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>
71 #include "af_netlink.h"
75 unsigned long masks[];
79 #define NETLINK_S_CONGESTED 0x0
81 static inline int netlink_is_kernel(struct sock *sk)
83 return nlk_sk(sk)->flags & NETLINK_F_KERNEL_SOCKET;
86 struct netlink_table *nl_table __read_mostly;
87 EXPORT_SYMBOL_GPL(nl_table);
89 static DECLARE_WAIT_QUEUE_HEAD(nl_table_wait);
91 static struct lock_class_key nlk_cb_mutex_keys[MAX_LINKS];
93 static const char *const nlk_cb_mutex_key_strings[MAX_LINKS + 1] = {
96 "nlk_cb_mutex-USERSOCK",
97 "nlk_cb_mutex-FIREWALL",
98 "nlk_cb_mutex-SOCK_DIAG",
101 "nlk_cb_mutex-SELINUX",
102 "nlk_cb_mutex-ISCSI",
103 "nlk_cb_mutex-AUDIT",
104 "nlk_cb_mutex-FIB_LOOKUP",
105 "nlk_cb_mutex-CONNECTOR",
106 "nlk_cb_mutex-NETFILTER",
107 "nlk_cb_mutex-IP6_FW",
108 "nlk_cb_mutex-DNRTMSG",
109 "nlk_cb_mutex-KOBJECT_UEVENT",
110 "nlk_cb_mutex-GENERIC",
112 "nlk_cb_mutex-SCSITRANSPORT",
113 "nlk_cb_mutex-ECRYPTFS",
115 "nlk_cb_mutex-CRYPTO",
126 "nlk_cb_mutex-MAX_LINKS"
129 static int netlink_dump(struct sock *sk);
131 /* nl_table locking explained:
132 * Lookup and traversal are protected with an RCU read-side lock. Insertion
133 * and removal are protected with per bucket lock while using RCU list
134 * modification primitives and may run in parallel to RCU protected lookups.
135 * Destruction of the Netlink socket may only occur *after* nl_table_lock has
136 * been acquired * either during or after the socket has been removed from
137 * the list and after an RCU grace period.
139 DEFINE_RWLOCK(nl_table_lock);
140 EXPORT_SYMBOL_GPL(nl_table_lock);
141 static atomic_t nl_table_users = ATOMIC_INIT(0);
143 #define nl_deref_protected(X) rcu_dereference_protected(X, lockdep_is_held(&nl_table_lock));
145 static BLOCKING_NOTIFIER_HEAD(netlink_chain);
148 static const struct rhashtable_params netlink_rhashtable_params;
150 static inline u32 netlink_group_mask(u32 group)
154 return group ? 1 << (group - 1) : 0;
157 static struct sk_buff *netlink_to_full_skb(const struct sk_buff *skb,
160 unsigned int len = skb->len;
163 new = alloc_skb(len, gfp_mask);
167 NETLINK_CB(new).portid = NETLINK_CB(skb).portid;
168 NETLINK_CB(new).dst_group = NETLINK_CB(skb).dst_group;
169 NETLINK_CB(new).creds = NETLINK_CB(skb).creds;
171 skb_put_data(new, skb->data, len);
175 static unsigned int netlink_tap_net_id;
177 struct netlink_tap_net {
178 struct list_head netlink_tap_all;
179 struct mutex netlink_tap_lock;
182 int netlink_add_tap(struct netlink_tap *nt)
184 struct net *net = dev_net(nt->dev);
185 struct netlink_tap_net *nn = net_generic(net, netlink_tap_net_id);
187 if (unlikely(nt->dev->type != ARPHRD_NETLINK))
190 mutex_lock(&nn->netlink_tap_lock);
191 list_add_rcu(&nt->list, &nn->netlink_tap_all);
192 mutex_unlock(&nn->netlink_tap_lock);
194 __module_get(nt->module);
198 EXPORT_SYMBOL_GPL(netlink_add_tap);
200 static int __netlink_remove_tap(struct netlink_tap *nt)
202 struct net *net = dev_net(nt->dev);
203 struct netlink_tap_net *nn = net_generic(net, netlink_tap_net_id);
205 struct netlink_tap *tmp;
207 mutex_lock(&nn->netlink_tap_lock);
209 list_for_each_entry(tmp, &nn->netlink_tap_all, list) {
211 list_del_rcu(&nt->list);
217 pr_warn("__netlink_remove_tap: %p not found\n", nt);
219 mutex_unlock(&nn->netlink_tap_lock);
222 module_put(nt->module);
224 return found ? 0 : -ENODEV;
227 int netlink_remove_tap(struct netlink_tap *nt)
231 ret = __netlink_remove_tap(nt);
236 EXPORT_SYMBOL_GPL(netlink_remove_tap);
238 static __net_init int netlink_tap_init_net(struct net *net)
240 struct netlink_tap_net *nn = net_generic(net, netlink_tap_net_id);
242 INIT_LIST_HEAD(&nn->netlink_tap_all);
243 mutex_init(&nn->netlink_tap_lock);
247 static struct pernet_operations netlink_tap_net_ops = {
248 .init = netlink_tap_init_net,
249 .id = &netlink_tap_net_id,
250 .size = sizeof(struct netlink_tap_net),
253 static bool netlink_filter_tap(const struct sk_buff *skb)
255 struct sock *sk = skb->sk;
257 /* We take the more conservative approach and
258 * whitelist socket protocols that may pass.
260 switch (sk->sk_protocol) {
262 case NETLINK_USERSOCK:
263 case NETLINK_SOCK_DIAG:
266 case NETLINK_FIB_LOOKUP:
267 case NETLINK_NETFILTER:
268 case NETLINK_GENERIC:
275 static int __netlink_deliver_tap_skb(struct sk_buff *skb,
276 struct net_device *dev)
278 struct sk_buff *nskb;
279 struct sock *sk = skb->sk;
282 if (!net_eq(dev_net(dev), sock_net(sk)))
287 if (is_vmalloc_addr(skb->head))
288 nskb = netlink_to_full_skb(skb, GFP_ATOMIC);
290 nskb = skb_clone(skb, GFP_ATOMIC);
293 nskb->protocol = htons((u16) sk->sk_protocol);
294 nskb->pkt_type = netlink_is_kernel(sk) ?
295 PACKET_KERNEL : PACKET_USER;
296 skb_reset_network_header(nskb);
297 ret = dev_queue_xmit(nskb);
298 if (unlikely(ret > 0))
299 ret = net_xmit_errno(ret);
306 static void __netlink_deliver_tap(struct sk_buff *skb, struct netlink_tap_net *nn)
309 struct netlink_tap *tmp;
311 if (!netlink_filter_tap(skb))
314 list_for_each_entry_rcu(tmp, &nn->netlink_tap_all, list) {
315 ret = __netlink_deliver_tap_skb(skb, tmp->dev);
321 static void netlink_deliver_tap(struct net *net, struct sk_buff *skb)
323 struct netlink_tap_net *nn = net_generic(net, netlink_tap_net_id);
327 if (unlikely(!list_empty(&nn->netlink_tap_all)))
328 __netlink_deliver_tap(skb, nn);
333 static void netlink_deliver_tap_kernel(struct sock *dst, struct sock *src,
336 if (!(netlink_is_kernel(dst) && netlink_is_kernel(src)))
337 netlink_deliver_tap(sock_net(dst), skb);
340 static void netlink_overrun(struct sock *sk)
342 struct netlink_sock *nlk = nlk_sk(sk);
344 if (!(nlk->flags & NETLINK_F_RECV_NO_ENOBUFS)) {
345 if (!test_and_set_bit(NETLINK_S_CONGESTED,
346 &nlk_sk(sk)->state)) {
347 sk->sk_err = ENOBUFS;
348 sk->sk_error_report(sk);
351 atomic_inc(&sk->sk_drops);
354 static void netlink_rcv_wake(struct sock *sk)
356 struct netlink_sock *nlk = nlk_sk(sk);
358 if (skb_queue_empty_lockless(&sk->sk_receive_queue))
359 clear_bit(NETLINK_S_CONGESTED, &nlk->state);
360 if (!test_bit(NETLINK_S_CONGESTED, &nlk->state))
361 wake_up_interruptible(&nlk->wait);
364 static void netlink_skb_destructor(struct sk_buff *skb)
366 if (is_vmalloc_addr(skb->head)) {
368 !atomic_dec_return(&(skb_shinfo(skb)->dataref)))
369 vfree_atomic(skb->head);
377 static void netlink_skb_set_owner_r(struct sk_buff *skb, struct sock *sk)
379 WARN_ON(skb->sk != NULL);
381 skb->destructor = netlink_skb_destructor;
382 atomic_add(skb->truesize, &sk->sk_rmem_alloc);
383 sk_mem_charge(sk, skb->truesize);
386 static void netlink_sock_destruct(struct sock *sk)
388 struct netlink_sock *nlk = nlk_sk(sk);
390 if (nlk->cb_running) {
392 nlk->cb.done(&nlk->cb);
393 module_put(nlk->cb.module);
394 kfree_skb(nlk->cb.skb);
397 skb_queue_purge(&sk->sk_receive_queue);
399 if (!sock_flag(sk, SOCK_DEAD)) {
400 printk(KERN_ERR "Freeing alive netlink socket %p\n", sk);
404 WARN_ON(atomic_read(&sk->sk_rmem_alloc));
405 WARN_ON(refcount_read(&sk->sk_wmem_alloc));
406 WARN_ON(nlk_sk(sk)->groups);
409 static void netlink_sock_destruct_work(struct work_struct *work)
411 struct netlink_sock *nlk = container_of(work, struct netlink_sock,
417 /* This lock without WQ_FLAG_EXCLUSIVE is good on UP and it is _very_ bad on
418 * SMP. Look, when several writers sleep and reader wakes them up, all but one
419 * immediately hit write lock and grab all the cpus. Exclusive sleep solves
420 * this, _but_ remember, it adds useless work on UP machines.
