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, bool lock_taken);
135 /* nl_table locking explained:
136 * Lookup and traversal are protected with an RCU read-side lock. Insertion
137 * and removal are protected with per bucket lock while using RCU list
138 * modification primitives and may run in parallel to RCU protected lookups.
139 * Destruction of the Netlink socket may only occur *after* nl_table_lock has
140 * been acquired * either during or after the socket has been removed from
141 * the list and after an RCU grace period.
143 DEFINE_RWLOCK(nl_table_lock);
144 EXPORT_SYMBOL_GPL(nl_table_lock);
145 static atomic_t nl_table_users = ATOMIC_INIT(0);
147 #define nl_deref_protected(X) rcu_dereference_protected(X, lockdep_is_held(&nl_table_lock));
149 static BLOCKING_NOTIFIER_HEAD(netlink_chain);
152 static const struct rhashtable_params netlink_rhashtable_params;
154 void do_trace_netlink_extack(const char *msg)
156 trace_netlink_extack(msg);
158 EXPORT_SYMBOL(do_trace_netlink_extack);
160 static inline u32 netlink_group_mask(u32 group)
164 return group ? 1 << (group - 1) : 0;
167 static struct sk_buff *netlink_to_full_skb(const struct sk_buff *skb,
170 unsigned int len = skb->len;
173 new = alloc_skb(len, gfp_mask);
177 NETLINK_CB(new).portid = NETLINK_CB(skb).portid;
178 NETLINK_CB(new).dst_group = NETLINK_CB(skb).dst_group;
179 NETLINK_CB(new).creds = NETLINK_CB(skb).creds;
181 skb_put_data(new, skb->data, len);
185 static unsigned int netlink_tap_net_id;
187 struct netlink_tap_net {
188 struct list_head netlink_tap_all;
189 struct mutex netlink_tap_lock;
192 int netlink_add_tap(struct netlink_tap *nt)
194 struct net *net = dev_net(nt->dev);
195 struct netlink_tap_net *nn = net_generic(net, netlink_tap_net_id);
197 if (unlikely(nt->dev->type != ARPHRD_NETLINK))
200 mutex_lock(&nn->netlink_tap_lock);
201 list_add_rcu(&nt->list, &nn->netlink_tap_all);
202 mutex_unlock(&nn->netlink_tap_lock);
204 __module_get(nt->module);
208 EXPORT_SYMBOL_GPL(netlink_add_tap);
210 static int __netlink_remove_tap(struct netlink_tap *nt)
212 struct net *net = dev_net(nt->dev);
213 struct netlink_tap_net *nn = net_generic(net, netlink_tap_net_id);
215 struct netlink_tap *tmp;
217 mutex_lock(&nn->netlink_tap_lock);
219 list_for_each_entry(tmp, &nn->netlink_tap_all, list) {
221 list_del_rcu(&nt->list);
227 pr_warn("__netlink_remove_tap: %p not found\n", nt);
229 mutex_unlock(&nn->netlink_tap_lock);
232 module_put(nt->module);
234 return found ? 0 : -ENODEV;
237 int netlink_remove_tap(struct netlink_tap *nt)
241 ret = __netlink_remove_tap(nt);
246 EXPORT_SYMBOL_GPL(netlink_remove_tap);
248 static __net_init int netlink_tap_init_net(struct net *net)
250 struct netlink_tap_net *nn = net_generic(net, netlink_tap_net_id);
252 INIT_LIST_HEAD(&nn->netlink_tap_all);
253 mutex_init(&nn->netlink_tap_lock);
257 static struct pernet_operations netlink_tap_net_ops = {
258 .init = netlink_tap_init_net,
259 .id = &netlink_tap_net_id,
260 .size = sizeof(struct netlink_tap_net),
263 static bool netlink_filter_tap(const struct sk_buff *skb)
265 struct sock *sk = skb->sk;
267 /* We take the more conservative approach and
268 * whitelist socket protocols that may pass.
270 switch (sk->sk_protocol) {
272 case NETLINK_USERSOCK:
273 case NETLINK_SOCK_DIAG:
276 case NETLINK_FIB_LOOKUP:
277 case NETLINK_NETFILTER:
278 case NETLINK_GENERIC:
285 static int __netlink_deliver_tap_skb(struct sk_buff *skb,
286 struct net_device *dev)
288 struct sk_buff *nskb;
289 struct sock *sk = skb->sk;
292 if (!net_eq(dev_net(dev), sock_net(sk)))
297 if (is_vmalloc_addr(skb->head))
298 nskb = netlink_to_full_skb(skb, GFP_ATOMIC);
300 nskb = skb_clone(skb, GFP_ATOMIC);
303 nskb->protocol = htons((u16) sk->sk_protocol);
304 nskb->pkt_type = netlink_is_kernel(sk) ?
305 PACKET_KERNEL : PACKET_USER;
306 skb_reset_network_header(nskb);
307 ret = dev_queue_xmit(nskb);
308 if (unlikely(ret > 0))
309 ret = net_xmit_errno(ret);
316 static void __netlink_deliver_tap(struct sk_buff *skb, struct netlink_tap_net *nn)
319 struct netlink_tap *tmp;
321 if (!netlink_filter_tap(skb))
324 list_for_each_entry_rcu(tmp, &nn->netlink_tap_all, list) {
325 ret = __netlink_deliver_tap_skb(skb, tmp->dev);
331 static void netlink_deliver_tap(struct net *net, struct sk_buff *skb)
333 struct netlink_tap_net *nn = net_generic(net, netlink_tap_net_id);
337 if (unlikely(!list_empty(&nn->netlink_tap_all)))
338 __netlink_deliver_tap(skb, nn);
343 static void netlink_deliver_tap_kernel(struct sock *dst, struct sock *src,
346 if (!(netlink_is_kernel(dst) && netlink_is_kernel(src)))
347 netlink_deliver_tap(sock_net(dst), skb);
350 static void netlink_overrun(struct sock *sk)
352 if (!nlk_test_bit(RECV_NO_ENOBUFS, sk)) {
353 if (!test_and_set_bit(NETLINK_S_CONGESTED,
354 &nlk_sk(sk)->state)) {
355 WRITE_ONCE(sk->sk_err, ENOBUFS);
359 atomic_inc(&sk->sk_drops);
362 static void netlink_rcv_wake(struct sock *sk)
364 struct netlink_sock *nlk = nlk_sk(sk);
366 if (skb_queue_empty_lockless(&sk->sk_receive_queue))
367 clear_bit(NETLINK_S_CONGESTED, &nlk->state);
368 if (!test_bit(NETLINK_S_CONGESTED, &nlk->state))
369 wake_up_interruptible(&nlk->wait);
372 static void netlink_skb_destructor(struct sk_buff *skb)
374 if (is_vmalloc_addr(skb->head)) {
376 !atomic_dec_return(&(skb_shinfo(skb)->dataref)))
377 vfree_atomic(skb->head);
385 static void netlink_skb_set_owner_r(struct sk_buff *skb, struct sock *sk)
387 WARN_ON(skb->sk != NULL);
389 skb->destructor = netlink_skb_destructor;
390 atomic_add(skb->truesize, &sk->sk_rmem_alloc);
391 sk_mem_charge(sk, skb->truesize);
394 static void netlink_sock_destruct(struct sock *sk)
396 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 *dump_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);
654 mutex_init(&nlk->nl_cb_mutex);
655 lockdep_set_class_and_name(&nlk->nl_cb_mutex,
656 nlk_cb_mutex_keys + protocol,
657 nlk_cb_mutex_key_strings[protocol]);
658 nlk->dump_cb_mutex = dump_cb_mutex;
659 init_waitqueue_head(&nlk->wait);
661 sk->sk_destruct = netlink_sock_destruct;
662 sk->sk_protocol = protocol;
666 static int netlink_create(struct net *net, struct socket *sock, int protocol,
669 struct module *module = NULL;
670 struct mutex *cb_mutex;
671 struct netlink_sock *nlk;
672 int (*bind)(struct net *net, int group);
673 void (*unbind)(struct net *net, int group);
674 void (*release)(struct sock *sock, unsigned long *groups);
677 sock->state = SS_UNCONNECTED;
679 if (sock->type != SOCK_RAW && sock->type != SOCK_DGRAM)
680 return -ESOCKTNOSUPPORT;
682 if (protocol < 0 || protocol >= MAX_LINKS)
683 return -EPROTONOSUPPORT;
684 protocol = array_index_nospec(protocol, MAX_LINKS);
686 netlink_lock_table();
687 #ifdef CONFIG_MODULES
688 if (!nl_table[protocol].registered) {
689 netlink_unlock_table();
690 request_module("net-pf-%d-proto-%d", PF_NETLINK, protocol);
691 netlink_lock_table();
694 if (nl_table[protocol].registered &&
695 try_module_get(nl_table[protocol].module))
696 module = nl_table[protocol].module;
698 err = -EPROTONOSUPPORT;
699 cb_mutex = nl_table[protocol].cb_mutex;
700 bind = nl_table[protocol].bind;
701 unbind = nl_table[protocol].unbind;
702 release = nl_table[protocol].release;
703 netlink_unlock_table();
708 err = __netlink_create(net, sock, cb_mutex, protocol, kern);
712 sock_prot_inuse_add(net, &netlink_proto, 1);
714 nlk = nlk_sk(sock->sk);
715 nlk->module = module;
716 nlk->netlink_bind = bind;
717 nlk->netlink_unbind = unbind;
718 nlk->netlink_release = release;
727 static void deferred_put_nlk_sk(struct rcu_head *head)
729 struct netlink_sock *nlk = container_of(head, struct netlink_sock, rcu);
730 struct sock *sk = &nlk->sk;
735 if (!refcount_dec_and_test(&sk->sk_refcnt))
