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/bpf.h>
24 #include <linux/capability.h>
25 #include <linux/kernel.h>
26 #include <linux/filter.h>
27 #include <linux/init.h>
28 #include <linux/signal.h>
29 #include <linux/sched.h>
30 #include <linux/errno.h>
31 #include <linux/string.h>
32 #include <linux/stat.h>
33 #include <linux/socket.h>
35 #include <linux/fcntl.h>
36 #include <linux/termios.h>
37 #include <linux/sockios.h>
38 #include <linux/net.h>
40 #include <linux/slab.h>
41 #include <linux/uaccess.h>
42 #include <linux/skbuff.h>
43 #include <linux/netdevice.h>
44 #include <linux/rtnetlink.h>
45 #include <linux/proc_fs.h>
46 #include <linux/seq_file.h>
47 #include <linux/notifier.h>
48 #include <linux/security.h>
49 #include <linux/jhash.h>
50 #include <linux/jiffies.h>
51 #include <linux/random.h>
52 #include <linux/bitops.h>
54 #include <linux/types.h>
55 #include <linux/audit.h>
56 #include <linux/mutex.h>
57 #include <linux/vmalloc.h>
58 #include <linux/if_arp.h>
59 #include <linux/rhashtable.h>
60 #include <asm/cacheflush.h>
61 #include <linux/hash.h>
62 #include <linux/genetlink.h>
63 #include <linux/net_namespace.h>
64 #include <linux/nospec.h>
65 #include <linux/btf_ids.h>
67 #include <net/net_namespace.h>
68 #include <net/netns/generic.h>
71 #include <net/netlink.h>
72 #define CREATE_TRACE_POINTS
73 #include <trace/events/netlink.h>
75 #include "af_netlink.h"
79 unsigned long masks[];
83 #define NETLINK_S_CONGESTED 0x0
85 static inline int netlink_is_kernel(struct sock *sk)
87 return nlk_sk(sk)->flags & NETLINK_F_KERNEL_SOCKET;
90 struct netlink_table *nl_table __read_mostly;
91 EXPORT_SYMBOL_GPL(nl_table);
93 static DECLARE_WAIT_QUEUE_HEAD(nl_table_wait);
95 static struct lock_class_key nlk_cb_mutex_keys[MAX_LINKS];
97 static const char *const nlk_cb_mutex_key_strings[MAX_LINKS + 1] = {
100 "nlk_cb_mutex-USERSOCK",
101 "nlk_cb_mutex-FIREWALL",
102 "nlk_cb_mutex-SOCK_DIAG",
103 "nlk_cb_mutex-NFLOG",
105 "nlk_cb_mutex-SELINUX",
106 "nlk_cb_mutex-ISCSI",
107 "nlk_cb_mutex-AUDIT",
108 "nlk_cb_mutex-FIB_LOOKUP",
109 "nlk_cb_mutex-CONNECTOR",
110 "nlk_cb_mutex-NETFILTER",
111 "nlk_cb_mutex-IP6_FW",
112 "nlk_cb_mutex-DNRTMSG",
113 "nlk_cb_mutex-KOBJECT_UEVENT",
114 "nlk_cb_mutex-GENERIC",
116 "nlk_cb_mutex-SCSITRANSPORT",
117 "nlk_cb_mutex-ECRYPTFS",
119 "nlk_cb_mutex-CRYPTO",
130 "nlk_cb_mutex-MAX_LINKS"
133 static int netlink_dump(struct sock *sk);
135 /* nl_table locking explained:
136 * Lookup and traversal are protected with an RCU read-side lock. Insertion
137 * and removal are protected with per bucket lock while using RCU list
138 * modification primitives and may run in parallel to RCU protected lookups.
139 * Destruction of the Netlink socket may only occur *after* nl_table_lock has
140 * been acquired * either during or after the socket has been removed from
141 * the list and after an RCU grace period.
143 DEFINE_RWLOCK(nl_table_lock);
144 EXPORT_SYMBOL_GPL(nl_table_lock);
145 static atomic_t nl_table_users = ATOMIC_INIT(0);
147 #define nl_deref_protected(X) rcu_dereference_protected(X, lockdep_is_held(&nl_table_lock));
149 static BLOCKING_NOTIFIER_HEAD(netlink_chain);
152 static const struct rhashtable_params netlink_rhashtable_params;
154 void do_trace_netlink_extack(const char *msg)
156 trace_netlink_extack(msg);
158 EXPORT_SYMBOL(do_trace_netlink_extack);
160 static inline u32 netlink_group_mask(u32 group)
164 return group ? 1 << (group - 1) : 0;
167 static struct sk_buff *netlink_to_full_skb(const struct sk_buff *skb,
170 unsigned int len = skb_end_offset(skb);
173 new = alloc_skb(len, gfp_mask);
177 NETLINK_CB(new).portid = NETLINK_CB(skb).portid;
178 NETLINK_CB(new).dst_group = NETLINK_CB(skb).dst_group;
179 NETLINK_CB(new).creds = NETLINK_CB(skb).creds;
181 skb_put_data(new, skb->data, len);
185 static unsigned int netlink_tap_net_id;
187 struct netlink_tap_net {
188 struct list_head netlink_tap_all;
189 struct mutex netlink_tap_lock;
192 int netlink_add_tap(struct netlink_tap *nt)
194 struct net *net = dev_net(nt->dev);
195 struct netlink_tap_net *nn = net_generic(net, netlink_tap_net_id);
197 if (unlikely(nt->dev->type != ARPHRD_NETLINK))
200 mutex_lock(&nn->netlink_tap_lock);
201 list_add_rcu(&nt->list, &nn->netlink_tap_all);
202 mutex_unlock(&nn->netlink_tap_lock);
204 __module_get(nt->module);
208 EXPORT_SYMBOL_GPL(netlink_add_tap);
210 static int __netlink_remove_tap(struct netlink_tap *nt)
212 struct net *net = dev_net(nt->dev);
213 struct netlink_tap_net *nn = net_generic(net, netlink_tap_net_id);
215 struct netlink_tap *tmp;
217 mutex_lock(&nn->netlink_tap_lock);
219 list_for_each_entry(tmp, &nn->netlink_tap_all, list) {
221 list_del_rcu(&nt->list);
227 pr_warn("__netlink_remove_tap: %p not found\n", nt);
229 mutex_unlock(&nn->netlink_tap_lock);
232 module_put(nt->module);
234 return found ? 0 : -ENODEV;
237 int netlink_remove_tap(struct netlink_tap *nt)
241 ret = __netlink_remove_tap(nt);
246 EXPORT_SYMBOL_GPL(netlink_remove_tap);
248 static __net_init int netlink_tap_init_net(struct net *net)
250 struct netlink_tap_net *nn = net_generic(net, netlink_tap_net_id);
252 INIT_LIST_HEAD(&nn->netlink_tap_all);
253 mutex_init(&nn->netlink_tap_lock);
257 static struct pernet_operations netlink_tap_net_ops = {
258 .init = netlink_tap_init_net,
259 .id = &netlink_tap_net_id,
260 .size = sizeof(struct netlink_tap_net),
263 static bool netlink_filter_tap(const struct sk_buff *skb)
265 struct sock *sk = skb->sk;
267 /* We take the more conservative approach and
268 * whitelist socket protocols that may pass.
270 switch (sk->sk_protocol) {
272 case NETLINK_USERSOCK:
273 case NETLINK_SOCK_DIAG:
276 case NETLINK_FIB_LOOKUP:
277 case NETLINK_NETFILTER:
278 case NETLINK_GENERIC:
285 static int __netlink_deliver_tap_skb(struct sk_buff *skb,
286 struct net_device *dev)
288 struct sk_buff *nskb;
289 struct sock *sk = skb->sk;
292 if (!net_eq(dev_net(dev), sock_net(sk)))
297 if (is_vmalloc_addr(skb->head))
298 nskb = netlink_to_full_skb(skb, GFP_ATOMIC);
300 nskb = skb_clone(skb, GFP_ATOMIC);
303 nskb->protocol = htons((u16) sk->sk_protocol);
304 nskb->pkt_type = netlink_is_kernel(sk) ?
