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) {
331 nlk->cb.done(&nlk->cb);
333 module_put(nlk->cb.module);
334 kfree_skb(nlk->cb.skb);
337 skb_queue_purge(&sk->sk_receive_queue);
339 if (!sock_flag(sk, SOCK_DEAD)) {
340 printk(KERN_ERR "Freeing alive netlink socket %p\n", sk);
344 WARN_ON(atomic_read(&sk->sk_rmem_alloc));
345 WARN_ON(atomic_read(&sk->sk_wmem_alloc));
346 WARN_ON(nlk_sk(sk)->groups);
349 /* This lock without WQ_FLAG_EXCLUSIVE is good on UP and it is _very_ bad on
350 * SMP. Look, when several writers sleep and reader wakes them up, all but one
351 * immediately hit write lock and grab all the cpus. Exclusive sleep solves
352 * this, _but_ remember, it adds useless work on UP machines.
355 void netlink_table_grab(void)
356 __acquires(nl_table_lock)
360 write_lock_irq(&nl_table_lock);
362 if (atomic_read(&nl_table_users)) {
363 DECLARE_WAITQUEUE(wait, current);
365 add_wait_queue_exclusive(&nl_table_wait, &wait);
367 set_current_state(TASK_UNINTERRUPTIBLE);
368 if (atomic_read(&nl_table_users) == 0)
370 write_unlock_irq(&nl_table_lock);
372 write_lock_irq(&nl_table_lock);
375 __set_current_state(TASK_RUNNING);
376 remove_wait_queue(&nl_table_wait, &wait);
380 void netlink_table_ungrab(void)
381 __releases(nl_table_lock)
383 write_unlock_irq(&nl_table_lock);
384 wake_up(&nl_table_wait);
388 netlink_lock_table(void)
390 /* read_lock() synchronizes us to netlink_table_grab */
392 read_lock(&nl_table_lock);
393 atomic_inc(&nl_table_users);
394 read_unlock(&nl_table_lock);
398 netlink_unlock_table(void)
400 if (atomic_dec_and_test(&nl_table_users))
401 wake_up(&nl_table_wait);
404 struct netlink_compare_arg
410 /* Doing sizeof directly may yield 4 extra bytes on 64-bit. */
411 #define netlink_compare_arg_len \
412 (offsetof(struct netlink_compare_arg, portid) + sizeof(u32))
414 static inline int netlink_compare(struct rhashtable_compare_arg *arg,
417 const struct netlink_compare_arg *x = arg->key;
418 const struct netlink_sock *nlk = ptr;
420 return nlk->portid != x->portid ||
421 !net_eq(sock_net(&nlk->sk), read_pnet(&x->pnet));
424 static void netlink_compare_arg_init(struct netlink_compare_arg *arg,
425 struct net *net, u32 portid)
427 memset(arg, 0, sizeof(*arg));
428 write_pnet(&arg->pnet, net);
429 arg->portid = portid;
432 static struct sock *__netlink_lookup(struct netlink_table *table, u32 portid,
435 struct netlink_compare_arg arg;
437 netlink_compare_arg_init(&arg, net, portid);
438 return rhashtable_lookup_fast(&table->hash, &arg,
439 netlink_rhashtable_params);
442 static int __netlink_insert(struct netlink_table *table, struct sock *sk)
444 struct netlink_compare_arg arg;
446 netlink_compare_arg_init(&arg, sock_net(sk), nlk_sk(sk)->portid);
447 return rhashtable_lookup_insert_key(&table->hash, &arg,
449 netlink_rhashtable_params);
452 static struct sock *netlink_lookup(struct net *net, int protocol, u32 portid)
454 struct netlink_table *table = &nl_table[protocol];
458 sk = __netlink_lookup(table, portid, net);
466 static const struct proto_ops netlink_ops;
469 netlink_update_listeners(struct sock *sk)
471 struct netlink_table *tbl = &nl_table[sk->sk_protocol];
474 struct listeners *listeners;
476 listeners = nl_deref_protected(tbl->listeners);
480 for (i = 0; i < NLGRPLONGS(tbl->groups); i++) {
482 sk_for_each_bound(sk, &tbl->mc_list) {
483 if (i < NLGRPLONGS(nlk_sk(sk)->ngroups))
484 mask |= nlk_sk(sk)->groups[i];
486 listeners->masks[i] = mask;
488 /* this function is only called with the netlink table "grabbed", which
489 * makes sure updates are visible before bind or setsockopt return. */
492 static int netlink_insert(struct sock *sk, u32 portid)
494 struct netlink_table *table = &nl_table[sk->sk_protocol];
499 err = nlk_sk(sk)->portid == portid ? 0 : -EBUSY;
500 if (nlk_sk(sk)->bound)
504 if (BITS_PER_LONG > 32 &&
505 unlikely(atomic_read(&table->hash.nelems) >= UINT_MAX))
508 nlk_sk(sk)->portid = portid;
511 err = __netlink_insert(table, sk);
513 /* In case the hashtable backend returns with -EBUSY
514 * from here, it must not escape to the caller.
516 if (unlikely(err == -EBUSY))
524 /* We need to ensure that the socket is hashed and visible. */
526 nlk_sk(sk)->bound = portid;
533 static void netlink_remove(struct sock *sk)
535 struct netlink_table *table;
537 table = &nl_table[sk->sk_protocol];
538 if (!rhashtable_remove_fast(&table->hash, &nlk_sk(sk)->node,
539 netlink_rhashtable_params)) {
540 WARN_ON(atomic_read(&sk->sk_refcnt) == 1);
544 netlink_table_grab();
545 if (nlk_sk(sk)->subscriptions) {
546 __sk_del_bind_node(sk);
547 netlink_update_listeners(sk);
549 if (sk->sk_protocol == NETLINK_GENERIC)
550 atomic_inc(&genl_sk_destructing_cnt);
551 netlink_table_ungrab();
554 static struct proto netlink_proto = {
556 .owner = THIS_MODULE,
557 .obj_size = sizeof(struct netlink_sock),
560 static int __netlink_create(struct net *net, struct socket *sock,
561 struct mutex *cb_mutex, int protocol,
565 struct netlink_sock *nlk;
567 sock->ops = &netlink_ops;
569 sk = sk_alloc(net, PF_NETLINK, GFP_KERNEL, &netlink_proto, kern);
573 sock_init_data(sock, sk);
577 nlk->cb_mutex = cb_mutex;
579 nlk->cb_mutex = &nlk->cb_def_mutex;
580 mutex_init(nlk->cb_mutex);
582 init_waitqueue_head(&nlk->wait);
584 sk->sk_destruct = netlink_sock_destruct;
585 sk->sk_protocol = protocol;
589 static int netlink_create(struct net *net, struct socket *sock, int protocol,
592 struct module *module = NULL;
593 struct mutex *cb_mutex;
594 struct netlink_sock *nlk;
595 int (*bind)(struct net *net, int group);
596 void (*unbind)(struct net *net, int group);
599 sock->state = SS_UNCONNECTED;
601 if (sock->type != SOCK_RAW && sock->type != SOCK_DGRAM)
602 return -ESOCKTNOSUPPORT;
604 if (protocol < 0 || protocol >= MAX_LINKS)
605 return -EPROTONOSUPPORT;
607 netlink_lock_table();
608 #ifdef CONFIG_MODULES
609 if (!nl_table[protocol].registered) {
610 netlink_unlock_table();
611 request_module("net-pf-%d-proto-%d", PF_NETLINK, protocol);
612 netlink_lock_table();
615 if (nl_table[protocol].registered &&
616 try_module_get(nl_table[protocol].module))
617 module = nl_table[protocol].module;
619 err = -EPROTONOSUPPORT;
620 cb_mutex = nl_table[protocol].cb_mutex;
621 bind = nl_table[protocol].bind;
622 unbind = nl_table[protocol].unbind;
623 netlink_unlock_table();
628 err = __netlink_create(net, sock, cb_mutex, protocol, kern);
633 sock_prot_inuse_add(net, &netlink_proto, 1);
636 nlk = nlk_sk(sock->sk);
637 nlk->module = module;
638 nlk->netlink_bind = bind;
639 nlk->netlink_unbind = unbind;
648 static void deferred_put_nlk_sk(struct rcu_head *head)
650 struct netlink_sock *nlk = container_of(head, struct netlink_sock, rcu);
