1 // SPDX-License-Identifier: GPL-2.0-or-later
3 * NETLINK Kernel-user communication protocol.
9 * Tue Jun 26 14:36:48 MEST 2001 Herbert "herp" Rosmanith
10 * added netlink_proto_exit
12 * use nlk_sk, as sk->protinfo is on a diet 8)
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)
160 return group ? 1 << (group - 1) : 0;
163 static struct sk_buff *netlink_to_full_skb(const struct sk_buff *skb,
166 unsigned int len = skb_end_offset(skb);
169 new = alloc_skb(len, gfp_mask);
173 NETLINK_CB(new).portid = NETLINK_CB(skb).portid;
174 NETLINK_CB(new).dst_group = NETLINK_CB(skb).dst_group;
175 NETLINK_CB(new).creds = NETLINK_CB(skb).creds;
177 skb_put_data(new, skb->data, len);
181 static unsigned int netlink_tap_net_id;
183 struct netlink_tap_net {
184 struct list_head netlink_tap_all;
185 struct mutex netlink_tap_lock;
188 int netlink_add_tap(struct netlink_tap *nt)
190 struct net *net = dev_net(nt->dev);
191 struct netlink_tap_net *nn = net_generic(net, netlink_tap_net_id);
193 if (unlikely(nt->dev->type != ARPHRD_NETLINK))
196 mutex_lock(&nn->netlink_tap_lock);
197 list_add_rcu(&nt->list, &nn->netlink_tap_all);
198 mutex_unlock(&nn->netlink_tap_lock);
200 __module_get(nt->module);
204 EXPORT_SYMBOL_GPL(netlink_add_tap);
206 static int __netlink_remove_tap(struct netlink_tap *nt)
208 struct net *net = dev_net(nt->dev);
209 struct netlink_tap_net *nn = net_generic(net, netlink_tap_net_id);
211 struct netlink_tap *tmp;
213 mutex_lock(&nn->netlink_tap_lock);
215 list_for_each_entry(tmp, &nn->netlink_tap_all, list) {
217 list_del_rcu(&nt->list);
223 pr_warn("__netlink_remove_tap: %p not found\n", nt);
225 mutex_unlock(&nn->netlink_tap_lock);
228 module_put(nt->module);
230 return found ? 0 : -ENODEV;
233 int netlink_remove_tap(struct netlink_tap *nt)
237 ret = __netlink_remove_tap(nt);
242 EXPORT_SYMBOL_GPL(netlink_remove_tap);
244 static __net_init int netlink_tap_init_net(struct net *net)
246 struct netlink_tap_net *nn = net_generic(net, netlink_tap_net_id);
248 INIT_LIST_HEAD(&nn->netlink_tap_all);
249 mutex_init(&nn->netlink_tap_lock);
253 static struct pernet_operations netlink_tap_net_ops = {
254 .init = netlink_tap_init_net,
255 .id = &netlink_tap_net_id,
256 .size = sizeof(struct netlink_tap_net),
259 static bool netlink_filter_tap(const struct sk_buff *skb)
261 struct sock *sk = skb->sk;
263 /* We take the more conservative approach and
264 * whitelist socket protocols that may pass.
266 switch (sk->sk_protocol) {
268 case NETLINK_USERSOCK:
269 case NETLINK_SOCK_DIAG:
272 case NETLINK_FIB_LOOKUP:
273 case NETLINK_NETFILTER:
274 case NETLINK_GENERIC:
281 static int __netlink_deliver_tap_skb(struct sk_buff *skb,
282 struct net_device *dev)
284 struct sk_buff *nskb;
285 struct sock *sk = skb->sk;
288 if (!net_eq(dev_net(dev), sock_net(sk)))
293 if (is_vmalloc_addr(skb->head))
294 nskb = netlink_to_full_skb(skb, GFP_ATOMIC);
296 nskb = skb_clone(skb, GFP_ATOMIC);
299 nskb->protocol = htons((u16) sk->sk_protocol);
300 nskb->pkt_type = netlink_is_kernel(sk) ?
301 PACKET_KERNEL : PACKET_USER;
302 skb_reset_network_header(nskb);
303 ret = dev_queue_xmit(nskb);
304 if (unlikely(ret > 0))
305 ret = net_xmit_errno(ret);
312 static void __netlink_deliver_tap(struct sk_buff *skb, struct netlink_tap_net *nn)
315 struct netlink_tap *tmp;
317 if (!netlink_filter_tap(skb))
320 list_for_each_entry_rcu(tmp, &nn->netlink_tap_all, list) {
321 ret = __netlink_deliver_tap_skb(skb, tmp->dev);
327 static void netlink_deliver_tap(struct net *net, struct sk_buff *skb)
329 struct netlink_tap_net *nn = net_generic(net, netlink_tap_net_id);
333 if (unlikely(!list_empty(&nn->netlink_tap_all)))
334 __netlink_deliver_tap(skb, nn);
339 static void netlink_deliver_tap_kernel(struct sock *dst, struct sock *src,
342 if (!(netlink_is_kernel(dst) && netlink_is_kernel(src)))
343 netlink_deliver_tap(sock_net(dst), skb);
346 static void netlink_overrun(struct sock *sk)
348 struct netlink_sock *nlk = nlk_sk(sk);
350 if (!(nlk->flags & NETLINK_F_RECV_NO_ENOBUFS)) {
351 if (!test_and_set_bit(NETLINK_S_CONGESTED,
352 &nlk_sk(sk)->state)) {
353 sk->sk_err = ENOBUFS;
357 atomic_inc(&sk->sk_drops);
360 static void netlink_rcv_wake(struct sock *sk)
362 struct netlink_sock *nlk = nlk_sk(sk);
364 if (skb_queue_empty_lockless(&sk->sk_receive_queue))
365 clear_bit(NETLINK_S_CONGESTED, &nlk->state);
366 if (!test_bit(NETLINK_S_CONGESTED, &nlk->state))
367 wake_up_interruptible(&nlk->wait);
370 static void netlink_skb_destructor(struct sk_buff *skb)
372 if (is_vmalloc_addr(skb->head)) {
374 !atomic_dec_return(&(skb_shinfo(skb)->dataref)))
383 static void netlink_skb_set_owner_r(struct sk_buff *skb, struct sock *sk)
385 WARN_ON(skb->sk != NULL);
387 skb->destructor = netlink_skb_destructor;
388 atomic_add(skb->truesize, &sk->sk_rmem_alloc);
389 sk_mem_charge(sk, skb->truesize);
392 static void netlink_sock_destruct(struct sock *sk)
394 struct netlink_sock *nlk = nlk_sk(sk);
396 if (nlk->cb_running) {
398 nlk->cb.done(&nlk->cb);
399 module_put(nlk->cb.module);
400 kfree_skb(nlk->cb.skb);
403 skb_queue_purge(&sk->sk_receive_queue);
405 if (!sock_flag(sk, SOCK_DEAD)) {
406 printk(KERN_ERR "Freeing alive netlink socket %p\n", sk);
410 WARN_ON(atomic_read(&sk->sk_rmem_alloc));
411 WARN_ON(refcount_read(&sk->sk_wmem_alloc));
412 WARN_ON(nlk_sk(sk)->groups);
415 static void netlink_sock_destruct_work(struct work_struct *work)
417 struct netlink_sock *nlk = container_of(work, struct netlink_sock,
423 /* This lock without WQ_FLAG_EXCLUSIVE is good on UP and it is _very_ bad on
424 * SMP. Look, when several writers sleep and reader wakes them up, all but one
425 * immediately hit write lock and grab all the cpus. Exclusive sleep solves
426 * this, _but_ remember, it adds useless work on UP machines.
