Loading...
1/*
2 * NETLINK Kernel-user communication protocol.
3 *
4 * Authors: Alan Cox <alan@lxorguk.ukuu.org.uk>
5 * Alexey Kuznetsov <kuznet@ms2.inr.ac.ru>
6 * Patrick McHardy <kaber@trash.net>
7 *
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.
12 *
13 * Tue Jun 26 14:36:48 MEST 2001 Herbert "herp" Rosmanith
14 * added netlink_proto_exit
15 * Tue Jan 22 18:32:44 BRST 2002 Arnaldo C. de Melo <acme@conectiva.com.br>
16 * use nlk_sk, as sk->protinfo is on a diet 8)
17 * Fri Jul 22 19:51:12 MEST 2005 Harald Welte <laforge@gnumonks.org>
18 * - inc module use count of module that owns
19 * the kernel socket in case userspace opens
20 * socket of same protocol
21 * - remove all module support, since netlink is
22 * mandatory if CONFIG_NET=y these days
23 */
24
25#include <linux/module.h>
26
27#include <linux/capability.h>
28#include <linux/kernel.h>
29#include <linux/init.h>
30#include <linux/signal.h>
31#include <linux/sched.h>
32#include <linux/errno.h>
33#include <linux/string.h>
34#include <linux/stat.h>
35#include <linux/socket.h>
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 <linux/uaccess.h>
44#include <linux/skbuff.h>
45#include <linux/netdevice.h>
46#include <linux/rtnetlink.h>
47#include <linux/proc_fs.h>
48#include <linux/seq_file.h>
49#include <linux/notifier.h>
50#include <linux/security.h>
51#include <linux/jhash.h>
52#include <linux/jiffies.h>
53#include <linux/random.h>
54#include <linux/bitops.h>
55#include <linux/mm.h>
56#include <linux/types.h>
57#include <linux/audit.h>
58#include <linux/mutex.h>
59#include <linux/vmalloc.h>
60#include <linux/if_arp.h>
61#include <linux/rhashtable.h>
62#include <asm/cacheflush.h>
63#include <linux/hash.h>
64#include <linux/genetlink.h>
65
66#include <net/net_namespace.h>
67#include <net/sock.h>
68#include <net/scm.h>
69#include <net/netlink.h>
70
71#include "af_netlink.h"
72
73struct listeners {
74 struct rcu_head rcu;
75 unsigned long masks[0];
76};
77
78/* state bits */
79#define NETLINK_S_CONGESTED 0x0
80
81/* flags */
82#define NETLINK_F_KERNEL_SOCKET 0x1
83#define NETLINK_F_RECV_PKTINFO 0x2
84#define NETLINK_F_BROADCAST_SEND_ERROR 0x4
85#define NETLINK_F_RECV_NO_ENOBUFS 0x8
86#define NETLINK_F_LISTEN_ALL_NSID 0x10
87#define NETLINK_F_CAP_ACK 0x20
88
89static inline int netlink_is_kernel(struct sock *sk)
90{
91 return nlk_sk(sk)->flags & NETLINK_F_KERNEL_SOCKET;
92}
93
94struct netlink_table *nl_table __read_mostly;
95EXPORT_SYMBOL_GPL(nl_table);
96
97static DECLARE_WAIT_QUEUE_HEAD(nl_table_wait);
98
99static int netlink_dump(struct sock *sk);
100static void netlink_skb_destructor(struct sk_buff *skb);
101
102/* nl_table locking explained:
103 * Lookup and traversal are protected with an RCU read-side lock. Insertion
104 * and removal are protected with per bucket lock while using RCU list
105 * modification primitives and may run in parallel to RCU protected lookups.
106 * Destruction of the Netlink socket may only occur *after* nl_table_lock has
107 * been acquired * either during or after the socket has been removed from
108 * the list and after an RCU grace period.
109 */
110DEFINE_RWLOCK(nl_table_lock);
111EXPORT_SYMBOL_GPL(nl_table_lock);
112static atomic_t nl_table_users = ATOMIC_INIT(0);
113
114#define nl_deref_protected(X) rcu_dereference_protected(X, lockdep_is_held(&nl_table_lock));
115
116static BLOCKING_NOTIFIER_HEAD(netlink_chain);
117
118static DEFINE_SPINLOCK(netlink_tap_lock);
119static struct list_head netlink_tap_all __read_mostly;
120
121static const struct rhashtable_params netlink_rhashtable_params;
122
123static inline u32 netlink_group_mask(u32 group)
124{
125 return group ? 1 << (group - 1) : 0;
126}
127
128static 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
146int 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
155 __module_get(nt->module);
156
157 return 0;
158}
159EXPORT_SYMBOL_GPL(netlink_add_tap);
160
161static int __netlink_remove_tap(struct netlink_tap *nt)
162{
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
180 if (found)
181 module_put(nt->module);
182
183 return found ? 0 : -ENODEV;
184}
185
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
197static bool netlink_filter_tap(const struct sk_buff *skb)
198{
199 struct sock *sk = skb->sk;
200
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:
213 return true;
214 }
215
216 return false;
217}
218
219static int __netlink_deliver_tap_skb(struct sk_buff *skb,
220 struct net_device *dev)
221{
222 struct sk_buff *nskb;
223 struct sock *sk = skb->sk;
224 int ret = -ENOMEM;
225
226 dev_hold(dev);
227
228 if (is_vmalloc_addr(skb->head))
229 nskb = netlink_to_full_skb(skb, GFP_ATOMIC);
230 else
231 nskb = skb_clone(skb, GFP_ATOMIC);
232 if (nskb) {
233 nskb->dev = dev;
234 nskb->protocol = htons((u16) sk->sk_protocol);
235 nskb->pkt_type = netlink_is_kernel(sk) ?
236 PACKET_KERNEL : PACKET_USER;
237 skb_reset_network_header(nskb);
238 ret = dev_queue_xmit(nskb);
239 if (unlikely(ret > 0))
240 ret = net_xmit_errno(ret);
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
252 if (!netlink_filter_tap(skb))
253 return;
254
255 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
272static 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
279static void netlink_overrun(struct sock *sk)
280{
281 struct netlink_sock *nlk = nlk_sk(sk);
282
283 if (!(nlk->flags & NETLINK_F_RECV_NO_ENOBUFS)) {
284 if (!test_and_set_bit(NETLINK_S_CONGESTED,
285 &nlk_sk(sk)->state)) {
286 sk->sk_err = ENOBUFS;
287 sk->sk_error_report(sk);
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))
298 clear_bit(NETLINK_S_CONGESTED, &nlk->state);
299 if (!test_bit(NETLINK_S_CONGESTED, &nlk->state))
300 wake_up_interruptible(&nlk->wait);
301}
302
303static void netlink_skb_destructor(struct sk_buff *skb)
304{
305 if (is_vmalloc_addr(skb->head)) {
306 if (!skb->cloned ||
307 !atomic_dec_return(&(skb_shinfo(skb)->dataref)))
308 vfree(skb->head);
309
310 skb->head = NULL;
311 }
312 if (skb->sk != NULL)
313 sock_rfree(skb);
314}
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
325static void netlink_sock_destruct(struct sock *sk)
326{
327 struct netlink_sock *nlk = nlk_sk(sk);
328
329 if (nlk->cb_running) {
330 if (nlk->cb.done)
331 nlk->cb.done(&nlk->cb);
332 module_put(nlk->cb.module);
333 kfree_skb(nlk->cb.skb);
334 }
335
336 skb_queue_purge(&sk->sk_receive_queue);
337
338 if (!sock_flag(sk, SOCK_DEAD)) {
339 printk(KERN_ERR "Freeing alive netlink socket %p\n", sk);
340 return;
341 }
342
343 WARN_ON(atomic_read(&sk->sk_rmem_alloc));
344 WARN_ON(atomic_read(&sk->sk_wmem_alloc));
345 WARN_ON(nlk_sk(sk)->groups);
346}
347
348static void netlink_sock_destruct_work(struct work_struct *work)
349{
350 struct netlink_sock *nlk = container_of(work, struct netlink_sock,
351 work);
352
353 sk_free(&nlk->sk);
354}
355
356/* This lock without WQ_FLAG_EXCLUSIVE is good on UP and it is _very_ bad on
357 * SMP. Look, when several writers sleep and reader wakes them up, all but one
358 * immediately hit write lock and grab all the cpus. Exclusive sleep solves
359 * this, _but_ remember, it adds useless work on UP machines.
360 */
361
362void netlink_table_grab(void)
363 __acquires(nl_table_lock)
364{
365 might_sleep();
366
367 write_lock_irq(&nl_table_lock);
368
369 if (atomic_read(&nl_table_users)) {
370 DECLARE_WAITQUEUE(wait, current);
371
372 add_wait_queue_exclusive(&nl_table_wait, &wait);
373 for (;;) {
374 set_current_state(TASK_UNINTERRUPTIBLE);
375 if (atomic_read(&nl_table_users) == 0)
376 break;
377 write_unlock_irq(&nl_table_lock);
378 schedule();
379 write_lock_irq(&nl_table_lock);
380 }
381
382 __set_current_state(TASK_RUNNING);
383 remove_wait_queue(&nl_table_wait, &wait);
384 }
385}
386
387void netlink_table_ungrab(void)
388 __releases(nl_table_lock)
389{
390 write_unlock_irq(&nl_table_lock);
391 wake_up(&nl_table_wait);
392}
393
394static inline void
395netlink_lock_table(void)
396{
397 /* read_lock() synchronizes us to netlink_table_grab */
398
399 read_lock(&nl_table_lock);
400 atomic_inc(&nl_table_users);
401 read_unlock(&nl_table_lock);
402}
403
404static inline void
405netlink_unlock_table(void)
406{
407 if (atomic_dec_and_test(&nl_table_users))
408 wake_up(&nl_table_wait);
409}
410
411struct netlink_compare_arg
412{
413 possible_net_t pnet;
414 u32 portid;
415};
416
417/* Doing sizeof directly may yield 4 extra bytes on 64-bit. */
418#define netlink_compare_arg_len \
419 (offsetof(struct netlink_compare_arg, portid) + sizeof(u32))
420
421static inline int netlink_compare(struct rhashtable_compare_arg *arg,
422 const void *ptr)
423{
424 const struct netlink_compare_arg *x = arg->key;
425 const struct netlink_sock *nlk = ptr;
426
427 return nlk->portid != x->portid ||
428 !net_eq(sock_net(&nlk->sk), read_pnet(&x->pnet));
429}
430
431static void netlink_compare_arg_init(struct netlink_compare_arg *arg,
432 struct net *net, u32 portid)
433{
434 memset(arg, 0, sizeof(*arg));
435 write_pnet(&arg->pnet, net);
436 arg->portid = portid;
437}
438
439static struct sock *__netlink_lookup(struct netlink_table *table, u32 portid,
440 struct net *net)
441{
442 struct netlink_compare_arg arg;
443
444 netlink_compare_arg_init(&arg, net, portid);
445 return rhashtable_lookup_fast(&table->hash, &arg,
446 netlink_rhashtable_params);
447}
448
449static int __netlink_insert(struct netlink_table *table, struct sock *sk)
450{
451 struct netlink_compare_arg arg;
452
453 netlink_compare_arg_init(&arg, sock_net(sk), nlk_sk(sk)->portid);
454 return rhashtable_lookup_insert_key(&table->hash, &arg,
455 &nlk_sk(sk)->node,
456 netlink_rhashtable_params);
457}
458
459static struct sock *netlink_lookup(struct net *net, int protocol, u32 portid)
460{
461 struct netlink_table *table = &nl_table[protocol];
462 struct sock *sk;
463
464 rcu_read_lock();
465 sk = __netlink_lookup(table, portid, net);
466 if (sk)
467 sock_hold(sk);
468 rcu_read_unlock();
469
470 return sk;
471}
472
473static const struct proto_ops netlink_ops;
474
475static void
476netlink_update_listeners(struct sock *sk)
477{
478 struct netlink_table *tbl = &nl_table[sk->sk_protocol];
479 unsigned long mask;
480 unsigned int i;
481 struct listeners *listeners;
482
483 listeners = nl_deref_protected(tbl->listeners);
484 if (!listeners)
485 return;
486
487 for (i = 0; i < NLGRPLONGS(tbl->groups); i++) {
488 mask = 0;
489 sk_for_each_bound(sk, &tbl->mc_list) {
490 if (i < NLGRPLONGS(nlk_sk(sk)->ngroups))
491 mask |= nlk_sk(sk)->groups[i];
492 }
493 listeners->masks[i] = mask;
494 }
495 /* this function is only called with the netlink table "grabbed", which
496 * makes sure updates are visible before bind or setsockopt return. */
497}
498
499static int netlink_insert(struct sock *sk, u32 portid)
500{
501 struct netlink_table *table = &nl_table[sk->sk_protocol];
502 int err;
503
504 lock_sock(sk);
505
506 err = nlk_sk(sk)->portid == portid ? 0 : -EBUSY;
507 if (nlk_sk(sk)->bound)
508 goto err;
509
510 err = -ENOMEM;
511 if (BITS_PER_LONG > 32 &&
512 unlikely(atomic_read(&table->hash.nelems) >= UINT_MAX))
513 goto err;
514
515 nlk_sk(sk)->portid = portid;
516 sock_hold(sk);
517
518 err = __netlink_insert(table, sk);
519 if (err) {
520 /* In case the hashtable backend returns with -EBUSY
521 * from here, it must not escape to the caller.
522 */
523 if (unlikely(err == -EBUSY))
524 err = -EOVERFLOW;
525 if (err == -EEXIST)
526 err = -EADDRINUSE;
527 sock_put(sk);
528 goto err;
529 }
530
531 /* We need to ensure that the socket is hashed and visible. */
532 smp_wmb();
533 nlk_sk(sk)->bound = portid;
534
535err:
536 release_sock(sk);
537 return err;
538}
539
540static void netlink_remove(struct sock *sk)
541{
542 struct netlink_table *table;
543
544 table = &nl_table[sk->sk_protocol];
545 if (!rhashtable_remove_fast(&table->hash, &nlk_sk(sk)->node,
546 netlink_rhashtable_params)) {
547 WARN_ON(atomic_read(&sk->sk_refcnt) == 1);
548 __sock_put(sk);
549 }
550
551 netlink_table_grab();
552 if (nlk_sk(sk)->subscriptions) {
553 __sk_del_bind_node(sk);
554 netlink_update_listeners(sk);
555 }
556 if (sk->sk_protocol == NETLINK_GENERIC)
557 atomic_inc(&genl_sk_destructing_cnt);
558 netlink_table_ungrab();
559}
560
561static struct proto netlink_proto = {
562 .name = "NETLINK",
563 .owner = THIS_MODULE,
564 .obj_size = sizeof(struct netlink_sock),
565};
566
567static int __netlink_create(struct net *net, struct socket *sock,
568 struct mutex *cb_mutex, int protocol,
569 int kern)
570{
571 struct sock *sk;
572 struct netlink_sock *nlk;
573
574 sock->ops = &netlink_ops;
575
576 sk = sk_alloc(net, PF_NETLINK, GFP_KERNEL, &netlink_proto, kern);
577 if (!sk)
578 return -ENOMEM;
579
580 sock_init_data(sock, sk);
581
582 nlk = nlk_sk(sk);
583 if (cb_mutex) {
584 nlk->cb_mutex = cb_mutex;
585 } else {
586 nlk->cb_mutex = &nlk->cb_def_mutex;
587 mutex_init(nlk->cb_mutex);
588 }
589 init_waitqueue_head(&nlk->wait);
590
591 sk->sk_destruct = netlink_sock_destruct;
592 sk->sk_protocol = protocol;
593 return 0;
594}
595
596static int netlink_create(struct net *net, struct socket *sock, int protocol,
597 int kern)
598{
599 struct module *module = NULL;
600 struct mutex *cb_mutex;
601 struct netlink_sock *nlk;
602 int (*bind)(struct net *net, int group);
603 void (*unbind)(struct net *net, int group);
604 int err = 0;
605
606 sock->state = SS_UNCONNECTED;
607
608 if (sock->type != SOCK_RAW && sock->type != SOCK_DGRAM)
609 return -ESOCKTNOSUPPORT;
610
611 if (protocol < 0 || protocol >= MAX_LINKS)
612 return -EPROTONOSUPPORT;
613
614 netlink_lock_table();
615#ifdef CONFIG_MODULES
616 if (!nl_table[protocol].registered) {
617 netlink_unlock_table();
618 request_module("net-pf-%d-proto-%d", PF_NETLINK, protocol);
619 netlink_lock_table();
620 }
621#endif
622 if (nl_table[protocol].registered &&
623 try_module_get(nl_table[protocol].module))
624 module = nl_table[protocol].module;
625 else
626 err = -EPROTONOSUPPORT;
627 cb_mutex = nl_table[protocol].cb_mutex;
628 bind = nl_table[protocol].bind;
629 unbind = nl_table[protocol].unbind;
630 netlink_unlock_table();
631
632 if (err < 0)
633 goto out;
634
635 err = __netlink_create(net, sock, cb_mutex, protocol, kern);
636 if (err < 0)
637 goto out_module;
638
639 local_bh_disable();
640 sock_prot_inuse_add(net, &netlink_proto, 1);
641 local_bh_enable();
642
643 nlk = nlk_sk(sock->sk);
644 nlk->module = module;
645 nlk->netlink_bind = bind;
646 nlk->netlink_unbind = unbind;
647out:
648 return err;
649
650out_module:
651 module_put(module);
652 goto out;
653}
654
655static void deferred_put_nlk_sk(struct rcu_head *head)
656{
657 struct netlink_sock *nlk = container_of(head, struct netlink_sock, rcu);
658 struct sock *sk = &nlk->sk;
659
660 if (!atomic_dec_and_test(&sk->sk_refcnt))
661 return;
662
663 if (nlk->cb_running && nlk->cb.done) {
664 INIT_WORK(&nlk->work, netlink_sock_destruct_work);
665 schedule_work(&nlk->work);
666 return;
667 }
668
669 sk_free(sk);
670}
671
672static int netlink_release(struct socket *sock)
673{
674 struct sock *sk = sock->sk;
675 struct netlink_sock *nlk;
676
677 if (!sk)
678 return 0;
679
680 netlink_remove(sk);
681 sock_orphan(sk);
682 nlk = nlk_sk(sk);
683
684 /*
685 * OK. Socket is unlinked, any packets that arrive now
686 * will be purged.
687 */
688
689 /* must not acquire netlink_table_lock in any way again before unbind
690 * and notifying genetlink is done as otherwise it might deadlock
691 */
692 if (nlk->netlink_unbind) {
693 int i;
694
695 for (i = 0; i < nlk->ngroups; i++)
696 if (test_bit(i, nlk->groups))
697 nlk->netlink_unbind(sock_net(sk), i + 1);
698 }
699 if (sk->sk_protocol == NETLINK_GENERIC &&
700 atomic_dec_return(&genl_sk_destructing_cnt) == 0)
701 wake_up(&genl_sk_destructing_waitq);
702
703 sock->sk = NULL;
704 wake_up_interruptible_all(&nlk->wait);
705
706 skb_queue_purge(&sk->sk_write_queue);
707
708 if (nlk->portid && nlk->bound) {
709 struct netlink_notify n = {
710 .net = sock_net(sk),
711 .protocol = sk->sk_protocol,
712 .portid = nlk->portid,
713 };
714 blocking_notifier_call_chain(&netlink_chain,
715 NETLINK_URELEASE, &n);
716 }
717
718 module_put(nlk->module);
719
720 if (netlink_is_kernel(sk)) {
721 netlink_table_grab();
722 BUG_ON(nl_table[sk->sk_protocol].registered == 0);
723 if (--nl_table[sk->sk_protocol].registered == 0) {
724 struct listeners *old;
725
726 old = nl_deref_protected(nl_table[sk->sk_protocol].listeners);
727 RCU_INIT_POINTER(nl_table[sk->sk_protocol].listeners, NULL);
728 kfree_rcu(old, rcu);
729 nl_table[sk->sk_protocol].module = NULL;
730 nl_table[sk->sk_protocol].bind = NULL;
731 nl_table[sk->sk_protocol].unbind = NULL;
732 nl_table[sk->sk_protocol].flags = 0;
733 nl_table[sk->sk_protocol].registered = 0;
734 }
735 netlink_table_ungrab();
736 }
737
738 kfree(nlk->groups);
739 nlk->groups = NULL;
740
741 local_bh_disable();
742 sock_prot_inuse_add(sock_net(sk), &netlink_proto, -1);
743 local_bh_enable();
744 call_rcu(&nlk->rcu, deferred_put_nlk_sk);
745 return 0;
746}
747
748static int netlink_autobind(struct socket *sock)
749{
750 struct sock *sk = sock->sk;
751 struct net *net = sock_net(sk);
752 struct netlink_table *table = &nl_table[sk->sk_protocol];
753 s32 portid = task_tgid_vnr(current);
754 int err;
755 s32 rover = -4096;
756 bool ok;
757
758retry:
759 cond_resched();
760 rcu_read_lock();
761 ok = !__netlink_lookup(table, portid, net);
762 rcu_read_unlock();
763 if (!ok) {
764 /* Bind collision, search negative portid values. */
765 if (rover == -4096)
766 /* rover will be in range [S32_MIN, -4097] */
767 rover = S32_MIN + prandom_u32_max(-4096 - S32_MIN);
768 else if (rover >= -4096)
769 rover = -4097;
770 portid = rover--;
771 goto retry;
772 }
773
774 err = netlink_insert(sk, portid);
775 if (err == -EADDRINUSE)
776 goto retry;
777
778 /* If 2 threads race to autobind, that is fine. */
779 if (err == -EBUSY)
780 err = 0;
781
782 return err;
783}
784
785/**
786 * __netlink_ns_capable - General netlink message capability test
787 * @nsp: NETLINK_CB of the socket buffer holding a netlink command from userspace.
788 * @user_ns: The user namespace of the capability to use
789 * @cap: The capability to use
790 *
791 * Test to see if the opener of the socket we received the message
792 * from had when the netlink socket was created and the sender of the
793 * message has has the capability @cap in the user namespace @user_ns.
794 */
795bool __netlink_ns_capable(const struct netlink_skb_parms *nsp,
796 struct user_namespace *user_ns, int cap)
797{
798 return ((nsp->flags & NETLINK_SKB_DST) ||
799 file_ns_capable(nsp->sk->sk_socket->file, user_ns, cap)) &&
800 ns_capable(user_ns, cap);
801}
802EXPORT_SYMBOL(__netlink_ns_capable);
803
804/**
805 * netlink_ns_capable - General netlink message capability test
806 * @skb: socket buffer holding a netlink command from userspace
807 * @user_ns: The user namespace of the capability to use
808 * @cap: The capability to use
809 *
810 * Test to see if the opener of the socket we received the message
811 * from had when the netlink socket was created and the sender of the
812 * message has has the capability @cap in the user namespace @user_ns.
813 */
814bool netlink_ns_capable(const struct sk_buff *skb,
815 struct user_namespace *user_ns, int cap)
816{
817 return __netlink_ns_capable(&NETLINK_CB(skb), user_ns, cap);
818}
819EXPORT_SYMBOL(netlink_ns_capable);
820
821/**
822 * netlink_capable - Netlink global message capability test
823 * @skb: socket buffer holding a netlink command from userspace
824 * @cap: The capability to use
825 *
826 * Test to see if the opener of the socket we received the message
827 * from had when the netlink socket was created and the sender of the
828 * message has has the capability @cap in all user namespaces.
829 */
830bool netlink_capable(const struct sk_buff *skb, int cap)
831{
832 return netlink_ns_capable(skb, &init_user_ns, cap);
833}
834EXPORT_SYMBOL(netlink_capable);
835
836/**
837 * netlink_net_capable - Netlink network namespace message capability test
838 * @skb: socket buffer holding a netlink command from userspace
839 * @cap: The capability to use
840 *
841 * Test to see if the opener of the socket we received the message
842 * from had when the netlink socket was created and the sender of the
843 * message has has the capability @cap over the network namespace of
844 * the socket we received the message from.
