blob: 330ebd600f254d53e23a174273a504e395c0e7cf [file] [log] [blame]
Linus Torvalds1da177e2005-04-16 15:20:36 -07001/*
2 * NETLINK Kernel-user communication protocol.
3 *
Alan Cox113aa832008-10-13 19:01:08 -07004 * Authors: Alan Cox <alan@lxorguk.ukuu.org.uk>
Linus Torvalds1da177e2005-04-16 15:20:36 -07005 * Alexey Kuznetsov <kuznet@ms2.inr.ac.ru>
Patrick McHardycd1df522013-04-17 06:47:05 +00006 * Patrick McHardy <kaber@trash.net>
Linus Torvalds1da177e2005-04-16 15:20:36 -07007 *
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.
YOSHIFUJI Hideaki746fac42007-02-09 23:25:07 +090012 *
Linus Torvalds1da177e2005-04-16 15:20:36 -070013 * 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)
Harald Welte4fdb3bb2005-08-09 19:40:55 -070017 * 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
Linus Torvalds1da177e2005-04-16 15:20:36 -070023 */
24
Linus Torvalds1da177e2005-04-16 15:20:36 -070025#include <linux/module.h>
26
Randy Dunlap4fc268d2006-01-11 12:17:47 -080027#include <linux/capability.h>
Linus Torvalds1da177e2005-04-16 15:20:36 -070028#include <linux/kernel.h>
29#include <linux/init.h>
Linus Torvalds1da177e2005-04-16 15:20:36 -070030#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>
36#include <linux/un.h>
37#include <linux/fcntl.h>
38#include <linux/termios.h>
39#include <linux/sockios.h>
40#include <linux/net.h>
41#include <linux/fs.h>
42#include <linux/slab.h>
43#include <asm/uaccess.h>
44#include <linux/skbuff.h>
45#include <linux/netdevice.h>
46#include <linux/rtnetlink.h>
47#include <linux/proc_fs.h>
48#include <linux/seq_file.h>
Linus Torvalds1da177e2005-04-16 15:20:36 -070049#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>
55#include <linux/mm.h>
56#include <linux/types.h>
Andrew Morton54e0f522005-04-30 07:07:04 +010057#include <linux/audit.h>
Patrick McHardyaf65bdf2007-04-20 14:14:21 -070058#include <linux/mutex.h>
Patrick McHardyccdfcc32013-04-17 06:47:01 +000059#include <linux/vmalloc.h>
Daniel Borkmannbcbde0d2013-06-21 19:38:07 +020060#include <linux/if_arp.h>
Thomas Grafe3416942014-08-02 11:47:45 +020061#include <linux/rhashtable.h>
Patrick McHardy9652e932013-04-17 06:47:02 +000062#include <asm/cacheflush.h>
Thomas Grafe3416942014-08-02 11:47:45 +020063#include <linux/hash.h>
Johannes Bergee1c24422015-01-16 11:37:14 +010064#include <linux/genetlink.h>
Andrew Morton54e0f522005-04-30 07:07:04 +010065
Eric W. Biederman457c4cb2007-09-12 12:01:34 +020066#include <net/net_namespace.h>
Linus Torvalds1da177e2005-04-16 15:20:36 -070067#include <net/sock.h>
68#include <net/scm.h>
Thomas Graf82ace472005-11-10 02:25:53 +010069#include <net/netlink.h>
Linus Torvalds1da177e2005-04-16 15:20:36 -070070
Andrey Vagin0f29c762013-03-21 20:33:47 +040071#include "af_netlink.h"
Linus Torvalds1da177e2005-04-16 15:20:36 -070072
Eric Dumazet5c398dc2010-10-24 04:27:10 +000073struct listeners {
74 struct rcu_head rcu;
75 unsigned long masks[0];
Johannes Berg6c04bb12009-07-10 09:51:32 +000076};
77
Patrick McHardycd967e02013-04-17 06:46:56 +000078/* state bits */
Nicolas Dichtelcc3a5722015-05-07 11:02:52 +020079#define NETLINK_S_CONGESTED 0x0
Patrick McHardycd967e02013-04-17 06:46:56 +000080
81/* flags */
Nicolas Dichtelcc3a5722015-05-07 11:02:52 +020082#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
Nicolas Dichtel59324cf2015-05-07 11:02:53 +020086#define NETLINK_F_LISTEN_ALL_NSID 0x10
Christophe Ricard0a6a3a22015-08-28 07:07:48 +020087#define NETLINK_F_CAP_ACK 0x20
Patrick McHardy77247bb2005-08-14 19:27:13 -070088
David S. Miller035c4c12011-12-23 17:33:03 -050089static inline int netlink_is_kernel(struct sock *sk)
Denis V. Lunevaed81562007-10-10 21:14:32 -070090{
Nicolas Dichtelcc3a5722015-05-07 11:02:52 +020091 return nlk_sk(sk)->flags & NETLINK_F_KERNEL_SOCKET;
Denis V. Lunevaed81562007-10-10 21:14:32 -070092}
93
Eric Dumazet91dd93f2015-05-12 17:24:50 -070094struct netlink_table *nl_table __read_mostly;
Andrey Vagin0f29c762013-03-21 20:33:47 +040095EXPORT_SYMBOL_GPL(nl_table);
Linus Torvalds1da177e2005-04-16 15:20:36 -070096
97static DECLARE_WAIT_QUEUE_HEAD(nl_table_wait);
98
99static int netlink_dump(struct sock *sk);
Patrick McHardy9652e932013-04-17 06:47:02 +0000100static void netlink_skb_destructor(struct sk_buff *skb);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700101
Thomas Graf78fd1d02014-10-21 22:05:38 +0200102/* nl_table locking explained:
Thomas Graf21e49022015-01-02 23:00:22 +0100103 * Lookup and traversal are protected with an RCU read-side lock. Insertion
Ying Xuec5adde92015-01-12 14:52:23 +0800104 * and removal are protected with per bucket lock while using RCU list
Thomas Graf21e49022015-01-02 23:00:22 +0100105 * modification primitives and may run in parallel to RCU protected lookups.
106 * Destruction of the Netlink socket may only occur *after* nl_table_lock has
107 * been acquired * either during or after the socket has been removed from
108 * the list and after an RCU grace period.
Thomas Graf78fd1d02014-10-21 22:05:38 +0200109 */
Andrey Vagin0f29c762013-03-21 20:33:47 +0400110DEFINE_RWLOCK(nl_table_lock);
111EXPORT_SYMBOL_GPL(nl_table_lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700112static atomic_t nl_table_users = ATOMIC_INIT(0);
113
Eric Dumazet6d772ac2012-10-18 03:21:55 +0000114#define nl_deref_protected(X) rcu_dereference_protected(X, lockdep_is_held(&nl_table_lock));
115
Alan Sterne041c682006-03-27 01:16:30 -0800116static ATOMIC_NOTIFIER_HEAD(netlink_chain);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700117
Daniel Borkmannbcbde0d2013-06-21 19:38:07 +0200118static DEFINE_SPINLOCK(netlink_tap_lock);
119static struct list_head netlink_tap_all __read_mostly;
120
Herbert Xuc428ecd2015-03-20 21:57:01 +1100121static const struct rhashtable_params netlink_rhashtable_params;
122
stephen hemmingerb57ef81f2011-12-22 08:52:02 +0000123static inline u32 netlink_group_mask(u32 group)
Patrick McHardyd629b832005-08-14 19:27:50 -0700124{
125 return group ? 1 << (group - 1) : 0;
126}
127
Daniel Borkmann1853c942015-09-10 20:05:46 +0200128static struct sk_buff *netlink_to_full_skb(const struct sk_buff *skb,
129 gfp_t gfp_mask)
130{
131 unsigned int len = skb_end_offset(skb);
132 struct sk_buff *new;
133
134 new = alloc_skb(len, gfp_mask);
135 if (new == NULL)
136 return NULL;
137
138 NETLINK_CB(new).portid = NETLINK_CB(skb).portid;
139 NETLINK_CB(new).dst_group = NETLINK_CB(skb).dst_group;
140 NETLINK_CB(new).creds = NETLINK_CB(skb).creds;
141
142 memcpy(skb_put(new, len), skb->data, len);
143 return new;
144}
145
Daniel Borkmannbcbde0d2013-06-21 19:38:07 +0200146int netlink_add_tap(struct netlink_tap *nt)
147{
148 if (unlikely(nt->dev->type != ARPHRD_NETLINK))
149 return -EINVAL;
150
151 spin_lock(&netlink_tap_lock);
152 list_add_rcu(&nt->list, &netlink_tap_all);
153 spin_unlock(&netlink_tap_lock);
154
Markus Elfringfcd4d352014-11-18 21:03:13 +0100155 __module_get(nt->module);
Daniel Borkmannbcbde0d2013-06-21 19:38:07 +0200156
157 return 0;
158}
159EXPORT_SYMBOL_GPL(netlink_add_tap);
160
stephen hemminger2173f8d2013-12-30 10:49:22 -0800161static int __netlink_remove_tap(struct netlink_tap *nt)
Daniel Borkmannbcbde0d2013-06-21 19:38:07 +0200162{
163 bool found = false;
164 struct netlink_tap *tmp;
165
166 spin_lock(&netlink_tap_lock);
167
168 list_for_each_entry(tmp, &netlink_tap_all, list) {
169 if (nt == tmp) {
170 list_del_rcu(&nt->list);
171 found = true;
172 goto out;
173 }
174 }
175
176 pr_warn("__netlink_remove_tap: %p not found\n", nt);
177out:
178 spin_unlock(&netlink_tap_lock);
179
Markus Elfring92b80eb2015-07-02 18:38:12 +0200180 if (found)
Daniel Borkmannbcbde0d2013-06-21 19:38:07 +0200181 module_put(nt->module);
182
183 return found ? 0 : -ENODEV;
184}
Daniel Borkmannbcbde0d2013-06-21 19:38:07 +0200185
186int netlink_remove_tap(struct netlink_tap *nt)
187{
188 int ret;
189
190 ret = __netlink_remove_tap(nt);
191 synchronize_net();
192
193 return ret;
194}
195EXPORT_SYMBOL_GPL(netlink_remove_tap);
196
Daniel Borkmann5ffd5cd2013-09-05 17:48:47 +0200197static bool netlink_filter_tap(const struct sk_buff *skb)
198{
199 struct sock *sk = skb->sk;
Daniel Borkmann5ffd5cd2013-09-05 17:48:47 +0200200
201 /* We take the more conservative approach and
202 * whitelist socket protocols that may pass.
203 */
204 switch (sk->sk_protocol) {
205 case NETLINK_ROUTE:
206 case NETLINK_USERSOCK:
207 case NETLINK_SOCK_DIAG:
208 case NETLINK_NFLOG:
209 case NETLINK_XFRM:
210 case NETLINK_FIB_LOOKUP:
211 case NETLINK_NETFILTER:
212 case NETLINK_GENERIC:
Varka Bhadram498044b2014-07-16 10:59:47 +0530213 return true;
Daniel Borkmann5ffd5cd2013-09-05 17:48:47 +0200214 }
215
Varka Bhadram498044b2014-07-16 10:59:47 +0530216 return false;
Daniel Borkmann5ffd5cd2013-09-05 17:48:47 +0200217}
218
Daniel Borkmannbcbde0d2013-06-21 19:38:07 +0200219static int __netlink_deliver_tap_skb(struct sk_buff *skb,
220 struct net_device *dev)
221{
222 struct sk_buff *nskb;
Daniel Borkmann5ffd5cd2013-09-05 17:48:47 +0200223 struct sock *sk = skb->sk;
Daniel Borkmannbcbde0d2013-06-21 19:38:07 +0200224 int ret = -ENOMEM;
225
226 dev_hold(dev);
Daniel Borkmann1853c942015-09-10 20:05:46 +0200227
Florian Westphald1b4c682016-02-18 15:03:24 +0100228 if (is_vmalloc_addr(skb->head))
Daniel Borkmann1853c942015-09-10 20:05:46 +0200229 nskb = netlink_to_full_skb(skb, GFP_ATOMIC);
230 else
231 nskb = skb_clone(skb, GFP_ATOMIC);
Daniel Borkmannbcbde0d2013-06-21 19:38:07 +0200232 if (nskb) {
233 nskb->dev = dev;
Daniel Borkmann5ffd5cd2013-09-05 17:48:47 +0200234 nskb->protocol = htons((u16) sk->sk_protocol);
Daniel Borkmann604d13c2013-12-23 14:35:56 +0100235 nskb->pkt_type = netlink_is_kernel(sk) ?
236 PACKET_KERNEL : PACKET_USER;
Daniel Borkmann4e48ed82014-08-07 22:22:47 +0200237 skb_reset_network_header(nskb);
Daniel Borkmannbcbde0d2013-06-21 19:38:07 +0200238 ret = dev_queue_xmit(nskb);
239 if (unlikely(ret > 0))
240 ret = net_xmit_errno(ret);
241 }
242
243 dev_put(dev);
244 return ret;
245}
246
247static void __netlink_deliver_tap(struct sk_buff *skb)
248{
249 int ret;
250 struct netlink_tap *tmp;
251
Daniel Borkmann5ffd5cd2013-09-05 17:48:47 +0200252 if (!netlink_filter_tap(skb))
253 return;
254
Daniel Borkmannbcbde0d2013-06-21 19:38:07 +0200255 list_for_each_entry_rcu(tmp, &netlink_tap_all, list) {
256 ret = __netlink_deliver_tap_skb(skb, tmp->dev);
257 if (unlikely(ret))
258 break;
259 }
260}
261
262static void netlink_deliver_tap(struct sk_buff *skb)
263{
264 rcu_read_lock();
265
266 if (unlikely(!list_empty(&netlink_tap_all)))
267 __netlink_deliver_tap(skb);
268
269 rcu_read_unlock();
270}
271
Daniel Borkmann73bfd372013-12-23 14:35:55 +0100272static void netlink_deliver_tap_kernel(struct sock *dst, struct sock *src,
273 struct sk_buff *skb)
274{
275 if (!(netlink_is_kernel(dst) && netlink_is_kernel(src)))
276 netlink_deliver_tap(skb);
277}
278
Patrick McHardycd1df522013-04-17 06:47:05 +0000279static void netlink_overrun(struct sock *sk)
280{
281 struct netlink_sock *nlk = nlk_sk(sk);
282
Nicolas Dichtelcc3a5722015-05-07 11:02:52 +0200283 if (!(nlk->flags & NETLINK_F_RECV_NO_ENOBUFS)) {
284 if (!test_and_set_bit(NETLINK_S_CONGESTED,
285 &nlk_sk(sk)->state)) {
Patrick McHardycd1df522013-04-17 06:47:05 +0000286 sk->sk_err = ENOBUFS;
287 sk->sk_error_report(sk);
288 }
289 }
290 atomic_inc(&sk->sk_drops);
291}
292
293static void netlink_rcv_wake(struct sock *sk)
294{
295 struct netlink_sock *nlk = nlk_sk(sk);
296
297 if (skb_queue_empty(&sk->sk_receive_queue))
Nicolas Dichtelcc3a5722015-05-07 11:02:52 +0200298 clear_bit(NETLINK_S_CONGESTED, &nlk->state);
299 if (!test_bit(NETLINK_S_CONGESTED, &nlk->state))
Patrick McHardycd1df522013-04-17 06:47:05 +0000300 wake_up_interruptible(&nlk->wait);
301}
302
Patrick McHardycf0a0182013-04-17 06:47:00 +0000303static void netlink_skb_destructor(struct sk_buff *skb)
304{
Pablo Neira Ayusoc05cdb12013-06-03 09:46:28 +0000305 if (is_vmalloc_addr(skb->head)) {
Pablo Neira3a365152013-06-28 03:04:23 +0200306 if (!skb->cloned ||
307 !atomic_dec_return(&(skb_shinfo(skb)->dataref)))
308 vfree(skb->head);
309
Pablo Neira Ayusoc05cdb12013-06-03 09:46:28 +0000310 skb->head = NULL;
311 }
Patrick McHardy9652e932013-04-17 06:47:02 +0000312 if (skb->sk != NULL)
313 sock_rfree(skb);
Patrick McHardycf0a0182013-04-17 06:47:00 +0000314}
315
316static void netlink_skb_set_owner_r(struct sk_buff *skb, struct sock *sk)
317{
318 WARN_ON(skb->sk != NULL);
319 skb->sk = sk;
320 skb->destructor = netlink_skb_destructor;
321 atomic_add(skb->truesize, &sk->sk_rmem_alloc);
322 sk_mem_charge(sk, skb->truesize);
323}
324
Linus Torvalds1da177e2005-04-16 15:20:36 -0700325static void netlink_sock_destruct(struct sock *sk)
326{
Herbert Xu3f660d62007-05-03 03:17:14 -0700327 struct netlink_sock *nlk = nlk_sk(sk);
328
Pravin B Shelar16b304f2013-08-15 15:31:06 -0700329 if (nlk->cb_running) {
330 if (nlk->cb.done)
331 nlk->cb.done(&nlk->cb);
Gao feng6dc878a2012-10-04 20:15:48 +0000332
Pravin B Shelar16b304f2013-08-15 15:31:06 -0700333 module_put(nlk->cb.module);
334 kfree_skb(nlk->cb.skb);
Herbert Xu3f660d62007-05-03 03:17:14 -0700335 }
336
Linus Torvalds1da177e2005-04-16 15:20:36 -0700337 skb_queue_purge(&sk->sk_receive_queue);
338
339 if (!sock_flag(sk, SOCK_DEAD)) {
Patrick McHardy6ac552f2007-12-04 00:19:38 -0800340 printk(KERN_ERR "Freeing alive netlink socket %p\n", sk);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700341 return;
342 }
Ilpo Järvinen547b7922008-07-25 21:43:18 -0700343
344 WARN_ON(atomic_read(&sk->sk_rmem_alloc));
345 WARN_ON(atomic_read(&sk->sk_wmem_alloc));
346 WARN_ON(nlk_sk(sk)->groups);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700347}
348
Patrick McHardy6ac552f2007-12-04 00:19:38 -0800349/* This lock without WQ_FLAG_EXCLUSIVE is good on UP and it is _very_ bad on
350 * SMP. Look, when several writers sleep and reader wakes them up, all but one
Linus Torvalds1da177e2005-04-16 15:20:36 -0700351 * immediately hit write lock and grab all the cpus. Exclusive sleep solves
352 * this, _but_ remember, it adds useless work on UP machines.
