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 struct sock *netlink_lookup(struct net *net, int protocol, u32 portid)
863 struct nl_portid_hash *hash = &nl_table[protocol].hash;
864 struct hlist_head *head;
867 read_lock(&nl_table_lock);
868 head = nl_portid_hashfn(hash, portid);
869 sk_for_each(sk, head) {
870 if (net_eq(sock_net(sk), net) && (nlk_sk(sk)->portid == portid)) {
877 read_unlock(&nl_table_lock);
881 static struct hlist_head *nl_portid_hash_zalloc(size_t size)
883 if (size <= PAGE_SIZE)
884 return kzalloc(size, GFP_ATOMIC);
886 return (struct hlist_head *)
887 __get_free_pages(GFP_ATOMIC | __GFP_ZERO,
891 static void nl_portid_hash_free(struct hlist_head *table, size_t size)
893 if (size <= PAGE_SIZE)
896 free_pages((unsigned long)table, get_order(size));
899 static int nl_portid_hash_rehash(struct nl_portid_hash *hash, int grow)
901 unsigned int omask, mask, shift;
903 struct hlist_head *otable, *table;
906 omask = mask = hash->mask;
907 osize = size = (mask + 1) * sizeof(*table);
911 if (++shift > hash->max_shift)
917 table = nl_portid_hash_zalloc(size);
921 otable = hash->table;
925 get_random_bytes(&hash->rnd, sizeof(hash->rnd));
927 for (i = 0; i <= omask; i++) {
929 struct hlist_node *tmp;
931 sk_for_each_safe(sk, tmp, &otable[i])
932 __sk_add_node(sk, nl_portid_hashfn(hash, nlk_sk(sk)->portid));
935 nl_portid_hash_free(otable, osize);
936 hash->rehash_time = jiffies + 10 * 60 * HZ;
940 static inline int nl_portid_hash_dilute(struct nl_portid_hash *hash, int len)
942 int avg = hash->entries >> hash->shift;
944 if (unlikely(avg > 1) && nl_portid_hash_rehash(hash, 1))
947 if (unlikely(len > avg) && time_after(jiffies, hash->rehash_time)) {
948 nl_portid_hash_rehash(hash, 0);
955 static const struct proto_ops netlink_ops;
958 netlink_update_listeners(struct sock *sk)
960 struct netlink_table *tbl = &nl_table[sk->sk_protocol];
963 struct listeners *listeners;
965 listeners = nl_deref_protected(tbl->listeners);
969 for (i = 0; i < NLGRPLONGS(tbl->groups); i++) {
971 sk_for_each_bound(sk, &tbl->mc_list) {
972 if (i < NLGRPLONGS(nlk_sk(sk)->ngroups))
973 mask |= nlk_sk(sk)->groups[i];
975 listeners->masks[i] = mask;
977 /* this function is only called with the netlink table "grabbed", which
978 * makes sure updates are visible before bind or setsockopt return. */
981 static int netlink_insert(struct sock *sk, struct net *net, u32 portid)
983 struct nl_portid_hash *hash = &nl_table[sk->sk_protocol].hash;
984 struct hlist_head *head;
985 int err = -EADDRINUSE;
989 netlink_table_grab();
990 head = nl_portid_hashfn(hash, portid);
992 sk_for_each(osk, head) {
993 if (net_eq(sock_net(osk), net) && (nlk_sk(osk)->portid == portid))
1001 if (nlk_sk(sk)->portid)
1005 if (BITS_PER_LONG > 32 && unlikely(hash->entries >= UINT_MAX))
1008 if (len && nl_portid_hash_dilute(hash, len))
1009 head = nl_portid_hashfn(hash, portid);
1011 nlk_sk(sk)->portid = portid;
1012 sk_add_node(sk, head);
1016 netlink_table_ungrab();
1020 static void netlink_remove(struct sock *sk)
1022 netlink_table_grab();
1023 if (sk_del_node_init(sk))
1024 nl_table[sk->sk_protocol].hash.entries--;
1025 if (nlk_sk(sk)->subscriptions)
1026 __sk_del_bind_node(sk);
1027 netlink_table_ungrab();
1030 static struct proto netlink_proto = {
1032 .owner = THIS_MODULE,
1033 .obj_size = sizeof(struct netlink_sock),
1036 static int __netlink_create(struct net *net, struct socket *sock,
1037 struct mutex *cb_mutex, int protocol)
1040 struct netlink_sock *nlk;
1042 sock->ops = &netlink_ops;
1044 sk = sk_alloc(net, PF_NETLINK, GFP_KERNEL, &netlink_proto);
1048 sock_init_data(sock, sk);
1052 nlk->cb_mutex = cb_mutex;
1054 nlk->cb_mutex = &nlk->cb_def_mutex;
1055 mutex_init(nlk->cb_mutex);
1057 init_waitqueue_head(&nlk->wait);
1058 #ifdef CONFIG_NETLINK_MMAP
1059 mutex_init(&nlk->pg_vec_lock);
1062 sk->sk_destruct = netlink_sock_destruct;
1063 sk->sk_protocol = protocol;
1067 static int netlink_create(struct net *net, struct socket *sock, int protocol,
1070 struct module *module = NULL;
1071 struct mutex *cb_mutex;
1072 struct netlink_sock *nlk;
1073 void (*bind)(int group);
1076 sock->state = SS_UNCONNECTED;
1078 if (sock->type != SOCK_RAW && sock->type != SOCK_DGRAM)
1079 return -ESOCKTNOSUPPORT;
1081 if (protocol < 0 || protocol >= MAX_LINKS)
1082 return -EPROTONOSUPPORT;
1084 netlink_lock_table();
1085 #ifdef CONFIG_MODULES
1086 if (!nl_table[protocol].registered) {
1087 netlink_unlock_table();
1088 request_module("net-pf-%d-proto-%d", PF_NETLINK, protocol);
1089 netlink_lock_table();
1092 if (nl_table[protocol].registered &&
1093 try_module_get(nl_table[protocol].module))
1094 module = nl_table[protocol].module;
1096 err = -EPROTONOSUPPORT;
1097 cb_mutex = nl_table[protocol].cb_mutex;
1098 bind = nl_table[protocol].bind;
1099 netlink_unlock_table();
1104 err = __netlink_create(net, sock, cb_mutex, protocol);
1109 sock_prot_inuse_add(net, &netlink_proto, 1);
1112 nlk = nlk_sk(sock->sk);
1113 nlk->module = module;
1114 nlk->netlink_bind = bind;
1123 static int netlink_release(struct socket *sock)
1125 struct sock *sk = sock->sk;
1126 struct netlink_sock *nlk;
1136 * OK. Socket is unlinked, any packets that arrive now
1141 wake_up_interruptible_all(&nlk->wait);
1143 skb_queue_purge(&sk->sk_write_queue);
1146 struct netlink_notify n = {
1147 .net = sock_net(sk),
1148 .protocol = sk->sk_protocol,
1149 .portid = nlk->portid,
1151 atomic_notifier_call_chain(&netlink_chain,
1152 NETLINK_URELEASE, &n);
1155 module_put(nlk->module);
1157 netlink_table_grab();
1158 if (netlink_is_kernel(sk)) {
1159 BUG_ON(nl_table[sk->sk_protocol].registered == 0);
1160 if (--nl_table[sk->sk_protocol].registered == 0) {
1161 struct listeners *old;
1163 old = nl_deref_protected(nl_table[sk->sk_protocol].listeners);
1164 RCU_INIT_POINTER(nl_table[sk->sk_protocol].listeners, NULL);
