]> git.proxmox.com Git - mirror_ubuntu-eoan-kernel.git/blob - net/netlink/af_netlink.c
Merge branch 'cpuidle' of git://git.kernel.org/pub/scm/linux/kernel/git/lenb/linux...
[mirror_ubuntu-eoan-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 <asm/uaccess.h>
44 #include <linux/skbuff.h>
45 #include <linux/netdevice.h>
46 #include <linux/rtnetlink.h>
47 #include <linux/proc_fs.h>
48 #include <linux/seq_file.h>
49 #include <linux/notifier.h>
50 #include <linux/security.h>
51 #include <linux/jhash.h>
52 #include <linux/jiffies.h>
53 #include <linux/random.h>
54 #include <linux/bitops.h>
55 #include <linux/mm.h>
56 #include <linux/types.h>
57 #include <linux/audit.h>
58 #include <linux/mutex.h>
59 #include <linux/vmalloc.h>
60 #include <linux/if_arp.h>
61 #include <linux/rhashtable.h>
62 #include <asm/cacheflush.h>
63 #include <linux/hash.h>
64 #include <linux/genetlink.h>
65
66 #include <net/net_namespace.h>
67 #include <net/sock.h>
68 #include <net/scm.h>
69 #include <net/netlink.h>
70
71 #include "af_netlink.h"
72
73 struct listeners {
74 struct rcu_head rcu;
75 unsigned long masks[0];
76 };
77
78 /* state bits */
79 #define NETLINK_S_CONGESTED 0x0
80
81 /* flags */
82 #define NETLINK_F_KERNEL_SOCKET 0x1
83 #define NETLINK_F_RECV_PKTINFO 0x2
84 #define NETLINK_F_BROADCAST_SEND_ERROR 0x4
85 #define NETLINK_F_RECV_NO_ENOBUFS 0x8
86 #define NETLINK_F_LISTEN_ALL_NSID 0x10
87 #define NETLINK_F_CAP_ACK 0x20
88
89 static inline int netlink_is_kernel(struct sock *sk)
90 {
91 return nlk_sk(sk)->flags & NETLINK_F_KERNEL_SOCKET;
92 }
93
94 struct netlink_table *nl_table __read_mostly;
95 EXPORT_SYMBOL_GPL(nl_table);
96
97 static DECLARE_WAIT_QUEUE_HEAD(nl_table_wait);
98
99 static int netlink_dump(struct sock *sk);
100 static void netlink_skb_destructor(struct sk_buff *skb);
101
102 /* nl_table locking explained:
103 * Lookup and traversal are protected with an RCU read-side lock. Insertion
104 * and removal are protected with per bucket lock while using RCU list
105 * modification primitives and may run in parallel to RCU protected lookups.
106 * Destruction of the Netlink socket may only occur *after* nl_table_lock has
107 * been acquired * either during or after the socket has been removed from
108 * the list and after an RCU grace period.
109 */
110 DEFINE_RWLOCK(nl_table_lock);
111 EXPORT_SYMBOL_GPL(nl_table_lock);
112 static atomic_t nl_table_users = ATOMIC_INIT(0);
113
114 #define nl_deref_protected(X) rcu_dereference_protected(X, lockdep_is_held(&nl_table_lock));
115
116 static ATOMIC_NOTIFIER_HEAD(netlink_chain);
117
118 static DEFINE_SPINLOCK(netlink_tap_lock);
119 static struct list_head netlink_tap_all __read_mostly;
120
121 static const struct rhashtable_params netlink_rhashtable_params;
122
123 static inline u32 netlink_group_mask(u32 group)
124 {
125 return group ? 1 << (group - 1) : 0;
126 }
127
128 static struct sk_buff *netlink_to_full_skb(const struct sk_buff *skb,
129 gfp_t gfp_mask)
130 {
131 unsigned int len = skb_end_offset(skb);
132 struct sk_buff *new;
133
134 new = alloc_skb(len, gfp_mask);
135 if (new == NULL)
136 return NULL;
137
138 NETLINK_CB(new).portid = NETLINK_CB(skb).portid;
139 NETLINK_CB(new).dst_group = NETLINK_CB(skb).dst_group;
140 NETLINK_CB(new).creds = NETLINK_CB(skb).creds;
141
142 memcpy(skb_put(new, len), skb->data, len);
143 return new;
144 }
145
146 int netlink_add_tap(struct netlink_tap *nt)
147 {
148 if (unlikely(nt->dev->type != ARPHRD_NETLINK))
149 return -EINVAL;
150
151 spin_lock(&netlink_tap_lock);
152 list_add_rcu(&nt->list, &netlink_tap_all);
153 spin_unlock(&netlink_tap_lock);
154
155 __module_get(nt->module);
156
157 return 0;
158 }
159 EXPORT_SYMBOL_GPL(netlink_add_tap);
160
161 static int __netlink_remove_tap(struct netlink_tap *nt)
162 {
163 bool found = false;
164 struct netlink_tap *tmp;
165
166 spin_lock(&netlink_tap_lock);
167
168 list_for_each_entry(tmp, &netlink_tap_all, list) {
169 if (nt == tmp) {
170 list_del_rcu(&nt->list);
171 found = true;
172 goto out;
173 }
174 }
175
176 pr_warn("__netlink_remove_tap: %p not found\n", nt);
177 out:
178 spin_unlock(&netlink_tap_lock);
179
180 if (found)
181 module_put(nt->module);
182
183 return found ? 0 : -ENODEV;
184 }
185
186 int netlink_remove_tap(struct netlink_tap *nt)
187 {
188 int ret;
189
190 ret = __netlink_remove_tap(nt);
191 synchronize_net();
192
193 return ret;
194 }
195 EXPORT_SYMBOL_GPL(netlink_remove_tap);
196
197 static bool netlink_filter_tap(const struct sk_buff *skb)
198 {
199 struct sock *sk = skb->sk;
200
201 /* We take the more conservative approach and
202 * whitelist socket protocols that may pass.
203 */
204 switch (sk->sk_protocol) {
205 case NETLINK_ROUTE:
206 case NETLINK_USERSOCK:
207 case NETLINK_SOCK_DIAG:
208 case NETLINK_NFLOG:
209 case NETLINK_XFRM:
210 case NETLINK_FIB_LOOKUP:
211 case NETLINK_NETFILTER:
212 case NETLINK_GENERIC:
213 return true;
214 }
215
216 return false;
217 }
218
219 static int __netlink_deliver_tap_skb(struct sk_buff *skb,
220 struct net_device *dev)
221 {
222 struct sk_buff *nskb;
223 struct sock *sk = skb->sk;
224 int ret = -ENOMEM;
225
226 dev_hold(dev);
227
228 if (netlink_skb_is_mmaped(skb) || is_vmalloc_addr(skb->head))
229 nskb = netlink_to_full_skb(skb, GFP_ATOMIC);
230 else
231 nskb = skb_clone(skb, GFP_ATOMIC);
232 if (nskb) {
233 nskb->dev = dev;
234 nskb->protocol = htons((u16) sk->sk_protocol);
235 nskb->pkt_type = netlink_is_kernel(sk) ?
236 PACKET_KERNEL : PACKET_USER;
237 skb_reset_network_header(nskb);
238 ret = dev_queue_xmit(nskb);
239 if (unlikely(ret > 0))
240 ret = net_xmit_errno(ret);
241 }
242
243 dev_put(dev);
244 return ret;
245 }
246
247 static void __netlink_deliver_tap(struct sk_buff *skb)
248 {
249 int ret;
250 struct netlink_tap *tmp;
251
252 if (!netlink_filter_tap(skb))
253 return;
254
255 list_for_each_entry_rcu(tmp, &netlink_tap_all, list) {
256 ret = __netlink_deliver_tap_skb(skb, tmp->dev);
257 if (unlikely(ret))
258 break;
259 }
260 }
261
262 static void netlink_deliver_tap(struct sk_buff *skb)
263 {
264 rcu_read_lock();
265
266 if (unlikely(!list_empty(&netlink_tap_all)))
267 __netlink_deliver_tap(skb);
268
269 rcu_read_unlock();
270 }
271
272 static void netlink_deliver_tap_kernel(struct sock *dst, struct sock *src,
273 struct sk_buff *skb)
274 {
275 if (!(netlink_is_kernel(dst) && netlink_is_kernel(src)))
276 netlink_deliver_tap(skb);
277 }
278
279 static void netlink_overrun(struct sock *sk)
280 {
281 struct netlink_sock *nlk = nlk_sk(sk);
282
283 if (!(nlk->flags & NETLINK_F_RECV_NO_ENOBUFS)) {
284 if (!test_and_set_bit(NETLINK_S_CONGESTED,
285 &nlk_sk(sk)->state)) {
286 sk->sk_err = ENOBUFS;
287 sk->sk_error_report(sk);
288 }
289 }
290 atomic_inc(&sk->sk_drops);
291 }
292
293 static void netlink_rcv_wake(struct sock *sk)
294 {
295 struct netlink_sock *nlk = nlk_sk(sk);
296
297 if (skb_queue_empty(&sk->sk_receive_queue))
298 clear_bit(NETLINK_S_CONGESTED, &nlk->state);
299 if (!test_bit(NETLINK_S_CONGESTED, &nlk->state))
300 wake_up_interruptible(&nlk->wait);
301 }
302
303 #ifdef CONFIG_NETLINK_MMAP
304 static bool netlink_rx_is_mmaped(struct sock *sk)
305 {
306 return nlk_sk(sk)->rx_ring.pg_vec != NULL;
307 }
308
309 static bool netlink_tx_is_mmaped(struct sock *sk)
310 {
311 return nlk_sk(sk)->tx_ring.pg_vec != NULL;
312 }
313
314 static __pure struct page *pgvec_to_page(const void *addr)
315 {
316 if (is_vmalloc_addr(addr))
317 return vmalloc_to_page(addr);
318 else
319 return virt_to_page(addr);
320 }
321
322 static void free_pg_vec(void **pg_vec, unsigned int order, unsigned int len)
323 {
324 unsigned int i;
325
326 for (i = 0; i < len; i++) {
327 if (pg_vec[i] != NULL) {
328 if (is_vmalloc_addr(pg_vec[i]))
329 vfree(pg_vec[i]);
330 else
331 free_pages((unsigned long)pg_vec[i], order);
332 }
333 }
334 kfree(pg_vec);
335 }
336
337 static void *alloc_one_pg_vec_page(unsigned long order)
338 {
339 void *buffer;
340 gfp_t gfp_flags = GFP_KERNEL | __GFP_COMP | __GFP_ZERO |
341 __GFP_NOWARN | __GFP_NORETRY;
342
343 buffer = (void *)__get_free_pages(gfp_flags, order);
344 if (buffer != NULL)
345 return buffer;
346
347 buffer = vzalloc((1 << order) * PAGE_SIZE);
348 if (buffer != NULL)
349 return buffer;
350
351 gfp_flags &= ~__GFP_NORETRY;
352 return (void *)__get_free_pages(gfp_flags, order);
353 }
354
355 static void **alloc_pg_vec(struct netlink_sock *nlk,
356 struct nl_mmap_req *req, unsigned int order)
357 {
358 unsigned int block_nr = req->nm_block_nr;
359 unsigned int i;
360 void **pg_vec;
361
362 pg_vec = kcalloc(block_nr, sizeof(void *), GFP_KERNEL);
363 if (pg_vec == NULL)
364 return NULL;
365
366 for (i = 0; i < block_nr; i++) {
367 pg_vec[i] = alloc_one_pg_vec_page(order);
368 if (pg_vec[i] == NULL)
369 goto err1;
370 }
371
372 return pg_vec;
373 err1:
374 free_pg_vec(pg_vec, order, block_nr);
375 return NULL;
376 }
377
378
379 static void
380 __netlink_set_ring(struct sock *sk, struct nl_mmap_req *req, bool tx_ring, void **pg_vec,
381 unsigned int order)
382 {
383 struct netlink_sock *nlk = nlk_sk(sk);
384 struct sk_buff_head *queue;
385 struct netlink_ring *ring;
386
387 queue = tx_ring ? &sk->sk_write_queue : &sk->sk_receive_queue;
388 ring = tx_ring ? &nlk->tx_ring : &nlk->rx_ring;
389
390 spin_lock_bh(&queue->lock);
391
392 ring->frame_max = req->nm_frame_nr - 1;
393 ring->head = 0;
394 ring->frame_size = req->nm_frame_size;
395 ring->pg_vec_pages = req->nm_block_size / PAGE_SIZE;
396
397 swap(ring->pg_vec_len, req->nm_block_nr);
398 swap(ring->pg_vec_order, order);
399 swap(ring->pg_vec, pg_vec);
400
401 __skb_queue_purge(queue);
402 spin_unlock_bh(&queue->lock);
403
404 WARN_ON(atomic_read(&nlk->mapped));
405
406 if (pg_vec)
407 free_pg_vec(pg_vec, order, req->nm_block_nr);
408 }
409
410 static int netlink_set_ring(struct sock *sk, struct nl_mmap_req *req,
411 bool tx_ring)
412 {
413 struct netlink_sock *nlk = nlk_sk(sk);
414 struct netlink_ring *ring;
415 void **pg_vec = NULL;
416 unsigned int order = 0;
417
418 ring = tx_ring ? &nlk->tx_ring : &nlk->rx_ring;
419
420 if (atomic_read(&nlk->mapped))
421 return -EBUSY;
422 if (atomic_read(&ring->pending))
423 return -EBUSY;
424
425 if (req->nm_block_nr) {
426 if (ring->pg_vec != NULL)
427 return -EBUSY;
428
429 if ((int)req->nm_block_size <= 0)
430 return -EINVAL;
431 if (!PAGE_ALIGNED(req->nm_block_size))
432 return -EINVAL;
433 if (req->nm_frame_size < NL_MMAP_HDRLEN)
434 return -EINVAL;
435 if (!IS_ALIGNED(req->nm_frame_size, NL_MMAP_MSG_ALIGNMENT))
436 return -EINVAL;
437
438 ring->frames_per_block = req->nm_block_size /
439 req->nm_frame_size;
440 if (ring->frames_per_block == 0)
441 return -EINVAL;
442 if (ring->frames_per_block * req->nm_block_nr !=
443 req->nm_frame_nr)
444 return -EINVAL;
445
446 order = get_order(req->nm_block_size);
447 pg_vec = alloc_pg_vec(nlk, req, order);
448 if (pg_vec == NULL)
449 return -ENOMEM;
450 } else {
451 if (req->nm_frame_nr)
452 return -EINVAL;
453 }
454
455 mutex_lock(&nlk->pg_vec_lock);
456 if (atomic_read(&nlk->mapped) == 0) {
457 __netlink_set_ring(sk, req, tx_ring, pg_vec, order);
458 mutex_unlock(&nlk->pg_vec_lock);
459 return 0;
460 }
461
462 mutex_unlock(&nlk->pg_vec_lock);
463
464 if (pg_vec)
465 free_pg_vec(pg_vec, order, req->nm_block_nr);
466
467 return -EBUSY;
468 }
469
470 static void netlink_mm_open(struct vm_area_struct *vma)
471 {
472 struct file *file = vma->vm_file;
