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 /* portid can be read locklessly from netlink_getname(). */
582 WRITE_ONCE(nlk_sk(sk)->portid, portid);
586 err = __netlink_insert(table, sk);
588 /* In case the hashtable backend returns with -EBUSY
589 * from here, it must not escape to the caller.
591 if (unlikely(err == -EBUSY))
599 /* We need to ensure that the socket is hashed and visible. */
601 /* Paired with lockless reads from netlink_bind(),
602 * netlink_connect() and netlink_sendmsg().
604 WRITE_ONCE(nlk_sk(sk)->bound, portid);
611 static void netlink_remove(struct sock *sk)
613 struct netlink_table *table;
615 table = &nl_table[sk->sk_protocol];
616 if (!rhashtable_remove_fast(&table->hash, &nlk_sk(sk)->node,
617 netlink_rhashtable_params)) {
618 WARN_ON(refcount_read(&sk->sk_refcnt) == 1);
622 netlink_table_grab();
623 if (nlk_sk(sk)->subscriptions) {
624 __sk_del_bind_node(sk);
625 netlink_update_listeners(sk);
627 if (sk->sk_protocol == NETLINK_GENERIC)
628 atomic_inc(&genl_sk_destructing_cnt);
629 netlink_table_ungrab();
632 static struct proto netlink_proto = {
634 .owner = THIS_MODULE,
635 .obj_size = sizeof(struct netlink_sock),
638 static int __netlink_create(struct net *net, struct socket *sock,
639 struct mutex *cb_mutex, int protocol,
643 struct netlink_sock *nlk;
645 sock->ops = &netlink_ops;
647 sk = sk_alloc(net, PF_NETLINK, GFP_KERNEL, &netlink_proto, kern);
651 sock_init_data(sock, sk);
655 nlk->cb_mutex = cb_mutex;
657 nlk->cb_mutex = &nlk->cb_def_mutex;
658 mutex_init(nlk->cb_mutex);
659 lockdep_set_class_and_name(nlk->cb_mutex,
660 nlk_cb_mutex_keys + protocol,
661 nlk_cb_mutex_key_strings[protocol]);
663 init_waitqueue_head(&nlk->wait);
665 sk->sk_destruct = netlink_sock_destruct;
666 sk->sk_protocol = protocol;
670 static int netlink_create(struct net *net, struct socket *sock, int protocol,
673 struct module *module = NULL;
674 struct mutex *cb_mutex;
675 struct netlink_sock *nlk;
676 int (*bind)(struct net *net, int group);
677 void (*unbind)(struct net *net, int group);
680 sock->state = SS_UNCONNECTED;
682 if (sock->type != SOCK_RAW && sock->type != SOCK_DGRAM)
683 return -ESOCKTNOSUPPORT;
685 if (protocol < 0 || protocol >= MAX_LINKS)
686 return -EPROTONOSUPPORT;
687 protocol = array_index_nospec(protocol, MAX_LINKS);
689 netlink_lock_table();
690 #ifdef CONFIG_MODULES
691 if (!nl_table[protocol].registered) {
692 netlink_unlock_table();
693 request_module("net-pf-%d-proto-%d", PF_NETLINK, protocol);
694 netlink_lock_table();
697 if (nl_table[protocol].registered &&
698 try_module_get(nl_table[protocol].module))
699 module = nl_table[protocol].module;
701 err = -EPROTONOSUPPORT;
702 cb_mutex = nl_table[protocol].cb_mutex;
703 bind = nl_table[protocol].bind;
704 unbind = nl_table[protocol].unbind;
705 netlink_unlock_table();
710 err = __netlink_create(net, sock, cb_mutex, protocol, kern);
715 sock_prot_inuse_add(net, &netlink_proto, 1);
718 nlk = nlk_sk(sock->sk);
719 nlk->module = module;
720 nlk->netlink_bind = bind;
721 nlk->netlink_unbind = unbind;
730 static void deferred_put_nlk_sk(struct rcu_head *head)
732 struct netlink_sock *nlk = container_of(head, struct netlink_sock, rcu);
733 struct sock *sk = &nlk->sk;
738 if (!refcount_dec_and_test(&sk->sk_refcnt))
741 if (nlk->cb_running && nlk->cb.done) {
742 INIT_WORK(&nlk->work, netlink_sock_destruct_work);
743 schedule_work(&nlk->work);
750 static int netlink_release(struct socket *sock)
752 struct sock *sk = sock->sk;
753 struct netlink_sock *nlk;
763 * OK. Socket is unlinked, any packets that arrive now
767 /* must not acquire netlink_table_lock in any way again before unbind
768 * and notifying genetlink is done as otherwise it might deadlock
770 if (nlk->netlink_unbind) {
773 for (i = 0; i < nlk->ngroups; i++)
774 if (test_bit(i, nlk->groups))
775 nlk->netlink_unbind(sock_net(sk), i + 1);
777 if (sk->sk_protocol == NETLINK_GENERIC &&
778 atomic_dec_return(&genl_sk_destructing_cnt) == 0)
779 wake_up(&genl_sk_destructing_waitq);
782 wake_up_interruptible_all(&nlk->wait);
784 skb_queue_purge(&sk->sk_write_queue);
786 if (nlk->portid && nlk->bound) {
787 struct netlink_notify n = {
789 .protocol = sk->sk_protocol,
790 .portid = nlk->portid,
792 blocking_notifier_call_chain(&netlink_chain,
793 NETLINK_URELEASE, &n);
796 module_put(nlk->module);
798 if (netlink_is_kernel(sk)) {
799 netlink_table_grab();
800 BUG_ON(nl_table[sk->sk_protocol].registered == 0);
801 if (--nl_table[sk->sk_protocol].registered == 0) {
802 struct listeners *old;
804 old = nl_deref_protected(nl_table[sk->sk_protocol].listeners);
805 RCU_INIT_POINTER(nl_table[sk->sk_protocol].listeners, NULL);
807 nl_table[sk->sk_protocol].module = NULL;
808 nl_table[sk->sk_protocol].bind = NULL;
809 nl_table[sk->sk_protocol].unbind = NULL;
810 nl_table[sk->sk_protocol].flags = 0;
811 nl_table[sk->sk_protocol].registered = 0;
813 netlink_table_ungrab();
817 sock_prot_inuse_add(sock_net(sk), &netlink_proto, -1);
819 call_rcu(&nlk->rcu, deferred_put_nlk_sk);
823 static int netlink_autobind(struct socket *sock)
825 struct sock *sk = sock->sk;
826 struct net *net = sock_net(sk);
827 struct netlink_table *table = &nl_table[sk->sk_protocol];
828 s32 portid = task_tgid_vnr(current);
836 ok = !__netlink_lookup(table, portid, net);
839 /* Bind collision, search negative portid values. */
841 /* rover will be in range [S32_MIN, -4097] */
842 rover = S32_MIN + prandom_u32_max(-4096 - S32_MIN);
843 else if (rover >= -4096)
849 err = netlink_insert(sk, portid);
850 if (err == -EADDRINUSE)
853 /* If 2 threads race to autobind, that is fine. */
861 * __netlink_ns_capable - General netlink message capability test
862 * @nsp: NETLINK_CB of the socket buffer holding a netlink command from userspace.
863 * @user_ns: The user namespace of the capability to use
864 * @cap: The capability to use
866 * Test to see if the opener of the socket we received the message
867 * from had when the netlink socket was created and the sender of the
868 * message has the capability @cap in the user namespace @user_ns.
870 bool __netlink_ns_capable(const struct netlink_skb_parms *nsp,
871 struct user_namespace *user_ns, int cap)
873 return ((nsp->flags & NETLINK_SKB_DST) ||
874 file_ns_capable(nsp->sk->sk_socket->file, user_ns, cap)) &&
875 ns_capable(user_ns, cap);
877 EXPORT_SYMBOL(__netlink_ns_capable);
880 * netlink_ns_capable - General netlink message capability test
881 * @skb: socket buffer holding a netlink command from userspace
882 * @user_ns: The user namespace of the capability to use
883 * @cap: The capability to use
885 * Test to see if the opener of the socket we received the message
886 * from had when the netlink socket was created and the sender of the
887 * message has the capability @cap in the user namespace @user_ns.
