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 <linux/uaccess.h>
44 #include <linux/skbuff.h>
45 #include <linux/netdevice.h>
46 #include <linux/rtnetlink.h>
47 #include <linux/proc_fs.h>
48 #include <linux/seq_file.h>
49 #include <linux/notifier.h>
50 #include <linux/security.h>
51 #include <linux/jhash.h>
52 #include <linux/jiffies.h>
53 #include <linux/random.h>
54 #include <linux/bitops.h>
56 #include <linux/types.h>
57 #include <linux/audit.h>
58 #include <linux/mutex.h>
59 #include <linux/vmalloc.h>
60 #include <linux/if_arp.h>
61 #include <linux/rhashtable.h>
62 #include <asm/cacheflush.h>
63 #include <linux/hash.h>
64 #include <linux/genetlink.h>
65 #include <linux/net_namespace.h>
67 #include <net/net_namespace.h>
70 #include <net/netlink.h>
72 #include "af_netlink.h"
76 unsigned long masks[0];
80 #define NETLINK_S_CONGESTED 0x0
82 static inline int netlink_is_kernel(struct sock *sk)
84 return nlk_sk(sk)->flags & NETLINK_F_KERNEL_SOCKET;
87 struct netlink_table *nl_table __read_mostly;
88 EXPORT_SYMBOL_GPL(nl_table);
90 static DECLARE_WAIT_QUEUE_HEAD(nl_table_wait);
92 static struct lock_class_key nlk_cb_mutex_keys[MAX_LINKS];
94 static const char *const nlk_cb_mutex_key_strings[MAX_LINKS + 1] = {
97 "nlk_cb_mutex-USERSOCK",
98 "nlk_cb_mutex-FIREWALL",
99 "nlk_cb_mutex-SOCK_DIAG",
100 "nlk_cb_mutex-NFLOG",
102 "nlk_cb_mutex-SELINUX",
103 "nlk_cb_mutex-ISCSI",
104 "nlk_cb_mutex-AUDIT",
105 "nlk_cb_mutex-FIB_LOOKUP",
106 "nlk_cb_mutex-CONNECTOR",
107 "nlk_cb_mutex-NETFILTER",
108 "nlk_cb_mutex-IP6_FW",
109 "nlk_cb_mutex-DNRTMSG",
110 "nlk_cb_mutex-KOBJECT_UEVENT",
111 "nlk_cb_mutex-GENERIC",
113 "nlk_cb_mutex-SCSITRANSPORT",
114 "nlk_cb_mutex-ECRYPTFS",
116 "nlk_cb_mutex-CRYPTO",
127 "nlk_cb_mutex-MAX_LINKS"
130 static int netlink_dump(struct sock *sk);
132 /* nl_table locking explained:
133 * Lookup and traversal are protected with an RCU read-side lock. Insertion
134 * and removal are protected with per bucket lock while using RCU list
135 * modification primitives and may run in parallel to RCU protected lookups.
136 * Destruction of the Netlink socket may only occur *after* nl_table_lock has
137 * been acquired * either during or after the socket has been removed from
138 * the list and after an RCU grace period.
140 DEFINE_RWLOCK(nl_table_lock);
141 EXPORT_SYMBOL_GPL(nl_table_lock);
142 static atomic_t nl_table_users = ATOMIC_INIT(0);
144 #define nl_deref_protected(X) rcu_dereference_protected(X, lockdep_is_held(&nl_table_lock));
146 static BLOCKING_NOTIFIER_HEAD(netlink_chain);
148 static DEFINE_SPINLOCK(netlink_tap_lock);
149 static struct list_head netlink_tap_all __read_mostly;
151 static const struct rhashtable_params netlink_rhashtable_params;
153 static inline u32 netlink_group_mask(u32 group)
155 return group ? 1 << (group - 1) : 0;
158 static struct sk_buff *netlink_to_full_skb(const struct sk_buff *skb,
161 unsigned int len = skb_end_offset(skb);
164 new = alloc_skb(len, gfp_mask);
168 NETLINK_CB(new).portid = NETLINK_CB(skb).portid;
169 NETLINK_CB(new).dst_group = NETLINK_CB(skb).dst_group;
170 NETLINK_CB(new).creds = NETLINK_CB(skb).creds;
172 skb_put_data(new, skb->data, len);
176 int netlink_add_tap(struct netlink_tap *nt)
178 if (unlikely(nt->dev->type != ARPHRD_NETLINK))
181 spin_lock(&netlink_tap_lock);
182 list_add_rcu(&nt->list, &netlink_tap_all);
183 spin_unlock(&netlink_tap_lock);
185 __module_get(nt->module);
189 EXPORT_SYMBOL_GPL(netlink_add_tap);
191 static int __netlink_remove_tap(struct netlink_tap *nt)
194 struct netlink_tap *tmp;
196 spin_lock(&netlink_tap_lock);
198 list_for_each_entry(tmp, &netlink_tap_all, list) {
200 list_del_rcu(&nt->list);
206 pr_warn("__netlink_remove_tap: %p not found\n", nt);
208 spin_unlock(&netlink_tap_lock);
211 module_put(nt->module);
213 return found ? 0 : -ENODEV;
216 int netlink_remove_tap(struct netlink_tap *nt)
220 ret = __netlink_remove_tap(nt);
225 EXPORT_SYMBOL_GPL(netlink_remove_tap);
227 static bool netlink_filter_tap(const struct sk_buff *skb)
229 struct sock *sk = skb->sk;
231 /* We take the more conservative approach and
232 * whitelist socket protocols that may pass.
234 switch (sk->sk_protocol) {
236 case NETLINK_USERSOCK:
237 case NETLINK_SOCK_DIAG:
240 case NETLINK_FIB_LOOKUP:
241 case NETLINK_NETFILTER:
242 case NETLINK_GENERIC:
249 static int __netlink_deliver_tap_skb(struct sk_buff *skb,
250 struct net_device *dev)
252 struct sk_buff *nskb;
253 struct sock *sk = skb->sk;
256 if (!net_eq(dev_net(dev), sock_net(sk)))
261 if (is_vmalloc_addr(skb->head))
262 nskb = netlink_to_full_skb(skb, GFP_ATOMIC);
264 nskb = skb_clone(skb, GFP_ATOMIC);
267 nskb->protocol = htons((u16) sk->sk_protocol);
268 nskb->pkt_type = netlink_is_kernel(sk) ?
269 PACKET_KERNEL : PACKET_USER;
270 skb_reset_network_header(nskb);
271 ret = dev_queue_xmit(nskb);
272 if (unlikely(ret > 0))
273 ret = net_xmit_errno(ret);
280 static void __netlink_deliver_tap(struct sk_buff *skb)
283 struct netlink_tap *tmp;
285 if (!netlink_filter_tap(skb))
288 list_for_each_entry_rcu(tmp, &netlink_tap_all, list) {
289 ret = __netlink_deliver_tap_skb(skb, tmp->dev);
295 static void netlink_deliver_tap(struct sk_buff *skb)
299 if (unlikely(!list_empty(&netlink_tap_all)))
300 __netlink_deliver_tap(skb);
305 static void netlink_deliver_tap_kernel(struct sock *dst, struct sock *src,
308 if (!(netlink_is_kernel(dst) && netlink_is_kernel(src)))
309 netlink_deliver_tap(skb);
312 static void netlink_overrun(struct sock *sk)
314 struct netlink_sock *nlk = nlk_sk(sk);
316 if (!(nlk->flags & NETLINK_F_RECV_NO_ENOBUFS)) {
317 if (!test_and_set_bit(NETLINK_S_CONGESTED,
318 &nlk_sk(sk)->state)) {
319 sk->sk_err = ENOBUFS;
320 sk->sk_error_report(sk);
323 atomic_inc(&sk->sk_drops);
326 static void netlink_rcv_wake(struct sock *sk)
328 struct netlink_sock *nlk = nlk_sk(sk);
330 if (skb_queue_empty(&sk->sk_receive_queue))
331 clear_bit(NETLINK_S_CONGESTED, &nlk->state);
332 if (!test_bit(NETLINK_S_CONGESTED, &nlk->state))
333 wake_up_interruptible(&nlk->wait);
336 static void netlink_skb_destructor(struct sk_buff *skb)
338 if (is_vmalloc_addr(skb->head)) {
340 !atomic_dec_return(&(skb_shinfo(skb)->dataref)))
349 static void netlink_skb_set_owner_r(struct sk_buff *skb, struct sock *sk)
351 WARN_ON(skb->sk != NULL);
353 skb->destructor = netlink_skb_destructor;
354 atomic_add(skb->truesize, &sk->sk_rmem_alloc);
355 sk_mem_charge(sk, skb->truesize);
358 static void netlink_sock_destruct(struct sock *sk)
360 struct netlink_sock *nlk = nlk_sk(sk);
362 if (nlk->cb_running) {
364 nlk->cb.done(&nlk->cb);
365 module_put(nlk->cb.module);
366 kfree_skb(nlk->cb.skb);
369 skb_queue_purge(&sk->sk_receive_queue);
371 if (!sock_flag(sk, SOCK_DEAD)) {
372 printk(KERN_ERR "Freeing alive netlink socket %p\n", sk);
376 WARN_ON(atomic_read(&sk->sk_rmem_alloc));
377 WARN_ON(refcount_read(&sk->sk_wmem_alloc));
378 WARN_ON(nlk_sk(sk)->groups);
381 static void netlink_sock_destruct_work(struct work_struct *work)
383 struct netlink_sock *nlk = container_of(work, struct netlink_sock,
389 /* This lock without WQ_FLAG_EXCLUSIVE is good on UP and it is _very_ bad on
390 * SMP. Look, when several writers sleep and reader wakes them up, all but one
391 * immediately hit write lock and grab all the cpus. Exclusive sleep solves
392 * this, _but_ remember, it adds useless work on UP machines.
