1 // SPDX-License-Identifier: GPL-2.0-or-later
3 * NETLINK Kernel-user communication protocol.
5 * Authors: Alan Cox <alan@lxorguk.ukuu.org.uk>
6 * Alexey Kuznetsov <kuznet@ms2.inr.ac.ru>
7 * Patrick McHardy <kaber@trash.net>
9 * Tue Jun 26 14:36:48 MEST 2001 Herbert "herp" Rosmanith
10 * added netlink_proto_exit
11 * Tue Jan 22 18:32:44 BRST 2002 Arnaldo C. de Melo <acme@conectiva.com.br>
12 * use nlk_sk, as sk->protinfo is on a diet 8)
13 * Fri Jul 22 19:51:12 MEST 2005 Harald Welte <laforge@gnumonks.org>
14 * - inc module use count of module that owns
15 * the kernel socket in case userspace opens
16 * socket of same protocol
17 * - remove all module support, since netlink is
18 * mandatory if CONFIG_NET=y these days
21 #include <linux/module.h>
23 #include <linux/capability.h>
24 #include <linux/kernel.h>
25 #include <linux/init.h>
26 #include <linux/signal.h>
27 #include <linux/sched.h>
28 #include <linux/errno.h>
29 #include <linux/string.h>
30 #include <linux/stat.h>
31 #include <linux/socket.h>
33 #include <linux/fcntl.h>
34 #include <linux/termios.h>
35 #include <linux/sockios.h>
36 #include <linux/net.h>
38 #include <linux/slab.h>
39 #include <linux/uaccess.h>
40 #include <linux/skbuff.h>
41 #include <linux/netdevice.h>
42 #include <linux/rtnetlink.h>
43 #include <linux/proc_fs.h>
44 #include <linux/seq_file.h>
45 #include <linux/notifier.h>
46 #include <linux/security.h>
47 #include <linux/jhash.h>
48 #include <linux/jiffies.h>
49 #include <linux/random.h>
50 #include <linux/bitops.h>
52 #include <linux/types.h>
53 #include <linux/audit.h>
54 #include <linux/mutex.h>
55 #include <linux/vmalloc.h>
56 #include <linux/if_arp.h>
57 #include <linux/rhashtable.h>
58 #include <asm/cacheflush.h>
59 #include <linux/hash.h>
60 #include <linux/genetlink.h>
61 #include <linux/net_namespace.h>
62 #include <linux/nospec.h>
63 #include <linux/btf_ids.h>
65 #include <net/net_namespace.h>
66 #include <net/netns/generic.h>
69 #include <net/netlink.h>
70 #define CREATE_TRACE_POINTS
71 #include <trace/events/netlink.h>
73 #include "af_netlink.h"
77 unsigned long masks
[];
81 #define NETLINK_S_CONGESTED 0x0
83 static inline int netlink_is_kernel(struct sock
*sk
)
85 return nlk_sk(sk
)->flags
& NETLINK_F_KERNEL_SOCKET
;
88 struct netlink_table
*nl_table __read_mostly
;
89 EXPORT_SYMBOL_GPL(nl_table
);
91 static DECLARE_WAIT_QUEUE_HEAD(nl_table_wait
);
93 static struct lock_class_key nlk_cb_mutex_keys
[MAX_LINKS
];
95 static const char *const nlk_cb_mutex_key_strings
[MAX_LINKS
+ 1] = {
98 "nlk_cb_mutex-USERSOCK",
99 "nlk_cb_mutex-FIREWALL",
100 "nlk_cb_mutex-SOCK_DIAG",
101 "nlk_cb_mutex-NFLOG",
103 "nlk_cb_mutex-SELINUX",
104 "nlk_cb_mutex-ISCSI",
105 "nlk_cb_mutex-AUDIT",
106 "nlk_cb_mutex-FIB_LOOKUP",
107 "nlk_cb_mutex-CONNECTOR",
108 "nlk_cb_mutex-NETFILTER",
109 "nlk_cb_mutex-IP6_FW",
110 "nlk_cb_mutex-DNRTMSG",
111 "nlk_cb_mutex-KOBJECT_UEVENT",
112 "nlk_cb_mutex-GENERIC",
114 "nlk_cb_mutex-SCSITRANSPORT",
115 "nlk_cb_mutex-ECRYPTFS",
117 "nlk_cb_mutex-CRYPTO",
128 "nlk_cb_mutex-MAX_LINKS"
131 static int netlink_dump(struct sock
*sk
);
133 /* nl_table locking explained:
134 * Lookup and traversal are protected with an RCU read-side lock. Insertion
135 * and removal are protected with per bucket lock while using RCU list
136 * modification primitives and may run in parallel to RCU protected lookups.
137 * Destruction of the Netlink socket may only occur *after* nl_table_lock has
138 * been acquired * either during or after the socket has been removed from
139 * the list and after an RCU grace period.
141 DEFINE_RWLOCK(nl_table_lock
);
142 EXPORT_SYMBOL_GPL(nl_table_lock
);
143 static atomic_t nl_table_users
= ATOMIC_INIT(0);
145 #define nl_deref_protected(X) rcu_dereference_protected(X, lockdep_is_held(&nl_table_lock));
147 static BLOCKING_NOTIFIER_HEAD(netlink_chain
);
150 static const struct rhashtable_params netlink_rhashtable_params
;
152 void do_trace_netlink_extack(const char *msg
)
154 trace_netlink_extack(msg
);
156 EXPORT_SYMBOL(do_trace_netlink_extack
);
158 static inline u32
netlink_group_mask(u32 group
)
162 return group
? 1 << (group
- 1) : 0;
165 static struct sk_buff
*netlink_to_full_skb(const struct sk_buff
*skb
,
168 unsigned int len
= skb_end_offset(skb
);
171 new = alloc_skb(len
, gfp_mask
);
175 NETLINK_CB(new).portid
= NETLINK_CB(skb
).portid
;
176 NETLINK_CB(new).dst_group
= NETLINK_CB(skb
).dst_group
;
177 NETLINK_CB(new).creds
= NETLINK_CB(skb
).creds
;
179 skb_put_data(new, skb
->data
, len
);
183 static unsigned int netlink_tap_net_id
;
185 struct netlink_tap_net
{
186 struct list_head netlink_tap_all
;
187 struct mutex netlink_tap_lock
;
190 int netlink_add_tap(struct netlink_tap
*nt
)
192 struct net
*net
= dev_net(nt
->dev
);
193 struct netlink_tap_net
*nn
= net_generic(net
, netlink_tap_net_id
);
195 if (unlikely(nt
->dev
->type
!= ARPHRD_NETLINK
))
198 mutex_lock(&nn
->netlink_tap_lock
);
199 list_add_rcu(&nt
->list
, &nn
->netlink_tap_all
);
200 mutex_unlock(&nn
->netlink_tap_lock
);
202 __module_get(nt
->module
);
206 EXPORT_SYMBOL_GPL(netlink_add_tap
);
208 static int __netlink_remove_tap(struct netlink_tap
*nt
)
210 struct net
*net
= dev_net(nt
->dev
);
211 struct netlink_tap_net
*nn
= net_generic(net
, netlink_tap_net_id
);
213 struct netlink_tap
*tmp
;
215 mutex_lock(&nn
->netlink_tap_lock
);
217 list_for_each_entry(tmp
, &nn
->netlink_tap_all
, list
) {
219 list_del_rcu(&nt
->list
);
225 pr_warn("__netlink_remove_tap: %p not found\n", nt
);
227 mutex_unlock(&nn
->netlink_tap_lock
);
230 module_put(nt
->module
);
232 return found
? 0 : -ENODEV
;
235 int netlink_remove_tap(struct netlink_tap
*nt
)
239 ret
= __netlink_remove_tap(nt
);
244 EXPORT_SYMBOL_GPL(netlink_remove_tap
);
246 static __net_init
int netlink_tap_init_net(struct net
*net
)
248 struct netlink_tap_net
*nn
= net_generic(net
, netlink_tap_net_id
);
250 INIT_LIST_HEAD(&nn
->netlink_tap_all
);
251 mutex_init(&nn
->netlink_tap_lock
);
255 static struct pernet_operations netlink_tap_net_ops
= {
256 .init
= netlink_tap_init_net
,
257 .id
= &netlink_tap_net_id
,
258 .size
= sizeof(struct netlink_tap_net
),
261 static bool netlink_filter_tap(const struct sk_buff
*skb
)
263 struct sock
*sk
= skb
->sk
;
265 /* We take the more conservative approach and
266 * whitelist socket protocols that may pass.
268 switch (sk
->sk_protocol
) {
270 case NETLINK_USERSOCK
:
271 case NETLINK_SOCK_DIAG
:
274 case NETLINK_FIB_LOOKUP
:
275 case NETLINK_NETFILTER
:
276 case NETLINK_GENERIC
:
283 static int __netlink_deliver_tap_skb(struct sk_buff
*skb
,
284 struct net_device
*dev
)
286 struct sk_buff
*nskb
;
287 struct sock
*sk
= skb
->sk
;
290 if (!net_eq(dev_net(dev
), sock_net(sk
)))
295 if (is_vmalloc_addr(skb
->head
))
296 nskb
= netlink_to_full_skb(skb
, GFP_ATOMIC
);
298 nskb
= skb_clone(skb
, GFP_ATOMIC
);
301 nskb
->protocol
= htons((u16
) sk
->sk_protocol
);
302 nskb
->pkt_type
= netlink_is_kernel(sk
) ?
