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>
64 #include <linux/genetlink.h>
66 #include <net/net_namespace.h>
69 #include <net/netlink.h>
71 #include "af_netlink.h"
75 unsigned long masks[0];
79 #define NETLINK_CONGESTED 0x0
82 #define NETLINK_KERNEL_SOCKET 0x1
83 #define NETLINK_RECV_PKTINFO 0x2
84 #define NETLINK_BROADCAST_SEND_ERROR 0x4
85 #define NETLINK_RECV_NO_ENOBUFS 0x8
87 static inline int netlink_is_kernel(struct sock *sk)
89 return nlk_sk(sk)->flags & NETLINK_KERNEL_SOCKET;
92 struct netlink_table *nl_table;
93 EXPORT_SYMBOL_GPL(nl_table);
95 static DECLARE_WAIT_QUEUE_HEAD(nl_table_wait);
97 static int netlink_dump(struct sock *sk);
98 static void netlink_skb_destructor(struct sk_buff *skb);
100 /* nl_table locking explained:
101 * Lookup and traversal are protected with an RCU read-side lock. Insertion
102 * and removal are protected with per bucket lock while using RCU list
103 * modification primitives and may run in parallel to RCU protected lookups.
104 * Destruction of the Netlink socket may only occur *after* nl_table_lock has
105 * been acquired * either during or after the socket has been removed from
106 * the list and after an RCU grace period.
108 DEFINE_RWLOCK(nl_table_lock);
109 EXPORT_SYMBOL_GPL(nl_table_lock);
110 static atomic_t nl_table_users = ATOMIC_INIT(0);
112 #define nl_deref_protected(X) rcu_dereference_protected(X, lockdep_is_held(&nl_table_lock));
114 static ATOMIC_NOTIFIER_HEAD(netlink_chain);
116 static DEFINE_SPINLOCK(netlink_tap_lock);
117 static struct list_head netlink_tap_all __read_mostly;
119 static const struct rhashtable_params netlink_rhashtable_params;
121 static inline u32 netlink_group_mask(u32 group)
123 return group ? 1 << (group - 1) : 0;
126 int netlink_add_tap(struct netlink_tap *nt)
128 if (unlikely(nt->dev->type != ARPHRD_NETLINK))
131 spin_lock(&netlink_tap_lock);
132 list_add_rcu(&nt->list, &netlink_tap_all);
133 spin_unlock(&netlink_tap_lock);
135 __module_get(nt->module);
139 EXPORT_SYMBOL_GPL(netlink_add_tap);
141 static int __netlink_remove_tap(struct netlink_tap *nt)
144 struct netlink_tap *tmp;
146 spin_lock(&netlink_tap_lock);
148 list_for_each_entry(tmp, &netlink_tap_all, list) {
150 list_del_rcu(&nt->list);
156 pr_warn("__netlink_remove_tap: %p not found\n", nt);
158 spin_unlock(&netlink_tap_lock);
160 if (found && nt->module)
161 module_put(nt->module);
163 return found ? 0 : -ENODEV;
166 int netlink_remove_tap(struct netlink_tap *nt)
170 ret = __netlink_remove_tap(nt);
175 EXPORT_SYMBOL_GPL(netlink_remove_tap);
177 static bool netlink_filter_tap(const struct sk_buff *skb)
179 struct sock *sk = skb->sk;
181 /* We take the more conservative approach and
182 * whitelist socket protocols that may pass.
184 switch (sk->sk_protocol) {
186 case NETLINK_USERSOCK:
187 case NETLINK_SOCK_DIAG:
190 case NETLINK_FIB_LOOKUP:
191 case NETLINK_NETFILTER:
192 case NETLINK_GENERIC:
199 static int __netlink_deliver_tap_skb(struct sk_buff *skb,
200 struct net_device *dev)
202 struct sk_buff *nskb;
203 struct sock *sk = skb->sk;
207 nskb = skb_clone(skb, GFP_ATOMIC);
210 nskb->protocol = htons((u16) sk->sk_protocol);
211 nskb->pkt_type = netlink_is_kernel(sk) ?
212 PACKET_KERNEL : PACKET_USER;
213 skb_reset_network_header(nskb);
214 ret = dev_queue_xmit(nskb);
215 if (unlikely(ret > 0))
216 ret = net_xmit_errno(ret);
223 static void __netlink_deliver_tap(struct sk_buff *skb)
226 struct netlink_tap *tmp;
228 if (!netlink_filter_tap(skb))
231 list_for_each_entry_rcu(tmp, &netlink_tap_all, list) {
232 ret = __netlink_deliver_tap_skb(skb, tmp->dev);
238 static void netlink_deliver_tap(struct sk_buff *skb)
242 if (unlikely(!list_empty(&netlink_tap_all)))
243 __netlink_deliver_tap(skb);
248 static void netlink_deliver_tap_kernel(struct sock *dst, struct sock *src,
251 if (!(netlink_is_kernel(dst) && netlink_is_kernel(src)))
252 netlink_deliver_tap(skb);
255 static void netlink_overrun(struct sock *sk)
257 struct netlink_sock *nlk = nlk_sk(sk);
259 if (!(nlk->flags & NETLINK_RECV_NO_ENOBUFS)) {
260 if (!test_and_set_bit(NETLINK_CONGESTED, &nlk_sk(sk)->state)) {
261 sk->sk_err = ENOBUFS;
262 sk->sk_error_report(sk);
265 atomic_inc(&sk->sk_drops);
268 static void netlink_rcv_wake(struct sock *sk)
270 struct netlink_sock *nlk = nlk_sk(sk);
272 if (skb_queue_empty(&sk->sk_receive_queue))
273 clear_bit(NETLINK_CONGESTED, &nlk->state);
274 if (!test_bit(NETLINK_CONGESTED, &nlk->state))
275 wake_up_interruptible(&nlk->wait);
278 #ifdef CONFIG_NETLINK_MMAP
279 static bool netlink_skb_is_mmaped(const struct sk_buff *skb)
281 return NETLINK_CB(skb).flags & NETLINK_SKB_MMAPED;
284 static bool netlink_rx_is_mmaped(struct sock *sk)
286 return nlk_sk(sk)->rx_ring.pg_vec != NULL;
289 static bool netlink_tx_is_mmaped(struct sock *sk)
291 return nlk_sk(sk)->tx_ring.pg_vec != NULL;
294 static __pure struct page *pgvec_to_page(const void *addr)
296 if (is_vmalloc_addr(addr))
297 return vmalloc_to_page(addr);
299 return virt_to_page(addr);
302 static void free_pg_vec(void **pg_vec, unsigned int order, unsigned int len)
306 for (i = 0; i < len; i++) {
307 if (pg_vec[i] != NULL) {
308 if (is_vmalloc_addr(pg_vec[i]))
311 free_pages((unsigned long)pg_vec[i], order);
317 static void *alloc_one_pg_vec_page(unsigned long order)
320 gfp_t gfp_flags = GFP_KERNEL | __GFP_COMP | __GFP_ZERO |
321 __GFP_NOWARN | __GFP_NORETRY;
323 buffer = (void *)__get_free_pages(gfp_flags, order);
327 buffer = vzalloc((1 << order) * PAGE_SIZE);
331 gfp_flags &= ~__GFP_NORETRY;
332 return (void *)__get_free_pages(gfp_flags, order);
335 static void **alloc_pg_vec(struct netlink_sock *nlk,
336 struct nl_mmap_req *req, unsigned int order)
338 unsigned int block_nr = req->nm_block_nr;
342 pg_vec = kcalloc(block_nr, sizeof(void *), GFP_KERNEL);
346 for (i = 0; i < block_nr; i++) {
347 pg_vec[i] = alloc_one_pg_vec_page(order);
348 if (pg_vec[i] == NULL)
354 free_pg_vec(pg_vec, order, block_nr);
358 static int netlink_set_ring(struct sock *sk, struct nl_mmap_req *req,
359 bool closing, bool tx_ring)
361 struct netlink_sock *nlk = nlk_sk(sk);
362 struct netlink_ring *ring;
363 struct sk_buff_head *queue;
364 void **pg_vec = NULL;
365 unsigned int order = 0;
368 ring = tx_ring ? &nlk->tx_ring : &nlk->rx_ring;
369 queue = tx_ring ? &sk->sk_write_queue : &sk->sk_receive_queue;
372 if (atomic_read(&nlk->mapped))
374 if (atomic_read(&ring->pending))
378 if (req->nm_block_nr) {
379 if (ring->pg_vec != NULL)
382 if ((int)req->nm_block_size <= 0)
384 if (!PAGE_ALIGNED(req->nm_block_size))
386 if (req->nm_frame_size < NL_MMAP_HDRLEN)
388 if (!IS_ALIGNED(req->nm_frame_size, NL_MMAP_MSG_ALIGNMENT))
391 ring->frames_per_block = req->nm_block_size /
393 if (ring->frames_per_block == 0)
395 if (ring->frames_per_block * req->nm_block_nr !=
399 order = get_order(req->nm_block_size);
400 pg_vec = alloc_pg_vec(nlk, req, order);
404 if (req->nm_frame_nr)
409 mutex_lock(&nlk->pg_vec_lock);
410 if (closing || atomic_read(&nlk->mapped) == 0) {
412 spin_lock_bh(&queue->lock);
414 ring->frame_max = req->nm_frame_nr - 1;
416 ring->frame_size = req->nm_frame_size;
417 ring->pg_vec_pages = req->nm_block_size / PAGE_SIZE;
419 swap(ring->pg_vec_len, req->nm_block_nr);
420 swap(ring->pg_vec_order, order);
421 swap(ring->pg_vec, pg_vec);
423 __skb_queue_purge(queue);
424 spin_unlock_bh(&queue->lock);
426 WARN_ON(atomic_read(&nlk->mapped));
428 mutex_unlock(&nlk->pg_vec_lock);
431 free_pg_vec(pg_vec, order, req->nm_block_nr);
435 static void netlink_mm_open(struct vm_area_struct *vma)
437 struct file *file = vma->vm_file;
438 struct socket *sock = file->private_data;
439 struct sock *sk = sock->sk;
442 atomic_inc(&nlk_sk(sk)->mapped);
445 static void netlink_mm_close(struct vm_area_struct *vma)
447 struct file *file = vma->vm_file;
448 struct socket *sock = file->private_data;
449 struct sock *sk = sock->sk;
452 atomic_dec(&nlk_sk(sk)->mapped);
455 static const struct vm_operations_struct netlink_mmap_ops = {
456 .open = netlink_mm_open,
457 .close = netlink_mm_close,
460 static int netlink_mmap(struct file *file, struct socket *sock,
461 struct vm_area_struct *vma)
463 struct sock *sk = sock->sk;
464 struct netlink_sock *nlk = nlk_sk(sk);
465 struct netlink_ring *ring;
466 unsigned long start, size, expected;
473 mutex_lock(&nlk->pg_vec_lock);
476 for (ring = &nlk->rx_ring; ring <= &nlk->tx_ring; ring++) {
