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 <asm/cacheflush.h>
62 #include <net/net_namespace.h>
65 #include <net/netlink.h>
67 #include "af_netlink.h"
71 unsigned long masks[0];
75 #define NETLINK_CONGESTED 0x0
78 #define NETLINK_KERNEL_SOCKET 0x1
79 #define NETLINK_RECV_PKTINFO 0x2
80 #define NETLINK_BROADCAST_SEND_ERROR 0x4
81 #define NETLINK_RECV_NO_ENOBUFS 0x8
83 static inline int netlink_is_kernel(struct sock *sk)
85 return nlk_sk(sk)->flags & NETLINK_KERNEL_SOCKET;
88 struct netlink_table *nl_table;
89 EXPORT_SYMBOL_GPL(nl_table);
91 static DECLARE_WAIT_QUEUE_HEAD(nl_table_wait);
93 static int netlink_dump(struct sock *sk);
94 static void netlink_skb_destructor(struct sk_buff *skb);
96 DEFINE_RWLOCK(nl_table_lock);
97 EXPORT_SYMBOL_GPL(nl_table_lock);
98 static atomic_t nl_table_users = ATOMIC_INIT(0);
100 #define nl_deref_protected(X) rcu_dereference_protected(X, lockdep_is_held(&nl_table_lock));
102 static ATOMIC_NOTIFIER_HEAD(netlink_chain);
104 static inline u32 netlink_group_mask(u32 group)
106 return group ? 1 << (group - 1) : 0;
109 static inline struct hlist_head *nl_portid_hashfn(struct nl_portid_hash *hash, u32 portid)
111 return &hash->table[jhash_1word(portid, hash->rnd) & hash->mask];
114 static void netlink_overrun(struct sock *sk)
116 struct netlink_sock *nlk = nlk_sk(sk);
118 if (!(nlk->flags & NETLINK_RECV_NO_ENOBUFS)) {
119 if (!test_and_set_bit(NETLINK_CONGESTED, &nlk_sk(sk)->state)) {
120 sk->sk_err = ENOBUFS;
121 sk->sk_error_report(sk);
124 atomic_inc(&sk->sk_drops);
127 static void netlink_rcv_wake(struct sock *sk)
129 struct netlink_sock *nlk = nlk_sk(sk);
131 if (skb_queue_empty(&sk->sk_receive_queue))
132 clear_bit(NETLINK_CONGESTED, &nlk->state);
133 if (!test_bit(NETLINK_CONGESTED, &nlk->state))
134 wake_up_interruptible(&nlk->wait);
137 #ifdef CONFIG_NETLINK_MMAP
138 static bool netlink_skb_is_mmaped(const struct sk_buff *skb)
140 return NETLINK_CB(skb).flags & NETLINK_SKB_MMAPED;
143 static bool netlink_rx_is_mmaped(struct sock *sk)
145 return nlk_sk(sk)->rx_ring.pg_vec != NULL;
148 static bool netlink_tx_is_mmaped(struct sock *sk)
150 return nlk_sk(sk)->tx_ring.pg_vec != NULL;
153 static __pure struct page *pgvec_to_page(const void *addr)
155 if (is_vmalloc_addr(addr))
156 return vmalloc_to_page(addr);
158 return virt_to_page(addr);
161 static void free_pg_vec(void **pg_vec, unsigned int order, unsigned int len)
165 for (i = 0; i < len; i++) {
166 if (pg_vec[i] != NULL) {
167 if (is_vmalloc_addr(pg_vec[i]))
170 free_pages((unsigned long)pg_vec[i], order);
176 static void *alloc_one_pg_vec_page(unsigned long order)
179 gfp_t gfp_flags = GFP_KERNEL | __GFP_COMP | __GFP_ZERO |
180 __GFP_NOWARN | __GFP_NORETRY;
182 buffer = (void *)__get_free_pages(gfp_flags, order);
186 buffer = vzalloc((1 << order) * PAGE_SIZE);
190 gfp_flags &= ~__GFP_NORETRY;
191 return (void *)__get_free_pages(gfp_flags, order);
194 static void **alloc_pg_vec(struct netlink_sock *nlk,
195 struct nl_mmap_req *req, unsigned int order)
197 unsigned int block_nr = req->nm_block_nr;
201 pg_vec = kcalloc(block_nr, sizeof(void *), GFP_KERNEL);
205 for (i = 0; i < block_nr; i++) {
206 pg_vec[i] = ptr = alloc_one_pg_vec_page(order);
207 if (pg_vec[i] == NULL)
213 free_pg_vec(pg_vec, order, block_nr);
217 static int netlink_set_ring(struct sock *sk, struct nl_mmap_req *req,
218 bool closing, bool tx_ring)
220 struct netlink_sock *nlk = nlk_sk(sk);
221 struct netlink_ring *ring;
222 struct sk_buff_head *queue;
223 void **pg_vec = NULL;
224 unsigned int order = 0;
227 ring = tx_ring ? &nlk->tx_ring : &nlk->rx_ring;
228 queue = tx_ring ? &sk->sk_write_queue : &sk->sk_receive_queue;
231 if (atomic_read(&nlk->mapped))
233 if (atomic_read(&ring->pending))
237 if (req->nm_block_nr) {
238 if (ring->pg_vec != NULL)
241 if ((int)req->nm_block_size <= 0)
243 if (!IS_ALIGNED(req->nm_block_size, PAGE_SIZE))
245 if (req->nm_frame_size < NL_MMAP_HDRLEN)
247 if (!IS_ALIGNED(req->nm_frame_size, NL_MMAP_MSG_ALIGNMENT))
250 ring->frames_per_block = req->nm_block_size /
252 if (ring->frames_per_block == 0)
254 if (ring->frames_per_block * req->nm_block_nr !=
258 order = get_order(req->nm_block_size);
259 pg_vec = alloc_pg_vec(nlk, req, order);
263 if (req->nm_frame_nr)
268 mutex_lock(&nlk->pg_vec_lock);
269 if (closing || atomic_read(&nlk->mapped) == 0) {
271 spin_lock_bh(&queue->lock);
273 ring->frame_max = req->nm_frame_nr - 1;
275 ring->frame_size = req->nm_frame_size;
276 ring->pg_vec_pages = req->nm_block_size / PAGE_SIZE;
278 swap(ring->pg_vec_len, req->nm_block_nr);
279 swap(ring->pg_vec_order, order);
280 swap(ring->pg_vec, pg_vec);
282 __skb_queue_purge(queue);
283 spin_unlock_bh(&queue->lock);
285 WARN_ON(atomic_read(&nlk->mapped));
287 mutex_unlock(&nlk->pg_vec_lock);
290 free_pg_vec(pg_vec, order, req->nm_block_nr);
294 static void netlink_mm_open(struct vm_area_struct *vma)
296 struct file *file = vma->vm_file;
297 struct socket *sock = file->private_data;
298 struct sock *sk = sock->sk;
301 atomic_inc(&nlk_sk(sk)->mapped);
304 static void netlink_mm_close(struct vm_area_struct *vma)
306 struct file *file = vma->vm_file;
307 struct socket *sock = file->private_data;
308 struct sock *sk = sock->sk;
311 atomic_dec(&nlk_sk(sk)->mapped);
314 static const struct vm_operations_struct netlink_mmap_ops = {
315 .open = netlink_mm_open,
316 .close = netlink_mm_close,
319 static int netlink_mmap(struct file *file, struct socket *sock,
320 struct vm_area_struct *vma)
322 struct sock *sk = sock->sk;
323 struct netlink_sock *nlk = nlk_sk(sk);
324 struct netlink_ring *ring;
325 unsigned long start, size, expected;
332 mutex_lock(&nlk->pg_vec_lock);
335 for (ring = &nlk->rx_ring; ring <= &nlk->tx_ring; ring++) {
336 if (ring->pg_vec == NULL)
338 expected += ring->pg_vec_len * ring->pg_vec_pages * PAGE_SIZE;
344 size = vma->vm_end - vma->vm_start;
345 if (size != expected)
348 start = vma->vm_start;
349 for (ring = &nlk->rx_ring; ring <= &nlk->tx_ring; ring++) {
350 if (ring->pg_vec == NULL)
353 for (i = 0; i < ring->pg_vec_len; i++) {
355 void *kaddr = ring->pg_vec[i];
358 for (pg_num = 0; pg_num < ring->pg_vec_pages; pg_num++) {
359 page = pgvec_to_page(kaddr);
360 err = vm_insert_page(vma, start, page);
369 atomic_inc(&nlk->mapped);
370 vma->vm_ops = &netlink_mmap_ops;
373 mutex_unlock(&nlk->pg_vec_lock);
377 static void netlink_frame_flush_dcache(const struct nl_mmap_hdr *hdr)
379 #if ARCH_IMPLEMENTS_FLUSH_DCACHE_PAGE == 1
380 struct page *p_start, *p_end;
382 /* First page is flushed through netlink_{get,set}_status */
383 p_start = pgvec_to_page(hdr + PAGE_SIZE);
384 p_end = pgvec_to_page((void *)hdr + NL_MMAP_HDRLEN + hdr->nm_len - 1);
385 while (p_start <= p_end) {
386 flush_dcache_page(p_start);
392 static enum nl_mmap_status netlink_get_status(const struct nl_mmap_hdr *hdr)
395 flush_dcache_page(pgvec_to_page(hdr));
396 return hdr->nm_status;
399 static void netlink_set_status(struct nl_mmap_hdr *hdr,
400 enum nl_mmap_status status)
402 hdr->nm_status = status;
403 flush_dcache_page(pgvec_to_page(hdr));
407 static struct nl_mmap_hdr *
408 __netlink_lookup_frame(const struct netlink_ring *ring, unsigned int pos)
410 unsigned int pg_vec_pos, frame_off;
412 pg_vec_pos = pos / ring->frames_per_block;
413 frame_off = pos % ring->frames_per_block;
415 return ring->pg_vec[pg_vec_pos] + (frame_off * ring->frame_size);
418 static struct nl_mmap_hdr *
419 netlink_lookup_frame(const struct netlink_ring *ring, unsigned int pos,
420 enum nl_mmap_status status)
422 struct nl_mmap_hdr *hdr;
424 hdr = __netlink_lookup_frame(ring, pos);
425 if (netlink_get_status(hdr) != status)
431 static struct nl_mmap_hdr *
432 netlink_current_frame(const struct netlink_ring *ring,
433 enum nl_mmap_status status)
435 return netlink_lookup_frame(ring, ring->head, status);
438 static struct nl_mmap_hdr *
439 netlink_previous_frame(const struct netlink_ring *ring,
440 enum nl_mmap_status status)
444 prev = ring->head ? ring->head - 1 : ring->frame_max;
445 return netlink_lookup_frame(ring, prev, status);
448 static void netlink_increment_head(struct netlink_ring *ring)
450 ring->head = ring->head != ring->frame_max ? ring->head + 1 : 0;
453 static void netlink_forward_ring(struct netlink_ring *ring)
455 unsigned int head = ring->head, pos = head;
456 const struct nl_mmap_hdr *hdr;
459 hdr = __netlink_lookup_frame(ring, pos);
460 if (hdr->nm_status == NL_MMAP_STATUS_UNUSED)
462 if (hdr->nm_status != NL_MMAP_STATUS_SKIP)
464 netlink_increment_head(ring);
465 } while (ring->head != head);
468 static bool netlink_dump_space(struct netlink_sock *nlk)
470 struct netlink_ring *ring = &nlk->rx_ring;
471 struct nl_mmap_hdr *hdr;
474 hdr = netlink_current_frame(ring, NL_MMAP_STATUS_UNUSED);
478 n = ring->head + ring->frame_max / 2;
479 if (n > ring->frame_max)
480 n -= ring->frame_max;
482 hdr = __netlink_lookup_frame(ring, n);
484 return hdr->nm_status == NL_MMAP_STATUS_UNUSED;
487 static unsigned int netlink_poll(struct file *file, struct socket *sock,
490 struct sock *sk = sock->sk;
491 struct netlink_sock *nlk = nlk_sk(sk);
495 if (nlk->rx_ring.pg_vec != NULL) {
496 /* Memory mapped sockets don't call recvmsg(), so flow control
497 * for dumps is performed here. A dump is allowed to continue
498 * if at least half the ring is unused.
