2 * NETLINK Kernel-user communication protocol.
4 * Authors: Alan Cox <alan@lxorguk.ukuu.org.uk>
5 * Alexey Kuznetsov <kuznet@ms2.inr.ac.ru>
6 * Patrick McHardy <kaber@trash.net>
8 * This program is free software; you can redistribute it and/or
9 * modify it under the terms of the GNU General Public License
10 * as published by the Free Software Foundation; either version
11 * 2 of the License, or (at your option) any later version.
13 * Tue Jun 26 14:36:48 MEST 2001 Herbert "herp" Rosmanith
14 * added netlink_proto_exit
15 * Tue Jan 22 18:32:44 BRST 2002 Arnaldo C. de Melo <acme@conectiva.com.br>
16 * use nlk_sk, as sk->protinfo is on a diet 8)
17 * Fri Jul 22 19:51:12 MEST 2005 Harald Welte <laforge@gnumonks.org>
18 * - inc module use count of module that owns
19 * the kernel socket in case userspace opens
20 * socket of same protocol
21 * - remove all module support, since netlink is
22 * mandatory if CONFIG_NET=y these days
25 #include <linux/module.h>
27 #include <linux/capability.h>
28 #include <linux/kernel.h>
29 #include <linux/init.h>
30 #include <linux/signal.h>
31 #include <linux/sched.h>
32 #include <linux/errno.h>
33 #include <linux/string.h>
34 #include <linux/stat.h>
35 #include <linux/socket.h>
37 #include <linux/fcntl.h>
38 #include <linux/termios.h>
39 #include <linux/sockios.h>
40 #include <linux/net.h>
42 #include <linux/slab.h>
43 #include <asm/uaccess.h>
44 #include <linux/skbuff.h>
45 #include <linux/netdevice.h>
46 #include <linux/rtnetlink.h>
47 #include <linux/proc_fs.h>
48 #include <linux/seq_file.h>
49 #include <linux/notifier.h>
50 #include <linux/security.h>
51 #include <linux/jhash.h>
52 #include <linux/jiffies.h>
53 #include <linux/random.h>
54 #include <linux/bitops.h>
56 #include <linux/types.h>
57 #include <linux/audit.h>
58 #include <linux/mutex.h>
59 #include <linux/vmalloc.h>
60 #include <linux/if_arp.h>
61 #include <linux/rhashtable.h>
62 #include <asm/cacheflush.h>
63 #include <linux/hash.h>
65 #include <net/net_namespace.h>
68 #include <net/netlink.h>
70 #include "af_netlink.h"
74 unsigned long masks[0];
78 #define NETLINK_CONGESTED 0x0
81 #define NETLINK_KERNEL_SOCKET 0x1
82 #define NETLINK_RECV_PKTINFO 0x2
83 #define NETLINK_BROADCAST_SEND_ERROR 0x4
84 #define NETLINK_RECV_NO_ENOBUFS 0x8
86 static inline int netlink_is_kernel(struct sock *sk)
88 return nlk_sk(sk)->flags & NETLINK_KERNEL_SOCKET;
91 struct netlink_table *nl_table;
92 EXPORT_SYMBOL_GPL(nl_table);
94 static DECLARE_WAIT_QUEUE_HEAD(nl_table_wait);
96 static int netlink_dump(struct sock *sk);
97 static void netlink_skb_destructor(struct sk_buff *skb);
99 /* nl_table locking explained:
100 * Lookup and traversal are protected with nl_sk_hash_lock or nl_table_lock
101 * combined with an RCU read-side lock. Insertion and removal are protected
102 * with nl_sk_hash_lock while using RCU list modification primitives and may
103 * run in parallel to nl_table_lock protected lookups. Destruction of the
104 * Netlink socket may only occur *after* nl_table_lock has been acquired
105 * either during or after the socket has been removed from the list.
107 DEFINE_RWLOCK(nl_table_lock);
108 EXPORT_SYMBOL_GPL(nl_table_lock);
109 static atomic_t nl_table_users = ATOMIC_INIT(0);
111 #define nl_deref_protected(X) rcu_dereference_protected(X, lockdep_is_held(&nl_table_lock));
113 /* Protects netlink socket hash table mutations */
114 DEFINE_MUTEX(nl_sk_hash_lock);
115 EXPORT_SYMBOL_GPL(nl_sk_hash_lock);
117 static int lockdep_nl_sk_hash_is_held(void)
119 #ifdef CONFIG_LOCKDEP
121 return lockdep_is_held(&nl_sk_hash_lock) || lockdep_is_held(&nl_table_lock);
126 static ATOMIC_NOTIFIER_HEAD(netlink_chain);
128 static DEFINE_SPINLOCK(netlink_tap_lock);
129 static struct list_head netlink_tap_all __read_mostly;
131 static inline u32 netlink_group_mask(u32 group)
133 return group ? 1 << (group - 1) : 0;
136 int netlink_add_tap(struct netlink_tap *nt)
138 if (unlikely(nt->dev->type != ARPHRD_NETLINK))
141 spin_lock(&netlink_tap_lock);
142 list_add_rcu(&nt->list, &netlink_tap_all);
143 spin_unlock(&netlink_tap_lock);
146 __module_get(nt->module);
150 EXPORT_SYMBOL_GPL(netlink_add_tap);
152 static int __netlink_remove_tap(struct netlink_tap *nt)
155 struct netlink_tap *tmp;
157 spin_lock(&netlink_tap_lock);
159 list_for_each_entry(tmp, &netlink_tap_all, list) {
161 list_del_rcu(&nt->list);
167 pr_warn("__netlink_remove_tap: %p not found\n", nt);
169 spin_unlock(&netlink_tap_lock);
171 if (found && nt->module)
172 module_put(nt->module);
174 return found ? 0 : -ENODEV;
177 int netlink_remove_tap(struct netlink_tap *nt)
181 ret = __netlink_remove_tap(nt);
186 EXPORT_SYMBOL_GPL(netlink_remove_tap);
188 static bool netlink_filter_tap(const struct sk_buff *skb)
190 struct sock *sk = skb->sk;
192 /* We take the more conservative approach and
193 * whitelist socket protocols that may pass.
195 switch (sk->sk_protocol) {
197 case NETLINK_USERSOCK:
198 case NETLINK_SOCK_DIAG:
201 case NETLINK_FIB_LOOKUP:
202 case NETLINK_NETFILTER:
203 case NETLINK_GENERIC:
210 static int __netlink_deliver_tap_skb(struct sk_buff *skb,
211 struct net_device *dev)
213 struct sk_buff *nskb;
214 struct sock *sk = skb->sk;
218 nskb = skb_clone(skb, GFP_ATOMIC);
221 nskb->protocol = htons((u16) sk->sk_protocol);
222 nskb->pkt_type = netlink_is_kernel(sk) ?
223 PACKET_KERNEL : PACKET_USER;
224 skb_reset_network_header(nskb);
225 ret = dev_queue_xmit(nskb);
226 if (unlikely(ret > 0))
227 ret = net_xmit_errno(ret);
234 static void __netlink_deliver_tap(struct sk_buff *skb)
237 struct netlink_tap *tmp;
239 if (!netlink_filter_tap(skb))
242 list_for_each_entry_rcu(tmp, &netlink_tap_all, list) {
243 ret = __netlink_deliver_tap_skb(skb, tmp->dev);
249 static void netlink_deliver_tap(struct sk_buff *skb)
253 if (unlikely(!list_empty(&netlink_tap_all)))
254 __netlink_deliver_tap(skb);
259 static void netlink_deliver_tap_kernel(struct sock *dst, struct sock *src,
262 if (!(netlink_is_kernel(dst) && netlink_is_kernel(src)))
263 netlink_deliver_tap(skb);
266 static void netlink_overrun(struct sock *sk)
268 struct netlink_sock *nlk = nlk_sk(sk);
270 if (!(nlk->flags & NETLINK_RECV_NO_ENOBUFS)) {
271 if (!test_and_set_bit(NETLINK_CONGESTED, &nlk_sk(sk)->state)) {
272 sk->sk_err = ENOBUFS;
273 sk->sk_error_report(sk);
276 atomic_inc(&sk->sk_drops);
279 static void netlink_rcv_wake(struct sock *sk)
281 struct netlink_sock *nlk = nlk_sk(sk);
283 if (skb_queue_empty(&sk->sk_receive_queue))
284 clear_bit(NETLINK_CONGESTED, &nlk->state);
285 if (!test_bit(NETLINK_CONGESTED, &nlk->state))
286 wake_up_interruptible(&nlk->wait);
289 #ifdef CONFIG_NETLINK_MMAP
290 static bool netlink_skb_is_mmaped(const struct sk_buff *skb)
292 return NETLINK_CB(skb).flags & NETLINK_SKB_MMAPED;
295 static bool netlink_rx_is_mmaped(struct sock *sk)
297 return nlk_sk(sk)->rx_ring.pg_vec != NULL;
300 static bool netlink_tx_is_mmaped(struct sock *sk)
302 return nlk_sk(sk)->tx_ring.pg_vec != NULL;
305 static __pure struct page *pgvec_to_page(const void *addr)
307 if (is_vmalloc_addr(addr))
308 return vmalloc_to_page(addr);
310 return virt_to_page(addr);
313 static void free_pg_vec(void **pg_vec, unsigned int order, unsigned int len)
317 for (i = 0; i < len; i++) {
318 if (pg_vec[i] != NULL) {
319 if (is_vmalloc_addr(pg_vec[i]))
322 free_pages((unsigned long)pg_vec[i], order);
328 static void *alloc_one_pg_vec_page(unsigned long order)
331 gfp_t gfp_flags = GFP_KERNEL | __GFP_COMP | __GFP_ZERO |
332 __GFP_NOWARN | __GFP_NORETRY;
334 buffer = (void *)__get_free_pages(gfp_flags, order);
338 buffer = vzalloc((1 << order) * PAGE_SIZE);
342 gfp_flags &= ~__GFP_NORETRY;
343 return (void *)__get_free_pages(gfp_flags, order);
346 static void **alloc_pg_vec(struct netlink_sock *nlk,
347 struct nl_mmap_req *req, unsigned int order)
349 unsigned int block_nr = req->nm_block_nr;
353 pg_vec = kcalloc(block_nr, sizeof(void *), GFP_KERNEL);
