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);
753 if (is_vmalloc_addr(skb->head)) {
761 static void netlink_skb_set_owner_r(struct sk_buff *skb, struct sock *sk)
763 WARN_ON(skb->sk != NULL);
765 skb->destructor = netlink_skb_destructor;
766 atomic_add(skb->truesize, &sk->sk_rmem_alloc);
767 sk_mem_charge(sk, skb->truesize);
770 static void netlink_sock_destruct(struct sock *sk)
772 struct netlink_sock *nlk = nlk_sk(sk);
776 nlk->cb->done(nlk->cb);
778 module_put(nlk->cb->module);
779 netlink_destroy_callback(nlk->cb);
782 skb_queue_purge(&sk->sk_receive_queue);
783 #ifdef CONFIG_NETLINK_MMAP
785 struct nl_mmap_req req;
787 memset(&req, 0, sizeof(req));
788 if (nlk->rx_ring.pg_vec)
789 netlink_set_ring(sk, &req, true, false);
790 memset(&req, 0, sizeof(req));
791 if (nlk->tx_ring.pg_vec)
792 netlink_set_ring(sk, &req, true, true);
794 #endif /* CONFIG_NETLINK_MMAP */
796 if (!sock_flag(sk, SOCK_DEAD)) {
797 printk(KERN_ERR "Freeing alive netlink socket %p\n", sk);
801 WARN_ON(atomic_read(&sk->sk_rmem_alloc));
802 WARN_ON(atomic_read(&sk->sk_wmem_alloc));
803 WARN_ON(nlk_sk(sk)->groups);
806 /* This lock without WQ_FLAG_EXCLUSIVE is good on UP and it is _very_ bad on
807 * SMP. Look, when several writers sleep and reader wakes them up, all but one
808 * immediately hit write lock and grab all the cpus. Exclusive sleep solves
809 * this, _but_ remember, it adds useless work on UP machines.
812 void netlink_table_grab(void)
813 __acquires(nl_table_lock)
817 write_lock_irq(&nl_table_lock);
819 if (atomic_read(&nl_table_users)) {
820 DECLARE_WAITQUEUE(wait, current);
822 add_wait_queue_exclusive(&nl_table_wait, &wait);
824 set_current_state(TASK_UNINTERRUPTIBLE);
825 if (atomic_read(&nl_table_users) == 0)
827 write_unlock_irq(&nl_table_lock);
829 write_lock_irq(&nl_table_lock);
832 __set_current_state(TASK_RUNNING);
833 remove_wait_queue(&nl_table_wait, &wait);
837 void netlink_table_ungrab(void)
838 __releases(nl_table_lock)
840 write_unlock_irq(&nl_table_lock);
841 wake_up(&nl_table_wait);
845 netlink_lock_table(void)
847 /* read_lock() synchronizes us to netlink_table_grab */
849 read_lock(&nl_table_lock);
850 atomic_inc(&nl_table_users);
851 read_unlock(&nl_table_lock);
855 netlink_unlock_table(void)
857 if (atomic_dec_and_test(&nl_table_users))
858 wake_up(&nl_table_wait);
861 static bool netlink_compare(struct net *net, struct sock *sk)
863 return net_eq(sock_net(sk), net);
866 static struct sock *netlink_lookup(struct net *net, int protocol, u32 portid)
868 struct netlink_table *table = &nl_table[protocol];
869 struct nl_portid_hash *hash = &table->hash;
870 struct hlist_head *head;
873 read_lock(&nl_table_lock);
874 head = nl_portid_hashfn(hash, portid);
875 sk_for_each(sk, head) {
876 if (table->compare(net, sk) &&
877 (nlk_sk(sk)->portid == portid)) {
884 read_unlock(&nl_table_lock);
888 static struct hlist_head *nl_portid_hash_zalloc(size_t size)
890 if (size <= PAGE_SIZE)
891 return kzalloc(size, GFP_ATOMIC);
893 return (struct hlist_head *)
894 __get_free_pages(GFP_ATOMIC | __GFP_ZERO,
898 static void nl_portid_hash_free(struct hlist_head *table, size_t size)
900 if (size <= PAGE_SIZE)
903 free_pages((unsigned long)table, get_order(size));
906 static int nl_portid_hash_rehash(struct nl_portid_hash *hash, int grow)
908 unsigned int omask, mask, shift;
910 struct hlist_head *otable, *table;
913 omask = mask = hash->mask;
914 osize = size = (mask + 1) * sizeof(*table);
918 if (++shift > hash->max_shift)
924 table = nl_portid_hash_zalloc(size);
928 otable = hash->table;
932 get_random_bytes(&hash->rnd, sizeof(hash->rnd));
934 for (i = 0; i <= omask; i++) {
936 struct hlist_node *tmp;
938 sk_for_each_safe(sk, tmp, &otable[i])
939 __sk_add_node(sk, nl_portid_hashfn(hash, nlk_sk(sk)->portid));
942 nl_portid_hash_free(otable, osize);
943 hash->rehash_time = jiffies + 10 * 60 * HZ;
947 static inline int nl_portid_hash_dilute(struct nl_portid_hash *hash, int len)
949 int avg = hash->entries >> hash->shift;
951 if (unlikely(avg > 1) && nl_portid_hash_rehash(hash, 1))
954 if (unlikely(len > avg) && time_after(jiffies, hash->rehash_time)) {
955 nl_portid_hash_rehash(hash, 0);
962 static const struct proto_ops netlink_ops;
965 netlink_update_listeners(struct sock *sk)
967 struct netlink_table *tbl = &nl_table[sk->sk_protocol];
970 struct listeners *listeners;
972 listeners = nl_deref_protected(tbl->listeners);
976 for (i = 0; i < NLGRPLONGS(tbl->groups); i++) {
978 sk_for_each_bound(sk, &tbl->mc_list) {
979 if (i < NLGRPLONGS(nlk_sk(sk)->ngroups))
980 mask |= nlk_sk(sk)->groups[i];
982 listeners->masks[i] = mask;
984 /* this function is only called with the netlink table "grabbed", which
985 * makes sure updates are visible before bind or setsockopt return. */
988 static int netlink_insert(struct sock *sk, struct net *net, u32 portid)
990 struct netlink_table *table = &nl_table[sk->sk_protocol];
991 struct nl_portid_hash *hash = &table->hash;
992 struct hlist_head *head;
993 int err = -EADDRINUSE;
997 netlink_table_grab();
998 head = nl_portid_hashfn(hash, portid);
1000 sk_for_each(osk, head) {
1001 if (table->compare(net, osk) &&
1002 (nlk_sk(osk)->portid == portid))
1010 if (nlk_sk(sk)->portid)
1014 if (BITS_PER_LONG > 32 && unlikely(hash->entries >= UINT_MAX))
1017 if (len && nl_portid_hash_dilute(hash, len))
1018 head = nl_portid_hashfn(hash, portid);
1020 nlk_sk(sk)->portid = portid;
1021 sk_add_node(sk, head);
1025 netlink_table_ungrab();
1029 static void netlink_remove(struct sock *sk)
1031 netlink_table_grab();
1032 if (sk_del_node_init(sk))
1033 nl_table[sk->sk_protocol].hash.entries--;
