?  search:
summary | log | graphiclog1 | graphiclog2 | commit | commitdiff | tree | refs | edit | fork
blame | history | raw | HEAD
KVM: Introduce gfn_to_hva_memslot_prot
[linux/fpc-iii.git] / net / netlink / af_netlink.c
blobc416725d28c49f8b0c1b10bbf35a28594c646bc5
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
65 #include <net/net_namespace.h>
66 #include <net/sock.h>
67 #include <net/scm.h>
68 #include <net/netlink.h>
69
70 #include "af_netlink.h"
71
72 struct listeners {
73 struct rcu_head rcu;
74 unsigned long masks[0];
75 };
76
77 /* state bits */
78 #define NETLINK_CONGESTED 0x0
79
80 /* flags */
81 #define NETLINK_KERNEL_SOCKET 0x1
82 #define NETLINK_RECV_PKTINFO 0x2
83 #define NETLINK_BROADCAST_SEND_ERROR 0x4
84 #define NETLINK_RECV_NO_ENOBUFS 0x8
85
86 static inline int netlink_is_kernel(struct sock *sk)
87 {
88 return nlk_sk(sk)->flags & NETLINK_KERNEL_SOCKET;
89 }
90
91 struct netlink_table *nl_table;
92 EXPORT_SYMBOL_GPL(nl_table);
93
94 static DECLARE_WAIT_QUEUE_HEAD(nl_table_wait);
95
96 static int netlink_dump(struct sock *sk);
97 static void netlink_skb_destructor(struct sk_buff *skb);
98
99 DEFINE_RWLOCK(nl_table_lock);
100 EXPORT_SYMBOL_GPL(nl_table_lock);
101 static atomic_t nl_table_users = ATOMIC_INIT(0);
102
103 #define nl_deref_protected(X) rcu_dereference_protected(X, lockdep_is_held(&nl_table_lock));
104
105 /* Protects netlink socket hash table mutations */
106 DEFINE_MUTEX(nl_sk_hash_lock);
107 EXPORT_SYMBOL_GPL(nl_sk_hash_lock);
108
109 static int lockdep_nl_sk_hash_is_held(void)
110 {
111 #ifdef CONFIG_LOCKDEP
112 return (debug_locks) ? lockdep_is_held(&nl_sk_hash_lock) : 1;
113 #else
114 return 1;
115 #endif
116 }
117
118 static ATOMIC_NOTIFIER_HEAD(netlink_chain);
119
120 static DEFINE_SPINLOCK(netlink_tap_lock);
121 static struct list_head netlink_tap_all __read_mostly;
122
123 static inline u32 netlink_group_mask(u32 group)
124 {
125 return group ? 1 << (group - 1) : 0;
126 }
127
128 int netlink_add_tap(struct netlink_tap *nt)
129 {
130 if (unlikely(nt->dev->type != ARPHRD_NETLINK))
131 return -EINVAL;
132
133 spin_lock(&netlink_tap_lock);
134 list_add_rcu(&nt->list, &netlink_tap_all);
135 spin_unlock(&netlink_tap_lock);
136
137 if (nt->module)
138 __module_get(nt->module);
139
140 return 0;
141 }
142 EXPORT_SYMBOL_GPL(netlink_add_tap);
143
144 static int __netlink_remove_tap(struct netlink_tap *nt)
145 {
146 bool found = false;
147 struct netlink_tap *tmp;
148
149 spin_lock(&netlink_tap_lock);
150
151 list_for_each_entry(tmp, &netlink_tap_all, list) {
152 if (nt == tmp) {
153 list_del_rcu(&nt->list);
154 found = true;
155 goto out;
156 }
157 }
158
159 pr_warn("__netlink_remove_tap: %p not found\n", nt);
160 out:
161 spin_unlock(&netlink_tap_lock);
162
163 if (found && nt->module)
164 module_put(nt->module);
165
166 return found ? 0 : -ENODEV;
167 }
168
169 int netlink_remove_tap(struct netlink_tap *nt)
170 {
171 int ret;
172
173 ret = __netlink_remove_tap(nt);
174 synchronize_net();
175
176 return ret;
177 }
178 EXPORT_SYMBOL_GPL(netlink_remove_tap);
179
180 static bool netlink_filter_tap(const struct sk_buff *skb)
181 {
182 struct sock *sk = skb->sk;
183
184 /* We take the more conservative approach and
185 * whitelist socket protocols that may pass.
186 */
187 switch (sk->sk_protocol) {
188 case NETLINK_ROUTE:
189 case NETLINK_USERSOCK:
190 case NETLINK_SOCK_DIAG:
191 case NETLINK_NFLOG:
192 case NETLINK_XFRM:
193 case NETLINK_FIB_LOOKUP:
194 case NETLINK_NETFILTER:
195 case NETLINK_GENERIC:
196 return true;
197 }
198
199 return false;
200 }
201
202 static int __netlink_deliver_tap_skb(struct sk_buff *skb,
203 struct net_device *dev)
204 {
205 struct sk_buff *nskb;
206 struct sock *sk = skb->sk;
207 int ret = -ENOMEM;
208
209 dev_hold(dev);
210 nskb = skb_clone(skb, GFP_ATOMIC);
211 if (nskb) {
212 nskb->dev = dev;
213 nskb->protocol = htons((u16) sk->sk_protocol);
214 nskb->pkt_type = netlink_is_kernel(sk) ?
215 PACKET_KERNEL : PACKET_USER;
216 skb_reset_network_header(nskb);
217 ret = dev_queue_xmit(nskb);
218 if (unlikely(ret > 0))
219 ret = net_xmit_errno(ret);
220 }
221
222 dev_put(dev);
223 return ret;
224 }
225
226 static void __netlink_deliver_tap(struct sk_buff *skb)
227 {
228 int ret;
229 struct netlink_tap *tmp;
230
231 if (!netlink_filter_tap(skb))
232 return;
233
234 list_for_each_entry_rcu(tmp, &netlink_tap_all, list) {
235 ret = __netlink_deliver_tap_skb(skb, tmp->dev);
236 if (unlikely(ret))
237 break;
238 }
239 }
240
241 static void netlink_deliver_tap(struct sk_buff *skb)
242 {
243 rcu_read_lock();
244
245 if (unlikely(!list_empty(&netlink_tap_all)))
246 __netlink_deliver_tap(skb);
247
248 rcu_read_unlock();
249 }
250
251 static void netlink_deliver_tap_kernel(struct sock *dst, struct sock *src,
252 struct sk_buff *skb)
253 {
254 if (!(netlink_is_kernel(dst) && netlink_is_kernel(src)))
255 netlink_deliver_tap(skb);
256 }
257
258 static void netlink_overrun(struct sock *sk)
259 {
260 struct netlink_sock *nlk = nlk_sk(sk);
261
262 if (!(nlk->flags & NETLINK_RECV_NO_ENOBUFS)) {
263 if (!test_and_set_bit(NETLINK_CONGESTED, &nlk_sk(sk)->state)) {
264 sk->sk_err = ENOBUFS;
265 sk->sk_error_report(sk);
266 }
267 }
268 atomic_inc(&sk->sk_drops);
269 }
270
271 static void netlink_rcv_wake(struct sock *sk)
272 {
273 struct netlink_sock *nlk = nlk_sk(sk);
274
275 if (skb_queue_empty(&sk->sk_receive_queue))
276 clear_bit(NETLINK_CONGESTED, &nlk->state);
277 if (!test_bit(NETLINK_CONGESTED, &nlk->state))
278 wake_up_interruptible(&nlk->wait);
279 }
280
281 #ifdef CONFIG_NETLINK_MMAP
282 static bool netlink_skb_is_mmaped(const struct sk_buff *skb)
283 {
284 return NETLINK_CB(skb).flags & NETLINK_SKB_MMAPED;
285 }
286
287 static bool netlink_rx_is_mmaped(struct sock *sk)
288 {
289 return nlk_sk(sk)->rx_ring.pg_vec != NULL;
290 }
291
292 static bool netlink_tx_is_mmaped(struct sock *sk)
293 {
294 return nlk_sk(sk)->tx_ring.pg_vec != NULL;
295 }
296
297 static __pure struct page *pgvec_to_page(const void *addr)
298 {
299 if (is_vmalloc_addr(addr))
300 return vmalloc_to_page(addr);
301 else
302 return virt_to_page(addr);
303 }
304
305 static void free_pg_vec(void **pg_vec, unsigned int order, unsigned int len)
306 {
307 unsigned int i;
308
309 for (i = 0; i < len; i++) {
310 if (pg_vec[i] != NULL) {
311 if (is_vmalloc_addr(pg_vec[i]))
312 vfree(pg_vec[i]);
313 else
314 free_pages((unsigned long)pg_vec[i], order);
315 }
316 }
317 kfree(pg_vec);
318 }
319
320 static void *alloc_one_pg_vec_page(unsigned long order)
321 {
322 void *buffer;
323 gfp_t gfp_flags = GFP_KERNEL | __GFP_COMP | __GFP_ZERO |
324 __GFP_NOWARN | __GFP_NORETRY;
325
326 buffer = (void *)__get_free_pages(gfp_flags, order);
327 if (buffer != NULL)
328 return buffer;
329
330 buffer = vzalloc((1 << order) * PAGE_SIZE);
331 if (buffer != NULL)
332 return buffer;
333
334 gfp_flags &= ~__GFP_NORETRY;
335 return (void *)__get_free_pages(gfp_flags, order);
336 }
337
338 static void **alloc_pg_vec(struct netlink_sock *nlk,
339 struct nl_mmap_req *req, unsigned int order)
340 {
341 unsigned int block_nr = req->nm_block_nr;
342 unsigned int i;
343 void **pg_vec;
344
345 pg_vec = kcalloc(block_nr, sizeof(void *), GFP_KERNEL);
346 if (pg_vec == NULL)
347 return NULL;
348
349 for (i = 0; i < block_nr; i++) {
350 pg_vec[i] = alloc_one_pg_vec_page(order);
351 if (pg_vec[i] == NULL)
352 goto err1;
353 }
354
355 return pg_vec;
356 err1:
357 free_pg_vec(pg_vec, order, block_nr);
358 return NULL;
359 }
360
361 static int netlink_set_ring(struct sock *sk, struct nl_mmap_req *req,
362 bool closing, bool tx_ring)
363 {
364 struct netlink_sock *nlk = nlk_sk(sk);
365 struct netlink_ring *ring;
366 struct sk_buff_head *queue;
367 void **pg_vec = NULL;
368 unsigned int order = 0;
369 int err;
370
371 ring = tx_ring ? &nlk->tx_ring : &nlk->rx_ring;
372 queue = tx_ring ? &sk->sk_write_queue : &sk->sk_receive_queue;
373
374 if (!closing) {
375 if (atomic_read(&nlk->mapped))
376 return -EBUSY;
377 if (atomic_read(&ring->pending))
378 return -EBUSY;
379 }
380
381 if (req->nm_block_nr) {
382 if (ring->pg_vec != NULL)
383 return -EBUSY;
384
385 if ((int)req->nm_block_size <= 0)
386 return -EINVAL;
387 if (!PAGE_ALIGNED(req->nm_block_size))
388 return -EINVAL;
389 if (req->nm_frame_size < NL_MMAP_HDRLEN)
390 return -EINVAL;
391 if (!IS_ALIGNED(req->nm_frame_size, NL_MMAP_MSG_ALIGNMENT))
392 return -EINVAL;
393
394 ring->frames_per_block = req->nm_block_size /
395 req->nm_frame_size;
396 if (ring->frames_per_block == 0)
397 return -EINVAL;
398 if (ring->frames_per_block * req->nm_block_nr !=
399 req->nm_frame_nr)
400 return -EINVAL;
401
402 order = get_order(req->nm_block_size);
403 pg_vec = alloc_pg_vec(nlk, req, order);
404 if (pg_vec == NULL)
405 return -ENOMEM;
406 } else {
407 if (req->nm_frame_nr)
408 return -EINVAL;
409 }
410
411 err = -EBUSY;
412 mutex_lock(&nlk->pg_vec_lock);
413 if (closing || atomic_read(&nlk->mapped) == 0) {
414 err = 0;
415 spin_lock_bh(&queue->lock);
416
417 ring->frame_max = req->nm_frame_nr - 1;
418 ring->head = 0;
419 ring->frame_size = req->nm_frame_size;
420 ring->pg_vec_pages = req->nm_block_size / PAGE_SIZE;
421
422 swap(ring->pg_vec_len, req->nm_block_nr);
423 swap(ring->pg_vec_order, order);
424 swap(ring->pg_vec, pg_vec);
425
426 __skb_queue_purge(queue);
427 spin_unlock_bh(&queue->lock);
428
429 WARN_ON(atomic_read(&nlk->mapped));
430 }
431 mutex_unlock(&nlk->pg_vec_lock);
432
433 if (pg_vec)
434 free_pg_vec(pg_vec, order, req->nm_block_nr);
435 return err;
436 }
437
438 static void netlink_mm_open(struct vm_area_struct *vma)
439 {
440 struct file *file = vma->vm_file;
441 struct socket *sock = file->private_data;
442 struct sock *sk = sock->sk;
443
444 if (sk)
445 atomic_inc(&nlk_sk(sk)->mapped);
446 }
447
448 static void netlink_mm_close(struct vm_area_struct *vma)
449 {
450 struct file *file = vma->vm_file;
451 struct socket *sock = file->private_data;
452 struct sock *sk = sock->sk;
453
454 if (sk)
455 atomic_dec(&nlk_sk(sk)->mapped);
456 }
457
458 static const struct vm_operations_struct netlink_mmap_ops = {
459 .open = netlink_mm_open,
460 .close = netlink_mm_close,
461 };
462
463 static int netlink_mmap(struct file *file, struct socket *sock,
464 struct vm_area_struct *vma)
465 {
466 struct sock *sk = sock->sk;
467 struct netlink_sock *nlk = nlk_sk(sk);
468 struct netlink_ring *ring;
469 unsigned long start, size, expected;
470 unsigned int i;
471 int err = -EINVAL;
472
473 if (vma->vm_pgoff)
474 return -EINVAL;
475
476 mutex_lock(&nlk->pg_vec_lock);
477
478 expected = 0;
479 for (ring = &nlk->rx_ring; ring <= &nlk->tx_ring; ring++) {
480 if (ring->pg_vec == NULL)
481 continue;
482 expected += ring->pg_vec_len * ring->pg_vec_pages * PAGE_SIZE;
483 }
