projects / GitHub / mt8127 / android_kernel_alcatel_ttab.git / blob
? search:


6560635fd25cd4bf25cf7bdbcce50416216c08dd
[GitHub/mt8127/android_kernel_alcatel_ttab.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 *
7 * This program is free software; you can redistribute it and/or
8 * modify it under the terms of the GNU General Public License
9 * as published by the Free Software Foundation; either version
10 * 2 of the License, or (at your option) any later version.
11 *
12 * Tue Jun 26 14:36:48 MEST 2001 Herbert "herp" Rosmanith
13 * added netlink_proto_exit
14 * Tue Jan 22 18:32:44 BRST 2002 Arnaldo C. de Melo <acme@conectiva.com.br>
15 * use nlk_sk, as sk->protinfo is on a diet 8)
16 * Fri Jul 22 19:51:12 MEST 2005 Harald Welte <laforge@gnumonks.org>
17 * - inc module use count of module that owns
18 * the kernel socket in case userspace opens
19 * socket of same protocol
20 * - remove all module support, since netlink is
21 * mandatory if CONFIG_NET=y these days
22 */
23
24 #include <linux/module.h>
25
26 #include <linux/capability.h>
27 #include <linux/kernel.h>
28 #include <linux/init.h>
29 #include <linux/signal.h>
30 #include <linux/sched.h>
31 #include <linux/errno.h>
32 #include <linux/string.h>
33 #include <linux/stat.h>
34 #include <linux/socket.h>
35 #include <linux/un.h>
36 #include <linux/fcntl.h>
37 #include <linux/termios.h>
38 #include <linux/sockios.h>
39 #include <linux/net.h>
40 #include <linux/fs.h>
41 #include <linux/slab.h>
42 #include <asm/uaccess.h>
43 #include <linux/skbuff.h>
44 #include <linux/netdevice.h>
45 #include <linux/rtnetlink.h>
46 #include <linux/proc_fs.h>
47 #include <linux/seq_file.h>
48 #include <linux/notifier.h>
49 #include <linux/security.h>
50 #include <linux/jhash.h>
51 #include <linux/jiffies.h>
52 #include <linux/random.h>
53 #include <linux/bitops.h>
54 #include <linux/mm.h>
55 #include <linux/types.h>
56 #include <linux/audit.h>
57 #include <linux/mutex.h>
58 #include <linux/vmalloc.h>
59 #include <asm/cacheflush.h>
60
61 #include <net/net_namespace.h>
62 #include <net/sock.h>
63 #include <net/scm.h>
64 #include <net/netlink.h>
65
66 #include "af_netlink.h"
67
68 struct listeners {
69 struct rcu_head rcu;
70 unsigned long masks[0];
71 };
72
73 /* state bits */
74 #define NETLINK_CONGESTED 0x0
75
76 /* flags */
77 #define NETLINK_KERNEL_SOCKET 0x1
78 #define NETLINK_RECV_PKTINFO 0x2
79 #define NETLINK_BROADCAST_SEND_ERROR 0x4
80 #define NETLINK_RECV_NO_ENOBUFS 0x8
81
82 static inline int netlink_is_kernel(struct sock *sk)
83 {
84 return nlk_sk(sk)->flags & NETLINK_KERNEL_SOCKET;
85 }
86
87 struct netlink_table *nl_table;
88 EXPORT_SYMBOL_GPL(nl_table);
89
90 static DECLARE_WAIT_QUEUE_HEAD(nl_table_wait);
91
92 static int netlink_dump(struct sock *sk);
93 static void netlink_skb_destructor(struct sk_buff *skb);
94
95 DEFINE_RWLOCK(nl_table_lock);
96 EXPORT_SYMBOL_GPL(nl_table_lock);
97 static atomic_t nl_table_users = ATOMIC_INIT(0);
98
99 #define nl_deref_protected(X) rcu_dereference_protected(X, lockdep_is_held(&nl_table_lock));
100
101 static ATOMIC_NOTIFIER_HEAD(netlink_chain);
102
103 static inline u32 netlink_group_mask(u32 group)
104 {
105 return group ? 1 << (group - 1) : 0;
106 }
107
108 static inline struct hlist_head *nl_portid_hashfn(struct nl_portid_hash *hash, u32 portid)
109 {
110 return &hash->table[jhash_1word(portid, hash->rnd) & hash->mask];
111 }
112
113 #ifdef CONFIG_NETLINK_MMAP
114 static bool netlink_skb_is_mmaped(const struct sk_buff *skb)
115 {
116 return NETLINK_CB(skb).flags & NETLINK_SKB_MMAPED;
117 }
118
119 static __pure struct page *pgvec_to_page(const void *addr)
120 {
121 if (is_vmalloc_addr(addr))
122 return vmalloc_to_page(addr);
123 else
124 return virt_to_page(addr);
125 }
126
127 static void free_pg_vec(void **pg_vec, unsigned int order, unsigned int len)
128 {
129 unsigned int i;
130
131 for (i = 0; i < len; i++) {
132 if (pg_vec[i] != NULL) {
133 if (is_vmalloc_addr(pg_vec[i]))
134 vfree(pg_vec[i]);
135 else
136 free_pages((unsigned long)pg_vec[i], order);
137 }
138 }
139 kfree(pg_vec);
140 }
141
142 static void *alloc_one_pg_vec_page(unsigned long order)
143 {
144 void *buffer;
145 gfp_t gfp_flags = GFP_KERNEL | __GFP_COMP | __GFP_ZERO |
146 __GFP_NOWARN | __GFP_NORETRY;
147
148 buffer = (void *)__get_free_pages(gfp_flags, order);
149 if (buffer != NULL)
150 return buffer;
151
152 buffer = vzalloc((1 << order) * PAGE_SIZE);
153 if (buffer != NULL)
154 return buffer;
155
156 gfp_flags &= ~__GFP_NORETRY;
157 return (void *)__get_free_pages(gfp_flags, order);
158 }
159
160 static void **alloc_pg_vec(struct netlink_sock *nlk,
161 struct nl_mmap_req *req, unsigned int order)
162 {
163 unsigned int block_nr = req->nm_block_nr;
164 unsigned int i;
165 void **pg_vec, *ptr;
166
167 pg_vec = kcalloc(block_nr, sizeof(void *), GFP_KERNEL);
168 if (pg_vec == NULL)
169 return NULL;
170
171 for (i = 0; i < block_nr; i++) {
172 pg_vec[i] = ptr = alloc_one_pg_vec_page(order);
173 if (pg_vec[i] == NULL)
174 goto err1;
175 }
176
177 return pg_vec;
178 err1:
179 free_pg_vec(pg_vec, order, block_nr);
180 return NULL;
181 }
182
183 static int netlink_set_ring(struct sock *sk, struct nl_mmap_req *req,
184 bool closing, bool tx_ring)
185 {
186 struct netlink_sock *nlk = nlk_sk(sk);
187 struct netlink_ring *ring;
188 struct sk_buff_head *queue;
189 void **pg_vec = NULL;
190 unsigned int order = 0;
191 int err;
192
193 ring = tx_ring ? &nlk->tx_ring : &nlk->rx_ring;
194 queue = tx_ring ? &sk->sk_write_queue : &sk->sk_receive_queue;
195
196 if (!closing) {
197 if (atomic_read(&nlk->mapped))
198 return -EBUSY;
199 if (atomic_read(&ring->pending))
200 return -EBUSY;
201 }
202
203 if (req->nm_block_nr) {
204 if (ring->pg_vec != NULL)
205 return -EBUSY;
206
207 if ((int)req->nm_block_size <= 0)
208 return -EINVAL;
209 if (!IS_ALIGNED(req->nm_block_size, PAGE_SIZE))
210 return -EINVAL;
211 if (req->nm_frame_size < NL_MMAP_HDRLEN)
212 return -EINVAL;
213 if (!IS_ALIGNED(req->nm_frame_size, NL_MMAP_MSG_ALIGNMENT))
214 return -EINVAL;
215
216 ring->frames_per_block = req->nm_block_size /
217 req->nm_frame_size;
218 if (ring->frames_per_block == 0)
219 return -EINVAL;
220 if (ring->frames_per_block * req->nm_block_nr !=
221 req->nm_frame_nr)
222 return -EINVAL;
223
224 order = get_order(req->nm_block_size);
225 pg_vec = alloc_pg_vec(nlk, req, order);
226 if (pg_vec == NULL)
227 return -ENOMEM;
228 } else {
229 if (req->nm_frame_nr)
230 return -EINVAL;
231 }
232
233 err = -EBUSY;
234 mutex_lock(&nlk->pg_vec_lock);
235 if (closing || atomic_read(&nlk->mapped) == 0) {
236 err = 0;
237 spin_lock_bh(&queue->lock);
238
239 ring->frame_max = req->nm_frame_nr - 1;
240 ring->head = 0;
241 ring->frame_size = req->nm_frame_size;
242 ring->pg_vec_pages = req->nm_block_size / PAGE_SIZE;
243
244 swap(ring->pg_vec_len, req->nm_block_nr);
245 swap(ring->pg_vec_order, order);
246 swap(ring->pg_vec, pg_vec);
247
248 __skb_queue_purge(queue);
249 spin_unlock_bh(&queue->lock);
250
251 WARN_ON(atomic_read(&nlk->mapped));
252 }
253 mutex_unlock(&nlk->pg_vec_lock);
254
255 if (pg_vec)
256 free_pg_vec(pg_vec, order, req->nm_block_nr);
257 return err;
258 }
259
260 static void netlink_mm_open(struct vm_area_struct *vma)
261 {
262 struct file *file = vma->vm_file;
263 struct socket *sock = file->private_data;
264 struct sock *sk = sock->sk;
265
266 if (sk)
267 atomic_inc(&nlk_sk(sk)->mapped);
268 }
269
270 static void netlink_mm_close(struct vm_area_struct *vma)
271 {
272 struct file *file = vma->vm_file;
273 struct socket *sock = file->private_data;
274 struct sock *sk = sock->sk;
275
276 if (sk)
277 atomic_dec(&nlk_sk(sk)->mapped);
278 }
279
280 static const struct vm_operations_struct netlink_mmap_ops = {
281 .open = netlink_mm_open,
282 .close = netlink_mm_close,
283 };
284
285 static int netlink_mmap(struct file *file, struct socket *sock,
286 struct vm_area_struct *vma)
287 {
288 struct sock *sk = sock->sk;
289 struct netlink_sock *nlk = nlk_sk(sk);
290 struct netlink_ring *ring;
291 unsigned long start, size, expected;
