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>
37 #include <linux/fcntl.h>
38 #include <linux/termios.h>
39 #include <linux/sockios.h>
40 #include <linux/net.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>
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 <asm/cacheflush.h>
63 #include <net/net_namespace.h>
66 #include <net/netlink.h>
68 #include "af_netlink.h"
72 unsigned long masks[0];
76 #define NETLINK_CONGESTED 0x0
79 #define NETLINK_KERNEL_SOCKET 0x1
80 #define NETLINK_RECV_PKTINFO 0x2
81 #define NETLINK_BROADCAST_SEND_ERROR 0x4
82 #define NETLINK_RECV_NO_ENOBUFS 0x8
84 static inline int netlink_is_kernel(struct sock *sk)
86 return nlk_sk(sk)->flags & NETLINK_KERNEL_SOCKET;
89 struct netlink_table *nl_table;
90 EXPORT_SYMBOL_GPL(nl_table);
92 static DECLARE_WAIT_QUEUE_HEAD(nl_table_wait);
94 static int netlink_dump(struct sock *sk);
95 static void netlink_skb_destructor(struct sk_buff *skb);
97 DEFINE_RWLOCK(nl_table_lock);
98 EXPORT_SYMBOL_GPL(nl_table_lock);
99 static atomic_t nl_table_users = ATOMIC_INIT(0);
101 #define nl_deref_protected(X) rcu_dereference_protected(X, lockdep_is_held(&nl_table_lock));
103 static ATOMIC_NOTIFIER_HEAD(netlink_chain);
105 static DEFINE_SPINLOCK(netlink_tap_lock);
106 static struct list_head netlink_tap_all __read_mostly;
108 static inline u32 netlink_group_mask(u32 group)
110 return group ? 1 << (group - 1) : 0;
113 static inline struct hlist_head *nl_portid_hashfn(struct nl_portid_hash *hash, u32 portid)
115 return &hash->table[jhash_1word(portid, hash->rnd) & hash->mask];
118 int netlink_add_tap(struct netlink_tap *nt)
120 if (unlikely(nt->dev->type != ARPHRD_NETLINK))
123 spin_lock(&netlink_tap_lock);
124 list_add_rcu(&nt->list, &netlink_tap_all);
125 spin_unlock(&netlink_tap_lock);
128 __module_get(nt->module);
132 EXPORT_SYMBOL_GPL(netlink_add_tap);
134 int __netlink_remove_tap(struct netlink_tap *nt)
137 struct netlink_tap *tmp;
139 spin_lock(&netlink_tap_lock);
141 list_for_each_entry(tmp, &netlink_tap_all, list) {
143 list_del_rcu(&nt->list);
149 pr_warn("__netlink_remove_tap: %p not found\n", nt);
151 spin_unlock(&netlink_tap_lock);
153 if (found && nt->module)
154 module_put(nt->module);
156 return found ? 0 : -ENODEV;
158 EXPORT_SYMBOL_GPL(__netlink_remove_tap);
160 int netlink_remove_tap(struct netlink_tap *nt)
164 ret = __netlink_remove_tap(nt);
169 EXPORT_SYMBOL_GPL(netlink_remove_tap);
171 static int __netlink_deliver_tap_skb(struct sk_buff *skb,
172 struct net_device *dev)
174 struct sk_buff *nskb;
178 nskb = skb_clone(skb, GFP_ATOMIC);
181 ret = dev_queue_xmit(nskb);
182 if (unlikely(ret > 0))
183 ret = net_xmit_errno(ret);
190 static void __netlink_deliver_tap(struct sk_buff *skb)
193 struct netlink_tap *tmp;
195 list_for_each_entry_rcu(tmp, &netlink_tap_all, list) {
196 ret = __netlink_deliver_tap_skb(skb, tmp->dev);
202 static void netlink_deliver_tap(struct sk_buff *skb)
206 if (unlikely(!list_empty(&netlink_tap_all)))
207 __netlink_deliver_tap(skb);
212 static void netlink_overrun(struct sock *sk)
214 struct netlink_sock *nlk = nlk_sk(sk);
216 if (!(nlk->flags & NETLINK_RECV_NO_ENOBUFS)) {
217 if (!test_and_set_bit(NETLINK_CONGESTED, &nlk_sk(sk)->state)) {
218 sk->sk_err = ENOBUFS;
219 sk->sk_error_report(sk);
222 atomic_inc(&sk->sk_drops);
225 static void netlink_rcv_wake(struct sock *sk)
227 struct netlink_sock *nlk = nlk_sk(sk);
229 if (skb_queue_empty(&sk->sk_receive_queue))
230 clear_bit(NETLINK_CONGESTED, &nlk->state);
231 if (!test_bit(NETLINK_CONGESTED, &nlk->state))
232 wake_up_interruptible(&nlk->wait);
235 #ifdef CONFIG_NETLINK_MMAP
236 static bool netlink_skb_is_mmaped(const struct sk_buff *skb)
238 return NETLINK_CB(skb).flags & NETLINK_SKB_MMAPED;
241 static bool netlink_rx_is_mmaped(struct sock *sk)
243 return nlk_sk(sk)->rx_ring.pg_vec != NULL;
246 static bool netlink_tx_is_mmaped(struct sock *sk)
248 return nlk_sk(sk)->tx_ring.pg_vec != NULL;
251 static __pure struct page *pgvec_to_page(const void *addr)
253 if (is_vmalloc_addr(addr))
254 return vmalloc_to_page(addr);
256 return virt_to_page(addr);
259 static void free_pg_vec(void **pg_vec, unsigned int order, unsigned int len)
263 for (i = 0; i < len; i++) {
264 if (pg_vec[i] != NULL) {
265 if (is_vmalloc_addr(pg_vec[i]))
268 free_pages((unsigned long)pg_vec[i], order);
274 static void *alloc_one_pg_vec_page(unsigned long order)
277 gfp_t gfp_flags = GFP_KERNEL | __GFP_COMP | __GFP_ZERO |
278 __GFP_NOWARN | __GFP_NORETRY;
280 buffer = (void *)__get_free_pages(gfp_flags, order);
284 buffer = vzalloc((1 << order) * PAGE_SIZE);
288 gfp_flags &= ~__GFP_NORETRY;
289 return (void *)__get_free_pages(gfp_flags, order);
292 static void **alloc_pg_vec(struct netlink_sock *nlk,
293 struct nl_mmap_req *req, unsigned int order)
295 unsigned int block_nr = req->nm_block_nr;
299 pg_vec = kcalloc(block_nr, sizeof(void *), GFP_KERNEL);
303 for (i = 0; i < block_nr; i++) {
304 pg_vec[i] = ptr = alloc_one_pg_vec_page(order);
305 if (pg_vec[i] == NULL)
311 free_pg_vec(pg_vec, order, block_nr);
315 static int netlink_set_ring(struct sock *sk, struct nl_mmap_req *req,
316 bool closing, bool tx_ring)
318 struct netlink_sock *nlk = nlk_sk(sk);
319 struct netlink_ring *ring;
320 struct sk_buff_head *queue;
321 void **pg_vec = NULL;
322 unsigned int order = 0;
325 ring = tx_ring ? &nlk->tx_ring : &nlk->rx_ring;
326 queue = tx_ring ? &sk->sk_write_queue : &sk->sk_receive_queue;
329 if (atomic_read(&nlk->mapped))
331 if (atomic_read(&ring->pending))
335 if (req->nm_block_nr) {
336 if (ring->pg_vec != NULL)
339 if ((int)req->nm_block_size <= 0)
341 if (!IS_ALIGNED(req->nm_block_size, PAGE_SIZE))
343 if (req->nm_frame_size < NL_MMAP_HDRLEN)
345 if (!IS_ALIGNED(req->nm_frame_size, NL_MMAP_MSG_ALIGNMENT))
348 ring->frames_per_block = req->nm_block_size /
350 if (ring->frames_per_block == 0)
352 if (ring->frames_per_block * req->nm_block_nr !=
356 order = get_order(req->nm_block_size);
357 pg_vec = alloc_pg_vec(nlk, req, order);
361 if (req->nm_frame_nr)
366 mutex_lock(&nlk->pg_vec_lock);
367 if (closing || atomic_read(&nlk->mapped) == 0) {
369 spin_lock_bh(&queue->lock);
371 ring->frame_max = req->nm_frame_nr - 1;
373 ring->frame_size = req->nm_frame_size;
374 ring->pg_vec_pages = req->nm_block_size / PAGE_SIZE;
376 swap(ring->pg_vec_len, req->nm_block_nr);
377 swap(ring->pg_vec_order, order);
378 swap(ring->pg_vec, pg_vec);
380 __skb_queue_purge(queue);
381 spin_unlock_bh(&queue->lock);
383 WARN_ON(atomic_read(&nlk->mapped));
385 mutex_unlock(&nlk->pg_vec_lock);
388 free_pg_vec(pg_vec, order, req->nm_block_nr);
392 static void netlink_mm_open(struct vm_area_struct *vma)
394 struct file *file = vma->vm_file;
395 struct socket *sock = file->private_data;
396 struct sock *sk = sock->sk;
399 atomic_inc(&nlk_sk(sk)->mapped);
402 static void netlink_mm_close(struct vm_area_struct *vma)
404 struct file *file = vma->vm_file;
405 struct socket *sock = file->private_data;
406 struct sock *sk = sock->sk;
409 atomic_dec(&nlk_sk(sk)->mapped);
412 static const struct vm_operations_struct netlink_mmap_ops = {
413 .open = netlink_mm_open,
414 .close = netlink_mm_close,
417 static int netlink_mmap(struct file *file, struct socket *sock,
418 struct vm_area_struct *vma)
420 struct sock *sk = sock->sk;
421 struct netlink_sock *nlk = nlk_sk(sk);
422 struct netlink_ring *ring;
423 unsigned long start, size, expected;
430 mutex_lock(&nlk->pg_vec_lock);
433 for (ring = &nlk->rx_ring; ring <= &nlk->tx_ring; ring++) {
434 if (ring->pg_vec == NULL)
436 expected += ring->pg_vec_len * ring->pg_vec_pages * PAGE_SIZE;
442 size = vma->vm_end - vma->vm_start;
443 if (size != expected)
446 start = vma->vm_start;
447 for (ring = &nlk->rx_ring; ring <= &nlk->tx_ring; ring++) {
448 if (ring->pg_vec == NULL)
451 for (i = 0; i < ring->pg_vec_len; i++) {
453 void *kaddr = ring->pg_vec[i];
456 for (pg_num = 0; pg_num < ring->pg_vec_pages; pg_num++) {
457 page = pgvec_to_page(kaddr);
458 err = vm_insert_page(vma, start, page);
467 atomic_inc(&nlk->mapped);
468 vma->vm_ops = &netlink_mmap_ops;
471 mutex_unlock(&nlk->pg_vec_lock);
475 static void netlink_frame_flush_dcache(const struct nl_mmap_hdr *hdr)
477 #if ARCH_IMPLEMENTS_FLUSH_DCACHE_PAGE == 1
478 struct page *p_start, *p_end;
480 /* First page is flushed through netlink_{get,set}_status */