889 bool netlink_ns_capable(const struct sk_buff *skb,
890 struct user_namespace *user_ns, int cap)
892 return __netlink_ns_capable(&NETLINK_CB(skb), user_ns, cap);
894 EXPORT_SYMBOL(netlink_ns_capable);
897 * netlink_capable - Netlink global message capability test
898 * @skb: socket buffer holding a netlink command from userspace
899 * @cap: The capability to use
901 * Test to see if the opener of the socket we received the message
902 * from had when the netlink socket was created and the sender of the
903 * message has the capability @cap in all user namespaces.
905 bool netlink_capable(const struct sk_buff *skb, int cap)
907 return netlink_ns_capable(skb, &init_user_ns, cap);
909 EXPORT_SYMBOL(netlink_capable);
912 * netlink_net_capable - Netlink network namespace message capability test
913 * @skb: socket buffer holding a netlink command from userspace
914 * @cap: The capability to use
916 * Test to see if the opener of the socket we received the message
917 * from had when the netlink socket was created and the sender of the
918 * message has the capability @cap over the network namespace of
919 * the socket we received the message from.
921 bool netlink_net_capable(const struct sk_buff *skb, int cap)
923 return netlink_ns_capable(skb, sock_net(skb->sk)->user_ns, cap);
925 EXPORT_SYMBOL(netlink_net_capable);
927 static inline int netlink_allowed(const struct socket *sock, unsigned int flag)
929 return (nl_table[sock->sk->sk_protocol].flags & flag) ||
930 ns_capable(sock_net(sock->sk)->user_ns, CAP_NET_ADMIN);
934 netlink_update_subscriptions(struct sock *sk, unsigned int subscriptions)
936 struct netlink_sock *nlk = nlk_sk(sk);
938 if (nlk->subscriptions && !subscriptions)
939 __sk_del_bind_node(sk);
940 else if (!nlk->subscriptions && subscriptions)
941 sk_add_bind_node(sk, &nl_table[sk->sk_protocol].mc_list);
942 nlk->subscriptions = subscriptions;
945 static int netlink_realloc_groups(struct sock *sk)
947 struct netlink_sock *nlk = nlk_sk(sk);
949 unsigned long *new_groups;
952 netlink_table_grab();
954 groups = nl_table[sk->sk_protocol].groups;
955 if (!nl_table[sk->sk_protocol].registered) {
960 if (nlk->ngroups >= groups)
963 new_groups = krealloc(nlk->groups, NLGRPSZ(groups), GFP_ATOMIC);
964 if (new_groups == NULL) {
968 memset((char *)new_groups + NLGRPSZ(nlk->ngroups), 0,
969 NLGRPSZ(groups) - NLGRPSZ(nlk->ngroups));
971 nlk->groups = new_groups;
972 nlk->ngroups = groups;
974 netlink_table_ungrab();
978 static void netlink_undo_bind(int group, long unsigned int groups,
981 struct netlink_sock *nlk = nlk_sk(sk);
984 if (!nlk->netlink_unbind)
987 for (undo = 0; undo < group; undo++)
988 if (test_bit(undo, &groups))
989 nlk->netlink_unbind(sock_net(sk), undo + 1);
992 static int netlink_bind(struct socket *sock, struct sockaddr *addr,
995 struct sock *sk = sock->sk;
996 struct net *net = sock_net(sk);
997 struct netlink_sock *nlk = nlk_sk(sk);
998 struct sockaddr_nl *nladdr = (struct sockaddr_nl *)addr;
1000 unsigned long groups;
1003 if (addr_len < sizeof(struct sockaddr_nl))
1006 if (nladdr->nl_family != AF_NETLINK)
1008 groups = nladdr->nl_groups;
1010 /* Only superuser is allowed to listen multicasts */
1012 if (!netlink_allowed(sock, NL_CFG_F_NONROOT_RECV))
1014 err = netlink_realloc_groups(sk);
1019 if (nlk->ngroups < BITS_PER_LONG)
1020 groups &= (1UL << nlk->ngroups) - 1;
1022 /* Paired with WRITE_ONCE() in netlink_insert() */
1023 bound = READ_ONCE(nlk->bound);
1025 /* Ensure nlk->portid is up-to-date. */
1028 if (nladdr->nl_pid != nlk->portid)
1032 if (nlk->netlink_bind && groups) {
1035 /* nl_groups is a u32, so cap the maximum groups we can bind */
1036 for (group = 0; group < BITS_PER_TYPE(u32); group++) {
1037 if (!test_bit(group, &groups))
1039 err = nlk->netlink_bind(net, group + 1);
1042 netlink_undo_bind(group, groups, sk);
1047 /* No need for barriers here as we return to user-space without
1048 * using any of the bound attributes.
1050 netlink_lock_table();
1052 err = nladdr->nl_pid ?
1053 netlink_insert(sk, nladdr->nl_pid) :
1054 netlink_autobind(sock);
1056 netlink_undo_bind(BITS_PER_TYPE(u32), groups, sk);
1061 if (!groups && (nlk->groups == NULL || !(u32)nlk->groups[0]))
1063 netlink_unlock_table();
1065 netlink_table_grab();
1066 netlink_update_subscriptions(sk, nlk->subscriptions +
1068 hweight32(nlk->groups[0]));
1069 nlk->groups[0] = (nlk->groups[0] & ~0xffffffffUL) | groups;
1070 netlink_update_listeners(sk);
1071 netlink_table_ungrab();
1076 netlink_unlock_table();
1080 static int netlink_connect(struct socket *sock, struct sockaddr *addr,
1081 int alen, int flags)
1084 struct sock *sk = sock->sk;
1085 struct netlink_sock *nlk = nlk_sk(sk);
1086 struct sockaddr_nl *nladdr = (struct sockaddr_nl *)addr;
1088 if (alen < sizeof(addr->sa_family))
1091 if (addr->sa_family == AF_UNSPEC) {
1092 sk->sk_state = NETLINK_UNCONNECTED;
1093 /* dst_portid and dst_group can be read locklessly */
1094 WRITE_ONCE(nlk->dst_portid, 0);
1095 WRITE_ONCE(nlk->dst_group, 0);
1098 if (addr->sa_family != AF_NETLINK)
1101 if (alen < sizeof(struct sockaddr_nl))
1104 if ((nladdr->nl_groups || nladdr->nl_pid) &&
1105 !netlink_allowed(sock, NL_CFG_F_NONROOT_SEND))
1108 /* No need for barriers here as we return to user-space without
1109 * using any of the bound attributes.
1110 * Paired with WRITE_ONCE() in netlink_insert().
1112 if (!READ_ONCE(nlk->bound))
1113 err = netlink_autobind(sock);
1116 sk->sk_state = NETLINK_CONNECTED;
1117 /* dst_portid and dst_group can be read locklessly */
1118 WRITE_ONCE(nlk->dst_portid, nladdr->nl_pid);
1119 WRITE_ONCE(nlk->dst_group, ffs(nladdr->nl_groups));
1125 static int netlink_getname(struct socket *sock, struct sockaddr *addr,
1128 struct sock *sk = sock->sk;
1129 struct netlink_sock *nlk = nlk_sk(sk);
1130 DECLARE_SOCKADDR(struct sockaddr_nl *, nladdr, addr);
1132 nladdr->nl_family = AF_NETLINK;
1136 /* Paired with WRITE_ONCE() in netlink_connect() */
1137 nladdr->nl_pid = READ_ONCE(nlk->dst_portid);
1138 nladdr->nl_groups = netlink_group_mask(READ_ONCE(nlk->dst_group));
1140 /* Paired with WRITE_ONCE() in netlink_insert() */
1141 nladdr->nl_pid = READ_ONCE(nlk->portid);
1142 netlink_lock_table();
1143 nladdr->nl_groups = nlk->groups ? nlk->groups[0] : 0;
1144 netlink_unlock_table();
1146 return sizeof(*nladdr);
1149 static int netlink_ioctl(struct socket *sock, unsigned int cmd,
1152 /* try to hand this ioctl down to the NIC drivers.