292 unsigned int i;
293 int err = -EINVAL;
294
295 if (vma->vm_pgoff)
296 return -EINVAL;
297
298 mutex_lock(&nlk->pg_vec_lock);
299
300 expected = 0;
301 for (ring = &nlk->rx_ring; ring <= &nlk->tx_ring; ring++) {
302 if (ring->pg_vec == NULL)
303 continue;
304 expected += ring->pg_vec_len * ring->pg_vec_pages * PAGE_SIZE;
305 }
306
307 if (expected == 0)
308 goto out;
309
310 size = vma->vm_end - vma->vm_start;
311 if (size != expected)
312 goto out;
313
314 start = vma->vm_start;
315 for (ring = &nlk->rx_ring; ring <= &nlk->tx_ring; ring++) {
316 if (ring->pg_vec == NULL)
317 continue;
318
319 for (i = 0; i < ring->pg_vec_len; i++) {
320 struct page *page;
321 void *kaddr = ring->pg_vec[i];
322 unsigned int pg_num;
323
324 for (pg_num = 0; pg_num < ring->pg_vec_pages; pg_num++) {
325 page = pgvec_to_page(kaddr);
326 err = vm_insert_page(vma, start, page);
327 if (err < 0)
328 goto out;
329 start += PAGE_SIZE;
330 kaddr += PAGE_SIZE;
331 }
332 }
333 }
334
335 atomic_inc(&nlk->mapped);
336 vma->vm_ops = &netlink_mmap_ops;
337 err = 0;
338 out:
339 mutex_unlock(&nlk->pg_vec_lock);
340 return 0;
341 }
342
343 static void netlink_frame_flush_dcache(const struct nl_mmap_hdr *hdr)
344 {
345 #if ARCH_IMPLEMENTS_FLUSH_DCACHE_PAGE == 1
346 struct page *p_start, *p_end;
347
348 /* First page is flushed through netlink_{get,set}_status */
349 p_start = pgvec_to_page(hdr + PAGE_SIZE);
350 p_end = pgvec_to_page((void *)hdr + NL_MMAP_MSG_HDRLEN + hdr->nm_len - 1);
351 while (p_start <= p_end) {
352 flush_dcache_page(p_start);
353 p_start++;
354 }
355 #endif
356 }
357
358 static enum nl_mmap_status netlink_get_status(const struct nl_mmap_hdr *hdr)
359 {
360 smp_rmb();
361 flush_dcache_page(pgvec_to_page(hdr));
362 return hdr->nm_status;
363 }
364
365 static void netlink_set_status(struct nl_mmap_hdr *hdr,
366 enum nl_mmap_status status)
367 {
368 hdr->nm_status = status;
369 flush_dcache_page(pgvec_to_page(hdr));
370 smp_wmb();
371 }
372
373 static struct nl_mmap_hdr *
374 __netlink_lookup_frame(const struct netlink_ring *ring, unsigned int pos)
375 {
376 unsigned int pg_vec_pos, frame_off;
377
378 pg_vec_pos = pos / ring->frames_per_block;
379 frame_off = pos % ring->frames_per_block;
380
381 return ring->pg_vec[pg_vec_pos] + (frame_off * ring->frame_size);
382 }
383
384 static struct nl_mmap_hdr *
385 netlink_lookup_frame(const struct netlink_ring *ring, unsigned int pos,
386 enum nl_mmap_status status)
387 {
388 struct nl_mmap_hdr *hdr;
389
390 hdr = __netlink_lookup_frame(ring, pos);
391 if (netlink_get_status(hdr) != status)
392 return NULL;
393
394 return hdr;
395 }
396
397 static struct nl_mmap_hdr *
398 netlink_current_frame(const struct netlink_ring *ring,
399 enum nl_mmap_status status)
400 {
401 return netlink_lookup_frame(ring, ring->head, status);
402 }
403
404 static struct nl_mmap_hdr *
405 netlink_previous_frame(const struct netlink_ring *ring,
406 enum nl_mmap_status status)
407 {
408 unsigned int prev;
409
410 prev = ring->head ? ring->head - 1 : ring->frame_max;
411 return netlink_lookup_frame(ring, prev, status);
412 }
413
414 static void netlink_increment_head(struct netlink_ring *ring)
415 {
416 ring->head = ring->head != ring->frame_max ? ring->head + 1 : 0;
417 }
418
419 static void netlink_forward_ring(struct netlink_ring *ring)
420 {
421 unsigned int head = ring->head, pos = head;
422 const struct nl_mmap_hdr *hdr;
423
424 do {
425 hdr = __netlink_lookup_frame(ring, pos);
426 if (hdr->nm_status == NL_MMAP_STATUS_UNUSED)
427 break;
428 if (hdr->nm_status != NL_MMAP_STATUS_SKIP)
429 break;
430 netlink_increment_head(ring);
431 } while (ring->head != head);
432 }
433
434 static unsigned int netlink_poll(struct file *file, struct socket *sock,
435 poll_table *wait)
436 {
437 struct sock *sk = sock->sk;
438 struct netlink_sock *nlk = nlk_sk(sk);
439 unsigned int mask;
440
441 mask = datagram_poll(file, sock, wait);
442
443 spin_lock_bh(&sk->sk_receive_queue.lock);
444 if (nlk->rx_ring.pg_vec) {
445 netlink_forward_ring(&nlk->rx_ring);
446 if (!netlink_previous_frame(&nlk->rx_ring, NL_MMAP_STATUS_UNUSED))
447 mask |= POLLIN | POLLRDNORM;
448 }
449 spin_unlock_bh(&sk->sk_receive_queue.lock);
450
451 spin_lock_bh(&sk->sk_write_queue.lock);
452 if (nlk->tx_ring.pg_vec) {
453 if (netlink_current_frame(&nlk->tx_ring, NL_MMAP_STATUS_UNUSED))
454 mask |= POLLOUT | POLLWRNORM;
455 }
456 spin_unlock_bh(&sk->sk_write_queue.lock);
457
458 return mask;
459 }
460
461 static struct nl_mmap_hdr *netlink_mmap_hdr(struct sk_buff *skb)
462 {
463 return (struct nl_mmap_hdr *)(skb->head - NL_MMAP_HDRLEN);
464 }
465
466 static void netlink_ring_setup_skb(struct sk_buff *skb, struct sock *sk,
467 struct netlink_ring *ring,
468 struct nl_mmap_hdr *hdr)
469 {
470 unsigned int size;
471 void *data;
472
473 size = ring->frame_size - NL_MMAP_HDRLEN;
474 data = (void *)hdr + NL_MMAP_HDRLEN;
475
476 skb->head = data;
477 skb->data = data;
478 skb_reset_tail_pointer(skb);
479 skb->end = skb->tail + size;
480 skb->len = 0;
481
482 skb->destructor = netlink_skb_destructor;
483 NETLINK_CB(skb).flags |= NETLINK_SKB_MMAPED;
484 NETLINK_CB(skb).sk = sk;
485 }
486 #else /* CONFIG_NETLINK_MMAP */
487 #define netlink_skb_is_mmaped(skb) false
488 #define netlink_mmap sock_no_mmap
489 #define netlink_poll datagram_poll
490 #endif /* CONFIG_NETLINK_MMAP */
491
492 static void netlink_destroy_callback(struct netlink_callback *cb)
493 {
494 kfree_skb(cb->skb);
495 kfree(cb);
496 }
497
498 static void netlink_consume_callback(struct netlink_callback *cb)
499 {
500 consume_skb(cb->skb);
501 kfree(cb);
502 }
503
504 static void netlink_skb_destructor(struct sk_buff *skb)
505 {
506 #ifdef CONFIG_NETLINK_MMAP
507 struct nl_mmap_hdr *hdr;
508 struct netlink_ring *ring;
509 struct sock *sk;
510
511 /* If a packet from the kernel to userspace was freed because of an
512 * error without being delivered to userspace, the kernel must reset
513 * the status. In the direction userspace to kernel, the status is
514 * always reset here after the packet was processed and freed.
515 */
516 if (netlink_skb_is_mmaped(skb)) {
517 hdr = netlink_mmap_hdr(skb);
518 sk = NETLINK_CB(skb).sk;
519
520 if (!(NETLINK_CB(skb).flags & NETLINK_SKB_DELIVERED)) {
521 hdr->nm_len = 0;
522 netlink_set_status(hdr, NL_MMAP_STATUS_VALID);
523 }
524 ring = &nlk_sk(sk)->rx_ring;
525
526 WARN_ON(atomic_read(&ring->pending) == 0);
527 atomic_dec(&ring->pending);
528 sock_put(sk);
529
530 skb->data = NULL;
531 }
532 #endif
533 if (skb->sk != NULL)
534 sock_rfree(skb);
535 }
536
537 static void netlink_skb_set_owner_r(struct sk_buff *skb, struct sock *sk)
538 {
539 WARN_ON(skb->sk != NULL);
540 skb->sk = sk;
541 skb->destructor = netlink_skb_destructor;
542 atomic_add(skb->truesize, &sk->sk_rmem_alloc);
543 sk_mem_charge(sk, skb->truesize);
544 }
545
546 static void netlink_sock_destruct(struct sock *sk)
547 {
548 struct netlink_sock *nlk = nlk_sk(sk);
549
550 if (nlk->cb) {
551 if (nlk->cb->done)
552 nlk->cb->done(nlk->cb);
553
554 module_put(nlk->cb->module);
555 netlink_destroy_callback(nlk->cb);
556 }
557
558 skb_queue_purge(&sk->sk_receive_queue);
559 #ifdef CONFIG_NETLINK_MMAP
560 if (1) {
561 struct nl_mmap_req req;
562
563 memset(&req, 0, sizeof(req));
564 if (nlk->rx_ring.pg_vec)
565 netlink_set_ring(sk, &req, true, false);
566 memset(&req, 0, sizeof(req));
567 if (nlk->tx_ring.pg_vec)
568 netlink_set_ring(sk, &req, true, true);
569 }
570 #endif /* CONFIG_NETLINK_MMAP */
571
572 if (!sock_flag(sk, SOCK_DEAD)) {
573 printk(KERN_ERR "Freeing alive netlink socket %p\n", sk);
574 return;
575 }
576
577 WARN_ON(atomic_read(&sk->sk_rmem_alloc));
578 WARN_ON(atomic_read(&sk->sk_wmem_alloc));
579 WARN_ON(nlk_sk(sk)->groups);
580 }
581
582 /* This lock without WQ_FLAG_EXCLUSIVE is good on UP and it is _very_ bad on
583 * SMP. Look, when several writers sleep and reader wakes them up, all but one
584 * immediately hit write lock and grab all the cpus. Exclusive sleep solves
585 * this, _but_ remember, it adds useless work on UP machines.