473 struct socket *sock = file->private_data;
474 struct sock *sk = sock->sk;
475
476 if (sk)
477 atomic_inc(&nlk_sk(sk)->mapped);
478 }
479
480 static void netlink_mm_close(struct vm_area_struct *vma)
481 {
482 struct file *file = vma->vm_file;
483 struct socket *sock = file->private_data;
484 struct sock *sk = sock->sk;
485
486 if (sk)
487 atomic_dec(&nlk_sk(sk)->mapped);
488 }
489
490 static const struct vm_operations_struct netlink_mmap_ops = {
491 .open = netlink_mm_open,
492 .close = netlink_mm_close,
493 };
494
495 static int netlink_mmap(struct file *file, struct socket *sock,
496 struct vm_area_struct *vma)
497 {
498 struct sock *sk = sock->sk;
499 struct netlink_sock *nlk = nlk_sk(sk);
500 struct netlink_ring *ring;
501 unsigned long start, size, expected;
502 unsigned int i;
503 int err = -EINVAL;
504
505 if (vma->vm_pgoff)
506 return -EINVAL;
507
508 mutex_lock(&nlk->pg_vec_lock);
509
510 expected = 0;
511 for (ring = &nlk->rx_ring; ring <= &nlk->tx_ring; ring++) {
512 if (ring->pg_vec == NULL)
513 continue;
514 expected += ring->pg_vec_len * ring->pg_vec_pages * PAGE_SIZE;
515 }
516
517 if (expected == 0)
518 goto out;
519
520 size = vma->vm_end - vma->vm_start;
521 if (size != expected)
522 goto out;
523
524 start = vma->vm_start;
525 for (ring = &nlk->rx_ring; ring <= &nlk->tx_ring; ring++) {
526 if (ring->pg_vec == NULL)
527 continue;
528
529 for (i = 0; i < ring->pg_vec_len; i++) {
530 struct page *page;
531 void *kaddr = ring->pg_vec[i];
532 unsigned int pg_num;
533
534 for (pg_num = 0; pg_num < ring->pg_vec_pages; pg_num++) {
535 page = pgvec_to_page(kaddr);
536 err = vm_insert_page(vma, start, page);
537 if (err < 0)
538 goto out;
539 start += PAGE_SIZE;
540 kaddr += PAGE_SIZE;
541 }
542 }
543 }
544
545 atomic_inc(&nlk->mapped);
546 vma->vm_ops = &netlink_mmap_ops;
547 err = 0;
548 out:
549 mutex_unlock(&nlk->pg_vec_lock);
550 return err;
551 }
552
553 static void netlink_frame_flush_dcache(const struct nl_mmap_hdr *hdr, unsigned int nm_len)
554 {
555 #if ARCH_IMPLEMENTS_FLUSH_DCACHE_PAGE == 1
556 struct page *p_start, *p_end;
557
558 /* First page is flushed through netlink_{get,set}_status */
559 p_start = pgvec_to_page(hdr + PAGE_SIZE);
560 p_end = pgvec_to_page((void *)hdr + NL_MMAP_HDRLEN + nm_len - 1);
561 while (p_start <= p_end) {
562 flush_dcache_page(p_start);
563 p_start++;
564 }
565 #endif
566 }
567
568 static enum nl_mmap_status netlink_get_status(const struct nl_mmap_hdr *hdr)
569 {
570 smp_rmb();
571 flush_dcache_page(pgvec_to_page(hdr));
572 return hdr->nm_status;
573 }
574
575 static void netlink_set_status(struct nl_mmap_hdr *hdr,
576 enum nl_mmap_status status)
577 {
578 smp_mb();
579 hdr->nm_status = status;
580 flush_dcache_page(pgvec_to_page(hdr));
581 }
582
583 static struct nl_mmap_hdr *
584 __netlink_lookup_frame(const struct netlink_ring *ring, unsigned int pos)
585 {
586 unsigned int pg_vec_pos, frame_off;
587
588 pg_vec_pos = pos / ring->frames_per_block;
589 frame_off = pos % ring->frames_per_block;
590
591 return ring->pg_vec[pg_vec_pos] + (frame_off * ring->frame_size);
592 }
593
594 static struct nl_mmap_hdr *
595 netlink_lookup_frame(const struct netlink_ring *ring, unsigned int pos,
596 enum nl_mmap_status status)
597 {
598 struct nl_mmap_hdr *hdr;
599
600 hdr = __netlink_lookup_frame(ring, pos);
601 if (netlink_get_status(hdr) != status)
602 return NULL;
603
604 return hdr;
605 }
606
607 static struct nl_mmap_hdr *
608 netlink_current_frame(const struct netlink_ring *ring,
609 enum nl_mmap_status status)
610 {
611 return netlink_lookup_frame(ring, ring->head, status);
612 }
613
614 static void netlink_increment_head(struct netlink_ring *ring)
615 {
616 ring->head = ring->head != ring->frame_max ? ring->head + 1 : 0;
617 }
618
619 static void netlink_forward_ring(struct netlink_ring *ring)
620 {
621 unsigned int head = ring->head;
622 const struct nl_mmap_hdr *hdr;
623
624 do {
625 hdr = __netlink_lookup_frame(ring, ring->head);
626 if (hdr->nm_status == NL_MMAP_STATUS_UNUSED)
627 break;
628 if (hdr->nm_status != NL_MMAP_STATUS_SKIP)
629 break;
630 netlink_increment_head(ring);
631 } while (ring->head != head);
632 }
633
634 static bool netlink_has_valid_frame(struct netlink_ring *ring)
635 {
636 unsigned int head = ring->head, pos = head;
637 const struct nl_mmap_hdr *hdr;
638
639 do {
640 hdr = __netlink_lookup_frame(ring, pos);
641 if (hdr->nm_status == NL_MMAP_STATUS_VALID)
642 return true;
643 pos = pos != 0 ? pos - 1 : ring->frame_max;
644 } while (pos != head);
645
646 return false;
647 }
648
649 static bool netlink_dump_space(struct netlink_sock *nlk)
650 {
651 struct netlink_ring *ring = &nlk->rx_ring;
652 struct nl_mmap_hdr *hdr;
653 unsigned int n;
654
655 hdr = netlink_current_frame(ring, NL_MMAP_STATUS_UNUSED);
656 if (hdr == NULL)
657 return false;
658
659 n = ring->head + ring->frame_max / 2;
660 if (n > ring->frame_max)
661 n -= ring->frame_max;
662
663 hdr = __netlink_lookup_frame(ring, n);
664
665 return hdr->nm_status == NL_MMAP_STATUS_UNUSED;
666 }
667
668 static unsigned int netlink_poll(struct file *file, struct socket *sock,
669 poll_table *wait)
670 {
671 struct sock *sk = sock->sk;
672 struct netlink_sock *nlk = nlk_sk(sk);
673 unsigned int mask;
674 int err;
675
676 if (nlk->rx_ring.pg_vec != NULL) {
677 /* Memory mapped sockets don't call recvmsg(), so flow control
678 * for dumps is performed here. A dump is allowed to continue
679 * if at least half the ring is unused.
680 */
681 while (nlk->cb_running && netlink_dump_space(nlk)) {
682 err = netlink_dump(sk);
683 if (err < 0) {
684 sk->sk_err = -err;
685 sk->sk_error_report(sk);
686 break;
687 }
688 }
689 netlink_rcv_wake(sk);
690 }
691
692 mask = datagram_poll(file, sock, wait);
693
694 /* We could already have received frames in the normal receive
695 * queue, that will show up as NL_MMAP_STATUS_COPY in the ring,
696 * so if mask contains pollin/etc already, there's no point
697 * walking the ring.
698 */
699 if ((mask & (POLLIN | POLLRDNORM)) != (POLLIN | POLLRDNORM)) {
700 spin_lock_bh(&sk->sk_receive_queue.lock);
701 if (nlk->rx_ring.pg_vec) {
702 if (netlink_has_valid_frame(&nlk->rx_ring))
703 mask |= POLLIN | POLLRDNORM;
704 }
705 spin_unlock_bh(&sk->sk_receive_queue.lock);
706 }
707
708 spin_lock_bh(&sk->sk_write_queue.lock);
709 if (nlk->tx_ring.pg_vec) {
710 if (netlink_current_frame(&nlk->tx_ring, NL_MMAP_STATUS_UNUSED))
711 mask |= POLLOUT | POLLWRNORM;
712 }
713 spin_unlock_bh(&sk->sk_write_queue.lock);
714
715 return mask;
716 }
717
718 static struct nl_mmap_hdr *netlink_mmap_hdr(struct sk_buff *skb)
719 {
720 return (struct nl_mmap_hdr *)(skb->head - NL_MMAP_HDRLEN);
721 }
722
723 static void netlink_ring_setup_skb(struct sk_buff *skb, struct sock *sk,
724 struct netlink_ring *ring,
725 struct nl_mmap_hdr *hdr)
726 {
727 unsigned int size;
728 void *data;
729
730 size = ring->frame_size - NL_MMAP_HDRLEN;
731 data = (void *)hdr + NL_MMAP_HDRLEN;
732
733 skb->head = data;
734 skb->data = data;
735 skb_reset_tail_pointer(skb);
736 skb->end = skb->tail + size;
737 skb->len = 0;
738
739 skb->destructor = netlink_skb_destructor;
740 NETLINK_CB(skb).flags |= NETLINK_SKB_MMAPED;
741 NETLINK_CB(skb).sk = sk;
742 }
743
744 static int netlink_mmap_sendmsg(struct sock *sk, struct msghdr *msg,
745 u32 dst_portid, u32 dst_group,
746 struct scm_cookie *scm)
747 {
748 struct netlink_sock *nlk = nlk_sk(sk);
749 struct netlink_ring *ring;
750 struct nl_mmap_hdr *hdr;
751 struct sk_buff *skb;
752 unsigned int maxlen;
753 int err = 0, len = 0;
754
755 mutex_lock(&nlk->pg_vec_lock);
756
757 ring = &nlk->tx_ring;
758 maxlen = ring->frame_size - NL_MMAP_HDRLEN;
759
760 do {
761 unsigned int nm_len;
762
763 hdr = netlink_current_frame(ring, NL_MMAP_STATUS_VALID);
764 if (hdr == NULL) {
765 if (!(msg->msg_flags & MSG_DONTWAIT) &&
766 atomic_read(&nlk->tx_ring.pending))
767 schedule();
768 continue;
769 }
770
771 nm_len = ACCESS_ONCE(hdr->nm_len);
772 if (nm_len > maxlen) {
773 err = -EINVAL;
774 goto out;
775 }
776
777 netlink_frame_flush_dcache(hdr, nm_len);
778
779 skb = alloc_skb(nm_len, GFP_KERNEL);
780 if (skb == NULL) {
781 err = -ENOBUFS;
782 goto out;
783 }
784 __skb_put(skb, nm_len);
785 memcpy(skb->data, (void *)hdr + NL_MMAP_HDRLEN, nm_len);
786 netlink_set_status(hdr, NL_MMAP_STATUS_UNUSED);
787
788 netlink_increment_head(ring);
789
790 NETLINK_CB(skb).portid = nlk->portid;
791 NETLINK_CB(skb).dst_group = dst_group;
792 NETLINK_CB(skb).creds = scm->creds;
793
794 err = security_netlink_send(sk, skb);
795 if (err) {
796 kfree_skb(skb);
797 goto out;
798 }
799
800 if (unlikely(dst_group)) {
801 atomic_inc(&skb->users);
802 netlink_broadcast(sk, skb, dst_portid, dst_group,
803 GFP_KERNEL);
804 }
805 err = netlink_unicast(sk, skb, dst_portid,
806 msg->msg_flags & MSG_DONTWAIT);
807 if (err < 0)
808 goto out;
809 len += err;
810
811 } while (hdr != NULL ||
812 (!(msg->msg_flags & MSG_DONTWAIT) &&
813 atomic_read(&nlk->tx_ring.pending)));
814
815 if (len > 0)
816 err = len;
817 out:
818 mutex_unlock(&nlk->pg_vec_lock);
819 return err;
820 }
821
822 static void netlink_queue_mmaped_skb(struct sock *sk, struct sk_buff *skb)
823 {
824 struct nl_mmap_hdr *hdr;
825
826 hdr = netlink_mmap_hdr(skb);
827 hdr->nm_len = skb->len;
828 hdr->nm_group = NETLINK_CB(skb).dst_group;
829 hdr->nm_pid = NETLINK_CB(skb).creds.pid;
830 hdr->nm_uid = from_kuid(sk_user_ns(sk), NETLINK_CB(skb).creds.uid);
831 hdr->nm_gid = from_kgid(sk_user_ns(sk), NETLINK_CB(skb).creds.gid);
832 netlink_frame_flush_dcache(hdr, hdr->nm_len);
833 netlink_set_status(hdr, NL_MMAP_STATUS_VALID);
834
835 NETLINK_CB(skb).flags |= NETLINK_SKB_DELIVERED;
836 kfree_skb(skb);
837 }
838
839 static void netlink_ring_set_copied(struct sock *sk, struct sk_buff *skb)
840 {
841 struct netlink_sock *nlk = nlk_sk(sk);
842 struct netlink_ring *ring = &nlk->rx_ring;
843 struct nl_mmap_hdr *hdr;
844
845 spin_lock_bh(&sk->sk_receive_queue.lock);
846 hdr = netlink_current_frame(ring, NL_MMAP_STATUS_UNUSED);
847 if (hdr == NULL) {
848 spin_unlock_bh(&sk->sk_receive_queue.lock);
849 kfree_skb(skb);
850 netlink_overrun(sk);
851 return;
852 }
853 netlink_increment_head(ring);
854 __skb_queue_tail(&sk->sk_receive_queue, skb);
855 spin_unlock_bh(&sk->sk_receive_queue.lock);
856
857 hdr->nm_len = skb->len;
858 hdr->nm_group = NETLINK_CB(skb).dst_group;
859 hdr->nm_pid = NETLINK_CB(skb).creds.pid;
860 hdr->nm_uid = from_kuid(sk_user_ns(sk), NETLINK_CB(skb).creds.uid);
861 hdr->nm_gid = from_kgid(sk_user_ns(sk), NETLINK_CB(skb).creds.gid);
862 netlink_set_status(hdr, NL_MMAP_STATUS_COPY);
863 }
864
865 #else /* CONFIG_NETLINK_MMAP */
866 #define netlink_rx_is_mmaped(sk) false
867 #define netlink_tx_is_mmaped(sk) false
868 #define netlink_mmap sock_no_mmap
869 #define netlink_poll datagram_poll
870 #define netlink_mmap_sendmsg(sk, msg, dst_portid, dst_group, scm) 0
871 #endif /* CONFIG_NETLINK_MMAP */
872
873 static void netlink_skb_destructor(struct sk_buff *skb)
874 {
875 #ifdef CONFIG_NETLINK_MMAP
876 struct nl_mmap_hdr *hdr;
877 struct netlink_ring *ring;
878 struct sock *sk;
879
880 /* If a packet from the kernel to userspace was freed because of an
881 * error without being delivered to userspace, the kernel must reset
882 * the status. In the direction userspace to kernel, the status is
883 * always reset here after the packet was processed and freed.
