]> git.proxmox.com Git - mirror_ubuntu-bionic-kernel.git/blob - net/netlink/af_netlink.c
projects / mirror_ubuntu-bionic-kernel.git / blob
? search:
summary | shortlog | log | commit | commitdiff | tree
blame | history | raw | HEAD
ipv4: Return EINVAL when ping_group_range sysctl doesn't map to user ns
[mirror_ubuntu-bionic-kernel.git] / net / netlink / af_netlink.c
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 #include <linux/net_namespace.h>
66
67 #include <net/net_namespace.h>
68 #include <net/sock.h>
69 #include <net/scm.h>
70 #include <net/netlink.h>
71
72 #include "af_netlink.h"
73
74 struct listeners {
75 struct rcu_head rcu;
76 unsigned long masks[0];
77 };
78
79 /* state bits */
80 #define NETLINK_S_CONGESTED 0x0
81
82 static inline int netlink_is_kernel(struct sock *sk)
83 {
84 return nlk_sk(sk)->flags & NETLINK_F_KERNEL_SOCKET;
85 }
86
87 struct netlink_table *nl_table __read_mostly;
88 EXPORT_SYMBOL_GPL(nl_table);
89
90 static DECLARE_WAIT_QUEUE_HEAD(nl_table_wait);
91
92 static struct lock_class_key nlk_cb_mutex_keys[MAX_LINKS];
93
94 static const char *const nlk_cb_mutex_key_strings[MAX_LINKS + 1] = {
95 "nlk_cb_mutex-ROUTE",
96 "nlk_cb_mutex-1",
97 "nlk_cb_mutex-USERSOCK",
98 "nlk_cb_mutex-FIREWALL",
99 "nlk_cb_mutex-SOCK_DIAG",
100 "nlk_cb_mutex-NFLOG",
101 "nlk_cb_mutex-XFRM",
102 "nlk_cb_mutex-SELINUX",
103 "nlk_cb_mutex-ISCSI",
104 "nlk_cb_mutex-AUDIT",
105 "nlk_cb_mutex-FIB_LOOKUP",
106 "nlk_cb_mutex-CONNECTOR",
107 "nlk_cb_mutex-NETFILTER",
108 "nlk_cb_mutex-IP6_FW",
109 "nlk_cb_mutex-DNRTMSG",
110 "nlk_cb_mutex-KOBJECT_UEVENT",
111 "nlk_cb_mutex-GENERIC",
112 "nlk_cb_mutex-17",
113 "nlk_cb_mutex-SCSITRANSPORT",
114 "nlk_cb_mutex-ECRYPTFS",
115 "nlk_cb_mutex-RDMA",
116 "nlk_cb_mutex-CRYPTO",
117 "nlk_cb_mutex-SMC",
118 "nlk_cb_mutex-23",
119 "nlk_cb_mutex-24",
120 "nlk_cb_mutex-25",
121 "nlk_cb_mutex-26",
122 "nlk_cb_mutex-27",
123 "nlk_cb_mutex-28",
124 "nlk_cb_mutex-29",
125 "nlk_cb_mutex-30",
126 "nlk_cb_mutex-31",
127 "nlk_cb_mutex-MAX_LINKS"
128 };
129
130 static int netlink_dump(struct sock *sk);
131
132 /* nl_table locking explained:
133 * Lookup and traversal are protected with an RCU read-side lock. Insertion
134 * and removal are protected with per bucket lock while using RCU list
135 * modification primitives and may run in parallel to RCU protected lookups.
136 * Destruction of the Netlink socket may only occur *after* nl_table_lock has
137 * been acquired * either during or after the socket has been removed from
138 * the list and after an RCU grace period.
139 */
140 DEFINE_RWLOCK(nl_table_lock);
141 EXPORT_SYMBOL_GPL(nl_table_lock);
142 static atomic_t nl_table_users = ATOMIC_INIT(0);
143
144 #define nl_deref_protected(X) rcu_dereference_protected(X, lockdep_is_held(&nl_table_lock));
145
146 static BLOCKING_NOTIFIER_HEAD(netlink_chain);
147
148 static DEFINE_SPINLOCK(netlink_tap_lock);
149 static struct list_head netlink_tap_all __read_mostly;
150
151 static const struct rhashtable_params netlink_rhashtable_params;
152
153 static inline u32 netlink_group_mask(u32 group)
154 {
155 return group ? 1 << (group - 1) : 0;
156 }
157
158 static struct sk_buff *netlink_to_full_skb(const struct sk_buff *skb,
159 gfp_t gfp_mask)
160 {
161 unsigned int len = skb_end_offset(skb);
162 struct sk_buff *new;
163
164 new = alloc_skb(len, gfp_mask);
165 if (new == NULL)
166 return NULL;
167
168 NETLINK_CB(new).portid = NETLINK_CB(skb).portid;
169 NETLINK_CB(new).dst_group = NETLINK_CB(skb).dst_group;
170 NETLINK_CB(new).creds = NETLINK_CB(skb).creds;
171
172 skb_put_data(new, skb->data, len);
173 return new;
174 }
175
176 int netlink_add_tap(struct netlink_tap *nt)
177 {
178 if (unlikely(nt->dev->type != ARPHRD_NETLINK))
179 return -EINVAL;
180
181 spin_lock(&netlink_tap_lock);
182 list_add_rcu(&nt->list, &netlink_tap_all);
183 spin_unlock(&netlink_tap_lock);
184
185 __module_get(nt->module);
186
187 return 0;
188 }
189 EXPORT_SYMBOL_GPL(netlink_add_tap);
190
191 static int __netlink_remove_tap(struct netlink_tap *nt)
192 {
193 bool found = false;
194 struct netlink_tap *tmp;
195
196 spin_lock(&netlink_tap_lock);
197
198 list_for_each_entry(tmp, &netlink_tap_all, list) {
199 if (nt == tmp) {
200 list_del_rcu(&nt->list);
201 found = true;
202 goto out;
203 }
204 }
205
206 pr_warn("__netlink_remove_tap: %p not found\n", nt);
207 out:
208 spin_unlock(&netlink_tap_lock);
209
210 if (found)
211 module_put(nt->module);
212
213 return found ? 0 : -ENODEV;
214 }
215
216 int netlink_remove_tap(struct netlink_tap *nt)
217 {
218 int ret;
219
220 ret = __netlink_remove_tap(nt);
221 synchronize_net();
222
223 return ret;
224 }
225 EXPORT_SYMBOL_GPL(netlink_remove_tap);
226
227 static bool netlink_filter_tap(const struct sk_buff *skb)
228 {
229 struct sock *sk = skb->sk;
230
231 /* We take the more conservative approach and
232 * whitelist socket protocols that may pass.
233 */
234 switch (sk->sk_protocol) {
235 case NETLINK_ROUTE:
236 case NETLINK_USERSOCK:
237 case NETLINK_SOCK_DIAG:
238 case NETLINK_NFLOG:
239 case NETLINK_XFRM:
240 case NETLINK_FIB_LOOKUP:
241 case NETLINK_NETFILTER:
242 case NETLINK_GENERIC:
243 return true;
244 }
245
246 return false;
247 }
248
249 static int __netlink_deliver_tap_skb(struct sk_buff *skb,
250 struct net_device *dev)
251 {
252 struct sk_buff *nskb;
253 struct sock *sk = skb->sk;
254 int ret = -ENOMEM;
255
256 if (!net_eq(dev_net(dev), sock_net(sk)))
257 return 0;
258
259 dev_hold(dev);
260
261 if (is_vmalloc_addr(skb->head))
262 nskb = netlink_to_full_skb(skb, GFP_ATOMIC);
263 else
264 nskb = skb_clone(skb, GFP_ATOMIC);
265 if (nskb) {
266 nskb->dev = dev;
267 nskb->protocol = htons((u16) sk->sk_protocol);
268 nskb->pkt_type = netlink_is_kernel(sk) ?
269 PACKET_KERNEL : PACKET_USER;
270 skb_reset_network_header(nskb);
271 ret = dev_queue_xmit(nskb);
272 if (unlikely(ret > 0))
273 ret = net_xmit_errno(ret);
274 }
275
276 dev_put(dev);
277 return ret;
278 }
279
280 static void __netlink_deliver_tap(struct sk_buff *skb)
281 {
282 int ret;
283 struct netlink_tap *tmp;
284
285 if (!netlink_filter_tap(skb))
286 return;
287
288 list_for_each_entry_rcu(tmp, &netlink_tap_all, list) {
289 ret = __netlink_deliver_tap_skb(skb, tmp->dev);
290 if (unlikely(ret))
291 break;
292 }
293 }
294
295 static void netlink_deliver_tap(struct sk_buff *skb)
296 {
297 rcu_read_lock();
298
299 if (unlikely(!list_empty(&netlink_tap_all)))
300 __netlink_deliver_tap(skb);
301
302 rcu_read_unlock();
303 }
304
305 static void netlink_deliver_tap_kernel(struct sock *dst, struct sock *src,
306 struct sk_buff *skb)
307 {
308 if (!(netlink_is_kernel(dst) && netlink_is_kernel(src)))
309 netlink_deliver_tap(skb);
310 }
311
312 static void netlink_overrun(struct sock *sk)
313 {
314 struct netlink_sock *nlk = nlk_sk(sk);
315
316 if (!(nlk->flags & NETLINK_F_RECV_NO_ENOBUFS)) {
317 if (!test_and_set_bit(NETLINK_S_CONGESTED,
318 &nlk_sk(sk)->state)) {
319 sk->sk_err = ENOBUFS;
320 sk->sk_error_report(sk);
321 }
322 }
323 atomic_inc(&sk->sk_drops);
324 }
325
326 static void netlink_rcv_wake(struct sock *sk)
327 {
328 struct netlink_sock *nlk = nlk_sk(sk);
329
330 if (skb_queue_empty(&sk->sk_receive_queue))
331 clear_bit(NETLINK_S_CONGESTED, &nlk->state);
332 if (!test_bit(NETLINK_S_CONGESTED, &nlk->state))
333 wake_up_interruptible(&nlk->wait);
334 }
335
336 static void netlink_skb_destructor(struct sk_buff *skb)
337 {
338 if (is_vmalloc_addr(skb->head)) {
339 if (!skb->cloned ||
340 !atomic_dec_return(&(skb_shinfo(skb)->dataref)))
341 vfree(skb->head);
342
343 skb->head = NULL;
344 }
345 if (skb->sk != NULL)
346 sock_rfree(skb);
347 }
348
349 static void netlink_skb_set_owner_r(struct sk_buff *skb, struct sock *sk)
350 {
351 WARN_ON(skb->sk != NULL);
352 skb->sk = sk;
353 skb->destructor = netlink_skb_destructor;
354 atomic_add(skb->truesize, &sk->sk_rmem_alloc);
355 sk_mem_charge(sk, skb->truesize);
356 }
357
358 static void netlink_sock_destruct(struct sock *sk)
359 {
360 struct netlink_sock *nlk = nlk_sk(sk);
361
362 if (nlk->cb_running) {
363 if (nlk->cb.done)
364 nlk->cb.done(&nlk->cb);
365 module_put(nlk->cb.module);
366 kfree_skb(nlk->cb.skb);
367 }
368
369 skb_queue_purge(&sk->sk_receive_queue);
370
371 if (!sock_flag(sk, SOCK_DEAD)) {
372 printk(KERN_ERR "Freeing alive netlink socket %p\n", sk);
373 return;
374 }
375
376 WARN_ON(atomic_read(&sk->sk_rmem_alloc));
377 WARN_ON(refcount_read(&sk->sk_wmem_alloc));
378 WARN_ON(nlk_sk(sk)->groups);
379 }
380
381 static void netlink_sock_destruct_work(struct work_struct *work)
382 {
383 struct netlink_sock *nlk = container_of(work, struct netlink_sock,
384 work);
385
386 sk_free(&nlk->sk);
387 }
388
389 /* This lock without WQ_FLAG_EXCLUSIVE is good on UP and it is _very_ bad on
390 * SMP. Look, when several writers sleep and reader wakes them up, all but one
391 * immediately hit write lock and grab all the cpus. Exclusive sleep solves
392 * this, _but_ remember, it adds useless work on UP machines.
