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/bpf.h>
24 #include <linux/capability.h>
25 #include <linux/kernel.h>
26 #include <linux/filter.h>
27 #include <linux/init.h>
28 #include <linux/signal.h>
29 #include <linux/sched.h>
30 #include <linux/errno.h>
31 #include <linux/string.h>
32 #include <linux/stat.h>
33 #include <linux/socket.h>
35 #include <linux/fcntl.h>
36 #include <linux/termios.h>
37 #include <linux/sockios.h>
38 #include <linux/net.h>
40 #include <linux/slab.h>
41 #include <linux/uaccess.h>
42 #include <linux/skbuff.h>
43 #include <linux/netdevice.h>
44 #include <linux/rtnetlink.h>
45 #include <linux/proc_fs.h>
46 #include <linux/seq_file.h>
47 #include <linux/notifier.h>
48 #include <linux/security.h>
49 #include <linux/jhash.h>
50 #include <linux/jiffies.h>
51 #include <linux/random.h>
52 #include <linux/bitops.h>
54 #include <linux/types.h>
55 #include <linux/audit.h>
56 #include <linux/mutex.h>
57 #include <linux/vmalloc.h>
58 #include <linux/if_arp.h>
59 #include <linux/rhashtable.h>
60 #include <asm/cacheflush.h>
61 #include <linux/hash.h>
62 #include <linux/genetlink.h>
63 #include <linux/net_namespace.h>
64 #include <linux/nospec.h>
65 #include <linux/btf_ids.h>
67 #include <net/net_namespace.h>
68 #include <net/netns/generic.h>
71 #include <net/netlink.h>
72 #define CREATE_TRACE_POINTS
73 #include <trace/events/netlink.h>
75 #include "af_netlink.h"
79 unsigned long masks[];
83 #define NETLINK_S_CONGESTED 0x0
85 static inline int netlink_is_kernel(struct sock *sk)
87 return nlk_test_bit(KERNEL_SOCKET, sk);
90 struct netlink_table *nl_table __read_mostly;
91 EXPORT_SYMBOL_GPL(nl_table);
93 static DECLARE_WAIT_QUEUE_HEAD(nl_table_wait);
95 static struct lock_class_key nlk_cb_mutex_keys[MAX_LINKS];
97 static const char *const nlk_cb_mutex_key_strings[MAX_LINKS + 1] = {
100 "nlk_cb_mutex-USERSOCK",
101 "nlk_cb_mutex-FIREWALL",
102 "nlk_cb_mutex-SOCK_DIAG",
103 "nlk_cb_mutex-NFLOG",
105 "nlk_cb_mutex-SELINUX",
106 "nlk_cb_mutex-ISCSI",
107 "nlk_cb_mutex-AUDIT",
108 "nlk_cb_mutex-FIB_LOOKUP",
109 "nlk_cb_mutex-CONNECTOR",
110 "nlk_cb_mutex-NETFILTER",
111 "nlk_cb_mutex-IP6_FW",
112 "nlk_cb_mutex-DNRTMSG",
113 "nlk_cb_mutex-KOBJECT_UEVENT",
114 "nlk_cb_mutex-GENERIC",
116 "nlk_cb_mutex-SCSITRANSPORT",
117 "nlk_cb_mutex-ECRYPTFS",
119 "nlk_cb_mutex-CRYPTO",
130 "nlk_cb_mutex-MAX_LINKS"
133 static int netlink_dump(struct sock *sk);
135 /* nl_table locking explained:
136 * Lookup and traversal are protected with an RCU read-side lock. Insertion
137 * and removal are protected with per bucket lock while using RCU list
138 * modification primitives and may run in parallel to RCU protected lookups.
139 * Destruction of the Netlink socket may only occur *after* nl_table_lock has
140 * been acquired * either during or after the socket has been removed from
141 * the list and after an RCU grace period.
143 DEFINE_RWLOCK(nl_table_lock);
144 EXPORT_SYMBOL_GPL(nl_table_lock);
145 static atomic_t nl_table_users = ATOMIC_INIT(0);
147 #define nl_deref_protected(X) rcu_dereference_protected(X, lockdep_is_held(&nl_table_lock));
149 static BLOCKING_NOTIFIER_HEAD(netlink_chain);
152 static const struct rhashtable_params netlink_rhashtable_params;
154 void do_trace_netlink_extack(const char *msg)
156 trace_netlink_extack(msg);
158 EXPORT_SYMBOL(do_trace_netlink_extack);
160 static inline u32 netlink_group_mask(u32 group)
164 return group ? 1 << (group - 1) : 0;
167 static struct sk_buff *netlink_to_full_skb(const struct sk_buff *skb,
170 unsigned int len = skb_end_offset(skb);
173 new = alloc_skb(len, gfp_mask);
177 NETLINK_CB(new).portid = NETLINK_CB(skb).portid;
178 NETLINK_CB(new).dst_group = NETLINK_CB(skb).dst_group;
179 NETLINK_CB(new).creds = NETLINK_CB(skb).creds;
181 skb_put_data(new, skb->data, len);
185 static unsigned int netlink_tap_net_id;
187 struct netlink_tap_net {
188 struct list_head netlink_tap_all;
189 struct mutex netlink_tap_lock;
192 int netlink_add_tap(struct netlink_tap *nt)
194 struct net *net = dev_net(nt->dev);
195 struct netlink_tap_net *nn = net_generic(net, netlink_tap_net_id);
197 if (unlikely(nt->dev->type != ARPHRD_NETLINK))
200 mutex_lock(&nn->netlink_tap_lock);
201 list_add_rcu(&nt->list, &nn->netlink_tap_all);
202 mutex_unlock(&nn->netlink_tap_lock);
204 __module_get(nt->module);
208 EXPORT_SYMBOL_GPL(netlink_add_tap);
210 static int __netlink_remove_tap(struct netlink_tap *nt)
212 struct net *net = dev_net(nt->dev);
213 struct netlink_tap_net *nn = net_generic(net, netlink_tap_net_id);
215 struct netlink_tap *tmp;
217 mutex_lock(&nn->netlink_tap_lock);
219 list_for_each_entry(tmp, &nn->netlink_tap_all, list) {
221 list_del_rcu(&nt->list);
227 pr_warn("__netlink_remove_tap: %p not found\n", nt);
229 mutex_unlock(&nn->netlink_tap_lock);
232 module_put(nt->module);
234 return found ? 0 : -ENODEV;
237 int netlink_remove_tap(struct netlink_tap *nt)
241 ret = __netlink_remove_tap(nt);
246 EXPORT_SYMBOL_GPL(netlink_remove_tap);
248 static __net_init int netlink_tap_init_net(struct net *net)
250 struct netlink_tap_net *nn = net_generic(net, netlink_tap_net_id);
252 INIT_LIST_HEAD(&nn->netlink_tap_all);
253 mutex_init(&nn->netlink_tap_lock);
257 static struct pernet_operations netlink_tap_net_ops = {
258 .init = netlink_tap_init_net,
259 .id = &netlink_tap_net_id,
260 .size = sizeof(struct netlink_tap_net),
263 static bool netlink_filter_tap(const struct sk_buff *skb)
265 struct sock *sk = skb->sk;
267 /* We take the more conservative approach and
268 * whitelist socket protocols that may pass.
270 switch (sk->sk_protocol) {
272 case NETLINK_USERSOCK:
273 case NETLINK_SOCK_DIAG:
276 case NETLINK_FIB_LOOKUP:
277 case NETLINK_NETFILTER:
278 case NETLINK_GENERIC:
285 static int __netlink_deliver_tap_skb(struct sk_buff *skb,
286 struct net_device *dev)
288 struct sk_buff *nskb;
289 struct sock *sk = skb->sk;
292 if (!net_eq(dev_net(dev), sock_net(sk)))
297 if (is_vmalloc_addr(skb->head))
298 nskb = netlink_to_full_skb(skb, GFP_ATOMIC);
300 nskb = skb_clone(skb, GFP_ATOMIC);
303 nskb->protocol = htons((u16) sk->sk_protocol);
304 nskb->pkt_type = netlink_is_kernel(sk) ?
305 PACKET_KERNEL : PACKET_USER;
306 skb_reset_network_header(nskb);
307 ret = dev_queue_xmit(nskb);
308 if (unlikely(ret > 0))
309 ret = net_xmit_errno(ret);
316 static void __netlink_deliver_tap(struct sk_buff *skb, struct netlink_tap_net *nn)
319 struct netlink_tap *tmp;
321 if (!netlink_filter_tap(skb))
324 list_for_each_entry_rcu(tmp, &nn->netlink_tap_all, list) {
325 ret = __netlink_deliver_tap_skb(skb, tmp->dev);
331 static void netlink_deliver_tap(struct net *net, struct sk_buff *skb)
333 struct netlink_tap_net *nn = net_generic(net, netlink_tap_net_id);
337 if (unlikely(!list_empty(&nn->netlink_tap_all)))
338 __netlink_deliver_tap(skb, nn);
343 static void netlink_deliver_tap_kernel(struct sock *dst, struct sock *src,
346 if (!(netlink_is_kernel(dst) && netlink_is_kernel(src)))
347 netlink_deliver_tap(sock_net(dst), skb);
350 static void netlink_overrun(struct sock *sk)
352 if (!nlk_test_bit(RECV_NO_ENOBUFS, sk)) {
353 if (!test_and_set_bit(NETLINK_S_CONGESTED,
354 &nlk_sk(sk)->state)) {
355 sk->sk_err = ENOBUFS;
359 atomic_inc(&sk->sk_drops);
362 static void netlink_rcv_wake(struct sock *sk)
364 struct netlink_sock *nlk = nlk_sk(sk);
366 if (skb_queue_empty_lockless(&sk->sk_receive_queue))
367 clear_bit(NETLINK_S_CONGESTED, &nlk->state);
368 if (!test_bit(NETLINK_S_CONGESTED, &nlk->state))
369 wake_up_interruptible(&nlk->wait);
372 static void netlink_skb_destructor(struct sk_buff *skb)
374 if (is_vmalloc_addr(skb->head)) {
376 !atomic_dec_return(&(skb_shinfo(skb)->dataref)))
385 static void netlink_skb_set_owner_r(struct sk_buff *skb, struct sock *sk)
387 WARN_ON(skb->sk != NULL);
389 skb->destructor = netlink_skb_destructor;
390 atomic_add(skb->truesize, &sk->sk_rmem_alloc);
391 sk_mem_charge(sk, skb->truesize);
394 static void netlink_sock_destruct(struct sock *sk)
396 struct netlink_sock *nlk = nlk_sk(sk);
398 if (nlk->cb_running) {
400 nlk->cb.done(&nlk->cb);
401 module_put(nlk->cb.module);
402 kfree_skb(nlk->cb.skb);
405 skb_queue_purge(&sk->sk_receive_queue);
407 if (!sock_flag(sk, SOCK_DEAD)) {
408 printk(KERN_ERR "Freeing alive netlink socket %p\n", sk);
412 WARN_ON(atomic_read(&sk->sk_rmem_alloc));
413 WARN_ON(refcount_read(&sk->sk_wmem_alloc));
414 WARN_ON(nlk_sk(sk)->groups);
417 static void netlink_sock_destruct_work(struct work_struct *work)
419 struct netlink_sock *nlk = container_of(work, struct netlink_sock,
425 /* This lock without WQ_FLAG_EXCLUSIVE is good on UP and it is _very_ bad on
426 * SMP. Look, when several writers sleep and reader wakes them up, all but one
427 * immediately hit write lock and grab all the cpus. Exclusive sleep solves
428 * this, _but_ remember, it adds useless work on UP machines.
