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 inline u32 netlink_group_mask(u32 group)
121 return group ? 1 << (group - 1) : 0;
124 int netlink_add_tap(struct netlink_tap *nt)
126 if (unlikely(nt->dev->type != ARPHRD_NETLINK))
129 spin_lock(&netlink_tap_lock);
130 list_add_rcu(&nt->list, &netlink_tap_all);
131 spin_unlock(&netlink_tap_lock);
133 __module_get(nt->module);
137 EXPORT_SYMBOL_GPL(netlink_add_tap);
139 static int __netlink_remove_tap(struct netlink_tap *nt)
142 struct netlink_tap *tmp;
144 spin_lock(&netlink_tap_lock);
146 list_for_each_entry(tmp, &netlink_tap_all, list) {
148 list_del_rcu(&nt->list);
154 pr_warn("__netlink_remove_tap: %p not found\n", nt);
156 spin_unlock(&netlink_tap_lock);
158 if (found && nt->module)
159 module_put(nt->module);
161 return found ? 0 : -ENODEV;
164 int netlink_remove_tap(struct netlink_tap *nt)
168 ret = __netlink_remove_tap(nt);
173 EXPORT_SYMBOL_GPL(netlink_remove_tap);
175 static bool netlink_filter_tap(const struct sk_buff *skb)
177 struct sock *sk = skb->sk;
179 /* We take the more conservative approach and
180 * whitelist socket protocols that may pass.
182 switch (sk->sk_protocol) {
184 case NETLINK_USERSOCK:
185 case NETLINK_SOCK_DIAG:
188 case NETLINK_FIB_LOOKUP:
189 case NETLINK_NETFILTER:
190 case NETLINK_GENERIC:
197 static int __netlink_deliver_tap_skb(struct sk_buff *skb,
198 struct net_device *dev)
200 struct sk_buff *nskb;
201 struct sock *sk = skb->sk;
205 nskb = skb_clone(skb, GFP_ATOMIC);
208 nskb->protocol = htons((u16) sk->sk_protocol);
209 nskb->pkt_type = netlink_is_kernel(sk) ?
210 PACKET_KERNEL : PACKET_USER;
211 skb_reset_network_header(nskb);
212 ret = dev_queue_xmit(nskb);
213 if (unlikely(ret > 0))
214 ret = net_xmit_errno(ret);
221 static void __netlink_deliver_tap(struct sk_buff *skb)
224 struct netlink_tap *tmp;
226 if (!netlink_filter_tap(skb))
229 list_for_each_entry_rcu(tmp, &netlink_tap_all, list) {
230 ret = __netlink_deliver_tap_skb(skb, tmp->dev);
236 static void netlink_deliver_tap(struct sk_buff *skb)
240 if (unlikely(!list_empty(&netlink_tap_all)))
241 __netlink_deliver_tap(skb);
246 static void netlink_deliver_tap_kernel(struct sock *dst, struct sock *src,
249 if (!(netlink_is_kernel(dst) && netlink_is_kernel(src)))
250 netlink_deliver_tap(skb);
253 static void netlink_overrun(struct sock *sk)
255 struct netlink_sock *nlk = nlk_sk(sk);
257 if (!(nlk->flags & NETLINK_RECV_NO_ENOBUFS)) {
258 if (!test_and_set_bit(NETLINK_CONGESTED, &nlk_sk(sk)->state)) {
259 sk->sk_err = ENOBUFS;
260 sk->sk_error_report(sk);
263 atomic_inc(&sk->sk_drops);
266 static void netlink_rcv_wake(struct sock *sk)
268 struct netlink_sock *nlk = nlk_sk(sk);
270 if (skb_queue_empty(&sk->sk_receive_queue))
271 clear_bit(NETLINK_CONGESTED, &nlk->state);
272 if (!test_bit(NETLINK_CONGESTED, &nlk->state))
273 wake_up_interruptible(&nlk->wait);
276 #ifdef CONFIG_NETLINK_MMAP
277 static bool netlink_skb_is_mmaped(const struct sk_buff *skb)
279 return NETLINK_CB(skb).flags & NETLINK_SKB_MMAPED;
282 static bool netlink_rx_is_mmaped(struct sock *sk)
284 return nlk_sk(sk)->rx_ring.pg_vec != NULL;
287 static bool netlink_tx_is_mmaped(struct sock *sk)
289 return nlk_sk(sk)->tx_ring.pg_vec != NULL;
292 static __pure struct page *pgvec_to_page(const void *addr)
294 if (is_vmalloc_addr(addr))
295 return vmalloc_to_page(addr);
297 return virt_to_page(addr);
300 static void free_pg_vec(void **pg_vec, unsigned int order, unsigned int len)
304 for (i = 0; i < len; i++) {
305 if (pg_vec[i] != NULL) {
306 if (is_vmalloc_addr(pg_vec[i]))
309 free_pages((unsigned long)pg_vec[i], order);
315 static void *alloc_one_pg_vec_page(unsigned long order)
318 gfp_t gfp_flags = GFP_KERNEL | __GFP_COMP | __GFP_ZERO |
319 __GFP_NOWARN | __GFP_NORETRY;
321 buffer = (void *)__get_free_pages(gfp_flags, order);
325 buffer = vzalloc((1 << order) * PAGE_SIZE);
329 gfp_flags &= ~__GFP_NORETRY;
330 return (void *)__get_free_pages(gfp_flags, order);
333 static void **alloc_pg_vec(struct netlink_sock *nlk,
334 struct nl_mmap_req *req, unsigned int order)
336 unsigned int block_nr = req->nm_block_nr;
340 pg_vec = kcalloc(block_nr, sizeof(void *), GFP_KERNEL);
344 for (i = 0; i < block_nr; i++) {
345 pg_vec[i] = alloc_one_pg_vec_page(order);
346 if (pg_vec[i] == NULL)
352 free_pg_vec(pg_vec, order, block_nr);
356 static int netlink_set_ring(struct sock *sk, struct nl_mmap_req *req,
357 bool closing, bool tx_ring)
359 struct netlink_sock *nlk = nlk_sk(sk);
360 struct netlink_ring *ring;
361 struct sk_buff_head *queue;
362 void **pg_vec = NULL;
363 unsigned int order = 0;
366 ring = tx_ring ? &nlk->tx_ring : &nlk->rx_ring;
367 queue = tx_ring ? &sk->sk_write_queue : &sk->sk_receive_queue;
370 if (atomic_read(&nlk->mapped))
372 if (atomic_read(&ring->pending))
376 if (req->nm_block_nr) {
377 if (ring->pg_vec != NULL)
380 if ((int)req->nm_block_size <= 0)
382 if (!PAGE_ALIGNED(req->nm_block_size))
384 if (req->nm_frame_size < NL_MMAP_HDRLEN)
386 if (!IS_ALIGNED(req->nm_frame_size, NL_MMAP_MSG_ALIGNMENT))
389 ring->frames_per_block = req->nm_block_size /
391 if (ring->frames_per_block == 0)
393 if (ring->frames_per_block * req->nm_block_nr !=
397 order = get_order(req->nm_block_size);
398 pg_vec = alloc_pg_vec(nlk, req, order);
402 if (req->nm_frame_nr)
407 mutex_lock(&nlk->pg_vec_lock);
408 if (closing || atomic_read(&nlk->mapped) == 0) {
410 spin_lock_bh(&queue->lock);
412 ring->frame_max = req->nm_frame_nr - 1;
414 ring->frame_size = req->nm_frame_size;
415 ring->pg_vec_pages = req->nm_block_size / PAGE_SIZE;
417 swap(ring->pg_vec_len, req->nm_block_nr);
418 swap(ring->pg_vec_order, order);
419 swap(ring->pg_vec, pg_vec);
421 __skb_queue_purge(queue);
422 spin_unlock_bh(&queue->lock);
424 WARN_ON(atomic_read(&nlk->mapped));
426 mutex_unlock(&nlk->pg_vec_lock);
429 free_pg_vec(pg_vec, order, req->nm_block_nr);
433 static void netlink_mm_open(struct vm_area_struct *vma)
435 struct file *file = vma->vm_file;
436 struct socket *sock = file->private_data;
437 struct sock *sk = sock->sk;
440 atomic_inc(&nlk_sk(sk)->mapped);
443 static void netlink_mm_close(struct vm_area_struct *vma)
445 struct file *file = vma->vm_file;
446 struct socket *sock = file->private_data;
447 struct sock *sk = sock->sk;
450 atomic_dec(&nlk_sk(sk)->mapped);
453 static const struct vm_operations_struct netlink_mmap_ops = {
454 .open = netlink_mm_open,
455 .close = netlink_mm_close,
458 static int netlink_mmap(struct file *file, struct socket *sock,
459 struct vm_area_struct *vma)
461 struct sock *sk = sock->sk;
462 struct netlink_sock *nlk = nlk_sk(sk);
463 struct netlink_ring *ring;
464 unsigned long start, size, expected;
471 mutex_lock(&nlk->pg_vec_lock);
474 for (ring = &nlk->rx_ring; ring <= &nlk->tx_ring; ring++) {
475 if (ring->pg_vec == NULL)
477 expected += ring->pg_vec_len * ring->pg_vec_pages * PAGE_SIZE;
483 size = vma->vm_end - vma->vm_start;
484 if (size != expected)
487 start = vma->vm_start;
488 for (ring = &nlk->rx_ring; ring <= &nlk->tx_ring; ring++) {
489 if (ring->pg_vec == NULL)
492 for (i = 0; i < ring->pg_vec_len; i++) {
494 void *kaddr = ring->pg_vec[i];
497 for (pg_num = 0; pg_num < ring->pg_vec_pages; pg_num++) {
498 page = pgvec_to_page(kaddr);
499 err = vm_insert_page(vma, start, page);
508 atomic_inc(&nlk->mapped);
509 vma->vm_ops = &netlink_mmap_ops;
512 mutex_unlock(&nlk->pg_vec_lock);
516 static void netlink_frame_flush_dcache(const struct nl_mmap_hdr *hdr, unsigned int nm_len)
518 #if ARCH_IMPLEMENTS_FLUSH_DCACHE_PAGE == 1
519 struct page *p_start, *p_end;
