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 DEFINE_RWLOCK(nl_table_lock
);
100 EXPORT_SYMBOL_GPL(nl_table_lock
);
101 static atomic_t nl_table_users
= ATOMIC_INIT(0);
103 #define nl_deref_protected(X) rcu_dereference_protected(X, lockdep_is_held(&nl_table_lock));
105 /* Protects netlink socket hash table mutations */
106 DEFINE_MUTEX(nl_sk_hash_lock
);
108 static int lockdep_nl_sk_hash_is_held(void)
110 #ifdef CONFIG_LOCKDEP
111 return (debug_locks
) ? lockdep_is_held(&nl_sk_hash_lock
) : 1;
117 static ATOMIC_NOTIFIER_HEAD(netlink_chain
);
119 static DEFINE_SPINLOCK(netlink_tap_lock
);
120 static struct list_head netlink_tap_all __read_mostly
;
122 static inline u32
netlink_group_mask(u32 group
)
124 return group
? 1 << (group
- 1) : 0;
127 int netlink_add_tap(struct netlink_tap
*nt
)
129 if (unlikely(nt
->dev
->type
!= ARPHRD_NETLINK
))
132 spin_lock(&netlink_tap_lock
);
133 list_add_rcu(&nt
->list
, &netlink_tap_all
);
134 spin_unlock(&netlink_tap_lock
);
137 __module_get(nt
->module
);
141 EXPORT_SYMBOL_GPL(netlink_add_tap
);
143 static int __netlink_remove_tap(struct netlink_tap
*nt
)
146 struct netlink_tap
*tmp
;
148 spin_lock(&netlink_tap_lock
);
150 list_for_each_entry(tmp
, &netlink_tap_all
, list
) {
152 list_del_rcu(&nt
->list
);
158 pr_warn("__netlink_remove_tap: %p not found\n", nt
);
160 spin_unlock(&netlink_tap_lock
);
162 if (found
&& nt
->module
)
163 module_put(nt
->module
);
165 return found
? 0 : -ENODEV
;
168 int netlink_remove_tap(struct netlink_tap
*nt
)
172 ret
= __netlink_remove_tap(nt
);
177 EXPORT_SYMBOL_GPL(netlink_remove_tap
);
179 static bool netlink_filter_tap(const struct sk_buff
*skb
)
181 struct sock
*sk
= skb
->sk
;
183 /* We take the more conservative approach and
184 * whitelist socket protocols that may pass.
186 switch (sk
->sk_protocol
) {
188 case NETLINK_USERSOCK
:
189 case NETLINK_SOCK_DIAG
:
192 case NETLINK_FIB_LOOKUP
:
193 case NETLINK_NETFILTER
:
194 case NETLINK_GENERIC
:
201 static int __netlink_deliver_tap_skb(struct sk_buff
*skb
,
202 struct net_device
*dev
)
204 struct sk_buff
*nskb
;
205 struct sock
*sk
= skb
->sk
;
209 nskb
= skb_clone(skb
, GFP_ATOMIC
);
212 nskb
->protocol
= htons((u16
) sk
->sk_protocol
);
213 nskb
->pkt_type
= netlink_is_kernel(sk
) ?
214 PACKET_KERNEL
: PACKET_USER
;
216 ret
= dev_queue_xmit(nskb
);
217 if (unlikely(ret
> 0))
218 ret
= net_xmit_errno(ret
);
225 static void __netlink_deliver_tap(struct sk_buff
*skb
)
228 struct netlink_tap
*tmp
;
230 if (!netlink_filter_tap(skb
))
233 list_for_each_entry_rcu(tmp
, &netlink_tap_all
, list
) {
234 ret
= __netlink_deliver_tap_skb(skb
, tmp
->dev
);
240 static void netlink_deliver_tap(struct sk_buff
*skb
)
244 if (unlikely(!list_empty(&netlink_tap_all
)))
245 __netlink_deliver_tap(skb
);
250 static void netlink_deliver_tap_kernel(struct sock
*dst
, struct sock
*src
,
253 if (!(netlink_is_kernel(dst
) && netlink_is_kernel(src
)))
254 netlink_deliver_tap(skb
);
257 static void netlink_overrun(struct sock
*sk
)
259 struct netlink_sock
*nlk
= nlk_sk(sk
);
261 if (!(nlk
->flags
& NETLINK_RECV_NO_ENOBUFS
)) {
262 if (!test_and_set_bit(NETLINK_CONGESTED
, &nlk_sk(sk
)->state
)) {
263 sk
->sk_err
= ENOBUFS
;
264 sk
->sk_error_report(sk
);
267 atomic_inc(&sk
->sk_drops
);
270 static void netlink_rcv_wake(struct sock
*sk
)
272 struct netlink_sock
*nlk
= nlk_sk(sk
);
274 if (skb_queue_empty(&sk
->sk_receive_queue
))
275 clear_bit(NETLINK_CONGESTED
, &nlk
->state
);
276 if (!test_bit(NETLINK_CONGESTED
, &nlk
->state
))
277 wake_up_interruptible(&nlk
->wait
);
280 #ifdef CONFIG_NETLINK_MMAP
281 static bool netlink_skb_is_mmaped(const struct sk_buff
*skb
)
283 return NETLINK_CB(skb
).flags
& NETLINK_SKB_MMAPED
;
286 static bool netlink_rx_is_mmaped(struct sock
*sk
)
288 return nlk_sk(sk
)->rx_ring
.pg_vec
!= NULL
;
291 static bool netlink_tx_is_mmaped(struct sock
*sk
)
293 return nlk_sk(sk
)->tx_ring
.pg_vec
!= NULL
;
296 static __pure
struct page
*pgvec_to_page(const void *addr
)
298 if (is_vmalloc_addr(addr
))
299 return vmalloc_to_page(addr
);
301 return virt_to_page(addr
);
304 static void free_pg_vec(void **pg_vec
, unsigned int order
, unsigned int len
)
308 for (i
= 0; i
< len
; i
++) {
309 if (pg_vec
[i
] != NULL
) {
310 if (is_vmalloc_addr(pg_vec
[i
]))
313 free_pages((unsigned long)pg_vec
[i
], order
);
319 static void *alloc_one_pg_vec_page(unsigned long order
)
322 gfp_t gfp_flags
= GFP_KERNEL
| __GFP_COMP
| __GFP_ZERO
|
323 __GFP_NOWARN
| __GFP_NORETRY
;
325 buffer
= (void *)__get_free_pages(gfp_flags
, order
);
329 buffer
= vzalloc((1 << order
) * PAGE_SIZE
);
333 gfp_flags
&= ~__GFP_NORETRY
;
334 return (void *)__get_free_pages(gfp_flags
, order
);
337 static void **alloc_pg_vec(struct netlink_sock
*nlk
,
338 struct nl_mmap_req
*req
, unsigned int order
)
340 unsigned int block_nr
= req
->nm_block_nr
;
344 pg_vec
= kcalloc(block_nr
, sizeof(void *), GFP_KERNEL
);
348 for (i
= 0; i
< block_nr
; i
++) {
349 pg_vec
[i
] = alloc_one_pg_vec_page(order
);
350 if (pg_vec
[i
] == NULL
)
356 free_pg_vec(pg_vec
, order
, block_nr
);
360 static int netlink_set_ring(struct sock
*sk
, struct nl_mmap_req
*req
,
361 bool closing
, bool tx_ring
)
363 struct netlink_sock
*nlk
= nlk_sk(sk
);
364 struct netlink_ring
*ring
;
365 struct sk_buff_head
*queue
;
366 void **pg_vec
= NULL
;
367 unsigned int order
= 0;
370 ring
= tx_ring
? &nlk
->tx_ring
: &nlk
->rx_ring
;
371 queue
= tx_ring
? &sk
->sk_write_queue
: &sk
->sk_receive_queue
;
374 if (atomic_read(&nlk
->mapped
))
376 if (atomic_read(&ring
->pending
))
380 if (req
->nm_block_nr
) {
381 if (ring
->pg_vec
!= NULL
)
384 if ((int)req
->nm_block_size
<= 0)
386 if (!PAGE_ALIGNED(req
->nm_block_size
))
388 if (req
->nm_frame_size
< NL_MMAP_HDRLEN
)
390 if (!IS_ALIGNED(req
->nm_frame_size
, NL_MMAP_MSG_ALIGNMENT
))
393 ring
->frames_per_block
= req
->nm_block_size
/
395 if (ring
->frames_per_block
== 0)
397 if (ring
->frames_per_block
* req
->nm_block_nr
!=
401 order
= get_order(req
->nm_block_size
);
402 pg_vec
= alloc_pg_vec(nlk
, req
, order
);
406 if (req
->nm_frame_nr
)
411 mutex_lock(&nlk
->pg_vec_lock
);
412 if (closing
|| atomic_read(&nlk
->mapped
) == 0) {
414 spin_lock_bh(&queue
->lock
);
416 ring
->frame_max
= req
->nm_frame_nr
- 1;
418 ring
->frame_size
= req
->nm_frame_size
;
419 ring
->pg_vec_pages
= req
->nm_block_size
/ PAGE_SIZE
;
421 swap(ring
->pg_vec_len
, req
->nm_block_nr
);
422 swap(ring
->pg_vec_order
, order
);
423 swap(ring
->pg_vec
, pg_vec
);
425 __skb_queue_purge(queue
);
426 spin_unlock_bh(&queue
->lock
);
428 WARN_ON(atomic_read(&nlk
->mapped
));
430 mutex_unlock(&nlk
->pg_vec_lock
);
433 free_pg_vec(pg_vec
, order
, req
->nm_block_nr
);
437 static void netlink_mm_open(struct vm_area_struct
*vma
)
439 struct file
*file
= vma
->vm_file
;
440 struct socket
*sock
= file
->private_data
;
441 struct sock
*sk
= sock
->sk
;
444 atomic_inc(&nlk_sk(sk
)->mapped
);
447 static void netlink_mm_close(struct vm_area_struct
*vma
)
449 struct file
*file
= vma
->vm_file
;
450 struct socket
*sock
= file
->private_data
;
451 struct sock
*sk
= sock
->sk
;
454 atomic_dec(&nlk_sk(sk
)->mapped
);
457 static const struct vm_operations_struct netlink_mmap_ops
= {
458 .open
= netlink_mm_open
,
459 .close
= netlink_mm_close
,
462 static int netlink_mmap(struct file
*file
, struct socket
*sock
,
463 struct vm_area_struct
*vma
)
465 struct sock
*sk
= sock
->sk
;
466 struct netlink_sock
*nlk
= nlk_sk(sk
);
467 struct netlink_ring
*ring
;
468 unsigned long start
, size
, expected
;
475 mutex_lock(&nlk
->pg_vec_lock
);
478 for (ring
= &nlk
->rx_ring
; ring
<= &nlk
->tx_ring
; ring
++) {
479 if (ring
->pg_vec
== NULL
)
481 expected
+= ring
->pg_vec_len
* ring
->pg_vec_pages
* PAGE_SIZE
;
487 size
= vma
->vm_end
- vma
->vm_start
;
488 if (size
!= expected
)
491 start
= vma
->vm_start
;
492 for (ring
= &nlk
->rx_ring
; ring
<= &nlk
->tx_ring
; ring
++) {
493 if (ring
->pg_vec
== NULL
)
496 for (i
= 0; i
< ring
->pg_vec_len
; i
++) {
498 void *kaddr
= ring
->pg_vec
[i
];
501 for (pg_num
= 0; pg_num
< ring
->pg_vec_pages
; pg_num
++) {
502 page
= pgvec_to_page(kaddr
);
503 err
= vm_insert_page(vma
, start
, page
);
512 atomic_inc(&nlk
->mapped
);
513 vma
->vm_ops
= &netlink_mmap_ops
;
516 mutex_unlock(&nlk
->pg_vec_lock
);
520 static void netlink_frame_flush_dcache(const struct nl_mmap_hdr
*hdr
)
522 #if ARCH_IMPLEMENTS_FLUSH_DCACHE_PAGE == 1
523 struct page