353 */
354
Johannes Bergd136f1b2009-09-12 03:03:15 +0000355void netlink_table_grab(void)
Eric Dumazet9a429c42008-01-01 21:58:02 -0800356 __acquires(nl_table_lock)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700357{
Johannes Bergd136f1b2009-09-12 03:03:15 +0000358 might_sleep();
359
Arjan van de Ven6abd2192006-07-03 00:24:07 -0700360 write_lock_irq(&nl_table_lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700361
362 if (atomic_read(&nl_table_users)) {
363 DECLARE_WAITQUEUE(wait, current);
364
365 add_wait_queue_exclusive(&nl_table_wait, &wait);
Patrick McHardy6ac552f2007-12-04 00:19:38 -0800366 for (;;) {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700367 set_current_state(TASK_UNINTERRUPTIBLE);
368 if (atomic_read(&nl_table_users) == 0)
369 break;
Arjan van de Ven6abd2192006-07-03 00:24:07 -0700370 write_unlock_irq(&nl_table_lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700371 schedule();
Arjan van de Ven6abd2192006-07-03 00:24:07 -0700372 write_lock_irq(&nl_table_lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700373 }
374
375 __set_current_state(TASK_RUNNING);
376 remove_wait_queue(&nl_table_wait, &wait);
377 }
378}
379
Johannes Bergd136f1b2009-09-12 03:03:15 +0000380void netlink_table_ungrab(void)
Eric Dumazet9a429c42008-01-01 21:58:02 -0800381 __releases(nl_table_lock)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700382{
Arjan van de Ven6abd2192006-07-03 00:24:07 -0700383 write_unlock_irq(&nl_table_lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700384 wake_up(&nl_table_wait);
385}
386
Patrick McHardy6ac552f2007-12-04 00:19:38 -0800387static inline void
Linus Torvalds1da177e2005-04-16 15:20:36 -0700388netlink_lock_table(void)
389{
390 /* read_lock() synchronizes us to netlink_table_grab */
391
392 read_lock(&nl_table_lock);
393 atomic_inc(&nl_table_users);
394 read_unlock(&nl_table_lock);
395}
396
Patrick McHardy6ac552f2007-12-04 00:19:38 -0800397static inline void
Linus Torvalds1da177e2005-04-16 15:20:36 -0700398netlink_unlock_table(void)
399{
400 if (atomic_dec_and_test(&nl_table_users))
401 wake_up(&nl_table_wait);
402}
403
Thomas Grafe3416942014-08-02 11:47:45 +0200404struct netlink_compare_arg
Gao fengda12c902013-06-06 14:49:11 +0800405{
Herbert Xuc428ecd2015-03-20 21:57:01 +1100406 possible_net_t pnet;
Thomas Grafe3416942014-08-02 11:47:45 +0200407 u32 portid;
408};
409
Herbert Xu8f2ddaa2015-03-21 14:14:03 +1100410/* Doing sizeof directly may yield 4 extra bytes on 64-bit. */
411#define netlink_compare_arg_len \
412 (offsetof(struct netlink_compare_arg, portid) + sizeof(u32))
Thomas Grafe3416942014-08-02 11:47:45 +0200413
Herbert Xuc428ecd2015-03-20 21:57:01 +1100414static inline int netlink_compare(struct rhashtable_compare_arg *arg,
415 const void *ptr)
416{
417 const struct netlink_compare_arg *x = arg->key;
418 const struct netlink_sock *nlk = ptr;
419
Herbert Xuda314c92015-09-22 11:38:56 +0800420 return nlk->portid != x->portid ||
Herbert Xuc428ecd2015-03-20 21:57:01 +1100421 !net_eq(sock_net(&nlk->sk), read_pnet(&x->pnet));
422}
423
424static void netlink_compare_arg_init(struct netlink_compare_arg *arg,
425 struct net *net, u32 portid)
426{
427 memset(arg, 0, sizeof(*arg));
428 write_pnet(&arg->pnet, net);
429 arg->portid = portid;
Thomas Grafe3416942014-08-02 11:47:45 +0200430}
431
432static struct sock *__netlink_lookup(struct netlink_table *table, u32 portid,
433 struct net *net)
434{
Herbert Xuc428ecd2015-03-20 21:57:01 +1100435 struct netlink_compare_arg arg;
Thomas Grafe3416942014-08-02 11:47:45 +0200436
Herbert Xuc428ecd2015-03-20 21:57:01 +1100437 netlink_compare_arg_init(&arg, net, portid);
438 return rhashtable_lookup_fast(&table->hash, &arg,
439 netlink_rhashtable_params);
Gao fengda12c902013-06-06 14:49:11 +0800440}
441
Herbert Xuc428ecd2015-03-20 21:57:01 +1100442static int __netlink_insert(struct netlink_table *table, struct sock *sk)
Ying Xuec5adde92015-01-12 14:52:23 +0800443{
Herbert Xuc428ecd2015-03-20 21:57:01 +1100444 struct netlink_compare_arg arg;
Ying Xuec5adde92015-01-12 14:52:23 +0800445
Herbert Xuda314c92015-09-22 11:38:56 +0800446 netlink_compare_arg_init(&arg, sock_net(sk), nlk_sk(sk)->portid);
Herbert Xuc428ecd2015-03-20 21:57:01 +1100447 return rhashtable_lookup_insert_key(&table->hash, &arg,
448 &nlk_sk(sk)->node,
449 netlink_rhashtable_params);
Ying Xuec5adde92015-01-12 14:52:23 +0800450}
451
Eric W. Biederman15e47302012-09-07 20:12:54 +0000452static struct sock *netlink_lookup(struct net *net, int protocol, u32 portid)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700453{
Gao fengda12c902013-06-06 14:49:11 +0800454 struct netlink_table *table = &nl_table[protocol];
Linus Torvalds1da177e2005-04-16 15:20:36 -0700455 struct sock *sk;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700456
Thomas Grafe3416942014-08-02 11:47:45 +0200457 rcu_read_lock();
458 sk = __netlink_lookup(table, portid, net);
459 if (sk)
460 sock_hold(sk);
461 rcu_read_unlock();
462
Linus Torvalds1da177e2005-04-16 15:20:36 -0700463 return sk;
464}
465
Eric Dumazet90ddc4f2005-12-22 12:49:22 -0800466static const struct proto_ops netlink_ops;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700467
Patrick McHardy4277a082006-03-20 18:52:01 -0800468static void
469netlink_update_listeners(struct sock *sk)
470{
471 struct netlink_table *tbl = &nl_table[sk->sk_protocol];
Patrick McHardy4277a082006-03-20 18:52:01 -0800472 unsigned long mask;
473 unsigned int i;
Eric Dumazet6d772ac2012-10-18 03:21:55 +0000474 struct listeners *listeners;
475
476 listeners = nl_deref_protected(tbl->listeners);
477 if (!listeners)
478 return;
Patrick McHardy4277a082006-03-20 18:52:01 -0800479
Johannes Bergb4ff4f02007-07-18 15:46:06 -0700480 for (i = 0; i < NLGRPLONGS(tbl->groups); i++) {
Patrick McHardy4277a082006-03-20 18:52:01 -0800481 mask = 0;
Sasha Levinb67bfe02013-02-27 17:06:00 -0800482 sk_for_each_bound(sk, &tbl->mc_list) {
Johannes Bergb4ff4f02007-07-18 15:46:06 -0700483 if (i < NLGRPLONGS(nlk_sk(sk)->ngroups))
484 mask |= nlk_sk(sk)->groups[i];
485 }
Eric Dumazet6d772ac2012-10-18 03:21:55 +0000486 listeners->masks[i] = mask;
Patrick McHardy4277a082006-03-20 18:52:01 -0800487 }
488 /* this function is only called with the netlink table "grabbed", which
489 * makes sure updates are visible before bind or setsockopt return. */
490}
491
Herbert Xu8ea65f42015-01-26 14:02:56 +1100492static int netlink_insert(struct sock *sk, u32 portid)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700493{
Gao fengda12c902013-06-06 14:49:11 +0800494 struct netlink_table *table = &nl_table[sk->sk_protocol];
Herbert Xu919d9db2015-01-16 17:23:48 +1100495 int err;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700496
Ying Xuec5adde92015-01-12 14:52:23 +0800497 lock_sock(sk);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700498
Herbert Xuda314c92015-09-22 11:38:56 +0800499 err = nlk_sk(sk)->portid == portid ? 0 : -EBUSY;
500 if (nlk_sk(sk)->bound)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700501 goto err;
502
503 err = -ENOMEM;
Thomas Graf97defe12015-01-02 23:00:20 +0100504 if (BITS_PER_LONG > 32 &&
505 unlikely(atomic_read(&table->hash.nelems) >= UINT_MAX))
Linus Torvalds1da177e2005-04-16 15:20:36 -0700506 goto err;
507
Herbert Xuda314c92015-09-22 11:38:56 +0800508 nlk_sk(sk)->portid = portid;
Thomas Grafe3416942014-08-02 11:47:45 +0200509 sock_hold(sk);
Herbert Xu919d9db2015-01-16 17:23:48 +1100510
Herbert Xuc428ecd2015-03-20 21:57:01 +1100511 err = __netlink_insert(table, sk);
512 if (err) {
Daniel Borkmann4e7c1332015-08-07 00:26:41 +0200513 /* In case the hashtable backend returns with -EBUSY
514 * from here, it must not escape to the caller.
515 */
516 if (unlikely(err == -EBUSY))
517 err = -EOVERFLOW;
Herbert Xuc428ecd2015-03-20 21:57:01 +1100518 if (err == -EEXIST)
519 err = -EADDRINUSE;
Ying Xuec5adde92015-01-12 14:52:23 +0800520 sock_put(sk);
Herbert Xu1f770c02015-09-18 19:16:50 +0800521 goto err;
Herbert Xu919d9db2015-01-16 17:23:48 +1100522 }
523
Herbert Xuda314c92015-09-22 11:38:56 +0800524 /* We need to ensure that the socket is hashed and visible. */
525 smp_wmb();
526 nlk_sk(sk)->bound = portid;
Herbert Xu1f770c02015-09-18 19:16:50 +0800527
Linus Torvalds1da177e2005-04-16 15:20:36 -0700528err:
Ying Xuec5adde92015-01-12 14:52:23 +0800529 release_sock(sk);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700530 return err;
531}
532
533static void netlink_remove(struct sock *sk)
534{
Thomas Grafe3416942014-08-02 11:47:45 +0200535 struct netlink_table *table;
536
Thomas Grafe3416942014-08-02 11:47:45 +0200537 table = &nl_table[sk->sk_protocol];
Herbert Xuc428ecd2015-03-20 21:57:01 +1100538 if (!rhashtable_remove_fast(&table->hash, &nlk_sk(sk)->node,
539 netlink_rhashtable_params)) {
Thomas Grafe3416942014-08-02 11:47:45 +0200540 WARN_ON(atomic_read(&sk->sk_refcnt) == 1);
541 __sock_put(sk);
542 }
Thomas Grafe3416942014-08-02 11:47:45 +0200543
Linus Torvalds1da177e2005-04-16 15:20:36 -0700544 netlink_table_grab();
Johannes Bergb10dcb32014-12-22 18:56:37 +0100545 if (nlk_sk(sk)->subscriptions) {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700546 __sk_del_bind_node(sk);
Johannes Bergb10dcb32014-12-22 18:56:37 +0100547 netlink_update_listeners(sk);
548 }
Johannes Bergee1c24422015-01-16 11:37:14 +0100549 if (sk->sk_protocol == NETLINK_GENERIC)
550 atomic_inc(&genl_sk_destructing_cnt);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700551 netlink_table_ungrab();
552}
553
554static struct proto netlink_proto = {
555 .name = "NETLINK",
556 .owner = THIS_MODULE,
557 .obj_size = sizeof(struct netlink_sock),
558};
559
Eric W. Biederman1b8d7ae2007-10-08 23:24:22 -0700560static int __netlink_create(struct net *net, struct socket *sock,
Eric W. Biederman11aa9c22015-05-08 21:09:13 -0500561 struct mutex *cb_mutex, int protocol,
562 int kern)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700563{
564 struct sock *sk;
565 struct netlink_sock *nlk;
Patrick McHardyab33a172005-08-14 19:31:36 -0700566
567 sock->ops = &netlink_ops;
568
Eric W. Biederman11aa9c22015-05-08 21:09:13 -0500569 sk = sk_alloc(net, PF_NETLINK, GFP_KERNEL, &netlink_proto, kern);
Patrick McHardyab33a172005-08-14 19:31:36 -0700570 if (!sk)
571 return -ENOMEM;
572
573 sock_init_data(sock, sk);
574
575 nlk = nlk_sk(sk);
Eric Dumazet658cb352012-04-22 21:30:21 +0000576 if (cb_mutex) {
Patrick McHardyffa4d722007-04-25 14:01:17 -0700577 nlk->cb_mutex = cb_mutex;
Eric Dumazet658cb352012-04-22 21:30:21 +0000578 } else {
Patrick McHardyffa4d722007-04-25 14:01:17 -0700579 nlk->cb_mutex = &nlk->cb_def_mutex;
580 mutex_init(nlk->cb_mutex);
581 }
Patrick McHardyab33a172005-08-14 19:31:36 -0700582 init_waitqueue_head(&nlk->wait);
583
584 sk->sk_destruct = netlink_sock_destruct;
585 sk->sk_protocol = protocol;
586 return 0;
587}
588
Eric Paris3f378b62009-11-05 22:18:14 -0800589static int netlink_create(struct net *net, struct socket *sock, int protocol,
590 int kern)
Patrick McHardyab33a172005-08-14 19:31:36 -0700591{
592 struct module *module = NULL;
Patrick McHardyaf65bdf2007-04-20 14:14:21 -0700593 struct mutex *cb_mutex;
Patrick McHardyf7fa9b12005-08-15 12:29:13 -0700594 struct netlink_sock *nlk;
Johannes Berg023e2cf2014-12-23 21:00:06 +0100595 int (*bind)(struct net *net, int group);
596 void (*unbind)(struct net *net, int group);
Patrick McHardyab33a172005-08-14 19:31:36 -0700597 int err = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700598
599 sock->state = SS_UNCONNECTED;
600
601 if (sock->type != SOCK_RAW && sock->type != SOCK_DGRAM)
602 return -ESOCKTNOSUPPORT;
603
Patrick McHardy6ac552f2007-12-04 00:19:38 -0800604 if (protocol < 0 || protocol >= MAX_LINKS)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700605 return -EPROTONOSUPPORT;
606
Patrick McHardy77247bb2005-08-14 19:27:13 -0700607 netlink_lock_table();
Johannes Berg95a5afc2008-10-16 15:24:51 -0700608#ifdef CONFIG_MODULES
Patrick McHardyab33a172005-08-14 19:31:36 -0700609 if (!nl_table[protocol].registered) {
Patrick McHardy77247bb2005-08-14 19:27:13 -0700610 netlink_unlock_table();
Harald Welte4fdb3bb2005-08-09 19:40:55 -0700611 request_module("net-pf-%d-proto-%d", PF_NETLINK, protocol);
Patrick McHardy77247bb2005-08-14 19:27:13 -0700612 netlink_lock_table();
Harald Welte4fdb3bb2005-08-09 19:40:55 -0700613 }
Patrick McHardyab33a172005-08-14 19:31:36 -0700614#endif
615 if (nl_table[protocol].registered &&
616 try_module_get(nl_table[protocol].module))
617 module = nl_table[protocol].module;
Alexey Dobriyan974c37e2010-01-30 10:05:05 +0000618 else
619 err = -EPROTONOSUPPORT;
Patrick McHardyaf65bdf2007-04-20 14:14:21 -0700620 cb_mutex = nl_table[protocol].cb_mutex;
Pablo Neira Ayuso03292742012-06-29 06:15:22 +0000621 bind = nl_table[protocol].bind;
Richard Guy Briggs4f520902014-04-22 21:31:54 -0400622 unbind = nl_table[protocol].unbind;
Patrick McHardy77247bb2005-08-14 19:27:13 -0700623 netlink_unlock_table();
Harald Welte4fdb3bb2005-08-09 19:40:55 -0700624
Alexey Dobriyan974c37e2010-01-30 10:05:05 +0000625 if (err < 0)
626 goto out;
627
Eric W. Biederman11aa9c22015-05-08 21:09:13 -0500628 err = __netlink_create(net, sock, cb_mutex, protocol, kern);
Patrick McHardy6ac552f2007-12-04 00:19:38 -0800629 if (err < 0)
Patrick McHardyab33a172005-08-14 19:31:36 -0700630 goto out_module;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700631
David S. Miller6f756a82008-11-23 17:34:03 -0800632 local_bh_disable();
Eric Dumazetc1fd3b92008-11-23 15:48:22 -0800633 sock_prot_inuse_add(net, &netlink_proto, 1);
David S. Miller6f756a82008-11-23 17:34:03 -0800634 local_bh_enable();
635
Patrick McHardyf7fa9b12005-08-15 12:29:13 -0700636 nlk = nlk_sk(sock->sk);
Patrick McHardyf7fa9b12005-08-15 12:29:13 -0700637 nlk->module = module;
Pablo Neira Ayuso03292742012-06-29 06:15:22 +0000638 nlk->netlink_bind = bind;
Richard Guy Briggs4f520902014-04-22 21:31:54 -0400639 nlk->netlink_unbind = unbind;
Patrick McHardyab33a172005-08-14 19:31:36 -0700640out:
641 return err;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700642
Patrick McHardyab33a172005-08-14 19:31:36 -0700643out_module:
644 module_put(module);
645 goto out;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700646}
647
Thomas Graf21e49022015-01-02 23:00:22 +0100648static void deferred_put_nlk_sk(struct rcu_head *head)
649{
650 struct netlink_sock *nlk = container_of(head, struct netlink_sock, rcu);
651
652 sock_put(&nlk->sk);
653}
654
Linus Torvalds1da177e2005-04-16 15:20:36 -0700655static int netlink_release(struct socket *sock)
656{
657 struct sock *sk = sock->sk;
658 struct netlink_sock *nlk;
659
660 if (!sk)
661 return 0;
662
663 netlink_remove(sk);
Denis Lunevac57b3a2007-04-18 17:05:58 -0700664 sock_orphan(sk);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700665 nlk = nlk_sk(sk);
666
Herbert Xu3f660d62007-05-03 03:17:14 -0700667 /*
668 * OK. Socket is unlinked, any packets that arrive now
669 * will be purged.