423 void netlink_table_grab(void)
424 __acquires(nl_table_lock)
428 write_lock_irq(&nl_table_lock);
430 if (atomic_read(&nl_table_users)) {
431 DECLARE_WAITQUEUE(wait, current);
433 add_wait_queue_exclusive(&nl_table_wait, &wait);
435 set_current_state(TASK_UNINTERRUPTIBLE);
436 if (atomic_read(&nl_table_users) == 0)
438 write_unlock_irq(&nl_table_lock);
440 write_lock_irq(&nl_table_lock);
443 __set_current_state(TASK_RUNNING);
444 remove_wait_queue(&nl_table_wait, &wait);
448 void netlink_table_ungrab(void)
449 __releases(nl_table_lock)
451 write_unlock_irq(&nl_table_lock);
452 wake_up(&nl_table_wait);
456 netlink_lock_table(void)
460 /* read_lock() synchronizes us to netlink_table_grab */
462 read_lock_irqsave(&nl_table_lock, flags);
463 atomic_inc(&nl_table_users);
464 read_unlock_irqrestore(&nl_table_lock, flags);
468 netlink_unlock_table(void)
470 if (atomic_dec_and_test(&nl_table_users))
471 wake_up(&nl_table_wait);
474 struct netlink_compare_arg
480 /* Doing sizeof directly may yield 4 extra bytes on 64-bit. */
481 #define netlink_compare_arg_len \
482 (offsetof(struct netlink_compare_arg, portid) + sizeof(u32))
484 static inline int netlink_compare(struct rhashtable_compare_arg *arg,
487 const struct netlink_compare_arg *x = arg->key;
488 const struct netlink_sock *nlk = ptr;
490 return nlk->portid != x->portid ||
491 !net_eq(sock_net(&nlk->sk), read_pnet(&x->pnet));
494 static void netlink_compare_arg_init(struct netlink_compare_arg *arg,
495 struct net *net, u32 portid)
497 memset(arg, 0, sizeof(*arg));
498 write_pnet(&arg->pnet, net);
499 arg->portid = portid;
502 static struct sock *__netlink_lookup(struct netlink_table *table, u32 portid,
505 struct netlink_compare_arg arg;
507 netlink_compare_arg_init(&arg, net, portid);
508 return rhashtable_lookup_fast(&table->hash, &arg,
509 netlink_rhashtable_params);
512 static int __netlink_insert(struct netlink_table *table, struct sock *sk)
514 struct netlink_compare_arg arg;
516 netlink_compare_arg_init(&arg, sock_net(sk), nlk_sk(sk)->portid);
517 return rhashtable_lookup_insert_key(&table->hash, &arg,
519 netlink_rhashtable_params);
522 static struct sock *netlink_lookup(struct net *net, int protocol, u32 portid)
524 struct netlink_table *table = &nl_table[protocol];
528 sk = __netlink_lookup(table, portid, net);
536 static const struct proto_ops netlink_ops;
539 netlink_update_listeners(struct sock *sk)
541 struct netlink_table *tbl = &nl_table[sk->sk_protocol];
544 struct listeners *listeners;
546 listeners = nl_deref_protected(tbl->listeners);
550 for (i = 0; i < NLGRPLONGS(tbl->groups); i++) {
552 sk_for_each_bound(sk, &tbl->mc_list) {
553 if (i < NLGRPLONGS(nlk_sk(sk)->ngroups))
554 mask |= nlk_sk(sk)->groups[i];
556 listeners->masks[i] = mask;
558 /* this function is only called with the netlink table "grabbed", which
559 * makes sure updates are visible before bind or setsockopt return. */
562 static int netlink_insert(struct sock *sk, u32 portid)
564 struct netlink_table *table = &nl_table[sk->sk_protocol];
569 err = nlk_sk(sk)->portid == portid ? 0 : -EBUSY;
570 if (nlk_sk(sk)->bound)
573 /* portid can be read locklessly from netlink_getname(). */
574 WRITE_ONCE(nlk_sk(sk)->portid, portid);
578 err = __netlink_insert(table, sk);
580 /* In case the hashtable backend returns with -EBUSY
581 * from here, it must not escape to the caller.
583 if (unlikely(err == -EBUSY))
591 /* We need to ensure that the socket is hashed and visible. */
593 /* Paired with lockless reads from netlink_bind(),
594 * netlink_connect() and netlink_sendmsg().
596 WRITE_ONCE(nlk_sk(sk)->bound, portid);
603 static void netlink_remove(struct sock *sk)
605 struct netlink_table *table;
607 table = &nl_table[sk->sk_protocol];
608 if (!rhashtable_remove_fast(&table->hash, &nlk_sk(sk)->node,
609 netlink_rhashtable_params)) {
610 WARN_ON(refcount_read(&sk->sk_refcnt) == 1);
614 netlink_table_grab();
615 if (nlk_sk(sk)->subscriptions) {
616 __sk_del_bind_node(sk);
617 netlink_update_listeners(sk);
619 if (sk->sk_protocol == NETLINK_GENERIC)
620 atomic_inc(&genl_sk_destructing_cnt);
621 netlink_table_ungrab();
624 static struct proto netlink_proto = {
626 .owner = THIS_MODULE,
627 .obj_size = sizeof(struct netlink_sock),
630 static int __netlink_create(struct net *net, struct socket *sock,
631 struct mutex *cb_mutex, int protocol,
635 struct netlink_sock *nlk;
637 sock->ops = &netlink_ops;
639 sk = sk_alloc(net, PF_NETLINK, GFP_KERNEL, &netlink_proto, kern);
643 sock_init_data(sock, sk);
647 nlk->cb_mutex = cb_mutex;
649 nlk->cb_mutex = &nlk->cb_def_mutex;
650 mutex_init(nlk->cb_mutex);
651 lockdep_set_class_and_name(nlk->cb_mutex,
652 nlk_cb_mutex_keys + protocol,
653 nlk_cb_mutex_key_strings[protocol]);
655 init_waitqueue_head(&nlk->wait);
657 sk->sk_destruct = netlink_sock_destruct;
658 sk->sk_protocol = protocol;
662 static int netlink_create(struct net *net, struct socket *sock, int protocol,
665 struct module *module = NULL;
666 struct mutex *cb_mutex;
667 struct netlink_sock *nlk;
668 int (*bind)(struct net *net, int group);
669 void (*unbind)(struct net *net, int group);
672 sock->state = SS_UNCONNECTED;
674 if (sock->type != SOCK_RAW && sock->type != SOCK_DGRAM)
675 return -ESOCKTNOSUPPORT;
677 if (protocol < 0 || protocol >= MAX_LINKS)
678 return -EPROTONOSUPPORT;
679 protocol = array_index_nospec(protocol, MAX_LINKS);
681 netlink_lock_table();
682 #ifdef CONFIG_MODULES
683 if (!nl_table[protocol].registered) {
684 netlink_unlock_table();
685 request_module("net-pf-%d-proto-%d", PF_NETLINK, protocol);
686 netlink_lock_table();
689 if (nl_table[protocol].registered &&
690 try_module_get(nl_table[protocol].module))
691 module = nl_table[protocol].module;
693 err = -EPROTONOSUPPORT;
694 cb_mutex = nl_table[protocol].cb_mutex;
695 bind = nl_table[protocol].bind;
696 unbind = nl_table[protocol].unbind;
697 netlink_unlock_table();
702 err = __netlink_create(net, sock, cb_mutex, protocol, kern);
707 sock_prot_inuse_add(net, &netlink_proto, 1);
710 nlk = nlk_sk(sock->sk);
711 nlk->module = module;
712 nlk->netlink_bind = bind;
713 nlk->netlink_unbind = unbind;
722 static void deferred_put_nlk_sk(struct rcu_head *head)
724 struct netlink_sock *nlk = container_of(head, struct netlink_sock, rcu);
725 struct sock *sk = &nlk->sk;
730 if (!refcount_dec_and_test(&sk->sk_refcnt))
733 if (nlk->cb_running && nlk->cb.done) {
734 INIT_WORK(&nlk->work, netlink_sock_destruct_work);
735 schedule_work(&nlk->work);
742 static int netlink_release(struct socket *sock)
744 struct sock *sk = sock->sk;
745 struct netlink_sock *nlk;
755 * OK. Socket is unlinked, any packets that arrive now
759 /* must not acquire netlink_table_lock in any way again before unbind
760 * and notifying genetlink is done as otherwise it might deadlock
762 if (nlk->netlink_unbind) {
765 for (i = 0; i < nlk->ngroups; i++)
766 if (test_bit(i, nlk->groups))
767 nlk->netlink_unbind(sock_net(sk), i + 1);
769 if (sk->sk_protocol == NETLINK_GENERIC &&
770 atomic_dec_return(&genl_sk_destructing_cnt) == 0)
771 wake_up(&genl_sk_destructing_waitq);
774 wake_up_interruptible_all(&nlk->wait);
776 skb_queue_purge(&sk->sk_write_queue);
778 if (nlk->portid && nlk->bound) {
779 struct netlink_notify n = {
781 .protocol = sk->sk_protocol,
782 .portid = nlk->portid,
784 blocking_notifier_call_chain(&netlink_chain,
785 NETLINK_URELEASE, &n);
788 module_put(nlk->module);
790 if (netlink_is_kernel(sk)) {
791 netlink_table_grab();
792 BUG_ON(nl_table[sk->sk_protocol].registered == 0);
793 if (--nl_table[sk->sk_protocol].registered == 0) {
794 struct listeners *old;
796 old = nl_deref_protected(nl_table[sk->sk_protocol].listeners);
797 RCU_INIT_POINTER(nl_table[sk->sk_protocol].listeners, NULL);
799 nl_table[sk->sk_protocol].module = NULL;
800 nl_table[sk->sk_protocol].bind = NULL;
801 nl_table[sk->sk_protocol].unbind = NULL;
802 nl_table[sk->sk_protocol].flags = 0;
803 nl_table[sk->sk_protocol].registered = 0;
805 netlink_table_ungrab();
809 sock_prot_inuse_add(sock_net(sk), &netlink_proto, -1);
811 call_rcu(&nlk->rcu, deferred_put_nlk_sk);
815 static int netlink_autobind(struct socket *sock)
817 struct sock *sk = sock->sk;
818 struct net *net = sock_net(sk);
819 struct netlink_table *table = &nl_table[sk->sk_protocol];
820 s32 portid = task_tgid_vnr(current);
828 ok = !__netlink_lookup(table, portid, net);
831 /* Bind collision, search negative portid values. */
833 /* rover will be in range [S32_MIN, -4097] */
834 rover = S32_MIN + prandom_u32_max(-4096 - S32_MIN);
835 else if (rover >= -4096)
841 err = netlink_insert(sk, portid);
842 if (err == -EADDRINUSE)
845 /* If 2 threads race to autobind, that is fine. */
853 * __netlink_ns_capable - General netlink message capability test
854 * @nsp: NETLINK_CB of the socket buffer holding a netlink command from userspace.