395 void netlink_table_grab(void)
396 __acquires(nl_table_lock)
400 write_lock_irq(&nl_table_lock);
402 if (atomic_read(&nl_table_users)) {
403 DECLARE_WAITQUEUE(wait, current);
405 add_wait_queue_exclusive(&nl_table_wait, &wait);
407 set_current_state(TASK_UNINTERRUPTIBLE);
408 if (atomic_read(&nl_table_users) == 0)
410 write_unlock_irq(&nl_table_lock);
412 write_lock_irq(&nl_table_lock);
415 __set_current_state(TASK_RUNNING);
416 remove_wait_queue(&nl_table_wait, &wait);
420 void netlink_table_ungrab(void)
421 __releases(nl_table_lock)
423 write_unlock_irq(&nl_table_lock);
424 wake_up(&nl_table_wait);
428 netlink_lock_table(void)
430 /* read_lock() synchronizes us to netlink_table_grab */
432 read_lock(&nl_table_lock);
433 atomic_inc(&nl_table_users);
434 read_unlock(&nl_table_lock);
438 netlink_unlock_table(void)
440 if (atomic_dec_and_test(&nl_table_users))
441 wake_up(&nl_table_wait);
444 struct netlink_compare_arg
450 /* Doing sizeof directly may yield 4 extra bytes on 64-bit. */
451 #define netlink_compare_arg_len \
452 (offsetof(struct netlink_compare_arg, portid) + sizeof(u32))
454 static inline int netlink_compare(struct rhashtable_compare_arg *arg,
457 const struct netlink_compare_arg *x = arg->key;
458 const struct netlink_sock *nlk = ptr;
460 return nlk->portid != x->portid ||
461 !net_eq(sock_net(&nlk->sk), read_pnet(&x->pnet));
464 static void netlink_compare_arg_init(struct netlink_compare_arg *arg,
465 struct net *net, u32 portid)
467 memset(arg, 0, sizeof(*arg));
468 write_pnet(&arg->pnet, net);
469 arg->portid = portid;
472 static struct sock *__netlink_lookup(struct netlink_table *table, u32 portid,
475 struct netlink_compare_arg arg;
477 netlink_compare_arg_init(&arg, net, portid);
478 return rhashtable_lookup_fast(&table->hash, &arg,
479 netlink_rhashtable_params);
482 static int __netlink_insert(struct netlink_table *table, struct sock *sk)
484 struct netlink_compare_arg arg;
486 netlink_compare_arg_init(&arg, sock_net(sk), nlk_sk(sk)->portid);
487 return rhashtable_lookup_insert_key(&table->hash, &arg,
489 netlink_rhashtable_params);
492 static struct sock *netlink_lookup(struct net *net, int protocol, u32 portid)
494 struct netlink_table *table = &nl_table[protocol];
498 sk = __netlink_lookup(table, portid, net);
506 static const struct proto_ops netlink_ops;
509 netlink_update_listeners(struct sock *sk)
511 struct netlink_table *tbl = &nl_table[sk->sk_protocol];
514 struct listeners *listeners;
516 listeners = nl_deref_protected(tbl->listeners);
520 for (i = 0; i < NLGRPLONGS(tbl->groups); i++) {
522 sk_for_each_bound(sk, &tbl->mc_list) {
523 if (i < NLGRPLONGS(nlk_sk(sk)->ngroups))
524 mask |= nlk_sk(sk)->groups[i];
526 listeners->masks[i] = mask;
528 /* this function is only called with the netlink table "grabbed", which
529 * makes sure updates are visible before bind or setsockopt return. */
532 static int netlink_insert(struct sock *sk, u32 portid)
534 struct netlink_table *table = &nl_table[sk->sk_protocol];
539 err = nlk_sk(sk)->portid == portid ? 0 : -EBUSY;
540 if (nlk_sk(sk)->bound)
544 if (BITS_PER_LONG > 32 &&
545 unlikely(atomic_read(&table->hash.nelems) >= UINT_MAX))
548 nlk_sk(sk)->portid = portid;
551 err = __netlink_insert(table, sk);
553 /* In case the hashtable backend returns with -EBUSY
554 * from here, it must not escape to the caller.
556 if (unlikely(err == -EBUSY))
564 /* We need to ensure that the socket is hashed and visible. */
566 nlk_sk(sk)->bound = portid;
573 static void netlink_remove(struct sock *sk)
575 struct netlink_table *table;
577 table = &nl_table[sk->sk_protocol];
578 if (!rhashtable_remove_fast(&table->hash, &nlk_sk(sk)->node,
579 netlink_rhashtable_params)) {
580 WARN_ON(refcount_read(&sk->sk_refcnt) == 1);
584 netlink_table_grab();
585 if (nlk_sk(sk)->subscriptions) {
586 __sk_del_bind_node(sk);
587 netlink_update_listeners(sk);
589 if (sk->sk_protocol == NETLINK_GENERIC)
590 atomic_inc(&genl_sk_destructing_cnt);
591 netlink_table_ungrab();
594 static struct proto netlink_proto = {
596 .owner = THIS_MODULE,
597 .obj_size = sizeof(struct netlink_sock),
600 static int __netlink_create(struct net *net, struct socket *sock,
601 struct mutex *cb_mutex, int protocol,
605 struct netlink_sock *nlk;
607 sock->ops = &netlink_ops;
609 sk = sk_alloc(net, PF_NETLINK, GFP_KERNEL, &netlink_proto, kern);
613 sock_init_data(sock, sk);
617 nlk->cb_mutex = cb_mutex;
619 nlk->cb_mutex = &nlk->cb_def_mutex;
620 mutex_init(nlk->cb_mutex);
621 lockdep_set_class_and_name(nlk->cb_mutex,
622 nlk_cb_mutex_keys + protocol,
623 nlk_cb_mutex_key_strings[protocol]);
625 init_waitqueue_head(&nlk->wait);
627 sk->sk_destruct = netlink_sock_destruct;
628 sk->sk_protocol = protocol;
632 static int netlink_create(struct net *net, struct socket *sock, int protocol,
635 struct module *module = NULL;
636 struct mutex *cb_mutex;
637 struct netlink_sock *nlk;
638 int (*bind)(struct net *net, int group);
639 void (*unbind)(struct net *net, int group);
642 sock->state = SS_UNCONNECTED;
644 if (sock->type != SOCK_RAW && sock->type != SOCK_DGRAM)
645 return -ESOCKTNOSUPPORT;
647 if (protocol < 0 || protocol >= MAX_LINKS)
648 return -EPROTONOSUPPORT;
650 netlink_lock_table();
651 #ifdef CONFIG_MODULES
652 if (!nl_table[protocol].registered) {
653 netlink_unlock_table();
654 request_module("net-pf-%d-proto-%d", PF_NETLINK, protocol);
655 netlink_lock_table();
658 if (nl_table[protocol].registered &&
659 try_module_get(nl_table[protocol].module))
660 module = nl_table[protocol].module;
662 err = -EPROTONOSUPPORT;
663 cb_mutex = nl_table[protocol].cb_mutex;
664 bind = nl_table[protocol].bind;
665 unbind = nl_table[protocol].unbind;
666 netlink_unlock_table();
671 err = __netlink_create(net, sock, cb_mutex, protocol, kern);
676 sock_prot_inuse_add(net, &netlink_proto, 1);
679 nlk = nlk_sk(sock->sk);
680 nlk->module = module;
681 nlk->netlink_bind = bind;
682 nlk->netlink_unbind = unbind;
691 static void deferred_put_nlk_sk(struct rcu_head *head)
693 struct netlink_sock *nlk = container_of(head, struct netlink_sock, rcu);
694 struct sock *sk = &nlk->sk;
699 if (!refcount_dec_and_test(&sk->sk_refcnt))
702 if (nlk->cb_running && nlk->cb.done) {
703 INIT_WORK(&nlk->work, netlink_sock_destruct_work);
704 schedule_work(&nlk->work);
711 static int netlink_release(struct socket *sock)
713 struct sock *sk = sock->sk;
714 struct netlink_sock *nlk;
724 * OK. Socket is unlinked, any packets that arrive now
728 /* must not acquire netlink_table_lock in any way again before unbind
729 * and notifying genetlink is done as otherwise it might deadlock
731 if (nlk->netlink_unbind) {
734 for (i = 0; i < nlk->ngroups; i++)
735 if (test_bit(i, nlk->groups))
736 nlk->netlink_unbind(sock_net(sk), i + 1);
738 if (sk->sk_protocol == NETLINK_GENERIC &&
739 atomic_dec_return(&genl_sk_destructing_cnt) == 0)
740 wake_up(&genl_sk_destructing_waitq);
743 wake_up_interruptible_all(&nlk->wait);
745 skb_queue_purge(&sk->sk_write_queue);
747 if (nlk->portid && nlk->bound) {
748 struct netlink_notify n = {
750 .protocol = sk->sk_protocol,
751 .portid = nlk->portid,
753 blocking_notifier_call_chain(&netlink_chain,
754 NETLINK_URELEASE, &n);
757 module_put(nlk->module);
759 if (netlink_is_kernel(sk)) {
760 netlink_table_grab();
761 BUG_ON(nl_table[sk->sk_protocol].registered == 0);
762 if (--nl_table[sk->sk_protocol].registered == 0) {
763 struct listeners *old;
765 old = nl_deref_protected(nl_table[sk->sk_protocol].listeners);
766 RCU_INIT_POINTER(nl_table[sk->sk_protocol].listeners, NULL);
768 nl_table[sk->sk_protocol].module = NULL;
769 nl_table[sk->sk_protocol].bind = NULL;
770 nl_table[sk->sk_protocol].unbind = NULL;
771 nl_table[sk->sk_protocol].flags = 0;
772 nl_table[sk->sk_protocol].registered = 0;
774 netlink_table_ungrab();
778 sock_prot_inuse_add(sock_net(sk), &netlink_proto, -1);
780 call_rcu(&nlk->rcu, deferred_put_nlk_sk);
784 static int netlink_autobind(struct socket *sock)
786 struct sock *sk = sock->sk;
787 struct net *net = sock_net(sk);
788 struct netlink_table *table = &nl_table[sk->sk_protocol];
789 s32 portid = task_tgid_vnr(current);
797 ok = !__netlink_lookup(table, portid, net);
800 /* Bind collision, search negative portid values. */
802 /* rover will be in range [S32_MIN, -4097] */
803 rover = S32_MIN + prandom_u32_max(-4096 - S32_MIN);
804 else if (rover >= -4096)
810 err = netlink_insert(sk, portid);
811 if (err == -EADDRINUSE)
814 /* If 2 threads race to autobind, that is fine. */
822 * __netlink_ns_capable - General netlink message capability test
823 * @nsp: NETLINK_CB of the socket buffer holding a netlink command from userspace.
