2 * NETLINK Kernel-user communication protocol.
4 * Authors: Alan Cox <alan@lxorguk.ukuu.org.uk>
5 * Alexey Kuznetsov <kuznet@ms2.inr.ac.ru>
6 * Patrick McHardy <kaber@trash.net>
8 * This program is free software; you can redistribute it and/or
9 * modify it under the terms of the GNU General Public License
10 * as published by the Free Software Foundation; either version
11 * 2 of the License, or (at your option) any later version.
13 * Tue Jun 26 14:36:48 MEST 2001 Herbert "herp" Rosmanith
14 * added netlink_proto_exit
15 * Tue Jan 22 18:32:44 BRST 2002 Arnaldo C. de Melo <acme@conectiva.com.br>
16 * use nlk_sk, as sk->protinfo is on a diet 8)
17 * Fri Jul 22 19:51:12 MEST 2005 Harald Welte <laforge@gnumonks.org>
18 * - inc module use count of module that owns
19 * the kernel socket in case userspace opens
20 * socket of same protocol
21 * - remove all module support, since netlink is
22 * mandatory if CONFIG_NET=y these days
25 #include <linux/module.h>
27 #include <linux/capability.h>
28 #include <linux/kernel.h>
29 #include <linux/init.h>
30 #include <linux/signal.h>
31 #include <linux/sched.h>
32 #include <linux/errno.h>
33 #include <linux/string.h>
34 #include <linux/stat.h>
35 #include <linux/socket.h>
37 #include <linux/fcntl.h>
38 #include <linux/termios.h>
39 #include <linux/sockios.h>
40 #include <linux/net.h>
42 #include <linux/slab.h>
43 #include <asm/uaccess.h>
44 #include <linux/skbuff.h>
45 #include <linux/netdevice.h>
46 #include <linux/rtnetlink.h>
47 #include <linux/proc_fs.h>
48 #include <linux/seq_file.h>
49 #include <linux/notifier.h>
50 #include <linux/security.h>
51 #include <linux/jhash.h>
52 #include <linux/jiffies.h>
53 #include <linux/random.h>
54 #include <linux/bitops.h>
56 #include <linux/types.h>
57 #include <linux/audit.h>
58 #include <linux/mutex.h>
59 #include <linux/vmalloc.h>
60 #include <linux/if_arp.h>
61 #include <linux/rhashtable.h>
62 #include <asm/cacheflush.h>
63 #include <linux/hash.h>
64 #include <linux/genetlink.h>
65 #include <linux/nospec.h>
67 #include <net/net_namespace.h>
70 #include <net/netlink.h>
72 #include "af_netlink.h"
76 unsigned long masks[0];
80 #define NETLINK_S_CONGESTED 0x0
83 #define NETLINK_F_KERNEL_SOCKET 0x1
84 #define NETLINK_F_RECV_PKTINFO 0x2
85 #define NETLINK_F_BROADCAST_SEND_ERROR 0x4
86 #define NETLINK_F_RECV_NO_ENOBUFS 0x8
87 #define NETLINK_F_LISTEN_ALL_NSID 0x10
88 #define NETLINK_F_CAP_ACK 0x20
90 static inline int netlink_is_kernel(struct sock *sk)
92 return nlk_sk(sk)->flags & NETLINK_F_KERNEL_SOCKET;
95 struct netlink_table *nl_table __read_mostly;
96 EXPORT_SYMBOL_GPL(nl_table);
98 static DECLARE_WAIT_QUEUE_HEAD(nl_table_wait);
100 static struct lock_class_key nlk_cb_mutex_keys[MAX_LINKS];
102 static const char *const nlk_cb_mutex_key_strings[MAX_LINKS + 1] = {
103 "nlk_cb_mutex-ROUTE",
105 "nlk_cb_mutex-USERSOCK",
106 "nlk_cb_mutex-FIREWALL",
107 "nlk_cb_mutex-SOCK_DIAG",
108 "nlk_cb_mutex-NFLOG",
110 "nlk_cb_mutex-SELINUX",
111 "nlk_cb_mutex-ISCSI",
112 "nlk_cb_mutex-AUDIT",
113 "nlk_cb_mutex-FIB_LOOKUP",
114 "nlk_cb_mutex-CONNECTOR",
115 "nlk_cb_mutex-NETFILTER",
116 "nlk_cb_mutex-IP6_FW",
117 "nlk_cb_mutex-DNRTMSG",
118 "nlk_cb_mutex-KOBJECT_UEVENT",
119 "nlk_cb_mutex-GENERIC",
121 "nlk_cb_mutex-SCSITRANSPORT",
122 "nlk_cb_mutex-ECRYPTFS",
124 "nlk_cb_mutex-CRYPTO",
135 "nlk_cb_mutex-MAX_LINKS"
138 static int netlink_dump(struct sock *sk);
139 static void netlink_skb_destructor(struct sk_buff *skb);
141 /* nl_table locking explained:
142 * Lookup and traversal are protected with an RCU read-side lock. Insertion
143 * and removal are protected with per bucket lock while using RCU list
144 * modification primitives and may run in parallel to RCU protected lookups.
145 * Destruction of the Netlink socket may only occur *after* nl_table_lock has
146 * been acquired * either during or after the socket has been removed from
147 * the list and after an RCU grace period.
149 DEFINE_RWLOCK(nl_table_lock);
150 EXPORT_SYMBOL_GPL(nl_table_lock);
151 static atomic_t nl_table_users = ATOMIC_INIT(0);
153 #define nl_deref_protected(X) rcu_dereference_protected(X, lockdep_is_held(&nl_table_lock));
155 static ATOMIC_NOTIFIER_HEAD(netlink_chain);
157 static DEFINE_SPINLOCK(netlink_tap_lock);
158 static struct list_head netlink_tap_all __read_mostly;
160 static const struct rhashtable_params netlink_rhashtable_params;
162 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 memcpy(skb_put(new, len), skb->data, len);
185 int netlink_add_tap(struct netlink_tap *nt)
187 if (unlikely(nt->dev->type != ARPHRD_NETLINK))
190 spin_lock(&netlink_tap_lock);
191 list_add_rcu(&nt->list, &netlink_tap_all);
192 spin_unlock(&netlink_tap_lock);
194 __module_get(nt->module);
198 EXPORT_SYMBOL_GPL(netlink_add_tap);
200 static int __netlink_remove_tap(struct netlink_tap *nt)
203 struct netlink_tap *tmp;
205 spin_lock(&netlink_tap_lock);
207 list_for_each_entry(tmp, &netlink_tap_all, list) {
209 list_del_rcu(&nt->list);
215 pr_warn("__netlink_remove_tap: %p not found\n", nt);
217 spin_unlock(&netlink_tap_lock);
220 module_put(nt->module);
222 return found ? 0 : -ENODEV;
225 int netlink_remove_tap(struct netlink_tap *nt)
229 ret = __netlink_remove_tap(nt);
234 EXPORT_SYMBOL_GPL(netlink_remove_tap);
236 static bool netlink_filter_tap(const struct sk_buff *skb)
238 struct sock *sk = skb->sk;
240 /* We take the more conservative approach and
241 * whitelist socket protocols that may pass.
243 switch (sk->sk_protocol) {
245 case NETLINK_USERSOCK:
246 case NETLINK_SOCK_DIAG:
249 case NETLINK_FIB_LOOKUP:
250 case NETLINK_NETFILTER:
251 case NETLINK_GENERIC:
258 static int __netlink_deliver_tap_skb(struct sk_buff *skb,
259 struct net_device *dev)
261 struct sk_buff *nskb;
262 struct sock *sk = skb->sk;
265 if (!net_eq(dev_net(dev), sock_net(sk)))
270 if (is_vmalloc_addr(skb->head))
271 nskb = netlink_to_full_skb(skb, GFP_ATOMIC);
273 nskb = skb_clone(skb, GFP_ATOMIC);
276 nskb->protocol = htons((u16) sk->sk_protocol);
277 nskb->pkt_type = netlink_is_kernel(sk) ?