1165 kfree_rcu(old, rcu);
1166 nl_table[sk->sk_protocol].module = NULL;
1167 nl_table[sk->sk_protocol].bind = NULL;
1168 nl_table[sk->sk_protocol].flags = 0;
1169 nl_table[sk->sk_protocol].registered = 0;
1171 } else if (nlk->subscriptions) {
1172 netlink_update_listeners(sk);
1174 netlink_table_ungrab();
1180 sock_prot_inuse_add(sock_net(sk), &netlink_proto, -1);
1186 static int netlink_autobind(struct socket *sock)
1188 struct sock *sk = sock->sk;
1189 struct net *net = sock_net(sk);
1190 struct nl_portid_hash *hash = &nl_table[sk->sk_protocol].hash;
1191 struct hlist_head *head;
1193 s32 portid = task_tgid_vnr(current);
1195 static s32 rover = -4097;
1199 netlink_table_grab();
1200 head = nl_portid_hashfn(hash, portid);
1201 sk_for_each(osk, head) {
1202 if (!net_eq(sock_net(osk), net))
1204 if (nlk_sk(osk)->portid == portid) {
1205 /* Bind collision, search negative portid values. */
1209 netlink_table_ungrab();
1213 netlink_table_ungrab();
1215 err = netlink_insert(sk, net, portid);
1216 if (err == -EADDRINUSE)
1219 /* If 2 threads race to autobind, that is fine. */
1226 static inline int netlink_capable(const struct socket *sock, unsigned int flag)
1228 return (nl_table[sock->sk->sk_protocol].flags & flag) ||
1229 ns_capable(sock_net(sock->sk)->user_ns, CAP_NET_ADMIN);
1233 netlink_update_subscriptions(struct sock *sk, unsigned int subscriptions)
1235 struct netlink_sock *nlk = nlk_sk(sk);
1237 if (nlk->subscriptions && !subscriptions)
1238 __sk_del_bind_node(sk);
1239 else if (!nlk->subscriptions && subscriptions)
1240 sk_add_bind_node(sk, &nl_table[sk->sk_protocol].mc_list);
1241 nlk->subscriptions = subscriptions;
1244 static int netlink_realloc_groups(struct sock *sk)
1246 struct netlink_sock *nlk = nlk_sk(sk);
1247 unsigned int groups;
1248 unsigned long *new_groups;
1251 netlink_table_grab();
1253 groups = nl_table[sk->sk_protocol].groups;
1254 if (!nl_table[sk->sk_protocol].registered) {
1259 if (nlk->ngroups >= groups)
1262 new_groups = krealloc(nlk->groups, NLGRPSZ(groups), GFP_ATOMIC);
1263 if (new_groups == NULL) {
1267 memset((char *)new_groups + NLGRPSZ(nlk->ngroups), 0,
1268 NLGRPSZ(groups) - NLGRPSZ(nlk->ngroups));
1270 nlk->groups = new_groups;
1271 nlk->ngroups = groups;
1273 netlink_table_ungrab();
1277 static int netlink_bind(struct socket *sock, struct sockaddr *addr,
1280 struct sock *sk = sock->sk;
1281 struct net *net = sock_net(sk);
1282 struct netlink_sock *nlk = nlk_sk(sk);
1283 struct sockaddr_nl *nladdr = (struct sockaddr_nl *)addr;
1286 if (addr_len < sizeof(struct sockaddr_nl))
1289 if (nladdr->nl_family != AF_NETLINK)
1292 /* Only superuser is allowed to listen multicasts */
1293 if (nladdr->nl_groups) {
1294 if (!netlink_capable(sock, NL_CFG_F_NONROOT_RECV))
1296 err = netlink_realloc_groups(sk);
1302 if (nladdr->nl_pid != nlk->portid)
1305 err = nladdr->nl_pid ?
1306 netlink_insert(sk, net, nladdr->nl_pid) :
1307 netlink_autobind(sock);
1312 if (!nladdr->nl_groups && (nlk->groups == NULL || !(u32)nlk->groups[0]))
1315 netlink_table_grab();
1316 netlink_update_subscriptions(sk, nlk->subscriptions +
1317 hweight32(nladdr->nl_groups) -
1318 hweight32(nlk->groups[0]));
1319 nlk->groups[0] = (nlk->groups[0] & ~0xffffffffUL) | nladdr->nl_groups;
1320 netlink_update_listeners(sk);
1321 netlink_table_ungrab();
1323 if (nlk->netlink_bind && nlk->groups[0]) {
1326 for (i=0; i<nlk->ngroups; i++) {
1327 if (test_bit(i, nlk->groups))
1328 nlk->netlink_bind(i);
1335 static int netlink_connect(struct socket *sock, struct sockaddr *addr,
1336 int alen, int flags)
1339 struct sock *sk = sock->sk;
1340 struct netlink_sock *nlk = nlk_sk(sk);
1341 struct sockaddr_nl *nladdr = (struct sockaddr_nl *)addr;
1343 if (alen < sizeof(addr->sa_family))
1346 if (addr->sa_family == AF_UNSPEC) {
1347 sk->sk_state = NETLINK_UNCONNECTED;
1348 nlk->dst_portid = 0;
1352 if (addr->sa_family != AF_NETLINK)
1355 /* Only superuser is allowed to send multicasts */
1356 if (nladdr->nl_groups && !netlink_capable(sock, NL_CFG_F_NONROOT_SEND))
1360 err = netlink_autobind(sock);
1363 sk->sk_state = NETLINK_CONNECTED;
1364 nlk->dst_portid = nladdr->nl_pid;
1365 nlk->dst_group = ffs(nladdr->nl_groups);
1371 static int netlink_getname(struct socket *sock, struct sockaddr *addr,
1372 int *addr_len, int peer)
1374 struct sock *sk = sock->sk;
1375 struct netlink_sock *nlk = nlk_sk(sk);
1376 DECLARE_SOCKADDR(struct sockaddr_nl *, nladdr, addr);
1378 nladdr->nl_family = AF_NETLINK;
1380 *addr_len = sizeof(*nladdr);
1383 nladdr->nl_pid = nlk->dst_portid;
1384 nladdr->nl_groups = netlink_group_mask(nlk->dst_group);
1386 nladdr->nl_pid = nlk->portid;
1387 nladdr->nl_groups = nlk->groups ? nlk->groups[0] : 0;
1392 static struct sock *netlink_getsockbyportid(struct sock *ssk, u32 portid)
1395 struct netlink_sock *nlk;
1397 sock = netlink_lookup(sock_net(ssk), ssk->sk_protocol, portid);
1399 return ERR_PTR(-ECONNREFUSED);
1401 /* Don't bother queuing skb if kernel socket has no input function */
1403 if (sock->sk_state == NETLINK_CONNECTED &&
1404 nlk->dst_portid != nlk_sk(ssk)->portid) {
1406 return ERR_PTR(-ECONNREFUSED);
1411 struct sock *netlink_getsockbyfilp(struct file *filp)
1413 struct inode *inode = file_inode(filp);
1416 if (!S_ISSOCK(inode->i_mode))
1417 return ERR_PTR(-ENOTSOCK);
1419 sock = SOCKET_I(inode)->sk;
1420 if (sock->sk_family != AF_NETLINK)
1421 return ERR_PTR(-EINVAL);
1427 static struct sk_buff *netlink_alloc_large_skb(unsigned int size)
1429 struct sk_buff *skb;
1432 if (size <= NLMSG_GOODSIZE)
1433 return alloc_skb(size, GFP_KERNEL);
1435 skb = alloc_skb_head(GFP_KERNEL);
1439 data = vmalloc(size);
1445 skb_reset_tail_pointer(skb);
1446 skb->end = skb->tail + size;
1448 skb->destructor = netlink_skb_destructor;
1457 * Attach a skb to a netlink socket.