484
485 if (expected == 0)
486 goto out;
487
488 size = vma->vm_end - vma->vm_start;
489 if (size != expected)
490 goto out;
491
492 start = vma->vm_start;
493 for (ring = &nlk->rx_ring; ring <= &nlk->tx_ring; ring++) {
494 if (ring->pg_vec == NULL)
495 continue;
496
497 for (i = 0; i < ring->pg_vec_len; i++) {
498 struct page *page;
499 void *kaddr = ring->pg_vec[i];
500 unsigned int pg_num;
501
502 for (pg_num = 0; pg_num < ring->pg_vec_pages; pg_num++) {
503 page = pgvec_to_page(kaddr);
504 err = vm_insert_page(vma, start, page);
505 if (err < 0)
506 goto out;
507 start += PAGE_SIZE;
508 kaddr += PAGE_SIZE;
509 }
510 }
511 }
512
513 atomic_inc(&nlk->mapped);
514 vma->vm_ops = &netlink_mmap_ops;
515 err = 0;
516 out:
517 mutex_unlock(&nlk->pg_vec_lock);
518 return err;
519 }
520
521 static void netlink_frame_flush_dcache(const struct nl_mmap_hdr *hdr)
522 {
523 #if ARCH_IMPLEMENTS_FLUSH_DCACHE_PAGE == 1
524 struct page *p_start, *p_end;
525
526 /* First page is flushed through netlink_{get,set}_status */
527 p_start = pgvec_to_page(hdr + PAGE_SIZE);
528 p_end = pgvec_to_page((void *)hdr + NL_MMAP_HDRLEN + hdr->nm_len - 1);
529 while (p_start <= p_end) {
530 flush_dcache_page(p_start);
531 p_start++;
532 }
533 #endif
534 }
535
536 static enum nl_mmap_status netlink_get_status(const struct nl_mmap_hdr *hdr)
537 {
538 smp_rmb();
539 flush_dcache_page(pgvec_to_page(hdr));
540 return hdr->nm_status;
541 }
542
543 static void netlink_set_status(struct nl_mmap_hdr *hdr,
544 enum nl_mmap_status status)
545 {
546 hdr->nm_status = status;
547 flush_dcache_page(pgvec_to_page(hdr));
548 smp_wmb();
549 }
550
551 static struct nl_mmap_hdr *
552 __netlink_lookup_frame(const struct netlink_ring *ring, unsigned int pos)
553 {
554 unsigned int pg_vec_pos, frame_off;
555
556 pg_vec_pos = pos / ring->frames_per_block;
557 frame_off = pos % ring->frames_per_block;
558
559 return ring->pg_vec[pg_vec_pos] + (frame_off * ring->frame_size);
560 }
561
562 static struct nl_mmap_hdr *
563 netlink_lookup_frame(const struct netlink_ring *ring, unsigned int pos,
564 enum nl_mmap_status status)
565 {
566 struct nl_mmap_hdr *hdr;
567
568 hdr = __netlink_lookup_frame(ring, pos);
569 if (netlink_get_status(hdr) != status)
570 return NULL;
571
572 return hdr;
573 }
574
575 static struct nl_mmap_hdr *
576 netlink_current_frame(const struct netlink_ring *ring,
577 enum nl_mmap_status status)
578 {
579 return netlink_lookup_frame(ring, ring->head, status);
580 }
581
582 static struct nl_mmap_hdr *
583 netlink_previous_frame(const struct netlink_ring *ring,
584 enum nl_mmap_status status)
585 {
586 unsigned int prev;
587
588 prev = ring->head ? ring->head - 1 : ring->frame_max;
589 return netlink_lookup_frame(ring, prev, status);
590 }
591
592 static void netlink_increment_head(struct netlink_ring *ring)
593 {
594 ring->head = ring->head != ring->frame_max ? ring->head + 1 : 0;
595 }
596
597 static void netlink_forward_ring(struct netlink_ring *ring)
598 {
599 unsigned int head = ring->head, pos = head;
600 const struct nl_mmap_hdr *hdr;
601
602 do {
603 hdr = __netlink_lookup_frame(ring, pos);
604 if (hdr->nm_status == NL_MMAP_STATUS_UNUSED)
605 break;
606 if (hdr->nm_status != NL_MMAP_STATUS_SKIP)
607 break;
608 netlink_increment_head(ring);
609 } while (ring->head != head);
610 }
611
612 static bool netlink_dump_space(struct netlink_sock *nlk)
613 {
614 struct netlink_ring *ring = &nlk->rx_ring;
615 struct nl_mmap_hdr *hdr;
616 unsigned int n;
617
618 hdr = netlink_current_frame(ring, NL_MMAP_STATUS_UNUSED);
619 if (hdr == NULL)
620 return false;
621
622 n = ring->head + ring->frame_max / 2;
623 if (n > ring->frame_max)
624 n -= ring->frame_max;
625
626 hdr = __netlink_lookup_frame(ring, n);
627
628 return hdr->nm_status == NL_MMAP_STATUS_UNUSED;
629 }
630
631 static unsigned int netlink_poll(struct file *file, struct socket *sock,
632 poll_table *wait)
633 {
634 struct sock *sk = sock->sk;
635 struct netlink_sock *nlk = nlk_sk(sk);
636 unsigned int mask;
637 int err;
638
639 if (nlk->rx_ring.pg_vec != NULL) {
640 /* Memory mapped sockets don't call recvmsg(), so flow control
641 * for dumps is performed here. A dump is allowed to continue
642 * if at least half the ring is unused.
643 */
644 while (nlk->cb_running && netlink_dump_space(nlk)) {
645 err = netlink_dump(sk);
646 if (err < 0) {
647 sk->sk_err = -err;
648 sk->sk_error_report(sk);
649 break;
650 }
651 }
652 netlink_rcv_wake(sk);
653 }
654
655 mask = datagram_poll(file, sock, wait);
656
657 spin_lock_bh(&sk->sk_receive_queue.lock);
658 if (nlk->rx_ring.pg_vec) {
659 netlink_forward_ring(&nlk->rx_ring);
660 if (!netlink_previous_frame(&nlk->rx_ring, NL_MMAP_STATUS_UNUSED))
661 mask |= POLLIN | POLLRDNORM;
662 }
663 spin_unlock_bh(&sk->sk_receive_queue.lock);
664
665 spin_lock_bh(&sk->sk_write_queue.lock);
666 if (nlk->tx_ring.pg_vec) {
667 if (netlink_current_frame(&nlk->tx_ring, NL_MMAP_STATUS_UNUSED))
668 mask |= POLLOUT | POLLWRNORM;
669 }
670 spin_unlock_bh(&sk->sk_write_queue.lock);
671
672 return mask;
673 }
674
675 static struct nl_mmap_hdr *netlink_mmap_hdr(struct sk_buff *skb)
676 {
677 return (struct nl_mmap_hdr *)(skb->head - NL_MMAP_HDRLEN);
678 }
679
680 static void netlink_ring_setup_skb(struct sk_buff *skb, struct sock *sk,
681 struct netlink_ring *ring,
682 struct nl_mmap_hdr *hdr)
683 {
684 unsigned int size;
685 void *data;
686
687 size = ring->frame_size - NL_MMAP_HDRLEN;
688 data = (void *)hdr + NL_MMAP_HDRLEN;
689
690 skb->head = data;
691 skb->data = data;
692 skb_reset_tail_pointer(skb);
693 skb->end = skb->tail + size;
694 skb->len = 0;
695
696 skb->destructor = netlink_skb_destructor;
697 NETLINK_CB(skb).flags |= NETLINK_SKB_MMAPED;
698 NETLINK_CB(skb).sk = sk;
699 }
700
701 static int netlink_mmap_sendmsg(struct sock *sk, struct msghdr *msg,
702 u32 dst_portid, u32 dst_group,
703 struct sock_iocb *siocb)
704 {
705 struct netlink_sock *nlk = nlk_sk(sk);
706 struct netlink_ring *ring;
707 struct nl_mmap_hdr *hdr;
708 struct sk_buff *skb;
709 unsigned int maxlen;
710 bool excl = true;
711 int err = 0, len = 0;
712
713 /* Netlink messages are validated by the receiver before processing.
714 * In order to avoid userspace changing the contents of the message
715 * after validation, the socket and the ring may only be used by a
716 * single process, otherwise we fall back to copying.
717 */
718 if (atomic_long_read(&sk->sk_socket->file->f_count) > 2 ||
719 atomic_read(&nlk->mapped) > 1)
720 excl = false;
721
722 mutex_lock(&nlk->pg_vec_lock);
723
724 ring = &nlk->tx_ring;
725 maxlen = ring->frame_size - NL_MMAP_HDRLEN;
726
727 do {
728 hdr = netlink_current_frame(ring, NL_MMAP_STATUS_VALID);
729 if (hdr == NULL) {
730 if (!(msg->msg_flags & MSG_DONTWAIT) &&
731 atomic_read(&nlk->tx_ring.pending))
732 schedule();
733 continue;
734 }
735 if (hdr->nm_len > maxlen) {
736 err = -EINVAL;
737 goto out;
738 }
739
740 netlink_frame_flush_dcache(hdr);
741
742 if (likely(dst_portid == 0 && dst_group == 0 && excl)) {
743 skb = alloc_skb_head(GFP_KERNEL);
744 if (skb == NULL) {
745 err = -ENOBUFS;
746 goto out;
747 }
748 sock_hold(sk);
749 netlink_ring_setup_skb(skb, sk, ring, hdr);
750 NETLINK_CB(skb).flags |= NETLINK_SKB_TX;
751 __skb_put(skb, hdr->nm_len);
752 netlink_set_status(hdr, NL_MMAP_STATUS_RESERVED);
753 atomic_inc(&ring->pending);
754 } else {
755 skb = alloc_skb(hdr->nm_len, GFP_KERNEL);
756 if (skb == NULL) {
757 err = -ENOBUFS;
758 goto out;
759 }
760 __skb_put(skb, hdr->nm_len);
761 memcpy(skb->data, (void *)hdr + NL_MMAP_HDRLEN, hdr->nm_len);
762 netlink_set_status(hdr, NL_MMAP_STATUS_UNUSED);
763 }
764
765 netlink_increment_head(ring);
766
767 NETLINK_CB(skb).portid = nlk->portid;
768 NETLINK_CB(skb).dst_group = dst_group;
769 NETLINK_CB(skb).creds = siocb->scm->creds;
770
771 err = security_netlink_send(sk, skb);
772 if (err) {
773 kfree_skb(skb);
774 goto out;
775 }
776
777 if (unlikely(dst_group)) {
778 atomic_inc(&skb->users);
779 netlink_broadcast(sk, skb, dst_portid, dst_group,
780 GFP_KERNEL);
781 }
782 err = netlink_unicast(sk, skb, dst_portid,
783 msg->msg_flags & MSG_DONTWAIT);
784 if (err < 0)
785 goto out;
786 len += err;
787
788 } while (hdr != NULL ||
789 (!(msg->msg_flags & MSG_DONTWAIT) &&
790 atomic_read(&nlk->tx_ring.pending)));
791
792 if (len > 0)
793 err = len;
794 out:
795 mutex_unlock(&nlk->pg_vec_lock);
796 return err;
797 }
798
799 static void netlink_queue_mmaped_skb(struct sock *sk, struct sk_buff *skb)
800 {
801 struct nl_mmap_hdr *hdr;
802
803 hdr = netlink_mmap_hdr(skb);
804 hdr->nm_len = skb->len;
805 hdr->nm_group = NETLINK_CB(skb).dst_group;
806 hdr->nm_pid = NETLINK_CB(skb).creds.pid;
807 hdr->nm_uid = from_kuid(sk_user_ns(sk), NETLINK_CB(skb).creds.uid);
808 hdr->nm_gid = from_kgid(sk_user_ns(sk), NETLINK_CB(skb).creds.gid);
809 netlink_frame_flush_dcache(hdr);
810 netlink_set_status(hdr, NL_MMAP_STATUS_VALID);
811
812 NETLINK_CB(skb).flags |= NETLINK_SKB_DELIVERED;
813 kfree_skb(skb);
814 }
815
816 static void netlink_ring_set_copied(struct sock *sk, struct sk_buff *skb)
817 {
818 struct netlink_sock *nlk = nlk_sk(sk);
819 struct netlink_ring *ring = &nlk->rx_ring;
820 struct nl_mmap_hdr *hdr;
821
822 spin_lock_bh(&sk->sk_receive_queue.lock);
823 hdr = netlink_current_frame(ring, NL_MMAP_STATUS_UNUSED);
824 if (hdr == NULL) {
825 spin_unlock_bh(&sk->sk_receive_queue.lock);
826 kfree_skb(skb);
827 netlink_overrun(sk);
828 return;
829 }
830 netlink_increment_head(ring);
831 __skb_queue_tail(&sk->sk_receive_queue, skb);
832 spin_unlock_bh(&sk->sk_receive_queue.lock);
833
834 hdr->nm_len = skb->len;
835 hdr->nm_group = NETLINK_CB(skb).dst_group;
836 hdr->nm_pid = NETLINK_CB(skb).creds.pid;
837 hdr->nm_uid = from_kuid(sk_user_ns(sk), NETLINK_CB(skb).creds.uid);
838 hdr->nm_gid = from_kgid(sk_user_ns(sk), NETLINK_CB(skb).creds.gid);
839 netlink_set_status(hdr, NL_MMAP_STATUS_COPY);
840 }
841
842 #else /* CONFIG_NETLINK_MMAP */
843 #define netlink_skb_is_mmaped(skb) false
844 #define netlink_rx_is_mmaped(sk) false
845 #define netlink_tx_is_mmaped(sk) false
846 #define netlink_mmap sock_no_mmap
847 #define netlink_poll datagram_poll
848 #define netlink_mmap_sendmsg(sk, msg, dst_portid, dst_group, siocb) 0
849 #endif /* CONFIG_NETLINK_MMAP */
850
851 static void netlink_skb_destructor(struct sk_buff *skb)
852 {
853 #ifdef CONFIG_NETLINK_MMAP
854 struct nl_mmap_hdr *hdr;
855 struct netlink_ring *ring;
856 struct sock *sk;
857
858 /* If a packet from the kernel to userspace was freed because of an
859 * error without being delivered to userspace, the kernel must reset
860 * the status. In the direction userspace to kernel, the status is
861 * always reset here after the packet was processed and freed.