305 PACKET_KERNEL : PACKET_USER;
306 skb_reset_network_header(nskb);
307 ret = dev_queue_xmit(nskb);
308 if (unlikely(ret > 0))
309 ret = net_xmit_errno(ret);
316 static void __netlink_deliver_tap(struct sk_buff *skb, struct netlink_tap_net *nn)
319 struct netlink_tap *tmp;
321 if (!netlink_filter_tap(skb))
324 list_for_each_entry_rcu(tmp, &nn->netlink_tap_all, list) {
325 ret = __netlink_deliver_tap_skb(skb, tmp->dev);
331 static void netlink_deliver_tap(struct net *net, struct sk_buff *skb)
333 struct netlink_tap_net *nn = net_generic(net, netlink_tap_net_id);
337 if (unlikely(!list_empty(&nn->netlink_tap_all)))
338 __netlink_deliver_tap(skb, nn);
343 static void netlink_deliver_tap_kernel(struct sock *dst, struct sock *src,
346 if (!(netlink_is_kernel(dst) && netlink_is_kernel(src)))
347 netlink_deliver_tap(sock_net(dst), skb);
350 static void netlink_overrun(struct sock *sk)
352 struct netlink_sock *nlk = nlk_sk(sk);
354 if (!(nlk->flags & NETLINK_F_RECV_NO_ENOBUFS)) {
355 if (!test_and_set_bit(NETLINK_S_CONGESTED,
356 &nlk_sk(sk)->state)) {
357 sk->sk_err = ENOBUFS;
361 atomic_inc(&sk->sk_drops);
364 static void netlink_rcv_wake(struct sock *sk)
366 struct netlink_sock *nlk = nlk_sk(sk);
368 if (skb_queue_empty_lockless(&sk->sk_receive_queue))
369 clear_bit(NETLINK_S_CONGESTED, &nlk->state);
370 if (!test_bit(NETLINK_S_CONGESTED, &nlk->state))
371 wake_up_interruptible(&nlk->wait);
374 static void netlink_skb_destructor(struct sk_buff *skb)
376 if (is_vmalloc_addr(skb->head)) {
378 !atomic_dec_return(&(skb_shinfo(skb)->dataref)))
387 static void netlink_skb_set_owner_r(struct sk_buff *skb, struct sock *sk)
389 WARN_ON(skb->sk != NULL);
391 skb->destructor = netlink_skb_destructor;
392 atomic_add(skb->truesize, &sk->sk_rmem_alloc);
393 sk_mem_charge(sk, skb->truesize);
396 static void netlink_sock_destruct(struct sock *sk)
398 struct netlink_sock *nlk = nlk_sk(sk);
400 if (nlk->cb_running) {
402 nlk->cb.done(&nlk->cb);
403 module_put(nlk->cb.module);
404 kfree_skb(nlk->cb.skb);
407 skb_queue_purge(&sk->sk_receive_queue);
409 if (!sock_flag(sk, SOCK_DEAD)) {
410 printk(KERN_ERR "Freeing alive netlink socket %p\n", sk);
414 WARN_ON(atomic_read(&sk->sk_rmem_alloc));
415 WARN_ON(refcount_read(&sk->sk_wmem_alloc));
416 WARN_ON(nlk_sk(sk)->groups);
419 static void netlink_sock_destruct_work(struct work_struct *work)
421 struct netlink_sock *nlk = container_of(work, struct netlink_sock,
427 /* This lock without WQ_FLAG_EXCLUSIVE is good on UP and it is _very_ bad on
428 * SMP. Look, when several writers sleep and reader wakes them up, all but one
429 * immediately hit write lock and grab all the cpus. Exclusive sleep solves
430 * this, _but_ remember, it adds useless work on UP machines.
433 void netlink_table_grab(void)
434 __acquires(nl_table_lock)
438 write_lock_irq(&nl_table_lock);
440 if (atomic_read(&nl_table_users)) {
441 DECLARE_WAITQUEUE(wait, current);
443 add_wait_queue_exclusive(&nl_table_wait, &wait);
445 set_current_state(TASK_UNINTERRUPTIBLE);
446 if (atomic_read(&nl_table_users) == 0)
448 write_unlock_irq(&nl_table_lock);
450 write_lock_irq(&nl_table_lock);
453 __set_current_state(TASK_RUNNING);
454 remove_wait_queue(&nl_table_wait, &wait);
458 void netlink_table_ungrab(void)
459 __releases(nl_table_lock)
461 write_unlock_irq(&nl_table_lock);
462 wake_up(&nl_table_wait);
466 netlink_lock_table(void)
470 /* read_lock() synchronizes us to netlink_table_grab */
472 read_lock_irqsave(&nl_table_lock, flags);
473 atomic_inc(&nl_table_users);
474 read_unlock_irqrestore(&nl_table_lock, flags);
478 netlink_unlock_table(void)
480 if (atomic_dec_and_test(&nl_table_users))
481 wake_up(&nl_table_wait);
484 struct netlink_compare_arg
490 /* Doing sizeof directly may yield 4 extra bytes on 64-bit. */
491 #define netlink_compare_arg_len \
492 (offsetof(struct netlink_compare_arg, portid) + sizeof(u32))
494 static inline int netlink_compare(struct rhashtable_compare_arg *arg,
497 const struct netlink_compare_arg *x = arg->key;
498 const struct netlink_sock *nlk = ptr;
500 return nlk->portid != x->portid ||
501 !net_eq(sock_net(&nlk->sk), read_pnet(&x->pnet));
504 static void netlink_compare_arg_init(struct netlink_compare_arg *arg,
505 struct net *net, u32 portid)
507 memset(arg, 0, sizeof(*arg));
508 write_pnet(&arg->pnet, net);
509 arg->portid = portid;
512 static struct sock *__netlink_lookup(struct netlink_table *table, u32 portid,
515 struct netlink_compare_arg arg;
517 netlink_compare_arg_init(&arg, net, portid);
518 return rhashtable_lookup_fast(&table->hash, &arg,
519 netlink_rhashtable_params);
522 static int __netlink_insert(struct netlink_table *table, struct sock *sk)
524 struct netlink_compare_arg arg;
526 netlink_compare_arg_init(&arg, sock_net(sk), nlk_sk(sk)->portid);
527 return rhashtable_lookup_insert_key(&table->hash, &arg,
529 netlink_rhashtable_params);
532 static struct sock *netlink_lookup(struct net *net, int protocol, u32 portid)
534 struct netlink_table *table = &nl_table[protocol];
538 sk = __netlink_lookup(table, portid, net);
546 static const struct proto_ops netlink_ops;
549 netlink_update_listeners(struct sock *sk)
551 struct netlink_table *tbl = &nl_table[sk->sk_protocol];
554 struct listeners *listeners;
556 listeners = nl_deref_protected(tbl->listeners);
560 for (i = 0; i < NLGRPLONGS(tbl->groups); i++) {
562 sk_for_each_bound(sk, &tbl->mc_list) {
563 if (i < NLGRPLONGS(nlk_sk(sk)->ngroups))
564 mask |= nlk_sk(sk)->groups[i];
566 listeners->masks[i] = mask;
568 /* this function is only called with the netlink table "grabbed", which
569 * makes sure updates are visible before bind or setsockopt return. */
572 static int netlink_insert(struct sock *sk, u32 portid)
574 struct netlink_table *table = &nl_table[sk->sk_protocol];
579 err = nlk_sk(sk)->portid == portid ? 0 : -EBUSY;
580 if (nlk_sk(sk)->bound)
583 /* portid can be read locklessly from netlink_getname(). */
584 WRITE_ONCE(nlk_sk(sk)->portid, portid);
588 err = __netlink_insert(table, sk);
590 /* In case the hashtable backend returns with -EBUSY
591 * from here, it must not escape to the caller.
593 if (unlikely(err == -EBUSY))
601 /* We need to ensure that the socket is hashed and visible. */
603 /* Paired with lockless reads from netlink_bind(),
604 * netlink_connect() and netlink_sendmsg().
606 WRITE_ONCE(nlk_sk(sk)->bound, portid);
613 static void netlink_remove(struct sock *sk)
615 struct netlink_table *table;
617 table = &nl_table[sk->sk_protocol];
618 if (!rhashtable_remove_fast(&table->hash, &nlk_sk(sk)->node,
619 netlink_rhashtable_params)) {
620 WARN_ON(refcount_read(&sk->sk_refcnt) == 1);
624 netlink_table_grab();
625 if (nlk_sk(sk)->subscriptions) {
626 __sk_del_bind_node(sk);
627 netlink_update_listeners(sk);
629 if (sk->sk_protocol == NETLINK_GENERIC)
630 atomic_inc(&genl_sk_destructing_cnt);
631 netlink_table_ungrab();
634 static struct proto netlink_proto = {
636 .owner = THIS_MODULE,
637 .obj_size = sizeof(struct netlink_sock),
640 static int __netlink_create(struct net *net, struct socket *sock,
641 struct mutex *cb_mutex, int protocol,
645 struct netlink_sock *nlk;
647 sock->ops = &netlink_ops;
649 sk = sk_alloc(net, PF_NETLINK, GFP_KERNEL, &netlink_proto, kern);
653 sock_init_data(sock, sk);
657 nlk->cb_mutex = cb_mutex;
659 nlk->cb_mutex = &nlk->cb_def_mutex;
660 mutex_init(nlk->cb_mutex);
661 lockdep_set_class_and_name(nlk->cb_mutex,
662 nlk_cb_mutex_keys + protocol,
663 nlk_cb_mutex_key_strings[protocol]);
665 init_waitqueue_head(&nlk->wait);
667 sk->sk_destruct = netlink_sock_destruct;
668 sk->sk_protocol = protocol;
672 static int netlink_create(struct net *net, struct socket *sock, int protocol,
675 struct module *module = NULL;
676 struct mutex *cb_mutex;
677 struct netlink_sock *nlk;
678 int (*bind)(struct net *net, int group);
679 void (*unbind)(struct net *net, int group);
682 sock->state = SS_UNCONNECTED;
684 if (sock->type != SOCK_RAW && sock->type != SOCK_DGRAM)
685 return -ESOCKTNOSUPPORT;
687 if (protocol < 0 || protocol >= MAX_LINKS)
688 return -EPROTONOSUPPORT;
689 protocol = array_index_nospec(protocol, MAX_LINKS);
691 netlink_lock_table();
692 #ifdef CONFIG_MODULES
693 if (!nl_table[protocol].registered) {
694 netlink_unlock_table();
695 request_module("net-pf-%d-proto-%d", PF_NETLINK, protocol);
696 netlink_lock_table();
699 if (nl_table[protocol].registered &&
700 try_module_get(nl_table[protocol].module))
701 module = nl_table[protocol].module;
703 err = -EPROTONOSUPPORT;
704 cb_mutex = nl_table[protocol].cb_mutex;
705 bind = nl_table[protocol].bind;
706 unbind = nl_table[protocol].unbind;
707 netlink_unlock_table();
712 err = __netlink_create(net, sock, cb_mutex, protocol, kern);
716 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();
816 sock_prot_inuse_add(sock_net(sk), &netlink_proto, -1);
817 call_rcu(&nlk->rcu, deferred_put_nlk_sk);
821 static int netlink_autobind(struct socket *sock)
823 struct sock *sk = sock->sk;
824 struct net *net = sock_net(sk);
825 struct netlink_table *table = &nl_table[sk->sk_protocol];
826 s32 portid = task_tgid_vnr(current);
834 ok = !__netlink_lookup(table, portid, net);
837 /* Bind collision, search negative portid values. */
839 /* rover will be in range [S32_MIN, -4097] */
840 rover = S32_MIN + prandom_u32_max(-4096 - S32_MIN);
841 else if (rover >= -4096)
847 err = netlink_insert(sk, portid);
848 if (err == -EADDRINUSE)
851 /* If 2 threads race to autobind, that is fine. */
859 * __netlink_ns_capable - General netlink message capability test
860 * @nsp: NETLINK_CB of the socket buffer holding a netlink command from userspace.