1458 * The caller must hold a reference to the destination socket. On error, the
1459 * reference is dropped. The skb is not send to the destination, just all
1460 * all error checks are performed and memory in the queue is reserved.
1462 * < 0: error. skb freed, reference to sock dropped.
1464 * 1: repeat lookup - reference dropped while waiting for socket memory.
1466 int netlink_attachskb(struct sock *sk, struct sk_buff *skb,
1467 long *timeo, struct sock *ssk)
1469 struct netlink_sock *nlk;
1473 if ((atomic_read(&sk->sk_rmem_alloc) > sk->sk_rcvbuf ||
1474 test_bit(NETLINK_CONGESTED, &nlk->state)) &&
1475 !netlink_skb_is_mmaped(skb)) {
1476 DECLARE_WAITQUEUE(wait, current);
1478 if (!ssk || netlink_is_kernel(ssk))
1479 netlink_overrun(sk);
1485 __set_current_state(TASK_INTERRUPTIBLE);
1486 add_wait_queue(&nlk->wait, &wait);
1488 if ((atomic_read(&sk->sk_rmem_alloc) > sk->sk_rcvbuf ||
1489 test_bit(NETLINK_CONGESTED, &nlk->state)) &&
1490 !sock_flag(sk, SOCK_DEAD))
1491 *timeo = schedule_timeout(*timeo);
1493 __set_current_state(TASK_RUNNING);
1494 remove_wait_queue(&nlk->wait, &wait);
1497 if (signal_pending(current)) {
1499 return sock_intr_errno(*timeo);
1503 netlink_skb_set_owner_r(skb, sk);
1507 static int __netlink_sendskb(struct sock *sk, struct sk_buff *skb)
1511 #ifdef CONFIG_NETLINK_MMAP
1512 if (netlink_skb_is_mmaped(skb))
1513 netlink_queue_mmaped_skb(sk, skb);
1514 else if (netlink_rx_is_mmaped(sk))
1515 netlink_ring_set_copied(sk, skb);
1517 #endif /* CONFIG_NETLINK_MMAP */
1518 skb_queue_tail(&sk->sk_receive_queue, skb);
1519 sk->sk_data_ready(sk, len);
1523 int netlink_sendskb(struct sock *sk, struct sk_buff *skb)
1525 int len = __netlink_sendskb(sk, skb);
1531 void netlink_detachskb(struct sock *sk, struct sk_buff *skb)
1537 static struct sk_buff *netlink_trim(struct sk_buff *skb, gfp_t allocation)
1541 WARN_ON(skb->sk != NULL);
1542 if (netlink_skb_is_mmaped(skb))
1545 delta = skb->end - skb->tail;
1546 if (is_vmalloc_addr(skb->head) || delta * 2 < skb->truesize)
1549 if (skb_shared(skb)) {
1550 struct sk_buff *nskb = skb_clone(skb, allocation);
1557 if (!pskb_expand_head(skb, 0, -delta, allocation))
1558 skb->truesize -= delta;
1563 static int netlink_unicast_kernel(struct sock *sk, struct sk_buff *skb,
1567 struct netlink_sock *nlk = nlk_sk(sk);
1569 ret = -ECONNREFUSED;
1570 if (nlk->netlink_rcv != NULL) {
1572 netlink_skb_set_owner_r(skb, sk);
1573 NETLINK_CB(skb).sk = ssk;
1574 nlk->netlink_rcv(skb);
1583 int netlink_unicast(struct sock *ssk, struct sk_buff *skb,
1584 u32 portid, int nonblock)
1590 skb = netlink_trim(skb, gfp_any());
1592 timeo = sock_sndtimeo(ssk, nonblock);
1594 sk = netlink_getsockbyportid(ssk, portid);
1599 if (netlink_is_kernel(sk))
1600 return netlink_unicast_kernel(sk, skb, ssk);
1602 if (sk_filter(sk, skb)) {
1609 err = netlink_attachskb(sk, skb, &timeo, ssk);
1615 return netlink_sendskb(sk, skb);
1617 EXPORT_SYMBOL(netlink_unicast);
1619 struct sk_buff *netlink_alloc_skb(struct sock *ssk, unsigned int size,
1620 u32 dst_portid, gfp_t gfp_mask)
1622 #ifdef CONFIG_NETLINK_MMAP
1623 struct sock *sk = NULL;
1624 struct sk_buff *skb;
1625 struct netlink_ring *ring;
1626 struct nl_mmap_hdr *hdr;
1627 unsigned int maxlen;
1629 sk = netlink_getsockbyportid(ssk, dst_portid);
1633 ring = &nlk_sk(sk)->rx_ring;
1634 /* fast-path without atomic ops for common case: non-mmaped receiver */
1635 if (ring->pg_vec == NULL)
1638 skb = alloc_skb_head(gfp_mask);
1642 spin_lock_bh(&sk->sk_receive_queue.lock);
1643 /* check again under lock */
1644 if (ring->pg_vec == NULL)
1647 maxlen = ring->frame_size - NL_MMAP_HDRLEN;
1651 netlink_forward_ring(ring);
1652 hdr = netlink_current_frame(ring, NL_MMAP_STATUS_UNUSED);
1655 netlink_ring_setup_skb(skb, sk, ring, hdr);
1656 netlink_set_status(hdr, NL_MMAP_STATUS_RESERVED);
1657 atomic_inc(&ring->pending);
1658 netlink_increment_head(ring);
1660 spin_unlock_bh(&sk->sk_receive_queue.lock);
1665 spin_unlock_bh(&sk->sk_receive_queue.lock);
1666 netlink_overrun(sk);
1673 spin_unlock_bh(&sk->sk_receive_queue.lock);
1678 return alloc_skb(size, gfp_mask);
1680 EXPORT_SYMBOL_GPL(netlink_alloc_skb);