481 p_start = pgvec_to_page(hdr + PAGE_SIZE);
482 p_end = pgvec_to_page((void *)hdr + NL_MMAP_HDRLEN + hdr->nm_len - 1);
483 while (p_start <= p_end) {
484 flush_dcache_page(p_start);
490 static enum nl_mmap_status netlink_get_status(const struct nl_mmap_hdr *hdr)
493 flush_dcache_page(pgvec_to_page(hdr));
494 return hdr->nm_status;
497 static void netlink_set_status(struct nl_mmap_hdr *hdr,
498 enum nl_mmap_status status)
500 hdr->nm_status = status;
501 flush_dcache_page(pgvec_to_page(hdr));
505 static struct nl_mmap_hdr *
506 __netlink_lookup_frame(const struct netlink_ring *ring, unsigned int pos)
508 unsigned int pg_vec_pos, frame_off;
510 pg_vec_pos = pos / ring->frames_per_block;
511 frame_off = pos % ring->frames_per_block;
513 return ring->pg_vec[pg_vec_pos] + (frame_off * ring->frame_size);
516 static struct nl_mmap_hdr *
517 netlink_lookup_frame(const struct netlink_ring *ring, unsigned int pos,
518 enum nl_mmap_status status)
520 struct nl_mmap_hdr *hdr;
522 hdr = __netlink_lookup_frame(ring, pos);
523 if (netlink_get_status(hdr) != status)
529 static struct nl_mmap_hdr *
530 netlink_current_frame(const struct netlink_ring *ring,
531 enum nl_mmap_status status)
533 return netlink_lookup_frame(ring, ring->head, status);
536 static struct nl_mmap_hdr *
537 netlink_previous_frame(const struct netlink_ring *ring,
538 enum nl_mmap_status status)
542 prev = ring->head ? ring->head - 1 : ring->frame_max;
543 return netlink_lookup_frame(ring, prev, status);
546 static void netlink_increment_head(struct netlink_ring *ring)
548 ring->head = ring->head != ring->frame_max ? ring->head + 1 : 0;
551 static void netlink_forward_ring(struct netlink_ring *ring)
553 unsigned int head = ring->head, pos = head;
554 const struct nl_mmap_hdr *hdr;
557 hdr = __netlink_lookup_frame(ring, pos);
558 if (hdr->nm_status == NL_MMAP_STATUS_UNUSED)
560 if (hdr->nm_status != NL_MMAP_STATUS_SKIP)
562 netlink_increment_head(ring);
563 } while (ring->head != head);
566 static bool netlink_dump_space(struct netlink_sock *nlk)
568 struct netlink_ring *ring = &nlk->rx_ring;
569 struct nl_mmap_hdr *hdr;
572 hdr = netlink_current_frame(ring, NL_MMAP_STATUS_UNUSED);
576 n = ring->head + ring->frame_max / 2;
577 if (n > ring->frame_max)
578 n -= ring->frame_max;
580 hdr = __netlink_lookup_frame(ring, n);
582 return hdr->nm_status == NL_MMAP_STATUS_UNUSED;
585 static unsigned int netlink_poll(struct file *file, struct socket *sock,
588 struct sock *sk = sock->sk;
589 struct netlink_sock *nlk = nlk_sk(sk);
593 if (nlk->rx_ring.pg_vec != NULL) {
594 /* Memory mapped sockets don't call recvmsg(), so flow control
595 * for dumps is performed here. A dump is allowed to continue
596 * if at least half the ring is unused.
598 while (nlk->cb != NULL && netlink_dump_space(nlk)) {
599 err = netlink_dump(sk);
602 sk->sk_error_report(sk);
606 netlink_rcv_wake(sk);
609 mask = datagram_poll(file, sock, wait);
611 spin_lock_bh(&sk->sk_receive_queue.lock);
612 if (nlk->rx_ring.pg_vec) {
613 netlink_forward_ring(&nlk->rx_ring);
614 if (!netlink_previous_frame(&nlk->rx_ring, NL_MMAP_STATUS_UNUSED))
615 mask |= POLLIN | POLLRDNORM;
617 spin_unlock_bh(&sk->sk_receive_queue.lock);
619 spin_lock_bh(&sk->sk_write_queue.lock);
620 if (nlk->tx_ring.pg_vec) {
621 if (netlink_current_frame(&nlk->tx_ring, NL_MMAP_STATUS_UNUSED))
622 mask |= POLLOUT | POLLWRNORM;
624 spin_unlock_bh(&sk->sk_write_queue.lock);
629 static struct nl_mmap_hdr *netlink_mmap_hdr(struct sk_buff *skb)
631 return (struct nl_mmap_hdr *)(skb->head - NL_MMAP_HDRLEN);
634 static void netlink_ring_setup_skb(struct sk_buff *skb, struct sock *sk,
635 struct netlink_ring *ring,
636 struct nl_mmap_hdr *hdr)
641 size = ring->frame_size - NL_MMAP_HDRLEN;
642 data = (void *)hdr + NL_MMAP_HDRLEN;
646 skb_reset_tail_pointer(skb);
647 skb->end = skb->tail + size;
650 skb->destructor = netlink_skb_destructor;
651 NETLINK_CB(skb).flags |= NETLINK_SKB_MMAPED;
652 NETLINK_CB(skb).sk = sk;
655 static int netlink_mmap_sendmsg(struct sock *sk, struct msghdr *msg,
656 u32 dst_portid, u32 dst_group,
657 struct sock_iocb *siocb)
659 struct netlink_sock *nlk = nlk_sk(sk);
660 struct netlink_ring *ring;
661 struct nl_mmap_hdr *hdr;
665 int err = 0, len = 0;
667 /* Netlink messages are validated by the receiver before processing.
668 * In order to avoid userspace changing the contents of the message
669 * after validation, the socket and the ring may only be used by a
670 * single process, otherwise we fall back to copying.
672 if (atomic_long_read(&sk->sk_socket->file->f_count) > 2 ||
673 atomic_read(&nlk->mapped) > 1)
676 mutex_lock(&nlk->pg_vec_lock);
678 ring = &nlk->tx_ring;
679 maxlen = ring->frame_size - NL_MMAP_HDRLEN;
682 hdr = netlink_current_frame(ring, NL_MMAP_STATUS_VALID);
684 if (!(msg->msg_flags & MSG_DONTWAIT) &&
685 atomic_read(&nlk->tx_ring.pending))
689 if (hdr->nm_len > maxlen) {
694 netlink_frame_flush_dcache(hdr);
696 if (likely(dst_portid == 0 && dst_group == 0 && excl)) {
697 skb = alloc_skb_head(GFP_KERNEL);
703 netlink_ring_setup_skb(skb, sk, ring, hdr);
704 NETLINK_CB(skb).flags |= NETLINK_SKB_TX;
705 __skb_put(skb, hdr->nm_len);
706 netlink_set_status(hdr, NL_MMAP_STATUS_RESERVED);
707 atomic_inc(&ring->pending);
709 skb = alloc_skb(hdr->nm_len, GFP_KERNEL);
714 __skb_put(skb, hdr->nm_len);
715 memcpy(skb->data, (void *)hdr + NL_MMAP_HDRLEN, hdr->nm_len);
716 netlink_set_status(hdr, NL_MMAP_STATUS_UNUSED);
719 netlink_increment_head(ring);
721 NETLINK_CB(skb).portid = nlk->portid;
722 NETLINK_CB(skb).dst_group = dst_group;
723 NETLINK_CB(skb).creds = siocb->scm->creds;
725 err = security_netlink_send(sk, skb);
731 if (unlikely(dst_group)) {
732 atomic_inc(&skb->users);
733 netlink_broadcast(sk, skb, dst_portid, dst_group,
736 err = netlink_unicast(sk, skb, dst_portid,
737 msg->msg_flags & MSG_DONTWAIT);
742 } while (hdr != NULL ||
743 (!(msg->msg_flags & MSG_DONTWAIT) &&
744 atomic_read(&nlk->tx_ring.pending)));
749 mutex_unlock(&nlk->pg_vec_lock);
753 static void netlink_queue_mmaped_skb(struct sock *sk, struct sk_buff *skb)
755 struct nl_mmap_hdr *hdr;
757 hdr = netlink_mmap_hdr(skb);
758 hdr->nm_len = skb->len;
759 hdr->nm_group = NETLINK_CB(skb).dst_group;
760 hdr->nm_pid = NETLINK_CB(skb).creds.pid;
761 hdr->nm_uid = from_kuid(sk_user_ns(sk), NETLINK_CB(skb).creds.uid);
762 hdr->nm_gid = from_kgid(sk_user_ns(sk), NETLINK_CB(skb).creds.gid);
763 netlink_frame_flush_dcache(hdr);
764 netlink_set_status(hdr, NL_MMAP_STATUS_VALID);
766 NETLINK_CB(skb).flags |= NETLINK_SKB_DELIVERED;
770 static void netlink_ring_set_copied(struct sock *sk, struct sk_buff *skb)
772 struct netlink_sock *nlk = nlk_sk(sk);
773 struct netlink_ring *ring = &nlk->rx_ring;
774 struct nl_mmap_hdr *hdr;
776 spin_lock_bh(&sk->sk_receive_queue.lock);
777 hdr = netlink_current_frame(ring, NL_MMAP_STATUS_UNUSED);
779 spin_unlock_bh(&sk->sk_receive_queue.lock);
784 netlink_increment_head(ring);
785 __skb_queue_tail(&sk->sk_receive_queue, skb);
786 spin_unlock_bh(&sk->sk_receive_queue.lock);
788 hdr->nm_len = skb->len;
789 hdr->nm_group = NETLINK_CB(skb).dst_group;
790 hdr->nm_pid = NETLINK_CB(skb).creds.pid;
791 hdr->nm_uid = from_kuid(sk_user_ns(sk), NETLINK_CB(skb).creds.uid);
792 hdr->nm_gid = from_kgid(sk_user_ns(sk), NETLINK_CB(skb).creds.gid);
793 netlink_set_status(hdr, NL_MMAP_STATUS_COPY);
796 #else /* CONFIG_NETLINK_MMAP */
797 #define netlink_skb_is_mmaped(skb) false
798 #define netlink_rx_is_mmaped(sk) false
799 #define netlink_tx_is_mmaped(sk) false
800 #define netlink_mmap sock_no_mmap
801 #define netlink_poll datagram_poll
802 #define netlink_mmap_sendmsg(sk, msg, dst_portid, dst_group, siocb) 0
803 #endif /* CONFIG_NETLINK_MMAP */
805 static void netlink_destroy_callback(struct netlink_callback *cb)
811 static void netlink_consume_callback(struct netlink_callback *cb)
813 consume_skb(cb->skb);
817 static void netlink_skb_destructor(struct sk_buff *skb)
819 #ifdef CONFIG_NETLINK_MMAP
820 struct nl_mmap_hdr *hdr;
821 struct netlink_ring *ring;
824 /* If a packet from the kernel to userspace was freed because of an
825 * error without being delivered to userspace, the kernel must reset
826 * the status. In the direction userspace to kernel, the status is
827 * always reset here after the packet was processed and freed.