357 for (i = 0; i < block_nr; i++) {
358 pg_vec[i] = alloc_one_pg_vec_page(order);
359 if (pg_vec[i] == NULL)
365 free_pg_vec(pg_vec, order, block_nr);
369 static int netlink_set_ring(struct sock *sk, struct nl_mmap_req *req,
370 bool closing, bool tx_ring)
372 struct netlink_sock *nlk = nlk_sk(sk);
373 struct netlink_ring *ring;
374 struct sk_buff_head *queue;
375 void **pg_vec = NULL;
376 unsigned int order = 0;
379 ring = tx_ring ? &nlk->tx_ring : &nlk->rx_ring;
380 queue = tx_ring ? &sk->sk_write_queue : &sk->sk_receive_queue;
383 if (atomic_read(&nlk->mapped))
385 if (atomic_read(&ring->pending))
389 if (req->nm_block_nr) {
390 if (ring->pg_vec != NULL)
393 if ((int)req->nm_block_size <= 0)
395 if (!PAGE_ALIGNED(req->nm_block_size))
397 if (req->nm_frame_size < NL_MMAP_HDRLEN)
399 if (!IS_ALIGNED(req->nm_frame_size, NL_MMAP_MSG_ALIGNMENT))
402 ring->frames_per_block = req->nm_block_size /
404 if (ring->frames_per_block == 0)
406 if (ring->frames_per_block * req->nm_block_nr !=
410 order = get_order(req->nm_block_size);
411 pg_vec = alloc_pg_vec(nlk, req, order);
415 if (req->nm_frame_nr)
420 mutex_lock(&nlk->pg_vec_lock);
421 if (closing || atomic_read(&nlk->mapped) == 0) {
423 spin_lock_bh(&queue->lock);
425 ring->frame_max = req->nm_frame_nr - 1;
427 ring->frame_size = req->nm_frame_size;
428 ring->pg_vec_pages = req->nm_block_size / PAGE_SIZE;
430 swap(ring->pg_vec_len, req->nm_block_nr);
431 swap(ring->pg_vec_order, order);
432 swap(ring->pg_vec, pg_vec);
434 __skb_queue_purge(queue);
435 spin_unlock_bh(&queue->lock);
437 WARN_ON(atomic_read(&nlk->mapped));
439 mutex_unlock(&nlk->pg_vec_lock);
442 free_pg_vec(pg_vec, order, req->nm_block_nr);
446 static void netlink_mm_open(struct vm_area_struct *vma)
448 struct file *file = vma->vm_file;
449 struct socket *sock = file->private_data;
450 struct sock *sk = sock->sk;
453 atomic_inc(&nlk_sk(sk)->mapped);
456 static void netlink_mm_close(struct vm_area_struct *vma)
458 struct file *file = vma->vm_file;
459 struct socket *sock = file->private_data;
460 struct sock *sk = sock->sk;
463 atomic_dec(&nlk_sk(sk)->mapped);
466 static const struct vm_operations_struct netlink_mmap_ops = {
467 .open = netlink_mm_open,
468 .close = netlink_mm_close,
471 static int netlink_mmap(struct file *file, struct socket *sock,
472 struct vm_area_struct *vma)
474 struct sock *sk = sock->sk;
475 struct netlink_sock *nlk = nlk_sk(sk);
476 struct netlink_ring *ring;
477 unsigned long start, size, expected;
484 mutex_lock(&nlk->pg_vec_lock);
487 for (ring = &nlk->rx_ring; ring <= &nlk->tx_ring; ring++) {
488 if (ring->pg_vec == NULL)
490 expected += ring->pg_vec_len * ring->pg_vec_pages * PAGE_SIZE;
496 size = vma->vm_end - vma->vm_start;
497 if (size != expected)
500 start = vma->vm_start;
501 for (ring = &nlk->rx_ring; ring <= &nlk->tx_ring; ring++) {
502 if (ring->pg_vec == NULL)
505 for (i = 0; i < ring->pg_vec_len; i++) {
507 void *kaddr = ring->pg_vec[i];
510 for (pg_num = 0; pg_num < ring->pg_vec_pages; pg_num++) {
511 page = pgvec_to_page(kaddr);
512 err = vm_insert_page(vma, start, page);
521 atomic_inc(&nlk->mapped);
522 vma->vm_ops = &netlink_mmap_ops;
525 mutex_unlock(&nlk->pg_vec_lock);
529 static void netlink_frame_flush_dcache(const struct nl_mmap_hdr *hdr)
531 #if ARCH_IMPLEMENTS_FLUSH_DCACHE_PAGE == 1
532 struct page *p_start, *p_end;
534 /* First page is flushed through netlink_{get,set}_status */
535 p_start = pgvec_to_page(hdr + PAGE_SIZE);
536 p_end = pgvec_to_page((void *)hdr + NL_MMAP_HDRLEN + hdr->nm_len - 1);
537 while (p_start <= p_end) {
538 flush_dcache_page(p_start);
544 static enum nl_mmap_status netlink_get_status(const struct nl_mmap_hdr *hdr)
547 flush_dcache_page(pgvec_to_page(hdr));
548 return hdr->nm_status;
551 static void netlink_set_status(struct nl_mmap_hdr *hdr,
552 enum nl_mmap_status status)
554 hdr->nm_status = status;
555 flush_dcache_page(pgvec_to_page(hdr));
559 static struct nl_mmap_hdr *
560 __netlink_lookup_frame(const struct netlink_ring *ring, unsigned int pos)
562 unsigned int pg_vec_pos, frame_off;
564 pg_vec_pos = pos / ring->frames_per_block;
565 frame_off = pos % ring->frames_per_block;
567 return ring->pg_vec[pg_vec_pos] + (frame_off * ring->frame_size);
570 static struct nl_mmap_hdr *
571 netlink_lookup_frame(const struct netlink_ring *ring, unsigned int pos,
572 enum nl_mmap_status status)
574 struct nl_mmap_hdr *hdr;
576 hdr = __netlink_lookup_frame(ring, pos);
577 if (netlink_get_status(hdr) != status)
583 static struct nl_mmap_hdr *
584 netlink_current_frame(const struct netlink_ring *ring,
585 enum nl_mmap_status status)
587 return netlink_lookup_frame(ring, ring->head, status);
590 static struct nl_mmap_hdr *
591 netlink_previous_frame(const struct netlink_ring *ring,
592 enum nl_mmap_status status)
596 prev = ring->head ? ring->head - 1 : ring->frame_max;
597 return netlink_lookup_frame(ring, prev, status);
600 static void netlink_increment_head(struct netlink_ring *ring)
602 ring->head = ring->head != ring->frame_max ? ring->head + 1 : 0;
605 static void netlink_forward_ring(struct netlink_ring *ring)
607 unsigned int head = ring->head, pos = head;
608 const struct nl_mmap_hdr *hdr;
611 hdr = __netlink_lookup_frame(ring, pos);
612 if (hdr->nm_status == NL_MMAP_STATUS_UNUSED)
614 if (hdr->nm_status != NL_MMAP_STATUS_SKIP)
616 netlink_increment_head(ring);
617 } while (ring->head != head);
620 static bool netlink_dump_space(struct netlink_sock *nlk)
622 struct netlink_ring *ring = &nlk->rx_ring;
623 struct nl_mmap_hdr *hdr;
626 hdr = netlink_current_frame(ring, NL_MMAP_STATUS_UNUSED);
630 n = ring->head + ring->frame_max / 2;
631 if (n > ring->frame_max)
632 n -= ring->frame_max;
634 hdr = __netlink_lookup_frame(ring, n);
636 return hdr->nm_status == NL_MMAP_STATUS_UNUSED;
639 static unsigned int netlink_poll(struct file *file, struct socket *sock,
642 struct sock *sk = sock->sk;
643 struct netlink_sock *nlk = nlk_sk(sk);
647 if (nlk->rx_ring.pg_vec != NULL) {
648 /* Memory mapped sockets don't call recvmsg(), so flow control
649 * for dumps is performed here. A dump is allowed to continue
650 * if at least half the ring is unused.
652 while (nlk->cb_running && netlink_dump_space(nlk)) {
653 err = netlink_dump(sk);
656 sk->sk_error_report(sk);
660 netlink_rcv_wake(sk);
663 mask = datagram_poll(file, sock, wait);
665 spin_lock_bh(&sk->sk_receive_queue.lock);
666 if (nlk->rx_ring.pg_vec) {
667 netlink_forward_ring(&nlk->rx_ring);
668 if (!netlink_previous_frame(&nlk->rx_ring, NL_MMAP_STATUS_UNUSED))
669 mask |= POLLIN | POLLRDNORM;
671 spin_unlock_bh(&sk->sk_receive_queue.lock);
673 spin_lock_bh(&sk->sk_write_queue.lock);
674 if (nlk->tx_ring.pg_vec) {
675 if (netlink_current_frame(&nlk->tx_ring, NL_MMAP_STATUS_UNUSED))
676 mask |= POLLOUT | POLLWRNORM;
678 spin_unlock_bh(&sk->sk_write_queue.lock);
683 static struct nl_mmap_hdr *netlink_mmap_hdr(struct sk_buff *skb)
685 return (struct nl_mmap_hdr *)(skb->head - NL_MMAP_HDRLEN);
688 static void netlink_ring_setup_skb(struct sk_buff *skb, struct sock *sk,
689 struct netlink_ring *ring,
690 struct nl_mmap_hdr *hdr)
695 size = ring->frame_size - NL_MMAP_HDRLEN;
696 data = (void *)hdr + NL_MMAP_HDRLEN;
700 skb_reset_tail_pointer(skb);
701 skb->end = skb->tail + size;
704 skb->destructor = netlink_skb_destructor;
705 NETLINK_CB(skb).flags |= NETLINK_SKB_MMAPED;
706 NETLINK_CB(skb).sk = sk;
709 static int netlink_mmap_sendmsg(struct sock *sk, struct msghdr *msg,
710 u32 dst_portid, u32 dst_group,
711 struct sock_iocb *siocb)
713 struct netlink_sock *nlk = nlk_sk(sk);
714 struct netlink_ring *ring;
715 struct nl_mmap_hdr *hdr;
719 int err = 0, len = 0;
721 /* Netlink messages are validated by the receiver before processing.
722 * In order to avoid userspace changing the contents of the message
723 * after validation, the socket and the ring may only be used by a
724 * single process, otherwise we fall back to copying.