*p_start
, *p_end
;
525 /* First page is flushed through netlink_{get,set}_status */
526 p_start
= pgvec_to_page(hdr
+ PAGE_SIZE
);
527 p_end
= pgvec_to_page((void *)hdr
+ NL_MMAP_HDRLEN
+ hdr
->nm_len
- 1);
528 while (p_start
<= p_end
) {
529 flush_dcache_page(p_start
);
535 static enum nl_mmap_status
netlink_get_status(const struct nl_mmap_hdr
*hdr
)
538 flush_dcache_page(pgvec_to_page(hdr
));
539 return hdr
->nm_status
;
542 static void netlink_set_status(struct nl_mmap_hdr
*hdr
,
543 enum nl_mmap_status status
)
545 hdr
->nm_status
= status
;
546 flush_dcache_page(pgvec_to_page(hdr
));
550 static struct nl_mmap_hdr
*
551 __netlink_lookup_frame(const struct netlink_ring
*ring
, unsigned int pos
)
553 unsigned int pg_vec_pos
, frame_off
;
555 pg_vec_pos
= pos
/ ring
->frames_per_block
;
556 frame_off
= pos
% ring
->frames_per_block
;
558 return ring
->pg_vec
[pg_vec_pos
] + (frame_off
* ring
->frame_size
);
561 static struct nl_mmap_hdr
*
562 netlink_lookup_frame(const struct netlink_ring
*ring
, unsigned int pos
,
563 enum nl_mmap_status status
)
565 struct nl_mmap_hdr
*hdr
;
567 hdr
= __netlink_lookup_frame(ring
, pos
);
568 if (netlink_get_status(hdr
) != status
)
574 static struct nl_mmap_hdr
*
575 netlink_current_frame(const struct netlink_ring
*ring
,
576 enum nl_mmap_status status
)
578 return netlink_lookup_frame(ring
, ring
->head
, status
);
581 static struct nl_mmap_hdr
*
582 netlink_previous_frame(const struct netlink_ring
*ring
,
583 enum nl_mmap_status status
)
587 prev
= ring
->head
? ring
->head
- 1 : ring
->frame_max
;
588 return netlink_lookup_frame(ring
, prev
, status
);
591 static void netlink_increment_head(struct netlink_ring
*ring
)
593 ring
->head
= ring
->head
!= ring
->frame_max
? ring
->head
+ 1 : 0;
596 static void netlink_forward_ring(struct netlink_ring
*ring
)
598 unsigned int head
= ring
->head
, pos
= head
;
599 const struct nl_mmap_hdr
*hdr
;
602 hdr
= __netlink_lookup_frame(ring
, pos
);
603 if (hdr
->nm_status
== NL_MMAP_STATUS_UNUSED
)
605 if (hdr
->nm_status
!= NL_MMAP_STATUS_SKIP
)
607 netlink_increment_head(ring
);
608 } while (ring
->head
!= head
);
611 static bool netlink_dump_space(struct netlink_sock
*nlk
)
613 struct netlink_ring
*ring
= &nlk
->rx_ring
;
614 struct nl_mmap_hdr
*hdr
;
617 hdr
= netlink_current_frame(ring
, NL_MMAP_STATUS_UNUSED
);
621 n
= ring
->head
+ ring
->frame_max
/ 2;
622 if (n
> ring
->frame_max
)
623 n
-= ring
->frame_max
;
625 hdr
= __netlink_lookup_frame(ring
, n
);
627 return hdr
->nm_status
== NL_MMAP_STATUS_UNUSED
;
630 static unsigned int netlink_poll(struct file
*file
, struct socket
*sock
,
633 struct sock
*sk
= sock
->sk
;
634 struct netlink_sock
*nlk
= nlk_sk(sk
);
638 if (nlk
->rx_ring
.pg_vec
!= NULL
) {
639 /* Memory mapped sockets don't call recvmsg(), so flow control
640 * for dumps is performed here. A dump is allowed to continue
641 * if at least half the ring is unused.
643 while (nlk
->cb_running
&& netlink_dump_space(nlk
)) {
644 err
= netlink_dump(sk
);
647 sk
->sk_error_report(sk
);
651 netlink_rcv_wake(sk
);
654 mask
= datagram_poll(file
, sock
, wait
);
656 spin_lock_bh(&sk
->sk_receive_queue
.lock
);
657 if (nlk
->rx_ring
.pg_vec
) {
658 netlink_forward_ring(&nlk
->rx_ring
);
659 if (!netlink_previous_frame(&nlk
->rx_ring
, NL_MMAP_STATUS_UNUSED
))
660 mask
|= POLLIN
| POLLRDNORM
;
662 spin_unlock_bh(&sk
->sk_receive_queue
.lock
);
664 spin_lock_bh(&sk
->sk_write_queue
.lock
);
665 if (nlk
->tx_ring
.pg_vec
) {
666 if (netlink_current_frame(&nlk
->tx_ring
, NL_MMAP_STATUS_UNUSED
))
667 mask
|= POLLOUT
| POLLWRNORM
;
669 spin_unlock_bh(&sk
->sk_write_queue
.lock
);
674 static struct nl_mmap_hdr
*netlink_mmap_hdr(struct sk_buff
*skb
)
676 return (struct nl_mmap_hdr
*)(skb
->head
- NL_MMAP_HDRLEN
);
679 static void netlink_ring_setup_skb(struct sk_buff
*skb
, struct sock
*sk
,
680 struct netlink_ring
*ring
,
681 struct nl_mmap_hdr
*hdr
)
686 size
= ring
->frame_size
- NL_MMAP_HDRLEN
;
687 data
= (void *)hdr
+ NL_MMAP_HDRLEN
;
691 skb_reset_tail_pointer(skb
);
692 skb
->end
= skb
->tail
+ size
;
695 skb
->destructor
= netlink_skb_destructor
;
696 NETLINK_CB(skb
).flags
|= NETLINK_SKB_MMAPED
;
697 NETLINK_CB(skb
).sk
= sk
;
700 static int netlink_mmap_sendmsg(struct sock
*sk
, struct msghdr
*msg
,
701 u32 dst_portid
, u32 dst_group
,
702 struct sock_iocb
*siocb
)
704 struct netlink_sock
*nlk
= nlk_sk(sk
);
705 struct netlink_ring
*ring
;
706 struct nl_mmap_hdr
*hdr
;
710 int err
= 0, len
= 0;
712 /* Netlink messages are validated by the receiver before processing.
713 * In order to avoid userspace changing the contents of the message
714 * after validation, the socket and the ring may only be used by a
715 * single process, otherwise we fall back to copying.
717 if (atomic_long_read(&sk
->sk_socket
->file
->f_count
) > 2 ||
718 atomic_read(&nlk
->mapped
) > 1)
721 mutex_lock(&nlk
->pg_vec_lock
);
723 ring
= &nlk
->tx_ring
;
724 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
))
734 if (hdr
->nm_len
> maxlen
) {
739 netlink_frame_flush_dcache(hdr
);
741 if (likely(dst_portid
== 0 && dst_group
== 0 && excl
)) {
742 skb
= alloc_skb_head(GFP_KERNEL
);
748 netlink_ring_setup_skb(skb
, sk
, ring
, hdr
);
749 NETLINK_CB(skb
).flags
|= NETLINK_SKB_TX
;
750 __skb_put(skb
, hdr
->nm_len
);
751 netlink_set_status(hdr
, NL_MMAP_STATUS_RESERVED
);
752 atomic_inc(&ring
->pending
);
754 skb
= alloc_skb(hdr
->nm_len
, GFP_KERNEL
);
759 __skb_put(skb
, hdr
->nm_len
);
760 memcpy(skb
->data
, (void *)hdr
+ NL_MMAP_HDRLEN
, hdr
->nm_len
);
761 netlink_set_status(hdr
, NL_MMAP_STATUS_UNUSED
);
764 netlink_increment_head(ring
);
766 NETLINK_CB(skb
).portid
= nlk
->portid
;
767 NETLINK_CB(skb
).dst_group
= dst_group
;
768 NETLINK_CB(skb
).creds
= siocb
->scm
->creds
;
770 err
= security_netlink_send(sk
, skb
);
776 if (unlikely(dst_group
)) {
777 atomic_inc(&skb
->users
);
778 netlink_broadcast(sk
, skb
, dst_portid
, dst_group
,
781 err
= netlink_unicast(sk
, skb
, dst_portid
,
782 msg
->msg_flags
& MSG_DONTWAIT
);
787 } while (hdr
!= NULL
||
788 (!(msg
->msg_flags
& MSG_DONTWAIT
) &&
789 atomic_read(&nlk
->tx_ring
.pending
)));
794 mutex_unlock(&nlk
->pg_vec_lock
);
798 static void netlink_queue_mmaped_skb(struct sock
*sk
, struct sk_buff
*skb
)
800 struct nl_mmap_hdr
*hdr
;
802 hdr
= netlink_mmap_hdr(skb
);
803 hdr
->nm_len
= skb
->len
;
804 hdr
->nm_group
= NETLINK_CB(skb
).dst_group
;
805 hdr
->nm_pid
= NETLINK_CB(skb
).creds
.pid
;
806 hdr
->nm_uid
= from_kuid(sk_user_ns(sk
), NETLINK_CB(skb
).creds
.uid
);
807 hdr
->nm_gid
= from_kgid(sk_user_ns(sk
), NETLINK_CB(skb
).creds
.gid
);
808 netlink_frame_flush_dcache(hdr
);
809 netlink_set_status(hdr
, NL_MMAP_STATUS_VALID
);
811 NETLINK_CB(skb
).flags
|= NETLINK_SKB_DELIVERED
;
815 static void netlink_ring_set_copied(struct sock
*sk
, struct sk_buff
*skb
)
817 struct netlink_sock
*nlk
= nlk_sk(sk
);
818 struct netlink_ring
*ring
= &nlk
->rx_ring
;
819 struct nl_mmap_hdr
*hdr
;
821 spin_lock_bh(&sk
->sk_receive_queue
.lock
);
822 hdr
= netlink_current_frame(ring
, NL_MMAP_STATUS_UNUSED
);
824 spin_unlock_bh(&sk
->sk_receive_queue
.lock
);
829 netlink_increment_head(ring
);
830 __skb_queue_tail(&sk
->sk_receive_queue
, skb
);
831 spin_unlock_bh(&sk
->sk_receive_queue
.lock
);
833 hdr
->nm_len
= skb
->len
;
834 hdr
->nm_group
= NETLINK_CB(skb
).dst_group
;
835 hdr
->nm_pid
= NETLINK_CB(skb
).creds
.pid
;
836 hdr
->nm_uid
= from_kuid(sk_user_ns(sk
), NETLINK_CB(skb
).creds
.uid
);
837 hdr
->nm_gid
= from_kgid(sk_user_ns(sk
), NETLINK_CB(skb
).creds
.gid
);
838 netlink_set_status(hdr
, NL_MMAP_STATUS_COPY
);
841 #else /* CONFIG_NETLINK_MMAP */
842 #define netlink_skb_is_mmaped(skb) false
843 #define netlink_rx_is_mmaped(sk) false
844 #define netlink_tx_is_mmaped(sk) false
845 #define netlink_mmap sock_no_mmap
846 #define netlink_poll datagram_poll
847 #define netlink_mmap_sendmsg(sk, msg, dst_portid, dst_group, siocb) 0
848 #endif /* CONFIG_NETLINK_MMAP */
850 static void netlink_skb_destructor(struct sk_buff
*skb
)
852 #ifdef CONFIG_NETLINK_MMAP
853 struct nl_mmap_hdr
*hdr
;
854 struct netlink_ring
*ring
;
857 /* If a packet from the kernel to userspace was freed because of an
858 * error without being delivered to userspace, the kernel must reset
859 * the status. In the direction userspace to kernel, the status is
860 * always reset here after the packet was processed and freed.