861 * @user_ns: The user namespace of the capability to use
862 * @cap: The capability to use
864 * Test to see if the opener of the socket we received the message
865 * from had when the netlink socket was created and the sender of the
866 * message has the capability @cap in the user namespace @user_ns.
868 bool __netlink_ns_capable(const struct netlink_skb_parms *nsp,
869 struct user_namespace *user_ns, int cap)
871 return ((nsp->flags & NETLINK_SKB_DST) ||
872 file_ns_capable(nsp->sk->sk_socket->file, user_ns, cap)) &&
873 ns_capable(user_ns, cap);
875 EXPORT_SYMBOL(__netlink_ns_capable);
878 * netlink_ns_capable - General netlink message capability test
879 * @skb: socket buffer holding a netlink command from userspace
880 * @user_ns: The user namespace of the capability to use
881 * @cap: The capability to use
883 * Test to see if the opener of the socket we received the message
884 * from had when the netlink socket was created and the sender of the
885 * message has the capability @cap in the user namespace @user_ns.
887 bool netlink_ns_capable(const struct sk_buff *skb,
888 struct user_namespace *user_ns, int cap)
890 return __netlink_ns_capable(&NETLINK_CB(skb), user_ns, cap);
892 EXPORT_SYMBOL(netlink_ns_capable);
895 * netlink_capable - Netlink global message capability test
896 * @skb: socket buffer holding a netlink command from userspace
897 * @cap: The capability to use
899 * Test to see if the opener of the socket we received the message
900 * from had when the netlink socket was created and the sender of the
901 * message has the capability @cap in all user namespaces.
903 bool netlink_capable(const struct sk_buff *skb, int cap)
905 return netlink_ns_capable(skb, &init_user_ns, cap);
907 EXPORT_SYMBOL(netlink_capable);
910 * netlink_net_capable - Netlink network namespace message capability test
911 * @skb: socket buffer holding a netlink command from userspace
912 * @cap: The capability to use
914 * Test to see if the opener of the socket we received the message
915 * from had when the netlink socket was created and the sender of the
916 * message has the capability @cap over the network namespace of
917 * the socket we received the message from.
919 bool netlink_net_capable(const struct sk_buff *skb, int cap)
921 return netlink_ns_capable(skb, sock_net(skb->sk)->user_ns, cap);
923 EXPORT_SYMBOL(netlink_net_capable);
925 static inline int netlink_allowed(const struct socket *sock, unsigned int flag)
927 return (nl_table[sock->sk->sk_protocol].flags & flag) ||
928 ns_capable(sock_net(sock->sk)->user_ns, CAP_NET_ADMIN);
932 netlink_update_subscriptions(struct sock *sk, unsigned int subscriptions)
934 struct netlink_sock *nlk = nlk_sk(sk);
936 if (nlk->subscriptions && !subscriptions)
937 __sk_del_bind_node(sk);
938 else if (!nlk->subscriptions && subscriptions)
939 sk_add_bind_node(sk, &nl_table[sk->sk_protocol].mc_list);
940 nlk->subscriptions = subscriptions;
943 static int netlink_realloc_groups(struct sock *sk)
945 struct netlink_sock *nlk = nlk_sk(sk);
947 unsigned long *new_groups;
950 netlink_table_grab();
952 groups = nl_table[sk->sk_protocol].groups;
953 if (!nl_table[sk->sk_protocol].registered) {
958 if (nlk->ngroups >= groups)
961 new_groups = krealloc(nlk->groups, NLGRPSZ(groups), GFP_ATOMIC);
962 if (new_groups == NULL) {
966 memset((char *)new_groups + NLGRPSZ(nlk->ngroups), 0,
967 NLGRPSZ(groups) - NLGRPSZ(nlk->ngroups));
969 nlk->groups = new_groups;
970 nlk->ngroups = groups;
972 netlink_table_ungrab();
976 static void netlink_undo_bind(int group, long unsigned int groups,
979 struct netlink_sock *nlk = nlk_sk(sk);
982 if (!nlk->netlink_unbind)
985 for (undo = 0; undo < group; undo++)
986 if (test_bit(undo, &groups))
987 nlk->netlink_unbind(sock_net(sk), undo + 1);
990 static int netlink_bind(struct socket *sock, struct sockaddr *addr,
993 struct sock *sk = sock->sk;
994 struct net *net = sock_net(sk);
995 struct netlink_sock *nlk = nlk_sk(sk);
996 struct sockaddr_nl *nladdr = (struct sockaddr_nl *)addr;
998 unsigned long groups;
1001 if (addr_len < sizeof(struct sockaddr_nl))
1004 if (nladdr->nl_family != AF_NETLINK)
1006 groups = nladdr->nl_groups;
1008 /* Only superuser is allowed to listen multicasts */
1010 if (!netlink_allowed(sock, NL_CFG_F_NONROOT_RECV))
1012 err = netlink_realloc_groups(sk);
1017 if (nlk->ngroups < BITS_PER_LONG)
1018 groups &= (1UL << nlk->ngroups) - 1;
1020 /* Paired with WRITE_ONCE() in netlink_insert() */
1021 bound = READ_ONCE(nlk->bound);
1023 /* Ensure nlk->portid is up-to-date. */
1026 if (nladdr->nl_pid != nlk->portid)
1030 if (nlk->netlink_bind && groups) {
1033 /* nl_groups is a u32, so cap the maximum groups we can bind */
1034 for (group = 0; group < BITS_PER_TYPE(u32); group++) {
1035 if (!test_bit(group, &groups))
1037 err = nlk->netlink_bind(net, group + 1);
1040 netlink_undo_bind(group, groups, sk);
1045 /* No need for barriers here as we return to user-space without
1046 * using any of the bound attributes.
1048 netlink_lock_table();
1050 err = nladdr->nl_pid ?
1051 netlink_insert(sk, nladdr->nl_pid) :
1052 netlink_autobind(sock);
1054 netlink_undo_bind(BITS_PER_TYPE(u32), groups, sk);
1059 if (!groups && (nlk->groups == NULL || !(u32)nlk->groups[0]))
1061 netlink_unlock_table();
1063 netlink_table_grab();
1064 netlink_update_subscriptions(sk, nlk->subscriptions +
1066 hweight32(nlk->groups[0]));
1067 nlk->groups[0] = (nlk->groups[0] & ~0xffffffffUL) | groups;
1068 netlink_update_listeners(sk);
1069 netlink_table_ungrab();
1074 netlink_unlock_table();
1078 static int netlink_connect(struct socket *sock, struct sockaddr *addr,
1079 int alen, int flags)
1082 struct sock *sk = sock->sk;
1083 struct netlink_sock *nlk = nlk_sk(sk);
1084 struct sockaddr_nl *nladdr = (struct sockaddr_nl *)addr;
1086 if (alen < sizeof(addr->sa_family))
1089 if (addr->sa_family == AF_UNSPEC) {
1090 /* paired with READ_ONCE() in netlink_getsockbyportid() */
1091 WRITE_ONCE(sk->sk_state, NETLINK_UNCONNECTED);
1092 /* dst_portid and dst_group can be read locklessly */
1093 WRITE_ONCE(nlk->dst_portid, 0);
1094 WRITE_ONCE(nlk->dst_group, 0);
1097 if (addr->sa_family != AF_NETLINK)
1100 if (alen < sizeof(struct sockaddr_nl))
1103 if ((nladdr->nl_groups || nladdr->nl_pid) &&
1104 !netlink_allowed(sock, NL_CFG_F_NONROOT_SEND))
1107 /* No need for barriers here as we return to user-space without
1108 * using any of the bound attributes.