429 void netlink_table_grab(void)
430 __acquires(nl_table_lock)
434 write_lock_irq(&nl_table_lock);
436 if (atomic_read(&nl_table_users)) {
437 DECLARE_WAITQUEUE(wait, current);
439 add_wait_queue_exclusive(&nl_table_wait, &wait);
441 set_current_state(TASK_UNINTERRUPTIBLE);
442 if (atomic_read(&nl_table_users) == 0)
444 write_unlock_irq(&nl_table_lock);
446 write_lock_irq(&nl_table_lock);
449 __set_current_state(TASK_RUNNING);
450 remove_wait_queue(&nl_table_wait, &wait);
454 void netlink_table_ungrab(void)
455 __releases(nl_table_lock)
457 write_unlock_irq(&nl_table_lock);
458 wake_up(&nl_table_wait);
462 netlink_lock_table(void)
466 /* read_lock() synchronizes us to netlink_table_grab */
468 read_lock_irqsave(&nl_table_lock, flags);
469 atomic_inc(&nl_table_users);
470 read_unlock_irqrestore(&nl_table_lock, flags);
474 netlink_unlock_table(void)
476 if (atomic_dec_and_test(&nl_table_users))
477 wake_up(&nl_table_wait);
480 struct netlink_compare_arg
486 /* Doing sizeof directly may yield 4 extra bytes on 64-bit. */
487 #define netlink_compare_arg_len \
488 (offsetof(struct netlink_compare_arg, portid) + sizeof(u32))
490 static inline int netlink_compare(struct rhashtable_compare_arg *arg,
493 const struct netlink_compare_arg *x = arg->key;
494 const struct netlink_sock *nlk = ptr;
496 return nlk->portid != x->portid ||
497 !net_eq(sock_net(&nlk->sk), read_pnet(&x->pnet));
500 static void netlink_compare_arg_init(struct netlink_compare_arg *arg,
501 struct net *net, u32 portid)
503 memset(arg, 0, sizeof(*arg));
504 write_pnet(&arg->pnet, net);
505 arg->portid = portid;
508 static struct sock *__netlink_lookup(struct netlink_table *table, u32 portid,
511 struct netlink_compare_arg arg;
513 netlink_compare_arg_init(&arg, net, portid);
514 return rhashtable_lookup_fast(&table->hash, &arg,
515 netlink_rhashtable_params);
518 static int __netlink_insert(struct netlink_table *table, struct sock *sk)
520 struct netlink_compare_arg arg;
522 netlink_compare_arg_init(&arg, sock_net(sk), nlk_sk(sk)->portid);
523 return rhashtable_lookup_insert_key(&table->hash, &arg,
525 netlink_rhashtable_params);
528 static struct sock *netlink_lookup(struct net *net, int protocol, u32 portid)
530 struct netlink_table *table = &nl_table[protocol];
534 sk = __netlink_lookup(table, portid, net);
542 static const struct proto_ops netlink_ops;
545 netlink_update_listeners(struct sock *sk)
547 struct netlink_table *tbl = &nl_table[sk->sk_protocol];
550 struct listeners *listeners;
552 listeners = nl_deref_protected(tbl->listeners);
556 for (i = 0; i < NLGRPLONGS(tbl->groups); i++) {
558 sk_for_each_bound(sk, &tbl->mc_list) {
559 if (i < NLGRPLONGS(nlk_sk(sk)->ngroups))
560 mask |= nlk_sk(sk)->groups[i];
562 listeners->masks[i] = mask;
564 /* this function is only called with the netlink table "grabbed", which
565 * makes sure updates are visible before bind or setsockopt return. */
568 static int netlink_insert(struct sock *sk, u32 portid)
570 struct netlink_table *table = &nl_table[sk->sk_protocol];
575 err = nlk_sk(sk)->portid == portid ? 0 : -EBUSY;
576 if (nlk_sk(sk)->bound)
579 nlk_sk(sk)->portid = portid;
582 err = __netlink_insert(table, sk);
584 /* In case the hashtable backend returns with -EBUSY
585 * from here, it must not escape to the caller.
587 if (unlikely(err == -EBUSY))
595 /* We need to ensure that the socket is hashed and visible. */
597 /* Paired with lockless reads from netlink_bind(),
598 * netlink_connect() and netlink_sendmsg().
600 WRITE_ONCE(nlk_sk(sk)->bound, portid);
607 static void netlink_remove(struct sock *sk)
609 struct netlink_table *table;
611 table = &nl_table[sk->sk_protocol];
612 if (!rhashtable_remove_fast(&table->hash, &nlk_sk(sk)->node,
613 netlink_rhashtable_params)) {
614 WARN_ON(refcount_read(&sk->sk_refcnt) == 1);
618 netlink_table_grab();
619 if (nlk_sk(sk)->subscriptions) {
620 __sk_del_bind_node(sk);
621 netlink_update_listeners(sk);
623 if (sk->sk_protocol == NETLINK_GENERIC)
624 atomic_inc(&genl_sk_destructing_cnt);
625 netlink_table_ungrab();
628 static struct proto netlink_proto = {
630 .owner = THIS_MODULE,
631 .obj_size = sizeof(struct netlink_sock),
634 static int __netlink_create(struct net *net, struct socket *sock,
635 struct mutex *cb_mutex, int protocol,
639 struct netlink_sock *nlk;
641 sock->ops = &netlink_ops;
643 sk = sk_alloc(net, PF_NETLINK, GFP_KERNEL, &netlink_proto, kern);
647 sock_init_data(sock, sk);
651 nlk->cb_mutex = cb_mutex;
653 nlk->cb_mutex = &nlk->cb_def_mutex;
654 mutex_init(nlk->cb_mutex);
655 lockdep_set_class_and_name(nlk->cb_mutex,
656 nlk_cb_mutex_keys + protocol,
657 nlk_cb_mutex_key_strings[protocol]);
659 init_waitqueue_head(&nlk->wait);
661 sk->sk_destruct = netlink_sock_destruct;
662 sk->sk_protocol = protocol;
666 static int netlink_create(struct net *net, struct socket *sock, int protocol,
669 struct module *module = NULL;
670 struct mutex *cb_mutex;
671 struct netlink_sock *nlk;
672 int (*bind)(struct net *net, int group);
673 void (*unbind)(struct net *net, int group);
676 sock->state = SS_UNCONNECTED;
678 if (sock->type != SOCK_RAW && sock->type != SOCK_DGRAM)
679 return -ESOCKTNOSUPPORT;
681 if (protocol < 0 || protocol >= MAX_LINKS)
682 return -EPROTONOSUPPORT;
683 protocol = array_index_nospec(protocol, MAX_LINKS);
685 netlink_lock_table();
686 #ifdef CONFIG_MODULES
687 if (!nl_table[protocol].registered) {
688 netlink_unlock_table();
689 request_module("net-pf-%d-proto-%d", PF_NETLINK, protocol);
690 netlink_lock_table();
693 if (nl_table[protocol].registered &&
694 try_module_get(nl_table[protocol].module))
695 module = nl_table[protocol].module;
697 err = -EPROTONOSUPPORT;
698 cb_mutex = nl_table[protocol].cb_mutex;
699 bind = nl_table[protocol].bind;
700 unbind = nl_table[protocol].unbind;
701 netlink_unlock_table();
706 err = __netlink_create(net, sock, cb_mutex, protocol, kern);
711 sock_prot_inuse_add(net, &netlink_proto, 1);
714 nlk = nlk_sk(sock->sk);
715 nlk->module = module;
716 nlk->netlink_bind = bind;
717 nlk->netlink_unbind = unbind;
726 static void deferred_put_nlk_sk(struct rcu_head *head)
728 struct netlink_sock *nlk = container_of(head, struct netlink_sock, rcu);
729 struct sock *sk = &nlk->sk;
734 if (!refcount_dec_and_test(&sk->sk_refcnt))
737 if (nlk->cb_running && nlk->cb.done) {
738 INIT_WORK(&nlk->work, netlink_sock_destruct_work);
739 schedule_work(&nlk->work);
746 static int netlink_release(struct socket *sock)
748 struct sock *sk = sock->sk;
749 struct netlink_sock *nlk;
759 * OK. Socket is unlinked, any packets that arrive now
763 /* must not acquire netlink_table_lock in any way again before unbind
764 * and notifying genetlink is done as otherwise it might deadlock
766 if (nlk->netlink_unbind) {
769 for (i = 0; i < nlk->ngroups; i++)
770 if (test_bit(i, nlk->groups))
771 nlk->netlink_unbind(sock_net(sk), i + 1);
773 if (sk->sk_protocol == NETLINK_GENERIC &&
774 atomic_dec_return(&genl_sk_destructing_cnt) == 0)
775 wake_up(&genl_sk_destructing_waitq);
778 wake_up_interruptible_all(&nlk->wait);
780 skb_queue_purge(&sk->sk_write_queue);
782 if (nlk->portid && nlk->bound) {
783 struct netlink_notify n = {
785 .protocol = sk->sk_protocol,
786 .portid = nlk->portid,
788 blocking_notifier_call_chain(&netlink_chain,
789 NETLINK_URELEASE, &n);
792 module_put(nlk->module);
794 if (netlink_is_kernel(sk)) {
795 netlink_table_grab();
796 BUG_ON(nl_table[sk->sk_protocol].registered == 0);
797 if (--nl_table[sk->sk_protocol].registered == 0) {
798 struct listeners *old;
800 old = nl_deref_protected(nl_table[sk->sk_protocol].listeners);
801 RCU_INIT_POINTER(nl_table[sk->sk_protocol].listeners, NULL);
803 nl_table[sk->sk_protocol].module = NULL;
804 nl_table[sk->sk_protocol].bind = NULL;
805 nl_table[sk->sk_protocol].unbind = NULL;
806 nl_table[sk->sk_protocol].flags = 0;
807 nl_table[sk->sk_protocol].registered = 0;
809 netlink_table_ungrab();
813 sock_prot_inuse_add(sock_net(sk), &netlink_proto, -1);
815 call_rcu(&nlk->rcu, deferred_put_nlk_sk);
819 static int netlink_autobind(struct socket *sock)
821 struct sock *sk = sock->sk;
822 struct net *net = sock_net(sk);
823 struct netlink_table *table = &nl_table[sk->sk_protocol];
824 s32 portid = task_tgid_vnr(current);
832 ok = !__netlink_lookup(table, portid, net);
835 /* Bind collision, search negative portid values. */
837 /* rover will be in range [S32_MIN, -4097] */
838 rover = S32_MIN + prandom_u32_max(-4096 - S32_MIN);
839 else if (rover >= -4096)
845 err = netlink_insert(sk, portid);
846 if (err == -EADDRINUSE)
849 /* If 2 threads race to autobind, that is fine. */
857 * __netlink_ns_capable - General netlink message capability test
858 * @nsp: NETLINK_CB of the socket buffer holding a netlink command from userspace.
859 * @user_ns: The user namespace of the capability to use
860 * @cap: The capability to use
862 * Test to see if the opener of the socket we received the message
863 * from had when the netlink socket was created and the sender of the
864 * message has the capability @cap in the user namespace @user_ns.
866 bool __netlink_ns_capable(const struct netlink_skb_parms *nsp,
867 struct user_namespace *user_ns, int cap)
869 return ((nsp->flags & NETLINK_SKB_DST) ||
870 file_ns_capable(nsp->sk->sk_socket->file, user_ns, cap)) &&
871 ns_capable(user_ns, cap);
873 EXPORT_SYMBOL(__netlink_ns_capable);
876 * netlink_ns_capable - General netlink message capability test
877 * @skb: socket buffer holding a netlink command from userspace
878 * @user_ns: The user namespace of the capability to use
879 * @cap: The capability to use
881 * Test to see if the opener of the socket we received the message
882 * from had when the netlink socket was created and the sender of the
883 * message has the capability @cap in the user namespace @user_ns.