393 */
394
395 void netlink_table_grab(void)
396 __acquires(nl_table_lock)
397 {
398 might_sleep();
399
400 write_lock_irq(&nl_table_lock);
401
402 if (atomic_read(&nl_table_users)) {
403 DECLARE_WAITQUEUE(wait, current);
404
405 add_wait_queue_exclusive(&nl_table_wait, &wait);
406 for (;;) {
407 set_current_state(TASK_UNINTERRUPTIBLE);
408 if (atomic_read(&nl_table_users) == 0)
409 break;
410 write_unlock_irq(&nl_table_lock);
411 schedule();
412 write_lock_irq(&nl_table_lock);
413 }
414
415 __set_current_state(TASK_RUNNING);
416 remove_wait_queue(&nl_table_wait, &wait);
417 }
418 }
419
420 void netlink_table_ungrab(void)
421 __releases(nl_table_lock)
422 {
423 write_unlock_irq(&nl_table_lock);
424 wake_up(&nl_table_wait);
425 }
426
427 static inline void
428 netlink_lock_table(void)
429 {
430 /* read_lock() synchronizes us to netlink_table_grab */
431
432 read_lock(&nl_table_lock);
433 atomic_inc(&nl_table_users);
434 read_unlock(&nl_table_lock);
435 }
436
437 static inline void
438 netlink_unlock_table(void)
439 {
440 if (atomic_dec_and_test(&nl_table_users))
441 wake_up(&nl_table_wait);
442 }
443
444 struct netlink_compare_arg
445 {
446 possible_net_t pnet;
447 u32 portid;
448 };
449
450 /* Doing sizeof directly may yield 4 extra bytes on 64-bit. */
451 #define netlink_compare_arg_len \
452 (offsetof(struct netlink_compare_arg, portid) + sizeof(u32))
453
454 static inline int netlink_compare(struct rhashtable_compare_arg *arg,
455 const void *ptr)
456 {
457 const struct netlink_compare_arg *x = arg->key;
458 const struct netlink_sock *nlk = ptr;
459
460 return nlk->portid != x->portid ||
461 !net_eq(sock_net(&nlk->sk), read_pnet(&x->pnet));
462 }
463
464 static void netlink_compare_arg_init(struct netlink_compare_arg *arg,
465 struct net *net, u32 portid)
466 {
467 memset(arg, 0, sizeof(*arg));
468 write_pnet(&arg->pnet, net);
469 arg->portid = portid;
470 }
471
472 static struct sock *__netlink_lookup(struct netlink_table *table, u32 portid,
473 struct net *net)
474 {
475 struct netlink_compare_arg arg;
476
477 netlink_compare_arg_init(&arg, net, portid);
478 return rhashtable_lookup_fast(&table->hash, &arg,
479 netlink_rhashtable_params);
480 }
481
482 static int __netlink_insert(struct netlink_table *table, struct sock *sk)
483 {
484 struct netlink_compare_arg arg;
485
486 netlink_compare_arg_init(&arg, sock_net(sk), nlk_sk(sk)->portid);
487 return rhashtable_lookup_insert_key(&table->hash, &arg,
488 &nlk_sk(sk)->node,
489 netlink_rhashtable_params);
490 }
491
492 static struct sock *netlink_lookup(struct net *net, int protocol, u32 portid)
493 {
494 struct netlink_table *table = &nl_table[protocol];
495 struct sock *sk;
496
497 rcu_read_lock();
498 sk = __netlink_lookup(table, portid, net);
499 if (sk)
500 sock_hold(sk);
501 rcu_read_unlock();
502
503 return sk;
504 }
505
506 static const struct proto_ops netlink_ops;
507
508 static void
509 netlink_update_listeners(struct sock *sk)
510 {
511 struct netlink_table *tbl = &nl_table[sk->sk_protocol];
512 unsigned long mask;
513 unsigned int i;
514 struct listeners *listeners;
515
516 listeners = nl_deref_protected(tbl->listeners);
517 if (!listeners)
518 return;
519
520 for (i = 0; i < NLGRPLONGS(tbl->groups); i++) {
521 mask = 0;
522 sk_for_each_bound(sk, &tbl->mc_list) {
523 if (i < NLGRPLONGS(nlk_sk(sk)->ngroups))
524 mask |= nlk_sk(sk)->groups[i];
525 }
526 listeners->masks[i] = mask;
527 }
528 /* this function is only called with the netlink table "grabbed", which
529 * makes sure updates are visible before bind or setsockopt return. */
530 }
531
532 static int netlink_insert(struct sock *sk, u32 portid)
533 {
534 struct netlink_table *table = &nl_table[sk->sk_protocol];
535 int err;
536
537 lock_sock(sk);
538
539 err = nlk_sk(sk)->portid == portid ? 0 : -EBUSY;
540 if (nlk_sk(sk)->bound)
541 goto err;
542
543 err = -ENOMEM;
544 if (BITS_PER_LONG > 32 &&
545 unlikely(atomic_read(&table->hash.nelems) >= UINT_MAX))
546 goto err;
547
548 nlk_sk(sk)->portid = portid;
549 sock_hold(sk);
550
551 err = __netlink_insert(table, sk);
552 if (err) {
553 /* In case the hashtable backend returns with -EBUSY
554 * from here, it must not escape to the caller.
555 */
556 if (unlikely(err == -EBUSY))
557 err = -EOVERFLOW;
558 if (err == -EEXIST)
559 err = -EADDRINUSE;
560 sock_put(sk);
561 goto err;
562 }
563
564 /* We need to ensure that the socket is hashed and visible. */
565 smp_wmb();
566 nlk_sk(sk)->bound = portid;
567
568 err:
569 release_sock(sk);
570 return err;
571 }
572
573 static void netlink_remove(struct sock *sk)
574 {
575 struct netlink_table *table;
576
577 table = &nl_table[sk->sk_protocol];
578 if (!rhashtable_remove_fast(&table->hash, &nlk_sk(sk)->node,
579 netlink_rhashtable_params)) {
580 WARN_ON(refcount_read(&sk->sk_refcnt) == 1);
581 __sock_put(sk);
582 }
583
584 netlink_table_grab();
585 if (nlk_sk(sk)->subscriptions) {
586 __sk_del_bind_node(sk);
587 netlink_update_listeners(sk);
588 }
589 if (sk->sk_protocol == NETLINK_GENERIC)
590 atomic_inc(&genl_sk_destructing_cnt);
591 netlink_table_ungrab();
592 }
593
594 static struct proto netlink_proto = {
595 .name = "NETLINK",
596 .owner = THIS_MODULE,
597 .obj_size = sizeof(struct netlink_sock),
598 };
599
600 static int __netlink_create(struct net *net, struct socket *sock,
601 struct mutex *cb_mutex, int protocol,
602 int kern)
603 {
604 struct sock *sk;
605 struct netlink_sock *nlk;
606
607 sock->ops = &netlink_ops;
608
609 sk = sk_alloc(net, PF_NETLINK, GFP_KERNEL, &netlink_proto, kern);
610 if (!sk)
611 return -ENOMEM;
612
613 sock_init_data(sock, sk);
614
615 nlk = nlk_sk(sk);
616 if (cb_mutex) {
617 nlk->cb_mutex = cb_mutex;
618 } else {
619 nlk->cb_mutex = &nlk->cb_def_mutex;
620 mutex_init(nlk->cb_mutex);
621 lockdep_set_class_and_name(nlk->cb_mutex,
622 nlk_cb_mutex_keys + protocol,
623 nlk_cb_mutex_key_strings[protocol]);
624 }
625 init_waitqueue_head(&nlk->wait);
626
627 sk->sk_destruct = netlink_sock_destruct;
628 sk->sk_protocol = protocol;
629 return 0;
630 }
631
632 static int netlink_create(struct net *net, struct socket *sock, int protocol,
633 int kern)
634 {
635 struct module *module = NULL;
636 struct mutex *cb_mutex;
637 struct netlink_sock *nlk;
638 int (*bind)(struct net *net, int group);
639 void (*unbind)(struct net *net, int group);
640 int err = 0;
641
642 sock->state = SS_UNCONNECTED;
643
644 if (sock->type != SOCK_RAW && sock->type != SOCK_DGRAM)
645 return -ESOCKTNOSUPPORT;
646
647 if (protocol < 0 || protocol >= MAX_LINKS)
648 return -EPROTONOSUPPORT;
649
650 netlink_lock_table();
651 #ifdef CONFIG_MODULES
652 if (!nl_table[protocol].registered) {
653 netlink_unlock_table();
654 request_module("net-pf-%d-proto-%d", PF_NETLINK, protocol);
655 netlink_lock_table();
656 }
657 #endif
658 if (nl_table[protocol].registered &&
659 try_module_get(nl_table[protocol].module))
660 module = nl_table[protocol].module;
661 else
662 err = -EPROTONOSUPPORT;
663 cb_mutex = nl_table[protocol].cb_mutex;
664 bind = nl_table[protocol].bind;
665 unbind = nl_table[protocol].unbind;
666 netlink_unlock_table();
667
668 if (err < 0)
669 goto out;
670
671 err = __netlink_create(net, sock, cb_mutex, protocol, kern);
672 if (err < 0)
673 goto out_module;
674
675 local_bh_disable();
676 sock_prot_inuse_add(net, &netlink_proto, 1);
677 local_bh_enable();
678
679 nlk = nlk_sk(sock->sk);
680 nlk->module = module;
681 nlk->netlink_bind = bind;
682 nlk->netlink_unbind = unbind;
683 out:
684 return err;
685
686 out_module:
687 module_put(module);
688 goto out;
689 }
690
691 static void deferred_put_nlk_sk(struct rcu_head *head)
692 {
693 struct netlink_sock *nlk = container_of(head, struct netlink_sock, rcu);
694 struct sock *sk = &nlk->sk;
695
696 kfree(nlk->groups);
697 nlk->groups = NULL;
698
699 if (!refcount_dec_and_test(&sk->sk_refcnt))
700 return;
701
702 if (nlk->cb_running && nlk->cb.done) {
703 INIT_WORK(&nlk->work, netlink_sock_destruct_work);
704 schedule_work(&nlk->work);
705 return;
706 }
707
708 sk_free(sk);
709 }
710
711 static int netlink_release(struct socket *sock)
712 {
713 struct sock *sk = sock->sk;
714 struct netlink_sock *nlk;
715
716 if (!sk)
717 return 0;
718
719 netlink_remove(sk);
720 sock_orphan(sk);
721 nlk = nlk_sk(sk);
722
723 /*
724 * OK. Socket is unlinked, any packets that arrive now
725 * will be purged.