586 */
587
588 void netlink_table_grab(void)
589 __acquires(nl_table_lock)
590 {
591 might_sleep();
592
593 write_lock_irq(&nl_table_lock);
594
595 if (atomic_read(&nl_table_users)) {
596 DECLARE_WAITQUEUE(wait, current);
597
598 add_wait_queue_exclusive(&nl_table_wait, &wait);
599 for (;;) {
600 set_current_state(TASK_UNINTERRUPTIBLE);
601 if (atomic_read(&nl_table_users) == 0)
602 break;
603 write_unlock_irq(&nl_table_lock);
604 schedule();
605 write_lock_irq(&nl_table_lock);
606 }
607
608 __set_current_state(TASK_RUNNING);
609 remove_wait_queue(&nl_table_wait, &wait);
610 }
611 }
612
613 void netlink_table_ungrab(void)
614 __releases(nl_table_lock)
615 {
616 write_unlock_irq(&nl_table_lock);
617 wake_up(&nl_table_wait);
618 }
619
620 static inline void
621 netlink_lock_table(void)
622 {
623 /* read_lock() synchronizes us to netlink_table_grab */
624
625 read_lock(&nl_table_lock);
626 atomic_inc(&nl_table_users);
627 read_unlock(&nl_table_lock);
628 }
629
630 static inline void
631 netlink_unlock_table(void)
632 {
633 if (atomic_dec_and_test(&nl_table_users))
634 wake_up(&nl_table_wait);
635 }
636
637 static struct sock *netlink_lookup(struct net *net, int protocol, u32 portid)
638 {
639 struct nl_portid_hash *hash = &nl_table[protocol].hash;
640 struct hlist_head *head;
641 struct sock *sk;
642
643 read_lock(&nl_table_lock);
644 head = nl_portid_hashfn(hash, portid);
645 sk_for_each(sk, head) {
646 if (net_eq(sock_net(sk), net) && (nlk_sk(sk)->portid == portid)) {
647 sock_hold(sk);
648 goto found;
649 }
650 }
651 sk = NULL;
652 found:
653 read_unlock(&nl_table_lock);
654 return sk;
655 }
656
657 static struct hlist_head *nl_portid_hash_zalloc(size_t size)
658 {
659 if (size <= PAGE_SIZE)
660 return kzalloc(size, GFP_ATOMIC);
661 else
662 return (struct hlist_head *)
663 __get_free_pages(GFP_ATOMIC | __GFP_ZERO,
664 get_order(size));
665 }
666
667 static void nl_portid_hash_free(struct hlist_head *table, size_t size)
668 {
669 if (size <= PAGE_SIZE)
670 kfree(table);
671 else
672 free_pages((unsigned long)table, get_order(size));
673 }
674
675 static int nl_portid_hash_rehash(struct nl_portid_hash *hash, int grow)
676 {
677 unsigned int omask, mask, shift;
678 size_t osize, size;
679 struct hlist_head *otable, *table;
680 int i;
681
682 omask = mask = hash->mask;
683 osize = size = (mask + 1) * sizeof(*table);
684 shift = hash->shift;
685
686 if (grow) {
687 if (++shift > hash->max_shift)
688 return 0;
689 mask = mask * 2 + 1;
690 size *= 2;
691 }
692
693 table = nl_portid_hash_zalloc(size);
694 if (!table)
695 return 0;
696
697 otable = hash->table;
698 hash->table = table;
699 hash->mask = mask;
700 hash->shift = shift;
701 get_random_bytes(&hash->rnd, sizeof(hash->rnd));
702
703 for (i = 0; i <= omask; i++) {
704 struct sock *sk;
705 struct hlist_node *tmp;
706
707 sk_for_each_safe(sk, tmp, &otable[i])
708 __sk_add_node(sk, nl_portid_hashfn(hash, nlk_sk(sk)->portid));
709 }
710
711 nl_portid_hash_free(otable, osize);
712 hash->rehash_time = jiffies + 10 * 60 * HZ;
713 return 1;
714 }
715
716 static inline int nl_portid_hash_dilute(struct nl_portid_hash *hash, int len)
717 {
718 int avg = hash->entries >> hash->shift;
719
720 if (unlikely(avg > 1) && nl_portid_hash_rehash(hash, 1))
721 return 1;
722
723 if (unlikely(len > avg) && time_after(jiffies, hash->rehash_time)) {
724 nl_portid_hash_rehash(hash, 0);
725 return 1;
726 }
727
728 return 0;
729 }
730
731 static const struct proto_ops netlink_ops;
732
733 static void
734 netlink_update_listeners(struct sock *sk)
735 {
736 struct netlink_table *tbl = &nl_table[sk->sk_protocol];
737 unsigned long mask;
738 unsigned int i;
739 struct listeners *listeners;
740
741 listeners = nl_deref_protected(tbl->listeners);
742 if (!listeners)
743 return;
744
745 for (i = 0; i < NLGRPLONGS(tbl->groups); i++) {
746 mask = 0;
747 sk_for_each_bound(sk, &tbl->mc_list) {
748 if (i < NLGRPLONGS(nlk_sk(sk)->ngroups))
749 mask |= nlk_sk(sk)->groups[i];
750 }
751 listeners->masks[i] = mask;
752 }
753 /* this function is only called with the netlink table "grabbed", which
754 * makes sure updates are visible before bind or setsockopt return. */
755 }
756
757 static int netlink_insert(struct sock *sk, struct net *net, u32 portid)
758 {
759 struct nl_portid_hash *hash = &nl_table[sk->sk_protocol].hash;
760 struct hlist_head *head;
761 int err = -EADDRINUSE;
762 struct sock *osk;
763 int len;
764
765 netlink_table_grab();
766 head = nl_portid_hashfn(hash, portid);
767 len = 0;
768 sk_for_each(osk, head) {
769 if (net_eq(sock_net(osk), net) && (nlk_sk(osk)->portid == portid))
770 break;
771 len++;
772 }
773 if (osk)
774 goto err;
775
776 err = -EBUSY;
777 if (nlk_sk(sk)->portid)
778 goto err;
779
780 err = -ENOMEM;
781 if (BITS_PER_LONG > 32 && unlikely(hash->entries >= UINT_MAX))
782 goto err;
783
784 if (len && nl_portid_hash_dilute(hash, len))
785 head = nl_portid_hashfn(hash, portid);
786 hash->entries++;
787 nlk_sk(sk)->portid = portid;
788 sk_add_node(sk, head);
789 err = 0;
790
791 err:
792 netlink_table_ungrab();
793 return err;
794 }
795
796 static void netlink_remove(struct sock *sk)
797 {
798 netlink_table_grab();
799 if (sk_del_node_init(sk))
800 nl_table[sk->sk_protocol].hash.entries--;
801 if (nlk_sk(sk)->subscriptions)
802 __sk_del_bind_node(sk);
803 netlink_table_ungrab();
804 }
805
806 static struct proto netlink_proto = {
807 .name = "NETLINK",
808 .owner = THIS_MODULE,
809 .obj_size = sizeof(struct netlink_sock),
810 };
811
812 static int __netlink_create(struct net *net, struct socket *sock,
813 struct mutex *cb_mutex, int protocol)
814 {
815 struct sock *sk;
816 struct netlink_sock *nlk;
817
818 sock->ops = &netlink_ops;
819
820 sk = sk_alloc(net, PF_NETLINK, GFP_KERNEL, &netlink_proto);
821 if (!sk)
822 return -ENOMEM;
823
824 sock_init_data(sock, sk);
825
826 nlk = nlk_sk(sk);
827 if (cb_mutex) {
828 nlk->cb_mutex = cb_mutex;
829 } else {
830 nlk->cb_mutex = &nlk->cb_def_mutex;
831 mutex_init(nlk->cb_mutex);
832 }
833 init_waitqueue_head(&nlk->wait);
834 #ifdef CONFIG_NETLINK_MMAP
835 mutex_init(&nlk->pg_vec_lock);
836 #endif
837
838 sk->sk_destruct = netlink_sock_destruct;
839 sk->sk_protocol = protocol;
840 return 0;
841 }
842
843 static int netlink_create(struct net *net, struct socket *sock, int protocol,
844 int kern)
845 {
846 struct module *module = NULL;
847 struct mutex *cb_mutex;
848 struct netlink_sock *nlk;
849 void (*bind)(int group);
850 int err = 0;
851
852 sock->state = SS_UNCONNECTED;
853
854 if (sock->type != SOCK_RAW && sock->type != SOCK_DGRAM)
855 return -ESOCKTNOSUPPORT;
856
857 if (protocol < 0 || protocol >= MAX_LINKS)
858 return -EPROTONOSUPPORT;
859
860 netlink_lock_table();
861 #ifdef CONFIG_MODULES
862 if (!nl_table[protocol].registered) {
863 netlink_unlock_table();
864 request_module("net-pf-%d-proto-%d", PF_NETLINK, protocol);
865 netlink_lock_table();
866 }
867 #endif
868 if (nl_table[protocol].registered &&
869 try_module_get(nl_table[protocol].module))
870 module = nl_table[protocol].module;
871 else
872 err = -EPROTONOSUPPORT;
873 cb_mutex = nl_table[protocol].cb_mutex;
874 bind = nl_table[protocol].bind;
875 netlink_unlock_table();
876
877 if (err < 0)
878 goto out;
879
880 err = __netlink_create(net, sock, cb_mutex, protocol);
881 if (err < 0)
882 goto out_module;
883
884 local_bh_disable();
885 sock_prot_inuse_add(net, &netlink_proto, 1);
886 local_bh_enable();
887
888 nlk = nlk_sk(sock->sk);
889 nlk->module = module;
890 nlk->netlink_bind = bind;
891 out:
892 return err;
893
894 out_module:
895 module_put(module);
896 goto out;
897 }
898
899 static int netlink_release(struct socket *sock)
900 {
901 struct sock *sk = sock->sk;
902 struct netlink_sock *nlk;
903
904 if (!sk)
905 return 0;
906
907 netlink_remove(sk);
908 sock_orphan(sk);
909 nlk = nlk_sk(sk);
910
911 /*
912 * OK. Socket is unlinked, any packets that arrive now
913 * will be purged.
914 */
915
916 sock->sk = NULL;
917 wake_up_interruptible_all(&nlk->wait);
918
919 skb_queue_purge(&sk->sk_write_queue);
920
921 if (nlk->portid) {
922 struct netlink_notify n = {
923 .net = sock_net(sk),
924 .protocol = sk->sk_protocol,
925 .portid = nlk->portid,
926 };
927 atomic_notifier_call_chain(&netlink_chain,
928 NETLINK_URELEASE, &n);
929 }
930
931 module_put(nlk->module);
932
933 netlink_table_grab();
934 if (netlink_is_kernel(sk)) {
935 BUG_ON(nl_table[sk->sk_protocol].registered == 0);
936 if (--nl_table[sk->sk_protocol].registered == 0) {
937 struct listeners *old;
938
939 old = nl_deref_protected(nl_table[sk->sk_protocol].listeners);
940 RCU_INIT_POINTER(nl_table[sk->sk_protocol].listeners, NULL);
941 kfree_rcu(old, rcu);
942 nl_table[sk->sk_protocol].module = NULL;
943 nl_table[sk->sk_protocol].bind = NULL;
944 nl_table[sk->sk_protocol].flags = 0;
945 nl_table[sk->sk_protocol].registered = 0;
946 }
947 } else if (nlk->subscriptions) {
948 netlink_update_listeners(sk);
949 }
950 netlink_table_ungrab();
951
952 kfree(nlk->groups);
953 nlk->groups = NULL;
954
955 local_bh_disable();
956 sock_prot_inuse_add(sock_net(sk), &netlink_proto, -1);
957 local_bh_enable();
958 sock_put(sk);
959 return 0;
960 }
961
962 static int netlink_autobind(struct socket *sock)
963 {
964 struct sock *sk = sock->sk;
965 struct net *net = sock_net(sk);
966 struct nl_portid_hash *hash = &nl_table[sk->sk_protocol].hash;
967 struct hlist_head *head;
968 struct sock *osk;
969 s32 portid = task_tgid_vnr(current);
970 int err;
971 static s32 rover = -4097;
972
973 retry:
974 cond_resched();
975 netlink_table_grab();
976 head = nl_portid_hashfn(hash, portid);
977 sk_for_each(osk, head) {
978 if (!net_eq(sock_net(osk), net))
979 continue;
980 if (nlk_sk(osk)->portid == portid) {
981 /* Bind collision, search negative portid values. */
982 portid = rover--;
983 if (rover > -4097)
984 rover = -4097;
985 netlink_table_ungrab();
986 goto retry;
987 }
988 }
989 netlink_table_ungrab();
990
991 err = netlink_insert(sk, net, portid);
992 if (err == -EADDRINUSE)
993 goto retry;
994
995 /* If 2 threads race to autobind, that is fine. */
996 if (err == -EBUSY)
997 err = 0;
998
999 return err;
1000 }
1001
1002 static inline int netlink_capable(const struct socket *sock, unsigned int flag)
1003 {
1004 return (nl_table[sock->sk->sk_protocol].flags & flag) ||
1005 ns_capable(sock_net(sock->sk)->user_ns, CAP_NET_ADMIN);
1006 }
1007
1008 static void
1009 netlink_update_subscriptions(struct sock *sk, unsigned int subscriptions)
1010 {
1011 struct netlink_sock *nlk = nlk_sk(sk);
1012
1013 if (nlk->subscriptions && !subscriptions)
1014 __sk_del_bind_node(sk);
1015 else if (!nlk->subscriptions && subscriptions)
1016 sk_add_bind_node(sk, &nl_table[sk->sk_protocol].mc_list);
1017 nlk->subscriptions = subscriptions;
1018 }
1019
1020 static int netlink_realloc_groups(struct sock *sk)
1021 {
1022 struct netlink_sock *nlk = nlk_sk(sk);
1023 unsigned int groups;
1024 unsigned long *new_groups;
1025 int err = 0;
1026
1027 netlink_table_grab();
1028
1029 groups = nl_table[sk->sk_protocol].groups;
1030 if (!nl_table[sk->sk_protocol].registered) {
1031 err = -ENOENT;
1032 goto out_unlock;
1033 }
1034
1035 if (nlk->ngroups >= groups)
1036 goto out_unlock;
1037
1038 new_groups = krealloc(nlk->groups, NLGRPSZ(groups), GFP_ATOMIC);
1039 if (new_groups == NULL) {
1040 err = -ENOMEM;
1041 goto out_unlock;
1042 }
1043 memset((char *)new_groups + NLGRPSZ(nlk->ngroups), 0,
1044 NLGRPSZ(groups) - NLGRPSZ(nlk->ngroups));
1045
1046 nlk->groups = new_groups;
1047 nlk->ngroups = groups;
1048 out_unlock:
1049 netlink_table_ungrab();
1050 return err;
1051 }
1052
1053 static int netlink_bind(struct socket *sock, struct sockaddr *addr,
1054 int addr_len)
1055 {
1056 struct sock *sk = sock->sk;
1057 struct net *net = sock_net(sk);
1058 struct netlink_sock *nlk = nlk_sk(sk);
1059 struct sockaddr_nl *nladdr = (struct sockaddr_nl *)addr;
1060 int err;
1061
1062 if (addr_len < sizeof(struct sockaddr_nl))
1063 return -EINVAL;
1064
1065 if (nladdr->nl_family != AF_NETLINK)
1066 return -EINVAL;
1067
1068 /* Only superuser is allowed to listen multicasts */
1069 if (nladdr->nl_groups) {
1070 if (!netlink_capable(sock, NL_CFG_F_NONROOT_RECV))
1071 return -EPERM;
1072 err = netlink_realloc_groups(sk);
1073 if (err)
1074 return err;
1075 }
1076
1077 if (nlk->portid) {
1078 if (nladdr->nl_pid != nlk->portid)
1079 return -EINVAL;
1080 } else {
1081 err = nladdr->nl_pid ?
