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_S_CONGESTED 0x0
82 #define NETLINK_F_KERNEL_SOCKET 0x1
83 #define NETLINK_F_RECV_PKTINFO 0x2
84 #define NETLINK_F_BROADCAST_SEND_ERROR 0x4
85 #define NETLINK_F_RECV_NO_ENOBUFS 0x8
86 #define NETLINK_F_LISTEN_ALL_NSID 0x10
87 #define NETLINK_F_CAP_ACK 0x20
89 static inline int netlink_is_kernel(struct sock *sk)
91 return nlk_sk(sk)->flags & NETLINK_F_KERNEL_SOCKET;
94 struct netlink_table *nl_table __read_mostly;
95 EXPORT_SYMBOL_GPL(nl_table);
97 static DECLARE_WAIT_QUEUE_HEAD(nl_table_wait);
99 static int netlink_dump(struct sock *sk);
100 static void netlink_skb_destructor(struct sk_buff *skb);
102 /* nl_table locking explained:
103 * Lookup and traversal are protected with an RCU read-side lock. Insertion
104 * and removal are protected with per bucket lock while using RCU list
105 * modification primitives and may run in parallel to RCU protected lookups.
106 * Destruction of the Netlink socket may only occur *after* nl_table_lock has
107 * been acquired * either during or after the socket has been removed from
108 * the list and after an RCU grace period.
110 DEFINE_RWLOCK(nl_table_lock);
111 EXPORT_SYMBOL_GPL(nl_table_lock);
112 static atomic_t nl_table_users = ATOMIC_INIT(0);
114 #define nl_deref_protected(X) rcu_dereference_protected(X, lockdep_is_held(&nl_table_lock));
116 static ATOMIC_NOTIFIER_HEAD(netlink_chain);
118 static DEFINE_SPINLOCK(netlink_tap_lock);
119 static struct list_head netlink_tap_all __read_mostly;
121 static const struct rhashtable_params netlink_rhashtable_params;
123 static inline u32 netlink_group_mask(u32 group)
125 return group ? 1 << (group - 1) : 0;
128 int netlink_add_tap(struct netlink_tap *nt)
130 if (unlikely(nt->dev->type != ARPHRD_NETLINK))
133 spin_lock(&netlink_tap_lock);
134 list_add_rcu(&nt->list, &netlink_tap_all);
135 spin_unlock(&netlink_tap_lock);
137 __module_get(nt->module);
141 EXPORT_SYMBOL_GPL(netlink_add_tap);
143 static int __netlink_remove_tap(struct netlink_tap *nt)
146 struct netlink_tap *tmp;
148 spin_lock(&netlink_tap_lock);
150 list_for_each_entry(tmp, &netlink_tap_all, list) {
152 list_del_rcu(&nt->list);
158 pr_warn("__netlink_remove_tap: %p not found\n", nt);
160 spin_unlock(&netlink_tap_lock);
163 module_put(nt->module);
165 return found ? 0 : -ENODEV;
168 int netlink_remove_tap(struct netlink_tap *nt)
172 ret = __netlink_remove_tap(nt);
177 EXPORT_SYMBOL_GPL(netlink_remove_tap);
179 static bool netlink_filter_tap(const struct sk_buff *skb)
181 struct sock *sk = skb->sk;
183 /* We take the more conservative approach and
184 * whitelist socket protocols that may pass.
186 switch (sk->sk_protocol) {
188 case NETLINK_USERSOCK:
189 case NETLINK_SOCK_DIAG:
192 case NETLINK_FIB_LOOKUP:
193 case NETLINK_NETFILTER:
194 case NETLINK_GENERIC:
201 static int __netlink_deliver_tap_skb(struct sk_buff *skb,
202 struct net_device *dev)
204 struct sk_buff *nskb;
205 struct sock *sk = skb->sk;
209 nskb = skb_clone(skb, GFP_ATOMIC);
212 nskb->protocol = htons((u16) sk->sk_protocol);
213 nskb->pkt_type = netlink_is_kernel(sk) ?
214 PACKET_KERNEL : PACKET_USER;
215 skb_reset_network_header(nskb);
216 ret = dev_queue_xmit(nskb);
217 if (unlikely(ret > 0))
218 ret = net_xmit_errno(ret);
225 static void __netlink_deliver_tap(struct sk_buff *skb)
228 struct netlink_tap *tmp;
230 if (!netlink_filter_tap(skb))
233 list_for_each_entry_rcu(tmp, &netlink_tap_all, list) {
234 ret = __netlink_deliver_tap_skb(skb, tmp->dev);
240 static void netlink_deliver_tap(struct sk_buff *skb)
244 if (unlikely(!list_empty(&netlink_tap_all)))
245 __netlink_deliver_tap(skb);
250 static void netlink_deliver_tap_kernel(struct sock *dst, struct sock *src,
253 if (!(netlink_is_kernel(dst) && netlink_is_kernel(src)))
254 netlink_deliver_tap(skb);
257 static void netlink_overrun(struct sock *sk)
259 struct netlink_sock *nlk = nlk_sk(sk);
261 if (!(nlk->flags & NETLINK_F_RECV_NO_ENOBUFS)) {
262 if (!test_and_set_bit(NETLINK_S_CONGESTED,
263 &nlk_sk(sk)->state)) {
264 sk->sk_err = ENOBUFS;
265 sk->sk_error_report(sk);
268 atomic_inc(&sk->sk_drops);
271 static void netlink_rcv_wake(struct sock *sk)
273 struct netlink_sock *nlk = nlk_sk(sk);
275 if (skb_queue_empty(&sk->sk_receive_queue))
276 clear_bit(NETLINK_S_CONGESTED, &nlk->state);
277 if (!test_bit(NETLINK_S_CONGESTED, &nlk->state))
278 wake_up_interruptible(&nlk->wait);
281 #ifdef CONFIG_NETLINK_MMAP
282 static bool netlink_skb_is_mmaped(const struct sk_buff *skb)
284 return NETLINK_CB(skb).flags & NETLINK_SKB_MMAPED;
287 static bool netlink_rx_is_mmaped(struct sock *sk)
289 return nlk_sk(sk)->rx_ring.pg_vec != NULL;
292 static bool netlink_tx_is_mmaped(struct sock *sk)
294 return nlk_sk(sk)->tx_ring.pg_vec != NULL;
297 static __pure struct page *pgvec_to_page(const void *addr)
299 if (is_vmalloc_addr(addr))
300 return vmalloc_to_page(addr);
302 return virt_to_page(addr);
305 static void free_pg_vec(void **pg_vec, unsigned int order, unsigned int len)
309 for (i = 0; i < len; i++) {
310 if (pg_vec[i] != NULL) {
311 if (is_vmalloc_addr(pg_vec[i]))
314 free_pages((unsigned long)pg_vec[i], order);
320 static void *alloc_one_pg_vec_page(unsigned long order)
323 gfp_t gfp_flags = GFP_KERNEL | __GFP_COMP | __GFP_ZERO |
324 __GFP_NOWARN | __GFP_NORETRY;
326 buffer = (void *)__get_free_pages(gfp_flags, order);
330 buffer = vzalloc((1 << order) * PAGE_SIZE);
334 gfp_flags &= ~__GFP_NORETRY;
335 return (void *)__get_free_pages(gfp_flags, order);
338 static void **alloc_pg_vec(struct netlink_sock *nlk,
339 struct nl_mmap_req *req, unsigned int order)
341 unsigned int block_nr = req->nm_block_nr;
345 pg_vec = kcalloc(block_nr, sizeof(void *), GFP_KERNEL);
349 for (i = 0; i < block_nr; i++) {
350 pg_vec[i] = alloc_one_pg_vec_page(order);
351 if (pg_vec[i] == NULL)
357 free_pg_vec(pg_vec, order, block_nr);
363 __netlink_set_ring(struct sock *sk, struct nl_mmap_req *req, bool tx_ring, void **pg_vec,
366 struct netlink_sock *nlk = nlk_sk(sk);
367 struct sk_buff_head *queue;
368 struct netlink_ring *ring;
370 queue = tx_ring ? &sk->sk_write_queue : &sk->sk_receive_queue;
371 ring = tx_ring ? &nlk->tx_ring : &nlk->rx_ring;
373 spin_lock_bh(&queue->lock);
375 ring->frame_max = req->nm_frame_nr - 1;
377 ring->frame_size = req->nm_frame_size;
378 ring->pg_vec_pages = req->nm_block_size / PAGE_SIZE;
380 swap(ring->pg_vec_len, req->nm_block_nr);
381 swap(ring->pg_vec_order, order);
382 swap(ring->pg_vec, pg_vec);
384 __skb_queue_purge(queue);
385 spin_unlock_bh(&queue->lock);
387 WARN_ON(atomic_read(&nlk->mapped));
390 free_pg_vec(pg_vec, order, req->nm_block_nr);
393 static int netlink_set_ring(struct sock *sk, struct nl_mmap_req *req,
396 struct netlink_sock *nlk = nlk_sk(sk);
397 struct netlink_ring *ring;
398 void **pg_vec = NULL;
399 unsigned int order = 0;
401 ring = tx_ring ? &nlk->tx_ring : &nlk->rx_ring;
403 if (atomic_read(&nlk->mapped))
405 if (atomic_read(&ring->pending))
408 if (req->nm_block_nr) {
409 if (ring->pg_vec != NULL)
412 if ((int)req->nm_block_size <= 0)
414 if (!PAGE_ALIGNED(req->nm_block_size))
416 if (req->nm_frame_size < NL_MMAP_HDRLEN)
418 if (!IS_ALIGNED(req->nm_frame_size, NL_MMAP_MSG_ALIGNMENT))
421 ring->frames_per_block = req->nm_block_size /
423 if (ring->frames_per_block == 0)
425 if (ring->frames_per_block * req->nm_block_nr !=
429 order = get_order(req->nm_block_size);
430 pg_vec = alloc_pg_vec(nlk, req, order);
434 if (req->nm_frame_nr)
438 mutex_lock(&nlk->pg_vec_lock);
439 if (atomic_read(&nlk->mapped) == 0) {
440 __netlink_set_ring(sk, req, tx_ring, pg_vec, order);
441 mutex_unlock(&nlk->pg_vec_lock);
445 mutex_unlock(&nlk->pg_vec_lock);
448 free_pg_vec(pg_vec, order, req->nm_block_nr);
453 static void netlink_mm_open(struct vm_area_struct *vma)
455 struct file *file = vma->vm_file;
456 struct socket *sock = file->private_data;
457 struct sock *sk = sock->sk;
460 atomic_inc(&nlk_sk(sk)->mapped);
463 static void netlink_mm_close(struct vm_area_struct *vma)
465 struct file *file = vma->vm_file;
466 struct socket *sock = file->private_data;
467 struct sock *sk = sock->sk;
470 atomic_dec(&nlk_sk(sk)->mapped);
473 static const struct vm_operations_struct netlink_mmap_ops = {
474 .open = netlink_mm_open,
475 .close = netlink_mm_close,
478 static int netlink_mmap(struct file *file, struct socket *sock,
479 struct vm_area_struct *vma)
481 struct sock *sk = sock->sk;
482 struct netlink_sock *nlk = nlk_sk(sk);
483 struct netlink_ring *ring;