1682 int netlink_has_listeners(struct sock *sk, unsigned int group)
1685 struct listeners *listeners;
1687 BUG_ON(!netlink_is_kernel(sk));
1690 listeners = rcu_dereference(nl_table[sk->sk_protocol].listeners);
1692 if (listeners && group - 1 < nl_table[sk->sk_protocol].groups)
1693 res = test_bit(group - 1, listeners->masks);
1699 EXPORT_SYMBOL_GPL(netlink_has_listeners);
1701 static int netlink_broadcast_deliver(struct sock *sk, struct sk_buff *skb)
1703 struct netlink_sock *nlk = nlk_sk(sk);
1705 if (atomic_read(&sk->sk_rmem_alloc) <= sk->sk_rcvbuf &&
1706 !test_bit(NETLINK_CONGESTED, &nlk->state)) {
1707 netlink_skb_set_owner_r(skb, sk);
1708 __netlink_sendskb(sk, skb);
1709 return atomic_read(&sk->sk_rmem_alloc) > (sk->sk_rcvbuf >> 1);
1714 struct netlink_broadcast_data {
1715 struct sock *exclude_sk;
1720 int delivery_failure;
1724 struct sk_buff *skb, *skb2;
1725 int (*tx_filter)(struct sock *dsk, struct sk_buff *skb, void *data);
1729 static int do_one_broadcast(struct sock *sk,
1730 struct netlink_broadcast_data *p)
1732 struct netlink_sock *nlk = nlk_sk(sk);
1735 if (p->exclude_sk == sk)
1738 if (nlk->portid == p->portid || p->group - 1 >= nlk->ngroups ||
1739 !test_bit(p->group - 1, nlk->groups))
1742 if (!net_eq(sock_net(sk), p->net))
1746 netlink_overrun(sk);
1751 if (p->skb2 == NULL) {
1752 if (skb_shared(p->skb)) {
1753 p->skb2 = skb_clone(p->skb, p->allocation);
1755 p->skb2 = skb_get(p->skb);
1757 * skb ownership may have been set when
1758 * delivered to a previous socket.
1760 skb_orphan(p->skb2);
1763 if (p->skb2 == NULL) {
1764 netlink_overrun(sk);
1765 /* Clone failed. Notify ALL listeners. */
1767 if (nlk->flags & NETLINK_BROADCAST_SEND_ERROR)
1768 p->delivery_failure = 1;
1769 } else if (p->tx_filter && p->tx_filter(sk, p->skb2, p->tx_data)) {
1772 } else if (sk_filter(sk, p->skb2)) {
1775 } else if ((val = netlink_broadcast_deliver(sk, p->skb2)) < 0) {
1776 netlink_overrun(sk);
1777 if (nlk->flags & NETLINK_BROADCAST_SEND_ERROR)
1778 p->delivery_failure = 1;
1780 p->congested |= val;
1790 int netlink_broadcast_filtered(struct sock *ssk, struct sk_buff *skb, u32 portid,
1791 u32 group, gfp_t allocation,
1792 int (*filter)(struct sock *dsk, struct sk_buff *skb, void *data),
1795 struct net *net = sock_net(ssk);
1796 struct netlink_broadcast_data info;
1799 skb = netlink_trim(skb, allocation);
1801 info.exclude_sk = ssk;
1803 info.portid = portid;
1806 info.delivery_failure = 0;
1809 info.allocation = allocation;
1812 info.tx_filter = filter;
1813 info.tx_data = filter_data;
1815 /* While we sleep in clone, do not allow to change socket list */
1817 netlink_lock_table();
1819 sk_for_each_bound(sk, &nl_table[ssk->sk_protocol].mc_list)
1820 do_one_broadcast(sk, &info);
1824 netlink_unlock_table();
1826 if (info.delivery_failure) {
1827 kfree_skb(info.skb2);
1830 consume_skb(info.skb2);
1832 if (info.delivered) {
1833 if (info.congested && (allocation & __GFP_WAIT))
1839 EXPORT_SYMBOL(netlink_broadcast_filtered);
1841 int netlink_broadcast(struct sock *ssk, struct sk_buff *skb, u32 portid,
1842 u32 group, gfp_t allocation)
1844 return netlink_broadcast_filtered(ssk, skb, portid, group, allocation,
1847 EXPORT_SYMBOL(netlink_broadcast);
1849 struct netlink_set_err_data {
1850 struct sock *exclude_sk;
1856 static int do_one_set_err(struct sock *sk, struct netlink_set_err_data *p)
1858 struct netlink_sock *nlk = nlk_sk(sk);
1861 if (sk == p->exclude_sk)
1864 if (!net_eq(sock_net(sk), sock_net(p->exclude_sk)))
1867 if (nlk->portid == p->portid || p->group - 1 >= nlk->ngroups ||
1868 !test_bit(p->group - 1, nlk->groups))
1871 if (p->code == ENOBUFS && nlk->flags & NETLINK_RECV_NO_ENOBUFS) {
1876 sk->sk_err = p->code;
1877 sk->sk_error_report(sk);
1883 * netlink_set_err - report error to broadcast listeners
1884 * @ssk: the kernel netlink socket, as returned by netlink_kernel_create()
1885 * @portid: the PORTID of a process that we want to skip (if any)
1886 * @groups: the broadcast group that will notice the error
1887 * @code: error code, must be negative (as usual in kernelspace)
1889 * This function returns the number of broadcast listeners that have set the
1890 * NETLINK_RECV_NO_ENOBUFS socket option.