500 while (nlk->cb != NULL && netlink_dump_space(nlk)) {
501 err = netlink_dump(sk);
504 sk->sk_error_report(sk);
508 netlink_rcv_wake(sk);
511 mask = datagram_poll(file, sock, wait);
513 spin_lock_bh(&sk->sk_receive_queue.lock);
514 if (nlk->rx_ring.pg_vec) {
515 netlink_forward_ring(&nlk->rx_ring);
516 if (!netlink_previous_frame(&nlk->rx_ring, NL_MMAP_STATUS_UNUSED))
517 mask |= POLLIN | POLLRDNORM;
519 spin_unlock_bh(&sk->sk_receive_queue.lock);
521 spin_lock_bh(&sk->sk_write_queue.lock);
522 if (nlk->tx_ring.pg_vec) {
523 if (netlink_current_frame(&nlk->tx_ring, NL_MMAP_STATUS_UNUSED))
524 mask |= POLLOUT | POLLWRNORM;
526 spin_unlock_bh(&sk->sk_write_queue.lock);
531 static struct nl_mmap_hdr *netlink_mmap_hdr(struct sk_buff *skb)
533 return (struct nl_mmap_hdr *)(skb->head - NL_MMAP_HDRLEN);
536 static void netlink_ring_setup_skb(struct sk_buff *skb, struct sock *sk,
537 struct netlink_ring *ring,
538 struct nl_mmap_hdr *hdr)
543 size = ring->frame_size - NL_MMAP_HDRLEN;
544 data = (void *)hdr + NL_MMAP_HDRLEN;
548 skb_reset_tail_pointer(skb);
549 skb->end = skb->tail + size;
552 skb->destructor = netlink_skb_destructor;
553 NETLINK_CB(skb).flags |= NETLINK_SKB_MMAPED;
554 NETLINK_CB(skb).sk = sk;
557 static int netlink_mmap_sendmsg(struct sock *sk, struct msghdr *msg,
558 u32 dst_portid, u32 dst_group,
559 struct sock_iocb *siocb)
561 struct netlink_sock *nlk = nlk_sk(sk);
562 struct netlink_ring *ring;
563 struct nl_mmap_hdr *hdr;
567 int err = 0, len = 0;
569 /* Netlink messages are validated by the receiver before processing.
570 * In order to avoid userspace changing the contents of the message
571 * after validation, the socket and the ring may only be used by a
572 * single process, otherwise we fall back to copying.
574 if (atomic_long_read(&sk->sk_socket->file->f_count) > 2 ||
575 atomic_read(&nlk->mapped) > 1)
578 mutex_lock(&nlk->pg_vec_lock);
580 ring = &nlk->tx_ring;
581 maxlen = ring->frame_size - NL_MMAP_HDRLEN;
584 hdr = netlink_current_frame(ring, NL_MMAP_STATUS_VALID);
586 if (!(msg->msg_flags & MSG_DONTWAIT) &&
587 atomic_read(&nlk->tx_ring.pending))
591 if (hdr->nm_len > maxlen) {
596 netlink_frame_flush_dcache(hdr);
598 if (likely(dst_portid == 0 && dst_group == 0 && excl)) {
599 skb = alloc_skb_head(GFP_KERNEL);
605 netlink_ring_setup_skb(skb, sk, ring, hdr);
606 NETLINK_CB(skb).flags |= NETLINK_SKB_TX;
607 __skb_put(skb, hdr->nm_len);
608 netlink_set_status(hdr, NL_MMAP_STATUS_RESERVED);
609 atomic_inc(&ring->pending);
611 skb = alloc_skb(hdr->nm_len, GFP_KERNEL);
616 __skb_put(skb, hdr->nm_len);
617 memcpy(skb->data, (void *)hdr + NL_MMAP_HDRLEN, hdr->nm_len);
618 netlink_set_status(hdr, NL_MMAP_STATUS_UNUSED);
621 netlink_increment_head(ring);
623 NETLINK_CB(skb).portid = nlk->portid;
624 NETLINK_CB(skb).dst_group = dst_group;
625 NETLINK_CB(skb).creds = siocb->scm->creds;
627 err = security_netlink_send(sk, skb);
633 if (unlikely(dst_group)) {
634 atomic_inc(&skb->users);
635 netlink_broadcast(sk, skb, dst_portid, dst_group,
638 err = netlink_unicast(sk, skb, dst_portid,
639 msg->msg_flags & MSG_DONTWAIT);
644 } while (hdr != NULL ||
645 (!(msg->msg_flags & MSG_DONTWAIT) &&
646 atomic_read(&nlk->tx_ring.pending)));
651 mutex_unlock(&nlk->pg_vec_lock);
655 static void netlink_queue_mmaped_skb(struct sock *sk, struct sk_buff *skb)
657 struct nl_mmap_hdr *hdr;
659 hdr = netlink_mmap_hdr(skb);
660 hdr->nm_len = skb->len;
661 hdr->nm_group = NETLINK_CB(skb).dst_group;
662 hdr->nm_pid = NETLINK_CB(skb).creds.pid;
663 hdr->nm_uid = from_kuid(sk_user_ns(sk), NETLINK_CB(skb).creds.uid);
664 hdr->nm_gid = from_kgid(sk_user_ns(sk), NETLINK_CB(skb).creds.gid);
665 netlink_frame_flush_dcache(hdr);
666 netlink_set_status(hdr, NL_MMAP_STATUS_VALID);
668 NETLINK_CB(skb).flags |= NETLINK_SKB_DELIVERED;
672 static void netlink_ring_set_copied(struct sock *sk, struct sk_buff *skb)
674 struct netlink_sock *nlk = nlk_sk(sk);
675 struct netlink_ring *ring = &nlk->rx_ring;
676 struct nl_mmap_hdr *hdr;
678 spin_lock_bh(&sk->sk_receive_queue.lock);
679 hdr = netlink_current_frame(ring, NL_MMAP_STATUS_UNUSED);
681 spin_unlock_bh(&sk->sk_receive_queue.lock);
686 netlink_increment_head(ring);
687 __skb_queue_tail(&sk->sk_receive_queue, skb);
688 spin_unlock_bh(&sk->sk_receive_queue.lock);
690 hdr->nm_len = skb->len;
691 hdr->nm_group = NETLINK_CB(skb).dst_group;
692 hdr->nm_pid = NETLINK_CB(skb).creds.pid;
693 hdr->nm_uid = from_kuid(sk_user_ns(sk), NETLINK_CB(skb).creds.uid);
694 hdr->nm_gid = from_kgid(sk_user_ns(sk), NETLINK_CB(skb).creds.gid);
695 netlink_set_status(hdr, NL_MMAP_STATUS_COPY);
698 #else /* CONFIG_NETLINK_MMAP */
699 #define netlink_skb_is_mmaped(skb) false
700 #define netlink_rx_is_mmaped(sk) false
701 #define netlink_tx_is_mmaped(sk) false
702 #define netlink_mmap sock_no_mmap
703 #define netlink_poll datagram_poll
704 #define netlink_mmap_sendmsg(sk, msg, dst_portid, dst_group, siocb) 0
705 #endif /* CONFIG_NETLINK_MMAP */
707 static void netlink_destroy_callback(struct netlink_callback *cb)
713 static void netlink_consume_callback(struct netlink_callback *cb)
715 consume_skb(cb->skb);
719 static void netlink_skb_destructor(struct sk_buff *skb)
721 #ifdef CONFIG_NETLINK_MMAP
722 struct nl_mmap_hdr *hdr;
723 struct netlink_ring *ring;
726 /* If a packet from the kernel to userspace was freed because of an
727 * error without being delivered to userspace, the kernel must reset
728 * the status. In the direction userspace to kernel, the status is
729 * always reset here after the packet was processed and freed.