670 */
Linus Torvalds1da177e2005-04-16 15:20:36 -0700671
Johannes Bergee1c24422015-01-16 11:37:14 +0100672 /* must not acquire netlink_table_lock in any way again before unbind
673 * and notifying genetlink is done as otherwise it might deadlock
674 */
675 if (nlk->netlink_unbind) {
676 int i;
677
678 for (i = 0; i < nlk->ngroups; i++)
679 if (test_bit(i, nlk->groups))
680 nlk->netlink_unbind(sock_net(sk), i + 1);
681 }
682 if (sk->sk_protocol == NETLINK_GENERIC &&
683 atomic_dec_return(&genl_sk_destructing_cnt) == 0)
684 wake_up(&genl_sk_destructing_waitq);
685
Linus Torvalds1da177e2005-04-16 15:20:36 -0700686 sock->sk = NULL;
687 wake_up_interruptible_all(&nlk->wait);
688
689 skb_queue_purge(&sk->sk_write_queue);
690
Dmitry Ivanove2726022016-04-07 09:31:38 +0200691 if (nlk->portid && nlk->bound) {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700692 struct netlink_notify n = {
YOSHIFUJI Hideaki3b1e0a62008-03-26 02:26:21 +0900693 .net = sock_net(sk),
Linus Torvalds1da177e2005-04-16 15:20:36 -0700694 .protocol = sk->sk_protocol,
Eric W. Biederman15e47302012-09-07 20:12:54 +0000695 .portid = nlk->portid,
Linus Torvalds1da177e2005-04-16 15:20:36 -0700696 };
Alan Sterne041c682006-03-27 01:16:30 -0800697 atomic_notifier_call_chain(&netlink_chain,
698 NETLINK_URELEASE, &n);
YOSHIFUJI Hideaki746fac42007-02-09 23:25:07 +0900699 }
Harald Welte4fdb3bb2005-08-09 19:40:55 -0700700
Mariusz Kozlowski5e7c0012007-01-02 15:24:30 -0800701 module_put(nlk->module);
Harald Welte4fdb3bb2005-08-09 19:40:55 -0700702
Denis V. Lunevaed81562007-10-10 21:14:32 -0700703 if (netlink_is_kernel(sk)) {
Johannes Bergb10dcb32014-12-22 18:56:37 +0100704 netlink_table_grab();
Denis V. Lunev869e58f2008-01-18 23:53:31 -0800705 BUG_ON(nl_table[sk->sk_protocol].registered == 0);
706 if (--nl_table[sk->sk_protocol].registered == 0) {
Eric Dumazet6d772ac2012-10-18 03:21:55 +0000707 struct listeners *old;
708
709 old = nl_deref_protected(nl_table[sk->sk_protocol].listeners);
710 RCU_INIT_POINTER(nl_table[sk->sk_protocol].listeners, NULL);
711 kfree_rcu(old, rcu);
Denis V. Lunev869e58f2008-01-18 23:53:31 -0800712 nl_table[sk->sk_protocol].module = NULL;
Pablo Neira Ayuso9785e102012-09-08 02:53:53 +0000713 nl_table[sk->sk_protocol].bind = NULL;
Richard Guy Briggs4f520902014-04-22 21:31:54 -0400714 nl_table[sk->sk_protocol].unbind = NULL;
Pablo Neira Ayuso9785e102012-09-08 02:53:53 +0000715 nl_table[sk->sk_protocol].flags = 0;
Denis V. Lunev869e58f2008-01-18 23:53:31 -0800716 nl_table[sk->sk_protocol].registered = 0;
717 }
Johannes Bergb10dcb32014-12-22 18:56:37 +0100718 netlink_table_ungrab();
Eric Dumazet658cb352012-04-22 21:30:21 +0000719 }
Patrick McHardy77247bb2005-08-14 19:27:13 -0700720
Patrick McHardyf7fa9b12005-08-15 12:29:13 -0700721 kfree(nlk->groups);
722 nlk->groups = NULL;
723
Eric Dumazet37558102008-11-24 14:05:22 -0800724 local_bh_disable();
Eric Dumazetc1fd3b92008-11-23 15:48:22 -0800725 sock_prot_inuse_add(sock_net(sk), &netlink_proto, -1);
Eric Dumazet37558102008-11-24 14:05:22 -0800726 local_bh_enable();
Thomas Graf21e49022015-01-02 23:00:22 +0100727 call_rcu(&nlk->rcu, deferred_put_nlk_sk);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700728 return 0;
729}
730
731static int netlink_autobind(struct socket *sock)
732{
733 struct sock *sk = sock->sk;
YOSHIFUJI Hideaki3b1e0a62008-03-26 02:26:21 +0900734 struct net *net = sock_net(sk);
Gao fengda12c902013-06-06 14:49:11 +0800735 struct netlink_table *table = &nl_table[sk->sk_protocol];
Eric W. Biederman15e47302012-09-07 20:12:54 +0000736 s32 portid = task_tgid_vnr(current);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700737 int err;
Herbert Xub9fbe702015-05-17 10:45:34 +0800738 s32 rover = -4096;
739 bool ok;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700740
741retry:
742 cond_resched();
Thomas Grafe3416942014-08-02 11:47:45 +0200743 rcu_read_lock();
Herbert Xub9fbe702015-05-17 10:45:34 +0800744 ok = !__netlink_lookup(table, portid, net);
745 rcu_read_unlock();
746 if (!ok) {
Thomas Grafe3416942014-08-02 11:47:45 +0200747 /* Bind collision, search negative portid values. */
Herbert Xub9fbe702015-05-17 10:45:34 +0800748 if (rover == -4096)
749 /* rover will be in range [S32_MIN, -4097] */
750 rover = S32_MIN + prandom_u32_max(-4096 - S32_MIN);
751 else if (rover >= -4096)
Thomas Grafe3416942014-08-02 11:47:45 +0200752 rover = -4097;
Herbert Xub9fbe702015-05-17 10:45:34 +0800753 portid = rover--;
Thomas Grafe3416942014-08-02 11:47:45 +0200754 goto retry;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700755 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700756
Herbert Xu8ea65f42015-01-26 14:02:56 +1100757 err = netlink_insert(sk, portid);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700758 if (err == -EADDRINUSE)
759 goto retry;
David S. Millerd470e3b2005-06-26 15:31:51 -0700760
761 /* If 2 threads race to autobind, that is fine. */
762 if (err == -EBUSY)
763 err = 0;
764
765 return err;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700766}
767
Eric W. Biedermanaa4cf942014-04-23 14:28:03 -0700768/**
769 * __netlink_ns_capable - General netlink message capability test
770 * @nsp: NETLINK_CB of the socket buffer holding a netlink command from userspace.
771 * @user_ns: The user namespace of the capability to use
772 * @cap: The capability to use
773 *
774 * Test to see if the opener of the socket we received the message
775 * from had when the netlink socket was created and the sender of the
776 * message has has the capability @cap in the user namespace @user_ns.
777 */
778bool __netlink_ns_capable(const struct netlink_skb_parms *nsp,
779 struct user_namespace *user_ns, int cap)
780{
Eric W. Biederman2d7a85f2014-05-30 11:04:00 -0700781 return ((nsp->flags & NETLINK_SKB_DST) ||
782 file_ns_capable(nsp->sk->sk_socket->file, user_ns, cap)) &&
783 ns_capable(user_ns, cap);
Eric W. Biedermanaa4cf942014-04-23 14:28:03 -0700784}
785EXPORT_SYMBOL(__netlink_ns_capable);
786
787/**
788 * netlink_ns_capable - General netlink message capability test
789 * @skb: socket buffer holding a netlink command from userspace
790 * @user_ns: The user namespace of the capability to use
791 * @cap: The capability to use
792 *
793 * Test to see if the opener of the socket we received the message
794 * from had when the netlink socket was created and the sender of the
795 * message has has the capability @cap in the user namespace @user_ns.
796 */
797bool netlink_ns_capable(const struct sk_buff *skb,
798 struct user_namespace *user_ns, int cap)
799{
800 return __netlink_ns_capable(&NETLINK_CB(skb), user_ns, cap);
801}
802EXPORT_SYMBOL(netlink_ns_capable);
803
804/**
805 * netlink_capable - Netlink global message capability test
806 * @skb: socket buffer holding a netlink command from userspace
807 * @cap: The capability to use
808 *
809 * Test to see if the opener of the socket we received the message
810 * from had when the netlink socket was created and the sender of the
811 * message has has the capability @cap in all user namespaces.
812 */
813bool netlink_capable(const struct sk_buff *skb, int cap)
814{
815 return netlink_ns_capable(skb, &init_user_ns, cap);
816}
817EXPORT_SYMBOL(netlink_capable);
818
819/**
820 * netlink_net_capable - Netlink network namespace message capability test
821 * @skb: socket buffer holding a netlink command from userspace
822 * @cap: The capability to use
823 *
824 * Test to see if the opener of the socket we received the message
825 * from had when the netlink socket was created and the sender of the
826 * message has has the capability @cap over the network namespace of
827 * the socket we received the message from.
828 */
829bool netlink_net_capable(const struct sk_buff *skb, int cap)
830{
831 return netlink_ns_capable(skb, sock_net(skb->sk)->user_ns, cap);
832}
833EXPORT_SYMBOL(netlink_net_capable);
834
Eric W. Biederman5187cd02014-04-23 14:25:48 -0700835static inline int netlink_allowed(const struct socket *sock, unsigned int flag)
YOSHIFUJI Hideaki746fac42007-02-09 23:25:07 +0900836{
Pablo Neira Ayuso9785e102012-09-08 02:53:53 +0000837 return (nl_table[sock->sk->sk_protocol].flags & flag) ||
Eric W. Biedermandf008c92012-11-16 03:03:07 +0000838 ns_capable(sock_net(sock->sk)->user_ns, CAP_NET_ADMIN);
YOSHIFUJI Hideaki746fac42007-02-09 23:25:07 +0900839}
Linus Torvalds1da177e2005-04-16 15:20:36 -0700840
Patrick McHardyf7fa9b12005-08-15 12:29:13 -0700841static void
842netlink_update_subscriptions(struct sock *sk, unsigned int subscriptions)
843{
844 struct netlink_sock *nlk = nlk_sk(sk);
845
846 if (nlk->subscriptions && !subscriptions)
847 __sk_del_bind_node(sk);
848 else if (!nlk->subscriptions && subscriptions)
849 sk_add_bind_node(sk, &nl_table[sk->sk_protocol].mc_list);
850 nlk->subscriptions = subscriptions;
851}
852
Johannes Bergb4ff4f02007-07-18 15:46:06 -0700853static int netlink_realloc_groups(struct sock *sk)
Patrick McHardy513c2502005-09-06 15:43:59 -0700854{
855 struct netlink_sock *nlk = nlk_sk(sk);
856 unsigned int groups;
Johannes Bergb4ff4f02007-07-18 15:46:06 -0700857 unsigned long *new_groups;
Patrick McHardy513c2502005-09-06 15:43:59 -0700858 int err = 0;
859
Johannes Bergb4ff4f02007-07-18 15:46:06 -0700860 netlink_table_grab();
861
Patrick McHardy513c2502005-09-06 15:43:59 -0700862 groups = nl_table[sk->sk_protocol].groups;
Johannes Bergb4ff4f02007-07-18 15:46:06 -0700863 if (!nl_table[sk->sk_protocol].registered) {
Patrick McHardy513c2502005-09-06 15:43:59 -0700864 err = -ENOENT;
Johannes Bergb4ff4f02007-07-18 15:46:06 -0700865 goto out_unlock;
866 }
Patrick McHardy513c2502005-09-06 15:43:59 -0700867
Johannes Bergb4ff4f02007-07-18 15:46:06 -0700868 if (nlk->ngroups >= groups)
869 goto out_unlock;
Patrick McHardy513c2502005-09-06 15:43:59 -0700870
Johannes Bergb4ff4f02007-07-18 15:46:06 -0700871 new_groups = krealloc(nlk->groups, NLGRPSZ(groups), GFP_ATOMIC);
872 if (new_groups == NULL) {
873 err = -ENOMEM;
874 goto out_unlock;
875 }
Patrick McHardy6ac552f2007-12-04 00:19:38 -0800876 memset((char *)new_groups + NLGRPSZ(nlk->ngroups), 0,
Johannes Bergb4ff4f02007-07-18 15:46:06 -0700877 NLGRPSZ(groups) - NLGRPSZ(nlk->ngroups));
878
879 nlk->groups = new_groups;
Patrick McHardy513c2502005-09-06 15:43:59 -0700880 nlk->ngroups = groups;
Johannes Bergb4ff4f02007-07-18 15:46:06 -0700881 out_unlock:
882 netlink_table_ungrab();
883 return err;
Patrick McHardy513c2502005-09-06 15:43:59 -0700884}
885
Johannes Berg02c81ab2014-12-22 18:56:35 +0100886static void netlink_undo_bind(int group, long unsigned int groups,
Johannes Berg023e2cf2014-12-23 21:00:06 +0100887 struct sock *sk)
Richard Guy Briggs4f520902014-04-22 21:31:54 -0400888{
Johannes Berg023e2cf2014-12-23 21:00:06 +0100889 struct netlink_sock *nlk = nlk_sk(sk);
Richard Guy Briggs4f520902014-04-22 21:31:54 -0400890 int undo;
891
892 if (!nlk->netlink_unbind)
893 return;
894
895 for (undo = 0; undo < group; undo++)
Hiroaki SHIMODA6251edd2014-11-13 04:24:10 +0900896 if (test_bit(undo, &groups))
Pablo Neira8b7c36d2015-01-29 10:51:53 +0100897 nlk->netlink_unbind(sock_net(sk), undo + 1);
Richard Guy Briggs4f520902014-04-22 21:31:54 -0400898}
899
Patrick McHardy6ac552f2007-12-04 00:19:38 -0800900static int netlink_bind(struct socket *sock, struct sockaddr *addr,
901 int addr_len)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700902{
903 struct sock *sk = sock->sk;
YOSHIFUJI Hideaki3b1e0a62008-03-26 02:26:21 +0900904 struct net *net = sock_net(sk);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700905 struct netlink_sock *nlk = nlk_sk(sk);
906 struct sockaddr_nl *nladdr = (struct sockaddr_nl *)addr;
907 int err;
Richard Guy Briggs4f520902014-04-22 21:31:54 -0400908 long unsigned int groups = nladdr->nl_groups;
Herbert Xuda314c92015-09-22 11:38:56 +0800909 bool bound;
YOSHIFUJI Hideaki746fac42007-02-09 23:25:07 +0900910
Hannes Frederic Sowa4e4b5372012-12-15 15:42:19 +0000911 if (addr_len < sizeof(struct sockaddr_nl))
912 return -EINVAL;
913
Linus Torvalds1da177e2005-04-16 15:20:36 -0700914 if (nladdr->nl_family != AF_NETLINK)
915 return -EINVAL;
916
917 /* Only superuser is allowed to listen multicasts */
Richard Guy Briggs4f520902014-04-22 21:31:54 -0400918 if (groups) {
Eric W. Biederman5187cd02014-04-23 14:25:48 -0700919 if (!netlink_allowed(sock, NL_CFG_F_NONROOT_RECV))
Patrick McHardy513c2502005-09-06 15:43:59 -0700920 return -EPERM;
Johannes Bergb4ff4f02007-07-18 15:46:06 -0700921 err = netlink_realloc_groups(sk);
922 if (err)
923 return err;
Patrick McHardy513c2502005-09-06 15:43:59 -0700924 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700925
Herbert Xuda314c92015-09-22 11:38:56 +0800926 bound = nlk->bound;
927 if (bound) {
928 /* Ensure nlk->portid is up-to-date. */
929 smp_rmb();
930
Eric W. Biederman15e47302012-09-07 20:12:54 +0000931 if (nladdr->nl_pid != nlk->portid)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700932 return -EINVAL;
Herbert Xuda314c92015-09-22 11:38:56 +0800933 }
Richard Guy Briggs4f520902014-04-22 21:31:54 -0400934
935 if (nlk->netlink_bind && groups) {
936 int group;
937
938 for (group = 0; group < nlk->ngroups; group++) {
939 if (!test_bit(group, &groups))
940 continue;
Pablo Neira8b7c36d2015-01-29 10:51:53 +0100941 err = nlk->netlink_bind(net, group + 1);
Richard Guy Briggs4f520902014-04-22 21:31:54 -0400942 if (!err)
943 continue;
Johannes Berg023e2cf2014-12-23 21:00:06 +0100944 netlink_undo_bind(group, groups, sk);
Richard Guy Briggs4f520902014-04-22 21:31:54 -0400945 return err;
946 }
947 }
948
Herbert Xuda314c92015-09-22 11:38:56 +0800949 /* No need for barriers here as we return to user-space without
950 * using any of the bound attributes.
951 */
952 if (!bound) {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700953 err = nladdr->nl_pid ?
Herbert Xu8ea65f42015-01-26 14:02:56 +1100954 netlink_insert(sk, nladdr->nl_pid) :
Linus Torvalds1da177e2005-04-16 15:20:36 -0700955 netlink_autobind(sock);
Richard Guy Briggs4f520902014-04-22 21:31:54 -0400956 if (err) {
Johannes Berg023e2cf2014-12-23 21:00:06 +0100957 netlink_undo_bind(nlk->ngroups, groups, sk);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700958 return err;
Richard Guy Briggs4f520902014-04-22 21:31:54 -0400959 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700960 }
961
Richard Guy Briggs4f520902014-04-22 21:31:54 -0400962 if (!groups && (nlk->groups == NULL || !(u32)nlk->groups[0]))
Linus Torvalds1da177e2005-04-16 15:20:36 -0700963 return 0;
964
965 netlink_table_grab();
Patrick McHardyf7fa9b12005-08-15 12:29:13 -0700966 netlink_update_subscriptions(sk, nlk->subscriptions +
Richard Guy Briggs4f520902014-04-22 21:31:54 -0400967 hweight32(groups) -
YOSHIFUJI Hideaki746fac42007-02-09 23:25:07 +0900968 hweight32(nlk->groups[0]));
Richard Guy Briggs4f520902014-04-22 21:31:54 -0400969 nlk->groups[0] = (nlk->groups[0] & ~0xffffffffUL) | groups;
Patrick McHardy4277a082006-03-20 18:52:01 -0800970 netlink_update_listeners(sk);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700971 netlink_table_ungrab();
972
973 return 0;
974}
975
976static int netlink_connect(struct socket *sock, struct sockaddr *addr,
977 int alen, int flags)
978{
979 int err = 0;
980 struct sock *sk = sock->sk;
981 struct netlink_sock *nlk = nlk_sk(sk);
Patrick McHardy6ac552f2007-12-04 00:19:38 -0800982 struct sockaddr_nl *nladdr = (struct sockaddr_nl *)addr;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700983
Changli Gao6503d962010-03-31 22:58:26 +0000984 if (alen < sizeof(addr->sa_family))
985 return -EINVAL;
986
Linus Torvalds1da177e2005-04-16 15:20:36 -0700987 if (addr->sa_family == AF_UNSPEC) {
988 sk->sk_state = NETLINK_UNCONNECTED;
Eric W. Biederman15e47302012-09-07 20:12:54 +0000989 nlk->dst_portid = 0;
Patrick McHardyd629b832005-08-14 19:27:50 -0700990 nlk->dst_group = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700991 return 0;
992 }
993 if (addr->sa_family != AF_NETLINK)
994 return -EINVAL;
995
Mike Pecovnik46833a82014-02-24 21:11:16 +0100996 if ((nladdr->nl_groups || nladdr->nl_pid) &&
Eric W. Biederman5187cd02014-04-23 14:25:48 -0700997 !netlink_allowed(sock, NL_CFG_F_NONROOT_SEND))
Linus Torvalds1da177e2005-04-16 15:20:36 -0700998 return -EPERM;
999
Herbert Xuda314c92015-09-22 11:38:56 +08001000 /* No need for barriers here as we return to user-space without
1001 * using any of the bound attributes.
1002 */
1003 if (!nlk->bound)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001004 err = netlink_autobind(sock);
1005
1006 if (err == 0) {
1007 sk->sk_state = NETLINK_CONNECTED;
Eric W. Biederman15e47302012-09-07 20:12:54 +00001008 nlk->dst_portid = nladdr->nl_pid;
Patrick McHardyd629b832005-08-14 19:27:50 -07001009 nlk->dst_group = ffs(nladdr->nl_groups);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001010 }
1011
1012 return err;
1013}
1014
Patrick McHardy6ac552f2007-12-04 00:19:38 -08001015static int netlink_getname(struct socket *sock, struct sockaddr *addr,
1016 int *addr_len, int peer)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001017{
1018 struct sock *sk = sock->sk;
1019 struct netlink_sock *nlk = nlk_sk(sk);
Cyrill Gorcunov13cfa972009-11-08 05:51:19 +00001020 DECLARE_SOCKADDR(struct sockaddr_nl *, nladdr, addr);
YOSHIFUJI Hideaki746fac42007-02-09 23:25:07 +09001021
Linus Torvalds1da177e2005-04-16 15:20:36 -07001022 nladdr->nl_family = AF_NETLINK;
1023 nladdr->nl_pad = 0;
1024 *addr_len = sizeof(*nladdr);
1025
1026 if (peer) {
Eric W. Biederman15e47302012-09-07 20:12:54 +00001027 nladdr->nl_pid = nlk->dst_portid;
Patrick McHardyd629b832005-08-14 19:27:50 -07001028 nladdr->nl_groups = netlink_group_mask(nlk->dst_group);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001029 } else {
Eric W. Biederman15e47302012-09-07 20:12:54 +00001030 nladdr->nl_pid = nlk->portid;
Patrick McHardy513c2502005-09-06 15:43:59 -07001031 nladdr->nl_groups = nlk->groups ? nlk->groups[0] : 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001032 }
1033 return 0;
1034}
1035
David Decotigny025c6812016-03-21 10:15:35 -07001036static int netlink_ioctl(struct socket *sock, unsigned int cmd,
1037 unsigned long arg)
1038{
1039 /* try to hand this ioctl down to the NIC drivers.
