Linux 4.1.16
[linux/fpc-iii.git] / net / netlink / af_netlink.c
blob0d6038c87bef76ef43aa30fce88726cb93288761
1 /*
2 * NETLINK Kernel-user communication protocol.
4 * Authors: Alan Cox <alan@lxorguk.ukuu.org.uk>
5 * Alexey Kuznetsov <kuznet@ms2.inr.ac.ru>
6 * Patrick McHardy <kaber@trash.net>
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.
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
25 #include <linux/module.h>
27 #include <linux/capability.h>
28 #include <linux/kernel.h>
29 #include <linux/init.h>
30 #include <linux/signal.h>
31 #include <linux/sched.h>
32 #include <linux/errno.h>
33 #include <linux/string.h>
34 #include <linux/stat.h>
35 #include <linux/socket.h>
36 #include <linux/un.h>
37 #include <linux/fcntl.h>
38 #include <linux/termios.h>
39 #include <linux/sockios.h>
40 #include <linux/net.h>
41 #include <linux/fs.h>
42 #include <linux/slab.h>
43 #include <asm/uaccess.h>
44 #include <linux/skbuff.h>
45 #include <linux/netdevice.h>
46 #include <linux/rtnetlink.h>
47 #include <linux/proc_fs.h>
48 #include <linux/seq_file.h>
49 #include <linux/notifier.h>
50 #include <linux/security.h>
51 #include <linux/jhash.h>
52 #include <linux/jiffies.h>
53 #include <linux/random.h>
54 #include <linux/bitops.h>
55 #include <linux/mm.h>
56 #include <linux/types.h>
57 #include <linux/audit.h>
58 #include <linux/mutex.h>
59 #include <linux/vmalloc.h>
60 #include <linux/if_arp.h>
61 #include <linux/rhashtable.h>
62 #include <asm/cacheflush.h>
63 #include <linux/hash.h>
64 #include <linux/genetlink.h>
66 #include <net/net_namespace.h>
67 #include <net/sock.h>
68 #include <net/scm.h>
69 #include <net/netlink.h>
71 #include "af_netlink.h"
73 struct listeners {
74 struct rcu_head rcu;
75 unsigned long masks[0];
78 /* state bits */
79 #define NETLINK_CONGESTED 0x0
81 /* flags */
82 #define NETLINK_KERNEL_SOCKET 0x1
83 #define NETLINK_RECV_PKTINFO 0x2
84 #define NETLINK_BROADCAST_SEND_ERROR 0x4
85 #define NETLINK_RECV_NO_ENOBUFS 0x8
87 static inline int netlink_is_kernel(struct sock *sk)
89 return nlk_sk(sk)->flags & NETLINK_KERNEL_SOCKET;
92 struct netlink_table *nl_table __read_mostly;
93 EXPORT_SYMBOL_GPL(nl_table);
95 static DECLARE_WAIT_QUEUE_HEAD(nl_table_wait);
97 static int netlink_dump(struct sock *sk);
98 static void netlink_skb_destructor(struct sk_buff *skb);
100 /* nl_table locking explained:
101 * Lookup and traversal are protected with an RCU read-side lock. Insertion
102 * and removal are protected with per bucket lock while using RCU list
103 * modification primitives and may run in parallel to RCU protected lookups.
104 * Destruction of the Netlink socket may only occur *after* nl_table_lock has
105 * been acquired * either during or after the socket has been removed from
106 * the list and after an RCU grace period.
108 DEFINE_RWLOCK(nl_table_lock);
109 EXPORT_SYMBOL_GPL(nl_table_lock);
110 static atomic_t nl_table_users = ATOMIC_INIT(0);
112 #define nl_deref_protected(X) rcu_dereference_protected(X, lockdep_is_held(&nl_table_lock));
114 static ATOMIC_NOTIFIER_HEAD(netlink_chain);
116 static DEFINE_SPINLOCK(netlink_tap_lock);
117 static struct list_head netlink_tap_all __read_mostly;
119 static const struct rhashtable_params netlink_rhashtable_params;
121 static inline u32 netlink_group_mask(u32 group)
123 return group ? 1 << (group - 1) : 0;
126 static struct sk_buff *netlink_to_full_skb(const struct sk_buff *skb,
127 gfp_t gfp_mask)
129 unsigned int len = skb_end_offset(skb);
130 struct sk_buff *new;
132 new = alloc_skb(len, gfp_mask);
133 if (new == NULL)
134 return NULL;
136 NETLINK_CB(new).portid = NETLINK_CB(skb).portid;
137 NETLINK_CB(new).dst_group = NETLINK_CB(skb).dst_group;
138 NETLINK_CB(new).creds = NETLINK_CB(skb).creds;
140 memcpy(skb_put(new, len), skb->data, len);
141 return new;
144 int netlink_add_tap(struct netlink_tap *nt)
146 if (unlikely(nt->dev->type != ARPHRD_NETLINK))
147 return -EINVAL;
149 spin_lock(&netlink_tap_lock);
150 list_add_rcu(&nt->list, &netlink_tap_all);
151 spin_unlock(&netlink_tap_lock);
153 __module_get(nt->module);
155 return 0;
157 EXPORT_SYMBOL_GPL(netlink_add_tap);
159 static int __netlink_remove_tap(struct netlink_tap *nt)
161 bool found = false;
162 struct netlink_tap *tmp;
164 spin_lock(&netlink_tap_lock);
166 list_for_each_entry(tmp, &netlink_tap_all, list) {
167 if (nt == tmp) {
168 list_del_rcu(&nt->list);
169 found = true;
170 goto out;
174 pr_warn("__netlink_remove_tap: %p not found\n", nt);
175 out:
176 spin_unlock(&netlink_tap_lock);
178 if (found && nt->module)
179 module_put(nt->module);
181 return found ? 0 : -ENODEV;
184 int netlink_remove_tap(struct netlink_tap *nt)
186 int ret;
188 ret = __netlink_remove_tap(nt);
189 synchronize_net();
191 return ret;
193 EXPORT_SYMBOL_GPL(netlink_remove_tap);
195 static bool netlink_filter_tap(const struct sk_buff *skb)
197 struct sock *sk = skb->sk;
199 /* We take the more conservative approach and
200 * whitelist socket protocols that may pass.
202 switch (sk->sk_protocol) {
203 case NETLINK_ROUTE:
204 case NETLINK_USERSOCK:
205 case NETLINK_SOCK_DIAG:
206 case NETLINK_NFLOG:
207 case NETLINK_XFRM:
208 case NETLINK_FIB_LOOKUP:
209 case NETLINK_NETFILTER:
210 case NETLINK_GENERIC:
211 return true;
214 return false;
217 static int __netlink_deliver_tap_skb(struct sk_buff *skb,
218 struct net_device *dev)
220 struct sk_buff *nskb;
221 struct sock *sk = skb->sk;
222 int ret = -ENOMEM;
224 dev_hold(dev);
226 if (netlink_skb_is_mmaped(skb) || is_vmalloc_addr(skb->head))
227 nskb = netlink_to_full_skb(skb, GFP_ATOMIC);
228 else
229 nskb = skb_clone(skb, GFP_ATOMIC);
230 if (nskb) {
231 nskb->dev = dev;
232 nskb->protocol = htons((u16) sk->sk_protocol);
233 nskb->pkt_type = netlink_is_kernel(sk) ?
234 PACKET_KERNEL : PACKET_USER;
235 skb_reset_network_header(nskb);
236 ret = dev_queue_xmit(nskb);
237 if (unlikely(ret > 0))
238 ret = net_xmit_errno(ret);
241 dev_put(dev);
242 return ret;
245 static void __netlink_deliver_tap(struct sk_buff *skb)
247 int ret;
248 struct netlink_tap *tmp;
250 if (!netlink_filter_tap(skb))
251 return;
253 list_for_each_entry_rcu(tmp, &netlink_tap_all, list) {
254 ret = __netlink_deliver_tap_skb(skb, tmp->dev);
255 if (unlikely(ret))
256 break;
260 static void netlink_deliver_tap(struct sk_buff *skb)
262 rcu_read_lock();
264 if (unlikely(!list_empty(&netlink_tap_all)))
265 __netlink_deliver_tap(skb);
267 rcu_read_unlock();
270 static void netlink_deliver_tap_kernel(struct sock *dst, struct sock *src,
271 struct sk_buff *skb)
273 if (!(netlink_is_kernel(dst) && netlink_is_kernel(src)))
274 netlink_deliver_tap(skb);
277 static void netlink_overrun(struct sock *sk)
279 struct netlink_sock *nlk = nlk_sk(sk);
281 if (!(nlk->flags & NETLINK_RECV_NO_ENOBUFS)) {
282 if (!test_and_set_bit(NETLINK_CONGESTED, &nlk_sk(sk)->state)) {
283 sk->sk_err = ENOBUFS;
284 sk->sk_error_report(sk);
287 atomic_inc(&sk->sk_drops);
290 static void netlink_rcv_wake(struct sock *sk)
292 struct netlink_sock *nlk = nlk_sk(sk);
294 if (skb_queue_empty(&sk->sk_receive_queue))
295 clear_bit(NETLINK_CONGESTED, &nlk->state);
296 if (!test_bit(NETLINK_CONGESTED, &nlk->state))
297 wake_up_interruptible(&nlk->wait);
300 #ifdef CONFIG_NETLINK_MMAP
301 static bool netlink_rx_is_mmaped(struct sock *sk)
303 return nlk_sk(sk)->rx_ring.pg_vec != NULL;
306 static bool netlink_tx_is_mmaped(struct sock *sk)
308 return nlk_sk(sk)->tx_ring.pg_vec != NULL;
311 static __pure struct page *pgvec_to_page(const void *addr)
313 if (is_vmalloc_addr(addr))
314 return vmalloc_to_page(addr);
315 else
316 return virt_to_page(addr);
319 static void free_pg_vec(void **pg_vec, unsigned int order, unsigned int len)
321 unsigned int i;
323 for (i = 0; i < len; i++) {
324 if (pg_vec[i] != NULL) {
325 if (is_vmalloc_addr(pg_vec[i]))
326 vfree(pg_vec[i]);
327 else
328 free_pages((unsigned long)pg_vec[i], order);
331 kfree(pg_vec);
334 static void *alloc_one_pg_vec_page(unsigned long order)
336 void *buffer;
337 gfp_t gfp_flags = GFP_KERNEL | __GFP_COMP | __GFP_ZERO |
338 __GFP_NOWARN | __GFP_NORETRY;
340 buffer = (void *)__get_free_pages(gfp_flags, order);
341 if (buffer != NULL)
342 return buffer;
344 buffer = vzalloc((1 << order) * PAGE_SIZE);
345 if (buffer != NULL)
346 return buffer;
348 gfp_flags &= ~__GFP_NORETRY;
349 return (void *)__get_free_pages(gfp_flags, order);
352 static void **alloc_pg_vec(struct netlink_sock *nlk,
353 struct nl_mmap_req *req, unsigned int order)
355 unsigned int block_nr = req->nm_block_nr;
356 unsigned int i;
357 void **pg_vec;
359 pg_vec = kcalloc(block_nr, sizeof(void *), GFP_KERNEL);
360 if (pg_vec == NULL)
361 return NULL;
363 for (i = 0; i < block_nr; i++) {
364 pg_vec[i] = alloc_one_pg_vec_page(order);
365 if (pg_vec[i] == NULL)
366 goto err1;
369 return pg_vec;
370 err1:
371 free_pg_vec(pg_vec, order, block_nr);
372 return NULL;
376 static void
377 __netlink_set_ring(struct sock *sk, struct nl_mmap_req *req, bool tx_ring, void **pg_vec,
378 unsigned int order)
380 struct netlink_sock *nlk = nlk_sk(sk);
381 struct sk_buff_head *queue;
382 struct netlink_ring *ring;
384 queue = tx_ring ? &sk->sk_write_queue : &sk->sk_receive_queue;
385 ring = tx_ring ? &nlk->tx_ring : &nlk->rx_ring;
387 spin_lock_bh(&queue->lock);
389 ring->frame_max = req->nm_frame_nr - 1;
390 ring->head = 0;
391 ring->frame_size = req->nm_frame_size;
392 ring->pg_vec_pages = req->nm_block_size / PAGE_SIZE;
394 swap(ring->pg_vec_len, req->nm_block_nr);
395 swap(ring->pg_vec_order, order);
396 swap(ring->pg_vec, pg_vec);
398 __skb_queue_purge(queue);
399 spin_unlock_bh(&queue->lock);
401 WARN_ON(atomic_read(&nlk->mapped));
403 if (pg_vec)
404 free_pg_vec(pg_vec, order, req->nm_block_nr);
407 static int netlink_set_ring(struct sock *sk, struct nl_mmap_req *req,
408 bool tx_ring)
410 struct netlink_sock *nlk = nlk_sk(sk);
411 struct netlink_ring *ring;
412 void **pg_vec = NULL;
413 unsigned int order = 0;
415 ring = tx_ring ? &nlk->tx_ring : &nlk->rx_ring;
417 if (atomic_read(&nlk->mapped))
418 return -EBUSY;
419 if (atomic_read(&ring->pending))
420 return -EBUSY;
422 if (req->nm_block_nr) {
423 if (ring->pg_vec != NULL)
424 return -EBUSY;
426 if ((int)req->nm_block_size <= 0)
427 return -EINVAL;
428 if (!PAGE_ALIGNED(req->nm_block_size))
429 return -EINVAL;
430 if (req->nm_frame_size < NL_MMAP_HDRLEN)
431 return -EINVAL;
432 if (!IS_ALIGNED(req->nm_frame_size, NL_MMAP_MSG_ALIGNMENT))
433 return -EINVAL;
435 ring->frames_per_block = req->nm_block_size /
436 req->nm_frame_size;
437 if (ring->frames_per_block == 0)
438 return -EINVAL;
439 if (ring->frames_per_block * req->nm_block_nr !=
440 req->nm_frame_nr)
441 return -EINVAL;
443 order = get_order(req->nm_block_size);
444 pg_vec = alloc_pg_vec(nlk, req, order);
445 if (pg_vec == NULL)
446 return -ENOMEM;
447 } else {
448 if (req->nm_frame_nr)
449 return -EINVAL;