726 */
727
728 /* must not acquire netlink_table_lock in any way again before unbind
729 * and notifying genetlink is done as otherwise it might deadlock
730 */
731 if (nlk->netlink_unbind) {
732 int i;
733
734 for (i = 0; i < nlk->ngroups; i++)
735 if (test_bit(i, nlk->groups))
736 nlk->netlink_unbind(sock_net(sk), i + 1);
737 }
738 if (sk->sk_protocol == NETLINK_GENERIC &&
739 atomic_dec_return(&genl_sk_destructing_cnt) == 0)
740 wake_up(&genl_sk_destructing_waitq);
741
742 sock->sk = NULL;
743 wake_up_interruptible_all(&nlk->wait);
744
745 skb_queue_purge(&sk->sk_write_queue);
746
747 if (nlk->portid && nlk->bound) {
748 struct netlink_notify n = {
749 .net = sock_net(sk),
750 .protocol = sk->sk_protocol,
751 .portid = nlk->portid,
752 };
753 blocking_notifier_call_chain(&netlink_chain,
754 NETLINK_URELEASE, &n);
755 }
756
757 module_put(nlk->module);
758
759 if (netlink_is_kernel(sk)) {
760 netlink_table_grab();
761 BUG_ON(nl_table[sk->sk_protocol].registered == 0);
762 if (--nl_table[sk->sk_protocol].registered == 0) {
763 struct listeners *old;
764
765 old = nl_deref_protected(nl_table[sk->sk_protocol].listeners);
766 RCU_INIT_POINTER(nl_table[sk->sk_protocol].listeners, NULL);
767 kfree_rcu(old, rcu);
768 nl_table[sk->sk_protocol].module = NULL;
769 nl_table[sk->sk_protocol].bind = NULL;
770 nl_table[sk->sk_protocol].unbind = NULL;
771 nl_table[sk->sk_protocol].flags = 0;
772 nl_table[sk->sk_protocol].registered = 0;
773 }
774 netlink_table_ungrab();
775 }
776
777 local_bh_disable();
778 sock_prot_inuse_add(sock_net(sk), &netlink_proto, -1);
779 local_bh_enable();
780 call_rcu(&nlk->rcu, deferred_put_nlk_sk);
781 return 0;
782 }
783
784 static int netlink_autobind(struct socket *sock)
785 {
786 struct sock *sk = sock->sk;
787 struct net *net = sock_net(sk);
788 struct netlink_table *table = &nl_table[sk->sk_protocol];
789 s32 portid = task_tgid_vnr(current);
790 int err;
791 s32 rover = -4096;
792 bool ok;
793
794 retry:
795 cond_resched();
796 rcu_read_lock();
797 ok = !__netlink_lookup(table, portid, net);
798 rcu_read_unlock();
799 if (!ok) {
800 /* Bind collision, search negative portid values. */
801 if (rover == -4096)
802 /* rover will be in range [S32_MIN, -4097] */
803 rover = S32_MIN + prandom_u32_max(-4096 - S32_MIN);
804 else if (rover >= -4096)
805 rover = -4097;
806 portid = rover--;
807 goto retry;
808 }
809
810 err = netlink_insert(sk, portid);
811 if (err == -EADDRINUSE)
812 goto retry;
813
814 /* If 2 threads race to autobind, that is fine. */
815 if (err == -EBUSY)
816 err = 0;
817
818 return err;
819 }
820
821 /**
822 * __netlink_ns_capable - General netlink message capability test
823 * @nsp: NETLINK_CB of the socket buffer holding a netlink command from userspace.
824 * @user_ns: The user namespace of the capability to use
825 * @cap: The capability to use
826 *
827 * Test to see if the opener of the socket we received the message
828 * from had when the netlink socket was created and the sender of the
829 * message has has the capability @cap in the user namespace @user_ns.
830 */
831 bool __netlink_ns_capable(const struct netlink_skb_parms *nsp,
832 struct user_namespace *user_ns, int cap)
833 {
834 return ((nsp->flags & NETLINK_SKB_DST) ||
835 file_ns_capable(nsp->sk->sk_socket->file, user_ns, cap)) &&
836 ns_capable(user_ns, cap);
837 }
838 EXPORT_SYMBOL(__netlink_ns_capable);
839
840 /**
841 * netlink_ns_capable - General netlink message capability test
842 * @skb: socket buffer holding a netlink command from userspace
843 * @user_ns: The user namespace of the capability to use
844 * @cap: The capability to use
845 *
846 * Test to see if the opener of the socket we received the message
847 * from had when the netlink socket was created and the sender of the
848 * message has has the capability @cap in the user namespace @user_ns.
849 */
850 bool netlink_ns_capable(const struct sk_buff *skb,
851 struct user_namespace *user_ns, int cap)
852 {
853 return __netlink_ns_capable(&NETLINK_CB(skb), user_ns, cap);
854 }
855 EXPORT_SYMBOL(netlink_ns_capable);
856
857 /**
858 * netlink_capable - Netlink global message capability test
859 * @skb: socket buffer holding a netlink command from userspace
860 * @cap: The capability to use
861 *
862 * Test to see if the opener of the socket we received the message
863 * from had when the netlink socket was created and the sender of the
864 * message has has the capability @cap in all user namespaces.
865 */
866 bool netlink_capable(const struct sk_buff *skb, int cap)
867 {
868 return netlink_ns_capable(skb, &init_user_ns, cap);
869 }
870 EXPORT_SYMBOL(netlink_capable);
871
872 /**
873 * netlink_net_capable - Netlink network namespace message capability test
874 * @skb: socket buffer holding a netlink command from userspace
875 * @cap: The capability to use
876 *
877 * Test to see if the opener of the socket we received the message
878 * from had when the netlink socket was created and the sender of the
879 * message has has the capability @cap over the network namespace of
880 * the socket we received the message from.
881 */
882 bool netlink_net_capable(const struct sk_buff *skb, int cap)
883 {
884 return netlink_ns_capable(skb, sock_net(skb->sk)->user_ns, cap);
885 }
886 EXPORT_SYMBOL(netlink_net_capable);
887
888 static inline int netlink_allowed(const struct socket *sock, unsigned int flag)
889 {
890 return (nl_table[sock->sk->sk_protocol].flags & flag) ||
891 ns_capable(sock_net(sock->sk)->user_ns, CAP_NET_ADMIN);
892 }
893
894 static void
895 netlink_update_subscriptions(struct sock *sk, unsigned int subscriptions)
896 {
897 struct netlink_sock *nlk = nlk_sk(sk);
898
899 if (nlk->subscriptions && !subscriptions)
900 __sk_del_bind_node(sk);
901 else if (!nlk->subscriptions && subscriptions)
902 sk_add_bind_node(sk, &nl_table[sk->sk_protocol].mc_list);
903 nlk->subscriptions = subscriptions;
904 }
905
906 static int netlink_realloc_groups(struct sock *sk)
907 {
908 struct netlink_sock *nlk = nlk_sk(sk);
909 unsigned int groups;
910 unsigned long *new_groups;
911 int err = 0;
912
913 netlink_table_grab();
914
915 groups = nl_table[sk->sk_protocol].groups;
916 if (!nl_table[sk->sk_protocol].registered) {
917 err = -ENOENT;
918 goto out_unlock;
919 }
920
921 if (nlk->ngroups >= groups)
922 goto out_unlock;
923
924 new_groups = krealloc(nlk->groups, NLGRPSZ(groups), GFP_ATOMIC);
925 if (new_groups == NULL) {
926 err = -ENOMEM;
927 goto out_unlock;
928 }
929 memset((char *)new_groups + NLGRPSZ(nlk->ngroups), 0,
930 NLGRPSZ(groups) - NLGRPSZ(nlk->ngroups));
931
932 nlk->groups = new_groups;
933 nlk->ngroups = groups;
934 out_unlock:
935 netlink_table_ungrab();
936 return err;
937 }
938
939 static void netlink_undo_bind(int group, long unsigned int groups,
940 struct sock *sk)
941 {
942 struct netlink_sock *nlk = nlk_sk(sk);
943 int undo;
944
945 if (!nlk->netlink_unbind)
946 return;
947
948 for (undo = 0; undo < group; undo++)
949 if (test_bit(undo, &groups))
950 nlk->netlink_unbind(sock_net(sk), undo + 1);
951 }
952
953 static int netlink_bind(struct socket *sock, struct sockaddr *addr,
954 int addr_len)
955 {
956 struct sock *sk = sock->sk;
957 struct net *net = sock_net(sk);
958 struct netlink_sock *nlk = nlk_sk(sk);
959 struct sockaddr_nl *nladdr = (struct sockaddr_nl *)addr;
960 int err = 0;
961 long unsigned int groups = nladdr->nl_groups;
962 bool bound;
963
964 if (addr_len < sizeof(struct sockaddr_nl))
965 return -EINVAL;
966
967 if (nladdr->nl_family != AF_NETLINK)
968 return -EINVAL;
969
970 /* Only superuser is allowed to listen multicasts */
971 if (groups) {
972 if (!netlink_allowed(sock, NL_CFG_F_NONROOT_RECV))
973 return -EPERM;
974 err = netlink_realloc_groups(sk);
975 if (err)
976 return err;
977 }
978
979 bound = nlk->bound;
980 if (bound) {
981 /* Ensure nlk->portid is up-to-date. */
982 smp_rmb();
983
984 if (nladdr->nl_pid != nlk->portid)
985 return -EINVAL;
986 }
987
988 netlink_lock_table();
989 if (nlk->netlink_bind && groups) {
990 int group;
991
992 for (group = 0; group < nlk->ngroups; group++) {
993 if (!test_bit(group, &groups))
994 continue;
995 err = nlk->netlink_bind(net, group + 1);
996 if (!err)
997 continue;
998 netlink_undo_bind(group, groups, sk);
999 goto unlock;
1000 }
1001 }
1002
1003 /* No need for barriers here as we return to user-space without
1004 * using any of the bound attributes.
1005 */
1006 if (!bound) {
1007 err = nladdr->nl_pid ?
1008 netlink_insert(sk, nladdr->nl_pid) :
1009 netlink_autobind(sock);
1010 if (err) {
1011 netlink_undo_bind(nlk->ngroups, groups, sk);
1012 goto unlock;
1013 }
1014 }
1015
1016 if (!groups && (nlk->groups == NULL || !(u32)nlk->groups[0]))
1017 goto unlock;
1018 netlink_unlock_table();
1019
1020 netlink_table_grab();
1021 netlink_update_subscriptions(sk, nlk->subscriptions +
1022 hweight32(groups) -
1023 hweight32(nlk->groups[0]));
1024 nlk->groups[0] = (nlk->groups[0] & ~0xffffffffUL) | groups;
1025 netlink_update_listeners(sk);
1026 netlink_table_ungrab();
1027
1028 return 0;
1029
1030 unlock:
1031 netlink_unlock_table();
1032 return err;
1033 }
1034
1035 static int netlink_connect(struct socket *sock, struct sockaddr *addr,
1036 int alen, int flags)
1037 {
1038 int err = 0;
1039 struct sock *sk = sock->sk;
1040 struct netlink_sock *nlk = nlk_sk(sk);
1041 struct sockaddr_nl *nladdr = (struct sockaddr_nl *)addr;
1042
1043 if (alen < sizeof(addr->sa_family))
1044 return -EINVAL;
1045
1046 if (addr->sa_family == AF_UNSPEC) {
1047 sk->sk_state = NETLINK_UNCONNECTED;
1048 nlk->dst_portid = 0;
1049 nlk->dst_group = 0;
1050 return 0;
1051 }
1052 if (addr->sa_family != AF_NETLINK)
1053 return -EINVAL;
1054
1055 if (alen < sizeof(struct sockaddr_nl))
1056 return -EINVAL;
1057
1058 if ((nladdr->nl_groups || nladdr->nl_pid) &&
1059 !netlink_allowed(sock, NL_CFG_F_NONROOT_SEND))
1060 return -EPERM;
1061
1062 /* No need for barriers here as we return to user-space without
1063 * using any of the bound attributes.
1064 */
1065 if (!nlk->bound)
1066 err = netlink_autobind(sock);
1067
1068 if (err == 0) {
1069 sk->sk_state = NETLINK_CONNECTED;
1070 nlk->dst_portid = nladdr->nl_pid;
1071 nlk->dst_group = ffs(nladdr->nl_groups);
1072 }
1073
1074 return err;
1075 }
1076
1077 static int netlink_getname(struct socket *sock, struct sockaddr *addr,
1078 int *addr_len, int peer)
1079 {
1080 struct sock *sk = sock->sk;
1081 struct netlink_sock *nlk = nlk_sk(sk);
1082 DECLARE_SOCKADDR(struct sockaddr_nl *, nladdr, addr);
1083
1084 nladdr->nl_family = AF_NETLINK;
1085 nladdr->nl_pad = 0;
1086 *addr_len = sizeof(*nladdr);
1087
1088 if (peer) {
1089 nladdr->nl_pid = nlk->dst_portid;
1090 nladdr->nl_groups = netlink_group_mask(nlk->dst_group);
1091 } else {
1092 nladdr->nl_pid = nlk->portid;
1093 netlink_lock_table();
1094 nladdr->nl_groups = nlk->groups ? nlk->groups[0] : 0;
1095 netlink_unlock_table();
1096 }
1097 return 0;
1098 }
1099
1100 static int netlink_ioctl(struct socket *sock, unsigned int cmd,
1101 unsigned long arg)
1102 {
1103 /* try to hand this ioctl down to the NIC drivers.