1892 int netlink_set_err(struct sock *ssk, u32 portid, u32 group, int code)
1894 struct netlink_set_err_data info;
1898 info.exclude_sk = ssk;
1899 info.portid = portid;
1901 /* sk->sk_err wants a positive error value */
1904 read_lock(&nl_table_lock);
1906 sk_for_each_bound(sk, &nl_table[ssk->sk_protocol].mc_list)
1907 ret += do_one_set_err(sk, &info);
1909 read_unlock(&nl_table_lock);
1912 EXPORT_SYMBOL(netlink_set_err);
1914 /* must be called with netlink table grabbed */
1915 static void netlink_update_socket_mc(struct netlink_sock *nlk,
1919 int old, new = !!is_new, subscriptions;
1921 old = test_bit(group - 1, nlk->groups);
1922 subscriptions = nlk->subscriptions - old + new;
1924 __set_bit(group - 1, nlk->groups);
1926 __clear_bit(group - 1, nlk->groups);
1927 netlink_update_subscriptions(&nlk->sk, subscriptions);
1928 netlink_update_listeners(&nlk->sk);
1931 static int netlink_setsockopt(struct socket *sock, int level, int optname,
1932 char __user *optval, unsigned int optlen)
1934 struct sock *sk = sock->sk;
1935 struct netlink_sock *nlk = nlk_sk(sk);
1936 unsigned int val = 0;
1939 if (level != SOL_NETLINK)
1940 return -ENOPROTOOPT;
1942 if (optname != NETLINK_RX_RING && optname != NETLINK_TX_RING &&
1943 optlen >= sizeof(int) &&
1944 get_user(val, (unsigned int __user *)optval))
1948 case NETLINK_PKTINFO:
1950 nlk->flags |= NETLINK_RECV_PKTINFO;
1952 nlk->flags &= ~NETLINK_RECV_PKTINFO;
1955 case NETLINK_ADD_MEMBERSHIP:
1956 case NETLINK_DROP_MEMBERSHIP: {
1957 if (!netlink_capable(sock, NL_CFG_F_NONROOT_RECV))
1959 err = netlink_realloc_groups(sk);
1962 if (!val || val - 1 >= nlk->ngroups)
1964 netlink_table_grab();
1965 netlink_update_socket_mc(nlk, val,
1966 optname == NETLINK_ADD_MEMBERSHIP);
1967 netlink_table_ungrab();
1969 if (nlk->netlink_bind)
1970 nlk->netlink_bind(val);
1975 case NETLINK_BROADCAST_ERROR:
1977 nlk->flags |= NETLINK_BROADCAST_SEND_ERROR;
1979 nlk->flags &= ~NETLINK_BROADCAST_SEND_ERROR;
1982 case NETLINK_NO_ENOBUFS:
1984 nlk->flags |= NETLINK_RECV_NO_ENOBUFS;
1985 clear_bit(NETLINK_CONGESTED, &nlk->state);
1986 wake_up_interruptible(&nlk->wait);
1988 nlk->flags &= ~NETLINK_RECV_NO_ENOBUFS;
1992 #ifdef CONFIG_NETLINK_MMAP
1993 case NETLINK_RX_RING:
1994 case NETLINK_TX_RING: {
1995 struct nl_mmap_req req;
1997 /* Rings might consume more memory than queue limits, require
2000 if (!capable(CAP_NET_ADMIN))
2002 if (optlen < sizeof(req))
2004 if (copy_from_user(&req, optval, sizeof(req)))
2006 err = netlink_set_ring(sk, &req, false,
2007 optname == NETLINK_TX_RING);
2010 #endif /* CONFIG_NETLINK_MMAP */
2017 static int netlink_getsockopt(struct socket *sock, int level, int optname,
2018 char __user *optval, int __user *optlen)
2020 struct sock *sk = sock->sk;
2021 struct netlink_sock *nlk = nlk_sk(sk);
2024 if (level != SOL_NETLINK)
2025 return -ENOPROTOOPT;
2027 if (get_user(len, optlen))
2033 case NETLINK_PKTINFO:
2034 if (len < sizeof(int))
2037 val = nlk->flags & NETLINK_RECV_PKTINFO ? 1 : 0;
2038 if (put_user(len, optlen) ||
2039 put_user(val, optval))
2043 case NETLINK_BROADCAST_ERROR:
2044 if (len < sizeof(int))
2047 val = nlk->flags & NETLINK_BROADCAST_SEND_ERROR ? 1 : 0;
2048 if (put_user(len, optlen) ||
2049 put_user(val, optval))
2053 case NETLINK_NO_ENOBUFS:
2054 if (len < sizeof(int))
2057 val = nlk->flags & NETLINK_RECV_NO_ENOBUFS ? 1 : 0;
2058 if (put_user(len, optlen) ||
2059 put_user(val, optval))
2069 static void netlink_cmsg_recv_pktinfo(struct msghdr *msg, struct sk_buff *skb)
2071 struct nl_pktinfo info;
2073 info.group = NETLINK_CB(skb).dst_group;
2074 put_cmsg(msg, SOL_NETLINK, NETLINK_PKTINFO, sizeof(info), &info);
2077 static int netlink_sendmsg(struct kiocb *kiocb, struct socket *sock,
2078 struct msghdr *msg, size_t len)
2080 struct sock_iocb *siocb = kiocb_to_siocb(kiocb);
2081 struct sock *sk = sock->sk;
2082 struct netlink_sock *nlk = nlk_sk(sk);
2083 struct sockaddr_nl *addr = msg->msg_name;
2086 struct sk_buff *skb;
2088 struct scm_cookie scm;
2090 if (msg->msg_flags&MSG_OOB)
2093 if (NULL == siocb->scm)
2096 err = scm_send(sock, msg, siocb->scm, true);
2100 if (msg->msg_namelen) {
2102 if (addr->nl_family != AF_NETLINK)
2104 dst_portid = addr->nl_pid;
2105 dst_group = ffs(addr->nl_groups);
2107 if ((dst_group || dst_portid) &&
2108 !netlink_capable(sock, NL_CFG_F_NONROOT_SEND))
2111 dst_portid = nlk->dst_portid;
2112 dst_group = nlk->dst_group;
2116 err = netlink_autobind(sock);
2121 if (netlink_tx_is_mmaped(sk) &&
2122 msg->msg_iov->iov_base == NULL) {
2123 err = netlink_mmap_sendmsg(sk, msg, dst_portid, dst_group,
2129 if (len > sk->sk_sndbuf - 32)
2132 skb = netlink_alloc_large_skb(len);
2136 NETLINK_CB(skb).portid = nlk->portid;
2137 NETLINK_CB(skb).dst_group = dst_group;
2138 NETLINK_CB(skb).creds = siocb->scm->creds;
2141 if (memcpy_fromiovec(skb_put(skb, len), msg->msg_iov, len)) {
2146 err = security_netlink_send(sk, skb);
2153 atomic_inc(&skb->users);
2154 netlink_broadcast(sk, skb, dst_portid, dst_group, GFP_KERNEL);
2156 err = netlink_unicast(sk, skb, dst_portid, msg->msg_flags&MSG_DONTWAIT);
2159 scm_destroy(siocb->scm);
2163 static int netlink_recvmsg(struct kiocb *kiocb, struct socket *sock,
2164 struct msghdr *msg, size_t len,
2167 struct sock_iocb *siocb = kiocb_to_siocb(kiocb);
2168 struct scm_cookie scm;
2169 struct sock *sk = sock->sk;
2170 struct netlink_sock *nlk = nlk_sk(sk);
2171 int noblock = flags&MSG_DONTWAIT;
2173 struct sk_buff *skb, *data_skb;
2181 skb = skb_recv_datagram(sk, flags, noblock, &err);
2187 #ifdef CONFIG_COMPAT_NETLINK_MESSAGES
2188 if (unlikely(skb_shinfo(skb)->frag_list)) {
2190 * If this skb has a frag_list, then here that means that we
2191 * will have to use the frag_list skb's data for compat tasks
2192 * and the regular skb's data for normal (non-compat) tasks.
2194 * If we need to send the compat skb, assign it to the
2195 * 'data_skb' variable so that it will be used below for data
2196 * copying. We keep 'skb' for everything else, including
2197 * freeing both later.