431 void netlink_table_grab(void)
432 __acquires(nl_table_lock)
436 write_lock_irq(&nl_table_lock);
438 if (atomic_read(&nl_table_users)) {
439 DECLARE_WAITQUEUE(wait, current);
441 add_wait_queue_exclusive(&nl_table_wait, &wait);
443 set_current_state(TASK_UNINTERRUPTIBLE);
444 if (atomic_read(&nl_table_users) == 0)
446 write_unlock_irq(&nl_table_lock);
448 write_lock_irq(&nl_table_lock);
451 __set_current_state(TASK_RUNNING);
452 remove_wait_queue(&nl_table_wait, &wait);
456 void netlink_table_ungrab(void)
457 __releases(nl_table_lock)
459 write_unlock_irq(&nl_table_lock);
460 wake_up(&nl_table_wait);
464 netlink_lock_table(void)
468 /* read_lock() synchronizes us to netlink_table_grab */
470 read_lock_irqsave(&nl_table_lock, flags);
471 atomic_inc(&nl_table_users);
472 read_unlock_irqrestore(&nl_table_lock, flags);
476 netlink_unlock_table(void)
478 if (atomic_dec_and_test(&nl_table_users))
479 wake_up(&nl_table_wait);
482 struct netlink_compare_arg
488 /* Doing sizeof directly may yield 4 extra bytes on 64-bit. */
489 #define netlink_compare_arg_len \
490 (offsetof(struct netlink_compare_arg, portid) + sizeof(u32))
492 static inline int netlink_compare(struct rhashtable_compare_arg *arg,
495 const struct netlink_compare_arg *x = arg->key;
496 const struct netlink_sock *nlk = ptr;
498 return nlk->portid != x->portid ||
499 !net_eq(sock_net(&nlk->sk), read_pnet(&x->pnet));
502 static void netlink_compare_arg_init(struct netlink_compare_arg *arg,
503 struct net *net, u32 portid)
505 memset(arg, 0, sizeof(*arg));
506 write_pnet(&arg->pnet, net);
507 arg->portid = portid;
510 static struct sock *__netlink_lookup(struct netlink_table *table, u32 portid,
513 struct netlink_compare_arg arg;
515 netlink_compare_arg_init(&arg, net, portid);
516 return rhashtable_lookup_fast(&table->hash, &arg,
517 netlink_rhashtable_params);
520 static int __netlink_insert(struct netlink_table *table, struct sock *sk)
522 struct netlink_compare_arg arg;
524 netlink_compare_arg_init(&arg, sock_net(sk), nlk_sk(sk)->portid);
525 return rhashtable_lookup_insert_key(&table->hash, &arg,
527 netlink_rhashtable_params);
530 static struct sock *netlink_lookup(struct net *net, int protocol, u32 portid)
532 struct netlink_table *table = &nl_table[protocol];
536 sk = __netlink_lookup(table, portid, net);
544 static const struct proto_ops netlink_ops;
547 netlink_update_listeners(struct sock *sk)
549 struct netlink_table *tbl = &nl_table[sk->sk_protocol];
552 struct listeners *listeners;
554 listeners = nl_deref_protected(tbl->listeners);
558 for (i = 0; i < NLGRPLONGS(tbl->groups); i++) {
560 sk_for_each_bound(sk, &tbl->mc_list) {
561 if (i < NLGRPLONGS(nlk_sk(sk)->ngroups))
562 mask |= nlk_sk(sk)->groups[i];
564 listeners->masks[i] = mask;
566 /* this function is only called with the netlink table "grabbed", which
567 * makes sure updates are visible before bind or setsockopt return. */
570 static int netlink_insert(struct sock *sk, u32 portid)
572 struct netlink_table *table = &nl_table[sk->sk_protocol];
577 err = nlk_sk(sk)->portid == portid ? 0 : -EBUSY;
578 if (nlk_sk(sk)->bound)
581 /* portid can be read locklessly from netlink_getname(). */
582 WRITE_ONCE(nlk_sk(sk)->portid, portid);
586 err = __netlink_insert(table, sk);
588 /* In case the hashtable backend returns with -EBUSY
589 * from here, it must not escape to the caller.
591 if (unlikely(err == -EBUSY))
599 /* We need to ensure that the socket is hashed and visible. */
601 /* Paired with lockless reads from netlink_bind(),
602 * netlink_connect() and netlink_sendmsg().
604 WRITE_ONCE(nlk_sk(sk)->bound, portid);
611 static void netlink_remove(struct sock *sk)
613 struct netlink_table *table;
615 table = &nl_table[sk->sk_protocol];
616 if (!rhashtable_remove_fast(&table->hash, &nlk_sk(sk)->node,
617 netlink_rhashtable_params)) {
618 WARN_ON(refcount_read(&sk->sk_refcnt) == 1);
622 netlink_table_grab();
623 if (nlk_sk(sk)->subscriptions) {
624 __sk_del_bind_node(sk);
625 netlink_update_listeners(sk);
627 if (sk->sk_protocol == NETLINK_GENERIC)
628 atomic_inc(&genl_sk_destructing_cnt);
629 netlink_table_ungrab();
632 static struct proto netlink_proto = {
634 .owner = THIS_MODULE,
635 .obj_size = sizeof(struct netlink_sock),
638 static int __netlink_create(struct net *net, struct socket *sock,
639 struct mutex *cb_mutex, int protocol,
643 struct netlink_sock *nlk;
645 sock->ops = &netlink_ops;
647 sk = sk_alloc(net, PF_NETLINK, GFP_KERNEL, &netlink_proto, kern);
651 sock_init_data(sock, sk);
655 nlk->cb_mutex = cb_mutex;
657 nlk->cb_mutex = &nlk->cb_def_mutex;
658 mutex_init(nlk->cb_mutex);
659 lockdep_set_class_and_name(nlk->cb_mutex,
660 nlk_cb_mutex_keys + protocol,
661 nlk_cb_mutex_key_strings[protocol]);
663 init_waitqueue_head(&nlk->wait);
665 sk->sk_destruct = netlink_sock_destruct;
666 sk->sk_protocol = protocol;
670 static int netlink_create(struct net *net, struct socket *sock, int protocol,
673 struct module *module = NULL;
674 struct mutex *cb_mutex;
675 struct netlink_sock *nlk;
676 int (*bind)(struct net *net, int group);
677 void (*unbind)(struct net *net, int group);
678 void (*release)(struct sock *sock, unsigned long *groups);
681 sock->state = SS_UNCONNECTED;
683 if (sock->type != SOCK_RAW && sock->type != SOCK_DGRAM)
684 return -ESOCKTNOSUPPORT;
686 if (protocol < 0 || protocol >= MAX_LINKS)
687 return -EPROTONOSUPPORT;
688 protocol = array_index_nospec(protocol, MAX_LINKS);
690 netlink_lock_table();
691 #ifdef CONFIG_MODULES
692 if (!nl_table[protocol].registered) {
693 netlink_unlock_table();
694 request_module("net-pf-%d-proto-%d", PF_NETLINK, protocol);
695 netlink_lock_table();
698 if (nl_table[protocol].registered &&
699 try_module_get(nl_table[protocol].module))
700 module = nl_table[protocol].module;
702 err = -EPROTONOSUPPORT;
703 cb_mutex = nl_table[protocol].cb_mutex;
704 bind = nl_table[protocol].bind;
705 unbind = nl_table[protocol].unbind;
706 release = nl_table[protocol].release;
707 netlink_unlock_table();
712 err = __netlink_create(net, sock, cb_mutex, protocol, kern);
716 sock_prot_inuse_add(net, &netlink_proto, 1);
718 nlk = nlk_sk(sock->sk);
719 nlk->module = module;
720 nlk->netlink_bind = bind;
721 nlk->netlink_unbind = unbind;
722 nlk->netlink_release = release;
731 static void deferred_put_nlk_sk(struct rcu_head *head)
733 struct netlink_sock *nlk = container_of(head, struct netlink_sock, rcu);
734 struct sock *sk = &nlk->sk;
739 if (!refcount_dec_and_test(&sk->sk_refcnt))
742 if (nlk->cb_running && nlk->cb.done) {
743 INIT_WORK(&nlk->work, netlink_sock_destruct_work);
744 schedule_work(&nlk->work);
751 static int netlink_release(struct socket *sock)
753 struct sock *sk = sock->sk;
754 struct netlink_sock *nlk;
764 * OK. Socket is unlinked, any packets that arrive now
767 if (nlk->netlink_release)
768 nlk->netlink_release(sk, nlk->groups);
770 /* must not acquire netlink_table_lock in any way again before unbind
771 * and notifying genetlink is done as otherwise it might deadlock
773 if (nlk->netlink_unbind) {
776 for (i = 0; i < nlk->ngroups; i++)
777 if (test_bit(i, nlk->groups))
778 nlk->netlink_unbind(sock_net(sk), i + 1);
780 if (sk->sk_protocol == NETLINK_GENERIC &&
781 atomic_dec_return(&genl_sk_destructing_cnt) == 0)
782 wake_up(&genl_sk_destructing_waitq);
785 wake_up_interruptible_all(&nlk->wait);
787 skb_queue_purge(&sk->sk_write_queue);
789 if (nlk->portid && nlk->bound) {
790 struct netlink_notify n = {
792 .protocol = sk->sk_protocol,
793 .portid = nlk->portid,
795 blocking_notifier_call_chain(&netlink_chain,
796 NETLINK_URELEASE, &n);
799 module_put(nlk->module);
801 if (netlink_is_kernel(sk)) {
802 netlink_table_grab();
803 BUG_ON(nl_table[sk->sk_protocol].registered == 0);
804 if (--nl_table[sk->sk_protocol].registered == 0) {
805 struct listeners *old;
807 old = nl_deref_protected(nl_table[sk->sk_protocol].listeners);
808 RCU_INIT_POINTER(nl_table[sk->sk_protocol].listeners, NULL);
810 nl_table[sk->sk_protocol].module = NULL;
811 nl_table[sk->sk_protocol].bind = NULL;
812 nl_table[sk->sk_protocol].unbind = NULL;
813 nl_table[sk->sk_protocol].flags = 0;
814 nl_table[sk->sk_protocol].registered = 0;
816 netlink_table_ungrab();
819 sock_prot_inuse_add(sock_net(sk), &netlink_proto, -1);
821 /* Because struct net might disappear soon, do not keep a pointer. */
822 if (!sk->sk_net_refcnt && sock_net(sk) != &init_net) {
823 __netns_tracker_free(sock_net(sk), &sk->ns_tracker, false);
824 /* Because of deferred_put_nlk_sk and use of work queue,
825 * it is possible netns will be freed before this socket.