452 mutex_lock(&nlk->pg_vec_lock);
453 if (atomic_read(&nlk->mapped) == 0) {
454 __netlink_set_ring(sk, req, tx_ring, pg_vec, order);
455 mutex_unlock(&nlk->pg_vec_lock);
456 return 0;
459 mutex_unlock(&nlk->pg_vec_lock);
461 if (pg_vec)
462 free_pg_vec(pg_vec, order, req->nm_block_nr);
464 return -EBUSY;
467 static void netlink_mm_open(struct vm_area_struct *vma)
469 struct file *file = vma->vm_file;
470 struct socket *sock = file->private_data;
471 struct sock *sk = sock->sk;
473 if (sk)
474 atomic_inc(&nlk_sk(sk)->mapped);
477 static void netlink_mm_close(struct vm_area_struct *vma)
479 struct file *file = vma->vm_file;
480 struct socket *sock = file->private_data;
481 struct sock *sk = sock->sk;
483 if (sk)
484 atomic_dec(&nlk_sk(sk)->mapped);
487 static const struct vm_operations_struct netlink_mmap_ops = {
488 .open = netlink_mm_open,
489 .close = netlink_mm_close,
492 static int netlink_mmap(struct file *file, struct socket *sock,
493 struct vm_area_struct *vma)
495 struct sock *sk = sock->sk;
496 struct netlink_sock *nlk = nlk_sk(sk);
497 struct netlink_ring *ring;
498 unsigned long start, size, expected;
499 unsigned int i;
500 int err = -EINVAL;
502 if (vma->vm_pgoff)
503 return -EINVAL;
505 mutex_lock(&nlk->pg_vec_lock);
507 expected = 0;
508 for (ring = &nlk->rx_ring; ring <= &nlk->tx_ring; ring++) {
509 if (ring->pg_vec == NULL)
510 continue;
511 expected += ring->pg_vec_len * ring->pg_vec_pages * PAGE_SIZE;
514 if (expected == 0)
515 goto out;
517 size = vma->vm_end - vma->vm_start;
518 if (size != expected)
519 goto out;
521 start = vma->vm_start;
522 for (ring = &nlk->rx_ring; ring <= &nlk->tx_ring; ring++) {
523 if (ring->pg_vec == NULL)
524 continue;
526 for (i = 0; i < ring->pg_vec_len; i++) {
527 struct page *page;
528 void *kaddr = ring->pg_vec[i];
529 unsigned int pg_num;
531 for (pg_num = 0; pg_num < ring->pg_vec_pages; pg_num++) {
532 page = pgvec_to_page(kaddr);
533 err = vm_insert_page(vma, start, page);
534 if (err < 0)
535 goto out;
536 start += PAGE_SIZE;
537 kaddr += PAGE_SIZE;
542 atomic_inc(&nlk->mapped);
543 vma->vm_ops = &netlink_mmap_ops;
544 err = 0;
545 out:
546 mutex_unlock(&nlk->pg_vec_lock);
547 return err;
550 static void netlink_frame_flush_dcache(const struct nl_mmap_hdr *hdr, unsigned int nm_len)
552 #if ARCH_IMPLEMENTS_FLUSH_DCACHE_PAGE == 1
553 struct page *p_start, *p_end;
555 /* First page is flushed through netlink_{get,set}_status */
556 p_start = pgvec_to_page(hdr + PAGE_SIZE);
557 p_end = pgvec_to_page((void *)hdr + NL_MMAP_HDRLEN + nm_len - 1);
558 while (p_start <= p_end) {
559 flush_dcache_page(p_start);
560 p_start++;
562 #endif
565 static enum nl_mmap_status netlink_get_status(const struct nl_mmap_hdr *hdr)
567 smp_rmb();
568 flush_dcache_page(pgvec_to_page(hdr));
569 return hdr->nm_status;
572 static void netlink_set_status(struct nl_mmap_hdr *hdr,
573 enum nl_mmap_status status)
575 smp_mb();
576 hdr->nm_status = status;
577 flush_dcache_page(pgvec_to_page(hdr));
580 static struct nl_mmap_hdr *
581 __netlink_lookup_frame(const struct netlink_ring *ring, unsigned int pos)
583 unsigned int pg_vec_pos, frame_off;
585 pg_vec_pos = pos / ring->frames_per_block;
586 frame_off = pos % ring->frames_per_block;
588 return ring->pg_vec[pg_vec_pos] + (frame_off * ring->frame_size);
591 static struct nl_mmap_hdr *
592 netlink_lookup_frame(const struct netlink_ring *ring, unsigned int pos,
593 enum nl_mmap_status status)
595 struct nl_mmap_hdr *hdr;
597 hdr = __netlink_lookup_frame(ring, pos);
598 if (netlink_get_status(hdr) != status)
599 return NULL;
601 return hdr;
604 static struct nl_mmap_hdr *
605 netlink_current_frame(const struct netlink_ring *ring,
606 enum nl_mmap_status status)
608 return netlink_lookup_frame(ring, ring->head, status);
611 static struct nl_mmap_hdr *
612 netlink_previous_frame(const struct netlink_ring *ring,
613 enum nl_mmap_status status)
615 unsigned int prev;
617 prev = ring->head ? ring->head - 1 : ring->frame_max;
618 return netlink_lookup_frame(ring, prev, status);
621 static void netlink_increment_head(struct netlink_ring *ring)
623 ring->head = ring->head != ring->frame_max ? ring->head + 1 : 0;
626 static void netlink_forward_ring(struct netlink_ring *ring)
628 unsigned int head = ring->head, pos = head;
629 const struct nl_mmap_hdr *hdr;
631 do {
632 hdr = __netlink_lookup_frame(ring, pos);
633 if (hdr->nm_status == NL_MMAP_STATUS_UNUSED)
634 break;
635 if (hdr->nm_status != NL_MMAP_STATUS_SKIP)
636 break;
637 netlink_increment_head(ring);
638 } while (ring->head != head);
641 static bool netlink_dump_space(struct netlink_sock *nlk)
643 struct netlink_ring *ring = &nlk->rx_ring;
644 struct nl_mmap_hdr *hdr;
645 unsigned int n;
647 hdr = netlink_current_frame(ring, NL_MMAP_STATUS_UNUSED);
648 if (hdr == NULL)
649 return false;
651 n = ring->head + ring->frame_max / 2;
652 if (n > ring->frame_max)
653 n -= ring->frame_max;
655 hdr = __netlink_lookup_frame(ring, n);
657 return hdr->nm_status == NL_MMAP_STATUS_UNUSED;
660 static unsigned int netlink_poll(struct file *file, struct socket *sock,
661 poll_table *wait)
663 struct sock *sk = sock->sk;
664 struct netlink_sock *nlk = nlk_sk(sk);
665 unsigned int mask;
666 int err;
668 if (nlk->rx_ring.pg_vec != NULL) {
669 /* Memory mapped sockets don't call recvmsg(), so flow control
670 * for dumps is performed here. A dump is allowed to continue
671 * if at least half the ring is unused.
673 while (nlk->cb_running && netlink_dump_space(nlk)) {
674 err = netlink_dump(sk);
675 if (err < 0) {
676 sk->sk_err = -err;
677 sk->sk_error_report(sk);
678 break;
681 netlink_rcv_wake(sk);
684 mask = datagram_poll(file, sock, wait);
686 spin_lock_bh(&sk->sk_receive_queue.lock);
687 if (nlk->rx_ring.pg_vec) {
688 netlink_forward_ring(&nlk->rx_ring);
689 if (!netlink_previous_frame(&nlk->rx_ring, NL_MMAP_STATUS_UNUSED))
690 mask |= POLLIN | POLLRDNORM;
692 spin_unlock_bh(&sk->sk_receive_queue.lock);
694 spin_lock_bh(&sk->sk_write_queue.lock);
695 if (nlk->tx_ring.pg_vec) {
696 if (netlink_current_frame(&nlk->tx_ring, NL_MMAP_STATUS_UNUSED))
697 mask |= POLLOUT | POLLWRNORM;
699 spin_unlock_bh(&sk->sk_write_queue.lock);
701 return mask;
704 static struct nl_mmap_hdr *netlink_mmap_hdr(struct sk_buff *skb)
706 return (struct nl_mmap_hdr *)(skb->head - NL_MMAP_HDRLEN);
709 static void netlink_ring_setup_skb(struct sk_buff *skb, struct sock *sk,
710 struct netlink_ring *ring,
711 struct nl_mmap_hdr *hdr)
713 unsigned int size;
714 void *data;
716 size = ring->frame_size - NL_MMAP_HDRLEN;
717 data = (void *)hdr + NL_MMAP_HDRLEN;
719 skb->head = data;
720 skb->data = data;
721 skb_reset_tail_pointer(skb);
722 skb->end = skb->tail + size;
723 skb->len = 0;
725 skb->destructor = netlink_skb_destructor;
726 NETLINK_CB(skb).flags |= NETLINK_SKB_MMAPED;
727 NETLINK_CB(skb).sk = sk;
730 static int netlink_mmap_sendmsg(struct sock *sk, struct msghdr *msg,
731 u32 dst_portid, u32 dst_group,
732 struct scm_cookie *scm)
734 struct netlink_sock *nlk = nlk_sk(sk);
735 struct netlink_ring *ring;
736 struct nl_mmap_hdr *hdr;
737 struct sk_buff *skb;
738 unsigned int maxlen;
739 int err = 0, len = 0;
741 mutex_lock(&nlk->pg_vec_lock);
743 ring = &nlk->tx_ring;
744 maxlen = ring->frame_size - NL_MMAP_HDRLEN;
746 do {
747 unsigned int nm_len;
749 hdr = netlink_current_frame(ring, NL_MMAP_STATUS_VALID);
750 if (hdr == NULL) {
751 if (!(msg->msg_flags & MSG_DONTWAIT) &&
752 atomic_read(&nlk->tx_ring.pending))
753 schedule();
754 continue;
757 nm_len = ACCESS_ONCE(hdr->nm_len);
758 if (nm_len > maxlen) {
759 err = -EINVAL;
760 goto out;
763 netlink_frame_flush_dcache(hdr, nm_len);
765 skb = alloc_skb(nm_len, GFP_KERNEL);
766 if (skb == NULL) {
767 err = -ENOBUFS;
768 goto out;
770 __skb_put(skb, nm_len);
771 memcpy(skb->data, (void *)hdr + NL_MMAP_HDRLEN, nm_len);
772 netlink_set_status(hdr, NL_MMAP_STATUS_UNUSED);
774 netlink_increment_head(ring);
776 NETLINK_CB(skb).portid = nlk->portid;
777 NETLINK_CB(skb).dst_group = dst_group;
778 NETLINK_CB(skb).creds = scm->creds;
780 err = security_netlink_send(sk, skb);
781 if (err) {
782 kfree_skb(skb);
783 goto out;
786 if (unlikely(dst_group)) {
787 atomic_inc(&skb->users);
788 netlink_broadcast(sk, skb, dst_portid, dst_group,
789 GFP_KERNEL);
791 err = netlink_unicast(sk, skb, dst_portid,
792 msg->msg_flags & MSG_DONTWAIT);
793 if (err < 0)
794 goto out;
795 len += err;
797 } while (hdr != NULL ||
798 (!(msg->msg_flags & MSG_DONTWAIT) &&
799 atomic_read(&nlk->tx_ring.pending)));
801 if (len > 0)
802 err = len;
803 out:
804 mutex_unlock(&nlk->pg_vec_lock);
805 return err;
808 static void netlink_queue_mmaped_skb(struct sock *sk, struct sk_buff *skb)
810 struct nl_mmap_hdr *hdr;
812 hdr = netlink_mmap_hdr(skb);
813 hdr->nm_len = skb->len;
814 hdr->nm_group = NETLINK_CB(skb).dst_group;
815 hdr->nm_pid = NETLINK_CB(skb).creds.pid;
816 hdr->nm_uid = from_kuid(sk_user_ns(sk), NETLINK_CB(skb).creds.uid);
817 hdr->nm_gid = from_kgid(sk_user_ns(sk), NETLINK_CB(skb).creds.gid);
818 netlink_frame_flush_dcache(hdr, hdr->nm_len);
819 netlink_set_status(hdr, NL_MMAP_STATUS_VALID);
821 NETLINK_CB(skb).flags |= NETLINK_SKB_DELIVERED;
822 kfree_skb(skb);
825 static void netlink_ring_set_copied(struct sock *sk, struct sk_buff *skb)
827 struct netlink_sock *nlk = nlk_sk(sk);
828 struct netlink_ring *ring = &nlk->rx_ring;
829 struct nl_mmap_hdr *hdr;
831 spin_lock_bh(&sk->sk_receive_queue.lock);
832 hdr = netlink_current_frame(ring, NL_MMAP_STATUS_UNUSED);
833 if (hdr == NULL) {
834 spin_unlock_bh(&sk->sk_receive_queue.lock);
835 kfree_skb(skb);
836 netlink_overrun(sk);
837 return;
839 netlink_increment_head(ring);
840 __skb_queue_tail(&sk->sk_receive_queue, skb);
841 spin_unlock_bh(&sk->sk_receive_queue.lock);
843 hdr->nm_len = skb->len;
844 hdr->nm_group = NETLINK_CB(skb).dst_group;
845 hdr->nm_pid = NETLINK_CB(skb).creds.pid;
846 hdr->nm_uid = from_kuid(sk_user_ns(sk), NETLINK_CB(skb).creds.uid);
847 hdr->nm_gid = from_kgid(sk_user_ns(sk), NETLINK_CB(skb).creds.gid);
848 netlink_set_status(hdr, NL_MMAP_STATUS_COPY);
851 #else /* CONFIG_NETLINK_MMAP */
852 #define netlink_rx_is_mmaped(sk) false
853 #define netlink_tx_is_mmaped(sk) false
854 #define netlink_mmap sock_no_mmap
855 #define netlink_poll datagram_poll
856 #define netlink_mmap_sendmsg(sk, msg, dst_portid, dst_group, scm) 0
857 #endif /* CONFIG_NETLINK_MMAP */
859 static void netlink_skb_destructor(struct sk_buff *skb)
861 #ifdef CONFIG_NETLINK_MMAP
862 struct nl_mmap_hdr *hdr;
863 struct netlink_ring *ring;
864 struct sock *sk;
866 /* If a packet from the kernel to userspace was freed because of an
867 * error without being delivered to userspace, the kernel must reset
868 * the status. In the direction userspace to kernel, the status is
869 * always reset here after the packet was processed and freed.