862 if (netlink_skb_is_mmaped(skb
)) {
863 hdr
= netlink_mmap_hdr(skb
);
864 sk
= NETLINK_CB(skb
).sk
;
866 if (NETLINK_CB(skb
).flags
& NETLINK_SKB_TX
) {
867 netlink_set_status(hdr
, NL_MMAP_STATUS_UNUSED
);
868 ring
= &nlk_sk(sk
)->tx_ring
;
870 if (!(NETLINK_CB(skb
).flags
& NETLINK_SKB_DELIVERED
)) {
872 netlink_set_status(hdr
, NL_MMAP_STATUS_VALID
);
874 ring
= &nlk_sk(sk
)->rx_ring
;
877 WARN_ON(atomic_read(&ring
->pending
) == 0);
878 atomic_dec(&ring
->pending
);
884 if (is_vmalloc_addr(skb
->head
)) {
886 !atomic_dec_return(&(skb_shinfo(skb
)->dataref
)))
895 static void netlink_skb_set_owner_r(struct sk_buff
*skb
, struct sock
*sk
)
897 WARN_ON(skb
->sk
!= NULL
);
899 skb
->destructor
= netlink_skb_destructor
;
900 atomic_add(skb
->truesize
, &sk
->sk_rmem_alloc
);
901 sk_mem_charge(sk
, skb
->truesize
);
904 static void netlink_sock_destruct(struct sock
*sk
)
906 struct netlink_sock
*nlk
= nlk_sk(sk
);
908 if (nlk
->cb_running
) {
910 nlk
->cb
.done(&nlk
->cb
);
912 module_put(nlk
->cb
.module
);
913 kfree_skb(nlk
->cb
.skb
);
916 skb_queue_purge(&sk
->sk_receive_queue
);
917 #ifdef CONFIG_NETLINK_MMAP
919 struct nl_mmap_req req
;
921 memset(&req
, 0, sizeof(req
));
922 if (nlk
->rx_ring
.pg_vec
)
923 netlink_set_ring(sk
, &req
, true, false);
924 memset(&req
, 0, sizeof(req
));
925 if (nlk
->tx_ring
.pg_vec
)
926 netlink_set_ring(sk
, &req
, true, true);
928 #endif /* CONFIG_NETLINK_MMAP */
930 if (!sock_flag(sk
, SOCK_DEAD
)) {
931 printk(KERN_ERR
"Freeing alive netlink socket %p\n", sk
);
935 WARN_ON(atomic_read(&sk
->sk_rmem_alloc
));
936 WARN_ON(atomic_read(&sk
->sk_wmem_alloc
));
937 WARN_ON(nlk_sk(sk
)->groups
);
940 /* This lock without WQ_FLAG_EXCLUSIVE is good on UP and it is _very_ bad on
941 * SMP. Look, when several writers sleep and reader wakes them up, all but one
942 * immediately hit write lock and grab all the cpus. Exclusive sleep solves
943 * this, _but_ remember, it adds useless work on UP machines.
946 void netlink_table_grab(void)
947 __acquires(nl_table_lock
)
951 write_lock_irq(&nl_table_lock
);
953 if (atomic_read(&nl_table_users
)) {
954 DECLARE_WAITQUEUE(wait
, current
);
956 add_wait_queue_exclusive(&nl_table_wait
, &wait
);
958 set_current_state(TASK_UNINTERRUPTIBLE
);
959 if (atomic_read(&nl_table_users
) == 0)
961 write_unlock_irq(&nl_table_lock
);
963 write_lock_irq(&nl_table_lock
);
966 __set_current_state(TASK_RUNNING
);
967 remove_wait_queue(&nl_table_wait
, &wait
);
971 void netlink_table_ungrab(void)
972 __releases(nl_table_lock
)
974 write_unlock_irq(&nl_table_lock
);
975 wake_up(&nl_table_wait
);
979 netlink_lock_table(void)
981 /* read_lock() synchronizes us to netlink_table_grab */
983 read_lock(&nl_table_lock
);
984 atomic_inc(&nl_table_users
);
985 read_unlock(&nl_table_lock
);
989 netlink_unlock_table(void)
991 if (atomic_dec_and_test(&nl_table_users
))
992 wake_up(&nl_table_wait
);
995 struct netlink_compare_arg
1001 static bool netlink_compare(void *ptr
, void *arg
)
1003 struct netlink_compare_arg
*x
= arg
;
1004 struct sock
*sk
= ptr
;
1006 return nlk_sk(sk
)->portid
== x
->portid
&&
1007 net_eq(sock_net(sk
), x
->net
);
1010 static struct sock
*__netlink_lookup(struct netlink_table
*table
, u32 portid
,
1013 struct netlink_compare_arg arg
= {
1019 hash
= rhashtable_hashfn(&table
->hash
, &portid
, sizeof(portid
));
1021 return rhashtable_lookup_compare(&table
->hash
, hash
,
1022 &netlink_compare
, &arg
);
1025 static struct sock
*netlink_lookup(struct net
*net
, int protocol
, u32 portid
)
1027 struct netlink_table
*table
= &nl_table
[protocol
];
1031 sk
= __netlink_lookup(table
, portid
, net
);
1039 static const struct proto_ops netlink_ops
;
1042 netlink_update_listeners(struct sock
*sk
)
1044 struct netlink_table
*tbl
= &nl_table
[sk
->sk_protocol
];
1047 struct listeners
*listeners
;
1049 listeners
= nl_deref_protected(tbl
->listeners
);
1053 for (i
= 0; i
< NLGRPLONGS(tbl
->groups
); i
++) {
1055 sk_for_each_bound(sk
, &tbl
->mc_list
) {
1056 if (i
< NLGRPLONGS(nlk_sk(sk
)->ngroups
))
1057 mask
|= nlk_sk(sk
)->groups
[i
];
1059 listeners
->masks
[i
] = mask
;
1061 /* this function is only called with the netlink table "grabbed", which
1062 * makes sure updates are visible before bind or setsockopt return. */
1065 static int netlink_insert(struct sock
*sk
, struct net
*net
, u32 portid
)
1067 struct netlink_table
*table
= &nl_table
[sk
->sk_protocol
];
1068 int err
= -EADDRINUSE
;
1070 mutex_lock(&nl_sk_hash_lock
);
1071 if (__netlink_lookup(table
, portid
, net
))
1075 if (nlk_sk(sk
)->portid
)
1079 if (BITS_PER_LONG
> 32 && unlikely(table
->hash
.nelems
>= UINT_MAX
))
1082 nlk_sk(sk
)->portid
= portid
;
1084 rhashtable_insert(&table
->hash
, &nlk_sk(sk
)->node
, GFP_KERNEL
);
1087 mutex_unlock(&nl_sk_hash_lock
);
1091 static void netlink_remove(struct sock
*sk
)
1093 struct netlink_table
*table
;
1095 mutex_lock(&nl_sk_hash_lock
);
1096 table
= &nl_table
[sk
->sk_protocol
];
1097 if (rhashtable_remove(&table
->hash
, &nlk_sk(sk
)->node
, GFP_KERNEL
)) {
1098 WARN_ON(atomic_read(&sk
->sk_refcnt
) == 1);
1101 mutex_unlock(&nl_sk_hash_lock
);
1103 netlink_table_grab();
1104 if (nlk_sk(sk
)->subscriptions
)
1105 __sk_del_bind_node(sk
);
1106 netlink_table_ungrab();
1109 static struct proto netlink_proto
= {
1111 .owner
= THIS_MODULE
,
1112 .obj_size
= sizeof(struct netlink_sock
),
1115 static int __netlink_create(struct net
*net
, struct socket
*sock
,
1116 struct mutex
*cb_mutex
, int protocol
)
1119 struct netlink_sock
*nlk
;
1121 sock
->ops
= &netlink_ops
;
1123 sk
= sk_alloc(net
, PF_NETLINK
, GFP_KERNEL
, &netlink_proto
);
1127 sock_init_data(sock
, sk
);
1131 nlk
->cb_mutex
= cb_mutex
;
1133 nlk
->cb_mutex
= &nlk
->cb_def_mutex
;
1134 mutex_init(nlk
->cb_mutex
);
1136 init_waitqueue_head(&nlk
->wait
);
1137 #ifdef CONFIG_NETLINK_MMAP
1138 mutex_init(&nlk
->pg_vec_lock
);
1141 sk
->sk_destruct
= netlink_sock_destruct
;
1142 sk
->sk_protocol
= protocol
;
1146 static int netlink_create(struct net
*net
, struct socket
*sock
, int protocol
,
1149 struct module
*module
= NULL
;
1150 struct mutex
*cb_mutex
;
1151 struct netlink_sock
*nlk
;
1152 int (*bind
)(int group
);
1153 void (*unbind
)(int group
);
1156 sock
->state
= SS_UNCONNECTED
;
1158 if (sock
->type
!= SOCK_RAW
&& sock
->type
!= SOCK_DGRAM
)
1159 return -ESOCKTNOSUPPORT
;
1161 if (protocol
< 0 || protocol
>= MAX_LINKS
)
1162 return -EPROTONOSUPPORT
;
1164 netlink_lock_table();
1165 #ifdef CONFIG_MODULES
1166 if (!nl_table
[protocol
].registered
) {
1167 netlink_unlock_table();
1168 request_module("net-pf-%d-proto-%d", PF_NETLINK
, protocol
);
1169 netlink_lock_table();
1172 if (nl_table
[protocol
].registered
&&
1173 try_module_get(nl_table
[protocol
].module
))
1174 module
= nl_table
[protocol
].module
;
1176 err
= -EPROTONOSUPPORT
;
1177 cb_mutex
= nl_table
[protocol
].cb_mutex
;
1178 bind
= nl_table
[protocol
].bind
;
1179 unbind
= nl_table
[protocol
].unbind
;
1180 netlink_unlock_table();
1185 err
= __netlink_create(net
, sock
, cb_mutex
, protocol
);
1190 sock_prot_inuse_add(net
, &netlink_proto
, 1);
1193 nlk
= nlk_sk(sock
->sk
);
1194 nlk
->module
= module
;
1195 nlk
->netlink_bind
= bind
;
1196 nlk
->netlink_unbind
= unbind
;
1205 static int netlink_release(struct socket
*sock
)
1207 struct sock
*sk
= sock
->sk
;
1208 struct netlink_sock
*nlk
;
1218 * OK. Socket is unlinked, any packets that arrive now
1223 wake_up_interruptible_all(&nlk
->wait
);
1225 skb_queue_purge(&sk
->sk_write_queue
);
1228 struct netlink_notify n
= {
1229 .net
= sock_net(sk
),
1230 .protocol
= sk
->sk_protocol
,
1231 .portid
= nlk
->portid
,
1233 atomic_notifier_call_chain(&netlink_chain
,
1234 NETLINK_URELEASE
, &n
);
1237 module_put(nlk
->module
);
1239 netlink_table_grab();
1240 if (netlink_is_kernel(sk
)) {
1241 BUG_ON(nl_table
[sk
->sk_protocol
].registered
== 0);
1242 if (--nl_table
[sk
->sk_protocol
].registered
== 0) {
1243 struct listeners
*old
;
1245 old
= nl_deref_protected(nl_table
[sk
->sk_protocol
].listeners
);
1246 RCU_INIT_POINTER(nl_table
[sk
->sk_protocol
].listeners
, NULL
);
1247 kfree_rcu(old
, rcu
);
1248 nl_table
[sk
->sk_protocol
].module
= NULL
;
1249 nl_table
[sk
->sk_protocol
].bind
= NULL
;
1250 nl_table
[sk
->sk_protocol
].unbind
= NULL
;
1251 nl_table
[sk
->sk_protocol
].flags
= 0;
1252 nl_table
[sk
->sk_protocol
].registered
= 0;
1254 } else if (nlk
->subscriptions
) {
1255 netlink_update_listeners(sk
);
1257 netlink_table_ungrab();
1259 /* Wait for readers to complete */
1266 sock_prot_inuse_add(sock_net(sk
), &netlink_proto
, -1);
1272 static int netlink_autobind(struct socket
*sock
)
1274 struct sock
*sk
= sock
->sk
;
1275 struct net
*net
= sock_net(sk
);
1276 struct netlink_table
*table
= &nl_table
[sk
->sk_protocol
];
1277 s32 portid
= task_tgid_vnr(current
);
1279 static s32 rover
= -4097;
1284 if (__netlink_lookup(table
, portid
, net
)) {
1285 /* Bind collision, search negative portid values. */
1294 err
= netlink_insert(sk
, net
, portid
);
1295 if (err
== -EADDRINUSE
)
1298 /* If 2 threads race to autobind, that is fine. */
1306 * __netlink_ns_capable - General netlink message capability test
1307 * @nsp: NETLINK_CB of the socket buffer holding a netlink command from userspace.