278 PACKET_KERNEL : PACKET_USER;
279 skb_reset_network_header(nskb);
280 ret = dev_queue_xmit(nskb);
281 if (unlikely(ret > 0))
282 ret = net_xmit_errno(ret);
289 static void __netlink_deliver_tap(struct sk_buff *skb)
292 struct netlink_tap *tmp;
294 if (!netlink_filter_tap(skb))
297 list_for_each_entry_rcu(tmp, &netlink_tap_all, list) {
298 ret = __netlink_deliver_tap_skb(skb, tmp->dev);
304 static void netlink_deliver_tap(struct sk_buff *skb)
308 if (unlikely(!list_empty(&netlink_tap_all)))
309 __netlink_deliver_tap(skb);
314 static void netlink_deliver_tap_kernel(struct sock *dst, struct sock *src,
317 if (!(netlink_is_kernel(dst) && netlink_is_kernel(src)))
318 netlink_deliver_tap(skb);
321 static void netlink_overrun(struct sock *sk)
323 struct netlink_sock *nlk = nlk_sk(sk);
325 if (!(nlk->flags & NETLINK_F_RECV_NO_ENOBUFS)) {
326 if (!test_and_set_bit(NETLINK_S_CONGESTED,
327 &nlk_sk(sk)->state)) {
328 sk->sk_err = ENOBUFS;
329 sk->sk_error_report(sk);
332 atomic_inc(&sk->sk_drops);
335 static void netlink_rcv_wake(struct sock *sk)
337 struct netlink_sock *nlk = nlk_sk(sk);
339 if (skb_queue_empty(&sk->sk_receive_queue))
340 clear_bit(NETLINK_S_CONGESTED, &nlk->state);
341 if (!test_bit(NETLINK_S_CONGESTED, &nlk->state))
342 wake_up_interruptible(&nlk->wait);
345 static void netlink_skb_destructor(struct sk_buff *skb)
347 if (is_vmalloc_addr(skb->head)) {
349 !atomic_dec_return(&(skb_shinfo(skb)->dataref)))
358 static void netlink_skb_set_owner_r(struct sk_buff *skb, struct sock *sk)
360 WARN_ON(skb->sk != NULL);
362 skb->destructor = netlink_skb_destructor;
363 atomic_add(skb->truesize, &sk->sk_rmem_alloc);
364 sk_mem_charge(sk, skb->truesize);
367 static void netlink_sock_destruct(struct sock *sk)
369 struct netlink_sock *nlk = nlk_sk(sk);
371 if (nlk->cb_running) {
373 nlk->cb.done(&nlk->cb);
374 module_put(nlk->cb.module);
375 kfree_skb(nlk->cb.skb);
378 skb_queue_purge(&sk->sk_receive_queue);
380 if (!sock_flag(sk, SOCK_DEAD)) {
381 printk(KERN_ERR "Freeing alive netlink socket %p\n", sk);
385 WARN_ON(atomic_read(&sk->sk_rmem_alloc));
386 WARN_ON(atomic_read(&sk->sk_wmem_alloc));
387 WARN_ON(nlk_sk(sk)->groups);
390 static void netlink_sock_destruct_work(struct work_struct *work)
392 struct netlink_sock *nlk = container_of(work, struct netlink_sock,
398 /* This lock without WQ_FLAG_EXCLUSIVE is good on UP and it is _very_ bad on
399 * SMP. Look, when several writers sleep and reader wakes them up, all but one
400 * immediately hit write lock and grab all the cpus. Exclusive sleep solves
401 * this, _but_ remember, it adds useless work on UP machines.
404 void netlink_table_grab(void)
405 __acquires(nl_table_lock)
409 write_lock_irq(&nl_table_lock);
411 if (atomic_read(&nl_table_users)) {
412 DECLARE_WAITQUEUE(wait, current);
414 add_wait_queue_exclusive(&nl_table_wait, &wait);
416 set_current_state(TASK_UNINTERRUPTIBLE);
417 if (atomic_read(&nl_table_users) == 0)
419 write_unlock_irq(&nl_table_lock);
421 write_lock_irq(&nl_table_lock);
424 __set_current_state(TASK_RUNNING);
425 remove_wait_queue(&nl_table_wait, &wait);
429 void netlink_table_ungrab(void)
430 __releases(nl_table_lock)
432 write_unlock_irq(&nl_table_lock);
433 wake_up(&nl_table_wait);
437 netlink_lock_table(void)
441 /* read_lock() synchronizes us to netlink_table_grab */
443 read_lock_irqsave(&nl_table_lock, flags);
444 atomic_inc(&nl_table_users);
445 read_unlock_irqrestore(&nl_table_lock, flags);
449 netlink_unlock_table(void)
451 if (atomic_dec_and_test(&nl_table_users))
452 wake_up(&nl_table_wait);
455 struct netlink_compare_arg
461 /* Doing sizeof directly may yield 4 extra bytes on 64-bit. */
462 #define netlink_compare_arg_len \
463 (offsetof(struct netlink_compare_arg, portid) + sizeof(u32))
465 static inline int netlink_compare(struct rhashtable_compare_arg *arg,
468 const struct netlink_compare_arg *x = arg->key;
469 const struct netlink_sock *nlk = ptr;
471 return nlk->portid != x->portid ||
472 !net_eq(sock_net(&nlk->sk), read_pnet(&x->pnet));
475 static void netlink_compare_arg_init(struct netlink_compare_arg *arg,
476 struct net *net, u32 portid)
478 memset(arg, 0, sizeof(*arg));
479 write_pnet(&arg->pnet, net);
480 arg->portid = portid;
483 static struct sock *__netlink_lookup(struct netlink_table *table, u32 portid,
486 struct netlink_compare_arg arg;
488 netlink_compare_arg_init(&arg, net, portid);
489 return rhashtable_lookup_fast(&table->hash, &arg,
490 netlink_rhashtable_params);
493 static int __netlink_insert(struct netlink_table *table, struct sock *sk)
495 struct netlink_compare_arg arg;
497 netlink_compare_arg_init(&arg, sock_net(sk), nlk_sk(sk)->portid);
498 return rhashtable_lookup_insert_key(&table->hash, &arg,
500 netlink_rhashtable_params);
503 static struct sock *netlink_lookup(struct net *net, int protocol, u32 portid)
505 struct netlink_table *table = &nl_table[protocol];
509 sk = __netlink_lookup(table, portid, net);
517 static const struct proto_ops netlink_ops;
520 netlink_update_listeners(struct sock *sk)
522 struct netlink_table *tbl = &nl_table[sk->sk_protocol];
525 struct listeners *listeners;
527 listeners = nl_deref_protected(tbl->listeners);
531 for (i = 0; i < NLGRPLONGS(tbl->groups); i++) {
533 sk_for_each_bound(sk, &tbl->mc_list) {
534 if (i < NLGRPLONGS(nlk_sk(sk)->ngroups))
535 mask |= nlk_sk(sk)->groups[i];
537 listeners->masks[i] = mask;
539 /* this function is only called with the netlink table "grabbed", which
540 * makes sure updates are visible before bind or setsockopt return. */
543 static int netlink_insert(struct sock *sk, u32 portid)
545 struct netlink_table *table = &nl_table[sk->sk_protocol];
550 err = nlk_sk(sk)->portid == portid ? 0 : -EBUSY;
551 if (nlk_sk(sk)->bound)
555 if (BITS_PER_LONG > 32 &&
556 unlikely(atomic_read(&table->hash.nelems) >= UINT_MAX))
559 nlk_sk(sk)->portid = portid;
562 err = __netlink_insert(table, sk);
564 /* In case the hashtable backend returns with -EBUSY
565 * from here, it must not escape to the caller.
567 if (unlikely(err == -EBUSY))
575 /* We need to ensure that the socket is hashed and visible. */
577 /* Paired with lockless reads from netlink_bind(),
578 * netlink_connect() and netlink_sendmsg().