655 static int netlink_release(struct socket *sock)
657 struct sock *sk = sock->sk;
658 struct netlink_sock *nlk;
668 * OK. Socket is unlinked, any packets that arrive now
672 /* must not acquire netlink_table_lock in any way again before unbind
673 * and notifying genetlink is done as otherwise it might deadlock
675 if (nlk->netlink_unbind) {
678 for (i = 0; i < nlk->ngroups; i++)
679 if (test_bit(i, nlk->groups))
680 nlk->netlink_unbind(sock_net(sk), i + 1);
682 if (sk->sk_protocol == NETLINK_GENERIC &&
683 atomic_dec_return(&genl_sk_destructing_cnt) == 0)
684 wake_up(&genl_sk_destructing_waitq);
687 wake_up_interruptible_all(&nlk->wait);
689 skb_queue_purge(&sk->sk_write_queue);
692 struct netlink_notify n = {
694 .protocol = sk->sk_protocol,
695 .portid = nlk->portid,
697 atomic_notifier_call_chain(&netlink_chain,
698 NETLINK_URELEASE, &n);
701 module_put(nlk->module);
703 if (netlink_is_kernel(sk)) {
704 netlink_table_grab();
705 BUG_ON(nl_table[sk->sk_protocol].registered == 0);
706 if (--nl_table[sk->sk_protocol].registered == 0) {
707 struct listeners *old;
709 old = nl_deref_protected(nl_table[sk->sk_protocol].listeners);
710 RCU_INIT_POINTER(nl_table[sk->sk_protocol].listeners, NULL);
712 nl_table[sk->sk_protocol].module = NULL;
713 nl_table[sk->sk_protocol].bind = NULL;
714 nl_table[sk->sk_protocol].unbind = NULL;
715 nl_table[sk->sk_protocol].flags = 0;
716 nl_table[sk->sk_protocol].registered = 0;
718 netlink_table_ungrab();
725 sock_prot_inuse_add(sock_net(sk), &netlink_proto, -1);
727 call_rcu(&nlk->rcu, deferred_put_nlk_sk);
731 static int netlink_autobind(struct socket *sock)
733 struct sock *sk = sock->sk;
734 struct net *net = sock_net(sk);
735 struct netlink_table *table = &nl_table[sk->sk_protocol];
736 s32 portid = task_tgid_vnr(current);
744 ok = !__netlink_lookup(table, portid, net);
747 /* Bind collision, search negative portid values. */
749 /* rover will be in range [S32_MIN, -4097] */
750 rover = S32_MIN + prandom_u32_max(-4096 - S32_MIN);
751 else if (rover >= -4096)
757 err = netlink_insert(sk, portid);
758 if (err == -EADDRINUSE)
761 /* If 2 threads race to autobind, that is fine. */
769 * __netlink_ns_capable - General netlink message capability test
770 * @nsp: NETLINK_CB of the socket buffer holding a netlink command from userspace.
771 * @user_ns: The user namespace of the capability to use
772 * @cap: The capability to use
774 * Test to see if the opener of the socket we received the message
775 * from had when the netlink socket was created and the sender of the
776 * message has has the capability @cap in the user namespace @user_ns.
778 bool __netlink_ns_capable(const struct netlink_skb_parms *nsp,
779 struct user_namespace *user_ns, int cap)
781 return ((nsp->flags & NETLINK_SKB_DST) ||
782 file_ns_capable(nsp->sk->sk_socket->file, user_ns, cap)) &&
783 ns_capable(user_ns, cap);
785 EXPORT_SYMBOL(__netlink_ns_capable);
788 * netlink_ns_capable - General netlink message capability test
789 * @skb: 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 sk_buff *skb,
798 struct user_namespace *user_ns, int cap)
800 return __netlink_ns_capable(&NETLINK_CB(skb), user_ns, cap);
802 EXPORT_SYMBOL(netlink_ns_capable);
805 * netlink_capable - Netlink global message capability test
806 * @skb: socket buffer holding a netlink command from userspace
807 * @cap: The capability to use
809 * Test to see if the opener of the socket we received the message
810 * from had when the netlink socket was created and the sender of the
811 * message has has the capability @cap in all user namespaces.
813 bool netlink_capable(const struct sk_buff *skb, int cap)
815 return netlink_ns_capable(skb, &init_user_ns, cap);
817 EXPORT_SYMBOL(netlink_capable);
820 * netlink_net_capable - Netlink network namespace message capability test
821 * @skb: socket buffer holding a netlink command from userspace
822 * @cap: The capability to use
824 * Test to see if the opener of the socket we received the message
825 * from had when the netlink socket was created and the sender of the
826 * message has has the capability @cap over the network namespace of
827 * the socket we received the message from.
829 bool netlink_net_capable(const struct sk_buff *skb, int cap)
831 return netlink_ns_capable(skb, sock_net(skb->sk)->user_ns, cap);
833 EXPORT_SYMBOL(netlink_net_capable);
835 static inline int netlink_allowed(const struct socket *sock, unsigned int flag)
837 return (nl_table[sock->sk->sk_protocol].flags & flag) ||
838 ns_capable(sock_net(sock->sk)->user_ns, CAP_NET_ADMIN);
842 netlink_update_subscriptions(struct sock *sk, unsigned int subscriptions)
844 struct netlink_sock *nlk = nlk_sk(sk);
846 if (nlk->subscriptions && !subscriptions)
847 __sk_del_bind_node(sk);
848 else if (!nlk->subscriptions && subscriptions)
849 sk_add_bind_node(sk, &nl_table[sk->sk_protocol].mc_list);
850 nlk->subscriptions = subscriptions;
853 static int netlink_realloc_groups(struct sock *sk)
855 struct netlink_sock *nlk = nlk_sk(sk);
857 unsigned long *new_groups;
860 netlink_table_grab();
862 groups = nl_table[sk->sk_protocol].groups;
863 if (!nl_table[sk->sk_protocol].registered) {
868 if (nlk->ngroups >= groups)
871 new_groups = krealloc(nlk->groups, NLGRPSZ(groups), GFP_ATOMIC);
872 if (new_groups == NULL) {
876 memset((char *)new_groups + NLGRPSZ(nlk->ngroups), 0,
877 NLGRPSZ(groups) - NLGRPSZ(nlk->ngroups));
879 nlk->groups = new_groups;
880 nlk->ngroups = groups;
882 netlink_table_ungrab();
886 static void netlink_undo_bind(int group, long unsigned int groups,
889 struct netlink_sock *nlk = nlk_sk(sk);
892 if (!nlk->netlink_unbind)
895 for (undo = 0; undo < group; undo++)
896 if (test_bit(undo, &groups))
897 nlk->netlink_unbind(sock_net(sk), undo + 1);
900 static int netlink_bind(struct socket *sock, struct sockaddr *addr,
903 struct sock *sk = sock->sk;
904 struct net *net = sock_net(sk);
905 struct netlink_sock *nlk = nlk_sk(sk);
906 struct sockaddr_nl *nladdr = (struct sockaddr_nl *)addr;
908 long unsigned int groups = nladdr->nl_groups;
911 if (addr_len < sizeof(struct sockaddr_nl))
914 if (nladdr->nl_family != AF_NETLINK)
917 /* Only superuser is allowed to listen multicasts */
919 if (!netlink_allowed(sock, NL_CFG_F_NONROOT_RECV))
921 err = netlink_realloc_groups(sk);
928 /* Ensure nlk->portid is up-to-date. */
931 if (nladdr->nl_pid != nlk->portid)
935 if (nlk->netlink_bind && groups) {
938 for (group = 0; group < nlk->ngroups; group++) {
939 if (!test_bit(group, &groups))
941 err = nlk->netlink_bind(net, group + 1);
944 netlink_undo_bind(group, groups, sk);
949 /* No need for barriers here as we return to user-space without
950 * using any of the bound attributes.