1082 netlink_insert(sk, net, nladdr->nl_pid) :
1083 netlink_autobind(sock);
1084 if (err)
1085 return err;
1086 }
1087
1088 if (!nladdr->nl_groups && (nlk->groups == NULL || !(u32)nlk->groups[0]))
1089 return 0;
1090
1091 netlink_table_grab();
1092 netlink_update_subscriptions(sk, nlk->subscriptions +
1093 hweight32(nladdr->nl_groups) -
1094 hweight32(nlk->groups[0]));
1095 nlk->groups[0] = (nlk->groups[0] & ~0xffffffffUL) | nladdr->nl_groups;
1096 netlink_update_listeners(sk);
1097 netlink_table_ungrab();
1098
1099 if (nlk->netlink_bind && nlk->groups[0]) {
1100 int i;
1101
1102 for (i=0; i<nlk->ngroups; i++) {
1103 if (test_bit(i, nlk->groups))
1104 nlk->netlink_bind(i);
1105 }
1106 }
1107
1108 return 0;
1109 }
1110
1111 static int netlink_connect(struct socket *sock, struct sockaddr *addr,
1112 int alen, int flags)
1113 {
1114 int err = 0;
1115 struct sock *sk = sock->sk;
1116 struct netlink_sock *nlk = nlk_sk(sk);
1117 struct sockaddr_nl *nladdr = (struct sockaddr_nl *)addr;
1118
1119 if (alen < sizeof(addr->sa_family))
1120 return -EINVAL;
1121
1122 if (addr->sa_family == AF_UNSPEC) {
1123 sk->sk_state = NETLINK_UNCONNECTED;
1124 nlk->dst_portid = 0;
1125 nlk->dst_group = 0;
1126 return 0;
1127 }
1128 if (addr->sa_family != AF_NETLINK)
1129 return -EINVAL;
1130
1131 /* Only superuser is allowed to send multicasts */
1132 if (nladdr->nl_groups && !netlink_capable(sock, NL_CFG_F_NONROOT_SEND))
1133 return -EPERM;
1134
1135 if (!nlk->portid)
1136 err = netlink_autobind(sock);
1137
1138 if (err == 0) {
1139 sk->sk_state = NETLINK_CONNECTED;
1140 nlk->dst_portid = nladdr->nl_pid;
1141 nlk->dst_group = ffs(nladdr->nl_groups);
1142 }
1143
1144 return err;
1145 }
1146
1147 static int netlink_getname(struct socket *sock, struct sockaddr *addr,
1148 int *addr_len, int peer)
1149 {
1150 struct sock *sk = sock->sk;
1151 struct netlink_sock *nlk = nlk_sk(sk);
1152 DECLARE_SOCKADDR(struct sockaddr_nl *, nladdr, addr);
1153
1154 nladdr->nl_family = AF_NETLINK;
1155 nladdr->nl_pad = 0;
1156 *addr_len = sizeof(*nladdr);
1157
1158 if (peer) {
1159 nladdr->nl_pid = nlk->dst_portid;
1160 nladdr->nl_groups = netlink_group_mask(nlk->dst_group);
1161 } else {
1162 nladdr->nl_pid = nlk->portid;
1163 nladdr->nl_groups = nlk->groups ? nlk->groups[0] : 0;
1164 }
1165 return 0;
1166 }
1167
1168 static void netlink_overrun(struct sock *sk)
1169 {
1170 struct netlink_sock *nlk = nlk_sk(sk);
1171
1172 if (!(nlk->flags & NETLINK_RECV_NO_ENOBUFS)) {
1173 if (!test_and_set_bit(NETLINK_CONGESTED, &nlk_sk(sk)->state)) {
1174 sk->sk_err = ENOBUFS;
1175 sk->sk_error_report(sk);
1176 }
1177 }
1178 atomic_inc(&sk->sk_drops);
1179 }
1180
1181 static struct sock *netlink_getsockbyportid(struct sock *ssk, u32 portid)
1182 {
1183 struct sock *sock;
1184 struct netlink_sock *nlk;
1185
1186 sock = netlink_lookup(sock_net(ssk), ssk->sk_protocol, portid);
1187 if (!sock)
1188 return ERR_PTR(-ECONNREFUSED);
1189
1190 /* Don't bother queuing skb if kernel socket has no input function */
1191 nlk = nlk_sk(sock);
1192 if (sock->sk_state == NETLINK_CONNECTED &&
1193 nlk->dst_portid != nlk_sk(ssk)->portid) {
1194 sock_put(sock);
1195 return ERR_PTR(-ECONNREFUSED);
1196 }
1197 return sock;
1198 }
1199
1200 struct sock *netlink_getsockbyfilp(struct file *filp)
1201 {
1202 struct inode *inode = file_inode(filp);
1203 struct sock *sock;
1204
1205 if (!S_ISSOCK(inode->i_mode))
1206 return ERR_PTR(-ENOTSOCK);
1207
1208 sock = SOCKET_I(inode)->sk;
1209 if (sock->sk_family != AF_NETLINK)
1210 return ERR_PTR(-EINVAL);
1211
1212 sock_hold(sock);
1213 return sock;
1214 }
1215
1216 /*
1217 * Attach a skb to a netlink socket.
1218 * The caller must hold a reference to the destination socket. On error, the
1219 * reference is dropped. The skb is not send to the destination, just all
1220 * all error checks are performed and memory in the queue is reserved.
1221 * Return values:
1222 * < 0: error. skb freed, reference to sock dropped.
1223 * 0: continue
1224 * 1: repeat lookup - reference dropped while waiting for socket memory.
1225 */
1226 int netlink_attachskb(struct sock *sk, struct sk_buff *skb,
1227 long *timeo, struct sock *ssk)
1228 {
1229 struct netlink_sock *nlk;
1230
1231 nlk = nlk_sk(sk);
1232
1233 if (atomic_read(&sk->sk_rmem_alloc) > sk->sk_rcvbuf ||
1234 test_bit(NETLINK_CONGESTED, &nlk->state)) {
1235 DECLARE_WAITQUEUE(wait, current);
1236 if (!*timeo) {
1237 if (!ssk || netlink_is_kernel(ssk))
1238 netlink_overrun(sk);
1239 sock_put(sk);
1240 kfree_skb(skb);
1241 return -EAGAIN;
1242 }
1243
1244 __set_current_state(TASK_INTERRUPTIBLE);
1245 add_wait_queue(&nlk->wait, &wait);
1246
1247 if ((atomic_read(&sk->sk_rmem_alloc) > sk->sk_rcvbuf ||
1248 test_bit(NETLINK_CONGESTED, &nlk->state)) &&
1249 !sock_flag(sk, SOCK_DEAD))
1250 *timeo = schedule_timeout(*timeo);
1251
1252 __set_current_state(TASK_RUNNING);
1253 remove_wait_queue(&nlk->wait, &wait);
1254 sock_put(sk);
1255
1256 if (signal_pending(current)) {
1257 kfree_skb(skb);
1258 return sock_intr_errno(*timeo);
1259 }
1260 return 1;
1261 }
1262 netlink_skb_set_owner_r(skb, sk);
1263 return 0;
1264 }
1265
1266 static int __netlink_sendskb(struct sock *sk, struct sk_buff *skb)
1267 {
1268 int len = skb->len;
1269
1270 skb_queue_tail(&sk->sk_receive_queue, skb);
1271 sk->sk_data_ready(sk, len);
1272 return len;
1273 }
1274
1275 int netlink_sendskb(struct sock *sk, struct sk_buff *skb)
1276 {
1277 int len = __netlink_sendskb(sk, skb);
1278
1279 sock_put(sk);
1280 return len;
1281 }
1282
1283 void netlink_detachskb(struct sock *sk, struct sk_buff *skb)
1284 {
1285 kfree_skb(skb);
1286 sock_put(sk);
1287 }
1288
1289 static struct sk_buff *netlink_trim(struct sk_buff *skb, gfp_t allocation)
1290 {
1291 int delta;
1292
1293 WARN_ON(skb->sk != NULL);
1294
1295 delta = skb->end - skb->tail;
1296 if (delta * 2 < skb->truesize)
1297 return skb;
1298
1299 if (skb_shared(skb)) {
1300 struct sk_buff *nskb = skb_clone(skb, allocation);
1301 if (!nskb)
1302 return skb;
1303 consume_skb(skb);
1304 skb = nskb;
1305 }
1306
1307 if (!pskb_expand_head(skb, 0, -delta, allocation))
1308 skb->truesize -= delta;
1309
1310 return skb;
1311 }
1312
1313 static void netlink_rcv_wake(struct sock *sk)
1314 {
1315 struct netlink_sock *nlk = nlk_sk(sk);
1316
1317 if (skb_queue_empty(&sk->sk_receive_queue))
1318 clear_bit(NETLINK_CONGESTED, &nlk->state);
1319 if (!test_bit(NETLINK_CONGESTED, &nlk->state))
1320 wake_up_interruptible(&nlk->wait);
1321 }
1322
1323 static int netlink_unicast_kernel(struct sock *sk, struct sk_buff *skb,
1324 struct sock *ssk)
1325 {
1326 int ret;
1327 struct netlink_sock *nlk = nlk_sk(sk);
1328
1329 ret = -ECONNREFUSED;
1330 if (nlk->netlink_rcv != NULL) {
1331 ret = skb->len;
1332 netlink_skb_set_owner_r(skb, sk);
1333 NETLINK_CB(skb).sk = ssk;
1334 nlk->netlink_rcv(skb);
1335 consume_skb(skb);
1336 } else {
1337 kfree_skb(skb);
1338 }
1339 sock_put(sk);
1340 return ret;
1341 }
1342
1343 int netlink_unicast(struct sock *ssk, struct sk_buff *skb,
1344 u32 portid, int nonblock)
1345 {
1346 struct sock *sk;
1347 int err;
1348 long timeo;
1349
1350 skb = netlink_trim(skb, gfp_any());
1351
1352 timeo = sock_sndtimeo(ssk, nonblock);
1353 retry:
1354 sk = netlink_getsockbyportid(ssk, portid);
1355 if (IS_ERR(sk)) {
1356 kfree_skb(skb);
1357 return PTR_ERR(sk);
1358 }
1359 if (netlink_is_kernel(sk))
1360 return netlink_unicast_kernel(sk, skb, ssk);
1361
1362 if (sk_filter(sk, skb)) {
1363 err = skb->len;
1364 kfree_skb(skb);
1365 sock_put(sk);
1366 return err;
1367 }
1368
1369 err = netlink_attachskb(sk, skb, &timeo, ssk);
1370 if (err == 1)
1371 goto retry;
1372 if (err)
1373 return err;
1374
1375 return netlink_sendskb(sk, skb);
1376 }
1377 EXPORT_SYMBOL(netlink_unicast);
1378
1379 int netlink_has_listeners(struct sock *sk, unsigned int group)
1380 {
1381 int res = 0;
1382 struct listeners *listeners;
1383
1384 BUG_ON(!netlink_is_kernel(sk));
1385
1386 rcu_read_lock();
1387 listeners = rcu_dereference(nl_table[sk->sk_protocol].listeners);
1388
1389 if (listeners && group - 1 < nl_table[sk->sk_protocol].groups)
1390 res = test_bit(group - 1, listeners->masks);
1391
1392 rcu_read_unlock();
1393
1394 return res;
1395 }