871 if (netlink_skb_is_mmaped(skb)) {
872 hdr = netlink_mmap_hdr(skb);
873 sk = NETLINK_CB(skb).sk;
875 if (NETLINK_CB(skb).flags & NETLINK_SKB_TX) {
876 netlink_set_status(hdr, NL_MMAP_STATUS_UNUSED);
877 ring = &nlk_sk(sk)->tx_ring;
878 } else {
879 if (!(NETLINK_CB(skb).flags & NETLINK_SKB_DELIVERED)) {
880 hdr->nm_len = 0;
881 netlink_set_status(hdr, NL_MMAP_STATUS_VALID);
883 ring = &nlk_sk(sk)->rx_ring;
886 WARN_ON(atomic_read(&ring->pending) == 0);
887 atomic_dec(&ring->pending);
888 sock_put(sk);
890 skb->head = NULL;
892 #endif
893 if (is_vmalloc_addr(skb->head)) {
894 if (!skb->cloned ||
895 !atomic_dec_return(&(skb_shinfo(skb)->dataref)))
896 vfree(skb->head);
898 skb->head = NULL;
900 if (skb->sk != NULL)
901 sock_rfree(skb);
904 static void netlink_skb_set_owner_r(struct sk_buff *skb, struct sock *sk)
906 WARN_ON(skb->sk != NULL);
907 skb->sk = sk;
908 skb->destructor = netlink_skb_destructor;
909 atomic_add(skb->truesize, &sk->sk_rmem_alloc);
910 sk_mem_charge(sk, skb->truesize);
913 static void netlink_sock_destruct(struct sock *sk)
915 struct netlink_sock *nlk = nlk_sk(sk);
917 if (nlk->cb_running) {
918 if (nlk->cb.done)
919 nlk->cb.done(&nlk->cb);
921 module_put(nlk->cb.module);
922 kfree_skb(nlk->cb.skb);
925 skb_queue_purge(&sk->sk_receive_queue);
926 #ifdef CONFIG_NETLINK_MMAP
927 if (1) {
928 struct nl_mmap_req req;
930 memset(&req, 0, sizeof(req));
931 if (nlk->rx_ring.pg_vec)
932 __netlink_set_ring(sk, &req, false, NULL, 0);
933 memset(&req, 0, sizeof(req));
934 if (nlk->tx_ring.pg_vec)
935 __netlink_set_ring(sk, &req, true, NULL, 0);
937 #endif /* CONFIG_NETLINK_MMAP */
939 if (!sock_flag(sk, SOCK_DEAD)) {
940 printk(KERN_ERR "Freeing alive netlink socket %p\n", sk);
941 return;
944 WARN_ON(atomic_read(&sk->sk_rmem_alloc));
945 WARN_ON(atomic_read(&sk->sk_wmem_alloc));
946 WARN_ON(nlk_sk(sk)->groups);
949 /* This lock without WQ_FLAG_EXCLUSIVE is good on UP and it is _very_ bad on
950 * SMP. Look, when several writers sleep and reader wakes them up, all but one
951 * immediately hit write lock and grab all the cpus. Exclusive sleep solves
952 * this, _but_ remember, it adds useless work on UP machines.
955 void netlink_table_grab(void)
956 __acquires(nl_table_lock)
958 might_sleep();
960 write_lock_irq(&nl_table_lock);
962 if (atomic_read(&nl_table_users)) {
963 DECLARE_WAITQUEUE(wait, current);
965 add_wait_queue_exclusive(&nl_table_wait, &wait);
966 for (;;) {
967 set_current_state(TASK_UNINTERRUPTIBLE);
968 if (atomic_read(&nl_table_users) == 0)
969 break;
970 write_unlock_irq(&nl_table_lock);
971 schedule();
972 write_lock_irq(&nl_table_lock);
975 __set_current_state(TASK_RUNNING);
976 remove_wait_queue(&nl_table_wait, &wait);
980 void netlink_table_ungrab(void)
981 __releases(nl_table_lock)
983 write_unlock_irq(&nl_table_lock);
984 wake_up(&nl_table_wait);
987 static inline void
988 netlink_lock_table(void)
990 /* read_lock() synchronizes us to netlink_table_grab */
992 read_lock(&nl_table_lock);
993 atomic_inc(&nl_table_users);
994 read_unlock(&nl_table_lock);
997 static inline void
998 netlink_unlock_table(void)
1000 if (atomic_dec_and_test(&nl_table_users))
1001 wake_up(&nl_table_wait);
1004 struct netlink_compare_arg
1006 possible_net_t pnet;
1007 u32 portid;
1010 /* Doing sizeof directly may yield 4 extra bytes on 64-bit. */
1011 #define netlink_compare_arg_len \
1012 (offsetof(struct netlink_compare_arg, portid) + sizeof(u32))
1014 static inline int netlink_compare(struct rhashtable_compare_arg *arg,
1015 const void *ptr)
1017 const struct netlink_compare_arg *x = arg->key;
1018 const struct netlink_sock *nlk = ptr;
1020 return nlk->portid != x->portid ||
1021 !net_eq(sock_net(&nlk->sk), read_pnet(&x->pnet));
1024 static void netlink_compare_arg_init(struct netlink_compare_arg *arg,
1025 struct net *net, u32 portid)
1027 memset(arg, 0, sizeof(*arg));
1028 write_pnet(&arg->pnet, net);
1029 arg->portid = portid;
1032 static struct sock *__netlink_lookup(struct netlink_table *table, u32 portid,
1033 struct net *net)
1035 struct netlink_compare_arg arg;
1037 netlink_compare_arg_init(&arg, net, portid);
1038 return rhashtable_lookup_fast(&table->hash, &arg,
1039 netlink_rhashtable_params);
1042 static int __netlink_insert(struct netlink_table *table, struct sock *sk)
1044 struct netlink_compare_arg arg;
1046 netlink_compare_arg_init(&arg, sock_net(sk), nlk_sk(sk)->portid);
1047 return rhashtable_lookup_insert_key(&table->hash, &arg,
1048 &nlk_sk(sk)->node,
1049 netlink_rhashtable_params);
1052 static struct sock *netlink_lookup(struct net *net, int protocol, u32 portid)
1054 struct netlink_table *table = &nl_table[protocol];
1055 struct sock *sk;
1057 rcu_read_lock();
1058 sk = __netlink_lookup(table, portid, net);
1059 if (sk)
1060 sock_hold(sk);
1061 rcu_read_unlock();
1063 return sk;
1066 static const struct proto_ops netlink_ops;
1068 static void
1069 netlink_update_listeners(struct sock *sk)
1071 struct netlink_table *tbl = &nl_table[sk->sk_protocol];
1072 unsigned long mask;
1073 unsigned int i;
1074 struct listeners *listeners;
1076 listeners = nl_deref_protected(tbl->listeners);
1077 if (!listeners)
1078 return;
1080 for (i = 0; i < NLGRPLONGS(tbl->groups); i++) {
1081 mask = 0;
1082 sk_for_each_bound(sk, &tbl->mc_list) {
1083 if (i < NLGRPLONGS(nlk_sk(sk)->ngroups))
1084 mask |= nlk_sk(sk)->groups[i];
1086 listeners->masks[i] = mask;
1088 /* this function is only called with the netlink table "grabbed", which
1089 * makes sure updates are visible before bind or setsockopt return. */
1092 static int netlink_insert(struct sock *sk, u32 portid)
1094 struct netlink_table *table = &nl_table[sk->sk_protocol];
1095 int err;
1097 lock_sock(sk);
1099 err = nlk_sk(sk)->portid == portid ? 0 : -EBUSY;
1100 if (nlk_sk(sk)->bound)
1101 goto err;
1103 err = -ENOMEM;
1104 if (BITS_PER_LONG > 32 &&
1105 unlikely(atomic_read(&table->hash.nelems) >= UINT_MAX))
1106 goto err;
1108 nlk_sk(sk)->portid = portid;
1109 sock_hold(sk);
1111 err = __netlink_insert(table, sk);
1112 if (err) {
1113 /* In case the hashtable backend returns with -EBUSY
1114 * from here, it must not escape to the caller.
1116 if (unlikely(err == -EBUSY))
1117 err = -EOVERFLOW;
1118 if (err == -EEXIST)
1119 err = -EADDRINUSE;
1120 sock_put(sk);
1121 goto err;
1124 /* We need to ensure that the socket is hashed and visible. */
1125 smp_wmb();
1126 nlk_sk(sk)->bound = portid;
1128 err:
1129 release_sock(sk);
1130 return err;
1133 static void netlink_remove(struct sock *sk)
1135 struct netlink_table *table;
1137 table = &nl_table[sk->sk_protocol];
1138 if (!rhashtable_remove_fast(&table->hash, &nlk_sk(sk)->node,
1139 netlink_rhashtable_params)) {
1140 WARN_ON(atomic_read(&sk->sk_refcnt) == 1);
1141 __sock_put(sk);
1144 netlink_table_grab();
1145 if (nlk_sk(sk)->subscriptions) {
1146 __sk_del_bind_node(sk);
1147 netlink_update_listeners(sk);
1149 if (sk->sk_protocol == NETLINK_GENERIC)
1150 atomic_inc(&genl_sk_destructing_cnt);
1151 netlink_table_ungrab();
1154 static struct proto netlink_proto = {
1155 .name = "NETLINK",
1156 .owner = THIS_MODULE,
1157 .obj_size = sizeof(struct netlink_sock),
1160 static int __netlink_create(struct net *net, struct socket *sock,
1161 struct mutex *cb_mutex, int protocol)
1163 struct sock *sk;
1164 struct netlink_sock *nlk;
1166 sock->ops = &netlink_ops;
1168 sk = sk_alloc(net, PF_NETLINK, GFP_KERNEL, &netlink_proto);
1169 if (!sk)
1170 return -ENOMEM;
1172 sock_init_data(sock, sk);
1174 nlk = nlk_sk(sk);
1175 if (cb_mutex) {
1176 nlk->cb_mutex = cb_mutex;
1177 } else {
1178 nlk->cb_mutex = &nlk->cb_def_mutex;
1179 mutex_init(nlk->cb_mutex);
1181 init_waitqueue_head(&nlk->wait);
1182 #ifdef CONFIG_NETLINK_MMAP
1183 mutex_init(&nlk->pg_vec_lock);
1184 #endif
1186 sk->sk_destruct = netlink_sock_destruct;
1187 sk->sk_protocol = protocol;
1188 return 0;
1191 static int netlink_create(struct net *net, struct socket *sock, int protocol,
1192 int kern)
1194 struct module *module = NULL;
1195 struct mutex *cb_mutex;
1196 struct netlink_sock *nlk;
1197 int (*bind)(struct net *net, int group);
1198 void (*unbind)(struct net *net, int group);
1199 int err = 0;
1201 sock->state = SS_UNCONNECTED;
1203 if (sock->type != SOCK_RAW && sock->type != SOCK_DGRAM)
1204 return -ESOCKTNOSUPPORT;
1206 if (protocol < 0 || protocol >= MAX_LINKS)
1207 return -EPROTONOSUPPORT;
1209 netlink_lock_table();
1210 #ifdef CONFIG_MODULES
1211 if (!nl_table[protocol].registered) {
1212 netlink_unlock_table();
1213 request_module("net-pf-%d-proto-%d", PF_NETLINK, protocol);
1214 netlink_lock_table();
1216 #endif
1217 if (nl_table[protocol].registered &&
1218 try_module_get(nl_table[protocol].module))
1219 module = nl_table[protocol].module;
1220 else
1221 err = -EPROTONOSUPPORT;
1222 cb_mutex = nl_table[protocol].cb_mutex;
1223 bind = nl_table[protocol].bind;
1224 unbind = nl_table[protocol].unbind;
1225 netlink_unlock_table();
1227 if (err < 0)
1228 goto out;
1230 err = __netlink_create(net, sock, cb_mutex, protocol);
1231 if (err < 0)
1232 goto out_module;
1234 local_bh_disable();
1235 sock_prot_inuse_add(net, &netlink_proto, 1);
1236 local_bh_enable();
1238 nlk = nlk_sk(sock->sk);
1239 nlk->module = module;
1240 nlk->netlink_bind = bind;
1241 nlk->netlink_unbind = unbind;
1242 out:
1243 return err;
1245 out_module:
1246 module_put(module);
1247 goto out;
1250 static void deferred_put_nlk_sk(struct rcu_head *head)
1252 struct netlink_sock *nlk = container_of(head, struct netlink_sock, rcu);
1254 sock_put(&nlk->sk);
1257 static int netlink_release(struct socket *sock)
1259 struct sock *sk = sock->sk;
1260 struct netlink_sock *nlk;
1262 if (!sk)
1263 return 0;
1265 netlink_remove(sk);
1266 sock_orphan(sk);
1267 nlk = nlk_sk(sk);
1270 * OK. Socket is unlinked, any packets that arrive now
1271 * will be purged.