953 err = nladdr->nl_pid ?
954 netlink_insert(sk, nladdr->nl_pid) :
955 netlink_autobind(sock);
957 netlink_undo_bind(nlk->ngroups, groups, sk);
962 if (!groups && (nlk->groups == NULL || !(u32)nlk->groups[0]))
965 netlink_table_grab();
966 netlink_update_subscriptions(sk, nlk->subscriptions +
968 hweight32(nlk->groups[0]));
969 nlk->groups[0] = (nlk->groups[0] & ~0xffffffffUL) | groups;
970 netlink_update_listeners(sk);
971 netlink_table_ungrab();
976 static int netlink_connect(struct socket *sock, struct sockaddr *addr,
980 struct sock *sk = sock->sk;
981 struct netlink_sock *nlk = nlk_sk(sk);
982 struct sockaddr_nl *nladdr = (struct sockaddr_nl *)addr;
984 if (alen < sizeof(addr->sa_family))
987 if (addr->sa_family == AF_UNSPEC) {
988 sk->sk_state = NETLINK_UNCONNECTED;
993 if (addr->sa_family != AF_NETLINK)
996 if ((nladdr->nl_groups || nladdr->nl_pid) &&
997 !netlink_allowed(sock, NL_CFG_F_NONROOT_SEND))
1000 /* No need for barriers here as we return to user-space without
1001 * using any of the bound attributes.
1004 err = netlink_autobind(sock);
1007 sk->sk_state = NETLINK_CONNECTED;
1008 nlk->dst_portid = nladdr->nl_pid;
1009 nlk->dst_group = ffs(nladdr->nl_groups);
1015 static int netlink_getname(struct socket *sock, struct sockaddr *addr,
1016 int *addr_len, int peer)
1018 struct sock *sk = sock->sk;
1019 struct netlink_sock *nlk = nlk_sk(sk);
1020 DECLARE_SOCKADDR(struct sockaddr_nl *, nladdr, addr);
1022 nladdr->nl_family = AF_NETLINK;
1024 *addr_len = sizeof(*nladdr);
1027 nladdr->nl_pid = nlk->dst_portid;
1028 nladdr->nl_groups = netlink_group_mask(nlk->dst_group);
1030 nladdr->nl_pid = nlk->portid;
1031 nladdr->nl_groups = nlk->groups ? nlk->groups[0] : 0;
1036 static struct sock *netlink_getsockbyportid(struct sock *ssk, u32 portid)
1039 struct netlink_sock *nlk;
1041 sock = netlink_lookup(sock_net(ssk), ssk->sk_protocol, portid);
1043 return ERR_PTR(-ECONNREFUSED);
1045 /* Don't bother queuing skb if kernel socket has no input function */
1047 if (sock->sk_state == NETLINK_CONNECTED &&
1048 nlk->dst_portid != nlk_sk(ssk)->portid) {
1050 return ERR_PTR(-ECONNREFUSED);
1055 struct sock *netlink_getsockbyfilp(struct file *filp)
1057 struct inode *inode = file_inode(filp);
1060 if (!S_ISSOCK(inode->i_mode))
1061 return ERR_PTR(-ENOTSOCK);
1063 sock = SOCKET_I(inode)->sk;
1064 if (sock->sk_family != AF_NETLINK)
1065 return ERR_PTR(-EINVAL);
1071 static struct sk_buff *netlink_alloc_large_skb(unsigned int size,
1074 struct sk_buff *skb;
1077 if (size <= NLMSG_GOODSIZE || broadcast)
1078 return alloc_skb(size, GFP_KERNEL);
1080 size = SKB_DATA_ALIGN(size) +
1081 SKB_DATA_ALIGN(sizeof(struct skb_shared_info));
1083 data = vmalloc(size);
1087 skb = __build_skb(data, size);
1091 skb->destructor = netlink_skb_destructor;
1097 * Attach a skb to a netlink socket.
1098 * The caller must hold a reference to the destination socket. On error, the
1099 * reference is dropped. The skb is not send to the destination, just all
1100 * all error checks are performed and memory in the queue is reserved.
1102 * < 0: error. skb freed, reference to sock dropped.
1104 * 1: repeat lookup - reference dropped while waiting for socket memory.
1106 int netlink_attachskb(struct sock *sk, struct sk_buff *skb,
1107 long *timeo, struct sock *ssk)
1109 struct netlink_sock *nlk;
1113 if ((atomic_read(&sk->sk_rmem_alloc) > sk->sk_rcvbuf ||
1114 test_bit(NETLINK_S_CONGESTED, &nlk->state))) {
1115 DECLARE_WAITQUEUE(wait, current);
1117 if (!ssk || netlink_is_kernel(ssk))
1118 netlink_overrun(sk);
1124 __set_current_state(TASK_INTERRUPTIBLE);
1125 add_wait_queue(&nlk->wait, &wait);
1127 if ((atomic_read(&sk->sk_rmem_alloc) > sk->sk_rcvbuf ||
1128 test_bit(NETLINK_S_CONGESTED, &nlk->state)) &&
1129 !sock_flag(sk, SOCK_DEAD))
1130 *timeo = schedule_timeout(*timeo);
1132 __set_current_state(TASK_RUNNING);
1133 remove_wait_queue(&nlk->wait, &wait);
1136 if (signal_pending(current)) {
1138 return sock_intr_errno(*timeo);
1142 netlink_skb_set_owner_r(skb, sk);
1146 static int __netlink_sendskb(struct sock *sk, struct sk_buff *skb)
1150 netlink_deliver_tap(skb);
1152 skb_queue_tail(&sk->sk_receive_queue, skb);
1153 sk->sk_data_ready(sk);
1157 int netlink_sendskb(struct sock *sk, struct sk_buff *skb)
1159 int len = __netlink_sendskb(sk, skb);
1165 void netlink_detachskb(struct sock *sk, struct sk_buff *skb)
1171 static struct sk_buff *netlink_trim(struct sk_buff *skb, gfp_t allocation)
1175 WARN_ON(skb->sk != NULL);
1176 delta = skb->end - skb->tail;
1177 if (is_vmalloc_addr(skb->head) || delta * 2 < skb->truesize)
1180 if (skb_shared(skb)) {
1181 struct sk_buff *nskb = skb_clone(skb, allocation);
1188 if (!pskb_expand_head(skb, 0, -delta, allocation))
1189 skb->truesize -= delta;
1194 static int netlink_unicast_kernel(struct sock *sk, struct sk_buff *skb,
1198 struct netlink_sock *nlk = nlk_sk(sk);
1200 ret = -ECONNREFUSED;
1201 if (nlk->netlink_rcv != NULL) {
1203 netlink_skb_set_owner_r(skb, sk);
1204 NETLINK_CB(skb).sk = ssk;