477 if (ring->pg_vec == NULL)
479 expected += ring->pg_vec_len * ring->pg_vec_pages * PAGE_SIZE;
485 size = vma->vm_end - vma->vm_start;
486 if (size != expected)
489 start = vma->vm_start;
490 for (ring = &nlk->rx_ring; ring <= &nlk->tx_ring; ring++) {
491 if (ring->pg_vec == NULL)
494 for (i = 0; i < ring->pg_vec_len; i++) {
496 void *kaddr = ring->pg_vec[i];
499 for (pg_num = 0; pg_num < ring->pg_vec_pages; pg_num++) {
500 page = pgvec_to_page(kaddr);
501 err = vm_insert_page(vma, start, page);
510 atomic_inc(&nlk->mapped);
511 vma->vm_ops = &netlink_mmap_ops;
514 mutex_unlock(&nlk->pg_vec_lock);
518 static void netlink_frame_flush_dcache(const struct nl_mmap_hdr *hdr, unsigned int nm_len)
520 #if ARCH_IMPLEMENTS_FLUSH_DCACHE_PAGE == 1
521 struct page *p_start, *p_end;
523 /* First page is flushed through netlink_{get,set}_status */
524 p_start = pgvec_to_page(hdr + PAGE_SIZE);
525 p_end = pgvec_to_page((void *)hdr + NL_MMAP_HDRLEN + nm_len - 1);
526 while (p_start <= p_end) {
527 flush_dcache_page(p_start);
533 static enum nl_mmap_status netlink_get_status(const struct nl_mmap_hdr *hdr)
536 flush_dcache_page(pgvec_to_page(hdr));
537 return hdr->nm_status;
540 static void netlink_set_status(struct nl_mmap_hdr *hdr,
541 enum nl_mmap_status status)
544 hdr->nm_status = status;
545 flush_dcache_page(pgvec_to_page(hdr));
548 static struct nl_mmap_hdr *
549 __netlink_lookup_frame(const struct netlink_ring *ring, unsigned int pos)
551 unsigned int pg_vec_pos, frame_off;
553 pg_vec_pos = pos / ring->frames_per_block;
554 frame_off = pos % ring->frames_per_block;
556 return ring->pg_vec[pg_vec_pos] + (frame_off * ring->frame_size);
559 static struct nl_mmap_hdr *
560 netlink_lookup_frame(const struct netlink_ring *ring, unsigned int pos,
561 enum nl_mmap_status status)
563 struct nl_mmap_hdr *hdr;
565 hdr = __netlink_lookup_frame(ring, pos);
566 if (netlink_get_status(hdr) != status)
572 static struct nl_mmap_hdr *
573 netlink_current_frame(const struct netlink_ring *ring,
574 enum nl_mmap_status status)
576 return netlink_lookup_frame(ring, ring->head, status);
579 static struct nl_mmap_hdr *
580 netlink_previous_frame(const struct netlink_ring *ring,
581 enum nl_mmap_status status)
585 prev = ring->head ? ring->head - 1 : ring->frame_max;
586 return netlink_lookup_frame(ring, prev, status);
589 static void netlink_increment_head(struct netlink_ring *ring)
591 ring->head = ring->head != ring->frame_max ? ring->head + 1 : 0;
594 static void netlink_forward_ring(struct netlink_ring *ring)
596 unsigned int head = ring->head, pos = head;
597 const struct nl_mmap_hdr *hdr;
600 hdr = __netlink_lookup_frame(ring, pos);
601 if (hdr->nm_status == NL_MMAP_STATUS_UNUSED)
603 if (hdr->nm_status != NL_MMAP_STATUS_SKIP)
605 netlink_increment_head(ring);
606 } while (ring->head != head);
609 static bool netlink_dump_space(struct netlink_sock *nlk)
611 struct netlink_ring *ring = &nlk->rx_ring;
612 struct nl_mmap_hdr *hdr;
615 hdr = netlink_current_frame(ring, NL_MMAP_STATUS_UNUSED);
619 n = ring->head + ring->frame_max / 2;
620 if (n > ring->frame_max)
621 n -= ring->frame_max;
623 hdr = __netlink_lookup_frame(ring, n);
625 return hdr->nm_status == NL_MMAP_STATUS_UNUSED;
628 static unsigned int netlink_poll(struct file *file, struct socket *sock,
631 struct sock *sk = sock->sk;
632 struct netlink_sock *nlk = nlk_sk(sk);
636 if (nlk->rx_ring.pg_vec != NULL) {
637 /* Memory mapped sockets don't call recvmsg(), so flow control
638 * for dumps is performed here. A dump is allowed to continue
639 * if at least half the ring is unused.
641 while (nlk->cb_running && netlink_dump_space(nlk)) {
642 err = netlink_dump(sk);
645 sk->sk_error_report(sk);
649 netlink_rcv_wake(sk);
652 mask = datagram_poll(file, sock, wait);
654 spin_lock_bh(&sk->sk_receive_queue.lock);
655 if (nlk->rx_ring.pg_vec) {
656 netlink_forward_ring(&nlk->rx_ring);
657 if (!netlink_previous_frame(&nlk->rx_ring, NL_MMAP_STATUS_UNUSED))
658 mask |= POLLIN | POLLRDNORM;
660 spin_unlock_bh(&sk->sk_receive_queue.lock);
662 spin_lock_bh(&sk->sk_write_queue.lock);
663 if (nlk->tx_ring.pg_vec) {
664 if (netlink_current_frame(&nlk->tx_ring, NL_MMAP_STATUS_UNUSED))
665 mask |= POLLOUT | POLLWRNORM;
667 spin_unlock_bh(&sk->sk_write_queue.lock);
672 static struct nl_mmap_hdr *netlink_mmap_hdr(struct sk_buff *skb)
674 return (struct nl_mmap_hdr *)(skb->head - NL_MMAP_HDRLEN);
677 static void netlink_ring_setup_skb(struct sk_buff *skb, struct sock *sk,
678 struct netlink_ring *ring,
679 struct nl_mmap_hdr *hdr)
684 size = ring->frame_size - NL_MMAP_HDRLEN;
685 data = (void *)hdr + NL_MMAP_HDRLEN;
689 skb_reset_tail_pointer(skb);
690 skb->end = skb->tail + size;
693 skb->destructor = netlink_skb_destructor;
694 NETLINK_CB(skb).flags |= NETLINK_SKB_MMAPED;
695 NETLINK_CB(skb).sk = sk;
698 static int netlink_mmap_sendmsg(struct sock *sk, struct msghdr *msg,
699 u32 dst_portid, u32 dst_group,
700 struct scm_cookie *scm)
702 struct netlink_sock *nlk = nlk_sk(sk);
703 struct netlink_ring *ring;
704 struct nl_mmap_hdr *hdr;
707 int err = 0, len = 0;
709 mutex_lock(&nlk->pg_vec_lock);
711 ring = &nlk->tx_ring;
712 maxlen = ring->frame_size - NL_MMAP_HDRLEN;
717 hdr = netlink_current_frame(ring, NL_MMAP_STATUS_VALID);
719 if (!(msg->msg_flags & MSG_DONTWAIT) &&
720 atomic_read(&nlk->tx_ring.pending))
725 nm_len = ACCESS_ONCE(hdr->nm_len);
726 if (nm_len > maxlen) {
731 netlink_frame_flush_dcache(hdr, nm_len);
733 skb = alloc_skb(nm_len, GFP_KERNEL);
738 __skb_put(skb, nm_len);
739 memcpy(skb->data, (void *)hdr + NL_MMAP_HDRLEN, nm_len);
740 netlink_set_status(hdr, NL_MMAP_STATUS_UNUSED);
742 netlink_increment_head(ring);
744 NETLINK_CB(skb).portid = nlk->portid;
745 NETLINK_CB(skb).dst_group = dst_group;
746 NETLINK_CB(skb).creds = scm->creds;
748 err = security_netlink_send(sk, skb);
754 if (unlikely(dst_group)) {
755 atomic_inc(&skb->users);
756 netlink_broadcast(sk, skb, dst_portid, dst_group,
759 err = netlink_unicast(sk, skb, dst_portid,
760 msg->msg_flags & MSG_DONTWAIT);
765 } while (hdr != NULL ||
766 (!(msg->msg_flags & MSG_DONTWAIT) &&
767 atomic_read(&nlk->tx_ring.pending)));
772 mutex_unlock(&nlk->pg_vec_lock);
776 static void netlink_queue_mmaped_skb(struct sock *sk, struct sk_buff *skb)
778 struct nl_mmap_hdr *hdr;
780 hdr = netlink_mmap_hdr(skb);
781 hdr->nm_len = skb->len;
782 hdr->nm_group = NETLINK_CB(skb).dst_group;
783 hdr->nm_pid = NETLINK_CB(skb).creds.pid;
784 hdr->nm_uid = from_kuid(sk_user_ns(sk), NETLINK_CB(skb).creds.uid);
785 hdr->nm_gid = from_kgid(sk_user_ns(sk), NETLINK_CB(skb).creds.gid);
786 netlink_frame_flush_dcache(hdr, hdr->nm_len);
787 netlink_set_status(hdr, NL_MMAP_STATUS_VALID);
789 NETLINK_CB(skb).flags |= NETLINK_SKB_DELIVERED;
793 static void netlink_ring_set_copied(struct sock *sk, struct sk_buff *skb)
795 struct netlink_sock *nlk = nlk_sk(sk);
796 struct netlink_ring *ring = &nlk->rx_ring;
797 struct nl_mmap_hdr *hdr;
799 spin_lock_bh(&sk->sk_receive_queue.lock);
800 hdr = netlink_current_frame(ring, NL_MMAP_STATUS_UNUSED);
802 spin_unlock_bh(&sk->sk_receive_queue.lock);
807 netlink_increment_head(ring);
808 __skb_queue_tail(&sk->sk_receive_queue, skb);
809 spin_unlock_bh(&sk->sk_receive_queue.lock);
811 hdr->nm_len = skb->len;
812 hdr->nm_group = NETLINK_CB(skb).dst_group;
813 hdr->nm_pid = NETLINK_CB(skb).creds.pid;
814 hdr->nm_uid = from_kuid(sk_user_ns(sk), NETLINK_CB(skb).creds.uid);
815 hdr->nm_gid = from_kgid(sk_user_ns(sk), NETLINK_CB(skb).creds.gid);
816 netlink_set_status(hdr, NL_MMAP_STATUS_COPY);
819 #else /* CONFIG_NETLINK_MMAP */
820 #define netlink_skb_is_mmaped(skb) false
821 #define netlink_rx_is_mmaped(sk) false
822 #define netlink_tx_is_mmaped(sk) false
823 #define netlink_mmap sock_no_mmap
824 #define netlink_poll datagram_poll
825 #define netlink_mmap_sendmsg(sk, msg, dst_portid, dst_group, scm) 0
826 #endif /* CONFIG_NETLINK_MMAP */
828 static void netlink_skb_destructor(struct sk_buff *skb)
830 #ifdef CONFIG_NETLINK_MMAP
831 struct nl_mmap_hdr *hdr;
832 struct netlink_ring *ring;
835 /* If a packet from the kernel to userspace was freed because of an
836 * error without being delivered to userspace, the kernel must reset
837 * the status. In the direction userspace to kernel, the status is
838 * always reset here after the packet was processed and freed.
