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>
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 struct lock_class_key nlk_cb_mutex_keys[MAX_LINKS];
101 static const char *const nlk_cb_mutex_key_strings[MAX_LINKS + 1] = {
102 "nlk_cb_mutex-ROUTE",
104 "nlk_cb_mutex-USERSOCK",
105 "nlk_cb_mutex-FIREWALL",
106 "nlk_cb_mutex-SOCK_DIAG",
107 "nlk_cb_mutex-NFLOG",
109 "nlk_cb_mutex-SELINUX",
110 "nlk_cb_mutex-ISCSI",
111 "nlk_cb_mutex-AUDIT",
112 "nlk_cb_mutex-FIB_LOOKUP",
113 "nlk_cb_mutex-CONNECTOR",
114 "nlk_cb_mutex-NETFILTER",
115 "nlk_cb_mutex-IP6_FW",
116 "nlk_cb_mutex-DNRTMSG",
117 "nlk_cb_mutex-KOBJECT_UEVENT",
118 "nlk_cb_mutex-GENERIC",
120 "nlk_cb_mutex-SCSITRANSPORT",
121 "nlk_cb_mutex-ECRYPTFS",
123 "nlk_cb_mutex-CRYPTO",
134 "nlk_cb_mutex-MAX_LINKS"
137 static int netlink_dump(struct sock *sk);
138 static void netlink_skb_destructor(struct sk_buff *skb);
140 /* nl_table locking explained:
141 * Lookup and traversal are protected with an RCU read-side lock. Insertion
142 * and removal are protected with per bucket lock while using RCU list
143 * modification primitives and may run in parallel to RCU protected lookups.
144 * Destruction of the Netlink socket may only occur *after* nl_table_lock has
145 * been acquired * either during or after the socket has been removed from
146 * the list and after an RCU grace period.
148 DEFINE_RWLOCK(nl_table_lock);
149 EXPORT_SYMBOL_GPL(nl_table_lock);
150 static atomic_t nl_table_users = ATOMIC_INIT(0);
152 #define nl_deref_protected(X) rcu_dereference_protected(X, lockdep_is_held(&nl_table_lock));
154 static BLOCKING_NOTIFIER_HEAD(netlink_chain);
156 static DEFINE_SPINLOCK(netlink_tap_lock);
157 static struct list_head netlink_tap_all __read_mostly;
159 static const struct rhashtable_params netlink_rhashtable_params;
161 static inline u32 netlink_group_mask(u32 group)
163 return group ? 1 << (group - 1) : 0;
166 static struct sk_buff *netlink_to_full_skb(const struct sk_buff *skb,
169 unsigned int len = skb_end_offset(skb);
172 new = alloc_skb(len, gfp_mask);
176 NETLINK_CB(new).portid = NETLINK_CB(skb).portid;
177 NETLINK_CB(new).dst_group = NETLINK_CB(skb).dst_group;
178 NETLINK_CB(new).creds = NETLINK_CB(skb).creds;
180 memcpy(skb_put(new, len), skb->data, len);
184 int netlink_add_tap(struct netlink_tap *nt)
186 if (unlikely(nt->dev->type != ARPHRD_NETLINK))
189 spin_lock(&netlink_tap_lock);
190 list_add_rcu(&nt->list, &netlink_tap_all);
191 spin_unlock(&netlink_tap_lock);
193 __module_get(nt->module);
197 EXPORT_SYMBOL_GPL(netlink_add_tap);
199 static int __netlink_remove_tap(struct netlink_tap *nt)
202 struct netlink_tap *tmp;
204 spin_lock(&netlink_tap_lock);
206 list_for_each_entry(tmp, &netlink_tap_all, list) {
208 list_del_rcu(&nt->list);
214 pr_warn("__netlink_remove_tap: %p not found\n", nt);
216 spin_unlock(&netlink_tap_lock);
219 module_put(nt->module);
221 return found ? 0 : -ENODEV;
224 int netlink_remove_tap(struct netlink_tap *nt)
228 ret = __netlink_remove_tap(nt);
233 EXPORT_SYMBOL_GPL(netlink_remove_tap);
235 static bool netlink_filter_tap(const struct sk_buff *skb)
237 struct sock *sk = skb->sk;
239 /* We take the more conservative approach and
240 * whitelist socket protocols that may pass.
242 switch (sk->sk_protocol) {
244 case NETLINK_USERSOCK:
245 case NETLINK_SOCK_DIAG:
248 case NETLINK_FIB_LOOKUP:
249 case NETLINK_NETFILTER:
250 case NETLINK_GENERIC:
257 static int __netlink_deliver_tap_skb(struct sk_buff *skb,
258 struct net_device *dev)
260 struct sk_buff *nskb;
261 struct sock *sk = skb->sk;
266 if (is_vmalloc_addr(skb->head))
267 nskb = netlink_to_full_skb(skb, GFP_ATOMIC);
269 nskb = skb_clone(skb, GFP_ATOMIC);
272 nskb->protocol = htons((u16) sk->sk_protocol);
273 nskb->pkt_type = netlink_is_kernel(sk) ?
274 PACKET_KERNEL : PACKET_USER;
275 skb_reset_network_header(nskb);
276 ret = dev_queue_xmit(nskb);
277 if (unlikely(ret > 0))
278 ret = net_xmit_errno(ret);
285 static void __netlink_deliver_tap(struct sk_buff *skb)
288 struct netlink_tap *tmp;
290 if (!netlink_filter_tap(skb))
293 list_for_each_entry_rcu(tmp, &netlink_tap_all, list) {
294 ret = __netlink_deliver_tap_skb(skb, tmp->dev);
300 static void netlink_deliver_tap(struct sk_buff *skb)
304 if (unlikely(!list_empty(&netlink_tap_all)))
305 __netlink_deliver_tap(skb);
310 static void netlink_deliver_tap_kernel(struct sock *dst, struct sock *src,
313 if (!(netlink_is_kernel(dst) && netlink_is_kernel(src)))
314 netlink_deliver_tap(skb);
317 static void netlink_overrun(struct sock *sk)
319 struct netlink_sock *nlk = nlk_sk(sk);
321 if (!(nlk->flags & NETLINK_F_RECV_NO_ENOBUFS)) {
322 if (!test_and_set_bit(NETLINK_S_CONGESTED,
323 &nlk_sk(sk)->state)) {
324 sk->sk_err = ENOBUFS;
325 sk->sk_error_report(sk);
328 atomic_inc(&sk->sk_drops);
331 static void netlink_rcv_wake(struct sock *sk)
333 struct netlink_sock *nlk = nlk_sk(sk);
335 if (skb_queue_empty(&sk->sk_receive_queue))
336 clear_bit(NETLINK_S_CONGESTED, &nlk->state);
337 if (!test_bit(NETLINK_S_CONGESTED, &nlk->state))
338 wake_up_interruptible(&nlk->wait);
341 static void netlink_skb_destructor(struct sk_buff *skb)
343 if (is_vmalloc_addr(skb->head)) {
345 !atomic_dec_return(&(skb_shinfo(skb)->dataref)))
354 static void netlink_skb_set_owner_r(struct sk_buff *skb, struct sock *sk)
356 WARN_ON(skb->sk != NULL);
358 skb->destructor = netlink_skb_destructor;
359 atomic_add(skb->truesize, &sk->sk_rmem_alloc);
360 sk_mem_charge(sk, skb->truesize);
363 static void netlink_sock_destruct(struct sock *sk)
365 struct netlink_sock *nlk = nlk_sk(sk);
367 if (nlk->cb_running) {
369 nlk->cb.done(&nlk->cb);
370 module_put(nlk->cb.module);
371 kfree_skb(nlk->cb.skb);
374 skb_queue_purge(&sk->sk_receive_queue);
376 if (!sock_flag(sk, SOCK_DEAD)) {
377 printk(KERN_ERR "Freeing alive netlink socket %p\n", sk);
381 WARN_ON(atomic_read(&sk->sk_rmem_alloc));
382 WARN_ON(atomic_read(&sk->sk_wmem_alloc));
383 WARN_ON(nlk_sk(sk)->groups);
386 static void netlink_sock_destruct_work(struct work_struct *work)
388 struct netlink_sock *nlk = container_of(work, struct netlink_sock,
394 /* This lock without WQ_FLAG_EXCLUSIVE is good on UP and it is _very_ bad on
395 * SMP. Look, when several writers sleep and reader wakes them up, all but one
396 * immediately hit write lock and grab all the cpus. Exclusive sleep solves
397 * this, _but_ remember, it adds useless work on UP machines.
