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 nlk_sk(sk)->bound = portid;
584 static void netlink_remove(struct sock *sk)
586 struct netlink_table *table;
588 table = &nl_table[sk->sk_protocol];
589 if (!rhashtable_remove_fast(&table->hash, &nlk_sk(sk)->node,
590 netlink_rhashtable_params)) {
591 WARN_ON(atomic_read(&sk->sk_refcnt) == 1);
595 netlink_table_grab();
596 if (nlk_sk(sk)->subscriptions) {
597 __sk_del_bind_node(sk);
598 netlink_update_listeners(sk);
600 if (sk->sk_protocol == NETLINK_GENERIC)
601 atomic_inc(&genl_sk_destructing_cnt);
602 netlink_table_ungrab();
605 static struct proto netlink_proto = {
607 .owner = THIS_MODULE,
608 .obj_size = sizeof(struct netlink_sock),
611 static int __netlink_create(struct net *net, struct socket *sock,
612 struct mutex *cb_mutex, int protocol,
616 struct netlink_sock *nlk;
618 sock->ops = &netlink_ops;
620 sk = sk_alloc(net, PF_NETLINK, GFP_KERNEL, &netlink_proto, kern);
624 sock_init_data(sock, sk);
628 nlk->cb_mutex = cb_mutex;
630 nlk->cb_mutex = &nlk->cb_def_mutex;
631 mutex_init(nlk->cb_mutex);
632 lockdep_set_class_and_name(nlk->cb_mutex,
633 nlk_cb_mutex_keys + protocol,
634 nlk_cb_mutex_key_strings[protocol]);
636 init_waitqueue_head(&nlk->wait);
638 sk->sk_destruct = netlink_sock_destruct;
639 sk->sk_protocol = protocol;
643 static int netlink_create(struct net *net, struct socket *sock, int protocol,
646 struct module *module = NULL;
647 struct mutex *cb_mutex;
648 struct netlink_sock *nlk;
649 int (*bind)(struct net *net, int group);
650 void (*unbind)(struct net *net, int group);
653 sock->state = SS_UNCONNECTED;
655 if (sock->type != SOCK_RAW && sock->type != SOCK_DGRAM)
656 return -ESOCKTNOSUPPORT;
658 if (protocol < 0 || protocol >= MAX_LINKS)
659 return -EPROTONOSUPPORT;
660 protocol = array_index_nospec(protocol, MAX_LINKS);
662 netlink_lock_table();
663 #ifdef CONFIG_MODULES
664 if (!nl_table[protocol].registered) {
665 netlink_unlock_table();
666 request_module("net-pf-%d-proto-%d", PF_NETLINK, protocol);
667 netlink_lock_table();
670 if (nl_table[protocol].registered &&
671 try_module_get(nl_table[protocol].module))
672 module = nl_table[protocol].module;
674 err = -EPROTONOSUPPORT;
675 cb_mutex = nl_table[protocol].cb_mutex;
676 bind = nl_table[protocol].bind;
677 unbind = nl_table[protocol].unbind;
678 netlink_unlock_table();
683 err = __netlink_create(net, sock, cb_mutex, protocol, kern);
688 sock_prot_inuse_add(net, &netlink_proto, 1);
691 nlk = nlk_sk(sock->sk);
692 nlk->module = module;
693 nlk->netlink_bind = bind;
694 nlk->netlink_unbind = unbind;
703 static void deferred_put_nlk_sk(struct rcu_head *head)
705 struct netlink_sock *nlk = container_of(head, struct netlink_sock, rcu);
706 struct sock *sk = &nlk->sk;
708 if (!atomic_dec_and_test(&sk->sk_refcnt))
711 if (nlk->cb_running && nlk->cb.done) {
712 INIT_WORK(&nlk->work, netlink_sock_destruct_work);
713 schedule_work(&nlk->work);
720 static int netlink_release(struct socket *sock)
722 struct sock *sk = sock->sk;
723 struct netlink_sock *nlk;
733 * OK. Socket is unlinked, any packets that arrive now
737 /* must not acquire netlink_table_lock in any way again before unbind
738 * and notifying genetlink is done as otherwise it might deadlock
740 if (nlk->netlink_unbind) {
743 for (i = 0; i < nlk->ngroups; i++)
744 if (test_bit(i, nlk->groups))
745 nlk->netlink_unbind(sock_net(sk), i + 1);
747 if (sk->sk_protocol == NETLINK_GENERIC &&
748 atomic_dec_return(&genl_sk_destructing_cnt) == 0)
749 wake_up(&genl_sk_destructing_waitq);
752 wake_up_interruptible_all(&nlk->wait);
754 skb_queue_purge(&sk->sk_write_queue);
756 if (nlk->portid && nlk->bound) {
757 struct netlink_notify n = {
759 .protocol = sk->sk_protocol,
760 .portid = nlk->portid,
762 atomic_notifier_call_chain(&netlink_chain,
763 NETLINK_URELEASE, &n);
766 module_put(nlk->module);
768 if (netlink_is_kernel(sk)) {
769 netlink_table_grab();
770 BUG_ON(nl_table[sk->sk_protocol].registered == 0);
771 if (--nl_table[sk->sk_protocol].registered == 0) {
772 struct listeners *old;
774 old = nl_deref_protected(nl_table[sk->sk_protocol].listeners);
775 RCU_INIT_POINTER(nl_table[sk->sk_protocol].listeners, NULL);
777 nl_table[sk->sk_protocol].module = NULL;
778 nl_table[sk->sk_protocol].bind = NULL;
779 nl_table[sk->sk_protocol].unbind = NULL;
780 nl_table[sk->sk_protocol].flags = 0;
781 nl_table[sk->sk_protocol].registered = 0;
783 netlink_table_ungrab();
790 sock_prot_inuse_add(sock_net(sk), &netlink_proto, -1);
792 call_rcu(&nlk->rcu, deferred_put_nlk_sk);
796 static int netlink_autobind(struct socket *sock)
798 struct sock *sk = sock->sk;
799 struct net *net = sock_net(sk);
800 struct netlink_table *table = &nl_table[sk->sk_protocol];
801 s32 portid = task_tgid_vnr(current);
809 ok = !__netlink_lookup(table, portid, net);
812 /* Bind collision, search negative portid values. */
814 /* rover will be in range [S32_MIN, -4097] */
815 rover = S32_MIN + prandom_u32_max(-4096 - S32_MIN);
816 else if (rover >= -4096)
822 err = netlink_insert(sk, portid);
823 if (err == -EADDRINUSE)
826 /* If 2 threads race to autobind, that is fine. */
834 * __netlink_ns_capable - General netlink message capability test
835 * @nsp: NETLINK_CB of the socket buffer holding a netlink command from userspace.
836 * @user_ns: The user namespace of the capability to use
837 * @cap: The capability to use
839 * Test to see if the opener of the socket we received the message
840 * from had when the netlink socket was created and the sender of the
841 * message has has the capability @cap in the user namespace @user_ns.