484 unsigned long start, size, expected;
491 mutex_lock(&nlk->pg_vec_lock);
494 for (ring = &nlk->rx_ring; ring <= &nlk->tx_ring; ring++) {
495 if (ring->pg_vec == NULL)
497 expected += ring->pg_vec_len * ring->pg_vec_pages * PAGE_SIZE;
503 size = vma->vm_end - vma->vm_start;
504 if (size != expected)
507 start = vma->vm_start;
508 for (ring = &nlk->rx_ring; ring <= &nlk->tx_ring; ring++) {
509 if (ring->pg_vec == NULL)
512 for (i = 0; i < ring->pg_vec_len; i++) {
514 void *kaddr = ring->pg_vec[i];
517 for (pg_num = 0; pg_num < ring->pg_vec_pages; pg_num++) {
518 page = pgvec_to_page(kaddr);
519 err = vm_insert_page(vma, start, page);
528 atomic_inc(&nlk->mapped);
529 vma->vm_ops = &netlink_mmap_ops;
532 mutex_unlock(&nlk->pg_vec_lock);
536 static void netlink_frame_flush_dcache(const struct nl_mmap_hdr *hdr, unsigned int nm_len)
538 #if ARCH_IMPLEMENTS_FLUSH_DCACHE_PAGE == 1
539 struct page *p_start, *p_end;
541 /* First page is flushed through netlink_{get,set}_status */
542 p_start = pgvec_to_page(hdr + PAGE_SIZE);
543 p_end = pgvec_to_page((void *)hdr + NL_MMAP_HDRLEN + nm_len - 1);
544 while (p_start <= p_end) {
545 flush_dcache_page(p_start);
551 static enum nl_mmap_status netlink_get_status(const struct nl_mmap_hdr *hdr)
554 flush_dcache_page(pgvec_to_page(hdr));
555 return hdr->nm_status;
558 static void netlink_set_status(struct nl_mmap_hdr *hdr,
559 enum nl_mmap_status status)
562 hdr->nm_status = status;
563 flush_dcache_page(pgvec_to_page(hdr));
566 static struct nl_mmap_hdr *
567 __netlink_lookup_frame(const struct netlink_ring *ring, unsigned int pos)
569 unsigned int pg_vec_pos, frame_off;
571 pg_vec_pos = pos / ring->frames_per_block;
572 frame_off = pos % ring->frames_per_block;
574 return ring->pg_vec[pg_vec_pos] + (frame_off * ring->frame_size);
577 static struct nl_mmap_hdr *
578 netlink_lookup_frame(const struct netlink_ring *ring, unsigned int pos,
579 enum nl_mmap_status status)
581 struct nl_mmap_hdr *hdr;
583 hdr = __netlink_lookup_frame(ring, pos);
584 if (netlink_get_status(hdr) != status)
590 static struct nl_mmap_hdr *
591 netlink_current_frame(const struct netlink_ring *ring,
592 enum nl_mmap_status status)
594 return netlink_lookup_frame(ring, ring->head, status);
597 static void netlink_increment_head(struct netlink_ring *ring)
599 ring->head = ring->head != ring->frame_max ? ring->head + 1 : 0;
602 static void netlink_forward_ring(struct netlink_ring *ring)
604 unsigned int head = ring->head;
605 const struct nl_mmap_hdr *hdr;
608 hdr = __netlink_lookup_frame(ring, ring->head);
609 if (hdr->nm_status == NL_MMAP_STATUS_UNUSED)
611 if (hdr->nm_status != NL_MMAP_STATUS_SKIP)
613 netlink_increment_head(ring);
614 } while (ring->head != head);
617 static bool netlink_has_valid_frame(struct netlink_ring *ring)
619 unsigned int head = ring->head, pos = head;
620 const struct nl_mmap_hdr *hdr;
623 hdr = __netlink_lookup_frame(ring, pos);
624 if (hdr->nm_status == NL_MMAP_STATUS_VALID)
626 pos = pos != 0 ? pos - 1 : ring->frame_max;
627 } while (pos != head);
632 static bool netlink_dump_space(struct netlink_sock *nlk)
634 struct netlink_ring *ring = &nlk->rx_ring;
635 struct nl_mmap_hdr *hdr;
638 hdr = netlink_current_frame(ring, NL_MMAP_STATUS_UNUSED);
642 n = ring->head + ring->frame_max / 2;
643 if (n > ring->frame_max)
644 n -= ring->frame_max;
646 hdr = __netlink_lookup_frame(ring, n);
648 return hdr->nm_status == NL_MMAP_STATUS_UNUSED;
651 static unsigned int netlink_poll(struct file *file, struct socket *sock,
654 struct sock *sk = sock->sk;
655 struct netlink_sock *nlk = nlk_sk(sk);
659 if (nlk->rx_ring.pg_vec != NULL) {
660 /* Memory mapped sockets don't call recvmsg(), so flow control
661 * for dumps is performed here. A dump is allowed to continue
662 * if at least half the ring is unused.
664 while (nlk->cb_running && netlink_dump_space(nlk)) {
665 err = netlink_dump(sk);
668 sk->sk_error_report(sk);
672 netlink_rcv_wake(sk);
675 mask = datagram_poll(file, sock, wait);
677 spin_lock_bh(&sk->sk_receive_queue.lock);
678 if (nlk->rx_ring.pg_vec) {
679 if (netlink_has_valid_frame(&nlk->rx_ring))
680 mask |= POLLIN | POLLRDNORM;
682 spin_unlock_bh(&sk->sk_receive_queue.lock);
684 spin_lock_bh(&sk->sk_write_queue.lock);
685 if (nlk->tx_ring.pg_vec) {
686 if (netlink_current_frame(&nlk->tx_ring, NL_MMAP_STATUS_UNUSED))
687 mask |= POLLOUT | POLLWRNORM;
689 spin_unlock_bh(&sk->sk_write_queue.lock);
694 static struct nl_mmap_hdr *netlink_mmap_hdr(struct sk_buff *skb)
696 return (struct nl_mmap_hdr *)(skb->head - NL_MMAP_HDRLEN);
699 static void netlink_ring_setup_skb(struct sk_buff *skb, struct sock *sk,
700 struct netlink_ring *ring,
701 struct nl_mmap_hdr *hdr)
706 size = ring->frame_size - NL_MMAP_HDRLEN;
707 data = (void *)hdr + NL_MMAP_HDRLEN;
711 skb_reset_tail_pointer(skb);
712 skb->end = skb->tail + size;
715 skb->destructor = netlink_skb_destructor;
716 NETLINK_CB(skb).flags |= NETLINK_SKB_MMAPED;
717 NETLINK_CB(skb).sk = sk;
720 static int netlink_mmap_sendmsg(struct sock *sk, struct msghdr *msg,
721 u32 dst_portid, u32 dst_group,
722 struct scm_cookie *scm)
724 struct netlink_sock *nlk = nlk_sk(sk);
725 struct netlink_ring *ring;
726 struct nl_mmap_hdr *hdr;
729 int err = 0, len = 0;
731 mutex_lock(&nlk->pg_vec_lock);
733 ring = &nlk->tx_ring;
734 maxlen = ring->frame_size - NL_MMAP_HDRLEN;
739 hdr = netlink_current_frame(ring, NL_MMAP_STATUS_VALID);
741 if (!(msg->msg_flags & MSG_DONTWAIT) &&
742 atomic_read(&nlk->tx_ring.pending))
747 nm_len = ACCESS_ONCE(hdr->nm_len);
748 if (nm_len > maxlen) {
753 netlink_frame_flush_dcache(hdr, nm_len);
755 skb = alloc_skb(nm_len, GFP_KERNEL);
760 __skb_put(skb, nm_len);
761 memcpy(skb->data, (void *)hdr + NL_MMAP_HDRLEN, nm_len);
762 netlink_set_status(hdr, NL_MMAP_STATUS_UNUSED);
764 netlink_increment_head(ring);
766 NETLINK_CB(skb).portid = nlk->portid;
767 NETLINK_CB(skb).dst_group = dst_group;
768 NETLINK_CB(skb).creds = scm->creds;
770 err = security_netlink_send(sk, skb);
776 if (unlikely(dst_group)) {
777 atomic_inc(&skb->users);
778 netlink_broadcast(sk, skb, dst_portid, dst_group,
781 err = netlink_unicast(sk, skb, dst_portid,
782 msg->msg_flags & MSG_DONTWAIT);
787 } while (hdr != NULL ||
788 (!(msg->msg_flags & MSG_DONTWAIT) &&
789 atomic_read(&nlk->tx_ring.pending)));
794 mutex_unlock(&nlk->pg_vec_lock);
798 static void netlink_queue_mmaped_skb(struct sock *sk, struct sk_buff *skb)
800 struct nl_mmap_hdr *hdr;
802 hdr = netlink_mmap_hdr(skb);
803 hdr->nm_len = skb->len;
804 hdr->nm_group = NETLINK_CB(skb).dst_group;
805 hdr->nm_pid = NETLINK_CB(skb).creds.pid;
806 hdr->nm_uid = from_kuid(sk_user_ns(sk), NETLINK_CB(skb).creds.uid);
807 hdr->nm_gid = from_kgid(sk_user_ns(sk), NETLINK_CB(skb).creds.gid);
808 netlink_frame_flush_dcache(hdr, hdr->nm_len);
809 netlink_set_status(hdr, NL_MMAP_STATUS_VALID);
811 NETLINK_CB(skb).flags |= NETLINK_SKB_DELIVERED;
815 static void netlink_ring_set_copied(struct sock *sk, struct sk_buff *skb)
817 struct netlink_sock *nlk = nlk_sk(sk);
818 struct netlink_ring *ring = &nlk->rx_ring;
819 struct nl_mmap_hdr *hdr;
821 spin_lock_bh(&sk->sk_receive_queue.lock);
822 hdr = netlink_current_frame(ring, NL_MMAP_STATUS_UNUSED);
824 spin_unlock_bh(&sk->sk_receive_queue.lock);
829 netlink_increment_head(ring);
830 __skb_queue_tail(&sk->sk_receive_queue, skb);
831 spin_unlock_bh(&sk->sk_receive_queue.lock);
833 hdr->nm_len = skb->len;
834 hdr->nm_group = NETLINK_CB(skb).dst_group;
835 hdr->nm_pid = NETLINK_CB(skb).creds.pid;
836 hdr->nm_uid = from_kuid(sk_user_ns(sk), NETLINK_CB(skb).creds.uid);
837 hdr->nm_gid = from_kgid(sk_user_ns(sk), NETLINK_CB(skb).creds.gid);
838 netlink_set_status(hdr, NL_MMAP_STATUS_COPY);
841 #else /* CONFIG_NETLINK_MMAP */
842 #define netlink_skb_is_mmaped(skb) false
843 #define netlink_rx_is_mmaped(sk) false
844 #define netlink_tx_is_mmaped(sk) false
845 #define netlink_mmap sock_no_mmap
846 #define netlink_poll datagram_poll
847 #define netlink_mmap_sendmsg(sk, msg, dst_portid, dst_group, scm) 0
848 #endif /* CONFIG_NETLINK_MMAP */
850 static void netlink_skb_destructor(struct sk_buff *skb)
852 #ifdef CONFIG_NETLINK_MMAP
853 struct nl_mmap_hdr *hdr;
854 struct netlink_ring *ring;
857 /* If a packet from the kernel to userspace was freed because of an
858 * error without being delivered to userspace, the kernel must reset
859 * the status. In the direction userspace to kernel, the status is
860 * always reset here after the packet was processed and freed.