1034 if (nlk_sk(sk)->subscriptions)
1035 __sk_del_bind_node(sk);
1036 netlink_table_ungrab();
1039 static struct proto netlink_proto = {
1041 .owner = THIS_MODULE,
1042 .obj_size = sizeof(struct netlink_sock),
1045 static int __netlink_create(struct net *net, struct socket *sock,
1046 struct mutex *cb_mutex, int protocol)
1049 struct netlink_sock *nlk;
1051 sock->ops = &netlink_ops;
1053 sk = sk_alloc(net, PF_NETLINK, GFP_KERNEL, &netlink_proto);
1057 sock_init_data(sock, sk);
1061 nlk->cb_mutex = cb_mutex;
1063 nlk->cb_mutex = &nlk->cb_def_mutex;
1064 mutex_init(nlk->cb_mutex);
1066 init_waitqueue_head(&nlk->wait);
1067 #ifdef CONFIG_NETLINK_MMAP
1068 mutex_init(&nlk->pg_vec_lock);
1071 sk->sk_destruct = netlink_sock_destruct;
1072 sk->sk_protocol = protocol;
1076 static int netlink_create(struct net *net, struct socket *sock, int protocol,
1079 struct module *module = NULL;
1080 struct mutex *cb_mutex;
1081 struct netlink_sock *nlk;
1082 void (*bind)(int group);
1085 sock->state = SS_UNCONNECTED;
1087 if (sock->type != SOCK_RAW && sock->type != SOCK_DGRAM)
1088 return -ESOCKTNOSUPPORT;
1090 if (protocol < 0 || protocol >= MAX_LINKS)
1091 return -EPROTONOSUPPORT;
1093 netlink_lock_table();
1094 #ifdef CONFIG_MODULES
1095 if (!nl_table[protocol].registered) {
1096 netlink_unlock_table();
1097 request_module("net-pf-%d-proto-%d", PF_NETLINK, protocol);
1098 netlink_lock_table();
1101 if (nl_table[protocol].registered &&
1102 try_module_get(nl_table[protocol].module))
1103 module = nl_table[protocol].module;
1105 err = -EPROTONOSUPPORT;
1106 cb_mutex = nl_table[protocol].cb_mutex;
1107 bind = nl_table[protocol].bind;
1108 netlink_unlock_table();
1113 err = __netlink_create(net, sock, cb_mutex, protocol);
1118 sock_prot_inuse_add(net, &netlink_proto, 1);
1121 nlk = nlk_sk(sock->sk);
1122 nlk->module = module;
1123 nlk->netlink_bind = bind;
1132 static int netlink_release(struct socket *sock)
1134 struct sock *sk = sock->sk;
1135 struct netlink_sock *nlk;
1145 * OK. Socket is unlinked, any packets that arrive now
1150 wake_up_interruptible_all(&nlk->wait);
1152 skb_queue_purge(&sk->sk_write_queue);
1155 struct netlink_notify n = {
1156 .net = sock_net(sk),
1157 .protocol = sk->sk_protocol,
1158 .portid = nlk->portid,
1160 atomic_notifier_call_chain(&netlink_chain,
1161 NETLINK_URELEASE, &n);
1164 module_put(nlk->module);
1166 netlink_table_grab();
1167 if (netlink_is_kernel(sk)) {
1168 BUG_ON(nl_table[sk->sk_protocol].registered == 0);
1169 if (--nl_table[sk->sk_protocol].registered == 0) {
1170 struct listeners *old;
1172 old = nl_deref_protected(nl_table[sk->sk_protocol].listeners);
1173 RCU_INIT_POINTER(nl_table[sk->sk_protocol].listeners, NULL);
1174 kfree_rcu(old, rcu);
1175 nl_table[sk->sk_protocol].module = NULL;
1176 nl_table[sk->sk_protocol].bind = NULL;
1177 nl_table[sk->sk_protocol].flags = 0;
1178 nl_table[sk->sk_protocol].registered = 0;
1180 } else if (nlk->subscriptions) {
1181 netlink_update_listeners(sk);
1183 netlink_table_ungrab();
1189 sock_prot_inuse_add(sock_net(sk), &netlink_proto, -1);
1195 static int netlink_autobind(struct socket *sock)
1197 struct sock *sk = sock->sk;
1198 struct net *net = sock_net(sk);
1199 struct netlink_table *table = &nl_table[sk->sk_protocol];
1200 struct nl_portid_hash *hash = &table->hash;
1201 struct hlist_head *head;
1203 s32 portid = task_tgid_vnr(current);
1205 static s32 rover = -4097;
1209 netlink_table_grab();
1210 head = nl_portid_hashfn(hash, portid);
1211 sk_for_each(osk, head) {
1212 if (!table->compare(net, osk))
1214 if (nlk_sk(osk)->portid == portid) {
1215 /* Bind collision, search negative portid values. */
1219 netlink_table_ungrab();
1223 netlink_table_ungrab();
1225 err = netlink_insert(sk, net, portid);
1226 if (err == -EADDRINUSE)
1229 /* If 2 threads race to autobind, that is fine. */
1236 static inline int netlink_capable(const struct socket *sock, unsigned int flag)
1238 return (nl_table[sock->sk->sk_protocol].flags & flag) ||
1239 ns_capable(sock_net(sock->sk)->user_ns, CAP_NET_ADMIN);
1243 netlink_update_subscriptions(struct sock *sk, unsigned int subscriptions)
1245 struct netlink_sock *nlk = nlk_sk(sk);
1247 if (nlk->subscriptions && !subscriptions)
1248 __sk_del_bind_node(sk);
1249 else if (!nlk->subscriptions && subscriptions)
1250 sk_add_bind_node(sk, &nl_table[sk->sk_protocol].mc_list);
1251 nlk->subscriptions = subscriptions;
1254 static int netlink_realloc_groups(struct sock *sk)
1256 struct netlink_sock *nlk = nlk_sk(sk);
1257 unsigned int groups;
1258 unsigned long *new_groups;
1261 netlink_table_grab();
1263 groups = nl_table[sk->sk_protocol].groups;
1264 if (!nl_table[sk->sk_protocol].registered) {
1269 if (nlk->ngroups >= groups)
1272 new_groups = krealloc(nlk->groups, NLGRPSZ(groups), GFP_ATOMIC);
1273 if (new_groups == NULL) {
1277 memset((char *)new_groups + NLGRPSZ(nlk->ngroups), 0,
1278 NLGRPSZ(groups) - NLGRPSZ(nlk->ngroups));
1280 nlk->groups = new_groups;
1281 nlk->ngroups = groups;
1283 netlink_table_ungrab();
1287 static int netlink_bind(struct socket *sock, struct sockaddr *addr,
1290 struct sock *sk = sock->sk;
1291 struct net *net = sock_net(sk);
1292 struct netlink_sock *nlk = nlk_sk(sk);
1293 struct sockaddr_nl *nladdr = (struct sockaddr_nl *)addr;
1296 if (addr_len < sizeof(struct sockaddr_nl))
1299 if (nladdr->nl_family != AF_NETLINK)
1302 /* Only superuser is allowed to listen multicasts */
1303 if (nladdr->nl_groups) {
1304 if (!netlink_capable(sock, NL_CFG_F_NONROOT_RECV))
1306 err = netlink_realloc_groups(sk);
1312 if (nladdr->nl_pid != nlk->portid)
1315 err = nladdr->nl_pid ?