827 sock_net_set(sk, &init_net);
828 __netns_tracker_alloc(&init_net, &sk->ns_tracker,
831 call_rcu(&nlk->rcu, deferred_put_nlk_sk);
835 static int netlink_autobind(struct socket *sock)
837 struct sock *sk = sock->sk;
838 struct net *net = sock_net(sk);
839 struct netlink_table *table = &nl_table[sk->sk_protocol];
840 s32 portid = task_tgid_vnr(current);
848 ok = !__netlink_lookup(table, portid, net);
851 /* Bind collision, search negative portid values. */
853 /* rover will be in range [S32_MIN, -4097] */
854 rover = S32_MIN + get_random_u32_below(-4096 - S32_MIN);
855 else if (rover >= -4096)
861 err = netlink_insert(sk, portid);
862 if (err == -EADDRINUSE)
865 /* If 2 threads race to autobind, that is fine. */
873 * __netlink_ns_capable - General netlink message capability test
874 * @nsp: NETLINK_CB of the 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 netlink_skb_parms *nsp,
883 struct user_namespace *user_ns, int cap)
885 return ((nsp->flags & NETLINK_SKB_DST) ||
886 file_ns_capable(nsp->sk->sk_socket->file, user_ns, cap)) &&
887 ns_capable(user_ns, cap);
889 EXPORT_SYMBOL(__netlink_ns_capable);
892 * netlink_ns_capable - General netlink message capability test
893 * @skb: socket buffer holding a netlink command from userspace
894 * @user_ns: The user namespace of the capability to use
895 * @cap: The capability to use
897 * Test to see if the opener of the socket we received the message
898 * from had when the netlink socket was created and the sender of the
899 * message has the capability @cap in the user namespace @user_ns.
901 bool netlink_ns_capable(const struct sk_buff *skb,
902 struct user_namespace *user_ns, int cap)
904 return __netlink_ns_capable(&NETLINK_CB(skb), user_ns, cap);
906 EXPORT_SYMBOL(netlink_ns_capable);
909 * netlink_capable - Netlink global message capability test
910 * @skb: socket buffer holding a netlink command from userspace
911 * @cap: The capability to use
913 * Test to see if the opener of the socket we received the message
914 * from had when the netlink socket was created and the sender of the
915 * message has the capability @cap in all user namespaces.
917 bool netlink_capable(const struct sk_buff *skb, int cap)
919 return netlink_ns_capable(skb, &init_user_ns, cap);
921 EXPORT_SYMBOL(netlink_capable);
924 * netlink_net_capable - Netlink network namespace message capability test
925 * @skb: socket buffer holding a netlink command from userspace
926 * @cap: The capability to use
928 * Test to see if the opener of the socket we received the message
929 * from had when the netlink socket was created and the sender of the
930 * message has the capability @cap over the network namespace of
931 * the socket we received the message from.
933 bool netlink_net_capable(const struct sk_buff *skb, int cap)
935 return netlink_ns_capable(skb, sock_net(skb->sk)->user_ns, cap);
937 EXPORT_SYMBOL(netlink_net_capable);
939 static inline int netlink_allowed(const struct socket *sock, unsigned int flag)
941 return (nl_table[sock->sk->sk_protocol].flags & flag) ||
942 ns_capable(sock_net(sock->sk)->user_ns, CAP_NET_ADMIN);
946 netlink_update_subscriptions(struct sock *sk, unsigned int subscriptions)
948 struct netlink_sock *nlk = nlk_sk(sk);
950 if (nlk->subscriptions && !subscriptions)
951 __sk_del_bind_node(sk);
952 else if (!nlk->subscriptions && subscriptions)
953 sk_add_bind_node(sk, &nl_table[sk->sk_protocol].mc_list);
954 nlk->subscriptions = subscriptions;
957 static int netlink_realloc_groups(struct sock *sk)
959 struct netlink_sock *nlk = nlk_sk(sk);
961 unsigned long *new_groups;
964 netlink_table_grab();
966 groups = nl_table[sk->sk_protocol].groups;
967 if (!nl_table[sk->sk_protocol].registered) {
972 if (nlk->ngroups >= groups)
975 new_groups = krealloc(nlk->groups, NLGRPSZ(groups), GFP_ATOMIC);
976 if (new_groups == NULL) {
980 memset((char *)new_groups + NLGRPSZ(nlk->ngroups), 0,
981 NLGRPSZ(groups) - NLGRPSZ(nlk->ngroups));
983 nlk->groups = new_groups;
984 nlk->ngroups = groups;
986 netlink_table_ungrab();
990 static void netlink_undo_bind(int group, long unsigned int groups,
993 struct netlink_sock *nlk = nlk_sk(sk);
996 if (!nlk->netlink_unbind)
999 for (undo = 0; undo < group; undo++)
1000 if (test_bit(undo, &groups))
1001 nlk->netlink_unbind(sock_net(sk), undo + 1);
1004 static int netlink_bind(struct socket *sock, struct sockaddr *addr,
1007 struct sock *sk = sock->sk;
1008 struct net *net = sock_net(sk);
1009 struct netlink_sock *nlk = nlk_sk(sk);
1010 struct sockaddr_nl *nladdr = (struct sockaddr_nl *)addr;
1012 unsigned long groups;
1015 if (addr_len < sizeof(struct sockaddr_nl))
1018 if (nladdr->nl_family != AF_NETLINK)
1020 groups = nladdr->nl_groups;
1022 /* Only superuser is allowed to listen multicasts */
1024 if (!netlink_allowed(sock, NL_CFG_F_NONROOT_RECV))
1026 err = netlink_realloc_groups(sk);
1031 if (nlk->ngroups < BITS_PER_LONG)
1032 groups &= (1UL << nlk->ngroups) - 1;
1034 /* Paired with WRITE_ONCE() in netlink_insert() */
1035 bound = READ_ONCE(nlk->bound);
1037 /* Ensure nlk->portid is up-to-date. */
1040 if (nladdr->nl_pid != nlk->portid)
1044 if (nlk->netlink_bind && groups) {
1047 /* nl_groups is a u32, so cap the maximum groups we can bind */
1048 for (group = 0; group < BITS_PER_TYPE(u32); group++) {
1049 if (!test_bit(group, &groups))
1051 err = nlk->netlink_bind(net, group + 1);
1054 netlink_undo_bind(group, groups, sk);
1059 /* No need for barriers here as we return to user-space without
1060 * using any of the bound attributes.
1062 netlink_lock_table();
1064 err = nladdr->nl_pid ?
1065 netlink_insert(sk, nladdr->nl_pid) :
1066 netlink_autobind(sock);
1068 netlink_undo_bind(BITS_PER_TYPE(u32), groups, sk);
1073 if (!groups && (nlk->groups == NULL || !(u32)nlk->groups[0]))
1075 netlink_unlock_table();
1077 netlink_table_grab();
1078 netlink_update_subscriptions(sk, nlk->subscriptions +
1080 hweight32(nlk->groups[0]));
1081 nlk->groups[0] = (nlk->groups[0] & ~0xffffffffUL) | groups;
1082 netlink_update_listeners(sk);
1083 netlink_table_ungrab();
1088 netlink_unlock_table();
1092 static int netlink_connect(struct socket *sock, struct sockaddr *addr,
1093 int alen, int flags)
1096 struct sock *sk = sock->sk;
1097 struct netlink_sock *nlk = nlk_sk(sk);
1098 struct sockaddr_nl *nladdr = (struct sockaddr_nl *)addr;
1100 if (alen < sizeof(addr->sa_family))
1103 if (addr->sa_family == AF_UNSPEC) {
1104 /* paired with READ_ONCE() in netlink_getsockbyportid() */
1105 WRITE_ONCE(sk->sk_state, NETLINK_UNCONNECTED);
1106 /* dst_portid and dst_group can be read locklessly */
1107 WRITE_ONCE(nlk->dst_portid, 0);
1108 WRITE_ONCE(nlk->dst_group, 0);
1111 if (addr->sa_family != AF_NETLINK)
1114 if (alen < sizeof(struct sockaddr_nl))
1117 if ((nladdr->nl_groups || nladdr->nl_pid) &&
1118 !netlink_allowed(sock, NL_CFG_F_NONROOT_SEND))
1121 /* No need for barriers here as we return to user-space without
1122 * using any of the bound attributes.
1123 * Paired with WRITE_ONCE() in netlink_insert().