845 */
846bool netlink_net_capable(const struct sk_buff *skb, int cap)
847{
848 return netlink_ns_capable(skb, sock_net(skb->sk)->user_ns, cap);
849}
850EXPORT_SYMBOL(netlink_net_capable);
851
852static inline int netlink_allowed(const struct socket *sock, unsigned int flag)
853{
854 return (nl_table[sock->sk->sk_protocol].flags & flag) ||
855 ns_capable(sock_net(sock->sk)->user_ns, CAP_NET_ADMIN);
856}
857
858static void
859netlink_update_subscriptions(struct sock *sk, unsigned int subscriptions)
860{
861 struct netlink_sock *nlk = nlk_sk(sk);
862
863 if (nlk->subscriptions && !subscriptions)
864 __sk_del_bind_node(sk);
865 else if (!nlk->subscriptions && subscriptions)
866 sk_add_bind_node(sk, &nl_table[sk->sk_protocol].mc_list);
867 nlk->subscriptions = subscriptions;
868}
869
870static int netlink_realloc_groups(struct sock *sk)
871{
872 struct netlink_sock *nlk = nlk_sk(sk);
873 unsigned int groups;
874 unsigned long *new_groups;
875 int err = 0;
876
877 netlink_table_grab();
878
879 groups = nl_table[sk->sk_protocol].groups;
880 if (!nl_table[sk->sk_protocol].registered) {
881 err = -ENOENT;
882 goto out_unlock;
883 }
884
885 if (nlk->ngroups >= groups)
886 goto out_unlock;
887
888 new_groups = krealloc(nlk->groups, NLGRPSZ(groups), GFP_ATOMIC);
889 if (new_groups == NULL) {
890 err = -ENOMEM;
891 goto out_unlock;
892 }
893 memset((char *)new_groups + NLGRPSZ(nlk->ngroups), 0,
894 NLGRPSZ(groups) - NLGRPSZ(nlk->ngroups));
895
896 nlk->groups = new_groups;
897 nlk->ngroups = groups;
898 out_unlock:
899 netlink_table_ungrab();
900 return err;
901}
902
903static void netlink_undo_bind(int group, long unsigned int groups,
904 struct sock *sk)
905{
906 struct netlink_sock *nlk = nlk_sk(sk);
907 int undo;
908
909 if (!nlk->netlink_unbind)
910 return;
911
912 for (undo = 0; undo < group; undo++)
913 if (test_bit(undo, &groups))
914 nlk->netlink_unbind(sock_net(sk), undo + 1);
915}
916
917static int netlink_bind(struct socket *sock, struct sockaddr *addr,
918 int addr_len)
919{
920 struct sock *sk = sock->sk;
921 struct net *net = sock_net(sk);
922 struct netlink_sock *nlk = nlk_sk(sk);
923 struct sockaddr_nl *nladdr = (struct sockaddr_nl *)addr;
924 int err;
925 long unsigned int groups = nladdr->nl_groups;
926 bool bound;
927
928 if (addr_len < sizeof(struct sockaddr_nl))
929 return -EINVAL;
930
931 if (nladdr->nl_family != AF_NETLINK)
932 return -EINVAL;
933
934 /* Only superuser is allowed to listen multicasts */
935 if (groups) {
936 if (!netlink_allowed(sock, NL_CFG_F_NONROOT_RECV))
937 return -EPERM;
938 err = netlink_realloc_groups(sk);
939 if (err)
940 return err;
941 }
942
943 bound = nlk->bound;
944 if (bound) {
945 /* Ensure nlk->portid is up-to-date. */
946 smp_rmb();
947
948 if (nladdr->nl_pid != nlk->portid)
949 return -EINVAL;
950 }
951
952 if (nlk->netlink_bind && groups) {
953 int group;
954
955 for (group = 0; group < nlk->ngroups; group++) {
956 if (!test_bit(group, &groups))
957 continue;
958 err = nlk->netlink_bind(net, group + 1);
959 if (!err)
960 continue;
961 netlink_undo_bind(group, groups, sk);
962 return err;
963 }
964 }
965
966 /* No need for barriers here as we return to user-space without
967 * using any of the bound attributes.
968 */
969 if (!bound) {
970 err = nladdr->nl_pid ?
971 netlink_insert(sk, nladdr->nl_pid) :
972 netlink_autobind(sock);
973 if (err) {
974 netlink_undo_bind(nlk->ngroups, groups, sk);
975 return err;
976 }
977 }
978
979 if (!groups && (nlk->groups == NULL || !(u32)nlk->groups[0]))
980 return 0;
981
982 netlink_table_grab();
983 netlink_update_subscriptions(sk, nlk->subscriptions +
984 hweight32(groups) -
985 hweight32(nlk->groups[0]));
986 nlk->groups[0] = (nlk->groups[0] & ~0xffffffffUL) | groups;
987 netlink_update_listeners(sk);
988 netlink_table_ungrab();
989
990 return 0;
991}
992
993static int netlink_connect(struct socket *sock, struct sockaddr *addr,
994 int alen, int flags)
995{
996 int err = 0;
997 struct sock *sk = sock->sk;
998 struct netlink_sock *nlk = nlk_sk(sk);
999 struct sockaddr_nl *nladdr = (struct sockaddr_nl *)addr;
1000
1001 if (alen < sizeof(addr->sa_family))
1002 return -EINVAL;
1003
1004 if (addr->sa_family == AF_UNSPEC) {
1005 sk->sk_state = NETLINK_UNCONNECTED;
1006 nlk->dst_portid = 0;
1007 nlk->dst_group = 0;
1008 return 0;
1009 }
1010 if (addr->sa_family != AF_NETLINK)
1011 return -EINVAL;
1012
1013 if ((nladdr->nl_groups || nladdr->nl_pid) &&
1014 !netlink_allowed(sock, NL_CFG_F_NONROOT_SEND))
1015 return -EPERM;
1016
1017 /* No need for barriers here as we return to user-space without
1018 * using any of the bound attributes.
1019 */
1020 if (!nlk->bound)
1021 err = netlink_autobind(sock);
1022
1023 if (err == 0) {
1024 sk->sk_state = NETLINK_CONNECTED;
1025 nlk->dst_portid = nladdr->nl_pid;
1026 nlk->dst_group = ffs(nladdr->nl_groups);
1027 }
1028
1029 return err;
1030}
1031
1032static int netlink_getname(struct socket *sock, struct sockaddr *addr,
1033 int *addr_len, int peer)
1034{
1035 struct sock *sk = sock->sk;
1036 struct netlink_sock *nlk = nlk_sk(sk);
1037 DECLARE_SOCKADDR(struct sockaddr_nl *, nladdr, addr);
1038
1039 nladdr->nl_family = AF_NETLINK;
1040 nladdr->nl_pad = 0;
1041 *addr_len = sizeof(*nladdr);
1042
1043 if (peer) {
1044 nladdr->nl_pid = nlk->dst_portid;
1045 nladdr->nl_groups = netlink_group_mask(nlk->dst_group);
1046 } else {
1047 nladdr->nl_pid = nlk->portid;
1048 nladdr->nl_groups = nlk->groups ? nlk->groups[0] : 0;
1049 }
1050 return 0;
1051}
1052
1053static int netlink_ioctl(struct socket *sock, unsigned int cmd,
1054 unsigned long arg)
1055{
1056 /* try to hand this ioctl down to the NIC drivers.
1057 */
1058 return -ENOIOCTLCMD;
1059}
1060
1061static struct sock *netlink_getsockbyportid(struct sock *ssk, u32 portid)
1062{
1063 struct sock *sock;
1064 struct netlink_sock *nlk;
1065
1066 sock = netlink_lookup(sock_net(ssk), ssk->sk_protocol, portid);
1067 if (!sock)
1068 return ERR_PTR(-ECONNREFUSED);
1069
1070 /* Don't bother queuing skb if kernel socket has no input function */
1071 nlk = nlk_sk(sock);
1072 if (sock->sk_state == NETLINK_CONNECTED &&
1073 nlk->dst_portid != nlk_sk(ssk)->portid) {
1074 sock_put(sock);
1075 return ERR_PTR(-ECONNREFUSED);
1076 }
1077 return sock;
1078}
1079
1080struct sock *netlink_getsockbyfilp(struct file *filp)
1081{
1082 struct inode *inode = file_inode(filp);
1083 struct sock *sock;
1084
1085 if (!S_ISSOCK(inode->i_mode))
1086 return ERR_PTR(-ENOTSOCK);
1087
1088 sock = SOCKET_I(inode)->sk;
1089 if (sock->sk_family != AF_NETLINK)
1090 return ERR_PTR(-EINVAL);
1091
1092 sock_hold(sock);
1093 return sock;
1094}
1095
1096static struct sk_buff *netlink_alloc_large_skb(unsigned int size,
1097 int broadcast)
1098{
1099 struct sk_buff *skb;
1100 void *data;
1101
1102 if (size <= NLMSG_GOODSIZE || broadcast)
1103 return alloc_skb(size, GFP_KERNEL);
1104
1105 size = SKB_DATA_ALIGN(size) +
1106 SKB_DATA_ALIGN(sizeof(struct skb_shared_info));
1107
1108 data = vmalloc(size);
1109 if (data == NULL)
1110 return NULL;
1111
1112 skb = __build_skb(data, size);
1113 if (skb == NULL)
1114 vfree(data);
1115 else
1116 skb->destructor = netlink_skb_destructor;
1117
1118 return skb;
1119}
1120
1121/*
1122 * Attach a skb to a netlink socket.
1123 * The caller must hold a reference to the destination socket. On error, the
1124 * reference is dropped. The skb is not send to the destination, just all
1125 * all error checks are performed and memory in the queue is reserved.
1126 * Return values:
1127 * < 0: error. skb freed, reference to sock dropped.
1128 * 0: continue
1129 * 1: repeat lookup - reference dropped while waiting for socket memory.
1130 */
1131int netlink_attachskb(struct sock *sk, struct sk_buff *skb,
1132 long *timeo, struct sock *ssk)
1133{
1134 struct netlink_sock *nlk;
1135
1136 nlk = nlk_sk(sk);
1137
1138 if ((atomic_read(&sk->sk_rmem_alloc) > sk->sk_rcvbuf ||
1139 test_bit(NETLINK_S_CONGESTED, &nlk->state))) {
1140 DECLARE_WAITQUEUE(wait, current);
1141 if (!*timeo) {
1142 if (!ssk || netlink_is_kernel(ssk))
1143 netlink_overrun(sk);
1144 sock_put(sk);
1145 kfree_skb(skb);
1146 return -EAGAIN;
1147 }
1148
1149 __set_current_state(TASK_INTERRUPTIBLE);
1150 add_wait_queue(&nlk->wait, &wait);
1151
1152 if ((atomic_read(&sk->sk_rmem_alloc) > sk->sk_rcvbuf ||
1153 test_bit(NETLINK_S_CONGESTED, &nlk->state)) &&
1154 !sock_flag(sk, SOCK_DEAD))
1155 *timeo = schedule_timeout(*timeo);
1156
1157 __set_current_state(TASK_RUNNING);
1158 remove_wait_queue(&nlk->wait, &wait);
1159 sock_put(sk);
1160
1161 if (signal_pending(current)) {
1162 kfree_skb(skb);
1163 return sock_intr_errno(*timeo);
1164 }
1165 return 1;
1166 }
1167 netlink_skb_set_owner_r(skb, sk);
1168 return 0;
1169}
1170
1171static int __netlink_sendskb(struct sock *sk, struct sk_buff *skb)
1172{
1173 int len = skb->len;
1174
1175 netlink_deliver_tap(skb);
1176
1177 skb_queue_tail(&sk->sk_receive_queue, skb);
1178 sk->sk_data_ready(sk);
1179 return len;
1180}
1181
1182int netlink_sendskb(struct sock *sk, struct sk_buff *skb)
1183{
1184 int len = __netlink_sendskb(sk, skb);
1185
1186 sock_put(sk);
1187 return len;
1188}
1189
1190void netlink_detachskb(struct sock *sk, struct sk_buff *skb)
1191{
1192 kfree_skb(skb);
1193 sock_put(sk);
1194}
1195
1196static struct sk_buff *netlink_trim(struct sk_buff *skb, gfp_t allocation)
1197{
1198 int delta;
1199
1200 WARN_ON(skb->sk != NULL);
1201 delta = skb->end - skb->tail;
1202 if (is_vmalloc_addr(skb->head) || delta * 2 < skb->truesize)
1203 return skb;
1204
1205 if (skb_shared(skb)) {
1206 struct sk_buff *nskb = skb_clone(skb, allocation);
1207 if (!nskb)
1208 return skb;
1209 consume_skb(skb);
1210 skb = nskb;
1211 }
1212
1213 if (!pskb_expand_head(skb, 0, -delta, allocation))
1214 skb->truesize -= delta;
1215
1216 return skb;
1217}
1218
1219static int netlink_unicast_kernel(struct sock *sk, struct sk_buff *skb,
1220 struct sock *ssk)
1221{
1222 int ret;
1223 struct netlink_sock *nlk = nlk_sk(sk);
1224
1225 ret = -ECONNREFUSED;
1226 if (nlk->netlink_rcv != NULL) {
1227 ret = skb->len;
1228 netlink_skb_set_owner_r(skb, sk);
1229 NETLINK_CB(skb).sk = ssk;
1230 netlink_deliver_tap_kernel(sk, ssk, skb);
1231 nlk->netlink_rcv(skb);
1232 consume_skb(skb);
1233 } else {
1234 kfree_skb(skb);
1235 }
1236 sock_put(sk);
1237 return ret;
1238}
1239
1240int netlink_unicast(struct sock *ssk, struct sk_buff *skb,
1241 u32 portid, int nonblock)
1242{
1243 struct sock *sk;
1244 int err;
1245 long timeo;
1246
1247 skb = netlink_trim(skb, gfp_any());
1248
1249 timeo = sock_sndtimeo(ssk, nonblock);
1250retry:
1251 sk = netlink_getsockbyportid(ssk, portid);
1252 if (IS_ERR(sk)) {
1253 kfree_skb(skb);
1254 return PTR_ERR(sk);
1255 }
1256 if (netlink_is_kernel(sk))
1257 return netlink_unicast_kernel(sk, skb, ssk);
1258
1259 if (sk_filter(sk, skb)) {
1260 err = skb->len;
1261 kfree_skb(skb);
1262 sock_put(sk);
1263 return err;
1264 }
1265
1266 err = netlink_attachskb(sk, skb, &timeo, ssk);
1267 if (err == 1)
1268 goto retry;
1269 if (err)
1270 return err;
1271
1272 return netlink_sendskb(sk, skb);
1273}
1274EXPORT_SYMBOL(netlink_unicast);
1275
1276int netlink_has_listeners(struct sock *sk, unsigned int group)
1277{
1278 int res = 0;
1279 struct listeners *listeners;
1280
1281 BUG_ON(!netlink_is_kernel(sk));
1282
1283 rcu_read_lock();
1284 listeners = rcu_dereference(nl_table[sk->sk_protocol].listeners);
1285
1286 if (listeners && group - 1 < nl_table[sk->sk_protocol].groups)
1287 res = test_bit(group - 1, listeners->masks);
1288
1289 rcu_read_unlock();
1290
1291 return res;
1292}
1293EXPORT_SYMBOL_GPL(netlink_has_listeners);
1294
1295static int netlink_broadcast_deliver(struct sock *sk, struct sk_buff *skb)
1296{
1297 struct netlink_sock *nlk = nlk_sk(sk);
1298
1299 if (atomic_read(&sk->sk_rmem_alloc) <= sk->sk_rcvbuf &&
1300 !test_bit(NETLINK_S_CONGESTED, &nlk->state)) {
1301 netlink_skb_set_owner_r(skb, sk);
1302 __netlink_sendskb(sk, skb);
1303 return atomic_read(&sk->sk_rmem_alloc) > (sk->sk_rcvbuf >> 1);
1304 }
1305 return -1;
1306}
1307
1308struct netlink_broadcast_data {
1309 struct sock *exclude_sk;
1310 struct net *net;
1311 u32 portid;
1312 u32 group;
1313 int failure;
1314 int delivery_failure;
1315 int congested;
1316 int delivered;
1317 gfp_t allocation;
1318 struct sk_buff *skb, *skb2;
1319 int (*tx_filter)(struct sock *dsk, struct sk_buff *skb, void *data);
1320 void *tx_data;
1321};
1322
1323static void do_one_broadcast(struct sock *sk,
1324 struct netlink_broadcast_data *p)
1325{
1326 struct netlink_sock *nlk = nlk_sk(sk);
1327 int val;
1328
1329 if (p->exclude_sk == sk)
1330 return;
1331
1332 if (nlk->portid == p->portid || p->group - 1 >= nlk->ngroups ||
1333 !test_bit(p->group - 1, nlk->groups))
1334 return;
1335
1336 if (!net_eq(sock_net(sk), p->net)) {
1337 if (!(nlk->flags & NETLINK_F_LISTEN_ALL_NSID))
1338 return;
1339
1340 if (!peernet_has_id(sock_net(sk), p->net))
1341 return;
1342
1343 if (!file_ns_capable(sk->sk_socket->file, p->net->user_ns,
1344 CAP_NET_BROADCAST))
1345 return;
1346 }
1347
1348 if (p->failure) {
1349 netlink_overrun(sk);
1350 return;
1351 }
1352
1353 sock_hold(sk);
1354 if (p->skb2 == NULL) {
1355 if (skb_shared(p->skb)) {
1356 p->skb2 = skb_clone(p->skb, p->allocation);
1357 } else {
1358 p->skb2 = skb_get(p->skb);
1359 /*
1360 * skb ownership may have been set when
1361 * delivered to a previous socket.
1362 */
1363 skb_orphan(p->skb2);
1364 }
1365 }
1366 if (p->skb2 == NULL) {
1367 netlink_overrun(sk);
1368 /* Clone failed. Notify ALL listeners. */
1369 p->failure = 1;
1370 if (nlk->flags & NETLINK_F_BROADCAST_SEND_ERROR)
1371 p->delivery_failure = 1;
1372 goto out;
1373 }
1374 if (p->tx_filter && p->tx_filter(sk, p->skb2, p->tx_data)) {
1375 kfree_skb(p->skb2);
1376 p->skb2 = NULL;
1377 goto out;
1378 }
1379 if (sk_filter(sk, p->skb2)) {
1380 kfree_skb(p->skb2);
1381 p->skb2 = NULL;
1382 goto out;
1383 }
1384 NETLINK_CB(p->skb2).nsid = peernet2id(sock_net(sk), p->net);
1385 NETLINK_CB(p->skb2).nsid_is_set = true;
1386 val = netlink_broadcast_deliver(sk, p->skb2);
1387 if (val < 0) {
1388 netlink_overrun(sk);
1389 if (nlk->flags & NETLINK_F_BROADCAST_SEND_ERROR)
1390 p->delivery_failure = 1;
1391 } else {
1392 p->congested |= val;
1393 p->delivered = 1;
1394 p->skb2 = NULL;
1395 }
1396out:
1397 sock_put(sk);
1398}
1399
1400int netlink_broadcast_filtered(struct sock *ssk, struct sk_buff *skb, u32 portid,
1401 u32 group, gfp_t allocation,
1402 int (*filter)(struct sock *dsk, struct sk_buff *skb, void *data),
1403 void *filter_data)
1404{
1405 struct net *net = sock_net(ssk);
1406 struct netlink_broadcast_data info;
1407 struct sock *sk;
1408
1409 skb = netlink_trim(skb, allocation);
1410
1411 info.exclude_sk = ssk;
1412 info.net = net;
1413 info.portid = portid;
1414 info.group = group;
1415 info.failure = 0;
1416 info.delivery_failure = 0;
1417 info.congested = 0;
1418 info.delivered = 0;
1419 info.allocation = allocation;
1420 info.skb = skb;
1421 info.skb2 = NULL;
1422 info.tx_filter = filter;
1423 info.tx_data = filter_data;
1424
1425 /* While we sleep in clone, do not allow to change socket list */
1426
1427 netlink_lock_table();
1428
1429 sk_for_each_bound(sk, &nl_table[ssk->sk_protocol].mc_list)
1430 do_one_broadcast(sk, &info);
1431
1432 consume_skb(skb);
1433
1434 netlink_unlock_table();
1435
1436 if (info.delivery_failure) {
1437 kfree_skb(info.skb2);
1438 return -ENOBUFS;
1439 }
1440 consume_skb(info.skb2);
1441
1442 if (info.delivered) {
1443 if (info.congested && gfpflags_allow_blocking(allocation))
1444 yield();
1445 return 0;
1446 }
1447 return -ESRCH;
1448}
1449EXPORT_SYMBOL(netlink_broadcast_filtered);
1450
1451int netlink_broadcast(struct sock *ssk, struct sk_buff *skb, u32 portid,
1452 u32 group, gfp_t allocation)
1453{
1454 return netlink_broadcast_filtered(ssk, skb, portid, group, allocation,
1455 NULL, NULL);
1456}
1457EXPORT_SYMBOL(netlink_broadcast);
1458
1459struct netlink_set_err_data {
1460 struct sock *exclude_sk;
1461 u32 portid;
1462 u32 group;
1463 int code;
1464};
1465
1466static int do_one_set_err(struct sock *sk, struct netlink_set_err_data *p)
1467{
1468 struct netlink_sock *nlk = nlk_sk(sk);
1469 int ret = 0;
1470
1471 if (sk == p->exclude_sk)
1472 goto out;
1473
1474 if (!net_eq(sock_net(sk), sock_net(p->exclude_sk)))
1475 goto out;
1476
1477 if (nlk->portid == p->portid || p->group - 1 >= nlk->ngroups ||
1478 !test_bit(p->group - 1, nlk->groups))
1479 goto out;
1480
1481 if (p->code == ENOBUFS && nlk->flags & NETLINK_F_RECV_NO_ENOBUFS) {
1482 ret = 1;
1483 goto out;
1484 }
1485
1486 sk->sk_err = p->code;
1487 sk->sk_error_report(sk);
1488out:
1489 return ret;
1490}
1491
1492/**
1493 * netlink_set_err - report error to broadcast listeners
1494 * @ssk: the kernel netlink socket, as returned by netlink_kernel_create()
1495 * @portid: the PORTID of a process that we want to skip (if any)
1496 * @group: the broadcast group that will notice the error
1497 * @code: error code, must be negative (as usual in kernelspace)
1498 *
1499 * This function returns the number of broadcast listeners that have set the
1500 * NETLINK_NO_ENOBUFS socket option.