824 * @user_ns: The user namespace of the capability to use
825 * @cap: The capability to use
827 * Test to see if the opener of the socket we received the message
828 * from had when the netlink socket was created and the sender of the
829 * message has has the capability @cap in the user namespace @user_ns.
831 bool __netlink_ns_capable(const struct netlink_skb_parms *nsp,
832 struct user_namespace *user_ns, int cap)
834 return ((nsp->flags & NETLINK_SKB_DST) ||
835 file_ns_capable(nsp->sk->sk_socket->file, user_ns, cap)) &&
836 ns_capable(user_ns, cap);
838 EXPORT_SYMBOL(__netlink_ns_capable);
841 * netlink_ns_capable - General netlink message capability test
842 * @skb: socket buffer holding a netlink command from userspace
843 * @user_ns: The user namespace of the capability to use
844 * @cap: The capability to use
846 * Test to see if the opener of the socket we received the message
847 * from had when the netlink socket was created and the sender of the
848 * message has has the capability @cap in the user namespace @user_ns.
850 bool netlink_ns_capable(const struct sk_buff *skb,
851 struct user_namespace *user_ns, int cap)
853 return __netlink_ns_capable(&NETLINK_CB(skb), user_ns, cap);
855 EXPORT_SYMBOL(netlink_ns_capable);
858 * netlink_capable - Netlink global message capability test
859 * @skb: socket buffer holding a netlink command from userspace
860 * @cap: The capability to use
862 * Test to see if the opener of the socket we received the message
863 * from had when the netlink socket was created and the sender of the
864 * message has has the capability @cap in all user namespaces.
866 bool netlink_capable(const struct sk_buff *skb, int cap)
868 return netlink_ns_capable(skb, &init_user_ns, cap);
870 EXPORT_SYMBOL(netlink_capable);
873 * netlink_net_capable - Netlink network namespace message capability test
874 * @skb: socket buffer holding a netlink command from userspace
875 * @cap: The capability to use
877 * Test to see if the opener of the socket we received the message
878 * from had when the netlink socket was created and the sender of the
879 * message has has the capability @cap over the network namespace of
880 * the socket we received the message from.
882 bool netlink_net_capable(const struct sk_buff *skb, int cap)
884 return netlink_ns_capable(skb, sock_net(skb->sk)->user_ns, cap);
886 EXPORT_SYMBOL(netlink_net_capable);
888 static inline int netlink_allowed(const struct socket *sock, unsigned int flag)
890 return (nl_table[sock->sk->sk_protocol].flags & flag) ||
891 ns_capable(sock_net(sock->sk)->user_ns, CAP_NET_ADMIN);
895 netlink_update_subscriptions(struct sock *sk, unsigned int subscriptions)
897 struct netlink_sock *nlk = nlk_sk(sk);
899 if (nlk->subscriptions && !subscriptions)
900 __sk_del_bind_node(sk);
901 else if (!nlk->subscriptions && subscriptions)
902 sk_add_bind_node(sk, &nl_table[sk->sk_protocol].mc_list);
903 nlk->subscriptions = subscriptions;
906 static int netlink_realloc_groups(struct sock *sk)
908 struct netlink_sock *nlk = nlk_sk(sk);
910 unsigned long *new_groups;
913 netlink_table_grab();
915 groups = nl_table[sk->sk_protocol].groups;
916 if (!nl_table[sk->sk_protocol].registered) {
921 if (nlk->ngroups >= groups)
924 new_groups = krealloc(nlk->groups, NLGRPSZ(groups), GFP_ATOMIC);
925 if (new_groups == NULL) {
929 memset((char *)new_groups + NLGRPSZ(nlk->ngroups), 0,
930 NLGRPSZ(groups) - NLGRPSZ(nlk->ngroups));
932 nlk->groups = new_groups;
933 nlk->ngroups = groups;
935 netlink_table_ungrab();
939 static void netlink_undo_bind(int group, long unsigned int groups,
942 struct netlink_sock *nlk = nlk_sk(sk);
945 if (!nlk->netlink_unbind)
948 for (undo = 0; undo < group; undo++)
949 if (test_bit(undo, &groups))
950 nlk->netlink_unbind(sock_net(sk), undo + 1);
953 static int netlink_bind(struct socket *sock, struct sockaddr *addr,
956 struct sock *sk = sock->sk;
957 struct net *net = sock_net(sk);
958 struct netlink_sock *nlk = nlk_sk(sk);
959 struct sockaddr_nl *nladdr = (struct sockaddr_nl *)addr;
961 long unsigned int groups = nladdr->nl_groups;
964 if (addr_len < sizeof(struct sockaddr_nl))
967 if (nladdr->nl_family != AF_NETLINK)
970 /* Only superuser is allowed to listen multicasts */
972 if (!netlink_allowed(sock, NL_CFG_F_NONROOT_RECV))
974 err = netlink_realloc_groups(sk);
981 /* Ensure nlk->portid is up-to-date. */
984 if (nladdr->nl_pid != nlk->portid)
988 netlink_lock_table();
989 if (nlk->netlink_bind && groups) {
992 for (group = 0; group < nlk->ngroups; group++) {
993 if (!test_bit(group, &groups))
995 err = nlk->netlink_bind(net, group + 1);
998 netlink_undo_bind(group, groups, sk);
1003 /* No need for barriers here as we return to user-space without
1004 * using any of the bound attributes.
1007 err = nladdr->nl_pid ?
1008 netlink_insert(sk, nladdr->nl_pid) :
1009 netlink_autobind(sock);
1011 netlink_undo_bind(nlk->ngroups, groups, sk);
1016 if (!groups && (nlk->groups == NULL || !(u32)nlk->groups[0]))
1018 netlink_unlock_table();
1020 netlink_table_grab();
1021 netlink_update_subscriptions(sk, nlk->subscriptions +
1023 hweight32(nlk->groups[0]));
1024 nlk->groups[0] = (nlk->groups[0] & ~0xffffffffUL) | groups;
1025 netlink_update_listeners(sk);
1026 netlink_table_ungrab();
1031 netlink_unlock_table();
1035 static int netlink_connect(struct socket *sock, struct sockaddr *addr,
1036 int alen, int flags)
1039 struct sock *sk = sock->sk;
1040 struct netlink_sock *nlk = nlk_sk(sk);
1041 struct sockaddr_nl *nladdr = (struct sockaddr_nl *)addr;
1043 if (alen < sizeof(addr->sa_family))
1046 if (addr->sa_family == AF_UNSPEC) {
1047 sk->sk_state = NETLINK_UNCONNECTED;
1048 nlk->dst_portid = 0;
1052 if (addr->sa_family != AF_NETLINK)
1055 if ((nladdr->nl_groups || nladdr->nl_pid) &&
1056 !netlink_allowed(sock, NL_CFG_F_NONROOT_SEND))
1059 /* No need for barriers here as we return to user-space without
1060 * using any of the bound attributes.