884 */
885 if (netlink_skb_is_mmaped(skb)) {
886 hdr = netlink_mmap_hdr(skb);
887 sk = NETLINK_CB(skb).sk;
888
889 if (NETLINK_CB(skb).flags & NETLINK_SKB_TX) {
890 netlink_set_status(hdr, NL_MMAP_STATUS_UNUSED);
891 ring = &nlk_sk(sk)->tx_ring;
892 } else {
893 if (!(NETLINK_CB(skb).flags & NETLINK_SKB_DELIVERED)) {
894 hdr->nm_len = 0;
895 netlink_set_status(hdr, NL_MMAP_STATUS_VALID);
896 }
897 ring = &nlk_sk(sk)->rx_ring;
898 }
899
900 WARN_ON(atomic_read(&ring->pending) == 0);
901 atomic_dec(&ring->pending);
902 sock_put(sk);
903
904 skb->head = NULL;
905 }
906 #endif
907 if (is_vmalloc_addr(skb->head)) {
908 if (!skb->cloned ||
909 !atomic_dec_return(&(skb_shinfo(skb)->dataref)))
910 vfree(skb->head);
911
912 skb->head = NULL;
913 }
914 if (skb->sk != NULL)
915 sock_rfree(skb);
916 }
917
918 static void netlink_skb_set_owner_r(struct sk_buff *skb, struct sock *sk)
919 {
920 WARN_ON(skb->sk != NULL);
921 skb->sk = sk;
922 skb->destructor = netlink_skb_destructor;
923 atomic_add(skb->truesize, &sk->sk_rmem_alloc);
924 sk_mem_charge(sk, skb->truesize);
925 }
926
927 static void netlink_sock_destruct(struct sock *sk)
928 {
929 struct netlink_sock *nlk = nlk_sk(sk);
930
931 if (nlk->cb_running) {
932 if (nlk->cb.done)
933 nlk->cb.done(&nlk->cb);
934
935 module_put(nlk->cb.module);
936 kfree_skb(nlk->cb.skb);
937 }
938
939 skb_queue_purge(&sk->sk_receive_queue);
940 #ifdef CONFIG_NETLINK_MMAP
941 if (1) {
942 struct nl_mmap_req req;
943
944 memset(&req, 0, sizeof(req));
945 if (nlk->rx_ring.pg_vec)
946 __netlink_set_ring(sk, &req, false, NULL, 0);
947 memset(&req, 0, sizeof(req));
948 if (nlk->tx_ring.pg_vec)
949 __netlink_set_ring(sk, &req, true, NULL, 0);
950 }
951 #endif /* CONFIG_NETLINK_MMAP */
952
953 if (!sock_flag(sk, SOCK_DEAD)) {
954 printk(KERN_ERR "Freeing alive netlink socket %p\n", sk);
955 return;
956 }
957
958 WARN_ON(atomic_read(&sk->sk_rmem_alloc));
959 WARN_ON(atomic_read(&sk->sk_wmem_alloc));
960 WARN_ON(nlk_sk(sk)->groups);
961 }
962
963 /* This lock without WQ_FLAG_EXCLUSIVE is good on UP and it is _very_ bad on
964 * SMP. Look, when several writers sleep and reader wakes them up, all but one
965 * immediately hit write lock and grab all the cpus. Exclusive sleep solves
966 * this, _but_ remember, it adds useless work on UP machines.
967 */
968
969 void netlink_table_grab(void)
970 __acquires(nl_table_lock)
971 {
972 might_sleep();
973
974 write_lock_irq(&nl_table_lock);
975
976 if (atomic_read(&nl_table_users)) {
977 DECLARE_WAITQUEUE(wait, current);
978
979 add_wait_queue_exclusive(&nl_table_wait, &wait);
980 for (;;) {
981 set_current_state(TASK_UNINTERRUPTIBLE);
982 if (atomic_read(&nl_table_users) == 0)
983 break;
984 write_unlock_irq(&nl_table_lock);
985 schedule();
986 write_lock_irq(&nl_table_lock);
987 }
988
989 __set_current_state(TASK_RUNNING);
990 remove_wait_queue(&nl_table_wait, &wait);
991 }
992 }
993
994 void netlink_table_ungrab(void)
995 __releases(nl_table_lock)
996 {
997 write_unlock_irq(&nl_table_lock);
998 wake_up(&nl_table_wait);
999 }
1000
1001 static inline void
1002 netlink_lock_table(void)
1003 {
1004 /* read_lock() synchronizes us to netlink_table_grab */
1005
1006 read_lock(&nl_table_lock);
1007 atomic_inc(&nl_table_users);
1008 read_unlock(&nl_table_lock);
1009 }
1010
1011 static inline void
1012 netlink_unlock_table(void)
1013 {
1014 if (atomic_dec_and_test(&nl_table_users))
1015 wake_up(&nl_table_wait);
1016 }
1017
1018 struct netlink_compare_arg
1019 {
1020 possible_net_t pnet;
1021 u32 portid;
1022 };
1023
1024 /* Doing sizeof directly may yield 4 extra bytes on 64-bit. */
1025 #define netlink_compare_arg_len \
1026 (offsetof(struct netlink_compare_arg, portid) + sizeof(u32))
1027
1028 static inline int netlink_compare(struct rhashtable_compare_arg *arg,
1029 const void *ptr)
1030 {
1031 const struct netlink_compare_arg *x = arg->key;
1032 const struct netlink_sock *nlk = ptr;
1033
1034 return nlk->portid != x->portid ||
1035 !net_eq(sock_net(&nlk->sk), read_pnet(&x->pnet));
1036 }
1037
1038 static void netlink_compare_arg_init(struct netlink_compare_arg *arg,
1039 struct net *net, u32 portid)
1040 {
1041 memset(arg, 0, sizeof(*arg));
1042 write_pnet(&arg->pnet, net);
1043 arg->portid = portid;
1044 }
1045
1046 static struct sock *__netlink_lookup(struct netlink_table *table, u32 portid,
1047 struct net *net)
1048 {
1049 struct netlink_compare_arg arg;
1050
1051 netlink_compare_arg_init(&arg, net, portid);
1052 return rhashtable_lookup_fast(&table->hash, &arg,
1053 netlink_rhashtable_params);
1054 }
1055
1056 static int __netlink_insert(struct netlink_table *table, struct sock *sk)
1057 {
1058 struct netlink_compare_arg arg;
1059
1060 netlink_compare_arg_init(&arg, sock_net(sk), nlk_sk(sk)->portid);
1061 return rhashtable_lookup_insert_key(&table->hash, &arg,
1062 &nlk_sk(sk)->node,
1063 netlink_rhashtable_params);
1064 }
1065
1066 static struct sock *netlink_lookup(struct net *net, int protocol, u32 portid)
1067 {
1068 struct netlink_table *table = &nl_table[protocol];
1069 struct sock *sk;
1070
1071 rcu_read_lock();
1072 sk = __netlink_lookup(table, portid, net);
1073 if (sk)
1074 sock_hold(sk);
1075 rcu_read_unlock();
1076
1077 return sk;
1078 }
1079
1080 static const struct proto_ops netlink_ops;
1081
1082 static void
1083 netlink_update_listeners(struct sock *sk)
1084 {
1085 struct netlink_table *tbl = &nl_table[sk->sk_protocol];
1086 unsigned long mask;
1087 unsigned int i;
1088 struct listeners *listeners;
1089
1090 listeners = nl_deref_protected(tbl->listeners);
1091 if (!listeners)
1092 return;
1093
1094 for (i = 0; i < NLGRPLONGS(tbl->groups); i++) {
1095 mask = 0;
1096 sk_for_each_bound(sk, &tbl->mc_list) {
1097 if (i < NLGRPLONGS(nlk_sk(sk)->ngroups))
1098 mask |= nlk_sk(sk)->groups[i];
1099 }
1100 listeners->masks[i] = mask;
1101 }
1102 /* this function is only called with the netlink table "grabbed", which
1103 * makes sure updates are visible before bind or setsockopt return. */
1104 }
1105
1106 static int netlink_insert(struct sock *sk, u32 portid)
1107 {
1108 struct netlink_table *table = &nl_table[sk->sk_protocol];
1109 int err;
1110
1111 lock_sock(sk);
1112
1113 err = nlk_sk(sk)->portid == portid ? 0 : -EBUSY;
1114 if (nlk_sk(sk)->bound)
1115 goto err;
1116
1117 err = -ENOMEM;
1118 if (BITS_PER_LONG > 32 &&
1119 unlikely(atomic_read(&table->hash.nelems) >= UINT_MAX))
1120 goto err;
1121
1122 nlk_sk(sk)->portid = portid;
1123 sock_hold(sk);
1124
1125 err = __netlink_insert(table, sk);
1126 if (err) {
1127 /* In case the hashtable backend returns with -EBUSY
1128 * from here, it must not escape to the caller.
1129 */
1130 if (unlikely(err == -EBUSY))
1131 err = -EOVERFLOW;
1132 if (err == -EEXIST)
1133 err = -EADDRINUSE;
1134 sock_put(sk);
1135 goto err;
1136 }
1137
1138 /* We need to ensure that the socket is hashed and visible. */
1139 smp_wmb();
1140 nlk_sk(sk)->bound = portid;
1141
1142 err:
1143 release_sock(sk);
1144 return err;
1145 }
1146
1147 static void netlink_remove(struct sock *sk)
1148 {
1149 struct netlink_table *table;
1150
1151 table = &nl_table[sk->sk_protocol];
1152 if (!rhashtable_remove_fast(&table->hash, &nlk_sk(sk)->node,
1153 netlink_rhashtable_params)) {
1154 WARN_ON(atomic_read(&sk->sk_refcnt) == 1);
1155 __sock_put(sk);
1156 }
1157
1158 netlink_table_grab();
1159 if (nlk_sk(sk)->subscriptions) {
1160 __sk_del_bind_node(sk);
1161 netlink_update_listeners(sk);
1162 }
1163 if (sk->sk_protocol == NETLINK_GENERIC)
1164 atomic_inc(&genl_sk_destructing_cnt);
1165 netlink_table_ungrab();
1166 }
1167
1168 static struct proto netlink_proto = {
1169 .name = "NETLINK",
1170 .owner = THIS_MODULE,
1171 .obj_size = sizeof(struct netlink_sock),
1172 };
1173
1174 static int __netlink_create(struct net *net, struct socket *sock,
1175 struct mutex *cb_mutex, int protocol,
1176 int kern)
1177 {
1178 struct sock *sk;
1179 struct netlink_sock *nlk;
1180
1181 sock->ops = &netlink_ops;
1182
1183 sk = sk_alloc(net, PF_NETLINK, GFP_KERNEL, &netlink_proto, kern);
1184 if (!sk)
1185 return -ENOMEM;
1186
1187 sock_init_data(sock, sk);
1188
1189 nlk = nlk_sk(sk);
1190 if (cb_mutex) {
1191 nlk->cb_mutex = cb_mutex;
1192 } else {
1193 nlk->cb_mutex = &nlk->cb_def_mutex;
1194 mutex_init(nlk->cb_mutex);
1195 }
1196 init_waitqueue_head(&nlk->wait);
1197 #ifdef CONFIG_NETLINK_MMAP
1198 mutex_init(&nlk->pg_vec_lock);
1199 #endif
1200
1201 sk->sk_destruct = netlink_sock_destruct;
1202 sk->sk_protocol = protocol;
1203 return 0;
1204 }
1205
1206 static int netlink_create(struct net *net, struct socket *sock, int protocol,
1207 int kern)
1208 {
1209 struct module *module = NULL;
1210 struct mutex *cb_mutex;
1211 struct netlink_sock *nlk;
1212 int (*bind)(struct net *net, int group);
1213 void (*unbind)(struct net *net, int group);
1214 int err = 0;
1215
1216 sock->state = SS_UNCONNECTED;
1217
1218 if (sock->type != SOCK_RAW && sock->type != SOCK_DGRAM)
1219 return -ESOCKTNOSUPPORT;
1220
1221 if (protocol < 0 || protocol >= MAX_LINKS)
1222 return -EPROTONOSUPPORT;
1223
1224 netlink_lock_table();
1225 #ifdef CONFIG_MODULES
1226 if (!nl_table[protocol].registered) {
1227 netlink_unlock_table();
1228 request_module("net-pf-%d-proto-%d", PF_NETLINK, protocol);
1229 netlink_lock_table();
1230 }
1231 #endif
1232 if (nl_table[protocol].registered &&
1233 try_module_get(nl_table[protocol].module))
1234 module = nl_table[protocol].module;
1235 else
1236 err = -EPROTONOSUPPORT;
1237 cb_mutex = nl_table[protocol].cb_mutex;
1238 bind = nl_table[protocol].bind;
1239 unbind = nl_table[protocol].unbind;
1240 netlink_unlock_table();
1241
1242 if (err < 0)
1243 goto out;
1244
1245 err = __netlink_create(net, sock, cb_mutex, protocol, kern);
1246 if (err < 0)
1247 goto out_module;
1248
1249 local_bh_disable();
1250 sock_prot_inuse_add(net, &netlink_proto, 1);
1251 local_bh_enable();
1252
1253 nlk = nlk_sk(sock->sk);
1254 nlk->module = module;
1255 nlk->netlink_bind = bind;
1256 nlk->netlink_unbind = unbind;
1257 out:
1258 return err;
1259
1260 out_module:
1261 module_put(module);
1262 goto out;
1263 }
1264
1265 static void deferred_put_nlk_sk(struct rcu_head *head)
1266 {
1267 struct netlink_sock *nlk = container_of(head, struct netlink_sock, rcu);
1268
1269 sock_put(&nlk->sk);
1270 }
1271
1272 static int netlink_release(struct socket *sock)
1273 {
1274 struct sock *sk = sock->sk;
1275 struct netlink_sock *nlk;
1276
1277 if (!sk)
1278 return 0;
1279
1280 netlink_remove(sk);
1281 sock_orphan(sk);
1282 nlk = nlk_sk(sk);
1283
1284 /*
1285 * OK. Socket is unlinked, any packets that arrive now
1286 * will be purged.
