1 // SPDX-License-Identifier: GPL-2.0-or-later
3 * NETLINK Kernel-user communication protocol.
5 * Authors: Alan Cox <alan@lxorguk.ukuu.org.uk>
6 * Alexey Kuznetsov <kuznet@ms2.inr.ac.ru>
7 * Patrick McHardy <kaber@trash.net>
9 * Tue Jun 26 14:36:48 MEST 2001 Herbert "herp" Rosmanith
10 * added netlink_proto_exit
11 * Tue Jan 22 18:32:44 BRST 2002 Arnaldo C. de Melo <acme@conectiva.com.br>
12 * use nlk_sk, as sk->protinfo is on a diet 8)
13 * Fri Jul 22 19:51:12 MEST 2005 Harald Welte <laforge@gnumonks.org>
14 * - inc module use count of module that owns
15 * the kernel socket in case userspace opens
16 * socket of same protocol
17 * - remove all module support, since netlink is
18 * mandatory if CONFIG_NET=y these days
21 #include <linux/module.h>
23 #include <linux/capability.h>
24 #include <linux/kernel.h>
25 #include <linux/init.h>
26 #include <linux/signal.h>
27 #include <linux/sched.h>
28 #include <linux/errno.h>
29 #include <linux/string.h>
30 #include <linux/stat.h>
31 #include <linux/socket.h>
33 #include <linux/fcntl.h>
34 #include <linux/termios.h>
35 #include <linux/sockios.h>
36 #include <linux/net.h>
38 #include <linux/slab.h>
39 #include <linux/uaccess.h>
40 #include <linux/skbuff.h>
41 #include <linux/netdevice.h>
42 #include <linux/rtnetlink.h>
43 #include <linux/proc_fs.h>
44 #include <linux/seq_file.h>
45 #include <linux/notifier.h>
46 #include <linux/security.h>
47 #include <linux/jhash.h>
48 #include <linux/jiffies.h>
49 #include <linux/random.h>
50 #include <linux/bitops.h>
52 #include <linux/types.h>
53 #include <linux/audit.h>
54 #include <linux/mutex.h>
55 #include <linux/vmalloc.h>
56 #include <linux/if_arp.h>
57 #include <linux/rhashtable.h>
58 #include <asm/cacheflush.h>
59 #include <linux/hash.h>
60 #include <linux/genetlink.h>
61 #include <linux/net_namespace.h>
62 #include <linux/nospec.h>
63 #include <linux/btf_ids.h>
65 #include <net/net_namespace.h>
66 #include <net/netns/generic.h>
69 #include <net/netlink.h>
70 #define CREATE_TRACE_POINTS
71 #include <trace/events/netlink.h>
73 #include "af_netlink.h"
77 unsigned long masks[];
81 #define NETLINK_S_CONGESTED 0x0
83 static inline int netlink_is_kernel(struct sock *sk)
85 return nlk_sk(sk)->flags & NETLINK_F_KERNEL_SOCKET;
88 struct netlink_table *nl_table __read_mostly;
89 EXPORT_SYMBOL_GPL(nl_table);
91 static DECLARE_WAIT_QUEUE_HEAD(nl_table_wait);
93 static struct lock_class_key nlk_cb_mutex_keys[MAX_LINKS];
95 static const char *const nlk_cb_mutex_key_strings[MAX_LINKS + 1] = {
98 "nlk_cb_mutex-USERSOCK",
99 "nlk_cb_mutex-FIREWALL",
100 "nlk_cb_mutex-SOCK_DIAG",
101 "nlk_cb_mutex-NFLOG",
103 "nlk_cb_mutex-SELINUX",
104 "nlk_cb_mutex-ISCSI",
105 "nlk_cb_mutex-AUDIT",
106 "nlk_cb_mutex-FIB_LOOKUP",
107 "nlk_cb_mutex-CONNECTOR",
108 "nlk_cb_mutex-NETFILTER",
109 "nlk_cb_mutex-IP6_FW",
110 "nlk_cb_mutex-DNRTMSG",
111 "nlk_cb_mutex-KOBJECT_UEVENT",
112 "nlk_cb_mutex-GENERIC",
114 "nlk_cb_mutex-SCSITRANSPORT",
115 "nlk_cb_mutex-ECRYPTFS",
117 "nlk_cb_mutex-CRYPTO",
128 "nlk_cb_mutex-MAX_LINKS"
131 static int netlink_dump(struct sock *sk);
133 /* nl_table locking explained:
134 * Lookup and traversal are protected with an RCU read-side lock. Insertion
135 * and removal are protected with per bucket lock while using RCU list
136 * modification primitives and may run in parallel to RCU protected lookups.
137 * Destruction of the Netlink socket may only occur *after* nl_table_lock has
138 * been acquired * either during or after the socket has been removed from
139 * the list and after an RCU grace period.
141 DEFINE_RWLOCK(nl_table_lock);
142 EXPORT_SYMBOL_GPL(nl_table_lock);
143 static atomic_t nl_table_users = ATOMIC_INIT(0);
145 #define nl_deref_protected(X) rcu_dereference_protected(X, lockdep_is_held(&nl_table_lock));
147 static BLOCKING_NOTIFIER_HEAD(netlink_chain);
150 static const struct rhashtable_params netlink_rhashtable_params;
152 void do_trace_netlink_extack(const char *msg)
154 trace_netlink_extack(msg);
156 EXPORT_SYMBOL(do_trace_netlink_extack);
158 static inline u32 netlink_group_mask(u32 group)
160 return group ? 1 << (group - 1) : 0;
163 static struct sk_buff *netlink_to_full_skb(const struct sk_buff *skb,
166 unsigned int len = skb_end_offset(skb);
169 new = alloc_skb(len, gfp_mask);
173 NETLINK_CB(new).portid = NETLINK_CB(skb).portid;
174 NETLINK_CB(new).dst_group = NETLINK_CB(skb).dst_group;
175 NETLINK_CB(new).creds = NETLINK_CB(skb).creds;
177 skb_put_data(new, skb->data, len);
181 static unsigned int netlink_tap_net_id;
183 struct netlink_tap_net {
184 struct list_head netlink_tap_all;
185 struct mutex netlink_tap_lock;
188 int netlink_add_tap(struct netlink_tap *nt)
190 struct net *net = dev_net(nt->dev);
191 struct netlink_tap_net *nn = net_generic(net, netlink_tap_net_id);
193 if (unlikely(nt->dev->type != ARPHRD_NETLINK))
196 mutex_lock(&nn->netlink_tap_lock);
197 list_add_rcu(&nt->list, &nn->netlink_tap_all);
198 mutex_unlock(&nn->netlink_tap_lock);
200 __module_get(nt->module);
204 EXPORT_SYMBOL_GPL(netlink_add_tap);
206 static int __netlink_remove_tap(struct netlink_tap *nt)
208 struct net *net = dev_net(nt->dev);
209 struct netlink_tap_net *nn = net_generic(net, netlink_tap_net_id);
211 struct netlink_tap *tmp;
213 mutex_lock(&nn->netlink_tap_lock);
215 list_for_each_entry(tmp, &nn->netlink_tap_all, list) {
217 list_del_rcu(&nt->list);
223 pr_warn("__netlink_remove_tap: %p not found\n", nt);
225 mutex_unlock(&nn->netlink_tap_lock);
228 module_put(nt->module);
230 return found ? 0 : -ENODEV;
233 int netlink_remove_tap(struct netlink_tap *nt)
237 ret = __netlink_remove_tap(nt);
242 EXPORT_SYMBOL_GPL(netlink_remove_tap);
244 static __net_init int netlink_tap_init_net(struct net *net)
246 struct netlink_tap_net *nn = net_generic(net, netlink_tap_net_id);
248 INIT_LIST_HEAD(&nn->netlink_tap_all);
249 mutex_init(&nn->netlink_tap_lock);
253 static struct pernet_operations netlink_tap_net_ops = {
254 .init = netlink_tap_init_net,
255 .id = &netlink_tap_net_id,
256 .size = sizeof(struct netlink_tap_net),
259 static bool netlink_filter_tap(const struct sk_buff *skb)
261 struct sock *sk = skb->sk;
263 /* We take the more conservative approach and
264 * whitelist socket protocols that may pass.
266 switch (sk->sk_protocol) {
268 case NETLINK_USERSOCK:
269 case NETLINK_SOCK_DIAG:
272 case NETLINK_FIB_LOOKUP:
273 case NETLINK_NETFILTER:
274 case NETLINK_GENERIC:
281 static int __netlink_deliver_tap_skb(struct sk_buff *skb,
282 struct net_device *dev)
284 struct sk_buff *nskb;
285 struct sock *sk = skb->sk;
288 if (!net_eq(dev_net(dev), sock_net(sk)))
293 if (is_vmalloc_addr(skb->head))
294 nskb = netlink_to_full_skb(skb, GFP_ATOMIC);
296 nskb = skb_clone(skb, GFP_ATOMIC);
299 nskb->protocol = htons((u16) sk->sk_protocol);
300 nskb->pkt_type = netlink_is_kernel(sk) ?