1063 err = netlink_autobind(sock);
1066 sk->sk_state = NETLINK_CONNECTED;
1067 nlk->dst_portid = nladdr->nl_pid;
1068 nlk->dst_group = ffs(nladdr->nl_groups);
1074 static int netlink_getname(struct socket *sock, struct sockaddr *addr,
1075 int *addr_len, int peer)
1077 struct sock *sk = sock->sk;
1078 struct netlink_sock *nlk = nlk_sk(sk);
1079 DECLARE_SOCKADDR(struct sockaddr_nl *, nladdr, addr);
1081 nladdr->nl_family = AF_NETLINK;
1083 *addr_len = sizeof(*nladdr);
1086 nladdr->nl_pid = nlk->dst_portid;
1087 nladdr->nl_groups = netlink_group_mask(nlk->dst_group);
1089 nladdr->nl_pid = nlk->portid;
1090 netlink_lock_table();
1091 nladdr->nl_groups = nlk->groups ? nlk->groups[0] : 0;
1092 netlink_unlock_table();
1097 static int netlink_ioctl(struct socket *sock, unsigned int cmd,
1100 /* try to hand this ioctl down to the NIC drivers.
1102 return -ENOIOCTLCMD;
1105 static struct sock *netlink_getsockbyportid(struct sock *ssk, u32 portid)
1108 struct netlink_sock *nlk;
1110 sock = netlink_lookup(sock_net(ssk), ssk->sk_protocol, portid);
1112 return ERR_PTR(-ECONNREFUSED);
1114 /* Don't bother queuing skb if kernel socket has no input function */
1116 if (sock->sk_state == NETLINK_CONNECTED &&
1117 nlk->dst_portid != nlk_sk(ssk)->portid) {
1119 return ERR_PTR(-ECONNREFUSED);
1124 struct sock *netlink_getsockbyfilp(struct file *filp)
1126 struct inode *inode = file_inode(filp);
1129 if (!S_ISSOCK(inode->i_mode))
1130 return ERR_PTR(-ENOTSOCK);
1132 sock = SOCKET_I(inode)->sk;
1133 if (sock->sk_family != AF_NETLINK)
1134 return ERR_PTR(-EINVAL);
1140 static struct sk_buff *netlink_alloc_large_skb(unsigned int size,
1143 struct sk_buff *skb;
1146 if (size <= NLMSG_GOODSIZE || broadcast)
1147 return alloc_skb(size, GFP_KERNEL);
1149 size = SKB_DATA_ALIGN(size) +
1150 SKB_DATA_ALIGN(sizeof(struct skb_shared_info));
1152 data = vmalloc(size);
1156 skb = __build_skb(data, size);
1160 skb->destructor = netlink_skb_destructor;
1166 * Attach a skb to a netlink socket.
1167 * The caller must hold a reference to the destination socket. On error, the
1168 * reference is dropped. The skb is not send to the destination, just all
1169 * all error checks are performed and memory in the queue is reserved.
1171 * < 0: error. skb freed, reference to sock dropped.
1173 * 1: repeat lookup - reference dropped while waiting for socket memory.
1175 int netlink_attachskb(struct sock *sk, struct sk_buff *skb,
1176 long *timeo, struct sock *ssk)
1178 struct netlink_sock *nlk;
1182 if ((atomic_read(&sk->sk_rmem_alloc) > sk->sk_rcvbuf ||
1183 test_bit(NETLINK_S_CONGESTED, &nlk->state))) {
1184 DECLARE_WAITQUEUE(wait, current);
1186 if (!ssk || netlink_is_kernel(ssk))
1187 netlink_overrun(sk);
1193 __set_current_state(TASK_INTERRUPTIBLE);
1194 add_wait_queue(&nlk->wait, &wait);
1196 if ((atomic_read(&sk->sk_rmem_alloc) > sk->sk_rcvbuf ||
1197 test_bit(NETLINK_S_CONGESTED, &nlk->state)) &&
1198 !sock_flag(sk, SOCK_DEAD))
1199 *timeo = schedule_timeout(*timeo);
1201 __set_current_state(TASK_RUNNING);
1202 remove_wait_queue(&nlk->wait, &wait);
1205 if (signal_pending(current)) {
1207 return sock_intr_errno(*timeo);
1211 netlink_skb_set_owner_r(skb, sk);
1215 static int __netlink_sendskb(struct sock *sk, struct sk_buff *skb)
1219 netlink_deliver_tap(skb);
1221 skb_queue_tail(&sk->sk_receive_queue, skb);
1222 sk->sk_data_ready(sk);
1226 int netlink_sendskb(struct sock *sk, struct sk_buff *skb)
1228 int len = __netlink_sendskb(sk, skb);
1234 void netlink_detachskb(struct sock *sk, struct sk_buff *skb)
1240 static struct sk_buff *netlink_trim(struct sk_buff *skb, gfp_t allocation)
1244 WARN_ON(skb->sk != NULL);
1245 delta = skb->end - skb->tail;
1246 if (is_vmalloc_addr(skb->head) || delta * 2 < skb->truesize)
1249 if (skb_shared(skb)) {
1250 struct sk_buff *nskb = skb_clone(skb, allocation);
1257 pskb_expand_head(skb, 0, -delta,
1258 (allocation & ~__GFP_DIRECT_RECLAIM) |
1259 __GFP_NOWARN | __GFP_NORETRY);
1263 static int netlink_unicast_kernel(struct sock *sk, struct sk_buff *skb,
1267 struct netlink_sock *nlk = nlk_sk(sk);
1269 ret = -ECONNREFUSED;
1270 if (nlk->netlink_rcv != NULL) {
1272 netlink_skb_set_owner_r(skb, sk);
1273 NETLINK_CB(skb).sk = ssk;
1274 netlink_deliver_tap_kernel(sk, ssk, skb);
1275 nlk->netlink_rcv(skb);
1284 int netlink_unicast(struct sock *ssk, struct sk_buff *skb,
1285 u32 portid, int nonblock)
1291 skb = netlink_trim(skb, gfp_any());
1293 timeo = sock_sndtimeo(ssk, nonblock);
1295 sk = netlink_getsockbyportid(ssk, portid);
1300 if (netlink_is_kernel(sk))
1301 return netlink_unicast_kernel(sk, skb, ssk);
1303 if (sk_filter(sk, skb)) {
1310 err = netlink_attachskb(sk, skb, &timeo, ssk);
1316 return netlink_sendskb(sk, skb);
1318 EXPORT_SYMBOL(netlink_unicast);
1320 int netlink_has_listeners(struct sock *sk, unsigned int group)
1323 struct listeners *listeners;
1325 BUG_ON(!netlink_is_kernel(sk));
1328 listeners = rcu_dereference(nl_table[sk->sk_protocol].listeners);
1330 if (listeners && group - 1 < nl_table[sk->sk_protocol].groups)
1331 res = test_bit(group - 1, listeners->masks);
1337 EXPORT_SYMBOL_GPL(netlink_has_listeners);
1339 static int netlink_broadcast_deliver(struct sock *sk, struct sk_buff *skb)
1341 struct netlink_sock *nlk = nlk_sk(sk);
1343 if (atomic_read(&sk->sk_rmem_alloc) <= sk->sk_rcvbuf &&
1344 !test_bit(NETLINK_S_CONGESTED, &nlk->state)) {
1345 netlink_skb_set_owner_r(skb, sk);
1346 __netlink_sendskb(sk, skb);
1347 return atomic_read(&sk->sk_rmem_alloc) > (sk->sk_rcvbuf >> 1);
1352 struct netlink_broadcast_data {
1353 struct sock *exclude_sk;
1358 int delivery_failure;
1362 struct sk_buff *skb, *skb2;
1363 int (*tx_filter)(struct sock *dsk, struct sk_buff *skb, void *data);
1367 static void do_one_broadcast(struct sock *sk,
1368 struct netlink_broadcast_data *p)
1370 struct netlink_sock *nlk = nlk_sk(sk);
1373 if (p->exclude_sk == sk)
1376 if (nlk->portid == p->portid || p->group - 1 >= nlk->ngroups ||
1377 !test_bit(p->group - 1, nlk->groups))
1380 if (!net_eq(sock_net(sk), p->net)) {
1381 if (!(nlk->flags & NETLINK_F_LISTEN_ALL_NSID))
1384 if (!peernet_has_id(sock_net(sk), p->net))
1387 if (!file_ns_capable(sk->sk_socket->file, p->net->user_ns,
1393 netlink_overrun(sk);
1398 if (p->skb2 == NULL) {
1399 if (skb_shared(p->skb)) {
1400 p->skb2 = skb_clone(p->skb, p->allocation);
1402 p->skb2 = skb_get(p->skb);
1404 * skb ownership may have been set when
1405 * delivered to a previous socket.