1287 */
1288
1289 /* must not acquire netlink_table_lock in any way again before unbind
1290 * and notifying genetlink is done as otherwise it might deadlock
1291 */
1292 if (nlk->netlink_unbind) {
1293 int i;
1294
1295 for (i = 0; i < nlk->ngroups; i++)
1296 if (test_bit(i, nlk->groups))
1297 nlk->netlink_unbind(sock_net(sk), i + 1);
1298 }
1299 if (sk->sk_protocol == NETLINK_GENERIC &&
1300 atomic_dec_return(&genl_sk_destructing_cnt) == 0)
1301 wake_up(&genl_sk_destructing_waitq);
1302
1303 sock->sk = NULL;
1304 wake_up_interruptible_all(&nlk->wait);
1305
1306 skb_queue_purge(&sk->sk_write_queue);
1307
1308 if (nlk->portid) {
1309 struct netlink_notify n = {
1310 .net = sock_net(sk),
1311 .protocol = sk->sk_protocol,
1312 .portid = nlk->portid,
1313 };
1314 atomic_notifier_call_chain(&netlink_chain,
1315 NETLINK_URELEASE, &n);
1316 }
1317
1318 module_put(nlk->module);
1319
1320 if (netlink_is_kernel(sk)) {
1321 netlink_table_grab();
1322 BUG_ON(nl_table[sk->sk_protocol].registered == 0);
1323 if (--nl_table[sk->sk_protocol].registered == 0) {
1324 struct listeners *old;
1325
1326 old = nl_deref_protected(nl_table[sk->sk_protocol].listeners);
1327 RCU_INIT_POINTER(nl_table[sk->sk_protocol].listeners, NULL);
1328 kfree_rcu(old, rcu);
1329 nl_table[sk->sk_protocol].module = NULL;
1330 nl_table[sk->sk_protocol].bind = NULL;
1331 nl_table[sk->sk_protocol].unbind = NULL;
1332 nl_table[sk->sk_protocol].flags = 0;
1333 nl_table[sk->sk_protocol].registered = 0;
1334 }
1335 netlink_table_ungrab();
1336 }
1337
1338 kfree(nlk->groups);
1339 nlk->groups = NULL;
1340
1341 local_bh_disable();
1342 sock_prot_inuse_add(sock_net(sk), &netlink_proto, -1);
1343 local_bh_enable();
1344 call_rcu(&nlk->rcu, deferred_put_nlk_sk);
1345 return 0;
1346 }
1347
1348 static int netlink_autobind(struct socket *sock)
1349 {
1350 struct sock *sk = sock->sk;
1351 struct net *net = sock_net(sk);
1352 struct netlink_table *table = &nl_table[sk->sk_protocol];
1353 s32 portid = task_tgid_vnr(current);
1354 int err;
1355 s32 rover = -4096;
1356 bool ok;
1357
1358 retry:
1359 cond_resched();
1360 rcu_read_lock();
1361 ok = !__netlink_lookup(table, portid, net);
1362 rcu_read_unlock();
1363 if (!ok) {
1364 /* Bind collision, search negative portid values. */
1365 if (rover == -4096)
1366 /* rover will be in range [S32_MIN, -4097] */
1367 rover = S32_MIN + prandom_u32_max(-4096 - S32_MIN);
1368 else if (rover >= -4096)
1369 rover = -4097;
1370 portid = rover--;
1371 goto retry;
1372 }
1373
1374 err = netlink_insert(sk, portid);
1375 if (err == -EADDRINUSE)
1376 goto retry;
1377
1378 /* If 2 threads race to autobind, that is fine. */
1379 if (err == -EBUSY)
1380 err = 0;
1381
1382 return err;
1383 }
1384
1385 /**
1386 * __netlink_ns_capable - General netlink message capability test
1387 * @nsp: NETLINK_CB of the socket buffer holding a netlink command from userspace.
1388 * @user_ns: The user namespace of the capability to use
1389 * @cap: The capability to use
1390 *
1391 * Test to see if the opener of the socket we received the message
1392 * from had when the netlink socket was created and the sender of the
1393 * message has has the capability @cap in the user namespace @user_ns.
1394 */
1395 bool __netlink_ns_capable(const struct netlink_skb_parms *nsp,
1396 struct user_namespace *user_ns, int cap)
1397 {
1398 return ((nsp->flags & NETLINK_SKB_DST) ||
1399 file_ns_capable(nsp->sk->sk_socket->file, user_ns, cap)) &&
1400 ns_capable(user_ns, cap);
1401 }
1402 EXPORT_SYMBOL(__netlink_ns_capable);
1403
1404 /**
1405 * netlink_ns_capable - General netlink message capability test
1406 * @skb: socket buffer holding a netlink command from userspace
1407 * @user_ns: The user namespace of the capability to use
1408 * @cap: The capability to use
1409 *
1410 * Test to see if the opener of the socket we received the message
1411 * from had when the netlink socket was created and the sender of the
1412 * message has has the capability @cap in the user namespace @user_ns.
1413 */
1414 bool netlink_ns_capable(const struct sk_buff *skb,
1415 struct user_namespace *user_ns, int cap)
1416 {
1417 return __netlink_ns_capable(&NETLINK_CB(skb), user_ns, cap);
1418 }
1419 EXPORT_SYMBOL(netlink_ns_capable);
1420
1421 /**
1422 * netlink_capable - Netlink global message capability test
1423 * @skb: socket buffer holding a netlink command from userspace
1424 * @cap: The capability to use
1425 *
1426 * Test to see if the opener of the socket we received the message
1427 * from had when the netlink socket was created and the sender of the
1428 * message has has the capability @cap in all user namespaces.
1429 */
1430 bool netlink_capable(const struct sk_buff *skb, int cap)
1431 {
1432 return netlink_ns_capable(skb, &init_user_ns, cap);
1433 }
1434 EXPORT_SYMBOL(netlink_capable);
1435
1436 /**
1437 * netlink_net_capable - Netlink network namespace message capability test
1438 * @skb: socket buffer holding a netlink command from userspace
1439 * @cap: The capability to use
1440 *
1441 * Test to see if the opener of the socket we received the message
1442 * from had when the netlink socket was created and the sender of the
1443 * message has has the capability @cap over the network namespace of
1444 * the socket we received the message from.
1445 */
1446 bool netlink_net_capable(const struct sk_buff *skb, int cap)
1447 {
1448 return netlink_ns_capable(skb, sock_net(skb->sk)->user_ns, cap);
1449 }
1450 EXPORT_SYMBOL(netlink_net_capable);
1451
1452 static inline int netlink_allowed(const struct socket *sock, unsigned int flag)
1453 {
1454 return (nl_table[sock->sk->sk_protocol].flags & flag) ||
1455 ns_capable(sock_net(sock->sk)->user_ns, CAP_NET_ADMIN);
1456 }
1457
1458 static void
1459 netlink_update_subscriptions(struct sock *sk, unsigned int subscriptions)
1460 {
1461 struct netlink_sock *nlk = nlk_sk(sk);
1462
1463 if (nlk->subscriptions && !subscriptions)
1464 __sk_del_bind_node(sk);
1465 else if (!nlk->subscriptions && subscriptions)
1466 sk_add_bind_node(sk, &nl_table[sk->sk_protocol].mc_list);
1467 nlk->subscriptions = subscriptions;
1468 }
1469
1470 static int netlink_realloc_groups(struct sock *sk)
1471 {
1472 struct netlink_sock *nlk = nlk_sk(sk);
1473 unsigned int groups;
1474 unsigned long *new_groups;
1475 int err = 0;
1476
1477 netlink_table_grab();
1478
1479 groups = nl_table[sk->sk_protocol].groups;
1480 if (!nl_table[sk->sk_protocol].registered) {
1481 err = -ENOENT;
1482 goto out_unlock;
1483 }
1484
1485 if (nlk->ngroups >= groups)
1486 goto out_unlock;
1487
1488 new_groups = krealloc(nlk->groups, NLGRPSZ(groups), GFP_ATOMIC);
1489 if (new_groups == NULL) {
1490 err = -ENOMEM;
1491 goto out_unlock;
1492 }
1493 memset((char *)new_groups + NLGRPSZ(nlk->ngroups), 0,
1494 NLGRPSZ(groups) - NLGRPSZ(nlk->ngroups));
1495
1496 nlk->groups = new_groups;
1497 nlk->ngroups = groups;
1498 out_unlock:
1499 netlink_table_ungrab();
1500 return err;
1501 }
1502
1503 static void netlink_undo_bind(int group, long unsigned int groups,
1504 struct sock *sk)
1505 {
1506 struct netlink_sock *nlk = nlk_sk(sk);
1507 int undo;
1508
1509 if (!nlk->netlink_unbind)
1510 return;
1511
1512 for (undo = 0; undo < group; undo++)
1513 if (test_bit(undo, &groups))
1514 nlk->netlink_unbind(sock_net(sk), undo + 1);
1515 }
1516
1517 static int netlink_bind(struct socket *sock, struct sockaddr *addr,
1518 int addr_len)
1519 {
1520 struct sock *sk = sock->sk;
1521 struct net *net = sock_net(sk);
1522 struct netlink_sock *nlk = nlk_sk(sk);
1523 struct sockaddr_nl *nladdr = (struct sockaddr_nl *)addr;
1524 int err;
1525 long unsigned int groups = nladdr->nl_groups;
1526 bool bound;
1527
1528 if (addr_len < sizeof(struct sockaddr_nl))
1529 return -EINVAL;
1530
1531 if (nladdr->nl_family != AF_NETLINK)
1532 return -EINVAL;
1533
1534 /* Only superuser is allowed to listen multicasts */
1535 if (groups) {
1536 if (!netlink_allowed(sock, NL_CFG_F_NONROOT_RECV))
1537 return -EPERM;
1538 err = netlink_realloc_groups(sk);
1539 if (err)
1540 return err;
1541 }
1542
1543 bound = nlk->bound;
1544 if (bound) {
1545 /* Ensure nlk->portid is up-to-date. */
1546 smp_rmb();
1547
1548 if (nladdr->nl_pid != nlk->portid)
1549 return -EINVAL;
1550 }
1551
1552 if (nlk->netlink_bind && groups) {
1553 int group;
1554
1555 for (group = 0; group < nlk->ngroups; group++) {
1556 if (!test_bit(group, &groups))
1557 continue;
1558 err = nlk->netlink_bind(net, group + 1);
1559 if (!err)
1560 continue;
1561 netlink_undo_bind(group, groups, sk);
1562 return err;
1563 }
1564 }
1565
1566 /* No need for barriers here as we return to user-space without
1567 * using any of the bound attributes.
1568 */
1569 if (!bound) {
1570 err = nladdr->nl_pid ?
1571 netlink_insert(sk, nladdr->nl_pid) :
1572 netlink_autobind(sock);
1573 if (err) {
1574 netlink_undo_bind(nlk->ngroups, groups, sk);
1575 return err;
1576 }
1577 }
1578
1579 if (!groups && (nlk->groups == NULL || !(u32)nlk->groups[0]))
1580 return 0;
1581
1582 netlink_table_grab();
1583 netlink_update_subscriptions(sk, nlk->subscriptions +
1584 hweight32(groups) -
1585 hweight32(nlk->groups[0]));
1586 nlk->groups[0] = (nlk->groups[0] & ~0xffffffffUL) | groups;
1587 netlink_update_listeners(sk);
1588 netlink_table_ungrab();
1589
1590 return 0;
1591 }
1592
1593 static int netlink_connect(struct socket *sock, struct sockaddr *addr,
1594 int alen, int flags)
1595 {
1596 int err = 0;
1597 struct sock *sk = sock->sk;
1598 struct netlink_sock *nlk = nlk_sk(sk);
1599 struct sockaddr_nl *nladdr = (struct sockaddr_nl *)addr;
1600
1601 if (alen < sizeof(addr->sa_family))
1602 return -EINVAL;
1603
1604 if (addr->sa_family == AF_UNSPEC) {
1605 sk->sk_state = NETLINK_UNCONNECTED;
1606 nlk->dst_portid = 0;
1607 nlk->dst_group = 0;
1608 return 0;
1609 }
1610 if (addr->sa_family != AF_NETLINK)
1611 return -EINVAL;
1612
1613 if ((nladdr->nl_groups || nladdr->nl_pid) &&
1614 !netlink_allowed(sock, NL_CFG_F_NONROOT_SEND))
1615 return -EPERM;
1616
1617 /* No need for barriers here as we return to user-space without
1618 * using any of the bound attributes.
1619 */
1620 if (!nlk->bound)
1621 err = netlink_autobind(sock);
1622
1623 if (err == 0) {
1624 sk->sk_state = NETLINK_CONNECTED;
1625 nlk->dst_portid = nladdr->nl_pid;
1626 nlk->dst_group = ffs(nladdr->nl_groups);
1627 }
1628
1629 return err;
1630 }
1631
1632 static int netlink_getname(struct socket *sock, struct sockaddr *addr,
1633 int *addr_len, int peer)
1634 {
1635 struct sock *sk = sock->sk;
1636 struct netlink_sock *nlk = nlk_sk(sk);
1637 DECLARE_SOCKADDR(struct sockaddr_nl *, nladdr, addr);
1638
1639 nladdr->nl_family = AF_NETLINK;
1640 nladdr->nl_pad = 0;
1641 *addr_len = sizeof(*nladdr);
1642
1643 if (peer) {
1644 nladdr->nl_pid = nlk->dst_portid;
1645 nladdr->nl_groups = netlink_group_mask(nlk->dst_group);
1646 } else {
1647 nladdr->nl_pid = nlk->portid;
1648 nladdr->nl_groups = nlk->groups ? nlk->groups[0] : 0;
1649 }
1650 return 0;
1651 }
1652
1653 static struct sock *netlink_getsockbyportid(struct sock *ssk, u32 portid)
1654 {
1655 struct sock *sock;
1656 struct netlink_sock *nlk;
1657
1658 sock = netlink_lookup(sock_net(ssk), ssk->sk_protocol, portid);
1659 if (!sock)
1660 return ERR_PTR(-ECONNREFUSED);
1661
1662 /* Don't bother queuing skb if kernel socket has no input function */
1663 nlk = nlk_sk(sock);
1664 if (sock->sk_state == NETLINK_CONNECTED &&
1665 nlk->dst_portid != nlk_sk(ssk)->portid) {
1666 sock_put(sock);
1667 return ERR_PTR(-ECONNREFUSED);
1668 }
1669 return sock;
1670 }
1671
1672 struct sock *netlink_getsockbyfilp(struct file *filp)
1673 {
1674 struct inode *inode = file_inode(filp);
1675 struct sock *sock;
1676
1677 if (!S_ISSOCK(inode->i_mode))
1678 return ERR_PTR(-ENOTSOCK);
1679
1680 sock = SOCKET_I(inode)->sk;
1681 if (sock->sk_family != AF_NETLINK)
1682 return ERR_PTR(-EINVAL);
1683
1684 sock_hold(sock);
1685 return sock;
1686 }
1687
1688 static struct sk_buff *netlink_alloc_large_skb(unsigned int size,
1689 int broadcast)
1690 {
1691 struct sk_buff *skb;
1692 void *data;
1693
1694 if (size <= NLMSG_GOODSIZE || broadcast)
1695 return alloc_skb(size, GFP_KERNEL);
1696
1697 size = SKB_DATA_ALIGN(size) +
1698 SKB_DATA_ALIGN(sizeof(struct skb_shared_info));
1699
1700 data = vmalloc(size);
1701 if (data == NULL)
1702 return NULL;
1703
1704 skb = __build_skb(data, size);
1705 if (skb == NULL)
1706 vfree(data);
1707 else
1708 skb->destructor = netlink_skb_destructor;
1709
1710 return skb;
1711 }
1712
1713 /*
1714 * Attach a skb to a netlink socket.