843 bool __netlink_ns_capable(const struct netlink_skb_parms *nsp,
844 struct user_namespace *user_ns, int cap)
846 return ((nsp->flags & NETLINK_SKB_DST) ||
847 file_ns_capable(nsp->sk->sk_socket->file, user_ns, cap)) &&
848 ns_capable(user_ns, cap);
850 EXPORT_SYMBOL(__netlink_ns_capable);
853 * netlink_ns_capable - General netlink message capability test
854 * @skb: socket buffer holding a netlink command from userspace
855 * @user_ns: The user namespace of the capability to use
856 * @cap: The capability to use
858 * Test to see if the opener of the socket we received the message
859 * from had when the netlink socket was created and the sender of the
860 * message has has the capability @cap in the user namespace @user_ns.
862 bool netlink_ns_capable(const struct sk_buff *skb,
863 struct user_namespace *user_ns, int cap)
865 return __netlink_ns_capable(&NETLINK_CB(skb), user_ns, cap);
867 EXPORT_SYMBOL(netlink_ns_capable);
870 * netlink_capable - Netlink global message capability test
871 * @skb: socket buffer holding a netlink command from userspace
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 has the capability @cap in all user namespaces.
878 bool netlink_capable(const struct sk_buff *skb, int cap)
880 return netlink_ns_capable(skb, &init_user_ns, cap);
882 EXPORT_SYMBOL(netlink_capable);
885 * netlink_net_capable - Netlink network namespace message capability test
886 * @skb: socket buffer holding a netlink command from userspace
887 * @cap: The capability to use
889 * Test to see if the opener of the socket we received the message
890 * from had when the netlink socket was created and the sender of the
891 * message has has the capability @cap over the network namespace of
892 * the socket we received the message from.
894 bool netlink_net_capable(const struct sk_buff *skb, int cap)
896 return netlink_ns_capable(skb, sock_net(skb->sk)->user_ns, cap);
898 EXPORT_SYMBOL(netlink_net_capable);
900 static inline int netlink_allowed(const struct socket *sock, unsigned int flag)
902 return (nl_table[sock->sk->sk_protocol].flags & flag) ||
903 ns_capable(sock_net(sock->sk)->user_ns, CAP_NET_ADMIN);
907 netlink_update_subscriptions(struct sock *sk, unsigned int subscriptions)
909 struct netlink_sock *nlk = nlk_sk(sk);
911 if (nlk->subscriptions && !subscriptions)
912 __sk_del_bind_node(sk);
913 else if (!nlk->subscriptions && subscriptions)
914 sk_add_bind_node(sk, &nl_table[sk->sk_protocol].mc_list);
915 nlk->subscriptions = subscriptions;
918 static int netlink_realloc_groups(struct sock *sk)
920 struct netlink_sock *nlk = nlk_sk(sk);
922 unsigned long *new_groups;
925 netlink_table_grab();
927 groups = nl_table[sk->sk_protocol].groups;
928 if (!nl_table[sk->sk_protocol].registered) {
933 if (nlk->ngroups >= groups)
936 new_groups = krealloc(nlk->groups, NLGRPSZ(groups), GFP_ATOMIC);
937 if (new_groups == NULL) {
941 memset((char *)new_groups + NLGRPSZ(nlk->ngroups), 0,
942 NLGRPSZ(groups) - NLGRPSZ(nlk->ngroups));
944 nlk->groups = new_groups;
945 nlk->ngroups = groups;
947 netlink_table_ungrab();
951 static void netlink_undo_bind(int group, long unsigned int groups,
954 struct netlink_sock *nlk = nlk_sk(sk);
957 if (!nlk->netlink_unbind)
960 for (undo = 0; undo < group; undo++)
961 if (test_bit(undo, &groups))
962 nlk->netlink_unbind(sock_net(sk), undo + 1);
965 static int netlink_bind(struct socket *sock, struct sockaddr *addr,
968 struct sock *sk = sock->sk;
969 struct net *net = sock_net(sk);
970 struct netlink_sock *nlk = nlk_sk(sk);
971 struct sockaddr_nl *nladdr = (struct sockaddr_nl *)addr;
973 long unsigned int groups = nladdr->nl_groups;
976 if (addr_len < sizeof(struct sockaddr_nl))
979 if (nladdr->nl_family != AF_NETLINK)
982 /* Only superuser is allowed to listen multicasts */
984 if (!netlink_allowed(sock, NL_CFG_F_NONROOT_RECV))
986 err = netlink_realloc_groups(sk);
991 if (nlk->ngroups == 0)
993 else if (nlk->ngroups < 8*sizeof(groups))
994 groups &= (1UL << nlk->ngroups) - 1;
998 /* Ensure nlk->portid is up-to-date. */
1001 if (nladdr->nl_pid != nlk->portid)
1005 if (nlk->netlink_bind && groups) {
1008 /* nl_groups is a u32, so cap the maximum groups we can bind */
1009 for (group = 0; group < BITS_PER_TYPE(u32); group++) {
1010 if (!test_bit(group, &groups))
1012 err = nlk->netlink_bind(net, group + 1);
1015 netlink_undo_bind(group, groups, sk);
1020 /* No need for barriers here as we return to user-space without
1021 * using any of the bound attributes.
1024 err = nladdr->nl_pid ?
1025 netlink_insert(sk, nladdr->nl_pid) :
1026 netlink_autobind(sock);
1028 netlink_undo_bind(BITS_PER_TYPE(u32), groups, sk);
1033 if (!groups && (nlk->groups == NULL || !(u32)nlk->groups[0]))
1036 netlink_table_grab();
1037 netlink_update_subscriptions(sk, nlk->subscriptions +
1039 hweight32(nlk->groups[0]));
1040 nlk->groups[0] = (nlk->groups[0] & ~0xffffffffUL) | groups;
1041 netlink_update_listeners(sk);
1042 netlink_table_ungrab();
1047 static int netlink_connect(struct socket *sock, struct sockaddr *addr,
1048 int alen, int flags)
1051 struct sock *sk = sock->sk;
1052 struct netlink_sock *nlk = nlk_sk(sk);
1053 struct sockaddr_nl *nladdr = (struct sockaddr_nl *)addr;
1055 if (alen < sizeof(addr->sa_family))
1058 if (addr->sa_family == AF_UNSPEC) {
1059 sk->sk_state = NETLINK_UNCONNECTED;
1060 nlk->dst_portid = 0;
1064 if (addr->sa_family != AF_NETLINK)
1067 if (alen < sizeof(struct sockaddr_nl))
1070 if ((nladdr->nl_groups || nladdr->nl_pid) &&
1071 !netlink_allowed(sock, NL_CFG_F_NONROOT_SEND))
1074 /* No need for barriers here as we return to user-space without
1075 * using any of the bound attributes.