1040 */
1041 return -ENOIOCTLCMD;
1042}
1043
Eric W. Biederman15e47302012-09-07 20:12:54 +00001044static struct sock *netlink_getsockbyportid(struct sock *ssk, u32 portid)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001045{
Linus Torvalds1da177e2005-04-16 15:20:36 -07001046 struct sock *sock;
1047 struct netlink_sock *nlk;
1048
Eric W. Biederman15e47302012-09-07 20:12:54 +00001049 sock = netlink_lookup(sock_net(ssk), ssk->sk_protocol, portid);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001050 if (!sock)
1051 return ERR_PTR(-ECONNREFUSED);
1052
1053 /* Don't bother queuing skb if kernel socket has no input function */
1054 nlk = nlk_sk(sock);
Denis V. Lunevcd40b7d2007-10-10 21:15:29 -07001055 if (sock->sk_state == NETLINK_CONNECTED &&
Eric W. Biederman15e47302012-09-07 20:12:54 +00001056 nlk->dst_portid != nlk_sk(ssk)->portid) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07001057 sock_put(sock);
1058 return ERR_PTR(-ECONNREFUSED);
1059 }
1060 return sock;
1061}
1062
1063struct sock *netlink_getsockbyfilp(struct file *filp)
1064{
Al Viro496ad9a2013-01-23 17:07:38 -05001065 struct inode *inode = file_inode(filp);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001066 struct sock *sock;
1067
1068 if (!S_ISSOCK(inode->i_mode))
1069 return ERR_PTR(-ENOTSOCK);
1070
1071 sock = SOCKET_I(inode)->sk;
1072 if (sock->sk_family != AF_NETLINK)
1073 return ERR_PTR(-EINVAL);
1074
1075 sock_hold(sock);
1076 return sock;
1077}
1078
Pablo Neira3a365152013-06-28 03:04:23 +02001079static struct sk_buff *netlink_alloc_large_skb(unsigned int size,
1080 int broadcast)
Pablo Neira Ayusoc05cdb12013-06-03 09:46:28 +00001081{
1082 struct sk_buff *skb;
1083 void *data;
1084
Pablo Neira3a365152013-06-28 03:04:23 +02001085 if (size <= NLMSG_GOODSIZE || broadcast)
Pablo Neira Ayusoc05cdb12013-06-03 09:46:28 +00001086 return alloc_skb(size, GFP_KERNEL);
1087
Pablo Neira3a365152013-06-28 03:04:23 +02001088 size = SKB_DATA_ALIGN(size) +
1089 SKB_DATA_ALIGN(sizeof(struct skb_shared_info));
Pablo Neira Ayusoc05cdb12013-06-03 09:46:28 +00001090
1091 data = vmalloc(size);
1092 if (data == NULL)
Pablo Neira3a365152013-06-28 03:04:23 +02001093 return NULL;
Pablo Neira Ayusoc05cdb12013-06-03 09:46:28 +00001094
Eric Dumazet2ea2f622015-04-24 16:05:01 -07001095 skb = __build_skb(data, size);
Pablo Neira3a365152013-06-28 03:04:23 +02001096 if (skb == NULL)
1097 vfree(data);
Eric Dumazet2ea2f622015-04-24 16:05:01 -07001098 else
Pablo Neira3a365152013-06-28 03:04:23 +02001099 skb->destructor = netlink_skb_destructor;
Pablo Neira Ayusoc05cdb12013-06-03 09:46:28 +00001100
1101 return skb;
Pablo Neira Ayusoc05cdb12013-06-03 09:46:28 +00001102}
1103
Linus Torvalds1da177e2005-04-16 15:20:36 -07001104/*
1105 * Attach a skb to a netlink socket.
1106 * The caller must hold a reference to the destination socket. On error, the
1107 * reference is dropped. The skb is not send to the destination, just all
1108 * all error checks are performed and memory in the queue is reserved.
1109 * Return values:
1110 * < 0: error. skb freed, reference to sock dropped.
1111 * 0: continue
1112 * 1: repeat lookup - reference dropped while waiting for socket memory.
1113 */
Denis V. Lunev9457afe2008-06-05 11:23:39 -07001114int netlink_attachskb(struct sock *sk, struct sk_buff *skb,
Patrick McHardyc3d8d1e2007-11-07 02:42:09 -08001115 long *timeo, struct sock *ssk)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001116{
1117 struct netlink_sock *nlk;
1118
1119 nlk = nlk_sk(sk);
1120
Patrick McHardy5fd96122013-04-17 06:47:03 +00001121 if ((atomic_read(&sk->sk_rmem_alloc) > sk->sk_rcvbuf ||
Florian Westphald1b4c682016-02-18 15:03:24 +01001122 test_bit(NETLINK_S_CONGESTED, &nlk->state))) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07001123 DECLARE_WAITQUEUE(wait, current);
Patrick McHardyc3d8d1e2007-11-07 02:42:09 -08001124 if (!*timeo) {
Denis V. Lunevaed81562007-10-10 21:14:32 -07001125 if (!ssk || netlink_is_kernel(ssk))
Linus Torvalds1da177e2005-04-16 15:20:36 -07001126 netlink_overrun(sk);
1127 sock_put(sk);
1128 kfree_skb(skb);
1129 return -EAGAIN;
1130 }
1131
1132 __set_current_state(TASK_INTERRUPTIBLE);
1133 add_wait_queue(&nlk->wait, &wait);
1134
1135 if ((atomic_read(&sk->sk_rmem_alloc) > sk->sk_rcvbuf ||
Nicolas Dichtelcc3a5722015-05-07 11:02:52 +02001136 test_bit(NETLINK_S_CONGESTED, &nlk->state)) &&
Linus Torvalds1da177e2005-04-16 15:20:36 -07001137 !sock_flag(sk, SOCK_DEAD))
Patrick McHardyc3d8d1e2007-11-07 02:42:09 -08001138 *timeo = schedule_timeout(*timeo);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001139
1140 __set_current_state(TASK_RUNNING);
1141 remove_wait_queue(&nlk->wait, &wait);
1142 sock_put(sk);
1143
1144 if (signal_pending(current)) {
1145 kfree_skb(skb);
Patrick McHardyc3d8d1e2007-11-07 02:42:09 -08001146 return sock_intr_errno(*timeo);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001147 }
1148 return 1;
1149 }
Patrick McHardycf0a0182013-04-17 06:47:00 +00001150 netlink_skb_set_owner_r(skb, sk);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001151 return 0;
1152}
1153
Eric Dumazet4a7e7c22012-04-05 22:17:46 +00001154static int __netlink_sendskb(struct sock *sk, struct sk_buff *skb)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001155{
Linus Torvalds1da177e2005-04-16 15:20:36 -07001156 int len = skb->len;
1157
Daniel Borkmannbcbde0d2013-06-21 19:38:07 +02001158 netlink_deliver_tap(skb);
1159
Florian Westphald1b4c682016-02-18 15:03:24 +01001160 skb_queue_tail(&sk->sk_receive_queue, skb);
David S. Miller676d2362014-04-11 16:15:36 -04001161 sk->sk_data_ready(sk);
Eric Dumazet4a7e7c22012-04-05 22:17:46 +00001162 return len;
1163}
1164
1165int netlink_sendskb(struct sock *sk, struct sk_buff *skb)
1166{
1167 int len = __netlink_sendskb(sk, skb);
1168
Linus Torvalds1da177e2005-04-16 15:20:36 -07001169 sock_put(sk);
1170 return len;
1171}
1172
1173void netlink_detachskb(struct sock *sk, struct sk_buff *skb)
1174{
1175 kfree_skb(skb);
1176 sock_put(sk);
1177}
1178
stephen hemmingerb57ef81f2011-12-22 08:52:02 +00001179static struct sk_buff *netlink_trim(struct sk_buff *skb, gfp_t allocation)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001180{
1181 int delta;
1182
Patrick McHardy1298ca42013-04-17 06:46:59 +00001183 WARN_ON(skb->sk != NULL);
Arnaldo Carvalho de Melo4305b542007-04-19 20:43:29 -07001184 delta = skb->end - skb->tail;
Pablo Neira Ayusoc05cdb12013-06-03 09:46:28 +00001185 if (is_vmalloc_addr(skb->head) || delta * 2 < skb->truesize)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001186 return skb;
1187
1188 if (skb_shared(skb)) {
1189 struct sk_buff *nskb = skb_clone(skb, allocation);
1190 if (!nskb)
1191 return skb;
Eric Dumazet8460c002012-04-19 02:24:28 +00001192 consume_skb(skb);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001193 skb = nskb;
1194 }
1195
1196 if (!pskb_expand_head(skb, 0, -delta, allocation))
1197 skb->truesize -= delta;
1198
1199 return skb;
1200}
1201
Eric W. Biederman3fbc2902012-05-24 17:21:27 -06001202static int netlink_unicast_kernel(struct sock *sk, struct sk_buff *skb,
1203 struct sock *ssk)
Denis V. Lunevcd40b7d2007-10-10 21:15:29 -07001204{
1205 int ret;
1206 struct netlink_sock *nlk = nlk_sk(sk);
1207
1208 ret = -ECONNREFUSED;
1209 if (nlk->netlink_rcv != NULL) {
1210 ret = skb->len;
Patrick McHardycf0a0182013-04-17 06:47:00 +00001211 netlink_skb_set_owner_r(skb, sk);
Patrick McHardye32123e2013-04-17 06:46:57 +00001212 NETLINK_CB(skb).sk = ssk;
Daniel Borkmann73bfd372013-12-23 14:35:55 +01001213 netlink_deliver_tap_kernel(sk, ssk, skb);
Denis V. Lunevcd40b7d2007-10-10 21:15:29 -07001214 nlk->netlink_rcv(skb);
Eric Dumazetbfb253c2012-04-22 21:30:29 +00001215 consume_skb(skb);
1216 } else {
1217 kfree_skb(skb);
Denis V. Lunevcd40b7d2007-10-10 21:15:29 -07001218 }
Denis V. Lunevcd40b7d2007-10-10 21:15:29 -07001219 sock_put(sk);
1220 return ret;
1221}
1222
1223int netlink_unicast(struct sock *ssk, struct sk_buff *skb,
Eric W. Biederman15e47302012-09-07 20:12:54 +00001224 u32 portid, int nonblock)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001225{
1226 struct sock *sk;
1227 int err;
1228 long timeo;
1229
1230 skb = netlink_trim(skb, gfp_any());
1231
1232 timeo = sock_sndtimeo(ssk, nonblock);
1233retry:
Eric W. Biederman15e47302012-09-07 20:12:54 +00001234 sk = netlink_getsockbyportid(ssk, portid);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001235 if (IS_ERR(sk)) {
1236 kfree_skb(skb);
1237 return PTR_ERR(sk);
1238 }
Denis V. Lunevcd40b7d2007-10-10 21:15:29 -07001239 if (netlink_is_kernel(sk))
Eric W. Biederman3fbc2902012-05-24 17:21:27 -06001240 return netlink_unicast_kernel(sk, skb, ssk);
Denis V. Lunevcd40b7d2007-10-10 21:15:29 -07001241
Stephen Hemmingerb1153f22008-03-21 15:46:12 -07001242 if (sk_filter(sk, skb)) {
Wang Chen84874602008-07-01 19:55:09 -07001243 err = skb->len;
Stephen Hemmingerb1153f22008-03-21 15:46:12 -07001244 kfree_skb(skb);
1245 sock_put(sk);
1246 return err;
1247 }
1248
Denis V. Lunev9457afe2008-06-05 11:23:39 -07001249 err = netlink_attachskb(sk, skb, &timeo, ssk);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001250 if (err == 1)
1251 goto retry;
1252 if (err)
1253 return err;
1254
Denis V. Lunev7ee015e2007-10-10 21:14:03 -07001255 return netlink_sendskb(sk, skb);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001256}
Patrick McHardy6ac552f2007-12-04 00:19:38 -08001257EXPORT_SYMBOL(netlink_unicast);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001258
Patrick McHardy4277a082006-03-20 18:52:01 -08001259int netlink_has_listeners(struct sock *sk, unsigned int group)
1260{
1261 int res = 0;
Eric Dumazet5c398dc2010-10-24 04:27:10 +00001262 struct listeners *listeners;
Patrick McHardy4277a082006-03-20 18:52:01 -08001263
Denis V. Lunevaed81562007-10-10 21:14:32 -07001264 BUG_ON(!netlink_is_kernel(sk));
Johannes Bergb4ff4f02007-07-18 15:46:06 -07001265
1266 rcu_read_lock();
1267 listeners = rcu_dereference(nl_table[sk->sk_protocol].listeners);
1268
Eric Dumazet6d772ac2012-10-18 03:21:55 +00001269 if (listeners && group - 1 < nl_table[sk->sk_protocol].groups)
Eric Dumazet5c398dc2010-10-24 04:27:10 +00001270 res = test_bit(group - 1, listeners->masks);
Johannes Bergb4ff4f02007-07-18 15:46:06 -07001271
1272 rcu_read_unlock();
1273
Patrick McHardy4277a082006-03-20 18:52:01 -08001274 return res;
1275}
1276EXPORT_SYMBOL_GPL(netlink_has_listeners);
1277
stephen hemmingerb57ef81f2011-12-22 08:52:02 +00001278static int netlink_broadcast_deliver(struct sock *sk, struct sk_buff *skb)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001279{
1280 struct netlink_sock *nlk = nlk_sk(sk);
1281
1282 if (atomic_read(&sk->sk_rmem_alloc) <= sk->sk_rcvbuf &&
Nicolas Dichtelcc3a5722015-05-07 11:02:52 +02001283 !test_bit(NETLINK_S_CONGESTED, &nlk->state)) {
Patrick McHardycf0a0182013-04-17 06:47:00 +00001284 netlink_skb_set_owner_r(skb, sk);
Eric Dumazet4a7e7c22012-04-05 22:17:46 +00001285 __netlink_sendskb(sk, skb);
stephen hemminger2c6458002011-12-22 08:52:03 +00001286 return atomic_read(&sk->sk_rmem_alloc) > (sk->sk_rcvbuf >> 1);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001287 }
1288 return -1;
1289}
1290
1291struct netlink_broadcast_data {
1292 struct sock *exclude_sk;
Eric W. Biedermanb4b51022007-09-12 13:05:38 +02001293 struct net *net;
Eric W. Biederman15e47302012-09-07 20:12:54 +00001294 u32 portid;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001295 u32 group;
1296 int failure;
Pablo Neira Ayusoff491a72009-02-05 23:56:36 -08001297 int delivery_failure;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001298 int congested;
1299 int delivered;
Al Viro7d877f32005-10-21 03:20:43 -04001300 gfp_t allocation;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001301 struct sk_buff *skb, *skb2;
Eric W. Biederman910a7e92010-05-04 17:36:46 -07001302 int (*tx_filter)(struct sock *dsk, struct sk_buff *skb, void *data);
1303 void *tx_data;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001304};
1305
Rami Rosen46c95212014-07-01 21:17:35 +03001306static void do_one_broadcast(struct sock *sk,
1307 struct netlink_broadcast_data *p)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001308{
1309 struct netlink_sock *nlk = nlk_sk(sk);
1310 int val;
1311
1312 if (p->exclude_sk == sk)
Rami Rosen46c95212014-07-01 21:17:35 +03001313 return;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001314
Eric W. Biederman15e47302012-09-07 20:12:54 +00001315 if (nlk->portid == p->portid || p->group - 1 >= nlk->ngroups ||
Patrick McHardyf7fa9b12005-08-15 12:29:13 -07001316 !test_bit(p->group - 1, nlk->groups))
Rami Rosen46c95212014-07-01 21:17:35 +03001317 return;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001318
Nicolas Dichtel59324cf2015-05-07 11:02:53 +02001319 if (!net_eq(sock_net(sk), p->net)) {
1320 if (!(nlk->flags & NETLINK_F_LISTEN_ALL_NSID))
1321 return;
1322
1323 if (!peernet_has_id(sock_net(sk), p->net))
1324 return;
1325
1326 if (!file_ns_capable(sk->sk_socket->file, p->net->user_ns,
1327 CAP_NET_BROADCAST))
1328 return;
1329 }
Eric W. Biedermanb4b51022007-09-12 13:05:38 +02001330
Linus Torvalds1da177e2005-04-16 15:20:36 -07001331 if (p->failure) {
1332 netlink_overrun(sk);
Rami Rosen46c95212014-07-01 21:17:35 +03001333 return;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001334 }
1335
1336 sock_hold(sk);
1337 if (p->skb2 == NULL) {
Tommy S. Christensen68acc022005-05-19 13:06:35 -07001338 if (skb_shared(p->skb)) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07001339 p->skb2 = skb_clone(p->skb, p->allocation);
1340 } else {
Tommy S. Christensen68acc022005-05-19 13:06:35 -07001341 p->skb2 = skb_get(p->skb);
1342 /*
1343 * skb ownership may have been set when
1344 * delivered to a previous socket.