521 /* First page is flushed through netlink_{get,set}_status */
522 p_start = pgvec_to_page(hdr + PAGE_SIZE);
523 p_end = pgvec_to_page((void *)hdr + NL_MMAP_HDRLEN + nm_len - 1);
524 while (p_start <= p_end) {
525 flush_dcache_page(p_start);
531 static enum nl_mmap_status netlink_get_status(const struct nl_mmap_hdr *hdr)
534 flush_dcache_page(pgvec_to_page(hdr));
535 return hdr->nm_status;
538 static void netlink_set_status(struct nl_mmap_hdr *hdr,
539 enum nl_mmap_status status)
542 hdr->nm_status = status;
543 flush_dcache_page(pgvec_to_page(hdr));
546 static struct nl_mmap_hdr *
547 __netlink_lookup_frame(const struct netlink_ring *ring, unsigned int pos)
549 unsigned int pg_vec_pos, frame_off;
551 pg_vec_pos = pos / ring->frames_per_block;
552 frame_off = pos % ring->frames_per_block;
554 return ring->pg_vec[pg_vec_pos] + (frame_off * ring->frame_size);
557 static struct nl_mmap_hdr *
558 netlink_lookup_frame(const struct netlink_ring *ring, unsigned int pos,
559 enum nl_mmap_status status)
561 struct nl_mmap_hdr *hdr;
563 hdr = __netlink_lookup_frame(ring, pos);
564 if (netlink_get_status(hdr) != status)
570 static struct nl_mmap_hdr *
571 netlink_current_frame(const struct netlink_ring *ring,
572 enum nl_mmap_status status)
574 return netlink_lookup_frame(ring, ring->head, status);
577 static struct nl_mmap_hdr *
578 netlink_previous_frame(const struct netlink_ring *ring,
579 enum nl_mmap_status status)
583 prev = ring->head ? ring->head - 1 : ring->frame_max;
584 return netlink_lookup_frame(ring, prev, status);
587 static void netlink_increment_head(struct netlink_ring *ring)
589 ring->head = ring->head != ring->frame_max ? ring->head + 1 : 0;
592 static void netlink_forward_ring(struct netlink_ring *ring)
594 unsigned int head = ring->head, pos = head;
595 const struct nl_mmap_hdr *hdr;
598 hdr = __netlink_lookup_frame(ring, pos);
599 if (hdr->nm_status == NL_MMAP_STATUS_UNUSED)
601 if (hdr->nm_status != NL_MMAP_STATUS_SKIP)
603 netlink_increment_head(ring);
604 } while (ring->head != head);
607 static bool netlink_dump_space(struct netlink_sock *nlk)
609 struct netlink_ring *ring = &nlk->rx_ring;
610 struct nl_mmap_hdr *hdr;
613 hdr = netlink_current_frame(ring, NL_MMAP_STATUS_UNUSED);
617 n = ring->head + ring->frame_max / 2;
618 if (n > ring->frame_max)
619 n -= ring->frame_max;
621 hdr = __netlink_lookup_frame(ring, n);
623 return hdr->nm_status == NL_MMAP_STATUS_UNUSED;
626 static unsigned int netlink_poll(struct file *file, struct socket *sock,
629 struct sock *sk = sock->sk;
630 struct netlink_sock *nlk = nlk_sk(sk);
634 if (nlk->rx_ring.pg_vec != NULL) {
635 /* Memory mapped sockets don't call recvmsg(), so flow control
636 * for dumps is performed here. A dump is allowed to continue
637 * if at least half the ring is unused.
639 while (nlk->cb_running && netlink_dump_space(nlk)) {
640 err = netlink_dump(sk);
643 sk->sk_error_report(sk);
647 netlink_rcv_wake(sk);
650 mask = datagram_poll(file, sock, wait);
652 spin_lock_bh(&sk->sk_receive_queue.lock);
653 if (nlk->rx_ring.pg_vec) {
654 netlink_forward_ring(&nlk->rx_ring);
655 if (!netlink_previous_frame(&nlk->rx_ring, NL_MMAP_STATUS_UNUSED))
656 mask |= POLLIN | POLLRDNORM;
658 spin_unlock_bh(&sk->sk_receive_queue.lock);
660 spin_lock_bh(&sk->sk_write_queue.lock);
661 if (nlk->tx_ring.pg_vec) {
662 if (netlink_current_frame(&nlk->tx_ring, NL_MMAP_STATUS_UNUSED))
663 mask |= POLLOUT | POLLWRNORM;
665 spin_unlock_bh(&sk->sk_write_queue.lock);
670 static struct nl_mmap_hdr *netlink_mmap_hdr(struct sk_buff *skb)
672 return (struct nl_mmap_hdr *)(skb->head - NL_MMAP_HDRLEN);
675 static void netlink_ring_setup_skb(struct sk_buff *skb, struct sock *sk,
676 struct netlink_ring *ring,
677 struct nl_mmap_hdr *hdr)
682 size = ring->frame_size - NL_MMAP_HDRLEN;
683 data = (void *)hdr + NL_MMAP_HDRLEN;
687 skb_reset_tail_pointer(skb);
688 skb->end = skb->tail + size;
691 skb->destructor = netlink_skb_destructor;
692 NETLINK_CB(skb).flags |= NETLINK_SKB_MMAPED;
693 NETLINK_CB(skb).sk = sk;
696 static int netlink_mmap_sendmsg(struct sock *sk, struct msghdr *msg,
697 u32 dst_portid, u32 dst_group,
698 struct scm_cookie *scm)
700 struct netlink_sock *nlk = nlk_sk(sk);
701 struct netlink_ring *ring;
702 struct nl_mmap_hdr *hdr;
705 int err = 0, len = 0;
707 mutex_lock(&nlk->pg_vec_lock);
709 ring = &nlk->tx_ring;
710 maxlen = ring->frame_size - NL_MMAP_HDRLEN;
715 hdr = netlink_current_frame(ring, NL_MMAP_STATUS_VALID);
717 if (!(msg->msg_flags & MSG_DONTWAIT) &&
718 atomic_read(&nlk->tx_ring.pending))
723 nm_len = ACCESS_ONCE(hdr->nm_len);
724 if (nm_len > maxlen) {
729 netlink_frame_flush_dcache(hdr, nm_len);
731 skb = alloc_skb(nm_len, GFP_KERNEL);
736 __skb_put(skb, nm_len);
737 memcpy(skb->data, (void *)hdr + NL_MMAP_HDRLEN, nm_len);
738 netlink_set_status(hdr, NL_MMAP_STATUS_UNUSED);
740 netlink_increment_head(ring);
742 NETLINK_CB(skb).portid = nlk->portid;
743 NETLINK_CB(skb).dst_group = dst_group;
744 NETLINK_CB(skb).creds = scm->creds;
746 err = security_netlink_send(sk, skb);
752 if (unlikely(dst_group)) {
753 atomic_inc(&skb->users);
754 netlink_broadcast(sk, skb, dst_portid, dst_group,
757 err = netlink_unicast(sk, skb, dst_portid,
758 msg->msg_flags & MSG_DONTWAIT);
763 } while (hdr != NULL ||
764 (!(msg->msg_flags & MSG_DONTWAIT) &&
765 atomic_read(&nlk->tx_ring.pending)));
770 mutex_unlock(&nlk->pg_vec_lock);
774 static void netlink_queue_mmaped_skb(struct sock *sk, struct sk_buff *skb)
776 struct nl_mmap_hdr *hdr;
778 hdr = netlink_mmap_hdr(skb);
779 hdr->nm_len = skb->len;
780 hdr->nm_group = NETLINK_CB(skb).dst_group;
781 hdr->nm_pid = NETLINK_CB(skb).creds.pid;
782 hdr->nm_uid = from_kuid(sk_user_ns(sk), NETLINK_CB(skb).creds.uid);
783 hdr->nm_gid = from_kgid(sk_user_ns(sk), NETLINK_CB(skb).creds.gid);
784 netlink_frame_flush_dcache(hdr, hdr->nm_len);
785 netlink_set_status(hdr, NL_MMAP_STATUS_VALID);
787 NETLINK_CB(skb).flags |= NETLINK_SKB_DELIVERED;
791 static void netlink_ring_set_copied(struct sock *sk, struct sk_buff *skb)
793 struct netlink_sock *nlk = nlk_sk(sk);
794 struct netlink_ring *ring = &nlk->rx_ring;
795 struct nl_mmap_hdr *hdr;
797 spin_lock_bh(&sk->sk_receive_queue.lock);
798 hdr = netlink_current_frame(ring, NL_MMAP_STATUS_UNUSED);
800 spin_unlock_bh(&sk->sk_receive_queue.lock);
805 netlink_increment_head(ring);
806 __skb_queue_tail(&sk->sk_receive_queue, skb);
807 spin_unlock_bh(&sk->sk_receive_queue.lock);
809 hdr->nm_len = skb->len;
810 hdr->nm_group = NETLINK_CB(skb).dst_group;
811 hdr->nm_pid = NETLINK_CB(skb).creds.pid;
812 hdr->nm_uid = from_kuid(sk_user_ns(sk), NETLINK_CB(skb).creds.uid);
813 hdr->nm_gid = from_kgid(sk_user_ns(sk), NETLINK_CB(skb).creds.gid);
814 netlink_set_status(hdr, NL_MMAP_STATUS_COPY);
817 #else /* CONFIG_NETLINK_MMAP */
818 #define netlink_skb_is_mmaped(skb) false
819 #define netlink_rx_is_mmaped(sk) false
820 #define netlink_tx_is_mmaped(sk) false
821 #define netlink_mmap sock_no_mmap
822 #define netlink_poll datagram_poll
823 #define netlink_mmap_sendmsg(sk, msg, dst_portid, dst_group, scm) 0
824 #endif /* CONFIG_NETLINK_MMAP */
826 static void netlink_skb_destructor(struct sk_buff *skb)
828 #ifdef CONFIG_NETLINK_MMAP
829 struct nl_mmap_hdr *hdr;
830 struct netlink_ring *ring;
833 /* If a packet from the kernel to userspace was freed because of an
834 * error without being delivered to userspace, the kernel must reset
835 * the status. In the direction userspace to kernel, the status is
836 * always reset here after the packet was processed and freed.