1308 * @user_ns: The user namespace of the capability to use
1309 * @cap: The capability to use
1311 * Test to see if the opener of the socket we received the message
1312 * from had when the netlink socket was created and the sender of the
1313 * message has has the capability @cap in the user namespace @user_ns.
1315 bool __netlink_ns_capable(const struct netlink_skb_parms
*nsp
,
1316 struct user_namespace
*user_ns
, int cap
)
1318 return ((nsp
->flags
& NETLINK_SKB_DST
) ||
1319 file_ns_capable(nsp
->sk
->sk_socket
->file
, user_ns
, cap
)) &&
1320 ns_capable(user_ns
, cap
);
1322 EXPORT_SYMBOL(__netlink_ns_capable
);
1325 * netlink_ns_capable - General netlink message capability test
1326 * @skb: socket buffer holding a netlink command from userspace
1327 * @user_ns: The user namespace of the capability to use
1328 * @cap: The capability to use
1330 * Test to see if the opener of the socket we received the message
1331 * from had when the netlink socket was created and the sender of the
1332 * message has has the capability @cap in the user namespace @user_ns.
1334 bool netlink_ns_capable(const struct sk_buff
*skb
,
1335 struct user_namespace
*user_ns
, int cap
)
1337 return __netlink_ns_capable(&NETLINK_CB(skb
), user_ns
, cap
);
1339 EXPORT_SYMBOL(netlink_ns_capable
);
1342 * netlink_capable - Netlink global message capability test
1343 * @skb: socket buffer holding a netlink command from userspace
1344 * @cap: The capability to use
1346 * Test to see if the opener of the socket we received the message
1347 * from had when the netlink socket was created and the sender of the
1348 * message has has the capability @cap in all user namespaces.
1350 bool netlink_capable(const struct sk_buff
*skb
, int cap
)
1352 return netlink_ns_capable(skb
, &init_user_ns
, cap
);
1354 EXPORT_SYMBOL(netlink_capable
);
1357 * netlink_net_capable - Netlink network namespace message capability test
1358 * @skb: socket buffer holding a netlink command from userspace
1359 * @cap: The capability to use
1361 * Test to see if the opener of the socket we received the message
1362 * from had when the netlink socket was created and the sender of the
1363 * message has has the capability @cap over the network namespace of
1364 * the socket we received the message from.
1366 bool netlink_net_capable(const struct sk_buff
*skb
, int cap
)
1368 return netlink_ns_capable(skb
, sock_net(skb
->sk
)->user_ns
, cap
);
1370 EXPORT_SYMBOL(netlink_net_capable
);
1372 static inline int netlink_allowed(const struct socket
*sock
, unsigned int flag
)
1374 return (nl_table
[sock
->sk
->sk_protocol
].flags
& flag
) ||
1375 ns_capable(sock_net(sock
->sk
)->user_ns
, CAP_NET_ADMIN
);
1379 netlink_update_subscriptions(struct sock
*sk
, unsigned int subscriptions
)
1381 struct netlink_sock
*nlk
= nlk_sk(sk
);
1383 if (nlk
->subscriptions
&& !subscriptions
)
1384 __sk_del_bind_node(sk
);
1385 else if (!nlk
->subscriptions
&& subscriptions
)
1386 sk_add_bind_node(sk
, &nl_table
[sk
->sk_protocol
].mc_list
);
1387 nlk
->subscriptions
= subscriptions
;
1390 static int netlink_realloc_groups(struct sock
*sk
)
1392 struct netlink_sock
*nlk
= nlk_sk(sk
);
1393 unsigned int groups
;
1394 unsigned long *new_groups
;
1397 netlink_table_grab();
1399 groups
= nl_table
[sk
->sk_protocol
].groups
;
1400 if (!nl_table
[sk
->sk_protocol
].registered
) {
1405 if (nlk
->ngroups
>= groups
)
1408 new_groups
= krealloc(nlk
->groups
, NLGRPSZ(groups
), GFP_ATOMIC
);
1409 if (new_groups
== NULL
) {
1413 memset((char *)new_groups
+ NLGRPSZ(nlk
->ngroups
), 0,
1414 NLGRPSZ(groups
) - NLGRPSZ(nlk
->ngroups
));
1416 nlk
->groups
= new_groups
;
1417 nlk
->ngroups
= groups
;
1419 netlink_table_ungrab();
1423 static void netlink_unbind(int group
, long unsigned int groups
,
1424 struct netlink_sock
*nlk
)
1428 if (!nlk
->netlink_unbind
)
1431 for (undo
= 0; undo
< group
; undo
++)
1432 if (test_bit(group
, &groups
))
1433 nlk
->netlink_unbind(undo
);
1436 static int netlink_bind(struct socket
*sock
, struct sockaddr
*addr
,
1439 struct sock
*sk
= sock
->sk
;
1440 struct net
*net
= sock_net(sk
);
1441 struct netlink_sock
*nlk
= nlk_sk(sk
);
1442 struct sockaddr_nl
*nladdr
= (struct sockaddr_nl
*)addr
;
1444 long unsigned int groups
= nladdr
->nl_groups
;
1446 if (addr_len
< sizeof(struct sockaddr_nl
))
1449 if (nladdr
->nl_family
!= AF_NETLINK
)
1452 /* Only superuser is allowed to listen multicasts */
1454 if (!netlink_allowed(sock
, NL_CFG_F_NONROOT_RECV
))
1456 err
= netlink_realloc_groups(sk
);
1462 if (nladdr
->nl_pid
!= nlk
->portid
)
1465 if (nlk
->netlink_bind
&& groups
) {
1468 for (group
= 0; group
< nlk
->ngroups
; group
++) {
1469 if (!test_bit(group
, &groups
))
1471 err
= nlk
->netlink_bind(group
);
1474 netlink_unbind(group
, groups
, nlk
);
1480 err
= nladdr
->nl_pid
?
1481 netlink_insert(sk
, net
, nladdr
->nl_pid
) :
1482 netlink_autobind(sock
);
1484 netlink_unbind(nlk
->ngroups
- 1, groups
, nlk
);
1489 if (!groups
&& (nlk
->groups
== NULL
|| !(u32
)nlk
->groups
[0]))
1492 netlink_table_grab();
1493 netlink_update_subscriptions(sk
, nlk
->subscriptions
+
1495 hweight32(nlk
->groups
[0]));
1496 nlk
->groups
[0] = (nlk
->groups
[0] & ~0xffffffffUL
) | groups
;
1497 netlink_update_listeners(sk
);
1498 netlink_table_ungrab();
1503 static int netlink_connect(struct socket
*sock
, struct sockaddr
*addr
,
1504 int alen
, int flags
)
1507 struct sock
*sk
= sock
->sk
;
1508 struct netlink_sock
*nlk
= nlk_sk(sk
);
1509 struct sockaddr_nl
*nladdr
= (struct sockaddr_nl
*)addr
;
1511 if (alen
< sizeof(addr
->sa_family
))
1514 if (addr
->sa_family
== AF_UNSPEC
) {
1515 sk
->sk_state
= NETLINK_UNCONNECTED
;
1516 nlk
->dst_portid
= 0;
1520 if (addr
->sa_family
!= AF_NETLINK
)
1523 if ((nladdr
->nl_groups
|| nladdr
->nl_pid
) &&
1524 !netlink_allowed(sock
, NL_CFG_F_NONROOT_SEND
))
1528 err
= netlink_autobind(sock
);
1531 sk
->sk_state
= NETLINK_CONNECTED
;
1532 nlk
->dst_portid
= nladdr
->nl_pid
;
1533 nlk
->dst_group
= ffs(nladdr
->nl_groups
);
1539 static int netlink_getname(struct socket
*sock
, struct sockaddr
*addr
,
1540 int *addr_len
, int peer
)
1542 struct sock
*sk
= sock
->sk
;
1543 struct netlink_sock
*nlk
= nlk_sk(sk
);
1544 DECLARE_SOCKADDR(struct sockaddr_nl
*, nladdr
, addr
);
1546 nladdr
->nl_family
= AF_NETLINK
;
1548 *addr_len
= sizeof(*nladdr
);
1551 nladdr
->nl_pid
= nlk
->dst_portid
;
1552 nladdr
->nl_groups
= netlink_group_mask(nlk
->dst_group
);
1554 nladdr
->nl_pid
= nlk
->portid
;
1555 nladdr
->nl_groups
= nlk
->groups
? nlk
->groups
[0] : 0;
1560 static struct sock
*netlink_getsockbyportid(struct sock
*ssk
, u32 portid
)
1563 struct netlink_sock
*nlk
;
1565 sock
= netlink_lookup(sock_net(ssk
), ssk
->sk_protocol
, portid
);
1567 return ERR_PTR(-ECONNREFUSED
);
1569 /* Don't bother queuing skb if kernel socket has no input function */
1571 if (sock
->sk_state
== NETLINK_CONNECTED
&&
1572 nlk
->dst_portid
!= nlk_sk(ssk
)->portid
) {
1574 return ERR_PTR(-ECONNREFUSED
);
1579 struct sock
*netlink_getsockbyfilp(struct file
*filp
)
1581 struct inode
*inode
= file_inode(filp
);
1584 if (!S_ISSOCK(inode
->i_mode
))
1585 return ERR_PTR(-ENOTSOCK
);
1587 sock
= SOCKET_I(inode
)->sk
;
1588 if (sock
->sk_family
!= AF_NETLINK
)
1589 return ERR_PTR(-EINVAL
);
1595 static struct sk_buff
*netlink_alloc_large_skb(unsigned int size
,
1598 struct sk_buff
*skb
;
1601 if (size
<= NLMSG_GOODSIZE
|| broadcast
)
1602 return alloc_skb(size
, GFP_KERNEL
);
1604 size
= SKB_DATA_ALIGN(size
) +
1605 SKB_DATA_ALIGN(sizeof(struct skb_shared_info
));
1607 data
= vmalloc(size
);
1611 skb
= build_skb(data
, size
);
1616 skb
->destructor
= netlink_skb_destructor
;
1623 * Attach a skb to a netlink socket.