303 PACKET_KERNEL
: PACKET_USER
;
304 skb_reset_network_header(nskb
);
305 ret
= dev_queue_xmit(nskb
);
306 if (unlikely(ret
> 0))
307 ret
= net_xmit_errno(ret
);
314 static void __netlink_deliver_tap(struct sk_buff
*skb
, struct netlink_tap_net
*nn
)
317 struct netlink_tap
*tmp
;
319 if (!netlink_filter_tap(skb
))
322 list_for_each_entry_rcu(tmp
, &nn
->netlink_tap_all
, list
) {
323 ret
= __netlink_deliver_tap_skb(skb
, tmp
->dev
);
329 static void netlink_deliver_tap(struct net
*net
, struct sk_buff
*skb
)
331 struct netlink_tap_net
*nn
= net_generic(net
, netlink_tap_net_id
);
335 if (unlikely(!list_empty(&nn
->netlink_tap_all
)))
336 __netlink_deliver_tap(skb
, nn
);
341 static void netlink_deliver_tap_kernel(struct sock
*dst
, struct sock
*src
,
344 if (!(netlink_is_kernel(dst
) && netlink_is_kernel(src
)))
345 netlink_deliver_tap(sock_net(dst
), skb
);
348 static void netlink_overrun(struct sock
*sk
)
350 struct netlink_sock
*nlk
= nlk_sk(sk
);
352 if (!(nlk
->flags
& NETLINK_F_RECV_NO_ENOBUFS
)) {
353 if (!test_and_set_bit(NETLINK_S_CONGESTED
,
354 &nlk_sk(sk
)->state
)) {
355 sk
->sk_err
= ENOBUFS
;
359 atomic_inc(&sk
->sk_drops
);
362 static void netlink_rcv_wake(struct sock
*sk
)
364 struct netlink_sock
*nlk
= nlk_sk(sk
);
366 if (skb_queue_empty_lockless(&sk
->sk_receive_queue
))
367 clear_bit(NETLINK_S_CONGESTED
, &nlk
->state
);
368 if (!test_bit(NETLINK_S_CONGESTED
, &nlk
->state
))
369 wake_up_interruptible(&nlk
->wait
);
372 static void netlink_skb_destructor(struct sk_buff
*skb
)
374 if (is_vmalloc_addr(skb
->head
)) {
376 !atomic_dec_return(&(skb_shinfo(skb
)->dataref
)))
385 static void netlink_skb_set_owner_r(struct sk_buff
*skb
, struct sock
*sk
)
387 WARN_ON(skb
->sk
!= NULL
);
389 skb
->destructor
= netlink_skb_destructor
;
390 atomic_add(skb
->truesize
, &sk
->sk_rmem_alloc
);
391 sk_mem_charge(sk
, skb
->truesize
);
394 static void netlink_sock_destruct(struct sock
*sk
)
396 struct netlink_sock
*nlk
= nlk_sk(sk
);
398 if (nlk
->cb_running
) {
400 nlk
->cb
.done(&nlk
->cb
);
401 module_put(nlk
->cb
.module
);
402 kfree_skb(nlk
->cb
.skb
);
405 skb_queue_purge(&sk
->sk_receive_queue
);
407 if (!sock_flag(sk
, SOCK_DEAD
)) {
408 printk(KERN_ERR
"Freeing alive netlink socket %p\n", sk
);
412 WARN_ON(atomic_read(&sk
->sk_rmem_alloc
));
413 WARN_ON(refcount_read(&sk
->sk_wmem_alloc
));
414 WARN_ON(nlk_sk(sk
)->groups
);
417 static void netlink_sock_destruct_work(struct work_struct
*work
)
419 struct netlink_sock
*nlk
= container_of(work
, struct netlink_sock
,
425 /* This lock without WQ_FLAG_EXCLUSIVE is good on UP and it is _very_ bad on
426 * SMP. Look, when several writers sleep and reader wakes them up, all but one
427 * immediately hit write lock and grab all the cpus. Exclusive sleep solves
428 * this, _but_ remember, it adds useless work on UP machines.
431 void netlink_table_grab(void)
432 __acquires(nl_table_lock
)
436 write_lock_irq(&nl_table_lock
);
438 if (atomic_read(&nl_table_users
)) {
439 DECLARE_WAITQUEUE(wait
, current
);
441 add_wait_queue_exclusive(&nl_table_wait
, &wait
);
443 set_current_state(TASK_UNINTERRUPTIBLE
);
444 if (atomic_read(&nl_table_users
) == 0)
446 write_unlock_irq(&nl_table_lock
);
448 write_lock_irq(&nl_table_lock
);
451 __set_current_state(TASK_RUNNING
);
452 remove_wait_queue(&nl_table_wait
, &wait
);
456 void netlink_table_ungrab(void)
457 __releases(nl_table_lock
)
459 write_unlock_irq(&nl_table_lock
);
460 wake_up(&nl_table_wait
);
464 netlink_lock_table(void)
468 /* read_lock() synchronizes us to netlink_table_grab */
470 read_lock_irqsave(&nl_table_lock
, flags
);
471 atomic_inc(&nl_table_users
);
472 read_unlock_irqrestore(&nl_table_lock
, flags
);
476 netlink_unlock_table(void)
478 if (atomic_dec_and_test(&nl_table_users
))
479 wake_up(&nl_table_wait
);
482 struct netlink_compare_arg
488 /* Doing sizeof directly may yield 4 extra bytes on 64-bit. */
489 #define netlink_compare_arg_len \
490 (offsetof(struct netlink_compare_arg, portid) + sizeof(u32))
492 static inline int netlink_compare(struct rhashtable_compare_arg
*arg
,
495 const struct netlink_compare_arg
*x
= arg
->key
;
496 const struct netlink_sock
*nlk
= ptr
;
498 return nlk
->portid
!= x
->portid
||
499 !net_eq(sock_net(&nlk
->sk
), read_pnet(&x
->pnet
));
502 static void netlink_compare_arg_init(struct netlink_compare_arg
*arg
,
503 struct net
*net
, u32 portid
)
505 memset(arg
, 0, sizeof(*arg
));
506 write_pnet(&arg
->pnet
, net
);
507 arg
->portid
= portid
;
510 static struct sock
*__netlink_lookup(struct netlink_table
*table
, u32 portid
,
513 struct netlink_compare_arg arg
;
515 netlink_compare_arg_init(&arg
, net
, portid
);
516 return rhashtable_lookup_fast(&table
->hash
, &arg
,
517 netlink_rhashtable_params
);
520 static int __netlink_insert(struct netlink_table
*table
, struct sock
*sk
)
522 struct netlink_compare_arg arg
;
524 netlink_compare_arg_init(&arg
, sock_net(sk
), nlk_sk(sk
)->portid
);
525 return rhashtable_lookup_insert_key(&table
->hash
, &arg
,
527 netlink_rhashtable_params
);
530 static struct sock
*netlink_lookup(struct net
*net
, int protocol
, u32 portid
)
532 struct netlink_table
*table
= &nl_table
[protocol
];
536 sk
= __netlink_lookup(table
, portid
, net
);
544 static const struct proto_ops netlink_ops
;
547 netlink_update_listeners(struct sock
*sk
)
549 struct netlink_table
*tbl
= &nl_table
[sk
->sk_protocol
];
552 struct listeners
*listeners
;
554 listeners
= nl_deref_protected(tbl
->listeners
);
558 for (i
= 0; i
< NLGRPLONGS(tbl
->groups
); i
++) {
560 sk_for_each_bound(sk
, &tbl
->mc_list
) {
561 if (i
< NLGRPLONGS(nlk_sk(sk
)->ngroups
))
562 mask
|= nlk_sk(sk
)->groups
[i
];
564 listeners
->masks
[i
] = mask
;
566 /* this function is only called with the netlink table "grabbed", which
567 * makes sure updates are visible before bind or setsockopt return. */
570 static int netlink_insert(struct sock
*sk
, u32 portid
)
572 struct netlink_table
*table
= &nl_table
[sk
->sk_protocol
];
577 err
= nlk_sk(sk
)->portid
== portid
? 0 : -EBUSY
;
578 if (nlk_sk(sk
)->bound
)
581 nlk_sk(sk
)->portid
= portid
;
584 err
= __netlink_insert(table
, sk
);
586 /* In case the hashtable backend returns with -EBUSY
587 * from here, it must not escape to the caller.
589 if (unlikely(err
== -EBUSY
))
597 /* We need to ensure that the socket is hashed and visible. */
599 /* Paired with lockless reads from netlink_bind(),
600 * netlink_connect() and netlink_sendmsg().
602 WRITE_ONCE(nlk_sk(sk
)->bound
, portid
);
609 static void netlink_remove(struct sock
*sk
)
611 struct netlink_table
*table
;
613 table
= &nl_table
[sk
->sk_protocol
];
614 if (!rhashtable_remove_fast(&table
->hash
, &nlk_sk(sk
)->node
,
615 netlink_rhashtable_params
)) {
616 WARN_ON(refcount_read(&sk
->sk_refcnt
) == 1);
620 netlink_table_grab();
621 if (nlk_sk(sk
)->subscriptions
) {
622 __sk_del_bind_node(sk
);
623 netlink_update_listeners(sk
);
625 if (sk
->sk_protocol
== NETLINK_GENERIC
)
626 atomic_inc(&genl_sk_destructing_cnt
);
627 netlink_table_ungrab();
630 static struct proto netlink_proto
= {
632 .owner
= THIS_MODULE
,
633 .obj_size
= sizeof(struct netlink_sock
),
636 static int __netlink_create(struct net
*net
, struct socket
*sock
,
637 struct mutex
*cb_mutex
, int protocol
,
641 struct netlink_sock
*nlk
;
643 sock
->ops
= &netlink_ops
;
645 sk
= sk_alloc(net
, PF_NETLINK
, GFP_KERNEL
, &netlink_proto
, kern
);
649 sock_init_data(sock
, sk
);
653 nlk
->cb_mutex
= cb_mutex
;
655 nlk
->cb_mutex
= &nlk
->cb_def_mutex
;
656 mutex_init(nlk
->cb_mutex
);
657 lockdep_set_class_and_name(nlk
->cb_mutex
,
658 nlk_cb_mutex_keys
+ protocol
,
659 nlk_cb_mutex_key_strings
[protocol
]);
661 init_waitqueue_head(&nlk
->wait
);
663 sk
->sk_destruct
= netlink_sock_destruct
;
664 sk
->sk_protocol
= protocol
;
668 static int netlink_create(struct net
*net
, struct socket
*sock
, int protocol
,
671 struct module
*module
= NULL
;
672 struct mutex
*cb_mutex
;
673 struct netlink_sock
*nlk
;
674 int (*bind
)(struct net
*net
, int group
);
675 void (*unbind
)(struct net
*net
, int group
);
678 sock
->state
= SS_UNCONNECTED
;
680 if (sock
->type
!= SOCK_RAW
&& sock
->type
!= SOCK_DGRAM
)
681 return -ESOCKTNOSUPPORT
;
683 if (protocol
< 0 || protocol
>= MAX_LINKS
)
684 return -EPROTONOSUPPORT
;
685 protocol
= array_index_nospec(protocol
, MAX_LINKS
);
687 netlink_lock_table();
688 #ifdef CONFIG_MODULES
689 if (!nl_table
[protocol
].registered
) {
690 netlink_unlock_table();
691 request_module("net-pf-%d-proto-%d", PF_NETLINK
, protocol
);
692 netlink_lock_table();
695 if (nl_table
[protocol
].registered
&&
696 try_module_get(nl_table
[protocol
].module
))
697 module
= nl_table
[protocol
].module
;
699 err
= -EPROTONOSUPPORT
;
700 cb_mutex
= nl_table
[protocol
].cb_mutex
;
701 bind
= nl_table
[protocol
].bind
;
702 unbind
= nl_table
[protocol
].unbind
;
703 netlink_unlock_table();
708 err
= __netlink_create(net
, sock
, cb_mutex
, protocol
, kern
);
713 sock_prot_inuse_add(net
, &netlink_proto
, 1);
716 nlk
= nlk_sk(sock
->sk
);
717 nlk
->module
= module
;
718 nlk
->netlink_bind
= bind
;
719 nlk
->netlink_unbind
= unbind
;
728 static void deferred_put_nlk_sk(struct rcu_head
*head
)
730 struct netlink_sock
*nlk
= container_of(head
, struct netlink_sock
, rcu
);
731 struct sock
*sk
= &nlk
->sk
;
736 if (!refcount_dec_and_test(&sk
->sk_refcnt
))
739 if (nlk
->cb_running
&& nlk
->cb
.done
) {
740 INIT_WORK(&nlk
->work
, netlink_sock_destruct_work
);
741 schedule_work(&nlk
->work
);
748 static int netlink_release(struct socket
*sock
)
750 struct sock
*sk
= sock
->sk
;
751 struct netlink_sock
*nlk
;
761 * OK. Socket is unlinked, any packets that arrive now
765 /* must not acquire netlink_table_lock in any way again before unbind
766 * and notifying genetlink is done as otherwise it might deadlock
768 if (nlk
->netlink_unbind
) {
771 for (i
= 0; i
< nlk
->ngroups
; i
++)
772 if (test_bit(i
, nlk
->groups
))
773 nlk
->netlink_unbind(sock_net(sk
), i
+ 1);
775 if (sk
->sk_protocol
== NETLINK_GENERIC
&&
776 atomic_dec_return(&genl_sk_destructing_cnt
) == 0)
777 wake_up(&genl_sk_destructing_waitq
);
780 wake_up_interruptible_all(&nlk
->wait
);
782 skb_queue_purge(&sk
->sk_write_queue
);
784 if (nlk
->portid
&& nlk
->bound
) {
785 struct netlink_notify n
= {
787 .protocol
= sk
->sk_protocol
,
788 .portid
= nlk
->portid
,
790 blocking_notifier_call_chain(&netlink_chain
,
791 NETLINK_URELEASE
, &n
);
794 module_put(nlk
->module
);
796 if (netlink_is_kernel(sk
)) {
797 netlink_table_grab();
798 BUG_ON(nl_table
[sk
->sk_protocol
].registered
== 0);
799 if (--nl_table
[sk
->sk_protocol
].registered
== 0) {
800 struct listeners
*old
;
802 old
= nl_deref_protected(nl_table
[sk
->sk_protocol
].listeners
);
803 RCU_INIT_POINTER(nl_table
[sk
->sk_protocol
].listeners
, NULL
);
805 nl_table
[sk
->sk_protocol
].module
= NULL
;
806 nl_table
[sk
->sk_protocol
].bind
= NULL
;
807 nl_table
[sk
->sk_protocol
].unbind
= NULL
;
808 nl_table
[sk
->sk_protocol
].flags
= 0;
809 nl_table
[sk
->sk_protocol
].registered
= 0;
811 netlink_table_ungrab();
815 sock_prot_inuse_add(sock_net(sk
), &netlink_proto
, -1);
817 call_rcu(&nlk
->rcu
, deferred_put_nlk_sk
);
821 static int netlink_autobind(struct socket
*sock
)
823 struct sock
*sk
= sock
->sk
;
824 struct net
*net
= sock_net(sk
);
825 struct netlink_table
*table
= &nl_table
[sk
->sk_protocol
];
826 s32 portid
= task_tgid_vnr(current
);
834 ok
= !__netlink_lookup(table
, portid
, net
);
837 /* Bind collision, search negative portid values. */
839 /* rover will be in range [S32_MIN, -4097] */
840 rover
= S32_MIN
+ prandom_u32_max(-4096 - S32_MIN
);
841 else if (rover
>= -4096)
847 err
= netlink_insert(sk
, portid
);
848 if (err
== -EADDRINUSE
)
851 /* If 2 threads race to autobind, that is fine. */
859 * __netlink_ns_capable - General netlink message capability test
860 * @nsp: NETLINK_CB of the socket buffer holding a netlink command from userspace.
