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 __read_mostly
;
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
);
1086 nlk_sk(sk
)->portid
= 0;
1095 static void netlink_remove(struct sock
*sk
)
1097 struct netlink_table
*table
;
1099 table
= &nl_table
[sk
->sk_protocol
];
1100 if (!rhashtable_remove_fast(&table
->hash
, &nlk_sk(sk
)->node
,
1101 netlink_rhashtable_params
)) {
1102 WARN_ON(atomic_read(&sk
->sk_refcnt
) == 1);
1106 netlink_table_grab();
1107 if (nlk_sk(sk
)->subscriptions
) {
1108 __sk_del_bind_node(sk
);
1109 netlink_update_listeners(sk
);
1111 if (sk
->sk_protocol
== NETLINK_GENERIC
)
1112 atomic_inc(&genl_sk_destructing_cnt
);
1113 netlink_table_ungrab();
1116 static struct proto netlink_proto
= {
1118 .owner
= THIS_MODULE
,
1119 .obj_size
= sizeof(struct netlink_sock
),
1122 static int __netlink_create(struct net
*net
, struct socket
*sock
,
1123 struct mutex
*cb_mutex
, int protocol
,
1127 struct netlink_sock
*nlk
;
1129 sock
->ops
= &netlink_ops
;
1131 sk
= sk_alloc(net
, PF_NETLINK
, GFP_KERNEL
, &netlink_proto
, kern
);
1135 sock_init_data(sock
, sk
);
1139 nlk
->cb_mutex
= cb_mutex
;
1141 nlk
->cb_mutex
= &nlk
->cb_def_mutex
;
1142 mutex_init(nlk
->cb_mutex
);
1144 init_waitqueue_head(&nlk
->wait
);
1145 #ifdef CONFIG_NETLINK_MMAP
1146 mutex_init(&nlk
->pg_vec_lock
);
1149 sk
->sk_destruct
= netlink_sock_destruct
;
1150 sk
->sk_protocol
= protocol
;
1154 static int netlink_create(struct net
*net
, struct socket
*sock
, int protocol
,
1157 struct module
*module
= NULL
;
1158 struct mutex
*cb_mutex
;
1159 struct netlink_sock
*nlk
;
1160 int (*bind
)(struct net
*net
, int group
);
1161 void (*unbind
)(struct net
*net
, int group
);
1164 sock
->state
= SS_UNCONNECTED
;
1166 if (sock
->type
!= SOCK_RAW
&& sock
->type
!= SOCK_DGRAM
)
1167 return -ESOCKTNOSUPPORT
;
1169 if (protocol
< 0 || protocol
>= MAX_LINKS
)
1170 return -EPROTONOSUPPORT
;
1172 netlink_lock_table();
1173 #ifdef CONFIG_MODULES
1174 if (!nl_table
[protocol
].registered
) {
1175 netlink_unlock_table();
1176 request_module("net-pf-%d-proto-%d", PF_NETLINK
, protocol
);
1177 netlink_lock_table();
1180 if (nl_table
[protocol
].registered
&&
1181 try_module_get(nl_table
[protocol
].module
))
1182 module
= nl_table
[protocol
].module
;
1184 err
= -EPROTONOSUPPORT
;
1185 cb_mutex
= nl_table
[protocol
].cb_mutex
;
1186 bind
= nl_table
[protocol
].bind
;
1187 unbind
= nl_table
[protocol
].unbind
;
1188 netlink_unlock_table();
1193 err
= __netlink_create(net
, sock
, cb_mutex
, protocol
, kern
);
1198 sock_prot_inuse_add(net
, &netlink_proto
, 1);
1201 nlk
= nlk_sk(sock
->sk
);
1202 nlk
->module
= module
;
1203 nlk
->netlink_bind
= bind
;
1204 nlk
->netlink_unbind
= unbind
;
1213 static void deferred_put_nlk_sk(struct rcu_head
*head
)
1215 struct netlink_sock
*nlk
= container_of(head
, struct netlink_sock
, rcu
);
1220 static int netlink_release(struct socket
*sock
)
1222 struct sock
*sk
= sock
->sk
;
1223 struct netlink_sock
*nlk
;
1233 * OK. Socket is unlinked, any packets that arrive now
1237 /* must not acquire netlink_table_lock in any way again before unbind
1238 * and notifying genetlink is done as otherwise it might deadlock
1240 if (nlk
->netlink_unbind
) {
1243 for (i
= 0; i
< nlk
->ngroups
; i
++)
1244 if (test_bit(i
, nlk
->groups
))
1245 nlk
->netlink_unbind(sock_net(sk
), i
+ 1);
1247 if (sk
->sk_protocol
== NETLINK_GENERIC
&&
1248 atomic_dec_return(&genl_sk_destructing_cnt
) == 0)
1249 wake_up(&genl_sk_destructing_waitq
);
1252 wake_up_interruptible_all(&nlk
->wait
);
1254 skb_queue_purge(&sk
->sk_write_queue
);
1257 struct netlink_notify n
= {
1258 .net
= sock_net(sk
),
1259 .protocol
= sk
->sk_protocol
,
1260 .portid
= nlk
->portid
,
1262 atomic_notifier_call_chain(&netlink_chain
,
1263 NETLINK_URELEASE
, &n
);
1266 module_put(nlk
->module
);
1268 if (netlink_is_kernel(sk
)) {
1269 netlink_table_grab();
1270 BUG_ON(nl_table
[sk
->sk_protocol
].registered
== 0);
1271 if (--nl_table
[sk
->sk_protocol
].registered
== 0) {
1272 struct listeners
*old
;
1274 old
= nl_deref_protected(nl_table
[sk
->sk_protocol
].listeners
);
1275 RCU_INIT_POINTER(nl_table
[sk
->sk_protocol
].listeners
, NULL
);
1276 kfree_rcu(old
, rcu
);
1277 nl_table
[sk
->sk_protocol
].module
= NULL
;
1278 nl_table
[sk
->sk_protocol
].bind
= NULL
;
1279 nl_table
[sk
->sk_protocol
].unbind
= NULL
;
1280 nl_table
[sk
->sk_protocol
].flags
= 0;
1281 nl_table
[sk
->sk_protocol
].registered
= 0;
1283 netlink_table_ungrab();
1290 sock_prot_inuse_add(sock_net(sk
), &netlink_proto
, -1);
1292 call_rcu(&nlk
->rcu
, deferred_put_nlk_sk
);
1296 static int netlink_autobind(struct socket
*sock
)
1298 struct sock
*sk
= sock
->sk
;
1299 struct net
*net
= sock_net(sk
);
1300 struct netlink_table
*table
= &nl_table
[sk
->sk_protocol
];
1301 s32 portid
= task_tgid_vnr(current
);
1309 ok
= !__netlink_lookup(table
, portid
, net
);
1312 /* Bind collision, search negative portid values. */
1314 /* rover will be in range [S32_MIN, -4097] */
1315 rover
= S32_MIN
+ prandom_u32_max(-4096 - S32_MIN
);
1316 else if (rover
>= -4096)
1322 err
= netlink_insert(sk
, portid
);
1323 if (err
== -EADDRINUSE
)
1326 /* If 2 threads race to autobind, that is fine. */
1334 * __netlink_ns_capable - General netlink message capability test
1335 * @nsp: NETLINK_CB of the socket buffer holding a netlink command from userspace.
1336 * @user_ns: The user namespace of the capability to use
1337 * @cap: The capability to use
1339 * Test to see if the opener of the socket we received the message
1340 * from had when the netlink socket was created and the sender of the
1341 * message has has the capability @cap in the user namespace @user_ns.
1343 bool __netlink_ns_capable(const struct netlink_skb_parms
*nsp
,
1344 struct user_namespace
*user_ns
, int cap
)
1346 return ((nsp
->flags
& NETLINK_SKB_DST
) ||
1347 file_ns_capable(nsp
->sk
->sk_socket
->file
, user_ns
, cap
)) &&
1348 ns_capable(user_ns
, cap
);
1350 EXPORT_SYMBOL(__netlink_ns_capable
);
1353 * netlink_ns_capable - General netlink message capability test
1354 * @skb: socket buffer holding a netlink command from userspace
1355 * @user_ns: The user namespace of the capability to use
1356 * @cap: The capability to use
1358 * Test to see if the opener of the socket we received the message
1359 * from had when the netlink socket was created and the sender of the
1360 * message has has the capability @cap in the user namespace @user_ns.