1078 err = netlink_autobind(sock);
1081 sk->sk_state = NETLINK_CONNECTED;
1082 nlk->dst_portid = nladdr->nl_pid;
1083 nlk->dst_group = ffs(nladdr->nl_groups);
1089 static int netlink_getname(struct socket *sock, struct sockaddr *addr,
1090 int *addr_len, int peer)
1092 struct sock *sk = sock->sk;
1093 struct netlink_sock *nlk = nlk_sk(sk);
1094 DECLARE_SOCKADDR(struct sockaddr_nl *, nladdr, addr);
1096 nladdr->nl_family = AF_NETLINK;
1098 *addr_len = sizeof(*nladdr);
1101 nladdr->nl_pid = nlk->dst_portid;
1102 nladdr->nl_groups = netlink_group_mask(nlk->dst_group);
1104 nladdr->nl_pid = nlk->portid;
1105 nladdr->nl_groups = nlk->groups ? nlk->groups[0] : 0;
1110 static struct sock *netlink_getsockbyportid(struct sock *ssk, u32 portid)
1113 struct netlink_sock *nlk;
1115 sock = netlink_lookup(sock_net(ssk), ssk->sk_protocol, portid);
1117 return ERR_PTR(-ECONNREFUSED);
1119 /* Don't bother queuing skb if kernel socket has no input function */
1121 if (sock->sk_state == NETLINK_CONNECTED &&
1122 nlk->dst_portid != nlk_sk(ssk)->portid) {
1124 return ERR_PTR(-ECONNREFUSED);
1129 struct sock *netlink_getsockbyfilp(struct file *filp)
1131 struct inode *inode = file_inode(filp);
1134 if (!S_ISSOCK(inode->i_mode))
1135 return ERR_PTR(-ENOTSOCK);
1137 sock = SOCKET_I(inode)->sk;
1138 if (sock->sk_family != AF_NETLINK)
1139 return ERR_PTR(-EINVAL);
1145 static struct sk_buff *netlink_alloc_large_skb(unsigned int size,
1148 struct sk_buff *skb;
1151 if (size <= NLMSG_GOODSIZE || broadcast)
1152 return alloc_skb(size, GFP_KERNEL);
1154 size = SKB_DATA_ALIGN(size) +
1155 SKB_DATA_ALIGN(sizeof(struct skb_shared_info));
1157 data = vmalloc(size);
1161 skb = __build_skb(data, size);
1165 skb->destructor = netlink_skb_destructor;
1171 * Attach a skb to a netlink socket.
1172 * The caller must hold a reference to the destination socket. On error, the
1173 * reference is dropped. The skb is not send to the destination, just all
1174 * all error checks are performed and memory in the queue is reserved.
1176 * < 0: error. skb freed, reference to sock dropped.
1178 * 1: repeat lookup - reference dropped while waiting for socket memory.
1180 int netlink_attachskb(struct sock *sk, struct sk_buff *skb,
1181 long *timeo, struct sock *ssk)
1183 struct netlink_sock *nlk;
1187 if ((atomic_read(&sk->sk_rmem_alloc) > sk->sk_rcvbuf ||
1188 test_bit(NETLINK_S_CONGESTED, &nlk->state))) {
1189 DECLARE_WAITQUEUE(wait, current);
1191 if (!ssk || netlink_is_kernel(ssk))
1192 netlink_overrun(sk);
1198 __set_current_state(TASK_INTERRUPTIBLE);
1199 add_wait_queue(&nlk->wait, &wait);
1201 if ((atomic_read(&sk->sk_rmem_alloc) > sk->sk_rcvbuf ||
1202 test_bit(NETLINK_S_CONGESTED, &nlk->state)) &&
1203 !sock_flag(sk, SOCK_DEAD))
1204 *timeo = schedule_timeout(*timeo);
1206 __set_current_state(TASK_RUNNING);
1207 remove_wait_queue(&nlk->wait, &wait);
1210 if (signal_pending(current)) {
1212 return sock_intr_errno(*timeo);
1216 netlink_skb_set_owner_r(skb, sk);
1220 static int __netlink_sendskb(struct sock *sk, struct sk_buff *skb)
1224 netlink_deliver_tap(skb);
1226 skb_queue_tail(&sk->sk_receive_queue, skb);
1227 sk->sk_data_ready(sk);
1231 int netlink_sendskb(struct sock *sk, struct sk_buff *skb)
1233 int len = __netlink_sendskb(sk, skb);
1239 void netlink_detachskb(struct sock *sk, struct sk_buff *skb)
1245 static struct sk_buff *netlink_trim(struct sk_buff *skb, gfp_t allocation)
1249 WARN_ON(skb->sk != NULL);
1250 delta = skb->end - skb->tail;
1251 if (is_vmalloc_addr(skb->head) || delta * 2 < skb->truesize)
1254 if (skb_shared(skb)) {
1255 struct sk_buff *nskb = skb_clone(skb, allocation);
1262 if (!pskb_expand_head(skb, 0, -delta, allocation))
1263 skb->truesize -= delta;
1268 static int netlink_unicast_kernel(struct sock *sk, struct sk_buff *skb,
1272 struct netlink_sock *nlk = nlk_sk(sk);
1274 ret = -ECONNREFUSED;
1275 if (nlk->netlink_rcv != NULL) {
1277 netlink_skb_set_owner_r(skb, sk);
1278 NETLINK_CB(skb).sk = ssk;
1279 netlink_deliver_tap_kernel(sk, ssk, skb);
1280 nlk->netlink_rcv(skb);
1289 int netlink_unicast(struct sock *ssk, struct sk_buff *skb,
1290 u32 portid, int nonblock)
1296 skb = netlink_trim(skb, gfp_any());
1298 timeo = sock_sndtimeo(ssk, nonblock);
1300 sk = netlink_getsockbyportid(ssk, portid);
1305 if (netlink_is_kernel(sk))
1306 return netlink_unicast_kernel(sk, skb, ssk);
1308 if (sk_filter(sk, skb)) {
1315 err = netlink_attachskb(sk, skb, &timeo, ssk);
1321 return netlink_sendskb(sk, skb);
1323 EXPORT_SYMBOL(netlink_unicast);
1325 struct sk_buff *__netlink_alloc_skb(struct sock *ssk, unsigned int size,
1326 unsigned int ldiff, u32 dst_portid,
1329 return alloc_skb(size, gfp_mask);
1331 EXPORT_SYMBOL_GPL(__netlink_alloc_skb);
1333 int netlink_has_listeners(struct sock *sk, unsigned int group)
1336 struct listeners *listeners;
1338 BUG_ON(!netlink_is_kernel(sk));
1341 listeners = rcu_dereference(nl_table[sk->sk_protocol].listeners);
1343 if (listeners && group - 1 < nl_table[sk->sk_protocol].groups)
1344 res = test_bit(group - 1, listeners->masks);
1350 EXPORT_SYMBOL_GPL(netlink_has_listeners);
1352 static int netlink_broadcast_deliver(struct sock *sk, struct sk_buff *skb)
1354 struct netlink_sock *nlk = nlk_sk(sk);
1356 if (atomic_read(&sk->sk_rmem_alloc) <= sk->sk_rcvbuf &&
1357 !test_bit(NETLINK_S_CONGESTED, &nlk->state)) {
1358 netlink_skb_set_owner_r(skb, sk);
1359 __netlink_sendskb(sk, skb);
1360 return atomic_read(&sk->sk_rmem_alloc) > (sk->sk_rcvbuf >> 1);
1365 struct netlink_broadcast_data {
1366 struct sock *exclude_sk;
1371 int delivery_failure;
1375 struct sk_buff *skb, *skb2;
1376 int (*tx_filter)(struct sock *dsk, struct sk_buff *skb, void *data);
1380 static void do_one_broadcast(struct sock *sk,
1381 struct netlink_broadcast_data *p)
1383 struct netlink_sock *nlk = nlk_sk(sk);
1386 if (p->exclude_sk == sk)
1389 if (nlk->portid == p->portid || p->group - 1 >= nlk->ngroups ||
1390 !test_bit(p->group - 1, nlk->groups))
1393 if (!net_eq(sock_net(sk), p->net)) {
1394 if (!(nlk->flags & NETLINK_F_LISTEN_ALL_NSID))
1397 if (!peernet_has_id(sock_net(sk), p->net))
1400 if (!file_ns_capable(sk->sk_socket->file, p->net->user_ns,
1406 netlink_overrun(sk);
1411 if (p->skb2 == NULL) {
1412 if (skb_shared(p->skb)) {
1413 p->skb2 = skb_clone(p->skb, p->allocation);
1415 p->skb2 = skb_get(p->skb);
1417 * skb ownership may have been set when
1418 * delivered to a previous socket.
