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/net_namespace.h>
63 #include <linux/nospec.h>
64 #include <linux/btf_ids.h>
66 #include <net/net_namespace.h>
67 #include <net/netns/generic.h>
70 #include <net/netlink.h>
71 #define CREATE_TRACE_POINTS
72 #include <trace/events/netlink.h>
74 #include "af_netlink.h"
75 #include "genetlink.h"
79 unsigned long masks[];
83 #define NETLINK_S_CONGESTED 0x0
85 static inline int netlink_is_kernel(struct sock *sk)
87 return nlk_test_bit(KERNEL_SOCKET, sk);
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, bool lock_taken);
135 /* nl_table locking explained:
136 * Lookup and traversal are protected with an RCU read-side lock. Insertion
137 * and removal are protected with per bucket lock while using RCU list
138 * modification primitives and may run in parallel to RCU protected lookups.
139 * Destruction of the Netlink socket may only occur *after* nl_table_lock has
140 * been acquired * either during or after the socket has been removed from
141 * the list and after an RCU grace period.
143 DEFINE_RWLOCK(nl_table_lock);
144 EXPORT_SYMBOL_GPL(nl_table_lock);
145 static atomic_t nl_table_users = ATOMIC_INIT(0);
147 #define nl_deref_protected(X) rcu_dereference_protected(X, lockdep_is_held(&nl_table_lock));
149 static BLOCKING_NOTIFIER_HEAD(netlink_chain);
152 static const struct rhashtable_params netlink_rhashtable_params;
154 void do_trace_netlink_extack(const char *msg)
156 trace_netlink_extack(msg);
158 EXPORT_SYMBOL(do_trace_netlink_extack);
160 static inline u32 netlink_group_mask(u32 group)
164 return group ? 1 << (group - 1) : 0;
167 static struct sk_buff *netlink_to_full_skb(const struct sk_buff *skb,
170 unsigned int len = skb->len;
173 new = alloc_skb(len, gfp_mask);
177 NETLINK_CB(new).portid = NETLINK_CB(skb).portid;
178 NETLINK_CB(new).dst_group = NETLINK_CB(skb).dst_group;
179 NETLINK_CB(new).creds = NETLINK_CB(skb).creds;
181 skb_put_data(new, skb->data, len);
185 static unsigned int netlink_tap_net_id;
187 struct netlink_tap_net {
188 struct list_head netlink_tap_all;
189 struct mutex netlink_tap_lock;
192 int netlink_add_tap(struct netlink_tap *nt)
194 struct net *net = dev_net(nt->dev);
195 struct netlink_tap_net *nn = net_generic(net, netlink_tap_net_id);
197 if (unlikely(nt->dev->type != ARPHRD_NETLINK))
200 mutex_lock(&nn->netlink_tap_lock);
201 list_add_rcu(&nt->list, &nn->netlink_tap_all);
202 mutex_unlock(&nn->netlink_tap_lock);
204 __module_get(nt->module);
208 EXPORT_SYMBOL_GPL(netlink_add_tap);
210 static int __netlink_remove_tap(struct netlink_tap *nt)
212 struct net *net = dev_net(nt->dev);
213 struct netlink_tap_net *nn = net_generic(net, netlink_tap_net_id);
215 struct netlink_tap *tmp;
217 mutex_lock(&nn->netlink_tap_lock);
219 list_for_each_entry(tmp, &nn->netlink_tap_all, list) {
221 list_del_rcu(&nt->list);
227 pr_warn("__netlink_remove_tap: %p not found\n", nt);
229 mutex_unlock(&nn->netlink_tap_lock);
232 module_put(nt->module);
234 return found ? 0 : -ENODEV;
237 int netlink_remove_tap(struct netlink_tap *nt)
241 ret = __netlink_remove_tap(nt);
246 EXPORT_SYMBOL_GPL(netlink_remove_tap);
248 static __net_init int netlink_tap_init_net(struct net *net)
250 struct netlink_tap_net *nn = net_generic(net, netlink_tap_net_id);
252 INIT_LIST_HEAD(&nn->netlink_tap_all);
253 mutex_init(&nn->netlink_tap_lock);
257 static struct pernet_operations netlink_tap_net_ops = {
258 .init = netlink_tap_init_net,
259 .id = &netlink_tap_net_id,
260 .size = sizeof(struct netlink_tap_net),
263 static bool netlink_filter_tap(const struct sk_buff *skb)
265 struct sock *sk = skb->sk;
267 /* We take the more conservative approach and
268 * whitelist socket protocols that may pass.
270 switch (sk->sk_protocol) {
272 case NETLINK_USERSOCK:
273 case NETLINK_SOCK_DIAG:
276 case NETLINK_FIB_LOOKUP:
277 case NETLINK_NETFILTER:
278 case NETLINK_GENERIC:
285 static int __netlink_deliver_tap_skb(struct sk_buff *skb,
286 struct net_device *dev)
288 struct sk_buff *nskb;
289 struct sock *sk = skb->sk;
292 if (!net_eq(dev_net(dev), sock_net(sk)))
297 if (is_vmalloc_addr(skb->head))
298 nskb = netlink_to_full_skb(skb, GFP_ATOMIC);
300 nskb = skb_clone(skb, GFP_ATOMIC);
303 nskb->protocol = htons((u16) sk->sk_protocol);
304 nskb->pkt_type = netlink_is_kernel(sk) ?
305 PACKET_KERNEL : PACKET_USER;
306 skb_reset_network_header(nskb);
307 ret = dev_queue_xmit(nskb);
308 if (unlikely(ret > 0))
309 ret = net_xmit_errno(ret);
316 static void __netlink_deliver_tap(struct sk_buff *skb, struct netlink_tap_net *nn)
319 struct netlink_tap *tmp;
321 if (!netlink_filter_tap(skb))
324 list_for_each_entry_rcu(tmp, &nn->netlink_tap_all, list) {
325 ret = __netlink_deliver_tap_skb(skb, tmp->dev);
331 static void netlink_deliver_tap(struct net *net, struct sk_buff *skb)
333 struct netlink_tap_net *nn = net_generic(net, netlink_tap_net_id);
337 if (unlikely(!list_empty(&nn->netlink_tap_all)))
338 __netlink_deliver_tap(skb, nn);
343 static void netlink_deliver_tap_kernel(struct sock *dst, struct sock *src,
346 if (!(netlink_is_kernel(dst) && netlink_is_kernel(src)))
347 netlink_deliver_tap(sock_net(dst), skb);
350 static void netlink_overrun(struct sock *sk)
352 if (!nlk_test_bit(RECV_NO_ENOBUFS, sk)) {
353 if (!test_and_set_bit(NETLINK_S_CONGESTED,
354 &nlk_sk(sk)->state)) {
355 WRITE_ONCE(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)))
377 vfree_atomic(skb->head);
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 skb_queue_purge(&sk->sk_receive_queue);
398 if (!sock_flag(sk, SOCK_DEAD)) {
399 printk(KERN_ERR "Freeing alive netlink socket %p\n", sk);
403 WARN_ON(atomic_read(&sk->sk_rmem_alloc));
404 WARN_ON(refcount_read(&sk->sk_wmem_alloc));
405 WARN_ON(nlk_sk(sk)->groups);
408 /* This lock without WQ_FLAG_EXCLUSIVE is good on UP and it is _very_ bad on
409 * SMP. Look, when several writers sleep and reader wakes them up, all but one
410 * immediately hit write lock and grab all the cpus. Exclusive sleep solves
411 * this, _but_ remember, it adds useless work on UP machines.
414 void netlink_table_grab(void)
415 __acquires(nl_table_lock)
419 write_lock_irq(&nl_table_lock);
421 if (atomic_read(&nl_table_users)) {
422 DECLARE_WAITQUEUE(wait, current);
424 add_wait_queue_exclusive(&nl_table_wait, &wait);
426 set_current_state(TASK_UNINTERRUPTIBLE);
427 if (atomic_read(&nl_table_users) == 0)
429 write_unlock_irq(&nl_table_lock);
431 write_lock_irq(&nl_table_lock);
434 __set_current_state(TASK_RUNNING);
435 remove_wait_queue(&nl_table_wait, &wait);
439 void netlink_table_ungrab(void)
440 __releases(nl_table_lock)
442 write_unlock_irq(&nl_table_lock);
443 wake_up(&nl_table_wait);
447 netlink_lock_table(void)
451 /* read_lock() synchronizes us to netlink_table_grab */
453 read_lock_irqsave(&nl_table_lock, flags);
454 atomic_inc(&nl_table_users);
455 read_unlock_irqrestore(&nl_table_lock, flags);
459 netlink_unlock_table(void)
461 if (atomic_dec_and_test(&nl_table_users))
462 wake_up(&nl_table_wait);
465 struct netlink_compare_arg
471 /* Doing sizeof directly may yield 4 extra bytes on 64-bit. */
472 #define netlink_compare_arg_len \
473 (offsetof(struct netlink_compare_arg, portid) + sizeof(u32))
475 static inline int netlink_compare(struct rhashtable_compare_arg *arg,
478 const struct netlink_compare_arg *x = arg->key;
479 const struct netlink_sock *nlk = ptr;
481 return nlk->portid != x->portid ||
482 !net_eq(sock_net(&nlk->sk), read_pnet(&x->pnet));
485 static void netlink_compare_arg_init(struct netlink_compare_arg *arg,
486 struct net *net, u32 portid)
488 memset(arg, 0, sizeof(*arg));
489 write_pnet(&arg->pnet, net);
490 arg->portid = portid;
493 static struct sock *__netlink_lookup(struct netlink_table *table, u32 portid,
496 struct netlink_compare_arg arg;
498 netlink_compare_arg_init(&arg, net, portid);
499 return rhashtable_lookup_fast(&table->hash, &arg,
500 netlink_rhashtable_params);
503 static int __netlink_insert(struct netlink_table *table, struct sock *sk)
505 struct netlink_compare_arg arg;
507 netlink_compare_arg_init(&arg, sock_net(sk), nlk_sk(sk)->portid);
508 return rhashtable_lookup_insert_key(&table->hash, &arg,
510 netlink_rhashtable_params);
513 static struct sock *netlink_lookup(struct net *net, int protocol, u32 portid)
515 struct netlink_table *table = &nl_table[protocol];
519 sk = __netlink_lookup(table, portid, net);
527 static const struct proto_ops netlink_ops;
530 netlink_update_listeners(struct sock *sk)
532 struct netlink_table *tbl = &nl_table[sk->sk_protocol];
535 struct listeners *listeners;
537 listeners = nl_deref_protected(tbl->listeners);
541 for (i = 0; i < NLGRPLONGS(tbl->groups); i++) {
543 sk_for_each_bound(sk, &tbl->mc_list) {
544 if (i < NLGRPLONGS(nlk_sk(sk)->ngroups))
545 mask |= nlk_sk(sk)->groups[i];
547 listeners->masks[i] = mask;
549 /* this function is only called with the netlink table "grabbed", which
550 * makes sure updates are visible before bind or setsockopt return. */
553 static int netlink_insert(struct sock *sk, u32 portid)
555 struct netlink_table *table = &nl_table[sk->sk_protocol];
560 err = nlk_sk(sk)->portid == portid ? 0 : -EBUSY;
561 if (nlk_sk(sk)->bound)
564 /* portid can be read locklessly from netlink_getname(). */
565 WRITE_ONCE(nlk_sk(sk)->portid, portid);
569 err = __netlink_insert(table, sk);
571 /* In case the hashtable backend returns with -EBUSY
572 * from here, it must not escape to the caller.