1109 * Paired with WRITE_ONCE() in netlink_insert().
1111 if (!READ_ONCE(nlk->bound))
1112 err = netlink_autobind(sock);
1115 /* paired with READ_ONCE() in netlink_getsockbyportid() */
1116 WRITE_ONCE(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 and sk_state can be changed in netlink_connect() */
1169 if (READ_ONCE(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;
1426 static void do_one_broadcast(struct sock *sk,
1427 struct netlink_broadcast_data *p)
1429 struct netlink_sock *nlk = nlk_sk(sk);
1432 if (p->exclude_sk == sk)
1435 if (nlk->portid == p->portid || p->group - 1 >= nlk->ngroups ||
1436 !test_bit(p->group - 1, nlk->groups))
1439 if (!net_eq(sock_net(sk), p->net)) {
1440 if (!(nlk->flags & NETLINK_F_LISTEN_ALL_NSID))
1443 if (!peernet_has_id(sock_net(sk), p->net))
1446 if (!file_ns_capable(sk->sk_socket->file, p->net->user_ns,
1452 netlink_overrun(sk);
1457 if (p->skb2 == NULL) {
1458 if (skb_shared(p->skb)) {
1459 p->skb2 = skb_clone(p->skb, p->allocation);
1461 p->skb2 = skb_get(p->skb);
1463 * skb ownership may have been set when
1464 * delivered to a previous socket.
1466 skb_orphan(p->skb2);
1469 if (p->skb2 == NULL) {
1470 netlink_overrun(sk);
1471 /* Clone failed. Notify ALL listeners. */
1473 if (nlk->flags & NETLINK_F_BROADCAST_SEND_ERROR)
1474 p->delivery_failure = 1;
1477 if (sk_filter(sk, p->skb2)) {
1482 NETLINK_CB(p->skb2).nsid = peernet2id(sock_net(sk), p->net);
1483 if (NETLINK_CB(p->skb2).nsid != NETNSA_NSID_NOT_ASSIGNED)
1484 NETLINK_CB(p->skb2).nsid_is_set = true;
1485 val = netlink_broadcast_deliver(sk, p->skb2);
1487 netlink_overrun(sk);
1488 if (nlk->flags & NETLINK_F_BROADCAST_SEND_ERROR)
1489 p->delivery_failure = 1;
1491 p->congested |= val;
1499 int netlink_broadcast(struct sock *ssk, struct sk_buff *skb, u32 portid,
1500 u32 group, gfp_t allocation)
1502 struct net *net = sock_net(ssk);
1503 struct netlink_broadcast_data info;
1506 skb = netlink_trim(skb, allocation);
1508 info.exclude_sk = ssk;
1510 info.portid = portid;
1513 info.delivery_failure = 0;
1516 info.allocation = allocation;
1520 /* While we sleep in clone, do not allow to change socket list */
1522 netlink_lock_table();
1524 sk_for_each_bound(sk, &nl_table[ssk->sk_protocol].mc_list)
1525 do_one_broadcast(sk, &info);
1529 netlink_unlock_table();
1531 if (info.delivery_failure) {
1532 kfree_skb(info.skb2);
1535 consume_skb(info.skb2);
1537 if (info.delivered) {
1538 if (info.congested && gfpflags_allow_blocking(allocation))
1544 EXPORT_SYMBOL(netlink_broadcast);
1546 struct netlink_set_err_data {
1547 struct sock *exclude_sk;
1553 static int do_one_set_err(struct sock *sk, struct netlink_set_err_data *p)
1555 struct netlink_sock *nlk = nlk_sk(sk);
1558 if (sk == p->exclude_sk)
1561 if (!net_eq(sock_net(sk), sock_net(p->exclude_sk)))
1564 if (nlk->portid == p->portid || p->group - 1 >= nlk->ngroups ||
1565 !test_bit(p->group - 1, nlk->groups))
1568 if (p->code == ENOBUFS && nlk->flags & NETLINK_F_RECV_NO_ENOBUFS) {
1573 sk->sk_err = p->code;
1574 sk_error_report(sk);
1580 * netlink_set_err - report error to broadcast listeners
1581 * @ssk: the kernel netlink socket, as returned by netlink_kernel_create()
1582 * @portid: the PORTID of a process that we want to skip (if any)
1583 * @group: the broadcast group that will notice the error
1584 * @code: error code, must be negative (as usual in kernelspace)
1586 * This function returns the number of broadcast listeners that have set the
1587 * NETLINK_NO_ENOBUFS socket option.
1589 int netlink_set_err(struct sock *ssk, u32 portid, u32 group, int code)
1591 struct netlink_set_err_data info;
1595 info.exclude_sk = ssk;
1596 info.portid = portid;
1598 /* sk->sk_err wants a positive error value */
1601 read_lock(&nl_table_lock);
1603 sk_for_each_bound(sk, &nl_table[ssk->sk_protocol].mc_list)
1604 ret += do_one_set_err(sk, &info);
1606 read_unlock(&nl_table_lock);
1609 EXPORT_SYMBOL(netlink_set_err);
1611 /* must be called with netlink table grabbed */
1612 static void netlink_update_socket_mc(struct netlink_sock *nlk,
1616 int old, new = !!is_new, subscriptions;
1618 old = test_bit(group - 1, nlk->groups);
1619 subscriptions = nlk->subscriptions - old + new;
1621 __set_bit(group - 1, nlk->groups);
1623 __clear_bit(group - 1, nlk->groups);
1624 netlink_update_subscriptions(&nlk->sk, subscriptions);
1625 netlink_update_listeners(&nlk->sk);
1628 static int netlink_setsockopt(struct socket *sock, int level, int optname,
1629 sockptr_t optval, unsigned int optlen)
1631 struct sock *sk = sock->sk;
1632 struct netlink_sock *nlk = nlk_sk(sk);
1633 unsigned int val = 0;
1636 if (level != SOL_NETLINK)
1637 return -ENOPROTOOPT;
1639 if (optlen >= sizeof(int) &&
1640 copy_from_sockptr(&val, optval, sizeof(val)))
1644 case NETLINK_PKTINFO:
1646 nlk->flags |= NETLINK_F_RECV_PKTINFO;
1648 nlk->flags &= ~NETLINK_F_RECV_PKTINFO;
1651 case NETLINK_ADD_MEMBERSHIP:
1652 case NETLINK_DROP_MEMBERSHIP: {
1653 if (!netlink_allowed(sock, NL_CFG_F_NONROOT_RECV))
1655 err = netlink_realloc_groups(sk);
1658 if (!val || val - 1 >= nlk->ngroups)
1660 if (optname == NETLINK_ADD_MEMBERSHIP && nlk->netlink_bind) {
1661 err = nlk->netlink_bind(sock_net(sk), val);
1665 netlink_table_grab();
1666 netlink_update_socket_mc(nlk, val,
1667 optname == NETLINK_ADD_MEMBERSHIP);
1668 netlink_table_ungrab();
1669 if (optname == NETLINK_DROP_MEMBERSHIP && nlk->netlink_unbind)
1670 nlk->netlink_unbind(sock_net(sk), val);
1675 case NETLINK_BROADCAST_ERROR:
1677 nlk->flags |= NETLINK_F_BROADCAST_SEND_ERROR;
