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;
357 atomic_inc(&sk->sk_drops);
360 static void netlink_rcv_wake(struct sock *sk)
362 struct netlink_sock *nlk = nlk_sk(sk);
364 if (skb_queue_empty_lockless(&sk->sk_receive_queue))
365 clear_bit(NETLINK_S_CONGESTED, &nlk->state);
366 if (!test_bit(NETLINK_S_CONGESTED, &nlk->state))
367 wake_up_interruptible(&nlk->wait);
370 static void netlink_skb_destructor(struct sk_buff *skb)
372 if (is_vmalloc_addr(skb->head)) {
374 !atomic_dec_return(&(skb_shinfo(skb)->dataref)))
383 static void netlink_skb_set_owner_r(struct sk_buff *skb, struct sock *sk)
385 WARN_ON(skb->sk != NULL);
387 skb->destructor = netlink_skb_destructor;
388 atomic_add(skb->truesize, &sk->sk_rmem_alloc);
389 sk_mem_charge(sk, skb->truesize);
392 static void netlink_sock_destruct(struct sock *sk)
394 struct netlink_sock *nlk = nlk_sk(sk);
396 if (nlk->cb_running) {
398 nlk->cb.done(&nlk->cb);
399 module_put(nlk->cb.module);
400 kfree_skb(nlk->cb.skb);
403 skb_queue_purge(&sk->sk_receive_queue);
405 if (!sock_flag(sk, SOCK_DEAD)) {
406 printk(KERN_ERR "Freeing alive netlink socket %p\n", sk);
410 WARN_ON(atomic_read(&sk->sk_rmem_alloc));
411 WARN_ON(refcount_read(&sk->sk_wmem_alloc));
412 WARN_ON(nlk_sk(sk)->groups);
415 static void netlink_sock_destruct_work(struct work_struct *work)
417 struct netlink_sock *nlk = container_of(work, struct netlink_sock,
423 /* This lock without WQ_FLAG_EXCLUSIVE is good on UP and it is _very_ bad on
424 * SMP. Look, when several writers sleep and reader wakes them up, all but one
425 * immediately hit write lock and grab all the cpus. Exclusive sleep solves
426 * this, _but_ remember, it adds useless work on UP machines.
429 void netlink_table_grab(void)
430 __acquires(nl_table_lock)
434 write_lock_irq(&nl_table_lock);
436 if (atomic_read(&nl_table_users)) {
437 DECLARE_WAITQUEUE(wait, current);
439 add_wait_queue_exclusive(&nl_table_wait, &wait);
441 set_current_state(TASK_UNINTERRUPTIBLE);
442 if (atomic_read(&nl_table_users) == 0)
444 write_unlock_irq(&nl_table_lock);
446 write_lock_irq(&nl_table_lock);
449 __set_current_state(TASK_RUNNING);
450 remove_wait_queue(&nl_table_wait, &wait);
454 void netlink_table_ungrab(void)
455 __releases(nl_table_lock)
457 write_unlock_irq(&nl_table_lock);
458 wake_up(&nl_table_wait);
462 netlink_lock_table(void)
466 /* read_lock() synchronizes us to netlink_table_grab */
468 read_lock_irqsave(&nl_table_lock, flags);
469 atomic_inc(&nl_table_users);
470 read_unlock_irqrestore(&nl_table_lock, flags);
474 netlink_unlock_table(void)
476 if (atomic_dec_and_test(&nl_table_users))
477 wake_up(&nl_table_wait);
480 struct netlink_compare_arg
486 /* Doing sizeof directly may yield 4 extra bytes on 64-bit. */
487 #define netlink_compare_arg_len \
488 (offsetof(struct netlink_compare_arg, portid) + sizeof(u32))
490 static inline int netlink_compare(struct rhashtable_compare_arg *arg,
493 const struct netlink_compare_arg *x = arg->key;
494 const struct netlink_sock *nlk = ptr;
496 return nlk->portid != x->portid ||
497 !net_eq(sock_net(&nlk->sk), read_pnet(&x->pnet));
500 static void netlink_compare_arg_init(struct netlink_compare_arg *arg,
501 struct net *net, u32 portid)
503 memset(arg, 0, sizeof(*arg));
504 write_pnet(&arg->pnet, net);
505 arg->portid = portid;
508 static struct sock *__netlink_lookup(struct netlink_table *table, u32 portid,
511 struct netlink_compare_arg arg;
513 netlink_compare_arg_init(&arg, net, portid);
514 return rhashtable_lookup_fast(&table->hash, &arg,
515 netlink_rhashtable_params);
518 static int __netlink_insert(struct netlink_table *table, struct sock *sk)
520 struct netlink_compare_arg arg;
522 netlink_compare_arg_init(&arg, sock_net(sk), nlk_sk(sk)->portid);
523 return rhashtable_lookup_insert_key(&table->hash, &arg,
525 netlink_rhashtable_params);
528 static struct sock *netlink_lookup(struct net *net, int protocol, u32 portid)
530 struct netlink_table *table = &nl_table[protocol];
534 sk = __netlink_lookup(table, portid, net);
542 static const struct proto_ops netlink_ops;
545 netlink_update_listeners(struct sock *sk)
547 struct netlink_table *tbl = &nl_table[sk->sk_protocol];
550 struct listeners *listeners;
552 listeners = nl_deref_protected(tbl->listeners);
556 for (i = 0; i < NLGRPLONGS(tbl->groups); i++) {
558 sk_for_each_bound(sk, &tbl->mc_list) {
559 if (i < NLGRPLONGS(nlk_sk(sk)->ngroups))
560 mask |= nlk_sk(sk)->groups[i];
562 listeners->masks[i] = mask;
564 /* this function is only called with the netlink table "grabbed", which
565 * makes sure updates are visible before bind or setsockopt return. */
568 static int netlink_insert(struct sock *sk, u32 portid)
570 struct netlink_table *table = &nl_table[sk->sk_protocol];
575 err = nlk_sk(sk)->portid == portid ? 0 : -EBUSY;
576 if (nlk_sk(sk)->bound)
579 nlk_sk(sk)->portid = portid;
582 err = __netlink_insert(table, sk);
584 /* In case the hashtable backend returns with -EBUSY
585 * from here, it must not escape to the caller.
587 if (unlikely(err == -EBUSY))
595 /* We need to ensure that the socket is hashed and visible. */
597 nlk_sk(sk)->bound = portid;
604 static void netlink_remove(struct sock *sk)
606 struct netlink_table *table;
608 table = &nl_table[sk->sk_protocol];
609 if (!rhashtable_remove_fast(&table->hash, &nlk_sk(sk)->node,
610 netlink_rhashtable_params)) {
611 WARN_ON(refcount_read(&sk->sk_refcnt) == 1);
615 netlink_table_grab();
616 if (nlk_sk(sk)->subscriptions) {
617 __sk_del_bind_node(sk);
618 netlink_update_listeners(sk);
620 if (sk->sk_protocol == NETLINK_GENERIC)
621 atomic_inc(&genl_sk_destructing_cnt);
622 netlink_table_ungrab();
625 static struct proto netlink_proto = {
627 .owner = THIS_MODULE,
628 .obj_size = sizeof(struct netlink_sock),
631 static int __netlink_create(struct net *net, struct socket *sock,
632 struct mutex *cb_mutex, int protocol,
636 struct netlink_sock *nlk;
638 sock->ops = &netlink_ops;
640 sk = sk_alloc(net, PF_NETLINK, GFP_KERNEL, &netlink_proto, kern);
644 sock_init_data(sock, sk);
648 nlk->cb_mutex = cb_mutex;
650 nlk->cb_mutex = &nlk->cb_def_mutex;
651 mutex_init(nlk->cb_mutex);
652 lockdep_set_class_and_name(nlk->cb_mutex,
653 nlk_cb_mutex_keys + protocol,
654 nlk_cb_mutex_key_strings[protocol]);
656 init_waitqueue_head(&nlk->wait);
658 sk->sk_destruct = netlink_sock_destruct;
659 sk->sk_protocol = protocol;
663 static int netlink_create(struct net *net, struct socket *sock, int protocol,
666 struct module *module = NULL;
667 struct mutex *cb_mutex;
668 struct netlink_sock *nlk;
669 int (*bind)(struct net *net, int group);
670 void (*unbind)(struct net *net, int group);
673 sock->state = SS_UNCONNECTED;
675 if (sock->type != SOCK_RAW && sock->type != SOCK_DGRAM)
676 return -ESOCKTNOSUPPORT;
678 if (protocol < 0 || protocol >= MAX_LINKS)
679 return -EPROTONOSUPPORT;
680 protocol = array_index_nospec(protocol, MAX_LINKS);
682 netlink_lock_table();
683 #ifdef CONFIG_MODULES
684 if (!nl_table[protocol].registered) {
685 netlink_unlock_table();
686 request_module("net-pf-%d-proto-%d", PF_NETLINK, protocol);
687 netlink_lock_table();
690 if (nl_table[protocol].registered &&
691 try_module_get(nl_table[protocol].module))
692 module = nl_table[protocol].module;
694 err = -EPROTONOSUPPORT;
695 cb_mutex = nl_table[protocol].cb_mutex;
696 bind = nl_table[protocol].bind;
697 unbind = nl_table[protocol].unbind;
698 netlink_unlock_table();
703 err = __netlink_create(net, sock, cb_mutex, protocol, kern);
708 sock_prot_inuse_add(net, &netlink_proto, 1);
711 nlk = nlk_sk(sock->sk);
712 nlk->module = module;
713 nlk->netlink_bind = bind;