1396 EXPORT_SYMBOL_GPL(netlink_has_listeners);
1397
1398 static int netlink_broadcast_deliver(struct sock *sk, struct sk_buff *skb)
1399 {
1400 struct netlink_sock *nlk = nlk_sk(sk);
1401
1402 if (atomic_read(&sk->sk_rmem_alloc) <= sk->sk_rcvbuf &&
1403 !test_bit(NETLINK_CONGESTED, &nlk->state)) {
1404 netlink_skb_set_owner_r(skb, sk);
1405 __netlink_sendskb(sk, skb);
1406 return atomic_read(&sk->sk_rmem_alloc) > (sk->sk_rcvbuf >> 1);
1407 }
1408 return -1;
1409 }
1410
1411 struct netlink_broadcast_data {
1412 struct sock *exclude_sk;
1413 struct net *net;
1414 u32 portid;
1415 u32 group;
1416 int failure;
1417 int delivery_failure;
1418 int congested;
1419 int delivered;
1420 gfp_t allocation;
1421 struct sk_buff *skb, *skb2;
1422 int (*tx_filter)(struct sock *dsk, struct sk_buff *skb, void *data);
1423 void *tx_data;
1424 };
1425
1426 static int do_one_broadcast(struct sock *sk,
1427 struct netlink_broadcast_data *p)
1428 {
1429 struct netlink_sock *nlk = nlk_sk(sk);
1430 int val;
1431
1432 if (p->exclude_sk == sk)
1433 goto out;
1434
1435 if (nlk->portid == p->portid || p->group - 1 >= nlk->ngroups ||
1436 !test_bit(p->group - 1, nlk->groups))
1437 goto out;
1438
1439 if (!net_eq(sock_net(sk), p->net))
1440 goto out;
1441
1442 if (p->failure) {
1443 netlink_overrun(sk);
1444 goto out;
1445 }
1446
1447 sock_hold(sk);
1448 if (p->skb2 == NULL) {
1449 if (skb_shared(p->skb)) {
1450 p->skb2 = skb_clone(p->skb, p->allocation);
1451 } else {
1452 p->skb2 = skb_get(p->skb);
1453 /*
1454 * skb ownership may have been set when
1455 * delivered to a previous socket.
1456 */
1457 skb_orphan(p->skb2);
1458 }
1459 }
1460 if (p->skb2 == NULL) {
1461 netlink_overrun(sk);
1462 /* Clone failed. Notify ALL listeners. */
1463 p->failure = 1;
1464 if (nlk->flags & NETLINK_BROADCAST_SEND_ERROR)
1465 p->delivery_failure = 1;
1466 } else if (p->tx_filter && p->tx_filter(sk, p->skb2, p->tx_data)) {
1467 kfree_skb(p->skb2);
1468 p->skb2 = NULL;
1469 } else if (sk_filter(sk, p->skb2)) {
1470 kfree_skb(p->skb2);
1471 p->skb2 = NULL;
1472 } else if ((val = netlink_broadcast_deliver(sk, p->skb2)) < 0) {
1473 netlink_overrun(sk);
1474 if (nlk->flags & NETLINK_BROADCAST_SEND_ERROR)
1475 p->delivery_failure = 1;
1476 } else {
1477 p->congested |= val;
1478 p->delivered = 1;
1479 p->skb2 = NULL;
1480 }
1481 sock_put(sk);
1482
1483 out:
1484 return 0;
1485 }
1486
1487 int netlink_broadcast_filtered(struct sock *ssk, struct sk_buff *skb, u32 portid,
1488 u32 group, gfp_t allocation,
1489 int (*filter)(struct sock *dsk, struct sk_buff *skb, void *data),
1490 void *filter_data)
1491 {
1492 struct net *net = sock_net(ssk);
1493 struct netlink_broadcast_data info;
1494 struct sock *sk;
1495
1496 skb = netlink_trim(skb, allocation);
1497
1498 info.exclude_sk = ssk;
1499 info.net = net;
1500 info.portid = portid;
1501 info.group = group;
1502 info.failure = 0;
1503 info.delivery_failure = 0;
1504 info.congested = 0;
1505 info.delivered = 0;
1506 info.allocation = allocation;
1507 info.skb = skb;
1508 info.skb2 = NULL;
1509 info.tx_filter = filter;
1510 info.tx_data = filter_data;
1511
1512 /* While we sleep in clone, do not allow to change socket list */
1513
1514 netlink_lock_table();
1515
1516 sk_for_each_bound(sk, &nl_table[ssk->sk_protocol].mc_list)
1517 do_one_broadcast(sk, &info);
1518
1519 consume_skb(skb);
1520
1521 netlink_unlock_table();
1522
1523 if (info.delivery_failure) {
1524 kfree_skb(info.skb2);
1525 return -ENOBUFS;
1526 }
1527 consume_skb(info.skb2);
1528
1529 if (info.delivered) {
1530 if (info.congested && (allocation & __GFP_WAIT))
1531 yield();
1532 return 0;
1533 }
1534 return -ESRCH;
1535 }
1536 EXPORT_SYMBOL(netlink_broadcast_filtered);
1537
1538 int netlink_broadcast(struct sock *ssk, struct sk_buff *skb, u32 portid,
1539 u32 group, gfp_t allocation)
1540 {
1541 return netlink_broadcast_filtered(ssk, skb, portid, group, allocation,
1542 NULL, NULL);
1543 }
1544 EXPORT_SYMBOL(netlink_broadcast);
1545
1546 struct netlink_set_err_data {
1547 struct sock *exclude_sk;
1548 u32 portid;
1549 u32 group;
1550 int code;
1551 };
1552
1553 static int do_one_set_err(struct sock *sk, struct netlink_set_err_data *p)
1554 {
1555 struct netlink_sock *nlk = nlk_sk(sk);
1556 int ret = 0;
1557
1558 if (sk == p->exclude_sk)
1559 goto out;
1560
1561 if (!net_eq(sock_net(sk), sock_net(p->exclude_sk)))
1562 goto out;
1563
1564 if (nlk->portid == p->portid || p->group - 1 >= nlk->ngroups ||
1565 !test_bit(p->group - 1, nlk->groups))
1566 goto out;
1567
1568 if (p->code == ENOBUFS && nlk->flags & NETLINK_RECV_NO_ENOBUFS) {
1569 ret = 1;
1570 goto out;
1571 }
1572
1573 sk->sk_err = p->code;
1574 sk->sk_error_report(sk);
1575 out:
1576 return ret;
1577 }
1578
1579 /**
1580 * netlink_set_err - report error to broadcast listeners
1581 * @ssk: the kernel netlink socket, as returned by netlink_kernel_create()
1582 * @portid: the PORTID of a process that we want to skip (if any)
1583 * @groups: the broadcast group that will notice the error
1584 * @code: error code, must be negative (as usual in kernelspace)
1585 *
1586 * This function returns the number of broadcast listeners that have set the
1587 * NETLINK_RECV_NO_ENOBUFS socket option.
1588 */
1589 int netlink_set_err(struct sock *ssk, u32 portid, u32 group, int code)
1590 {
1591 struct netlink_set_err_data info;
1592 struct sock *sk;
1593 int ret = 0;
1594
1595 info.exclude_sk = ssk;
1596 info.portid = portid;
1597 info.group = group;
1598 /* sk->sk_err wants a positive error value */
1599 info.code = -code;
1600
1601 read_lock(&nl_table_lock);
1602
1603 sk_for_each_bound(sk, &nl_table[ssk->sk_protocol].mc_list)
1604 ret += do_one_set_err(sk, &info);
1605
1606 read_unlock(&nl_table_lock);
1607 return ret;
1608 }
1609 EXPORT_SYMBOL(netlink_set_err);
1610
1611 /* must be called with netlink table grabbed */
1612 static void netlink_update_socket_mc(struct netlink_sock *nlk,
1613 unsigned int group,
1614 int is_new)
1615 {
1616 int old, new = !!is_new, subscriptions;
1617
1618 old = test_bit(group - 1, nlk->groups);
1619 subscriptions = nlk->subscriptions - old + new;
1620 if (new)
1621 __set_bit(group - 1, nlk->groups);
1622 else
1623 __clear_bit(group - 1, nlk->groups);
1624 netlink_update_subscriptions(&nlk->sk, subscriptions);
1625 netlink_update_listeners(&nlk->sk);
1626 }
1627
1628 static int netlink_setsockopt(struct socket *sock, int level, int optname,
1629 char __user *optval, unsigned int optlen)
1630 {
1631 struct sock *sk = sock->sk;
1632 struct netlink_sock *nlk = nlk_sk(sk);
1633 unsigned int val = 0;
1634 int err;
1635
1636 if (level != SOL_NETLINK)
1637 return -ENOPROTOOPT;
1638
1639 if (optname != NETLINK_RX_RING && optname != NETLINK_TX_RING &&
1640 optlen >= sizeof(int) &&
1641 get_user(val, (unsigned int __user *)optval))
1642 return -EFAULT;
1643
1644 switch (optname) {
1645 case NETLINK_PKTINFO:
1646 if (val)
1647 nlk->flags |= NETLINK_RECV_PKTINFO;
1648 else
1649 nlk->flags &= ~NETLINK_RECV_PKTINFO;
1650 err = 0;
1651 break;
1652 case NETLINK_ADD_MEMBERSHIP:
1653 case NETLINK_DROP_MEMBERSHIP: {
1654 if (!netlink_capable(sock, NL_CFG_F_NONROOT_RECV))
1655 return -EPERM;
1656 err = netlink_realloc_groups(sk);
1657 if (err)
1658 return err;
1659 if (!val || val - 1 >= nlk->ngroups)
1660 return -EINVAL;
1661 netlink_table_grab();
1662 netlink_update_socket_mc(nlk, val,
1663 optname == NETLINK_ADD_MEMBERSHIP);
1664 netlink_table_ungrab();
1665
1666 if (nlk->netlink_bind)
1667 nlk->netlink_bind(val);
1668
1669 err = 0;
1670 break;
1671 }
1672 case NETLINK_BROADCAST_ERROR:
1673 if (val)
1674 nlk->flags |= NETLINK_BROADCAST_SEND_ERROR;
1675 else
1676 nlk->flags &= ~NETLINK_BROADCAST_SEND_ERROR;
1677 err = 0;
1678 break;
1679 case NETLINK_NO_ENOBUFS:
1680 if (val) {
1681 nlk->flags |= NETLINK_RECV_NO_ENOBUFS;
1682 clear_bit(NETLINK_CONGESTED, &nlk->state);
1683 wake_up_interruptible(&nlk->wait);
1684 } else {
1685 nlk->flags &= ~NETLINK_RECV_NO_ENOBUFS;
1686 }
1687 err = 0;
1688 break;
1689 #ifdef CONFIG_NETLINK_MMAP
1690 case NETLINK_RX_RING:
1691 case NETLINK_TX_RING: {
1692 struct nl_mmap_req req;
1693
1694 /* Rings might consume more memory than queue limits, require
1695 * CAP_NET_ADMIN.