840 if (netlink_skb_is_mmaped(skb)) {
841 hdr = netlink_mmap_hdr(skb);
842 sk = NETLINK_CB(skb).sk;
844 if (NETLINK_CB(skb).flags & NETLINK_SKB_TX) {
845 netlink_set_status(hdr, NL_MMAP_STATUS_UNUSED);
846 ring = &nlk_sk(sk)->tx_ring;
848 if (!(NETLINK_CB(skb).flags & NETLINK_SKB_DELIVERED)) {
850 netlink_set_status(hdr, NL_MMAP_STATUS_VALID);
852 ring = &nlk_sk(sk)->rx_ring;
855 WARN_ON(atomic_read(&ring->pending) == 0);
856 atomic_dec(&ring->pending);
862 if (is_vmalloc_addr(skb->head)) {
864 !atomic_dec_return(&(skb_shinfo(skb)->dataref)))
873 static void netlink_skb_set_owner_r(struct sk_buff *skb, struct sock *sk)
875 WARN_ON(skb->sk != NULL);
877 skb->destructor = netlink_skb_destructor;
878 atomic_add(skb->truesize, &sk->sk_rmem_alloc);
879 sk_mem_charge(sk, skb->truesize);
882 static void netlink_sock_destruct(struct sock *sk)
884 struct netlink_sock *nlk = nlk_sk(sk);
886 if (nlk->cb_running) {
888 nlk->cb.done(&nlk->cb);
890 module_put(nlk->cb.module);
891 kfree_skb(nlk->cb.skb);
894 skb_queue_purge(&sk->sk_receive_queue);
895 #ifdef CONFIG_NETLINK_MMAP
897 struct nl_mmap_req req;
899 memset(&req, 0, sizeof(req));
900 if (nlk->rx_ring.pg_vec)
901 netlink_set_ring(sk, &req, true, false);
902 memset(&req, 0, sizeof(req));
903 if (nlk->tx_ring.pg_vec)
904 netlink_set_ring(sk, &req, true, true);
906 #endif /* CONFIG_NETLINK_MMAP */
908 if (!sock_flag(sk, SOCK_DEAD)) {
909 printk(KERN_ERR "Freeing alive netlink socket %p\n", sk);
913 WARN_ON(atomic_read(&sk->sk_rmem_alloc));
914 WARN_ON(atomic_read(&sk->sk_wmem_alloc));
915 WARN_ON(nlk_sk(sk)->groups);
918 /* This lock without WQ_FLAG_EXCLUSIVE is good on UP and it is _very_ bad on
919 * SMP. Look, when several writers sleep and reader wakes them up, all but one
920 * immediately hit write lock and grab all the cpus. Exclusive sleep solves
921 * this, _but_ remember, it adds useless work on UP machines.
924 void netlink_table_grab(void)
925 __acquires(nl_table_lock)
929 write_lock_irq(&nl_table_lock);
931 if (atomic_read(&nl_table_users)) {
932 DECLARE_WAITQUEUE(wait, current);
934 add_wait_queue_exclusive(&nl_table_wait, &wait);
936 set_current_state(TASK_UNINTERRUPTIBLE);
937 if (atomic_read(&nl_table_users) == 0)
939 write_unlock_irq(&nl_table_lock);
941 write_lock_irq(&nl_table_lock);
944 __set_current_state(TASK_RUNNING);
945 remove_wait_queue(&nl_table_wait, &wait);
949 void netlink_table_ungrab(void)
950 __releases(nl_table_lock)
952 write_unlock_irq(&nl_table_lock);
953 wake_up(&nl_table_wait);
957 netlink_lock_table(void)
959 /* read_lock() synchronizes us to netlink_table_grab */
961 read_lock(&nl_table_lock);
962 atomic_inc(&nl_table_users);
963 read_unlock(&nl_table_lock);
967 netlink_unlock_table(void)
969 if (atomic_dec_and_test(&nl_table_users))
970 wake_up(&nl_table_wait);
973 struct netlink_compare_arg
979 /* Doing sizeof directly may yield 4 extra bytes on 64-bit. */
980 #define netlink_compare_arg_len \
981 (offsetof(struct netlink_compare_arg, portid) + sizeof(u32))
983 static inline int netlink_compare(struct rhashtable_compare_arg *arg,
986 const struct netlink_compare_arg *x = arg->key;
987 const struct netlink_sock *nlk = ptr;
989 return nlk->portid != x->portid ||
990 !net_eq(sock_net(&nlk->sk), read_pnet(&x->pnet));
993 static void netlink_compare_arg_init(struct netlink_compare_arg *arg,
994 struct net *net, u32 portid)
996 memset(arg, 0, sizeof(*arg));
997 write_pnet(&arg->pnet, net);
998 arg->portid = portid;
1001 static struct sock *__netlink_lookup(struct netlink_table *table, u32 portid,
1004 struct netlink_compare_arg arg;
1006 netlink_compare_arg_init(&arg, net, portid);
1007 return rhashtable_lookup_fast(&table->hash, &arg,
1008 netlink_rhashtable_params);
1011 static int __netlink_insert(struct netlink_table *table, struct sock *sk)
1013 struct netlink_compare_arg arg;
1015 netlink_compare_arg_init(&arg, sock_net(sk), nlk_sk(sk)->portid);
1016 return rhashtable_lookup_insert_key(&table->hash, &arg,
1018 netlink_rhashtable_params);
1021 static struct sock *netlink_lookup(struct net *net, int protocol, u32 portid)
1023 struct netlink_table *table = &nl_table[protocol];
1027 sk = __netlink_lookup(table, portid, net);
1035 static const struct proto_ops netlink_ops;
1038 netlink_update_listeners(struct sock *sk)
1040 struct netlink_table *tbl = &nl_table[sk->sk_protocol];
1043 struct listeners *listeners;
1045 listeners = nl_deref_protected(tbl->listeners);
1049 for (i = 0; i < NLGRPLONGS(tbl->groups); i++) {
1051 sk_for_each_bound(sk, &tbl->mc_list) {
1052 if (i < NLGRPLONGS(nlk_sk(sk)->ngroups))
1053 mask |= nlk_sk(sk)->groups[i];
1055 listeners->masks[i] = mask;
1057 /* this function is only called with the netlink table "grabbed", which
1058 * makes sure updates are visible before bind or setsockopt return. */
1061 static int netlink_insert(struct sock *sk, u32 portid)
1063 struct netlink_table *table = &nl_table[sk->sk_protocol];
1069 if (nlk_sk(sk)->portid)
1073 if (BITS_PER_LONG > 32 &&
1074 unlikely(atomic_read(&table->hash.nelems) >= UINT_MAX))
1077 nlk_sk(sk)->portid = portid;
1080 err = __netlink_insert(table, sk);
1092 static void netlink_remove(struct sock *sk)
1094 struct netlink_table *table;
1096 table = &nl_table[sk->sk_protocol];
1097 if (!rhashtable_remove_fast(&table->hash, &nlk_sk(sk)->node,
1098 netlink_rhashtable_params)) {
1099 WARN_ON(atomic_read(&sk->sk_refcnt) == 1);
1103 netlink_table_grab();
1104 if (nlk_sk(sk)->subscriptions) {
1105 __sk_del_bind_node(sk);
1106 netlink_update_listeners(sk);
1108 if (sk->sk_protocol == NETLINK_GENERIC)
1109 atomic_inc(&genl_sk_destructing_cnt);
1110 netlink_table_ungrab();
1113 static struct proto netlink_proto = {
1115 .owner = THIS_MODULE,
1116 .obj_size = sizeof(struct netlink_sock),
1119 static int __netlink_create(struct net *net, struct socket *sock,
1120 struct mutex *cb_mutex, int protocol)
1123 struct netlink_sock *nlk;
1125 sock->ops = &netlink_ops;
1127 sk = sk_alloc(net, PF_NETLINK, GFP_KERNEL, &netlink_proto);
1131 sock_init_data(sock, sk);
1135 nlk->cb_mutex = cb_mutex;
1137 nlk->cb_mutex = &nlk->cb_def_mutex;
1138 mutex_init(nlk->cb_mutex);
1140 init_waitqueue_head(&nlk->wait);
1141 #ifdef CONFIG_NETLINK_MMAP
1142 mutex_init(&nlk->pg_vec_lock);
1145 sk->sk_destruct = netlink_sock_destruct;
1146 sk->sk_protocol = protocol;
1150 static int netlink_create(struct net *net, struct socket *sock, int protocol,
1153 struct module *module = NULL;
1154 struct mutex *cb_mutex;
1155 struct netlink_sock *nlk;
1156 int (*bind)(struct net *net, int group);
1157 void (*unbind)(struct net *net, int group);
1160 sock->state = SS_UNCONNECTED;
1162 if (sock->type != SOCK_RAW && sock->type != SOCK_DGRAM)
1163 return -ESOCKTNOSUPPORT;
1165 if (protocol < 0 || protocol >= MAX_LINKS)
1166 return -EPROTONOSUPPORT;
1168 netlink_lock_table();
1169 #ifdef CONFIG_MODULES
1170 if (!nl_table[protocol].registered) {
1171 netlink_unlock_table();
1172 request_module("net-pf-%d-proto-%d", PF_NETLINK, protocol);
1173 netlink_lock_table();
1176 if (nl_table[protocol].registered &&
1177 try_module_get(nl_table[protocol].module))
1178 module = nl_table[protocol].module;
1180 err = -EPROTONOSUPPORT;
1181 cb_mutex = nl_table[protocol].cb_mutex;
1182 bind = nl_table[protocol].bind;
1183 unbind = nl_table[protocol].unbind;
1184 netlink_unlock_table();
1189 err = __netlink_create(net, sock, cb_mutex, protocol);
1194 sock_prot_inuse_add(net, &netlink_proto, 1);
1197 nlk = nlk_sk(sock->sk);
1198 nlk->module = module;
1199 nlk->netlink_bind = bind;
1200 nlk->netlink_unbind = unbind;
1209 static void deferred_put_nlk_sk(struct rcu_head *head)
1211 struct netlink_sock *nlk = container_of(head, struct netlink_sock, rcu);
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
1233 /* must not acquire netlink_table_lock in any way again before unbind
1234 * and notifying genetlink is done as otherwise it might deadlock
1236 if (nlk->netlink_unbind) {
1239 for (i = 0; i < nlk->ngroups; i++)
1240 if (test_bit(i, nlk->groups))
1241 nlk->netlink_unbind(sock_net(sk), i + 1);
1243 if (sk->sk_protocol == NETLINK_GENERIC &&
1244 atomic_dec_return(&genl_sk_destructing_cnt) == 0)
1245 wake_up(&genl_sk_destructing_waitq);
1248 wake_up_interruptible_all(&nlk->wait);
1250 skb_queue_purge(&sk->sk_write_queue);
1253 struct netlink_notify n = {
1254 .net = sock_net(sk),
1255 .protocol = sk->sk_protocol,
1256 .portid = nlk->portid,
1258 atomic_notifier_call_chain(&netlink_chain,
1259 NETLINK_URELEASE, &n);
1262 module_put(nlk->module);
1264 if (netlink_is_kernel(sk)) {
1265 netlink_table_grab();
1266 BUG_ON(nl_table[sk->sk_protocol].registered == 0);
1267 if (--nl_table[sk->sk_protocol].registered == 0) {
1268 struct listeners *old;
1270 old = nl_deref_protected(nl_table[sk->sk_protocol].listeners);
1271 RCU_INIT_POINTER(nl_table[sk->sk_protocol].listeners, NULL);
1272 kfree_rcu(old, rcu);
1273 nl_table[sk->sk_protocol].module = NULL;
1274 nl_table[sk->sk_protocol].bind = NULL;
1275 nl_table[sk->sk_protocol].unbind = NULL;
1276 nl_table[sk->sk_protocol].flags = 0;
1277 nl_table[sk->sk_protocol].registered = 0;
1279 netlink_table_ungrab();
1286 sock_prot_inuse_add(sock_net(sk), &netlink_proto, -1);
1288 call_rcu(&nlk->rcu, deferred_put_nlk_sk);
1292 static int netlink_autobind(struct socket *sock)
1294 struct sock *sk = sock->sk;
1295 struct net *net = sock_net(sk);
1296 struct netlink_table *table = &nl_table[sk->sk_protocol];
1297 s32 portid = task_tgid_vnr(current);
1299 static s32 rover = -4097;
1304 if (__netlink_lookup(table, portid, net)) {
1305 /* Bind collision, search negative portid values. */
1314 err = netlink_insert(sk, portid);
1315 if (err == -EADDRINUSE)
1318 /* If 2 threads race to autobind, that is fine. */
1326 * __netlink_ns_capable - General netlink message capability test
1327 * @nsp: NETLINK_CB of the socket buffer holding a netlink command from userspace.