1420 skb_orphan(p->skb2);
1423 if (p->skb2 == NULL) {
1424 netlink_overrun(sk);
1425 /* Clone failed. Notify ALL listeners. */
1427 if (nlk->flags & NETLINK_F_BROADCAST_SEND_ERROR)
1428 p->delivery_failure = 1;
1431 if (p->tx_filter && p->tx_filter(sk, p->skb2, p->tx_data)) {
1436 if (sk_filter(sk, p->skb2)) {
1441 NETLINK_CB(p->skb2).nsid = peernet2id(sock_net(sk), p->net);
1442 NETLINK_CB(p->skb2).nsid_is_set = true;
1443 val = netlink_broadcast_deliver(sk, p->skb2);
1445 netlink_overrun(sk);
1446 if (nlk->flags & NETLINK_F_BROADCAST_SEND_ERROR)
1447 p->delivery_failure = 1;
1449 p->congested |= val;
1457 int netlink_broadcast_filtered(struct sock *ssk, struct sk_buff *skb, u32 portid,
1458 u32 group, gfp_t allocation,
1459 int (*filter)(struct sock *dsk, struct sk_buff *skb, void *data),
1462 struct net *net = sock_net(ssk);
1463 struct netlink_broadcast_data info;
1466 skb = netlink_trim(skb, allocation);
1468 info.exclude_sk = ssk;
1470 info.portid = portid;
1473 info.delivery_failure = 0;
1476 info.allocation = allocation;
1479 info.tx_filter = filter;
1480 info.tx_data = filter_data;
1482 /* While we sleep in clone, do not allow to change socket list */
1484 netlink_lock_table();
1486 sk_for_each_bound(sk, &nl_table[ssk->sk_protocol].mc_list)
1487 do_one_broadcast(sk, &info);
1491 netlink_unlock_table();
1493 if (info.delivery_failure) {
1494 kfree_skb(info.skb2);
1497 consume_skb(info.skb2);
1499 if (info.delivered) {
1500 if (info.congested && gfpflags_allow_blocking(allocation))
1506 EXPORT_SYMBOL(netlink_broadcast_filtered);
1508 int netlink_broadcast(struct sock *ssk, struct sk_buff *skb, u32 portid,
1509 u32 group, gfp_t allocation)
1511 return netlink_broadcast_filtered(ssk, skb, portid, group, allocation,
1514 EXPORT_SYMBOL(netlink_broadcast);
1516 struct netlink_set_err_data {
1517 struct sock *exclude_sk;
1523 static int do_one_set_err(struct sock *sk, struct netlink_set_err_data *p)
1525 struct netlink_sock *nlk = nlk_sk(sk);
1528 if (sk == p->exclude_sk)
1531 if (!net_eq(sock_net(sk), sock_net(p->exclude_sk)))
1534 if (nlk->portid == p->portid || p->group - 1 >= nlk->ngroups ||
1535 !test_bit(p->group - 1, nlk->groups))
1538 if (p->code == ENOBUFS && nlk->flags & NETLINK_F_RECV_NO_ENOBUFS) {
1543 sk->sk_err = p->code;
1544 sk->sk_error_report(sk);
1550 * netlink_set_err - report error to broadcast listeners
1551 * @ssk: the kernel netlink socket, as returned by netlink_kernel_create()
1552 * @portid: the PORTID of a process that we want to skip (if any)
1553 * @group: the broadcast group that will notice the error
1554 * @code: error code, must be negative (as usual in kernelspace)
1556 * This function returns the number of broadcast listeners that have set the
1557 * NETLINK_NO_ENOBUFS socket option.
1559 int netlink_set_err(struct sock *ssk, u32 portid, u32 group, int code)
1561 struct netlink_set_err_data info;
1565 info.exclude_sk = ssk;
1566 info.portid = portid;
1568 /* sk->sk_err wants a positive error value */
1571 read_lock(&nl_table_lock);
1573 sk_for_each_bound(sk, &nl_table[ssk->sk_protocol].mc_list)
1574 ret += do_one_set_err(sk, &info);
1576 read_unlock(&nl_table_lock);
1579 EXPORT_SYMBOL(netlink_set_err);
1581 /* must be called with netlink table grabbed */
1582 static void netlink_update_socket_mc(struct netlink_sock *nlk,
1586 int old, new = !!is_new, subscriptions;
1588 old = test_bit(group - 1, nlk->groups);
1589 subscriptions = nlk->subscriptions - old + new;
1591 __set_bit(group - 1, nlk->groups);
1593 __clear_bit(group - 1, nlk->groups);
1594 netlink_update_subscriptions(&nlk->sk, subscriptions);
1595 netlink_update_listeners(&nlk->sk);
1598 static int netlink_setsockopt(struct socket *sock, int level, int optname,
1599 char __user *optval, unsigned int optlen)
1601 struct sock *sk = sock->sk;
1602 struct netlink_sock *nlk = nlk_sk(sk);
1603 unsigned int val = 0;
1606 if (level != SOL_NETLINK)
1607 return -ENOPROTOOPT;
1609 if (optlen >= sizeof(int) &&
1610 get_user(val, (unsigned int __user *)optval))
1614 case NETLINK_PKTINFO:
1616 nlk->flags |= NETLINK_F_RECV_PKTINFO;
1618 nlk->flags &= ~NETLINK_F_RECV_PKTINFO;
1621 case NETLINK_ADD_MEMBERSHIP:
1622 case NETLINK_DROP_MEMBERSHIP: {
1623 if (!netlink_allowed(sock, NL_CFG_F_NONROOT_RECV))
1625 err = netlink_realloc_groups(sk);
1628 if (!val || val - 1 >= nlk->ngroups)
1630 if (optname == NETLINK_ADD_MEMBERSHIP && nlk->netlink_bind) {
1631 err = nlk->netlink_bind(sock_net(sk), val);
1635 netlink_table_grab();
1636 netlink_update_socket_mc(nlk, val,
1637 optname == NETLINK_ADD_MEMBERSHIP);
1638 netlink_table_ungrab();
1639 if (optname == NETLINK_DROP_MEMBERSHIP && nlk->netlink_unbind)
1640 nlk->netlink_unbind(sock_net(sk), val);
1645 case NETLINK_BROADCAST_ERROR:
1647 nlk->flags |= NETLINK_F_BROADCAST_SEND_ERROR;
1649 nlk->flags &= ~NETLINK_F_BROADCAST_SEND_ERROR;
1652 case NETLINK_NO_ENOBUFS:
1654 nlk->flags |= NETLINK_F_RECV_NO_ENOBUFS;
1655 clear_bit(NETLINK_S_CONGESTED, &nlk->state);
1656 wake_up_interruptible(&nlk->wait);
1658 nlk->flags &= ~NETLINK_F_RECV_NO_ENOBUFS;
1662 case NETLINK_LISTEN_ALL_NSID:
1663 if (!ns_capable(sock_net(sk)->user_ns, CAP_NET_BROADCAST))
1667 nlk->flags |= NETLINK_F_LISTEN_ALL_NSID;
1669 nlk->flags &= ~NETLINK_F_LISTEN_ALL_NSID;
1672 case NETLINK_CAP_ACK:
1674 nlk->flags |= NETLINK_F_CAP_ACK;
1676 nlk->flags &= ~NETLINK_F_CAP_ACK;
1685 static int netlink_getsockopt(struct socket *sock, int level, int optname,
1686 char __user *optval, int __user *optlen)
1688 struct sock *sk = sock->sk;
1689 struct netlink_sock *nlk = nlk_sk(sk);
1692 if (level != SOL_NETLINK)
1693 return -ENOPROTOOPT;
1695 if (get_user(len, optlen))
1701 case NETLINK_PKTINFO:
1702 if (len < sizeof(int))
1705 val = nlk->flags & NETLINK_F_RECV_PKTINFO ? 1 : 0;
1706 if (put_user(len, optlen) ||
1707 put_user(val, optval))
1711 case NETLINK_BROADCAST_ERROR:
1712 if (len < sizeof(int))
1715 val = nlk->flags & NETLINK_F_BROADCAST_SEND_ERROR ? 1 : 0;
1716 if (put_user(len, optlen) ||
1717 put_user(val, optval))
1721 case NETLINK_NO_ENOBUFS:
1722 if (len < sizeof(int))
1725 val = nlk->flags & NETLINK_F_RECV_NO_ENOBUFS ? 1 : 0;
1726 if (put_user(len, optlen) ||
1727 put_user(val, optval))
1731 case NETLINK_LIST_MEMBERSHIPS: {
1732 int pos, idx, shift;
1735 netlink_lock_table();
1736 for (pos = 0; pos * 8 < nlk->ngroups; pos += sizeof(u32)) {
1737 if (len - pos < sizeof(u32))
1740 idx = pos / sizeof(unsigned long);
1741 shift = (pos % sizeof(unsigned long)) * 8;
1742 if (put_user((u32)(nlk->groups[idx] >> shift),
1743 (u32 __user *)(optval + pos))) {
1748 if (put_user(ALIGN(nlk->ngroups / 8, sizeof(u32)), optlen))
1750 netlink_unlock_table();
1753 case NETLINK_CAP_ACK:
1754 if (len < sizeof(int))
1757 val = nlk->flags & NETLINK_F_CAP_ACK ? 1 : 0;
1758 if (put_user(len, optlen) ||
1759 put_user(val, optval))
1769 static void netlink_cmsg_recv_pktinfo(struct msghdr *msg, struct sk_buff *skb)
1771 struct nl_pktinfo info;
1773 info.group = NETLINK_CB(skb).dst_group;
1774 put_cmsg(msg, SOL_NETLINK, NETLINK_PKTINFO, sizeof(info), &info);
1777 static void netlink_cmsg_listen_all_nsid(struct sock *sk, struct msghdr *msg,
1778 struct sk_buff *skb)
1780 if (!NETLINK_CB(skb).nsid_is_set)
1783 put_cmsg(msg, SOL_NETLINK, NETLINK_LISTEN_ALL_NSID, sizeof(int),
1784 &NETLINK_CB(skb).nsid);
1787 static int netlink_sendmsg(struct socket *sock, struct msghdr *msg, size_t len)
1789 struct sock *sk = sock->sk;
1790 struct netlink_sock *nlk = nlk_sk(sk);
1791 DECLARE_SOCKADDR(struct sockaddr_nl *, addr, msg->msg_name);
1794 struct sk_buff *skb;
1796 struct scm_cookie scm;
1797 u32 netlink_skb_flags = 0;
1799 if (msg->msg_flags&MSG_OOB)
1802 err = scm_send(sock, msg, &scm, true);
1806 if (msg->msg_namelen) {
1808 if (msg->msg_namelen < sizeof(struct sockaddr_nl))
1810 if (addr->nl_family != AF_NETLINK)
1812 dst_portid = addr->nl_pid;
1813 dst_group = ffs(addr->nl_groups);
1815 if ((dst_group || dst_portid) &&
1816 !netlink_allowed(sock, NL_CFG_F_NONROOT_SEND))
1818 netlink_skb_flags |= NETLINK_SKB_DST;
1820 dst_portid = nlk->dst_portid;
1821 dst_group = nlk->dst_group;
1825 err = netlink_autobind(sock);
1829 /* Ensure nlk is hashed and visible. */
1834 if (len > sk->sk_sndbuf - 32)
1837 skb = netlink_alloc_large_skb(len, dst_group);
1841 NETLINK_CB(skb).portid = nlk->portid;
1842 NETLINK_CB(skb).dst_group = dst_group;
1843 NETLINK_CB(skb).creds = scm.creds;
1844 NETLINK_CB(skb).flags = netlink_skb_flags;
1847 if (memcpy_from_msg(skb_put(skb, len), msg, len)) {
1852 err = security_netlink_send(sk, skb);
1859 atomic_inc(&skb->users);
1860 netlink_broadcast(sk, skb, dst_portid, dst_group, GFP_KERNEL);
1862 err = netlink_unicast(sk, skb, dst_portid, msg->msg_flags&MSG_DONTWAIT);
1869 static int netlink_recvmsg(struct socket *sock, struct msghdr *msg, size_t len,
1872 struct scm_cookie scm;
1873 struct sock *sk = sock->sk;
1874 struct netlink_sock *nlk = nlk_sk(sk);
1875 int noblock = flags&MSG_DONTWAIT;
1877 struct sk_buff *skb, *data_skb;
1885 skb = skb_recv_datagram(sk, flags, noblock, &err);
1891 #ifdef CONFIG_COMPAT_NETLINK_MESSAGES
1892 if (unlikely(skb_shinfo(skb)->frag_list)) {
1894 * If this skb has a frag_list, then here that means that we
1895 * will have to use the frag_list skb's data for compat tasks
1896 * and the regular skb's data for normal (non-compat) tasks.
1898 * If we need to send the compat skb, assign it to the
1899 * 'data_skb' variable so that it will be used below for data
1900 * copying. We keep 'skb' for everything else, including
1901 * freeing both later.