829 if (netlink_skb_is_mmaped(skb)) {
830 hdr = netlink_mmap_hdr(skb);
831 sk = NETLINK_CB(skb).sk;
833 if (NETLINK_CB(skb).flags & NETLINK_SKB_TX) {
834 netlink_set_status(hdr, NL_MMAP_STATUS_UNUSED);
835 ring = &nlk_sk(sk)->tx_ring;
837 if (!(NETLINK_CB(skb).flags & NETLINK_SKB_DELIVERED)) {
839 netlink_set_status(hdr, NL_MMAP_STATUS_VALID);
841 ring = &nlk_sk(sk)->rx_ring;
844 WARN_ON(atomic_read(&ring->pending) == 0);
845 atomic_dec(&ring->pending);
851 if (is_vmalloc_addr(skb->head)) {
859 static void netlink_skb_set_owner_r(struct sk_buff *skb, struct sock *sk)
861 WARN_ON(skb->sk != NULL);
863 skb->destructor = netlink_skb_destructor;
864 atomic_add(skb->truesize, &sk->sk_rmem_alloc);
865 sk_mem_charge(sk, skb->truesize);
868 static void netlink_sock_destruct(struct sock *sk)
870 struct netlink_sock *nlk = nlk_sk(sk);
874 nlk->cb->done(nlk->cb);
876 module_put(nlk->cb->module);
877 netlink_destroy_callback(nlk->cb);
880 skb_queue_purge(&sk->sk_receive_queue);
881 #ifdef CONFIG_NETLINK_MMAP
883 struct nl_mmap_req req;
885 memset(&req, 0, sizeof(req));
886 if (nlk->rx_ring.pg_vec)
887 netlink_set_ring(sk, &req, true, false);
888 memset(&req, 0, sizeof(req));
889 if (nlk->tx_ring.pg_vec)
890 netlink_set_ring(sk, &req, true, true);
892 #endif /* CONFIG_NETLINK_MMAP */
894 if (!sock_flag(sk, SOCK_DEAD)) {
895 printk(KERN_ERR "Freeing alive netlink socket %p\n", sk);
899 WARN_ON(atomic_read(&sk->sk_rmem_alloc));
900 WARN_ON(atomic_read(&sk->sk_wmem_alloc));
901 WARN_ON(nlk_sk(sk)->groups);
904 /* This lock without WQ_FLAG_EXCLUSIVE is good on UP and it is _very_ bad on
905 * SMP. Look, when several writers sleep and reader wakes them up, all but one
906 * immediately hit write lock and grab all the cpus. Exclusive sleep solves
907 * this, _but_ remember, it adds useless work on UP machines.
910 void netlink_table_grab(void)
911 __acquires(nl_table_lock)
915 write_lock_irq(&nl_table_lock);
917 if (atomic_read(&nl_table_users)) {
918 DECLARE_WAITQUEUE(wait, current);
920 add_wait_queue_exclusive(&nl_table_wait, &wait);
922 set_current_state(TASK_UNINTERRUPTIBLE);
923 if (atomic_read(&nl_table_users) == 0)
925 write_unlock_irq(&nl_table_lock);
927 write_lock_irq(&nl_table_lock);
930 __set_current_state(TASK_RUNNING);
931 remove_wait_queue(&nl_table_wait, &wait);
935 void netlink_table_ungrab(void)
936 __releases(nl_table_lock)
938 write_unlock_irq(&nl_table_lock);
939 wake_up(&nl_table_wait);
943 netlink_lock_table(void)
945 /* read_lock() synchronizes us to netlink_table_grab */
947 read_lock(&nl_table_lock);
948 atomic_inc(&nl_table_users);
949 read_unlock(&nl_table_lock);
953 netlink_unlock_table(void)
955 if (atomic_dec_and_test(&nl_table_users))
956 wake_up(&nl_table_wait);
959 static bool netlink_compare(struct net *net, struct sock *sk)
961 return net_eq(sock_net(sk), net);
964 static struct sock *netlink_lookup(struct net *net, int protocol, u32 portid)
966 struct netlink_table *table = &nl_table[protocol];
967 struct nl_portid_hash *hash = &table->hash;
968 struct hlist_head *head;
971 read_lock(&nl_table_lock);
972 head = nl_portid_hashfn(hash, portid);
973 sk_for_each(sk, head) {
974 if (table->compare(net, sk) &&
975 (nlk_sk(sk)->portid == portid)) {
982 read_unlock(&nl_table_lock);
986 static struct hlist_head *nl_portid_hash_zalloc(size_t size)
988 if (size <= PAGE_SIZE)
989 return kzalloc(size, GFP_ATOMIC);
991 return (struct hlist_head *)
992 __get_free_pages(GFP_ATOMIC | __GFP_ZERO,
996 static void nl_portid_hash_free(struct hlist_head *table, size_t size)
998 if (size <= PAGE_SIZE)
1001 free_pages((unsigned long)table, get_order(size));
1004 static int nl_portid_hash_rehash(struct nl_portid_hash *hash, int grow)
1006 unsigned int omask, mask, shift;
1008 struct hlist_head *otable, *table;
1011 omask = mask = hash->mask;
1012 osize = size = (mask + 1) * sizeof(*table);
1013 shift = hash->shift;
1016 if (++shift > hash->max_shift)
1018 mask = mask * 2 + 1;
1022 table = nl_portid_hash_zalloc(size);
1026 otable = hash->table;
1027 hash->table = table;
1029 hash->shift = shift;
1030 get_random_bytes(&hash->rnd, sizeof(hash->rnd));
1032 for (i = 0; i <= omask; i++) {
1034 struct hlist_node *tmp;
1036 sk_for_each_safe(sk, tmp, &otable[i])
1037 __sk_add_node(sk, nl_portid_hashfn(hash, nlk_sk(sk)->portid));
1040 nl_portid_hash_free(otable, osize);
1041 hash->rehash_time = jiffies + 10 * 60 * HZ;
1045 static inline int nl_portid_hash_dilute(struct nl_portid_hash *hash, int len)
1047 int avg = hash->entries >> hash->shift;
1049 if (unlikely(avg > 1) && nl_portid_hash_rehash(hash, 1))
1052 if (unlikely(len > avg) && time_after(jiffies, hash->rehash_time)) {
1053 nl_portid_hash_rehash(hash, 0);
1060 static const struct proto_ops netlink_ops;
1063 netlink_update_listeners(struct sock *sk)
1065 struct netlink_table *tbl = &nl_table[sk->sk_protocol];
1068 struct listeners *listeners;
1070 listeners = nl_deref_protected(tbl->listeners);
1074 for (i = 0; i < NLGRPLONGS(tbl->groups); i++) {
1076 sk_for_each_bound(sk, &tbl->mc_list) {
1077 if (i < NLGRPLONGS(nlk_sk(sk)->ngroups))
1078 mask |= nlk_sk(sk)->groups[i];
1080 listeners->masks[i] = mask;
1082 /* this function is only called with the netlink table "grabbed", which
1083 * makes sure updates are visible before bind or setsockopt return. */
1086 static int netlink_insert(struct sock *sk, struct net *net, u32 portid)
1088 struct netlink_table *table = &nl_table[sk->sk_protocol];
1089 struct nl_portid_hash *hash = &table->hash;
1090 struct hlist_head *head;
1091 int err = -EADDRINUSE;
1095 netlink_table_grab();
1096 head = nl_portid_hashfn(hash, portid);
1098 sk_for_each(osk, head) {
1099 if (table->compare(net, osk) &&
1100 (nlk_sk(osk)->portid == portid))
1108 if (nlk_sk(sk)->portid)
1112 if (BITS_PER_LONG > 32 && unlikely(hash->entries >= UINT_MAX))
1115 if (len && nl_portid_hash_dilute(hash, len))
1116 head = nl_portid_hashfn(hash, portid);
1118 nlk_sk(sk)->portid = portid;
1119 sk_add_node(sk, head);
1123 netlink_table_ungrab();
1127 static void netlink_remove(struct sock *sk)
1129 netlink_table_grab();
1130 if (sk_del_node_init(sk))
1131 nl_table[sk->sk_protocol].hash.entries--;
1132 if (nlk_sk(sk)->subscriptions)
1133 __sk_del_bind_node(sk);
1134 netlink_table_ungrab();
1137 static struct proto netlink_proto = {
1139 .owner = THIS_MODULE,
1140 .obj_size = sizeof(struct netlink_sock),
1143 static int __netlink_create(struct net *net, struct socket *sock,
1144 struct mutex *cb_mutex, int protocol)
1147 struct netlink_sock *nlk;
1149 sock->ops = &netlink_ops;
1151 sk = sk_alloc(net, PF_NETLINK, GFP_KERNEL, &netlink_proto);
1155 sock_init_data(sock, sk);
1159 nlk->cb_mutex = cb_mutex;
1161 nlk->cb_mutex = &nlk->cb_def_mutex;
1162 mutex_init(nlk->cb_mutex);
1164 init_waitqueue_head(&nlk->wait);
1165 #ifdef CONFIG_NETLINK_MMAP
1166 mutex_init(&nlk->pg_vec_lock);
1169 sk->sk_destruct = netlink_sock_destruct;
1170 sk->sk_protocol = protocol;
1174 static int netlink_create(struct net *net, struct socket *sock, int protocol,
1177 struct module *module = NULL;
1178 struct mutex *cb_mutex;
1179 struct netlink_sock *nlk;
1180 void (*bind)(int group);
1183 sock->state = SS_UNCONNECTED;
1185 if (sock->type != SOCK_RAW && sock->type != SOCK_DGRAM)
1186 return -ESOCKTNOSUPPORT;
1188 if (protocol < 0 || protocol >= MAX_LINKS)
1189 return -EPROTONOSUPPORT;