1407 skb_orphan(p->skb2);
1410 if (p->skb2 == NULL) {
1411 netlink_overrun(sk);
1412 /* Clone failed. Notify ALL listeners. */
1414 if (nlk->flags & NETLINK_F_BROADCAST_SEND_ERROR)
1415 p->delivery_failure = 1;
1418 if (p->tx_filter && p->tx_filter(sk, p->skb2, p->tx_data)) {
1423 if (sk_filter(sk, p->skb2)) {
1428 NETLINK_CB(p->skb2).nsid = peernet2id(sock_net(sk), p->net);
1429 if (NETLINK_CB(p->skb2).nsid != NETNSA_NSID_NOT_ASSIGNED)
1430 NETLINK_CB(p->skb2).nsid_is_set = true;
1431 val = netlink_broadcast_deliver(sk, p->skb2);
1433 netlink_overrun(sk);
1434 if (nlk->flags & NETLINK_F_BROADCAST_SEND_ERROR)
1435 p->delivery_failure = 1;
1437 p->congested |= val;
1445 int netlink_broadcast_filtered(struct sock *ssk, struct sk_buff *skb, u32 portid,
1446 u32 group, gfp_t allocation,
1447 int (*filter)(struct sock *dsk, struct sk_buff *skb, void *data),
1450 struct net *net = sock_net(ssk);
1451 struct netlink_broadcast_data info;
1454 skb = netlink_trim(skb, allocation);
1456 info.exclude_sk = ssk;
1458 info.portid = portid;
1461 info.delivery_failure = 0;
1464 info.allocation = allocation;
1467 info.tx_filter = filter;
1468 info.tx_data = filter_data;
1470 /* While we sleep in clone, do not allow to change socket list */
1472 netlink_lock_table();
1474 sk_for_each_bound(sk, &nl_table[ssk->sk_protocol].mc_list)
1475 do_one_broadcast(sk, &info);
1479 netlink_unlock_table();
1481 if (info.delivery_failure) {
1482 kfree_skb(info.skb2);
1485 consume_skb(info.skb2);
1487 if (info.delivered) {
1488 if (info.congested && gfpflags_allow_blocking(allocation))
1494 EXPORT_SYMBOL(netlink_broadcast_filtered);
1496 int netlink_broadcast(struct sock *ssk, struct sk_buff *skb, u32 portid,
1497 u32 group, gfp_t allocation)
1499 return netlink_broadcast_filtered(ssk, skb, portid, group, allocation,
1502 EXPORT_SYMBOL(netlink_broadcast);
1504 struct netlink_set_err_data {
1505 struct sock *exclude_sk;
1511 static int do_one_set_err(struct sock *sk, struct netlink_set_err_data *p)
1513 struct netlink_sock *nlk = nlk_sk(sk);
1516 if (sk == p->exclude_sk)
1519 if (!net_eq(sock_net(sk), sock_net(p->exclude_sk)))
1522 if (nlk->portid == p->portid || p->group - 1 >= nlk->ngroups ||
1523 !test_bit(p->group - 1, nlk->groups))
1526 if (p->code == ENOBUFS && nlk->flags & NETLINK_F_RECV_NO_ENOBUFS) {
1531 sk->sk_err = p->code;
1532 sk->sk_error_report(sk);
1538 * netlink_set_err - report error to broadcast listeners
1539 * @ssk: the kernel netlink socket, as returned by netlink_kernel_create()
1540 * @portid: the PORTID of a process that we want to skip (if any)
1541 * @group: the broadcast group that will notice the error
1542 * @code: error code, must be negative (as usual in kernelspace)
1544 * This function returns the number of broadcast listeners that have set the
1545 * NETLINK_NO_ENOBUFS socket option.
1547 int netlink_set_err(struct sock *ssk, u32 portid, u32 group, int code)
1549 struct netlink_set_err_data info;
1553 info.exclude_sk = ssk;
1554 info.portid = portid;
1556 /* sk->sk_err wants a positive error value */
1559 read_lock(&nl_table_lock);
1561 sk_for_each_bound(sk, &nl_table[ssk->sk_protocol].mc_list)
1562 ret += do_one_set_err(sk, &info);
1564 read_unlock(&nl_table_lock);
1567 EXPORT_SYMBOL(netlink_set_err);
1569 /* must be called with netlink table grabbed */
1570 static void netlink_update_socket_mc(struct netlink_sock *nlk,
1574 int old, new = !!is_new, subscriptions;
1576 old = test_bit(group - 1, nlk->groups);
1577 subscriptions = nlk->subscriptions - old + new;
1579 __set_bit(group - 1, nlk->groups);
1581 __clear_bit(group - 1, nlk->groups);
1582 netlink_update_subscriptions(&nlk->sk, subscriptions);
1583 netlink_update_listeners(&nlk->sk);
1586 static int netlink_setsockopt(struct socket *sock, int level, int optname,
1587 char __user *optval, unsigned int optlen)
1589 struct sock *sk = sock->sk;
1590 struct netlink_sock *nlk = nlk_sk(sk);
1591 unsigned int val = 0;
1594 if (level != SOL_NETLINK)
1595 return -ENOPROTOOPT;
1597 if (optlen >= sizeof(int) &&
1598 get_user(val, (unsigned int __user *)optval))
1602 case NETLINK_PKTINFO:
1604 nlk->flags |= NETLINK_F_RECV_PKTINFO;
1606 nlk->flags &= ~NETLINK_F_RECV_PKTINFO;
1609 case NETLINK_ADD_MEMBERSHIP:
1610 case NETLINK_DROP_MEMBERSHIP: {
1611 if (!netlink_allowed(sock, NL_CFG_F_NONROOT_RECV))
1613 err = netlink_realloc_groups(sk);
1616 if (!val || val - 1 >= nlk->ngroups)
1618 if (optname == NETLINK_ADD_MEMBERSHIP && nlk->netlink_bind) {
1619 err = nlk->netlink_bind(sock_net(sk), val);
1623 netlink_table_grab();
1624 netlink_update_socket_mc(nlk, val,
1625 optname == NETLINK_ADD_MEMBERSHIP);
1626 netlink_table_ungrab();
1627 if (optname == NETLINK_DROP_MEMBERSHIP && nlk->netlink_unbind)
1628 nlk->netlink_unbind(sock_net(sk), val);
1633 case NETLINK_BROADCAST_ERROR:
1635 nlk->flags |= NETLINK_F_BROADCAST_SEND_ERROR;
1637 nlk->flags &= ~NETLINK_F_BROADCAST_SEND_ERROR;
1640 case NETLINK_NO_ENOBUFS:
1642 nlk->flags |= NETLINK_F_RECV_NO_ENOBUFS;
1643 clear_bit(NETLINK_S_CONGESTED, &nlk->state);
1644 wake_up_interruptible(&nlk->wait);
1646 nlk->flags &= ~NETLINK_F_RECV_NO_ENOBUFS;
1650 case NETLINK_LISTEN_ALL_NSID:
1651 if (!ns_capable(sock_net(sk)->user_ns, CAP_NET_BROADCAST))
1655 nlk->flags |= NETLINK_F_LISTEN_ALL_NSID;
1657 nlk->flags &= ~NETLINK_F_LISTEN_ALL_NSID;
1660 case NETLINK_CAP_ACK:
1662 nlk->flags |= NETLINK_F_CAP_ACK;
1664 nlk->flags &= ~NETLINK_F_CAP_ACK;
1667 case NETLINK_EXT_ACK:
1669 nlk->flags |= NETLINK_F_EXT_ACK;
1671 nlk->flags &= ~NETLINK_F_EXT_ACK;
1680 static int netlink_getsockopt(struct socket *sock, int level, int optname,
1681 char __user *optval, int __user *optlen)
1683 struct sock *sk = sock->sk;
1684 struct netlink_sock *nlk = nlk_sk(sk);
1687 if (level != SOL_NETLINK)
1688 return -ENOPROTOOPT;
1690 if (get_user(len, optlen))
1696 case NETLINK_PKTINFO:
1697 if (len < sizeof(int))
1700 val = nlk->flags & NETLINK_F_RECV_PKTINFO ? 1 : 0;
1701 if (put_user(len, optlen) ||
1702 put_user(val, optval))
1706 case NETLINK_BROADCAST_ERROR:
1707 if (len < sizeof(int))
1710 val = nlk->flags & NETLINK_F_BROADCAST_SEND_ERROR ? 1 : 0;
1711 if (put_user(len, optlen) ||
1712 put_user(val, optval))
1716 case NETLINK_NO_ENOBUFS:
1717 if (len < sizeof(int))
1720 val = nlk->flags & NETLINK_F_RECV_NO_ENOBUFS ? 1 : 0;
1721 if (put_user(len, optlen) ||
1722 put_user(val, optval))
1726 case NETLINK_LIST_MEMBERSHIPS: {
1727 int pos, idx, shift;
1730 netlink_lock_table();
1731 for (pos = 0; pos * 8 < nlk->ngroups; pos += sizeof(u32)) {
1732 if (len - pos < sizeof(u32))
1735 idx = pos / sizeof(unsigned long);
1736 shift = (pos % sizeof(unsigned long)) * 8;
1737 if (put_user((u32)(nlk->groups[idx] >> shift),
1738 (u32 __user *)(optval + pos))) {
1743 if (put_user(ALIGN(nlk->ngroups / 8, sizeof(u32)), optlen))
1745 netlink_unlock_table();
1748 case NETLINK_CAP_ACK:
1749 if (len < sizeof(int))
1752 val = nlk->flags & NETLINK_F_CAP_ACK ? 1 : 0;
1753 if (put_user(len, optlen) ||
1754 put_user(val, optval))