838 if (netlink_skb_is_mmaped(skb)) {
839 hdr = netlink_mmap_hdr(skb);
840 sk = NETLINK_CB(skb).sk;
842 if (NETLINK_CB(skb).flags & NETLINK_SKB_TX) {
843 netlink_set_status(hdr, NL_MMAP_STATUS_UNUSED);
844 ring = &nlk_sk(sk)->tx_ring;
846 if (!(NETLINK_CB(skb).flags & NETLINK_SKB_DELIVERED)) {
848 netlink_set_status(hdr, NL_MMAP_STATUS_VALID);
850 ring = &nlk_sk(sk)->rx_ring;
853 WARN_ON(atomic_read(&ring->pending) == 0);
854 atomic_dec(&ring->pending);
860 if (is_vmalloc_addr(skb->head)) {
862 !atomic_dec_return(&(skb_shinfo(skb)->dataref)))
871 static void netlink_skb_set_owner_r(struct sk_buff *skb, struct sock *sk)
873 WARN_ON(skb->sk != NULL);
875 skb->destructor = netlink_skb_destructor;
876 atomic_add(skb->truesize, &sk->sk_rmem_alloc);
877 sk_mem_charge(sk, skb->truesize);
880 static void netlink_sock_destruct(struct sock *sk)
882 struct netlink_sock *nlk = nlk_sk(sk);
884 if (nlk->cb_running) {
886 nlk->cb.done(&nlk->cb);
888 module_put(nlk->cb.module);
889 kfree_skb(nlk->cb.skb);
892 skb_queue_purge(&sk->sk_receive_queue);
893 #ifdef CONFIG_NETLINK_MMAP
895 struct nl_mmap_req req;
897 memset(&req, 0, sizeof(req));
898 if (nlk->rx_ring.pg_vec)
899 netlink_set_ring(sk, &req, true, false);
900 memset(&req, 0, sizeof(req));
901 if (nlk->tx_ring.pg_vec)
902 netlink_set_ring(sk, &req, true, true);
904 #endif /* CONFIG_NETLINK_MMAP */
906 if (!sock_flag(sk, SOCK_DEAD)) {
907 printk(KERN_ERR "Freeing alive netlink socket %p\n", sk);
911 WARN_ON(atomic_read(&sk->sk_rmem_alloc));
912 WARN_ON(atomic_read(&sk->sk_wmem_alloc));
913 WARN_ON(nlk_sk(sk)->groups);
916 /* This lock without WQ_FLAG_EXCLUSIVE is good on UP and it is _very_ bad on
917 * SMP. Look, when several writers sleep and reader wakes them up, all but one
918 * immediately hit write lock and grab all the cpus. Exclusive sleep solves
919 * this, _but_ remember, it adds useless work on UP machines.
922 void netlink_table_grab(void)
923 __acquires(nl_table_lock)
927 write_lock_irq(&nl_table_lock);
929 if (atomic_read(&nl_table_users)) {
930 DECLARE_WAITQUEUE(wait, current);
932 add_wait_queue_exclusive(&nl_table_wait, &wait);
934 set_current_state(TASK_UNINTERRUPTIBLE);
935 if (atomic_read(&nl_table_users) == 0)
937 write_unlock_irq(&nl_table_lock);
939 write_lock_irq(&nl_table_lock);
942 __set_current_state(TASK_RUNNING);
943 remove_wait_queue(&nl_table_wait, &wait);
947 void netlink_table_ungrab(void)
948 __releases(nl_table_lock)
950 write_unlock_irq(&nl_table_lock);
951 wake_up(&nl_table_wait);
955 netlink_lock_table(void)
957 /* read_lock() synchronizes us to netlink_table_grab */
959 read_lock(&nl_table_lock);
960 atomic_inc(&nl_table_users);
961 read_unlock(&nl_table_lock);
965 netlink_unlock_table(void)
967 if (atomic_dec_and_test(&nl_table_users))
968 wake_up(&nl_table_wait);
971 struct netlink_compare_arg
977 static bool netlink_compare(void *ptr, void *arg)
979 struct netlink_compare_arg *x = arg;
980 struct sock *sk = ptr;
982 return nlk_sk(sk)->portid == x->portid &&
983 net_eq(sock_net(sk), x->net);
986 static struct sock *__netlink_lookup(struct netlink_table *table, u32 portid,
989 struct netlink_compare_arg arg = {
994 return rhashtable_lookup_compare(&table->hash, &portid,
995 &netlink_compare, &arg);
998 static bool __netlink_insert(struct netlink_table *table, struct sock *sk)
1000 struct netlink_compare_arg arg = {
1001 .net = sock_net(sk),
1002 .portid = nlk_sk(sk)->portid,
1005 return rhashtable_lookup_compare_insert(&table->hash,
1007 &netlink_compare, &arg);
1010 static struct sock *netlink_lookup(struct net *net, int protocol, u32 portid)
1012 struct netlink_table *table = &nl_table[protocol];
1016 sk = __netlink_lookup(table, portid, net);
1024 static const struct proto_ops netlink_ops;
1027 netlink_update_listeners(struct sock *sk)
1029 struct netlink_table *tbl = &nl_table[sk->sk_protocol];
1032 struct listeners *listeners;
1034 listeners = nl_deref_protected(tbl->listeners);
1038 for (i = 0; i < NLGRPLONGS(tbl->groups); i++) {
1040 sk_for_each_bound(sk, &tbl->mc_list) {
1041 if (i < NLGRPLONGS(nlk_sk(sk)->ngroups))
1042 mask |= nlk_sk(sk)->groups[i];
1044 listeners->masks[i] = mask;
1046 /* this function is only called with the netlink table "grabbed", which
1047 * makes sure updates are visible before bind or setsockopt return. */
1050 static int netlink_insert(struct sock *sk, u32 portid)
1052 struct netlink_table *table = &nl_table[sk->sk_protocol];
1058 if (nlk_sk(sk)->portid)
1062 if (BITS_PER_LONG > 32 &&
1063 unlikely(atomic_read(&table->hash.nelems) >= UINT_MAX))
1066 nlk_sk(sk)->portid = portid;
1070 if (!__netlink_insert(table, sk)) {
1080 static void netlink_remove(struct sock *sk)
1082 struct netlink_table *table;
1084 table = &nl_table[sk->sk_protocol];
1085 if (rhashtable_remove(&table->hash, &nlk_sk(sk)->node)) {
1086 WARN_ON(atomic_read(&sk->sk_refcnt) == 1);
1090 netlink_table_grab();
1091 if (nlk_sk(sk)->subscriptions) {
1092 __sk_del_bind_node(sk);
1093 netlink_update_listeners(sk);
1095 if (sk->sk_protocol == NETLINK_GENERIC)
1096 atomic_inc(&genl_sk_destructing_cnt);
1097 netlink_table_ungrab();
1100 static struct proto netlink_proto = {
1102 .owner = THIS_MODULE,
1103 .obj_size = sizeof(struct netlink_sock),
1106 static int __netlink_create(struct net *net, struct socket *sock,
1107 struct mutex *cb_mutex, int protocol)
1110 struct netlink_sock *nlk;
1112 sock->ops = &netlink_ops;
1114 sk = sk_alloc(net, PF_NETLINK, GFP_KERNEL, &netlink_proto);
1118 sock_init_data(sock, sk);
1122 nlk->cb_mutex = cb_mutex;
1124 nlk->cb_mutex = &nlk->cb_def_mutex;
1125 mutex_init(nlk->cb_mutex);
1127 init_waitqueue_head(&nlk->wait);
1128 #ifdef CONFIG_NETLINK_MMAP
1129 mutex_init(&nlk->pg_vec_lock);
1132 sk->sk_destruct = netlink_sock_destruct;
1133 sk->sk_protocol = protocol;
1137 static int netlink_create(struct net *net, struct socket *sock, int protocol,
1140 struct module *module = NULL;
1141 struct mutex *cb_mutex;
1142 struct netlink_sock *nlk;
1143 int (*bind)(struct net *net, int group);
1144 void (*unbind)(struct net *net, int group);
1147 sock->state = SS_UNCONNECTED;
1149 if (sock->type != SOCK_RAW && sock->type != SOCK_DGRAM)
1150 return -ESOCKTNOSUPPORT;
1152 if (protocol < 0 || protocol >= MAX_LINKS)
1153 return -EPROTONOSUPPORT;
1155 netlink_lock_table();
1156 #ifdef CONFIG_MODULES
1157 if (!nl_table[protocol].registered) {
1158 netlink_unlock_table();
1159 request_module("net-pf-%d-proto-%d", PF_NETLINK, protocol);
1160 netlink_lock_table();
1163 if (nl_table[protocol].registered &&
1164 try_module_get(nl_table[protocol].module))
1165 module = nl_table[protocol].module;
1167 err = -EPROTONOSUPPORT;
1168 cb_mutex = nl_table[protocol].cb_mutex;
1169 bind = nl_table[protocol].bind;
1170 unbind = nl_table[protocol].unbind;
1171 netlink_unlock_table();
1176 err = __netlink_create(net, sock, cb_mutex, protocol);
1181 sock_prot_inuse_add(net, &netlink_proto, 1);
1184 nlk = nlk_sk(sock->sk);
1185 nlk->module = module;
1186 nlk->netlink_bind = bind;
1187 nlk->netlink_unbind = unbind;
1196 static void deferred_put_nlk_sk(struct rcu_head *head)