1328 * @user_ns: The user namespace of the capability to use
1329 * @cap: The capability to use
1331 * Test to see if the opener of the socket we received the message
1332 * from had when the netlink socket was created and the sender of the
1333 * message has has the capability @cap in the user namespace @user_ns.
1335 bool __netlink_ns_capable(const struct netlink_skb_parms *nsp,
1336 struct user_namespace *user_ns, int cap)
1338 return ((nsp->flags & NETLINK_SKB_DST) ||
1339 file_ns_capable(nsp->sk->sk_socket->file, user_ns, cap)) &&
1340 ns_capable(user_ns, cap);
1342 EXPORT_SYMBOL(__netlink_ns_capable);
1345 * netlink_ns_capable - General netlink message capability test
1346 * @skb: socket buffer holding a netlink command from userspace
1347 * @user_ns: The user namespace of the capability to use
1348 * @cap: The capability to use
1350 * Test to see if the opener of the socket we received the message
1351 * from had when the netlink socket was created and the sender of the
1352 * message has has the capability @cap in the user namespace @user_ns.
1354 bool netlink_ns_capable(const struct sk_buff *skb,
1355 struct user_namespace *user_ns, int cap)
1357 return __netlink_ns_capable(&NETLINK_CB(skb), user_ns, cap);
1359 EXPORT_SYMBOL(netlink_ns_capable);
1362 * netlink_capable - Netlink global message capability test
1363 * @skb: socket buffer holding a netlink command from userspace
1364 * @cap: The capability to use
1366 * Test to see if the opener of the socket we received the message
1367 * from had when the netlink socket was created and the sender of the
1368 * message has has the capability @cap in all user namespaces.
1370 bool netlink_capable(const struct sk_buff *skb, int cap)
1372 return netlink_ns_capable(skb, &init_user_ns, cap);
1374 EXPORT_SYMBOL(netlink_capable);
1377 * netlink_net_capable - Netlink network namespace message capability test
1378 * @skb: socket buffer holding a netlink command from userspace
1379 * @cap: The capability to use
1381 * Test to see if the opener of the socket we received the message
1382 * from had when the netlink socket was created and the sender of the
1383 * message has has the capability @cap over the network namespace of
1384 * the socket we received the message from.
1386 bool netlink_net_capable(const struct sk_buff *skb, int cap)
1388 return netlink_ns_capable(skb, sock_net(skb->sk)->user_ns, cap);
1390 EXPORT_SYMBOL(netlink_net_capable);
1392 static inline int netlink_allowed(const struct socket *sock, unsigned int flag)
1394 return (nl_table[sock->sk->sk_protocol].flags & flag) ||
1395 ns_capable(sock_net(sock->sk)->user_ns, CAP_NET_ADMIN);
1399 netlink_update_subscriptions(struct sock *sk, unsigned int subscriptions)
1401 struct netlink_sock *nlk = nlk_sk(sk);
1403 if (nlk->subscriptions && !subscriptions)
1404 __sk_del_bind_node(sk);
1405 else if (!nlk->subscriptions && subscriptions)
1406 sk_add_bind_node(sk, &nl_table[sk->sk_protocol].mc_list);
1407 nlk->subscriptions = subscriptions;
1410 static int netlink_realloc_groups(struct sock *sk)
1412 struct netlink_sock *nlk = nlk_sk(sk);
1413 unsigned int groups;
1414 unsigned long *new_groups;
1417 netlink_table_grab();
1419 groups = nl_table[sk->sk_protocol].groups;
1420 if (!nl_table[sk->sk_protocol].registered) {
1425 if (nlk->ngroups >= groups)
1428 new_groups = krealloc(nlk->groups, NLGRPSZ(groups), GFP_ATOMIC);
1429 if (new_groups == NULL) {
1433 memset((char *)new_groups + NLGRPSZ(nlk->ngroups), 0,
1434 NLGRPSZ(groups) - NLGRPSZ(nlk->ngroups));
1436 nlk->groups = new_groups;
1437 nlk->ngroups = groups;
1439 netlink_table_ungrab();
1443 static void netlink_undo_bind(int group, long unsigned int groups,
1446 struct netlink_sock *nlk = nlk_sk(sk);
1449 if (!nlk->netlink_unbind)
1452 for (undo = 0; undo < group; undo++)
1453 if (test_bit(undo, &groups))
1454 nlk->netlink_unbind(sock_net(sk), undo + 1);
1457 static int netlink_bind(struct socket *sock, struct sockaddr *addr,
1460 struct sock *sk = sock->sk;
1461 struct net *net = sock_net(sk);
1462 struct netlink_sock *nlk = nlk_sk(sk);
1463 struct sockaddr_nl *nladdr = (struct sockaddr_nl *)addr;
1465 long unsigned int groups = nladdr->nl_groups;
1467 if (addr_len < sizeof(struct sockaddr_nl))
1470 if (nladdr->nl_family != AF_NETLINK)
1473 /* Only superuser is allowed to listen multicasts */
1475 if (!netlink_allowed(sock, NL_CFG_F_NONROOT_RECV))
1477 err = netlink_realloc_groups(sk);
1483 if (nladdr->nl_pid != nlk->portid)
1486 if (nlk->netlink_bind && groups) {
1489 for (group = 0; group < nlk->ngroups; group++) {
1490 if (!test_bit(group, &groups))
1492 err = nlk->netlink_bind(net, group + 1);
1495 netlink_undo_bind(group, groups, sk);
1501 err = nladdr->nl_pid ?
1502 netlink_insert(sk, nladdr->nl_pid) :
1503 netlink_autobind(sock);
1505 netlink_undo_bind(nlk->ngroups, groups, sk);
1510 if (!groups && (nlk->groups == NULL || !(u32)nlk->groups[0]))
1513 netlink_table_grab();
1514 netlink_update_subscriptions(sk, nlk->subscriptions +
1516 hweight32(nlk->groups[0]));
1517 nlk->groups[0] = (nlk->groups[0] & ~0xffffffffUL) | groups;
1518 netlink_update_listeners(sk);
1519 netlink_table_ungrab();
1524 static int netlink_connect(struct socket *sock, struct sockaddr *addr,
1525 int alen, int flags)
1528 struct sock *sk = sock->sk;
1529 struct netlink_sock *nlk = nlk_sk(sk);
1530 struct sockaddr_nl *nladdr = (struct sockaddr_nl *)addr;
1532 if (alen < sizeof(addr->sa_family))
1535 if (addr->sa_family == AF_UNSPEC) {
1536 sk->sk_state = NETLINK_UNCONNECTED;
1537 nlk->dst_portid = 0;
1541 if (addr->sa_family != AF_NETLINK)
1544 if ((nladdr->nl_groups || nladdr->nl_pid) &&
1545 !netlink_allowed(sock, NL_CFG_F_NONROOT_SEND))
1549 err = netlink_autobind(sock);
1552 sk->sk_state = NETLINK_CONNECTED;
1553 nlk->dst_portid = nladdr->nl_pid;
1554 nlk->dst_group = ffs(nladdr->nl_groups);
1560 static int netlink_getname(struct socket *sock, struct sockaddr *addr,
1561 int *addr_len, int peer)
1563 struct sock *sk = sock->sk;
1564 struct netlink_sock *nlk = nlk_sk(sk);
1565 DECLARE_SOCKADDR(struct sockaddr_nl *, nladdr, addr);
1567 nladdr->nl_family = AF_NETLINK;
1569 *addr_len = sizeof(*nladdr);
1572 nladdr->nl_pid = nlk->dst_portid;
1573 nladdr->nl_groups = netlink_group_mask(nlk->dst_group);
1575 nladdr->nl_pid = nlk->portid;
1576 nladdr->nl_groups = nlk->groups ? nlk->groups[0] : 0;
1581 static struct sock *netlink_getsockbyportid(struct sock *ssk, u32 portid)
1584 struct netlink_sock *nlk;
1586 sock = netlink_lookup(sock_net(ssk), ssk->sk_protocol, portid);
1588 return ERR_PTR(-ECONNREFUSED);
1590 /* Don't bother queuing skb if kernel socket has no input function */
1592 if (sock->sk_state == NETLINK_CONNECTED &&
1593 nlk->dst_portid != nlk_sk(ssk)->portid) {
1595 return ERR_PTR(-ECONNREFUSED);
1600 struct sock *netlink_getsockbyfilp(struct file *filp)
1602 struct inode *inode = file_inode(filp);
1605 if (!S_ISSOCK(inode->i_mode))
1606 return ERR_PTR(-ENOTSOCK);
1608 sock = SOCKET_I(inode)->sk;
1609 if (sock->sk_family != AF_NETLINK)
1610 return ERR_PTR(-EINVAL);
1616 static struct sk_buff *netlink_alloc_large_skb(unsigned int size,
1619 struct sk_buff *skb;
1622 if (size <= NLMSG_GOODSIZE || broadcast)
1623 return alloc_skb(size, GFP_KERNEL);
1625 size = SKB_DATA_ALIGN(size) +
1626 SKB_DATA_ALIGN(sizeof(struct skb_shared_info));
1628 data = vmalloc(size);
1632 skb = __build_skb(data, size);
1636 skb->destructor = netlink_skb_destructor;
1642 * Attach a skb to a netlink socket.