580 WRITE_ONCE(nlk_sk(sk)->bound, portid);
587 static void netlink_remove(struct sock *sk)
589 struct netlink_table *table;
591 table = &nl_table[sk->sk_protocol];
592 if (!rhashtable_remove_fast(&table->hash, &nlk_sk(sk)->node,
593 netlink_rhashtable_params)) {
594 WARN_ON(atomic_read(&sk->sk_refcnt) == 1);
598 netlink_table_grab();
599 if (nlk_sk(sk)->subscriptions) {
600 __sk_del_bind_node(sk);
601 netlink_update_listeners(sk);
603 if (sk->sk_protocol == NETLINK_GENERIC)
604 atomic_inc(&genl_sk_destructing_cnt);
605 netlink_table_ungrab();
608 static struct proto netlink_proto = {
610 .owner = THIS_MODULE,
611 .obj_size = sizeof(struct netlink_sock),
614 static int __netlink_create(struct net *net, struct socket *sock,
615 struct mutex *cb_mutex, int protocol,
619 struct netlink_sock *nlk;
621 sock->ops = &netlink_ops;
623 sk = sk_alloc(net, PF_NETLINK, GFP_KERNEL, &netlink_proto, kern);
627 sock_init_data(sock, sk);
631 nlk->cb_mutex = cb_mutex;
633 nlk->cb_mutex = &nlk->cb_def_mutex;
634 mutex_init(nlk->cb_mutex);
635 lockdep_set_class_and_name(nlk->cb_mutex,
636 nlk_cb_mutex_keys + protocol,
637 nlk_cb_mutex_key_strings[protocol]);
639 init_waitqueue_head(&nlk->wait);
641 sk->sk_destruct = netlink_sock_destruct;
642 sk->sk_protocol = protocol;
646 static int netlink_create(struct net *net, struct socket *sock, int protocol,
649 struct module *module = NULL;
650 struct mutex *cb_mutex;
651 struct netlink_sock *nlk;
652 int (*bind)(struct net *net, int group);
653 void (*unbind)(struct net *net, int group);
656 sock->state = SS_UNCONNECTED;
658 if (sock->type != SOCK_RAW && sock->type != SOCK_DGRAM)
659 return -ESOCKTNOSUPPORT;
661 if (protocol < 0 || protocol >= MAX_LINKS)
662 return -EPROTONOSUPPORT;
663 protocol = array_index_nospec(protocol, MAX_LINKS);
665 netlink_lock_table();
666 #ifdef CONFIG_MODULES
667 if (!nl_table[protocol].registered) {
668 netlink_unlock_table();
669 request_module("net-pf-%d-proto-%d", PF_NETLINK, protocol);
670 netlink_lock_table();
673 if (nl_table[protocol].registered &&
674 try_module_get(nl_table[protocol].module))
675 module = nl_table[protocol].module;
677 err = -EPROTONOSUPPORT;
678 cb_mutex = nl_table[protocol].cb_mutex;
679 bind = nl_table[protocol].bind;
680 unbind = nl_table[protocol].unbind;
681 netlink_unlock_table();
686 err = __netlink_create(net, sock, cb_mutex, protocol, kern);
691 sock_prot_inuse_add(net, &netlink_proto, 1);
694 nlk = nlk_sk(sock->sk);
695 nlk->module = module;
696 nlk->netlink_bind = bind;
697 nlk->netlink_unbind = unbind;
706 static void deferred_put_nlk_sk(struct rcu_head *head)
708 struct netlink_sock *nlk = container_of(head, struct netlink_sock, rcu);
709 struct sock *sk = &nlk->sk;
711 if (!atomic_dec_and_test(&sk->sk_refcnt))
714 if (nlk->cb_running && nlk->cb.done) {
715 INIT_WORK(&nlk->work, netlink_sock_destruct_work);
716 schedule_work(&nlk->work);
723 static int netlink_release(struct socket *sock)
725 struct sock *sk = sock->sk;
726 struct netlink_sock *nlk;
736 * OK. Socket is unlinked, any packets that arrive now
740 /* must not acquire netlink_table_lock in any way again before unbind
741 * and notifying genetlink is done as otherwise it might deadlock
743 if (nlk->netlink_unbind) {
746 for (i = 0; i < nlk->ngroups; i++)
747 if (test_bit(i, nlk->groups))
748 nlk->netlink_unbind(sock_net(sk), i + 1);
750 if (sk->sk_protocol == NETLINK_GENERIC &&
751 atomic_dec_return(&genl_sk_destructing_cnt) == 0)
752 wake_up(&genl_sk_destructing_waitq);
755 wake_up_interruptible_all(&nlk->wait);
757 skb_queue_purge(&sk->sk_write_queue);
759 if (nlk->portid && nlk->bound) {
760 struct netlink_notify n = {
762 .protocol = sk->sk_protocol,
763 .portid = nlk->portid,
765 atomic_notifier_call_chain(&netlink_chain,
766 NETLINK_URELEASE, &n);
769 module_put(nlk->module);
771 if (netlink_is_kernel(sk)) {
772 netlink_table_grab();
773 BUG_ON(nl_table[sk->sk_protocol].registered == 0);
774 if (--nl_table[sk->sk_protocol].registered == 0) {
775 struct listeners *old;
777 old = nl_deref_protected(nl_table[sk->sk_protocol].listeners);
778 RCU_INIT_POINTER(nl_table[sk->sk_protocol].listeners, NULL);
780 nl_table[sk->sk_protocol].module = NULL;
781 nl_table[sk->sk_protocol].bind = NULL;
782 nl_table[sk->sk_protocol].unbind = NULL;
783 nl_table[sk->sk_protocol].flags = 0;
784 nl_table[sk->sk_protocol].registered = 0;
786 netlink_table_ungrab();
793 sock_prot_inuse_add(sock_net(sk), &netlink_proto, -1);
795 call_rcu(&nlk->rcu, deferred_put_nlk_sk);
799 static int netlink_autobind(struct socket *sock)
801 struct sock *sk = sock->sk;
802 struct net *net = sock_net(sk);
803 struct netlink_table *table = &nl_table[sk->sk_protocol];
804 s32 portid = task_tgid_vnr(current);
812 ok = !__netlink_lookup(table, portid, net);
815 /* Bind collision, search negative portid values. */
817 /* rover will be in range [S32_MIN, -4097] */
818 rover = S32_MIN + prandom_u32_max(-4096 - S32_MIN);
819 else if (rover >= -4096)
825 err = netlink_insert(sk, portid);
826 if (err == -EADDRINUSE)
829 /* If 2 threads race to autobind, that is fine. */
837 * __netlink_ns_capable - General netlink message capability test
838 * @nsp: NETLINK_CB of the socket buffer holding a netlink command from userspace.
839 * @user_ns: The user namespace of the capability to use
840 * @cap: The capability to use
842 * Test to see if the opener of the socket we received the message
843 * from had when the netlink socket was created and the sender of the
844 * message has has the capability @cap in the user namespace @user_ns.
846 bool __netlink_ns_capable(const struct netlink_skb_parms *nsp,
847 struct user_namespace *user_ns, int cap)
849 return ((nsp->flags & NETLINK_SKB_DST) ||
850 file_ns_capable(nsp->sk->sk_socket->file, user_ns, cap)) &&
851 ns_capable(user_ns, cap);
853 EXPORT_SYMBOL(__netlink_ns_capable);
856 * netlink_ns_capable - General netlink message capability test
857 * @skb: socket buffer holding a netlink command from userspace
858 * @user_ns: The user namespace of the capability to use
859 * @cap: The capability to use
861 * Test to see if the opener of the socket we received the message
862 * from had when the netlink socket was created and the sender of the
863 * message has has the capability @cap in the user namespace @user_ns.
865 bool netlink_ns_capable(const struct sk_buff *skb,
866 struct user_namespace *user_ns, int cap)
868 return __netlink_ns_capable(&NETLINK_CB(skb), user_ns, cap);
870 EXPORT_SYMBOL(netlink_ns_capable);
873 * netlink_capable - Netlink global message capability test
874 * @skb: socket buffer holding a netlink command from userspace
875 * @cap: The capability to use
877 * Test to see if the opener of the socket we received the message
878 * from had when the netlink socket was created and the sender of the
879 * message has has the capability @cap in all user namespaces.
881 bool netlink_capable(const struct sk_buff *skb, int cap)
883 return netlink_ns_capable(skb, &init_user_ns, cap);
885 EXPORT_SYMBOL(netlink_capable);
888 * netlink_net_capable - Netlink network namespace message capability test
889 * @skb: socket buffer holding a netlink command from userspace
890 * @cap: The capability to use
892 * Test to see if the opener of the socket we received the message
893 * from had when the netlink socket was created and the sender of the
894 * message has has the capability @cap over the network namespace of
895 * the socket we received the message from.
897 bool netlink_net_capable(const struct sk_buff *skb, int cap)
899 return netlink_ns_capable(skb, sock_net(skb->sk)->user_ns, cap);
901 EXPORT_SYMBOL(netlink_net_capable);
903 static inline int netlink_allowed(const struct socket *sock, unsigned int flag)
905 return (nl_table[sock->sk->sk_protocol].flags & flag) ||
906 ns_capable(sock_net(sock->sk)->user_ns, CAP_NET_ADMIN);
910 netlink_update_subscriptions(struct sock *sk, unsigned int subscriptions)
912 struct netlink_sock *nlk = nlk_sk(sk);
914 if (nlk->subscriptions && !subscriptions)
915 __sk_del_bind_node(sk);
916 else if (!nlk->subscriptions && subscriptions)
917 sk_add_bind_node(sk, &nl_table[sk->sk_protocol].mc_list);
918 nlk->subscriptions = subscriptions;
921 static int netlink_realloc_groups(struct sock *sk)
923 struct netlink_sock *nlk = nlk_sk(sk);
925 unsigned long *new_groups;
928 netlink_table_grab();
930 groups = nl_table[sk->sk_protocol].groups;
931 if (!nl_table[sk->sk_protocol].registered) {
936 if (nlk->ngroups >= groups)
939 new_groups = krealloc(nlk->groups, NLGRPSZ(groups), GFP_ATOMIC);
940 if (new_groups == NULL) {
944 memset((char *)new_groups + NLGRPSZ(nlk->ngroups), 0,
945 NLGRPSZ(groups) - NLGRPSZ(nlk->ngroups));
947 nlk->groups = new_groups;
948 nlk->ngroups = groups;
950 netlink_table_ungrab();
954 static void netlink_undo_bind(int group, long unsigned int groups,
957 struct netlink_sock *nlk = nlk_sk(sk);
960 if (!nlk->netlink_unbind)
963 for (undo = 0; undo < group; undo++)
964 if (test_bit(undo, &groups))
965 nlk->netlink_unbind(sock_net(sk), undo + 1);
968 static int netlink_bind(struct socket *sock, struct sockaddr *addr,
971 struct sock *sk = sock->sk;
972 struct net *net = sock_net(sk);
973 struct netlink_sock *nlk = nlk_sk(sk);
974 struct sockaddr_nl *nladdr = (struct sockaddr_nl *)addr;
976 long unsigned int groups = nladdr->nl_groups;
979 if (addr_len < sizeof(struct sockaddr_nl))
982 if (nladdr->nl_family != AF_NETLINK)
985 /* Only superuser is allowed to listen multicasts */
987 if (!netlink_allowed(sock, NL_CFG_F_NONROOT_RECV))
989 err = netlink_realloc_groups(sk);
994 if (nlk->ngroups == 0)
996 else if (nlk->ngroups < 8*sizeof(groups))
997 groups &= (1UL << nlk->ngroups) - 1;
999 /* Paired with WRITE_ONCE() in netlink_insert() */
1000 bound = READ_ONCE(nlk->bound);
1002 /* Ensure nlk->portid is up-to-date. */
1005 if (nladdr->nl_pid != nlk->portid)
1009 if (nlk->netlink_bind && groups) {
1012 /* nl_groups is a u32, so cap the maximum groups we can bind */
1013 for (group = 0; group < BITS_PER_TYPE(u32); group++) {
1014 if (!test_bit(group, &groups))
1016 err = nlk->netlink_bind(net, group + 1);
1019 netlink_undo_bind(group, groups, sk);
1024 /* No need for barriers here as we return to user-space without
1025 * using any of the bound attributes.