1316 netlink_insert(sk, net, nladdr->nl_pid) :
1317 netlink_autobind(sock);
1322 if (!nladdr->nl_groups && (nlk->groups == NULL || !(u32)nlk->groups[0]))
1325 netlink_table_grab();
1326 netlink_update_subscriptions(sk, nlk->subscriptions +
1327 hweight32(nladdr->nl_groups) -
1328 hweight32(nlk->groups[0]));
1329 nlk->groups[0] = (nlk->groups[0] & ~0xffffffffUL) | nladdr->nl_groups;
1330 netlink_update_listeners(sk);
1331 netlink_table_ungrab();
1333 if (nlk->netlink_bind && nlk->groups[0]) {
1336 for (i=0; i<nlk->ngroups; i++) {
1337 if (test_bit(i, nlk->groups))
1338 nlk->netlink_bind(i);
1345 static int netlink_connect(struct socket *sock, struct sockaddr *addr,
1346 int alen, int flags)
1349 struct sock *sk = sock->sk;
1350 struct netlink_sock *nlk = nlk_sk(sk);
1351 struct sockaddr_nl *nladdr = (struct sockaddr_nl *)addr;
1353 if (alen < sizeof(addr->sa_family))
1356 if (addr->sa_family == AF_UNSPEC) {
1357 sk->sk_state = NETLINK_UNCONNECTED;
1358 nlk->dst_portid = 0;
1362 if (addr->sa_family != AF_NETLINK)
1365 /* Only superuser is allowed to send multicasts */
1366 if (nladdr->nl_groups && !netlink_capable(sock, NL_CFG_F_NONROOT_SEND))
1370 err = netlink_autobind(sock);
1373 sk->sk_state = NETLINK_CONNECTED;
1374 nlk->dst_portid = nladdr->nl_pid;
1375 nlk->dst_group = ffs(nladdr->nl_groups);
1381 static int netlink_getname(struct socket *sock, struct sockaddr *addr,
1382 int *addr_len, int peer)
1384 struct sock *sk = sock->sk;
1385 struct netlink_sock *nlk = nlk_sk(sk);
1386 DECLARE_SOCKADDR(struct sockaddr_nl *, nladdr, addr);
1388 nladdr->nl_family = AF_NETLINK;
1390 *addr_len = sizeof(*nladdr);
1393 nladdr->nl_pid = nlk->dst_portid;
1394 nladdr->nl_groups = netlink_group_mask(nlk->dst_group);
1396 nladdr->nl_pid = nlk->portid;
1397 nladdr->nl_groups = nlk->groups ? nlk->groups[0] : 0;
1402 static struct sock *netlink_getsockbyportid(struct sock *ssk, u32 portid)
1405 struct netlink_sock *nlk;
1407 sock = netlink_lookup(sock_net(ssk), ssk->sk_protocol, portid);
1409 return ERR_PTR(-ECONNREFUSED);
1411 /* Don't bother queuing skb if kernel socket has no input function */
1413 if (sock->sk_state == NETLINK_CONNECTED &&
1414 nlk->dst_portid != nlk_sk(ssk)->portid) {
1416 return ERR_PTR(-ECONNREFUSED);
1421 struct sock *netlink_getsockbyfilp(struct file *filp)
1423 struct inode *inode = file_inode(filp);
1426 if (!S_ISSOCK(inode->i_mode))
1427 return ERR_PTR(-ENOTSOCK);
1429 sock = SOCKET_I(inode)->sk;
1430 if (sock->sk_family != AF_NETLINK)
1431 return ERR_PTR(-EINVAL);
1437 static struct sk_buff *netlink_alloc_large_skb(unsigned int size)
1439 struct sk_buff *skb;
1442 if (size <= NLMSG_GOODSIZE)
1443 return alloc_skb(size, GFP_KERNEL);
1445 skb = alloc_skb_head(GFP_KERNEL);
1449 data = vmalloc(size);
1455 skb_reset_tail_pointer(skb);
1456 skb->end = skb->tail + size;
1458 skb->destructor = netlink_skb_destructor;
1467 * Attach a skb to a netlink socket.
1468 * The caller must hold a reference to the destination socket. On error, the
1469 * reference is dropped. The skb is not send to the destination, just all
1470 * all error checks are performed and memory in the queue is reserved.
1472 * < 0: error. skb freed, reference to sock dropped.
1474 * 1: repeat lookup - reference dropped while waiting for socket memory.
1476 int netlink_attachskb(struct sock *sk, struct sk_buff *skb,
1477 long *timeo, struct sock *ssk)
1479 struct netlink_sock *nlk;
1483 if ((atomic_read(&sk->sk_rmem_alloc) > sk->sk_rcvbuf ||
1484 test_bit(NETLINK_CONGESTED, &nlk->state)) &&
1485 !netlink_skb_is_mmaped(skb)) {
1486 DECLARE_WAITQUEUE(wait, current);
1488 if (!ssk || netlink_is_kernel(ssk))
1489 netlink_overrun(sk);
1495 __set_current_state(TASK_INTERRUPTIBLE);
1496 add_wait_queue(&nlk->wait, &wait);
1498 if ((atomic_read(&sk->sk_rmem_alloc) > sk->sk_rcvbuf ||
1499 test_bit(NETLINK_CONGESTED, &nlk->state)) &&
1500 !sock_flag(sk, SOCK_DEAD))
1501 *timeo = schedule_timeout(*timeo);
1503 __set_current_state(TASK_RUNNING);
1504 remove_wait_queue(&nlk->wait, &wait);
1507 if (signal_pending(current)) {
1509 return sock_intr_errno(*timeo);
1513 netlink_skb_set_owner_r(skb, sk);
1517 static int __netlink_sendskb(struct sock *sk, struct sk_buff *skb)
1521 #ifdef CONFIG_NETLINK_MMAP
1522 if (netlink_skb_is_mmaped(skb))
1523 netlink_queue_mmaped_skb(sk, skb);
1524 else if (netlink_rx_is_mmaped(sk))
1525 netlink_ring_set_copied(sk, skb);
1527 #endif /* CONFIG_NETLINK_MMAP */
1528 skb_queue_tail(&sk->sk_receive_queue, skb);
1529 sk->sk_data_ready(sk, len);
1533 int netlink_sendskb(struct sock *sk, struct sk_buff *skb)
1535 int len = __netlink_sendskb(sk, skb);
1541 void netlink_detachskb(struct sock *sk, struct sk_buff *skb)
1547 static struct sk_buff *netlink_trim(struct sk_buff *skb, gfp_t allocation)
1551 WARN_ON(skb->sk != NULL);
1552 if (netlink_skb_is_mmaped(skb))
1555 delta = skb->end - skb->tail;
1556 if (is_vmalloc_addr(skb->head) || delta * 2 < skb->truesize)
1559 if (skb_shared(skb)) {
1560 struct sk_buff *nskb = skb_clone(skb, allocation);
1567 if (!pskb_expand_head(skb, 0, -delta, allocation))
1568 skb->truesize -= delta;
1573 static int netlink_unicast_kernel(struct sock *sk, struct sk_buff *skb,
1577 struct netlink_sock *nlk = nlk_sk(sk);
1579 ret = -ECONNREFUSED;
1580 if (nlk->netlink_rcv != NULL) {
1582 netlink_skb_set_owner_r(skb, sk);
1583 NETLINK_CB(skb).sk = ssk;
1584 nlk->netlink_rcv(skb);
1593 int netlink_unicast(struct sock *ssk, struct sk_buff *skb,
1594 u32 portid, int nonblock)
1600 skb = netlink_trim(skb, gfp_any());
1602 timeo = sock_sndtimeo(ssk, nonblock);
1604 sk = netlink_getsockbyportid(ssk, portid);
