink_kernel_create().
2560 * @groups: The new number of groups.
2562 int netlink_change_ngroups(struct sock
*sk
, unsigned int groups
)
2566 netlink_table_grab();
2567 err
= __netlink_change_ngroups(sk
, groups
);
2568 netlink_table_ungrab();
2573 void __netlink_clear_multicast_users(struct sock
*ksk
, unsigned int group
)
2576 struct netlink_table
*tbl
= &nl_table
[ksk
->sk_protocol
];
2578 sk_for_each_bound(sk
, &tbl
->mc_list
)
2579 netlink_update_socket_mc(nlk_sk(sk
), group
, 0);
2583 __nlmsg_put(struct sk_buff
*skb
, u32 portid
, u32 seq
, int type
, int len
, int flags
)
2585 struct nlmsghdr
*nlh
;
2586 int size
= nlmsg_msg_size(len
);
2588 nlh
= (struct nlmsghdr
*)skb_put(skb
, NLMSG_ALIGN(size
));
2589 nlh
->nlmsg_type
= type
;
2590 nlh
->nlmsg_len
= size
;
2591 nlh
->nlmsg_flags
= flags
;
2592 nlh
->nlmsg_pid
= portid
;
2593 nlh
->nlmsg_seq
= seq
;
2594 if (!__builtin_constant_p(size
) || NLMSG_ALIGN(size
) - size
!= 0)
2595 memset(nlmsg_data(nlh
) + len
, 0, NLMSG_ALIGN(size
) - size
);
2598 EXPORT_SYMBOL(__nlmsg_put
);
2601 * It looks a bit ugly.
2602 * It would be better to create kernel thread.
2605 static int netlink_dump(struct sock
*sk
)
2607 struct netlink_sock
*nlk
= nlk_sk(sk
);
2608 struct netlink_callback
*cb
;
2609 struct sk_buff
*skb
= NULL
;
2610 struct nlmsghdr
*nlh
;
2611 int len
, err
= -ENOBUFS
;
2614 mutex_lock(nlk
->cb_mutex
);
2615 if (!nlk
->cb_running
) {
2621 alloc_size
= max_t(int, cb
->min_dump_alloc
, NLMSG_GOODSIZE
);
2623 if (!netlink_rx_is_mmaped(sk
) &&
2624 atomic_read(&sk
->sk_rmem_alloc
) >= sk
->sk_rcvbuf
)
2627 /* NLMSG_GOODSIZE is small to avoid high order allocations being
2628 * required, but it makes sense to _attempt_ a 16K bytes allocation
2629 * to reduce number of system calls on dump operations, if user
2630 * ever provided a big enough buffer.
2632 if (alloc_size
< nlk
->max_recvmsg_len
) {
2633 skb
= netlink_alloc_skb(sk
,
2634 nlk
->max_recvmsg_len
,
2639 /* available room should be exact amount to avoid MSG_TRUNC */
2641 skb_reserve(skb
, skb_tailroom(skb
) -
2642 nlk
->max_recvmsg_len
);
2645 skb
= netlink_alloc_skb(sk
, alloc_size
, nlk
->portid
,
2649 netlink_skb_set_owner_r(skb
, sk
);
2651 len
= cb
->dump(skb
, cb
);
2654 mutex_unlock(nlk
->cb_mutex
);
2656 if (sk_filter(sk
, skb
))
2659 __netlink_sendskb(sk
, skb
);
2663 nlh
= nlmsg_put_answer(skb
, cb
, NLMSG_DONE
, sizeof(len
), NLM_F_MULTI
);
2667 nl_dump_check_consistent(cb
, nlh
);
2669 memcpy(nlmsg_data(nlh
), &len
, sizeof(len
));
2671 if (sk_filter(sk
, skb
))
2674 __netlink_sendskb(sk
, skb
);
2679 nlk
->cb_running
= false;
2680 mutex_unlock(nlk
->cb_mutex
);
2681 module_put(cb
->module
);
2682 consume_skb(cb
->skb
);
2686 mutex_unlock(nlk
->cb_mutex
);
2691 int __netlink_dump_start(struct sock
*ssk
, struct sk_buff
*skb
,
2692 const struct nlmsghdr
*nlh
,
2693 struct netlink_dump_control
*control
)
2695 struct netlink_callback
*cb
;
2697 struct netlink_sock
*nlk
;
2700 /* Memory mapped dump requests need to be copied to avoid looping
2701 * on the pending state in netlink_mmap_sendmsg() while the CB hold
2702 * a reference to the skb.