1028 err = nladdr->nl_pid ?
1029 netlink_insert(sk, nladdr->nl_pid) :
1030 netlink_autobind(sock);
1032 netlink_undo_bind(BITS_PER_TYPE(u32), groups, sk);
1037 if (!groups && (nlk->groups == NULL || !(u32)nlk->groups[0]))
1040 netlink_table_grab();
1041 netlink_update_subscriptions(sk, nlk->subscriptions +
1043 hweight32(nlk->groups[0]));
1044 nlk->groups[0] = (nlk->groups[0] & ~0xffffffffUL) | groups;
1045 netlink_update_listeners(sk);
1046 netlink_table_ungrab();
1051 static int netlink_connect(struct socket *sock, struct sockaddr *addr,
1052 int alen, int flags)
1055 struct sock *sk = sock->sk;
1056 struct netlink_sock *nlk = nlk_sk(sk);
1057 struct sockaddr_nl *nladdr = (struct sockaddr_nl *)addr;
1059 if (alen < sizeof(addr->sa_family))
1062 if (addr->sa_family == AF_UNSPEC) {
1063 sk->sk_state = NETLINK_UNCONNECTED;
1064 nlk->dst_portid = 0;
1068 if (addr->sa_family != AF_NETLINK)
1071 if (alen < sizeof(struct sockaddr_nl))
1074 if ((nladdr->nl_groups || nladdr->nl_pid) &&
1075 !netlink_allowed(sock, NL_CFG_F_NONROOT_SEND))
1078 /* No need for barriers here as we return to user-space without
1079 * using any of the bound attributes.
1080 * Paired with WRITE_ONCE() in netlink_insert().
1082 if (!READ_ONCE(nlk->bound))
1083 err = netlink_autobind(sock);
1086 sk->sk_state = NETLINK_CONNECTED;
1087 nlk->dst_portid = nladdr->nl_pid;
1088 nlk->dst_group = ffs(nladdr->nl_groups);
1094 static int netlink_getname(struct socket *sock, struct sockaddr *addr,
1095 int *addr_len, int peer)
1097 struct sock *sk = sock->sk;
1098 struct netlink_sock *nlk = nlk_sk(sk);
1099 DECLARE_SOCKADDR(struct sockaddr_nl *, nladdr, addr);
1101 nladdr->nl_family = AF_NETLINK;
1103 *addr_len = sizeof(*nladdr);
1106 nladdr->nl_pid = nlk->dst_portid;
1107 nladdr->nl_groups = netlink_group_mask(nlk->dst_group);
1109 nladdr->nl_pid = nlk->portid;
1110 nladdr->nl_groups = nlk->groups ? nlk->groups[0] : 0;
1115 static int netlink_ioctl(struct socket *sock, unsigned int cmd,
1118 /* try to hand this ioctl down to the NIC drivers.
1120 return -ENOIOCTLCMD;
1123 static struct sock *netlink_getsockbyportid(struct sock *ssk, u32 portid)
1126 struct netlink_sock *nlk;
1128 sock = netlink_lookup(sock_net(ssk), ssk->sk_protocol, portid);
1130 return ERR_PTR(-ECONNREFUSED);
1132 /* Don't bother queuing skb if kernel socket has no input function */
1134 if (sock->sk_state == NETLINK_CONNECTED &&
1135 nlk->dst_portid != nlk_sk(ssk)->portid) {
1137 return ERR_PTR(-ECONNREFUSED);
1142 struct sock *netlink_getsockbyfilp(struct file *filp)
1144 struct inode *inode = file_inode(filp);
1147 if (!S_ISSOCK(inode->i_mode))
1148 return ERR_PTR(-ENOTSOCK);
1150 sock = SOCKET_I(inode)->sk;
1151 if (sock->sk_family != AF_NETLINK)
1152 return ERR_PTR(-EINVAL);
1158 static struct sk_buff *netlink_alloc_large_skb(unsigned int size,
1161 struct sk_buff *skb;
1164 if (size <= NLMSG_GOODSIZE || broadcast)
1165 return alloc_skb(size, GFP_KERNEL);
1167 size = SKB_DATA_ALIGN(size) +
1168 SKB_DATA_ALIGN(sizeof(struct skb_shared_info));
1170 data = vmalloc(size);
1174 skb = __build_skb(data, size);
1178 skb->destructor = netlink_skb_destructor;
1184 * Attach a skb to a netlink socket.
1185 * The caller must hold a reference to the destination socket. On error, the
1186 * reference is dropped. The skb is not send to the destination, just all
1187 * all error checks are performed and memory in the queue is reserved.
1189 * < 0: error. skb freed, reference to sock dropped.
1191 * 1: repeat lookup - reference dropped while waiting for socket memory.
1193 int netlink_attachskb(struct sock *sk, struct sk_buff *skb,
1194 long *timeo, struct sock *ssk)
1196 struct netlink_sock *nlk;
1200 if ((atomic_read(&sk->sk_rmem_alloc) > sk->sk_rcvbuf ||
1201 test_bit(NETLINK_S_CONGESTED, &nlk->state))) {
1202 DECLARE_WAITQUEUE(wait, current);
1204 if (!ssk || netlink_is_kernel(ssk))
1205 netlink_overrun(sk);
1211 __set_current_state(TASK_INTERRUPTIBLE);
1212 add_wait_queue(&nlk->wait, &wait);
1214 if ((atomic_read(&sk->sk_rmem_alloc) > sk->sk_rcvbuf ||
1215 test_bit(NETLINK_S_CONGESTED, &nlk->state)) &&
1216 !sock_flag(sk, SOCK_DEAD))
1217 *timeo = schedule_timeout(*timeo);
1219 __set_current_state(TASK_RUNNING);
1220 remove_wait_queue(&nlk->wait, &wait);
1223 if (signal_pending(current)) {
1225 return sock_intr_errno(*timeo);
1229 netlink_skb_set_owner_r(skb, sk);
1233 static int __netlink_sendskb(struct sock *sk, struct sk_buff *skb)
1237 netlink_deliver_tap(skb);
1239 skb_queue_tail(&sk->sk_receive_queue, skb);
1240 sk->sk_data_ready(sk);
1244 int netlink_sendskb(struct sock *sk, struct sk_buff *skb)
1246 int len = __netlink_sendskb(sk, skb);
1252 void netlink_detachskb(struct sock *sk, struct sk_buff *skb)
1258 static struct sk_buff *netlink_trim(struct sk_buff *skb, gfp_t allocation)
1262 WARN_ON(skb->sk != NULL);
1263 delta = skb->end - skb->tail;
1264 if (is_vmalloc_addr(skb->head) || delta * 2 < skb->truesize)
1267 if (skb_shared(skb)) {
1268 struct sk_buff *nskb = skb_clone(skb, allocation);
1275 if (!pskb_expand_head(skb, 0, -delta, allocation))
1276 skb->truesize -= delta;
1281 static int netlink_unicast_kernel(struct sock *sk, struct sk_buff *skb,
1285 struct netlink_sock *nlk = nlk_sk(sk);
1287 ret = -ECONNREFUSED;
1288 if (nlk->netlink_rcv != NULL) {
1290 netlink_skb_set_owner_r(skb, sk);
1291 NETLINK_CB(skb).sk = ssk;
1292 netlink_deliver_tap_kernel(sk, ssk, skb);
1293 nlk->netlink_rcv(skb);
1302 int netlink_unicast(struct sock *ssk, struct sk_buff *skb,
1303 u32 portid, int nonblock)
1309 skb = netlink_trim(skb, gfp_any());
1311 timeo = sock_sndtimeo(ssk, nonblock);
1313 sk = netlink_getsockbyportid(ssk, portid);
1318 if (netlink_is_kernel(sk))
1319 return netlink_unicast_kernel(sk, skb, ssk);
1321 if (sk_filter(sk, skb)) {
1328 err = netlink_attachskb(sk, skb, &timeo, ssk);
1334 return netlink_sendskb(sk, skb);
1336 EXPORT_SYMBOL(netlink_unicast);
1338 int netlink_has_listeners(struct sock *sk, unsigned int group)
1341 struct listeners *listeners;
1343 BUG_ON(!netlink_is_kernel(sk));
1346 listeners = rcu_dereference(nl_table[sk->sk_protocol].listeners);
1348 if (listeners && group - 1 < nl_table[sk->sk_protocol].groups)
1349 res = test_bit(group - 1, listeners->masks);
1355 EXPORT_SYMBOL_GPL(netlink_has_listeners);
1357 static int netlink_broadcast_deliver(struct sock *sk, struct sk_buff *skb)
1359 struct netlink_sock *nlk = nlk_sk(sk);
1361 if (atomic_read(&sk->sk_rmem_alloc) <= sk->sk_rcvbuf &&
1362 !test_bit(NETLINK_S_CONGESTED, &nlk->state)) {
1363 netlink_skb_set_owner_r(skb, sk);
1364 __netlink_sendskb(sk, skb);
1365 return atomic_read(&sk->sk_rmem_alloc) > (sk->sk_rcvbuf >> 1);
1370 struct netlink_broadcast_data {
1371 struct sock *exclude_sk;
1376 int delivery_failure;
1380 struct sk_buff *skb, *skb2;
1381 int (*tx_filter)(struct sock *dsk, struct sk_buff *skb, void *data);
1385 static void do_one_broadcast(struct sock *sk,
1386 struct netlink_broadcast_data *p)
1388 struct netlink_sock *nlk = nlk_sk(sk);
1391 if (p->exclude_sk == sk)
1394 if (nlk->portid == p->portid || p->group - 1 >= nlk->ngroups ||
1395 !test_bit(p->group - 1, nlk->groups))
1398 if (!net_eq(sock_net(sk), p->net)) {
1399 if (!(nlk->flags & NETLINK_F_LISTEN_ALL_NSID))
1402 if (!peernet_has_id(sock_net(sk), p->net))
1405 if (!file_ns_capable(sk->sk_socket->file, p->net->user_ns,
1411 netlink_overrun(sk);
1416 if (p->skb2 == NULL) {
1417 if (skb_shared(p->skb)) {
1418 p->skb2 = skb_clone(p->skb, p->allocation);
1420 p->skb2 = skb_get(p->skb);
1422 * skb ownership may have been set when
1423 * delivered to a previous socket.