862 */
863 if (netlink_skb_is_mmaped(skb)) {
864 hdr = netlink_mmap_hdr(skb);
865 sk = NETLINK_CB(skb).sk;
866
867 if (NETLINK_CB(skb).flags & NETLINK_SKB_TX) {
868 netlink_set_status(hdr, NL_MMAP_STATUS_UNUSED);
869 ring = &nlk_sk(sk)->tx_ring;
870 } else {
871 if (!(NETLINK_CB(skb).flags & NETLINK_SKB_DELIVERED)) {
872 hdr->nm_len = 0;
873 netlink_set_status(hdr, NL_MMAP_STATUS_VALID);
874 }
875 ring = &nlk_sk(sk)->rx_ring;
876 }
877
878 WARN_ON(atomic_read(&ring->pending) == 0);
879 atomic_dec(&ring->pending);
880 sock_put(sk);
881
882 skb->head = NULL;
883 }
884 #endif
885 if (is_vmalloc_addr(skb->head)) {
886 if (!skb->cloned ||
887 !atomic_dec_return(&(skb_shinfo(skb)->dataref)))
888 vfree(skb->head);
889
890 skb->head = NULL;
891 }
892 if (skb->sk != NULL)
893 sock_rfree(skb);
894 }
895
896 static void netlink_skb_set_owner_r(struct sk_buff *skb, struct sock *sk)
897 {
898 WARN_ON(skb->sk != NULL);
899 skb->sk = sk;
900 skb->destructor = netlink_skb_destructor;
901 atomic_add(skb->truesize, &sk->sk_rmem_alloc);
902 sk_mem_charge(sk, skb->truesize);
903 }
904
905 static void netlink_sock_destruct(struct sock *sk)
906 {
907 struct netlink_sock *nlk = nlk_sk(sk);
908
909 if (nlk->cb_running) {
910 if (nlk->cb.done)
911 nlk->cb.done(&nlk->cb);
912
913 module_put(nlk->cb.module);
914 kfree_skb(nlk->cb.skb);
915 }
916
917 skb_queue_purge(&sk->sk_receive_queue);
918 #ifdef CONFIG_NETLINK_MMAP
919 if (1) {
920 struct nl_mmap_req req;
921
922 memset(&req, 0, sizeof(req));
923 if (nlk->rx_ring.pg_vec)
924 netlink_set_ring(sk, &req, true, false);
925 memset(&req, 0, sizeof(req));
926 if (nlk->tx_ring.pg_vec)
927 netlink_set_ring(sk, &req, true, true);
928 }
929 #endif /* CONFIG_NETLINK_MMAP */
930
931 if (!sock_flag(sk, SOCK_DEAD)) {
932 printk(KERN_ERR "Freeing alive netlink socket %p\n", sk);
933 return;
934 }
935
936 WARN_ON(atomic_read(&sk->sk_rmem_alloc));
937 WARN_ON(atomic_read(&sk->sk_wmem_alloc));
938 WARN_ON(nlk_sk(sk)->groups);
939 }
940
941 /* This lock without WQ_FLAG_EXCLUSIVE is good on UP and it is _very_ bad on
942 * SMP. Look, when several writers sleep and reader wakes them up, all but one
943 * immediately hit write lock and grab all the cpus. Exclusive sleep solves
944 * this, _but_ remember, it adds useless work on UP machines.
945 */
946
947 void netlink_table_grab(void)
948 __acquires(nl_table_lock)
949 {
950 might_sleep();
951
952 write_lock_irq(&nl_table_lock);
953
954 if (atomic_read(&nl_table_users)) {
955 DECLARE_WAITQUEUE(wait, current);
956
957 add_wait_queue_exclusive(&nl_table_wait, &wait);
958 for (;;) {
959 set_current_state(TASK_UNINTERRUPTIBLE);
960 if (atomic_read(&nl_table_users) == 0)
961 break;
962 write_unlock_irq(&nl_table_lock);
963 schedule();
964 write_lock_irq(&nl_table_lock);
965 }
966
967 __set_current_state(TASK_RUNNING);
968 remove_wait_queue(&nl_table_wait, &wait);
969 }
970 }
971
972 void netlink_table_ungrab(void)
973 __releases(nl_table_lock)
974 {
975 write_unlock_irq(&nl_table_lock);
976 wake_up(&nl_table_wait);
977 }
978
979 static inline void
980 netlink_lock_table(void)
981 {
982 /* read_lock() synchronizes us to netlink_table_grab */
983
984 read_lock(&nl_table_lock);
985 atomic_inc(&nl_table_users);
986 read_unlock(&nl_table_lock);
987 }
988
989 static inline void
990 netlink_unlock_table(void)
991 {
992 if (atomic_dec_and_test(&nl_table_users))
993 wake_up(&nl_table_wait);
994 }
995
996 struct netlink_compare_arg
997 {
998 struct net *net;
999 u32 portid;
1000 };
1001
1002 static bool netlink_compare(void *ptr, void *arg)
1003 {
1004 struct netlink_compare_arg *x = arg;
1005 struct sock *sk = ptr;
1006
1007 return nlk_sk(sk)->portid == x->portid &&
1008 net_eq(sock_net(sk), x->net);
1009 }
1010
1011 static struct sock *__netlink_lookup(struct netlink_table *table, u32 portid,
1012 struct net *net)
1013 {
1014 struct netlink_compare_arg arg = {
1015 .net = net,
1016 .portid = portid,
1017 };
1018 u32 hash;
1019
1020 hash = rhashtable_hashfn(&table->hash, &portid, sizeof(portid));
1021
1022 return rhashtable_lookup_compare(&table->hash, hash,
1023 &netlink_compare, &arg);
1024 }
1025
1026 static struct sock *netlink_lookup(struct net *net, int protocol, u32 portid)
1027 {
1028 struct netlink_table *table = &nl_table[protocol];
1029 struct sock *sk;
1030
1031 rcu_read_lock();
1032 sk = __netlink_lookup(table, portid, net);
1033 if (sk)
1034 sock_hold(sk);
1035 rcu_read_unlock();
1036
1037 return sk;
1038 }
1039
1040 static const struct proto_ops netlink_ops;
1041
1042 static void
1043 netlink_update_listeners(struct sock *sk)
1044 {
1045 struct netlink_table *tbl = &nl_table[sk->sk_protocol];
1046 unsigned long mask;
1047 unsigned int i;
1048 struct listeners *listeners;
1049
1050 listeners = nl_deref_protected(tbl->listeners);
1051 if (!listeners)
1052 return;
1053
1054 for (i = 0; i < NLGRPLONGS(tbl->groups); i++) {
1055 mask = 0;
1056 sk_for_each_bound(sk, &tbl->mc_list) {
1057 if (i < NLGRPLONGS(nlk_sk(sk)->ngroups))
1058 mask |= nlk_sk(sk)->groups[i];
1059 }
1060 listeners->masks[i] = mask;
1061 }
1062 /* this function is only called with the netlink table "grabbed", which
1063 * makes sure updates are visible before bind or setsockopt return. */
1064 }
1065
1066 static int netlink_insert(struct sock *sk, struct net *net, u32 portid)
1067 {
1068 struct netlink_table *table = &nl_table[sk->sk_protocol];
1069 int err = -EADDRINUSE;
1070
1071 mutex_lock(&nl_sk_hash_lock);
1072 if (__netlink_lookup(table, portid, net))
1073 goto err;
1074
1075 err = -EBUSY;
1076 if (nlk_sk(sk)->portid)
1077 goto err;
1078
1079 err = -ENOMEM;
1080 if (BITS_PER_LONG > 32 && unlikely(table->hash.nelems >= UINT_MAX))
1081 goto err;
1082
1083 nlk_sk(sk)->portid = portid;
1084 sock_hold(sk);
1085 rhashtable_insert(&table->hash, &nlk_sk(sk)->node, GFP_KERNEL);
1086 err = 0;
1087 err:
1088 mutex_unlock(&nl_sk_hash_lock);
1089 return err;
1090 }
1091
1092 static void netlink_remove(struct sock *sk)
1093 {
1094 struct netlink_table *table;
1095
1096 mutex_lock(&nl_sk_hash_lock);
1097 table = &nl_table[sk->sk_protocol];
1098 if (rhashtable_remove(&table->hash, &nlk_sk(sk)->node, GFP_KERNEL)) {
1099 WARN_ON(atomic_read(&sk->sk_refcnt) == 1);
1100 __sock_put(sk);
1101 }
1102 mutex_unlock(&nl_sk_hash_lock);
1103
1104 netlink_table_grab();
1105 if (nlk_sk(sk)->subscriptions)
1106 __sk_del_bind_node(sk);
1107 netlink_table_ungrab();
1108 }
1109
1110 static struct proto netlink_proto = {
1111 .name = "NETLINK",
1112 .owner = THIS_MODULE,
1113 .obj_size = sizeof(struct netlink_sock),
1114 };
1115
1116 static int __netlink_create(struct net *net, struct socket *sock,
1117 struct mutex *cb_mutex, int protocol)
1118 {
1119 struct sock *sk;
1120 struct netlink_sock *nlk;
1121
1122 sock->ops = &netlink_ops;
1123
1124 sk = sk_alloc(net, PF_NETLINK, GFP_KERNEL, &netlink_proto);
1125 if (!sk)
1126 return -ENOMEM;
1127
1128 sock_init_data(sock, sk);
1129
1130 nlk = nlk_sk(sk);
1131 if (cb_mutex) {
1132 nlk->cb_mutex = cb_mutex;
1133 } else {
1134 nlk->cb_mutex = &nlk->cb_def_mutex;
1135 mutex_init(nlk->cb_mutex);
1136 }
1137 init_waitqueue_head(&nlk->wait);
1138 #ifdef CONFIG_NETLINK_MMAP
1139 mutex_init(&nlk->pg_vec_lock);
1140 #endif
1141
1142 sk->sk_destruct = netlink_sock_destruct;
1143 sk->sk_protocol = protocol;
1144 return 0;
1145 }
1146
1147 static int netlink_create(struct net *net, struct socket *sock, int protocol,
1148 int kern)
1149 {
1150 struct module *module = NULL;
1151 struct mutex *cb_mutex;
1152 struct netlink_sock *nlk;
1153 int (*bind)(int group);
1154 void (*unbind)(int group);
1155 int err = 0;
1156
1157 sock->state = SS_UNCONNECTED;
1158
1159 if (sock->type != SOCK_RAW && sock->type != SOCK_DGRAM)
1160 return -ESOCKTNOSUPPORT;
1161
1162 if (protocol < 0 || protocol >= MAX_LINKS)
1163 return -EPROTONOSUPPORT;
1164
1165 netlink_lock_table();
1166 #ifdef CONFIG_MODULES
1167 if (!nl_table[protocol].registered) {
1168 netlink_unlock_table();
1169 request_module("net-pf-%d-proto-%d", PF_NETLINK, protocol);
1170 netlink_lock_table();
1171 }
1172 #endif
1173 if (nl_table[protocol].registered &&
1174 try_module_get(nl_table[protocol].module))
1175 module = nl_table[protocol].module;
1176 else
1177 err = -EPROTONOSUPPORT;
1178 cb_mutex = nl_table[protocol].cb_mutex;
1179 bind = nl_table[protocol].bind;
1180 unbind = nl_table[protocol].unbind;
1181 netlink_unlock_table();
1182
1183 if (err < 0)
1184 goto out;
1185
1186 err = __netlink_create(net, sock, cb_mutex, protocol);
1187 if (err < 0)
1188 goto out_module;
1189
1190 local_bh_disable();
1191 sock_prot_inuse_add(net, &netlink_proto, 1);
1192 local_bh_enable();
1193
1194 nlk = nlk_sk(sock->sk);
1195 nlk->module = module;
1196 nlk->netlink_bind = bind;
1197 nlk->netlink_unbind = unbind;
1198 out:
1199 return err;
1200
1201 out_module:
1202 module_put(module);
1203 goto out;
1204 }
1205
1206 static int netlink_release(struct socket *sock)
1207 {
1208 struct sock *sk = sock->sk;
1209 struct netlink_sock *nlk;
1210
1211 if (!sk)
1212 return 0;
1213
1214 netlink_remove(sk);
1215 sock_orphan(sk);
1216 nlk = nlk_sk(sk);
1217
1218 /*
1219 * OK. Socket is unlinked, any packets that arrive now
1220 * will be purged.
