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>
66 #include <net/net_namespace.h>
69 #include <net/netlink.h>
71 #include "af_netlink.h"
75 unsigned long masks[0];
79 #define NETLINK_S_CONGESTED 0x0
82 #define NETLINK_F_KERNEL_SOCKET 0x1
83 #define NETLINK_F_RECV_PKTINFO 0x2
84 #define NETLINK_F_BROADCAST_SEND_ERROR 0x4
85 #define NETLINK_F_RECV_NO_ENOBUFS 0x8
86 #define NETLINK_F_LISTEN_ALL_NSID 0x10
87 #define NETLINK_F_CAP_ACK 0x20
89 static inline int netlink_is_kernel(struct sock *sk)
91 return nlk_sk(sk)->flags & NETLINK_F_KERNEL_SOCKET;
94 struct netlink_table *nl_table __read_mostly;
95 EXPORT_SYMBOL_GPL(nl_table);
97 static DECLARE_WAIT_QUEUE_HEAD(nl_table_wait);
99 static int netlink_dump(struct sock *sk);
100 static void netlink_skb_destructor(struct sk_buff *skb);
102 /* nl_table locking explained:
103 * Lookup and traversal are protected with an RCU read-side lock. Insertion
104 * and removal are protected with per bucket lock while using RCU list
105 * modification primitives and may run in parallel to RCU protected lookups.
106 * Destruction of the Netlink socket may only occur *after* nl_table_lock has
107 * been acquired * either during or after the socket has been removed from
108 * the list and after an RCU grace period.
110 DEFINE_RWLOCK(nl_table_lock);
111 EXPORT_SYMBOL_GPL(nl_table_lock);
112 static atomic_t nl_table_users = ATOMIC_INIT(0);
114 #define nl_deref_protected(X) rcu_dereference_protected(X, lockdep_is_held(&nl_table_lock));
116 static ATOMIC_NOTIFIER_HEAD(netlink_chain);
118 static DEFINE_SPINLOCK(netlink_tap_lock);
119 static struct list_head netlink_tap_all __read_mostly;
121 static const struct rhashtable_params netlink_rhashtable_params;
123 static inline u32 netlink_group_mask(u32 group)
125 return group ? 1 << (group - 1) : 0;
128 static struct sk_buff *netlink_to_full_skb(const struct sk_buff *skb,
131 unsigned int len = skb_end_offset(skb);
134 new = alloc_skb(len, gfp_mask);
138 NETLINK_CB(new).portid = NETLINK_CB(skb).portid;
139 NETLINK_CB(new).dst_group = NETLINK_CB(skb).dst_group;
140 NETLINK_CB(new).creds = NETLINK_CB(skb).creds;
142 memcpy(skb_put(new, len), skb->data, len);
146 int netlink_add_tap(struct netlink_tap *nt)
148 if (unlikely(nt->dev->type != ARPHRD_NETLINK))
151 spin_lock(&netlink_tap_lock);
152 list_add_rcu(&nt->list, &netlink_tap_all);
153 spin_unlock(&netlink_tap_lock);
155 __module_get(nt->module);
159 EXPORT_SYMBOL_GPL(netlink_add_tap);
161 static int __netlink_remove_tap(struct netlink_tap *nt)
164 struct netlink_tap *tmp;
166 spin_lock(&netlink_tap_lock);
168 list_for_each_entry(tmp, &netlink_tap_all, list) {
170 list_del_rcu(&nt->list);
176 pr_warn("__netlink_remove_tap: %p not found\n", nt);
178 spin_unlock(&netlink_tap_lock);
181 module_put(nt->module);
183 return found ? 0 : -ENODEV;
186 int netlink_remove_tap(struct netlink_tap *nt)
190 ret = __netlink_remove_tap(nt);
195 EXPORT_SYMBOL_GPL(netlink_remove_tap);
197 static bool netlink_filter_tap(const struct sk_buff *skb)
199 struct sock *sk = skb->sk;
201 /* We take the more conservative approach and
202 * whitelist socket protocols that may pass.
204 switch (sk->sk_protocol) {
206 case NETLINK_USERSOCK:
207 case NETLINK_SOCK_DIAG:
210 case NETLINK_FIB_LOOKUP:
211 case NETLINK_NETFILTER:
212 case NETLINK_GENERIC:
219 static int __netlink_deliver_tap_skb(struct sk_buff *skb,
220 struct net_device *dev)
222 struct sk_buff *nskb;
223 struct sock *sk = skb->sk;
228 if (is_vmalloc_addr(skb->head))
229 nskb = netlink_to_full_skb(skb, GFP_ATOMIC);
231 nskb = skb_clone(skb, GFP_ATOMIC);
234 nskb->protocol = htons((u16) sk->sk_protocol);
235 nskb->pkt_type = netlink_is_kernel(sk) ?
236 PACKET_KERNEL : PACKET_USER;
237 skb_reset_network_header(nskb);
238 ret = dev_queue_xmit(nskb);
239 if (unlikely(ret > 0))
240 ret = net_xmit_errno(ret);
247 static void __netlink_deliver_tap(struct sk_buff *skb)
250 struct netlink_tap *tmp;
252 if (!netlink_filter_tap(skb))
255 list_for_each_entry_rcu(tmp, &netlink_tap_all, list) {
256 ret = __netlink_deliver_tap_skb(skb, tmp->dev);
262 static void netlink_deliver_tap(struct sk_buff *skb)
266 if (unlikely(!list_empty(&netlink_tap_all)))
267 __netlink_deliver_tap(skb);
272 static void netlink_deliver_tap_kernel(struct sock *dst, struct sock *src,
275 if (!(netlink_is_kernel(dst) && netlink_is_kernel(src)))
276 netlink_deliver_tap(skb);
279 static void netlink_overrun(struct sock *sk)
281 struct netlink_sock *nlk = nlk_sk(sk);
283 if (!(nlk->flags & NETLINK_F_RECV_NO_ENOBUFS)) {
284 if (!test_and_set_bit(NETLINK_S_CONGESTED,
285 &nlk_sk(sk)->state)) {
286 sk->sk_err = ENOBUFS;
287 sk->sk_error_report(sk);
290 atomic_inc(&sk->sk_drops);
293 static void netlink_rcv_wake(struct sock *sk)
295 struct netlink_sock *nlk = nlk_sk(sk);
297 if (skb_queue_empty(&sk->sk_receive_queue))
298 clear_bit(NETLINK_S_CONGESTED, &nlk->state);
299 if (!test_bit(NETLINK_S_CONGESTED, &nlk->state))
300 wake_up_interruptible(&nlk->wait);
303 static void netlink_skb_destructor(struct sk_buff *skb)
305 if (is_vmalloc_addr(skb->head)) {
307 !atomic_dec_return(&(skb_shinfo(skb)->dataref)))
316 static void netlink_skb_set_owner_r(struct sk_buff *skb, struct sock *sk)
318 WARN_ON(skb->sk != NULL);
320 skb->destructor = netlink_skb_destructor;
321 atomic_add(skb->truesize, &sk->sk_rmem_alloc);
322 sk_mem_charge(sk, skb->truesize);
325 static void __netlink_sock_destruct(struct sock *sk)
327 struct netlink_sock *nlk = nlk_sk(sk);
329 if (nlk->cb_running) {
330 module_put(nlk->cb.module);
331 kfree_skb(nlk->cb.skb);
334 skb_queue_purge(&sk->sk_receive_queue);
336 if (!sock_flag(sk, SOCK_DEAD)) {
337 printk(KERN_ERR "Freeing alive netlink socket %p\n", sk);
341 WARN_ON(atomic_read(&sk->sk_rmem_alloc));
342 WARN_ON(atomic_read(&sk->sk_wmem_alloc));
343 WARN_ON(nlk_sk(sk)->groups);
346 static void netlink_sock_destruct_work(struct work_struct *work)
348 struct netlink_sock *nlk = container_of(work, struct netlink_sock,
351 nlk->cb.done(&nlk->cb);
352 __netlink_sock_destruct(&nlk->sk);
355 static void netlink_sock_destruct(struct sock *sk)
357 struct netlink_sock *nlk = nlk_sk(sk);
359 if (nlk->cb_running && nlk->cb.done) {
360 INIT_WORK(&nlk->work, netlink_sock_destruct_work);
361 schedule_work(&nlk->work);
365 __netlink_sock_destruct(sk);
368 /* This lock without WQ_FLAG_EXCLUSIVE is good on UP and it is _very_ bad on
369 * SMP. Look, when several writers sleep and reader wakes them up, all but one
370 * immediately hit write lock and grab all the cpus. Exclusive sleep solves
371 * this, _but_ remember, it adds useless work on UP machines.