861 * @user_ns: The user namespace of the capability to use
862 * @cap: The capability to use
864 * Test to see if the opener of the socket we received the message
865 * from had when the netlink socket was created and the sender of the
866 * message has the capability @cap in the user namespace @user_ns.
868 bool __netlink_ns_capable(const struct netlink_skb_parms
*nsp
,
869 struct user_namespace
*user_ns
, int cap
)
871 return ((nsp
->flags
& NETLINK_SKB_DST
) ||
872 file_ns_capable(nsp
->sk
->sk_socket
->file
, user_ns
, cap
)) &&
873 ns_capable(user_ns
, cap
);
875 EXPORT_SYMBOL(__netlink_ns_capable
);
878 * netlink_ns_capable - General netlink message capability test
879 * @skb: socket buffer holding a netlink command from userspace
880 * @user_ns: The user namespace of the capability to use
881 * @cap: The capability to use
883 * Test to see if the opener of the socket we received the message
884 * from had when the netlink socket was created and the sender of the
885 * message has the capability @cap in the user namespace @user_ns.
887 bool netlink_ns_capable(const struct sk_buff
*skb
,
888 struct user_namespace
*user_ns
, int cap
)
890 return __netlink_ns_capable(&NETLINK_CB(skb
), user_ns
, cap
);
892 EXPORT_SYMBOL(netlink_ns_capable
);
895 * netlink_capable - Netlink global message capability test
896 * @skb: socket buffer holding a netlink command from userspace
897 * @cap: The capability to use
899 * Test to see if the opener of the socket we received the message
900 * from had when the netlink socket was created and the sender of the
901 * message has the capability @cap in all user namespaces.
903 bool netlink_capable(const struct sk_buff
*skb
, int cap
)
905 return netlink_ns_capable(skb
, &init_user_ns
, cap
);
907 EXPORT_SYMBOL(netlink_capable
);
910 * netlink_net_capable - Netlink network namespace message capability test
911 * @skb: socket buffer holding a netlink command from userspace
912 * @cap: The capability to use
914 * Test to see if the opener of the socket we received the message
915 * from had when the netlink socket was created and the sender of the
916 * message has the capability @cap over the network namespace of
917 * the socket we received the message from.
919 bool netlink_net_capable(const struct sk_buff
*skb
, int cap
)
921 return netlink_ns_capable(skb
, sock_net(skb
->sk
)->user_ns
, cap
);
923 EXPORT_SYMBOL(netlink_net_capable
);
925 static inline int netlink_allowed(const struct socket
*sock
, unsigned int flag
)
927 return (nl_table
[sock
->sk
->sk_protocol
].flags
& flag
) ||
928 ns_capable(sock_net(sock
->sk
)->user_ns
, CAP_NET_ADMIN
);
932 netlink_update_subscriptions(struct sock
*sk
, unsigned int subscriptions
)
934 struct netlink_sock
*nlk
= nlk_sk(sk
);
936 if (nlk
->subscriptions
&& !subscriptions
)
937 __sk_del_bind_node(sk
);
938 else if (!nlk
->subscriptions
&& subscriptions
)
939 sk_add_bind_node(sk
, &nl_table
[sk
->sk_protocol
].mc_list
);
940 nlk
->subscriptions
= subscriptions
;
943 static int netlink_realloc_groups(struct sock
*sk
)
945 struct netlink_sock
*nlk
= nlk_sk(sk
);
947 unsigned long *new_groups
;
950 netlink_table_grab();
952 groups
= nl_table
[sk
->sk_protocol
].groups
;
953 if (!nl_table
[sk
->sk_protocol
].registered
) {
958 if (nlk
->ngroups
>= groups
)
961 new_groups
= krealloc(nlk
->groups
, NLGRPSZ(groups
), GFP_ATOMIC
);
962 if (new_groups
== NULL
) {
966 memset((char *)new_groups
+ NLGRPSZ(nlk
->ngroups
), 0,
967 NLGRPSZ(groups
) - NLGRPSZ(nlk
->ngroups
));
969 nlk
->groups
= new_groups
;
970 nlk
->ngroups
= groups
;
972 netlink_table_ungrab();
976 static void netlink_undo_bind(int group
, long unsigned int groups
,
979 struct netlink_sock
*nlk
= nlk_sk(sk
);
982 if (!nlk
->netlink_unbind
)
985 for (undo
= 0; undo
< group
; undo
++)
986 if (test_bit(undo
, &groups
))
987 nlk
->netlink_unbind(sock_net(sk
), undo
+ 1);
990 static int netlink_bind(struct socket
*sock
, struct sockaddr
*addr
,
993 struct sock
*sk
= sock
->sk
;
994 struct net
*net
= sock_net(sk
);
995 struct netlink_sock
*nlk
= nlk_sk(sk
);
996 struct sockaddr_nl
*nladdr
= (struct sockaddr_nl
*)addr
;
998 unsigned long groups
;
1001 if (addr_len
< sizeof(struct sockaddr_nl
))
1004 if (nladdr
->nl_family
!= AF_NETLINK
)
1006 groups
= nladdr
->nl_groups
;
1008 /* Only superuser is allowed to listen multicasts */
1010 if (!netlink_allowed(sock
, NL_CFG_F_NONROOT_RECV
))
1012 err
= netlink_realloc_groups(sk
);
1017 if (nlk
->ngroups
< BITS_PER_LONG
)
1018 groups
&= (1UL << nlk
->ngroups
) - 1;
1020 /* Paired with WRITE_ONCE() in netlink_insert() */
1021 bound
= READ_ONCE(nlk
->bound
);
1023 /* Ensure nlk->portid is up-to-date. */
1026 if (nladdr
->nl_pid
!= nlk
->portid
)
1030 if (nlk
->netlink_bind
&& groups
) {
1033 /* nl_groups is a u32, so cap the maximum groups we can bind */
1034 for (group
= 0; group
< BITS_PER_TYPE(u32
); group
++) {
1035 if (!test_bit(group
, &groups
))
1037 err
= nlk
->netlink_bind(net
, group
+ 1);
1040 netlink_undo_bind(group
, groups
, sk
);
1045 /* No need for barriers here as we return to user-space without
1046 * using any of the bound attributes.
1048 netlink_lock_table();
1050 err
= nladdr
->nl_pid
?
1051 netlink_insert(sk
, nladdr
->nl_pid
) :
1052 netlink_autobind(sock
);
1054 netlink_undo_bind(BITS_PER_TYPE(u32
), groups
, sk
);
1059 if (!groups
&& (nlk
->groups
== NULL
|| !(u32
)nlk
->groups
[0]))
1061 netlink_unlock_table();
1063 netlink_table_grab();
1064 netlink_update_subscriptions(sk
, nlk
->subscriptions
+
1066 hweight32(nlk
->groups
[0]));
1067 nlk
->groups
[0] = (nlk
->groups
[0] & ~0xffffffffUL
) | groups
;
1068 netlink_update_listeners(sk
);
1069 netlink_table_ungrab();
1074 netlink_unlock_table();
1078 static int netlink_connect(struct socket
*sock
, struct sockaddr
*addr
,
1079 int alen
, int flags
)
1082 struct sock
*sk
= sock
->sk
;
1083 struct netlink_sock
*nlk
= nlk_sk(sk
);
1084 struct sockaddr_nl
*nladdr
= (struct sockaddr_nl
*)addr
;
1086 if (alen
< sizeof(addr
->sa_family
))
1089 if (addr
->sa_family
== AF_UNSPEC
) {
1090 sk
->sk_state
= NETLINK_UNCONNECTED
;
1091 nlk
->dst_portid
= 0;
1095 if (addr
->sa_family
!= AF_NETLINK
)
1098 if (alen
< sizeof(struct sockaddr_nl
))
1101 if ((nladdr
->nl_groups
|| nladdr
->nl_pid
) &&
1102 !netlink_allowed(sock
, NL_CFG_F_NONROOT_SEND
))
1105 /* No need for barriers here as we return to user-space without
1106 * using any of the bound attributes.
1107 * Paired with WRITE_ONCE() in netlink_insert().