1345 */
1346 skb_orphan(p->skb2);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001347 }
1348 }
1349 if (p->skb2 == NULL) {
1350 netlink_overrun(sk);
1351 /* Clone failed. Notify ALL listeners. */
1352 p->failure = 1;
Nicolas Dichtelcc3a5722015-05-07 11:02:52 +02001353 if (nlk->flags & NETLINK_F_BROADCAST_SEND_ERROR)
Pablo Neira Ayusobe0c22a2009-02-18 01:40:43 +00001354 p->delivery_failure = 1;
Nicolas Dichtel59324cf2015-05-07 11:02:53 +02001355 goto out;
1356 }
1357 if (p->tx_filter && p->tx_filter(sk, p->skb2, p->tx_data)) {
Eric W. Biederman910a7e92010-05-04 17:36:46 -07001358 kfree_skb(p->skb2);
1359 p->skb2 = NULL;
Nicolas Dichtel59324cf2015-05-07 11:02:53 +02001360 goto out;
1361 }
1362 if (sk_filter(sk, p->skb2)) {
Stephen Hemmingerb1153f22008-03-21 15:46:12 -07001363 kfree_skb(p->skb2);
1364 p->skb2 = NULL;
Nicolas Dichtel59324cf2015-05-07 11:02:53 +02001365 goto out;
1366 }
1367 NETLINK_CB(p->skb2).nsid = peernet2id(sock_net(sk), p->net);
1368 NETLINK_CB(p->skb2).nsid_is_set = true;
1369 val = netlink_broadcast_deliver(sk, p->skb2);
1370 if (val < 0) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07001371 netlink_overrun(sk);
Nicolas Dichtelcc3a5722015-05-07 11:02:52 +02001372 if (nlk->flags & NETLINK_F_BROADCAST_SEND_ERROR)
Pablo Neira Ayusobe0c22a2009-02-18 01:40:43 +00001373 p->delivery_failure = 1;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001374 } else {
1375 p->congested |= val;
1376 p->delivered = 1;
1377 p->skb2 = NULL;
1378 }
Nicolas Dichtel59324cf2015-05-07 11:02:53 +02001379out:
Linus Torvalds1da177e2005-04-16 15:20:36 -07001380 sock_put(sk);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001381}
1382
Eric W. Biederman15e47302012-09-07 20:12:54 +00001383int netlink_broadcast_filtered(struct sock *ssk, struct sk_buff *skb, u32 portid,
Eric W. Biederman910a7e92010-05-04 17:36:46 -07001384 u32 group, gfp_t allocation,
1385 int (*filter)(struct sock *dsk, struct sk_buff *skb, void *data),
1386 void *filter_data)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001387{
YOSHIFUJI Hideaki3b1e0a62008-03-26 02:26:21 +09001388 struct net *net = sock_net(ssk);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001389 struct netlink_broadcast_data info;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001390 struct sock *sk;
1391
1392 skb = netlink_trim(skb, allocation);
1393
1394 info.exclude_sk = ssk;
Eric W. Biedermanb4b51022007-09-12 13:05:38 +02001395 info.net = net;
Eric W. Biederman15e47302012-09-07 20:12:54 +00001396 info.portid = portid;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001397 info.group = group;
1398 info.failure = 0;
Pablo Neira Ayusoff491a72009-02-05 23:56:36 -08001399 info.delivery_failure = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001400 info.congested = 0;
1401 info.delivered = 0;
1402 info.allocation = allocation;
1403 info.skb = skb;
1404 info.skb2 = NULL;
Eric W. Biederman910a7e92010-05-04 17:36:46 -07001405 info.tx_filter = filter;
1406 info.tx_data = filter_data;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001407
1408 /* While we sleep in clone, do not allow to change socket list */
1409
1410 netlink_lock_table();
1411
Sasha Levinb67bfe02013-02-27 17:06:00 -08001412 sk_for_each_bound(sk, &nl_table[ssk->sk_protocol].mc_list)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001413 do_one_broadcast(sk, &info);
1414
Neil Horman70d4bf62010-07-20 06:45:56 +00001415 consume_skb(skb);
Tommy S. Christensenaa1c6a62005-05-19 13:07:32 -07001416
Linus Torvalds1da177e2005-04-16 15:20:36 -07001417 netlink_unlock_table();
1418
Neil Horman70d4bf62010-07-20 06:45:56 +00001419 if (info.delivery_failure) {
1420 kfree_skb(info.skb2);
Pablo Neira Ayusoff491a72009-02-05 23:56:36 -08001421 return -ENOBUFS;
Eric Dumazet658cb352012-04-22 21:30:21 +00001422 }
1423 consume_skb(info.skb2);
Pablo Neira Ayusoff491a72009-02-05 23:56:36 -08001424
Linus Torvalds1da177e2005-04-16 15:20:36 -07001425 if (info.delivered) {
Mel Gormand0164ad2015-11-06 16:28:21 -08001426 if (info.congested && gfpflags_allow_blocking(allocation))
Linus Torvalds1da177e2005-04-16 15:20:36 -07001427 yield();
1428 return 0;
1429 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001430 return -ESRCH;
1431}
Eric W. Biederman910a7e92010-05-04 17:36:46 -07001432EXPORT_SYMBOL(netlink_broadcast_filtered);
1433
Eric W. Biederman15e47302012-09-07 20:12:54 +00001434int netlink_broadcast(struct sock *ssk, struct sk_buff *skb, u32 portid,
Eric W. Biederman910a7e92010-05-04 17:36:46 -07001435 u32 group, gfp_t allocation)
1436{
Eric W. Biederman15e47302012-09-07 20:12:54 +00001437 return netlink_broadcast_filtered(ssk, skb, portid, group, allocation,
Eric W. Biederman910a7e92010-05-04 17:36:46 -07001438 NULL, NULL);
1439}
Patrick McHardy6ac552f2007-12-04 00:19:38 -08001440EXPORT_SYMBOL(netlink_broadcast);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001441
1442struct netlink_set_err_data {
1443 struct sock *exclude_sk;
Eric W. Biederman15e47302012-09-07 20:12:54 +00001444 u32 portid;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001445 u32 group;
1446 int code;
1447};
1448
stephen hemmingerb57ef81f2011-12-22 08:52:02 +00001449static int do_one_set_err(struct sock *sk, struct netlink_set_err_data *p)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001450{
1451 struct netlink_sock *nlk = nlk_sk(sk);
Pablo Neira Ayuso1a503072010-03-18 14:24:42 +00001452 int ret = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001453
1454 if (sk == p->exclude_sk)
1455 goto out;
1456
Octavian Purdila09ad9bc2009-11-25 15:14:13 -08001457 if (!net_eq(sock_net(sk), sock_net(p->exclude_sk)))
Eric W. Biedermanb4b51022007-09-12 13:05:38 +02001458 goto out;
1459
Eric W. Biederman15e47302012-09-07 20:12:54 +00001460 if (nlk->portid == p->portid || p->group - 1 >= nlk->ngroups ||
Patrick McHardyf7fa9b12005-08-15 12:29:13 -07001461 !test_bit(p->group - 1, nlk->groups))
Linus Torvalds1da177e2005-04-16 15:20:36 -07001462 goto out;
1463
Nicolas Dichtelcc3a5722015-05-07 11:02:52 +02001464 if (p->code == ENOBUFS && nlk->flags & NETLINK_F_RECV_NO_ENOBUFS) {
Pablo Neira Ayuso1a503072010-03-18 14:24:42 +00001465 ret = 1;
1466 goto out;
1467 }
1468
Linus Torvalds1da177e2005-04-16 15:20:36 -07001469 sk->sk_err = p->code;
1470 sk->sk_error_report(sk);
1471out:
Pablo Neira Ayuso1a503072010-03-18 14:24:42 +00001472 return ret;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001473}
1474
Pablo Neira Ayuso4843b932009-03-03 23:37:30 -08001475/**
1476 * netlink_set_err - report error to broadcast listeners
1477 * @ssk: the kernel netlink socket, as returned by netlink_kernel_create()
Eric W. Biederman15e47302012-09-07 20:12:54 +00001478 * @portid: the PORTID of a process that we want to skip (if any)
Johannes Berg840e93f22013-11-19 10:35:40 +01001479 * @group: the broadcast group that will notice the error
Pablo Neira Ayuso4843b932009-03-03 23:37:30 -08001480 * @code: error code, must be negative (as usual in kernelspace)
Pablo Neira Ayuso1a503072010-03-18 14:24:42 +00001481 *
1482 * This function returns the number of broadcast listeners that have set the
Nicolas Dichtelcc3a5722015-05-07 11:02:52 +02001483 * NETLINK_NO_ENOBUFS socket option.
Pablo Neira Ayuso4843b932009-03-03 23:37:30 -08001484 */
Eric W. Biederman15e47302012-09-07 20:12:54 +00001485int netlink_set_err(struct sock *ssk, u32 portid, u32 group, int code)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001486{
1487 struct netlink_set_err_data info;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001488 struct sock *sk;
Pablo Neira Ayuso1a503072010-03-18 14:24:42 +00001489 int ret = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001490
1491 info.exclude_sk = ssk;
Eric W. Biederman15e47302012-09-07 20:12:54 +00001492 info.portid = portid;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001493 info.group = group;
Pablo Neira Ayuso4843b932009-03-03 23:37:30 -08001494 /* sk->sk_err wants a positive error value */
1495 info.code = -code;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001496
1497 read_lock(&nl_table_lock);
1498
Sasha Levinb67bfe02013-02-27 17:06:00 -08001499 sk_for_each_bound(sk, &nl_table[ssk->sk_protocol].mc_list)
Pablo Neira Ayuso1a503072010-03-18 14:24:42 +00001500 ret += do_one_set_err(sk, &info);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001501
1502 read_unlock(&nl_table_lock);
Pablo Neira Ayuso1a503072010-03-18 14:24:42 +00001503 return ret;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001504}
Pablo Neira Ayusodd5b6ce2009-03-23 13:21:06 +01001505EXPORT_SYMBOL(netlink_set_err);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001506
Johannes Berg84659eb2007-07-18 15:47:05 -07001507/* must be called with netlink table grabbed */
1508static void netlink_update_socket_mc(struct netlink_sock *nlk,
1509 unsigned int group,
1510 int is_new)
1511{
1512 int old, new = !!is_new, subscriptions;
1513
1514 old = test_bit(group - 1, nlk->groups);
1515 subscriptions = nlk->subscriptions - old + new;
1516 if (new)
1517 __set_bit(group - 1, nlk->groups);
1518 else
1519 __clear_bit(group - 1, nlk->groups);
1520 netlink_update_subscriptions(&nlk->sk, subscriptions);
1521 netlink_update_listeners(&nlk->sk);
1522}
1523
Patrick McHardy9a4595b2005-08-15 12:32:15 -07001524static int netlink_setsockopt(struct socket *sock, int level, int optname,
David S. Millerb7058842009-09-30 16:12:20 -07001525 char __user *optval, unsigned int optlen)
Patrick McHardy9a4595b2005-08-15 12:32:15 -07001526{
1527 struct sock *sk = sock->sk;
1528 struct netlink_sock *nlk = nlk_sk(sk);
Johannes Bergeb496532007-07-18 02:07:51 -07001529 unsigned int val = 0;
1530 int err;
Patrick McHardy9a4595b2005-08-15 12:32:15 -07001531
1532 if (level != SOL_NETLINK)
1533 return -ENOPROTOOPT;
1534
Florian Westphald1b4c682016-02-18 15:03:24 +01001535 if (optlen >= sizeof(int) &&
Johannes Bergeb496532007-07-18 02:07:51 -07001536 get_user(val, (unsigned int __user *)optval))
Patrick McHardy9a4595b2005-08-15 12:32:15 -07001537 return -EFAULT;
1538
1539 switch (optname) {
1540 case NETLINK_PKTINFO:
1541 if (val)
Nicolas Dichtelcc3a5722015-05-07 11:02:52 +02001542 nlk->flags |= NETLINK_F_RECV_PKTINFO;
Patrick McHardy9a4595b2005-08-15 12:32:15 -07001543 else
Nicolas Dichtelcc3a5722015-05-07 11:02:52 +02001544 nlk->flags &= ~NETLINK_F_RECV_PKTINFO;
Patrick McHardy9a4595b2005-08-15 12:32:15 -07001545 err = 0;
1546 break;
1547 case NETLINK_ADD_MEMBERSHIP:
1548 case NETLINK_DROP_MEMBERSHIP: {
Eric W. Biederman5187cd02014-04-23 14:25:48 -07001549 if (!netlink_allowed(sock, NL_CFG_F_NONROOT_RECV))
Patrick McHardy9a4595b2005-08-15 12:32:15 -07001550 return -EPERM;
Johannes Bergb4ff4f02007-07-18 15:46:06 -07001551 err = netlink_realloc_groups(sk);
1552 if (err)
1553 return err;
Patrick McHardy9a4595b2005-08-15 12:32:15 -07001554 if (!val || val - 1 >= nlk->ngroups)
1555 return -EINVAL;
Richard Guy Briggs7774d5e2014-04-22 21:31:55 -04001556 if (optname == NETLINK_ADD_MEMBERSHIP && nlk->netlink_bind) {
Johannes Berg023e2cf2014-12-23 21:00:06 +01001557 err = nlk->netlink_bind(sock_net(sk), val);
Richard Guy Briggs4f520902014-04-22 21:31:54 -04001558 if (err)
1559 return err;
1560 }
Patrick McHardy9a4595b2005-08-15 12:32:15 -07001561 netlink_table_grab();
Johannes Berg84659eb2007-07-18 15:47:05 -07001562 netlink_update_socket_mc(nlk, val,
1563 optname == NETLINK_ADD_MEMBERSHIP);
Patrick McHardy9a4595b2005-08-15 12:32:15 -07001564 netlink_table_ungrab();
Richard Guy Briggs7774d5e2014-04-22 21:31:55 -04001565 if (optname == NETLINK_DROP_MEMBERSHIP && nlk->netlink_unbind)
Johannes Berg023e2cf2014-12-23 21:00:06 +01001566 nlk->netlink_unbind(sock_net(sk), val);
Pablo Neira Ayuso03292742012-06-29 06:15:22 +00001567
Patrick McHardy9a4595b2005-08-15 12:32:15 -07001568 err = 0;
1569 break;
1570 }
Pablo Neira Ayusobe0c22a2009-02-18 01:40:43 +00001571 case NETLINK_BROADCAST_ERROR:
1572 if (val)
Nicolas Dichtelcc3a5722015-05-07 11:02:52 +02001573 nlk->flags |= NETLINK_F_BROADCAST_SEND_ERROR;
Pablo Neira Ayusobe0c22a2009-02-18 01:40:43 +00001574 else
Nicolas Dichtelcc3a5722015-05-07 11:02:52 +02001575 nlk->flags &= ~NETLINK_F_BROADCAST_SEND_ERROR;
Pablo Neira Ayusobe0c22a2009-02-18 01:40:43 +00001576 err = 0;
1577 break;
Pablo Neira Ayuso38938bf2009-03-24 16:37:55 -07001578 case NETLINK_NO_ENOBUFS:
1579 if (val) {
Nicolas Dichtelcc3a5722015-05-07 11:02:52 +02001580 nlk->flags |= NETLINK_F_RECV_NO_ENOBUFS;
1581 clear_bit(NETLINK_S_CONGESTED, &nlk->state);
Pablo Neira Ayuso38938bf2009-03-24 16:37:55 -07001582 wake_up_interruptible(&nlk->wait);
Eric Dumazet658cb352012-04-22 21:30:21 +00001583 } else {
Nicolas Dichtelcc3a5722015-05-07 11:02:52 +02001584 nlk->flags &= ~NETLINK_F_RECV_NO_ENOBUFS;
Eric Dumazet658cb352012-04-22 21:30:21 +00001585 }
Pablo Neira Ayuso38938bf2009-03-24 16:37:55 -07001586 err = 0;
1587 break;
Nicolas Dichtel59324cf2015-05-07 11:02:53 +02001588 case NETLINK_LISTEN_ALL_NSID:
1589 if (!ns_capable(sock_net(sk)->user_ns, CAP_NET_BROADCAST))
1590 return -EPERM;
1591
1592 if (val)
1593 nlk->flags |= NETLINK_F_LISTEN_ALL_NSID;
1594 else
1595 nlk->flags &= ~NETLINK_F_LISTEN_ALL_NSID;
1596 err = 0;
1597 break;
Christophe Ricard0a6a3a22015-08-28 07:07:48 +02001598 case NETLINK_CAP_ACK:
1599 if (val)
1600 nlk->flags |= NETLINK_F_CAP_ACK;
1601 else
1602 nlk->flags &= ~NETLINK_F_CAP_ACK;
1603 err = 0;
1604 break;
Patrick McHardy9a4595b2005-08-15 12:32:15 -07001605 default:
1606 err = -ENOPROTOOPT;
1607 }
1608 return err;
1609}
1610
1611static int netlink_getsockopt(struct socket *sock, int level, int optname,
YOSHIFUJI Hideaki746fac42007-02-09 23:25:07 +09001612 char __user *optval, int __user *optlen)
Patrick McHardy9a4595b2005-08-15 12:32:15 -07001613{
1614 struct sock *sk = sock->sk;
1615 struct netlink_sock *nlk = nlk_sk(sk);
1616 int len, val, err;
1617
1618 if (level != SOL_NETLINK)
1619 return -ENOPROTOOPT;
1620
1621 if (get_user(len, optlen))
1622 return -EFAULT;
1623 if (len < 0)
1624 return -EINVAL;
1625
1626 switch (optname) {
1627 case NETLINK_PKTINFO:
1628 if (len < sizeof(int))
1629 return -EINVAL;
1630 len = sizeof(int);
Nicolas Dichtelcc3a5722015-05-07 11:02:52 +02001631 val = nlk->flags & NETLINK_F_RECV_PKTINFO ? 1 : 0;
Heiko Carstensa27b58f2006-10-30 15:06:12 -08001632 if (put_user(len, optlen) ||
1633 put_user(val, optval))
1634 return -EFAULT;
Patrick McHardy9a4595b2005-08-15 12:32:15 -07001635 err = 0;
1636 break;
Pablo Neira Ayusobe0c22a2009-02-18 01:40:43 +00001637 case NETLINK_BROADCAST_ERROR:
1638 if (len < sizeof(int))
1639 return -EINVAL;
1640 len = sizeof(int);
Nicolas Dichtelcc3a5722015-05-07 11:02:52 +02001641 val = nlk->flags & NETLINK_F_BROADCAST_SEND_ERROR ? 1 : 0;
Pablo Neira Ayusobe0c22a2009-02-18 01:40:43 +00001642 if (put_user(len, optlen) ||
1643 put_user(val, optval))
1644 return -EFAULT;
1645 err = 0;
1646 break;
Pablo Neira Ayuso38938bf2009-03-24 16:37:55 -07001647 case NETLINK_NO_ENOBUFS:
1648 if (len < sizeof(int))
1649 return -EINVAL;
1650 len = sizeof(int);
Nicolas Dichtelcc3a5722015-05-07 11:02:52 +02001651 val = nlk->flags & NETLINK_F_RECV_NO_ENOBUFS ? 1 : 0;
Pablo Neira Ayuso38938bf2009-03-24 16:37:55 -07001652 if (put_user(len, optlen) ||
1653 put_user(val, optval))
1654 return -EFAULT;
1655 err = 0;
1656 break;
David Herrmannb42be382015-06-17 17:14:33 +02001657 case NETLINK_LIST_MEMBERSHIPS: {
1658 int pos, idx, shift;
1659
1660 err = 0;
David Herrmann47191d62015-10-21 11:47:43 +02001661 netlink_lock_table();
David Herrmannb42be382015-06-17 17:14:33 +02001662 for (pos = 0; pos * 8 < nlk->ngroups; pos += sizeof(u32)) {
1663 if (len - pos < sizeof(u32))
1664 break;
1665
1666 idx = pos / sizeof(unsigned long);
1667 shift = (pos % sizeof(unsigned long)) * 8;
1668 if (put_user((u32)(nlk->groups[idx] >> shift),
1669 (u32 __user *)(optval + pos))) {
1670 err = -EFAULT;
1671 break;
1672 }
1673 }
1674 if (put_user(ALIGN(nlk->ngroups / 8, sizeof(u32)), optlen))
1675 err = -EFAULT;
David Herrmann47191d62015-10-21 11:47:43 +02001676 netlink_unlock_table();
David Herrmannb42be382015-06-17 17:14:33 +02001677 break;
1678 }
Christophe Ricard0a6a3a22015-08-28 07:07:48 +02001679 case NETLINK_CAP_ACK:
1680 if (len < sizeof(int))
1681 return -EINVAL;
1682 len = sizeof(int);
1683 val = nlk->flags & NETLINK_F_CAP_ACK ? 1 : 0;
1684 if (put_user(len, optlen) ||
1685 put_user(val, optval))
1686 return -EFAULT;
1687 err = 0;
1688 break;
Patrick McHardy9a4595b2005-08-15 12:32:15 -07001689 default:
1690 err = -ENOPROTOOPT;