885 bool netlink_ns_capable(const struct sk_buff *skb,
886 struct user_namespace *user_ns, int cap)
888 return __netlink_ns_capable(&NETLINK_CB(skb), user_ns, cap);
890 EXPORT_SYMBOL(netlink_ns_capable);
893 * netlink_capable - Netlink global message capability test
894 * @skb: socket buffer holding a netlink command from userspace
895 * @cap: The capability to use
897 * Test to see if the opener of the socket we received the message
898 * from had when the netlink socket was created and the sender of the
899 * message has the capability @cap in all user namespaces.
901 bool netlink_capable(const struct sk_buff *skb, int cap)
903 return netlink_ns_capable(skb, &init_user_ns, cap);
905 EXPORT_SYMBOL(netlink_capable);
908 * netlink_net_capable - Netlink network namespace message capability test
909 * @skb: socket buffer holding a netlink command from userspace
910 * @cap: The capability to use
912 * Test to see if the opener of the socket we received the message
913 * from had when the netlink socket was created and the sender of the
914 * message has the capability @cap over the network namespace of
915 * the socket we received the message from.
917 bool netlink_net_capable(const struct sk_buff *skb, int cap)
919 return netlink_ns_capable(skb, sock_net(skb->sk)->user_ns, cap);
921 EXPORT_SYMBOL(netlink_net_capable);
923 static inline int netlink_allowed(const struct socket *sock, unsigned int flag)
925 return (nl_table[sock->sk->sk_protocol].flags & flag) ||
926 ns_capable(sock_net(sock->sk)->user_ns, CAP_NET_ADMIN);
930 netlink_update_subscriptions(struct sock *sk, unsigned int subscriptions)
932 struct netlink_sock *nlk = nlk_sk(sk);
934 if (nlk->subscriptions && !subscriptions)
935 __sk_del_bind_node(sk);
936 else if (!nlk->subscriptions && subscriptions)
937 sk_add_bind_node(sk, &nl_table[sk->sk_protocol].mc_list);
938 nlk->subscriptions = subscriptions;
941 static int netlink_realloc_groups(struct sock *sk)
943 struct netlink_sock *nlk = nlk_sk(sk);
945 unsigned long *new_groups;
948 netlink_table_grab();
950 groups = nl_table[sk->sk_protocol].groups;
951 if (!nl_table[sk->sk_protocol].registered) {
956 if (nlk->ngroups >= groups)
959 new_groups = krealloc(nlk->groups, NLGRPSZ(groups), GFP_ATOMIC);
960 if (new_groups == NULL) {
964 memset((char *)new_groups + NLGRPSZ(nlk->ngroups), 0,
965 NLGRPSZ(groups) - NLGRPSZ(nlk->ngroups));
967 nlk->groups = new_groups;
968 nlk->ngroups = groups;
970 netlink_table_ungrab();
974 static void netlink_undo_bind(int group, long unsigned int groups,
977 struct netlink_sock *nlk = nlk_sk(sk);
980 if (!nlk->netlink_unbind)
983 for (undo = 0; undo < group; undo++)
984 if (test_bit(undo, &groups))
985 nlk->netlink_unbind(sock_net(sk), undo + 1);
988 static int netlink_bind(struct socket *sock, struct sockaddr *addr,
991 struct sock *sk = sock->sk;
992 struct net *net = sock_net(sk);
993 struct netlink_sock *nlk = nlk_sk(sk);
994 struct sockaddr_nl *nladdr = (struct sockaddr_nl *)addr;
996 unsigned long groups;
999 if (addr_len < sizeof(struct sockaddr_nl))
1002 if (nladdr->nl_family != AF_NETLINK)
1004 groups = nladdr->nl_groups;
1006 /* Only superuser is allowed to listen multicasts */
1008 if (!netlink_allowed(sock, NL_CFG_F_NONROOT_RECV))
1010 err = netlink_realloc_groups(sk);
1015 if (nlk->ngroups < BITS_PER_LONG)
1016 groups &= (1UL << nlk->ngroups) - 1;
1018 /* Paired with WRITE_ONCE() in netlink_insert() */
1019 bound = READ_ONCE(nlk->bound);
1021 /* Ensure nlk->portid is up-to-date. */
1024 if (nladdr->nl_pid != nlk->portid)
1028 if (nlk->netlink_bind && groups) {
1031 /* nl_groups is a u32, so cap the maximum groups we can bind */
1032 for (group = 0; group < BITS_PER_TYPE(u32); group++) {
1033 if (!test_bit(group, &groups))
1035 err = nlk->netlink_bind(net, group + 1);
1038 netlink_undo_bind(group, groups, sk);
1043 /* No need for barriers here as we return to user-space without
1044 * using any of the bound attributes.
1046 netlink_lock_table();
1048 err = nladdr->nl_pid ?
1049 netlink_insert(sk, nladdr->nl_pid) :
1050 netlink_autobind(sock);
1052 netlink_undo_bind(BITS_PER_TYPE(u32), groups, sk);
1057 if (!groups && (nlk->groups == NULL || !(u32)nlk->groups[0]))
1059 netlink_unlock_table();
1061 netlink_table_grab();
1062 netlink_update_subscriptions(sk, nlk->subscriptions +
1064 hweight32(nlk->groups[0]));
1065 nlk->groups[0] = (nlk->groups[0] & ~0xffffffffUL) | groups;
1066 netlink_update_listeners(sk);
1067 netlink_table_ungrab();
1072 netlink_unlock_table();
1076 static int netlink_connect(struct socket *sock, struct sockaddr *addr,
1077 int alen, int flags)
1080 struct sock *sk = sock->sk;
1081 struct netlink_sock *nlk = nlk_sk(sk);
1082 struct sockaddr_nl *nladdr = (struct sockaddr_nl *)addr;
1084 if (alen < sizeof(addr->sa_family))
1087 if (addr->sa_family == AF_UNSPEC) {
1088 sk->sk_state = NETLINK_UNCONNECTED;
1089 nlk->dst_portid = 0;
1093 if (addr->sa_family != AF_NETLINK)
1096 if (alen < sizeof(struct sockaddr_nl))
1099 if ((nladdr->nl_groups || nladdr->nl_pid) &&
1100 !netlink_allowed(sock, NL_CFG_F_NONROOT_SEND))
1103 /* No need for barriers here as we return to user-space without
1104 * using any of the bound attributes.
1105 * Paired with WRITE_ONCE() in netlink_insert().
1107 if (!READ_ONCE(nlk->bound))
1108 err = netlink_autobind(sock);
1111 sk->sk_state = NETLINK_CONNECTED;
1112 nlk->dst_portid = nladdr->nl_pid;
1113 nlk->dst_group = ffs(nladdr->nl_groups);
1119 static int netlink_getname(struct socket *sock, struct sockaddr *addr,
1122 struct sock *sk = sock->sk;
1123 struct netlink_sock *nlk = nlk_sk(sk);
1124 DECLARE_SOCKADDR(struct sockaddr_nl *, nladdr, addr);
1126 nladdr->nl_family = AF_NETLINK;
1130 nladdr->nl_pid = nlk->dst_portid;
1131 nladdr->nl_groups = netlink_group_mask(nlk->dst_group);
1133 nladdr->nl_pid = nlk->portid;
1134 netlink_lock_table();
1135 nladdr->nl_groups = nlk->groups ? nlk->groups[0] : 0;
1136 netlink_unlock_table();
1138 return sizeof(*nladdr);
1141 static int netlink_ioctl(struct socket *sock, unsigned int cmd,
1144 /* try to hand this ioctl down to the NIC drivers.
1146 return -ENOIOCTLCMD;
1149 static struct sock *netlink_getsockbyportid(struct sock *ssk, u32 portid)
1152 struct netlink_sock *nlk;
1154 sock = netlink_lookup(sock_net(ssk), ssk->sk_protocol, portid);
1156 return ERR_PTR(-ECONNREFUSED);
1158 /* Don't bother queuing skb if kernel socket has no input function */
1160 if (sock->sk_state == NETLINK_CONNECTED &&
1161 nlk->dst_portid != nlk_sk(ssk)->portid) {
1163 return ERR_PTR(-ECONNREFUSED);
1168 struct sock *netlink_getsockbyfilp(struct file *filp)
1170 struct inode *inode = file_inode(filp);
1173 if (!S_ISSOCK(inode->i_mode))
1174 return ERR_PTR(-ENOTSOCK);
1176 sock = SOCKET_I(inode)->sk;
1177 if (sock->sk_family != AF_NETLINK)
1178 return ERR_PTR(-EINVAL);
1184 static struct sk_buff *netlink_alloc_large_skb(unsigned int size,
1187 struct sk_buff *skb;
1190 if (size <= NLMSG_GOODSIZE || broadcast)
1191 return alloc_skb(size, GFP_KERNEL);
1193 size = SKB_DATA_ALIGN(size) +
1194 SKB_DATA_ALIGN(sizeof(struct skb_shared_info));
1196 data = vmalloc(size);
1200 skb = __build_skb(data, size);
1204 skb->destructor = netlink_skb_destructor;
1210 * Attach a skb to a netlink socket.
1211 * The caller must hold a reference to the destination socket. On error, the
1212 * reference is dropped. The skb is not send to the destination, just all
1213 * all error checks are performed and memory in the queue is reserved.