1715 * The caller must hold a reference to the destination socket. On error, the
1716 * reference is dropped. The skb is not send to the destination, just all
1717 * all error checks are performed and memory in the queue is reserved.
1718 * Return values:
1719 * < 0: error. skb freed, reference to sock dropped.
1720 * 0: continue
1721 * 1: repeat lookup - reference dropped while waiting for socket memory.
1722 */
1723 int netlink_attachskb(struct sock *sk, struct sk_buff *skb,
1724 long *timeo, struct sock *ssk)
1725 {
1726 struct netlink_sock *nlk;
1727
1728 nlk = nlk_sk(sk);
1729
1730 if ((atomic_read(&sk->sk_rmem_alloc) > sk->sk_rcvbuf ||
1731 test_bit(NETLINK_S_CONGESTED, &nlk->state)) &&
1732 !netlink_skb_is_mmaped(skb)) {
1733 DECLARE_WAITQUEUE(wait, current);
1734 if (!*timeo) {
1735 if (!ssk || netlink_is_kernel(ssk))
1736 netlink_overrun(sk);
1737 sock_put(sk);
1738 kfree_skb(skb);
1739 return -EAGAIN;
1740 }
1741
1742 __set_current_state(TASK_INTERRUPTIBLE);
1743 add_wait_queue(&nlk->wait, &wait);
1744
1745 if ((atomic_read(&sk->sk_rmem_alloc) > sk->sk_rcvbuf ||
1746 test_bit(NETLINK_S_CONGESTED, &nlk->state)) &&
1747 !sock_flag(sk, SOCK_DEAD))
1748 *timeo = schedule_timeout(*timeo);
1749
1750 __set_current_state(TASK_RUNNING);
1751 remove_wait_queue(&nlk->wait, &wait);
1752 sock_put(sk);
1753
1754 if (signal_pending(current)) {
1755 kfree_skb(skb);
1756 return sock_intr_errno(*timeo);
1757 }
1758 return 1;
1759 }
1760 netlink_skb_set_owner_r(skb, sk);
1761 return 0;
1762 }
1763
1764 static int __netlink_sendskb(struct sock *sk, struct sk_buff *skb)
1765 {
1766 int len = skb->len;
1767
1768 netlink_deliver_tap(skb);
1769
1770 #ifdef CONFIG_NETLINK_MMAP
1771 if (netlink_skb_is_mmaped(skb))
1772 netlink_queue_mmaped_skb(sk, skb);
1773 else if (netlink_rx_is_mmaped(sk))
1774 netlink_ring_set_copied(sk, skb);
1775 else
1776 #endif /* CONFIG_NETLINK_MMAP */
1777 skb_queue_tail(&sk->sk_receive_queue, skb);
1778 sk->sk_data_ready(sk);
1779 return len;
1780 }
1781
1782 int netlink_sendskb(struct sock *sk, struct sk_buff *skb)
1783 {
1784 int len = __netlink_sendskb(sk, skb);
1785
1786 sock_put(sk);
1787 return len;
1788 }
1789
1790 void netlink_detachskb(struct sock *sk, struct sk_buff *skb)
1791 {
1792 kfree_skb(skb);
1793 sock_put(sk);
1794 }
1795
1796 static struct sk_buff *netlink_trim(struct sk_buff *skb, gfp_t allocation)
1797 {
1798 int delta;
1799
1800 WARN_ON(skb->sk != NULL);
1801 if (netlink_skb_is_mmaped(skb))
1802 return skb;
1803
1804 delta = skb->end - skb->tail;
1805 if (is_vmalloc_addr(skb->head) || delta * 2 < skb->truesize)
1806 return skb;
1807
1808 if (skb_shared(skb)) {
1809 struct sk_buff *nskb = skb_clone(skb, allocation);
1810 if (!nskb)
1811 return skb;
1812 consume_skb(skb);
1813 skb = nskb;
1814 }
1815
1816 if (!pskb_expand_head(skb, 0, -delta, allocation))
1817 skb->truesize -= delta;
1818
1819 return skb;
1820 }
1821
1822 static int netlink_unicast_kernel(struct sock *sk, struct sk_buff *skb,
1823 struct sock *ssk)
1824 {
1825 int ret;
1826 struct netlink_sock *nlk = nlk_sk(sk);
1827
1828 ret = -ECONNREFUSED;
1829 if (nlk->netlink_rcv != NULL) {
1830 ret = skb->len;
1831 netlink_skb_set_owner_r(skb, sk);
1832 NETLINK_CB(skb).sk = ssk;
1833 netlink_deliver_tap_kernel(sk, ssk, skb);
1834 nlk->netlink_rcv(skb);
1835 consume_skb(skb);
1836 } else {
1837 kfree_skb(skb);
1838 }
1839 sock_put(sk);
1840 return ret;
1841 }
1842
1843 int netlink_unicast(struct sock *ssk, struct sk_buff *skb,
1844 u32 portid, int nonblock)
1845 {
1846 struct sock *sk;
1847 int err;
1848 long timeo;
1849
1850 skb = netlink_trim(skb, gfp_any());
1851
1852 timeo = sock_sndtimeo(ssk, nonblock);
1853 retry:
1854 sk = netlink_getsockbyportid(ssk, portid);
1855 if (IS_ERR(sk)) {
1856 kfree_skb(skb);
1857 return PTR_ERR(sk);
1858 }
1859 if (netlink_is_kernel(sk))
1860 return netlink_unicast_kernel(sk, skb, ssk);
1861
1862 if (sk_filter(sk, skb)) {
1863 err = skb->len;
1864 kfree_skb(skb);
1865 sock_put(sk);
1866 return err;
1867 }
1868
1869 err = netlink_attachskb(sk, skb, &timeo, ssk);
1870 if (err == 1)
1871 goto retry;
1872 if (err)
1873 return err;
1874
1875 return netlink_sendskb(sk, skb);
1876 }
1877 EXPORT_SYMBOL(netlink_unicast);
1878
1879 struct sk_buff *__netlink_alloc_skb(struct sock *ssk, unsigned int size,
1880 unsigned int ldiff, u32 dst_portid,
1881 gfp_t gfp_mask)
1882 {
1883 #ifdef CONFIG_NETLINK_MMAP
1884 unsigned int maxlen, linear_size;
1885 struct sock *sk = NULL;
1886 struct sk_buff *skb;
1887 struct netlink_ring *ring;
1888 struct nl_mmap_hdr *hdr;
1889
1890 sk = netlink_getsockbyportid(ssk, dst_portid);
1891 if (IS_ERR(sk))
1892 goto out;
1893
1894 ring = &nlk_sk(sk)->rx_ring;
1895 /* fast-path without atomic ops for common case: non-mmaped receiver */
1896 if (ring->pg_vec == NULL)
1897 goto out_put;
1898
1899 /* We need to account the full linear size needed as a ring
1900 * slot cannot have non-linear parts.
1901 */
1902 linear_size = size + ldiff;
1903 if (ring->frame_size - NL_MMAP_HDRLEN < linear_size)
1904 goto out_put;
1905
1906 skb = alloc_skb_head(gfp_mask);
1907 if (skb == NULL)
1908 goto err1;
1909
1910 spin_lock_bh(&sk->sk_receive_queue.lock);
1911 /* check again under lock */
1912 if (ring->pg_vec == NULL)
1913 goto out_free;
1914
1915 /* check again under lock */
1916 maxlen = ring->frame_size - NL_MMAP_HDRLEN;
1917 if (maxlen < linear_size)
1918 goto out_free;
1919
1920 netlink_forward_ring(ring);
1921 hdr = netlink_current_frame(ring, NL_MMAP_STATUS_UNUSED);
1922 if (hdr == NULL)
1923 goto err2;
1924
1925 netlink_ring_setup_skb(skb, sk, ring, hdr);
1926 netlink_set_status(hdr, NL_MMAP_STATUS_RESERVED);
1927 atomic_inc(&ring->pending);
1928 netlink_increment_head(ring);
1929
1930 spin_unlock_bh(&sk->sk_receive_queue.lock);
1931 return skb;
1932
1933 err2:
1934 kfree_skb(skb);
1935 spin_unlock_bh(&sk->sk_receive_queue.lock);
1936 netlink_overrun(sk);
1937 err1:
1938 sock_put(sk);
1939 return NULL;
1940
1941 out_free:
1942 kfree_skb(skb);
1943 spin_unlock_bh(&sk->sk_receive_queue.lock);
1944 out_put:
1945 sock_put(sk);
1946 out:
1947 #endif
1948 return alloc_skb(size, gfp_mask);
1949 }
1950 EXPORT_SYMBOL_GPL(__netlink_alloc_skb);
1951
1952 int netlink_has_listeners(struct sock *sk, unsigned int group)
1953 {
1954 int res = 0;
1955 struct listeners *listeners;
1956
1957 BUG_ON(!netlink_is_kernel(sk));
1958
1959 rcu_read_lock();
1960 listeners = rcu_dereference(nl_table[sk->sk_protocol].listeners);
1961
1962 if (listeners && group - 1 < nl_table[sk->sk_protocol].groups)
1963 res = test_bit(group - 1, listeners->masks);
1964
1965 rcu_read_unlock();
1966
1967 return res;
1968 }
1969 EXPORT_SYMBOL_GPL(netlink_has_listeners);
1970
1971 static int netlink_broadcast_deliver(struct sock *sk, struct sk_buff *skb)
1972 {
1973 struct netlink_sock *nlk = nlk_sk(sk);
1974
1975 if (atomic_read(&sk->sk_rmem_alloc) <= sk->sk_rcvbuf &&
1976 !test_bit(NETLINK_S_CONGESTED, &nlk->state)) {
1977 netlink_skb_set_owner_r(skb, sk);
1978 __netlink_sendskb(sk, skb);
1979 return atomic_read(&sk->sk_rmem_alloc) > (sk->sk_rcvbuf >> 1);
1980 }
1981 return -1;
1982 }
1983
1984 struct netlink_broadcast_data {
1985 struct sock *exclude_sk;
1986 struct net *net;
1987 u32 portid;
1988 u32 group;
1989 int failure;
1990 int delivery_failure;
1991 int congested;
1992 int delivered;
1993 gfp_t allocation;
1994 struct sk_buff *skb, *skb2;
1995 int (*tx_filter)(struct sock *dsk, struct sk_buff *skb, void *data);
1996 void *tx_data;
1997 };
1998
1999 static void do_one_broadcast(struct sock *sk,
2000 struct netlink_broadcast_data *p)
2001 {
2002 struct netlink_sock *nlk = nlk_sk(sk);
2003 int val;
2004
2005 if (p->exclude_sk == sk)
2006 return;
2007
2008 if (nlk->portid == p->portid || p->group - 1 >= nlk->ngroups ||
2009 !test_bit(p->group - 1, nlk->groups))
2010 return;
2011
2012 if (!net_eq(sock_net(sk), p->net)) {
2013 if (!(nlk->flags & NETLINK_F_LISTEN_ALL_NSID))
2014 return;
2015
2016 if (!peernet_has_id(sock_net(sk), p->net))
2017 return;
2018
2019 if (!file_ns_capable(sk->sk_socket->file, p->net->user_ns,
2020 CAP_NET_BROADCAST))
2021 return;
2022 }
2023
2024 if (p->failure) {
2025 netlink_overrun(sk);
2026 return;
2027 }
2028
2029 sock_hold(sk);
2030 if (p->skb2 == NULL) {
2031 if (skb_shared(p->skb)) {
2032 p->skb2 = skb_clone(p->skb, p->allocation);
2033 } else {
2034 p->skb2 = skb_get(p->skb);
2035 /*
2036 * skb ownership may have been set when
2037 * delivered to a previous socket.
2038 */
2039 skb_orphan(p->skb2);
2040 }
2041 }
2042 if (p->skb2 == NULL) {
2043 netlink_overrun(sk);
2044 /* Clone failed. Notify ALL listeners. */
2045 p->failure = 1;
2046 if (nlk->flags & NETLINK_F_BROADCAST_SEND_ERROR)
2047 p->delivery_failure = 1;
2048 goto out;
2049 }
2050 if (p->tx_filter && p->tx_filter(sk, p->skb2, p->tx_data)) {
2051 kfree_skb(p->skb2);
2052 p->skb2 = NULL;
2053 goto out;
2054 }
2055 if (sk_filter(sk, p->skb2)) {
2056 kfree_skb(p->skb2);
2057 p->skb2 = NULL;
2058 goto out;
2059 }
2060 NETLINK_CB(p->skb2).nsid = peernet2id(sock_net(sk), p->net);
2061 NETLINK_CB(p->skb2).nsid_is_set = true;
2062 val = netlink_broadcast_deliver(sk, p->skb2);
2063 if (val < 0) {
2064 netlink_overrun(sk);
2065 if (nlk->flags & NETLINK_F_BROADCAST_SEND_ERROR)
2066 p->delivery_failure = 1;
2067 } else {
2068 p->congested |= val;
2069 p->delivered = 1;
2070 p->skb2 = NULL;
2071 }
2072 out:
2073 sock_put(sk);
2074 }
2075
2076 int netlink_broadcast_filtered(struct sock *ssk, struct sk_buff *skb, u32 portid,
2077 u32 group, gfp_t allocation,
2078 int (*filter)(struct sock *dsk, struct sk_buff *skb, void *data),
2079 void *filter_data)
2080 {
2081 struct net *net = sock_net(ssk);
2082 struct netlink_broadcast_data info;
2083 struct sock *sk;
2084
2085 skb = netlink_trim(skb, allocation);
2086
2087 info.exclude_sk = ssk;
2088 info.net = net;
2089 info.portid = portid;
2090 info.group = group;
2091 info.failure = 0;
2092 info.delivery_failure = 0;
2093 info.congested = 0;
2094 info.delivered = 0;
2095 info.allocation = allocation;
2096 info.skb = skb;
2097 info.skb2 = NULL;
2098 info.tx_filter = filter;
2099 info.tx_data = filter_data;
2100
2101 /* While we sleep in clone, do not allow to change socket list */
2102
2103 netlink_lock_table();
2104
2105 sk_for_each_bound(sk, &nl_table[ssk->sk_protocol].mc_list)
2106 do_one_broadcast(sk, &info);
2107
2108 consume_skb(skb);
2109
2110 netlink_unlock_table();
2111
2112 if (info.delivery_failure) {
2113 kfree_skb(info.skb2);
2114 return -ENOBUFS;
2115 }
2116 consume_skb(info.skb2);
2117
2118 if (info.delivered) {
2119 if (info.congested && gfpflags_allow_blocking(allocation))
2120 yield();
2121 return 0;
2122 }
2123 return -ESRCH;
2124 }
2125 EXPORT_SYMBOL(netlink_broadcast_filtered);
2126
2127 int netlink_broadcast(struct sock *ssk, struct sk_buff *skb, u32 portid,
2128 u32 group, gfp_t allocation)
2129 {
2130 return netlink_broadcast_filtered(ssk, skb, portid, group, allocation,
2131 NULL, NULL);
2132 }
2133 EXPORT_SYMBOL(netlink_broadcast);
2134
2135 struct netlink_set_err_data {
2136 struct sock *exclude_sk;
2137 u32 portid;
2138 u32 group;
2139 int code;
2140 };
2141
2142 static int do_one_set_err(struct sock *sk, struct netlink_set_err_data *p)
2143 {
2144 struct netlink_sock *nlk = nlk_sk(sk);
2145 int ret = 0;
2146
2147 if (sk == p->exclude_sk)
2148 goto out;
2149
2150 if (!net_eq(sock_net(sk), sock_net(p->exclude_sk)))
2151 goto out;
2152
2153 if (nlk->portid == p->portid || p->group - 1 >= nlk->ngroups ||
2154 !test_bit(p->group - 1, nlk->groups))
2155 goto out;
2156
2157 if (p->code == ENOBUFS && nlk->flags & NETLINK_F_RECV_NO_ENOBUFS) {
2158 ret = 1;
2159 goto out;
2160 }
2161
2162 sk->sk_err = p->code;
2163 sk->sk_error_report(sk);
2164 out:
2165 return ret;
2166 }
2167
2168 /**
2169 * netlink_set_err - report error to broadcast listeners
2170 * @ssk: the kernel netlink socket, as returned by netlink_kernel_create()
2171 * @portid: the PORTID of a process that we want to skip (if any)
2172 * @group: the broadcast group that will notice the error
2173 * @code: error code, must be negative (as usual in kernelspace)
2174 *
2175 * This function returns the number of broadcast listeners that have set the
2176 * NETLINK_NO_ENOBUFS socket option.