1362 bool netlink_ns_capable(const struct sk_buff
*skb
,
1363 struct user_namespace
*user_ns
, int cap
)
1365 return __netlink_ns_capable(&NETLINK_CB(skb
), user_ns
, cap
);
1367 EXPORT_SYMBOL(netlink_ns_capable
);
1370 * netlink_capable - Netlink global message capability test
1371 * @skb: socket buffer holding a netlink command from userspace
1372 * @cap: The capability to use
1374 * Test to see if the opener of the socket we received the message
1375 * from had when the netlink socket was created and the sender of the
1376 * message has has the capability @cap in all user namespaces.
1378 bool netlink_capable(const struct sk_buff
*skb
, int cap
)
1380 return netlink_ns_capable(skb
, &init_user_ns
, cap
);
1382 EXPORT_SYMBOL(netlink_capable
);
1385 * netlink_net_capable - Netlink network namespace message capability test
1386 * @skb: socket buffer holding a netlink command from userspace
1387 * @cap: The capability to use
1389 * Test to see if the opener of the socket we received the message
1390 * from had when the netlink socket was created and the sender of the
1391 * message has has the capability @cap over the network namespace of
1392 * the socket we received the message from.
1394 bool netlink_net_capable(const struct sk_buff
*skb
, int cap
)
1396 return netlink_ns_capable(skb
, sock_net(skb
->sk
)->user_ns
, cap
);
1398 EXPORT_SYMBOL(netlink_net_capable
);
1400 static inline int netlink_allowed(const struct socket
*sock
, unsigned int flag
)
1402 return (nl_table
[sock
->sk
->sk_protocol
].flags
& flag
) ||
1403 ns_capable(sock_net(sock
->sk
)->user_ns
, CAP_NET_ADMIN
);
1407 netlink_update_subscriptions(struct sock
*sk
, unsigned int subscriptions
)
1409 struct netlink_sock
*nlk
= nlk_sk(sk
);
1411 if (nlk
->subscriptions
&& !subscriptions
)
1412 __sk_del_bind_node(sk
);
1413 else if (!nlk
->subscriptions
&& subscriptions
)
1414 sk_add_bind_node(sk
, &nl_table
[sk
->sk_protocol
].mc_list
);
1415 nlk
->subscriptions
= subscriptions
;
1418 static int netlink_realloc_groups(struct sock
*sk
)
1420 struct netlink_sock
*nlk
= nlk_sk(sk
);
1421 unsigned int groups
;
1422 unsigned long *new_groups
;
1425 netlink_table_grab();
1427 groups
= nl_table
[sk
->sk_protocol
].groups
;
1428 if (!nl_table
[sk
->sk_protocol
].registered
) {
1433 if (nlk
->ngroups
>= groups
)
1436 new_groups
= krealloc(nlk
->groups
, NLGRPSZ(groups
), GFP_ATOMIC
);
1437 if (new_groups
== NULL
) {
1441 memset((char *)new_groups
+ NLGRPSZ(nlk
->ngroups
), 0,
1442 NLGRPSZ(groups
) - NLGRPSZ(nlk
->ngroups
));
1444 nlk
->groups
= new_groups
;
1445 nlk
->ngroups
= groups
;
1447 netlink_table_ungrab();
1451 static void netlink_undo_bind(int group
, long unsigned int groups
,
1454 struct netlink_sock
*nlk
= nlk_sk(sk
);
1457 if (!nlk
->netlink_unbind
)
1460 for (undo
= 0; undo
< group
; undo
++)
1461 if (test_bit(undo
, &groups
))
1462 nlk
->netlink_unbind(sock_net(sk
), undo
+ 1);
1465 static int netlink_bind(struct socket
*sock
, struct sockaddr
*addr
,
1468 struct sock
*sk
= sock
->sk
;
1469 struct net
*net
= sock_net(sk
);
1470 struct netlink_sock
*nlk
= nlk_sk(sk
);
1471 struct sockaddr_nl
*nladdr
= (struct sockaddr_nl
*)addr
;
1473 long unsigned int groups
= nladdr
->nl_groups
;
1475 if (addr_len
< sizeof(struct sockaddr_nl
))
1478 if (nladdr
->nl_family
!= AF_NETLINK
)
1481 /* Only superuser is allowed to listen multicasts */
1483 if (!netlink_allowed(sock
, NL_CFG_F_NONROOT_RECV
))
1485 err
= netlink_realloc_groups(sk
);
1491 if (nladdr
->nl_pid
!= nlk
->portid
)
1494 if (nlk
->netlink_bind
&& groups
) {
1497 for (group
= 0; group
< nlk
->ngroups
; group
++) {
1498 if (!test_bit(group
, &groups
))
1500 err
= nlk
->netlink_bind(net
, group
+ 1);
1503 netlink_undo_bind(group
, groups
, sk
);
1509 err
= nladdr
->nl_pid
?
1510 netlink_insert(sk
, nladdr
->nl_pid
) :
1511 netlink_autobind(sock
);
1513 netlink_undo_bind(nlk
->ngroups
, groups
, sk
);
1518 if (!groups
&& (nlk
->groups
== NULL
|| !(u32
)nlk
->groups
[0]))
1521 netlink_table_grab();
1522 netlink_update_subscriptions(sk
, nlk
->subscriptions
+
1524 hweight32(nlk
->groups
[0]));
1525 nlk
->groups
[0] = (nlk
->groups
[0] & ~0xffffffffUL
) | groups
;
1526 netlink_update_listeners(sk
);
1527 netlink_table_ungrab();
1532 static int netlink_connect(struct socket
*sock
, struct sockaddr
*addr
,
1533 int alen
, int flags
)
1536 struct sock
*sk
= sock
->sk
;
1537 struct netlink_sock
*nlk
= nlk_sk(sk
);
1538 struct sockaddr_nl
*nladdr
= (struct sockaddr_nl
*)addr
;
1540 if (alen
< sizeof(addr
->sa_family
))
1543 if (addr
->sa_family
== AF_UNSPEC
) {
1544 sk
->sk_state
= NETLINK_UNCONNECTED
;
1545 nlk
->dst_portid
= 0;
1549 if (addr
->sa_family
!= AF_NETLINK
)
1552 if ((nladdr
->nl_groups
|| nladdr
->nl_pid
) &&
1553 !netlink_allowed(sock
, NL_CFG_F_NONROOT_SEND
))
1557 err
= netlink_autobind(sock
);
1560 sk
->sk_state
= NETLINK_CONNECTED
;
1561 nlk
->dst_portid
= nladdr
->nl_pid
;
1562 nlk
->dst_group
= ffs(nladdr
->nl_groups
);
1568 static int netlink_getname(struct socket
*sock
, struct sockaddr
*addr
,
1569 int *addr_len
, int peer
)
1571 struct sock
*sk
= sock
->sk
;
1572 struct netlink_sock
*nlk
= nlk_sk(sk
);
1573 DECLARE_SOCKADDR(struct sockaddr_nl
*, nladdr
, addr
);
1575 nladdr
->nl_family
= AF_NETLINK
;
1577 *addr_len
= sizeof(*nladdr
);
1580 nladdr
->nl_pid
= nlk
->dst_portid
;
1581 nladdr
->nl_groups
= netlink_group_mask(nlk
->dst_group
);
1583 nladdr
->nl_pid
= nlk
->portid
;
1584 nladdr
->nl_groups
= nlk
->groups
? nlk
->groups
[0] : 0;
1589 static struct sock
*netlink_getsockbyportid(struct sock
*ssk
, u32 portid
)
1592 struct netlink_sock
*nlk
;
1594 sock
= netlink_lookup(sock_net(ssk
), ssk
->sk_protocol
, portid
);
1596 return ERR_PTR(-ECONNREFUSED
);
1598 /* Don't bother queuing skb if kernel socket has no input function */
1600 if (sock
->sk_state
== NETLINK_CONNECTED
&&
1601 nlk
->dst_portid
!= nlk_sk(ssk
)->portid
) {
1603 return ERR_PTR(-ECONNREFUSED
);
1608 struct sock
*netlink_getsockbyfilp(struct file
*filp
)
1610 struct inode
*inode
= file_inode(filp
);
1613 if (!S_ISSOCK(inode
->i_mode
))
1614 return ERR_PTR(-ENOTSOCK
);
1616 sock
= SOCKET_I(inode
)->sk
;
1617 if (sock
->sk_family
!= AF_NETLINK
)
1618 return ERR_PTR(-EINVAL
);
1624 static struct sk_buff
*netlink_alloc_large_skb(unsigned int size
,
1627 struct sk_buff
*skb
;
1630 if (size
<= NLMSG_GOODSIZE
|| broadcast
)
1631 return alloc_skb(size
, GFP_KERNEL
);
1633 size
= SKB_DATA_ALIGN(size
) +
1634 SKB_DATA_ALIGN(sizeof(struct skb_shared_info
));
1636 data
= vmalloc(size
);
1640 skb
= __build_skb(data
, size
);
1644 skb
->destructor
= netlink_skb_destructor
;
1650 * Attach a skb to a netlink socket.