1624 * The caller must hold a reference to the destination socket. On error, the
1625 * reference is dropped. The skb is not send to the destination, just all
1626 * all error checks are performed and memory in the queue is reserved.
1628 * < 0: error. skb freed, reference to sock dropped.
1630 * 1: repeat lookup - reference dropped while waiting for socket memory.
1632 int netlink_attachskb(struct sock
*sk
, struct sk_buff
*skb
,
1633 long *timeo
, struct sock
*ssk
)
1635 struct netlink_sock
*nlk
;
1639 if ((atomic_read(&sk
->sk_rmem_alloc
) > sk
->sk_rcvbuf
||
1640 test_bit(NETLINK_CONGESTED
, &nlk
->state
)) &&
1641 !netlink_skb_is_mmaped(skb
)) {
1642 DECLARE_WAITQUEUE(wait
, current
);
1644 if (!ssk
|| netlink_is_kernel(ssk
))
1645 netlink_overrun(sk
);
1651 __set_current_state(TASK_INTERRUPTIBLE
);
1652 add_wait_queue(&nlk
->wait
, &wait
);
1654 if ((atomic_read(&sk
->sk_rmem_alloc
) > sk
->sk_rcvbuf
||
1655 test_bit(NETLINK_CONGESTED
, &nlk
->state
)) &&
1656 !sock_flag(sk
, SOCK_DEAD
))
1657 *timeo
= schedule_timeout(*timeo
);
1659 __set_current_state(TASK_RUNNING
);
1660 remove_wait_queue(&nlk
->wait
, &wait
);
1663 if (signal_pending(current
)) {
1665 return sock_intr_errno(*timeo
);
1669 netlink_skb_set_owner_r(skb
, sk
);
1673 static int __netlink_sendskb(struct sock
*sk
, struct sk_buff
*skb
)
1677 netlink_deliver_tap(skb
);
1679 #ifdef CONFIG_NETLINK_MMAP
1680 if (netlink_skb_is_mmaped(skb
))
1681 netlink_queue_mmaped_skb(sk
, skb
);
1682 else if (netlink_rx_is_mmaped(sk
))
1683 netlink_ring_set_copied(sk
, skb
);
1685 #endif /* CONFIG_NETLINK_MMAP */
1686 skb_queue_tail(&sk
->sk_receive_queue
, skb
);
1687 sk
->sk_data_ready(sk
);
1691 int netlink_sendskb(struct sock
*sk
, struct sk_buff
*skb
)
1693 int len
= __netlink_sendskb(sk
, skb
);
1699 void netlink_detachskb(struct sock
*sk
, struct sk_buff
*skb
)
1705 static struct sk_buff
*netlink_trim(struct sk_buff
*skb
, gfp_t allocation
)
1709 WARN_ON(skb
->sk
!= NULL
);
1710 if (netlink_skb_is_mmaped(skb
))
1713 delta
= skb
->end
- skb
->tail
;
1714 if (is_vmalloc_addr(skb
->head
) || delta
* 2 < skb
->truesize
)
1717 if (skb_shared(skb
)) {
1718 struct sk_buff
*nskb
= skb_clone(skb
, allocation
);
1725 if (!pskb_expand_head(skb
, 0, -delta
, allocation
))
1726 skb
->truesize
-= delta
;
1731 static int netlink_unicast_kernel(struct sock
*sk
, struct sk_buff
*skb
,
1735 struct netlink_sock
*nlk
= nlk_sk(sk
);
1737 ret
= -ECONNREFUSED
;
1738 if (nlk
->netlink_rcv
!= NULL
) {
1740 netlink_skb_set_owner_r(skb
, sk
);
1741 NETLINK_CB(skb
).sk
= ssk
;
1742 netlink_deliver_tap_kernel(sk
, ssk
, skb
);
1743 nlk
->netlink_rcv(skb
);
1752 int netlink_unicast(struct sock
*ssk
, struct sk_buff
*skb
,
1753 u32 portid
, int nonblock
)
1759 skb
= netlink_trim(skb
, gfp_any());
1761 timeo
= sock_sndtimeo(ssk
, nonblock
);
1763 sk
= netlink_getsockbyportid(ssk
, portid
);
1768 if (netlink_is_kernel(sk
))
1769 return netlink_unicast_kernel(sk
, skb
, ssk
);
1771 if (sk_filter(sk
, skb
)) {
1778 err
= netlink_attachskb(sk
, skb
, &timeo
, ssk
);
1784 return netlink_sendskb(sk
, skb
);
1786 EXPORT_SYMBOL(netlink_unicast
);
1788 struct sk_buff
*netlink_alloc_skb(struct sock
*ssk
, unsigned int size
,
1789 u32 dst_portid
, gfp_t gfp_mask
)
1791 #ifdef CONFIG_NETLINK_MMAP
1792 struct sock
*sk
= NULL
;
1793 struct sk_buff
*skb
;
1794 struct netlink_ring
*ring
;
1795 struct nl_mmap_hdr
*hdr
;
1796 unsigned int maxlen
;
1798 sk
= netlink_getsockbyportid(ssk
, dst_portid
);
1802 ring
= &nlk_sk(sk
)->rx_ring
;
1803 /* fast-path without atomic ops for common case: non-mmaped receiver */
1804 if (ring
->pg_vec
== NULL
)
1807 if (ring
->frame_size
- NL_MMAP_HDRLEN
< size
)
1810 skb
= alloc_skb_head(gfp_mask
);
1814 spin_lock_bh(&sk
->sk_receive_queue
.lock
);
1815 /* check again under lock */
1816 if (ring
->pg_vec
== NULL
)
1819 /* check again under lock */
1820 maxlen
= ring
->frame_size
- NL_MMAP_HDRLEN
;
1824 netlink_forward_ring(ring
);
1825 hdr
= netlink_current_frame(ring
, NL_MMAP_STATUS_UNUSED
);
1828 netlink_ring_setup_skb(skb
, sk
, ring
, hdr
);
1829 netlink_set_status(hdr
, NL_MMAP_STATUS_RESERVED
);
1830 atomic_inc(&ring
->pending
);
1831 netlink_increment_head(ring
);
1833 spin_unlock_bh(&sk
->sk_receive_queue
.lock
);
1838 spin_unlock_bh(&sk
->sk_receive_queue
.lock
);
1839 netlink_overrun(sk
);
1846 spin_unlock_bh(&sk
->sk_receive_queue
.lock
);
1851 return alloc_skb(size
, gfp_mask
);
1853 EXPORT_SYMBOL_GPL(netlink_alloc_skb
);
1855 int netlink_has_listeners(struct sock
*sk
, unsigned int group
)
1858 struct listeners
*listeners
;
1860 BUG_ON(!netlink_is_kernel(sk
));
1863 listeners
= rcu_dereference(nl_table
[sk
->sk_protocol
].listeners
);
1865 if (listeners
&& group
- 1 < nl_table
[sk
->sk_protocol
].groups
)
1866 res
= test_bit(group
- 1, listeners
->masks
);
1872 EXPORT_SYMBOL_GPL(netlink_has_listeners
);
1874 static int netlink_broadcast_deliver(struct sock
*sk
, struct sk_buff
*skb
)
1876 struct netlink_sock
*nlk
= nlk_sk(sk
);
1878 if (atomic_read(&sk
->sk_rmem_alloc
) <= sk
->sk_rcvbuf
&&
1879 !test_bit(NETLINK_CONGESTED
, &nlk
->state
)) {
1880 netlink_skb_set_owner_r(skb
, sk
);
1881 __netlink_sendskb(sk
, skb
);
1882 return atomic_read(&sk
->sk_rmem_alloc
) > (sk
->sk_rcvbuf
>> 1);
1887 struct netlink_broadcast_data
{
1888 struct sock
*exclude_sk
;
1893 int delivery_failure
;
1897 struct sk_buff
*skb
, *skb2
;
1898 int (*tx_filter
)(struct sock
*dsk
, struct sk_buff
*skb
, void *data
);
1902 static void do_one_broadcast(struct sock
*sk
,
1903 struct netlink_broadcast_data
*p
)
1905 struct netlink_sock
*nlk
= nlk_sk(sk
);
1908 if (p
->exclude_sk
== sk
)
1911 if (nlk
->portid
== p
->portid
|| p
->group
- 1 >= nlk
->ngroups
||
1912 !test_bit(p
->group
- 1, nlk
->groups
))
1915 if (!net_eq(sock_net(sk
), p
->net
))
1919 netlink_overrun(sk
);
1924 if (p
->skb2
== NULL
) {
1925 if (skb_shared(p
->skb
)) {
1926 p
->skb2
= skb_clone(p
->skb
, p
->allocation
);
1928 p
->skb2
= skb_get(p
->skb
);
1930 * skb ownership may have been set when
1931 * delivered to a previous socket.
1933 skb_orphan(p
->skb2
);
1936 if (p
->skb2
== NULL
) {
1937 netlink_overrun(sk
);
1938 /* Clone failed. Notify ALL listeners. */
1940 if (nlk
->flags
& NETLINK_BROADCAST_SEND_ERROR
)
1941 p
->delivery_failure
= 1;
1942 } else if (p
->tx_filter
&& p
->tx_filter(sk
, p
->skb2
, p
->tx_data
)) {
1945 } else if (sk_filter(sk
, p
->skb2
)) {
1948 } else if ((val
= netlink_broadcast_deliver(sk
, p
->skb2
)) < 0) {
1949 netlink_overrun(sk
);
1950 if (nlk
->flags
& NETLINK_BROADCAST_SEND_ERROR
)
1951 p
->delivery_failure
= 1;
1953 p
->congested
|= val
;
1960 int netlink_broadcast_filtered(struct sock
*ssk
, struct sk_buff
*skb
, u32 portid
,
1961 u32 group
, gfp_t allocation
,
1962 int (*filter
)(struct sock
*dsk
, struct sk_buff
*skb
, void *data
),
1965 struct net
*net
= sock_net(ssk
);
1966 struct netlink_broadcast_data info
;
1969 skb
= netlink_trim(skb
, allocation
);
1971 info
.exclude_sk
= ssk
;
1973 info
.portid
= portid
;
1976 info
.delivery_failure
= 0;
1979 info
.allocation
= allocation
;
1982 info
.tx_filter
= filter
;
1983 info
.tx_data
= filter_data
;
1985 /* While we sleep in clone, do not allow to change socket list */
1987 netlink_lock_table();
1989 sk_for_each_bound(sk
, &nl_table
[ssk
->sk_protocol
].mc_list
)
1990 do_one_broadcast(sk
, &info
);
1994 netlink_unlock_table();
1996 if (info
.delivery_failure
) {
1997 kfree_skb(info
.skb2
);
2000 consume_skb(info
.skb2
);
2002 if (info
.delivered
) {
2003 if (info
.congested
&& (allocation
& __GFP_WAIT
))
2009 EXPORT_SYMBOL(netlink_broadcast_filtered
);
2011 int netlink_broadcast(struct sock
*ssk
, struct sk_buff
*skb
, u32 portid
,
2012 u32 group
, gfp_t allocation
)
2014 return netlink_broadcast_filtered(ssk
, skb
, portid
, group
, allocation
,
2017 EXPORT_SYMBOL(netlink_broadcast
);
2019 struct netlink_set_err_data
{
2020 struct sock
*exclude_sk
;
2026 static int do_one_set_err(struct sock
*sk
, struct netlink_set_err_data
*p
)
2028 struct netlink_sock
*nlk
= nlk_sk(sk
);
2031 if (sk
== p
->exclude_sk
)
2034 if (!net_eq(sock_net(sk
), sock_net(p
->exclude_sk
)))
2037 if (nlk
->portid
== p
->portid
|| p
->group
- 1 >= nlk
->ngroups
||
2038 !test_bit(p
->group
- 1, nlk
->groups
))
2041 if (p
->code
== ENOBUFS
&& nlk
->flags
& NETLINK_RECV_NO_ENOBUFS
) {
2046 sk
->sk_err
= p
->code
;
2047 sk
->sk_error_report(sk
);
2053 * netlink_set_err - report error to broadcast listeners
2054 * @ssk: the kernel netlink socket, as returned by netlink_kernel_create()
2055 * @portid: the PORTID of a process that we want to skip (if any)
2056 * @group: the broadcast group that will notice the error
2057 * @code: error code, must be negative (as usual in kernelspace)
2059 * This function returns the number of broadcast listeners that have set the
2060 * NETLINK_RECV_NO_ENOBUFS socket option.