1274 /* must not acquire netlink_table_lock in any way again before unbind
1275 * and notifying genetlink is done as otherwise it might deadlock
1277 if (nlk->netlink_unbind) {
1278 int i;
1280 for (i = 0; i < nlk->ngroups; i++)
1281 if (test_bit(i, nlk->groups))
1282 nlk->netlink_unbind(sock_net(sk), i + 1);
1284 if (sk->sk_protocol == NETLINK_GENERIC &&
1285 atomic_dec_return(&genl_sk_destructing_cnt) == 0)
1286 wake_up(&genl_sk_destructing_waitq);
1288 sock->sk = NULL;
1289 wake_up_interruptible_all(&nlk->wait);
1291 skb_queue_purge(&sk->sk_write_queue);
1293 if (nlk->portid) {
1294 struct netlink_notify n = {
1295 .net = sock_net(sk),
1296 .protocol = sk->sk_protocol,
1297 .portid = nlk->portid,
1299 atomic_notifier_call_chain(&netlink_chain,
1300 NETLINK_URELEASE, &n);
1303 module_put(nlk->module);
1305 if (netlink_is_kernel(sk)) {
1306 netlink_table_grab();
1307 BUG_ON(nl_table[sk->sk_protocol].registered == 0);
1308 if (--nl_table[sk->sk_protocol].registered == 0) {
1309 struct listeners *old;
1311 old = nl_deref_protected(nl_table[sk->sk_protocol].listeners);
1312 RCU_INIT_POINTER(nl_table[sk->sk_protocol].listeners, NULL);
1313 kfree_rcu(old, rcu);
1314 nl_table[sk->sk_protocol].module = NULL;
1315 nl_table[sk->sk_protocol].bind = NULL;
1316 nl_table[sk->sk_protocol].unbind = NULL;
1317 nl_table[sk->sk_protocol].flags = 0;
1318 nl_table[sk->sk_protocol].registered = 0;
1320 netlink_table_ungrab();
1323 kfree(nlk->groups);
1324 nlk->groups = NULL;
1326 local_bh_disable();
1327 sock_prot_inuse_add(sock_net(sk), &netlink_proto, -1);
1328 local_bh_enable();
1329 call_rcu(&nlk->rcu, deferred_put_nlk_sk);
1330 return 0;
1333 static int netlink_autobind(struct socket *sock)
1335 struct sock *sk = sock->sk;
1336 struct net *net = sock_net(sk);
1337 struct netlink_table *table = &nl_table[sk->sk_protocol];
1338 s32 portid = task_tgid_vnr(current);
1339 int err;
1340 static s32 rover = -4097;
1342 retry:
1343 cond_resched();
1344 rcu_read_lock();
1345 if (__netlink_lookup(table, portid, net)) {
1346 /* Bind collision, search negative portid values. */
1347 portid = rover--;
1348 if (rover > -4097)
1349 rover = -4097;
1350 rcu_read_unlock();
1351 goto retry;
1353 rcu_read_unlock();
1355 err = netlink_insert(sk, portid);
1356 if (err == -EADDRINUSE)
1357 goto retry;
1359 /* If 2 threads race to autobind, that is fine. */
1360 if (err == -EBUSY)
1361 err = 0;
1363 return err;
1367 * __netlink_ns_capable - General netlink message capability test
1368 * @nsp: NETLINK_CB of the socket buffer holding a netlink command from userspace.
1369 * @user_ns: The user namespace of the capability to use
1370 * @cap: The capability to use
1372 * Test to see if the opener of the socket we received the message
1373 * from had when the netlink socket was created and the sender of the
1374 * message has has the capability @cap in the user namespace @user_ns.
1376 bool __netlink_ns_capable(const struct netlink_skb_parms *nsp,
1377 struct user_namespace *user_ns, int cap)
1379 return ((nsp->flags & NETLINK_SKB_DST) ||
1380 file_ns_capable(nsp->sk->sk_socket->file, user_ns, cap)) &&
1381 ns_capable(user_ns, cap);
1383 EXPORT_SYMBOL(__netlink_ns_capable);
1386 * netlink_ns_capable - General netlink message capability test
1387 * @skb: socket buffer holding a netlink command from userspace
1388 * @user_ns: The user namespace of the capability to use
1389 * @cap: The capability to use
1391 * Test to see if the opener of the socket we received the message
1392 * from had when the netlink socket was created and the sender of the
1393 * message has has the capability @cap in the user namespace @user_ns.
1395 bool netlink_ns_capable(const struct sk_buff *skb,
1396 struct user_namespace *user_ns, int cap)
1398 return __netlink_ns_capable(&NETLINK_CB(skb), user_ns, cap);
1400 EXPORT_SYMBOL(netlink_ns_capable);
1403 * netlink_capable - Netlink global message capability test
1404 * @skb: socket buffer holding a netlink command from userspace
1405 * @cap: The capability to use
1407 * Test to see if the opener of the socket we received the message
1408 * from had when the netlink socket was created and the sender of the
1409 * message has has the capability @cap in all user namespaces.
1411 bool netlink_capable(const struct sk_buff *skb, int cap)
1413 return netlink_ns_capable(skb, &init_user_ns, cap);
1415 EXPORT_SYMBOL(netlink_capable);
1418 * netlink_net_capable - Netlink network namespace message capability test
1419 * @skb: socket buffer holding a netlink command from userspace
1420 * @cap: The capability to use
1422 * Test to see if the opener of the socket we received the message
1423 * from had when the netlink socket was created and the sender of the
1424 * message has has the capability @cap over the network namespace of
1425 * the socket we received the message from.
1427 bool netlink_net_capable(const struct sk_buff *skb, int cap)
1429 return netlink_ns_capable(skb, sock_net(skb->sk)->user_ns, cap);
1431 EXPORT_SYMBOL(netlink_net_capable);
1433 static inline int netlink_allowed(const struct socket *sock, unsigned int flag)
1435 return (nl_table[sock->sk->sk_protocol].flags & flag) ||
1436 ns_capable(sock_net(sock->sk)->user_ns, CAP_NET_ADMIN);
1439 static void
1440 netlink_update_subscriptions(struct sock *sk, unsigned int subscriptions)
1442 struct netlink_sock *nlk = nlk_sk(sk);
1444 if (nlk->subscriptions && !subscriptions)
1445 __sk_del_bind_node(sk);
1446 else if (!nlk->subscriptions && subscriptions)
1447 sk_add_bind_node(sk, &nl_table[sk->sk_protocol].mc_list);
1448 nlk->subscriptions = subscriptions;
1451 static int netlink_realloc_groups(struct sock *sk)
1453 struct netlink_sock *nlk = nlk_sk(sk);
1454 unsigned int groups;
1455 unsigned long *new_groups;
1456 int err = 0;
1458 netlink_table_grab();
1460 groups = nl_table[sk->sk_protocol].groups;
1461 if (!nl_table[sk->sk_protocol].registered) {
1462 err = -ENOENT;
1463 goto out_unlock;
1466 if (nlk->ngroups >= groups)
1467 goto out_unlock;
1469 new_groups = krealloc(nlk->groups, NLGRPSZ(groups), GFP_ATOMIC);
1470 if (new_groups == NULL) {
1471 err = -ENOMEM;
1472 goto out_unlock;
1474 memset((char *)new_groups + NLGRPSZ(nlk->ngroups), 0,
1475 NLGRPSZ(groups) - NLGRPSZ(nlk->ngroups));
1477 nlk->groups = new_groups;
1478 nlk->ngroups = groups;
1479 out_unlock:
1480 netlink_table_ungrab();
1481 return err;
1484 static void netlink_undo_bind(int group, long unsigned int groups,
1485 struct sock *sk)
1487 struct netlink_sock *nlk = nlk_sk(sk);
1488 int undo;
1490 if (!nlk->netlink_unbind)
1491 return;
1493 for (undo = 0; undo < group; undo++)
1494 if (test_bit(undo, &groups))
1495 nlk->netlink_unbind(sock_net(sk), undo + 1);
1498 static int netlink_bind(struct socket *sock, struct sockaddr *addr,
1499 int addr_len)
1501 struct sock *sk = sock->sk;
1502 struct net *net = sock_net(sk);
1503 struct netlink_sock *nlk = nlk_sk(sk);
1504 struct sockaddr_nl *nladdr = (struct sockaddr_nl *)addr;
1505 int err;
1506 long unsigned int groups = nladdr->nl_groups;
1507 bool bound;
1509 if (addr_len < sizeof(struct sockaddr_nl))
1510 return -EINVAL;
1512 if (nladdr->nl_family != AF_NETLINK)
1513 return -EINVAL;
1515 /* Only superuser is allowed to listen multicasts */
1516 if (groups) {
1517 if (!netlink_allowed(sock, NL_CFG_F_NONROOT_RECV))
1518 return -EPERM;
1519 err = netlink_realloc_groups(sk);
1520 if (err)
1521 return err;
1524 bound = nlk->bound;
1525 if (bound) {
1526 /* Ensure nlk->portid is up-to-date. */
1527 smp_rmb();
1529 if (nladdr->nl_pid != nlk->portid)
1530 return -EINVAL;
1533 if (nlk->netlink_bind && groups) {
1534 int group;
1536 for (group = 0; group < nlk->ngroups; group++) {
1537 if (!test_bit(group, &groups))
1538 continue;
1539 err = nlk->netlink_bind(net, group + 1);
1540 if (!err)
1541 continue;
1542 netlink_undo_bind(group, groups, sk);
1543 return err;
1547 /* No need for barriers here as we return to user-space without
1548 * using any of the bound attributes.
1550 if (!bound) {
1551 err = nladdr->nl_pid ?
1552 netlink_insert(sk, nladdr->nl_pid) :
1553 netlink_autobind(sock);
1554 if (err) {
1555 netlink_undo_bind(nlk->ngroups, groups, sk);
1556 return err;
1560 if (!groups && (nlk->groups == NULL || !(u32)nlk->groups[0]))
1561 return 0;
1563 netlink_table_grab();
1564 netlink_update_subscriptions(sk, nlk->subscriptions +
1565 hweight32(groups) -
1566 hweight32(nlk->groups[0]));
1567 nlk->groups[0] = (nlk->groups[0] & ~0xffffffffUL) | groups;
1568 netlink_update_listeners(sk);
1569 netlink_table_ungrab();
1571 return 0;
1574 static int netlink_connect(struct socket *sock, struct sockaddr *addr,
1575 int alen, int flags)
1577 int err = 0;
1578 struct sock *sk = sock->sk;
1579 struct netlink_sock *nlk = nlk_sk(sk);
1580 struct sockaddr_nl *nladdr = (struct sockaddr_nl *)addr;
1582 if (alen < sizeof(addr->sa_family))
1583 return -EINVAL;
1585 if (addr->sa_family == AF_UNSPEC) {
1586 sk->sk_state = NETLINK_UNCONNECTED;
1587 nlk->dst_portid = 0;
1588 nlk->dst_group = 0;
1589 return 0;
1591 if (addr->sa_family != AF_NETLINK)
1592 return -EINVAL;
1594 if ((nladdr->nl_groups || nladdr->nl_pid) &&
1595 !netlink_allowed(sock, NL_CFG_F_NONROOT_SEND))
1596 return -EPERM;
1598 /* No need for barriers here as we return to user-space without
1599 * using any of the bound attributes.
