2 * NETLINK Kernel-user communication protocol.
4 * Authors: Alan Cox <alan@lxorguk.ukuu.org.uk>
5 * Alexey Kuznetsov <kuznet@ms2.inr.ac.ru>
6 * Patrick McHardy <kaber@trash.net>
8 * This program is free software; you can redistribute it and/or
9 * modify it under the terms of the GNU General Public License
10 * as published by the Free Software Foundation; either version
11 * 2 of the License, or (at your option) any later version.
13 * Tue Jun 26 14:36:48 MEST 2001 Herbert "herp" Rosmanith
14 * added netlink_proto_exit
15 * Tue Jan 22 18:32:44 BRST 2002 Arnaldo C. de Melo <acme@conectiva.com.br>
16 * use nlk_sk, as sk->protinfo is on a diet 8)
17 * Fri Jul 22 19:51:12 MEST 2005 Harald Welte <laforge@gnumonks.org>
18 * - inc module use count of module that owns
19 * the kernel socket in case userspace opens
20 * socket of same protocol
21 * - remove all module support, since netlink is
22 * mandatory if CONFIG_NET=y these days
25 #include <linux/module.h>
27 #include <linux/capability.h>
28 #include <linux/kernel.h>
29 #include <linux/init.h>
30 #include <linux/signal.h>
31 #include <linux/sched.h>
32 #include <linux/errno.h>
33 #include <linux/string.h>
34 #include <linux/stat.h>
35 #include <linux/socket.h>
37 #include <linux/fcntl.h>
38 #include <linux/termios.h>
39 #include <linux/sockios.h>
40 #include <linux/net.h>
42 #include <linux/slab.h>
43 #include <asm/uaccess.h>
44 #include <linux/skbuff.h>
45 #include <linux/netdevice.h>
46 #include <linux/rtnetlink.h>
47 #include <linux/proc_fs.h>
48 #include <linux/seq_file.h>
49 #include <linux/notifier.h>
50 #include <linux/security.h>
51 #include <linux/jhash.h>
52 #include <linux/jiffies.h>
53 #include <linux/random.h>
54 #include <linux/bitops.h>
56 #include <linux/types.h>
57 #include <linux/audit.h>
58 #include <linux/mutex.h>
59 #include <linux/vmalloc.h>
60 #include <linux/if_arp.h>
61 #include <linux/rhashtable.h>
62 #include <asm/cacheflush.h>
63 #include <linux/hash.h>
64 #include <linux/genetlink.h>
66 #include <net/net_namespace.h>
69 #include <net/netlink.h>
71 #include "af_netlink.h"
75 unsigned long masks
[0];
79 #define NETLINK_S_CONGESTED 0x0
82 #define NETLINK_F_KERNEL_SOCKET 0x1
83 #define NETLINK_F_RECV_PKTINFO 0x2
84 #define NETLINK_F_BROADCAST_SEND_ERROR 0x4
85 #define NETLINK_F_RECV_NO_ENOBUFS 0x8
86 #define NETLINK_F_LISTEN_ALL_NSID 0x10
88 static inline int netlink_is_kernel(struct sock
*sk
)
90 return nlk_sk(sk
)->flags
& NETLINK_F_KERNEL_SOCKET
;
93 struct netlink_table
*nl_table
;
94 EXPORT_SYMBOL_GPL(nl_table
);
96 static DECLARE_WAIT_QUEUE_HEAD(nl_table_wait
);
98 static int netlink_dump(struct sock
*sk
);
99 static void netlink_skb_destructor(struct sk_buff
*skb
);
101 /* nl_table locking explained:
102 * Lookup and traversal are protected with an RCU read-side lock. Insertion
103 * and removal are protected with per bucket lock while using RCU list
104 * modification primitives and may run in parallel to RCU protected lookups.
105 * Destruction of the Netlink socket may only occur *after* nl_table_lock has
106 * been acquired * either during or after the socket has been removed from
107 * the list and after an RCU grace period.
109 DEFINE_RWLOCK(nl_table_lock
);
110 EXPORT_SYMBOL_GPL(nl_table_lock
);
111 static atomic_t nl_table_users
= ATOMIC_INIT(0);
113 #define nl_deref_protected(X) rcu_dereference_protected(X, lockdep_is_held(&nl_table_lock));
115 static ATOMIC_NOTIFIER_HEAD(netlink_chain
);
117 static DEFINE_SPINLOCK(netlink_tap_lock
);
118 static struct list_head netlink_tap_all __read_mostly
;
120 static const struct rhashtable_params netlink_rhashtable_params
;
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
);
136 __module_get(nt
->module
);
140 EXPORT_SYMBOL_GPL(netlink_add_tap
);
142 static int __netlink_remove_tap(struct netlink_tap
*nt
)
145 struct netlink_tap
*tmp
;
147 spin_lock(&netlink_tap_lock
);
149 list_for_each_entry(tmp
, &netlink_tap_all
, list
) {
151 list_del_rcu(&nt
->list
);
157 pr_warn("__netlink_remove_tap: %p not found\n", nt
);
159 spin_unlock(&netlink_tap_lock
);
161 if (found
&& nt
->module
)
162 module_put(nt
->module
);
164 return found
? 0 : -ENODEV
;
167 int netlink_remove_tap(struct netlink_tap
*nt
)
171 ret
= __netlink_remove_tap(nt
);
176 EXPORT_SYMBOL_GPL(netlink_remove_tap
);
178 static bool netlink_filter_tap(const struct sk_buff
*skb
)
180 struct sock
*sk
= skb
->sk
;
182 /* We take the more conservative approach and
183 * whitelist socket protocols that may pass.
185 switch (sk
->sk_protocol
) {
187 case NETLINK_USERSOCK
:
188 case NETLINK_SOCK_DIAG
:
191 case NETLINK_FIB_LOOKUP
:
192 case NETLINK_NETFILTER
:
193 case NETLINK_GENERIC
:
200 static int __netlink_deliver_tap_skb(struct sk_buff
*skb
,
201 struct net_device
*dev
)
203 struct sk_buff
*nskb
;
204 struct sock
*sk
= skb
->sk
;
208 nskb
= skb_clone(skb
, GFP_ATOMIC
);
211 nskb
->protocol
= htons((u16
) sk
->sk_protocol
);
212 nskb
->pkt_type
= netlink_is_kernel(sk
) ?
213 PACKET_KERNEL
: PACKET_USER
;
214 skb_reset_network_header(nskb
);
215 ret
= dev_queue_xmit(nskb
);
216 if (unlikely(ret
> 0))
217 ret
= net_xmit_errno(ret
);
224 static void __netlink_deliver_tap(struct sk_buff
*skb
)
227 struct netlink_tap
*tmp
;
229 if (!netlink_filter_tap(skb
))
232 list_for_each_entry_rcu(tmp
, &netlink_tap_all
, list
) {
233 ret
= __netlink_deliver_tap_skb(skb
, tmp
->dev
);
239 static void netlink_deliver_tap(struct sk_buff
*skb
)
243 if (unlikely(!list_empty(&netlink_tap_all
)))
244 __netlink_deliver_tap(skb
);
249 static void netlink_deliver_tap_kernel(struct sock
*dst
, struct sock
*src
,
252 if (!(netlink_is_kernel(dst
) && netlink_is_kernel(src
)))
253 netlink_deliver_tap(skb
);
256 static void netlink_overrun(struct sock
*sk
)
258 struct netlink_sock
*nlk
= nlk_sk(sk
);
260 if (!(nlk
->flags
& NETLINK_F_RECV_NO_ENOBUFS
)) {
261 if (!test_and_set_bit(NETLINK_S_CONGESTED
,
262 &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_S_CONGESTED
, &nlk
->state
);
276 if (!test_bit(NETLINK_S_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
, unsigned int nm_len
)
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
+ 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
)
546 hdr
->nm_status
= status
;
547 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 scm_cookie
*scm
)
704 struct netlink_sock
*nlk
= nlk_sk(sk
);
705 struct netlink_ring
*ring
;
706 struct nl_mmap_hdr
*hdr
;
709 int err
= 0, len
= 0;
711 mutex_lock(&nlk
->pg_vec_lock
);
713 ring
= &nlk
->tx_ring
;
714 maxlen
= ring
->frame_size
- NL_MMAP_HDRLEN
;
719 hdr
= netlink_current_frame(ring
, NL_MMAP_STATUS_VALID
);
721 if (!(msg
->msg_flags
& MSG_DONTWAIT
) &&
722 atomic_read(&nlk
->tx_ring
.pending
))
727 nm_len
= ACCESS_ONCE(hdr
->nm_len
);
728 if (nm_len
> maxlen
) {
733 netlink_frame_flush_dcache(hdr
, nm_len
);
735 skb
= alloc_skb(nm_len
, GFP_KERNEL
);
740 __skb_put(skb
, nm_len
);
741 memcpy(skb
->data
, (void *)hdr
+ NL_MMAP_HDRLEN
, nm_len
);
742 netlink_set_status(hdr
, NL_MMAP_STATUS_UNUSED
);
744 netlink_increment_head(ring
);
746 NETLINK_CB(skb
).portid
= nlk
->portid
;
747 NETLINK_CB(skb
).dst_group
= dst_group
;
748 NETLINK_CB(skb
).creds
= scm
->creds
;
750 err
= security_netlink_send(sk
, skb
);
756 if (unlikely(dst_group
)) {
757 atomic_inc(&skb
->users
);
758 netlink_broadcast(sk
, skb
, dst_portid
, dst_group
,
761 err
= netlink_unicast(sk
, skb
, dst_portid
,
762 msg
->msg_flags
& MSG_DONTWAIT
);
767 } while (hdr
!= NULL
||
768 (!(msg
->msg_flags
& MSG_DONTWAIT
) &&
769 atomic_read(&nlk
->tx_ring
.pending
)));
774 mutex_unlock(&nlk
->pg_vec_lock
);
778 static void netlink_queue_mmaped_skb(struct sock
*sk
, struct sk_buff
*skb
)
780 struct nl_mmap_hdr
*hdr
;
782 hdr
= netlink_mmap_hdr(skb
);
783 hdr
->nm_len
= skb
->len
;
784 hdr
->nm_group
= NETLINK_CB(skb
).dst_group
;
785 hdr
->nm_pid
= NETLINK_CB(skb
).creds
.pid
;
786 hdr
->nm_uid
= from_kuid(sk_user_ns(sk
), NETLINK_CB(skb
).creds
.uid
);
787 hdr
->nm_gid
= from_kgid(sk_user_ns(sk
), NETLINK_CB(skb
).creds
.gid
);
788 netlink_frame_flush_dcache(hdr
, hdr
->nm_len
);
789 netlink_set_status(hdr
, NL_MMAP_STATUS_VALID
);
791 NETLINK_CB(skb
).flags
|= NETLINK_SKB_DELIVERED
;
795 static void netlink_ring_set_copied(struct sock
*sk
, struct sk_buff
*skb
)
797 struct netlink_sock
*nlk
= nlk_sk(sk
);
798 struct netlink_ring
*ring
= &nlk
->rx_ring
;
799 struct nl_mmap_hdr
*hdr
;
801 spin_lock_bh(&sk
->sk_receive_queue
.lock
);
802 hdr
= netlink_current_frame(ring
, NL_MMAP_STATUS_UNUSED
);
804 spin_unlock_bh(&sk
->sk_receive_queue
.lock
);
809 netlink_increment_head(ring
);
810 __skb_queue_tail(&sk
->sk_receive_queue
, skb
);
811 spin_unlock_bh(&sk
->sk_receive_queue
.lock
);
813 hdr
->nm_len
= skb
->len
;
814 hdr
->nm_group
= NETLINK_CB(skb
).dst_group
;
815 hdr
->nm_pid
= NETLINK_CB(skb
).creds
.pid
;
816 hdr
->nm_uid
= from_kuid(sk_user_ns(sk
), NETLINK_CB(skb
).creds
.uid
);
817 hdr
->nm_gid
= from_kgid(sk_user_ns(sk
), NETLINK_CB(skb
).creds
.gid
);
818 netlink_set_status(hdr
, NL_MMAP_STATUS_COPY
);
821 #else /* CONFIG_NETLINK_MMAP */
822 #define netlink_skb_is_mmaped(skb) false
823 #define netlink_rx_is_mmaped(sk) false
824 #define netlink_tx_is_mmaped(sk) false
825 #define netlink_mmap sock_no_mmap
826 #define netlink_poll datagram_poll
827 #define netlink_mmap_sendmsg(sk, msg, dst_portid, dst_group, scm) 0
828 #endif /* CONFIG_NETLINK_MMAP */
830 static void netlink_skb_destructor(struct sk_buff
*skb
)
832 #ifdef CONFIG_NETLINK_MMAP
833 struct nl_mmap_hdr
*hdr
;
834 struct netlink_ring
*ring
;
837 /* If a packet from the kernel to userspace was freed because of an
838 * error without being delivered to userspace, the kernel must reset
839 * the status. In the direction userspace to kernel, the status is
840 * always reset here after the packet was processed and freed.