1198 struct netlink_sock *nlk = container_of(head, struct netlink_sock, rcu);
1203 static int netlink_release(struct socket *sock)
1205 struct sock *sk = sock->sk;
1206 struct netlink_sock *nlk;
1216 * OK. Socket is unlinked, any packets that arrive now
1220 /* must not acquire netlink_table_lock in any way again before unbind
1221 * and notifying genetlink is done as otherwise it might deadlock
1223 if (nlk->netlink_unbind) {
1226 for (i = 0; i < nlk->ngroups; i++)
1227 if (test_bit(i, nlk->groups))
1228 nlk->netlink_unbind(sock_net(sk), i + 1);
1230 if (sk->sk_protocol == NETLINK_GENERIC &&
1231 atomic_dec_return(&genl_sk_destructing_cnt) == 0)
1232 wake_up(&genl_sk_destructing_waitq);
1235 wake_up_interruptible_all(&nlk->wait);
1237 skb_queue_purge(&sk->sk_write_queue);
1240 struct netlink_notify n = {
1241 .net = sock_net(sk),
1242 .protocol = sk->sk_protocol,
1243 .portid = nlk->portid,
1245 atomic_notifier_call_chain(&netlink_chain,
1246 NETLINK_URELEASE, &n);
1249 module_put(nlk->module);
1251 if (netlink_is_kernel(sk)) {
1252 netlink_table_grab();
1253 BUG_ON(nl_table[sk->sk_protocol].registered == 0);
1254 if (--nl_table[sk->sk_protocol].registered == 0) {
1255 struct listeners *old;
1257 old = nl_deref_protected(nl_table[sk->sk_protocol].listeners);
1258 RCU_INIT_POINTER(nl_table[sk->sk_protocol].listeners, NULL);
1259 kfree_rcu(old, rcu);
1260 nl_table[sk->sk_protocol].module = NULL;
1261 nl_table[sk->sk_protocol].bind = NULL;
1262 nl_table[sk->sk_protocol].unbind = NULL;
1263 nl_table[sk->sk_protocol].flags = 0;
1264 nl_table[sk->sk_protocol].registered = 0;
1266 netlink_table_ungrab();
1273 sock_prot_inuse_add(sock_net(sk), &netlink_proto, -1);
1275 call_rcu(&nlk->rcu, deferred_put_nlk_sk);
1279 static int netlink_autobind(struct socket *sock)
1281 struct sock *sk = sock->sk;
1282 struct net *net = sock_net(sk);
1283 struct netlink_table *table = &nl_table[sk->sk_protocol];
1284 s32 portid = task_tgid_vnr(current);
1286 static s32 rover = -4097;
1291 if (__netlink_lookup(table, portid, net)) {
1292 /* Bind collision, search negative portid values. */
1301 err = netlink_insert(sk, portid);
1302 if (err == -EADDRINUSE)
1305 /* If 2 threads race to autobind, that is fine. */
1313 * __netlink_ns_capable - General netlink message capability test
1314 * @nsp: NETLINK_CB of the socket buffer holding a netlink command from userspace.
1315 * @user_ns: The user namespace of the capability to use
1316 * @cap: The capability to use
1318 * Test to see if the opener of the socket we received the message
1319 * from had when the netlink socket was created and the sender of the
1320 * message has has the capability @cap in the user namespace @user_ns.
1322 bool __netlink_ns_capable(const struct netlink_skb_parms *nsp,
1323 struct user_namespace *user_ns, int cap)
1325 return ((nsp->flags & NETLINK_SKB_DST) ||
1326 file_ns_capable(nsp->sk->sk_socket->file, user_ns, cap)) &&
1327 ns_capable(user_ns, cap);
1329 EXPORT_SYMBOL(__netlink_ns_capable);
1332 * netlink_ns_capable - General netlink message capability test
1333 * @skb: socket buffer holding a netlink command from userspace
1334 * @user_ns: The user namespace of the capability to use
1335 * @cap: The capability to use
1337 * Test to see if the opener of the socket we received the message
1338 * from had when the netlink socket was created and the sender of the
1339 * message has has the capability @cap in the user namespace @user_ns.
1341 bool netlink_ns_capable(const struct sk_buff *skb,
1342 struct user_namespace *user_ns, int cap)
1344 return __netlink_ns_capable(&NETLINK_CB(skb), user_ns, cap);
1346 EXPORT_SYMBOL(netlink_ns_capable);
1349 * netlink_capable - Netlink global message capability test
1350 * @skb: socket buffer holding a netlink command from userspace
1351 * @cap: The capability to use
1353 * Test to see if the opener of the socket we received the message
1354 * from had when the netlink socket was created and the sender of the
1355 * message has has the capability @cap in all user namespaces.
1357 bool netlink_capable(const struct sk_buff *skb, int cap)
1359 return netlink_ns_capable(skb, &init_user_ns, cap);
1361 EXPORT_SYMBOL(netlink_capable);
1364 * netlink_net_capable - Netlink network namespace message capability test
1365 * @skb: socket buffer holding a netlink command from userspace
1366 * @cap: The capability to use
1368 * Test to see if the opener of the socket we received the message
1369 * from had when the netlink socket was created and the sender of the
1370 * message has has the capability @cap over the network namespace of
1371 * the socket we received the message from.
1373 bool netlink_net_capable(const struct sk_buff *skb, int cap)
1375 return netlink_ns_capable(skb, sock_net(skb->sk)->user_ns, cap);
1377 EXPORT_SYMBOL(netlink_net_capable);
1379 static inline int netlink_allowed(const struct socket *sock, unsigned int flag)
1381 return (nl_table[sock->sk->sk_protocol].flags & flag) ||
1382 ns_capable(sock_net(sock->sk)->user_ns, CAP_NET_ADMIN);
1386 netlink_update_subscriptions(struct sock *sk, unsigned int subscriptions)
1388 struct netlink_sock *nlk = nlk_sk(sk);
1390 if (nlk->subscriptions && !subscriptions)
1391 __sk_del_bind_node(sk);
1392 else if (!nlk->subscriptions && subscriptions)
1393 sk_add_bind_node(sk, &nl_table[sk->sk_protocol].mc_list);
1394 nlk->subscriptions = subscriptions;
1397 static int netlink_realloc_groups(struct sock *sk)
1399 struct netlink_sock *nlk = nlk_sk(sk);
1400 unsigned int groups;
1401 unsigned long *new_groups;
1404 netlink_table_grab();
1406 groups = nl_table[sk->sk_protocol].groups;
1407 if (!nl_table[sk->sk_protocol].registered) {
1412 if (nlk->ngroups >= groups)
1415 new_groups = krealloc(nlk->groups, NLGRPSZ(groups), GFP_ATOMIC);
1416 if (new_groups == NULL) {
1420 memset((char *)new_groups + NLGRPSZ(nlk->ngroups), 0,
1421 NLGRPSZ(groups) - NLGRPSZ(nlk->ngroups));
1423 nlk->groups = new_groups;
1424 nlk->ngroups = groups;
1426 netlink_table_ungrab();
1430 static void netlink_undo_bind(int group, long unsigned int groups,
1433 struct netlink_sock *nlk = nlk_sk(sk);
1436 if (!nlk->netlink_unbind)
1439 for (undo = 0; undo < group; undo++)
1440 if (test_bit(undo, &groups))
1441 nlk->netlink_unbind(sock_net(sk), undo + 1);
1444 static int netlink_bind(struct socket *sock, struct sockaddr *addr,
1447 struct sock *sk = sock->sk;
1448 struct net *net = sock_net(sk);
1449 struct netlink_sock *nlk = nlk_sk(sk);
1450 struct sockaddr_nl *nladdr = (struct sockaddr_nl *)addr;
1452 long unsigned int groups = nladdr->nl_groups;
1454 if (addr_len < sizeof(struct sockaddr_nl))
1457 if (nladdr->nl_family != AF_NETLINK)
1460 /* Only superuser is allowed to listen multicasts */
1462 if (!netlink_allowed(sock, NL_CFG_F_NONROOT_RECV))
1464 err = netlink_realloc_groups(sk);
1470 if (nladdr->nl_pid != nlk->portid)
1473 if (nlk->netlink_bind && groups) {
1476 for (group = 0; group < nlk->ngroups; group++) {
1477 if (!test_bit(group, &groups))
1479 err = nlk->netlink_bind(net, group + 1);
1482 netlink_undo_bind(group, groups, sk);
1488 err = nladdr->nl_pid ?