400 void netlink_table_grab(void)
401 __acquires(nl_table_lock)
405 write_lock_irq(&nl_table_lock);
407 if (atomic_read(&nl_table_users)) {
408 DECLARE_WAITQUEUE(wait, current);
410 add_wait_queue_exclusive(&nl_table_wait, &wait);
412 set_current_state(TASK_UNINTERRUPTIBLE);
413 if (atomic_read(&nl_table_users) == 0)
415 write_unlock_irq(&nl_table_lock);
417 write_lock_irq(&nl_table_lock);
420 __set_current_state(TASK_RUNNING);
421 remove_wait_queue(&nl_table_wait, &wait);
425 void netlink_table_ungrab(void)
426 __releases(nl_table_lock)
428 write_unlock_irq(&nl_table_lock);
429 wake_up(&nl_table_wait);
433 netlink_lock_table(void)
435 /* read_lock() synchronizes us to netlink_table_grab */
437 read_lock(&nl_table_lock);
438 atomic_inc(&nl_table_users);
439 read_unlock(&nl_table_lock);
443 netlink_unlock_table(void)
445 if (atomic_dec_and_test(&nl_table_users))
446 wake_up(&nl_table_wait);
449 struct netlink_compare_arg
455 /* Doing sizeof directly may yield 4 extra bytes on 64-bit. */
456 #define netlink_compare_arg_len \
457 (offsetof(struct netlink_compare_arg, portid) + sizeof(u32))
459 static inline int netlink_compare(struct rhashtable_compare_arg *arg,
462 const struct netlink_compare_arg *x = arg->key;
463 const struct netlink_sock *nlk = ptr;
465 return nlk->portid != x->portid ||
466 !net_eq(sock_net(&nlk->sk), read_pnet(&x->pnet));
469 static void netlink_compare_arg_init(struct netlink_compare_arg *arg,
470 struct net *net, u32 portid)
472 memset(arg, 0, sizeof(*arg));
473 write_pnet(&arg->pnet, net);
474 arg->portid = portid;
477 static struct sock *__netlink_lookup(struct netlink_table *table, u32 portid,
480 struct netlink_compare_arg arg;
482 netlink_compare_arg_init(&arg, net, portid);
483 return rhashtable_lookup_fast(&table->hash, &arg,
484 netlink_rhashtable_params);
487 static int __netlink_insert(struct netlink_table *table, struct sock *sk)
489 struct netlink_compare_arg arg;
491 netlink_compare_arg_init(&arg, sock_net(sk), nlk_sk(sk)->portid);
492 return rhashtable_lookup_insert_key(&table->hash, &arg,
494 netlink_rhashtable_params);
497 static struct sock *netlink_lookup(struct net *net, int protocol, u32 portid)
499 struct netlink_table *table = &nl_table[protocol];
503 sk = __netlink_lookup(table, portid, net);
511 static const struct proto_ops netlink_ops;
514 netlink_update_listeners(struct sock *sk)
516 struct netlink_table *tbl = &nl_table[sk->sk_protocol];
519 struct listeners *listeners;
521 listeners = nl_deref_protected(tbl->listeners);
525 for (i = 0; i < NLGRPLONGS(tbl->groups); i++) {
527 sk_for_each_bound(sk, &tbl->mc_list) {
528 if (i < NLGRPLONGS(nlk_sk(sk)->ngroups))
529 mask |= nlk_sk(sk)->groups[i];
531 listeners->masks[i] = mask;
533 /* this function is only called with the netlink table "grabbed", which
534 * makes sure updates are visible before bind or setsockopt return. */
537 static int netlink_insert(struct sock *sk, u32 portid)
539 struct netlink_table *table = &nl_table[sk->sk_protocol];
544 err = nlk_sk(sk)->portid == portid ? 0 : -EBUSY;
545 if (nlk_sk(sk)->bound)
549 if (BITS_PER_LONG > 32 &&
550 unlikely(atomic_read(&table->hash.nelems) >= UINT_MAX))
553 nlk_sk(sk)->portid = portid;
556 err = __netlink_insert(table, sk);
558 /* In case the hashtable backend returns with -EBUSY
559 * from here, it must not escape to the caller.
561 if (unlikely(err == -EBUSY))
569 /* We need to ensure that the socket is hashed and visible. */
571 nlk_sk(sk)->bound = portid;
578 static void netlink_remove(struct sock *sk)
580 struct netlink_table *table;
582 table = &nl_table[sk->sk_protocol];
583 if (!rhashtable_remove_fast(&table->hash, &nlk_sk(sk)->node,
584 netlink_rhashtable_params)) {
585 WARN_ON(atomic_read(&sk->sk_refcnt) == 1);
589 netlink_table_grab();
590 if (nlk_sk(sk)->subscriptions) {
591 __sk_del_bind_node(sk);
592 netlink_update_listeners(sk);
594 if (sk->sk_protocol == NETLINK_GENERIC)
595 atomic_inc(&genl_sk_destructing_cnt);
596 netlink_table_ungrab();
599 static struct proto netlink_proto = {
601 .owner = THIS_MODULE,
602 .obj_size = sizeof(struct netlink_sock),
605 static int __netlink_create(struct net *net, struct socket *sock,
606 struct mutex *cb_mutex, int protocol,
610 struct netlink_sock *nlk;
612 sock->ops = &netlink_ops;
614 sk = sk_alloc(net, PF_NETLINK, GFP_KERNEL, &netlink_proto, kern);
618 sock_init_data(sock, sk);
622 nlk->cb_mutex = cb_mutex;
624 nlk->cb_mutex = &nlk->cb_def_mutex;
625 mutex_init(nlk->cb_mutex);
626 lockdep_set_class_and_name(nlk->cb_mutex,
627 nlk_cb_mutex_keys + protocol,
628 nlk_cb_mutex_key_strings[protocol]);
630 init_waitqueue_head(&nlk->wait);
632 sk->sk_destruct = netlink_sock_destruct;
633 sk->sk_protocol = protocol;
637 static int netlink_create(struct net *net, struct socket *sock, int protocol,
640 struct module *module = NULL;
641 struct mutex *cb_mutex;
642 struct netlink_sock *nlk;
643 int (*bind)(struct net *net, int group);
644 void (*unbind)(struct net *net, int group);
647 sock->state = SS_UNCONNECTED;
649 if (sock->type != SOCK_RAW && sock->type != SOCK_DGRAM)
650 return -ESOCKTNOSUPPORT;
652 if (protocol < 0 || protocol >= MAX_LINKS)
653 return -EPROTONOSUPPORT;
655 netlink_lock_table();
656 #ifdef CONFIG_MODULES
657 if (!nl_table[protocol].registered) {
658 netlink_unlock_table();
659 request_module("net-pf-%d-proto-%d", PF_NETLINK, protocol);
660 netlink_lock_table();
663 if (nl_table[protocol].registered &&
664 try_module_get(nl_table[protocol].module))
665 module = nl_table[protocol].module;
667 err = -EPROTONOSUPPORT;
668 cb_mutex = nl_table[protocol].cb_mutex;
669 bind = nl_table[protocol].bind;
670 unbind = nl_table[protocol].unbind;
671 netlink_unlock_table();
676 err = __netlink_create(net, sock, cb_mutex, protocol, kern);
681 sock_prot_inuse_add(net, &netlink_proto, 1);
684 nlk = nlk_sk(sock->sk);
685 nlk->module = module;
686 nlk->netlink_bind = bind;
687 nlk->netlink_unbind = unbind;
696 static void deferred_put_nlk_sk(struct rcu_head *head)
698 struct netlink_sock *nlk = container_of(head, struct netlink_sock, rcu);
699 struct sock *sk = &nlk->sk;
701 if (!atomic_dec_and_test(&sk->sk_refcnt))
704 if (nlk->cb_running && nlk->cb.done) {
705 INIT_WORK(&nlk->work, netlink_sock_destruct_work);
706 schedule_work(&nlk->work);
713 static int netlink_release(struct socket *sock)
715 struct sock *sk = sock->sk;
716 struct netlink_sock *nlk;
726 * OK. Socket is unlinked, any packets that arrive now
730 /* must not acquire netlink_table_lock in any way again before unbind
731 * and notifying genetlink is done as otherwise it might deadlock
733 if (nlk->netlink_unbind) {
736 for (i = 0; i < nlk->ngroups; i++)
737 if (test_bit(i, nlk->groups))
738 nlk->netlink_unbind(sock_net(sk), i + 1);
740 if (sk->sk_protocol == NETLINK_GENERIC &&
741 atomic_dec_return(&genl_sk_destructing_cnt) == 0)
742 wake_up(&genl_sk_destructing_waitq);
745 wake_up_interruptible_all(&nlk->wait);
747 skb_queue_purge(&sk->sk_write_queue);
749 if (nlk->portid && nlk->bound) {
750 struct netlink_notify n = {
752 .protocol = sk->sk_protocol,
753 .portid = nlk->portid,
755 blocking_notifier_call_chain(&netlink_chain,
756 NETLINK_URELEASE, &n);
759 module_put(nlk->module);
761 if (netlink_is_kernel(sk)) {
762 netlink_table_grab();
763 BUG_ON(nl_table[sk->sk_protocol].registered == 0);
764 if (--nl_table[sk->sk_protocol].registered == 0) {
765 struct listeners *old;
767 old = nl_deref_protected(nl_table[sk->sk_protocol].listeners);
768 RCU_INIT_POINTER(nl_table[sk->sk_protocol].listeners, NULL);
770 nl_table[sk->sk_protocol].module = NULL;
771 nl_table[sk->sk_protocol].bind = NULL;
772 nl_table[sk->sk_protocol].unbind = NULL;
773 nl_table[sk->sk_protocol].flags = 0;
774 nl_table[sk->sk_protocol].registered = 0;
776 netlink_table_ungrab();
783 sock_prot_inuse_add(sock_net(sk), &netlink_proto, -1);
785 call_rcu(&nlk->rcu, deferred_put_nlk_sk);
789 static int netlink_autobind(struct socket *sock)
791 struct sock *sk = sock->sk;
792 struct net *net = sock_net(sk);
793 struct netlink_table *table = &nl_table[sk->sk_protocol];
794 s32 portid = task_tgid_vnr(current);
802 ok = !__netlink_lookup(table, portid, net);
805 /* Bind collision, search negative portid values. */
807 /* rover will be in range [S32_MIN, -4097] */
808 rover = S32_MIN + prandom_u32_max(-4096 - S32_MIN);
809 else if (rover >= -4096)
815 err = netlink_insert(sk, portid);
816 if (err == -EADDRINUSE)
819 /* If 2 threads race to autobind, that is fine. */
827 * __netlink_ns_capable - General netlink message capability test
828 * @nsp: NETLINK_CB of the socket buffer holding a netlink command from userspace.