574 if (unlikely(err == -EBUSY))
582 /* We need to ensure that the socket is hashed and visible. */
584 /* Paired with lockless reads from netlink_bind(),
585 * netlink_connect() and netlink_sendmsg().
587 WRITE_ONCE(nlk_sk(sk)->bound, portid);
594 static void netlink_remove(struct sock *sk)
596 struct netlink_table *table;
598 table = &nl_table[sk->sk_protocol];
599 if (!rhashtable_remove_fast(&table->hash, &nlk_sk(sk)->node,
600 netlink_rhashtable_params)) {
601 WARN_ON(refcount_read(&sk->sk_refcnt) == 1);
605 netlink_table_grab();
606 if (nlk_sk(sk)->subscriptions) {
607 __sk_del_bind_node(sk);
608 netlink_update_listeners(sk);
610 if (sk->sk_protocol == NETLINK_GENERIC)
611 atomic_inc(&genl_sk_destructing_cnt);
612 netlink_table_ungrab();
615 static struct proto netlink_proto = {
617 .owner = THIS_MODULE,
618 .obj_size = sizeof(struct netlink_sock),
621 static int __netlink_create(struct net *net, struct socket *sock,
622 int protocol, int kern)
625 struct netlink_sock *nlk;
627 sock->ops = &netlink_ops;
629 sk = sk_alloc(net, PF_NETLINK, GFP_KERNEL, &netlink_proto, kern);
633 sock_init_data(sock, sk);
636 mutex_init(&nlk->nl_cb_mutex);
637 lockdep_set_class_and_name(&nlk->nl_cb_mutex,
638 nlk_cb_mutex_keys + protocol,
639 nlk_cb_mutex_key_strings[protocol]);
640 init_waitqueue_head(&nlk->wait);
642 sk->sk_destruct = netlink_sock_destruct;
643 sk->sk_protocol = protocol;
647 static int netlink_create(struct net *net, struct socket *sock, int protocol,
650 struct module *module = NULL;
651 struct netlink_sock *nlk;
652 int (*bind)(struct net *net, int group);
653 void (*unbind)(struct net *net, int group);
654 void (*release)(struct sock *sock, unsigned long *groups);
657 sock->state = SS_UNCONNECTED;
659 if (sock->type != SOCK_RAW && sock->type != SOCK_DGRAM)
660 return -ESOCKTNOSUPPORT;
662 if (protocol < 0 || protocol >= MAX_LINKS)
663 return -EPROTONOSUPPORT;
664 protocol = array_index_nospec(protocol, MAX_LINKS);
666 netlink_lock_table();
667 #ifdef CONFIG_MODULES
668 if (!nl_table[protocol].registered) {
669 netlink_unlock_table();
670 request_module("net-pf-%d-proto-%d", PF_NETLINK, protocol);
671 netlink_lock_table();
674 if (nl_table[protocol].registered &&
675 try_module_get(nl_table[protocol].module))
676 module = nl_table[protocol].module;
678 err = -EPROTONOSUPPORT;
679 bind = nl_table[protocol].bind;
680 unbind = nl_table[protocol].unbind;
681 release = nl_table[protocol].release;
682 netlink_unlock_table();
687 err = __netlink_create(net, sock, protocol, kern);
691 sock_prot_inuse_add(net, &netlink_proto, 1);
693 nlk = nlk_sk(sock->sk);
694 nlk->module = module;
695 nlk->netlink_bind = bind;
696 nlk->netlink_unbind = unbind;
697 nlk->netlink_release = release;
706 static void deferred_put_nlk_sk(struct rcu_head *head)
708 struct netlink_sock *nlk = container_of(head, struct netlink_sock, rcu);
709 struct sock *sk = &nlk->sk;
714 if (!refcount_dec_and_test(&sk->sk_refcnt))
720 static int netlink_release(struct socket *sock)
722 struct sock *sk = sock->sk;
723 struct netlink_sock *nlk;
733 * OK. Socket is unlinked, any packets that arrive now
736 if (nlk->netlink_release)
737 nlk->netlink_release(sk, nlk->groups);
739 /* must not acquire netlink_table_lock in any way again before unbind
740 * and notifying genetlink is done as otherwise it might deadlock
742 if (nlk->netlink_unbind) {
745 for (i = 0; i < nlk->ngroups; i++)
746 if (test_bit(i, nlk->groups))
747 nlk->netlink_unbind(sock_net(sk), i + 1);
749 if (sk->sk_protocol == NETLINK_GENERIC &&
750 atomic_dec_return(&genl_sk_destructing_cnt) == 0)
751 wake_up(&genl_sk_destructing_waitq);
754 wake_up_interruptible_all(&nlk->wait);
756 skb_queue_purge(&sk->sk_write_queue);
758 if (nlk->portid && nlk->bound) {
759 struct netlink_notify n = {
761 .protocol = sk->sk_protocol,
762 .portid = nlk->portid,
764 blocking_notifier_call_chain(&netlink_chain,
765 NETLINK_URELEASE, &n);
768 /* Terminate any outstanding dump */
769 if (nlk->cb_running) {
771 nlk->cb.done(&nlk->cb);
772 module_put(nlk->cb.module);
773 kfree_skb(nlk->cb.skb);
776 module_put(nlk->module);
778 if (netlink_is_kernel(sk)) {
779 netlink_table_grab();
780 BUG_ON(nl_table[sk->sk_protocol].registered == 0);
781 if (--nl_table[sk->sk_protocol].registered == 0) {
782 struct listeners *old;
784 old = nl_deref_protected(nl_table[sk->sk_protocol].listeners);
785 RCU_INIT_POINTER(nl_table[sk->sk_protocol].listeners, NULL);
787 nl_table[sk->sk_protocol].module = NULL;
788 nl_table[sk->sk_protocol].bind = NULL;
789 nl_table[sk->sk_protocol].unbind = NULL;
790 nl_table[sk->sk_protocol].flags = 0;
791 nl_table[sk->sk_protocol].registered = 0;
793 netlink_table_ungrab();
796 sock_prot_inuse_add(sock_net(sk), &netlink_proto, -1);
798 /* Because struct net might disappear soon, do not keep a pointer. */
799 if (!sk->sk_net_refcnt && sock_net(sk) != &init_net) {
800 __netns_tracker_free(sock_net(sk), &sk->ns_tracker, false);
801 /* Because of deferred_put_nlk_sk and use of work queue,
802 * it is possible netns will be freed before this socket.
804 sock_net_set(sk, &init_net);
805 __netns_tracker_alloc(&init_net, &sk->ns_tracker,
808 call_rcu(&nlk->rcu, deferred_put_nlk_sk);
812 static int netlink_autobind(struct socket *sock)
814 struct sock *sk = sock->sk;
815 struct net *net = sock_net(sk);
816 struct netlink_table *table = &nl_table[sk->sk_protocol];
817 s32 portid = task_tgid_vnr(current);
825 ok = !__netlink_lookup(table, portid, net);
828 /* Bind collision, search negative portid values. */
830 /* rover will be in range [S32_MIN, -4097] */
831 rover = S32_MIN + get_random_u32_below(-4096 - S32_MIN);
832 else if (rover >= -4096)
838 err = netlink_insert(sk, portid);
839 if (err == -EADDRINUSE)
842 /* If 2 threads race to autobind, that is fine. */
850 * __netlink_ns_capable - General netlink message capability test
851 * @nsp: NETLINK_CB of the socket buffer holding a netlink command from userspace.
852 * @user_ns: The user namespace of the capability to use
853 * @cap: The capability to use
855 * Test to see if the opener of the socket we received the message
856 * from had when the netlink socket was created and the sender of the
857 * message has the capability @cap in the user namespace @user_ns.
859 bool __netlink_ns_capable(const struct netlink_skb_parms *nsp,
860 struct user_namespace *user_ns, int cap)
862 return ((nsp->flags & NETLINK_SKB_DST) ||
863 file_ns_capable(nsp->sk->sk_socket->file, user_ns, cap)) &&
864 ns_capable(user_ns, cap);
866 EXPORT_SYMBOL(__netlink_ns_capable);
869 * netlink_ns_capable - General netlink message capability test
870 * @skb: socket buffer holding a netlink command from userspace
871 * @user_ns: The user namespace of the capability to use
872 * @cap: The capability to use
874 * Test to see if the opener of the socket we received the message
875 * from had when the netlink socket was created and the sender of the
876 * message has the capability @cap in the user namespace @user_ns.