374 void netlink_table_grab(void)
375 __acquires(nl_table_lock)
379 write_lock_irq(&nl_table_lock);
381 if (atomic_read(&nl_table_users)) {
382 DECLARE_WAITQUEUE(wait, current);
384 add_wait_queue_exclusive(&nl_table_wait, &wait);
386 set_current_state(TASK_UNINTERRUPTIBLE);
387 if (atomic_read(&nl_table_users) == 0)
389 write_unlock_irq(&nl_table_lock);
391 write_lock_irq(&nl_table_lock);
394 __set_current_state(TASK_RUNNING);
395 remove_wait_queue(&nl_table_wait, &wait);
399 void netlink_table_ungrab(void)
400 __releases(nl_table_lock)
402 write_unlock_irq(&nl_table_lock);
403 wake_up(&nl_table_wait);
407 netlink_lock_table(void)
409 /* read_lock() synchronizes us to netlink_table_grab */
411 read_lock(&nl_table_lock);
412 atomic_inc(&nl_table_users);
413 read_unlock(&nl_table_lock);
417 netlink_unlock_table(void)
419 if (atomic_dec_and_test(&nl_table_users))
420 wake_up(&nl_table_wait);
423 struct netlink_compare_arg
429 /* Doing sizeof directly may yield 4 extra bytes on 64-bit. */
430 #define netlink_compare_arg_len \
431 (offsetof(struct netlink_compare_arg, portid) + sizeof(u32))
433 static inline int netlink_compare(struct rhashtable_compare_arg *arg,
436 const struct netlink_compare_arg *x = arg->key;
437 const struct netlink_sock *nlk = ptr;
439 return nlk->portid != x->portid ||
440 !net_eq(sock_net(&nlk->sk), read_pnet(&x->pnet));
443 static void netlink_compare_arg_init(struct netlink_compare_arg *arg,
444 struct net *net, u32 portid)
446 memset(arg, 0, sizeof(*arg));
447 write_pnet(&arg->pnet, net);
448 arg->portid = portid;
451 static struct sock *__netlink_lookup(struct netlink_table *table, u32 portid,
454 struct netlink_compare_arg arg;
456 netlink_compare_arg_init(&arg, net, portid);
457 return rhashtable_lookup_fast(&table->hash, &arg,
458 netlink_rhashtable_params);
461 static int __netlink_insert(struct netlink_table *table, struct sock *sk)
463 struct netlink_compare_arg arg;
465 netlink_compare_arg_init(&arg, sock_net(sk), nlk_sk(sk)->portid);
466 return rhashtable_lookup_insert_key(&table->hash, &arg,
468 netlink_rhashtable_params);
471 static struct sock *netlink_lookup(struct net *net, int protocol, u32 portid)
473 struct netlink_table *table = &nl_table[protocol];
477 sk = __netlink_lookup(table, portid, net);
485 static const struct proto_ops netlink_ops;
488 netlink_update_listeners(struct sock *sk)
490 struct netlink_table *tbl = &nl_table[sk->sk_protocol];
493 struct listeners *listeners;
495 listeners = nl_deref_protected(tbl->listeners);
499 for (i = 0; i < NLGRPLONGS(tbl->groups); i++) {
501 sk_for_each_bound(sk, &tbl->mc_list) {
502 if (i < NLGRPLONGS(nlk_sk(sk)->ngroups))
503 mask |= nlk_sk(sk)->groups[i];
505 listeners->masks[i] = mask;
507 /* this function is only called with the netlink table "grabbed", which
508 * makes sure updates are visible before bind or setsockopt return. */
511 static int netlink_insert(struct sock *sk, u32 portid)
513 struct netlink_table *table = &nl_table[sk->sk_protocol];
518 err = nlk_sk(sk)->portid == portid ? 0 : -EBUSY;
519 if (nlk_sk(sk)->bound)
523 if (BITS_PER_LONG > 32 &&
524 unlikely(atomic_read(&table->hash.nelems) >= UINT_MAX))
527 nlk_sk(sk)->portid = portid;
530 err = __netlink_insert(table, sk);
532 /* In case the hashtable backend returns with -EBUSY
533 * from here, it must not escape to the caller.
535 if (unlikely(err == -EBUSY))
543 /* We need to ensure that the socket is hashed and visible. */
545 nlk_sk(sk)->bound = portid;
552 static void netlink_remove(struct sock *sk)
554 struct netlink_table *table;
556 table = &nl_table[sk->sk_protocol];
557 if (!rhashtable_remove_fast(&table->hash, &nlk_sk(sk)->node,
558 netlink_rhashtable_params)) {
559 WARN_ON(atomic_read(&sk->sk_refcnt) == 1);
563 netlink_table_grab();
564 if (nlk_sk(sk)->subscriptions) {
565 __sk_del_bind_node(sk);
566 netlink_update_listeners(sk);
568 if (sk->sk_protocol == NETLINK_GENERIC)
569 atomic_inc(&genl_sk_destructing_cnt);
570 netlink_table_ungrab();
573 static struct proto netlink_proto = {
575 .owner = THIS_MODULE,
576 .obj_size = sizeof(struct netlink_sock),
579 static int __netlink_create(struct net *net, struct socket *sock,
580 struct mutex *cb_mutex, int protocol,
584 struct netlink_sock *nlk;
586 sock->ops = &netlink_ops;
588 sk = sk_alloc(net, PF_NETLINK, GFP_KERNEL, &netlink_proto, kern);
592 sock_init_data(sock, sk);
596 nlk->cb_mutex = cb_mutex;
598 nlk->cb_mutex = &nlk->cb_def_mutex;
599 mutex_init(nlk->cb_mutex);
601 init_waitqueue_head(&nlk->wait);
603 sk->sk_destruct = netlink_sock_destruct;
604 sk->sk_protocol = protocol;
608 static int netlink_create(struct net *net, struct socket *sock, int protocol,
611 struct module *module = NULL;
612 struct mutex *cb_mutex;
613 struct netlink_sock *nlk;
614 int (*bind)(struct net *net, int group);
615 void (*unbind)(struct net *net, int group);
618 sock->state = SS_UNCONNECTED;
620 if (sock->type != SOCK_RAW && sock->type != SOCK_DGRAM)
621 return -ESOCKTNOSUPPORT;
623 if (protocol < 0 || protocol >= MAX_LINKS)
624 return -EPROTONOSUPPORT;
626 netlink_lock_table();
627 #ifdef CONFIG_MODULES
628 if (!nl_table[protocol].registered) {
629 netlink_unlock_table();
630 request_module("net-pf-%d-proto-%d", PF_NETLINK, protocol);
631 netlink_lock_table();
634 if (nl_table[protocol].registered &&
635 try_module_get(nl_table[protocol].module))
636 module = nl_table[protocol].module;
638 err = -EPROTONOSUPPORT;
639 cb_mutex = nl_table[protocol].cb_mutex;
640 bind = nl_table[protocol].bind;
641 unbind = nl_table[protocol].unbind;
642 netlink_unlock_table();
647 err = __netlink_create(net, sock, cb_mutex, protocol, kern);
652 sock_prot_inuse_add(net, &netlink_proto, 1);
655 nlk = nlk_sk(sock->sk);
656 nlk->module = module;
657 nlk->netlink_bind = bind;
658 nlk->netlink_unbind = unbind;
667 static void deferred_put_nlk_sk(struct rcu_head *head)
669 struct netlink_sock *nlk = container_of(head, struct netlink_sock, rcu);
674 static int netlink_release(struct socket *sock)
676 struct sock *sk = sock->sk;
677 struct netlink_sock *nlk;
687 * OK. Socket is unlinked, any packets that arrive now
691 /* must not acquire netlink_table_lock in any way again before unbind
692 * and notifying genetlink is done as otherwise it might deadlock
694 if (nlk->netlink_unbind) {
697 for (i = 0; i < nlk->ngroups; i++)
698 if (test_bit(i, nlk->groups))
699 nlk->netlink_unbind(sock_net(sk), i + 1);
701 if (sk->sk_protocol == NETLINK_GENERIC &&
702 atomic_dec_return(&genl_sk_destructing_cnt) == 0)
703 wake_up(&genl_sk_destructing_waitq);
706 wake_up_interruptible_all(&nlk->wait);
708 skb_queue_purge(&sk->sk_write_queue);
710 if (nlk->portid && nlk->bound) {
711 struct netlink_notify n = {
713 .protocol = sk->sk_protocol,
714 .portid = nlk->portid,
716 atomic_notifier_call_chain(&netlink_chain,
717 NETLINK_URELEASE, &n);
720 module_put(nlk->module);
722 if (netlink_is_kernel(sk)) {
723 netlink_table_grab();
724 BUG_ON(nl_table[sk->sk_protocol].registered == 0);
725 if (--nl_table[sk->sk_protocol].registered == 0) {
726 struct listeners *old;
728 old = nl_deref_protected(nl_table[sk->sk_protocol].listeners);
729 RCU_INIT_POINTER(nl_table[sk->sk_protocol].listeners, NULL);
731 nl_table[sk->sk_protocol].module = NULL;
732 nl_table[sk->sk_protocol].bind = NULL;
733 nl_table[sk->sk_protocol].unbind = NULL;
734 nl_table[sk->sk_protocol].flags = 0;
735 nl_table[sk->sk_protocol].registered = 0;
737 netlink_table_ungrab();
744 sock_prot_inuse_add(sock_net(sk), &netlink_proto, -1);
746 call_rcu(&nlk->rcu, deferred_put_nlk_sk);
750 static int netlink_autobind(struct socket *sock)
752 struct sock *sk = sock->sk;
753 struct net *net = sock_net(sk);
754 struct netlink_table *table = &nl_table[sk->sk_protocol];
755 s32 portid = task_tgid_vnr(current);
763 ok = !__netlink_lookup(table, portid, net);
766 /* Bind collision, search negative portid values. */
768 /* rover will be in range [S32_MIN, -4097] */
769 rover = S32_MIN + prandom_u32_max(-4096 - S32_MIN);
770 else if (rover >= -4096)
776 err = netlink_insert(sk, portid);
777 if (err == -EADDRINUSE)
780 /* If 2 threads race to autobind, that is fine. */
788 * __netlink_ns_capable - General netlink message capability test
789 * @nsp: NETLINK_CB of the socket buffer holding a netlink command from userspace.