1609 if (netlink_is_kernel(sk))
1610 return netlink_unicast_kernel(sk, skb, ssk);
1612 if (sk_filter(sk, skb)) {
1619 err = netlink_attachskb(sk, skb, &timeo, ssk);
1625 return netlink_sendskb(sk, skb);
1627 EXPORT_SYMBOL(netlink_unicast);
1629 struct sk_buff *netlink_alloc_skb(struct sock *ssk, unsigned int size,
1630 u32 dst_portid, gfp_t gfp_mask)
1632 #ifdef CONFIG_NETLINK_MMAP
1633 struct sock *sk = NULL;
1634 struct sk_buff *skb;
1635 struct netlink_ring *ring;
1636 struct nl_mmap_hdr *hdr;
1637 unsigned int maxlen;
1639 sk = netlink_getsockbyportid(ssk, dst_portid);
1643 ring = &nlk_sk(sk)->rx_ring;
1644 /* fast-path without atomic ops for common case: non-mmaped receiver */
1645 if (ring->pg_vec == NULL)
1648 skb = alloc_skb_head(gfp_mask);
1652 spin_lock_bh(&sk->sk_receive_queue.lock);
1653 /* check again under lock */
1654 if (ring->pg_vec == NULL)
1657 maxlen = ring->frame_size - NL_MMAP_HDRLEN;
1661 netlink_forward_ring(ring);
1662 hdr = netlink_current_frame(ring, NL_MMAP_STATUS_UNUSED);
1665 netlink_ring_setup_skb(skb, sk, ring, hdr);
1666 netlink_set_status(hdr, NL_MMAP_STATUS_RESERVED);
1667 atomic_inc(&ring->pending);
1668 netlink_increment_head(ring);
1670 spin_unlock_bh(&sk->sk_receive_queue.lock);
1675 spin_unlock_bh(&sk->sk_receive_queue.lock);
1676 netlink_overrun(sk);
1683 spin_unlock_bh(&sk->sk_receive_queue.lock);
1688 return alloc_skb(size, gfp_mask);
1690 EXPORT_SYMBOL_GPL(netlink_alloc_skb);
1692 int netlink_has_listeners(struct sock *sk, unsigned int group)
1695 struct listeners *listeners;
1697 BUG_ON(!netlink_is_kernel(sk));
1700 listeners = rcu_dereference(nl_table[sk->sk_protocol].listeners);
1702 if (listeners && group - 1 < nl_table[sk->sk_protocol].groups)
1703 res = test_bit(group - 1, listeners->masks);
1709 EXPORT_SYMBOL_GPL(netlink_has_listeners);
1711 static int netlink_broadcast_deliver(struct sock *sk, struct sk_buff *skb)
1713 struct netlink_sock *nlk = nlk_sk(sk);
1715 if (atomic_read(&sk->sk_rmem_alloc) <= sk->sk_rcvbuf &&
1716 !test_bit(NETLINK_CONGESTED, &nlk->state)) {
1717 netlink_skb_set_owner_r(skb, sk);
1718 __netlink_sendskb(sk, skb);
1719 return atomic_read(&sk->sk_rmem_alloc) > (sk->sk_rcvbuf >> 1);
1724 struct netlink_broadcast_data {
1725 struct sock *exclude_sk;
1730 int delivery_failure;
1734 struct sk_buff *skb, *skb2;
1735 int (*tx_filter)(struct sock *dsk, struct sk_buff *skb, void *data);
1739 static int do_one_broadcast(struct sock *sk,
1740 struct netlink_broadcast_data *p)
1742 struct netlink_sock *nlk = nlk_sk(sk);
1745 if (p->exclude_sk == sk)
1748 if (nlk->portid == p->portid || p->group - 1 >= nlk->ngroups ||
1749 !test_bit(p->group - 1, nlk->groups))
1752 if (!net_eq(sock_net(sk), p->net))
1756 netlink_overrun(sk);
1761 if (p->skb2 == NULL) {
1762 if (skb_shared(p->skb)) {
1763 p->skb2 = skb_clone(p->skb, p->allocation);
1765 p->skb2 = skb_get(p->skb);
1767 * skb ownership may have been set when
1768 * delivered to a previous socket.
1770 skb_orphan(p->skb2);
1773 if (p->skb2 == NULL) {
1774 netlink_overrun(sk);
1775 /* Clone failed. Notify ALL listeners. */
1777 if (nlk->flags & NETLINK_BROADCAST_SEND_ERROR)
1778 p->delivery_failure = 1;
1779 } else if (p->tx_filter && p->tx_filter(sk, p->skb2, p->tx_data)) {
1782 } else if (sk_filter(sk, p->skb2)) {
1785 } else if ((val = netlink_broadcast_deliver(sk, p->skb2)) < 0) {
1786 netlink_overrun(sk);
1787 if (nlk->flags & NETLINK_BROADCAST_SEND_ERROR)
1788 p->delivery_failure = 1;
1790 p->congested |= val;
1800 int netlink_broadcast_filtered(struct sock *ssk, struct sk_buff *skb, u32 portid,
1801 u32 group, gfp_t allocation,
1802 int (*filter)(struct sock *dsk, struct sk_buff *skb, void *data),
1805 struct net *net = sock_net(ssk);
1806 struct netlink_broadcast_data info;
1809 skb = netlink_trim(skb, allocation);
1811 info.exclude_sk = ssk;
1813 info.portid = portid;
1816 info.delivery_failure = 0;
1819 info.allocation = allocation;
1822 info.tx_filter = filter;
1823 info.tx_data = filter_data;
1825 /* While we sleep in clone, do not allow to change socket list */
1827 netlink_lock_table();
1829 sk_for_each_bound(sk, &nl_table[ssk->sk_protocol].mc_list)
1830 do_one_broadcast(sk, &info);
1834 netlink_unlock_table();
1836 if (info.delivery_failure) {
1837 kfree_skb(info.skb2);
1840 consume_skb(info.skb2);
1842 if (info.delivered) {
1843 if (info.congested && (allocation & __GFP_WAIT))
1849 EXPORT_SYMBOL(netlink_broadcast_filtered);
1851 int netlink_broadcast(struct sock *ssk, struct sk_buff *skb, u32 portid,
1852 u32 group, gfp_t allocation)
1854 return netlink_broadcast_filtered(ssk, skb, portid, group, allocation,
1857 EXPORT_SYMBOL(netlink_broadcast);
1859 struct netlink_set_err_data {
1860 struct sock *exclude_sk;
1866 static int do_one_set_err(struct sock *sk, struct netlink_set_err_data *p)
1868 struct netlink_sock *nlk = nlk_sk(sk);
1871 if (sk == p->exclude_sk)
1874 if (!net_eq(sock_net(sk), sock_net(p->exclude_sk)))
1877 if (nlk->portid == p->portid || p->group - 1 >= nlk->ngroups ||
1878 !test_bit(p->group - 1, nlk->groups))
1881 if (p->code == ENOBUFS && nlk->flags & NETLINK_RECV_NO_ENOBUFS) {
1886 sk->sk_err = p->code;
1887 sk->sk_error_report(sk);
1893 * netlink_set_err - report error to broadcast listeners
1894 * @ssk: the kernel netlink socket, as returned by netlink_kernel_create()
1895 * @portid: the PORTID of a process that we want to skip (if any)
1896 * @groups: the broadcast group that will notice the error
1897 * @code: error code, must be negative (as usual in kernelspace)
1899 * This function returns the number of broadcast listeners that have set the
1900 * NETLINK_RECV_NO_ENOBUFS socket option.