1691 }
1692 return err;
1693}
1694
1695static void netlink_cmsg_recv_pktinfo(struct msghdr *msg, struct sk_buff *skb)
1696{
1697 struct nl_pktinfo info;
1698
1699 info.group = NETLINK_CB(skb).dst_group;
1700 put_cmsg(msg, SOL_NETLINK, NETLINK_PKTINFO, sizeof(info), &info);
1701}
1702
Nicolas Dichtel59324cf2015-05-07 11:02:53 +02001703static void netlink_cmsg_listen_all_nsid(struct sock *sk, struct msghdr *msg,
1704 struct sk_buff *skb)
1705{
1706 if (!NETLINK_CB(skb).nsid_is_set)
1707 return;
1708
1709 put_cmsg(msg, SOL_NETLINK, NETLINK_LISTEN_ALL_NSID, sizeof(int),
1710 &NETLINK_CB(skb).nsid);
1711}
1712
Ying Xue1b784142015-03-02 15:37:48 +08001713static int netlink_sendmsg(struct socket *sock, struct msghdr *msg, size_t len)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001714{
Linus Torvalds1da177e2005-04-16 15:20:36 -07001715 struct sock *sk = sock->sk;
1716 struct netlink_sock *nlk = nlk_sk(sk);
Steffen Hurrle342dfc32014-01-17 22:53:15 +01001717 DECLARE_SOCKADDR(struct sockaddr_nl *, addr, msg->msg_name);
Eric W. Biederman15e47302012-09-07 20:12:54 +00001718 u32 dst_portid;
Patrick McHardyd629b832005-08-14 19:27:50 -07001719 u32 dst_group;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001720 struct sk_buff *skb;
1721 int err;
1722 struct scm_cookie scm;
Eric W. Biederman2d7a85f2014-05-30 11:04:00 -07001723 u32 netlink_skb_flags = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001724
1725 if (msg->msg_flags&MSG_OOB)
1726 return -EOPNOTSUPP;
1727
Christoph Hellwig7cc05662015-01-28 18:04:53 +01001728 err = scm_send(sock, msg, &scm, true);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001729 if (err < 0)
1730 return err;
1731
1732 if (msg->msg_namelen) {
Eric W. Biedermanb47030c2010-06-13 03:31:06 +00001733 err = -EINVAL;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001734 if (addr->nl_family != AF_NETLINK)
Eric W. Biedermanb47030c2010-06-13 03:31:06 +00001735 goto out;
Eric W. Biederman15e47302012-09-07 20:12:54 +00001736 dst_portid = addr->nl_pid;
Patrick McHardyd629b832005-08-14 19:27:50 -07001737 dst_group = ffs(addr->nl_groups);
Eric W. Biedermanb47030c2010-06-13 03:31:06 +00001738 err = -EPERM;
Eric W. Biederman15e47302012-09-07 20:12:54 +00001739 if ((dst_group || dst_portid) &&
Eric W. Biederman5187cd02014-04-23 14:25:48 -07001740 !netlink_allowed(sock, NL_CFG_F_NONROOT_SEND))
Eric W. Biedermanb47030c2010-06-13 03:31:06 +00001741 goto out;
Eric W. Biederman2d7a85f2014-05-30 11:04:00 -07001742 netlink_skb_flags |= NETLINK_SKB_DST;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001743 } else {
Eric W. Biederman15e47302012-09-07 20:12:54 +00001744 dst_portid = nlk->dst_portid;
Patrick McHardyd629b832005-08-14 19:27:50 -07001745 dst_group = nlk->dst_group;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001746 }
1747
Herbert Xuda314c92015-09-22 11:38:56 +08001748 if (!nlk->bound) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07001749 err = netlink_autobind(sock);
1750 if (err)
1751 goto out;
Herbert Xuda314c92015-09-22 11:38:56 +08001752 } else {
1753 /* Ensure nlk is hashed and visible. */
1754 smp_rmb();
Linus Torvalds1da177e2005-04-16 15:20:36 -07001755 }
1756
1757 err = -EMSGSIZE;
1758 if (len > sk->sk_sndbuf - 32)
1759 goto out;
1760 err = -ENOBUFS;
Pablo Neira3a365152013-06-28 03:04:23 +02001761 skb = netlink_alloc_large_skb(len, dst_group);
Patrick McHardy6ac552f2007-12-04 00:19:38 -08001762 if (skb == NULL)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001763 goto out;
1764
Eric W. Biederman15e47302012-09-07 20:12:54 +00001765 NETLINK_CB(skb).portid = nlk->portid;
Patrick McHardyd629b832005-08-14 19:27:50 -07001766 NETLINK_CB(skb).dst_group = dst_group;
Christoph Hellwig7cc05662015-01-28 18:04:53 +01001767 NETLINK_CB(skb).creds = scm.creds;
Eric W. Biederman2d7a85f2014-05-30 11:04:00 -07001768 NETLINK_CB(skb).flags = netlink_skb_flags;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001769
Linus Torvalds1da177e2005-04-16 15:20:36 -07001770 err = -EFAULT;
Al Viro6ce8e9c2014-04-06 21:25:44 -04001771 if (memcpy_from_msg(skb_put(skb, len), msg, len)) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07001772 kfree_skb(skb);
1773 goto out;
1774 }
1775
1776 err = security_netlink_send(sk, skb);
1777 if (err) {
1778 kfree_skb(skb);
1779 goto out;
1780 }
1781
Patrick McHardyd629b832005-08-14 19:27:50 -07001782 if (dst_group) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07001783 atomic_inc(&skb->users);
Eric W. Biederman15e47302012-09-07 20:12:54 +00001784 netlink_broadcast(sk, skb, dst_portid, dst_group, GFP_KERNEL);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001785 }
Eric W. Biederman15e47302012-09-07 20:12:54 +00001786 err = netlink_unicast(sk, skb, dst_portid, msg->msg_flags&MSG_DONTWAIT);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001787
1788out:
Christoph Hellwig7cc05662015-01-28 18:04:53 +01001789 scm_destroy(&scm);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001790 return err;
1791}
1792
Ying Xue1b784142015-03-02 15:37:48 +08001793static int netlink_recvmsg(struct socket *sock, struct msghdr *msg, size_t len,
Linus Torvalds1da177e2005-04-16 15:20:36 -07001794 int flags)
1795{
Linus Torvalds1da177e2005-04-16 15:20:36 -07001796 struct scm_cookie scm;
1797 struct sock *sk = sock->sk;
1798 struct netlink_sock *nlk = nlk_sk(sk);
1799 int noblock = flags&MSG_DONTWAIT;
1800 size_t copied;
Johannes Berg68d6ac62010-08-15 21:20:44 +00001801 struct sk_buff *skb, *data_skb;
Andrey Vaginb44d2112011-02-21 02:40:47 +00001802 int err, ret;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001803
1804 if (flags&MSG_OOB)
1805 return -EOPNOTSUPP;
1806
1807 copied = 0;
1808
Patrick McHardy6ac552f2007-12-04 00:19:38 -08001809 skb = skb_recv_datagram(sk, flags, noblock, &err);
1810 if (skb == NULL)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001811 goto out;
1812
Johannes Berg68d6ac62010-08-15 21:20:44 +00001813 data_skb = skb;
1814
Johannes Berg1dacc762009-07-01 11:26:02 +00001815#ifdef CONFIG_COMPAT_NETLINK_MESSAGES
1816 if (unlikely(skb_shinfo(skb)->frag_list)) {
Johannes Berg1dacc762009-07-01 11:26:02 +00001817 /*
Johannes Berg68d6ac62010-08-15 21:20:44 +00001818 * If this skb has a frag_list, then here that means that we
1819 * will have to use the frag_list skb's data for compat tasks
1820 * and the regular skb's data for normal (non-compat) tasks.
Johannes Berg1dacc762009-07-01 11:26:02 +00001821 *
Johannes Berg68d6ac62010-08-15 21:20:44 +00001822 * If we need to send the compat skb, assign it to the
1823 * 'data_skb' variable so that it will be used below for data
1824 * copying. We keep 'skb' for everything else, including
1825 * freeing both later.
Johannes Berg1dacc762009-07-01 11:26:02 +00001826 */
Johannes Berg68d6ac62010-08-15 21:20:44 +00001827 if (flags & MSG_CMSG_COMPAT)
1828 data_skb = skb_shinfo(skb)->frag_list;
Johannes Berg1dacc762009-07-01 11:26:02 +00001829 }
1830#endif
1831
Eric Dumazet9063e212014-03-07 12:02:33 -08001832 /* Record the max length of recvmsg() calls for future allocations */
1833 nlk->max_recvmsg_len = max(nlk->max_recvmsg_len, len);
1834 nlk->max_recvmsg_len = min_t(size_t, nlk->max_recvmsg_len,
1835 16384);
1836
Johannes Berg68d6ac62010-08-15 21:20:44 +00001837 copied = data_skb->len;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001838 if (len < copied) {
1839 msg->msg_flags |= MSG_TRUNC;
1840 copied = len;
1841 }
1842
Johannes Berg68d6ac62010-08-15 21:20:44 +00001843 skb_reset_transport_header(data_skb);
David S. Miller51f3d022014-11-05 16:46:40 -05001844 err = skb_copy_datagram_msg(data_skb, 0, msg, copied);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001845
1846 if (msg->msg_name) {
Steffen Hurrle342dfc32014-01-17 22:53:15 +01001847 DECLARE_SOCKADDR(struct sockaddr_nl *, addr, msg->msg_name);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001848 addr->nl_family = AF_NETLINK;
1849 addr->nl_pad = 0;
Eric W. Biederman15e47302012-09-07 20:12:54 +00001850 addr->nl_pid = NETLINK_CB(skb).portid;
Patrick McHardyd629b832005-08-14 19:27:50 -07001851 addr->nl_groups = netlink_group_mask(NETLINK_CB(skb).dst_group);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001852 msg->msg_namelen = sizeof(*addr);
1853 }
1854
Nicolas Dichtelcc3a5722015-05-07 11:02:52 +02001855 if (nlk->flags & NETLINK_F_RECV_PKTINFO)
Patrick McHardycc9a06c2006-03-12 20:34:27 -08001856 netlink_cmsg_recv_pktinfo(msg, skb);
Nicolas Dichtel59324cf2015-05-07 11:02:53 +02001857 if (nlk->flags & NETLINK_F_LISTEN_ALL_NSID)
1858 netlink_cmsg_listen_all_nsid(sk, msg, skb);
Patrick McHardycc9a06c2006-03-12 20:34:27 -08001859
Christoph Hellwig7cc05662015-01-28 18:04:53 +01001860 memset(&scm, 0, sizeof(scm));
1861 scm.creds = *NETLINK_CREDS(skb);
Patrick McHardy188ccb52007-05-03 03:27:01 -07001862 if (flags & MSG_TRUNC)
Johannes Berg68d6ac62010-08-15 21:20:44 +00001863 copied = data_skb->len;
David S. Millerdaa37662010-08-15 23:21:50 -07001864
Linus Torvalds1da177e2005-04-16 15:20:36 -07001865 skb_free_datagram(sk, skb);
1866
Pravin B Shelar16b304f2013-08-15 15:31:06 -07001867 if (nlk->cb_running &&
1868 atomic_read(&sk->sk_rmem_alloc) <= sk->sk_rcvbuf / 2) {
Andrey Vaginb44d2112011-02-21 02:40:47 +00001869 ret = netlink_dump(sk);
1870 if (ret) {
Ben Pfaffac30ef82014-07-09 10:31:22 -07001871 sk->sk_err = -ret;
Andrey Vaginb44d2112011-02-21 02:40:47 +00001872 sk->sk_error_report(sk);
1873 }
1874 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001875
Christoph Hellwig7cc05662015-01-28 18:04:53 +01001876 scm_recv(sock, msg, &scm, flags);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001877out:
1878 netlink_rcv_wake(sk);
1879 return err ? : copied;
1880}
1881
David S. Miller676d2362014-04-11 16:15:36 -04001882static void netlink_data_ready(struct sock *sk)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001883{
Denis V. Lunevcd40b7d2007-10-10 21:15:29 -07001884 BUG();
Linus Torvalds1da177e2005-04-16 15:20:36 -07001885}
1886
1887/*
YOSHIFUJI Hideaki746fac42007-02-09 23:25:07 +09001888 * We export these functions to other modules. They provide a
Linus Torvalds1da177e2005-04-16 15:20:36 -07001889 * complete set of kernel non-blocking support for message
1890 * queueing.
1891 */
1892
1893struct sock *
Pablo Neira Ayuso9f00d972012-09-08 02:53:54 +00001894__netlink_kernel_create(struct net *net, int unit, struct module *module,
1895 struct netlink_kernel_cfg *cfg)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001896{
1897 struct socket *sock;
1898 struct sock *sk;
Patrick McHardy77247bb2005-08-14 19:27:13 -07001899 struct netlink_sock *nlk;
Eric Dumazet5c398dc2010-10-24 04:27:10 +00001900 struct listeners *listeners = NULL;
Pablo Neira Ayusoa31f2d12012-06-29 06:15:21 +00001901 struct mutex *cb_mutex = cfg ? cfg->cb_mutex : NULL;
1902 unsigned int groups;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001903
Akinobu Mitafab2caf2006-08-29 02:15:24 -07001904 BUG_ON(!nl_table);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001905
Patrick McHardy6ac552f2007-12-04 00:19:38 -08001906 if (unit < 0 || unit >= MAX_LINKS)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001907 return NULL;
1908
1909 if (sock_create_lite(PF_NETLINK, SOCK_DGRAM, unit, &sock))
1910 return NULL;
Eric W. Biederman13d30782015-05-08 21:11:33 -05001911
1912 if (__netlink_create(net, sock, cb_mutex, unit, 1) < 0)
Pavel Emelyanov23fe1862008-01-30 19:31:06 -08001913 goto out_sock_release_nosk;
1914
1915 sk = sock->sk;
Harald Welte4fdb3bb2005-08-09 19:40:55 -07001916
Pablo Neira Ayusoa31f2d12012-06-29 06:15:21 +00001917 if (!cfg || cfg->groups < 32)
Patrick McHardy4277a082006-03-20 18:52:01 -08001918 groups = 32;
Pablo Neira Ayusoa31f2d12012-06-29 06:15:21 +00001919 else
1920 groups = cfg->groups;
Patrick McHardy4277a082006-03-20 18:52:01 -08001921
Eric Dumazet5c398dc2010-10-24 04:27:10 +00001922 listeners = kzalloc(sizeof(*listeners) + NLGRPSZ(groups), GFP_KERNEL);
Patrick McHardy4277a082006-03-20 18:52:01 -08001923 if (!listeners)
1924 goto out_sock_release;
1925
Linus Torvalds1da177e2005-04-16 15:20:36 -07001926 sk->sk_data_ready = netlink_data_ready;
Pablo Neira Ayusoa31f2d12012-06-29 06:15:21 +00001927 if (cfg && cfg->input)
1928 nlk_sk(sk)->netlink_rcv = cfg->input;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001929
Herbert Xu8ea65f42015-01-26 14:02:56 +11001930 if (netlink_insert(sk, 0))
Patrick McHardy77247bb2005-08-14 19:27:13 -07001931 goto out_sock_release;
1932
1933 nlk = nlk_sk(sk);
Nicolas Dichtelcc3a5722015-05-07 11:02:52 +02001934 nlk->flags |= NETLINK_F_KERNEL_SOCKET;
Patrick McHardy77247bb2005-08-14 19:27:13 -07001935
1936 netlink_table_grab();
Eric W. Biedermanb4b51022007-09-12 13:05:38 +02001937 if (!nl_table[unit].registered) {
1938 nl_table[unit].groups = groups;
Eric Dumazet5c398dc2010-10-24 04:27:10 +00001939 rcu_assign_pointer(nl_table[unit].listeners, listeners);
Eric W. Biedermanb4b51022007-09-12 13:05:38 +02001940 nl_table[unit].cb_mutex = cb_mutex;
1941 nl_table[unit].module = module;
Pablo Neira Ayuso9785e102012-09-08 02:53:53 +00001942 if (cfg) {
1943 nl_table[unit].bind = cfg->bind;
Hiroaki SHIMODA6251edd2014-11-13 04:24:10 +09001944 nl_table[unit].unbind = cfg->unbind;
Pablo Neira Ayuso9785e102012-09-08 02:53:53 +00001945 nl_table[unit].flags = cfg->flags;
Gao fengda12c902013-06-06 14:49:11 +08001946 if (cfg->compare)
1947 nl_table[unit].compare = cfg->compare;
Pablo Neira Ayuso9785e102012-09-08 02:53:53 +00001948 }
Eric W. Biedermanb4b51022007-09-12 13:05:38 +02001949 nl_table[unit].registered = 1;
Jesper Juhlf937f1f462007-10-15 01:39:12 -07001950 } else {
1951 kfree(listeners);
Denis V. Lunev869e58f2008-01-18 23:53:31 -08001952 nl_table[unit].registered++;
Eric W. Biedermanb4b51022007-09-12 13:05:38 +02001953 }
Patrick McHardy77247bb2005-08-14 19:27:13 -07001954 netlink_table_ungrab();
Harald Welte4fdb3bb2005-08-09 19:40:55 -07001955 return sk;
1956
Harald Welte4fdb3bb2005-08-09 19:40:55 -07001957out_sock_release:
Patrick McHardy4277a082006-03-20 18:52:01 -08001958 kfree(listeners);
Denis V. Lunev9dfbec12008-02-29 11:17:56 -08001959 netlink_kernel_release(sk);
Pavel Emelyanov23fe1862008-01-30 19:31:06 -08001960 return NULL;
1961
1962out_sock_release_nosk:
Harald Welte4fdb3bb2005-08-09 19:40:55 -07001963 sock_release(sock);
Patrick McHardy77247bb2005-08-14 19:27:13 -07001964 return NULL;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001965}
Pablo Neira Ayuso9f00d972012-09-08 02:53:54 +00001966EXPORT_SYMBOL(__netlink_kernel_create);
Denis V. Lunevb7c6ba62008-01-28 14:41:19 -08001967
1968void
1969netlink_kernel_release(struct sock *sk)
1970{
Eric W. Biederman13d30782015-05-08 21:11:33 -05001971 if (sk == NULL || sk->sk_socket == NULL)
1972 return;
1973
1974 sock_release(sk->sk_socket);
Denis V. Lunevb7c6ba62008-01-28 14:41:19 -08001975}
1976EXPORT_SYMBOL(netlink_kernel_release);
1977
Johannes Bergd136f1b2009-09-12 03:03:15 +00001978int __netlink_change_ngroups(struct sock *sk, unsigned int groups)
Johannes Bergb4ff4f02007-07-18 15:46:06 -07001979{
Eric Dumazet5c398dc2010-10-24 04:27:10 +00001980 struct listeners *new, *old;
Johannes Bergb4ff4f02007-07-18 15:46:06 -07001981 struct netlink_table *tbl = &nl_table[sk->sk_protocol];
Johannes Bergb4ff4f02007-07-18 15:46:06 -07001982
1983 if (groups < 32)
1984 groups = 32;
1985
Johannes Bergb4ff4f02007-07-18 15:46:06 -07001986 if (NLGRPSZ(tbl->groups) < NLGRPSZ(groups)) {
Eric Dumazet5c398dc2010-10-24 04:27:10 +00001987 new = kzalloc(sizeof(*new) + NLGRPSZ(groups), GFP_ATOMIC);
1988 if (!new)
Johannes Bergd136f1b2009-09-12 03:03:15 +00001989 return -ENOMEM;
Eric Dumazet6d772ac2012-10-18 03:21:55 +00001990 old = nl_deref_protected(tbl->listeners);
Eric Dumazet5c398dc2010-10-24 04:27:10 +00001991 memcpy(new->masks, old->masks, NLGRPSZ(tbl->groups));
1992 rcu_assign_pointer(tbl->listeners, new);
1993
Lai Jiangshan37b6b932011-03-15 18:01:42 +08001994 kfree_rcu(old, rcu);
Johannes Bergb4ff4f02007-07-18 15:46:06 -07001995 }
1996 tbl->groups = groups;
1997
Johannes Bergd136f1b2009-09-12 03:03:15 +00001998 return 0;
1999}
2000
2001/**
2002 * netlink_change_ngroups - change number of multicast groups
2003 *
2004 * This changes the number of multicast groups that are available
2005 * on a certain netlink family. Note that it is not possible to
2006 * change the number of groups to below 32. Also note that it does
2007 * not implicitly call netlink_clear_multicast_users() when the
2008 * number of groups is reduced.