2062 int netlink_set_err(struct sock
*ssk
, u32 portid
, u32 group
, int code
)
2064 struct netlink_set_err_data info
;
2068 info
.exclude_sk
= ssk
;
2069 info
.portid
= portid
;
2071 /* sk->sk_err wants a positive error value */
2074 read_lock(&nl_table_lock
);
2076 sk_for_each_bound(sk
, &nl_table
[ssk
->sk_protocol
].mc_list
)
2077 ret
+= do_one_set_err(sk
, &info
);
2079 read_unlock(&nl_table_lock
);
2082 EXPORT_SYMBOL(netlink_set_err
);
2084 /* must be called with netlink table grabbed */
2085 static void netlink_update_socket_mc(struct netlink_sock
*nlk
,
2089 int old
, new = !!is_new
, subscriptions
;
2091 old
= test_bit(group
- 1, nlk
->groups
);
2092 subscriptions
= nlk
->subscriptions
- old
+ new;
2094 __set_bit(group
- 1, nlk
->groups
);
2096 __clear_bit(group
- 1, nlk
->groups
);
2097 netlink_update_subscriptions(&nlk
->sk
, subscriptions
);
2098 netlink_update_listeners(&nlk
->sk
);
2101 static int netlink_setsockopt(struct socket
*sock
, int level
, int optname
,
2102 char __user
*optval
, unsigned int optlen
)
2104 struct sock
*sk
= sock
->sk
;
2105 struct netlink_sock
*nlk
= nlk_sk(sk
);
2106 unsigned int val
= 0;
2109 if (level
!= SOL_NETLINK
)
2110 return -ENOPROTOOPT
;
2112 if (optname
!= NETLINK_RX_RING
&& optname
!= NETLINK_TX_RING
&&
2113 optlen
>= sizeof(int) &&
2114 get_user(val
, (unsigned int __user
*)optval
))
2118 case NETLINK_PKTINFO
:
2120 nlk
->flags
|= NETLINK_RECV_PKTINFO
;
2122 nlk
->flags
&= ~NETLINK_RECV_PKTINFO
;
2125 case NETLINK_ADD_MEMBERSHIP
:
2126 case NETLINK_DROP_MEMBERSHIP
: {
2127 if (!netlink_allowed(sock
, NL_CFG_F_NONROOT_RECV
))
2129 err
= netlink_realloc_groups(sk
);
2132 if (!val
|| val
- 1 >= nlk
->ngroups
)
2134 if (optname
== NETLINK_ADD_MEMBERSHIP
&& nlk
->netlink_bind
) {
2135 err
= nlk
->netlink_bind(val
);
2139 netlink_table_grab();
2140 netlink_update_socket_mc(nlk
, val
,
2141 optname
== NETLINK_ADD_MEMBERSHIP
);
2142 netlink_table_ungrab();
2143 if (optname
== NETLINK_DROP_MEMBERSHIP
&& nlk
->netlink_unbind
)
2144 nlk
->netlink_unbind(val
);
2149 case NETLINK_BROADCAST_ERROR
:
2151 nlk
->flags
|= NETLINK_BROADCAST_SEND_ERROR
;
2153 nlk
->flags
&= ~NETLINK_BROADCAST_SEND_ERROR
;
2156 case NETLINK_NO_ENOBUFS
:
2158 nlk
->flags
|= NETLINK_RECV_NO_ENOBUFS
;
2159 clear_bit(NETLINK_CONGESTED
, &nlk
->state
);
2160 wake_up_interruptible(&nlk
->wait
);
2162 nlk
->flags
&= ~NETLINK_RECV_NO_ENOBUFS
;
2166 #ifdef CONFIG_NETLINK_MMAP
2167 case NETLINK_RX_RING
:
2168 case NETLINK_TX_RING
: {
2169 struct nl_mmap_req req
;
2171 /* Rings might consume more memory than queue limits, require
2174 if (!capable(CAP_NET_ADMIN
))
2176 if (optlen
< sizeof(req
))
2178 if (copy_from_user(&req
, optval
, sizeof(req
)))
2180 err
= netlink_set_ring(sk
, &req
, false,
2181 optname
== NETLINK_TX_RING
);
2184 #endif /* CONFIG_NETLINK_MMAP */
2191 static int netlink_getsockopt(struct socket
*sock
, int level
, int optname
,
2192 char __user
*optval
, int __user
*optlen
)
2194 struct sock
*sk
= sock
->sk
;
2195 struct netlink_sock
*nlk
= nlk_sk(sk
);
2198 if (level
!= SOL_NETLINK
)
2199 return -ENOPROTOOPT
;
2201 if (get_user(len
, optlen
))
2207 case NETLINK_PKTINFO
:
2208 if (len
< sizeof(int))
2211 val
= nlk
->flags
& NETLINK_RECV_PKTINFO
? 1 : 0;
2212 if (put_user(len
, optlen
) ||
2213 put_user(val
, optval
))
2217 case NETLINK_BROADCAST_ERROR
:
2218 if (len
< sizeof(int))
2221 val
= nlk
->flags
& NETLINK_BROADCAST_SEND_ERROR
? 1 : 0;
2222 if (put_user(len
, optlen
) ||
2223 put_user(val
, optval
))
2227 case NETLINK_NO_ENOBUFS
:
2228 if (len
< sizeof(int))
2231 val
= nlk
->flags
& NETLINK_RECV_NO_ENOBUFS
? 1 : 0;
2232 if (put_user(len
, optlen
) ||
2233 put_user(val
, optval
))
2243 static void netlink_cmsg_recv_pktinfo(struct msghdr
*msg
, struct sk_buff
*skb
)
2245 struct nl_pktinfo info
;
2247 info
.group
= NETLINK_CB(skb
).dst_group
;
2248 put_cmsg(msg
, SOL_NETLINK
, NETLINK_PKTINFO
, sizeof(info
), &info
);
2251 static int netlink_sendmsg(struct kiocb
*kiocb
, struct socket
*sock
,
2252 struct msghdr
*msg
, size_t len
)
2254 struct sock_iocb
*siocb
= kiocb_to_siocb(kiocb
);
2255 struct sock
*sk
= sock
->sk
;
2256 struct netlink_sock
*nlk
= nlk_sk(sk
);
2257 DECLARE_SOCKADDR(struct sockaddr_nl
*, addr
, msg
->msg_name
);
2260 struct sk_buff
*skb
;
2262 struct scm_cookie scm
;
2263 u32 netlink_skb_flags
= 0;
2265 if (msg
->msg_flags
&MSG_OOB
)
2268 if (NULL
== siocb
->scm
)
2271 err
= scm_send(sock
, msg
, siocb
->scm
, true);
2275 if (msg
->msg_namelen
) {
2277 if (addr
->nl_family
!= AF_NETLINK
)
2279 dst_portid
= addr
->nl_pid
;
2280 dst_group
= ffs(addr
->nl_groups
);
2282 if ((dst_group
|| dst_portid
) &&
2283 !netlink_allowed(sock
, NL_CFG_F_NONROOT_SEND
))
2285 netlink_skb_flags
|= NETLINK_SKB_DST
;
2287 dst_portid
= nlk
->dst_portid
;
2288 dst_group
= nlk
->dst_group
;
2292 err
= netlink_autobind(sock
);
2297 if (netlink_tx_is_mmaped(sk
) &&
2298 msg
->msg_iov
->iov_base
== NULL
) {
2299 err
= netlink_mmap_sendmsg(sk
, msg
, dst_portid
, dst_group
,
2305 if (len
> sk
->sk_sndbuf
- 32)
2308 skb
= netlink_alloc_large_skb(len
, dst_group
);
2312 NETLINK_CB(skb
).portid
= nlk
->portid
;
2313 NETLINK_CB(skb
).dst_group
= dst_group
;
2314 NETLINK_CB(skb
).creds
= siocb
->scm
->creds
;
2315 NETLINK_CB(skb
).flags
= netlink_skb_flags
;
2318 if (memcpy_fromiovec(skb_put(skb
, len
), msg
->msg_iov
, len
)) {
2323 err
= security_netlink_send(sk
, skb
);
2330 atomic_inc(&skb
->users
);
2331 netlink_broadcast(sk
, skb
, dst_portid
, dst_group
, GFP_KERNEL
);
2333 err
= netlink_unicast(sk
, skb
, dst_portid
, msg
->msg_flags
&MSG_DONTWAIT
);
2336 scm_destroy(siocb
->scm
);
2340 static int netlink_recvmsg(struct kiocb
*kiocb
, struct socket
*sock
,
2341 struct msghdr
*msg
, size_t len
,
2344 struct sock_iocb
*siocb
= kiocb_to_siocb(kiocb
);
2345 struct scm_cookie scm
;
2346 struct sock
*sk
= sock
->sk
;
2347 struct netlink_sock
*nlk
= nlk_sk(sk
);
2348 int noblock
= flags
&MSG_DONTWAIT
;
2350 struct sk_buff
*skb
, *data_skb
;
2358 skb
= skb_recv_datagram(sk
, flags
, noblock
, &err
);
2364 #ifdef CONFIG_COMPAT_NETLINK_MESSAGES
2365 if (unlikely(skb_shinfo(skb
)->frag_list
)) {
2367 * If this skb has a frag_list, then here that means that we
2368 * will have to use the frag_list skb's data for compat tasks
2369 * and the regular skb's data for normal (non-compat) tasks.
2371 * If we need to send the compat skb, assign it to the
2372 * 'data_skb' variable so that it will be used below for data
2373 * copying. We keep 'skb' for everything else, including
2374 * freeing both later.