726 if (atomic_long_read(&sk->sk_socket->file->f_count) > 1 ||
727 atomic_read(&nlk->mapped) > 1)
730 mutex_lock(&nlk->pg_vec_lock);
732 ring = &nlk->tx_ring;
733 maxlen = ring->frame_size - NL_MMAP_HDRLEN;
736 hdr = netlink_current_frame(ring, NL_MMAP_STATUS_VALID);
738 if (!(msg->msg_flags & MSG_DONTWAIT) &&
739 atomic_read(&nlk->tx_ring.pending))
743 if (hdr->nm_len > maxlen) {
748 netlink_frame_flush_dcache(hdr);
750 if (likely(dst_portid == 0 && dst_group == 0 && excl)) {
751 skb = alloc_skb_head(GFP_KERNEL);
757 netlink_ring_setup_skb(skb, sk, ring, hdr);
758 NETLINK_CB(skb).flags |= NETLINK_SKB_TX;
759 __skb_put(skb, hdr->nm_len);
760 netlink_set_status(hdr, NL_MMAP_STATUS_RESERVED);
761 atomic_inc(&ring->pending);
763 skb = alloc_skb(hdr->nm_len, GFP_KERNEL);
768 __skb_put(skb, hdr->nm_len);
769 memcpy(skb->data, (void *)hdr + NL_MMAP_HDRLEN, hdr->nm_len);
770 netlink_set_status(hdr, NL_MMAP_STATUS_UNUSED);
773 netlink_increment_head(ring);
775 NETLINK_CB(skb).portid = nlk->portid;
776 NETLINK_CB(skb).dst_group = dst_group;
777 NETLINK_CB(skb).creds = siocb->scm->creds;
779 err = security_netlink_send(sk, skb);
785 if (unlikely(dst_group)) {
786 atomic_inc(&skb->users);
787 netlink_broadcast(sk, skb, dst_portid, dst_group,
790 err = netlink_unicast(sk, skb, dst_portid,
791 msg->msg_flags & MSG_DONTWAIT);
796 } while (hdr != NULL ||
797 (!(msg->msg_flags & MSG_DONTWAIT) &&
798 atomic_read(&nlk->tx_ring.pending)));
803 mutex_unlock(&nlk->pg_vec_lock);
807 static void netlink_queue_mmaped_skb(struct sock *sk, struct sk_buff *skb)
809 struct nl_mmap_hdr *hdr;
811 hdr = netlink_mmap_hdr(skb);
812 hdr->nm_len = skb->len;
813 hdr->nm_group = NETLINK_CB(skb).dst_group;
814 hdr->nm_pid = NETLINK_CB(skb).creds.pid;
815 hdr->nm_uid = from_kuid(sk_user_ns(sk), NETLINK_CB(skb).creds.uid);
816 hdr->nm_gid = from_kgid(sk_user_ns(sk), NETLINK_CB(skb).creds.gid);
817 netlink_frame_flush_dcache(hdr);
818 netlink_set_status(hdr, NL_MMAP_STATUS_VALID);
820 NETLINK_CB(skb).flags |= NETLINK_SKB_DELIVERED;
824 static void netlink_ring_set_copied(struct sock *sk, struct sk_buff *skb)
826 struct netlink_sock *nlk = nlk_sk(sk);
827 struct netlink_ring *ring = &nlk->rx_ring;
828 struct nl_mmap_hdr *hdr;
830 spin_lock_bh(&sk->sk_receive_queue.lock);
831 hdr = netlink_current_frame(ring, NL_MMAP_STATUS_UNUSED);
833 spin_unlock_bh(&sk->sk_receive_queue.lock);
838 netlink_increment_head(ring);
839 __skb_queue_tail(&sk->sk_receive_queue, skb);
840 spin_unlock_bh(&sk->sk_receive_queue.lock);
842 hdr->nm_len = skb->len;
843 hdr->nm_group = NETLINK_CB(skb).dst_group;
844 hdr->nm_pid = NETLINK_CB(skb).creds.pid;
845 hdr->nm_uid = from_kuid(sk_user_ns(sk), NETLINK_CB(skb).creds.uid);
846 hdr->nm_gid = from_kgid(sk_user_ns(sk), NETLINK_CB(skb).creds.gid);
847 netlink_set_status(hdr, NL_MMAP_STATUS_COPY);
850 #else /* CONFIG_NETLINK_MMAP */
851 #define netlink_skb_is_mmaped(skb) false
852 #define netlink_rx_is_mmaped(sk) false
853 #define netlink_tx_is_mmaped(sk) false
854 #define netlink_mmap sock_no_mmap
855 #define netlink_poll datagram_poll
856 #define netlink_mmap_sendmsg(sk, msg, dst_portid, dst_group, siocb) 0
857 #endif /* CONFIG_NETLINK_MMAP */
859 static void netlink_skb_destructor(struct sk_buff *skb)
861 #ifdef CONFIG_NETLINK_MMAP
862 struct nl_mmap_hdr *hdr;
863 struct netlink_ring *ring;
866 /* If a packet from the kernel to userspace was freed because of an
867 * error without being delivered to userspace, the kernel must reset
868 * the status. In the direction userspace to kernel, the status is
869 * always reset here after the packet was processed and freed.
871 if (netlink_skb_is_mmaped(skb)) {
872 hdr = netlink_mmap_hdr(skb);
873 sk = NETLINK_CB(skb).sk;
875 if (NETLINK_CB(skb).flags & NETLINK_SKB_TX) {
876 netlink_set_status(hdr, NL_MMAP_STATUS_UNUSED);
877 ring = &nlk_sk(sk)->tx_ring;
879 if (!(NETLINK_CB(skb).flags & NETLINK_SKB_DELIVERED)) {
881 netlink_set_status(hdr, NL_MMAP_STATUS_VALID);
883 ring = &nlk_sk(sk)->rx_ring;
886 WARN_ON(atomic_read(&ring->pending) == 0);
887 atomic_dec(&ring->pending);
893 if (is_vmalloc_addr(skb->head)) {
895 !atomic_dec_return(&(skb_shinfo(skb)->dataref)))
904 static void netlink_skb_set_owner_r(struct sk_buff *skb, struct sock *sk)
906 WARN_ON(skb->sk != NULL);
908 skb->destructor = netlink_skb_destructor;
909 atomic_add(skb->truesize, &sk->sk_rmem_alloc);
910 sk_mem_charge(sk, skb->truesize);
913 static void netlink_sock_destruct(struct sock *sk)
915 struct netlink_sock *nlk = nlk_sk(sk);
917 if (nlk->cb_running) {
919 nlk->cb.done(&nlk->cb);
921 module_put(nlk->cb.module);
922 kfree_skb(nlk->cb.skb);
925 skb_queue_purge(&sk->sk_receive_queue);
926 #ifdef CONFIG_NETLINK_MMAP
928 struct nl_mmap_req req;
930 memset(&req, 0, sizeof(req));
931 if (nlk->rx_ring.pg_vec)
932 netlink_set_ring(sk, &req, true, false);
933 memset(&req, 0, sizeof(req));
934 if (nlk->tx_ring.pg_vec)
935 netlink_set_ring(sk, &req, true, true);
937 #endif /* CONFIG_NETLINK_MMAP */
939 if (!sock_flag(sk, SOCK_DEAD)) {
940 printk(KERN_ERR "Freeing alive netlink socket %p\n", sk);
944 WARN_ON(atomic_read(&sk->sk_rmem_alloc));
945 WARN_ON(atomic_read(&sk->sk_wmem_alloc));
946 WARN_ON(nlk_sk(sk)->groups);
949 /* This lock without WQ_FLAG_EXCLUSIVE is good on UP and it is _very_ bad on
950 * SMP. Look, when several writers sleep and reader wakes them up, all but one
951 * immediately hit write lock and grab all the cpus. Exclusive sleep solves
952 * this, _but_ remember, it adds useless work on UP machines.
955 void netlink_table_grab(void)
956 __acquires(nl_table_lock)
960 write_lock_irq(&nl_table_lock);
962 if (atomic_read(&nl_table_users)) {
963 DECLARE_WAITQUEUE(wait, current);
965 add_wait_queue_exclusive(&nl_table_wait, &wait);
967 set_current_state(TASK_UNINTERRUPTIBLE);
968 if (atomic_read(&nl_table_users) == 0)
970 write_unlock_irq(&nl_table_lock);
972 write_lock_irq(&nl_table_lock);
975 __set_current_state(TASK_RUNNING);
976 remove_wait_queue(&nl_table_wait, &wait);
980 void netlink_table_ungrab(void)
981 __releases(nl_table_lock)
983 write_unlock_irq(&nl_table_lock);
984 wake_up(&nl_table_wait);
988 netlink_lock_table(void)
990 /* read_lock() synchronizes us to netlink_table_grab */
992 read_lock(&nl_table_lock);
993 atomic_inc(&nl_table_users);
994 read_unlock(&nl_table_lock);
998 netlink_unlock_table(void)
1000 if (atomic_dec_and_test(&nl_table_users))
1001 wake_up(&nl_table_wait);
1004 struct netlink_compare_arg
1010 static bool netlink_compare(void *ptr, void *arg)
1012 struct netlink_compare_arg *x = arg;
1013 struct sock *sk = ptr;
1015 return nlk_sk(sk)->portid == x->portid &&
1016 net_eq(sock_net(sk), x->net);
1019 static struct sock *__netlink_lookup(struct netlink_table *table, u32 portid,
1022 struct netlink_compare_arg arg = {
1028 hash = rhashtable_hashfn(&table->hash, &portid, sizeof(portid));
1030 return rhashtable_lookup_compare(&table->hash, hash,
1031 &netlink_compare, &arg);
1034 static struct sock *netlink_lookup(struct net *net, int protocol, u32 portid)
1036 struct netlink_table *table = &nl_table[protocol];
1039 read_lock(&nl_table_lock);
1041 sk = __netlink_lookup(table, portid, net);
1045 read_unlock(&nl_table_lock);
1050 static const struct proto_ops netlink_ops;
1053 netlink_update_listeners(struct sock *sk)
1055 struct netlink_table *tbl = &nl_table[sk->sk_protocol];
1058 struct listeners *listeners;
1060 listeners = nl_deref_protected(tbl->listeners);
1064 for (i = 0; i < NLGRPLONGS(tbl->groups); i++) {
1066 sk_for_each_bound(sk, &tbl->mc_list) {
1067 if (i < NLGRPLONGS(nlk_sk(sk)->ngroups))
1068 mask |= nlk_sk(sk)->groups[i];
1070 listeners->masks[i] = mask;
1072 /* this function is only called with the netlink table "grabbed", which
1073 * makes sure updates are visible before bind or setsockopt return. */
1076 static int netlink_insert(struct sock *sk, struct net *net, u32 portid)
1078 struct netlink_table *table = &nl_table[sk->sk_protocol];
1079 int err = -EADDRINUSE;
1081 mutex_lock(&nl_sk_hash_lock);
1082 if (__netlink_lookup(table, portid, net))
1086 if (nlk_sk(sk)->portid)
1090 if (BITS_PER_LONG > 32 && unlikely(table->hash.nelems >= UINT_MAX))
1093 nlk_sk(sk)->portid = portid;
1095 rhashtable_insert(&table->hash, &nlk_sk(sk)->node, GFP_KERNEL);
1098 mutex_unlock(&nl_sk_hash_lock);
1102 static void netlink_remove(struct sock *sk)
1104 struct netlink_table *table;
1106 mutex_lock(&nl_sk_hash_lock);
1107 table = &nl_table[sk->sk_protocol];
1108 if (rhashtable_remove(&table->hash, &nlk_sk(sk)->node, GFP_KERNEL)) {
1109 WARN_ON(atomic_read(&sk->sk_refcnt) == 1);
1112 mutex_unlock(&nl_sk_hash_lock);
1114 netlink_table_grab();