731 if (netlink_skb_is_mmaped(skb)) {
732 hdr = netlink_mmap_hdr(skb);
733 sk = NETLINK_CB(skb).sk;
735 if (NETLINK_CB(skb).flags & NETLINK_SKB_TX) {
736 netlink_set_status(hdr, NL_MMAP_STATUS_UNUSED);
737 ring = &nlk_sk(sk)->tx_ring;
739 if (!(NETLINK_CB(skb).flags & NETLINK_SKB_DELIVERED)) {
741 netlink_set_status(hdr, NL_MMAP_STATUS_VALID);
743 ring = &nlk_sk(sk)->rx_ring;
746 WARN_ON(atomic_read(&ring->pending) == 0);
747 atomic_dec(&ring->pending);
757 static void netlink_skb_set_owner_r(struct sk_buff *skb, struct sock *sk)
759 WARN_ON(skb->sk != NULL);
761 skb->destructor = netlink_skb_destructor;
762 atomic_add(skb->truesize, &sk->sk_rmem_alloc);
763 sk_mem_charge(sk, skb->truesize);
766 static void netlink_sock_destruct(struct sock *sk)
768 struct netlink_sock *nlk = nlk_sk(sk);
772 nlk->cb->done(nlk->cb);
774 module_put(nlk->cb->module);
775 netlink_destroy_callback(nlk->cb);
778 skb_queue_purge(&sk->sk_receive_queue);
779 #ifdef CONFIG_NETLINK_MMAP
781 struct nl_mmap_req req;
783 memset(&req, 0, sizeof(req));
784 if (nlk->rx_ring.pg_vec)
785 netlink_set_ring(sk, &req, true, false);
786 memset(&req, 0, sizeof(req));
787 if (nlk->tx_ring.pg_vec)
788 netlink_set_ring(sk, &req, true, true);
790 #endif /* CONFIG_NETLINK_MMAP */
792 if (!sock_flag(sk, SOCK_DEAD)) {
793 printk(KERN_ERR "Freeing alive netlink socket %p\n", sk);
797 WARN_ON(atomic_read(&sk->sk_rmem_alloc));
798 WARN_ON(atomic_read(&sk->sk_wmem_alloc));
799 WARN_ON(nlk_sk(sk)->groups);
802 /* This lock without WQ_FLAG_EXCLUSIVE is good on UP and it is _very_ bad on
803 * SMP. Look, when several writers sleep and reader wakes them up, all but one
804 * immediately hit write lock and grab all the cpus. Exclusive sleep solves
805 * this, _but_ remember, it adds useless work on UP machines.
808 void netlink_table_grab(void)
809 __acquires(nl_table_lock)
813 write_lock_irq(&nl_table_lock);
815 if (atomic_read(&nl_table_users)) {
816 DECLARE_WAITQUEUE(wait, current);
818 add_wait_queue_exclusive(&nl_table_wait, &wait);
820 set_current_state(TASK_UNINTERRUPTIBLE);
821 if (atomic_read(&nl_table_users) == 0)
823 write_unlock_irq(&nl_table_lock);
825 write_lock_irq(&nl_table_lock);
828 __set_current_state(TASK_RUNNING);
829 remove_wait_queue(&nl_table_wait, &wait);
833 void netlink_table_ungrab(void)
834 __releases(nl_table_lock)
836 write_unlock_irq(&nl_table_lock);
837 wake_up(&nl_table_wait);
841 netlink_lock_table(void)
843 /* read_lock() synchronizes us to netlink_table_grab */
845 read_lock(&nl_table_lock);
846 atomic_inc(&nl_table_users);
847 read_unlock(&nl_table_lock);
851 netlink_unlock_table(void)
853 if (atomic_dec_and_test(&nl_table_users))
854 wake_up(&nl_table_wait);
857 static struct sock *netlink_lookup(struct net *net, int protocol, u32 portid)
859 struct nl_portid_hash *hash = &nl_table[protocol].hash;
860 struct hlist_head *head;
863 read_lock(&nl_table_lock);
864 head = nl_portid_hashfn(hash, portid);
865 sk_for_each(sk, head) {
866 if (net_eq(sock_net(sk), net) && (nlk_sk(sk)->portid == portid)) {
873 read_unlock(&nl_table_lock);
877 static struct hlist_head *nl_portid_hash_zalloc(size_t size)
879 if (size <= PAGE_SIZE)
880 return kzalloc(size, GFP_ATOMIC);
882 return (struct hlist_head *)
883 __get_free_pages(GFP_ATOMIC | __GFP_ZERO,
887 static void nl_portid_hash_free(struct hlist_head *table, size_t size)
889 if (size <= PAGE_SIZE)
892 free_pages((unsigned long)table, get_order(size));
895 static int nl_portid_hash_rehash(struct nl_portid_hash *hash, int grow)
897 unsigned int omask, mask, shift;
899 struct hlist_head *otable, *table;
902 omask = mask = hash->mask;
903 osize = size = (mask + 1) * sizeof(*table);
907 if (++shift > hash->max_shift)
913 table = nl_portid_hash_zalloc(size);
917 otable = hash->table;
921 get_random_bytes(&hash->rnd, sizeof(hash->rnd));
923 for (i = 0; i <= omask; i++) {
925 struct hlist_node *tmp;
927 sk_for_each_safe(sk, tmp, &otable[i])
928 __sk_add_node(sk, nl_portid_hashfn(hash, nlk_sk(sk)->portid));
931 nl_portid_hash_free(otable, osize);
932 hash->rehash_time = jiffies + 10 * 60 * HZ;
936 static inline int nl_portid_hash_dilute(struct nl_portid_hash *hash, int len)
938 int avg = hash->entries >> hash->shift;
940 if (unlikely(avg > 1) && nl_portid_hash_rehash(hash, 1))
943 if (unlikely(len > avg) && time_after(jiffies, hash->rehash_time)) {
944 nl_portid_hash_rehash(hash, 0);
951 static const struct proto_ops netlink_ops;
954 netlink_update_listeners(struct sock *sk)
956 struct netlink_table *tbl = &nl_table[sk->sk_protocol];
959 struct listeners *listeners;
961 listeners = nl_deref_protected(tbl->listeners);
965 for (i = 0; i < NLGRPLONGS(tbl->groups); i++) {
967 sk_for_each_bound(sk, &tbl->mc_list) {
968 if (i < NLGRPLONGS(nlk_sk(sk)->ngroups))
969 mask |= nlk_sk(sk)->groups[i];
971 listeners->masks[i] = mask;
973 /* this function is only called with the netlink table "grabbed", which
974 * makes sure updates are visible before bind or setsockopt return. */
977 static int netlink_insert(struct sock *sk, struct net *net, u32 portid)
979 struct nl_portid_hash *hash = &nl_table[sk->sk_protocol].hash;
980 struct hlist_head *head;
981 int err = -EADDRINUSE;
985 netlink_table_grab();
986 head = nl_portid_hashfn(hash, portid);
988 sk_for_each(osk, head) {
989 if (net_eq(sock_net(osk), net) && (nlk_sk(osk)->portid == portid))
997 if (nlk_sk(sk)->portid)
1001 if (BITS_PER_LONG > 32 && unlikely(hash->entries >= UINT_MAX))
1004 if (len && nl_portid_hash_dilute(hash, len))
1005 head = nl_portid_hashfn(hash, portid);
1007 nlk_sk(sk)->portid = portid;
1008 sk_add_node(sk, head);
1012 netlink_table_ungrab();
1016 static void netlink_remove(struct sock *sk)
1018 netlink_table_grab();
1019 if (sk_del_node_init(sk))
1020 nl_table[sk->sk_protocol].hash.entries--;
1021 if (nlk_sk(sk)->subscriptions)
1022 __sk_del_bind_node(sk);
1023 netlink_table_ungrab();
1026 static struct proto netlink_proto = {
1028 .owner = THIS_MODULE,
1029 .obj_size = sizeof(struct netlink_sock),
1032 static int __netlink_create(struct net *net, struct socket *sock,
1033 struct mutex *cb_mutex, int protocol)
1036 struct netlink_sock *nlk;
1038 sock->ops = &netlink_ops;
1040 sk = sk_alloc(net, PF_NETLINK, GFP_KERNEL, &netlink_proto);
1044 sock_init_data(sock, sk);
1048 nlk->cb_mutex = cb_mutex;
1050 nlk->cb_mutex = &nlk->cb_def_mutex;
1051 mutex_init(nlk->cb_mutex);
1053 init_waitqueue_head(&nlk->wait);
1054 #ifdef CONFIG_NETLINK_MMAP
1055 mutex_init(&nlk->pg_vec_lock);
1058 sk->sk_destruct = netlink_sock_destruct;
1059 sk->sk_protocol = protocol;
1063 static int netlink_create(struct net *net, struct socket *sock, int protocol,
1066 struct module *module = NULL;
1067 struct mutex *cb_mutex;
1068 struct netlink_sock *nlk;
1069 void (*bind)(int group);
1072 sock->state = SS_UNCONNECTED;
1074 if (sock->type != SOCK_RAW && sock->type != SOCK_DGRAM)
1075 return -ESOCKTNOSUPPORT;
1077 if (protocol < 0 || protocol >= MAX_LINKS)
1078 return -EPROTONOSUPPORT;
1080 netlink_lock_table();
1081 #ifdef CONFIG_MODULES
1082 if (!nl_table[protocol].registered) {
1083 netlink_unlock_table();
1084 request_module("net-pf-%d-proto-%d", PF_NETLINK, protocol);
1085 netlink_lock_table();
1088 if (nl_table[protocol].registered &&
1089 try_module_get(nl_table[protocol].module))
1090 module = nl_table[protocol].module;
1092 err = -EPROTONOSUPPORT;
1093 cb_mutex = nl_table[protocol].cb_mutex;
1094 bind = nl_table[protocol].bind;
1095 netlink_unlock_table();
1100 err = __netlink_create(net, sock, cb_mutex, protocol);
1105 sock_prot_inuse_add(net, &netlink_proto, 1);
1108 nlk = nlk_sk(sock->sk);
1109 nlk->module = module;
1110 nlk->netlink_bind = bind;
1119 static int netlink_release(struct socket *sock)
1121 struct sock *sk = sock->sk;
1122 struct netlink_sock *nlk;
1132 * OK. Socket is unlinked, any packets that arrive now
1137 wake_up_interruptible_all(&nlk->wait);
1139 skb_queue_purge(&sk->sk_write_queue);
1142 struct netlink_notify n = {
1143 .net = sock_net(sk),
1144 .protocol = sk->sk_protocol,
1145 .portid = nlk->portid,
1147 atomic_notifier_call_chain(&netlink_chain,
1148 NETLINK_URELEASE, &n);
1151 module_put(nlk->module);
1153 netlink_table_grab();
1154 if (netlink_is_kernel(sk)) {
1155 BUG_ON(nl_table[sk->sk_protocol].registered == 0);
1156 if (--nl_table[sk->sk_protocol].registered == 0) {
1157 struct listeners *old;
1159 old = nl_deref_protected(nl_table[sk->sk_protocol].listeners);
1160 RCU_INIT_POINTER(nl_table[sk->sk_protocol].listeners, NULL);
1161 kfree_rcu(old, rcu);
1162 nl_table[sk->sk_protocol].module = NULL;
1163 nl_table[sk->sk_protocol].bind = NULL;
1164 nl_table[sk->sk_protocol].flags = 0;
1165 nl_table[sk->sk_protocol].registered = 0;
1167 } else if (nlk->subscriptions) {
1168 netlink_update_listeners(sk);
1170 netlink_table_ungrab();
1176 sock_prot_inuse_add(sock_net(sk), &netlink_proto, -1);
1182 static int netlink_autobind(struct socket *sock)
1184 struct sock *sk = sock->sk;
1185 struct net *net = sock_net(sk);
1186 struct nl_portid_hash *hash = &nl_table[sk->sk_protocol].hash;
1187 struct hlist_head *head;
1189 s32 portid = task_tgid_vnr(current);
1191 static s32 rover = -4097;
1195 netlink_table_grab();
1196 head = nl_portid_hashfn(hash, portid);
1197 sk_for_each(osk, head) {
1198 if (!net_eq(sock_net(osk), net))
1200 if (nlk_sk(osk)->portid == portid) {
1201 /* Bind collision, search negative portid values. */
1205 netlink_table_ungrab();
1209 netlink_table_ungrab();
1211 err = netlink_insert(sk, net, portid);
1212 if (err == -EADDRINUSE)
1215 /* If 2 threads race to autobind, that is fine. */
1223 * __netlink_ns_capable - General netlink message capability test
1224 * @nsp: NETLINK_CB of the socket buffer holding a netlink command from userspace.