855 * @user_ns: The user namespace of the capability to use
856 * @cap: The capability to use
858 * Test to see if the opener of the socket we received the message
859 * from had when the netlink socket was created and the sender of the
860 * message has the capability @cap in the user namespace @user_ns.
862 bool __netlink_ns_capable(const struct netlink_skb_parms *nsp,
863 struct user_namespace *user_ns, int cap)
865 return ((nsp->flags & NETLINK_SKB_DST) ||
866 file_ns_capable(nsp->sk->sk_socket->file, user_ns, cap)) &&
867 ns_capable(user_ns, cap);
869 EXPORT_SYMBOL(__netlink_ns_capable);
872 * netlink_ns_capable - General netlink message capability test
873 * @skb: socket buffer holding a netlink command from userspace
874 * @user_ns: The user namespace of the capability to use
875 * @cap: The capability to use
877 * Test to see if the opener of the socket we received the message
878 * from had when the netlink socket was created and the sender of the
879 * message has the capability @cap in the user namespace @user_ns.
881 bool netlink_ns_capable(const struct sk_buff *skb,
882 struct user_namespace *user_ns, int cap)
884 return __netlink_ns_capable(&NETLINK_CB(skb), user_ns, cap);
886 EXPORT_SYMBOL(netlink_ns_capable);
889 * netlink_capable - Netlink global message capability test
890 * @skb: socket buffer holding a netlink command from userspace
891 * @cap: The capability to use
893 * Test to see if the opener of the socket we received the message
894 * from had when the netlink socket was created and the sender of the
895 * message has the capability @cap in all user namespaces.
897 bool netlink_capable(const struct sk_buff *skb, int cap)
899 return netlink_ns_capable(skb, &init_user_ns, cap);
901 EXPORT_SYMBOL(netlink_capable);
904 * netlink_net_capable - Netlink network namespace message capability test
905 * @skb: socket buffer holding a netlink command from userspace
906 * @cap: The capability to use
908 * Test to see if the opener of the socket we received the message
909 * from had when the netlink socket was created and the sender of the
910 * message has the capability @cap over the network namespace of
911 * the socket we received the message from.
913 bool netlink_net_capable(const struct sk_buff *skb, int cap)
915 return netlink_ns_capable(skb, sock_net(skb->sk)->user_ns, cap);
917 EXPORT_SYMBOL(netlink_net_capable);
919 static inline int netlink_allowed(const struct socket *sock, unsigned int flag)
921 return (nl_table[sock->sk->sk_protocol].flags & flag) ||
922 ns_capable(sock_net(sock->sk)->user_ns, CAP_NET_ADMIN);
926 netlink_update_subscriptions(struct sock *sk, unsigned int subscriptions)
928 struct netlink_sock *nlk = nlk_sk(sk);
930 if (nlk->subscriptions && !subscriptions)
931 __sk_del_bind_node(sk);
932 else if (!nlk->subscriptions && subscriptions)
933 sk_add_bind_node(sk, &nl_table[sk->sk_protocol].mc_list);
934 nlk->subscriptions = subscriptions;
937 static int netlink_realloc_groups(struct sock *sk)
939 struct netlink_sock *nlk = nlk_sk(sk);
941 unsigned long *new_groups;
944 netlink_table_grab();
946 groups = nl_table[sk->sk_protocol].groups;
947 if (!nl_table[sk->sk_protocol].registered) {
952 if (nlk->ngroups >= groups)
955 new_groups = krealloc(nlk->groups, NLGRPSZ(groups), GFP_ATOMIC);
956 if (new_groups == NULL) {
960 memset((char *)new_groups + NLGRPSZ(nlk->ngroups), 0,
961 NLGRPSZ(groups) - NLGRPSZ(nlk->ngroups));
963 nlk->groups = new_groups;
964 nlk->ngroups = groups;
966 netlink_table_ungrab();
970 static void netlink_undo_bind(int group, long unsigned int groups,
973 struct netlink_sock *nlk = nlk_sk(sk);
976 if (!nlk->netlink_unbind)
979 for (undo = 0; undo < group; undo++)
980 if (test_bit(undo, &groups))
981 nlk->netlink_unbind(sock_net(sk), undo + 1);
984 static int netlink_bind(struct socket *sock, struct sockaddr *addr,
987 struct sock *sk = sock->sk;
988 struct net *net = sock_net(sk);
989 struct netlink_sock *nlk = nlk_sk(sk);
990 struct sockaddr_nl *nladdr = (struct sockaddr_nl *)addr;
992 unsigned long groups;
995 if (addr_len < sizeof(struct sockaddr_nl))
998 if (nladdr->nl_family != AF_NETLINK)
1000 groups = nladdr->nl_groups;
1002 /* Only superuser is allowed to listen multicasts */
1004 if (!netlink_allowed(sock, NL_CFG_F_NONROOT_RECV))
1006 err = netlink_realloc_groups(sk);
1011 if (nlk->ngroups < BITS_PER_LONG)
1012 groups &= (1UL << nlk->ngroups) - 1;
1014 /* Paired with WRITE_ONCE() in netlink_insert() */
1015 bound = READ_ONCE(nlk->bound);
1017 /* Ensure nlk->portid is up-to-date. */
1020 if (nladdr->nl_pid != nlk->portid)
1024 if (nlk->netlink_bind && groups) {
1027 /* nl_groups is a u32, so cap the maximum groups we can bind */
1028 for (group = 0; group < BITS_PER_TYPE(u32); group++) {
1029 if (!test_bit(group, &groups))
1031 err = nlk->netlink_bind(net, group + 1);
1034 netlink_undo_bind(group, groups, sk);
1039 /* No need for barriers here as we return to user-space without
1040 * using any of the bound attributes.
1042 netlink_lock_table();
1044 err = nladdr->nl_pid ?
1045 netlink_insert(sk, nladdr->nl_pid) :
1046 netlink_autobind(sock);
1048 netlink_undo_bind(BITS_PER_TYPE(u32), groups, sk);
1053 if (!groups && (nlk->groups == NULL || !(u32)nlk->groups[0]))
1055 netlink_unlock_table();
1057 netlink_table_grab();
1058 netlink_update_subscriptions(sk, nlk->subscriptions +
1060 hweight32(nlk->groups[0]));
1061 nlk->groups[0] = (nlk->groups[0] & ~0xffffffffUL) | groups;
1062 netlink_update_listeners(sk);
1063 netlink_table_ungrab();
1068 netlink_unlock_table();
1072 static int netlink_connect(struct socket *sock, struct sockaddr *addr,
1073 int alen, int flags)
1076 struct sock *sk = sock->sk;
1077 struct netlink_sock *nlk = nlk_sk(sk);
1078 struct sockaddr_nl *nladdr = (struct sockaddr_nl *)addr;
1080 if (alen < sizeof(addr->sa_family))
1083 if (addr->sa_family == AF_UNSPEC) {
1084 /* paired with READ_ONCE() in netlink_getsockbyportid() */
1085 WRITE_ONCE(sk->sk_state, NETLINK_UNCONNECTED);
1086 /* dst_portid and dst_group can be read locklessly */
1087 WRITE_ONCE(nlk->dst_portid, 0);
1088 WRITE_ONCE(nlk->dst_group, 0);
1091 if (addr->sa_family != AF_NETLINK)
1094 if (alen < sizeof(struct sockaddr_nl))
1097 if ((nladdr->nl_groups || nladdr->nl_pid) &&
1098 !netlink_allowed(sock, NL_CFG_F_NONROOT_SEND))
1101 /* No need for barriers here as we return to user-space without
1102 * using any of the bound attributes.
1103 * Paired with WRITE_ONCE() in netlink_insert().
1105 if (!READ_ONCE(nlk->bound))
1106 err = netlink_autobind(sock);
1109 /* paired with READ_ONCE() in netlink_getsockbyportid() */
1110 WRITE_ONCE(sk->sk_state, NETLINK_CONNECTED);
1111 /* dst_portid and dst_group can be read locklessly */
1112 WRITE_ONCE(nlk->dst_portid, nladdr->nl_pid);
1113 WRITE_ONCE(nlk->dst_group, ffs(nladdr->nl_groups));
1119 static int netlink_getname(struct socket *sock, struct sockaddr *addr,
1122 struct sock *sk = sock->sk;
1123 struct netlink_sock *nlk = nlk_sk(sk);
1124 DECLARE_SOCKADDR(struct sockaddr_nl *, nladdr, addr);
1126 nladdr->nl_family = AF_NETLINK;
1130 /* Paired with WRITE_ONCE() in netlink_connect() */
1131 nladdr->nl_pid = READ_ONCE(nlk->dst_portid);
1132 nladdr->nl_groups = netlink_group_mask(READ_ONCE(nlk->dst_group));
1134 /* Paired with WRITE_ONCE() in netlink_insert() */
1135 nladdr->nl_pid = READ_ONCE(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 /* dst_portid and sk_state can be changed in netlink_connect() */
1163 if (READ_ONCE(sock->sk_state) == NETLINK_CONNECTED &&
1164 READ_ONCE(nlk->dst_portid) != nlk_sk(ssk)->portid) {
1166 return ERR_PTR(-ECONNREFUSED);
1171 struct sock *netlink_getsockbyfilp(struct file *filp)
1173 struct inode *inode = file_inode(filp);
1176 if (!S_ISSOCK(inode->i_mode))
1177 return ERR_PTR(-ENOTSOCK);
1179 sock = SOCKET_I(inode)->sk;
1180 if (sock->sk_family != AF_NETLINK)
1181 return ERR_PTR(-EINVAL);
1187 static struct sk_buff *netlink_alloc_large_skb(unsigned int size,
1190 struct sk_buff *skb;
1193 if (size <= NLMSG_GOODSIZE || broadcast)
1194 return alloc_skb(size, GFP_KERNEL);
1196 size = SKB_DATA_ALIGN(size) +
1197 SKB_DATA_ALIGN(sizeof(struct skb_shared_info));
1199 data = vmalloc(size);
1203 skb = __build_skb(data, size);
1207 skb->destructor = netlink_skb_destructor;
1213 * Attach a skb to a netlink socket.
