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/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)
162 return group ? 1 << (group - 1) : 0;
165 static struct sk_buff *netlink_to_full_skb(const struct sk_buff *skb,
168 unsigned int len = skb_end_offset(skb);
171 new = alloc_skb(len, gfp_mask);
175 NETLINK_CB(new).portid = NETLINK_CB(skb).portid;
176 NETLINK_CB(new).dst_group = NETLINK_CB(skb).dst_group;
177 NETLINK_CB(new).creds = NETLINK_CB(skb).creds;
179 skb_put_data(new, skb->data, len);
183 static unsigned int netlink_tap_net_id;
185 struct netlink_tap_net {
186 struct list_head netlink_tap_all;
187 struct mutex netlink_tap_lock;
190 int netlink_add_tap(struct netlink_tap *nt)
192 struct net *net = dev_net(nt->dev);
193 struct netlink_tap_net *nn = net_generic(net, netlink_tap_net_id);
195 if (unlikely(nt->dev->type != ARPHRD_NETLINK))
198 mutex_lock(&nn->netlink_tap_lock);
199 list_add_rcu(&nt->list, &nn->netlink_tap_all);
200 mutex_unlock(&nn->netlink_tap_lock);
202 __module_get(nt->module);
206 EXPORT_SYMBOL_GPL(netlink_add_tap);
208 static int __netlink_remove_tap(struct netlink_tap *nt)
210 struct net *net = dev_net(nt->dev);
211 struct netlink_tap_net *nn = net_generic(net, netlink_tap_net_id);
213 struct netlink_tap *tmp;
215 mutex_lock(&nn->netlink_tap_lock);
217 list_for_each_entry(tmp, &nn->netlink_tap_all, list) {
219 list_del_rcu(&nt->list);
225 pr_warn("__netlink_remove_tap: %p not found\n", nt);
227 mutex_unlock(&nn->netlink_tap_lock);
230 module_put(nt->module);
232 return found ? 0 : -ENODEV;
235 int netlink_remove_tap(struct netlink_tap *nt)
239 ret = __netlink_remove_tap(nt);
244 EXPORT_SYMBOL_GPL(netlink_remove_tap);
246 static __net_init int netlink_tap_init_net(struct net *net)
248 struct netlink_tap_net *nn = net_generic(net, netlink_tap_net_id);
250 INIT_LIST_HEAD(&nn->netlink_tap_all);
251 mutex_init(&nn->netlink_tap_lock);
255 static struct pernet_operations netlink_tap_net_ops = {
256 .init = netlink_tap_init_net,
257 .id = &netlink_tap_net_id,
258 .size = sizeof(struct netlink_tap_net),
261 static bool netlink_filter_tap(const struct sk_buff *skb)
263 struct sock *sk = skb->sk;
265 /* We take the more conservative approach and
266 * whitelist socket protocols that may pass.
268 switch (sk->sk_protocol) {
270 case NETLINK_USERSOCK:
271 case NETLINK_SOCK_DIAG:
274 case NETLINK_FIB_LOOKUP:
275 case NETLINK_NETFILTER:
276 case NETLINK_GENERIC:
283 static int __netlink_deliver_tap_skb(struct sk_buff *skb,
284 struct net_device *dev)
286 struct sk_buff *nskb;
287 struct sock *sk = skb->sk;
290 if (!net_eq(dev_net(dev), sock_net(sk)))
295 if (is_vmalloc_addr(skb->head))
296 nskb = netlink_to_full_skb(skb, GFP_ATOMIC);
298 nskb = skb_clone(skb, GFP_ATOMIC);
301 nskb->protocol = htons((u16) sk->sk_protocol);
302 nskb->pkt_type = netlink_is_kernel(sk) ?
303 PACKET_KERNEL : PACKET_USER;
304 skb_reset_network_header(nskb);
305 ret = dev_queue_xmit(nskb);
306 if (unlikely(ret > 0))
307 ret = net_xmit_errno(ret);
314 static void __netlink_deliver_tap(struct sk_buff *skb, struct netlink_tap_net *nn)
317 struct netlink_tap *tmp;
319 if (!netlink_filter_tap(skb))
322 list_for_each_entry_rcu(tmp, &nn->netlink_tap_all, list) {
323 ret = __netlink_deliver_tap_skb(skb, tmp->dev);
329 static void netlink_deliver_tap(struct net *net, struct sk_buff *skb)
331 struct netlink_tap_net *nn = net_generic(net, netlink_tap_net_id);
335 if (unlikely(!list_empty(&nn->netlink_tap_all)))
336 __netlink_deliver_tap(skb, nn);
341 static void netlink_deliver_tap_kernel(struct sock *dst, struct sock *src,
344 if (!(netlink_is_kernel(dst) && netlink_is_kernel(src)))
345 netlink_deliver_tap(sock_net(dst), skb);
348 static void netlink_overrun(struct sock *sk)
350 struct netlink_sock *nlk = nlk_sk(sk);
352 if (!(nlk->flags & NETLINK_F_RECV_NO_ENOBUFS)) {
353 if (!test_and_set_bit(NETLINK_S_CONGESTED,
354 &nlk_sk(sk)->state)) {
355 sk->sk_err = ENOBUFS;
359 atomic_inc(&sk->sk_drops);
362 static void netlink_rcv_wake(struct sock *sk)
364 struct netlink_sock *nlk = nlk_sk(sk);
366 if (skb_queue_empty_lockless(&sk->sk_receive_queue))
367 clear_bit(NETLINK_S_CONGESTED, &nlk->state);
368 if (!test_bit(NETLINK_S_CONGESTED, &nlk->state))
369 wake_up_interruptible(&nlk->wait);
372 static void netlink_skb_destructor(struct sk_buff *skb)
374 if (is_vmalloc_addr(skb->head)) {
376 !atomic_dec_return(&(skb_shinfo(skb)->dataref)))
385 static void netlink_skb_set_owner_r(struct sk_buff *skb, struct sock *sk)
387 WARN_ON(skb->sk != NULL);
389 skb->destructor = netlink_skb_destructor;
390 atomic_add(skb->truesize, &sk->sk_rmem_alloc);
391 sk_mem_charge(sk, skb->truesize);
394 static void netlink_sock_destruct(struct sock *sk)
396 struct netlink_sock *nlk = nlk_sk(sk);
398 if (nlk->cb_running) {
400 nlk->cb.done(&nlk->cb);
401 module_put(nlk->cb.module);
402 kfree_skb(nlk->cb.skb);
405 skb_queue_purge(&sk->sk_receive_queue);
407 if (!sock_flag(sk, SOCK_DEAD)) {
408 printk(KERN_ERR "Freeing alive netlink socket %p\n", sk);
412 WARN_ON(atomic_read(&sk->sk_rmem_alloc));
413 WARN_ON(refcount_read(&sk->sk_wmem_alloc));
414 WARN_ON(nlk_sk(sk)->groups);
417 static void netlink_sock_destruct_work(struct work_struct *work)
419 struct netlink_sock *nlk = container_of(work, struct netlink_sock,
425 /* This lock without WQ_FLAG_EXCLUSIVE is good on UP and it is _very_ bad on
426 * SMP. Look, when several writers sleep and reader wakes them up, all but one
427 * immediately hit write lock and grab all the cpus. Exclusive sleep solves
428 * this, _but_ remember, it adds useless work on UP machines.