1225 * @user_ns: The user namespace of the capability to use
1226 * @cap: The capability to use
1228 * Test to see if the opener of the socket we received the message
1229 * from had when the netlink socket was created and the sender of the
1230 * message has has the capability @cap in the user namespace @user_ns.
1232 bool __netlink_ns_capable(const struct netlink_skb_parms *nsp,
1233 struct user_namespace *user_ns, int cap)
1235 return sk_ns_capable(nsp->sk, user_ns, cap);
1237 EXPORT_SYMBOL(__netlink_ns_capable);
1240 * netlink_ns_capable - General netlink message capability test
1241 * @skb: socket buffer holding a netlink command from userspace
1242 * @user_ns: The user namespace of the capability to use
1243 * @cap: The capability to use
1245 * Test to see if the opener of the socket we received the message
1246 * from had when the netlink socket was created and the sender of the
1247 * message has has the capability @cap in the user namespace @user_ns.
1249 bool netlink_ns_capable(const struct sk_buff *skb,
1250 struct user_namespace *user_ns, int cap)
1252 return __netlink_ns_capable(&NETLINK_CB(skb), user_ns, cap);
1254 EXPORT_SYMBOL(netlink_ns_capable);
1257 * netlink_capable - Netlink global message capability test
1258 * @skb: socket buffer holding a netlink command from userspace
1259 * @cap: The capability to use
1261 * Test to see if the opener of the socket we received the message
1262 * from had when the netlink socket was created and the sender of the
1263 * message has has the capability @cap in all user namespaces.
1265 bool netlink_capable(const struct sk_buff *skb, int cap)
1267 return netlink_ns_capable(skb, &init_user_ns, cap);
1269 EXPORT_SYMBOL(netlink_capable);
1272 * netlink_net_capable - Netlink network namespace message capability test
1273 * @skb: socket buffer holding a netlink command from userspace
1274 * @cap: The capability to use
1276 * Test to see if the opener of the socket we received the message
1277 * from had when the netlink socket was created and the sender of the
1278 * message has has the capability @cap over the network namespace of
1279 * the socket we received the message from.
1281 bool netlink_net_capable(const struct sk_buff *skb, int cap)
1283 return netlink_ns_capable(skb, sock_net(skb->sk)->user_ns, cap);
1285 EXPORT_SYMBOL(netlink_net_capable);
1287 static inline int netlink_allowed(const struct socket *sock, unsigned int flag)
1289 return (nl_table[sock->sk->sk_protocol].flags & flag) ||
1290 ns_capable(sock_net(sock->sk)->user_ns, CAP_NET_ADMIN);
1294 netlink_update_subscriptions(struct sock *sk, unsigned int subscriptions)
1296 struct netlink_sock *nlk = nlk_sk(sk);
1298 if (nlk->subscriptions && !subscriptions)
1299 __sk_del_bind_node(sk);
1300 else if (!nlk->subscriptions && subscriptions)
1301 sk_add_bind_node(sk, &nl_table[sk->sk_protocol].mc_list);
1302 nlk->subscriptions = subscriptions;
1305 static int netlink_realloc_groups(struct sock *sk)
1307 struct netlink_sock *nlk = nlk_sk(sk);
1308 unsigned int groups;
1309 unsigned long *new_groups;
1312 netlink_table_grab();
1314 groups = nl_table[sk->sk_protocol].groups;
1315 if (!nl_table[sk->sk_protocol].registered) {
1320 if (nlk->ngroups >= groups)
1323 new_groups = krealloc(nlk->groups, NLGRPSZ(groups), GFP_ATOMIC);
1324 if (new_groups == NULL) {
1328 memset((char *)new_groups + NLGRPSZ(nlk->ngroups), 0,
1329 NLGRPSZ(groups) - NLGRPSZ(nlk->ngroups));
1331 nlk->groups = new_groups;
1332 nlk->ngroups = groups;
1334 netlink_table_ungrab();
1338 static int netlink_bind(struct socket *sock, struct sockaddr *addr,
1341 struct sock *sk = sock->sk;
1342 struct net *net = sock_net(sk);
1343 struct netlink_sock *nlk = nlk_sk(sk);
1344 struct sockaddr_nl *nladdr = (struct sockaddr_nl *)addr;
1347 if (addr_len < sizeof(struct sockaddr_nl))
1350 if (nladdr->nl_family != AF_NETLINK)
1353 /* Only superuser is allowed to listen multicasts */
1354 if (nladdr->nl_groups) {
1355 if (!netlink_allowed(sock, NL_CFG_F_NONROOT_RECV))
1357 err = netlink_realloc_groups(sk);
1363 if (nladdr->nl_pid != nlk->portid)
1366 err = nladdr->nl_pid ?
1367 netlink_insert(sk, net, nladdr->nl_pid) :
1368 netlink_autobind(sock);
1373 if (!nladdr->nl_groups && (nlk->groups == NULL || !(u32)nlk->groups[0]))
1376 netlink_table_grab();
1377 netlink_update_subscriptions(sk, nlk->subscriptions +
1378 hweight32(nladdr->nl_groups) -
1379 hweight32(nlk->groups[0]));
1380 nlk->groups[0] = (nlk->groups[0] & ~0xffffffffUL) | nladdr->nl_groups;
1381 netlink_update_listeners(sk);
1382 netlink_table_ungrab();
1384 if (nlk->netlink_bind && nlk->groups[0]) {
1387 for (i=0; i<nlk->ngroups; i++) {
1388 if (test_bit(i, nlk->groups))
1389 nlk->netlink_bind(i);
1396 static int netlink_connect(struct socket *sock, struct sockaddr *addr,
1397 int alen, int flags)
1400 struct sock *sk = sock->sk;
1401 struct netlink_sock *nlk = nlk_sk(sk);
1402 struct sockaddr_nl *nladdr = (struct sockaddr_nl *)addr;
1404 if (alen < sizeof(addr->sa_family))
1407 if (addr->sa_family == AF_UNSPEC) {
1408 sk->sk_state = NETLINK_UNCONNECTED;
1409 nlk->dst_portid = 0;
1413 if (addr->sa_family != AF_NETLINK)
1416 /* Only superuser is allowed to send multicasts */
1417 if (nladdr->nl_groups && !netlink_allowed(sock, NL_CFG_F_NONROOT_SEND))
1421 err = netlink_autobind(sock);
1424 sk->sk_state = NETLINK_CONNECTED;
1425 nlk->dst_portid = nladdr->nl_pid;
1426 nlk->dst_group = ffs(nladdr->nl_groups);
1432 static int netlink_getname(struct socket *sock, struct sockaddr *addr,
1433 int *addr_len, int peer)
1435 struct sock *sk = sock->sk;
1436 struct netlink_sock *nlk = nlk_sk(sk);
1437 DECLARE_SOCKADDR(struct sockaddr_nl *, nladdr, addr);
1439 nladdr->nl_family = AF_NETLINK;
1441 *addr_len = sizeof(*nladdr);
1444 nladdr->nl_pid = nlk->dst_portid;
1445 nladdr->nl_groups = netlink_group_mask(nlk->dst_group);
1447 nladdr->nl_pid = nlk->portid;
1448 nladdr->nl_groups = nlk->groups ? nlk->groups[0] : 0;
1453 static struct sock *netlink_getsockbyportid(struct sock *ssk, u32 portid)
1456 struct netlink_sock *nlk;
1458 sock = netlink_lookup(sock_net(ssk), ssk->sk_protocol, portid);
1460 return ERR_PTR(-ECONNREFUSED);
1462 /* Don't bother queuing skb if kernel socket has no input function */
1464 if (sock->sk_state == NETLINK_CONNECTED &&
1465 nlk->dst_portid != nlk_sk(ssk)->portid) {
1467 return ERR_PTR(-ECONNREFUSED);
1472 struct sock *netlink_getsockbyfilp(struct file *filp)
1474 struct inode *inode = file_inode(filp);
1477 if (!S_ISSOCK(inode->i_mode))
1478 return ERR_PTR(-ENOTSOCK);
1480 sock = SOCKET_I(inode)->sk;
1481 if (sock->sk_family != AF_NETLINK)
1482 return ERR_PTR(-EINVAL);
1489 * Attach a skb to a netlink socket.