1109 if (!READ_ONCE(nlk
->bound
))
1110 err
= netlink_autobind(sock
);
1113 sk
->sk_state
= NETLINK_CONNECTED
;
1114 nlk
->dst_portid
= nladdr
->nl_pid
;
1115 nlk
->dst_group
= ffs(nladdr
->nl_groups
);
1121 static int netlink_getname(struct socket
*sock
, struct sockaddr
*addr
,
1124 struct sock
*sk
= sock
->sk
;
1125 struct netlink_sock
*nlk
= nlk_sk(sk
);
1126 DECLARE_SOCKADDR(struct sockaddr_nl
*, nladdr
, addr
);
1128 nladdr
->nl_family
= AF_NETLINK
;
1132 nladdr
->nl_pid
= nlk
->dst_portid
;
1133 nladdr
->nl_groups
= netlink_group_mask(nlk
->dst_group
);
1135 nladdr
->nl_pid
= nlk
->portid
;
1136 netlink_lock_table();
1137 nladdr
->nl_groups
= nlk
->groups
? nlk
->groups
[0] : 0;
1138 netlink_unlock_table();
1140 return sizeof(*nladdr
);
1143 static int netlink_ioctl(struct socket
*sock
, unsigned int cmd
,
1146 /* try to hand this ioctl down to the NIC drivers.
1148 return -ENOIOCTLCMD
;
1151 static struct sock
*netlink_getsockbyportid(struct sock
*ssk
, u32 portid
)
1154 struct netlink_sock
*nlk
;
1156 sock
= netlink_lookup(sock_net(ssk
), ssk
->sk_protocol
, portid
);
1158 return ERR_PTR(-ECONNREFUSED
);
1160 /* Don't bother queuing skb if kernel socket has no input function */
1162 if (sock
->sk_state
== NETLINK_CONNECTED
&&
1163 nlk
->dst_portid
!= nlk_sk(ssk
)->portid
) {
1165 return ERR_PTR(-ECONNREFUSED
);
1170 struct sock
*netlink_getsockbyfilp(struct file
*filp
)
1172 struct inode
*inode
= file_inode(filp
);
1175 if (!S_ISSOCK(inode
->i_mode
))
1176 return ERR_PTR(-ENOTSOCK
);
1178 sock
= SOCKET_I(inode
)->sk
;
1179 if (sock
->sk_family
!= AF_NETLINK
)
1180 return ERR_PTR(-EINVAL
);
1186 static struct sk_buff
*netlink_alloc_large_skb(unsigned int size
,
1189 struct sk_buff
*skb
;
1192 if (size
<= NLMSG_GOODSIZE
|| broadcast
)
1193 return alloc_skb(size
, GFP_KERNEL
);
1195 size
= SKB_DATA_ALIGN(size
) +
1196 SKB_DATA_ALIGN(sizeof(struct skb_shared_info
));
1198 data
= vmalloc(size
);
1202 skb
= __build_skb(data
, size
);
1206 skb
->destructor
= netlink_skb_destructor
;
1212 * Attach a skb to a netlink socket.
1213 * The caller must hold a reference to the destination socket. On error, the
1214 * reference is dropped. The skb is not send to the destination, just all
1215 * all error checks are performed and memory in the queue is reserved.
1217 * < 0: error. skb freed, reference to sock dropped.
1219 * 1: repeat lookup - reference dropped while waiting for socket memory.
1221 int netlink_attachskb(struct sock
*sk
, struct sk_buff
*skb
,
1222 long *timeo
, struct sock
*ssk
)
1224 struct netlink_sock
*nlk
;
1228 if ((atomic_read(&sk
->sk_rmem_alloc
) > sk
->sk_rcvbuf
||
1229 test_bit(NETLINK_S_CONGESTED
, &nlk
->state
))) {
1230 DECLARE_WAITQUEUE(wait
, current
);
1232 if (!ssk
|| netlink_is_kernel(ssk
))
1233 netlink_overrun(sk
);
1239 __set_current_state(TASK_INTERRUPTIBLE
);
1240 add_wait_queue(&nlk
->wait
, &wait
);
1242 if ((atomic_read(&sk
->sk_rmem_alloc
) > sk
->sk_rcvbuf
||
1243 test_bit(NETLINK_S_CONGESTED
, &nlk
->state
)) &&
1244 !sock_flag(sk
, SOCK_DEAD
))
1245 *timeo
= schedule_timeout(*timeo
);
1247 __set_current_state(TASK_RUNNING
);
1248 remove_wait_queue(&nlk
->wait
, &wait
);
1251 if (signal_pending(current
)) {
1253 return sock_intr_errno(*timeo
);
1257 netlink_skb_set_owner_r(skb
, sk
);
1261 static int __netlink_sendskb(struct sock
*sk
, struct sk_buff
*skb
)
1265 netlink_deliver_tap(sock_net(sk
), skb
);
1267 skb_queue_tail(&sk
->sk_receive_queue
, skb
);
1268 sk
->sk_data_ready(sk
);
1272 int netlink_sendskb(struct sock
*sk
, struct sk_buff
*skb
)
1274 int len
= __netlink_sendskb(sk
, skb
);
1280 void netlink_detachskb(struct sock
*sk
, struct sk_buff
*skb
)
1286 static struct sk_buff
*netlink_trim(struct sk_buff
*skb
, gfp_t allocation
)
1290 WARN_ON(skb
->sk
!= NULL
);
1291 delta
= skb
->end
- skb
->tail
;
1292 if (is_vmalloc_addr(skb
->head
) || delta
* 2 < skb
->truesize
)
1295 if (skb_shared(skb
)) {
1296 struct sk_buff
*nskb
= skb_clone(skb
, allocation
);
1303 pskb_expand_head(skb
, 0, -delta
,
1304 (allocation
& ~__GFP_DIRECT_RECLAIM
) |
1305 __GFP_NOWARN
| __GFP_NORETRY
);
1309 static int netlink_unicast_kernel(struct sock
*sk
, struct sk_buff
*skb
,
1313 struct netlink_sock
*nlk
= nlk_sk(sk
);
1315 ret
= -ECONNREFUSED
;
1316 if (nlk
->netlink_rcv
!= NULL
) {
1318 netlink_skb_set_owner_r(skb
, sk
);
1319 NETLINK_CB(skb
).sk
= ssk
;
1320 netlink_deliver_tap_kernel(sk
, ssk
, skb
);
1321 nlk
->netlink_rcv(skb
);
1330 int netlink_unicast(struct sock
*ssk
, struct sk_buff
*skb
,
1331 u32 portid
, int nonblock
)
1337 skb
= netlink_trim(skb
, gfp_any());
1339 timeo
= sock_sndtimeo(ssk
, nonblock
);
1341 sk
= netlink_getsockbyportid(ssk
, portid
);
1346 if (netlink_is_kernel(sk
))
1347 return netlink_unicast_kernel(sk
, skb
, ssk
);
1349 if (sk_filter(sk
, skb
)) {
1356 err
= netlink_attachskb(sk
, skb
, &timeo
, ssk
);
1362 return netlink_sendskb(sk
, skb
);
1364 EXPORT_SYMBOL(netlink_unicast
);
1366 int netlink_has_listeners(struct sock
*sk
, unsigned int group
)
1369 struct listeners
*listeners
;
1371 BUG_ON(!netlink_is_kernel(sk
));
1374 listeners
= rcu_dereference(nl_table
[sk
->sk_protocol
].listeners
);
1376 if (listeners
&& group
- 1 < nl_table
[sk
->sk_protocol
].groups
)
1377 res
= test_bit(group
- 1, listeners
->masks
);
1383 EXPORT_SYMBOL_GPL(netlink_has_listeners
);
1385 bool netlink_strict_get_check(struct sk_buff
*skb
)
1387 const struct netlink_sock
*nlk
= nlk_sk(NETLINK_CB(skb
).sk
);
1389 return nlk
->flags
& NETLINK_F_STRICT_CHK
;
1391 EXPORT_SYMBOL_GPL(netlink_strict_get_check
);
1393 static int netlink_broadcast_deliver(struct sock
*sk
, struct sk_buff
*skb
)
1395 struct netlink_sock
*nlk
= nlk_sk(sk
);
1397 if (atomic_read(&sk
->sk_rmem_alloc
) <= sk
->sk_rcvbuf
&&
1398 !test_bit(NETLINK_S_CONGESTED
, &nlk
->state
)) {
1399 netlink_skb_set_owner_r(skb
, sk
);
1400 __netlink_sendskb(sk
, skb
);
1401 return atomic_read(&sk
->sk_rmem_alloc
) > (sk
->sk_rcvbuf
>> 1);
1406 struct netlink_broadcast_data
{
1407 struct sock
*exclude_sk
;
1412 int delivery_failure
;
1416 struct sk_buff
*skb
, *skb2
;
1417 int (*tx_filter
)(struct sock
*dsk
, struct sk_buff
*skb
, void *data
);
1421 static void do_one_broadcast(struct sock
*sk
,
1422 struct netlink_broadcast_data
*p
)
1424 struct netlink_sock
*nlk
= nlk_sk(sk
);
1427 if (p
->exclude_sk
== sk
)
1430 if (nlk
->portid
== p
->portid
|| p
->group
- 1 >= nlk
->ngroups
||
1431 !test_bit(p
->group
- 1, nlk
->groups
))
1434 if (!net_eq(sock_net(sk
), p
->net
)) {
1435 if (!(nlk
->flags
& NETLINK_F_LISTEN_ALL_NSID
))
1438 if (!peernet_has_id(sock_net(sk
), p
->net
))
1441 if (!file_ns_capable(sk
->sk_socket
->file
, p
->net
->user_ns
,
1447 netlink_overrun(sk
);
1452 if (p
->skb2
== NULL
) {
1453 if (skb_shared(p
->skb
)) {
1454 p
->skb2
= skb_clone(p
->skb
, p
->allocation
);
1456 p
->skb2
= skb_get(p
->skb
);
1458 * skb ownership may have been set when
1459 * delivered to a previous socket.