431 void netlink_table_grab(void)
432 __acquires(nl_table_lock)
436 write_lock_irq(&nl_table_lock);
438 if (atomic_read(&nl_table_users)) {
439 DECLARE_WAITQUEUE(wait, current);
441 add_wait_queue_exclusive(&nl_table_wait, &wait);
443 set_current_state(TASK_UNINTERRUPTIBLE);
444 if (atomic_read(&nl_table_users) == 0)
446 write_unlock_irq(&nl_table_lock);
448 write_lock_irq(&nl_table_lock);
451 __set_current_state(TASK_RUNNING);
452 remove_wait_queue(&nl_table_wait, &wait);
456 void netlink_table_ungrab(void)
457 __releases(nl_table_lock)
459 write_unlock_irq(&nl_table_lock);
460 wake_up(&nl_table_wait);
464 netlink_lock_table(void)
468 /* read_lock() synchronizes us to netlink_table_grab */
470 read_lock_irqsave(&nl_table_lock, flags);
471 atomic_inc(&nl_table_users);
472 read_unlock_irqrestore(&nl_table_lock, flags);
476 netlink_unlock_table(void)
478 if (atomic_dec_and_test(&nl_table_users))
479 wake_up(&nl_table_wait);
482 struct netlink_compare_arg
488 /* Doing sizeof directly may yield 4 extra bytes on 64-bit. */
489 #define netlink_compare_arg_len \
490 (offsetof(struct netlink_compare_arg, portid) + sizeof(u32))
492 static inline int netlink_compare(struct rhashtable_compare_arg *arg,
495 const struct netlink_compare_arg *x = arg->key;
496 const struct netlink_sock *nlk = ptr;
498 return nlk->portid != x->portid ||
499 !net_eq(sock_net(&nlk->sk), read_pnet(&x->pnet));
502 static void netlink_compare_arg_init(struct netlink_compare_arg *arg,
503 struct net *net, u32 portid)
505 memset(arg, 0, sizeof(*arg));
506 write_pnet(&arg->pnet, net);
507 arg->portid = portid;
510 static struct sock *__netlink_lookup(struct netlink_table *table, u32 portid,
513 struct netlink_compare_arg arg;
515 netlink_compare_arg_init(&arg, net, portid);
516 return rhashtable_lookup_fast(&table->hash, &arg,
517 netlink_rhashtable_params);
520 static int __netlink_insert(struct netlink_table *table, struct sock *sk)
522 struct netlink_compare_arg arg;
524 netlink_compare_arg_init(&arg, sock_net(sk), nlk_sk(sk)->portid);
525 return rhashtable_lookup_insert_key(&table->hash, &arg,
527 netlink_rhashtable_params);
530 static struct sock *netlink_lookup(struct net *net, int protocol, u32 portid)
532 struct netlink_table *table = &nl_table[protocol];
536 sk = __netlink_lookup(table, portid, net);
544 static const struct proto_ops netlink_ops;
547 netlink_update_listeners(struct sock *sk)
549 struct netlink_table *tbl = &nl_table[sk->sk_protocol];
552 struct listeners *listeners;
554 listeners = nl_deref_protected(tbl->listeners);
558 for (i = 0; i < NLGRPLONGS(tbl->groups); i++) {
560 sk_for_each_bound(sk, &tbl->mc_list) {
561 if (i < NLGRPLONGS(nlk_sk(sk)->ngroups))
562 mask |= nlk_sk(sk)->groups[i];
564 listeners->masks[i] = mask;
566 /* this function is only called with the netlink table "grabbed", which
567 * makes sure updates are visible before bind or setsockopt return. */
570 static int netlink_insert(struct sock *sk, u32 portid)
572 struct netlink_table *table = &nl_table[sk->sk_protocol];
577 err = nlk_sk(sk)->portid == portid ? 0 : -EBUSY;
578 if (nlk_sk(sk)->bound)
581 nlk_sk(sk)->portid = portid;
584 err = __netlink_insert(table, sk);
586 /* In case the hashtable backend returns with -EBUSY
587 * from here, it must not escape to the caller.
589 if (unlikely(err == -EBUSY))
597 /* We need to ensure that the socket is hashed and visible. */
599 /* Paired with lockless reads from netlink_bind(),
600 * netlink_connect() and netlink_sendmsg().
602 WRITE_ONCE(nlk_sk(sk)->bound, portid);
609 static void netlink_remove(struct sock *sk)
611 struct netlink_table *table;
613 table = &nl_table[sk->sk_protocol];
614 if (!rhashtable_remove_fast(&table->hash, &nlk_sk(sk)->node,
615 netlink_rhashtable_params)) {
616 WARN_ON(refcount_read(&sk->sk_refcnt) == 1);
620 netlink_table_grab();
621 if (nlk_sk(sk)->subscriptions) {
622 __sk_del_bind_node(sk);
623 netlink_update_listeners(sk);
625 if (sk->sk_protocol == NETLINK_GENERIC)
626 atomic_inc(&genl_sk_destructing_cnt);
627 netlink_table_ungrab();
630 static struct proto netlink_proto = {
632 .owner = THIS_MODULE,
633 .obj_size = sizeof(struct netlink_sock),
636 static int __netlink_create(struct net *net, struct socket *sock,
637 struct mutex *cb_mutex, int protocol,
641 struct netlink_sock *nlk;
643 sock->ops = &netlink_ops;
645 sk = sk_alloc(net, PF_NETLINK, GFP_KERNEL, &netlink_proto, kern);
649 sock_init_data(sock, sk);
653 nlk->cb_mutex = cb_mutex;
655 nlk->cb_mutex = &nlk->cb_def_mutex;
656 mutex_init(nlk->cb_mutex);
657 lockdep_set_class_and_name(nlk->cb_mutex,
658 nlk_cb_mutex_keys + protocol,
659 nlk_cb_mutex_key_strings[protocol]);
661 init_waitqueue_head(&nlk->wait);
663 sk->sk_destruct = netlink_sock_destruct;
664 sk->sk_protocol = protocol;
668 static int netlink_create(struct net *net, struct socket *sock, int protocol,
671 struct module *module = NULL;
672 struct mutex *cb_mutex;
673 struct netlink_sock *nlk;
674 int (*bind)(struct net *net, int group);
675 void (*unbind)(struct net *net, int group);
678 sock->state = SS_UNCONNECTED;
680 if (sock->type != SOCK_RAW && sock->type != SOCK_DGRAM)
681 return -ESOCKTNOSUPPORT;
683 if (protocol < 0 || protocol >= MAX_LINKS)
684 return -EPROTONOSUPPORT;
685 protocol = array_index_nospec(protocol, MAX_LINKS);
687 netlink_lock_table();
688 #ifdef CONFIG_MODULES
689 if (!nl_table[protocol].registered) {
690 netlink_unlock_table();
691 request_module("net-pf-%d-proto-%d", PF_NETLINK, protocol);
692 netlink_lock_table();
695 if (nl_table[protocol].registered &&
696 try_module_get(nl_table[protocol].module))
697 module = nl_table[protocol].module;
699 err = -EPROTONOSUPPORT;
700 cb_mutex = nl_table[protocol].cb_mutex;
701 bind = nl_table[protocol].bind;
702 unbind = nl_table[protocol].unbind;
703 netlink_unlock_table();
708 err = __netlink_create(net, sock, cb_mutex, protocol, kern);
712 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();
812 sock_prot_inuse_add(sock_net(sk), &netlink_proto, -1);
813 call_rcu(&nlk->rcu, deferred_put_nlk_sk);
817 static int netlink_autobind(struct socket *sock)
819 struct sock *sk = sock->sk;
820 struct net *net = sock_net(sk);
821 struct netlink_table *table = &nl_table[sk->sk_protocol];
822 s32 portid = task_tgid_vnr(current);
830 ok = !__netlink_lookup(table, portid, net);
833 /* Bind collision, search negative portid values. */
835 /* rover will be in range [S32_MIN, -4097] */
836 rover = S32_MIN + prandom_u32_max(-4096 - S32_MIN);
837 else if (rover >= -4096)
843 err = netlink_insert(sk, portid);
844 if (err == -EADDRINUSE)
847 /* If 2 threads race to autobind, that is fine. */
855 * __netlink_ns_capable - General netlink message capability test
856 * @nsp: NETLINK_CB of the socket buffer holding a netlink command from userspace.
857 * @user_ns: The user namespace of the capability to use
858 * @cap: The capability to use
860 * Test to see if the opener of the socket we received the message
861 * from had when the netlink socket was created and the sender of the
862 * message has the capability @cap in the user namespace @user_ns.
864 bool __netlink_ns_capable(const struct netlink_skb_parms *nsp,
865 struct user_namespace *user_ns, int cap)
867 return ((nsp->flags & NETLINK_SKB_DST) ||
868 file_ns_capable(nsp->sk->sk_socket->file, user_ns, cap)) &&
869 ns_capable(user_ns, cap);
871 EXPORT_SYMBOL(__netlink_ns_capable);
874 * netlink_ns_capable - General netlink message capability test
875 * @skb: socket buffer holding a netlink command from userspace
876 * @user_ns: The user namespace of the capability to use
877 * @cap: The capability to use
879 * Test to see if the opener of the socket we received the message
880 * from had when the netlink socket was created and the sender of the
881 * message has the capability @cap in the user namespace @user_ns.
883 bool netlink_ns_capable(const struct sk_buff *skb,
884 struct user_namespace *user_ns, int cap)
886 return __netlink_ns_capable(&NETLINK_CB(skb), user_ns, cap);
888 EXPORT_SYMBOL(netlink_ns_capable);
891 * netlink_capable - Netlink global message capability test
892 * @skb: socket buffer holding a netlink command from userspace
893 * @cap: The capability to use
895 * Test to see if the opener of the socket we received the message
896 * from had when the netlink socket was created and the sender of the
897 * message has the capability @cap in all user namespaces.
899 bool netlink_capable(const struct sk_buff *skb, int cap)
901 return netlink_ns_capable(skb, &init_user_ns, cap);
903 EXPORT_SYMBOL(netlink_capable);
906 * netlink_net_capable - Netlink network namespace message capability test
907 * @skb: socket buffer holding a netlink command from userspace
908 * @cap: The capability to use
910 * Test to see if the opener of the socket we received the message
911 * from had when the netlink socket was created and the sender of the
912 * message has the capability @cap over the network namespace of
913 * the socket we received the message from.