2009 *
2010 * @sk: The kernel netlink socket, as returned by netlink_kernel_create().
2011 * @groups: The new number of groups.
2012 */
2013int netlink_change_ngroups(struct sock *sk, unsigned int groups)
2014{
2015 int err;
2016
2017 netlink_table_grab();
2018 err = __netlink_change_ngroups(sk, groups);
Johannes Bergb4ff4f02007-07-18 15:46:06 -07002019 netlink_table_ungrab();
Johannes Bergd136f1b2009-09-12 03:03:15 +00002020
Johannes Bergb4ff4f02007-07-18 15:46:06 -07002021 return err;
2022}
Johannes Bergb4ff4f02007-07-18 15:46:06 -07002023
Johannes Bergb8273572009-09-24 15:44:05 -07002024void __netlink_clear_multicast_users(struct sock *ksk, unsigned int group)
2025{
2026 struct sock *sk;
Johannes Bergb8273572009-09-24 15:44:05 -07002027 struct netlink_table *tbl = &nl_table[ksk->sk_protocol];
2028
Sasha Levinb67bfe02013-02-27 17:06:00 -08002029 sk_for_each_bound(sk, &tbl->mc_list)
Johannes Bergb8273572009-09-24 15:44:05 -07002030 netlink_update_socket_mc(nlk_sk(sk), group, 0);
2031}
2032
Denys Vlasenkoa46621a2012-01-30 15:22:06 -05002033struct nlmsghdr *
Eric W. Biederman15e47302012-09-07 20:12:54 +00002034__nlmsg_put(struct sk_buff *skb, u32 portid, u32 seq, int type, int len, int flags)
Denys Vlasenkoa46621a2012-01-30 15:22:06 -05002035{
2036 struct nlmsghdr *nlh;
Hong zhi guo573ce262013-03-27 06:47:04 +00002037 int size = nlmsg_msg_size(len);
Denys Vlasenkoa46621a2012-01-30 15:22:06 -05002038
Wang Yufen23b45672014-02-17 16:53:32 +08002039 nlh = (struct nlmsghdr *)skb_put(skb, NLMSG_ALIGN(size));
Denys Vlasenkoa46621a2012-01-30 15:22:06 -05002040 nlh->nlmsg_type = type;
2041 nlh->nlmsg_len = size;
2042 nlh->nlmsg_flags = flags;
Eric W. Biederman15e47302012-09-07 20:12:54 +00002043 nlh->nlmsg_pid = portid;
Denys Vlasenkoa46621a2012-01-30 15:22:06 -05002044 nlh->nlmsg_seq = seq;
2045 if (!__builtin_constant_p(size) || NLMSG_ALIGN(size) - size != 0)
Hong zhi guo573ce262013-03-27 06:47:04 +00002046 memset(nlmsg_data(nlh) + len, 0, NLMSG_ALIGN(size) - size);
Denys Vlasenkoa46621a2012-01-30 15:22:06 -05002047 return nlh;
2048}
2049EXPORT_SYMBOL(__nlmsg_put);
2050
Linus Torvalds1da177e2005-04-16 15:20:36 -07002051/*
2052 * It looks a bit ugly.
2053 * It would be better to create kernel thread.
2054 */
2055
2056static int netlink_dump(struct sock *sk)
2057{
2058 struct netlink_sock *nlk = nlk_sk(sk);
2059 struct netlink_callback *cb;
Greg Rosec7ac8672011-06-10 01:27:09 +00002060 struct sk_buff *skb = NULL;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002061 struct nlmsghdr *nlh;
Thomas Grafbf8b79e2006-08-04 23:03:29 -07002062 int len, err = -ENOBUFS;
Arad, Ronendb65a3a2015-10-15 01:55:17 -07002063 int alloc_min_size;
Greg Rosec7ac8672011-06-10 01:27:09 +00002064 int alloc_size;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002065
Patrick McHardyaf65bdf2007-04-20 14:14:21 -07002066 mutex_lock(nlk->cb_mutex);
Pravin B Shelar16b304f2013-08-15 15:31:06 -07002067 if (!nlk->cb_running) {
Thomas Grafbf8b79e2006-08-04 23:03:29 -07002068 err = -EINVAL;
2069 goto errout_skb;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002070 }
2071
Florian Westphald1b4c682016-02-18 15:03:24 +01002072 if (atomic_read(&sk->sk_rmem_alloc) >= sk->sk_rcvbuf)
Patrick McHardyf9c22882013-04-17 06:47:04 +00002073 goto errout_skb;
Eric Dumazet9063e212014-03-07 12:02:33 -08002074
2075 /* NLMSG_GOODSIZE is small to avoid high order allocations being
2076 * required, but it makes sense to _attempt_ a 16K bytes allocation
2077 * to reduce number of system calls on dump operations, if user
2078 * ever provided a big enough buffer.
2079 */
Arad, Ronendb65a3a2015-10-15 01:55:17 -07002080 cb = &nlk->cb;
2081 alloc_min_size = max_t(int, cb->min_dump_alloc, NLMSG_GOODSIZE);
2082
2083 if (alloc_min_size < nlk->max_recvmsg_len) {
2084 alloc_size = nlk->max_recvmsg_len;
Florian Westphalc5b0db32016-02-18 15:03:28 +01002085 skb = alloc_skb(alloc_size, GFP_KERNEL |
2086 __GFP_NOWARN | __GFP_NORETRY);
Eric Dumazet9063e212014-03-07 12:02:33 -08002087 }
Arad, Ronendb65a3a2015-10-15 01:55:17 -07002088 if (!skb) {
2089 alloc_size = alloc_min_size;
Florian Westphalc5b0db32016-02-18 15:03:28 +01002090 skb = alloc_skb(alloc_size, GFP_KERNEL);
Arad, Ronendb65a3a2015-10-15 01:55:17 -07002091 }
Greg Rosec7ac8672011-06-10 01:27:09 +00002092 if (!skb)
Dan Carpenterc63d6ea2011-06-15 03:11:42 +00002093 goto errout_skb;
Arad, Ronendb65a3a2015-10-15 01:55:17 -07002094
2095 /* Trim skb to allocated size. User is expected to provide buffer as
2096 * large as max(min_dump_alloc, 16KiB (mac_recvmsg_len capped at
2097 * netlink_recvmsg())). dump will pack as many smaller messages as
2098 * could fit within the allocated skb. skb is typically allocated
2099 * with larger space than required (could be as much as near 2x the
2100 * requested size with align to next power of 2 approach). Allowing
2101 * dump to use the excess space makes it difficult for a user to have a
2102 * reasonable static buffer based on the expected largest dump of a
2103 * single netdev. The outcome is MSG_TRUNC error.
2104 */
Florian Westphald1b4c682016-02-18 15:03:24 +01002105 skb_reserve(skb, skb_tailroom(skb) - alloc_size);
Patrick McHardyf9c22882013-04-17 06:47:04 +00002106 netlink_skb_set_owner_r(skb, sk);
Greg Rosec7ac8672011-06-10 01:27:09 +00002107
Linus Torvalds1da177e2005-04-16 15:20:36 -07002108 len = cb->dump(skb, cb);
2109
2110 if (len > 0) {
Patrick McHardyaf65bdf2007-04-20 14:14:21 -07002111 mutex_unlock(nlk->cb_mutex);
Stephen Hemmingerb1153f22008-03-21 15:46:12 -07002112
2113 if (sk_filter(sk, skb))
2114 kfree_skb(skb);
Eric Dumazet4a7e7c22012-04-05 22:17:46 +00002115 else
2116 __netlink_sendskb(sk, skb);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002117 return 0;
2118 }
2119
Thomas Grafbf8b79e2006-08-04 23:03:29 -07002120 nlh = nlmsg_put_answer(skb, cb, NLMSG_DONE, sizeof(len), NLM_F_MULTI);
2121 if (!nlh)
2122 goto errout_skb;
2123
Johannes Berg670dc282011-06-20 13:40:46 +02002124 nl_dump_check_consistent(cb, nlh);
2125
Thomas Grafbf8b79e2006-08-04 23:03:29 -07002126 memcpy(nlmsg_data(nlh), &len, sizeof(len));
2127
Stephen Hemmingerb1153f22008-03-21 15:46:12 -07002128 if (sk_filter(sk, skb))
2129 kfree_skb(skb);
Eric Dumazet4a7e7c22012-04-05 22:17:46 +00002130 else
2131 __netlink_sendskb(sk, skb);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002132
Thomas Grafa8f74b22005-11-10 02:25:52 +01002133 if (cb->done)
2134 cb->done(cb);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002135
Pravin B Shelar16b304f2013-08-15 15:31:06 -07002136 nlk->cb_running = false;
2137 mutex_unlock(nlk->cb_mutex);
Gao feng6dc878a2012-10-04 20:15:48 +00002138 module_put(cb->module);
Pravin B Shelar16b304f2013-08-15 15:31:06 -07002139 consume_skb(cb->skb);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002140 return 0;
Thomas Graf17977542005-06-18 22:53:48 -07002141
Thomas Grafbf8b79e2006-08-04 23:03:29 -07002142errout_skb:
Patrick McHardyaf65bdf2007-04-20 14:14:21 -07002143 mutex_unlock(nlk->cb_mutex);
Thomas Grafbf8b79e2006-08-04 23:03:29 -07002144 kfree_skb(skb);
Thomas Grafbf8b79e2006-08-04 23:03:29 -07002145 return err;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002146}
2147
Gao feng6dc878a2012-10-04 20:15:48 +00002148int __netlink_dump_start(struct sock *ssk, struct sk_buff *skb,
2149 const struct nlmsghdr *nlh,
2150 struct netlink_dump_control *control)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002151{
2152 struct netlink_callback *cb;
2153 struct sock *sk;
2154 struct netlink_sock *nlk;
Andrey Vaginb44d2112011-02-21 02:40:47 +00002155 int ret;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002156
Florian Westphald1b4c682016-02-18 15:03:24 +01002157 atomic_inc(&skb->users);
Patrick McHardyf9c22882013-04-17 06:47:04 +00002158
Pravin B Shelar16b304f2013-08-15 15:31:06 -07002159 sk = netlink_lookup(sock_net(ssk), ssk->sk_protocol, NETLINK_CB(skb).portid);
2160 if (sk == NULL) {
2161 ret = -ECONNREFUSED;
2162 goto error_free;
2163 }
2164
2165 nlk = nlk_sk(sk);
2166 mutex_lock(nlk->cb_mutex);
2167 /* A dump is in progress... */
2168 if (nlk->cb_running) {
2169 ret = -EBUSY;
2170 goto error_unlock;
2171 }
2172 /* add reference of module which cb->dump belongs to */
2173 if (!try_module_get(control->module)) {
2174 ret = -EPROTONOSUPPORT;
2175 goto error_unlock;
2176 }
2177
2178 cb = &nlk->cb;
2179 memset(cb, 0, sizeof(*cb));
Tom Herbertfc9e50f2015-12-15 15:41:37 -08002180 cb->start = control->start;
Pablo Neira Ayuso80d326f2012-02-24 14:30:15 +00002181 cb->dump = control->dump;
2182 cb->done = control->done;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002183 cb->nlh = nlh;
Pablo Neira Ayuso7175c882012-02-24 14:30:16 +00002184 cb->data = control->data;
Gao feng6dc878a2012-10-04 20:15:48 +00002185 cb->module = control->module;
Pablo Neira Ayuso80d326f2012-02-24 14:30:15 +00002186 cb->min_dump_alloc = control->min_dump_alloc;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002187 cb->skb = skb;
2188
Pravin B Shelar16b304f2013-08-15 15:31:06 -07002189 nlk->cb_running = true;
Gao feng6dc878a2012-10-04 20:15:48 +00002190
Patrick McHardyaf65bdf2007-04-20 14:14:21 -07002191 mutex_unlock(nlk->cb_mutex);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002192
Tom Herbertfc9e50f2015-12-15 15:41:37 -08002193 if (cb->start)
2194 cb->start(cb);
2195
Andrey Vaginb44d2112011-02-21 02:40:47 +00002196 ret = netlink_dump(sk);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002197 sock_put(sk);
Denis V. Lunev5c582982007-10-23 20:29:25 -07002198
Andrey Vaginb44d2112011-02-21 02:40:47 +00002199 if (ret)
2200 return ret;
2201
Denis V. Lunev5c582982007-10-23 20:29:25 -07002202 /* We successfully started a dump, by returning -EINTR we
2203 * signal not to send ACK even if it was requested.
2204 */
2205 return -EINTR;
Pravin B Shelar16b304f2013-08-15 15:31:06 -07002206
2207error_unlock:
2208 sock_put(sk);
2209 mutex_unlock(nlk->cb_mutex);
2210error_free:
2211 kfree_skb(skb);
2212 return ret;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002213}
Gao feng6dc878a2012-10-04 20:15:48 +00002214EXPORT_SYMBOL(__netlink_dump_start);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002215
2216void netlink_ack(struct sk_buff *in_skb, struct nlmsghdr *nlh, int err)
2217{
2218 struct sk_buff *skb;
2219 struct nlmsghdr *rep;
2220 struct nlmsgerr *errmsg;
Thomas Graf339bf982006-11-10 14:10:15 -08002221 size_t payload = sizeof(*errmsg);
Christophe Ricard0a6a3a22015-08-28 07:07:48 +02002222 struct netlink_sock *nlk = nlk_sk(NETLINK_CB(in_skb).sk);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002223
Christophe Ricard0a6a3a22015-08-28 07:07:48 +02002224 /* Error messages get the original request appened, unless the user
2225 * requests to cap the error message.