1601 if (!nlk->bound)
1602 err = netlink_autobind(sock);
1604 if (err == 0) {
1605 sk->sk_state = NETLINK_CONNECTED;
1606 nlk->dst_portid = nladdr->nl_pid;
1607 nlk->dst_group = ffs(nladdr->nl_groups);
1610 return err;
1613 static int netlink_getname(struct socket *sock, struct sockaddr *addr,
1614 int *addr_len, int peer)
1616 struct sock *sk = sock->sk;
1617 struct netlink_sock *nlk = nlk_sk(sk);
1618 DECLARE_SOCKADDR(struct sockaddr_nl *, nladdr, addr);
1620 nladdr->nl_family = AF_NETLINK;
1621 nladdr->nl_pad = 0;
1622 *addr_len = sizeof(*nladdr);
1624 if (peer) {
1625 nladdr->nl_pid = nlk->dst_portid;
1626 nladdr->nl_groups = netlink_group_mask(nlk->dst_group);
1627 } else {
1628 nladdr->nl_pid = nlk->portid;
1629 nladdr->nl_groups = nlk->groups ? nlk->groups[0] : 0;
1631 return 0;
1634 static struct sock *netlink_getsockbyportid(struct sock *ssk, u32 portid)
1636 struct sock *sock;
1637 struct netlink_sock *nlk;
1639 sock = netlink_lookup(sock_net(ssk), ssk->sk_protocol, portid);
1640 if (!sock)
1641 return ERR_PTR(-ECONNREFUSED);
1643 /* Don't bother queuing skb if kernel socket has no input function */
1644 nlk = nlk_sk(sock);
1645 if (sock->sk_state == NETLINK_CONNECTED &&
1646 nlk->dst_portid != nlk_sk(ssk)->portid) {
1647 sock_put(sock);
1648 return ERR_PTR(-ECONNREFUSED);
1650 return sock;
1653 struct sock *netlink_getsockbyfilp(struct file *filp)
1655 struct inode *inode = file_inode(filp);
1656 struct sock *sock;
1658 if (!S_ISSOCK(inode->i_mode))
1659 return ERR_PTR(-ENOTSOCK);
1661 sock = SOCKET_I(inode)->sk;
1662 if (sock->sk_family != AF_NETLINK)
1663 return ERR_PTR(-EINVAL);
1665 sock_hold(sock);
1666 return sock;
1669 static struct sk_buff *netlink_alloc_large_skb(unsigned int size,
1670 int broadcast)
1672 struct sk_buff *skb;
1673 void *data;
1675 if (size <= NLMSG_GOODSIZE || broadcast)
1676 return alloc_skb(size, GFP_KERNEL);
1678 size = SKB_DATA_ALIGN(size) +
1679 SKB_DATA_ALIGN(sizeof(struct skb_shared_info));
1681 data = vmalloc(size);
1682 if (data == NULL)
1683 return NULL;
1685 skb = __build_skb(data, size);
1686 if (skb == NULL)
1687 vfree(data);
1688 else
1689 skb->destructor = netlink_skb_destructor;
1691 return skb;
1695 * Attach a skb to a netlink socket.
1696 * The caller must hold a reference to the destination socket. On error, the
1697 * reference is dropped. The skb is not send to the destination, just all
1698 * all error checks are performed and memory in the queue is reserved.
1699 * Return values:
1700 * < 0: error. skb freed, reference to sock dropped.
1701 * 0: continue
1702 * 1: repeat lookup - reference dropped while waiting for socket memory.
1704 int netlink_attachskb(struct sock *sk, struct sk_buff *skb,
1705 long *timeo, struct sock *ssk)
1707 struct netlink_sock *nlk;
1709 nlk = nlk_sk(sk);
1711 if ((atomic_read(&sk->sk_rmem_alloc) > sk->sk_rcvbuf ||
1712 test_bit(NETLINK_CONGESTED, &nlk->state)) &&
1713 !netlink_skb_is_mmaped(skb)) {
1714 DECLARE_WAITQUEUE(wait, current);
1715 if (!*timeo) {
1716 if (!ssk || netlink_is_kernel(ssk))
1717 netlink_overrun(sk);
1718 sock_put(sk);
1719 kfree_skb(skb);
1720 return -EAGAIN;
1723 __set_current_state(TASK_INTERRUPTIBLE);
1724 add_wait_queue(&nlk->wait, &wait);
1726 if ((atomic_read(&sk->sk_rmem_alloc) > sk->sk_rcvbuf ||
1727 test_bit(NETLINK_CONGESTED, &nlk->state)) &&
1728 !sock_flag(sk, SOCK_DEAD))
1729 *timeo = schedule_timeout(*timeo);
1731 __set_current_state(TASK_RUNNING);
1732 remove_wait_queue(&nlk->wait, &wait);
1733 sock_put(sk);
1735 if (signal_pending(current)) {
1736 kfree_skb(skb);
1737 return sock_intr_errno(*timeo);
1739 return 1;
1741 netlink_skb_set_owner_r(skb, sk);
1742 return 0;
1745 static int __netlink_sendskb(struct sock *sk, struct sk_buff *skb)
1747 int len = skb->len;
1749 netlink_deliver_tap(skb);
1751 #ifdef CONFIG_NETLINK_MMAP
1752 if (netlink_skb_is_mmaped(skb))
1753 netlink_queue_mmaped_skb(sk, skb);
1754 else if (netlink_rx_is_mmaped(sk))
1755 netlink_ring_set_copied(sk, skb);
1756 else
1757 #endif /* CONFIG_NETLINK_MMAP */
1758 skb_queue_tail(&sk->sk_receive_queue, skb);
1759 sk->sk_data_ready(sk);
1760 return len;
1763 int netlink_sendskb(struct sock *sk, struct sk_buff *skb)
1765 int len = __netlink_sendskb(sk, skb);
1767 sock_put(sk);
1768 return len;
1771 void netlink_detachskb(struct sock *sk, struct sk_buff *skb)
1773 kfree_skb(skb);
1774 sock_put(sk);
1777 static struct sk_buff *netlink_trim(struct sk_buff *skb, gfp_t allocation)
1779 int delta;
1781 WARN_ON(skb->sk != NULL);
1782 if (netlink_skb_is_mmaped(skb))
1783 return skb;
1785 delta = skb->end - skb->tail;
1786 if (is_vmalloc_addr(skb->head) || delta * 2 < skb->truesize)
1787 return skb;
1789 if (skb_shared(skb)) {
1790 struct sk_buff *nskb = skb_clone(skb, allocation);
1791 if (!nskb)
1792 return skb;
1793 consume_skb(skb);
1794 skb = nskb;
1797 if (!pskb_expand_head(skb, 0, -delta, allocation))
1798 skb->truesize -= delta;
1800 return skb;
1803 static int netlink_unicast_kernel(struct sock *sk, struct sk_buff *skb,
1804 struct sock *ssk)
1806 int ret;
1807 struct netlink_sock *nlk = nlk_sk(sk);
1809 ret = -ECONNREFUSED;
1810 if (nlk->netlink_rcv != NULL) {
1811 ret = skb->len;
1812 netlink_skb_set_owner_r(skb, sk);
1813 NETLINK_CB(skb).sk = ssk;
1814 netlink_deliver_tap_kernel(sk, ssk, skb);
1815 nlk->netlink_rcv(skb);
1816 consume_skb(skb);
1817 } else {
1818 kfree_skb(skb);
1820 sock_put(sk);
1821 return ret;
1824 int netlink_unicast(struct sock *ssk, struct sk_buff *skb,
1825 u32 portid, int nonblock)
1827 struct sock *sk;
1828 int err;
1829 long timeo;
1831 skb = netlink_trim(skb, gfp_any());
1833 timeo = sock_sndtimeo(ssk, nonblock);
1834 retry:
1835 sk = netlink_getsockbyportid(ssk, portid);
1836 if (IS_ERR(sk)) {
1837 kfree_skb(skb);
1838 return PTR_ERR(sk);
1840 if (netlink_is_kernel(sk))
1841 return netlink_unicast_kernel(sk, skb, ssk);
1843 if (sk_filter(sk, skb)) {
1844 err = skb->len;
1845 kfree_skb(skb);
1846 sock_put(sk);
1847 return err;
1850 err = netlink_attachskb(sk, skb, &timeo, ssk);
1851 if (err == 1)
1852 goto retry;
1853 if (err)
1854 return err;
1856 return netlink_sendskb(sk, skb);
1858 EXPORT_SYMBOL(netlink_unicast);
1860 struct sk_buff *netlink_alloc_skb(struct sock *ssk, unsigned int size,
1861 u32 dst_portid, gfp_t gfp_mask)
1863 #ifdef CONFIG_NETLINK_MMAP
1864 struct sock *sk = NULL;
1865 struct sk_buff *skb;
1866 struct netlink_ring *ring;
1867 struct nl_mmap_hdr *hdr;
1868 unsigned int maxlen;
1870 sk = netlink_getsockbyportid(ssk, dst_portid);
1871 if (IS_ERR(sk))
1872 goto out;
1874 ring = &nlk_sk(sk)->rx_ring;
1875 /* fast-path without atomic ops for common case: non-mmaped receiver */
1876 if (ring->pg_vec == NULL)
1877 goto out_put;
1879 if (ring->frame_size - NL_MMAP_HDRLEN < size)
1880 goto out_put;
1882 skb = alloc_skb_head(gfp_mask);
1883 if (skb == NULL)
1884 goto err1;
1886 spin_lock_bh(&sk->sk_receive_queue.lock);
1887 /* check again under lock */
1888 if (ring->pg_vec == NULL)
1889 goto out_free;
1891 /* check again under lock */
1892 maxlen = ring->frame_size - NL_MMAP_HDRLEN;
1893 if (maxlen < size)
1894 goto out_free;
1896 netlink_forward_ring(ring);
1897 hdr = netlink_current_frame(ring, NL_MMAP_STATUS_UNUSED);
1898 if (hdr == NULL)
1899 goto err2;
1900 netlink_ring_setup_skb(skb, sk, ring, hdr);
1901 netlink_set_status(hdr, NL_MMAP_STATUS_RESERVED);
1902 atomic_inc(&ring->pending);
1903 netlink_increment_head(ring);
1905 spin_unlock_bh(&sk->sk_receive_queue.lock);
1906 return skb;
1908 err2:
1909 kfree_skb(skb);
1910 spin_unlock_bh(&sk->sk_receive_queue.lock);
1911 netlink_overrun(sk);
1912 err1:
1913 sock_put(sk);
1914 return NULL;
1916 out_free:
1917 kfree_skb(skb);
1918 spin_unlock_bh(&sk->sk_receive_queue.lock);
1919 out_put:
1920 sock_put(sk);
1921 out:
1922 #endif
1923 return alloc_skb(size, gfp_mask);
1925 EXPORT_SYMBOL_GPL(netlink_alloc_skb);
1927 int netlink_has_listeners(struct sock *sk, unsigned int group)
1929 int res = 0;
1930 struct listeners *listeners;
1932 BUG_ON(!netlink_is_kernel(sk));
1934 rcu_read_lock();
1935 listeners = rcu_dereference(nl_table[sk->sk_protocol].listeners);
1937 if (listeners && group - 1 < nl_table[sk->sk_protocol].groups)
1938 res = test_bit(group - 1, listeners->masks);
1940 rcu_read_unlock();
1942 return res;
1944 EXPORT_SYMBOL_GPL(netlink_has_listeners);
1946 static int netlink_broadcast_deliver(struct sock *sk, struct sk_buff *skb)
1948 struct netlink_sock *nlk = nlk_sk(sk);
1950 if (atomic_read(&sk->sk_rmem_alloc) <= sk->sk_rcvbuf &&
1951 !test_bit(NETLINK_CONGESTED, &nlk->state)) {
1952 netlink_skb_set_owner_r(skb, sk);
1953 __netlink_sendskb(sk, skb);
1954 return atomic_read(&sk->sk_rmem_alloc) > (sk->sk_rcvbuf >> 1);
1956 return -1;
1959 struct netlink_broadcast_data {
1960 struct sock *exclude_sk;
1961 struct net *net;
1962 u32 portid;
1963 u32 group;
1964 int failure;
1965 int delivery_failure;
1966 int congested;
1967 int delivered;
1968 gfp_t allocation;
1969 struct sk_buff *skb, *skb2;
1970 int (*tx_filter)(struct sock *dsk, struct sk_buff *skb, void *data);
1971 void *tx_data;
1974 static void do_one_broadcast(struct sock *sk,
1975 struct netlink_broadcast_data *p)
1977 struct netlink_sock *nlk = nlk_sk(sk);
1978 int val;
1980 if (p->exclude_sk == sk)
1981 return;
1983 if (nlk->portid == p->portid || p->group - 1 >= nlk->ngroups ||
1984 !test_bit(p->group - 1, nlk->groups))
1985 return;
1987 if (!net_eq(sock_net(sk), p->net))
1988 return;
1990 if (p->failure) {
1991 netlink_overrun(sk);
1992 return;
1995 sock_hold(sk);
1996 if (p->skb2 == NULL) {
1997 if (skb_shared(p->skb)) {
1998 p->skb2 = skb_clone(p->skb, p->allocation);
1999 } else {
2000 p->skb2 = skb_get(p->skb);
2002 * skb ownership may have been set when
2003 * delivered to a previous socket.