1205 netlink_deliver_tap_kernel(sk, ssk, skb);
1206 nlk->netlink_rcv(skb);
1215 int netlink_unicast(struct sock *ssk, struct sk_buff *skb,
1216 u32 portid, int nonblock)
1222 skb = netlink_trim(skb, gfp_any());
1224 timeo = sock_sndtimeo(ssk, nonblock);
1226 sk = netlink_getsockbyportid(ssk, portid);
1231 if (netlink_is_kernel(sk))
1232 return netlink_unicast_kernel(sk, skb, ssk);
1234 if (sk_filter(sk, skb)) {
1241 err = netlink_attachskb(sk, skb, &timeo, ssk);
1247 return netlink_sendskb(sk, skb);
1249 EXPORT_SYMBOL(netlink_unicast);
1251 int netlink_has_listeners(struct sock *sk, unsigned int group)
1254 struct listeners *listeners;
1256 BUG_ON(!netlink_is_kernel(sk));
1259 listeners = rcu_dereference(nl_table[sk->sk_protocol].listeners);
1261 if (listeners && group - 1 < nl_table[sk->sk_protocol].groups)
1262 res = test_bit(group - 1, listeners->masks);
1268 EXPORT_SYMBOL_GPL(netlink_has_listeners);
1270 static int netlink_broadcast_deliver(struct sock *sk, struct sk_buff *skb)
1272 struct netlink_sock *nlk = nlk_sk(sk);
1274 if (atomic_read(&sk->sk_rmem_alloc) <= sk->sk_rcvbuf &&
1275 !test_bit(NETLINK_S_CONGESTED, &nlk->state)) {
1276 netlink_skb_set_owner_r(skb, sk);
1277 __netlink_sendskb(sk, skb);
1278 return atomic_read(&sk->sk_rmem_alloc) > (sk->sk_rcvbuf >> 1);
1283 struct netlink_broadcast_data {
1284 struct sock *exclude_sk;
1289 int delivery_failure;
1293 struct sk_buff *skb, *skb2;
1294 int (*tx_filter)(struct sock *dsk, struct sk_buff *skb, void *data);
1298 static void do_one_broadcast(struct sock *sk,
1299 struct netlink_broadcast_data *p)
1301 struct netlink_sock *nlk = nlk_sk(sk);
1304 if (p->exclude_sk == sk)
1307 if (nlk->portid == p->portid || p->group - 1 >= nlk->ngroups ||
1308 !test_bit(p->group - 1, nlk->groups))
1311 if (!net_eq(sock_net(sk), p->net)) {
1312 if (!(nlk->flags & NETLINK_F_LISTEN_ALL_NSID))
1315 if (!peernet_has_id(sock_net(sk), p->net))
1318 if (!file_ns_capable(sk->sk_socket->file, p->net->user_ns,
1324 netlink_overrun(sk);
1329 if (p->skb2 == NULL) {
1330 if (skb_shared(p->skb)) {
1331 p->skb2 = skb_clone(p->skb, p->allocation);
1333 p->skb2 = skb_get(p->skb);
1335 * skb ownership may have been set when
1336 * delivered to a previous socket.
1338 skb_orphan(p->skb2);
1341 if (p->skb2 == NULL) {
1342 netlink_overrun(sk);
1343 /* Clone failed. Notify ALL listeners. */
1345 if (nlk->flags & NETLINK_F_BROADCAST_SEND_ERROR)
1346 p->delivery_failure = 1;
1349 if (p->tx_filter && p->tx_filter(sk, p->skb2, p->tx_data)) {
1354 if (sk_filter(sk, p->skb2)) {
1359 NETLINK_CB(p->skb2).nsid = peernet2id(sock_net(sk), p->net);
1360 NETLINK_CB(p->skb2).nsid_is_set = true;
1361 val = netlink_broadcast_deliver(sk, p->skb2);
1363 netlink_overrun(sk);
1364 if (nlk->flags & NETLINK_F_BROADCAST_SEND_ERROR)
1365 p->delivery_failure = 1;
1367 p->congested |= val;
1375 int netlink_broadcast_filtered(struct sock *ssk, struct sk_buff *skb, u32 portid,
1376 u32 group, gfp_t allocation,
1377 int (*filter)(struct sock *dsk, struct sk_buff *skb, void *data),
1380 struct net *net = sock_net(ssk);
1381 struct netlink_broadcast_data info;
1384 skb = netlink_trim(skb, allocation);
1386 info.exclude_sk = ssk;
1388 info.portid = portid;
1391 info.delivery_failure = 0;
1394 info.allocation = allocation;
1397 info.tx_filter = filter;
1398 info.tx_data = filter_data;
1400 /* While we sleep in clone, do not allow to change socket list */
1402 netlink_lock_table();
1404 sk_for_each_bound(sk, &nl_table[ssk->sk_protocol].mc_list)
1405 do_one_broadcast(sk, &info);
1409 netlink_unlock_table();
1411 if (info.delivery_failure) {
1412 kfree_skb(info.skb2);
1415 consume_skb(info.skb2);
1417 if (info.delivered) {
1418 if (info.congested && gfpflags_allow_blocking(allocation))
1424 EXPORT_SYMBOL(netlink_broadcast_filtered);
1426 int netlink_broadcast(struct sock *ssk, struct sk_buff *skb, u32 portid,
1427 u32 group, gfp_t allocation)
1429 return netlink_broadcast_filtered(ssk, skb, portid, group, allocation,
1432 EXPORT_SYMBOL(netlink_broadcast);
1434 struct netlink_set_err_data {
1435 struct sock *exclude_sk;
1441 static int do_one_set_err(struct sock *sk, struct netlink_set_err_data *p)
1443 struct netlink_sock *nlk = nlk_sk(sk);
1446 if (sk == p->exclude_sk)
1449 if (!net_eq(sock_net(sk), sock_net(p->exclude_sk)))
1452 if (nlk->portid == p->portid || p->group - 1 >= nlk->ngroups ||
1453 !test_bit(p->group - 1, nlk->groups))
1456 if (p->code == ENOBUFS && nlk->flags & NETLINK_F_RECV_NO_ENOBUFS) {
1461 sk->sk_err = p->code;
1462 sk->sk_error_report(sk);
1468 * netlink_set_err - report error to broadcast listeners
1469 * @ssk: the kernel netlink socket, as returned by netlink_kernel_create()
1470 * @portid: the PORTID of a process that we want to skip (if any)
1471 * @group: the broadcast group that will notice the error
1472 * @code: error code, must be negative (as usual in kernelspace)
1474 * This function returns the number of broadcast listeners that have set the
1475 * NETLINK_NO_ENOBUFS socket option.