1903 if (flags & MSG_CMSG_COMPAT)
1904 data_skb = skb_shinfo(skb)->frag_list;
1908 /* Record the max length of recvmsg() calls for future allocations */
1909 nlk->max_recvmsg_len = max(nlk->max_recvmsg_len, len);
1910 nlk->max_recvmsg_len = min_t(size_t, nlk->max_recvmsg_len,
1911 SKB_WITH_OVERHEAD(32768));
1913 copied = data_skb->len;
1915 msg->msg_flags |= MSG_TRUNC;
1919 skb_reset_transport_header(data_skb);
1920 err = skb_copy_datagram_msg(data_skb, 0, msg, copied);
1922 if (msg->msg_name) {
1923 DECLARE_SOCKADDR(struct sockaddr_nl *, addr, msg->msg_name);
1924 addr->nl_family = AF_NETLINK;
1926 addr->nl_pid = NETLINK_CB(skb).portid;
1927 addr->nl_groups = netlink_group_mask(NETLINK_CB(skb).dst_group);
1928 msg->msg_namelen = sizeof(*addr);
1931 if (nlk->flags & NETLINK_F_RECV_PKTINFO)
1932 netlink_cmsg_recv_pktinfo(msg, skb);
1933 if (nlk->flags & NETLINK_F_LISTEN_ALL_NSID)
1934 netlink_cmsg_listen_all_nsid(sk, msg, skb);
1936 memset(&scm, 0, sizeof(scm));
1937 scm.creds = *NETLINK_CREDS(skb);
1938 if (flags & MSG_TRUNC)
1939 copied = data_skb->len;
1941 skb_free_datagram(sk, skb);
1943 if (nlk->cb_running &&
1944 atomic_read(&sk->sk_rmem_alloc) <= sk->sk_rcvbuf / 2) {
1945 ret = netlink_dump(sk);
1948 sk->sk_error_report(sk);
1952 scm_recv(sock, msg, &scm, flags);
1954 netlink_rcv_wake(sk);
1955 return err ? : copied;
1958 static void netlink_data_ready(struct sock *sk)
1964 * We export these functions to other modules. They provide a
1965 * complete set of kernel non-blocking support for message
1970 __netlink_kernel_create(struct net *net, int unit, struct module *module,
1971 struct netlink_kernel_cfg *cfg)
1973 struct socket *sock;
1975 struct netlink_sock *nlk;
1976 struct listeners *listeners = NULL;
1977 struct mutex *cb_mutex = cfg ? cfg->cb_mutex : NULL;
1978 unsigned int groups;
1982 if (unit < 0 || unit >= MAX_LINKS)
1985 if (sock_create_lite(PF_NETLINK, SOCK_DGRAM, unit, &sock))
1988 if (__netlink_create(net, sock, cb_mutex, unit, 1) < 0)
1989 goto out_sock_release_nosk;
1993 if (!cfg || cfg->groups < 32)
1996 groups = cfg->groups;
1998 listeners = kzalloc(sizeof(*listeners) + NLGRPSZ(groups), GFP_KERNEL);
2000 goto out_sock_release;
2002 sk->sk_data_ready = netlink_data_ready;
2003 if (cfg && cfg->input)
2004 nlk_sk(sk)->netlink_rcv = cfg->input;
2006 if (netlink_insert(sk, 0))
2007 goto out_sock_release;
2010 nlk->flags |= NETLINK_F_KERNEL_SOCKET;
2012 netlink_table_grab();
2013 if (!nl_table[unit].registered) {
2014 nl_table[unit].groups = groups;
2015 rcu_assign_pointer(nl_table[unit].listeners, listeners);
2016 nl_table[unit].cb_mutex = cb_mutex;
2017 nl_table[unit].module = module;
2019 nl_table[unit].bind = cfg->bind;
2020 nl_table[unit].unbind = cfg->unbind;
2021 nl_table[unit].flags = cfg->flags;
2023 nl_table[unit].compare = cfg->compare;
2025 nl_table[unit].registered = 1;
2028 nl_table[unit].registered++;
2030 netlink_table_ungrab();
2035 netlink_kernel_release(sk);
2038 out_sock_release_nosk:
2042 EXPORT_SYMBOL(__netlink_kernel_create);
2045 netlink_kernel_release(struct sock *sk)
2047 if (sk == NULL || sk->sk_socket == NULL)
2050 sock_release(sk->sk_socket);
2052 EXPORT_SYMBOL(netlink_kernel_release);
2054 int __netlink_change_ngroups(struct sock *sk, unsigned int groups)
2056 struct listeners *new, *old;
2057 struct netlink_table *tbl = &nl_table[sk->sk_protocol];
2062 if (NLGRPSZ(tbl->groups) < NLGRPSZ(groups)) {
2063 new = kzalloc(sizeof(*new) + NLGRPSZ(groups), GFP_ATOMIC);
2066 old = nl_deref_protected(tbl->listeners);
2067 memcpy(new->masks, old->masks, NLGRPSZ(tbl->groups));
2068 rcu_assign_pointer(tbl->listeners, new);
2070 kfree_rcu(old, rcu);
2072 tbl->groups = groups;
2078 * netlink_change_ngroups - change number of multicast groups
2080 * This changes the number of multicast groups that are available
2081 * on a certain netlink family. Note that it is not possible to
2082 * change the number of groups to below 32. Also note that it does
2083 * not implicitly call netlink_clear_multicast_users() when the
2084 * number of groups is reduced.
2086 * @sk: The kernel netlink socket, as returned by netlink_kernel_create().
2087 * @groups: The new number of groups.
2089 int netlink_change_ngroups(struct sock *sk, unsigned int groups)
2093 netlink_table_grab();
2094 err = __netlink_change_ngroups(sk, groups);
2095 netlink_table_ungrab();
2100 void __netlink_clear_multicast_users(struct sock *ksk, unsigned int group)
2103 struct netlink_table *tbl = &nl_table[ksk->sk_protocol];
2105 sk_for_each_bound(sk, &tbl->mc_list)
2106 netlink_update_socket_mc(nlk_sk(sk), group, 0);
2110 __nlmsg_put(struct sk_buff *skb, u32 portid, u32 seq, int type, int len, int flags)
2112 struct nlmsghdr *nlh;
2113 int size = nlmsg_msg_size(len);
2115 nlh = (struct nlmsghdr *)skb_put(skb, NLMSG_ALIGN(size));
2116 nlh->nlmsg_type = type;
2117 nlh->nlmsg_len = size;
2118 nlh->nlmsg_flags = flags;
2119 nlh->nlmsg_pid = portid;
2120 nlh->nlmsg_seq = seq;
2121 if (!__builtin_constant_p(size) || NLMSG_ALIGN(size) - size != 0)
2122 memset(nlmsg_data(nlh) + len, 0, NLMSG_ALIGN(size) - size);
2125 EXPORT_SYMBOL(__nlmsg_put);
2128 * It looks a bit ugly.
2129 * It would be better to create kernel thread.
2132 static int netlink_dump(struct sock *sk)
2134 struct netlink_sock *nlk = nlk_sk(sk);
2135 struct netlink_callback *cb;
2136 struct sk_buff *skb = NULL;
2137 struct nlmsghdr *nlh;
2138 struct module *module;
2143 mutex_lock(nlk->cb_mutex);
2144 if (!nlk->cb_running) {
2149 if (atomic_read(&sk->sk_rmem_alloc) >= sk->sk_rcvbuf)
2152 /* NLMSG_GOODSIZE is small to avoid high order allocations being
2153 * required, but it makes sense to _attempt_ a 16K bytes allocation
2154 * to reduce number of system calls on dump operations, if user
2155 * ever provided a big enough buffer.
2158 alloc_min_size = max_t(int, cb->min_dump_alloc, NLMSG_GOODSIZE);
2160 if (alloc_min_size < nlk->max_recvmsg_len) {
2161 alloc_size = nlk->max_recvmsg_len;
2162 skb = netlink_alloc_skb(sk, alloc_size, nlk->portid,
2163 (GFP_KERNEL & ~__GFP_DIRECT_RECLAIM) |
2164 __GFP_NOWARN | __GFP_NORETRY);
2167 alloc_size = alloc_min_size;
2168 skb = netlink_alloc_skb(sk, alloc_size, nlk->portid,
2174 /* Trim skb to allocated size. User is expected to provide buffer as
2175 * large as max(min_dump_alloc, 16KiB (mac_recvmsg_len capped at
2176 * netlink_recvmsg())). dump will pack as many smaller messages as
2177 * could fit within the allocated skb. skb is typically allocated
2178 * with larger space than required (could be as much as near 2x the
2179 * requested size with align to next power of 2 approach). Allowing
2180 * dump to use the excess space makes it difficult for a user to have a
2181 * reasonable static buffer based on the expected largest dump of a
2182 * single netdev. The outcome is MSG_TRUNC error.