1115 if (nlk_sk(sk)->subscriptions)
1116 __sk_del_bind_node(sk);
1117 netlink_table_ungrab();
1120 static struct proto netlink_proto = {
1122 .owner = THIS_MODULE,
1123 .obj_size = sizeof(struct netlink_sock),
1126 static int __netlink_create(struct net *net, struct socket *sock,
1127 struct mutex *cb_mutex, int protocol)
1130 struct netlink_sock *nlk;
1132 sock->ops = &netlink_ops;
1134 sk = sk_alloc(net, PF_NETLINK, GFP_KERNEL, &netlink_proto);
1138 sock_init_data(sock, sk);
1142 nlk->cb_mutex = cb_mutex;
1144 nlk->cb_mutex = &nlk->cb_def_mutex;
1145 mutex_init(nlk->cb_mutex);
1147 init_waitqueue_head(&nlk->wait);
1148 #ifdef CONFIG_NETLINK_MMAP
1149 mutex_init(&nlk->pg_vec_lock);
1152 sk->sk_destruct = netlink_sock_destruct;
1153 sk->sk_protocol = protocol;
1157 static int netlink_create(struct net *net, struct socket *sock, int protocol,
1160 struct module *module = NULL;
1161 struct mutex *cb_mutex;
1162 struct netlink_sock *nlk;
1163 int (*bind)(int group);
1164 void (*unbind)(int group);
1167 sock->state = SS_UNCONNECTED;
1169 if (sock->type != SOCK_RAW && sock->type != SOCK_DGRAM)
1170 return -ESOCKTNOSUPPORT;
1172 if (protocol < 0 || protocol >= MAX_LINKS)
1173 return -EPROTONOSUPPORT;
1175 netlink_lock_table();
1176 #ifdef CONFIG_MODULES
1177 if (!nl_table[protocol].registered) {
1178 netlink_unlock_table();
1179 request_module("net-pf-%d-proto-%d", PF_NETLINK, protocol);
1180 netlink_lock_table();
1183 if (nl_table[protocol].registered &&
1184 try_module_get(nl_table[protocol].module))
1185 module = nl_table[protocol].module;
1187 err = -EPROTONOSUPPORT;
1188 cb_mutex = nl_table[protocol].cb_mutex;
1189 bind = nl_table[protocol].bind;
1190 unbind = nl_table[protocol].unbind;
1191 netlink_unlock_table();
1196 err = __netlink_create(net, sock, cb_mutex, protocol);
1201 sock_prot_inuse_add(net, &netlink_proto, 1);
1204 nlk = nlk_sk(sock->sk);
1205 nlk->module = module;
1206 nlk->netlink_bind = bind;
1207 nlk->netlink_unbind = unbind;
1216 static int netlink_release(struct socket *sock)
1218 struct sock *sk = sock->sk;
1219 struct netlink_sock *nlk;
1229 * OK. Socket is unlinked, any packets that arrive now
1234 wake_up_interruptible_all(&nlk->wait);
1236 skb_queue_purge(&sk->sk_write_queue);
1239 struct netlink_notify n = {
1240 .net = sock_net(sk),
1241 .protocol = sk->sk_protocol,
1242 .portid = nlk->portid,
1244 atomic_notifier_call_chain(&netlink_chain,
1245 NETLINK_URELEASE, &n);
1248 module_put(nlk->module);
1250 netlink_table_grab();
1251 if (netlink_is_kernel(sk)) {
1252 BUG_ON(nl_table[sk->sk_protocol].registered == 0);
1253 if (--nl_table[sk->sk_protocol].registered == 0) {
1254 struct listeners *old;
1256 old = nl_deref_protected(nl_table[sk->sk_protocol].listeners);
1257 RCU_INIT_POINTER(nl_table[sk->sk_protocol].listeners, NULL);
1258 kfree_rcu(old, rcu);
1259 nl_table[sk->sk_protocol].module = NULL;
1260 nl_table[sk->sk_protocol].bind = NULL;
1261 nl_table[sk->sk_protocol].unbind = NULL;
1262 nl_table[sk->sk_protocol].flags = 0;
1263 nl_table[sk->sk_protocol].registered = 0;
1265 } else if (nlk->subscriptions) {
1266 netlink_update_listeners(sk);
1268 netlink_table_ungrab();
1274 sock_prot_inuse_add(sock_net(sk), &netlink_proto, -1);
1280 static int netlink_autobind(struct socket *sock)
1282 struct sock *sk = sock->sk;
1283 struct net *net = sock_net(sk);
1284 struct netlink_table *table = &nl_table[sk->sk_protocol];
1285 s32 portid = task_tgid_vnr(current);
1287 static s32 rover = -4097;
1291 netlink_table_grab();
1293 if (__netlink_lookup(table, portid, net)) {
1294 /* Bind collision, search negative portid values. */
1299 netlink_table_ungrab();
1303 netlink_table_ungrab();
1305 err = netlink_insert(sk, net, portid);
1306 if (err == -EADDRINUSE)
1309 /* If 2 threads race to autobind, that is fine. */
1317 * __netlink_ns_capable - General netlink message capability test
1318 * @nsp: NETLINK_CB of the socket buffer holding a netlink command from userspace.
1319 * @user_ns: The user namespace of the capability to use
1320 * @cap: The capability to use
1322 * Test to see if the opener of the socket we received the message
1323 * from had when the netlink socket was created and the sender of the
1324 * message has has the capability @cap in the user namespace @user_ns.
1326 bool __netlink_ns_capable(const struct netlink_skb_parms *nsp,
1327 struct user_namespace *user_ns, int cap)
1329 return ((nsp->flags & NETLINK_SKB_DST) ||
1330 file_ns_capable(nsp->sk->sk_socket->file, user_ns, cap)) &&
1331 ns_capable(user_ns, cap);
1333 EXPORT_SYMBOL(__netlink_ns_capable);
1336 * netlink_ns_capable - General netlink message capability test
1337 * @skb: socket buffer holding a netlink command from userspace
1338 * @user_ns: The user namespace of the capability to use
1339 * @cap: The capability to use
1341 * Test to see if the opener of the socket we received the message
1342 * from had when the netlink socket was created and the sender of the
1343 * message has has the capability @cap in the user namespace @user_ns.
1345 bool netlink_ns_capable(const struct sk_buff *skb,
1346 struct user_namespace *user_ns, int cap)
1348 return __netlink_ns_capable(&NETLINK_CB(skb), user_ns, cap);
1350 EXPORT_SYMBOL(netlink_ns_capable);
1353 * netlink_capable - Netlink global message capability test
1354 * @skb: socket buffer holding a netlink command from userspace
1355 * @cap: The capability to use
1357 * Test to see if the opener of the socket we received the message
1358 * from had when the netlink socket was created and the sender of the
1359 * message has has the capability @cap in all user namespaces.
1361 bool netlink_capable(const struct sk_buff *skb, int cap)
1363 return netlink_ns_capable(skb, &init_user_ns, cap);
1365 EXPORT_SYMBOL(netlink_capable);
1368 * netlink_net_capable - Netlink network namespace message capability test
1369 * @skb: socket buffer holding a netlink command from userspace
1370 * @cap: The capability to use
1372 * Test to see if the opener of the socket we received the message
1373 * from had when the netlink socket was created and the sender of the
1374 * message has has the capability @cap over the network namespace of
1375 * the socket we received the message from.
1377 bool netlink_net_capable(const struct sk_buff *skb, int cap)
1379 return netlink_ns_capable(skb, sock_net(skb->sk)->user_ns, cap);
1381 EXPORT_SYMBOL(netlink_net_capable);
1383 static inline int netlink_allowed(const struct socket *sock, unsigned int flag)
1385 return (nl_table[sock->sk->sk_protocol].flags & flag) ||
1386 ns_capable(sock_net(sock->sk)->user_ns, CAP_NET_ADMIN);
1390 netlink_update_subscriptions(struct sock *sk, unsigned int subscriptions)
1392 struct netlink_sock *nlk = nlk_sk(sk);
1394 if (nlk->subscriptions && !subscriptions)
1395 __sk_del_bind_node(sk);
1396 else if (!nlk->subscriptions && subscriptions)
1397 sk_add_bind_node(sk, &nl_table[sk->sk_protocol].mc_list);
1398 nlk->subscriptions = subscriptions;
1401 static int netlink_realloc_groups(struct sock *sk)
1403 struct netlink_sock *nlk = nlk_sk(sk);
1404 unsigned int groups;
1405 unsigned long *new_groups;
1408 netlink_table_grab();
1410 groups = nl_table[sk->sk_protocol].groups;
1411 if (!nl_table[sk->sk_protocol].registered) {
1416 if (nlk->ngroups >= groups)
1419 new_groups = krealloc(nlk->groups, NLGRPSZ(groups), GFP_ATOMIC);
1420 if (new_groups == NULL) {
1424 memset((char *)new_groups + NLGRPSZ(nlk->ngroups), 0,
1425 NLGRPSZ(groups) - NLGRPSZ(nlk->ngroups));
1427 nlk->groups = new_groups;
1428 nlk->ngroups = groups;
1430 netlink_table_ungrab();
1434 static void netlink_unbind(int group, long unsigned int groups,
1435 struct netlink_sock *nlk)
1439 if (!nlk->netlink_unbind)
1442 for (undo = 0; undo < group; undo++)
1443 if (test_bit(group, &groups))
1444 nlk->netlink_unbind(undo);
1447 static int netlink_bind(struct socket *sock, struct sockaddr *addr,
1450 struct sock *sk = sock->sk;
1451 struct net *net = sock_net(sk);
1452 struct netlink_sock *nlk = nlk_sk(sk);
1453 struct sockaddr_nl *nladdr = (struct sockaddr_nl *)addr;
1455 long unsigned int groups = nladdr->nl_groups;
1457 if (addr_len < sizeof(struct sockaddr_nl))
1460 if (nladdr->nl_family != AF_NETLINK)
1463 /* Only superuser is allowed to listen multicasts */
1465 if (!netlink_allowed(sock, NL_CFG_F_NONROOT_RECV))
1467 err = netlink_realloc_groups(sk);
1473 if (nladdr->nl_pid != nlk->portid)
1476 if (nlk->netlink_bind && groups) {
1479 for (group = 0; group < nlk->ngroups; group++) {
1480 if (!test_bit(group, &groups))
1482 err = nlk->netlink_bind(group);
1485 netlink_unbind(group, groups, nlk);
1491 err = nladdr->nl_pid ?