1758 case NETLINK_EXT_ACK:
1759 if (len < sizeof(int))
1762 val = nlk->flags & NETLINK_F_EXT_ACK ? 1 : 0;
1763 if (put_user(len, optlen) || put_user(val, optval))
1773 static void netlink_cmsg_recv_pktinfo(struct msghdr *msg, struct sk_buff *skb)
1775 struct nl_pktinfo info;
1777 info.group = NETLINK_CB(skb).dst_group;
1778 put_cmsg(msg, SOL_NETLINK, NETLINK_PKTINFO, sizeof(info), &info);
1781 static void netlink_cmsg_listen_all_nsid(struct sock *sk, struct msghdr *msg,
1782 struct sk_buff *skb)
1784 if (!NETLINK_CB(skb).nsid_is_set)
1787 put_cmsg(msg, SOL_NETLINK, NETLINK_LISTEN_ALL_NSID, sizeof(int),
1788 &NETLINK_CB(skb).nsid);
1791 static int netlink_sendmsg(struct socket *sock, struct msghdr *msg, size_t len)
1793 struct sock *sk = sock->sk;
1794 struct netlink_sock *nlk = nlk_sk(sk);
1795 DECLARE_SOCKADDR(struct sockaddr_nl *, addr, msg->msg_name);
1798 struct sk_buff *skb;
1800 struct scm_cookie scm;
1801 u32 netlink_skb_flags = 0;
1803 if (msg->msg_flags&MSG_OOB)
1806 err = scm_send(sock, msg, &scm, true);
1810 if (msg->msg_namelen) {
1812 if (addr->nl_family != AF_NETLINK)
1814 dst_portid = addr->nl_pid;
1815 dst_group = ffs(addr->nl_groups);
1817 if ((dst_group || dst_portid) &&
1818 !netlink_allowed(sock, NL_CFG_F_NONROOT_SEND))
1820 netlink_skb_flags |= NETLINK_SKB_DST;
1822 dst_portid = nlk->dst_portid;
1823 dst_group = nlk->dst_group;
1827 err = netlink_autobind(sock);
1831 /* Ensure nlk is hashed and visible. */
1836 if (len > sk->sk_sndbuf - 32)
1839 skb = netlink_alloc_large_skb(len, dst_group);
1843 NETLINK_CB(skb).portid = nlk->portid;
1844 NETLINK_CB(skb).dst_group = dst_group;
1845 NETLINK_CB(skb).creds = scm.creds;
1846 NETLINK_CB(skb).flags = netlink_skb_flags;
1849 if (memcpy_from_msg(skb_put(skb, len), msg, len)) {
1854 err = security_netlink_send(sk, skb);
1861 refcount_inc(&skb->users);
1862 netlink_broadcast(sk, skb, dst_portid, dst_group, GFP_KERNEL);
1864 err = netlink_unicast(sk, skb, dst_portid, msg->msg_flags&MSG_DONTWAIT);
1871 static int netlink_recvmsg(struct socket *sock, struct msghdr *msg, size_t len,
1874 struct scm_cookie scm;
1875 struct sock *sk = sock->sk;
1876 struct netlink_sock *nlk = nlk_sk(sk);
1877 int noblock = flags&MSG_DONTWAIT;
1879 struct sk_buff *skb, *data_skb;
1887 skb = skb_recv_datagram(sk, flags, noblock, &err);
1893 #ifdef CONFIG_COMPAT_NETLINK_MESSAGES
1894 if (unlikely(skb_shinfo(skb)->frag_list)) {
1896 * If this skb has a frag_list, then here that means that we
1897 * will have to use the frag_list skb's data for compat tasks
1898 * and the regular skb's data for normal (non-compat) tasks.
1900 * If we need to send the compat skb, assign it to the
1901 * 'data_skb' variable so that it will be used below for data
1902 * copying. We keep 'skb' for everything else, including
1903 * freeing both later.
1905 if (flags & MSG_CMSG_COMPAT)
1906 data_skb = skb_shinfo(skb)->frag_list;
1910 /* Record the max length of recvmsg() calls for future allocations */
1911 nlk->max_recvmsg_len = max(nlk->max_recvmsg_len, len);
1912 nlk->max_recvmsg_len = min_t(size_t, nlk->max_recvmsg_len,
1913 SKB_WITH_OVERHEAD(32768));
1915 copied = data_skb->len;
1917 msg->msg_flags |= MSG_TRUNC;
1921 skb_reset_transport_header(data_skb);
1922 err = skb_copy_datagram_msg(data_skb, 0, msg, copied);
1924 if (msg->msg_name) {
1925 DECLARE_SOCKADDR(struct sockaddr_nl *, addr, msg->msg_name);
1926 addr->nl_family = AF_NETLINK;
1928 addr->nl_pid = NETLINK_CB(skb).portid;
1929 addr->nl_groups = netlink_group_mask(NETLINK_CB(skb).dst_group);
1930 msg->msg_namelen = sizeof(*addr);
1933 if (nlk->flags & NETLINK_F_RECV_PKTINFO)
1934 netlink_cmsg_recv_pktinfo(msg, skb);
1935 if (nlk->flags & NETLINK_F_LISTEN_ALL_NSID)
1936 netlink_cmsg_listen_all_nsid(sk, msg, skb);
1938 memset(&scm, 0, sizeof(scm));
1939 scm.creds = *NETLINK_CREDS(skb);
1940 if (flags & MSG_TRUNC)
1941 copied = data_skb->len;
1943 skb_free_datagram(sk, skb);
1945 if (nlk->cb_running &&
1946 atomic_read(&sk->sk_rmem_alloc) <= sk->sk_rcvbuf / 2) {
1947 ret = netlink_dump(sk);
1950 sk->sk_error_report(sk);
1954 scm_recv(sock, msg, &scm, flags);
1956 netlink_rcv_wake(sk);
1957 return err ? : copied;
1960 static void netlink_data_ready(struct sock *sk)
1966 * We export these functions to other modules. They provide a
1967 * complete set of kernel non-blocking support for message
1972 __netlink_kernel_create(struct net *net, int unit, struct module *module,
1973 struct netlink_kernel_cfg *cfg)
1975 struct socket *sock;
1977 struct netlink_sock *nlk;
1978 struct listeners *listeners = NULL;
1979 struct mutex *cb_mutex = cfg ? cfg->cb_mutex : NULL;
1980 unsigned int groups;
1984 if (unit < 0 || unit >= MAX_LINKS)
1987 if (sock_create_lite(PF_NETLINK, SOCK_DGRAM, unit, &sock))
1990 if (__netlink_create(net, sock, cb_mutex, unit, 1) < 0)
1991 goto out_sock_release_nosk;
1995 if (!cfg || cfg->groups < 32)
1998 groups = cfg->groups;
2000 listeners = kzalloc(sizeof(*listeners) + NLGRPSZ(groups), GFP_KERNEL);
2002 goto out_sock_release;
2004 sk->sk_data_ready = netlink_data_ready;
2005 if (cfg && cfg->input)
2006 nlk_sk(sk)->netlink_rcv = cfg->input;
2008 if (netlink_insert(sk, 0))
2009 goto out_sock_release;
2012 nlk->flags |= NETLINK_F_KERNEL_SOCKET;
2014 netlink_table_grab();
2015 if (!nl_table[unit].registered) {
2016 nl_table[unit].groups = groups;
2017 rcu_assign_pointer(nl_table[unit].listeners, listeners);
2018 nl_table[unit].cb_mutex = cb_mutex;
2019 nl_table[unit].module = module;
2021 nl_table[unit].bind = cfg->bind;
2022 nl_table[unit].unbind = cfg->unbind;
2023 nl_table[unit].flags = cfg->flags;
2025 nl_table[unit].compare = cfg->compare;
2027 nl_table[unit].registered = 1;
2030 nl_table[unit].registered++;
2032 netlink_table_ungrab();
2037 netlink_kernel_release(sk);
2040 out_sock_release_nosk:
2044 EXPORT_SYMBOL(__netlink_kernel_create);
2047 netlink_kernel_release(struct sock *sk)
2049 if (sk == NULL || sk->sk_socket == NULL)
2052 sock_release(sk->sk_socket);
2054 EXPORT_SYMBOL(netlink_kernel_release);
2056 int __netlink_change_ngroups(struct sock *sk, unsigned int groups)
2058 struct listeners *new, *old;
2059 struct netlink_table *tbl = &nl_table[sk->sk_protocol];
2064 if (NLGRPSZ(tbl->groups) < NLGRPSZ(groups)) {
2065 new = kzalloc(sizeof(*new) + NLGRPSZ(groups), GFP_ATOMIC);
2068 old = nl_deref_protected(tbl->listeners);
2069 memcpy(new->masks, old->masks, NLGRPSZ(tbl->groups));
2070 rcu_assign_pointer(tbl->listeners, new);
2072 kfree_rcu(old, rcu);
2074 tbl->groups = groups;
2080 * netlink_change_ngroups - change number of multicast groups
2082 * This changes the number of multicast groups that are available
2083 * on a certain netlink family. Note that it is not possible to
2084 * change the number of groups to below 32. Also note that it does
2085 * not implicitly call netlink_clear_multicast_users() when the
2086 * number of groups is reduced.
2088 * @sk: The kernel netlink socket, as returned by netlink_kernel_create().
2089 * @groups: The new number of groups.