790 * @user_ns: The user namespace of the capability to use
791 * @cap: The capability to use
793 * Test to see if the opener of the socket we received the message
794 * from had when the netlink socket was created and the sender of the
795 * message has has the capability @cap in the user namespace @user_ns.
797 bool __netlink_ns_capable(const struct netlink_skb_parms *nsp,
798 struct user_namespace *user_ns, int cap)
800 return ((nsp->flags & NETLINK_SKB_DST) ||
801 file_ns_capable(nsp->sk->sk_socket->file, user_ns, cap)) &&
802 ns_capable(user_ns, cap);
804 EXPORT_SYMBOL(__netlink_ns_capable);
807 * netlink_ns_capable - General netlink message capability test
808 * @skb: socket buffer holding a netlink command from userspace
809 * @user_ns: The user namespace of the capability to use
810 * @cap: The capability to use
812 * Test to see if the opener of the socket we received the message
813 * from had when the netlink socket was created and the sender of the
814 * message has has the capability @cap in the user namespace @user_ns.
816 bool netlink_ns_capable(const struct sk_buff *skb,
817 struct user_namespace *user_ns, int cap)
819 return __netlink_ns_capable(&NETLINK_CB(skb), user_ns, cap);
821 EXPORT_SYMBOL(netlink_ns_capable);
824 * netlink_capable - Netlink global message capability test
825 * @skb: socket buffer holding a netlink command from userspace
826 * @cap: The capability to use
828 * Test to see if the opener of the socket we received the message
829 * from had when the netlink socket was created and the sender of the
830 * message has has the capability @cap in all user namespaces.
832 bool netlink_capable(const struct sk_buff *skb, int cap)
834 return netlink_ns_capable(skb, &init_user_ns, cap);
836 EXPORT_SYMBOL(netlink_capable);
839 * netlink_net_capable - Netlink network namespace message capability test
840 * @skb: socket buffer holding a netlink command from userspace
841 * @cap: The capability to use
843 * Test to see if the opener of the socket we received the message
844 * from had when the netlink socket was created and the sender of the
845 * message has has the capability @cap over the network namespace of
846 * the socket we received the message from.
848 bool netlink_net_capable(const struct sk_buff *skb, int cap)
850 return netlink_ns_capable(skb, sock_net(skb->sk)->user_ns, cap);
852 EXPORT_SYMBOL(netlink_net_capable);
854 static inline int netlink_allowed(const struct socket *sock, unsigned int flag)
856 return (nl_table[sock->sk->sk_protocol].flags & flag) ||
857 ns_capable(sock_net(sock->sk)->user_ns, CAP_NET_ADMIN);
861 netlink_update_subscriptions(struct sock *sk, unsigned int subscriptions)
863 struct netlink_sock *nlk = nlk_sk(sk);
865 if (nlk->subscriptions && !subscriptions)
866 __sk_del_bind_node(sk);
867 else if (!nlk->subscriptions && subscriptions)
868 sk_add_bind_node(sk, &nl_table[sk->sk_protocol].mc_list);
869 nlk->subscriptions = subscriptions;
872 static int netlink_realloc_groups(struct sock *sk)
874 struct netlink_sock *nlk = nlk_sk(sk);
876 unsigned long *new_groups;
879 netlink_table_grab();
881 groups = nl_table[sk->sk_protocol].groups;
882 if (!nl_table[sk->sk_protocol].registered) {
887 if (nlk->ngroups >= groups)
890 new_groups = krealloc(nlk->groups, NLGRPSZ(groups), GFP_ATOMIC);
891 if (new_groups == NULL) {
895 memset((char *)new_groups + NLGRPSZ(nlk->ngroups), 0,
896 NLGRPSZ(groups) - NLGRPSZ(nlk->ngroups));
898 nlk->groups = new_groups;
899 nlk->ngroups = groups;
901 netlink_table_ungrab();
905 static void netlink_undo_bind(int group, long unsigned int groups,
908 struct netlink_sock *nlk = nlk_sk(sk);
911 if (!nlk->netlink_unbind)
914 for (undo = 0; undo < group; undo++)
915 if (test_bit(undo, &groups))
916 nlk->netlink_unbind(sock_net(sk), undo + 1);
919 static int netlink_bind(struct socket *sock, struct sockaddr *addr,
922 struct sock *sk = sock->sk;
923 struct net *net = sock_net(sk);
924 struct netlink_sock *nlk = nlk_sk(sk);
925 struct sockaddr_nl *nladdr = (struct sockaddr_nl *)addr;
927 long unsigned int groups = nladdr->nl_groups;
930 if (addr_len < sizeof(struct sockaddr_nl))
933 if (nladdr->nl_family != AF_NETLINK)
936 /* Only superuser is allowed to listen multicasts */
938 if (!netlink_allowed(sock, NL_CFG_F_NONROOT_RECV))
940 err = netlink_realloc_groups(sk);
947 /* Ensure nlk->portid is up-to-date. */
950 if (nladdr->nl_pid != nlk->portid)
954 if (nlk->netlink_bind && groups) {
957 for (group = 0; group < nlk->ngroups; group++) {
958 if (!test_bit(group, &groups))
960 err = nlk->netlink_bind(net, group + 1);
963 netlink_undo_bind(group, groups, sk);
968 /* No need for barriers here as we return to user-space without
969 * using any of the bound attributes.
972 err = nladdr->nl_pid ?
973 netlink_insert(sk, nladdr->nl_pid) :
974 netlink_autobind(sock);
976 netlink_undo_bind(nlk->ngroups, groups, sk);
981 if (!groups && (nlk->groups == NULL || !(u32)nlk->groups[0]))
984 netlink_table_grab();
985 netlink_update_subscriptions(sk, nlk->subscriptions +
987 hweight32(nlk->groups[0]));
988 nlk->groups[0] = (nlk->groups[0] & ~0xffffffffUL) | groups;
989 netlink_update_listeners(sk);
990 netlink_table_ungrab();
995 static int netlink_connect(struct socket *sock, struct sockaddr *addr,
999 struct sock *sk = sock->sk;
1000 struct netlink_sock *nlk = nlk_sk(sk);
1001 struct sockaddr_nl *nladdr = (struct sockaddr_nl *)addr;
1003 if (alen < sizeof(addr->sa_family))
1006 if (addr->sa_family == AF_UNSPEC) {
1007 sk->sk_state = NETLINK_UNCONNECTED;
1008 nlk->dst_portid = 0;
1012 if (addr->sa_family != AF_NETLINK)
1015 if ((nladdr->nl_groups || nladdr->nl_pid) &&
1016 !netlink_allowed(sock, NL_CFG_F_NONROOT_SEND))
1019 /* No need for barriers here as we return to user-space without
1020 * using any of the bound attributes.