714 nlk->netlink_unbind = unbind;
723 static void deferred_put_nlk_sk(struct rcu_head *head)
725 struct netlink_sock *nlk = container_of(head, struct netlink_sock, rcu);
726 struct sock *sk = &nlk->sk;
731 if (!refcount_dec_and_test(&sk->sk_refcnt))
734 if (nlk->cb_running && nlk->cb.done) {
735 INIT_WORK(&nlk->work, netlink_sock_destruct_work);
736 schedule_work(&nlk->work);
743 static int netlink_release(struct socket *sock)
745 struct sock *sk = sock->sk;
746 struct netlink_sock *nlk;
756 * OK. Socket is unlinked, any packets that arrive now
760 /* must not acquire netlink_table_lock in any way again before unbind
761 * and notifying genetlink is done as otherwise it might deadlock
763 if (nlk->netlink_unbind) {
766 for (i = 0; i < nlk->ngroups; i++)
767 if (test_bit(i, nlk->groups))
768 nlk->netlink_unbind(sock_net(sk), i + 1);
770 if (sk->sk_protocol == NETLINK_GENERIC &&
771 atomic_dec_return(&genl_sk_destructing_cnt) == 0)
772 wake_up(&genl_sk_destructing_waitq);
775 wake_up_interruptible_all(&nlk->wait);
777 skb_queue_purge(&sk->sk_write_queue);
779 if (nlk->portid && nlk->bound) {
780 struct netlink_notify n = {
782 .protocol = sk->sk_protocol,
783 .portid = nlk->portid,
785 blocking_notifier_call_chain(&netlink_chain,
786 NETLINK_URELEASE, &n);
789 module_put(nlk->module);
791 if (netlink_is_kernel(sk)) {
792 netlink_table_grab();
793 BUG_ON(nl_table[sk->sk_protocol].registered == 0);
794 if (--nl_table[sk->sk_protocol].registered == 0) {
795 struct listeners *old;
797 old = nl_deref_protected(nl_table[sk->sk_protocol].listeners);
798 RCU_INIT_POINTER(nl_table[sk->sk_protocol].listeners, NULL);
800 nl_table[sk->sk_protocol].module = NULL;
801 nl_table[sk->sk_protocol].bind = NULL;
802 nl_table[sk->sk_protocol].unbind = NULL;
803 nl_table[sk->sk_protocol].flags = 0;
804 nl_table[sk->sk_protocol].registered = 0;
806 netlink_table_ungrab();
810 sock_prot_inuse_add(sock_net(sk), &netlink_proto, -1);
812 call_rcu(&nlk->rcu, deferred_put_nlk_sk);
816 static int netlink_autobind(struct socket *sock)
818 struct sock *sk = sock->sk;
819 struct net *net = sock_net(sk);
820 struct netlink_table *table = &nl_table[sk->sk_protocol];
821 s32 portid = task_tgid_vnr(current);
829 ok = !__netlink_lookup(table, portid, net);
832 /* Bind collision, search negative portid values. */
834 /* rover will be in range [S32_MIN, -4097] */
835 rover = S32_MIN + prandom_u32_max(-4096 - S32_MIN);
836 else if (rover >= -4096)
842 err = netlink_insert(sk, portid);
843 if (err == -EADDRINUSE)
846 /* If 2 threads race to autobind, that is fine. */
854 * __netlink_ns_capable - General netlink message capability test
855 * @nsp: NETLINK_CB of the socket buffer holding a netlink command from userspace.
856 * @user_ns: The user namespace of the capability to use
857 * @cap: The capability to use
859 * Test to see if the opener of the socket we received the message
860 * from had when the netlink socket was created and the sender of the
861 * message has the capability @cap in the user namespace @user_ns.
863 bool __netlink_ns_capable(const struct netlink_skb_parms *nsp,
864 struct user_namespace *user_ns, int cap)
866 return ((nsp->flags & NETLINK_SKB_DST) ||
867 file_ns_capable(nsp->sk->sk_socket->file, user_ns, cap)) &&
868 ns_capable(user_ns, cap);
870 EXPORT_SYMBOL(__netlink_ns_capable);
873 * netlink_ns_capable - General netlink message capability test
874 * @skb: socket buffer holding a netlink command from userspace
875 * @user_ns: The user namespace of the capability to use
876 * @cap: The capability to use
878 * Test to see if the opener of the socket we received the message
879 * from had when the netlink socket was created and the sender of the
880 * message has the capability @cap in the user namespace @user_ns.
882 bool netlink_ns_capable(const struct sk_buff *skb,
883 struct user_namespace *user_ns, int cap)
885 return __netlink_ns_capable(&NETLINK_CB(skb), user_ns, cap);
887 EXPORT_SYMBOL(netlink_ns_capable);
890 * netlink_capable - Netlink global message capability test
891 * @skb: socket buffer holding a netlink command from userspace
892 * @cap: The capability to use
894 * Test to see if the opener of the socket we received the message
895 * from had when the netlink socket was created and the sender of the
896 * message has the capability @cap in all user namespaces.
898 bool netlink_capable(const struct sk_buff *skb, int cap)
900 return netlink_ns_capable(skb, &init_user_ns, cap);
902 EXPORT_SYMBOL(netlink_capable);
905 * netlink_net_capable - Netlink network namespace message capability test
906 * @skb: socket buffer holding a netlink command from userspace
907 * @cap: The capability to use
909 * Test to see if the opener of the socket we received the message
910 * from had when the netlink socket was created and the sender of the
911 * message has the capability @cap over the network namespace of
912 * the socket we received the message from.
914 bool netlink_net_capable(const struct sk_buff *skb, int cap)
916 return netlink_ns_capable(skb, sock_net(skb->sk)->user_ns, cap);
918 EXPORT_SYMBOL(netlink_net_capable);
920 static inline int netlink_allowed(const struct socket *sock, unsigned int flag)
922 return (nl_table[sock->sk->sk_protocol].flags & flag) ||
923 ns_capable(sock_net(sock->sk)->user_ns, CAP_NET_ADMIN);
927 netlink_update_subscriptions(struct sock *sk, unsigned int subscriptions)
929 struct netlink_sock *nlk = nlk_sk(sk);
931 if (nlk->subscriptions && !subscriptions)
932 __sk_del_bind_node(sk);
933 else if (!nlk->subscriptions && subscriptions)
934 sk_add_bind_node(sk, &nl_table[sk->sk_protocol].mc_list);
935 nlk->subscriptions = subscriptions;
938 static int netlink_realloc_groups(struct sock *sk)
940 struct netlink_sock *nlk = nlk_sk(sk);
942 unsigned long *new_groups;
945 netlink_table_grab();
947 groups = nl_table[sk->sk_protocol].groups;
948 if (!nl_table[sk->sk_protocol].registered) {
953 if (nlk->ngroups >= groups)
956 new_groups = krealloc(nlk->groups, NLGRPSZ(groups), GFP_ATOMIC);
957 if (new_groups == NULL) {
961 memset((char *)new_groups + NLGRPSZ(nlk->ngroups), 0,
962 NLGRPSZ(groups) - NLGRPSZ(nlk->ngroups));
964 nlk->groups = new_groups;
965 nlk->ngroups = groups;
967 netlink_table_ungrab();
971 static void netlink_undo_bind(int group, long unsigned int groups,
974 struct netlink_sock *nlk = nlk_sk(sk);
977 if (!nlk->netlink_unbind)
980 for (undo = 0; undo < group; undo++)
981 if (test_bit(undo, &groups))
982 nlk->netlink_unbind(sock_net(sk), undo + 1);
985 static int netlink_bind(struct socket *sock, struct sockaddr *addr,
988 struct sock *sk = sock->sk;
989 struct net *net = sock_net(sk);
990 struct netlink_sock *nlk = nlk_sk(sk);
991 struct sockaddr_nl *nladdr = (struct sockaddr_nl *)addr;
993 unsigned long groups;
996 if (addr_len < sizeof(struct sockaddr_nl))
999 if (nladdr->nl_family != AF_NETLINK)
1001 groups = nladdr->nl_groups;
1003 /* Only superuser is allowed to listen multicasts */
1005 if (!netlink_allowed(sock, NL_CFG_F_NONROOT_RECV))
1007 err = netlink_realloc_groups(sk);
1012 if (nlk->ngroups < BITS_PER_LONG)
1013 groups &= (1UL << nlk->ngroups) - 1;
1017 /* Ensure nlk->portid is up-to-date. */
1020 if (nladdr->nl_pid != nlk->portid)
1024 if (nlk->netlink_bind && groups) {
1027 /* nl_groups is a u32, so cap the maximum groups we can bind */
1028 for (group = 0; group < BITS_PER_TYPE(u32); group++) {
1029 if (!test_bit(group, &groups))
1031 err = nlk->netlink_bind(net, group + 1);
1034 netlink_undo_bind(group, groups, sk);
1039 /* No need for barriers here as we return to user-space without
1040 * using any of the bound attributes.