1221 */
1222
1223 sock->sk = NULL;
1224 wake_up_interruptible_all(&nlk->wait);
1225
1226 skb_queue_purge(&sk->sk_write_queue);
1227
1228 if (nlk->portid) {
1229 struct netlink_notify n = {
1230 .net = sock_net(sk),
1231 .protocol = sk->sk_protocol,
1232 .portid = nlk->portid,
1233 };
1234 atomic_notifier_call_chain(&netlink_chain,
1235 NETLINK_URELEASE, &n);
1236 }
1237
1238 module_put(nlk->module);
1239
1240 netlink_table_grab();
1241 if (netlink_is_kernel(sk)) {
1242 BUG_ON(nl_table[sk->sk_protocol].registered == 0);
1243 if (--nl_table[sk->sk_protocol].registered == 0) {
1244 struct listeners *old;
1245
1246 old = nl_deref_protected(nl_table[sk->sk_protocol].listeners);
1247 RCU_INIT_POINTER(nl_table[sk->sk_protocol].listeners, NULL);
1248 kfree_rcu(old, rcu);
1249 nl_table[sk->sk_protocol].module = NULL;
1250 nl_table[sk->sk_protocol].bind = NULL;
1251 nl_table[sk->sk_protocol].unbind = NULL;
1252 nl_table[sk->sk_protocol].flags = 0;
1253 nl_table[sk->sk_protocol].registered = 0;
1254 }
1255 } else if (nlk->subscriptions) {
1256 netlink_update_listeners(sk);
1257 }
1258 netlink_table_ungrab();
1259
1260 /* Wait for readers to complete */
1261 synchronize_net();
1262
1263 kfree(nlk->groups);
1264 nlk->groups = NULL;
1265
1266 local_bh_disable();
1267 sock_prot_inuse_add(sock_net(sk), &netlink_proto, -1);
1268 local_bh_enable();
1269 sock_put(sk);
1270 return 0;
1271 }
1272
1273 static int netlink_autobind(struct socket *sock)
1274 {
1275 struct sock *sk = sock->sk;
1276 struct net *net = sock_net(sk);
1277 struct netlink_table *table = &nl_table[sk->sk_protocol];
1278 s32 portid = task_tgid_vnr(current);
1279 int err;
1280 static s32 rover = -4097;
1281
1282 retry:
1283 cond_resched();
1284 rcu_read_lock();
1285 if (__netlink_lookup(table, portid, net)) {
1286 /* Bind collision, search negative portid values. */
1287 portid = rover--;
1288 if (rover > -4097)
1289 rover = -4097;
1290 rcu_read_unlock();
1291 goto retry;
1292 }
1293 rcu_read_unlock();
1294
1295 err = netlink_insert(sk, net, portid);
1296 if (err == -EADDRINUSE)
1297 goto retry;
1298
1299 /* If 2 threads race to autobind, that is fine. */
1300 if (err == -EBUSY)
1301 err = 0;
1302
1303 return err;
1304 }
1305
1306 /**
1307 * __netlink_ns_capable - General netlink message capability test
1308 * @nsp: NETLINK_CB of the socket buffer holding a netlink command from userspace.
1309 * @user_ns: The user namespace of the capability to use
1310 * @cap: The capability to use
1311 *
1312 * Test to see if the opener of the socket we received the message
1313 * from had when the netlink socket was created and the sender of the
1314 * message has has the capability @cap in the user namespace @user_ns.
1315 */
1316 bool __netlink_ns_capable(const struct netlink_skb_parms *nsp,
1317 struct user_namespace *user_ns, int cap)
1318 {
1319 return ((nsp->flags & NETLINK_SKB_DST) ||
1320 file_ns_capable(nsp->sk->sk_socket->file, user_ns, cap)) &&
1321 ns_capable(user_ns, cap);
1322 }
1323 EXPORT_SYMBOL(__netlink_ns_capable);
1324
1325 /**
1326 * netlink_ns_capable - General netlink message capability test
1327 * @skb: socket buffer holding a netlink command from userspace
1328 * @user_ns: The user namespace of the capability to use
1329 * @cap: The capability to use
1330 *
1331 * Test to see if the opener of the socket we received the message
1332 * from had when the netlink socket was created and the sender of the
1333 * message has has the capability @cap in the user namespace @user_ns.
1334 */
1335 bool netlink_ns_capable(const struct sk_buff *skb,
1336 struct user_namespace *user_ns, int cap)
1337 {
1338 return __netlink_ns_capable(&NETLINK_CB(skb), user_ns, cap);
1339 }
1340 EXPORT_SYMBOL(netlink_ns_capable);
1341
1342 /**
1343 * netlink_capable - Netlink global message capability test
1344 * @skb: socket buffer holding a netlink command from userspace
1345 * @cap: The capability to use
1346 *
1347 * Test to see if the opener of the socket we received the message
1348 * from had when the netlink socket was created and the sender of the
1349 * message has has the capability @cap in all user namespaces.
1350 */
1351 bool netlink_capable(const struct sk_buff *skb, int cap)
1352 {
1353 return netlink_ns_capable(skb, &init_user_ns, cap);
1354 }
1355 EXPORT_SYMBOL(netlink_capable);
1356
1357 /**
1358 * netlink_net_capable - Netlink network namespace message capability test
1359 * @skb: socket buffer holding a netlink command from userspace
1360 * @cap: The capability to use
1361 *
1362 * Test to see if the opener of the socket we received the message
1363 * from had when the netlink socket was created and the sender of the
1364 * message has has the capability @cap over the network namespace of
1365 * the socket we received the message from.
1366 */
1367 bool netlink_net_capable(const struct sk_buff *skb, int cap)
1368 {
1369 return netlink_ns_capable(skb, sock_net(skb->sk)->user_ns, cap);
1370 }
1371 EXPORT_SYMBOL(netlink_net_capable);
1372
1373 static inline int netlink_allowed(const struct socket *sock, unsigned int flag)
1374 {
1375 return (nl_table[sock->sk->sk_protocol].flags & flag) ||
1376 ns_capable(sock_net(sock->sk)->user_ns, CAP_NET_ADMIN);
1377 }
1378
1379 static void
1380 netlink_update_subscriptions(struct sock *sk, unsigned int subscriptions)
1381 {
1382 struct netlink_sock *nlk = nlk_sk(sk);
1383
1384 if (nlk->subscriptions && !subscriptions)
1385 __sk_del_bind_node(sk);
1386 else if (!nlk->subscriptions && subscriptions)
1387 sk_add_bind_node(sk, &nl_table[sk->sk_protocol].mc_list);
1388 nlk->subscriptions = subscriptions;
1389 }
1390
1391 static int netlink_realloc_groups(struct sock *sk)
1392 {
1393 struct netlink_sock *nlk = nlk_sk(sk);
1394 unsigned int groups;
1395 unsigned long *new_groups;
1396 int err = 0;
1397
1398 netlink_table_grab();
1399
1400 groups = nl_table[sk->sk_protocol].groups;
1401 if (!nl_table[sk->sk_protocol].registered) {
1402 err = -ENOENT;
1403 goto out_unlock;
1404 }
1405
1406 if (nlk->ngroups >= groups)
1407 goto out_unlock;
1408
1409 new_groups = krealloc(nlk->groups, NLGRPSZ(groups), GFP_ATOMIC);
1410 if (new_groups == NULL) {
1411 err = -ENOMEM;
1412 goto out_unlock;
1413 }
1414 memset((char *)new_groups + NLGRPSZ(nlk->ngroups), 0,
1415 NLGRPSZ(groups) - NLGRPSZ(nlk->ngroups));
1416
1417 nlk->groups = new_groups;
1418 nlk->ngroups = groups;
1419 out_unlock:
1420 netlink_table_ungrab();
1421 return err;
1422 }
1423
1424 static void netlink_unbind(int group, long unsigned int groups,
1425 struct netlink_sock *nlk)
1426 {
1427 int undo;
1428
1429 if (!nlk->netlink_unbind)
1430 return;
1431
1432 for (undo = 0; undo < group; undo++)
1433 if (test_bit(group, &groups))
1434 nlk->netlink_unbind(undo);
1435 }
1436
1437 static int netlink_bind(struct socket *sock, struct sockaddr *addr,
1438 int addr_len)
1439 {
1440 struct sock *sk = sock->sk;
1441 struct net *net = sock_net(sk);
1442 struct netlink_sock *nlk = nlk_sk(sk);
1443 struct sockaddr_nl *nladdr = (struct sockaddr_nl *)addr;
1444 int err;
1445 long unsigned int groups = nladdr->nl_groups;
1446
1447 if (addr_len < sizeof(struct sockaddr_nl))
1448 return -EINVAL;
1449
1450 if (nladdr->nl_family != AF_NETLINK)
1451 return -EINVAL;
1452
1453 /* Only superuser is allowed to listen multicasts */
1454 if (groups) {
1455 if (!netlink_allowed(sock, NL_CFG_F_NONROOT_RECV))
1456 return -EPERM;
1457 err = netlink_realloc_groups(sk);
1458 if (err)
1459 return err;
1460 }
1461
1462 if (nlk->portid)
1463 if (nladdr->nl_pid != nlk->portid)
1464 return -EINVAL;
1465
1466 if (nlk->netlink_bind && groups) {
1467 int group;
1468
1469 for (group = 0; group < nlk->ngroups; group++) {
1470 if (!test_bit(group, &groups))
1471 continue;
1472 err = nlk->netlink_bind(group);
1473 if (!err)
1474 continue;
1475 netlink_unbind(group, groups, nlk);
1476 return err;
1477 }
1478 }
1479
1480 if (!nlk->portid) {
1481 err = nladdr->nl_pid ?
1482 netlink_insert(sk, net, nladdr->nl_pid) :
1483 netlink_autobind(sock);
1484 if (err) {
1485 netlink_unbind(nlk->ngroups - 1, groups, nlk);
1486 return err;
1487 }
1488 }
1489
1490 if (!groups && (nlk->groups == NULL || !(u32)nlk->groups[0]))
1491 return 0;
1492
1493 netlink_table_grab();
1494 netlink_update_subscriptions(sk, nlk->subscriptions +
1495 hweight32(groups) -
1496 hweight32(nlk->groups[0]));
1497 nlk->groups[0] = (nlk->groups[0] & ~0xffffffffUL) | groups;
1498 netlink_update_listeners(sk);
1499 netlink_table_ungrab();
1500
1501 return 0;
1502 }
1503
1504 static int netlink_connect(struct socket *sock, struct sockaddr *addr,
1505 int alen, int flags)
1506 {
1507 int err = 0;
1508 struct sock *sk = sock->sk;
1509 struct netlink_sock *nlk = nlk_sk(sk);
1510 struct sockaddr_nl *nladdr = (struct sockaddr_nl *)addr;
1511
1512 if (alen < sizeof(addr->sa_family))
1513 return -EINVAL;
1514
1515 if (addr->sa_family == AF_UNSPEC) {
1516 sk->sk_state = NETLINK_UNCONNECTED;
1517 nlk->dst_portid = 0;
1518 nlk->dst_group = 0;
1519 return 0;
1520 }
1521 if (addr->sa_family != AF_NETLINK)
1522 return -EINVAL;
1523
1524 if ((nladdr->nl_groups || nladdr->nl_pid) &&
1525 !netlink_allowed(sock, NL_CFG_F_NONROOT_SEND))
1526 return -EPERM;
1527
1528 if (!nlk->portid)
1529 err = netlink_autobind(sock);
1530
1531 if (err == 0) {
1532 sk->sk_state = NETLINK_CONNECTED;
1533 nlk->dst_portid = nladdr->nl_pid;
1534 nlk->dst_group = ffs(nladdr->nl_groups);
1535 }
1536
1537 return err;
1538 }
1539
1540 static int netlink_getname(struct socket *sock, struct sockaddr *addr,
1541 int *addr_len, int peer)
1542 {
1543 struct sock *sk = sock->sk;
1544 struct netlink_sock *nlk = nlk_sk(sk);
1545 DECLARE_SOCKADDR(struct sockaddr_nl *, nladdr, addr);
1546
1547 nladdr->nl_family = AF_NETLINK;
1548 nladdr->nl_pad = 0;
1549 *addr_len = sizeof(*nladdr);
1550
1551 if (peer) {
1552 nladdr->nl_pid = nlk->dst_portid;
1553 nladdr->nl_groups = netlink_group_mask(nlk->dst_group);
1554 } else {
1555 nladdr->nl_pid = nlk->portid;
1556 nladdr->nl_groups = nlk->groups ? nlk->groups[0] : 0;
1557 }
1558 return 0;
1559 }
1560
1561 static struct sock *netlink_getsockbyportid(struct sock *ssk, u32 portid)
1562 {
1563 struct sock *sock;
1564 struct netlink_sock *nlk;
1565
1566 sock = netlink_lookup(sock_net(ssk), ssk->sk_protocol, portid);
1567 if (!sock)
1568 return ERR_PTR(-ECONNREFUSED);
1569
1570 /* Don't bother queuing skb if kernel socket has no input function */
1571 nlk = nlk_sk(sock);
1572 if (sock->sk_state == NETLINK_CONNECTED &&
1573 nlk->dst_portid != nlk_sk(ssk)->portid) {
1574 sock_put(sock);
1575 return ERR_PTR(-ECONNREFUSED);
1576 }
1577 return sock;
1578 }
1579
1580 struct sock *netlink_getsockbyfilp(struct file *filp)
1581 {
1582 struct inode *inode = file_inode(filp);
1583 struct sock *sock;
1584
1585 if (!S_ISSOCK(inode->i_mode))
1586 return ERR_PTR(-ENOTSOCK);
1587
1588 sock = SOCKET_I(inode)->sk;
1589 if (sock->sk_family != AF_NETLINK)
1590 return ERR_PTR(-EINVAL);
1591
1592 sock_hold(sock);
1593 return sock;
1594 }
1595
1596 static struct sk_buff *netlink_alloc_large_skb(unsigned int size,
1597 int broadcast)
1598 {
1599 struct sk_buff *skb;
1600 void *data;
1601
1602 if (size <= NLMSG_GOODSIZE || broadcast)
1603 return alloc_skb(size, GFP_KERNEL);
1604
1605 size = SKB_DATA_ALIGN(size) +
1606 SKB_DATA_ALIGN(sizeof(struct skb_shared_info));
1607
1608 data = vmalloc(size);
1609 if (data == NULL)
1610 return NULL;
1611
1612 skb = build_skb(data, size);
1613 if (skb == NULL)
1614 vfree(data);
1615 else {
1616 skb->head_frag = 0;
1617 skb->destructor = netlink_skb_destructor;
1618 }
1619
1620 return skb;
1621 }
1622
1623 /*
1624 * Attach a skb to a netlink socket.