1489 netlink_insert(sk, nladdr->nl_pid) :
1490 netlink_autobind(sock);
1492 netlink_undo_bind(nlk->ngroups, groups, sk);
1497 if (!groups && (nlk->groups == NULL || !(u32)nlk->groups[0]))
1500 netlink_table_grab();
1501 netlink_update_subscriptions(sk, nlk->subscriptions +
1503 hweight32(nlk->groups[0]));
1504 nlk->groups[0] = (nlk->groups[0] & ~0xffffffffUL) | groups;
1505 netlink_update_listeners(sk);
1506 netlink_table_ungrab();
1511 static int netlink_connect(struct socket *sock, struct sockaddr *addr,
1512 int alen, int flags)
1515 struct sock *sk = sock->sk;
1516 struct netlink_sock *nlk = nlk_sk(sk);
1517 struct sockaddr_nl *nladdr = (struct sockaddr_nl *)addr;
1519 if (alen < sizeof(addr->sa_family))
1522 if (addr->sa_family == AF_UNSPEC) {
1523 sk->sk_state = NETLINK_UNCONNECTED;
1524 nlk->dst_portid = 0;
1528 if (addr->sa_family != AF_NETLINK)
1531 if ((nladdr->nl_groups || nladdr->nl_pid) &&
1532 !netlink_allowed(sock, NL_CFG_F_NONROOT_SEND))
1536 err = netlink_autobind(sock);
1539 sk->sk_state = NETLINK_CONNECTED;
1540 nlk->dst_portid = nladdr->nl_pid;
1541 nlk->dst_group = ffs(nladdr->nl_groups);
1547 static int netlink_getname(struct socket *sock, struct sockaddr *addr,
1548 int *addr_len, int peer)
1550 struct sock *sk = sock->sk;
1551 struct netlink_sock *nlk = nlk_sk(sk);
1552 DECLARE_SOCKADDR(struct sockaddr_nl *, nladdr, addr);
1554 nladdr->nl_family = AF_NETLINK;
1556 *addr_len = sizeof(*nladdr);
1559 nladdr->nl_pid = nlk->dst_portid;
1560 nladdr->nl_groups = netlink_group_mask(nlk->dst_group);
1562 nladdr->nl_pid = nlk->portid;
1563 nladdr->nl_groups = nlk->groups ? nlk->groups[0] : 0;
1568 static struct sock *netlink_getsockbyportid(struct sock *ssk, u32 portid)
1571 struct netlink_sock *nlk;
1573 sock = netlink_lookup(sock_net(ssk), ssk->sk_protocol, portid);
1575 return ERR_PTR(-ECONNREFUSED);
1577 /* Don't bother queuing skb if kernel socket has no input function */
1579 if (sock->sk_state == NETLINK_CONNECTED &&
1580 nlk->dst_portid != nlk_sk(ssk)->portid) {
1582 return ERR_PTR(-ECONNREFUSED);
1587 struct sock *netlink_getsockbyfilp(struct file *filp)
1589 struct inode *inode = file_inode(filp);
1592 if (!S_ISSOCK(inode->i_mode))
1593 return ERR_PTR(-ENOTSOCK);
1595 sock = SOCKET_I(inode)->sk;
1596 if (sock->sk_family != AF_NETLINK)
1597 return ERR_PTR(-EINVAL);
1603 static struct sk_buff *netlink_alloc_large_skb(unsigned int size,
1606 struct sk_buff *skb;
1609 if (size <= NLMSG_GOODSIZE || broadcast)
1610 return alloc_skb(size, GFP_KERNEL);
1612 size = SKB_DATA_ALIGN(size) +
1613 SKB_DATA_ALIGN(sizeof(struct skb_shared_info));
1615 data = vmalloc(size);
1619 skb = build_skb(data, size);
1624 skb->destructor = netlink_skb_destructor;
1631 * Attach a skb to a netlink socket.
1632 * The caller must hold a reference to the destination socket. On error, the
1633 * reference is dropped. The skb is not send to the destination, just all
1634 * all error checks are performed and memory in the queue is reserved.
1636 * < 0: error. skb freed, reference to sock dropped.
1638 * 1: repeat lookup - reference dropped while waiting for socket memory.
1640 int netlink_attachskb(struct sock *sk, struct sk_buff *skb,
1641 long *timeo, struct sock *ssk)
1643 struct netlink_sock *nlk;
1647 if ((atomic_read(&sk->sk_rmem_alloc) > sk->sk_rcvbuf ||
1648 test_bit(NETLINK_CONGESTED, &nlk->state)) &&
1649 !netlink_skb_is_mmaped(skb)) {
1650 DECLARE_WAITQUEUE(wait, current);
1652 if (!ssk || netlink_is_kernel(ssk))
1653 netlink_overrun(sk);
1659 __set_current_state(TASK_INTERRUPTIBLE);
1660 add_wait_queue(&nlk->wait, &wait);
1662 if ((atomic_read(&sk->sk_rmem_alloc) > sk->sk_rcvbuf ||
1663 test_bit(NETLINK_CONGESTED, &nlk->state)) &&
1664 !sock_flag(sk, SOCK_DEAD))
1665 *timeo = schedule_timeout(*timeo);
1667 __set_current_state(TASK_RUNNING);
1668 remove_wait_queue(&nlk->wait, &wait);
1671 if (signal_pending(current)) {
1673 return sock_intr_errno(*timeo);
1677 netlink_skb_set_owner_r(skb, sk);
1681 static int __netlink_sendskb(struct sock *sk, struct sk_buff *skb)
1685 netlink_deliver_tap(skb);
1687 #ifdef CONFIG_NETLINK_MMAP
1688 if (netlink_skb_is_mmaped(skb))
1689 netlink_queue_mmaped_skb(sk, skb);
1690 else if (netlink_rx_is_mmaped(sk))
1691 netlink_ring_set_copied(sk, skb);
1693 #endif /* CONFIG_NETLINK_MMAP */
1694 skb_queue_tail(&sk->sk_receive_queue, skb);
1695 sk->sk_data_ready(sk);
1699 int netlink_sendskb(struct sock *sk, struct sk_buff *skb)
1701 int len = __netlink_sendskb(sk, skb);
1707 void netlink_detachskb(struct sock *sk, struct sk_buff *skb)
1713 static struct sk_buff *netlink_trim(struct sk_buff *skb, gfp_t allocation)
1717 WARN_ON(skb->sk != NULL);
1718 if (netlink_skb_is_mmaped(skb))
1721 delta = skb->end - skb->tail;
1722 if (is_vmalloc_addr(skb->head) || delta * 2 < skb->truesize)
1725 if (skb_shared(skb)) {
1726 struct sk_buff *nskb = skb_clone(skb, allocation);
1733 if (!pskb_expand_head(skb, 0, -delta, allocation))
1734 skb->truesize -= delta;
1739 static int netlink_unicast_kernel(struct sock *sk, struct sk_buff *skb,
1743 struct netlink_sock *nlk = nlk_sk(sk);
1745 ret = -ECONNREFUSED;
1746 if (nlk->netlink_rcv != NULL) {
1748 netlink_skb_set_owner_r(skb, sk);
1749 NETLINK_CB(skb).sk = ssk;
1750 netlink_deliver_tap_kernel(sk, ssk, skb);
1751 nlk->netlink_rcv(skb);
1760 int netlink_unicast(struct sock *ssk, struct sk_buff *skb,
1761 u32 portid, int nonblock)
1767 skb = netlink_trim(skb, gfp_any());
1769 timeo = sock_sndtimeo(ssk, nonblock);
1771 sk = netlink_getsockbyportid(ssk, portid);
1776 if (netlink_is_kernel(sk))
1777 return netlink_unicast_kernel(sk, skb, ssk);
1779 if (sk_filter(sk, skb)) {
1786 err = netlink_attachskb(sk, skb, &timeo, ssk);
1792 return netlink_sendskb(sk, skb);
1794 EXPORT_SYMBOL(netlink_unicast);
1796 struct sk_buff *netlink_alloc_skb(struct sock *ssk, unsigned int size,
1797 u32 dst_portid, gfp_t gfp_mask)
1799 #ifdef CONFIG_NETLINK_MMAP
1800 struct sock *sk = NULL;
1801 struct sk_buff *skb;
1802 struct netlink_ring *ring;
1803 struct nl_mmap_hdr *hdr;
1804 unsigned int maxlen;
1806 sk = netlink_getsockbyportid(ssk, dst_portid);
1810 ring = &nlk_sk(sk)->rx_ring;
1811 /* fast-path without atomic ops for common case: non-mmaped receiver */
1812 if (ring->pg_vec == NULL)
1815 if (ring->frame_size - NL_MMAP_HDRLEN < size)
1818 skb = alloc_skb_head(gfp_mask);
1822 spin_lock_bh(&sk->sk_receive_queue.lock);
1823 /* check again under lock */
1824 if (ring->pg_vec == NULL)
1827 /* check again under lock */
1828 maxlen = ring->frame_size - NL_MMAP_HDRLEN;
1832 netlink_forward_ring(ring);
1833 hdr = netlink_current_frame(ring, NL_MMAP_STATUS_UNUSED);
1836 netlink_ring_setup_skb(skb, sk, ring, hdr);
1837 netlink_set_status(hdr, NL_MMAP_STATUS_RESERVED);
1838 atomic_inc(&ring->pending);
1839 netlink_increment_head(ring);
1841 spin_unlock_bh(&sk->sk_receive_queue.lock);
1846 spin_unlock_bh(&sk->sk_receive_queue.lock);
1847 netlink_overrun(sk);
1854 spin_unlock_bh(&sk->sk_receive_queue.lock);
1859 return alloc_skb(size, gfp_mask);
1861 EXPORT_SYMBOL_GPL(netlink_alloc_skb);
1863 int netlink_has_listeners(struct sock *sk, unsigned int group)
1866 struct listeners *listeners;
1868 BUG_ON(!netlink_is_kernel(sk));
1871 listeners = rcu_dereference(nl_table[sk->sk_protocol].listeners);
1873 if (listeners && group - 1 < nl_table[sk->sk_protocol].groups)
1874 res = test_bit(group - 1, listeners->masks);
1880 EXPORT_SYMBOL_GPL(netlink_has_listeners);
1882 static int netlink_broadcast_deliver(struct sock *sk, struct sk_buff *skb)
1884 struct netlink_sock *nlk = nlk_sk(sk);
1886 if (atomic_read(&sk->sk_rmem_alloc) <= sk->sk_rcvbuf &&
1887 !test_bit(NETLINK_CONGESTED, &nlk->state)) {
1888 netlink_skb_set_owner_r(skb, sk);
1889 __netlink_sendskb(sk, skb);
1890 return atomic_read(&sk->sk_rmem_alloc) > (sk->sk_rcvbuf >> 1);
1895 struct netlink_broadcast_data {
1896 struct sock *exclude_sk;
1901 int delivery_failure;
1905 struct sk_buff *skb, *skb2;
1906 int (*tx_filter)(struct sock *dsk, struct sk_buff *skb, void *data);
1910 static void do_one_broadcast(struct sock *sk,
1911 struct netlink_broadcast_data *p)
1913 struct netlink_sock *nlk = nlk_sk(sk);
1916 if (p->exclude_sk == sk)
1919 if (nlk->portid == p->portid || p->group - 1 >= nlk->ngroups ||
1920 !test_bit(p->group - 1, nlk->groups))
1923 if (!net_eq(sock_net(sk), p->net))
1927 netlink_overrun(sk);
1932 if (p->skb2 == NULL) {
1933 if (skb_shared(p->skb)) {
1934 p->skb2 = skb_clone(p->skb, p->allocation);
1936 p->skb2 = skb_get(p->skb);
1938 * skb ownership may have been set when
1939 * delivered to a previous socket.