1490 * The caller must hold a reference to the destination socket. On error, the
1491 * reference is dropped. The skb is not send to the destination, just all
1492 * all error checks are performed and memory in the queue is reserved.
1494 * < 0: error. skb freed, reference to sock dropped.
1496 * 1: repeat lookup - reference dropped while waiting for socket memory.
1498 int netlink_attachskb(struct sock *sk, struct sk_buff *skb,
1499 long *timeo, struct sock *ssk)
1501 struct netlink_sock *nlk;
1505 if ((atomic_read(&sk->sk_rmem_alloc) > sk->sk_rcvbuf ||
1506 test_bit(NETLINK_CONGESTED, &nlk->state)) &&
1507 !netlink_skb_is_mmaped(skb)) {
1508 DECLARE_WAITQUEUE(wait, current);
1510 if (!ssk || netlink_is_kernel(ssk))
1511 netlink_overrun(sk);
1517 __set_current_state(TASK_INTERRUPTIBLE);
1518 add_wait_queue(&nlk->wait, &wait);
1520 if ((atomic_read(&sk->sk_rmem_alloc) > sk->sk_rcvbuf ||
1521 test_bit(NETLINK_CONGESTED, &nlk->state)) &&
1522 !sock_flag(sk, SOCK_DEAD))
1523 *timeo = schedule_timeout(*timeo);
1525 __set_current_state(TASK_RUNNING);
1526 remove_wait_queue(&nlk->wait, &wait);
1529 if (signal_pending(current)) {
1531 return sock_intr_errno(*timeo);
1535 netlink_skb_set_owner_r(skb, sk);
1539 static int __netlink_sendskb(struct sock *sk, struct sk_buff *skb)
1543 #ifdef CONFIG_NETLINK_MMAP
1544 if (netlink_skb_is_mmaped(skb))
1545 netlink_queue_mmaped_skb(sk, skb);
1546 else if (netlink_rx_is_mmaped(sk))
1547 netlink_ring_set_copied(sk, skb);
1549 #endif /* CONFIG_NETLINK_MMAP */
1550 skb_queue_tail(&sk->sk_receive_queue, skb);
1551 sk->sk_data_ready(sk, len);
1555 int netlink_sendskb(struct sock *sk, struct sk_buff *skb)
1557 int len = __netlink_sendskb(sk, skb);
1563 void netlink_detachskb(struct sock *sk, struct sk_buff *skb)
1569 static struct sk_buff *netlink_trim(struct sk_buff *skb, gfp_t allocation)
1573 WARN_ON(skb->sk != NULL);
1574 if (netlink_skb_is_mmaped(skb))
1577 delta = skb->end - skb->tail;
1578 if (delta * 2 < skb->truesize)
1581 if (skb_shared(skb)) {
1582 struct sk_buff *nskb = skb_clone(skb, allocation);
1589 if (!pskb_expand_head(skb, 0, -delta, allocation))
1590 skb->truesize -= delta;
1595 static int netlink_unicast_kernel(struct sock *sk, struct sk_buff *skb,
1599 struct netlink_sock *nlk = nlk_sk(sk);
1601 ret = -ECONNREFUSED;
1602 if (nlk->netlink_rcv != NULL) {
1604 netlink_skb_set_owner_r(skb, sk);
1605 NETLINK_CB(skb).sk = ssk;
1606 nlk->netlink_rcv(skb);
1615 int netlink_unicast(struct sock *ssk, struct sk_buff *skb,
1616 u32 portid, int nonblock)
1622 skb = netlink_trim(skb, gfp_any());
1624 timeo = sock_sndtimeo(ssk, nonblock);
1626 sk = netlink_getsockbyportid(ssk, portid);
1631 if (netlink_is_kernel(sk))
1632 return netlink_unicast_kernel(sk, skb, ssk);
1634 if (sk_filter(sk, skb)) {
1641 err = netlink_attachskb(sk, skb, &timeo, ssk);
1647 return netlink_sendskb(sk, skb);
1649 EXPORT_SYMBOL(netlink_unicast);
1651 struct sk_buff *netlink_alloc_skb(struct sock *ssk, unsigned int size,
1652 u32 dst_portid, gfp_t gfp_mask)
1654 #ifdef CONFIG_NETLINK_MMAP
1655 struct sock *sk = NULL;
1656 struct sk_buff *skb;
1657 struct netlink_ring *ring;
1658 struct nl_mmap_hdr *hdr;
1659 unsigned int maxlen;
1661 sk = netlink_getsockbyportid(ssk, dst_portid);
1665 ring = &nlk_sk(sk)->rx_ring;
1666 /* fast-path without atomic ops for common case: non-mmaped receiver */
1667 if (ring->pg_vec == NULL)
1670 skb = alloc_skb_head(gfp_mask);
1674 spin_lock_bh(&sk->sk_receive_queue.lock);
1675 /* check again under lock */
1676 if (ring->pg_vec == NULL)
1679 maxlen = ring->frame_size - NL_MMAP_HDRLEN;
1683 netlink_forward_ring(ring);
1684 hdr = netlink_current_frame(ring, NL_MMAP_STATUS_UNUSED);
1687 netlink_ring_setup_skb(skb, sk, ring, hdr);
1688 netlink_set_status(hdr, NL_MMAP_STATUS_RESERVED);
1689 atomic_inc(&ring->pending);
1690 netlink_increment_head(ring);
1692 spin_unlock_bh(&sk->sk_receive_queue.lock);
1697 spin_unlock_bh(&sk->sk_receive_queue.lock);
1698 netlink_overrun(sk);
1705 spin_unlock_bh(&sk->sk_receive_queue.lock);
1710 return alloc_skb(size, gfp_mask);
1712 EXPORT_SYMBOL_GPL(netlink_alloc_skb);
1714 int netlink_has_listeners(struct sock *sk, unsigned int group)
1717 struct listeners *listeners;
1719 BUG_ON(!netlink_is_kernel(sk));
1722 listeners = rcu_dereference(nl_table[sk->sk_protocol].listeners);
1724 if (listeners && group - 1 < nl_table[sk->sk_protocol].groups)
1725 res = test_bit(group - 1, listeners->masks);
1731 EXPORT_SYMBOL_GPL(netlink_has_listeners);
1733 static int netlink_broadcast_deliver(struct sock *sk, struct sk_buff *skb)
1735 struct netlink_sock *nlk = nlk_sk(sk);
1737 if (atomic_read(&sk->sk_rmem_alloc) <= sk->sk_rcvbuf &&
1738 !test_bit(NETLINK_CONGESTED, &nlk->state)) {
1739 netlink_skb_set_owner_r(skb, sk);
1740 __netlink_sendskb(sk, skb);
1741 return atomic_read(&sk->sk_rmem_alloc) > (sk->sk_rcvbuf >> 1);
1746 struct netlink_broadcast_data {
1747 struct sock *exclude_sk;
1752 int delivery_failure;
1756 struct sk_buff *skb, *skb2;
1757 int (*tx_filter)(struct sock *dsk, struct sk_buff *skb, void *data);
1761 static int do_one_broadcast(struct sock *sk,
1762 struct netlink_broadcast_data *p)
1764 struct netlink_sock *nlk = nlk_sk(sk);
1767 if (p->exclude_sk == sk)
1770 if (nlk->portid == p->portid || p->group - 1 >= nlk->ngroups ||
1771 !test_bit(p->group - 1, nlk->groups))
1774 if (!net_eq(sock_net(sk), p->net))
1778 netlink_overrun(sk);
1783 if (p->skb2 == NULL) {
1784 if (skb_shared(p->skb)) {
1785 p->skb2 = skb_clone(p->skb, p->allocation);
1787 p->skb2 = skb_get(p->skb);
1789 * skb ownership may have been set when
1790 * delivered to a previous socket.
1792 skb_orphan(p->skb2);
1795 if (p->skb2 == NULL) {
1796 netlink_overrun(sk);
1797 /* Clone failed. Notify ALL listeners. */
1799 if (nlk->flags & NETLINK_BROADCAST_SEND_ERROR)
1800 p->delivery_failure = 1;
1801 } else if (p->tx_filter && p->tx_filter(sk, p->skb2, p->tx_data)) {
1804 } else if (sk_filter(sk, p->skb2)) {
1807 } else if ((val = netlink_broadcast_deliver(sk, p->skb2)) < 0) {
1808 netlink_overrun(sk);
1809 if (nlk->flags & NETLINK_BROADCAST_SEND_ERROR)
1810 p->delivery_failure = 1;
1812 p->congested |= val;
1822 int netlink_broadcast_filtered(struct sock *ssk, struct sk_buff *skb, u32 portid,
1823 u32 group, gfp_t allocation,
1824 int (*filter)(struct sock *dsk, struct sk_buff *skb, void *data),
1827 struct net *net = sock_net(ssk);
1828 struct netlink_broadcast_data info;
1831 skb = netlink_trim(skb, allocation);
1833 info.exclude_sk = ssk;
1835 info.portid = portid;
1838 info.delivery_failure = 0;
1841 info.allocation = allocation;
1844 info.tx_filter = filter;
1845 info.tx_data = filter_data;
1847 /* While we sleep in clone, do not allow to change socket list */
1849 netlink_lock_table();
1851 sk_for_each_bound(sk, &nl_table[ssk->sk_protocol].mc_list)
1852 do_one_broadcast(sk, &info);
1856 netlink_unlock_table();
1858 if (info.delivery_failure) {
1859 kfree_skb(info.skb2);
1862 consume_skb(info.skb2);
1864 if (info.delivered) {
1865 if (info.congested && (allocation & __GFP_WAIT))
1871 EXPORT_SYMBOL(netlink_broadcast_filtered);
1873 int netlink_broadcast(struct sock *ssk, struct sk_buff *skb, u32 portid,
1874 u32 group, gfp_t allocation)
1876 return netlink_broadcast_filtered(ssk, skb, portid, group, allocation,
1879 EXPORT_SYMBOL(netlink_broadcast);
1881 struct netlink_set_err_data {
1882 struct sock *exclude_sk;
1888 static int do_one_set_err(struct sock *sk, struct netlink_set_err_data *p)
1890 struct netlink_sock *nlk = nlk_sk(sk);
1893 if (sk == p->exclude_sk)
1896 if (!net_eq(sock_net(sk), sock_net(p->exclude_sk)))
1899 if (nlk->portid == p->portid || p->group - 1 >= nlk->ngroups ||
1900 !test_bit(p->group - 1, nlk->groups))
1903 if (p->code == ENOBUFS && nlk->flags & NETLINK_RECV_NO_ENOBUFS) {
1908 sk->sk_err = p->code;
1909 sk->sk_error_report(sk);
1915 * netlink_set_err - report error to broadcast listeners
1916 * @ssk: the kernel netlink socket, as returned by netlink_kernel_create()
1917 * @portid: the PORTID of a process that we want to skip (if any)
1918 * @groups: the broadcast group that will notice the error
1919 * @code: error code, must be negative (as usual in kernelspace)
1921 * This function returns the number of broadcast listeners that have set the
1922 * NETLINK_RECV_NO_ENOBUFS socket option.