301 PACKET_KERNEL : PACKET_USER;
302 skb_reset_network_header(nskb);
303 ret = dev_queue_xmit(nskb);
304 if (unlikely(ret > 0))
305 ret = net_xmit_errno(ret);
312 static void __netlink_deliver_tap(struct sk_buff *skb, struct netlink_tap_net *nn)
315 struct netlink_tap *tmp;
317 if (!netlink_filter_tap(skb))
320 list_for_each_entry_rcu(tmp, &nn->netlink_tap_all, list) {
321 ret = __netlink_deliver_tap_skb(skb, tmp->dev);
327 static void netlink_deliver_tap(struct net *net, struct sk_buff *skb)
329 struct netlink_tap_net *nn = net_generic(net, netlink_tap_net_id);
333 if (unlikely(!list_empty(&nn->netlink_tap_all)))
334 __netlink_deliver_tap(skb, nn);
339 static void netlink_deliver_tap_kernel(struct sock *dst, struct sock *src,
342 if (!(netlink_is_kernel(dst) && netlink_is_kernel(src)))
343 netlink_deliver_tap(sock_net(dst), skb);
346 static void netlink_overrun(struct sock *sk)
348 struct netlink_sock *nlk = nlk_sk(sk);
350 if (!(nlk->flags & NETLINK_F_RECV_NO_ENOBUFS)) {
351 if (!test_and_set_bit(NETLINK_S_CONGESTED,
352 &nlk_sk(sk)->state)) {
353 sk->sk_err = ENOBUFS;
354 sk->sk_error_report(sk);
357 atomic_inc(&sk->sk_drops);
360 static void netlink_rcv_wake(struct sock *sk)
362 struct netlink_sock *nlk = nlk_sk(sk);
364 if (skb_queue_empty_lockless(&sk->sk_receive_queue))
365 clear_bit(NETLINK_S_CONGESTED, &nlk->state);
366 if (!test_bit(NETLINK_S_CONGESTED, &nlk->state))
367 wake_up_interruptible(&nlk->wait);
370 static void netlink_skb_destructor(struct sk_buff *skb)
372 if (is_vmalloc_addr(skb->head)) {
374 !atomic_dec_return(&(skb_shinfo(skb)->dataref)))
383 static void netlink_skb_set_owner_r(struct sk_buff *skb, struct sock *sk)
385 WARN_ON(skb->sk != NULL);
387 skb->destructor = netlink_skb_destructor;
388 atomic_add(skb->truesize, &sk->sk_rmem_alloc);
389 sk_mem_charge(sk, skb->truesize);
392 static void netlink_sock_destruct(struct sock *sk)
394 struct netlink_sock *nlk = nlk_sk(sk);
396 if (nlk->cb_running) {
398 nlk->cb.done(&nlk->cb);
399 module_put(nlk->cb.module);
400 kfree_skb(nlk->cb.skb);
403 skb_queue_purge(&sk->sk_receive_queue);
405 if (!sock_flag(sk, SOCK_DEAD)) {
406 printk(KERN_ERR "Freeing alive netlink socket %p\n", sk);
410 WARN_ON(atomic_read(&sk->sk_rmem_alloc));
411 WARN_ON(refcount_read(&sk->sk_wmem_alloc));
412 WARN_ON(nlk_sk(sk)->groups);
415 static void netlink_sock_destruct_work(struct work_struct *work)
417 struct netlink_sock *nlk = container_of(work, struct netlink_sock,
423 /* This lock without WQ_FLAG_EXCLUSIVE is good on UP and it is _very_ bad on
424 * SMP. Look, when several writers sleep and reader wakes them up, all but one
425 * immediately hit write lock and grab all the cpus. Exclusive sleep solves
426 * this, _but_ remember, it adds useless work on UP machines.
429 void netlink_table_grab(void)
430 __acquires(nl_table_lock)
434 write_lock_irq(&nl_table_lock);
436 if (atomic_read(&nl_table_users)) {
437 DECLARE_WAITQUEUE(wait, current);
439 add_wait_queue_exclusive(&nl_table_wait, &wait);
441 set_current_state(TASK_UNINTERRUPTIBLE);
442 if (atomic_read(&nl_table_users) == 0)
444 write_unlock_irq(&nl_table_lock);
446 write_lock_irq(&nl_table_lock);
449 __set_current_state(TASK_RUNNING);
450 remove_wait_queue(&nl_table_wait, &wait);
454 void netlink_table_ungrab(void)
455 __releases(nl_table_lock)
457 write_unlock_irq(&nl_table_lock);
458 wake_up(&nl_table_wait);
462 netlink_lock_table(void)
464 /* read_lock() synchronizes us to netlink_table_grab */
466 read_lock(&nl_table_lock);
467 atomic_inc(&nl_table_users);
468 read_unlock(&nl_table_lock);
472 netlink_unlock_table(void)
474 if (atomic_dec_and_test(&nl_table_users))
475 wake_up(&nl_table_wait);
478 struct netlink_compare_arg
484 /* Doing sizeof directly may yield 4 extra bytes on 64-bit. */
485 #define netlink_compare_arg_len \
486 (offsetof(struct netlink_compare_arg, portid) + sizeof(u32))
488 static inline int netlink_compare(struct rhashtable_compare_arg *arg,
491 const struct netlink_compare_arg *x = arg->key;
492 const struct netlink_sock *nlk = ptr;
494 return nlk->portid != x->portid ||
495 !net_eq(sock_net(&nlk->sk), read_pnet(&x->pnet));
498 static void netlink_compare_arg_init(struct netlink_compare_arg *arg,
499 struct net *net, u32 portid)
501 memset(arg, 0, sizeof(*arg));
502 write_pnet(&arg->pnet, net);
503 arg->portid = portid;
506 static struct sock *__netlink_lookup(struct netlink_table *table, u32 portid,
509 struct netlink_compare_arg arg;
511 netlink_compare_arg_init(&arg, net, portid);
512 return rhashtable_lookup_fast(&table->hash, &arg,
513 netlink_rhashtable_params);
516 static int __netlink_insert(struct netlink_table *table, struct sock *sk)
518 struct netlink_compare_arg arg;
520 netlink_compare_arg_init(&arg, sock_net(sk), nlk_sk(sk)->portid);
521 return rhashtable_lookup_insert_key(&table->hash, &arg,
523 netlink_rhashtable_params);
526 static struct sock *netlink_lookup(struct net *net, int protocol, u32 portid)
528 struct netlink_table *table = &nl_table[protocol];
532 sk = __netlink_lookup(table, portid, net);
540 static const struct proto_ops netlink_ops;
543 netlink_update_listeners(struct sock *sk)
545 struct netlink_table *tbl = &nl_table[sk->sk_protocol];
548 struct listeners *listeners;
550 listeners = nl_deref_protected(tbl->listeners);
554 for (i = 0; i < NLGRPLONGS(tbl->groups); i++) {
556 sk_for_each_bound(sk, &tbl->mc_list) {
557 if (i < NLGRPLONGS(nlk_sk(sk)->ngroups))
558 mask |= nlk_sk(sk)->groups[i];
560 listeners->masks[i] = mask;
562 /* this function is only called with the netlink table "grabbed", which
563 * makes sure updates are visible before bind or setsockopt return. */
566 static int netlink_insert(struct sock *sk, u32 portid)
568 struct netlink_table *table = &nl_table[sk->sk_protocol];
573 err = nlk_sk(sk)->portid == portid ? 0 : -EBUSY;
574 if (nlk_sk(sk)->bound)
577 nlk_sk(sk)->portid = portid;
580 err = __netlink_insert(table, sk);
582 /* In case the hashtable backend returns with -EBUSY
583 * from here, it must not escape to the caller.
585 if (unlikely(err == -EBUSY))
593 /* We need to ensure that the socket is hashed and visible. */
595 nlk_sk(sk)->bound = portid;
602 static void netlink_remove(struct sock *sk)
604 struct netlink_table *table;
606 table = &nl_table[sk->sk_protocol];
607 if (!rhashtable_remove_fast(&table->hash, &nlk_sk(sk)->node,
608 netlink_rhashtable_params)) {
609 WARN_ON(refcount_read(&sk->sk_refcnt) == 1);
613 netlink_table_grab();
614 if (nlk_sk(sk)->subscriptions) {
615 __sk_del_bind_node(sk);
616 netlink_update_listeners(sk);
618 if (sk->sk_protocol == NETLINK_GENERIC)
619 atomic_inc(&genl_sk_destructing_cnt);
620 netlink_table_ungrab();
623 static struct proto netlink_proto = {
625 .owner = THIS_MODULE,
626 .obj_size = sizeof(struct netlink_sock),
629 static int __netlink_create(struct net *net, struct socket *sock,
630 struct mutex *cb_mutex, int protocol,
634 struct netlink_sock *nlk;
636 sock->ops = &netlink_ops;
638 sk = sk_alloc(net, PF_NETLINK, GFP_KERNEL, &netlink_proto, kern);
642 sock_init_data(sock, sk);
646 nlk->cb_mutex = cb_mutex;
648 nlk->cb_mutex = &nlk->cb_def_mutex;
649 mutex_init(nlk->cb_mutex);
650 lockdep_set_class_and_name(nlk->cb_mutex,
651 nlk_cb_mutex_keys + protocol,
652 nlk_cb_mutex_key_strings[protocol]);
654 init_waitqueue_head(&nlk->wait);
656 sk->sk_destruct = netlink_sock_destruct;
657 sk->sk_protocol = protocol;
661 static int netlink_create(struct net *net, struct socket *sock, int protocol,
664 struct module *module = NULL;
665 struct mutex *cb_mutex;
666 struct netlink_sock *nlk;
667 int (*bind)(struct net *net, int group);
668 void (*unbind)(struct net *net, int group);
671 sock->state = SS_UNCONNECTED;
673 if (sock->type != SOCK_RAW && sock->type != SOCK_DGRAM)
674 return -ESOCKTNOSUPPORT;
676 if (protocol < 0 || protocol >= MAX_LINKS)
677 return -EPROTONOSUPPORT;
678 protocol = array_index_nospec(protocol, MAX_LINKS);
680 netlink_lock_table();
681 #ifdef CONFIG_MODULES
682 if (!nl_table[protocol].registered) {
683 netlink_unlock_table();
684 request_module("net-pf-%d-proto-%d", PF_NETLINK, protocol);
685 netlink_lock_table();
688 if (nl_table[protocol].registered &&
689 try_module_get(nl_table[protocol].module))
690 module = nl_table[protocol].module;
692 err = -EPROTONOSUPPORT;
693 cb_mutex = nl_table[protocol].cb_mutex;
694 bind = nl_table[protocol].bind;
695 unbind = nl_table[protocol].unbind;
696 netlink_unlock_table();
701 err = __netlink_create(net, sock, cb_mutex, protocol, kern);
706 sock_prot_inuse_add(net, &netlink_proto, 1);
709 nlk = nlk_sk(sock->sk);
710 nlk->module = module;
711 nlk->netlink_bind = bind;
712 nlk->netlink_unbind = unbind;
721 static void deferred_put_nlk_sk(struct rcu_head *head)
723 struct netlink_sock *nlk = container_of(head, struct netlink_sock, rcu);
724 struct sock *sk = &nlk->sk;
729 if (!refcount_dec_and_test(&sk->sk_refcnt))
732 if (nlk->cb_running && nlk->cb.done) {
733 INIT_WORK(&nlk->work, netlink_sock_destruct_work);
734 schedule_work(&nlk->work);
741 static int netlink_release(struct socket *sock)
743 struct sock *sk = sock->sk;
744 struct netlink_sock *nlk;
754 * OK. Socket is unlinked, any packets that arrive now
758 /* must not acquire netlink_table_lock in any way again before unbind
759 * and notifying genetlink is done as otherwise it might deadlock
761 if (nlk->netlink_unbind) {
764 for (i = 0; i < nlk->ngroups; i++)
765 if (test_bit(i, nlk->groups))
766 nlk->netlink_unbind(sock_net(sk), i + 1);
768 if (sk->sk_protocol == NETLINK_GENERIC &&
769 atomic_dec_return(&genl_sk_destructing_cnt) == 0)
770 wake_up(&genl_sk_destructing_waitq);
773 wake_up_interruptible_all(&nlk->wait);
775 skb_queue_purge(&sk->sk_write_queue);
777 if (nlk->portid && nlk->bound) {
778 struct netlink_notify n = {
780 .protocol = sk->sk_protocol,
781 .portid = nlk->portid,
783 blocking_notifier_call_chain(&netlink_chain,
784 NETLINK_URELEASE, &n);
787 module_put(nlk->module);
789 if (netlink_is_kernel(sk)) {
790 netlink_table_grab();
791 BUG_ON(nl_table[sk->sk_protocol].registered == 0);
792 if (--nl_table[sk->sk_protocol].registered == 0) {
793 struct listeners *old;
795 old = nl_deref_protected(nl_table[sk->sk_protocol].listeners);
796 RCU_INIT_POINTER(nl_table[sk->sk_protocol].listeners, NULL);
798 nl_table[sk->sk_protocol].module = NULL;
799 nl_table[sk->sk_protocol].bind = NULL;
800 nl_table[sk->sk_protocol].unbind = NULL;
801 nl_table[sk->sk_protocol].flags = 0;
802 nl_table[sk->sk_protocol].registered = 0;
804 netlink_table_ungrab();
808 sock_prot_inuse_add(sock_net(sk), &netlink_proto, -1);
810 call_rcu(&nlk->rcu, deferred_put_nlk_sk);
814 static int netlink_autobind(struct socket *sock)
816 struct sock *sk = sock->sk;
817 struct net *net = sock_net(sk);
818 struct netlink_table *table = &nl_table[sk->sk_protocol];
819 s32 portid = task_tgid_vnr(current);
827 ok = !__netlink_lookup(table, portid, net);
830 /* Bind collision, search negative portid values. */
832 /* rover will be in range [S32_MIN, -4097] */
833 rover = S32_MIN + prandom_u32_max(-4096 - S32_MIN);
834 else if (rover >= -4096)
840 err = netlink_insert(sk, portid);
841 if (err == -EADDRINUSE)
844 /* If 2 threads race to autobind, that is fine. */
852 * __netlink_ns_capable - General netlink message capability test
853 * @nsp: NETLINK_CB of the socket buffer holding a netlink command from userspace.