1215 * < 0: error. skb freed, reference to sock dropped.
1217 * 1: repeat lookup - reference dropped while waiting for socket memory.
1219 int netlink_attachskb(struct sock *sk, struct sk_buff *skb,
1220 long *timeo, struct sock *ssk)
1222 struct netlink_sock *nlk;
1226 if ((atomic_read(&sk->sk_rmem_alloc) > sk->sk_rcvbuf ||
1227 test_bit(NETLINK_S_CONGESTED, &nlk->state))) {
1228 DECLARE_WAITQUEUE(wait, current);
1230 if (!ssk || netlink_is_kernel(ssk))
1231 netlink_overrun(sk);
1237 __set_current_state(TASK_INTERRUPTIBLE);
1238 add_wait_queue(&nlk->wait, &wait);
1240 if ((atomic_read(&sk->sk_rmem_alloc) > sk->sk_rcvbuf ||
1241 test_bit(NETLINK_S_CONGESTED, &nlk->state)) &&
1242 !sock_flag(sk, SOCK_DEAD))
1243 *timeo = schedule_timeout(*timeo);
1245 __set_current_state(TASK_RUNNING);
1246 remove_wait_queue(&nlk->wait, &wait);
1249 if (signal_pending(current)) {
1251 return sock_intr_errno(*timeo);
1255 netlink_skb_set_owner_r(skb, sk);
1259 static int __netlink_sendskb(struct sock *sk, struct sk_buff *skb)
1263 netlink_deliver_tap(sock_net(sk), skb);
1265 skb_queue_tail(&sk->sk_receive_queue, skb);
1266 sk->sk_data_ready(sk);
1270 int netlink_sendskb(struct sock *sk, struct sk_buff *skb)
1272 int len = __netlink_sendskb(sk, skb);
1278 void netlink_detachskb(struct sock *sk, struct sk_buff *skb)
1284 static struct sk_buff *netlink_trim(struct sk_buff *skb, gfp_t allocation)
1288 WARN_ON(skb->sk != NULL);
1289 delta = skb->end - skb->tail;
1290 if (is_vmalloc_addr(skb->head) || delta * 2 < skb->truesize)
1293 if (skb_shared(skb)) {
1294 struct sk_buff *nskb = skb_clone(skb, allocation);
1301 pskb_expand_head(skb, 0, -delta,
1302 (allocation & ~__GFP_DIRECT_RECLAIM) |
1303 __GFP_NOWARN | __GFP_NORETRY);
1307 static int netlink_unicast_kernel(struct sock *sk, struct sk_buff *skb,
1311 struct netlink_sock *nlk = nlk_sk(sk);
1313 ret = -ECONNREFUSED;
1314 if (nlk->netlink_rcv != NULL) {
1316 netlink_skb_set_owner_r(skb, sk);
1317 NETLINK_CB(skb).sk = ssk;
1318 netlink_deliver_tap_kernel(sk, ssk, skb);
1319 nlk->netlink_rcv(skb);
1328 int netlink_unicast(struct sock *ssk, struct sk_buff *skb,
1329 u32 portid, int nonblock)
1335 skb = netlink_trim(skb, gfp_any());
1337 timeo = sock_sndtimeo(ssk, nonblock);
1339 sk = netlink_getsockbyportid(ssk, portid);
1344 if (netlink_is_kernel(sk))
1345 return netlink_unicast_kernel(sk, skb, ssk);
1347 if (sk_filter(sk, skb)) {
1354 err = netlink_attachskb(sk, skb, &timeo, ssk);
1360 return netlink_sendskb(sk, skb);
1362 EXPORT_SYMBOL(netlink_unicast);
1364 int netlink_has_listeners(struct sock *sk, unsigned int group)
1367 struct listeners *listeners;
1369 BUG_ON(!netlink_is_kernel(sk));
1372 listeners = rcu_dereference(nl_table[sk->sk_protocol].listeners);
1374 if (listeners && group - 1 < nl_table[sk->sk_protocol].groups)
1375 res = test_bit(group - 1, listeners->masks);
1381 EXPORT_SYMBOL_GPL(netlink_has_listeners);
1383 bool netlink_strict_get_check(struct sk_buff *skb)
1385 const struct netlink_sock *nlk = nlk_sk(NETLINK_CB(skb).sk);
1387 return nlk->flags & NETLINK_F_STRICT_CHK;
1389 EXPORT_SYMBOL_GPL(netlink_strict_get_check);
1391 static int netlink_broadcast_deliver(struct sock *sk, struct sk_buff *skb)
1393 struct netlink_sock *nlk = nlk_sk(sk);
1395 if (atomic_read(&sk->sk_rmem_alloc) <= sk->sk_rcvbuf &&
1396 !test_bit(NETLINK_S_CONGESTED, &nlk->state)) {
1397 netlink_skb_set_owner_r(skb, sk);
1398 __netlink_sendskb(sk, skb);
1399 return atomic_read(&sk->sk_rmem_alloc) > (sk->sk_rcvbuf >> 1);
1404 struct netlink_broadcast_data {
1405 struct sock *exclude_sk;
1410 int delivery_failure;
1414 struct sk_buff *skb, *skb2;
1417 static void do_one_broadcast(struct sock *sk,
1418 struct netlink_broadcast_data *p)
1420 struct netlink_sock *nlk = nlk_sk(sk);
1423 if (p->exclude_sk == sk)
1426 if (nlk->portid == p->portid || p->group - 1 >= nlk->ngroups ||
1427 !test_bit(p->group - 1, nlk->groups))
1430 if (!net_eq(sock_net(sk), p->net)) {
1431 if (!(nlk->flags & NETLINK_F_LISTEN_ALL_NSID))
1434 if (!peernet_has_id(sock_net(sk), p->net))
1437 if (!file_ns_capable(sk->sk_socket->file, p->net->user_ns,
1443 netlink_overrun(sk);
1448 if (p->skb2 == NULL) {
1449 if (skb_shared(p->skb)) {
1450 p->skb2 = skb_clone(p->skb, p->allocation);
1452 p->skb2 = skb_get(p->skb);
1454 * skb ownership may have been set when
1455 * delivered to a previous socket.
1457 skb_orphan(p->skb2);
1460 if (p->skb2 == NULL) {
1461 netlink_overrun(sk);
1462 /* Clone failed. Notify ALL listeners. */
1464 if (nlk->flags & NETLINK_F_BROADCAST_SEND_ERROR)
1465 p->delivery_failure = 1;
1468 if (sk_filter(sk, p->skb2)) {
1473 NETLINK_CB(p->skb2).nsid = peernet2id(sock_net(sk), p->net);
1474 if (NETLINK_CB(p->skb2).nsid != NETNSA_NSID_NOT_ASSIGNED)
1475 NETLINK_CB(p->skb2).nsid_is_set = true;
1476 val = netlink_broadcast_deliver(sk, p->skb2);
1478 netlink_overrun(sk);
1479 if (nlk->flags & NETLINK_F_BROADCAST_SEND_ERROR)
1480 p->delivery_failure = 1;
1482 p->congested |= val;
1490 int netlink_broadcast(struct sock *ssk, struct sk_buff *skb, u32 portid,
1491 u32 group, gfp_t allocation)
1493 struct net *net = sock_net(ssk);
1494 struct netlink_broadcast_data info;
1497 skb = netlink_trim(skb, allocation);
1499 info.exclude_sk = ssk;
1501 info.portid = portid;
1504 info.delivery_failure = 0;
1507 info.allocation = allocation;
1511 /* While we sleep in clone, do not allow to change socket list */
1513 netlink_lock_table();
1515 sk_for_each_bound(sk, &nl_table[ssk->sk_protocol].mc_list)
1516 do_one_broadcast(sk, &info);
1520 netlink_unlock_table();
1522 if (info.delivery_failure) {
1523 kfree_skb(info.skb2);
1526 consume_skb(info.skb2);
1528 if (info.delivered) {
1529 if (info.congested && gfpflags_allow_blocking(allocation))
1535 EXPORT_SYMBOL(netlink_broadcast);
1537 struct netlink_set_err_data {
1538 struct sock *exclude_sk;
1544 static int do_one_set_err(struct sock *sk, struct netlink_set_err_data *p)
1546 struct netlink_sock *nlk = nlk_sk(sk);
1549 if (sk == p->exclude_sk)
1552 if (!net_eq(sock_net(sk), sock_net(p->exclude_sk)))
1555 if (nlk->portid == p->portid || p->group - 1 >= nlk->ngroups ||
1556 !test_bit(p->group - 1, nlk->groups))
1559 if (p->code == ENOBUFS && nlk->flags & NETLINK_F_RECV_NO_ENOBUFS) {
1564 sk->sk_err = p->code;
1565 sk_error_report(sk);
1571 * netlink_set_err - report error to broadcast listeners
1572 * @ssk: the kernel netlink socket, as returned by netlink_kernel_create()
1573 * @portid: the PORTID of a process that we want to skip (if any)
1574 * @group: the broadcast group that will notice the error
1575 * @code: error code, must be negative (as usual in kernelspace)
1577 * This function returns the number of broadcast listeners that have set the
1578 * NETLINK_NO_ENOBUFS socket option.