1924 int netlink_set_err(struct sock *ssk, u32 portid, u32 group, int code)
1926 struct netlink_set_err_data info;
1930 info.exclude_sk = ssk;
1931 info.portid = portid;
1933 /* sk->sk_err wants a positive error value */
1936 read_lock(&nl_table_lock);
1938 sk_for_each_bound(sk, &nl_table[ssk->sk_protocol].mc_list)
1939 ret += do_one_set_err(sk, &info);
1941 read_unlock(&nl_table_lock);
1944 EXPORT_SYMBOL(netlink_set_err);
1946 /* must be called with netlink table grabbed */
1947 static void netlink_update_socket_mc(struct netlink_sock *nlk,
1951 int old, new = !!is_new, subscriptions;
1953 old = test_bit(group - 1, nlk->groups);
1954 subscriptions = nlk->subscriptions - old + new;
1956 __set_bit(group - 1, nlk->groups);
1958 __clear_bit(group - 1, nlk->groups);
1959 netlink_update_subscriptions(&nlk->sk, subscriptions);
1960 netlink_update_listeners(&nlk->sk);
1963 static int netlink_setsockopt(struct socket *sock, int level, int optname,
1964 char __user *optval, unsigned int optlen)
1966 struct sock *sk = sock->sk;
1967 struct netlink_sock *nlk = nlk_sk(sk);
1968 unsigned int val = 0;
1971 if (level != SOL_NETLINK)
1972 return -ENOPROTOOPT;
1974 if (optname != NETLINK_RX_RING && optname != NETLINK_TX_RING &&
1975 optlen >= sizeof(int) &&
1976 get_user(val, (unsigned int __user *)optval))
1980 case NETLINK_PKTINFO:
1982 nlk->flags |= NETLINK_RECV_PKTINFO;
1984 nlk->flags &= ~NETLINK_RECV_PKTINFO;
1987 case NETLINK_ADD_MEMBERSHIP:
1988 case NETLINK_DROP_MEMBERSHIP: {
1989 if (!netlink_allowed(sock, NL_CFG_F_NONROOT_RECV))
1991 err = netlink_realloc_groups(sk);
1994 if (!val || val - 1 >= nlk->ngroups)
1996 netlink_table_grab();
1997 netlink_update_socket_mc(nlk, val,
1998 optname == NETLINK_ADD_MEMBERSHIP);
1999 netlink_table_ungrab();
2001 if (nlk->netlink_bind)
2002 nlk->netlink_bind(val);
2007 case NETLINK_BROADCAST_ERROR:
2009 nlk->flags |= NETLINK_BROADCAST_SEND_ERROR;
2011 nlk->flags &= ~NETLINK_BROADCAST_SEND_ERROR;
2014 case NETLINK_NO_ENOBUFS:
2016 nlk->flags |= NETLINK_RECV_NO_ENOBUFS;
2017 clear_bit(NETLINK_CONGESTED, &nlk->state);
2018 wake_up_interruptible(&nlk->wait);
2020 nlk->flags &= ~NETLINK_RECV_NO_ENOBUFS;
2024 #ifdef CONFIG_NETLINK_MMAP
2025 case NETLINK_RX_RING:
2026 case NETLINK_TX_RING: {
2027 struct nl_mmap_req req;
2029 /* Rings might consume more memory than queue limits, require
2032 if (!capable(CAP_NET_ADMIN))
2034 if (optlen < sizeof(req))
2036 if (copy_from_user(&req, optval, sizeof(req)))
2038 err = netlink_set_ring(sk, &req, false,
2039 optname == NETLINK_TX_RING);
2042 #endif /* CONFIG_NETLINK_MMAP */
2049 static int netlink_getsockopt(struct socket *sock, int level, int optname,
2050 char __user *optval, int __user *optlen)
2052 struct sock *sk = sock->sk;
2053 struct netlink_sock *nlk = nlk_sk(sk);
2056 if (level != SOL_NETLINK)
2057 return -ENOPROTOOPT;
2059 if (get_user(len, optlen))
2065 case NETLINK_PKTINFO:
2066 if (len < sizeof(int))
2069 val = nlk->flags & NETLINK_RECV_PKTINFO ? 1 : 0;
2070 if (put_user(len, optlen) ||
2071 put_user(val, optval))
2075 case NETLINK_BROADCAST_ERROR:
2076 if (len < sizeof(int))
2079 val = nlk->flags & NETLINK_BROADCAST_SEND_ERROR ? 1 : 0;
2080 if (put_user(len, optlen) ||
2081 put_user(val, optval))
2085 case NETLINK_NO_ENOBUFS:
2086 if (len < sizeof(int))
2089 val = nlk->flags & NETLINK_RECV_NO_ENOBUFS ? 1 : 0;
2090 if (put_user(len, optlen) ||
2091 put_user(val, optval))
2101 static void netlink_cmsg_recv_pktinfo(struct msghdr *msg, struct sk_buff *skb)
2103 struct nl_pktinfo info;
2105 info.group = NETLINK_CB(skb).dst_group;
2106 put_cmsg(msg, SOL_NETLINK, NETLINK_PKTINFO, sizeof(info), &info);
2109 static int netlink_sendmsg(struct kiocb *kiocb, struct socket *sock,
2110 struct msghdr *msg, size_t len)
2112 struct sock_iocb *siocb = kiocb_to_siocb(kiocb);
2113 struct sock *sk = sock->sk;
2114 struct netlink_sock *nlk = nlk_sk(sk);
2115 struct sockaddr_nl *addr = msg->msg_name;
2118 struct sk_buff *skb;
2120 struct scm_cookie scm;
2122 if (msg->msg_flags&MSG_OOB)
2125 if (NULL == siocb->scm)
2128 err = scm_send(sock, msg, siocb->scm, true);
2132 if (msg->msg_namelen) {
2134 if (addr->nl_family != AF_NETLINK)
2136 dst_portid = addr->nl_pid;
2137 dst_group = ffs(addr->nl_groups);
2139 if ((dst_group || dst_portid) &&
2140 !netlink_allowed(sock, NL_CFG_F_NONROOT_SEND))
2143 dst_portid = nlk->dst_portid;
2144 dst_group = nlk->dst_group;
2148 err = netlink_autobind(sock);
2153 if (netlink_tx_is_mmaped(sk) &&
2154 msg->msg_iov->iov_base == NULL) {
2155 err = netlink_mmap_sendmsg(sk, msg, dst_portid, dst_group,
2161 if (len > sk->sk_sndbuf - 32)
2164 skb = alloc_skb(len, GFP_KERNEL);
2168 NETLINK_CB(skb).portid = nlk->portid;
2169 NETLINK_CB(skb).dst_group = dst_group;
2170 NETLINK_CB(skb).creds = siocb->scm->creds;
2173 if (memcpy_fromiovec(skb_put(skb, len), msg->msg_iov, len)) {
2178 err = security_netlink_send(sk, skb);
2185 atomic_inc(&skb->users);
2186 netlink_broadcast(sk, skb, dst_portid, dst_group, GFP_KERNEL);
2188 err = netlink_unicast(sk, skb, dst_portid, msg->msg_flags&MSG_DONTWAIT);
2191 scm_destroy(siocb->scm);
2195 static int netlink_recvmsg(struct kiocb *kiocb, struct socket *sock,
2196 struct msghdr *msg, size_t len,
2199 struct sock_iocb *siocb = kiocb_to_siocb(kiocb);
2200 struct scm_cookie scm;
2201 struct sock *sk = sock->sk;
2202 struct netlink_sock *nlk = nlk_sk(sk);
2203 int noblock = flags&MSG_DONTWAIT;
2205 struct sk_buff *skb, *data_skb;
2213 skb = skb_recv_datagram(sk, flags, noblock, &err);
2219 #ifdef CONFIG_COMPAT_NETLINK_MESSAGES
2220 if (unlikely(skb_shinfo(skb)->frag_list)) {
2222 * If this skb has a frag_list, then here that means that we
2223 * will have to use the frag_list skb's data for compat tasks
2224 * and the regular skb's data for normal (non-compat) tasks.
2226 * If we need to send the compat skb, assign it to the
2227 * 'data_skb' variable so that it will be used below for data
2228 * copying. We keep 'skb' for everything else, including
2229 * freeing both later.