2091 int netlink_change_ngroups(struct sock *sk, unsigned int groups)
2095 netlink_table_grab();
2096 err = __netlink_change_ngroups(sk, groups);
2097 netlink_table_ungrab();
2102 void __netlink_clear_multicast_users(struct sock *ksk, unsigned int group)
2105 struct netlink_table *tbl = &nl_table[ksk->sk_protocol];
2107 sk_for_each_bound(sk, &tbl->mc_list)
2108 netlink_update_socket_mc(nlk_sk(sk), group, 0);
2112 __nlmsg_put(struct sk_buff *skb, u32 portid, u32 seq, int type, int len, int flags)
2114 struct nlmsghdr *nlh;
2115 int size = nlmsg_msg_size(len);
2117 nlh = skb_put(skb, NLMSG_ALIGN(size));
2118 nlh->nlmsg_type = type;
2119 nlh->nlmsg_len = size;
2120 nlh->nlmsg_flags = flags;
2121 nlh->nlmsg_pid = portid;
2122 nlh->nlmsg_seq = seq;
2123 if (!__builtin_constant_p(size) || NLMSG_ALIGN(size) - size != 0)
2124 memset(nlmsg_data(nlh) + len, 0, NLMSG_ALIGN(size) - size);
2127 EXPORT_SYMBOL(__nlmsg_put);
2130 * It looks a bit ugly.
2131 * It would be better to create kernel thread.
2134 static int netlink_dump(struct sock *sk)
2136 struct netlink_sock *nlk = nlk_sk(sk);
2137 struct netlink_callback *cb;
2138 struct sk_buff *skb = NULL;
2139 struct nlmsghdr *nlh;
2140 struct module *module;
2145 mutex_lock(nlk->cb_mutex);
2146 if (!nlk->cb_running) {
2151 if (atomic_read(&sk->sk_rmem_alloc) >= sk->sk_rcvbuf)
2154 /* NLMSG_GOODSIZE is small to avoid high order allocations being
2155 * required, but it makes sense to _attempt_ a 16K bytes allocation
2156 * to reduce number of system calls on dump operations, if user
2157 * ever provided a big enough buffer.
2160 alloc_min_size = max_t(int, cb->min_dump_alloc, NLMSG_GOODSIZE);
2162 if (alloc_min_size < nlk->max_recvmsg_len) {
2163 alloc_size = nlk->max_recvmsg_len;
2164 skb = alloc_skb(alloc_size,
2165 (GFP_KERNEL & ~__GFP_DIRECT_RECLAIM) |
2166 __GFP_NOWARN | __GFP_NORETRY);
2169 alloc_size = alloc_min_size;
2170 skb = alloc_skb(alloc_size, GFP_KERNEL);
2175 /* Trim skb to allocated size. User is expected to provide buffer as
2176 * large as max(min_dump_alloc, 16KiB (mac_recvmsg_len capped at
2177 * netlink_recvmsg())). dump will pack as many smaller messages as
2178 * could fit within the allocated skb. skb is typically allocated
2179 * with larger space than required (could be as much as near 2x the
2180 * requested size with align to next power of 2 approach). Allowing
2181 * dump to use the excess space makes it difficult for a user to have a
2182 * reasonable static buffer based on the expected largest dump of a
2183 * single netdev. The outcome is MSG_TRUNC error.
2185 skb_reserve(skb, skb_tailroom(skb) - alloc_size);
2186 netlink_skb_set_owner_r(skb, sk);
2188 if (nlk->dump_done_errno > 0)
2189 nlk->dump_done_errno = cb->dump(skb, cb);
2191 if (nlk->dump_done_errno > 0 ||
2192 skb_tailroom(skb) < nlmsg_total_size(sizeof(nlk->dump_done_errno))) {
2193 mutex_unlock(nlk->cb_mutex);
2195 if (sk_filter(sk, skb))
2198 __netlink_sendskb(sk, skb);
2202 nlh = nlmsg_put_answer(skb, cb, NLMSG_DONE,
2203 sizeof(nlk->dump_done_errno), NLM_F_MULTI);
2207 nl_dump_check_consistent(cb, nlh);
2209 memcpy(nlmsg_data(nlh), &nlk->dump_done_errno,
2210 sizeof(nlk->dump_done_errno));
2212 if (sk_filter(sk, skb))
2215 __netlink_sendskb(sk, skb);
2220 nlk->cb_running = false;
2221 module = cb->module;
2223 mutex_unlock(nlk->cb_mutex);
2229 mutex_unlock(nlk->cb_mutex);
2234 int __netlink_dump_start(struct sock *ssk, struct sk_buff *skb,
2235 const struct nlmsghdr *nlh,
2236 struct netlink_dump_control *control)
2238 struct netlink_callback *cb;
2240 struct netlink_sock *nlk;
2243 refcount_inc(&skb->users);
2245 sk = netlink_lookup(sock_net(ssk), ssk->sk_protocol, NETLINK_CB(skb).portid);
2247 ret = -ECONNREFUSED;
2252 mutex_lock(nlk->cb_mutex);
2253 /* A dump is in progress... */
2254 if (nlk->cb_running) {
2258 /* add reference of module which cb->dump belongs to */
2259 if (!try_module_get(control->module)) {
2260 ret = -EPROTONOSUPPORT;
2265 memset(cb, 0, sizeof(*cb));
2266 cb->start = control->start;
2267 cb->dump = control->dump;
2268 cb->done = control->done;
2270 cb->data = control->data;
2271 cb->module = control->module;
2272 cb->min_dump_alloc = control->min_dump_alloc;
2276 ret = cb->start(cb);
2281 nlk->cb_running = true;
2282 nlk->dump_done_errno = INT_MAX;
2284 mutex_unlock(nlk->cb_mutex);
2286 ret = netlink_dump(sk);
2293 /* We successfully started a dump, by returning -EINTR we
2294 * signal not to send ACK even if it was requested.
2300 mutex_unlock(nlk->cb_mutex);
2305 EXPORT_SYMBOL(__netlink_dump_start);
2307 void netlink_ack(struct sk_buff *in_skb, struct nlmsghdr *nlh, int err,
2308 const struct netlink_ext_ack *extack)
2310 struct sk_buff *skb;
2311 struct nlmsghdr *rep;
2312 struct nlmsgerr *errmsg;
2313 size_t payload = sizeof(*errmsg);
2315 struct netlink_sock *nlk = nlk_sk(NETLINK_CB(in_skb).sk);
2316 unsigned int flags = 0;
2317 bool nlk_has_extack = nlk->flags & NETLINK_F_EXT_ACK;
2319 /* Error messages get the original request appened, unless the user
2320 * requests to cap the error message, and get extra error data if
2323 if (nlk_has_extack && extack && extack->_msg)
2324 tlvlen += nla_total_size(strlen(extack->_msg) + 1);
2327 if (!(nlk->flags & NETLINK_F_CAP_ACK))
2328 payload += nlmsg_len(nlh);
2330 flags |= NLM_F_CAPPED;
2331 if (nlk_has_extack && extack && extack->bad_attr)
2332 tlvlen += nla_total_size(sizeof(u32));
2334 flags |= NLM_F_CAPPED;
2336 if (nlk_has_extack && extack && extack->cookie_len)
2337 tlvlen += nla_total_size(extack->cookie_len);
2341 flags |= NLM_F_ACK_TLVS;
2343 skb = nlmsg_new(payload + tlvlen, GFP_KERNEL);
2345 NETLINK_CB(in_skb).sk->sk_err = ENOBUFS;
2346 NETLINK_CB(in_skb).sk->sk_error_report(NETLINK_CB(in_skb).sk);
2350 rep = __nlmsg_put(skb, NETLINK_CB(in_skb).portid, nlh->nlmsg_seq,
2351 NLMSG_ERROR, payload, flags);
2352 errmsg = nlmsg_data(rep);
2353 errmsg->error = err;
2354 memcpy(&errmsg->msg, nlh, payload > sizeof(*errmsg) ? nlh->nlmsg_len : sizeof(*nlh));
2356 if (nlk_has_extack && extack) {
2358 WARN_ON(nla_put_string(skb, NLMSGERR_ATTR_MSG,
2362 if (extack->bad_attr &&
2363 !WARN_ON((u8 *)extack->bad_attr < in_skb->data ||
2364 (u8 *)extack->bad_attr >= in_skb->data +
2366 WARN_ON(nla_put_u32(skb, NLMSGERR_ATTR_OFFS,
2367 (u8 *)extack->bad_attr -
2370 if (extack->cookie_len)
2371 WARN_ON(nla_put(skb, NLMSGERR_ATTR_COOKIE,
2377 nlmsg_end(skb, rep);
2379 netlink_unicast(in_skb->sk, skb, NETLINK_CB(in_skb).portid, MSG_DONTWAIT);
2381 EXPORT_SYMBOL(netlink_ack);
2383 int netlink_rcv_skb(struct sk_buff *skb, int (*cb)(struct sk_buff *,
2385 struct netlink_ext_ack *))
2387 struct netlink_ext_ack extack;
2388 struct nlmsghdr *nlh;
2391 while (skb->len >= nlmsg_total_size(0)) {
2394 memset(&extack, 0, sizeof(extack));
2395 nlh = nlmsg_hdr(skb);
2398 if (nlh->nlmsg_len < NLMSG_HDRLEN || skb->len < nlh->nlmsg_len)
2401 /* Only requests are handled by the kernel */
2402 if (!(nlh->nlmsg_flags & NLM_F_REQUEST))
2405 /* Skip control messages */
2406 if (nlh->nlmsg_type < NLMSG_MIN_TYPE)
2409 err = cb(skb, nlh, &extack);