2005 skb_orphan(p->skb2);
2008 if (p->skb2 == NULL) {
2009 netlink_overrun(sk);
2010 /* Clone failed. Notify ALL listeners. */
2011 p->failure = 1;
2012 if (nlk->flags & NETLINK_BROADCAST_SEND_ERROR)
2013 p->delivery_failure = 1;
2014 } else if (p->tx_filter && p->tx_filter(sk, p->skb2, p->tx_data)) {
2015 kfree_skb(p->skb2);
2016 p->skb2 = NULL;
2017 } else if (sk_filter(sk, p->skb2)) {
2018 kfree_skb(p->skb2);
2019 p->skb2 = NULL;
2020 } else if ((val = netlink_broadcast_deliver(sk, p->skb2)) < 0) {
2021 netlink_overrun(sk);
2022 if (nlk->flags & NETLINK_BROADCAST_SEND_ERROR)
2023 p->delivery_failure = 1;
2024 } else {
2025 p->congested |= val;
2026 p->delivered = 1;
2027 p->skb2 = NULL;
2029 sock_put(sk);
2032 int netlink_broadcast_filtered(struct sock *ssk, struct sk_buff *skb, u32 portid,
2033 u32 group, gfp_t allocation,
2034 int (*filter)(struct sock *dsk, struct sk_buff *skb, void *data),
2035 void *filter_data)
2037 struct net *net = sock_net(ssk);
2038 struct netlink_broadcast_data info;
2039 struct sock *sk;
2041 skb = netlink_trim(skb, allocation);
2043 info.exclude_sk = ssk;
2044 info.net = net;
2045 info.portid = portid;
2046 info.group = group;
2047 info.failure = 0;
2048 info.delivery_failure = 0;
2049 info.congested = 0;
2050 info.delivered = 0;
2051 info.allocation = allocation;
2052 info.skb = skb;
2053 info.skb2 = NULL;
2054 info.tx_filter = filter;
2055 info.tx_data = filter_data;
2057 /* While we sleep in clone, do not allow to change socket list */
2059 netlink_lock_table();
2061 sk_for_each_bound(sk, &nl_table[ssk->sk_protocol].mc_list)
2062 do_one_broadcast(sk, &info);
2064 consume_skb(skb);
2066 netlink_unlock_table();
2068 if (info.delivery_failure) {
2069 kfree_skb(info.skb2);
2070 return -ENOBUFS;
2072 consume_skb(info.skb2);
2074 if (info.delivered) {
2075 if (info.congested && (allocation & __GFP_WAIT))
2076 yield();
2077 return 0;
2079 return -ESRCH;
2081 EXPORT_SYMBOL(netlink_broadcast_filtered);
2083 int netlink_broadcast(struct sock *ssk, struct sk_buff *skb, u32 portid,
2084 u32 group, gfp_t allocation)
2086 return netlink_broadcast_filtered(ssk, skb, portid, group, allocation,
2087 NULL, NULL);
2089 EXPORT_SYMBOL(netlink_broadcast);
2091 struct netlink_set_err_data {
2092 struct sock *exclude_sk;
2093 u32 portid;
2094 u32 group;
2095 int code;
2098 static int do_one_set_err(struct sock *sk, struct netlink_set_err_data *p)
2100 struct netlink_sock *nlk = nlk_sk(sk);
2101 int ret = 0;
2103 if (sk == p->exclude_sk)
2104 goto out;
2106 if (!net_eq(sock_net(sk), sock_net(p->exclude_sk)))
2107 goto out;
2109 if (nlk->portid == p->portid || p->group - 1 >= nlk->ngroups ||
2110 !test_bit(p->group - 1, nlk->groups))
2111 goto out;
2113 if (p->code == ENOBUFS && nlk->flags & NETLINK_RECV_NO_ENOBUFS) {
2114 ret = 1;
2115 goto out;
2118 sk->sk_err = p->code;
2119 sk->sk_error_report(sk);
2120 out:
2121 return ret;
2125 * netlink_set_err - report error to broadcast listeners
2126 * @ssk: the kernel netlink socket, as returned by netlink_kernel_create()
2127 * @portid: the PORTID of a process that we want to skip (if any)
2128 * @group: the broadcast group that will notice the error
2129 * @code: error code, must be negative (as usual in kernelspace)
2131 * This function returns the number of broadcast listeners that have set the
2132 * NETLINK_RECV_NO_ENOBUFS socket option.
2134 int netlink_set_err(struct sock *ssk, u32 portid, u32 group, int code)
2136 struct netlink_set_err_data info;
2137 struct sock *sk;
2138 int ret = 0;
2140 info.exclude_sk = ssk;
2141 info.portid = portid;
2142 info.group = group;
2143 /* sk->sk_err wants a positive error value */
2144 info.code = -code;
2146 read_lock(&nl_table_lock);
2148 sk_for_each_bound(sk, &nl_table[ssk->sk_protocol].mc_list)
2149 ret += do_one_set_err(sk, &info);
2151 read_unlock(&nl_table_lock);
2152 return ret;
2154 EXPORT_SYMBOL(netlink_set_err);
2156 /* must be called with netlink table grabbed */
2157 static void netlink_update_socket_mc(struct netlink_sock *nlk,
2158 unsigned int group,
2159 int is_new)
2161 int old, new = !!is_new, subscriptions;
2163 old = test_bit(group - 1, nlk->groups);
2164 subscriptions = nlk->subscriptions - old + new;
2165 if (new)
2166 __set_bit(group - 1, nlk->groups);
2167 else
2168 __clear_bit(group - 1, nlk->groups);
2169 netlink_update_subscriptions(&nlk->sk, subscriptions);
2170 netlink_update_listeners(&nlk->sk);
2173 static int netlink_setsockopt(struct socket *sock, int level, int optname,
2174 char __user *optval, unsigned int optlen)
2176 struct sock *sk = sock->sk;
2177 struct netlink_sock *nlk = nlk_sk(sk);
2178 unsigned int val = 0;
2179 int err;
2181 if (level != SOL_NETLINK)
2182 return -ENOPROTOOPT;
2184 if (optname != NETLINK_RX_RING && optname != NETLINK_TX_RING &&
2185 optlen >= sizeof(int) &&
2186 get_user(val, (unsigned int __user *)optval))
2187 return -EFAULT;
2189 switch (optname) {
2190 case NETLINK_PKTINFO:
2191 if (val)
2192 nlk->flags |= NETLINK_RECV_PKTINFO;
2193 else
2194 nlk->flags &= ~NETLINK_RECV_PKTINFO;
2195 err = 0;
2196 break;
2197 case NETLINK_ADD_MEMBERSHIP:
2198 case NETLINK_DROP_MEMBERSHIP: {
2199 if (!netlink_allowed(sock, NL_CFG_F_NONROOT_RECV))
2200 return -EPERM;
2201 err = netlink_realloc_groups(sk);
2202 if (err)
2203 return err;
2204 if (!val || val - 1 >= nlk->ngroups)
2205 return -EINVAL;
2206 if (optname == NETLINK_ADD_MEMBERSHIP && nlk->netlink_bind) {
2207 err = nlk->netlink_bind(sock_net(sk), val);
2208 if (err)
2209 return err;
2211 netlink_table_grab();
2212 netlink_update_socket_mc(nlk, val,
2213 optname == NETLINK_ADD_MEMBERSHIP);
2214 netlink_table_ungrab();
2215 if (optname == NETLINK_DROP_MEMBERSHIP && nlk->netlink_unbind)
2216 nlk->netlink_unbind(sock_net(sk), val);
2218 err = 0;
2219 break;
2221 case NETLINK_BROADCAST_ERROR:
2222 if (val)
2223 nlk->flags |= NETLINK_BROADCAST_SEND_ERROR;
2224 else
2225 nlk->flags &= ~NETLINK_BROADCAST_SEND_ERROR;
2226 err = 0;
2227 break;
2228 case NETLINK_NO_ENOBUFS:
2229 if (val) {
2230 nlk->flags |= NETLINK_RECV_NO_ENOBUFS;
2231 clear_bit(NETLINK_CONGESTED, &nlk->state);
2232 wake_up_interruptible(&nlk->wait);
2233 } else {
2234 nlk->flags &= ~NETLINK_RECV_NO_ENOBUFS;
2236 err = 0;
2237 break;
2238 #ifdef CONFIG_NETLINK_MMAP
2239 case NETLINK_RX_RING:
2240 case NETLINK_TX_RING: {
2241 struct nl_mmap_req req;
2243 /* Rings might consume more memory than queue limits, require
2244 * CAP_NET_ADMIN.
2246 if (!capable(CAP_NET_ADMIN))
2247 return -EPERM;
2248 if (optlen < sizeof(req))
2249 return -EINVAL;
2250 if (copy_from_user(&req, optval, sizeof(req)))
2251 return -EFAULT;
2252 err = netlink_set_ring(sk, &req,
2253 optname == NETLINK_TX_RING);
2254 break;
2256 #endif /* CONFIG_NETLINK_MMAP */
2257 default:
2258 err = -ENOPROTOOPT;
2260 return err;
2263 static int netlink_getsockopt(struct socket *sock, int level, int optname,
2264 char __user *optval, int __user *optlen)
2266 struct sock *sk = sock->sk;
2267 struct netlink_sock *nlk = nlk_sk(sk);
2268 int len, val, err;
2270 if (level != SOL_NETLINK)
2271 return -ENOPROTOOPT;
2273 if (get_user(len, optlen))
2274 return -EFAULT;
2275 if (len < 0)
2276 return -EINVAL;
2278 switch (optname) {
2279 case NETLINK_PKTINFO:
2280 if (len < sizeof(int))
2281 return -EINVAL;
2282 len = sizeof(int);
2283 val = nlk->flags & NETLINK_RECV_PKTINFO ? 1 : 0;
2284 if (put_user(len, optlen) ||
2285 put_user(val, optval))
2286 return -EFAULT;
2287 err = 0;
2288 break;
2289 case NETLINK_BROADCAST_ERROR:
2290 if (len < sizeof(int))
2291 return -EINVAL;
2292 len = sizeof(int);
2293 val = nlk->flags & NETLINK_BROADCAST_SEND_ERROR ? 1 : 0;
2294 if (put_user(len, optlen) ||
2295 put_user(val, optval))
2296 return -EFAULT;
2297 err = 0;
2298 break;
2299 case NETLINK_NO_ENOBUFS:
2300 if (len < sizeof(int))
2301 return -EINVAL;
2302 len = sizeof(int);
2303 val = nlk->flags & NETLINK_RECV_NO_ENOBUFS ? 1 : 0;
2304 if (put_user(len, optlen) ||
2305 put_user(val, optval))
2306 return -EFAULT;
2307 err = 0;
2308 break;
2309 default:
2310 err = -ENOPROTOOPT;
2312 return err;
2315 static void netlink_cmsg_recv_pktinfo(struct msghdr *msg, struct sk_buff *skb)
2317 struct nl_pktinfo info;
2319 info.group = NETLINK_CB(skb).dst_group;
2320 put_cmsg(msg, SOL_NETLINK, NETLINK_PKTINFO, sizeof(info), &info);
2323 static int netlink_sendmsg(struct socket *sock, struct msghdr *msg, size_t len)
2325 struct sock *sk = sock->sk;
2326 struct netlink_sock *nlk = nlk_sk(sk);
2327 DECLARE_SOCKADDR(struct sockaddr_nl *, addr, msg->msg_name);
2328 u32 dst_portid;
2329 u32 dst_group;
2330 struct sk_buff *skb;
2331 int err;
2332 struct scm_cookie scm;
2333 u32 netlink_skb_flags = 0;
2335 if (msg->msg_flags&MSG_OOB)
2336 return -EOPNOTSUPP;
2338 err = scm_send(sock, msg, &scm, true);
2339 if (err < 0)
2340 return err;
2342 if (msg->msg_namelen) {
2343 err = -EINVAL;
2344 if (addr->nl_family != AF_NETLINK)
2345 goto out;
2346 dst_portid = addr->nl_pid;
2347 dst_group = ffs(addr->nl_groups);
2348 err = -EPERM;
2349 if ((dst_group || dst_portid) &&
2350 !netlink_allowed(sock, NL_CFG_F_NONROOT_SEND))
2351 goto out;
2352 netlink_skb_flags |= NETLINK_SKB_DST;
2353 } else {
2354 dst_portid = nlk->dst_portid;
2355 dst_group = nlk->dst_group;
2358 if (!nlk->bound) {
2359 err = netlink_autobind(sock);
2360 if (err)
2361 goto out;
2362 } else {
2363 /* Ensure nlk is hashed and visible. */
2364 smp_rmb();
2367 /* It's a really convoluted way for userland to ask for mmaped
2368 * sendmsg(), but that's what we've got...
2370 if (netlink_tx_is_mmaped(sk) &&
2371 msg->msg_iter.type == ITER_IOVEC &&
2372 msg->msg_iter.nr_segs == 1 &&
2373 msg->msg_iter.iov->iov_base == NULL) {
2374 err = netlink_mmap_sendmsg(sk, msg, dst_portid, dst_group,
2375 &scm);
2376 goto out;
2379 err = -EMSGSIZE;
2380 if (len > sk->sk_sndbuf - 32)
2381 goto out;
2382 err = -ENOBUFS;
2383 skb = netlink_alloc_large_skb(len, dst_group);
2384 if (skb == NULL)
2385 goto out;
2387 NETLINK_CB(skb).portid = nlk->portid;
2388 NETLINK_CB(skb).dst_group = dst_group;
2389 NETLINK_CB(skb).creds = scm.creds;
2390 NETLINK_CB(skb).flags = netlink_skb_flags;
2392 err = -EFAULT;
2393 if (memcpy_from_msg(skb_put(skb, len), msg, len)) {
2394 kfree_skb(skb);
2395 goto out;
2398 err = security_netlink_send(sk, skb);
2399 if (err) {
2400 kfree_skb(skb);
2401 goto out;
2404 if (dst_group) {
2405 atomic_inc(&skb->users);
2406 netlink_broadcast(sk, skb, dst_portid, dst_group, GFP_KERNEL);
2408 err = netlink_unicast(sk, skb, dst_portid, msg->msg_flags&MSG_DONTWAIT);
2410 out:
2411 scm_destroy(&scm);
2412 return err;
2415 static int netlink_recvmsg(struct socket *sock, struct msghdr *msg, size_t len,
2416 int flags)
2418 struct scm_cookie scm;
2419 struct sock *sk = sock->sk;
2420 struct netlink_sock *nlk = nlk_sk(sk);
2421 int noblock = flags&MSG_DONTWAIT;
2422 size_t copied;
2423 struct sk_buff *skb, *data_skb;
2424 int err, ret;
2426 if (flags&MSG_OOB)
2427 return -EOPNOTSUPP;
2429 copied = 0;
2431 skb = skb_recv_datagram(sk, flags, noblock, &err);
2432 if (skb == NULL)
2433 goto out;
2435 data_skb = skb;
2437 #ifdef CONFIG_COMPAT_NETLINK_MESSAGES
2438 if (unlikely(skb_shinfo(skb)->frag_list)) {
2440 * If this skb has a frag_list, then here that means that we
2441 * will have to use the frag_list skb's data for compat tasks
2442 * and the regular skb's data for normal (non-compat) tasks.