1679 nlk->flags &= ~NETLINK_F_BROADCAST_SEND_ERROR;
1682 case NETLINK_NO_ENOBUFS:
1684 nlk->flags |= NETLINK_F_RECV_NO_ENOBUFS;
1685 clear_bit(NETLINK_S_CONGESTED, &nlk->state);
1686 wake_up_interruptible(&nlk->wait);
1688 nlk->flags &= ~NETLINK_F_RECV_NO_ENOBUFS;
1692 case NETLINK_LISTEN_ALL_NSID:
1693 if (!ns_capable(sock_net(sk)->user_ns, CAP_NET_BROADCAST))
1697 nlk->flags |= NETLINK_F_LISTEN_ALL_NSID;
1699 nlk->flags &= ~NETLINK_F_LISTEN_ALL_NSID;
1702 case NETLINK_CAP_ACK:
1704 nlk->flags |= NETLINK_F_CAP_ACK;
1706 nlk->flags &= ~NETLINK_F_CAP_ACK;
1709 case NETLINK_EXT_ACK:
1711 nlk->flags |= NETLINK_F_EXT_ACK;
1713 nlk->flags &= ~NETLINK_F_EXT_ACK;
1716 case NETLINK_GET_STRICT_CHK:
1718 nlk->flags |= NETLINK_F_STRICT_CHK;
1720 nlk->flags &= ~NETLINK_F_STRICT_CHK;
1729 static int netlink_getsockopt(struct socket *sock, int level, int optname,
1730 char __user *optval, int __user *optlen)
1732 struct sock *sk = sock->sk;
1733 struct netlink_sock *nlk = nlk_sk(sk);
1736 if (level != SOL_NETLINK)
1737 return -ENOPROTOOPT;
1739 if (get_user(len, optlen))
1745 case NETLINK_PKTINFO:
1746 if (len < sizeof(int))
1749 val = nlk->flags & NETLINK_F_RECV_PKTINFO ? 1 : 0;
1750 if (put_user(len, optlen) ||
1751 put_user(val, optval))
1755 case NETLINK_BROADCAST_ERROR:
1756 if (len < sizeof(int))
1759 val = nlk->flags & NETLINK_F_BROADCAST_SEND_ERROR ? 1 : 0;
1760 if (put_user(len, optlen) ||
1761 put_user(val, optval))
1765 case NETLINK_NO_ENOBUFS:
1766 if (len < sizeof(int))
1769 val = nlk->flags & NETLINK_F_RECV_NO_ENOBUFS ? 1 : 0;
1770 if (put_user(len, optlen) ||
1771 put_user(val, optval))
1775 case NETLINK_LIST_MEMBERSHIPS: {
1776 int pos, idx, shift;
1779 netlink_lock_table();
1780 for (pos = 0; pos * 8 < nlk->ngroups; pos += sizeof(u32)) {
1781 if (len - pos < sizeof(u32))
1784 idx = pos / sizeof(unsigned long);
1785 shift = (pos % sizeof(unsigned long)) * 8;
1786 if (put_user((u32)(nlk->groups[idx] >> shift),
1787 (u32 __user *)(optval + pos))) {
1792 if (put_user(ALIGN(nlk->ngroups / 8, sizeof(u32)), optlen))
1794 netlink_unlock_table();
1797 case NETLINK_CAP_ACK:
1798 if (len < sizeof(int))
1801 val = nlk->flags & NETLINK_F_CAP_ACK ? 1 : 0;
1802 if (put_user(len, optlen) ||
1803 put_user(val, optval))
1807 case NETLINK_EXT_ACK:
1808 if (len < sizeof(int))
1811 val = nlk->flags & NETLINK_F_EXT_ACK ? 1 : 0;
1812 if (put_user(len, optlen) || put_user(val, optval))
1816 case NETLINK_GET_STRICT_CHK:
1817 if (len < sizeof(int))
1820 val = nlk->flags & NETLINK_F_STRICT_CHK ? 1 : 0;
1821 if (put_user(len, optlen) || put_user(val, optval))
1831 static void netlink_cmsg_recv_pktinfo(struct msghdr *msg, struct sk_buff *skb)
1833 struct nl_pktinfo info;
1835 info.group = NETLINK_CB(skb).dst_group;
1836 put_cmsg(msg, SOL_NETLINK, NETLINK_PKTINFO, sizeof(info), &info);
1839 static void netlink_cmsg_listen_all_nsid(struct sock *sk, struct msghdr *msg,
1840 struct sk_buff *skb)
1842 if (!NETLINK_CB(skb).nsid_is_set)
1845 put_cmsg(msg, SOL_NETLINK, NETLINK_LISTEN_ALL_NSID, sizeof(int),
1846 &NETLINK_CB(skb).nsid);
1849 static int netlink_sendmsg(struct socket *sock, struct msghdr *msg, size_t len)
1851 struct sock *sk = sock->sk;
1852 struct netlink_sock *nlk = nlk_sk(sk);
1853 DECLARE_SOCKADDR(struct sockaddr_nl *, addr, msg->msg_name);
1856 struct sk_buff *skb;
1858 struct scm_cookie scm;
1859 u32 netlink_skb_flags = 0;
1861 if (msg->msg_flags & MSG_OOB)
1865 pr_warn_once("Zero length message leads to an empty skb\n");
1869 err = scm_send(sock, msg, &scm, true);
1873 if (msg->msg_namelen) {
1875 if (msg->msg_namelen < sizeof(struct sockaddr_nl))
1877 if (addr->nl_family != AF_NETLINK)
1879 dst_portid = addr->nl_pid;
1880 dst_group = ffs(addr->nl_groups);
1882 if ((dst_group || dst_portid) &&
1883 !netlink_allowed(sock, NL_CFG_F_NONROOT_SEND))
1885 netlink_skb_flags |= NETLINK_SKB_DST;
1887 /* Paired with WRITE_ONCE() in netlink_connect() */
1888 dst_portid = READ_ONCE(nlk->dst_portid);
1889 dst_group = READ_ONCE(nlk->dst_group);
1892 /* Paired with WRITE_ONCE() in netlink_insert() */
1893 if (!READ_ONCE(nlk->bound)) {
1894 err = netlink_autobind(sock);
1898 /* Ensure nlk is hashed and visible. */
1903 if (len > sk->sk_sndbuf - 32)
1906 skb = netlink_alloc_large_skb(len, dst_group);
1910 NETLINK_CB(skb).portid = nlk->portid;
1911 NETLINK_CB(skb).dst_group = dst_group;
1912 NETLINK_CB(skb).creds = scm.creds;
1913 NETLINK_CB(skb).flags = netlink_skb_flags;
1916 if (memcpy_from_msg(skb_put(skb, len), msg, len)) {
1921 err = security_netlink_send(sk, skb);
1928 refcount_inc(&skb->users);
1929 netlink_broadcast(sk, skb, dst_portid, dst_group, GFP_KERNEL);
1931 err = netlink_unicast(sk, skb, dst_portid, msg->msg_flags & MSG_DONTWAIT);
1938 static int netlink_recvmsg(struct socket *sock, struct msghdr *msg, size_t len,
1941 struct scm_cookie scm;
1942 struct sock *sk = sock->sk;
1943 struct netlink_sock *nlk = nlk_sk(sk);
1944 size_t copied, max_recvmsg_len;
1945 struct sk_buff *skb, *data_skb;
1948 if (flags & MSG_OOB)
1953 skb = skb_recv_datagram(sk, flags, &err);
1959 #ifdef CONFIG_COMPAT_NETLINK_MESSAGES
1960 if (unlikely(skb_shinfo(skb)->frag_list)) {
1962 * If this skb has a frag_list, then here that means that we
1963 * will have to use the frag_list skb's data for compat tasks
1964 * and the regular skb's data for normal (non-compat) tasks.
1966 * If we need to send the compat skb, assign it to the
1967 * 'data_skb' variable so that it will be used below for data
1968 * copying. We keep 'skb' for everything else, including
1969 * freeing both later.