2184 skb_reserve(skb, skb_tailroom(skb) - alloc_size);
2185 netlink_skb_set_owner_r(skb, sk);
2187 if (nlk->dump_done_errno > 0)
2188 nlk->dump_done_errno = cb->dump(skb, cb);
2190 if (nlk->dump_done_errno > 0 ||
2191 skb_tailroom(skb) < nlmsg_total_size(sizeof(nlk->dump_done_errno))) {
2192 mutex_unlock(nlk->cb_mutex);
2194 if (sk_filter(sk, skb))
2197 __netlink_sendskb(sk, skb);
2201 nlh = nlmsg_put_answer(skb, cb, NLMSG_DONE,
2202 sizeof(nlk->dump_done_errno), NLM_F_MULTI);
2206 nl_dump_check_consistent(cb, nlh);
2208 memcpy(nlmsg_data(nlh), &nlk->dump_done_errno,
2209 sizeof(nlk->dump_done_errno));
2211 if (sk_filter(sk, skb))
2214 __netlink_sendskb(sk, skb);
2219 nlk->cb_running = false;
2220 module = cb->module;
2222 mutex_unlock(nlk->cb_mutex);
2228 mutex_unlock(nlk->cb_mutex);
2233 int __netlink_dump_start(struct sock *ssk, struct sk_buff *skb,
2234 const struct nlmsghdr *nlh,
2235 struct netlink_dump_control *control)
2237 struct netlink_callback *cb;
2239 struct netlink_sock *nlk;
2242 atomic_inc(&skb->users);
2244 sk = netlink_lookup(sock_net(ssk), ssk->sk_protocol, NETLINK_CB(skb).portid);
2246 ret = -ECONNREFUSED;
2251 mutex_lock(nlk->cb_mutex);
2252 /* A dump is in progress... */
2253 if (nlk->cb_running) {
2257 /* add reference of module which cb->dump belongs to */
2258 if (!try_module_get(control->module)) {
2259 ret = -EPROTONOSUPPORT;
2264 memset(cb, 0, sizeof(*cb));
2265 cb->start = control->start;
2266 cb->dump = control->dump;
2267 cb->done = control->done;
2269 cb->data = control->data;
2270 cb->module = control->module;
2271 cb->min_dump_alloc = control->min_dump_alloc;
2274 nlk->cb_running = true;
2275 nlk->dump_done_errno = INT_MAX;
2277 mutex_unlock(nlk->cb_mutex);
2282 ret = netlink_dump(sk);
2288 /* We successfully started a dump, by returning -EINTR we
2289 * signal not to send ACK even if it was requested.
2295 mutex_unlock(nlk->cb_mutex);
2300 EXPORT_SYMBOL(__netlink_dump_start);
2302 void netlink_ack(struct sk_buff *in_skb, struct nlmsghdr *nlh, int err)
2304 struct sk_buff *skb;
2305 struct nlmsghdr *rep;
2306 struct nlmsgerr *errmsg;
2307 size_t payload = sizeof(*errmsg);
2308 struct netlink_sock *nlk = nlk_sk(NETLINK_CB(in_skb).sk);
2310 /* Error messages get the original request appened, unless the user
2311 * requests to cap the error message.
2313 if (!(nlk->flags & NETLINK_F_CAP_ACK) && err)
2314 payload += nlmsg_len(nlh);
2316 skb = netlink_alloc_skb(in_skb->sk, nlmsg_total_size(payload),
2317 NETLINK_CB(in_skb).portid, GFP_KERNEL);
2321 sk = netlink_lookup(sock_net(in_skb->sk),
2322 in_skb->sk->sk_protocol,
2323 NETLINK_CB(in_skb).portid);
2325 sk->sk_err = ENOBUFS;
2326 sk->sk_error_report(sk);
2332 rep = __nlmsg_put(skb, NETLINK_CB(in_skb).portid, nlh->nlmsg_seq,
2333 NLMSG_ERROR, payload, 0);
2334 errmsg = nlmsg_data(rep);
2335 errmsg->error = err;
2336 memcpy(&errmsg->msg, nlh, payload > sizeof(*errmsg) ? nlh->nlmsg_len : sizeof(*nlh));
2337 netlink_unicast(in_skb->sk, skb, NETLINK_CB(in_skb).portid, MSG_DONTWAIT);
2339 EXPORT_SYMBOL(netlink_ack);
2341 int netlink_rcv_skb(struct sk_buff *skb, int (*cb)(struct sk_buff *,
2344 struct nlmsghdr *nlh;
2347 while (skb->len >= nlmsg_total_size(0)) {
2350 nlh = nlmsg_hdr(skb);
2353 if (nlh->nlmsg_len < NLMSG_HDRLEN || skb->len < nlh->nlmsg_len)
2356 /* Only requests are handled by the kernel */
2357 if (!(nlh->nlmsg_flags & NLM_F_REQUEST))
2360 /* Skip control messages */
2361 if (nlh->nlmsg_type < NLMSG_MIN_TYPE)
2369 if (nlh->nlmsg_flags & NLM_F_ACK || err)
2370 netlink_ack(skb, nlh, err);
2373 msglen = NLMSG_ALIGN(nlh->nlmsg_len);
2374 if (msglen > skb->len)
2376 skb_pull(skb, msglen);
2381 EXPORT_SYMBOL(netlink_rcv_skb);
2384 * nlmsg_notify - send a notification netlink message
2385 * @sk: netlink socket to use
2386 * @skb: notification message
2387 * @portid: destination netlink portid for reports or 0
2388 * @group: destination multicast group or 0
2389 * @report: 1 to report back, 0 to disable
2390 * @flags: allocation flags
2392 int nlmsg_notify(struct sock *sk, struct sk_buff *skb, u32 portid,
2393 unsigned int group, int report, gfp_t flags)
2398 int exclude_portid = 0;
2401 atomic_inc(&skb->users);
2402 exclude_portid = portid;
2405 /* errors reported via destination sk->sk_err, but propagate
2406 * delivery errors if NETLINK_BROADCAST_ERROR flag is set */
2407 err = nlmsg_multicast(sk, skb, exclude_portid, group, flags);
2415 err2 = nlmsg_unicast(sk, skb, portid);
2422 EXPORT_SYMBOL(nlmsg_notify);
2424 #ifdef CONFIG_PROC_FS
2425 struct nl_seq_iter {
2426 struct seq_net_private p;
2427 struct rhashtable_iter hti;
2431 static int netlink_walk_start(struct nl_seq_iter *iter)
2435 err = rhashtable_walk_init(&nl_table[iter->link].hash, &iter->hti);
2437 iter->link = MAX_LINKS;
2441 err = rhashtable_walk_start(&iter->hti);
2442 return err == -EAGAIN ? 0 : err;
2445 static void netlink_walk_stop(struct nl_seq_iter *iter)
2447 rhashtable_walk_stop(&iter->hti);
2448 rhashtable_walk_exit(&iter->hti);
2451 static void *__netlink_seq_next(struct seq_file *seq)
2453 struct nl_seq_iter *iter = seq->private;
2454 struct netlink_sock *nlk;
2460 nlk = rhashtable_walk_next(&iter->hti);
2463 if (PTR_ERR(nlk) == -EAGAIN)
2472 netlink_walk_stop(iter);