915 bool netlink_net_capable(const struct sk_buff *skb, int cap)
917 return netlink_ns_capable(skb, sock_net(skb->sk)->user_ns, cap);
919 EXPORT_SYMBOL(netlink_net_capable);
921 static inline int netlink_allowed(const struct socket *sock, unsigned int flag)
923 return (nl_table[sock->sk->sk_protocol].flags & flag) ||
924 ns_capable(sock_net(sock->sk)->user_ns, CAP_NET_ADMIN);
928 netlink_update_subscriptions(struct sock *sk, unsigned int subscriptions)
930 struct netlink_sock *nlk = nlk_sk(sk);
932 if (nlk->subscriptions && !subscriptions)
933 __sk_del_bind_node(sk);
934 else if (!nlk->subscriptions && subscriptions)
935 sk_add_bind_node(sk, &nl_table[sk->sk_protocol].mc_list);
936 nlk->subscriptions = subscriptions;
939 static int netlink_realloc_groups(struct sock *sk)
941 struct netlink_sock *nlk = nlk_sk(sk);
943 unsigned long *new_groups;
946 netlink_table_grab();
948 groups = nl_table[sk->sk_protocol].groups;
949 if (!nl_table[sk->sk_protocol].registered) {
954 if (nlk->ngroups >= groups)
957 new_groups = krealloc(nlk->groups, NLGRPSZ(groups), GFP_ATOMIC);
958 if (new_groups == NULL) {
962 memset((char *)new_groups + NLGRPSZ(nlk->ngroups), 0,
963 NLGRPSZ(groups) - NLGRPSZ(nlk->ngroups));
965 nlk->groups = new_groups;
966 nlk->ngroups = groups;
968 netlink_table_ungrab();
972 static void netlink_undo_bind(int group, long unsigned int groups,
975 struct netlink_sock *nlk = nlk_sk(sk);
978 if (!nlk->netlink_unbind)
981 for (undo = 0; undo < group; undo++)
982 if (test_bit(undo, &groups))
983 nlk->netlink_unbind(sock_net(sk), undo + 1);
986 static int netlink_bind(struct socket *sock, struct sockaddr *addr,
989 struct sock *sk = sock->sk;
990 struct net *net = sock_net(sk);
991 struct netlink_sock *nlk = nlk_sk(sk);
992 struct sockaddr_nl *nladdr = (struct sockaddr_nl *)addr;
994 unsigned long groups;
997 if (addr_len < sizeof(struct sockaddr_nl))
1000 if (nladdr->nl_family != AF_NETLINK)
1002 groups = nladdr->nl_groups;
1004 /* Only superuser is allowed to listen multicasts */
1006 if (!netlink_allowed(sock, NL_CFG_F_NONROOT_RECV))
1008 err = netlink_realloc_groups(sk);
1013 if (nlk->ngroups < BITS_PER_LONG)
1014 groups &= (1UL << nlk->ngroups) - 1;
1016 /* Paired with WRITE_ONCE() in netlink_insert() */
1017 bound = READ_ONCE(nlk->bound);
1019 /* Ensure nlk->portid is up-to-date. */
1022 if (nladdr->nl_pid != nlk->portid)
1026 if (nlk->netlink_bind && groups) {
1029 /* nl_groups is a u32, so cap the maximum groups we can bind */
1030 for (group = 0; group < BITS_PER_TYPE(u32); group++) {
1031 if (!test_bit(group, &groups))
1033 err = nlk->netlink_bind(net, group + 1);
1036 netlink_undo_bind(group, groups, sk);
1041 /* No need for barriers here as we return to user-space without
1042 * using any of the bound attributes.
1044 netlink_lock_table();
1046 err = nladdr->nl_pid ?
1047 netlink_insert(sk, nladdr->nl_pid) :
1048 netlink_autobind(sock);
1050 netlink_undo_bind(BITS_PER_TYPE(u32), groups, sk);
1055 if (!groups && (nlk->groups == NULL || !(u32)nlk->groups[0]))
1057 netlink_unlock_table();
1059 netlink_table_grab();
1060 netlink_update_subscriptions(sk, nlk->subscriptions +
1062 hweight32(nlk->groups[0]));
1063 nlk->groups[0] = (nlk->groups[0] & ~0xffffffffUL) | groups;
1064 netlink_update_listeners(sk);
1065 netlink_table_ungrab();
1070 netlink_unlock_table();
1074 static int netlink_connect(struct socket *sock, struct sockaddr *addr,
1075 int alen, int flags)
1078 struct sock *sk = sock->sk;
1079 struct netlink_sock *nlk = nlk_sk(sk);
1080 struct sockaddr_nl *nladdr = (struct sockaddr_nl *)addr;
1082 if (alen < sizeof(addr->sa_family))
1085 if (addr->sa_family == AF_UNSPEC) {
1086 sk->sk_state = NETLINK_UNCONNECTED;
1087 nlk->dst_portid = 0;
1091 if (addr->sa_family != AF_NETLINK)
1094 if (alen < sizeof(struct sockaddr_nl))
1097 if ((nladdr->nl_groups || nladdr->nl_pid) &&
1098 !netlink_allowed(sock, NL_CFG_F_NONROOT_SEND))
1101 /* No need for barriers here as we return to user-space without
1102 * using any of the bound attributes.
1103 * Paired with WRITE_ONCE() in netlink_insert().
1105 if (!READ_ONCE(nlk->bound))
1106 err = netlink_autobind(sock);
1109 sk->sk_state = NETLINK_CONNECTED;
1110 nlk->dst_portid = nladdr->nl_pid;
1111 nlk->dst_group = ffs(nladdr->nl_groups);
1117 static int netlink_getname(struct socket *sock, struct sockaddr *addr,
1120 struct sock *sk = sock->sk;
1121 struct netlink_sock *nlk = nlk_sk(sk);
1122 DECLARE_SOCKADDR(struct sockaddr_nl *, nladdr, addr);
1124 nladdr->nl_family = AF_NETLINK;
1128 nladdr->nl_pid = nlk->dst_portid;
1129 nladdr->nl_groups = netlink_group_mask(nlk->dst_group);
1131 nladdr->nl_pid = nlk->portid;
1132 netlink_lock_table();
1133 nladdr->nl_groups = nlk->groups ? nlk->groups[0] : 0;
1134 netlink_unlock_table();
1136 return sizeof(*nladdr);
1139 static int netlink_ioctl(struct socket *sock, unsigned int cmd,
1142 /* try to hand this ioctl down to the NIC drivers.
1144 return -ENOIOCTLCMD;
1147 static struct sock *netlink_getsockbyportid(struct sock *ssk, u32 portid)
1150 struct netlink_sock *nlk;
1152 sock = netlink_lookup(sock_net(ssk), ssk->sk_protocol, portid);
1154 return ERR_PTR(-ECONNREFUSED);
1156 /* Don't bother queuing skb if kernel socket has no input function */
1158 if (sock->sk_state == NETLINK_CONNECTED &&
1159 nlk->dst_portid != nlk_sk(ssk)->portid) {
1161 return ERR_PTR(-ECONNREFUSED);
1166 struct sock *netlink_getsockbyfilp(struct file *filp)
1168 struct inode *inode = file_inode(filp);
1171 if (!S_ISSOCK(inode->i_mode))
1172 return ERR_PTR(-ENOTSOCK);
1174 sock = SOCKET_I(inode)->sk;
1175 if (sock->sk_family != AF_NETLINK)
1176 return ERR_PTR(-EINVAL);
1182 static struct sk_buff *netlink_alloc_large_skb(unsigned int size,
1185 struct sk_buff *skb;
1188 if (size <= NLMSG_GOODSIZE || broadcast)
1189 return alloc_skb(size, GFP_KERNEL);
1191 size = SKB_DATA_ALIGN(size) +
1192 SKB_DATA_ALIGN(sizeof(struct skb_shared_info));
1194 data = vmalloc(size);
1198 skb = __build_skb(data, size);
1202 skb->destructor = netlink_skb_destructor;
1208 * Attach a skb to a netlink socket.
1209 * The caller must hold a reference to the destination socket. On error, the
1210 * reference is dropped. The skb is not send to the destination, just all
1211 * all error checks are performed and memory in the queue is reserved.
1213 * < 0: error. skb freed, reference to sock dropped.
1215 * 1: repeat lookup - reference dropped while waiting for socket memory.