1104 */
1105 return -ENOIOCTLCMD;
1106 }
1107
1108 static struct sock *netlink_getsockbyportid(struct sock *ssk, u32 portid)
1109 {
1110 struct sock *sock;
1111 struct netlink_sock *nlk;
1112
1113 sock = netlink_lookup(sock_net(ssk), ssk->sk_protocol, portid);
1114 if (!sock)
1115 return ERR_PTR(-ECONNREFUSED);
1116
1117 /* Don't bother queuing skb if kernel socket has no input function */
1118 nlk = nlk_sk(sock);
1119 if (sock->sk_state == NETLINK_CONNECTED &&
1120 nlk->dst_portid != nlk_sk(ssk)->portid) {
1121 sock_put(sock);
1122 return ERR_PTR(-ECONNREFUSED);
1123 }
1124 return sock;
1125 }
1126
1127 struct sock *netlink_getsockbyfilp(struct file *filp)
1128 {
1129 struct inode *inode = file_inode(filp);
1130 struct sock *sock;
1131
1132 if (!S_ISSOCK(inode->i_mode))
1133 return ERR_PTR(-ENOTSOCK);
1134
1135 sock = SOCKET_I(inode)->sk;
1136 if (sock->sk_family != AF_NETLINK)
1137 return ERR_PTR(-EINVAL);
1138
1139 sock_hold(sock);
1140 return sock;
1141 }
1142
1143 static struct sk_buff *netlink_alloc_large_skb(unsigned int size,
1144 int broadcast)
1145 {
1146 struct sk_buff *skb;
1147 void *data;
1148
1149 if (size <= NLMSG_GOODSIZE || broadcast)
1150 return alloc_skb(size, GFP_KERNEL);
1151
1152 size = SKB_DATA_ALIGN(size) +
1153 SKB_DATA_ALIGN(sizeof(struct skb_shared_info));
1154
1155 data = vmalloc(size);
1156 if (data == NULL)
1157 return NULL;
1158
1159 skb = __build_skb(data, size);
1160 if (skb == NULL)
1161 vfree(data);
1162 else
1163 skb->destructor = netlink_skb_destructor;
1164
1165 return skb;
1166 }
1167
1168 /*
1169 * Attach a skb to a netlink socket.
1170 * The caller must hold a reference to the destination socket. On error, the
1171 * reference is dropped. The skb is not send to the destination, just all
1172 * all error checks are performed and memory in the queue is reserved.
1173 * Return values:
1174 * < 0: error. skb freed, reference to sock dropped.
1175 * 0: continue
1176 * 1: repeat lookup - reference dropped while waiting for socket memory.
1177 */
1178 int netlink_attachskb(struct sock *sk, struct sk_buff *skb,
1179 long *timeo, struct sock *ssk)
1180 {
1181 struct netlink_sock *nlk;
1182
1183 nlk = nlk_sk(sk);
1184
1185 if ((atomic_read(&sk->sk_rmem_alloc) > sk->sk_rcvbuf ||
1186 test_bit(NETLINK_S_CONGESTED, &nlk->state))) {
1187 DECLARE_WAITQUEUE(wait, current);
1188 if (!*timeo) {
1189 if (!ssk || netlink_is_kernel(ssk))
1190 netlink_overrun(sk);
1191 sock_put(sk);
1192 kfree_skb(skb);
1193 return -EAGAIN;
1194 }
1195
1196 __set_current_state(TASK_INTERRUPTIBLE);
1197 add_wait_queue(&nlk->wait, &wait);
1198
1199 if ((atomic_read(&sk->sk_rmem_alloc) > sk->sk_rcvbuf ||
1200 test_bit(NETLINK_S_CONGESTED, &nlk->state)) &&
1201 !sock_flag(sk, SOCK_DEAD))
1202 *timeo = schedule_timeout(*timeo);
1203
1204 __set_current_state(TASK_RUNNING);
1205 remove_wait_queue(&nlk->wait, &wait);
1206 sock_put(sk);
1207
1208 if (signal_pending(current)) {
1209 kfree_skb(skb);
1210 return sock_intr_errno(*timeo);
1211 }
1212 return 1;
1213 }
1214 netlink_skb_set_owner_r(skb, sk);
1215 return 0;
1216 }
1217
1218 static int __netlink_sendskb(struct sock *sk, struct sk_buff *skb)
1219 {
1220 int len = skb->len;
1221
1222 netlink_deliver_tap(skb);
1223
1224 skb_queue_tail(&sk->sk_receive_queue, skb);
1225 sk->sk_data_ready(sk);
1226 return len;
1227 }
1228
1229 int netlink_sendskb(struct sock *sk, struct sk_buff *skb)
1230 {
1231 int len = __netlink_sendskb(sk, skb);
1232
1233 sock_put(sk);
1234 return len;
1235 }
1236
1237 void netlink_detachskb(struct sock *sk, struct sk_buff *skb)
1238 {
1239 kfree_skb(skb);
1240 sock_put(sk);
1241 }
1242
1243 static struct sk_buff *netlink_trim(struct sk_buff *skb, gfp_t allocation)
1244 {
1245 int delta;
1246
1247 WARN_ON(skb->sk != NULL);
1248 delta = skb->end - skb->tail;
1249 if (is_vmalloc_addr(skb->head) || delta * 2 < skb->truesize)
1250 return skb;
1251
1252 if (skb_shared(skb)) {
1253 struct sk_buff *nskb = skb_clone(skb, allocation);
1254 if (!nskb)
1255 return skb;
1256 consume_skb(skb);
1257 skb = nskb;
1258 }
1259
1260 pskb_expand_head(skb, 0, -delta,
1261 (allocation & ~__GFP_DIRECT_RECLAIM) |
1262 __GFP_NOWARN | __GFP_NORETRY);
1263 return skb;
1264 }
1265
1266 static int netlink_unicast_kernel(struct sock *sk, struct sk_buff *skb,
1267 struct sock *ssk)
1268 {
1269 int ret;
1270 struct netlink_sock *nlk = nlk_sk(sk);
1271
1272 ret = -ECONNREFUSED;
1273 if (nlk->netlink_rcv != NULL) {
1274 ret = skb->len;
1275 netlink_skb_set_owner_r(skb, sk);
1276 NETLINK_CB(skb).sk = ssk;
1277 netlink_deliver_tap_kernel(sk, ssk, skb);
1278 nlk->netlink_rcv(skb);
1279 consume_skb(skb);
1280 } else {
1281 kfree_skb(skb);
1282 }
1283 sock_put(sk);
1284 return ret;
1285 }
1286
1287 int netlink_unicast(struct sock *ssk, struct sk_buff *skb,
1288 u32 portid, int nonblock)
1289 {
1290 struct sock *sk;
1291 int err;
1292 long timeo;
1293
1294 skb = netlink_trim(skb, gfp_any());
1295
1296 timeo = sock_sndtimeo(ssk, nonblock);
1297 retry:
1298 sk = netlink_getsockbyportid(ssk, portid);
1299 if (IS_ERR(sk)) {
1300 kfree_skb(skb);
1301 return PTR_ERR(sk);
1302 }
1303 if (netlink_is_kernel(sk))
1304 return netlink_unicast_kernel(sk, skb, ssk);
1305
1306 if (sk_filter(sk, skb)) {
1307 err = skb->len;
1308 kfree_skb(skb);
1309 sock_put(sk);
1310 return err;
1311 }
1312
1313 err = netlink_attachskb(sk, skb, &timeo, ssk);
1314 if (err == 1)
1315 goto retry;
1316 if (err)
1317 return err;
1318
1319 return netlink_sendskb(sk, skb);
1320 }
1321 EXPORT_SYMBOL(netlink_unicast);
1322
1323 int netlink_has_listeners(struct sock *sk, unsigned int group)
1324 {
1325 int res = 0;
1326 struct listeners *listeners;
1327
1328 BUG_ON(!netlink_is_kernel(sk));
1329
1330 rcu_read_lock();
1331 listeners = rcu_dereference(nl_table[sk->sk_protocol].listeners);
1332
1333 if (listeners && group - 1 < nl_table[sk->sk_protocol].groups)
1334 res = test_bit(group - 1, listeners->masks);
1335
1336 rcu_read_unlock();
1337
1338 return res;
1339 }
1340 EXPORT_SYMBOL_GPL(netlink_has_listeners);
1341
1342 static int netlink_broadcast_deliver(struct sock *sk, struct sk_buff *skb)
1343 {
1344 struct netlink_sock *nlk = nlk_sk(sk);
1345
1346 if (atomic_read(&sk->sk_rmem_alloc) <= sk->sk_rcvbuf &&
1347 !test_bit(NETLINK_S_CONGESTED, &nlk->state)) {
1348 netlink_skb_set_owner_r(skb, sk);
1349 __netlink_sendskb(sk, skb);
1350 return atomic_read(&sk->sk_rmem_alloc) > (sk->sk_rcvbuf >> 1);
1351 }
1352 return -1;
1353 }
1354
1355 struct netlink_broadcast_data {
1356 struct sock *exclude_sk;
1357 struct net *net;
1358 u32 portid;
1359 u32 group;
1360 int failure;
1361 int delivery_failure;
1362 int congested;
1363 int delivered;
1364 gfp_t allocation;
1365 struct sk_buff *skb, *skb2;
1366 int (*tx_filter)(struct sock *dsk, struct sk_buff *skb, void *data);
1367 void *tx_data;
1368 };
1369
1370 static void do_one_broadcast(struct sock *sk,
1371 struct netlink_broadcast_data *p)
1372 {
1373 struct netlink_sock *nlk = nlk_sk(sk);
1374 int val;
1375
1376 if (p->exclude_sk == sk)
1377 return;
1378
1379 if (nlk->portid == p->portid || p->group - 1 >= nlk->ngroups ||
1380 !test_bit(p->group - 1, nlk->groups))
1381 return;
1382
1383 if (!net_eq(sock_net(sk), p->net)) {
1384 if (!(nlk->flags & NETLINK_F_LISTEN_ALL_NSID))
1385 return;
1386
1387 if (!peernet_has_id(sock_net(sk), p->net))
1388 return;
1389
1390 if (!file_ns_capable(sk->sk_socket->file, p->net->user_ns,
1391 CAP_NET_BROADCAST))
1392 return;
1393 }
1394
1395 if (p->failure) {
1396 netlink_overrun(sk);
1397 return;
1398 }
1399
1400 sock_hold(sk);
1401 if (p->skb2 == NULL) {
1402 if (skb_shared(p->skb)) {
1403 p->skb2 = skb_clone(p->skb, p->allocation);
1404 } else {
1405 p->skb2 = skb_get(p->skb);
1406 /*
1407 * skb ownership may have been set when
1408 * delivered to a previous socket.