1696 */
1697 if (!capable(CAP_NET_ADMIN))
1698 return -EPERM;
1699 if (optlen < sizeof(req))
1700 return -EINVAL;
1701 if (copy_from_user(&req, optval, sizeof(req)))
1702 return -EFAULT;
1703 err = netlink_set_ring(sk, &req, false,
1704 optname == NETLINK_TX_RING);
1705 break;
1706 }
1707 #endif /* CONFIG_NETLINK_MMAP */
1708 default:
1709 err = -ENOPROTOOPT;
1710 }
1711 return err;
1712 }
1713
1714 static int netlink_getsockopt(struct socket *sock, int level, int optname,
1715 char __user *optval, int __user *optlen)
1716 {
1717 struct sock *sk = sock->sk;
1718 struct netlink_sock *nlk = nlk_sk(sk);
1719 int len, val, err;
1720
1721 if (level != SOL_NETLINK)
1722 return -ENOPROTOOPT;
1723
1724 if (get_user(len, optlen))
1725 return -EFAULT;
1726 if (len < 0)
1727 return -EINVAL;
1728
1729 switch (optname) {
1730 case NETLINK_PKTINFO:
1731 if (len < sizeof(int))
1732 return -EINVAL;
1733 len = sizeof(int);
1734 val = nlk->flags & NETLINK_RECV_PKTINFO ? 1 : 0;
1735 if (put_user(len, optlen) ||
1736 put_user(val, optval))
1737 return -EFAULT;
1738 err = 0;
1739 break;
1740 case NETLINK_BROADCAST_ERROR:
1741 if (len < sizeof(int))
1742 return -EINVAL;
1743 len = sizeof(int);
1744 val = nlk->flags & NETLINK_BROADCAST_SEND_ERROR ? 1 : 0;
1745 if (put_user(len, optlen) ||
1746 put_user(val, optval))
1747 return -EFAULT;
1748 err = 0;
1749 break;
1750 case NETLINK_NO_ENOBUFS:
1751 if (len < sizeof(int))
1752 return -EINVAL;
1753 len = sizeof(int);
1754 val = nlk->flags & NETLINK_RECV_NO_ENOBUFS ? 1 : 0;
1755 if (put_user(len, optlen) ||
1756 put_user(val, optval))
1757 return -EFAULT;
1758 err = 0;
1759 break;
1760 default:
1761 err = -ENOPROTOOPT;
1762 }
1763 return err;
1764 }
1765
1766 static void netlink_cmsg_recv_pktinfo(struct msghdr *msg, struct sk_buff *skb)
1767 {
1768 struct nl_pktinfo info;
1769
1770 info.group = NETLINK_CB(skb).dst_group;
1771 put_cmsg(msg, SOL_NETLINK, NETLINK_PKTINFO, sizeof(info), &info);
1772 }
1773
1774 static int netlink_sendmsg(struct kiocb *kiocb, struct socket *sock,
1775 struct msghdr *msg, size_t len)
1776 {
1777 struct sock_iocb *siocb = kiocb_to_siocb(kiocb);
1778 struct sock *sk = sock->sk;
1779 struct netlink_sock *nlk = nlk_sk(sk);
1780 struct sockaddr_nl *addr = msg->msg_name;
1781 u32 dst_portid;
1782 u32 dst_group;
1783 struct sk_buff *skb;
1784 int err;
1785 struct scm_cookie scm;
1786
1787 if (msg->msg_flags&MSG_OOB)
1788 return -EOPNOTSUPP;
1789
1790 if (NULL == siocb->scm)
1791 siocb->scm = &scm;
1792
1793 err = scm_send(sock, msg, siocb->scm, true);
1794 if (err < 0)
1795 return err;
1796
1797 if (msg->msg_namelen) {
1798 err = -EINVAL;
1799 if (addr->nl_family != AF_NETLINK)
1800 goto out;
1801 dst_portid = addr->nl_pid;
1802 dst_group = ffs(addr->nl_groups);
1803 err = -EPERM;
1804 if ((dst_group || dst_portid) &&
1805 !netlink_capable(sock, NL_CFG_F_NONROOT_SEND))
1806 goto out;
1807 } else {
1808 dst_portid = nlk->dst_portid;
1809 dst_group = nlk->dst_group;
1810 }
1811
1812 if (!nlk->portid) {
1813 err = netlink_autobind(sock);
1814 if (err)
1815 goto out;
1816 }
1817
1818 err = -EMSGSIZE;
1819 if (len > sk->sk_sndbuf - 32)
1820 goto out;
1821 err = -ENOBUFS;
1822 skb = alloc_skb(len, GFP_KERNEL);
1823 if (skb == NULL)
1824 goto out;
1825
1826 NETLINK_CB(skb).portid = nlk->portid;
1827 NETLINK_CB(skb).dst_group = dst_group;
1828 NETLINK_CB(skb).creds = siocb->scm->creds;
1829
1830 err = -EFAULT;
1831 if (memcpy_fromiovec(skb_put(skb, len), msg->msg_iov, len)) {
1832 kfree_skb(skb);
1833 goto out;
1834 }
1835
1836 err = security_netlink_send(sk, skb);
1837 if (err) {
1838 kfree_skb(skb);
1839 goto out;
1840 }
1841
1842 if (dst_group) {
1843 atomic_inc(&skb->users);
1844 netlink_broadcast(sk, skb, dst_portid, dst_group, GFP_KERNEL);
1845 }
1846 err = netlink_unicast(sk, skb, dst_portid, msg->msg_flags&MSG_DONTWAIT);
1847
1848 out:
1849 scm_destroy(siocb->scm);
1850 return err;
1851 }
1852
1853 static int netlink_recvmsg(struct kiocb *kiocb, struct socket *sock,
1854 struct msghdr *msg, size_t len,
1855 int flags)
1856 {
1857 struct sock_iocb *siocb = kiocb_to_siocb(kiocb);
1858 struct scm_cookie scm;
1859 struct sock *sk = sock->sk;
1860 struct netlink_sock *nlk = nlk_sk(sk);
1861 int noblock = flags&MSG_DONTWAIT;
1862 size_t copied;
1863 struct sk_buff *skb, *data_skb;
1864 int err, ret;
1865
1866 if (flags&MSG_OOB)
1867 return -EOPNOTSUPP;
1868
1869 copied = 0;
1870
1871 skb = skb_recv_datagram(sk, flags, noblock, &err);
1872 if (skb == NULL)
1873 goto out;
1874
1875 data_skb = skb;
1876
1877 #ifdef CONFIG_COMPAT_NETLINK_MESSAGES
1878 if (unlikely(skb_shinfo(skb)->frag_list)) {
1879 /*
1880 * If this skb has a frag_list, then here that means that we
1881 * will have to use the frag_list skb's data for compat tasks
1882 * and the regular skb's data for normal (non-compat) tasks.
1883 *
1884 * If we need to send the compat skb, assign it to the
1885 * 'data_skb' variable so that it will be used below for data
1886 * copying. We keep 'skb' for everything else, including
1887 * freeing both later.
1888 */
1889 if (flags & MSG_CMSG_COMPAT)
1890 data_skb = skb_shinfo(skb)->frag_list;
1891 }
1892 #endif
1893
1894 msg->msg_namelen = 0;
1895
1896 copied = data_skb->len;
1897 if (len < copied) {
1898 msg->msg_flags |= MSG_TRUNC;
1899 copied = len;
1900 }
1901
1902 skb_reset_transport_header(data_skb);
1903 err = skb_copy_datagram_iovec(data_skb, 0, msg->msg_iov, copied);
1904
1905 if (msg->msg_name) {
1906 struct sockaddr_nl *addr = (struct sockaddr_nl *)msg->msg_name;
1907 addr->nl_family = AF_NETLINK;
1908 addr->nl_pad = 0;
1909 addr->nl_pid = NETLINK_CB(skb).portid;
1910 addr->nl_groups = netlink_group_mask(NETLINK_CB(skb).dst_group);
1911 msg->msg_namelen = sizeof(*addr);
1912 }
1913
1914 if (nlk->flags & NETLINK_RECV_PKTINFO)
1915 netlink_cmsg_recv_pktinfo(msg, skb);
1916
1917 if (NULL == siocb->scm) {
1918 memset(&scm, 0, sizeof(scm));
1919 siocb->scm = &scm;
1920 }
1921 siocb->scm->creds = *NETLINK_CREDS(skb);
1922 if (flags & MSG_TRUNC)
1923 copied = data_skb->len;
1924
1925 skb_free_datagram(sk, skb);
1926
1927 if (nlk->cb && atomic_read(&sk->sk_rmem_alloc) <= sk->sk_rcvbuf / 2) {
1928 ret = netlink_dump(sk);
1929 if (ret) {
1930 sk->sk_err = ret;
1931 sk->sk_error_report(sk);
1932 }
1933 }
1934
1935 scm_recv(sock, msg, siocb->scm, flags);
1936 out:
1937 netlink_rcv_wake(sk);
1938 return err ? : copied;
1939 }
1940
1941 static void netlink_data_ready(struct sock *sk, int len)
1942 {
1943 BUG();
1944 }
1945
1946 /*
1947 * We export these functions to other modules. They provide a
1948 * complete set of kernel non-blocking support for message
1949 * queueing.
1950 */
1951
1952 struct sock *
1953 __netlink_kernel_create(struct net *net, int unit, struct module *module,
1954 struct netlink_kernel_cfg *cfg)
1955 {
1956 struct socket *sock;
1957 struct sock *sk;
1958 struct netlink_sock *nlk;
1959 struct listeners *listeners = NULL;
1960 struct mutex *cb_mutex = cfg ? cfg->cb_mutex : NULL;
1961 unsigned int groups;
1962
1963 BUG_ON(!nl_table);
1964
1965 if (unit < 0 || unit >= MAX_LINKS)
1966 return NULL;
1967
1968 if (sock_create_lite(PF_NETLINK, SOCK_DGRAM, unit, &sock))
1969 return NULL;
1970
1971 /*
1972 * We have to just have a reference on the net from sk, but don't
1973 * get_net it. Besides, we cannot get and then put the net here.
1974 * So we create one inside init_net and the move it to net.