1042 netlink_lock_table();
1044 err = nladdr->nl_pid ?
1045 netlink_insert(sk, nladdr->nl_pid) :
1046 netlink_autobind(sock);
1048 netlink_undo_bind(BITS_PER_TYPE(u32), groups, sk);
1053 if (!groups && (nlk->groups == NULL || !(u32)nlk->groups[0]))
1055 netlink_unlock_table();
1057 netlink_table_grab();
1058 netlink_update_subscriptions(sk, nlk->subscriptions +
1060 hweight32(nlk->groups[0]));
1061 nlk->groups[0] = (nlk->groups[0] & ~0xffffffffUL) | groups;
1062 netlink_update_listeners(sk);
1063 netlink_table_ungrab();
1068 netlink_unlock_table();
1072 static int netlink_connect(struct socket *sock, struct sockaddr *addr,
1073 int alen, int flags)
1076 struct sock *sk = sock->sk;
1077 struct netlink_sock *nlk = nlk_sk(sk);
1078 struct sockaddr_nl *nladdr = (struct sockaddr_nl *)addr;
1080 if (alen < sizeof(addr->sa_family))
1083 if (addr->sa_family == AF_UNSPEC) {
1084 sk->sk_state = NETLINK_UNCONNECTED;
1085 nlk->dst_portid = 0;
1089 if (addr->sa_family != AF_NETLINK)
1092 if (alen < sizeof(struct sockaddr_nl))
1095 if ((nladdr->nl_groups || nladdr->nl_pid) &&
1096 !netlink_allowed(sock, NL_CFG_F_NONROOT_SEND))
1099 /* No need for barriers here as we return to user-space without
1100 * using any of the bound attributes.
1103 err = netlink_autobind(sock);
1106 sk->sk_state = NETLINK_CONNECTED;
1107 nlk->dst_portid = nladdr->nl_pid;
1108 nlk->dst_group = ffs(nladdr->nl_groups);
1114 static int netlink_getname(struct socket *sock, struct sockaddr *addr,
1117 struct sock *sk = sock->sk;
1118 struct netlink_sock *nlk = nlk_sk(sk);
1119 DECLARE_SOCKADDR(struct sockaddr_nl *, nladdr, addr);
1121 nladdr->nl_family = AF_NETLINK;
1125 nladdr->nl_pid = nlk->dst_portid;
1126 nladdr->nl_groups = netlink_group_mask(nlk->dst_group);
1128 nladdr->nl_pid = nlk->portid;
1129 netlink_lock_table();
1130 nladdr->nl_groups = nlk->groups ? nlk->groups[0] : 0;
1131 netlink_unlock_table();
1133 return sizeof(*nladdr);
1136 static int netlink_ioctl(struct socket *sock, unsigned int cmd,
1139 /* try to hand this ioctl down to the NIC drivers.
1141 return -ENOIOCTLCMD;
1144 static struct sock *netlink_getsockbyportid(struct sock *ssk, u32 portid)
1147 struct netlink_sock *nlk;
1149 sock = netlink_lookup(sock_net(ssk), ssk->sk_protocol, portid);
1151 return ERR_PTR(-ECONNREFUSED);
1153 /* Don't bother queuing skb if kernel socket has no input function */
1155 if (sock->sk_state == NETLINK_CONNECTED &&
1156 nlk->dst_portid != nlk_sk(ssk)->portid) {
1158 return ERR_PTR(-ECONNREFUSED);
1163 struct sock *netlink_getsockbyfilp(struct file *filp)
1165 struct inode *inode = file_inode(filp);
1168 if (!S_ISSOCK(inode->i_mode))
1169 return ERR_PTR(-ENOTSOCK);
1171 sock = SOCKET_I(inode)->sk;
1172 if (sock->sk_family != AF_NETLINK)
1173 return ERR_PTR(-EINVAL);
1179 static struct sk_buff *netlink_alloc_large_skb(unsigned int size,
1182 struct sk_buff *skb;
1185 if (size <= NLMSG_GOODSIZE || broadcast)
1186 return alloc_skb(size, GFP_KERNEL);
1188 size = SKB_DATA_ALIGN(size) +
1189 SKB_DATA_ALIGN(sizeof(struct skb_shared_info));
1191 data = vmalloc(size);
1195 skb = __build_skb(data, size);
1199 skb->destructor = netlink_skb_destructor;
1205 * Attach a skb to a netlink socket.
1206 * The caller must hold a reference to the destination socket. On error, the
1207 * reference is dropped. The skb is not send to the destination, just all
1208 * all error checks are performed and memory in the queue is reserved.
1210 * < 0: error. skb freed, reference to sock dropped.
1212 * 1: repeat lookup - reference dropped while waiting for socket memory.
1214 int netlink_attachskb(struct sock *sk, struct sk_buff *skb,
1215 long *timeo, struct sock *ssk)
1217 struct netlink_sock *nlk;
1221 if ((atomic_read(&sk->sk_rmem_alloc) > sk->sk_rcvbuf ||
1222 test_bit(NETLINK_S_CONGESTED, &nlk->state))) {
1223 DECLARE_WAITQUEUE(wait, current);
1225 if (!ssk || netlink_is_kernel(ssk))
1226 netlink_overrun(sk);
1232 __set_current_state(TASK_INTERRUPTIBLE);
1233 add_wait_queue(&nlk->wait, &wait);
1235 if ((atomic_read(&sk->sk_rmem_alloc) > sk->sk_rcvbuf ||
1236 test_bit(NETLINK_S_CONGESTED, &nlk->state)) &&
1237 !sock_flag(sk, SOCK_DEAD))
1238 *timeo = schedule_timeout(*timeo);
1240 __set_current_state(TASK_RUNNING);
1241 remove_wait_queue(&nlk->wait, &wait);
1244 if (signal_pending(current)) {
1246 return sock_intr_errno(*timeo);
1250 netlink_skb_set_owner_r(skb, sk);
1254 static int __netlink_sendskb(struct sock *sk, struct sk_buff *skb)
1258 netlink_deliver_tap(sock_net(sk), skb);
1260 skb_queue_tail(&sk->sk_receive_queue, skb);
1261 sk->sk_data_ready(sk);
1265 int netlink_sendskb(struct sock *sk, struct sk_buff *skb)
1267 int len = __netlink_sendskb(sk, skb);
1273 void netlink_detachskb(struct sock *sk, struct sk_buff *skb)
1279 static struct sk_buff *netlink_trim(struct sk_buff *skb, gfp_t allocation)
1283 WARN_ON(skb->sk != NULL);
1284 delta = skb->end - skb->tail;
1285 if (is_vmalloc_addr(skb->head) || delta * 2 < skb->truesize)
1288 if (skb_shared(skb)) {
1289 struct sk_buff *nskb = skb_clone(skb, allocation);
1296 pskb_expand_head(skb, 0, -delta,
1297 (allocation & ~__GFP_DIRECT_RECLAIM) |
1298 __GFP_NOWARN | __GFP_NORETRY);
1302 static int netlink_unicast_kernel(struct sock *sk, struct sk_buff *skb,
1306 struct netlink_sock *nlk = nlk_sk(sk);
1308 ret = -ECONNREFUSED;
1309 if (nlk->netlink_rcv != NULL) {
1311 netlink_skb_set_owner_r(skb, sk);
1312 NETLINK_CB(skb).sk = ssk;
1313 netlink_deliver_tap_kernel(sk, ssk, skb);
1314 nlk->netlink_rcv(skb);
1323 int netlink_unicast(struct sock *ssk, struct sk_buff *skb,
1324 u32 portid, int nonblock)
1330 skb = netlink_trim(skb, gfp_any());
1332 timeo = sock_sndtimeo(ssk, nonblock);
1334 sk = netlink_getsockbyportid(ssk, portid);
1339 if (netlink_is_kernel(sk))
1340 return netlink_unicast_kernel(sk, skb, ssk);
1342 if (sk_filter(sk, skb)) {
1349 err = netlink_attachskb(sk, skb, &timeo, ssk);
1355 return netlink_sendskb(sk, skb);
1357 EXPORT_SYMBOL(netlink_unicast);
1359 int netlink_has_listeners(struct sock *sk, unsigned int group)
1362 struct listeners *listeners;
1364 BUG_ON(!netlink_is_kernel(sk));
1367 listeners = rcu_dereference(nl_table[sk->sk_protocol].listeners);
1369 if (listeners && group - 1 < nl_table[sk->sk_protocol].groups)
1370 res = test_bit(group - 1, listeners->masks);
1376 EXPORT_SYMBOL_GPL(netlink_has_listeners);
1378 bool netlink_strict_get_check(struct sk_buff *skb)
1380 const struct netlink_sock *nlk = nlk_sk(NETLINK_CB(skb).sk);
1382 return nlk->flags & NETLINK_F_STRICT_CHK;
1384 EXPORT_SYMBOL_GPL(netlink_strict_get_check);
1386 static int netlink_broadcast_deliver(struct sock *sk, struct sk_buff *skb)
1388 struct netlink_sock *nlk = nlk_sk(sk);
1390 if (atomic_read(&sk->sk_rmem_alloc) <= sk->sk_rcvbuf &&
1391 !test_bit(NETLINK_S_CONGESTED, &nlk->state)) {
1392 netlink_skb_set_owner_r(skb, sk);
1393 __netlink_sendskb(sk, skb);
1394 return atomic_read(&sk->sk_rmem_alloc) > (sk->sk_rcvbuf >> 1);
1399 struct netlink_broadcast_data {
1400 struct sock *exclude_sk;
1405 int delivery_failure;
1409 struct sk_buff *skb, *skb2;
1410 int (*tx_filter)(struct sock *dsk, struct sk_buff *skb, void *data);
1414 static void do_one_broadcast(struct sock *sk,
1415 struct netlink_broadcast_data *p)
1417 struct netlink_sock *nlk = nlk_sk(sk);
1420 if (p->exclude_sk == sk)
1423 if (nlk->portid == p->portid || p->group - 1 >= nlk->ngroups ||
1424 !test_bit(p->group - 1, nlk->groups))
1427 if (!net_eq(sock_net(sk), p->net)) {
1428 if (!(nlk->flags & NETLINK_F_LISTEN_ALL_NSID))
1431 if (!peernet_has_id(sock_net(sk), p->net))
1434 if (!file_ns_capable(sk->sk_socket->file, p->net->user_ns,
1440 netlink_overrun(sk);
1445 if (p->skb2 == NULL) {
1446 if (skb_shared(p->skb)) {
1447 p->skb2 = skb_clone(p->skb, p->allocation);
1449 p->skb2 = skb_get(p->skb);
1451 * skb ownership may have been set when
1452 * delivered to a previous socket.