1409 */
1410 skb_orphan(p->skb2);
1411 }
1412 }
1413 if (p->skb2 == NULL) {
1414 netlink_overrun(sk);
1415 /* Clone failed. Notify ALL listeners. */
1416 p->failure = 1;
1417 if (nlk->flags & NETLINK_F_BROADCAST_SEND_ERROR)
1418 p->delivery_failure = 1;
1419 goto out;
1420 }
1421 if (p->tx_filter && p->tx_filter(sk, p->skb2, p->tx_data)) {
1422 kfree_skb(p->skb2);
1423 p->skb2 = NULL;
1424 goto out;
1425 }
1426 if (sk_filter(sk, p->skb2)) {
1427 kfree_skb(p->skb2);
1428 p->skb2 = NULL;
1429 goto out;
1430 }
1431 NETLINK_CB(p->skb2).nsid = peernet2id(sock_net(sk), p->net);
1432 if (NETLINK_CB(p->skb2).nsid != NETNSA_NSID_NOT_ASSIGNED)
1433 NETLINK_CB(p->skb2).nsid_is_set = true;
1434 val = netlink_broadcast_deliver(sk, p->skb2);
1435 if (val < 0) {
1436 netlink_overrun(sk);
1437 if (nlk->flags & NETLINK_F_BROADCAST_SEND_ERROR)
1438 p->delivery_failure = 1;
1439 } else {
1440 p->congested |= val;
1441 p->delivered = 1;
1442 p->skb2 = NULL;
1443 }
1444 out:
1445 sock_put(sk);
1446 }
1447
1448 int netlink_broadcast_filtered(struct sock *ssk, struct sk_buff *skb, u32 portid,
1449 u32 group, gfp_t allocation,
1450 int (*filter)(struct sock *dsk, struct sk_buff *skb, void *data),
1451 void *filter_data)
1452 {
1453 struct net *net = sock_net(ssk);
1454 struct netlink_broadcast_data info;
1455 struct sock *sk;
1456
1457 skb = netlink_trim(skb, allocation);
1458
1459 info.exclude_sk = ssk;
1460 info.net = net;
1461 info.portid = portid;
1462 info.group = group;
1463 info.failure = 0;
1464 info.delivery_failure = 0;
1465 info.congested = 0;
1466 info.delivered = 0;
1467 info.allocation = allocation;
1468 info.skb = skb;
1469 info.skb2 = NULL;
1470 info.tx_filter = filter;
1471 info.tx_data = filter_data;
1472
1473 /* While we sleep in clone, do not allow to change socket list */
1474
1475 netlink_lock_table();
1476
1477 sk_for_each_bound(sk, &nl_table[ssk->sk_protocol].mc_list)
1478 do_one_broadcast(sk, &info);
1479
1480 consume_skb(skb);
1481
1482 netlink_unlock_table();
1483
1484 if (info.delivery_failure) {
1485 kfree_skb(info.skb2);
1486 return -ENOBUFS;
1487 }
1488 consume_skb(info.skb2);
1489
1490 if (info.delivered) {
1491 if (info.congested && gfpflags_allow_blocking(allocation))
1492 yield();
1493 return 0;
1494 }
1495 return -ESRCH;
1496 }
1497 EXPORT_SYMBOL(netlink_broadcast_filtered);
1498
1499 int netlink_broadcast(struct sock *ssk, struct sk_buff *skb, u32 portid,
1500 u32 group, gfp_t allocation)
1501 {
1502 return netlink_broadcast_filtered(ssk, skb, portid, group, allocation,
1503 NULL, NULL);
1504 }
1505 EXPORT_SYMBOL(netlink_broadcast);
1506
1507 struct netlink_set_err_data {
1508 struct sock *exclude_sk;
1509 u32 portid;
1510 u32 group;
1511 int code;
1512 };
1513
1514 static int do_one_set_err(struct sock *sk, struct netlink_set_err_data *p)
1515 {
1516 struct netlink_sock *nlk = nlk_sk(sk);
1517 int ret = 0;
1518
1519 if (sk == p->exclude_sk)
1520 goto out;
1521
1522 if (!net_eq(sock_net(sk), sock_net(p->exclude_sk)))
1523 goto out;
1524
1525 if (nlk->portid == p->portid || p->group - 1 >= nlk->ngroups ||
1526 !test_bit(p->group - 1, nlk->groups))
1527 goto out;
1528
1529 if (p->code == ENOBUFS && nlk->flags & NETLINK_F_RECV_NO_ENOBUFS) {
1530 ret = 1;
1531 goto out;
1532 }
1533
1534 sk->sk_err = p->code;
1535 sk->sk_error_report(sk);
1536 out:
1537 return ret;
1538 }
1539
1540 /**
1541 * netlink_set_err - report error to broadcast listeners
1542 * @ssk: the kernel netlink socket, as returned by netlink_kernel_create()
1543 * @portid: the PORTID of a process that we want to skip (if any)
1544 * @group: the broadcast group that will notice the error
1545 * @code: error code, must be negative (as usual in kernelspace)
1546 *
1547 * This function returns the number of broadcast listeners that have set the
1548 * NETLINK_NO_ENOBUFS socket option.
1549 */
1550 int netlink_set_err(struct sock *ssk, u32 portid, u32 group, int code)
1551 {
1552 struct netlink_set_err_data info;
1553 struct sock *sk;
1554 int ret = 0;
1555
1556 info.exclude_sk = ssk;
1557 info.portid = portid;
1558 info.group = group;
1559 /* sk->sk_err wants a positive error value */
1560 info.code = -code;
1561
1562 read_lock(&nl_table_lock);
1563
1564 sk_for_each_bound(sk, &nl_table[ssk->sk_protocol].mc_list)
1565 ret += do_one_set_err(sk, &info);
1566
1567 read_unlock(&nl_table_lock);
1568 return ret;
1569 }
1570 EXPORT_SYMBOL(netlink_set_err);
1571
1572 /* must be called with netlink table grabbed */
1573 static void netlink_update_socket_mc(struct netlink_sock *nlk,
1574 unsigned int group,
1575 int is_new)
1576 {
1577 int old, new = !!is_new, subscriptions;
1578
1579 old = test_bit(group - 1, nlk->groups);
1580 subscriptions = nlk->subscriptions - old + new;
1581 if (new)
1582 __set_bit(group - 1, nlk->groups);
1583 else
1584 __clear_bit(group - 1, nlk->groups);
1585 netlink_update_subscriptions(&nlk->sk, subscriptions);
1586 netlink_update_listeners(&nlk->sk);
1587 }
1588
1589 static int netlink_setsockopt(struct socket *sock, int level, int optname,
1590 char __user *optval, unsigned int optlen)
1591 {
1592 struct sock *sk = sock->sk;
1593 struct netlink_sock *nlk = nlk_sk(sk);
1594 unsigned int val = 0;
1595 int err;
1596
1597 if (level != SOL_NETLINK)
1598 return -ENOPROTOOPT;
1599
1600 if (optlen >= sizeof(int) &&
1601 get_user(val, (unsigned int __user *)optval))
1602 return -EFAULT;
1603
1604 switch (optname) {
1605 case NETLINK_PKTINFO:
1606 if (val)
1607 nlk->flags |= NETLINK_F_RECV_PKTINFO;
1608 else
1609 nlk->flags &= ~NETLINK_F_RECV_PKTINFO;
1610 err = 0;
1611 break;
1612 case NETLINK_ADD_MEMBERSHIP:
1613 case NETLINK_DROP_MEMBERSHIP: {
1614 if (!netlink_allowed(sock, NL_CFG_F_NONROOT_RECV))
1615 return -EPERM;
1616 err = netlink_realloc_groups(sk);
1617 if (err)
1618 return err;
1619 if (!val || val - 1 >= nlk->ngroups)
1620 return -EINVAL;
1621 if (optname == NETLINK_ADD_MEMBERSHIP && nlk->netlink_bind) {
1622 err = nlk->netlink_bind(sock_net(sk), val);
1623 if (err)
1624 return err;
1625 }
1626 netlink_table_grab();
1627 netlink_update_socket_mc(nlk, val,
1628 optname == NETLINK_ADD_MEMBERSHIP);
1629 netlink_table_ungrab();
1630 if (optname == NETLINK_DROP_MEMBERSHIP && nlk->netlink_unbind)
1631 nlk->netlink_unbind(sock_net(sk), val);
1632
1633 err = 0;
1634 break;
1635 }
1636 case NETLINK_BROADCAST_ERROR:
1637 if (val)
1638 nlk->flags |= NETLINK_F_BROADCAST_SEND_ERROR;
1639 else
1640 nlk->flags &= ~NETLINK_F_BROADCAST_SEND_ERROR;
1641 err = 0;
1642 break;
1643 case NETLINK_NO_ENOBUFS:
1644 if (val) {
1645 nlk->flags |= NETLINK_F_RECV_NO_ENOBUFS;
1646 clear_bit(NETLINK_S_CONGESTED, &nlk->state);
1647 wake_up_interruptible(&nlk->wait);
1648 } else {
1649 nlk->flags &= ~NETLINK_F_RECV_NO_ENOBUFS;
1650 }
1651 err = 0;
1652 break;
1653 case NETLINK_LISTEN_ALL_NSID:
1654 if (!ns_capable(sock_net(sk)->user_ns, CAP_NET_BROADCAST))
1655 return -EPERM;
1656
1657 if (val)
1658 nlk->flags |= NETLINK_F_LISTEN_ALL_NSID;
1659 else
1660 nlk->flags &= ~NETLINK_F_LISTEN_ALL_NSID;
1661 err = 0;
1662 break;
1663 case NETLINK_CAP_ACK:
1664 if (val)
1665 nlk->flags |= NETLINK_F_CAP_ACK;
1666 else
1667 nlk->flags &= ~NETLINK_F_CAP_ACK;
1668 err = 0;
1669 break;
1670 case NETLINK_EXT_ACK:
1671 if (val)
1672 nlk->flags |= NETLINK_F_EXT_ACK;
1673 else
1674 nlk->flags &= ~NETLINK_F_EXT_ACK;
1675 err = 0;
1676 break;
1677 default:
1678 err = -ENOPROTOOPT;
1679 }
1680 return err;
1681 }
1682
1683 static int netlink_getsockopt(struct socket *sock, int level, int optname,
1684 char __user *optval, int __user *optlen)
1685 {
1686 struct sock *sk = sock->sk;
1687 struct netlink_sock *nlk = nlk_sk(sk);
1688 int len, val, err;
1689
1690 if (level != SOL_NETLINK)
1691 return -ENOPROTOOPT;
1692
1693 if (get_user(len, optlen))
1694 return -EFAULT;
1695 if (len < 0)
1696 return -EINVAL;
1697
1698 switch (optname) {
1699 case NETLINK_PKTINFO:
1700 if (len < sizeof(int))
1701 return -EINVAL;
1702 len = sizeof(int);
1703 val = nlk->flags & NETLINK_F_RECV_PKTINFO ? 1 : 0;
1704 if (put_user(len, optlen) ||
1705 put_user(val, optval))
1706 return -EFAULT;
1707 err = 0;
1708 break;
1709 case NETLINK_BROADCAST_ERROR:
1710 if (len < sizeof(int))
1711 return -EINVAL;
1712 len = sizeof(int);
1713 val = nlk->flags & NETLINK_F_BROADCAST_SEND_ERROR ? 1 : 0;
1714 if (put_user(len, optlen) ||
1715 put_user(val, optval))
1716 return -EFAULT;
1717 err = 0;
1718 break;
1719 case NETLINK_NO_ENOBUFS:
1720 if (len < sizeof(int))
1721 return -EINVAL;
1722 len = sizeof(int);
1723 val = nlk->flags & NETLINK_F_RECV_NO_ENOBUFS ? 1 : 0;
1724 if (put_user(len, optlen) ||
1725 put_user(val, optval))
1726 return -EFAULT;
1727 err = 0;
1728 break;
1729 case NETLINK_LIST_MEMBERSHIPS: {
1730 int pos, idx, shift;
1731
1732 err = 0;
1733 netlink_lock_table();
1734 for (pos = 0; pos * 8 < nlk->ngroups; pos += sizeof(u32)) {
1735 if (len - pos < sizeof(u32))
1736 break;
1737
1738 idx = pos / sizeof(unsigned long);
1739 shift = (pos % sizeof(unsigned long)) * 8;
1740 if (put_user((u32)(nlk->groups[idx] >> shift),
1741 (u32 __user *)(optval + pos))) {
1742 err = -EFAULT;
1743 break;
1744 }
1745 }
1746 if (put_user(ALIGN(nlk->ngroups / 8, sizeof(u32)), optlen))
1747 err = -EFAULT;
1748 netlink_unlock_table();
1749 break;
1750 }
1751 case NETLINK_CAP_ACK:
1752 if (len < sizeof(int))
1753 return -EINVAL;
1754 len = sizeof(int);