1501 */
1502int netlink_set_err(struct sock *ssk, u32 portid, u32 group, int code)
1503{
1504 struct netlink_set_err_data info;
1505 struct sock *sk;
1506 int ret = 0;
1507
1508 info.exclude_sk = ssk;
1509 info.portid = portid;
1510 info.group = group;
1511 /* sk->sk_err wants a positive error value */
1512 info.code = -code;
1513
1514 read_lock(&nl_table_lock);
1515
1516 sk_for_each_bound(sk, &nl_table[ssk->sk_protocol].mc_list)
1517 ret += do_one_set_err(sk, &info);
1518
1519 read_unlock(&nl_table_lock);
1520 return ret;
1521}
1522EXPORT_SYMBOL(netlink_set_err);
1523
1524/* must be called with netlink table grabbed */
1525static void netlink_update_socket_mc(struct netlink_sock *nlk,
1526 unsigned int group,
1527 int is_new)
1528{
1529 int old, new = !!is_new, subscriptions;
1530
1531 old = test_bit(group - 1, nlk->groups);
1532 subscriptions = nlk->subscriptions - old + new;
1533 if (new)
1534 __set_bit(group - 1, nlk->groups);
1535 else
1536 __clear_bit(group - 1, nlk->groups);
1537 netlink_update_subscriptions(&nlk->sk, subscriptions);
1538 netlink_update_listeners(&nlk->sk);
1539}
1540
1541static int netlink_setsockopt(struct socket *sock, int level, int optname,
1542 char __user *optval, unsigned int optlen)
1543{
1544 struct sock *sk = sock->sk;
1545 struct netlink_sock *nlk = nlk_sk(sk);
1546 unsigned int val = 0;
1547 int err;
1548
1549 if (level != SOL_NETLINK)
1550 return -ENOPROTOOPT;
1551
1552 if (optlen >= sizeof(int) &&
1553 get_user(val, (unsigned int __user *)optval))
1554 return -EFAULT;
1555
1556 switch (optname) {
1557 case NETLINK_PKTINFO:
1558 if (val)
1559 nlk->flags |= NETLINK_F_RECV_PKTINFO;
1560 else
1561 nlk->flags &= ~NETLINK_F_RECV_PKTINFO;
1562 err = 0;
1563 break;
1564 case NETLINK_ADD_MEMBERSHIP:
1565 case NETLINK_DROP_MEMBERSHIP: {
1566 if (!netlink_allowed(sock, NL_CFG_F_NONROOT_RECV))
1567 return -EPERM;
1568 err = netlink_realloc_groups(sk);
1569 if (err)
1570 return err;
1571 if (!val || val - 1 >= nlk->ngroups)
1572 return -EINVAL;
1573 if (optname == NETLINK_ADD_MEMBERSHIP && nlk->netlink_bind) {
1574 err = nlk->netlink_bind(sock_net(sk), val);
1575 if (err)
1576 return err;
1577 }
1578 netlink_table_grab();
1579 netlink_update_socket_mc(nlk, val,
1580 optname == NETLINK_ADD_MEMBERSHIP);
1581 netlink_table_ungrab();
1582 if (optname == NETLINK_DROP_MEMBERSHIP && nlk->netlink_unbind)
1583 nlk->netlink_unbind(sock_net(sk), val);
1584
1585 err = 0;
1586 break;
1587 }
1588 case NETLINK_BROADCAST_ERROR:
1589 if (val)
1590 nlk->flags |= NETLINK_F_BROADCAST_SEND_ERROR;
1591 else
1592 nlk->flags &= ~NETLINK_F_BROADCAST_SEND_ERROR;
1593 err = 0;
1594 break;
1595 case NETLINK_NO_ENOBUFS:
1596 if (val) {
1597 nlk->flags |= NETLINK_F_RECV_NO_ENOBUFS;
1598 clear_bit(NETLINK_S_CONGESTED, &nlk->state);
1599 wake_up_interruptible(&nlk->wait);
1600 } else {
1601 nlk->flags &= ~NETLINK_F_RECV_NO_ENOBUFS;
1602 }
1603 err = 0;
1604 break;
1605 case NETLINK_LISTEN_ALL_NSID:
1606 if (!ns_capable(sock_net(sk)->user_ns, CAP_NET_BROADCAST))
1607 return -EPERM;
1608
1609 if (val)
1610 nlk->flags |= NETLINK_F_LISTEN_ALL_NSID;
1611 else
1612 nlk->flags &= ~NETLINK_F_LISTEN_ALL_NSID;
1613 err = 0;
1614 break;
1615 case NETLINK_CAP_ACK:
1616 if (val)
1617 nlk->flags |= NETLINK_F_CAP_ACK;
1618 else
1619 nlk->flags &= ~NETLINK_F_CAP_ACK;
1620 err = 0;
1621 break;
1622 default:
1623 err = -ENOPROTOOPT;
1624 }
1625 return err;
1626}
1627
1628static int netlink_getsockopt(struct socket *sock, int level, int optname,
1629 char __user *optval, int __user *optlen)
1630{
1631 struct sock *sk = sock->sk;
1632 struct netlink_sock *nlk = nlk_sk(sk);
1633 int len, val, err;
1634
1635 if (level != SOL_NETLINK)
1636 return -ENOPROTOOPT;
1637
1638 if (get_user(len, optlen))
1639 return -EFAULT;
1640 if (len < 0)
1641 return -EINVAL;
1642
1643 switch (optname) {
1644 case NETLINK_PKTINFO:
1645 if (len < sizeof(int))
1646 return -EINVAL;
1647 len = sizeof(int);
1648 val = nlk->flags & NETLINK_F_RECV_PKTINFO ? 1 : 0;
1649 if (put_user(len, optlen) ||
1650 put_user(val, optval))
1651 return -EFAULT;
1652 err = 0;
1653 break;
1654 case NETLINK_BROADCAST_ERROR:
1655 if (len < sizeof(int))
1656 return -EINVAL;
1657 len = sizeof(int);
1658 val = nlk->flags & NETLINK_F_BROADCAST_SEND_ERROR ? 1 : 0;
1659 if (put_user(len, optlen) ||
1660 put_user(val, optval))
1661 return -EFAULT;
1662 err = 0;
1663 break;
1664 case NETLINK_NO_ENOBUFS:
1665 if (len < sizeof(int))
1666 return -EINVAL;
1667 len = sizeof(int);
1668 val = nlk->flags & NETLINK_F_RECV_NO_ENOBUFS ? 1 : 0;
1669 if (put_user(len, optlen) ||
1670 put_user(val, optval))
1671 return -EFAULT;
1672 err = 0;
1673 break;
1674 case NETLINK_LIST_MEMBERSHIPS: {
1675 int pos, idx, shift;
1676
1677 err = 0;
1678 netlink_lock_table();
1679 for (pos = 0; pos * 8 < nlk->ngroups; pos += sizeof(u32)) {
1680 if (len - pos < sizeof(u32))
1681 break;
1682
1683 idx = pos / sizeof(unsigned long);
1684 shift = (pos % sizeof(unsigned long)) * 8;
1685 if (put_user((u32)(nlk->groups[idx] >> shift),
1686 (u32 __user *)(optval + pos))) {
1687 err = -EFAULT;
1688 break;
1689 }
1690 }
1691 if (put_user(ALIGN(nlk->ngroups / 8, sizeof(u32)), optlen))
1692 err = -EFAULT;
1693 netlink_unlock_table();
1694 break;
1695 }
1696 case NETLINK_CAP_ACK:
1697 if (len < sizeof(int))
1698 return -EINVAL;
1699 len = sizeof(int);
1700 val = nlk->flags & NETLINK_F_CAP_ACK ? 1 : 0;
1701 if (put_user(len, optlen) ||
1702 put_user(val, optval))
1703 return -EFAULT;
1704 err = 0;
1705 break;
1706 default:
1707 err = -ENOPROTOOPT;
1708 }
1709 return err;
1710}
1711
1712static void netlink_cmsg_recv_pktinfo(struct msghdr *msg, struct sk_buff *skb)
1713{
1714 struct nl_pktinfo info;
1715
1716 info.group = NETLINK_CB(skb).dst_group;
1717 put_cmsg(msg, SOL_NETLINK, NETLINK_PKTINFO, sizeof(info), &info);
1718}
1719
1720static void netlink_cmsg_listen_all_nsid(struct sock *sk, struct msghdr *msg,
1721 struct sk_buff *skb)
1722{
1723 if (!NETLINK_CB(skb).nsid_is_set)
1724 return;
1725
1726 put_cmsg(msg, SOL_NETLINK, NETLINK_LISTEN_ALL_NSID, sizeof(int),
1727 &NETLINK_CB(skb).nsid);
1728}
1729
1730static int netlink_sendmsg(struct socket *sock, struct msghdr *msg, size_t len)
1731{
1732 struct sock *sk = sock->sk;
1733 struct netlink_sock *nlk = nlk_sk(sk);
1734 DECLARE_SOCKADDR(struct sockaddr_nl *, addr, msg->msg_name);
1735 u32 dst_portid;
1736 u32 dst_group;
1737 struct sk_buff *skb;
1738 int err;
1739 struct scm_cookie scm;
1740 u32 netlink_skb_flags = 0;
1741
1742 if (msg->msg_flags&MSG_OOB)
1743 return -EOPNOTSUPP;
1744
1745 err = scm_send(sock, msg, &scm, true);
1746 if (err < 0)
1747 return err;
1748
1749 if (msg->msg_namelen) {
1750 err = -EINVAL;
1751 if (addr->nl_family != AF_NETLINK)
1752 goto out;
1753 dst_portid = addr->nl_pid;
1754 dst_group = ffs(addr->nl_groups);
1755 err = -EPERM;
1756 if ((dst_group || dst_portid) &&
1757 !netlink_allowed(sock, NL_CFG_F_NONROOT_SEND))
1758 goto out;
1759 netlink_skb_flags |= NETLINK_SKB_DST;
1760 } else {
1761 dst_portid = nlk->dst_portid;
1762 dst_group = nlk->dst_group;
1763 }
1764
1765 if (!nlk->bound) {
1766 err = netlink_autobind(sock);
1767 if (err)
1768 goto out;
1769 } else {
1770 /* Ensure nlk is hashed and visible. */
1771 smp_rmb();
1772 }
1773
1774 err = -EMSGSIZE;
1775 if (len > sk->sk_sndbuf - 32)
1776 goto out;
1777 err = -ENOBUFS;
1778 skb = netlink_alloc_large_skb(len, dst_group);
1779 if (skb == NULL)
1780 goto out;
1781
1782 NETLINK_CB(skb).portid = nlk->portid;
1783 NETLINK_CB(skb).dst_group = dst_group;
1784 NETLINK_CB(skb).creds = scm.creds;
1785 NETLINK_CB(skb).flags = netlink_skb_flags;
1786
1787 err = -EFAULT;
1788 if (memcpy_from_msg(skb_put(skb, len), msg, len)) {
1789 kfree_skb(skb);
1790 goto out;
1791 }
1792
1793 err = security_netlink_send(sk, skb);
1794 if (err) {
1795 kfree_skb(skb);
1796 goto out;
1797 }
1798
1799 if (dst_group) {
1800 atomic_inc(&skb->users);
1801 netlink_broadcast(sk, skb, dst_portid, dst_group, GFP_KERNEL);
1802 }
1803 err = netlink_unicast(sk, skb, dst_portid, msg->msg_flags&MSG_DONTWAIT);
1804
1805out:
1806 scm_destroy(&scm);
1807 return err;
1808}
1809
1810static int netlink_recvmsg(struct socket *sock, struct msghdr *msg, size_t len,
1811 int flags)
1812{
1813 struct scm_cookie scm;
1814 struct sock *sk = sock->sk;
1815 struct netlink_sock *nlk = nlk_sk(sk);
1816 int noblock = flags&MSG_DONTWAIT;
1817 size_t copied;
1818 struct sk_buff *skb, *data_skb;
1819 int err, ret;
1820
1821 if (flags&MSG_OOB)
1822 return -EOPNOTSUPP;
1823
1824 copied = 0;
1825
1826 skb = skb_recv_datagram(sk, flags, noblock, &err);
1827 if (skb == NULL)
1828 goto out;
1829
1830 data_skb = skb;
1831
1832#ifdef CONFIG_COMPAT_NETLINK_MESSAGES
1833 if (unlikely(skb_shinfo(skb)->frag_list)) {
1834 /*
1835 * If this skb has a frag_list, then here that means that we
1836 * will have to use the frag_list skb's data for compat tasks
1837 * and the regular skb's data for normal (non-compat) tasks.
1838 *
1839 * If we need to send the compat skb, assign it to the
1840 * 'data_skb' variable so that it will be used below for data
1841 * copying. We keep 'skb' for everything else, including
1842 * freeing both later.
1843 */
1844 if (flags & MSG_CMSG_COMPAT)
1845 data_skb = skb_shinfo(skb)->frag_list;
1846 }
1847#endif
1848
1849 /* Record the max length of recvmsg() calls for future allocations */
1850 nlk->max_recvmsg_len = max(nlk->max_recvmsg_len, len);
1851 nlk->max_recvmsg_len = min_t(size_t, nlk->max_recvmsg_len,
1852 SKB_WITH_OVERHEAD(32768));
1853
1854 copied = data_skb->len;
1855 if (len < copied) {
1856 msg->msg_flags |= MSG_TRUNC;
1857 copied = len;
1858 }
1859
1860 skb_reset_transport_header(data_skb);
1861 err = skb_copy_datagram_msg(data_skb, 0, msg, copied);
1862
1863 if (msg->msg_name) {
1864 DECLARE_SOCKADDR(struct sockaddr_nl *, addr, msg->msg_name);
1865 addr->nl_family = AF_NETLINK;
1866 addr->nl_pad = 0;
1867 addr->nl_pid = NETLINK_CB(skb).portid;
1868 addr->nl_groups = netlink_group_mask(NETLINK_CB(skb).dst_group);
1869 msg->msg_namelen = sizeof(*addr);
1870 }
1871
1872 if (nlk->flags & NETLINK_F_RECV_PKTINFO)
1873 netlink_cmsg_recv_pktinfo(msg, skb);
1874 if (nlk->flags & NETLINK_F_LISTEN_ALL_NSID)
1875 netlink_cmsg_listen_all_nsid(sk, msg, skb);
1876
1877 memset(&scm, 0, sizeof(scm));
1878 scm.creds = *NETLINK_CREDS(skb);
1879 if (flags & MSG_TRUNC)
1880 copied = data_skb->len;
1881
1882 skb_free_datagram(sk, skb);
1883
1884 if (nlk->cb_running &&
1885 atomic_read(&sk->sk_rmem_alloc) <= sk->sk_rcvbuf / 2) {
1886 ret = netlink_dump(sk);
1887 if (ret) {
1888 sk->sk_err = -ret;
1889 sk->sk_error_report(sk);
1890 }
1891 }
1892
1893 scm_recv(sock, msg, &scm, flags);
1894out:
1895 netlink_rcv_wake(sk);
1896 return err ? : copied;
1897}
1898
1899static void netlink_data_ready(struct sock *sk)
1900{
1901 BUG();
1902}
1903
1904/*
1905 * We export these functions to other modules. They provide a
1906 * complete set of kernel non-blocking support for message
1907 * queueing.
1908 */
1909
1910struct sock *
1911__netlink_kernel_create(struct net *net, int unit, struct module *module,
1912 struct netlink_kernel_cfg *cfg)
1913{
1914 struct socket *sock;
1915 struct sock *sk;
1916 struct netlink_sock *nlk;
1917 struct listeners *listeners = NULL;
1918 struct mutex *cb_mutex = cfg ? cfg->cb_mutex : NULL;
1919 unsigned int groups;
1920
1921 BUG_ON(!nl_table);
1922
1923 if (unit < 0 || unit >= MAX_LINKS)
1924 return NULL;
1925
1926 if (sock_create_lite(PF_NETLINK, SOCK_DGRAM, unit, &sock))
1927 return NULL;
1928
1929 if (__netlink_create(net, sock, cb_mutex, unit, 1) < 0)
1930 goto out_sock_release_nosk;
1931
1932 sk = sock->sk;
1933
1934 if (!cfg || cfg->groups < 32)
1935 groups = 32;
1936 else
1937 groups = cfg->groups;
1938
1939 listeners = kzalloc(sizeof(*listeners) + NLGRPSZ(groups), GFP_KERNEL);
1940 if (!listeners)
1941 goto out_sock_release;
1942
1943 sk->sk_data_ready = netlink_data_ready;
1944 if (cfg && cfg->input)
1945 nlk_sk(sk)->netlink_rcv = cfg->input;
1946
1947 if (netlink_insert(sk, 0))
1948 goto out_sock_release;
1949
1950 nlk = nlk_sk(sk);
1951 nlk->flags |= NETLINK_F_KERNEL_SOCKET;
1952
1953 netlink_table_grab();
1954 if (!nl_table[unit].registered) {
1955 nl_table[unit].groups = groups;
1956 rcu_assign_pointer(nl_table[unit].listeners, listeners);
1957 nl_table[unit].cb_mutex = cb_mutex;
1958 nl_table[unit].module = module;
1959 if (cfg) {
1960 nl_table[unit].bind = cfg->bind;
1961 nl_table[unit].unbind = cfg->unbind;
1962 nl_table[unit].flags = cfg->flags;
1963 if (cfg->compare)
1964 nl_table[unit].compare = cfg->compare;
1965 }
1966 nl_table[unit].registered = 1;
1967 } else {
1968 kfree(listeners);
1969 nl_table[unit].registered++;
1970 }
1971 netlink_table_ungrab();
1972 return sk;
1973
1974out_sock_release:
1975 kfree(listeners);
1976 netlink_kernel_release(sk);
1977 return NULL;
1978
1979out_sock_release_nosk:
1980 sock_release(sock);
1981 return NULL;
1982}
1983EXPORT_SYMBOL(__netlink_kernel_create);
1984
1985void
1986netlink_kernel_release(struct sock *sk)
1987{
1988 if (sk == NULL || sk->sk_socket == NULL)
1989 return;
1990
1991 sock_release(sk->sk_socket);
1992}
1993EXPORT_SYMBOL(netlink_kernel_release);
1994
1995int __netlink_change_ngroups(struct sock *sk, unsigned int groups)
1996{
1997 struct listeners *new, *old;
1998 struct netlink_table *tbl = &nl_table[sk->sk_protocol];
1999
2000 if (groups < 32)
2001 groups = 32;
2002
2003 if (NLGRPSZ(tbl->groups) < NLGRPSZ(groups)) {
2004 new = kzalloc(sizeof(*new) + NLGRPSZ(groups), GFP_ATOMIC);
2005 if (!new)
2006 return -ENOMEM;
2007 old = nl_deref_protected(tbl->listeners);
2008 memcpy(new->masks, old->masks, NLGRPSZ(tbl->groups));
2009 rcu_assign_pointer(tbl->listeners, new);
2010
2011 kfree_rcu(old, rcu);
2012 }
2013 tbl->groups = groups;
2014
2015 return 0;
2016}
2017
2018/**
2019 * netlink_change_ngroups - change number of multicast groups
2020 *
2021 * This changes the number of multicast groups that are available
2022 * on a certain netlink family. Note that it is not possible to
2023 * change the number of groups to below 32. Also note that it does
2024 * not implicitly call netlink_clear_multicast_users() when the
2025 * number of groups is reduced.
2026 *
2027 * @sk: The kernel netlink socket, as returned by netlink_kernel_create().
2028 * @groups: The new number of groups.
2029 */
2030int netlink_change_ngroups(struct sock *sk, unsigned int groups)
2031{
2032 int err;
2033
2034 netlink_table_grab();
2035 err = __netlink_change_ngroups(sk, groups);
2036 netlink_table_ungrab();
2037
2038 return err;
2039}
2040
2041void __netlink_clear_multicast_users(struct sock *ksk, unsigned int group)
2042{
2043 struct sock *sk;
2044 struct netlink_table *tbl = &nl_table[ksk->sk_protocol];
2045
2046 sk_for_each_bound(sk, &tbl->mc_list)
2047 netlink_update_socket_mc(nlk_sk(sk), group, 0);
2048}
2049
2050struct nlmsghdr *
2051__nlmsg_put(struct sk_buff *skb, u32 portid, u32 seq, int type, int len, int flags)
2052{
2053 struct nlmsghdr *nlh;
2054 int size = nlmsg_msg_size(len);
2055
2056 nlh = (struct nlmsghdr *)skb_put(skb, NLMSG_ALIGN(size));
2057 nlh->nlmsg_type = type;
2058 nlh->nlmsg_len = size;
2059 nlh->nlmsg_flags = flags;
2060 nlh->nlmsg_pid = portid;
2061 nlh->nlmsg_seq = seq;
2062 if (!__builtin_constant_p(size) || NLMSG_ALIGN(size) - size != 0)
2063 memset(nlmsg_data(nlh) + len, 0, NLMSG_ALIGN(size) - size);
2064 return nlh;
2065}
2066EXPORT_SYMBOL(__nlmsg_put);
2067
2068/*
2069 * It looks a bit ugly.
2070 * It would be better to create kernel thread.
2071 */
2072
2073static int netlink_dump(struct sock *sk)
2074{
2075 struct netlink_sock *nlk = nlk_sk(sk);
2076 struct netlink_callback *cb;
2077 struct sk_buff *skb = NULL;
2078 struct nlmsghdr *nlh;
2079 struct module *module;
2080 int len, err = -ENOBUFS;
2081 int alloc_min_size;
2082 int alloc_size;
2083
2084 mutex_lock(nlk->cb_mutex);
2085 if (!nlk->cb_running) {
2086 err = -EINVAL;
2087 goto errout_skb;
2088 }
2089
2090 if (atomic_read(&sk->sk_rmem_alloc) >= sk->sk_rcvbuf)
2091 goto errout_skb;
2092
2093 /* NLMSG_GOODSIZE is small to avoid high order allocations being
2094 * required, but it makes sense to _attempt_ a 16K bytes allocation
2095 * to reduce number of system calls on dump operations, if user
2096 * ever provided a big enough buffer.
2097 */
2098 cb = &nlk->cb;
2099 alloc_min_size = max_t(int, cb->min_dump_alloc, NLMSG_GOODSIZE);
2100
2101 if (alloc_min_size < nlk->max_recvmsg_len) {
2102 alloc_size = nlk->max_recvmsg_len;
2103 skb = alloc_skb(alloc_size,
2104 (GFP_KERNEL & ~__GFP_DIRECT_RECLAIM) |
2105 __GFP_NOWARN | __GFP_NORETRY);
2106 }
2107 if (!skb) {
2108 alloc_size = alloc_min_size;
2109 skb = alloc_skb(alloc_size, GFP_KERNEL);
2110 }
2111 if (!skb)
2112 goto errout_skb;
2113
2114 /* Trim skb to allocated size. User is expected to provide buffer as
2115 * large as max(min_dump_alloc, 16KiB (mac_recvmsg_len capped at
2116 * netlink_recvmsg())). dump will pack as many smaller messages as
2117 * could fit within the allocated skb. skb is typically allocated
2118 * with larger space than required (could be as much as near 2x the
2119 * requested size with align to next power of 2 approach). Allowing
2120 * dump to use the excess space makes it difficult for a user to have a
2121 * reasonable static buffer based on the expected largest dump of a
2122 * single netdev. The outcome is MSG_TRUNC error.
2123 */
2124 skb_reserve(skb, skb_tailroom(skb) - alloc_size);
2125 netlink_skb_set_owner_r(skb, sk);
2126
2127 len = cb->dump(skb, cb);
2128
2129 if (len > 0) {
2130 mutex_unlock(nlk->cb_mutex);
2131
2132 if (sk_filter(sk, skb))
2133 kfree_skb(skb);
2134 else
2135 __netlink_sendskb(sk, skb);
2136 return 0;
2137 }
2138
2139 nlh = nlmsg_put_answer(skb, cb, NLMSG_DONE, sizeof(len), NLM_F_MULTI);
2140 if (!nlh)
2141 goto errout_skb;
2142
2143 nl_dump_check_consistent(cb, nlh);
2144
2145 memcpy(nlmsg_data(nlh), &len, sizeof(len));
2146
2147 if (sk_filter(sk, skb))
2148 kfree_skb(skb);
2149 else
2150 __netlink_sendskb(sk, skb);
2151
2152 if (cb->done)
2153 cb->done(cb);
2154
2155 nlk->cb_running = false;
2156 module = cb->module;
2157 skb = cb->skb;
2158 mutex_unlock(nlk->cb_mutex);
2159 module_put(module);
2160 consume_skb(skb);
2161 return 0;
2162
2163errout_skb:
2164 mutex_unlock(nlk->cb_mutex);
2165 kfree_skb(skb);
2166 return err;
2167}
2168
2169int __netlink_dump_start(struct sock *ssk, struct sk_buff *skb,
2170 const struct nlmsghdr *nlh,
2171 struct netlink_dump_control *control)
2172{
2173 struct netlink_callback *cb;
2174 struct sock *sk;
2175 struct netlink_sock *nlk;
2176 int ret;
2177
2178 atomic_inc(&skb->users);
2179
2180 sk = netlink_lookup(sock_net(ssk), ssk->sk_protocol, NETLINK_CB(skb).portid);
2181 if (sk == NULL) {
2182 ret = -ECONNREFUSED;
2183 goto error_free;
2184 }
2185
2186 nlk = nlk_sk(sk);
2187 mutex_lock(nlk->cb_mutex);
2188 /* A dump is in progress... */
2189 if (nlk->cb_running) {
2190 ret = -EBUSY;
2191 goto error_unlock;
2192 }
2193 /* add reference of module which cb->dump belongs to */
2194 if (!try_module_get(control->module)) {
2195 ret = -EPROTONOSUPPORT;
2196 goto error_unlock;
2197 }
2198
2199 cb = &nlk->cb;
2200 memset(cb, 0, sizeof(*cb));
2201 cb->start = control->start;
2202 cb->dump = control->dump;
2203 cb->done = control->done;
2204 cb->nlh = nlh;
2205 cb->data = control->data;
2206 cb->module = control->module;
2207 cb->min_dump_alloc = control->min_dump_alloc;
2208 cb->skb = skb;
2209
2210 nlk->cb_running = true;
2211
2212 mutex_unlock(nlk->cb_mutex);
2213
2214 if (cb->start)
2215 cb->start(cb);
2216
2217 ret = netlink_dump(sk);
2218 sock_put(sk);
2219
2220 if (ret)
2221 return ret;
2222
2223 /* We successfully started a dump, by returning -EINTR we
2224 * signal not to send ACK even if it was requested.