1454 skb_orphan(p->skb2);
1457 if (p->skb2 == NULL) {
1458 netlink_overrun(sk);
1459 /* Clone failed. Notify ALL listeners. */
1461 if (nlk->flags & NETLINK_F_BROADCAST_SEND_ERROR)
1462 p->delivery_failure = 1;
1465 if (p->tx_filter && p->tx_filter(sk, p->skb2, p->tx_data)) {
1470 if (sk_filter(sk, p->skb2)) {
1475 NETLINK_CB(p->skb2).nsid = peernet2id(sock_net(sk), p->net);
1476 if (NETLINK_CB(p->skb2).nsid != NETNSA_NSID_NOT_ASSIGNED)
1477 NETLINK_CB(p->skb2).nsid_is_set = true;
1478 val = netlink_broadcast_deliver(sk, p->skb2);
1480 netlink_overrun(sk);
1481 if (nlk->flags & NETLINK_F_BROADCAST_SEND_ERROR)
1482 p->delivery_failure = 1;
1484 p->congested |= val;
1492 int netlink_broadcast_filtered(struct sock *ssk, struct sk_buff *skb, u32 portid,
1493 u32 group, gfp_t allocation,
1494 int (*filter)(struct sock *dsk, struct sk_buff *skb, void *data),
1497 struct net *net = sock_net(ssk);
1498 struct netlink_broadcast_data info;
1501 skb = netlink_trim(skb, allocation);
1503 info.exclude_sk = ssk;
1505 info.portid = portid;
1508 info.delivery_failure = 0;
1511 info.allocation = allocation;
1514 info.tx_filter = filter;
1515 info.tx_data = filter_data;
1517 /* While we sleep in clone, do not allow to change socket list */
1519 netlink_lock_table();
1521 sk_for_each_bound(sk, &nl_table[ssk->sk_protocol].mc_list)
1522 do_one_broadcast(sk, &info);
1526 netlink_unlock_table();
1528 if (info.delivery_failure) {
1529 kfree_skb(info.skb2);
1532 consume_skb(info.skb2);
1534 if (info.delivered) {
1535 if (info.congested && gfpflags_allow_blocking(allocation))
1541 EXPORT_SYMBOL(netlink_broadcast_filtered);
1543 int netlink_broadcast(struct sock *ssk, struct sk_buff *skb, u32 portid,
1544 u32 group, gfp_t allocation)
1546 return netlink_broadcast_filtered(ssk, skb, portid, group, allocation,
1549 EXPORT_SYMBOL(netlink_broadcast);
1551 struct netlink_set_err_data {
1552 struct sock *exclude_sk;
1558 static int do_one_set_err(struct sock *sk, struct netlink_set_err_data *p)
1560 struct netlink_sock *nlk = nlk_sk(sk);
1563 if (sk == p->exclude_sk)
1566 if (!net_eq(sock_net(sk), sock_net(p->exclude_sk)))
1569 if (nlk->portid == p->portid || p->group - 1 >= nlk->ngroups ||
1570 !test_bit(p->group - 1, nlk->groups))
1573 if (p->code == ENOBUFS && nlk->flags & NETLINK_F_RECV_NO_ENOBUFS) {
1578 sk->sk_err = p->code;
1579 sk_error_report(sk);
1585 * netlink_set_err - report error to broadcast listeners
1586 * @ssk: the kernel netlink socket, as returned by netlink_kernel_create()
1587 * @portid: the PORTID of a process that we want to skip (if any)
1588 * @group: the broadcast group that will notice the error
1589 * @code: error code, must be negative (as usual in kernelspace)
1591 * This function returns the number of broadcast listeners that have set the
1592 * NETLINK_NO_ENOBUFS socket option.
1594 int netlink_set_err(struct sock *ssk, u32 portid, u32 group, int code)
1596 struct netlink_set_err_data info;
1600 info.exclude_sk = ssk;
1601 info.portid = portid;
1603 /* sk->sk_err wants a positive error value */
1606 read_lock(&nl_table_lock);
1608 sk_for_each_bound(sk, &nl_table[ssk->sk_protocol].mc_list)
1609 ret += do_one_set_err(sk, &info);
1611 read_unlock(&nl_table_lock);
1614 EXPORT_SYMBOL(netlink_set_err);
1616 /* must be called with netlink table grabbed */
1617 static void netlink_update_socket_mc(struct netlink_sock *nlk,
1621 int old, new = !!is_new, subscriptions;
1623 old = test_bit(group - 1, nlk->groups);
1624 subscriptions = nlk->subscriptions - old + new;
1626 __set_bit(group - 1, nlk->groups);
1628 __clear_bit(group - 1, nlk->groups);
1629 netlink_update_subscriptions(&nlk->sk, subscriptions);
1630 netlink_update_listeners(&nlk->sk);
1633 static int netlink_setsockopt(struct socket *sock, int level, int optname,
1634 sockptr_t optval, unsigned int optlen)
1636 struct sock *sk = sock->sk;
1637 struct netlink_sock *nlk = nlk_sk(sk);
1638 unsigned int val = 0;
1641 if (level != SOL_NETLINK)
1642 return -ENOPROTOOPT;
1644 if (optlen >= sizeof(int) &&
1645 copy_from_sockptr(&val, optval, sizeof(val)))
1649 case NETLINK_PKTINFO:
1651 nlk->flags |= NETLINK_F_RECV_PKTINFO;
1653 nlk->flags &= ~NETLINK_F_RECV_PKTINFO;
1656 case NETLINK_ADD_MEMBERSHIP:
1657 case NETLINK_DROP_MEMBERSHIP: {
1658 if (!netlink_allowed(sock, NL_CFG_F_NONROOT_RECV))
1660 err = netlink_realloc_groups(sk);
1663 if (!val || val - 1 >= nlk->ngroups)
1665 if (optname == NETLINK_ADD_MEMBERSHIP && nlk->netlink_bind) {
1666 err = nlk->netlink_bind(sock_net(sk), val);
1670 netlink_table_grab();
1671 netlink_update_socket_mc(nlk, val,
1672 optname == NETLINK_ADD_MEMBERSHIP);
1673 netlink_table_ungrab();
1674 if (optname == NETLINK_DROP_MEMBERSHIP && nlk->netlink_unbind)
1675 nlk->netlink_unbind(sock_net(sk), val);
1680 case NETLINK_BROADCAST_ERROR:
1682 nlk->flags |= NETLINK_F_BROADCAST_SEND_ERROR;
1684 nlk->flags &= ~NETLINK_F_BROADCAST_SEND_ERROR;
1687 case NETLINK_NO_ENOBUFS:
1689 nlk->flags |= NETLINK_F_RECV_NO_ENOBUFS;
1690 clear_bit(NETLINK_S_CONGESTED, &nlk->state);
1691 wake_up_interruptible(&nlk->wait);