1461 skb_orphan(p
->skb2
);
1464 if (p
->skb2
== NULL
) {
1465 netlink_overrun(sk
);
1466 /* Clone failed. Notify ALL listeners. */
1468 if (nlk
->flags
& NETLINK_F_BROADCAST_SEND_ERROR
)
1469 p
->delivery_failure
= 1;
1472 if (p
->tx_filter
&& p
->tx_filter(sk
, p
->skb2
, p
->tx_data
)) {
1477 if (sk_filter(sk
, p
->skb2
)) {
1482 NETLINK_CB(p
->skb2
).nsid
= peernet2id(sock_net(sk
), p
->net
);
1483 if (NETLINK_CB(p
->skb2
).nsid
!= NETNSA_NSID_NOT_ASSIGNED
)
1484 NETLINK_CB(p
->skb2
).nsid_is_set
= true;
1485 val
= netlink_broadcast_deliver(sk
, p
->skb2
);
1487 netlink_overrun(sk
);
1488 if (nlk
->flags
& NETLINK_F_BROADCAST_SEND_ERROR
)
1489 p
->delivery_failure
= 1;
1491 p
->congested
|= val
;
1499 int netlink_broadcast_filtered(struct sock
*ssk
, struct sk_buff
*skb
, u32 portid
,
1500 u32 group
, gfp_t allocation
,
1501 int (*filter
)(struct sock
*dsk
, struct sk_buff
*skb
, void *data
),
1504 struct net
*net
= sock_net(ssk
);
1505 struct netlink_broadcast_data info
;
1508 skb
= netlink_trim(skb
, allocation
);
1510 info
.exclude_sk
= ssk
;
1512 info
.portid
= portid
;
1515 info
.delivery_failure
= 0;
1518 info
.allocation
= allocation
;
1521 info
.tx_filter
= filter
;
1522 info
.tx_data
= filter_data
;
1524 /* While we sleep in clone, do not allow to change socket list */
1526 netlink_lock_table();
1528 sk_for_each_bound(sk
, &nl_table
[ssk
->sk_protocol
].mc_list
)
1529 do_one_broadcast(sk
, &info
);
1533 netlink_unlock_table();
1535 if (info
.delivery_failure
) {
1536 kfree_skb(info
.skb2
);
1539 consume_skb(info
.skb2
);
1541 if (info
.delivered
) {
1542 if (info
.congested
&& gfpflags_allow_blocking(allocation
))
1548 EXPORT_SYMBOL(netlink_broadcast_filtered
);
1550 int netlink_broadcast(struct sock
*ssk
, struct sk_buff
*skb
, u32 portid
,
1551 u32 group
, gfp_t allocation
)
1553 return netlink_broadcast_filtered(ssk
, skb
, portid
, group
, allocation
,
1556 EXPORT_SYMBOL(netlink_broadcast
);
1558 struct netlink_set_err_data
{
1559 struct sock
*exclude_sk
;
1565 static int do_one_set_err(struct sock
*sk
, struct netlink_set_err_data
*p
)
1567 struct netlink_sock
*nlk
= nlk_sk(sk
);
1570 if (sk
== p
->exclude_sk
)
1573 if (!net_eq(sock_net(sk
), sock_net(p
->exclude_sk
)))
1576 if (nlk
->portid
== p
->portid
|| p
->group
- 1 >= nlk
->ngroups
||
1577 !test_bit(p
->group
- 1, nlk
->groups
))
1580 if (p
->code
== ENOBUFS
&& nlk
->flags
& NETLINK_F_RECV_NO_ENOBUFS
) {
1585 sk
->sk_err
= p
->code
;
1586 sk_error_report(sk
);
1592 * netlink_set_err - report error to broadcast listeners
1593 * @ssk: the kernel netlink socket, as returned by netlink_kernel_create()
1594 * @portid: the PORTID of a process that we want to skip (if any)
1595 * @group: the broadcast group that will notice the error
1596 * @code: error code, must be negative (as usual in kernelspace)
1598 * This function returns the number of broadcast listeners that have set the
1599 * NETLINK_NO_ENOBUFS socket option.
1601 int netlink_set_err(struct sock
*ssk
, u32 portid
, u32 group
, int code
)
1603 struct netlink_set_err_data info
;
1607 info
.exclude_sk
= ssk
;
1608 info
.portid
= portid
;
1610 /* sk->sk_err wants a positive error value */
1613 read_lock(&nl_table_lock
);
1615 sk_for_each_bound(sk
, &nl_table
[ssk
->sk_protocol
].mc_list
)
1616 ret
+= do_one_set_err(sk
, &info
);
1618 read_unlock(&nl_table_lock
);
1621 EXPORT_SYMBOL(netlink_set_err
);
1623 /* must be called with netlink table grabbed */
1624 static void netlink_update_socket_mc(struct netlink_sock
*nlk
,
1628 int old
, new = !!is_new
, subscriptions
;
1630 old
= test_bit(group
- 1, nlk
->groups
);
1631 subscriptions
= nlk
->subscriptions
- old
+ new;
1633 __set_bit(group
- 1, nlk
->groups
);
1635 __clear_bit(group
- 1, nlk
->groups
);
1636 netlink_update_subscriptions(&nlk
->sk
, subscriptions
);
1637 netlink_update_listeners(&nlk
->sk
);
1640 static int netlink_setsockopt(struct socket
*sock
, int level
, int optname
,
1641 sockptr_t optval
, unsigned int optlen
)
1643 struct sock
*sk
= sock
->sk
;
1644 struct netlink_sock
*nlk
= nlk_sk(sk
);
1645 unsigned int val
= 0;
1648 if (level
!= SOL_NETLINK
)
1649 return -ENOPROTOOPT
;
1651 if (optlen
>= sizeof(int) &&
1652 copy_from_sockptr(&val
, optval
, sizeof(val
)))
1656 case NETLINK_PKTINFO
:
1658 nlk
->flags
|= NETLINK_F_RECV_PKTINFO
;
1660 nlk
->flags
&= ~NETLINK_F_RECV_PKTINFO
;
1663 case NETLINK_ADD_MEMBERSHIP
:
1664 case NETLINK_DROP_MEMBERSHIP
: {
1665 if (!netlink_allowed(sock
, NL_CFG_F_NONROOT_RECV
))
1667 err
= netlink_realloc_groups(sk
);
1670 if (!val
|| val
- 1 >= nlk
->ngroups
)
1672 if (optname
== NETLINK_ADD_MEMBERSHIP
&& nlk
->netlink_bind
) {
1673 err
= nlk
->netlink_bind(sock_net(sk
), val
);
1677 netlink_table_grab();
1678 netlink_update_socket_mc(nlk
, val
,
1679 optname
== NETLINK_ADD_MEMBERSHIP
);
1680 netlink_table_ungrab();
1681 if (optname
== NETLINK_DROP_MEMBERSHIP
&& nlk
->netlink_unbind
)
1682 nlk
->netlink_unbind(sock_net(sk
), val
);
1687 case NETLINK_BROADCAST_ERROR
:
1689 nlk
->flags
|= NETLINK_F_BROADCAST_SEND_ERROR
;
1691 nlk
->flags
&= ~NETLINK_F_BROADCAST_SEND_ERROR
;
1694 case NETLINK_NO_ENOBUFS
:
1696 nlk
->flags
|= NETLINK_F_RECV_NO_ENOBUFS
;
1697 clear_bit(NETLINK_S_CONGESTED
, &nlk
->state
);
1698 wake_up_interruptible(&nlk
->wait
);
1700 nlk
->flags
&= ~NETLINK_F_RECV_NO_ENOBUFS
;
1704 case NETLINK_LISTEN_ALL_NSID
:
1705 if (!ns_capable(sock_net(sk
)->user_ns
, CAP_NET_BROADCAST
))
1709 nlk
->flags
|= NETLINK_F_LISTEN_ALL_NSID
;
1711 nlk
->flags
&= ~NETLINK_F_LISTEN_ALL_NSID
;
1714 case NETLINK_CAP_ACK
:
1716 nlk
->flags
|= NETLINK_F_CAP_ACK
;
1718 nlk
->flags
&= ~NETLINK_F_CAP_ACK
;
1721 case NETLINK_EXT_ACK
:
1723 nlk
->flags
|= NETLINK_F_EXT_ACK
;
1725 nlk
->flags
&= ~NETLINK_F_EXT_ACK
;
1728 case NETLINK_GET_STRICT_CHK
:
1730 nlk
->flags
|= NETLINK_F_STRICT_CHK
;
1732 nlk
->flags
&= ~NETLINK_F_STRICT_CHK
;
1741 static int netlink_getsockopt(struct socket
*sock
, int level
, int optname
,
1742 char __user
*optval
, int __user
*optlen
)
1744 struct sock
*sk
= sock
->sk
;
1745 struct netlink_sock
*nlk
= nlk_sk(sk
);
1748 if (level
!= SOL_NETLINK
)
1749 return -ENOPROTOOPT
;
1751 if (get_user(len
, optlen
))
1757 case NETLINK_PKTINFO
:
1758 if (len
< sizeof(int))
1761 val
= nlk
->flags
& NETLINK_F_RECV_PKTINFO
? 1 : 0;
1762 if (put_user(len
, optlen
) ||
1763 put_user(val
, optval
))
1767 case NETLINK_BROADCAST_ERROR
:
1768 if (len
< sizeof(int))
1771 val
= nlk
->flags
& NETLINK_F_BROADCAST_SEND_ERROR
? 1 : 0;
1772 if (put_user(len
, optlen
) ||
1773 put_user(val
, optval
))
1777 case NETLINK_NO_ENOBUFS
:
1778 if (len
< sizeof(int))
1781 val
= nlk
->flags
& NETLINK_F_RECV_NO_ENOBUFS
? 1 : 0;
1782 if (put_user(len
, optlen
) ||
1783 put_user(val
, optval
))
1787 case NETLINK_LIST_MEMBERSHIPS
: {
1788 int pos
, idx
, shift
;
1791 netlink_lock_table();
1792 for (pos
= 0; pos
* 8 < nlk
->ngroups
; pos
+= sizeof(u32
)) {
1793 if (len
- pos
< sizeof(u32
))
1796 idx
= pos
/ sizeof(unsigned long);
1797 shift
= (pos
% sizeof(unsigned long)) * 8;
1798 if (put_user((u32
)(nlk
->groups
[idx
] >> shift
),
1799 (u32 __user
*)(optval
+ pos
))) {
1804 if (put_user(ALIGN(nlk
->ngroups
/ 8, sizeof(u32
)), optlen
))
1806 netlink_unlock_table();
1809 case NETLINK_CAP_ACK
:
1810 if (len
< sizeof(int))
1813 val
= nlk
->flags
& NETLINK_F_CAP_ACK
? 1 : 0;
1814 if (put_user(len
, optlen
) ||
1815 put_user(val
, optval
))
1819 case NETLINK_EXT_ACK
:
1820 if (len
< sizeof(int))
1823 val
= nlk
->flags
& NETLINK_F_EXT_ACK
? 1 : 0;
1824 if (put_user(len
, optlen
) || put_user(val
, optval
))
1828 case NETLINK_GET_STRICT_CHK
:
1829 if (len
< sizeof(int))
1832 val
= nlk
->flags
& NETLINK_F_STRICT_CHK
? 1 : 0;
1833 if (put_user(len
, optlen
) || put_user(val
, optval
))
1843 static void netlink_cmsg_recv_pktinfo(struct msghdr
*msg
, struct sk_buff
*skb
)
1845 struct nl_pktinfo info
;
1847 info
.group
= NETLINK_CB(skb
).dst_group
;
1848 put_cmsg(msg
, SOL_NETLINK
, NETLINK_PKTINFO
, sizeof(info
), &info
);
1851 static void netlink_cmsg_listen_all_nsid(struct sock
*sk
, struct msghdr
*msg
,
1852 struct sk_buff
*skb
)
1854 if (!NETLINK_CB(skb
).nsid_is_set
)
1857 put_cmsg(msg
, SOL_NETLINK
, NETLINK_LISTEN_ALL_NSID
, sizeof(int),
1858 &NETLINK_CB(skb
).nsid
);
1861 static int netlink_sendmsg(struct socket
*sock
, struct msghdr
*msg
, size_t len
)
1863 struct sock
*sk
= sock
->sk
;
1864 struct netlink_sock
*nlk
= nlk_sk(sk
);
1865 DECLARE_SOCKADDR(struct sockaddr_nl
*, addr
, msg
->msg_name
);
1868 struct sk_buff
*skb
;
1870 struct scm_cookie scm
;
1871 u32 netlink_skb_flags
= 0;
1873 if (msg
->msg_flags
& MSG_OOB
)
1877 pr_warn_once("Zero length message leads to an empty skb\n");
1881 err
= scm_send(sock
, msg
, &scm
, true);
1885 if (msg
->msg_namelen
) {
1887 if (msg
->msg_namelen
< sizeof(struct sockaddr_nl
))
1889 if (addr
->nl_family
!= AF_NETLINK
)
1891 dst_portid
= addr
->nl_pid
;
1892 dst_group
= ffs(addr
->nl_groups
);
1894 if ((dst_group
|| dst_portid
) &&
1895 !netlink_allowed(sock
, NL_CFG_F_NONROOT_SEND
))
1897 netlink_skb_flags
|= NETLINK_SKB_DST
;
1899 dst_portid
= nlk
->dst_portid
;
1900 dst_group
= nlk
->dst_group
;
1903 /* Paired with WRITE_ONCE() in netlink_insert() */
1904 if (!READ_ONCE(nlk
->bound
)) {
1905 err
= netlink_autobind(sock
);
1909 /* Ensure nlk is hashed and visible. */
1914 if (len
> sk
->sk_sndbuf
- 32)
1917 skb
= netlink_alloc_large_skb(len
, dst_group
);
1921 NETLINK_CB(skb
).portid
= nlk
->portid
;
1922 NETLINK_CB(skb
).dst_group
= dst_group
;
1923 NETLINK_CB(skb
).creds
= scm
.creds
;
1924 NETLINK_CB(skb
).flags
= netlink_skb_flags
;
1927 if (memcpy_from_msg(skb_put(skb
, len
), msg
, len
)) {
1932 err
= security_netlink_send(sk
, skb
);
1939 refcount_inc(&skb
->users
);
1940 netlink_broadcast(sk
, skb
, dst_portid
, dst_group
, GFP_KERNEL
);
1942 err
= netlink_unicast(sk
, skb
, dst_portid
, msg
->msg_flags
& MSG_DONTWAIT
);
1949 static int netlink_recvmsg(struct socket
*sock
, struct msghdr
*msg
, size_t len
,
1952 struct scm_cookie scm
;
1953 struct sock
*sk
= sock
->sk
;
1954 struct netlink_sock
*nlk
= nlk_sk(sk
);
1955 int noblock
= flags
& MSG_DONTWAIT
;
1957 struct sk_buff
*skb
, *data_skb
;
1960 if (flags
& MSG_OOB
)
1965 skb
= skb_recv_datagram(sk
, flags
, noblock
, &err
);
1971 #ifdef CONFIG_COMPAT_NETLINK_MESSAGES
1972 if (unlikely(skb_shinfo(skb
)->frag_list
)) {
1974 * If this skb has a frag_list, then here that means that we
1975 * will have to use the frag_list skb's data for compat tasks
1976 * and the regular skb's data for normal (non-compat) tasks.