1971 if (flags & MSG_CMSG_COMPAT)
1972 data_skb = skb_shinfo(skb)->frag_list;
1976 /* Record the max length of recvmsg() calls for future allocations */
1977 max_recvmsg_len = max(READ_ONCE(nlk->max_recvmsg_len), len);
1978 max_recvmsg_len = min_t(size_t, max_recvmsg_len,
1979 SKB_WITH_OVERHEAD(32768));
1980 WRITE_ONCE(nlk->max_recvmsg_len, max_recvmsg_len);
1982 copied = data_skb->len;
1984 msg->msg_flags |= MSG_TRUNC;
1988 err = skb_copy_datagram_msg(data_skb, 0, msg, copied);
1990 if (msg->msg_name) {
1991 DECLARE_SOCKADDR(struct sockaddr_nl *, addr, msg->msg_name);
1992 addr->nl_family = AF_NETLINK;
1994 addr->nl_pid = NETLINK_CB(skb).portid;
1995 addr->nl_groups = netlink_group_mask(NETLINK_CB(skb).dst_group);
1996 msg->msg_namelen = sizeof(*addr);
1999 if (nlk->flags & NETLINK_F_RECV_PKTINFO)
2000 netlink_cmsg_recv_pktinfo(msg, skb);
2001 if (nlk->flags & NETLINK_F_LISTEN_ALL_NSID)
2002 netlink_cmsg_listen_all_nsid(sk, msg, skb);
2004 memset(&scm, 0, sizeof(scm));
2005 scm.creds = *NETLINK_CREDS(skb);
2006 if (flags & MSG_TRUNC)
2007 copied = data_skb->len;
2009 skb_free_datagram(sk, skb);
2011 if (nlk->cb_running &&
2012 atomic_read(&sk->sk_rmem_alloc) <= sk->sk_rcvbuf / 2) {
2013 ret = netlink_dump(sk);
2016 sk_error_report(sk);
2020 scm_recv(sock, msg, &scm, flags);
2022 netlink_rcv_wake(sk);
2023 return err ? : copied;
2026 static void netlink_data_ready(struct sock *sk)
2032 * We export these functions to other modules. They provide a
2033 * complete set of kernel non-blocking support for message
2038 __netlink_kernel_create(struct net *net, int unit, struct module *module,
2039 struct netlink_kernel_cfg *cfg)
2041 struct socket *sock;
2043 struct netlink_sock *nlk;
2044 struct listeners *listeners = NULL;
2045 struct mutex *cb_mutex = cfg ? cfg->cb_mutex : NULL;
2046 unsigned int groups;
2050 if (unit < 0 || unit >= MAX_LINKS)
2053 if (sock_create_lite(PF_NETLINK, SOCK_DGRAM, unit, &sock))
2056 if (__netlink_create(net, sock, cb_mutex, unit, 1) < 0)
2057 goto out_sock_release_nosk;
2061 if (!cfg || cfg->groups < 32)
2064 groups = cfg->groups;
2066 listeners = kzalloc(sizeof(*listeners) + NLGRPSZ(groups), GFP_KERNEL);
2068 goto out_sock_release;
2070 sk->sk_data_ready = netlink_data_ready;
2071 if (cfg && cfg->input)
2072 nlk_sk(sk)->netlink_rcv = cfg->input;
2074 if (netlink_insert(sk, 0))
2075 goto out_sock_release;
2078 nlk->flags |= NETLINK_F_KERNEL_SOCKET;
2080 netlink_table_grab();
2081 if (!nl_table[unit].registered) {
2082 nl_table[unit].groups = groups;
2083 rcu_assign_pointer(nl_table[unit].listeners, listeners);
2084 nl_table[unit].cb_mutex = cb_mutex;
2085 nl_table[unit].module = module;
2087 nl_table[unit].bind = cfg->bind;
2088 nl_table[unit].unbind = cfg->unbind;
2089 nl_table[unit].flags = cfg->flags;
2091 nl_table[unit].compare = cfg->compare;
2093 nl_table[unit].registered = 1;
2096 nl_table[unit].registered++;
2098 netlink_table_ungrab();
2103 netlink_kernel_release(sk);
2106 out_sock_release_nosk:
2110 EXPORT_SYMBOL(__netlink_kernel_create);
2113 netlink_kernel_release(struct sock *sk)
2115 if (sk == NULL || sk->sk_socket == NULL)
2118 sock_release(sk->sk_socket);
2120 EXPORT_SYMBOL(netlink_kernel_release);
2122 int __netlink_change_ngroups(struct sock *sk, unsigned int groups)
2124 struct listeners *new, *old;
2125 struct netlink_table *tbl = &nl_table[sk->sk_protocol];
2130 if (NLGRPSZ(tbl->groups) < NLGRPSZ(groups)) {
2131 new = kzalloc(sizeof(*new) + NLGRPSZ(groups), GFP_ATOMIC);
2134 old = nl_deref_protected(tbl->listeners);
2135 memcpy(new->masks, old->masks, NLGRPSZ(tbl->groups));
2136 rcu_assign_pointer(tbl->listeners, new);
2138 kfree_rcu(old, rcu);
2140 tbl->groups = groups;
2146 * netlink_change_ngroups - change number of multicast groups
2148 * This changes the number of multicast groups that are available
2149 * on a certain netlink family. Note that it is not possible to
2150 * change the number of groups to below 32. Also note that it does
2151 * not implicitly call netlink_clear_multicast_users() when the
2152 * number of groups is reduced.
2154 * @sk: The kernel netlink socket, as returned by netlink_kernel_create().
2155 * @groups: The new number of groups.
2157 int netlink_change_ngroups(struct sock *sk, unsigned int groups)
2161 netlink_table_grab();
2162 err = __netlink_change_ngroups(sk, groups);
2163 netlink_table_ungrab();
2168 void __netlink_clear_multicast_users(struct sock *ksk, unsigned int group)
2171 struct netlink_table *tbl = &nl_table[ksk->sk_protocol];
2173 sk_for_each_bound(sk, &tbl->mc_list)
2174 netlink_update_socket_mc(nlk_sk(sk), group, 0);
2178 __nlmsg_put(struct sk_buff *skb, u32 portid, u32 seq, int type, int len, int flags)
2180 struct nlmsghdr *nlh;
2181 int size = nlmsg_msg_size(len);
2183 nlh = skb_put(skb, NLMSG_ALIGN(size));
2184 nlh->nlmsg_type = type;
2185 nlh->nlmsg_len = size;
2186 nlh->nlmsg_flags = flags;
2187 nlh->nlmsg_pid = portid;
2188 nlh->nlmsg_seq = seq;
2189 if (!__builtin_constant_p(size) || NLMSG_ALIGN(size) - size != 0)
2190 memset(nlmsg_data(nlh) + len, 0, NLMSG_ALIGN(size) - size);
2193 EXPORT_SYMBOL(__nlmsg_put);
2196 * It looks a bit ugly.
2197 * It would be better to create kernel thread.
2200 static int netlink_dump_done(struct netlink_sock *nlk, struct sk_buff *skb,
2201 struct netlink_callback *cb,
2202 struct netlink_ext_ack *extack)
2204 struct nlmsghdr *nlh;
2206 nlh = nlmsg_put_answer(skb, cb, NLMSG_DONE, sizeof(nlk->dump_done_errno),
2207 NLM_F_MULTI | cb->answer_flags);
2211 nl_dump_check_consistent(cb, nlh);
2212 memcpy(nlmsg_data(nlh), &nlk->dump_done_errno, sizeof(nlk->dump_done_errno));
2214 if (extack->_msg && nlk->flags & NETLINK_F_EXT_ACK) {
2215 nlh->nlmsg_flags |= NLM_F_ACK_TLVS;
2216 if (!nla_put_string(skb, NLMSGERR_ATTR_MSG, extack->_msg))
2217 nlmsg_end(skb, nlh);
2223 static int netlink_dump(struct sock *sk)
2225 struct netlink_sock *nlk = nlk_sk(sk);
2226 struct netlink_ext_ack extack = {};
2227 struct netlink_callback *cb;
2228 struct sk_buff *skb = NULL;
2229 size_t max_recvmsg_len;
2230 struct module *module;
2235 mutex_lock(nlk->cb_mutex);
2236 if (!nlk->cb_running) {
2241 if (atomic_read(&sk->sk_rmem_alloc) >= sk->sk_rcvbuf)
2244 /* NLMSG_GOODSIZE is small to avoid high order allocations being
2245 * required, but it makes sense to _attempt_ a 16K bytes allocation
2246 * to reduce number of system calls on dump operations, if user
2247 * ever provided a big enough buffer.
2250 alloc_min_size = max_t(int, cb->min_dump_alloc, NLMSG_GOODSIZE);
2252 max_recvmsg_len = READ_ONCE(nlk->max_recvmsg_len);
2253 if (alloc_min_size < max_recvmsg_len) {
2254 alloc_size = max_recvmsg_len;
2255 skb = alloc_skb(alloc_size,
2256 (GFP_KERNEL & ~__GFP_DIRECT_RECLAIM) |
2257 __GFP_NOWARN | __GFP_NORETRY);
2260 alloc_size = alloc_min_size;
2261 skb = alloc_skb(alloc_size, GFP_KERNEL);
2266 /* Trim skb to allocated size. User is expected to provide buffer as
2267 * large as max(min_dump_alloc, 16KiB (mac_recvmsg_len capped at
2268 * netlink_recvmsg())). dump will pack as many smaller messages as
2269 * could fit within the allocated skb. skb is typically allocated
2270 * with larger space than required (could be as much as near 2x the
2271 * requested size with align to next power of 2 approach). Allowing
2272 * dump to use the excess space makes it difficult for a user to have a
2273 * reasonable static buffer based on the expected largest dump of a
2274 * single netdev. The outcome is MSG_TRUNC error.
2276 skb_reserve(skb, skb_tailroom(skb) - alloc_size);
2278 /* Make sure malicious BPF programs can not read unitialized memory
2279 * from skb->head -> skb->data
2281 skb_reset_network_header(skb);
2282 skb_reset_mac_header(skb);
2284 netlink_skb_set_owner_r(skb, sk);
2286 if (nlk->dump_done_errno > 0) {
2287 cb->extack = &extack;
2288 nlk->dump_done_errno = cb->dump(skb, cb);
2292 if (nlk->dump_done_errno > 0 ||
2293 skb_tailroom(skb) < nlmsg_total_size(sizeof(nlk->dump_done_errno))) {
2294 mutex_unlock(nlk->cb_mutex);
2296 if (sk_filter(sk, skb))
2299 __netlink_sendskb(sk, skb);
2303 if (netlink_dump_done(nlk, skb, cb, &extack))
2306 #ifdef CONFIG_COMPAT_NETLINK_MESSAGES
2307 /* frag_list skb's data is used for compat tasks
2308 * and the regular skb's data for normal (non-compat) tasks.
2309 * See netlink_recvmsg().