1902 int netlink_set_err(struct sock *ssk, u32 portid, u32 group, int code)
1904 struct netlink_set_err_data info;
1908 info.exclude_sk = ssk;
1909 info.portid = portid;
1911 /* sk->sk_err wants a positive error value */
1914 read_lock(&nl_table_lock);
1916 sk_for_each_bound(sk, &nl_table[ssk->sk_protocol].mc_list)
1917 ret += do_one_set_err(sk, &info);
1919 read_unlock(&nl_table_lock);
1922 EXPORT_SYMBOL(netlink_set_err);
1924 /* must be called with netlink table grabbed */
1925 static void netlink_update_socket_mc(struct netlink_sock *nlk,
1929 int old, new = !!is_new, subscriptions;
1931 old = test_bit(group - 1, nlk->groups);
1932 subscriptions = nlk->subscriptions - old + new;
1934 __set_bit(group - 1, nlk->groups);
1936 __clear_bit(group - 1, nlk->groups);
1937 netlink_update_subscriptions(&nlk->sk, subscriptions);
1938 netlink_update_listeners(&nlk->sk);
1941 static int netlink_setsockopt(struct socket *sock, int level, int optname,
1942 char __user *optval, unsigned int optlen)
1944 struct sock *sk = sock->sk;
1945 struct netlink_sock *nlk = nlk_sk(sk);
1946 unsigned int val = 0;
1949 if (level != SOL_NETLINK)
1950 return -ENOPROTOOPT;
1952 if (optname != NETLINK_RX_RING && optname != NETLINK_TX_RING &&
1953 optlen >= sizeof(int) &&
1954 get_user(val, (unsigned int __user *)optval))
1958 case NETLINK_PKTINFO:
1960 nlk->flags |= NETLINK_RECV_PKTINFO;
1962 nlk->flags &= ~NETLINK_RECV_PKTINFO;
1965 case NETLINK_ADD_MEMBERSHIP:
1966 case NETLINK_DROP_MEMBERSHIP: {
1967 if (!netlink_capable(sock, NL_CFG_F_NONROOT_RECV))
1969 err = netlink_realloc_groups(sk);
1972 if (!val || val - 1 >= nlk->ngroups)
1974 netlink_table_grab();
1975 netlink_update_socket_mc(nlk, val,
1976 optname == NETLINK_ADD_MEMBERSHIP);
1977 netlink_table_ungrab();
1979 if (nlk->netlink_bind)
1980 nlk->netlink_bind(val);
1985 case NETLINK_BROADCAST_ERROR:
1987 nlk->flags |= NETLINK_BROADCAST_SEND_ERROR;
1989 nlk->flags &= ~NETLINK_BROADCAST_SEND_ERROR;
1992 case NETLINK_NO_ENOBUFS:
1994 nlk->flags |= NETLINK_RECV_NO_ENOBUFS;
1995 clear_bit(NETLINK_CONGESTED, &nlk->state);
1996 wake_up_interruptible(&nlk->wait);
1998 nlk->flags &= ~NETLINK_RECV_NO_ENOBUFS;
2002 #ifdef CONFIG_NETLINK_MMAP
2003 case NETLINK_RX_RING:
2004 case NETLINK_TX_RING: {
2005 struct nl_mmap_req req;
2007 /* Rings might consume more memory than queue limits, require
2010 if (!capable(CAP_NET_ADMIN))
2012 if (optlen < sizeof(req))
2014 if (copy_from_user(&req, optval, sizeof(req)))
2016 err = netlink_set_ring(sk, &req, false,
2017 optname == NETLINK_TX_RING);
2020 #endif /* CONFIG_NETLINK_MMAP */
2027 static int netlink_getsockopt(struct socket *sock, int level, int optname,
2028 char __user *optval, int __user *optlen)
2030 struct sock *sk = sock->sk;
2031 struct netlink_sock *nlk = nlk_sk(sk);
2034 if (level != SOL_NETLINK)
2035 return -ENOPROTOOPT;
2037 if (get_user(len, optlen))
2043 case NETLINK_PKTINFO:
2044 if (len < sizeof(int))
2047 val = nlk->flags & NETLINK_RECV_PKTINFO ? 1 : 0;
2048 if (put_user(len, optlen) ||
2049 put_user(val, optval))
2053 case NETLINK_BROADCAST_ERROR:
2054 if (len < sizeof(int))
2057 val = nlk->flags & NETLINK_BROADCAST_SEND_ERROR ? 1 : 0;
2058 if (put_user(len, optlen) ||
2059 put_user(val, optval))
2063 case NETLINK_NO_ENOBUFS:
2064 if (len < sizeof(int))
2067 val = nlk->flags & NETLINK_RECV_NO_ENOBUFS ? 1 : 0;
2068 if (put_user(len, optlen) ||
2069 put_user(val, optval))
2079 static void netlink_cmsg_recv_pktinfo(struct msghdr *msg, struct sk_buff *skb)
2081 struct nl_pktinfo info;
2083 info.group = NETLINK_CB(skb).dst_group;
2084 put_cmsg(msg, SOL_NETLINK, NETLINK_PKTINFO, sizeof(info), &info);
2087 static int netlink_sendmsg(struct kiocb *kiocb, struct socket *sock,
2088 struct msghdr *msg, size_t len)
2090 struct sock_iocb *siocb = kiocb_to_siocb(kiocb);
2091 struct sock *sk = sock->sk;
2092 struct netlink_sock *nlk = nlk_sk(sk);
2093 struct sockaddr_nl *addr = msg->msg_name;
2096 struct sk_buff *skb;
2098 struct scm_cookie scm;
2100 if (msg->msg_flags&MSG_OOB)
2103 if (NULL == siocb->scm)
2106 err = scm_send(sock, msg, siocb->scm, true);
2110 if (msg->msg_namelen) {
2112 if (addr->nl_family != AF_NETLINK)
2114 dst_portid = addr->nl_pid;
2115 dst_group = ffs(addr->nl_groups);
2117 if ((dst_group || dst_portid) &&
2118 !netlink_capable(sock, NL_CFG_F_NONROOT_SEND))
2121 dst_portid = nlk->dst_portid;
2122 dst_group = nlk->dst_group;
2126 err = netlink_autobind(sock);
2131 if (netlink_tx_is_mmaped(sk) &&
2132 msg->msg_iov->iov_base == NULL) {
2133 err = netlink_mmap_sendmsg(sk, msg, dst_portid, dst_group,
2139 if (len > sk->sk_sndbuf - 32)
2142 skb = netlink_alloc_large_skb(len);
2146 NETLINK_CB(skb).portid = nlk->portid;
2147 NETLINK_CB(skb).dst_group = dst_group;
2148 NETLINK_CB(skb).creds = siocb->scm->creds;
2151 if (memcpy_fromiovec(skb_put(skb, len), msg->msg_iov, len)) {
2156 err = security_netlink_send(sk, skb);
2163 atomic_inc(&skb->users);
2164 netlink_broadcast(sk, skb, dst_portid, dst_group, GFP_KERNEL);
2166 err = netlink_unicast(sk, skb, dst_portid, msg->msg_flags&MSG_DONTWAIT);
2169 scm_destroy(siocb->scm);
2173 static int netlink_recvmsg(struct kiocb *kiocb, struct socket *sock,
2174 struct msghdr *msg, size_t len,
2177 struct sock_iocb *siocb = kiocb_to_siocb(kiocb);
2178 struct scm_cookie scm;
2179 struct sock *sk = sock->sk;
2180 struct netlink_sock *nlk = nlk_sk(sk);
2181 int noblock = flags&MSG_DONTWAIT;
2183 struct sk_buff *skb, *data_skb;
2191 skb = skb_recv_datagram(sk, flags, noblock, &err);
2197 #ifdef CONFIG_COMPAT_NETLINK_MESSAGES
2198 if (unlikely(skb_shinfo(skb)->frag_list)) {
2200 * If this skb has a frag_list, then here that means that we
2201 * will have to use the frag_list skb's data for compat tasks
2202 * and the regular skb's data for normal (non-compat) tasks.