2414 if (nlh->nlmsg_flags & NLM_F_ACK || err)
2415 netlink_ack(skb, nlh, err, &extack);
2418 msglen = NLMSG_ALIGN(nlh->nlmsg_len);
2419 if (msglen > skb->len)
2421 skb_pull(skb, msglen);
2426 EXPORT_SYMBOL(netlink_rcv_skb);
2429 * nlmsg_notify - send a notification netlink message
2430 * @sk: netlink socket to use
2431 * @skb: notification message
2432 * @portid: destination netlink portid for reports or 0
2433 * @group: destination multicast group or 0
2434 * @report: 1 to report back, 0 to disable
2435 * @flags: allocation flags
2437 int nlmsg_notify(struct sock *sk, struct sk_buff *skb, u32 portid,
2438 unsigned int group, int report, gfp_t flags)
2443 int exclude_portid = 0;
2446 refcount_inc(&skb->users);
2447 exclude_portid = portid;
2450 /* errors reported via destination sk->sk_err, but propagate
2451 * delivery errors if NETLINK_BROADCAST_ERROR flag is set */
2452 err = nlmsg_multicast(sk, skb, exclude_portid, group, flags);
2458 err2 = nlmsg_unicast(sk, skb, portid);
2459 if (!err || err == -ESRCH)
2465 EXPORT_SYMBOL(nlmsg_notify);
2467 #ifdef CONFIG_PROC_FS
2468 struct nl_seq_iter {
2469 struct seq_net_private p;
2470 struct rhashtable_iter hti;
2474 static int netlink_walk_start(struct nl_seq_iter *iter)
2478 err = rhashtable_walk_init(&nl_table[iter->link].hash, &iter->hti,
2481 iter->link = MAX_LINKS;
2485 err = rhashtable_walk_start(&iter->hti);
2486 return err == -EAGAIN ? 0 : err;
2489 static void netlink_walk_stop(struct nl_seq_iter *iter)
2491 rhashtable_walk_stop(&iter->hti);
2492 rhashtable_walk_exit(&iter->hti);
2495 static void *__netlink_seq_next(struct seq_file *seq)
2497 struct nl_seq_iter *iter = seq->private;
2498 struct netlink_sock *nlk;
2504 nlk = rhashtable_walk_next(&iter->hti);
2507 if (PTR_ERR(nlk) == -EAGAIN)
2516 netlink_walk_stop(iter);
2517 if (++iter->link >= MAX_LINKS)
2520 err = netlink_walk_start(iter);
2522 return ERR_PTR(err);
2524 } while (sock_net(&nlk->sk) != seq_file_net(seq));
2529 static void *netlink_seq_start(struct seq_file *seq, loff_t *posp)
2531 struct nl_seq_iter *iter = seq->private;
2532 void *obj = SEQ_START_TOKEN;
2538 err = netlink_walk_start(iter);
2540 return ERR_PTR(err);
2542 for (pos = *posp; pos && obj && !IS_ERR(obj); pos--)
2543 obj = __netlink_seq_next(seq);
2548 static void *netlink_seq_next(struct seq_file *seq, void *v, loff_t *pos)
2551 return __netlink_seq_next(seq);
2554 static void netlink_seq_stop(struct seq_file *seq, void *v)
2556 struct nl_seq_iter *iter = seq->private;
2558 if (iter->link >= MAX_LINKS)
2561 netlink_walk_stop(iter);
2565 static int netlink_seq_show(struct seq_file *seq, void *v)
2567 if (v == SEQ_START_TOKEN) {
2569 "sk Eth Pid Groups "
2570 "Rmem Wmem Dump Locks Drops Inode\n");
2573 struct netlink_sock *nlk = nlk_sk(s);
2575 seq_printf(seq, "%pK %-3d %-6u %08x %-8d %-8d %d %-8d %-8d %-8lu\n",
2579 nlk->groups ? (u32)nlk->groups[0] : 0,
2580 sk_rmem_alloc_get(s),
2581 sk_wmem_alloc_get(s),
2583 refcount_read(&s->sk_refcnt),
2584 atomic_read(&s->sk_drops),
2592 static const struct seq_operations netlink_seq_ops = {
2593 .start = netlink_seq_start,
2594 .next = netlink_seq_next,
2595 .stop = netlink_seq_stop,
2596 .show = netlink_seq_show,
2600 static int netlink_seq_open(struct inode *inode, struct file *file)
2602 return seq_open_net(inode, file, &netlink_seq_ops,
2603 sizeof(struct nl_seq_iter));
2606 static const struct file_operations netlink_seq_fops = {
2607 .owner = THIS_MODULE,
2608 .open = netlink_seq_open,
2610 .llseek = seq_lseek,
2611 .release = seq_release_net,
2616 int netlink_register_notifier(struct notifier_block *nb)
2618 return blocking_notifier_chain_register(&netlink_chain, nb);
2620 EXPORT_SYMBOL(netlink_register_notifier);
2622 int netlink_unregister_notifier(struct notifier_block *nb)
2624 return blocking_notifier_chain_unregister(&netlink_chain, nb);
2626 EXPORT_SYMBOL(netlink_unregister_notifier);
2628 static const struct proto_ops netlink_ops = {
2629 .family = PF_NETLINK,
2630 .owner = THIS_MODULE,
2631 .release = netlink_release,
2632 .bind = netlink_bind,
2633 .connect = netlink_connect,
2634 .socketpair = sock_no_socketpair,
2635 .accept = sock_no_accept,
2636 .getname = netlink_getname,
2637 .poll = datagram_poll,
2638 .ioctl = netlink_ioctl,
2639 .listen = sock_no_listen,
2640 .shutdown = sock_no_shutdown,
2641 .setsockopt = netlink_setsockopt,
2642 .getsockopt = netlink_getsockopt,
2643 .sendmsg = netlink_sendmsg,
2644 .recvmsg = netlink_recvmsg,
2645 .mmap = sock_no_mmap,
2646 .sendpage = sock_no_sendpage,
2649 static const struct net_proto_family netlink_family_ops = {
2650 .family = PF_NETLINK,
2651 .create = netlink_create,
2652 .owner = THIS_MODULE, /* for consistency 8) */
2655 static int __net_init netlink_net_init(struct net *net)
2657 #ifdef CONFIG_PROC_FS
2658 if (!proc_create("netlink", 0, net->proc_net, &netlink_seq_fops))
2664 static void __net_exit netlink_net_exit(struct net *net)
2666 #ifdef CONFIG_PROC_FS
2667 remove_proc_entry("netlink", net->proc_net);
2671 static void __init netlink_add_usersock_entry(void)
2673 struct listeners *listeners;
2676 listeners = kzalloc(sizeof(*listeners) + NLGRPSZ(groups), GFP_KERNEL);
2678 panic("netlink_add_usersock_entry: Cannot allocate listeners\n");
2680 netlink_table_grab();
2682 nl_table[NETLINK_USERSOCK].groups = groups;
2683 rcu_assign_pointer(nl_table[NETLINK_USERSOCK].listeners, listeners);
2684 nl_table[NETLINK_USERSOCK].module = THIS_MODULE;
2685 nl_table[NETLINK_USERSOCK].registered = 1;
2686 nl_table[NETLINK_USERSOCK].flags = NL_CFG_F_NONROOT_SEND;
2688 netlink_table_ungrab();
2691 static struct pernet_operations __net_initdata netlink_net_ops = {
2692 .init = netlink_net_init,
2693 .exit = netlink_net_exit,
2696 static inline u32 netlink_hash(const void *data, u32 len, u32 seed)
2698 const struct netlink_sock *nlk = data;
2699 struct netlink_compare_arg arg;
2701 netlink_compare_arg_init(&arg, sock_net(&nlk->sk), nlk->portid);
2702 return jhash2((u32 *)&arg, netlink_compare_arg_len / sizeof(u32), seed);
2705 static const struct rhashtable_params netlink_rhashtable_params = {
2706 .head_offset = offsetof(struct netlink_sock, node),
2707 .key_len = netlink_compare_arg_len,
2708 .obj_hashfn = netlink_hash,
2709 .obj_cmpfn = netlink_compare,
2710 .automatic_shrinking = true,
2713 static int __init netlink_proto_init(void)
2716 int err = proto_register(&netlink_proto, 0);
2721 BUILD_BUG_ON(sizeof(struct netlink_skb_parms) > FIELD_SIZEOF(struct sk_buff, cb));
2723 nl_table = kcalloc(MAX_LINKS, sizeof(*nl_table), GFP_KERNEL);
2727 for (i = 0; i < MAX_LINKS; i++) {
2728 if (rhashtable_init(&nl_table[i].hash,
2729 &netlink_rhashtable_params) < 0) {
2731 rhashtable_destroy(&nl_table[i].hash);
2737 INIT_LIST_HEAD(&netlink_tap_all);
2739 netlink_add_usersock_entry();
2741 sock_register(&netlink_family_ops);
2742 register_pernet_subsys(&netlink_net_ops);
2743 /* The netlink device handler may be needed early. */
2748 panic("netlink_init: Cannot allocate nl_table\n");
2751 core_initcall(netlink_proto_init);