1625 * The caller must hold a reference to the destination socket. On error, the
1626 * reference is dropped. The skb is not send to the destination, just all
1627 * all error checks are performed and memory in the queue is reserved.
1628 * Return values:
1629 * < 0: error. skb freed, reference to sock dropped.
1630 * 0: continue
1631 * 1: repeat lookup - reference dropped while waiting for socket memory.
1632 */
1633 int netlink_attachskb(struct sock *sk, struct sk_buff *skb,
1634 long *timeo, struct sock *ssk)
1635 {
1636 struct netlink_sock *nlk;
1637
1638 nlk = nlk_sk(sk);
1639
1640 if ((atomic_read(&sk->sk_rmem_alloc) > sk->sk_rcvbuf ||
1641 test_bit(NETLINK_CONGESTED, &nlk->state)) &&
1642 !netlink_skb_is_mmaped(skb)) {
1643 DECLARE_WAITQUEUE(wait, current);
1644 if (!*timeo) {
1645 if (!ssk || netlink_is_kernel(ssk))
1646 netlink_overrun(sk);
1647 sock_put(sk);
1648 kfree_skb(skb);
1649 return -EAGAIN;
1650 }
1651
1652 __set_current_state(TASK_INTERRUPTIBLE);
1653 add_wait_queue(&nlk->wait, &wait);
1654
1655 if ((atomic_read(&sk->sk_rmem_alloc) > sk->sk_rcvbuf ||
1656 test_bit(NETLINK_CONGESTED, &nlk->state)) &&
1657 !sock_flag(sk, SOCK_DEAD))
1658 *timeo = schedule_timeout(*timeo);
1659
1660 __set_current_state(TASK_RUNNING);
1661 remove_wait_queue(&nlk->wait, &wait);
1662 sock_put(sk);
1663
1664 if (signal_pending(current)) {
1665 kfree_skb(skb);
1666 return sock_intr_errno(*timeo);
1667 }
1668 return 1;
1669 }
1670 netlink_skb_set_owner_r(skb, sk);
1671 return 0;
1672 }
1673
1674 static int __netlink_sendskb(struct sock *sk, struct sk_buff *skb)
1675 {
1676 int len = skb->len;
1677
1678 netlink_deliver_tap(skb);
1679
1680 #ifdef CONFIG_NETLINK_MMAP
1681 if (netlink_skb_is_mmaped(skb))
1682 netlink_queue_mmaped_skb(sk, skb);
1683 else if (netlink_rx_is_mmaped(sk))
1684 netlink_ring_set_copied(sk, skb);
1685 else
1686 #endif /* CONFIG_NETLINK_MMAP */
1687 skb_queue_tail(&sk->sk_receive_queue, skb);
1688 sk->sk_data_ready(sk);
1689 return len;
1690 }
1691
1692 int netlink_sendskb(struct sock *sk, struct sk_buff *skb)
1693 {
1694 int len = __netlink_sendskb(sk, skb);
1695
1696 sock_put(sk);
1697 return len;
1698 }
1699
1700 void netlink_detachskb(struct sock *sk, struct sk_buff *skb)
1701 {
1702 kfree_skb(skb);
1703 sock_put(sk);
1704 }
1705
1706 static struct sk_buff *netlink_trim(struct sk_buff *skb, gfp_t allocation)
1707 {
1708 int delta;
1709
1710 WARN_ON(skb->sk != NULL);
1711 if (netlink_skb_is_mmaped(skb))
1712 return skb;
1713
1714 delta = skb->end - skb->tail;
1715 if (is_vmalloc_addr(skb->head) || delta * 2 < skb->truesize)
1716 return skb;
1717
1718 if (skb_shared(skb)) {
1719 struct sk_buff *nskb = skb_clone(skb, allocation);
1720 if (!nskb)
1721 return skb;
1722 consume_skb(skb);
1723 skb = nskb;
1724 }
1725
1726 if (!pskb_expand_head(skb, 0, -delta, allocation))
1727 skb->truesize -= delta;
1728
1729 return skb;
1730 }
1731
1732 static int netlink_unicast_kernel(struct sock *sk, struct sk_buff *skb,
1733 struct sock *ssk)
1734 {
1735 int ret;
1736 struct netlink_sock *nlk = nlk_sk(sk);
1737
1738 ret = -ECONNREFUSED;
1739 if (nlk->netlink_rcv != NULL) {
1740 ret = skb->len;
1741 netlink_skb_set_owner_r(skb, sk);
1742 NETLINK_CB(skb).sk = ssk;
1743 netlink_deliver_tap_kernel(sk, ssk, skb);
1744 nlk->netlink_rcv(skb);
1745 consume_skb(skb);
1746 } else {
1747 kfree_skb(skb);
1748 }
1749 sock_put(sk);
1750 return ret;
1751 }
1752
1753 int netlink_unicast(struct sock *ssk, struct sk_buff *skb,
1754 u32 portid, int nonblock)
1755 {
1756 struct sock *sk;
1757 int err;
1758 long timeo;
1759
1760 skb = netlink_trim(skb, gfp_any());
1761
1762 timeo = sock_sndtimeo(ssk, nonblock);
1763 retry:
1764 sk = netlink_getsockbyportid(ssk, portid);
1765 if (IS_ERR(sk)) {
1766 kfree_skb(skb);
1767 return PTR_ERR(sk);
1768 }
1769 if (netlink_is_kernel(sk))
1770 return netlink_unicast_kernel(sk, skb, ssk);
1771
1772 if (sk_filter(sk, skb)) {
1773 err = skb->len;
1774 kfree_skb(skb);
1775 sock_put(sk);
1776 return err;
1777 }
1778
1779 err = netlink_attachskb(sk, skb, &timeo, ssk);
1780 if (err == 1)
1781 goto retry;
1782 if (err)
1783 return err;
1784
1785 return netlink_sendskb(sk, skb);
1786 }
1787 EXPORT_SYMBOL(netlink_unicast);
1788
1789 struct sk_buff *netlink_alloc_skb(struct sock *ssk, unsigned int size,
1790 u32 dst_portid, gfp_t gfp_mask)
1791 {
1792 #ifdef CONFIG_NETLINK_MMAP
1793 struct sock *sk = NULL;
1794 struct sk_buff *skb;
1795 struct netlink_ring *ring;
1796 struct nl_mmap_hdr *hdr;
1797 unsigned int maxlen;
1798
1799 sk = netlink_getsockbyportid(ssk, dst_portid);
1800 if (IS_ERR(sk))
1801 goto out;
1802
1803 ring = &nlk_sk(sk)->rx_ring;
1804 /* fast-path without atomic ops for common case: non-mmaped receiver */
1805 if (ring->pg_vec == NULL)
1806 goto out_put;
1807
1808 if (ring->frame_size - NL_MMAP_HDRLEN < size)
1809 goto out_put;
1810
1811 skb = alloc_skb_head(gfp_mask);
1812 if (skb == NULL)
1813 goto err1;
1814
1815 spin_lock_bh(&sk->sk_receive_queue.lock);
1816 /* check again under lock */
1817 if (ring->pg_vec == NULL)
1818 goto out_free;
1819
1820 /* check again under lock */
1821 maxlen = ring->frame_size - NL_MMAP_HDRLEN;
1822 if (maxlen < size)
1823 goto out_free;
1824
1825 netlink_forward_ring(ring);
1826 hdr = netlink_current_frame(ring, NL_MMAP_STATUS_UNUSED);
1827 if (hdr == NULL)
1828 goto err2;
1829 netlink_ring_setup_skb(skb, sk, ring, hdr);
1830 netlink_set_status(hdr, NL_MMAP_STATUS_RESERVED);
1831 atomic_inc(&ring->pending);
1832 netlink_increment_head(ring);
1833
1834 spin_unlock_bh(&sk->sk_receive_queue.lock);
1835 return skb;
1836
1837 err2:
1838 kfree_skb(skb);
1839 spin_unlock_bh(&sk->sk_receive_queue.lock);
1840 netlink_overrun(sk);
1841 err1:
1842 sock_put(sk);
1843 return NULL;
1844
1845 out_free:
1846 kfree_skb(skb);
1847 spin_unlock_bh(&sk->sk_receive_queue.lock);
1848 out_put:
1849 sock_put(sk);
1850 out:
1851 #endif
1852 return alloc_skb(size, gfp_mask);
1853 }
1854 EXPORT_SYMBOL_GPL(netlink_alloc_skb);
1855
1856 int netlink_has_listeners(struct sock *sk, unsigned int group)
1857 {
1858 int res = 0;
1859 struct listeners *listeners;
1860
1861 BUG_ON(!netlink_is_kernel(sk));
1862
1863 rcu_read_lock();
1864 listeners = rcu_dereference(nl_table[sk->sk_protocol].listeners);
1865
1866 if (listeners && group - 1 < nl_table[sk->sk_protocol].groups)
1867 res = test_bit(group - 1, listeners->masks);
1868
1869 rcu_read_unlock();
1870
1871 return res;
1872 }
1873 EXPORT_SYMBOL_GPL(netlink_has_listeners);
1874
1875 static int netlink_broadcast_deliver(struct sock *sk, struct sk_buff *skb)
1876 {
1877 struct netlink_sock *nlk = nlk_sk(sk);
1878
1879 if (atomic_read(&sk->sk_rmem_alloc) <= sk->sk_rcvbuf &&
1880 !test_bit(NETLINK_CONGESTED, &nlk->state)) {
1881 netlink_skb_set_owner_r(skb, sk);
1882 __netlink_sendskb(sk, skb);
1883 return atomic_read(&sk->sk_rmem_alloc) > (sk->sk_rcvbuf >> 1);
1884 }
1885 return -1;
1886 }
1887
1888 struct netlink_broadcast_data {
1889 struct sock *exclude_sk;
1890 struct net *net;
1891 u32 portid;
1892 u32 group;
1893 int failure;
1894 int delivery_failure;
1895 int congested;
1896 int delivered;
1897 gfp_t allocation;
1898 struct sk_buff *skb, *skb2;
1899 int (*tx_filter)(struct sock *dsk, struct sk_buff *skb, void *data);
1900 void *tx_data;
1901 };
1902
1903 static void do_one_broadcast(struct sock *sk,
1904 struct netlink_broadcast_data *p)
1905 {
1906 struct netlink_sock *nlk = nlk_sk(sk);
1907 int val;
1908
1909 if (p->exclude_sk == sk)
1910 return;
1911
1912 if (nlk->portid == p->portid || p->group - 1 >= nlk->ngroups ||
1913 !test_bit(p->group - 1, nlk->groups))
1914 return;
1915
1916 if (!net_eq(sock_net(sk), p->net))
1917 return;
1918
1919 if (p->failure) {
1920 netlink_overrun(sk);
1921 return;
1922 }
1923
1924 sock_hold(sk);
1925 if (p->skb2 == NULL) {
1926 if (skb_shared(p->skb)) {
1927 p->skb2 = skb_clone(p->skb, p->allocation);
1928 } else {
1929 p->skb2 = skb_get(p->skb);
1930 /*
1931 * skb ownership may have been set when
1932 * delivered to a previous socket.
1933 */
1934 skb_orphan(p->skb2);
1935 }
1936 }
1937 if (p->skb2 == NULL) {
1938 netlink_overrun(sk);
1939 /* Clone failed. Notify ALL listeners. */
1940 p->failure = 1;
1941 if (nlk->flags & NETLINK_BROADCAST_SEND_ERROR)
1942 p->delivery_failure = 1;
1943 } else if (p->tx_filter && p->tx_filter(sk, p->skb2, p->tx_data)) {
1944 kfree_skb(p->skb2);
1945 p->skb2 = NULL;
1946 } else if (sk_filter(sk, p->skb2)) {
1947 kfree_skb(p->skb2);
1948 p->skb2 = NULL;
1949 } else if ((val = netlink_broadcast_deliver(sk, p->skb2)) < 0) {
1950 netlink_overrun(sk);
1951 if (nlk->flags & NETLINK_BROADCAST_SEND_ERROR)
1952 p->delivery_failure = 1;
1953 } else {
1954 p->congested |= val;
1955 p->delivered = 1;
1956 p->skb2 = NULL;
1957 }
1958 sock_put(sk);
1959 }
1960
1961 int netlink_broadcast_filtered(struct sock *ssk, struct sk_buff *skb, u32 portid,
1962 u32 group, gfp_t allocation,
1963 int (*filter)(struct sock *dsk, struct sk_buff *skb, void *data),
1964 void *filter_data)
1965 {
1966 struct net *net = sock_net(ssk);
1967 struct netlink_broadcast_data info;
1968 struct sock *sk;
1969
1970 skb = netlink_trim(skb, allocation);
1971
1972 info.exclude_sk = ssk;
1973 info.net = net;
1974 info.portid = portid;
1975 info.group = group;
1976 info.failure = 0;
1977 info.delivery_failure = 0;
1978 info.congested = 0;
1979 info.delivered = 0;
1980 info.allocation = allocation;
1981 info.skb = skb;
1982 info.skb2 = NULL;
1983 info.tx_filter = filter;
1984 info.tx_data = filter_data;
1985
1986 /* While we sleep in clone, do not allow to change socket list */
1987
1988 netlink_lock_table();
1989
1990 sk_for_each_bound(sk, &nl_table[ssk->sk_protocol].mc_list)
1991 do_one_broadcast(sk, &info);
1992
1993 consume_skb(skb);
1994
1995 netlink_unlock_table();
1996
1997 if (info.delivery_failure) {
1998 kfree_skb(info.skb2);
1999 return -ENOBUFS;
2000 }
2001 consume_skb(info.skb2);
2002
2003 if (info.delivered) {
2004 if (info.congested && (allocation & __GFP_WAIT))
2005 yield();
2006 return 0;
2007 }
2008 return -ESRCH;
2009 }
2010 EXPORT_SYMBOL(netlink_broadcast_filtered);
2011
2012 int netlink_broadcast(struct sock *ssk, struct sk_buff *skb, u32 portid,
2013 u32 group, gfp_t allocation)
2014 {
2015 return netlink_broadcast_filtered(ssk, skb, portid, group, allocation,
2016 NULL, NULL);
2017 }
2018 EXPORT_SYMBOL(netlink_broadcast);
2019
2020 struct netlink_set_err_data {
2021 struct sock *exclude_sk;
2022 u32 portid;
2023 u32 group;
2024 int code;
2025 };
2026
2027 static int do_one_set_err(struct sock *sk, struct netlink_set_err_data *p)
2028 {
2029 struct netlink_sock *nlk = nlk_sk(sk);
2030 int ret = 0;
2031
2032 if (sk == p->exclude_sk)
2033 goto out;
2034
2035 if (!net_eq(sock_net(sk), sock_net(p->exclude_sk)))
2036 goto out;
2037
2038 if (nlk->portid == p->portid || p->group - 1 >= nlk->ngroups ||
2039 !test_bit(p->group - 1, nlk->groups))
2040 goto out;
2041
2042 if (p->code == ENOBUFS && nlk->flags & NETLINK_RECV_NO_ENOBUFS) {
2043 ret = 1;
2044 goto out;
2045 }
2046
2047 sk->sk_err = p->code;
2048 sk->sk_error_report(sk);
2049 out:
2050 return ret;
2051 }
2052
2053 /**
2054 * netlink_set_err - report error to broadcast listeners
2055 * @ssk: the kernel netlink socket, as returned by netlink_kernel_create()
2056 * @portid: the PORTID of a process that we want to skip (if any)
2057 * @group: the broadcast group that will notice the error
2058 * @code: error code, must be negative (as usual in kernelspace)
2059 *
2060 * This function returns the number of broadcast listeners that have set the
2061 * NETLINK_RECV_NO_ENOBUFS socket option.