1023 err = netlink_autobind(sock);
1026 sk->sk_state = NETLINK_CONNECTED;
1027 nlk->dst_portid = nladdr->nl_pid;
1028 nlk->dst_group = ffs(nladdr->nl_groups);
1034 static int netlink_getname(struct socket *sock, struct sockaddr *addr,
1035 int *addr_len, int peer)
1037 struct sock *sk = sock->sk;
1038 struct netlink_sock *nlk = nlk_sk(sk);
1039 DECLARE_SOCKADDR(struct sockaddr_nl *, nladdr, addr);
1041 nladdr->nl_family = AF_NETLINK;
1043 *addr_len = sizeof(*nladdr);
1046 nladdr->nl_pid = nlk->dst_portid;
1047 nladdr->nl_groups = netlink_group_mask(nlk->dst_group);
1049 nladdr->nl_pid = nlk->portid;
1050 nladdr->nl_groups = nlk->groups ? nlk->groups[0] : 0;
1055 static int netlink_ioctl(struct socket *sock, unsigned int cmd,
1058 /* try to hand this ioctl down to the NIC drivers.
1060 return -ENOIOCTLCMD;
1063 static struct sock *netlink_getsockbyportid(struct sock *ssk, u32 portid)
1066 struct netlink_sock *nlk;
1068 sock = netlink_lookup(sock_net(ssk), ssk->sk_protocol, portid);
1070 return ERR_PTR(-ECONNREFUSED);
1072 /* Don't bother queuing skb if kernel socket has no input function */
1074 if (sock->sk_state == NETLINK_CONNECTED &&
1075 nlk->dst_portid != nlk_sk(ssk)->portid) {
1077 return ERR_PTR(-ECONNREFUSED);
1082 struct sock *netlink_getsockbyfilp(struct file *filp)
1084 struct inode *inode = file_inode(filp);
1087 if (!S_ISSOCK(inode->i_mode))
1088 return ERR_PTR(-ENOTSOCK);
1090 sock = SOCKET_I(inode)->sk;
1091 if (sock->sk_family != AF_NETLINK)
1092 return ERR_PTR(-EINVAL);
1098 static struct sk_buff *netlink_alloc_large_skb(unsigned int size,
1101 struct sk_buff *skb;
1104 if (size <= NLMSG_GOODSIZE || broadcast)
1105 return alloc_skb(size, GFP_KERNEL);
1107 size = SKB_DATA_ALIGN(size) +
1108 SKB_DATA_ALIGN(sizeof(struct skb_shared_info));
1110 data = vmalloc(size);
1114 skb = __build_skb(data, size);
1118 skb->destructor = netlink_skb_destructor;
1124 * Attach a skb to a netlink socket.
1125 * The caller must hold a reference to the destination socket. On error, the
1126 * reference is dropped. The skb is not send to the destination, just all
1127 * all error checks are performed and memory in the queue is reserved.
1129 * < 0: error. skb freed, reference to sock dropped.
1131 * 1: repeat lookup - reference dropped while waiting for socket memory.
1133 int netlink_attachskb(struct sock *sk, struct sk_buff *skb,
1134 long *timeo, struct sock *ssk)
1136 struct netlink_sock *nlk;
1140 if ((atomic_read(&sk->sk_rmem_alloc) > sk->sk_rcvbuf ||
1141 test_bit(NETLINK_S_CONGESTED, &nlk->state))) {
1142 DECLARE_WAITQUEUE(wait, current);
1144 if (!ssk || netlink_is_kernel(ssk))
1145 netlink_overrun(sk);
1151 __set_current_state(TASK_INTERRUPTIBLE);
1152 add_wait_queue(&nlk->wait, &wait);
1154 if ((atomic_read(&sk->sk_rmem_alloc) > sk->sk_rcvbuf ||
1155 test_bit(NETLINK_S_CONGESTED, &nlk->state)) &&
1156 !sock_flag(sk, SOCK_DEAD))
1157 *timeo = schedule_timeout(*timeo);
1159 __set_current_state(TASK_RUNNING);
1160 remove_wait_queue(&nlk->wait, &wait);
1163 if (signal_pending(current)) {
1165 return sock_intr_errno(*timeo);
1169 netlink_skb_set_owner_r(skb, sk);
1173 static int __netlink_sendskb(struct sock *sk, struct sk_buff *skb)
1177 netlink_deliver_tap(skb);
1179 skb_queue_tail(&sk->sk_receive_queue, skb);
1180 sk->sk_data_ready(sk);
1184 int netlink_sendskb(struct sock *sk, struct sk_buff *skb)
1186 int len = __netlink_sendskb(sk, skb);
1192 void netlink_detachskb(struct sock *sk, struct sk_buff *skb)
1198 static struct sk_buff *netlink_trim(struct sk_buff *skb, gfp_t allocation)
1202 WARN_ON(skb->sk != NULL);
1203 delta = skb->end - skb->tail;
1204 if (is_vmalloc_addr(skb->head) || delta * 2 < skb->truesize)
1207 if (skb_shared(skb)) {
1208 struct sk_buff *nskb = skb_clone(skb, allocation);
1215 if (!pskb_expand_head(skb, 0, -delta, allocation))
1216 skb->truesize -= delta;
1221 static int netlink_unicast_kernel(struct sock *sk, struct sk_buff *skb,
1225 struct netlink_sock *nlk = nlk_sk(sk);
1227 ret = -ECONNREFUSED;
1228 if (nlk->netlink_rcv != NULL) {
1230 netlink_skb_set_owner_r(skb, sk);
1231 NETLINK_CB(skb).sk = ssk;
1232 netlink_deliver_tap_kernel(sk, ssk, skb);
1233 nlk->netlink_rcv(skb);
1242 int netlink_unicast(struct sock *ssk, struct sk_buff *skb,
1243 u32 portid, int nonblock)
1249 skb = netlink_trim(skb, gfp_any());
1251 timeo = sock_sndtimeo(ssk, nonblock);
1253 sk = netlink_getsockbyportid(ssk, portid);
1258 if (netlink_is_kernel(sk))
1259 return netlink_unicast_kernel(sk, skb, ssk);
1261 if (sk_filter(sk, skb)) {
1268 err = netlink_attachskb(sk, skb, &timeo, ssk);
1274 return netlink_sendskb(sk, skb);
1276 EXPORT_SYMBOL(netlink_unicast);
1278 int netlink_has_listeners(struct sock *sk, unsigned int group)
1281 struct listeners *listeners;
1283 BUG_ON(!netlink_is_kernel(sk));
1286 listeners = rcu_dereference(nl_table[sk->sk_protocol].listeners);
1288 if (listeners && group - 1 < nl_table[sk->sk_protocol].groups)
1289 res = test_bit(group - 1, listeners->masks);
1295 EXPORT_SYMBOL_GPL(netlink_has_listeners);
1297 static int netlink_broadcast_deliver(struct sock *sk, struct sk_buff *skb)
1299 struct netlink_sock *nlk = nlk_sk(sk);
1301 if (atomic_read(&sk->sk_rmem_alloc) <= sk->sk_rcvbuf &&
1302 !test_bit(NETLINK_S_CONGESTED, &nlk->state)) {
1303 netlink_skb_set_owner_r(skb, sk);
1304 __netlink_sendskb(sk, skb);
1305 return atomic_read(&sk->sk_rmem_alloc) > (sk->sk_rcvbuf >> 1);
1310 struct netlink_broadcast_data {
1311 struct sock *exclude_sk;
1316 int delivery_failure;
1320 struct sk_buff *skb, *skb2;
1321 int (*tx_filter)(struct sock *dsk, struct sk_buff *skb, void *data);
1325 static void do_one_broadcast(struct sock *sk,
1326 struct netlink_broadcast_data *p)
1328 struct netlink_sock *nlk = nlk_sk(sk);
1331 if (p->exclude_sk == sk)
1334 if (nlk->portid == p->portid || p->group - 1 >= nlk->ngroups ||
1335 !test_bit(p->group - 1, nlk->groups))
1338 if (!net_eq(sock_net(sk), p->net)) {
1339 if (!(nlk->flags & NETLINK_F_LISTEN_ALL_NSID))
1342 if (!peernet_has_id(sock_net(sk), p->net))
1345 if (!file_ns_capable(sk->sk_socket->file, p->net->user_ns,
1351 netlink_overrun(sk);
1356 if (p->skb2 == NULL) {
1357 if (skb_shared(p->skb)) {
1358 p->skb2 = skb_clone(p->skb, p->allocation);
1360 p->skb2 = skb_get(p->skb);
1362 * skb ownership may have been set when
1363 * delivered to a previous socket.