1214 * The caller must hold a reference to the destination socket. On error, the
1215 * reference is dropped. The skb is not send to the destination, just all
1216 * all error checks are performed and memory in the queue is reserved.
1218 * < 0: error. skb freed, reference to sock dropped.
1220 * 1: repeat lookup - reference dropped while waiting for socket memory.
1222 int netlink_attachskb(struct sock *sk, struct sk_buff *skb,
1223 long *timeo, struct sock *ssk)
1225 struct netlink_sock *nlk;
1229 if ((atomic_read(&sk->sk_rmem_alloc) > sk->sk_rcvbuf ||
1230 test_bit(NETLINK_S_CONGESTED, &nlk->state))) {
1231 DECLARE_WAITQUEUE(wait, current);
1233 if (!ssk || netlink_is_kernel(ssk))
1234 netlink_overrun(sk);
1240 __set_current_state(TASK_INTERRUPTIBLE);
1241 add_wait_queue(&nlk->wait, &wait);
1243 if ((atomic_read(&sk->sk_rmem_alloc) > sk->sk_rcvbuf ||
1244 test_bit(NETLINK_S_CONGESTED, &nlk->state)) &&
1245 !sock_flag(sk, SOCK_DEAD))
1246 *timeo = schedule_timeout(*timeo);
1248 __set_current_state(TASK_RUNNING);
1249 remove_wait_queue(&nlk->wait, &wait);
1252 if (signal_pending(current)) {
1254 return sock_intr_errno(*timeo);
1258 netlink_skb_set_owner_r(skb, sk);
1262 static int __netlink_sendskb(struct sock *sk, struct sk_buff *skb)
1266 netlink_deliver_tap(sock_net(sk), skb);
1268 skb_queue_tail(&sk->sk_receive_queue, skb);
1269 sk->sk_data_ready(sk);
1273 int netlink_sendskb(struct sock *sk, struct sk_buff *skb)
1275 int len = __netlink_sendskb(sk, skb);
1281 void netlink_detachskb(struct sock *sk, struct sk_buff *skb)
1287 static struct sk_buff *netlink_trim(struct sk_buff *skb, gfp_t allocation)
1291 WARN_ON(skb->sk != NULL);
1292 delta = skb->end - skb->tail;
1293 if (is_vmalloc_addr(skb->head) || delta * 2 < skb->truesize)
1296 if (skb_shared(skb)) {
1297 struct sk_buff *nskb = skb_clone(skb, allocation);
1304 pskb_expand_head(skb, 0, -delta,
1305 (allocation & ~__GFP_DIRECT_RECLAIM) |
1306 __GFP_NOWARN | __GFP_NORETRY);
1310 static int netlink_unicast_kernel(struct sock *sk, struct sk_buff *skb,
1314 struct netlink_sock *nlk = nlk_sk(sk);
1316 ret = -ECONNREFUSED;
1317 if (nlk->netlink_rcv != NULL) {
1319 netlink_skb_set_owner_r(skb, sk);
1320 NETLINK_CB(skb).sk = ssk;
1321 netlink_deliver_tap_kernel(sk, ssk, skb);
1322 nlk->netlink_rcv(skb);
1331 int netlink_unicast(struct sock *ssk, struct sk_buff *skb,
1332 u32 portid, int nonblock)
1338 skb = netlink_trim(skb, gfp_any());
1340 timeo = sock_sndtimeo(ssk, nonblock);
1342 sk = netlink_getsockbyportid(ssk, portid);
1347 if (netlink_is_kernel(sk))
1348 return netlink_unicast_kernel(sk, skb, ssk);
1350 if (sk_filter(sk, skb)) {
1357 err = netlink_attachskb(sk, skb, &timeo, ssk);
1363 return netlink_sendskb(sk, skb);
1365 EXPORT_SYMBOL(netlink_unicast);
1367 int netlink_has_listeners(struct sock *sk, unsigned int group)
1370 struct listeners *listeners;
1372 BUG_ON(!netlink_is_kernel(sk));
1375 listeners = rcu_dereference(nl_table[sk->sk_protocol].listeners);
1377 if (listeners && group - 1 < nl_table[sk->sk_protocol].groups)
1378 res = test_bit(group - 1, listeners->masks);
1384 EXPORT_SYMBOL_GPL(netlink_has_listeners);
1386 bool netlink_strict_get_check(struct sk_buff *skb)
1388 const struct netlink_sock *nlk = nlk_sk(NETLINK_CB(skb).sk);
1390 return nlk->flags & NETLINK_F_STRICT_CHK;
1392 EXPORT_SYMBOL_GPL(netlink_strict_get_check);
1394 static int netlink_broadcast_deliver(struct sock *sk, struct sk_buff *skb)
1396 struct netlink_sock *nlk = nlk_sk(sk);
1398 if (atomic_read(&sk->sk_rmem_alloc) <= sk->sk_rcvbuf &&
1399 !test_bit(NETLINK_S_CONGESTED, &nlk->state)) {
1400 netlink_skb_set_owner_r(skb, sk);
1401 __netlink_sendskb(sk, skb);
1402 return atomic_read(&sk->sk_rmem_alloc) > (sk->sk_rcvbuf >> 1);
1407 struct netlink_broadcast_data {
1408 struct sock *exclude_sk;
1413 int delivery_failure;
1417 struct sk_buff *skb, *skb2;
1418 int (*tx_filter)(struct sock *dsk, struct sk_buff *skb, void *data);
1422 static void do_one_broadcast(struct sock *sk,
1423 struct netlink_broadcast_data *p)
1425 struct netlink_sock *nlk = nlk_sk(sk);
1428 if (p->exclude_sk == sk)
1431 if (nlk->portid == p->portid || p->group - 1 >= nlk->ngroups ||
1432 !test_bit(p->group - 1, nlk->groups))
1435 if (!net_eq(sock_net(sk), p->net)) {
1436 if (!(nlk->flags & NETLINK_F_LISTEN_ALL_NSID))
1439 if (!peernet_has_id(sock_net(sk), p->net))
1442 if (!file_ns_capable(sk->sk_socket->file, p->net->user_ns,
1448 netlink_overrun(sk);
1453 if (p->skb2 == NULL) {
1454 if (skb_shared(p->skb)) {
1455 p->skb2 = skb_clone(p->skb, p->allocation);
1457 p->skb2 = skb_get(p->skb);
1459 * skb ownership may have been set when
1460 * delivered to a previous socket.
1462 skb_orphan(p->skb2);
1465 if (p->skb2 == NULL) {
1466 netlink_overrun(sk);
1467 /* Clone failed. Notify ALL listeners. */
1469 if (nlk->flags & NETLINK_F_BROADCAST_SEND_ERROR)
1470 p->delivery_failure = 1;
1473 if (p->tx_filter && p->tx_filter(sk, p->skb2, p->tx_data)) {
1478 if (sk_filter(sk, p->skb2)) {
1483 NETLINK_CB(p->skb2).nsid = peernet2id(sock_net(sk), p->net);
1484 if (NETLINK_CB(p->skb2).nsid != NETNSA_NSID_NOT_ASSIGNED)
1485 NETLINK_CB(p->skb2).nsid_is_set = true;
1486 val = netlink_broadcast_deliver(sk, p->skb2);
1488 netlink_overrun(sk);
1489 if (nlk->flags & NETLINK_F_BROADCAST_SEND_ERROR)
1490 p->delivery_failure = 1;
1492 p->congested |= val;
1500 int netlink_broadcast_filtered(struct sock *ssk, struct sk_buff *skb, u32 portid,
1501 u32 group, gfp_t allocation,
1502 int (*filter)(struct sock *dsk, struct sk_buff *skb, void *data),
1505 struct net *net = sock_net(ssk);
1506 struct netlink_broadcast_data info;
1509 skb = netlink_trim(skb, allocation);
1511 info.exclude_sk = ssk;
1513 info.portid = portid;
1516 info.delivery_failure = 0;
1519 info.allocation = allocation;
1522 info.tx_filter = filter;
1523 info.tx_data = filter_data;
1525 /* While we sleep in clone, do not allow to change socket list */
1527 netlink_lock_table();
1529 sk_for_each_bound(sk, &nl_table[ssk->sk_protocol].mc_list)
1530 do_one_broadcast(sk, &info);
1534 netlink_unlock_table();
1536 if (info.delivery_failure) {
1537 kfree_skb(info.skb2);
1540 consume_skb(info.skb2);
1542 if (info.delivered) {
1543 if (info.congested && gfpflags_allow_blocking(allocation))
1549 EXPORT_SYMBOL(netlink_broadcast_filtered);
1551 int netlink_broadcast(struct sock *ssk, struct sk_buff *skb, u32 portid,
1552 u32 group, gfp_t allocation)
1554 return netlink_broadcast_filtered(ssk, skb, portid, group, allocation,
1557 EXPORT_SYMBOL(netlink_broadcast);
1559 struct netlink_set_err_data {
1560 struct sock *exclude_sk;
1566 static int do_one_set_err(struct sock *sk, struct netlink_set_err_data *p)
1568 struct netlink_sock *nlk = nlk_sk(sk);
1571 if (sk == p->exclude_sk)
1574 if (!net_eq(sock_net(sk), sock_net(p->exclude_sk)))
1577 if (nlk->portid == p->portid || p->group - 1 >= nlk->ngroups ||
1578 !test_bit(p->group - 1, nlk->groups))
1581 if (p->code == ENOBUFS && nlk->flags & NETLINK_F_RECV_NO_ENOBUFS) {
1586 sk->sk_err = p->code;
1587 sk->sk_error_report(sk);
1593 * netlink_set_err - report error to broadcast listeners
1594 * @ssk: the kernel netlink socket, as returned by netlink_kernel_create()
1595 * @portid: the PORTID of a process that we want to skip (if any)
1596 * @group: the broadcast group that will notice the error
1597 * @code: error code, must be negative (as usual in kernelspace)
1599 * This function returns the number of broadcast listeners that have set the
1600 * NETLINK_NO_ENOBUFS socket option.