2062 */
2063 int netlink_set_err(struct sock *ssk, u32 portid, u32 group, int code)
2064 {
2065 struct netlink_set_err_data info;
2066 struct sock *sk;
2067 int ret = 0;
2068
2069 info.exclude_sk = ssk;
2070 info.portid = portid;
2071 info.group = group;
2072 /* sk->sk_err wants a positive error value */
2073 info.code = -code;
2074
2075 read_lock(&nl_table_lock);
2076
2077 sk_for_each_bound(sk, &nl_table[ssk->sk_protocol].mc_list)
2078 ret += do_one_set_err(sk, &info);
2079
2080 read_unlock(&nl_table_lock);
2081 return ret;
2082 }
2083 EXPORT_SYMBOL(netlink_set_err);
2084
2085 /* must be called with netlink table grabbed */
2086 static void netlink_update_socket_mc(struct netlink_sock *nlk,
2087 unsigned int group,
2088 int is_new)
2089 {
2090 int old, new = !!is_new, subscriptions;
2091
2092 old = test_bit(group - 1, nlk->groups);
2093 subscriptions = nlk->subscriptions - old + new;
2094 if (new)
2095 __set_bit(group - 1, nlk->groups);
2096 else
2097 __clear_bit(group - 1, nlk->groups);
2098 netlink_update_subscriptions(&nlk->sk, subscriptions);
2099 netlink_update_listeners(&nlk->sk);
2100 }
2101
2102 static int netlink_setsockopt(struct socket *sock, int level, int optname,
2103 char __user *optval, unsigned int optlen)
2104 {
2105 struct sock *sk = sock->sk;
2106 struct netlink_sock *nlk = nlk_sk(sk);
2107 unsigned int val = 0;
2108 int err;
2109
2110 if (level != SOL_NETLINK)
2111 return -ENOPROTOOPT;
2112
2113 if (optname != NETLINK_RX_RING && optname != NETLINK_TX_RING &&
2114 optlen >= sizeof(int) &&
2115 get_user(val, (unsigned int __user *)optval))
2116 return -EFAULT;
2117
2118 switch (optname) {
2119 case NETLINK_PKTINFO:
2120 if (val)
2121 nlk->flags |= NETLINK_RECV_PKTINFO;
2122 else
2123 nlk->flags &= ~NETLINK_RECV_PKTINFO;
2124 err = 0;
2125 break;
2126 case NETLINK_ADD_MEMBERSHIP:
2127 case NETLINK_DROP_MEMBERSHIP: {
2128 if (!netlink_allowed(sock, NL_CFG_F_NONROOT_RECV))
2129 return -EPERM;
2130 err = netlink_realloc_groups(sk);
2131 if (err)
2132 return err;
2133 if (!val || val - 1 >= nlk->ngroups)
2134 return -EINVAL;
2135 if (optname == NETLINK_ADD_MEMBERSHIP && nlk->netlink_bind) {
2136 err = nlk->netlink_bind(val);
2137 if (err)
2138 return err;
2139 }
2140 netlink_table_grab();
2141 netlink_update_socket_mc(nlk, val,
2142 optname == NETLINK_ADD_MEMBERSHIP);
2143 netlink_table_ungrab();
2144 if (optname == NETLINK_DROP_MEMBERSHIP && nlk->netlink_unbind)
2145 nlk->netlink_unbind(val);
2146
2147 err = 0;
2148 break;
2149 }
2150 case NETLINK_BROADCAST_ERROR:
2151 if (val)
2152 nlk->flags |= NETLINK_BROADCAST_SEND_ERROR;
2153 else
2154 nlk->flags &= ~NETLINK_BROADCAST_SEND_ERROR;
2155 err = 0;
2156 break;
2157 case NETLINK_NO_ENOBUFS:
2158 if (val) {
2159 nlk->flags |= NETLINK_RECV_NO_ENOBUFS;
2160 clear_bit(NETLINK_CONGESTED, &nlk->state);
2161 wake_up_interruptible(&nlk->wait);
2162 } else {
2163 nlk->flags &= ~NETLINK_RECV_NO_ENOBUFS;
2164 }
2165 err = 0;
2166 break;
2167 #ifdef CONFIG_NETLINK_MMAP
2168 case NETLINK_RX_RING:
2169 case NETLINK_TX_RING: {
2170 struct nl_mmap_req req;
2171
2172 /* Rings might consume more memory than queue limits, require
2173 * CAP_NET_ADMIN.
2174 */
2175 if (!capable(CAP_NET_ADMIN))
2176 return -EPERM;
2177 if (optlen < sizeof(req))
2178 return -EINVAL;
2179 if (copy_from_user(&req, optval, sizeof(req)))
2180 return -EFAULT;
2181 err = netlink_set_ring(sk, &req, false,
2182 optname == NETLINK_TX_RING);
2183 break;
2184 }
2185 #endif /* CONFIG_NETLINK_MMAP */
2186 default:
2187 err = -ENOPROTOOPT;
2188 }
2189 return err;
2190 }
2191
2192 static int netlink_getsockopt(struct socket *sock, int level, int optname,
2193 char __user *optval, int __user *optlen)
2194 {
2195 struct sock *sk = sock->sk;
2196 struct netlink_sock *nlk = nlk_sk(sk);
2197 int len, val, err;
2198
2199 if (level != SOL_NETLINK)
2200 return -ENOPROTOOPT;
2201
2202 if (get_user(len, optlen))
2203 return -EFAULT;
2204 if (len < 0)
2205 return -EINVAL;
2206
2207 switch (optname) {
2208 case NETLINK_PKTINFO:
2209 if (len < sizeof(int))
2210 return -EINVAL;
2211 len = sizeof(int);
2212 val = nlk->flags & NETLINK_RECV_PKTINFO ? 1 : 0;
2213 if (put_user(len, optlen) ||
2214 put_user(val, optval))
2215 return -EFAULT;
2216 err = 0;
2217 break;
2218 case NETLINK_BROADCAST_ERROR:
2219 if (len < sizeof(int))
2220 return -EINVAL;
2221 len = sizeof(int);
2222 val = nlk->flags & NETLINK_BROADCAST_SEND_ERROR ? 1 : 0;
2223 if (put_user(len, optlen) ||
2224 put_user(val, optval))
2225 return -EFAULT;
2226 err = 0;
2227 break;
2228 case NETLINK_NO_ENOBUFS:
2229 if (len < sizeof(int))
2230 return -EINVAL;
2231 len = sizeof(int);
2232 val = nlk->flags & NETLINK_RECV_NO_ENOBUFS ? 1 : 0;
2233 if (put_user(len, optlen) ||
2234 put_user(val, optval))
2235 return -EFAULT;
2236 err = 0;
2237 break;
2238 default:
2239 err = -ENOPROTOOPT;
2240 }
2241 return err;
2242 }
2243
2244 static void netlink_cmsg_recv_pktinfo(struct msghdr *msg, struct sk_buff *skb)
2245 {
2246 struct nl_pktinfo info;
2247
2248 info.group = NETLINK_CB(skb).dst_group;
2249 put_cmsg(msg, SOL_NETLINK, NETLINK_PKTINFO, sizeof(info), &info);
2250 }
2251
2252 static int netlink_sendmsg(struct kiocb *kiocb, struct socket *sock,
2253 struct msghdr *msg, size_t len)
2254 {
2255 struct sock_iocb *siocb = kiocb_to_siocb(kiocb);
2256 struct sock *sk = sock->sk;
2257 struct netlink_sock *nlk = nlk_sk(sk);
2258 DECLARE_SOCKADDR(struct sockaddr_nl *, addr, msg->msg_name);
2259 u32 dst_portid;
2260 u32 dst_group;
2261 struct sk_buff *skb;
2262 int err;
2263 struct scm_cookie scm;
2264 u32 netlink_skb_flags = 0;
2265
2266 if (msg->msg_flags&MSG_OOB)
2267 return -EOPNOTSUPP;
2268
2269 if (NULL == siocb->scm)
2270 siocb->scm = &scm;
2271
2272 err = scm_send(sock, msg, siocb->scm, true);
2273 if (err < 0)
2274 return err;
2275
2276 if (msg->msg_namelen) {
2277 err = -EINVAL;
2278 if (addr->nl_family != AF_NETLINK)
2279 goto out;
2280 dst_portid = addr->nl_pid;
2281 dst_group = ffs(addr->nl_groups);
2282 err = -EPERM;
2283 if ((dst_group || dst_portid) &&
2284 !netlink_allowed(sock, NL_CFG_F_NONROOT_SEND))
2285 goto out;
2286 netlink_skb_flags |= NETLINK_SKB_DST;
2287 } else {
2288 dst_portid = nlk->dst_portid;
2289 dst_group = nlk->dst_group;
2290 }
2291
2292 if (!nlk->portid) {
2293 err = netlink_autobind(sock);
2294 if (err)
2295 goto out;
2296 }
2297
2298 if (netlink_tx_is_mmaped(sk) &&
2299 msg->msg_iov->iov_base == NULL) {
2300 err = netlink_mmap_sendmsg(sk, msg, dst_portid, dst_group,
2301 siocb);
2302 goto out;
2303 }
2304
2305 err = -EMSGSIZE;
2306 if (len > sk->sk_sndbuf - 32)
2307 goto out;
2308 err = -ENOBUFS;
2309 skb = netlink_alloc_large_skb(len, dst_group);
2310 if (skb == NULL)
2311 goto out;
2312
2313 NETLINK_CB(skb).portid = nlk->portid;
2314 NETLINK_CB(skb).dst_group = dst_group;
2315 NETLINK_CB(skb).creds = siocb->scm->creds;
2316 NETLINK_CB(skb).flags = netlink_skb_flags;
2317
2318 err = -EFAULT;
2319 if (memcpy_fromiovec(skb_put(skb, len), msg->msg_iov, len)) {
2320 kfree_skb(skb);
2321 goto out;
2322 }
2323
2324 err = security_netlink_send(sk, skb);
2325 if (err) {
2326 kfree_skb(skb);
2327 goto out;
2328 }
2329
2330 if (dst_group) {
2331 atomic_inc(&skb->users);
2332 netlink_broadcast(sk, skb, dst_portid, dst_group, GFP_KERNEL);
2333 }
2334 err = netlink_unicast(sk, skb, dst_portid, msg->msg_flags&MSG_DONTWAIT);
2335
2336 out:
2337 scm_destroy(siocb->scm);
2338 return err;
2339 }
2340
2341 static int netlink_recvmsg(struct kiocb *kiocb, struct socket *sock,
2342 struct msghdr *msg, size_t len,
2343 int flags)
2344 {
2345 struct sock_iocb *siocb = kiocb_to_siocb(kiocb);
2346 struct scm_cookie scm;
2347 struct sock *sk = sock->sk;
2348 struct netlink_sock *nlk = nlk_sk(sk);
2349 int noblock = flags&MSG_DONTWAIT;
2350 size_t copied;
2351 struct sk_buff *skb, *data_skb;
2352 int err, ret;
2353
2354 if (flags&MSG_OOB)
2355 return -EOPNOTSUPP;
2356
2357 copied = 0;
2358
2359 skb = skb_recv_datagram(sk, flags, noblock, &err);
2360 if (skb == NULL)
2361 goto out;
2362
2363 data_skb = skb;
2364
2365 #ifdef CONFIG_COMPAT_NETLINK_MESSAGES
2366 if (unlikely(skb_shinfo(skb)->frag_list)) {
2367 /*
2368 * If this skb has a frag_list, then here that means that we
2369 * will have to use the frag_list skb's data for compat tasks
2370 * and the regular skb's data for normal (non-compat) tasks.
2371 *
2372 * If we need to send the compat skb, assign it to the
2373 * 'data_skb' variable so that it will be used below for data
2374 * copying. We keep 'skb' for everything else, including
2375 * freeing both later.