878 bool netlink_ns_capable(const struct sk_buff *skb,
879 struct user_namespace *user_ns, int cap)
881 return __netlink_ns_capable(&NETLINK_CB(skb), user_ns, cap);
883 EXPORT_SYMBOL(netlink_ns_capable);
886 * netlink_capable - Netlink global message capability test
887 * @skb: socket buffer holding a netlink command from userspace
888 * @cap: The capability to use
890 * Test to see if the opener of the socket we received the message
891 * from had when the netlink socket was created and the sender of the
892 * message has the capability @cap in all user namespaces.
894 bool netlink_capable(const struct sk_buff *skb, int cap)
896 return netlink_ns_capable(skb, &init_user_ns, cap);
898 EXPORT_SYMBOL(netlink_capable);
901 * netlink_net_capable - Netlink network namespace message capability test
902 * @skb: socket buffer holding a netlink command from userspace
903 * @cap: The capability to use
905 * Test to see if the opener of the socket we received the message
906 * from had when the netlink socket was created and the sender of the
907 * message has the capability @cap over the network namespace of
908 * the socket we received the message from.
910 bool netlink_net_capable(const struct sk_buff *skb, int cap)
912 return netlink_ns_capable(skb, sock_net(skb->sk)->user_ns, cap);
914 EXPORT_SYMBOL(netlink_net_capable);
916 static inline int netlink_allowed(const struct socket *sock, unsigned int flag)
918 return (nl_table[sock->sk->sk_protocol].flags & flag) ||
919 ns_capable(sock_net(sock->sk)->user_ns, CAP_NET_ADMIN);
923 netlink_update_subscriptions(struct sock *sk, unsigned int subscriptions)
925 struct netlink_sock *nlk = nlk_sk(sk);
927 if (nlk->subscriptions && !subscriptions)
928 __sk_del_bind_node(sk);
929 else if (!nlk->subscriptions && subscriptions)
930 sk_add_bind_node(sk, &nl_table[sk->sk_protocol].mc_list);
931 nlk->subscriptions = subscriptions;
934 static int netlink_realloc_groups(struct sock *sk)
936 struct netlink_sock *nlk = nlk_sk(sk);
938 unsigned long *new_groups;
941 netlink_table_grab();
943 groups = nl_table[sk->sk_protocol].groups;
944 if (!nl_table[sk->sk_protocol].registered) {
949 if (nlk->ngroups >= groups)
952 new_groups = krealloc(nlk->groups, NLGRPSZ(groups), GFP_ATOMIC);
953 if (new_groups == NULL) {
957 memset((char *)new_groups + NLGRPSZ(nlk->ngroups), 0,
958 NLGRPSZ(groups) - NLGRPSZ(nlk->ngroups));
960 nlk->groups = new_groups;
961 nlk->ngroups = groups;
963 netlink_table_ungrab();
967 static void netlink_undo_bind(int group, long unsigned int groups,
970 struct netlink_sock *nlk = nlk_sk(sk);
973 if (!nlk->netlink_unbind)
976 for (undo = 0; undo < group; undo++)
977 if (test_bit(undo, &groups))
978 nlk->netlink_unbind(sock_net(sk), undo + 1);
981 static int netlink_bind(struct socket *sock, struct sockaddr *addr,
984 struct sock *sk = sock->sk;
985 struct net *net = sock_net(sk);
986 struct netlink_sock *nlk = nlk_sk(sk);
987 struct sockaddr_nl *nladdr = (struct sockaddr_nl *)addr;
989 unsigned long groups;
992 if (addr_len < sizeof(struct sockaddr_nl))
995 if (nladdr->nl_family != AF_NETLINK)
997 groups = nladdr->nl_groups;
999 /* Only superuser is allowed to listen multicasts */
1001 if (!netlink_allowed(sock, NL_CFG_F_NONROOT_RECV))
1003 err = netlink_realloc_groups(sk);
1008 if (nlk->ngroups < BITS_PER_LONG)
1009 groups &= (1UL << nlk->ngroups) - 1;
1011 /* Paired with WRITE_ONCE() in netlink_insert() */
1012 bound = READ_ONCE(nlk->bound);
1014 /* Ensure nlk->portid is up-to-date. */
1017 if (nladdr->nl_pid != nlk->portid)
1021 if (nlk->netlink_bind && groups) {
1024 /* nl_groups is a u32, so cap the maximum groups we can bind */
1025 for (group = 0; group < BITS_PER_TYPE(u32); group++) {
1026 if (!test_bit(group, &groups))
1028 err = nlk->netlink_bind(net, group + 1);
1031 netlink_undo_bind(group, groups, sk);
1036 /* No need for barriers here as we return to user-space without
1037 * using any of the bound attributes.
1039 netlink_lock_table();
1041 err = nladdr->nl_pid ?
1042 netlink_insert(sk, nladdr->nl_pid) :
1043 netlink_autobind(sock);
1045 netlink_undo_bind(BITS_PER_TYPE(u32), groups, sk);
1050 if (!groups && (nlk->groups == NULL || !(u32)nlk->groups[0]))
1052 netlink_unlock_table();
1054 netlink_table_grab();
1055 netlink_update_subscriptions(sk, nlk->subscriptions +
1057 hweight32(nlk->groups[0]));
1058 nlk->groups[0] = (nlk->groups[0] & ~0xffffffffUL) | groups;
1059 netlink_update_listeners(sk);
1060 netlink_table_ungrab();
1065 netlink_unlock_table();
1069 static int netlink_connect(struct socket *sock, struct sockaddr *addr,
1070 int alen, int flags)
1073 struct sock *sk = sock->sk;
1074 struct netlink_sock *nlk = nlk_sk(sk);
1075 struct sockaddr_nl *nladdr = (struct sockaddr_nl *)addr;
1077 if (alen < sizeof(addr->sa_family))
1080 if (addr->sa_family == AF_UNSPEC) {
1081 /* paired with READ_ONCE() in netlink_getsockbyportid() */
1082 WRITE_ONCE(sk->sk_state, NETLINK_UNCONNECTED);
1083 /* dst_portid and dst_group can be read locklessly */
1084 WRITE_ONCE(nlk->dst_portid, 0);
1085 WRITE_ONCE(nlk->dst_group, 0);
1088 if (addr->sa_family != AF_NETLINK)
1091 if (alen < sizeof(struct sockaddr_nl))
1094 if ((nladdr->nl_groups || nladdr->nl_pid) &&
1095 !netlink_allowed(sock, NL_CFG_F_NONROOT_SEND))
1098 /* No need for barriers here as we return to user-space without
1099 * using any of the bound attributes.
1100 * Paired with WRITE_ONCE() in netlink_insert().
1102 if (!READ_ONCE(nlk->bound))
1103 err = netlink_autobind(sock);
1106 /* paired with READ_ONCE() in netlink_getsockbyportid() */
1107 WRITE_ONCE(sk->sk_state, NETLINK_CONNECTED);
1108 /* dst_portid and dst_group can be read locklessly */
1109 WRITE_ONCE(nlk->dst_portid, nladdr->nl_pid);
1110 WRITE_ONCE(nlk->dst_group, ffs(nladdr->nl_groups));
1116 static int netlink_getname(struct socket *sock, struct sockaddr *addr,
1119 struct sock *sk = sock->sk;
1120 struct netlink_sock *nlk = nlk_sk(sk);
1121 DECLARE_SOCKADDR(struct sockaddr_nl *, nladdr, addr);
1123 nladdr->nl_family = AF_NETLINK;
1127 /* Paired with WRITE_ONCE() in netlink_connect() */
1128 nladdr->nl_pid = READ_ONCE(nlk->dst_portid);
1129 nladdr->nl_groups = netlink_group_mask(READ_ONCE(nlk->dst_group));
1131 /* Paired with WRITE_ONCE() in netlink_insert() */
1132 nladdr->nl_pid = READ_ONCE(nlk->portid);
1133 netlink_lock_table();
1134 nladdr->nl_groups = nlk->groups ? nlk->groups[0] : 0;
1135 netlink_unlock_table();
1137 return sizeof(*nladdr);
1140 static int netlink_ioctl(struct socket *sock, unsigned int cmd,
1143 /* try to hand this ioctl down to the NIC drivers.
1145 return -ENOIOCTLCMD;
1148 static struct sock *netlink_getsockbyportid(struct sock *ssk, u32 portid)
1151 struct netlink_sock *nlk;
1153 sock = netlink_lookup(sock_net(ssk), ssk->sk_protocol, portid);
1155 return ERR_PTR(-ECONNREFUSED);
1157 /* Don't bother queuing skb if kernel socket has no input function */
1159 /* dst_portid and sk_state can be changed in netlink_connect() */
1160 if (READ_ONCE(sock->sk_state) == NETLINK_CONNECTED &&
1161 READ_ONCE(nlk->dst_portid) != nlk_sk(ssk)->portid) {
1163 return ERR_PTR(-ECONNREFUSED);
1168 struct sock *netlink_getsockbyfd(int fd)
1171 struct inode *inode;
1175 return ERR_PTR(-EBADF);
1177 inode = file_inode(fd_file(f));
1178 if (!S_ISSOCK(inode->i_mode))
1179 return ERR_PTR(-ENOTSOCK);
1181 sock = SOCKET_I(inode)->sk;
1182 if (sock->sk_family != AF_NETLINK)
1183 return ERR_PTR(-EINVAL);
1189 struct sk_buff *netlink_alloc_large_skb(unsigned int size, int broadcast)
1191 size_t head_size = SKB_HEAD_ALIGN(size);
1192 struct sk_buff *skb;
1195 if (head_size <= PAGE_SIZE || broadcast)
1196 return alloc_skb(size, GFP_KERNEL);
1198 data = kvmalloc(head_size, GFP_KERNEL);
1202 skb = __build_skb(data, head_size);
1205 else if (is_vmalloc_addr(data))
1206 skb->destructor = netlink_skb_destructor;
1212 * Attach a skb to a netlink socket.