1492 netlink_insert(sk, net, nladdr->nl_pid) :
1493 netlink_autobind(sock);
1495 netlink_unbind(nlk->ngroups - 1, groups, nlk);
1500 if (!groups && (nlk->groups == NULL || !(u32)nlk->groups[0]))
1503 netlink_table_grab();
1504 netlink_update_subscriptions(sk, nlk->subscriptions +
1506 hweight32(nlk->groups[0]));
1507 nlk->groups[0] = (nlk->groups[0] & ~0xffffffffUL) | groups;
1508 netlink_update_listeners(sk);
1509 netlink_table_ungrab();
1514 static int netlink_connect(struct socket *sock, struct sockaddr *addr,
1515 int alen, int flags)
1518 struct sock *sk = sock->sk;
1519 struct netlink_sock *nlk = nlk_sk(sk);
1520 struct sockaddr_nl *nladdr = (struct sockaddr_nl *)addr;
1522 if (alen < sizeof(addr->sa_family))
1525 if (addr->sa_family == AF_UNSPEC) {
1526 sk->sk_state = NETLINK_UNCONNECTED;
1527 nlk->dst_portid = 0;
1531 if (addr->sa_family != AF_NETLINK)
1534 if ((nladdr->nl_groups || nladdr->nl_pid) &&
1535 !netlink_allowed(sock, NL_CFG_F_NONROOT_SEND))
1539 err = netlink_autobind(sock);
1542 sk->sk_state = NETLINK_CONNECTED;
1543 nlk->dst_portid = nladdr->nl_pid;
1544 nlk->dst_group = ffs(nladdr->nl_groups);
1550 static int netlink_getname(struct socket *sock, struct sockaddr *addr,
1551 int *addr_len, int peer)
1553 struct sock *sk = sock->sk;
1554 struct netlink_sock *nlk = nlk_sk(sk);
1555 DECLARE_SOCKADDR(struct sockaddr_nl *, nladdr, addr);
1557 nladdr->nl_family = AF_NETLINK;
1559 *addr_len = sizeof(*nladdr);
1562 nladdr->nl_pid = nlk->dst_portid;
1563 nladdr->nl_groups = netlink_group_mask(nlk->dst_group);
1565 nladdr->nl_pid = nlk->portid;
1566 nladdr->nl_groups = nlk->groups ? nlk->groups[0] : 0;
1571 static struct sock *netlink_getsockbyportid(struct sock *ssk, u32 portid)
1574 struct netlink_sock *nlk;
1576 sock = netlink_lookup(sock_net(ssk), ssk->sk_protocol, portid);
1578 return ERR_PTR(-ECONNREFUSED);
1580 /* Don't bother queuing skb if kernel socket has no input function */
1582 if (sock->sk_state == NETLINK_CONNECTED &&
1583 nlk->dst_portid != nlk_sk(ssk)->portid) {
1585 return ERR_PTR(-ECONNREFUSED);
1590 struct sock *netlink_getsockbyfilp(struct file *filp)
1592 struct inode *inode = file_inode(filp);
1595 if (!S_ISSOCK(inode->i_mode))
1596 return ERR_PTR(-ENOTSOCK);
1598 sock = SOCKET_I(inode)->sk;
1599 if (sock->sk_family != AF_NETLINK)
1600 return ERR_PTR(-EINVAL);
1606 static struct sk_buff *netlink_alloc_large_skb(unsigned int size,
1609 struct sk_buff *skb;
1612 if (size <= NLMSG_GOODSIZE || broadcast)
1613 return alloc_skb(size, GFP_KERNEL);
1615 size = SKB_DATA_ALIGN(size) +
1616 SKB_DATA_ALIGN(sizeof(struct skb_shared_info));
1618 data = vmalloc(size);
1622 skb = build_skb(data, size);
1627 skb->destructor = netlink_skb_destructor;
1634 * Attach a skb to a netlink socket.
1635 * The caller must hold a reference to the destination socket. On error, the
1636 * reference is dropped. The skb is not send to the destination, just all
1637 * all error checks are performed and memory in the queue is reserved.
1639 * < 0: error. skb freed, reference to sock dropped.
1641 * 1: repeat lookup - reference dropped while waiting for socket memory.
1643 int netlink_attachskb(struct sock *sk, struct sk_buff *skb,
1644 long *timeo, struct sock *ssk)
1646 struct netlink_sock *nlk;
1650 if ((atomic_read(&sk->sk_rmem_alloc) > sk->sk_rcvbuf ||
1651 test_bit(NETLINK_CONGESTED, &nlk->state)) &&
1652 !netlink_skb_is_mmaped(skb)) {
1653 DECLARE_WAITQUEUE(wait, current);
1655 if (!ssk || netlink_is_kernel(ssk))
1656 netlink_overrun(sk);
1662 __set_current_state(TASK_INTERRUPTIBLE);
1663 add_wait_queue(&nlk->wait, &wait);
1665 if ((atomic_read(&sk->sk_rmem_alloc) > sk->sk_rcvbuf ||
1666 test_bit(NETLINK_CONGESTED, &nlk->state)) &&
1667 !sock_flag(sk, SOCK_DEAD))
1668 *timeo = schedule_timeout(*timeo);
1670 __set_current_state(TASK_RUNNING);
1671 remove_wait_queue(&nlk->wait, &wait);
1674 if (signal_pending(current)) {
1676 return sock_intr_errno(*timeo);
1680 netlink_skb_set_owner_r(skb, sk);
1684 static int __netlink_sendskb(struct sock *sk, struct sk_buff *skb)
1688 netlink_deliver_tap(skb);
1690 #ifdef CONFIG_NETLINK_MMAP
1691 if (netlink_skb_is_mmaped(skb))
1692 netlink_queue_mmaped_skb(sk, skb);
1693 else if (netlink_rx_is_mmaped(sk))
1694 netlink_ring_set_copied(sk, skb);
1696 #endif /* CONFIG_NETLINK_MMAP */
1697 skb_queue_tail(&sk->sk_receive_queue, skb);
1698 sk->sk_data_ready(sk);
1702 int netlink_sendskb(struct sock *sk, struct sk_buff *skb)
1704 int len = __netlink_sendskb(sk, skb);
1710 void netlink_detachskb(struct sock *sk, struct sk_buff *skb)
1716 static struct sk_buff *netlink_trim(struct sk_buff *skb, gfp_t allocation)
1720 WARN_ON(skb->sk != NULL);
1721 if (netlink_skb_is_mmaped(skb))
1724 delta = skb->end - skb->tail;
1725 if (is_vmalloc_addr(skb->head) || delta * 2 < skb->truesize)
1728 if (skb_shared(skb)) {
1729 struct sk_buff *nskb = skb_clone(skb, allocation);
1736 if (!pskb_expand_head(skb, 0, -delta, allocation))
1737 skb->truesize -= delta;
1742 static int netlink_unicast_kernel(struct sock *sk, struct sk_buff *skb,
1746 struct netlink_sock *nlk = nlk_sk(sk);
1748 ret = -ECONNREFUSED;
1749 if (nlk->netlink_rcv != NULL) {
1751 netlink_skb_set_owner_r(skb, sk);
1752 NETLINK_CB(skb).sk = ssk;
1753 netlink_deliver_tap_kernel(sk, ssk, skb);
1754 nlk->netlink_rcv(skb);
1763 int netlink_unicast(struct sock *ssk, struct sk_buff *skb,
1764 u32 portid, int nonblock)
1770 skb = netlink_trim(skb, gfp_any());
1772 timeo = sock_sndtimeo(ssk, nonblock);
1774 sk = netlink_getsockbyportid(ssk, portid);
1779 if (netlink_is_kernel(sk))
1780 return netlink_unicast_kernel(sk, skb, ssk);
1782 if (sk_filter(sk, skb)) {
1789 err = netlink_attachskb(sk, skb, &timeo, ssk);
1795 return netlink_sendskb(sk, skb);
1797 EXPORT_SYMBOL(netlink_unicast);
1799 struct sk_buff *netlink_alloc_skb(struct sock *ssk, unsigned int size,
1800 u32 dst_portid, gfp_t gfp_mask)
1802 #ifdef CONFIG_NETLINK_MMAP
1803 struct sock *sk = NULL;
1804 struct sk_buff *skb;
1805 struct netlink_ring *ring;
1806 struct nl_mmap_hdr *hdr;
1807 unsigned int maxlen;
1809 sk = netlink_getsockbyportid(ssk, dst_portid);
1813 ring = &nlk_sk(sk)->rx_ring;
1814 /* fast-path without atomic ops for common case: non-mmaped receiver */
1815 if (ring->pg_vec == NULL)
1818 if (ring->frame_size - NL_MMAP_HDRLEN < size)
1821 skb = alloc_skb_head(gfp_mask);
1825 spin_lock_bh(&sk->sk_receive_queue.lock);
1826 /* check again under lock */
1827 if (ring->pg_vec == NULL)
1830 /* check again under lock */
1831 maxlen = ring->frame_size - NL_MMAP_HDRLEN;
1835 netlink_forward_ring(ring);
1836 hdr = netlink_current_frame(ring, NL_MMAP_STATUS_UNUSED);
1839 netlink_ring_setup_skb(skb, sk, ring, hdr);
1840 netlink_set_status(hdr, NL_MMAP_STATUS_RESERVED);
1841 atomic_inc(&ring->pending);
1842 netlink_increment_head(ring);
1844 spin_unlock_bh(&sk->sk_receive_queue.lock);
1849 spin_unlock_bh(&sk->sk_receive_queue.lock);
1850 netlink_overrun(sk);
1857 spin_unlock_bh(&sk->sk_receive_queue.lock);
1862 return alloc_skb(size, gfp_mask);
1864 EXPORT_SYMBOL_GPL(netlink_alloc_skb);
1866 int netlink_has_listeners(struct sock *sk, unsigned int group)
1869 struct listeners *listeners;
1871 BUG_ON(!netlink_is_kernel(sk));
1874 listeners = rcu_dereference(nl_table[sk->sk_protocol].listeners);
1876 if (listeners && group - 1 < nl_table[sk->sk_protocol].groups)
1877 res = test_bit(group - 1, listeners->masks);
1883 EXPORT_SYMBOL_GPL(netlink_has_listeners);
1885 static int netlink_broadcast_deliver(struct sock *sk, struct sk_buff *skb)
1887 struct netlink_sock *nlk = nlk_sk(sk);
1889 if (atomic_read(&sk->sk_rmem_alloc) <= sk->sk_rcvbuf &&
1890 !test_bit(NETLINK_CONGESTED, &nlk->state)) {
1891 netlink_skb_set_owner_r(skb, sk);
1892 __netlink_sendskb(sk, skb);
1893 return atomic_read(&sk->sk_rmem_alloc) > (sk->sk_rcvbuf >> 1);
1898 struct netlink_broadcast_data {
1899 struct sock *exclude_sk;
1904 int delivery_failure;
1908 struct sk_buff *skb, *skb2;
1909 int (*tx_filter)(struct sock *dsk, struct sk_buff *skb, void *data);
1913 static void do_one_broadcast(struct sock *sk,
1914 struct netlink_broadcast_data *p)
1916 struct netlink_sock *nlk = nlk_sk(sk);
1919 if (p->exclude_sk == sk)
1922 if (nlk->portid == p->portid || p->group - 1 >= nlk->ngroups ||
1923 !test_bit(p->group - 1, nlk->groups))
1926 if (!net_eq(sock_net(sk), p->net))
1930 netlink_overrun(sk);
1935 if (p->skb2 == NULL) {
1936 if (skb_shared(p->skb)) {
1937 p->skb2 = skb_clone(p->skb, p->allocation);
1939 p->skb2 = skb_get(p->skb);
1941 * skb ownership may have been set when
1942 * delivered to a previous socket.
