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