1154 return -ENOIOCTLCMD;
1157 static struct sock *netlink_getsockbyportid(struct sock *ssk, u32 portid)
1160 struct netlink_sock *nlk;
1162 sock = netlink_lookup(sock_net(ssk), ssk->sk_protocol, portid);
1164 return ERR_PTR(-ECONNREFUSED);
1166 /* Don't bother queuing skb if kernel socket has no input function */
1168 /* dst_portid can be changed in netlink_connect() */
1169 if (sock->sk_state == NETLINK_CONNECTED &&
1170 READ_ONCE(nlk->dst_portid) != nlk_sk(ssk)->portid) {
1172 return ERR_PTR(-ECONNREFUSED);
1177 struct sock *netlink_getsockbyfilp(struct file *filp)
1179 struct inode *inode = file_inode(filp);
1182 if (!S_ISSOCK(inode->i_mode))
1183 return ERR_PTR(-ENOTSOCK);
1185 sock = SOCKET_I(inode)->sk;
1186 if (sock->sk_family != AF_NETLINK)
1187 return ERR_PTR(-EINVAL);
1193 static struct sk_buff *netlink_alloc_large_skb(unsigned int size,
1196 struct sk_buff *skb;
1199 if (size <= NLMSG_GOODSIZE || broadcast)
1200 return alloc_skb(size, GFP_KERNEL);
1202 size = SKB_DATA_ALIGN(size) +
1203 SKB_DATA_ALIGN(sizeof(struct skb_shared_info));
1205 data = vmalloc(size);
1209 skb = __build_skb(data, size);
1213 skb->destructor = netlink_skb_destructor;
1219 * Attach a skb to a netlink socket.
1220 * The caller must hold a reference to the destination socket. On error, the
1221 * reference is dropped. The skb is not send to the destination, just all
1222 * all error checks are performed and memory in the queue is reserved.
1224 * < 0: error. skb freed, reference to sock dropped.
1226 * 1: repeat lookup - reference dropped while waiting for socket memory.
1228 int netlink_attachskb(struct sock *sk, struct sk_buff *skb,
1229 long *timeo, struct sock *ssk)
1231 struct netlink_sock *nlk;
1235 if ((atomic_read(&sk->sk_rmem_alloc) > sk->sk_rcvbuf ||
1236 test_bit(NETLINK_S_CONGESTED, &nlk->state))) {
1237 DECLARE_WAITQUEUE(wait, current);
1239 if (!ssk || netlink_is_kernel(ssk))
1240 netlink_overrun(sk);
1246 __set_current_state(TASK_INTERRUPTIBLE);
1247 add_wait_queue(&nlk->wait, &wait);
1249 if ((atomic_read(&sk->sk_rmem_alloc) > sk->sk_rcvbuf ||
1250 test_bit(NETLINK_S_CONGESTED, &nlk->state)) &&
1251 !sock_flag(sk, SOCK_DEAD))
1252 *timeo = schedule_timeout(*timeo);
1254 __set_current_state(TASK_RUNNING);
1255 remove_wait_queue(&nlk->wait, &wait);
1258 if (signal_pending(current)) {
1260 return sock_intr_errno(*timeo);
1264 netlink_skb_set_owner_r(skb, sk);
1268 static int __netlink_sendskb(struct sock *sk, struct sk_buff *skb)
1272 netlink_deliver_tap(sock_net(sk), skb);
1274 skb_queue_tail(&sk->sk_receive_queue, skb);
1275 sk->sk_data_ready(sk);
1279 int netlink_sendskb(struct sock *sk, struct sk_buff *skb)
1281 int len = __netlink_sendskb(sk, skb);
1287 void netlink_detachskb(struct sock *sk, struct sk_buff *skb)
1293 static struct sk_buff *netlink_trim(struct sk_buff *skb, gfp_t allocation)
1297 WARN_ON(skb->sk != NULL);
1298 delta = skb->end - skb->tail;
1299 if (is_vmalloc_addr(skb->head) || delta * 2 < skb->truesize)
1302 if (skb_shared(skb)) {
1303 struct sk_buff *nskb = skb_clone(skb, allocation);
1310 pskb_expand_head(skb, 0, -delta,
1311 (allocation & ~__GFP_DIRECT_RECLAIM) |
1312 __GFP_NOWARN | __GFP_NORETRY);
1316 static int netlink_unicast_kernel(struct sock *sk, struct sk_buff *skb,
1320 struct netlink_sock *nlk = nlk_sk(sk);
1322 ret = -ECONNREFUSED;
1323 if (nlk->netlink_rcv != NULL) {
1325 netlink_skb_set_owner_r(skb, sk);
1326 NETLINK_CB(skb).sk = ssk;
1327 netlink_deliver_tap_kernel(sk, ssk, skb);
1328 nlk->netlink_rcv(skb);
1337 int netlink_unicast(struct sock *ssk, struct sk_buff *skb,
1338 u32 portid, int nonblock)
1344 skb = netlink_trim(skb, gfp_any());
1346 timeo = sock_sndtimeo(ssk, nonblock);
1348 sk = netlink_getsockbyportid(ssk, portid);
1353 if (netlink_is_kernel(sk))
1354 return netlink_unicast_kernel(sk, skb, ssk);
1356 if (sk_filter(sk, skb)) {
1363 err = netlink_attachskb(sk, skb, &timeo, ssk);
1369 return netlink_sendskb(sk, skb);
1371 EXPORT_SYMBOL(netlink_unicast);
1373 int netlink_has_listeners(struct sock *sk, unsigned int group)
1376 struct listeners *listeners;
1378 BUG_ON(!netlink_is_kernel(sk));
1381 listeners = rcu_dereference(nl_table[sk->sk_protocol].listeners);
1383 if (listeners && group - 1 < nl_table[sk->sk_protocol].groups)
1384 res = test_bit(group - 1, listeners->masks);
1390 EXPORT_SYMBOL_GPL(netlink_has_listeners);
1392 bool netlink_strict_get_check(struct sk_buff *skb)
1394 const struct netlink_sock *nlk = nlk_sk(NETLINK_CB(skb).sk);
1396 return nlk->flags & NETLINK_F_STRICT_CHK;
1398 EXPORT_SYMBOL_GPL(netlink_strict_get_check);
1400 static int netlink_broadcast_deliver(struct sock *sk, struct sk_buff *skb)
1402 struct netlink_sock *nlk = nlk_sk(sk);
1404 if (atomic_read(&sk->sk_rmem_alloc) <= sk->sk_rcvbuf &&
1405 !test_bit(NETLINK_S_CONGESTED, &nlk->state)) {
1406 netlink_skb_set_owner_r(skb, sk);
1407 __netlink_sendskb(sk, skb);
1408 return atomic_read(&sk->sk_rmem_alloc) > (sk->sk_rcvbuf >> 1);
1413 struct netlink_broadcast_data {
1414 struct sock *exclude_sk;
1419 int delivery_failure;
1423 struct sk_buff *skb, *skb2;
1424 int (*tx_filter)(struct sock *dsk, struct sk_buff *skb, void *data);
1428 static void do_one_broadcast(struct sock *sk,
1429 struct netlink_broadcast_data *p)
1431 struct netlink_sock *nlk = nlk_sk(sk);
1434 if (p->exclude_sk == sk)
1437 if (nlk->portid == p->portid || p->group - 1 >= nlk->ngroups ||
1438 !test_bit(p->group - 1, nlk->groups))
1441 if (!net_eq(sock_net(sk), p->net)) {
1442 if (!(nlk->flags & NETLINK_F_LISTEN_ALL_NSID))
1445 if (!peernet_has_id(sock_net(sk), p->net))
1448 if (!file_ns_capable(sk->sk_socket->file, p->net->user_ns,
1454 netlink_overrun(sk);
1459 if (p->skb2 == NULL) {
1460 if (skb_shared(p->skb)) {
1461 p->skb2 = skb_clone(p->skb, p->allocation);
1463 p->skb2 = skb_get(p->skb);
1465 * skb ownership may have been set when
1466 * delivered to a previous socket.