1651 * The caller must hold a reference to the destination socket. On error, the
1652 * reference is dropped. The skb is not send to the destination, just all
1653 * all error checks are performed and memory in the queue is reserved.
1655 * < 0: error. skb freed, reference to sock dropped.
1657 * 1: repeat lookup - reference dropped while waiting for socket memory.
1659 int netlink_attachskb(struct sock
*sk
, struct sk_buff
*skb
,
1660 long *timeo
, struct sock
*ssk
)
1662 struct netlink_sock
*nlk
;
1666 if ((atomic_read(&sk
->sk_rmem_alloc
) > sk
->sk_rcvbuf
||
1667 test_bit(NETLINK_S_CONGESTED
, &nlk
->state
)) &&
1668 !netlink_skb_is_mmaped(skb
)) {
1669 DECLARE_WAITQUEUE(wait
, current
);
1671 if (!ssk
|| netlink_is_kernel(ssk
))
1672 netlink_overrun(sk
);
1678 __set_current_state(TASK_INTERRUPTIBLE
);
1679 add_wait_queue(&nlk
->wait
, &wait
);
1681 if ((atomic_read(&sk
->sk_rmem_alloc
) > sk
->sk_rcvbuf
||
1682 test_bit(NETLINK_S_CONGESTED
, &nlk
->state
)) &&
1683 !sock_flag(sk
, SOCK_DEAD
))
1684 *timeo
= schedule_timeout(*timeo
);
1686 __set_current_state(TASK_RUNNING
);
1687 remove_wait_queue(&nlk
->wait
, &wait
);
1690 if (signal_pending(current
)) {
1692 return sock_intr_errno(*timeo
);
1696 netlink_skb_set_owner_r(skb
, sk
);
1700 static int __netlink_sendskb(struct sock
*sk
, struct sk_buff
*skb
)
1704 netlink_deliver_tap(skb
);
1706 #ifdef CONFIG_NETLINK_MMAP
1707 if (netlink_skb_is_mmaped(skb
))
1708 netlink_queue_mmaped_skb(sk
, skb
);
1709 else if (netlink_rx_is_mmaped(sk
))
1710 netlink_ring_set_copied(sk
, skb
);
1712 #endif /* CONFIG_NETLINK_MMAP */
1713 skb_queue_tail(&sk
->sk_receive_queue
, skb
);
1714 sk
->sk_data_ready(sk
);
1718 int netlink_sendskb(struct sock
*sk
, struct sk_buff
*skb
)
1720 int len
= __netlink_sendskb(sk
, skb
);
1726 void netlink_detachskb(struct sock
*sk
, struct sk_buff
*skb
)
1732 static struct sk_buff
*netlink_trim(struct sk_buff
*skb
, gfp_t allocation
)
1736 WARN_ON(skb
->sk
!= NULL
);
1737 if (netlink_skb_is_mmaped(skb
))
1740 delta
= skb
->end
- skb
->tail
;
1741 if (is_vmalloc_addr(skb
->head
) || delta
* 2 < skb
->truesize
)
1744 if (skb_shared(skb
)) {
1745 struct sk_buff
*nskb
= skb_clone(skb
, allocation
);
1752 if (!pskb_expand_head(skb
, 0, -delta
, allocation
))
1753 skb
->truesize
-= delta
;
1758 static int netlink_unicast_kernel(struct sock
*sk
, struct sk_buff
*skb
,
1762 struct netlink_sock
*nlk
= nlk_sk(sk
);
1764 ret
= -ECONNREFUSED
;
1765 if (nlk
->netlink_rcv
!= NULL
) {
1767 netlink_skb_set_owner_r(skb
, sk
);
1768 NETLINK_CB(skb
).sk
= ssk
;
1769 netlink_deliver_tap_kernel(sk
, ssk
, skb
);
1770 nlk
->netlink_rcv(skb
);
1779 int netlink_unicast(struct sock
*ssk
, struct sk_buff
*skb
,
1780 u32 portid
, int nonblock
)
1786 skb
= netlink_trim(skb
, gfp_any());
1788 timeo
= sock_sndtimeo(ssk
, nonblock
);
1790 sk
= netlink_getsockbyportid(ssk
, portid
);
1795 if (netlink_is_kernel(sk
))
1796 return netlink_unicast_kernel(sk
, skb
, ssk
);
1798 if (sk_filter(sk
, skb
)) {
1805 err
= netlink_attachskb(sk
, skb
, &timeo
, ssk
);
1811 return netlink_sendskb(sk
, skb
);
1813 EXPORT_SYMBOL(netlink_unicast
);
1815 struct sk_buff
*netlink_alloc_skb(struct sock
*ssk
, unsigned int size
,
1816 u32 dst_portid
, gfp_t gfp_mask
)
1818 #ifdef CONFIG_NETLINK_MMAP
1819 struct sock
*sk
= NULL
;
1820 struct sk_buff
*skb
;
1821 struct netlink_ring
*ring
;
1822 struct nl_mmap_hdr
*hdr
;
1823 unsigned int maxlen
;
1825 sk
= netlink_getsockbyportid(ssk
, dst_portid
);
1829 ring
= &nlk_sk(sk
)->rx_ring
;
1830 /* fast-path without atomic ops for common case: non-mmaped receiver */
1831 if (ring
->pg_vec
== NULL
)
1834 if (ring
->frame_size
- NL_MMAP_HDRLEN
< size
)
1837 skb
= alloc_skb_head(gfp_mask
);
1841 spin_lock_bh(&sk
->sk_receive_queue
.lock
);
1842 /* check again under lock */
1843 if (ring
->pg_vec
== NULL
)
1846 /* check again under lock */
1847 maxlen
= ring
->frame_size
- NL_MMAP_HDRLEN
;
1851 netlink_forward_ring(ring
);
1852 hdr
= netlink_current_frame(ring
, NL_MMAP_STATUS_UNUSED
);
1855 netlink_ring_setup_skb(skb
, sk
, ring
, hdr
);
1856 netlink_set_status(hdr
, NL_MMAP_STATUS_RESERVED
);
1857 atomic_inc(&ring
->pending
);
1858 netlink_increment_head(ring
);
1860 spin_unlock_bh(&sk
->sk_receive_queue
.lock
);
1865 spin_unlock_bh(&sk
->sk_receive_queue
.lock
);
1866 netlink_overrun(sk
);
1873 spin_unlock_bh(&sk
->sk_receive_queue
.lock
);
1878 return alloc_skb(size
, gfp_mask
);
1880 EXPORT_SYMBOL_GPL(netlink_alloc_skb
);
1882 int netlink_has_listeners(struct sock
*sk
, unsigned int group
)
1885 struct listeners
*listeners
;
1887 BUG_ON(!netlink_is_kernel(sk
));
1890 listeners
= rcu_dereference(nl_table
[sk
->sk_protocol
].listeners
);
1892 if (listeners
&& group
- 1 < nl_table
[sk
->sk_protocol
].groups
)
1893 res
= test_bit(group
- 1, listeners
->masks
);
1899 EXPORT_SYMBOL_GPL(netlink_has_listeners
);
1901 static int netlink_broadcast_deliver(struct sock
*sk
, struct sk_buff
*skb
)
1903 struct netlink_sock
*nlk
= nlk_sk(sk
);
1905 if (atomic_read(&sk
->sk_rmem_alloc
) <= sk
->sk_rcvbuf
&&
1906 !test_bit(NETLINK_S_CONGESTED
, &nlk
->state
)) {
1907 netlink_skb_set_owner_r(skb
, sk
);
1908 __netlink_sendskb(sk
, skb
);
1909 return atomic_read(&sk
->sk_rmem_alloc
) > (sk
->sk_rcvbuf
>> 1);
1914 struct netlink_broadcast_data
{
1915 struct sock
*exclude_sk
;
1920 int delivery_failure
;
1924 struct sk_buff
*skb
, *skb2
;
1925 int (*tx_filter
)(struct sock
*dsk
, struct sk_buff
*skb
, void *data
);
1929 static void do_one_broadcast(struct sock
*sk
,
1930 struct netlink_broadcast_data
*p
)
1932 struct netlink_sock
*nlk
= nlk_sk(sk
);
1935 if (p
->exclude_sk
== sk
)
1938 if (nlk
->portid
== p
->portid
|| p
->group
- 1 >= nlk
->ngroups
||
1939 !test_bit(p
->group
- 1, nlk
->groups
))
1942 if (!net_eq(sock_net(sk
), p
->net
)) {
1943 if (!(nlk
->flags
& NETLINK_F_LISTEN_ALL_NSID
))
1946 if (!peernet_has_id(sock_net(sk
), p
->net
))
1949 if (!file_ns_capable(sk
->sk_socket
->file
, p
->net
->user_ns
,
1955 netlink_overrun(sk
);
1960 if (p
->skb2
== NULL
) {
1961 if (skb_shared(p
->skb
)) {
1962 p
->skb2
= skb_clone(p
->skb
, p
->allocation
);
1964 p
->skb2
= skb_get(p
->skb
);
1966 * skb ownership may have been set when
1967 * delivered to a previous socket.