1975 */
1976
1977 if (__netlink_create(&init_net, sock, cb_mutex, unit) < 0)
1978 goto out_sock_release_nosk;
1979
1980 sk = sock->sk;
1981 sk_change_net(sk, net);
1982
1983 if (!cfg || cfg->groups < 32)
1984 groups = 32;
1985 else
1986 groups = cfg->groups;
1987
1988 listeners = kzalloc(sizeof(*listeners) + NLGRPSZ(groups), GFP_KERNEL);
1989 if (!listeners)
1990 goto out_sock_release;
1991
1992 sk->sk_data_ready = netlink_data_ready;
1993 if (cfg && cfg->input)
1994 nlk_sk(sk)->netlink_rcv = cfg->input;
1995
1996 if (netlink_insert(sk, net, 0))
1997 goto out_sock_release;
1998
1999 nlk = nlk_sk(sk);
2000 nlk->flags |= NETLINK_KERNEL_SOCKET;
2001
2002 netlink_table_grab();
2003 if (!nl_table[unit].registered) {
2004 nl_table[unit].groups = groups;
2005 rcu_assign_pointer(nl_table[unit].listeners, listeners);
2006 nl_table[unit].cb_mutex = cb_mutex;
2007 nl_table[unit].module = module;
2008 if (cfg) {
2009 nl_table[unit].bind = cfg->bind;
2010 nl_table[unit].flags = cfg->flags;
2011 }
2012 nl_table[unit].registered = 1;
2013 } else {
2014 kfree(listeners);
2015 nl_table[unit].registered++;
2016 }
2017 netlink_table_ungrab();
2018 return sk;
2019
2020 out_sock_release:
2021 kfree(listeners);
2022 netlink_kernel_release(sk);
2023 return NULL;
2024
2025 out_sock_release_nosk:
2026 sock_release(sock);
2027 return NULL;
2028 }
2029 EXPORT_SYMBOL(__netlink_kernel_create);
2030
2031 void
2032 netlink_kernel_release(struct sock *sk)
2033 {
2034 sk_release_kernel(sk);
2035 }
2036 EXPORT_SYMBOL(netlink_kernel_release);
2037
2038 int __netlink_change_ngroups(struct sock *sk, unsigned int groups)
2039 {
2040 struct listeners *new, *old;
2041 struct netlink_table *tbl = &nl_table[sk->sk_protocol];
2042
2043 if (groups < 32)
2044 groups = 32;
2045
2046 if (NLGRPSZ(tbl->groups) < NLGRPSZ(groups)) {
2047 new = kzalloc(sizeof(*new) + NLGRPSZ(groups), GFP_ATOMIC);
2048 if (!new)
2049 return -ENOMEM;
2050 old = nl_deref_protected(tbl->listeners);
2051 memcpy(new->masks, old->masks, NLGRPSZ(tbl->groups));
2052 rcu_assign_pointer(tbl->listeners, new);
2053
2054 kfree_rcu(old, rcu);
2055 }
2056 tbl->groups = groups;
2057
2058 return 0;
2059 }
2060
2061 /**
2062 * netlink_change_ngroups - change number of multicast groups
2063 *
2064 * This changes the number of multicast groups that are available
2065 * on a certain netlink family. Note that it is not possible to
2066 * change the number of groups to below 32. Also note that it does
2067 * not implicitly call netlink_clear_multicast_users() when the
2068 * number of groups is reduced.
2069 *
2070 * @sk: The kernel netlink socket, as returned by netlink_kernel_create().
2071 * @groups: The new number of groups.
2072 */
2073 int netlink_change_ngroups(struct sock *sk, unsigned int groups)
2074 {
2075 int err;
2076
2077 netlink_table_grab();
2078 err = __netlink_change_ngroups(sk, groups);
2079 netlink_table_ungrab();
2080
2081 return err;
2082 }
2083
2084 void __netlink_clear_multicast_users(struct sock *ksk, unsigned int group)
2085 {
2086 struct sock *sk;
2087 struct netlink_table *tbl = &nl_table[ksk->sk_protocol];
2088
2089 sk_for_each_bound(sk, &tbl->mc_list)
2090 netlink_update_socket_mc(nlk_sk(sk), group, 0);
2091 }
2092
2093 /**
2094 * netlink_clear_multicast_users - kick off multicast listeners
2095 *
2096 * This function removes all listeners from the given group.
2097 * @ksk: The kernel netlink socket, as returned by
2098 * netlink_kernel_create().
2099 * @group: The multicast group to clear.
2100 */
2101 void netlink_clear_multicast_users(struct sock *ksk, unsigned int group)
2102 {
2103 netlink_table_grab();
2104 __netlink_clear_multicast_users(ksk, group);
2105 netlink_table_ungrab();
2106 }
2107
2108 struct nlmsghdr *
2109 __nlmsg_put(struct sk_buff *skb, u32 portid, u32 seq, int type, int len, int flags)
2110 {
2111 struct nlmsghdr *nlh;
2112 int size = nlmsg_msg_size(len);
2113
2114 nlh = (struct nlmsghdr*)skb_put(skb, NLMSG_ALIGN(size));
2115 nlh->nlmsg_type = type;
2116 nlh->nlmsg_len = size;
2117 nlh->nlmsg_flags = flags;
2118 nlh->nlmsg_pid = portid;
2119 nlh->nlmsg_seq = seq;
2120 if (!__builtin_constant_p(size) || NLMSG_ALIGN(size) - size != 0)
2121 memset(nlmsg_data(nlh) + len, 0, NLMSG_ALIGN(size) - size);
2122 return nlh;
2123 }
2124 EXPORT_SYMBOL(__nlmsg_put);
2125
2126 /*
2127 * It looks a bit ugly.
2128 * It would be better to create kernel thread.
2129 */
2130
2131 static int netlink_dump(struct sock *sk)
2132 {
2133 struct netlink_sock *nlk = nlk_sk(sk);
2134 struct netlink_callback *cb;
2135 struct sk_buff *skb = NULL;
2136 struct nlmsghdr *nlh;
2137 int len, err = -ENOBUFS;
2138 int alloc_size;
2139
2140 mutex_lock(nlk->cb_mutex);
2141
2142 cb = nlk->cb;
2143 if (cb == NULL) {
2144 err = -EINVAL;
2145 goto errout_skb;
2146 }
2147
2148 alloc_size = max_t(int, cb->min_dump_alloc, NLMSG_GOODSIZE);
2149
2150 skb = sock_rmalloc(sk, alloc_size, 0, GFP_KERNEL);
2151 if (!skb)
2152 goto errout_skb;
2153
2154 len = cb->dump(skb, cb);
2155
2156 if (len > 0) {
2157 mutex_unlock(nlk->cb_mutex);
2158
2159 if (sk_filter(sk, skb))
2160 kfree_skb(skb);
2161 else
2162 __netlink_sendskb(sk, skb);
2163 return 0;
2164 }
2165
2166 nlh = nlmsg_put_answer(skb, cb, NLMSG_DONE, sizeof(len), NLM_F_MULTI);
2167 if (!nlh)
2168 goto errout_skb;
2169
2170 nl_dump_check_consistent(cb, nlh);
2171
2172 memcpy(nlmsg_data(nlh), &len, sizeof(len));
2173
2174 if (sk_filter(sk, skb))
2175 kfree_skb(skb);
2176 else
2177 __netlink_sendskb(sk, skb);
2178
2179 if (cb->done)
2180 cb->done(cb);
2181 nlk->cb = NULL;
2182 mutex_unlock(nlk->cb_mutex);
2183
2184 module_put(cb->module);
2185 netlink_consume_callback(cb);
2186 return 0;
2187
2188 errout_skb:
2189 mutex_unlock(nlk->cb_mutex);
2190 kfree_skb(skb);
2191 return err;
2192 }
2193
2194 int __netlink_dump_start(struct sock *ssk, struct sk_buff *skb,
2195 const struct nlmsghdr *nlh,
2196 struct netlink_dump_control *control)
2197 {
2198 struct netlink_callback *cb;
2199 struct sock *sk;
2200 struct netlink_sock *nlk;
2201 int ret;
2202
2203 cb = kzalloc(sizeof(*cb), GFP_KERNEL);
2204 if (cb == NULL)
2205 return -ENOBUFS;
2206
2207 cb->dump = control->dump;
2208 cb->done = control->done;
2209 cb->nlh = nlh;
2210 cb->data = control->data;
2211 cb->module = control->module;
2212 cb->min_dump_alloc = control->min_dump_alloc;
2213 atomic_inc(&skb->users);
2214 cb->skb = skb;
2215
2216 sk = netlink_lookup(sock_net(ssk), ssk->sk_protocol, NETLINK_CB(skb).portid);
2217 if (sk == NULL) {
2218 netlink_destroy_callback(cb);
2219 return -ECONNREFUSED;
2220 }
2221 nlk = nlk_sk(sk);
2222
2223 mutex_lock(nlk->cb_mutex);
2224 /* A dump is in progress... */
2225 if (nlk->cb) {
2226 mutex_unlock(nlk->cb_mutex);
2227 netlink_destroy_callback(cb);
2228 ret = -EBUSY;
2229 goto out;
2230 }
2231 /* add reference of module which cb->dump belongs to */
2232 if (!try_module_get(cb->module)) {
2233 mutex_unlock(nlk->cb_mutex);
2234 netlink_destroy_callback(cb);
2235 ret = -EPROTONOSUPPORT;
2236 goto out;
2237 }
2238
2239 nlk->cb = cb;
2240 mutex_unlock(nlk->cb_mutex);
2241
2242 ret = netlink_dump(sk);
2243 out:
2244 sock_put(sk);
2245
2246 if (ret)
2247 return ret;
2248
2249 /* We successfully started a dump, by returning -EINTR we
2250 * signal not to send ACK even if it was requested.