2204 * If we need to send the compat skb, assign it to the
2205 * 'data_skb' variable so that it will be used below for data
2206 * copying. We keep 'skb' for everything else, including
2207 * freeing both later.
2209 if (flags & MSG_CMSG_COMPAT)
2210 data_skb = skb_shinfo(skb)->frag_list;
2214 msg->msg_namelen = 0;
2216 copied = data_skb->len;
2218 msg->msg_flags |= MSG_TRUNC;
2222 skb_reset_transport_header(data_skb);
2223 err = skb_copy_datagram_iovec(data_skb, 0, msg->msg_iov, copied);
2225 if (msg->msg_name) {
2226 struct sockaddr_nl *addr = (struct sockaddr_nl *)msg->msg_name;
2227 addr->nl_family = AF_NETLINK;
2229 addr->nl_pid = NETLINK_CB(skb).portid;
2230 addr->nl_groups = netlink_group_mask(NETLINK_CB(skb).dst_group);
2231 msg->msg_namelen = sizeof(*addr);
2234 if (nlk->flags & NETLINK_RECV_PKTINFO)
2235 netlink_cmsg_recv_pktinfo(msg, skb);
2237 if (NULL == siocb->scm) {
2238 memset(&scm, 0, sizeof(scm));
2241 siocb->scm->creds = *NETLINK_CREDS(skb);
2242 if (flags & MSG_TRUNC)
2243 copied = data_skb->len;
2245 skb_free_datagram(sk, skb);
2247 if (nlk->cb && atomic_read(&sk->sk_rmem_alloc) <= sk->sk_rcvbuf / 2) {
2248 ret = netlink_dump(sk);
2251 sk->sk_error_report(sk);
2255 scm_recv(sock, msg, siocb->scm, flags);
2257 netlink_rcv_wake(sk);
2258 return err ? : copied;
2261 static void netlink_data_ready(struct sock *sk, int len)
2267 * We export these functions to other modules. They provide a
2268 * complete set of kernel non-blocking support for message
2273 __netlink_kernel_create(struct net *net, int unit, struct module *module,
2274 struct netlink_kernel_cfg *cfg)
2276 struct socket *sock;
2278 struct netlink_sock *nlk;
2279 struct listeners *listeners = NULL;
2280 struct mutex *cb_mutex = cfg ? cfg->cb_mutex : NULL;
2281 unsigned int groups;
2285 if (unit < 0 || unit >= MAX_LINKS)
2288 if (sock_create_lite(PF_NETLINK, SOCK_DGRAM, unit, &sock))
2292 * We have to just have a reference on the net from sk, but don't
2293 * get_net it. Besides, we cannot get and then put the net here.
2294 * So we create one inside init_net and the move it to net.
2297 if (__netlink_create(&init_net, sock, cb_mutex, unit) < 0)
2298 goto out_sock_release_nosk;
2301 sk_change_net(sk, net);
2303 if (!cfg || cfg->groups < 32)
2306 groups = cfg->groups;
2308 listeners = kzalloc(sizeof(*listeners) + NLGRPSZ(groups), GFP_KERNEL);
2310 goto out_sock_release;
2312 sk->sk_data_ready = netlink_data_ready;
2313 if (cfg && cfg->input)
2314 nlk_sk(sk)->netlink_rcv = cfg->input;
2316 if (netlink_insert(sk, net, 0))
2317 goto out_sock_release;
2320 nlk->flags |= NETLINK_KERNEL_SOCKET;
2322 netlink_table_grab();
2323 if (!nl_table[unit].registered) {
2324 nl_table[unit].groups = groups;
2325 rcu_assign_pointer(nl_table[unit].listeners, listeners);
2326 nl_table[unit].cb_mutex = cb_mutex;
2327 nl_table[unit].module = module;
2329 nl_table[unit].bind = cfg->bind;
2330 nl_table[unit].flags = cfg->flags;
2332 nl_table[unit].compare = cfg->compare;
2334 nl_table[unit].registered = 1;
2337 nl_table[unit].registered++;
2339 netlink_table_ungrab();
2344 netlink_kernel_release(sk);
2347 out_sock_release_nosk:
2351 EXPORT_SYMBOL(__netlink_kernel_create);
2354 netlink_kernel_release(struct sock *sk)
2356 sk_release_kernel(sk);
2358 EXPORT_SYMBOL(netlink_kernel_release);
2360 int __netlink_change_ngroups(struct sock *sk, unsigned int groups)
2362 struct listeners *new, *old;
2363 struct netlink_table *tbl = &nl_table[sk->sk_protocol];
2368 if (NLGRPSZ(tbl->groups) < NLGRPSZ(groups)) {
2369 new = kzalloc(sizeof(*new) + NLGRPSZ(groups), GFP_ATOMIC);
2372 old = nl_deref_protected(tbl->listeners);
2373 memcpy(new->masks, old->masks, NLGRPSZ(tbl->groups));
2374 rcu_assign_pointer(tbl->listeners, new);
2376 kfree_rcu(old, rcu);
2378 tbl->groups = groups;
2384 * netlink_change_ngroups - change number of multicast groups
2386 * This changes the number of multicast groups that are available
2387 * on a certain netlink family. Note that it is not possible to
2388 * change the number of groups to below 32. Also note that it does
2389 * not implicitly call netlink_clear_multicast_users() when the
2390 * number of groups is reduced.
2392 * @sk: The kernel netlink socket, as returned by netlink_kernel_create().
2393 * @groups: The new number of groups.
2395 int netlink_change_ngroups(struct sock *sk, unsigned int groups)
2399 netlink_table_grab();
2400 err = __netlink_change_ngroups(sk, groups);
2401 netlink_table_ungrab();
2406 void __netlink_clear_multicast_users(struct sock *ksk, unsigned int group)
2409 struct netlink_table *tbl = &nl_table[ksk->sk_protocol];
2411 sk_for_each_bound(sk, &tbl->mc_list)
2412 netlink_update_socket_mc(nlk_sk(sk), group, 0);
2416 * netlink_clear_multicast_users - kick off multicast listeners
2418 * This function removes all listeners from the given group.
2419 * @ksk: The kernel netlink socket, as returned by
2420 * netlink_kernel_create().
2421 * @group: The multicast group to clear.