2704 if (netlink_skb_is_mmaped(skb
)) {
2705 skb
= skb_copy(skb
, GFP_KERNEL
);
2709 atomic_inc(&skb
->users
);
2711 sk
= netlink_lookup(sock_net(ssk
), ssk
->sk_protocol
, NETLINK_CB(skb
).portid
);
2713 ret
= -ECONNREFUSED
;
2718 mutex_lock(nlk
->cb_mutex
);
2719 /* A dump is in progress... */
2720 if (nlk
->cb_running
) {
2724 /* add reference of module which cb->dump belongs to */
2725 if (!try_module_get(control
->module
)) {
2726 ret
= -EPROTONOSUPPORT
;
2731 memset(cb
, 0, sizeof(*cb
));
2732 cb
->dump
= control
->dump
;
2733 cb
->done
= control
->done
;
2735 cb
->data
= control
->data
;
2736 cb
->module
= control
->module
;
2737 cb
->min_dump_alloc
= control
->min_dump_alloc
;
2740 nlk
->cb_running
= true;
2742 mutex_unlock(nlk
->cb_mutex
);
2744 ret
= netlink_dump(sk
);
2750 /* We successfully started a dump, by returning -EINTR we
2751 * signal not to send ACK even if it was requested.
2757 mutex_unlock(nlk
->cb_mutex
);
2762 EXPORT_SYMBOL(__netlink_dump_start
);
2764 void netlink_ack(struct sk_buff
*in_skb
, struct nlmsghdr
*nlh
, int err
)
2766 struct sk_buff
*skb
;
2767 struct nlmsghdr
*rep
;
2768 struct nlmsgerr
*errmsg
;
2769 size_t payload
= sizeof(*errmsg
);
2771 /* error messages get the original request appened */
2773 payload
+= nlmsg_len(nlh
);
2775 skb
= netlink_alloc_skb(in_skb
->sk
, nlmsg_total_size(payload
),
2776 NETLINK_CB(in_skb
).portid
, GFP_KERNEL
);
2780 sk
= netlink_lookup(sock_net(in_skb
->sk
),
2781 in_skb
->sk
->sk_protocol
,
2782 NETLINK_CB(in_skb
).portid
);
2784 sk
->sk_err
= ENOBUFS
;
2785 sk
->sk_error_report(sk
);
2791 rep
= __nlmsg_put(skb
, NETLINK_CB(in_skb
).portid
, nlh
->nlmsg_seq
,
2792 NLMSG_ERROR
, payload
, 0);
2793 errmsg
= nlmsg_data(rep
);
2794 errmsg
->error
= err
;
2795 memcpy(&errmsg
->msg
, nlh
, err
? nlh
->nlmsg_len
: sizeof(*nlh
));
2796 netlink_unicast(in_skb
->sk
, skb
, NETLINK_CB(in_skb
).portid
, MSG_DONTWAIT
);
2798 EXPORT_SYMBOL(netlink_ack
);
2800 int netlink_rcv_skb(struct sk_buff
*skb
, int (*cb
)(struct sk_buff
*,
2803 struct nlmsghdr
*nlh
;
2806 while (skb
->len
>= nlmsg_total_size(0)) {
2809 nlh
= nlmsg_hdr(skb
);
2812 if (nlh
->nlmsg_len
< NLMSG_HDRLEN
|| skb
->len
< nlh
->nlmsg_len
)
2815 /* Only requests are handled by the kernel */
2816 if (!(nlh
->nlmsg_flags
& NLM_F_REQUEST
))
2819 /* Skip control messages */
2820 if (nlh
->nlmsg_type
< NLMSG_MIN_TYPE
)
2828 if (nlh
->nlmsg_flags
& NLM_F_ACK
|| err
)
2829 netlink_ack(skb
, nlh
, err
);
2832 msglen
= NLMSG_ALIGN(nlh
->nlmsg_len
);
2833 if (msglen
> skb
->len
)
2835 skb_pull(skb
, msglen
);
2840 EXPORT_SYMBOL(netlink_rcv_skb
);
2843 * nlmsg_notify - send a notification netlink message
2844 * @sk: netlink socket to use
2845 * @skb: notification message
2846 * @portid: destination netlink portid for reports or 0
2847 * @group: destination multicast group or 0
2848 * @report: 1 to report back, 0 to disable
2849 * @flags: allocation flags
2851 int nlmsg_notify(struct sock
*sk
, struct sk_buff
*skb
, u32 portid
,
2852 unsigned int group
, int report
, gfp_t flags
)
2857 int exclude_portid
= 0;
2860 atomic_inc(&skb
->users
);
2861 exclude_portid
= portid
;
2864 /* errors reported via destination sk->sk_err, but propagate