842 if (netlink_skb_is_mmaped(skb
)) {
843 hdr
= netlink_mmap_hdr(skb
);
844 sk
= NETLINK_CB(skb
).sk
;
846 if (NETLINK_CB(skb
).flags
& NETLINK_SKB_TX
) {
847 netlink_set_status(hdr
, NL_MMAP_STATUS_UNUSED
);
848 ring
= &nlk_sk(sk
)->tx_ring
;
850 if (!(NETLINK_CB(skb
).flags
& NETLINK_SKB_DELIVERED
)) {
852 netlink_set_status(hdr
, NL_MMAP_STATUS_VALID
);
854 ring
= &nlk_sk(sk
)->rx_ring
;
857 WARN_ON(atomic_read(&ring
->pending
) == 0);
858 atomic_dec(&ring
->pending
);
864 if (is_vmalloc_addr(skb
->head
)) {
866 !atomic_dec_return(&(skb_shinfo(skb
)->dataref
)))
875 static void netlink_skb_set_owner_r(struct sk_buff
*skb
, struct sock
*sk
)
877 WARN_ON(skb
->sk
!= NULL
);
879 skb
->destructor
= netlink_skb_destructor
;
880 atomic_add(skb
->truesize
, &sk
->sk_rmem_alloc
);
881 sk_mem_charge(sk
, skb
->truesize
);
884 static void netlink_sock_destruct(struct sock
*sk
)
886 struct netlink_sock
*nlk
= nlk_sk(sk
);
888 if (nlk
->cb_running
) {
890 nlk
->cb
.done(&nlk
->cb
);
892 module_put(nlk
->cb
.module
);
893 kfree_skb(nlk
->cb
.skb
);
896 skb_queue_purge(&sk
->sk_receive_queue
);
897 #ifdef CONFIG_NETLINK_MMAP
899 struct nl_mmap_req req
;
901 memset(&req
, 0, sizeof(req
));
902 if (nlk
->rx_ring
.pg_vec
)
903 netlink_set_ring(sk
, &req
, true, false);
904 memset(&req
, 0, sizeof(req
));
905 if (nlk
->tx_ring
.pg_vec
)
906 netlink_set_ring(sk
, &req
, true, true);
908 #endif /* CONFIG_NETLINK_MMAP */
910 if (!sock_flag(sk
, SOCK_DEAD
)) {
911 printk(KERN_ERR
"Freeing alive netlink socket %p\n", sk
);
915 WARN_ON(atomic_read(&sk
->sk_rmem_alloc
));
916 WARN_ON(atomic_read(&sk
->sk_wmem_alloc
));
917 WARN_ON(nlk_sk(sk
)->groups
);
920 /* This lock without WQ_FLAG_EXCLUSIVE is good on UP and it is _very_ bad on
921 * SMP. Look, when several writers sleep and reader wakes them up, all but one
922 * immediately hit write lock and grab all the cpus. Exclusive sleep solves
923 * this, _but_ remember, it adds useless work on UP machines.
926 void netlink_table_grab(void)
927 __acquires(nl_table_lock
)
931 write_lock_irq(&nl_table_lock
);
933 if (atomic_read(&nl_table_users
)) {
934 DECLARE_WAITQUEUE(wait
, current
);
936 add_wait_queue_exclusive(&nl_table_wait
, &wait
);
938 set_current_state(TASK_UNINTERRUPTIBLE
);
939 if (atomic_read(&nl_table_users
) == 0)
941 write_unlock_irq(&nl_table_lock
);
943 write_lock_irq(&nl_table_lock
);
946 __set_current_state(TASK_RUNNING
);
947 remove_wait_queue(&nl_table_wait
, &wait
);
951 void netlink_table_ungrab(void)
952 __releases(nl_table_lock
)
954 write_unlock_irq(&nl_table_lock
);
955 wake_up(&nl_table_wait
);
959 netlink_lock_table(void)
961 /* read_lock() synchronizes us to netlink_table_grab */
963 read_lock(&nl_table_lock
);
964 atomic_inc(&nl_table_users
);
965 read_unlock(&nl_table_lock
);
969 netlink_unlock_table(void)
971 if (atomic_dec_and_test(&nl_table_users
))
972 wake_up(&nl_table_wait
);
975 struct netlink_compare_arg
981 /* Doing sizeof directly may yield 4 extra bytes on 64-bit. */
982 #define netlink_compare_arg_len \
983 (offsetof(struct netlink_compare_arg, portid) + sizeof(u32))
985 static inline int netlink_compare(struct rhashtable_compare_arg
*arg
,
988 const struct netlink_compare_arg
*x
= arg
->key
;
989 const struct netlink_sock
*nlk
= ptr
;
991 return nlk
->portid
!= x
->portid
||
992 !net_eq(sock_net(&nlk
->sk
), read_pnet(&x
->pnet
));
995 static void netlink_compare_arg_init(struct netlink_compare_arg
*arg
,
996 struct net
*net
, u32 portid
)
998 memset(arg
, 0, sizeof(*arg
));
999 write_pnet(&arg
->pnet
, net
);
1000 arg
->portid
= portid
;
1003 static struct sock
*__netlink_lookup(struct netlink_table
*table
, u32 portid
,
1006 struct netlink_compare_arg arg
;
1008 netlink_compare_arg_init(&arg
, net
, portid
);
1009 return rhashtable_lookup_fast(&table
->hash
, &arg
,
1010 netlink_rhashtable_params
);
1013 static int __netlink_insert(struct netlink_table
*table
, struct sock
*sk
)
1015 struct netlink_compare_arg arg
;
1017 netlink_compare_arg_init(&arg
, sock_net(sk
), nlk_sk(sk
)->portid
);
1018 return rhashtable_lookup_insert_key(&table
->hash
, &arg
,
1020 netlink_rhashtable_params
);
1023 static struct sock
*netlink_lookup(struct net
*net
, int protocol
, u32 portid
)
1025 struct netlink_table
*table
= &nl_table
[protocol
];
1029 sk
= __netlink_lookup(table
, portid
, net
);
1037 static const struct proto_ops netlink_ops
;
1040 netlink_update_listeners(struct sock
*sk
)
1042 struct netlink_table
*tbl
= &nl_table
[sk
->sk_protocol
];
1045 struct listeners
*listeners
;
1047 listeners
= nl_deref_protected(tbl
->listeners
);
1051 for (i
= 0; i
< NLGRPLONGS(tbl
->groups
); i
++) {
1053 sk_for_each_bound(sk
, &tbl
->mc_list
) {
1054 if (i
< NLGRPLONGS(nlk_sk(sk
)->ngroups
))
1055 mask
|= nlk_sk(sk
)->groups
[i
];
1057 listeners
->masks
[i
] = mask
;
1059 /* this function is only called with the netlink table "grabbed", which
1060 * makes sure updates are visible before bind or setsockopt return. */
1063 static int netlink_insert(struct sock
*sk
, u32 portid
)
1065 struct netlink_table
*table
= &nl_table
[sk
->sk_protocol
];
1071 if (nlk_sk(sk
)->portid
)
1075 if (BITS_PER_LONG
> 32 &&
1076 unlikely(atomic_read(&table
->hash
.nelems
) >= UINT_MAX
))
1079 nlk_sk(sk
)->portid
= portid
;
1082 err
= __netlink_insert(table
, sk
);
1094 static void netlink_remove(struct sock
*sk
)
1096 struct netlink_table
*table
;
1098 table
= &nl_table
[sk
->sk_protocol
];
1099 if (!rhashtable_remove_fast(&table
->hash
, &nlk_sk(sk
)->node
,
1100 netlink_rhashtable_params
)) {
1101 WARN_ON(atomic_read(&sk
->sk_refcnt
) == 1);
1105 netlink_table_grab();
1106 if (nlk_sk(sk
)->subscriptions
) {
1107 __sk_del_bind_node(sk
);
1108 netlink_update_listeners(sk
);
1110 if (sk
->sk_protocol
== NETLINK_GENERIC
)
1111 atomic_inc(&genl_sk_destructing_cnt
);
1112 netlink_table_ungrab();
1115 static struct proto netlink_proto
= {
1117 .owner
= THIS_MODULE
,
1118 .obj_size
= sizeof(struct netlink_sock
),
1121 static int __netlink_create(struct net
*net
, struct socket
*sock
,
1122 struct mutex
*cb_mutex
, int protocol
)
1125 struct netlink_sock
*nlk
;
1127 sock
->ops
= &netlink_ops
;
1129 sk
= sk_alloc(net
, PF_NETLINK
, GFP_KERNEL
, &netlink_proto
);
1133 sock_init_data(sock
, sk
);
1137 nlk
->cb_mutex
= cb_mutex
;
1139 nlk
->cb_mutex
= &nlk
->cb_def_mutex
;
1140 mutex_init(nlk
->cb_mutex
);
1142 init_waitqueue_head(&nlk
->wait
);
1143 #ifdef CONFIG_NETLINK_MMAP
1144 mutex_init(&nlk
->pg_vec_lock
);
1147 sk
->sk_destruct
= netlink_sock_destruct
;
1148 sk
->sk_protocol
= protocol
;
1152 static int netlink_create(struct net
*net
, struct socket
*sock
, int protocol
,
1155 struct module
*module
= NULL
;
1156 struct mutex
*cb_mutex
;
1157 struct netlink_sock
*nlk
;
1158 int (*bind
)(struct net
*net
, int group
);
1159 void (*unbind
)(struct net
*net
, int group
);
1162 sock
->state
= SS_UNCONNECTED
;
1164 if (sock
->type
!= SOCK_RAW
&& sock
->type
!= SOCK_DGRAM
)
1165 return -ESOCKTNOSUPPORT
;
1167 if (protocol
< 0 || protocol
>= MAX_LINKS
)
1168 return -EPROTONOSUPPORT
;
1170 netlink_lock_table();
1171 #ifdef CONFIG_MODULES
1172 if (!nl_table
[protocol
].registered
) {
1173 netlink_unlock_table();
1174 request_module("net-pf-%d-proto-%d", PF_NETLINK
, protocol
);
1175 netlink_lock_table();
1178 if (nl_table
[protocol
].registered
&&
1179 try_module_get(nl_table
[protocol
].module
))
1180 module
= nl_table
[protocol
].module
;
1182 err
= -EPROTONOSUPPORT
;
1183 cb_mutex
= nl_table
[protocol
].cb_mutex
;
1184 bind
= nl_table
[protocol
].bind
;
1185 unbind
= nl_table
[protocol
].unbind
;
1186 netlink_unlock_table();
1191 err