2423 void netlink_clear_multicast_users(struct sock *ksk, unsigned int group)
2425 netlink_table_grab();
2426 __netlink_clear_multicast_users(ksk, group);
2427 netlink_table_ungrab();
2431 __nlmsg_put(struct sk_buff *skb, u32 portid, u32 seq, int type, int len, int flags)
2433 struct nlmsghdr *nlh;
2434 int size = nlmsg_msg_size(len);
2436 nlh = (struct nlmsghdr*)skb_put(skb, NLMSG_ALIGN(size));
2437 nlh->nlmsg_type = type;
2438 nlh->nlmsg_len = size;
2439 nlh->nlmsg_flags = flags;
2440 nlh->nlmsg_pid = portid;
2441 nlh->nlmsg_seq = seq;
2442 if (!__builtin_constant_p(size) || NLMSG_ALIGN(size) - size != 0)
2443 memset(nlmsg_data(nlh) + len, 0, NLMSG_ALIGN(size) - size);
2446 EXPORT_SYMBOL(__nlmsg_put);
2449 * It looks a bit ugly.
2450 * It would be better to create kernel thread.
2453 static int netlink_dump(struct sock *sk)
2455 struct netlink_sock *nlk = nlk_sk(sk);
2456 struct netlink_callback *cb;
2457 struct sk_buff *skb = NULL;
2458 struct nlmsghdr *nlh;
2459 int len, err = -ENOBUFS;
2462 mutex_lock(nlk->cb_mutex);
2470 alloc_size = max_t(int, cb->min_dump_alloc, NLMSG_GOODSIZE);
2472 if (!netlink_rx_is_mmaped(sk) &&
2473 atomic_read(&sk->sk_rmem_alloc) >= sk->sk_rcvbuf)
2475 skb = netlink_alloc_skb(sk, alloc_size, nlk->portid, GFP_KERNEL);
2478 netlink_skb_set_owner_r(skb, sk);
2480 len = cb->dump(skb, cb);
2483 mutex_unlock(nlk->cb_mutex);
2485 if (sk_filter(sk, skb))
2488 __netlink_sendskb(sk, skb);
2492 nlh = nlmsg_put_answer(skb, cb, NLMSG_DONE, sizeof(len), NLM_F_MULTI);
2496 nl_dump_check_consistent(cb, nlh);
2498 memcpy(nlmsg_data(nlh), &len, sizeof(len));
2500 if (sk_filter(sk, skb))
2503 __netlink_sendskb(sk, skb);
2508 mutex_unlock(nlk->cb_mutex);
2510 module_put(cb->module);
2511 netlink_consume_callback(cb);
2515 mutex_unlock(nlk->cb_mutex);
2520 int __netlink_dump_start(struct sock *ssk, struct sk_buff *skb,
2521 const struct nlmsghdr *nlh,
2522 struct netlink_dump_control *control)
2524 struct netlink_callback *cb;
2526 struct netlink_sock *nlk;
2529 cb = kzalloc(sizeof(*cb), GFP_KERNEL);
2533 /* Memory mapped dump requests need to be copied to avoid looping
2534 * on the pending state in netlink_mmap_sendmsg() while the CB hold
2535 * a reference to the skb.
2537 if (netlink_skb_is_mmaped(skb)) {
2538 skb = skb_copy(skb, GFP_KERNEL);
2544 atomic_inc(&skb->users);
2546 cb->dump = control->dump;
2547 cb->done = control->done;
2549 cb->data = control->data;
2550 cb->module = control->module;
2551 cb->min_dump_alloc = control->min_dump_alloc;
2554 sk = netlink_lookup(sock_net(ssk), ssk->sk_protocol, NETLINK_CB(skb).portid);
2556 netlink_destroy_callback(cb);
2557 return -ECONNREFUSED;
2561 mutex_lock(nlk->cb_mutex);
2562 /* A dump is in progress... */
2564 mutex_unlock(nlk->cb_mutex);
2565 netlink_destroy_callback(cb);
2569 /* add reference of module which cb->dump belongs to */
2570 if (!try_module_get(cb->module)) {
2571 mutex_unlock(nlk->cb_mutex);
2572 netlink_destroy_callback(cb);
2573 ret = -EPROTONOSUPPORT;
2578 mutex_unlock(nlk->cb_mutex);
2580 ret = netlink_dump(sk);
2587 /* We successfully started a dump, by returning -EINTR we
2588 * signal not to send ACK even if it was requested.
2592 EXPORT_SYMBOL(__netlink_dump_start);
2594 void netlink_ack(struct sk_buff *in_skb, struct nlmsghdr *nlh, int err)
2596 struct sk_buff *skb;
2597 struct nlmsghdr *rep;
2598 struct nlmsgerr *errmsg;
2599 size_t payload = sizeof(*errmsg);
2601 /* error messages get the original request appened */
2603 payload += nlmsg_len(nlh);
2605 skb = netlink_alloc_skb(in_skb->sk, nlmsg_total_size(payload),
2606 NETLINK_CB(in_skb).portid, GFP_KERNEL);
2610 sk = netlink_lookup(sock_net(in_skb->sk),
2611 in_skb->sk->sk_protocol,
2612 NETLINK_CB(in_skb).portid);
2614 sk->sk_err = ENOBUFS;
2615 sk->sk_error_report(sk);
2621 rep = __nlmsg_put(skb, NETLINK_CB(in_skb).portid, nlh->nlmsg_seq,
2622 NLMSG_ERROR, payload, 0);
2623 errmsg = nlmsg_data(rep);
2624 errmsg->error = err;
2625 memcpy(&errmsg->msg, nlh, err ? nlh->nlmsg_len : sizeof(*nlh));
2626 netlink_unicast(in_skb->sk, skb, NETLINK_CB(in_skb).portid, MSG_DONTWAIT);
2628 EXPORT_SYMBOL(netlink_ack);
2630 int netlink_rcv_skb(struct sk_buff *skb, int (*cb)(struct sk_buff *,
2633 struct nlmsghdr *nlh;
2636 while (skb->len >= nlmsg_total_size(0)) {
2639 nlh = nlmsg_hdr(skb);
2642 if (nlh->nlmsg_len < NLMSG_HDRLEN || skb->len < nlh->nlmsg_len)
2645 /* Only requests are handled by the kernel */
2646 if (!(nlh->nlmsg_flags & NLM_F_REQUEST))
2649 /* Skip control messages */
2650 if (nlh->nlmsg_type < NLMSG_MIN_TYPE)
2658 if (nlh->nlmsg_flags & NLM_F_ACK || err)
2659 netlink_ack(skb, nlh, err);
2662 msglen = NLMSG_ALIGN(nlh->nlmsg_len);
2663 if (msglen > skb->len)
2665 skb_pull(skb, msglen);
2670 EXPORT_SYMBOL(netlink_rcv_skb);
2673 * nlmsg_notify - send a notification netlink message
2674 * @sk: netlink socket to use
2675 * @skb: notification message
2676 * @portid: destination netlink portid for reports or 0
2677 * @group: destination multicast group or 0
2678 * @report: 1 to report back, 0 to disable
2679 * @flags: allocation flags
2681 int nlmsg_notify(struct sock *sk, struct sk_buff *skb, u32 portid,
2682 unsigned int group, int report, gfp_t flags)
2687 int exclude_portid = 0;
2690 atomic_inc(&skb->users);
2691 exclude_portid = portid;
2694 /* errors reported via destination sk->sk_err, but propagate
2695 * delivery errors if NETLINK_BROADCAST_ERROR flag is set */
2696 err = nlmsg_multicast(sk, skb, exclude_portid, group, flags);
2702 err2 = nlmsg_unicast(sk, skb, portid);
2703 if (!err || err == -ESRCH)
2709 EXPORT_SYMBOL(nlmsg_notify);
2711 #ifdef CONFIG_PROC_FS
2712 struct nl_seq_iter {