1755 val = nlk->flags & NETLINK_F_CAP_ACK ? 1 : 0;
1756 if (put_user(len, optlen) ||
1757 put_user(val, optval))
1758 return -EFAULT;
1759 err = 0;
1760 break;
1761 case NETLINK_EXT_ACK:
1762 if (len < sizeof(int))
1763 return -EINVAL;
1764 len = sizeof(int);
1765 val = nlk->flags & NETLINK_F_EXT_ACK ? 1 : 0;
1766 if (put_user(len, optlen) || put_user(val, optval))
1767 return -EFAULT;
1768 err = 0;
1769 break;
1770 default:
1771 err = -ENOPROTOOPT;
1772 }
1773 return err;
1774 }
1775
1776 static void netlink_cmsg_recv_pktinfo(struct msghdr *msg, struct sk_buff *skb)
1777 {
1778 struct nl_pktinfo info;
1779
1780 info.group = NETLINK_CB(skb).dst_group;
1781 put_cmsg(msg, SOL_NETLINK, NETLINK_PKTINFO, sizeof(info), &info);
1782 }
1783
1784 static void netlink_cmsg_listen_all_nsid(struct sock *sk, struct msghdr *msg,
1785 struct sk_buff *skb)
1786 {
1787 if (!NETLINK_CB(skb).nsid_is_set)
1788 return;
1789
1790 put_cmsg(msg, SOL_NETLINK, NETLINK_LISTEN_ALL_NSID, sizeof(int),
1791 &NETLINK_CB(skb).nsid);
1792 }
1793
1794 static int netlink_sendmsg(struct socket *sock, struct msghdr *msg, size_t len)
1795 {
1796 struct sock *sk = sock->sk;
1797 struct netlink_sock *nlk = nlk_sk(sk);
1798 DECLARE_SOCKADDR(struct sockaddr_nl *, addr, msg->msg_name);
1799 u32 dst_portid;
1800 u32 dst_group;
1801 struct sk_buff *skb;
1802 int err;
1803 struct scm_cookie scm;
1804 u32 netlink_skb_flags = 0;
1805
1806 if (msg->msg_flags&MSG_OOB)
1807 return -EOPNOTSUPP;
1808
1809 err = scm_send(sock, msg, &scm, true);
1810 if (err < 0)
1811 return err;
1812
1813 if (msg->msg_namelen) {
1814 err = -EINVAL;
1815 if (msg->msg_namelen < sizeof(struct sockaddr_nl))
1816 goto out;
1817 if (addr->nl_family != AF_NETLINK)
1818 goto out;
1819 dst_portid = addr->nl_pid;
1820 dst_group = ffs(addr->nl_groups);
1821 err = -EPERM;
1822 if ((dst_group || dst_portid) &&
1823 !netlink_allowed(sock, NL_CFG_F_NONROOT_SEND))
1824 goto out;
1825 netlink_skb_flags |= NETLINK_SKB_DST;
1826 } else {
1827 dst_portid = nlk->dst_portid;
1828 dst_group = nlk->dst_group;
1829 }
1830
1831 if (!nlk->bound) {
1832 err = netlink_autobind(sock);
1833 if (err)
1834 goto out;
1835 } else {
1836 /* Ensure nlk is hashed and visible. */
1837 smp_rmb();
1838 }
1839
1840 err = -EMSGSIZE;
1841 if (len > sk->sk_sndbuf - 32)
1842 goto out;
1843 err = -ENOBUFS;
1844 skb = netlink_alloc_large_skb(len, dst_group);
1845 if (skb == NULL)
1846 goto out;
1847
1848 NETLINK_CB(skb).portid = nlk->portid;
1849 NETLINK_CB(skb).dst_group = dst_group;
1850 NETLINK_CB(skb).creds = scm.creds;
1851 NETLINK_CB(skb).flags = netlink_skb_flags;
1852
1853 err = -EFAULT;
1854 if (memcpy_from_msg(skb_put(skb, len), msg, len)) {
1855 kfree_skb(skb);
1856 goto out;
1857 }
1858
1859 err = security_netlink_send(sk, skb);
1860 if (err) {
1861 kfree_skb(skb);
1862 goto out;
1863 }
1864
1865 if (dst_group) {
1866 refcount_inc(&skb->users);
1867 netlink_broadcast(sk, skb, dst_portid, dst_group, GFP_KERNEL);
1868 }
1869 err = netlink_unicast(sk, skb, dst_portid, msg->msg_flags&MSG_DONTWAIT);
1870
1871 out:
1872 scm_destroy(&scm);
1873 return err;
1874 }
1875
1876 static int netlink_recvmsg(struct socket *sock, struct msghdr *msg, size_t len,
1877 int flags)
1878 {
1879 struct scm_cookie scm;
1880 struct sock *sk = sock->sk;
1881 struct netlink_sock *nlk = nlk_sk(sk);
1882 int noblock = flags&MSG_DONTWAIT;
1883 size_t copied;
1884 struct sk_buff *skb, *data_skb;
1885 int err, ret;
1886
1887 if (flags&MSG_OOB)
1888 return -EOPNOTSUPP;
1889
1890 copied = 0;
1891
1892 skb = skb_recv_datagram(sk, flags, noblock, &err);
1893 if (skb == NULL)
1894 goto out;
1895
1896 data_skb = skb;
1897
1898 #ifdef CONFIG_COMPAT_NETLINK_MESSAGES
1899 if (unlikely(skb_shinfo(skb)->frag_list)) {
1900 /*
1901 * If this skb has a frag_list, then here that means that we
1902 * will have to use the frag_list skb's data for compat tasks
1903 * and the regular skb's data for normal (non-compat) tasks.
1904 *
1905 * If we need to send the compat skb, assign it to the
1906 * 'data_skb' variable so that it will be used below for data
1907 * copying. We keep 'skb' for everything else, including
1908 * freeing both later.
1909 */
1910 if (flags & MSG_CMSG_COMPAT)
1911 data_skb = skb_shinfo(skb)->frag_list;
1912 }
1913 #endif
1914
1915 /* Record the max length of recvmsg() calls for future allocations */
1916 nlk->max_recvmsg_len = max(nlk->max_recvmsg_len, len);
1917 nlk->max_recvmsg_len = min_t(size_t, nlk->max_recvmsg_len,
1918 SKB_WITH_OVERHEAD(32768));
1919
1920 copied = data_skb->len;
1921 if (len < copied) {
1922 msg->msg_flags |= MSG_TRUNC;
1923 copied = len;
1924 }
1925
1926 skb_reset_transport_header(data_skb);
1927 err = skb_copy_datagram_msg(data_skb, 0, msg, copied);
1928
1929 if (msg->msg_name) {
1930 DECLARE_SOCKADDR(struct sockaddr_nl *, addr, msg->msg_name);
1931 addr->nl_family = AF_NETLINK;
1932 addr->nl_pad = 0;
1933 addr->nl_pid = NETLINK_CB(skb).portid;
1934 addr->nl_groups = netlink_group_mask(NETLINK_CB(skb).dst_group);
1935 msg->msg_namelen = sizeof(*addr);
1936 }
1937
1938 if (nlk->flags & NETLINK_F_RECV_PKTINFO)
1939 netlink_cmsg_recv_pktinfo(msg, skb);
1940 if (nlk->flags & NETLINK_F_LISTEN_ALL_NSID)
1941 netlink_cmsg_listen_all_nsid(sk, msg, skb);
1942
1943 memset(&scm, 0, sizeof(scm));
1944 scm.creds = *NETLINK_CREDS(skb);
1945 if (flags & MSG_TRUNC)
1946 copied = data_skb->len;
1947
1948 skb_free_datagram(sk, skb);
1949
1950 if (nlk->cb_running &&
1951 atomic_read(&sk->sk_rmem_alloc) <= sk->sk_rcvbuf / 2) {
1952 ret = netlink_dump(sk);
1953 if (ret) {
1954 sk->sk_err = -ret;
1955 sk->sk_error_report(sk);
1956 }
1957 }
1958
1959 scm_recv(sock, msg, &scm, flags);
1960 out:
1961 netlink_rcv_wake(sk);
1962 return err ? : copied;
1963 }
1964
1965 static void netlink_data_ready(struct sock *sk)
1966 {
1967 BUG();
1968 }
1969
1970 /*
1971 * We export these functions to other modules. They provide a
1972 * complete set of kernel non-blocking support for message
1973 * queueing.
1974 */
1975
1976 struct sock *
1977 __netlink_kernel_create(struct net *net, int unit, struct module *module,
1978 struct netlink_kernel_cfg *cfg)
1979 {
1980 struct socket *sock;
1981 struct sock *sk;
1982 struct netlink_sock *nlk;
1983 struct listeners *listeners = NULL;
1984 struct mutex *cb_mutex = cfg ? cfg->cb_mutex : NULL;
1985 unsigned int groups;
1986
1987 BUG_ON(!nl_table);
1988
1989 if (unit < 0 || unit >= MAX_LINKS)
1990 return NULL;
1991
1992 if (sock_create_lite(PF_NETLINK, SOCK_DGRAM, unit, &sock))
1993 return NULL;
1994
1995 if (__netlink_create(net, sock, cb_mutex, unit, 1) < 0)
1996 goto out_sock_release_nosk;
1997
1998 sk = sock->sk;
1999
2000 if (!cfg || cfg->groups < 32)
2001 groups = 32;
2002 else
2003 groups = cfg->groups;
2004
2005 listeners = kzalloc(sizeof(*listeners) + NLGRPSZ(groups), GFP_KERNEL);
2006 if (!listeners)
2007 goto out_sock_release;
2008
2009 sk->sk_data_ready = netlink_data_ready;
2010 if (cfg && cfg->input)
2011 nlk_sk(sk)->netlink_rcv = cfg->input;
2012
2013 if (netlink_insert(sk, 0))
2014 goto out_sock_release;
2015
2016 nlk = nlk_sk(sk);
2017 nlk->flags |= NETLINK_F_KERNEL_SOCKET;
2018
2019 netlink_table_grab();
2020 if (!nl_table[unit].registered) {
2021 nl_table[unit].groups = groups;
2022 rcu_assign_pointer(nl_table[unit].listeners, listeners);
2023 nl_table[unit].cb_mutex = cb_mutex;
2024 nl_table[unit].module = module;
2025 if (cfg) {
2026 nl_table[unit].bind = cfg->bind;
2027 nl_table[unit].unbind = cfg->unbind;
2028 nl_table[unit].flags = cfg->flags;
2029 if (cfg->compare)
2030 nl_table[unit].compare = cfg->compare;
2031 }
2032 nl_table[unit].registered = 1;
2033 } else {
2034 kfree(listeners);
2035 nl_table[unit].registered++;
2036 }
2037 netlink_table_ungrab();
2038 return sk;
2039
2040 out_sock_release:
2041 kfree(listeners);
2042 netlink_kernel_release(sk);
2043 return NULL;
2044
2045 out_sock_release_nosk:
2046 sock_release(sock);
2047 return NULL;
2048 }
2049 EXPORT_SYMBOL(__netlink_kernel_create);
2050
2051 void
2052 netlink_kernel_release(struct sock *sk)
2053 {
2054 if (sk == NULL || sk->sk_socket == NULL)
2055 return;
2056
2057 sock_release(sk->sk_socket);
2058 }
2059 EXPORT_SYMBOL(netlink_kernel_release);
2060
2061 int __netlink_change_ngroups(struct sock *sk, unsigned int groups)
2062 {
2063 struct listeners *new, *old;
2064 struct netlink_table *tbl = &nl_table[sk->sk_protocol];
2065
2066 if (groups < 32)
2067 groups = 32;
2068
2069 if (NLGRPSZ(tbl->groups) < NLGRPSZ(groups)) {
2070 new = kzalloc(sizeof(*new) + NLGRPSZ(groups), GFP_ATOMIC);
2071 if (!new)
2072 return -ENOMEM;
2073 old = nl_deref_protected(tbl->listeners);
2074 memcpy(new->masks, old->masks, NLGRPSZ(tbl->groups));
2075 rcu_assign_pointer(tbl->listeners, new);
2076
2077 kfree_rcu(old, rcu);
2078 }
2079 tbl->groups = groups;
2080
2081 return 0;
2082 }
2083
2084 /**
2085 * netlink_change_ngroups - change number of multicast groups
2086 *
2087 * This changes the number of multicast groups that are available
2088 * on a certain netlink family. Note that it is not possible to
2089 * change the number of groups to below 32. Also note that it does
2090 * not implicitly call netlink_clear_multicast_users() when the
2091 * number of groups is reduced.
2092 *
2093 * @sk: The kernel netlink socket, as returned by netlink_kernel_create().
2094 * @groups: The new number of groups.