1643 * The caller must hold a reference to the destination socket. On error, the
1644 * reference is dropped. The skb is not send to the destination, just all
1645 * all error checks are performed and memory in the queue is reserved.
1647 * < 0: error. skb freed, reference to sock dropped.
1649 * 1: repeat lookup - reference dropped while waiting for socket memory.
1651 int netlink_attachskb(struct sock *sk, struct sk_buff *skb,
1652 long *timeo, struct sock *ssk)
1654 struct netlink_sock *nlk;
1658 if ((atomic_read(&sk->sk_rmem_alloc) > sk->sk_rcvbuf ||
1659 test_bit(NETLINK_CONGESTED, &nlk->state)) &&
1660 !netlink_skb_is_mmaped(skb)) {
1661 DECLARE_WAITQUEUE(wait, current);
1663 if (!ssk || netlink_is_kernel(ssk))
1664 netlink_overrun(sk);
1670 __set_current_state(TASK_INTERRUPTIBLE);
1671 add_wait_queue(&nlk->wait, &wait);
1673 if ((atomic_read(&sk->sk_rmem_alloc) > sk->sk_rcvbuf ||
1674 test_bit(NETLINK_CONGESTED, &nlk->state)) &&
1675 !sock_flag(sk, SOCK_DEAD))
1676 *timeo = schedule_timeout(*timeo);
1678 __set_current_state(TASK_RUNNING);
1679 remove_wait_queue(&nlk->wait, &wait);
1682 if (signal_pending(current)) {
1684 return sock_intr_errno(*timeo);
1688 netlink_skb_set_owner_r(skb, sk);
1692 static int __netlink_sendskb(struct sock *sk, struct sk_buff *skb)
1696 netlink_deliver_tap(skb);
1698 #ifdef CONFIG_NETLINK_MMAP
1699 if (netlink_skb_is_mmaped(skb))
1700 netlink_queue_mmaped_skb(sk, skb);
1701 else if (netlink_rx_is_mmaped(sk))
1702 netlink_ring_set_copied(sk, skb);
1704 #endif /* CONFIG_NETLINK_MMAP */
1705 skb_queue_tail(&sk->sk_receive_queue, skb);
1706 sk->sk_data_ready(sk);
1710 int netlink_sendskb(struct sock *sk, struct sk_buff *skb)
1712 int len = __netlink_sendskb(sk, skb);
1718 void netlink_detachskb(struct sock *sk, struct sk_buff *skb)
1724 static struct sk_buff *netlink_trim(struct sk_buff *skb, gfp_t allocation)
1728 WARN_ON(skb->sk != NULL);
1729 if (netlink_skb_is_mmaped(skb))
1732 delta = skb->end - skb->tail;
1733 if (is_vmalloc_addr(skb->head) || delta * 2 < skb->truesize)
1736 if (skb_shared(skb)) {
1737 struct sk_buff *nskb = skb_clone(skb, allocation);
1744 if (!pskb_expand_head(skb, 0, -delta, allocation))
1745 skb->truesize -= delta;
1750 static int netlink_unicast_kernel(struct sock *sk, struct sk_buff *skb,
1754 struct netlink_sock *nlk = nlk_sk(sk);
1756 ret = -ECONNREFUSED;
1757 if (nlk->netlink_rcv != NULL) {
1759 netlink_skb_set_owner_r(skb, sk);
1760 NETLINK_CB(skb).sk = ssk;
1761 netlink_deliver_tap_kernel(sk, ssk, skb);
1762 nlk->netlink_rcv(skb);
1771 int netlink_unicast(struct sock *ssk, struct sk_buff *skb,
1772 u32 portid, int nonblock)
1778 skb = netlink_trim(skb, gfp_any());
1780 timeo = sock_sndtimeo(ssk, nonblock);
1782 sk = netlink_getsockbyportid(ssk, portid);
1787 if (netlink_is_kernel(sk))
1788 return netlink_unicast_kernel(sk, skb, ssk);
1790 if (sk_filter(sk, skb)) {
1797 err = netlink_attachskb(sk, skb, &timeo, ssk);
1803 return netlink_sendskb(sk, skb);
1805 EXPORT_SYMBOL(netlink_unicast);
1807 struct sk_buff *netlink_alloc_skb(struct sock *ssk, unsigned int size,
1808 u32 dst_portid, gfp_t gfp_mask)
1810 #ifdef CONFIG_NETLINK_MMAP
1811 struct sock *sk = NULL;
1812 struct sk_buff *skb;
1813 struct netlink_ring *ring;
1814 struct nl_mmap_hdr *hdr;
1815 unsigned int maxlen;
1817 sk = netlink_getsockbyportid(ssk, dst_portid);
1821 ring = &nlk_sk(sk)->rx_ring;
1822 /* fast-path without atomic ops for common case: non-mmaped receiver */
1823 if (ring->pg_vec == NULL)
1826 if (ring->frame_size - NL_MMAP_HDRLEN < size)
1829 skb = alloc_skb_head(gfp_mask);
1833 spin_lock_bh(&sk->sk_receive_queue.lock);
1834 /* check again under lock */
1835 if (ring->pg_vec == NULL)
1838 /* check again under lock */
1839 maxlen = ring->frame_size - NL_MMAP_HDRLEN;
1843 netlink_forward_ring(ring);
1844 hdr = netlink_current_frame(ring, NL_MMAP_STATUS_UNUSED);
1847 netlink_ring_setup_skb(skb, sk, ring, hdr);
1848 netlink_set_status(hdr, NL_MMAP_STATUS_RESERVED);
1849 atomic_inc(&ring->pending);
1850 netlink_increment_head(ring);
1852 spin_unlock_bh(&sk->sk_receive_queue.lock);
1857 spin_unlock_bh(&sk->sk_receive_queue.lock);
1858 netlink_overrun(sk);
1865 spin_unlock_bh(&sk->sk_receive_queue.lock);
1870 return alloc_skb(size, gfp_mask);
1872 EXPORT_SYMBOL_GPL(netlink_alloc_skb);
1874 int netlink_has_listeners(struct sock *sk, unsigned int group)
1877 struct listeners *listeners;
1879 BUG_ON(!netlink_is_kernel(sk));
1882 listeners = rcu_dereference(nl_table[sk->sk_protocol].listeners);
1884 if (listeners && group - 1 < nl_table[sk->sk_protocol].groups)
1885 res = test_bit(group - 1, listeners->masks);
1891 EXPORT_SYMBOL_GPL(netlink_has_listeners);
1893 static int netlink_broadcast_deliver(struct sock *sk, struct sk_buff *skb)
1895 struct netlink_sock *nlk = nlk_sk(sk);
1897 if (atomic_read(&sk->sk_rmem_alloc) <= sk->sk_rcvbuf &&
1898 !test_bit(NETLINK_CONGESTED, &nlk->state)) {
1899 netlink_skb_set_owner_r(skb, sk);
1900 __netlink_sendskb(sk, skb);
1901 return atomic_read(&sk->sk_rmem_alloc) > (sk->sk_rcvbuf >> 1);
1906 struct netlink_broadcast_data {
1907 struct sock *exclude_sk;
1912 int delivery_failure;
1916 struct sk_buff *skb, *skb2;
1917 int (*tx_filter)(struct sock *dsk, struct sk_buff *skb, void *data);
1921 static void do_one_broadcast(struct sock *sk,
1922 struct netlink_broadcast_data *p)
1924 struct netlink_sock *nlk = nlk_sk(sk);
1927 if (p->exclude_sk == sk)
1930 if (nlk->portid == p->portid || p->group - 1 >= nlk->ngroups ||
1931 !test_bit(p->group - 1, nlk->groups))
1934 if (!net_eq(sock_net(sk), p->net))
1938 netlink_overrun(sk);
1943 if (p->skb2 == NULL) {
1944 if (skb_shared(p->skb)) {
1945 p->skb2 = skb_clone(p->skb, p->allocation);
1947 p->skb2 = skb_get(p->skb);
1949 * skb ownership may have been set when
1950 * delivered to a previous socket.