2713 struct seq_net_private p;
2718 static struct sock *netlink_seq_socket_idx(struct seq_file *seq, loff_t pos)
2720 struct nl_seq_iter *iter = seq->private;
2725 for (i = 0; i < MAX_LINKS; i++) {
2726 struct nl_portid_hash *hash = &nl_table[i].hash;
2728 for (j = 0; j <= hash->mask; j++) {
2729 sk_for_each(s, &hash->table[j]) {
2730 if (sock_net(s) != seq_file_net(seq))
2744 static void *netlink_seq_start(struct seq_file *seq, loff_t *pos)
2745 __acquires(nl_table_lock)
2747 read_lock(&nl_table_lock);
2748 return *pos ? netlink_seq_socket_idx(seq, *pos - 1) : SEQ_START_TOKEN;
2751 static void *netlink_seq_next(struct seq_file *seq, void *v, loff_t *pos)
2754 struct nl_seq_iter *iter;
2760 if (v == SEQ_START_TOKEN)
2761 return netlink_seq_socket_idx(seq, 0);
2763 net = seq_file_net(seq);
2764 iter = seq->private;
2768 } while (s && !nl_table[s->sk_protocol].compare(net, s));
2773 j = iter->hash_idx + 1;
2776 struct nl_portid_hash *hash = &nl_table[i].hash;
2778 for (; j <= hash->mask; j++) {
2779 s = sk_head(&hash->table[j]);
2781 while (s && !nl_table[s->sk_protocol].compare(net, s))
2791 } while (++i < MAX_LINKS);
2796 static void netlink_seq_stop(struct seq_file *seq, void *v)
2797 __releases(nl_table_lock)
2799 read_unlock(&nl_table_lock);
2803 static int netlink_seq_show(struct seq_file *seq, void *v)
2805 if (v == SEQ_START_TOKEN) {
2807 "sk Eth Pid Groups "
2808 "Rmem Wmem Dump Locks Drops Inode\n");
2811 struct netlink_sock *nlk = nlk_sk(s);
2813 seq_printf(seq, "%pK %-3d %-6u %08x %-8d %-8d %pK %-8d %-8d %-8lu\n",
2817 nlk->groups ? (u32)nlk->groups[0] : 0,
2818 sk_rmem_alloc_get(s),
2819 sk_wmem_alloc_get(s),
2821 atomic_read(&s->sk_refcnt),
2822 atomic_read(&s->sk_drops),
2830 static const struct seq_operations netlink_seq_ops = {
2831 .start = netlink_seq_start,
2832 .next = netlink_seq_next,
2833 .stop = netlink_seq_stop,
2834 .show = netlink_seq_show,
2838 static int netlink_seq_open(struct inode *inode, struct file *file)
2840 return seq_open_net(inode, file, &netlink_seq_ops,
2841 sizeof(struct nl_seq_iter));
2844 static const struct file_operations netlink_seq_fops = {
2845 .owner = THIS_MODULE,
2846 .open = netlink_seq_open,
2848 .llseek = seq_lseek,
2849 .release = seq_release_net,
2854 int netlink_register_notifier(struct notifier_block *nb)
2856 return atomic_notifier_chain_register(&netlink_chain, nb);
2858 EXPORT_SYMBOL(netlink_register_notifier);
2860 int netlink_unregister_notifier(struct notifier_block *nb)
2862 return atomic_notifier_chain_unregister(&netlink_chain, nb);
2864 EXPORT_SYMBOL(netlink_unregister_notifier);
2866 static const struct proto_ops netlink_ops = {
2867 .family = PF_NETLINK,
2868 .owner = THIS_MODULE,
2869 .release = netlink_release,
2870 .bind = netlink_bind,
2871 .connect = netlink_connect,
2872 .socketpair = sock_no_socketpair,
2873 .accept = sock_no_accept,
2874 .getname = netlink_getname,
2875 .poll = netlink_poll,
2876 .ioctl = sock_no_ioctl,
2877 .listen = sock_no_listen,
2878 .shutdown = sock_no_shutdown,
2879 .setsockopt = netlink_setsockopt,
2880 .getsockopt = netlink_getsockopt,
2881 .sendmsg = netlink_sendmsg,
2882 .recvmsg = netlink_recvmsg,
2883 .mmap = netlink_mmap,
2884 .sendpage = sock_no_sendpage,
2887 static const struct net_proto_family netlink_family_ops = {
2888 .family = PF_NETLINK,
2889 .create = netlink_create,
2890 .owner = THIS_MODULE, /* for consistency 8) */
2893 static int __net_init netlink_net_init(struct net *net)
2895 #ifdef CONFIG_PROC_FS
2896 if (!proc_create("netlink", 0, net->proc_net, &netlink_seq_fops))
2902 static void __net_exit netlink_net_exit(struct net *net)
2904 #ifdef CONFIG_PROC_FS
2905 remove_proc_entry("netlink", net->proc_net);
2909 static void __init netlink_add_usersock_entry(void)
2911 struct listeners *listeners;
2914 listeners = kzalloc(sizeof(*listeners) + NLGRPSZ(groups), GFP_KERNEL);
2916 panic("netlink_add_usersock_entry: Cannot allocate listeners\n");
2918 netlink_table_grab();
2920 nl_table[NETLINK_USERSOCK].groups = groups;
2921 rcu_assign_pointer(nl_table[NETLINK_USERSOCK].listeners, listeners);
2922 nl_table[NETLINK_USERSOCK].module = THIS_MODULE;
2923 nl_table[NETLINK_USERSOCK].registered = 1;
2924 nl_table[NETLINK_USERSOCK].flags = NL_CFG_F_NONROOT_SEND;
2926 netlink_table_ungrab();
2929 static struct pernet_operations __net_initdata netlink_net_ops = {
2930 .init = netlink_net_init,
2931 .exit = netlink_net_exit,
2934 static int __init netlink_proto_init(void)
2937 unsigned long limit;
2939 int err = proto_register(&netlink_proto, 0);
2944 BUILD_BUG_ON(sizeof(struct netlink_skb_parms) > FIELD_SIZEOF(struct sk_buff, cb));
2946 nl_table = kcalloc(MAX_LINKS, sizeof(*nl_table), GFP_KERNEL);
2950 if (totalram_pages >= (128 * 1024))
2951 limit = totalram_pages >> (21 - PAGE_SHIFT);
2953 limit = totalram_pages >> (23 - PAGE_SHIFT);
2955 order = get_bitmask_order(limit) - 1 + PAGE_SHIFT;
2956 limit = (1UL << order) / sizeof(struct hlist_head);
2957 order = get_bitmask_order(min(limit, (unsigned long)UINT_MAX)) - 1;
2959 for (i = 0; i < MAX_LINKS; i++) {
2960 struct nl_portid_hash *hash = &nl_table[i].hash;
2962 hash->table = nl_portid_hash_zalloc(1 * sizeof(*hash->table));
2965 nl_portid_hash_free(nl_table[i].hash.table,
2966 1 * sizeof(*hash->table));
2970 hash->max_shift = order;
2973 hash->rehash_time = jiffies;
2975 nl_table[i].compare = netlink_compare;
2978 netlink_add_usersock_entry();
2980 sock_register(&netlink_family_ops);
2981 register_pernet_subsys(&netlink_net_ops);
2982 /* The netlink device handler may be needed early. */
2987 panic("netlink_init: Cannot allocate nl_table\n");
2990 core_initcall(netlink_proto_init);