1944 skb_orphan(p->skb2);
1947 if (p->skb2 == NULL) {
1948 netlink_overrun(sk);
1949 /* Clone failed. Notify ALL listeners. */
1951 if (nlk->flags & NETLINK_BROADCAST_SEND_ERROR)
1952 p->delivery_failure = 1;
1953 } else if (p->tx_filter && p->tx_filter(sk, p->skb2, p->tx_data)) {
1956 } else if (sk_filter(sk, p->skb2)) {
1959 } else if ((val = netlink_broadcast_deliver(sk, p->skb2)) < 0) {
1960 netlink_overrun(sk);
1961 if (nlk->flags & NETLINK_BROADCAST_SEND_ERROR)
1962 p->delivery_failure = 1;
1964 p->congested |= val;
1971 int netlink_broadcast_filtered(struct sock *ssk, struct sk_buff *skb, u32 portid,
1972 u32 group, gfp_t allocation,
1973 int (*filter)(struct sock *dsk, struct sk_buff *skb, void *data),
1976 struct net *net = sock_net(ssk);
1977 struct netlink_broadcast_data info;
1980 skb = netlink_trim(skb, allocation);
1982 info.exclude_sk = ssk;
1984 info.portid = portid;
1987 info.delivery_failure = 0;
1990 info.allocation = allocation;
1993 info.tx_filter = filter;
1994 info.tx_data = filter_data;
1996 /* While we sleep in clone, do not allow to change socket list */
1998 netlink_lock_table();
2000 sk_for_each_bound(sk, &nl_table[ssk->sk_protocol].mc_list)
2001 do_one_broadcast(sk, &info);
2005 netlink_unlock_table();
2007 if (info.delivery_failure) {
2008 kfree_skb(info.skb2);
2011 consume_skb(info.skb2);
2013 if (info.delivered) {
2014 if (info.congested && (allocation & __GFP_WAIT))
2020 EXPORT_SYMBOL(netlink_broadcast_filtered);
2022 int netlink_broadcast(struct sock *ssk, struct sk_buff *skb, u32 portid,
2023 u32 group, gfp_t allocation)
2025 return netlink_broadcast_filtered(ssk, skb, portid, group, allocation,
2028 EXPORT_SYMBOL(netlink_broadcast);
2030 struct netlink_set_err_data {
2031 struct sock *exclude_sk;
2037 static int do_one_set_err(struct sock *sk, struct netlink_set_err_data *p)
2039 struct netlink_sock *nlk = nlk_sk(sk);
2042 if (sk == p->exclude_sk)
2045 if (!net_eq(sock_net(sk), sock_net(p->exclude_sk)))
2048 if (nlk->portid == p->portid || p->group - 1 >= nlk->ngroups ||
2049 !test_bit(p->group - 1, nlk->groups))
2052 if (p->code == ENOBUFS && nlk->flags & NETLINK_RECV_NO_ENOBUFS) {
2057 sk->sk_err = p->code;
2058 sk->sk_error_report(sk);
2064 * netlink_set_err - report error to broadcast listeners
2065 * @ssk: the kernel netlink socket, as returned by netlink_kernel_create()
2066 * @portid: the PORTID of a process that we want to skip (if any)
2067 * @group: the broadcast group that will notice the error
2068 * @code: error code, must be negative (as usual in kernelspace)
2070 * This function returns the number of broadcast listeners that have set the
2071 * NETLINK_RECV_NO_ENOBUFS socket option.
2073 int netlink_set_err(struct sock *ssk, u32 portid, u32 group, int code)
2075 struct netlink_set_err_data info;
2079 info.exclude_sk = ssk;
2080 info.portid = portid;
2082 /* sk->sk_err wants a positive error value */
2085 read_lock(&nl_table_lock);
2087 sk_for_each_bound(sk, &nl_table[ssk->sk_protocol].mc_list)
2088 ret += do_one_set_err(sk, &info);
2090 read_unlock(&nl_table_lock);
2093 EXPORT_SYMBOL(netlink_set_err);
2095 /* must be called with netlink table grabbed */
2096 static void netlink_update_socket_mc(struct netlink_sock *nlk,
2100 int old, new = !!is_new, subscriptions;
2102 old = test_bit(group - 1, nlk->groups);
2103 subscriptions = nlk->subscriptions - old + new;
2105 __set_bit(group - 1, nlk->groups);
2107 __clear_bit(group - 1, nlk->groups);
2108 netlink_update_subscriptions(&nlk->sk, subscriptions);
2109 netlink_update_listeners(&nlk->sk);
2112 static int netlink_setsockopt(struct socket *sock, int level, int optname,
2113 char __user *optval, unsigned int optlen)
2115 struct sock *sk = sock->sk;
2116 struct netlink_sock *nlk = nlk_sk(sk);
2117 unsigned int val = 0;
2120 if (level != SOL_NETLINK)
2121 return -ENOPROTOOPT;
2123 if (optname != NETLINK_RX_RING && optname != NETLINK_TX_RING &&
2124 optlen >= sizeof(int) &&
2125 get_user(val, (unsigned int __user *)optval))
2129 case NETLINK_PKTINFO:
2131 nlk->flags |= NETLINK_RECV_PKTINFO;
2133 nlk->flags &= ~NETLINK_RECV_PKTINFO;
2136 case NETLINK_ADD_MEMBERSHIP:
2137 case NETLINK_DROP_MEMBERSHIP: {
2138 if (!netlink_allowed(sock, NL_CFG_F_NONROOT_RECV))
2140 err = netlink_realloc_groups(sk);
2143 if (!val || val - 1 >= nlk->ngroups)
2145 if (optname == NETLINK_ADD_MEMBERSHIP && nlk->netlink_bind) {
2146 err = nlk->netlink_bind(val);
2150 netlink_table_grab();
2151 netlink_update_socket_mc(nlk, val,
2152 optname == NETLINK_ADD_MEMBERSHIP);
2153 netlink_table_ungrab();
2154 if (optname == NETLINK_DROP_MEMBERSHIP && nlk->netlink_unbind)
2155 nlk->netlink_unbind(val);
2160 case NETLINK_BROADCAST_ERROR:
2162 nlk->flags |= NETLINK_BROADCAST_SEND_ERROR;
2164 nlk->flags &= ~NETLINK_BROADCAST_SEND_ERROR;
2167 case NETLINK_NO_ENOBUFS:
2169 nlk->flags |= NETLINK_RECV_NO_ENOBUFS;
2170 clear_bit(NETLINK_CONGESTED, &nlk->state);
2171 wake_up_interruptible(&nlk->wait);
2173 nlk->flags &= ~NETLINK_RECV_NO_ENOBUFS;
2177 #ifdef CONFIG_NETLINK_MMAP
2178 case NETLINK_RX_RING:
2179 case NETLINK_TX_RING: {
2180 struct nl_mmap_req req;
2182 /* Rings might consume more memory than queue limits, require
2185 if (!capable(CAP_NET_ADMIN))
2187 if (optlen < sizeof(req))
2189 if (copy_from_user(&req, optval, sizeof(req)))
2191 err = netlink_set_ring(sk, &req, false,
2192 optname == NETLINK_TX_RING);
2195 #endif /* CONFIG_NETLINK_MMAP */
2202 static int netlink_getsockopt(struct socket *sock, int level, int optname,
2203 char __user *optval, int __user *optlen)
2205 struct sock *sk = sock->sk;
2206 struct netlink_sock *nlk = nlk_sk(sk);
2209 if (level != SOL_NETLINK)
2210 return -ENOPROTOOPT;
2212 if (get_user(len, optlen))
2218 case NETLINK_PKTINFO:
2219 if (len < sizeof(int))
2222 val = nlk->flags & NETLINK_RECV_PKTINFO ? 1 : 0;
2223 if (put_user(len, optlen) ||
2224 put_user(val, optval))
2228 case NETLINK_BROADCAST_ERROR:
2229 if (len < sizeof(int))
2232 val = nlk->flags & NETLINK_BROADCAST_SEND_ERROR ? 1 : 0;
2233 if (put_user(len, optlen) ||
2234 put_user(val, optval))
2238 case NETLINK_NO_ENOBUFS:
2239 if (len < sizeof(int))
2242 val = nlk->flags & NETLINK_RECV_NO_ENOBUFS ? 1 : 0;
2243 if (put_user(len, optlen) ||
2244 put_user(val, optval))
2254 static void netlink_cmsg_recv_pktinfo(struct msghdr *msg, struct sk_buff *skb)
2256 struct nl_pktinfo info;
2258 info.group = NETLINK_CB(skb).dst_group;
2259 put_cmsg(msg, SOL_NETLINK, NETLINK_PKTINFO, sizeof(info), &info);
2262 static int netlink_sendmsg(struct kiocb *kiocb, struct socket *sock,
2263 struct msghdr *msg, size_t len)
2265 struct sock_iocb *siocb = kiocb_to_siocb(kiocb);
2266 struct sock *sk = sock->sk;
2267 struct netlink_sock *nlk = nlk_sk(sk);
2268 DECLARE_SOCKADDR(struct sockaddr_nl *, addr, msg->msg_name);
2271 struct sk_buff *skb;
2273 struct scm_cookie scm;
2274 u32 netlink_skb_flags = 0;
2276 if (msg->msg_flags&MSG_OOB)
2279 if (NULL == siocb->scm)
2282 err = scm_send(sock, msg, siocb->scm, true);
2286 if (msg->msg_namelen) {
2288 if (addr->nl_family != AF_NETLINK)
2290 dst_portid = addr->nl_pid;
2291 dst_group = ffs(addr->nl_groups);
2293 if ((dst_group || dst_portid) &&
2294 !netlink_allowed(sock, NL_CFG_F_NONROOT_SEND))
2296 netlink_skb_flags |= NETLINK_SKB_DST;
2298 dst_portid = nlk->dst_portid;
2299 dst_group = nlk->dst_group;
2303 err = netlink_autobind(sock);
2308 if (netlink_tx_is_mmaped(sk) &&
2309 msg->msg_iov->iov_base == NULL) {
2310 err = netlink_mmap_sendmsg(sk, msg, dst_portid, dst_group,
2316 if (len > sk->sk_sndbuf - 32)
2319 skb = netlink_alloc_large_skb(len, dst_group);
2323 NETLINK_CB(skb).portid = nlk->portid;
2324 NETLINK_CB(skb).dst_group = dst_group;
2325 NETLINK_CB(skb).creds = siocb->scm->creds;
2326 NETLINK_CB(skb).flags = netlink_skb_flags;
2329 if (memcpy_fromiovec(skb_put(skb, len), msg->msg_iov, len)) {
2334 err = security_netlink_send(sk, skb);
2341 atomic_inc(&skb->users);
2342 netlink_broadcast(sk, skb, dst_portid, dst_group, GFP_KERNEL);
2344 err = netlink_unicast(sk, skb, dst_portid, msg->msg_flags&MSG_DONTWAIT);
2347 scm_destroy(siocb->scm);
2351 static int netlink_recvmsg(struct kiocb *kiocb, struct socket *sock,
2352 struct msghdr *msg, size_t len,
2355 struct sock_iocb *siocb = kiocb_to_siocb(kiocb);
2356 struct scm_cookie scm;
2357 struct sock *sk = sock->sk;
2358 struct netlink_sock *nlk = nlk_sk(sk);
2359 int noblock = flags&MSG_DONTWAIT;
2361 struct sk_buff *skb, *data_skb;
2369 skb = skb_recv_datagram(sk, flags, noblock, &err);
2375 #ifdef CONFIG_COMPAT_NETLINK_MESSAGES
2376 if (unlikely(skb_shinfo(skb)->frag_list)) {
2378 * If this skb has a frag_list, then here that means that we
2379 * will have to use the frag_list skb's data for compat tasks
2380 * and the regular skb's data for normal (non-compat) tasks.