2376 */
2377 if (flags & MSG_CMSG_COMPAT)
2378 data_skb = skb_shinfo(skb)->frag_list;
2379 }
2380 #endif
2381
2382 /* Record the max length of recvmsg() calls for future allocations */
2383 nlk->max_recvmsg_len = max(nlk->max_recvmsg_len, len);
2384 nlk->max_recvmsg_len = min_t(size_t, nlk->max_recvmsg_len,
2385 16384);
2386
2387 copied = data_skb->len;
2388 if (len < copied) {
2389 msg->msg_flags |= MSG_TRUNC;
2390 copied = len;
2391 }
2392
2393 skb_reset_transport_header(data_skb);
2394 err = skb_copy_datagram_iovec(data_skb, 0, msg->msg_iov, copied);
2395
2396 if (msg->msg_name) {
2397 DECLARE_SOCKADDR(struct sockaddr_nl *, addr, msg->msg_name);
2398 addr->nl_family = AF_NETLINK;
2399 addr->nl_pad = 0;
2400 addr->nl_pid = NETLINK_CB(skb).portid;
2401 addr->nl_groups = netlink_group_mask(NETLINK_CB(skb).dst_group);
2402 msg->msg_namelen = sizeof(*addr);
2403 }
2404
2405 if (nlk->flags & NETLINK_RECV_PKTINFO)
2406 netlink_cmsg_recv_pktinfo(msg, skb);
2407
2408 if (NULL == siocb->scm) {
2409 memset(&scm, 0, sizeof(scm));
2410 siocb->scm = &scm;
2411 }
2412 siocb->scm->creds = *NETLINK_CREDS(skb);
2413 if (flags & MSG_TRUNC)
2414 copied = data_skb->len;
2415
2416 skb_free_datagram(sk, skb);
2417
2418 if (nlk->cb_running &&
2419 atomic_read(&sk->sk_rmem_alloc) <= sk->sk_rcvbuf / 2) {
2420 ret = netlink_dump(sk);
2421 if (ret) {
2422 sk->sk_err = -ret;
2423 sk->sk_error_report(sk);
2424 }
2425 }
2426
2427 scm_recv(sock, msg, siocb->scm, flags);
2428 out:
2429 netlink_rcv_wake(sk);
2430 return err ? : copied;
2431 }
2432
2433 static void netlink_data_ready(struct sock *sk)
2434 {
2435 BUG();
2436 }
2437
2438 /*
2439 * We export these functions to other modules. They provide a
2440 * complete set of kernel non-blocking support for message
2441 * queueing.
2442 */
2443
2444 struct sock *
2445 __netlink_kernel_create(struct net *net, int unit, struct module *module,
2446 struct netlink_kernel_cfg *cfg)
2447 {
2448 struct socket *sock;
2449 struct sock *sk;
2450 struct netlink_sock *nlk;
2451 struct listeners *listeners = NULL;
2452 struct mutex *cb_mutex = cfg ? cfg->cb_mutex : NULL;
2453 unsigned int groups;
2454
2455 BUG_ON(!nl_table);
2456
2457 if (unit < 0 || unit >= MAX_LINKS)
2458 return NULL;
2459
2460 if (sock_create_lite(PF_NETLINK, SOCK_DGRAM, unit, &sock))
2461 return NULL;
2462
2463 /*
2464 * We have to just have a reference on the net from sk, but don't
2465 * get_net it. Besides, we cannot get and then put the net here.
2466 * So we create one inside init_net and the move it to net.
2467 */
2468
2469 if (__netlink_create(&init_net, sock, cb_mutex, unit) < 0)
2470 goto out_sock_release_nosk;
2471
2472 sk = sock->sk;
2473 sk_change_net(sk, net);
2474
2475 if (!cfg || cfg->groups < 32)
2476 groups = 32;
2477 else
2478 groups = cfg->groups;
2479
2480 listeners = kzalloc(sizeof(*listeners) + NLGRPSZ(groups), GFP_KERNEL);
2481 if (!listeners)
2482 goto out_sock_release;
2483
2484 sk->sk_data_ready = netlink_data_ready;
2485 if (cfg && cfg->input)
2486 nlk_sk(sk)->netlink_rcv = cfg->input;
2487
2488 if (netlink_insert(sk, net, 0))
2489 goto out_sock_release;
2490
2491 nlk = nlk_sk(sk);
2492 nlk->flags |= NETLINK_KERNEL_SOCKET;
2493
2494 netlink_table_grab();
2495 if (!nl_table[unit].registered) {
2496 nl_table[unit].groups = groups;
2497 rcu_assign_pointer(nl_table[unit].listeners, listeners);
2498 nl_table[unit].cb_mutex = cb_mutex;
2499 nl_table[unit].module = module;
2500 if (cfg) {
2501 nl_table[unit].bind = cfg->bind;
2502 nl_table[unit].flags = cfg->flags;
2503 if (cfg->compare)
2504 nl_table[unit].compare = cfg->compare;
2505 }
2506 nl_table[unit].registered = 1;
2507 } else {
2508 kfree(listeners);
2509 nl_table[unit].registered++;
2510 }
2511 netlink_table_ungrab();
2512 return sk;
2513
2514 out_sock_release:
2515 kfree(listeners);
2516 netlink_kernel_release(sk);
2517 return NULL;
2518
2519 out_sock_release_nosk:
2520 sock_release(sock);
2521 return NULL;
2522 }
2523 EXPORT_SYMBOL(__netlink_kernel_create);
2524
2525 void
2526 netlink_kernel_release(struct sock *sk)
2527 {
2528 sk_release_kernel(sk);
2529 }
2530 EXPORT_SYMBOL(netlink_kernel_release);
2531
2532 int __netlink_change_ngroups(struct sock *sk, unsigned int groups)
2533 {
2534 struct listeners *new, *old;
2535 struct netlink_table *tbl = &nl_table[sk->sk_protocol];
2536
2537 if (groups < 32)
2538 groups = 32;
2539
2540 if (NLGRPSZ(tbl->groups) < NLGRPSZ(groups)) {
2541 new = kzalloc(sizeof(*new) + NLGRPSZ(groups), GFP_ATOMIC);
2542 if (!new)
2543 return -ENOMEM;
2544 old = nl_deref_protected(tbl->listeners);
2545 memcpy(new->masks, old->masks, NLGRPSZ(tbl->groups));
2546 rcu_assign_pointer(tbl->listeners, new);
2547
2548 kfree_rcu(old, rcu);
2549 }
2550 tbl->groups = groups;
2551
2552 return 0;
2553 }
2554
2555 /**
2556 * netlink_change_ngroups - change number of multicast groups
2557 *
2558 * This changes the number of multicast groups that are available
2559 * on a certain netlink family. Note that it is not possible to
2560 * change the number of groups to below 32. Also note that it does
2561 * not implicitly call netlink_clear_multicast_users() when the
2562 * number of groups is reduced.
2563 *
2564 * @sk: The kernel netlink socket, as returned by netlink_kernel_create().
2565 * @groups: The new number of groups.
2566 */
2567 int netlink_change_ngroups(struct sock *sk, unsigned int groups)
2568 {
2569 int err;
2570
2571 netlink_table_grab();
2572 err = __netlink_change_ngroups(sk, groups);
2573 netlink_table_ungrab();
2574
2575 return err;
2576 }
2577
2578 void __netlink_clear_multicast_users(struct sock *ksk, unsigned int group)
2579 {
2580 struct sock *sk;
2581 struct netlink_table *tbl = &nl_table[ksk->sk_protocol];
2582
2583 sk_for_each_bound(sk, &tbl->mc_list)
2584 netlink_update_socket_mc(nlk_sk(sk), group, 0);
2585 }
2586
2587 struct nlmsghdr *
2588 __nlmsg_put(struct sk_buff *skb, u32 portid, u32 seq, int type, int len, int flags)
2589 {
2590 struct nlmsghdr *nlh;
2591 int size = nlmsg_msg_size(len);
2592
2593 nlh = (struct nlmsghdr *)skb_put(skb, NLMSG_ALIGN(size));
2594 nlh->nlmsg_type = type;
2595 nlh->nlmsg_len = size;
2596 nlh->nlmsg_flags = flags;
2597 nlh->nlmsg_pid = portid;
2598 nlh->nlmsg_seq = seq;
2599 if (!__builtin_constant_p(size) || NLMSG_ALIGN(size) - size != 0)
2600 memset(nlmsg_data(nlh) + len, 0, NLMSG_ALIGN(size) - size);
2601 return nlh;
2602 }
2603 EXPORT_SYMBOL(__nlmsg_put);
2604
2605 /*
2606 * It looks a bit ugly.
2607 * It would be better to create kernel thread.
2608 */
2609
2610 static int netlink_dump(struct sock *sk)
2611 {
2612 struct netlink_sock *nlk = nlk_sk(sk);
2613 struct netlink_callback *cb;
2614 struct sk_buff *skb = NULL;
2615 struct nlmsghdr *nlh;
2616 int len, err = -ENOBUFS;
2617 int alloc_size;
2618
2619 mutex_lock(nlk->cb_mutex);
2620 if (!nlk->cb_running) {
2621 err = -EINVAL;
2622 goto errout_skb;
2623 }
2624
2625 cb = &nlk->cb;
2626 alloc_size = max_t(int, cb->min_dump_alloc, NLMSG_GOODSIZE);
2627
2628 if (!netlink_rx_is_mmaped(sk) &&
2629 atomic_read(&sk->sk_rmem_alloc) >= sk->sk_rcvbuf)
2630 goto errout_skb;
2631
2632 /* NLMSG_GOODSIZE is small to avoid high order allocations being
2633 * required, but it makes sense to _attempt_ a 16K bytes allocation
2634 * to reduce number of system calls on dump operations, if user
2635 * ever provided a big enough buffer.
2636 */
2637 if (alloc_size < nlk->max_recvmsg_len) {
2638 skb = netlink_alloc_skb(sk,
2639 nlk->max_recvmsg_len,
2640 nlk->portid,
2641 GFP_KERNEL |
2642 __GFP_NOWARN |
2643 __GFP_NORETRY);
2644 /* available room should be exact amount to avoid MSG_TRUNC */
2645 if (skb)
2646 skb_reserve(skb, skb_tailroom(skb) -
2647 nlk->max_recvmsg_len);
2648 }
2649 if (!skb)
2650 skb = netlink_alloc_skb(sk, alloc_size, nlk->portid,
2651 GFP_KERNEL);
2652 if (!skb)
2653 goto errout_skb;
2654 netlink_skb_set_owner_r(skb, sk);
2655
2656 len = cb->dump(skb, cb);
2657
2658 if (len > 0) {
2659 mutex_unlock(nlk->cb_mutex);
2660
2661 if (sk_filter(sk, skb))
2662 kfree_skb(skb);
2663 else
2664 __netlink_sendskb(sk, skb);
2665 return 0;
2666 }
2667
2668 nlh = nlmsg_put_answer(skb, cb, NLMSG_DONE, sizeof(len), NLM_F_MULTI);
2669 if (!nlh)
2670 goto errout_skb;
2671
2672 nl_dump_check_consistent(cb, nlh);
2673
2674 memcpy(nlmsg_data(nlh), &len, sizeof(len));
2675
2676 if (sk_filter(sk, skb))
2677 kfree_skb(skb);
2678 else
2679 __netlink_sendskb(sk, skb);
2680
2681 if (cb->done)
2682 cb->done(cb);
2683
2684 nlk->cb_running = false;
2685 mutex_unlock(nlk->cb_mutex);
2686 module_put(cb->module);
2687 consume_skb(cb->skb);
2688 return 0;
2689
2690 errout_skb:
2691 mutex_unlock(nlk->cb_mutex);
2692 kfree_skb(skb);
2693 return err;
2694 }
2695
2696 int __netlink_dump_start(struct sock *ssk, struct sk_buff *skb,
2697 const struct nlmsghdr *nlh,
2698 struct netlink_dump_control *control)
2699 {
2700 struct netlink_callback *cb;
2701 struct sock *sk;
2702 struct netlink_sock *nlk;
2703 int ret;
2704
2705 /* Memory mapped dump requests need to be copied to avoid looping
2706 * on the pending state in netlink_mmap_sendmsg() while the CB hold
2707 * a reference to the skb.
2708 */
2709 if (netlink_skb_is_mmaped(skb)) {
2710 skb = skb_copy(skb, GFP_KERNEL);
2711 if (skb == NULL)
2712 return -ENOBUFS;
2713 } else
2714 atomic_inc(&skb->users);
2715
2716 sk = netlink_lookup(sock_net(ssk), ssk->sk_protocol, NETLINK_CB(skb).portid);
2717 if (sk == NULL) {
2718 ret = -ECONNREFUSED;
2719 goto error_free;
2720 }
2721
2722 nlk = nlk_sk(sk);
2723 mutex_lock(nlk->cb_mutex);
2724 /* A dump is in progress... */
2725 if (nlk->cb_running) {
2726 ret = -EBUSY;
2727 goto error_unlock;
2728 }
2729 /* add reference of module which cb->dump belongs to */
2730 if (!try_module_get(control->module)) {
2731 ret = -EPROTONOSUPPORT;
2732 goto error_unlock;
2733 }
2734
2735 cb = &nlk->cb;
2736 memset(cb, 0, sizeof(*cb));
2737 cb->dump = control->dump;
2738 cb->done = control->done;
2739 cb->nlh = nlh;
2740 cb->data = control->data;
2741 cb->module = control->module;
2742 cb->min_dump_alloc = control->min_dump_alloc;
2743 cb->skb = skb;
2744
2745 nlk->cb_running = true;
2746
2747 mutex_unlock(nlk->cb_mutex);
2748
2749 ret = netlink_dump(sk);
2750 sock_put(sk);
2751
2752 if (ret)
2753 return ret;
2754
2755 /* We successfully started a dump, by returning -EINTR we
2756 * signal not to send ACK even if it was requested.