1477 int netlink_set_err(struct sock *ssk, u32 portid, u32 group, int code)
1479 struct netlink_set_err_data info;
1483 info.exclude_sk = ssk;
1484 info.portid = portid;
1486 /* sk->sk_err wants a positive error value */
1489 read_lock(&nl_table_lock);
1491 sk_for_each_bound(sk, &nl_table[ssk->sk_protocol].mc_list)
1492 ret += do_one_set_err(sk, &info);
1494 read_unlock(&nl_table_lock);
1497 EXPORT_SYMBOL(netlink_set_err);
1499 /* must be called with netlink table grabbed */
1500 static void netlink_update_socket_mc(struct netlink_sock *nlk,
1504 int old, new = !!is_new, subscriptions;
1506 old = test_bit(group - 1, nlk->groups);
1507 subscriptions = nlk->subscriptions - old + new;
1509 __set_bit(group - 1, nlk->groups);
1511 __clear_bit(group - 1, nlk->groups);
1512 netlink_update_subscriptions(&nlk->sk, subscriptions);
1513 netlink_update_listeners(&nlk->sk);
1516 static int netlink_setsockopt(struct socket *sock, int level, int optname,
1517 char __user *optval, unsigned int optlen)
1519 struct sock *sk = sock->sk;
1520 struct netlink_sock *nlk = nlk_sk(sk);
1521 unsigned int val = 0;
1524 if (level != SOL_NETLINK)
1525 return -ENOPROTOOPT;
1527 if (optlen >= sizeof(int) &&
1528 get_user(val, (unsigned int __user *)optval))
1532 case NETLINK_PKTINFO:
1534 nlk->flags |= NETLINK_F_RECV_PKTINFO;
1536 nlk->flags &= ~NETLINK_F_RECV_PKTINFO;
1539 case NETLINK_ADD_MEMBERSHIP:
1540 case NETLINK_DROP_MEMBERSHIP: {
1541 if (!netlink_allowed(sock, NL_CFG_F_NONROOT_RECV))
1543 err = netlink_realloc_groups(sk);
1546 if (!val || val - 1 >= nlk->ngroups)
1548 if (optname == NETLINK_ADD_MEMBERSHIP && nlk->netlink_bind) {
1549 err = nlk->netlink_bind(sock_net(sk), val);
1553 netlink_table_grab();
1554 netlink_update_socket_mc(nlk, val,
1555 optname == NETLINK_ADD_MEMBERSHIP);
1556 netlink_table_ungrab();
1557 if (optname == NETLINK_DROP_MEMBERSHIP && nlk->netlink_unbind)
1558 nlk->netlink_unbind(sock_net(sk), val);
1563 case NETLINK_BROADCAST_ERROR:
1565 nlk->flags |= NETLINK_F_BROADCAST_SEND_ERROR;
1567 nlk->flags &= ~NETLINK_F_BROADCAST_SEND_ERROR;
1570 case NETLINK_NO_ENOBUFS:
1572 nlk->flags |= NETLINK_F_RECV_NO_ENOBUFS;
1573 clear_bit(NETLINK_S_CONGESTED, &nlk->state);
1574 wake_up_interruptible(&nlk->wait);
1576 nlk->flags &= ~NETLINK_F_RECV_NO_ENOBUFS;
1580 case NETLINK_LISTEN_ALL_NSID:
1581 if (!ns_capable(sock_net(sk)->user_ns, CAP_NET_BROADCAST))
1585 nlk->flags |= NETLINK_F_LISTEN_ALL_NSID;
1587 nlk->flags &= ~NETLINK_F_LISTEN_ALL_NSID;
1590 case NETLINK_CAP_ACK:
1592 nlk->flags |= NETLINK_F_CAP_ACK;
1594 nlk->flags &= ~NETLINK_F_CAP_ACK;
1603 static int netlink_getsockopt(struct socket *sock, int level, int optname,
1604 char __user *optval, int __user *optlen)
1606 struct sock *sk = sock->sk;
1607 struct netlink_sock *nlk = nlk_sk(sk);
1610 if (level != SOL_NETLINK)
1611 return -ENOPROTOOPT;
1613 if (get_user(len, optlen))
1619 case NETLINK_PKTINFO:
1620 if (len < sizeof(int))
1623 val = nlk->flags & NETLINK_F_RECV_PKTINFO ? 1 : 0;
1624 if (put_user(len, optlen) ||
1625 put_user(val, optval))
1629 case NETLINK_BROADCAST_ERROR:
1630 if (len < sizeof(int))
1633 val = nlk->flags & NETLINK_F_BROADCAST_SEND_ERROR ? 1 : 0;
1634 if (put_user(len, optlen) ||
1635 put_user(val, optval))
1639 case NETLINK_NO_ENOBUFS:
1640 if (len < sizeof(int))
1643 val = nlk->flags & NETLINK_F_RECV_NO_ENOBUFS ? 1 : 0;
1644 if (put_user(len, optlen) ||
1645 put_user(val, optval))
1649 case NETLINK_LIST_MEMBERSHIPS: {
1650 int pos, idx, shift;
1653 netlink_lock_table();
1654 for (pos = 0; pos * 8 < nlk->ngroups; pos += sizeof(u32)) {
1655 if (len - pos < sizeof(u32))
1658 idx = pos / sizeof(unsigned long);
1659 shift = (pos % sizeof(unsigned long)) * 8;
1660 if (put_user((u32)(nlk->groups[idx] >> shift),
1661 (u32 __user *)(optval + pos))) {
1666 if (put_user(ALIGN(nlk->ngroups / 8, sizeof(u32)), optlen))
1668 netlink_unlock_table();
1671 case NETLINK_CAP_ACK:
1672 if (len < sizeof(int))
1675 val = nlk->flags & NETLINK_F_CAP_ACK ? 1 : 0;
1676 if (put_user(len, optlen) ||
1677 put_user(val, optval))
1687 static void netlink_cmsg_recv_pktinfo(struct msghdr *msg, struct sk_buff *skb)
1689 struct nl_pktinfo info;
1691 info.group = NETLINK_CB(skb).dst_group;
1692 put_cmsg(msg, SOL_NETLINK, NETLINK_PKTINFO, sizeof(info), &info);
1695 static void netlink_cmsg_listen_all_nsid(struct sock *sk, struct msghdr *msg,
1696 struct sk_buff *skb)
1698 if (!NETLINK_CB(skb).nsid_is_set)
1701 put_cmsg(msg, SOL_NETLINK, NETLINK_LISTEN_ALL_NSID, sizeof(int),
1702 &NETLINK_CB(skb).nsid);
1705 static int netlink_sendmsg(struct socket *sock, struct msghdr *msg, size_t len)
1707 struct sock *sk = sock->sk;
1708 struct netlink_sock *nlk = nlk_sk(sk);
1709 DECLARE_SOCKADDR(struct sockaddr_nl *, addr, msg->msg_name);
1712 struct sk_buff *skb;
1714 struct scm_cookie scm;
1715 u32 netlink_skb_flags = 0;
1717 if (msg->msg_flags&MSG_OOB)
1720 err = scm_send(sock, msg, &scm, true);
1724 if (msg->msg_namelen) {
1726 if (addr->nl_family != AF_NETLINK)
1728 dst_portid = addr->nl_pid;
1729 dst_group = ffs(addr->nl_groups);
1731 if ((dst_group || dst_portid) &&
1732 !netlink_allowed(sock, NL_CFG_F_NONROOT_SEND))
1734 netlink_skb_flags |= NETLINK_SKB_DST;
1736 dst_portid = nlk->dst_portid;
1737 dst_group = nlk->dst_group;
1741 err = netlink_autobind(sock);
1745 /* Ensure nlk is hashed and visible. */
1750 if (len > sk->sk_sndbuf - 32)
1753 skb = netlink_alloc_large_skb(len, dst_group);
1757 NETLINK_CB(skb).portid = nlk->portid;
1758 NETLINK_CB(skb).dst_group = dst_group;
1759 NETLINK_CB(skb).creds = scm.creds;
1760 NETLINK_CB(skb).flags = netlink_skb_flags;
1763 if (memcpy_from_msg(skb_put(skb, len), msg, len)) {
1768 err = security_netlink_send(sk, skb);
1775 atomic_inc(&skb->users);
1776 netlink_broadcast(sk, skb, dst_portid, dst_group, GFP_KERNEL);
1778 err = netlink_unicast(sk, skb, dst_portid, msg->msg_flags&MSG_DONTWAIT);
1785 static int netlink_recvmsg(struct socket *sock, struct msghdr *msg, size_t len,
1788 struct scm_cookie scm;
1789 struct sock *sk = sock->sk;
1790 struct netlink_sock *nlk = nlk_sk(sk);
1791 int noblock = flags&MSG_DONTWAIT;
1793 struct sk_buff *skb, *data_skb;
1801 skb = skb_recv_datagram(sk, flags, noblock, &err);
1807 #ifdef CONFIG_COMPAT_NETLINK_MESSAGES
1808 if (unlikely(skb_shinfo(skb)->frag_list)) {
1810 * If this skb has a frag_list, then here that means that we
1811 * will have to use the frag_list skb's data for compat tasks
1812 * and the regular skb's data for normal (non-compat) tasks.