1580 int netlink_set_err(struct sock *ssk, u32 portid, u32 group, int code)
1582 struct netlink_set_err_data info;
1586 info.exclude_sk = ssk;
1587 info.portid = portid;
1589 /* sk->sk_err wants a positive error value */
1592 read_lock(&nl_table_lock);
1594 sk_for_each_bound(sk, &nl_table[ssk->sk_protocol].mc_list)
1595 ret += do_one_set_err(sk, &info);
1597 read_unlock(&nl_table_lock);
1600 EXPORT_SYMBOL(netlink_set_err);
1602 /* must be called with netlink table grabbed */
1603 static void netlink_update_socket_mc(struct netlink_sock *nlk,
1607 int old, new = !!is_new, subscriptions;
1609 old = test_bit(group - 1, nlk->groups);
1610 subscriptions = nlk->subscriptions - old + new;
1612 __set_bit(group - 1, nlk->groups);
1614 __clear_bit(group - 1, nlk->groups);
1615 netlink_update_subscriptions(&nlk->sk, subscriptions);
1616 netlink_update_listeners(&nlk->sk);
1619 static int netlink_setsockopt(struct socket *sock, int level, int optname,
1620 sockptr_t optval, unsigned int optlen)
1622 struct sock *sk = sock->sk;
1623 struct netlink_sock *nlk = nlk_sk(sk);
1624 unsigned int val = 0;
1627 if (level != SOL_NETLINK)
1628 return -ENOPROTOOPT;
1630 if (optlen >= sizeof(int) &&
1631 copy_from_sockptr(&val, optval, sizeof(val)))
1635 case NETLINK_PKTINFO:
1637 nlk->flags |= NETLINK_F_RECV_PKTINFO;
1639 nlk->flags &= ~NETLINK_F_RECV_PKTINFO;
1642 case NETLINK_ADD_MEMBERSHIP:
1643 case NETLINK_DROP_MEMBERSHIP: {
1644 if (!netlink_allowed(sock, NL_CFG_F_NONROOT_RECV))
1646 err = netlink_realloc_groups(sk);
1649 if (!val || val - 1 >= nlk->ngroups)
1651 if (optname == NETLINK_ADD_MEMBERSHIP && nlk->netlink_bind) {
1652 err = nlk->netlink_bind(sock_net(sk), val);
1656 netlink_table_grab();
1657 netlink_update_socket_mc(nlk, val,
1658 optname == NETLINK_ADD_MEMBERSHIP);
1659 netlink_table_ungrab();
1660 if (optname == NETLINK_DROP_MEMBERSHIP && nlk->netlink_unbind)
1661 nlk->netlink_unbind(sock_net(sk), val);
1666 case NETLINK_BROADCAST_ERROR:
1668 nlk->flags |= NETLINK_F_BROADCAST_SEND_ERROR;
1670 nlk->flags &= ~NETLINK_F_BROADCAST_SEND_ERROR;
1673 case NETLINK_NO_ENOBUFS:
1675 nlk->flags |= NETLINK_F_RECV_NO_ENOBUFS;
1676 clear_bit(NETLINK_S_CONGESTED, &nlk->state);
1677 wake_up_interruptible(&nlk->wait);
1679 nlk->flags &= ~NETLINK_F_RECV_NO_ENOBUFS;
1683 case NETLINK_LISTEN_ALL_NSID:
1684 if (!ns_capable(sock_net(sk)->user_ns, CAP_NET_BROADCAST))
1688 nlk->flags |= NETLINK_F_LISTEN_ALL_NSID;
1690 nlk->flags &= ~NETLINK_F_LISTEN_ALL_NSID;
1693 case NETLINK_CAP_ACK:
1695 nlk->flags |= NETLINK_F_CAP_ACK;
1697 nlk->flags &= ~NETLINK_F_CAP_ACK;
1700 case NETLINK_EXT_ACK:
1702 nlk->flags |= NETLINK_F_EXT_ACK;
1704 nlk->flags &= ~NETLINK_F_EXT_ACK;
1707 case NETLINK_GET_STRICT_CHK:
1709 nlk->flags |= NETLINK_F_STRICT_CHK;
1711 nlk->flags &= ~NETLINK_F_STRICT_CHK;
1720 static int netlink_getsockopt(struct socket *sock, int level, int optname,
1721 char __user *optval, int __user *optlen)
1723 struct sock *sk = sock->sk;
1724 struct netlink_sock *nlk = nlk_sk(sk);
1727 if (level != SOL_NETLINK)
1728 return -ENOPROTOOPT;
1730 if (get_user(len, optlen))
1736 case NETLINK_PKTINFO:
1737 if (len < sizeof(int))
1740 val = nlk->flags & NETLINK_F_RECV_PKTINFO ? 1 : 0;
1741 if (put_user(len, optlen) ||
1742 put_user(val, optval))
1746 case NETLINK_BROADCAST_ERROR:
1747 if (len < sizeof(int))
1750 val = nlk->flags & NETLINK_F_BROADCAST_SEND_ERROR ? 1 : 0;
1751 if (put_user(len, optlen) ||
1752 put_user(val, optval))
1756 case NETLINK_NO_ENOBUFS:
1757 if (len < sizeof(int))
1760 val = nlk->flags & NETLINK_F_RECV_NO_ENOBUFS ? 1 : 0;
1761 if (put_user(len, optlen) ||
1762 put_user(val, optval))
1766 case NETLINK_LIST_MEMBERSHIPS: {
1767 int pos, idx, shift;
1770 netlink_lock_table();
1771 for (pos = 0; pos * 8 < nlk->ngroups; pos += sizeof(u32)) {
1772 if (len - pos < sizeof(u32))
1775 idx = pos / sizeof(unsigned long);
1776 shift = (pos % sizeof(unsigned long)) * 8;
1777 if (put_user((u32)(nlk->groups[idx] >> shift),
1778 (u32 __user *)(optval + pos))) {
1783 if (put_user(ALIGN(nlk->ngroups / 8, sizeof(u32)), optlen))
1785 netlink_unlock_table();
1788 case NETLINK_CAP_ACK:
1789 if (len < sizeof(int))
1792 val = nlk->flags & NETLINK_F_CAP_ACK ? 1 : 0;
1793 if (put_user(len, optlen) ||
1794 put_user(val, optval))
1798 case NETLINK_EXT_ACK:
1799 if (len < sizeof(int))
1802 val = nlk->flags & NETLINK_F_EXT_ACK ? 1 : 0;
1803 if (put_user(len, optlen) || put_user(val, optval))
1807 case NETLINK_GET_STRICT_CHK:
1808 if (len < sizeof(int))
1811 val = nlk->flags & NETLINK_F_STRICT_CHK ? 1 : 0;
1812 if (put_user(len, optlen) || put_user(val, optval))
1822 static void netlink_cmsg_recv_pktinfo(struct msghdr *msg, struct sk_buff *skb)
1824 struct nl_pktinfo info;
1826 info.group = NETLINK_CB(skb).dst_group;
1827 put_cmsg(msg, SOL_NETLINK, NETLINK_PKTINFO, sizeof(info), &info);
1830 static void netlink_cmsg_listen_all_nsid(struct sock *sk, struct msghdr *msg,
1831 struct sk_buff *skb)
1833 if (!NETLINK_CB(skb).nsid_is_set)
1836 put_cmsg(msg, SOL_NETLINK, NETLINK_LISTEN_ALL_NSID, sizeof(int),
1837 &NETLINK_CB(skb).nsid);
1840 static int netlink_sendmsg(struct socket *sock, struct msghdr *msg, size_t len)
1842 struct sock *sk = sock->sk;
1843 struct netlink_sock *nlk = nlk_sk(sk);
1844 DECLARE_SOCKADDR(struct sockaddr_nl *, addr, msg->msg_name);
1847 struct sk_buff *skb;
1849 struct scm_cookie scm;
1850 u32 netlink_skb_flags = 0;
1852 if (msg->msg_flags & MSG_OOB)
1856 pr_warn_once("Zero length message leads to an empty skb\n");
1860 err = scm_send(sock, msg, &scm, true);
1864 if (msg->msg_namelen) {
1866 if (msg->msg_namelen < sizeof(struct sockaddr_nl))
1868 if (addr->nl_family != AF_NETLINK)
1870 dst_portid = addr->nl_pid;
1871 dst_group = ffs(addr->nl_groups);
1873 if ((dst_group || dst_portid) &&
1874 !netlink_allowed(sock, NL_CFG_F_NONROOT_SEND))
1876 netlink_skb_flags |= NETLINK_SKB_DST;
1878 dst_portid = nlk->dst_portid;
1879 dst_group = nlk->dst_group;
1882 /* Paired with WRITE_ONCE() in netlink_insert() */
1883 if (!READ_ONCE(nlk->bound)) {
1884 err = netlink_autobind(sock);
1888 /* Ensure nlk is hashed and visible. */
1893 if (len > sk->sk_sndbuf - 32)
1896 skb = netlink_alloc_large_skb(len, dst_group);
1900 NETLINK_CB(skb).portid = nlk->portid;
1901 NETLINK_CB(skb).dst_group = dst_group;
1902 NETLINK_CB(skb).creds = scm.creds;
1903 NETLINK_CB(skb).flags = netlink_skb_flags;
1906 if (memcpy_from_msg(skb_put(skb, len), msg, len)) {
1911 err = security_netlink_send(sk, skb);
1918 refcount_inc(&skb->users);
1919 netlink_broadcast(sk, skb, dst_portid, dst_group, GFP_KERNEL);
1921 err = netlink_unicast(sk, skb, dst_portid, msg->msg_flags & MSG_DONTWAIT);
1928 static int netlink_recvmsg(struct socket *sock, struct msghdr *msg, size_t len,
1931 struct scm_cookie scm;
1932 struct sock *sk = sock->sk;
1933 struct netlink_sock *nlk = nlk_sk(sk);
1934 int noblock = flags & MSG_DONTWAIT;
1936 struct sk_buff *skb, *data_skb;
1939 if (flags & MSG_OOB)
1944 skb = skb_recv_datagram(sk, flags, noblock, &err);
1950 #ifdef CONFIG_COMPAT_NETLINK_MESSAGES
1951 if (unlikely(skb_shinfo(skb)->frag_list)) {
1953 * If this skb has a frag_list, then here that means that we
1954 * will have to use the frag_list skb's data for compat tasks
1955 * and the regular skb's data for normal (non-compat) tasks.