1602 int netlink_set_err(struct sock *ssk, u32 portid, u32 group, int code)
1604 struct netlink_set_err_data info;
1605 unsigned long flags;
1609 info.exclude_sk = ssk;
1610 info.portid = portid;
1612 /* sk->sk_err wants a positive error value */
1615 read_lock_irqsave(&nl_table_lock, flags);
1617 sk_for_each_bound(sk, &nl_table[ssk->sk_protocol].mc_list)
1618 ret += do_one_set_err(sk, &info);
1620 read_unlock_irqrestore(&nl_table_lock, flags);
1623 EXPORT_SYMBOL(netlink_set_err);
1625 /* must be called with netlink table grabbed */
1626 static void netlink_update_socket_mc(struct netlink_sock *nlk,
1630 int old, new = !!is_new, subscriptions;
1632 old = test_bit(group - 1, nlk->groups);
1633 subscriptions = nlk->subscriptions - old + new;
1635 __set_bit(group - 1, nlk->groups);
1637 __clear_bit(group - 1, nlk->groups);
1638 netlink_update_subscriptions(&nlk->sk, subscriptions);
1639 netlink_update_listeners(&nlk->sk);
1642 static int netlink_setsockopt(struct socket *sock, int level, int optname,
1643 sockptr_t optval, unsigned int optlen)
1645 struct sock *sk = sock->sk;
1646 struct netlink_sock *nlk = nlk_sk(sk);
1647 unsigned int val = 0;
1650 if (level != SOL_NETLINK)
1651 return -ENOPROTOOPT;
1653 if (optlen >= sizeof(int) &&
1654 copy_from_sockptr(&val, optval, sizeof(val)))
1658 case NETLINK_PKTINFO:
1660 nlk->flags |= NETLINK_F_RECV_PKTINFO;
1662 nlk->flags &= ~NETLINK_F_RECV_PKTINFO;
1665 case NETLINK_ADD_MEMBERSHIP:
1666 case NETLINK_DROP_MEMBERSHIP: {
1667 if (!netlink_allowed(sock, NL_CFG_F_NONROOT_RECV))
1669 err = netlink_realloc_groups(sk);
1672 if (!val || val - 1 >= nlk->ngroups)
1674 if (optname == NETLINK_ADD_MEMBERSHIP && nlk->netlink_bind) {
1675 err = nlk->netlink_bind(sock_net(sk), val);
1679 netlink_table_grab();
1680 netlink_update_socket_mc(nlk, val,
1681 optname == NETLINK_ADD_MEMBERSHIP);
1682 netlink_table_ungrab();
1683 if (optname == NETLINK_DROP_MEMBERSHIP && nlk->netlink_unbind)
1684 nlk->netlink_unbind(sock_net(sk), val);
1689 case NETLINK_BROADCAST_ERROR:
1691 nlk->flags |= NETLINK_F_BROADCAST_SEND_ERROR;
1693 nlk->flags &= ~NETLINK_F_BROADCAST_SEND_ERROR;
1696 case NETLINK_NO_ENOBUFS:
1698 nlk->flags |= NETLINK_F_RECV_NO_ENOBUFS;
1699 clear_bit(NETLINK_S_CONGESTED, &nlk->state);
1700 wake_up_interruptible(&nlk->wait);
1702 nlk->flags &= ~NETLINK_F_RECV_NO_ENOBUFS;
1706 case NETLINK_LISTEN_ALL_NSID:
1707 if (!ns_capable(sock_net(sk)->user_ns, CAP_NET_BROADCAST))
1711 nlk->flags |= NETLINK_F_LISTEN_ALL_NSID;
1713 nlk->flags &= ~NETLINK_F_LISTEN_ALL_NSID;
1716 case NETLINK_CAP_ACK:
1718 nlk->flags |= NETLINK_F_CAP_ACK;
1720 nlk->flags &= ~NETLINK_F_CAP_ACK;
1723 case NETLINK_EXT_ACK:
1725 nlk->flags |= NETLINK_F_EXT_ACK;
1727 nlk->flags &= ~NETLINK_F_EXT_ACK;
1730 case NETLINK_GET_STRICT_CHK:
1732 nlk->flags |= NETLINK_F_STRICT_CHK;
1734 nlk->flags &= ~NETLINK_F_STRICT_CHK;
1743 static int netlink_getsockopt(struct socket *sock, int level, int optname,
1744 char __user *optval, int __user *optlen)
1746 struct sock *sk = sock->sk;
1747 struct netlink_sock *nlk = nlk_sk(sk);
1751 if (level != SOL_NETLINK)
1752 return -ENOPROTOOPT;
1754 if (get_user(len, optlen))
1760 case NETLINK_PKTINFO:
1761 flag = NETLINK_F_RECV_PKTINFO;
1763 case NETLINK_BROADCAST_ERROR:
1764 flag = NETLINK_F_BROADCAST_SEND_ERROR;
1766 case NETLINK_NO_ENOBUFS:
1767 flag = NETLINK_F_RECV_NO_ENOBUFS;
1769 case NETLINK_LIST_MEMBERSHIPS: {
1770 int pos, idx, shift, err = 0;
1772 netlink_lock_table();
1773 for (pos = 0; pos * 8 < nlk->ngroups; pos += sizeof(u32)) {
1774 if (len - pos < sizeof(u32))
1777 idx = pos / sizeof(unsigned long);
1778 shift = (pos % sizeof(unsigned long)) * 8;
1779 if (put_user((u32)(nlk->groups[idx] >> shift),
1780 (u32 __user *)(optval + pos))) {
1785 if (put_user(ALIGN(BITS_TO_BYTES(nlk->ngroups), sizeof(u32)), optlen))
1787 netlink_unlock_table();
1790 case NETLINK_CAP_ACK:
1791 flag = NETLINK_F_CAP_ACK;
1793 case NETLINK_EXT_ACK:
1794 flag = NETLINK_F_EXT_ACK;
1796 case NETLINK_GET_STRICT_CHK:
1797 flag = NETLINK_F_STRICT_CHK;
1800 return -ENOPROTOOPT;
1803 if (len < sizeof(int))
1807 val = nlk->flags & flag ? 1 : 0;
1809 if (put_user(len, optlen) ||
1810 copy_to_user(optval, &val, len))
1816 static void netlink_cmsg_recv_pktinfo(struct msghdr *msg, struct sk_buff *skb)
1818 struct nl_pktinfo info;
1820 info.group = NETLINK_CB(skb).dst_group;
1821 put_cmsg(msg, SOL_NETLINK, NETLINK_PKTINFO, sizeof(info), &info);
1824 static void netlink_cmsg_listen_all_nsid(struct sock *sk, struct msghdr *msg,
1825 struct sk_buff *skb)
1827 if (!NETLINK_CB(skb).nsid_is_set)
1830 put_cmsg(msg, SOL_NETLINK, NETLINK_LISTEN_ALL_NSID, sizeof(int),
1831 &NETLINK_CB(skb).nsid);
1834 static int netlink_sendmsg(struct socket *sock, struct msghdr *msg, size_t len)
1836 struct sock *sk = sock->sk;
1837 struct netlink_sock *nlk = nlk_sk(sk);
1838 DECLARE_SOCKADDR(struct sockaddr_nl *, addr, msg->msg_name);
1841 struct sk_buff *skb;
1843 struct scm_cookie scm;
1844 u32 netlink_skb_flags = 0;
1846 if (msg->msg_flags & MSG_OOB)
1850 pr_warn_once("Zero length message leads to an empty skb\n");
1854 err = scm_send(sock, msg, &scm, true);
1858 if (msg->msg_namelen) {
1860 if (msg->msg_namelen < sizeof(struct sockaddr_nl))
1862 if (addr->nl_family != AF_NETLINK)
1864 dst_portid = addr->nl_pid;
1865 dst_group = ffs(addr->nl_groups);
1867 if ((dst_group || dst_portid) &&
1868 !netlink_allowed(sock, NL_CFG_F_NONROOT_SEND))
1870 netlink_skb_flags |= NETLINK_SKB_DST;
1872 /* Paired with WRITE_ONCE() in netlink_connect() */
1873 dst_portid = READ_ONCE(nlk->dst_portid);
1874 dst_group = READ_ONCE(nlk->dst_group);
1877 /* Paired with WRITE_ONCE() in netlink_insert() */
1878 if (!READ_ONCE(nlk->bound)) {
1879 err = netlink_autobind(sock);
1883 /* Ensure nlk is hashed and visible. */
1888 if (len > sk->sk_sndbuf - 32)
1891 skb = netlink_alloc_large_skb(len, dst_group);
1895 NETLINK_CB(skb).portid = nlk->portid;
1896 NETLINK_CB(skb).dst_group = dst_group;
1897 NETLINK_CB(skb).creds = scm.creds;
1898 NETLINK_CB(skb).flags = netlink_skb_flags;
1901 if (memcpy_from_msg(skb_put(skb, len), msg, len)) {
1906 err = security_netlink_send(sk, skb);
1913 refcount_inc(&skb->users);
1914 netlink_broadcast(sk, skb, dst_portid, dst_group, GFP_KERNEL);
1916 err = netlink_unicast(sk, skb, dst_portid, msg->msg_flags & MSG_DONTWAIT);
1923 static int netlink_recvmsg(struct socket *sock, struct msghdr *msg, size_t len,
1926 struct scm_cookie scm;
1927 struct sock *sk = sock->sk;
1928 struct netlink_sock *nlk = nlk_sk(sk);
1929 int noblock = flags & MSG_DONTWAIT;
1931 struct sk_buff *skb, *data_skb;
1934 if (flags & MSG_OOB)
1939 skb = skb_recv_datagram(sk, flags, noblock, &err);
1945 #ifdef CONFIG_COMPAT_NETLINK_MESSAGES
1946 if (unlikely(skb_shinfo(skb)->frag_list)) {
1948 * If this skb has a frag_list, then here that means that we
1949 * will have to use the frag_list skb's data for compat tasks
1950 * and the regular skb's data for normal (non-compat) tasks.