862 if (netlink_skb_is_mmaped(skb)) {
863 hdr = netlink_mmap_hdr(skb);
864 sk = NETLINK_CB(skb).sk;
866 if (NETLINK_CB(skb).flags & NETLINK_SKB_TX) {
867 netlink_set_status(hdr, NL_MMAP_STATUS_UNUSED);
868 ring = &nlk_sk(sk)->tx_ring;
870 if (!(NETLINK_CB(skb).flags & NETLINK_SKB_DELIVERED)) {
872 netlink_set_status(hdr, NL_MMAP_STATUS_VALID);
874 ring = &nlk_sk(sk)->rx_ring;
877 WARN_ON(atomic_read(&ring->pending) == 0);
878 atomic_dec(&ring->pending);
884 if (is_vmalloc_addr(skb->head)) {
886 !atomic_dec_return(&(skb_shinfo(skb)->dataref)))
895 static void netlink_skb_set_owner_r(struct sk_buff *skb, struct sock *sk)
897 WARN_ON(skb->sk != NULL);
899 skb->destructor = netlink_skb_destructor;
900 atomic_add(skb->truesize, &sk->sk_rmem_alloc);
901 sk_mem_charge(sk, skb->truesize);
904 static void netlink_sock_destruct(struct sock *sk)
906 struct netlink_sock *nlk = nlk_sk(sk);
908 if (nlk->cb_running) {
910 nlk->cb.done(&nlk->cb);
912 module_put(nlk->cb.module);
913 kfree_skb(nlk->cb.skb);
916 skb_queue_purge(&sk->sk_receive_queue);
917 #ifdef CONFIG_NETLINK_MMAP
919 struct nl_mmap_req req;
921 memset(&req, 0, sizeof(req));
922 if (nlk->rx_ring.pg_vec)
923 __netlink_set_ring(sk, &req, false, NULL, 0);
924 memset(&req, 0, sizeof(req));
925 if (nlk->tx_ring.pg_vec)
926 __netlink_set_ring(sk, &req, true, NULL, 0);
928 #endif /* CONFIG_NETLINK_MMAP */
930 if (!sock_flag(sk, SOCK_DEAD)) {
931 printk(KERN_ERR "Freeing alive netlink socket %p\n", sk);
935 WARN_ON(atomic_read(&sk->sk_rmem_alloc));
936 WARN_ON(atomic_read(&sk->sk_wmem_alloc));
937 WARN_ON(nlk_sk(sk)->groups);
940 /* This lock without WQ_FLAG_EXCLUSIVE is good on UP and it is _very_ bad on
941 * SMP. Look, when several writers sleep and reader wakes them up, all but one
942 * immediately hit write lock and grab all the cpus. Exclusive sleep solves
943 * this, _but_ remember, it adds useless work on UP machines.
946 void netlink_table_grab(void)
947 __acquires(nl_table_lock)
951 write_lock_irq(&nl_table_lock);
953 if (atomic_read(&nl_table_users)) {
954 DECLARE_WAITQUEUE(wait, current);
956 add_wait_queue_exclusive(&nl_table_wait, &wait);
958 set_current_state(TASK_UNINTERRUPTIBLE);
959 if (atomic_read(&nl_table_users) == 0)
961 write_unlock_irq(&nl_table_lock);
963 write_lock_irq(&nl_table_lock);
966 __set_current_state(TASK_RUNNING);
967 remove_wait_queue(&nl_table_wait, &wait);
971 void netlink_table_ungrab(void)
972 __releases(nl_table_lock)
974 write_unlock_irq(&nl_table_lock);
975 wake_up(&nl_table_wait);
979 netlink_lock_table(void)
981 /* read_lock() synchronizes us to netlink_table_grab */
983 read_lock(&nl_table_lock);
984 atomic_inc(&nl_table_users);
985 read_unlock(&nl_table_lock);
989 netlink_unlock_table(void)
991 if (atomic_dec_and_test(&nl_table_users))
992 wake_up(&nl_table_wait);
995 struct netlink_compare_arg
1001 /* Doing sizeof directly may yield 4 extra bytes on 64-bit. */
1002 #define netlink_compare_arg_len \
1003 (offsetof(struct netlink_compare_arg, portid) + sizeof(u32))
1005 static inline int netlink_compare(struct rhashtable_compare_arg *arg,
1008 const struct netlink_compare_arg *x = arg->key;
1009 const struct netlink_sock *nlk = ptr;
1011 return nlk->portid != x->portid ||
1012 !net_eq(sock_net(&nlk->sk), read_pnet(&x->pnet));
1015 static void netlink_compare_arg_init(struct netlink_compare_arg *arg,
1016 struct net *net, u32 portid)
1018 memset(arg, 0, sizeof(*arg));
1019 write_pnet(&arg->pnet, net);
1020 arg->portid = portid;
1023 static struct sock *__netlink_lookup(struct netlink_table *table, u32 portid,
1026 struct netlink_compare_arg arg;
1028 netlink_compare_arg_init(&arg, net, portid);
1029 return rhashtable_lookup_fast(&table->hash, &arg,
1030 netlink_rhashtable_params);
1033 static int __netlink_insert(struct netlink_table *table, struct sock *sk)
1035 struct netlink_compare_arg arg;
1037 netlink_compare_arg_init(&arg, sock_net(sk), nlk_sk(sk)->portid);
1038 return rhashtable_lookup_insert_key(&table->hash, &arg,
1040 netlink_rhashtable_params);
1043 static struct sock *netlink_lookup(struct net *net, int protocol, u32 portid)
1045 struct netlink_table *table = &nl_table[protocol];
1049 sk = __netlink_lookup(table, portid, net);
1057 static const struct proto_ops netlink_ops;
1060 netlink_update_listeners(struct sock *sk)
1062 struct netlink_table *tbl = &nl_table[sk->sk_protocol];
1065 struct listeners *listeners;
1067 listeners = nl_deref_protected(tbl->listeners);
1071 for (i = 0; i < NLGRPLONGS(tbl->groups); i++) {
1073 sk_for_each_bound(sk, &tbl->mc_list) {
1074 if (i < NLGRPLONGS(nlk_sk(sk)->ngroups))
1075 mask |= nlk_sk(sk)->groups[i];
1077 listeners->masks[i] = mask;
1079 /* this function is only called with the netlink table "grabbed", which
1080 * makes sure updates are visible before bind or setsockopt return. */
1083 static int netlink_insert(struct sock *sk, u32 portid)
1085 struct netlink_table *table = &nl_table[sk->sk_protocol];
1091 if (nlk_sk(sk)->portid)
1095 if (BITS_PER_LONG > 32 &&
1096 unlikely(atomic_read(&table->hash.nelems) >= UINT_MAX))
1099 nlk_sk(sk)->portid = portid;
1102 err = __netlink_insert(table, sk);
1104 /* In case the hashtable backend returns with -EBUSY
1105 * from here, it must not escape to the caller.
1107 if (unlikely(err == -EBUSY))
1111 nlk_sk(sk)->portid = 0;
1120 static void netlink_remove(struct sock *sk)
1122 struct netlink_table *table;
1124 table = &nl_table[sk->sk_protocol];
1125 if (!rhashtable_remove_fast(&table->hash, &nlk_sk(sk)->node,
1126 netlink_rhashtable_params)) {
1127 WARN_ON(atomic_read(&sk->sk_refcnt) == 1);
1131 netlink_table_grab();
1132 if (nlk_sk(sk)->subscriptions) {
1133 __sk_del_bind_node(sk);
1134 netlink_update_listeners(sk);
1136 if (sk->sk_protocol == NETLINK_GENERIC)
1137 atomic_inc(&genl_sk_destructing_cnt);
1138 netlink_table_ungrab();
1141 static struct proto netlink_proto = {
1143 .owner = THIS_MODULE,
1144 .obj_size = sizeof(struct netlink_sock),
1147 static int __netlink_create(struct net *net, struct socket *sock,
1148 struct mutex *cb_mutex, int protocol,
1152 struct netlink_sock *nlk;
1154 sock->ops = &netlink_ops;
1156 sk = sk_alloc(net, PF_NETLINK, GFP_KERNEL, &netlink_proto, kern);
1160 sock_init_data(sock, sk);
1164 nlk->cb_mutex = cb_mutex;
1166 nlk->cb_mutex = &nlk->cb_def_mutex;
1167 mutex_init(nlk->cb_mutex);
1169 init_waitqueue_head(&nlk->wait);
1170 #ifdef CONFIG_NETLINK_MMAP
1171 mutex_init(&nlk->pg_vec_lock);
1174 sk->sk_destruct = netlink_sock_destruct;
1175 sk->sk_protocol = protocol;
1179 static int netlink_create(struct net *net, struct socket *sock, int protocol,
1182 struct module *module = NULL;
1183 struct mutex *cb_mutex;
1184 struct netlink_sock *nlk;
1185 int (*bind)(struct net *net, int group);
1186 void (*unbind)(struct net *net, int group);
1189 sock->state = SS_UNCONNECTED;
1191 if (sock->type != SOCK_RAW && sock->type != SOCK_DGRAM)
1192 return -ESOCKTNOSUPPORT;
1194 if (protocol < 0 || protocol >= MAX_LINKS)
1195 return -EPROTONOSUPPORT;
1197 netlink_lock_table();
1198 #ifdef CONFIG_MODULES
1199 if (!nl_table[protocol].registered) {
1200 netlink_unlock_table();
1201 request_module("net-pf-%d-proto-%d", PF_NETLINK, protocol);
1202 netlink_lock_table();
1205 if (nl_table[protocol].registered &&
1206 try_module_get(nl_table[protocol].module))
1207 module = nl_table[protocol].module;
1209 err = -EPROTONOSUPPORT;
1210 cb_mutex = nl_table[protocol].cb_mutex;
1211 bind = nl_table[protocol].bind;
1212 unbind = nl_table[protocol].unbind;
1213 netlink_unlock_table();
1218 err = __netlink_create(net, sock, cb_mutex, protocol, kern);
1223 sock_prot_inuse_add(net, &netlink_proto, 1);
1226 nlk = nlk_sk(sock->sk);
1227 nlk->module = module;
1228 nlk->netlink_bind = bind;
1229 nlk->netlink_unbind = unbind;
1238 static void deferred_put_nlk_sk(struct rcu_head *head)
1240 struct netlink_sock *nlk = container_of(head, struct netlink_sock, rcu);
1245 static int netlink_release(struct socket *sock)
1247 struct sock *sk = sock->sk;
1248 struct netlink_sock *nlk;
1258 * OK. Socket is unlinked, any packets that arrive now
1262 /* must not acquire netlink_table_lock in any way again before unbind
1263 * and notifying genetlink is done as otherwise it might deadlock
1265 if (nlk->netlink_unbind) {
1268 for (i = 0; i < nlk->ngroups; i++)
1269 if (test_bit(i, nlk->groups))
1270 nlk->netlink_unbind(sock_net(sk), i + 1);
1272 if (sk->sk_protocol == NETLINK_GENERIC &&
1273 atomic_dec_return(&genl_sk_destructing_cnt) == 0)
1274 wake_up(&genl_sk_destructing_waitq);
1277 wake_up_interruptible_all(&nlk->wait);
1279 skb_queue_purge(&sk->sk_write_queue);
1282 struct netlink_notify n = {
1283 .net = sock_net(sk),
1284 .protocol = sk->sk_protocol,
1285 .portid = nlk->portid,
1287 atomic_notifier_call_chain(&netlink_chain,
1288 NETLINK_URELEASE, &n);
1291 module_put(nlk->module);
1293 if (netlink_is_kernel(sk)) {
1294 netlink_table_grab();
1295 BUG_ON(nl_table[sk->sk_protocol].registered == 0);
1296 if (--nl_table[sk->sk_protocol].registered == 0) {
1297 struct listeners *old;
1299 old = nl_deref_protected(nl_table[sk->sk_protocol].listeners);
1300 RCU_INIT_POINTER(nl_table[sk->sk_protocol].listeners, NULL);
1301 kfree_rcu(old, rcu);
1302 nl_table[sk->sk_protocol].module = NULL;
1303 nl_table[sk->sk_protocol].bind = NULL;
1304 nl_table[sk->sk_protocol].unbind = NULL;
1305 nl_table[sk->sk_protocol].flags = 0;
1306 nl_table[sk->sk_protocol].registered = 0;
1308 netlink_table_ungrab();
1315 sock_prot_inuse_add(sock_net(sk), &netlink_proto, -1);
1317 call_rcu(&nlk->rcu, deferred_put_nlk_sk);
1321 static int netlink_autobind(struct socket *sock)
1323 struct sock *sk = sock->sk;
1324 struct net *net = sock_net(sk);
1325 struct netlink_table *table = &nl_table[sk->sk_protocol];
1326 s32 portid = task_tgid_vnr(current);
1334 ok = !__netlink_lookup(table, portid, net);
1337 /* Bind collision, search negative portid values. */
1339 /* rover will be in range [S32_MIN, -4097] */
1340 rover = S32_MIN + prandom_u32_max(-4096 - S32_MIN);
1341 else if (rover >= -4096)
1347 err = netlink_insert(sk, portid);
1348 if (err == -EADDRINUSE)
1351 /* If 2 threads race to autobind, that is fine. */
1359 * __netlink_ns_capable - General netlink message capability test
1360 * @nsp: NETLINK_CB of the socket buffer holding a netlink command from userspace.