2177 */
2178 int netlink_set_err(struct sock *ssk, u32 portid, u32 group, int code)
2179 {
2180 struct netlink_set_err_data info;
2181 struct sock *sk;
2182 int ret = 0;
2183
2184 info.exclude_sk = ssk;
2185 info.portid = portid;
2186 info.group = group;
2187 /* sk->sk_err wants a positive error value */
2188 info.code = -code;
2189
2190 read_lock(&nl_table_lock);
2191
2192 sk_for_each_bound(sk, &nl_table[ssk->sk_protocol].mc_list)
2193 ret += do_one_set_err(sk, &info);
2194
2195 read_unlock(&nl_table_lock);
2196 return ret;
2197 }
2198 EXPORT_SYMBOL(netlink_set_err);
2199
2200 /* must be called with netlink table grabbed */
2201 static void netlink_update_socket_mc(struct netlink_sock *nlk,
2202 unsigned int group,
2203 int is_new)
2204 {
2205 int old, new = !!is_new, subscriptions;
2206
2207 old = test_bit(group - 1, nlk->groups);
2208 subscriptions = nlk->subscriptions - old + new;
2209 if (new)
2210 __set_bit(group - 1, nlk->groups);
2211 else
2212 __clear_bit(group - 1, nlk->groups);
2213 netlink_update_subscriptions(&nlk->sk, subscriptions);
2214 netlink_update_listeners(&nlk->sk);
2215 }
2216
2217 static int netlink_setsockopt(struct socket *sock, int level, int optname,
2218 char __user *optval, unsigned int optlen)
2219 {
2220 struct sock *sk = sock->sk;
2221 struct netlink_sock *nlk = nlk_sk(sk);
2222 unsigned int val = 0;
2223 int err;
2224
2225 if (level != SOL_NETLINK)
2226 return -ENOPROTOOPT;
2227
2228 if (optname != NETLINK_RX_RING && optname != NETLINK_TX_RING &&
2229 optlen >= sizeof(int) &&
2230 get_user(val, (unsigned int __user *)optval))
2231 return -EFAULT;
2232
2233 switch (optname) {
2234 case NETLINK_PKTINFO:
2235 if (val)
2236 nlk->flags |= NETLINK_F_RECV_PKTINFO;
2237 else
2238 nlk->flags &= ~NETLINK_F_RECV_PKTINFO;
2239 err = 0;
2240 break;
2241 case NETLINK_ADD_MEMBERSHIP:
2242 case NETLINK_DROP_MEMBERSHIP: {
2243 if (!netlink_allowed(sock, NL_CFG_F_NONROOT_RECV))
2244 return -EPERM;
2245 err = netlink_realloc_groups(sk);
2246 if (err)
2247 return err;
2248 if (!val || val - 1 >= nlk->ngroups)
2249 return -EINVAL;
2250 if (optname == NETLINK_ADD_MEMBERSHIP && nlk->netlink_bind) {
2251 err = nlk->netlink_bind(sock_net(sk), val);
2252 if (err)
2253 return err;
2254 }
2255 netlink_table_grab();
2256 netlink_update_socket_mc(nlk, val,
2257 optname == NETLINK_ADD_MEMBERSHIP);
2258 netlink_table_ungrab();
2259 if (optname == NETLINK_DROP_MEMBERSHIP && nlk->netlink_unbind)
2260 nlk->netlink_unbind(sock_net(sk), val);
2261
2262 err = 0;
2263 break;
2264 }
2265 case NETLINK_BROADCAST_ERROR:
2266 if (val)
2267 nlk->flags |= NETLINK_F_BROADCAST_SEND_ERROR;
2268 else
2269 nlk->flags &= ~NETLINK_F_BROADCAST_SEND_ERROR;
2270 err = 0;
2271 break;
2272 case NETLINK_NO_ENOBUFS:
2273 if (val) {
2274 nlk->flags |= NETLINK_F_RECV_NO_ENOBUFS;
2275 clear_bit(NETLINK_S_CONGESTED, &nlk->state);
2276 wake_up_interruptible(&nlk->wait);
2277 } else {
2278 nlk->flags &= ~NETLINK_F_RECV_NO_ENOBUFS;
2279 }
2280 err = 0;
2281 break;
2282 #ifdef CONFIG_NETLINK_MMAP
2283 case NETLINK_RX_RING:
2284 case NETLINK_TX_RING: {
2285 struct nl_mmap_req req;
2286
2287 /* Rings might consume more memory than queue limits, require
2288 * CAP_NET_ADMIN.
2289 */
2290 if (!capable(CAP_NET_ADMIN))
2291 return -EPERM;
2292 if (optlen < sizeof(req))
2293 return -EINVAL;
2294 if (copy_from_user(&req, optval, sizeof(req)))
2295 return -EFAULT;
2296 err = netlink_set_ring(sk, &req,
2297 optname == NETLINK_TX_RING);
2298 break;
2299 }
2300 #endif /* CONFIG_NETLINK_MMAP */
2301 case NETLINK_LISTEN_ALL_NSID:
2302 if (!ns_capable(sock_net(sk)->user_ns, CAP_NET_BROADCAST))
2303 return -EPERM;
2304
2305 if (val)
2306 nlk->flags |= NETLINK_F_LISTEN_ALL_NSID;
2307 else
2308 nlk->flags &= ~NETLINK_F_LISTEN_ALL_NSID;
2309 err = 0;
2310 break;
2311 case NETLINK_CAP_ACK:
2312 if (val)
2313 nlk->flags |= NETLINK_F_CAP_ACK;
2314 else
2315 nlk->flags &= ~NETLINK_F_CAP_ACK;
2316 err = 0;
2317 break;
2318 default:
2319 err = -ENOPROTOOPT;
2320 }
2321 return err;
2322 }
2323
2324 static int netlink_getsockopt(struct socket *sock, int level, int optname,
2325 char __user *optval, int __user *optlen)
2326 {
2327 struct sock *sk = sock->sk;
2328 struct netlink_sock *nlk = nlk_sk(sk);
2329 int len, val, err;
2330
2331 if (level != SOL_NETLINK)
2332 return -ENOPROTOOPT;
2333
2334 if (get_user(len, optlen))
2335 return -EFAULT;
2336 if (len < 0)
2337 return -EINVAL;
2338
2339 switch (optname) {
2340 case NETLINK_PKTINFO:
2341 if (len < sizeof(int))
2342 return -EINVAL;
2343 len = sizeof(int);
2344 val = nlk->flags & NETLINK_F_RECV_PKTINFO ? 1 : 0;
2345 if (put_user(len, optlen) ||
2346 put_user(val, optval))
2347 return -EFAULT;
2348 err = 0;
2349 break;
2350 case NETLINK_BROADCAST_ERROR:
2351 if (len < sizeof(int))
2352 return -EINVAL;
2353 len = sizeof(int);
2354 val = nlk->flags & NETLINK_F_BROADCAST_SEND_ERROR ? 1 : 0;
2355 if (put_user(len, optlen) ||
2356 put_user(val, optval))
2357 return -EFAULT;
2358 err = 0;
2359 break;
2360 case NETLINK_NO_ENOBUFS:
2361 if (len < sizeof(int))
2362 return -EINVAL;
2363 len = sizeof(int);
2364 val = nlk->flags & NETLINK_F_RECV_NO_ENOBUFS ? 1 : 0;
2365 if (put_user(len, optlen) ||
2366 put_user(val, optval))
2367 return -EFAULT;
2368 err = 0;
2369 break;
2370 case NETLINK_LIST_MEMBERSHIPS: {
2371 int pos, idx, shift;
2372
2373 err = 0;
2374 netlink_lock_table();
2375 for (pos = 0; pos * 8 < nlk->ngroups; pos += sizeof(u32)) {
2376 if (len - pos < sizeof(u32))
2377 break;
2378
2379 idx = pos / sizeof(unsigned long);
2380 shift = (pos % sizeof(unsigned long)) * 8;
2381 if (put_user((u32)(nlk->groups[idx] >> shift),
2382 (u32 __user *)(optval + pos))) {
2383 err = -EFAULT;
2384 break;
2385 }
2386 }
2387 if (put_user(ALIGN(nlk->ngroups / 8, sizeof(u32)), optlen))
2388 err = -EFAULT;
2389 netlink_unlock_table();
2390 break;
2391 }
2392 case NETLINK_CAP_ACK:
2393 if (len < sizeof(int))
2394 return -EINVAL;
2395 len = sizeof(int);
2396 val = nlk->flags & NETLINK_F_CAP_ACK ? 1 : 0;
2397 if (put_user(len, optlen) ||
2398 put_user(val, optval))
2399 return -EFAULT;
2400 err = 0;
2401 break;
2402 default:
2403 err = -ENOPROTOOPT;
2404 }
2405 return err;
2406 }
2407
2408 static void netlink_cmsg_recv_pktinfo(struct msghdr *msg, struct sk_buff *skb)
2409 {
2410 struct nl_pktinfo info;
2411
2412 info.group = NETLINK_CB(skb).dst_group;
2413 put_cmsg(msg, SOL_NETLINK, NETLINK_PKTINFO, sizeof(info), &info);
2414 }
2415
2416 static void netlink_cmsg_listen_all_nsid(struct sock *sk, struct msghdr *msg,
2417 struct sk_buff *skb)
2418 {
2419 if (!NETLINK_CB(skb).nsid_is_set)
2420 return;
2421
2422 put_cmsg(msg, SOL_NETLINK, NETLINK_LISTEN_ALL_NSID, sizeof(int),
2423 &NETLINK_CB(skb).nsid);
2424 }
2425
2426 static int netlink_sendmsg(struct socket *sock, struct msghdr *msg, size_t len)
2427 {
2428 struct sock *sk = sock->sk;
2429 struct netlink_sock *nlk = nlk_sk(sk);
2430 DECLARE_SOCKADDR(struct sockaddr_nl *, addr, msg->msg_name);
2431 u32 dst_portid;
2432 u32 dst_group;
2433 struct sk_buff *skb;
2434 int err;
2435 struct scm_cookie scm;
2436 u32 netlink_skb_flags = 0;
2437
2438 if (msg->msg_flags&MSG_OOB)
2439 return -EOPNOTSUPP;
2440
2441 err = scm_send(sock, msg, &scm, true);
2442 if (err < 0)
2443 return err;
2444
2445 if (msg->msg_namelen) {
2446 err = -EINVAL;
2447 if (addr->nl_family != AF_NETLINK)
2448 goto out;
2449 dst_portid = addr->nl_pid;
2450 dst_group = ffs(addr->nl_groups);
2451 err = -EPERM;
2452 if ((dst_group || dst_portid) &&
2453 !netlink_allowed(sock, NL_CFG_F_NONROOT_SEND))
2454 goto out;
2455 netlink_skb_flags |= NETLINK_SKB_DST;
2456 } else {
2457 dst_portid = nlk->dst_portid;
2458 dst_group = nlk->dst_group;
2459 }
2460
2461 if (!nlk->bound) {
2462 err = netlink_autobind(sock);
2463 if (err)
2464 goto out;
2465 } else {
2466 /* Ensure nlk is hashed and visible. */
2467 smp_rmb();
2468 }
2469
2470 /* It's a really convoluted way for userland to ask for mmaped
2471 * sendmsg(), but that's what we've got...
2472 */
2473 if (netlink_tx_is_mmaped(sk) &&
2474 iter_is_iovec(&msg->msg_iter) &&
2475 msg->msg_iter.nr_segs == 1 &&
2476 msg->msg_iter.iov->iov_base == NULL) {
2477 err = netlink_mmap_sendmsg(sk, msg, dst_portid, dst_group,
2478 &scm);
2479 goto out;
2480 }
2481
2482 err = -EMSGSIZE;
2483 if (len > sk->sk_sndbuf - 32)
2484 goto out;
2485 err = -ENOBUFS;
2486 skb = netlink_alloc_large_skb(len, dst_group);
2487 if (skb == NULL)
2488 goto out;
2489
2490 NETLINK_CB(skb).portid = nlk->portid;
2491 NETLINK_CB(skb).dst_group = dst_group;
2492 NETLINK_CB(skb).creds = scm.creds;
2493 NETLINK_CB(skb).flags = netlink_skb_flags;
2494
2495 err = -EFAULT;
2496 if (memcpy_from_msg(skb_put(skb, len), msg, len)) {
2497 kfree_skb(skb);
2498 goto out;
2499 }
2500
2501 err = security_netlink_send(sk, skb);
2502 if (err) {
2503 kfree_skb(skb);
2504 goto out;
2505 }
2506
2507 if (dst_group) {
2508 atomic_inc(&skb->users);
2509 netlink_broadcast(sk, skb, dst_portid, dst_group, GFP_KERNEL);
2510 }
2511 err = netlink_unicast(sk, skb, dst_portid, msg->msg_flags&MSG_DONTWAIT);
2512
2513 out:
2514 scm_destroy(&scm);
2515 return err;
2516 }
2517
2518 static int netlink_recvmsg(struct socket *sock, struct msghdr *msg, size_t len,
2519 int flags)
2520 {
2521 struct scm_cookie scm;
2522 struct sock *sk = sock->sk;
2523 struct netlink_sock *nlk = nlk_sk(sk);
2524 int noblock = flags&MSG_DONTWAIT;
2525 size_t copied;
2526 struct sk_buff *skb, *data_skb;
2527 int err, ret;
2528
2529 if (flags&MSG_OOB)
2530 return -EOPNOTSUPP;
2531
2532 copied = 0;
2533
2534 skb = skb_recv_datagram(sk, flags, noblock, &err);
2535 if (skb == NULL)
2536 goto out;
2537
2538 data_skb = skb;
2539
2540 #ifdef CONFIG_COMPAT_NETLINK_MESSAGES
2541 if (unlikely(skb_shinfo(skb)->frag_list)) {
2542 /*
2543 * If this skb has a frag_list, then here that means that we
2544 * will have to use the frag_list skb's data for compat tasks
2545 * and the regular skb's data for normal (non-compat) tasks.