1217 int netlink_attachskb(struct sock *sk, struct sk_buff *skb,
1218 long *timeo, struct sock *ssk)
1220 struct netlink_sock *nlk;
1224 if ((atomic_read(&sk->sk_rmem_alloc) > sk->sk_rcvbuf ||
1225 test_bit(NETLINK_S_CONGESTED, &nlk->state))) {
1226 DECLARE_WAITQUEUE(wait, current);
1228 if (!ssk || netlink_is_kernel(ssk))
1229 netlink_overrun(sk);
1235 __set_current_state(TASK_INTERRUPTIBLE);
1236 add_wait_queue(&nlk->wait, &wait);
1238 if ((atomic_read(&sk->sk_rmem_alloc) > sk->sk_rcvbuf ||
1239 test_bit(NETLINK_S_CONGESTED, &nlk->state)) &&
1240 !sock_flag(sk, SOCK_DEAD))
1241 *timeo = schedule_timeout(*timeo);
1243 __set_current_state(TASK_RUNNING);
1244 remove_wait_queue(&nlk->wait, &wait);
1247 if (signal_pending(current)) {
1249 return sock_intr_errno(*timeo);
1253 netlink_skb_set_owner_r(skb, sk);
1257 static int __netlink_sendskb(struct sock *sk, struct sk_buff *skb)
1261 netlink_deliver_tap(sock_net(sk), skb);
1263 skb_queue_tail(&sk->sk_receive_queue, skb);
1264 sk->sk_data_ready(sk);
1268 int netlink_sendskb(struct sock *sk, struct sk_buff *skb)
1270 int len = __netlink_sendskb(sk, skb);
1276 void netlink_detachskb(struct sock *sk, struct sk_buff *skb)
1282 static struct sk_buff *netlink_trim(struct sk_buff *skb, gfp_t allocation)
1286 WARN_ON(skb->sk != NULL);
1287 delta = skb->end - skb->tail;
1288 if (is_vmalloc_addr(skb->head) || delta * 2 < skb->truesize)
1291 if (skb_shared(skb)) {
1292 struct sk_buff *nskb = skb_clone(skb, allocation);
1299 pskb_expand_head(skb, 0, -delta,
1300 (allocation & ~__GFP_DIRECT_RECLAIM) |
1301 __GFP_NOWARN | __GFP_NORETRY);
1305 static int netlink_unicast_kernel(struct sock *sk, struct sk_buff *skb,
1309 struct netlink_sock *nlk = nlk_sk(sk);
1311 ret = -ECONNREFUSED;
1312 if (nlk->netlink_rcv != NULL) {
1314 netlink_skb_set_owner_r(skb, sk);
1315 NETLINK_CB(skb).sk = ssk;
1316 netlink_deliver_tap_kernel(sk, ssk, skb);
1317 nlk->netlink_rcv(skb);
1326 int netlink_unicast(struct sock *ssk, struct sk_buff *skb,
1327 u32 portid, int nonblock)
1333 skb = netlink_trim(skb, gfp_any());
1335 timeo = sock_sndtimeo(ssk, nonblock);
1337 sk = netlink_getsockbyportid(ssk, portid);
1342 if (netlink_is_kernel(sk))
1343 return netlink_unicast_kernel(sk, skb, ssk);
1345 if (sk_filter(sk, skb)) {
1352 err = netlink_attachskb(sk, skb, &timeo, ssk);
1358 return netlink_sendskb(sk, skb);
1360 EXPORT_SYMBOL(netlink_unicast);
1362 int netlink_has_listeners(struct sock *sk, unsigned int group)
1365 struct listeners *listeners;
1367 BUG_ON(!netlink_is_kernel(sk));
1370 listeners = rcu_dereference(nl_table[sk->sk_protocol].listeners);
1372 if (listeners && group - 1 < nl_table[sk->sk_protocol].groups)
1373 res = test_bit(group - 1, listeners->masks);
1379 EXPORT_SYMBOL_GPL(netlink_has_listeners);
1381 bool netlink_strict_get_check(struct sk_buff *skb)
1383 const struct netlink_sock *nlk = nlk_sk(NETLINK_CB(skb).sk);
1385 return nlk->flags & NETLINK_F_STRICT_CHK;
1387 EXPORT_SYMBOL_GPL(netlink_strict_get_check);
1389 static int netlink_broadcast_deliver(struct sock *sk, struct sk_buff *skb)
1391 struct netlink_sock *nlk = nlk_sk(sk);
1393 if (atomic_read(&sk->sk_rmem_alloc) <= sk->sk_rcvbuf &&
1394 !test_bit(NETLINK_S_CONGESTED, &nlk->state)) {
1395 netlink_skb_set_owner_r(skb, sk);
1396 __netlink_sendskb(sk, skb);
1397 return atomic_read(&sk->sk_rmem_alloc) > (sk->sk_rcvbuf >> 1);
1402 struct netlink_broadcast_data {
1403 struct sock *exclude_sk;
1408 int delivery_failure;
1412 struct sk_buff *skb, *skb2;
1415 static void do_one_broadcast(struct sock *sk,
1416 struct netlink_broadcast_data *p)
1418 struct netlink_sock *nlk = nlk_sk(sk);
1421 if (p->exclude_sk == sk)
1424 if (nlk->portid == p->portid || p->group - 1 >= nlk->ngroups ||
1425 !test_bit(p->group - 1, nlk->groups))
1428 if (!net_eq(sock_net(sk), p->net)) {
1429 if (!(nlk->flags & NETLINK_F_LISTEN_ALL_NSID))
1432 if (!peernet_has_id(sock_net(sk), p->net))
1435 if (!file_ns_capable(sk->sk_socket->file, p->net->user_ns,
1441 netlink_overrun(sk);
1446 if (p->skb2 == NULL) {
1447 if (skb_shared(p->skb)) {
1448 p->skb2 = skb_clone(p->skb, p->allocation);
1450 p->skb2 = skb_get(p->skb);
1452 * skb ownership may have been set when
1453 * delivered to a previous socket.
1455 skb_orphan(p->skb2);
1458 if (p->skb2 == NULL) {
1459 netlink_overrun(sk);
1460 /* Clone failed. Notify ALL listeners. */
1462 if (nlk->flags & NETLINK_F_BROADCAST_SEND_ERROR)
1463 p->delivery_failure = 1;
1466 if (sk_filter(sk, p->skb2)) {
1471 NETLINK_CB(p->skb2).nsid = peernet2id(sock_net(sk), p->net);
1472 if (NETLINK_CB(p->skb2).nsid != NETNSA_NSID_NOT_ASSIGNED)
1473 NETLINK_CB(p->skb2).nsid_is_set = true;
1474 val = netlink_broadcast_deliver(sk, p->skb2);
1476 netlink_overrun(sk);
1477 if (nlk->flags & NETLINK_F_BROADCAST_SEND_ERROR)
1478 p->delivery_failure = 1;
1480 p->congested |= val;
1488 int netlink_broadcast(struct sock *ssk, struct sk_buff *skb, u32 portid,
1489 u32 group, gfp_t allocation)
1491 struct net *net = sock_net(ssk);
1492 struct netlink_broadcast_data info;
1495 skb = netlink_trim(skb, allocation);
1497 info.exclude_sk = ssk;
1499 info.portid = portid;
1502 info.delivery_failure = 0;
1505 info.allocation = allocation;
1509 /* While we sleep in clone, do not allow to change socket list */
1511 netlink_lock_table();
1513 sk_for_each_bound(sk, &nl_table[ssk->sk_protocol].mc_list)
1514 do_one_broadcast(sk, &info);
1518 netlink_unlock_table();
1520 if (info.delivery_failure) {
1521 kfree_skb(info.skb2);
1524 consume_skb(info.skb2);
1526 if (info.delivered) {
1527 if (info.congested && gfpflags_allow_blocking(allocation))
1533 EXPORT_SYMBOL(netlink_broadcast);
1535 struct netlink_set_err_data {
1536 struct sock *exclude_sk;
1542 static int do_one_set_err(struct sock *sk, struct netlink_set_err_data *p)
1544 struct netlink_sock *nlk = nlk_sk(sk);
1547 if (sk == p->exclude_sk)
1550 if (!net_eq(sock_net(sk), sock_net(p->exclude_sk)))
1553 if (nlk->portid == p->portid || p->group - 1 >= nlk->ngroups ||
1554 !test_bit(p->group - 1, nlk->groups))
1557 if (p->code == ENOBUFS && nlk->flags & NETLINK_F_RECV_NO_ENOBUFS) {
1562 sk->sk_err = p->code;
1563 sk_error_report(sk);
1569 * netlink_set_err - report error to broadcast listeners
1570 * @ssk: the kernel netlink socket, as returned by netlink_kernel_create()
1571 * @portid: the PORTID of a process that we want to skip (if any)
1572 * @group: the broadcast group that will notice the error
1573 * @code: error code, must be negative (as usual in kernelspace)
1575 * This function returns the number of broadcast listeners that have set the
1576 * NETLINK_NO_ENOBUFS socket option.