1969 skb_orphan(p
->skb2
);
1972 if (p
->skb2
== NULL
) {
1973 netlink_overrun(sk
);
1974 /* Clone failed. Notify ALL listeners. */
1976 if (nlk
->flags
& NETLINK_F_BROADCAST_SEND_ERROR
)
1977 p
->delivery_failure
= 1;
1980 if (p
->tx_filter
&& p
->tx_filter(sk
, p
->skb2
, p
->tx_data
)) {
1985 if (sk_filter(sk
, p
->skb2
)) {
1990 NETLINK_CB(p
->skb2
).nsid
= peernet2id(sock_net(sk
), p
->net
);
1991 NETLINK_CB(p
->skb2
).nsid_is_set
= true;
1992 val
= netlink_broadcast_deliver(sk
, p
->skb2
);
1994 netlink_overrun(sk
);
1995 if (nlk
->flags
& NETLINK_F_BROADCAST_SEND_ERROR
)
1996 p
->delivery_failure
= 1;
1998 p
->congested
|= val
;
2006 int netlink_broadcast_filtered(struct sock
*ssk
, struct sk_buff
*skb
, u32 portid
,
2007 u32 group
, gfp_t allocation
,
2008 int (*filter
)(struct sock
*dsk
, struct sk_buff
*skb
, void *data
),
2011 struct net
*net
= sock_net(ssk
);
2012 struct netlink_broadcast_data info
;
2015 skb
= netlink_trim(skb
, allocation
);
2017 info
.exclude_sk
= ssk
;
2019 info
.portid
= portid
;
2022 info
.delivery_failure
= 0;
2025 info
.allocation
= allocation
;
2028 info
.tx_filter
= filter
;
2029 info
.tx_data
= filter_data
;
2031 /* While we sleep in clone, do not allow to change socket list */
2033 netlink_lock_table();
2035 sk_for_each_bound(sk
, &nl_table
[ssk
->sk_protocol
].mc_list
)
2036 do_one_broadcast(sk
, &info
);
2040 netlink_unlock_table();
2042 if (info
.delivery_failure
) {
2043 kfree_skb(info
.skb2
);
2046 consume_skb(info
.skb2
);
2048 if (info
.delivered
) {
2049 if (info
.congested
&& (allocation
& __GFP_WAIT
))
2055 EXPORT_SYMBOL(netlink_broadcast_filtered
);
2057 int netlink_broadcast(struct sock
*ssk
, struct sk_buff
*skb
, u32 portid
,
2058 u32 group
, gfp_t allocation
)
2060 return netlink_broadcast_filtered(ssk
, skb
, portid
, group
, allocation
,
2063 EXPORT_SYMBOL(netlink_broadcast
);
2065 struct netlink_set_err_data
{
2066 struct sock
*exclude_sk
;
2072 static int do_one_set_err(struct sock
*sk
, struct netlink_set_err_data
*p
)
2074 struct netlink_sock
*nlk
= nlk_sk(sk
);
2077 if (sk
== p
->exclude_sk
)
2080 if (!net_eq(sock_net(sk
), sock_net(p
->exclude_sk
)))
2083 if (nlk
->portid
== p
->portid
|| p
->group
- 1 >= nlk
->ngroups
||
2084 !test_bit(p
->group
- 1, nlk
->groups
))
2087 if (p
->code
== ENOBUFS
&& nlk
->flags
& NETLINK_F_RECV_NO_ENOBUFS
) {
2092 sk
->sk_err
= p
->code
;
2093 sk
->sk_error_report(sk
);
2099 * netlink_set_err - report error to broadcast listeners
2100 * @ssk: the kernel netlink socket, as returned by netlink_kernel_create()
2101 * @portid: the PORTID of a process that we want to skip (if any)
2102 * @group: the broadcast group that will notice the error
2103 * @code: error code, must be negative (as usual in kernelspace)
2105 * This function returns the number of broadcast listeners that have set the
2106 * NETLINK_NO_ENOBUFS socket option.