2546 *
2547 * If we need to send the compat skb, assign it to the
2548 * 'data_skb' variable so that it will be used below for data
2549 * copying. We keep 'skb' for everything else, including
2550 * freeing both later.
2551 */
2552 if (flags & MSG_CMSG_COMPAT)
2553 data_skb = skb_shinfo(skb)->frag_list;
2554 }
2555 #endif
2556
2557 /* Record the max length of recvmsg() calls for future allocations */
2558 nlk->max_recvmsg_len = max(nlk->max_recvmsg_len, len);
2559 nlk->max_recvmsg_len = min_t(size_t, nlk->max_recvmsg_len,
2560 16384);
2561
2562 copied = data_skb->len;
2563 if (len < copied) {
2564 msg->msg_flags |= MSG_TRUNC;
2565 copied = len;
2566 }
2567
2568 skb_reset_transport_header(data_skb);
2569 err = skb_copy_datagram_msg(data_skb, 0, msg, copied);
2570
2571 if (msg->msg_name) {
2572 DECLARE_SOCKADDR(struct sockaddr_nl *, addr, msg->msg_name);
2573 addr->nl_family = AF_NETLINK;
2574 addr->nl_pad = 0;
2575 addr->nl_pid = NETLINK_CB(skb).portid;
2576 addr->nl_groups = netlink_group_mask(NETLINK_CB(skb).dst_group);
2577 msg->msg_namelen = sizeof(*addr);
2578 }
2579
2580 if (nlk->flags & NETLINK_F_RECV_PKTINFO)
2581 netlink_cmsg_recv_pktinfo(msg, skb);
2582 if (nlk->flags & NETLINK_F_LISTEN_ALL_NSID)
2583 netlink_cmsg_listen_all_nsid(sk, msg, skb);
2584
2585 memset(&scm, 0, sizeof(scm));
2586 scm.creds = *NETLINK_CREDS(skb);
2587 if (flags & MSG_TRUNC)
2588 copied = data_skb->len;
2589
2590 skb_free_datagram(sk, skb);
2591
2592 if (nlk->cb_running &&
2593 atomic_read(&sk->sk_rmem_alloc) <= sk->sk_rcvbuf / 2) {
2594 ret = netlink_dump(sk);
2595 if (ret) {
2596 sk->sk_err = -ret;
2597 sk->sk_error_report(sk);
2598 }
2599 }
2600
2601 scm_recv(sock, msg, &scm, flags);
2602 out:
2603 netlink_rcv_wake(sk);
2604 return err ? : copied;
2605 }
2606
2607 static void netlink_data_ready(struct sock *sk)
2608 {
2609 BUG();
2610 }
2611
2612 /*
2613 * We export these functions to other modules. They provide a
2614 * complete set of kernel non-blocking support for message
2615 * queueing.
2616 */
2617
2618 struct sock *
2619 __netlink_kernel_create(struct net *net, int unit, struct module *module,
2620 struct netlink_kernel_cfg *cfg)
2621 {
2622 struct socket *sock;
2623 struct sock *sk;
2624 struct netlink_sock *nlk;
2625 struct listeners *listeners = NULL;
2626 struct mutex *cb_mutex = cfg ? cfg->cb_mutex : NULL;
2627 unsigned int groups;
2628
2629 BUG_ON(!nl_table);
2630
2631 if (unit < 0 || unit >= MAX_LINKS)
2632 return NULL;
2633
2634 if (sock_create_lite(PF_NETLINK, SOCK_DGRAM, unit, &sock))
2635 return NULL;
2636
2637 if (__netlink_create(net, sock, cb_mutex, unit, 1) < 0)
2638 goto out_sock_release_nosk;
2639
2640 sk = sock->sk;
2641
2642 if (!cfg || cfg->groups < 32)
2643 groups = 32;
2644 else
2645 groups = cfg->groups;
2646
2647 listeners = kzalloc(sizeof(*listeners) + NLGRPSZ(groups), GFP_KERNEL);
2648 if (!listeners)
2649 goto out_sock_release;
2650
2651 sk->sk_data_ready = netlink_data_ready;
2652 if (cfg && cfg->input)
2653 nlk_sk(sk)->netlink_rcv = cfg->input;
2654
2655 if (netlink_insert(sk, 0))
2656 goto out_sock_release;
2657
2658 nlk = nlk_sk(sk);
2659 nlk->flags |= NETLINK_F_KERNEL_SOCKET;
2660
2661 netlink_table_grab();
2662 if (!nl_table[unit].registered) {
2663 nl_table[unit].groups = groups;
2664 rcu_assign_pointer(nl_table[unit].listeners, listeners);
2665 nl_table[unit].cb_mutex = cb_mutex;
2666 nl_table[unit].module = module;
2667 if (cfg) {
2668 nl_table[unit].bind = cfg->bind;
2669 nl_table[unit].unbind = cfg->unbind;
2670 nl_table[unit].flags = cfg->flags;
2671 if (cfg->compare)
2672 nl_table[unit].compare = cfg->compare;
2673 }
2674 nl_table[unit].registered = 1;
2675 } else {
2676 kfree(listeners);
2677 nl_table[unit].registered++;
2678 }
2679 netlink_table_ungrab();
2680 return sk;
2681
2682 out_sock_release:
2683 kfree(listeners);
2684 netlink_kernel_release(sk);
2685 return NULL;
2686
2687 out_sock_release_nosk:
2688 sock_release(sock);
2689 return NULL;
2690 }
2691 EXPORT_SYMBOL(__netlink_kernel_create);
2692
2693 void
2694 netlink_kernel_release(struct sock *sk)
2695 {
2696 if (sk == NULL || sk->sk_socket == NULL)
2697 return;
2698
2699 sock_release(sk->sk_socket);
2700 }
2701 EXPORT_SYMBOL(netlink_kernel_release);
2702
2703 int __netlink_change_ngroups(struct sock *sk, unsigned int groups)
2704 {
2705 struct listeners *new, *old;
2706 struct netlink_table *tbl = &nl_table[sk->sk_protocol];
2707
2708 if (groups < 32)
2709 groups = 32;
2710
2711 if (NLGRPSZ(tbl->groups) < NLGRPSZ(groups)) {
2712 new = kzalloc(sizeof(*new) + NLGRPSZ(groups), GFP_ATOMIC);
2713 if (!new)
2714 return -ENOMEM;
2715 old = nl_deref_protected(tbl->listeners);
2716 memcpy(new->masks, old->masks, NLGRPSZ(tbl->groups));
2717 rcu_assign_pointer(tbl->listeners, new);
2718
2719 kfree_rcu(old, rcu);
2720 }
2721 tbl->groups = groups;
2722
2723 return 0;
2724 }
2725
2726 /**
2727 * netlink_change_ngroups - change number of multicast groups
2728 *
2729 * This changes the number of multicast groups that are available
2730 * on a certain netlink family. Note that it is not possible to
2731 * change the number of groups to below 32. Also note that it does
2732 * not implicitly call netlink_clear_multicast_users() when the
2733 * number of groups is reduced.
2734 *
2735 * @sk: The kernel netlink socket, as returned by netlink_kernel_create().
2736 * @groups: The new number of groups.
2737 */
2738 int netlink_change_ngroups(struct sock *sk, unsigned int groups)
2739 {
2740 int err;
2741
2742 netlink_table_grab();
2743 err = __netlink_change_ngroups(sk, groups);
2744 netlink_table_ungrab();
2745
2746 return err;
2747 }
2748
2749 void __netlink_clear_multicast_users(struct sock *ksk, unsigned int group)
2750 {
2751 struct sock *sk;
2752 struct netlink_table *tbl = &nl_table[ksk->sk_protocol];
2753
2754 sk_for_each_bound(sk, &tbl->mc_list)
2755 netlink_update_socket_mc(nlk_sk(sk), group, 0);
2756 }
2757
2758 struct nlmsghdr *
2759 __nlmsg_put(struct sk_buff *skb, u32 portid, u32 seq, int type, int len, int flags)
2760 {
2761 struct nlmsghdr *nlh;
2762 int size = nlmsg_msg_size(len);
2763
2764 nlh = (struct nlmsghdr *)skb_put(skb, NLMSG_ALIGN(size));
2765 nlh->nlmsg_type = type;
2766 nlh->nlmsg_len = size;
2767 nlh->nlmsg_flags = flags;
2768 nlh->nlmsg_pid = portid;
2769 nlh->nlmsg_seq = seq;
2770 if (!__builtin_constant_p(size) || NLMSG_ALIGN(size) - size != 0)
2771 memset(nlmsg_data(nlh) + len, 0, NLMSG_ALIGN(size) - size);
2772 return nlh;
2773 }
2774 EXPORT_SYMBOL(__nlmsg_put);
2775
2776 /*
2777 * It looks a bit ugly.
2778 * It would be better to create kernel thread.
2779 */
2780
2781 static int netlink_dump(struct sock *sk)
2782 {
2783 struct netlink_sock *nlk = nlk_sk(sk);
2784 struct netlink_callback *cb;
2785 struct sk_buff *skb = NULL;
2786 struct nlmsghdr *nlh;
2787 int len, err = -ENOBUFS;
2788 int alloc_min_size;
2789 int alloc_size;
2790
2791 mutex_lock(nlk->cb_mutex);
2792 if (!nlk->cb_running) {
2793 err = -EINVAL;
2794 goto errout_skb;
2795 }
2796
2797 if (!netlink_rx_is_mmaped(sk) &&
2798 atomic_read(&sk->sk_rmem_alloc) >= sk->sk_rcvbuf)
2799 goto errout_skb;
2800
2801 /* NLMSG_GOODSIZE is small to avoid high order allocations being
2802 * required, but it makes sense to _attempt_ a 16K bytes allocation
2803 * to reduce number of system calls on dump operations, if user
2804 * ever provided a big enough buffer.
2805 */
2806 cb = &nlk->cb;
2807 alloc_min_size = max_t(int, cb->min_dump_alloc, NLMSG_GOODSIZE);
2808
2809 if (alloc_min_size < nlk->max_recvmsg_len) {
2810 alloc_size = nlk->max_recvmsg_len;
2811 skb = netlink_alloc_skb(sk, alloc_size, nlk->portid,
2812 GFP_KERNEL |
2813 __GFP_NOWARN |
2814 __GFP_NORETRY);
2815 }
2816 if (!skb) {
2817 alloc_size = alloc_min_size;
2818 skb = netlink_alloc_skb(sk, alloc_size, nlk->portid,
2819 GFP_KERNEL);
2820 }
2821 if (!skb)
2822 goto errout_skb;
2823
2824 /* Trim skb to allocated size. User is expected to provide buffer as
2825 * large as max(min_dump_alloc, 16KiB (mac_recvmsg_len capped at
2826 * netlink_recvmsg())). dump will pack as many smaller messages as
2827 * could fit within the allocated skb. skb is typically allocated
2828 * with larger space than required (could be as much as near 2x the
2829 * requested size with align to next power of 2 approach). Allowing
2830 * dump to use the excess space makes it difficult for a user to have a
2831 * reasonable static buffer based on the expected largest dump of a
2832 * single netdev. The outcome is MSG_TRUNC error.
2833 */
2834 if (!netlink_rx_is_mmaped(sk))
2835 skb_reserve(skb, skb_tailroom(skb) - alloc_size);
2836 netlink_skb_set_owner_r(skb, sk);
2837
2838 len = cb->dump(skb, cb);
2839
2840 if (len > 0) {
2841 mutex_unlock(nlk->cb_mutex);
2842
2843 if (sk_filter(sk, skb))
2844 kfree_skb(skb);
2845 else
2846 __netlink_sendskb(sk, skb);
2847 return 0;
2848 }
2849
2850 nlh = nlmsg_put_answer(skb, cb, NLMSG_DONE, sizeof(len), NLM_F_MULTI);
2851 if (!nlh)
2852 goto errout_skb;
2853
2854 nl_dump_check_consistent(cb, nlh);
2855
2856 memcpy(nlmsg_data(nlh), &len, sizeof(len));
2857
2858 if (sk_filter(sk, skb))
2859 kfree_skb(skb);
2860 else
2861 __netlink_sendskb(sk, skb);
2862
2863 if (cb->done)
2864 cb->done(cb);
2865
2866 nlk->cb_running = false;
2867 mutex_unlock(nlk->cb_mutex);
2868 module_put(cb->module);
2869 consume_skb(cb->skb);
2870 return 0;
2871
2872 errout_skb:
2873 mutex_unlock(nlk->cb_mutex);
2874 kfree_skb(skb);
2875 return err;
2876 }
2877
2878 int __netlink_dump_start(struct sock *ssk, struct sk_buff *skb,
2879 const struct nlmsghdr *nlh,
2880 struct netlink_dump_control *control)
2881 {
2882 struct netlink_callback *cb;
2883 struct sock *sk;
2884 struct netlink_sock *nlk;
2885 int ret;
2886
2887 /* Memory mapped dump requests need to be copied to avoid looping
2888 * on the pending state in netlink_mmap_sendmsg() while the CB hold
2889 * a reference to the skb.
2890 */
2891 if (netlink_skb_is_mmaped(skb)) {
2892 skb = skb_copy(skb, GFP_KERNEL);
2893 if (skb == NULL)
2894 return -ENOBUFS;
2895 } else
2896 atomic_inc(&skb->users);
2897
2898 sk = netlink_lookup(sock_net(ssk), ssk->sk_protocol, NETLINK_CB(skb).portid);
2899 if (sk == NULL) {
2900 ret = -ECONNREFUSED;
2901 goto error_free;
2902 }
2903
2904 nlk = nlk_sk(sk);
2905 mutex_lock(nlk->cb_mutex);
2906 /* A dump is in progress... */
2907 if (nlk->cb_running) {
2908 ret = -EBUSY;
2909 goto error_unlock;
2910 }
2911 /* add reference of module which cb->dump belongs to */
2912 if (!try_module_get(control->module)) {
2913 ret = -EPROTONOSUPPORT;
2914 goto error_unlock;
2915 }
2916
2917 cb = &nlk->cb;
2918 memset(cb, 0, sizeof(*cb));
2919 cb->start = control->start;
2920 cb->dump = control->dump;
2921 cb->done = control->done;
2922 cb->nlh = nlh;
2923 cb->data = control->data;
2924 cb->module = control->module;
2925 cb->min_dump_alloc = control->min_dump_alloc;
2926 cb->skb = skb;
2927
2928 nlk->cb_running = true;
2929
2930 mutex_unlock(nlk->cb_mutex);
2931
2932 if (cb->start)
2933 cb->start(cb);
2934
2935 ret = netlink_dump(sk);
2936 sock_put(sk);
2937
2938 if (ret)
2939 return ret;
2940
2941 /* We successfully started a dump, by returning -EINTR we
2942 * signal not to send ACK even if it was requested.