2225 */
2226 return -EINTR;
2227
2228error_unlock:
2229 sock_put(sk);
2230 mutex_unlock(nlk->cb_mutex);
2231error_free:
2232 kfree_skb(skb);
2233 return ret;
2234}
2235EXPORT_SYMBOL(__netlink_dump_start);
2236
2237void netlink_ack(struct sk_buff *in_skb, struct nlmsghdr *nlh, int err)
2238{
2239 struct sk_buff *skb;
2240 struct nlmsghdr *rep;
2241 struct nlmsgerr *errmsg;
2242 size_t payload = sizeof(*errmsg);
2243 struct netlink_sock *nlk = nlk_sk(NETLINK_CB(in_skb).sk);
2244
2245 /* Error messages get the original request appened, unless the user
2246 * requests to cap the error message.
2247 */
2248 if (!(nlk->flags & NETLINK_F_CAP_ACK) && err)
2249 payload += nlmsg_len(nlh);
2250
2251 skb = nlmsg_new(payload, GFP_KERNEL);
2252 if (!skb) {
2253 struct sock *sk;
2254
2255 sk = netlink_lookup(sock_net(in_skb->sk),
2256 in_skb->sk->sk_protocol,
2257 NETLINK_CB(in_skb).portid);
2258 if (sk) {
2259 sk->sk_err = ENOBUFS;
2260 sk->sk_error_report(sk);
2261 sock_put(sk);
2262 }
2263 return;
2264 }
2265
2266 rep = __nlmsg_put(skb, NETLINK_CB(in_skb).portid, nlh->nlmsg_seq,
2267 NLMSG_ERROR, payload, 0);
2268 errmsg = nlmsg_data(rep);
2269 errmsg->error = err;
2270 memcpy(&errmsg->msg, nlh, payload > sizeof(*errmsg) ? nlh->nlmsg_len : sizeof(*nlh));
2271 netlink_unicast(in_skb->sk, skb, NETLINK_CB(in_skb).portid, MSG_DONTWAIT);
2272}
2273EXPORT_SYMBOL(netlink_ack);
2274
2275int netlink_rcv_skb(struct sk_buff *skb, int (*cb)(struct sk_buff *,
2276 struct nlmsghdr *))
2277{
2278 struct nlmsghdr *nlh;
2279 int err;
2280
2281 while (skb->len >= nlmsg_total_size(0)) {
2282 int msglen;
2283
2284 nlh = nlmsg_hdr(skb);
2285 err = 0;
2286
2287 if (nlh->nlmsg_len < NLMSG_HDRLEN || skb->len < nlh->nlmsg_len)
2288 return 0;
2289
2290 /* Only requests are handled by the kernel */
2291 if (!(nlh->nlmsg_flags & NLM_F_REQUEST))
2292 goto ack;
2293
2294 /* Skip control messages */
2295 if (nlh->nlmsg_type < NLMSG_MIN_TYPE)
2296 goto ack;
2297
2298 err = cb(skb, nlh);
2299 if (err == -EINTR)
2300 goto skip;
2301
2302ack:
2303 if (nlh->nlmsg_flags & NLM_F_ACK || err)
2304 netlink_ack(skb, nlh, err);
2305
2306skip:
2307 msglen = NLMSG_ALIGN(nlh->nlmsg_len);
2308 if (msglen > skb->len)
2309 msglen = skb->len;
2310 skb_pull(skb, msglen);
2311 }
2312
2313 return 0;
2314}
2315EXPORT_SYMBOL(netlink_rcv_skb);
2316
2317/**
2318 * nlmsg_notify - send a notification netlink message
2319 * @sk: netlink socket to use
2320 * @skb: notification message
2321 * @portid: destination netlink portid for reports or 0
2322 * @group: destination multicast group or 0
2323 * @report: 1 to report back, 0 to disable
2324 * @flags: allocation flags
2325 */
2326int nlmsg_notify(struct sock *sk, struct sk_buff *skb, u32 portid,
2327 unsigned int group, int report, gfp_t flags)
2328{
2329 int err = 0;
2330
2331 if (group) {
2332 int exclude_portid = 0;
2333
2334 if (report) {
2335 atomic_inc(&skb->users);
2336 exclude_portid = portid;
2337 }
2338
2339 /* errors reported via destination sk->sk_err, but propagate
2340 * delivery errors if NETLINK_BROADCAST_ERROR flag is set */
2341 err = nlmsg_multicast(sk, skb, exclude_portid, group, flags);
2342 }
2343
2344 if (report) {
2345 int err2;
2346
2347 err2 = nlmsg_unicast(sk, skb, portid);
2348 if (!err || err == -ESRCH)
2349 err = err2;
2350 }
2351
2352 return err;
2353}
2354EXPORT_SYMBOL(nlmsg_notify);
2355
2356#ifdef CONFIG_PROC_FS
2357struct nl_seq_iter {
2358 struct seq_net_private p;
2359 struct rhashtable_iter hti;
2360 int link;
2361};
2362
2363static int netlink_walk_start(struct nl_seq_iter *iter)
2364{
2365 int err;
2366
2367 err = rhashtable_walk_init(&nl_table[iter->link].hash, &iter->hti,
2368 GFP_KERNEL);
2369 if (err) {
2370 iter->link = MAX_LINKS;
2371 return err;
2372 }
2373
2374 err = rhashtable_walk_start(&iter->hti);
2375 return err == -EAGAIN ? 0 : err;
2376}
2377
2378static void netlink_walk_stop(struct nl_seq_iter *iter)
2379{
2380 rhashtable_walk_stop(&iter->hti);
2381 rhashtable_walk_exit(&iter->hti);
2382}
2383
2384static void *__netlink_seq_next(struct seq_file *seq)
2385{
2386 struct nl_seq_iter *iter = seq->private;
2387 struct netlink_sock *nlk;
2388
2389 do {
2390 for (;;) {
2391 int err;
2392
2393 nlk = rhashtable_walk_next(&iter->hti);
2394
2395 if (IS_ERR(nlk)) {
2396 if (PTR_ERR(nlk) == -EAGAIN)
2397 continue;
2398
2399 return nlk;
2400 }
2401
2402 if (nlk)
2403 break;
2404
2405 netlink_walk_stop(iter);
2406 if (++iter->link >= MAX_LINKS)
2407 return NULL;
2408
2409 err = netlink_walk_start(iter);
2410 if (err)
2411 return ERR_PTR(err);
2412 }
2413 } while (sock_net(&nlk->sk) != seq_file_net(seq));
2414
2415 return nlk;
2416}
2417
2418static void *netlink_seq_start(struct seq_file *seq, loff_t *posp)
2419{
2420 struct nl_seq_iter *iter = seq->private;
2421 void *obj = SEQ_START_TOKEN;
2422 loff_t pos;
2423 int err;
2424
2425 iter->link = 0;
2426
2427 err = netlink_walk_start(iter);
2428 if (err)
2429 return ERR_PTR(err);
2430
2431 for (pos = *posp; pos && obj && !IS_ERR(obj); pos--)
2432 obj = __netlink_seq_next(seq);
2433
2434 return obj;
2435}
2436
2437static void *netlink_seq_next(struct seq_file *seq, void *v, loff_t *pos)
2438{
2439 ++*pos;
2440 return __netlink_seq_next(seq);
2441}
2442
2443static void netlink_seq_stop(struct seq_file *seq, void *v)
2444{
2445 struct nl_seq_iter *iter = seq->private;
2446
2447 if (iter->link >= MAX_LINKS)
2448 return;
2449
2450 netlink_walk_stop(iter);
2451}
2452
2453
2454static int netlink_seq_show(struct seq_file *seq, void *v)
2455{
2456 if (v == SEQ_START_TOKEN) {
2457 seq_puts(seq,
2458 "sk Eth Pid Groups "
2459 "Rmem Wmem Dump Locks Drops Inode\n");
2460 } else {
2461 struct sock *s = v;
2462 struct netlink_sock *nlk = nlk_sk(s);
2463
2464 seq_printf(seq, "%pK %-3d %-6u %08x %-8d %-8d %d %-8d %-8d %-8lu\n",
2465 s,
2466 s->sk_protocol,
2467 nlk->portid,
2468 nlk->groups ? (u32)nlk->groups[0] : 0,
2469 sk_rmem_alloc_get(s),
2470 sk_wmem_alloc_get(s),
2471 nlk->cb_running,
2472 atomic_read(&s->sk_refcnt),
2473 atomic_read(&s->sk_drops),
2474 sock_i_ino(s)
2475 );
2476
2477 }
2478 return 0;
2479}
2480
2481static const struct seq_operations netlink_seq_ops = {
2482 .start = netlink_seq_start,
2483 .next = netlink_seq_next,
2484 .stop = netlink_seq_stop,
2485 .show = netlink_seq_show,
2486};
2487
2488
2489static int netlink_seq_open(struct inode *inode, struct file *file)
2490{
2491 return seq_open_net(inode, file, &netlink_seq_ops,
2492 sizeof(struct nl_seq_iter));
2493}
2494
2495static const struct file_operations netlink_seq_fops = {
2496 .owner = THIS_MODULE,
2497 .open = netlink_seq_open,
2498 .read = seq_read,
2499 .llseek = seq_lseek,
2500 .release = seq_release_net,
2501};
2502
2503#endif
2504
2505int netlink_register_notifier(struct notifier_block *nb)
2506{
2507 return blocking_notifier_chain_register(&netlink_chain, nb);
2508}
2509EXPORT_SYMBOL(netlink_register_notifier);
2510
2511int netlink_unregister_notifier(struct notifier_block *nb)
2512{
2513 return blocking_notifier_chain_unregister(&netlink_chain, nb);
2514}
2515EXPORT_SYMBOL(netlink_unregister_notifier);
2516
2517static const struct proto_ops netlink_ops = {
2518 .family = PF_NETLINK,
2519 .owner = THIS_MODULE,
2520 .release = netlink_release,
2521 .bind = netlink_bind,
2522 .connect = netlink_connect,
2523 .socketpair = sock_no_socketpair,
2524 .accept = sock_no_accept,
2525 .getname = netlink_getname,
2526 .poll = datagram_poll,
2527 .ioctl = netlink_ioctl,
2528 .listen = sock_no_listen,
2529 .shutdown = sock_no_shutdown,
2530 .setsockopt = netlink_setsockopt,
2531 .getsockopt = netlink_getsockopt,
2532 .sendmsg = netlink_sendmsg,
2533 .recvmsg = netlink_recvmsg,
2534 .mmap = sock_no_mmap,
2535 .sendpage = sock_no_sendpage,
2536};
2537
2538static const struct net_proto_family netlink_family_ops = {
2539 .family = PF_NETLINK,
2540 .create = netlink_create,
2541 .owner = THIS_MODULE, /* for consistency 8) */
2542};
2543
2544static int __net_init netlink_net_init(struct net *net)
2545{
2546#ifdef CONFIG_PROC_FS
2547 if (!proc_create("netlink", 0, net->proc_net, &netlink_seq_fops))
2548 return -ENOMEM;
2549#endif
2550 return 0;
2551}
2552
2553static void __net_exit netlink_net_exit(struct net *net)
2554{
2555#ifdef CONFIG_PROC_FS
2556 remove_proc_entry("netlink", net->proc_net);
2557#endif
2558}
2559
2560static void __init netlink_add_usersock_entry(void)
2561{
2562 struct listeners *listeners;
2563 int groups = 32;
2564
2565 listeners = kzalloc(sizeof(*listeners) + NLGRPSZ(groups), GFP_KERNEL);
2566 if (!listeners)
2567 panic("netlink_add_usersock_entry: Cannot allocate listeners\n");
2568
2569 netlink_table_grab();
2570
2571 nl_table[NETLINK_USERSOCK].groups = groups;
2572 rcu_assign_pointer(nl_table[NETLINK_USERSOCK].listeners, listeners);
2573 nl_table[NETLINK_USERSOCK].module = THIS_MODULE;
2574 nl_table[NETLINK_USERSOCK].registered = 1;
2575 nl_table[NETLINK_USERSOCK].flags = NL_CFG_F_NONROOT_SEND;
2576
2577 netlink_table_ungrab();
2578}
2579
2580static struct pernet_operations __net_initdata netlink_net_ops = {
2581 .init = netlink_net_init,
2582 .exit = netlink_net_exit,
2583};
2584
2585static inline u32 netlink_hash(const void *data, u32 len, u32 seed)
2586{
2587 const struct netlink_sock *nlk = data;
2588 struct netlink_compare_arg arg;
2589
2590 netlink_compare_arg_init(&arg, sock_net(&nlk->sk), nlk->portid);
2591 return jhash2((u32 *)&arg, netlink_compare_arg_len / sizeof(u32), seed);
2592}
2593
2594static const struct rhashtable_params netlink_rhashtable_params = {
2595 .head_offset = offsetof(struct netlink_sock, node),
2596 .key_len = netlink_compare_arg_len,
2597 .obj_hashfn = netlink_hash,
2598 .obj_cmpfn = netlink_compare,
2599 .automatic_shrinking = true,
2600};
2601
2602static int __init netlink_proto_init(void)
2603{
2604 int i;
2605 int err = proto_register(&netlink_proto, 0);
2606
2607 if (err != 0)
2608 goto out;
2609
2610 BUILD_BUG_ON(sizeof(struct netlink_skb_parms) > FIELD_SIZEOF(struct sk_buff, cb));
2611
2612 nl_table = kcalloc(MAX_LINKS, sizeof(*nl_table), GFP_KERNEL);
2613 if (!nl_table)
2614 goto panic;
2615
2616 for (i = 0; i < MAX_LINKS; i++) {
2617 if (rhashtable_init(&nl_table[i].hash,
2618 &netlink_rhashtable_params) < 0) {
2619 while (--i > 0)
2620 rhashtable_destroy(&nl_table[i].hash);
2621 kfree(nl_table);
2622 goto panic;
2623 }
2624 }
2625
2626 INIT_LIST_HEAD(&netlink_tap_all);
2627
2628 netlink_add_usersock_entry();
2629
2630 sock_register(&netlink_family_ops);
2631 register_pernet_subsys(&netlink_net_ops);
2632 /* The netlink device handler may be needed early. */
2633 rtnetlink_init();
2634out:
2635 return err;
2636panic:
2637 panic("netlink_init: Cannot allocate nl_table\n");
2638}
2639
2640core_initcall(netlink_proto_init);
1// SPDX-License-Identifier: GPL-2.0-or-later
2/*
3 * NETLINK Kernel-user communication protocol.
4 *
5 * Authors: Alan Cox <alan@lxorguk.ukuu.org.uk>
6 * Alexey Kuznetsov <kuznet@ms2.inr.ac.ru>
7 * Patrick McHardy <kaber@trash.net>
8 *
9 * Tue Jun 26 14:36:48 MEST 2001 Herbert "herp" Rosmanith
10 * added netlink_proto_exit
11 * Tue Jan 22 18:32:44 BRST 2002 Arnaldo C. de Melo <acme@conectiva.com.br>
12 * use nlk_sk, as sk->protinfo is on a diet 8)
13 * Fri Jul 22 19:51:12 MEST 2005 Harald Welte <laforge@gnumonks.org>
14 * - inc module use count of module that owns
15 * the kernel socket in case userspace opens
16 * socket of same protocol
17 * - remove all module support, since netlink is
18 * mandatory if CONFIG_NET=y these days
19 */
20
21#include <linux/module.h>
22
23#include <linux/bpf.h>
24#include <linux/capability.h>
25#include <linux/kernel.h>
26#include <linux/filter.h>
27#include <linux/init.h>
28#include <linux/signal.h>
29#include <linux/sched.h>
30#include <linux/errno.h>
31#include <linux/string.h>
32#include <linux/stat.h>
33#include <linux/socket.h>
34#include <linux/un.h>
35#include <linux/fcntl.h>
36#include <linux/termios.h>
37#include <linux/sockios.h>
38#include <linux/net.h>
39#include <linux/fs.h>
40#include <linux/slab.h>
41#include <linux/uaccess.h>
42#include <linux/skbuff.h>
43#include <linux/netdevice.h>
44#include <linux/rtnetlink.h>
45#include <linux/proc_fs.h>
46#include <linux/seq_file.h>
47#include <linux/notifier.h>
48#include <linux/security.h>
49#include <linux/jhash.h>
50#include <linux/jiffies.h>
51#include <linux/random.h>
52#include <linux/bitops.h>
53#include <linux/mm.h>
54#include <linux/types.h>
55#include <linux/audit.h>
56#include <linux/mutex.h>
57#include <linux/vmalloc.h>
58#include <linux/if_arp.h>
59#include <linux/rhashtable.h>
60#include <asm/cacheflush.h>
61#include <linux/hash.h>
62#include <linux/genetlink.h>
63#include <linux/net_namespace.h>
64#include <linux/nospec.h>
65#include <linux/btf_ids.h>
66
67#include <net/net_namespace.h>
68#include <net/netns/generic.h>
69#include <net/sock.h>
70#include <net/scm.h>
71#include <net/netlink.h>
72#define CREATE_TRACE_POINTS
73#include <trace/events/netlink.h>
74
75#include "af_netlink.h"
76
77struct listeners {
78 struct rcu_head rcu;
79 unsigned long masks[];
80};
81
82/* state bits */
83#define NETLINK_S_CONGESTED 0x0
84
85static inline int netlink_is_kernel(struct sock *sk)
86{
87 return nlk_sk(sk)->flags & NETLINK_F_KERNEL_SOCKET;
88}
89
90struct netlink_table *nl_table __read_mostly;
91EXPORT_SYMBOL_GPL(nl_table);
92
93static DECLARE_WAIT_QUEUE_HEAD(nl_table_wait);
94
95static struct lock_class_key nlk_cb_mutex_keys[MAX_LINKS];
96
97static const char *const nlk_cb_mutex_key_strings[MAX_LINKS + 1] = {
98 "nlk_cb_mutex-ROUTE",
99 "nlk_cb_mutex-1",
100 "nlk_cb_mutex-USERSOCK",
101 "nlk_cb_mutex-FIREWALL",
102 "nlk_cb_mutex-SOCK_DIAG",
103 "nlk_cb_mutex-NFLOG",
104 "nlk_cb_mutex-XFRM",
105 "nlk_cb_mutex-SELINUX",
106 "nlk_cb_mutex-ISCSI",
107 "nlk_cb_mutex-AUDIT",
108 "nlk_cb_mutex-FIB_LOOKUP",
109 "nlk_cb_mutex-CONNECTOR",
110 "nlk_cb_mutex-NETFILTER",
111 "nlk_cb_mutex-IP6_FW",
112 "nlk_cb_mutex-DNRTMSG",
113 "nlk_cb_mutex-KOBJECT_UEVENT",
114 "nlk_cb_mutex-GENERIC",
115 "nlk_cb_mutex-17",
116 "nlk_cb_mutex-SCSITRANSPORT",
117 "nlk_cb_mutex-ECRYPTFS",
118 "nlk_cb_mutex-RDMA",
119 "nlk_cb_mutex-CRYPTO",
120 "nlk_cb_mutex-SMC",
121 "nlk_cb_mutex-23",
122 "nlk_cb_mutex-24",
123 "nlk_cb_mutex-25",
124 "nlk_cb_mutex-26",
125 "nlk_cb_mutex-27",
126 "nlk_cb_mutex-28",
127 "nlk_cb_mutex-29",
128 "nlk_cb_mutex-30",
129 "nlk_cb_mutex-31",
130 "nlk_cb_mutex-MAX_LINKS"
131};
132
133static int netlink_dump(struct sock *sk);
134
135/* nl_table locking explained:
136 * Lookup and traversal are protected with an RCU read-side lock. Insertion
137 * and removal are protected with per bucket lock while using RCU list
138 * modification primitives and may run in parallel to RCU protected lookups.
139 * Destruction of the Netlink socket may only occur *after* nl_table_lock has
140 * been acquired * either during or after the socket has been removed from
141 * the list and after an RCU grace period.
142 */
143DEFINE_RWLOCK(nl_table_lock);
144EXPORT_SYMBOL_GPL(nl_table_lock);
145static atomic_t nl_table_users = ATOMIC_INIT(0);
146
147#define nl_deref_protected(X) rcu_dereference_protected(X, lockdep_is_held(&nl_table_lock));
148
149static BLOCKING_NOTIFIER_HEAD(netlink_chain);
150
151
152static const struct rhashtable_params netlink_rhashtable_params;
153
154void do_trace_netlink_extack(const char *msg)
155{
156 trace_netlink_extack(msg);
157}
158EXPORT_SYMBOL(do_trace_netlink_extack);
159
160static inline u32 netlink_group_mask(u32 group)
161{
162 if (group > 32)
163 return 0;
164 return group ? 1 << (group - 1) : 0;
165}
166
167static struct sk_buff *netlink_to_full_skb(const struct sk_buff *skb,
168 gfp_t gfp_mask)
169{
170 unsigned int len = skb_end_offset(skb);
171 struct sk_buff *new;
172
173 new = alloc_skb(len, gfp_mask);
174 if (new == NULL)
175 return NULL;
176
177 NETLINK_CB(new).portid = NETLINK_CB(skb).portid;
178 NETLINK_CB(new).dst_group = NETLINK_CB(skb).dst_group;
179 NETLINK_CB(new).creds = NETLINK_CB(skb).creds;
180
181 skb_put_data(new, skb->data, len);
182 return new;
183}
184
185static unsigned int netlink_tap_net_id;
186
187struct netlink_tap_net {
188 struct list_head netlink_tap_all;
189 struct mutex netlink_tap_lock;
190};
191
192int netlink_add_tap(struct netlink_tap *nt)
193{
194 struct net *net = dev_net(nt->dev);
195 struct netlink_tap_net *nn = net_generic(net, netlink_tap_net_id);
196
197 if (unlikely(nt->dev->type != ARPHRD_NETLINK))
198 return -EINVAL;
199
200 mutex_lock(&nn->netlink_tap_lock);
201 list_add_rcu(&nt->list, &nn->netlink_tap_all);
202 mutex_unlock(&nn->netlink_tap_lock);
203
204 __module_get(nt->module);
205
206 return 0;
207}
208EXPORT_SYMBOL_GPL(netlink_add_tap);
209
210static int __netlink_remove_tap(struct netlink_tap *nt)
211{
212 struct net *net = dev_net(nt->dev);
213 struct netlink_tap_net *nn = net_generic(net, netlink_tap_net_id);
214 bool found = false;
215 struct netlink_tap *tmp;
216
217 mutex_lock(&nn->netlink_tap_lock);
218
219 list_for_each_entry(tmp, &nn->netlink_tap_all, list) {
220 if (nt == tmp) {
221 list_del_rcu(&nt->list);
222 found = true;
223 goto out;
224 }
225 }
226
227 pr_warn("__netlink_remove_tap: %p not found\n", nt);
228out:
229 mutex_unlock(&nn->netlink_tap_lock);
230
231 if (found)
232 module_put(nt->module);
233
234 return found ? 0 : -ENODEV;
235}
236
237int netlink_remove_tap(struct netlink_tap *nt)
238{
239 int ret;
240
241 ret = __netlink_remove_tap(nt);
242 synchronize_net();
243
244 return ret;
245}
246EXPORT_SYMBOL_GPL(netlink_remove_tap);
247
248static __net_init int netlink_tap_init_net(struct net *net)
249{
250 struct netlink_tap_net *nn = net_generic(net, netlink_tap_net_id);
251
252 INIT_LIST_HEAD(&nn->netlink_tap_all);
253 mutex_init(&nn->netlink_tap_lock);
254 return 0;
255}
256
257static struct pernet_operations netlink_tap_net_ops = {
258 .init = netlink_tap_init_net,
259 .id = &netlink_tap_net_id,
260 .size = sizeof(struct netlink_tap_net),
261};
262
263static bool netlink_filter_tap(const struct sk_buff *skb)
264{
265 struct sock *sk = skb->sk;
266
267 /* We take the more conservative approach and
268 * whitelist socket protocols that may pass.