2108 int netlink_set_err(struct sock
*ssk
, u32 portid
, u32 group
, int code
)
2110 struct netlink_set_err_data info
;
2114 info
.exclude_sk
= ssk
;
2115 info
.portid
= portid
;
2117 /* sk->sk_err wants a positive error value */
2120 read_lock(&nl_table_lock
);
2122 sk_for_each_bound(sk
, &nl_table
[ssk
->sk_protocol
].mc_list
)
2123 ret
+= do_one_set_err(sk
, &info
);
2125 read_unlock(&nl_table_lock
);
2128 EXPORT_SYMBOL(netlink_set_err
);
2130 /* must be called with netlink table grabbed */
2131 static void netlink_update_socket_mc(struct netlink_sock
*nlk
,
2135 int old
, new = !!is_new
, subscriptions
;
2137 old
= test_bit(group
- 1, nlk
->groups
);
2138 subscriptions
= nlk
->subscriptions
- old
+ new;
2140 __set_bit(group
- 1, nlk
->groups
);
2142 __clear_bit(group
- 1, nlk
->groups
);
2143 netlink_update_subscriptions(&nlk
->sk
, subscriptions
);
2144 netlink_update_listeners(&nlk
->sk
);
2147 static int netlink_setsockopt(struct socket
*sock
, int level
, int optname
,
2148 char __user
*optval
, unsigned int optlen
)
2150 struct sock
*sk
= sock
->sk
;
2151 struct netlink_sock
*nlk
= nlk_sk(sk
);
2152 unsigned int val
= 0;
2155 if (level
!= SOL_NETLINK
)
2156 return -ENOPROTOOPT
;
2158 if (optname
!= NETLINK_RX_RING
&& optname
!= NETLINK_TX_RING
&&
2159 optlen
>= sizeof(int) &&
2160 get_user(val
, (unsigned int __user
*)optval
))
2164 case NETLINK_PKTINFO
:
2166 nlk
->flags
|= NETLINK_F_RECV_PKTINFO
;
2168 nlk
->flags
&= ~NETLINK_F_RECV_PKTINFO
;
2171 case NETLINK_ADD_MEMBERSHIP
:
2172 case NETLINK_DROP_MEMBERSHIP
: {
2173 if (!netlink_allowed(sock
, NL_CFG_F_NONROOT_RECV
))
2175 err
= netlink_realloc_groups(sk
);
2178 if (!val
|| val
- 1 >= nlk
->ngroups
)
2180 if (optname
== NETLINK_ADD_MEMBERSHIP
&& nlk
->netlink_bind
) {
2181 err
= nlk
->netlink_bind(sock_net(sk
), val
);
2185 netlink_table_grab();
2186 netlink_update_socket_mc(nlk
, val
,
2187 optname
== NETLINK_ADD_MEMBERSHIP
);
2188 netlink_table_ungrab();
2189 if (optname
== NETLINK_DROP_MEMBERSHIP
&& nlk
->netlink_unbind
)
2190 nlk
->netlink_unbind(sock_net(sk
), val
);
2195 case NETLINK_BROADCAST_ERROR
:
2197 nlk
->flags
|= NETLINK_F_BROADCAST_SEND_ERROR
;
2199 nlk
->flags
&= ~NETLINK_F_BROADCAST_SEND_ERROR
;
2202 case NETLINK_NO_ENOBUFS
:
2204 nlk
->flags
|= NETLINK_F_RECV_NO_ENOBUFS
;
2205 clear_bit(NETLINK_S_CONGESTED
, &nlk
->state
);
2206 wake_up_interruptible(&nlk
->wait
);
2208 nlk
->flags
&= ~NETLINK_F_RECV_NO_ENOBUFS
;
2212 #ifdef CONFIG_NETLINK_MMAP
2213 case NETLINK_RX_RING
:
2214 case NETLINK_TX_RING
: {
2215 struct nl_mmap_req req
;
2217 /* Rings might consume more memory than queue limits, require
2220 if (!capable(CAP_NET_ADMIN
))
2222 if (optlen
< sizeof(req
))
2224 if (copy_from_user(&req
, optval
, sizeof(req
)))
2226 err
= netlink_set_ring(sk
, &req
, false,
2227 optname
== NETLINK_TX_RING
);
2230 #endif /* CONFIG_NETLINK_MMAP */
2231 case NETLINK_LISTEN_ALL_NSID
:
2232 if (!ns_capable(sock_net(sk
)->user_ns
, CAP_NET_BROADCAST
))
2236 nlk
->flags
|= NETLINK_F_LISTEN_ALL_NSID
;
2238 nlk
->flags
&= ~NETLINK_F_LISTEN_ALL_NSID
;
2247 static int netlink_getsockopt(struct socket
*sock
, int level
, int optname
,
2248 char __user
*optval
, int __user
*optlen
)
2250 struct sock
*sk
= sock
->sk
;
2251 struct netlink_sock
*nlk
= nlk_sk(sk
);
2254 if (level
!= SOL_NETLINK
)
2255 return -ENOPROTOOPT
;
2257 if (get_user(len
, optlen
))
2263 case NETLINK_PKTINFO
:
2264 if (len
< sizeof(int))
2267 val
= nlk
->flags
& NETLINK_F_RECV_PKTINFO
? 1 : 0;
2268 if (put_user(len
, optlen
) ||
2269 put_user(val
, optval
))
2273 case NETLINK_BROADCAST_ERROR
:
2274 if (len
< sizeof(int))
2277 val
= nlk
->flags
& NETLINK_F_BROADCAST_SEND_ERROR
? 1 : 0;
2278 if (put_user(len
, optlen
) ||
2279 put_user(val
, optval
))
2283 case NETLINK_NO_ENOBUFS
:
2284 if (len
< sizeof(int))
2287 val
= nlk
->flags
& NETLINK_F_RECV_NO_ENOBUFS
? 1 : 0;
2288 if (put_user(len
, optlen
) ||
2289 put_user(val
, optval
))
2293 case NETLINK_LIST_MEMBERSHIPS
: {
2294 int pos
, idx
, shift
;
2297 netlink_table_grab();
2298 for (pos
= 0; pos
* 8 < nlk
->ngroups
; pos
+= sizeof(u32
)) {
2299 if (len
- pos
< sizeof(u32
))
2302 idx
= pos
/ sizeof(unsigned long);
2303 shift
= (pos
% sizeof(unsigned long)) * 8;
2304 if (put_user((u32
)(nlk
->groups
[idx
] >> shift
),
2305 (u32 __user
*)(optval
+ pos
))) {
2310 if (put_user(ALIGN(nlk
->ngroups
/ 8, sizeof(u32
)), optlen
))
2312 netlink_table_ungrab();
2321 static void netlink_cmsg_recv_pktinfo(struct msghdr
*msg
, struct sk_buff
*skb
)
2323 struct nl_pktinfo info
;
2325 info
.group
= NETLINK_CB(skb
).dst_group
;
2326 put_cmsg(msg
, SOL_NETLINK
, NETLINK_PKTINFO
, sizeof(info
), &info
);
2329 static void netlink_cmsg_listen_all_nsid(struct sock
*sk
, struct msghdr
*msg
,
2330 struct sk_buff
*skb
)
2332 if (!NETLINK_CB(skb
).nsid_is_set
)
2335 put_cmsg(msg
, SOL_NETLINK
, NETLINK_LISTEN_ALL_NSID
, sizeof(int),
2336 &NETLINK_CB(skb
).nsid
);
2339 static int netlink_sendmsg(struct socket
*sock
, struct msghdr
*msg
, size_t len
)
2341 struct sock
*sk
= sock
->sk
;
2342 struct netlink_sock
*nlk
= nlk_sk(sk
);
2343 DECLARE_SOCKADDR(struct sockaddr_nl
*, addr
, msg
->msg_name
);
2346 struct sk_buff
*skb
;
2348 struct scm_cookie scm
;
2349 u32 netlink_skb_flags
= 0;
2351 if (msg
->msg_flags
&MSG_OOB
)
2354 err
= scm_send(sock
, msg
, &scm
, true);
2358 if (msg
->msg_namelen
) {
2360 if (addr
->nl_family
!= AF_NETLINK
)
2362 dst_portid
= addr
->nl_pid
;
2363 dst_group
= ffs(addr
->nl_groups
);
2365 if ((dst_group
|| dst_portid
) &&
2366 !netlink_allowed(sock
, NL_CFG_F_NONROOT_SEND
))
2368 netlink_skb_flags
|= NETLINK_SKB_DST
;
2370 dst_portid
= nlk
->dst_portid
;
2371 dst_group
= nlk
->dst_group
;
2375 err
= netlink_autobind(sock
);
2380 /* It's a really convoluted way for userland to ask for mmaped
2381 * sendmsg(), but that's what we've got...
2383 if (netlink_tx_is_mmaped(sk
) &&
2384 msg
->msg_iter
.type
== ITER_IOVEC
&&
2385 msg
->msg_iter
.nr_segs
== 1 &&
2386 msg
->msg_iter
.iov
->iov_base
== NULL
) {
2387 err
= netlink_mmap_sendmsg(sk
, msg
, dst_portid
, dst_group
,
2393 if (len
> sk
->sk_sndbuf
- 32)
2396 skb
= netlink_alloc_large_skb(len
, dst_group
);
2400 NETLINK_CB(skb
).portid
= nlk
->portid
;
2401 NETLINK_CB(skb
).dst_group
= dst_group
;
2402 NETLINK_CB(skb
).creds
= scm
.creds
;
2403 NETLINK_CB(skb
).flags
= netlink_skb_flags
;
2406 if (memcpy_from_msg(skb_put(skb
, len
), msg
, len
)) {
2411 err
= security_netlink_send(sk
, skb
);
2418 atomic_inc(&skb
->users
);
2419 netlink_broadcast(sk
, skb
, dst_portid
, dst_group
, GFP_KERNEL
);
2421 err
= netlink_unicast(sk
, skb
, dst_portid
, msg
->msg_flags
&MSG_DONTWAIT
);
2428 static int netlink_recvmsg(struct socket
*sock
, struct msghdr
*msg
, size_t len
,
2431 struct scm_cookie scm
;
2432 struct sock
*sk
= sock
->sk
;
2433 struct netlink_sock
*nlk
= nlk_sk(sk
);
2434 int noblock
= flags
&MSG_DONTWAIT
;
2436 struct sk_buff
*skb
, *data_skb
;
2444 skb
= skb_recv_datagram(sk
, flags
, noblock
, &err
);
2450 #ifdef CONFIG_COMPAT_NETLINK_MESSAGES
2451 if (unlikely(skb_shinfo(skb
)->frag_list
)) {
2453 * If this skb has a frag_list, then here that means that we
2454 * will have to use the frag_list skb's data for compat tasks
2455 * and the regular skb's data for normal (non-compat) tasks.