= __netlink_create(net
, sock
, cb_mutex
, protocol
);
1196 sock_prot_inuse_add(net
, &netlink_proto
, 1);
1199 nlk
= nlk_sk(sock
->sk
);
1200 nlk
->module
= module
;
1201 nlk
->netlink_bind
= bind
;
1202 nlk
->netlink_unbind
= unbind
;
1211 static void deferred_put_nlk_sk(struct rcu_head
*head
)
1213 struct netlink_sock
*nlk
= container_of(head
, struct netlink_sock
, rcu
);
1218 static int netlink_release(struct socket
*sock
)
1220 struct sock
*sk
= sock
->sk
;
1221 struct netlink_sock
*nlk
;
1231 * OK. Socket is unlinked, any packets that arrive now
1235 /* must not acquire netlink_table_lock in any way again before unbind
1236 * and notifying genetlink is done as otherwise it might deadlock
1238 if (nlk
->netlink_unbind
) {
1241 for (i
= 0; i
< nlk
->ngroups
; i
++)
1242 if (test_bit(i
, nlk
->groups
))
1243 nlk
->netlink_unbind(sock_net(sk
), i
+ 1);
1245 if (sk
->sk_protocol
== NETLINK_GENERIC
&&
1246 atomic_dec_return(&genl_sk_destructing_cnt
) == 0)
1247 wake_up(&genl_sk_destructing_waitq
);
1250 wake_up_interruptible_all(&nlk
->wait
);
1252 skb_queue_purge(&sk
->sk_write_queue
);
1255 struct netlink_notify n
= {
1256 .net
= sock_net(sk
),
1257 .protocol
= sk
->sk_protocol
,
1258 .portid
= nlk
->portid
,
1260 atomic_notifier_call_chain(&netlink_chain
,
1261 NETLINK_URELEASE
, &n
);
1264 module_put(nlk
->module
);
1266 if (netlink_is_kernel(sk
)) {
1267 netlink_table_grab();
1268 BUG_ON(nl_table
[sk
->sk_protocol
].registered
== 0);
1269 if (--nl_table
[sk
->sk_protocol
].registered
== 0) {
1270 struct listeners
*old
;
1272 old
= nl_deref_protected(nl_table
[sk
->sk_protocol
].listeners
);
1273 RCU_INIT_POINTER(nl_table
[sk
->sk_protocol
].listeners
, NULL
);
1274 kfree_rcu(old
, rcu
);
1275 nl_table
[sk
->sk_protocol
].module
= NULL
;
1276 nl_table
[sk
->sk_protocol
].bind
= NULL
;
1277 nl_table
[sk
->sk_protocol
].unbind
= NULL
;
1278 nl_table
[sk
->sk_protocol
].flags
= 0;
1279 nl_table
[sk
->sk_protocol
].registered
= 0;
1281 netlink_table_ungrab();
1288 sock_prot_inuse_add(sock_net(sk
), &netlink_proto
, -1);
1290 call_rcu(&nlk
->rcu
, deferred_put_nlk_sk
);
1294 static int netlink_autobind(struct socket
*sock
)
1296 struct sock
*sk
= sock
->sk
;
1297 struct net
*net
= sock_net(sk
);
1298 struct netlink_table
*table
= &nl_table
[sk
->sk_protocol
];
1299 s32 portid
= task_tgid_vnr(current
);
1301 static s32 rover
= -4097;
1306 if (__netlink_lookup(table
, portid
, net
)) {
1307 /* Bind collision, search negative portid values. */
1316 err
= netlink_insert(sk
, portid
);
1317 if (err
== -EADDRINUSE
)
1320 /* If 2 threads race to autobind, that is fine. */
1328 * __netlink_ns_capable - General netlink message capability test
1329 * @nsp: NETLINK_CB of the socket buffer holding a netlink command from userspace.
1330 * @user_ns: The user namespace of the capability to use
1331 * @cap: The capability to use
1333 * Test to see if the opener of the socket we received the message
1334 * from had when the netlink socket was created and the sender of the
1335 * message has has the capability @cap in the user namespace @user_ns.
1337 bool __netlink_ns_capable(const struct netlink_skb_parms
*nsp
,
1338 struct user_namespace
*user_ns
, int cap
)
1340 return ((nsp
->flags
& NETLINK_SKB_DST
) ||
1341 file_ns_capable(nsp
->sk
->sk_socket
->file
, user_ns
, cap
)) &&
1342 ns_capable(user_ns
, cap
);
1344 EXPORT_SYMBOL(__netlink_ns_capable
);
1347 * netlink_ns_capable - General netlink message capability test
1348 * @skb: socket buffer holding a netlink command from userspace
1349 * @user_ns: The user namespace of the capability to use
1350 * @cap: The capability to use
1352 * Test to see if the opener of the socket we received the message
1353 * from had when the netlink socket was created and the sender of the
1354 * message has has the capability @cap in the user namespace @user_ns.
1356 bool netlink_ns_capable(const struct sk_buff
*skb
,
1357 struct user_namespace
*user_ns
, int cap
)
1359 return __netlink_ns_capable(&NETLINK_CB(skb
), user_ns
, cap
);
1361 EXPORT_SYMBOL(netlink_ns_capable
);
1364 * netlink_capable - Netlink global message capability test
1365 * @skb: socket buffer holding a netlink command from userspace
1366 * @cap: The capability to use
1368 * Test to see if the opener of the socket we received the message
1369 * from had when the netlink socket was created and the sender of the
1370 * message has has the capability @cap in all user namespaces.
1372 bool netlink_capable(const struct sk_buff
*skb
, int cap
)
1374 return netlink_ns_capable(skb
, &init_user_ns
, cap
);
1376 EXPORT_SYMBOL(netlink_capable
);
1379 * netlink_net_capable - Netlink network namespace message capability test
1380 * @skb: socket buffer holding a netlink command from userspace
1381 * @cap: The capability to use
1383 * Test to see if the opener of the socket we received the message
1384 * from had when the netlink socket was created and the sender of the
1385 * message has has the capability @cap over the network namespace of
1386 * the socket we received the message from.
1388 bool netlink_net_capable(const struct sk_buff
*skb
, int cap
)
1390 return netlink_ns_capable(skb
, sock_net(skb
->sk
)->user_ns
, cap
);
1392 EXPORT_SYMBOL(netlink_net_capable
);
1394 static inline int netlink_allowed(const struct socket
*sock
, unsigned int flag
)
1396 return (nl_table
[sock
->sk
->sk_protocol
].flags
& flag
) ||
1397 ns_capable(sock_net(sock
->sk
)->user_ns
, CAP_NET_ADMIN
);
1401 netlink_update_subscriptions(struct sock
*sk
, unsigned int subscriptions
)
1403 struct netlink_sock
*nlk
= nlk_sk(sk
);
1405 if (nlk
->subscriptions
&& !subscriptions
)
1406 __sk_del_bind_node(sk
);
1407 else if (!nlk
->subscriptions
&& subscriptions
)
1408 sk_add_bind_node(sk
, &nl_table
[sk
->sk_protocol
].mc_list
);
1409 nlk
->subscriptions
= subscriptions
;
1412 static int netlink_realloc_groups(struct sock
*sk
)
1414 struct netlink_sock
*nlk
= nlk_sk(sk
);
1415 unsigned int groups
;
1416 unsigned long *new_groups
;
1419 netlink_table_grab();
1421 groups
= nl_table
[sk
->sk_protocol
].groups
;
1422 if (!nl_table
[sk
->sk_protocol
].registered
) {
1427 if (nlk
->ngroups
>= groups
)
1430 new_groups
= krealloc(nlk
->groups
, NLGRPSZ(groups
), GFP_ATOMIC
);
1431 if (new_groups
== NULL
) {
1435 memset((char *)new_groups
+ NLGRPSZ(nlk
->ngroups
), 0,
1436 NLGRPSZ(groups
) - NLGRPSZ(nlk
->ngroups
));
1438 nlk
->groups
= new_groups
;
1439 nlk
->ngroups
= groups
;
1441 netlink_table_ungrab();
1445 static void netlink_undo_bind(int group
, long unsigned int groups
,
1448 struct netlink_sock
*nlk
= nlk_sk(sk
);
1451 if (!nlk
->netlink_unbind
)
1454 for (undo
= 0; undo
< group
; undo
++)
1455 if (test_bit(undo
, &groups
))
1456 nlk
->netlink_unbind(sock_net(sk
), undo
+ 1);
1459 static int netlink_bind(struct socket
*sock
, struct sockaddr
*addr
,
1462 struct sock
*sk
= sock
->sk
;
1463 struct net
*net
= sock_net(sk
);
1464 struct netlink_sock
*nlk
= nlk_sk(sk
);
1465 struct sockaddr_nl
*nladdr
= (struct sockaddr_nl
*)addr
;
1467 long unsigned int groups
= nladdr
->nl_groups
;
1469 if (addr_len
< sizeof(struct sockaddr_nl
))
1472 if (nladdr
->nl_family
!= AF_NETLINK
)
1475 /* Only superuser is allowed to listen multicasts */
1477 if (!netlink_allowed(sock
, NL_CFG_F_NONROOT_RECV
))
1479 err
= netlink_realloc_groups(sk
);
1485 if (nladdr
->nl_pid
!= nlk
->portid
)
1488 if (nlk
->netlink_bind
&& groups
) {
1491 for (group
= 0; group
< nlk
->ngroups
; group
++) {
1492 if (!test_bit(group
, &groups
))
1494 err
= nlk
->netlink_bind(net
, group
+ 1);
1497 netlink_undo_bind(group
, groups
, sk
);
1503 err
= nladdr
->nl_pid
?