1125 if (!READ_ONCE(nlk->bound))
1126 err = netlink_autobind(sock);
1129 /* paired with READ_ONCE() in netlink_getsockbyportid() */
1130 WRITE_ONCE(sk->sk_state, NETLINK_CONNECTED);
1131 /* dst_portid and dst_group can be read locklessly */
1132 WRITE_ONCE(nlk->dst_portid, nladdr->nl_pid);
1133 WRITE_ONCE(nlk->dst_group, ffs(nladdr->nl_groups));
1139 static int netlink_getname(struct socket *sock, struct sockaddr *addr,
1142 struct sock *sk = sock->sk;
1143 struct netlink_sock *nlk = nlk_sk(sk);
1144 DECLARE_SOCKADDR(struct sockaddr_nl *, nladdr, addr);
1146 nladdr->nl_family = AF_NETLINK;
1150 /* Paired with WRITE_ONCE() in netlink_connect() */
1151 nladdr->nl_pid = READ_ONCE(nlk->dst_portid);
1152 nladdr->nl_groups = netlink_group_mask(READ_ONCE(nlk->dst_group));
1154 /* Paired with WRITE_ONCE() in netlink_insert() */
1155 nladdr->nl_pid = READ_ONCE(nlk->portid);
1156 netlink_lock_table();
1157 nladdr->nl_groups = nlk->groups ? nlk->groups[0] : 0;
1158 netlink_unlock_table();
1160 return sizeof(*nladdr);
1163 static int netlink_ioctl(struct socket *sock, unsigned int cmd,
1166 /* try to hand this ioctl down to the NIC drivers.
1168 return -ENOIOCTLCMD;
1171 static struct sock *netlink_getsockbyportid(struct sock *ssk, u32 portid)
1174 struct netlink_sock *nlk;
1176 sock = netlink_lookup(sock_net(ssk), ssk->sk_protocol, portid);
1178 return ERR_PTR(-ECONNREFUSED);
1180 /* Don't bother queuing skb if kernel socket has no input function */
1182 /* dst_portid and sk_state can be changed in netlink_connect() */
1183 if (READ_ONCE(sock->sk_state) == NETLINK_CONNECTED &&
1184 READ_ONCE(nlk->dst_portid) != nlk_sk(ssk)->portid) {
1186 return ERR_PTR(-ECONNREFUSED);
1191 struct sock *netlink_getsockbyfilp(struct file *filp)
1193 struct inode *inode = file_inode(filp);
1196 if (!S_ISSOCK(inode->i_mode))
1197 return ERR_PTR(-ENOTSOCK);
1199 sock = SOCKET_I(inode)->sk;
1200 if (sock->sk_family != AF_NETLINK)
1201 return ERR_PTR(-EINVAL);
1207 static struct sk_buff *netlink_alloc_large_skb(unsigned int size,
1210 struct sk_buff *skb;
1213 if (size <= NLMSG_GOODSIZE || broadcast)
1214 return alloc_skb(size, GFP_KERNEL);
1216 size = SKB_DATA_ALIGN(size) +
1217 SKB_DATA_ALIGN(sizeof(struct skb_shared_info));
1219 data = vmalloc(size);
1223 skb = __build_skb(data, size);
1227 skb->destructor = netlink_skb_destructor;
1233 * Attach a skb to a netlink socket.
1234 * The caller must hold a reference to the destination socket. On error, the
1235 * reference is dropped. The skb is not send to the destination, just all
1236 * all error checks are performed and memory in the queue is reserved.
1238 * < 0: error. skb freed, reference to sock dropped.
1240 * 1: repeat lookup - reference dropped while waiting for socket memory.
1242 int netlink_attachskb(struct sock *sk, struct sk_buff *skb,
1243 long *timeo, struct sock *ssk)
1245 struct netlink_sock *nlk;
1249 if ((atomic_read(&sk->sk_rmem_alloc) > sk->sk_rcvbuf ||
1250 test_bit(NETLINK_S_CONGESTED, &nlk->state))) {
1251 DECLARE_WAITQUEUE(wait, current);
1253 if (!ssk || netlink_is_kernel(ssk))
1254 netlink_overrun(sk);
1260 __set_current_state(TASK_INTERRUPTIBLE);
1261 add_wait_queue(&nlk->wait, &wait);
1263 if ((atomic_read(&sk->sk_rmem_alloc) > sk->sk_rcvbuf ||
1264 test_bit(NETLINK_S_CONGESTED, &nlk->state)) &&
1265 !sock_flag(sk, SOCK_DEAD))
1266 *timeo = schedule_timeout(*timeo);
1268 __set_current_state(TASK_RUNNING);
1269 remove_wait_queue(&nlk->wait, &wait);
1272 if (signal_pending(current)) {
1274 return sock_intr_errno(*timeo);
1278 netlink_skb_set_owner_r(skb, sk);
1282 static int __netlink_sendskb(struct sock *sk, struct sk_buff *skb)
1286 netlink_deliver_tap(sock_net(sk), skb);
1288 skb_queue_tail(&sk->sk_receive_queue, skb);
1289 sk->sk_data_ready(sk);
1293 int netlink_sendskb(struct sock *sk, struct sk_buff *skb)
1295 int len = __netlink_sendskb(sk, skb);
1301 void netlink_detachskb(struct sock *sk, struct sk_buff *skb)
1307 static struct sk_buff *netlink_trim(struct sk_buff *skb, gfp_t allocation)
1311 WARN_ON(skb->sk != NULL);
1312 delta = skb->end - skb->tail;
1313 if (is_vmalloc_addr(skb->head) || delta * 2 < skb->truesize)
1316 if (skb_shared(skb)) {
1317 struct sk_buff *nskb = skb_clone(skb, allocation);
1324 pskb_expand_head(skb, 0, -delta,
1325 (allocation & ~__GFP_DIRECT_RECLAIM) |
1326 __GFP_NOWARN | __GFP_NORETRY);
1330 static int netlink_unicast_kernel(struct sock *sk, struct sk_buff *skb,
1334 struct netlink_sock *nlk = nlk_sk(sk);
1336 ret = -ECONNREFUSED;
1337 if (nlk->netlink_rcv != NULL) {
1339 netlink_skb_set_owner_r(skb, sk);
1340 NETLINK_CB(skb).sk = ssk;
1341 netlink_deliver_tap_kernel(sk, ssk, skb);
1342 nlk->netlink_rcv(skb);
1351 int netlink_unicast(struct sock *ssk, struct sk_buff *skb,
1352 u32 portid, int nonblock)
1358 skb = netlink_trim(skb, gfp_any());
1360 timeo = sock_sndtimeo(ssk, nonblock);
1362 sk = netlink_getsockbyportid(ssk, portid);
1367 if (netlink_is_kernel(sk))
1368 return netlink_unicast_kernel(sk, skb, ssk);
1370 if (sk_filter(sk, skb)) {
1377 err = netlink_attachskb(sk, skb, &timeo, ssk);
1383 return netlink_sendskb(sk, skb);
1385 EXPORT_SYMBOL(netlink_unicast);
1387 int netlink_has_listeners(struct sock *sk, unsigned int group)
1390 struct listeners *listeners;
1392 BUG_ON(!netlink_is_kernel(sk));
1395 listeners = rcu_dereference(nl_table[sk->sk_protocol].listeners);
1397 if (listeners && group - 1 < nl_table[sk->sk_protocol].groups)
1398 res = test_bit(group - 1, listeners->masks);
1404 EXPORT_SYMBOL_GPL(netlink_has_listeners);
1406 bool netlink_strict_get_check(struct sk_buff *skb)
1408 return nlk_test_bit(STRICT_CHK, NETLINK_CB(skb).sk);
1410 EXPORT_SYMBOL_GPL(netlink_strict_get_check);
1412 static int netlink_broadcast_deliver(struct sock *sk, struct sk_buff *skb)
1414 struct netlink_sock *nlk = nlk_sk(sk);
1416 if (atomic_read(&sk->sk_rmem_alloc) <= sk->sk_rcvbuf &&
1417 !test_bit(NETLINK_S_CONGESTED, &nlk->state)) {
1418 netlink_skb_set_owner_r(skb, sk);
1419 __netlink_sendskb(sk, skb);
1420 return atomic_read(&sk->sk_rmem_alloc) > (sk->sk_rcvbuf >> 1);
1425 struct netlink_broadcast_data {
1426 struct sock *exclude_sk;
1431 int delivery_failure;
1435 struct sk_buff *skb, *skb2;
1436 int (*tx_filter)(struct sock *dsk, struct sk_buff *skb, void *data);
1440 static void do_one_broadcast(struct sock *sk,
1441 struct netlink_broadcast_data *p)
1443 struct netlink_sock *nlk = nlk_sk(sk);
1446 if (p->exclude_sk == sk)
1449 if (nlk->portid == p->portid || p->group - 1 >= nlk->ngroups ||
1450 !test_bit(p->group - 1, nlk->groups))
1453 if (!net_eq(sock_net(sk), p->net)) {
1454 if (!nlk_test_bit(LISTEN_ALL_NSID, sk))
1457 if (!peernet_has_id(sock_net(sk), p->net))
1460 if (!file_ns_capable(sk->sk_socket->file, p->net->user_ns,
1466 netlink_overrun(sk);
1471 if (p->skb2 == NULL) {
1472 if (skb_shared(p->skb)) {
1473 p->skb2 = skb_clone(p->skb, p->allocation);
1475 p->skb2 = skb_get(p->skb);
1477 * skb ownership may have been set when
1478 * delivered to a previous socket.