1361 * @user_ns: The user namespace of the capability to use
1362 * @cap: The capability to use
1364 * Test to see if the opener of the socket we received the message
1365 * from had when the netlink socket was created and the sender of the
1366 * message has has the capability @cap in the user namespace @user_ns.
1368 bool __netlink_ns_capable(const struct netlink_skb_parms *nsp,
1369 struct user_namespace *user_ns, int cap)
1371 return ((nsp->flags & NETLINK_SKB_DST) ||
1372 file_ns_capable(nsp->sk->sk_socket->file, user_ns, cap)) &&
1373 ns_capable(user_ns, cap);
1375 EXPORT_SYMBOL(__netlink_ns_capable);
1378 * netlink_ns_capable - General netlink message capability test
1379 * @skb: socket buffer holding a netlink command from userspace
1380 * @user_ns: The user namespace of the capability to use
1381 * @cap: The capability to use
1383 * Test to see if the opener of the socket we received the message
1384 * from had when the netlink socket was created and the sender of the
1385 * message has has the capability @cap in the user namespace @user_ns.
1387 bool netlink_ns_capable(const struct sk_buff *skb,
1388 struct user_namespace *user_ns, int cap)
1390 return __netlink_ns_capable(&NETLINK_CB(skb), user_ns, cap);
1392 EXPORT_SYMBOL(netlink_ns_capable);
1395 * netlink_capable - Netlink global message capability test
1396 * @skb: socket buffer holding a netlink command from userspace
1397 * @cap: The capability to use
1399 * Test to see if the opener of the socket we received the message
1400 * from had when the netlink socket was created and the sender of the
1401 * message has has the capability @cap in all user namespaces.
1403 bool netlink_capable(const struct sk_buff *skb, int cap)
1405 return netlink_ns_capable(skb, &init_user_ns, cap);
1407 EXPORT_SYMBOL(netlink_capable);
1410 * netlink_net_capable - Netlink network namespace message capability test
1411 * @skb: socket buffer holding a netlink command from userspace
1412 * @cap: The capability to use
1414 * Test to see if the opener of the socket we received the message
1415 * from had when the netlink socket was created and the sender of the
1416 * message has has the capability @cap over the network namespace of
1417 * the socket we received the message from.
1419 bool netlink_net_capable(const struct sk_buff *skb, int cap)
1421 return netlink_ns_capable(skb, sock_net(skb->sk)->user_ns, cap);
1423 EXPORT_SYMBOL(netlink_net_capable);
1425 static inline int netlink_allowed(const struct socket *sock, unsigned int flag)
1427 return (nl_table[sock->sk->sk_protocol].flags & flag) ||
1428 ns_capable(sock_net(sock->sk)->user_ns, CAP_NET_ADMIN);
1432 netlink_update_subscriptions(struct sock *sk, unsigned int subscriptions)
1434 struct netlink_sock *nlk = nlk_sk(sk);
1436 if (nlk->subscriptions && !subscriptions)
1437 __sk_del_bind_node(sk);
1438 else if (!nlk->subscriptions && subscriptions)
1439 sk_add_bind_node(sk, &nl_table[sk->sk_protocol].mc_list);
1440 nlk->subscriptions = subscriptions;
1443 static int netlink_realloc_groups(struct sock *sk)
1445 struct netlink_sock *nlk = nlk_sk(sk);
1446 unsigned int groups;
1447 unsigned long *new_groups;
1450 netlink_table_grab();
1452 groups = nl_table[sk->sk_protocol].groups;
1453 if (!nl_table[sk->sk_protocol].registered) {
1458 if (nlk->ngroups >= groups)
1461 new_groups = krealloc(nlk->groups, NLGRPSZ(groups), GFP_ATOMIC);
1462 if (new_groups == NULL) {
1466 memset((char *)new_groups + NLGRPSZ(nlk->ngroups), 0,
1467 NLGRPSZ(groups) - NLGRPSZ(nlk->ngroups));
1469 nlk->groups = new_groups;
1470 nlk->ngroups = groups;
1472 netlink_table_ungrab();
1476 static void netlink_undo_bind(int group, long unsigned int groups,
1479 struct netlink_sock *nlk = nlk_sk(sk);
1482 if (!nlk->netlink_unbind)
1485 for (undo = 0; undo < group; undo++)
1486 if (test_bit(undo, &groups))
1487 nlk->netlink_unbind(sock_net(sk), undo + 1);
1490 static int netlink_bind(struct socket *sock, struct sockaddr *addr,
1493 struct sock *sk = sock->sk;
1494 struct net *net = sock_net(sk);
1495 struct netlink_sock *nlk = nlk_sk(sk);
1496 struct sockaddr_nl *nladdr = (struct sockaddr_nl *)addr;
1498 long unsigned int groups = nladdr->nl_groups;
1500 if (addr_len < sizeof(struct sockaddr_nl))
1503 if (nladdr->nl_family != AF_NETLINK)
1506 /* Only superuser is allowed to listen multicasts */
1508 if (!netlink_allowed(sock, NL_CFG_F_NONROOT_RECV))
1510 err = netlink_realloc_groups(sk);
1516 if (nladdr->nl_pid != nlk->portid)
1519 if (nlk->netlink_bind && groups) {
1522 for (group = 0; group < nlk->ngroups; group++) {
1523 if (!test_bit(group, &groups))
1525 err = nlk->netlink_bind(net, group + 1);
1528 netlink_undo_bind(group, groups, sk);
1534 err = nladdr->nl_pid ?
1535 netlink_insert(sk, nladdr->nl_pid) :
1536 netlink_autobind(sock);
1538 netlink_undo_bind(nlk->ngroups, groups, sk);
1543 if (!groups && (nlk->groups == NULL || !(u32)nlk->groups[0]))
1546 netlink_table_grab();
1547 netlink_update_subscriptions(sk, nlk->subscriptions +
1549 hweight32(nlk->groups[0]));
1550 nlk->groups[0] = (nlk->groups[0] & ~0xffffffffUL) | groups;
1551 netlink_update_listeners(sk);
1552 netlink_table_ungrab();
1557 static int netlink_connect(struct socket *sock, struct sockaddr *addr,
1558 int alen, int flags)
1561 struct sock *sk = sock->sk;
1562 struct netlink_sock *nlk = nlk_sk(sk);
1563 struct sockaddr_nl *nladdr = (struct sockaddr_nl *)addr;
1565 if (alen < sizeof(addr->sa_family))
1568 if (addr->sa_family == AF_UNSPEC) {
1569 sk->sk_state = NETLINK_UNCONNECTED;
1570 nlk->dst_portid = 0;
1574 if (addr->sa_family != AF_NETLINK)
1577 if ((nladdr->nl_groups || nladdr->nl_pid) &&
1578 !netlink_allowed(sock, NL_CFG_F_NONROOT_SEND))
1582 err = netlink_autobind(sock);
1585 sk->sk_state = NETLINK_CONNECTED;
1586 nlk->dst_portid = nladdr->nl_pid;
1587 nlk->dst_group = ffs(nladdr->nl_groups);
1593 static int netlink_getname(struct socket *sock, struct sockaddr *addr,
1594 int *addr_len, int peer)
1596 struct sock *sk = sock->sk;
1597 struct netlink_sock *nlk = nlk_sk(sk);
1598 DECLARE_SOCKADDR(struct sockaddr_nl *, nladdr, addr);
1600 nladdr->nl_family = AF_NETLINK;
1602 *addr_len = sizeof(*nladdr);
1605 nladdr->nl_pid = nlk->dst_portid;
1606 nladdr->nl_groups = netlink_group_mask(nlk->dst_group);
1608 nladdr->nl_pid = nlk->portid;
1609 nladdr->nl_groups = nlk->groups ? nlk->groups[0] : 0;
1614 static struct sock *netlink_getsockbyportid(struct sock *ssk, u32 portid)
1617 struct netlink_sock *nlk;
1619 sock = netlink_lookup(sock_net(ssk), ssk->sk_protocol, portid);
1621 return ERR_PTR(-ECONNREFUSED);
1623 /* Don't bother queuing skb if kernel socket has no input function */
1625 if (sock->sk_state == NETLINK_CONNECTED &&
1626 nlk->dst_portid != nlk_sk(ssk)->portid) {
1628 return ERR_PTR(-ECONNREFUSED);
1633 struct sock *netlink_getsockbyfilp(struct file *filp)
1635 struct inode *inode = file_inode(filp);
1638 if (!S_ISSOCK(inode->i_mode))
1639 return ERR_PTR(-ENOTSOCK);
1641 sock = SOCKET_I(inode)->sk;
1642 if (sock->sk_family != AF_NETLINK)
1643 return ERR_PTR(-EINVAL);
1649 static struct sk_buff *netlink_alloc_large_skb(unsigned int size,
1652 struct sk_buff *skb;
1655 if (size <= NLMSG_GOODSIZE || broadcast)
1656 return alloc_skb(size, GFP_KERNEL);
1658 size = SKB_DATA_ALIGN(size) +
1659 SKB_DATA_ALIGN(sizeof(struct skb_shared_info));
1661 data = vmalloc(size);
1665 skb = __build_skb(data, size);
1669 skb->destructor = netlink_skb_destructor;
1675 * Attach a skb to a netlink socket.