1504 netlink_insert(sk
, nladdr
->nl_pid
) :
1505 netlink_autobind(sock
);
1507 netlink_undo_bind(nlk
->ngroups
, groups
, sk
);
1512 if (!groups
&& (nlk
->groups
== NULL
|| !(u32
)nlk
->groups
[0]))
1515 netlink_table_grab();
1516 netlink_update_subscriptions(sk
, nlk
->subscriptions
+
1518 hweight32(nlk
->groups
[0]));
1519 nlk
->groups
[0] = (nlk
->groups
[0] & ~0xffffffffUL
) | groups
;
1520 netlink_update_listeners(sk
);
1521 netlink_table_ungrab();
1526 static int netlink_connect(struct socket
*sock
, struct sockaddr
*addr
,
1527 int alen
, int flags
)
1530 struct sock
*sk
= sock
->sk
;
1531 struct netlink_sock
*nlk
= nlk_sk(sk
);
1532 struct sockaddr_nl
*nladdr
= (struct sockaddr_nl
*)addr
;
1534 if (alen
< sizeof(addr
->sa_family
))
1537 if (addr
->sa_family
== AF_UNSPEC
) {
1538 sk
->sk_state
= NETLINK_UNCONNECTED
;
1539 nlk
->dst_portid
= 0;
1543 if (addr
->sa_family
!= AF_NETLINK
)
1546 if ((nladdr
->nl_groups
|| nladdr
->nl_pid
) &&
1547 !netlink_allowed(sock
, NL_CFG_F_NONROOT_SEND
))
1551 err
= netlink_autobind(sock
);
1554 sk
->sk_state
= NETLINK_CONNECTED
;
1555 nlk
->dst_portid
= nladdr
->nl_pid
;
1556 nlk
->dst_group
= ffs(nladdr
->nl_groups
);
1562 static int netlink_getname(struct socket
*sock
, struct sockaddr
*addr
,
1563 int *addr_len
, int peer
)
1565 struct sock
*sk
= sock
->sk
;
1566 struct netlink_sock
*nlk
= nlk_sk(sk
);
1567 DECLARE_SOCKADDR(struct sockaddr_nl
*, nladdr
, addr
);
1569 nladdr
->nl_family
= AF_NETLINK
;
1571 *addr_len
= sizeof(*nladdr
);
1574 nladdr
->nl_pid
= nlk
->dst_portid
;
1575 nladdr
->nl_groups
= netlink_group_mask(nlk
->dst_group
);
1577 nladdr
->nl_pid
= nlk
->portid
;
1578 nladdr
->nl_groups
= nlk
->groups
? nlk
->groups
[0] : 0;
1583 static struct sock
*netlink_getsockbyportid(struct sock
*ssk
, u32 portid
)
1586 struct netlink_sock
*nlk
;
1588 sock
= netlink_lookup(sock_net(ssk
), ssk
->sk_protocol
, portid
);
1590 return ERR_PTR(-ECONNREFUSED
);
1592 /* Don't bother queuing skb if kernel socket has no input function */
1594 if (sock
->sk_state
== NETLINK_CONNECTED
&&
1595 nlk
->dst_portid
!= nlk_sk(ssk
)->portid
) {
1597 return ERR_PTR(-ECONNREFUSED
);
1602 struct sock
*netlink_getsockbyfilp(struct file
*filp
)
1604 struct inode
*inode
= file_inode(filp
);
1607 if (!S_ISSOCK(inode
->i_mode
))
1608 return ERR_PTR(-ENOTSOCK
);
1610 sock
= SOCKET_I(inode
)->sk
;
1611 if (sock
->sk_family
!= AF_NETLINK
)
1612 return ERR_PTR(-EINVAL
);
1618 static struct sk_buff
*netlink_alloc_large_skb(unsigned int size
,
1621 struct sk_buff
*skb
;
1624 if (size
<= NLMSG_GOODSIZE
|| broadcast
)
1625 return alloc_skb(size
, GFP_KERNEL
);
1627 size
= SKB_DATA_ALIGN(size
) +
1628 SKB_DATA_ALIGN(sizeof(struct skb_shared_info
));
1630 data
= vmalloc(size
);
1634 skb
= __build_skb(data
, size
);
1638 skb
->destructor
= netlink_skb_destructor
;
1644 * Attach a skb to a netlink socket.
1645 * The caller must hold a reference to the destination socket. On error, the
1646 * reference is dropped. The skb is not send to the destination, just all
1647 * all error checks are performed and memory in the queue is reserved.
1649 * < 0: error. skb freed, reference to sock dropped.
1651 * 1: repeat lookup - reference dropped while waiting for socket memory.
1653 int netlink_attachskb(struct sock
*sk
, struct sk_buff
*skb
,
1654 long *timeo
, struct sock
*ssk
)
1656 struct netlink_sock
*nlk
;
1660 if ((atomic_read(&sk
->sk_rmem_alloc
) > sk
->sk_rcvbuf
||
1661 test_bit(NETLINK_S_CONGESTED
, &nlk
->state
)) &&
1662 !netlink_skb_is_mmaped(skb
)) {
1663 DECLARE_WAITQUEUE(wait
, current
);
1665 if (!ssk
|| netlink_is_kernel(ssk
))
1666 netlink_overrun(sk
);
1672 __set_current_state(TASK_INTERRUPTIBLE
);
1673 add_wait_queue(&nlk
->wait
, &wait
);
1675 if ((atomic_read(&sk
->sk_rmem_alloc
) > sk
->sk_rcvbuf
||
1676 test_bit(NETLINK_S_CONGESTED
, &nlk
->state
)) &&
1677 !sock_flag(sk
, SOCK_DEAD
))
1678 *timeo
= schedule_timeout(*timeo
);
1680 __set_current_state(TASK_RUNNING
);
1681 remove_wait_queue(&nlk
->wait
, &wait
);
1684 if (signal_pending(current
)) {
1686 return sock_intr_errno(*timeo
);
1690 netlink_skb_set_owner_r(skb
, sk
);
1694 static int __netlink_sendskb(struct sock
*sk
, struct sk_buff
*skb
)
1698 netlink_deliver_tap(skb
);
1700 #ifdef CONFIG_NETLINK_MMAP
1701 if (netlink_skb_is_mmaped(skb
))
1702 netlink_queue_mmaped_skb(sk
, skb
);
1703 else if (netlink_rx_is_mmaped(sk
))
1704 netlink_ring_set_copied(sk
, skb
);
1706 #endif /* CONFIG_NETLINK_MMAP */
1707 skb_queue_tail(&sk
->sk_receive_queue
, skb
);
1708 sk
->sk_data_ready(sk
);
1712 int netlink_sendskb(struct sock
*sk
, struct sk_buff
*skb
)
1714 int len
= __netlink_sendskb(sk
, skb
);
1720 void netlink_detachskb(struct sock
*sk
, struct sk_buff
*skb
)
1726 static struct sk_buff
*netlink_trim(struct sk_buff
*skb
, gfp_t allocation
)
1730 WARN_ON(skb
->sk
!= NULL
);
1731 if (netlink_skb_is_mmaped(skb
))
1734 delta
= skb
->end
- skb
->tail
;
1735 if (is_vmalloc_addr(skb
->head
) || delta
* 2 < skb
->truesize
)
1738 if (skb_shared(skb
)) {
1739 struct sk_buff
*nskb
= skb_clone(skb
, allocation
);
1746 if (!pskb_expand_head(skb
, 0, -delta
, allocation
))
1747 skb
->truesize
-= delta
;
1752 static int netlink_unicast_kernel(struct sock
*sk
, struct sk_buff
*skb
,
1756 struct netlink_sock
*nlk
= nlk_sk(sk
);
1758 ret
= -ECONNREFUSED
;
1759 if (nlk
->netlink_rcv
!= NULL
) {
1761 netlink_skb_set_owner_r(skb
, sk
);
1762 NETLINK_CB(skb
).sk
= ssk
;
1763 netlink_deliver_tap_kernel(sk
, ssk
, skb
);
1764 nlk
->netlink_rcv(skb
);
1773 int netlink_unicast(struct sock
*ssk
, struct sk_buff
*skb
,
1774 u32 portid
, int nonblock
)
1780 skb
= netlink_trim(skb
, gfp_any());
1782 timeo
= sock_sndtimeo(ssk
, nonblock
);
1784 sk
= netlink_getsockbyportid(ssk
, portid
);
1789 if (netlink_is_kernel(sk
))
1790 return netlink_unicast_kernel(sk
, skb
, ssk
);
1792 if (sk_filter(sk
, skb
)) {
1799 err
= netlink_attachskb(sk
, skb
, &timeo
, ssk
);
1805 return netlink_sendskb(sk
, skb
);
1807 EXPORT_SYMBOL(netlink_unicast