1480 skb_orphan(p->skb2);
1483 if (p->skb2 == NULL) {
1484 netlink_overrun(sk);
1485 /* Clone failed. Notify ALL listeners. */
1487 if (nlk_test_bit(BROADCAST_SEND_ERROR, sk))
1488 p->delivery_failure = 1;
1492 if (p->tx_filter && p->tx_filter(sk, p->skb2, p->tx_data)) {
1498 if (sk_filter(sk, p->skb2)) {
1503 NETLINK_CB(p->skb2).nsid = peernet2id(sock_net(sk), p->net);
1504 if (NETLINK_CB(p->skb2).nsid != NETNSA_NSID_NOT_ASSIGNED)
1505 NETLINK_CB(p->skb2).nsid_is_set = true;
1506 val = netlink_broadcast_deliver(sk, p->skb2);
1508 netlink_overrun(sk);
1509 if (nlk_test_bit(BROADCAST_SEND_ERROR, sk))
1510 p->delivery_failure = 1;
1512 p->congested |= val;
1520 int netlink_broadcast_filtered(struct sock *ssk, struct sk_buff *skb,
1522 u32 group, gfp_t allocation,
1523 int (*filter)(struct sock *dsk,
1524 struct sk_buff *skb, void *data),
1527 struct net *net = sock_net(ssk);
1528 struct netlink_broadcast_data info;
1531 skb = netlink_trim(skb, allocation);
1533 info.exclude_sk = ssk;
1535 info.portid = portid;
1538 info.delivery_failure = 0;
1541 info.allocation = allocation;
1544 info.tx_filter = filter;
1545 info.tx_data = filter_data;
1547 /* While we sleep in clone, do not allow to change socket list */
1549 netlink_lock_table();
1551 sk_for_each_bound(sk, &nl_table[ssk->sk_protocol].mc_list)
1552 do_one_broadcast(sk, &info);
1556 netlink_unlock_table();
1558 if (info.delivery_failure) {
1559 kfree_skb(info.skb2);
1562 consume_skb(info.skb2);
1564 if (info.delivered) {
1565 if (info.congested && gfpflags_allow_blocking(allocation))
1571 EXPORT_SYMBOL(netlink_broadcast_filtered);
1573 int netlink_broadcast(struct sock *ssk, struct sk_buff *skb, u32 portid,
1574 u32 group, gfp_t allocation)
1576 return netlink_broadcast_filtered(ssk, skb, portid, group, allocation,
1579 EXPORT_SYMBOL(netlink_broadcast);
1581 struct netlink_set_err_data {
1582 struct sock *exclude_sk;
1588 static int do_one_set_err(struct sock *sk, struct netlink_set_err_data *p)
1590 struct netlink_sock *nlk = nlk_sk(sk);
1593 if (sk == p->exclude_sk)
1596 if (!net_eq(sock_net(sk), sock_net(p->exclude_sk)))
1599 if (nlk->portid == p->portid || p->group - 1 >= nlk->ngroups ||
1600 !test_bit(p->group - 1, nlk->groups))
1603 if (p->code == ENOBUFS && nlk_test_bit(RECV_NO_ENOBUFS, sk)) {
1608 sk->sk_err = p->code;
1609 sk_error_report(sk);
1615 * netlink_set_err - report error to broadcast listeners
1616 * @ssk: the kernel netlink socket, as returned by netlink_kernel_create()
1617 * @portid: the PORTID of a process that we want to skip (if any)
1618 * @group: the broadcast group that will notice the error
1619 * @code: error code, must be negative (as usual in kernelspace)
1621 * This function returns the number of broadcast listeners that have set the
1622 * NETLINK_NO_ENOBUFS socket option.
1624 int netlink_set_err(struct sock *ssk, u32 portid, u32 group, int code)
1626 struct netlink_set_err_data info;
1627 unsigned long flags;
1631 info.exclude_sk = ssk;
1632 info.portid = portid;
1634 /* sk->sk_err wants a positive error value */
1637 read_lock_irqsave(&nl_table_lock, flags);
1639 sk_for_each_bound(sk, &nl_table[ssk->sk_protocol].mc_list)
1640 ret += do_one_set_err(sk, &info);
1642 read_unlock_irqrestore(&nl_table_lock, flags);
1645 EXPORT_SYMBOL(netlink_set_err);
1647 /* must be called with netlink table grabbed */
1648 static void netlink_update_socket_mc(struct netlink_sock *nlk,
1652 int old, new = !!is_new, subscriptions;
1654 old = test_bit(group - 1, nlk->groups);
1655 subscriptions = nlk->subscriptions - old + new;
1656 __assign_bit(group - 1, nlk->groups, new);
1657 netlink_update_subscriptions(&nlk->sk, subscriptions);
1658 netlink_update_listeners(&nlk->sk);
1661 static int netlink_setsockopt(struct socket *sock, int level, int optname,
1662 sockptr_t optval, unsigned int optlen)
1664 struct sock *sk = sock->sk;
1665 struct netlink_sock *nlk = nlk_sk(sk);
1666 unsigned int val = 0;
1669 if (level != SOL_NETLINK)
1670 return -ENOPROTOOPT;
1672 if (optlen >= sizeof(int) &&
1673 copy_from_sockptr(&val, optval, sizeof(val)))
1677 case NETLINK_PKTINFO:
1678 nr = NETLINK_F_RECV_PKTINFO;
1680 case NETLINK_ADD_MEMBERSHIP:
1681 case NETLINK_DROP_MEMBERSHIP: {
1684 if (!netlink_allowed(sock, NL_CFG_F_NONROOT_RECV))
1686 err = netlink_realloc_groups(sk);
1689 if (!val || val - 1 >= nlk->ngroups)
1691 if (optname == NETLINK_ADD_MEMBERSHIP && nlk->netlink_bind) {
1692 err = nlk->netlink_bind(sock_net(sk), val);
1696 netlink_table_grab();
1697 netlink_update_socket_mc(nlk, val,
1698 optname == NETLINK_ADD_MEMBERSHIP);
1699 netlink_table_ungrab();
1700 if (optname == NETLINK_DROP_MEMBERSHIP && nlk->netlink_unbind)
1701 nlk->netlink_unbind(sock_net(sk), val);
1705 case NETLINK_BROADCAST_ERROR:
1706 nr = NETLINK_F_BROADCAST_SEND_ERROR;
1708 case NETLINK_NO_ENOBUFS:
1709 assign_bit(NETLINK_F_RECV_NO_ENOBUFS, &nlk->flags, val);
1711 clear_bit(NETLINK_S_CONGESTED, &nlk->state);
1712 wake_up_interruptible(&nlk->wait);
1715 case NETLINK_LISTEN_ALL_NSID:
1716 if (!ns_capable(sock_net(sk)->user_ns, CAP_NET_BROADCAST))
1718 nr = NETLINK_F_LISTEN_ALL_NSID;
1720 case NETLINK_CAP_ACK:
1721 nr = NETLINK_F_CAP_ACK;
1723 case NETLINK_EXT_ACK:
1724 nr = NETLINK_F_EXT_ACK;
1726 case NETLINK_GET_STRICT_CHK:
1727 nr = NETLINK_F_STRICT_CHK;
1730 return -ENOPROTOOPT;
1733 assign_bit(nr, &nlk->flags, val);
1737 static int netlink_getsockopt(struct socket *sock, int level, int optname,
1738 char __user *optval, int __user *optlen)
1740 struct sock *sk = sock->sk;
1741 struct netlink_sock *nlk = nlk_sk(sk);
1745 if (level != SOL_NETLINK)
1746 return -ENOPROTOOPT;
1748 if (get_user(len, optlen))
1754 case NETLINK_PKTINFO:
1755 flag = NETLINK_F_RECV_PKTINFO;
1757 case NETLINK_BROADCAST_ERROR:
1758 flag = NETLINK_F_BROADCAST_SEND_ERROR;
1760 case NETLINK_NO_ENOBUFS:
1761 flag = NETLINK_F_RECV_NO_ENOBUFS;
1763 case NETLINK_LIST_MEMBERSHIPS: {
1764 int pos, idx, shift, err = 0;
1766 netlink_lock_table();
1767 for (pos = 0; pos * 8 < nlk->ngroups; pos += sizeof(u32)) {
1768 if (len - pos < sizeof(u32))
1771 idx = pos / sizeof(unsigned long);
1772 shift = (pos % sizeof(unsigned long)) * 8;
1773 if (put_user((u32)(nlk->groups[idx] >> shift),
1774 (u32 __user *)(optval + pos))) {
1779 if (put_user(ALIGN(BITS_TO_BYTES(nlk->ngroups), sizeof(u32)), optlen))
1781 netlink_unlock_table();
1784 case NETLINK_CAP_ACK:
1785 flag = NETLINK_F_CAP_ACK;
1787 case NETLINK_EXT_ACK:
1788 flag = NETLINK_F_EXT_ACK;
1790 case NETLINK_GET_STRICT_CHK:
1791 flag = NETLINK_F_STRICT_CHK;
1794 return -ENOPROTOOPT;
1797 if (len < sizeof(int))
1801 val = test_bit(flag, &nlk->flags);
1803 if (put_user(len, optlen) ||
1804 copy_to_user(optval, &val, len))
1810 static void netlink_cmsg_recv_pktinfo(struct msghdr *msg, struct sk_buff *skb)
1812 struct nl_pktinfo info;
1814 info.group = NETLINK_CB(skb).dst_group;
1815 put_cmsg(msg, SOL_NETLINK, NETLINK_PKTINFO, sizeof(info), &info);
1818 static void netlink_cmsg_listen_all_nsid(struct sock *sk, struct msghdr *msg,
1819 struct sk_buff *skb)
1821 if (!NETLINK_CB(skb).nsid_is_set)
1824 put_cmsg(msg, SOL_NETLINK, NETLINK_LISTEN_ALL_NSID, sizeof(int),
1825 &NETLINK_CB(skb).nsid);
1828 static int netlink_sendmsg(struct socket *sock, struct msghdr *msg, size_t len)
1830 struct sock *sk = sock->sk;
1831 struct netlink_sock *nlk = nlk_sk(sk);
1832 DECLARE_SOCKADDR(struct sockaddr_nl *, addr, msg->msg_name);
1835 struct sk_buff *skb;
1837 struct scm_cookie scm;
1838 u32 netlink_skb_flags = 0;
1840 if (msg->msg_flags & MSG_OOB)
1844 pr_warn_once("Zero length message leads to an empty skb\n");
1848 err = scm_send(sock, msg, &scm, true);
1852 if (msg->msg_namelen) {
1854 if (msg->msg_namelen < sizeof(struct sockaddr_nl))
1856 if (addr->nl_family != AF_NETLINK)
1858 dst_portid = addr->nl_pid;
1859 dst_group = ffs(addr->nl_groups);
1861 if ((dst_group || dst_portid) &&
1862 !netlink_allowed(sock, NL_CFG_F_NONROOT_SEND))
1864 netlink_skb_flags |= NETLINK_SKB_DST;
1866 /* Paired with WRITE_ONCE() in netlink_connect() */
1867 dst_portid = READ_ONCE(nlk->dst_portid);
1868 dst_group = READ_ONCE(nlk->dst_group);
1871 /* Paired with WRITE_ONCE() in netlink_insert() */
1872 if (!READ_ONCE(nlk->bound)) {
1873 err = netlink_autobind(sock);
1877 /* Ensure nlk is hashed and visible. */
1882 if (len > sk->sk_sndbuf - 32)
1885 skb = netlink_alloc_large_skb(len, dst_group);
1889 NETLINK_CB(skb).portid = nlk->portid;
1890 NETLINK_CB(skb).dst_group = dst_group;
1891 NETLINK_CB(skb).creds = scm.creds;
1892 NETLINK_CB(skb).flags = netlink_skb_flags;
1895 if (memcpy_from_msg(skb_put(skb, len), msg, len)) {
1900 err = security_netlink_send(sk, skb);
1907 refcount_inc(&skb->users);
1908 netlink_broadcast(sk, skb, dst_portid, dst_group, GFP_KERNEL);
1910 err = netlink_unicast(sk, skb, dst_portid, msg->msg_flags & MSG_DONTWAIT);
1917 static int netlink_recvmsg(struct socket *sock, struct msghdr *msg, size_t len,
1920 struct scm_cookie scm;
1921 struct sock *sk = sock->sk;
1922 struct netlink_sock *nlk = nlk_sk(sk);
1923 size_t copied, max_recvmsg_len;
1924 struct sk_buff *skb, *data_skb;
1927 if (flags & MSG_OOB)
1932 skb = skb_recv_datagram(sk, flags, &err);
1938 #ifdef CONFIG_COMPAT_NETLINK_MESSAGES
1939 if (unlikely(skb_shinfo(skb)->frag_list)) {
1941 * If this skb has a frag_list, then here that means that we
1942 * will have to use the frag_list skb's data for compat tasks
1943 * and the regular skb's data for normal (non-compat) tasks.