2199 if (flags & MSG_CMSG_COMPAT)
2200 data_skb = skb_shinfo(skb)->frag_list;
2204 msg->msg_namelen = 0;
2206 copied = data_skb->len;
2208 msg->msg_flags |= MSG_TRUNC;
2212 skb_reset_transport_header(data_skb);
2213 err = skb_copy_datagram_iovec(data_skb, 0, msg->msg_iov, copied);
2215 if (msg->msg_name) {
2216 struct sockaddr_nl *addr = (struct sockaddr_nl *)msg->msg_name;
2217 addr->nl_family = AF_NETLINK;
2219 addr->nl_pid = NETLINK_CB(skb).portid;
2220 addr->nl_groups = netlink_group_mask(NETLINK_CB(skb).dst_group);
2221 msg->msg_namelen = sizeof(*addr);
2224 if (nlk->flags & NETLINK_RECV_PKTINFO)
2225 netlink_cmsg_recv_pktinfo(msg, skb);
2227 if (NULL == siocb->scm) {
2228 memset(&scm, 0, sizeof(scm));
2231 siocb->scm->creds = *NETLINK_CREDS(skb);
2232 if (flags & MSG_TRUNC)
2233 copied = data_skb->len;
2235 skb_free_datagram(sk, skb);
2237 if (nlk->cb && atomic_read(&sk->sk_rmem_alloc) <= sk->sk_rcvbuf / 2) {
2238 ret = netlink_dump(sk);
2241 sk->sk_error_report(sk);
2245 scm_recv(sock, msg, siocb->scm, flags);
2247 netlink_rcv_wake(sk);
2248 return err ? : copied;
2251 static void netlink_data_ready(struct sock *sk, int len)
2257 * We export these functions to other modules. They provide a
2258 * complete set of kernel non-blocking support for message
2263 __netlink_kernel_create(struct net *net, int unit, struct module *module,
2264 struct netlink_kernel_cfg *cfg)
2266 struct socket *sock;
2268 struct netlink_sock *nlk;
2269 struct listeners *listeners = NULL;
2270 struct mutex *cb_mutex = cfg ? cfg->cb_mutex : NULL;
2271 unsigned int groups;
2275 if (unit < 0 || unit >= MAX_LINKS)
2278 if (sock_create_lite(PF_NETLINK, SOCK_DGRAM, unit, &sock))
2282 * We have to just have a reference on the net from sk, but don't
2283 * get_net it. Besides, we cannot get and then put the net here.
2284 * So we create one inside init_net and the move it to net.
2287 if (__netlink_create(&init_net, sock, cb_mutex, unit) < 0)
2288 goto out_sock_release_nosk;
2291 sk_change_net(sk, net);
2293 if (!cfg || cfg->groups < 32)
2296 groups = cfg->groups;
2298 listeners = kzalloc(sizeof(*listeners) + NLGRPSZ(groups), GFP_KERNEL);
2300 goto out_sock_release;
2302 sk->sk_data_ready = netlink_data_ready;
2303 if (cfg && cfg->input)
2304 nlk_sk(sk)->netlink_rcv = cfg->input;
2306 if (netlink_insert(sk, net, 0))
2307 goto out_sock_release;
2310 nlk->flags |= NETLINK_KERNEL_SOCKET;
2312 netlink_table_grab();
2313 if (!nl_table[unit].registered) {
2314 nl_table[unit].groups = groups;
2315 rcu_assign_pointer(nl_table[unit].listeners, listeners);
2316 nl_table[unit].cb_mutex = cb_mutex;
2317 nl_table[unit].module = module;
2319 nl_table[unit].bind = cfg->bind;
2320 nl_table[unit].flags = cfg->flags;
2322 nl_table[unit].registered = 1;
2325 nl_table[unit].registered++;
2327 netlink_table_ungrab();
2332 netlink_kernel_release(sk);
2335 out_sock_release_nosk:
2339 EXPORT_SYMBOL(__netlink_kernel_create);
2342 netlink_kernel_release(struct sock *sk)
2344 sk_release_kernel(sk);
2346 EXPORT_SYMBOL(netlink_kernel_release);
2348 int __netlink_change_ngroups(struct sock *sk, unsigned int groups)
2350 struct listeners *new, *old;
2351 struct netlink_table *tbl = &nl_table[sk->sk_protocol];
2356 if (NLGRPSZ(tbl->groups) < NLGRPSZ(groups)) {
2357 new = kzalloc(sizeof(*new) + NLGRPSZ(groups), GFP_ATOMIC);
2360 old = nl_deref_protected(tbl->listeners);
2361 memcpy(new->masks, old->masks, NLGRPSZ(tbl->groups));
2362 rcu_assign_pointer(tbl->listeners, new);
2364 kfree_rcu(old, rcu);
2366 tbl->groups = groups;
2372 * netlink_change_ngroups - change number of multicast groups
2374 * This changes the number of multicast groups that are available
2375 * on a certain netlink family. Note that it is not possible to
2376 * change the number of groups to below 32. Also note that it does
2377 * not implicitly call netlink_clear_multicast_users() when the
2378 * number of groups is reduced.
2380 * @sk: The kernel netlink socket, as returned by netlink_kernel_create().
2381 * @groups: The new number of groups.
2383 int netlink_change_ngroups(struct sock *sk, unsigned int groups)
2387 netlink_table_grab();
2388 err = __netlink_change_ngroups(sk, groups);
2389 netlink_table_ungrab();
2394 void __netlink_clear_multicast_users(struct sock *ksk, unsigned int group)
2397 struct netlink_table *tbl = &nl_table[ksk->sk_protocol];
2399 sk_for_each_bound(sk, &tbl->mc_list)
2400 netlink_update_socket_mc(nlk_sk(sk), group, 0);
2404 * netlink_clear_multicast_users - kick off multicast listeners
2406 * This function removes all listeners from the given group.
2407 * @ksk: The kernel netlink socket, as returned by
2408 * netlink_kernel_create().
2409 * @group: The multicast group to clear.
2411 void netlink_clear_multicast_users(struct sock *ksk, unsigned int group)
2413 netlink_table_grab();
2414 __netlink_clear_multicast_users(ksk, group);
2415 netlink_table_ungrab();
2419 __nlmsg_put(struct sk_buff *skb, u32 portid, u32 seq, int type, int len, int flags)
2421 struct nlmsghdr *nlh;
2422 int size = nlmsg_msg_size(len);
2424 nlh = (struct nlmsghdr*)skb_put(skb, NLMSG_ALIGN(size));
2425 nlh->nlmsg_type = type;
2426 nlh->nlmsg_len = size;
2427 nlh->nlmsg_flags = flags;
2428 nlh->nlmsg_pid = portid;
2429 nlh->nlmsg_seq = seq;
2430 if (!__builtin_constant_p(size) || NLMSG_ALIGN(size) - size != 0)
2431 memset(nlmsg_data(nlh) + len, 0, NLMSG_ALIGN(size) - size);
2434 EXPORT_SYMBOL(__nlmsg_put);
2437 * It looks a bit ugly.
2438 * It would be better to create kernel thread.