2376 if (flags
& MSG_CMSG_COMPAT
)
2377 data_skb
= skb_shinfo(skb
)->frag_list
;
2381 /* Record the max length of recvmsg() calls for future allocations */
2382 nlk
->max_recvmsg_len
= max(nlk
->max_recvmsg_len
, len
);
2383 nlk
->max_recvmsg_len
= min_t(size_t, nlk
->max_recvmsg_len
,
2386 copied
= data_skb
->len
;
2388 msg
->msg_flags
|= MSG_TRUNC
;
2392 skb_reset_transport_header(data_skb
);
2393 err
= skb_copy_datagram_iovec(data_skb
, 0, msg
->msg_iov
, copied
);
2395 if (msg
->msg_name
) {
2396 DECLARE_SOCKADDR(struct sockaddr_nl
*, addr
, msg
->msg_name
);
2397 addr
->nl_family
= AF_NETLINK
;
2399 addr
->nl_pid
= NETLINK_CB(skb
).portid
;
2400 addr
->nl_groups
= netlink_group_mask(NETLINK_CB(skb
).dst_group
);
2401 msg
->msg_namelen
= sizeof(*addr
);
2404 if (nlk
->flags
& NETLINK_RECV_PKTINFO
)
2405 netlink_cmsg_recv_pktinfo(msg
, skb
);
2407 if (NULL
== siocb
->scm
) {
2408 memset(&scm
, 0, sizeof(scm
));
2411 siocb
->scm
->creds
= *NETLINK_CREDS(skb
);
2412 if (flags
& MSG_TRUNC
)
2413 copied
= data_skb
->len
;
2415 skb_free_datagram(sk
, skb
);
2417 if (nlk
->cb_running
&&
2418 atomic_read(&sk
->sk_rmem_alloc
) <= sk
->sk_rcvbuf
/ 2) {
2419 ret
= netlink_dump(sk
);
2422 sk
->sk_error_report(sk
);
2426 scm_recv(sock
, msg
, siocb
->scm
, flags
);
2428 netlink_rcv_wake(sk
);
2429 return err
? : copied
;
2432 static void netlink_data_ready(struct sock
*sk
)
2438 * We export these functions to other modules. They provide a
2439 * complete set of kernel non-blocking support for message
2444 __netlink_kernel_create(struct net
*net
, int unit
, struct module
*module
,
2445 struct netlink_kernel_cfg
*cfg
)
2447 struct socket
*sock
;
2449 struct netlink_sock
*nlk
;
2450 struct listeners
*listeners
= NULL
;
2451 struct mutex
*cb_mutex
= cfg
? cfg
->cb_mutex
: NULL
;
2452 unsigned int groups
;
2456 if (unit
< 0 || unit
>= MAX_LINKS
)
2459 if (sock_create_lite(PF_NETLINK
, SOCK_DGRAM
, unit
, &sock
))
2463 * We have to just have a reference on the net from sk, but don't
2464 * get_net it. Besides, we cannot get and then put the net here.
2465 * So we create one inside init_net and the move it to net.
2468 if (__netlink_create(&init_net
, sock
, cb_mutex
, unit
) < 0)
2469 goto out_sock_release_nosk
;
2472 sk_change_net(sk
, net
);
2474 if (!cfg
|| cfg
->groups
< 32)
2477 groups
= cfg
->groups
;
2479 listeners
= kzalloc(sizeof(*listeners
) + NLGRPSZ(groups
), GFP_KERNEL
);
2481 goto out_sock_release
;
2483 sk
->sk_data_ready
= netlink_data_ready
;
2484 if (cfg
&& cfg
->input
)
2485 nlk_sk(sk
)->netlink_rcv
= cfg
->input
;
2487 if (netlink_insert(sk
, net
, 0))
2488 goto out_sock_release
;
2491 nlk
->flags
|= NETLINK_KERNEL_SOCKET
;
2493 netlink_table_grab();
2494 if (!nl_table
[unit
].registered
) {
2495 nl_table
[unit
].groups
= groups
;
2496 rcu_assign_pointer(nl_table
[unit
].listeners
, listeners
);
2497 nl_table
[unit
].cb_mutex
= cb_mutex
;
2498 nl_table
[unit
].module
= module
;
2500 nl_table
[unit
].bind
= cfg
->bind
;
2501 nl_table
[unit
].flags
= cfg
->flags
;
2503 nl_table
[unit
].compare
= cfg
->compare
;
2505 nl_table
[unit
].registered
= 1;
2508 nl_table
[unit
].registered
++;
2510 netlink_table_ungrab();
2515 netlink_kernel_release(sk
);
2518 out_sock_release_nosk
:
2522 EXPORT_SYMBOL(__netlink_kernel_create
);
2525 netlink_kernel_release(struct sock
*sk
)
2527 sk_release_kernel(sk
);
2529 EXPORT_SYMBOL(netlink_kernel_release
);
2531 int __netlink_change_ngroups(struct sock
*sk
, unsigned int groups
)
2533 struct listeners
*new, *old
;
2534 struct netlink_table
*tbl
= &nl_table
[sk
->sk_protocol
];
2539 if (NLGRPSZ(tbl
->groups
) < NLGRPSZ(groups
)) {
2540 new = kzalloc(sizeof(*new) + NLGRPSZ(groups
), GFP_ATOMIC
);
2543 old
= nl_deref_protected(tbl
->listeners
);
2544 memcpy(new->masks
, old
->masks
, NLGRPSZ(tbl
->groups
));
2545 rcu_assign_pointer(tbl
->listeners
, new);
2547 kfree_rcu(old
, rcu
);
2549 tbl
->groups
= groups
;
2555 * netlink_change_ngroups - change number of multicast groups
2557 * This changes the number of multicast groups that are available
2558 * on a certain netlink family. Note that it is not possible to
2559 * change the number of groups to below 32. Also note that it does
2560 * not implicitly call netlink_clear_multicast_users() when the
2561 * number of groups is reduced.
2563 * @sk: The kernel netlink socket, as returned by netlink_kernel_create().
2564 * @groups: The new number of groups.
2566 int netlink_change_ngroups(struct sock
*sk
, unsigned int groups
)
2570 netlink_table_grab();
2571 err
= __netlink_change_ngroups(sk
, groups
);
2572 netlink_table_ungrab();
2577 void __netlink_clear_multicast_users(struct sock
*ksk
, unsigned int group
)
2580 struct netlink_table
*tbl
= &nl_table
[ksk
->sk_protocol
];
2582 sk_for_each_bound(sk
, &tbl
->mc_list
)
2583 netlink_update_socket_mc(nlk_sk(sk
), group
, 0);
2587 __nlmsg_put(struct sk_buff
*skb
, u32 portid
, u32 seq
, int type
, int len
, int flags
)
2589 struct nlmsghdr
*nlh
;
2590 int size
= nlmsg_msg_size(len
);
2592 nlh
= (struct nlmsghdr
*)skb_put(skb
, NLMSG_ALIGN(size
));
2593 nlh
->nlmsg_type
= type
;
2594 nlh
->nlmsg_len
= size
;
2595 nlh
->nlmsg_flags
= flags
;
2596 nlh
->nlmsg_pid
= portid
;
2597 nlh
->nlmsg_seq
= seq
;
2598 if (!__builtin_constant_p(size
) || NLMSG_ALIGN(size
) - size
!= 0)
2599 memset(nlmsg_data(nlh
) + len
, 0, NLMSG_ALIGN(size
) - size
);
2602 EXPORT_SYMBOL(__nlmsg_put
);
2605 * It looks a bit ugly.
2606 * It would be better to create kernel thread.
2609 static int netlink_dump(struct sock
*sk
)
2611 struct netlink_sock
*nlk
= nlk_sk(sk
);
2612 struct netlink_callback
*cb
;
2613 struct sk_buff
*skb
= NULL
;
2614 struct nlmsghdr
*nlh
;
2615 int len
, err
= -ENOBUFS
;
2618 mutex_lock(nlk
->cb_mutex
);
2619 if (!nlk
->cb_running
) {
2625 alloc_size
= max_t(int, cb
->min_dump_alloc
, NLMSG_GOODSIZE
);
2627 if (!netlink_rx_is_mmaped(sk
) &&
2628 atomic_read(&sk
->sk_rmem_alloc
) >= sk
->sk_rcvbuf
)
2631 /* NLMSG_GOODSIZE is small to avoid high order allocations being
2632 * required, but it makes sense to _attempt_ a 16K bytes allocation
2633 * to reduce number of system calls on dump operations, if user
2634 * ever provided a big enough buffer.
2636 if (alloc_size
< nlk
->max_recvmsg_len
) {
2637 skb
= netlink_alloc_skb(sk
,
2638 nlk
->max_recvmsg_len
,
2643 /* available room should be exact amount to avoid MSG_TRUNC */
2645 skb_reserve(skb
, skb_tailroom(skb
) -
2646 nlk
->max_recvmsg_len
);
2649 skb
= netlink_alloc_skb(sk
, alloc_size
, nlk
->portid
,
2653 netlink_skb_set_owner_r(skb
, sk
);
2655 len
= cb
->dump(skb
, cb
);
2658 mutex_unlock(nlk
->cb_mutex
);
2660 if (sk_filter(sk
, skb
))
2663 __netlink_sendskb(sk
, skb
);
2667 nlh
= nlmsg_put_answer(skb
, cb
, NLMSG_DONE
, sizeof(len
), NLM_F_MULTI
);
2671 nl_dump_check_consistent(cb
, nlh
);
2673 memcpy(nlmsg_data(nlh
), &len
, sizeof(len
));
2675 if (sk_filter(sk
, skb
))
2678 __netlink_sendskb(sk
, skb
);
2683 nlk
->cb_running
= false;
2684 mutex_unlock(nlk
->cb_mutex
);
2685 module_put(cb
->module
);
2686 consume_skb(cb
->skb
);
2690 mutex_unlock(nlk
->cb_mutex
);
2695 int __netlink_dump_start(struct sock
*ssk
, struct sk_buff
*skb
,
2696 const struct nlmsghdr
*nlh
,
2697 struct netlink_dump_control
*control
)
2699 struct netlink_callback
*cb
;
2701 struct netlink_sock
*nlk
;
2704 /* Memory mapped dump requests need to be copied to avoid looping
2705 * on the pending state in netlink_mmap_sendmsg() while the CB hold
2706 * a reference to the skb.
2708 if (netlink_skb_is_mmaped(skb
)) {
2709 skb
= skb_copy(skb
, GFP_KERNEL
);
2713 atomic_inc(&skb
->users
);
2715 sk
= netlink_lookup(sock_net(ssk
), ssk
->sk_protocol
, NETLINK_CB(skb
).portid
);
2717 ret
= -ECONNREFUSED
;
2722 mutex_lock(nlk
->cb_mutex
);
2723 /* A dump is in progress... */
2724 if (nlk
->cb_running
) {
2728 /* add reference of module which cb->dump belongs to */
2729 if (!try_module_get(control
->module
)) {
2730 ret
= -EPROTONOSUPPORT
;
2735 memset(cb
, 0, sizeof(*cb
));
2736 cb
->dump
= control
->dump
;
2737 cb
->done
= control
->done
;
2739 cb
->data
= control
->data
;
2740 cb
->module
= control
->module
;
2741 cb
->min_dump_alloc
= control
->min_dump_alloc
;
2744 nlk
->cb_running
= true;
2746 mutex_unlock(nlk
->cb_mutex
);
2748 ret
= netlink_dump(sk
);
2754 /* We successfully started a dump, by returning -EINTR we
2755 * signal not to send ACK even if it was requested.