1468 skb_orphan(p->skb2);
1471 if (p->skb2 == NULL) {
1472 netlink_overrun(sk);
1473 /* Clone failed. Notify ALL listeners. */
1475 if (nlk->flags & NETLINK_F_BROADCAST_SEND_ERROR)
1476 p->delivery_failure = 1;
1479 if (p->tx_filter && p->tx_filter(sk, p->skb2, p->tx_data)) {
1484 if (sk_filter(sk, p->skb2)) {
1489 NETLINK_CB(p->skb2).nsid = peernet2id(sock_net(sk), p->net);
1490 if (NETLINK_CB(p->skb2).nsid != NETNSA_NSID_NOT_ASSIGNED)
1491 NETLINK_CB(p->skb2).nsid_is_set = true;
1492 val = netlink_broadcast_deliver(sk, p->skb2);
1494 netlink_overrun(sk);
1495 if (nlk->flags & NETLINK_F_BROADCAST_SEND_ERROR)
1496 p->delivery_failure = 1;
1498 p->congested |= val;
1506 int netlink_broadcast_filtered(struct sock *ssk, struct sk_buff *skb, u32 portid,
1507 u32 group, gfp_t allocation,
1508 int (*filter)(struct sock *dsk, struct sk_buff *skb, void *data),
1511 struct net *net = sock_net(ssk);
1512 struct netlink_broadcast_data info;
1515 skb = netlink_trim(skb, allocation);
1517 info.exclude_sk = ssk;
1519 info.portid = portid;
1522 info.delivery_failure = 0;
1525 info.allocation = allocation;
1528 info.tx_filter = filter;
1529 info.tx_data = filter_data;
1531 /* While we sleep in clone, do not allow to change socket list */
1533 netlink_lock_table();
1535 sk_for_each_bound(sk, &nl_table[ssk->sk_protocol].mc_list)
1536 do_one_broadcast(sk, &info);
1540 netlink_unlock_table();
1542 if (info.delivery_failure) {
1543 kfree_skb(info.skb2);
1546 consume_skb(info.skb2);
1548 if (info.delivered) {
1549 if (info.congested && gfpflags_allow_blocking(allocation))
1555 EXPORT_SYMBOL(netlink_broadcast_filtered);
1557 int netlink_broadcast(struct sock *ssk, struct sk_buff *skb, u32 portid,
1558 u32 group, gfp_t allocation)
1560 return netlink_broadcast_filtered(ssk, skb, portid, group, allocation,
1563 EXPORT_SYMBOL(netlink_broadcast);
1565 struct netlink_set_err_data {
1566 struct sock *exclude_sk;
1572 static int do_one_set_err(struct sock *sk, struct netlink_set_err_data *p)
1574 struct netlink_sock *nlk = nlk_sk(sk);
1577 if (sk == p->exclude_sk)
1580 if (!net_eq(sock_net(sk), sock_net(p->exclude_sk)))
1583 if (nlk->portid == p->portid || p->group - 1 >= nlk->ngroups ||
1584 !test_bit(p->group - 1, nlk->groups))
1587 if (p->code == ENOBUFS && nlk->flags & NETLINK_F_RECV_NO_ENOBUFS) {
1592 sk->sk_err = p->code;
1593 sk_error_report(sk);
1599 * netlink_set_err - report error to broadcast listeners
1600 * @ssk: the kernel netlink socket, as returned by netlink_kernel_create()
1601 * @portid: the PORTID of a process that we want to skip (if any)
1602 * @group: the broadcast group that will notice the error
1603 * @code: error code, must be negative (as usual in kernelspace)
1605 * This function returns the number of broadcast listeners that have set the
1606 * NETLINK_NO_ENOBUFS socket option.
1608 int netlink_set_err(struct sock *ssk, u32 portid, u32 group, int code)
1610 struct netlink_set_err_data info;
1614 info.exclude_sk = ssk;
1615 info.portid = portid;
1617 /* sk->sk_err wants a positive error value */
1620 read_lock(&nl_table_lock);
1622 sk_for_each_bound(sk, &nl_table[ssk->sk_protocol].mc_list)
1623 ret += do_one_set_err(sk, &info);
1625 read_unlock(&nl_table_lock);
1628 EXPORT_SYMBOL(netlink_set_err);
1630 /* must be called with netlink table grabbed */
1631 static void netlink_update_socket_mc(struct netlink_sock *nlk,
1635 int old, new = !!is_new, subscriptions;
1637 old = test_bit(group - 1, nlk->groups);
1638 subscriptions = nlk->subscriptions - old + new;
1640 __set_bit(group - 1, nlk->groups);
1642 __clear_bit(group - 1, nlk->groups);
1643 netlink_update_subscriptions(&nlk->sk, subscriptions);
1644 netlink_update_listeners(&nlk->sk);
1647 static int netlink_setsockopt(struct socket *sock, int level, int optname,
1648 sockptr_t optval, unsigned int optlen)
1650 struct sock *sk = sock->sk;
1651 struct netlink_sock *nlk = nlk_sk(sk);
1652 unsigned int val = 0;
1655 if (level != SOL_NETLINK)
1656 return -ENOPROTOOPT;
1658 if (optlen >= sizeof(int) &&
1659 copy_from_sockptr(&val, optval, sizeof(val)))
1663 case NETLINK_PKTINFO:
1665 nlk->flags |= NETLINK_F_RECV_PKTINFO;
1667 nlk->flags &= ~NETLINK_F_RECV_PKTINFO;
1670 case NETLINK_ADD_MEMBERSHIP:
1671 case NETLINK_DROP_MEMBERSHIP: {
1672 if (!netlink_allowed(sock, NL_CFG_F_NONROOT_RECV))
1674 err = netlink_realloc_groups(sk);
1677 if (!val || val - 1 >= nlk->ngroups)
1679 if (optname == NETLINK_ADD_MEMBERSHIP && nlk->netlink_bind) {
1680 err = nlk->netlink_bind(sock_net(sk), val);
1684 netlink_table_grab();
1685 netlink_update_socket_mc(nlk, val,
1686 optname == NETLINK_ADD_MEMBERSHIP);
1687 netlink_table_ungrab();
1688 if (optname == NETLINK_DROP_MEMBERSHIP && nlk->netlink_unbind)
1689 nlk->netlink_unbind(sock_net(sk), val);
1694 case NETLINK_BROADCAST_ERROR:
1696 nlk->flags |= NETLINK_F_BROADCAST_SEND_ERROR;
1698 nlk->flags &= ~NETLINK_F_BROADCAST_SEND_ERROR;
1701 case NETLINK_NO_ENOBUFS:
1703 nlk->flags |= NETLINK_F_RECV_NO_ENOBUFS;
1704 clear_bit(NETLINK_S_CONGESTED, &nlk->state);
1705 wake_up_interruptible(&nlk->wait);
1707 nlk->flags &= ~NETLINK_F_RECV_NO_ENOBUFS;
1711 case NETLINK_LISTEN_ALL_NSID:
1712 if (!ns_capable(sock_net(sk)->user_ns, CAP_NET_BROADCAST))
1716 nlk->flags |= NETLINK_F_LISTEN_ALL_NSID;
1718 nlk->flags &= ~NETLINK_F_LISTEN_ALL_NSID;
1721 case NETLINK_CAP_ACK:
1723 nlk->flags |= NETLINK_F_CAP_ACK;
1725 nlk->flags &= ~NETLINK_F_CAP_ACK;
1728 case NETLINK_EXT_ACK:
1730 nlk->flags |= NETLINK_F_EXT_ACK;
1732 nlk->flags &= ~NETLINK_F_EXT_ACK;
1735 case NETLINK_GET_STRICT_CHK:
1737 nlk->flags |= NETLINK_F_STRICT_CHK;
1739 nlk->flags &= ~NETLINK_F_STRICT_CHK;
1748 static int netlink_getsockopt(struct socket *sock, int level, int optname,
1749 char __user *optval, int __user *optlen)
1751 struct sock *sk = sock->sk;
1752 struct netlink_sock *nlk = nlk_sk(sk);
1755 if (level != SOL_NETLINK)
1756 return -ENOPROTOOPT;
1758 if (get_user(len, optlen))
1764 case NETLINK_PKTINFO:
1765 if (len < sizeof(int))
1768 val = nlk->flags & NETLINK_F_RECV_PKTINFO ? 1 : 0;
1769 if (put_user(len, optlen) ||
1770 put_user(val, optval))
1774 case NETLINK_BROADCAST_ERROR:
1775 if (len < sizeof(int))
1778 val = nlk->flags & NETLINK_F_BROADCAST_SEND_ERROR ? 1 : 0;
1779 if (put_user(len, optlen) ||
1780 put_user(val, optval))
1784 case NETLINK_NO_ENOBUFS:
1785 if (len < sizeof(int))
1788 val = nlk->flags & NETLINK_F_RECV_NO_ENOBUFS ? 1 : 0;
1789 if (put_user(len, optlen) ||
1790 put_user(val, optval))
1794 case NETLINK_LIST_MEMBERSHIPS: {
1795 int pos, idx, shift;
1798 netlink_lock_table();
1799 for (pos = 0; pos * 8 < nlk->ngroups; pos += sizeof(u32)) {
1800 if (len - pos < sizeof(u32))
1803 idx = pos / sizeof(unsigned long);
1804 shift = (pos % sizeof(unsigned long)) * 8;
1805 if (put_user((u32)(nlk->groups[idx] >> shift),
1806 (u32 __user *)(optval + pos))) {
1811 if (put_user(ALIGN(nlk->ngroups / 8, sizeof(u32)), optlen))
1813 netlink_unlock_table();
1816 case NETLINK_CAP_ACK:
1817 if (len < sizeof(int))
1820 val = nlk->flags & NETLINK_F_CAP_ACK ? 1 : 0;
1821 if (put_user(len, optlen) ||
1822 put_user(val, optval))
1826 case NETLINK_EXT_ACK:
1827 if (len < sizeof(int))
1830 val = nlk->flags & NETLINK_F_EXT_ACK ? 1 : 0;