1945 * If we need to send the compat skb, assign it to the
1946 * 'data_skb' variable so that it will be used below for data
1947 * copying. We keep 'skb' for everything else, including
1948 * freeing both later.
1950 if (flags & MSG_CMSG_COMPAT)
1951 data_skb = skb_shinfo(skb)->frag_list;
1955 /* Record the max length of recvmsg() calls for future allocations */
1956 max_recvmsg_len = max(READ_ONCE(nlk->max_recvmsg_len), len);
1957 max_recvmsg_len = min_t(size_t, max_recvmsg_len,
1958 SKB_WITH_OVERHEAD(32768));
1959 WRITE_ONCE(nlk->max_recvmsg_len, max_recvmsg_len);
1961 copied = data_skb->len;
1963 msg->msg_flags |= MSG_TRUNC;
1967 err = skb_copy_datagram_msg(data_skb, 0, msg, copied);
1969 if (msg->msg_name) {
1970 DECLARE_SOCKADDR(struct sockaddr_nl *, addr, msg->msg_name);
1971 addr->nl_family = AF_NETLINK;
1973 addr->nl_pid = NETLINK_CB(skb).portid;
1974 addr->nl_groups = netlink_group_mask(NETLINK_CB(skb).dst_group);
1975 msg->msg_namelen = sizeof(*addr);
1978 if (nlk_test_bit(RECV_PKTINFO, sk))
1979 netlink_cmsg_recv_pktinfo(msg, skb);
1980 if (nlk_test_bit(LISTEN_ALL_NSID, sk))
1981 netlink_cmsg_listen_all_nsid(sk, msg, skb);
1983 memset(&scm, 0, sizeof(scm));
1984 scm.creds = *NETLINK_CREDS(skb);
1985 if (flags & MSG_TRUNC)
1986 copied = data_skb->len;
1988 skb_free_datagram(sk, skb);
1990 if (READ_ONCE(nlk->cb_running) &&
1991 atomic_read(&sk->sk_rmem_alloc) <= sk->sk_rcvbuf / 2) {
1992 ret = netlink_dump(sk);
1995 sk_error_report(sk);
1999 scm_recv(sock, msg, &scm, flags);
2001 netlink_rcv_wake(sk);
2002 return err ? : copied;
2005 static void netlink_data_ready(struct sock *sk)
2011 * We export these functions to other modules. They provide a
2012 * complete set of kernel non-blocking support for message
2017 __netlink_kernel_create(struct net *net, int unit, struct module *module,
2018 struct netlink_kernel_cfg *cfg)
2020 struct socket *sock;
2022 struct netlink_sock *nlk;
2023 struct listeners *listeners = NULL;
2024 struct mutex *cb_mutex = cfg ? cfg->cb_mutex : NULL;
2025 unsigned int groups;
2029 if (unit < 0 || unit >= MAX_LINKS)
2032 if (sock_create_lite(PF_NETLINK, SOCK_DGRAM, unit, &sock))
2035 if (__netlink_create(net, sock, cb_mutex, unit, 1) < 0)
2036 goto out_sock_release_nosk;
2040 if (!cfg || cfg->groups < 32)
2043 groups = cfg->groups;
2045 listeners = kzalloc(sizeof(*listeners) + NLGRPSZ(groups), GFP_KERNEL);
2047 goto out_sock_release;
2049 sk->sk_data_ready = netlink_data_ready;
2050 if (cfg && cfg->input)
2051 nlk_sk(sk)->netlink_rcv = cfg->input;
2053 if (netlink_insert(sk, 0))
2054 goto out_sock_release;
2057 set_bit(NETLINK_F_KERNEL_SOCKET, &nlk->flags);
2059 netlink_table_grab();
2060 if (!nl_table[unit].registered) {
2061 nl_table[unit].groups = groups;
2062 rcu_assign_pointer(nl_table[unit].listeners, listeners);
2063 nl_table[unit].cb_mutex = cb_mutex;
2064 nl_table[unit].module = module;
2066 nl_table[unit].bind = cfg->bind;
2067 nl_table[unit].unbind = cfg->unbind;
2068 nl_table[unit].release = cfg->release;
2069 nl_table[unit].flags = cfg->flags;
2071 nl_table[unit].registered = 1;
2074 nl_table[unit].registered++;
2076 netlink_table_ungrab();
2081 netlink_kernel_release(sk);
2084 out_sock_release_nosk:
2088 EXPORT_SYMBOL(__netlink_kernel_create);
2091 netlink_kernel_release(struct sock *sk)
2093 if (sk == NULL || sk->sk_socket == NULL)
2096 sock_release(sk->sk_socket);
2098 EXPORT_SYMBOL(netlink_kernel_release);
2100 int __netlink_change_ngroups(struct sock *sk, unsigned int groups)
2102 struct listeners *new, *old;
2103 struct netlink_table *tbl = &nl_table[sk->sk_protocol];
2108 if (NLGRPSZ(tbl->groups) < NLGRPSZ(groups)) {
2109 new = kzalloc(sizeof(*new) + NLGRPSZ(groups), GFP_ATOMIC);
2112 old = nl_deref_protected(tbl->listeners);
2113 memcpy(new->masks, old->masks, NLGRPSZ(tbl->groups));
2114 rcu_assign_pointer(tbl->listeners, new);
2116 kfree_rcu(old, rcu);
2118 tbl->groups = groups;
2124 * netlink_change_ngroups - change number of multicast groups
2126 * This changes the number of multicast groups that are available
2127 * on a certain netlink family. Note that it is not possible to
2128 * change the number of groups to below 32. Also note that it does
2129 * not implicitly call netlink_clear_multicast_users() when the
2130 * number of groups is reduced.
2132 * @sk: The kernel netlink socket, as returned by netlink_kernel_create().
2133 * @groups: The new number of groups.
2135 int netlink_change_ngroups(struct sock *sk, unsigned int groups)
2139 netlink_table_grab();
2140 err = __netlink_change_ngroups(sk, groups);
2141 netlink_table_ungrab();
2146 void __netlink_clear_multicast_users(struct sock *ksk, unsigned int group)
2149 struct netlink_table *tbl = &nl_table[ksk->sk_protocol];
2151 sk_for_each_bound(sk, &tbl->mc_list)
2152 netlink_update_socket_mc(nlk_sk(sk), group, 0);
2156 __nlmsg_put(struct sk_buff *skb, u32 portid, u32 seq, int type, int len, int flags)
2158 struct nlmsghdr *nlh;
2159 int size = nlmsg_msg_size(len);
2161 nlh = skb_put(skb, NLMSG_ALIGN(size));
2162 nlh->nlmsg_type = type;
2163 nlh->nlmsg_len = size;
2164 nlh->nlmsg_flags = flags;
2165 nlh->nlmsg_pid = portid;
2166 nlh->nlmsg_seq = seq;
2167 if (!__builtin_constant_p(size) || NLMSG_ALIGN(size) - size != 0)
2168 memset(nlmsg_data(nlh) + len, 0, NLMSG_ALIGN(size) - size);
2171 EXPORT_SYMBOL(__nlmsg_put);
2174 * It looks a bit ugly.
2175 * It would be better to create kernel thread.
2178 static int netlink_dump_done(struct netlink_sock *nlk, struct sk_buff *skb,
2179 struct netlink_callback *cb,
2180 struct netlink_ext_ack *extack)
2182 struct nlmsghdr *nlh;
2184 nlh = nlmsg_put_answer(skb, cb, NLMSG_DONE, sizeof(nlk->dump_done_errno),
2185 NLM_F_MULTI | cb->answer_flags);
2189 nl_dump_check_consistent(cb, nlh);
2190 memcpy(nlmsg_data(nlh), &nlk->dump_done_errno, sizeof(nlk->dump_done_errno));
2192 if (extack->_msg && test_bit(NETLINK_F_EXT_ACK, &nlk->flags)) {
2193 nlh->nlmsg_flags |= NLM_F_ACK_TLVS;
2194 if (!nla_put_string(skb, NLMSGERR_ATTR_MSG, extack->_msg))
2195 nlmsg_end(skb, nlh);
2201 static int netlink_dump(struct sock *sk)
2203 struct netlink_sock *nlk = nlk_sk(sk);
2204 struct netlink_ext_ack extack = {};
2205 struct netlink_callback *cb;
2206 struct sk_buff *skb = NULL;
2207 size_t max_recvmsg_len;
2208 struct module *module;
2213 mutex_lock(nlk->cb_mutex);
2214 if (!nlk->cb_running) {
2219 if (atomic_read(&sk->sk_rmem_alloc) >= sk->sk_rcvbuf)
2222 /* NLMSG_GOODSIZE is small to avoid high order allocations being
2223 * required, but it makes sense to _attempt_ a 16K bytes allocation
2224 * to reduce number of system calls on dump operations, if user
2225 * ever provided a big enough buffer.