1957 * If we need to send the compat skb, assign it to the
1958 * 'data_skb' variable so that it will be used below for data
1959 * copying. We keep 'skb' for everything else, including
1960 * freeing both later.
1962 if (flags & MSG_CMSG_COMPAT)
1963 data_skb = skb_shinfo(skb)->frag_list;
1967 /* Record the max length of recvmsg() calls for future allocations */
1968 nlk->max_recvmsg_len = max(nlk->max_recvmsg_len, len);
1969 nlk->max_recvmsg_len = min_t(size_t, nlk->max_recvmsg_len,
1970 SKB_WITH_OVERHEAD(32768));
1972 copied = data_skb->len;
1974 msg->msg_flags |= MSG_TRUNC;
1978 skb_reset_transport_header(data_skb);
1979 err = skb_copy_datagram_msg(data_skb, 0, msg, copied);
1981 if (msg->msg_name) {
1982 DECLARE_SOCKADDR(struct sockaddr_nl *, addr, msg->msg_name);
1983 addr->nl_family = AF_NETLINK;
1985 addr->nl_pid = NETLINK_CB(skb).portid;
1986 addr->nl_groups = netlink_group_mask(NETLINK_CB(skb).dst_group);
1987 msg->msg_namelen = sizeof(*addr);
1990 if (nlk->flags & NETLINK_F_RECV_PKTINFO)
1991 netlink_cmsg_recv_pktinfo(msg, skb);
1992 if (nlk->flags & NETLINK_F_LISTEN_ALL_NSID)
1993 netlink_cmsg_listen_all_nsid(sk, msg, skb);
1995 memset(&scm, 0, sizeof(scm));
1996 scm.creds = *NETLINK_CREDS(skb);
1997 if (flags & MSG_TRUNC)
1998 copied = data_skb->len;
2000 skb_free_datagram(sk, skb);
2002 if (nlk->cb_running &&
2003 atomic_read(&sk->sk_rmem_alloc) <= sk->sk_rcvbuf / 2) {
2004 ret = netlink_dump(sk);
2007 sk_error_report(sk);
2011 scm_recv(sock, msg, &scm, flags);
2013 netlink_rcv_wake(sk);
2014 return err ? : copied;
2017 static void netlink_data_ready(struct sock *sk)
2023 * We export these functions to other modules. They provide a
2024 * complete set of kernel non-blocking support for message
2029 __netlink_kernel_create(struct net *net, int unit, struct module *module,
2030 struct netlink_kernel_cfg *cfg)
2032 struct socket *sock;
2034 struct netlink_sock *nlk;
2035 struct listeners *listeners = NULL;
2036 struct mutex *cb_mutex = cfg ? cfg->cb_mutex : NULL;
2037 unsigned int groups;
2041 if (unit < 0 || unit >= MAX_LINKS)
2044 if (sock_create_lite(PF_NETLINK, SOCK_DGRAM, unit, &sock))
2047 if (__netlink_create(net, sock, cb_mutex, unit, 1) < 0)
2048 goto out_sock_release_nosk;
2052 if (!cfg || cfg->groups < 32)
2055 groups = cfg->groups;
2057 listeners = kzalloc(sizeof(*listeners) + NLGRPSZ(groups), GFP_KERNEL);
2059 goto out_sock_release;
2061 sk->sk_data_ready = netlink_data_ready;
2062 if (cfg && cfg->input)
2063 nlk_sk(sk)->netlink_rcv = cfg->input;
2065 if (netlink_insert(sk, 0))
2066 goto out_sock_release;
2069 nlk->flags |= NETLINK_F_KERNEL_SOCKET;
2071 netlink_table_grab();
2072 if (!nl_table[unit].registered) {
2073 nl_table[unit].groups = groups;
2074 rcu_assign_pointer(nl_table[unit].listeners, listeners);
2075 nl_table[unit].cb_mutex = cb_mutex;
2076 nl_table[unit].module = module;
2078 nl_table[unit].bind = cfg->bind;
2079 nl_table[unit].unbind = cfg->unbind;
2080 nl_table[unit].flags = cfg->flags;
2082 nl_table[unit].compare = cfg->compare;
2084 nl_table[unit].registered = 1;
2087 nl_table[unit].registered++;
2089 netlink_table_ungrab();
2094 netlink_kernel_release(sk);
2097 out_sock_release_nosk:
2101 EXPORT_SYMBOL(__netlink_kernel_create);
2104 netlink_kernel_release(struct sock *sk)
2106 if (sk == NULL || sk->sk_socket == NULL)
2109 sock_release(sk->sk_socket);
2111 EXPORT_SYMBOL(netlink_kernel_release);
2113 int __netlink_change_ngroups(struct sock *sk, unsigned int groups)
2115 struct listeners *new, *old;
2116 struct netlink_table *tbl = &nl_table[sk->sk_protocol];
2121 if (NLGRPSZ(tbl->groups) < NLGRPSZ(groups)) {
2122 new = kzalloc(sizeof(*new) + NLGRPSZ(groups), GFP_ATOMIC);
2125 old = nl_deref_protected(tbl->listeners);
2126 memcpy(new->masks, old->masks, NLGRPSZ(tbl->groups));
2127 rcu_assign_pointer(tbl->listeners, new);
2129 kfree_rcu(old, rcu);
2131 tbl->groups = groups;
2137 * netlink_change_ngroups - change number of multicast groups
2139 * This changes the number of multicast groups that are available
2140 * on a certain netlink family. Note that it is not possible to
2141 * change the number of groups to below 32. Also note that it does
2142 * not implicitly call netlink_clear_multicast_users() when the
2143 * number of groups is reduced.
2145 * @sk: The kernel netlink socket, as returned by netlink_kernel_create().
2146 * @groups: The new number of groups.
2148 int netlink_change_ngroups(struct sock *sk, unsigned int groups)
2152 netlink_table_grab();
2153 err = __netlink_change_ngroups(sk, groups);
2154 netlink_table_ungrab();
2159 void __netlink_clear_multicast_users(struct sock *ksk, unsigned int group)
2162 struct netlink_table *tbl = &nl_table[ksk->sk_protocol];
2164 sk_for_each_bound(sk, &tbl->mc_list)
2165 netlink_update_socket_mc(nlk_sk(sk), group, 0);
2169 __nlmsg_put(struct sk_buff *skb, u32 portid, u32 seq, int type, int len, int flags)
2171 struct nlmsghdr *nlh;
2172 int size = nlmsg_msg_size(len);
2174 nlh = skb_put(skb, NLMSG_ALIGN(size));
2175 nlh->nlmsg_type = type;
2176 nlh->nlmsg_len = size;
2177 nlh->nlmsg_flags = flags;
2178 nlh->nlmsg_pid = portid;
2179 nlh->nlmsg_seq = seq;
2180 if (!__builtin_constant_p(size) || NLMSG_ALIGN(size) - size != 0)
2181 memset(nlmsg_data(nlh) + len, 0, NLMSG_ALIGN(size) - size);
2184 EXPORT_SYMBOL(__nlmsg_put);
2187 * It looks a bit ugly.
2188 * It would be better to create kernel thread.
2191 static int netlink_dump_done(struct netlink_sock *nlk, struct sk_buff *skb,
2192 struct netlink_callback *cb,
2193 struct netlink_ext_ack *extack)
2195 struct nlmsghdr *nlh;
2197 nlh = nlmsg_put_answer(skb, cb, NLMSG_DONE, sizeof(nlk->dump_done_errno),
2198 NLM_F_MULTI | cb->answer_flags);
2202 nl_dump_check_consistent(cb, nlh);
2203 memcpy(nlmsg_data(nlh), &nlk->dump_done_errno, sizeof(nlk->dump_done_errno));
2205 if (extack->_msg && nlk->flags & NETLINK_F_EXT_ACK) {
2206 nlh->nlmsg_flags |= NLM_F_ACK_TLVS;
2207 if (!nla_put_string(skb, NLMSGERR_ATTR_MSG, extack->_msg))
2208 nlmsg_end(skb, nlh);
2214 static int netlink_dump(struct sock *sk)
2216 struct netlink_sock *nlk = nlk_sk(sk);
2217 struct netlink_ext_ack extack = {};
2218 struct netlink_callback *cb;
2219 struct sk_buff *skb = NULL;
2220 struct module *module;
2225 mutex_lock(nlk->cb_mutex);
2226 if (!nlk->cb_running) {
2231 if (atomic_read(&sk->sk_rmem_alloc) >= sk->sk_rcvbuf)
2234 /* NLMSG_GOODSIZE is small to avoid high order allocations being
2235 * required, but it makes sense to _attempt_ a 16K bytes allocation
2236 * to reduce number of system calls on dump operations, if user
2237 * ever provided a big enough buffer.