1425 skb_orphan(p->skb2);
1428 if (p->skb2 == NULL) {
1429 netlink_overrun(sk);
1430 /* Clone failed. Notify ALL listeners. */
1432 if (nlk->flags & NETLINK_F_BROADCAST_SEND_ERROR)
1433 p->delivery_failure = 1;
1436 if (p->tx_filter && p->tx_filter(sk, p->skb2, p->tx_data)) {
1441 if (sk_filter(sk, p->skb2)) {
1446 NETLINK_CB(p->skb2).nsid = peernet2id(sock_net(sk), p->net);
1447 NETLINK_CB(p->skb2).nsid_is_set = true;
1448 val = netlink_broadcast_deliver(sk, p->skb2);
1450 netlink_overrun(sk);
1451 if (nlk->flags & NETLINK_F_BROADCAST_SEND_ERROR)
1452 p->delivery_failure = 1;
1454 p->congested |= val;
1462 int netlink_broadcast_filtered(struct sock *ssk, struct sk_buff *skb, u32 portid,
1463 u32 group, gfp_t allocation,
1464 int (*filter)(struct sock *dsk, struct sk_buff *skb, void *data),
1467 struct net *net = sock_net(ssk);
1468 struct netlink_broadcast_data info;
1471 skb = netlink_trim(skb, allocation);
1473 info.exclude_sk = ssk;
1475 info.portid = portid;
1478 info.delivery_failure = 0;
1481 info.allocation = allocation;
1484 info.tx_filter = filter;
1485 info.tx_data = filter_data;
1487 /* While we sleep in clone, do not allow to change socket list */
1489 netlink_lock_table();
1491 sk_for_each_bound(sk, &nl_table[ssk->sk_protocol].mc_list)
1492 do_one_broadcast(sk, &info);
1496 netlink_unlock_table();
1498 if (info.delivery_failure) {
1499 kfree_skb(info.skb2);
1502 consume_skb(info.skb2);
1504 if (info.delivered) {
1505 if (info.congested && gfpflags_allow_blocking(allocation))
1511 EXPORT_SYMBOL(netlink_broadcast_filtered);
1513 int netlink_broadcast(struct sock *ssk, struct sk_buff *skb, u32 portid,
1514 u32 group, gfp_t allocation)
1516 return netlink_broadcast_filtered(ssk, skb, portid, group, allocation,
1519 EXPORT_SYMBOL(netlink_broadcast);
1521 struct netlink_set_err_data {
1522 struct sock *exclude_sk;
1528 static int do_one_set_err(struct sock *sk, struct netlink_set_err_data *p)
1530 struct netlink_sock *nlk = nlk_sk(sk);
1533 if (sk == p->exclude_sk)
1536 if (!net_eq(sock_net(sk), sock_net(p->exclude_sk)))
1539 if (nlk->portid == p->portid || p->group - 1 >= nlk->ngroups ||
1540 !test_bit(p->group - 1, nlk->groups))
1543 if (p->code == ENOBUFS && nlk->flags & NETLINK_F_RECV_NO_ENOBUFS) {
1548 sk->sk_err = p->code;
1549 sk->sk_error_report(sk);
1555 * netlink_set_err - report error to broadcast listeners
1556 * @ssk: the kernel netlink socket, as returned by netlink_kernel_create()
1557 * @portid: the PORTID of a process that we want to skip (if any)
1558 * @group: the broadcast group that will notice the error
1559 * @code: error code, must be negative (as usual in kernelspace)
1561 * This function returns the number of broadcast listeners that have set the
1562 * NETLINK_NO_ENOBUFS socket option.
1564 int netlink_set_err(struct sock *ssk, u32 portid, u32 group, int code)
1566 struct netlink_set_err_data info;
1570 info.exclude_sk = ssk;
1571 info.portid = portid;
1573 /* sk->sk_err wants a positive error value */
1576 read_lock(&nl_table_lock);
1578 sk_for_each_bound(sk, &nl_table[ssk->sk_protocol].mc_list)
1579 ret += do_one_set_err(sk, &info);
1581 read_unlock(&nl_table_lock);
1584 EXPORT_SYMBOL(netlink_set_err);
1586 /* must be called with netlink table grabbed */
1587 static void netlink_update_socket_mc(struct netlink_sock *nlk,
1591 int old, new = !!is_new, subscriptions;
1593 old = test_bit(group - 1, nlk->groups);
1594 subscriptions = nlk->subscriptions - old + new;
1596 __set_bit(group - 1, nlk->groups);
1598 __clear_bit(group - 1, nlk->groups);
1599 netlink_update_subscriptions(&nlk->sk, subscriptions);
1600 netlink_update_listeners(&nlk->sk);
1603 static int netlink_setsockopt(struct socket *sock, int level, int optname,
1604 char __user *optval, unsigned int optlen)
1606 struct sock *sk = sock->sk;
1607 struct netlink_sock *nlk = nlk_sk(sk);
1608 unsigned int val = 0;
1611 if (level != SOL_NETLINK)
1612 return -ENOPROTOOPT;
1614 if (optlen >= sizeof(int) &&
1615 get_user(val, (unsigned int __user *)optval))
1619 case NETLINK_PKTINFO:
1621 nlk->flags |= NETLINK_F_RECV_PKTINFO;
1623 nlk->flags &= ~NETLINK_F_RECV_PKTINFO;
1626 case NETLINK_ADD_MEMBERSHIP:
1627 case NETLINK_DROP_MEMBERSHIP: {
1628 if (!netlink_allowed(sock, NL_CFG_F_NONROOT_RECV))
1630 err = netlink_realloc_groups(sk);
1633 if (!val || val - 1 >= nlk->ngroups)
1635 if (optname == NETLINK_ADD_MEMBERSHIP && nlk->netlink_bind) {
1636 err = nlk->netlink_bind(sock_net(sk), val);
1640 netlink_table_grab();
1641 netlink_update_socket_mc(nlk, val,
1642 optname == NETLINK_ADD_MEMBERSHIP);
1643 netlink_table_ungrab();
1644 if (optname == NETLINK_DROP_MEMBERSHIP && nlk->netlink_unbind)
1645 nlk->netlink_unbind(sock_net(sk), val);
1650 case NETLINK_BROADCAST_ERROR:
1652 nlk->flags |= NETLINK_F_BROADCAST_SEND_ERROR;
1654 nlk->flags &= ~NETLINK_F_BROADCAST_SEND_ERROR;
1657 case NETLINK_NO_ENOBUFS:
1659 nlk->flags |= NETLINK_F_RECV_NO_ENOBUFS;
1660 clear_bit(NETLINK_S_CONGESTED, &nlk->state);
1661 wake_up_interruptible(&nlk->wait);
1663 nlk->flags &= ~NETLINK_F_RECV_NO_ENOBUFS;
1667 case NETLINK_LISTEN_ALL_NSID:
1668 if (!ns_capable(sock_net(sk)->user_ns, CAP_NET_BROADCAST))
1672 nlk->flags |= NETLINK_F_LISTEN_ALL_NSID;
1674 nlk->flags &= ~NETLINK_F_LISTEN_ALL_NSID;
1677 case NETLINK_CAP_ACK:
1679 nlk->flags |= NETLINK_F_CAP_ACK;
1681 nlk->flags &= ~NETLINK_F_CAP_ACK;
1690 static int netlink_getsockopt(struct socket *sock, int level, int optname,
1691 char __user *optval, int __user *optlen)
1693 struct sock *sk = sock->sk;
1694 struct netlink_sock *nlk = nlk_sk(sk);
1697 if (level != SOL_NETLINK)
1698 return -ENOPROTOOPT;
1700 if (get_user(len, optlen))
1706 case NETLINK_PKTINFO:
1707 if (len < sizeof(int))
1710 val = nlk->flags & NETLINK_F_RECV_PKTINFO ? 1 : 0;
1711 if (put_user(len, optlen) ||
1712 put_user(val, optval))
1716 case NETLINK_BROADCAST_ERROR:
1717 if (len < sizeof(int))
1720 val = nlk->flags & NETLINK_F_BROADCAST_SEND_ERROR ? 1 : 0;
1721 if (put_user(len, optlen) ||
1722 put_user(val, optval))
1726 case NETLINK_NO_ENOBUFS:
1727 if (len < sizeof(int))
1730 val = nlk->flags & NETLINK_F_RECV_NO_ENOBUFS ? 1 : 0;
1731 if (put_user(len, optlen) ||
1732 put_user(val, optval))
1736 case NETLINK_LIST_MEMBERSHIPS: {
1737 int pos, idx, shift;
1740 netlink_lock_table();
1741 for (pos = 0; pos * 8 < nlk->ngroups; pos += sizeof(u32)) {
1742 if (len - pos < sizeof(u32))
1745 idx = pos / sizeof(unsigned long);
1746 shift = (pos % sizeof(unsigned long)) * 8;
1747 if (put_user((u32)(nlk->groups[idx] >> shift),
1748 (u32 __user *)(optval + pos))) {
1753 if (put_user(ALIGN(nlk->ngroups / 8, sizeof(u32)), optlen))
1755 netlink_unlock_table();
1758 case NETLINK_CAP_ACK:
1759 if (len < sizeof(int))
1762 val = nlk->flags & NETLINK_F_CAP_ACK ? 1 : 0;
1763 if (put_user(len, optlen) ||
1764 put_user(val, optval))
1774 static void netlink_cmsg_recv_pktinfo(struct msghdr *msg, struct sk_buff *skb)