2382 * If we need to send the compat skb, assign it to the
2383 * 'data_skb' variable so that it will be used below for data
2384 * copying. We keep 'skb' for everything else, including
2385 * freeing both later.
2387 if (flags & MSG_CMSG_COMPAT)
2388 data_skb = skb_shinfo(skb)->frag_list;
2392 /* Record the max length of recvmsg() calls for future allocations */
2393 nlk->max_recvmsg_len = max(nlk->max_recvmsg_len, len);
2394 nlk->max_recvmsg_len = min_t(size_t, nlk->max_recvmsg_len,
2397 copied = data_skb->len;
2399 msg->msg_flags |= MSG_TRUNC;
2403 skb_reset_transport_header(data_skb);
2404 err = skb_copy_datagram_iovec(data_skb, 0, msg->msg_iov, copied);
2406 if (msg->msg_name) {
2407 DECLARE_SOCKADDR(struct sockaddr_nl *, addr, msg->msg_name);
2408 addr->nl_family = AF_NETLINK;
2410 addr->nl_pid = NETLINK_CB(skb).portid;
2411 addr->nl_groups = netlink_group_mask(NETLINK_CB(skb).dst_group);
2412 msg->msg_namelen = sizeof(*addr);
2415 if (nlk->flags & NETLINK_RECV_PKTINFO)
2416 netlink_cmsg_recv_pktinfo(msg, skb);
2418 if (NULL == siocb->scm) {
2419 memset(&scm, 0, sizeof(scm));
2422 siocb->scm->creds = *NETLINK_CREDS(skb);
2423 if (flags & MSG_TRUNC)
2424 copied = data_skb->len;
2426 skb_free_datagram(sk, skb);
2428 if (nlk->cb_running &&
2429 atomic_read(&sk->sk_rmem_alloc) <= sk->sk_rcvbuf / 2) {
2430 ret = netlink_dump(sk);
2433 sk->sk_error_report(sk);
2437 scm_recv(sock, msg, siocb->scm, flags);
2439 netlink_rcv_wake(sk);
2440 return err ? : copied;
2443 static void netlink_data_ready(struct sock *sk)
2449 * We export these functions to other modules. They provide a
2450 * complete set of kernel non-blocking support for message
2455 __netlink_kernel_create(struct net *net, int unit, struct module *module,
2456 struct netlink_kernel_cfg *cfg)
2458 struct socket *sock;
2460 struct netlink_sock *nlk;
2461 struct listeners *listeners = NULL;
2462 struct mutex *cb_mutex = cfg ? cfg->cb_mutex : NULL;
2463 unsigned int groups;
2467 if (unit < 0 || unit >= MAX_LINKS)
2470 if (sock_create_lite(PF_NETLINK, SOCK_DGRAM, unit, &sock))
2474 * We have to just have a reference on the net from sk, but don't
2475 * get_net it. Besides, we cannot get and then put the net here.
2476 * So we create one inside init_net and the move it to net.
2479 if (__netlink_create(&init_net, sock, cb_mutex, unit) < 0)
2480 goto out_sock_release_nosk;
2483 sk_change_net(sk, net);
2485 if (!cfg || cfg->groups < 32)
2488 groups = cfg->groups;
2490 listeners = kzalloc(sizeof(*listeners) + NLGRPSZ(groups), GFP_KERNEL);
2492 goto out_sock_release;
2494 sk->sk_data_ready = netlink_data_ready;
2495 if (cfg && cfg->input)
2496 nlk_sk(sk)->netlink_rcv = cfg->input;
2498 if (netlink_insert(sk, net, 0))
2499 goto out_sock_release;
2502 nlk->flags |= NETLINK_KERNEL_SOCKET;
2504 netlink_table_grab();
2505 if (!nl_table[unit].registered) {
2506 nl_table[unit].groups = groups;
2507 rcu_assign_pointer(nl_table[unit].listeners, listeners);
2508 nl_table[unit].cb_mutex = cb_mutex;
2509 nl_table[unit].module = module;
2511 nl_table[unit].bind = cfg->bind;
2512 nl_table[unit].flags = cfg->flags;
2514 nl_table[unit].compare = cfg->compare;
2516 nl_table[unit].registered = 1;
2519 nl_table[unit].registered++;
2521 netlink_table_ungrab();
2526 netlink_kernel_release(sk);
2529 out_sock_release_nosk:
2533 EXPORT_SYMBOL(__netlink_kernel_create);
2536 netlink_kernel_release(struct sock *sk)
2538 sk_release_kernel(sk);
2540 EXPORT_SYMBOL(netlink_kernel_release);
2542 int __netlink_change_ngroups(struct sock *sk, unsigned int groups)
2544 struct listeners *new, *old;
2545 struct netlink_table *tbl = &nl_table[sk->sk_protocol];
2550 if (NLGRPSZ(tbl->groups) < NLGRPSZ(groups)) {
2551 new = kzalloc(sizeof(*new) + NLGRPSZ(groups), GFP_ATOMIC);
2554 old = nl_deref_protected(tbl->listeners);
2555 memcpy(new->masks, old->masks, NLGRPSZ(tbl->groups));
2556 rcu_assign_pointer(tbl->listeners, new);
2558 kfree_rcu(old, rcu);
2560 tbl->groups = groups;
2566 * netlink_change_ngroups - change number of multicast groups
2568 * This changes the number of multicast groups that are available
2569 * on a certain netlink family. Note that it is not possible to
2570 * change the number of groups to below 32. Also note that it does
2571 * not implicitly call netlink_clear_multicast_users() when the
2572 * number of groups is reduced.
2574 * @sk: The kernel netlink socket, as returned by netlink_kernel_create().
2575 * @groups: The new number of groups.
2577 int netlink_change_ngroups(struct sock *sk, unsigned int groups)
2581 netlink_table_grab();
2582 err = __netlink_change_ngroups(sk, groups);
2583 netlink_table_ungrab();
2588 void __netlink_clear_multicast_users(struct sock *ksk, unsigned int group)
2591 struct netlink_table *tbl = &nl_table[ksk->sk_protocol];
2593 sk_for_each_bound(sk, &tbl->mc_list)
2594 netlink_update_socket_mc(nlk_sk(sk), group, 0);
2598 __nlmsg_put(struct sk_buff *skb, u32 portid, u32 seq, int type, int len, int flags)
2600 struct nlmsghdr *nlh;
2601 int size = nlmsg_msg_size(len);
2603 nlh = (struct nlmsghdr *)skb_put(skb, NLMSG_ALIGN(size));
2604 nlh->nlmsg_type = type;
2605 nlh->nlmsg_len = size;
2606 nlh->nlmsg_flags = flags;
2607 nlh->nlmsg_pid = portid;
2608 nlh->nlmsg_seq = seq;
2609 if (!__builtin_constant_p(size) || NLMSG_ALIGN(size) - size != 0)
2610 memset(nlmsg_data(nlh) + len, 0, NLMSG_ALIGN(size) - size);
2613 EXPORT_SYMBOL(__nlmsg_put);
2616 * It looks a bit ugly.
2617 * It would be better to create kernel thread.
2620 static int netlink_dump(struct sock *sk)
2622 struct netlink_sock *nlk = nlk_sk(sk);
2623 struct netlink_callback *cb;
2624 struct sk_buff *skb = NULL;
2625 struct nlmsghdr *nlh;
2626 int len, err = -ENOBUFS;
2629 mutex_lock(nlk->cb_mutex);
2630 if (!nlk->cb_running) {
2636 alloc_size = max_t(int, cb->min_dump_alloc, NLMSG_GOODSIZE);
2638 if (!netlink_rx_is_mmaped(sk) &&
2639 atomic_read(&sk->sk_rmem_alloc) >= sk->sk_rcvbuf)
2642 /* NLMSG_GOODSIZE is small to avoid high order allocations being
2643 * required, but it makes sense to _attempt_ a 16K bytes allocation
2644 * to reduce number of system calls on dump operations, if user
2645 * ever provided a big enough buffer.
2647 if (alloc_size < nlk->max_recvmsg_len) {
2648 skb = netlink_alloc_skb(sk,
2649 nlk->max_recvmsg_len,
2654 /* available room should be exact amount to avoid MSG_TRUNC */
2656 skb_reserve(skb, skb_tailroom(skb) -
2657 nlk->max_recvmsg_len);
2660 skb = netlink_alloc_skb(sk, alloc_size, nlk->portid,
2664 netlink_skb_set_owner_r(skb, sk);
2666 len = cb->dump(skb, cb);
2669 mutex_unlock(nlk->cb_mutex);
2671 if (sk_filter(sk, skb))
2674 __netlink_sendskb(sk, skb);
2678 nlh = nlmsg_put_answer(skb, cb, NLMSG_DONE, sizeof(len), NLM_F_MULTI);
2682 nl_dump_check_consistent(cb, nlh);
2684 memcpy(nlmsg_data(nlh), &len, sizeof(len));
2686 if (sk_filter(sk, skb))
2689 __netlink_sendskb(sk, skb);
2694 nlk->cb_running = false;
2695 mutex_unlock(nlk->cb_mutex);
2696 module_put(cb->module);
2697 consume_skb(cb->skb);
2701 mutex_unlock(nlk->cb_mutex);
2706 int __netlink_dump_start(struct sock *ssk, struct sk_buff *skb,
2707 const struct nlmsghdr *nlh,
2708 struct netlink_dump_control *control)
2710 struct netlink_callback *cb;
2712 struct netlink_sock *nlk;
2715 /* Memory mapped dump requests need to be copied to avoid looping
2716 * on the pending state in netlink_mmap_sendmsg() while the CB hold
2717 * a reference to the skb.
2719 if (netlink_skb_is_mmaped(skb)) {
2720 skb = skb_copy(skb, GFP_KERNEL);
2724 atomic_inc(&skb->users);
2726 sk = netlink_lookup(sock_net(ssk), ssk->sk_protocol, NETLINK_CB(skb).portid);
2728 ret = -ECONNREFUSED;
2733 mutex_lock(nlk->cb_mutex);
2734 /* A dump is in progress... */
2735 if (nlk->cb_running) {
2739 /* add reference of module which cb->dump belongs to */
2740 if (!try_module_get(control->module)) {
2741 ret = -EPROTONOSUPPORT;
2746 memset(cb, 0, sizeof(*cb));
2747 cb->dump = control->dump;
2748 cb->done = control->done;
2750 cb->data = control->data;
2751 cb->module = control->module;
2752 cb->min_dump_alloc = control->min_dump_alloc;
2755 nlk->cb_running = true;
2757 mutex_unlock(nlk->cb_mutex);
2759 ret = netlink_dump(sk);
2765 /* We successfully started a dump, by returning -EINTR we
2766 * signal not to send ACK even if it was requested.