1814 * If we need to send the compat skb, assign it to the
1815 * 'data_skb' variable so that it will be used below for data
1816 * copying. We keep 'skb' for everything else, including
1817 * freeing both later.
1819 if (flags & MSG_CMSG_COMPAT)
1820 data_skb = skb_shinfo(skb)->frag_list;
1824 /* Record the max length of recvmsg() calls for future allocations */
1825 nlk->max_recvmsg_len = max(nlk->max_recvmsg_len, len);
1826 nlk->max_recvmsg_len = min_t(size_t, nlk->max_recvmsg_len,
1829 copied = data_skb->len;
1831 msg->msg_flags |= MSG_TRUNC;
1835 skb_reset_transport_header(data_skb);
1836 err = skb_copy_datagram_msg(data_skb, 0, msg, copied);
1838 if (msg->msg_name) {
1839 DECLARE_SOCKADDR(struct sockaddr_nl *, addr, msg->msg_name);
1840 addr->nl_family = AF_NETLINK;
1842 addr->nl_pid = NETLINK_CB(skb).portid;
1843 addr->nl_groups = netlink_group_mask(NETLINK_CB(skb).dst_group);
1844 msg->msg_namelen = sizeof(*addr);
1847 if (nlk->flags & NETLINK_F_RECV_PKTINFO)
1848 netlink_cmsg_recv_pktinfo(msg, skb);
1849 if (nlk->flags & NETLINK_F_LISTEN_ALL_NSID)
1850 netlink_cmsg_listen_all_nsid(sk, msg, skb);
1852 memset(&scm, 0, sizeof(scm));
1853 scm.creds = *NETLINK_CREDS(skb);
1854 if (flags & MSG_TRUNC)
1855 copied = data_skb->len;
1857 skb_free_datagram(sk, skb);
1859 if (nlk->cb_running &&
1860 atomic_read(&sk->sk_rmem_alloc) <= sk->sk_rcvbuf / 2) {
1861 ret = netlink_dump(sk);
1864 sk->sk_error_report(sk);
1868 scm_recv(sock, msg, &scm, flags);
1870 netlink_rcv_wake(sk);
1871 return err ? : copied;
1874 static void netlink_data_ready(struct sock *sk)
1880 * We export these functions to other modules. They provide a
1881 * complete set of kernel non-blocking support for message
1886 __netlink_kernel_create(struct net *net, int unit, struct module *module,
1887 struct netlink_kernel_cfg *cfg)
1889 struct socket *sock;
1891 struct netlink_sock *nlk;
1892 struct listeners *listeners = NULL;
1893 struct mutex *cb_mutex = cfg ? cfg->cb_mutex : NULL;
1894 unsigned int groups;
1898 if (unit < 0 || unit >= MAX_LINKS)
1901 if (sock_create_lite(PF_NETLINK, SOCK_DGRAM, unit, &sock))
1904 if (__netlink_create(net, sock, cb_mutex, unit, 1) < 0)
1905 goto out_sock_release_nosk;
1909 if (!cfg || cfg->groups < 32)
1912 groups = cfg->groups;
1914 listeners = kzalloc(sizeof(*listeners) + NLGRPSZ(groups), GFP_KERNEL);
1916 goto out_sock_release;
1918 sk->sk_data_ready = netlink_data_ready;
1919 if (cfg && cfg->input)
1920 nlk_sk(sk)->netlink_rcv = cfg->input;
1922 if (netlink_insert(sk, 0))
1923 goto out_sock_release;
1926 nlk->flags |= NETLINK_F_KERNEL_SOCKET;
1928 netlink_table_grab();
1929 if (!nl_table[unit].registered) {
1930 nl_table[unit].groups = groups;
1931 rcu_assign_pointer(nl_table[unit].listeners, listeners);
1932 nl_table[unit].cb_mutex = cb_mutex;
1933 nl_table[unit].module = module;
1935 nl_table[unit].bind = cfg->bind;
1936 nl_table[unit].unbind = cfg->unbind;
1937 nl_table[unit].flags = cfg->flags;
1939 nl_table[unit].compare = cfg->compare;
1941 nl_table[unit].registered = 1;
1944 nl_table[unit].registered++;
1946 netlink_table_ungrab();
1951 netlink_kernel_release(sk);
1954 out_sock_release_nosk:
1958 EXPORT_SYMBOL(__netlink_kernel_create);
1961 netlink_kernel_release(struct sock *sk)
1963 if (sk == NULL || sk->sk_socket == NULL)
1966 sock_release(sk->sk_socket);
1968 EXPORT_SYMBOL(netlink_kernel_release);
1970 int __netlink_change_ngroups(struct sock *sk, unsigned int groups)
1972 struct listeners *new, *old;
1973 struct netlink_table *tbl = &nl_table[sk->sk_protocol];
1978 if (NLGRPSZ(tbl->groups) < NLGRPSZ(groups)) {
1979 new = kzalloc(sizeof(*new) + NLGRPSZ(groups), GFP_ATOMIC);
1982 old = nl_deref_protected(tbl->listeners);
1983 memcpy(new->masks, old->masks, NLGRPSZ(tbl->groups));
1984 rcu_assign_pointer(tbl->listeners, new);
1986 kfree_rcu(old, rcu);
1988 tbl->groups = groups;
1994 * netlink_change_ngroups - change number of multicast groups
1996 * This changes the number of multicast groups that are available
1997 * on a certain netlink family. Note that it is not possible to
1998 * change the number of groups to below 32. Also note that it does
1999 * not implicitly call netlink_clear_multicast_users() when the
2000 * number of groups is reduced.
2002 * @sk: The kernel netlink socket, as returned by netlink_kernel_create().
2003 * @groups: The new number of groups.
2005 int netlink_change_ngroups(struct sock *sk, unsigned int groups)
2009 netlink_table_grab();
2010 err = __netlink_change_ngroups(sk, groups);
2011 netlink_table_ungrab();
2016 void __netlink_clear_multicast_users(struct sock *ksk, unsigned int group)
2019 struct netlink_table *tbl = &nl_table[ksk->sk_protocol];
2021 sk_for_each_bound(sk, &tbl->mc_list)
2022 netlink_update_socket_mc(nlk_sk(sk), group, 0);
2026 __nlmsg_put(struct sk_buff *skb, u32 portid, u32 seq, int type, int len, int flags)
2028 struct nlmsghdr *nlh;
2029 int size = nlmsg_msg_size(len);
2031 nlh = (struct nlmsghdr *)skb_put(skb, NLMSG_ALIGN(size));
2032 nlh->nlmsg_type = type;
2033 nlh->nlmsg_len = size;
2034 nlh->nlmsg_flags = flags;
2035 nlh->nlmsg_pid = portid;
2036 nlh->nlmsg_seq = seq;
2037 if (!__builtin_constant_p(size) || NLMSG_ALIGN(size) - size != 0)
2038 memset(nlmsg_data(nlh) + len, 0, NLMSG_ALIGN(size) - size);
2041 EXPORT_SYMBOL(__nlmsg_put);
2044 * It looks a bit ugly.
2045 * It would be better to create kernel thread.