269 */
270 switch (sk->sk_protocol) {
271 case NETLINK_ROUTE:
272 case NETLINK_USERSOCK:
273 case NETLINK_SOCK_DIAG:
274 case NETLINK_NFLOG:
275 case NETLINK_XFRM:
276 case NETLINK_FIB_LOOKUP:
277 case NETLINK_NETFILTER:
278 case NETLINK_GENERIC:
279 return true;
280 }
281
282 return false;
283}
284
285static int __netlink_deliver_tap_skb(struct sk_buff *skb,
286 struct net_device *dev)
287{
288 struct sk_buff *nskb;
289 struct sock *sk = skb->sk;
290 int ret = -ENOMEM;
291
292 if (!net_eq(dev_net(dev), sock_net(sk)))
293 return 0;
294
295 dev_hold(dev);
296
297 if (is_vmalloc_addr(skb->head))
298 nskb = netlink_to_full_skb(skb, GFP_ATOMIC);
299 else
300 nskb = skb_clone(skb, GFP_ATOMIC);
301 if (nskb) {
302 nskb->dev = dev;
303 nskb->protocol = htons((u16) sk->sk_protocol);
304 nskb->pkt_type = netlink_is_kernel(sk) ?
305 PACKET_KERNEL : PACKET_USER;
306 skb_reset_network_header(nskb);
307 ret = dev_queue_xmit(nskb);
308 if (unlikely(ret > 0))
309 ret = net_xmit_errno(ret);
310 }
311
312 dev_put(dev);
313 return ret;
314}
315
316static void __netlink_deliver_tap(struct sk_buff *skb, struct netlink_tap_net *nn)
317{
318 int ret;
319 struct netlink_tap *tmp;
320
321 if (!netlink_filter_tap(skb))
322 return;
323
324 list_for_each_entry_rcu(tmp, &nn->netlink_tap_all, list) {
325 ret = __netlink_deliver_tap_skb(skb, tmp->dev);
326 if (unlikely(ret))
327 break;
328 }
329}
330
331static void netlink_deliver_tap(struct net *net, struct sk_buff *skb)
332{
333 struct netlink_tap_net *nn = net_generic(net, netlink_tap_net_id);
334
335 rcu_read_lock();
336
337 if (unlikely(!list_empty(&nn->netlink_tap_all)))
338 __netlink_deliver_tap(skb, nn);
339
340 rcu_read_unlock();
341}
342
343static void netlink_deliver_tap_kernel(struct sock *dst, struct sock *src,
344 struct sk_buff *skb)
345{
346 if (!(netlink_is_kernel(dst) && netlink_is_kernel(src)))
347 netlink_deliver_tap(sock_net(dst), skb);
348}
349
350static void netlink_overrun(struct sock *sk)
351{
352 struct netlink_sock *nlk = nlk_sk(sk);
353
354 if (!(nlk->flags & NETLINK_F_RECV_NO_ENOBUFS)) {
355 if (!test_and_set_bit(NETLINK_S_CONGESTED,
356 &nlk_sk(sk)->state)) {
357 sk->sk_err = ENOBUFS;
358 sk_error_report(sk);
359 }
360 }
361 atomic_inc(&sk->sk_drops);
362}
363
364static void netlink_rcv_wake(struct sock *sk)
365{
366 struct netlink_sock *nlk = nlk_sk(sk);
367
368 if (skb_queue_empty_lockless(&sk->sk_receive_queue))
369 clear_bit(NETLINK_S_CONGESTED, &nlk->state);
370 if (!test_bit(NETLINK_S_CONGESTED, &nlk->state))
371 wake_up_interruptible(&nlk->wait);
372}
373
374static void netlink_skb_destructor(struct sk_buff *skb)
375{
376 if (is_vmalloc_addr(skb->head)) {
377 if (!skb->cloned ||
378 !atomic_dec_return(&(skb_shinfo(skb)->dataref)))
379 vfree(skb->head);
380
381 skb->head = NULL;
382 }
383 if (skb->sk != NULL)
384 sock_rfree(skb);
385}
386
387static void netlink_skb_set_owner_r(struct sk_buff *skb, struct sock *sk)
388{
389 WARN_ON(skb->sk != NULL);
390 skb->sk = sk;
391 skb->destructor = netlink_skb_destructor;
392 atomic_add(skb->truesize, &sk->sk_rmem_alloc);
393 sk_mem_charge(sk, skb->truesize);
394}
395
396static void netlink_sock_destruct(struct sock *sk)
397{
398 struct netlink_sock *nlk = nlk_sk(sk);
399
400 if (nlk->cb_running) {
401 if (nlk->cb.done)
402 nlk->cb.done(&nlk->cb);
403 module_put(nlk->cb.module);
404 kfree_skb(nlk->cb.skb);
405 }
406
407 skb_queue_purge(&sk->sk_receive_queue);
408
409 if (!sock_flag(sk, SOCK_DEAD)) {
410 printk(KERN_ERR "Freeing alive netlink socket %p\n", sk);
411 return;
412 }
413
414 WARN_ON(atomic_read(&sk->sk_rmem_alloc));
415 WARN_ON(refcount_read(&sk->sk_wmem_alloc));
416 WARN_ON(nlk_sk(sk)->groups);
417}
418
419static void netlink_sock_destruct_work(struct work_struct *work)
420{
421 struct netlink_sock *nlk = container_of(work, struct netlink_sock,
422 work);
423
424 sk_free(&nlk->sk);
425}
426
427/* This lock without WQ_FLAG_EXCLUSIVE is good on UP and it is _very_ bad on
428 * SMP. Look, when several writers sleep and reader wakes them up, all but one
429 * immediately hit write lock and grab all the cpus. Exclusive sleep solves
430 * this, _but_ remember, it adds useless work on UP machines.
431 */
432
433void netlink_table_grab(void)
434 __acquires(nl_table_lock)
435{
436 might_sleep();
437
438 write_lock_irq(&nl_table_lock);
439
440 if (atomic_read(&nl_table_users)) {
441 DECLARE_WAITQUEUE(wait, current);
442
443 add_wait_queue_exclusive(&nl_table_wait, &wait);
444 for (;;) {
445 set_current_state(TASK_UNINTERRUPTIBLE);
446 if (atomic_read(&nl_table_users) == 0)
447 break;
448 write_unlock_irq(&nl_table_lock);
449 schedule();
450 write_lock_irq(&nl_table_lock);
451 }
452
453 __set_current_state(TASK_RUNNING);
454 remove_wait_queue(&nl_table_wait, &wait);
455 }
456}
457
458void netlink_table_ungrab(void)
459 __releases(nl_table_lock)
460{
461 write_unlock_irq(&nl_table_lock);
462 wake_up(&nl_table_wait);
463}
464
465static inline void
466netlink_lock_table(void)
467{
468 unsigned long flags;
469
470 /* read_lock() synchronizes us to netlink_table_grab */
471
472 read_lock_irqsave(&nl_table_lock, flags);
473 atomic_inc(&nl_table_users);
474 read_unlock_irqrestore(&nl_table_lock, flags);
475}
476
477static inline void
478netlink_unlock_table(void)
479{
480 if (atomic_dec_and_test(&nl_table_users))
481 wake_up(&nl_table_wait);
482}
483
484struct netlink_compare_arg
485{
486 possible_net_t pnet;
487 u32 portid;
488};
489
490/* Doing sizeof directly may yield 4 extra bytes on 64-bit. */
491#define netlink_compare_arg_len \
492 (offsetof(struct netlink_compare_arg, portid) + sizeof(u32))
493
494static inline int netlink_compare(struct rhashtable_compare_arg *arg,
495 const void *ptr)
496{
497 const struct netlink_compare_arg *x = arg->key;
498 const struct netlink_sock *nlk = ptr;
499
500 return nlk->portid != x->portid ||
501 !net_eq(sock_net(&nlk->sk), read_pnet(&x->pnet));
502}
503
504static void netlink_compare_arg_init(struct netlink_compare_arg *arg,
505 struct net *net, u32 portid)
506{
507 memset(arg, 0, sizeof(*arg));
508 write_pnet(&arg->pnet, net);
509 arg->portid = portid;
510}
511
512static struct sock *__netlink_lookup(struct netlink_table *table, u32 portid,
513 struct net *net)
514{
515 struct netlink_compare_arg arg;
516
517 netlink_compare_arg_init(&arg, net, portid);
518 return rhashtable_lookup_fast(&table->hash, &arg,
519 netlink_rhashtable_params);
520}
521
522static int __netlink_insert(struct netlink_table *table, struct sock *sk)
523{
524 struct netlink_compare_arg arg;
525
526 netlink_compare_arg_init(&arg, sock_net(sk), nlk_sk(sk)->portid);
527 return rhashtable_lookup_insert_key(&table->hash, &arg,
528 &nlk_sk(sk)->node,
529 netlink_rhashtable_params);
530}
531
532static struct sock *netlink_lookup(struct net *net, int protocol, u32 portid)
533{
534 struct netlink_table *table = &nl_table[protocol];
535 struct sock *sk;
536
537 rcu_read_lock();
538 sk = __netlink_lookup(table, portid, net);
539 if (sk)
540 sock_hold(sk);
541 rcu_read_unlock();
542
543 return sk;
544}
545
546static const struct proto_ops netlink_ops;
547
548static void
549netlink_update_listeners(struct sock *sk)
550{
551 struct netlink_table *tbl = &nl_table[sk->sk_protocol];
552 unsigned long mask;
553 unsigned int i;
554 struct listeners *listeners;
555
556 listeners = nl_deref_protected(tbl->listeners);
557 if (!listeners)
558 return;
559
560 for (i = 0; i < NLGRPLONGS(tbl->groups); i++) {
561 mask = 0;
562 sk_for_each_bound(sk, &tbl->mc_list) {
563 if (i < NLGRPLONGS(nlk_sk(sk)->ngroups))
564 mask |= nlk_sk(sk)->groups[i];
565 }
566 listeners->masks[i] = mask;
567 }
568 /* this function is only called with the netlink table "grabbed", which
569 * makes sure updates are visible before bind or setsockopt return. */
570}
571
572static int netlink_insert(struct sock *sk, u32 portid)
573{
574 struct netlink_table *table = &nl_table[sk->sk_protocol];
575 int err;
576
577 lock_sock(sk);
578
579 err = nlk_sk(sk)->portid == portid ? 0 : -EBUSY;
580 if (nlk_sk(sk)->bound)
581 goto err;
582
583 /* portid can be read locklessly from netlink_getname(). */
584 WRITE_ONCE(nlk_sk(sk)->portid, portid);
585
586 sock_hold(sk);
587
588 err = __netlink_insert(table, sk);
589 if (err) {
590 /* In case the hashtable backend returns with -EBUSY
591 * from here, it must not escape to the caller.
592 */
593 if (unlikely(err == -EBUSY))
594 err = -EOVERFLOW;
595 if (err == -EEXIST)
596 err = -EADDRINUSE;
597 sock_put(sk);
598 goto err;
599 }
600
601 /* We need to ensure that the socket is hashed and visible. */
602 smp_wmb();
603 /* Paired with lockless reads from netlink_bind(),
604 * netlink_connect() and netlink_sendmsg().
605 */
606 WRITE_ONCE(nlk_sk(sk)->bound, portid);
607
608err:
609 release_sock(sk);
610 return err;
611}
612
613static void netlink_remove(struct sock *sk)
614{
615 struct netlink_table *table;
616
617 table = &nl_table[sk->sk_protocol];
618 if (!rhashtable_remove_fast(&table->hash, &nlk_sk(sk)->node,
619 netlink_rhashtable_params)) {
620 WARN_ON(refcount_read(&sk->sk_refcnt) == 1);
621 __sock_put(sk);
622 }
623
624 netlink_table_grab();
625 if (nlk_sk(sk)->subscriptions) {
626 __sk_del_bind_node(sk);
627 netlink_update_listeners(sk);
628 }
629 if (sk->sk_protocol == NETLINK_GENERIC)
630 atomic_inc(&genl_sk_destructing_cnt);
631 netlink_table_ungrab();
632}
633
634static struct proto netlink_proto = {
635 .name = "NETLINK",
636 .owner = THIS_MODULE,
637 .obj_size = sizeof(struct netlink_sock),
638};
639
640static int __netlink_create(struct net *net, struct socket *sock,
641 struct mutex *cb_mutex, int protocol,
642 int kern)
643{
644 struct sock *sk;
645 struct netlink_sock *nlk;
646
647 sock->ops = &netlink_ops;
648
649 sk = sk_alloc(net, PF_NETLINK, GFP_KERNEL, &netlink_proto, kern);
650 if (!sk)
651 return -ENOMEM;
652
653 sock_init_data(sock, sk);
654
655 nlk = nlk_sk(sk);
656 if (cb_mutex) {
657 nlk->cb_mutex = cb_mutex;
658 } else {
659 nlk->cb_mutex = &nlk->cb_def_mutex;
660 mutex_init(nlk->cb_mutex);
661 lockdep_set_class_and_name(nlk->cb_mutex,
662 nlk_cb_mutex_keys + protocol,
663 nlk_cb_mutex_key_strings[protocol]);
664 }
665 init_waitqueue_head(&nlk->wait);
666
667 sk->sk_destruct = netlink_sock_destruct;
668 sk->sk_protocol = protocol;
669 return 0;
670}
671
672static int netlink_create(struct net *net, struct socket *sock, int protocol,
673 int kern)
674{
675 struct module *module = NULL;
676 struct mutex *cb_mutex;
677 struct netlink_sock *nlk;
678 int (*bind)(struct net *net, int group);
679 void (*unbind)(struct net *net, int group);
680 int err = 0;
681
682 sock->state = SS_UNCONNECTED;
683
684 if (sock->type != SOCK_RAW && sock->type != SOCK_DGRAM)
685 return -ESOCKTNOSUPPORT;
686
687 if (protocol < 0 || protocol >= MAX_LINKS)
688 return -EPROTONOSUPPORT;
689 protocol = array_index_nospec(protocol, MAX_LINKS);
690
691 netlink_lock_table();
692#ifdef CONFIG_MODULES
693 if (!nl_table[protocol].registered) {
694 netlink_unlock_table();
695 request_module("net-pf-%d-proto-%d", PF_NETLINK, protocol);
696 netlink_lock_table();
697 }
698#endif
699 if (nl_table[protocol].registered &&
700 try_module_get(nl_table[protocol].module))
701 module = nl_table[protocol].module;
702 else
703 err = -EPROTONOSUPPORT;
704 cb_mutex = nl_table[protocol].cb_mutex;
705 bind = nl_table[protocol].bind;
706 unbind = nl_table[protocol].unbind;
707 netlink_unlock_table();
708
709 if (err < 0)
710 goto out;
711
712 err = __netlink_create(net, sock, cb_mutex, protocol, kern);
713 if (err < 0)
714 goto out_module;
715
716 sock_prot_inuse_add(net, &netlink_proto, 1);
717
718 nlk = nlk_sk(sock->sk);
719 nlk->module = module;
720 nlk->netlink_bind = bind;
721 nlk->netlink_unbind = unbind;
722out:
723 return err;
724
725out_module:
726 module_put(module);
727 goto out;
728}
729
730static void deferred_put_nlk_sk(struct rcu_head *head)
731{
732 struct netlink_sock *nlk = container_of(head, struct netlink_sock, rcu);
733 struct sock *sk = &nlk->sk;
734
735 kfree(nlk->groups);
736 nlk->groups = NULL;
737
738 if (!refcount_dec_and_test(&sk->sk_refcnt))
739 return;
740
741 if (nlk->cb_running && nlk->cb.done) {
742 INIT_WORK(&nlk->work, netlink_sock_destruct_work);
743 schedule_work(&nlk->work);
744 return;
745 }
746
747 sk_free(sk);
748}
749
750static int netlink_release(struct socket *sock)
751{
752 struct sock *sk = sock->sk;
753 struct netlink_sock *nlk;
754
755 if (!sk)
756 return 0;
757
758 netlink_remove(sk);
759 sock_orphan(sk);
760 nlk = nlk_sk(sk);
761
762 /*
763 * OK. Socket is unlinked, any packets that arrive now
764 * will be purged.
765 */
766
767 /* must not acquire netlink_table_lock in any way again before unbind
768 * and notifying genetlink is done as otherwise it might deadlock
769 */
770 if (nlk->netlink_unbind) {
771 int i;
772
773 for (i = 0; i < nlk->ngroups; i++)
774 if (test_bit(i, nlk->groups))
775 nlk->netlink_unbind(sock_net(sk), i + 1);
776 }
777 if (sk->sk_protocol == NETLINK_GENERIC &&
778 atomic_dec_return(&genl_sk_destructing_cnt) == 0)
779 wake_up(&genl_sk_destructing_waitq);
780
781 sock->sk = NULL;
782 wake_up_interruptible_all(&nlk->wait);
783
784 skb_queue_purge(&sk->sk_write_queue);
785
786 if (nlk->portid && nlk->bound) {
787 struct netlink_notify n = {
788 .net = sock_net(sk),
789 .protocol = sk->sk_protocol,
790 .portid = nlk->portid,
791 };
792 blocking_notifier_call_chain(&netlink_chain,
793 NETLINK_URELEASE, &n);
794 }
795
796 module_put(nlk->module);
797
798 if (netlink_is_kernel(sk)) {
799 netlink_table_grab();
800 BUG_ON(nl_table[sk->sk_protocol].registered == 0);
801 if (--nl_table[sk->sk_protocol].registered == 0) {
802 struct listeners *old;
803
804 old = nl_deref_protected(nl_table[sk->sk_protocol].listeners);
805 RCU_INIT_POINTER(nl_table[sk->sk_protocol].listeners, NULL);
806 kfree_rcu(old, rcu);
807 nl_table[sk->sk_protocol].module = NULL;
808 nl_table[sk->sk_protocol].bind = NULL;
809 nl_table[sk->sk_protocol].unbind = NULL;
810 nl_table[sk->sk_protocol].flags = 0;
811 nl_table[sk->sk_protocol].registered = 0;
812 }
813 netlink_table_ungrab();
814 }
815
816 sock_prot_inuse_add(sock_net(sk), &netlink_proto, -1);
817
818 /* Because struct net might disappear soon, do not keep a pointer. */
819 if (!sk->sk_net_refcnt && sock_net(sk) != &init_net) {
820 __netns_tracker_free(sock_net(sk), &sk->ns_tracker, false);
821 /* Because of deferred_put_nlk_sk and use of work queue,
822 * it is possible netns will be freed before this socket.
823 */
824 sock_net_set(sk, &init_net);
825 __netns_tracker_alloc(&init_net, &sk->ns_tracker,
826 false, GFP_KERNEL);
827 }
828 call_rcu(&nlk->rcu, deferred_put_nlk_sk);
829 return 0;
830}
831
832static int netlink_autobind(struct socket *sock)
833{
834 struct sock *sk = sock->sk;
835 struct net *net = sock_net(sk);
836 struct netlink_table *table = &nl_table[sk->sk_protocol];
837 s32 portid = task_tgid_vnr(current);
838 int err;
839 s32 rover = -4096;
840 bool ok;
841
842retry:
843 cond_resched();
844 rcu_read_lock();
845 ok = !__netlink_lookup(table, portid, net);
846 rcu_read_unlock();
847 if (!ok) {
848 /* Bind collision, search negative portid values. */
849 if (rover == -4096)
850 /* rover will be in range [S32_MIN, -4097] */
851 rover = S32_MIN + get_random_u32_below(-4096 - S32_MIN);
852 else if (rover >= -4096)
853 rover = -4097;
854 portid = rover--;
855 goto retry;
856 }
857
858 err = netlink_insert(sk, portid);
859 if (err == -EADDRINUSE)
860 goto retry;
861
862 /* If 2 threads race to autobind, that is fine. */
863 if (err == -EBUSY)
864 err = 0;
865
866 return err;
867}
868
869/**
870 * __netlink_ns_capable - General netlink message capability test
871 * @nsp: NETLINK_CB of the socket buffer holding a netlink command from userspace.
872 * @user_ns: The user namespace of the capability to use
873 * @cap: The capability to use
874 *
875 * Test to see if the opener of the socket we received the message
876 * from had when the netlink socket was created and the sender of the
877 * message has the capability @cap in the user namespace @user_ns.
878 */
879bool __netlink_ns_capable(const struct netlink_skb_parms *nsp,
880 struct user_namespace *user_ns, int cap)
881{
882 return ((nsp->flags & NETLINK_SKB_DST) ||
883 file_ns_capable(nsp->sk->sk_socket->file, user_ns, cap)) &&
884 ns_capable(user_ns, cap);
885}
886EXPORT_SYMBOL(__netlink_ns_capable);
887
888/**
889 * netlink_ns_capable - General netlink message capability test
890 * @skb: socket buffer holding a netlink command from userspace
891 * @user_ns: The user namespace of the capability to use
892 * @cap: The capability to use
893 *
894 * Test to see if the opener of the socket we received the message
895 * from had when the netlink socket was created and the sender of the
896 * message has the capability @cap in the user namespace @user_ns.
897 */
898bool netlink_ns_capable(const struct sk_buff *skb,
899 struct user_namespace *user_ns, int cap)
900{
901 return __netlink_ns_capable(&NETLINK_CB(skb), user_ns, cap);
902}
903EXPORT_SYMBOL(netlink_ns_capable);
904
905/**
906 * netlink_capable - Netlink global message capability test
907 * @skb: socket buffer holding a netlink command from userspace
908 * @cap: The capability to use
909 *
910 * Test to see if the opener of the socket we received the message
911 * from had when the netlink socket was created and the sender of the
912 * message has the capability @cap in all user namespaces.
913 */
914bool netlink_capable(const struct sk_buff *skb, int cap)
915{
916 return netlink_ns_capable(skb, &init_user_ns, cap);
917}
918EXPORT_SYMBOL(netlink_capable);
919
920/**
921 * netlink_net_capable - Netlink network namespace message capability test
922 * @skb: socket buffer holding a netlink command from userspace
923 * @cap: The capability to use
924 *
925 * Test to see if the opener of the socket we received the message
926 * from had when the netlink socket was created and the sender of the
927 * message has the capability @cap over the network namespace of
928 * the socket we received the message from.