2095 */
2096 int netlink_change_ngroups(struct sock *sk, unsigned int groups)
2097 {
2098 int err;
2099
2100 netlink_table_grab();
2101 err = __netlink_change_ngroups(sk, groups);
2102 netlink_table_ungrab();
2103
2104 return err;
2105 }
2106
2107 void __netlink_clear_multicast_users(struct sock *ksk, unsigned int group)
2108 {
2109 struct sock *sk;
2110 struct netlink_table *tbl = &nl_table[ksk->sk_protocol];
2111
2112 sk_for_each_bound(sk, &tbl->mc_list)
2113 netlink_update_socket_mc(nlk_sk(sk), group, 0);
2114 }
2115
2116 struct nlmsghdr *
2117 __nlmsg_put(struct sk_buff *skb, u32 portid, u32 seq, int type, int len, int flags)
2118 {
2119 struct nlmsghdr *nlh;
2120 int size = nlmsg_msg_size(len);
2121
2122 nlh = skb_put(skb, NLMSG_ALIGN(size));
2123 nlh->nlmsg_type = type;
2124 nlh->nlmsg_len = size;
2125 nlh->nlmsg_flags = flags;
2126 nlh->nlmsg_pid = portid;
2127 nlh->nlmsg_seq = seq;
2128 if (!__builtin_constant_p(size) || NLMSG_ALIGN(size) - size != 0)
2129 memset(nlmsg_data(nlh) + len, 0, NLMSG_ALIGN(size) - size);
2130 return nlh;
2131 }
2132 EXPORT_SYMBOL(__nlmsg_put);
2133
2134 /*
2135 * It looks a bit ugly.
2136 * It would be better to create kernel thread.
2137 */
2138
2139 static int netlink_dump(struct sock *sk)
2140 {
2141 struct netlink_sock *nlk = nlk_sk(sk);
2142 struct netlink_callback *cb;
2143 struct sk_buff *skb = NULL;
2144 struct nlmsghdr *nlh;
2145 struct module *module;
2146 int err = -ENOBUFS;
2147 int alloc_min_size;
2148 int alloc_size;
2149
2150 mutex_lock(nlk->cb_mutex);
2151 if (!nlk->cb_running) {
2152 err = -EINVAL;
2153 goto errout_skb;
2154 }
2155
2156 if (atomic_read(&sk->sk_rmem_alloc) >= sk->sk_rcvbuf)
2157 goto errout_skb;
2158
2159 /* NLMSG_GOODSIZE is small to avoid high order allocations being
2160 * required, but it makes sense to _attempt_ a 16K bytes allocation
2161 * to reduce number of system calls on dump operations, if user
2162 * ever provided a big enough buffer.
2163 */
2164 cb = &nlk->cb;
2165 alloc_min_size = max_t(int, cb->min_dump_alloc, NLMSG_GOODSIZE);
2166
2167 if (alloc_min_size < nlk->max_recvmsg_len) {
2168 alloc_size = nlk->max_recvmsg_len;
2169 skb = alloc_skb(alloc_size,
2170 (GFP_KERNEL & ~__GFP_DIRECT_RECLAIM) |
2171 __GFP_NOWARN | __GFP_NORETRY);
2172 }
2173 if (!skb) {
2174 alloc_size = alloc_min_size;
2175 skb = alloc_skb(alloc_size, GFP_KERNEL);
2176 }
2177 if (!skb)
2178 goto errout_skb;
2179
2180 /* Trim skb to allocated size. User is expected to provide buffer as
2181 * large as max(min_dump_alloc, 16KiB (mac_recvmsg_len capped at
2182 * netlink_recvmsg())). dump will pack as many smaller messages as
2183 * could fit within the allocated skb. skb is typically allocated
2184 * with larger space than required (could be as much as near 2x the
2185 * requested size with align to next power of 2 approach). Allowing
2186 * dump to use the excess space makes it difficult for a user to have a
2187 * reasonable static buffer based on the expected largest dump of a
2188 * single netdev. The outcome is MSG_TRUNC error.
2189 */
2190 skb_reserve(skb, skb_tailroom(skb) - alloc_size);
2191 netlink_skb_set_owner_r(skb, sk);
2192
2193 if (nlk->dump_done_errno > 0)
2194 nlk->dump_done_errno = cb->dump(skb, cb);
2195
2196 if (nlk->dump_done_errno > 0 ||
2197 skb_tailroom(skb) < nlmsg_total_size(sizeof(nlk->dump_done_errno))) {
2198 mutex_unlock(nlk->cb_mutex);
2199
2200 if (sk_filter(sk, skb))
2201 kfree_skb(skb);
2202 else
2203 __netlink_sendskb(sk, skb);
2204 return 0;
2205 }
2206
2207 nlh = nlmsg_put_answer(skb, cb, NLMSG_DONE,
2208 sizeof(nlk->dump_done_errno), NLM_F_MULTI);
2209 if (WARN_ON(!nlh))
2210 goto errout_skb;
2211
2212 nl_dump_check_consistent(cb, nlh);
2213
2214 memcpy(nlmsg_data(nlh), &nlk->dump_done_errno,
2215 sizeof(nlk->dump_done_errno));
2216
2217 if (sk_filter(sk, skb))
2218 kfree_skb(skb);
2219 else
2220 __netlink_sendskb(sk, skb);
2221
2222 if (cb->done)
2223 cb->done(cb);
2224
2225 nlk->cb_running = false;
2226 module = cb->module;
2227 skb = cb->skb;
2228 mutex_unlock(nlk->cb_mutex);
2229 module_put(module);
2230 consume_skb(skb);
2231 return 0;
2232
2233 errout_skb:
2234 mutex_unlock(nlk->cb_mutex);
2235 kfree_skb(skb);
2236 return err;
2237 }
2238
2239 int __netlink_dump_start(struct sock *ssk, struct sk_buff *skb,
2240 const struct nlmsghdr *nlh,
2241 struct netlink_dump_control *control)
2242 {
2243 struct netlink_callback *cb;
2244 struct sock *sk;
2245 struct netlink_sock *nlk;
2246 int ret;
2247
2248 refcount_inc(&skb->users);
2249
2250 sk = netlink_lookup(sock_net(ssk), ssk->sk_protocol, NETLINK_CB(skb).portid);
2251 if (sk == NULL) {
2252 ret = -ECONNREFUSED;
2253 goto error_free;
2254 }
2255
2256 nlk = nlk_sk(sk);
2257 mutex_lock(nlk->cb_mutex);
2258 /* A dump is in progress... */
2259 if (nlk->cb_running) {
2260 ret = -EBUSY;
2261 goto error_unlock;
2262 }
2263 /* add reference of module which cb->dump belongs to */
2264 if (!try_module_get(control->module)) {
2265 ret = -EPROTONOSUPPORT;
2266 goto error_unlock;
2267 }
2268
2269 cb = &nlk->cb;
2270 memset(cb, 0, sizeof(*cb));
2271 cb->start = control->start;
2272 cb->dump = control->dump;
2273 cb->done = control->done;
2274 cb->nlh = nlh;
2275 cb->data = control->data;
2276 cb->module = control->module;
2277 cb->min_dump_alloc = control->min_dump_alloc;
2278 cb->skb = skb;
2279
2280 if (cb->start) {
2281 ret = cb->start(cb);
2282 if (ret)
2283 goto error_put;
2284 }
2285
2286 nlk->cb_running = true;
2287 nlk->dump_done_errno = INT_MAX;
2288
2289 mutex_unlock(nlk->cb_mutex);
2290
2291 ret = netlink_dump(sk);
2292
2293 sock_put(sk);
2294
2295 if (ret)
2296 return ret;
2297
2298 /* We successfully started a dump, by returning -EINTR we
2299 * signal not to send ACK even if it was requested.
2300 */
2301 return -EINTR;
2302
2303 error_put:
2304 module_put(control->module);
2305 error_unlock:
2306 sock_put(sk);
2307 mutex_unlock(nlk->cb_mutex);
2308 error_free:
2309 kfree_skb(skb);
2310 return ret;
2311 }
2312 EXPORT_SYMBOL(__netlink_dump_start);
2313
2314 void netlink_ack(struct sk_buff *in_skb, struct nlmsghdr *nlh, int err,
2315 const struct netlink_ext_ack *extack)
2316 {
2317 struct sk_buff *skb;
2318 struct nlmsghdr *rep;
2319 struct nlmsgerr *errmsg;
2320 size_t payload = sizeof(*errmsg);
2321 size_t tlvlen = 0;
2322 struct netlink_sock *nlk = nlk_sk(NETLINK_CB(in_skb).sk);
2323 unsigned int flags = 0;
2324 bool nlk_has_extack = nlk->flags & NETLINK_F_EXT_ACK;
2325
2326 /* Error messages get the original request appened, unless the user
2327 * requests to cap the error message, and get extra error data if
2328 * requested.