2943 */
2944 return -EINTR;
2945
2946 error_unlock:
2947 sock_put(sk);
2948 mutex_unlock(nlk->cb_mutex);
2949 error_free:
2950 kfree_skb(skb);
2951 return ret;
2952 }
2953 EXPORT_SYMBOL(__netlink_dump_start);
2954
2955 void netlink_ack(struct sk_buff *in_skb, struct nlmsghdr *nlh, int err)
2956 {
2957 struct sk_buff *skb;
2958 struct nlmsghdr *rep;
2959 struct nlmsgerr *errmsg;
2960 size_t payload = sizeof(*errmsg);
2961 struct netlink_sock *nlk = nlk_sk(NETLINK_CB(in_skb).sk);
2962
2963 /* Error messages get the original request appened, unless the user
2964 * requests to cap the error message.
2965 */
2966 if (!(nlk->flags & NETLINK_F_CAP_ACK) && err)
2967 payload += nlmsg_len(nlh);
2968
2969 skb = netlink_alloc_skb(in_skb->sk, nlmsg_total_size(payload),
2970 NETLINK_CB(in_skb).portid, GFP_KERNEL);
2971 if (!skb) {
2972 struct sock *sk;
2973
2974 sk = netlink_lookup(sock_net(in_skb->sk),
2975 in_skb->sk->sk_protocol,
2976 NETLINK_CB(in_skb).portid);
2977 if (sk) {
2978 sk->sk_err = ENOBUFS;
2979 sk->sk_error_report(sk);
2980 sock_put(sk);
2981 }
2982 return;
2983 }
2984
2985 rep = __nlmsg_put(skb, NETLINK_CB(in_skb).portid, nlh->nlmsg_seq,
2986 NLMSG_ERROR, payload, 0);
2987 errmsg = nlmsg_data(rep);
2988 errmsg->error = err;
2989 memcpy(&errmsg->msg, nlh, payload > sizeof(*errmsg) ? nlh->nlmsg_len : sizeof(*nlh));
2990 netlink_unicast(in_skb->sk, skb, NETLINK_CB(in_skb).portid, MSG_DONTWAIT);
2991 }
2992 EXPORT_SYMBOL(netlink_ack);
2993
2994 int netlink_rcv_skb(struct sk_buff *skb, int (*cb)(struct sk_buff *,
2995 struct nlmsghdr *))
2996 {
2997 struct nlmsghdr *nlh;
2998 int err;
2999
3000 while (skb->len >= nlmsg_total_size(0)) {
3001 int msglen;
3002
3003 nlh = nlmsg_hdr(skb);
3004 err = 0;
3005
3006 if (nlh->nlmsg_len < NLMSG_HDRLEN || skb->len < nlh->nlmsg_len)
3007 return 0;
3008
3009 /* Only requests are handled by the kernel */
3010 if (!(nlh->nlmsg_flags & NLM_F_REQUEST))
3011 goto ack;
3012
3013 /* Skip control messages */
3014 if (nlh->nlmsg_type < NLMSG_MIN_TYPE)
3015 goto ack;
3016
3017 err = cb(skb, nlh);
3018 if (err == -EINTR)
3019 goto skip;
3020
3021 ack:
3022 if (nlh->nlmsg_flags & NLM_F_ACK || err)
3023 netlink_ack(skb, nlh, err);
3024
3025 skip:
3026 msglen = NLMSG_ALIGN(nlh->nlmsg_len);
3027 if (msglen > skb->len)
3028 msglen = skb->len;
3029 skb_pull(skb, msglen);
3030 }
3031
3032 return 0;
3033 }
3034 EXPORT_SYMBOL(netlink_rcv_skb);
3035
3036 /**
3037 * nlmsg_notify - send a notification netlink message
3038 * @sk: netlink socket to use
3039 * @skb: notification message
3040 * @portid: destination netlink portid for reports or 0
3041 * @group: destination multicast group or 0
3042 * @report: 1 to report back, 0 to disable
3043 * @flags: allocation flags
3044 */
3045 int nlmsg_notify(struct sock *sk, struct sk_buff *skb, u32 portid,
3046 unsigned int group, int report, gfp_t flags)
3047 {
3048 int err = 0;
3049
3050 if (group) {
3051 int exclude_portid = 0;
3052
3053 if (report) {
3054 atomic_inc(&skb->users);
3055 exclude_portid = portid;
3056 }
3057
3058 /* errors reported via destination sk->sk_err, but propagate
3059 * delivery errors if NETLINK_BROADCAST_ERROR flag is set */
3060 err = nlmsg_multicast(sk, skb, exclude_portid, group, flags);
3061 }
3062
3063 if (report) {
3064 int err2;
3065
3066 err2 = nlmsg_unicast(sk, skb, portid);
3067 if (!err || err == -ESRCH)
3068 err = err2;
3069 }
3070
3071 return err;
3072 }
3073 EXPORT_SYMBOL(nlmsg_notify);
3074
3075 #ifdef CONFIG_PROC_FS
3076 struct nl_seq_iter {
3077 struct seq_net_private p;
3078 struct rhashtable_iter hti;
3079 int link;
3080 };
3081
3082 static int netlink_walk_start(struct nl_seq_iter *iter)
3083 {
3084 int err;
3085
3086 err = rhashtable_walk_init(&nl_table[iter->link].hash, &iter->hti);
3087 if (err) {
3088 iter->link = MAX_LINKS;
3089 return err;
3090 }
3091
3092 err = rhashtable_walk_start(&iter->hti);
3093 return err == -EAGAIN ? 0 : err;
3094 }
3095
3096 static void netlink_walk_stop(struct nl_seq_iter *iter)
3097 {
3098 rhashtable_walk_stop(&iter->hti);
3099 rhashtable_walk_exit(&iter->hti);
3100 }
3101
3102 static void *__netlink_seq_next(struct seq_file *seq)
3103 {
3104 struct nl_seq_iter *iter = seq->private;
3105 struct netlink_sock *nlk;
3106
3107 do {
3108 for (;;) {
3109 int err;
3110
3111 nlk = rhashtable_walk_next(&iter->hti);
3112
3113 if (IS_ERR(nlk)) {
3114 if (PTR_ERR(nlk) == -EAGAIN)
3115 continue;
3116
3117 return nlk;
3118 }
3119
3120 if (nlk)
3121 break;
3122
3123 netlink_walk_stop(iter);
3124 if (++iter->link >= MAX_LINKS)
3125 return NULL;
3126
3127 err = netlink_walk_start(iter);
3128 if (err)
3129 return ERR_PTR(err);
3130 }
3131 } while (sock_net(&nlk->sk) != seq_file_net(seq));
3132
3133 return nlk;
3134 }
3135
3136 static void *netlink_seq_start(struct seq_file *seq, loff_t *posp)
3137 {
3138 struct nl_seq_iter *iter = seq->private;
3139 void *obj = SEQ_START_TOKEN;
3140 loff_t pos;
3141 int err;
3142
3143 iter->link = 0;
3144
3145 err = netlink_walk_start(iter);
3146 if (err)
3147 return ERR_PTR(err);
3148
3149 for (pos = *posp; pos && obj && !IS_ERR(obj); pos--)
3150 obj = __netlink_seq_next(seq);
3151
3152 return obj;
3153 }
3154
3155 static void *netlink_seq_next(struct seq_file *seq, void *v, loff_t *pos)
3156 {
3157 ++*pos;
3158 return __netlink_seq_next(seq);
3159 }
3160
3161 static void netlink_seq_stop(struct seq_file *seq, void *v)
3162 {
3163 struct nl_seq_iter *iter = seq->private;
3164
3165 if (iter->link >= MAX_LINKS)
3166 return;
3167
3168 netlink_walk_stop(iter);
3169 }
3170
3171
3172 static int netlink_seq_show(struct seq_file *seq, void *v)
3173 {
3174 if (v == SEQ_START_TOKEN) {
3175 seq_puts(seq,
3176 "sk Eth Pid Groups "
3177 "Rmem Wmem Dump Locks Drops Inode\n");
3178 } else {
3179 struct sock *s = v;
3180 struct netlink_sock *nlk = nlk_sk(s);
3181
3182 seq_printf(seq, "%pK %-3d %-6u %08x %-8d %-8d %d %-8d %-8d %-8lu\n",
3183 s,
3184 s->sk_protocol,
3185 nlk->portid,
3186 nlk->groups ? (u32)nlk->groups[0] : 0,
3187 sk_rmem_alloc_get(s),
3188 sk_wmem_alloc_get(s),
3189 nlk->cb_running,
3190 atomic_read(&s->sk_refcnt),
3191 atomic_read(&s->sk_drops),
3192 sock_i_ino(s)
3193 );
3194
3195 }
3196 return 0;
3197 }
3198
3199 static const struct seq_operations netlink_seq_ops = {
3200 .start = netlink_seq_start,
3201 .next = netlink_seq_next,
3202 .stop = netlink_seq_stop,
3203 .show = netlink_seq_show,
3204 };
3205
3206
3207 static int netlink_seq_open(struct inode *inode, struct file *file)
3208 {
3209 return seq_open_net(inode, file, &netlink_seq_ops,
3210 sizeof(struct nl_seq_iter));
3211 }
3212
3213 static const struct file_operations netlink_seq_fops = {
3214 .owner = THIS_MODULE,
3215 .open = netlink_seq_open,
3216 .read = seq_read,
3217 .llseek = seq_lseek,
3218 .release = seq_release_net,
3219 };
3220
3221 #endif
3222
3223 int netlink_register_notifier(struct notifier_block *nb)
3224 {
3225 return atomic_notifier_chain_register(&netlink_chain, nb);
3226 }
3227 EXPORT_SYMBOL(netlink_register_notifier);
3228
3229 int netlink_unregister_notifier(struct notifier_block *nb)
3230 {
3231 return atomic_notifier_chain_unregister(&netlink_chain, nb);
3232 }
3233 EXPORT_SYMBOL(netlink_unregister_notifier);
3234
3235 static const struct proto_ops netlink_ops = {
3236 .family = PF_NETLINK,
3237 .owner = THIS_MODULE,
3238 .release = netlink_release,
3239 .bind = netlink_bind,
3240 .connect = netlink_connect,
3241 .socketpair = sock_no_socketpair,
3242 .accept = sock_no_accept,
3243 .getname = netlink_getname,
3244 .poll = netlink_poll,
3245 .ioctl = sock_no_ioctl,
3246 .listen = sock_no_listen,
3247 .shutdown = sock_no_shutdown,
3248 .setsockopt = netlink_setsockopt,
3249 .getsockopt = netlink_getsockopt,
3250 .sendmsg = netlink_sendmsg,
3251 .recvmsg = netlink_recvmsg,
3252 .mmap = netlink_mmap,
3253 .sendpage = sock_no_sendpage,
3254 };
3255
3256 static const struct net_proto_family netlink_family_ops = {
3257 .family = PF_NETLINK,
3258 .create = netlink_create,
3259 .owner = THIS_MODULE, /* for consistency 8) */
3260 };
3261
3262 static int __net_init netlink_net_init(struct net *net)
3263 {
3264 #ifdef CONFIG_PROC_FS
3265 if (!proc_create("netlink", 0, net->proc_net, &netlink_seq_fops))
3266 return -ENOMEM;
3267 #endif
3268 return 0;
3269 }
3270
3271 static void __net_exit netlink_net_exit(struct net *net)
3272 {
3273 #ifdef CONFIG_PROC_FS
3274 remove_proc_entry("netlink", net->proc_net);
3275 #endif
3276 }
3277
3278 static void __init netlink_add_usersock_entry(void)
3279 {
3280 struct listeners *listeners;
3281 int groups = 32;
3282
3283 listeners = kzalloc(sizeof(*listeners) + NLGRPSZ(groups), GFP_KERNEL);
3284 if (!listeners)
3285 panic("netlink_add_usersock_entry: Cannot allocate listeners\n");
3286
3287 netlink_table_grab();
3288
3289 nl_table[NETLINK_USERSOCK].groups = groups;
3290 rcu_assign_pointer(nl_table[NETLINK_USERSOCK].listeners, listeners);
3291 nl_table[NETLINK_USERSOCK].module = THIS_MODULE;
3292 nl_table[NETLINK_USERSOCK].registered = 1;
3293 nl_table[NETLINK_USERSOCK].flags = NL_CFG_F_NONROOT_SEND;
3294
3295 netlink_table_ungrab();
3296 }
3297
3298 static struct pernet_operations __net_initdata netlink_net_ops = {
3299 .init = netlink_net_init,
3300 .exit = netlink_net_exit,
3301 };
3302
3303 static inline u32 netlink_hash(const void *data, u32 len, u32 seed)
3304 {
3305 const struct netlink_sock *nlk = data;
3306 struct netlink_compare_arg arg;
3307
3308 netlink_compare_arg_init(&arg, sock_net(&nlk->sk), nlk->portid);
3309 return jhash2((u32 *)&arg, netlink_compare_arg_len / sizeof(u32), seed);
3310 }
3311
3312 static const struct rhashtable_params netlink_rhashtable_params = {
3313 .head_offset = offsetof(struct netlink_sock, node),
3314 .key_len = netlink_compare_arg_len,
3315 .obj_hashfn = netlink_hash,
3316 .obj_cmpfn = netlink_compare,
3317 .automatic_shrinking = true,
3318 };
3319
3320 static int __init netlink_proto_init(void)
3321 {
3322 int i;
3323 int err = proto_register(&netlink_proto, 0);
3324
3325 if (err != 0)
3326 goto out;
3327
3328 BUILD_BUG_ON(sizeof(struct netlink_skb_parms) > FIELD_SIZEOF(struct sk_buff, cb));
3329
3330 nl_table = kcalloc(MAX_LINKS, sizeof(*nl_table), GFP_KERNEL);
3331 if (!nl_table)
3332 goto panic;
3333
3334 for (i = 0; i < MAX_LINKS; i++) {
3335 if (rhashtable_init(&nl_table[i].hash,
3336 &netlink_rhashtable_params) < 0) {
3337 while (--i > 0)
3338 rhashtable_destroy(&nl_table[i].hash);
3339 kfree(nl_table);
3340 goto panic;
3341 }
3342 }
3343
3344 INIT_LIST_HEAD(&netlink_tap_all);
3345
3346 netlink_add_usersock_entry();
3347
3348 sock_register(&netlink_family_ops);
3349 register_pernet_subsys(&netlink_net_ops);
3350 /* The netlink device handler may be needed early. */
3351 rtnetlink_init();
3352 out:
3353 return err;
3354 panic:
3355 panic("netlink_init: Cannot allocate nl_table\n");
3356 }
3357
3358 core_initcall(netlink_proto_init);