1578 int netlink_set_err(struct sock *ssk, u32 portid, u32 group, int code)
1580 struct netlink_set_err_data info;
1584 info.exclude_sk = ssk;
1585 info.portid = portid;
1587 /* sk->sk_err wants a positive error value */
1590 read_lock(&nl_table_lock);
1592 sk_for_each_bound(sk, &nl_table[ssk->sk_protocol].mc_list)
1593 ret += do_one_set_err(sk, &info);
1595 read_unlock(&nl_table_lock);
1598 EXPORT_SYMBOL(netlink_set_err);
1600 /* must be called with netlink table grabbed */
1601 static void netlink_update_socket_mc(struct netlink_sock *nlk,
1605 int old, new = !!is_new, subscriptions;
1607 old = test_bit(group - 1, nlk->groups);
1608 subscriptions = nlk->subscriptions - old + new;
1610 __set_bit(group - 1, nlk->groups);
1612 __clear_bit(group - 1, nlk->groups);
1613 netlink_update_subscriptions(&nlk->sk, subscriptions);
1614 netlink_update_listeners(&nlk->sk);
1617 static int netlink_setsockopt(struct socket *sock, int level, int optname,
1618 sockptr_t optval, unsigned int optlen)
1620 struct sock *sk = sock->sk;
1621 struct netlink_sock *nlk = nlk_sk(sk);
1622 unsigned int val = 0;
1625 if (level != SOL_NETLINK)
1626 return -ENOPROTOOPT;
1628 if (optlen >= sizeof(int) &&
1629 copy_from_sockptr(&val, optval, sizeof(val)))
1633 case NETLINK_PKTINFO:
1635 nlk->flags |= NETLINK_F_RECV_PKTINFO;
1637 nlk->flags &= ~NETLINK_F_RECV_PKTINFO;
1640 case NETLINK_ADD_MEMBERSHIP:
1641 case NETLINK_DROP_MEMBERSHIP: {
1642 if (!netlink_allowed(sock, NL_CFG_F_NONROOT_RECV))
1644 err = netlink_realloc_groups(sk);
1647 if (!val || val - 1 >= nlk->ngroups)
1649 if (optname == NETLINK_ADD_MEMBERSHIP && nlk->netlink_bind) {
1650 err = nlk->netlink_bind(sock_net(sk), val);
1654 netlink_table_grab();
1655 netlink_update_socket_mc(nlk, val,
1656 optname == NETLINK_ADD_MEMBERSHIP);
1657 netlink_table_ungrab();
1658 if (optname == NETLINK_DROP_MEMBERSHIP && nlk->netlink_unbind)
1659 nlk->netlink_unbind(sock_net(sk), val);
1664 case NETLINK_BROADCAST_ERROR:
1666 nlk->flags |= NETLINK_F_BROADCAST_SEND_ERROR;
1668 nlk->flags &= ~NETLINK_F_BROADCAST_SEND_ERROR;
1671 case NETLINK_NO_ENOBUFS:
1673 nlk->flags |= NETLINK_F_RECV_NO_ENOBUFS;
1674 clear_bit(NETLINK_S_CONGESTED, &nlk->state);
1675 wake_up_interruptible(&nlk->wait);
1677 nlk->flags &= ~NETLINK_F_RECV_NO_ENOBUFS;
1681 case NETLINK_LISTEN_ALL_NSID:
1682 if (!ns_capable(sock_net(sk)->user_ns, CAP_NET_BROADCAST))
1686 nlk->flags |= NETLINK_F_LISTEN_ALL_NSID;
1688 nlk->flags &= ~NETLINK_F_LISTEN_ALL_NSID;
1691 case NETLINK_CAP_ACK:
1693 nlk->flags |= NETLINK_F_CAP_ACK;
1695 nlk->flags &= ~NETLINK_F_CAP_ACK;
1698 case NETLINK_EXT_ACK:
1700 nlk->flags |= NETLINK_F_EXT_ACK;
1702 nlk->flags &= ~NETLINK_F_EXT_ACK;
1705 case NETLINK_GET_STRICT_CHK:
1707 nlk->flags |= NETLINK_F_STRICT_CHK;
1709 nlk->flags &= ~NETLINK_F_STRICT_CHK;
1718 static int netlink_getsockopt(struct socket *sock, int level, int optname,
1719 char __user *optval, int __user *optlen)
1721 struct sock *sk = sock->sk;
1722 struct netlink_sock *nlk = nlk_sk(sk);
1725 if (level != SOL_NETLINK)
1726 return -ENOPROTOOPT;
1728 if (get_user(len, optlen))
1734 case NETLINK_PKTINFO:
1735 if (len < sizeof(int))
1738 val = nlk->flags & NETLINK_F_RECV_PKTINFO ? 1 : 0;
1739 if (put_user(len, optlen) ||
1740 put_user(val, optval))
1744 case NETLINK_BROADCAST_ERROR:
1745 if (len < sizeof(int))
1748 val = nlk->flags & NETLINK_F_BROADCAST_SEND_ERROR ? 1 : 0;
1749 if (put_user(len, optlen) ||
1750 put_user(val, optval))
1754 case NETLINK_NO_ENOBUFS:
1755 if (len < sizeof(int))
1758 val = nlk->flags & NETLINK_F_RECV_NO_ENOBUFS ? 1 : 0;
1759 if (put_user(len, optlen) ||
1760 put_user(val, optval))
1764 case NETLINK_LIST_MEMBERSHIPS: {
1765 int pos, idx, shift;
1768 netlink_lock_table();
1769 for (pos = 0; pos * 8 < nlk->ngroups; pos += sizeof(u32)) {
1770 if (len - pos < sizeof(u32))
1773 idx = pos / sizeof(unsigned long);
1774 shift = (pos % sizeof(unsigned long)) * 8;
1775 if (put_user((u32)(nlk->groups[idx] >> shift),
1776 (u32 __user *)(optval + pos))) {
1781 if (put_user(ALIGN(nlk->ngroups / 8, sizeof(u32)), optlen))
1783 netlink_unlock_table();
1786 case NETLINK_CAP_ACK:
1787 if (len < sizeof(int))
1790 val = nlk->flags & NETLINK_F_CAP_ACK ? 1 : 0;
1791 if (put_user(len, optlen) ||
1792 put_user(val, optval))
1796 case NETLINK_EXT_ACK:
1797 if (len < sizeof(int))
1800 val = nlk->flags & NETLINK_F_EXT_ACK ? 1 : 0;
1801 if (put_user(len, optlen) || put_user(val, optval))
1805 case NETLINK_GET_STRICT_CHK:
1806 if (len < sizeof(int))
1809 val = nlk->flags & NETLINK_F_STRICT_CHK ? 1 : 0;
1810 if (put_user(len, optlen) || put_user(val, optval))
1820 static void netlink_cmsg_recv_pktinfo(struct msghdr *msg, struct sk_buff *skb)
1822 struct nl_pktinfo info;
1824 info.group = NETLINK_CB(skb).dst_group;
1825 put_cmsg(msg, SOL_NETLINK, NETLINK_PKTINFO, sizeof(info), &info);
1828 static void netlink_cmsg_listen_all_nsid(struct sock *sk, struct msghdr *msg,
1829 struct sk_buff *skb)
1831 if (!NETLINK_CB(skb).nsid_is_set)
1834 put_cmsg(msg, SOL_NETLINK, NETLINK_LISTEN_ALL_NSID, sizeof(int),
1835 &NETLINK_CB(skb).nsid);
1838 static int netlink_sendmsg(struct socket *sock, struct msghdr *msg, size_t len)
1840 struct sock *sk = sock->sk;
1841 struct netlink_sock *nlk = nlk_sk(sk);
1842 DECLARE_SOCKADDR(struct sockaddr_nl *, addr, msg->msg_name);
1845 struct sk_buff *skb;
1847 struct scm_cookie scm;
1848 u32 netlink_skb_flags = 0;
1850 if (msg->msg_flags & MSG_OOB)
1854 pr_warn_once("Zero length message leads to an empty skb\n");
1858 err = scm_send(sock, msg, &scm, true);
1862 if (msg->msg_namelen) {
1864 if (msg->msg_namelen < sizeof(struct sockaddr_nl))
1866 if (addr->nl_family != AF_NETLINK)
1868 dst_portid = addr->nl_pid;
1869 dst_group = ffs(addr->nl_groups);
1871 if ((dst_group || dst_portid) &&
1872 !netlink_allowed(sock, NL_CFG_F_NONROOT_SEND))
1874 netlink_skb_flags |= NETLINK_SKB_DST;
1876 dst_portid = nlk->dst_portid;
1877 dst_group = nlk->dst_group;
1880 /* Paired with WRITE_ONCE() in netlink_insert() */
1881 if (!READ_ONCE(nlk->bound)) {
1882 err = netlink_autobind(sock);
1886 /* Ensure nlk is hashed and visible. */
1891 if (len > sk->sk_sndbuf - 32)
1894 skb = netlink_alloc_large_skb(len, dst_group);
1898 NETLINK_CB(skb).portid = nlk->portid;
1899 NETLINK_CB(skb).dst_group = dst_group;
1900 NETLINK_CB(skb).creds = scm.creds;
1901 NETLINK_CB(skb).flags = netlink_skb_flags;
1904 if (memcpy_from_msg(skb_put(skb, len), msg, len)) {
1909 err = security_netlink_send(sk, skb);
1916 refcount_inc(&skb->users);
1917 netlink_broadcast(sk, skb, dst_portid, dst_group, GFP_KERNEL);
1919 err = netlink_unicast(sk, skb, dst_portid, msg->msg_flags & MSG_DONTWAIT);
1926 static int netlink_recvmsg(struct socket *sock, struct msghdr *msg, size_t len,
1929 struct scm_cookie scm;
1930 struct sock *sk = sock->sk;
1931 struct netlink_sock *nlk = nlk_sk(sk);
1932 int noblock = flags & MSG_DONTWAIT;
1934 struct sk_buff *skb, *data_skb;
1937 if (flags & MSG_OOB)
1942 skb = skb_recv_datagram(sk, flags, noblock, &err);
1948 #ifdef CONFIG_COMPAT_NETLINK_MESSAGES
1949 if (unlikely(skb_shinfo(skb)->frag_list)) {
1951 * If this skb has a frag_list, then here that means that we
1952 * will have to use the frag_list skb's data for compat tasks
1953 * and the regular skb's data for normal (non-compat) tasks.
1955 * If we need to send the compat skb, assign it to the
1956 * 'data_skb' variable so that it will be used below for data
1957 * copying. We keep 'skb' for everything else, including
1958 * freeing both later.