2311 if (unlikely(skb_shinfo(skb)->frag_list)) {
2312 if (netlink_dump_done(nlk, skb_shinfo(skb)->frag_list, cb, &extack))
2317 if (sk_filter(sk, skb))
2320 __netlink_sendskb(sk, skb);
2325 nlk->cb_running = false;
2326 module = cb->module;
2328 mutex_unlock(nlk->cb_mutex);
2334 mutex_unlock(nlk->cb_mutex);
2339 int __netlink_dump_start(struct sock *ssk, struct sk_buff *skb,
2340 const struct nlmsghdr *nlh,
2341 struct netlink_dump_control *control)
2343 struct netlink_sock *nlk, *nlk2;
2344 struct netlink_callback *cb;
2348 refcount_inc(&skb->users);
2350 sk = netlink_lookup(sock_net(ssk), ssk->sk_protocol, NETLINK_CB(skb).portid);
2352 ret = -ECONNREFUSED;
2357 mutex_lock(nlk->cb_mutex);
2358 /* A dump is in progress... */
2359 if (nlk->cb_running) {
2363 /* add reference of module which cb->dump belongs to */
2364 if (!try_module_get(control->module)) {
2365 ret = -EPROTONOSUPPORT;
2370 memset(cb, 0, sizeof(*cb));
2371 cb->dump = control->dump;
2372 cb->done = control->done;
2374 cb->data = control->data;
2375 cb->module = control->module;
2376 cb->min_dump_alloc = control->min_dump_alloc;
2379 nlk2 = nlk_sk(NETLINK_CB(skb).sk);
2380 cb->strict_check = !!(nlk2->flags & NETLINK_F_STRICT_CHK);
2382 if (control->start) {
2383 ret = control->start(cb);
2388 nlk->cb_running = true;
2389 nlk->dump_done_errno = INT_MAX;
2391 mutex_unlock(nlk->cb_mutex);
2393 ret = netlink_dump(sk);
2400 /* We successfully started a dump, by returning -EINTR we
2401 * signal not to send ACK even if it was requested.
2406 module_put(control->module);
2409 mutex_unlock(nlk->cb_mutex);
2414 EXPORT_SYMBOL(__netlink_dump_start);
2417 netlink_ack_tlv_len(struct netlink_sock *nlk, int err,
2418 const struct netlink_ext_ack *extack)
2422 if (!extack || !(nlk->flags & NETLINK_F_EXT_ACK))
2427 tlvlen += nla_total_size(strlen(extack->_msg) + 1);
2428 if (extack->cookie_len)
2429 tlvlen += nla_total_size(extack->cookie_len);
2431 /* Following attributes are only reported as error (not warning) */
2435 if (extack->bad_attr)
2436 tlvlen += nla_total_size(sizeof(u32));
2438 tlvlen += netlink_policy_dump_attr_size_estimate(extack->policy);
2439 if (extack->miss_type)
2440 tlvlen += nla_total_size(sizeof(u32));
2441 if (extack->miss_nest)
2442 tlvlen += nla_total_size(sizeof(u32));
2448 netlink_ack_tlv_fill(struct sk_buff *in_skb, struct sk_buff *skb,
2449 struct nlmsghdr *nlh, int err,
2450 const struct netlink_ext_ack *extack)
2453 WARN_ON(nla_put_string(skb, NLMSGERR_ATTR_MSG, extack->_msg));
2454 if (extack->cookie_len)
2455 WARN_ON(nla_put(skb, NLMSGERR_ATTR_COOKIE,
2456 extack->cookie_len, extack->cookie));
2461 if (extack->bad_attr &&
2462 !WARN_ON((u8 *)extack->bad_attr < in_skb->data ||
2463 (u8 *)extack->bad_attr >= in_skb->data + in_skb->len))
2464 WARN_ON(nla_put_u32(skb, NLMSGERR_ATTR_OFFS,
2465 (u8 *)extack->bad_attr - (u8 *)nlh));
2467 netlink_policy_dump_write_attr(skb, extack->policy,
2468 NLMSGERR_ATTR_POLICY);
2469 if (extack->miss_type)
2470 WARN_ON(nla_put_u32(skb, NLMSGERR_ATTR_MISS_TYPE,
2471 extack->miss_type));
2472 if (extack->miss_nest &&
2473 !WARN_ON((u8 *)extack->miss_nest < in_skb->data ||
2474 (u8 *)extack->miss_nest > in_skb->data + in_skb->len))
2475 WARN_ON(nla_put_u32(skb, NLMSGERR_ATTR_MISS_NEST,
2476 (u8 *)extack->miss_nest - (u8 *)nlh));
2479 void netlink_ack(struct sk_buff *in_skb, struct nlmsghdr *nlh, int err,
2480 const struct netlink_ext_ack *extack)
2482 struct sk_buff *skb;
2483 struct nlmsghdr *rep;
2484 struct nlmsgerr *errmsg;
2485 size_t payload = sizeof(*errmsg);
2486 struct netlink_sock *nlk = nlk_sk(NETLINK_CB(in_skb).sk);
2487 unsigned int flags = 0;
2490 /* Error messages get the original request appened, unless the user
2491 * requests to cap the error message, and get extra error data if
2494 if (err && !(nlk->flags & NETLINK_F_CAP_ACK))
2495 payload += nlmsg_len(nlh);
2497 flags |= NLM_F_CAPPED;
2499 tlvlen = netlink_ack_tlv_len(nlk, err, extack);
2501 flags |= NLM_F_ACK_TLVS;
2503 skb = nlmsg_new(payload + tlvlen, GFP_KERNEL);
2505 NETLINK_CB(in_skb).sk->sk_err = ENOBUFS;
2506 sk_error_report(NETLINK_CB(in_skb).sk);
2510 rep = nlmsg_put(skb, NETLINK_CB(in_skb).portid, nlh->nlmsg_seq,
2511 NLMSG_ERROR, payload, flags);
2512 errmsg = nlmsg_data(rep);
2513 errmsg->error = err;
2514 unsafe_memcpy(&errmsg->msg, nlh, payload > sizeof(*errmsg)
2515 ? nlh->nlmsg_len : sizeof(*nlh),
2516 /* Bounds checked by the skb layer. */);
2519 netlink_ack_tlv_fill(in_skb, skb, nlh, err, extack);
2521 nlmsg_end(skb, rep);
2523 nlmsg_unicast(in_skb->sk, skb, NETLINK_CB(in_skb).portid);
2525 EXPORT_SYMBOL(netlink_ack);
2527 int netlink_rcv_skb(struct sk_buff *skb, int (*cb)(struct sk_buff *,
2529 struct netlink_ext_ack *))
2531 struct netlink_ext_ack extack;
2532 struct nlmsghdr *nlh;
2535 while (skb->len >= nlmsg_total_size(0)) {
2538 memset(&extack, 0, sizeof(extack));
2539 nlh = nlmsg_hdr(skb);
2542 if (nlh->nlmsg_len < NLMSG_HDRLEN || skb->len < nlh->nlmsg_len)
2545 /* Only requests are handled by the kernel */
2546 if (!(nlh->nlmsg_flags & NLM_F_REQUEST))
2549 /* Skip control messages */
2550 if (nlh->nlmsg_type < NLMSG_MIN_TYPE)
2553 err = cb(skb, nlh, &extack);
2558 if (nlh->nlmsg_flags & NLM_F_ACK || err)
2559 netlink_ack(skb, nlh, err, &extack);
2562 msglen = NLMSG_ALIGN(nlh->nlmsg_len);
2563 if (msglen > skb->len)
2565 skb_pull(skb, msglen);
2570 EXPORT_SYMBOL(netlink_rcv_skb);
2573 * nlmsg_notify - send a notification netlink message
2574 * @sk: netlink socket to use
2575 * @skb: notification message
2576 * @portid: destination netlink portid for reports or 0
2577 * @group: destination multicast group or 0
2578 * @report: 1 to report back, 0 to disable
2579 * @flags: allocation flags
2581 int nlmsg_notify(struct sock *sk, struct sk_buff *skb, u32 portid,
2582 unsigned int group, int report, gfp_t flags)
2587 int exclude_portid = 0;
2590 refcount_inc(&skb->users);
2591 exclude_portid = portid;
2594 /* errors reported via destination sk->sk_err, but propagate
2595 * delivery errors if NETLINK_BROADCAST_ERROR flag is set */
2596 err = nlmsg_multicast(sk, skb, exclude_portid, group, flags);
2604 err2 = nlmsg_unicast(sk, skb, portid);
2611 EXPORT_SYMBOL(nlmsg_notify);
2613 #ifdef CONFIG_PROC_FS
2614 struct nl_seq_iter {
2615 struct seq_net_private p;
2616 struct rhashtable_iter hti;
2620 static void netlink_walk_start(struct nl_seq_iter *iter)
2622 rhashtable_walk_enter(&nl_table[iter->link].hash, &iter->hti);
2623 rhashtable_walk_start(&iter->hti);
2626 static void netlink_walk_stop(struct nl_seq_iter *iter)
2628 rhashtable_walk_stop(&iter->hti);
2629 rhashtable_walk_exit(&iter->hti);
2632 static void *__netlink_seq_next(struct seq_file *seq)
2634 struct nl_seq_iter *iter = seq->private;
2635 struct netlink_sock *nlk;
2639 nlk = rhashtable_walk_next(&iter->hti);
2642 if (PTR_ERR(nlk) == -EAGAIN)
2651 netlink_walk_stop(iter);
2652 if (++iter->link >= MAX_LINKS)
2655 netlink_walk_start(iter);