829 * @user_ns: The user namespace of the capability to use
830 * @cap: The capability to use
832 * Test to see if the opener of the socket we received the message
833 * from had when the netlink socket was created and the sender of the
834 * message has has the capability @cap in the user namespace @user_ns.
836 bool __netlink_ns_capable(const struct netlink_skb_parms *nsp,
837 struct user_namespace *user_ns, int cap)
839 return ((nsp->flags & NETLINK_SKB_DST) ||
840 file_ns_capable(nsp->sk->sk_socket->file, user_ns, cap)) &&
841 ns_capable(user_ns, cap);
843 EXPORT_SYMBOL(__netlink_ns_capable);
846 * netlink_ns_capable - General netlink message capability test
847 * @skb: socket buffer holding a netlink command from userspace
848 * @user_ns: The user namespace of the capability to use
849 * @cap: The capability to use
851 * Test to see if the opener of the socket we received the message
852 * from had when the netlink socket was created and the sender of the
853 * message has has the capability @cap in the user namespace @user_ns.
855 bool netlink_ns_capable(const struct sk_buff *skb,
856 struct user_namespace *user_ns, int cap)
858 return __netlink_ns_capable(&NETLINK_CB(skb), user_ns, cap);
860 EXPORT_SYMBOL(netlink_ns_capable);
863 * netlink_capable - Netlink global message capability test
864 * @skb: socket buffer holding a netlink command from userspace
865 * @cap: The capability to use
867 * Test to see if the opener of the socket we received the message
868 * from had when the netlink socket was created and the sender of the
869 * message has has the capability @cap in all user namespaces.
871 bool netlink_capable(const struct sk_buff *skb, int cap)
873 return netlink_ns_capable(skb, &init_user_ns, cap);
875 EXPORT_SYMBOL(netlink_capable);
878 * netlink_net_capable - Netlink network namespace message capability test
879 * @skb: socket buffer holding a netlink command from userspace
880 * @cap: The capability to use
882 * Test to see if the opener of the socket we received the message
883 * from had when the netlink socket was created and the sender of the
884 * message has has the capability @cap over the network namespace of
885 * the socket we received the message from.
887 bool netlink_net_capable(const struct sk_buff *skb, int cap)
889 return netlink_ns_capable(skb, sock_net(skb->sk)->user_ns, cap);
891 EXPORT_SYMBOL(netlink_net_capable);
893 static inline int netlink_allowed(const struct socket *sock, unsigned int flag)
895 return (nl_table[sock->sk->sk_protocol].flags & flag) ||
896 ns_capable(sock_net(sock->sk)->user_ns, CAP_NET_ADMIN);
900 netlink_update_subscriptions(struct sock *sk, unsigned int subscriptions)
902 struct netlink_sock *nlk = nlk_sk(sk);
904 if (nlk->subscriptions && !subscriptions)
905 __sk_del_bind_node(sk);
906 else if (!nlk->subscriptions && subscriptions)
907 sk_add_bind_node(sk, &nl_table[sk->sk_protocol].mc_list);
908 nlk->subscriptions = subscriptions;
911 static int netlink_realloc_groups(struct sock *sk)
913 struct netlink_sock *nlk = nlk_sk(sk);
915 unsigned long *new_groups;
918 netlink_table_grab();
920 groups = nl_table[sk->sk_protocol].groups;
921 if (!nl_table[sk->sk_protocol].registered) {
926 if (nlk->ngroups >= groups)
929 new_groups = krealloc(nlk->groups, NLGRPSZ(groups), GFP_ATOMIC);
930 if (new_groups == NULL) {
934 memset((char *)new_groups + NLGRPSZ(nlk->ngroups), 0,
935 NLGRPSZ(groups) - NLGRPSZ(nlk->ngroups));
937 nlk->groups = new_groups;
938 nlk->ngroups = groups;
940 netlink_table_ungrab();
944 static void netlink_undo_bind(int group, long unsigned int groups,
947 struct netlink_sock *nlk = nlk_sk(sk);
950 if (!nlk->netlink_unbind)
953 for (undo = 0; undo < group; undo++)
954 if (test_bit(undo, &groups))
955 nlk->netlink_unbind(sock_net(sk), undo + 1);
958 static int netlink_bind(struct socket *sock, struct sockaddr *addr,
961 struct sock *sk = sock->sk;
962 struct net *net = sock_net(sk);
963 struct netlink_sock *nlk = nlk_sk(sk);
964 struct sockaddr_nl *nladdr = (struct sockaddr_nl *)addr;
966 long unsigned int groups = nladdr->nl_groups;
969 if (addr_len < sizeof(struct sockaddr_nl))
972 if (nladdr->nl_family != AF_NETLINK)
975 /* Only superuser is allowed to listen multicasts */
977 if (!netlink_allowed(sock, NL_CFG_F_NONROOT_RECV))
979 err = netlink_realloc_groups(sk);
986 /* Ensure nlk->portid is up-to-date. */
989 if (nladdr->nl_pid != nlk->portid)
993 if (nlk->netlink_bind && groups) {
996 for (group = 0; group < nlk->ngroups; group++) {
997 if (!test_bit(group, &groups))
999 err = nlk->netlink_bind(net, group + 1);
1002 netlink_undo_bind(group, groups, sk);
1007 /* No need for barriers here as we return to user-space without
1008 * using any of the bound attributes.
1011 err = nladdr->nl_pid ?
1012 netlink_insert(sk, nladdr->nl_pid) :
1013 netlink_autobind(sock);
1015 netlink_undo_bind(nlk->ngroups, groups, sk);
1020 if (!groups && (nlk->groups == NULL || !(u32)nlk->groups[0]))
1023 netlink_table_grab();
1024 netlink_update_subscriptions(sk, nlk->subscriptions +
1026 hweight32(nlk->groups[0]));
1027 nlk->groups[0] = (nlk->groups[0] & ~0xffffffffUL) | groups;
1028 netlink_update_listeners(sk);
1029 netlink_table_ungrab();
1034 static int netlink_connect(struct socket *sock, struct sockaddr *addr,
1035 int alen, int flags)
1038 struct sock *sk = sock->sk;
1039 struct netlink_sock *nlk = nlk_sk(sk);
1040 struct sockaddr_nl *nladdr = (struct sockaddr_nl *)addr;
1042 if (alen < sizeof(addr->sa_family))
1045 if (addr->sa_family == AF_UNSPEC) {
1046 sk->sk_state = NETLINK_UNCONNECTED;
1047 nlk->dst_portid = 0;
1051 if (addr->sa_family != AF_NETLINK)
1054 if ((nladdr->nl_groups || nladdr->nl_pid) &&
1055 !netlink_allowed(sock, NL_CFG_F_NONROOT_SEND))
1058 /* No need for barriers here as we return to user-space without
1059 * using any of the bound attributes.