738 if (nlk->cb_running && nlk->cb.done) {
739 INIT_WORK(&nlk->work, netlink_sock_destruct_work);
740 schedule_work(&nlk->work);
747 static int netlink_release(struct socket *sock)
749 struct sock *sk = sock->sk;
750 struct netlink_sock *nlk;
760 * OK. Socket is unlinked, any packets that arrive now
763 if (nlk->netlink_release)
764 nlk->netlink_release(sk, nlk->groups);
766 /* must not acquire netlink_table_lock in any way again before unbind
767 * and notifying genetlink is done as otherwise it might deadlock
769 if (nlk->netlink_unbind) {
772 for (i = 0; i < nlk->ngroups; i++)
773 if (test_bit(i, nlk->groups))
774 nlk->netlink_unbind(sock_net(sk), i + 1);
776 if (sk->sk_protocol == NETLINK_GENERIC &&
777 atomic_dec_return(&genl_sk_destructing_cnt) == 0)
778 wake_up(&genl_sk_destructing_waitq);
781 wake_up_interruptible_all(&nlk->wait);
783 skb_queue_purge(&sk->sk_write_queue);
785 if (nlk->portid && nlk->bound) {
786 struct netlink_notify n = {
788 .protocol = sk->sk_protocol,
789 .portid = nlk->portid,
791 blocking_notifier_call_chain(&netlink_chain,
792 NETLINK_URELEASE, &n);
795 module_put(nlk->module);
797 if (netlink_is_kernel(sk)) {
798 netlink_table_grab();
799 BUG_ON(nl_table[sk->sk_protocol].registered == 0);
800 if (--nl_table[sk->sk_protocol].registered == 0) {
801 struct listeners *old;
803 old = nl_deref_protected(nl_table[sk->sk_protocol].listeners);
804 RCU_INIT_POINTER(nl_table[sk->sk_protocol].listeners, NULL);
806 nl_table[sk->sk_protocol].module = NULL;
807 nl_table[sk->sk_protocol].bind = NULL;
808 nl_table[sk->sk_protocol].unbind = NULL;
809 nl_table[sk->sk_protocol].flags = 0;
810 nl_table[sk->sk_protocol].registered = 0;
812 netlink_table_ungrab();
815 sock_prot_inuse_add(sock_net(sk), &netlink_proto, -1);
817 /* Because struct net might disappear soon, do not keep a pointer. */
818 if (!sk->sk_net_refcnt && sock_net(sk) != &init_net) {
819 __netns_tracker_free(sock_net(sk), &sk->ns_tracker, false);
820 /* Because of deferred_put_nlk_sk and use of work queue,
821 * it is possible netns will be freed before this socket.
823 sock_net_set(sk, &init_net);
824 __netns_tracker_alloc(&init_net, &sk->ns_tracker,
827 call_rcu(&nlk->rcu, deferred_put_nlk_sk);
831 static int netlink_autobind(struct socket *sock)
833 struct sock *sk = sock->sk;
834 struct net *net = sock_net(sk);
835 struct netlink_table *table = &nl_table[sk->sk_protocol];
836 s32 portid = task_tgid_vnr(current);
844 ok = !__netlink_lookup(table, portid, net);
847 /* Bind collision, search negative portid values. */
849 /* rover will be in range [S32_MIN, -4097] */
850 rover = S32_MIN + get_random_u32_below(-4096 - S32_MIN);
851 else if (rover >= -4096)
857 err = netlink_insert(sk, portid);
858 if (err == -EADDRINUSE)
861 /* If 2 threads race to autobind, that is fine. */
869 * __netlink_ns_capable - General netlink message capability test
870 * @nsp: NETLINK_CB of the socket buffer holding a netlink command from userspace.
871 * @user_ns: The user namespace of the capability to use
872 * @cap: The capability to use
874 * Test to see if the opener of the socket we received the message
875 * from had when the netlink socket was created and the sender of the
876 * message has the capability @cap in the user namespace @user_ns.
878 bool __netlink_ns_capable(const struct netlink_skb_parms *nsp,
879 struct user_namespace *user_ns, int cap)
881 return ((nsp->flags & NETLINK_SKB_DST) ||
882 file_ns_capable(nsp->sk->sk_socket->file, user_ns, cap)) &&
883 ns_capable(user_ns, cap);
885 EXPORT_SYMBOL(__netlink_ns_capable);
888 * netlink_ns_capable - General netlink message capability test
889 * @skb: socket buffer holding a netlink command from userspace
890 * @user_ns: The user namespace of the capability to use
891 * @cap: The capability to use
893 * Test to see if the opener of the socket we received the message
894 * from had when the netlink socket was created and the sender of the
895 * message has the capability @cap in the user namespace @user_ns.
897 bool netlink_ns_capable(const struct sk_buff *skb,
898 struct user_namespace *user_ns, int cap)
900 return __netlink_ns_capable(&NETLINK_CB(skb), user_ns, cap);
902 EXPORT_SYMBOL(netlink_ns_capable);
905 * netlink_capable - Netlink global 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 in all user namespaces.
913 bool netlink_capable(const struct sk_buff *skb, int cap)
915 return netlink_ns_capable(skb, &init_user_ns, cap);
917 EXPORT_SYMBOL(netlink_capable);
920 * netlink_net_capable - Netlink network namespace message capability test
921 * @skb: socket buffer holding a netlink command from userspace
922 * @cap: The capability to use
924 * Test to see if the opener of the socket we received the message
925 * from had when the netlink socket was created and the sender of the
926 * message has the capability @cap over the network namespace of
927 * the socket we received the message from.
929 bool netlink_net_capable(const struct sk_buff *skb, int cap)
931 return netlink_ns_capable(skb, sock_net(skb->sk)->user_ns, cap);
933 EXPORT_SYMBOL(netlink_net_capable);
935 static inline int netlink_allowed(const struct socket *sock, unsigned int flag)
937 return (nl_table[sock->sk->sk_protocol].flags & flag) ||
938 ns_capable(sock_net(sock->sk)->user_ns, CAP_NET_ADMIN);
942 netlink_update_subscriptions(struct sock *sk, unsigned int subscriptions)
944 struct netlink_sock *nlk = nlk_sk(sk);
946 if (nlk->subscriptions && !subscriptions)
947 __sk_del_bind_node(sk);
948 else if (!nlk->subscriptions && subscriptions)
949 sk_add_bind_node(sk, &nl_table[sk->sk_protocol].mc_list);
950 nlk->subscriptions = subscriptions;
953 static int netlink_realloc_groups(struct sock *sk)
955 struct netlink_sock *nlk = nlk_sk(sk);
957 unsigned long *new_groups;
960 netlink_table_grab();
962 groups = nl_table[sk->sk_protocol].groups;
963 if (!nl_table[sk->sk_protocol].registered) {
968 if (nlk->ngroups >= groups)
971 new_groups = krealloc(nlk->groups, NLGRPSZ(groups), GFP_ATOMIC);
972 if (new_groups == NULL) {
976 memset((char *)new_groups + NLGRPSZ(nlk->ngroups), 0,
977 NLGRPSZ(groups) - NLGRPSZ(nlk->ngroups));
979 nlk->groups = new_groups;
980 nlk->ngroups = groups;
982 netlink_table_ungrab();
986 static void netlink_undo_bind(int group, long unsigned int groups,
989 struct netlink_sock *nlk = nlk_sk(sk);
992 if (!nlk->netlink_unbind)
995 for (undo = 0; undo < group; undo++)
996 if (test_bit(undo, &groups))
997 nlk->netlink_unbind(sock_net(sk), undo + 1);
1000 static int netlink_bind(struct socket *sock, struct sockaddr *addr,
1003 struct sock *sk = sock->sk;
1004 struct net *net = sock_net(sk);
1005 struct netlink_sock *nlk = nlk_sk(sk);
1006 struct sockaddr_nl *nladdr = (struct sockaddr_nl *)addr;
1008 unsigned long groups;
1011 if (addr_len < sizeof(struct sockaddr_nl))
1014 if (nladdr->nl_family != AF_NETLINK)
1016 groups = nladdr->nl_groups;
1018 /* Only superuser is allowed to listen multicasts */
1020 if (!netlink_allowed(sock, NL_CFG_F_NONROOT_RECV))
1022 err = netlink_realloc_groups(sk);
1027 if (nlk->ngroups < BITS_PER_LONG)
1028 groups &= (1UL << nlk->ngroups) - 1;
1030 /* Paired with WRITE_ONCE() in netlink_insert() */
1031 bound = READ_ONCE(nlk->bound);
1033 /* Ensure nlk->portid is up-to-date. */
1036 if (nladdr->nl_pid != nlk->portid)
1040 if (nlk->netlink_bind && groups) {
1043 /* nl_groups is a u32, so cap the maximum groups we can bind */
1044 for (group = 0; group < BITS_PER_TYPE(u32); group++) {
1045 if (!test_bit(group, &groups))
1047 err = nlk->netlink_bind(net, group + 1);
1050 netlink_undo_bind(group, groups, sk);
1055 /* No need for barriers here as we return to user-space without
1056 * using any of the bound attributes.