1978 * If we need to send the compat skb, assign it to the
1979 * 'data_skb' variable so that it will be used below for data
1980 * copying. We keep 'skb' for everything else, including
1981 * freeing both later.
1983 if (flags
& MSG_CMSG_COMPAT
)
1984 data_skb
= skb_shinfo(skb
)->frag_list
;
1988 /* Record the max length of recvmsg() calls for future allocations */
1989 nlk
->max_recvmsg_len
= max(nlk
->max_recvmsg_len
, len
);
1990 nlk
->max_recvmsg_len
= min_t(size_t, nlk
->max_recvmsg_len
,
1991 SKB_WITH_OVERHEAD(32768));
1993 copied
= data_skb
->len
;
1995 msg
->msg_flags
|= MSG_TRUNC
;
1999 skb_reset_transport_header(data_skb
);
2000 err
= skb_copy_datagram_msg(data_skb
, 0, msg
, copied
);
2002 if (msg
->msg_name
) {
2003 DECLARE_SOCKADDR(struct sockaddr_nl
*, addr
, msg
->msg_name
);
2004 addr
->nl_family
= AF_NETLINK
;
2006 addr
->nl_pid
= NETLINK_CB(skb
).portid
;
2007 addr
->nl_groups
= netlink_group_mask(NETLINK_CB(skb
).dst_group
);
2008 msg
->msg_namelen
= sizeof(*addr
);
2011 if (nlk
->flags
& NETLINK_F_RECV_PKTINFO
)
2012 netlink_cmsg_recv_pktinfo(msg
, skb
);
2013 if (nlk
->flags
& NETLINK_F_LISTEN_ALL_NSID
)
2014 netlink_cmsg_listen_all_nsid(sk
, msg
, skb
);
2016 memset(&scm
, 0, sizeof(scm
));
2017 scm
.creds
= *NETLINK_CREDS(skb
);
2018 if (flags
& MSG_TRUNC
)
2019 copied
= data_skb
->len
;
2021 skb_free_datagram(sk
, skb
);
2023 if (nlk
->cb_running
&&
2024 atomic_read(&sk
->sk_rmem_alloc
) <= sk
->sk_rcvbuf
/ 2) {
2025 ret
= netlink_dump(sk
);
2028 sk_error_report(sk
);
2032 scm_recv(sock
, msg
, &scm
, flags
);
2034 netlink_rcv_wake(sk
);
2035 return err
? : copied
;
2038 static void netlink_data_ready(struct sock
*sk
)
2044 * We export these functions to other modules. They provide a
2045 * complete set of kernel non-blocking support for message
2050 __netlink_kernel_create(struct net
*net
, int unit
, struct module
*module
,
2051 struct netlink_kernel_cfg
*cfg
)
2053 struct socket
*sock
;
2055 struct netlink_sock
*nlk
;
2056 struct listeners
*listeners
= NULL
;
2057 struct mutex
*cb_mutex
= cfg
? cfg
->cb_mutex
: NULL
;
2058 unsigned int groups
;
2062 if (unit
< 0 || unit
>= MAX_LINKS
)
2065 if (sock_create_lite(PF_NETLINK
, SOCK_DGRAM
, unit
, &sock
))
2068 if (__netlink_create(net
, sock
, cb_mutex
, unit
, 1) < 0)
2069 goto out_sock_release_nosk
;
2073 if (!cfg
|| cfg
->groups
< 32)
2076 groups
= cfg
->groups
;
2078 listeners
= kzalloc(sizeof(*listeners
) + NLGRPSZ(groups
), GFP_KERNEL
);
2080 goto out_sock_release
;
2082 sk
->sk_data_ready
= netlink_data_ready
;
2083 if (cfg
&& cfg
->input
)
2084 nlk_sk(sk
)->netlink_rcv
= cfg
->input
;
2086 if (netlink_insert(sk
, 0))
2087 goto out_sock_release
;
2090 nlk
->flags
|= NETLINK_F_KERNEL_SOCKET
;
2092 netlink_table_grab();
2093 if (!nl_table
[unit
].registered
) {
2094 nl_table
[unit
].groups
= groups
;
2095 rcu_assign_pointer(nl_table
[unit
].listeners
, listeners
);
2096 nl_table
[unit
].cb_mutex
= cb_mutex
;
2097 nl_table
[unit
].module
= module
;
2099 nl_table
[unit
].bind
= cfg
->bind
;
2100 nl_table
[unit
].unbind
= cfg
->unbind
;
2101 nl_table
[unit
].flags
= cfg
->flags
;
2103 nl_table
[unit
].compare
= cfg
->compare
;
2105 nl_table
[unit
].registered
= 1;
2108 nl_table
[unit
].registered
++;
2110 netlink_table_ungrab();
2115 netlink_kernel_release(sk
);
2118 out_sock_release_nosk
:
2122 EXPORT_SYMBOL(__netlink_kernel_create
);
2125 netlink_kernel_release(struct sock
*sk
)
2127 if (sk
== NULL
|| sk
->sk_socket
== NULL
)
2130 sock_release(sk
->sk_socket
);
2132 EXPORT_SYMBOL(netlink_kernel_release
);
2134 int __netlink_change_ngroups(struct sock
*sk
, unsigned int groups
)
2136 struct listeners
*new, *old
;
2137 struct netlink_table
*tbl
= &nl_table
[sk
->sk_protocol
];
2142 if (NLGRPSZ(tbl
->groups
) < NLGRPSZ(groups
)) {
2143 new = kzalloc(sizeof(*new) + NLGRPSZ(groups
), GFP_ATOMIC
);
2146 old
= nl_deref_protected(tbl
->listeners
);
2147 memcpy(new->masks
, old
->masks
, NLGRPSZ(tbl
->groups
));
2148 rcu_assign_pointer(tbl
->listeners
, new);
2150 kfree_rcu(old
, rcu
);
2152 tbl
->groups
= groups
;
2158 * netlink_change_ngroups - change number of multicast groups
2160 * This changes the number of multicast groups that are available
2161 * on a certain netlink family. Note that it is not possible to
2162 * change the number of groups to below 32. Also note that it does
2163 * not implicitly call netlink_clear_multicast_users() when the
2164 * number of groups is reduced.
2166 * @sk: The kernel netlink socket, as returned by netlink_kernel_create().
2167 * @groups: The new number of groups.
2169 int netlink_change_ngroups(struct sock
*sk
, unsigned int groups
)
2173 netlink_table_grab();
2174 err
= __netlink_change_ngroups(sk
, groups
);
2175 netlink_table_ungrab();
2180 void __netlink_clear_multicast_users(struct sock
*ksk
, unsigned int group
)
2183 struct netlink_table
*tbl
= &nl_table
[ksk
->sk_protocol
];
2185 sk_for_each_bound(sk
, &tbl
->mc_list
)
2186 netlink_update_socket_mc(nlk_sk(sk
), group
, 0);
2190 __nlmsg_put(struct sk_buff
*skb
, u32 portid
, u32 seq
, int type
, int len
, int flags
)
2192 struct nlmsghdr
*nlh
;
2193 int size
= nlmsg_msg_size(len
);
2195 nlh
= skb_put(skb
, NLMSG_ALIGN(size
));
2196 nlh
->nlmsg_type
= type
;
2197 nlh
->nlmsg_len
= size
;
2198 nlh
->nlmsg_flags
= flags
;
2199 nlh
->nlmsg_pid
= portid
;
2200 nlh
->nlmsg_seq
= seq
;
2201 if (!__builtin_constant_p(size
) || NLMSG_ALIGN(size
) - size
!= 0)
2202 memset(nlmsg_data(nlh
) + len
, 0, NLMSG_ALIGN(size
) - size
);
2205 EXPORT_SYMBOL(__nlmsg_put
);
2208 * It looks a bit ugly.
2209 * It would be better to create kernel thread.