2048 static int netlink_dump(struct sock *sk)
2050 struct netlink_sock *nlk = nlk_sk(sk);
2051 struct netlink_callback *cb;
2052 struct sk_buff *skb = NULL;
2053 struct nlmsghdr *nlh;
2054 int len, err = -ENOBUFS;
2058 mutex_lock(nlk->cb_mutex);
2059 if (!nlk->cb_running) {
2064 if (atomic_read(&sk->sk_rmem_alloc) >= sk->sk_rcvbuf)
2067 /* NLMSG_GOODSIZE is small to avoid high order allocations being
2068 * required, but it makes sense to _attempt_ a 16K bytes allocation
2069 * to reduce number of system calls on dump operations, if user
2070 * ever provided a big enough buffer.
2073 alloc_min_size = max_t(int, cb->min_dump_alloc, NLMSG_GOODSIZE);
2075 if (alloc_min_size < nlk->max_recvmsg_len) {
2076 alloc_size = nlk->max_recvmsg_len;
2077 skb = alloc_skb(alloc_size, GFP_KERNEL |
2078 __GFP_NOWARN | __GFP_NORETRY);
2081 alloc_size = alloc_min_size;
2082 skb = alloc_skb(alloc_size, GFP_KERNEL);
2087 /* Trim skb to allocated size. User is expected to provide buffer as
2088 * large as max(min_dump_alloc, 16KiB (mac_recvmsg_len capped at
2089 * netlink_recvmsg())). dump will pack as many smaller messages as
2090 * could fit within the allocated skb. skb is typically allocated
2091 * with larger space than required (could be as much as near 2x the
2092 * requested size with align to next power of 2 approach). Allowing
2093 * dump to use the excess space makes it difficult for a user to have a
2094 * reasonable static buffer based on the expected largest dump of a
2095 * single netdev. The outcome is MSG_TRUNC error.
2097 skb_reserve(skb, skb_tailroom(skb) - alloc_size);
2098 netlink_skb_set_owner_r(skb, sk);
2100 len = cb->dump(skb, cb);
2103 mutex_unlock(nlk->cb_mutex);
2105 if (sk_filter(sk, skb))
2108 __netlink_sendskb(sk, skb);
2112 nlh = nlmsg_put_answer(skb, cb, NLMSG_DONE, sizeof(len), NLM_F_MULTI);
2116 nl_dump_check_consistent(cb, nlh);
2118 memcpy(nlmsg_data(nlh), &len, sizeof(len));
2120 if (sk_filter(sk, skb))
2123 __netlink_sendskb(sk, skb);
2128 nlk->cb_running = false;
2129 mutex_unlock(nlk->cb_mutex);
2130 module_put(cb->module);
2131 consume_skb(cb->skb);
2135 mutex_unlock(nlk->cb_mutex);
2140 int __netlink_dump_start(struct sock *ssk, struct sk_buff *skb,
2141 const struct nlmsghdr *nlh,
2142 struct netlink_dump_control *control)
2144 struct netlink_callback *cb;
2146 struct netlink_sock *nlk;
2149 atomic_inc(&skb->users);
2151 sk = netlink_lookup(sock_net(ssk), ssk->sk_protocol, NETLINK_CB(skb).portid);
2153 ret = -ECONNREFUSED;
2158 mutex_lock(nlk->cb_mutex);
2159 /* A dump is in progress... */
2160 if (nlk->cb_running) {
2164 /* add reference of module which cb->dump belongs to */
2165 if (!try_module_get(control->module)) {
2166 ret = -EPROTONOSUPPORT;
2171 memset(cb, 0, sizeof(*cb));
2172 cb->start = control->start;
2173 cb->dump = control->dump;
2174 cb->done = control->done;
2176 cb->data = control->data;
2177 cb->module = control->module;
2178 cb->min_dump_alloc = control->min_dump_alloc;
2181 nlk->cb_running = true;
2183 mutex_unlock(nlk->cb_mutex);
2188 ret = netlink_dump(sk);
2194 /* We successfully started a dump, by returning -EINTR we
2195 * signal not to send ACK even if it was requested.
2201 mutex_unlock(nlk->cb_mutex);
2206 EXPORT_SYMBOL(__netlink_dump_start);
2208 void netlink_ack(struct sk_buff *in_skb, struct nlmsghdr *nlh, int err)
2210 struct sk_buff *skb;
2211 struct nlmsghdr *rep;
2212 struct nlmsgerr *errmsg;
2213 size_t payload = sizeof(*errmsg);
2214 struct netlink_sock *nlk = nlk_sk(NETLINK_CB(in_skb).sk);
2216 /* Error messages get the original request appened, unless the user
2217 * requests to cap the error message.
2219 if (!(nlk->flags & NETLINK_F_CAP_ACK) && err)
2220 payload += nlmsg_len(nlh);
2222 skb = nlmsg_new(payload, GFP_KERNEL);
2226 sk = netlink_lookup(sock_net(in_skb->sk),
2227 in_skb->sk->sk_protocol,
2228 NETLINK_CB(in_skb).portid);
2230 sk->sk_err = ENOBUFS;
2231 sk->sk_error_report(sk);
2237 rep = __nlmsg_put(skb, NETLINK_CB(in_skb).portid, nlh->nlmsg_seq,
2238 NLMSG_ERROR, payload, 0);
2239 errmsg = nlmsg_data(rep);
2240 errmsg->error = err;
2241 memcpy(&errmsg->msg, nlh, payload > sizeof(*errmsg) ? nlh->nlmsg_len : sizeof(*nlh));
2242 netlink_unicast(in_skb->sk, skb, NETLINK_CB(in_skb).portid, MSG_DONTWAIT);
2244 EXPORT_SYMBOL(netlink_ack);
2246 int netlink_rcv_skb(struct sk_buff *skb, int (*cb)(struct sk_buff *,
2249 struct nlmsghdr *nlh;
2252 while (skb->len >= nlmsg_total_size(0)) {
2255 nlh = nlmsg_hdr(skb);
2258 if (nlh->nlmsg_len < NLMSG_HDRLEN || skb->len < nlh->nlmsg_len)
2261 /* Only requests are handled by the kernel */
2262 if (!(nlh->nlmsg_flags & NLM_F_REQUEST))
2265 /* Skip control messages */
2266 if (nlh->nlmsg_type < NLMSG_MIN_TYPE)
2274 if (nlh->nlmsg_flags & NLM_F_ACK || err)
2275 netlink_ack(skb, nlh, err);
2278 msglen = NLMSG_ALIGN(nlh->nlmsg_len);
2279 if (msglen > skb->len)
2281 skb_pull(skb, msglen);
2286 EXPORT_SYMBOL(netlink_rcv_skb);
2289 * nlmsg_notify - send a notification netlink message
2290 * @sk: netlink socket to use
2291 * @skb: notification message
2292 * @portid: destination netlink portid for reports or 0
2293 * @group: destination multicast group or 0
2294 * @report: 1 to report back, 0 to disable
2295 * @flags: allocation flags
2297 int nlmsg_notify(struct sock *sk, struct sk_buff *skb, u32 portid,
2298 unsigned int group, int report, gfp_t flags)
2303 int exclude_portid = 0;
2306 atomic_inc(&skb->users);
2307 exclude_portid = portid;
2310 /* errors reported via destination sk->sk_err, but propagate
2311 * delivery errors if NETLINK_BROADCAST_ERROR flag is set */
2312 err = nlmsg_multicast(sk, skb, exclude_portid, group, flags);
2318 err2 = nlmsg_unicast(sk, skb, portid);
2319 if (!err || err == -ESRCH)
2325 EXPORT_SYMBOL(nlmsg_notify);
2327 #ifdef CONFIG_PROC_FS
2328 struct nl_seq_iter {
2329 struct seq_net_private p;
2330 struct rhashtable_iter hti;