1058 netlink_lock_table();
1060 err = nladdr->nl_pid ?
1061 netlink_insert(sk, nladdr->nl_pid) :
1062 netlink_autobind(sock);
1064 netlink_undo_bind(BITS_PER_TYPE(u32), groups, sk);
1069 if (!groups && (nlk->groups == NULL || !(u32)nlk->groups[0]))
1071 netlink_unlock_table();
1073 netlink_table_grab();
1074 netlink_update_subscriptions(sk, nlk->subscriptions +
1076 hweight32(nlk->groups[0]));
1077 nlk->groups[0] = (nlk->groups[0] & ~0xffffffffUL) | groups;
1078 netlink_update_listeners(sk);
1079 netlink_table_ungrab();
1084 netlink_unlock_table();
1088 static int netlink_connect(struct socket *sock, struct sockaddr *addr,
1089 int alen, int flags)
1092 struct sock *sk = sock->sk;
1093 struct netlink_sock *nlk = nlk_sk(sk);
1094 struct sockaddr_nl *nladdr = (struct sockaddr_nl *)addr;
1096 if (alen < sizeof(addr->sa_family))
1099 if (addr->sa_family == AF_UNSPEC) {
1100 /* paired with READ_ONCE() in netlink_getsockbyportid() */
1101 WRITE_ONCE(sk->sk_state, NETLINK_UNCONNECTED);
1102 /* dst_portid and dst_group can be read locklessly */
1103 WRITE_ONCE(nlk->dst_portid, 0);
1104 WRITE_ONCE(nlk->dst_group, 0);
1107 if (addr->sa_family != AF_NETLINK)
1110 if (alen < sizeof(struct sockaddr_nl))
1113 if ((nladdr->nl_groups || nladdr->nl_pid) &&
1114 !netlink_allowed(sock, NL_CFG_F_NONROOT_SEND))
1117 /* No need for barriers here as we return to user-space without
1118 * using any of the bound attributes.
1119 * Paired with WRITE_ONCE() in netlink_insert().
1121 if (!READ_ONCE(nlk->bound))
1122 err = netlink_autobind(sock);
1125 /* paired with READ_ONCE() in netlink_getsockbyportid() */
1126 WRITE_ONCE(sk->sk_state, NETLINK_CONNECTED);
1127 /* dst_portid and dst_group can be read locklessly */
1128 WRITE_ONCE(nlk->dst_portid, nladdr->nl_pid);
1129 WRITE_ONCE(nlk->dst_group, ffs(nladdr->nl_groups));
1135 static int netlink_getname(struct socket *sock, struct sockaddr *addr,
1138 struct sock *sk = sock->sk;
1139 struct netlink_sock *nlk = nlk_sk(sk);
1140 DECLARE_SOCKADDR(struct sockaddr_nl *, nladdr, addr);
1142 nladdr->nl_family = AF_NETLINK;
1146 /* Paired with WRITE_ONCE() in netlink_connect() */
1147 nladdr->nl_pid = READ_ONCE(nlk->dst_portid);
1148 nladdr->nl_groups = netlink_group_mask(READ_ONCE(nlk->dst_group));
1150 /* Paired with WRITE_ONCE() in netlink_insert() */
1151 nladdr->nl_pid = READ_ONCE(nlk->portid);
1152 netlink_lock_table();
1153 nladdr->nl_groups = nlk->groups ? nlk->groups[0] : 0;
1154 netlink_unlock_table();
1156 return sizeof(*nladdr);
1159 static int netlink_ioctl(struct socket *sock, unsigned int cmd,
1162 /* try to hand this ioctl down to the NIC drivers.
1164 return -ENOIOCTLCMD;
1167 static struct sock *netlink_getsockbyportid(struct sock *ssk, u32 portid)
1170 struct netlink_sock *nlk;
1172 sock = netlink_lookup(sock_net(ssk), ssk->sk_protocol, portid);
1174 return ERR_PTR(-ECONNREFUSED);
1176 /* Don't bother queuing skb if kernel socket has no input function */
1178 /* dst_portid and sk_state can be changed in netlink_connect() */
1179 if (READ_ONCE(sock->sk_state) == NETLINK_CONNECTED &&
1180 READ_ONCE(nlk->dst_portid) != nlk_sk(ssk)->portid) {
1182 return ERR_PTR(-ECONNREFUSED);
1187 struct sock *netlink_getsockbyfilp(struct file *filp)
1189 struct inode *inode = file_inode(filp);
1192 if (!S_ISSOCK(inode->i_mode))
1193 return ERR_PTR(-ENOTSOCK);
1195 sock = SOCKET_I(inode)->sk;
1196 if (sock->sk_family != AF_NETLINK)
1197 return ERR_PTR(-EINVAL);
1203 struct sk_buff *netlink_alloc_large_skb(unsigned int size, int broadcast)
1205 size_t head_size = SKB_HEAD_ALIGN(size);
1206 struct sk_buff *skb;
1209 if (head_size <= PAGE_SIZE || broadcast)
1210 return alloc_skb(size, GFP_KERNEL);
1212 data = kvmalloc(head_size, GFP_KERNEL);
1216 skb = __build_skb(data, head_size);
1219 else if (is_vmalloc_addr(data))
1220 skb->destructor = netlink_skb_destructor;
1226 * Attach a skb to a netlink socket.
1227 * The caller must hold a reference to the destination socket. On error, the
1228 * reference is dropped. The skb is not send to the destination, just all
1229 * all error checks are performed and memory in the queue is reserved.
1231 * < 0: error. skb freed, reference to sock dropped.
1233 * 1: repeat lookup - reference dropped while waiting for socket memory.
1235 int netlink_attachskb(struct sock *sk, struct sk_buff *skb,
1236 long *timeo, struct sock *ssk)
1238 struct netlink_sock *nlk;
1242 if ((atomic_read(&sk->sk_rmem_alloc) > sk->sk_rcvbuf ||
1243 test_bit(NETLINK_S_CONGESTED, &nlk->state))) {
1244 DECLARE_WAITQUEUE(wait, current);
1246 if (!ssk || netlink_is_kernel(ssk))
1247 netlink_overrun(sk);
1253 __set_current_state(TASK_INTERRUPTIBLE);
1254 add_wait_queue(&nlk->wait, &wait);
1256 if ((atomic_read(&sk->sk_rmem_alloc) > sk->sk_rcvbuf ||
1257 test_bit(NETLINK_S_CONGESTED, &nlk->state)) &&
1258 !sock_flag(sk, SOCK_DEAD))
1259 *timeo = schedule_timeout(*timeo);
1261 __set_current_state(TASK_RUNNING);
1262 remove_wait_queue(&nlk->wait, &wait);
1265 if (signal_pending(current)) {
1267 return sock_intr_errno(*timeo);
1271 netlink_skb_set_owner_r(skb, sk);
1275 static int __netlink_sendskb(struct sock *sk, struct sk_buff *skb)
1279 netlink_deliver_tap(sock_net(sk), skb);
1281 skb_queue_tail(&sk->sk_receive_queue, skb);
1282 sk->sk_data_ready(sk);
1286 int netlink_sendskb(struct sock *sk, struct sk_buff *skb)
1288 int len = __netlink_sendskb(sk, skb);
1294 void netlink_detachskb(struct sock *sk, struct sk_buff *skb)
1300 static struct sk_buff *netlink_trim(struct sk_buff *skb, gfp_t allocation)
1304 WARN_ON(skb->sk != NULL);
1305 delta = skb->end - skb->tail;
1306 if (is_vmalloc_addr(skb->head) || delta * 2 < skb->truesize)
1309 if (skb_shared(skb)) {
1310 struct sk_buff *nskb = skb_clone(skb, allocation);
1317 pskb_expand_head(skb, 0, -delta,
1318 (allocation & ~__GFP_DIRECT_RECLAIM) |
1319 __GFP_NOWARN | __GFP_NORETRY);
1323 static int netlink_unicast_kernel(struct sock *sk, struct sk_buff *skb,
1327 struct netlink_sock *nlk = nlk_sk(sk);
1329 ret = -ECONNREFUSED;
1330 if (nlk->netlink_rcv != NULL) {
1332 netlink_skb_set_owner_r(skb, sk);
1333 NETLINK_CB(skb).sk = ssk;
1334 netlink_deliver_tap_kernel(sk, ssk, skb);
1335 nlk->netlink_rcv(skb);
1344 int netlink_unicast(struct sock *ssk, struct sk_buff *skb,
1345 u32 portid, int nonblock)
1351 skb = netlink_trim(skb, gfp_any());
1353 timeo = sock_sndtimeo(ssk, nonblock);
1355 sk = netlink_getsockbyportid(ssk, portid);
1360 if (netlink_is_kernel(sk))
1361 return netlink_unicast_kernel(sk, skb, ssk);
1363 if (sk_filter(sk, skb)) {
1370 err = netlink_attachskb(sk, skb, &timeo, ssk);
1376 return netlink_sendskb(sk, skb);
1378 EXPORT_SYMBOL(netlink_unicast);
1380 int netlink_has_listeners(struct sock *sk, unsigned int group)
1383 struct listeners *listeners;
1385 BUG_ON(!netlink_is_kernel(sk));
1388 listeners = rcu_dereference(nl_table[sk->sk_protocol].listeners);
1390 if (listeners && group - 1 < nl_table[sk->sk_protocol].groups)
1391 res = test_bit(group - 1, listeners->masks);
1397 EXPORT_SYMBOL_GPL(netlink_has_listeners);
1399 bool netlink_strict_get_check(struct sk_buff *skb)
1401 return nlk_test_bit(STRICT_CHK, NETLINK_CB(skb).sk);
1403 EXPORT_SYMBOL_GPL(netlink_strict_get_check);
1405 static int netlink_broadcast_deliver(struct sock *sk, struct sk_buff *skb)
1407 struct netlink_sock *nlk = nlk_sk(sk);
1409 if (atomic_read(&sk->sk_rmem_alloc) <= sk->sk_rcvbuf &&
1410 !test_bit(NETLINK_S_CONGESTED, &nlk->state)) {
1411 netlink_skb_set_owner_r(skb, sk);
1412 __netlink_sendskb(sk, skb);
1413 return atomic_read(&sk->sk_rmem_alloc) > (sk->sk_rcvbuf >> 1);
1418 struct netlink_broadcast_data {
1419 struct sock *exclude_sk;
1424 int delivery_failure;
1428 struct sk_buff *skb, *skb2;
1429 int (*tx_filter)(struct sock *dsk, struct sk_buff *skb, void *data);
1433 static void do_one_broadcast(struct sock *sk,
1434 struct netlink_broadcast_data *p)
1436 struct netlink_sock *nlk = nlk_sk(sk);
1439 if (p->exclude_sk == sk)
1442 if (nlk->portid == p->portid || p->group - 1 >= nlk->ngroups ||
1443 !test_bit(p->group - 1, nlk->groups))
1446 if (!net_eq(sock_net(sk), p->net)) {
1447 if (!nlk_test_bit(LISTEN_ALL_NSID, sk))
1450 if (!peernet_has_id(sock_net(sk), p->net))
1453 if (!file_ns_capable(sk->sk_socket->file, p->net->user_ns,
1459 netlink_overrun(sk);
1464 if (p->skb2 == NULL) {
1465 if (skb_shared(p->skb)) {
1466 p->skb2 = skb_clone(p->skb, p->allocation);
1468 p->skb2 = skb_get(p->skb);
1470 * skb ownership may have been set when
1471 * delivered to a previous socket.