1213 * The caller must hold a reference to the destination socket. On error, the
1214 * reference is dropped. The skb is not send to the destination, just all
1215 * all error checks are performed and memory in the queue is reserved.
1217 * < 0: error. skb freed, reference to sock dropped.
1219 * 1: repeat lookup - reference dropped while waiting for socket memory.
1221 int netlink_attachskb(struct sock *sk, struct sk_buff *skb,
1222 long *timeo, struct sock *ssk)
1224 struct netlink_sock *nlk;
1228 if ((atomic_read(&sk->sk_rmem_alloc) > sk->sk_rcvbuf ||
1229 test_bit(NETLINK_S_CONGESTED, &nlk->state))) {
1230 DECLARE_WAITQUEUE(wait, current);
1232 if (!ssk || netlink_is_kernel(ssk))
1233 netlink_overrun(sk);
1239 __set_current_state(TASK_INTERRUPTIBLE);
1240 add_wait_queue(&nlk->wait, &wait);
1242 if ((atomic_read(&sk->sk_rmem_alloc) > sk->sk_rcvbuf ||
1243 test_bit(NETLINK_S_CONGESTED, &nlk->state)) &&
1244 !sock_flag(sk, SOCK_DEAD))
1245 *timeo = schedule_timeout(*timeo);
1247 __set_current_state(TASK_RUNNING);
1248 remove_wait_queue(&nlk->wait, &wait);
1251 if (signal_pending(current)) {
1253 return sock_intr_errno(*timeo);
1257 netlink_skb_set_owner_r(skb, sk);
1261 static int __netlink_sendskb(struct sock *sk, struct sk_buff *skb)
1265 netlink_deliver_tap(sock_net(sk), skb);
1267 skb_queue_tail(&sk->sk_receive_queue, skb);
1268 sk->sk_data_ready(sk);
1272 int netlink_sendskb(struct sock *sk, struct sk_buff *skb)
1274 int len = __netlink_sendskb(sk, skb);
1280 void netlink_detachskb(struct sock *sk, struct sk_buff *skb)
1286 static struct sk_buff *netlink_trim(struct sk_buff *skb, gfp_t allocation)
1290 WARN_ON(skb->sk != NULL);
1291 delta = skb->end - skb->tail;
1292 if (is_vmalloc_addr(skb->head) || delta * 2 < skb->truesize)
1295 if (skb_shared(skb)) {
1296 struct sk_buff *nskb = skb_clone(skb, allocation);
1303 pskb_expand_head(skb, 0, -delta,
1304 (allocation & ~__GFP_DIRECT_RECLAIM) |
1305 __GFP_NOWARN | __GFP_NORETRY);
1309 static int netlink_unicast_kernel(struct sock *sk, struct sk_buff *skb,
1313 struct netlink_sock *nlk = nlk_sk(sk);
1315 ret = -ECONNREFUSED;
1316 if (nlk->netlink_rcv != NULL) {
1318 netlink_skb_set_owner_r(skb, sk);
1319 NETLINK_CB(skb).sk = ssk;
1320 netlink_deliver_tap_kernel(sk, ssk, skb);
1321 nlk->netlink_rcv(skb);
1330 int netlink_unicast(struct sock *ssk, struct sk_buff *skb,
1331 u32 portid, int nonblock)
1337 skb = netlink_trim(skb, gfp_any());
1339 timeo = sock_sndtimeo(ssk, nonblock);
1341 sk = netlink_getsockbyportid(ssk, portid);
1346 if (netlink_is_kernel(sk))
1347 return netlink_unicast_kernel(sk, skb, ssk);
1349 if (sk_filter(sk, skb)) {
1356 err = netlink_attachskb(sk, skb, &timeo, ssk);
1362 return netlink_sendskb(sk, skb);
1364 EXPORT_SYMBOL(netlink_unicast);
1366 int netlink_has_listeners(struct sock *sk, unsigned int group)
1369 struct listeners *listeners;
1371 BUG_ON(!netlink_is_kernel(sk));
1374 listeners = rcu_dereference(nl_table[sk->sk_protocol].listeners);
1376 if (listeners && group - 1 < nl_table[sk->sk_protocol].groups)
1377 res = test_bit(group - 1, listeners->masks);
1383 EXPORT_SYMBOL_GPL(netlink_has_listeners);
1385 bool netlink_strict_get_check(struct sk_buff *skb)
1387 return nlk_test_bit(STRICT_CHK, NETLINK_CB(skb).sk);
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;
1415 int (*tx_filter)(struct sock *dsk, struct sk_buff *skb, void *data);
1419 static void do_one_broadcast(struct sock *sk,
1420 struct netlink_broadcast_data *p)
1422 struct netlink_sock *nlk = nlk_sk(sk);
1425 if (p->exclude_sk == sk)
1428 if (nlk->portid == p->portid || p->group - 1 >= nlk->ngroups ||
1429 !test_bit(p->group - 1, nlk->groups))
1432 if (!net_eq(sock_net(sk), p->net)) {
1433 if (!nlk_test_bit(LISTEN_ALL_NSID, sk))
1436 if (!peernet_has_id(sock_net(sk), p->net))
1439 if (!file_ns_capable(sk->sk_socket->file, p->net->user_ns,
1445 netlink_overrun(sk);
1450 if (p->skb2 == NULL) {
1451 if (skb_shared(p->skb)) {
1452 p->skb2 = skb_clone(p->skb, p->allocation);
1454 p->skb2 = skb_get(p->skb);
1456 * skb ownership may have been set when
1457 * delivered to a previous socket.
1459 skb_orphan(p->skb2);
1462 if (p->skb2 == NULL) {
1463 netlink_overrun(sk);
1464 /* Clone failed. Notify ALL listeners. */
1466 if (nlk_test_bit(BROADCAST_SEND_ERROR, sk))
1467 p->delivery_failure = 1;
1471 if (p->tx_filter && p->tx_filter(sk, p->skb2, p->tx_data)) {
1477 if (sk_filter(sk, p->skb2)) {
1482 NETLINK_CB(p->skb2).nsid = peernet2id(sock_net(sk), p->net);
1483 if (NETLINK_CB(p->skb2).nsid != NETNSA_NSID_NOT_ASSIGNED)
1484 NETLINK_CB(p->skb2).nsid_is_set = true;
1485 val = netlink_broadcast_deliver(sk, p->skb2);
1487 netlink_overrun(sk);
1488 if (nlk_test_bit(BROADCAST_SEND_ERROR, sk))
1489 p->delivery_failure = 1;
1491 p->congested |= val;
1499 int netlink_broadcast_filtered(struct sock *ssk, struct sk_buff *skb,
1501 u32 group, gfp_t allocation,
1502 netlink_filter_fn filter,
1505 struct net *net = sock_net(ssk);
1506 struct netlink_broadcast_data info;
1509 skb = netlink_trim(skb, allocation);
1511 info.exclude_sk = ssk;
1513 info.portid = portid;
1516 info.delivery_failure = 0;
1519 info.allocation = allocation;
1522 info.tx_filter = filter;
1523 info.tx_data = filter_data;
1525 /* While we sleep in clone, do not allow to change socket list */
1527 netlink_lock_table();
1529 sk_for_each_bound(sk, &nl_table[ssk->sk_protocol].mc_list)
1530 do_one_broadcast(sk, &info);
1534 netlink_unlock_table();
1536 if (info.delivery_failure) {
1537 kfree_skb(info.skb2);
1540 consume_skb(info.skb2);
1542 if (info.delivered) {
1543 if (info.congested && gfpflags_allow_blocking(allocation))
1549 EXPORT_SYMBOL(netlink_broadcast_filtered);
1551 int netlink_broadcast(struct sock *ssk, struct sk_buff *skb, u32 portid,
1552 u32 group, gfp_t allocation)
1554 return netlink_broadcast_filtered(ssk, skb, portid, group, allocation,
1557 EXPORT_SYMBOL(netlink_broadcast);
1559 struct netlink_set_err_data {
1560 struct sock *exclude_sk;
1566 static int do_one_set_err(struct sock *sk, struct netlink_set_err_data *p)
1568 struct netlink_sock *nlk = nlk_sk(sk);
1571 if (sk == p->exclude_sk)
1574 if (!net_eq(sock_net(sk), sock_net(p->exclude_sk)))
1577 if (nlk->portid == p->portid || p->group - 1 >= nlk->ngroups ||
1578 !test_bit(p->group - 1, nlk->groups))
1581 if (p->code == ENOBUFS && nlk_test_bit(RECV_NO_ENOBUFS, sk)) {
1586 WRITE_ONCE(sk->sk_err, p->code);
1587 sk_error_report(sk);
1593 * netlink_set_err - report error to broadcast listeners
1594 * @ssk: the kernel netlink socket, as returned by netlink_kernel_create()
1595 * @portid: the PORTID of a process that we want to skip (if any)
1596 * @group: the broadcast group that will notice the error
1597 * @code: error code, must be negative (as usual in kernelspace)
1599 * This function returns the number of broadcast listeners that have set the
1600 * NETLINK_NO_ENOBUFS socket option.