1062 err = netlink_autobind(sock);
1065 sk->sk_state = NETLINK_CONNECTED;
1066 nlk->dst_portid = nladdr->nl_pid;
1067 nlk->dst_group = ffs(nladdr->nl_groups);
1073 static int netlink_getname(struct socket *sock, struct sockaddr *addr,
1074 int *addr_len, int peer)
1076 struct sock *sk = sock->sk;
1077 struct netlink_sock *nlk = nlk_sk(sk);
1078 DECLARE_SOCKADDR(struct sockaddr_nl *, nladdr, addr);
1080 nladdr->nl_family = AF_NETLINK;
1082 *addr_len = sizeof(*nladdr);
1085 nladdr->nl_pid = nlk->dst_portid;
1086 nladdr->nl_groups = netlink_group_mask(nlk->dst_group);
1088 nladdr->nl_pid = nlk->portid;
1089 nladdr->nl_groups = nlk->groups ? nlk->groups[0] : 0;
1094 static int netlink_ioctl(struct socket *sock, unsigned int cmd,
1097 /* try to hand this ioctl down to the NIC drivers.
1099 return -ENOIOCTLCMD;
1102 static struct sock *netlink_getsockbyportid(struct sock *ssk, u32 portid)
1105 struct netlink_sock *nlk;
1107 sock = netlink_lookup(sock_net(ssk), ssk->sk_protocol, portid);
1109 return ERR_PTR(-ECONNREFUSED);
1111 /* Don't bother queuing skb if kernel socket has no input function */
1113 if (sock->sk_state == NETLINK_CONNECTED &&
1114 nlk->dst_portid != nlk_sk(ssk)->portid) {
1116 return ERR_PTR(-ECONNREFUSED);
1121 struct sock *netlink_getsockbyfilp(struct file *filp)
1123 struct inode *inode = file_inode(filp);
1126 if (!S_ISSOCK(inode->i_mode))
1127 return ERR_PTR(-ENOTSOCK);
1129 sock = SOCKET_I(inode)->sk;
1130 if (sock->sk_family != AF_NETLINK)
1131 return ERR_PTR(-EINVAL);
1137 static struct sk_buff *netlink_alloc_large_skb(unsigned int size,
1140 struct sk_buff *skb;
1143 if (size <= NLMSG_GOODSIZE || broadcast)
1144 return alloc_skb(size, GFP_KERNEL);
1146 size = SKB_DATA_ALIGN(size) +
1147 SKB_DATA_ALIGN(sizeof(struct skb_shared_info));
1149 data = vmalloc(size);
1153 skb = __build_skb(data, size);
1157 skb->destructor = netlink_skb_destructor;
1163 * Attach a skb to a netlink socket.
1164 * The caller must hold a reference to the destination socket. On error, the
1165 * reference is dropped. The skb is not send to the destination, just all
1166 * all error checks are performed and memory in the queue is reserved.
1168 * < 0: error. skb freed, reference to sock dropped.
1170 * 1: repeat lookup - reference dropped while waiting for socket memory.
1172 int netlink_attachskb(struct sock *sk, struct sk_buff *skb,
1173 long *timeo, struct sock *ssk)
1175 struct netlink_sock *nlk;
1179 if ((atomic_read(&sk->sk_rmem_alloc) > sk->sk_rcvbuf ||
1180 test_bit(NETLINK_S_CONGESTED, &nlk->state))) {
1181 DECLARE_WAITQUEUE(wait, current);
1183 if (!ssk || netlink_is_kernel(ssk))
1184 netlink_overrun(sk);
1190 __set_current_state(TASK_INTERRUPTIBLE);
1191 add_wait_queue(&nlk->wait, &wait);
1193 if ((atomic_read(&sk->sk_rmem_alloc) > sk->sk_rcvbuf ||
1194 test_bit(NETLINK_S_CONGESTED, &nlk->state)) &&
1195 !sock_flag(sk, SOCK_DEAD))
1196 *timeo = schedule_timeout(*timeo);
1198 __set_current_state(TASK_RUNNING);
1199 remove_wait_queue(&nlk->wait, &wait);
1202 if (signal_pending(current)) {
1204 return sock_intr_errno(*timeo);
1208 netlink_skb_set_owner_r(skb, sk);
1212 static int __netlink_sendskb(struct sock *sk, struct sk_buff *skb)
1216 netlink_deliver_tap(skb);
1218 skb_queue_tail(&sk->sk_receive_queue, skb);
1219 sk->sk_data_ready(sk);
1223 int netlink_sendskb(struct sock *sk, struct sk_buff *skb)
1225 int len = __netlink_sendskb(sk, skb);
1231 void netlink_detachskb(struct sock *sk, struct sk_buff *skb)
1237 static struct sk_buff *netlink_trim(struct sk_buff *skb, gfp_t allocation)
1241 WARN_ON(skb->sk != NULL);
1242 delta = skb->end - skb->tail;
1243 if (is_vmalloc_addr(skb->head) || delta * 2 < skb->truesize)
1246 if (skb_shared(skb)) {
1247 struct sk_buff *nskb = skb_clone(skb, allocation);
1254 pskb_expand_head(skb, 0, -delta,
1255 (allocation & ~__GFP_DIRECT_RECLAIM) |
1256 __GFP_NOWARN | __GFP_NORETRY);
1260 static int netlink_unicast_kernel(struct sock *sk, struct sk_buff *skb,
1264 struct netlink_sock *nlk = nlk_sk(sk);
1266 ret = -ECONNREFUSED;
1267 if (nlk->netlink_rcv != NULL) {
1269 netlink_skb_set_owner_r(skb, sk);
1270 NETLINK_CB(skb).sk = ssk;
1271 netlink_deliver_tap_kernel(sk, ssk, skb);
1272 nlk->netlink_rcv(skb);
1281 int netlink_unicast(struct sock *ssk, struct sk_buff *skb,
1282 u32 portid, int nonblock)
1288 skb = netlink_trim(skb, gfp_any());
1290 timeo = sock_sndtimeo(ssk, nonblock);
1292 sk = netlink_getsockbyportid(ssk, portid);
1297 if (netlink_is_kernel(sk))
1298 return netlink_unicast_kernel(sk, skb, ssk);
1300 if (sk_filter(sk, skb)) {
1307 err = netlink_attachskb(sk, skb, &timeo, ssk);
1313 return netlink_sendskb(sk, skb);
1315 EXPORT_SYMBOL(netlink_unicast);
1317 int netlink_has_listeners(struct sock *sk, unsigned int group)
1320 struct listeners *listeners;
1322 BUG_ON(!netlink_is_kernel(sk));
1325 listeners = rcu_dereference(nl_table[sk->sk_protocol].listeners);
1327 if (listeners && group - 1 < nl_table[sk->sk_protocol].groups)
1328 res = test_bit(group - 1, listeners->masks);
1334 EXPORT_SYMBOL_GPL(netlink_has_listeners);
1336 static int netlink_broadcast_deliver(struct sock *sk, struct sk_buff *skb)
1338 struct netlink_sock *nlk = nlk_sk(sk);
1340 if (atomic_read(&sk->sk_rmem_alloc) <= sk->sk_rcvbuf &&
1341 !test_bit(NETLINK_S_CONGESTED, &nlk->state)) {
1342 netlink_skb_set_owner_r(skb, sk);
1343 __netlink_sendskb(sk, skb);
1344 return atomic_read(&sk->sk_rmem_alloc) > (sk->sk_rcvbuf >> 1);
1349 struct netlink_broadcast_data {
1350 struct sock *exclude_sk;
1355 int delivery_failure;
1359 struct sk_buff *skb, *skb2;
1360 int (*tx_filter)(struct sock *dsk, struct sk_buff *skb, void *data);
1364 static void do_one_broadcast(struct sock *sk,
1365 struct netlink_broadcast_data *p)
1367 struct netlink_sock *nlk = nlk_sk(sk);
1370 if (p->exclude_sk == sk)
1373 if (nlk->portid == p->portid || p->group - 1 >= nlk->ngroups ||
1374 !test_bit(p->group - 1, nlk->groups))
1377 if (!net_eq(sock_net(sk), p->net)) {
1378 if (!(nlk->flags & NETLINK_F_LISTEN_ALL_NSID))
1381 if (!peernet_has_id(sock_net(sk), p->net))
1384 if (!file_ns_capable(sk->sk_socket->file, p->net->user_ns,
1390 netlink_overrun(sk);
1395 if (p->skb2 == NULL) {
1396 if (skb_shared(p->skb)) {
1397 p->skb2 = skb_clone(p->skb, p->allocation);
1399 p->skb2 = skb_get(p->skb);
1401 * skb ownership may have been set when
1402 * delivered to a previous socket.