854 * @user_ns: The user namespace of the capability to use
855 * @cap: The capability to use
857 * Test to see if the opener of the socket we received the message
858 * from had when the netlink socket was created and the sender of the
859 * message has the capability @cap in the user namespace @user_ns.
861 bool __netlink_ns_capable(const struct netlink_skb_parms *nsp,
862 struct user_namespace *user_ns, int cap)
864 return ((nsp->flags & NETLINK_SKB_DST) ||
865 file_ns_capable(nsp->sk->sk_socket->file, user_ns, cap)) &&
866 ns_capable(user_ns, cap);
868 EXPORT_SYMBOL(__netlink_ns_capable);
871 * netlink_ns_capable - General netlink message capability test
872 * @skb: socket buffer holding a netlink command from userspace
873 * @user_ns: The user namespace of the capability to use
874 * @cap: The capability to use
876 * Test to see if the opener of the socket we received the message
877 * from had when the netlink socket was created and the sender of the
878 * message has the capability @cap in the user namespace @user_ns.
880 bool netlink_ns_capable(const struct sk_buff *skb,
881 struct user_namespace *user_ns, int cap)
883 return __netlink_ns_capable(&NETLINK_CB(skb), user_ns, cap);
885 EXPORT_SYMBOL(netlink_ns_capable);
888 * netlink_capable - Netlink global message capability test
889 * @skb: socket buffer holding a netlink command from userspace
890 * @cap: The capability to use
892 * Test to see if the opener of the socket we received the message
893 * from had when the netlink socket was created and the sender of the
894 * message has the capability @cap in all user namespaces.
896 bool netlink_capable(const struct sk_buff *skb, int cap)
898 return netlink_ns_capable(skb, &init_user_ns, cap);
900 EXPORT_SYMBOL(netlink_capable);
903 * netlink_net_capable - Netlink network namespace message capability test
904 * @skb: socket buffer holding a netlink command from userspace
905 * @cap: The capability to use
907 * Test to see if the opener of the socket we received the message
908 * from had when the netlink socket was created and the sender of the
909 * message has the capability @cap over the network namespace of
910 * the socket we received the message from.
912 bool netlink_net_capable(const struct sk_buff *skb, int cap)
914 return netlink_ns_capable(skb, sock_net(skb->sk)->user_ns, cap);
916 EXPORT_SYMBOL(netlink_net_capable);
918 static inline int netlink_allowed(const struct socket *sock, unsigned int flag)
920 return (nl_table[sock->sk->sk_protocol].flags & flag) ||
921 ns_capable(sock_net(sock->sk)->user_ns, CAP_NET_ADMIN);
925 netlink_update_subscriptions(struct sock *sk, unsigned int subscriptions)
927 struct netlink_sock *nlk = nlk_sk(sk);
929 if (nlk->subscriptions && !subscriptions)
930 __sk_del_bind_node(sk);
931 else if (!nlk->subscriptions && subscriptions)
932 sk_add_bind_node(sk, &nl_table[sk->sk_protocol].mc_list);
933 nlk->subscriptions = subscriptions;
936 static int netlink_realloc_groups(struct sock *sk)
938 struct netlink_sock *nlk = nlk_sk(sk);
940 unsigned long *new_groups;
943 netlink_table_grab();
945 groups = nl_table[sk->sk_protocol].groups;
946 if (!nl_table[sk->sk_protocol].registered) {
951 if (nlk->ngroups >= groups)
954 new_groups = krealloc(nlk->groups, NLGRPSZ(groups), GFP_ATOMIC);
955 if (new_groups == NULL) {
959 memset((char *)new_groups + NLGRPSZ(nlk->ngroups), 0,
960 NLGRPSZ(groups) - NLGRPSZ(nlk->ngroups));
962 nlk->groups = new_groups;
963 nlk->ngroups = groups;
965 netlink_table_ungrab();
969 static void netlink_undo_bind(int group, long unsigned int groups,
972 struct netlink_sock *nlk = nlk_sk(sk);
975 if (!nlk->netlink_unbind)
978 for (undo = 0; undo < group; undo++)
979 if (test_bit(undo, &groups))
980 nlk->netlink_unbind(sock_net(sk), undo + 1);
983 static int netlink_bind(struct socket *sock, struct sockaddr *addr,
986 struct sock *sk = sock->sk;
987 struct net *net = sock_net(sk);
988 struct netlink_sock *nlk = nlk_sk(sk);
989 struct sockaddr_nl *nladdr = (struct sockaddr_nl *)addr;
991 unsigned long groups;
994 if (addr_len < sizeof(struct sockaddr_nl))
997 if (nladdr->nl_family != AF_NETLINK)
999 groups = nladdr->nl_groups;
1001 /* Only superuser is allowed to listen multicasts */
1003 if (!netlink_allowed(sock, NL_CFG_F_NONROOT_RECV))
1005 err = netlink_realloc_groups(sk);
1010 if (nlk->ngroups < BITS_PER_LONG)
1011 groups &= (1UL << nlk->ngroups) - 1;
1015 /* Ensure nlk->portid is up-to-date. */
1018 if (nladdr->nl_pid != nlk->portid)
1022 if (nlk->netlink_bind && groups) {
1025 /* nl_groups is a u32, so cap the maximum groups we can bind */
1026 for (group = 0; group < BITS_PER_TYPE(u32); group++) {
1027 if (!test_bit(group, &groups))
1029 err = nlk->netlink_bind(net, group + 1);
1032 netlink_undo_bind(group, groups, sk);
1037 /* No need for barriers here as we return to user-space without
1038 * using any of the bound attributes.
1040 netlink_lock_table();
1042 err = nladdr->nl_pid ?
1043 netlink_insert(sk, nladdr->nl_pid) :
1044 netlink_autobind(sock);
1046 netlink_undo_bind(BITS_PER_TYPE(u32), groups, sk);
1051 if (!groups && (nlk->groups == NULL || !(u32)nlk->groups[0]))
1053 netlink_unlock_table();
1055 netlink_table_grab();
1056 netlink_update_subscriptions(sk, nlk->subscriptions +
1058 hweight32(nlk->groups[0]));
1059 nlk->groups[0] = (nlk->groups[0] & ~0xffffffffUL) | groups;
1060 netlink_update_listeners(sk);
1061 netlink_table_ungrab();
1066 netlink_unlock_table();
1070 static int netlink_connect(struct socket *sock, struct sockaddr *addr,
1071 int alen, int flags)
1074 struct sock *sk = sock->sk;
1075 struct netlink_sock *nlk = nlk_sk(sk);
1076 struct sockaddr_nl *nladdr = (struct sockaddr_nl *)addr;
1078 if (alen < sizeof(addr->sa_family))
1081 if (addr->sa_family == AF_UNSPEC) {
1082 sk->sk_state = NETLINK_UNCONNECTED;
1083 nlk->dst_portid = 0;
1087 if (addr->sa_family != AF_NETLINK)
1090 if (alen < sizeof(struct sockaddr_nl))
1093 if ((nladdr->nl_groups || nladdr->nl_pid) &&
1094 !netlink_allowed(sock, NL_CFG_F_NONROOT_SEND))
1097 /* No need for barriers here as we return to user-space without
1098 * using any of the bound attributes.