1602 int netlink_set_err(struct sock *ssk, u32 portid, u32 group, int code)
1604 struct netlink_set_err_data info;
1605 unsigned long flags;
1609 info.exclude_sk = ssk;
1610 info.portid = portid;
1612 /* sk->sk_err wants a positive error value */
1615 read_lock_irqsave(&nl_table_lock, flags);
1617 sk_for_each_bound(sk, &nl_table[ssk->sk_protocol].mc_list)
1618 ret += do_one_set_err(sk, &info);
1620 read_unlock_irqrestore(&nl_table_lock, flags);
1623 EXPORT_SYMBOL(netlink_set_err);
1625 /* must be called with netlink table grabbed */
1626 static void netlink_update_socket_mc(struct netlink_sock *nlk,
1630 int old, new = !!is_new, subscriptions;
1632 old = test_bit(group - 1, nlk->groups);
1633 subscriptions = nlk->subscriptions - old + new;
1634 __assign_bit(group - 1, nlk->groups, new);
1635 netlink_update_subscriptions(&nlk->sk, subscriptions);
1636 netlink_update_listeners(&nlk->sk);
1639 static int netlink_setsockopt(struct socket *sock, int level, int optname,
1640 sockptr_t optval, unsigned int optlen)
1642 struct sock *sk = sock->sk;
1643 struct netlink_sock *nlk = nlk_sk(sk);
1644 unsigned int val = 0;
1647 if (level != SOL_NETLINK)
1648 return -ENOPROTOOPT;
1650 if (optlen >= sizeof(int) &&
1651 copy_from_sockptr(&val, optval, sizeof(val)))
1655 case NETLINK_PKTINFO:
1656 nr = NETLINK_F_RECV_PKTINFO;
1658 case NETLINK_ADD_MEMBERSHIP:
1659 case NETLINK_DROP_MEMBERSHIP: {
1662 if (!netlink_allowed(sock, NL_CFG_F_NONROOT_RECV))
1664 err = netlink_realloc_groups(sk);
1667 if (!val || val - 1 >= nlk->ngroups)
1669 if (optname == NETLINK_ADD_MEMBERSHIP && nlk->netlink_bind) {
1670 err = nlk->netlink_bind(sock_net(sk), val);
1674 netlink_table_grab();
1675 netlink_update_socket_mc(nlk, val,
1676 optname == NETLINK_ADD_MEMBERSHIP);
1677 netlink_table_ungrab();
1678 if (optname == NETLINK_DROP_MEMBERSHIP && nlk->netlink_unbind)
1679 nlk->netlink_unbind(sock_net(sk), val);
1683 case NETLINK_BROADCAST_ERROR:
1684 nr = NETLINK_F_BROADCAST_SEND_ERROR;
1686 case NETLINK_NO_ENOBUFS:
1687 assign_bit(NETLINK_F_RECV_NO_ENOBUFS, &nlk->flags, val);
1689 clear_bit(NETLINK_S_CONGESTED, &nlk->state);
1690 wake_up_interruptible(&nlk->wait);
1693 case NETLINK_LISTEN_ALL_NSID:
1694 if (!ns_capable(sock_net(sk)->user_ns, CAP_NET_BROADCAST))
1696 nr = NETLINK_F_LISTEN_ALL_NSID;
1698 case NETLINK_CAP_ACK:
1699 nr = NETLINK_F_CAP_ACK;
1701 case NETLINK_EXT_ACK:
1702 nr = NETLINK_F_EXT_ACK;
1704 case NETLINK_GET_STRICT_CHK:
1705 nr = NETLINK_F_STRICT_CHK;
1708 return -ENOPROTOOPT;
1711 assign_bit(nr, &nlk->flags, val);
1715 static int netlink_getsockopt(struct socket *sock, int level, int optname,
1716 char __user *optval, int __user *optlen)
1718 struct sock *sk = sock->sk;
1719 struct netlink_sock *nlk = nlk_sk(sk);
1723 if (level != SOL_NETLINK)
1724 return -ENOPROTOOPT;
1726 if (get_user(len, optlen))
1732 case NETLINK_PKTINFO:
1733 flag = NETLINK_F_RECV_PKTINFO;
1735 case NETLINK_BROADCAST_ERROR:
1736 flag = NETLINK_F_BROADCAST_SEND_ERROR;
1738 case NETLINK_NO_ENOBUFS:
1739 flag = NETLINK_F_RECV_NO_ENOBUFS;
1741 case NETLINK_LIST_MEMBERSHIPS: {
1742 int pos, idx, shift, err = 0;
1744 netlink_lock_table();
1745 for (pos = 0; pos * 8 < nlk->ngroups; pos += sizeof(u32)) {
1746 if (len - pos < sizeof(u32))
1749 idx = pos / sizeof(unsigned long);
1750 shift = (pos % sizeof(unsigned long)) * 8;
1751 if (put_user((u32)(nlk->groups[idx] >> shift),
1752 (u32 __user *)(optval + pos))) {
1757 if (put_user(ALIGN(BITS_TO_BYTES(nlk->ngroups), sizeof(u32)), optlen))
1759 netlink_unlock_table();
1762 case NETLINK_LISTEN_ALL_NSID:
1763 flag = NETLINK_F_LISTEN_ALL_NSID;
1765 case NETLINK_CAP_ACK:
1766 flag = NETLINK_F_CAP_ACK;
1768 case NETLINK_EXT_ACK:
1769 flag = NETLINK_F_EXT_ACK;
1771 case NETLINK_GET_STRICT_CHK:
1772 flag = NETLINK_F_STRICT_CHK;
1775 return -ENOPROTOOPT;
1778 if (len < sizeof(int))
1782 val = test_bit(flag, &nlk->flags);
1784 if (put_user(len, optlen) ||
1785 copy_to_user(optval, &val, len))
1791 static void netlink_cmsg_recv_pktinfo(struct msghdr *msg, struct sk_buff *skb)
1793 struct nl_pktinfo info;
1795 info.group = NETLINK_CB(skb).dst_group;
1796 put_cmsg(msg, SOL_NETLINK, NETLINK_PKTINFO, sizeof(info), &info);
1799 static void netlink_cmsg_listen_all_nsid(struct sock *sk, struct msghdr *msg,
1800 struct sk_buff *skb)
1802 if (!NETLINK_CB(skb).nsid_is_set)
1805 put_cmsg(msg, SOL_NETLINK, NETLINK_LISTEN_ALL_NSID, sizeof(int),
1806 &NETLINK_CB(skb).nsid);
1809 static int netlink_sendmsg(struct socket *sock, struct msghdr *msg, size_t len)
1811 struct sock *sk = sock->sk;
1812 struct netlink_sock *nlk = nlk_sk(sk);
1813 DECLARE_SOCKADDR(struct sockaddr_nl *, addr, msg->msg_name);
1816 struct sk_buff *skb;
1818 struct scm_cookie scm;
1819 u32 netlink_skb_flags = 0;
1821 if (msg->msg_flags & MSG_OOB)
1825 pr_warn_once("Zero length message leads to an empty skb\n");
1829 err = scm_send(sock, msg, &scm, true);
1833 if (msg->msg_namelen) {
1835 if (msg->msg_namelen < sizeof(struct sockaddr_nl))
1837 if (addr->nl_family != AF_NETLINK)
1839 dst_portid = addr->nl_pid;
1840 dst_group = ffs(addr->nl_groups);
1842 if ((dst_group || dst_portid) &&
1843 !netlink_allowed(sock, NL_CFG_F_NONROOT_SEND))
1845 netlink_skb_flags |= NETLINK_SKB_DST;
1847 /* Paired with WRITE_ONCE() in netlink_connect() */
1848 dst_portid = READ_ONCE(nlk->dst_portid);
1849 dst_group = READ_ONCE(nlk->dst_group);
1852 /* Paired with WRITE_ONCE() in netlink_insert() */
1853 if (!READ_ONCE(nlk->bound)) {
1854 err = netlink_autobind(sock);
1858 /* Ensure nlk is hashed and visible. */
1863 if (len > sk->sk_sndbuf - 32)
1866 skb = netlink_alloc_large_skb(len, dst_group);
1870 NETLINK_CB(skb).portid = nlk->portid;
1871 NETLINK_CB(skb).dst_group = dst_group;
1872 NETLINK_CB(skb).creds = scm.creds;
1873 NETLINK_CB(skb).flags = netlink_skb_flags;
1876 if (memcpy_from_msg(skb_put(skb, len), msg, len)) {
1881 err = security_netlink_send(sk, skb);
1888 refcount_inc(&skb->users);
1889 netlink_broadcast(sk, skb, dst_portid, dst_group, GFP_KERNEL);
1891 err = netlink_unicast(sk, skb, dst_portid, msg->msg_flags & MSG_DONTWAIT);
1898 static int netlink_recvmsg(struct socket *sock, struct msghdr *msg, size_t len,
1901 struct scm_cookie scm;
1902 struct sock *sk = sock->sk;
1903 struct netlink_sock *nlk = nlk_sk(sk);
1904 size_t copied, max_recvmsg_len;
1905 struct sk_buff *skb, *data_skb;
1908 if (flags & MSG_OOB)
1913 skb = skb_recv_datagram(sk, flags, &err);
1919 #ifdef CONFIG_COMPAT_NETLINK_MESSAGES
1920 if (unlikely(skb_shinfo(skb)->frag_list)) {
1922 * If this skb has a frag_list, then here that means that we
1923 * will have to use the frag_list skb's data for compat tasks
1924 * and the regular skb's data for normal (non-compat) tasks.
1926 * If we need to send the compat skb, assign it to the
1927 * 'data_skb' variable so that it will be used below for data
1928 * copying. We keep 'skb' for everything else, including
1929 * freeing both later.