1776 struct nl_pktinfo info;
1778 info.group = NETLINK_CB(skb).dst_group;
1779 put_cmsg(msg, SOL_NETLINK, NETLINK_PKTINFO, sizeof(info), &info);
1782 static void netlink_cmsg_listen_all_nsid(struct sock *sk, struct msghdr *msg,
1783 struct sk_buff *skb)
1785 if (!NETLINK_CB(skb).nsid_is_set)
1788 put_cmsg(msg, SOL_NETLINK, NETLINK_LISTEN_ALL_NSID, sizeof(int),
1789 &NETLINK_CB(skb).nsid);
1792 static int netlink_sendmsg(struct socket *sock, struct msghdr *msg, size_t len)
1794 struct sock *sk = sock->sk;
1795 struct netlink_sock *nlk = nlk_sk(sk);
1796 DECLARE_SOCKADDR(struct sockaddr_nl *, addr, msg->msg_name);
1799 struct sk_buff *skb;
1801 struct scm_cookie scm;
1802 u32 netlink_skb_flags = 0;
1804 if (msg->msg_flags&MSG_OOB)
1808 pr_warn_once("Zero length message leads to an empty skb\n");
1812 err = scm_send(sock, msg, &scm, true);
1816 if (msg->msg_namelen) {
1818 if (msg->msg_namelen < sizeof(struct sockaddr_nl))
1820 if (addr->nl_family != AF_NETLINK)
1822 dst_portid = addr->nl_pid;
1823 dst_group = ffs(addr->nl_groups);
1825 if ((dst_group || dst_portid) &&
1826 !netlink_allowed(sock, NL_CFG_F_NONROOT_SEND))
1828 netlink_skb_flags |= NETLINK_SKB_DST;
1830 dst_portid = nlk->dst_portid;
1831 dst_group = nlk->dst_group;
1834 /* Paired with WRITE_ONCE() in netlink_insert() */
1835 if (!READ_ONCE(nlk->bound)) {
1836 err = netlink_autobind(sock);
1840 /* Ensure nlk is hashed and visible. */
1845 if (len > sk->sk_sndbuf - 32)
1848 skb = netlink_alloc_large_skb(len, dst_group);
1852 NETLINK_CB(skb).portid = nlk->portid;
1853 NETLINK_CB(skb).dst_group = dst_group;
1854 NETLINK_CB(skb).creds = scm.creds;
1855 NETLINK_CB(skb).flags = netlink_skb_flags;
1858 if (memcpy_from_msg(skb_put(skb, len), msg, len)) {
1863 err = security_netlink_send(sk, skb);
1870 atomic_inc(&skb->users);
1871 netlink_broadcast(sk, skb, dst_portid, dst_group, GFP_KERNEL);
1873 err = netlink_unicast(sk, skb, dst_portid, msg->msg_flags&MSG_DONTWAIT);
1880 static int netlink_recvmsg(struct socket *sock, struct msghdr *msg, size_t len,
1883 struct scm_cookie scm;
1884 struct sock *sk = sock->sk;
1885 struct netlink_sock *nlk = nlk_sk(sk);
1886 int noblock = flags&MSG_DONTWAIT;
1888 struct sk_buff *skb, *data_skb;
1896 skb = skb_recv_datagram(sk, flags, noblock, &err);
1902 #ifdef CONFIG_COMPAT_NETLINK_MESSAGES
1903 if (unlikely(skb_shinfo(skb)->frag_list)) {
1905 * If this skb has a frag_list, then here that means that we
1906 * will have to use the frag_list skb's data for compat tasks
1907 * and the regular skb's data for normal (non-compat) tasks.
1909 * If we need to send the compat skb, assign it to the
1910 * 'data_skb' variable so that it will be used below for data
1911 * copying. We keep 'skb' for everything else, including
1912 * freeing both later.
1914 if (flags & MSG_CMSG_COMPAT)
1915 data_skb = skb_shinfo(skb)->frag_list;
1919 /* Record the max length of recvmsg() calls for future allocations */
1920 nlk->max_recvmsg_len = max(nlk->max_recvmsg_len, len);
1921 nlk->max_recvmsg_len = min_t(size_t, nlk->max_recvmsg_len,
1922 SKB_WITH_OVERHEAD(32768));
1924 copied = data_skb->len;
1926 msg->msg_flags |= MSG_TRUNC;
1930 skb_reset_transport_header(data_skb);
1931 err = skb_copy_datagram_msg(data_skb, 0, msg, copied);
1933 if (msg->msg_name) {
1934 DECLARE_SOCKADDR(struct sockaddr_nl *, addr, msg->msg_name);
1935 addr->nl_family = AF_NETLINK;
1937 addr->nl_pid = NETLINK_CB(skb).portid;
1938 addr->nl_groups = netlink_group_mask(NETLINK_CB(skb).dst_group);
1939 msg->msg_namelen = sizeof(*addr);
1942 if (nlk->flags & NETLINK_F_RECV_PKTINFO)
1943 netlink_cmsg_recv_pktinfo(msg, skb);
1944 if (nlk->flags & NETLINK_F_LISTEN_ALL_NSID)
1945 netlink_cmsg_listen_all_nsid(sk, msg, skb);
1947 memset(&scm, 0, sizeof(scm));
1948 scm.creds = *NETLINK_CREDS(skb);
1949 if (flags & MSG_TRUNC)
1950 copied = data_skb->len;
1952 skb_free_datagram(sk, skb);
1954 if (nlk->cb_running &&
1955 atomic_read(&sk->sk_rmem_alloc) <= sk->sk_rcvbuf / 2) {
1956 ret = netlink_dump(sk);
1959 sk->sk_error_report(sk);
1963 scm_recv(sock, msg, &scm, flags);
1965 netlink_rcv_wake(sk);
1966 return err ? : copied;
1969 static void netlink_data_ready(struct sock *sk)
1975 * We export these functions to other modules. They provide a
1976 * complete set of kernel non-blocking support for message
1981 __netlink_kernel_create(struct net *net, int unit, struct module *module,
1982 struct netlink_kernel_cfg *cfg)
1984 struct socket *sock;
1986 struct netlink_sock *nlk;
1987 struct listeners *listeners = NULL;
1988 struct mutex *cb_mutex = cfg ? cfg->cb_mutex : NULL;
1989 unsigned int groups;
1993 if (unit < 0 || unit >= MAX_LINKS)
1996 if (sock_create_lite(PF_NETLINK, SOCK_DGRAM, unit, &sock))
1999 if (__netlink_create(net, sock, cb_mutex, unit, 1) < 0)
2000 goto out_sock_release_nosk;
2004 if (!cfg || cfg->groups < 32)
2007 groups = cfg->groups;
2009 listeners = kzalloc(sizeof(*listeners) + NLGRPSZ(groups), GFP_KERNEL);
2011 goto out_sock_release;
2013 sk->sk_data_ready = netlink_data_ready;
2014 if (cfg && cfg->input)
2015 nlk_sk(sk)->netlink_rcv = cfg->input;
2017 if (netlink_insert(sk, 0))
2018 goto out_sock_release;
2021 nlk->flags |= NETLINK_F_KERNEL_SOCKET;
2023 netlink_table_grab();
2024 if (!nl_table[unit].registered) {
2025 nl_table[unit].groups = groups;
2026 rcu_assign_pointer(nl_table[unit].listeners, listeners);
2027 nl_table[unit].cb_mutex = cb_mutex;
2028 nl_table[unit].module = module;
2030 nl_table[unit].bind = cfg->bind;
2031 nl_table[unit].unbind = cfg->unbind;
2032 nl_table[unit].flags = cfg->flags;
2034 nl_table[unit].compare = cfg->compare;
2036 nl_table[unit].registered = 1;
2039 nl_table[unit].registered++;
2041 netlink_table_ungrab();
2046 netlink_kernel_release(sk);
2049 out_sock_release_nosk:
2053 EXPORT_SYMBOL(__netlink_kernel_create);
2056 netlink_kernel_release(struct sock *sk)
2058 if (sk == NULL || sk->sk_socket == NULL)
2061 sock_release(sk->sk_socket);
2063 EXPORT_SYMBOL(netlink_kernel_release);
2065 int __netlink_change_ngroups(struct sock *sk, unsigned int groups)
2067 struct listeners *new, *old;
2068 struct netlink_table *tbl = &nl_table[sk->sk_protocol];
2073 if (NLGRPSZ(tbl->groups) < NLGRPSZ(groups)) {
2074 new = kzalloc(sizeof(*new) + NLGRPSZ(groups), GFP_ATOMIC);
2077 old = nl_deref_protected(tbl->listeners);
2078 memcpy(new->masks, old->masks, NLGRPSZ(tbl->groups));
2079 rcu_assign_pointer(tbl->listeners, new);
2081 kfree_rcu(old, rcu);
2083 tbl->groups = groups;
2089 * netlink_change_ngroups - change number of multicast groups
2091 * This changes the number of multicast groups that are available
2092 * on a certain netlink family. Note that it is not possible to
2093 * change the number of groups to below 32. Also note that it does
2094 * not implicitly call netlink_clear_multicast_users() when the
2095 * number of groups is reduced.