1952 skb_orphan(p->skb2);
1955 if (p->skb2 == NULL) {
1956 netlink_overrun(sk);
1957 /* Clone failed. Notify ALL listeners. */
1959 if (nlk->flags & NETLINK_BROADCAST_SEND_ERROR)
1960 p->delivery_failure = 1;
1961 } else if (p->tx_filter && p->tx_filter(sk, p->skb2, p->tx_data)) {
1964 } else if (sk_filter(sk, p->skb2)) {
1967 } else if ((val = netlink_broadcast_deliver(sk, p->skb2)) < 0) {
1968 netlink_overrun(sk);
1969 if (nlk->flags & NETLINK_BROADCAST_SEND_ERROR)
1970 p->delivery_failure = 1;
1972 p->congested |= val;
1979 int netlink_broadcast_filtered(struct sock *ssk, struct sk_buff *skb, u32 portid,
1980 u32 group, gfp_t allocation,
1981 int (*filter)(struct sock *dsk, struct sk_buff *skb, void *data),
1984 struct net *net = sock_net(ssk);
1985 struct netlink_broadcast_data info;
1988 skb = netlink_trim(skb, allocation);
1990 info.exclude_sk = ssk;
1992 info.portid = portid;
1995 info.delivery_failure = 0;
1998 info.allocation = allocation;
2001 info.tx_filter = filter;
2002 info.tx_data = filter_data;
2004 /* While we sleep in clone, do not allow to change socket list */
2006 netlink_lock_table();
2008 sk_for_each_bound(sk, &nl_table[ssk->sk_protocol].mc_list)
2009 do_one_broadcast(sk, &info);
2013 netlink_unlock_table();
2015 if (info.delivery_failure) {
2016 kfree_skb(info.skb2);
2019 consume_skb(info.skb2);
2021 if (info.delivered) {
2022 if (info.congested && (allocation & __GFP_WAIT))
2028 EXPORT_SYMBOL(netlink_broadcast_filtered);
2030 int netlink_broadcast(struct sock *ssk, struct sk_buff *skb, u32 portid,
2031 u32 group, gfp_t allocation)
2033 return netlink_broadcast_filtered(ssk, skb, portid, group, allocation,
2036 EXPORT_SYMBOL(netlink_broadcast);
2038 struct netlink_set_err_data {
2039 struct sock *exclude_sk;
2045 static int do_one_set_err(struct sock *sk, struct netlink_set_err_data *p)
2047 struct netlink_sock *nlk = nlk_sk(sk);
2050 if (sk == p->exclude_sk)
2053 if (!net_eq(sock_net(sk), sock_net(p->exclude_sk)))
2056 if (nlk->portid == p->portid || p->group - 1 >= nlk->ngroups ||
2057 !test_bit(p->group - 1, nlk->groups))
2060 if (p->code == ENOBUFS && nlk->flags & NETLINK_RECV_NO_ENOBUFS) {
2065 sk->sk_err = p->code;
2066 sk->sk_error_report(sk);
2072 * netlink_set_err - report error to broadcast listeners
2073 * @ssk: the kernel netlink socket, as returned by netlink_kernel_create()
2074 * @portid: the PORTID of a process that we want to skip (if any)
2075 * @group: the broadcast group that will notice the error
2076 * @code: error code, must be negative (as usual in kernelspace)
2078 * This function returns the number of broadcast listeners that have set the
2079 * NETLINK_RECV_NO_ENOBUFS socket option.
2081 int netlink_set_err(struct sock *ssk, u32 portid, u32 group, int code)
2083 struct netlink_set_err_data info;
2087 info.exclude_sk = ssk;
2088 info.portid = portid;
2090 /* sk->sk_err wants a positive error value */
2093 read_lock(&nl_table_lock);
2095 sk_for_each_bound(sk, &nl_table[ssk->sk_protocol].mc_list)
2096 ret += do_one_set_err(sk, &info);
2098 read_unlock(&nl_table_lock);
2101 EXPORT_SYMBOL(netlink_set_err);
2103 /* must be called with netlink table grabbed */
2104 static void netlink_update_socket_mc(struct netlink_sock *nlk,
2108 int old, new = !!is_new, subscriptions;
2110 old = test_bit(group - 1, nlk->groups);
2111 subscriptions = nlk->subscriptions - old + new;
2113 __set_bit(group - 1, nlk->groups);
2115 __clear_bit(group - 1, nlk->groups);
2116 netlink_update_subscriptions(&nlk->sk, subscriptions);
2117 netlink_update_listeners(&nlk->sk);
2120 static int netlink_setsockopt(struct socket *sock, int level, int optname,
2121 char __user *optval, unsigned int optlen)
2123 struct sock *sk = sock->sk;
2124 struct netlink_sock *nlk = nlk_sk(sk);
2125 unsigned int val = 0;
2128 if (level != SOL_NETLINK)
2129 return -ENOPROTOOPT;
2131 if (optname != NETLINK_RX_RING && optname != NETLINK_TX_RING &&
2132 optlen >= sizeof(int) &&
2133 get_user(val, (unsigned int __user *)optval))
2137 case NETLINK_PKTINFO:
2139 nlk->flags |= NETLINK_RECV_PKTINFO;
2141 nlk->flags &= ~NETLINK_RECV_PKTINFO;
2144 case NETLINK_ADD_MEMBERSHIP:
2145 case NETLINK_DROP_MEMBERSHIP: {
2146 if (!netlink_allowed(sock, NL_CFG_F_NONROOT_RECV))
2148 err = netlink_realloc_groups(sk);
2151 if (!val || val - 1 >= nlk->ngroups)
2153 if (optname == NETLINK_ADD_MEMBERSHIP && nlk->netlink_bind) {
2154 err = nlk->netlink_bind(sock_net(sk), val);
2158 netlink_table_grab();
2159 netlink_update_socket_mc(nlk, val,
2160 optname == NETLINK_ADD_MEMBERSHIP);
2161 netlink_table_ungrab();
2162 if (optname == NETLINK_DROP_MEMBERSHIP && nlk->netlink_unbind)
2163 nlk->netlink_unbind(sock_net(sk), val);
2168 case NETLINK_BROADCAST_ERROR:
2170 nlk->flags |= NETLINK_BROADCAST_SEND_ERROR;
2172 nlk->flags &= ~NETLINK_BROADCAST_SEND_ERROR;
2175 case NETLINK_NO_ENOBUFS:
2177 nlk->flags |= NETLINK_RECV_NO_ENOBUFS;
2178 clear_bit(NETLINK_CONGESTED, &nlk->state);
2179 wake_up_interruptible(&nlk->wait);
2181 nlk->flags &= ~NETLINK_RECV_NO_ENOBUFS;
2185 #ifdef CONFIG_NETLINK_MMAP
2186 case NETLINK_RX_RING:
2187 case NETLINK_TX_RING: {
2188 struct nl_mmap_req req;
2190 /* Rings might consume more memory than queue limits, require
2193 if (!capable(CAP_NET_ADMIN))
2195 if (optlen < sizeof(req))
2197 if (copy_from_user(&req, optval, sizeof(req)))
2199 err = netlink_set_ring(sk, &req, false,
2200 optname == NETLINK_TX_RING);
2203 #endif /* CONFIG_NETLINK_MMAP */
2210 static int netlink_getsockopt(struct socket *sock, int level, int optname,
2211 char __user *optval, int __user *optlen)
2213 struct sock *sk = sock->sk;
2214 struct netlink_sock *nlk = nlk_sk(sk);
2217 if (level != SOL_NETLINK)
2218 return -ENOPROTOOPT;
2220 if (get_user(len, optlen))
2226 case NETLINK_PKTINFO:
2227 if (len < sizeof(int))
2230 val = nlk->flags & NETLINK_RECV_PKTINFO ? 1 : 0;
2231 if (put_user(len, optlen) ||
2232 put_user(val, optval))
2236 case NETLINK_BROADCAST_ERROR:
2237 if (len < sizeof(int))
2240 val = nlk->flags & NETLINK_BROADCAST_SEND_ERROR ? 1 : 0;
2241 if (put_user(len, optlen) ||
2242 put_user(val, optval))
2246 case NETLINK_NO_ENOBUFS:
2247 if (len < sizeof(int))
2250 val = nlk->flags & NETLINK_RECV_NO_ENOBUFS ? 1 : 0;
2251 if (put_user(len, optlen) ||
2252 put_user(val, optval))
2262 static void netlink_cmsg_recv_pktinfo(struct msghdr *msg, struct sk_buff *skb)
2264 struct nl_pktinfo info;
2266 info.group = NETLINK_CB(skb).dst_group;
2267 put_cmsg(msg, SOL_NETLINK, NETLINK_PKTINFO, sizeof(info), &info);
2270 static int netlink_sendmsg(struct socket *sock, struct msghdr *msg, size_t len)
2272 struct sock *sk = sock->sk;
2273 struct netlink_sock *nlk = nlk_sk(sk);
2274 DECLARE_SOCKADDR(struct sockaddr_nl *, addr, msg->msg_name);
2277 struct sk_buff *skb;
2279 struct scm_cookie scm;
2280 u32 netlink_skb_flags = 0;
2282 if (msg->msg_flags&MSG_OOB)
2285 err = scm_send(sock, msg, &scm, true);
2289 if (msg->msg_namelen) {
2291 if (addr->nl_family != AF_NETLINK)
2293 dst_portid = addr->nl_pid;
2294 dst_group = ffs(addr->nl_groups);
2296 if ((dst_group || dst_portid) &&
2297 !netlink_allowed(sock, NL_CFG_F_NONROOT_SEND))
2299 netlink_skb_flags |= NETLINK_SKB_DST;
2301 dst_portid = nlk->dst_portid;
2302 dst_group = nlk->dst_group;
2306 err = netlink_autobind(sock);
2311 /* It's a really convoluted way for userland to ask for mmaped
2312 * sendmsg(), but that's what we've got...
2314 if (netlink_tx_is_mmaped(sk) &&
2315 msg->msg_iter.type == ITER_IOVEC &&
2316 msg->msg_iter.nr_segs == 1 &&
2317 msg->msg_iter.iov->iov_base == NULL) {
2318 err = netlink_mmap_sendmsg(sk, msg, dst_portid, dst_group,
2324 if (len > sk->sk_sndbuf - 32)
2327 skb = netlink_alloc_large_skb(len, dst_group);
2331 NETLINK_CB(skb).portid = nlk->portid;
2332 NETLINK_CB(skb).dst_group = dst_group;
2333 NETLINK_CB(skb).creds = scm.creds;
2334 NETLINK_CB(skb).flags = netlink_skb_flags;
2337 if (memcpy_from_msg(skb_put(skb, len), msg, len)) {
2342 err = security_netlink_send(sk, skb);
2349 atomic_inc(&skb->users);
2350 netlink_broadcast(sk, skb, dst_portid, dst_group, GFP_KERNEL);
2352 err = netlink_unicast(sk, skb, dst_portid, msg->msg_flags&MSG_DONTWAIT);
2359 static int netlink_recvmsg(struct socket *sock, struct msghdr *msg, size_t len,
2362 struct scm_cookie scm;
2363 struct sock *sk = sock->sk;
2364 struct netlink_sock *nlk = nlk_sk(sk);
2365 int noblock = flags&MSG_DONTWAIT;
2367 struct sk_buff *skb, *data_skb;
2375 skb = skb_recv_datagram(sk, flags, noblock, &err);
2381 #ifdef CONFIG_COMPAT_NETLINK_MESSAGES
2382 if (unlikely(skb_shinfo(skb)->frag_list)) {
2384 * If this skb has a frag_list, then here that means that we
2385 * will have to use the frag_list skb's data for compat tasks
2386 * and the regular skb's data for normal (non-compat) tasks.
2388 * If we need to send the compat skb, assign it to the
2389 * 'data_skb' variable so that it will be used below for data
2390 * copying. We keep 'skb' for everything else, including
2391 * freeing both later.