2212 static int netlink_dump_done(struct netlink_sock
*nlk
, struct sk_buff
*skb
,
2213 struct netlink_callback
*cb
,
2214 struct netlink_ext_ack
*extack
)
2216 struct nlmsghdr
*nlh
;
2218 nlh
= nlmsg_put_answer(skb
, cb
, NLMSG_DONE
, sizeof(nlk
->dump_done_errno
),
2219 NLM_F_MULTI
| cb
->answer_flags
);
2223 nl_dump_check_consistent(cb
, nlh
);
2224 memcpy(nlmsg_data(nlh
), &nlk
->dump_done_errno
, sizeof(nlk
->dump_done_errno
));
2226 if (extack
->_msg
&& nlk
->flags
& NETLINK_F_EXT_ACK
) {
2227 nlh
->nlmsg_flags
|= NLM_F_ACK_TLVS
;
2228 if (!nla_put_string(skb
, NLMSGERR_ATTR_MSG
, extack
->_msg
))
2229 nlmsg_end(skb
, nlh
);
2235 static int netlink_dump(struct sock
*sk
)
2237 struct netlink_sock
*nlk
= nlk_sk(sk
);
2238 struct netlink_ext_ack extack
= {};
2239 struct netlink_callback
*cb
;
2240 struct sk_buff
*skb
= NULL
;
2241 struct module
*module
;
2246 mutex_lock(nlk
->cb_mutex
);
2247 if (!nlk
->cb_running
) {
2252 if (atomic_read(&sk
->sk_rmem_alloc
) >= sk
->sk_rcvbuf
)
2255 /* NLMSG_GOODSIZE is small to avoid high order allocations being
2256 * required, but it makes sense to _attempt_ a 16K bytes allocation
2257 * to reduce number of system calls on dump operations, if user
2258 * ever provided a big enough buffer.
2261 alloc_min_size
= max_t(int, cb
->min_dump_alloc
, NLMSG_GOODSIZE
);
2263 if (alloc_min_size
< nlk
->max_recvmsg_len
) {
2264 alloc_size
= nlk
->max_recvmsg_len
;
2265 skb
= alloc_skb(alloc_size
,
2266 (GFP_KERNEL
& ~__GFP_DIRECT_RECLAIM
) |
2267 __GFP_NOWARN
| __GFP_NORETRY
);
2270 alloc_size
= alloc_min_size
;
2271 skb
= alloc_skb(alloc_size
, GFP_KERNEL
);
2276 /* Trim skb to allocated size. User is expected to provide buffer as
2277 * large as max(min_dump_alloc, 16KiB (mac_recvmsg_len capped at
2278 * netlink_recvmsg())). dump will pack as many smaller messages as
2279 * could fit within the allocated skb. skb is typically allocated
2280 * with larger space than required (could be as much as near 2x the
2281 * requested size with align to next power of 2 approach). Allowing
2282 * dump to use the excess space makes it difficult for a user to have a
2283 * reasonable static buffer based on the expected largest dump of a
2284 * single netdev. The outcome is MSG_TRUNC error.
2286 skb_reserve(skb
, skb_tailroom(skb
) - alloc_size
);
2287 netlink_skb_set_owner_r(skb
, sk
);
2289 if (nlk
->dump_done_errno
> 0) {
2290 cb
->extack
= &extack
;
2291 nlk
->dump_done_errno
= cb
->dump(skb
, cb
);
2295 if (nlk
->dump_done_errno
> 0 ||
2296 skb_tailroom(skb
) < nlmsg_total_size(sizeof(nlk
->dump_done_errno
))) {
2297 mutex_unlock(nlk
->cb_mutex
);
2299 if (sk_filter(sk
, skb
))
2302 __netlink_sendskb(sk
, skb
);
2306 if (netlink_dump_done(nlk
, skb
, cb
, &extack
))
2309 #ifdef CONFIG_COMPAT_NETLINK_MESSAGES
2310 /* frag_list skb's data is used for compat tasks
2311 * and the regular skb's data for normal (non-compat) tasks.
2312 * See netlink_recvmsg().
2314 if (unlikely(skb_shinfo(skb
)->frag_list
)) {
2315 if (netlink_dump_done(nlk
, skb_shinfo(skb
)->frag_list
, cb
, &extack
))
2320 if (sk_filter(sk
, skb
))
2323 __netlink_sendskb(sk
, skb
);
2328 nlk
->cb_running
= false;
2329 module
= cb
->module
;
2331 mutex_unlock(nlk
->cb_mutex
);
2337 mutex_unlock(nlk
->cb_mutex
);
2342 int __netlink_dump_start(struct sock
*ssk
, struct sk_buff
*skb
,
2343 const struct nlmsghdr
*nlh
,
2344 struct netlink_dump_control
*control
)
2346 struct netlink_sock
*nlk
, *nlk2
;
2347 struct netlink_callback
*cb
;
2351 refcount_inc(&skb
->users
);
2353 sk
= netlink_lookup(sock_net(ssk
), ssk
->sk_protocol
, NETLINK_CB(skb
).portid
);
2355 ret
= -ECONNREFUSED
;
2360 mutex_lock(nlk
->cb_mutex
);
2361 /* A dump is in progress... */
2362 if (nlk
->cb_running
) {
2366 /* add reference of module which cb->dump belongs to */
2367 if (!try_module_get(control
->module
)) {
2368 ret
= -EPROTONOSUPPORT
;
2373 memset(cb
, 0, sizeof(*cb
));
2374 cb
->dump
= control
->dump
;
2375 cb
->done
= control
->done
;
2377 cb
->data
= control
->data
;
2378 cb
->module
= control
->module
;
2379 cb
->min_dump_alloc
= control
->min_dump_alloc
;
2382 nlk2
= nlk_sk(NETLINK_CB(skb
).sk
);
2383 cb
->strict_check
= !!(nlk2
->flags
& NETLINK_F_STRICT_CHK
);
2385 if (control
->start
) {
2386 ret
= control
->start(cb
);
2391 nlk
->cb_running
= true;
2392 nlk
->dump_done_errno
= INT_MAX
;
2394 mutex_unlock(nlk
->cb_mutex
);
2396 ret
= netlink_dump(sk
);
2403 /* We successfully started a dump, by returning -EINTR we
2404 * signal not to send ACK even if it was requested.
2409 module_put(control
->module
);
2412 mutex_unlock(nlk
->cb_mutex
);
2417 EXPORT_SYMBOL(__netlink_dump_start
);
2419 void netlink_ack(struct sk_buff
*in_skb
, struct nlmsghdr
*nlh
, int err
,
2420 const struct netlink_ext_ack
*extack
)
2422 struct sk_buff
*skb
;
2423 struct nlmsghdr
*rep
;
2424 struct nlmsgerr
*errmsg
;
2425 size_t payload
= sizeof(*errmsg
);
2427 struct netlink_sock
*nlk
= nlk_sk(NETLINK_CB(in_skb
).sk
);
2428 unsigned int flags
= 0;
2429 bool nlk_has_extack
= nlk
->flags
& NETLINK_F_EXT_ACK
;
2431 /* Error messages get the original request appened, unless the user
2432 * requests to cap the error message, and get extra error data if
2435 if (nlk_has_extack
&& extack
&& extack
->_msg
)
2436 tlvlen
+= nla_total_size(strlen(extack
->_msg
) + 1);
2438 if (err
&& !(nlk
->flags
& NETLINK_F_CAP_ACK
))
2439 payload
+= nlmsg_len(nlh
);
2441 flags
|= NLM_F_CAPPED
;
2442 if (err
&& nlk_has_extack
&& extack
&& extack
->bad_attr
)
2443 tlvlen
+= nla_total_size(sizeof(u32
));
2444 if (nlk_has_extack
&& extack
&& extack
->cookie_len
)
2445 tlvlen
+= nla_total_size(extack
->cookie_len
);
2446 if (err
&& nlk_has_extack
&& extack
&& extack
->policy
)
2447 tlvlen
+= netlink_policy_dump_attr_size_estimate(extack
->policy
);
2450 flags
|= NLM_F_ACK_TLVS
;
2452 skb
= nlmsg_new(payload
+ tlvlen
, GFP_KERNEL
);
2454 NETLINK_CB(in_skb
).sk
->sk_err
= ENOBUFS
;
2455 sk_error_report(NETLINK_CB(in_skb
).sk
);
2459 rep
= __nlmsg_put(skb
, NETLINK_CB(in_skb
).portid
, nlh
->nlmsg_seq
,
2460 NLMSG_ERROR
, payload
, flags
);
2461 errmsg
= nlmsg_data(rep
);
2462 errmsg
->error
= err
;
2463 memcpy(&errmsg
->msg
, nlh
, payload
> sizeof(*errmsg
) ? nlh
->nlmsg_len
: sizeof(*nlh
));
2465 if (nlk_has_extack
&& extack
) {
2467 WARN_ON(nla_put_string(skb
, NLMSGERR_ATTR_MSG
,
2470 if (err
&& extack
->bad_attr
&&
2471 !WARN_ON((u8
*)extack
->bad_attr
< in_skb
->data
||
2472 (u8
*)extack
->bad_attr
>= in_skb
->data
+
2474 WARN_ON(nla_put_u32(skb
, NLMSGERR_ATTR_OFFS
,
2475 (u8
*)extack
->bad_attr
-
2477 if (extack
->cookie_len
)
2478 WARN_ON(nla_put(skb
, NLMSGERR_ATTR_COOKIE
,
2479 extack
->cookie_len
, extack
->cookie
));
2481 netlink_policy_dump_write_attr(skb
, extack
->policy
,
2482 NLMSGERR_ATTR_POLICY
);
2485 nlmsg_end(skb
, rep
);
2487 nlmsg_unicast(in_skb
->sk
, skb
, NETLINK_CB(in_skb
).portid
);
2489 EXPORT_SYMBOL(netlink_ack
);
2491 int netlink_rcv_skb(struct sk_buff
*skb
, int (*cb
)(struct sk_buff
*,
2493 struct netlink_ext_ack
*))
2495 struct netlink_ext_ack extack
;
2496 struct nlmsghdr
*nlh
;
2499 while (skb
->len
>= nlmsg_total_size(0)) {
2502 memset(&extack
, 0, sizeof(extack
));
2503 nlh
= nlmsg_hdr(skb
);
2506 if (nlh
->nlmsg_len
< NLMSG_HDRLEN
|| skb
->len
< nlh
->nlmsg_len
)
2509 /* Only requests are handled by the kernel */
2510 if (!(nlh
->nlmsg_flags
& NLM_F_REQUEST
))
2513 /* Skip control messages */
2514 if (nlh
->nlmsg_type
< NLMSG_MIN_TYPE
)
2517 err
= cb(skb
, nlh
, &extack
);
2522 if (nlh
->nlmsg_flags
& NLM_F_ACK
|| err
)
2523 netlink_ack(skb
, nlh
, err
, &extack
);
2526 msglen
= NLMSG_ALIGN(nlh
->nlmsg_len
);
2527 if (msglen
> skb
->len
)
2529 skb_pull(skb
, msglen
);
2534 EXPORT_SYMBOL(netlink_rcv_skb
);
2537 * nlmsg_notify - send a notification netlink message
2538 * @sk: netlink socket to use
2539 * @skb: notification message
2540 * @portid: destination netlink portid for reports or 0
2541 * @group: destination multicast group or 0
2542 * @report: 1 to report back, 0 to disable
2543 * @flags: allocation flags
2545 int nlmsg_notify(struct sock
*sk
, struct sk_buff
*skb
, u32 portid
,
2546 unsigned int group
, int report
, gfp_t flags
)
2551 int exclude_portid
= 0;
2554 refcount_inc(&skb
->users
);