1676 * The caller must hold a reference to the destination socket. On error, the
1677 * reference is dropped. The skb is not send to the destination, just all
1678 * all error checks are performed and memory in the queue is reserved.
1680 * < 0: error. skb freed, reference to sock dropped.
1682 * 1: repeat lookup - reference dropped while waiting for socket memory.
1684 int netlink_attachskb(struct sock *sk, struct sk_buff *skb,
1685 long *timeo, struct sock *ssk)
1687 struct netlink_sock *nlk;
1691 if ((atomic_read(&sk->sk_rmem_alloc) > sk->sk_rcvbuf ||
1692 test_bit(NETLINK_S_CONGESTED, &nlk->state)) &&
1693 !netlink_skb_is_mmaped(skb)) {
1694 DECLARE_WAITQUEUE(wait, current);
1696 if (!ssk || netlink_is_kernel(ssk))
1697 netlink_overrun(sk);
1703 __set_current_state(TASK_INTERRUPTIBLE);
1704 add_wait_queue(&nlk->wait, &wait);
1706 if ((atomic_read(&sk->sk_rmem_alloc) > sk->sk_rcvbuf ||
1707 test_bit(NETLINK_S_CONGESTED, &nlk->state)) &&
1708 !sock_flag(sk, SOCK_DEAD))
1709 *timeo = schedule_timeout(*timeo);
1711 __set_current_state(TASK_RUNNING);
1712 remove_wait_queue(&nlk->wait, &wait);
1715 if (signal_pending(current)) {
1717 return sock_intr_errno(*timeo);
1721 netlink_skb_set_owner_r(skb, sk);
1725 static int __netlink_sendskb(struct sock *sk, struct sk_buff *skb)
1729 netlink_deliver_tap(skb);
1731 #ifdef CONFIG_NETLINK_MMAP
1732 if (netlink_skb_is_mmaped(skb))
1733 netlink_queue_mmaped_skb(sk, skb);
1734 else if (netlink_rx_is_mmaped(sk))
1735 netlink_ring_set_copied(sk, skb);
1737 #endif /* CONFIG_NETLINK_MMAP */
1738 skb_queue_tail(&sk->sk_receive_queue, skb);
1739 sk->sk_data_ready(sk);
1743 int netlink_sendskb(struct sock *sk, struct sk_buff *skb)
1745 int len = __netlink_sendskb(sk, skb);
1751 void netlink_detachskb(struct sock *sk, struct sk_buff *skb)
1757 static struct sk_buff *netlink_trim(struct sk_buff *skb, gfp_t allocation)
1761 WARN_ON(skb->sk != NULL);
1762 if (netlink_skb_is_mmaped(skb))
1765 delta = skb->end - skb->tail;
1766 if (is_vmalloc_addr(skb->head) || delta * 2 < skb->truesize)
1769 if (skb_shared(skb)) {
1770 struct sk_buff *nskb = skb_clone(skb, allocation);
1777 if (!pskb_expand_head(skb, 0, -delta, allocation))
1778 skb->truesize -= delta;
1783 static int netlink_unicast_kernel(struct sock *sk, struct sk_buff *skb,
1787 struct netlink_sock *nlk = nlk_sk(sk);
1789 ret = -ECONNREFUSED;
1790 if (nlk->netlink_rcv != NULL) {
1792 netlink_skb_set_owner_r(skb, sk);
1793 NETLINK_CB(skb).sk = ssk;
1794 netlink_deliver_tap_kernel(sk, ssk, skb);
1795 nlk->netlink_rcv(skb);
1804 int netlink_unicast(struct sock *ssk, struct sk_buff *skb,
1805 u32 portid, int nonblock)
1811 skb = netlink_trim(skb, gfp_any());
1813 timeo = sock_sndtimeo(ssk, nonblock);
1815 sk = netlink_getsockbyportid(ssk, portid);
1820 if (netlink_is_kernel(sk))
1821 return netlink_unicast_kernel(sk, skb, ssk);
1823 if (sk_filter(sk, skb)) {
1830 err = netlink_attachskb(sk, skb, &timeo, ssk);
1836 return netlink_sendskb(sk, skb);
1838 EXPORT_SYMBOL(netlink_unicast);
1840 struct sk_buff *netlink_alloc_skb(struct sock *ssk, unsigned int size,
1841 u32 dst_portid, gfp_t gfp_mask)
1843 #ifdef CONFIG_NETLINK_MMAP
1844 struct sock *sk = NULL;
1845 struct sk_buff *skb;
1846 struct netlink_ring *ring;
1847 struct nl_mmap_hdr *hdr;
1848 unsigned int maxlen;
1850 sk = netlink_getsockbyportid(ssk, dst_portid);
1854 ring = &nlk_sk(sk)->rx_ring;
1855 /* fast-path without atomic ops for common case: non-mmaped receiver */
1856 if (ring->pg_vec == NULL)
1859 if (ring->frame_size - NL_MMAP_HDRLEN < size)
1862 skb = alloc_skb_head(gfp_mask);
1866 spin_lock_bh(&sk->sk_receive_queue.lock);
1867 /* check again under lock */
1868 if (ring->pg_vec == NULL)
1871 /* check again under lock */
1872 maxlen = ring->frame_size - NL_MMAP_HDRLEN;
1876 netlink_forward_ring(ring);
1877 hdr = netlink_current_frame(ring, NL_MMAP_STATUS_UNUSED);
1880 netlink_ring_setup_skb(skb, sk, ring, hdr);
1881 netlink_set_status(hdr, NL_MMAP_STATUS_RESERVED);
1882 atomic_inc(&ring->pending);
1883 netlink_increment_head(ring);
1885 spin_unlock_bh(&sk->sk_receive_queue.lock);
1890 spin_unlock_bh(&sk->sk_receive_queue.lock);
1891 netlink_overrun(sk);
1898 spin_unlock_bh(&sk->sk_receive_queue.lock);
1903 return alloc_skb(size, gfp_mask);
1905 EXPORT_SYMBOL_GPL(netlink_alloc_skb);
1907 int netlink_has_listeners(struct sock *sk, unsigned int group)
1910 struct listeners *listeners;
1912 BUG_ON(!netlink_is_kernel(sk));
1915 listeners = rcu_dereference(nl_table[sk->sk_protocol].listeners);
1917 if (listeners && group - 1 < nl_table[sk->sk_protocol].groups)
1918 res = test_bit(group - 1, listeners->masks);
1924 EXPORT_SYMBOL_GPL(netlink_has_listeners);
1926 static int netlink_broadcast_deliver(struct sock *sk, struct sk_buff *skb)
1928 struct netlink_sock *nlk = nlk_sk(sk);
1930 if (atomic_read(&sk->sk_rmem_alloc) <= sk->sk_rcvbuf &&
1931 !test_bit(NETLINK_S_CONGESTED, &nlk->state)) {
1932 netlink_skb_set_owner_r(skb, sk);
1933 __netlink_sendskb(sk, skb);
1934 return atomic_read(&sk->sk_rmem_alloc) > (sk->sk_rcvbuf >> 1);
1939 struct netlink_broadcast_data {
1940 struct sock *exclude_sk;
1945 int delivery_failure;
1949 struct sk_buff *skb, *skb2;
1950 int (*tx_filter)(struct sock *dsk, struct sk_buff *skb, void *data);
1954 static void do_one_broadcast(struct sock *sk,
1955 struct netlink_broadcast_data *p)
1957 struct netlink_sock *nlk = nlk_sk(sk);
1960 if (p->exclude_sk == sk)
1963 if (nlk->portid == p->portid || p->group - 1 >= nlk->ngroups ||
1964 !test_bit(p->group - 1, nlk->groups))
1967 if (!net_eq(sock_net(sk), p->net)) {
1968 if (!(nlk->flags & NETLINK_F_LISTEN_ALL_NSID))
1971 if (!peernet_has_id(sock_net(sk), p->net))
1974 if (!file_ns_capable(sk->sk_socket->file, p->net->user_ns,
1980 netlink_overrun(sk);
1985 if (p->skb2 == NULL) {
1986 if (skb_shared(p->skb)) {
1987 p->skb2 = skb_clone(p->skb, p->allocation);
1989 p->skb2 = skb_get(p->skb);
1991 * skb ownership may have been set when
1992 * delivered to a previous socket.
1994 skb_orphan(p->skb2);
1997 if (p->skb2 == NULL) {
1998 netlink_overrun(sk);
1999 /* Clone failed. Notify ALL listeners. */
2001 if (nlk->flags & NETLINK_F_BROADCAST_SEND_ERROR)
2002 p->delivery_failure = 1;
2005 if (p->tx_filter && p->tx_filter(sk, p->skb2, p->tx_data)) {
2010 if (sk_filter(sk, p->skb2)) {
2015 NETLINK_CB(p->skb2).nsid = peernet2id(sock_net(sk), p->net);
2016 NETLINK_CB(p->skb2).nsid_is_set = true;
2017 val = netlink_broadcast_deliver(sk, p->skb2);
2019 netlink_overrun(sk);
2020 if (nlk->flags & NETLINK_F_BROADCAST_SEND_ERROR)
2021 p->delivery_failure = 1;
2023 p->congested |= val;
2031 int netlink_broadcast_filtered(struct sock *ssk, struct sk_buff *skb, u32 portid,
2032 u32 group, gfp_t allocation,
2033 int (*filter)(struct sock *dsk, struct sk_buff *skb, void *data),
2036 struct net *net = sock_net(ssk);
2037 struct netlink_broadcast_data info;
2040 skb = netlink_trim(skb, allocation);
2042 info.exclude_sk = ssk;
2044 info.portid = portid;
2047 info.delivery_failure = 0;
2050 info.allocation = allocation;
2053 info.tx_filter = filter;
2054 info.tx_data = filter_data;
2056 /* While we sleep in clone, do not allow to change socket list */
2058 netlink_lock_table();
2060 sk_for_each_bound(sk, &nl_table[ssk->sk_protocol].mc_list)
2061 do_one_broadcast(sk, &info);
2065 netlink_unlock_table();
2067 if (info.delivery_failure) {
2068 kfree_skb(info.skb2);
2071 consume_skb(info.skb2);
2073 if (info.delivered) {
2074 if (info.congested && (allocation & __GFP_WAIT))
2080 EXPORT_SYMBOL(netlink_broadcast_filtered);
2082 int netlink_broadcast(struct sock *ssk, struct sk_buff *skb, u32 portid,
2083 u32 group, gfp_t allocation)
2085 return netlink_broadcast_filtered(ssk, skb, portid, group, allocation,
2088 EXPORT_SYMBOL(netlink_broadcast);
2090 struct netlink_set_err_data {
2091 struct sock *exclude_sk;
2097 static int do_one_set_err(struct sock *sk, struct netlink_set_err_data *p)
2099 struct netlink_sock *nlk = nlk_sk(sk);
2102 if (sk == p->exclude_sk)
2105 if (!net_eq(sock_net(sk), sock_net(p->exclude_sk)))
2108 if (nlk->portid == p->portid || p->group - 1 >= nlk->ngroups ||
2109 !test_bit(p->group - 1, nlk->groups))
2112 if (p->code == ENOBUFS && nlk->flags & NETLINK_F_RECV_NO_ENOBUFS) {
2117 sk->sk_err = p->code;
2118 sk->sk_error_report(sk);
2124 * netlink_set_err - report error to broadcast listeners
2125 * @ssk: the kernel netlink socket, as returned by netlink_kernel_create()
2126 * @portid: the PORTID of a process that we want to skip (if any)
2127 * @group: the broadcast group that will notice the error
2128 * @code: error code, must be negative (as usual in kernelspace)
2130 * This function returns the number of broadcast listeners that have set the
2131 * NETLINK_NO_ENOBUFS socket option.