1473 skb_orphan(p->skb2);
1476 if (p->skb2 == NULL) {
1477 netlink_overrun(sk);
1478 /* Clone failed. Notify ALL listeners. */
1480 if (nlk_test_bit(BROADCAST_SEND_ERROR, sk))
1481 p->delivery_failure = 1;
1485 if (p->tx_filter && p->tx_filter(sk, p->skb2, p->tx_data)) {
1491 if (sk_filter(sk, p->skb2)) {
1496 NETLINK_CB(p->skb2).nsid = peernet2id(sock_net(sk), p->net);
1497 if (NETLINK_CB(p->skb2).nsid != NETNSA_NSID_NOT_ASSIGNED)
1498 NETLINK_CB(p->skb2).nsid_is_set = true;
1499 val = netlink_broadcast_deliver(sk, p->skb2);
1501 netlink_overrun(sk);
1502 if (nlk_test_bit(BROADCAST_SEND_ERROR, sk))
1503 p->delivery_failure = 1;
1505 p->congested |= val;
1513 int netlink_broadcast_filtered(struct sock *ssk, struct sk_buff *skb,
1515 u32 group, gfp_t allocation,
1516 netlink_filter_fn filter,
1519 struct net *net = sock_net(ssk);
1520 struct netlink_broadcast_data info;
1523 skb = netlink_trim(skb, allocation);
1525 info.exclude_sk = ssk;
1527 info.portid = portid;
1530 info.delivery_failure = 0;
1533 info.allocation = allocation;
1536 info.tx_filter = filter;
1537 info.tx_data = filter_data;
1539 /* While we sleep in clone, do not allow to change socket list */
1541 netlink_lock_table();
1543 sk_for_each_bound(sk, &nl_table[ssk->sk_protocol].mc_list)
1544 do_one_broadcast(sk, &info);
1548 netlink_unlock_table();
1550 if (info.delivery_failure) {
1551 kfree_skb(info.skb2);
1554 consume_skb(info.skb2);
1556 if (info.delivered) {
1557 if (info.congested && gfpflags_allow_blocking(allocation))
1563 EXPORT_SYMBOL(netlink_broadcast_filtered);
1565 int netlink_broadcast(struct sock *ssk, struct sk_buff *skb, u32 portid,
1566 u32 group, gfp_t allocation)
1568 return netlink_broadcast_filtered(ssk, skb, portid, group, allocation,
1571 EXPORT_SYMBOL(netlink_broadcast);
1573 struct netlink_set_err_data {
1574 struct sock *exclude_sk;
1580 static int do_one_set_err(struct sock *sk, struct netlink_set_err_data *p)
1582 struct netlink_sock *nlk = nlk_sk(sk);
1585 if (sk == p->exclude_sk)
1588 if (!net_eq(sock_net(sk), sock_net(p->exclude_sk)))
1591 if (nlk->portid == p->portid || p->group - 1 >= nlk->ngroups ||
1592 !test_bit(p->group - 1, nlk->groups))
1595 if (p->code == ENOBUFS && nlk_test_bit(RECV_NO_ENOBUFS, sk)) {
1600 WRITE_ONCE(sk->sk_err, p->code);
1601 sk_error_report(sk);
1607 * netlink_set_err - report error to broadcast listeners
1608 * @ssk: the kernel netlink socket, as returned by netlink_kernel_create()
1609 * @portid: the PORTID of a process that we want to skip (if any)
1610 * @group: the broadcast group that will notice the error
1611 * @code: error code, must be negative (as usual in kernelspace)
1613 * This function returns the number of broadcast listeners that have set the
1614 * NETLINK_NO_ENOBUFS socket option.
1616 int netlink_set_err(struct sock *ssk, u32 portid, u32 group, int code)
1618 struct netlink_set_err_data info;
1619 unsigned long flags;
1623 info.exclude_sk = ssk;
1624 info.portid = portid;
1626 /* sk->sk_err wants a positive error value */
1629 read_lock_irqsave(&nl_table_lock, flags);
1631 sk_for_each_bound(sk, &nl_table[ssk->sk_protocol].mc_list)
1632 ret += do_one_set_err(sk, &info);
1634 read_unlock_irqrestore(&nl_table_lock, flags);
1637 EXPORT_SYMBOL(netlink_set_err);
1639 /* must be called with netlink table grabbed */
1640 static void netlink_update_socket_mc(struct netlink_sock *nlk,
1644 int old, new = !!is_new, subscriptions;
1646 old = test_bit(group - 1, nlk->groups);
1647 subscriptions = nlk->subscriptions - old + new;
1648 __assign_bit(group - 1, nlk->groups, new);
1649 netlink_update_subscriptions(&nlk->sk, subscriptions);
1650 netlink_update_listeners(&nlk->sk);
1653 static int netlink_setsockopt(struct socket *sock, int level, int optname,
1654 sockptr_t optval, unsigned int optlen)
1656 struct sock *sk = sock->sk;
1657 struct netlink_sock *nlk = nlk_sk(sk);
1658 unsigned int val = 0;
1661 if (level != SOL_NETLINK)
1662 return -ENOPROTOOPT;
1664 if (optlen >= sizeof(int) &&
1665 copy_from_sockptr(&val, optval, sizeof(val)))
1669 case NETLINK_PKTINFO:
1670 nr = NETLINK_F_RECV_PKTINFO;
1672 case NETLINK_ADD_MEMBERSHIP:
1673 case NETLINK_DROP_MEMBERSHIP: {
1676 if (!netlink_allowed(sock, NL_CFG_F_NONROOT_RECV))
1678 err = netlink_realloc_groups(sk);
1681 if (!val || val - 1 >= nlk->ngroups)
1683 if (optname == NETLINK_ADD_MEMBERSHIP && nlk->netlink_bind) {
1684 err = nlk->netlink_bind(sock_net(sk), val);
1688 netlink_table_grab();
1689 netlink_update_socket_mc(nlk, val,
1690 optname == NETLINK_ADD_MEMBERSHIP);
1691 netlink_table_ungrab();
1692 if (optname == NETLINK_DROP_MEMBERSHIP && nlk->netlink_unbind)
1693 nlk->netlink_unbind(sock_net(sk), val);
1697 case NETLINK_BROADCAST_ERROR:
1698 nr = NETLINK_F_BROADCAST_SEND_ERROR;
1700 case NETLINK_NO_ENOBUFS:
1701 assign_bit(NETLINK_F_RECV_NO_ENOBUFS, &nlk->flags, val);