929 */
930bool netlink_net_capable(const struct sk_buff *skb, int cap)
931{
932 return netlink_ns_capable(skb, sock_net(skb->sk)->user_ns, cap);
933}
934EXPORT_SYMBOL(netlink_net_capable);
935
936static inline int netlink_allowed(const struct socket *sock, unsigned int flag)
937{
938 return (nl_table[sock->sk->sk_protocol].flags & flag) ||
939 ns_capable(sock_net(sock->sk)->user_ns, CAP_NET_ADMIN);
940}
941
942static void
943netlink_update_subscriptions(struct sock *sk, unsigned int subscriptions)
944{
945 struct netlink_sock *nlk = nlk_sk(sk);
946
947 if (nlk->subscriptions && !subscriptions)
948 __sk_del_bind_node(sk);
949 else if (!nlk->subscriptions && subscriptions)
950 sk_add_bind_node(sk, &nl_table[sk->sk_protocol].mc_list);
951 nlk->subscriptions = subscriptions;
952}
953
954static int netlink_realloc_groups(struct sock *sk)
955{
956 struct netlink_sock *nlk = nlk_sk(sk);
957 unsigned int groups;
958 unsigned long *new_groups;
959 int err = 0;
960
961 netlink_table_grab();
962
963 groups = nl_table[sk->sk_protocol].groups;
964 if (!nl_table[sk->sk_protocol].registered) {
965 err = -ENOENT;
966 goto out_unlock;
967 }
968
969 if (nlk->ngroups >= groups)
970 goto out_unlock;
971
972 new_groups = krealloc(nlk->groups, NLGRPSZ(groups), GFP_ATOMIC);
973 if (new_groups == NULL) {
974 err = -ENOMEM;
975 goto out_unlock;
976 }
977 memset((char *)new_groups + NLGRPSZ(nlk->ngroups), 0,
978 NLGRPSZ(groups) - NLGRPSZ(nlk->ngroups));
979
980 nlk->groups = new_groups;
981 nlk->ngroups = groups;
982 out_unlock:
983 netlink_table_ungrab();
984 return err;
985}
986
987static void netlink_undo_bind(int group, long unsigned int groups,
988 struct sock *sk)
989{
990 struct netlink_sock *nlk = nlk_sk(sk);
991 int undo;
992
993 if (!nlk->netlink_unbind)
994 return;
995
996 for (undo = 0; undo < group; undo++)
997 if (test_bit(undo, &groups))
998 nlk->netlink_unbind(sock_net(sk), undo + 1);
999}
1000
1001static int netlink_bind(struct socket *sock, struct sockaddr *addr,
1002 int addr_len)
1003{
1004 struct sock *sk = sock->sk;
1005 struct net *net = sock_net(sk);
1006 struct netlink_sock *nlk = nlk_sk(sk);
1007 struct sockaddr_nl *nladdr = (struct sockaddr_nl *)addr;
1008 int err = 0;
1009 unsigned long groups;
1010 bool bound;
1011
1012 if (addr_len < sizeof(struct sockaddr_nl))
1013 return -EINVAL;
1014
1015 if (nladdr->nl_family != AF_NETLINK)
1016 return -EINVAL;
1017 groups = nladdr->nl_groups;
1018
1019 /* Only superuser is allowed to listen multicasts */
1020 if (groups) {
1021 if (!netlink_allowed(sock, NL_CFG_F_NONROOT_RECV))
1022 return -EPERM;
1023 err = netlink_realloc_groups(sk);
1024 if (err)
1025 return err;
1026 }
1027
1028 if (nlk->ngroups < BITS_PER_LONG)
1029 groups &= (1UL << nlk->ngroups) - 1;
1030
1031 /* Paired with WRITE_ONCE() in netlink_insert() */
1032 bound = READ_ONCE(nlk->bound);
1033 if (bound) {
1034 /* Ensure nlk->portid is up-to-date. */
1035 smp_rmb();
1036
1037 if (nladdr->nl_pid != nlk->portid)
1038 return -EINVAL;
1039 }
1040
1041 if (nlk->netlink_bind && groups) {
1042 int group;
1043
1044 /* nl_groups is a u32, so cap the maximum groups we can bind */
1045 for (group = 0; group < BITS_PER_TYPE(u32); group++) {
1046 if (!test_bit(group, &groups))
1047 continue;
1048 err = nlk->netlink_bind(net, group + 1);
1049 if (!err)
1050 continue;
1051 netlink_undo_bind(group, groups, sk);
1052 return err;
1053 }
1054 }
1055
1056 /* No need for barriers here as we return to user-space without
1057 * using any of the bound attributes.
1058 */
1059 netlink_lock_table();
1060 if (!bound) {
1061 err = nladdr->nl_pid ?
1062 netlink_insert(sk, nladdr->nl_pid) :
1063 netlink_autobind(sock);
1064 if (err) {
1065 netlink_undo_bind(BITS_PER_TYPE(u32), groups, sk);
1066 goto unlock;
1067 }
1068 }
1069
1070 if (!groups && (nlk->groups == NULL || !(u32)nlk->groups[0]))
1071 goto unlock;
1072 netlink_unlock_table();
1073
1074 netlink_table_grab();
1075 netlink_update_subscriptions(sk, nlk->subscriptions +
1076 hweight32(groups) -
1077 hweight32(nlk->groups[0]));
1078 nlk->groups[0] = (nlk->groups[0] & ~0xffffffffUL) | groups;
1079 netlink_update_listeners(sk);
1080 netlink_table_ungrab();
1081
1082 return 0;
1083
1084unlock:
1085 netlink_unlock_table();
1086 return err;
1087}
1088
1089static int netlink_connect(struct socket *sock, struct sockaddr *addr,
1090 int alen, int flags)
1091{
1092 int err = 0;
1093 struct sock *sk = sock->sk;
1094 struct netlink_sock *nlk = nlk_sk(sk);
1095 struct sockaddr_nl *nladdr = (struct sockaddr_nl *)addr;
1096
1097 if (alen < sizeof(addr->sa_family))
1098 return -EINVAL;
1099
1100 if (addr->sa_family == AF_UNSPEC) {
1101 /* paired with READ_ONCE() in netlink_getsockbyportid() */
1102 WRITE_ONCE(sk->sk_state, NETLINK_UNCONNECTED);
1103 /* dst_portid and dst_group can be read locklessly */
1104 WRITE_ONCE(nlk->dst_portid, 0);
1105 WRITE_ONCE(nlk->dst_group, 0);
1106 return 0;
1107 }
1108 if (addr->sa_family != AF_NETLINK)
1109 return -EINVAL;
1110
1111 if (alen < sizeof(struct sockaddr_nl))
1112 return -EINVAL;
1113
1114 if ((nladdr->nl_groups || nladdr->nl_pid) &&
1115 !netlink_allowed(sock, NL_CFG_F_NONROOT_SEND))
1116 return -EPERM;
1117
1118 /* No need for barriers here as we return to user-space without
1119 * using any of the bound attributes.
1120 * Paired with WRITE_ONCE() in netlink_insert().
1121 */
1122 if (!READ_ONCE(nlk->bound))
1123 err = netlink_autobind(sock);
1124
1125 if (err == 0) {
1126 /* paired with READ_ONCE() in netlink_getsockbyportid() */
1127 WRITE_ONCE(sk->sk_state, NETLINK_CONNECTED);
1128 /* dst_portid and dst_group can be read locklessly */
1129 WRITE_ONCE(nlk->dst_portid, nladdr->nl_pid);
1130 WRITE_ONCE(nlk->dst_group, ffs(nladdr->nl_groups));
1131 }
1132
1133 return err;
1134}
1135
1136static int netlink_getname(struct socket *sock, struct sockaddr *addr,
1137 int peer)
1138{
1139 struct sock *sk = sock->sk;
1140 struct netlink_sock *nlk = nlk_sk(sk);
1141 DECLARE_SOCKADDR(struct sockaddr_nl *, nladdr, addr);
1142
1143 nladdr->nl_family = AF_NETLINK;
1144 nladdr->nl_pad = 0;
1145
1146 if (peer) {
1147 /* Paired with WRITE_ONCE() in netlink_connect() */
1148 nladdr->nl_pid = READ_ONCE(nlk->dst_portid);
1149 nladdr->nl_groups = netlink_group_mask(READ_ONCE(nlk->dst_group));
1150 } else {
1151 /* Paired with WRITE_ONCE() in netlink_insert() */
1152 nladdr->nl_pid = READ_ONCE(nlk->portid);
1153 netlink_lock_table();
1154 nladdr->nl_groups = nlk->groups ? nlk->groups[0] : 0;
1155 netlink_unlock_table();
1156 }
1157 return sizeof(*nladdr);
1158}
1159
1160static int netlink_ioctl(struct socket *sock, unsigned int cmd,
1161 unsigned long arg)
1162{
1163 /* try to hand this ioctl down to the NIC drivers.
1164 */
1165 return -ENOIOCTLCMD;
1166}
1167
1168static struct sock *netlink_getsockbyportid(struct sock *ssk, u32 portid)
1169{
1170 struct sock *sock;
1171 struct netlink_sock *nlk;
1172
1173 sock = netlink_lookup(sock_net(ssk), ssk->sk_protocol, portid);
1174 if (!sock)
1175 return ERR_PTR(-ECONNREFUSED);
1176
1177 /* Don't bother queuing skb if kernel socket has no input function */
1178 nlk = nlk_sk(sock);
1179 /* dst_portid and sk_state can be changed in netlink_connect() */
1180 if (READ_ONCE(sock->sk_state) == NETLINK_CONNECTED &&
1181 READ_ONCE(nlk->dst_portid) != nlk_sk(ssk)->portid) {
1182 sock_put(sock);
1183 return ERR_PTR(-ECONNREFUSED);
1184 }
1185 return sock;
1186}
1187
1188struct sock *netlink_getsockbyfilp(struct file *filp)
1189{
1190 struct inode *inode = file_inode(filp);
1191 struct sock *sock;
1192
1193 if (!S_ISSOCK(inode->i_mode))
1194 return ERR_PTR(-ENOTSOCK);
1195
1196 sock = SOCKET_I(inode)->sk;
1197 if (sock->sk_family != AF_NETLINK)
1198 return ERR_PTR(-EINVAL);
1199
1200 sock_hold(sock);
1201 return sock;
1202}
1203
1204static struct sk_buff *netlink_alloc_large_skb(unsigned int size,
1205 int broadcast)
1206{
1207 struct sk_buff *skb;
1208 void *data;
1209
1210 if (size <= NLMSG_GOODSIZE || broadcast)
1211 return alloc_skb(size, GFP_KERNEL);
1212
1213 size = SKB_DATA_ALIGN(size) +
1214 SKB_DATA_ALIGN(sizeof(struct skb_shared_info));
1215
1216 data = vmalloc(size);
1217 if (data == NULL)
1218 return NULL;
1219
1220 skb = __build_skb(data, size);
1221 if (skb == NULL)
1222 vfree(data);
1223 else
1224 skb->destructor = netlink_skb_destructor;
1225
1226 return skb;
1227}
1228
1229/*
1230 * Attach a skb to a netlink socket.
1231 * The caller must hold a reference to the destination socket. On error, the
1232 * reference is dropped. The skb is not send to the destination, just all
1233 * all error checks are performed and memory in the queue is reserved.
1234 * Return values:
1235 * < 0: error. skb freed, reference to sock dropped.
1236 * 0: continue
1237 * 1: repeat lookup - reference dropped while waiting for socket memory.
1238 */
1239int netlink_attachskb(struct sock *sk, struct sk_buff *skb,
1240 long *timeo, struct sock *ssk)
1241{
1242 struct netlink_sock *nlk;
1243
1244 nlk = nlk_sk(sk);
1245
1246 if ((atomic_read(&sk->sk_rmem_alloc) > sk->sk_rcvbuf ||
1247 test_bit(NETLINK_S_CONGESTED, &nlk->state))) {
1248 DECLARE_WAITQUEUE(wait, current);
1249 if (!*timeo) {
1250 if (!ssk || netlink_is_kernel(ssk))
1251 netlink_overrun(sk);
1252 sock_put(sk);
1253 kfree_skb(skb);
1254 return -EAGAIN;
1255 }
1256
1257 __set_current_state(TASK_INTERRUPTIBLE);
1258 add_wait_queue(&nlk->wait, &wait);
1259
1260 if ((atomic_read(&sk->sk_rmem_alloc) > sk->sk_rcvbuf ||
1261 test_bit(NETLINK_S_CONGESTED, &nlk->state)) &&
1262 !sock_flag(sk, SOCK_DEAD))
1263 *timeo = schedule_timeout(*timeo);
1264
1265 __set_current_state(TASK_RUNNING);
1266 remove_wait_queue(&nlk->wait, &wait);
1267 sock_put(sk);
1268
1269 if (signal_pending(current)) {
1270 kfree_skb(skb);
1271 return sock_intr_errno(*timeo);
1272 }
1273 return 1;
1274 }
1275 netlink_skb_set_owner_r(skb, sk);
1276 return 0;
1277}
1278
1279static int __netlink_sendskb(struct sock *sk, struct sk_buff *skb)
1280{
1281 int len = skb->len;
1282
1283 netlink_deliver_tap(sock_net(sk), skb);
1284
1285 skb_queue_tail(&sk->sk_receive_queue, skb);
1286 sk->sk_data_ready(sk);
1287 return len;
1288}
1289
1290int netlink_sendskb(struct sock *sk, struct sk_buff *skb)
1291{
1292 int len = __netlink_sendskb(sk, skb);
1293
1294 sock_put(sk);
1295 return len;
1296}
1297
1298void netlink_detachskb(struct sock *sk, struct sk_buff *skb)
1299{
1300 kfree_skb(skb);
1301 sock_put(sk);
1302}
1303
1304static struct sk_buff *netlink_trim(struct sk_buff *skb, gfp_t allocation)
1305{
1306 int delta;
1307
1308 WARN_ON(skb->sk != NULL);
1309 delta = skb->end - skb->tail;
1310 if (is_vmalloc_addr(skb->head) || delta * 2 < skb->truesize)
1311 return skb;
1312
1313 if (skb_shared(skb)) {
1314 struct sk_buff *nskb = skb_clone(skb, allocation);
1315 if (!nskb)
1316 return skb;
1317 consume_skb(skb);
1318 skb = nskb;
1319 }
1320
1321 pskb_expand_head(skb, 0, -delta,
1322 (allocation & ~__GFP_DIRECT_RECLAIM) |
1323 __GFP_NOWARN | __GFP_NORETRY);
1324 return skb;
1325}
1326
1327static int netlink_unicast_kernel(struct sock *sk, struct sk_buff *skb,
1328 struct sock *ssk)
1329{
1330 int ret;
1331 struct netlink_sock *nlk = nlk_sk(sk);
1332
1333 ret = -ECONNREFUSED;
1334 if (nlk->netlink_rcv != NULL) {
1335 ret = skb->len;
1336 netlink_skb_set_owner_r(skb, sk);
1337 NETLINK_CB(skb).sk = ssk;
1338 netlink_deliver_tap_kernel(sk, ssk, skb);
1339 nlk->netlink_rcv(skb);
1340 consume_skb(skb);
1341 } else {
1342 kfree_skb(skb);
1343 }
1344 sock_put(sk);
1345 return ret;
1346}
1347
1348int netlink_unicast(struct sock *ssk, struct sk_buff *skb,
1349 u32 portid, int nonblock)
1350{
1351 struct sock *sk;
1352 int err;
1353 long timeo;
1354
1355 skb = netlink_trim(skb, gfp_any());
1356
1357 timeo = sock_sndtimeo(ssk, nonblock);
1358retry:
1359 sk = netlink_getsockbyportid(ssk, portid);
1360 if (IS_ERR(sk)) {
1361 kfree_skb(skb);
1362 return PTR_ERR(sk);
1363 }
1364 if (netlink_is_kernel(sk))
1365 return netlink_unicast_kernel(sk, skb, ssk);
1366
1367 if (sk_filter(sk, skb)) {
1368 err = skb->len;
1369 kfree_skb(skb);
1370 sock_put(sk);
1371 return err;
1372 }
1373
1374 err = netlink_attachskb(sk, skb, &timeo, ssk);
1375 if (err == 1)
1376 goto retry;
1377 if (err)
1378 return err;
1379
1380 return netlink_sendskb(sk, skb);
1381}
1382EXPORT_SYMBOL(netlink_unicast);
1383
1384int netlink_has_listeners(struct sock *sk, unsigned int group)
1385{
1386 int res = 0;
1387 struct listeners *listeners;
1388
1389 BUG_ON(!netlink_is_kernel(sk));
1390
1391 rcu_read_lock();
1392 listeners = rcu_dereference(nl_table[sk->sk_protocol].listeners);
1393
1394 if (listeners && group - 1 < nl_table[sk->sk_protocol].groups)
1395 res = test_bit(group - 1, listeners->masks);
1396
1397 rcu_read_unlock();
1398
1399 return res;
1400}
1401EXPORT_SYMBOL_GPL(netlink_has_listeners);
1402
1403bool netlink_strict_get_check(struct sk_buff *skb)
1404{
1405 const struct netlink_sock *nlk = nlk_sk(NETLINK_CB(skb).sk);
1406
1407 return nlk->flags & NETLINK_F_STRICT_CHK;
1408}
1409EXPORT_SYMBOL_GPL(netlink_strict_get_check);
1410
1411static int netlink_broadcast_deliver(struct sock *sk, struct sk_buff *skb)
1412{
1413 struct netlink_sock *nlk = nlk_sk(sk);
1414
1415 if (atomic_read(&sk->sk_rmem_alloc) <= sk->sk_rcvbuf &&
1416 !test_bit(NETLINK_S_CONGESTED, &nlk->state)) {
1417 netlink_skb_set_owner_r(skb, sk);
1418 __netlink_sendskb(sk, skb);
1419 return atomic_read(&sk->sk_rmem_alloc) > (sk->sk_rcvbuf >> 1);
1420 }
1421 return -1;
1422}
1423
1424struct netlink_broadcast_data {
1425 struct sock *exclude_sk;
1426 struct net *net;
1427 u32 portid;
1428 u32 group;
1429 int failure;
1430 int delivery_failure;
1431 int congested;
1432 int delivered;
1433 gfp_t allocation;
1434 struct sk_buff *skb, *skb2;
1435};
1436
1437static void do_one_broadcast(struct sock *sk,
1438 struct netlink_broadcast_data *p)
1439{
1440 struct netlink_sock *nlk = nlk_sk(sk);
1441 int val;
1442
1443 if (p->exclude_sk == sk)
1444 return;
1445
1446 if (nlk->portid == p->portid || p->group - 1 >= nlk->ngroups ||
1447 !test_bit(p->group - 1, nlk->groups))
1448 return;
1449
1450 if (!net_eq(sock_net(sk), p->net)) {
1451 if (!(nlk->flags & NETLINK_F_LISTEN_ALL_NSID))
1452 return;
1453
1454 if (!peernet_has_id(sock_net(sk), p->net))
1455 return;
1456
1457 if (!file_ns_capable(sk->sk_socket->file, p->net->user_ns,
1458 CAP_NET_BROADCAST))
1459 return;
1460 }
1461
1462 if (p->failure) {
1463 netlink_overrun(sk);
1464 return;
1465 }
1466
1467 sock_hold(sk);
1468 if (p->skb2 == NULL) {
1469 if (skb_shared(p->skb)) {
1470 p->skb2 = skb_clone(p->skb, p->allocation);
1471 } else {
1472 p->skb2 = skb_get(p->skb);
1473 /*
1474 * skb ownership may have been set when
1475 * delivered to a previous socket.
1476 */
1477 skb_orphan(p->skb2);
1478 }
1479 }
1480 if (p->skb2 == NULL) {
1481 netlink_overrun(sk);
1482 /* Clone failed. Notify ALL listeners. */
1483 p->failure = 1;
1484 if (nlk->flags & NETLINK_F_BROADCAST_SEND_ERROR)
1485 p->delivery_failure = 1;
1486 goto out;
1487 }
1488 if (sk_filter(sk, p->skb2)) {
1489 kfree_skb(p->skb2);
1490 p->skb2 = NULL;
1491 goto out;
1492 }
1493 NETLINK_CB(p->skb2).nsid = peernet2id(sock_net(sk), p->net);
1494 if (NETLINK_CB(p->skb2).nsid != NETNSA_NSID_NOT_ASSIGNED)
1495 NETLINK_CB(p->skb2).nsid_is_set = true;
1496 val = netlink_broadcast_deliver(sk, p->skb2);
1497 if (val < 0) {
1498 netlink_overrun(sk);
1499 if (nlk->flags & NETLINK_F_BROADCAST_SEND_ERROR)
1500 p->delivery_failure = 1;
1501 } else {
1502 p->congested |= val;
1503 p->delivered = 1;
1504 p->skb2 = NULL;
1505 }
1506out:
1507 sock_put(sk);
1508}
1509
1510int netlink_broadcast(struct sock *ssk, struct sk_buff *skb, u32 portid,
1511 u32 group, gfp_t allocation)
1512{
1513 struct net *net = sock_net(ssk);
1514 struct netlink_broadcast_data info;
1515 struct sock *sk;
1516
1517 skb = netlink_trim(skb, allocation);
1518
1519 info.exclude_sk = ssk;
1520 info.net = net;
1521 info.portid = portid;
1522 info.group = group;
1523 info.failure = 0;
1524 info.delivery_failure = 0;
1525 info.congested = 0;
1526 info.delivered = 0;
1527 info.allocation = allocation;
1528 info.skb = skb;
1529 info.skb2 = NULL;
1530
1531 /* While we sleep in clone, do not allow to change socket list */
1532
1533 netlink_lock_table();
1534
1535 sk_for_each_bound(sk, &nl_table[ssk->sk_protocol].mc_list)
1536 do_one_broadcast(sk, &info);
1537
1538 consume_skb(skb);
1539
1540 netlink_unlock_table();
1541
1542 if (info.delivery_failure) {
1543 kfree_skb(info.skb2);
1544 return -ENOBUFS;
1545 }
1546 consume_skb(info.skb2);
1547
1548 if (info.delivered) {
1549 if (info.congested && gfpflags_allow_blocking(allocation))
1550 yield();
1551 return 0;
1552 }
1553 return -ESRCH;
1554}
1555EXPORT_SYMBOL(netlink_broadcast);
1556
1557struct netlink_set_err_data {
1558 struct sock *exclude_sk;
1559 u32 portid;
1560 u32 group;
1561 int code;
1562};
1563
1564static int do_one_set_err(struct sock *sk, struct netlink_set_err_data *p)
1565{
1566 struct netlink_sock *nlk = nlk_sk(sk);
1567 int ret = 0;
1568
1569 if (sk == p->exclude_sk)
1570 goto out;
1571
1572 if (!net_eq(sock_net(sk), sock_net(p->exclude_sk)))
1573 goto out;
1574
1575 if (nlk->portid == p->portid || p->group - 1 >= nlk->ngroups ||
1576 !test_bit(p->group - 1, nlk->groups))
1577 goto out;
1578
1579 if (p->code == ENOBUFS && nlk->flags & NETLINK_F_RECV_NO_ENOBUFS) {
1580 ret = 1;
1581 goto out;
1582 }
1583
1584 sk->sk_err = p->code;
1585 sk_error_report(sk);
1586out:
1587 return ret;
1588}
1589
1590/**
1591 * netlink_set_err - report error to broadcast listeners
1592 * @ssk: the kernel netlink socket, as returned by netlink_kernel_create()
1593 * @portid: the PORTID of a process that we want to skip (if any)
1594 * @group: the broadcast group that will notice the error
1595 * @code: error code, must be negative (as usual in kernelspace)
1596 *
1597 * This function returns the number of broadcast listeners that have set the
1598 * NETLINK_NO_ENOBUFS socket option.