1365 skb_orphan(p->skb2);
1368 if (p->skb2 == NULL) {
1369 netlink_overrun(sk);
1370 /* Clone failed. Notify ALL listeners. */
1372 if (nlk->flags & NETLINK_F_BROADCAST_SEND_ERROR)
1373 p->delivery_failure = 1;
1376 if (p->tx_filter && p->tx_filter(sk, p->skb2, p->tx_data)) {
1381 if (sk_filter(sk, p->skb2)) {
1386 NETLINK_CB(p->skb2).nsid = peernet2id(sock_net(sk), p->net);
1387 NETLINK_CB(p->skb2).nsid_is_set = true;
1388 val = netlink_broadcast_deliver(sk, p->skb2);
1390 netlink_overrun(sk);
1391 if (nlk->flags & NETLINK_F_BROADCAST_SEND_ERROR)
1392 p->delivery_failure = 1;
1394 p->congested |= val;
1402 int netlink_broadcast_filtered(struct sock *ssk, struct sk_buff *skb, u32 portid,
1403 u32 group, gfp_t allocation,
1404 int (*filter)(struct sock *dsk, struct sk_buff *skb, void *data),
1407 struct net *net = sock_net(ssk);
1408 struct netlink_broadcast_data info;
1411 skb = netlink_trim(skb, allocation);
1413 info.exclude_sk = ssk;
1415 info.portid = portid;
1418 info.delivery_failure = 0;
1421 info.allocation = allocation;
1424 info.tx_filter = filter;
1425 info.tx_data = filter_data;
1427 /* While we sleep in clone, do not allow to change socket list */
1429 netlink_lock_table();
1431 sk_for_each_bound(sk, &nl_table[ssk->sk_protocol].mc_list)
1432 do_one_broadcast(sk, &info);
1436 netlink_unlock_table();
1438 if (info.delivery_failure) {
1439 kfree_skb(info.skb2);
1442 consume_skb(info.skb2);
1444 if (info.delivered) {
1445 if (info.congested && gfpflags_allow_blocking(allocation))
1451 EXPORT_SYMBOL(netlink_broadcast_filtered);
1453 int netlink_broadcast(struct sock *ssk, struct sk_buff *skb, u32 portid,
1454 u32 group, gfp_t allocation)
1456 return netlink_broadcast_filtered(ssk, skb, portid, group, allocation,
1459 EXPORT_SYMBOL(netlink_broadcast);
1461 struct netlink_set_err_data {
1462 struct sock *exclude_sk;
1468 static int do_one_set_err(struct sock *sk, struct netlink_set_err_data *p)
1470 struct netlink_sock *nlk = nlk_sk(sk);
1473 if (sk == p->exclude_sk)
1476 if (!net_eq(sock_net(sk), sock_net(p->exclude_sk)))
1479 if (nlk->portid == p->portid || p->group - 1 >= nlk->ngroups ||
1480 !test_bit(p->group - 1, nlk->groups))
1483 if (p->code == ENOBUFS && nlk->flags & NETLINK_F_RECV_NO_ENOBUFS) {
1488 sk->sk_err = p->code;
1489 sk->sk_error_report(sk);
1495 * netlink_set_err - report error to broadcast listeners
1496 * @ssk: the kernel netlink socket, as returned by netlink_kernel_create()
1497 * @portid: the PORTID of a process that we want to skip (if any)
1498 * @group: the broadcast group that will notice the error
1499 * @code: error code, must be negative (as usual in kernelspace)
1501 * This function returns the number of broadcast listeners that have set the
1502 * NETLINK_NO_ENOBUFS socket option.
1504 int netlink_set_err(struct sock *ssk, u32 portid, u32 group, int code)
1506 struct netlink_set_err_data info;
1510 info.exclude_sk = ssk;
1511 info.portid = portid;
1513 /* sk->sk_err wants a positive error value */
1516 read_lock(&nl_table_lock);
1518 sk_for_each_bound(sk, &nl_table[ssk->sk_protocol].mc_list)
1519 ret += do_one_set_err(sk, &info);
1521 read_unlock(&nl_table_lock);
1524 EXPORT_SYMBOL(netlink_set_err);
1526 /* must be called with netlink table grabbed */
1527 static void netlink_update_socket_mc(struct netlink_sock *nlk,
1531 int old, new = !!is_new, subscriptions;
1533 old = test_bit(group - 1, nlk->groups);
1534 subscriptions = nlk->subscriptions - old + new;
1536 __set_bit(group - 1, nlk->groups);
1538 __clear_bit(group - 1, nlk->groups);
1539 netlink_update_subscriptions(&nlk->sk, subscriptions);
1540 netlink_update_listeners(&nlk->sk);
1543 static int netlink_setsockopt(struct socket *sock, int level, int optname,
1544 char __user *optval, unsigned int optlen)
1546 struct sock *sk = sock->sk;
1547 struct netlink_sock *nlk = nlk_sk(sk);
1548 unsigned int val = 0;
1551 if (level != SOL_NETLINK)
1552 return -ENOPROTOOPT;
1554 if (optlen >= sizeof(int) &&
1555 get_user(val, (unsigned int __user *)optval))
1559 case NETLINK_PKTINFO:
1561 nlk->flags |= NETLINK_F_RECV_PKTINFO;
1563 nlk->flags &= ~NETLINK_F_RECV_PKTINFO;
1566 case NETLINK_ADD_MEMBERSHIP:
1567 case NETLINK_DROP_MEMBERSHIP: {
1568 if (!netlink_allowed(sock, NL_CFG_F_NONROOT_RECV))
1570 err = netlink_realloc_groups(sk);
1573 if (!val || val - 1 >= nlk->ngroups)
1575 if (optname == NETLINK_ADD_MEMBERSHIP && nlk->netlink_bind) {
1576 err = nlk->netlink_bind(sock_net(sk), val);
1580 netlink_table_grab();
1581 netlink_update_socket_mc(nlk, val,
1582 optname == NETLINK_ADD_MEMBERSHIP);
1583 netlink_table_ungrab();
1584 if (optname == NETLINK_DROP_MEMBERSHIP && nlk->netlink_unbind)
1585 nlk->netlink_unbind(sock_net(sk), val);
1590 case NETLINK_BROADCAST_ERROR:
1592 nlk->flags |= NETLINK_F_BROADCAST_SEND_ERROR;
1594 nlk->flags &= ~NETLINK_F_BROADCAST_SEND_ERROR;
1597 case NETLINK_NO_ENOBUFS:
1599 nlk->flags |= NETLINK_F_RECV_NO_ENOBUFS;
1600 clear_bit(NETLINK_S_CONGESTED, &nlk->state);
1601 wake_up_interruptible(&nlk->wait);
1603 nlk->flags &= ~NETLINK_F_RECV_NO_ENOBUFS;
1607 case NETLINK_LISTEN_ALL_NSID:
1608 if (!ns_capable(sock_net(sk)->user_ns, CAP_NET_BROADCAST))
1612 nlk->flags |= NETLINK_F_LISTEN_ALL_NSID;
1614 nlk->flags &= ~NETLINK_F_LISTEN_ALL_NSID;
1617 case NETLINK_CAP_ACK:
1619 nlk->flags |= NETLINK_F_CAP_ACK;
1621 nlk->flags &= ~NETLINK_F_CAP_ACK;
1630 static int netlink_getsockopt(struct socket *sock, int level, int optname,
1631 char __user *optval, int __user *optlen)
1633 struct sock *sk = sock->sk;
1634 struct netlink_sock *nlk = nlk_sk(sk);
1637 if (level != SOL_NETLINK)
1638 return -ENOPROTOOPT;
1640 if (get_user(len, optlen))
1646 case NETLINK_PKTINFO:
1647 if (len < sizeof(int))
1650 val = nlk->flags & NETLINK_F_RECV_PKTINFO ? 1 : 0;
1651 if (put_user(len, optlen) ||
1652 put_user(val, optval))
1656 case NETLINK_BROADCAST_ERROR:
1657 if (len < sizeof(int))
1660 val = nlk->flags & NETLINK_F_BROADCAST_SEND_ERROR ? 1 : 0;
1661 if (put_user(len, optlen) ||
1662 put_user(val, optval))
1666 case NETLINK_NO_ENOBUFS:
1667 if (len < sizeof(int))
1670 val = nlk->flags & NETLINK_F_RECV_NO_ENOBUFS ? 1 : 0;
1671 if (put_user(len, optlen) ||
1672 put_user(val, optval))
1676 case NETLINK_LIST_MEMBERSHIPS: {
1677 int pos, idx, shift;
1680 netlink_lock_table();
1681 for (pos = 0; pos * 8 < nlk->ngroups; pos += sizeof(u32)) {
1682 if (len - pos < sizeof(u32))
1685 idx = pos / sizeof(unsigned long);
1686 shift = (pos % sizeof(unsigned long)) * 8;
1687 if (put_user((u32)(nlk->groups[idx] >> shift),
1688 (u32 __user *)(optval + pos))) {
1693 if (put_user(ALIGN(nlk->ngroups / 8, sizeof(u32)), optlen))
1695 netlink_unlock_table();
1698 case NETLINK_CAP_ACK:
1699 if (len < sizeof(int))
1702 val = nlk->flags & NETLINK_F_CAP_ACK ? 1 : 0;