2329 */
2330 if (nlk_has_extack && extack && extack->_msg)
2331 tlvlen += nla_total_size(strlen(extack->_msg) + 1);
2332
2333 if (err) {
2334 if (!(nlk->flags & NETLINK_F_CAP_ACK))
2335 payload += nlmsg_len(nlh);
2336 else
2337 flags |= NLM_F_CAPPED;
2338 if (nlk_has_extack && extack && extack->bad_attr)
2339 tlvlen += nla_total_size(sizeof(u32));
2340 } else {
2341 flags |= NLM_F_CAPPED;
2342
2343 if (nlk_has_extack && extack && extack->cookie_len)
2344 tlvlen += nla_total_size(extack->cookie_len);
2345 }
2346
2347 if (tlvlen)
2348 flags |= NLM_F_ACK_TLVS;
2349
2350 skb = nlmsg_new(payload + tlvlen, GFP_KERNEL);
2351 if (!skb) {
2352 NETLINK_CB(in_skb).sk->sk_err = ENOBUFS;
2353 NETLINK_CB(in_skb).sk->sk_error_report(NETLINK_CB(in_skb).sk);
2354 return;
2355 }
2356
2357 rep = __nlmsg_put(skb, NETLINK_CB(in_skb).portid, nlh->nlmsg_seq,
2358 NLMSG_ERROR, payload, flags);
2359 errmsg = nlmsg_data(rep);
2360 errmsg->error = err;
2361 memcpy(&errmsg->msg, nlh, payload > sizeof(*errmsg) ? nlh->nlmsg_len : sizeof(*nlh));
2362
2363 if (nlk_has_extack && extack) {
2364 if (extack->_msg) {
2365 WARN_ON(nla_put_string(skb, NLMSGERR_ATTR_MSG,
2366 extack->_msg));
2367 }
2368 if (err) {
2369 if (extack->bad_attr &&
2370 !WARN_ON((u8 *)extack->bad_attr < in_skb->data ||
2371 (u8 *)extack->bad_attr >= in_skb->data +
2372 in_skb->len))
2373 WARN_ON(nla_put_u32(skb, NLMSGERR_ATTR_OFFS,
2374 (u8 *)extack->bad_attr -
2375 in_skb->data));
2376 } else {
2377 if (extack->cookie_len)
2378 WARN_ON(nla_put(skb, NLMSGERR_ATTR_COOKIE,
2379 extack->cookie_len,
2380 extack->cookie));
2381 }
2382 }
2383
2384 nlmsg_end(skb, rep);
2385
2386 netlink_unicast(in_skb->sk, skb, NETLINK_CB(in_skb).portid, MSG_DONTWAIT);
2387 }
2388 EXPORT_SYMBOL(netlink_ack);
2389
2390 int netlink_rcv_skb(struct sk_buff *skb, int (*cb)(struct sk_buff *,
2391 struct nlmsghdr *,
2392 struct netlink_ext_ack *))
2393 {
2394 struct netlink_ext_ack extack;
2395 struct nlmsghdr *nlh;
2396 int err;
2397
2398 while (skb->len >= nlmsg_total_size(0)) {
2399 int msglen;
2400
2401 memset(&extack, 0, sizeof(extack));
2402 nlh = nlmsg_hdr(skb);
2403 err = 0;
2404
2405 if (nlh->nlmsg_len < NLMSG_HDRLEN || skb->len < nlh->nlmsg_len)
2406 return 0;
2407
2408 /* Only requests are handled by the kernel */
2409 if (!(nlh->nlmsg_flags & NLM_F_REQUEST))
2410 goto ack;
2411
2412 /* Skip control messages */
2413 if (nlh->nlmsg_type < NLMSG_MIN_TYPE)
2414 goto ack;
2415
2416 err = cb(skb, nlh, &extack);
2417 if (err == -EINTR)
2418 goto skip;
2419
2420 ack:
2421 if (nlh->nlmsg_flags & NLM_F_ACK || err)
2422 netlink_ack(skb, nlh, err, &extack);
2423
2424 skip:
2425 msglen = NLMSG_ALIGN(nlh->nlmsg_len);
2426 if (msglen > skb->len)
2427 msglen = skb->len;
2428 skb_pull(skb, msglen);
2429 }
2430
2431 return 0;
2432 }
2433 EXPORT_SYMBOL(netlink_rcv_skb);
2434
2435 /**
2436 * nlmsg_notify - send a notification netlink message
2437 * @sk: netlink socket to use
2438 * @skb: notification message
2439 * @portid: destination netlink portid for reports or 0
2440 * @group: destination multicast group or 0
2441 * @report: 1 to report back, 0 to disable
2442 * @flags: allocation flags
2443 */
2444 int nlmsg_notify(struct sock *sk, struct sk_buff *skb, u32 portid,
2445 unsigned int group, int report, gfp_t flags)
2446 {
2447 int err = 0;
2448
2449 if (group) {
2450 int exclude_portid = 0;
2451
2452 if (report) {
2453 refcount_inc(&skb->users);
2454 exclude_portid = portid;
2455 }
2456
2457 /* errors reported via destination sk->sk_err, but propagate
2458 * delivery errors if NETLINK_BROADCAST_ERROR flag is set */
2459 err = nlmsg_multicast(sk, skb, exclude_portid, group, flags);
2460 }
2461
2462 if (report) {
2463 int err2;
2464
2465 err2 = nlmsg_unicast(sk, skb, portid);
2466 if (!err || err == -ESRCH)
2467 err = err2;
2468 }
2469
2470 return err;
2471 }
2472 EXPORT_SYMBOL(nlmsg_notify);
2473
2474 #ifdef CONFIG_PROC_FS
2475 struct nl_seq_iter {
2476 struct seq_net_private p;
2477 struct rhashtable_iter hti;
2478 int link;
2479 };
2480
2481 static int netlink_walk_start(struct nl_seq_iter *iter)
2482 {
2483 int err;
2484
2485 err = rhashtable_walk_init(&nl_table[iter->link].hash, &iter->hti,
2486 GFP_KERNEL);
2487 if (err) {
2488 iter->link = MAX_LINKS;
2489 return err;
2490 }
2491
2492 err = rhashtable_walk_start(&iter->hti);
2493 return err == -EAGAIN ? 0 : err;
2494 }
2495
2496 static void netlink_walk_stop(struct nl_seq_iter *iter)
2497 {
2498 rhashtable_walk_stop(&iter->hti);
2499 rhashtable_walk_exit(&iter->hti);
2500 }
2501
2502 static void *__netlink_seq_next(struct seq_file *seq)
2503 {
2504 struct nl_seq_iter *iter = seq->private;
2505 struct netlink_sock *nlk;
2506
2507 do {
2508 for (;;) {
2509 int err;
2510
2511 nlk = rhashtable_walk_next(&iter->hti);
2512
2513 if (IS_ERR(nlk)) {
2514 if (PTR_ERR(nlk) == -EAGAIN)
2515 continue;
2516
2517 return nlk;
2518 }
2519
2520 if (nlk)
2521 break;
2522
2523 netlink_walk_stop(iter);
2524 if (++iter->link >= MAX_LINKS)
2525 return NULL;
2526
2527 err = netlink_walk_start(iter);
2528 if (err)
2529 return ERR_PTR(err);
2530 }
2531 } while (sock_net(&nlk->sk) != seq_file_net(seq));
2532
2533 return nlk;
2534 }
2535
2536 static void *netlink_seq_start(struct seq_file *seq, loff_t *posp)
2537 {
2538 struct nl_seq_iter *iter = seq->private;
2539 void *obj = SEQ_START_TOKEN;
2540 loff_t pos;
2541 int err;
2542
2543 iter->link = 0;
2544
2545 err = netlink_walk_start(iter);
2546 if (err)
2547 return ERR_PTR(err);
2548
2549 for (pos = *posp; pos && obj && !IS_ERR(obj); pos--)
2550 obj = __netlink_seq_next(seq);
2551
2552 return obj;
2553 }
2554
2555 static void *netlink_seq_next(struct seq_file *seq, void *v, loff_t *pos)
2556 {
2557 ++*pos;
2558 return __netlink_seq_next(seq);
2559 }
2560
2561 static void netlink_seq_stop(struct seq_file *seq, void *v)
2562 {
2563 struct nl_seq_iter *iter = seq->private;
2564
2565 if (iter->link >= MAX_LINKS)
2566 return;
2567
2568 netlink_walk_stop(iter);
2569 }
2570
2571
2572 static int netlink_seq_show(struct seq_file *seq, void *v)
2573 {
2574 if (v == SEQ_START_TOKEN) {
2575 seq_puts(seq,
2576 "sk Eth Pid Groups "
2577 "Rmem Wmem Dump Locks Drops Inode\n");
2578 } else {
2579 struct sock *s = v;
2580 struct netlink_sock *nlk = nlk_sk(s);
2581
2582 seq_printf(seq, "%pK %-3d %-6u %08x %-8d %-8d %d %-8d %-8d %-8lu\n",
2583 s,
2584 s->sk_protocol,
2585 nlk->portid,
2586 nlk->groups ? (u32)nlk->groups[0] : 0,
2587 sk_rmem_alloc_get(s),
2588 sk_wmem_alloc_get(s),
2589 nlk->cb_running,
2590 refcount_read(&s->sk_refcnt),
2591 atomic_read(&s->sk_drops),
2592 sock_i_ino(s)
2593 );
2594
2595 }
2596 return 0;
2597 }
2598
2599 static const struct seq_operations netlink_seq_ops = {
2600 .start = netlink_seq_start,
2601 .next = netlink_seq_next,
2602 .stop = netlink_seq_stop,
2603 .show = netlink_seq_show,
2604 };
2605
2606
2607 static int netlink_seq_open(struct inode *inode, struct file *file)
2608 {
2609 return seq_open_net(inode, file, &netlink_seq_ops,
2610 sizeof(struct nl_seq_iter));
2611 }
2612
2613 static const struct file_operations netlink_seq_fops = {
2614 .owner = THIS_MODULE,
2615 .open = netlink_seq_open,
2616 .read = seq_read,
2617 .llseek = seq_lseek,
2618 .release = seq_release_net,
2619 };
2620
2621 #endif
2622
2623 int netlink_register_notifier(struct notifier_block *nb)
2624 {
2625 return blocking_notifier_chain_register(&netlink_chain, nb);
2626 }
2627 EXPORT_SYMBOL(netlink_register_notifier);
2628
2629 int netlink_unregister_notifier(struct notifier_block *nb)
2630 {
2631 return blocking_notifier_chain_unregister(&netlink_chain, nb);
2632 }
2633 EXPORT_SYMBOL(netlink_unregister_notifier);
2634
2635 static const struct proto_ops netlink_ops = {
2636 .family = PF_NETLINK,
2637 .owner = THIS_MODULE,
2638 .release = netlink_release,
2639 .bind = netlink_bind,
2640 .connect = netlink_connect,
2641 .socketpair = sock_no_socketpair,
2642 .accept = sock_no_accept,
2643 .getname = netlink_getname,
2644 .poll = datagram_poll,
2645 .ioctl = netlink_ioctl,
2646 .listen = sock_no_listen,
2647 .shutdown = sock_no_shutdown,
2648 .setsockopt = netlink_setsockopt,
2649 .getsockopt = netlink_getsockopt,
2650 .sendmsg = netlink_sendmsg,
2651 .recvmsg = netlink_recvmsg,
2652 .mmap = sock_no_mmap,
2653 .sendpage = sock_no_sendpage,
2654 };
2655
2656 static const struct net_proto_family netlink_family_ops = {
2657 .family = PF_NETLINK,
2658 .create = netlink_create,
2659 .owner = THIS_MODULE, /* for consistency 8) */
2660 };
2661
2662 static int __net_init netlink_net_init(struct net *net)
2663 {
2664 #ifdef CONFIG_PROC_FS
2665 if (!proc_create("netlink", 0, net->proc_net, &netlink_seq_fops))
2666 return -ENOMEM;
2667 #endif
2668 return 0;
2669 }
2670
2671 static void __net_exit netlink_net_exit(struct net *net)
2672 {
2673 #ifdef CONFIG_PROC_FS
2674 remove_proc_entry("netlink", net->proc_net);
2675 #endif
2676 }
2677
2678 static void __init netlink_add_usersock_entry(void)
2679 {
2680 struct listeners *listeners;
2681 int groups = 32;
2682
2683 listeners = kzalloc(sizeof(*listeners) + NLGRPSZ(groups), GFP_KERNEL);
2684 if (!listeners)
2685 panic("netlink_add_usersock_entry: Cannot allocate listeners\n");
2686
2687 netlink_table_grab();
2688
2689 nl_table[NETLINK_USERSOCK].groups = groups;
2690 rcu_assign_pointer(nl_table[NETLINK_USERSOCK].listeners, listeners);
2691 nl_table[NETLINK_USERSOCK].module = THIS_MODULE;
2692 nl_table[NETLINK_USERSOCK].registered = 1;
2693 nl_table[NETLINK_USERSOCK].flags = NL_CFG_F_NONROOT_SEND;
2694
2695 netlink_table_ungrab();
2696 }
2697
2698 static struct pernet_operations __net_initdata netlink_net_ops = {
2699 .init = netlink_net_init,
2700 .exit = netlink_net_exit,
2701 };
2702
2703 static inline u32 netlink_hash(const void *data, u32 len, u32 seed)
2704 {
2705 const struct netlink_sock *nlk = data;
2706 struct netlink_compare_arg arg;
2707
2708 netlink_compare_arg_init(&arg, sock_net(&nlk->sk), nlk->portid);
2709 return jhash2((u32 *)&arg, netlink_compare_arg_len / sizeof(u32), seed);
2710 }
2711
2712 static const struct rhashtable_params netlink_rhashtable_params = {
2713 .head_offset = offsetof(struct netlink_sock, node),
2714 .key_len = netlink_compare_arg_len,
2715 .obj_hashfn = netlink_hash,
2716 .obj_cmpfn = netlink_compare,
2717 .automatic_shrinking = true,
2718 };
2719
2720 static int __init netlink_proto_init(void)
2721 {
2722 int i;
2723 int err = proto_register(&netlink_proto, 0);
2724
2725 if (err != 0)
2726 goto out;
2727
2728 BUILD_BUG_ON(sizeof(struct netlink_skb_parms) > FIELD_SIZEOF(struct sk_buff, cb));
2729
2730 nl_table = kcalloc(MAX_LINKS, sizeof(*nl_table), GFP_KERNEL);
2731 if (!nl_table)
2732 goto panic;
2733
2734 for (i = 0; i < MAX_LINKS; i++) {
2735 if (rhashtable_init(&nl_table[i].hash,
2736 &netlink_rhashtable_params) < 0) {
2737 while (--i > 0)
2738 rhashtable_destroy(&nl_table[i].hash);
2739 kfree(nl_table);
2740 goto panic;
2741 }
2742 }
2743
2744 INIT_LIST_HEAD(&netlink_tap_all);
2745
2746 netlink_add_usersock_entry();
2747
2748 sock_register(&netlink_family_ops);
2749 register_pernet_subsys(&netlink_net_ops);
2750 /* The netlink device handler may be needed early. */
2751 rtnetlink_init();
2752 out:
2753 return err;
2754 panic:
2755 panic("netlink_init: Cannot allocate nl_table\n");
2756 }
2757
2758 core_initcall(netlink_proto_init);
ubuntu-bionic-kernel mirror