1960 if (flags & MSG_CMSG_COMPAT)
1961 data_skb = skb_shinfo(skb)->frag_list;
1965 /* Record the max length of recvmsg() calls for future allocations */
1966 nlk->max_recvmsg_len = max(nlk->max_recvmsg_len, len);
1967 nlk->max_recvmsg_len = min_t(size_t, nlk->max_recvmsg_len,
1968 SKB_WITH_OVERHEAD(32768));
1970 copied = data_skb->len;
1972 msg->msg_flags |= MSG_TRUNC;
1976 skb_reset_transport_header(data_skb);
1977 err = skb_copy_datagram_msg(data_skb, 0, msg, copied);
1979 if (msg->msg_name) {
1980 DECLARE_SOCKADDR(struct sockaddr_nl *, addr, msg->msg_name);
1981 addr->nl_family = AF_NETLINK;
1983 addr->nl_pid = NETLINK_CB(skb).portid;
1984 addr->nl_groups = netlink_group_mask(NETLINK_CB(skb).dst_group);
1985 msg->msg_namelen = sizeof(*addr);
1988 if (nlk->flags & NETLINK_F_RECV_PKTINFO)
1989 netlink_cmsg_recv_pktinfo(msg, skb);
1990 if (nlk->flags & NETLINK_F_LISTEN_ALL_NSID)
1991 netlink_cmsg_listen_all_nsid(sk, msg, skb);
1993 memset(&scm, 0, sizeof(scm));
1994 scm.creds = *NETLINK_CREDS(skb);
1995 if (flags & MSG_TRUNC)
1996 copied = data_skb->len;
1998 skb_free_datagram(sk, skb);
2000 if (nlk->cb_running &&
2001 atomic_read(&sk->sk_rmem_alloc) <= sk->sk_rcvbuf / 2) {
2002 ret = netlink_dump(sk);
2005 sk_error_report(sk);
2009 scm_recv(sock, msg, &scm, flags);
2011 netlink_rcv_wake(sk);
2012 return err ? : copied;
2015 static void netlink_data_ready(struct sock *sk)
2021 * We export these functions to other modules. They provide a
2022 * complete set of kernel non-blocking support for message
2027 __netlink_kernel_create(struct net *net, int unit, struct module *module,
2028 struct netlink_kernel_cfg *cfg)
2030 struct socket *sock;
2032 struct netlink_sock *nlk;
2033 struct listeners *listeners = NULL;
2034 struct mutex *cb_mutex = cfg ? cfg->cb_mutex : NULL;
2035 unsigned int groups;
2039 if (unit < 0 || unit >= MAX_LINKS)
2042 if (sock_create_lite(PF_NETLINK, SOCK_DGRAM, unit, &sock))
2045 if (__netlink_create(net, sock, cb_mutex, unit, 1) < 0)
2046 goto out_sock_release_nosk;
2050 if (!cfg || cfg->groups < 32)
2053 groups = cfg->groups;
2055 listeners = kzalloc(sizeof(*listeners) + NLGRPSZ(groups), GFP_KERNEL);
2057 goto out_sock_release;
2059 sk->sk_data_ready = netlink_data_ready;
2060 if (cfg && cfg->input)
2061 nlk_sk(sk)->netlink_rcv = cfg->input;
2063 if (netlink_insert(sk, 0))
2064 goto out_sock_release;
2067 nlk->flags |= NETLINK_F_KERNEL_SOCKET;
2069 netlink_table_grab();
2070 if (!nl_table[unit].registered) {
2071 nl_table[unit].groups = groups;
2072 rcu_assign_pointer(nl_table[unit].listeners, listeners);
2073 nl_table[unit].cb_mutex = cb_mutex;
2074 nl_table[unit].module = module;
2076 nl_table[unit].bind = cfg->bind;
2077 nl_table[unit].unbind = cfg->unbind;
2078 nl_table[unit].flags = cfg->flags;
2080 nl_table[unit].compare = cfg->compare;
2082 nl_table[unit].registered = 1;
2085 nl_table[unit].registered++;
2087 netlink_table_ungrab();
2092 netlink_kernel_release(sk);
2095 out_sock_release_nosk:
2099 EXPORT_SYMBOL(__netlink_kernel_create);
2102 netlink_kernel_release(struct sock *sk)
2104 if (sk == NULL || sk->sk_socket == NULL)
2107 sock_release(sk->sk_socket);
2109 EXPORT_SYMBOL(netlink_kernel_release);
2111 int __netlink_change_ngroups(struct sock *sk, unsigned int groups)
2113 struct listeners *new, *old;
2114 struct netlink_table *tbl = &nl_table[sk->sk_protocol];
2119 if (NLGRPSZ(tbl->groups) < NLGRPSZ(groups)) {
2120 new = kzalloc(sizeof(*new) + NLGRPSZ(groups), GFP_ATOMIC);
2123 old = nl_deref_protected(tbl->listeners);
2124 memcpy(new->masks, old->masks, NLGRPSZ(tbl->groups));
2125 rcu_assign_pointer(tbl->listeners, new);
2127 kfree_rcu(old, rcu);
2129 tbl->groups = groups;
2135 * netlink_change_ngroups - change number of multicast groups
2137 * This changes the number of multicast groups that are available
2138 * on a certain netlink family. Note that it is not possible to
2139 * change the number of groups to below 32. Also note that it does
2140 * not implicitly call netlink_clear_multicast_users() when the
2141 * number of groups is reduced.
2143 * @sk: The kernel netlink socket, as returned by netlink_kernel_create().
2144 * @groups: The new number of groups.
2146 int netlink_change_ngroups(struct sock *sk, unsigned int groups)
2150 netlink_table_grab();
2151 err = __netlink_change_ngroups(sk, groups);
2152 netlink_table_ungrab();
2157 void __netlink_clear_multicast_users(struct sock *ksk, unsigned int group)
2160 struct netlink_table *tbl = &nl_table[ksk->sk_protocol];
2162 sk_for_each_bound(sk, &tbl->mc_list)
2163 netlink_update_socket_mc(nlk_sk(sk), group, 0);
2167 __nlmsg_put(struct sk_buff *skb, u32 portid, u32 seq, int type, int len, int flags)
2169 struct nlmsghdr *nlh;
2170 int size = nlmsg_msg_size(len);
2172 nlh = skb_put(skb, NLMSG_ALIGN(size));
2173 nlh->nlmsg_type = type;
2174 nlh->nlmsg_len = size;
2175 nlh->nlmsg_flags = flags;
2176 nlh->nlmsg_pid = portid;
2177 nlh->nlmsg_seq = seq;
2178 if (!__builtin_constant_p(size) || NLMSG_ALIGN(size) - size != 0)
2179 memset(nlmsg_data(nlh) + len, 0, NLMSG_ALIGN(size) - size);
2182 EXPORT_SYMBOL(__nlmsg_put);
2185 * It looks a bit ugly.
2186 * It would be better to create kernel thread.
2189 static int netlink_dump_done(struct netlink_sock *nlk, struct sk_buff *skb,
2190 struct netlink_callback *cb,
2191 struct netlink_ext_ack *extack)
2193 struct nlmsghdr *nlh;
2195 nlh = nlmsg_put_answer(skb, cb, NLMSG_DONE, sizeof(nlk->dump_done_errno),
2196 NLM_F_MULTI | cb->answer_flags);
2200 nl_dump_check_consistent(cb, nlh);
2201 memcpy(nlmsg_data(nlh), &nlk->dump_done_errno, sizeof(nlk->dump_done_errno));
2203 if (extack->_msg && nlk->flags & NETLINK_F_EXT_ACK) {
2204 nlh->nlmsg_flags |= NLM_F_ACK_TLVS;
2205 if (!nla_put_string(skb, NLMSGERR_ATTR_MSG, extack->_msg))
2206 nlmsg_end(skb, nlh);
2212 static int netlink_dump(struct sock *sk)
2214 struct netlink_sock *nlk = nlk_sk(sk);
2215 struct netlink_ext_ack extack = {};
2216 struct netlink_callback *cb;
2217 struct sk_buff *skb = NULL;
2218 struct module *module;
2223 mutex_lock(nlk->cb_mutex);
2224 if (!nlk->cb_running) {
2229 if (atomic_read(&sk->sk_rmem_alloc) >= sk->sk_rcvbuf)
2232 /* NLMSG_GOODSIZE is small to avoid high order allocations being
2233 * required, but it makes sense to _attempt_ a 16K bytes allocation
2234 * to reduce number of system calls on dump operations, if user
2235 * ever provided a big enough buffer.
2238 alloc_min_size = max_t(int, cb->min_dump_alloc, NLMSG_GOODSIZE);
2240 if (alloc_min_size < nlk->max_recvmsg_len) {
2241 alloc_size = nlk->max_recvmsg_len;
2242 skb = alloc_skb(alloc_size,
2243 (GFP_KERNEL & ~__GFP_DIRECT_RECLAIM) |
2244 __GFP_NOWARN | __GFP_NORETRY);
2247 alloc_size = alloc_min_size;
2248 skb = alloc_skb(alloc_size, GFP_KERNEL);
2253 /* Trim skb to allocated size. User is expected to provide buffer as
2254 * large as max(min_dump_alloc, 16KiB (mac_recvmsg_len capped at
2255 * netlink_recvmsg())). dump will pack as many smaller messages as
2256 * could fit within the allocated skb. skb is typically allocated
2257 * with larger space than required (could be as much as near 2x the
2258 * requested size with align to next power of 2 approach). Allowing
2259 * dump to use the excess space makes it difficult for a user to have a
2260 * reasonable static buffer based on the expected largest dump of a
2261 * single netdev. The outcome is MSG_TRUNC error.