);
1809 struct sk_buff
*netlink_alloc_skb(struct sock
*ssk
, unsigned int size
,
1810 u32 dst_portid
, gfp_t gfp_mask
)
1812 #ifdef CONFIG_NETLINK_MMAP
1813 struct sock
*sk
= NULL
;
1814 struct sk_buff
*skb
;
1815 struct netlink_ring
*ring
;
1816 struct nl_mmap_hdr
*hdr
;
1817 unsigned int maxlen
;
1819 sk
= netlink_getsockbyportid(ssk
, dst_portid
);
1823 ring
= &nlk_sk(sk
)->rx_ring
;
1824 /* fast-path without atomic ops for common case: non-mmaped receiver */
1825 if (ring
->pg_vec
== NULL
)
1828 if (ring
->frame_size
- NL_MMAP_HDRLEN
< size
)
1831 skb
= alloc_skb_head(gfp_mask
);
1835 spin_lock_bh(&sk
->sk_receive_queue
.lock
);
1836 /* check again under lock */
1837 if (ring
->pg_vec
== NULL
)
1840 /* check again under lock */
1841 maxlen
= ring
->frame_size
- NL_MMAP_HDRLEN
;
1845 netlink_forward_ring(ring
);
1846 hdr
= netlink_current_frame(ring
, NL_MMAP_STATUS_UNUSED
);
1849 netlink_ring_setup_skb(skb
, sk
, ring
, hdr
);
1850 netlink_set_status(hdr
, NL_MMAP_STATUS_RESERVED
);
1851 atomic_inc(&ring
->pending
);
1852 netlink_increment_head(ring
);
1854 spin_unlock_bh(&sk
->sk_receive_queue
.lock
);
1859 spin_unlock_bh(&sk
->sk_receive_queue
.lock
);
1860 netlink_overrun(sk
);
1867 spin_unlock_bh(&sk
->sk_receive_queue
.lock
);
1872 return alloc_skb(size
, gfp_mask
);
1874 EXPORT_SYMBOL_GPL(netlink_alloc_skb
);
1876 int netlink_has_listeners(struct sock
*sk
, unsigned int group
)
1879 struct listeners
*listeners
;
1881 BUG_ON(!netlink_is_kernel(sk
));
1884 listeners
= rcu_dereference(nl_table
[sk
->sk_protocol
].listeners
);
1886 if (listeners
&& group
- 1 < nl_table
[sk
->sk_protocol
].groups
)
1887 res
= test_bit(group
- 1, listeners
->masks
);
1893 EXPORT_SYMBOL_GPL(netlink_has_listeners
);
1895 static int netlink_broadcast_deliver(struct sock
*sk
, struct sk_buff
*skb
)
1897 struct netlink_sock
*nlk
= nlk_sk(sk
);
1899 if (atomic_read(&sk
->sk_rmem_alloc
) <= sk
->sk_rcvbuf
&&
1900 !test_bit(NETLINK_S_CONGESTED
, &nlk
->state
)) {
1901 netlink_skb_set_owner_r(skb
, sk
);
1902 __netlink_sendskb(sk
, skb
);
1903 return atomic_read(&sk
->sk_rmem_alloc
) > (sk
->sk_rcvbuf
>> 1);
1908 struct netlink_broadcast_data
{
1909 struct sock
*exclude_sk
;
1914 int delivery_failure
;
1918 struct sk_buff
*skb
, *skb2
;
1919 int (*tx_filter
)(struct sock
*dsk
, struct sk_buff
*skb
, void *data
);
1923 static void do_one_broadcast(struct sock
*sk
,
1924 struct netlink_broadcast_data
*p
)
1926 struct netlink_sock
*nlk
= nlk_sk(sk
);
1929 if (p
->exclude_sk
== sk
)
1932 if (nlk
->portid
== p
->portid
|| p
->group
- 1 >= nlk
->ngroups
||
1933 !test_bit(p
->group
- 1, nlk
->groups
))
1936 if (!net_eq(sock_net(sk
), p
->net
)) {
1937 if (!(nlk
->flags
& NETLINK_F_LISTEN_ALL_NSID
))
1940 if (!peernet_has_id(sock_net(sk
), p
->net
))
1943 if (!file_ns_capable(sk
->sk_socket
->file
, p
->net
->user_ns
,
1949 netlink_overrun(sk
);
1954 if (p
->skb2
== NULL
) {
1955 if (skb_shared(p
->skb
)) {
1956 p
->skb2
= skb_clone(p
->skb
, p
->allocation
);
1958 p
->skb2
= skb_get(p
->skb
);
1960 * skb ownership may have been set when
1961 * delivered to a previous socket.
1963 skb_orphan(p
->skb2
);
1966 if (p
->skb2
== NULL
) {
1967 netlink_overrun(sk
);
1968 /* Clone failed. Notify ALL listeners. */
1970 if (nlk
->flags
& NETLINK_F_BROADCAST_SEND_ERROR
)
1971 p
->delivery_failure
= 1;
1974 if (p
->tx_filter
&& p
->tx_filter(sk
, p
->skb2
, p
->tx_data
)) {
1979 if (sk_filter(sk
, p
->skb2
)) {
1984 NETLINK_CB(p
->skb2
).nsid
= peernet2id(sock_net(sk
), p
->net
);
1985 NETLINK_CB(p
->skb2
).nsid_is_set
= true;
1986 val
= netlink_broadcast_deliver(sk
, p
->skb2
);
1988 netlink_overrun(sk
);
1989 if (nlk
->flags
& NETLINK_F_BROADCAST_SEND_ERROR
)
1990 p
->delivery_failure
= 1;
1992 p
->congested
|= val
;
2000 int netlink_broadcast_filtered(struct sock
*ssk
, struct sk_buff
*skb
, u32 portid
,
2001 u32 group
, gfp_t allocation
,
2002 int (*filter
)(struct sock
*dsk
, struct sk_buff
*skb
, void *data
),
2005 struct net
*net
= sock_net(ssk
);
2006 struct netlink_broadcast_data info
;
2009 skb
= netlink_trim(skb
, allocation
);
2011 info
.exclude_sk
= ssk
;
2013 info
.portid
= portid
;
2016 info
.delivery_failure
= 0;
2019 info
.allocation
= allocation
;
2022 info
.tx_filter
= filter
;
2023 info
.tx_data
= filter_data
;
2025 /* While we sleep in clone, do not allow to change socket list */
2027 netlink_lock_table();
2029 sk_for_each_bound(sk
, &nl_table
[ssk
->sk_protocol
].mc_list
)
2030 do_one_broadcast(sk
, &info
);
2034 netlink_unlock_table();
2036 if (info
.delivery_failure
) {
2037 kfree_skb(info
.skb2
);
2040 consume_skb(info
.skb2
);
2042 if (info
.delivered
) {
2043 if (info
.congested
&& (allocation
& __GFP_WAIT
))
2049 EXPORT_SYMBOL(netlink_broadcast_filtered
);
2051 int netlink_broadcast(struct sock
*ssk
, struct sk_buff
*skb
, u32 portid
,
2052 u32 group
, gfp_t allocation
)
2054 return netlink_broadcast_filtered(ssk
, skb
, portid
, group
, allocation
,
2057 EXPORT_SYMBOL(netlink_broadcast
);
2059 struct netlink_set_err_data
{
2060 struct sock
*exclude_sk
;
2066 static int do_one_set_err(struct sock
*sk
, struct netlink_set_err_data
*p
)
2068 struct netlink_sock
*nlk
= nlk_sk(sk
);
2071 if (sk
== p
->exclude_sk
)
2074 if (!net_eq(sock_net(sk
), sock_net(p
->exclude_sk
)))
2077 if (nlk
->portid
== p
->portid
|| p
->group
- 1 >= nlk
->ngroups
||
2078 !test_bit(p
->group
- 1, nlk
->groups
))
2081 if (p
->code
== ENOBUFS
&& nlk
->flags
& NETLINK_F_RECV_NO_ENOBUFS
) {
2086 sk
->sk_err
= p
->code
;
2087 sk
->sk_error_report(sk
);
2093 * netlink_set_err - report error to broadcast listeners
2094 * @ssk: the kernel netlink socket, as returned by netlink_kernel_create()
2095 * @portid: the PORTID of a process that we want to skip (if any)
2096 * @group: the broadcast group that will notice the error
2097 * @code: error code, must be negative (as usual in kernelspace)
2099 * This function returns the number of broadcast listeners that have set the
2100 * NETLINK_NO_ENOBUFS socket option.