2097 * @sk: The kernel netlink socket, as returned by netlink_kernel_create().
2098 * @groups: The new number of groups.
2100 int netlink_change_ngroups(struct sock *sk, unsigned int groups)
2104 netlink_table_grab();
2105 err = __netlink_change_ngroups(sk, groups);
2106 netlink_table_ungrab();
2111 void __netlink_clear_multicast_users(struct sock *ksk, unsigned int group)
2114 struct netlink_table *tbl = &nl_table[ksk->sk_protocol];
2116 sk_for_each_bound(sk, &tbl->mc_list)
2117 netlink_update_socket_mc(nlk_sk(sk), group, 0);
2121 __nlmsg_put(struct sk_buff *skb, u32 portid, u32 seq, int type, int len, int flags)
2123 struct nlmsghdr *nlh;
2124 int size = nlmsg_msg_size(len);
2126 nlh = (struct nlmsghdr *)skb_put(skb, NLMSG_ALIGN(size));
2127 nlh->nlmsg_type = type;
2128 nlh->nlmsg_len = size;
2129 nlh->nlmsg_flags = flags;
2130 nlh->nlmsg_pid = portid;
2131 nlh->nlmsg_seq = seq;
2132 if (!__builtin_constant_p(size) || NLMSG_ALIGN(size) - size != 0)
2133 memset(nlmsg_data(nlh) + len, 0, NLMSG_ALIGN(size) - size);
2136 EXPORT_SYMBOL(__nlmsg_put);
2139 * It looks a bit ugly.
2140 * It would be better to create kernel thread.
2143 static int netlink_dump(struct sock *sk)
2145 struct netlink_sock *nlk = nlk_sk(sk);
2146 struct netlink_callback *cb;
2147 struct sk_buff *skb = NULL;
2148 struct nlmsghdr *nlh;
2149 struct module *module;
2154 mutex_lock(nlk->cb_mutex);
2155 if (!nlk->cb_running) {
2160 if (atomic_read(&sk->sk_rmem_alloc) >= sk->sk_rcvbuf)
2163 /* NLMSG_GOODSIZE is small to avoid high order allocations being
2164 * required, but it makes sense to _attempt_ a 16K bytes allocation
2165 * to reduce number of system calls on dump operations, if user
2166 * ever provided a big enough buffer.
2169 alloc_min_size = max_t(int, cb->min_dump_alloc, NLMSG_GOODSIZE);
2171 if (alloc_min_size < nlk->max_recvmsg_len) {
2172 alloc_size = nlk->max_recvmsg_len;
2173 skb = alloc_skb(alloc_size,
2174 (GFP_KERNEL & ~__GFP_DIRECT_RECLAIM) |
2175 __GFP_NOWARN | __GFP_NORETRY);
2178 alloc_size = alloc_min_size;
2179 skb = alloc_skb(alloc_size, GFP_KERNEL);
2184 /* Trim skb to allocated size. User is expected to provide buffer as
2185 * large as max(min_dump_alloc, 16KiB (mac_recvmsg_len capped at
2186 * netlink_recvmsg())). dump will pack as many smaller messages as
2187 * could fit within the allocated skb. skb is typically allocated
2188 * with larger space than required (could be as much as near 2x the
2189 * requested size with align to next power of 2 approach). Allowing
2190 * dump to use the excess space makes it difficult for a user to have a
2191 * reasonable static buffer based on the expected largest dump of a
2192 * single netdev. The outcome is MSG_TRUNC error.
2194 skb_reserve(skb, skb_tailroom(skb) - alloc_size);
2195 netlink_skb_set_owner_r(skb, sk);
2197 if (nlk->dump_done_errno > 0)
2198 nlk->dump_done_errno = cb->dump(skb, cb);
2200 if (nlk->dump_done_errno > 0 ||
2201 skb_tailroom(skb) < nlmsg_total_size(sizeof(nlk->dump_done_errno))) {
2202 mutex_unlock(nlk->cb_mutex);
2204 if (sk_filter(sk, skb))
2207 __netlink_sendskb(sk, skb);
2211 nlh = nlmsg_put_answer(skb, cb, NLMSG_DONE,
2212 sizeof(nlk->dump_done_errno), NLM_F_MULTI);
2216 nl_dump_check_consistent(cb, nlh);
2218 memcpy(nlmsg_data(nlh), &nlk->dump_done_errno,
2219 sizeof(nlk->dump_done_errno));
2221 if (sk_filter(sk, skb))
2224 __netlink_sendskb(sk, skb);
2229 nlk->cb_running = false;
2230 module = cb->module;
2232 mutex_unlock(nlk->cb_mutex);
2238 mutex_unlock(nlk->cb_mutex);
2243 int __netlink_dump_start(struct sock *ssk, struct sk_buff *skb,
2244 const struct nlmsghdr *nlh,
2245 struct netlink_dump_control *control)
2247 struct netlink_callback *cb;
2249 struct netlink_sock *nlk;
2252 atomic_inc(&skb->users);
2254 sk = netlink_lookup(sock_net(ssk), ssk->sk_protocol, NETLINK_CB(skb).portid);
2256 ret = -ECONNREFUSED;
2261 mutex_lock(nlk->cb_mutex);
2262 /* A dump is in progress... */
2263 if (nlk->cb_running) {
2267 /* add reference of module which cb->dump belongs to */
2268 if (!try_module_get(control->module)) {
2269 ret = -EPROTONOSUPPORT;
2274 memset(cb, 0, sizeof(*cb));
2275 cb->start = control->start;
2276 cb->dump = control->dump;
2277 cb->done = control->done;
2279 cb->data = control->data;
2280 cb->module = control->module;
2281 cb->min_dump_alloc = control->min_dump_alloc;
2285 ret = cb->start(cb);
2290 nlk->cb_running = true;
2291 nlk->dump_done_errno = INT_MAX;
2293 mutex_unlock(nlk->cb_mutex);
2295 ret = netlink_dump(sk);
2302 /* We successfully started a dump, by returning -EINTR we
2303 * signal not to send ACK even if it was requested.
2308 module_put(control->module);
2311 mutex_unlock(nlk->cb_mutex);
2316 EXPORT_SYMBOL(__netlink_dump_start);
2318 void netlink_ack(struct sk_buff *in_skb, struct nlmsghdr *nlh, int err)
2320 struct sk_buff *skb;
2321 struct nlmsghdr *rep;
2322 struct nlmsgerr *errmsg;
2323 size_t payload = sizeof(*errmsg);
2324 struct netlink_sock *nlk = nlk_sk(NETLINK_CB(in_skb).sk);
2326 /* Error messages get the original request appened, unless the user
2327 * requests to cap the error message.