1101 err = netlink_autobind(sock);
1104 sk->sk_state = NETLINK_CONNECTED;
1105 nlk->dst_portid = nladdr->nl_pid;
1106 nlk->dst_group = ffs(nladdr->nl_groups);
1112 static int netlink_getname(struct socket *sock, struct sockaddr *addr,
1115 struct sock *sk = sock->sk;
1116 struct netlink_sock *nlk = nlk_sk(sk);
1117 DECLARE_SOCKADDR(struct sockaddr_nl *, nladdr, addr);
1119 nladdr->nl_family = AF_NETLINK;
1123 nladdr->nl_pid = nlk->dst_portid;
1124 nladdr->nl_groups = netlink_group_mask(nlk->dst_group);
1126 nladdr->nl_pid = nlk->portid;
1127 netlink_lock_table();
1128 nladdr->nl_groups = nlk->groups ? nlk->groups[0] : 0;
1129 netlink_unlock_table();
1131 return sizeof(*nladdr);
1134 static int netlink_ioctl(struct socket *sock, unsigned int cmd,
1137 /* try to hand this ioctl down to the NIC drivers.
1139 return -ENOIOCTLCMD;
1142 static struct sock *netlink_getsockbyportid(struct sock *ssk, u32 portid)
1145 struct netlink_sock *nlk;
1147 sock = netlink_lookup(sock_net(ssk), ssk->sk_protocol, portid);
1149 return ERR_PTR(-ECONNREFUSED);
1151 /* Don't bother queuing skb if kernel socket has no input function */
1153 if (sock->sk_state == NETLINK_CONNECTED &&
1154 nlk->dst_portid != nlk_sk(ssk)->portid) {
1156 return ERR_PTR(-ECONNREFUSED);
1161 struct sock *netlink_getsockbyfilp(struct file *filp)
1163 struct inode *inode = file_inode(filp);
1166 if (!S_ISSOCK(inode->i_mode))
1167 return ERR_PTR(-ENOTSOCK);
1169 sock = SOCKET_I(inode)->sk;
1170 if (sock->sk_family != AF_NETLINK)
1171 return ERR_PTR(-EINVAL);
1177 static struct sk_buff *netlink_alloc_large_skb(unsigned int size,
1180 struct sk_buff *skb;
1183 if (size <= NLMSG_GOODSIZE || broadcast)
1184 return alloc_skb(size, GFP_KERNEL);
1186 size = SKB_DATA_ALIGN(size) +
1187 SKB_DATA_ALIGN(sizeof(struct skb_shared_info));
1189 data = vmalloc(size);
1193 skb = __build_skb(data, size);
1197 skb->destructor = netlink_skb_destructor;
1203 * Attach a skb to a netlink socket.
1204 * The caller must hold a reference to the destination socket. On error, the
1205 * reference is dropped. The skb is not send to the destination, just all
1206 * all error checks are performed and memory in the queue is reserved.
1208 * < 0: error. skb freed, reference to sock dropped.
1210 * 1: repeat lookup - reference dropped while waiting for socket memory.
1212 int netlink_attachskb(struct sock *sk, struct sk_buff *skb,
1213 long *timeo, struct sock *ssk)
1215 struct netlink_sock *nlk;
1219 if ((atomic_read(&sk->sk_rmem_alloc) > sk->sk_rcvbuf ||
1220 test_bit(NETLINK_S_CONGESTED, &nlk->state))) {
1221 DECLARE_WAITQUEUE(wait, current);
1223 if (!ssk || netlink_is_kernel(ssk))
1224 netlink_overrun(sk);
1230 __set_current_state(TASK_INTERRUPTIBLE);
1231 add_wait_queue(&nlk->wait, &wait);
1233 if ((atomic_read(&sk->sk_rmem_alloc) > sk->sk_rcvbuf ||
1234 test_bit(NETLINK_S_CONGESTED, &nlk->state)) &&
1235 !sock_flag(sk, SOCK_DEAD))
1236 *timeo = schedule_timeout(*timeo);
1238 __set_current_state(TASK_RUNNING);
1239 remove_wait_queue(&nlk->wait, &wait);
1242 if (signal_pending(current)) {
1244 return sock_intr_errno(*timeo);
1248 netlink_skb_set_owner_r(skb, sk);
1252 static int __netlink_sendskb(struct sock *sk, struct sk_buff *skb)
1256 netlink_deliver_tap(sock_net(sk), skb);
1258 skb_queue_tail(&sk->sk_receive_queue, skb);
1259 sk->sk_data_ready(sk);
1263 int netlink_sendskb(struct sock *sk, struct sk_buff *skb)
1265 int len = __netlink_sendskb(sk, skb);
1271 void netlink_detachskb(struct sock *sk, struct sk_buff *skb)
1277 static struct sk_buff *netlink_trim(struct sk_buff *skb, gfp_t allocation)
1281 WARN_ON(skb->sk != NULL);
1282 delta = skb->end - skb->tail;
1283 if (is_vmalloc_addr(skb->head) || delta * 2 < skb->truesize)
1286 if (skb_shared(skb)) {
1287 struct sk_buff *nskb = skb_clone(skb, allocation);
1294 pskb_expand_head(skb, 0, -delta,
1295 (allocation & ~__GFP_DIRECT_RECLAIM) |
1296 __GFP_NOWARN | __GFP_NORETRY);
1300 static int netlink_unicast_kernel(struct sock *sk, struct sk_buff *skb,
1304 struct netlink_sock *nlk = nlk_sk(sk);
1306 ret = -ECONNREFUSED;
1307 if (nlk->netlink_rcv != NULL) {
1309 netlink_skb_set_owner_r(skb, sk);
1310 NETLINK_CB(skb).sk = ssk;
1311 netlink_deliver_tap_kernel(sk, ssk, skb);
1312 nlk->netlink_rcv(skb);
1321 int netlink_unicast(struct sock *ssk, struct sk_buff *skb,
1322 u32 portid, int nonblock)
1328 skb = netlink_trim(skb, gfp_any());
1330 timeo = sock_sndtimeo(ssk, nonblock);
1332 sk = netlink_getsockbyportid(ssk, portid);
1337 if (netlink_is_kernel(sk))
1338 return netlink_unicast_kernel(sk, skb, ssk);
1340 if (sk_filter(sk, skb)) {
1347 err = netlink_attachskb(sk, skb, &timeo, ssk);
1353 return netlink_sendskb(sk, skb);
1355 EXPORT_SYMBOL(netlink_unicast);
1357 int netlink_has_listeners(struct sock *sk, unsigned int group)
1360 struct listeners *listeners;
1362 BUG_ON(!netlink_is_kernel(sk));
1365 listeners = rcu_dereference(nl_table[sk->sk_protocol].listeners);
1367 if (listeners && group - 1 < nl_table[sk->sk_protocol].groups)
1368 res = test_bit(group - 1, listeners->masks);
1374 EXPORT_SYMBOL_GPL(netlink_has_listeners);
1376 bool netlink_strict_get_check(struct sk_buff *skb)
1378 const struct netlink_sock *nlk = nlk_sk(NETLINK_CB(skb).sk);
1380 return nlk->flags & NETLINK_F_STRICT_CHK;
1382 EXPORT_SYMBOL_GPL(netlink_strict_get_check);
1384 static int netlink_broadcast_deliver(struct sock *sk, struct sk_buff *skb)
1386 struct netlink_sock *nlk = nlk_sk(sk);
1388 if (atomic_read(&sk->sk_rmem_alloc) <= sk->sk_rcvbuf &&
1389 !test_bit(NETLINK_S_CONGESTED, &nlk->state)) {
1390 netlink_skb_set_owner_r(skb, sk);
1391 __netlink_sendskb(sk, skb);
1392 return atomic_read(&sk->sk_rmem_alloc) > (sk->sk_rcvbuf >> 1);
1397 struct netlink_broadcast_data {
1398 struct sock *exclude_sk;
1403 int delivery_failure;
1407 struct sk_buff *skb, *skb2;
1408 int (*tx_filter)(struct sock *dsk, struct sk_buff *skb, void *data);
1412 static void do_one_broadcast(struct sock *sk,
1413 struct netlink_broadcast_data *p)
1415 struct netlink_sock *nlk = nlk_sk(sk);
1418 if (p->exclude_sk == sk)
1421 if (nlk->portid == p->portid || p->group - 1 >= nlk->ngroups ||
1422 !test_bit(p->group - 1, nlk->groups))
1425 if (!net_eq(sock_net(sk), p->net)) {
1426 if (!(nlk->flags & NETLINK_F_LISTEN_ALL_NSID))
1429 if (!peernet_has_id(sock_net(sk), p->net))
1432 if (!file_ns_capable(sk->sk_socket->file, p->net->user_ns,
1438 netlink_overrun(sk);
1443 if (p->skb2 == NULL) {
1444 if (skb_shared(p->skb)) {
1445 p->skb2 = skb_clone(p->skb, p->allocation);
1447 p->skb2 = skb_get(p->skb);
1449 * skb ownership may have been set when
1450 * delivered to a previous socket.