2133 int netlink_set_err(struct sock *ssk, u32 portid, u32 group, int code)
2135 struct netlink_set_err_data info;
2139 info.exclude_sk = ssk;
2140 info.portid = portid;
2142 /* sk->sk_err wants a positive error value */
2145 read_lock(&nl_table_lock);
2147 sk_for_each_bound(sk, &nl_table[ssk->sk_protocol].mc_list)
2148 ret += do_one_set_err(sk, &info);
2150 read_unlock(&nl_table_lock);
2153 EXPORT_SYMBOL(netlink_set_err);
2155 /* must be called with netlink table grabbed */
2156 static void netlink_update_socket_mc(struct netlink_sock *nlk,
2160 int old, new = !!is_new, subscriptions;
2162 old = test_bit(group - 1, nlk->groups);
2163 subscriptions = nlk->subscriptions - old + new;
2165 __set_bit(group - 1, nlk->groups);
2167 __clear_bit(group - 1, nlk->groups);
2168 netlink_update_subscriptions(&nlk->sk, subscriptions);
2169 netlink_update_listeners(&nlk->sk);
2172 static int netlink_setsockopt(struct socket *sock, int level, int optname,
2173 char __user *optval, unsigned int optlen)
2175 struct sock *sk = sock->sk;
2176 struct netlink_sock *nlk = nlk_sk(sk);
2177 unsigned int val = 0;
2180 if (level != SOL_NETLINK)
2181 return -ENOPROTOOPT;
2183 if (optname != NETLINK_RX_RING && optname != NETLINK_TX_RING &&
2184 optlen >= sizeof(int) &&
2185 get_user(val, (unsigned int __user *)optval))
2189 case NETLINK_PKTINFO:
2191 nlk->flags |= NETLINK_F_RECV_PKTINFO;
2193 nlk->flags &= ~NETLINK_F_RECV_PKTINFO;
2196 case NETLINK_ADD_MEMBERSHIP:
2197 case NETLINK_DROP_MEMBERSHIP: {
2198 if (!netlink_allowed(sock, NL_CFG_F_NONROOT_RECV))
2200 err = netlink_realloc_groups(sk);
2203 if (!val || val - 1 >= nlk->ngroups)
2205 if (optname == NETLINK_ADD_MEMBERSHIP && nlk->netlink_bind) {
2206 err = nlk->netlink_bind(sock_net(sk), val);
2210 netlink_table_grab();
2211 netlink_update_socket_mc(nlk, val,
2212 optname == NETLINK_ADD_MEMBERSHIP);
2213 netlink_table_ungrab();
2214 if (optname == NETLINK_DROP_MEMBERSHIP && nlk->netlink_unbind)
2215 nlk->netlink_unbind(sock_net(sk), val);
2220 case NETLINK_BROADCAST_ERROR:
2222 nlk->flags |= NETLINK_F_BROADCAST_SEND_ERROR;
2224 nlk->flags &= ~NETLINK_F_BROADCAST_SEND_ERROR;
2227 case NETLINK_NO_ENOBUFS:
2229 nlk->flags |= NETLINK_F_RECV_NO_ENOBUFS;
2230 clear_bit(NETLINK_S_CONGESTED, &nlk->state);
2231 wake_up_interruptible(&nlk->wait);
2233 nlk->flags &= ~NETLINK_F_RECV_NO_ENOBUFS;
2237 #ifdef CONFIG_NETLINK_MMAP
2238 case NETLINK_RX_RING:
2239 case NETLINK_TX_RING: {
2240 struct nl_mmap_req req;
2242 /* Rings might consume more memory than queue limits, require
2245 if (!capable(CAP_NET_ADMIN))
2247 if (optlen < sizeof(req))
2249 if (copy_from_user(&req, optval, sizeof(req)))
2251 err = netlink_set_ring(sk, &req,
2252 optname == NETLINK_TX_RING);
2255 #endif /* CONFIG_NETLINK_MMAP */
2256 case NETLINK_LISTEN_ALL_NSID:
2257 if (!ns_capable(sock_net(sk)->user_ns, CAP_NET_BROADCAST))
2261 nlk->flags |= NETLINK_F_LISTEN_ALL_NSID;
2263 nlk->flags &= ~NETLINK_F_LISTEN_ALL_NSID;
2266 case NETLINK_CAP_ACK:
2268 nlk->flags |= NETLINK_F_CAP_ACK;
2270 nlk->flags &= ~NETLINK_F_CAP_ACK;
2279 static int netlink_getsockopt(struct socket *sock, int level, int optname,
2280 char __user *optval, int __user *optlen)
2282 struct sock *sk = sock->sk;
2283 struct netlink_sock *nlk = nlk_sk(sk);
2286 if (level != SOL_NETLINK)
2287 return -ENOPROTOOPT;
2289 if (get_user(len, optlen))
2295 case NETLINK_PKTINFO:
2296 if (len < sizeof(int))
2299 val = nlk->flags & NETLINK_F_RECV_PKTINFO ? 1 : 0;
2300 if (put_user(len, optlen) ||
2301 put_user(val, optval))
2305 case NETLINK_BROADCAST_ERROR:
2306 if (len < sizeof(int))
2309 val = nlk->flags & NETLINK_F_BROADCAST_SEND_ERROR ? 1 : 0;
2310 if (put_user(len, optlen) ||
2311 put_user(val, optval))
2315 case NETLINK_NO_ENOBUFS:
2316 if (len < sizeof(int))
2319 val = nlk->flags & NETLINK_F_RECV_NO_ENOBUFS ? 1 : 0;
2320 if (put_user(len, optlen) ||
2321 put_user(val, optval))
2325 case NETLINK_LIST_MEMBERSHIPS: {
2326 int pos, idx, shift;
2329 netlink_table_grab();
2330 for (pos = 0; pos * 8 < nlk->ngroups; pos += sizeof(u32)) {
2331 if (len - pos < sizeof(u32))
2334 idx = pos / sizeof(unsigned long);
2335 shift = (pos % sizeof(unsigned long)) * 8;
2336 if (put_user((u32)(nlk->groups[idx] >> shift),
2337 (u32 __user *)(optval + pos))) {
2342 if (put_user(ALIGN(nlk->ngroups / 8, sizeof(u32)), optlen))
2344 netlink_table_ungrab();
2347 case NETLINK_CAP_ACK:
2348 if (len < sizeof(int))
2351 val = nlk->flags & NETLINK_F_CAP_ACK ? 1 : 0;
2352 if (put_user(len, optlen) ||
2353 put_user(val, optval))
2363 static void netlink_cmsg_recv_pktinfo(struct msghdr *msg, struct sk_buff *skb)
2365 struct nl_pktinfo info;
2367 info.group = NETLINK_CB(skb).dst_group;
2368 put_cmsg(msg, SOL_NETLINK, NETLINK_PKTINFO, sizeof(info), &info);
2371 static void netlink_cmsg_listen_all_nsid(struct sock *sk, struct msghdr *msg,
2372 struct sk_buff *skb)
2374 if (!NETLINK_CB(skb).nsid_is_set)
2377 put_cmsg(msg, SOL_NETLINK, NETLINK_LISTEN_ALL_NSID, sizeof(int),
2378 &NETLINK_CB(skb).nsid);
2381 static int netlink_sendmsg(struct socket *sock, struct msghdr *msg, size_t len)
2383 struct sock *sk = sock->sk;
2384 struct netlink_sock *nlk = nlk_sk(sk);
2385 DECLARE_SOCKADDR(struct sockaddr_nl *, addr, msg->msg_name);
2388 struct sk_buff *skb;
2390 struct scm_cookie scm;
2391 u32 netlink_skb_flags = 0;
2393 if (msg->msg_flags&MSG_OOB)
2396 err = scm_send(sock, msg, &scm, true);
2400 if (msg->msg_namelen) {
2402 if (addr->nl_family != AF_NETLINK)
2404 dst_portid = addr->nl_pid;
2405 dst_group = ffs(addr->nl_groups);
2407 if ((dst_group || dst_portid) &&
2408 !netlink_allowed(sock, NL_CFG_F_NONROOT_SEND))
2410 netlink_skb_flags |= NETLINK_SKB_DST;
2412 dst_portid = nlk->dst_portid;
2413 dst_group = nlk->dst_group;
2417 err = netlink_autobind(sock);
2422 /* It's a really convoluted way for userland to ask for mmaped
2423 * sendmsg(), but that's what we've got...
2425 if (netlink_tx_is_mmaped(sk) &&
2426 iter_is_iovec(&msg->msg_iter) &&
2427 msg->msg_iter.nr_segs == 1 &&
2428 msg->msg_iter.iov->iov_base == NULL) {
2429 err = netlink_mmap_sendmsg(sk, msg, dst_portid, dst_group,
2435 if (len > sk->sk_sndbuf - 32)
2438 skb = netlink_alloc_large_skb(len, dst_group);
2442 NETLINK_CB(skb).portid = nlk->portid;
2443 NETLINK_CB(skb).dst_group = dst_group;
2444 NETLINK_CB(skb).creds = scm.creds;
2445 NETLINK_CB(skb).flags = netlink_skb_flags;
2448 if (memcpy_from_msg(skb_put(skb, len), msg, len)) {
2453 err = security_netlink_send(sk, skb);
2460 atomic_inc(&skb->users);
2461 netlink_broadcast(sk, skb, dst_portid, dst_group, GFP_KERNEL);
2463 err = netlink_unicast(sk, skb, dst_portid, msg->msg_flags&MSG_DONTWAIT);
2470 static int netlink_recvmsg(struct socket *sock, struct msghdr *msg, size_t len,
2473 struct scm_cookie scm;
2474 struct sock *sk = sock->sk;
2475 struct netlink_sock *nlk = nlk_sk(sk);
2476 int noblock = flags&MSG_DONTWAIT;
2478 struct sk_buff *skb, *data_skb;
2486 skb = skb_recv_datagram(sk, flags, noblock, &err);
2492 #ifdef CONFIG_COMPAT_NETLINK_MESSAGES
2493 if (unlikely(skb_shinfo(skb)->frag_list)) {
2495 * If this skb has a frag_list, then here that means that we
2496 * will have to use the frag_list skb's data for compat tasks
2497 * and the regular skb's data for normal (non-compat) tasks.