1831 if (put_user(len, optlen) || put_user(val, optval))
1835 case NETLINK_GET_STRICT_CHK:
1836 if (len < sizeof(int))
1839 val = nlk->flags & NETLINK_F_STRICT_CHK ? 1 : 0;
1840 if (put_user(len, optlen) || put_user(val, optval))
1850 static void netlink_cmsg_recv_pktinfo(struct msghdr *msg, struct sk_buff *skb)
1852 struct nl_pktinfo info;
1854 info.group = NETLINK_CB(skb).dst_group;
1855 put_cmsg(msg, SOL_NETLINK, NETLINK_PKTINFO, sizeof(info), &info);
1858 static void netlink_cmsg_listen_all_nsid(struct sock *sk, struct msghdr *msg,
1859 struct sk_buff *skb)
1861 if (!NETLINK_CB(skb).nsid_is_set)
1864 put_cmsg(msg, SOL_NETLINK, NETLINK_LISTEN_ALL_NSID, sizeof(int),
1865 &NETLINK_CB(skb).nsid);
1868 static int netlink_sendmsg(struct socket *sock, struct msghdr *msg, size_t len)
1870 struct sock *sk = sock->sk;
1871 struct netlink_sock *nlk = nlk_sk(sk);
1872 DECLARE_SOCKADDR(struct sockaddr_nl *, addr, msg->msg_name);
1875 struct sk_buff *skb;
1877 struct scm_cookie scm;
1878 u32 netlink_skb_flags = 0;
1880 if (msg->msg_flags & MSG_OOB)
1884 pr_warn_once("Zero length message leads to an empty skb\n");
1888 err = scm_send(sock, msg, &scm, true);
1892 if (msg->msg_namelen) {
1894 if (msg->msg_namelen < sizeof(struct sockaddr_nl))
1896 if (addr->nl_family != AF_NETLINK)
1898 dst_portid = addr->nl_pid;
1899 dst_group = ffs(addr->nl_groups);
1901 if ((dst_group || dst_portid) &&
1902 !netlink_allowed(sock, NL_CFG_F_NONROOT_SEND))
1904 netlink_skb_flags |= NETLINK_SKB_DST;
1906 /* Paired with WRITE_ONCE() in netlink_connect() */
1907 dst_portid = READ_ONCE(nlk->dst_portid);
1908 dst_group = READ_ONCE(nlk->dst_group);
1911 /* Paired with WRITE_ONCE() in netlink_insert() */
1912 if (!READ_ONCE(nlk->bound)) {
1913 err = netlink_autobind(sock);
1917 /* Ensure nlk is hashed and visible. */
1922 if (len > sk->sk_sndbuf - 32)
1925 skb = netlink_alloc_large_skb(len, dst_group);
1929 NETLINK_CB(skb).portid = nlk->portid;
1930 NETLINK_CB(skb).dst_group = dst_group;
1931 NETLINK_CB(skb).creds = scm.creds;
1932 NETLINK_CB(skb).flags = netlink_skb_flags;
1935 if (memcpy_from_msg(skb_put(skb, len), msg, len)) {
1940 err = security_netlink_send(sk, skb);
1947 refcount_inc(&skb->users);
1948 netlink_broadcast(sk, skb, dst_portid, dst_group, GFP_KERNEL);
1950 err = netlink_unicast(sk, skb, dst_portid, msg->msg_flags & MSG_DONTWAIT);
1957 static int netlink_recvmsg(struct socket *sock, struct msghdr *msg, size_t len,
1960 struct scm_cookie scm;
1961 struct sock *sk = sock->sk;
1962 struct netlink_sock *nlk = nlk_sk(sk);
1963 int noblock = flags & MSG_DONTWAIT;
1965 struct sk_buff *skb, *data_skb;
1968 if (flags & MSG_OOB)
1973 skb = skb_recv_datagram(sk, flags, noblock, &err);
1979 #ifdef CONFIG_COMPAT_NETLINK_MESSAGES
1980 if (unlikely(skb_shinfo(skb)->frag_list)) {
1982 * If this skb has a frag_list, then here that means that we
1983 * will have to use the frag_list skb's data for compat tasks
1984 * and the regular skb's data for normal (non-compat) tasks.
1986 * If we need to send the compat skb, assign it to the
1987 * 'data_skb' variable so that it will be used below for data
1988 * copying. We keep 'skb' for everything else, including
1989 * freeing both later.
1991 if (flags & MSG_CMSG_COMPAT)
1992 data_skb = skb_shinfo(skb)->frag_list;
1996 /* Record the max length of recvmsg() calls for future allocations */
1997 nlk->max_recvmsg_len = max(nlk->max_recvmsg_len, len);
1998 nlk->max_recvmsg_len = min_t(size_t, nlk->max_recvmsg_len,
1999 SKB_WITH_OVERHEAD(32768));
2001 copied = data_skb->len;
2003 msg->msg_flags |= MSG_TRUNC;
2007 err = skb_copy_datagram_msg(data_skb, 0, msg, copied);
2009 if (msg->msg_name) {
2010 DECLARE_SOCKADDR(struct sockaddr_nl *, addr, msg->msg_name);
2011 addr->nl_family = AF_NETLINK;
2013 addr->nl_pid = NETLINK_CB(skb).portid;
2014 addr->nl_groups = netlink_group_mask(NETLINK_CB(skb).dst_group);
2015 msg->msg_namelen = sizeof(*addr);
2018 if (nlk->flags & NETLINK_F_RECV_PKTINFO)
2019 netlink_cmsg_recv_pktinfo(msg, skb);
2020 if (nlk->flags & NETLINK_F_LISTEN_ALL_NSID)
2021 netlink_cmsg_listen_all_nsid(sk, msg, skb);
2023 memset(&scm, 0, sizeof(scm));
2024 scm.creds = *NETLINK_CREDS(skb);
2025 if (flags & MSG_TRUNC)
2026 copied = data_skb->len;
2028 skb_free_datagram(sk, skb);
2030 if (nlk->cb_running &&
2031 atomic_read(&sk->sk_rmem_alloc) <= sk->sk_rcvbuf / 2) {
2032 ret = netlink_dump(sk);
2035 sk_error_report(sk);
2039 scm_recv(sock, msg, &scm, flags);
2041 netlink_rcv_wake(sk);
2042 return err ? : copied;
2045 static void netlink_data_ready(struct sock *sk)
2051 * We export these functions to other modules. They provide a
2052 * complete set of kernel non-blocking support for message
2057 __netlink_kernel_create(struct net *net, int unit, struct module *module,
2058 struct netlink_kernel_cfg *cfg)
2060 struct socket *sock;
2062 struct netlink_sock *nlk;
2063 struct listeners *listeners = NULL;
2064 struct mutex *cb_mutex = cfg ? cfg->cb_mutex : NULL;
2065 unsigned int groups;
2069 if (unit < 0 || unit >= MAX_LINKS)
2072 if (sock_create_lite(PF_NETLINK, SOCK_DGRAM, unit, &sock))
2075 if (__netlink_create(net, sock, cb_mutex, unit, 1) < 0)
2076 goto out_sock_release_nosk;
2080 if (!cfg || cfg->groups < 32)
2083 groups = cfg->groups;
2085 listeners = kzalloc(sizeof(*listeners) + NLGRPSZ(groups), GFP_KERNEL);
2087 goto out_sock_release;
2089 sk->sk_data_ready = netlink_data_ready;
2090 if (cfg && cfg->input)
2091 nlk_sk(sk)->netlink_rcv = cfg->input;
2093 if (netlink_insert(sk, 0))
2094 goto out_sock_release;
2097 nlk->flags |= NETLINK_F_KERNEL_SOCKET;
2099 netlink_table_grab();
2100 if (!nl_table[unit].registered) {
2101 nl_table[unit].groups = groups;
2102 rcu_assign_pointer(nl_table[unit].listeners, listeners);
2103 nl_table[unit].cb_mutex = cb_mutex;
2104 nl_table[unit].module = module;
2106 nl_table[unit].bind = cfg->bind;
2107 nl_table[unit].unbind = cfg->unbind;
2108 nl_table[unit].flags = cfg->flags;
2110 nl_table[unit].compare = cfg->compare;
2112 nl_table[unit].registered = 1;
2115 nl_table[unit].registered++;
2117 netlink_table_ungrab();
2122 netlink_kernel_release(sk);
2125 out_sock_release_nosk:
2129 EXPORT_SYMBOL(__netlink_kernel_create);
2132 netlink_kernel_release(struct sock *sk)
2134 if (sk == NULL || sk->sk_socket == NULL)
2137 sock_release(sk->sk_socket);
2139 EXPORT_SYMBOL(netlink_kernel_release);
2141 int __netlink_change_ngroups(struct sock *sk, unsigned int groups)
2143 struct listeners *new, *old;
2144 struct netlink_table *tbl = &nl_table[sk->sk_protocol];
2149 if (NLGRPSZ(tbl->groups) < NLGRPSZ(groups)) {
2150 new = kzalloc(sizeof(*new) + NLGRPSZ(groups), GFP_ATOMIC);
2153 old = nl_deref_protected(tbl->listeners);
2154 memcpy(new->masks, old->masks, NLGRPSZ(tbl->groups));
2155 rcu_assign_pointer(tbl->listeners, new);
2157 kfree_rcu(old, rcu);
2159 tbl->groups = groups;
2165 * netlink_change_ngroups - change number of multicast groups
2167 * This changes the number of multicast groups that are available
2168 * on a certain netlink family. Note that it is not possible to
2169 * change the number of groups to below 32. Also note that it does
2170 * not implicitly call netlink_clear_multicast_users() when the
2171 * number of groups is reduced.