1191 netlink_lock_table();
1192 #ifdef CONFIG_MODULES
1193 if (!nl_table[protocol].registered) {
1194 netlink_unlock_table();
1195 request_module("net-pf-%d-proto-%d", PF_NETLINK, protocol);
1196 netlink_lock_table();
1199 if (nl_table[protocol].registered &&
1200 try_module_get(nl_table[protocol].module))
1201 module = nl_table[protocol].module;
1203 err = -EPROTONOSUPPORT;
1204 cb_mutex = nl_table[protocol].cb_mutex;
1205 bind = nl_table[protocol].bind;
1206 netlink_unlock_table();
1211 err = __netlink_create(net, sock, cb_mutex, protocol);
1216 sock_prot_inuse_add(net, &netlink_proto, 1);
1219 nlk = nlk_sk(sock->sk);
1220 nlk->module = module;
1221 nlk->netlink_bind = bind;
1230 static int netlink_release(struct socket *sock)
1232 struct sock *sk = sock->sk;
1233 struct netlink_sock *nlk;
1243 * OK. Socket is unlinked, any packets that arrive now
1248 wake_up_interruptible_all(&nlk->wait);
1250 skb_queue_purge(&sk->sk_write_queue);
1253 struct netlink_notify n = {
1254 .net = sock_net(sk),
1255 .protocol = sk->sk_protocol,
1256 .portid = nlk->portid,
1258 atomic_notifier_call_chain(&netlink_chain,
1259 NETLINK_URELEASE, &n);
1262 module_put(nlk->module);
1264 netlink_table_grab();
1265 if (netlink_is_kernel(sk)) {
1266 BUG_ON(nl_table[sk->sk_protocol].registered == 0);
1267 if (--nl_table[sk->sk_protocol].registered == 0) {
1268 struct listeners *old;
1270 old = nl_deref_protected(nl_table[sk->sk_protocol].listeners);
1271 RCU_INIT_POINTER(nl_table[sk->sk_protocol].listeners, NULL);
1272 kfree_rcu(old, rcu);
1273 nl_table[sk->sk_protocol].module = NULL;
1274 nl_table[sk->sk_protocol].bind = NULL;
1275 nl_table[sk->sk_protocol].flags = 0;
1276 nl_table[sk->sk_protocol].registered = 0;
1278 } else if (nlk->subscriptions) {
1279 netlink_update_listeners(sk);
1281 netlink_table_ungrab();
1287 sock_prot_inuse_add(sock_net(sk), &netlink_proto, -1);
1293 static int netlink_autobind(struct socket *sock)
1295 struct sock *sk = sock->sk;
1296 struct net *net = sock_net(sk);
1297 struct netlink_table *table = &nl_table[sk->sk_protocol];
1298 struct nl_portid_hash *hash = &table->hash;
1299 struct hlist_head *head;
1301 s32 portid = task_tgid_vnr(current);
1303 static s32 rover = -4097;
1307 netlink_table_grab();
1308 head = nl_portid_hashfn(hash, portid);
1309 sk_for_each(osk, head) {
1310 if (!table->compare(net, osk))
1312 if (nlk_sk(osk)->portid == portid) {
1313 /* Bind collision, search negative portid values. */
1317 netlink_table_ungrab();
1321 netlink_table_ungrab();
1323 err = netlink_insert(sk, net, portid);
1324 if (err == -EADDRINUSE)
1327 /* If 2 threads race to autobind, that is fine. */
1334 static inline int netlink_capable(const struct socket *sock, unsigned int flag)
1336 return (nl_table[sock->sk->sk_protocol].flags & flag) ||
1337 ns_capable(sock_net(sock->sk)->user_ns, CAP_NET_ADMIN);
1341 netlink_update_subscriptions(struct sock *sk, unsigned int subscriptions)
1343 struct netlink_sock *nlk = nlk_sk(sk);
1345 if (nlk->subscriptions && !subscriptions)
1346 __sk_del_bind_node(sk);
1347 else if (!nlk->subscriptions && subscriptions)
1348 sk_add_bind_node(sk, &nl_table[sk->sk_protocol].mc_list);
1349 nlk->subscriptions = subscriptions;
1352 static int netlink_realloc_groups(struct sock *sk)
1354 struct netlink_sock *nlk = nlk_sk(sk);
1355 unsigned int groups;
1356 unsigned long *new_groups;
1359 netlink_table_grab();
1361 groups = nl_table[sk->sk_protocol].groups;
1362 if (!nl_table[sk->sk_protocol].registered) {
1367 if (nlk->ngroups >= groups)
1370 new_groups = krealloc(nlk->groups, NLGRPSZ(groups), GFP_ATOMIC);
1371 if (new_groups == NULL) {
1375 memset((char *)new_groups + NLGRPSZ(nlk->ngroups), 0,
1376 NLGRPSZ(groups) - NLGRPSZ(nlk->ngroups));
1378 nlk->groups = new_groups;
1379 nlk->ngroups = groups;
1381 netlink_table_ungrab();
1385 static int netlink_bind(struct socket *sock, struct sockaddr *addr,
1388 struct sock *sk = sock->sk;
1389 struct net *net = sock_net(sk);
1390 struct netlink_sock *nlk = nlk_sk(sk);
1391 struct sockaddr_nl *nladdr = (struct sockaddr_nl *)addr;
1394 if (addr_len < sizeof(struct sockaddr_nl))
1397 if (nladdr->nl_family != AF_NETLINK)
1400 /* Only superuser is allowed to listen multicasts */
1401 if (nladdr->nl_groups) {
1402 if (!netlink_capable(sock, NL_CFG_F_NONROOT_RECV))
1404 err = netlink_realloc_groups(sk);
1410 if (nladdr->nl_pid != nlk->portid)
1413 err = nladdr->nl_pid ?
1414 netlink_insert(sk, net, nladdr->nl_pid) :
1415 netlink_autobind(sock);
1420 if (!nladdr->nl_groups && (nlk->groups == NULL || !(u32)nlk->groups[0]))
1423 netlink_table_grab();
1424 netlink_update_subscriptions(sk, nlk->subscriptions +
1425 hweight32(nladdr->nl_groups) -
1426 hweight32(nlk->groups[0]));
1427 nlk->groups[0] = (nlk->groups[0] & ~0xffffffffUL) | nladdr->nl_groups;
1428 netlink_update_listeners(sk);
1429 netlink_table_ungrab();
1431 if (nlk->netlink_bind && nlk->groups[0]) {
1434 for (i=0; i<nlk->ngroups; i++) {
1435 if (test_bit(i, nlk->groups))
1436 nlk->netlink_bind(i);
1443 static int netlink_connect(struct socket *sock, struct sockaddr *addr,
1444 int alen, int flags)
1447 struct sock *sk = sock->sk;
1448 struct netlink_sock *nlk = nlk_sk(sk);
1449 struct sockaddr_nl *nladdr = (struct sockaddr_nl *)addr;
1451 if (alen < sizeof(addr->sa_family))
1454 if (addr->sa_family == AF_UNSPEC) {
1455 sk->sk_state = NETLINK_UNCONNECTED;
1456 nlk->dst_portid = 0;
1460 if (addr->sa_family != AF_NETLINK)
1463 /* Only superuser is allowed to send multicasts */
1464 if (nladdr->nl_groups && !netlink_capable(sock, NL_CFG_F_NONROOT_SEND))
1468 err = netlink_autobind(sock);
1471 sk->sk_state = NETLINK_CONNECTED;
1472 nlk->dst_portid = nladdr->nl_pid;
1473 nlk->dst_group = ffs(nladdr->nl_groups);
1479 static int netlink_getname(struct socket *sock, struct sockaddr *addr,
1480 int *addr_len, int peer)
1482 struct sock *sk = sock->sk;
1483 struct netlink_sock *nlk = nlk_sk(sk);
1484 DECLARE_SOCKADDR(struct sockaddr_nl *, nladdr, addr);
1486 nladdr->nl_family = AF_NETLINK;
1488 *addr_len = sizeof(*nladdr);
1491 nladdr->nl_pid = nlk->dst_portid;
1492 nladdr->nl_groups = netlink_group_mask(nlk->dst_group);
1494 nladdr->nl_pid = nlk->portid;
1495 nladdr->nl_groups = nlk->groups ? nlk->groups[0] : 0;
1500 static struct sock *netlink_getsockbyportid(struct sock *ssk, u32 portid)
1503 struct netlink_sock *nlk;
1505 sock = netlink_lookup(sock_net(ssk), ssk->sk_protocol, portid);
1507 return ERR_PTR(-ECONNREFUSED);
1509 /* Don't bother queuing skb if kernel socket has no input function */
1511 if (sock->sk_state == NETLINK_CONNECTED &&
1512 nlk->dst_portid != nlk_sk(ssk)->portid) {
1514 return ERR_PTR(-ECONNREFUSED);
1519 struct sock *netlink_getsockbyfilp(struct file *filp)
1521 struct inode *inode = file_inode(filp);
1524 if (!S_ISSOCK(inode->i_mode))
1525 return ERR_PTR(-ENOTSOCK);
1527 sock = SOCKET_I(inode)->sk;
1528 if (sock->sk_family != AF_NETLINK)
1529 return ERR_PTR(-EINVAL);
1535 static struct sk_buff *netlink_alloc_large_skb(unsigned int size)
1537 struct sk_buff *skb;
1540 if (size <= NLMSG_GOODSIZE)
1541 return alloc_skb(size, GFP_KERNEL);
1543 skb = alloc_skb_head(GFP_KERNEL);
1547 data = vmalloc(size);
1553 skb_reset_tail_pointer(skb);
1554 skb->end = skb->tail + size;
1556 skb->destructor = netlink_skb_destructor;
1565 * Attach a skb to a netlink socket.