2555 exclude_portid
= portid
;
2558 /* errors reported via destination sk->sk_err, but propagate
2559 * delivery errors if NETLINK_BROADCAST_ERROR flag is set */
2560 err
= nlmsg_multicast(sk
, skb
, exclude_portid
, group
, flags
);
2568 err2
= nlmsg_unicast(sk
, skb
, portid
);
2575 EXPORT_SYMBOL(nlmsg_notify
);
2577 #ifdef CONFIG_PROC_FS
2578 struct nl_seq_iter
{
2579 struct seq_net_private p
;
2580 struct rhashtable_iter hti
;
2584 static void netlink_walk_start(struct nl_seq_iter
*iter
)
2586 rhashtable_walk_enter(&nl_table
[iter
->link
].hash
, &iter
->hti
);
2587 rhashtable_walk_start(&iter
->hti
);
2590 static void netlink_walk_stop(struct nl_seq_iter
*iter
)
2592 rhashtable_walk_stop(&iter
->hti
);
2593 rhashtable_walk_exit(&iter
->hti
);
2596 static void *__netlink_seq_next(struct seq_file
*seq
)
2598 struct nl_seq_iter
*iter
= seq
->private;
2599 struct netlink_sock
*nlk
;
2603 nlk
= rhashtable_walk_next(&iter
->hti
);
2606 if (PTR_ERR(nlk
) == -EAGAIN
)
2615 netlink_walk_stop(iter
);
2616 if (++iter
->link
>= MAX_LINKS
)
2619 netlink_walk_start(iter
);
2621 } while (sock_net(&nlk
->sk
) != seq_file_net(seq
));
2626 static void *netlink_seq_start(struct seq_file
*seq
, loff_t
*posp
)
2629 struct nl_seq_iter
*iter
= seq
->private;
2630 void *obj
= SEQ_START_TOKEN
;
2635 netlink_walk_start(iter
);
2637 for (pos
= *posp
; pos
&& obj
&& !IS_ERR(obj
); pos
--)
2638 obj
= __netlink_seq_next(seq
);
2643 static void *netlink_seq_next(struct seq_file
*seq
, void *v
, loff_t
*pos
)
2646 return __netlink_seq_next(seq
);
2649 static void netlink_native_seq_stop(struct seq_file
*seq
, void *v
)
2651 struct nl_seq_iter
*iter
= seq
->private;
2653 if (iter
->link
>= MAX_LINKS
)
2656 netlink_walk_stop(iter
);
2660 static int netlink_native_seq_show(struct seq_file
*seq
, void *v
)
2662 if (v
== SEQ_START_TOKEN
) {
2664 "sk Eth Pid Groups "
2665 "Rmem Wmem Dump Locks Drops Inode\n");
2668 struct netlink_sock
*nlk
= nlk_sk(s
);
2670 seq_printf(seq
, "%pK %-3d %-10u %08x %-8d %-8d %-5d %-8d %-8u %-8lu\n",
2674 nlk
->groups
? (u32
)nlk
->groups
[0] : 0,
2675 sk_rmem_alloc_get(s
),
2676 sk_wmem_alloc_get(s
),
2678 refcount_read(&s
->sk_refcnt
),
2679 atomic_read(&s
->sk_drops
),
2687 #ifdef CONFIG_BPF_SYSCALL
2688 struct bpf_iter__netlink
{
2689 __bpf_md_ptr(struct bpf_iter_meta
*, meta
);
2690 __bpf_md_ptr(struct netlink_sock
*, sk
);
2693 DEFINE_BPF_ITER_FUNC(netlink
, struct bpf_iter_meta
*meta
, struct netlink_sock
*sk
)
2695 static int netlink_prog_seq_show(struct bpf_prog
*prog
,
2696 struct bpf_iter_meta
*meta
,
2699 struct bpf_iter__netlink ctx
;
2701 meta
->seq_num
--; /* skip SEQ_START_TOKEN */
2703 ctx
.sk
= nlk_sk((struct sock
*)v
);
2704 return bpf_iter_run_prog(prog
, &ctx
);
2707 static int netlink_seq_show(struct seq_file
*seq
, void *v
)
2709 struct bpf_iter_meta meta
;
2710 struct bpf_prog
*prog
;
2713 prog
= bpf_iter_get_info(&meta
, false);
2715 return netlink_native_seq_show(seq
, v
);
2717 if (v
!= SEQ_START_TOKEN
)
2718 return netlink_prog_seq_show(prog
, &meta
, v
);
2723 static void netlink_seq_stop(struct seq_file
*seq
, void *v
)
2725 struct bpf_iter_meta meta
;
2726 struct bpf_prog
*prog
;
2730 prog
= bpf_iter_get_info(&meta
, true);
2732 (void)netlink_prog_seq_show(prog
, &meta
, v
);
2735 netlink_native_seq_stop(seq
, v
);
2738 static int netlink_seq_show(struct seq_file
*seq
, void *v
)
2740 return netlink_native_seq_show(seq
, v
);
2743 static void netlink_seq_stop(struct seq_file
*seq
, void *v
)
2745 netlink_native_seq_stop(seq
, v
);
2749 static const struct seq_operations netlink_seq_ops
= {
2750 .start
= netlink_seq_start
,
2751 .next
= netlink_seq_next
,
2752 .stop
= netlink_seq_stop
,
2753 .show
= netlink_seq_show
,
2757 int netlink_register_notifier(struct notifier_block
*nb
)
2759 return blocking_notifier_chain_register(&netlink_chain
, nb
);
2761 EXPORT_SYMBOL(netlink_register_notifier
);
2763 int netlink_unregister_notifier(struct notifier_block
*nb
)
2765 return blocking_notifier_chain_unregister(&netlink_chain
, nb
);
2767 EXPORT_SYMBOL(netlink_unregister_notifier
);
2769 static const struct proto_ops netlink_ops
= {
2770 .family
= PF_NETLINK
,
2771 .owner
= THIS_MODULE
,
2772 .release
= netlink_release
,
2773 .bind
= netlink_bind
,
2774 .connect
= netlink_connect
,
2775 .socketpair
= sock_no_socketpair
,
2776 .accept
= sock_no_accept
,
2777 .getname
= netlink_getname
,
2778 .poll
= datagram_poll
,
2779 .ioctl
= netlink_ioctl
,
2780 .listen
= sock_no_listen
,
2781 .shutdown
= sock_no_shutdown
,
2782 .setsockopt
= netlink_setsockopt
,
2783 .getsockopt
= netlink_getsockopt
,
2784 .sendmsg
= netlink_sendmsg
,
2785 .recvmsg
= netlink_recvmsg
,
2786 .mmap
= sock_no_mmap
,
2787 .sendpage
= sock_no_sendpage
,
2790 static const struct net_proto_family netlink_family_ops
= {
2791 .family
= PF_NETLINK
,
2792 .create
= netlink_create
,
2793 .owner
= THIS_MODULE
, /* for consistency 8) */
2796 static int __net_init
netlink_net_init(struct net
*net
)
2798 #ifdef CONFIG_PROC_FS
2799 if (!proc_create_net("netlink", 0, net
->proc_net
, &netlink_seq_ops
,
2800 sizeof(struct nl_seq_iter
)))
2806 static void __net_exit
netlink_net_exit(struct net
*net
)
2808 #ifdef CONFIG_PROC_FS
2809 remove_proc_entry("netlink", net
->proc_net
);
2813 static void __init
netlink_add_usersock_entry(void)
2815 struct listeners
*listeners
;
2818 listeners
= kzalloc(sizeof(*listeners
) + NLGRPSZ(groups
), GFP_KERNEL
);
2820 panic("netlink_add_usersock_entry: Cannot allocate listeners\n");
2822 netlink_table_grab();
2824 nl_table
[NETLINK_USERSOCK
].groups
= groups
;
2825 rcu_assign_pointer(nl_table
[NETLINK_USERSOCK
].listeners
, listeners
);
2826 nl_table
[NETLINK_USERSOCK
].module
= THIS_MODULE
;
2827 nl_table
[NETLINK_USERSOCK
].registered
= 1;
2828 nl_table
[NETLINK_USERSOCK
].flags
= NL_CFG_F_NONROOT_SEND
;
2830 netlink_table_ungrab();
2833 static struct pernet_operations __net_initdata netlink_net_ops
= {
2834 .init
= netlink_net_init
,
2835 .exit
= netlink_net_exit
,
2838 static inline u32
netlink_hash(const void *data
, u32 len
, u32 seed
)
2840 const struct netlink_sock
*nlk
= data
;
2841 struct netlink_compare_arg arg
;
2843 netlink_compare_arg_init(&arg
, sock_net(&nlk
->sk
), nlk
->portid
);
2844 return jhash2((u32
*)&arg
, netlink_compare_arg_len
/ sizeof(u32
), seed
);
2847 static const struct rhashtable_params netlink_rhashtable_params
= {
2848 .head_offset
= offsetof(struct netlink_sock
, node
),
2849 .key_len
= netlink_compare_arg_len
,
2850 .obj_hashfn
= netlink_hash
,
2851 .obj_cmpfn
= netlink_compare
,
2852 .automatic_shrinking
= true,
2855 #if defined(CONFIG_BPF_SYSCALL) && defined(CONFIG_PROC_FS)
2856 BTF_ID_LIST(btf_netlink_sock_id
)
2857 BTF_ID(struct, netlink_sock
)
2859 static const struct bpf_iter_seq_info netlink_seq_info
= {
2860 .seq_ops
= &netlink_seq_ops
,
2861 .init_seq_private
= bpf_iter_init_seq_net
,
2862 .fini_seq_private
= bpf_iter_fini_seq_net
,
2863 .seq_priv_size
= sizeof(struct nl_seq_iter
),
2866 static struct bpf_iter_reg netlink_reg_info
= {
2867 .target
= "netlink",
2868 .ctx_arg_info_size
= 1,
2870 { offsetof(struct bpf_iter__netlink
, sk
),
2871 PTR_TO_BTF_ID_OR_NULL
},
2873 .seq_info
= &netlink_seq_info
,
2876 static int __init
bpf_iter_register(void)
2878 netlink_reg_info
.ctx_arg_info
[0].btf_id
= *btf_netlink_sock_id
;
2879 return bpf_iter_reg_target(&netlink_reg_info
);
2883 static int __init
netlink_proto_init(void)
2886 int err
= proto_register(&netlink_proto
, 0);
2891 #if defined(CONFIG_BPF_SYSCALL) && defined(CONFIG_PROC_FS)
2892 err
= bpf_iter_register();
2897 BUILD_BUG_ON(sizeof(struct netlink_skb_parms
) > sizeof_field(struct sk_buff
, cb
));
2899 nl_table
= kcalloc(MAX_LINKS
, sizeof(*nl_table
), GFP_KERNEL
);
2903 for (i
= 0; i
< MAX_LINKS
; i
++) {
2904 if (rhashtable_init(&nl_table
[i
].hash
,
2905 &netlink_rhashtable_params
) < 0) {
2907 rhashtable_destroy(&nl_table
[i
].hash
);
2913 netlink_add_usersock_entry();
2915 sock_register(&netlink_family_ops
);
2916 register_pernet_subsys(&netlink_net_ops
);
2917 register_pernet_subsys(&netlink_tap_net_ops
);
2918 /* The netlink device handler may be needed early. */
2923 panic("netlink_init: Cannot allocate nl_table\n");
2926 core_initcall(netlink_proto_init
);