1452 skb_orphan(p->skb2);
1455 if (p->skb2 == NULL) {
1456 netlink_overrun(sk);
1457 /* Clone failed. Notify ALL listeners. */
1459 if (nlk->flags & NETLINK_F_BROADCAST_SEND_ERROR)
1460 p->delivery_failure = 1;
1463 if (p->tx_filter && p->tx_filter(sk, p->skb2, p->tx_data)) {
1468 if (sk_filter(sk, p->skb2)) {
1473 NETLINK_CB(p->skb2).nsid = peernet2id(sock_net(sk), p->net);
1474 if (NETLINK_CB(p->skb2).nsid != NETNSA_NSID_NOT_ASSIGNED)
1475 NETLINK_CB(p->skb2).nsid_is_set = true;
1476 val = netlink_broadcast_deliver(sk, p->skb2);
1478 netlink_overrun(sk);
1479 if (nlk->flags & NETLINK_F_BROADCAST_SEND_ERROR)
1480 p->delivery_failure = 1;
1482 p->congested |= val;
1490 int netlink_broadcast_filtered(struct sock *ssk, struct sk_buff *skb, u32 portid,
1491 u32 group, gfp_t allocation,
1492 int (*filter)(struct sock *dsk, struct sk_buff *skb, void *data),
1495 struct net *net = sock_net(ssk);
1496 struct netlink_broadcast_data info;
1499 skb = netlink_trim(skb, allocation);
1501 info.exclude_sk = ssk;
1503 info.portid = portid;
1506 info.delivery_failure = 0;
1509 info.allocation = allocation;
1512 info.tx_filter = filter;
1513 info.tx_data = filter_data;
1515 /* While we sleep in clone, do not allow to change socket list */
1517 netlink_lock_table();
1519 sk_for_each_bound(sk, &nl_table[ssk->sk_protocol].mc_list)
1520 do_one_broadcast(sk, &info);
1524 netlink_unlock_table();
1526 if (info.delivery_failure) {
1527 kfree_skb(info.skb2);
1530 consume_skb(info.skb2);
1532 if (info.delivered) {
1533 if (info.congested && gfpflags_allow_blocking(allocation))
1539 EXPORT_SYMBOL(netlink_broadcast_filtered);
1541 int netlink_broadcast(struct sock *ssk, struct sk_buff *skb, u32 portid,
1542 u32 group, gfp_t allocation)
1544 return netlink_broadcast_filtered(ssk, skb, portid, group, allocation,
1547 EXPORT_SYMBOL(netlink_broadcast);
1549 struct netlink_set_err_data {
1550 struct sock *exclude_sk;
1556 static int do_one_set_err(struct sock *sk, struct netlink_set_err_data *p)
1558 struct netlink_sock *nlk = nlk_sk(sk);
1561 if (sk == p->exclude_sk)
1564 if (!net_eq(sock_net(sk), sock_net(p->exclude_sk)))
1567 if (nlk->portid == p->portid || p->group - 1 >= nlk->ngroups ||
1568 !test_bit(p->group - 1, nlk->groups))
1571 if (p->code == ENOBUFS && nlk->flags & NETLINK_F_RECV_NO_ENOBUFS) {
1576 sk->sk_err = p->code;
1577 sk->sk_error_report(sk);
1583 * netlink_set_err - report error to broadcast listeners
1584 * @ssk: the kernel netlink socket, as returned by netlink_kernel_create()
1585 * @portid: the PORTID of a process that we want to skip (if any)
1586 * @group: the broadcast group that will notice the error
1587 * @code: error code, must be negative (as usual in kernelspace)
1589 * This function returns the number of broadcast listeners that have set the
1590 * NETLINK_NO_ENOBUFS socket option.
1592 int netlink_set_err(struct sock *ssk, u32 portid, u32 group, int code)
1594 struct netlink_set_err_data info;
1598 info.exclude_sk = ssk;
1599 info.portid = portid;
1601 /* sk->sk_err wants a positive error value */
1604 read_lock(&nl_table_lock);
1606 sk_for_each_bound(sk, &nl_table[ssk->sk_protocol].mc_list)
1607 ret += do_one_set_err(sk, &info);
1609 read_unlock(&nl_table_lock);
1612 EXPORT_SYMBOL(netlink_set_err);
1614 /* must be called with netlink table grabbed */
1615 static void netlink_update_socket_mc(struct netlink_sock *nlk,
1619 int old, new = !!is_new, subscriptions;
1621 old = test_bit(group - 1, nlk->groups);
1622 subscriptions = nlk->subscriptions - old + new;
1624 __set_bit(group - 1, nlk->groups);
1626 __clear_bit(group - 1, nlk->groups);
1627 netlink_update_subscriptions(&nlk->sk, subscriptions);
1628 netlink_update_listeners(&nlk->sk);
1631 static int netlink_setsockopt(struct socket *sock, int level, int optname,
1632 sockptr_t optval, unsigned int optlen)
1634 struct sock *sk = sock->sk;
1635 struct netlink_sock *nlk = nlk_sk(sk);
1636 unsigned int val = 0;
1639 if (level != SOL_NETLINK)
1640 return -ENOPROTOOPT;
1642 if (optlen >= sizeof(int) &&
1643 copy_from_sockptr(&val, optval, sizeof(val)))
1647 case NETLINK_PKTINFO:
1649 nlk->flags |= NETLINK_F_RECV_PKTINFO;
1651 nlk->flags &= ~NETLINK_F_RECV_PKTINFO;
1654 case NETLINK_ADD_MEMBERSHIP:
1655 case NETLINK_DROP_MEMBERSHIP: {
1656 if (!netlink_allowed(sock, NL_CFG_F_NONROOT_RECV))
1658 err = netlink_realloc_groups(sk);
1661 if (!val || val - 1 >= nlk->ngroups)
1663 if (optname == NETLINK_ADD_MEMBERSHIP && nlk->netlink_bind) {
1664 err = nlk->netlink_bind(sock_net(sk), val);
1668 netlink_table_grab();
1669 netlink_update_socket_mc(nlk, val,
1670 optname == NETLINK_ADD_MEMBERSHIP);
1671 netlink_table_ungrab();
1672 if (optname == NETLINK_DROP_MEMBERSHIP && nlk->netlink_unbind)
1673 nlk->netlink_unbind(sock_net(sk), val);
1678 case NETLINK_BROADCAST_ERROR:
1680 nlk->flags |= NETLINK_F_BROADCAST_SEND_ERROR;
1682 nlk->flags &= ~NETLINK_F_BROADCAST_SEND_ERROR;
1685 case NETLINK_NO_ENOBUFS:
1687 nlk->flags |= NETLINK_F_RECV_NO_ENOBUFS;
1688 clear_bit(NETLINK_S_CONGESTED, &nlk->state);
1689 wake_up_interruptible(&nlk->wait);
1691 nlk->flags &= ~NETLINK_F_RECV_NO_ENOBUFS;
1695 case NETLINK_LISTEN_ALL_NSID:
1696 if (!ns_capable(sock_net(sk)->user_ns, CAP_NET_BROADCAST))
1700 nlk->flags |= NETLINK_F_LISTEN_ALL_NSID;
1702 nlk->flags &= ~NETLINK_F_LISTEN_ALL_NSID;
1705 case NETLINK_CAP_ACK:
1707 nlk->flags |= NETLINK_F_CAP_ACK;
1709 nlk->flags &= ~NETLINK_F_CAP_ACK;
1712 case NETLINK_EXT_ACK:
1714 nlk->flags |= NETLINK_F_EXT_ACK;
1716 nlk->flags &= ~NETLINK_F_EXT_ACK;
1719 case NETLINK_GET_STRICT_CHK:
1721 nlk->flags |= NETLINK_F_STRICT_CHK;
1723 nlk->flags &= ~NETLINK_F_STRICT_CHK;
1732 static int netlink_getsockopt(struct socket *sock, int level, int optname,
1733 char __user *optval, int __user *optlen)
1735 struct sock *sk = sock->sk;
1736 struct netlink_sock *nlk = nlk_sk(sk);
1739 if (level != SOL_NETLINK)
1740 return -ENOPROTOOPT;
1742 if (get_user(len, optlen))
1748 case NETLINK_PKTINFO:
1749 if (len < sizeof(int))
1752 val = nlk->flags & NETLINK_F_RECV_PKTINFO ? 1 : 0;
1753 if (put_user(len, optlen) ||
1754 put_user(val, optval))
1758 case NETLINK_BROADCAST_ERROR:
1759 if (len < sizeof(int))
1762 val = nlk->flags & NETLINK_F_BROADCAST_SEND_ERROR ? 1 : 0;
1763 if (put_user(len, optlen) ||
1764 put_user(val, optval))
1768 case NETLINK_NO_ENOBUFS:
1769 if (len < sizeof(int))
1772 val = nlk->flags & NETLINK_F_RECV_NO_ENOBUFS ? 1 : 0;
1773 if (put_user(len, optlen) ||
1774 put_user(val, optval))
1778 case NETLINK_LIST_MEMBERSHIPS: {
1779 int pos, idx, shift;
1782 netlink_lock_table();
1783 for (pos = 0; pos * 8 < nlk->ngroups; pos += sizeof(u32)) {
1784 if (len - pos < sizeof(u32))
1787 idx = pos / sizeof(unsigned long);
1788 shift = (pos % sizeof(unsigned long)) * 8;
1789 if (put_user((u32)(nlk->groups[idx] >> shift),
1790 (u32 __user *)(optval + pos))) {
1795 if (put_user(ALIGN(nlk->ngroups / 8, sizeof(u32)), optlen))
1797 netlink_unlock_table();
1800 case NETLINK_CAP_ACK:
1801 if (len < sizeof(int))
1804 val = nlk->flags & NETLINK_F_CAP_ACK ? 1 : 0;
1805 if (put_user(len, optlen) ||
1806 put_user(val, optval))
1810 case NETLINK_EXT_ACK:
1811 if (len < sizeof(int))
1814 val = nlk->flags & NETLINK_F_EXT_ACK ? 1 : 0;
1815 if (put_user(len, optlen) || put_user(val, optval))
1819 case NETLINK_GET_STRICT_CHK:
1820 if (len < sizeof(int))
1823 val = nlk->flags & NETLINK_F_STRICT_CHK ? 1 : 0;
1824 if (put_user(len, optlen) || put_user(val, optval))
1834 static void netlink_cmsg_recv_pktinfo(struct msghdr *msg, struct sk_buff *skb)
1836 struct nl_pktinfo info;
1838 info.group = NETLINK_CB(skb).dst_group;
1839 put_cmsg(msg, SOL_NETLINK, NETLINK_PKTINFO, sizeof(info), &info);
1842 static void netlink_cmsg_listen_all_nsid(struct sock *sk, struct msghdr *msg,
1843 struct sk_buff *skb)
1845 if (!NETLINK_CB(skb).nsid_is_set)
1848 put_cmsg(msg, SOL_NETLINK, NETLINK_LISTEN_ALL_NSID, sizeof(int),
1849 &NETLINK_CB(skb).nsid);
1852 static int netlink_sendmsg(struct socket *sock, struct msghdr *msg, size_t len)
1854 struct sock *sk = sock->sk;
1855 struct netlink_sock *nlk = nlk_sk(sk);
1856 DECLARE_SOCKADDR(struct sockaddr_nl *, addr, msg->msg_name);
1859 struct sk_buff *skb;
1861 struct scm_cookie scm;
1862 u32 netlink_skb_flags = 0;
1864 if (msg->msg_flags & MSG_OOB)
1867 err = scm_send(sock, msg, &scm, true);
1871 if (msg->msg_namelen) {
1873 if (msg->msg_namelen < sizeof(struct sockaddr_nl))
1875 if (addr->nl_family != AF_NETLINK)
1877 dst_portid = addr->nl_pid;
1878 dst_group = ffs(addr->nl_groups);
1880 if ((dst_group || dst_portid) &&
1881 !netlink_allowed(sock, NL_CFG_F_NONROOT_SEND))
1883 netlink_skb_flags |= NETLINK_SKB_DST;
1885 dst_portid = nlk->dst_portid;
1886 dst_group = nlk->dst_group;
1890 err = netlink_autobind(sock);
1894 /* Ensure nlk is hashed and visible. */
1899 if (len > sk->sk_sndbuf - 32)
1902 skb = netlink_alloc_large_skb(len, dst_group);
1906 NETLINK_CB(skb).portid = nlk->portid;
1907 NETLINK_CB(skb).dst_group = dst_group;
1908 NETLINK_CB(skb).creds = scm.creds;
1909 NETLINK_CB(skb).flags = netlink_skb_flags;
1912 if (memcpy_from_msg(skb_put(skb, len), msg, len)) {
1917 err = security_netlink_send(sk, skb);
1924 refcount_inc(&skb->users);
1925 netlink_broadcast(sk, skb, dst_portid, dst_group, GFP_KERNEL);
1927 err = netlink_unicast(sk, skb, dst_portid, msg->msg_flags & MSG_DONTWAIT);
1934 static int netlink_recvmsg(struct socket *sock, struct msghdr *msg, size_t len,
1937 struct scm_cookie scm;
1938 struct sock *sk = sock->sk;
1939 struct netlink_sock *nlk = nlk_sk(sk);
1940 int noblock = flags & MSG_DONTWAIT;
1942 struct sk_buff *skb, *data_skb;
1945 if (flags & MSG_OOB)
1950 skb = skb_recv_datagram(sk, flags, noblock, &err);
1956 #ifdef CONFIG_COMPAT_NETLINK_MESSAGES
1957 if (unlikely(skb_shinfo(skb)->frag_list)) {
1959 * If this skb has a frag_list, then here that means that we
1960 * will have to use the frag_list skb's data for compat tasks
1961 * and the regular skb's data for normal (non-compat) tasks.