2173 * @sk: The kernel netlink socket, as returned by netlink_kernel_create().
2174 * @groups: The new number of groups.
2176 int netlink_change_ngroups(struct sock *sk, unsigned int groups)
2180 netlink_table_grab();
2181 err = __netlink_change_ngroups(sk, groups);
2182 netlink_table_ungrab();
2187 void __netlink_clear_multicast_users(struct sock *ksk, unsigned int group)
2190 struct netlink_table *tbl = &nl_table[ksk->sk_protocol];
2192 sk_for_each_bound(sk, &tbl->mc_list)
2193 netlink_update_socket_mc(nlk_sk(sk), group, 0);
2197 __nlmsg_put(struct sk_buff *skb, u32 portid, u32 seq, int type, int len, int flags)
2199 struct nlmsghdr *nlh;
2200 int size = nlmsg_msg_size(len);
2202 nlh = skb_put(skb, NLMSG_ALIGN(size));
2203 nlh->nlmsg_type = type;
2204 nlh->nlmsg_len = size;
2205 nlh->nlmsg_flags = flags;
2206 nlh->nlmsg_pid = portid;
2207 nlh->nlmsg_seq = seq;
2208 if (!__builtin_constant_p(size) || NLMSG_ALIGN(size) - size != 0)
2209 memset(nlmsg_data(nlh) + len, 0, NLMSG_ALIGN(size) - size);
2212 EXPORT_SYMBOL(__nlmsg_put);
2215 * It looks a bit ugly.
2216 * It would be better to create kernel thread.
2219 static int netlink_dump_done(struct netlink_sock *nlk, struct sk_buff *skb,
2220 struct netlink_callback *cb,
2221 struct netlink_ext_ack *extack)
2223 struct nlmsghdr *nlh;
2225 nlh = nlmsg_put_answer(skb, cb, NLMSG_DONE, sizeof(nlk->dump_done_errno),
2226 NLM_F_MULTI | cb->answer_flags);
2230 nl_dump_check_consistent(cb, nlh);
2231 memcpy(nlmsg_data(nlh), &nlk->dump_done_errno, sizeof(nlk->dump_done_errno));
2233 if (extack->_msg && nlk->flags & NETLINK_F_EXT_ACK) {
2234 nlh->nlmsg_flags |= NLM_F_ACK_TLVS;
2235 if (!nla_put_string(skb, NLMSGERR_ATTR_MSG, extack->_msg))
2236 nlmsg_end(skb, nlh);
2242 static int netlink_dump(struct sock *sk)
2244 struct netlink_sock *nlk = nlk_sk(sk);
2245 struct netlink_ext_ack extack = {};
2246 struct netlink_callback *cb;
2247 struct sk_buff *skb = NULL;
2248 struct module *module;
2253 mutex_lock(nlk->cb_mutex);
2254 if (!nlk->cb_running) {
2259 if (atomic_read(&sk->sk_rmem_alloc) >= sk->sk_rcvbuf)
2262 /* NLMSG_GOODSIZE is small to avoid high order allocations being
2263 * required, but it makes sense to _attempt_ a 16K bytes allocation
2264 * to reduce number of system calls on dump operations, if user
2265 * ever provided a big enough buffer.
2268 alloc_min_size = max_t(int, cb->min_dump_alloc, NLMSG_GOODSIZE);
2270 if (alloc_min_size < nlk->max_recvmsg_len) {
2271 alloc_size = nlk->max_recvmsg_len;
2272 skb = alloc_skb(alloc_size,
2273 (GFP_KERNEL & ~__GFP_DIRECT_RECLAIM) |
2274 __GFP_NOWARN | __GFP_NORETRY);
2277 alloc_size = alloc_min_size;
2278 skb = alloc_skb(alloc_size, GFP_KERNEL);
2283 /* Trim skb to allocated size. User is expected to provide buffer as
2284 * large as max(min_dump_alloc, 16KiB (mac_recvmsg_len capped at
2285 * netlink_recvmsg())). dump will pack as many smaller messages as
2286 * could fit within the allocated skb. skb is typically allocated
2287 * with larger space than required (could be as much as near 2x the
2288 * requested size with align to next power of 2 approach). Allowing
2289 * dump to use the excess space makes it difficult for a user to have a
2290 * reasonable static buffer based on the expected largest dump of a
2291 * single netdev. The outcome is MSG_TRUNC error.
2293 skb_reserve(skb, skb_tailroom(skb) - alloc_size);
2295 /* Make sure malicious BPF programs can not read unitialized memory
2296 * from skb->head -> skb->data
2298 skb_reset_network_header(skb);
2299 skb_reset_mac_header(skb);
2301 netlink_skb_set_owner_r(skb, sk);
2303 if (nlk->dump_done_errno > 0) {
2304 cb->extack = &extack;
2305 nlk->dump_done_errno = cb->dump(skb, cb);
2309 if (nlk->dump_done_errno > 0 ||
2310 skb_tailroom(skb) < nlmsg_total_size(sizeof(nlk->dump_done_errno))) {
2311 mutex_unlock(nlk->cb_mutex);
2313 if (sk_filter(sk, skb))
2316 __netlink_sendskb(sk, skb);
2320 if (netlink_dump_done(nlk, skb, cb, &extack))
2323 #ifdef CONFIG_COMPAT_NETLINK_MESSAGES
2324 /* frag_list skb's data is used for compat tasks
2325 * and the regular skb's data for normal (non-compat) tasks.
2326 * See netlink_recvmsg().