2657 } while (sock_net(&nlk->sk) != seq_file_net(seq));
2662 static void *netlink_seq_start(struct seq_file *seq, loff_t *posp)
2665 struct nl_seq_iter *iter = seq->private;
2666 void *obj = SEQ_START_TOKEN;
2671 netlink_walk_start(iter);
2673 for (pos = *posp; pos && obj && !IS_ERR(obj); pos--)
2674 obj = __netlink_seq_next(seq);
2679 static void *netlink_seq_next(struct seq_file *seq, void *v, loff_t *pos)
2682 return __netlink_seq_next(seq);
2685 static void netlink_native_seq_stop(struct seq_file *seq, void *v)
2687 struct nl_seq_iter *iter = seq->private;
2689 if (iter->link >= MAX_LINKS)
2692 netlink_walk_stop(iter);
2696 static int netlink_native_seq_show(struct seq_file *seq, void *v)
2698 if (v == SEQ_START_TOKEN) {
2700 "sk Eth Pid Groups "
2701 "Rmem Wmem Dump Locks Drops Inode\n");
2704 struct netlink_sock *nlk = nlk_sk(s);
2706 seq_printf(seq, "%pK %-3d %-10u %08x %-8d %-8d %-5d %-8d %-8u %-8lu\n",
2710 nlk->groups ? (u32)nlk->groups[0] : 0,
2711 sk_rmem_alloc_get(s),
2712 sk_wmem_alloc_get(s),
2714 refcount_read(&s->sk_refcnt),
2715 atomic_read(&s->sk_drops),
2723 #ifdef CONFIG_BPF_SYSCALL
2724 struct bpf_iter__netlink {
2725 __bpf_md_ptr(struct bpf_iter_meta *, meta);
2726 __bpf_md_ptr(struct netlink_sock *, sk);
2729 DEFINE_BPF_ITER_FUNC(netlink, struct bpf_iter_meta *meta, struct netlink_sock *sk)
2731 static int netlink_prog_seq_show(struct bpf_prog *prog,
2732 struct bpf_iter_meta *meta,
2735 struct bpf_iter__netlink ctx;
2737 meta->seq_num--; /* skip SEQ_START_TOKEN */
2739 ctx.sk = nlk_sk((struct sock *)v);
2740 return bpf_iter_run_prog(prog, &ctx);
2743 static int netlink_seq_show(struct seq_file *seq, void *v)
2745 struct bpf_iter_meta meta;
2746 struct bpf_prog *prog;
2749 prog = bpf_iter_get_info(&meta, false);
2751 return netlink_native_seq_show(seq, v);
2753 if (v != SEQ_START_TOKEN)
2754 return netlink_prog_seq_show(prog, &meta, v);
2759 static void netlink_seq_stop(struct seq_file *seq, void *v)
2761 struct bpf_iter_meta meta;
2762 struct bpf_prog *prog;
2766 prog = bpf_iter_get_info(&meta, true);
2768 (void)netlink_prog_seq_show(prog, &meta, v);
2771 netlink_native_seq_stop(seq, v);
2774 static int netlink_seq_show(struct seq_file *seq, void *v)
2776 return netlink_native_seq_show(seq, v);
2779 static void netlink_seq_stop(struct seq_file *seq, void *v)
2781 netlink_native_seq_stop(seq, v);
2785 static const struct seq_operations netlink_seq_ops = {
2786 .start = netlink_seq_start,
2787 .next = netlink_seq_next,
2788 .stop = netlink_seq_stop,
2789 .show = netlink_seq_show,
2793 int netlink_register_notifier(struct notifier_block *nb)
2795 return blocking_notifier_chain_register(&netlink_chain, nb);
2797 EXPORT_SYMBOL(netlink_register_notifier);
2799 int netlink_unregister_notifier(struct notifier_block *nb)
2801 return blocking_notifier_chain_unregister(&netlink_chain, nb);
2803 EXPORT_SYMBOL(netlink_unregister_notifier);
2805 static const struct proto_ops netlink_ops = {
2806 .family = PF_NETLINK,
2807 .owner = THIS_MODULE,
2808 .release = netlink_release,
2809 .bind = netlink_bind,
2810 .connect = netlink_connect,
2811 .socketpair = sock_no_socketpair,
2812 .accept = sock_no_accept,
2813 .getname = netlink_getname,
2814 .poll = datagram_poll,
2815 .ioctl = netlink_ioctl,
2816 .listen = sock_no_listen,
2817 .shutdown = sock_no_shutdown,
2818 .setsockopt = netlink_setsockopt,
2819 .getsockopt = netlink_getsockopt,
2820 .sendmsg = netlink_sendmsg,
2821 .recvmsg = netlink_recvmsg,
2822 .mmap = sock_no_mmap,
2823 .sendpage = sock_no_sendpage,
2826 static const struct net_proto_family netlink_family_ops = {
2827 .family = PF_NETLINK,
2828 .create = netlink_create,
2829 .owner = THIS_MODULE, /* for consistency 8) */
2832 static int __net_init netlink_net_init(struct net *net)
2834 #ifdef CONFIG_PROC_FS
2835 if (!proc_create_net("netlink", 0, net->proc_net, &netlink_seq_ops,
2836 sizeof(struct nl_seq_iter)))
2842 static void __net_exit netlink_net_exit(struct net *net)
2844 #ifdef CONFIG_PROC_FS
2845 remove_proc_entry("netlink", net->proc_net);
2849 static void __init netlink_add_usersock_entry(void)
2851 struct listeners *listeners;
2854 listeners = kzalloc(sizeof(*listeners) + NLGRPSZ(groups), GFP_KERNEL);
2856 panic("netlink_add_usersock_entry: Cannot allocate listeners\n");
2858 netlink_table_grab();
2860 nl_table[NETLINK_USERSOCK].groups = groups;
2861 rcu_assign_pointer(nl_table[NETLINK_USERSOCK].listeners, listeners);
2862 nl_table[NETLINK_USERSOCK].module = THIS_MODULE;
2863 nl_table[NETLINK_USERSOCK].registered = 1;
2864 nl_table[NETLINK_USERSOCK].flags = NL_CFG_F_NONROOT_SEND;
2866 netlink_table_ungrab();
2869 static struct pernet_operations __net_initdata netlink_net_ops = {
2870 .init = netlink_net_init,
2871 .exit = netlink_net_exit,
2874 static inline u32 netlink_hash(const void *data, u32 len, u32 seed)
2876 const struct netlink_sock *nlk = data;
2877 struct netlink_compare_arg arg;
2879 netlink_compare_arg_init(&arg, sock_net(&nlk->sk), nlk->portid);
2880 return jhash2((u32 *)&arg, netlink_compare_arg_len / sizeof(u32), seed);
2883 static const struct rhashtable_params netlink_rhashtable_params = {
2884 .head_offset = offsetof(struct netlink_sock, node),
2885 .key_len = netlink_compare_arg_len,
2886 .obj_hashfn = netlink_hash,
2887 .obj_cmpfn = netlink_compare,
2888 .automatic_shrinking = true,
2891 #if defined(CONFIG_BPF_SYSCALL) && defined(CONFIG_PROC_FS)
2892 BTF_ID_LIST(btf_netlink_sock_id)
2893 BTF_ID(struct, netlink_sock)
2895 static const struct bpf_iter_seq_info netlink_seq_info = {
2896 .seq_ops = &netlink_seq_ops,
2897 .init_seq_private = bpf_iter_init_seq_net,
2898 .fini_seq_private = bpf_iter_fini_seq_net,
2899 .seq_priv_size = sizeof(struct nl_seq_iter),
2902 static struct bpf_iter_reg netlink_reg_info = {
2903 .target = "netlink",
2904 .ctx_arg_info_size = 1,
2906 { offsetof(struct bpf_iter__netlink, sk),
2907 PTR_TO_BTF_ID_OR_NULL },
2909 .seq_info = &netlink_seq_info,
2912 static int __init bpf_iter_register(void)
2914 netlink_reg_info.ctx_arg_info[0].btf_id = *btf_netlink_sock_id;
2915 return bpf_iter_reg_target(&netlink_reg_info);
2919 static int __init netlink_proto_init(void)
2922 int err = proto_register(&netlink_proto, 0);
2927 #if defined(CONFIG_BPF_SYSCALL) && defined(CONFIG_PROC_FS)
2928 err = bpf_iter_register();
2933 BUILD_BUG_ON(sizeof(struct netlink_skb_parms) > sizeof_field(struct sk_buff, cb));
2935 nl_table = kcalloc(MAX_LINKS, sizeof(*nl_table), GFP_KERNEL);
2939 for (i = 0; i < MAX_LINKS; i++) {
2940 if (rhashtable_init(&nl_table[i].hash,
2941 &netlink_rhashtable_params) < 0) {
2943 rhashtable_destroy(&nl_table[i].hash);
2949 netlink_add_usersock_entry();
2951 sock_register(&netlink_family_ops);
2952 register_pernet_subsys(&netlink_net_ops);
2953 register_pernet_subsys(&netlink_tap_net_ops);
2954 /* The netlink device handler may be needed early. */
2959 panic("netlink_init: Cannot allocate nl_table\n");
2962 core_initcall(netlink_proto_init);