2772 mutex_unlock(nlk->cb_mutex);
2777 EXPORT_SYMBOL(__netlink_dump_start);
2779 void netlink_ack(struct sk_buff *in_skb, struct nlmsghdr *nlh, int err)
2781 struct sk_buff *skb;
2782 struct nlmsghdr *rep;
2783 struct nlmsgerr *errmsg;
2784 size_t payload = sizeof(*errmsg);
2786 /* error messages get the original request appened */
2788 payload += nlmsg_len(nlh);
2790 skb = netlink_alloc_skb(in_skb->sk, nlmsg_total_size(payload),
2791 NETLINK_CB(in_skb).portid, GFP_KERNEL);
2795 sk = netlink_lookup(sock_net(in_skb->sk),
2796 in_skb->sk->sk_protocol,
2797 NETLINK_CB(in_skb).portid);
2799 sk->sk_err = ENOBUFS;
2800 sk->sk_error_report(sk);
2806 rep = __nlmsg_put(skb, NETLINK_CB(in_skb).portid, nlh->nlmsg_seq,
2807 NLMSG_ERROR, payload, 0);
2808 errmsg = nlmsg_data(rep);
2809 errmsg->error = err;
2810 memcpy(&errmsg->msg, nlh, err ? nlh->nlmsg_len : sizeof(*nlh));
2811 netlink_unicast(in_skb->sk, skb, NETLINK_CB(in_skb).portid, MSG_DONTWAIT);
2813 EXPORT_SYMBOL(netlink_ack);
2815 int netlink_rcv_skb(struct sk_buff *skb, int (*cb)(struct sk_buff *,
2818 struct nlmsghdr *nlh;
2821 while (skb->len >= nlmsg_total_size(0)) {
2824 nlh = nlmsg_hdr(skb);
2827 if (nlh->nlmsg_len < NLMSG_HDRLEN || skb->len < nlh->nlmsg_len)
2830 /* Only requests are handled by the kernel */
2831 if (!(nlh->nlmsg_flags & NLM_F_REQUEST))
2834 /* Skip control messages */
2835 if (nlh->nlmsg_type < NLMSG_MIN_TYPE)
2843 if (nlh->nlmsg_flags & NLM_F_ACK || err)
2844 netlink_ack(skb, nlh, err);
2847 msglen = NLMSG_ALIGN(nlh->nlmsg_len);
2848 if (msglen > skb->len)
2850 skb_pull(skb, msglen);
2855 EXPORT_SYMBOL(netlink_rcv_skb);
2858 * nlmsg_notify - send a notification netlink message
2859 * @sk: netlink socket to use
2860 * @skb: notification message
2861 * @portid: destination netlink portid for reports or 0
2862 * @group: destination multicast group or 0
2863 * @report: 1 to report back, 0 to disable
2864 * @flags: allocation flags
2866 int nlmsg_notify(struct sock *sk, struct sk_buff *skb, u32 portid,
2867 unsigned int group, int report, gfp_t flags)
2872 int exclude_portid = 0;
2875 atomic_inc(&skb->users);
2876 exclude_portid = portid;
2879 /* errors reported via destination sk->sk_err, but propagate
2880 * delivery errors if NETLINK_BROADCAST_ERROR flag is set */
2881 err = nlmsg_multicast(sk, skb, exclude_portid, group, flags);
2887 err2 = nlmsg_unicast(sk, skb, portid);
2888 if (!err || err == -ESRCH)
2894 EXPORT_SYMBOL(nlmsg_notify);
2896 #ifdef CONFIG_PROC_FS
2897 struct nl_seq_iter {
2898 struct seq_net_private p;
2903 static struct sock *netlink_seq_socket_idx(struct seq_file *seq, loff_t pos)
2905 struct nl_seq_iter *iter = seq->private;
2907 struct netlink_sock *nlk;
2911 for (i = 0; i < MAX_LINKS; i++) {
2912 struct rhashtable *ht = &nl_table[i].hash;
2913 const struct bucket_table *tbl = rht_dereference_rcu(ht->tbl, ht);
2915 for (j = 0; j < tbl->size; j++) {
2916 rht_for_each_entry_rcu(nlk, tbl->buckets[j], node) {
2917 s = (struct sock *)nlk;
2919 if (sock_net(s) != seq_file_net(seq))
2933 static void *netlink_seq_start(struct seq_file *seq, loff_t *pos)
2934 __acquires(nl_table_lock) __acquires(RCU)
2936 read_lock(&nl_table_lock);
2938 return *pos ? netlink_seq_socket_idx(seq, *pos - 1) : SEQ_START_TOKEN;
2941 static void *netlink_seq_next(struct seq_file *seq, void *v, loff_t *pos)
2943 struct rhashtable *ht;
2944 struct netlink_sock *nlk;
2945 struct nl_seq_iter *iter;
2951 if (v == SEQ_START_TOKEN)
2952 return netlink_seq_socket_idx(seq, 0);
2954 net = seq_file_net(seq);
2955 iter = seq->private;
2959 ht = &nl_table[i].hash;
2960 rht_for_each_entry(nlk, nlk->node.next, ht, node)
2961 if (net_eq(sock_net((struct sock *)nlk), net))
2964 j = iter->hash_idx + 1;
2967 const struct bucket_table *tbl = rht_dereference_rcu(ht->tbl, ht);
2969 for (; j < tbl->size; j++) {
2970 rht_for_each_entry(nlk, tbl->buckets[j], ht, node) {
2971 if (net_eq(sock_net((struct sock *)nlk), net)) {
2980 } while (++i < MAX_LINKS);
2985 static void netlink_seq_stop(struct seq_file *seq, void *v)
2986 __releases(RCU) __releases(nl_table_lock)
2989 read_unlock(&nl_table_lock);
2993 static int netlink_seq_show(struct seq_file *seq, void *v)
2995 if (v == SEQ_START_TOKEN) {
2997 "sk Eth Pid Groups "
2998 "Rmem Wmem Dump Locks Drops Inode\n");
3001 struct netlink_sock *nlk = nlk_sk(s);
3003 seq_printf(seq, "%pK %-3d %-6u %08x %-8d %-8d %d %-8d %-8d %-8lu\n",
3007 nlk->groups ? (u32)nlk->groups[0] : 0,
3008 sk_rmem_alloc_get(s),
3009 sk_wmem_alloc_get(s),
3011 atomic_read(&s->sk_refcnt),
3012 atomic_read(&s->sk_drops),
3020 static const struct seq_operations netlink_seq_ops = {
3021 .start = netlink_seq_start,
3022 .next = netlink_seq_next,
3023 .stop = netlink_seq_stop,
3024 .show = netlink_seq_show,
3028 static int netlink_seq_open(struct inode *inode, struct file *file)
3030 return seq_open_net(inode, file, &netlink_seq_ops,
3031 sizeof(struct nl_seq_iter));
3034 static const struct file_operations netlink_seq_fops = {
3035 .owner = THIS_MODULE,
3036 .open = netlink_seq_open,
3038 .llseek = seq_lseek,
3039 .release = seq_release_net,
3044 int netlink_register_notifier(struct notifier_block *nb)
3046 return atomic_notifier_chain_register(&netlink_chain, nb);
3048 EXPORT_SYMBOL(netlink_register_notifier);
3050 int netlink_unregister_notifier(struct notifier_block *nb)
3052 return atomic_notifier_chain_unregister(&netlink_chain, nb);
3054 EXPORT_SYMBOL(netlink_unregister_notifier);
3056 static const struct proto_ops netlink_ops = {
3057 .family = PF_NETLINK,
3058 .owner = THIS_MODULE,
3059 .release = netlink_release,
3060 .bind = netlink_bind,
3061 .connect = netlink_connect,
3062 .socketpair = sock_no_socketpair,
3063 .accept = sock_no_accept,
3064 .getname = netlink_getname,
3065 .poll = netlink_poll,
3066 .ioctl = sock_no_ioctl,
3067 .listen = sock_no_listen,
3068 .shutdown = sock_no_shutdown,
3069 .setsockopt = netlink_setsockopt,
3070 .getsockopt = netlink_getsockopt,
3071 .sendmsg = netlink_sendmsg,
3072 .recvmsg = netlink_recvmsg,
3073 .mmap = netlink_mmap,
3074 .sendpage = sock_no_sendpage,
3077 static const struct net_proto_family netlink_family_ops = {
3078 .family = PF_NETLINK,
3079 .create = netlink_create,
3080 .owner = THIS_MODULE, /* for consistency 8) */
3083 static int __net_init netlink_net_init(struct net *net)
3085 #ifdef CONFIG_PROC_FS
3086 if (!proc_create("netlink", 0, net->proc_net, &netlink_seq_fops))
3092 static void __net_exit netlink_net_exit(struct net *net)
3094 #ifdef CONFIG_PROC_FS
3095 remove_proc_entry("netlink", net->proc_net);
3099 static void __init netlink_add_usersock_entry(void)
3101 struct listeners *listeners;
3104 listeners = kzalloc(sizeof(*listeners) + NLGRPSZ(groups), GFP_KERNEL);
3106 panic("netlink_add_usersock_entry: Cannot allocate listeners\n");
3108 netlink_table_grab();
3110 nl_table[NETLINK_USERSOCK].groups = groups;
3111 rcu_assign_pointer(nl_table[NETLINK_USERSOCK].listeners, listeners);
3112 nl_table[NETLINK_USERSOCK].module = THIS_MODULE;
3113 nl_table[NETLINK_USERSOCK].registered = 1;
3114 nl_table[NETLINK_USERSOCK].flags = NL_CFG_F_NONROOT_SEND;
3116 netlink_table_ungrab();
3119 static struct pernet_operations __net_initdata netlink_net_ops = {
3120 .init = netlink_net_init,
3121 .exit = netlink_net_exit,
3124 static int __init netlink_proto_init(void)
3127 int err = proto_register(&netlink_proto, 0);
3128 struct rhashtable_params ht_params = {
3129 .head_offset = offsetof(struct netlink_sock, node),
3130 .key_offset = offsetof(struct netlink_sock, portid),
3131 .key_len = sizeof(u32), /* portid */
3132 .hashfn = arch_fast_hash,
3133 .max_shift = 16, /* 64K */
3134 .grow_decision = rht_grow_above_75,
3135 .shrink_decision = rht_shrink_below_30,
3136 .mutex_is_held = lockdep_nl_sk_hash_is_held,
3142 BUILD_BUG_ON(sizeof(struct netlink_skb_parms) > FIELD_SIZEOF(struct sk_buff, cb));
3144 nl_table = kcalloc(MAX_LINKS, sizeof(*nl_table), GFP_KERNEL);
3148 for (i = 0; i < MAX_LINKS; i++) {
3149 if (rhashtable_init(&nl_table[i].hash, &ht_params) < 0) {
3151 rhashtable_destroy(&nl_table[i].hash);
3157 INIT_LIST_HEAD(&netlink_tap_all);
3159 netlink_add_usersock_entry();
3161 sock_register(&netlink_family_ops);
3162 register_pernet_subsys(&netlink_net_ops);
3163 /* The netlink device handler may be needed early. */
3168 panic("netlink_init: Cannot allocate nl_table\n");
3171 core_initcall(netlink_proto_init);