2457 * If we need to send the compat skb, assign it to the
2458 * 'data_skb' variable so that it will be used below for data
2459 * copying. We keep 'skb' for everything else, including
2460 * freeing both later.
2462 if (flags
& MSG_CMSG_COMPAT
)
2463 data_skb
= skb_shinfo(skb
)->frag_list
;
2467 /* Record the max length of recvmsg() calls for future allocations */
2468 nlk
->max_recvmsg_len
= max(nlk
->max_recvmsg_len
, len
);
2469 nlk
->max_recvmsg_len
= min_t(size_t, nlk
->max_recvmsg_len
,
2472 copied
= data_skb
->len
;
2474 msg
->msg_flags
|= MSG_TRUNC
;
2478 skb_reset_transport_header(data_skb
);
2479 err
= skb_copy_datagram_msg(data_skb
, 0, msg
, copied
);
2481 if (msg
->msg_name
) {
2482 DECLARE_SOCKADDR(struct sockaddr_nl
*, addr
, msg
->msg_name
);
2483 addr
->nl_family
= AF_NETLINK
;
2485 addr
->nl_pid
= NETLINK_CB(skb
).portid
;
2486 addr
->nl_groups
= netlink_group_mask(NETLINK_CB(skb
).dst_group
);
2487 msg
->msg_namelen
= sizeof(*addr
);
2490 if (nlk
->flags
& NETLINK_F_RECV_PKTINFO
)
2491 netlink_cmsg_recv_pktinfo(msg
, skb
);
2492 if (nlk
->flags
& NETLINK_F_LISTEN_ALL_NSID
)
2493 netlink_cmsg_listen_all_nsid(sk
, msg
, skb
);
2495 memset(&scm
, 0, sizeof(scm
));
2496 scm
.creds
= *NETLINK_CREDS(skb
);
2497 if (flags
& MSG_TRUNC
)
2498 copied
= data_skb
->len
;
2500 skb_free_datagram(sk
, skb
);
2502 if (nlk
->cb_running
&&
2503 atomic_read(&sk
->sk_rmem_alloc
) <= sk
->sk_rcvbuf
/ 2) {
2504 ret
= netlink_dump(sk
);
2507 sk
->sk_error_report(sk
);
2511 scm_recv(sock
, msg
, &scm
, flags
);
2513 netlink_rcv_wake(sk
);
2514 return err
? : copied
;
2517 static void netlink_data_ready(struct sock
*sk
)
2523 * We export these functions to other modules. They provide a
2524 * complete set of kernel non-blocking support for message
2529 __netlink_kernel_create(struct net
*net
, int unit
, struct module
*module
,
2530 struct netlink_kernel_cfg
*cfg
)
2532 struct socket
*sock
;
2534 struct netlink_sock
*nlk
;
2535 struct listeners
*listeners
= NULL
;
2536 struct mutex
*cb_mutex
= cfg
? cfg
->cb_mutex
: NULL
;
2537 unsigned int groups
;
2541 if (unit
< 0 || unit
>= MAX_LINKS
)
2544 if (sock_create_lite(PF_NETLINK
, SOCK_DGRAM
, unit
, &sock
))
2547 if (__netlink_create(net
, sock
, cb_mutex
, unit
, 1) < 0)
2548 goto out_sock_release_nosk
;
2552 if (!cfg
|| cfg
->groups
< 32)
2555 groups
= cfg
->groups
;
2557 listeners
= kzalloc(sizeof(*listeners
) + NLGRPSZ(groups
), GFP_KERNEL
);
2559 goto out_sock_release
;
2561 sk
->sk_data_ready
= netlink_data_ready
;
2562 if (cfg
&& cfg
->input
)
2563 nlk_sk(sk
)->netlink_rcv
= cfg
->input
;
2565 if (netlink_insert(sk
, 0))
2566 goto out_sock_release
;
2569 nlk
->flags
|= NETLINK_F_KERNEL_SOCKET
;
2571 netlink_table_grab();
2572 if (!nl_table
[unit
].registered
) {
2573 nl_table
[unit
].groups
= groups
;
2574 rcu_assign_pointer(nl_table
[unit
].listeners
, listeners
);
2575 nl_table
[unit
].cb_mutex
= cb_mutex
;
2576 nl_table
[unit
].module
= module
;
2578 nl_table
[unit
].bind
= cfg
->bind
;
2579 nl_table
[unit
].unbind
= cfg
->unbind
;
2580 nl_table
[unit
].flags
= cfg
->flags
;
2582 nl_table
[unit
].compare
= cfg
->compare
;
2584 nl_table
[unit
].registered
= 1;
2587 nl_table
[unit
].registered
++;
2589 netlink_table_ungrab();
2594 netlink_kernel_release(sk
);
2597 out_sock_release_nosk
:
2601 EXPORT_SYMBOL(__netlink_kernel_create
);
2604 netlink_kernel_release(struct sock
*sk
)
2606 if (sk
== NULL
|| sk
->sk_socket
== NULL
)
2609 sock_release(sk
->sk_socket
);
2611 EXPORT_SYMBOL(netlink_kernel_release
);
2613 int __netlink_change_ngroups(struct sock
*sk
, unsigned int groups
)
2615 struct listeners
*new, *old
;
2616 struct netlink_table
*tbl
= &nl_table
[sk
->sk_protocol
];
2621 if (NLGRPSZ(tbl
->groups
) < NLGRPSZ(groups
)) {
2622 new = kzalloc(sizeof(*new) + NLGRPSZ(groups
), GFP_ATOMIC
);
2625 old
= nl_deref_protected(tbl
->listeners
);
2626 memcpy(new->masks
, old
->masks
, NLGRPSZ(tbl
->groups
));
2627 rcu_assign_pointer(tbl
->listeners
, new);
2629 kfree_rcu(old
, rcu
);
2631 tbl
->groups
= groups
;
2637 * netlink_change_ngroups - change number of multicast groups
2639 * This changes the number of multicast groups that are available
2640 * on a certain netlink family. Note that it is not possible to
2641 * change the number of groups to below 32. Also note that it does
2642 * not implicitly call netlink_clear_multicast_users() when the
2643 * number of groups is reduced.
2645 * @sk: The kernel netlink socket, as returned by netlink_kernel_create().
2646 * @groups: The new number of groups.
2648 int netlink_change_ngroups(struct sock
*sk
, unsigned int groups
)
2652 netlink_table_grab();
2653 err
= __netlink_change_ngroups(sk
, groups
);
2654 netlink_table_ungrab();
2659 void __netlink_clear_multicast_users(struct sock
*ksk
, unsigned int group
)
2662 struct netlink_table
*tbl
= &nl_table
[ksk
->sk_protocol
];
2664 sk_for_each_bound(sk
, &tbl
->mc_list
)
2665 netlink_update_socket_mc(nlk_sk(sk
), group
, 0);
2669 __nlmsg_put(struct sk_buff
*skb
, u32 portid
, u32 seq
, int type
, int len
, int flags
)
2671 struct nlmsghdr
*nlh
;
2672 int size
= nlmsg_msg_size(len
);
2674 nlh
= (struct nlmsghdr
*)skb_put(skb
, NLMSG_ALIGN(size
));
2675 nlh
->nlmsg_type
= type
;
2676 nlh
->nlmsg_len
= size
;
2677 nlh
->nlmsg_flags
= flags
;
2678 nlh
->nlmsg_pid
= portid
;
2679 nlh
->nlmsg_seq
= seq
;
2680 if (!__builtin_constant_p(size
) || NLMSG_ALIGN(size
) - size
!= 0)
2681 memset(nlmsg_data(nlh
) + len
, 0, NLMSG_ALIGN(size
) - size
);
2684 EXPORT_SYMBOL(__nlmsg_put
);
2687 * It looks a bit ugly.