2240 alloc_min_size = max_t(int, cb->min_dump_alloc, NLMSG_GOODSIZE);
2242 if (alloc_min_size < nlk->max_recvmsg_len) {
2243 alloc_size = nlk->max_recvmsg_len;
2244 skb = alloc_skb(alloc_size,
2245 (GFP_KERNEL & ~__GFP_DIRECT_RECLAIM) |
2246 __GFP_NOWARN | __GFP_NORETRY);
2249 alloc_size = alloc_min_size;
2250 skb = alloc_skb(alloc_size, GFP_KERNEL);
2255 /* Trim skb to allocated size. User is expected to provide buffer as
2256 * large as max(min_dump_alloc, 16KiB (mac_recvmsg_len capped at
2257 * netlink_recvmsg())). dump will pack as many smaller messages as
2258 * could fit within the allocated skb. skb is typically allocated
2259 * with larger space than required (could be as much as near 2x the
2260 * requested size with align to next power of 2 approach). Allowing
2261 * dump to use the excess space makes it difficult for a user to have a
2262 * reasonable static buffer based on the expected largest dump of a
2263 * single netdev. The outcome is MSG_TRUNC error.
2265 skb_reserve(skb, skb_tailroom(skb) - alloc_size);
2266 netlink_skb_set_owner_r(skb, sk);
2268 if (nlk->dump_done_errno > 0) {
2269 cb->extack = &extack;
2270 nlk->dump_done_errno = cb->dump(skb, cb);
2274 if (nlk->dump_done_errno > 0 ||
2275 skb_tailroom(skb) < nlmsg_total_size(sizeof(nlk->dump_done_errno))) {
2276 mutex_unlock(nlk->cb_mutex);
2278 if (sk_filter(sk, skb))
2281 __netlink_sendskb(sk, skb);
2285 if (netlink_dump_done(nlk, skb, cb, &extack))
2288 #ifdef CONFIG_COMPAT_NETLINK_MESSAGES
2289 /* frag_list skb's data is used for compat tasks
2290 * and the regular skb's data for normal (non-compat) tasks.
2291 * See netlink_recvmsg().
2293 if (unlikely(skb_shinfo(skb)->frag_list)) {
2294 if (netlink_dump_done(nlk, skb_shinfo(skb)->frag_list, cb, &extack))
2299 if (sk_filter(sk, skb))
2302 __netlink_sendskb(sk, skb);
2307 nlk->cb_running = false;
2308 module = cb->module;
2310 mutex_unlock(nlk->cb_mutex);
2316 mutex_unlock(nlk->cb_mutex);
2321 int __netlink_dump_start(struct sock *ssk, struct sk_buff *skb,
2322 const struct nlmsghdr *nlh,
2323 struct netlink_dump_control *control)
2325 struct netlink_sock *nlk, *nlk2;
2326 struct netlink_callback *cb;
2330 refcount_inc(&skb->users);
2332 sk = netlink_lookup(sock_net(ssk), ssk->sk_protocol, NETLINK_CB(skb).portid);
2334 ret = -ECONNREFUSED;
2339 mutex_lock(nlk->cb_mutex);
2340 /* A dump is in progress... */
2341 if (nlk->cb_running) {
2345 /* add reference of module which cb->dump belongs to */
2346 if (!try_module_get(control->module)) {
2347 ret = -EPROTONOSUPPORT;
2352 memset(cb, 0, sizeof(*cb));
2353 cb->dump = control->dump;
2354 cb->done = control->done;
2356 cb->data = control->data;
2357 cb->module = control->module;
2358 cb->min_dump_alloc = control->min_dump_alloc;
2361 nlk2 = nlk_sk(NETLINK_CB(skb).sk);
2362 cb->strict_check = !!(nlk2->flags & NETLINK_F_STRICT_CHK);
2364 if (control->start) {
2365 ret = control->start(cb);
2370 nlk->cb_running = true;
2371 nlk->dump_done_errno = INT_MAX;
2373 mutex_unlock(nlk->cb_mutex);
2375 ret = netlink_dump(sk);
2382 /* We successfully started a dump, by returning -EINTR we
2383 * signal not to send ACK even if it was requested.
2388 module_put(control->module);
2391 mutex_unlock(nlk->cb_mutex);
2396 EXPORT_SYMBOL(__netlink_dump_start);
2398 void netlink_ack(struct sk_buff *in_skb, struct nlmsghdr *nlh, int err,
2399 const struct netlink_ext_ack *extack)
2401 struct sk_buff *skb;
2402 struct nlmsghdr *rep;
2403 struct nlmsgerr *errmsg;
2404 size_t payload = sizeof(*errmsg);
2406 struct netlink_sock *nlk = nlk_sk(NETLINK_CB(in_skb).sk);
2407 unsigned int flags = 0;
2408 bool nlk_has_extack = nlk->flags & NETLINK_F_EXT_ACK;
2410 /* Error messages get the original request appened, unless the user
2411 * requests to cap the error message, and get extra error data if
2414 if (nlk_has_extack && extack && extack->_msg)
2415 tlvlen += nla_total_size(strlen(extack->_msg) + 1);
2417 if (err && !(nlk->flags & NETLINK_F_CAP_ACK))
2418 payload += nlmsg_len(nlh);
2420 flags |= NLM_F_CAPPED;
2421 if (err && nlk_has_extack && extack && extack->bad_attr)
2422 tlvlen += nla_total_size(sizeof(u32));
2423 if (nlk_has_extack && extack && extack->cookie_len)
2424 tlvlen += nla_total_size(extack->cookie_len);
2425 if (err && nlk_has_extack && extack && extack->policy)
2426 tlvlen += netlink_policy_dump_attr_size_estimate(extack->policy);
2429 flags |= NLM_F_ACK_TLVS;
2431 skb = nlmsg_new(payload + tlvlen, GFP_KERNEL);
2433 NETLINK_CB(in_skb).sk->sk_err = ENOBUFS;
2434 sk_error_report(NETLINK_CB(in_skb).sk);
2438 rep = __nlmsg_put(skb, NETLINK_CB(in_skb).portid, nlh->nlmsg_seq,
2439 NLMSG_ERROR, payload, flags);
2440 errmsg = nlmsg_data(rep);
2441 errmsg->error = err;
2442 memcpy(&errmsg->msg, nlh, payload > sizeof(*errmsg) ? nlh->nlmsg_len : sizeof(*nlh));
2444 if (nlk_has_extack && extack) {
2446 WARN_ON(nla_put_string(skb, NLMSGERR_ATTR_MSG,
2449 if (err && extack->bad_attr &&
2450 !WARN_ON((u8 *)extack->bad_attr < in_skb->data ||
2451 (u8 *)extack->bad_attr >= in_skb->data +
2453 WARN_ON(nla_put_u32(skb, NLMSGERR_ATTR_OFFS,
2454 (u8 *)extack->bad_attr -
2456 if (extack->cookie_len)
2457 WARN_ON(nla_put(skb, NLMSGERR_ATTR_COOKIE,
2458 extack->cookie_len, extack->cookie));
2460 netlink_policy_dump_write_attr(skb, extack->policy,
2461 NLMSGERR_ATTR_POLICY);
2464 nlmsg_end(skb, rep);
2466 nlmsg_unicast(in_skb->sk, skb, NETLINK_CB(in_skb).portid);
2468 EXPORT_SYMBOL(netlink_ack);
2470 int netlink_rcv_skb(struct sk_buff *skb, int (*cb)(struct sk_buff *,
2472 struct netlink_ext_ack *))
2474 struct netlink_ext_ack extack;
2475 struct nlmsghdr *nlh;
2478 while (skb->len >= nlmsg_total_size(0)) {
2481 memset(&extack, 0, sizeof(extack));
2482 nlh = nlmsg_hdr(skb);
2485 if (nlh->nlmsg_len < NLMSG_HDRLEN || skb->len < nlh->nlmsg_len)
2488 /* Only requests are handled by the kernel */
2489 if (!(nlh->nlmsg_flags & NLM_F_REQUEST))
2492 /* Skip control messages */
2493 if (nlh->nlmsg_type < NLMSG_MIN_TYPE)
2496 err = cb(skb, nlh, &extack);
2501 if (nlh->nlmsg_flags & NLM_F_ACK || err)
2502 netlink_ack(skb, nlh, err, &extack);
2505 msglen = NLMSG_ALIGN(nlh->nlmsg_len);
2506 if (msglen > skb->len)
2508 skb_pull(skb, msglen);
2513 EXPORT_SYMBOL(netlink_rcv_skb);
2516 * nlmsg_notify - send a notification netlink message
2517 * @sk: netlink socket to use
2518 * @skb: notification message
2519 * @portid: destination netlink portid for reports or 0
2520 * @group: destination multicast group or 0
2521 * @report: 1 to report back, 0 to disable
2522 * @flags: allocation flags
2524 int nlmsg_notify(struct sock *sk, struct sk_buff *skb, u32 portid,
2525 unsigned int group, int report, gfp_t flags)
2530 int exclude_portid = 0;
2533 refcount_inc(&skb->users);
2534 exclude_portid = portid;
2537 /* errors reported via destination sk->sk_err, but propagate
2538 * delivery errors if NETLINK_BROADCAST_ERROR flag is set */
2539 err = nlmsg_multicast(sk, skb, exclude_portid, group, flags);
2547 err2 = nlmsg_unicast(sk, skb, portid);
2554 EXPORT_SYMBOL(nlmsg_notify);
2556 #ifdef CONFIG_PROC_FS
2557 struct nl_seq_iter {
2558 struct seq_net_private p;
2559 struct rhashtable_iter hti;
2563 static void netlink_walk_start(struct nl_seq_iter *iter)
2565 rhashtable_walk_enter(&nl_table[iter->link].hash, &iter->hti);
2566 rhashtable_walk_start(&iter->hti);
2569 static void netlink_walk_stop(struct nl_seq_iter *iter)