2263 skb_reserve(skb, skb_tailroom(skb) - alloc_size);
2264 netlink_skb_set_owner_r(skb, sk);
2266 if (nlk->dump_done_errno > 0) {
2267 cb->extack = &extack;
2268 nlk->dump_done_errno = cb->dump(skb, cb);
2272 if (nlk->dump_done_errno > 0 ||
2273 skb_tailroom(skb) < nlmsg_total_size(sizeof(nlk->dump_done_errno))) {
2274 mutex_unlock(nlk->cb_mutex);
2276 if (sk_filter(sk, skb))
2279 __netlink_sendskb(sk, skb);
2283 if (netlink_dump_done(nlk, skb, cb, &extack))
2286 #ifdef CONFIG_COMPAT_NETLINK_MESSAGES
2287 /* frag_list skb's data is used for compat tasks
2288 * and the regular skb's data for normal (non-compat) tasks.
2289 * See netlink_recvmsg().
2291 if (unlikely(skb_shinfo(skb)->frag_list)) {
2292 if (netlink_dump_done(nlk, skb_shinfo(skb)->frag_list, cb, &extack))
2297 if (sk_filter(sk, skb))
2300 __netlink_sendskb(sk, skb);
2305 nlk->cb_running = false;
2306 module = cb->module;
2308 mutex_unlock(nlk->cb_mutex);
2314 mutex_unlock(nlk->cb_mutex);
2319 int __netlink_dump_start(struct sock *ssk, struct sk_buff *skb,
2320 const struct nlmsghdr *nlh,
2321 struct netlink_dump_control *control)
2323 struct netlink_sock *nlk, *nlk2;
2324 struct netlink_callback *cb;
2328 refcount_inc(&skb->users);
2330 sk = netlink_lookup(sock_net(ssk), ssk->sk_protocol, NETLINK_CB(skb).portid);
2332 ret = -ECONNREFUSED;
2337 mutex_lock(nlk->cb_mutex);
2338 /* A dump is in progress... */
2339 if (nlk->cb_running) {
2343 /* add reference of module which cb->dump belongs to */
2344 if (!try_module_get(control->module)) {
2345 ret = -EPROTONOSUPPORT;
2350 memset(cb, 0, sizeof(*cb));
2351 cb->dump = control->dump;
2352 cb->done = control->done;
2354 cb->data = control->data;
2355 cb->module = control->module;
2356 cb->min_dump_alloc = control->min_dump_alloc;
2359 nlk2 = nlk_sk(NETLINK_CB(skb).sk);
2360 cb->strict_check = !!(nlk2->flags & NETLINK_F_STRICT_CHK);
2362 if (control->start) {
2363 ret = control->start(cb);
2368 nlk->cb_running = true;
2369 nlk->dump_done_errno = INT_MAX;
2371 mutex_unlock(nlk->cb_mutex);
2373 ret = netlink_dump(sk);
2380 /* We successfully started a dump, by returning -EINTR we
2381 * signal not to send ACK even if it was requested.
2386 module_put(control->module);
2389 mutex_unlock(nlk->cb_mutex);
2394 EXPORT_SYMBOL(__netlink_dump_start);
2396 void netlink_ack(struct sk_buff *in_skb, struct nlmsghdr *nlh, int err,
2397 const struct netlink_ext_ack *extack)
2399 struct sk_buff *skb;
2400 struct nlmsghdr *rep;
2401 struct nlmsgerr *errmsg;
2402 size_t payload = sizeof(*errmsg);
2404 struct netlink_sock *nlk = nlk_sk(NETLINK_CB(in_skb).sk);
2405 unsigned int flags = 0;
2406 bool nlk_has_extack = nlk->flags & NETLINK_F_EXT_ACK;
2408 /* Error messages get the original request appened, unless the user
2409 * requests to cap the error message, and get extra error data if
2412 if (nlk_has_extack && extack && extack->_msg)
2413 tlvlen += nla_total_size(strlen(extack->_msg) + 1);
2415 if (err && !(nlk->flags & NETLINK_F_CAP_ACK))
2416 payload += nlmsg_len(nlh);
2418 flags |= NLM_F_CAPPED;
2419 if (err && nlk_has_extack && extack && extack->bad_attr)
2420 tlvlen += nla_total_size(sizeof(u32));
2421 if (nlk_has_extack && extack && extack->cookie_len)
2422 tlvlen += nla_total_size(extack->cookie_len);
2423 if (err && nlk_has_extack && extack && extack->policy)
2424 tlvlen += netlink_policy_dump_attr_size_estimate(extack->policy);
2427 flags |= NLM_F_ACK_TLVS;
2429 skb = nlmsg_new(payload + tlvlen, GFP_KERNEL);
2431 NETLINK_CB(in_skb).sk->sk_err = ENOBUFS;
2432 sk_error_report(NETLINK_CB(in_skb).sk);
2436 rep = __nlmsg_put(skb, NETLINK_CB(in_skb).portid, nlh->nlmsg_seq,
2437 NLMSG_ERROR, payload, flags);
2438 errmsg = nlmsg_data(rep);
2439 errmsg->error = err;
2440 memcpy(&errmsg->msg, nlh, payload > sizeof(*errmsg) ? nlh->nlmsg_len : sizeof(*nlh));
2442 if (nlk_has_extack && extack) {
2444 WARN_ON(nla_put_string(skb, NLMSGERR_ATTR_MSG,
2447 if (err && extack->bad_attr &&
2448 !WARN_ON((u8 *)extack->bad_attr < in_skb->data ||
2449 (u8 *)extack->bad_attr >= in_skb->data +
2451 WARN_ON(nla_put_u32(skb, NLMSGERR_ATTR_OFFS,
2452 (u8 *)extack->bad_attr -
2454 if (extack->cookie_len)
2455 WARN_ON(nla_put(skb, NLMSGERR_ATTR_COOKIE,
2456 extack->cookie_len, extack->cookie));
2458 netlink_policy_dump_write_attr(skb, extack->policy,
2459 NLMSGERR_ATTR_POLICY);
2462 nlmsg_end(skb, rep);
2464 nlmsg_unicast(in_skb->sk, skb, NETLINK_CB(in_skb).portid);
2466 EXPORT_SYMBOL(netlink_ack);
2468 int netlink_rcv_skb(struct sk_buff *skb, int (*cb)(struct sk_buff *,
2470 struct netlink_ext_ack *))
2472 struct netlink_ext_ack extack;
2473 struct nlmsghdr *nlh;
2476 while (skb->len >= nlmsg_total_size(0)) {
2479 memset(&extack, 0, sizeof(extack));
2480 nlh = nlmsg_hdr(skb);
2483 if (nlh->nlmsg_len < NLMSG_HDRLEN || skb->len < nlh->nlmsg_len)
2486 /* Only requests are handled by the kernel */
2487 if (!(nlh->nlmsg_flags & NLM_F_REQUEST))
2490 /* Skip control messages */
2491 if (nlh->nlmsg_type < NLMSG_MIN_TYPE)
2494 err = cb(skb, nlh, &extack);
2499 if (nlh->nlmsg_flags & NLM_F_ACK || err)
2500 netlink_ack(skb, nlh, err, &extack);
2503 msglen = NLMSG_ALIGN(nlh->nlmsg_len);
2504 if (msglen > skb->len)
2506 skb_pull(skb, msglen);
2511 EXPORT_SYMBOL(netlink_rcv_skb);
2514 * nlmsg_notify - send a notification netlink message
2515 * @sk: netlink socket to use
2516 * @skb: notification message
2517 * @portid: destination netlink portid for reports or 0
2518 * @group: destination multicast group or 0
2519 * @report: 1 to report back, 0 to disable
2520 * @flags: allocation flags
2522 int nlmsg_notify(struct sock *sk, struct sk_buff *skb, u32 portid,
2523 unsigned int group, int report, gfp_t flags)
2528 int exclude_portid = 0;
2531 refcount_inc(&skb->users);
2532 exclude_portid = portid;
2535 /* errors reported via destination sk->sk_err, but propagate
2536 * delivery errors if NETLINK_BROADCAST_ERROR flag is set */
2537 err = nlmsg_multicast(sk, skb, exclude_portid, group, flags);
2545 err2 = nlmsg_unicast(sk, skb, portid);
2552 EXPORT_SYMBOL(nlmsg_notify);
2554 #ifdef CONFIG_PROC_FS
2555 struct nl_seq_iter {
2556 struct seq_net_private p;
2557 struct rhashtable_iter hti;
2561 static void netlink_walk_start(struct nl_seq_iter *iter)
2563 rhashtable_walk_enter(&nl_table[iter->link].hash, &iter->hti);
2564 rhashtable_walk_start(&iter->hti);
2567 static void netlink_walk_stop(struct nl_seq_iter *iter)
2569 rhashtable_walk_stop(&iter->hti);
2570 rhashtable_walk_exit(&iter->hti);
2573 static void *__netlink_seq_next(struct seq_file *seq)
2575 struct nl_seq_iter *iter = seq->private;
2576 struct netlink_sock *nlk;
2580 nlk = rhashtable_walk_next(&iter->hti);
2583 if (PTR_ERR(nlk) == -EAGAIN)
2592 netlink_walk_stop(iter);
2593 if (++iter->link >= MAX_LINKS)
2596 netlink_walk_start(iter);
2598 } while (sock_net(&nlk->sk) != seq_file_net(seq));