2688 * It would be better to create kernel thread.
2691 static int netlink_dump(struct sock
*sk
)
2693 struct netlink_sock
*nlk
= nlk_sk(sk
);
2694 struct netlink_callback
*cb
;
2695 struct sk_buff
*skb
= NULL
;
2696 struct nlmsghdr
*nlh
;
2697 int len
, err
= -ENOBUFS
;
2700 mutex_lock(nlk
->cb_mutex
);
2701 if (!nlk
->cb_running
) {
2707 alloc_size
= max_t(int, cb
->min_dump_alloc
, NLMSG_GOODSIZE
);
2709 if (!netlink_rx_is_mmaped(sk
) &&
2710 atomic_read(&sk
->sk_rmem_alloc
) >= sk
->sk_rcvbuf
)
2713 /* NLMSG_GOODSIZE is small to avoid high order allocations being
2714 * required, but it makes sense to _attempt_ a 16K bytes allocation
2715 * to reduce number of system calls on dump operations, if user
2716 * ever provided a big enough buffer.
2718 if (alloc_size
< nlk
->max_recvmsg_len
) {
2719 skb
= netlink_alloc_skb(sk
,
2720 nlk
->max_recvmsg_len
,
2725 /* available room should be exact amount to avoid MSG_TRUNC */
2727 skb_reserve(skb
, skb_tailroom(skb
) -
2728 nlk
->max_recvmsg_len
);
2731 skb
= netlink_alloc_skb(sk
, alloc_size
, nlk
->portid
,
2735 netlink_skb_set_owner_r(skb
, sk
);
2737 len
= cb
->dump(skb
, cb
);
2740 mutex_unlock(nlk
->cb_mutex
);
2742 if (sk_filter(sk
, skb
))
2745 __netlink_sendskb(sk
, skb
);
2749 nlh
= nlmsg_put_answer(skb
, cb
, NLMSG_DONE
, sizeof(len
), NLM_F_MULTI
);
2753 nl_dump_check_consistent(cb
, nlh
);
2755 memcpy(nlmsg_data(nlh
), &len
, sizeof(len
));
2757 if (sk_filter(sk
, skb
))
2760 __netlink_sendskb(sk
, skb
);
2765 nlk
->cb_running
= false;
2766 mutex_unlock(nlk
->cb_mutex
);
2767 module_put(cb
->module
);
2768 consume_skb(cb
->skb
);
2772 mutex_unlock(nlk
->cb_mutex
);
2777 int __netlink_dump_start(struct sock
*ssk
, struct sk_buff
*skb
,
2778 const struct nlmsghdr
*nlh
,
2779 struct netlink_dump_control
*control
)
2781 struct netlink_callback
*cb
;
2783 struct netlink_sock
*nlk
;
2786 /* Memory mapped dump requests need to be copied to avoid looping
2787 * on the pending state in netlink_mmap_sendmsg() while the CB hold
2788 * a reference to the skb.
2790 if (netlink_skb_is_mmaped(skb
)) {
2791 skb
= skb_copy(skb
, GFP_KERNEL
);
2795 atomic_inc(&skb
->users
);
2797 sk
= netlink_lookup(sock_net(ssk
), ssk
->sk_protocol
, NETLINK_CB(skb
).portid
);
2799 ret
= -ECONNREFUSED
;
2804 mutex_lock(nlk
->cb_mutex
);
2805 /* A dump is in progress... */
2806 if (nlk
->cb_running
) {
2810 /* add reference of module which cb->dump belongs to */
2811 if (!try_module_get(control
->module
)) {
2812 ret
= -EPROTONOSUPPORT
;
2817 memset(cb
, 0, sizeof(*cb
));
2818 cb
->dump
= control
->dump
;
2819 cb
->done
= control
->done
;
2821 cb
->data
= control
->data
;
2822 cb
->module
= control
->module
;
2823 cb
->min_dump_alloc
= control
->min_dump_alloc
;
2826 nlk
->cb_running
= true;
2828 mutex_unlock(nlk
->cb_mutex
);
2830 ret
= netlink_dump(sk
);
2836 /* We successfully started a dump, by returning -EINTR we
2837 * signal not to send ACK even if it was requested.
2843 mutex_unlock(nlk
->cb_mutex
);
2848 EXPORT_SYMBOL(__netlink_dump_start
);
2850 void netlink_ack(struct sk_buff
*in_skb
, struct nlmsghdr
*nlh
, int err
)
2852 struct sk_buff
*skb
;
2853 struct nlmsghdr
*rep
;
2854 struct nlmsgerr
*errmsg
;
2855 size_t payload
= sizeof(*errmsg
);
2857 /* error messages get the original request appened */
2859 payload
+= nlmsg_len(nlh
);
2861 skb
= netlink_alloc_skb(in_skb
->sk
, nlmsg_total_size(payload
),
2862 NETLINK_CB(in_skb
).portid
, GFP_KERNEL
);
2866 sk
= netlink_lookup(sock_net(in_skb
->sk
),
2867 in_skb
->sk
->sk_protocol
,
2868 NETLINK_CB(in_skb
).portid
);
2870 sk
->sk_err
= ENOBUFS
;
2871 sk
->sk_error_report(sk
);
2877 rep
= __nlmsg_put(skb
, NETLINK_CB(in_skb
).portid
, nlh
->nlmsg_seq
,
2878 NLMSG_ERROR
, payload
, 0);
2879 errmsg
= nlmsg_data(rep
);
2880 errmsg
->error
= err
;
2881 memcpy(&errmsg
->msg
, nlh
, err
? nlh
->nlmsg_len
: sizeof(*nlh
));
2882 netlink_unicast(in_skb
->sk
, skb
, NETLINK_CB(in_skb
).portid
, MSG_DONTWAIT
);
2884 EXPORT_SYMBOL(netlink_ack
);
2886 int netlink_rcv_skb(struct sk_buff
*skb
, int (*cb
)(struct sk_buff
*,
2889 struct nlmsghdr
*nlh
;
2892 while (skb
->len
>= nlmsg_total_size(0)) {
2895 nlh
= nlmsg_hdr(skb
);
2898 if (nlh
->nlmsg_len
< NLMSG_HDRLEN
|| skb
->len
< nlh
->nlmsg_len
)
2901 /* Only requests are handled by the kernel */
2902 if (!(nlh
->nlmsg_flags
& NLM_F_REQUEST
))
2905 /* Skip control messages */
2906 if (nlh
->nlmsg_type
< NLMSG_MIN_TYPE
)
2914 if (nlh
->nlmsg_flags
& NLM_F_ACK
|| err
)
2915 netlink_ack(skb
, nlh
, err
);
2918 msglen
= NLMSG_ALIGN(nlh
->nlmsg_len
);
2919 if (msglen
> skb
->len
)
2921 skb_pull(skb
, msglen
);
2926 EXPORT_SYMBOL(netlink_rcv_skb
);
2929 * nlmsg_notify - send a notification netlink message
2930 * @sk: netlink socket to use
2931 * @skb: notification message
2932 * @portid: destination netlink portid for reports or 0
2933 * @group: destination multicast group or 0
2934 * @report: 1 to report back, 0 to disable
2935 * @flags: allocation flags
2937 int nlmsg_notify(struct sock
*sk
, struct sk_buff
*skb
, u32 portid
,
2938 unsigned int group
, int report
, gfp_t flags
)
2943 int exclude_portid
= 0;
2946 atomic_inc(&skb
->users
);
2947 exclude_portid
= portid
;
2950 /* errors reported via destination sk->sk_err, but propagate
2951 * delivery errors if NETLINK_BROADCAST_ERROR flag is set */
2952 err
= nlmsg_multicast(sk
, skb
, exclude_portid
, group
, flags
);
2958 err2
= nlmsg_unicast(sk
, skb
, portid
);
2959 if (!err
|| err
== -ESRCH
)
2965 EXPORT_SYMBOL(nlmsg_notify
);
2967 #ifdef CONFIG_PROC_FS
2968 struct nl_seq_iter
{
2969 struct seq_net_private p
;
2970 struct rhashtable_iter hti
;
2974 static int netlink_walk_start(struct nl_seq_iter
*iter
)
2978 err
= rhashtable_walk_init(&nl_table
[iter
->link
].hash
, &iter
->hti
);
2980 iter
->link
= MAX_LINKS
;
2984 err
= rhashtable_walk_start(&iter