1952 * If we need to send the compat skb, assign it to the
1953 * 'data_skb' variable so that it will be used below for data
1954 * copying. We keep 'skb' for everything else, including
1955 * freeing both later.
1957 if (flags & MSG_CMSG_COMPAT)
1958 data_skb = skb_shinfo(skb)->frag_list;
1962 /* Record the max length of recvmsg() calls for future allocations */
1963 nlk->max_recvmsg_len = max(nlk->max_recvmsg_len, len);
1964 nlk->max_recvmsg_len = min_t(size_t, nlk->max_recvmsg_len,
1965 SKB_WITH_OVERHEAD(32768));
1967 copied = data_skb->len;
1969 msg->msg_flags |= MSG_TRUNC;
1973 err = skb_copy_datagram_msg(data_skb, 0, msg, copied);
1975 if (msg->msg_name) {
1976 DECLARE_SOCKADDR(struct sockaddr_nl *, addr, msg->msg_name);
1977 addr->nl_family = AF_NETLINK;
1979 addr->nl_pid = NETLINK_CB(skb).portid;
1980 addr->nl_groups = netlink_group_mask(NETLINK_CB(skb).dst_group);
1981 msg->msg_namelen = sizeof(*addr);
1984 if (nlk->flags & NETLINK_F_RECV_PKTINFO)
1985 netlink_cmsg_recv_pktinfo(msg, skb);
1986 if (nlk->flags & NETLINK_F_LISTEN_ALL_NSID)
1987 netlink_cmsg_listen_all_nsid(sk, msg, skb);
1989 memset(&scm, 0, sizeof(scm));
1990 scm.creds = *NETLINK_CREDS(skb);
1991 if (flags & MSG_TRUNC)
1992 copied = data_skb->len;
1994 skb_free_datagram(sk, skb);
1996 if (READ_ONCE(nlk->cb_running) &&
1997 atomic_read(&sk->sk_rmem_alloc) <= sk->sk_rcvbuf / 2) {
1998 ret = netlink_dump(sk);
2001 sk->sk_error_report(sk);
2005 scm_recv(sock, msg, &scm, flags);
2007 netlink_rcv_wake(sk);
2008 return err ? : copied;
2011 static void netlink_data_ready(struct sock *sk)
2017 * We export these functions to other modules. They provide a
2018 * complete set of kernel non-blocking support for message
2023 __netlink_kernel_create(struct net *net, int unit, struct module *module,
2024 struct netlink_kernel_cfg *cfg)
2026 struct socket *sock;
2028 struct netlink_sock *nlk;
2029 struct listeners *listeners = NULL;
2030 struct mutex *cb_mutex = cfg ? cfg->cb_mutex : NULL;
2031 unsigned int groups;
2035 if (unit < 0 || unit >= MAX_LINKS)
2038 if (sock_create_lite(PF_NETLINK, SOCK_DGRAM, unit, &sock))
2041 if (__netlink_create(net, sock, cb_mutex, unit, 1) < 0)
2042 goto out_sock_release_nosk;
2046 if (!cfg || cfg->groups < 32)
2049 groups = cfg->groups;
2051 listeners = kzalloc(sizeof(*listeners) + NLGRPSZ(groups), GFP_KERNEL);
2053 goto out_sock_release;
2055 sk->sk_data_ready = netlink_data_ready;
2056 if (cfg && cfg->input)
2057 nlk_sk(sk)->netlink_rcv = cfg->input;
2059 if (netlink_insert(sk, 0))
2060 goto out_sock_release;
2063 nlk->flags |= NETLINK_F_KERNEL_SOCKET;
2065 netlink_table_grab();
2066 if (!nl_table[unit].registered) {
2067 nl_table[unit].groups = groups;
2068 rcu_assign_pointer(nl_table[unit].listeners, listeners);
2069 nl_table[unit].cb_mutex = cb_mutex;
2070 nl_table[unit].module = module;
2072 nl_table[unit].bind = cfg->bind;
2073 nl_table[unit].unbind = cfg->unbind;
2074 nl_table[unit].flags = cfg->flags;
2076 nl_table[unit].compare = cfg->compare;
2078 nl_table[unit].registered = 1;
2081 nl_table[unit].registered++;
2083 netlink_table_ungrab();
2088 netlink_kernel_release(sk);
2091 out_sock_release_nosk:
2095 EXPORT_SYMBOL(__netlink_kernel_create);
2098 netlink_kernel_release(struct sock *sk)
2100 if (sk == NULL || sk->sk_socket == NULL)
2103 sock_release(sk->sk_socket);
2105 EXPORT_SYMBOL(netlink_kernel_release);
2107 int __netlink_change_ngroups(struct sock *sk, unsigned int groups)
2109 struct listeners *new, *old;
2110 struct netlink_table *tbl = &nl_table[sk->sk_protocol];
2115 if (NLGRPSZ(tbl->groups) < NLGRPSZ(groups)) {
2116 new = kzalloc(sizeof(*new) + NLGRPSZ(groups), GFP_ATOMIC);
2119 old = nl_deref_protected(tbl->listeners);
2120 memcpy(new->masks, old->masks, NLGRPSZ(tbl->groups));
2121 rcu_assign_pointer(tbl->listeners, new);
2123 kfree_rcu(old, rcu);
2125 tbl->groups = groups;
2131 * netlink_change_ngroups - change number of multicast groups
2133 * This changes the number of multicast groups that are available
2134 * on a certain netlink family. Note that it is not possible to
2135 * change the number of groups to below 32. Also note that it does
2136 * not implicitly call netlink_clear_multicast_users() when the
2137 * number of groups is reduced.
2139 * @sk: The kernel netlink socket, as returned by netlink_kernel_create().
2140 * @groups: The new number of groups.
2142 int netlink_change_ngroups(struct sock *sk, unsigned int groups)
2146 netlink_table_grab();
2147 err = __netlink_change_ngroups(sk, groups);
2148 netlink_table_ungrab();
2153 void __netlink_clear_multicast_users(struct sock *ksk, unsigned int group)
2156 struct netlink_table *tbl = &nl_table[ksk->sk_protocol];
2158 sk_for_each_bound(sk, &tbl->mc_list)
2159 netlink_update_socket_mc(nlk_sk(sk), group, 0);
2163 __nlmsg_put(struct sk_buff *skb, u32 portid, u32 seq, int type, int len, int flags)
2165 struct nlmsghdr *nlh;
2166 int size = nlmsg_msg_size(len);
2168 nlh = skb_put(skb, NLMSG_ALIGN(size));
2169 nlh->nlmsg_type = type;
2170 nlh->nlmsg_len = size;
2171 nlh->nlmsg_flags = flags;
2172 nlh->nlmsg_pid = portid;
2173 nlh->nlmsg_seq = seq;
2174 if (!__builtin_constant_p(size) || NLMSG_ALIGN(size) - size != 0)
2175 memset(nlmsg_data(nlh) + len, 0, NLMSG_ALIGN(size) - size);
2178 EXPORT_SYMBOL(__nlmsg_put);
2181 * It looks a bit ugly.
2182 * It would be better to create kernel thread.
2185 static int netlink_dump_done(struct netlink_sock *nlk, struct sk_buff *skb,
2186 struct netlink_callback *cb,
2187 struct netlink_ext_ack *extack)
2189 struct nlmsghdr *nlh;
2191 nlh = nlmsg_put_answer(skb, cb, NLMSG_DONE, sizeof(nlk->dump_done_errno),
2192 NLM_F_MULTI | cb->answer_flags);
2196 nl_dump_check_consistent(cb, nlh);
2197 memcpy(nlmsg_data(nlh), &nlk->dump_done_errno, sizeof(nlk->dump_done_errno));
2199 if (extack->_msg && nlk->flags & NETLINK_F_EXT_ACK) {
2200 nlh->nlmsg_flags |= NLM_F_ACK_TLVS;
2201 if (!nla_put_string(skb, NLMSGERR_ATTR_MSG, extack->_msg))
2202 nlmsg_end(skb, nlh);
2208 static int netlink_dump(struct sock *sk)
2210 struct netlink_sock *nlk = nlk_sk(sk);
2211 struct netlink_ext_ack extack = {};
2212 struct netlink_callback *cb;
2213 struct sk_buff *skb = NULL;
2214 struct module *module;
2219 mutex_lock(nlk->cb_mutex);
2220 if (!nlk->cb_running) {
2225 if (atomic_read(&sk->sk_rmem_alloc) >= sk->sk_rcvbuf)
2228 /* NLMSG_GOODSIZE is small to avoid high order allocations being
2229 * required, but it makes sense to _attempt_ a 16K bytes allocation
2230 * to reduce number of system calls on dump operations, if user
2231 * ever provided a big enough buffer.