2334 static int netlink_walk_start(struct nl_seq_iter *iter)
2338 err = rhashtable_walk_init(&nl_table[iter->link].hash, &iter->hti);
2340 iter->link = MAX_LINKS;
2344 err = rhashtable_walk_start(&iter->hti);
2345 return err == -EAGAIN ? 0 : err;
2348 static void netlink_walk_stop(struct nl_seq_iter *iter)
2350 rhashtable_walk_stop(&iter->hti);
2351 rhashtable_walk_exit(&iter->hti);
2354 static void *__netlink_seq_next(struct seq_file *seq)
2356 struct nl_seq_iter *iter = seq->private;
2357 struct netlink_sock *nlk;
2363 nlk = rhashtable_walk_next(&iter->hti);
2366 if (PTR_ERR(nlk) == -EAGAIN)
2375 netlink_walk_stop(iter);
2376 if (++iter->link >= MAX_LINKS)
2379 err = netlink_walk_start(iter);
2381 return ERR_PTR(err);
2383 } while (sock_net(&nlk->sk) != seq_file_net(seq));
2388 static void *netlink_seq_start(struct seq_file *seq, loff_t *posp)
2390 struct nl_seq_iter *iter = seq->private;
2391 void *obj = SEQ_START_TOKEN;
2397 err = netlink_walk_start(iter);
2399 return ERR_PTR(err);
2401 for (pos = *posp; pos && obj && !IS_ERR(obj); pos--)
2402 obj = __netlink_seq_next(seq);
2407 static void *netlink_seq_next(struct seq_file *seq, void *v, loff_t *pos)
2410 return __netlink_seq_next(seq);
2413 static void netlink_seq_stop(struct seq_file *seq, void *v)
2415 struct nl_seq_iter *iter = seq->private;
2417 if (iter->link >= MAX_LINKS)
2420 netlink_walk_stop(iter);
2424 static int netlink_seq_show(struct seq_file *seq, void *v)
2426 if (v == SEQ_START_TOKEN) {
2428 "sk Eth Pid Groups "
2429 "Rmem Wmem Dump Locks Drops Inode\n");
2432 struct netlink_sock *nlk = nlk_sk(s);
2434 seq_printf(seq, "%pK %-3d %-6u %08x %-8d %-8d %d %-8d %-8d %-8lu\n",
2438 nlk->groups ? (u32)nlk->groups[0] : 0,
2439 sk_rmem_alloc_get(s),
2440 sk_wmem_alloc_get(s),
2442 atomic_read(&s->sk_refcnt),
2443 atomic_read(&s->sk_drops),
2451 static const struct seq_operations netlink_seq_ops = {
2452 .start = netlink_seq_start,
2453 .next = netlink_seq_next,
2454 .stop = netlink_seq_stop,
2455 .show = netlink_seq_show,
2459 static int netlink_seq_open(struct inode *inode, struct file *file)
2461 return seq_open_net(inode, file, &netlink_seq_ops,
2462 sizeof(struct nl_seq_iter));
2465 static const struct file_operations netlink_seq_fops = {
2466 .owner = THIS_MODULE,
2467 .open = netlink_seq_open,
2469 .llseek = seq_lseek,
2470 .release = seq_release_net,
2475 int netlink_register_notifier(struct notifier_block *nb)
2477 return atomic_notifier_chain_register(&netlink_chain, nb);
2479 EXPORT_SYMBOL(netlink_register_notifier);
2481 int netlink_unregister_notifier(struct notifier_block *nb)
2483 return atomic_notifier_chain_unregister(&netlink_chain, nb);
2485 EXPORT_SYMBOL(netlink_unregister_notifier);
2487 static const struct proto_ops netlink_ops = {
2488 .family = PF_NETLINK,
2489 .owner = THIS_MODULE,
2490 .release = netlink_release,
2491 .bind = netlink_bind,
2492 .connect = netlink_connect,
2493 .socketpair = sock_no_socketpair,
2494 .accept = sock_no_accept,
2495 .getname = netlink_getname,
2496 .poll = datagram_poll,
2497 .ioctl = sock_no_ioctl,
2498 .listen = sock_no_listen,
2499 .shutdown = sock_no_shutdown,
2500 .setsockopt = netlink_setsockopt,
2501 .getsockopt = netlink_getsockopt,
2502 .sendmsg = netlink_sendmsg,
2503 .recvmsg = netlink_recvmsg,
2504 .mmap = sock_no_mmap,
2505 .sendpage = sock_no_sendpage,
2508 static const struct net_proto_family netlink_family_ops = {
2509 .family = PF_NETLINK,
2510 .create = netlink_create,
2511 .owner = THIS_MODULE, /* for consistency 8) */
2514 static int __net_init netlink_net_init(struct net *net)
2516 #ifdef CONFIG_PROC_FS
2517 if (!proc_create("netlink", 0, net->proc_net, &netlink_seq_fops))
2523 static void __net_exit netlink_net_exit(struct net *net)
2525 #ifdef CONFIG_PROC_FS
2526 remove_proc_entry("netlink", net->proc_net);
2530 static void __init netlink_add_usersock_entry(void)
2532 struct listeners *listeners;
2535 listeners = kzalloc(sizeof(*listeners) + NLGRPSZ(groups), GFP_KERNEL);
2537 panic("netlink_add_usersock_entry: Cannot allocate listeners\n");
2539 netlink_table_grab();
2541 nl_table[NETLINK_USERSOCK].groups = groups;
2542 rcu_assign_pointer(nl_table[NETLINK_USERSOCK].listeners, listeners);
2543 nl_table[NETLINK_USERSOCK].module = THIS_MODULE;
2544 nl_table[NETLINK_USERSOCK].registered = 1;
2545 nl_table[NETLINK_USERSOCK].flags = NL_CFG_F_NONROOT_SEND;
2547 netlink_table_ungrab();
2550 static struct pernet_operations __net_initdata netlink_net_ops = {
2551 .init = netlink_net_init,
2552 .exit = netlink_net_exit,
2555 static inline u32 netlink_hash(const void *data, u32 len, u32 seed)
2557 const struct netlink_sock *nlk = data;
2558 struct netlink_compare_arg arg;
2560 netlink_compare_arg_init(&arg, sock_net(&nlk->sk), nlk->portid);
2561 return jhash2((u32 *)&arg, netlink_compare_arg_len / sizeof(u32), seed);
2564 static const struct rhashtable_params netlink_rhashtable_params = {
2565 .head_offset = offsetof(struct netlink_sock, node),
2566 .key_len = netlink_compare_arg_len,
2567 .obj_hashfn = netlink_hash,
2568 .obj_cmpfn = netlink_compare,
2569 .automatic_shrinking = true,
2572 static int __init netlink_proto_init(void)
2575 int err = proto_register(&netlink_proto, 0);
2580 BUILD_BUG_ON(sizeof(struct netlink_skb_parms) > FIELD_SIZEOF(struct sk_buff, cb));
2582 nl_table = kcalloc(MAX_LINKS, sizeof(*nl_table), GFP_KERNEL);
2586 for (i = 0; i < MAX_LINKS; i++) {
2587 if (rhashtable_init(&nl_table[i].hash,
2588 &netlink_rhashtable_params) < 0) {
2590 rhashtable_destroy(&nl_table[i].hash);
2596 INIT_LIST_HEAD(&netlink_tap_all);
2598 netlink_add_usersock_entry();
2600 sock_register(&netlink_family_ops);
2601 register_pernet_subsys(&netlink_net_ops);
2602 /* The netlink device handler may be needed early. */
2607 panic("netlink_init: Cannot allocate nl_table\n");
2610 core_initcall(netlink_proto_init);