2102 int netlink_set_err(struct sock
*ssk
, u32 portid
, u32 group
, int code
)
2104 struct netlink_set_err_data info
;
2108 info
.exclude_sk
= ssk
;
2109 info
.portid
= portid
;
2111 /* sk->sk_err wants a positive error value */
2114 read_lock(&nl_table_lock
);
2116 sk_for_each_bound(sk
, &nl_table
[ssk
->sk_protocol
].mc_list
)
2117 ret
+= do_one_set_err(sk
, &info
);
2119 read_unlock(&nl_table_lock
);
2122 EXPORT_SYMBOL(netlink_set_err
);
2124 /* must be called with netlink table grabbed */
2125 static void netlink_update_socket_mc(struct netlink_sock
*nlk
,
2129 int old
, new = !!is_new
, subscriptions
;
2131 old
= test_bit(group
- 1, nlk
->groups
);
2132 subscriptions
= nlk
->subscriptions
- old
+ new;
2134 __set_bit(group
- 1, nlk
->groups
);
2136 __clear_bit(group
- 1, nlk
->groups
);
2137 netlink_update_subscriptions(&nlk
->sk
, subscriptions
);
2138 netlink_update_listeners(&nlk
->sk
);
2141 static int netlink_setsockopt(struct socket
*sock
, int level
, int optname
,
2142 char __user
*optval
, unsigned int optlen
)
2144 struct sock
*sk
= sock
->sk
;
2145 struct netlink_sock
*nlk
= nlk_sk(sk
);
2146 unsigned int val
= 0;
2149 if (level
!= SOL_NETLINK
)
2150 return -ENOPROTOOPT
;
2152 if (optname
!= NETLINK_RX_RING
&& optname
!= NETLINK_TX_RING
&&
2153 optlen
>= sizeof(int) &&
2154 get_user(val
, (unsigned int __user
*)optval
))
2158 case NETLINK_PKTINFO
:
2160 nlk
->flags
|= NETLINK_F_RECV_PKTINFO
;
2162 nlk
->flags
&= ~NETLINK_F_RECV_PKTINFO
;
2165 case NETLINK_ADD_MEMBERSHIP
:
2166 case NETLINK_DROP_MEMBERSHIP
: {
2167 if (!netlink_allowed(sock
, NL_CFG_F_NONROOT_RECV
))
2169 err
= netlink_realloc_groups(sk
);
2172 if (!val
|| val
- 1 >= nlk
->ngroups
)
2174 if (optname
== NETLINK_ADD_MEMBERSHIP
&& nlk
->netlink_bind
) {
2175 err
= nlk
->netlink_bind(sock_net(sk
), val
);
2179 netlink_table_grab();
2180 netlink_update_socket_mc(nlk
, val
,
2181 optname
== NETLINK_ADD_MEMBERSHIP
);
2182 netlink_table_ungrab();
2183 if (optname
== NETLINK_DROP_MEMBERSHIP
&& nlk
->netlink_unbind
)
2184 nlk
->netlink_unbind(sock_net(sk
), val
);
2189 case NETLINK_BROADCAST_ERROR
:
2191 nlk
->flags
|= NETLINK_F_BROADCAST_SEND_ERROR
;
2193 nlk
->flags
&= ~NETLINK_F_BROADCAST_SEND_ERROR
;
2196 case NETLINK_NO_ENOBUFS
:
2198 nlk
->flags
|= NETLINK_F_RECV_NO_ENOBUFS
;
2199 clear_bit(NETLINK_S_CONGESTED
, &nlk
->state
);
2200 wake_up_interruptible(&nlk
->wait
);
2202 nlk
->flags
&= ~NETLINK_F_RECV_NO_ENOBUFS
;
2206 #ifdef CONFIG_NETLINK_MMAP
2207 case NETLINK_RX_RING
:
2208 case NETLINK_TX_RING
: {
2209 struct nl_mmap_req req
;
2211 /* Rings might consume more memory than queue limits, require
2214 if (!capable(CAP_NET_ADMIN
))
2216 if (optlen
< sizeof(req
))
2218 if (copy_from_user(&req
, optval
, sizeof(req
)))
2220 err
= netlink_set_ring(sk
, &req
, false,
2221 optname
== NETLINK_TX_RING
);
2224 #endif /* CONFIG_NETLINK_MMAP */
2225 case NETLINK_LISTEN_ALL_NSID
:
2226 if (!ns_capable(sock_net(sk
)->user_ns
, CAP_NET_BROADCAST
))
2230 nlk
->flags
|= NETLINK_F_LISTEN_ALL_NSID
;
2232 nlk
->flags
&= ~NETLINK_F_LISTEN_ALL_NSID
;
2241 static int netlink_getsockopt(struct socket
*sock
, int level
, int optname
,
2242 char __user
*optval
, int __user
*optlen
)
2244 struct sock
*sk
= sock
->sk
;
2245 struct netlink_sock
*nlk
= nlk_sk(sk
);
2248 if (level
!= SOL_NETLINK
)
2249 return -ENOPROTOOPT
;
2251 if (get_user(len
, optlen
))
2257 case NETLINK_PKTINFO
:
2258 if (len
< sizeof(int))
2261 val
= nlk
->flags
& NETLINK_F_RECV_PKTINFO
? 1 : 0;
2262 if (put_user(len
, optlen
) ||
2263 put_user(val
, optval
))
2267 case NETLINK_BROADCAST_ERROR
:
2268 if (len
< sizeof(int))
2271 val
= nlk
->flags
& NETLINK_F_BROADCAST_SEND_ERROR
? 1 : 0;
2272 if (put_user(len
, optlen
) ||
2273 put_user(val
, optval
))
2277 case NETLINK_NO_ENOBUFS
:
2278 if (len
< sizeof(int))
2281 val
= nlk
->flags
& NETLINK_F_RECV_NO_ENOBUFS
? 1 : 0;
2282 if (put_user(len
, optlen
) ||
2283 put_user(val
, optval
))
2293 static void netlink_cmsg_recv_pktinfo(struct msghdr
*msg
, struct sk_buff
*skb
)
2295 struct nl_pktinfo info
;
2297 info
.group
= NETLINK_CB(skb
).dst_group
;
2298 put_cmsg(msg
, SOL_NETLINK
, NETLINK_PKTINFO
, sizeof(info
), &info
);
2301 static void netlink_cmsg_listen_all_nsid(struct sock
*sk
, struct msghdr
*msg
,
2302 struct sk_buff
*skb
)
2304 if (!NETLINK_CB(skb
).nsid_is_set
)
2307 put_cmsg(msg
, SOL_NETLINK
, NETLINK_LISTEN_ALL_NSID
, sizeof(int),
2308 &NETLINK_CB(skb
).nsid
);
2311 static int netlink_sendmsg(struct socket
*sock
, struct msghdr
*msg
, size_t len
)
2313 struct sock
*sk
= sock
->sk
;
2314 struct netlink_sock
*nlk
= nlk_sk(sk
);
2315 DECLARE_SOCKADDR(struct sockaddr_nl
*, addr
, msg
->msg_name
);
2318 struct sk_buff
*skb
;
2320 struct scm_cookie scm
;
2321 u32 netlink_skb_flags
= 0;
2323 if (msg
->msg_flags
&MSG_OOB
)
2326 err
= scm_send(sock
, msg
, &scm
, true);
2330 if (msg
->msg_namelen
) {
2332 if (addr
->nl_family
!= AF_NETLINK
)
2334 dst_portid
= addr
->nl_pid
;
2335 dst_group
= ffs(addr
->nl_groups
);
2337 if ((dst_group
|| dst_portid
) &&
2338 !netlink_allowed(sock
, NL_CFG_F_NONROOT_SEND
))
2340 netlink_skb_flags
|= NETLINK_SKB_DST
;
2342 dst_portid
= nlk
->dst_portid
;
2343 dst_group
= nlk
->dst_group
;
2347 err
= netlink_autobind(sock
);
2352 /* It's a really convoluted way for userland to ask for mmaped
2353 * sendmsg(), but that's what we've got...
2355 if (netlink_tx_is_mmaped(sk
) &&
2356 msg
->msg_iter
.type
== ITER_IOVEC
&&
2357 msg
->msg_iter
.nr_segs
== 1 &&
2358 msg
->msg_iter
.iov
->iov_base
== NULL
) {
2359 err
= netlink_mmap_sendmsg(sk
, msg
, dst_portid
, dst_group
,
2365 if (len
> sk
->sk_sndbuf
- 32)
2368 skb
= netlink_alloc_large_skb(len
, dst_group
);
2372 NETLINK_CB(skb
).portid
= nlk
->portid
;
2373 NETLINK_CB(skb
).dst_group
= dst_group
;
2374 NETLINK_CB(skb
).creds
= scm
.creds
;
2375 NETLINK_CB(skb
).flags
= netlink_skb_flags
;
2378 if (memcpy_from_msg(skb_put(skb
, len
), msg
, len
)) {
2383 err
= security_netlink_send(sk
, skb
);
2390 atomic_inc(&skb
->users
);
2391 netlink_broadcast(sk
, skb
, dst_portid
, dst_group
, GFP_KERNEL
);
2393 err
= netlink_unicast(sk
, skb
, dst_portid
, msg
->msg_flags
&MSG_DONTWAIT
);
2400 static int netlink_recvmsg(struct socket
*sock
, struct msghdr
*msg
, size_t len
,
2403 struct scm_cookie scm
;
2404 struct sock
*sk
= sock
->sk
;
2405 struct netlink_sock
*nlk
= nlk_sk(sk
);
2406 int noblock
= flags
&MSG_DONTWAIT
;
2408 struct sk_buff
*skb
, *data_skb
;
2416 skb
= skb_recv_datagram(sk
, flags
, noblock
, &err
);
2422 #ifdef CONFIG_COMPAT_NETLINK_MESSAGES
2423 if (unlikely(skb_shinfo(skb
)->frag_list
)) {
2425 * If this skb has a frag_list, then here that means that we
2426 * will have to use the frag_list skb's data for compat tasks
2427 * and the regular skb's data for normal (non-compat) tasks.
2429 * If we need to send the compat skb, assign it to the
2430 * 'data_skb' variable so that it will be used below for data
2431 * copying. We keep 'skb' for everything else, including
2432 * freeing both later.
2434 if (flags
& MSG_CMSG_COMPAT
)
2435 data_skb
= skb_shinfo(skb
)->frag_list
;
2439 /* Record the max length of recvmsg() calls for future allocations */
2440 nlk
->max_recvmsg_len
= max(nlk
->max_recvmsg_len
, len
);
2441 nlk
->max_recvmsg_len
= min_t(size_t, nlk
->max_recvmsg_len
,
2444 copied
= data_skb
->len
;
2446 msg
->msg_flags
|= MSG_TRUNC
;
2450 skb_reset_transport_header(data_skb
);
2451 err
= skb_copy_datagram_msg(data_skb
, 0, msg
, copied
);
2453 if (msg
->msg_name
) {
2454 DECLARE_SOCKADDR(struct sockaddr_nl
*, addr
, msg
->msg_name
);
2455 addr
->nl_family
= AF_NETLINK
;
2457 addr
->nl_pid
= NETLINK_CB(skb
).portid
;
2458 addr
->nl_groups
= netlink_group_mask(NETLINK_CB(skb
).dst_group
);
2459 msg
->msg_namelen
= sizeof(*addr
);
2462 if (nlk
->flags
& NETLINK_F_RECV_PKTINFO
)
2463 netlink_cmsg_recv_pktinfo(msg
, skb
);
2464 if (nlk
->flags
& NETLINK_F_LISTEN_ALL_NSID
)
2465 netlink_cmsg_listen_all_nsid(sk
, msg
, skb
);
2467 memset(&scm
, 0, sizeof(scm
));
2468 scm
.creds
= *NETLINK_CREDS(skb
);
2469 if (flags
& MSG_TRUNC
)
2470 copied
= data_skb
->len
;
2472 skb_free_datagram(sk
, skb
);
2474 if (nlk
->cb_running
&&
2475 atomic_read(&sk
->sk_rmem_alloc
) <= sk
->sk_rcvbuf
/ 2) {
2476 ret
= netlink_dump(sk
);
2479 sk
->sk_error_report(sk
);
2483 scm_recv(sock
, msg
, &scm
, flags
);
2485 netlink_rcv_wake(sk
);
2486 return err
? : copied
;
2489 static void netlink_data_ready(struct sock
*sk
)
2495 * We export these functions to other modules. They provide a
2496 * complete set of kernel non-blocking support for message
2501 __netlink_kernel_create(struct net
*net
, int unit
, struct module
*module
,
2502 struct netlink_kernel_cfg
*cfg
)
2504 struct socket
*sock
;
2506 struct netlink_sock
*nlk
;
2507 struct listeners
*listeners
= NULL
;
2508 struct mutex
*cb_mutex
= cfg
? cfg
->cb_mutex
: NULL
;
2509 unsigned int groups
;
2513 if (unit
< 0 || unit
>= MAX_LINKS
)
2516 if (sock_create_lite(PF_NETLINK
, SOCK_DGRAM
, unit
, &sock
))
2520 * We have to just have a reference on the net from sk, but don't
2521 * get_net it. Besides, we cannot get and then put the net here.