1566 * The caller must hold a reference to the destination socket. On error, the
1567 * reference is dropped. The skb is not send to the destination, just all
1568 * all error checks are performed and memory in the queue is reserved.
1570 * < 0: error. skb freed, reference to sock dropped.
1572 * 1: repeat lookup - reference dropped while waiting for socket memory.
1574 int netlink_attachskb(struct sock *sk, struct sk_buff *skb,
1575 long *timeo, struct sock *ssk)
1577 struct netlink_sock *nlk;
1581 if ((atomic_read(&sk->sk_rmem_alloc) > sk->sk_rcvbuf ||
1582 test_bit(NETLINK_CONGESTED, &nlk->state)) &&
1583 !netlink_skb_is_mmaped(skb)) {
1584 DECLARE_WAITQUEUE(wait, current);
1586 if (!ssk || netlink_is_kernel(ssk))
1587 netlink_overrun(sk);
1593 __set_current_state(TASK_INTERRUPTIBLE);
1594 add_wait_queue(&nlk->wait, &wait);
1596 if ((atomic_read(&sk->sk_rmem_alloc) > sk->sk_rcvbuf ||
1597 test_bit(NETLINK_CONGESTED, &nlk->state)) &&
1598 !sock_flag(sk, SOCK_DEAD))
1599 *timeo = schedule_timeout(*timeo);
1601 __set_current_state(TASK_RUNNING);
1602 remove_wait_queue(&nlk->wait, &wait);
1605 if (signal_pending(current)) {
1607 return sock_intr_errno(*timeo);
1611 netlink_skb_set_owner_r(skb, sk);
1615 static int __netlink_sendskb(struct sock *sk, struct sk_buff *skb)
1619 netlink_deliver_tap(skb);
1621 #ifdef CONFIG_NETLINK_MMAP
1622 if (netlink_skb_is_mmaped(skb))
1623 netlink_queue_mmaped_skb(sk, skb);
1624 else if (netlink_rx_is_mmaped(sk))
1625 netlink_ring_set_copied(sk, skb);
1627 #endif /* CONFIG_NETLINK_MMAP */
1628 skb_queue_tail(&sk->sk_receive_queue, skb);
1629 sk->sk_data_ready(sk, len);
1633 int netlink_sendskb(struct sock *sk, struct sk_buff *skb)
1635 int len = __netlink_sendskb(sk, skb);
1641 void netlink_detachskb(struct sock *sk, struct sk_buff *skb)
1647 static struct sk_buff *netlink_trim(struct sk_buff *skb, gfp_t allocation)
1651 WARN_ON(skb->sk != NULL);
1652 if (netlink_skb_is_mmaped(skb))
1655 delta = skb->end - skb->tail;
1656 if (is_vmalloc_addr(skb->head) || delta * 2 < skb->truesize)
1659 if (skb_shared(skb)) {
1660 struct sk_buff *nskb = skb_clone(skb, allocation);
1667 if (!pskb_expand_head(skb, 0, -delta, allocation))
1668 skb->truesize -= delta;
1673 static int netlink_unicast_kernel(struct sock *sk, struct sk_buff *skb,
1677 struct netlink_sock *nlk = nlk_sk(sk);
1679 ret = -ECONNREFUSED;
1680 if (nlk->netlink_rcv != NULL) {
1681 /* We could do a netlink_deliver_tap(skb) here as well
1682 * but since this is intended for the kernel only, we
1683 * should rather let it stay under the hood.
1687 netlink_skb_set_owner_r(skb, sk);
1688 NETLINK_CB(skb).sk = ssk;
1689 nlk->netlink_rcv(skb);
1698 int netlink_unicast(struct sock *ssk, struct sk_buff *skb,
1699 u32 portid, int nonblock)
1705 skb = netlink_trim(skb, gfp_any());
1707 timeo = sock_sndtimeo(ssk, nonblock);
1709 sk = netlink_getsockbyportid(ssk, portid);
1714 if (netlink_is_kernel(sk))
1715 return netlink_unicast_kernel(sk, skb, ssk);
1717 if (sk_filter(sk, skb)) {
1724 err = netlink_attachskb(sk, skb, &timeo, ssk);
1730 return netlink_sendskb(sk, skb);
1732 EXPORT_SYMBOL(netlink_unicast);
1734 struct sk_buff *netlink_alloc_skb(struct sock *ssk, unsigned int size,
1735 u32 dst_portid, gfp_t gfp_mask)
1737 #ifdef CONFIG_NETLINK_MMAP
1738 struct sock *sk = NULL;
1739 struct sk_buff *skb;
1740 struct netlink_ring *ring;
1741 struct nl_mmap_hdr *hdr;
1742 unsigned int maxlen;
1744 sk = netlink_getsockbyportid(ssk, dst_portid);
1748 ring = &nlk_sk(sk)->rx_ring;
1749 /* fast-path without atomic ops for common case: non-mmaped receiver */
1750 if (ring->pg_vec == NULL)
1753 skb = alloc_skb_head(gfp_mask);
1757 spin_lock_bh(&sk->sk_receive_queue.lock);
1758 /* check again under lock */
1759 if (ring->pg_vec == NULL)
1762 maxlen = ring->frame_size - NL_MMAP_HDRLEN;
1766 netlink_forward_ring(ring);
1767 hdr = netlink_current_frame(ring, NL_MMAP_STATUS_UNUSED);
1770 netlink_ring_setup_skb(skb, sk, ring, hdr);
1771 netlink_set_status(hdr, NL_MMAP_STATUS_RESERVED);
1772 atomic_inc(&ring->pending);
1773 netlink_increment_head(ring);
1775 spin_unlock_bh(&sk->sk_receive_queue.lock);
1780 spin_unlock_bh(&sk->sk_receive_queue.lock);
1781 netlink_overrun(sk);
1788 spin_unlock_bh(&sk->sk_receive_queue.lock);
1793 return alloc_skb(size, gfp_mask);
1795 EXPORT_SYMBOL_GPL(netlink_alloc_skb);
1797 int netlink_has_listeners(struct sock *sk, unsigned int group)
1800 struct listeners *listeners;
1802 BUG_ON(!netlink_is_kernel(sk));
1805 listeners = rcu_dereference(nl_table[sk->sk_protocol].listeners);
1807 if (listeners && group - 1 < nl_table[sk->sk_protocol].groups)
1808 res = test_bit(group - 1, listeners->masks);
1814 EXPORT_SYMBOL_GPL(netlink_has_listeners);
1816 static int netlink_broadcast_deliver(struct sock *sk, struct sk_buff *skb)
1818 struct netlink_sock *nlk = nlk_sk(sk);
1820 if (atomic_read(&sk->sk_rmem_alloc) <= sk->sk_rcvbuf &&
1821 !test_bit(NETLINK_CONGESTED, &nlk->state)) {
1822 netlink_skb_set_owner_r(skb, sk);
1823 __netlink_sendskb(sk, skb);
1824 return atomic_read(&sk->sk_rmem_alloc) > (sk->sk_rcvbuf >> 1);
1829 struct netlink_broadcast_data {
1830 struct sock *exclude_sk;
1835 int delivery_failure;
1839 struct sk_buff *skb, *skb2;
1840 int (*tx_filter)(struct sock *dsk, struct sk_buff *skb, void *data);
1844 static int do_one_broadcast(struct sock *sk,
1845 struct netlink_broadcast_data *p)
1847 struct netlink_sock *nlk = nlk_sk(sk);
1850 if (p->exclude_sk == sk)
1853 if (nlk->portid == p->portid || p->group - 1 >= nlk->ngroups ||
1854 !test_bit(p->group - 1, nlk->groups))
1857 if (!net_eq(sock_net(sk), p->net))
1861 netlink_overrun(sk);
1866 if (p->skb2 == NULL) {
1867 if (skb_shared(p->skb)) {
1868 p->skb2 = skb_clone(p->skb, p->allocation);
1870 p->skb2 = skb_get(p->skb);
1872 * skb ownership may have been set when
1873 * delivered to a previous socket.
1875 skb_orphan(p->skb2);
1878 if (p->skb2 == NULL) {
1879 netlink_overrun(sk);
1880 /* Clone failed. Notify ALL listeners. */
1882 if (nlk->flags & NETLINK_BROADCAST_SEND_ERROR)
1883 p->delivery_failure = 1;
1884 } else if (p->tx_filter && p->tx_filter(sk, p->skb2, p->tx_data)) {
1887 } else if (sk_filter(sk, p->skb2)) {
1890 } else if ((val = netlink_broadcast_deliver(sk, p->skb2)) < 0) {
1891 netlink_overrun(sk);
1892 if (nlk->flags & NETLINK_BROADCAST_SEND_ERROR)
1893 p->delivery_failure = 1;
1895 p->congested |= val;
1905 int netlink_broadcast_filtered(struct sock *ssk, struct sk_buff *skb, u32 portid,
1906 u32 group, gfp_t allocation,
1907 int (*filter)(struct sock *dsk, struct sk_buff *skb, void *data),
1910 struct net *net = sock_net(ssk);
1911 struct netlink_broadcast_data info;
1914 skb = netlink_trim(skb, allocation);
1916 info.exclude_sk = ssk;
1918 info.portid = portid;
1921 info.delivery_failure = 0;
1924 info.allocation = allocation;
1927 info.tx_filter = filter;
1928 info.tx_data = filter_data;
1930 /* While we sleep in clone, do not allow to change socket list */
1932 netlink_lock_table();
1934 sk_for_each_bound(sk, &nl_table[ssk->sk_protocol].mc_list)
1935 do_one_broadcast(sk, &info);
1939 netlink_unlock_table();
1941 if (info.delivery_failure) {
1942 kfree_skb(info.skb2);
1945 consume_skb(info.skb2);
1947 if (info.delivered) {
1948 if (info.congested && (allocation & __GFP_WAIT))
1954 EXPORT_SYMBOL(netlink_broadcast_filtered);
1956 int netlink_broadcast(struct sock *ssk, struct sk_buff *skb, u32 portid,
1957 u32 group, gfp_t allocation)
1959 return netlink_broadcast_filtered(ssk, skb, portid, group, allocation,
1962 EXPORT_SYMBOL(netlink_broadcast);
1964 struct netlink_set_err_data {
1965 struct sock *exclude_sk;
1971 static int do_one_set_err(struct sock *sk, struct netlink_set_err_data *p)
1973 struct netlink_sock *nlk = nlk_sk(sk);
1976 if (sk == p->exclude_sk)
1979 if (!net_eq(sock_net(sk), sock_net(p->exclude_sk)))
1982 if (nlk->portid == p->portid || p->group - 1 >= nlk->ngroups ||
1983 !test_bit(p->group - 1, nlk->groups))
1986 if (p->code == ENOBUFS && nlk->flags & NETLINK_RECV_NO_ENOBUFS) {
1991 sk->sk_err = p->code;
1992 sk->sk_error_report(sk);
1998 * netlink_set_err - report error to broadcast listeners
1999 * @ssk: the kernel netlink socket, as returned by netlink_kernel_create()
2000 * @portid: the PORTID of a process that we want to skip (if any)
2001 * @groups: the broadcast group that will notice the error
2002 * @code: error code, must be negative (as usual in kernelspace)
2004 * This function returns the number of broadcast listeners that have set the
2005 * NETLINK_RECV_NO_ENOBUFS socket option.