1963 * If we need to send the compat skb, assign it to the
1964 * 'data_skb' variable so that it will be used below for data
1965 * copying. We keep 'skb' for everything else, including
1966 * freeing both later.
1968 if (flags & MSG_CMSG_COMPAT)
1969 data_skb = skb_shinfo(skb)->frag_list;
1973 /* Record the max length of recvmsg() calls for future allocations */
1974 nlk->max_recvmsg_len = max(nlk->max_recvmsg_len, len);
1975 nlk->max_recvmsg_len = min_t(size_t, nlk->max_recvmsg_len,
1976 SKB_WITH_OVERHEAD(32768));
1978 copied = data_skb->len;
1980 msg->msg_flags |= MSG_TRUNC;
1984 skb_reset_transport_header(data_skb);
1985 err = skb_copy_datagram_msg(data_skb, 0, msg, copied);
1987 if (msg->msg_name) {
1988 DECLARE_SOCKADDR(struct sockaddr_nl *, addr, msg->msg_name);
1989 addr->nl_family = AF_NETLINK;
1991 addr->nl_pid = NETLINK_CB(skb).portid;
1992 addr->nl_groups = netlink_group_mask(NETLINK_CB(skb).dst_group);
1993 msg->msg_namelen = sizeof(*addr);
1996 if (nlk->flags & NETLINK_F_RECV_PKTINFO)
1997 netlink_cmsg_recv_pktinfo(msg, skb);
1998 if (nlk->flags & NETLINK_F_LISTEN_ALL_NSID)
1999 netlink_cmsg_listen_all_nsid(sk, msg, skb);
2001 memset(&scm, 0, sizeof(scm));
2002 scm.creds = *NETLINK_CREDS(skb);
2003 if (flags & MSG_TRUNC)
2004 copied = data_skb->len;
2006 skb_free_datagram(sk, skb);
2008 if (nlk->cb_running &&
2009 atomic_read(&sk->sk_rmem_alloc) <= sk->sk_rcvbuf / 2) {
2010 ret = netlink_dump(sk);
2013 sk->sk_error_report(sk);
2017 scm_recv(sock, msg, &scm, flags);
2019 netlink_rcv_wake(sk);
2020 return err ? : copied;
2023 static void netlink_data_ready(struct sock *sk)
2029 * We export these functions to other modules. They provide a
2030 * complete set of kernel non-blocking support for message
2035 __netlink_kernel_create(struct net *net, int unit, struct module *module,
2036 struct netlink_kernel_cfg *cfg)
2038 struct socket *sock;
2040 struct netlink_sock *nlk;
2041 struct listeners *listeners = NULL;
2042 struct mutex *cb_mutex = cfg ? cfg->cb_mutex : NULL;
2043 unsigned int groups;
2047 if (unit < 0 || unit >= MAX_LINKS)
2050 if (sock_create_lite(PF_NETLINK, SOCK_DGRAM, unit, &sock))
2053 if (__netlink_create(net, sock, cb_mutex, unit, 1) < 0)
2054 goto out_sock_release_nosk;
2058 if (!cfg || cfg->groups < 32)
2061 groups = cfg->groups;
2063 listeners = kzalloc(sizeof(*listeners) + NLGRPSZ(groups), GFP_KERNEL);
2065 goto out_sock_release;
2067 sk->sk_data_ready = netlink_data_ready;
2068 if (cfg && cfg->input)
2069 nlk_sk(sk)->netlink_rcv = cfg->input;
2071 if (netlink_insert(sk, 0))
2072 goto out_sock_release;
2075 nlk->flags |= NETLINK_F_KERNEL_SOCKET;
2077 netlink_table_grab();
2078 if (!nl_table[unit].registered) {
2079 nl_table[unit].groups = groups;
2080 rcu_assign_pointer(nl_table[unit].listeners, listeners);
2081 nl_table[unit].cb_mutex = cb_mutex;
2082 nl_table[unit].module = module;
2084 nl_table[unit].bind = cfg->bind;
2085 nl_table[unit].unbind = cfg->unbind;
2086 nl_table[unit].flags = cfg->flags;
2088 nl_table[unit].compare = cfg->compare;
2090 nl_table[unit].registered = 1;
2093 nl_table[unit].registered++;
2095 netlink_table_ungrab();
2100 netlink_kernel_release(sk);
2103 out_sock_release_nosk:
2107 EXPORT_SYMBOL(__netlink_kernel_create);
2110 netlink_kernel_release(struct sock *sk)
2112 if (sk == NULL || sk->sk_socket == NULL)
2115 sock_release(sk->sk_socket);
2117 EXPORT_SYMBOL(netlink_kernel_release);
2119 int __netlink_change_ngroups(struct sock *sk, unsigned int groups)
2121 struct listeners *new, *old;
2122 struct netlink_table *tbl = &nl_table[sk->sk_protocol];
2127 if (NLGRPSZ(tbl->groups) < NLGRPSZ(groups)) {
2128 new = kzalloc(sizeof(*new) + NLGRPSZ(groups), GFP_ATOMIC);
2131 old = nl_deref_protected(tbl->listeners);
2132 memcpy(new->masks, old->masks, NLGRPSZ(tbl->groups));
2133 rcu_assign_pointer(tbl->listeners, new);
2135 kfree_rcu(old, rcu);
2137 tbl->groups = groups;
2143 * netlink_change_ngroups - change number of multicast groups
2145 * This changes the number of multicast groups that are available
2146 * on a certain netlink family. Note that it is not possible to
2147 * change the number of groups to below 32. Also note that it does
2148 * not implicitly call netlink_clear_multicast_users() when the
2149 * number of groups is reduced.
2151 * @sk: The kernel netlink socket, as returned by netlink_kernel_create().
2152 * @groups: The new number of groups.
2154 int netlink_change_ngroups(struct sock *sk, unsigned int groups)
2158 netlink_table_grab();
2159 err = __netlink_change_ngroups(sk, groups);
2160 netlink_table_ungrab();
2165 void __netlink_clear_multicast_users(struct sock *ksk, unsigned int group)
2168 struct netlink_table *tbl = &nl_table[ksk->sk_protocol];
2170 sk_for_each_bound(sk, &tbl->mc_list)
2171 netlink_update_socket_mc(nlk_sk(sk), group, 0);
2175 __nlmsg_put(struct sk_buff *skb, u32 portid, u32 seq, int type, int len, int flags)
2177 struct nlmsghdr *nlh;
2178 int size = nlmsg_msg_size(len);
2180 nlh = skb_put(skb, NLMSG_ALIGN(size));
2181 nlh->nlmsg_type = type;
2182 nlh->nlmsg_len = size;
2183 nlh->nlmsg_flags = flags;
2184 nlh->nlmsg_pid = portid;
2185 nlh->nlmsg_seq = seq;
2186 if (!__builtin_constant_p(size) || NLMSG_ALIGN(size) - size != 0)
2187 memset(nlmsg_data(nlh) + len, 0, NLMSG_ALIGN(size) - size);
2190 EXPORT_SYMBOL(__nlmsg_put);
2193 * It looks a bit ugly.
2194 * It would be better to create kernel thread.
2197 static int netlink_dump_done(struct netlink_sock *nlk, struct sk_buff *skb,
2198 struct netlink_callback *cb,
2199 struct netlink_ext_ack *extack)
2201 struct nlmsghdr *nlh;
2203 nlh = nlmsg_put_answer(skb, cb, NLMSG_DONE, sizeof(nlk->dump_done_errno),
2204 NLM_F_MULTI | cb->answer_flags);
2208 nl_dump_check_consistent(cb, nlh);
2209 memcpy(nlmsg_data(nlh), &nlk->dump_done_errno, sizeof(nlk->dump_done_errno));
2211 if (extack->_msg && nlk->flags & NETLINK_F_EXT_ACK) {
2212 nlh->nlmsg_flags |= NLM_F_ACK_TLVS;
2213 if (!nla_put_string(skb, NLMSGERR_ATTR_MSG, extack->_msg))
2214 nlmsg_end(skb, nlh);
2220 static int netlink_dump(struct sock *sk)
2222 struct netlink_sock *nlk = nlk_sk(sk);
2223 struct netlink_ext_ack extack = {};
2224 struct netlink_callback *cb;
2225 struct sk_buff *skb = NULL;
2226 struct module *module;
2231 mutex_lock(nlk->cb_mutex);
2232 if (!nlk->cb_running) {
2237 if (atomic_read(&sk->sk_rmem_alloc) >= sk->sk_rcvbuf)
2240 /* NLMSG_GOODSIZE is small to avoid high order allocations being
2241 * required, but it makes sense to _attempt_ a 16K bytes allocation
2242 * to reduce number of system calls on dump operations, if user
2243 * ever provided a big enough buffer.