1703 if (put_user(len, optlen) ||
1704 put_user(val, optval))
1714 static void netlink_cmsg_recv_pktinfo(struct msghdr *msg, struct sk_buff *skb)
1716 struct nl_pktinfo info;
1718 info.group = NETLINK_CB(skb).dst_group;
1719 put_cmsg(msg, SOL_NETLINK, NETLINK_PKTINFO, sizeof(info), &info);
1722 static void netlink_cmsg_listen_all_nsid(struct sock *sk, struct msghdr *msg,
1723 struct sk_buff *skb)
1725 if (!NETLINK_CB(skb).nsid_is_set)
1728 put_cmsg(msg, SOL_NETLINK, NETLINK_LISTEN_ALL_NSID, sizeof(int),
1729 &NETLINK_CB(skb).nsid);
1732 static int netlink_sendmsg(struct socket *sock, struct msghdr *msg, size_t len)
1734 struct sock *sk = sock->sk;
1735 struct netlink_sock *nlk = nlk_sk(sk);
1736 DECLARE_SOCKADDR(struct sockaddr_nl *, addr, msg->msg_name);
1739 struct sk_buff *skb;
1741 struct scm_cookie scm;
1742 u32 netlink_skb_flags = 0;
1744 if (msg->msg_flags&MSG_OOB)
1747 err = scm_send(sock, msg, &scm, true);
1751 if (msg->msg_namelen) {
1753 if (addr->nl_family != AF_NETLINK)
1755 dst_portid = addr->nl_pid;
1756 dst_group = ffs(addr->nl_groups);
1758 if ((dst_group || dst_portid) &&
1759 !netlink_allowed(sock, NL_CFG_F_NONROOT_SEND))
1761 netlink_skb_flags |= NETLINK_SKB_DST;
1763 dst_portid = nlk->dst_portid;
1764 dst_group = nlk->dst_group;
1768 err = netlink_autobind(sock);
1772 /* Ensure nlk is hashed and visible. */
1777 if (len > sk->sk_sndbuf - 32)
1780 skb = netlink_alloc_large_skb(len, dst_group);
1784 NETLINK_CB(skb).portid = nlk->portid;
1785 NETLINK_CB(skb).dst_group = dst_group;
1786 NETLINK_CB(skb).creds = scm.creds;
1787 NETLINK_CB(skb).flags = netlink_skb_flags;
1790 if (memcpy_from_msg(skb_put(skb, len), msg, len)) {
1795 err = security_netlink_send(sk, skb);
1802 atomic_inc(&skb->users);
1803 netlink_broadcast(sk, skb, dst_portid, dst_group, GFP_KERNEL);
1805 err = netlink_unicast(sk, skb, dst_portid, msg->msg_flags&MSG_DONTWAIT);
1812 static int netlink_recvmsg(struct socket *sock, struct msghdr *msg, size_t len,
1815 struct scm_cookie scm;
1816 struct sock *sk = sock->sk;
1817 struct netlink_sock *nlk = nlk_sk(sk);
1818 int noblock = flags&MSG_DONTWAIT;
1820 struct sk_buff *skb, *data_skb;
1828 skb = skb_recv_datagram(sk, flags, noblock, &err);
1834 #ifdef CONFIG_COMPAT_NETLINK_MESSAGES
1835 if (unlikely(skb_shinfo(skb)->frag_list)) {
1837 * If this skb has a frag_list, then here that means that we
1838 * will have to use the frag_list skb's data for compat tasks
1839 * and the regular skb's data for normal (non-compat) tasks.
1841 * If we need to send the compat skb, assign it to the
1842 * 'data_skb' variable so that it will be used below for data
1843 * copying. We keep 'skb' for everything else, including
1844 * freeing both later.
1846 if (flags & MSG_CMSG_COMPAT)
1847 data_skb = skb_shinfo(skb)->frag_list;
1851 /* Record the max length of recvmsg() calls for future allocations */
1852 nlk->max_recvmsg_len = max(nlk->max_recvmsg_len, len);
1853 nlk->max_recvmsg_len = min_t(size_t, nlk->max_recvmsg_len,
1854 SKB_WITH_OVERHEAD(32768));
1856 copied = data_skb->len;
1858 msg->msg_flags |= MSG_TRUNC;
1862 skb_reset_transport_header(data_skb);
1863 err = skb_copy_datagram_msg(data_skb, 0, msg, copied);
1865 if (msg->msg_name) {
1866 DECLARE_SOCKADDR(struct sockaddr_nl *, addr, msg->msg_name);
1867 addr->nl_family = AF_NETLINK;
1869 addr->nl_pid = NETLINK_CB(skb).portid;
1870 addr->nl_groups = netlink_group_mask(NETLINK_CB(skb).dst_group);
1871 msg->msg_namelen = sizeof(*addr);
1874 if (nlk->flags & NETLINK_F_RECV_PKTINFO)
1875 netlink_cmsg_recv_pktinfo(msg, skb);
1876 if (nlk->flags & NETLINK_F_LISTEN_ALL_NSID)
1877 netlink_cmsg_listen_all_nsid(sk, msg, skb);
1879 memset(&scm, 0, sizeof(scm));
1880 scm.creds = *NETLINK_CREDS(skb);
1881 if (flags & MSG_TRUNC)
1882 copied = data_skb->len;
1884 skb_free_datagram(sk, skb);
1886 if (nlk->cb_running &&
1887 atomic_read(&sk->sk_rmem_alloc) <= sk->sk_rcvbuf / 2) {
1888 ret = netlink_dump(sk);
1891 sk->sk_error_report(sk);
1895 scm_recv(sock, msg, &scm, flags);
1897 netlink_rcv_wake(sk);
1898 return err ? : copied;
1901 static void netlink_data_ready(struct sock *sk)
1907 * We export these functions to other modules. They provide a
1908 * complete set of kernel non-blocking support for message
1913 __netlink_kernel_create(struct net *net, int unit, struct module *module,
1914 struct netlink_kernel_cfg *cfg)
1916 struct socket *sock;
1918 struct netlink_sock *nlk;
1919 struct listeners *listeners = NULL;
1920 struct mutex *cb_mutex = cfg ? cfg->cb_mutex : NULL;
1921 unsigned int groups;
1925 if (unit < 0 || unit >= MAX_LINKS)
1928 if (sock_create_lite(PF_NETLINK, SOCK_DGRAM, unit, &sock))
1931 if (__netlink_create(net, sock, cb_mutex, unit, 1) < 0)
1932 goto out_sock_release_nosk;
1936 if (!cfg || cfg->groups < 32)
1939 groups = cfg->groups;
1941 listeners = kzalloc(sizeof(*listeners) + NLGRPSZ(groups), GFP_KERNEL);
1943 goto out_sock_release;
1945 sk->sk_data_ready = netlink_data_ready;
1946 if (cfg && cfg->input)
1947 nlk_sk(sk)->netlink_rcv = cfg->input;
1949 if (netlink_insert(sk, 0))
1950 goto out_sock_release;
1953 nlk->flags |= NETLINK_F_KERNEL_SOCKET;
1955 netlink_table_grab();
1956 if (!nl_table[unit].registered) {
1957 nl_table[unit].groups = groups;
1958 rcu_assign_pointer(nl_table[unit].listeners, listeners);
1959 nl_table[unit].cb_mutex = cb_mutex;
1960 nl_table[unit].module = module;
1962 nl_table[unit].bind = cfg->bind;
1963 nl_table[unit].unbind = cfg->unbind;
1964 nl_table[unit].flags = cfg->flags;
1966 nl_table[unit].compare = cfg->compare;
1968 nl_table[unit].registered = 1;
1971 nl_table[unit].registered++;
1973 netlink_table_ungrab();
1978 netlink_kernel_release(sk);
1981 out_sock_release_nosk:
1985 EXPORT_SYMBOL(__netlink_kernel_create);
1988 netlink_kernel_release(struct sock *sk)
1990 if (sk == NULL || sk->sk_socket == NULL)
1993 sock_release(sk->sk_socket);
1995 EXPORT_SYMBOL(netlink_kernel_release);
1997 int __netlink_change_ngroups(struct sock *sk, unsigned int groups)
1999 struct listeners *new, *old;
2000 struct netlink_table *tbl = &nl_table[sk->sk_protocol];
2005 if (NLGRPSZ(tbl->groups) < NLGRPSZ(groups)) {
2006 new = kzalloc(sizeof(*new) + NLGRPSZ(groups), GFP_ATOMIC);
2009 old = nl_deref_protected(tbl->listeners);
2010 memcpy(new->masks, old->masks, NLGRPSZ(tbl->groups));
2011 rcu_assign_pointer(tbl->listeners, new);
2013 kfree_rcu(old, rcu);
2015 tbl->groups = groups;
2021 * netlink_change_ngroups - change number of multicast groups
2023 * This changes the number of multicast groups that are available
2024 * on a certain netlink family. Note that it is not possible to
2025 * change the number of groups to below 32. Also note that it does
2026 * not implicitly call netlink_clear_multicast_users() when the
2027 * number of groups is reduced.