1941 skb_orphan(p->skb2);
1944 if (p->skb2 == NULL) {
1945 netlink_overrun(sk);
1946 /* Clone failed. Notify ALL listeners. */
1948 if (nlk->flags & NETLINK_BROADCAST_SEND_ERROR)
1949 p->delivery_failure = 1;
1950 } else if (p->tx_filter && p->tx_filter(sk, p->skb2, p->tx_data)) {
1953 } else if (sk_filter(sk, p->skb2)) {
1956 } else if ((val = netlink_broadcast_deliver(sk, p->skb2)) < 0) {
1957 netlink_overrun(sk);
1958 if (nlk->flags & NETLINK_BROADCAST_SEND_ERROR)
1959 p->delivery_failure = 1;
1961 p->congested |= val;
1968 int netlink_broadcast_filtered(struct sock *ssk, struct sk_buff *skb, u32 portid,
1969 u32 group, gfp_t allocation,
1970 int (*filter)(struct sock *dsk, struct sk_buff *skb, void *data),
1973 struct net *net = sock_net(ssk);
1974 struct netlink_broadcast_data info;
1977 skb = netlink_trim(skb, allocation);
1979 info.exclude_sk = ssk;
1981 info.portid = portid;
1984 info.delivery_failure = 0;
1987 info.allocation = allocation;
1990 info.tx_filter = filter;
1991 info.tx_data = filter_data;
1993 /* While we sleep in clone, do not allow to change socket list */
1995 netlink_lock_table();
1997 sk_for_each_bound(sk, &nl_table[ssk->sk_protocol].mc_list)
1998 do_one_broadcast(sk, &info);
2002 netlink_unlock_table();
2004 if (info.delivery_failure) {
2005 kfree_skb(info.skb2);
2008 consume_skb(info.skb2);
2010 if (info.delivered) {
2011 if (info.congested && (allocation & __GFP_WAIT))
2017 EXPORT_SYMBOL(netlink_broadcast_filtered);
2019 int netlink_broadcast(struct sock *ssk, struct sk_buff *skb, u32 portid,
2020 u32 group, gfp_t allocation)
2022 return netlink_broadcast_filtered(ssk, skb, portid, group, allocation,
2025 EXPORT_SYMBOL(netlink_broadcast);
2027 struct netlink_set_err_data {
2028 struct sock *exclude_sk;
2034 static int do_one_set_err(struct sock *sk, struct netlink_set_err_data *p)
2036 struct netlink_sock *nlk = nlk_sk(sk);
2039 if (sk == p->exclude_sk)
2042 if (!net_eq(sock_net(sk), sock_net(p->exclude_sk)))
2045 if (nlk->portid == p->portid || p->group - 1 >= nlk->ngroups ||
2046 !test_bit(p->group - 1, nlk->groups))
2049 if (p->code == ENOBUFS && nlk->flags & NETLINK_RECV_NO_ENOBUFS) {
2054 sk->sk_err = p->code;
2055 sk->sk_error_report(sk);
2061 * netlink_set_err - report error to broadcast listeners
2062 * @ssk: the kernel netlink socket, as returned by netlink_kernel_create()
2063 * @portid: the PORTID of a process that we want to skip (if any)
2064 * @group: the broadcast group that will notice the error
2065 * @code: error code, must be negative (as usual in kernelspace)
2067 * This function returns the number of broadcast listeners that have set the
2068 * NETLINK_RECV_NO_ENOBUFS socket option.
2070 int netlink_set_err(struct sock *ssk, u32 portid, u32 group, int code)
2072 struct netlink_set_err_data info;
2076 info.exclude_sk = ssk;
2077 info.portid = portid;
2079 /* sk->sk_err wants a positive error value */
2082 read_lock(&nl_table_lock);
2084 sk_for_each_bound(sk, &nl_table[ssk->sk_protocol].mc_list)
2085 ret += do_one_set_err(sk, &info);
2087 read_unlock(&nl_table_lock);
2090 EXPORT_SYMBOL(netlink_set_err);
2092 /* must be called with netlink table grabbed */
2093 static void netlink_update_socket_mc(struct netlink_sock *nlk,
2097 int old, new = !!is_new, subscriptions;
2099 old = test_bit(group - 1, nlk->groups);
2100 subscriptions = nlk->subscriptions - old + new;
2102 __set_bit(group - 1, nlk->groups);
2104 __clear_bit(group - 1, nlk->groups);
2105 netlink_update_subscriptions(&nlk->sk, subscriptions);
2106 netlink_update_listeners(&nlk->sk);
2109 static int netlink_setsockopt(struct socket *sock, int level, int optname,
2110 char __user *optval, unsigned int optlen)
2112 struct sock *sk = sock->sk;
2113 struct netlink_sock *nlk = nlk_sk(sk);
2114 unsigned int val = 0;
2117 if (level != SOL_NETLINK)
2118 return -ENOPROTOOPT;
2120 if (optname != NETLINK_RX_RING && optname != NETLINK_TX_RING &&
2121 optlen >= sizeof(int) &&
2122 get_user(val, (unsigned int __user *)optval))
2126 case NETLINK_PKTINFO:
2128 nlk->flags |= NETLINK_RECV_PKTINFO;
2130 nlk->flags &= ~NETLINK_RECV_PKTINFO;
2133 case NETLINK_ADD_MEMBERSHIP:
2134 case NETLINK_DROP_MEMBERSHIP: {
2135 if (!netlink_allowed(sock, NL_CFG_F_NONROOT_RECV))
2137 err = netlink_realloc_groups(sk);
2140 if (!val || val - 1 >= nlk->ngroups)
2142 if (optname == NETLINK_ADD_MEMBERSHIP && nlk->netlink_bind) {
2143 err = nlk->netlink_bind(sock_net(sk), val);
2147 netlink_table_grab();
2148 netlink_update_socket_mc(nlk, val,
2149 optname == NETLINK_ADD_MEMBERSHIP);
2150 netlink_table_ungrab();
2151 if (optname == NETLINK_DROP_MEMBERSHIP && nlk->netlink_unbind)
2152 nlk->netlink_unbind(sock_net(sk), val);
2157 case NETLINK_BROADCAST_ERROR:
2159 nlk->flags |= NETLINK_BROADCAST_SEND_ERROR;
2161 nlk->flags &= ~NETLINK_BROADCAST_SEND_ERROR;
2164 case NETLINK_NO_ENOBUFS:
2166 nlk->flags |= NETLINK_RECV_NO_ENOBUFS;
2167 clear_bit(NETLINK_CONGESTED, &nlk->state);
2168 wake_up_interruptible(&nlk->wait);
2170 nlk->flags &= ~NETLINK_RECV_NO_ENOBUFS;
2174 #ifdef CONFIG_NETLINK_MMAP
2175 case NETLINK_RX_RING:
2176 case NETLINK_TX_RING: {
2177 struct nl_mmap_req req;
2179 /* Rings might consume more memory than queue limits, require
2182 if (!capable(CAP_NET_ADMIN))
2184 if (optlen < sizeof(req))
2186 if (copy_from_user(&req, optval, sizeof(req)))
2188 err = netlink_set_ring(sk, &req, false,
2189 optname == NETLINK_TX_RING);
2192 #endif /* CONFIG_NETLINK_MMAP */
2199 static int netlink_getsockopt(struct socket *sock, int level, int optname,
2200 char __user *optval, int __user *optlen)
2202 struct sock *sk = sock->sk;
2203 struct netlink_sock *nlk = nlk_sk(sk);
2206 if (level != SOL_NETLINK)
2207 return -ENOPROTOOPT;
2209 if (get_user(len, optlen))
2215 case NETLINK_PKTINFO:
2216 if (len < sizeof(int))
2219 val = nlk->flags & NETLINK_RECV_PKTINFO ? 1 : 0;
2220 if (put_user(len, optlen) ||
2221 put_user(val, optval))
2225 case NETLINK_BROADCAST_ERROR:
2226 if (len < sizeof(int))
2229 val = nlk->flags & NETLINK_BROADCAST_SEND_ERROR ? 1 : 0;
2230 if (put_user(len, optlen) ||
2231 put_user(val, optval))
2235 case NETLINK_NO_ENOBUFS:
2236 if (len < sizeof(int))
2239 val = nlk->flags & NETLINK_RECV_NO_ENOBUFS ? 1 : 0;
2240 if (put_user(len, optlen) ||
2241 put_user(val, optval))
2251 static void netlink_cmsg_recv_pktinfo(struct msghdr *msg, struct sk_buff *skb)
2253 struct nl_pktinfo info;
2255 info.group = NETLINK_CB(skb).dst_group;
2256 put_cmsg(msg, SOL_NETLINK, NETLINK_PKTINFO, sizeof(info), &info);
2259 static int netlink_sendmsg(struct kiocb *kiocb, struct socket *sock,
2260 struct msghdr *msg, size_t len)
2262 struct sock *sk = sock->sk;
2263 struct netlink_sock *nlk = nlk_sk(sk);
2264 DECLARE_SOCKADDR(struct sockaddr_nl *, addr, msg->msg_name);
2267 struct sk_buff *skb;
2269 struct scm_cookie scm;
2270 u32 netlink_skb_flags = 0;
2272 if (msg->msg_flags&MSG_OOB)
2275 err = scm_send(sock, msg, &scm, true);
2279 if (msg->msg_namelen) {
2281 if (addr->nl_family != AF_NETLINK)
2283 dst_portid = addr->nl_pid;
2284 dst_group = ffs(addr->nl_groups);
2286 if ((dst_group || dst_portid) &&
2287 !netlink_allowed(sock, NL_CFG_F_NONROOT_SEND))
2289 netlink_skb_flags |= NETLINK_SKB_DST;
2291 dst_portid = nlk->dst_portid;
2292 dst_group = nlk->dst_group;
2296 err = netlink_autobind(sock);
2301 /* It's a really convoluted way for userland to ask for mmaped
2302 * sendmsg(), but that's what we've got...