2246 alloc_min_size = max_t(int, cb->min_dump_alloc, NLMSG_GOODSIZE);
2248 if (alloc_min_size < nlk->max_recvmsg_len) {
2249 alloc_size = nlk->max_recvmsg_len;
2250 skb = alloc_skb(alloc_size,
2251 (GFP_KERNEL & ~__GFP_DIRECT_RECLAIM) |
2252 __GFP_NOWARN | __GFP_NORETRY);
2255 alloc_size = alloc_min_size;
2256 skb = alloc_skb(alloc_size, GFP_KERNEL);
2261 /* Trim skb to allocated size. User is expected to provide buffer as
2262 * large as max(min_dump_alloc, 16KiB (mac_recvmsg_len capped at
2263 * netlink_recvmsg())). dump will pack as many smaller messages as
2264 * could fit within the allocated skb. skb is typically allocated
2265 * with larger space than required (could be as much as near 2x the
2266 * requested size with align to next power of 2 approach). Allowing
2267 * dump to use the excess space makes it difficult for a user to have a
2268 * reasonable static buffer based on the expected largest dump of a
2269 * single netdev. The outcome is MSG_TRUNC error.
2271 skb_reserve(skb, skb_tailroom(skb) - alloc_size);
2272 netlink_skb_set_owner_r(skb, sk);
2274 if (nlk->dump_done_errno > 0) {
2275 cb->extack = &extack;
2276 nlk->dump_done_errno = cb->dump(skb, cb);
2280 if (nlk->dump_done_errno > 0 ||
2281 skb_tailroom(skb) < nlmsg_total_size(sizeof(nlk->dump_done_errno))) {
2282 mutex_unlock(nlk->cb_mutex);
2284 if (sk_filter(sk, skb))
2287 __netlink_sendskb(sk, skb);
2291 if (netlink_dump_done(nlk, skb, cb, &extack))
2294 #ifdef CONFIG_COMPAT_NETLINK_MESSAGES
2295 /* frag_list skb's data is used for compat tasks
2296 * and the regular skb's data for normal (non-compat) tasks.
2297 * See netlink_recvmsg().
2299 if (unlikely(skb_shinfo(skb)->frag_list)) {
2300 if (netlink_dump_done(nlk, skb_shinfo(skb)->frag_list, cb, &extack))
2305 if (sk_filter(sk, skb))
2308 __netlink_sendskb(sk, skb);
2313 nlk->cb_running = false;
2314 module = cb->module;
2316 mutex_unlock(nlk->cb_mutex);
2322 mutex_unlock(nlk->cb_mutex);
2327 int __netlink_dump_start(struct sock *ssk, struct sk_buff *skb,
2328 const struct nlmsghdr *nlh,
2329 struct netlink_dump_control *control)
2331 struct netlink_sock *nlk, *nlk2;
2332 struct netlink_callback *cb;
2336 refcount_inc(&skb->users);
2338 sk = netlink_lookup(sock_net(ssk), ssk->sk_protocol, NETLINK_CB(skb).portid);
2340 ret = -ECONNREFUSED;
2345 mutex_lock(nlk->cb_mutex);
2346 /* A dump is in progress... */
2347 if (nlk->cb_running) {
2351 /* add reference of module which cb->dump belongs to */
2352 if (!try_module_get(control->module)) {
2353 ret = -EPROTONOSUPPORT;
2358 memset(cb, 0, sizeof(*cb));
2359 cb->dump = control->dump;
2360 cb->done = control->done;
2362 cb->data = control->data;
2363 cb->module = control->module;
2364 cb->min_dump_alloc = control->min_dump_alloc;
2367 nlk2 = nlk_sk(NETLINK_CB(skb).sk);
2368 cb->strict_check = !!(nlk2->flags & NETLINK_F_STRICT_CHK);
2370 if (control->start) {
2371 ret = control->start(cb);
2376 nlk->cb_running = true;
2377 nlk->dump_done_errno = INT_MAX;
2379 mutex_unlock(nlk->cb_mutex);
2381 ret = netlink_dump(sk);
2388 /* We successfully started a dump, by returning -EINTR we
2389 * signal not to send ACK even if it was requested.
2394 module_put(control->module);
2397 mutex_unlock(nlk->cb_mutex);
2402 EXPORT_SYMBOL(__netlink_dump_start);
2404 void netlink_ack(struct sk_buff *in_skb, struct nlmsghdr *nlh, int err,
2405 const struct netlink_ext_ack *extack)
2407 struct sk_buff *skb;
2408 struct nlmsghdr *rep;
2409 struct nlmsgerr *errmsg;
2410 size_t payload = sizeof(*errmsg);
2412 struct netlink_sock *nlk = nlk_sk(NETLINK_CB(in_skb).sk);
2413 unsigned int flags = 0;
2414 bool nlk_has_extack = nlk->flags & NETLINK_F_EXT_ACK;
2416 /* Error messages get the original request appened, unless the user
2417 * requests to cap the error message, and get extra error data if
2420 if (nlk_has_extack && extack && extack->_msg)
2421 tlvlen += nla_total_size(strlen(extack->_msg) + 1);
2423 if (err && !(nlk->flags & NETLINK_F_CAP_ACK))
2424 payload += nlmsg_len(nlh);
2426 flags |= NLM_F_CAPPED;
2427 if (err && nlk_has_extack && extack && extack->bad_attr)
2428 tlvlen += nla_total_size(sizeof(u32));
2429 if (nlk_has_extack && extack && extack->cookie_len)
2430 tlvlen += nla_total_size(extack->cookie_len);
2431 if (err && nlk_has_extack && extack && extack->policy)
2432 tlvlen += netlink_policy_dump_attr_size_estimate(extack->policy);
2435 flags |= NLM_F_ACK_TLVS;
2437 skb = nlmsg_new(payload + tlvlen, GFP_KERNEL);
2439 NETLINK_CB(in_skb).sk->sk_err = ENOBUFS;
2440 NETLINK_CB(in_skb).sk->sk_error_report(NETLINK_CB(in_skb).sk);
2444 rep = __nlmsg_put(skb, NETLINK_CB(in_skb).portid, nlh->nlmsg_seq,
2445 NLMSG_ERROR, payload, flags);
2446 errmsg = nlmsg_data(rep);
2447 errmsg->error = err;
2448 memcpy(&errmsg->msg, nlh, payload > sizeof(*errmsg) ? nlh->nlmsg_len : sizeof(*nlh));
2450 if (nlk_has_extack && extack) {
2452 WARN_ON(nla_put_string(skb, NLMSGERR_ATTR_MSG,
2455 if (err && extack->bad_attr &&
2456 !WARN_ON((u8 *)extack->bad_attr < in_skb->data ||
2457 (u8 *)extack->bad_attr >= in_skb->data +
2459 WARN_ON(nla_put_u32(skb, NLMSGERR_ATTR_OFFS,
2460 (u8 *)extack->bad_attr -
2462 if (extack->cookie_len)
2463 WARN_ON(nla_put(skb, NLMSGERR_ATTR_COOKIE,
2464 extack->cookie_len, extack->cookie));
2466 netlink_policy_dump_write_attr(skb, extack->policy,
2467 NLMSGERR_ATTR_POLICY);
2470 nlmsg_end(skb, rep);
2472 netlink_unicast(in_skb->sk, skb, NETLINK_CB(in_skb).portid, MSG_DONTWAIT);
2474 EXPORT_SYMBOL(netlink_ack);
2476 int netlink_rcv_skb(struct sk_buff *skb, int (*cb)(struct sk_buff *,
2478 struct netlink_ext_ack *))
2480 struct netlink_ext_ack extack;
2481 struct nlmsghdr *nlh;
2484 while (skb->len >= nlmsg_total_size(0)) {
2487 memset(&extack, 0, sizeof(extack));
2488 nlh = nlmsg_hdr(skb);
2491 if (nlh->nlmsg_len < NLMSG_HDRLEN || skb->len < nlh->nlmsg_len)
2494 /* Only requests are handled by the kernel */
2495 if (!(nlh->nlmsg_flags & NLM_F_REQUEST))
2498 /* Skip control messages */
2499 if (nlh->nlmsg_type < NLMSG_MIN_TYPE)
2502 err = cb(skb, nlh, &extack);
2507 if (nlh->nlmsg_flags & NLM_F_ACK || err)
2508 netlink_ack(skb, nlh, err, &extack);
2511 msglen = NLMSG_ALIGN(nlh->nlmsg_len);
2512 if (msglen > skb->len)
2514 skb_pull(skb, msglen);
2519 EXPORT_SYMBOL(netlink_rcv_skb);
2522 * nlmsg_notify - send a notification netlink message
2523 * @sk: netlink socket to use
2524 * @skb: notification message
2525 * @portid: destination netlink portid for reports or 0
2526 * @group: destination multicast group or 0
2527 * @report: 1 to report back, 0 to disable
2528 * @flags: allocation flags
2530 int nlmsg_notify(struct sock *sk, struct sk_buff *skb, u32 portid,
2531 unsigned int group, int report, gfp_t flags)
2536 int exclude_portid = 0;
2539 refcount_inc(&skb->users);
2540 exclude_portid = portid;
2543 /* errors reported via destination sk->sk_err, but propagate
2544 * delivery errors if NETLINK_BROADCAST_ERROR flag is set */
2545 err = nlmsg_multicast(sk, skb, exclude_portid, group, flags);
2551 err2 = nlmsg_unicast(sk, skb, portid);
2552 if (!err || err == -ESRCH)
2558 EXPORT_SYMBOL(nlmsg_notify);
2560 #ifdef CONFIG_PROC_FS
2561 struct nl_seq_iter {
2562 struct seq_net_private p;
2563 struct rhashtable_iter hti;
2567 static void netlink_walk_start(struct nl_seq_iter *iter)
2569 rhashtable_walk_enter(&nl_table[iter->link].hash, &iter->hti);
2570 rhashtable_walk_start(&iter->hti);