2007 int netlink_set_err(struct sock *ssk, u32 portid, u32 group, int code)
2009 struct netlink_set_err_data info;
2013 info.exclude_sk = ssk;
2014 info.portid = portid;
2016 /* sk->sk_err wants a positive error value */
2019 read_lock(&nl_table_lock);
2021 sk_for_each_bound(sk, &nl_table[ssk->sk_protocol].mc_list)
2022 ret += do_one_set_err(sk, &info);
2024 read_unlock(&nl_table_lock);
2027 EXPORT_SYMBOL(netlink_set_err);
2029 /* must be called with netlink table grabbed */
2030 static void netlink_update_socket_mc(struct netlink_sock *nlk,
2034 int old, new = !!is_new, subscriptions;
2036 old = test_bit(group - 1, nlk->groups);
2037 subscriptions = nlk->subscriptions - old + new;
2039 __set_bit(group - 1, nlk->groups);
2041 __clear_bit(group - 1, nlk->groups);
2042 netlink_update_subscriptions(&nlk->sk, subscriptions);
2043 netlink_update_listeners(&nlk->sk);
2046 static int netlink_setsockopt(struct socket *sock, int level, int optname,
2047 char __user *optval, unsigned int optlen)
2049 struct sock *sk = sock->sk;
2050 struct netlink_sock *nlk = nlk_sk(sk);
2051 unsigned int val = 0;
2054 if (level != SOL_NETLINK)
2055 return -ENOPROTOOPT;
2057 if (optname != NETLINK_RX_RING && optname != NETLINK_TX_RING &&
2058 optlen >= sizeof(int) &&
2059 get_user(val, (unsigned int __user *)optval))
2063 case NETLINK_PKTINFO:
2065 nlk->flags |= NETLINK_RECV_PKTINFO;
2067 nlk->flags &= ~NETLINK_RECV_PKTINFO;
2070 case NETLINK_ADD_MEMBERSHIP:
2071 case NETLINK_DROP_MEMBERSHIP: {
2072 if (!netlink_capable(sock, NL_CFG_F_NONROOT_RECV))
2074 err = netlink_realloc_groups(sk);
2077 if (!val || val - 1 >= nlk->ngroups)
2079 netlink_table_grab();
2080 netlink_update_socket_mc(nlk, val,
2081 optname == NETLINK_ADD_MEMBERSHIP);
2082 netlink_table_ungrab();
2084 if (nlk->netlink_bind)
2085 nlk->netlink_bind(val);
2090 case NETLINK_BROADCAST_ERROR:
2092 nlk->flags |= NETLINK_BROADCAST_SEND_ERROR;
2094 nlk->flags &= ~NETLINK_BROADCAST_SEND_ERROR;
2097 case NETLINK_NO_ENOBUFS:
2099 nlk->flags |= NETLINK_RECV_NO_ENOBUFS;
2100 clear_bit(NETLINK_CONGESTED, &nlk->state);
2101 wake_up_interruptible(&nlk->wait);
2103 nlk->flags &= ~NETLINK_RECV_NO_ENOBUFS;
2107 #ifdef CONFIG_NETLINK_MMAP
2108 case NETLINK_RX_RING:
2109 case NETLINK_TX_RING: {
2110 struct nl_mmap_req req;
2112 /* Rings might consume more memory than queue limits, require
2115 if (!capable(CAP_NET_ADMIN))
2117 if (optlen < sizeof(req))
2119 if (copy_from_user(&req, optval, sizeof(req)))
2121 err = netlink_set_ring(sk, &req, false,
2122 optname == NETLINK_TX_RING);
2125 #endif /* CONFIG_NETLINK_MMAP */
2132 static int netlink_getsockopt(struct socket *sock, int level, int optname,
2133 char __user *optval, int __user *optlen)
2135 struct sock *sk = sock->sk;
2136 struct netlink_sock *nlk = nlk_sk(sk);
2139 if (level != SOL_NETLINK)
2140 return -ENOPROTOOPT;
2142 if (get_user(len, optlen))
2148 case NETLINK_PKTINFO:
2149 if (len < sizeof(int))
2152 val = nlk->flags & NETLINK_RECV_PKTINFO ? 1 : 0;
2153 if (put_user(len, optlen) ||
2154 put_user(val, optval))
2158 case NETLINK_BROADCAST_ERROR:
2159 if (len < sizeof(int))
2162 val = nlk->flags & NETLINK_BROADCAST_SEND_ERROR ? 1 : 0;
2163 if (put_user(len, optlen) ||
2164 put_user(val, optval))
2168 case NETLINK_NO_ENOBUFS:
2169 if (len < sizeof(int))
2172 val = nlk->flags & NETLINK_RECV_NO_ENOBUFS ? 1 : 0;
2173 if (put_user(len, optlen) ||
2174 put_user(val, optval))
2184 static void netlink_cmsg_recv_pktinfo(struct msghdr *msg, struct sk_buff *skb)
2186 struct nl_pktinfo info;
2188 info.group = NETLINK_CB(skb).dst_group;
2189 put_cmsg(msg, SOL_NETLINK, NETLINK_PKTINFO, sizeof(info), &info);
2192 static int netlink_sendmsg(struct kiocb *kiocb, struct socket *sock,
2193 struct msghdr *msg, size_t len)
2195 struct sock_iocb *siocb = kiocb_to_siocb(kiocb);
2196 struct sock *sk = sock->sk;
2197 struct netlink_sock *nlk = nlk_sk(sk);
2198 struct sockaddr_nl *addr = msg->msg_name;
2201 struct sk_buff *skb;
2203 struct scm_cookie scm;
2205 if (msg->msg_flags&MSG_OOB)
2208 if (NULL == siocb->scm)
2211 err = scm_send(sock, msg, siocb->scm, true);
2215 if (msg->msg_namelen) {
2217 if (addr->nl_family != AF_NETLINK)
2219 dst_portid = addr->nl_pid;
2220 dst_group = ffs(addr->nl_groups);
2222 if ((dst_group || dst_portid) &&
2223 !netlink_capable(sock, NL_CFG_F_NONROOT_SEND))
2226 dst_portid = nlk->dst_portid;
2227 dst_group = nlk->dst_group;
2231 err = netlink_autobind(sock);
2236 if (netlink_tx_is_mmaped(sk) &&
2237 msg->msg_iov->iov_base == NULL) {
2238 err = netlink_mmap_sendmsg(sk, msg, dst_portid, dst_group,
2244 if (len > sk->sk_sndbuf - 32)
2247 skb = netlink_alloc_large_skb(len);
2251 NETLINK_CB(skb).portid = nlk->portid;
2252 NETLINK_CB(skb).dst_group = dst_group;
2253 NETLINK_CB(skb).creds = siocb->scm->creds;
2256 if (memcpy_fromiovec(skb_put(skb, len), msg->msg_iov, len)) {
2261 err = security_netlink_send(sk, skb);
2268 atomic_inc(&skb->users);
2269 netlink_broadcast(sk, skb, dst_portid, dst_group, GFP_KERNEL);
2271 err = netlink_unicast(sk, skb, dst_portid, msg->msg_flags&MSG_DONTWAIT);
2274 scm_destroy(siocb->scm);
2278 static int netlink_recvmsg(struct kiocb *kiocb, struct socket *sock,
2279 struct msghdr *msg, size_t len,
2282 struct sock_iocb *siocb = kiocb_to_siocb(kiocb);
2283 struct scm_cookie scm;
2284 struct sock *sk = sock->sk;
2285 struct netlink_sock *nlk = nlk_sk(sk);
2286 int noblock = flags&MSG_DONTWAIT;
2288 struct sk_buff *skb, *data_skb;
2296 skb = skb_recv_datagram(sk, flags, noblock, &err);
2302 #ifdef CONFIG_COMPAT_NETLINK_MESSAGES
2303 if (unlikely(skb_shinfo(skb)->frag_list)) {
2305 * If this skb has a frag_list, then here that means that we
2306 * will have to use the frag_list skb's data for compat tasks
2307 * and the regular skb's data for normal (non-compat) tasks.
2309 * If we need to send the compat skb, assign it to the
2310 * 'data_skb' variable so that it will be used below for data
2311 * copying. We keep 'skb' for everything else, including
2312 * freeing both later.