2304 if (netlink_tx_is_mmaped(sk) &&
2305 msg->msg_iter.type == ITER_IOVEC &&
2306 msg->msg_iter.nr_segs == 1 &&
2307 msg->msg_iter.iov->iov_base == NULL) {
2308 err = netlink_mmap_sendmsg(sk, msg, dst_portid, dst_group,
2314 if (len > sk->sk_sndbuf - 32)
2317 skb = netlink_alloc_large_skb(len, dst_group);
2321 NETLINK_CB(skb).portid = nlk->portid;
2322 NETLINK_CB(skb).dst_group = dst_group;
2323 NETLINK_CB(skb).creds = scm.creds;
2324 NETLINK_CB(skb).flags = netlink_skb_flags;
2327 if (memcpy_from_msg(skb_put(skb, len), msg, len)) {
2332 err = security_netlink_send(sk, skb);
2339 atomic_inc(&skb->users);
2340 netlink_broadcast(sk, skb, dst_portid, dst_group, GFP_KERNEL);
2342 err = netlink_unicast(sk, skb, dst_portid, msg->msg_flags&MSG_DONTWAIT);
2349 static int netlink_recvmsg(struct kiocb *kiocb, struct socket *sock,
2350 struct msghdr *msg, size_t len,
2353 struct scm_cookie scm;
2354 struct sock *sk = sock->sk;
2355 struct netlink_sock *nlk = nlk_sk(sk);
2356 int noblock = flags&MSG_DONTWAIT;
2358 struct sk_buff *skb, *data_skb;
2366 skb = skb_recv_datagram(sk, flags, noblock, &err);
2372 #ifdef CONFIG_COMPAT_NETLINK_MESSAGES
2373 if (unlikely(skb_shinfo(skb)->frag_list)) {
2375 * If this skb has a frag_list, then here that means that we
2376 * will have to use the frag_list skb's data for compat tasks
2377 * and the regular skb's data for normal (non-compat) tasks.
2379 * If we need to send the compat skb, assign it to the
2380 * 'data_skb' variable so that it will be used below for data
2381 * copying. We keep 'skb' for everything else, including
2382 * freeing both later.
2384 if (flags & MSG_CMSG_COMPAT)
2385 data_skb = skb_shinfo(skb)->frag_list;
2389 /* Record the max length of recvmsg() calls for future allocations */
2390 nlk->max_recvmsg_len = max(nlk->max_recvmsg_len, len);
2391 nlk->max_recvmsg_len = min_t(size_t, nlk->max_recvmsg_len,
2394 copied = data_skb->len;
2396 msg->msg_flags |= MSG_TRUNC;
2400 skb_reset_transport_header(data_skb);
2401 err = skb_copy_datagram_msg(data_skb, 0, msg, copied);
2403 if (msg->msg_name) {
2404 DECLARE_SOCKADDR(struct sockaddr_nl *, addr, msg->msg_name);
2405 addr->nl_family = AF_NETLINK;
2407 addr->nl_pid = NETLINK_CB(skb).portid;
2408 addr->nl_groups = netlink_group_mask(NETLINK_CB(skb).dst_group);
2409 msg->msg_namelen = sizeof(*addr);
2412 if (nlk->flags & NETLINK_RECV_PKTINFO)
2413 netlink_cmsg_recv_pktinfo(msg, skb);
2415 memset(&scm, 0, sizeof(scm));
2416 scm.creds = *NETLINK_CREDS(skb);
2417 if (flags & MSG_TRUNC)
2418 copied = data_skb->len;
2420 skb_free_datagram(sk, skb);
2422 if (nlk->cb_running &&
2423 atomic_read(&sk->sk_rmem_alloc) <= sk->sk_rcvbuf / 2) {
2424 ret = netlink_dump(sk);
2427 sk->sk_error_report(sk);
2431 scm_recv(sock, msg, &scm, flags);
2433 netlink_rcv_wake(sk);
2434 return err ? : copied;
2437 static void netlink_data_ready(struct sock *sk)
2443 * We export these functions to other modules. They provide a
2444 * complete set of kernel non-blocking support for message
2449 __netlink_kernel_create(struct net *net, int unit, struct module *module,
2450 struct netlink_kernel_cfg *cfg)
2452 struct socket *sock;
2454 struct netlink_sock *nlk;
2455 struct listeners *listeners = NULL;
2456 struct mutex *cb_mutex = cfg ? cfg->cb_mutex : NULL;
2457 unsigned int groups;
2461 if (unit < 0 || unit >= MAX_LINKS)
2464 if (sock_create_lite(PF_NETLINK, SOCK_DGRAM, unit, &sock))
2468 * We have to just have a reference on the net from sk, but don't
2469 * get_net it. Besides, we cannot get and then put the net here.
2470 * So we create one inside init_net and the move it to net.
2473 if (__netlink_create(&init_net, sock, cb_mutex, unit) < 0)
2474 goto out_sock_release_nosk;
2477 sk_change_net(sk, net);
2479 if (!cfg || cfg->groups < 32)
2482 groups = cfg->groups;
2484 listeners = kzalloc(sizeof(*listeners) + NLGRPSZ(groups), GFP_KERNEL);
2486 goto out_sock_release;
2488 sk->sk_data_ready = netlink_data_ready;
2489 if (cfg && cfg->input)
2490 nlk_sk(sk)->netlink_rcv = cfg->input;
2492 if (netlink_insert(sk, 0))
2493 goto out_sock_release;
2496 nlk->flags |= NETLINK_KERNEL_SOCKET;
2498 netlink_table_grab();
2499 if (!nl_table[unit].registered) {
2500 nl_table[unit].groups = groups;
2501 rcu_assign_pointer(nl_table[unit].listeners, listeners);
2502 nl_table[unit].cb_mutex = cb_mutex;
2503 nl_table[unit].module = module;
2505 nl_table[unit].bind = cfg->bind;
2506 nl_table[unit].unbind = cfg->unbind;
2507 nl_table[unit].flags = cfg->flags;
2509 nl_table[unit].compare = cfg->compare;
2511 nl_table[unit].registered = 1;
2514 nl_table[unit].registered++;
2516 netlink_table_ungrab();
2521 netlink_kernel_release(sk);
2524 out_sock_release_nosk:
2528 EXPORT_SYMBOL(__netlink_kernel_create);
2531 netlink_kernel_release(struct sock *sk)
2533 sk_release_kernel(sk);
2535 EXPORT_SYMBOL(netlink_kernel_release);
2537 int __netlink_change_ngroups(struct sock *sk, unsigned int groups)
2539 struct listeners *new, *old;
2540 struct netlink_table *tbl = &nl_table[sk->sk_protocol];
2545 if (NLGRPSZ(tbl->groups) < NLGRPSZ(groups)) {
2546 new = kzalloc(sizeof(*new) + NLGRPSZ(groups), GFP_ATOMIC);
2549 old = nl_deref_protected(tbl->listeners);
2550 memcpy(new->masks, old->masks, NLGRPSZ(tbl->groups));
2551 rcu_assign_pointer(tbl->listeners, new);
2553 kfree_rcu(old, rcu);
2555 tbl->groups = groups;
2561 * netlink_change_ngroups - change number of multicast groups
2563 * This changes the number of multicast groups that are available
2564 * on a certain netlink family. Note that it is not possible to
2565 * change the number of groups to below 32. Also note that it does
2566 * not implicitly call netlink_clear_multicast_users() when the
2567 * number of groups is reduced.