1404 skb_orphan(p->skb2);
1407 if (p->skb2 == NULL) {
1408 netlink_overrun(sk);
1409 /* Clone failed. Notify ALL listeners. */
1411 if (nlk->flags & NETLINK_F_BROADCAST_SEND_ERROR)
1412 p->delivery_failure = 1;
1415 if (p->tx_filter && p->tx_filter(sk, p->skb2, p->tx_data)) {
1420 if (sk_filter(sk, p->skb2)) {
1425 NETLINK_CB(p->skb2).nsid = peernet2id(sock_net(sk), p->net);
1426 NETLINK_CB(p->skb2).nsid_is_set = true;
1427 val = netlink_broadcast_deliver(sk, p->skb2);
1429 netlink_overrun(sk);
1430 if (nlk->flags & NETLINK_F_BROADCAST_SEND_ERROR)
1431 p->delivery_failure = 1;
1433 p->congested |= val;
1441 int netlink_broadcast_filtered(struct sock *ssk, struct sk_buff *skb, u32 portid,
1442 u32 group, gfp_t allocation,
1443 int (*filter)(struct sock *dsk, struct sk_buff *skb, void *data),
1446 struct net *net = sock_net(ssk);
1447 struct netlink_broadcast_data info;
1450 skb = netlink_trim(skb, allocation);
1452 info.exclude_sk = ssk;
1454 info.portid = portid;
1457 info.delivery_failure = 0;
1460 info.allocation = allocation;
1463 info.tx_filter = filter;
1464 info.tx_data = filter_data;
1466 /* While we sleep in clone, do not allow to change socket list */
1468 netlink_lock_table();
1470 sk_for_each_bound(sk, &nl_table[ssk->sk_protocol].mc_list)
1471 do_one_broadcast(sk, &info);
1475 netlink_unlock_table();
1477 if (info.delivery_failure) {
1478 kfree_skb(info.skb2);
1481 consume_skb(info.skb2);
1483 if (info.delivered) {
1484 if (info.congested && gfpflags_allow_blocking(allocation))
1490 EXPORT_SYMBOL(netlink_broadcast_filtered);
1492 int netlink_broadcast(struct sock *ssk, struct sk_buff *skb, u32 portid,
1493 u32 group, gfp_t allocation)
1495 return netlink_broadcast_filtered(ssk, skb, portid, group, allocation,
1498 EXPORT_SYMBOL(netlink_broadcast);
1500 struct netlink_set_err_data {
1501 struct sock *exclude_sk;
1507 static int do_one_set_err(struct sock *sk, struct netlink_set_err_data *p)
1509 struct netlink_sock *nlk = nlk_sk(sk);
1512 if (sk == p->exclude_sk)
1515 if (!net_eq(sock_net(sk), sock_net(p->exclude_sk)))
1518 if (nlk->portid == p->portid || p->group - 1 >= nlk->ngroups ||
1519 !test_bit(p->group - 1, nlk->groups))
1522 if (p->code == ENOBUFS && nlk->flags & NETLINK_F_RECV_NO_ENOBUFS) {
1527 sk->sk_err = p->code;
1528 sk->sk_error_report(sk);
1534 * netlink_set_err - report error to broadcast listeners
1535 * @ssk: the kernel netlink socket, as returned by netlink_kernel_create()
1536 * @portid: the PORTID of a process that we want to skip (if any)
1537 * @group: the broadcast group that will notice the error
1538 * @code: error code, must be negative (as usual in kernelspace)
1540 * This function returns the number of broadcast listeners that have set the
1541 * NETLINK_NO_ENOBUFS socket option.
1543 int netlink_set_err(struct sock *ssk, u32 portid, u32 group, int code)
1545 struct netlink_set_err_data info;
1549 info.exclude_sk = ssk;
1550 info.portid = portid;
1552 /* sk->sk_err wants a positive error value */
1555 read_lock(&nl_table_lock);
1557 sk_for_each_bound(sk, &nl_table[ssk->sk_protocol].mc_list)
1558 ret += do_one_set_err(sk, &info);
1560 read_unlock(&nl_table_lock);
1563 EXPORT_SYMBOL(netlink_set_err);
1565 /* must be called with netlink table grabbed */
1566 static void netlink_update_socket_mc(struct netlink_sock *nlk,
1570 int old, new = !!is_new, subscriptions;
1572 old = test_bit(group - 1, nlk->groups);
1573 subscriptions = nlk->subscriptions - old + new;
1575 __set_bit(group - 1, nlk->groups);
1577 __clear_bit(group - 1, nlk->groups);
1578 netlink_update_subscriptions(&nlk->sk, subscriptions);
1579 netlink_update_listeners(&nlk->sk);
1582 static int netlink_setsockopt(struct socket *sock, int level, int optname,
1583 char __user *optval, unsigned int optlen)
1585 struct sock *sk = sock->sk;
1586 struct netlink_sock *nlk = nlk_sk(sk);
1587 unsigned int val = 0;
1590 if (level != SOL_NETLINK)
1591 return -ENOPROTOOPT;
1593 if (optlen >= sizeof(int) &&
1594 get_user(val, (unsigned int __user *)optval))
1598 case NETLINK_PKTINFO:
1600 nlk->flags |= NETLINK_F_RECV_PKTINFO;
1602 nlk->flags &= ~NETLINK_F_RECV_PKTINFO;
1605 case NETLINK_ADD_MEMBERSHIP:
1606 case NETLINK_DROP_MEMBERSHIP: {
1607 if (!netlink_allowed(sock, NL_CFG_F_NONROOT_RECV))
1609 err = netlink_realloc_groups(sk);
1612 if (!val || val - 1 >= nlk->ngroups)
1614 if (optname == NETLINK_ADD_MEMBERSHIP && nlk->netlink_bind) {
1615 err = nlk->netlink_bind(sock_net(sk), val);
1619 netlink_table_grab();
1620 netlink_update_socket_mc(nlk, val,
1621 optname == NETLINK_ADD_MEMBERSHIP);
1622 netlink_table_ungrab();
1623 if (optname == NETLINK_DROP_MEMBERSHIP && nlk->netlink_unbind)
1624 nlk->netlink_unbind(sock_net(sk), val);
1629 case NETLINK_BROADCAST_ERROR:
1631 nlk->flags |= NETLINK_F_BROADCAST_SEND_ERROR;
1633 nlk->flags &= ~NETLINK_F_BROADCAST_SEND_ERROR;
1636 case NETLINK_NO_ENOBUFS:
1638 nlk->flags |= NETLINK_F_RECV_NO_ENOBUFS;
1639 clear_bit(NETLINK_S_CONGESTED, &nlk->state);
1640 wake_up_interruptible(&nlk->wait);
1642 nlk->flags &= ~NETLINK_F_RECV_NO_ENOBUFS;
1646 case NETLINK_LISTEN_ALL_NSID:
1647 if (!ns_capable(sock_net(sk)->user_ns, CAP_NET_BROADCAST))
1651 nlk->flags |= NETLINK_F_LISTEN_ALL_NSID;
1653 nlk->flags &= ~NETLINK_F_LISTEN_ALL_NSID;
1656 case NETLINK_CAP_ACK:
1658 nlk->flags |= NETLINK_F_CAP_ACK;
1660 nlk->flags &= ~NETLINK_F_CAP_ACK;
1669 static int netlink_getsockopt(struct socket *sock, int level, int optname,
1670 char __user *optval, int __user *optlen)
1672 struct sock *sk = sock->sk;
1673 struct netlink_sock *nlk = nlk_sk(sk);
1676 if (level != SOL_NETLINK)
1677 return -ENOPROTOOPT;
1679 if (get_user(len, optlen))
1685 case NETLINK_PKTINFO:
1686 if (len < sizeof(int))
1689 val = nlk->flags & NETLINK_F_RECV_PKTINFO ? 1 : 0;
1690 if (put_user(len, optlen) ||
1691 put_user(val, optval))
1695 case NETLINK_BROADCAST_ERROR:
1696 if (len < sizeof(int))
1699 val = nlk->flags & NETLINK_F_BROADCAST_SEND_ERROR ? 1 : 0;
1700 if (put_user(len, optlen) ||
1701 put_user(val, optval))
1705 case NETLINK_NO_ENOBUFS:
1706 if (len < sizeof(int))
1709 val = nlk->flags & NETLINK_F_RECV_NO_ENOBUFS ? 1 : 0;
1710 if (put_user(len, optlen) ||
1711 put_user(val, optval))
1715 case NETLINK_LIST_MEMBERSHIPS: {
1716 int pos, idx, shift;
1719 netlink_lock_table();
1720 for (pos = 0; pos * 8 < nlk->ngroups; pos += sizeof(u32)) {
1721 if (len - pos < sizeof(u32))
1724 idx = pos / sizeof(unsigned long);
1725 shift = (pos % sizeof(unsigned long)) * 8;
1726 if (put_user((u32)(nlk->groups[idx] >> shift),
1727 (u32 __user *)(optval + pos))) {
1732 if (put_user(ALIGN(nlk->ngroups / 8, sizeof(u32)), optlen))
1734 netlink_unlock_table();
1737 case NETLINK_CAP_ACK:
1738 if (len < sizeof(int))
1741 val = nlk->flags & NETLINK_F_CAP_ACK ? 1 : 0;