1931 if (flags & MSG_CMSG_COMPAT)
1932 data_skb = skb_shinfo(skb)->frag_list;
1936 /* Record the max length of recvmsg() calls for future allocations */
1937 max_recvmsg_len = max(READ_ONCE(nlk->max_recvmsg_len), len);
1938 max_recvmsg_len = min_t(size_t, max_recvmsg_len,
1939 SKB_WITH_OVERHEAD(32768));
1940 WRITE_ONCE(nlk->max_recvmsg_len, max_recvmsg_len);
1942 copied = data_skb->len;
1944 msg->msg_flags |= MSG_TRUNC;
1948 err = skb_copy_datagram_msg(data_skb, 0, msg, copied);
1950 if (msg->msg_name) {
1951 DECLARE_SOCKADDR(struct sockaddr_nl *, addr, msg->msg_name);
1952 addr->nl_family = AF_NETLINK;
1954 addr->nl_pid = NETLINK_CB(skb).portid;
1955 addr->nl_groups = netlink_group_mask(NETLINK_CB(skb).dst_group);
1956 msg->msg_namelen = sizeof(*addr);
1959 if (nlk_test_bit(RECV_PKTINFO, sk))
1960 netlink_cmsg_recv_pktinfo(msg, skb);
1961 if (nlk_test_bit(LISTEN_ALL_NSID, sk))
1962 netlink_cmsg_listen_all_nsid(sk, msg, skb);
1964 memset(&scm, 0, sizeof(scm));
1965 scm.creds = *NETLINK_CREDS(skb);
1966 if (flags & MSG_TRUNC)
1967 copied = data_skb->len;
1969 skb_free_datagram(sk, skb);
1971 if (READ_ONCE(nlk->cb_running) &&
1972 atomic_read(&sk->sk_rmem_alloc) <= sk->sk_rcvbuf / 2) {
1973 ret = netlink_dump(sk, false);
1975 WRITE_ONCE(sk->sk_err, -ret);
1976 sk_error_report(sk);
1980 scm_recv(sock, msg, &scm, flags);
1982 netlink_rcv_wake(sk);
1983 return err ? : copied;
1986 static void netlink_data_ready(struct sock *sk)
1992 * We export these functions to other modules. They provide a
1993 * complete set of kernel non-blocking support for message
1998 __netlink_kernel_create(struct net *net, int unit, struct module *module,
1999 struct netlink_kernel_cfg *cfg)
2001 struct socket *sock;
2003 struct netlink_sock *nlk;
2004 struct listeners *listeners = NULL;
2005 unsigned int groups;
2009 if (unit < 0 || unit >= MAX_LINKS)
2012 if (sock_create_lite(PF_NETLINK, SOCK_DGRAM, unit, &sock))
2015 if (__netlink_create(net, sock, unit, 1) < 0)
2016 goto out_sock_release_nosk;
2020 if (!cfg || cfg->groups < 32)
2023 groups = cfg->groups;
2025 listeners = kzalloc(sizeof(*listeners) + NLGRPSZ(groups), GFP_KERNEL);
2027 goto out_sock_release;
2029 sk->sk_data_ready = netlink_data_ready;
2030 if (cfg && cfg->input)
2031 nlk_sk(sk)->netlink_rcv = cfg->input;
2033 if (netlink_insert(sk, 0))
2034 goto out_sock_release;
2037 set_bit(NETLINK_F_KERNEL_SOCKET, &nlk->flags);
2039 netlink_table_grab();
2040 if (!nl_table[unit].registered) {
2041 nl_table[unit].groups = groups;
2042 rcu_assign_pointer(nl_table[unit].listeners, listeners);
2043 nl_table[unit].module = module;
2045 nl_table[unit].bind = cfg->bind;
2046 nl_table[unit].unbind = cfg->unbind;
2047 nl_table[unit].release = cfg->release;
2048 nl_table[unit].flags = cfg->flags;
2050 nl_table[unit].registered = 1;
2053 nl_table[unit].registered++;
2055 netlink_table_ungrab();
2060 netlink_kernel_release(sk);
2063 out_sock_release_nosk:
2067 EXPORT_SYMBOL(__netlink_kernel_create);
2070 netlink_kernel_release(struct sock *sk)
2072 if (sk == NULL || sk->sk_socket == NULL)
2075 sock_release(sk->sk_socket);
2077 EXPORT_SYMBOL(netlink_kernel_release);
2079 int __netlink_change_ngroups(struct sock *sk, unsigned int groups)
2081 struct listeners *new, *old;
2082 struct netlink_table *tbl = &nl_table[sk->sk_protocol];
2087 if (NLGRPSZ(tbl->groups) < NLGRPSZ(groups)) {
2088 new = kzalloc(sizeof(*new) + NLGRPSZ(groups), GFP_ATOMIC);
2091 old = nl_deref_protected(tbl->listeners);
2092 memcpy(new->masks, old->masks, NLGRPSZ(tbl->groups));
2093 rcu_assign_pointer(tbl->listeners, new);
2095 kfree_rcu(old, rcu);
2097 tbl->groups = groups;
2103 * netlink_change_ngroups - change number of multicast groups
2105 * This changes the number of multicast groups that are available
2106 * on a certain netlink family. Note that it is not possible to
2107 * change the number of groups to below 32. Also note that it does
2108 * not implicitly call netlink_clear_multicast_users() when the
2109 * number of groups is reduced.
2111 * @sk: The kernel netlink socket, as returned by netlink_kernel_create().
2112 * @groups: The new number of groups.
2114 int netlink_change_ngroups(struct sock *sk, unsigned int groups)
2118 netlink_table_grab();
2119 err = __netlink_change_ngroups(sk, groups);
2120 netlink_table_ungrab();
2125 void __netlink_clear_multicast_users(struct sock *ksk, unsigned int group)
2128 struct netlink_table *tbl = &nl_table[ksk->sk_protocol];
2129 struct hlist_node *tmp;
2131 sk_for_each_bound_safe(sk, tmp, &tbl->mc_list)
2132 netlink_update_socket_mc(nlk_sk(sk), group, 0);
2136 __nlmsg_put(struct sk_buff *skb, u32 portid, u32 seq, int type, int len, int flags)
2138 struct nlmsghdr *nlh;
2139 int size = nlmsg_msg_size(len);
2141 nlh = skb_put(skb, NLMSG_ALIGN(size));
2142 nlh->nlmsg_type = type;
2143 nlh->nlmsg_len = size;
2144 nlh->nlmsg_flags = flags;
2145 nlh->nlmsg_pid = portid;
2146 nlh->nlmsg_seq = seq;
2147 if (!__builtin_constant_p(size) || NLMSG_ALIGN(size) - size != 0)
2148 memset(nlmsg_data(nlh) + len, 0, NLMSG_ALIGN(size) - size);
2151 EXPORT_SYMBOL(__nlmsg_put);
2154 netlink_ack_tlv_len(struct netlink_sock *nlk, int err,
2155 const struct netlink_ext_ack *extack)
2159 if (!extack || !test_bit(NETLINK_F_EXT_ACK, &nlk->flags))
2164 tlvlen += nla_total_size(strlen(extack->_msg) + 1);
2165 if (extack->cookie_len)
2166 tlvlen += nla_total_size(extack->cookie_len);
2168 /* Following attributes are only reported as error (not warning) */
2172 if (extack->bad_attr)
2173 tlvlen += nla_total_size(sizeof(u32));
2175 tlvlen += netlink_policy_dump_attr_size_estimate(extack->policy);
2176 if (extack->miss_type)
2177 tlvlen += nla_total_size(sizeof(u32));
2178 if (extack->miss_nest)
2179 tlvlen += nla_total_size(sizeof(u32));
2184 static bool nlmsg_check_in_payload(const struct nlmsghdr *nlh, const void *addr)
2186 return !WARN_ON(addr < nlmsg_data(nlh) ||
2187 addr - (const void *) nlh >= nlh->nlmsg_len);
2191 netlink_ack_tlv_fill(struct sk_buff *skb, const struct nlmsghdr *nlh, int err,
2192 const struct netlink_ext_ack *extack)
2195 WARN_ON(nla_put_string(skb, NLMSGERR_ATTR_MSG, extack->_msg));
2196 if (extack->cookie_len)
2197 WARN_ON(nla_put(skb, NLMSGERR_ATTR_COOKIE,
2198 extack->cookie_len, extack->cookie));
2203 if (extack->bad_attr && nlmsg_check_in_payload(nlh, extack->bad_attr))
2204 WARN_ON(nla_put_u32(skb, NLMSGERR_ATTR_OFFS,
2205 (u8 *)extack->bad_attr - (const u8 *)nlh));
2207 netlink_policy_dump_write_attr(skb, extack->policy,
2208 NLMSGERR_ATTR_POLICY);
2209 if (extack->miss_type)
2210 WARN_ON(nla_put_u32(skb, NLMSGERR_ATTR_MISS_TYPE,
2211 extack->miss_type));
2212 if (extack->miss_nest && nlmsg_check_in_payload(nlh, extack->miss_nest))
2213 WARN_ON(nla_put_u32(skb, NLMSGERR_ATTR_MISS_NEST,
2214 (u8 *)extack->miss_nest - (const u8 *)nlh));
2218 * It looks a bit ugly.
2219 * It would be better to create kernel thread.
2222 static int netlink_dump_done(struct netlink_sock *nlk, struct sk_buff *skb,
2223 struct netlink_callback *cb,
2224 struct netlink_ext_ack *extack)
2226 struct nlmsghdr *nlh;
2229 nlh = nlmsg_put_answer(skb, cb, NLMSG_DONE, sizeof(nlk->dump_done_errno),
2230 NLM_F_MULTI | cb->answer_flags);
2234 nl_dump_check_consistent(cb, nlh);
2235 memcpy(nlmsg_data(nlh), &nlk->dump_done_errno, sizeof(nlk->dump_done_errno));
2237 extack_len = netlink_ack_tlv_len(nlk, nlk->dump_done_errno, extack);
2239 nlh->nlmsg_flags |= NLM_F_ACK_TLVS;
2240 if (skb_tailroom(skb) >= extack_len) {
2241 netlink_ack_tlv_fill(skb, cb->nlh,
2242 nlk->dump_done_errno, extack);
2243 nlmsg_end(skb, nlh);
2250 static int netlink_dump(struct sock *sk, bool lock_taken)
2252 struct netlink_sock *nlk = nlk_sk(sk);
2253 struct netlink_ext_ack extack = {};
2254 struct netlink_callback *cb;
2255 struct sk_buff *skb = NULL;
2256 size_t max_recvmsg_len;
2257 struct module *module;
2263 mutex_lock(&nlk->nl_cb_mutex);
2264 if (!nlk->cb_running) {
2269 if (atomic_read(&sk->sk_rmem_alloc) >= sk->sk_rcvbuf)
2272 /* NLMSG_GOODSIZE is small to avoid high order allocations being
2273 * required, but it makes sense to _attempt_ a 32KiB allocation
2274 * to reduce number of system calls on dump operations, if user
2275 * ever provided a big enough buffer.