1693 nlk->flags &= ~NETLINK_F_RECV_NO_ENOBUFS;
1697 case NETLINK_LISTEN_ALL_NSID:
1698 if (!ns_capable(sock_net(sk)->user_ns, CAP_NET_BROADCAST))
1702 nlk->flags |= NETLINK_F_LISTEN_ALL_NSID;
1704 nlk->flags &= ~NETLINK_F_LISTEN_ALL_NSID;
1707 case NETLINK_CAP_ACK:
1709 nlk->flags |= NETLINK_F_CAP_ACK;
1711 nlk->flags &= ~NETLINK_F_CAP_ACK;
1714 case NETLINK_EXT_ACK:
1716 nlk->flags |= NETLINK_F_EXT_ACK;
1718 nlk->flags &= ~NETLINK_F_EXT_ACK;
1721 case NETLINK_GET_STRICT_CHK:
1723 nlk->flags |= NETLINK_F_STRICT_CHK;
1725 nlk->flags &= ~NETLINK_F_STRICT_CHK;
1734 static int netlink_getsockopt(struct socket *sock, int level, int optname,
1735 char __user *optval, int __user *optlen)
1737 struct sock *sk = sock->sk;
1738 struct netlink_sock *nlk = nlk_sk(sk);
1741 if (level != SOL_NETLINK)
1742 return -ENOPROTOOPT;
1744 if (get_user(len, optlen))
1750 case NETLINK_PKTINFO:
1751 if (len < sizeof(int))
1754 val = nlk->flags & NETLINK_F_RECV_PKTINFO ? 1 : 0;
1755 if (put_user(len, optlen) ||
1756 put_user(val, optval))
1760 case NETLINK_BROADCAST_ERROR:
1761 if (len < sizeof(int))
1764 val = nlk->flags & NETLINK_F_BROADCAST_SEND_ERROR ? 1 : 0;
1765 if (put_user(len, optlen) ||
1766 put_user(val, optval))
1770 case NETLINK_NO_ENOBUFS:
1771 if (len < sizeof(int))
1774 val = nlk->flags & NETLINK_F_RECV_NO_ENOBUFS ? 1 : 0;
1775 if (put_user(len, optlen) ||
1776 put_user(val, optval))
1780 case NETLINK_LIST_MEMBERSHIPS: {
1781 int pos, idx, shift;
1784 netlink_lock_table();
1785 for (pos = 0; pos * 8 < nlk->ngroups; pos += sizeof(u32)) {
1786 if (len - pos < sizeof(u32))
1789 idx = pos / sizeof(unsigned long);
1790 shift = (pos % sizeof(unsigned long)) * 8;
1791 if (put_user((u32)(nlk->groups[idx] >> shift),
1792 (u32 __user *)(optval + pos))) {
1797 if (put_user(ALIGN(nlk->ngroups / 8, sizeof(u32)), optlen))
1799 netlink_unlock_table();
1802 case NETLINK_CAP_ACK:
1803 if (len < sizeof(int))
1806 val = nlk->flags & NETLINK_F_CAP_ACK ? 1 : 0;
1807 if (put_user(len, optlen) ||
1808 put_user(val, optval))
1812 case NETLINK_EXT_ACK:
1813 if (len < sizeof(int))
1816 val = nlk->flags & NETLINK_F_EXT_ACK ? 1 : 0;
1817 if (put_user(len, optlen) || put_user(val, optval))
1821 case NETLINK_GET_STRICT_CHK:
1822 if (len < sizeof(int))
1825 val = nlk->flags & NETLINK_F_STRICT_CHK ? 1 : 0;
1826 if (put_user(len, optlen) || put_user(val, optval))
1836 static void netlink_cmsg_recv_pktinfo(struct msghdr *msg, struct sk_buff *skb)
1838 struct nl_pktinfo info;
1840 info.group = NETLINK_CB(skb).dst_group;
1841 put_cmsg(msg, SOL_NETLINK, NETLINK_PKTINFO, sizeof(info), &info);
1844 static void netlink_cmsg_listen_all_nsid(struct sock *sk, struct msghdr *msg,
1845 struct sk_buff *skb)
1847 if (!NETLINK_CB(skb).nsid_is_set)
1850 put_cmsg(msg, SOL_NETLINK, NETLINK_LISTEN_ALL_NSID, sizeof(int),
1851 &NETLINK_CB(skb).nsid);
1854 static int netlink_sendmsg(struct socket *sock, struct msghdr *msg, size_t len)
1856 struct sock *sk = sock->sk;
1857 struct netlink_sock *nlk = nlk_sk(sk);
1858 DECLARE_SOCKADDR(struct sockaddr_nl *, addr, msg->msg_name);
1861 struct sk_buff *skb;
1863 struct scm_cookie scm;
1864 u32 netlink_skb_flags = 0;
1866 if (msg->msg_flags & MSG_OOB)
1869 err = scm_send(sock, msg, &scm, true);
1873 if (msg->msg_namelen) {
1875 if (msg->msg_namelen < sizeof(struct sockaddr_nl))
1877 if (addr->nl_family != AF_NETLINK)
1879 dst_portid = addr->nl_pid;
1880 dst_group = ffs(addr->nl_groups);
1882 if ((dst_group || dst_portid) &&
1883 !netlink_allowed(sock, NL_CFG_F_NONROOT_SEND))
1885 netlink_skb_flags |= NETLINK_SKB_DST;
1887 dst_portid = nlk->dst_portid;
1888 dst_group = nlk->dst_group;
1892 err = netlink_autobind(sock);
1896 /* Ensure nlk is hashed and visible. */
1901 if (len > sk->sk_sndbuf - 32)
1904 skb = netlink_alloc_large_skb(len, dst_group);
1908 NETLINK_CB(skb).portid = nlk->portid;
1909 NETLINK_CB(skb).dst_group = dst_group;
1910 NETLINK_CB(skb).creds = scm.creds;
1911 NETLINK_CB(skb).flags = netlink_skb_flags;
1914 if (memcpy_from_msg(skb_put(skb, len), msg, len)) {
1919 err = security_netlink_send(sk, skb);
1926 refcount_inc(&skb->users);
1927 netlink_broadcast(sk, skb, dst_portid, dst_group, GFP_KERNEL);
1929 err = netlink_unicast(sk, skb, dst_portid, msg->msg_flags & MSG_DONTWAIT);
1936 static int netlink_recvmsg(struct socket *sock, struct msghdr *msg, size_t len,
1939 struct scm_cookie scm;
1940 struct sock *sk = sock->sk;
1941 struct netlink_sock *nlk = nlk_sk(sk);
1942 int noblock = flags & MSG_DONTWAIT;
1944 struct sk_buff *skb, *data_skb;
1947 if (flags & MSG_OOB)
1952 skb = skb_recv_datagram(sk, flags, noblock, &err);
1958 #ifdef CONFIG_COMPAT_NETLINK_MESSAGES
1959 if (unlikely(skb_shinfo(skb)->frag_list)) {
1961 * If this skb has a frag_list, then here that means that we
1962 * will have to use the frag_list skb's data for compat tasks
1963 * and the regular skb's data for normal (non-compat) tasks.
1965 * If we need to send the compat skb, assign it to the
1966 * 'data_skb' variable so that it will be used below for data
1967 * copying. We keep 'skb' for everything else, including
1968 * freeing both later.