2499 * If we need to send the compat skb, assign it to the
2500 * 'data_skb' variable so that it will be used below for data
2501 * copying. We keep 'skb' for everything else, including
2502 * freeing both later.
2504 if (flags & MSG_CMSG_COMPAT)
2505 data_skb = skb_shinfo(skb)->frag_list;
2509 /* Record the max length of recvmsg() calls for future allocations */
2510 nlk->max_recvmsg_len = max(nlk->max_recvmsg_len, len);
2511 nlk->max_recvmsg_len = min_t(size_t, nlk->max_recvmsg_len,
2514 copied = data_skb->len;
2516 msg->msg_flags |= MSG_TRUNC;
2520 skb_reset_transport_header(data_skb);
2521 err = skb_copy_datagram_msg(data_skb, 0, msg, copied);
2523 if (msg->msg_name) {
2524 DECLARE_SOCKADDR(struct sockaddr_nl *, addr, msg->msg_name);
2525 addr->nl_family = AF_NETLINK;
2527 addr->nl_pid = NETLINK_CB(skb).portid;
2528 addr->nl_groups = netlink_group_mask(NETLINK_CB(skb).dst_group);
2529 msg->msg_namelen = sizeof(*addr);
2532 if (nlk->flags & NETLINK_F_RECV_PKTINFO)
2533 netlink_cmsg_recv_pktinfo(msg, skb);
2534 if (nlk->flags & NETLINK_F_LISTEN_ALL_NSID)
2535 netlink_cmsg_listen_all_nsid(sk, msg, skb);
2537 memset(&scm, 0, sizeof(scm));
2538 scm.creds = *NETLINK_CREDS(skb);
2539 if (flags & MSG_TRUNC)
2540 copied = data_skb->len;
2542 skb_free_datagram(sk, skb);
2544 if (nlk->cb_running &&
2545 atomic_read(&sk->sk_rmem_alloc) <= sk->sk_rcvbuf / 2) {
2546 ret = netlink_dump(sk);
2549 sk->sk_error_report(sk);
2553 scm_recv(sock, msg, &scm, flags);
2555 netlink_rcv_wake(sk);
2556 return err ? : copied;
2559 static void netlink_data_ready(struct sock *sk)
2565 * We export these functions to other modules. They provide a
2566 * complete set of kernel non-blocking support for message
2571 __netlink_kernel_create(struct net *net, int unit, struct module *module,
2572 struct netlink_kernel_cfg *cfg)
2574 struct socket *sock;
2576 struct netlink_sock *nlk;
2577 struct listeners *listeners = NULL;
2578 struct mutex *cb_mutex = cfg ? cfg->cb_mutex : NULL;
2579 unsigned int groups;
2583 if (unit < 0 || unit >= MAX_LINKS)
2586 if (sock_create_lite(PF_NETLINK, SOCK_DGRAM, unit, &sock))
2589 if (__netlink_create(net, sock, cb_mutex, unit, 1) < 0)
2590 goto out_sock_release_nosk;
2594 if (!cfg || cfg->groups < 32)
2597 groups = cfg->groups;
2599 listeners = kzalloc(sizeof(*listeners) + NLGRPSZ(groups), GFP_KERNEL);
2601 goto out_sock_release;
2603 sk->sk_data_ready = netlink_data_ready;
2604 if (cfg && cfg->input)
2605 nlk_sk(sk)->netlink_rcv = cfg->input;
2607 if (netlink_insert(sk, 0))
2608 goto out_sock_release;
2611 nlk->flags |= NETLINK_F_KERNEL_SOCKET;
2613 netlink_table_grab();
2614 if (!nl_table[unit].registered) {
2615 nl_table[unit].groups = groups;
2616 rcu_assign_pointer(nl_table[unit].listeners, listeners);
2617 nl_table[unit].cb_mutex = cb_mutex;
2618 nl_table[unit].module = module;
2620 nl_table[unit].bind = cfg->bind;
2621 nl_table[unit].unbind = cfg->unbind;
2622 nl_table[unit].flags = cfg->flags;
2624 nl_table[unit].compare = cfg->compare;
2626 nl_table[unit].registered = 1;
2629 nl_table[unit].registered++;
2631 netlink_table_ungrab();
2636 netlink_kernel_release(sk);
2639 out_sock_release_nosk:
2643 EXPORT_SYMBOL(__netlink_kernel_create);
2646 netlink_kernel_release(struct sock *sk)
2648 if (sk == NULL || sk->sk_socket == NULL)
2651 sock_release(sk->sk_socket);
2653 EXPORT_SYMBOL(netlink_kernel_release);
2655 int __netlink_change_ngroups(struct sock *sk, unsigned int groups)
2657 struct listeners *new, *old;
2658 struct netlink_table *tbl = &nl_table[sk->sk_protocol];
2663 if (NLGRPSZ(tbl->groups) < NLGRPSZ(groups)) {
2664 new = kzalloc(sizeof(*new) + NLGRPSZ(groups), GFP_ATOMIC);
2667 old = nl_deref_protected(tbl->listeners);
2668 memcpy(new->masks, old->masks, NLGRPSZ(tbl->groups));
2669 rcu_assign_pointer(tbl->listeners, new);
2671 kfree_rcu(old, rcu);
2673 tbl->groups = groups;
2679 * netlink_change_ngroups - change number of multicast groups
2681 * This changes the number of multicast groups that are available
2682 * on a certain netlink family. Note that it is not possible to
2683 * change the number of groups to below 32. Also note that it does
2684 * not implicitly call netlink_clear_multicast_users() when the
2685 * number of groups is reduced.
2687 * @sk: The kernel netlink socket, as returned by netlink_kernel_create().
2688 * @groups: The new number of groups.
2690 int netlink_change_ngroups(struct sock *sk, unsigned int groups)
2694 netlink_table_grab();
2695 err = __netlink_change_ngroups(sk, groups);
2696 netlink_table_ungrab();
2701 void __netlink_clear_multicast_users(struct sock *ksk, unsigned int group)
2704 struct netlink_table *tbl = &nl_table[ksk->sk_protocol];
2706 sk_for_each_bound(sk, &tbl->mc_list)
2707 netlink_update_socket_mc(nlk_sk(sk), group, 0);
2711 __nlmsg_put(struct sk_buff *skb, u32 portid, u32 seq, int type, int len, int flags)
2713 struct nlmsghdr *nlh;
2714 int size = nlmsg_msg_size(len);
2716 nlh = (struct nlmsghdr *)skb_put(skb, NLMSG_ALIGN(size));
2717 nlh->nlmsg_type = type;
2718 nlh->nlmsg_len = size;
2719 nlh->nlmsg_flags = flags;
2720 nlh->nlmsg_pid = portid;
2721 nlh->nlmsg_seq = seq;
2722 if (!__builtin_constant_p(size) || NLMSG_ALIGN(size) - size != 0)
2723 memset(nlmsg_data(nlh) + len, 0, NLMSG_ALIGN(size) - size);
2726 EXPORT_SYMBOL(__nlmsg_put);
2729 * It looks a bit ugly.
2730 * It would be better to create kernel thread.
2733 static int netlink_dump(struct sock *sk)
2735 struct netlink_sock *nlk = nlk_sk(sk);
2736 struct netlink_callback *cb;
2737 struct sk_buff *skb = NULL;
2738 struct nlmsghdr *nlh;
2739 int len, err = -ENOBUFS;
2742 mutex_lock(nlk->cb_mutex);
2743 if (!nlk->cb_running) {
2749 alloc_size = max_t(int, cb->min_dump_alloc, NLMSG_GOODSIZE);
2751 if (!netlink_rx_is_mmaped(sk) &&
2752 atomic_read(&sk->sk_rmem_alloc) >= sk->sk_rcvbuf)
2755 /* NLMSG_GOODSIZE is small to avoid high order allocations being
2756 * required, but it makes sense to _attempt_ a 16K bytes allocation
2757 * to reduce number of system calls on dump operations, if user
2758 * ever provided a big enough buffer.
2760 if (alloc_size < nlk->max_recvmsg_len) {
2761 skb = netlink_alloc_skb(sk,
2762 nlk->max_recvmsg_len,
2767 /* available room should be exact amount to avoid MSG_TRUNC */
2769 skb_reserve(skb, skb_tailroom(skb) -
2770 nlk->max_recvmsg_len);
2773 skb = netlink_alloc_skb(sk, alloc_size, nlk->portid,
2777 netlink_skb_set_owner_r(skb, sk);
2779 len = cb->dump(skb, cb);
2782 mutex_unlock(nlk->cb_mutex);
2784 if (sk_filter(sk, skb))
2787 __netlink_sendskb(sk, skb);
2791 nlh = nlmsg_put_answer(skb, cb, NLMSG_DONE, sizeof(len), NLM_F_MULTI);
2795 nl_dump_check_consistent(cb, nlh);
2797 memcpy(nlmsg_data(nlh), &len, sizeof(len));
2799 if (sk_filter(sk, skb))
2802 __netlink_sendskb(sk, skb);
2807 nlk->cb_running = false;
2808 mutex_unlock(nlk->cb_mutex);
2809 module_put(cb->module);
2810 consume_skb(cb->skb);
2814 mutex_unlock(nlk->cb_mutex);
2819 int __netlink_dump_start(struct sock *ssk, struct sk_buff *skb,
2820 const struct nlmsghdr *nlh,
2821 struct netlink_dump_control *control)
2823 struct netlink_callback *cb;
2825 struct netlink_sock *nlk;
2828 /* Memory mapped dump requests need to be copied to avoid looping
2829 * on the pending state in netlink_mmap_sendmsg() while the CB hold
2830 * a reference to the skb.
2832 if (netlink_skb_is_mmaped(skb)) {
2833 skb = skb_copy(skb, GFP_KERNEL);
2837 atomic_inc(&skb->users);
2839 sk = netlink_lookup(sock_net(ssk), ssk->sk_protocol, NETLINK_CB(skb).portid);
2841 ret = -ECONNREFUSED;
2846 mutex_lock(nlk->cb_mutex);
2847 /* A dump is in progress... */
2848 if (nlk->cb_running) {
2852 /* add reference of module which cb->dump belongs to */
2853 if (!try_module_get(control->module)) {
2854 ret = -EPROTONOSUPPORT;
2859 memset(cb, 0, sizeof(*cb));
2860 cb->dump = control->dump;
2861 cb->done = control->done;
2863 cb->data = control->data;
2864 cb->module = control->module;
2865 cb->min_dump_alloc = control->min_dump_alloc;
2868 nlk->cb_running = true;
2870 mutex_unlock(nlk->cb_mutex);
2872 ret = netlink_dump(sk);
2878 /* We successfully started a dump, by returning -EINTR we
2879 * signal not to send ACK even if it was requested.