1703 clear_bit(NETLINK_S_CONGESTED, &nlk->state);
1704 wake_up_interruptible(&nlk->wait);
1707 case NETLINK_LISTEN_ALL_NSID:
1708 if (!ns_capable(sock_net(sk)->user_ns, CAP_NET_BROADCAST))
1710 nr = NETLINK_F_LISTEN_ALL_NSID;
1712 case NETLINK_CAP_ACK:
1713 nr = NETLINK_F_CAP_ACK;
1715 case NETLINK_EXT_ACK:
1716 nr = NETLINK_F_EXT_ACK;
1718 case NETLINK_GET_STRICT_CHK:
1719 nr = NETLINK_F_STRICT_CHK;
1722 return -ENOPROTOOPT;
1725 assign_bit(nr, &nlk->flags, val);
1729 static int netlink_getsockopt(struct socket *sock, int level, int optname,
1730 char __user *optval, int __user *optlen)
1732 struct sock *sk = sock->sk;
1733 struct netlink_sock *nlk = nlk_sk(sk);
1737 if (level != SOL_NETLINK)
1738 return -ENOPROTOOPT;
1740 if (get_user(len, optlen))
1746 case NETLINK_PKTINFO:
1747 flag = NETLINK_F_RECV_PKTINFO;
1749 case NETLINK_BROADCAST_ERROR:
1750 flag = NETLINK_F_BROADCAST_SEND_ERROR;
1752 case NETLINK_NO_ENOBUFS:
1753 flag = NETLINK_F_RECV_NO_ENOBUFS;
1755 case NETLINK_LIST_MEMBERSHIPS: {
1756 int pos, idx, shift, err = 0;
1758 netlink_lock_table();
1759 for (pos = 0; pos * 8 < nlk->ngroups; pos += sizeof(u32)) {
1760 if (len - pos < sizeof(u32))
1763 idx = pos / sizeof(unsigned long);
1764 shift = (pos % sizeof(unsigned long)) * 8;
1765 if (put_user((u32)(nlk->groups[idx] >> shift),
1766 (u32 __user *)(optval + pos))) {
1771 if (put_user(ALIGN(BITS_TO_BYTES(nlk->ngroups), sizeof(u32)), optlen))
1773 netlink_unlock_table();
1776 case NETLINK_LISTEN_ALL_NSID:
1777 flag = NETLINK_F_LISTEN_ALL_NSID;
1779 case NETLINK_CAP_ACK:
1780 flag = NETLINK_F_CAP_ACK;
1782 case NETLINK_EXT_ACK:
1783 flag = NETLINK_F_EXT_ACK;
1785 case NETLINK_GET_STRICT_CHK:
1786 flag = NETLINK_F_STRICT_CHK;
1789 return -ENOPROTOOPT;
1792 if (len < sizeof(int))
1796 val = test_bit(flag, &nlk->flags);
1798 if (put_user(len, optlen) ||
1799 copy_to_user(optval, &val, len))
1805 static void netlink_cmsg_recv_pktinfo(struct msghdr *msg, struct sk_buff *skb)
1807 struct nl_pktinfo info;
1809 info.group = NETLINK_CB(skb).dst_group;
1810 put_cmsg(msg, SOL_NETLINK, NETLINK_PKTINFO, sizeof(info), &info);
1813 static void netlink_cmsg_listen_all_nsid(struct sock *sk, struct msghdr *msg,
1814 struct sk_buff *skb)
1816 if (!NETLINK_CB(skb).nsid_is_set)
1819 put_cmsg(msg, SOL_NETLINK, NETLINK_LISTEN_ALL_NSID, sizeof(int),
1820 &NETLINK_CB(skb).nsid);
1823 static int netlink_sendmsg(struct socket *sock, struct msghdr *msg, size_t len)
1825 struct sock *sk = sock->sk;
1826 struct netlink_sock *nlk = nlk_sk(sk);
1827 DECLARE_SOCKADDR(struct sockaddr_nl *, addr, msg->msg_name);
1830 struct sk_buff *skb;
1832 struct scm_cookie scm;
1833 u32 netlink_skb_flags = 0;
1835 if (msg->msg_flags & MSG_OOB)
1839 pr_warn_once("Zero length message leads to an empty skb\n");
1843 err = scm_send(sock, msg, &scm, true);
1847 if (msg->msg_namelen) {
1849 if (msg->msg_namelen < sizeof(struct sockaddr_nl))
1851 if (addr->nl_family != AF_NETLINK)
1853 dst_portid = addr->nl_pid;
1854 dst_group = ffs(addr->nl_groups);
1856 if ((dst_group || dst_portid) &&
1857 !netlink_allowed(sock, NL_CFG_F_NONROOT_SEND))
1859 netlink_skb_flags |= NETLINK_SKB_DST;
1861 /* Paired with WRITE_ONCE() in netlink_connect() */
1862 dst_portid = READ_ONCE(nlk->dst_portid);
1863 dst_group = READ_ONCE(nlk->dst_group);
1866 /* Paired with WRITE_ONCE() in netlink_insert() */
1867 if (!READ_ONCE(nlk->bound)) {
1868 err = netlink_autobind(sock);
1872 /* Ensure nlk is hashed and visible. */
1877 if (len > sk->sk_sndbuf - 32)
1880 skb = netlink_alloc_large_skb(len, dst_group);
1884 NETLINK_CB(skb).portid = nlk->portid;
1885 NETLINK_CB(skb).dst_group = dst_group;
1886 NETLINK_CB(skb).creds = scm.creds;
1887 NETLINK_CB(skb).flags = netlink_skb_flags;
1890 if (memcpy_from_msg(skb_put(skb, len), msg, len)) {
1895 err = security_netlink_send(sk, skb);
1902 refcount_inc(&skb->users);
1903 netlink_broadcast(sk, skb, dst_portid, dst_group, GFP_KERNEL);
1905 err = netlink_unicast(sk, skb, dst_portid, msg->msg_flags & MSG_DONTWAIT);
1912 static int netlink_recvmsg(struct socket *sock, struct msghdr *msg, size_t len,
1915 struct scm_cookie scm;
1916 struct sock *sk = sock->sk;
1917 struct netlink_sock *nlk = nlk_sk(sk);
1918 size_t copied, max_recvmsg_len;
1919 struct sk_buff *skb, *data_skb;
1922 if (flags & MSG_OOB)
1927 skb = skb_recv_datagram(sk, flags, &err);
1933 #ifdef CONFIG_COMPAT_NETLINK_MESSAGES
1934 if (unlikely(skb_shinfo(skb)->frag_list)) {
1936 * If this skb has a frag_list, then here that means that we
1937 * will have to use the frag_list skb's data for compat tasks
1938 * and the regular skb's data for normal (non-compat) tasks.
1940 * If we need to send the compat skb, assign it to the
1941 * 'data_skb' variable so that it will be used below for data
1942 * copying. We keep 'skb' for everything else, including
1943 * freeing both later.