1970 if (flags & MSG_CMSG_COMPAT)
1971 data_skb = skb_shinfo(skb)->frag_list;
1975 /* Record the max length of recvmsg() calls for future allocations */
1976 nlk->max_recvmsg_len = max(nlk->max_recvmsg_len, len);
1977 nlk->max_recvmsg_len = min_t(size_t, nlk->max_recvmsg_len,
1978 SKB_WITH_OVERHEAD(32768));
1980 copied = data_skb->len;
1982 msg->msg_flags |= MSG_TRUNC;
1986 skb_reset_transport_header(data_skb);
1987 err = skb_copy_datagram_msg(data_skb, 0, msg, copied);
1989 if (msg->msg_name) {
1990 DECLARE_SOCKADDR(struct sockaddr_nl *, addr, msg->msg_name);
1991 addr->nl_family = AF_NETLINK;
1993 addr->nl_pid = NETLINK_CB(skb).portid;
1994 addr->nl_groups = netlink_group_mask(NETLINK_CB(skb).dst_group);
1995 msg->msg_namelen = sizeof(*addr);
1998 if (nlk->flags & NETLINK_F_RECV_PKTINFO)
1999 netlink_cmsg_recv_pktinfo(msg, skb);
2000 if (nlk->flags & NETLINK_F_LISTEN_ALL_NSID)
2001 netlink_cmsg_listen_all_nsid(sk, msg, skb);
2003 memset(&scm, 0, sizeof(scm));
2004 scm.creds = *NETLINK_CREDS(skb);
2005 if (flags & MSG_TRUNC)
2006 copied = data_skb->len;
2008 skb_free_datagram(sk, skb);
2010 if (nlk->cb_running &&
2011 atomic_read(&sk->sk_rmem_alloc) <= sk->sk_rcvbuf / 2) {
2012 ret = netlink_dump(sk);
2015 sk_error_report(sk);
2019 scm_recv(sock, msg, &scm, flags);
2021 netlink_rcv_wake(sk);
2022 return err ? : copied;
2025 static void netlink_data_ready(struct sock *sk)
2031 * We export these functions to other modules. They provide a
2032 * complete set of kernel non-blocking support for message
2037 __netlink_kernel_create(struct net *net, int unit, struct module *module,
2038 struct netlink_kernel_cfg *cfg)
2040 struct socket *sock;
2042 struct netlink_sock *nlk;
2043 struct listeners *listeners = NULL;
2044 struct mutex *cb_mutex = cfg ? cfg->cb_mutex : NULL;
2045 unsigned int groups;
2049 if (unit < 0 || unit >= MAX_LINKS)
2052 if (sock_create_lite(PF_NETLINK, SOCK_DGRAM, unit, &sock))
2055 if (__netlink_create(net, sock, cb_mutex, unit, 1) < 0)
2056 goto out_sock_release_nosk;
2060 if (!cfg || cfg->groups < 32)
2063 groups = cfg->groups;
2065 listeners = kzalloc(sizeof(*listeners) + NLGRPSZ(groups), GFP_KERNEL);
2067 goto out_sock_release;
2069 sk->sk_data_ready = netlink_data_ready;
2070 if (cfg && cfg->input)
2071 nlk_sk(sk)->netlink_rcv = cfg->input;
2073 if (netlink_insert(sk, 0))
2074 goto out_sock_release;
2077 nlk->flags |= NETLINK_F_KERNEL_SOCKET;
2079 netlink_table_grab();
2080 if (!nl_table[unit].registered) {
2081 nl_table[unit].groups = groups;
2082 rcu_assign_pointer(nl_table[unit].listeners, listeners);
2083 nl_table[unit].cb_mutex = cb_mutex;
2084 nl_table[unit].module = module;
2086 nl_table[unit].bind = cfg->bind;
2087 nl_table[unit].unbind = cfg->unbind;
2088 nl_table[unit].flags = cfg->flags;
2090 nl_table[unit].compare = cfg->compare;
2092 nl_table[unit].registered = 1;
2095 nl_table[unit].registered++;
2097 netlink_table_ungrab();
2102 netlink_kernel_release(sk);
2105 out_sock_release_nosk:
2109 EXPORT_SYMBOL(__netlink_kernel_create);
2112 netlink_kernel_release(struct sock *sk)
2114 if (sk == NULL || sk->sk_socket == NULL)
2117 sock_release(sk->sk_socket);
2119 EXPORT_SYMBOL(netlink_kernel_release);
2121 int __netlink_change_ngroups(struct sock *sk, unsigned int groups)
2123 struct listeners *new, *old;
2124 struct netlink_table *tbl = &nl_table[sk->sk_protocol];
2129 if (NLGRPSZ(tbl->groups) < NLGRPSZ(groups)) {
2130 new = kzalloc(sizeof(*new) + NLGRPSZ(groups), GFP_ATOMIC);
2133 old = nl_deref_protected(tbl->listeners);
2134 memcpy(new->masks, old->masks, NLGRPSZ(tbl->groups));
2135 rcu_assign_pointer(tbl->listeners, new);
2137 kfree_rcu(old, rcu);
2139 tbl->groups = groups;
2145 * netlink_change_ngroups - change number of multicast groups
2147 * This changes the number of multicast groups that are available
2148 * on a certain netlink family. Note that it is not possible to
2149 * change the number of groups to below 32. Also note that it does
2150 * not implicitly call netlink_clear_multicast_users() when the
2151 * number of groups is reduced.
2153 * @sk: The kernel netlink socket, as returned by netlink_kernel_create().
2154 * @groups: The new number of groups.
2156 int netlink_change_ngroups(struct sock *sk, unsigned int groups)
2160 netlink_table_grab();
2161 err = __netlink_change_ngroups(sk, groups);
2162 netlink_table_ungrab();
2167 void __netlink_clear_multicast_users(struct sock *ksk, unsigned int group)
2170 struct netlink_table *tbl = &nl_table[ksk->sk_protocol];
2172 sk_for_each_bound(sk, &tbl->mc_list)
2173 netlink_update_socket_mc(nlk_sk(sk), group, 0);
2177 __nlmsg_put(struct sk_buff *skb, u32 portid, u32 seq, int type, int len, int flags)
2179 struct nlmsghdr *nlh;
2180 int size = nlmsg_msg_size(len);
2182 nlh = skb_put(skb, NLMSG_ALIGN(size));
2183 nlh->nlmsg_type = type;
2184 nlh->nlmsg_len = size;
2185 nlh->nlmsg_flags = flags;
2186 nlh->nlmsg_pid = portid;
2187 nlh->nlmsg_seq = seq;
2188 if (!__builtin_constant_p(size) || NLMSG_ALIGN(size) - size != 0)
2189 memset(nlmsg_data(nlh) + len, 0, NLMSG_ALIGN(size) - size);
2192 EXPORT_SYMBOL(__nlmsg_put);
2195 * It looks a bit ugly.
2196 * It would be better to create kernel thread.
2199 static int netlink_dump_done(struct netlink_sock *nlk, struct sk_buff *skb,
2200 struct netlink_callback *cb,
2201 struct netlink_ext_ack *extack)
2203 struct nlmsghdr *nlh;
2205 nlh = nlmsg_put_answer(skb, cb, NLMSG_DONE, sizeof(nlk->dump_done_errno),
2206 NLM_F_MULTI | cb->answer_flags);
2210 nl_dump_check_consistent(cb, nlh);
2211 memcpy(nlmsg_data(nlh), &nlk->dump_done_errno, sizeof(nlk->dump_done_errno));
2213 if (extack->_msg && nlk->flags & NETLINK_F_EXT_ACK) {
2214 nlh->nlmsg_flags |= NLM_F_ACK_TLVS;
2215 if (!nla_put_string(skb, NLMSGERR_ATTR_MSG, extack->_msg))
2216 nlmsg_end(skb, nlh);
2222 static int netlink_dump(struct sock *sk)
2224 struct netlink_sock *nlk = nlk_sk(sk);
2225 struct netlink_ext_ack extack = {};
2226 struct netlink_callback *cb;
2227 struct sk_buff *skb = NULL;
2228 struct module *module;
2233 mutex_lock(nlk->cb_mutex);
2234 if (!nlk->cb_running) {
2239 if (atomic_read(&sk->sk_rmem_alloc) >= sk->sk_rcvbuf)
2242 /* NLMSG_GOODSIZE is small to avoid high order allocations being
2243 * required, but it makes sense to _attempt_ a 16K bytes allocation
2244 * to reduce number of system calls on dump operations, if user
2245 * ever provided a big enough buffer.
2248 alloc_min_size = max_t(int, cb->min_dump_alloc, NLMSG_GOODSIZE);
2250 if (alloc_min_size < nlk->max_recvmsg_len) {
2251 alloc_size = nlk->max_recvmsg_len;
2252 skb = alloc_skb(alloc_size,
2253 (GFP_KERNEL & ~__GFP_DIRECT_RECLAIM) |
2254 __GFP_NOWARN | __GFP_NORETRY);
2257 alloc_size = alloc_min_size;
2258 skb = alloc_skb(alloc_size, GFP_KERNEL);
2263 /* Trim skb to allocated size. User is expected to provide buffer as
2264 * large as max(min_dump_alloc, 16KiB (mac_recvmsg_len capped at
2265 * netlink_recvmsg())). dump will pack as many smaller messages as
2266 * could fit within the allocated skb. skb is typically allocated
2267 * with larger space than required (could be as much as near 2x the
2268 * requested size with align to next power of 2 approach). Allowing
2269 * dump to use the excess space makes it difficult for a user to have a
2270 * reasonable static buffer based on the expected largest dump of a
2271 * single netdev. The outcome is MSG_TRUNC error.
2273 skb_reserve(skb, skb_tailroom(skb) - alloc_size);
2274 netlink_skb_set_owner_r(skb, sk);
2276 if (nlk->dump_done_errno > 0) {
2277 cb->extack = &extack;
2278 nlk->dump_done_errno = cb->dump(skb, cb);
2282 if (nlk->dump_done_errno > 0 ||
2283 skb_tailroom(skb) < nlmsg_total_size(sizeof(nlk->dump_done_errno))) {
2284 mutex_unlock(nlk->cb_mutex);
2286 if (sk_filter(sk, skb))
2289 __netlink_sendskb(sk, skb);
2293 if (netlink_dump_done(nlk, skb, cb, &extack))
2296 #ifdef CONFIG_COMPAT_NETLINK_MESSAGES
2297 /* frag_list skb's data is used for compat tasks
2298 * and the regular skb's data for normal (non-compat) tasks.