2885 mutex_unlock(nlk->cb_mutex);
2890 EXPORT_SYMBOL(__netlink_dump_start);
2892 void netlink_ack(struct sk_buff *in_skb, struct nlmsghdr *nlh, int err)
2894 struct sk_buff *skb;
2895 struct nlmsghdr *rep;
2896 struct nlmsgerr *errmsg;
2897 size_t payload = sizeof(*errmsg);
2898 struct netlink_sock *nlk = nlk_sk(NETLINK_CB(in_skb).sk);
2900 /* Error messages get the original request appened, unless the user
2901 * requests to cap the error message.
2903 if (!(nlk->flags & NETLINK_F_CAP_ACK) && err)
2904 payload += nlmsg_len(nlh);
2906 skb = netlink_alloc_skb(in_skb->sk, nlmsg_total_size(payload),
2907 NETLINK_CB(in_skb).portid, GFP_KERNEL);
2911 sk = netlink_lookup(sock_net(in_skb->sk),
2912 in_skb->sk->sk_protocol,
2913 NETLINK_CB(in_skb).portid);
2915 sk->sk_err = ENOBUFS;
2916 sk->sk_error_report(sk);
2922 rep = __nlmsg_put(skb, NETLINK_CB(in_skb).portid, nlh->nlmsg_seq,
2923 NLMSG_ERROR, payload, 0);
2924 errmsg = nlmsg_data(rep);
2925 errmsg->error = err;
2926 memcpy(&errmsg->msg, nlh, payload > sizeof(*errmsg) ? nlh->nlmsg_len : sizeof(*nlh));
2927 netlink_unicast(in_skb->sk, skb, NETLINK_CB(in_skb).portid, MSG_DONTWAIT);
2929 EXPORT_SYMBOL(netlink_ack);
2931 int netlink_rcv_skb(struct sk_buff *skb, int (*cb)(struct sk_buff *,
2934 struct nlmsghdr *nlh;
2937 while (skb->len >= nlmsg_total_size(0)) {
2940 nlh = nlmsg_hdr(skb);
2943 if (nlh->nlmsg_len < NLMSG_HDRLEN || skb->len < nlh->nlmsg_len)
2946 /* Only requests are handled by the kernel */
2947 if (!(nlh->nlmsg_flags & NLM_F_REQUEST))
2950 /* Skip control messages */
2951 if (nlh->nlmsg_type < NLMSG_MIN_TYPE)
2959 if (nlh->nlmsg_flags & NLM_F_ACK || err)
2960 netlink_ack(skb, nlh, err);
2963 msglen = NLMSG_ALIGN(nlh->nlmsg_len);
2964 if (msglen > skb->len)
2966 skb_pull(skb, msglen);
2971 EXPORT_SYMBOL(netlink_rcv_skb);
2974 * nlmsg_notify - send a notification netlink message
2975 * @sk: netlink socket to use
2976 * @skb: notification message
2977 * @portid: destination netlink portid for reports or 0
2978 * @group: destination multicast group or 0
2979 * @report: 1 to report back, 0 to disable
2980 * @flags: allocation flags
2982 int nlmsg_notify(struct sock *sk, struct sk_buff *skb, u32 portid,
2983 unsigned int group, int report, gfp_t flags)
2988 int exclude_portid = 0;
2991 atomic_inc(&skb->users);
2992 exclude_portid = portid;
2995 /* errors reported via destination sk->sk_err, but propagate
2996 * delivery errors if NETLINK_BROADCAST_ERROR flag is set */
2997 err = nlmsg_multicast(sk, skb, exclude_portid, group, flags);
3003 err2 = nlmsg_unicast(sk, skb, portid);
3004 if (!err || err == -ESRCH)
3010 EXPORT_SYMBOL(nlmsg_notify);
3012 #ifdef CONFIG_PROC_FS
3013 struct nl_seq_iter {
3014 struct seq_net_private p;
3015 struct rhashtable_iter hti;
3019 static int netlink_walk_start(struct nl_seq_iter *iter)
3023 err = rhashtable_walk_init(&nl_table[iter->link].hash, &iter->hti);
3025 iter->link = MAX_LINKS;
3029 err = rhashtable_walk_start(&iter->hti);
3030 return err == -EAGAIN ? 0 : err;
3033 static void netlink_walk_stop(struct nl_seq_iter *iter)
3035 rhashtable_walk_stop(&iter->hti);
3036 rhashtable_walk_exit(&iter->hti);
3039 static void *__netlink_seq_next(struct seq_file *seq)
3041 struct nl_seq_iter *iter = seq->private;
3042 struct netlink_sock *nlk;
3048 nlk = rhashtable_walk_next(&iter->hti);
3051 if (PTR_ERR(nlk) == -EAGAIN)
3060 netlink_walk_stop(iter);
3061 if (++iter->link >= MAX_LINKS)
3064 err = netlink_walk_start(iter);
3066 return ERR_PTR(err);
3068 } while (sock_net(&nlk->sk) != seq_file_net(seq));
3073 static void *netlink_seq_start(struct seq_file *seq, loff_t *posp)
3075 struct nl_seq_iter *iter = seq->private;
3076 void *obj = SEQ_START_TOKEN;
3082 err = netlink_walk_start(iter);
3084 return ERR_PTR(err);
3086 for (pos = *posp; pos && obj && !IS_ERR(obj); pos--)
3087 obj = __netlink_seq_next(seq);
3092 static void *netlink_seq_next(struct seq_file *seq, void *v, loff_t *pos)
3095 return __netlink_seq_next(seq);
3098 static void netlink_seq_stop(struct seq_file *seq, void *v)
3100 struct nl_seq_iter *iter = seq->private;
3102 if (iter->link >= MAX_LINKS)
3105 netlink_walk_stop(iter);
3109 static int netlink_seq_show(struct seq_file *seq, void *v)
3111 if (v == SEQ_START_TOKEN) {
3113 "sk Eth Pid Groups "
3114 "Rmem Wmem Dump Locks Drops Inode\n");
3117 struct netlink_sock *nlk = nlk_sk(s);
3119 seq_printf(seq, "%pK %-3d %-6u %08x %-8d %-8d %d %-8d %-8d %-8lu\n",
3123 nlk->groups ? (u32)nlk->groups[0] : 0,
3124 sk_rmem_alloc_get(s),
3125 sk_wmem_alloc_get(s),
3127 atomic_read(&s->sk_refcnt),
3128 atomic_read(&s->sk_drops),
3136 static const struct seq_operations netlink_seq_ops = {
3137 .start = netlink_seq_start,
3138 .next = netlink_seq_next,
3139 .stop = netlink_seq_stop,
3140 .show = netlink_seq_show,
3144 static int netlink_seq_open(struct inode *inode, struct file *file)
3146 return seq_open_net(inode, file, &netlink_seq_ops,
3147 sizeof(struct nl_seq_iter));
3150 static const struct file_operations netlink_seq_fops = {
3151 .owner = THIS_MODULE,
3152 .open = netlink_seq_open,
3154 .llseek = seq_lseek,
3155 .release = seq_release_net,
3160 int netlink_register_notifier(struct notifier_block *nb)
3162 return atomic_notifier_chain_register(&netlink_chain, nb);
3164 EXPORT_SYMBOL(netlink_register_notifier);
3166 int netlink_unregister_notifier(struct notifier_block *nb)
3168 return atomic_notifier_chain_unregister(&netlink_chain, nb);
3170 EXPORT_SYMBOL(netlink_unregister_notifier);
3172 static const struct proto_ops netlink_ops = {
3173 .family = PF_NETLINK,
3174 .owner = THIS_MODULE,
3175 .release = netlink_release,
3176 .bind = netlink_bind,
3177 .connect = netlink_connect,
3178 .socketpair = sock_no_socketpair,
3179 .accept = sock_no_accept,
3180 .getname = netlink_getname,
3181 .poll = netlink_poll,
3182 .ioctl = sock_no_ioctl,
3183 .listen = sock_no_listen,
3184 .shutdown = sock_no_shutdown,
3185 .setsockopt = netlink_setsockopt,
3186 .getsockopt = netlink_getsockopt,
3187 .sendmsg = netlink_sendmsg,
3188 .recvmsg = netlink_recvmsg,
3189 .mmap = netlink_mmap,
3190 .sendpage = sock_no_sendpage,
3193 static const struct net_proto_family netlink_family_ops = {
3194 .family = PF_NETLINK,
3195 .create = netlink_create,
3196 .owner = THIS_MODULE, /* for consistency 8) */
3199 static int __net_init netlink_net_init(struct net *net)
3201 #ifdef CONFIG_PROC_FS
3202 if (!proc_create("netlink", 0, net->proc_net, &netlink_seq_fops))
3208 static void __net_exit netlink_net_exit(struct net *net)
3210 #ifdef CONFIG_PROC_FS
3211 remove_proc_entry("netlink", net->proc_net);
3215 static void __init netlink_add_usersock_entry(void)
3217 struct listeners *listeners;
3220 listeners = kzalloc(sizeof(*listeners) + NLGRPSZ(groups), GFP_KERNEL);
3222 panic("netlink_add_usersock_entry: Cannot allocate listeners\n");
3224 netlink_table_grab();
3226 nl_table[NETLINK_USERSOCK].groups = groups;
3227 rcu_assign_pointer(nl_table[NETLINK_USERSOCK].listeners, listeners);
3228 nl_table[NETLINK_USERSOCK].module = THIS_MODULE;
3229 nl_table[NETLINK_USERSOCK].registered = 1;
3230 nl_table[NETLINK_USERSOCK].flags = NL_CFG_F_NONROOT_SEND;
3232 netlink_table_ungrab();
3235 static struct pernet_operations __net_initdata netlink_net_ops = {
3236 .init = netlink_net_init,
3237 .exit = netlink_net_exit,
3240 static inline u32 netlink_hash(const void *data, u32 len, u32 seed)
3242 const struct netlink_sock *nlk = data;
3243 struct netlink_compare_arg arg;
3245 netlink_compare_arg_init(&arg, sock_net(&nlk->sk), nlk->portid);
3246 return jhash2((u32 *)&arg, netlink_compare_arg_len / sizeof(u32), seed);
3249 static const struct rhashtable_params netlink_rhashtable_params = {
3250 .head_offset = offsetof(struct netlink_sock, node),
3251 .key_len = netlink_compare_arg_len,
3252 .obj_hashfn = netlink_hash,
3253 .obj_cmpfn = netlink_compare,
3254 .automatic_shrinking = true,
3257 static int __init netlink_proto_init(void)
3260 int err = proto_register(&netlink_proto, 0);
3265 BUILD_BUG_ON(sizeof(struct netlink_skb_parms) > FIELD_SIZEOF(struct sk_buff, cb));
3267 nl_table = kcalloc(MAX_LINKS, sizeof(*nl_table), GFP_KERNEL);
3271 for (i = 0; i < MAX_LINKS; i++) {
3272 if (rhashtable_init(&nl_table[i].hash,
3273 &netlink_rhashtable_params) < 0) {
3275 rhashtable_destroy(&nl_table[i].hash);
3281 INIT_LIST_HEAD(&netlink_tap_all);
3283 netlink_add_usersock_entry();
3285 sock_register(&netlink_family_ops);
3286 register_pernet_subsys(&netlink_net_ops);
3287 /* The netlink device handler may be needed early. */
3292 panic("netlink_init: Cannot allocate nl_table\n");
3295 core_initcall(netlink_proto_init);