2278 alloc_min_size = max_t(int, cb->min_dump_alloc, NLMSG_GOODSIZE);
2280 max_recvmsg_len = READ_ONCE(nlk->max_recvmsg_len);
2281 if (alloc_min_size < max_recvmsg_len) {
2282 alloc_size = max_recvmsg_len;
2283 skb = alloc_skb(alloc_size,
2284 (GFP_KERNEL & ~__GFP_DIRECT_RECLAIM) |
2285 __GFP_NOWARN | __GFP_NORETRY);
2288 alloc_size = alloc_min_size;
2289 skb = alloc_skb(alloc_size, GFP_KERNEL);
2294 /* Trim skb to allocated size. User is expected to provide buffer as
2295 * large as max(min_dump_alloc, 32KiB (max_recvmsg_len capped at
2296 * netlink_recvmsg())). dump will pack as many smaller messages as
2297 * could fit within the allocated skb. skb is typically allocated
2298 * with larger space than required (could be as much as near 2x the
2299 * requested size with align to next power of 2 approach). Allowing
2300 * dump to use the excess space makes it difficult for a user to have a
2301 * reasonable static buffer based on the expected largest dump of a
2302 * single netdev. The outcome is MSG_TRUNC error.
2304 skb_reserve(skb, skb_tailroom(skb) - alloc_size);
2306 /* Make sure malicious BPF programs can not read unitialized memory
2307 * from skb->head -> skb->data
2309 skb_reset_network_header(skb);
2310 skb_reset_mac_header(skb);
2312 netlink_skb_set_owner_r(skb, sk);
2314 if (nlk->dump_done_errno > 0) {
2315 cb->extack = &extack;
2317 nlk->dump_done_errno = cb->dump(skb, cb);
2319 /* EMSGSIZE plus something already in the skb means
2320 * that there's more to dump but current skb has filled up.
2321 * If the callback really wants to return EMSGSIZE to user space
2322 * it needs to do so again, on the next cb->dump() call,
2323 * without putting data in the skb.
2325 if (nlk->dump_done_errno == -EMSGSIZE && skb->len)
2326 nlk->dump_done_errno = skb->len;
2331 if (nlk->dump_done_errno > 0 ||
2332 skb_tailroom(skb) < nlmsg_total_size(sizeof(nlk->dump_done_errno))) {
2333 mutex_unlock(&nlk->nl_cb_mutex);
2335 if (sk_filter(sk, skb))
2338 __netlink_sendskb(sk, skb);
2342 if (netlink_dump_done(nlk, skb, cb, &extack))
2345 #ifdef CONFIG_COMPAT_NETLINK_MESSAGES
2346 /* frag_list skb's data is used for compat tasks
2347 * and the regular skb's data for normal (non-compat) tasks.
2348 * See netlink_recvmsg().
2350 if (unlikely(skb_shinfo(skb)->frag_list)) {
2351 if (netlink_dump_done(nlk, skb_shinfo(skb)->frag_list, cb, &extack))
2356 if (sk_filter(sk, skb))
2359 __netlink_sendskb(sk, skb);
2364 WRITE_ONCE(nlk->cb_running, false);
2365 module = cb->module;
2367 mutex_unlock(&nlk->nl_cb_mutex);
2373 mutex_unlock(&nlk->nl_cb_mutex);
2378 int __netlink_dump_start(struct sock *ssk, struct sk_buff *skb,
2379 const struct nlmsghdr *nlh,
2380 struct netlink_dump_control *control)
2382 struct netlink_callback *cb;
2383 struct netlink_sock *nlk;
2387 refcount_inc(&skb->users);
2389 sk = netlink_lookup(sock_net(ssk), ssk->sk_protocol, NETLINK_CB(skb).portid);
2391 ret = -ECONNREFUSED;
2396 mutex_lock(&nlk->nl_cb_mutex);
2397 /* A dump is in progress... */
2398 if (nlk->cb_running) {
2402 /* add reference of module which cb->dump belongs to */
2403 if (!try_module_get(control->module)) {
2404 ret = -EPROTONOSUPPORT;
2409 memset(cb, 0, sizeof(*cb));
2410 cb->dump = control->dump;
2411 cb->done = control->done;
2413 cb->data = control->data;
2414 cb->module = control->module;
2415 cb->min_dump_alloc = control->min_dump_alloc;
2416 cb->flags = control->flags;
2419 cb->strict_check = nlk_test_bit(STRICT_CHK, NETLINK_CB(skb).sk);
2421 if (control->start) {
2422 cb->extack = control->extack;
2423 ret = control->start(cb);
2429 WRITE_ONCE(nlk->cb_running, true);
2430 nlk->dump_done_errno = INT_MAX;
2432 ret = netlink_dump(sk, true);
2439 /* We successfully started a dump, by returning -EINTR we
2440 * signal not to send ACK even if it was requested.
2445 module_put(control->module);
2448 mutex_unlock(&nlk->nl_cb_mutex);
2453 EXPORT_SYMBOL(__netlink_dump_start);
2455 void netlink_ack(struct sk_buff *in_skb, struct nlmsghdr *nlh, int err,
2456 const struct netlink_ext_ack *extack)
2458 struct sk_buff *skb;
2459 struct nlmsghdr *rep;
2460 struct nlmsgerr *errmsg;
2461 size_t payload = sizeof(*errmsg);
2462 struct netlink_sock *nlk = nlk_sk(NETLINK_CB(in_skb).sk);
2463 unsigned int flags = 0;
2466 /* Error messages get the original request appened, unless the user
2467 * requests to cap the error message, and get extra error data if
2470 if (err && !test_bit(NETLINK_F_CAP_ACK, &nlk->flags))
2471 payload += nlmsg_len(nlh);
2473 flags |= NLM_F_CAPPED;
2475 tlvlen = netlink_ack_tlv_len(nlk, err, extack);
2477 flags |= NLM_F_ACK_TLVS;
2479 skb = nlmsg_new(payload + tlvlen, GFP_KERNEL);
2483 rep = nlmsg_put(skb, NETLINK_CB(in_skb).portid, nlh->nlmsg_seq,
2484 NLMSG_ERROR, sizeof(*errmsg), flags);
2487 errmsg = nlmsg_data(rep);
2488 errmsg->error = err;
2491 if (!(flags & NLM_F_CAPPED)) {
2492 if (!nlmsg_append(skb, nlmsg_len(nlh)))
2495 memcpy(nlmsg_data(&errmsg->msg), nlmsg_data(nlh),
2500 netlink_ack_tlv_fill(skb, nlh, err, extack);
2502 nlmsg_end(skb, rep);
2504 nlmsg_unicast(in_skb->sk, skb, NETLINK_CB(in_skb).portid);
2511 WRITE_ONCE(NETLINK_CB(in_skb).sk->sk_err, ENOBUFS);
2512 sk_error_report(NETLINK_CB(in_skb).sk);
2514 EXPORT_SYMBOL(netlink_ack);
2516 int netlink_rcv_skb(struct sk_buff *skb, int (*cb)(struct sk_buff *,
2518 struct netlink_ext_ack *))
2520 struct netlink_ext_ack extack;
2521 struct nlmsghdr *nlh;
2524 while (skb->len >= nlmsg_total_size(0)) {
2527 memset(&extack, 0, sizeof(extack));
2528 nlh = nlmsg_hdr(skb);
2531 if (nlh->nlmsg_len < NLMSG_HDRLEN || skb->len < nlh->nlmsg_len)
2534 /* Only requests are handled by the kernel */
2535 if (!(nlh->nlmsg_flags & NLM_F_REQUEST))
2538 /* Skip control messages */
2539 if (nlh->nlmsg_type < NLMSG_MIN_TYPE)
2542 err = cb(skb, nlh, &extack);
2547 if (nlh->nlmsg_flags & NLM_F_ACK || err)
2548 netlink_ack(skb, nlh, err, &extack);
2551 msglen = NLMSG_ALIGN(nlh->nlmsg_len);
2552 if (msglen > skb->len)
2554 skb_pull(skb, msglen);
2559 EXPORT_SYMBOL(netlink_rcv_skb);
2562 * nlmsg_notify - send a notification netlink message
2563 * @sk: netlink socket to use
2564 * @skb: notification message
2565 * @portid: destination netlink portid for reports or 0
2566 * @group: destination multicast group or 0
2567 * @report: 1 to report back, 0 to disable
2568 * @flags: allocation flags
2570 int nlmsg_notify(struct sock *sk, struct sk_buff *skb, u32 portid,
2571 unsigned int group, int report, gfp_t flags)
2576 int exclude_portid = 0;
2579 refcount_inc(&skb->users);
2580 exclude_portid = portid;
2583 /* errors reported via destination sk->sk_err, but propagate
2584 * delivery errors if NETLINK_BROADCAST_ERROR flag is set */
2585 err = nlmsg_multicast(sk, skb, exclude_portid, group, flags);
2593 err2 = nlmsg_unicast(sk, skb, portid);
2600 EXPORT_SYMBOL(nlmsg_notify);
2602 #ifdef CONFIG_PROC_FS
2603 struct nl_seq_iter {
2604 struct seq_net_private p;
2605 struct rhashtable_iter hti;
2609 static void netlink_walk_start(struct nl_seq_iter *iter)
2611 rhashtable_walk_enter(&nl_table[iter->link].hash, &iter->hti);
2612 rhashtable_walk_start(&iter->hti);
2615 static void netlink_walk_stop(struct nl_seq_iter *iter)
2617 rhashtable_walk_stop(&iter->hti);
2618 rhashtable_walk_exit(&iter->hti);