2757 */
2758 return -EINTR;
2759
2760 error_unlock:
2761 sock_put(sk);
2762 mutex_unlock(nlk->cb_mutex);
2763 error_free:
2764 kfree_skb(skb);
2765 return ret;
2766 }
2767 EXPORT_SYMBOL(__netlink_dump_start);
2768
2769 void netlink_ack(struct sk_buff *in_skb, struct nlmsghdr *nlh, int err)
2770 {
2771 struct sk_buff *skb;
2772 struct nlmsghdr *rep;
2773 struct nlmsgerr *errmsg;
2774 size_t payload = sizeof(*errmsg);
2775
2776 /* error messages get the original request appened */
2777 if (err)
2778 payload += nlmsg_len(nlh);
2779
2780 skb = netlink_alloc_skb(in_skb->sk, nlmsg_total_size(payload),
2781 NETLINK_CB(in_skb).portid, GFP_KERNEL);
2782 if (!skb) {
2783 struct sock *sk;
2784
2785 sk = netlink_lookup(sock_net(in_skb->sk),
2786 in_skb->sk->sk_protocol,
2787 NETLINK_CB(in_skb).portid);
2788 if (sk) {
2789 sk->sk_err = ENOBUFS;
2790 sk->sk_error_report(sk);
2791 sock_put(sk);
2792 }
2793 return;
2794 }
2795
2796 rep = __nlmsg_put(skb, NETLINK_CB(in_skb).portid, nlh->nlmsg_seq,
2797 NLMSG_ERROR, payload, 0);
2798 errmsg = nlmsg_data(rep);
2799 errmsg->error = err;
2800 memcpy(&errmsg->msg, nlh, err ? nlh->nlmsg_len : sizeof(*nlh));
2801 netlink_unicast(in_skb->sk, skb, NETLINK_CB(in_skb).portid, MSG_DONTWAIT);
2802 }
2803 EXPORT_SYMBOL(netlink_ack);
2804
2805 int netlink_rcv_skb(struct sk_buff *skb, int (*cb)(struct sk_buff *,
2806 struct nlmsghdr *))
2807 {
2808 struct nlmsghdr *nlh;
2809 int err;
2810
2811 while (skb->len >= nlmsg_total_size(0)) {
2812 int msglen;
2813
2814 nlh = nlmsg_hdr(skb);
2815 err = 0;
2816
2817 if (nlh->nlmsg_len < NLMSG_HDRLEN || skb->len < nlh->nlmsg_len)
2818 return 0;
2819
2820 /* Only requests are handled by the kernel */
2821 if (!(nlh->nlmsg_flags & NLM_F_REQUEST))
2822 goto ack;
2823
2824 /* Skip control messages */
2825 if (nlh->nlmsg_type < NLMSG_MIN_TYPE)
2826 goto ack;
2827
2828 err = cb(skb, nlh);
2829 if (err == -EINTR)
2830 goto skip;
2831
2832 ack:
2833 if (nlh->nlmsg_flags & NLM_F_ACK || err)
2834 netlink_ack(skb, nlh, err);
2835
2836 skip:
2837 msglen = NLMSG_ALIGN(nlh->nlmsg_len);
2838 if (msglen > skb->len)
2839 msglen = skb->len;
2840 skb_pull(skb, msglen);
2841 }
2842
2843 return 0;
2844 }
2845 EXPORT_SYMBOL(netlink_rcv_skb);
2846
2847 /**
2848 * nlmsg_notify - send a notification netlink message
2849 * @sk: netlink socket to use
2850 * @skb: notification message
2851 * @portid: destination netlink portid for reports or 0
2852 * @group: destination multicast group or 0
2853 * @report: 1 to report back, 0 to disable
2854 * @flags: allocation flags
2855 */
2856 int nlmsg_notify(struct sock *sk, struct sk_buff *skb, u32 portid,
2857 unsigned int group, int report, gfp_t flags)
2858 {
2859 int err = 0;
2860
2861 if (group) {
2862 int exclude_portid = 0;
2863
2864 if (report) {
2865 atomic_inc(&skb->users);
2866 exclude_portid = portid;
2867 }
2868
2869 /* errors reported via destination sk->sk_err, but propagate
2870 * delivery errors if NETLINK_BROADCAST_ERROR flag is set */
2871 err = nlmsg_multicast(sk, skb, exclude_portid, group, flags);
2872 }
2873
2874 if (report) {
2875 int err2;
2876
2877 err2 = nlmsg_unicast(sk, skb, portid);
2878 if (!err || err == -ESRCH)
2879 err = err2;
2880 }
2881
2882 return err;
2883 }
2884 EXPORT_SYMBOL(nlmsg_notify);
2885
2886 #ifdef CONFIG_PROC_FS
2887 struct nl_seq_iter {
2888 struct seq_net_private p;
2889 int link;
2890 int hash_idx;
2891 };
2892
2893 static struct sock *netlink_seq_socket_idx(struct seq_file *seq, loff_t pos)
2894 {
2895 struct nl_seq_iter *iter = seq->private;
2896 int i, j;
2897 struct netlink_sock *nlk;
2898 struct sock *s;
2899 loff_t off = 0;
2900
2901 for (i = 0; i < MAX_LINKS; i++) {
2902 struct rhashtable *ht = &nl_table[i].hash;
2903 const struct bucket_table *tbl = rht_dereference_rcu(ht->tbl, ht);
2904
2905 for (j = 0; j < tbl->size; j++) {
2906 rht_for_each_entry_rcu(nlk, tbl->buckets[j], node) {
2907 s = (struct sock *)nlk;
2908
2909 if (sock_net(s) != seq_file_net(seq))
2910 continue;
2911 if (off == pos) {
2912 iter->link = i;
2913 iter->hash_idx = j;
2914 return s;
2915 }
2916 ++off;
2917 }
2918 }
2919 }
2920 return NULL;
2921 }
2922
2923 static void *netlink_seq_start(struct seq_file *seq, loff_t *pos)
2924 __acquires(RCU)
2925 {
2926 rcu_read_lock();
2927 return *pos ? netlink_seq_socket_idx(seq, *pos - 1) : SEQ_START_TOKEN;
2928 }
2929
2930 static void *netlink_seq_next(struct seq_file *seq, void *v, loff_t *pos)
2931 {
2932 struct netlink_sock *nlk;
2933 struct nl_seq_iter *iter;
2934 struct net *net;
2935 int i, j;
2936
2937 ++*pos;
2938
2939 if (v == SEQ_START_TOKEN)
2940 return netlink_seq_socket_idx(seq, 0);
2941
2942 net = seq_file_net(seq);
2943 iter = seq->private;
2944 nlk = v;
2945
2946 rht_for_each_entry_rcu(nlk, nlk->node.next, node)
2947 if (net_eq(sock_net((struct sock *)nlk), net))
2948 return nlk;
2949
2950 i = iter->link;
2951 j = iter->hash_idx + 1;
2952
2953 do {
2954 struct rhashtable *ht = &nl_table[i].hash;
2955 const struct bucket_table *tbl = rht_dereference_rcu(ht->tbl, ht);
2956
2957 for (; j < tbl->size; j++) {
2958 rht_for_each_entry_rcu(nlk, tbl->buckets[j], node) {
2959 if (net_eq(sock_net((struct sock *)nlk), net)) {
2960 iter->link = i;
2961 iter->hash_idx = j;
2962 return nlk;
2963 }
2964 }
2965 }
2966
2967 j = 0;
2968 } while (++i < MAX_LINKS);
2969
2970 return NULL;
2971 }
2972
2973 static void netlink_seq_stop(struct seq_file *seq, void *v)
2974 __releases(RCU)
2975 {
2976 rcu_read_unlock();
2977 }
2978
2979
2980 static int netlink_seq_show(struct seq_file *seq, void *v)
2981 {
2982 if (v == SEQ_START_TOKEN) {
2983 seq_puts(seq,
2984 "sk Eth Pid Groups "
2985 "Rmem Wmem Dump Locks Drops Inode\n");
2986 } else {
2987 struct sock *s = v;
2988 struct netlink_sock *nlk = nlk_sk(s);
2989
2990 seq_printf(seq, "%pK %-3d %-6u %08x %-8d %-8d %d %-8d %-8d %-8lu\n",
2991 s,
2992 s->sk_protocol,
2993 nlk->portid,
2994 nlk->groups ? (u32)nlk->groups[0] : 0,
2995 sk_rmem_alloc_get(s),
2996 sk_wmem_alloc_get(s),
2997 nlk->cb_running,
2998 atomic_read(&s->sk_refcnt),
2999 atomic_read(&s->sk_drops),
3000 sock_i_ino(s)
3001 );
3002
3003 }
3004 return 0;
3005 }
3006
3007 static const struct seq_operations netlink_seq_ops = {
3008 .start = netlink_seq_start,
3009 .next = netlink_seq_next,
3010 .stop = netlink_seq_stop,
3011 .show = netlink_seq_show,
3012 };
3013
3014
3015 static int netlink_seq_open(struct inode *inode, struct file *file)
3016 {
3017 return seq_open_net(inode, file, &netlink_seq_ops,
3018 sizeof(struct nl_seq_iter));
3019 }
3020
3021 static const struct file_operations netlink_seq_fops = {
3022 .owner = THIS_MODULE,
3023 .open = netlink_seq_open,
3024 .read = seq_read,
3025 .llseek = seq_lseek,
3026 .release = seq_release_net,
3027 };
3028
3029 #endif
3030
3031 int netlink_register_notifier(struct notifier_block *nb)
3032 {
3033 return atomic_notifier_chain_register(&netlink_chain, nb);
3034 }
3035 EXPORT_SYMBOL(netlink_register_notifier);
3036
3037 int netlink_unregister_notifier(struct notifier_block *nb)
3038 {
3039 return atomic_notifier_chain_unregister(&netlink_chain, nb);
3040 }
3041 EXPORT_SYMBOL(netlink_unregister_notifier);
3042
3043 static const struct proto_ops netlink_ops = {
3044 .family = PF_NETLINK,
3045 .owner = THIS_MODULE,
3046 .release = netlink_release,
3047 .bind = netlink_bind,
3048 .connect = netlink_connect,
3049 .socketpair = sock_no_socketpair,
3050 .accept = sock_no_accept,
3051 .getname = netlink_getname,
3052 .poll = netlink_poll,
3053 .ioctl = sock_no_ioctl,
3054 .listen = sock_no_listen,
3055 .shutdown = sock_no_shutdown,
3056 .setsockopt = netlink_setsockopt,
3057 .getsockopt = netlink_getsockopt,
3058 .sendmsg = netlink_sendmsg,
3059 .recvmsg = netlink_recvmsg,
3060 .mmap = netlink_mmap,
3061 .sendpage = sock_no_sendpage,
3062 };
3063
3064 static const struct net_proto_family netlink_family_ops = {
3065 .family = PF_NETLINK,
3066 .create = netlink_create,
3067 .owner = THIS_MODULE, /* for consistency 8) */
3068 };
3069
3070 static int __net_init netlink_net_init(struct net *net)
3071 {
3072 #ifdef CONFIG_PROC_FS
3073 if (!proc_create("netlink", 0, net->proc_net, &netlink_seq_fops))
3074 return -ENOMEM;
3075 #endif
3076 return 0;
3077 }
3078
3079 static void __net_exit netlink_net_exit(struct net *net)
3080 {
3081 #ifdef CONFIG_PROC_FS
3082 remove_proc_entry("netlink", net->proc_net);
3083 #endif
3084 }
3085
3086 static void __init netlink_add_usersock_entry(void)
3087 {
3088 struct listeners *listeners;
3089 int groups = 32;
3090
3091 listeners = kzalloc(sizeof(*listeners) + NLGRPSZ(groups), GFP_KERNEL);
3092 if (!listeners)
3093 panic("netlink_add_usersock_entry: Cannot allocate listeners\n");
3094
3095 netlink_table_grab();
3096
3097 nl_table[NETLINK_USERSOCK].groups = groups;
3098 rcu_assign_pointer(nl_table[NETLINK_USERSOCK].listeners, listeners);
3099 nl_table[NETLINK_USERSOCK].module = THIS_MODULE;
3100 nl_table[NETLINK_USERSOCK].registered = 1;
3101 nl_table[NETLINK_USERSOCK].flags = NL_CFG_F_NONROOT_SEND;
3102
3103 netlink_table_ungrab();
3104 }
3105
3106 static struct pernet_operations __net_initdata netlink_net_ops = {
3107 .init = netlink_net_init,
3108 .exit = netlink_net_exit,
3109 };
3110
3111 static int __init netlink_proto_init(void)
3112 {
3113 int i;
3114 int err = proto_register(&netlink_proto, 0);
3115 struct rhashtable_params ht_params = {
3116 .head_offset = offsetof(struct netlink_sock, node),
3117 .key_offset = offsetof(struct netlink_sock, portid),
3118 .key_len = sizeof(u32), /* portid */
3119 .hashfn = arch_fast_hash,
3120 .max_shift = 16, /* 64K */
3121 .grow_decision = rht_grow_above_75,
3122 .shrink_decision = rht_shrink_below_30,
3123 .mutex_is_held = lockdep_nl_sk_hash_is_held,
3124 };
3125
3126 if (err != 0)
3127 goto out;
3128
3129 BUILD_BUG_ON(sizeof(struct netlink_skb_parms) > FIELD_SIZEOF(struct sk_buff, cb));
3130
3131 nl_table = kcalloc(MAX_LINKS, sizeof(*nl_table), GFP_KERNEL);
3132 if (!nl_table)
3133 goto panic;
3134
3135 for (i = 0; i < MAX_LINKS; i++) {
3136 if (rhashtable_init(&nl_table[i].hash, &ht_params) < 0) {
3137 while (--i > 0)
3138 rhashtable_destroy(&nl_table[i].hash);
3139 kfree(nl_table);
3140 goto panic;
3141 }
3142 }
3143
3144 INIT_LIST_HEAD(&netlink_tap_all);
3145
3146 netlink_add_usersock_entry();
3147
3148 sock_register(&netlink_family_ops);
3149 register_pernet_subsys(&netlink_net_ops);
3150 /* The netlink device handler may be needed early. */
3151 rtnetlink_init();
3152 out:
3153 return err;
3154 panic:
3155 panic("netlink_init: Cannot allocate nl_table\n");
3156 }
3157
3158 core_initcall(netlink_proto_init);
Linux with added FPC-III support