2314 if (flags & MSG_CMSG_COMPAT)
2315 data_skb = skb_shinfo(skb)->frag_list;
2319 msg->msg_namelen = 0;
2321 copied = data_skb->len;
2323 msg->msg_flags |= MSG_TRUNC;
2327 skb_reset_transport_header(data_skb);
2328 err = skb_copy_datagram_iovec(data_skb, 0, msg->msg_iov, copied);
2330 if (msg->msg_name) {
2331 struct sockaddr_nl *addr = (struct sockaddr_nl *)msg->msg_name;
2332 addr->nl_family = AF_NETLINK;
2334 addr->nl_pid = NETLINK_CB(skb).portid;
2335 addr->nl_groups = netlink_group_mask(NETLINK_CB(skb).dst_group);
2336 msg->msg_namelen = sizeof(*addr);
2339 if (nlk->flags & NETLINK_RECV_PKTINFO)
2340 netlink_cmsg_recv_pktinfo(msg, skb);
2342 if (NULL == siocb->scm) {
2343 memset(&scm, 0, sizeof(scm));
2346 siocb->scm->creds = *NETLINK_CREDS(skb);
2347 if (flags & MSG_TRUNC)
2348 copied = data_skb->len;
2350 skb_free_datagram(sk, skb);
2352 if (nlk->cb && atomic_read(&sk->sk_rmem_alloc) <= sk->sk_rcvbuf / 2) {
2353 ret = netlink_dump(sk);
2356 sk->sk_error_report(sk);
2360 scm_recv(sock, msg, siocb->scm, flags);
2362 netlink_rcv_wake(sk);
2363 return err ? : copied;
2366 static void netlink_data_ready(struct sock *sk, int len)
2372 * We export these functions to other modules. They provide a
2373 * complete set of kernel non-blocking support for message
2378 __netlink_kernel_create(struct net *net, int unit, struct module *module,
2379 struct netlink_kernel_cfg *cfg)
2381 struct socket *sock;
2383 struct netlink_sock *nlk;
2384 struct listeners *listeners = NULL;
2385 struct mutex *cb_mutex = cfg ? cfg->cb_mutex : NULL;
2386 unsigned int groups;
2390 if (unit < 0 || unit >= MAX_LINKS)
2393 if (sock_create_lite(PF_NETLINK, SOCK_DGRAM, unit, &sock))
2397 * We have to just have a reference on the net from sk, but don't
2398 * get_net it. Besides, we cannot get and then put the net here.
2399 * So we create one inside init_net and the move it to net.
2402 if (__netlink_create(&init_net, sock, cb_mutex, unit) < 0)
2403 goto out_sock_release_nosk;
2406 sk_change_net(sk, net);
2408 if (!cfg || cfg->groups < 32)
2411 groups = cfg->groups;
2413 listeners = kzalloc(sizeof(*listeners) + NLGRPSZ(groups), GFP_KERNEL);
2415 goto out_sock_release;
2417 sk->sk_data_ready = netlink_data_ready;
2418 if (cfg && cfg->input)
2419 nlk_sk(sk)->netlink_rcv = cfg->input;
2421 if (netlink_insert(sk, net, 0))
2422 goto out_sock_release;
2425 nlk->flags |= NETLINK_KERNEL_SOCKET;
2427 netlink_table_grab();
2428 if (!nl_table[unit].registered) {
2429 nl_table[unit].groups = groups;
2430 rcu_assign_pointer(nl_table[unit].listeners, listeners);
2431 nl_table[unit].cb_mutex = cb_mutex;
2432 nl_table[unit].module = module;
2434 nl_table[unit].bind = cfg->bind;
2435 nl_table[unit].flags = cfg->flags;
2437 nl_table[unit].compare = cfg->compare;
2439 nl_table[unit].registered = 1;
2442 nl_table[unit].registered++;
2444 netlink_table_ungrab();
2449 netlink_kernel_release(sk);
2452 out_sock_release_nosk:
2456 EXPORT_SYMBOL(__netlink_kernel_create);
2459 netlink_kernel_release(struct sock *sk)
2461 sk_release_kernel(sk);
2463 EXPORT_SYMBOL(netlink_kernel_release);
2465 int __netlink_change_ngroups(struct sock *sk, unsigned int groups)
2467 struct listeners *new, *old;
2468 struct netlink_table *tbl = &nl_table[sk->sk_protocol];
2473 if (NLGRPSZ(tbl->groups) < NLGRPSZ(groups)) {
2474 new = kzalloc(sizeof(*new) + NLGRPSZ(groups), GFP_ATOMIC);
2477 old = nl_deref_protected(tbl->listeners);
2478 memcpy(new->masks, old->masks, NLGRPSZ(tbl->groups));
2479 rcu_assign_pointer(tbl->listeners, new);
2481 kfree_rcu(old, rcu);
2483 tbl->groups = groups;
2489 * netlink_change_ngroups - change number of multicast groups
2491 * This changes the number of multicast groups that are available
2492 * on a certain netlink family. Note that it is not possible to
2493 * change the number of groups to below 32. Also note that it does
2494 * not implicitly call netlink_clear_multicast_users() when the
2495 * number of groups is reduced.
2497 * @sk: The kernel netlink socket, as returned by netlink_kernel_create().
2498 * @groups: The new number of groups.
2500 int netlink_change_ngroups(struct sock *sk, unsigned int groups)
2504 netlink_table_grab();
2505 err = __netlink_change_ngroups(sk, groups);
2506 netlink_table_ungrab();
2511 void __netlink_clear_multicast_users(struct sock *ksk, unsigned int group)
2514 struct netlink_table *tbl = &nl_table[ksk->sk_protocol];
2516 sk_for_each_bound(sk, &tbl->mc_list)
2517 netlink_update_socket_mc(nlk_sk(sk), group, 0);
2521 * netlink_clear_multicast_users - kick off multicast listeners
2523 * This function removes all listeners from the given group.
2524 * @ksk: The kernel netlink socket, as returned by
2525 * netlink_kernel_create().
2526 * @group: The multicast group to clear.
2528 void netlink_clear_multicast_users(struct sock *ksk, unsigned int group)
2530 netlink_table_grab();
2531 __netlink_clear_multicast_users(ksk, group);
2532 netlink_table_ungrab();
2536 __nlmsg_put(struct sk_buff *skb, u32 portid, u32 seq, int type, int len, int flags)
2538 struct nlmsghdr *nlh;
2539 int size = nlmsg_msg_size(len);
2541 nlh = (struct nlmsghdr*)skb_put(skb, NLMSG_ALIGN(size));
2542 nlh->nlmsg_type = type;
2543 nlh->nlmsg_len = size;
2544 nlh->nlmsg_flags = flags;
2545 nlh->nlmsg_pid = portid;
2546 nlh->nlmsg_seq = seq;
2547 if (!__builtin_constant_p(size) || NLMSG_ALIGN(size) - size != 0)
2548 memset(nlmsg_data(nlh) + len, 0, NLMSG_ALIGN(size) - size);
2551 EXPORT_SYMBOL(__nlmsg_put);
2554 * It looks a bit ugly.
2555 * It would be better to create kernel thread.
2558 static int netlink_dump(struct sock *sk)
2560 struct netlink_sock *nlk = nlk_sk(sk);
2561 struct netlink_callback *cb;
2562 struct sk_buff *skb = NULL;
2563 struct nlmsghdr *nlh;
2564 int len, err = -ENOBUFS;
2567 mutex_lock(nlk->cb_mutex);
2575 alloc_size = max_t(int, cb->min_dump_alloc, NLMSG_GOODSIZE);
2577 if (!netlink_rx_is_mmaped(sk) &&
2578 atomic_read(&sk->sk_rmem_alloc) >= sk->sk_rcvbuf)
2580 skb = netlink_alloc_skb(sk, alloc_size, nlk->portid, GFP_KERNEL);
2583 netlink_skb_set_owner_r(skb, sk);
2585 len = cb->dump(skb, cb);
2588 mutex_unlock(nlk->cb_mutex);
2590 if (sk_filter(sk, skb))
2593 __netlink_sendskb(sk, skb);
2597 nlh = nlmsg_put_answer(skb, cb, NLMSG_DONE, sizeof(len), NLM_F_MULTI);
2601 nl_dump_check_consistent(cb, nlh);
2603 memcpy(nlmsg_data(nlh), &len, sizeof(len));
2605 if (sk_filter(sk, skb))
2608 __netlink_sendskb(sk, skb);
2613 mutex_unlock(nlk->cb_mutex);
2615 module_put(cb->module);
2616 netlink_consume_callback(cb);
2620 mutex_unlock(nlk->cb_mutex);
2625 int __netlink_dump_start(struct sock *ssk, struct sk_buff *skb,
2626 const struct nlmsghdr *nlh,
2627 struct netlink_dump_control *control)
2629 struct netlink_callback *cb;
2631 struct netlink_sock *nlk;
2634 cb = kzalloc(sizeof(*cb), GFP_KERNEL);
2638 /* Memory mapped dump requests need to be copied to avoid looping
2639 * on the pending state in netlink_mmap_sendmsg() while the CB hold
2640 * a reference to the skb.
2642 if (netlink_skb_is_mmaped(skb)) {
2643 skb = skb_copy(skb, GFP_KERNEL);
2649 atomic_inc(&skb->users);
2651 cb->dump = control->dump;
2652 cb->done = control->done;
2654 cb->data = control->data;
2655 cb->module = control->module;
2656 cb->min_dump_alloc = control->min_dump_alloc;
2659 sk = netlink_lookup(sock_net(ssk), ssk->sk_protocol, NETLINK_CB(skb).portid);
2661 netlink_destroy_callback(cb);
2662 return -ECONNREFUSED;
2666 mutex_lock(nlk->cb_mutex);
2667 /* A dump is in progress... */
2669 mutex_unlock(nlk->cb_mutex);
2670 netlink_destroy_callback(cb);
2674 /* add reference of module which cb->dump belongs to */
2675 if (!try_module_get(cb->module)) {
2676 mutex_unlock(nlk->cb_mutex);
2677 netlink_destroy_callback(cb);
2678 ret = -EPROTONOSUPPORT;
2683 mutex_unlock(nlk->cb_mutex);
2685 ret = netlink_dump(sk);
2692 /* We successfully started a dump, by returning -EINTR we
2693 * signal not to send ACK even if it was requested.