2441 static int netlink_dump(struct sock *sk)
2443 struct netlink_sock *nlk = nlk_sk(sk);
2444 struct netlink_callback *cb;
2445 struct sk_buff *skb = NULL;
2446 struct nlmsghdr *nlh;
2447 int len, err = -ENOBUFS;
2450 mutex_lock(nlk->cb_mutex);
2458 alloc_size = max_t(int, cb->min_dump_alloc, NLMSG_GOODSIZE);
2460 if (!netlink_rx_is_mmaped(sk) &&
2461 atomic_read(&sk->sk_rmem_alloc) >= sk->sk_rcvbuf)
2463 skb = netlink_alloc_skb(sk, alloc_size, nlk->portid, GFP_KERNEL);
2466 netlink_skb_set_owner_r(skb, sk);
2468 len = cb->dump(skb, cb);
2471 mutex_unlock(nlk->cb_mutex);
2473 if (sk_filter(sk, skb))
2476 __netlink_sendskb(sk, skb);
2480 nlh = nlmsg_put_answer(skb, cb, NLMSG_DONE, sizeof(len), NLM_F_MULTI);
2484 nl_dump_check_consistent(cb, nlh);
2486 memcpy(nlmsg_data(nlh), &len, sizeof(len));
2488 if (sk_filter(sk, skb))
2491 __netlink_sendskb(sk, skb);
2496 mutex_unlock(nlk->cb_mutex);
2498 module_put(cb->module);
2499 netlink_consume_callback(cb);
2503 mutex_unlock(nlk->cb_mutex);
2508 int __netlink_dump_start(struct sock *ssk, struct sk_buff *skb,
2509 const struct nlmsghdr *nlh,
2510 struct netlink_dump_control *control)
2512 struct netlink_callback *cb;
2514 struct netlink_sock *nlk;
2517 cb = kzalloc(sizeof(*cb), GFP_KERNEL);
2521 /* Memory mapped dump requests need to be copied to avoid looping
2522 * on the pending state in netlink_mmap_sendmsg() while the CB hold
2523 * a reference to the skb.
2525 if (netlink_skb_is_mmaped(skb)) {
2526 skb = skb_copy(skb, GFP_KERNEL);
2532 atomic_inc(&skb->users);
2534 cb->dump = control->dump;
2535 cb->done = control->done;
2537 cb->data = control->data;
2538 cb->module = control->module;
2539 cb->min_dump_alloc = control->min_dump_alloc;
2542 sk = netlink_lookup(sock_net(ssk), ssk->sk_protocol, NETLINK_CB(skb).portid);
2544 netlink_destroy_callback(cb);
2545 return -ECONNREFUSED;
2549 mutex_lock(nlk->cb_mutex);
2550 /* A dump is in progress... */
2552 mutex_unlock(nlk->cb_mutex);
2553 netlink_destroy_callback(cb);
2557 /* add reference of module which cb->dump belongs to */
2558 if (!try_module_get(cb->module)) {
2559 mutex_unlock(nlk->cb_mutex);
2560 netlink_destroy_callback(cb);
2561 ret = -EPROTONOSUPPORT;
2566 mutex_unlock(nlk->cb_mutex);
2568 ret = netlink_dump(sk);
2575 /* We successfully started a dump, by returning -EINTR we
2576 * signal not to send ACK even if it was requested.
2580 EXPORT_SYMBOL(__netlink_dump_start);
2582 void netlink_ack(struct sk_buff *in_skb, struct nlmsghdr *nlh, int err)
2584 struct sk_buff *skb;
2585 struct nlmsghdr *rep;
2586 struct nlmsgerr *errmsg;
2587 size_t payload = sizeof(*errmsg);
2589 /* error messages get the original request appened */
2591 payload += nlmsg_len(nlh);
2593 skb = netlink_alloc_skb(in_skb->sk, nlmsg_total_size(payload),
2594 NETLINK_CB(in_skb).portid, GFP_KERNEL);
2598 sk = netlink_lookup(sock_net(in_skb->sk),
2599 in_skb->sk->sk_protocol,
2600 NETLINK_CB(in_skb).portid);
2602 sk->sk_err = ENOBUFS;
2603 sk->sk_error_report(sk);
2609 rep = __nlmsg_put(skb, NETLINK_CB(in_skb).portid, nlh->nlmsg_seq,
2610 NLMSG_ERROR, payload, 0);
2611 errmsg = nlmsg_data(rep);
2612 errmsg->error = err;
2613 memcpy(&errmsg->msg, nlh, err ? nlh->nlmsg_len : sizeof(*nlh));
2614 netlink_unicast(in_skb->sk, skb, NETLINK_CB(in_skb).portid, MSG_DONTWAIT);
2616 EXPORT_SYMBOL(netlink_ack);
2618 int netlink_rcv_skb(struct sk_buff *skb, int (*cb)(struct sk_buff *,
2621 struct nlmsghdr *nlh;
2624 while (skb->len >= nlmsg_total_size(0)) {
2627 nlh = nlmsg_hdr(skb);
2630 if (nlh->nlmsg_len < NLMSG_HDRLEN || skb->len < nlh->nlmsg_len)
2633 /* Only requests are handled by the kernel */
2634 if (!(nlh->nlmsg_flags & NLM_F_REQUEST))
2637 /* Skip control messages */
2638 if (nlh->nlmsg_type < NLMSG_MIN_TYPE)
2646 if (nlh->nlmsg_flags & NLM_F_ACK || err)
2647 netlink_ack(skb, nlh, err);
2650 msglen = NLMSG_ALIGN(nlh->nlmsg_len);
2651 if (msglen > skb->len)
2653 skb_pull(skb, msglen);
2658 EXPORT_SYMBOL(netlink_rcv_skb);
2661 * nlmsg_notify - send a notification netlink message
2662 * @sk: netlink socket to use
2663 * @skb: notification message
2664 * @portid: destination netlink portid for reports or 0
2665 * @group: destination multicast group or 0
2666 * @report: 1 to report back, 0 to disable
2667 * @flags: allocation flags
2669 int nlmsg_notify(struct sock *sk, struct sk_buff *skb, u32 portid,
2670 unsigned int group, int report, gfp_t flags)
2675 int exclude_portid = 0;
2678 atomic_inc(&skb->users);
2679 exclude_portid = portid;
2682 /* errors reported via destination sk->sk_err, but propagate
2683 * delivery errors if NETLINK_BROADCAST_ERROR flag is set */
2684 err = nlmsg_multicast(sk, skb, exclude_portid, group, flags);
2690 err2 = nlmsg_unicast(sk, skb, portid);
2691 if (!err || err == -ESRCH)
2697 EXPORT_SYMBOL(nlmsg_notify);
2699 #ifdef CONFIG_PROC_FS
2700 struct nl_seq_iter {
2701 struct seq_net_private p;
2706 static struct sock *netlink_seq_socket_idx(struct seq_file *seq, loff_t pos)
2708 struct nl_seq_iter *iter = seq->private;
2713 for (i = 0; i < MAX_LINKS; i++) {
2714 struct nl_portid_hash *hash = &nl_table[i].hash;
2716 for (j = 0; j <= hash->mask; j++) {
2717 sk_for_each(s, &hash->table[j]) {