2329 if (!(nlk->flags & NETLINK_F_CAP_ACK) && err)
2330 payload += nlmsg_len(nlh);
2332 skb = nlmsg_new(payload, GFP_KERNEL);
2336 sk = netlink_lookup(sock_net(in_skb->sk),
2337 in_skb->sk->sk_protocol,
2338 NETLINK_CB(in_skb).portid);
2340 sk->sk_err = ENOBUFS;
2341 sk->sk_error_report(sk);
2347 rep = __nlmsg_put(skb, NETLINK_CB(in_skb).portid, nlh->nlmsg_seq,
2348 NLMSG_ERROR, payload, 0);
2349 errmsg = nlmsg_data(rep);
2350 errmsg->error = err;
2351 memcpy(&errmsg->msg, nlh, payload > sizeof(*errmsg) ? nlh->nlmsg_len : sizeof(*nlh));
2352 netlink_unicast(in_skb->sk, skb, NETLINK_CB(in_skb).portid, MSG_DONTWAIT);
2354 EXPORT_SYMBOL(netlink_ack);
2356 int netlink_rcv_skb(struct sk_buff *skb, int (*cb)(struct sk_buff *,
2359 struct nlmsghdr *nlh;
2362 while (skb->len >= nlmsg_total_size(0)) {
2365 nlh = nlmsg_hdr(skb);
2368 if (nlh->nlmsg_len < NLMSG_HDRLEN || skb->len < nlh->nlmsg_len)
2371 /* Only requests are handled by the kernel */
2372 if (!(nlh->nlmsg_flags & NLM_F_REQUEST))
2375 /* Skip control messages */
2376 if (nlh->nlmsg_type < NLMSG_MIN_TYPE)
2384 if (nlh->nlmsg_flags & NLM_F_ACK || err)
2385 netlink_ack(skb, nlh, err);
2388 msglen = NLMSG_ALIGN(nlh->nlmsg_len);
2389 if (msglen > skb->len)
2391 skb_pull(skb, msglen);
2396 EXPORT_SYMBOL(netlink_rcv_skb);
2399 * nlmsg_notify - send a notification netlink message
2400 * @sk: netlink socket to use
2401 * @skb: notification message
2402 * @portid: destination netlink portid for reports or 0
2403 * @group: destination multicast group or 0
2404 * @report: 1 to report back, 0 to disable
2405 * @flags: allocation flags
2407 int nlmsg_notify(struct sock *sk, struct sk_buff *skb, u32 portid,
2408 unsigned int group, int report, gfp_t flags)
2413 int exclude_portid = 0;
2416 atomic_inc(&skb->users);
2417 exclude_portid = portid;
2420 /* errors reported via destination sk->sk_err, but propagate
2421 * delivery errors if NETLINK_BROADCAST_ERROR flag is set */
2422 err = nlmsg_multicast(sk, skb, exclude_portid, group, flags);
2430 err2 = nlmsg_unicast(sk, skb, portid);
2437 EXPORT_SYMBOL(nlmsg_notify);
2439 #ifdef CONFIG_PROC_FS
2440 struct nl_seq_iter {
2441 struct seq_net_private p;
2442 struct rhashtable_iter hti;
2446 static int netlink_walk_start(struct nl_seq_iter *iter)
2450 err = rhashtable_walk_init(&nl_table[iter->link].hash, &iter->hti,
2453 iter->link = MAX_LINKS;
2457 err = rhashtable_walk_start(&iter->hti);
2458 return err == -EAGAIN ? 0 : err;
2461 static void netlink_walk_stop(struct nl_seq_iter *iter)
2463 rhashtable_walk_stop(&iter->hti);
2464 rhashtable_walk_exit(&iter->hti);
2467 static void *__netlink_seq_next(struct seq_file *seq)
2469 struct nl_seq_iter *iter = seq->private;
2470 struct netlink_sock *nlk;
2476 nlk = rhashtable_walk_next(&iter->hti);
2479 if (PTR_ERR(nlk) == -EAGAIN)
2488 netlink_walk_stop(iter);
2489 if (++iter->link >= MAX_LINKS)
2492 err = netlink_walk_start(iter);
2494 return ERR_PTR(err);
2496 } while (sock_net(&nlk->sk) != seq_file_net(seq));
2501 static void *netlink_seq_start(struct seq_file *seq, loff_t *posp)
2503 struct nl_seq_iter *iter = seq->private;
2504 void *obj = SEQ_START_TOKEN;
2510 err = netlink_walk_start(iter);
2512 return ERR_PTR(err);
2514 for (pos = *posp; pos && obj && !IS_ERR(obj); pos--)
2515 obj = __netlink_seq_next(seq);
2520 static void *netlink_seq_next(struct seq_file *seq, void *v, loff_t *pos)
2523 return __netlink_seq_next(seq);
2526 static void netlink_seq_stop(struct seq_file *seq, void *v)
2528 struct nl_seq_iter *iter = seq->private;
2530 if (iter->link >= MAX_LINKS)
2533 netlink_walk_stop(iter);
2537 static int netlink_seq_show(struct seq_file *seq, void *v)
2539 if (v == SEQ_START_TOKEN) {
2541 "sk Eth Pid Groups "
2542 "Rmem Wmem Dump Locks Drops Inode\n");
2545 struct netlink_sock *nlk = nlk_sk(s);
2547 seq_printf(seq, "%pK %-3d %-6u %08x %-8d %-8d %d %-8d %-8d %-8lu\n",
2551 nlk->groups ? (u32)nlk->groups[0] : 0,
2552 sk_rmem_alloc_get(s),
2553 sk_wmem_alloc_get(s),
2555 atomic_read(&s->sk_refcnt),
2556 atomic_read(&s->sk_drops),
2564 static const struct seq_operations netlink_seq_ops = {
2565 .start = netlink_seq_start,
2566 .next = netlink_seq_next,
2567 .stop = netlink_seq_stop,
2568 .show = netlink_seq_show,
2572 static int netlink_seq_open(struct inode *inode, struct file *file)
2574 return seq_open_net(inode, file, &netlink_seq_ops,
2575 sizeof(struct nl_seq_iter));
2578 static const struct file_operations netlink_seq_fops = {
2579 .owner = THIS_MODULE,
2580 .open = netlink_seq_open,
2582 .llseek = seq_lseek,
2583 .release = seq_release_net,
2588 int netlink_register_notifier(struct notifier_block *nb)
2590 return atomic_notifier_chain_register(&netlink_chain, nb);
2592 EXPORT_SYMBOL(netlink_register_notifier);
2594 int netlink_unregister_notifier(struct notifier_block *nb)
2596 return atomic_notifier_chain_unregister(&netlink_chain, nb);
2598 EXPORT_SYMBOL(netlink_unregister_notifier);
2600 static const struct proto_ops netlink_ops = {
2601 .family = PF_NETLINK,
2602 .owner = THIS_MODULE,
2603 .release = netlink_release,
2604 .bind = netlink_bind,
2605 .connect = netlink_connect,
2606 .socketpair = sock_no_socketpair,
2607 .accept = sock_no_accept,
2608 .getname = netlink_getname,
2609 .poll = datagram_poll,
2610 .ioctl = netlink_ioctl,
2611 .listen = sock_no_listen,
2612 .shutdown = sock_no_shutdown,
2613 .setsockopt = netlink_setsockopt,
2614 .getsockopt = netlink_getsockopt,
2615 .sendmsg = netlink_sendmsg,
2616 .recvmsg = netlink_recvmsg,
2617 .mmap = sock_no_mmap,
2618 .sendpage = sock_no_sendpage,
2621 static const struct net_proto_family netlink_family_ops = {
2622 .family = PF_NETLINK,
2623 .create = netlink_create,
2624 .owner = THIS_MODULE, /* for consistency 8) */
2627 static int __net_init netlink_net_init(struct net *net)
2629 #ifdef CONFIG_PROC_FS
2630 if (!proc_create("netlink", 0, net->proc_net, &netlink_seq_fops))
2636 static void __net_exit netlink_net_exit(struct net *net)
2638 #ifdef CONFIG_PROC_FS
2639 remove_proc_entry("netlink", net->proc_net);
2643 static void __init netlink_add_usersock_entry(void)
2645 struct listeners *listeners;
2648 listeners = kzalloc(sizeof(*listeners) + NLGRPSZ(groups), GFP_KERNEL);
2650 panic("netlink_add_usersock_entry: Cannot allocate listeners\n");
2652 netlink_table_grab();
2654 nl_table[NETLINK_USERSOCK].groups = groups;
2655 rcu_assign_pointer(nl_table[NETLINK_USERSOCK].listeners, listeners);
2656 nl_table[NETLINK_USERSOCK].module = THIS_MODULE;
2657 nl_table[NETLINK_USERSOCK].registered = 1;
2658 nl_table[NETLINK_USERSOCK].flags = NL_CFG_F_NONROOT_SEND;
2660 netlink_table_ungrab();
2663 static struct pernet_operations __net_initdata netlink_net_ops = {
2664 .init = netlink_net_init,
2665 .exit = netlink_net_exit,
2668 static inline u32 netlink_hash(const void *data, u32 len, u32 seed)
2670 const struct netlink_sock *nlk = data;
2671 struct netlink_compare_arg arg;
2673 netlink_compare_arg_init(&arg, sock_net(&nlk->sk), nlk->portid);
2674 return jhash2((u32 *)&arg, netlink_compare_arg_len / sizeof(u32), seed);
2677 static const struct rhashtable_params netlink_rhashtable_params = {
2678 .head_offset = offsetof(struct netlink_sock, node),
2679 .key_len = netlink_compare_arg_len,
2680 .obj_hashfn = netlink_hash,
2681 .obj_cmpfn = netlink_compare,
2682 .automatic_shrinking = true,
2685 static int __init netlink_proto_init(void)
2688 int err = proto_register(&netlink_proto, 0);
2693 BUILD_BUG_ON(sizeof(struct netlink_skb_parms) > FIELD_SIZEOF(struct sk_buff, cb));
2695 nl_table = kcalloc(MAX_LINKS, sizeof(*nl_table), GFP_KERNEL);
2699 for (i = 0; i < MAX_LINKS; i++) {
2700 if (rhashtable_init(&nl_table[i].hash,
2701 &netlink_rhashtable_params) < 0) {
2703 rhashtable_destroy(&nl_table[i].hash);
2709 INIT_LIST_HEAD(&netlink_tap_all);
2711 netlink_add_usersock_entry();
2713 sock_register(&netlink_family_ops);
2714 register_pernet_subsys(&netlink_net_ops);
2715 /* The netlink device handler may be needed early. */
2720 panic("netlink_init: Cannot allocate nl_table\n");
2723 core_initcall(netlink_proto_init);