2393 if (flags & MSG_CMSG_COMPAT)
2394 data_skb = skb_shinfo(skb)->frag_list;
2398 /* Record the max length of recvmsg() calls for future allocations */
2399 nlk->max_recvmsg_len = max(nlk->max_recvmsg_len, len);
2400 nlk->max_recvmsg_len = min_t(size_t, nlk->max_recvmsg_len,
2403 copied = data_skb->len;
2405 msg->msg_flags |= MSG_TRUNC;
2409 skb_reset_transport_header(data_skb);
2410 err = skb_copy_datagram_msg(data_skb, 0, msg, copied);
2412 if (msg->msg_name) {
2413 DECLARE_SOCKADDR(struct sockaddr_nl *, addr, msg->msg_name);
2414 addr->nl_family = AF_NETLINK;
2416 addr->nl_pid = NETLINK_CB(skb).portid;
2417 addr->nl_groups = netlink_group_mask(NETLINK_CB(skb).dst_group);
2418 msg->msg_namelen = sizeof(*addr);
2421 if (nlk->flags & NETLINK_RECV_PKTINFO)
2422 netlink_cmsg_recv_pktinfo(msg, skb);
2424 memset(&scm, 0, sizeof(scm));
2425 scm.creds = *NETLINK_CREDS(skb);
2426 if (flags & MSG_TRUNC)
2427 copied = data_skb->len;
2429 skb_free_datagram(sk, skb);
2431 if (nlk->cb_running &&
2432 atomic_read(&sk->sk_rmem_alloc) <= sk->sk_rcvbuf / 2) {
2433 ret = netlink_dump(sk);
2436 sk->sk_error_report(sk);
2440 scm_recv(sock, msg, &scm, flags);
2442 netlink_rcv_wake(sk);
2443 return err ? : copied;
2446 static void netlink_data_ready(struct sock *sk)
2452 * We export these functions to other modules. They provide a
2453 * complete set of kernel non-blocking support for message
2458 __netlink_kernel_create(struct net *net, int unit, struct module *module,
2459 struct netlink_kernel_cfg *cfg)
2461 struct socket *sock;
2463 struct netlink_sock *nlk;
2464 struct listeners *listeners = NULL;
2465 struct mutex *cb_mutex = cfg ? cfg->cb_mutex : NULL;
2466 unsigned int groups;
2470 if (unit < 0 || unit >= MAX_LINKS)
2473 if (sock_create_lite(PF_NETLINK, SOCK_DGRAM, unit, &sock))
2477 * We have to just have a reference on the net from sk, but don't
2478 * get_net it. Besides, we cannot get and then put the net here.
2479 * So we create one inside init_net and the move it to net.
2482 if (__netlink_create(&init_net, sock, cb_mutex, unit) < 0)
2483 goto out_sock_release_nosk;
2486 sk_change_net(sk, net);
2488 if (!cfg || cfg->groups < 32)
2491 groups = cfg->groups;
2493 listeners = kzalloc(sizeof(*listeners) + NLGRPSZ(groups), GFP_KERNEL);
2495 goto out_sock_release;
2497 sk->sk_data_ready = netlink_data_ready;
2498 if (cfg && cfg->input)
2499 nlk_sk(sk)->netlink_rcv = cfg->input;
2501 if (netlink_insert(sk, 0))
2502 goto out_sock_release;
2505 nlk->flags |= NETLINK_KERNEL_SOCKET;
2507 netlink_table_grab();
2508 if (!nl_table[unit].registered) {
2509 nl_table[unit].groups = groups;
2510 rcu_assign_pointer(nl_table[unit].listeners, listeners);
2511 nl_table[unit].cb_mutex = cb_mutex;
2512 nl_table[unit].module = module;
2514 nl_table[unit].bind = cfg->bind;
2515 nl_table[unit].unbind = cfg->unbind;
2516 nl_table[unit].flags = cfg->flags;
2518 nl_table[unit].compare = cfg->compare;
2520 nl_table[unit].registered = 1;
2523 nl_table[unit].registered++;
2525 netlink_table_ungrab();
2530 netlink_kernel_release(sk);
2533 out_sock_release_nosk:
2537 EXPORT_SYMBOL(__netlink_kernel_create);
2540 netlink_kernel_release(struct sock *sk)
2542 sk_release_kernel(sk);
2544 EXPORT_SYMBOL(netlink_kernel_release);
2546 int __netlink_change_ngroups(struct sock *sk, unsigned int groups)
2548 struct listeners *new, *old;
2549 struct netlink_table *tbl = &nl_table[sk->sk_protocol];
2554 if (NLGRPSZ(tbl->groups) < NLGRPSZ(groups)) {
2555 new = kzalloc(sizeof(*new) + NLGRPSZ(groups), GFP_ATOMIC);
2558 old = nl_deref_protected(tbl->listeners);
2559 memcpy(new->masks, old->masks, NLGRPSZ(tbl->groups));
2560 rcu_assign_pointer(tbl->listeners, new);
2562 kfree_rcu(old, rcu);
2564 tbl->groups = groups;
2570 * netlink_change_ngroups - change number of multicast groups
2572 * This changes the number of multicast groups that are available
2573 * on a certain netlink family. Note that it is not possible to
2574 * change the number of groups to below 32. Also note that it does
2575 * not implicitly call netlink_clear_multicast_users() when the
2576 * number of groups is reduced.
2578 * @sk: The kernel netlink socket, as returned by netlink_kernel_create().
2579 * @groups: The new number of groups.
2581 int netlink_change_ngroups(struct sock *sk, unsigned int groups)
2585 netlink_table_grab();
2586 err = __netlink_change_ngroups(sk, groups);
2587 netlink_table_ungrab();
2592 void __netlink_clear_multicast_users(struct sock *ksk, unsigned int group)
2595 struct netlink_table *tbl = &nl_table[ksk->sk_protocol];
2597 sk_for_each_bound(sk, &tbl->mc_list)
2598 netlink_update_socket_mc(nlk_sk(sk), group, 0);
2602 __nlmsg_put(struct sk_buff *skb, u32 portid, u32 seq, int type, int len, int flags)
2604 struct nlmsghdr *nlh;
2605 int size = nlmsg_msg_size(len);
2607 nlh = (struct nlmsghdr *)skb_put(skb, NLMSG_ALIGN(size));
2608 nlh->nlmsg_type = type;
2609 nlh->nlmsg_len = size;
2610 nlh->nlmsg_flags = flags;
2611 nlh->nlmsg_pid = portid;
2612 nlh->nlmsg_seq = seq;
2613 if (!__builtin_constant_p(size) || NLMSG_ALIGN(size) - size != 0)
2614 memset(nlmsg_data(nlh) + len, 0, NLMSG_ALIGN(size) - size);
2617 EXPORT_SYMBOL(__nlmsg_put);
2620 * It looks a bit ugly.
2621 * It would be better to create kernel thread.
2624 static int netlink_dump(struct sock *sk)
2626 struct netlink_sock *nlk = nlk_sk(sk);
2627 struct netlink_callback *cb;
2628 struct sk_buff *skb = NULL;
2629 struct nlmsghdr *nlh;
2630 int len, err = -ENOBUFS;
2633 mutex_lock(nlk->cb_mutex);
2634 if (!nlk->cb_running) {
2640 alloc_size = max_t(int, cb->min_dump_alloc, NLMSG_GOODSIZE);
2642 if (!netlink_rx_is_mmaped(sk) &&
2643 atomic_read(&sk->sk_rmem_alloc) >= sk->sk_rcvbuf)
2646 /* NLMSG_GOODSIZE is small to avoid high order allocations being
2647 * required, but it makes sense to _attempt_ a 16K bytes allocation
2648 * to reduce number of system calls on dump operations, if user
2649 * ever provided a big enough buffer.
2651 if (alloc_size < nlk->max_recvmsg_len) {
2652 skb = netlink_alloc_skb(sk,
2653 nlk->max_recvmsg_len,
2658 /* available room should be exact amount to avoid MSG_TRUNC */
2660 skb_reserve(skb, skb_tailroom(skb) -
2661 nlk->max_recvmsg_len);
2664 skb = netlink_alloc_skb(sk, alloc_size, nlk->portid,
2668 netlink_skb_set_owner_r(skb, sk);
2670 len = cb->dump(skb, cb);
2673 mutex_unlock(nlk->cb_mutex);
2675 if (sk_filter(sk, skb))
2678 __netlink_sendskb(sk, skb);
2682 nlh = nlmsg_put_answer(skb, cb, NLMSG_DONE, sizeof(len), NLM_F_MULTI);
2686 nl_dump_check_consistent(cb, nlh);
2688 memcpy(nlmsg_data(nlh), &len, sizeof(len));
2690 if (sk_filter(sk, skb))
2693 __netlink_sendskb(sk, skb);
2698 nlk->cb_running = false;
2699 mutex_unlock(nlk->cb_mutex);
2700 module_put(cb->module);
2701 consume_skb(cb->skb);
2705 mutex_unlock(nlk->cb_mutex);
2710 int __netlink_dump_start(struct sock *ssk, struct sk_buff *skb,
2711 const struct nlmsghdr *nlh,
2712 struct netlink_dump_control *control)
2714 struct netlink_callback *cb;
2716 struct netlink_sock *nlk;
2719 /* Memory mapped dump requests need to be copied to avoid looping
2720 * on the pending state in netlink_mmap_sendmsg() while the CB hold
2721 * a reference to the skb.
2723 if (netlink_skb_is_mmaped(skb)) {
2724 skb = skb_copy(skb, GFP_KERNEL);
2728 atomic_inc(&skb->users);
2730 sk = netlink_lookup(sock_net(ssk), ssk->sk_protocol, NETLINK_CB(skb).portid);
2732 ret = -ECONNREFUSED;
2737 mutex_lock(nlk->cb_mutex);
2738 /* A dump is in progress... */
2739 if (nlk->cb_running) {
2743 /* add reference of module which cb->dump belongs to */
2744 if (!try_module_get(control->module)) {
2745 ret = -EPROTONOSUPPORT;
2750 memset(cb, 0, sizeof(*cb));
2751 cb->dump = control->dump;
2752 cb->done = control->done;
2754 cb->data = control->data;
2755 cb->module = control->module;
2756 cb->min_dump_alloc = control->min_dump_alloc;
2759 nlk->cb_running = true;
2761 mutex_unlock(nlk->cb_mutex);
2763 ret = netlink_dump(sk);
2769 /* We successfully started a dump, by returning -EINTR we
2770 * signal not to send ACK even if it was requested.