2234 alloc_min_size = max_t(int, cb->min_dump_alloc, NLMSG_GOODSIZE);
2236 if (alloc_min_size < nlk->max_recvmsg_len) {
2237 alloc_size = nlk->max_recvmsg_len;
2238 skb = alloc_skb(alloc_size,
2239 (GFP_KERNEL & ~__GFP_DIRECT_RECLAIM) |
2240 __GFP_NOWARN | __GFP_NORETRY);
2243 alloc_size = alloc_min_size;
2244 skb = alloc_skb(alloc_size, GFP_KERNEL);
2249 /* Trim skb to allocated size. User is expected to provide buffer as
2250 * large as max(min_dump_alloc, 16KiB (mac_recvmsg_len capped at
2251 * netlink_recvmsg())). dump will pack as many smaller messages as
2252 * could fit within the allocated skb. skb is typically allocated
2253 * with larger space than required (could be as much as near 2x the
2254 * requested size with align to next power of 2 approach). Allowing
2255 * dump to use the excess space makes it difficult for a user to have a
2256 * reasonable static buffer based on the expected largest dump of a
2257 * single netdev. The outcome is MSG_TRUNC error.
2259 skb_reserve(skb, skb_tailroom(skb) - alloc_size);
2261 /* Make sure malicious BPF programs can not read unitialized memory
2262 * from skb->head -> skb->data
2264 skb_reset_network_header(skb);
2265 skb_reset_mac_header(skb);
2267 netlink_skb_set_owner_r(skb, sk);
2269 if (nlk->dump_done_errno > 0) {
2270 cb->extack = &extack;
2271 nlk->dump_done_errno = cb->dump(skb, cb);
2275 if (nlk->dump_done_errno > 0 ||
2276 skb_tailroom(skb) < nlmsg_total_size(sizeof(nlk->dump_done_errno))) {
2277 mutex_unlock(nlk->cb_mutex);
2279 if (sk_filter(sk, skb))
2282 __netlink_sendskb(sk, skb);
2286 if (netlink_dump_done(nlk, skb, cb, &extack))
2289 #ifdef CONFIG_COMPAT_NETLINK_MESSAGES
2290 /* frag_list skb's data is used for compat tasks
2291 * and the regular skb's data for normal (non-compat) tasks.
2292 * See netlink_recvmsg().
2294 if (unlikely(skb_shinfo(skb)->frag_list)) {
2295 if (netlink_dump_done(nlk, skb_shinfo(skb)->frag_list, cb, &extack))
2300 if (sk_filter(sk, skb))
2303 __netlink_sendskb(sk, skb);
2308 WRITE_ONCE(nlk->cb_running, false);
2309 module = cb->module;
2311 mutex_unlock(nlk->cb_mutex);
2317 mutex_unlock(nlk->cb_mutex);
2322 int __netlink_dump_start(struct sock *ssk, struct sk_buff *skb,
2323 const struct nlmsghdr *nlh,
2324 struct netlink_dump_control *control)
2326 struct netlink_sock *nlk, *nlk2;
2327 struct netlink_callback *cb;
2331 refcount_inc(&skb->users);
2333 sk = netlink_lookup(sock_net(ssk), ssk->sk_protocol, NETLINK_CB(skb).portid);
2335 ret = -ECONNREFUSED;
2340 mutex_lock(nlk->cb_mutex);
2341 /* A dump is in progress... */
2342 if (nlk->cb_running) {
2346 /* add reference of module which cb->dump belongs to */
2347 if (!try_module_get(control->module)) {
2348 ret = -EPROTONOSUPPORT;
2353 memset(cb, 0, sizeof(*cb));
2354 cb->dump = control->dump;
2355 cb->done = control->done;
2357 cb->data = control->data;
2358 cb->module = control->module;
2359 cb->min_dump_alloc = control->min_dump_alloc;
2362 nlk2 = nlk_sk(NETLINK_CB(skb).sk);
2363 cb->strict_check = !!(nlk2->flags & NETLINK_F_STRICT_CHK);
2365 if (control->start) {
2366 ret = control->start(cb);
2371 WRITE_ONCE(nlk->cb_running, true);
2372 nlk->dump_done_errno = INT_MAX;
2374 mutex_unlock(nlk->cb_mutex);
2376 ret = netlink_dump(sk);
2383 /* We successfully started a dump, by returning -EINTR we
2384 * signal not to send ACK even if it was requested.
2389 module_put(control->module);
2392 mutex_unlock(nlk->cb_mutex);
2397 EXPORT_SYMBOL(__netlink_dump_start);
2399 void netlink_ack(struct sk_buff *in_skb, struct nlmsghdr *nlh, int err,
2400 const struct netlink_ext_ack *extack)
2402 struct sk_buff *skb;
2403 struct nlmsghdr *rep;
2404 struct nlmsgerr *errmsg;
2405 size_t payload = sizeof(*errmsg);
2407 struct netlink_sock *nlk = nlk_sk(NETLINK_CB(in_skb).sk);
2408 unsigned int flags = 0;
2409 bool nlk_has_extack = nlk->flags & NETLINK_F_EXT_ACK;
2411 /* Error messages get the original request appened, unless the user
2412 * requests to cap the error message, and get extra error data if
2415 if (nlk_has_extack && extack && extack->_msg)
2416 tlvlen += nla_total_size(strlen(extack->_msg) + 1);
2418 if (err && !(nlk->flags & NETLINK_F_CAP_ACK))
2419 payload += nlmsg_len(nlh);
2421 flags |= NLM_F_CAPPED;
2422 if (err && nlk_has_extack && extack && extack->bad_attr)
2423 tlvlen += nla_total_size(sizeof(u32));
2424 if (nlk_has_extack && extack && extack->cookie_len)
2425 tlvlen += nla_total_size(extack->cookie_len);
2426 if (err && nlk_has_extack && extack && extack->policy)
2427 tlvlen += netlink_policy_dump_attr_size_estimate(extack->policy);
2430 flags |= NLM_F_ACK_TLVS;
2432 skb = nlmsg_new(payload + tlvlen, GFP_KERNEL);
2434 NETLINK_CB(in_skb).sk->sk_err = ENOBUFS;
2435 NETLINK_CB(in_skb).sk->sk_error_report(NETLINK_CB(in_skb).sk);
2439 rep = __nlmsg_put(skb, NETLINK_CB(in_skb).portid, nlh->nlmsg_seq,
2440 NLMSG_ERROR, payload, flags);
2441 errmsg = nlmsg_data(rep);
2442 errmsg->error = err;
2443 memcpy(&errmsg->msg, nlh, payload > sizeof(*errmsg) ? nlh->nlmsg_len : sizeof(*nlh));
2445 if (nlk_has_extack && extack) {
2447 WARN_ON(nla_put_string(skb, NLMSGERR_ATTR_MSG,
2450 if (err && extack->bad_attr &&
2451 !WARN_ON((u8 *)extack->bad_attr < in_skb->data ||
2452 (u8 *)extack->bad_attr >= in_skb->data +
2454 WARN_ON(nla_put_u32(skb, NLMSGERR_ATTR_OFFS,
2455 (u8 *)extack->bad_attr -
2457 if (extack->cookie_len)
2458 WARN_ON(nla_put(skb, NLMSGERR_ATTR_COOKIE,
2459 extack->cookie_len, extack->cookie));
2461 netlink_policy_dump_write_attr(skb, extack->policy,
2462 NLMSGERR_ATTR_POLICY);
2465 nlmsg_end(skb, rep);
2467 netlink_unicast(in_skb->sk, skb, NETLINK_CB(in_skb).portid, MSG_DONTWAIT);
2469 EXPORT_SYMBOL(netlink_ack);
2471 int netlink_rcv_skb(struct sk_buff *skb, int (*cb)(struct sk_buff *,
2473 struct netlink_ext_ack *))
2475 struct netlink_ext_ack extack;
2476 struct nlmsghdr *nlh;
2479 while (skb->len >= nlmsg_total_size(0)) {
2482 memset(&extack, 0, sizeof(extack));
2483 nlh = nlmsg_hdr(skb);
2486 if (nlh->nlmsg_len < NLMSG_HDRLEN || skb->len < nlh->nlmsg_len)
2489 /* Only requests are handled by the kernel */
2490 if (!(nlh->nlmsg_flags & NLM_F_REQUEST))
2493 /* Skip control messages */
2494 if (nlh->nlmsg_type < NLMSG_MIN_TYPE)
2497 err = cb(skb, nlh, &extack);
2502 if (nlh->nlmsg_flags & NLM_F_ACK || err)
2503 netlink_ack(skb, nlh, err, &extack);
2506 msglen = NLMSG_ALIGN(nlh->nlmsg_len);
2507 if (msglen > skb->len)
2509 skb_pull(skb, msglen);
2514 EXPORT_SYMBOL(netlink_rcv_skb);
2517 * nlmsg_notify - send a notification netlink message
2518 * @sk: netlink socket to use
2519 * @skb: notification message
2520 * @portid: destination netlink portid for reports or 0
2521 * @group: destination multicast group or 0
2522 * @report: 1 to report back, 0 to disable
2523 * @flags: allocation flags
2525 int nlmsg_notify(struct sock *sk, struct sk_buff *skb, u32 portid,
2526 unsigned int group, int report, gfp_t flags)
2531 int exclude_portid = 0;
2534 refcount_inc(&skb->users);
2535 exclude_portid = portid;
2538 /* errors reported via destination sk->sk_err, but propagate
2539 * delivery errors if NETLINK_BROADCAST_ERROR flag is set */
2540 err = nlmsg_multicast(sk, skb, exclude_portid, group, flags);
2548 err2 = nlmsg_unicast(sk, skb, portid);
2555 EXPORT_SYMBOL(nlmsg_notify);
2557 #ifdef CONFIG_PROC_FS
2558 struct nl_seq_iter {
2559 struct seq_net_private p;
2560 struct rhashtable_iter hti;
2564 static void netlink_walk_start(struct nl_seq_iter *iter)
2566 rhashtable_walk_enter(&nl_table[iter->link].hash, &iter->hti);