1945 if (flags & MSG_CMSG_COMPAT)
1946 data_skb = skb_shinfo(skb)->frag_list;
1950 /* Record the max length of recvmsg() calls for future allocations */
1951 max_recvmsg_len = max(READ_ONCE(nlk->max_recvmsg_len), len);
1952 max_recvmsg_len = min_t(size_t, max_recvmsg_len,
1953 SKB_WITH_OVERHEAD(32768));
1954 WRITE_ONCE(nlk->max_recvmsg_len, max_recvmsg_len);
1956 copied = data_skb->len;
1958 msg->msg_flags |= MSG_TRUNC;
1962 err = skb_copy_datagram_msg(data_skb, 0, msg, copied);
1964 if (msg->msg_name) {
1965 DECLARE_SOCKADDR(struct sockaddr_nl *, addr, msg->msg_name);
1966 addr->nl_family = AF_NETLINK;
1968 addr->nl_pid = NETLINK_CB(skb).portid;
1969 addr->nl_groups = netlink_group_mask(NETLINK_CB(skb).dst_group);
1970 msg->msg_namelen = sizeof(*addr);
1973 if (nlk_test_bit(RECV_PKTINFO, sk))
1974 netlink_cmsg_recv_pktinfo(msg, skb);
1975 if (nlk_test_bit(LISTEN_ALL_NSID, sk))
1976 netlink_cmsg_listen_all_nsid(sk, msg, skb);
1978 memset(&scm, 0, sizeof(scm));
1979 scm.creds = *NETLINK_CREDS(skb);
1980 if (flags & MSG_TRUNC)
1981 copied = data_skb->len;
1983 skb_free_datagram(sk, skb);
1985 if (READ_ONCE(nlk->cb_running) &&
1986 atomic_read(&sk->sk_rmem_alloc) <= sk->sk_rcvbuf / 2) {
1987 ret = netlink_dump(sk, false);
1989 WRITE_ONCE(sk->sk_err, -ret);
1990 sk_error_report(sk);
1994 scm_recv(sock, msg, &scm, flags);
1996 netlink_rcv_wake(sk);
1997 return err ? : copied;
2000 static void netlink_data_ready(struct sock *sk)
2006 * We export these functions to other modules. They provide a
2007 * complete set of kernel non-blocking support for message
2012 __netlink_kernel_create(struct net *net, int unit, struct module *module,
2013 struct netlink_kernel_cfg *cfg)
2015 struct socket *sock;
2017 struct netlink_sock *nlk;
2018 struct listeners *listeners = NULL;
2019 struct mutex *cb_mutex = cfg ? cfg->cb_mutex : NULL;
2020 unsigned int groups;
2024 if (unit < 0 || unit >= MAX_LINKS)
2027 if (sock_create_lite(PF_NETLINK, SOCK_DGRAM, unit, &sock))
2030 if (__netlink_create(net, sock, cb_mutex, unit, 1) < 0)
2031 goto out_sock_release_nosk;
2035 if (!cfg || cfg->groups < 32)
2038 groups = cfg->groups;
2040 listeners = kzalloc(sizeof(*listeners) + NLGRPSZ(groups), GFP_KERNEL);
2042 goto out_sock_release;
2044 sk->sk_data_ready = netlink_data_ready;
2045 if (cfg && cfg->input)
2046 nlk_sk(sk)->netlink_rcv = cfg->input;
2048 if (netlink_insert(sk, 0))
2049 goto out_sock_release;
2052 set_bit(NETLINK_F_KERNEL_SOCKET, &nlk->flags);
2054 netlink_table_grab();
2055 if (!nl_table[unit].registered) {
2056 nl_table[unit].groups = groups;
2057 rcu_assign_pointer(nl_table[unit].listeners, listeners);
2058 nl_table[unit].cb_mutex = cb_mutex;
2059 nl_table[unit].module = module;
2061 nl_table[unit].bind = cfg->bind;
2062 nl_table[unit].unbind = cfg->unbind;
2063 nl_table[unit].release = cfg->release;
2064 nl_table[unit].flags = cfg->flags;
2066 nl_table[unit].registered = 1;
2069 nl_table[unit].registered++;
2071 netlink_table_ungrab();
2076 netlink_kernel_release(sk);
2079 out_sock_release_nosk:
2083 EXPORT_SYMBOL(__netlink_kernel_create);
2086 netlink_kernel_release(struct sock *sk)
2088 if (sk == NULL || sk->sk_socket == NULL)
2091 sock_release(sk->sk_socket);
2093 EXPORT_SYMBOL(netlink_kernel_release);
2095 int __netlink_change_ngroups(struct sock *sk, unsigned int groups)
2097 struct listeners *new, *old;
2098 struct netlink_table *tbl = &nl_table[sk->sk_protocol];
2103 if (NLGRPSZ(tbl->groups) < NLGRPSZ(groups)) {
2104 new = kzalloc(sizeof(*new) + NLGRPSZ(groups), GFP_ATOMIC);
2107 old = nl_deref_protected(tbl->listeners);
2108 memcpy(new->masks, old->masks, NLGRPSZ(tbl->groups));
2109 rcu_assign_pointer(tbl->listeners, new);
2111 kfree_rcu(old, rcu);
2113 tbl->groups = groups;
2119 * netlink_change_ngroups - change number of multicast groups
2121 * This changes the number of multicast groups that are available
2122 * on a certain netlink family. Note that it is not possible to
2123 * change the number of groups to below 32. Also note that it does
2124 * not implicitly call netlink_clear_multicast_users() when the
2125 * number of groups is reduced.
2127 * @sk: The kernel netlink socket, as returned by netlink_kernel_create().
2128 * @groups: The new number of groups.
2130 int netlink_change_ngroups(struct sock *sk, unsigned int groups)
2134 netlink_table_grab();
2135 err = __netlink_change_ngroups(sk, groups);
2136 netlink_table_ungrab();
2141 void __netlink_clear_multicast_users(struct sock *ksk, unsigned int group)
2144 struct netlink_table *tbl = &nl_table[ksk->sk_protocol];
2146 sk_for_each_bound(sk, &tbl->mc_list)
2147 netlink_update_socket_mc(nlk_sk(sk), group, 0);
2151 __nlmsg_put(struct sk_buff *skb, u32 portid, u32 seq, int type, int len, int flags)
2153 struct nlmsghdr *nlh;
2154 int size = nlmsg_msg_size(len);
2156 nlh = skb_put(skb, NLMSG_ALIGN(size));
2157 nlh->nlmsg_type = type;
2158 nlh->nlmsg_len = size;
2159 nlh->nlmsg_flags = flags;
2160 nlh->nlmsg_pid = portid;
2161 nlh->nlmsg_seq = seq;
2162 if (!__builtin_constant_p(size) || NLMSG_ALIGN(size) - size != 0)
2163 memset(nlmsg_data(nlh) + len, 0, NLMSG_ALIGN(size) - size);
2166 EXPORT_SYMBOL(__nlmsg_put);
2169 * It looks a bit ugly.
2170 * It would be better to create kernel thread.
2173 static int netlink_dump_done(struct netlink_sock *nlk, struct sk_buff *skb,
2174 struct netlink_callback *cb,
2175 struct netlink_ext_ack *extack)
2177 struct nlmsghdr *nlh;
2179 nlh = nlmsg_put_answer(skb, cb, NLMSG_DONE, sizeof(nlk->dump_done_errno),
2180 NLM_F_MULTI | cb->answer_flags);
2184 nl_dump_check_consistent(cb, nlh);
2185 memcpy(nlmsg_data(nlh), &nlk->dump_done_errno, sizeof(nlk->dump_done_errno));
2187 if (extack->_msg && test_bit(NETLINK_F_EXT_ACK, &nlk->flags)) {
2188 nlh->nlmsg_flags |= NLM_F_ACK_TLVS;
2189 if (!nla_put_string(skb, NLMSGERR_ATTR_MSG, extack->_msg))
2190 nlmsg_end(skb, nlh);
2196 static int netlink_dump(struct sock *sk, bool lock_taken)
2198 struct netlink_sock *nlk = nlk_sk(sk);
2199 struct netlink_ext_ack extack = {};
2200 struct netlink_callback *cb;
2201 struct sk_buff *skb = NULL;
2202 size_t max_recvmsg_len;
2203 struct module *module;
2209 mutex_lock(&nlk->nl_cb_mutex);
2210 if (!nlk->cb_running) {
2215 if (atomic_read(&sk->sk_rmem_alloc) >= sk->sk_rcvbuf)
2218 /* NLMSG_GOODSIZE is small to avoid high order allocations being
2219 * required, but it makes sense to _attempt_ a 16K bytes allocation
2220 * to reduce number of system calls on dump operations, if user
2221 * ever provided a big enough buffer.
2224 alloc_min_size = max_t(int, cb->min_dump_alloc, NLMSG_GOODSIZE);
2226 max_recvmsg_len = READ_ONCE(nlk->max_recvmsg_len);
2227 if (alloc_min_size < max_recvmsg_len) {
2228 alloc_size = max_recvmsg_len;
2229 skb = alloc_skb(alloc_size,
2230 (GFP_KERNEL & ~__GFP_DIRECT_RECLAIM) |
2231 __GFP_NOWARN | __GFP_NORETRY);
2234 alloc_size = alloc_min_size;
2235 skb = alloc_skb(alloc_size, GFP_KERNEL);
2240 /* Trim skb to allocated size. User is expected to provide buffer as
2241 * large as max(min_dump_alloc, 16KiB (mac_recvmsg_len capped at
2242 * netlink_recvmsg())). dump will pack as many smaller messages as
2243 * could fit within the allocated skb. skb is typically allocated
2244 * with larger space than required (could be as much as near 2x the
2245 * requested size with align to next power of 2 approach). Allowing
2246 * dump to use the excess space makes it difficult for a user to have a
2247 * reasonable static buffer based on the expected largest dump of a
2248 * single netdev. The outcome is MSG_TRUNC error.
2250 skb_reserve(skb, skb_tailroom(skb) - alloc_size);
2252 /* Make sure malicious BPF programs can not read unitialized memory
2253 * from skb->head -> skb->data
2255 skb_reset_network_header(skb);
2256 skb_reset_mac_header(skb);
2258 netlink_skb_set_owner_r(skb, sk);
2260 if (nlk->dump_done_errno > 0) {
2261 struct mutex *extra_mutex = nlk->dump_cb_mutex;
2263 cb->extack = &extack;
2265 if (cb->flags & RTNL_FLAG_DUMP_UNLOCKED)
2268 mutex_lock(extra_mutex);
2269 nlk->dump_done_errno = cb->dump(skb, cb);
2271 mutex_unlock(extra_mutex);
2273 /* EMSGSIZE plus something already in the skb means
2274 * that there's more to dump but current skb has filled up.
2275 * If the callback really wants to return EMSGSIZE to user space
2276 * it needs to do so again, on the next cb->dump() call,
2277 * without putting data in the skb.
2279 if (nlk->dump_done_errno == -EMSGSIZE && skb->len)
2280 nlk->dump_done_errno = skb->len;
2285 if (nlk->dump_done_errno > 0 ||
2286 skb_tailroom(skb) < nlmsg_total_size(sizeof(nlk->dump_done_errno))) {
2287 mutex_unlock(&nlk->nl_cb_mutex);
2289 if (sk_filter(sk, skb))
2292 __netlink_sendskb(sk, skb);
2296 if (netlink_dump_done(nlk, skb, cb, &extack))
2299 #ifdef CONFIG_COMPAT_NETLINK_MESSAGES
2300 /* frag_list skb's data is used for compat tasks
2301 * and the regular skb's data for normal (non-compat) tasks.
2302 * See netlink_recvmsg().