2718 if (sock_net(s) != seq_file_net(seq))
2732 static void *netlink_seq_start(struct seq_file *seq, loff_t *pos)
2733 __acquires(nl_table_lock)
2735 read_lock(&nl_table_lock);
2736 return *pos ? netlink_seq_socket_idx(seq, *pos - 1) : SEQ_START_TOKEN;
2739 static void *netlink_seq_next(struct seq_file *seq, void *v, loff_t *pos)
2742 struct nl_seq_iter *iter;
2747 if (v == SEQ_START_TOKEN)
2748 return netlink_seq_socket_idx(seq, 0);
2750 iter = seq->private;
2754 } while (s && sock_net(s) != seq_file_net(seq));
2759 j = iter->hash_idx + 1;
2762 struct nl_portid_hash *hash = &nl_table[i].hash;
2764 for (; j <= hash->mask; j++) {
2765 s = sk_head(&hash->table[j]);
2766 while (s && sock_net(s) != seq_file_net(seq))
2776 } while (++i < MAX_LINKS);
2781 static void netlink_seq_stop(struct seq_file *seq, void *v)
2782 __releases(nl_table_lock)
2784 read_unlock(&nl_table_lock);
2788 static int netlink_seq_show(struct seq_file *seq, void *v)
2790 if (v == SEQ_START_TOKEN) {
2792 "sk Eth Pid Groups "
2793 "Rmem Wmem Dump Locks Drops Inode\n");
2796 struct netlink_sock *nlk = nlk_sk(s);
2798 seq_printf(seq, "%pK %-3d %-6u %08x %-8d %-8d %pK %-8d %-8d %-8lu\n",
2802 nlk->groups ? (u32)nlk->groups[0] : 0,
2803 sk_rmem_alloc_get(s),
2804 sk_wmem_alloc_get(s),
2806 atomic_read(&s->sk_refcnt),
2807 atomic_read(&s->sk_drops),
2815 static const struct seq_operations netlink_seq_ops = {
2816 .start = netlink_seq_start,
2817 .next = netlink_seq_next,
2818 .stop = netlink_seq_stop,
2819 .show = netlink_seq_show,
2823 static int netlink_seq_open(struct inode *inode, struct file *file)
2825 return seq_open_net(inode, file, &netlink_seq_ops,
2826 sizeof(struct nl_seq_iter));
2829 static const struct file_operations netlink_seq_fops = {
2830 .owner = THIS_MODULE,
2831 .open = netlink_seq_open,
2833 .llseek = seq_lseek,
2834 .release = seq_release_net,
2839 int netlink_register_notifier(struct notifier_block *nb)
2841 return atomic_notifier_chain_register(&netlink_chain, nb);
2843 EXPORT_SYMBOL(netlink_register_notifier);
2845 int netlink_unregister_notifier(struct notifier_block *nb)
2847 return atomic_notifier_chain_unregister(&netlink_chain, nb);
2849 EXPORT_SYMBOL(netlink_unregister_notifier);
2851 static const struct proto_ops netlink_ops = {
2852 .family = PF_NETLINK,
2853 .owner = THIS_MODULE,
2854 .release = netlink_release,
2855 .bind = netlink_bind,
2856 .connect = netlink_connect,
2857 .socketpair = sock_no_socketpair,
2858 .accept = sock_no_accept,
2859 .getname = netlink_getname,
2860 .poll = netlink_poll,
2861 .ioctl = sock_no_ioctl,
2862 .listen = sock_no_listen,
2863 .shutdown = sock_no_shutdown,
2864 .setsockopt = netlink_setsockopt,
2865 .getsockopt = netlink_getsockopt,
2866 .sendmsg = netlink_sendmsg,
2867 .recvmsg = netlink_recvmsg,
2868 .mmap = netlink_mmap,
2869 .sendpage = sock_no_sendpage,
2872 static const struct net_proto_family netlink_family_ops = {
2873 .family = PF_NETLINK,
2874 .create = netlink_create,
2875 .owner = THIS_MODULE, /* for consistency 8) */
2878 static int __net_init netlink_net_init(struct net *net)
2880 #ifdef CONFIG_PROC_FS
2881 if (!proc_create("netlink", 0, net->proc_net, &netlink_seq_fops))
2887 static void __net_exit netlink_net_exit(struct net *net)
2889 #ifdef CONFIG_PROC_FS
2890 remove_proc_entry("netlink", net->proc_net);
2894 static void __init netlink_add_usersock_entry(void)
2896 struct listeners *listeners;
2899 listeners = kzalloc(sizeof(*listeners) + NLGRPSZ(groups), GFP_KERNEL);
2901 panic("netlink_add_usersock_entry: Cannot allocate listeners\n");
2903 netlink_table_grab();
2905 nl_table[NETLINK_USERSOCK].groups = groups;
2906 rcu_assign_pointer(nl_table[NETLINK_USERSOCK].listeners, listeners);
2907 nl_table[NETLINK_USERSOCK].module = THIS_MODULE;
2908 nl_table[NETLINK_USERSOCK].registered = 1;
2909 nl_table[NETLINK_USERSOCK].flags = NL_CFG_F_NONROOT_SEND;
2911 netlink_table_ungrab();
2914 static struct pernet_operations __net_initdata netlink_net_ops = {
2915 .init = netlink_net_init,
2916 .exit = netlink_net_exit,
2919 static int __init netlink_proto_init(void)
2922 unsigned long limit;
2924 int err = proto_register(&netlink_proto, 0);
2929 BUILD_BUG_ON(sizeof(struct netlink_skb_parms) > FIELD_SIZEOF(struct sk_buff, cb));
2931 nl_table = kcalloc(MAX_LINKS, sizeof(*nl_table), GFP_KERNEL);
2935 if (totalram_pages >= (128 * 1024))
2936 limit = totalram_pages >> (21 - PAGE_SHIFT);
2938 limit = totalram_pages >> (23 - PAGE_SHIFT);
2940 order = get_bitmask_order(limit) - 1 + PAGE_SHIFT;
2941 limit = (1UL << order) / sizeof(struct hlist_head);
2942 order = get_bitmask_order(min(limit, (unsigned long)UINT_MAX)) - 1;
2944 for (i = 0; i < MAX_LINKS; i++) {
2945 struct nl_portid_hash *hash = &nl_table[i].hash;
2947 hash->table = nl_portid_hash_zalloc(1 * sizeof(*hash->table));
2950 nl_portid_hash_free(nl_table[i].hash.table,
2951 1 * sizeof(*hash->table));
2955 hash->max_shift = order;
2958 hash->rehash_time = jiffies;
2961 netlink_add_usersock_entry();
2963 sock_register(&netlink_family_ops);
2964 register_pernet_subsys(&netlink_net_ops);
2965 /* The netlink device handler may be needed early. */
2970 panic("netlink_init: Cannot allocate nl_table\n");
2973 core_initcall(netlink_proto_init);