2228 alloc_min_size = max_t(int, cb->min_dump_alloc, NLMSG_GOODSIZE);
2230 max_recvmsg_len = READ_ONCE(nlk->max_recvmsg_len);
2231 if (alloc_min_size < max_recvmsg_len) {
2232 alloc_size = max_recvmsg_len;
2233 skb = alloc_skb(alloc_size,
2234 (GFP_KERNEL & ~__GFP_DIRECT_RECLAIM) |
2235 __GFP_NOWARN | __GFP_NORETRY);
2238 alloc_size = alloc_min_size;
2239 skb = alloc_skb(alloc_size, GFP_KERNEL);
2244 /* Trim skb to allocated size. User is expected to provide buffer as
2245 * large as max(min_dump_alloc, 16KiB (mac_recvmsg_len capped at
2246 * netlink_recvmsg())). dump will pack as many smaller messages as
2247 * could fit within the allocated skb. skb is typically allocated
2248 * with larger space than required (could be as much as near 2x the
2249 * requested size with align to next power of 2 approach). Allowing
2250 * dump to use the excess space makes it difficult for a user to have a
2251 * reasonable static buffer based on the expected largest dump of a
2252 * single netdev. The outcome is MSG_TRUNC error.
2254 skb_reserve(skb, skb_tailroom(skb) - alloc_size);
2256 /* Make sure malicious BPF programs can not read unitialized memory
2257 * from skb->head -> skb->data
2259 skb_reset_network_header(skb);
2260 skb_reset_mac_header(skb);
2262 netlink_skb_set_owner_r(skb, sk);
2264 if (nlk->dump_done_errno > 0) {
2265 cb->extack = &extack;
2266 nlk->dump_done_errno = cb->dump(skb, cb);
2270 if (nlk->dump_done_errno > 0 ||
2271 skb_tailroom(skb) < nlmsg_total_size(sizeof(nlk->dump_done_errno))) {
2272 mutex_unlock(nlk->cb_mutex);
2274 if (sk_filter(sk, skb))
2277 __netlink_sendskb(sk, skb);
2281 if (netlink_dump_done(nlk, skb, cb, &extack))
2284 #ifdef CONFIG_COMPAT_NETLINK_MESSAGES
2285 /* frag_list skb's data is used for compat tasks
2286 * and the regular skb's data for normal (non-compat) tasks.
2287 * See netlink_recvmsg().
2289 if (unlikely(skb_shinfo(skb)->frag_list)) {
2290 if (netlink_dump_done(nlk, skb_shinfo(skb)->frag_list, cb, &extack))
2295 if (sk_filter(sk, skb))
2298 __netlink_sendskb(sk, skb);
2303 WRITE_ONCE(nlk->cb_running, false);
2304 module = cb->module;
2306 mutex_unlock(nlk->cb_mutex);
2312 mutex_unlock(nlk->cb_mutex);
2317 int __netlink_dump_start(struct sock *ssk, struct sk_buff *skb,
2318 const struct nlmsghdr *nlh,
2319 struct netlink_dump_control *control)
2321 struct netlink_callback *cb;
2322 struct netlink_sock *nlk;
2326 refcount_inc(&skb->users);
2328 sk = netlink_lookup(sock_net(ssk), ssk->sk_protocol, NETLINK_CB(skb).portid);
2330 ret = -ECONNREFUSED;
2335 mutex_lock(nlk->cb_mutex);
2336 /* A dump is in progress... */
2337 if (nlk->cb_running) {
2341 /* add reference of module which cb->dump belongs to */
2342 if (!try_module_get(control->module)) {
2343 ret = -EPROTONOSUPPORT;
2348 memset(cb, 0, sizeof(*cb));
2349 cb->dump = control->dump;
2350 cb->done = control->done;
2352 cb->data = control->data;
2353 cb->module = control->module;
2354 cb->min_dump_alloc = control->min_dump_alloc;
2357 cb->strict_check = nlk_test_bit(STRICT_CHK, NETLINK_CB(skb).sk);
2359 if (control->start) {
2360 cb->extack = control->extack;
2361 ret = control->start(cb);
2367 WRITE_ONCE(nlk->cb_running, true);
2368 nlk->dump_done_errno = INT_MAX;
2370 mutex_unlock(nlk->cb_mutex);
2372 ret = netlink_dump(sk);
2379 /* We successfully started a dump, by returning -EINTR we
2380 * signal not to send ACK even if it was requested.
2385 module_put(control->module);
2388 mutex_unlock(nlk->cb_mutex);
2393 EXPORT_SYMBOL(__netlink_dump_start);
2396 netlink_ack_tlv_len(struct netlink_sock *nlk, int err,
2397 const struct netlink_ext_ack *extack)
2401 if (!extack || !test_bit(NETLINK_F_EXT_ACK, &nlk->flags))
2406 tlvlen += nla_total_size(strlen(extack->_msg) + 1);
2407 if (extack->cookie_len)
2408 tlvlen += nla_total_size(extack->cookie_len);
2410 /* Following attributes are only reported as error (not warning) */
2414 if (extack->bad_attr)
2415 tlvlen += nla_total_size(sizeof(u32));
2417 tlvlen += netlink_policy_dump_attr_size_estimate(extack->policy);
2418 if (extack->miss_type)
2419 tlvlen += nla_total_size(sizeof(u32));
2420 if (extack->miss_nest)
2421 tlvlen += nla_total_size(sizeof(u32));
2427 netlink_ack_tlv_fill(struct sk_buff *in_skb, struct sk_buff *skb,
2428 struct nlmsghdr *nlh, int err,
2429 const struct netlink_ext_ack *extack)
2432 WARN_ON(nla_put_string(skb, NLMSGERR_ATTR_MSG, extack->_msg));
2433 if (extack->cookie_len)
2434 WARN_ON(nla_put(skb, NLMSGERR_ATTR_COOKIE,
2435 extack->cookie_len, extack->cookie));
2440 if (extack->bad_attr &&
2441 !WARN_ON((u8 *)extack->bad_attr < in_skb->data ||
2442 (u8 *)extack->bad_attr >= in_skb->data + in_skb->len))
2443 WARN_ON(nla_put_u32(skb, NLMSGERR_ATTR_OFFS,
2444 (u8 *)extack->bad_attr - (u8 *)nlh));
2446 netlink_policy_dump_write_attr(skb, extack->policy,
2447 NLMSGERR_ATTR_POLICY);
2448 if (extack->miss_type)
2449 WARN_ON(nla_put_u32(skb, NLMSGERR_ATTR_MISS_TYPE,
2450 extack->miss_type));
2451 if (extack->miss_nest &&
2452 !WARN_ON((u8 *)extack->miss_nest < in_skb->data ||
2453 (u8 *)extack->miss_nest > in_skb->data + in_skb->len))
2454 WARN_ON(nla_put_u32(skb, NLMSGERR_ATTR_MISS_NEST,
2455 (u8 *)extack->miss_nest - (u8 *)nlh));
2458 void netlink_ack(struct sk_buff *in_skb, struct nlmsghdr *nlh, int err,
2459 const struct netlink_ext_ack *extack)
2461 struct sk_buff *skb;
2462 struct nlmsghdr *rep;
2463 struct nlmsgerr *errmsg;
2464 size_t payload = sizeof(*errmsg);
2465 struct netlink_sock *nlk = nlk_sk(NETLINK_CB(in_skb).sk);
2466 unsigned int flags = 0;
2469 /* Error messages get the original request appened, unless the user
2470 * requests to cap the error message, and get extra error data if
2473 if (err && !test_bit(NETLINK_F_CAP_ACK, &nlk->flags))
2474 payload += nlmsg_len(nlh);
2476 flags |= NLM_F_CAPPED;
2478 tlvlen = netlink_ack_tlv_len(nlk, err, extack);
2480 flags |= NLM_F_ACK_TLVS;
2482 skb = nlmsg_new(payload + tlvlen, GFP_KERNEL);
2486 rep = nlmsg_put(skb, NETLINK_CB(in_skb).portid, nlh->nlmsg_seq,
2487 NLMSG_ERROR, sizeof(*errmsg), flags);
2490 errmsg = nlmsg_data(rep);
2491 errmsg->error = err;
2494 if (!(flags & NLM_F_CAPPED)) {
2495 if (!nlmsg_append(skb, nlmsg_len(nlh)))
2498 memcpy(nlmsg_data(&errmsg->msg), nlmsg_data(nlh),
2503 netlink_ack_tlv_fill(in_skb, skb, nlh, err, extack);
2505 nlmsg_end(skb, rep);
2507 nlmsg_unicast(in_skb->sk, skb, NETLINK_CB(in_skb).portid);
2514 NETLINK_CB(in_skb).sk->sk_err = ENOBUFS;
2515 sk_error_report(NETLINK_CB(in_skb).sk);
2517 EXPORT_SYMBOL(netlink_ack);
2519 int netlink_rcv_skb(struct sk_buff *skb, int (*cb)(struct sk_buff *,
2521 struct netlink_ext_ack *))
2523 struct netlink_ext_ack extack;
2524 struct nlmsghdr *nlh;
2527 while (skb->len >= nlmsg_total_size(0)) {
2530 memset(&extack, 0, sizeof(extack));
2531 nlh = nlmsg_hdr(skb);
2534 if (nlh->nlmsg_len < NLMSG_HDRLEN || skb->len < nlh->nlmsg_len)
2537 /* Only requests are handled by the kernel */
2538 if (!(nlh->nlmsg_flags & NLM_F_REQUEST))
2541 /* Skip control messages */
2542 if (nlh->nlmsg_type < NLMSG_MIN_TYPE)
2545 err = cb(skb, nlh, &extack);
2550 if (nlh->nlmsg_flags & NLM_F_ACK || err)
2551 netlink_ack(skb, nlh, err, &extack);
2554 msglen = NLMSG_ALIGN(nlh->nlmsg_len);
2555 if (msglen > skb->len)
2557 skb_pull(skb, msglen);
2562 EXPORT_SYMBOL(netlink_rcv_skb);
2565 * nlmsg_notify - send a notification netlink message
2566 * @sk: netlink socket to use
2567 * @skb: notification message
2568 * @portid: destination netlink portid for reports or 0
2569 * @group: destination multicast group or 0
2570 * @report: 1 to report back, 0 to disable
2571 * @flags: allocation flags
2573 int nlmsg_notify(struct sock *sk, struct sk_buff *skb, u32 portid,
2574 unsigned int group, int report, gfp_t flags)
2579 int exclude_portid = 0;
2582 refcount_inc(&skb->users);
2583 exclude_portid = portid;
2586 /* errors reported via destination sk->sk_err, but propagate