2865 * delivery errors if NETLINK_BROADCAST_ERROR flag is set */
2866 err
= nlmsg_multicast(sk
, skb
, exclude_portid
, group
, flags
);
2872 err2
= nlmsg_unicast(sk
, skb
, portid
);
2873 if (!err
|| err
== -ESRCH
)
2879 EXPORT_SYMBOL(nlmsg_notify
);
2881 #ifdef CONFIG_PROC_FS
2882 struct nl_seq_iter
{
2883 struct seq_net_private p
;
2888 static struct sock
*netlink_seq_socket_idx(struct seq_file
*seq
, loff_t pos
)
2890 struct nl_seq_iter
*iter
= seq
->private;
2892 struct netlink_sock
*nlk
;
2896 for (i
= 0; i
< MAX_LINKS
; i
++) {
2897 struct rhashtable
*ht
= &nl_table
[i
].hash
;
2898 const struct bucket_table
*tbl
= rht_dereference_rcu(ht
->tbl
, ht
);
2900 for (j
= 0; j
< tbl
->size
; j
++) {
2901 struct rhash_head
*node
;
2903 rht_for_each_entry_rcu(nlk
, node
, tbl
, j
, node
) {
2904 s
= (struct sock
*)nlk
;
2906 if (sock_net(s
) != seq_file_net(seq
))
2920 static void *netlink_seq_start(struct seq_file
*seq
, loff_t
*pos
)
2921 __acquires(nl_table_lock
) __acquires(RCU
)
2923 read_lock(&nl_table_lock
);
2925 return *pos
? netlink_seq_socket_idx(seq
, *pos
- 1) : SEQ_START_TOKEN
;
2928 static void *netlink_seq_next(struct seq_file
*seq
, void *v
, loff_t
*pos
)
2930 struct rhashtable
*ht
;
2931 const struct bucket_table
*tbl
;
2932 struct rhash_head
*node
;
2933 struct netlink_sock
*nlk
;
2934 struct nl_seq_iter
*iter
;
2940 if (v
== SEQ_START_TOKEN
)
2941 return netlink_seq_socket_idx(seq
, 0);
2943 net
= seq_file_net(seq
);
2944 iter
= seq
->private;
2948 ht
= &nl_table
[i
].hash
;
2949 tbl
= rht_dereference_rcu(ht
->tbl
, ht
);
2950 rht_for_each_entry_rcu_continue(nlk
, node
, nlk
->node
.next
, tbl
, iter
->hash_idx
, node
)
2951 if (net_eq(sock_net((struct sock
*)nlk
), net
))
2954 j
= iter
->hash_idx
+ 1;
2958 for (; j
< tbl
->size
; j
++) {
2959 rht_for_each_entry_rcu(nlk
, node
, tbl
, j
, node
) {
2960 if (net_eq(sock_net((struct sock
*)nlk
), net
)) {
2969 } while (++i
< MAX_LINKS
);
2974 static void netlink_seq_stop(struct seq_file
*seq
, void *v
)
2975 __releases(RCU
) __releases(nl_table_lock
)
2978 read_unlock(&nl_table_lock
);
2982 static int netlink_seq_show(struct seq_file
*seq
, void *v
)
2984 if (v
== SEQ_START_TOKEN
) {
2986 "sk Eth Pid Groups "
2987 "Rmem Wmem Dump Locks Drops Inode\n");
2990 struct netlink_sock
*nlk
= nlk_sk(s
);
2992 seq_printf(seq
, "%pK %-3d %-6u %08x %-8d %-8d %d %-8d %-8d %-8lu\n",
2996 nlk
->groups
? (u32
)nlk
->groups
[0] : 0,
2997 sk_rmem_alloc_get(s
),
2998 sk_wmem_alloc_get(s
),
3000 atomic_read(&s
->sk_refcnt
),
3001 atomic_read(&s
->sk_drops
),
3009 static const struct seq_operations netlink_seq_ops
= {
3010 .start
= netlink_seq_start
,
3011 .next
= netlink_seq_next
,
3012 .stop
= netlink_seq_stop
,
3013 .show
= netlink_seq_show
,
3017 static int netlink_seq_open(struct inode
*inode
, struct file
*file
)
3019 return seq_open_net(inode
, file
, &netlink_seq_ops
,
3020 sizeof(struct nl_seq_iter
));
3023 static const struct file_operations netlink_seq_fops
= {
3024 .owner
= THIS_MODULE
,
3025 .open
= netlink_seq_open
,
3027 .llseek
= seq_lseek
,
3028 .release
= seq_release_net
,
3033 int netlink_register_notifier(struct notifier_block
*nb
)
3035 return atomic_notifier_chain_register(&netlink_chain
, nb
);