2328 if (unlikely(skb_shinfo(skb)->frag_list)) {
2329 if (netlink_dump_done(nlk, skb_shinfo(skb)->frag_list, cb, &extack))
2334 if (sk_filter(sk, skb))
2337 __netlink_sendskb(sk, skb);
2342 nlk->cb_running = false;
2343 module = cb->module;
2345 mutex_unlock(nlk->cb_mutex);
2351 mutex_unlock(nlk->cb_mutex);
2356 int __netlink_dump_start(struct sock *ssk, struct sk_buff *skb,
2357 const struct nlmsghdr *nlh,
2358 struct netlink_dump_control *control)
2360 struct netlink_sock *nlk, *nlk2;
2361 struct netlink_callback *cb;
2365 refcount_inc(&skb->users);
2367 sk = netlink_lookup(sock_net(ssk), ssk->sk_protocol, NETLINK_CB(skb).portid);
2369 ret = -ECONNREFUSED;
2374 mutex_lock(nlk->cb_mutex);
2375 /* A dump is in progress... */
2376 if (nlk->cb_running) {
2380 /* add reference of module which cb->dump belongs to */
2381 if (!try_module_get(control->module)) {
2382 ret = -EPROTONOSUPPORT;
2387 memset(cb, 0, sizeof(*cb));
2388 cb->dump = control->dump;
2389 cb->done = control->done;
2391 cb->data = control->data;
2392 cb->module = control->module;
2393 cb->min_dump_alloc = control->min_dump_alloc;
2396 nlk2 = nlk_sk(NETLINK_CB(skb).sk);
2397 cb->strict_check = !!(nlk2->flags & NETLINK_F_STRICT_CHK);
2399 if (control->start) {
2400 ret = control->start(cb);
2405 nlk->cb_running = true;
2406 nlk->dump_done_errno = INT_MAX;
2408 mutex_unlock(nlk->cb_mutex);
2410 ret = netlink_dump(sk);
2417 /* We successfully started a dump, by returning -EINTR we
2418 * signal not to send ACK even if it was requested.
2423 module_put(control->module);
2426 mutex_unlock(nlk->cb_mutex);
2431 EXPORT_SYMBOL(__netlink_dump_start);
2433 void netlink_ack(struct sk_buff *in_skb, struct nlmsghdr *nlh, int err,
2434 const struct netlink_ext_ack *extack)
2436 struct sk_buff *skb;
2437 struct nlmsghdr *rep;
2438 struct nlmsgerr *errmsg;
2439 size_t payload = sizeof(*errmsg);
2441 struct netlink_sock *nlk = nlk_sk(NETLINK_CB(in_skb).sk);
2442 unsigned int flags = 0;
2443 bool nlk_has_extack = nlk->flags & NETLINK_F_EXT_ACK;
2445 /* Error messages get the original request appened, unless the user
2446 * requests to cap the error message, and get extra error data if
2449 if (nlk_has_extack && extack && extack->_msg)
2450 tlvlen += nla_total_size(strlen(extack->_msg) + 1);
2452 if (err && !(nlk->flags & NETLINK_F_CAP_ACK))
2453 payload += nlmsg_len(nlh);
2455 flags |= NLM_F_CAPPED;
2456 if (err && nlk_has_extack && extack && extack->bad_attr)
2457 tlvlen += nla_total_size(sizeof(u32));
2458 if (nlk_has_extack && extack && extack->cookie_len)
2459 tlvlen += nla_total_size(extack->cookie_len);
2460 if (err && nlk_has_extack && extack && extack->policy)
2461 tlvlen += netlink_policy_dump_attr_size_estimate(extack->policy);
2464 flags |= NLM_F_ACK_TLVS;
2466 skb = nlmsg_new(payload + tlvlen, GFP_KERNEL);
2468 NETLINK_CB(in_skb).sk->sk_err = ENOBUFS;
2469 sk_error_report(NETLINK_CB(in_skb).sk);
2473 rep = __nlmsg_put(skb, NETLINK_CB(in_skb).portid, nlh->nlmsg_seq,
2474 NLMSG_ERROR, payload, flags);
2475 errmsg = nlmsg_data(rep);
2476 errmsg->error = err;
2477 memcpy(&errmsg->msg, nlh, payload > sizeof(*errmsg) ? nlh->nlmsg_len : sizeof(*nlh));
2479 if (nlk_has_extack && extack) {
2481 WARN_ON(nla_put_string(skb, NLMSGERR_ATTR_MSG,
2484 if (err && extack->bad_attr &&
2485 !WARN_ON((u8 *)extack->bad_attr < in_skb->data ||
2486 (u8 *)extack->bad_attr >= in_skb->data +
2488 WARN_ON(nla_put_u32(skb, NLMSGERR_ATTR_OFFS,
2489 (u8 *)extack->bad_attr -
2491 if (extack->cookie_len)
2492 WARN_ON(nla_put(skb, NLMSGERR_ATTR_COOKIE,
2493 extack->cookie_len, extack->cookie));
2495 netlink_policy_dump_write_attr(skb, extack->policy,
2496 NLMSGERR_ATTR_POLICY);
2499 nlmsg_end(skb, rep);
2501 nlmsg_unicast(in_skb->sk, skb, NETLINK_CB(in_skb).portid);
2503 EXPORT_SYMBOL(netlink_ack);
2505 int netlink_rcv_skb(struct sk_buff *skb, int (*cb)(struct sk_buff *,
2507 struct netlink_ext_ack *))
2509 struct netlink_ext_ack extack;
2510 struct nlmsghdr *nlh;
2513 while (skb->len >= nlmsg_total_size(0)) {
2516 memset(&extack, 0, sizeof(extack));
2517 nlh = nlmsg_hdr(skb);
2520 if (nlh->nlmsg_len < NLMSG_HDRLEN || skb->len < nlh->nlmsg_len)
2523 /* Only requests are handled by the kernel */
2524 if (!(nlh->nlmsg_flags & NLM_F_REQUEST))
2527 /* Skip control messages */
2528 if (nlh->nlmsg_type < NLMSG_MIN_TYPE)
2531 err = cb(skb, nlh, &extack);
2536 if (nlh->nlmsg_flags & NLM_F_ACK || err)
2537 netlink_ack(skb, nlh, err, &extack);
2540 msglen = NLMSG_ALIGN(nlh->nlmsg_len);
2541 if (msglen > skb->len)
2543 skb_pull(skb, msglen);
2548 EXPORT_SYMBOL(netlink_rcv_skb);
2551 * nlmsg_notify - send a notification netlink message
2552 * @sk: netlink socket to use
2553 * @skb: notification message
2554 * @portid: destination netlink portid for reports or 0
2555 * @group: destination multicast group or 0
2556 * @report: 1 to report back, 0 to disable
2557 * @flags: allocation flags
2559 int nlmsg_notify(struct sock *sk, struct sk_buff *skb, u32 portid,
2560 unsigned int group, int report, gfp_t flags)
2565 int exclude_portid = 0;
2568 refcount_inc(&skb->users);
2569 exclude_portid = portid;
2572 /* errors reported via destination sk->sk_err, but propagate
2573 * delivery errors if NETLINK_BROADCAST_ERROR flag is set */
2574 err = nlmsg_multicast(sk, skb, exclude_portid, group, flags);
2582 err2 = nlmsg_unicast(sk, skb, portid);
2589 EXPORT_SYMBOL(nlmsg_notify);
2591 #ifdef CONFIG_PROC_FS
2592 struct nl_seq_iter {
2593 struct seq_net_private p;
2594 struct rhashtable_iter hti;
2598 static void netlink_walk_start(struct nl_seq_iter *iter)
2600 rhashtable_walk_enter(&nl_table[iter->link].hash, &iter->hti);
2601 rhashtable_walk_start(&iter->hti);
2604 static void netlink_walk_stop(struct nl_seq_iter *iter)
2606 rhashtable_walk_stop(&iter->hti);
2607 rhashtable_walk_exit(&iter->hti);
2610 static void *__netlink_seq_next(struct seq_file *seq)
2612 struct nl_seq_iter *iter = seq->private;
2613 struct netlink_sock *nlk;
2617 nlk = rhashtable_walk_next(&iter->hti);
2620 if (PTR_ERR(nlk) == -EAGAIN)
2629 netlink_walk_stop(iter);
2630 if (++iter->link >= MAX_LINKS)
2633 netlink_walk_start(iter);
2635 } while (sock_net(&nlk->sk) != seq_file_net(seq));
2640 static void *netlink_seq_start(struct seq_file *seq, loff_t *posp)
2643 struct nl_seq_iter *iter = seq->private;
2644 void *obj = SEQ_START_TOKEN;
2649 netlink_walk_start(iter);
2651 for (pos = *posp; pos && obj && !IS_ERR(obj); pos--)
2652 obj = __netlink_seq_next(seq);
2657 static void *netlink_seq_next(struct seq_file *seq, void *v, loff_t *pos)