2603 static void *netlink_seq_start(struct seq_file *seq, loff_t *posp)
2606 struct nl_seq_iter *iter = seq->private;
2607 void *obj = SEQ_START_TOKEN;
2612 netlink_walk_start(iter);
2614 for (pos = *posp; pos && obj && !IS_ERR(obj); pos--)
2615 obj = __netlink_seq_next(seq);
2620 static void *netlink_seq_next(struct seq_file *seq, void *v, loff_t *pos)
2623 return __netlink_seq_next(seq);
2626 static void netlink_native_seq_stop(struct seq_file *seq, void *v)
2628 struct nl_seq_iter *iter = seq->private;
2630 if (iter->link >= MAX_LINKS)
2633 netlink_walk_stop(iter);
2637 static int netlink_native_seq_show(struct seq_file *seq, void *v)
2639 if (v == SEQ_START_TOKEN) {
2641 "sk Eth Pid Groups "
2642 "Rmem Wmem Dump Locks Drops Inode\n");
2645 struct netlink_sock *nlk = nlk_sk(s);
2647 seq_printf(seq, "%pK %-3d %-10u %08x %-8d %-8d %-5d %-8d %-8u %-8lu\n",
2651 nlk->groups ? (u32)nlk->groups[0] : 0,
2652 sk_rmem_alloc_get(s),
2653 sk_wmem_alloc_get(s),
2655 refcount_read(&s->sk_refcnt),
2656 atomic_read(&s->sk_drops),
2664 #ifdef CONFIG_BPF_SYSCALL
2665 struct bpf_iter__netlink {
2666 __bpf_md_ptr(struct bpf_iter_meta *, meta);
2667 __bpf_md_ptr(struct netlink_sock *, sk);
2670 DEFINE_BPF_ITER_FUNC(netlink, struct bpf_iter_meta *meta, struct netlink_sock *sk)
2672 static int netlink_prog_seq_show(struct bpf_prog *prog,
2673 struct bpf_iter_meta *meta,
2676 struct bpf_iter__netlink ctx;
2678 meta->seq_num--; /* skip SEQ_START_TOKEN */
2680 ctx.sk = nlk_sk((struct sock *)v);
2681 return bpf_iter_run_prog(prog, &ctx);
2684 static int netlink_seq_show(struct seq_file *seq, void *v)
2686 struct bpf_iter_meta meta;
2687 struct bpf_prog *prog;
2690 prog = bpf_iter_get_info(&meta, false);
2692 return netlink_native_seq_show(seq, v);
2694 if (v != SEQ_START_TOKEN)
2695 return netlink_prog_seq_show(prog, &meta, v);
2700 static void netlink_seq_stop(struct seq_file *seq, void *v)
2702 struct bpf_iter_meta meta;
2703 struct bpf_prog *prog;
2707 prog = bpf_iter_get_info(&meta, true);
2709 (void)netlink_prog_seq_show(prog, &meta, v);
2712 netlink_native_seq_stop(seq, v);
2715 static int netlink_seq_show(struct seq_file *seq, void *v)
2717 return netlink_native_seq_show(seq, v);
2720 static void netlink_seq_stop(struct seq_file *seq, void *v)
2722 netlink_native_seq_stop(seq, v);
2726 static const struct seq_operations netlink_seq_ops = {
2727 .start = netlink_seq_start,
2728 .next = netlink_seq_next,
2729 .stop = netlink_seq_stop,
2730 .show = netlink_seq_show,
2734 int netlink_register_notifier(struct notifier_block *nb)
2736 return blocking_notifier_chain_register(&netlink_chain, nb);
2738 EXPORT_SYMBOL(netlink_register_notifier);
2740 int netlink_unregister_notifier(struct notifier_block *nb)
2742 return blocking_notifier_chain_unregister(&netlink_chain, nb);
2744 EXPORT_SYMBOL(netlink_unregister_notifier);
2746 static const struct proto_ops netlink_ops = {
2747 .family = PF_NETLINK,
2748 .owner = THIS_MODULE,
2749 .release = netlink_release,
2750 .bind = netlink_bind,
2751 .connect = netlink_connect,
2752 .socketpair = sock_no_socketpair,
2753 .accept = sock_no_accept,
2754 .getname = netlink_getname,
2755 .poll = datagram_poll,
2756 .ioctl = netlink_ioctl,
2757 .listen = sock_no_listen,
2758 .shutdown = sock_no_shutdown,
2759 .setsockopt = netlink_setsockopt,
2760 .getsockopt = netlink_getsockopt,
2761 .sendmsg = netlink_sendmsg,
2762 .recvmsg = netlink_recvmsg,
2763 .mmap = sock_no_mmap,
2764 .sendpage = sock_no_sendpage,
2767 static const struct net_proto_family netlink_family_ops = {
2768 .family = PF_NETLINK,
2769 .create = netlink_create,
2770 .owner = THIS_MODULE, /* for consistency 8) */
2773 static int __net_init netlink_net_init(struct net *net)
2775 #ifdef CONFIG_PROC_FS
2776 if (!proc_create_net("netlink", 0, net->proc_net, &netlink_seq_ops,
2777 sizeof(struct nl_seq_iter)))
2783 static void __net_exit netlink_net_exit(struct net *net)
2785 #ifdef CONFIG_PROC_FS
2786 remove_proc_entry("netlink", net->proc_net);
2790 static void __init netlink_add_usersock_entry(void)
2792 struct listeners *listeners;
2795 listeners = kzalloc(sizeof(*listeners) + NLGRPSZ(groups), GFP_KERNEL);
2797 panic("netlink_add_usersock_entry: Cannot allocate listeners\n");
2799 netlink_table_grab();
2801 nl_table[NETLINK_USERSOCK].groups = groups;
2802 rcu_assign_pointer(nl_table[NETLINK_USERSOCK].listeners, listeners);
2803 nl_table[NETLINK_USERSOCK].module = THIS_MODULE;
2804 nl_table[NETLINK_USERSOCK].registered = 1;
2805 nl_table[NETLINK_USERSOCK].flags = NL_CFG_F_NONROOT_SEND;
2807 netlink_table_ungrab();
2810 static struct pernet_operations __net_initdata netlink_net_ops = {
2811 .init = netlink_net_init,
2812 .exit = netlink_net_exit,
2815 static inline u32 netlink_hash(const void *data, u32 len, u32 seed)
2817 const struct netlink_sock *nlk = data;
2818 struct netlink_compare_arg arg;
2820 netlink_compare_arg_init(&arg, sock_net(&nlk->sk), nlk->portid);
2821 return jhash2((u32 *)&arg, netlink_compare_arg_len / sizeof(u32), seed);
2824 static const struct rhashtable_params netlink_rhashtable_params = {
2825 .head_offset = offsetof(struct netlink_sock, node),
2826 .key_len = netlink_compare_arg_len,
2827 .obj_hashfn = netlink_hash,
2828 .obj_cmpfn = netlink_compare,
2829 .automatic_shrinking = true,
2832 #if defined(CONFIG_BPF_SYSCALL) && defined(CONFIG_PROC_FS)
2833 BTF_ID_LIST(btf_netlink_sock_id)
2834 BTF_ID(struct, netlink_sock)
2836 static const struct bpf_iter_seq_info netlink_seq_info = {
2837 .seq_ops = &netlink_seq_ops,
2838 .init_seq_private = bpf_iter_init_seq_net,
2839 .fini_seq_private = bpf_iter_fini_seq_net,
2840 .seq_priv_size = sizeof(struct nl_seq_iter),
2843 static struct bpf_iter_reg netlink_reg_info = {
2844 .target = "netlink",
2845 .ctx_arg_info_size = 1,
2847 { offsetof(struct bpf_iter__netlink, sk),
2848 PTR_TO_BTF_ID_OR_NULL },
2850 .seq_info = &netlink_seq_info,
2853 static int __init bpf_iter_register(void)
2855 netlink_reg_info.ctx_arg_info[0].btf_id = *btf_netlink_sock_id;
2856 return bpf_iter_reg_target(&netlink_reg_info);
2860 static int __init netlink_proto_init(void)
2863 int err = proto_register(&netlink_proto, 0);
2868 #if defined(CONFIG_BPF_SYSCALL) && defined(CONFIG_PROC_FS)
2869 err = bpf_iter_register();
2874 BUILD_BUG_ON(sizeof(struct netlink_skb_parms) > sizeof_field(struct sk_buff, cb));
2876 nl_table = kcalloc(MAX_LINKS, sizeof(*nl_table), GFP_KERNEL);
2880 for (i = 0; i < MAX_LINKS; i++) {
2881 if (rhashtable_init(&nl_table[i].hash,
2882 &netlink_rhashtable_params) < 0) {
2884 rhashtable_destroy(&nl_table[i].hash);
2890 netlink_add_usersock_entry();
2892 sock_register(&netlink_family_ops);
2893 register_pernet_subsys(&netlink_net_ops);
2894 register_pernet_subsys(&netlink_tap_net_ops);
2895 /* The netlink device handler may be needed early. */
2900 panic("netlink_init: Cannot allocate nl_table\n");
2903 core_initcall(netlink_proto_init);