2567 rhashtable_walk_start(&iter->hti);
2570 static void netlink_walk_stop(struct nl_seq_iter *iter)
2572 rhashtable_walk_stop(&iter->hti);
2573 rhashtable_walk_exit(&iter->hti);
2576 static void *__netlink_seq_next(struct seq_file *seq)
2578 struct nl_seq_iter *iter = seq->private;
2579 struct netlink_sock *nlk;
2583 nlk = rhashtable_walk_next(&iter->hti);
2586 if (PTR_ERR(nlk) == -EAGAIN)
2595 netlink_walk_stop(iter);
2596 if (++iter->link >= MAX_LINKS)
2599 netlink_walk_start(iter);
2601 } while (sock_net(&nlk->sk) != seq_file_net(seq));
2606 static void *netlink_seq_start(struct seq_file *seq, loff_t *posp)
2609 struct nl_seq_iter *iter = seq->private;
2610 void *obj = SEQ_START_TOKEN;
2615 netlink_walk_start(iter);
2617 for (pos = *posp; pos && obj && !IS_ERR(obj); pos--)
2618 obj = __netlink_seq_next(seq);
2623 static void *netlink_seq_next(struct seq_file *seq, void *v, loff_t *pos)
2626 return __netlink_seq_next(seq);
2629 static void netlink_native_seq_stop(struct seq_file *seq, void *v)
2631 struct nl_seq_iter *iter = seq->private;
2633 if (iter->link >= MAX_LINKS)
2636 netlink_walk_stop(iter);
2640 static int netlink_native_seq_show(struct seq_file *seq, void *v)
2642 if (v == SEQ_START_TOKEN) {
2644 "sk Eth Pid Groups "
2645 "Rmem Wmem Dump Locks Drops Inode\n");
2648 struct netlink_sock *nlk = nlk_sk(s);
2650 seq_printf(seq, "%pK %-3d %-10u %08x %-8d %-8d %-5d %-8d %-8u %-8lu\n",
2654 nlk->groups ? (u32)nlk->groups[0] : 0,
2655 sk_rmem_alloc_get(s),
2656 sk_wmem_alloc_get(s),
2657 READ_ONCE(nlk->cb_running),
2658 refcount_read(&s->sk_refcnt),
2659 atomic_read(&s->sk_drops),
2667 #ifdef CONFIG_BPF_SYSCALL
2668 struct bpf_iter__netlink {
2669 __bpf_md_ptr(struct bpf_iter_meta *, meta);
2670 __bpf_md_ptr(struct netlink_sock *, sk);
2673 DEFINE_BPF_ITER_FUNC(netlink, struct bpf_iter_meta *meta, struct netlink_sock *sk)
2675 static int netlink_prog_seq_show(struct bpf_prog *prog,
2676 struct bpf_iter_meta *meta,
2679 struct bpf_iter__netlink ctx;
2681 meta->seq_num--; /* skip SEQ_START_TOKEN */
2683 ctx.sk = nlk_sk((struct sock *)v);
2684 return bpf_iter_run_prog(prog, &ctx);
2687 static int netlink_seq_show(struct seq_file *seq, void *v)
2689 struct bpf_iter_meta meta;
2690 struct bpf_prog *prog;
2693 prog = bpf_iter_get_info(&meta, false);
2695 return netlink_native_seq_show(seq, v);
2697 if (v != SEQ_START_TOKEN)
2698 return netlink_prog_seq_show(prog, &meta, v);
2703 static void netlink_seq_stop(struct seq_file *seq, void *v)
2705 struct bpf_iter_meta meta;
2706 struct bpf_prog *prog;
2710 prog = bpf_iter_get_info(&meta, true);
2712 (void)netlink_prog_seq_show(prog, &meta, v);
2715 netlink_native_seq_stop(seq, v);
2718 static int netlink_seq_show(struct seq_file *seq, void *v)
2720 return netlink_native_seq_show(seq, v);
2723 static void netlink_seq_stop(struct seq_file *seq, void *v)
2725 netlink_native_seq_stop(seq, v);
2729 static const struct seq_operations netlink_seq_ops = {
2730 .start = netlink_seq_start,
2731 .next = netlink_seq_next,
2732 .stop = netlink_seq_stop,
2733 .show = netlink_seq_show,
2737 int netlink_register_notifier(struct notifier_block *nb)
2739 return blocking_notifier_chain_register(&netlink_chain, nb);
2741 EXPORT_SYMBOL(netlink_register_notifier);
2743 int netlink_unregister_notifier(struct notifier_block *nb)
2745 return blocking_notifier_chain_unregister(&netlink_chain, nb);
2747 EXPORT_SYMBOL(netlink_unregister_notifier);
2749 static const struct proto_ops netlink_ops = {
2750 .family = PF_NETLINK,
2751 .owner = THIS_MODULE,
2752 .release = netlink_release,
2753 .bind = netlink_bind,
2754 .connect = netlink_connect,
2755 .socketpair = sock_no_socketpair,
2756 .accept = sock_no_accept,
2757 .getname = netlink_getname,
2758 .poll = datagram_poll,
2759 .ioctl = netlink_ioctl,
2760 .listen = sock_no_listen,
2761 .shutdown = sock_no_shutdown,
2762 .setsockopt = netlink_setsockopt,
2763 .getsockopt = netlink_getsockopt,
2764 .sendmsg = netlink_sendmsg,
2765 .recvmsg = netlink_recvmsg,
2766 .mmap = sock_no_mmap,
2767 .sendpage = sock_no_sendpage,
2770 static const struct net_proto_family netlink_family_ops = {
2771 .family = PF_NETLINK,
2772 .create = netlink_create,
2773 .owner = THIS_MODULE, /* for consistency 8) */
2776 static int __net_init netlink_net_init(struct net *net)
2778 #ifdef CONFIG_PROC_FS
2779 if (!proc_create_net("netlink", 0, net->proc_net, &netlink_seq_ops,
2780 sizeof(struct nl_seq_iter)))
2786 static void __net_exit netlink_net_exit(struct net *net)
2788 #ifdef CONFIG_PROC_FS
2789 remove_proc_entry("netlink", net->proc_net);
2793 static void __init netlink_add_usersock_entry(void)
2795 struct listeners *listeners;
2798 listeners = kzalloc(sizeof(*listeners) + NLGRPSZ(groups), GFP_KERNEL);
2800 panic("netlink_add_usersock_entry: Cannot allocate listeners\n");
2802 netlink_table_grab();
2804 nl_table[NETLINK_USERSOCK].groups = groups;
2805 rcu_assign_pointer(nl_table[NETLINK_USERSOCK].listeners, listeners);
2806 nl_table[NETLINK_USERSOCK].module = THIS_MODULE;
2807 nl_table[NETLINK_USERSOCK].registered = 1;
2808 nl_table[NETLINK_USERSOCK].flags = NL_CFG_F_NONROOT_SEND;
2810 netlink_table_ungrab();
2813 static struct pernet_operations __net_initdata netlink_net_ops = {
2814 .init = netlink_net_init,
2815 .exit = netlink_net_exit,
2818 static inline u32 netlink_hash(const void *data, u32 len, u32 seed)
2820 const struct netlink_sock *nlk = data;
2821 struct netlink_compare_arg arg;
2823 netlink_compare_arg_init(&arg, sock_net(&nlk->sk), nlk->portid);
2824 return jhash2((u32 *)&arg, netlink_compare_arg_len / sizeof(u32), seed);
2827 static const struct rhashtable_params netlink_rhashtable_params = {
2828 .head_offset = offsetof(struct netlink_sock, node),
2829 .key_len = netlink_compare_arg_len,
2830 .obj_hashfn = netlink_hash,
2831 .obj_cmpfn = netlink_compare,
2832 .automatic_shrinking = true,
2835 #if defined(CONFIG_BPF_SYSCALL) && defined(CONFIG_PROC_FS)
2836 BTF_ID_LIST(btf_netlink_sock_id)
2837 BTF_ID(struct, netlink_sock)
2839 static const struct bpf_iter_seq_info netlink_seq_info = {
2840 .seq_ops = &netlink_seq_ops,
2841 .init_seq_private = bpf_iter_init_seq_net,
2842 .fini_seq_private = bpf_iter_fini_seq_net,
2843 .seq_priv_size = sizeof(struct nl_seq_iter),
2846 static struct bpf_iter_reg netlink_reg_info = {
2847 .target = "netlink",
2848 .ctx_arg_info_size = 1,
2850 { offsetof(struct bpf_iter__netlink, sk),
2851 PTR_TO_BTF_ID_OR_NULL },
2853 .seq_info = &netlink_seq_info,
2856 static int __init bpf_iter_register(void)
2858 netlink_reg_info.ctx_arg_info[0].btf_id = *btf_netlink_sock_id;
2859 return bpf_iter_reg_target(&netlink_reg_info);
2863 static int __init netlink_proto_init(void)
2866 int err = proto_register(&netlink_proto, 0);
2871 #if defined(CONFIG_BPF_SYSCALL) && defined(CONFIG_PROC_FS)
2872 err = bpf_iter_register();
2877 BUILD_BUG_ON(sizeof(struct netlink_skb_parms) > sizeof_field(struct sk_buff, cb));
2879 nl_table = kcalloc(MAX_LINKS, sizeof(*nl_table), GFP_KERNEL);
2883 for (i = 0; i < MAX_LINKS; i++) {
2884 if (rhashtable_init(&nl_table[i].hash,
2885 &netlink_rhashtable_params) < 0) {
2887 rhashtable_destroy(&nl_table[i].hash);
2893 netlink_add_usersock_entry();
2895 sock_register(&netlink_family_ops);
2896 register_pernet_subsys(&netlink_net_ops);
2897 register_pernet_subsys(&netlink_tap_net_ops);
2898 /* The netlink device handler may be needed early. */
2903 panic("netlink_init: Cannot allocate nl_table\n");
2906 core_initcall(netlink_proto_init);