2444 * If we need to send the compat skb, assign it to the
2445 * 'data_skb' variable so that it will be used below for data
2446 * copying. We keep 'skb' for everything else, including
2447 * freeing both later.
2449 if (flags & MSG_CMSG_COMPAT)
2450 data_skb = skb_shinfo(skb)->frag_list;
2452 #endif
2454 /* Record the max length of recvmsg() calls for future allocations */
2455 nlk->max_recvmsg_len = max(nlk->max_recvmsg_len, len);
2456 nlk->max_recvmsg_len = min_t(size_t, nlk->max_recvmsg_len,
2457 16384);
2459 copied = data_skb->len;
2460 if (len < copied) {
2461 msg->msg_flags |= MSG_TRUNC;
2462 copied = len;
2465 skb_reset_transport_header(data_skb);
2466 err = skb_copy_datagram_msg(data_skb, 0, msg, copied);
2468 if (msg->msg_name) {
2469 DECLARE_SOCKADDR(struct sockaddr_nl *, addr, msg->msg_name);
2470 addr->nl_family = AF_NETLINK;
2471 addr->nl_pad = 0;
2472 addr->nl_pid = NETLINK_CB(skb).portid;
2473 addr->nl_groups = netlink_group_mask(NETLINK_CB(skb).dst_group);
2474 msg->msg_namelen = sizeof(*addr);
2477 if (nlk->flags & NETLINK_RECV_PKTINFO)
2478 netlink_cmsg_recv_pktinfo(msg, skb);
2480 memset(&scm, 0, sizeof(scm));
2481 scm.creds = *NETLINK_CREDS(skb);
2482 if (flags & MSG_TRUNC)
2483 copied = data_skb->len;
2485 skb_free_datagram(sk, skb);
2487 if (nlk->cb_running &&
2488 atomic_read(&sk->sk_rmem_alloc) <= sk->sk_rcvbuf / 2) {
2489 ret = netlink_dump(sk);
2490 if (ret) {
2491 sk->sk_err = -ret;
2492 sk->sk_error_report(sk);
2496 scm_recv(sock, msg, &scm, flags);
2497 out:
2498 netlink_rcv_wake(sk);
2499 return err ? : copied;
2502 static void netlink_data_ready(struct sock *sk)
2504 BUG();
2508 * We export these functions to other modules. They provide a
2509 * complete set of kernel non-blocking support for message
2510 * queueing.
2513 struct sock *
2514 __netlink_kernel_create(struct net *net, int unit, struct module *module,
2515 struct netlink_kernel_cfg *cfg)
2517 struct socket *sock;
2518 struct sock *sk;
2519 struct netlink_sock *nlk;
2520 struct listeners *listeners = NULL;
2521 struct mutex *cb_mutex = cfg ? cfg->cb_mutex : NULL;
2522 unsigned int groups;
2524 BUG_ON(!nl_table);
2526 if (unit < 0 || unit >= MAX_LINKS)
2527 return NULL;
2529 if (sock_create_lite(PF_NETLINK, SOCK_DGRAM, unit, &sock))
2530 return NULL;
2533 * We have to just have a reference on the net from sk, but don't
2534 * get_net it. Besides, we cannot get and then put the net here.
2535 * So we create one inside init_net and the move it to net.
2538 if (__netlink_create(&init_net, sock, cb_mutex, unit) < 0)
2539 goto out_sock_release_nosk;
2541 sk = sock->sk;
2542 sk_change_net(sk, net);
2544 if (!cfg || cfg->groups < 32)
2545 groups = 32;
2546 else
2547 groups = cfg->groups;
2549 listeners = kzalloc(sizeof(*listeners) + NLGRPSZ(groups), GFP_KERNEL);
2550 if (!listeners)
2551 goto out_sock_release;
2553 sk->sk_data_ready = netlink_data_ready;
2554 if (cfg && cfg->input)
2555 nlk_sk(sk)->netlink_rcv = cfg->input;
2557 if (netlink_insert(sk, 0))
2558 goto out_sock_release;
2560 nlk = nlk_sk(sk);
2561 nlk->flags |= NETLINK_KERNEL_SOCKET;
2563 netlink_table_grab();
2564 if (!nl_table[unit].registered) {
2565 nl_table[unit].groups = groups;
2566 rcu_assign_pointer(nl_table[unit].listeners, listeners);
2567 nl_table[unit].cb_mutex = cb_mutex;
2568 nl_table[unit].module = module;
2569 if (cfg) {
2570 nl_table[unit].bind = cfg->bind;
2571 nl_table[unit].unbind = cfg->unbind;
2572 nl_table[unit].flags = cfg->flags;
2573 if (cfg->compare)
2574 nl_table[unit].compare = cfg->compare;
2576 nl_table[unit].registered = 1;
2577 } else {
2578 kfree(listeners);
2579 nl_table[unit].registered++;
2581 netlink_table_ungrab();
2582 return sk;
2584 out_sock_release:
2585 kfree(listeners);
2586 netlink_kernel_release(sk);
2587 return NULL;
2589 out_sock_release_nosk:
2590 sock_release(sock);
2591 return NULL;
2593 EXPORT_SYMBOL(__netlink_kernel_create);
2595 void
2596 netlink_kernel_release(struct sock *sk)
2598 sk_release_kernel(sk);
2600 EXPORT_SYMBOL(netlink_kernel_release);
2602 int __netlink_change_ngroups(struct sock *sk, unsigned int groups)
2604 struct listeners *new, *old;
2605 struct netlink_table *tbl = &nl_table[sk->sk_protocol];
2607 if (groups < 32)
2608 groups = 32;
2610 if (NLGRPSZ(tbl->groups) < NLGRPSZ(groups)) {
2611 new = kzalloc(sizeof(*new) + NLGRPSZ(groups), GFP_ATOMIC);
2612 if (!new)
2613 return -ENOMEM;
2614 old = nl_deref_protected(tbl->listeners);
2615 memcpy(new->masks, old->masks, NLGRPSZ(tbl->groups));
2616 rcu_assign_pointer(tbl->listeners, new);
2618 kfree_rcu(old, rcu);
2620 tbl->groups = groups;
2622 return 0;
2626 * netlink_change_ngroups - change number of multicast groups
2628 * This changes the number of multicast groups that are available
2629 * on a certain netlink family. Note that it is not possible to
2630 * change the number of groups to below 32. Also note that it does
2631 * not implicitly call netlink_clear_multicast_users() when the
2632 * number of groups is reduced.
2634 * @sk: The kernel netlink socket, as returned by netlink_kernel_create().
2635 * @groups: The new number of groups.
2637 int netlink_change_ngroups(struct sock *sk, unsigned int groups)
2639 int err;
2641 netlink_table_grab();
2642 err = __netlink_change_ngroups(sk, groups);
2643 netlink_table_ungrab();
2645 return err;
2648 void __netlink_clear_multicast_users(struct sock *ksk, unsigned int group)
2650 struct sock *sk;
2651 struct netlink_table *tbl = &nl_table[ksk->sk_protocol];
2653 sk_for_each_bound(sk, &tbl->mc_list)
2654 netlink_update_socket_mc(nlk_sk(sk), group, 0);
2657 struct nlmsghdr *
2658 __nlmsg_put(struct sk_buff *skb, u32 portid, u32 seq, int type, int len, int flags)
2660 struct nlmsghdr *nlh;
2661 int size = nlmsg_msg_size(len);
2663 nlh = (struct nlmsghdr *)skb_put(skb, NLMSG_ALIGN(size));
2664 nlh->nlmsg_type = type;
2665 nlh->nlmsg_len = size;
2666 nlh->nlmsg_flags = flags;
2667 nlh->nlmsg_pid = portid;
2668 nlh->nlmsg_seq = seq;
2669 if (!__builtin_constant_p(size) || NLMSG_ALIGN(size) - size != 0)
2670 memset(nlmsg_data(nlh) + len, 0, NLMSG_ALIGN(size) - size);
2671 return nlh;
2673 EXPORT_SYMBOL(__nlmsg_put);
2676 * It looks a bit ugly.
2677 * It would be better to create kernel thread.
2680 static int netlink_dump(struct sock *sk)
2682 struct netlink_sock *nlk = nlk_sk(sk);
2683 struct netlink_callback *cb;
2684 struct sk_buff *skb = NULL;
2685 struct nlmsghdr *nlh;
2686 int len, err = -ENOBUFS;
2687 int alloc_min_size;
2688 int alloc_size;
2690 mutex_lock(nlk->cb_mutex);
2691 if (!nlk->cb_running) {
2692 err = -EINVAL;
2693 goto errout_skb;
2696 if (!netlink_rx_is_mmaped(sk) &&
2697 atomic_read(&sk->sk_rmem_alloc) >= sk->sk_rcvbuf)
2698 goto errout_skb;
2700 /* NLMSG_GOODSIZE is small to avoid high order allocations being
2701 * required, but it makes sense to _attempt_ a 16K bytes allocation
2702 * to reduce number of system calls on dump operations, if user
2703 * ever provided a big enough buffer.
2705 cb = &nlk->cb;
2706 alloc_min_size = max_t(int, cb->min_dump_alloc, NLMSG_GOODSIZE);
2708 if (alloc_min_size < nlk->max_recvmsg_len) {
2709 alloc_size = nlk->max_recvmsg_len;
2710 skb = netlink_alloc_skb(sk, alloc_size, nlk->portid,
2711 GFP_KERNEL |
2712 __GFP_NOWARN |
2713 __GFP_NORETRY);
2715 if (!skb) {
2716 alloc_size = alloc_min_size;
2717 skb = netlink_alloc_skb(sk, alloc_size, nlk->portid,
2718 GFP_KERNEL);
2720 if (!skb)
2721 goto errout_skb;
2723 /* Trim skb to allocated size. User is expected to provide buffer as
2724 * large as max(min_dump_alloc, 16KiB (mac_recvmsg_len capped at
2725 * netlink_recvmsg())). dump will pack as many smaller messages as
2726 * could fit within the allocated skb. skb is typically allocated
2727 * with larger space than required (could be as much as near 2x the
2728 * requested size with align to next power of 2 approach). Allowing
2729 * dump to use the excess space makes it difficult for a user to have a
2730 * reasonable static buffer based on the expected largest dump of a
2731 * single netdev. The outcome is MSG_TRUNC error.
2733 skb_reserve(skb, skb_tailroom(skb) - alloc_size);
2734 netlink_skb_set_owner_r(skb, sk);
2736 len = cb->dump(skb, cb);
2738 if (len > 0) {
2739 mutex_unlock(nlk->cb_mutex);
2741 if (sk_filter(sk, skb))
2742 kfree_skb(skb);
2743 else
2744 __netlink_sendskb(sk, skb);
2745 return 0;
2748 nlh = nlmsg_put_answer(skb, cb, NLMSG_DONE, sizeof(len), NLM_F_MULTI);
2749 if (!nlh)
2750 goto errout_skb;
2752 nl_dump_check_consistent(cb, nlh);
2754 memcpy(nlmsg_data(nlh), &len, sizeof(len));
2756 if (sk_filter(sk, skb))
2757 kfree_skb(skb);
2758 else
2759 __netlink_sendskb(sk, skb);
2761 if (cb->done)
2762 cb->done(cb);
2764 nlk->cb_running = false;
2765 mutex_unlock(nlk->cb_mutex);
2766 module_put(cb->module);
2767 consume_skb(cb->skb);
2768 return 0;
2770 errout_skb:
2771 mutex_unlock(nlk->cb_mutex);
2772 kfree_skb(skb);
2773 return err;
2776 int __netlink_dump_start(struct sock *ssk, struct sk_buff *skb,
2777 const struct nlmsghdr *nlh,
2778 struct netlink_dump_control *control)
2780 struct netlink_callback *cb;
2781 struct sock *sk;
2782 struct netlink_sock *nlk;
2783 int ret;
2785 /* Memory mapped dump requests need to be copied to avoid looping
2786 * on the pending state in netlink_mmap_sendmsg() while the CB hold
2787 * a reference to the skb.
2789 if (netlink_skb_is_mmaped(skb)) {
2790 skb = skb_copy(skb, GFP_KERNEL);
2791 if (skb == NULL)
2792 return -ENOBUFS;
2793 } else
2794 atomic_inc(&skb->users);
2796 sk = netlink_lookup(sock_net(ssk), ssk->sk_protocol, NETLINK_CB(skb).portid);
2797 if (sk == NULL) {
2798 ret = -ECONNREFUSED;
2799 goto error_free;
2802 nlk = nlk_sk(sk);
2803 mutex_lock(nlk->cb_mutex);
2804 /* A dump is in progress... */
2805 if (nlk->cb_running) {
2806 ret = -EBUSY;
2807 goto error_unlock;
2809 /* add reference of module which cb->dump belongs to */
2810 if (!try_module_get(control->module)) {
2811 ret = -EPROTONOSUPPORT;
2812 goto error_unlock;
2815 cb = &nlk->cb;
2816 memset(cb, 0, sizeof(*cb));
2817 cb->dump = control->dump;
2818 cb->done = control->done;
2819 cb->nlh = nlh;
2820 cb->data = control->data;
2821 cb->module = control->module;
2822 cb->min_dump_alloc = control->min_dump_alloc;
2823 cb->skb = skb;
2825 nlk->cb_running = true;
2827 mutex_unlock(nlk->cb_mutex);
2829 ret = netlink_dump(sk);
2830 sock_put(sk);
2832 if (ret)
2833 return ret;
2835 /* We successfully started a dump, by returning -EINTR we
2836 * signal not to send ACK even if it was requested.