2029 * @sk: The kernel netlink socket, as returned by netlink_kernel_create().
2030 * @groups: The new number of groups.
2032 int netlink_change_ngroups(struct sock *sk, unsigned int groups)
2036 netlink_table_grab();
2037 err = __netlink_change_ngroups(sk, groups);
2038 netlink_table_ungrab();
2043 void __netlink_clear_multicast_users(struct sock *ksk, unsigned int group)
2046 struct netlink_table *tbl = &nl_table[ksk->sk_protocol];
2048 sk_for_each_bound(sk, &tbl->mc_list)
2049 netlink_update_socket_mc(nlk_sk(sk), group, 0);
2053 __nlmsg_put(struct sk_buff *skb, u32 portid, u32 seq, int type, int len, int flags)
2055 struct nlmsghdr *nlh;
2056 int size = nlmsg_msg_size(len);
2058 nlh = (struct nlmsghdr *)skb_put(skb, NLMSG_ALIGN(size));
2059 nlh->nlmsg_type = type;
2060 nlh->nlmsg_len = size;
2061 nlh->nlmsg_flags = flags;
2062 nlh->nlmsg_pid = portid;
2063 nlh->nlmsg_seq = seq;
2064 if (!__builtin_constant_p(size) || NLMSG_ALIGN(size) - size != 0)
2065 memset(nlmsg_data(nlh) + len, 0, NLMSG_ALIGN(size) - size);
2068 EXPORT_SYMBOL(__nlmsg_put);
2071 * It looks a bit ugly.
2072 * It would be better to create kernel thread.
2075 static int netlink_dump(struct sock *sk)
2077 struct netlink_sock *nlk = nlk_sk(sk);
2078 struct netlink_callback *cb;
2079 struct sk_buff *skb = NULL;
2080 struct nlmsghdr *nlh;
2081 struct module *module;
2082 int len, err = -ENOBUFS;
2086 mutex_lock(nlk->cb_mutex);
2087 if (!nlk->cb_running) {
2092 if (atomic_read(&sk->sk_rmem_alloc) >= sk->sk_rcvbuf)
2095 /* NLMSG_GOODSIZE is small to avoid high order allocations being
2096 * required, but it makes sense to _attempt_ a 16K bytes allocation
2097 * to reduce number of system calls on dump operations, if user
2098 * ever provided a big enough buffer.
2101 alloc_min_size = max_t(int, cb->min_dump_alloc, NLMSG_GOODSIZE);
2103 if (alloc_min_size < nlk->max_recvmsg_len) {
2104 alloc_size = nlk->max_recvmsg_len;
2105 skb = alloc_skb(alloc_size,
2106 (GFP_KERNEL & ~__GFP_DIRECT_RECLAIM) |
2107 __GFP_NOWARN | __GFP_NORETRY);
2110 alloc_size = alloc_min_size;
2111 skb = alloc_skb(alloc_size, GFP_KERNEL);
2116 /* Trim skb to allocated size. User is expected to provide buffer as
2117 * large as max(min_dump_alloc, 16KiB (mac_recvmsg_len capped at
2118 * netlink_recvmsg())). dump will pack as many smaller messages as
2119 * could fit within the allocated skb. skb is typically allocated
2120 * with larger space than required (could be as much as near 2x the
2121 * requested size with align to next power of 2 approach). Allowing
2122 * dump to use the excess space makes it difficult for a user to have a
2123 * reasonable static buffer based on the expected largest dump of a
2124 * single netdev. The outcome is MSG_TRUNC error.
2126 skb_reserve(skb, skb_tailroom(skb) - alloc_size);
2127 netlink_skb_set_owner_r(skb, sk);
2129 len = cb->dump(skb, cb);
2132 mutex_unlock(nlk->cb_mutex);
2134 if (sk_filter(sk, skb))
2137 __netlink_sendskb(sk, skb);
2141 nlh = nlmsg_put_answer(skb, cb, NLMSG_DONE, sizeof(len), NLM_F_MULTI);
2145 nl_dump_check_consistent(cb, nlh);
2147 memcpy(nlmsg_data(nlh), &len, sizeof(len));
2149 if (sk_filter(sk, skb))
2152 __netlink_sendskb(sk, skb);
2157 nlk->cb_running = false;
2158 module = cb->module;
2160 mutex_unlock(nlk->cb_mutex);
2166 mutex_unlock(nlk->cb_mutex);
2171 int __netlink_dump_start(struct sock *ssk, struct sk_buff *skb,
2172 const struct nlmsghdr *nlh,
2173 struct netlink_dump_control *control)
2175 struct netlink_callback *cb;
2177 struct netlink_sock *nlk;
2180 atomic_inc(&skb->users);
2182 sk = netlink_lookup(sock_net(ssk), ssk->sk_protocol, NETLINK_CB(skb).portid);
2184 ret = -ECONNREFUSED;
2189 mutex_lock(nlk->cb_mutex);
2190 /* A dump is in progress... */
2191 if (nlk->cb_running) {
2195 /* add reference of module which cb->dump belongs to */
2196 if (!try_module_get(control->module)) {
2197 ret = -EPROTONOSUPPORT;
2202 memset(cb, 0, sizeof(*cb));
2203 cb->start = control->start;
2204 cb->dump = control->dump;
2205 cb->done = control->done;
2207 cb->data = control->data;
2208 cb->module = control->module;
2209 cb->min_dump_alloc = control->min_dump_alloc;
2212 nlk->cb_running = true;
2214 mutex_unlock(nlk->cb_mutex);
2219 ret = netlink_dump(sk);
2225 /* We successfully started a dump, by returning -EINTR we
2226 * signal not to send ACK even if it was requested.
2232 mutex_unlock(nlk->cb_mutex);
2237 EXPORT_SYMBOL(__netlink_dump_start);
2239 void netlink_ack(struct sk_buff *in_skb, struct nlmsghdr *nlh, int err)
2241 struct sk_buff *skb;
2242 struct nlmsghdr *rep;
2243 struct nlmsgerr *errmsg;
2244 size_t payload = sizeof(*errmsg);
2245 struct netlink_sock *nlk = nlk_sk(NETLINK_CB(in_skb).sk);
2247 /* Error messages get the original request appened, unless the user
2248 * requests to cap the error message.