2231 if (flags & MSG_CMSG_COMPAT)
2232 data_skb = skb_shinfo(skb)->frag_list;
2236 copied = data_skb->len;
2238 msg->msg_flags |= MSG_TRUNC;
2242 skb_reset_transport_header(data_skb);
2243 err = skb_copy_datagram_iovec(data_skb, 0, msg->msg_iov, copied);
2245 if (msg->msg_name) {
2246 struct sockaddr_nl *addr = (struct sockaddr_nl *)msg->msg_name;
2247 addr->nl_family = AF_NETLINK;
2249 addr->nl_pid = NETLINK_CB(skb).portid;
2250 addr->nl_groups = netlink_group_mask(NETLINK_CB(skb).dst_group);
2251 msg->msg_namelen = sizeof(*addr);
2254 if (nlk->flags & NETLINK_RECV_PKTINFO)
2255 netlink_cmsg_recv_pktinfo(msg, skb);
2257 if (NULL == siocb->scm) {
2258 memset(&scm, 0, sizeof(scm));
2261 siocb->scm->creds = *NETLINK_CREDS(skb);
2262 if (flags & MSG_TRUNC)
2263 copied = data_skb->len;
2265 skb_free_datagram(sk, skb);
2267 if (nlk->cb && atomic_read(&sk->sk_rmem_alloc) <= sk->sk_rcvbuf / 2) {
2268 ret = netlink_dump(sk);
2271 sk->sk_error_report(sk);
2275 scm_recv(sock, msg, siocb->scm, flags);
2277 netlink_rcv_wake(sk);
2278 return err ? : copied;
2281 static void netlink_data_ready(struct sock *sk, int len)
2287 * We export these functions to other modules. They provide a
2288 * complete set of kernel non-blocking support for message
2293 __netlink_kernel_create(struct net *net, int unit, struct module *module,
2294 struct netlink_kernel_cfg *cfg)
2296 struct socket *sock;
2298 struct netlink_sock *nlk;
2299 struct listeners *listeners = NULL;
2300 struct mutex *cb_mutex = cfg ? cfg->cb_mutex : NULL;
2301 unsigned int groups;
2305 if (unit < 0 || unit >= MAX_LINKS)
2308 if (sock_create_lite(PF_NETLINK, SOCK_DGRAM, unit, &sock))
2312 * We have to just have a reference on the net from sk, but don't
2313 * get_net it. Besides, we cannot get and then put the net here.
2314 * So we create one inside init_net and the move it to net.
2317 if (__netlink_create(&init_net, sock, cb_mutex, unit) < 0)
2318 goto out_sock_release_nosk;
2321 sk_change_net(sk, net);
2323 if (!cfg || cfg->groups < 32)
2326 groups = cfg->groups;
2328 listeners = kzalloc(sizeof(*listeners) + NLGRPSZ(groups), GFP_KERNEL);
2330 goto out_sock_release;
2332 sk->sk_data_ready = netlink_data_ready;
2333 if (cfg && cfg->input)
2334 nlk_sk(sk)->netlink_rcv = cfg->input;
2336 if (netlink_insert(sk, net, 0))
2337 goto out_sock_release;
2340 nlk->flags |= NETLINK_KERNEL_SOCKET;
2342 netlink_table_grab();
2343 if (!nl_table[unit].registered) {
2344 nl_table[unit].groups = groups;
2345 rcu_assign_pointer(nl_table[unit].listeners, listeners);
2346 nl_table[unit].cb_mutex = cb_mutex;
2347 nl_table[unit].module = module;
2349 nl_table[unit].bind = cfg->bind;
2350 nl_table[unit].flags = cfg->flags;
2352 nl_table[unit].registered = 1;
2355 nl_table[unit].registered++;
2357 netlink_table_ungrab();
2362 netlink_kernel_release(sk);
2365 out_sock_release_nosk:
2369 EXPORT_SYMBOL(__netlink_kernel_create);
2372 netlink_kernel_release(struct sock *sk)
2374 sk_release_kernel(sk);
2376 EXPORT_SYMBOL(netlink_kernel_release);
2378 int __netlink_change_ngroups(struct sock *sk, unsigned int groups)
2380 struct listeners *new, *old;
2381 struct netlink_table *tbl = &nl_table[sk->sk_protocol];
2386 if (NLGRPSZ(tbl->groups) < NLGRPSZ(groups)) {
2387 new = kzalloc(sizeof(*new) + NLGRPSZ(groups), GFP_ATOMIC);
2390 old = nl_deref_protected(tbl->listeners);
2391 memcpy(new->masks, old->masks, NLGRPSZ(tbl->groups));
2392 rcu_assign_pointer(tbl->listeners, new);
2394 kfree_rcu(old, rcu);
2396 tbl->groups = groups;
2402 * netlink_change_ngroups - change number of multicast groups
2404 * This changes the number of multicast groups that are available
2405 * on a certain netlink family. Note that it is not possible to
2406 * change the number of groups to below 32. Also note that it does
2407 * not implicitly call netlink_clear_multicast_users() when the
2408 * number of groups is reduced.
2410 * @sk: The kernel netlink socket, as returned by netlink_kernel_create().
2411 * @groups: The new number of groups.
2413 int netlink_change_ngroups(struct sock *sk, unsigned int groups)
2417 netlink_table_grab();
2418 err = __netlink_change_ngroups(sk, groups);
2419 netlink_table_ungrab();
2424 void __netlink_clear_multicast_users(struct sock *ksk, unsigned int group)
2427 struct netlink_table *tbl = &nl_table[ksk->sk_protocol];
2429 sk_for_each_bound(sk, &tbl->mc_list)
2430 netlink_update_socket_mc(nlk_sk(sk), group, 0);
2434 * netlink_clear_multicast_users - kick off multicast listeners
2436 * This function removes all listeners from the given group.
2437 * @ksk: The kernel netlink socket, as returned by
2438 * netlink_kernel_create().
2439 * @group: The multicast group to clear.
2441 void netlink_clear_multicast_users(struct sock *ksk, unsigned int group)
2443 netlink_table_grab();
2444 __netlink_clear_multicast_users(ksk, group);
2445 netlink_table_ungrab();
2449 __nlmsg_put(struct sk_buff *skb, u32 portid, u32 seq, int type, int len, int flags)
2451 struct nlmsghdr *nlh;
2452 int size = nlmsg_msg_size(len);
2454 nlh = (struct nlmsghdr*)skb_put(skb, NLMSG_ALIGN(size));
2455 nlh->nlmsg_type = type;
2456 nlh->nlmsg_len = size;
2457 nlh->nlmsg_flags = flags;
2458 nlh->nlmsg_pid = portid;
2459 nlh->nlmsg_seq = seq;
2460 if (!__builtin_constant_p(size) || NLMSG_ALIGN(size) - size != 0)
2461 memset(nlmsg_data(nlh) + len, 0, NLMSG_ALIGN(size) - size);
2464 EXPORT_SYMBOL(__nlmsg_put);
2467 * It looks a bit ugly.
2468 * It would be better to create kernel thread.
2471 static int netlink_dump(struct sock *sk)
2473 struct netlink_sock *nlk = nlk_sk(sk);
2474 struct netlink_callback *cb;
2475 struct sk_buff *skb = NULL;
2476 struct nlmsghdr *nlh;
2477 int len, err = -ENOBUFS;
2480 mutex_lock(nlk->cb_mutex);
2488 alloc_size = max_t(int, cb->min_dump_alloc, NLMSG_GOODSIZE);
2490 if (!netlink_rx_is_mmaped(sk) &&
2491 atomic_read(&sk->sk_rmem_alloc) >= sk->sk_rcvbuf)
2493 skb = netlink_alloc_skb(sk, alloc_size, nlk->portid, GFP_KERNEL);
2496 netlink_skb_set_owner_r(skb, sk);
2498 len = cb->dump(skb, cb);
2501 mutex_unlock(nlk->cb_mutex);
2503 if (sk_filter(sk, skb))
2506 __netlink_sendskb(sk, skb);
2510 nlh = nlmsg_put_answer(skb, cb, NLMSG_DONE, sizeof(len), NLM_F_MULTI);
2514 nl_dump_check_consistent(cb, nlh);
2516 memcpy(nlmsg_data(nlh), &len, sizeof(len));
2518 if (sk_filter(sk, skb))
2521 __netlink_sendskb(sk, skb);
2526 mutex_unlock(nlk->cb_mutex);
2528 module_put(cb->module);
2529 netlink_consume_callback(cb);
2533 mutex_unlock(nlk->cb_mutex);
2538 int __netlink_dump_start(struct sock *ssk, struct sk_buff *skb,
2539 const struct nlmsghdr *nlh,
2540 struct netlink_dump_control *control)
2542 struct netlink_callback *cb;
2544 struct netlink_sock *nlk;
2547 cb = kzalloc(sizeof(*cb), GFP_KERNEL);
2551 /* Memory mapped dump requests need to be copied to avoid looping
2552 * on the pending state in netlink_mmap_sendmsg() while the CB hold
2553 * a reference to the skb.
2555 if (netlink_skb_is_mmaped(skb)) {
2556 skb = skb_copy(skb, GFP_KERNEL);
2562 atomic_inc(&skb->users);
2564 cb->dump = control->dump;
2565 cb->done = control->done;
2567 cb->data = control->data;
2568 cb->module = control->module;
2569 cb->min_dump_alloc = control->min_dump_alloc;
2572 sk = netlink_lookup(sock_net(ssk), ssk->sk_protocol, NETLINK_CB(skb).portid);
2574 netlink_destroy_callback(cb);
2575 return -ECONNREFUSED;
2579 mutex_lock(nlk->cb_mutex);
2580 /* A dump is in progress... */
2582 mutex_unlock(nlk->cb_mutex);
2583 netlink_destroy_callback(cb);
2587 /* add reference of module which cb->dump belongs to */
2588 if (!try_module_get(cb->module)) {
2589 mutex_unlock(nlk->cb_mutex);
2590 netlink_destroy_callback(cb);
2591 ret = -EPROTONOSUPPORT;
2596 mutex_unlock(nlk->cb_mutex);
2598 ret = netlink_dump(sk);
2605 /* We successfully started a dump, by returning -EINTR we
2606 * signal not to send ACK even if it was requested.