2299 * See netlink_recvmsg().
2301 if (unlikely(skb_shinfo(skb)->frag_list)) {
2302 if (netlink_dump_done(nlk, skb_shinfo(skb)->frag_list, cb, &extack))
2307 if (sk_filter(sk, skb))
2310 __netlink_sendskb(sk, skb);
2315 nlk->cb_running = false;
2316 module = cb->module;
2318 mutex_unlock(nlk->cb_mutex);
2324 mutex_unlock(nlk->cb_mutex);
2329 int __netlink_dump_start(struct sock *ssk, struct sk_buff *skb,
2330 const struct nlmsghdr *nlh,
2331 struct netlink_dump_control *control)
2333 struct netlink_sock *nlk, *nlk2;
2334 struct netlink_callback *cb;
2338 refcount_inc(&skb->users);
2340 sk = netlink_lookup(sock_net(ssk), ssk->sk_protocol, NETLINK_CB(skb).portid);
2342 ret = -ECONNREFUSED;
2347 mutex_lock(nlk->cb_mutex);
2348 /* A dump is in progress... */
2349 if (nlk->cb_running) {
2353 /* add reference of module which cb->dump belongs to */
2354 if (!try_module_get(control->module)) {
2355 ret = -EPROTONOSUPPORT;
2360 memset(cb, 0, sizeof(*cb));
2361 cb->dump = control->dump;
2362 cb->done = control->done;
2364 cb->data = control->data;
2365 cb->module = control->module;
2366 cb->min_dump_alloc = control->min_dump_alloc;
2369 nlk2 = nlk_sk(NETLINK_CB(skb).sk);
2370 cb->strict_check = !!(nlk2->flags & NETLINK_F_STRICT_CHK);
2372 if (control->start) {
2373 ret = control->start(cb);
2378 nlk->cb_running = true;
2379 nlk->dump_done_errno = INT_MAX;
2381 mutex_unlock(nlk->cb_mutex);
2383 ret = netlink_dump(sk);
2390 /* We successfully started a dump, by returning -EINTR we
2391 * signal not to send ACK even if it was requested.
2396 module_put(control->module);
2399 mutex_unlock(nlk->cb_mutex);
2404 EXPORT_SYMBOL(__netlink_dump_start);
2406 void netlink_ack(struct sk_buff *in_skb, struct nlmsghdr *nlh, int err,
2407 const struct netlink_ext_ack *extack)
2409 struct sk_buff *skb;
2410 struct nlmsghdr *rep;
2411 struct nlmsgerr *errmsg;
2412 size_t payload = sizeof(*errmsg);
2414 struct netlink_sock *nlk = nlk_sk(NETLINK_CB(in_skb).sk);
2415 unsigned int flags = 0;
2416 bool nlk_has_extack = nlk->flags & NETLINK_F_EXT_ACK;
2418 /* Error messages get the original request appened, unless the user
2419 * requests to cap the error message, and get extra error data if
2422 if (nlk_has_extack && extack && extack->_msg)
2423 tlvlen += nla_total_size(strlen(extack->_msg) + 1);
2425 if (err && !(nlk->flags & NETLINK_F_CAP_ACK))
2426 payload += nlmsg_len(nlh);
2428 flags |= NLM_F_CAPPED;
2429 if (err && nlk_has_extack && extack && extack->bad_attr)
2430 tlvlen += nla_total_size(sizeof(u32));
2431 if (nlk_has_extack && extack && extack->cookie_len)
2432 tlvlen += nla_total_size(extack->cookie_len);
2433 if (err && nlk_has_extack && extack && extack->policy)
2434 tlvlen += netlink_policy_dump_attr_size_estimate(extack->policy);
2437 flags |= NLM_F_ACK_TLVS;
2439 skb = nlmsg_new(payload + tlvlen, GFP_KERNEL);
2441 NETLINK_CB(in_skb).sk->sk_err = ENOBUFS;
2442 sk_error_report(NETLINK_CB(in_skb).sk);
2446 rep = __nlmsg_put(skb, NETLINK_CB(in_skb).portid, nlh->nlmsg_seq,
2447 NLMSG_ERROR, payload, flags);
2448 errmsg = nlmsg_data(rep);
2449 errmsg->error = err;
2450 memcpy(&errmsg->msg, nlh, payload > sizeof(*errmsg) ? nlh->nlmsg_len : sizeof(*nlh));
2452 if (nlk_has_extack && extack) {
2454 WARN_ON(nla_put_string(skb, NLMSGERR_ATTR_MSG,
2457 if (err && extack->bad_attr &&
2458 !WARN_ON((u8 *)extack->bad_attr < in_skb->data ||
2459 (u8 *)extack->bad_attr >= in_skb->data +
2461 WARN_ON(nla_put_u32(skb, NLMSGERR_ATTR_OFFS,
2462 (u8 *)extack->bad_attr -
2464 if (extack->cookie_len)
2465 WARN_ON(nla_put(skb, NLMSGERR_ATTR_COOKIE,
2466 extack->cookie_len, extack->cookie));
2468 netlink_policy_dump_write_attr(skb, extack->policy,
2469 NLMSGERR_ATTR_POLICY);
2472 nlmsg_end(skb, rep);
2474 nlmsg_unicast(in_skb->sk, skb, NETLINK_CB(in_skb).portid);
2476 EXPORT_SYMBOL(netlink_ack);
2478 int netlink_rcv_skb(struct sk_buff *skb, int (*cb)(struct sk_buff *,
2480 struct netlink_ext_ack *))
2482 struct netlink_ext_ack extack;
2483 struct nlmsghdr *nlh;
2486 while (skb->len >= nlmsg_total_size(0)) {
2489 memset(&extack, 0, sizeof(extack));
2490 nlh = nlmsg_hdr(skb);
2493 if (nlh->nlmsg_len < NLMSG_HDRLEN || skb->len < nlh->nlmsg_len)
2496 /* Only requests are handled by the kernel */
2497 if (!(nlh->nlmsg_flags & NLM_F_REQUEST))
2500 /* Skip control messages */
2501 if (nlh->nlmsg_type < NLMSG_MIN_TYPE)
2504 err = cb(skb, nlh, &extack);
2509 if (nlh->nlmsg_flags & NLM_F_ACK || err)
2510 netlink_ack(skb, nlh, err, &extack);
2513 msglen = NLMSG_ALIGN(nlh->nlmsg_len);
2514 if (msglen > skb->len)
2516 skb_pull(skb, msglen);
2521 EXPORT_SYMBOL(netlink_rcv_skb);
2524 * nlmsg_notify - send a notification netlink message
2525 * @sk: netlink socket to use
2526 * @skb: notification message
2527 * @portid: destination netlink portid for reports or 0
2528 * @group: destination multicast group or 0
2529 * @report: 1 to report back, 0 to disable
2530 * @flags: allocation flags
2532 int nlmsg_notify(struct sock *sk, struct sk_buff *skb, u32 portid,
2533 unsigned int group, int report, gfp_t flags)
2538 int exclude_portid = 0;
2541 refcount_inc(&skb->users);
2542 exclude_portid = portid;
2545 /* errors reported via destination sk->sk_err, but propagate
2546 * delivery errors if NETLINK_BROADCAST_ERROR flag is set */
2547 err = nlmsg_multicast(sk, skb, exclude_portid, group, flags);
2555 err2 = nlmsg_unicast(sk, skb, portid);
2562 EXPORT_SYMBOL(nlmsg_notify);
2564 #ifdef CONFIG_PROC_FS
2565 struct nl_seq_iter {
2566 struct seq_net_private p;
2567 struct rhashtable_iter hti;
2571 static void netlink_walk_start(struct nl_seq_iter *iter)
2573 rhashtable_walk_enter(&nl_table[iter->link].hash, &iter->hti);
2574 rhashtable_walk_start(&iter->hti);
2577 static void netlink_walk_stop(struct nl_seq_iter *iter)
2579 rhashtable_walk_stop(&iter->hti);
2580 rhashtable_walk_exit(&iter->hti);
2583 static void *__netlink_seq_next(struct seq_file *seq)
2585 struct nl_seq_iter *iter = seq->private;
2586 struct netlink_sock *nlk;
2590 nlk = rhashtable_walk_next(&iter->hti);
2593 if (PTR_ERR(nlk) == -EAGAIN)
2602 netlink_walk_stop(iter);
2603 if (++iter->link >= MAX_LINKS)
2606 netlink_walk_start(iter);
2608 } while (sock_net(&nlk->sk) != seq_file_net(seq));
2613 static void *netlink_seq_start(struct seq_file *seq, loff_t *posp)
2616 struct nl_seq_iter *iter = seq->private;
2617 void *obj = SEQ_START_TOKEN;
2622 netlink_walk_start(iter);
2624 for (pos = *posp; pos && obj && !IS_ERR(obj); pos--)
2625 obj = __netlink_seq_next(seq);
2630 static void *netlink_seq_next(struct seq_file *seq, void *v, loff_t *pos)