1742 if (put_user(len, optlen) ||
1743 put_user(val, optval))
1753 static void netlink_cmsg_recv_pktinfo(struct msghdr *msg, struct sk_buff *skb)
1755 struct nl_pktinfo info;
1757 info.group = NETLINK_CB(skb).dst_group;
1758 put_cmsg(msg, SOL_NETLINK, NETLINK_PKTINFO, sizeof(info), &info);
1761 static void netlink_cmsg_listen_all_nsid(struct sock *sk, struct msghdr *msg,
1762 struct sk_buff *skb)
1764 if (!NETLINK_CB(skb).nsid_is_set)
1767 put_cmsg(msg, SOL_NETLINK, NETLINK_LISTEN_ALL_NSID, sizeof(int),
1768 &NETLINK_CB(skb).nsid);
1771 static int netlink_sendmsg(struct socket *sock, struct msghdr *msg, size_t len)
1773 struct sock *sk = sock->sk;
1774 struct netlink_sock *nlk = nlk_sk(sk);
1775 DECLARE_SOCKADDR(struct sockaddr_nl *, addr, msg->msg_name);
1778 struct sk_buff *skb;
1780 struct scm_cookie scm;
1781 u32 netlink_skb_flags = 0;
1783 if (msg->msg_flags&MSG_OOB)
1786 err = scm_send(sock, msg, &scm, true);
1790 if (msg->msg_namelen) {
1792 if (addr->nl_family != AF_NETLINK)
1794 dst_portid = addr->nl_pid;
1795 dst_group = ffs(addr->nl_groups);
1797 if ((dst_group || dst_portid) &&
1798 !netlink_allowed(sock, NL_CFG_F_NONROOT_SEND))
1800 netlink_skb_flags |= NETLINK_SKB_DST;
1802 dst_portid = nlk->dst_portid;
1803 dst_group = nlk->dst_group;
1807 err = netlink_autobind(sock);
1811 /* Ensure nlk is hashed and visible. */
1816 if (len > sk->sk_sndbuf - 32)
1819 skb = netlink_alloc_large_skb(len, dst_group);
1823 NETLINK_CB(skb).portid = nlk->portid;
1824 NETLINK_CB(skb).dst_group = dst_group;
1825 NETLINK_CB(skb).creds = scm.creds;
1826 NETLINK_CB(skb).flags = netlink_skb_flags;
1829 if (memcpy_from_msg(skb_put(skb, len), msg, len)) {
1834 err = security_netlink_send(sk, skb);
1841 atomic_inc(&skb->users);
1842 netlink_broadcast(sk, skb, dst_portid, dst_group, GFP_KERNEL);
1844 err = netlink_unicast(sk, skb, dst_portid, msg->msg_flags&MSG_DONTWAIT);
1851 static int netlink_recvmsg(struct socket *sock, struct msghdr *msg, size_t len,
1854 struct scm_cookie scm;
1855 struct sock *sk = sock->sk;
1856 struct netlink_sock *nlk = nlk_sk(sk);
1857 int noblock = flags&MSG_DONTWAIT;
1859 struct sk_buff *skb, *data_skb;
1867 skb = skb_recv_datagram(sk, flags, noblock, &err);
1873 #ifdef CONFIG_COMPAT_NETLINK_MESSAGES
1874 if (unlikely(skb_shinfo(skb)->frag_list)) {
1876 * If this skb has a frag_list, then here that means that we
1877 * will have to use the frag_list skb's data for compat tasks
1878 * and the regular skb's data for normal (non-compat) tasks.
1880 * If we need to send the compat skb, assign it to the
1881 * 'data_skb' variable so that it will be used below for data
1882 * copying. We keep 'skb' for everything else, including
1883 * freeing both later.
1885 if (flags & MSG_CMSG_COMPAT)
1886 data_skb = skb_shinfo(skb)->frag_list;
1890 /* Record the max length of recvmsg() calls for future allocations */
1891 nlk->max_recvmsg_len = max(nlk->max_recvmsg_len, len);
1892 nlk->max_recvmsg_len = min_t(size_t, nlk->max_recvmsg_len,
1893 SKB_WITH_OVERHEAD(32768));
1895 copied = data_skb->len;
1897 msg->msg_flags |= MSG_TRUNC;
1901 skb_reset_transport_header(data_skb);
1902 err = skb_copy_datagram_msg(data_skb, 0, msg, copied);
1904 if (msg->msg_name) {
1905 DECLARE_SOCKADDR(struct sockaddr_nl *, addr, msg->msg_name);
1906 addr->nl_family = AF_NETLINK;
1908 addr->nl_pid = NETLINK_CB(skb).portid;
1909 addr->nl_groups = netlink_group_mask(NETLINK_CB(skb).dst_group);
1910 msg->msg_namelen = sizeof(*addr);
1913 if (nlk->flags & NETLINK_F_RECV_PKTINFO)
1914 netlink_cmsg_recv_pktinfo(msg, skb);
1915 if (nlk->flags & NETLINK_F_LISTEN_ALL_NSID)
1916 netlink_cmsg_listen_all_nsid(sk, msg, skb);
1918 memset(&scm, 0, sizeof(scm));
1919 scm.creds = *NETLINK_CREDS(skb);
1920 if (flags & MSG_TRUNC)
1921 copied = data_skb->len;
1923 skb_free_datagram(sk, skb);
1925 if (nlk->cb_running &&
1926 atomic_read(&sk->sk_rmem_alloc) <= sk->sk_rcvbuf / 2) {
1927 ret = netlink_dump(sk);
1930 sk->sk_error_report(sk);
1934 scm_recv(sock, msg, &scm, flags);
1936 netlink_rcv_wake(sk);
1937 return err ? : copied;
1940 static void netlink_data_ready(struct sock *sk)
1946 * We export these functions to other modules. They provide a
1947 * complete set of kernel non-blocking support for message
1952 __netlink_kernel_create(struct net *net, int unit, struct module *module,
1953 struct netlink_kernel_cfg *cfg)
1955 struct socket *sock;
1957 struct netlink_sock *nlk;
1958 struct listeners *listeners = NULL;
1959 struct mutex *cb_mutex = cfg ? cfg->cb_mutex : NULL;
1960 unsigned int groups;
1964 if (unit < 0 || unit >= MAX_LINKS)
1967 if (sock_create_lite(PF_NETLINK, SOCK_DGRAM, unit, &sock))
1970 if (__netlink_create(net, sock, cb_mutex, unit, 1) < 0)
1971 goto out_sock_release_nosk;
1975 if (!cfg || cfg->groups < 32)
1978 groups = cfg->groups;
1980 listeners = kzalloc(sizeof(*listeners) + NLGRPSZ(groups), GFP_KERNEL);
1982 goto out_sock_release;
1984 sk->sk_data_ready = netlink_data_ready;
1985 if (cfg && cfg->input)
1986 nlk_sk(sk)->netlink_rcv = cfg->input;
1988 if (netlink_insert(sk, 0))
1989 goto out_sock_release;
1992 nlk->flags |= NETLINK_F_KERNEL_SOCKET;
1994 netlink_table_grab();
1995 if (!nl_table[unit].registered) {
1996 nl_table[unit].groups = groups;
1997 rcu_assign_pointer(nl_table[unit].listeners, listeners);
1998 nl_table[unit].cb_mutex = cb_mutex;
1999 nl_table[unit].module = module;
2001 nl_table[unit].bind = cfg->bind;
2002 nl_table[unit].unbind = cfg->unbind;
2003 nl_table[unit].flags = cfg->flags;
2005 nl_table[unit].compare = cfg->compare;
2007 nl_table[unit].registered = 1;
2010 nl_table[unit].registered++;
2012 netlink_table_ungrab();
2017 netlink_kernel_release(sk);
2020 out_sock_release_nosk:
2024 EXPORT_SYMBOL(__netlink_kernel_create);
2027 netlink_kernel_release(struct sock *sk)
2029 if (sk == NULL || sk->sk_socket == NULL)
2032 sock_release(sk->sk_socket);
2034 EXPORT_SYMBOL(netlink_kernel_release);
2036 int __netlink_change_ngroups(struct sock *sk, unsigned int groups)
2038 struct listeners *new, *old;
2039 struct netlink_table *tbl = &nl_table[sk->sk_protocol];
2044 if (NLGRPSZ(tbl->groups) < NLGRPSZ(groups)) {
2045 new = kzalloc(sizeof(*new) + NLGRPSZ(groups), GFP_ATOMIC);
2048 old = nl_deref_protected(tbl->listeners);
2049 memcpy(new->masks, old->masks, NLGRPSZ(tbl->groups));
2050 rcu_assign_pointer(tbl->listeners, new);
2052 kfree_rcu(old, rcu);
2054 tbl->groups = groups;
2060 * netlink_change_ngroups - change number of multicast groups
2062 * This changes the number of multicast groups that are available
2063 * on a certain netlink family. Note that it is not possible to
2064 * change the number of groups to below 32. Also note that it does
2065 * not implicitly call netlink_clear_multicast_users() when the
2066 * number of groups is reduced.