->hti
);
2985 return err
== -EAGAIN
? 0 : err
;
2988 static void netlink_walk_stop(struct nl_seq_iter
*iter
)
2990 rhashtable_walk_stop(&iter
->hti
);
2991 rhashtable_walk_exit(&iter
->hti
);
2994 static void *__netlink_seq_next(struct seq_file
*seq
)
2996 struct nl_seq_iter
*iter
= seq
->private;
2997 struct netlink_sock
*nlk
;
3003 nlk
= rhashtable_walk_next(&iter
->hti
);
3006 if (PTR_ERR(nlk
) == -EAGAIN
)
3015 netlink_walk_stop(iter
);
3016 if (++iter
->link
>= MAX_LINKS
)
3019 err
= netlink_walk_start(iter
);
3021 return ERR_PTR(err
);
3023 } while (sock_net(&nlk
->sk
) != seq_file_net(seq
));
3028 static void *netlink_seq_start(struct seq_file
*seq
, loff_t
*posp
)
3030 struct nl_seq_iter
*iter
= seq
->private;
3031 void *obj
= SEQ_START_TOKEN
;
3037 err
= netlink_walk_start(iter
);
3039 return ERR_PTR(err
);
3041 for (pos
= *posp
; pos
&& obj
&& !IS_ERR(obj
); pos
--)
3042 obj
= __netlink_seq_next(seq
);
3047 static void *netlink_seq_next(struct seq_file
*seq
, void *v
, loff_t
*pos
)
3050 return __netlink_seq_next(seq
);
3053 static void netlink_seq_stop(struct seq_file
*seq
, void *v
)
3055 struct nl_seq_iter
*iter
= seq
->private;
3057 if (iter
->link
>= MAX_LINKS
)
3060 netlink_walk_stop(iter
);
3064 static int netlink_seq_show(struct seq_file
*seq
, void *v
)
3066 if (v
== SEQ_START_TOKEN
) {
3068 "sk Eth Pid Groups "
3069 "Rmem Wmem Dump Locks Drops Inode\n");
3072 struct netlink_sock
*nlk
= nlk_sk(s
);
3074 seq_printf(seq
, "%pK %-3d %-6u %08x %-8d %-8d %d %-8d %-8d %-8lu\n",
3078 nlk
->groups
? (u32
)nlk
->groups
[0] : 0,
3079 sk_rmem_alloc_get(s
),
3080 sk_wmem_alloc_get(s
),
3082 atomic_read(&s
->sk_refcnt
),
3083 atomic_read(&s
->sk_drops
),
3091 static const struct seq_operations netlink_seq_ops
= {
3092 .start
= netlink_seq_start
,
3093 .next
= netlink_seq_next
,
3094 .stop
= netlink_seq_stop
,
3095 .show
= netlink_seq_show
,
3099 static int netlink_seq_open(struct inode
*inode
, struct file
*file
)
3101 return seq_open_net(inode
, file
, &netlink_seq_ops
,
3102 sizeof(struct nl_seq_iter
));
3105 static const struct file_operations netlink_seq_fops
= {
3106 .owner
= THIS_MODULE
,
3107 .open
= netlink_seq_open
,
3109 .llseek
= seq_lseek
,
3110 .release
= seq_release_net
,
3115 int netlink_register_notifier(struct notifier_block
*nb
)
3117 return atomic_notifier_chain_register(&netlink_chain
, nb
);
3119 EXPORT_SYMBOL(netlink_register_notifier
);
3121 int netlink_unregister_notifier(struct notifier_block
*nb
)
3123 return atomic_notifier_chain_unregister(&netlink_chain
, nb
);
3125 EXPORT_SYMBOL(netlink_unregister_notifier
);
3127 static const struct proto_ops netlink_ops
= {
3128 .family
= PF_NETLINK
,
3129 .owner
= THIS_MODULE
,
3130 .release
= netlink_release
,
3131 .bind
= netlink_bind
,
3132 .connect
= netlink_connect
,
3133 .socketpair
= sock_no_socketpair
,
3134 .accept
= sock_no_accept
,
3135 .getname
= netlink_getname
,
3136 .poll
= netlink_poll
,
3137 .ioctl
= sock_no_ioctl
,
3138 .listen
= sock_no_listen
,
3139 .shutdown
= sock_no_shutdown
,
3140 .setsockopt
= netlink_setsockopt
,
3141 .getsockopt
= netlink_getsockopt
,
3142 .sendmsg
= netlink_sendmsg
,
3143 .recvmsg
= netlink_recvmsg
,
3144 .mmap
= netlink_mmap
,
3145 .sendpage
= sock_no_sendpage
,
3148 static const struct net_proto_family netlink_family_ops
= {
3149 .family
= PF_NETLINK
,
3150 .create
= netlink_create
,
3151 .owner
= THIS_MODULE
, /* for consistency 8) */
3154 static int __net_init
netlink_net_init(struct net
*net
)
3156 #ifdef CONFIG_PROC_FS
3157 if (!proc_create("netlink", 0, net
->proc_net
, &netlink_seq_fops
))
3163 static void __net_exit
netlink_net_exit(struct net
*net
)
3165 #ifdef CONFIG_PROC_FS
3166 remove_proc_entry("netlink", net
->proc_net
);
3170 static void __init
netlink_add_usersock_entry(void)
3172 struct listeners
*listeners
;
3175 listeners
= kzalloc(sizeof(*listeners
) + NLGRPSZ(groups
), GFP_KERNEL
);
3177 panic("netlink_add_usersock_entry: Cannot allocate listeners\n");
3179 netlink_table_grab();
3181 nl_table
[NETLINK_USERSOCK
].groups
= groups
;
3182 rcu_assign_pointer(nl_table
[NETLINK_USERSOCK
].listeners
, listeners
);
3183 nl_table
[NETLINK_USERSOCK
].module
= THIS_MODULE
;
3184 nl_table
[NETLINK_USERSOCK
].registered
= 1;
3185 nl_table
[NETLINK_USERSOCK
].flags
= NL_CFG_F_NONROOT_SEND
;
3187 netlink_table_ungrab();
3190 static struct pernet_operations __net_initdata netlink_net_ops
= {
3191 .init
= netlink_net_init
,
3192 .exit
= netlink_net_exit
,
3195 static inline u32
netlink_hash(const void *data
, u32 len
, u32 seed
)
3197 const struct netlink_sock
*nlk
= data
;
3198 struct netlink_compare_arg arg
;
3200 netlink_compare_arg_init(&arg
, sock_net(&nlk
->sk
), nlk
->portid
);
3201 return jhash2((u32
*)&arg
, netlink_compare_arg_len
/ sizeof(u32
), seed
);
3204 static const struct rhashtable_params netlink_rhashtable_params
= {
3205 .head_offset
= offsetof(struct netlink_sock
, node
),
3206 .key_len
= netlink_compare_arg_len
,
3207 .obj_hashfn
= netlink_hash
,
3208 .obj_cmpfn
= netlink_compare
,
3209 .automatic_shrinking
= true,
3212 static int __init
netlink_proto_init(void)
3215 int err
= proto_register(&netlink_proto
, 0);
3220 BUILD_BUG_ON(sizeof(struct netlink_skb_parms
) > FIELD_SIZEOF(struct sk_buff
, cb
));
3222 nl_table
= kcalloc(MAX_LINKS
, sizeof(*nl_table
), GFP_KERNEL
);
3226 for (i
= 0; i
< MAX_LINKS
; i
++) {
3227 if (rhashtable_init(&nl_table
[i
].hash
,
3228 &netlink_rhashtable_params
) < 0) {
3230 rhashtable_destroy(&nl_table
[i
].hash
);
3236 INIT_LIST_HEAD(&netlink_tap_all
);
3238 netlink_add_usersock_entry();
3240 sock_register(&netlink_family_ops
);
3241 register_pernet_subsys(&netlink_net_ops
);
3242 /* The netlink device handler may be needed early. */
3247 panic("netlink_init: Cannot allocate nl_table\n");
3250 core_initcall(netlink_proto_init
);