2761 mutex_unlock(nlk
->cb_mutex
);
2766 EXPORT_SYMBOL(__netlink_dump_start
);
2768 void netlink_ack(struct sk_buff
*in_skb
, struct nlmsghdr
*nlh
, int err
)
2770 struct sk_buff
*skb
;
2771 struct nlmsghdr
*rep
;
2772 struct nlmsgerr
*errmsg
;
2773 size_t payload
= sizeof(*errmsg
);
2775 /* error messages get the original request appened */
2777 payload
+= nlmsg_len(nlh
);
2779 skb
= netlink_alloc_skb(in_skb
->sk
, nlmsg_total_size(payload
),
2780 NETLINK_CB(in_skb
).portid
, GFP_KERNEL
);
2784 sk
= netlink_lookup(sock_net(in_skb
->sk
),
2785 in_skb
->sk
->sk_protocol
,
2786 NETLINK_CB(in_skb
).portid
);
2788 sk
->sk_err
= ENOBUFS
;
2789 sk
->sk_error_report(sk
);
2795 rep
= __nlmsg_put(skb
, NETLINK_CB(in_skb
).portid
, nlh
->nlmsg_seq
,
2796 NLMSG_ERROR
, payload
, 0);
2797 errmsg
= nlmsg_data(rep
);
2798 errmsg
->error
= err
;
2799 memcpy(&errmsg
->msg
, nlh
, err
? nlh
->nlmsg_len
: sizeof(*nlh
));
2800 netlink_unicast(in_skb
->sk
, skb
, NETLINK_CB(in_skb
).portid
, MSG_DONTWAIT
);
2802 EXPORT_SYMBOL(netlink_ack
);
2804 int netlink_rcv_skb(struct sk_buff
*skb
, int (*cb
)(struct sk_buff
*,
2807 struct nlmsghdr
*nlh
;
2810 while (skb
->len
>= nlmsg_total_size(0)) {
2813 nlh
= nlmsg_hdr(skb
);
2816 if (nlh
->nlmsg_len
< NLMSG_HDRLEN
|| skb
->len
< nlh
->nlmsg_len
)
2819 /* Only requests are handled by the kernel */
2820 if (!(nlh
->nlmsg_flags
& NLM_F_REQUEST
))
2823 /* Skip control messages */
2824 if (nlh
->nlmsg_type
< NLMSG_MIN_TYPE
)
2832 if (nlh
->nlmsg_flags
& NLM_F_ACK
|| err
)
2833 netlink_ack(skb
, nlh
, err
);
2836 msglen
= NLMSG_ALIGN(nlh
->nlmsg_len
);
2837 if (msglen
> skb
->len
)
2839 skb_pull(skb
, msglen
);
2844 EXPORT_SYMBOL(netlink_rcv_skb
);
2847 * nlmsg_notify - send a notification netlink message
2848 * @sk: netlink socket to use
2849 * @skb: notification message
2850 * @portid: destination netlink portid for reports or 0
2851 * @group: destination multicast group or 0
2852 * @report: 1 to report back, 0 to disable
2853 * @flags: allocation flags
2855 int nlmsg_notify(struct sock
*sk
, struct sk_buff
*skb
, u32 portid
,
2856 unsigned int group
, int report
, gfp_t flags
)
2861 int exclude_portid
= 0;
2864 atomic_inc(&skb
->users
);
2865 exclude_portid
= portid
;
2868 /* errors reported via destination sk->sk_err, but propagate
2869 * delivery errors if NETLINK_BROADCAST_ERROR flag is set */
2870 err
= nlmsg_multicast(sk
, skb
, exclude_portid
, group
, flags
);
2876 err2
= nlmsg_unicast(sk
, skb
, portid
);
2877 if (!err
|| err
== -ESRCH
)
2883 EXPORT_SYMBOL(nlmsg_notify
);
2885 #ifdef CONFIG_PROC_FS
2886 struct nl_seq_iter
{
2887 struct seq_net_private p
;
2892 static struct sock
*netlink_seq_socket_idx(struct seq_file
*seq
, loff_t pos
)
2894 struct nl_seq_iter
*iter
= seq
->private;
2896 struct netlink_sock
*nlk
;
2900 for (i
= 0; i
< MAX_LINKS
; i
++) {
2901 struct rhashtable
*ht
= &nl_table
[i
].hash
;
2902 const struct bucket_table
*tbl
= rht_dereference(ht
->tbl
, ht
);
2904 for (j
= 0; j
< tbl
->size
; j
++) {
2905 rht_for_each_entry_rcu(nlk
, tbl
->buckets
[j
], node
) {
2906 s
= (struct sock
*)nlk
;
2908 if (sock_net(s
) != seq_file_net(seq
))
2922 static void *netlink_seq_start(struct seq_file
*seq
, loff_t
*pos
)
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 netlink_sock
*nlk
;
2931 struct nl_seq_iter
*iter
;
2937 if (v
== SEQ_START_TOKEN
)
2938 return netlink_seq_socket_idx(seq
, 0);
2940 net
= seq_file_net(seq
);
2941 iter
= seq
->private;
2944 rht_for_each_entry_rcu(nlk
, nlk
->node
.next
, node
)
2945 if (net_eq(sock_net((struct sock
*)nlk
), net
))
2949 j
= iter
->hash_idx
+ 1;
2952 struct rhashtable
*ht
= &nl_table
[i
].hash
;
2953 const struct bucket_table
*tbl
= rht_dereference(ht
->tbl
, ht
);
2955 for (; j
< tbl
->size
; j
++) {
2956 rht_for_each_entry_rcu(nlk
, tbl
->buckets
[j
], node
) {
2957 if (net_eq(sock_net((struct sock
*)nlk
), net
)) {
2966 } while (++i
< MAX_LINKS
);
2971 static void netlink_seq_stop(struct seq_file
*seq
, void *v
)
2977 static int netlink_seq_show(struct seq_file
*seq
, void *v
)
2979 if (v
== SEQ_START_TOKEN
) {
2981 "sk Eth Pid Groups "
2982 "Rmem Wmem Dump Locks Drops Inode\n");
2985 struct netlink_sock
*nlk
= nlk_sk(s
);
2987 seq_printf(seq
, "%pK %-3d %-6u %08x %-8d %-8d %d %-8d %-8d %-8lu\n",
2991 nlk
->groups
? (u32
)nlk
->groups
[0] : 0,
2992 sk_rmem_alloc_get(s
),
2993 sk_wmem_alloc_get(s
),
2995 atomic_read(&s
->sk_refcnt
),
2996 atomic_read(&s
->sk_drops
),
3004 static const struct seq_operations netlink_seq_ops
= {
3005 .start
= netlink_seq_start
,
3006 .next
= netlink_seq_next
,
3007 .stop
= netlink_seq_stop
,
3008 .show
= netlink_seq_show
,
3012 static int netlink_seq_open(struct inode
*inode
, struct file
*file
)
3014 return seq_open_net(inode
, file
, &netlink_seq_ops
,
3015 sizeof(struct nl_seq_iter
));
3018 static const struct file_operations netlink_seq_fops
= {
3019 .owner
= THIS_MODULE
,
3020 .open
= netlink_seq_open
,
3022 .llseek
= seq_lseek
,
3023 .release
= seq_release_net
,
3028 int netlink_register_notifier(struct notifier_block
*nb
)
3030 return atomic_notifier_chain_register(&netlink_chain
, nb
);
3032 EXPORT_SYMBOL(netlink_register_notifier
);
3034 int netlink_unregister_notifier(struct notifier_block
*nb
)
3036 return atomic_notifier_chain_unregister(&netlink_chain
, nb
);
3038 EXPORT_SYMBOL(netlink_unregister_notifier
);
3040 static const struct proto_ops netlink_ops
= {
3041 .family
= PF_NETLINK
,
3042 .owner
= THIS_MODULE
,
3043 .release
= netlink_release
,
3044 .bind
= netlink_bind
,
3045 .connect
= netlink_connect
,
3046 .socketpair
= sock_no_socketpair
,
3047 .accept
= sock_no_accept
,
3048 .getname
= netlink_getname
,
3049 .poll
= netlink_poll
,
3050 .ioctl
= sock_no_ioctl
,
3051 .listen
= sock_no_listen
,
3052 .shutdown
= sock_no_shutdown
,
3053 .setsockopt
= netlink_setsockopt
,
3054 .getsockopt
= netlink_getsockopt
,
3055 .sendmsg
= netlink_sendmsg
,
3056 .recvmsg
= netlink_recvmsg
,
3057 .mmap
= netlink_mmap
,
3058 .sendpage
= sock_no_sendpage
,
3061 static const struct net_proto_family netlink_family_ops
= {
3062 .family
= PF_NETLINK
,
3063 .create
= netlink_create
,
3064 .owner
= THIS_MODULE
, /* for consistency 8) */
3067 static int __net_init
netlink_net_init(struct net
*net
)
3069 #ifdef CONFIG_PROC_FS
3070 if (!proc_create("netlink", 0, net
->proc_net
, &netlink_seq_fops
))
3076 static void __net_exit
netlink_net_exit(struct net
*net
)
3078 #ifdef CONFIG_PROC_FS
3079 remove_proc_entry("netlink", net
->proc_net
);
3083 static void __init
netlink_add_usersock_entry(void)
3085 struct listeners
*listeners
;
3088 listeners
= kzalloc(sizeof(*listeners
) + NLGRPSZ(groups
), GFP_KERNEL
);
3090 panic("netlink_add_usersock_entry: Cannot allocate listeners\n");
3092 netlink_table_grab();
3094 nl_table
[NETLINK_USERSOCK
].groups
= groups
;
3095 rcu_assign_pointer(nl_table
[NETLINK_USERSOCK
].listeners
, listeners
);
3096 nl_table
[NETLINK_USERSOCK
].module
= THIS_MODULE
;
3097 nl_table
[NETLINK_USERSOCK
].registered
= 1;
3098 nl_table
[NETLINK_USERSOCK
].flags
= NL_CFG_F_NONROOT_SEND
;
3100 netlink_table_ungrab();
3103 static struct pernet_operations __net_initdata netlink_net_ops
= {
3104 .init
= netlink_net_init
,
3105 .exit
= netlink_net_exit
,
3108 static int __init
netlink_proto_init(void)
3111 int err
= proto_register(&netlink_proto
, 0);
3112 struct rhashtable_params ht_params
= {
3113 .head_offset
= offsetof(struct netlink_sock
, node
),
3114 .key_offset
= offsetof(struct netlink_sock
, portid
),
3115 .key_len
= sizeof(u32
), /* portid */
3116 .hashfn
= arch_fast_hash
,
3117 .max_shift
= 16, /* 64K */
3118 .grow_decision
= rht_grow_above_75
,
3119 .shrink_decision
= rht_shrink_below_30
,
3120 .mutex_is_held
= lockdep_nl_sk_hash_is_held
,
3126 BUILD_BUG_ON(sizeof(struct netlink_skb_parms
) > FIELD_SIZEOF(struct sk_buff
, cb
));
3128 nl_table
= kcalloc(MAX_LINKS
, sizeof(*nl_table
), GFP_KERNEL
);
3132 for (i
= 0; i
< MAX_LINKS
; i
++) {
3133 if (rhashtable_init(&nl_table
[i
].hash
, &ht_params
) < 0) {
3135 rhashtable_destroy(&nl_table
[i
].hash
);
3141 INIT_LIST_HEAD(&netlink_tap_all
);
3143 netlink_add_usersock_entry();
3145 sock_register(&netlink_family_ops
);
3146 register_pernet_subsys(&netlink_net_ops
);
3147 /* The netlink device handler may be needed early. */
3152 panic("netlink_init: Cannot allocate nl_table\n");
3155 core_initcall(netlink_proto_init
);