2226 */
2227 if (!(nlk->flags & NETLINK_F_CAP_ACK) && err)
Thomas Graf339bf982006-11-10 14:10:15 -08002228 payload += nlmsg_len(nlh);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002229
Florian Westphalc5b0db32016-02-18 15:03:28 +01002230 skb = nlmsg_new(payload, GFP_KERNEL);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002231 if (!skb) {
2232 struct sock *sk;
2233
YOSHIFUJI Hideaki3b1e0a62008-03-26 02:26:21 +09002234 sk = netlink_lookup(sock_net(in_skb->sk),
Eric W. Biedermanb4b51022007-09-12 13:05:38 +02002235 in_skb->sk->sk_protocol,
Eric W. Biederman15e47302012-09-07 20:12:54 +00002236 NETLINK_CB(in_skb).portid);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002237 if (sk) {
2238 sk->sk_err = ENOBUFS;
2239 sk->sk_error_report(sk);
2240 sock_put(sk);
2241 }
2242 return;
2243 }
2244
Eric W. Biederman15e47302012-09-07 20:12:54 +00002245 rep = __nlmsg_put(skb, NETLINK_CB(in_skb).portid, nlh->nlmsg_seq,
John Fastabend5dba93a2009-09-25 13:11:44 +00002246 NLMSG_ERROR, payload, 0);
Thomas Grafbf8b79e2006-08-04 23:03:29 -07002247 errmsg = nlmsg_data(rep);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002248 errmsg->error = err;
Christophe Ricard0a6a3a22015-08-28 07:07:48 +02002249 memcpy(&errmsg->msg, nlh, payload > sizeof(*errmsg) ? nlh->nlmsg_len : sizeof(*nlh));
Eric W. Biederman15e47302012-09-07 20:12:54 +00002250 netlink_unicast(in_skb->sk, skb, NETLINK_CB(in_skb).portid, MSG_DONTWAIT);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002251}
Patrick McHardy6ac552f2007-12-04 00:19:38 -08002252EXPORT_SYMBOL(netlink_ack);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002253
Denis V. Lunevcd40b7d2007-10-10 21:15:29 -07002254int netlink_rcv_skb(struct sk_buff *skb, int (*cb)(struct sk_buff *,
Thomas Graf1d00a4e2007-03-22 23:30:12 -07002255 struct nlmsghdr *))
Thomas Graf82ace472005-11-10 02:25:53 +01002256{
Thomas Graf82ace472005-11-10 02:25:53 +01002257 struct nlmsghdr *nlh;
2258 int err;
2259
2260 while (skb->len >= nlmsg_total_size(0)) {
Denis V. Lunevcd40b7d2007-10-10 21:15:29 -07002261 int msglen;
2262
Arnaldo Carvalho de Melob529ccf2007-04-25 19:08:35 -07002263 nlh = nlmsg_hdr(skb);
Thomas Grafd35b6852007-03-22 23:28:46 -07002264 err = 0;
Thomas Graf82ace472005-11-10 02:25:53 +01002265
Martin Murrayad8e4b72006-01-10 13:02:29 -08002266 if (nlh->nlmsg_len < NLMSG_HDRLEN || skb->len < nlh->nlmsg_len)
Thomas Graf82ace472005-11-10 02:25:53 +01002267 return 0;
2268
Thomas Grafd35b6852007-03-22 23:28:46 -07002269 /* Only requests are handled by the kernel */
2270 if (!(nlh->nlmsg_flags & NLM_F_REQUEST))
Denis V. Lunev5c582982007-10-23 20:29:25 -07002271 goto ack;
Thomas Grafd35b6852007-03-22 23:28:46 -07002272
Thomas Graf45e7ae72007-03-22 23:29:10 -07002273 /* Skip control messages */
2274 if (nlh->nlmsg_type < NLMSG_MIN_TYPE)
Denis V. Lunev5c582982007-10-23 20:29:25 -07002275 goto ack;
Thomas Graf45e7ae72007-03-22 23:29:10 -07002276
Thomas Graf1d00a4e2007-03-22 23:30:12 -07002277 err = cb(skb, nlh);
Denis V. Lunev5c582982007-10-23 20:29:25 -07002278 if (err == -EINTR)
2279 goto skip;
2280
2281ack:
Thomas Grafd35b6852007-03-22 23:28:46 -07002282 if (nlh->nlmsg_flags & NLM_F_ACK || err)
Thomas Graf82ace472005-11-10 02:25:53 +01002283 netlink_ack(skb, nlh, err);
Thomas Graf82ace472005-11-10 02:25:53 +01002284
Denis V. Lunev5c582982007-10-23 20:29:25 -07002285skip:
Patrick McHardy6ac552f2007-12-04 00:19:38 -08002286 msglen = NLMSG_ALIGN(nlh->nlmsg_len);
Denis V. Lunevcd40b7d2007-10-10 21:15:29 -07002287 if (msglen > skb->len)
2288 msglen = skb->len;
2289 skb_pull(skb, msglen);
Thomas Graf82ace472005-11-10 02:25:53 +01002290 }
2291
2292 return 0;
2293}
Patrick McHardy6ac552f2007-12-04 00:19:38 -08002294EXPORT_SYMBOL(netlink_rcv_skb);
Thomas Graf82ace472005-11-10 02:25:53 +01002295
2296/**
Thomas Grafd387f6a2006-08-15 00:31:06 -07002297 * nlmsg_notify - send a notification netlink message
2298 * @sk: netlink socket to use
2299 * @skb: notification message
Eric W. Biederman15e47302012-09-07 20:12:54 +00002300 * @portid: destination netlink portid for reports or 0
Thomas Grafd387f6a2006-08-15 00:31:06 -07002301 * @group: destination multicast group or 0
2302 * @report: 1 to report back, 0 to disable
2303 * @flags: allocation flags
2304 */
Eric W. Biederman15e47302012-09-07 20:12:54 +00002305int nlmsg_notify(struct sock *sk, struct sk_buff *skb, u32 portid,
Thomas Grafd387f6a2006-08-15 00:31:06 -07002306 unsigned int group, int report, gfp_t flags)
2307{
2308 int err = 0;
2309
2310 if (group) {
Eric W. Biederman15e47302012-09-07 20:12:54 +00002311 int exclude_portid = 0;
Thomas Grafd387f6a2006-08-15 00:31:06 -07002312
2313 if (report) {
2314 atomic_inc(&skb->users);
Eric W. Biederman15e47302012-09-07 20:12:54 +00002315 exclude_portid = portid;
Thomas Grafd387f6a2006-08-15 00:31:06 -07002316 }
2317
Pablo Neira Ayuso1ce85fe2009-02-24 23:18:28 -08002318 /* errors reported via destination sk->sk_err, but propagate
2319 * delivery errors if NETLINK_BROADCAST_ERROR flag is set */
Eric W. Biederman15e47302012-09-07 20:12:54 +00002320 err = nlmsg_multicast(sk, skb, exclude_portid, group, flags);
Thomas Grafd387f6a2006-08-15 00:31:06 -07002321 }
2322
Pablo Neira Ayuso1ce85fe2009-02-24 23:18:28 -08002323 if (report) {
2324 int err2;
2325
Eric W. Biederman15e47302012-09-07 20:12:54 +00002326 err2 = nlmsg_unicast(sk, skb, portid);
Pablo Neira Ayuso1ce85fe2009-02-24 23:18:28 -08002327 if (!err || err == -ESRCH)
2328 err = err2;
2329 }
Thomas Grafd387f6a2006-08-15 00:31:06 -07002330
2331 return err;
2332}
Patrick McHardy6ac552f2007-12-04 00:19:38 -08002333EXPORT_SYMBOL(nlmsg_notify);
Thomas Grafd387f6a2006-08-15 00:31:06 -07002334
Linus Torvalds1da177e2005-04-16 15:20:36 -07002335#ifdef CONFIG_PROC_FS
2336struct nl_seq_iter {
Denis V. Luneve372c412007-11-19 22:31:54 -08002337 struct seq_net_private p;
Herbert Xu56d28b12015-02-04 07:33:24 +11002338 struct rhashtable_iter hti;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002339 int link;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002340};
2341
Herbert Xu56d28b12015-02-04 07:33:24 +11002342static int netlink_walk_start(struct nl_seq_iter *iter)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002343{
Herbert Xu56d28b12015-02-04 07:33:24 +11002344 int err;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002345
Herbert Xu56d28b12015-02-04 07:33:24 +11002346 err = rhashtable_walk_init(&nl_table[iter->link].hash, &iter->hti);
2347 if (err) {
2348 iter->link = MAX_LINKS;
2349 return err;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002350 }
Herbert Xu56d28b12015-02-04 07:33:24 +11002351
2352 err = rhashtable_walk_start(&iter->hti);
2353 return err == -EAGAIN ? 0 : err;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002354}
2355
Herbert Xu56d28b12015-02-04 07:33:24 +11002356static void netlink_walk_stop(struct nl_seq_iter *iter)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002357{
Herbert Xu56d28b12015-02-04 07:33:24 +11002358 rhashtable_walk_stop(&iter->hti);
2359 rhashtable_walk_exit(&iter->hti);
2360}
2361
2362static void *__netlink_seq_next(struct seq_file *seq)
2363{
2364 struct nl_seq_iter *iter = seq->private;
2365 struct netlink_sock *nlk;
2366
2367 do {
2368 for (;;) {
2369 int err;
2370
2371 nlk = rhashtable_walk_next(&iter->hti);
2372
2373 if (IS_ERR(nlk)) {
2374 if (PTR_ERR(nlk) == -EAGAIN)
2375 continue;
2376
2377 return nlk;
2378 }
2379
2380 if (nlk)
2381 break;
2382
2383 netlink_walk_stop(iter);
2384 if (++iter->link >= MAX_LINKS)
2385 return NULL;
2386
2387 err = netlink_walk_start(iter);
2388 if (err)
2389 return ERR_PTR(err);
2390 }
2391 } while (sock_net(&nlk->sk) != seq_file_net(seq));
2392
2393 return nlk;
2394}
2395
2396static void *netlink_seq_start(struct seq_file *seq, loff_t *posp)
2397{
2398 struct nl_seq_iter *iter = seq->private;
2399 void *obj = SEQ_START_TOKEN;
2400 loff_t pos;
2401 int err;
2402
2403 iter->link = 0;
2404
2405 err = netlink_walk_start(iter);
2406 if (err)
2407 return ERR_PTR(err);
2408
2409 for (pos = *posp; pos && obj && !IS_ERR(obj); pos--)
2410 obj = __netlink_seq_next(seq);
2411
2412 return obj;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002413}
2414
2415static void *netlink_seq_next(struct seq_file *seq, void *v, loff_t *pos)
2416{
Linus Torvalds1da177e2005-04-16 15:20:36 -07002417 ++*pos;
Herbert Xu56d28b12015-02-04 07:33:24 +11002418 return __netlink_seq_next(seq);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002419}
2420
2421static void netlink_seq_stop(struct seq_file *seq, void *v)
2422{
Herbert Xu56d28b12015-02-04 07:33:24 +11002423 struct nl_seq_iter *iter = seq->private;
2424
2425 if (iter->link >= MAX_LINKS)
2426 return;
2427
2428 netlink_walk_stop(iter);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002429}
2430
2431
2432static int netlink_seq_show(struct seq_file *seq, void *v)
2433{
Eric Dumazet658cb352012-04-22 21:30:21 +00002434 if (v == SEQ_START_TOKEN) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07002435 seq_puts(seq,
2436 "sk Eth Pid Groups "
Masatake YAMATOcf0aa4e2010-02-27 19:45:37 +00002437 "Rmem Wmem Dump Locks Drops Inode\n");
Eric Dumazet658cb352012-04-22 21:30:21 +00002438 } else {
Linus Torvalds1da177e2005-04-16 15:20:36 -07002439 struct sock *s = v;
2440 struct netlink_sock *nlk = nlk_sk(s);
2441
Pravin B Shelar16b304f2013-08-15 15:31:06 -07002442 seq_printf(seq, "%pK %-3d %-6u %08x %-8d %-8d %d %-8d %-8d %-8lu\n",
Linus Torvalds1da177e2005-04-16 15:20:36 -07002443 s,
2444 s->sk_protocol,
Eric W. Biederman15e47302012-09-07 20:12:54 +00002445 nlk->portid,
Patrick McHardy513c2502005-09-06 15:43:59 -07002446 nlk->groups ? (u32)nlk->groups[0] : 0,
Eric Dumazet31e6d362009-06-17 19:05:41 -07002447 sk_rmem_alloc_get(s),
2448 sk_wmem_alloc_get(s),
Pravin B Shelar16b304f2013-08-15 15:31:06 -07002449 nlk->cb_running,
Pablo Neira Ayuso38938bf2009-03-24 16:37:55 -07002450 atomic_read(&s->sk_refcnt),
Masatake YAMATOcf0aa4e2010-02-27 19:45:37 +00002451 atomic_read(&s->sk_drops),
2452 sock_i_ino(s)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002453 );
2454
2455 }
2456 return 0;
2457}
2458
Philippe De Muyter56b3d972007-07-10 23:07:31 -07002459static const struct seq_operations netlink_seq_ops = {
Linus Torvalds1da177e2005-04-16 15:20:36 -07002460 .start = netlink_seq_start,
2461 .next = netlink_seq_next,
2462 .stop = netlink_seq_stop,
2463 .show = netlink_seq_show,
2464};
2465
2466
2467static int netlink_seq_open(struct inode *inode, struct file *file)
2468{
Denis V. Luneve372c412007-11-19 22:31:54 -08002469 return seq_open_net(inode, file, &netlink_seq_ops,
2470 sizeof(struct nl_seq_iter));
Eric W. Biedermanb4b51022007-09-12 13:05:38 +02002471}
2472
Arjan van de Venda7071d2007-02-12 00:55:36 -08002473static const struct file_operations netlink_seq_fops = {
Linus Torvalds1da177e2005-04-16 15:20:36 -07002474 .owner = THIS_MODULE,
2475 .open = netlink_seq_open,
2476 .read = seq_read,
2477 .llseek = seq_lseek,
Denis V. Luneve372c412007-11-19 22:31:54 -08002478 .release = seq_release_net,
Linus Torvalds1da177e2005-04-16 15:20:36 -07002479};
2480
2481#endif
2482
2483int netlink_register_notifier(struct notifier_block *nb)
2484{
Alan Sterne041c682006-03-27 01:16:30 -08002485 return atomic_notifier_chain_register(&netlink_chain, nb);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002486}
Patrick McHardy6ac552f2007-12-04 00:19:38 -08002487EXPORT_SYMBOL(netlink_register_notifier);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002488
2489int netlink_unregister_notifier(struct notifier_block *nb)
2490{
Alan Sterne041c682006-03-27 01:16:30 -08002491 return atomic_notifier_chain_unregister(&netlink_chain, nb);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002492}
Patrick McHardy6ac552f2007-12-04 00:19:38 -08002493EXPORT_SYMBOL(netlink_unregister_notifier);
YOSHIFUJI Hideaki746fac42007-02-09 23:25:07 +09002494
Eric Dumazet90ddc4f2005-12-22 12:49:22 -08002495static const struct proto_ops netlink_ops = {
Linus Torvalds1da177e2005-04-16 15:20:36 -07002496 .family = PF_NETLINK,
2497 .owner = THIS_MODULE,
2498 .release = netlink_release,
2499 .bind = netlink_bind,
2500 .connect = netlink_connect,
2501 .socketpair = sock_no_socketpair,
2502 .accept = sock_no_accept,
2503 .getname = netlink_getname,
Florian Westphald1b4c682016-02-18 15:03:24 +01002504 .poll = datagram_poll,
David Decotigny025c6812016-03-21 10:15:35 -07002505 .ioctl = netlink_ioctl,
Linus Torvalds1da177e2005-04-16 15:20:36 -07002506 .listen = sock_no_listen,
2507 .shutdown = sock_no_shutdown,
Patrick McHardy9a4595b2005-08-15 12:32:15 -07002508 .setsockopt = netlink_setsockopt,
2509 .getsockopt = netlink_getsockopt,
Linus Torvalds1da177e2005-04-16 15:20:36 -07002510 .sendmsg = netlink_sendmsg,
2511 .recvmsg = netlink_recvmsg,
Florian Westphald1b4c682016-02-18 15:03:24 +01002512 .mmap = sock_no_mmap,
Linus Torvalds1da177e2005-04-16 15:20:36 -07002513 .sendpage = sock_no_sendpage,
2514};
2515
Stephen Hemmingerec1b4cf2009-10-05 05:58:39 +00002516static const struct net_proto_family netlink_family_ops = {
Linus Torvalds1da177e2005-04-16 15:20:36 -07002517 .family = PF_NETLINK,
2518 .create = netlink_create,
2519 .owner = THIS_MODULE, /* for consistency 8) */
2520};
2521
Pavel Emelyanov46650792007-10-08 20:38:39 -07002522static int __net_init netlink_net_init(struct net *net)
Eric W. Biedermanb4b51022007-09-12 13:05:38 +02002523{
2524#ifdef CONFIG_PROC_FS
Gao fengd4beaa62013-02-18 01:34:54 +00002525 if (!proc_create("netlink", 0, net->proc_net, &netlink_seq_fops))
Eric W. Biedermanb4b51022007-09-12 13:05:38 +02002526 return -ENOMEM;
2527#endif
2528 return 0;
2529}
2530
Pavel Emelyanov46650792007-10-08 20:38:39 -07002531static void __net_exit netlink_net_exit(struct net *net)
Eric W. Biedermanb4b51022007-09-12 13:05:38 +02002532{
2533#ifdef CONFIG_PROC_FS
Gao fengece31ff2013-02-18 01:34:56 +00002534 remove_proc_entry("netlink", net->proc_net);
Eric W. Biedermanb4b51022007-09-12 13:05:38 +02002535#endif
2536}
2537
David S. Millerb963ea82010-08-30 19:08:01 -07002538static void __init netlink_add_usersock_entry(void)
2539{
Eric Dumazet5c398dc2010-10-24 04:27:10 +00002540 struct listeners *listeners;
David S. Millerb963ea82010-08-30 19:08:01 -07002541 int groups = 32;
2542
Eric Dumazet5c398dc2010-10-24 04:27:10 +00002543 listeners = kzalloc(sizeof(*listeners) + NLGRPSZ(groups), GFP_KERNEL);
David S. Millerb963ea82010-08-30 19:08:01 -07002544 if (!listeners)
Eric Dumazet5c398dc2010-10-24 04:27:10 +00002545 panic("netlink_add_usersock_entry: Cannot allocate listeners\n");
David S. Millerb963ea82010-08-30 19:08:01 -07002546
2547 netlink_table_grab();
2548
2549 nl_table[NETLINK_USERSOCK].groups = groups;
Eric Dumazet5c398dc2010-10-24 04:27:10 +00002550 rcu_assign_pointer(nl_table[NETLINK_USERSOCK].listeners, listeners);
David S. Millerb963ea82010-08-30 19:08:01 -07002551 nl_table[NETLINK_USERSOCK].module = THIS_MODULE;
2552 nl_table[NETLINK_USERSOCK].registered = 1;
Pablo Neira Ayuso9785e102012-09-08 02:53:53 +00002553 nl_table[NETLINK_USERSOCK].flags = NL_CFG_F_NONROOT_SEND;
David S. Millerb963ea82010-08-30 19:08:01 -07002554
2555 netlink_table_ungrab();
2556}
2557
Denis V. Lunev022cbae2007-11-13 03:23:50 -08002558static struct pernet_operations __net_initdata netlink_net_ops = {
Eric W. Biedermanb4b51022007-09-12 13:05:38 +02002559 .init = netlink_net_init,
2560 .exit = netlink_net_exit,
2561};
2562
Patrick McHardy49f7b332015-03-25 13:07:45 +00002563static inline u32 netlink_hash(const void *data, u32 len, u32 seed)
Herbert Xuc428ecd2015-03-20 21:57:01 +11002564{
2565 const struct netlink_sock *nlk = data;
2566 struct netlink_compare_arg arg;
2567
Herbert Xuda314c92015-09-22 11:38:56 +08002568 netlink_compare_arg_init(&arg, sock_net(&nlk->sk), nlk->portid);
Herbert Xu11b58ba2015-03-24 00:50:22 +11002569 return jhash2((u32 *)&arg, netlink_compare_arg_len / sizeof(u32), seed);
Herbert Xuc428ecd2015-03-20 21:57:01 +11002570}
2571
2572static const struct rhashtable_params netlink_rhashtable_params = {
2573 .head_offset = offsetof(struct netlink_sock, node),
2574 .key_len = netlink_compare_arg_len,
Herbert Xuc428ecd2015-03-20 21:57:01 +11002575 .obj_hashfn = netlink_hash,
2576 .obj_cmpfn = netlink_compare,
Thomas Grafb5e2c152015-03-24 20:42:19 +00002577 .automatic_shrinking = true,
Herbert Xuc428ecd2015-03-20 21:57:01 +11002578};
2579
Linus Torvalds1da177e2005-04-16 15:20:36 -07002580static int __init netlink_proto_init(void)
2581{
Linus Torvalds1da177e2005-04-16 15:20:36 -07002582 int i;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002583 int err = proto_register(&netlink_proto, 0);
2584
2585 if (err != 0)
2586 goto out;
2587
YOSHIFUJI Hideaki / 吉藤英明fab25742013-01-09 07:19:48 +00002588 BUILD_BUG_ON(sizeof(struct netlink_skb_parms) > FIELD_SIZEOF(struct sk_buff, cb));
Linus Torvalds1da177e2005-04-16 15:20:36 -07002589
Panagiotis Issaris0da974f2006-07-21 14:51:30 -07002590 nl_table = kcalloc(MAX_LINKS, sizeof(*nl_table), GFP_KERNEL);
Akinobu Mitafab2caf2006-08-29 02:15:24 -07002591 if (!nl_table)
2592 goto panic;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002593
Linus Torvalds1da177e2005-04-16 15:20:36 -07002594 for (i = 0; i < MAX_LINKS; i++) {
Herbert Xuc428ecd2015-03-20 21:57:01 +11002595 if (rhashtable_init(&nl_table[i].hash,
2596 &netlink_rhashtable_params) < 0) {
Thomas Grafe3416942014-08-02 11:47:45 +02002597 while (--i > 0)
2598 rhashtable_destroy(&nl_table[i].hash);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002599 kfree(nl_table);
Akinobu Mitafab2caf2006-08-29 02:15:24 -07002600 goto panic;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002601 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07002602 }
2603
Daniel Borkmannbcbde0d2013-06-21 19:38:07 +02002604 INIT_LIST_HEAD(&netlink_tap_all);
2605
David S. Millerb963ea82010-08-30 19:08:01 -07002606 netlink_add_usersock_entry();
2607
Linus Torvalds1da177e2005-04-16 15:20:36 -07002608 sock_register(&netlink_family_ops);
Eric W. Biedermanb4b51022007-09-12 13:05:38 +02002609 register_pernet_subsys(&netlink_net_ops);
YOSHIFUJI Hideaki746fac42007-02-09 23:25:07 +09002610 /* The netlink device handler may be needed early. */
Linus Torvalds1da177e2005-04-16 15:20:36 -07002611 rtnetlink_init();
2612out:
2613 return err;
Akinobu Mitafab2caf2006-08-29 02:15:24 -07002614panic:
2615 panic("netlink_init: Cannot allocate nl_table\n");
Linus Torvalds1da177e2005-04-16 15:20:36 -07002616}
2617
Linus Torvalds1da177e2005-04-16 15:20:36 -07002618core_initcall(netlink_proto_init);