2304 if (unlikely(skb_shinfo(skb)->frag_list)) {
2305 if (netlink_dump_done(nlk, skb_shinfo(skb)->frag_list, cb, &extack))
2310 if (sk_filter(sk, skb))
2313 __netlink_sendskb(sk, skb);
2318 WRITE_ONCE(nlk->cb_running, false);
2319 module = cb->module;
2321 mutex_unlock(&nlk->nl_cb_mutex);
2327 mutex_unlock(&nlk->nl_cb_mutex);
2332 int __netlink_dump_start(struct sock *ssk, struct sk_buff *skb,
2333 const struct nlmsghdr *nlh,
2334 struct netlink_dump_control *control)
2336 struct netlink_callback *cb;
2337 struct netlink_sock *nlk;
2341 refcount_inc(&skb->users);
2343 sk = netlink_lookup(sock_net(ssk), ssk->sk_protocol, NETLINK_CB(skb).portid);
2345 ret = -ECONNREFUSED;
2350 mutex_lock(&nlk->nl_cb_mutex);
2351 /* A dump is in progress... */
2352 if (nlk->cb_running) {
2356 /* add reference of module which cb->dump belongs to */
2357 if (!try_module_get(control->module)) {
2358 ret = -EPROTONOSUPPORT;
2363 memset(cb, 0, sizeof(*cb));
2364 cb->dump = control->dump;
2365 cb->done = control->done;
2367 cb->data = control->data;
2368 cb->module = control->module;
2369 cb->min_dump_alloc = control->min_dump_alloc;
2370 cb->flags = control->flags;
2373 cb->strict_check = nlk_test_bit(STRICT_CHK, NETLINK_CB(skb).sk);
2375 if (control->start) {
2376 cb->extack = control->extack;
2377 ret = control->start(cb);
2383 WRITE_ONCE(nlk->cb_running, true);
2384 nlk->dump_done_errno = INT_MAX;
2386 ret = netlink_dump(sk, true);
2393 /* We successfully started a dump, by returning -EINTR we
2394 * signal not to send ACK even if it was requested.
2399 module_put(control->module);
2402 mutex_unlock(&nlk->nl_cb_mutex);
2407 EXPORT_SYMBOL(__netlink_dump_start);
2410 netlink_ack_tlv_len(struct netlink_sock *nlk, int err,
2411 const struct netlink_ext_ack *extack)
2415 if (!extack || !test_bit(NETLINK_F_EXT_ACK, &nlk->flags))
2420 tlvlen += nla_total_size(strlen(extack->_msg) + 1);
2421 if (extack->cookie_len)
2422 tlvlen += nla_total_size(extack->cookie_len);
2424 /* Following attributes are only reported as error (not warning) */
2428 if (extack->bad_attr)
2429 tlvlen += nla_total_size(sizeof(u32));
2431 tlvlen += netlink_policy_dump_attr_size_estimate(extack->policy);
2432 if (extack->miss_type)
2433 tlvlen += nla_total_size(sizeof(u32));
2434 if (extack->miss_nest)
2435 tlvlen += nla_total_size(sizeof(u32));
2441 netlink_ack_tlv_fill(struct sk_buff *in_skb, struct sk_buff *skb,
2442 struct nlmsghdr *nlh, int err,
2443 const struct netlink_ext_ack *extack)
2446 WARN_ON(nla_put_string(skb, NLMSGERR_ATTR_MSG, extack->_msg));
2447 if (extack->cookie_len)
2448 WARN_ON(nla_put(skb, NLMSGERR_ATTR_COOKIE,
2449 extack->cookie_len, extack->cookie));
2454 if (extack->bad_attr &&
2455 !WARN_ON((u8 *)extack->bad_attr < in_skb->data ||
2456 (u8 *)extack->bad_attr >= in_skb->data + in_skb->len))
2457 WARN_ON(nla_put_u32(skb, NLMSGERR_ATTR_OFFS,
2458 (u8 *)extack->bad_attr - (u8 *)nlh));
2460 netlink_policy_dump_write_attr(skb, extack->policy,
2461 NLMSGERR_ATTR_POLICY);
2462 if (extack->miss_type)
2463 WARN_ON(nla_put_u32(skb, NLMSGERR_ATTR_MISS_TYPE,
2464 extack->miss_type));
2465 if (extack->miss_nest &&
2466 !WARN_ON((u8 *)extack->miss_nest < in_skb->data ||
2467 (u8 *)extack->miss_nest > in_skb->data + in_skb->len))
2468 WARN_ON(nla_put_u32(skb, NLMSGERR_ATTR_MISS_NEST,
2469 (u8 *)extack->miss_nest - (u8 *)nlh));
2472 void netlink_ack(struct sk_buff *in_skb, struct nlmsghdr *nlh, int err,
2473 const struct netlink_ext_ack *extack)
2475 struct sk_buff *skb;
2476 struct nlmsghdr *rep;
2477 struct nlmsgerr *errmsg;
2478 size_t payload = sizeof(*errmsg);
2479 struct netlink_sock *nlk = nlk_sk(NETLINK_CB(in_skb).sk);
2480 unsigned int flags = 0;
2483 /* Error messages get the original request appened, unless the user
2484 * requests to cap the error message, and get extra error data if
2487 if (err && !test_bit(NETLINK_F_CAP_ACK, &nlk->flags))
2488 payload += nlmsg_len(nlh);
2490 flags |= NLM_F_CAPPED;
2492 tlvlen = netlink_ack_tlv_len(nlk, err, extack);
2494 flags |= NLM_F_ACK_TLVS;
2496 skb = nlmsg_new(payload + tlvlen, GFP_KERNEL);
2500 rep = nlmsg_put(skb, NETLINK_CB(in_skb).portid, nlh->nlmsg_seq,
2501 NLMSG_ERROR, sizeof(*errmsg), flags);
2504 errmsg = nlmsg_data(rep);
2505 errmsg->error = err;
2508 if (!(flags & NLM_F_CAPPED)) {
2509 if (!nlmsg_append(skb, nlmsg_len(nlh)))
2512 memcpy(nlmsg_data(&errmsg->msg), nlmsg_data(nlh),
2517 netlink_ack_tlv_fill(in_skb, skb, nlh, err, extack);
2519 nlmsg_end(skb, rep);
2521 nlmsg_unicast(in_skb->sk, skb, NETLINK_CB(in_skb).portid);
2528 WRITE_ONCE(NETLINK_CB(in_skb).sk->sk_err, ENOBUFS);
2529 sk_error_report(NETLINK_CB(in_skb).sk);
2531 EXPORT_SYMBOL(netlink_ack);
2533 int netlink_rcv_skb(struct sk_buff *skb, int (*cb)(struct sk_buff *,
2535 struct netlink_ext_ack *))
2537 struct netlink_ext_ack extack;
2538 struct nlmsghdr *nlh;
2541 while (skb->len >= nlmsg_total_size(0)) {
2544 memset(&extack, 0, sizeof(extack));
2545 nlh = nlmsg_hdr(skb);
2548 if (nlh->nlmsg_len < NLMSG_HDRLEN || skb->len < nlh->nlmsg_len)
2551 /* Only requests are handled by the kernel */
2552 if (!(nlh->nlmsg_flags & NLM_F_REQUEST))
2555 /* Skip control messages */
2556 if (nlh->nlmsg_type < NLMSG_MIN_TYPE)
2559 err = cb(skb, nlh, &extack);
2564 if (nlh->nlmsg_flags & NLM_F_ACK || err)
2565 netlink_ack(skb, nlh, err, &extack);
2568 msglen = NLMSG_ALIGN(nlh->nlmsg_len);
2569 if (msglen > skb->len)
2571 skb_pull(skb, msglen);
2576 EXPORT_SYMBOL(netlink_rcv_skb);
2579 * nlmsg_notify - send a notification netlink message
2580 * @sk: netlink socket to use
2581 * @skb: notification message
2582 * @portid: destination netlink portid for reports or 0
2583 * @group: destination multicast group or 0
2584 * @report: 1 to report back, 0 to disable
2585 * @flags: allocation flags
2587 int nlmsg_notify(struct sock *sk, struct sk_buff *skb, u32 portid,
2588 unsigned int group, int report, gfp_t flags)
2593 int exclude_portid = 0;
2596 refcount_inc(&skb->users);
2597 exclude_portid = portid;
2600 /* errors reported via destination sk->sk_err, but propagate
2601 * delivery errors if NETLINK_BROADCAST_ERROR flag is set */
2602 err = nlmsg_multicast(sk, skb, exclude_portid, group, flags);
2610 err2 = nlmsg_unicast(sk, skb, portid);
2617 EXPORT_SYMBOL(nlmsg_notify);
2619 #ifdef CONFIG_PROC_FS
2620 struct nl_seq_iter {
2621 struct seq_net_private p;
2622 struct rhashtable_iter hti;
2626 static void netlink_walk_start(struct nl_seq_iter *iter)
2628 rhashtable_walk_enter(&nl_table[iter->link].hash, &iter->hti);
2629 rhashtable_walk_start(&iter->hti);
2632 static void netlink_walk_stop(struct nl_seq_iter *iter)
2634 rhashtable_walk_stop(&iter->hti);
2635 rhashtable_walk_exit(&iter->hti);
2638 static void *__netlink_seq_next(struct seq_file *seq)
2640 struct nl_seq_iter *iter = seq->private;
2641 struct netlink_sock *nlk;
2645 nlk = rhashtable_walk_next(&iter->hti);
2648 if (PTR_ERR(nlk) == -EAGAIN)
2657 netlink_walk_stop(iter);
2658 if (++iter->link >= MAX_LINKS)