2776 mutex_unlock(nlk->cb_mutex);
2781 EXPORT_SYMBOL(__netlink_dump_start);
2783 void netlink_ack(struct sk_buff *in_skb, struct nlmsghdr *nlh, int err)
2785 struct sk_buff *skb;
2786 struct nlmsghdr *rep;
2787 struct nlmsgerr *errmsg;
2788 size_t payload = sizeof(*errmsg);
2790 /* error messages get the original request appened */
2792 payload += nlmsg_len(nlh);
2794 skb = netlink_alloc_skb(in_skb->sk, nlmsg_total_size(payload),
2795 NETLINK_CB(in_skb).portid, GFP_KERNEL);
2799 sk = netlink_lookup(sock_net(in_skb->sk),
2800 in_skb->sk->sk_protocol,
2801 NETLINK_CB(in_skb).portid);
2803 sk->sk_err = ENOBUFS;
2804 sk->sk_error_report(sk);
2810 rep = __nlmsg_put(skb, NETLINK_CB(in_skb).portid, nlh->nlmsg_seq,
2811 NLMSG_ERROR, payload, 0);
2812 errmsg = nlmsg_data(rep);
2813 errmsg->error = err;
2814 memcpy(&errmsg->msg, nlh, err ? nlh->nlmsg_len : sizeof(*nlh));
2815 netlink_unicast(in_skb->sk, skb, NETLINK_CB(in_skb).portid, MSG_DONTWAIT);
2817 EXPORT_SYMBOL(netlink_ack);
2819 int netlink_rcv_skb(struct sk_buff *skb, int (*cb)(struct sk_buff *,
2822 struct nlmsghdr *nlh;
2825 while (skb->len >= nlmsg_total_size(0)) {
2828 nlh = nlmsg_hdr(skb);
2831 if (nlh->nlmsg_len < NLMSG_HDRLEN || skb->len < nlh->nlmsg_len)
2834 /* Only requests are handled by the kernel */
2835 if (!(nlh->nlmsg_flags & NLM_F_REQUEST))
2838 /* Skip control messages */
2839 if (nlh->nlmsg_type < NLMSG_MIN_TYPE)
2847 if (nlh->nlmsg_flags & NLM_F_ACK || err)
2848 netlink_ack(skb, nlh, err);
2851 msglen = NLMSG_ALIGN(nlh->nlmsg_len);
2852 if (msglen > skb->len)
2854 skb_pull(skb, msglen);
2859 EXPORT_SYMBOL(netlink_rcv_skb);
2862 * nlmsg_notify - send a notification netlink message
2863 * @sk: netlink socket to use
2864 * @skb: notification message
2865 * @portid: destination netlink portid for reports or 0
2866 * @group: destination multicast group or 0
2867 * @report: 1 to report back, 0 to disable
2868 * @flags: allocation flags
2870 int nlmsg_notify(struct sock *sk, struct sk_buff *skb, u32 portid,
2871 unsigned int group, int report, gfp_t flags)
2876 int exclude_portid = 0;
2879 atomic_inc(&skb->users);
2880 exclude_portid = portid;
2883 /* errors reported via destination sk->sk_err, but propagate
2884 * delivery errors if NETLINK_BROADCAST_ERROR flag is set */
2885 err = nlmsg_multicast(sk, skb, exclude_portid, group, flags);
2891 err2 = nlmsg_unicast(sk, skb, portid);
2892 if (!err || err == -ESRCH)
2898 EXPORT_SYMBOL(nlmsg_notify);
2900 #ifdef CONFIG_PROC_FS
2901 struct nl_seq_iter {
2902 struct seq_net_private p;
2903 struct rhashtable_iter hti;
2907 static int netlink_walk_start(struct nl_seq_iter *iter)
2911 err = rhashtable_walk_init(&nl_table[iter->link].hash, &iter->hti);
2913 iter->link = MAX_LINKS;
2917 err = rhashtable_walk_start(&iter->hti);
2918 return err == -EAGAIN ? 0 : err;
2921 static void netlink_walk_stop(struct nl_seq_iter *iter)
2923 rhashtable_walk_stop(&iter->hti);
2924 rhashtable_walk_exit(&iter->hti);
2927 static void *__netlink_seq_next(struct seq_file *seq)
2929 struct nl_seq_iter *iter = seq->private;
2930 struct netlink_sock *nlk;
2936 nlk = rhashtable_walk_next(&iter->hti);
2939 if (PTR_ERR(nlk) == -EAGAIN)
2948 netlink_walk_stop(iter);
2949 if (++iter->link >= MAX_LINKS)
2952 err = netlink_walk_start(iter);
2954 return ERR_PTR(err);
2956 } while (sock_net(&nlk->sk) != seq_file_net(seq));
2961 static void *netlink_seq_start(struct seq_file *seq, loff_t *posp)
2963 struct nl_seq_iter *iter = seq->private;
2964 void *obj = SEQ_START_TOKEN;
2970 err = netlink_walk_start(iter);
2972 return ERR_PTR(err);
2974 for (pos = *posp; pos && obj && !IS_ERR(obj); pos--)
2975 obj = __netlink_seq_next(seq);
2980 static void *netlink_seq_next(struct seq_file *seq, void *v, loff_t *pos)
2983 return __netlink_seq_next(seq);
2986 static void netlink_seq_stop(struct seq_file *seq, void *v)
2988 struct nl_seq_iter *iter = seq->private;
2990 if (iter->link >= MAX_LINKS)
2993 netlink_walk_stop(iter);
2997 static int netlink_seq_show(struct seq_file *seq, void *v)
2999 if (v == SEQ_START_TOKEN) {
3001 "sk Eth Pid Groups "
3002 "Rmem Wmem Dump Locks Drops Inode\n");
3005 struct netlink_sock *nlk = nlk_sk(s);
3007 seq_printf(seq, "%pK %-3d %-6u %08x %-8d %-8d %d %-8d %-8d %-8lu\n",
3011 nlk->groups ? (u32)nlk->groups[0] : 0,
3012 sk_rmem_alloc_get(s),
3013 sk_wmem_alloc_get(s),
3015 atomic_read(&s->sk_refcnt),
3016 atomic_read(&s->sk_drops),
3024 static const struct seq_operations netlink_seq_ops = {
3025 .start = netlink_seq_start,
3026 .next = netlink_seq_next,
3027 .stop = netlink_seq_stop,
3028 .show = netlink_seq_show,
3032 static int netlink_seq_open(struct inode *inode, struct file *file)
3034 return seq_open_net(inode, file, &netlink_seq_ops,
3035 sizeof(struct nl_seq_iter));
3038 static const struct file_operations netlink_seq_fops = {
3039 .owner = THIS_MODULE,
3040 .open = netlink_seq_open,
3042 .llseek = seq_lseek,
3043 .release = seq_release_net,
3048 int netlink_register_notifier(struct notifier_block *nb)
3050 return atomic_notifier_chain_register(&netlink_chain, nb);
3052 EXPORT_SYMBOL(netlink_register_notifier);
3054 int netlink_unregister_notifier(struct notifier_block *nb)
3056 return atomic_notifier_chain_unregister(&netlink_chain, nb);
3058 EXPORT_SYMBOL(netlink_unregister_notifier);
3060 static const struct proto_ops netlink_ops = {
3061 .family = PF_NETLINK,
3062 .owner = THIS_MODULE,
3063 .release = netlink_release,
3064 .bind = netlink_bind,
3065 .connect = netlink_connect,
3066 .socketpair = sock_no_socketpair,
3067 .accept = sock_no_accept,
3068 .getname = netlink_getname,
3069 .poll = netlink_poll,
3070 .ioctl = sock_no_ioctl,
3071 .listen = sock_no_listen,
3072 .shutdown = sock_no_shutdown,
3073 .setsockopt = netlink_setsockopt,
3074 .getsockopt = netlink_getsockopt,
3075 .sendmsg = netlink_sendmsg,
3076 .recvmsg = netlink_recvmsg,
3077 .mmap = netlink_mmap,
3078 .sendpage = sock_no_sendpage,
3081 static const struct net_proto_family netlink_family_ops = {
3082 .family = PF_NETLINK,
3083 .create = netlink_create,
3084 .owner = THIS_MODULE, /* for consistency 8) */
3087 static int __net_init netlink_net_init(struct net *net)
3089 #ifdef CONFIG_PROC_FS
3090 if (!proc_create("netlink", 0, net->proc_net, &netlink_seq_fops))
3096 static void __net_exit netlink_net_exit(struct net *net)
3098 #ifdef CONFIG_PROC_FS
3099 remove_proc_entry("netlink", net->proc_net);
3103 static void __init netlink_add_usersock_entry(void)
3105 struct listeners *listeners;
3108 listeners = kzalloc(sizeof(*listeners) + NLGRPSZ(groups), GFP_KERNEL);
3110 panic("netlink_add_usersock_entry: Cannot allocate listeners\n");
3112 netlink_table_grab();
3114 nl_table[NETLINK_USERSOCK].groups = groups;
3115 rcu_assign_pointer(nl_table[NETLINK_USERSOCK].listeners, listeners);
3116 nl_table[NETLINK_USERSOCK].module = THIS_MODULE;
3117 nl_table[NETLINK_USERSOCK].registered = 1;
3118 nl_table[NETLINK_USERSOCK].flags = NL_CFG_F_NONROOT_SEND;
3120 netlink_table_ungrab();
3123 static struct pernet_operations __net_initdata netlink_net_ops = {
3124 .init = netlink_net_init,
3125 .exit = netlink_net_exit,
3128 static inline u32 netlink_hash(const void *data, u32 len, u32 seed)
3130 const struct netlink_sock *nlk = data;
3131 struct netlink_compare_arg arg;
3133 netlink_compare_arg_init(&arg, sock_net(&nlk->sk), nlk->portid);
3134 return jhash2((u32 *)&arg, netlink_compare_arg_len / sizeof(u32), seed);
3137 static const struct rhashtable_params netlink_rhashtable_params = {
3138 .head_offset = offsetof(struct netlink_sock, node),
3139 .key_len = netlink_compare_arg_len,
3140 .obj_hashfn = netlink_hash,
3141 .obj_cmpfn = netlink_compare,
3142 .automatic_shrinking = true,
3145 static int __init netlink_proto_init(void)
3148 int err = proto_register(&netlink_proto, 0);
3153 BUILD_BUG_ON(sizeof(struct netlink_skb_parms) > FIELD_SIZEOF(struct sk_buff, cb));
3155 nl_table = kcalloc(MAX_LINKS, sizeof(*nl_table), GFP_KERNEL);
3159 for (i = 0; i < MAX_LINKS; i++) {
3160 if (rhashtable_init(&nl_table[i].hash,
3161 &netlink_rhashtable_params) < 0) {
3163 rhashtable_destroy(&nl_table[i].hash);
3169 INIT_LIST_HEAD(&netlink_tap_all);
3171 netlink_add_usersock_entry();
3173 sock_register(&netlink_family_ops);
3174 register_pernet_subsys(&netlink_net_ops);
3175 /* The netlink device handler may be needed early. */
3180 panic("netlink_init: Cannot allocate nl_table\n");
3183 core_initcall(netlink_proto_init);