2587 * delivery errors if NETLINK_BROADCAST_ERROR flag is set */
2588 err = nlmsg_multicast(sk, skb, exclude_portid, group, flags);
2596 err2 = nlmsg_unicast(sk, skb, portid);
2603 EXPORT_SYMBOL(nlmsg_notify);
2605 #ifdef CONFIG_PROC_FS
2606 struct nl_seq_iter {
2607 struct seq_net_private p;
2608 struct rhashtable_iter hti;
2612 static void netlink_walk_start(struct nl_seq_iter *iter)
2614 rhashtable_walk_enter(&nl_table[iter->link].hash, &iter->hti);
2615 rhashtable_walk_start(&iter->hti);
2618 static void netlink_walk_stop(struct nl_seq_iter *iter)
2620 rhashtable_walk_stop(&iter->hti);
2621 rhashtable_walk_exit(&iter->hti);
2624 static void *__netlink_seq_next(struct seq_file *seq)
2626 struct nl_seq_iter *iter = seq->private;
2627 struct netlink_sock *nlk;
2631 nlk = rhashtable_walk_next(&iter->hti);
2634 if (PTR_ERR(nlk) == -EAGAIN)
2643 netlink_walk_stop(iter);
2644 if (++iter->link >= MAX_LINKS)
2647 netlink_walk_start(iter);
2649 } while (sock_net(&nlk->sk) != seq_file_net(seq));
2654 static void *netlink_seq_start(struct seq_file *seq, loff_t *posp)
2657 struct nl_seq_iter *iter = seq->private;
2658 void *obj = SEQ_START_TOKEN;
2663 netlink_walk_start(iter);
2665 for (pos = *posp; pos && obj && !IS_ERR(obj); pos--)
2666 obj = __netlink_seq_next(seq);
2671 static void *netlink_seq_next(struct seq_file *seq, void *v, loff_t *pos)
2674 return __netlink_seq_next(seq);
2677 static void netlink_native_seq_stop(struct seq_file *seq, void *v)
2679 struct nl_seq_iter *iter = seq->private;
2681 if (iter->link >= MAX_LINKS)
2684 netlink_walk_stop(iter);
2688 static int netlink_native_seq_show(struct seq_file *seq, void *v)
2690 if (v == SEQ_START_TOKEN) {
2692 "sk Eth Pid Groups "
2693 "Rmem Wmem Dump Locks Drops Inode\n");
2696 struct netlink_sock *nlk = nlk_sk(s);
2698 seq_printf(seq, "%pK %-3d %-10u %08x %-8d %-8d %-5d %-8d %-8u %-8lu\n",
2702 nlk->groups ? (u32)nlk->groups[0] : 0,
2703 sk_rmem_alloc_get(s),
2704 sk_wmem_alloc_get(s),
2705 READ_ONCE(nlk->cb_running),
2706 refcount_read(&s->sk_refcnt),
2707 atomic_read(&s->sk_drops),
2715 #ifdef CONFIG_BPF_SYSCALL
2716 struct bpf_iter__netlink {
2717 __bpf_md_ptr(struct bpf_iter_meta *, meta);
2718 __bpf_md_ptr(struct netlink_sock *, sk);
2721 DEFINE_BPF_ITER_FUNC(netlink, struct bpf_iter_meta *meta, struct netlink_sock *sk)
2723 static int netlink_prog_seq_show(struct bpf_prog *prog,
2724 struct bpf_iter_meta *meta,
2727 struct bpf_iter__netlink ctx;
2729 meta->seq_num--; /* skip SEQ_START_TOKEN */
2731 ctx.sk = nlk_sk((struct sock *)v);
2732 return bpf_iter_run_prog(prog, &ctx);
2735 static int netlink_seq_show(struct seq_file *seq, void *v)
2737 struct bpf_iter_meta meta;
2738 struct bpf_prog *prog;
2741 prog = bpf_iter_get_info(&meta, false);
2743 return netlink_native_seq_show(seq, v);
2745 if (v != SEQ_START_TOKEN)
2746 return netlink_prog_seq_show(prog, &meta, v);
2751 static void netlink_seq_stop(struct seq_file *seq, void *v)
2753 struct bpf_iter_meta meta;
2754 struct bpf_prog *prog;
2758 prog = bpf_iter_get_info(&meta, true);
2760 (void)netlink_prog_seq_show(prog, &meta, v);
2763 netlink_native_seq_stop(seq, v);
2766 static int netlink_seq_show(struct seq_file *seq, void *v)
2768 return netlink_native_seq_show(seq, v);
2771 static void netlink_seq_stop(struct seq_file *seq, void *v)
2773 netlink_native_seq_stop(seq, v);
2777 static const struct seq_operations netlink_seq_ops = {
2778 .start = netlink_seq_start,
2779 .next = netlink_seq_next,
2780 .stop = netlink_seq_stop,
2781 .show = netlink_seq_show,
2785 int netlink_register_notifier(struct notifier_block *nb)
2787 return blocking_notifier_chain_register(&netlink_chain, nb);
2789 EXPORT_SYMBOL(netlink_register_notifier);
2791 int netlink_unregister_notifier(struct notifier_block *nb)
2793 return blocking_notifier_chain_unregister(&netlink_chain, nb);
2795 EXPORT_SYMBOL(netlink_unregister_notifier);
2797 static const struct proto_ops netlink_ops = {
2798 .family = PF_NETLINK,
2799 .owner = THIS_MODULE,
2800 .release = netlink_release,
2801 .bind = netlink_bind,
2802 .connect = netlink_connect,
2803 .socketpair = sock_no_socketpair,
2804 .accept = sock_no_accept,
2805 .getname = netlink_getname,
2806 .poll = datagram_poll,
2807 .ioctl = netlink_ioctl,
2808 .listen = sock_no_listen,
2809 .shutdown = sock_no_shutdown,
2810 .setsockopt = netlink_setsockopt,
2811 .getsockopt = netlink_getsockopt,
2812 .sendmsg = netlink_sendmsg,
2813 .recvmsg = netlink_recvmsg,
2814 .mmap = sock_no_mmap,
2817 static const struct net_proto_family netlink_family_ops = {
2818 .family = PF_NETLINK,
2819 .create = netlink_create,
2820 .owner = THIS_MODULE, /* for consistency 8) */
2823 static int __net_init netlink_net_init(struct net *net)
2825 #ifdef CONFIG_PROC_FS
2826 if (!proc_create_net("netlink", 0, net->proc_net, &netlink_seq_ops,
2827 sizeof(struct nl_seq_iter)))
2833 static void __net_exit netlink_net_exit(struct net *net)
2835 #ifdef CONFIG_PROC_FS
2836 remove_proc_entry("netlink", net->proc_net);
2840 static void __init netlink_add_usersock_entry(void)
2842 struct listeners *listeners;
2845 listeners = kzalloc(sizeof(*listeners) + NLGRPSZ(groups), GFP_KERNEL);
2847 panic("netlink_add_usersock_entry: Cannot allocate listeners\n");
2849 netlink_table_grab();
2851 nl_table[NETLINK_USERSOCK].groups = groups;
2852 rcu_assign_pointer(nl_table[NETLINK_USERSOCK].listeners, listeners);
2853 nl_table[NETLINK_USERSOCK].module = THIS_MODULE;
2854 nl_table[NETLINK_USERSOCK].registered = 1;
2855 nl_table[NETLINK_USERSOCK].flags = NL_CFG_F_NONROOT_SEND;
2857 netlink_table_ungrab();
2860 static struct pernet_operations __net_initdata netlink_net_ops = {
2861 .init = netlink_net_init,
2862 .exit = netlink_net_exit,
2865 static inline u32 netlink_hash(const void *data, u32 len, u32 seed)
2867 const struct netlink_sock *nlk = data;
2868 struct netlink_compare_arg arg;
2870 netlink_compare_arg_init(&arg, sock_net(&nlk->sk), nlk->portid);
2871 return jhash2((u32 *)&arg, netlink_compare_arg_len / sizeof(u32), seed);
2874 static const struct rhashtable_params netlink_rhashtable_params = {
2875 .head_offset = offsetof(struct netlink_sock, node),
2876 .key_len = netlink_compare_arg_len,
2877 .obj_hashfn = netlink_hash,
2878 .obj_cmpfn = netlink_compare,
2879 .automatic_shrinking = true,
2882 #if defined(CONFIG_BPF_SYSCALL) && defined(CONFIG_PROC_FS)
2883 BTF_ID_LIST(btf_netlink_sock_id)
2884 BTF_ID(struct, netlink_sock)
2886 static const struct bpf_iter_seq_info netlink_seq_info = {
2887 .seq_ops = &netlink_seq_ops,
2888 .init_seq_private = bpf_iter_init_seq_net,
2889 .fini_seq_private = bpf_iter_fini_seq_net,
2890 .seq_priv_size = sizeof(struct nl_seq_iter),
2893 static struct bpf_iter_reg netlink_reg_info = {
2894 .target = "netlink",
2895 .ctx_arg_info_size = 1,
2897 { offsetof(struct bpf_iter__netlink, sk),
2898 PTR_TO_BTF_ID_OR_NULL },
2900 .seq_info = &netlink_seq_info,
2903 static int __init bpf_iter_register(void)
2905 netlink_reg_info.ctx_arg_info[0].btf_id = *btf_netlink_sock_id;
2906 return bpf_iter_reg_target(&netlink_reg_info);
2910 static int __init netlink_proto_init(void)
2913 int err = proto_register(&netlink_proto, 0);
2918 #if defined(CONFIG_BPF_SYSCALL) && defined(CONFIG_PROC_FS)
2919 err = bpf_iter_register();
2924 BUILD_BUG_ON(sizeof(struct netlink_skb_parms) > sizeof_field(struct sk_buff, cb));
2926 nl_table = kcalloc(MAX_LINKS, sizeof(*nl_table), GFP_KERNEL);
2930 for (i = 0; i < MAX_LINKS; i++) {
2931 if (rhashtable_init(&nl_table[i].hash,
2932 &netlink_rhashtable_params) < 0) {
2934 rhashtable_destroy(&nl_table[i].hash);
2940 netlink_add_usersock_entry();
2942 sock_register(&netlink_family_ops);
2943 register_pernet_subsys(&netlink_net_ops);
2944 register_pernet_subsys(&netlink_tap_net_ops);
2945 /* The netlink device handler may be needed early. */
2950 panic("netlink_init: Cannot allocate nl_table\n");
2953 core_initcall(netlink_proto_init);