1599 */
1600int netlink_set_err(struct sock *ssk, u32 portid, u32 group, int code)
1601{
1602 struct netlink_set_err_data info;
1603 struct sock *sk;
1604 int ret = 0;
1605
1606 info.exclude_sk = ssk;
1607 info.portid = portid;
1608 info.group = group;
1609 /* sk->sk_err wants a positive error value */
1610 info.code = -code;
1611
1612 read_lock(&nl_table_lock);
1613
1614 sk_for_each_bound(sk, &nl_table[ssk->sk_protocol].mc_list)
1615 ret += do_one_set_err(sk, &info);
1616
1617 read_unlock(&nl_table_lock);
1618 return ret;
1619}
1620EXPORT_SYMBOL(netlink_set_err);
1621
1622/* must be called with netlink table grabbed */
1623static void netlink_update_socket_mc(struct netlink_sock *nlk,
1624 unsigned int group,
1625 int is_new)
1626{
1627 int old, new = !!is_new, subscriptions;
1628
1629 old = test_bit(group - 1, nlk->groups);
1630 subscriptions = nlk->subscriptions - old + new;
1631 if (new)
1632 __set_bit(group - 1, nlk->groups);
1633 else
1634 __clear_bit(group - 1, nlk->groups);
1635 netlink_update_subscriptions(&nlk->sk, subscriptions);
1636 netlink_update_listeners(&nlk->sk);
1637}
1638
1639static int netlink_setsockopt(struct socket *sock, int level, int optname,
1640 sockptr_t optval, unsigned int optlen)
1641{
1642 struct sock *sk = sock->sk;
1643 struct netlink_sock *nlk = nlk_sk(sk);
1644 unsigned int val = 0;
1645 int err;
1646
1647 if (level != SOL_NETLINK)
1648 return -ENOPROTOOPT;
1649
1650 if (optlen >= sizeof(int) &&
1651 copy_from_sockptr(&val, optval, sizeof(val)))
1652 return -EFAULT;
1653
1654 switch (optname) {
1655 case NETLINK_PKTINFO:
1656 if (val)
1657 nlk->flags |= NETLINK_F_RECV_PKTINFO;
1658 else
1659 nlk->flags &= ~NETLINK_F_RECV_PKTINFO;
1660 err = 0;
1661 break;
1662 case NETLINK_ADD_MEMBERSHIP:
1663 case NETLINK_DROP_MEMBERSHIP: {
1664 if (!netlink_allowed(sock, NL_CFG_F_NONROOT_RECV))
1665 return -EPERM;
1666 err = netlink_realloc_groups(sk);
1667 if (err)
1668 return err;
1669 if (!val || val - 1 >= nlk->ngroups)
1670 return -EINVAL;
1671 if (optname == NETLINK_ADD_MEMBERSHIP && nlk->netlink_bind) {
1672 err = nlk->netlink_bind(sock_net(sk), val);
1673 if (err)
1674 return err;
1675 }
1676 netlink_table_grab();
1677 netlink_update_socket_mc(nlk, val,
1678 optname == NETLINK_ADD_MEMBERSHIP);
1679 netlink_table_ungrab();
1680 if (optname == NETLINK_DROP_MEMBERSHIP && nlk->netlink_unbind)
1681 nlk->netlink_unbind(sock_net(sk), val);
1682
1683 err = 0;
1684 break;
1685 }
1686 case NETLINK_BROADCAST_ERROR:
1687 if (val)
1688 nlk->flags |= NETLINK_F_BROADCAST_SEND_ERROR;
1689 else
1690 nlk->flags &= ~NETLINK_F_BROADCAST_SEND_ERROR;
1691 err = 0;
1692 break;
1693 case NETLINK_NO_ENOBUFS:
1694 if (val) {
1695 nlk->flags |= NETLINK_F_RECV_NO_ENOBUFS;
1696 clear_bit(NETLINK_S_CONGESTED, &nlk->state);
1697 wake_up_interruptible(&nlk->wait);
1698 } else {
1699 nlk->flags &= ~NETLINK_F_RECV_NO_ENOBUFS;
1700 }
1701 err = 0;
1702 break;
1703 case NETLINK_LISTEN_ALL_NSID:
1704 if (!ns_capable(sock_net(sk)->user_ns, CAP_NET_BROADCAST))
1705 return -EPERM;
1706
1707 if (val)
1708 nlk->flags |= NETLINK_F_LISTEN_ALL_NSID;
1709 else
1710 nlk->flags &= ~NETLINK_F_LISTEN_ALL_NSID;
1711 err = 0;
1712 break;
1713 case NETLINK_CAP_ACK:
1714 if (val)
1715 nlk->flags |= NETLINK_F_CAP_ACK;
1716 else
1717 nlk->flags &= ~NETLINK_F_CAP_ACK;
1718 err = 0;
1719 break;
1720 case NETLINK_EXT_ACK:
1721 if (val)
1722 nlk->flags |= NETLINK_F_EXT_ACK;
1723 else
1724 nlk->flags &= ~NETLINK_F_EXT_ACK;
1725 err = 0;
1726 break;
1727 case NETLINK_GET_STRICT_CHK:
1728 if (val)
1729 nlk->flags |= NETLINK_F_STRICT_CHK;
1730 else
1731 nlk->flags &= ~NETLINK_F_STRICT_CHK;
1732 err = 0;
1733 break;
1734 default:
1735 err = -ENOPROTOOPT;
1736 }
1737 return err;
1738}
1739
1740static int netlink_getsockopt(struct socket *sock, int level, int optname,
1741 char __user *optval, int __user *optlen)
1742{
1743 struct sock *sk = sock->sk;
1744 struct netlink_sock *nlk = nlk_sk(sk);
1745 int len, val, err;
1746
1747 if (level != SOL_NETLINK)
1748 return -ENOPROTOOPT;
1749
1750 if (get_user(len, optlen))
1751 return -EFAULT;
1752 if (len < 0)
1753 return -EINVAL;
1754
1755 switch (optname) {
1756 case NETLINK_PKTINFO:
1757 if (len < sizeof(int))
1758 return -EINVAL;
1759 len = sizeof(int);
1760 val = nlk->flags & NETLINK_F_RECV_PKTINFO ? 1 : 0;
1761 if (put_user(len, optlen) ||
1762 put_user(val, optval))
1763 return -EFAULT;
1764 err = 0;
1765 break;
1766 case NETLINK_BROADCAST_ERROR:
1767 if (len < sizeof(int))
1768 return -EINVAL;
1769 len = sizeof(int);
1770 val = nlk->flags & NETLINK_F_BROADCAST_SEND_ERROR ? 1 : 0;
1771 if (put_user(len, optlen) ||
1772 put_user(val, optval))
1773 return -EFAULT;
1774 err = 0;
1775 break;
1776 case NETLINK_NO_ENOBUFS:
1777 if (len < sizeof(int))
1778 return -EINVAL;
1779 len = sizeof(int);
1780 val = nlk->flags & NETLINK_F_RECV_NO_ENOBUFS ? 1 : 0;
1781 if (put_user(len, optlen) ||
1782 put_user(val, optval))
1783 return -EFAULT;
1784 err = 0;
1785 break;
1786 case NETLINK_LIST_MEMBERSHIPS: {
1787 int pos, idx, shift;
1788
1789 err = 0;
1790 netlink_lock_table();
1791 for (pos = 0; pos * 8 < nlk->ngroups; pos += sizeof(u32)) {
1792 if (len - pos < sizeof(u32))
1793 break;
1794
1795 idx = pos / sizeof(unsigned long);
1796 shift = (pos % sizeof(unsigned long)) * 8;
1797 if (put_user((u32)(nlk->groups[idx] >> shift),
1798 (u32 __user *)(optval + pos))) {
1799 err = -EFAULT;
1800 break;
1801 }
1802 }
1803 if (put_user(ALIGN(nlk->ngroups / 8, sizeof(u32)), optlen))
1804 err = -EFAULT;
1805 netlink_unlock_table();
1806 break;
1807 }
1808 case NETLINK_CAP_ACK:
1809 if (len < sizeof(int))
1810 return -EINVAL;
1811 len = sizeof(int);
1812 val = nlk->flags & NETLINK_F_CAP_ACK ? 1 : 0;
1813 if (put_user(len, optlen) ||
1814 put_user(val, optval))
1815 return -EFAULT;
1816 err = 0;
1817 break;
1818 case NETLINK_EXT_ACK:
1819 if (len < sizeof(int))
1820 return -EINVAL;
1821 len = sizeof(int);
1822 val = nlk->flags & NETLINK_F_EXT_ACK ? 1 : 0;
1823 if (put_user(len, optlen) || put_user(val, optval))
1824 return -EFAULT;
1825 err = 0;
1826 break;
1827 case NETLINK_GET_STRICT_CHK:
1828 if (len < sizeof(int))
1829 return -EINVAL;
1830 len = sizeof(int);
1831 val = nlk->flags & NETLINK_F_STRICT_CHK ? 1 : 0;
1832 if (put_user(len, optlen) || put_user(val, optval))
1833 return -EFAULT;
1834 err = 0;
1835 break;
1836 default:
1837 err = -ENOPROTOOPT;
1838 }
1839 return err;
1840}
1841
1842static void netlink_cmsg_recv_pktinfo(struct msghdr *msg, struct sk_buff *skb)
1843{
1844 struct nl_pktinfo info;
1845
1846 info.group = NETLINK_CB(skb).dst_group;
1847 put_cmsg(msg, SOL_NETLINK, NETLINK_PKTINFO, sizeof(info), &info);
1848}
1849
1850static void netlink_cmsg_listen_all_nsid(struct sock *sk, struct msghdr *msg,
1851 struct sk_buff *skb)
1852{
1853 if (!NETLINK_CB(skb).nsid_is_set)
1854 return;
1855
1856 put_cmsg(msg, SOL_NETLINK, NETLINK_LISTEN_ALL_NSID, sizeof(int),
1857 &NETLINK_CB(skb).nsid);
1858}
1859
1860static int netlink_sendmsg(struct socket *sock, struct msghdr *msg, size_t len)
1861{
1862 struct sock *sk = sock->sk;
1863 struct netlink_sock *nlk = nlk_sk(sk);
1864 DECLARE_SOCKADDR(struct sockaddr_nl *, addr, msg->msg_name);
1865 u32 dst_portid;
1866 u32 dst_group;
1867 struct sk_buff *skb;
1868 int err;
1869 struct scm_cookie scm;
1870 u32 netlink_skb_flags = 0;
1871
1872 if (msg->msg_flags & MSG_OOB)
1873 return -EOPNOTSUPP;
1874
1875 if (len == 0) {
1876 pr_warn_once("Zero length message leads to an empty skb\n");
1877 return -ENODATA;
1878 }
1879
1880 err = scm_send(sock, msg, &scm, true);
1881 if (err < 0)
1882 return err;
1883
1884 if (msg->msg_namelen) {
1885 err = -EINVAL;
1886 if (msg->msg_namelen < sizeof(struct sockaddr_nl))
1887 goto out;
1888 if (addr->nl_family != AF_NETLINK)
1889 goto out;
1890 dst_portid = addr->nl_pid;
1891 dst_group = ffs(addr->nl_groups);
1892 err = -EPERM;
1893 if ((dst_group || dst_portid) &&
1894 !netlink_allowed(sock, NL_CFG_F_NONROOT_SEND))
1895 goto out;
1896 netlink_skb_flags |= NETLINK_SKB_DST;
1897 } else {
1898 /* Paired with WRITE_ONCE() in netlink_connect() */
1899 dst_portid = READ_ONCE(nlk->dst_portid);
1900 dst_group = READ_ONCE(nlk->dst_group);
1901 }
1902
1903 /* Paired with WRITE_ONCE() in netlink_insert() */
1904 if (!READ_ONCE(nlk->bound)) {
1905 err = netlink_autobind(sock);
1906 if (err)
1907 goto out;
1908 } else {
1909 /* Ensure nlk is hashed and visible. */
1910 smp_rmb();
1911 }
1912
1913 err = -EMSGSIZE;
1914 if (len > sk->sk_sndbuf - 32)
1915 goto out;
1916 err = -ENOBUFS;
1917 skb = netlink_alloc_large_skb(len, dst_group);
1918 if (skb == NULL)
1919 goto out;
1920
1921 NETLINK_CB(skb).portid = nlk->portid;
1922 NETLINK_CB(skb).dst_group = dst_group;
1923 NETLINK_CB(skb).creds = scm.creds;
1924 NETLINK_CB(skb).flags = netlink_skb_flags;
1925
1926 err = -EFAULT;
1927 if (memcpy_from_msg(skb_put(skb, len), msg, len)) {
1928 kfree_skb(skb);
1929 goto out;
1930 }
1931
1932 err = security_netlink_send(sk, skb);
1933 if (err) {
1934 kfree_skb(skb);
1935 goto out;
1936 }
1937
1938 if (dst_group) {
1939 refcount_inc(&skb->users);
1940 netlink_broadcast(sk, skb, dst_portid, dst_group, GFP_KERNEL);
1941 }
1942 err = netlink_unicast(sk, skb, dst_portid, msg->msg_flags & MSG_DONTWAIT);
1943
1944out:
1945 scm_destroy(&scm);
1946 return err;
1947}
1948
1949static int netlink_recvmsg(struct socket *sock, struct msghdr *msg, size_t len,
1950 int flags)
1951{
1952 struct scm_cookie scm;
1953 struct sock *sk = sock->sk;
1954 struct netlink_sock *nlk = nlk_sk(sk);
1955 size_t copied;
1956 struct sk_buff *skb, *data_skb;
1957 int err, ret;
1958
1959 if (flags & MSG_OOB)
1960 return -EOPNOTSUPP;
1961
1962 copied = 0;
1963
1964 skb = skb_recv_datagram(sk, flags, &err);
1965 if (skb == NULL)
1966 goto out;
1967
1968 data_skb = skb;
1969
1970#ifdef CONFIG_COMPAT_NETLINK_MESSAGES
1971 if (unlikely(skb_shinfo(skb)->frag_list)) {
1972 /*
1973 * If this skb has a frag_list, then here that means that we
1974 * will have to use the frag_list skb's data for compat tasks
1975 * and the regular skb's data for normal (non-compat) tasks.
1976 *
1977 * If we need to send the compat skb, assign it to the
1978 * 'data_skb' variable so that it will be used below for data
1979 * copying. We keep 'skb' for everything else, including
1980 * freeing both later.
1981 */
1982 if (flags & MSG_CMSG_COMPAT)
1983 data_skb = skb_shinfo(skb)->frag_list;
1984 }
1985#endif
1986
1987 /* Record the max length of recvmsg() calls for future allocations */
1988 nlk->max_recvmsg_len = max(nlk->max_recvmsg_len, len);
1989 nlk->max_recvmsg_len = min_t(size_t, nlk->max_recvmsg_len,
1990 SKB_WITH_OVERHEAD(32768));
1991
1992 copied = data_skb->len;
1993 if (len < copied) {
1994 msg->msg_flags |= MSG_TRUNC;
1995 copied = len;
1996 }
1997
1998 err = skb_copy_datagram_msg(data_skb, 0, msg, copied);
1999
2000 if (msg->msg_name) {
2001 DECLARE_SOCKADDR(struct sockaddr_nl *, addr, msg->msg_name);
2002 addr->nl_family = AF_NETLINK;
2003 addr->nl_pad = 0;
2004 addr->nl_pid = NETLINK_CB(skb).portid;
2005 addr->nl_groups = netlink_group_mask(NETLINK_CB(skb).dst_group);
2006 msg->msg_namelen = sizeof(*addr);
2007 }
2008
2009 if (nlk->flags & NETLINK_F_RECV_PKTINFO)
2010 netlink_cmsg_recv_pktinfo(msg, skb);
2011 if (nlk->flags & NETLINK_F_LISTEN_ALL_NSID)
2012 netlink_cmsg_listen_all_nsid(sk, msg, skb);
2013
2014 memset(&scm, 0, sizeof(scm));
2015 scm.creds = *NETLINK_CREDS(skb);
2016 if (flags & MSG_TRUNC)
2017 copied = data_skb->len;
2018
2019 skb_free_datagram(sk, skb);
2020
2021 if (nlk->cb_running &&
2022 atomic_read(&sk->sk_rmem_alloc) <= sk->sk_rcvbuf / 2) {
2023 ret = netlink_dump(sk);
2024 if (ret) {
2025 sk->sk_err = -ret;
2026 sk_error_report(sk);
2027 }
2028 }
2029
2030 scm_recv(sock, msg, &scm, flags);
2031out:
2032 netlink_rcv_wake(sk);
2033 return err ? : copied;
2034}
2035
2036static void netlink_data_ready(struct sock *sk)
2037{
2038 BUG();
2039}
2040
2041/*
2042 * We export these functions to other modules. They provide a
2043 * complete set of kernel non-blocking support for message
2044 * queueing.
2045 */
2046
2047struct sock *
2048__netlink_kernel_create(struct net *net, int unit, struct module *module,
2049 struct netlink_kernel_cfg *cfg)
2050{
2051 struct socket *sock;
2052 struct sock *sk;
2053 struct netlink_sock *nlk;
2054 struct listeners *listeners = NULL;
2055 struct mutex *cb_mutex = cfg ? cfg->cb_mutex : NULL;
2056 unsigned int groups;
2057
2058 BUG_ON(!nl_table);
2059
2060 if (unit < 0 || unit >= MAX_LINKS)
2061 return NULL;
2062
2063 if (sock_create_lite(PF_NETLINK, SOCK_DGRAM, unit, &sock))
2064 return NULL;
2065
2066 if (__netlink_create(net, sock, cb_mutex, unit, 1) < 0)
2067 goto out_sock_release_nosk;
2068
2069 sk = sock->sk;
2070
2071 if (!cfg || cfg->groups < 32)
2072 groups = 32;
2073 else
2074 groups = cfg->groups;
2075
2076 listeners = kzalloc(sizeof(*listeners) + NLGRPSZ(groups), GFP_KERNEL);
2077 if (!listeners)
2078 goto out_sock_release;
2079
2080 sk->sk_data_ready = netlink_data_ready;
2081 if (cfg && cfg->input)
2082 nlk_sk(sk)->netlink_rcv = cfg->input;
2083
2084 if (netlink_insert(sk, 0))
2085 goto out_sock_release;
2086
2087 nlk = nlk_sk(sk);
2088 nlk->flags |= NETLINK_F_KERNEL_SOCKET;
2089
2090 netlink_table_grab();
2091 if (!nl_table[unit].registered) {
2092 nl_table[unit].groups = groups;
2093 rcu_assign_pointer(nl_table[unit].listeners, listeners);
2094 nl_table[unit].cb_mutex = cb_mutex;
2095 nl_table[unit].module = module;
2096 if (cfg) {
2097 nl_table[unit].bind = cfg->bind;
2098 nl_table[unit].unbind = cfg->unbind;
2099 nl_table[unit].flags = cfg->flags;
2100 if (cfg->compare)
2101 nl_table[unit].compare = cfg->compare;
2102 }
2103 nl_table[unit].registered = 1;
2104 } else {
2105 kfree(listeners);
2106 nl_table[unit].registered++;
2107 }
2108 netlink_table_ungrab();
2109 return sk;
2110
2111out_sock_release:
2112 kfree(listeners);
2113 netlink_kernel_release(sk);
2114 return NULL;
2115
2116out_sock_release_nosk:
2117 sock_release(sock);
2118 return NULL;
2119}
2120EXPORT_SYMBOL(__netlink_kernel_create);
2121
2122void
2123netlink_kernel_release(struct sock *sk)
2124{
2125 if (sk == NULL || sk->sk_socket == NULL)
2126 return;
2127
2128 sock_release(sk->sk_socket);
2129}
2130EXPORT_SYMBOL(netlink_kernel_release);
2131
2132int __netlink_change_ngroups(struct sock *sk, unsigned int groups)
2133{
2134 struct listeners *new, *old;
2135 struct netlink_table *tbl = &nl_table[sk->sk_protocol];
2136
2137 if (groups < 32)
2138 groups = 32;
2139
2140 if (NLGRPSZ(tbl->groups) < NLGRPSZ(groups)) {
2141 new = kzalloc(sizeof(*new) + NLGRPSZ(groups), GFP_ATOMIC);
2142 if (!new)
2143 return -ENOMEM;
2144 old = nl_deref_protected(tbl->listeners);
2145 memcpy(new->masks, old->masks, NLGRPSZ(tbl->groups));
2146 rcu_assign_pointer(tbl->listeners, new);
2147
2148 kfree_rcu(old, rcu);
2149 }
2150 tbl->groups = groups;
2151
2152 return 0;
2153}
2154
2155/**
2156 * netlink_change_ngroups - change number of multicast groups
2157 *
2158 * This changes the number of multicast groups that are available
2159 * on a certain netlink family. Note that it is not possible to
2160 * change the number of groups to below 32. Also note that it does
2161 * not implicitly call netlink_clear_multicast_users() when the
2162 * number of groups is reduced.
2163 *
2164 * @sk: The kernel netlink socket, as returned by netlink_kernel_create().
2165 * @groups: The new number of groups.
2166 */
2167int netlink_change_ngroups(struct sock *sk, unsigned int groups)
2168{
2169 int err;
2170
2171 netlink_table_grab();
2172 err = __netlink_change_ngroups(sk, groups);
2173 netlink_table_ungrab();
2174
2175 return err;
2176}
2177
2178void __netlink_clear_multicast_users(struct sock *ksk, unsigned int group)
2179{
2180 struct sock *sk;
2181 struct netlink_table *tbl = &nl_table[ksk->sk_protocol];
2182
2183 sk_for_each_bound(sk, &tbl->mc_list)
2184 netlink_update_socket_mc(nlk_sk(sk), group, 0);
2185}
2186
2187struct nlmsghdr *
2188__nlmsg_put(struct sk_buff *skb, u32 portid, u32 seq, int type, int len, int flags)
2189{
2190 struct nlmsghdr *nlh;
2191 int size = nlmsg_msg_size(len);
2192
2193 nlh = skb_put(skb, NLMSG_ALIGN(size));
2194 nlh->nlmsg_type = type;
2195 nlh->nlmsg_len = size;
2196 nlh->nlmsg_flags = flags;
2197 nlh->nlmsg_pid = portid;
2198 nlh->nlmsg_seq = seq;
2199 if (!__builtin_constant_p(size) || NLMSG_ALIGN(size) - size != 0)
2200 memset(nlmsg_data(nlh) + len, 0, NLMSG_ALIGN(size) - size);
2201 return nlh;
2202}
2203EXPORT_SYMBOL(__nlmsg_put);
2204
2205/*
2206 * It looks a bit ugly.
2207 * It would be better to create kernel thread.
2208 */
2209
2210static int netlink_dump_done(struct netlink_sock *nlk, struct sk_buff *skb,
2211 struct netlink_callback *cb,
2212 struct netlink_ext_ack *extack)
2213{
2214 struct nlmsghdr *nlh;
2215
2216 nlh = nlmsg_put_answer(skb, cb, NLMSG_DONE, sizeof(nlk->dump_done_errno),
2217 NLM_F_MULTI | cb->answer_flags);
2218 if (WARN_ON(!nlh))
2219 return -ENOBUFS;
2220
2221 nl_dump_check_consistent(cb, nlh);
2222 memcpy(nlmsg_data(nlh), &nlk->dump_done_errno, sizeof(nlk->dump_done_errno));
2223
2224 if (extack->_msg && nlk->flags & NETLINK_F_EXT_ACK) {
2225 nlh->nlmsg_flags |= NLM_F_ACK_TLVS;
2226 if (!nla_put_string(skb, NLMSGERR_ATTR_MSG, extack->_msg))
2227 nlmsg_end(skb, nlh);
2228 }
2229
2230 return 0;
2231}
2232
2233static int netlink_dump(struct sock *sk)
2234{
2235 struct netlink_sock *nlk = nlk_sk(sk);
2236 struct netlink_ext_ack extack = {};
2237 struct netlink_callback *cb;
2238 struct sk_buff *skb = NULL;
2239 struct module *module;
2240 int err = -ENOBUFS;
2241 int alloc_min_size;
2242 int alloc_size;
2243
2244 mutex_lock(nlk->cb_mutex);
2245 if (!nlk->cb_running) {
2246 err = -EINVAL;
2247 goto errout_skb;
2248 }
2249
2250 if (atomic_read(&sk->sk_rmem_alloc) >= sk->sk_rcvbuf)
2251 goto errout_skb;
2252
2253 /* NLMSG_GOODSIZE is small to avoid high order allocations being
2254 * required, but it makes sense to _attempt_ a 16K bytes allocation
2255 * to reduce number of system calls on dump operations, if user
2256 * ever provided a big enough buffer.
2257 */
2258 cb = &nlk->cb;
2259 alloc_min_size = max_t(int, cb->min_dump_alloc, NLMSG_GOODSIZE);
2260
2261 if (alloc_min_size < nlk->max_recvmsg_len) {
2262 alloc_size = nlk->max_recvmsg_len;
2263 skb = alloc_skb(alloc_size,
2264 (GFP_KERNEL & ~__GFP_DIRECT_RECLAIM) |
2265 __GFP_NOWARN | __GFP_NORETRY);
2266 }
2267 if (!skb) {
2268 alloc_size = alloc_min_size;
2269 skb = alloc_skb(alloc_size, GFP_KERNEL);
2270 }
2271 if (!skb)
2272 goto errout_skb;
2273
2274 /* Trim skb to allocated size. User is expected to provide buffer as
2275 * large as max(min_dump_alloc, 16KiB (mac_recvmsg_len capped at
2276 * netlink_recvmsg())). dump will pack as many smaller messages as
2277 * could fit within the allocated skb. skb is typically allocated
2278 * with larger space than required (could be as much as near 2x the
2279 * requested size with align to next power of 2 approach). Allowing
2280 * dump to use the excess space makes it difficult for a user to have a
2281 * reasonable static buffer based on the expected largest dump of a
2282 * single netdev. The outcome is MSG_TRUNC error.
2283 */
2284 skb_reserve(skb, skb_tailroom(skb) - alloc_size);
2285
2286 /* Make sure malicious BPF programs can not read unitialized memory
2287 * from skb->head -> skb->data
2288 */
2289 skb_reset_network_header(skb);
2290 skb_reset_mac_header(skb);
2291
2292 netlink_skb_set_owner_r(skb, sk);
2293
2294 if (nlk->dump_done_errno > 0) {
2295 cb->extack = &extack;
2296 nlk->dump_done_errno = cb->dump(skb, cb);
2297 cb->extack = NULL;
2298 }
2299
2300 if (nlk->dump_done_errno > 0 ||
2301 skb_tailroom(skb) < nlmsg_total_size(sizeof(nlk->dump_done_errno))) {
2302 mutex_unlock(nlk->cb_mutex);
2303
2304 if (sk_filter(sk, skb))
2305 kfree_skb(skb);
2306 else
2307 __netlink_sendskb(sk, skb);
2308 return 0;
2309 }
2310
2311 if (netlink_dump_done(nlk, skb, cb, &extack))
2312 goto errout_skb;
2313
2314#ifdef CONFIG_COMPAT_NETLINK_MESSAGES
2315 /* frag_list skb's data is used for compat tasks
2316 * and the regular skb's data for normal (non-compat) tasks.