2660 return __netlink_seq_next(seq);
2663 static void netlink_native_seq_stop(struct seq_file *seq, void *v)
2665 struct nl_seq_iter *iter = seq->private;
2667 if (iter->link >= MAX_LINKS)
2670 netlink_walk_stop(iter);
2674 static int netlink_native_seq_show(struct seq_file *seq, void *v)
2676 if (v == SEQ_START_TOKEN) {
2678 "sk Eth Pid Groups "
2679 "Rmem Wmem Dump Locks Drops Inode\n");
2682 struct netlink_sock *nlk = nlk_sk(s);
2684 seq_printf(seq, "%pK %-3d %-10u %08x %-8d %-8d %-5d %-8d %-8u %-8lu\n",
2688 nlk->groups ? (u32)nlk->groups[0] : 0,
2689 sk_rmem_alloc_get(s),
2690 sk_wmem_alloc_get(s),
2692 refcount_read(&s->sk_refcnt),
2693 atomic_read(&s->sk_drops),
2701 #ifdef CONFIG_BPF_SYSCALL
2702 struct bpf_iter__netlink {
2703 __bpf_md_ptr(struct bpf_iter_meta *, meta);
2704 __bpf_md_ptr(struct netlink_sock *, sk);
2707 DEFINE_BPF_ITER_FUNC(netlink, struct bpf_iter_meta *meta, struct netlink_sock *sk)
2709 static int netlink_prog_seq_show(struct bpf_prog *prog,
2710 struct bpf_iter_meta *meta,
2713 struct bpf_iter__netlink ctx;
2715 meta->seq_num--; /* skip SEQ_START_TOKEN */
2717 ctx.sk = nlk_sk((struct sock *)v);
2718 return bpf_iter_run_prog(prog, &ctx);
2721 static int netlink_seq_show(struct seq_file *seq, void *v)
2723 struct bpf_iter_meta meta;
2724 struct bpf_prog *prog;
2727 prog = bpf_iter_get_info(&meta, false);
2729 return netlink_native_seq_show(seq, v);
2731 if (v != SEQ_START_TOKEN)
2732 return netlink_prog_seq_show(prog, &meta, v);
2737 static void netlink_seq_stop(struct seq_file *seq, void *v)
2739 struct bpf_iter_meta meta;
2740 struct bpf_prog *prog;
2744 prog = bpf_iter_get_info(&meta, true);
2746 (void)netlink_prog_seq_show(prog, &meta, v);
2749 netlink_native_seq_stop(seq, v);
2752 static int netlink_seq_show(struct seq_file *seq, void *v)
2754 return netlink_native_seq_show(seq, v);
2757 static void netlink_seq_stop(struct seq_file *seq, void *v)
2759 netlink_native_seq_stop(seq, v);
2763 static const struct seq_operations netlink_seq_ops = {
2764 .start = netlink_seq_start,
2765 .next = netlink_seq_next,
2766 .stop = netlink_seq_stop,
2767 .show = netlink_seq_show,
2771 int netlink_register_notifier(struct notifier_block *nb)
2773 return blocking_notifier_chain_register(&netlink_chain, nb);
2775 EXPORT_SYMBOL(netlink_register_notifier);
2777 int netlink_unregister_notifier(struct notifier_block *nb)
2779 return blocking_notifier_chain_unregister(&netlink_chain, nb);
2781 EXPORT_SYMBOL(netlink_unregister_notifier);
2783 static const struct proto_ops netlink_ops = {
2784 .family = PF_NETLINK,
2785 .owner = THIS_MODULE,
2786 .release = netlink_release,
2787 .bind = netlink_bind,
2788 .connect = netlink_connect,
2789 .socketpair = sock_no_socketpair,
2790 .accept = sock_no_accept,
2791 .getname = netlink_getname,
2792 .poll = datagram_poll,
2793 .ioctl = netlink_ioctl,
2794 .listen = sock_no_listen,
2795 .shutdown = sock_no_shutdown,
2796 .setsockopt = netlink_setsockopt,
2797 .getsockopt = netlink_getsockopt,
2798 .sendmsg = netlink_sendmsg,
2799 .recvmsg = netlink_recvmsg,
2800 .mmap = sock_no_mmap,
2801 .sendpage = sock_no_sendpage,
2804 static const struct net_proto_family netlink_family_ops = {
2805 .family = PF_NETLINK,
2806 .create = netlink_create,
2807 .owner = THIS_MODULE, /* for consistency 8) */
2810 static int __net_init netlink_net_init(struct net *net)
2812 #ifdef CONFIG_PROC_FS
2813 if (!proc_create_net("netlink", 0, net->proc_net, &netlink_seq_ops,
2814 sizeof(struct nl_seq_iter)))
2820 static void __net_exit netlink_net_exit(struct net *net)
2822 #ifdef CONFIG_PROC_FS
2823 remove_proc_entry("netlink", net->proc_net);
2827 static void __init netlink_add_usersock_entry(void)
2829 struct listeners *listeners;
2832 listeners = kzalloc(sizeof(*listeners) + NLGRPSZ(groups), GFP_KERNEL);
2834 panic("netlink_add_usersock_entry: Cannot allocate listeners\n");
2836 netlink_table_grab();
2838 nl_table[NETLINK_USERSOCK].groups = groups;
2839 rcu_assign_pointer(nl_table[NETLINK_USERSOCK].listeners, listeners);
2840 nl_table[NETLINK_USERSOCK].module = THIS_MODULE;
2841 nl_table[NETLINK_USERSOCK].registered = 1;
2842 nl_table[NETLINK_USERSOCK].flags = NL_CFG_F_NONROOT_SEND;
2844 netlink_table_ungrab();
2847 static struct pernet_operations __net_initdata netlink_net_ops = {
2848 .init = netlink_net_init,
2849 .exit = netlink_net_exit,
2852 static inline u32 netlink_hash(const void *data, u32 len, u32 seed)
2854 const struct netlink_sock *nlk = data;
2855 struct netlink_compare_arg arg;
2857 netlink_compare_arg_init(&arg, sock_net(&nlk->sk), nlk->portid);
2858 return jhash2((u32 *)&arg, netlink_compare_arg_len / sizeof(u32), seed);
2861 static const struct rhashtable_params netlink_rhashtable_params = {
2862 .head_offset = offsetof(struct netlink_sock, node),
2863 .key_len = netlink_compare_arg_len,
2864 .obj_hashfn = netlink_hash,
2865 .obj_cmpfn = netlink_compare,
2866 .automatic_shrinking = true,
2869 #if defined(CONFIG_BPF_SYSCALL) && defined(CONFIG_PROC_FS)
2870 BTF_ID_LIST(btf_netlink_sock_id)
2871 BTF_ID(struct, netlink_sock)
2873 static const struct bpf_iter_seq_info netlink_seq_info = {
2874 .seq_ops = &netlink_seq_ops,
2875 .init_seq_private = bpf_iter_init_seq_net,
2876 .fini_seq_private = bpf_iter_fini_seq_net,
2877 .seq_priv_size = sizeof(struct nl_seq_iter),
2880 static struct bpf_iter_reg netlink_reg_info = {
2881 .target = "netlink",
2882 .ctx_arg_info_size = 1,
2884 { offsetof(struct bpf_iter__netlink, sk),
2885 PTR_TO_BTF_ID_OR_NULL },
2887 .seq_info = &netlink_seq_info,
2890 static int __init bpf_iter_register(void)
2892 netlink_reg_info.ctx_arg_info[0].btf_id = *btf_netlink_sock_id;
2893 return bpf_iter_reg_target(&netlink_reg_info);
2897 static int __init netlink_proto_init(void)
2900 int err = proto_register(&netlink_proto, 0);
2905 #if defined(CONFIG_BPF_SYSCALL) && defined(CONFIG_PROC_FS)
2906 err = bpf_iter_register();
2911 BUILD_BUG_ON(sizeof(struct netlink_skb_parms) > sizeof_field(struct sk_buff, cb));
2913 nl_table = kcalloc(MAX_LINKS, sizeof(*nl_table), GFP_KERNEL);
2917 for (i = 0; i < MAX_LINKS; i++) {
2918 if (rhashtable_init(&nl_table[i].hash,
2919 &netlink_rhashtable_params) < 0) {
2921 rhashtable_destroy(&nl_table[i].hash);
2927 netlink_add_usersock_entry();
2929 sock_register(&netlink_family_ops);
2930 register_pernet_subsys(&netlink_net_ops);
2931 register_pernet_subsys(&netlink_tap_net_ops);
2932 /* The netlink device handler may be needed early. */
2937 panic("netlink_init: Cannot allocate nl_table\n");
2940 core_initcall(netlink_proto_init);