2522 * So we create one inside init_net and the move it to net.
2525 if (__netlink_create(&init_net
, sock
, cb_mutex
, unit
) < 0)
2526 goto out_sock_release_nosk
;
2529 sk_change_net(sk
, net
);
2531 if (!cfg
|| cfg
->groups
< 32)
2534 groups
= cfg
->groups
;
2536 listeners
= kzalloc(sizeof(*listeners
) + NLGRPSZ(groups
), GFP_KERNEL
);
2538 goto out_sock_release
;
2540 sk
->sk_data_ready
= netlink_data_ready
;
2541 if (cfg
&& cfg
->input
)
2542 nlk_sk(sk
)->netlink_rcv
= cfg
->input
;
2544 if (netlink_insert(sk
, 0))
2545 goto out_sock_release
;
2548 nlk
->flags
|= NETLINK_F_KERNEL_SOCKET
;
2550 netlink_table_grab();
2551 if (!nl_table
[unit
].registered
) {
2552 nl_table
[unit
].groups
= groups
;
2553 rcu_assign_pointer(nl_table
[unit
].listeners
, listeners
);
2554 nl_table
[unit
].cb_mutex
= cb_mutex
;
2555 nl_table
[unit
].module
= module
;
2557 nl_table
[unit
].bind
= cfg
->bind
;
2558 nl_table
[unit
].unbind
= cfg
->unbind
;
2559 nl_table
[unit
].flags
= cfg
->flags
;
2561 nl_table
[unit
].compare
= cfg
->compare
;
2563 nl_table
[unit
].registered
= 1;
2566 nl_table
[unit
].registered
++;
2568 netlink_table_ungrab();
2573 netlink_kernel_release(sk
);
2576 out_sock_release_nosk
:
2580 EXPORT_SYMBOL(__netlink_kernel_create
);
2583 netlink_kernel_release(struct sock
*sk
)
2585 sk_release_kernel(sk
);
2587 EXPORT_SYMBOL(netlink_kernel_release
);
2589 int __netlink_change_ngroups(struct sock
*sk
, unsigned int groups
)
2591 struct listeners
*new, *old
;
2592 struct netlink_table
*tbl
= &nl_table
[sk
->sk_protocol
];
2597 if (NLGRPSZ(tbl
->groups
) < NLGRPSZ(groups
)) {
2598 new = kzalloc(sizeof(*new) + NLGRPSZ(groups
), GFP_ATOMIC
);
2601 old
= nl_deref_protected(tbl
->listeners
);
2602 memcpy(new->masks
, old
->masks
, NLGRPSZ(tbl
->groups
));
2603 rcu_assign_pointer(tbl
->listeners
, new);
2605 kfree_rcu(old
, rcu
);
2607 tbl
->groups
= groups
;
2613 * netlink_change_ngroups - change number of multicast groups
2615 * This changes the number of multicast groups that are available
2616 * on a certain netlink family. Note that it is not possible to
2617 * change the number of groups to below 32. Also note that it does
2618 * not implicitly call netlink_clear_multicast_users() when the
2619 * number of groups is reduced.
2621 * @sk: The kernel netlink socket, as returned by netlink_kernel_create().
2622 * @groups: The new number of groups.
2624 int netlink_change_ngroups(struct sock
*sk
, unsigned int groups
)
2628 netlink_table_grab();
2629 err
= __netlink_change_ngroups(sk
, groups
);
2630 netlink_table_ungrab();
2635 void __netlink_clear_multicast_users(struct sock
*ksk
, unsigned int group
)
2638 struct netlink_table
*tbl
= &nl_table
[ksk
->sk_protocol
];
2640 sk_for_each_bound(sk
, &tbl
->mc_list
)
2641 netlink_update_socket_mc(nlk_sk(sk
), group
, 0);
2645 __nlmsg_put(struct sk_buff
*skb
, u32 portid
, u32 seq
, int type
, int len
, int flags
)
2647 struct nlmsghdr
*nlh
;
2648 int size
= nlmsg_msg_size(len
);
2650 nlh
= (struct nlmsghdr
*)skb_put(skb
, NLMSG_ALIGN(size
));
2651 nlh
->nlmsg_type
= type
;
2652 nlh
->nlmsg_len
= size
;
2653 nlh
->nlmsg_flags
= flags
;
2654 nlh
->nlmsg_pid
= portid
;
2655 nlh
->nlmsg_seq
= seq
;
2656 if (!__builtin_constant_p(size
) || NLMSG_ALIGN(size
) - size
!= 0)
2657 memset(nlmsg_data(nlh
) + len
, 0, NLMSG_ALIGN(size
) - size
);
2660 EXPORT_SYMBOL(__nlmsg_put
);
2663 * It looks a bit ugly.
2664 * It would be better to create kernel thread.
2667 static int netlink_dump(struct sock
*sk
)
2669 struct netlink_sock
*nlk
= nlk_sk(sk
);
2670 struct netlink_callback
*cb
;
2671 struct sk_buff
*skb
= NULL
;
2672 struct nlmsghdr
*nlh
;
2673 int len
, err
= -ENOBUFS
;
2676 mutex_lock(nlk
->cb_mutex
);
2677 if (!nlk
->cb_running
) {
2683 alloc_size
= max_t(int, cb
->min_dump_alloc
, NLMSG_GOODSIZE
);
2685 if (!netlink_rx_is_mmaped(sk
) &&
2686 atomic_read(&sk
->sk_rmem_alloc
) >= sk
->sk_rcvbuf
)
2689 /* NLMSG_GOODSIZE is small to avoid high order allocations being
2690 * required, but it makes sense to _attempt_ a 16K bytes allocation
2691 * to reduce number of system calls on dump operations, if user
2692 * ever provided a big enough buffer.
2694 if (alloc_size
< nlk
->max_recvmsg_len
) {
2695 skb
= netlink_alloc_skb(sk
,
2696 nlk
->max_recvmsg_len
,
2701 /* available room should be exact amount to avoid MSG_TRUNC */
2703 skb_reserve(skb
, skb_tailroom(skb
) -
2704 nlk
->max_recvmsg_len
);
2707 skb
= netlink_alloc_skb(sk
, alloc_size
, nlk
->portid
,
2711 netlink_skb_set_owner_r(skb
, sk
);
2713 len
= cb
->dump(skb
, cb
);
2716 mutex_unlock(nlk
->cb_mutex
);
2718 if (sk_filter(sk
, skb
))
2721 __netlink_sendskb(sk
, skb
);
2725 nlh
= nlmsg_put_answer(skb
, cb
, NLMSG_DONE
, sizeof(len
), NLM_F_MULTI
);
2729 nl_dump_check_consistent(cb
, nlh
);
2731 memcpy(nlmsg_data(nlh
), &len
, sizeof(len
));
2733 if (sk_filter(sk
, skb
))
2736 __netlink_sendskb(sk
, skb
);
2741 nlk
->cb_running
= false;
2742 mutex_unlock(nlk
->cb_mutex
);
2743 module_put(cb
->module
);
2744 consume_skb(cb
->skb
);
2748 mutex_unlock(nlk
->cb_mutex
);
2753 int __netlink_dump_start(struct sock
*ssk
, struct sk_buff
*skb
,
2754 const struct nlmsghdr
*nlh
,
2755 struct netlink_dump_control
*control
)
2757 struct netlink_callback
*cb
;
2759 struct netlink_sock
*nlk
;
2762 /* Memory mapped dump requests need to be copied to avoid looping
2763 * on the pending state in netlink_mmap_sendmsg() while the CB hold
2764 * a reference to the skb.
2766 if (netlink_skb_is_mmaped(skb
)) {
2767 skb
= skb_copy(skb
, GFP_KERNEL
);
2771 atomic_inc(&skb
->users
);
2773 sk
= netlink_lookup(sock_net(ssk
), ssk
->sk_protocol
, NETLINK_CB(skb
).portid
);
2775 ret
= -ECONNREFUSED
;
2780 mutex_lock(nlk
->cb_mutex
);
2781 /* A dump is in progress... */
2782 if (nlk
->cb_running
) {
2786 /* add reference of module which cb->dump belongs to */
2787 if (!try_module_get(control
->module
)) {
2788 ret
= -EPROTONOSUPPORT
;
2793 memset(cb
, 0, sizeof(*cb
));
2794 cb
->dump
= control
->dump
;
2795 cb
->done
= control
->done
;
2797 cb
->data
= control
->data
;
2798 cb
->module
= control
->module
;
2799 cb
->min_dump_alloc
= control
->min_dump_alloc
;
2802 nlk
->cb_running
= true;
2804 mutex_unlock(nlk
->cb_mutex
);
2806 ret
= netlink_dump(sk
);
2812 /* We successfully started a dump, by returning -EINTR we
2813 * signal not to send ACK even if it was requested.