2251 */
2252 return -EINTR;
2253 }
2254 EXPORT_SYMBOL(__netlink_dump_start);
2255
2256 void netlink_ack(struct sk_buff *in_skb, struct nlmsghdr *nlh, int err)
2257 {
2258 struct sk_buff *skb;
2259 struct nlmsghdr *rep;
2260 struct nlmsgerr *errmsg;
2261 size_t payload = sizeof(*errmsg);
2262
2263 /* error messages get the original request appened */
2264 if (err)
2265 payload += nlmsg_len(nlh);
2266
2267 skb = nlmsg_new(payload, GFP_KERNEL);
2268 if (!skb) {
2269 struct sock *sk;
2270
2271 sk = netlink_lookup(sock_net(in_skb->sk),
2272 in_skb->sk->sk_protocol,
2273 NETLINK_CB(in_skb).portid);
2274 if (sk) {
2275 sk->sk_err = ENOBUFS;
2276 sk->sk_error_report(sk);
2277 sock_put(sk);
2278 }
2279 return;
2280 }
2281
2282 rep = __nlmsg_put(skb, NETLINK_CB(in_skb).portid, nlh->nlmsg_seq,
2283 NLMSG_ERROR, payload, 0);
2284 errmsg = nlmsg_data(rep);
2285 errmsg->error = err;
2286 memcpy(&errmsg->msg, nlh, err ? nlh->nlmsg_len : sizeof(*nlh));
2287 netlink_unicast(in_skb->sk, skb, NETLINK_CB(in_skb).portid, MSG_DONTWAIT);
2288 }
2289 EXPORT_SYMBOL(netlink_ack);
2290
2291 int netlink_rcv_skb(struct sk_buff *skb, int (*cb)(struct sk_buff *,
2292 struct nlmsghdr *))
2293 {
2294 struct nlmsghdr *nlh;
2295 int err;
2296
2297 while (skb->len >= nlmsg_total_size(0)) {
2298 int msglen;
2299
2300 nlh = nlmsg_hdr(skb);
2301 err = 0;
2302
2303 if (nlh->nlmsg_len < NLMSG_HDRLEN || skb->len < nlh->nlmsg_len)
2304 return 0;
2305
2306 /* Only requests are handled by the kernel */
2307 if (!(nlh->nlmsg_flags & NLM_F_REQUEST))
2308 goto ack;
2309
2310 /* Skip control messages */
2311 if (nlh->nlmsg_type < NLMSG_MIN_TYPE)
2312 goto ack;
2313
2314 err = cb(skb, nlh);
2315 if (err == -EINTR)
2316 goto skip;
2317
2318 ack:
2319 if (nlh->nlmsg_flags & NLM_F_ACK || err)
2320 netlink_ack(skb, nlh, err);
2321
2322 skip:
2323 msglen = NLMSG_ALIGN(nlh->nlmsg_len);
2324 if (msglen > skb->len)
2325 msglen = skb->len;
2326 skb_pull(skb, msglen);
2327 }
2328
2329 return 0;
2330 }
2331 EXPORT_SYMBOL(netlink_rcv_skb);
2332
2333 /**
2334 * nlmsg_notify - send a notification netlink message
2335 * @sk: netlink socket to use
2336 * @skb: notification message
2337 * @portid: destination netlink portid for reports or 0
2338 * @group: destination multicast group or 0
2339 * @report: 1 to report back, 0 to disable
2340 * @flags: allocation flags
2341 */
2342 int nlmsg_notify(struct sock *sk, struct sk_buff *skb, u32 portid,
2343 unsigned int group, int report, gfp_t flags)
2344 {
2345 int err = 0;
2346
2347 if (group) {
2348 int exclude_portid = 0;
2349
2350 if (report) {
2351 atomic_inc(&skb->users);
2352 exclude_portid = portid;
2353 }
2354
2355 /* errors reported via destination sk->sk_err, but propagate
2356 * delivery errors if NETLINK_BROADCAST_ERROR flag is set */
2357 err = nlmsg_multicast(sk, skb, exclude_portid, group, flags);
2358 }
2359
2360 if (report) {
2361 int err2;
2362
2363 err2 = nlmsg_unicast(sk, skb, portid);
2364 if (!err || err == -ESRCH)
2365 err = err2;
2366 }
2367
2368 return err;
2369 }
2370 EXPORT_SYMBOL(nlmsg_notify);
2371
2372 #ifdef CONFIG_PROC_FS
2373 struct nl_seq_iter {
2374 struct seq_net_private p;
2375 int link;
2376 int hash_idx;
2377 };
2378
2379 static struct sock *netlink_seq_socket_idx(struct seq_file *seq, loff_t pos)
2380 {
2381 struct nl_seq_iter *iter = seq->private;
2382 int i, j;
2383 struct sock *s;
2384 loff_t off = 0;
2385
2386 for (i = 0; i < MAX_LINKS; i++) {
2387 struct nl_portid_hash *hash = &nl_table[i].hash;
2388
2389 for (j = 0; j <= hash->mask; j++) {
2390 sk_for_each(s, &hash->table[j]) {
2391 if (sock_net(s) != seq_file_net(seq))
2392 continue;
2393 if (off == pos) {
2394 iter->link = i;
2395 iter->hash_idx = j;
2396 return s;
2397 }
2398 ++off;
2399 }
2400 }
2401 }
2402 return NULL;
2403 }
2404
2405 static void *netlink_seq_start(struct seq_file *seq, loff_t *pos)
2406 __acquires(nl_table_lock)
2407 {
2408 read_lock(&nl_table_lock);
2409 return *pos ? netlink_seq_socket_idx(seq, *pos - 1) : SEQ_START_TOKEN;
2410 }
2411
2412 static void *netlink_seq_next(struct seq_file *seq, void *v, loff_t *pos)
2413 {
2414 struct sock *s;
2415 struct nl_seq_iter *iter;
2416 int i, j;
2417
2418 ++*pos;
2419
2420 if (v == SEQ_START_TOKEN)
2421 return netlink_seq_socket_idx(seq, 0);
2422
2423 iter = seq->private;
2424 s = v;
2425 do {
2426 s = sk_next(s);
2427 } while (s && sock_net(s) != seq_file_net(seq));
2428 if (s)
2429 return s;
2430
2431 i = iter->link;
2432 j = iter->hash_idx + 1;
2433
2434 do {
2435 struct nl_portid_hash *hash = &nl_table[i].hash;
2436
2437 for (; j <= hash->mask; j++) {
2438 s = sk_head(&hash->table[j]);
2439 while (s && sock_net(s) != seq_file_net(seq))
2440 s = sk_next(s);
2441 if (s) {
2442 iter->link = i;
2443 iter->hash_idx = j;
2444 return s;
2445 }
2446 }
2447
2448 j = 0;
2449 } while (++i < MAX_LINKS);
2450
2451 return NULL;
2452 }
2453
2454 static void netlink_seq_stop(struct seq_file *seq, void *v)
2455 __releases(nl_table_lock)
2456 {
2457 read_unlock(&nl_table_lock);
2458 }
2459
2460
2461 static int netlink_seq_show(struct seq_file *seq, void *v)
2462 {
2463 if (v == SEQ_START_TOKEN) {
2464 seq_puts(seq,
2465 "sk Eth Pid Groups "
2466 "Rmem Wmem Dump Locks Drops Inode\n");
2467 } else {
2468 struct sock *s = v;
2469 struct netlink_sock *nlk = nlk_sk(s);
2470
2471 seq_printf(seq, "%pK %-3d %-6u %08x %-8d %-8d %pK %-8d %-8d %-8lu\n",
2472 s,
2473 s->sk_protocol,
2474 nlk->portid,
2475 nlk->groups ? (u32)nlk->groups[0] : 0,
2476 sk_rmem_alloc_get(s),
2477 sk_wmem_alloc_get(s),
2478 nlk->cb,
2479 atomic_read(&s->sk_refcnt),
2480 atomic_read(&s->sk_drops),
2481 sock_i_ino(s)
2482 );
2483
2484 }
2485 return 0;
2486 }
2487
2488 static const struct seq_operations netlink_seq_ops = {
2489 .start = netlink_seq_start,
2490 .next = netlink_seq_next,
2491 .stop = netlink_seq_stop,
2492 .show = netlink_seq_show,
2493 };
2494
2495
2496 static int netlink_seq_open(struct inode *inode, struct file *file)
2497 {
2498 return seq_open_net(inode, file, &netlink_seq_ops,
2499 sizeof(struct nl_seq_iter));
2500 }
2501
2502 static const struct file_operations netlink_seq_fops = {
2503 .owner = THIS_MODULE,
2504 .open = netlink_seq_open,
2505 .read = seq_read,
2506 .llseek = seq_lseek,
2507 .release = seq_release_net,
2508 };
2509
2510 #endif
2511
2512 int netlink_register_notifier(struct notifier_block *nb)
2513 {
2514 return atomic_notifier_chain_register(&netlink_chain, nb);
2515 }
2516 EXPORT_SYMBOL(netlink_register_notifier);
2517
2518 int netlink_unregister_notifier(struct notifier_block *nb)
2519 {
2520 return atomic_notifier_chain_unregister(&netlink_chain, nb);
2521 }
2522 EXPORT_SYMBOL(netlink_unregister_notifier);
2523
2524 static const struct proto_ops netlink_ops = {
2525 .family = PF_NETLINK,
2526 .owner = THIS_MODULE,
2527 .release = netlink_release,
2528 .bind = netlink_bind,
2529 .connect = netlink_connect,
2530 .socketpair = sock_no_socketpair,
2531 .accept = sock_no_accept,
2532 .getname = netlink_getname,
2533 .poll = netlink_poll,
2534 .ioctl = sock_no_ioctl,
2535 .listen = sock_no_listen,
2536 .shutdown = sock_no_shutdown,
2537 .setsockopt = netlink_setsockopt,
2538 .getsockopt = netlink_getsockopt,
2539 .sendmsg = netlink_sendmsg,
2540 .recvmsg = netlink_recvmsg,
2541 .mmap = netlink_mmap,
2542 .sendpage = sock_no_sendpage,
2543 };
2544
2545 static const struct net_proto_family netlink_family_ops = {
2546 .family = PF_NETLINK,
2547 .create = netlink_create,
2548 .owner = THIS_MODULE, /* for consistency 8) */
2549 };
2550
2551 static int __net_init netlink_net_init(struct net *net)
2552 {
2553 #ifdef CONFIG_PROC_FS
2554 if (!proc_create("netlink", 0, net->proc_net, &netlink_seq_fops))
2555 return -ENOMEM;
2556 #endif
2557 return 0;
2558 }
2559
2560 static void __net_exit netlink_net_exit(struct net *net)
2561 {
2562 #ifdef CONFIG_PROC_FS
2563 remove_proc_entry("netlink", net->proc_net);
2564 #endif
2565 }
2566
2567 static void __init netlink_add_usersock_entry(void)
2568 {
2569 struct listeners *listeners;
2570 int groups = 32;
2571
2572 listeners = kzalloc(sizeof(*listeners) + NLGRPSZ(groups), GFP_KERNEL);
2573 if (!listeners)
2574 panic("netlink_add_usersock_entry: Cannot allocate listeners\n");
2575
2576 netlink_table_grab();
2577
2578 nl_table[NETLINK_USERSOCK].groups = groups;
2579 rcu_assign_pointer(nl_table[NETLINK_USERSOCK].listeners, listeners);
2580 nl_table[NETLINK_USERSOCK].module = THIS_MODULE;
2581 nl_table[NETLINK_USERSOCK].registered = 1;
2582 nl_table[NETLINK_USERSOCK].flags = NL_CFG_F_NONROOT_SEND;
2583
2584 netlink_table_ungrab();
2585 }
2586
2587 static struct pernet_operations __net_initdata netlink_net_ops = {
2588 .init = netlink_net_init,
2589 .exit = netlink_net_exit,
2590 };
2591
2592 static int __init netlink_proto_init(void)
2593 {
2594 int i;
2595 unsigned long limit;
2596 unsigned int order;
2597 int err = proto_register(&netlink_proto, 0);
2598
2599 if (err != 0)
2600 goto out;
2601
2602 BUILD_BUG_ON(sizeof(struct netlink_skb_parms) > FIELD_SIZEOF(struct sk_buff, cb));
2603
2604 nl_table = kcalloc(MAX_LINKS, sizeof(*nl_table), GFP_KERNEL);
2605 if (!nl_table)
2606 goto panic;
2607
2608 if (totalram_pages >= (128 * 1024))
2609 limit = totalram_pages >> (21 - PAGE_SHIFT);
2610 else
2611 limit = totalram_pages >> (23 - PAGE_SHIFT);
2612
2613 order = get_bitmask_order(limit) - 1 + PAGE_SHIFT;
2614 limit = (1UL << order) / sizeof(struct hlist_head);
2615 order = get_bitmask_order(min(limit, (unsigned long)UINT_MAX)) - 1;
2616
2617 for (i = 0; i < MAX_LINKS; i++) {
2618 struct nl_portid_hash *hash = &nl_table[i].hash;
2619
2620 hash->table = nl_portid_hash_zalloc(1 * sizeof(*hash->table));
2621 if (!hash->table) {
2622 while (i-- > 0)
2623 nl_portid_hash_free(nl_table[i].hash.table,
2624 1 * sizeof(*hash->table));
2625 kfree(nl_table);
2626 goto panic;
2627 }
2628 hash->max_shift = order;
2629 hash->shift = 0;
2630 hash->mask = 0;
2631 hash->rehash_time = jiffies;
2632 }
2633
2634 netlink_add_usersock_entry();
2635
2636 sock_register(&netlink_family_ops);
2637 register_pernet_subsys(&netlink_net_ops);
2638 /* The netlink device handler may be needed early. */
2639 rtnetlink_init();
2640 out:
2641 return err;
2642 panic:
2643 panic("netlink_init: Cannot allocate nl_table\n");
2644 }
2645
2646 core_initcall(netlink_proto_init);
kernel source for telekom puls (alcatel/mediatek)