2838 return -EINTR;
2840 error_unlock:
2841 sock_put(sk);
2842 mutex_unlock(nlk->cb_mutex);
2843 error_free:
2844 kfree_skb(skb);
2845 return ret;
2847 EXPORT_SYMBOL(__netlink_dump_start);
2849 void netlink_ack(struct sk_buff *in_skb, struct nlmsghdr *nlh, int err)
2851 struct sk_buff *skb;
2852 struct nlmsghdr *rep;
2853 struct nlmsgerr *errmsg;
2854 size_t payload = sizeof(*errmsg);
2856 /* error messages get the original request appened */
2857 if (err)
2858 payload += nlmsg_len(nlh);
2860 skb = netlink_alloc_skb(in_skb->sk, nlmsg_total_size(payload),
2861 NETLINK_CB(in_skb).portid, GFP_KERNEL);
2862 if (!skb) {
2863 struct sock *sk;
2865 sk = netlink_lookup(sock_net(in_skb->sk),
2866 in_skb->sk->sk_protocol,
2867 NETLINK_CB(in_skb).portid);
2868 if (sk) {
2869 sk->sk_err = ENOBUFS;
2870 sk->sk_error_report(sk);
2871 sock_put(sk);
2873 return;
2876 rep = __nlmsg_put(skb, NETLINK_CB(in_skb).portid, nlh->nlmsg_seq,
2877 NLMSG_ERROR, payload, 0);
2878 errmsg = nlmsg_data(rep);
2879 errmsg->error = err;
2880 memcpy(&errmsg->msg, nlh, err ? nlh->nlmsg_len : sizeof(*nlh));
2881 netlink_unicast(in_skb->sk, skb, NETLINK_CB(in_skb).portid, MSG_DONTWAIT);
2883 EXPORT_SYMBOL(netlink_ack);
2885 int netlink_rcv_skb(struct sk_buff *skb, int (*cb)(struct sk_buff *,
2886 struct nlmsghdr *))
2888 struct nlmsghdr *nlh;
2889 int err;
2891 while (skb->len >= nlmsg_total_size(0)) {
2892 int msglen;
2894 nlh = nlmsg_hdr(skb);
2895 err = 0;
2897 if (nlh->nlmsg_len < NLMSG_HDRLEN || skb->len < nlh->nlmsg_len)
2898 return 0;
2900 /* Only requests are handled by the kernel */
2901 if (!(nlh->nlmsg_flags & NLM_F_REQUEST))
2902 goto ack;
2904 /* Skip control messages */
2905 if (nlh->nlmsg_type < NLMSG_MIN_TYPE)
2906 goto ack;
2908 err = cb(skb, nlh);
2909 if (err == -EINTR)
2910 goto skip;
2912 ack:
2913 if (nlh->nlmsg_flags & NLM_F_ACK || err)
2914 netlink_ack(skb, nlh, err);
2916 skip:
2917 msglen = NLMSG_ALIGN(nlh->nlmsg_len);
2918 if (msglen > skb->len)
2919 msglen = skb->len;
2920 skb_pull(skb, msglen);
2923 return 0;
2925 EXPORT_SYMBOL(netlink_rcv_skb);
2928 * nlmsg_notify - send a notification netlink message
2929 * @sk: netlink socket to use
2930 * @skb: notification message
2931 * @portid: destination netlink portid for reports or 0
2932 * @group: destination multicast group or 0
2933 * @report: 1 to report back, 0 to disable
2934 * @flags: allocation flags
2936 int nlmsg_notify(struct sock *sk, struct sk_buff *skb, u32 portid,
2937 unsigned int group, int report, gfp_t flags)
2939 int err = 0;
2941 if (group) {
2942 int exclude_portid = 0;
2944 if (report) {
2945 atomic_inc(&skb->users);
2946 exclude_portid = portid;
2949 /* errors reported via destination sk->sk_err, but propagate
2950 * delivery errors if NETLINK_BROADCAST_ERROR flag is set */
2951 err = nlmsg_multicast(sk, skb, exclude_portid, group, flags);
2954 if (report) {
2955 int err2;
2957 err2 = nlmsg_unicast(sk, skb, portid);
2958 if (!err || err == -ESRCH)
2959 err = err2;
2962 return err;
2964 EXPORT_SYMBOL(nlmsg_notify);
2966 #ifdef CONFIG_PROC_FS
2967 struct nl_seq_iter {
2968 struct seq_net_private p;
2969 struct rhashtable_iter hti;
2970 int link;
2973 static int netlink_walk_start(struct nl_seq_iter *iter)
2975 int err;
2977 err = rhashtable_walk_init(&nl_table[iter->link].hash, &iter->hti);
2978 if (err) {
2979 iter->link = MAX_LINKS;
2980 return err;
2983 err = rhashtable_walk_start(&iter->hti);
2984 return err == -EAGAIN ? 0 : err;
2987 static void netlink_walk_stop(struct nl_seq_iter *iter)
2989 rhashtable_walk_stop(&iter->hti);
2990 rhashtable_walk_exit(&iter->hti);
2993 static void *__netlink_seq_next(struct seq_file *seq)
2995 struct nl_seq_iter *iter = seq->private;
2996 struct netlink_sock *nlk;
2998 do {
2999 for (;;) {
3000 int err;
3002 nlk = rhashtable_walk_next(&iter->hti);
3004 if (IS_ERR(nlk)) {
3005 if (PTR_ERR(nlk) == -EAGAIN)
3006 continue;
3008 return nlk;
3011 if (nlk)
3012 break;
3014 netlink_walk_stop(iter);
3015 if (++iter->link >= MAX_LINKS)
3016 return NULL;
3018 err = netlink_walk_start(iter);
3019 if (err)
3020 return ERR_PTR(err);
3022 } while (sock_net(&nlk->sk) != seq_file_net(seq));
3024 return nlk;
3027 static void *netlink_seq_start(struct seq_file *seq, loff_t *posp)
3029 struct nl_seq_iter *iter = seq->private;
3030 void *obj = SEQ_START_TOKEN;
3031 loff_t pos;
3032 int err;
3034 iter->link = 0;
3036 err = netlink_walk_start(iter);
3037 if (err)
3038 return ERR_PTR(err);
3040 for (pos = *posp; pos && obj && !IS_ERR(obj); pos--)
3041 obj = __netlink_seq_next(seq);
3043 return obj;
3046 static void *netlink_seq_next(struct seq_file *seq, void *v, loff_t *pos)
3048 ++*pos;
3049 return __netlink_seq_next(seq);
3052 static void netlink_seq_stop(struct seq_file *seq, void *v)
3054 struct nl_seq_iter *iter = seq->private;
3056 if (iter->link >= MAX_LINKS)
3057 return;
3059 netlink_walk_stop(iter);
3063 static int netlink_seq_show(struct seq_file *seq, void *v)
3065 if (v == SEQ_START_TOKEN) {
3066 seq_puts(seq,
3067 "sk Eth Pid Groups "
3068 "Rmem Wmem Dump Locks Drops Inode\n");
3069 } else {
3070 struct sock *s = v;
3071 struct netlink_sock *nlk = nlk_sk(s);
3073 seq_printf(seq, "%pK %-3d %-6u %08x %-8d %-8d %d %-8d %-8d %-8lu\n",
3075 s->sk_protocol,
3076 nlk->portid,
3077 nlk->groups ? (u32)nlk->groups[0] : 0,
3078 sk_rmem_alloc_get(s),
3079 sk_wmem_alloc_get(s),
3080 nlk->cb_running,
3081 atomic_read(&s->sk_refcnt),
3082 atomic_read(&s->sk_drops),
3083 sock_i_ino(s)
3087 return 0;
3090 static const struct seq_operations netlink_seq_ops = {
3091 .start = netlink_seq_start,
3092 .next = netlink_seq_next,
3093 .stop = netlink_seq_stop,
3094 .show = netlink_seq_show,
3098 static int netlink_seq_open(struct inode *inode, struct file *file)
3100 return seq_open_net(inode, file, &netlink_seq_ops,
3101 sizeof(struct nl_seq_iter));
3104 static const struct file_operations netlink_seq_fops = {
3105 .owner = THIS_MODULE,
3106 .open = netlink_seq_open,
3107 .read = seq_read,
3108 .llseek = seq_lseek,
3109 .release = seq_release_net,
3112 #endif
3114 int netlink_register_notifier(struct notifier_block *nb)
3116 return atomic_notifier_chain_register(&netlink_chain, nb);
3118 EXPORT_SYMBOL(netlink_register_notifier);
3120 int netlink_unregister_notifier(struct notifier_block *nb)
3122 return atomic_notifier_chain_unregister(&netlink_chain, nb);
3124 EXPORT_SYMBOL(netlink_unregister_notifier);
3126 static const struct proto_ops netlink_ops = {
3127 .family = PF_NETLINK,
3128 .owner = THIS_MODULE,
3129 .release = netlink_release,
3130 .bind = netlink_bind,
3131 .connect = netlink_connect,
3132 .socketpair = sock_no_socketpair,
3133 .accept = sock_no_accept,
3134 .getname = netlink_getname,
3135 .poll = netlink_poll,
3136 .ioctl = sock_no_ioctl,
3137 .listen = sock_no_listen,
3138 .shutdown = sock_no_shutdown,
3139 .setsockopt = netlink_setsockopt,
3140 .getsockopt = netlink_getsockopt,
3141 .sendmsg = netlink_sendmsg,
3142 .recvmsg = netlink_recvmsg,
3143 .mmap = netlink_mmap,
3144 .sendpage = sock_no_sendpage,
3147 static const struct net_proto_family netlink_family_ops = {
3148 .family = PF_NETLINK,
3149 .create = netlink_create,
3150 .owner = THIS_MODULE, /* for consistency 8) */
3153 static int __net_init netlink_net_init(struct net *net)
3155 #ifdef CONFIG_PROC_FS
3156 if (!proc_create("netlink", 0, net->proc_net, &netlink_seq_fops))
3157 return -ENOMEM;
3158 #endif
3159 return 0;
3162 static void __net_exit netlink_net_exit(struct net *net)
3164 #ifdef CONFIG_PROC_FS
3165 remove_proc_entry("netlink", net->proc_net);
3166 #endif
3169 static void __init netlink_add_usersock_entry(void)
3171 struct listeners *listeners;
3172 int groups = 32;
3174 listeners = kzalloc(sizeof(*listeners) + NLGRPSZ(groups), GFP_KERNEL);
3175 if (!listeners)
3176 panic("netlink_add_usersock_entry: Cannot allocate listeners\n");
3178 netlink_table_grab();
3180 nl_table[NETLINK_USERSOCK].groups = groups;
3181 rcu_assign_pointer(nl_table[NETLINK_USERSOCK].listeners, listeners);
3182 nl_table[NETLINK_USERSOCK].module = THIS_MODULE;
3183 nl_table[NETLINK_USERSOCK].registered = 1;
3184 nl_table[NETLINK_USERSOCK].flags = NL_CFG_F_NONROOT_SEND;
3186 netlink_table_ungrab();
3189 static struct pernet_operations __net_initdata netlink_net_ops = {
3190 .init = netlink_net_init,
3191 .exit = netlink_net_exit,
3194 static inline u32 netlink_hash(const void *data, u32 len, u32 seed)
3196 const struct netlink_sock *nlk = data;
3197 struct netlink_compare_arg arg;
3199 netlink_compare_arg_init(&arg, sock_net(&nlk->sk), nlk->portid);
3200 return jhash2((u32 *)&arg, netlink_compare_arg_len / sizeof(u32), seed);
3203 static const struct rhashtable_params netlink_rhashtable_params = {
3204 .head_offset = offsetof(struct netlink_sock, node),
3205 .key_len = netlink_compare_arg_len,
3206 .obj_hashfn = netlink_hash,
3207 .obj_cmpfn = netlink_compare,
3208 .automatic_shrinking = true,
3211 static int __init netlink_proto_init(void)
3213 int i;
3214 int err = proto_register(&netlink_proto, 0);
3216 if (err != 0)
3217 goto out;
3219 BUILD_BUG_ON(sizeof(struct netlink_skb_parms) > FIELD_SIZEOF(struct sk_buff, cb));
3221 nl_table = kcalloc(MAX_LINKS, sizeof(*nl_table), GFP_KERNEL);
3222 if (!nl_table)
3223 goto panic;
3225 for (i = 0; i < MAX_LINKS; i++) {
3226 if (rhashtable_init(&nl_table[i].hash,
3227 &netlink_rhashtable_params) < 0) {
3228 while (--i > 0)
3229 rhashtable_destroy(&nl_table[i].hash);
3230 kfree(nl_table);
3231 goto panic;
3235 INIT_LIST_HEAD(&netlink_tap_all);
3237 netlink_add_usersock_entry();
3239 sock_register(&netlink_family_ops);
3240 register_pernet_subsys(&netlink_net_ops);
3241 /* The netlink device handler may be needed early. */
3242 rtnetlink_init();
3243 out:
3244 return err;
3245 panic:
3246 panic("netlink_init: Cannot allocate nl_table\n");
3249 core_initcall(netlink_proto_init);