2068 * @sk: The kernel netlink socket, as returned by netlink_kernel_create().
2069 * @groups: The new number of groups.
2071 int netlink_change_ngroups(struct sock *sk, unsigned int groups)
2075 netlink_table_grab();
2076 err = __netlink_change_ngroups(sk, groups);
2077 netlink_table_ungrab();
2082 void __netlink_clear_multicast_users(struct sock *ksk, unsigned int group)
2085 struct netlink_table *tbl = &nl_table[ksk->sk_protocol];
2087 sk_for_each_bound(sk, &tbl->mc_list)
2088 netlink_update_socket_mc(nlk_sk(sk), group, 0);
2092 __nlmsg_put(struct sk_buff *skb, u32 portid, u32 seq, int type, int len, int flags)
2094 struct nlmsghdr *nlh;
2095 int size = nlmsg_msg_size(len);
2097 nlh = (struct nlmsghdr *)skb_put(skb, NLMSG_ALIGN(size));
2098 nlh->nlmsg_type = type;
2099 nlh->nlmsg_len = size;
2100 nlh->nlmsg_flags = flags;
2101 nlh->nlmsg_pid = portid;
2102 nlh->nlmsg_seq = seq;
2103 if (!__builtin_constant_p(size) || NLMSG_ALIGN(size) - size != 0)
2104 memset(nlmsg_data(nlh) + len, 0, NLMSG_ALIGN(size) - size);
2107 EXPORT_SYMBOL(__nlmsg_put);
2110 * It looks a bit ugly.
2111 * It would be better to create kernel thread.
2114 static int netlink_dump(struct sock *sk)
2116 struct netlink_sock *nlk = nlk_sk(sk);
2117 struct netlink_callback *cb;
2118 struct sk_buff *skb = NULL;
2119 struct nlmsghdr *nlh;
2120 struct module *module;
2121 int len, err = -ENOBUFS;
2125 mutex_lock(nlk->cb_mutex);
2126 if (!nlk->cb_running) {
2131 if (atomic_read(&sk->sk_rmem_alloc) >= sk->sk_rcvbuf)
2134 /* NLMSG_GOODSIZE is small to avoid high order allocations being
2135 * required, but it makes sense to _attempt_ a 16K bytes allocation
2136 * to reduce number of system calls on dump operations, if user
2137 * ever provided a big enough buffer.
2140 alloc_min_size = max_t(int, cb->min_dump_alloc, NLMSG_GOODSIZE);
2142 if (alloc_min_size < nlk->max_recvmsg_len) {
2143 alloc_size = nlk->max_recvmsg_len;
2144 skb = alloc_skb(alloc_size,
2145 (GFP_KERNEL & ~__GFP_DIRECT_RECLAIM) |
2146 __GFP_NOWARN | __GFP_NORETRY);
2149 alloc_size = alloc_min_size;
2150 skb = alloc_skb(alloc_size, GFP_KERNEL);
2155 /* Trim skb to allocated size. User is expected to provide buffer as
2156 * large as max(min_dump_alloc, 16KiB (mac_recvmsg_len capped at
2157 * netlink_recvmsg())). dump will pack as many smaller messages as
2158 * could fit within the allocated skb. skb is typically allocated
2159 * with larger space than required (could be as much as near 2x the
2160 * requested size with align to next power of 2 approach). Allowing
2161 * dump to use the excess space makes it difficult for a user to have a
2162 * reasonable static buffer based on the expected largest dump of a
2163 * single netdev. The outcome is MSG_TRUNC error.
2165 skb_reserve(skb, skb_tailroom(skb) - alloc_size);
2166 netlink_skb_set_owner_r(skb, sk);
2168 len = cb->dump(skb, cb);
2171 mutex_unlock(nlk->cb_mutex);
2173 if (sk_filter(sk, skb))
2176 __netlink_sendskb(sk, skb);
2180 nlh = nlmsg_put_answer(skb, cb, NLMSG_DONE, sizeof(len), NLM_F_MULTI);
2184 nl_dump_check_consistent(cb, nlh);
2186 memcpy(nlmsg_data(nlh), &len, sizeof(len));
2188 if (sk_filter(sk, skb))
2191 __netlink_sendskb(sk, skb);
2196 nlk->cb_running = false;
2197 module = cb->module;
2199 mutex_unlock(nlk->cb_mutex);
2205 mutex_unlock(nlk->cb_mutex);
2210 int __netlink_dump_start(struct sock *ssk, struct sk_buff *skb,
2211 const struct nlmsghdr *nlh,
2212 struct netlink_dump_control *control)
2214 struct netlink_callback *cb;
2216 struct netlink_sock *nlk;
2219 atomic_inc(&skb->users);
2221 sk = netlink_lookup(sock_net(ssk), ssk->sk_protocol, NETLINK_CB(skb).portid);
2223 ret = -ECONNREFUSED;
2228 mutex_lock(nlk->cb_mutex);
2229 /* A dump is in progress... */
2230 if (nlk->cb_running) {
2234 /* add reference of module which cb->dump belongs to */
2235 if (!try_module_get(control->module)) {
2236 ret = -EPROTONOSUPPORT;
2241 memset(cb, 0, sizeof(*cb));
2242 cb->start = control->start;
2243 cb->dump = control->dump;
2244 cb->done = control->done;
2246 cb->data = control->data;
2247 cb->module = control->module;
2248 cb->min_dump_alloc = control->min_dump_alloc;
2251 nlk->cb_running = true;
2253 mutex_unlock(nlk->cb_mutex);
2258 ret = netlink_dump(sk);
2264 /* We successfully started a dump, by returning -EINTR we
2265 * signal not to send ACK even if it was requested.
2271 mutex_unlock(nlk->cb_mutex);
2276 EXPORT_SYMBOL(__netlink_dump_start);
2278 void netlink_ack(struct sk_buff *in_skb, struct nlmsghdr *nlh, int err)
2280 struct sk_buff *skb;
2281 struct nlmsghdr *rep;
2282 struct nlmsgerr *errmsg;
2283 size_t payload = sizeof(*errmsg);
2284 struct netlink_sock *nlk = nlk_sk(NETLINK_CB(in_skb).sk);
2286 /* Error messages get the original request appened, unless the user
2287 * requests to cap the error message.