2571 rhashtable_walk_stop(&iter->hti);
2572 rhashtable_walk_exit(&iter->hti);
2575 static void *__netlink_seq_next(struct seq_file *seq)
2577 struct nl_seq_iter *iter = seq->private;
2578 struct netlink_sock *nlk;
2582 nlk = rhashtable_walk_next(&iter->hti);
2585 if (PTR_ERR(nlk) == -EAGAIN)
2594 netlink_walk_stop(iter);
2595 if (++iter->link >= MAX_LINKS)
2598 netlink_walk_start(iter);
2600 } while (sock_net(&nlk->sk) != seq_file_net(seq));
2605 static void *netlink_seq_start(struct seq_file *seq, loff_t *posp)
2608 struct nl_seq_iter *iter = seq->private;
2609 void *obj = SEQ_START_TOKEN;
2614 netlink_walk_start(iter);
2616 for (pos = *posp; pos && obj && !IS_ERR(obj); pos--)
2617 obj = __netlink_seq_next(seq);
2622 static void *netlink_seq_next(struct seq_file *seq, void *v, loff_t *pos)
2625 return __netlink_seq_next(seq);
2628 static void netlink_native_seq_stop(struct seq_file *seq, void *v)
2630 struct nl_seq_iter *iter = seq->private;
2632 if (iter->link >= MAX_LINKS)
2635 netlink_walk_stop(iter);
2639 static int netlink_native_seq_show(struct seq_file *seq, void *v)
2641 if (v == SEQ_START_TOKEN) {
2643 "sk Eth Pid Groups "
2644 "Rmem Wmem Dump Locks Drops Inode\n");
2647 struct netlink_sock *nlk = nlk_sk(s);
2649 seq_printf(seq, "%pK %-3d %-10u %08x %-8d %-8d %-5d %-8d %-8u %-8lu\n",
2653 nlk->groups ? (u32)nlk->groups[0] : 0,
2654 sk_rmem_alloc_get(s),
2655 sk_wmem_alloc_get(s),
2657 refcount_read(&s->sk_refcnt),
2658 atomic_read(&s->sk_drops),
2666 #ifdef CONFIG_BPF_SYSCALL
2667 struct bpf_iter__netlink {
2668 __bpf_md_ptr(struct bpf_iter_meta *, meta);
2669 __bpf_md_ptr(struct netlink_sock *, sk);
2672 DEFINE_BPF_ITER_FUNC(netlink, struct bpf_iter_meta *meta, struct netlink_sock *sk)
2674 static int netlink_prog_seq_show(struct bpf_prog *prog,
2675 struct bpf_iter_meta *meta,
2678 struct bpf_iter__netlink ctx;
2680 meta->seq_num--; /* skip SEQ_START_TOKEN */
2682 ctx.sk = nlk_sk((struct sock *)v);
2683 return bpf_iter_run_prog(prog, &ctx);
2686 static int netlink_seq_show(struct seq_file *seq, void *v)
2688 struct bpf_iter_meta meta;
2689 struct bpf_prog *prog;
2692 prog = bpf_iter_get_info(&meta, false);
2694 return netlink_native_seq_show(seq, v);
2696 if (v != SEQ_START_TOKEN)
2697 return netlink_prog_seq_show(prog, &meta, v);
2702 static void netlink_seq_stop(struct seq_file *seq, void *v)
2704 struct bpf_iter_meta meta;
2705 struct bpf_prog *prog;
2709 prog = bpf_iter_get_info(&meta, true);
2711 (void)netlink_prog_seq_show(prog, &meta, v);
2714 netlink_native_seq_stop(seq, v);
2717 static int netlink_seq_show(struct seq_file *seq, void *v)
2719 return netlink_native_seq_show(seq, v);
2722 static void netlink_seq_stop(struct seq_file *seq, void *v)
2724 netlink_native_seq_stop(seq, v);
2728 static const struct seq_operations netlink_seq_ops = {
2729 .start = netlink_seq_start,
2730 .next = netlink_seq_next,
2731 .stop = netlink_seq_stop,
2732 .show = netlink_seq_show,
2736 int netlink_register_notifier(struct notifier_block *nb)
2738 return blocking_notifier_chain_register(&netlink_chain, nb);
2740 EXPORT_SYMBOL(netlink_register_notifier);
2742 int netlink_unregister_notifier(struct notifier_block *nb)
2744 return blocking_notifier_chain_unregister(&netlink_chain, nb);
2746 EXPORT_SYMBOL(netlink_unregister_notifier);
2748 static const struct proto_ops netlink_ops = {
2749 .family = PF_NETLINK,
2750 .owner = THIS_MODULE,
2751 .release = netlink_release,
2752 .bind = netlink_bind,
2753 .connect = netlink_connect,
2754 .socketpair = sock_no_socketpair,
2755 .accept = sock_no_accept,
2756 .getname = netlink_getname,
2757 .poll = datagram_poll,
2758 .ioctl = netlink_ioctl,
2759 .listen = sock_no_listen,
2760 .shutdown = sock_no_shutdown,
2761 .setsockopt = netlink_setsockopt,
2762 .getsockopt = netlink_getsockopt,
2763 .sendmsg = netlink_sendmsg,
2764 .recvmsg = netlink_recvmsg,
2765 .mmap = sock_no_mmap,
2766 .sendpage = sock_no_sendpage,
2769 static const struct net_proto_family netlink_family_ops = {
2770 .family = PF_NETLINK,
2771 .create = netlink_create,
2772 .owner = THIS_MODULE, /* for consistency 8) */
2775 static int __net_init netlink_net_init(struct net *net)
2777 #ifdef CONFIG_PROC_FS
2778 if (!proc_create_net("netlink", 0, net->proc_net, &netlink_seq_ops,
2779 sizeof(struct nl_seq_iter)))
2785 static void __net_exit netlink_net_exit(struct net *net)
2787 #ifdef CONFIG_PROC_FS
2788 remove_proc_entry("netlink", net->proc_net);
2792 static void __init netlink_add_usersock_entry(void)
2794 struct listeners *listeners;
2797 listeners = kzalloc(sizeof(*listeners) + NLGRPSZ(groups), GFP_KERNEL);
2799 panic("netlink_add_usersock_entry: Cannot allocate listeners\n");
2801 netlink_table_grab();
2803 nl_table[NETLINK_USERSOCK].groups = groups;
2804 rcu_assign_pointer(nl_table[NETLINK_USERSOCK].listeners, listeners);
2805 nl_table[NETLINK_USERSOCK].module = THIS_MODULE;
2806 nl_table[NETLINK_USERSOCK].registered = 1;
2807 nl_table[NETLINK_USERSOCK].flags = NL_CFG_F_NONROOT_SEND;
2809 netlink_table_ungrab();
2812 static struct pernet_operations __net_initdata netlink_net_ops = {
2813 .init = netlink_net_init,
2814 .exit = netlink_net_exit,
2817 static inline u32 netlink_hash(const void *data, u32 len, u32 seed)
2819 const struct netlink_sock *nlk = data;
2820 struct netlink_compare_arg arg;
2822 netlink_compare_arg_init(&arg, sock_net(&nlk->sk), nlk->portid);
2823 return jhash2((u32 *)&arg, netlink_compare_arg_len / sizeof(u32), seed);
2826 static const struct rhashtable_params netlink_rhashtable_params = {
2827 .head_offset = offsetof(struct netlink_sock, node),
2828 .key_len = netlink_compare_arg_len,
2829 .obj_hashfn = netlink_hash,
2830 .obj_cmpfn = netlink_compare,
2831 .automatic_shrinking = true,
2834 #if defined(CONFIG_BPF_SYSCALL) && defined(CONFIG_PROC_FS)
2835 BTF_ID_LIST(btf_netlink_sock_id)
2836 BTF_ID(struct, netlink_sock)
2838 static const struct bpf_iter_seq_info netlink_seq_info = {
2839 .seq_ops = &netlink_seq_ops,
2840 .init_seq_private = bpf_iter_init_seq_net,
2841 .fini_seq_private = bpf_iter_fini_seq_net,
2842 .seq_priv_size = sizeof(struct nl_seq_iter),
2845 static struct bpf_iter_reg netlink_reg_info = {
2846 .target = "netlink",
2847 .ctx_arg_info_size = 1,
2849 { offsetof(struct bpf_iter__netlink, sk),
2850 PTR_TO_BTF_ID_OR_NULL },
2852 .seq_info = &netlink_seq_info,
2855 static int __init bpf_iter_register(void)
2857 netlink_reg_info.ctx_arg_info[0].btf_id = *btf_netlink_sock_id;
2858 return bpf_iter_reg_target(&netlink_reg_info);
2862 static int __init netlink_proto_init(void)
2865 int err = proto_register(&netlink_proto, 0);
2870 #if defined(CONFIG_BPF_SYSCALL) && defined(CONFIG_PROC_FS)
2871 err = bpf_iter_register();
2876 BUILD_BUG_ON(sizeof(struct netlink_skb_parms) > sizeof_field(struct sk_buff, cb));
2878 nl_table = kcalloc(MAX_LINKS, sizeof(*nl_table), GFP_KERNEL);
2882 for (i = 0; i < MAX_LINKS; i++) {
2883 if (rhashtable_init(&nl_table[i].hash,
2884 &netlink_rhashtable_params) < 0) {
2886 rhashtable_destroy(&nl_table[i].hash);
2892 netlink_add_usersock_entry();
2894 sock_register(&netlink_family_ops);
2895 register_pernet_subsys(&netlink_net_ops);
2896 register_pernet_subsys(&netlink_tap_net_ops);
2897 /* The netlink device handler may be needed early. */
2902 panic("netlink_init: Cannot allocate nl_table\n");
2905 core_initcall(netlink_proto_init);