2317 * See netlink_recvmsg().
2318 */
2319 if (unlikely(skb_shinfo(skb)->frag_list)) {
2320 if (netlink_dump_done(nlk, skb_shinfo(skb)->frag_list, cb, &extack))
2321 goto errout_skb;
2322 }
2323#endif
2324
2325 if (sk_filter(sk, skb))
2326 kfree_skb(skb);
2327 else
2328 __netlink_sendskb(sk, skb);
2329
2330 if (cb->done)
2331 cb->done(cb);
2332
2333 nlk->cb_running = false;
2334 module = cb->module;
2335 skb = cb->skb;
2336 mutex_unlock(nlk->cb_mutex);
2337 module_put(module);
2338 consume_skb(skb);
2339 return 0;
2340
2341errout_skb:
2342 mutex_unlock(nlk->cb_mutex);
2343 kfree_skb(skb);
2344 return err;
2345}
2346
2347int __netlink_dump_start(struct sock *ssk, struct sk_buff *skb,
2348 const struct nlmsghdr *nlh,
2349 struct netlink_dump_control *control)
2350{
2351 struct netlink_sock *nlk, *nlk2;
2352 struct netlink_callback *cb;
2353 struct sock *sk;
2354 int ret;
2355
2356 refcount_inc(&skb->users);
2357
2358 sk = netlink_lookup(sock_net(ssk), ssk->sk_protocol, NETLINK_CB(skb).portid);
2359 if (sk == NULL) {
2360 ret = -ECONNREFUSED;
2361 goto error_free;
2362 }
2363
2364 nlk = nlk_sk(sk);
2365 mutex_lock(nlk->cb_mutex);
2366 /* A dump is in progress... */
2367 if (nlk->cb_running) {
2368 ret = -EBUSY;
2369 goto error_unlock;
2370 }
2371 /* add reference of module which cb->dump belongs to */
2372 if (!try_module_get(control->module)) {
2373 ret = -EPROTONOSUPPORT;
2374 goto error_unlock;
2375 }
2376
2377 cb = &nlk->cb;
2378 memset(cb, 0, sizeof(*cb));
2379 cb->dump = control->dump;
2380 cb->done = control->done;
2381 cb->nlh = nlh;
2382 cb->data = control->data;
2383 cb->module = control->module;
2384 cb->min_dump_alloc = control->min_dump_alloc;
2385 cb->skb = skb;
2386
2387 nlk2 = nlk_sk(NETLINK_CB(skb).sk);
2388 cb->strict_check = !!(nlk2->flags & NETLINK_F_STRICT_CHK);
2389
2390 if (control->start) {
2391 ret = control->start(cb);
2392 if (ret)
2393 goto error_put;
2394 }
2395
2396 nlk->cb_running = true;
2397 nlk->dump_done_errno = INT_MAX;
2398
2399 mutex_unlock(nlk->cb_mutex);
2400
2401 ret = netlink_dump(sk);
2402
2403 sock_put(sk);
2404
2405 if (ret)
2406 return ret;
2407
2408 /* We successfully started a dump, by returning -EINTR we
2409 * signal not to send ACK even if it was requested.
2410 */
2411 return -EINTR;
2412
2413error_put:
2414 module_put(control->module);
2415error_unlock:
2416 sock_put(sk);
2417 mutex_unlock(nlk->cb_mutex);
2418error_free:
2419 kfree_skb(skb);
2420 return ret;
2421}
2422EXPORT_SYMBOL(__netlink_dump_start);
2423
2424static size_t
2425netlink_ack_tlv_len(struct netlink_sock *nlk, int err,
2426 const struct netlink_ext_ack *extack)
2427{
2428 size_t tlvlen;
2429
2430 if (!extack || !(nlk->flags & NETLINK_F_EXT_ACK))
2431 return 0;
2432
2433 tlvlen = 0;
2434 if (extack->_msg)
2435 tlvlen += nla_total_size(strlen(extack->_msg) + 1);
2436 if (extack->cookie_len)
2437 tlvlen += nla_total_size(extack->cookie_len);
2438
2439 /* Following attributes are only reported as error (not warning) */
2440 if (!err)
2441 return tlvlen;
2442
2443 if (extack->bad_attr)
2444 tlvlen += nla_total_size(sizeof(u32));
2445 if (extack->policy)
2446 tlvlen += netlink_policy_dump_attr_size_estimate(extack->policy);
2447 if (extack->miss_type)
2448 tlvlen += nla_total_size(sizeof(u32));
2449 if (extack->miss_nest)
2450 tlvlen += nla_total_size(sizeof(u32));
2451
2452 return tlvlen;
2453}
2454
2455static void
2456netlink_ack_tlv_fill(struct sk_buff *in_skb, struct sk_buff *skb,
2457 struct nlmsghdr *nlh, int err,
2458 const struct netlink_ext_ack *extack)
2459{
2460 if (extack->_msg)
2461 WARN_ON(nla_put_string(skb, NLMSGERR_ATTR_MSG, extack->_msg));
2462 if (extack->cookie_len)
2463 WARN_ON(nla_put(skb, NLMSGERR_ATTR_COOKIE,
2464 extack->cookie_len, extack->cookie));
2465
2466 if (!err)
2467 return;
2468
2469 if (extack->bad_attr &&
2470 !WARN_ON((u8 *)extack->bad_attr < in_skb->data ||
2471 (u8 *)extack->bad_attr >= in_skb->data + in_skb->len))
2472 WARN_ON(nla_put_u32(skb, NLMSGERR_ATTR_OFFS,
2473 (u8 *)extack->bad_attr - (u8 *)nlh));
2474 if (extack->policy)
2475 netlink_policy_dump_write_attr(skb, extack->policy,
2476 NLMSGERR_ATTR_POLICY);
2477 if (extack->miss_type)
2478 WARN_ON(nla_put_u32(skb, NLMSGERR_ATTR_MISS_TYPE,
2479 extack->miss_type));
2480 if (extack->miss_nest &&
2481 !WARN_ON((u8 *)extack->miss_nest < in_skb->data ||
2482 (u8 *)extack->miss_nest > in_skb->data + in_skb->len))
2483 WARN_ON(nla_put_u32(skb, NLMSGERR_ATTR_MISS_NEST,
2484 (u8 *)extack->miss_nest - (u8 *)nlh));
2485}
2486
2487void netlink_ack(struct sk_buff *in_skb, struct nlmsghdr *nlh, int err,
2488 const struct netlink_ext_ack *extack)
2489{
2490 struct sk_buff *skb;
2491 struct nlmsghdr *rep;
2492 struct nlmsgerr *errmsg;
2493 size_t payload = sizeof(*errmsg);
2494 struct netlink_sock *nlk = nlk_sk(NETLINK_CB(in_skb).sk);
2495 unsigned int flags = 0;
2496 size_t tlvlen;
2497
2498 /* Error messages get the original request appened, unless the user
2499 * requests to cap the error message, and get extra error data if
2500 * requested.
2501 */
2502 if (err && !(nlk->flags & NETLINK_F_CAP_ACK))
2503 payload += nlmsg_len(nlh);
2504 else
2505 flags |= NLM_F_CAPPED;
2506
2507 tlvlen = netlink_ack_tlv_len(nlk, err, extack);
2508 if (tlvlen)
2509 flags |= NLM_F_ACK_TLVS;
2510
2511 skb = nlmsg_new(payload + tlvlen, GFP_KERNEL);
2512 if (!skb)
2513 goto err_skb;
2514
2515 rep = nlmsg_put(skb, NETLINK_CB(in_skb).portid, nlh->nlmsg_seq,
2516 NLMSG_ERROR, sizeof(*errmsg), flags);
2517 if (!rep)
2518 goto err_bad_put;
2519 errmsg = nlmsg_data(rep);
2520 errmsg->error = err;
2521 errmsg->msg = *nlh;
2522
2523 if (!(flags & NLM_F_CAPPED)) {
2524 if (!nlmsg_append(skb, nlmsg_len(nlh)))
2525 goto err_bad_put;
2526
2527 memcpy(nlmsg_data(&errmsg->msg), nlmsg_data(nlh),
2528 nlmsg_len(nlh));
2529 }
2530
2531 if (tlvlen)
2532 netlink_ack_tlv_fill(in_skb, skb, nlh, err, extack);
2533
2534 nlmsg_end(skb, rep);
2535
2536 nlmsg_unicast(in_skb->sk, skb, NETLINK_CB(in_skb).portid);
2537
2538 return;
2539
2540err_bad_put:
2541 nlmsg_free(skb);
2542err_skb:
2543 NETLINK_CB(in_skb).sk->sk_err = ENOBUFS;
2544 sk_error_report(NETLINK_CB(in_skb).sk);
2545}
2546EXPORT_SYMBOL(netlink_ack);
2547
2548int netlink_rcv_skb(struct sk_buff *skb, int (*cb)(struct sk_buff *,
2549 struct nlmsghdr *,
2550 struct netlink_ext_ack *))
2551{
2552 struct netlink_ext_ack extack;
2553 struct nlmsghdr *nlh;
2554 int err;
2555
2556 while (skb->len >= nlmsg_total_size(0)) {
2557 int msglen;
2558
2559 memset(&extack, 0, sizeof(extack));
2560 nlh = nlmsg_hdr(skb);
2561 err = 0;
2562
2563 if (nlh->nlmsg_len < NLMSG_HDRLEN || skb->len < nlh->nlmsg_len)
2564 return 0;
2565
2566 /* Only requests are handled by the kernel */
2567 if (!(nlh->nlmsg_flags & NLM_F_REQUEST))
2568 goto ack;
2569
2570 /* Skip control messages */
2571 if (nlh->nlmsg_type < NLMSG_MIN_TYPE)
2572 goto ack;
2573
2574 err = cb(skb, nlh, &extack);
2575 if (err == -EINTR)
2576 goto skip;
2577
2578ack:
2579 if (nlh->nlmsg_flags & NLM_F_ACK || err)
2580 netlink_ack(skb, nlh, err, &extack);
2581
2582skip:
2583 msglen = NLMSG_ALIGN(nlh->nlmsg_len);
2584 if (msglen > skb->len)
2585 msglen = skb->len;
2586 skb_pull(skb, msglen);
2587 }
2588
2589 return 0;
2590}
2591EXPORT_SYMBOL(netlink_rcv_skb);
2592
2593/**
2594 * nlmsg_notify - send a notification netlink message
2595 * @sk: netlink socket to use
2596 * @skb: notification message
2597 * @portid: destination netlink portid for reports or 0
2598 * @group: destination multicast group or 0
2599 * @report: 1 to report back, 0 to disable
2600 * @flags: allocation flags
2601 */
2602int nlmsg_notify(struct sock *sk, struct sk_buff *skb, u32 portid,
2603 unsigned int group, int report, gfp_t flags)
2604{
2605 int err = 0;
2606
2607 if (group) {
2608 int exclude_portid = 0;
2609
2610 if (report) {
2611 refcount_inc(&skb->users);
2612 exclude_portid = portid;
2613 }
2614
2615 /* errors reported via destination sk->sk_err, but propagate
2616 * delivery errors if NETLINK_BROADCAST_ERROR flag is set */
2617 err = nlmsg_multicast(sk, skb, exclude_portid, group, flags);
2618 if (err == -ESRCH)
2619 err = 0;
2620 }
2621
2622 if (report) {
2623 int err2;
2624
2625 err2 = nlmsg_unicast(sk, skb, portid);
2626 if (!err)
2627 err = err2;
2628 }
2629
2630 return err;
2631}
2632EXPORT_SYMBOL(nlmsg_notify);
2633
2634#ifdef CONFIG_PROC_FS
2635struct nl_seq_iter {
2636 struct seq_net_private p;
2637 struct rhashtable_iter hti;
2638 int link;
2639};
2640
2641static void netlink_walk_start(struct nl_seq_iter *iter)
2642{
2643 rhashtable_walk_enter(&nl_table[iter->link].hash, &iter->hti);
2644 rhashtable_walk_start(&iter->hti);
2645}
2646
2647static void netlink_walk_stop(struct nl_seq_iter *iter)
2648{
2649 rhashtable_walk_stop(&iter->hti);
2650 rhashtable_walk_exit(&iter->hti);
2651}
2652
2653static void *__netlink_seq_next(struct seq_file *seq)
2654{
2655 struct nl_seq_iter *iter = seq->private;
2656 struct netlink_sock *nlk;
2657
2658 do {
2659 for (;;) {
2660 nlk = rhashtable_walk_next(&iter->hti);
2661
2662 if (IS_ERR(nlk)) {
2663 if (PTR_ERR(nlk) == -EAGAIN)
2664 continue;
2665
2666 return nlk;
2667 }
2668
2669 if (nlk)
2670 break;
2671
2672 netlink_walk_stop(iter);
2673 if (++iter->link >= MAX_LINKS)
2674 return NULL;
2675
2676 netlink_walk_start(iter);
2677 }
2678 } while (sock_net(&nlk->sk) != seq_file_net(seq));
2679
2680 return nlk;
2681}
2682
2683static void *netlink_seq_start(struct seq_file *seq, loff_t *posp)
2684 __acquires(RCU)
2685{
2686 struct nl_seq_iter *iter = seq->private;
2687 void *obj = SEQ_START_TOKEN;
2688 loff_t pos;
2689
2690 iter->link = 0;
2691
2692 netlink_walk_start(iter);
2693
2694 for (pos = *posp; pos && obj && !IS_ERR(obj); pos--)
2695 obj = __netlink_seq_next(seq);
2696
2697 return obj;
2698}
2699
2700static void *netlink_seq_next(struct seq_file *seq, void *v, loff_t *pos)
2701{
2702 ++*pos;
2703 return __netlink_seq_next(seq);
2704}
2705
2706static void netlink_native_seq_stop(struct seq_file *seq, void *v)
2707{
2708 struct nl_seq_iter *iter = seq->private;
2709
2710 if (iter->link >= MAX_LINKS)
2711 return;
2712
2713 netlink_walk_stop(iter);
2714}
2715
2716
2717static int netlink_native_seq_show(struct seq_file *seq, void *v)
2718{
2719 if (v == SEQ_START_TOKEN) {
2720 seq_puts(seq,
2721 "sk Eth Pid Groups "
2722 "Rmem Wmem Dump Locks Drops Inode\n");
2723 } else {
2724 struct sock *s = v;
2725 struct netlink_sock *nlk = nlk_sk(s);
2726
2727 seq_printf(seq, "%pK %-3d %-10u %08x %-8d %-8d %-5d %-8d %-8u %-8lu\n",
2728 s,
2729 s->sk_protocol,
2730 nlk->portid,
2731 nlk->groups ? (u32)nlk->groups[0] : 0,
2732 sk_rmem_alloc_get(s),
2733 sk_wmem_alloc_get(s),
2734 nlk->cb_running,
2735 refcount_read(&s->sk_refcnt),
2736 atomic_read(&s->sk_drops),
2737 sock_i_ino(s)
2738 );
2739
2740 }
2741 return 0;
2742}
2743
2744#ifdef CONFIG_BPF_SYSCALL
2745struct bpf_iter__netlink {
2746 __bpf_md_ptr(struct bpf_iter_meta *, meta);
2747 __bpf_md_ptr(struct netlink_sock *, sk);
2748};
2749
2750DEFINE_BPF_ITER_FUNC(netlink, struct bpf_iter_meta *meta, struct netlink_sock *sk)
2751
2752static int netlink_prog_seq_show(struct bpf_prog *prog,
2753 struct bpf_iter_meta *meta,
2754 void *v)
2755{
2756 struct bpf_iter__netlink ctx;
2757
2758 meta->seq_num--; /* skip SEQ_START_TOKEN */
2759 ctx.meta = meta;
2760 ctx.sk = nlk_sk((struct sock *)v);
2761 return bpf_iter_run_prog(prog, &ctx);
2762}
2763
2764static int netlink_seq_show(struct seq_file *seq, void *v)
2765{
2766 struct bpf_iter_meta meta;
2767 struct bpf_prog *prog;
2768
2769 meta.seq = seq;
2770 prog = bpf_iter_get_info(&meta, false);
2771 if (!prog)
2772 return netlink_native_seq_show(seq, v);
2773
2774 if (v != SEQ_START_TOKEN)
2775 return netlink_prog_seq_show(prog, &meta, v);
2776
2777 return 0;
2778}
2779
2780static void netlink_seq_stop(struct seq_file *seq, void *v)
2781{
2782 struct bpf_iter_meta meta;
2783 struct bpf_prog *prog;
2784
2785 if (!v) {
2786 meta.seq = seq;
2787 prog = bpf_iter_get_info(&meta, true);
2788 if (prog)
2789 (void)netlink_prog_seq_show(prog, &meta, v);
2790 }
2791
2792 netlink_native_seq_stop(seq, v);
2793}
2794#else
2795static int netlink_seq_show(struct seq_file *seq, void *v)
2796{
2797 return netlink_native_seq_show(seq, v);
2798}
2799
2800static void netlink_seq_stop(struct seq_file *seq, void *v)
2801{
2802 netlink_native_seq_stop(seq, v);
2803}
2804#endif
2805
2806static const struct seq_operations netlink_seq_ops = {
2807 .start = netlink_seq_start,
2808 .next = netlink_seq_next,
2809 .stop = netlink_seq_stop,
2810 .show = netlink_seq_show,
2811};
2812#endif
2813
2814int netlink_register_notifier(struct notifier_block *nb)
2815{
2816 return blocking_notifier_chain_register(&netlink_chain, nb);
2817}
2818EXPORT_SYMBOL(netlink_register_notifier);
2819
2820int netlink_unregister_notifier(struct notifier_block *nb)
2821{
2822 return blocking_notifier_chain_unregister(&netlink_chain, nb);
2823}
2824EXPORT_SYMBOL(netlink_unregister_notifier);
2825
2826static const struct proto_ops netlink_ops = {
2827 .family = PF_NETLINK,
2828 .owner = THIS_MODULE,
2829 .release = netlink_release,
2830 .bind = netlink_bind,
2831 .connect = netlink_connect,
2832 .socketpair = sock_no_socketpair,
2833 .accept = sock_no_accept,
2834 .getname = netlink_getname,
2835 .poll = datagram_poll,
2836 .ioctl = netlink_ioctl,
2837 .listen = sock_no_listen,
2838 .shutdown = sock_no_shutdown,
2839 .setsockopt = netlink_setsockopt,
2840 .getsockopt = netlink_getsockopt,
2841 .sendmsg = netlink_sendmsg,
2842 .recvmsg = netlink_recvmsg,
2843 .mmap = sock_no_mmap,
2844 .sendpage = sock_no_sendpage,
2845};
2846
2847static const struct net_proto_family netlink_family_ops = {
2848 .family = PF_NETLINK,
2849 .create = netlink_create,
2850 .owner = THIS_MODULE, /* for consistency 8) */
2851};
2852
2853static int __net_init netlink_net_init(struct net *net)
2854{
2855#ifdef CONFIG_PROC_FS
2856 if (!proc_create_net("netlink", 0, net->proc_net, &netlink_seq_ops,
2857 sizeof(struct nl_seq_iter)))
2858 return -ENOMEM;
2859#endif
2860 return 0;
2861}
2862
2863static void __net_exit netlink_net_exit(struct net *net)
2864{
2865#ifdef CONFIG_PROC_FS
2866 remove_proc_entry("netlink", net->proc_net);
2867#endif
2868}
2869
2870static void __init netlink_add_usersock_entry(void)
2871{
2872 struct listeners *listeners;
2873 int groups = 32;
2874
2875 listeners = kzalloc(sizeof(*listeners) + NLGRPSZ(groups), GFP_KERNEL);
2876 if (!listeners)
2877 panic("netlink_add_usersock_entry: Cannot allocate listeners\n");
2878
2879 netlink_table_grab();
2880
2881 nl_table[NETLINK_USERSOCK].groups = groups;
2882 rcu_assign_pointer(nl_table[NETLINK_USERSOCK].listeners, listeners);
2883 nl_table[NETLINK_USERSOCK].module = THIS_MODULE;
2884 nl_table[NETLINK_USERSOCK].registered = 1;
2885 nl_table[NETLINK_USERSOCK].flags = NL_CFG_F_NONROOT_SEND;
2886
2887 netlink_table_ungrab();
2888}
2889
2890static struct pernet_operations __net_initdata netlink_net_ops = {
2891 .init = netlink_net_init,
2892 .exit = netlink_net_exit,
2893};
2894
2895static inline u32 netlink_hash(const void *data, u32 len, u32 seed)
2896{
2897 const struct netlink_sock *nlk = data;
2898 struct netlink_compare_arg arg;
2899
2900 netlink_compare_arg_init(&arg, sock_net(&nlk->sk), nlk->portid);
2901 return jhash2((u32 *)&arg, netlink_compare_arg_len / sizeof(u32), seed);
2902}
2903
2904static const struct rhashtable_params netlink_rhashtable_params = {
2905 .head_offset = offsetof(struct netlink_sock, node),
2906 .key_len = netlink_compare_arg_len,
2907 .obj_hashfn = netlink_hash,
2908 .obj_cmpfn = netlink_compare,
2909 .automatic_shrinking = true,
2910};
2911
2912#if defined(CONFIG_BPF_SYSCALL) && defined(CONFIG_PROC_FS)
2913BTF_ID_LIST(btf_netlink_sock_id)
2914BTF_ID(struct, netlink_sock)
2915
2916static const struct bpf_iter_seq_info netlink_seq_info = {
2917 .seq_ops = &netlink_seq_ops,
2918 .init_seq_private = bpf_iter_init_seq_net,
2919 .fini_seq_private = bpf_iter_fini_seq_net,
2920 .seq_priv_size = sizeof(struct nl_seq_iter),
2921};
2922
2923static struct bpf_iter_reg netlink_reg_info = {
2924 .target = "netlink",
2925 .ctx_arg_info_size = 1,
2926 .ctx_arg_info = {
2927 { offsetof(struct bpf_iter__netlink, sk),
2928 PTR_TO_BTF_ID_OR_NULL },
2929 },
2930 .seq_info = &netlink_seq_info,
2931};
2932
2933static int __init bpf_iter_register(void)
2934{
2935 netlink_reg_info.ctx_arg_info[0].btf_id = *btf_netlink_sock_id;
2936 return bpf_iter_reg_target(&netlink_reg_info);
2937}
2938#endif
2939
2940static int __init netlink_proto_init(void)
2941{
2942 int i;
2943 int err = proto_register(&netlink_proto, 0);
2944
2945 if (err != 0)
2946 goto out;
2947
2948#if defined(CONFIG_BPF_SYSCALL) && defined(CONFIG_PROC_FS)
2949 err = bpf_iter_register();
2950 if (err)
2951 goto out;
2952#endif
2953
2954 BUILD_BUG_ON(sizeof(struct netlink_skb_parms) > sizeof_field(struct sk_buff, cb));
2955
2956 nl_table = kcalloc(MAX_LINKS, sizeof(*nl_table), GFP_KERNEL);
2957 if (!nl_table)
2958 goto panic;
2959
2960 for (i = 0; i < MAX_LINKS; i++) {
2961 if (rhashtable_init(&nl_table[i].hash,
2962 &netlink_rhashtable_params) < 0) {
2963 while (--i > 0)
2964 rhashtable_destroy(&nl_table[i].hash);
2965 kfree(nl_table);
2966 goto panic;
2967 }
2968 }
2969
2970 netlink_add_usersock_entry();
2971
2972 sock_register(&netlink_family_ops);
2973 register_pernet_subsys(&netlink_net_ops);
2974 register_pernet_subsys(&netlink_tap_net_ops);
2975 /* The netlink device handler may be needed early. */
2976 rtnetlink_init();
2977out:
2978 return err;
2979panic:
2980 panic("netlink_init: Cannot allocate nl_table\n");
2981}
2982
2983core_initcall(netlink_proto_init);