2289 if (!(nlk->flags & NETLINK_F_CAP_ACK) && err)
2290 payload += nlmsg_len(nlh);
2292 skb = nlmsg_new(payload, GFP_KERNEL);
2296 sk = netlink_lookup(sock_net(in_skb->sk),
2297 in_skb->sk->sk_protocol,
2298 NETLINK_CB(in_skb).portid);
2300 sk->sk_err = ENOBUFS;
2301 sk->sk_error_report(sk);
2307 rep = __nlmsg_put(skb, NETLINK_CB(in_skb).portid, nlh->nlmsg_seq,
2308 NLMSG_ERROR, payload, 0);
2309 errmsg = nlmsg_data(rep);
2310 errmsg->error = err;
2311 memcpy(&errmsg->msg, nlh, payload > sizeof(*errmsg) ? nlh->nlmsg_len : sizeof(*nlh));
2312 netlink_unicast(in_skb->sk, skb, NETLINK_CB(in_skb).portid, MSG_DONTWAIT);
2314 EXPORT_SYMBOL(netlink_ack);
2316 int netlink_rcv_skb(struct sk_buff *skb, int (*cb)(struct sk_buff *,
2319 struct nlmsghdr *nlh;
2322 while (skb->len >= nlmsg_total_size(0)) {
2325 nlh = nlmsg_hdr(skb);
2328 if (nlh->nlmsg_len < NLMSG_HDRLEN || skb->len < nlh->nlmsg_len)
2331 /* Only requests are handled by the kernel */
2332 if (!(nlh->nlmsg_flags & NLM_F_REQUEST))
2335 /* Skip control messages */
2336 if (nlh->nlmsg_type < NLMSG_MIN_TYPE)
2344 if (nlh->nlmsg_flags & NLM_F_ACK || err)
2345 netlink_ack(skb, nlh, err);
2348 msglen = NLMSG_ALIGN(nlh->nlmsg_len);
2349 if (msglen > skb->len)
2351 skb_pull(skb, msglen);
2356 EXPORT_SYMBOL(netlink_rcv_skb);
2359 * nlmsg_notify - send a notification netlink message
2360 * @sk: netlink socket to use
2361 * @skb: notification message
2362 * @portid: destination netlink portid for reports or 0
2363 * @group: destination multicast group or 0
2364 * @report: 1 to report back, 0 to disable
2365 * @flags: allocation flags
2367 int nlmsg_notify(struct sock *sk, struct sk_buff *skb, u32 portid,
2368 unsigned int group, int report, gfp_t flags)
2373 int exclude_portid = 0;
2376 atomic_inc(&skb->users);
2377 exclude_portid = portid;
2380 /* errors reported via destination sk->sk_err, but propagate
2381 * delivery errors if NETLINK_BROADCAST_ERROR flag is set */
2382 err = nlmsg_multicast(sk, skb, exclude_portid, group, flags);
2388 err2 = nlmsg_unicast(sk, skb, portid);
2389 if (!err || err == -ESRCH)
2395 EXPORT_SYMBOL(nlmsg_notify);
2397 #ifdef CONFIG_PROC_FS
2398 struct nl_seq_iter {
2399 struct seq_net_private p;
2400 struct rhashtable_iter hti;
2404 static int netlink_walk_start(struct nl_seq_iter *iter)
2408 err = rhashtable_walk_init(&nl_table[iter->link].hash, &iter->hti,
2411 iter->link = MAX_LINKS;
2415 err = rhashtable_walk_start(&iter->hti);
2416 return err == -EAGAIN ? 0 : err;
2419 static void netlink_walk_stop(struct nl_seq_iter *iter)
2421 rhashtable_walk_stop(&iter->hti);
2422 rhashtable_walk_exit(&iter->hti);
2425 static void *__netlink_seq_next(struct seq_file *seq)
2427 struct nl_seq_iter *iter = seq->private;
2428 struct netlink_sock *nlk;
2434 nlk = rhashtable_walk_next(&iter->hti);
2437 if (PTR_ERR(nlk) == -EAGAIN)
2446 netlink_walk_stop(iter);
2447 if (++iter->link >= MAX_LINKS)
2450 err = netlink_walk_start(iter);
2452 return ERR_PTR(err);
2454 } while (sock_net(&nlk->sk) != seq_file_net(seq));
2459 static void *netlink_seq_start(struct seq_file *seq, loff_t *posp)
2461 struct nl_seq_iter *iter = seq->private;
2462 void *obj = SEQ_START_TOKEN;
2468 err = netlink_walk_start(iter);
2470 return ERR_PTR(err);
2472 for (pos = *posp; pos && obj && !IS_ERR(obj); pos--)
2473 obj = __netlink_seq_next(seq);
2478 static void *netlink_seq_next(struct seq_file *seq, void *v, loff_t *pos)
2481 return __netlink_seq_next(seq);
2484 static void netlink_seq_stop(struct seq_file *seq, void *v)
2486 struct nl_seq_iter *iter = seq->private;
2488 if (iter->link >= MAX_LINKS)
2491 netlink_walk_stop(iter);
2495 static int netlink_seq_show(struct seq_file *seq, void *v)
2497 if (v == SEQ_START_TOKEN) {
2499 "sk Eth Pid Groups "
2500 "Rmem Wmem Dump Locks Drops Inode\n");
2503 struct netlink_sock *nlk = nlk_sk(s);
2505 seq_printf(seq, "%pK %-3d %-6u %08x %-8d %-8d %d %-8d %-8d %-8lu\n",
2509 nlk->groups ? (u32)nlk->groups[0] : 0,
2510 sk_rmem_alloc_get(s),
2511 sk_wmem_alloc_get(s),
2513 atomic_read(&s->sk_refcnt),
2514 atomic_read(&s->sk_drops),
2522 static const struct seq_operations netlink_seq_ops = {
2523 .start = netlink_seq_start,
2524 .next = netlink_seq_next,
2525 .stop = netlink_seq_stop,
2526 .show = netlink_seq_show,
2530 static int netlink_seq_open(struct inode *inode, struct file *file)
2532 return seq_open_net(inode, file, &netlink_seq_ops,
2533 sizeof(struct nl_seq_iter));
2536 static const struct file_operations netlink_seq_fops = {
2537 .owner = THIS_MODULE,
2538 .open = netlink_seq_open,
2540 .llseek = seq_lseek,
2541 .release = seq_release_net,
2546 int netlink_register_notifier(struct notifier_block *nb)
2548 return blocking_notifier_chain_register(&netlink_chain, nb);
2550 EXPORT_SYMBOL(netlink_register_notifier);
2552 int netlink_unregister_notifier(struct notifier_block *nb)
2554 return blocking_notifier_chain_unregister(&netlink_chain, nb);
2556 EXPORT_SYMBOL(netlink_unregister_notifier);
2558 static const struct proto_ops netlink_ops = {
2559 .family = PF_NETLINK,
2560 .owner = THIS_MODULE,
2561 .release = netlink_release,
2562 .bind = netlink_bind,
2563 .connect = netlink_connect,
2564 .socketpair = sock_no_socketpair,
2565 .accept = sock_no_accept,
2566 .getname = netlink_getname,
2567 .poll = datagram_poll,
2568 .ioctl = netlink_ioctl,
2569 .listen = sock_no_listen,
2570 .shutdown = sock_no_shutdown,
2571 .setsockopt = netlink_setsockopt,
2572 .getsockopt = netlink_getsockopt,
2573 .sendmsg = netlink_sendmsg,
2574 .recvmsg = netlink_recvmsg,
2575 .mmap = sock_no_mmap,
2576 .sendpage = sock_no_sendpage,
2579 static const struct net_proto_family netlink_family_ops = {
2580 .family = PF_NETLINK,
2581 .create = netlink_create,
2582 .owner = THIS_MODULE, /* for consistency 8) */
2585 static int __net_init netlink_net_init(struct net *net)
2587 #ifdef CONFIG_PROC_FS
2588 if (!proc_create("netlink", 0, net->proc_net, &netlink_seq_fops))
2594 static void __net_exit netlink_net_exit(struct net *net)
2596 #ifdef CONFIG_PROC_FS
2597 remove_proc_entry("netlink", net->proc_net);
2601 static void __init netlink_add_usersock_entry(void)
2603 struct listeners *listeners;
2606 listeners = kzalloc(sizeof(*listeners) + NLGRPSZ(groups), GFP_KERNEL);
2608 panic("netlink_add_usersock_entry: Cannot allocate listeners\n");
2610 netlink_table_grab();
2612 nl_table[NETLINK_USERSOCK].groups = groups;
2613 rcu_assign_pointer(nl_table[NETLINK_USERSOCK].listeners, listeners);
2614 nl_table[NETLINK_USERSOCK].module = THIS_MODULE;
2615 nl_table[NETLINK_USERSOCK].registered = 1;
2616 nl_table[NETLINK_USERSOCK].flags = NL_CFG_F_NONROOT_SEND;
2618 netlink_table_ungrab();
2621 static struct pernet_operations __net_initdata netlink_net_ops = {
2622 .init = netlink_net_init,
2623 .exit = netlink_net_exit,
2626 static inline u32 netlink_hash(const void *data, u32 len, u32 seed)
2628 const struct netlink_sock *nlk = data;
2629 struct netlink_compare_arg arg;
2631 netlink_compare_arg_init(&arg, sock_net(&nlk->sk), nlk->portid);
2632 return jhash2((u32 *)&arg, netlink_compare_arg_len / sizeof(u32), seed);
2635 static const struct rhashtable_params netlink_rhashtable_params = {
2636 .head_offset = offsetof(struct netlink_sock, node),
2637 .key_len = netlink_compare_arg_len,
2638 .obj_hashfn = netlink_hash,
2639 .obj_cmpfn = netlink_compare,
2640 .automatic_shrinking = true,
2643 static int __init netlink_proto_init(void)
2646 int err = proto_register(&netlink_proto, 0);
2651 BUILD_BUG_ON(sizeof(struct netlink_skb_parms) > FIELD_SIZEOF(struct sk_buff, cb));
2653 nl_table = kcalloc(MAX_LINKS, sizeof(*nl_table), GFP_KERNEL);
2657 for (i = 0; i < MAX_LINKS; i++) {
2658 if (rhashtable_init(&nl_table[i].hash,
2659 &netlink_rhashtable_params) < 0) {
2661 rhashtable_destroy(&nl_table[i].hash);
2667 INIT_LIST_HEAD(&netlink_tap_all);
2669 netlink_add_usersock_entry();
2671 sock_register(&netlink_family_ops);
2672 register_pernet_subsys(&netlink_net_ops);
2673 /* The netlink device handler may be needed early. */
2678 panic("netlink_init: Cannot allocate nl_table\n");
2681 core_initcall(netlink_proto_init);