2250 if (!(nlk->flags & NETLINK_F_CAP_ACK) && err)
2251 payload += nlmsg_len(nlh);
2253 skb = nlmsg_new(payload, GFP_KERNEL);
2257 sk = netlink_lookup(sock_net(in_skb->sk),
2258 in_skb->sk->sk_protocol,
2259 NETLINK_CB(in_skb).portid);
2261 sk->sk_err = ENOBUFS;
2262 sk->sk_error_report(sk);
2268 rep = __nlmsg_put(skb, NETLINK_CB(in_skb).portid, nlh->nlmsg_seq,
2269 NLMSG_ERROR, payload, 0);
2270 errmsg = nlmsg_data(rep);
2271 errmsg->error = err;
2272 memcpy(&errmsg->msg, nlh, payload > sizeof(*errmsg) ? nlh->nlmsg_len : sizeof(*nlh));
2273 netlink_unicast(in_skb->sk, skb, NETLINK_CB(in_skb).portid, MSG_DONTWAIT);
2275 EXPORT_SYMBOL(netlink_ack);
2277 int netlink_rcv_skb(struct sk_buff *skb, int (*cb)(struct sk_buff *,
2280 struct nlmsghdr *nlh;
2283 while (skb->len >= nlmsg_total_size(0)) {
2286 nlh = nlmsg_hdr(skb);
2289 if (nlh->nlmsg_len < NLMSG_HDRLEN || skb->len < nlh->nlmsg_len)
2292 /* Only requests are handled by the kernel */
2293 if (!(nlh->nlmsg_flags & NLM_F_REQUEST))
2296 /* Skip control messages */
2297 if (nlh->nlmsg_type < NLMSG_MIN_TYPE)
2305 if (nlh->nlmsg_flags & NLM_F_ACK || err)
2306 netlink_ack(skb, nlh, err);
2309 msglen = NLMSG_ALIGN(nlh->nlmsg_len);
2310 if (msglen > skb->len)
2312 skb_pull(skb, msglen);
2317 EXPORT_SYMBOL(netlink_rcv_skb);
2320 * nlmsg_notify - send a notification netlink message
2321 * @sk: netlink socket to use
2322 * @skb: notification message
2323 * @portid: destination netlink portid for reports or 0
2324 * @group: destination multicast group or 0
2325 * @report: 1 to report back, 0 to disable
2326 * @flags: allocation flags
2328 int nlmsg_notify(struct sock *sk, struct sk_buff *skb, u32 portid,
2329 unsigned int group, int report, gfp_t flags)
2334 int exclude_portid = 0;
2337 atomic_inc(&skb->users);
2338 exclude_portid = portid;
2341 /* errors reported via destination sk->sk_err, but propagate
2342 * delivery errors if NETLINK_BROADCAST_ERROR flag is set */
2343 err = nlmsg_multicast(sk, skb, exclude_portid, group, flags);
2349 err2 = nlmsg_unicast(sk, skb, portid);
2350 if (!err || err == -ESRCH)
2356 EXPORT_SYMBOL(nlmsg_notify);
2358 #ifdef CONFIG_PROC_FS
2359 struct nl_seq_iter {
2360 struct seq_net_private p;
2361 struct rhashtable_iter hti;
2365 static int netlink_walk_start(struct nl_seq_iter *iter)
2369 err = rhashtable_walk_init(&nl_table[iter->link].hash, &iter->hti,
2372 iter->link = MAX_LINKS;
2376 err = rhashtable_walk_start(&iter->hti);
2377 return err == -EAGAIN ? 0 : err;
2380 static void netlink_walk_stop(struct nl_seq_iter *iter)
2382 rhashtable_walk_stop(&iter->hti);
2383 rhashtable_walk_exit(&iter->hti);
2386 static void *__netlink_seq_next(struct seq_file *seq)
2388 struct nl_seq_iter *iter = seq->private;
2389 struct netlink_sock *nlk;
2395 nlk = rhashtable_walk_next(&iter->hti);
2398 if (PTR_ERR(nlk) == -EAGAIN)
2407 netlink_walk_stop(iter);
2408 if (++iter->link >= MAX_LINKS)
2411 err = netlink_walk_start(iter);
2413 return ERR_PTR(err);
2415 } while (sock_net(&nlk->sk) != seq_file_net(seq));
2420 static void *netlink_seq_start(struct seq_file *seq, loff_t *posp)
2422 struct nl_seq_iter *iter = seq->private;
2423 void *obj = SEQ_START_TOKEN;
2429 err = netlink_walk_start(iter);
2431 return ERR_PTR(err);
2433 for (pos = *posp; pos && obj && !IS_ERR(obj); pos--)
2434 obj = __netlink_seq_next(seq);
2439 static void *netlink_seq_next(struct seq_file *seq, void *v, loff_t *pos)
2442 return __netlink_seq_next(seq);
2445 static void netlink_seq_stop(struct seq_file *seq, void *v)
2447 struct nl_seq_iter *iter = seq->private;
2449 if (iter->link >= MAX_LINKS)
2452 netlink_walk_stop(iter);
2456 static int netlink_seq_show(struct seq_file *seq, void *v)
2458 if (v == SEQ_START_TOKEN) {
2460 "sk Eth Pid Groups "
2461 "Rmem Wmem Dump Locks Drops Inode\n");
2464 struct netlink_sock *nlk = nlk_sk(s);
2466 seq_printf(seq, "%pK %-3d %-6u %08x %-8d %-8d %d %-8d %-8d %-8lu\n",
2470 nlk->groups ? (u32)nlk->groups[0] : 0,
2471 sk_rmem_alloc_get(s),
2472 sk_wmem_alloc_get(s),
2474 atomic_read(&s->sk_refcnt),
2475 atomic_read(&s->sk_drops),
2483 static const struct seq_operations netlink_seq_ops = {
2484 .start = netlink_seq_start,
2485 .next = netlink_seq_next,
2486 .stop = netlink_seq_stop,
2487 .show = netlink_seq_show,
2491 static int netlink_seq_open(struct inode *inode, struct file *file)
2493 return seq_open_net(inode, file, &netlink_seq_ops,
2494 sizeof(struct nl_seq_iter));
2497 static const struct file_operations netlink_seq_fops = {
2498 .owner = THIS_MODULE,
2499 .open = netlink_seq_open,
2501 .llseek = seq_lseek,
2502 .release = seq_release_net,
2507 int netlink_register_notifier(struct notifier_block *nb)
2509 return atomic_notifier_chain_register(&netlink_chain, nb);
2511 EXPORT_SYMBOL(netlink_register_notifier);
2513 int netlink_unregister_notifier(struct notifier_block *nb)
2515 return atomic_notifier_chain_unregister(&netlink_chain, nb);
2517 EXPORT_SYMBOL(netlink_unregister_notifier);
2519 static const struct proto_ops netlink_ops = {
2520 .family = PF_NETLINK,
2521 .owner = THIS_MODULE,
2522 .release = netlink_release,
2523 .bind = netlink_bind,
2524 .connect = netlink_connect,
2525 .socketpair = sock_no_socketpair,
2526 .accept = sock_no_accept,
2527 .getname = netlink_getname,
2528 .poll = datagram_poll,
2529 .ioctl = netlink_ioctl,
2530 .listen = sock_no_listen,
2531 .shutdown = sock_no_shutdown,
2532 .setsockopt = netlink_setsockopt,
2533 .getsockopt = netlink_getsockopt,
2534 .sendmsg = netlink_sendmsg,
2535 .recvmsg = netlink_recvmsg,
2536 .mmap = sock_no_mmap,
2537 .sendpage = sock_no_sendpage,
2540 static const struct net_proto_family netlink_family_ops = {
2541 .family = PF_NETLINK,
2542 .create = netlink_create,
2543 .owner = THIS_MODULE, /* for consistency 8) */
2546 static int __net_init netlink_net_init(struct net *net)
2548 #ifdef CONFIG_PROC_FS
2549 if (!proc_create("netlink", 0, net->proc_net, &netlink_seq_fops))
2555 static void __net_exit netlink_net_exit(struct net *net)
2557 #ifdef CONFIG_PROC_FS
2558 remove_proc_entry("netlink", net->proc_net);
2562 static void __init netlink_add_usersock_entry(void)
2564 struct listeners *listeners;
2567 listeners = kzalloc(sizeof(*listeners) + NLGRPSZ(groups), GFP_KERNEL);
2569 panic("netlink_add_usersock_entry: Cannot allocate listeners\n");
2571 netlink_table_grab();
2573 nl_table[NETLINK_USERSOCK].groups = groups;
2574 rcu_assign_pointer(nl_table[NETLINK_USERSOCK].listeners, listeners);
2575 nl_table[NETLINK_USERSOCK].module = THIS_MODULE;
2576 nl_table[NETLINK_USERSOCK].registered = 1;
2577 nl_table[NETLINK_USERSOCK].flags = NL_CFG_F_NONROOT_SEND;
2579 netlink_table_ungrab();
2582 static struct pernet_operations __net_initdata netlink_net_ops = {
2583 .init = netlink_net_init,
2584 .exit = netlink_net_exit,
2587 static inline u32 netlink_hash(const void *data, u32 len, u32 seed)
2589 const struct netlink_sock *nlk = data;
2590 struct netlink_compare_arg arg;
2592 netlink_compare_arg_init(&arg, sock_net(&nlk->sk), nlk->portid);
2593 return jhash2((u32 *)&arg, netlink_compare_arg_len / sizeof(u32), seed);
2596 static const struct rhashtable_params netlink_rhashtable_params = {
2597 .head_offset = offsetof(struct netlink_sock, node),
2598 .key_len = netlink_compare_arg_len,
2599 .obj_hashfn = netlink_hash,
2600 .obj_cmpfn = netlink_compare,
2601 .automatic_shrinking = true,
2604 static int __init netlink_proto_init(void)
2607 int err = proto_register(&netlink_proto, 0);
2612 BUILD_BUG_ON(sizeof(struct netlink_skb_parms) > FIELD_SIZEOF(struct sk_buff, cb));
2614 nl_table = kcalloc(MAX_LINKS, sizeof(*nl_table), GFP_KERNEL);
2618 for (i = 0; i < MAX_LINKS; i++) {
2619 if (rhashtable_init(&nl_table[i].hash,
2620 &netlink_rhashtable_params) < 0) {
2622 rhashtable_destroy(&nl_table[i].hash);
2628 INIT_LIST_HEAD(&netlink_tap_all);
2630 netlink_add_usersock_entry();
2632 sock_register(&netlink_family_ops);
2633 register_pernet_subsys(&netlink_net_ops);
2634 /* The netlink device handler may be needed early. */
2639 panic("netlink_init: Cannot allocate nl_table\n");
2642 core_initcall(netlink_proto_init);