2621 static void *__netlink_seq_next(struct seq_file *seq)
2623 struct nl_seq_iter *iter = seq->private;
2624 struct netlink_sock *nlk;
2628 nlk = rhashtable_walk_next(&iter->hti);
2631 if (PTR_ERR(nlk) == -EAGAIN)
2640 netlink_walk_stop(iter);
2641 if (++iter->link >= MAX_LINKS)
2644 netlink_walk_start(iter);
2646 } while (sock_net(&nlk->sk) != seq_file_net(seq));
2651 static void *netlink_seq_start(struct seq_file *seq, loff_t *posp)
2654 struct nl_seq_iter *iter = seq->private;
2655 void *obj = SEQ_START_TOKEN;
2660 netlink_walk_start(iter);
2662 for (pos = *posp; pos && obj && !IS_ERR(obj); pos--)
2663 obj = __netlink_seq_next(seq);
2668 static void *netlink_seq_next(struct seq_file *seq, void *v, loff_t *pos)
2671 return __netlink_seq_next(seq);
2674 static void netlink_native_seq_stop(struct seq_file *seq, void *v)
2676 struct nl_seq_iter *iter = seq->private;
2678 if (iter->link >= MAX_LINKS)
2681 netlink_walk_stop(iter);
2685 static int netlink_native_seq_show(struct seq_file *seq, void *v)
2687 if (v == SEQ_START_TOKEN) {
2689 "sk Eth Pid Groups "
2690 "Rmem Wmem Dump Locks Drops Inode\n");
2693 struct netlink_sock *nlk = nlk_sk(s);
2695 seq_printf(seq, "%pK %-3d %-10u %08x %-8d %-8d %-5d %-8d %-8u %-8lu\n",
2699 nlk->groups ? (u32)nlk->groups[0] : 0,
2700 sk_rmem_alloc_get(s),
2701 sk_wmem_alloc_get(s),
2702 READ_ONCE(nlk->cb_running),
2703 refcount_read(&s->sk_refcnt),
2704 atomic_read(&s->sk_drops),
2712 #ifdef CONFIG_BPF_SYSCALL
2713 struct bpf_iter__netlink {
2714 __bpf_md_ptr(struct bpf_iter_meta *, meta);
2715 __bpf_md_ptr(struct netlink_sock *, sk);
2718 DEFINE_BPF_ITER_FUNC(netlink, struct bpf_iter_meta *meta, struct netlink_sock *sk)
2720 static int netlink_prog_seq_show(struct bpf_prog *prog,
2721 struct bpf_iter_meta *meta,
2724 struct bpf_iter__netlink ctx;
2726 meta->seq_num--; /* skip SEQ_START_TOKEN */
2728 ctx.sk = nlk_sk((struct sock *)v);
2729 return bpf_iter_run_prog(prog, &ctx);
2732 static int netlink_seq_show(struct seq_file *seq, void *v)
2734 struct bpf_iter_meta meta;
2735 struct bpf_prog *prog;
2738 prog = bpf_iter_get_info(&meta, false);
2740 return netlink_native_seq_show(seq, v);
2742 if (v != SEQ_START_TOKEN)
2743 return netlink_prog_seq_show(prog, &meta, v);
2748 static void netlink_seq_stop(struct seq_file *seq, void *v)
2750 struct bpf_iter_meta meta;
2751 struct bpf_prog *prog;
2755 prog = bpf_iter_get_info(&meta, true);
2757 (void)netlink_prog_seq_show(prog, &meta, v);
2760 netlink_native_seq_stop(seq, v);
2763 static int netlink_seq_show(struct seq_file *seq, void *v)
2765 return netlink_native_seq_show(seq, v);
2768 static void netlink_seq_stop(struct seq_file *seq, void *v)
2770 netlink_native_seq_stop(seq, v);
2774 static const struct seq_operations netlink_seq_ops = {
2775 .start = netlink_seq_start,
2776 .next = netlink_seq_next,
2777 .stop = netlink_seq_stop,
2778 .show = netlink_seq_show,
2782 int netlink_register_notifier(struct notifier_block *nb)
2784 return blocking_notifier_chain_register(&netlink_chain, nb);
2786 EXPORT_SYMBOL(netlink_register_notifier);
2788 int netlink_unregister_notifier(struct notifier_block *nb)
2790 return blocking_notifier_chain_unregister(&netlink_chain, nb);
2792 EXPORT_SYMBOL(netlink_unregister_notifier);
2794 static const struct proto_ops netlink_ops = {
2795 .family = PF_NETLINK,
2796 .owner = THIS_MODULE,
2797 .release = netlink_release,
2798 .bind = netlink_bind,
2799 .connect = netlink_connect,
2800 .socketpair = sock_no_socketpair,
2801 .accept = sock_no_accept,
2802 .getname = netlink_getname,
2803 .poll = datagram_poll,
2804 .ioctl = netlink_ioctl,
2805 .listen = sock_no_listen,
2806 .shutdown = sock_no_shutdown,
2807 .setsockopt = netlink_setsockopt,
2808 .getsockopt = netlink_getsockopt,
2809 .sendmsg = netlink_sendmsg,
2810 .recvmsg = netlink_recvmsg,
2811 .mmap = sock_no_mmap,
2814 static const struct net_proto_family netlink_family_ops = {
2815 .family = PF_NETLINK,
2816 .create = netlink_create,
2817 .owner = THIS_MODULE, /* for consistency 8) */
2820 static int __net_init netlink_net_init(struct net *net)
2822 #ifdef CONFIG_PROC_FS
2823 if (!proc_create_net("netlink", 0, net->proc_net, &netlink_seq_ops,
2824 sizeof(struct nl_seq_iter)))
2830 static void __net_exit netlink_net_exit(struct net *net)
2832 #ifdef CONFIG_PROC_FS
2833 remove_proc_entry("netlink", net->proc_net);
2837 static void __init netlink_add_usersock_entry(void)
2839 struct listeners *listeners;
2842 listeners = kzalloc(sizeof(*listeners) + NLGRPSZ(groups), GFP_KERNEL);
2844 panic("netlink_add_usersock_entry: Cannot allocate listeners\n");
2846 netlink_table_grab();
2848 nl_table[NETLINK_USERSOCK].groups = groups;
2849 rcu_assign_pointer(nl_table[NETLINK_USERSOCK].listeners, listeners);
2850 nl_table[NETLINK_USERSOCK].module = THIS_MODULE;
2851 nl_table[NETLINK_USERSOCK].registered = 1;
2852 nl_table[NETLINK_USERSOCK].flags = NL_CFG_F_NONROOT_SEND;
2854 netlink_table_ungrab();
2857 static struct pernet_operations __net_initdata netlink_net_ops = {
2858 .init = netlink_net_init,
2859 .exit = netlink_net_exit,
2862 static inline u32 netlink_hash(const void *data, u32 len, u32 seed)
2864 const struct netlink_sock *nlk = data;
2865 struct netlink_compare_arg arg;
2867 netlink_compare_arg_init(&arg, sock_net(&nlk->sk), nlk->portid);
2868 return jhash2((u32 *)&arg, netlink_compare_arg_len / sizeof(u32), seed);
2871 static const struct rhashtable_params netlink_rhashtable_params = {
2872 .head_offset = offsetof(struct netlink_sock, node),
2873 .key_len = netlink_compare_arg_len,
2874 .obj_hashfn = netlink_hash,
2875 .obj_cmpfn = netlink_compare,
2876 .automatic_shrinking = true,
2879 #if defined(CONFIG_BPF_SYSCALL) && defined(CONFIG_PROC_FS)
2880 BTF_ID_LIST(btf_netlink_sock_id)
2881 BTF_ID(struct, netlink_sock)
2883 static const struct bpf_iter_seq_info netlink_seq_info = {
2884 .seq_ops = &netlink_seq_ops,
2885 .init_seq_private = bpf_iter_init_seq_net,
2886 .fini_seq_private = bpf_iter_fini_seq_net,
2887 .seq_priv_size = sizeof(struct nl_seq_iter),
2890 static struct bpf_iter_reg netlink_reg_info = {
2891 .target = "netlink",
2892 .ctx_arg_info_size = 1,
2894 { offsetof(struct bpf_iter__netlink, sk),
2895 PTR_TO_BTF_ID_OR_NULL },
2897 .seq_info = &netlink_seq_info,
2900 static int __init bpf_iter_register(void)
2902 netlink_reg_info.ctx_arg_info[0].btf_id = *btf_netlink_sock_id;
2903 return bpf_iter_reg_target(&netlink_reg_info);
2907 static int __init netlink_proto_init(void)
2910 int err = proto_register(&netlink_proto, 0);
2915 #if defined(CONFIG_BPF_SYSCALL) && defined(CONFIG_PROC_FS)
2916 err = bpf_iter_register();
2921 BUILD_BUG_ON(sizeof(struct netlink_skb_parms) > sizeof_field(struct sk_buff, cb));
2923 nl_table = kcalloc(MAX_LINKS, sizeof(*nl_table), GFP_KERNEL);
2927 for (i = 0; i < MAX_LINKS; i++) {
2928 if (rhashtable_init(&nl_table[i].hash,
2929 &netlink_rhashtable_params) < 0)
2933 netlink_add_usersock_entry();
2935 sock_register(&netlink_family_ops);
2936 register_pernet_subsys(&netlink_net_ops);
2937 register_pernet_subsys(&netlink_tap_net_ops);
2938 /* The netlink device handler may be needed early. */
2943 panic("netlink_init: Cannot allocate nl_table\n");
2946 core_initcall(netlink_proto_init);