2473 if (++iter->link >= MAX_LINKS)
2476 err = netlink_walk_start(iter);
2478 return ERR_PTR(err);
2480 } while (sock_net(&nlk->sk) != seq_file_net(seq));
2485 static void *netlink_seq_start(struct seq_file *seq, loff_t *posp)
2487 struct nl_seq_iter *iter = seq->private;
2488 void *obj = SEQ_START_TOKEN;
2494 err = netlink_walk_start(iter);
2496 return ERR_PTR(err);
2498 for (pos = *posp; pos && obj && !IS_ERR(obj); pos--)
2499 obj = __netlink_seq_next(seq);
2504 static void *netlink_seq_next(struct seq_file *seq, void *v, loff_t *pos)
2507 return __netlink_seq_next(seq);
2510 static void netlink_seq_stop(struct seq_file *seq, void *v)
2512 struct nl_seq_iter *iter = seq->private;
2514 if (iter->link >= MAX_LINKS)
2517 netlink_walk_stop(iter);
2521 static int netlink_seq_show(struct seq_file *seq, void *v)
2523 if (v == SEQ_START_TOKEN) {
2525 "sk Eth Pid Groups "
2526 "Rmem Wmem Dump Locks Drops Inode\n");
2529 struct netlink_sock *nlk = nlk_sk(s);
2531 seq_printf(seq, "%pK %-3d %-6u %08x %-8d %-8d %d %-8d %-8d %-8lu\n",
2535 nlk->groups ? (u32)nlk->groups[0] : 0,
2536 sk_rmem_alloc_get(s),
2537 sk_wmem_alloc_get(s),
2539 atomic_read(&s->sk_refcnt),
2540 atomic_read(&s->sk_drops),
2548 static const struct seq_operations netlink_seq_ops = {
2549 .start = netlink_seq_start,
2550 .next = netlink_seq_next,
2551 .stop = netlink_seq_stop,
2552 .show = netlink_seq_show,
2556 static int netlink_seq_open(struct inode *inode, struct file *file)
2558 return seq_open_net(inode, file, &netlink_seq_ops,
2559 sizeof(struct nl_seq_iter));
2562 static const struct file_operations netlink_seq_fops = {
2563 .owner = THIS_MODULE,
2564 .open = netlink_seq_open,
2566 .llseek = seq_lseek,
2567 .release = seq_release_net,
2572 int netlink_register_notifier(struct notifier_block *nb)
2574 return atomic_notifier_chain_register(&netlink_chain, nb);
2576 EXPORT_SYMBOL(netlink_register_notifier);
2578 int netlink_unregister_notifier(struct notifier_block *nb)
2580 return atomic_notifier_chain_unregister(&netlink_chain, nb);
2582 EXPORT_SYMBOL(netlink_unregister_notifier);
2584 static const struct proto_ops netlink_ops = {
2585 .family = PF_NETLINK,
2586 .owner = THIS_MODULE,
2587 .release = netlink_release,
2588 .bind = netlink_bind,
2589 .connect = netlink_connect,
2590 .socketpair = sock_no_socketpair,
2591 .accept = sock_no_accept,
2592 .getname = netlink_getname,
2593 .poll = datagram_poll,
2594 .ioctl = sock_no_ioctl,
2595 .listen = sock_no_listen,
2596 .shutdown = sock_no_shutdown,
2597 .setsockopt = netlink_setsockopt,
2598 .getsockopt = netlink_getsockopt,
2599 .sendmsg = netlink_sendmsg,
2600 .recvmsg = netlink_recvmsg,
2601 .mmap = sock_no_mmap,
2602 .sendpage = sock_no_sendpage,
2605 static const struct net_proto_family netlink_family_ops = {
2606 .family = PF_NETLINK,
2607 .create = netlink_create,
2608 .owner = THIS_MODULE, /* for consistency 8) */
2611 static int __net_init netlink_net_init(struct net *net)
2613 #ifdef CONFIG_PROC_FS
2614 if (!proc_create("netlink", 0, net->proc_net, &netlink_seq_fops))
2620 static void __net_exit netlink_net_exit(struct net *net)
2622 #ifdef CONFIG_PROC_FS
2623 remove_proc_entry("netlink", net->proc_net);
2627 static void __init netlink_add_usersock_entry(void)
2629 struct listeners *listeners;
2632 listeners = kzalloc(sizeof(*listeners) + NLGRPSZ(groups), GFP_KERNEL);
2634 panic("netlink_add_usersock_entry: Cannot allocate listeners\n");
2636 netlink_table_grab();
2638 nl_table[NETLINK_USERSOCK].groups = groups;
2639 rcu_assign_pointer(nl_table[NETLINK_USERSOCK].listeners, listeners);
2640 nl_table[NETLINK_USERSOCK].module = THIS_MODULE;
2641 nl_table[NETLINK_USERSOCK].registered = 1;
2642 nl_table[NETLINK_USERSOCK].flags = NL_CFG_F_NONROOT_SEND;
2644 netlink_table_ungrab();
2647 static struct pernet_operations __net_initdata netlink_net_ops = {
2648 .init = netlink_net_init,
2649 .exit = netlink_net_exit,
2652 static inline u32 netlink_hash(const void *data, u32 len, u32 seed)
2654 const struct netlink_sock *nlk = data;
2655 struct netlink_compare_arg arg;
2657 netlink_compare_arg_init(&arg, sock_net(&nlk->sk), nlk->portid);
2658 return jhash2((u32 *)&arg, netlink_compare_arg_len / sizeof(u32), seed);
2661 static const struct rhashtable_params netlink_rhashtable_params = {
2662 .head_offset = offsetof(struct netlink_sock, node),
2663 .key_len = netlink_compare_arg_len,
2664 .obj_hashfn = netlink_hash,
2665 .obj_cmpfn = netlink_compare,
2666 .automatic_shrinking = true,
2669 static int __init netlink_proto_init(void)
2672 int err = proto_register(&netlink_proto, 0);
2677 BUILD_BUG_ON(sizeof(struct netlink_skb_parms) > FIELD_SIZEOF(struct sk_buff, cb));
2679 nl_table = kcalloc(MAX_LINKS, sizeof(*nl_table), GFP_KERNEL);
2683 for (i = 0; i < MAX_LINKS; i++) {
2684 if (rhashtable_init(&nl_table[i].hash,
2685 &netlink_rhashtable_params) < 0) {
2687 rhashtable_destroy(&nl_table[i].hash);
2693 INIT_LIST_HEAD(&netlink_tap_all);
2695 netlink_add_usersock_entry();
2697 sock_register(&netlink_family_ops);
2698 register_pernet_subsys(&netlink_net_ops);
2699 /* The netlink device handler may be needed early. */
2704 panic("netlink_init: Cannot allocate nl_table\n");
2707 core_initcall(netlink_proto_init);