2697 EXPORT_SYMBOL(__netlink_dump_start);
2699 void netlink_ack(struct sk_buff *in_skb, struct nlmsghdr *nlh, int err)
2701 struct sk_buff *skb;
2702 struct nlmsghdr *rep;
2703 struct nlmsgerr *errmsg;
2704 size_t payload = sizeof(*errmsg);
2706 /* error messages get the original request appened */
2708 payload += nlmsg_len(nlh);
2710 skb = netlink_alloc_skb(in_skb->sk, nlmsg_total_size(payload),
2711 NETLINK_CB(in_skb).portid, GFP_KERNEL);
2715 sk = netlink_lookup(sock_net(in_skb->sk),
2716 in_skb->sk->sk_protocol,
2717 NETLINK_CB(in_skb).portid);
2719 sk->sk_err = ENOBUFS;
2720 sk->sk_error_report(sk);
2726 rep = __nlmsg_put(skb, NETLINK_CB(in_skb).portid, nlh->nlmsg_seq,
2727 NLMSG_ERROR, payload, 0);
2728 errmsg = nlmsg_data(rep);
2729 errmsg->error = err;
2730 memcpy(&errmsg->msg, nlh, err ? nlh->nlmsg_len : sizeof(*nlh));
2731 netlink_unicast(in_skb->sk, skb, NETLINK_CB(in_skb).portid, MSG_DONTWAIT);
2733 EXPORT_SYMBOL(netlink_ack);
2735 int netlink_rcv_skb(struct sk_buff *skb, int (*cb)(struct sk_buff *,
2738 struct nlmsghdr *nlh;
2741 while (skb->len >= nlmsg_total_size(0)) {
2744 nlh = nlmsg_hdr(skb);
2747 if (nlh->nlmsg_len < NLMSG_HDRLEN || skb->len < nlh->nlmsg_len)
2750 /* Only requests are handled by the kernel */
2751 if (!(nlh->nlmsg_flags & NLM_F_REQUEST))
2754 /* Skip control messages */
2755 if (nlh->nlmsg_type < NLMSG_MIN_TYPE)
2763 if (nlh->nlmsg_flags & NLM_F_ACK || err)
2764 netlink_ack(skb, nlh, err);
2767 msglen = NLMSG_ALIGN(nlh->nlmsg_len);
2768 if (msglen > skb->len)
2770 skb_pull(skb, msglen);
2775 EXPORT_SYMBOL(netlink_rcv_skb);
2778 * nlmsg_notify - send a notification netlink message
2779 * @sk: netlink socket to use
2780 * @skb: notification message
2781 * @portid: destination netlink portid for reports or 0
2782 * @group: destination multicast group or 0
2783 * @report: 1 to report back, 0 to disable
2784 * @flags: allocation flags
2786 int nlmsg_notify(struct sock *sk, struct sk_buff *skb, u32 portid,
2787 unsigned int group, int report, gfp_t flags)
2792 int exclude_portid = 0;
2795 atomic_inc(&skb->users);
2796 exclude_portid = portid;
2799 /* errors reported via destination sk->sk_err, but propagate
2800 * delivery errors if NETLINK_BROADCAST_ERROR flag is set */
2801 err = nlmsg_multicast(sk, skb, exclude_portid, group, flags);
2807 err2 = nlmsg_unicast(sk, skb, portid);
2808 if (!err || err == -ESRCH)
2814 EXPORT_SYMBOL(nlmsg_notify);
2816 #ifdef CONFIG_PROC_FS
2817 struct nl_seq_iter {
2818 struct seq_net_private p;
2823 static struct sock *netlink_seq_socket_idx(struct seq_file *seq, loff_t pos)
2825 struct nl_seq_iter *iter = seq->private;
2830 for (i = 0; i < MAX_LINKS; i++) {
2831 struct nl_portid_hash *hash = &nl_table[i].hash;
2833 for (j = 0; j <= hash->mask; j++) {
2834 sk_for_each(s, &hash->table[j]) {
2835 if (sock_net(s) != seq_file_net(seq))
2849 static void *netlink_seq_start(struct seq_file *seq, loff_t *pos)
2850 __acquires(nl_table_lock)
2852 read_lock(&nl_table_lock);
2853 return *pos ? netlink_seq_socket_idx(seq, *pos - 1) : SEQ_START_TOKEN;
2856 static void *netlink_seq_next(struct seq_file *seq, void *v, loff_t *pos)
2859 struct nl_seq_iter *iter;
2865 if (v == SEQ_START_TOKEN)
2866 return netlink_seq_socket_idx(seq, 0);
2868 net = seq_file_net(seq);
2869 iter = seq->private;
2873 } while (s && !nl_table[s->sk_protocol].compare(net, s));
2878 j = iter->hash_idx + 1;
2881 struct nl_portid_hash *hash = &nl_table[i].hash;
2883 for (; j <= hash->mask; j++) {
2884 s = sk_head(&hash->table[j]);
2886 while (s && !nl_table[s->sk_protocol].compare(net, s))
2896 } while (++i < MAX_LINKS);
2901 static void netlink_seq_stop(struct seq_file *seq, void *v)
2902 __releases(nl_table_lock)
2904 read_unlock(&nl_table_lock);
2908 static int netlink_seq_show(struct seq_file *seq, void *v)
2910 if (v == SEQ_START_TOKEN) {
2912 "sk Eth Pid Groups "
2913 "Rmem Wmem Dump Locks Drops Inode\n");
2916 struct netlink_sock *nlk = nlk_sk(s);
2918 seq_printf(seq, "%pK %-3d %-6u %08x %-8d %-8d %pK %-8d %-8d %-8lu\n",
2922 nlk->groups ? (u32)nlk->groups[0] : 0,
2923 sk_rmem_alloc_get(s),
2924 sk_wmem_alloc_get(s),
2926 atomic_read(&s->sk_refcnt),
2927 atomic_read(&s->sk_drops),
2935 static const struct seq_operations netlink_seq_ops = {
2936 .start = netlink_seq_start,
2937 .next = netlink_seq_next,
2938 .stop = netlink_seq_stop,
2939 .show = netlink_seq_show,
2943 static int netlink_seq_open(struct inode *inode, struct file *file)
2945 return seq_open_net(inode, file, &netlink_seq_ops,
2946 sizeof(struct nl_seq_iter));
2949 static const struct file_operations netlink_seq_fops = {
2950 .owner = THIS_MODULE,
2951 .open = netlink_seq_open,
2953 .llseek = seq_lseek,
2954 .release = seq_release_net,
2959 int netlink_register_notifier(struct notifier_block *nb)
2961 return atomic_notifier_chain_register(&netlink_chain, nb);
2963 EXPORT_SYMBOL(netlink_register_notifier);
2965 int netlink_unregister_notifier(struct notifier_block *nb)
2967 return atomic_notifier_chain_unregister(&netlink_chain, nb);
2969 EXPORT_SYMBOL(netlink_unregister_notifier);
2971 static const struct proto_ops netlink_ops = {
2972 .family = PF_NETLINK,
2973 .owner = THIS_MODULE,
2974 .release = netlink_release,
2975 .bind = netlink_bind,
2976 .connect = netlink_connect,
2977 .socketpair = sock_no_socketpair,
2978 .accept = sock_no_accept,
2979 .getname = netlink_getname,
2980 .poll = netlink_poll,
2981 .ioctl = sock_no_ioctl,
2982 .listen = sock_no_listen,
2983 .shutdown = sock_no_shutdown,
2984 .setsockopt = netlink_setsockopt,
2985 .getsockopt = netlink_getsockopt,
2986 .sendmsg = netlink_sendmsg,
2987 .recvmsg = netlink_recvmsg,
2988 .mmap = netlink_mmap,
2989 .sendpage = sock_no_sendpage,
2992 static const struct net_proto_family netlink_family_ops = {
2993 .family = PF_NETLINK,
2994 .create = netlink_create,
2995 .owner = THIS_MODULE, /* for consistency 8) */
2998 static int __net_init netlink_net_init(struct net *net)
3000 #ifdef CONFIG_PROC_FS
3001 if (!proc_create("netlink", 0, net->proc_net, &netlink_seq_fops))
3007 static void __net_exit netlink_net_exit(struct net *net)
3009 #ifdef CONFIG_PROC_FS
3010 remove_proc_entry("netlink", net->proc_net);
3014 static void __init netlink_add_usersock_entry(void)
3016 struct listeners *listeners;
3019 listeners = kzalloc(sizeof(*listeners) + NLGRPSZ(groups), GFP_KERNEL);
3021 panic("netlink_add_usersock_entry: Cannot allocate listeners\n");
3023 netlink_table_grab();
3025 nl_table[NETLINK_USERSOCK].groups = groups;
3026 rcu_assign_pointer(nl_table[NETLINK_USERSOCK].listeners, listeners);
3027 nl_table[NETLINK_USERSOCK].module = THIS_MODULE;
3028 nl_table[NETLINK_USERSOCK].registered = 1;
3029 nl_table[NETLINK_USERSOCK].flags = NL_CFG_F_NONROOT_SEND;
3031 netlink_table_ungrab();
3034 static struct pernet_operations __net_initdata netlink_net_ops = {
3035 .init = netlink_net_init,
3036 .exit = netlink_net_exit,
3039 static int __init netlink_proto_init(void)
3042 unsigned long limit;
3044 int err = proto_register(&netlink_proto, 0);
3049 BUILD_BUG_ON(sizeof(struct netlink_skb_parms) > FIELD_SIZEOF(struct sk_buff, cb));
3051 nl_table = kcalloc(MAX_LINKS, sizeof(*nl_table), GFP_KERNEL);
3055 if (totalram_pages >= (128 * 1024))
3056 limit = totalram_pages >> (21 - PAGE_SHIFT);
3058 limit = totalram_pages >> (23 - PAGE_SHIFT);
3060 order = get_bitmask_order(limit) - 1 + PAGE_SHIFT;
3061 limit = (1UL << order) / sizeof(struct hlist_head);
3062 order = get_bitmask_order(min(limit, (unsigned long)UINT_MAX)) - 1;
3064 for (i = 0; i < MAX_LINKS; i++) {
3065 struct nl_portid_hash *hash = &nl_table[i].hash;
3067 hash->table = nl_portid_hash_zalloc(1 * sizeof(*hash->table));
3070 nl_portid_hash_free(nl_table[i].hash.table,
3071 1 * sizeof(*hash->table));
3075 hash->max_shift = order;
3078 hash->rehash_time = jiffies;
3080 nl_table[i].compare = netlink_compare;
3083 INIT_LIST_HEAD(&netlink_tap_all);
3085 netlink_add_usersock_entry();
3087 sock_register(&netlink_family_ops);
3088 register_pernet_subsys(&netlink_net_ops);
3089 /* The netlink device handler may be needed early. */
3094 panic("netlink_init: Cannot allocate nl_table\n");
3097 core_initcall(netlink_proto_init);