2633 return __netlink_seq_next(seq);
2636 static void netlink_native_seq_stop(struct seq_file *seq, void *v)
2638 struct nl_seq_iter *iter = seq->private;
2640 if (iter->link >= MAX_LINKS)
2643 netlink_walk_stop(iter);
2647 static int netlink_native_seq_show(struct seq_file *seq, void *v)
2649 if (v == SEQ_START_TOKEN) {
2651 "sk Eth Pid Groups "
2652 "Rmem Wmem Dump Locks Drops Inode\n");
2655 struct netlink_sock *nlk = nlk_sk(s);
2657 seq_printf(seq, "%pK %-3d %-10u %08x %-8d %-8d %-5d %-8d %-8u %-8lu\n",
2661 nlk->groups ? (u32)nlk->groups[0] : 0,
2662 sk_rmem_alloc_get(s),
2663 sk_wmem_alloc_get(s),
2665 refcount_read(&s->sk_refcnt),
2666 atomic_read(&s->sk_drops),
2674 #ifdef CONFIG_BPF_SYSCALL
2675 struct bpf_iter__netlink {
2676 __bpf_md_ptr(struct bpf_iter_meta *, meta);
2677 __bpf_md_ptr(struct netlink_sock *, sk);
2680 DEFINE_BPF_ITER_FUNC(netlink, struct bpf_iter_meta *meta, struct netlink_sock *sk)
2682 static int netlink_prog_seq_show(struct bpf_prog *prog,
2683 struct bpf_iter_meta *meta,
2686 struct bpf_iter__netlink ctx;
2688 meta->seq_num--; /* skip SEQ_START_TOKEN */
2690 ctx.sk = nlk_sk((struct sock *)v);
2691 return bpf_iter_run_prog(prog, &ctx);
2694 static int netlink_seq_show(struct seq_file *seq, void *v)
2696 struct bpf_iter_meta meta;
2697 struct bpf_prog *prog;
2700 prog = bpf_iter_get_info(&meta, false);
2702 return netlink_native_seq_show(seq, v);
2704 if (v != SEQ_START_TOKEN)
2705 return netlink_prog_seq_show(prog, &meta, v);
2710 static void netlink_seq_stop(struct seq_file *seq, void *v)
2712 struct bpf_iter_meta meta;
2713 struct bpf_prog *prog;
2717 prog = bpf_iter_get_info(&meta, true);
2719 (void)netlink_prog_seq_show(prog, &meta, v);
2722 netlink_native_seq_stop(seq, v);
2725 static int netlink_seq_show(struct seq_file *seq, void *v)
2727 return netlink_native_seq_show(seq, v);
2730 static void netlink_seq_stop(struct seq_file *seq, void *v)
2732 netlink_native_seq_stop(seq, v);
2736 static const struct seq_operations netlink_seq_ops = {
2737 .start = netlink_seq_start,
2738 .next = netlink_seq_next,
2739 .stop = netlink_seq_stop,
2740 .show = netlink_seq_show,
2744 int netlink_register_notifier(struct notifier_block *nb)
2746 return blocking_notifier_chain_register(&netlink_chain, nb);
2748 EXPORT_SYMBOL(netlink_register_notifier);
2750 int netlink_unregister_notifier(struct notifier_block *nb)
2752 return blocking_notifier_chain_unregister(&netlink_chain, nb);
2754 EXPORT_SYMBOL(netlink_unregister_notifier);
2756 static const struct proto_ops netlink_ops = {
2757 .family = PF_NETLINK,
2758 .owner = THIS_MODULE,
2759 .release = netlink_release,
2760 .bind = netlink_bind,
2761 .connect = netlink_connect,
2762 .socketpair = sock_no_socketpair,
2763 .accept = sock_no_accept,
2764 .getname = netlink_getname,
2765 .poll = datagram_poll,
2766 .ioctl = netlink_ioctl,
2767 .listen = sock_no_listen,
2768 .shutdown = sock_no_shutdown,
2769 .setsockopt = netlink_setsockopt,
2770 .getsockopt = netlink_getsockopt,
2771 .sendmsg = netlink_sendmsg,
2772 .recvmsg = netlink_recvmsg,
2773 .mmap = sock_no_mmap,
2774 .sendpage = sock_no_sendpage,
2777 static const struct net_proto_family netlink_family_ops = {
2778 .family = PF_NETLINK,
2779 .create = netlink_create,
2780 .owner = THIS_MODULE, /* for consistency 8) */
2783 static int __net_init netlink_net_init(struct net *net)
2785 #ifdef CONFIG_PROC_FS
2786 if (!proc_create_net("netlink", 0, net->proc_net, &netlink_seq_ops,
2787 sizeof(struct nl_seq_iter)))
2793 static void __net_exit netlink_net_exit(struct net *net)
2795 #ifdef CONFIG_PROC_FS
2796 remove_proc_entry("netlink", net->proc_net);
2800 static void __init netlink_add_usersock_entry(void)
2802 struct listeners *listeners;
2805 listeners = kzalloc(sizeof(*listeners) + NLGRPSZ(groups), GFP_KERNEL);
2807 panic("netlink_add_usersock_entry: Cannot allocate listeners\n");
2809 netlink_table_grab();
2811 nl_table[NETLINK_USERSOCK].groups = groups;
2812 rcu_assign_pointer(nl_table[NETLINK_USERSOCK].listeners, listeners);
2813 nl_table[NETLINK_USERSOCK].module = THIS_MODULE;
2814 nl_table[NETLINK_USERSOCK].registered = 1;
2815 nl_table[NETLINK_USERSOCK].flags = NL_CFG_F_NONROOT_SEND;
2817 netlink_table_ungrab();
2820 static struct pernet_operations __net_initdata netlink_net_ops = {
2821 .init = netlink_net_init,
2822 .exit = netlink_net_exit,
2825 static inline u32 netlink_hash(const void *data, u32 len, u32 seed)
2827 const struct netlink_sock *nlk = data;
2828 struct netlink_compare_arg arg;
2830 netlink_compare_arg_init(&arg, sock_net(&nlk->sk), nlk->portid);
2831 return jhash2((u32 *)&arg, netlink_compare_arg_len / sizeof(u32), seed);
2834 static const struct rhashtable_params netlink_rhashtable_params = {
2835 .head_offset = offsetof(struct netlink_sock, node),
2836 .key_len = netlink_compare_arg_len,
2837 .obj_hashfn = netlink_hash,
2838 .obj_cmpfn = netlink_compare,
2839 .automatic_shrinking = true,
2842 #if defined(CONFIG_BPF_SYSCALL) && defined(CONFIG_PROC_FS)
2843 BTF_ID_LIST(btf_netlink_sock_id)
2844 BTF_ID(struct, netlink_sock)
2846 static const struct bpf_iter_seq_info netlink_seq_info = {
2847 .seq_ops = &netlink_seq_ops,
2848 .init_seq_private = bpf_iter_init_seq_net,
2849 .fini_seq_private = bpf_iter_fini_seq_net,
2850 .seq_priv_size = sizeof(struct nl_seq_iter),
2853 static struct bpf_iter_reg netlink_reg_info = {
2854 .target = "netlink",
2855 .ctx_arg_info_size = 1,
2857 { offsetof(struct bpf_iter__netlink, sk),
2858 PTR_TO_BTF_ID_OR_NULL },
2860 .seq_info = &netlink_seq_info,
2863 static int __init bpf_iter_register(void)
2865 netlink_reg_info.ctx_arg_info[0].btf_id = *btf_netlink_sock_id;
2866 return bpf_iter_reg_target(&netlink_reg_info);
2870 static int __init netlink_proto_init(void)
2873 int err = proto_register(&netlink_proto, 0);
2878 #if defined(CONFIG_BPF_SYSCALL) && defined(CONFIG_PROC_FS)
2879 err = bpf_iter_register();
2884 BUILD_BUG_ON(sizeof(struct netlink_skb_parms) > sizeof_field(struct sk_buff, cb));
2886 nl_table = kcalloc(MAX_LINKS, sizeof(*nl_table), GFP_KERNEL);
2890 for (i = 0; i < MAX_LINKS; i++) {
2891 if (rhashtable_init(&nl_table[i].hash,
2892 &netlink_rhashtable_params) < 0) {
2894 rhashtable_destroy(&nl_table[i].hash);
2900 netlink_add_usersock_entry();
2902 sock_register(&netlink_family_ops);
2903 register_pernet_subsys(&netlink_net_ops);
2904 register_pernet_subsys(&netlink_tap_net_ops);
2905 /* The netlink device handler may be needed early. */
2910 panic("netlink_init: Cannot allocate nl_table\n");
2913 core_initcall(netlink_proto_init);