2610 EXPORT_SYMBOL(__netlink_dump_start);
2612 void netlink_ack(struct sk_buff *in_skb, struct nlmsghdr *nlh, int err)
2614 struct sk_buff *skb;
2615 struct nlmsghdr *rep;
2616 struct nlmsgerr *errmsg;
2617 size_t payload = sizeof(*errmsg);
2619 /* error messages get the original request appened */
2621 payload += nlmsg_len(nlh);
2623 skb = netlink_alloc_skb(in_skb->sk, nlmsg_total_size(payload),
2624 NETLINK_CB(in_skb).portid, GFP_KERNEL);
2628 sk = netlink_lookup(sock_net(in_skb->sk),
2629 in_skb->sk->sk_protocol,
2630 NETLINK_CB(in_skb).portid);
2632 sk->sk_err = ENOBUFS;
2633 sk->sk_error_report(sk);
2639 rep = __nlmsg_put(skb, NETLINK_CB(in_skb).portid, nlh->nlmsg_seq,
2640 NLMSG_ERROR, payload, 0);
2641 errmsg = nlmsg_data(rep);
2642 errmsg->error = err;
2643 memcpy(&errmsg->msg, nlh, err ? nlh->nlmsg_len : sizeof(*nlh));
2644 netlink_unicast(in_skb->sk, skb, NETLINK_CB(in_skb).portid, MSG_DONTWAIT);
2646 EXPORT_SYMBOL(netlink_ack);
2648 int netlink_rcv_skb(struct sk_buff *skb, int (*cb)(struct sk_buff *,
2651 struct nlmsghdr *nlh;
2654 while (skb->len >= nlmsg_total_size(0)) {
2657 nlh = nlmsg_hdr(skb);
2660 if (nlh->nlmsg_len < NLMSG_HDRLEN || skb->len < nlh->nlmsg_len)
2663 /* Only requests are handled by the kernel */
2664 if (!(nlh->nlmsg_flags & NLM_F_REQUEST))
2667 /* Skip control messages */
2668 if (nlh->nlmsg_type < NLMSG_MIN_TYPE)
2676 if (nlh->nlmsg_flags & NLM_F_ACK || err)
2677 netlink_ack(skb, nlh, err);
2680 msglen = NLMSG_ALIGN(nlh->nlmsg_len);
2681 if (msglen > skb->len)
2683 skb_pull(skb, msglen);
2688 EXPORT_SYMBOL(netlink_rcv_skb);
2691 * nlmsg_notify - send a notification netlink message
2692 * @sk: netlink socket to use
2693 * @skb: notification message
2694 * @portid: destination netlink portid for reports or 0
2695 * @group: destination multicast group or 0
2696 * @report: 1 to report back, 0 to disable
2697 * @flags: allocation flags
2699 int nlmsg_notify(struct sock *sk, struct sk_buff *skb, u32 portid,
2700 unsigned int group, int report, gfp_t flags)
2705 int exclude_portid = 0;
2708 atomic_inc(&skb->users);
2709 exclude_portid = portid;
2712 /* errors reported via destination sk->sk_err, but propagate
2713 * delivery errors if NETLINK_BROADCAST_ERROR flag is set */
2714 err = nlmsg_multicast(sk, skb, exclude_portid, group, flags);
2720 err2 = nlmsg_unicast(sk, skb, portid);
2721 if (!err || err == -ESRCH)
2727 EXPORT_SYMBOL(nlmsg_notify);
2729 #ifdef CONFIG_PROC_FS
2730 struct nl_seq_iter {
2731 struct seq_net_private p;
2736 static struct sock *netlink_seq_socket_idx(struct seq_file *seq, loff_t pos)
2738 struct nl_seq_iter *iter = seq->private;
2743 for (i = 0; i < MAX_LINKS; i++) {
2744 struct nl_portid_hash *hash = &nl_table[i].hash;
2746 for (j = 0; j <= hash->mask; j++) {
2747 sk_for_each(s, &hash->table[j]) {
2748 if (sock_net(s) != seq_file_net(seq))
2762 static void *netlink_seq_start(struct seq_file *seq, loff_t *pos)
2763 __acquires(nl_table_lock)
2765 read_lock(&nl_table_lock);
2766 return *pos ? netlink_seq_socket_idx(seq, *pos - 1) : SEQ_START_TOKEN;
2769 static void *netlink_seq_next(struct seq_file *seq, void *v, loff_t *pos)
2772 struct nl_seq_iter *iter;
2777 if (v == SEQ_START_TOKEN)
2778 return netlink_seq_socket_idx(seq, 0);
2780 iter = seq->private;
2784 } while (s && sock_net(s) != seq_file_net(seq));
2789 j = iter->hash_idx + 1;
2792 struct nl_portid_hash *hash = &nl_table[i].hash;
2794 for (; j <= hash->mask; j++) {
2795 s = sk_head(&hash->table[j]);
2796 while (s && sock_net(s) != seq_file_net(seq))
2806 } while (++i < MAX_LINKS);
2811 static void netlink_seq_stop(struct seq_file *seq, void *v)
2812 __releases(nl_table_lock)
2814 read_unlock(&nl_table_lock);
2818 static int netlink_seq_show(struct seq_file *seq, void *v)
2820 if (v == SEQ_START_TOKEN) {
2822 "sk Eth Pid Groups "
2823 "Rmem Wmem Dump Locks Drops Inode\n");
2826 struct netlink_sock *nlk = nlk_sk(s);
2828 seq_printf(seq, "%pK %-3d %-6u %08x %-8d %-8d %pK %-8d %-8d %-8lu\n",
2832 nlk->groups ? (u32)nlk->groups[0] : 0,
2833 sk_rmem_alloc_get(s),
2834 sk_wmem_alloc_get(s),
2836 atomic_read(&s->sk_refcnt),
2837 atomic_read(&s->sk_drops),
2845 static const struct seq_operations netlink_seq_ops = {
2846 .start = netlink_seq_start,
2847 .next = netlink_seq_next,
2848 .stop = netlink_seq_stop,
2849 .show = netlink_seq_show,
2853 static int netlink_seq_open(struct inode *inode, struct file *file)
2855 return seq_open_net(inode, file, &netlink_seq_ops,
2856 sizeof(struct nl_seq_iter));
2859 static const struct file_operations netlink_seq_fops = {
2860 .owner = THIS_MODULE,
2861 .open = netlink_seq_open,
2863 .llseek = seq_lseek,
2864 .release = seq_release_net,
2869 int netlink_register_notifier(struct notifier_block *nb)
2871 return atomic_notifier_chain_register(&netlink_chain, nb);
2873 EXPORT_SYMBOL(netlink_register_notifier);
2875 int netlink_unregister_notifier(struct notifier_block *nb)
2877 return atomic_notifier_chain_unregister(&netlink_chain, nb);
2879 EXPORT_SYMBOL(netlink_unregister_notifier);
2881 static const struct proto_ops netlink_ops = {
2882 .family = PF_NETLINK,
2883 .owner = THIS_MODULE,
2884 .release = netlink_release,
2885 .bind = netlink_bind,
2886 .connect = netlink_connect,
2887 .socketpair = sock_no_socketpair,
2888 .accept = sock_no_accept,
2889 .getname = netlink_getname,
2890 .poll = netlink_poll,
2891 .ioctl = sock_no_ioctl,
2892 .listen = sock_no_listen,
2893 .shutdown = sock_no_shutdown,
2894 .setsockopt = netlink_setsockopt,
2895 .getsockopt = netlink_getsockopt,
2896 .sendmsg = netlink_sendmsg,
2897 .recvmsg = netlink_recvmsg,
2898 .mmap = netlink_mmap,
2899 .sendpage = sock_no_sendpage,
2902 static const struct net_proto_family netlink_family_ops = {
2903 .family = PF_NETLINK,
2904 .create = netlink_create,
2905 .owner = THIS_MODULE, /* for consistency 8) */
2908 static int __net_init netlink_net_init(struct net *net)
2910 #ifdef CONFIG_PROC_FS
2911 if (!proc_create("netlink", 0, net->proc_net, &netlink_seq_fops))
2917 static void __net_exit netlink_net_exit(struct net *net)
2919 #ifdef CONFIG_PROC_FS
2920 remove_proc_entry("netlink", net->proc_net);
2924 static void __init netlink_add_usersock_entry(void)
2926 struct listeners *listeners;
2929 listeners = kzalloc(sizeof(*listeners) + NLGRPSZ(groups), GFP_KERNEL);
2931 panic("netlink_add_usersock_entry: Cannot allocate listeners\n");
2933 netlink_table_grab();
2935 nl_table[NETLINK_USERSOCK].groups = groups;
2936 rcu_assign_pointer(nl_table[NETLINK_USERSOCK].listeners, listeners);
2937 nl_table[NETLINK_USERSOCK].module = THIS_MODULE;
2938 nl_table[NETLINK_USERSOCK].registered = 1;
2939 nl_table[NETLINK_USERSOCK].flags = NL_CFG_F_NONROOT_SEND;
2941 netlink_table_ungrab();
2944 static struct pernet_operations __net_initdata netlink_net_ops = {
2945 .init = netlink_net_init,
2946 .exit = netlink_net_exit,
2949 static int __init netlink_proto_init(void)
2952 unsigned long limit;
2954 int err = proto_register(&netlink_proto, 0);
2959 BUILD_BUG_ON(sizeof(struct netlink_skb_parms) > FIELD_SIZEOF(struct sk_buff, cb));
2961 nl_table = kcalloc(MAX_LINKS, sizeof(*nl_table), GFP_KERNEL);
2965 if (totalram_pages >= (128 * 1024))
2966 limit = totalram_pages >> (21 - PAGE_SHIFT);
2968 limit = totalram_pages >> (23 - PAGE_SHIFT);
2970 order = get_bitmask_order(limit) - 1 + PAGE_SHIFT;
2971 limit = (1UL << order) / sizeof(struct hlist_head);
2972 order = get_bitmask_order(min(limit, (unsigned long)UINT_MAX)) - 1;
2974 for (i = 0; i < MAX_LINKS; i++) {
2975 struct nl_portid_hash *hash = &nl_table[i].hash;
2977 hash->table = nl_portid_hash_zalloc(1 * sizeof(*hash->table));
2980 nl_portid_hash_free(nl_table[i].hash.table,
2981 1 * sizeof(*hash->table));
2985 hash->max_shift = order;
2988 hash->rehash_time = jiffies;
2991 netlink_add_usersock_entry();
2993 sock_register(&netlink_family_ops);
2994 register_pernet_subsys(&netlink_net_ops);
2995 /* The netlink device handler may be needed early. */
3000 panic("netlink_init: Cannot allocate nl_table\n");
3003 core_initcall(netlink_proto_init);