2573 static void netlink_walk_stop(struct nl_seq_iter *iter)
2575 rhashtable_walk_stop(&iter->hti);
2576 rhashtable_walk_exit(&iter->hti);
2579 static void *__netlink_seq_next(struct seq_file *seq)
2581 struct nl_seq_iter *iter = seq->private;
2582 struct netlink_sock *nlk;
2586 nlk = rhashtable_walk_next(&iter->hti);
2589 if (PTR_ERR(nlk) == -EAGAIN)
2598 netlink_walk_stop(iter);
2599 if (++iter->link >= MAX_LINKS)
2602 netlink_walk_start(iter);
2604 } while (sock_net(&nlk->sk) != seq_file_net(seq));
2609 static void *netlink_seq_start(struct seq_file *seq, loff_t *posp)
2612 struct nl_seq_iter *iter = seq->private;
2613 void *obj = SEQ_START_TOKEN;
2618 netlink_walk_start(iter);
2620 for (pos = *posp; pos && obj && !IS_ERR(obj); pos--)
2621 obj = __netlink_seq_next(seq);
2626 static void *netlink_seq_next(struct seq_file *seq, void *v, loff_t *pos)
2629 return __netlink_seq_next(seq);
2632 static void netlink_native_seq_stop(struct seq_file *seq, void *v)
2634 struct nl_seq_iter *iter = seq->private;
2636 if (iter->link >= MAX_LINKS)
2639 netlink_walk_stop(iter);
2643 static int netlink_native_seq_show(struct seq_file *seq, void *v)
2645 if (v == SEQ_START_TOKEN) {
2647 "sk Eth Pid Groups "
2648 "Rmem Wmem Dump Locks Drops Inode\n");
2651 struct netlink_sock *nlk = nlk_sk(s);
2653 seq_printf(seq, "%pK %-3d %-10u %08x %-8d %-8d %-5d %-8d %-8u %-8lu\n",
2657 nlk->groups ? (u32)nlk->groups[0] : 0,
2658 sk_rmem_alloc_get(s),
2659 sk_wmem_alloc_get(s),
2661 refcount_read(&s->sk_refcnt),
2662 atomic_read(&s->sk_drops),
2670 #ifdef CONFIG_BPF_SYSCALL
2671 struct bpf_iter__netlink {
2672 __bpf_md_ptr(struct bpf_iter_meta *, meta);
2673 __bpf_md_ptr(struct netlink_sock *, sk);
2676 DEFINE_BPF_ITER_FUNC(netlink, struct bpf_iter_meta *meta, struct netlink_sock *sk)
2678 static int netlink_prog_seq_show(struct bpf_prog *prog,
2679 struct bpf_iter_meta *meta,
2682 struct bpf_iter__netlink ctx;
2684 meta->seq_num--; /* skip SEQ_START_TOKEN */
2686 ctx.sk = nlk_sk((struct sock *)v);
2687 return bpf_iter_run_prog(prog, &ctx);
2690 static int netlink_seq_show(struct seq_file *seq, void *v)
2692 struct bpf_iter_meta meta;
2693 struct bpf_prog *prog;
2696 prog = bpf_iter_get_info(&meta, false);
2698 return netlink_native_seq_show(seq, v);
2700 if (v != SEQ_START_TOKEN)
2701 return netlink_prog_seq_show(prog, &meta, v);
2706 static void netlink_seq_stop(struct seq_file *seq, void *v)
2708 struct bpf_iter_meta meta;
2709 struct bpf_prog *prog;
2713 prog = bpf_iter_get_info(&meta, true);
2715 (void)netlink_prog_seq_show(prog, &meta, v);
2718 netlink_native_seq_stop(seq, v);
2721 static int netlink_seq_show(struct seq_file *seq, void *v)
2723 return netlink_native_seq_show(seq, v);
2726 static void netlink_seq_stop(struct seq_file *seq, void *v)
2728 netlink_native_seq_stop(seq, v);
2732 static const struct seq_operations netlink_seq_ops = {
2733 .start = netlink_seq_start,
2734 .next = netlink_seq_next,
2735 .stop = netlink_seq_stop,
2736 .show = netlink_seq_show,
2740 int netlink_register_notifier(struct notifier_block *nb)
2742 return blocking_notifier_chain_register(&netlink_chain, nb);
2744 EXPORT_SYMBOL(netlink_register_notifier);
2746 int netlink_unregister_notifier(struct notifier_block *nb)
2748 return blocking_notifier_chain_unregister(&netlink_chain, nb);
2750 EXPORT_SYMBOL(netlink_unregister_notifier);
2752 static const struct proto_ops netlink_ops = {
2753 .family = PF_NETLINK,
2754 .owner = THIS_MODULE,
2755 .release = netlink_release,
2756 .bind = netlink_bind,
2757 .connect = netlink_connect,
2758 .socketpair = sock_no_socketpair,
2759 .accept = sock_no_accept,
2760 .getname = netlink_getname,
2761 .poll = datagram_poll,
2762 .ioctl = netlink_ioctl,
2763 .listen = sock_no_listen,
2764 .shutdown = sock_no_shutdown,
2765 .setsockopt = netlink_setsockopt,
2766 .getsockopt = netlink_getsockopt,
2767 .sendmsg = netlink_sendmsg,
2768 .recvmsg = netlink_recvmsg,
2769 .mmap = sock_no_mmap,
2770 .sendpage = sock_no_sendpage,
2773 static const struct net_proto_family netlink_family_ops = {
2774 .family = PF_NETLINK,
2775 .create = netlink_create,
2776 .owner = THIS_MODULE, /* for consistency 8) */
2779 static int __net_init netlink_net_init(struct net *net)
2781 #ifdef CONFIG_PROC_FS
2782 if (!proc_create_net("netlink", 0, net->proc_net, &netlink_seq_ops,
2783 sizeof(struct nl_seq_iter)))
2789 static void __net_exit netlink_net_exit(struct net *net)
2791 #ifdef CONFIG_PROC_FS
2792 remove_proc_entry("netlink", net->proc_net);
2796 static void __init netlink_add_usersock_entry(void)
2798 struct listeners *listeners;
2801 listeners = kzalloc(sizeof(*listeners) + NLGRPSZ(groups), GFP_KERNEL);
2803 panic("netlink_add_usersock_entry: Cannot allocate listeners\n");
2805 netlink_table_grab();
2807 nl_table[NETLINK_USERSOCK].groups = groups;
2808 rcu_assign_pointer(nl_table[NETLINK_USERSOCK].listeners, listeners);
2809 nl_table[NETLINK_USERSOCK].module = THIS_MODULE;
2810 nl_table[NETLINK_USERSOCK].registered = 1;
2811 nl_table[NETLINK_USERSOCK].flags = NL_CFG_F_NONROOT_SEND;
2813 netlink_table_ungrab();
2816 static struct pernet_operations __net_initdata netlink_net_ops = {
2817 .init = netlink_net_init,
2818 .exit = netlink_net_exit,
2821 static inline u32 netlink_hash(const void *data, u32 len, u32 seed)
2823 const struct netlink_sock *nlk = data;
2824 struct netlink_compare_arg arg;
2826 netlink_compare_arg_init(&arg, sock_net(&nlk->sk), nlk->portid);
2827 return jhash2((u32 *)&arg, netlink_compare_arg_len / sizeof(u32), seed);
2830 static const struct rhashtable_params netlink_rhashtable_params = {
2831 .head_offset = offsetof(struct netlink_sock, node),
2832 .key_len = netlink_compare_arg_len,
2833 .obj_hashfn = netlink_hash,
2834 .obj_cmpfn = netlink_compare,
2835 .automatic_shrinking = true,
2838 #if defined(CONFIG_BPF_SYSCALL) && defined(CONFIG_PROC_FS)
2839 BTF_ID_LIST(btf_netlink_sock_id)
2840 BTF_ID(struct, netlink_sock)
2842 static const struct bpf_iter_seq_info netlink_seq_info = {
2843 .seq_ops = &netlink_seq_ops,
2844 .init_seq_private = bpf_iter_init_seq_net,
2845 .fini_seq_private = bpf_iter_fini_seq_net,
2846 .seq_priv_size = sizeof(struct nl_seq_iter),
2849 static struct bpf_iter_reg netlink_reg_info = {
2850 .target = "netlink",
2851 .ctx_arg_info_size = 1,
2853 { offsetof(struct bpf_iter__netlink, sk),
2854 PTR_TO_BTF_ID_OR_NULL },
2856 .seq_info = &netlink_seq_info,
2859 static int __init bpf_iter_register(void)
2861 netlink_reg_info.ctx_arg_info[0].btf_id = *btf_netlink_sock_id;
2862 return bpf_iter_reg_target(&netlink_reg_info);
2866 static int __init netlink_proto_init(void)
2869 int err = proto_register(&netlink_proto, 0);
2874 #if defined(CONFIG_BPF_SYSCALL) && defined(CONFIG_PROC_FS)
2875 err = bpf_iter_register();
2880 BUILD_BUG_ON(sizeof(struct netlink_skb_parms) > sizeof_field(struct sk_buff, cb));
2882 nl_table = kcalloc(MAX_LINKS, sizeof(*nl_table), GFP_KERNEL);
2886 for (i = 0; i < MAX_LINKS; i++) {
2887 if (rhashtable_init(&nl_table[i].hash,
2888 &netlink_rhashtable_params) < 0) {
2890 rhashtable_destroy(&nl_table[i].hash);
2896 netlink_add_usersock_entry();
2898 sock_register(&netlink_family_ops);
2899 register_pernet_subsys(&netlink_net_ops);
2900 register_pernet_subsys(&netlink_tap_net_ops);
2901 /* The netlink device handler may be needed early. */
2906 panic("netlink_init: Cannot allocate nl_table\n");
2909 core_initcall(netlink_proto_init);