2661 netlink_walk_start(iter);
2663 } while (sock_net(&nlk->sk) != seq_file_net(seq));
2668 static void *netlink_seq_start(struct seq_file *seq, loff_t *posp)
2671 struct nl_seq_iter *iter = seq->private;
2672 void *obj = SEQ_START_TOKEN;
2677 netlink_walk_start(iter);
2679 for (pos = *posp; pos && obj && !IS_ERR(obj); pos--)
2680 obj = __netlink_seq_next(seq);
2685 static void *netlink_seq_next(struct seq_file *seq, void *v, loff_t *pos)
2688 return __netlink_seq_next(seq);
2691 static void netlink_native_seq_stop(struct seq_file *seq, void *v)
2693 struct nl_seq_iter *iter = seq->private;
2695 if (iter->link >= MAX_LINKS)
2698 netlink_walk_stop(iter);
2702 static int netlink_native_seq_show(struct seq_file *seq, void *v)
2704 if (v == SEQ_START_TOKEN) {
2706 "sk Eth Pid Groups "
2707 "Rmem Wmem Dump Locks Drops Inode\n");
2710 struct netlink_sock *nlk = nlk_sk(s);
2712 seq_printf(seq, "%pK %-3d %-10u %08x %-8d %-8d %-5d %-8d %-8u %-8lu\n",
2716 nlk->groups ? (u32)nlk->groups[0] : 0,
2717 sk_rmem_alloc_get(s),
2718 sk_wmem_alloc_get(s),
2719 READ_ONCE(nlk->cb_running),
2720 refcount_read(&s->sk_refcnt),
2721 atomic_read(&s->sk_drops),
2729 #ifdef CONFIG_BPF_SYSCALL
2730 struct bpf_iter__netlink {
2731 __bpf_md_ptr(struct bpf_iter_meta *, meta);
2732 __bpf_md_ptr(struct netlink_sock *, sk);
2735 DEFINE_BPF_ITER_FUNC(netlink, struct bpf_iter_meta *meta, struct netlink_sock *sk)
2737 static int netlink_prog_seq_show(struct bpf_prog *prog,
2738 struct bpf_iter_meta *meta,
2741 struct bpf_iter__netlink ctx;
2743 meta->seq_num--; /* skip SEQ_START_TOKEN */
2745 ctx.sk = nlk_sk((struct sock *)v);
2746 return bpf_iter_run_prog(prog, &ctx);
2749 static int netlink_seq_show(struct seq_file *seq, void *v)
2751 struct bpf_iter_meta meta;
2752 struct bpf_prog *prog;
2755 prog = bpf_iter_get_info(&meta, false);
2757 return netlink_native_seq_show(seq, v);
2759 if (v != SEQ_START_TOKEN)
2760 return netlink_prog_seq_show(prog, &meta, v);
2765 static void netlink_seq_stop(struct seq_file *seq, void *v)
2767 struct bpf_iter_meta meta;
2768 struct bpf_prog *prog;
2772 prog = bpf_iter_get_info(&meta, true);
2774 (void)netlink_prog_seq_show(prog, &meta, v);
2777 netlink_native_seq_stop(seq, v);
2780 static int netlink_seq_show(struct seq_file *seq, void *v)
2782 return netlink_native_seq_show(seq, v);
2785 static void netlink_seq_stop(struct seq_file *seq, void *v)
2787 netlink_native_seq_stop(seq, v);
2791 static const struct seq_operations netlink_seq_ops = {
2792 .start = netlink_seq_start,
2793 .next = netlink_seq_next,
2794 .stop = netlink_seq_stop,
2795 .show = netlink_seq_show,
2799 int netlink_register_notifier(struct notifier_block *nb)
2801 return blocking_notifier_chain_register(&netlink_chain, nb);
2803 EXPORT_SYMBOL(netlink_register_notifier);
2805 int netlink_unregister_notifier(struct notifier_block *nb)
2807 return blocking_notifier_chain_unregister(&netlink_chain, nb);
2809 EXPORT_SYMBOL(netlink_unregister_notifier);
2811 static const struct proto_ops netlink_ops = {
2812 .family = PF_NETLINK,
2813 .owner = THIS_MODULE,
2814 .release = netlink_release,
2815 .bind = netlink_bind,
2816 .connect = netlink_connect,
2817 .socketpair = sock_no_socketpair,
2818 .accept = sock_no_accept,
2819 .getname = netlink_getname,
2820 .poll = datagram_poll,
2821 .ioctl = netlink_ioctl,
2822 .listen = sock_no_listen,
2823 .shutdown = sock_no_shutdown,
2824 .setsockopt = netlink_setsockopt,
2825 .getsockopt = netlink_getsockopt,
2826 .sendmsg = netlink_sendmsg,
2827 .recvmsg = netlink_recvmsg,
2828 .mmap = sock_no_mmap,
2831 static const struct net_proto_family netlink_family_ops = {
2832 .family = PF_NETLINK,
2833 .create = netlink_create,
2834 .owner = THIS_MODULE, /* for consistency 8) */
2837 static int __net_init netlink_net_init(struct net *net)
2839 #ifdef CONFIG_PROC_FS
2840 if (!proc_create_net("netlink", 0, net->proc_net, &netlink_seq_ops,
2841 sizeof(struct nl_seq_iter)))
2847 static void __net_exit netlink_net_exit(struct net *net)
2849 #ifdef CONFIG_PROC_FS
2850 remove_proc_entry("netlink", net->proc_net);
2854 static void __init netlink_add_usersock_entry(void)
2856 struct listeners *listeners;
2859 listeners = kzalloc(sizeof(*listeners) + NLGRPSZ(groups), GFP_KERNEL);
2861 panic("netlink_add_usersock_entry: Cannot allocate listeners\n");
2863 netlink_table_grab();
2865 nl_table[NETLINK_USERSOCK].groups = groups;
2866 rcu_assign_pointer(nl_table[NETLINK_USERSOCK].listeners, listeners);
2867 nl_table[NETLINK_USERSOCK].module = THIS_MODULE;
2868 nl_table[NETLINK_USERSOCK].registered = 1;
2869 nl_table[NETLINK_USERSOCK].flags = NL_CFG_F_NONROOT_SEND;
2871 netlink_table_ungrab();
2874 static struct pernet_operations __net_initdata netlink_net_ops = {
2875 .init = netlink_net_init,
2876 .exit = netlink_net_exit,
2879 static inline u32 netlink_hash(const void *data, u32 len, u32 seed)
2881 const struct netlink_sock *nlk = data;
2882 struct netlink_compare_arg arg;
2884 netlink_compare_arg_init(&arg, sock_net(&nlk->sk), nlk->portid);
2885 return jhash2((u32 *)&arg, netlink_compare_arg_len / sizeof(u32), seed);
2888 static const struct rhashtable_params netlink_rhashtable_params = {
2889 .head_offset = offsetof(struct netlink_sock, node),
2890 .key_len = netlink_compare_arg_len,
2891 .obj_hashfn = netlink_hash,
2892 .obj_cmpfn = netlink_compare,
2893 .automatic_shrinking = true,
2896 #if defined(CONFIG_BPF_SYSCALL) && defined(CONFIG_PROC_FS)
2897 BTF_ID_LIST(btf_netlink_sock_id)
2898 BTF_ID(struct, netlink_sock)
2900 static const struct bpf_iter_seq_info netlink_seq_info = {
2901 .seq_ops = &netlink_seq_ops,
2902 .init_seq_private = bpf_iter_init_seq_net,
2903 .fini_seq_private = bpf_iter_fini_seq_net,
2904 .seq_priv_size = sizeof(struct nl_seq_iter),
2907 static struct bpf_iter_reg netlink_reg_info = {
2908 .target = "netlink",
2909 .ctx_arg_info_size = 1,
2911 { offsetof(struct bpf_iter__netlink, sk),
2912 PTR_TO_BTF_ID_OR_NULL },
2914 .seq_info = &netlink_seq_info,
2917 static int __init bpf_iter_register(void)
2919 netlink_reg_info.ctx_arg_info[0].btf_id = *btf_netlink_sock_id;
2920 return bpf_iter_reg_target(&netlink_reg_info);
2924 static int __init netlink_proto_init(void)
2927 int err = proto_register(&netlink_proto, 0);
2932 #if defined(CONFIG_BPF_SYSCALL) && defined(CONFIG_PROC_FS)
2933 err = bpf_iter_register();
2938 BUILD_BUG_ON(sizeof(struct netlink_skb_parms) > sizeof_field(struct sk_buff, cb));
2940 nl_table = kcalloc(MAX_LINKS, sizeof(*nl_table), GFP_KERNEL);
2944 for (i = 0; i < MAX_LINKS; i++) {
2945 if (rhashtable_init(&nl_table[i].hash,
2946 &netlink_rhashtable_params) < 0) {
2948 rhashtable_destroy(&nl_table[i].hash);
2954 netlink_add_usersock_entry();
2956 sock_register(&netlink_family_ops);
2957 register_pernet_subsys(&netlink_net_ops);
2958 register_pernet_subsys(&netlink_tap_net_ops);
2959 /* The netlink device handler may be needed early. */
2964 panic("netlink_init: Cannot allocate nl_table\n");
2967 core_initcall(netlink_proto_init);