2819 mutex_unlock(nlk
->cb_mutex
);
2824 EXPORT_SYMBOL(__netlink_dump_start
);
2826 void netlink_ack(struct sk_buff
*in_skb
, struct nlmsghdr
*nlh
, int err
)
2828 struct sk_buff
*skb
;
2829 struct nlmsghdr
*rep
;
2830 struct nlmsgerr
*errmsg
;
2831 size_t payload
= sizeof(*errmsg
);
2833 /* error messages get the original request appened */
2835 payload
+= nlmsg_len(nlh
);
2837 skb
= netlink_alloc_skb(in_skb
->sk
, nlmsg_total_size(payload
),
2838 NETLINK_CB(in_skb
).portid
, GFP_KERNEL
);
2842 sk
= netlink_lookup(sock_net(in_skb
->sk
),
2843 in_skb
->sk
->sk_protocol
,
2844 NETLINK_CB(in_skb
).portid
);
2846 sk
->sk_err
= ENOBUFS
;
2847 sk
->sk_error_report(sk
);
2853 rep
= __nlmsg_put(skb
, NETLINK_CB(in_skb
).portid
, nlh
->nlmsg_seq
,
2854 NLMSG_ERROR
, payload
, 0);
2855 errmsg
= nlmsg_data(rep
);
2856 errmsg
->error
= err
;
2857 memcpy(&errmsg
->msg
, nlh
, err
? nlh
->nlmsg_len
: sizeof(*nlh
));
2858 netlink_unicast(in_skb
->sk
, skb
, NETLINK_CB(in_skb
).portid
, MSG_DONTWAIT
);
2860 EXPORT_SYMBOL(netlink_ack
);
2862 int netlink_rcv_skb(struct sk_buff
*skb
, int (*cb
)(struct sk_buff
*,
2865 struct nlmsghdr
*nlh
;
2868 while (skb
->len
>= nlmsg_total_size(0)) {
2871 nlh
= nlmsg_hdr(skb
);
2874 if (nlh
->nlmsg_len
< NLMSG_HDRLEN
|| skb
->len
< nlh
->nlmsg_len
)
2877 /* Only requests are handled by the kernel */
2878 if (!(nlh
->nlmsg_flags
& NLM_F_REQUEST
))
2881 /* Skip control messages */
2882 if (nlh
->nlmsg_type
< NLMSG_MIN_TYPE
)
2890 if (nlh
->nlmsg_flags
& NLM_F_ACK
|| err
)
2891 netlink_ack(skb
, nlh
, err
);
2894 msglen
= NLMSG_ALIGN(nlh
->nlmsg_len
);
2895 if (msglen
> skb
->len
)
2897 skb_pull(skb
, msglen
);
2902 EXPORT_SYMBOL(netlink_rcv_skb
);
2905 * nlmsg_notify - send a notification netlink message
2906 * @sk: netlink socket to use
2907 * @skb: notification message
2908 * @portid: destination netlink portid for reports or 0
2909 * @group: destination multicast group or 0
2910 * @report: 1 to report back, 0 to disable
2911 * @flags: allocation flags
2913 int nlmsg_notify(struct sock
*sk
, struct sk_buff
*skb
, u32 portid
,
2914 unsigned int group
, int report
, gfp_t flags
)
2919 int exclude_portid
= 0;
2922 atomic_inc(&skb
->users
);
2923 exclude_portid
= portid
;
2926 /* errors reported via destination sk->sk_err, but propagate
2927 * delivery errors if NETLINK_BROADCAST_ERROR flag is set */
2928 err
= nlmsg_multicast(sk
, skb
, exclude_portid
, group
, flags
);
2934 err2
= nlmsg_unicast(sk
, skb
, portid
);
2935 if (!err
|| err
== -ESRCH
)
2941 EXPORT_SYMBOL(nlmsg_notify
);
2943 #ifdef CONFIG_PROC_FS
2944 struct nl_seq_iter
{
2945 struct seq_net_private p
;
2946 struct rhashtable_iter hti
;
2950 static int netlink_walk_start(struct nl_seq_iter
*iter
)
2954 err
= rhashtable_walk_init(&nl_table
[iter
->link
].hash
, &iter
->hti
);
2956 iter
->link
= MAX_LINKS
;
2960 err
= rhashtable_walk_start(&iter
->hti
);
2961 return err
== -EAGAIN
? 0 : err
;
2964 static void netlink_walk_stop(struct nl_seq_iter
*iter
)
2966 rhashtable_walk_stop(&iter
->hti
);
2967 rhashtable_walk_exit(&iter
->hti
);
2970 static void *__netlink_seq_next(struct seq_file
*seq
)
2972 struct nl_seq_iter
*iter
= seq
->private;
2973 struct netlink_sock
*nlk
;
2979 nlk
= rhashtable_walk_next(&iter
->hti
);
2982 if (PTR_ERR(nlk
) == -EAGAIN
)
2991 netlink_walk_stop(iter
);
2992 if (++iter
->link
>= MAX_LINKS
)
2995 err
= netlink_walk_start(iter
);
2997 return ERR_PTR(err
);
2999 } while (sock_net(&nlk
->sk
) != seq_file_net(seq
));
3004 static void *netlink_seq_start(struct seq_file
*seq
, loff_t
*posp
)
3006 struct nl_seq_iter
*iter
= seq
->private;
3007 void *obj
= SEQ_START_TOKEN
;
3013 err
= netlink_walk_start(iter
);
3015 return ERR_PTR(err
);
3017 for (pos
= *posp
; pos
&& obj
&& !IS_ERR(obj
); pos
--)
3018 obj
= __netlink_seq_next(seq
);
3023 static void *netlink_seq_next(struct seq_file
*seq
, void *v
, loff_t
*pos
)
3026 return __netlink_seq_next(seq
);
3029 static void netlink_seq_stop(struct seq_file
*seq
, void *v
)
3031 struct nl_seq_iter
*iter
= seq
->private;
3033 if (iter
->link
>= MAX_LINKS
)
3036 netlink_walk_stop(iter
);
3040 static int netlink_seq_show(struct seq_file
*seq
, void *v
)
3042 if (v
== SEQ_START_TOKEN
) {
3044 "sk Eth Pid Groups "
3045 "Rmem Wmem Dump Locks Drops Inode\n");
3048 struct netlink_sock
*nlk
= nlk_sk(s
);
3050 seq_printf(seq
, "%pK %-3d %-6u %08x %-8d %-8d %d %-8d %-8d %-8lu\n",
3054 nlk
->groups
? (u32
)nlk
->groups
[0] : 0,
3055 sk_rmem_alloc_get(s
),
3056 sk_wmem_alloc_get(s
),
3058 atomic_read(&s
->sk_refcnt
),
3059 atomic_read(&s
->sk_drops
),
3067 static const struct seq_operations netlink_seq_ops
= {
3068 .start
= netlink_seq_start
,
3069 .next
= netlink_seq_next
,
3070 .stop
= netlink_seq_stop
,
3071 .show
= netlink_seq_show
,
3075 static int netlink_seq_open(struct inode
*inode
, struct file
*file
)
3077 return seq_open_net(inode
, file
, &netlink_seq_ops
,
3078 sizeof(struct nl_seq_iter
));
3081 static const struct file_operations netlink_seq_fops
= {
3082 .owner
= THIS_MODULE
,
3083 .open
= netlink_seq_open
,
3085 .llseek
= seq_lseek
,
3086 .release
= seq_release_net
,
3091 int netlink_register_notifier(struct notifier_block
*nb
)
3093 return atomic_notifier_chain_register(&netlink_chain
, nb
);
3095 EXPORT_SYMBOL(netlink_register_notifier
);
3097 int netlink_unregister_notifier(struct notifier_block
*nb
)
3099 return atomic_notifier_chain_unregister(&netlink_chain
, nb
);
3101 EXPORT_SYMBOL(netlink_unregister_notifier
);
3103 static const struct proto_ops netlink_ops
= {
3104 .family
= PF_NETLINK
,
3105 .owner
= THIS_MODULE
,
3106 .release
= netlink_release
,
3107 .bind
= netlink_bind
,
3108 .connect
= netlink_connect
,
3109 .socketpair
= sock_no_socketpair
,
3110 .accept
= sock_no_accept
,
3111 .getname
= netlink_getname
,
3112 .poll
= netlink_poll
,
3113 .ioctl
= sock_no_ioctl
,
3114 .listen
= sock_no_listen
,
3115 .shutdown
= sock_no_shutdown
,
3116 .setsockopt
= netlink_setsockopt
,
3117 .getsockopt
= netlink_getsockopt
,
3118 .sendmsg
= netlink_sendmsg
,
3119 .recvmsg
= netlink_recvmsg
,
3120 .mmap
= netlink_mmap
,
3121 .sendpage
= sock_no_sendpage
,
3124 static const struct net_proto_family netlink_family_ops
= {
3125 .family
= PF_NETLINK
,
3126 .create
= netlink_create
,
3127 .owner
= THIS_MODULE
, /* for consistency 8) */
3130 static int __net_init
netlink_net_init(struct net
*net
)
3132 #ifdef CONFIG_PROC_FS
3133 if (!proc_create("netlink", 0, net
->proc_net
, &netlink_seq_fops
))
3139 static void __net_exit
netlink_net_exit(struct net
*net
)
3141 #ifdef CONFIG_PROC_FS
3142 remove_proc_entry("netlink", net
->proc_net
);
3146 static void __init
netlink_add_usersock_entry(void)
3148 struct listeners
*listeners
;
3151 listeners
= kzalloc(sizeof(*listeners
) + NLGRPSZ(groups
), GFP_KERNEL
);
3153 panic("netlink_add_usersock_entry: Cannot allocate listeners\n");
3155 netlink_table_grab();
3157 nl_table
[NETLINK_USERSOCK
].groups
= groups
;
3158 rcu_assign_pointer(nl_table
[NETLINK_USERSOCK
].listeners
, listeners
);
3159 nl_table
[NETLINK_USERSOCK
].module
= THIS_MODULE
;
3160 nl_table
[NETLINK_USERSOCK
].registered
= 1;
3161 nl_table
[NETLINK_USERSOCK
].flags
= NL_CFG_F_NONROOT_SEND
;
3163 netlink_table_ungrab();
3166 static struct pernet_operations __net_initdata netlink_net_ops
= {
3167 .init
= netlink_net_init
,
3168 .exit
= netlink_net_exit
,
3171 static inline u32
netlink_hash(const void *data
, u32 len
, u32 seed
)
3173 const struct netlink_sock
*nlk
= data
;
3174 struct netlink_compare_arg arg
;
3176 netlink_compare_arg_init(&arg
, sock_net(&nlk
->sk
), nlk
->portid
);
3177 return jhash2((u32
*)&arg
, netlink_compare_arg_len
/ sizeof(u32
), seed
);
3180 static const struct rhashtable_params netlink_rhashtable_params
= {
3181 .head_offset
= offsetof(struct netlink_sock
, node
),
3182 .key_len
= netlink_compare_arg_len
,
3183 .obj_hashfn
= netlink_hash
,
3184 .obj_cmpfn
= netlink_compare
,
3186 .automatic_shrinking
= true,
3189 static int __init
netlink_proto_init(void)
3192 int err
= proto_register(&netlink_proto
, 0);
3197 BUILD_BUG_ON(sizeof(struct netlink_skb_parms
) > FIELD_SIZEOF(struct sk_buff
, cb
));
3199 nl_table
= kcalloc(MAX_LINKS
, sizeof(*nl_table
), GFP_KERNEL
);
3203 for (i
= 0; i
< MAX_LINKS
; i
++) {
3204 if (rhashtable_init(&nl_table
[i
].hash
,
3205 &netlink_rhashtable_params
) < 0) {
3207 rhashtable_destroy(&nl_table
[i
].hash
);
3213 INIT_LIST_HEAD(&netlink_tap_all
);
3215 netlink_add_usersock_entry();
3217 sock_register(&netlink_family_ops
);
3218 register_pernet_subsys(&netlink_net_ops
);
3219 /* The netlink device handler may be needed early. */
3224 panic("netlink_init: Cannot allocate nl_table\n");
3227 core_initcall(netlink_proto_init
);