3037 EXPORT_SYMBOL(netlink_register_notifier
);
3039 int netlink_unregister_notifier(struct notifier_block
*nb
)
3041 return atomic_notifier_chain_unregister(&netlink_chain
, nb
);
3043 EXPORT_SYMBOL(netlink_unregister_notifier
);
3045 static const struct proto_ops netlink_ops
= {
3046 .family
= PF_NETLINK
,
3047 .owner
= THIS_MODULE
,
3048 .release
= netlink_release
,
3049 .bind
= netlink_bind
,
3050 .connect
= netlink_connect
,
3051 .socketpair
= sock_no_socketpair
,
3052 .accept
= sock_no_accept
,
3053 .getname
= netlink_getname
,
3054 .poll
= netlink_poll
,
3055 .ioctl
= sock_no_ioctl
,
3056 .listen
= sock_no_listen
,
3057 .shutdown
= sock_no_shutdown
,
3058 .setsockopt
= netlink_setsockopt
,
3059 .getsockopt
= netlink_getsockopt
,
3060 .sendmsg
= netlink_sendmsg
,
3061 .recvmsg
= netlink_recvmsg
,
3062 .mmap
= netlink_mmap
,
3063 .sendpage
= sock_no_sendpage
,
3066 static const struct net_proto_family netlink_family_ops
= {
3067 .family
= PF_NETLINK
,
3068 .create
= netlink_create
,
3069 .owner
= THIS_MODULE
, /* for consistency 8) */
3072 static int __net_init
netlink_net_init(struct net
*net
)
3074 #ifdef CONFIG_PROC_FS
3075 if (!proc_create("netlink", 0, net
->proc_net
, &netlink_seq_fops
))
3081 static void __net_exit
netlink_net_exit(struct net
*net
)
3083 #ifdef CONFIG_PROC_FS
3084 remove_proc_entry("netlink", net
->proc_net
);
3088 static void __init
netlink_add_usersock_entry(void)
3090 struct listeners
*listeners
;
3093 listeners
= kzalloc(sizeof(*listeners
) + NLGRPSZ(groups
), GFP_KERNEL
);
3095 panic("netlink_add_usersock_entry: Cannot allocate listeners\n");
3097 netlink_table_grab();
3099 nl_table
[NETLINK_USERSOCK
].groups
= groups
;
3100 rcu_assign_pointer(nl_table
[NETLINK_USERSOCK
].listeners
, listeners
);
3101 nl_table
[NETLINK_USERSOCK
].module
= THIS_MODULE
;
3102 nl_table
[NETLINK_USERSOCK
].registered
= 1;
3103 nl_table
[NETLINK_USERSOCK
].flags
= NL_CFG_F_NONROOT_SEND
;
3105 netlink_table_ungrab();
3108 static struct pernet_operations __net_initdata netlink_net_ops
= {
3109 .init
= netlink_net_init
,
3110 .exit
= netlink_net_exit
,
3113 static int __init
netlink_proto_init(void)
3116 int err
= proto_register(&netlink_proto
, 0);
3117 struct rhashtable_params ht_params
= {
3118 .head_offset
= offsetof(struct netlink_sock
, node
),
3119 .key_offset
= offsetof(struct netlink_sock
, portid
),
3120 .key_len
= sizeof(u32
), /* portid */
3122 .max_shift
= 16, /* 64K */
3123 .grow_decision
= rht_grow_above_75
,
3124 .shrink_decision
= rht_shrink_below_30
,
3125 #ifdef CONFIG_PROVE_LOCKING
3126 .mutex_is_held
= lockdep_nl_sk_hash_is_held
,
3133 BUILD_BUG_ON(sizeof(struct netlink_skb_parms
) > FIELD_SIZEOF(struct sk_buff
, cb
));
3135 nl_table
= kcalloc(MAX_LINKS
, sizeof(*nl_table
), GFP_KERNEL
);
3139 for (i
= 0; i
< MAX_LINKS
; i
++) {
3140 if (rhashtable_init(&nl_table
[i
].hash
, &ht_params
) < 0) {
3142 rhashtable_destroy(&nl_table
[i
].hash
);
3148 INIT_LIST_HEAD(&netlink_tap_all
);
3150 netlink_add_usersock_entry();
3152 sock_register(&netlink_family_ops
);
3153 register_pernet_subsys(&netlink_net_ops
);
3154 /* The netlink device handler may be needed early. */
3159 panic("netlink_init: Cannot allocate nl_table\n");
3162 core_initcall(netlink_proto_init
);