2569 * @sk: The kernel netlink socket, as returned by netlink_kernel_create().
2570 * @groups: The new number of groups.
2572 int netlink_change_ngroups(struct sock *sk, unsigned int groups)
2576 netlink_table_grab();
2577 err = __netlink_change_ngroups(sk, groups);
2578 netlink_table_ungrab();
2583 void __netlink_clear_multicast_users(struct sock *ksk, unsigned int group)
2586 struct netlink_table *tbl = &nl_table[ksk->sk_protocol];
2588 sk_for_each_bound(sk, &tbl->mc_list)
2589 netlink_update_socket_mc(nlk_sk(sk), group, 0);
2593 __nlmsg_put(struct sk_buff *skb, u32 portid, u32 seq, int type, int len, int flags)
2595 struct nlmsghdr *nlh;
2596 int size = nlmsg_msg_size(len);
2598 nlh = (struct nlmsghdr *)skb_put(skb, NLMSG_ALIGN(size));
2599 nlh->nlmsg_type = type;
2600 nlh->nlmsg_len = size;
2601 nlh->nlmsg_flags = flags;
2602 nlh->nlmsg_pid = portid;
2603 nlh->nlmsg_seq = seq;
2604 if (!__builtin_constant_p(size) || NLMSG_ALIGN(size) - size != 0)
2605 memset(nlmsg_data(nlh) + len, 0, NLMSG_ALIGN(size) - size);
2608 EXPORT_SYMBOL(__nlmsg_put);
2611 * It looks a bit ugly.
2612 * It would be better to create kernel thread.
2615 static int netlink_dump(struct sock *sk)
2617 struct netlink_sock *nlk = nlk_sk(sk);
2618 struct netlink_callback *cb;
2619 struct sk_buff *skb = NULL;
2620 struct nlmsghdr *nlh;
2621 int len, err = -ENOBUFS;
2624 mutex_lock(nlk->cb_mutex);
2625 if (!nlk->cb_running) {
2631 alloc_size = max_t(int, cb->min_dump_alloc, NLMSG_GOODSIZE);
2633 if (!netlink_rx_is_mmaped(sk) &&
2634 atomic_read(&sk->sk_rmem_alloc) >= sk->sk_rcvbuf)
2637 /* NLMSG_GOODSIZE is small to avoid high order allocations being
2638 * required, but it makes sense to _attempt_ a 16K bytes allocation
2639 * to reduce number of system calls on dump operations, if user
2640 * ever provided a big enough buffer.
2642 if (alloc_size < nlk->max_recvmsg_len) {
2643 skb = netlink_alloc_skb(sk,
2644 nlk->max_recvmsg_len,
2649 /* available room should be exact amount to avoid MSG_TRUNC */
2651 skb_reserve(skb, skb_tailroom(skb) -
2652 nlk->max_recvmsg_len);
2655 skb = netlink_alloc_skb(sk, alloc_size, nlk->portid,
2659 netlink_skb_set_owner_r(skb, sk);
2661 len = cb->dump(skb, cb);
2664 mutex_unlock(nlk->cb_mutex);
2666 if (sk_filter(sk, skb))
2669 __netlink_sendskb(sk, skb);
2673 nlh = nlmsg_put_answer(skb, cb, NLMSG_DONE, sizeof(len), NLM_F_MULTI);
2677 nl_dump_check_consistent(cb, nlh);
2679 memcpy(nlmsg_data(nlh), &len, sizeof(len));
2681 if (sk_filter(sk, skb))
2684 __netlink_sendskb(sk, skb);
2689 nlk->cb_running = false;
2690 mutex_unlock(nlk->cb_mutex);
2691 module_put(cb->module);
2692 consume_skb(cb->skb);
2696 mutex_unlock(nlk->cb_mutex);
2701 int __netlink_dump_start(struct sock *ssk, struct sk_buff *skb,
2702 const struct nlmsghdr *nlh,
2703 struct netlink_dump_control *control)
2705 struct netlink_callback *cb;
2707 struct netlink_sock *nlk;
2710 /* Memory mapped dump requests need to be copied to avoid looping
2711 * on the pending state in netlink_mmap_sendmsg() while the CB hold
2712 * a reference to the skb.
2714 if (netlink_skb_is_mmaped(skb)) {
2715 skb = skb_copy(skb, GFP_KERNEL);
2719 atomic_inc(&skb->users);
2721 sk = netlink_lookup(sock_net(ssk), ssk->sk_protocol, NETLINK_CB(skb).portid);
2723 ret = -ECONNREFUSED;
2728 mutex_lock(nlk->cb_mutex);
2729 /* A dump is in progress... */
2730 if (nlk->cb_running) {
2734 /* add reference of module which cb->dump belongs to */
2735 if (!try_module_get(control->module)) {
2736 ret = -EPROTONOSUPPORT;
2741 memset(cb, 0, sizeof(*cb));
2742 cb->dump = control->dump;
2743 cb->done = control->done;
2745 cb->data = control->data;
2746 cb->module = control->module;
2747 cb->min_dump_alloc = control->min_dump_alloc;
2750 nlk->cb_running = true;
2752 mutex_unlock(nlk->cb_mutex);
2754 ret = netlink_dump(sk);
2760 /* We successfully started a dump, by returning -EINTR we
2761 * signal not to send ACK even if it was requested.
2767 mutex_unlock(nlk->cb_mutex);
2772 EXPORT_SYMBOL(__netlink_dump_start);
2774 void netlink_ack(struct sk_buff *in_skb, struct nlmsghdr *nlh, int err)
2776 struct sk_buff *skb;
2777 struct nlmsghdr *rep;
2778 struct nlmsgerr *errmsg;
2779 size_t payload = sizeof(*errmsg);
2781 /* error messages get the original request appened */
2783 payload += nlmsg_len(nlh);
2785 skb = netlink_alloc_skb(in_skb->sk, nlmsg_total_size(payload),
2786 NETLINK_CB(in_skb).portid, GFP_KERNEL);
2790 sk = netlink_lookup(sock_net(in_skb->sk),
2791 in_skb->sk->sk_protocol,
2792 NETLINK_CB(in_skb).portid);
2794 sk->sk_err = ENOBUFS;
2795 sk->sk_error_report(sk);
2801 rep = __nlmsg_put(skb, NETLINK_CB(in_skb).portid, nlh->nlmsg_seq,
2802 NLMSG_ERROR, payload, 0);
2803 errmsg = nlmsg_data(rep);
2804 errmsg->error = err;
2805 memcpy(&errmsg->msg, nlh, err ? nlh->nlmsg_len : sizeof(*nlh));
2806 netlink_unicast(in_skb->sk, skb, NETLINK_CB(in_skb).portid, MSG_DONTWAIT);
2808 EXPORT_SYMBOL(netlink_ack);
2810 int netlink_rcv_skb(struct sk_buff *skb, int (*cb)(struct sk_buff *,
2813 struct nlmsghdr *nlh;
2816 while (skb->len >= nlmsg_total_size(0)) {
2819 nlh = nlmsg_hdr(skb);
2822 if (nlh->nlmsg_len < NLMSG_HDRLEN || skb->len < nlh->nlmsg_len)
2825 /* Only requests are handled by the kernel */
2826 if (!(nlh->nlmsg_flags & NLM_F_REQUEST))
2829 /* Skip control messages */
2830 if (nlh->nlmsg_type < NLMSG_MIN_TYPE)
2838 if (nlh->nlmsg_flags & NLM_F_ACK || err)
2839 netlink_ack(skb, nlh, err);
2842 msglen = NLMSG_ALIGN(nlh->nlmsg_len);
2843 if (msglen > skb->len)
2845 skb_pull(skb, msglen);
2850 EXPORT_SYMBOL(netlink_rcv_skb);
2853 * nlmsg_notify - send a notification netlink message
2854 * @sk: netlink socket to use
2855 * @skb: notification message
2856 * @portid: destination netlink portid for reports or 0
2857 * @group: destination multicast group or 0
2858 * @report: 1 to report back, 0 to disable
2859 * @flags: allocation flags
2861 int nlmsg_notify(struct sock *sk, struct sk_buff *skb, u32 portid,
2862 unsigned int group, int report, gfp_t flags)
2867 int exclude_portid = 0;
2870 atomic_inc(&skb->users);
2871 exclude_portid = portid;
2874 /* errors reported via destination sk->sk_err, but propagate
2875 * delivery errors if NETLINK_BROADCAST_ERROR flag is set */
2876 err = nlmsg_multicast(sk, skb, exclude_portid, group, flags);
2882 err2 = nlmsg_unicast(sk, skb, portid);
2883 if (!err || err == -ESRCH)
2889 EXPORT_SYMBOL(nlmsg_notify);
2891 #ifdef CONFIG_PROC_FS
2892 struct nl_seq_iter {
2893 struct seq_net_private p;
2894 struct rhashtable_iter hti;
2898 static int netlink_walk_start(struct nl_seq_iter *iter)
2902 err = rhashtable_walk_init(&nl_table[iter->link].hash, &iter->hti);
2904 iter->link = MAX_LINKS;
2908 err = rhashtable_walk_start(&iter->hti);
2909 return err == -EAGAIN ? 0 : err;
2912 static void netlink_walk_stop(struct nl_seq_iter *iter)
2914 rhashtable_walk_stop(&iter->hti);
2915 rhashtable_walk_exit(&iter->hti);
2918 static void *__netlink_seq_next(struct seq_file *seq)
2920 struct nl_seq_iter *iter = seq->private;
2921 struct netlink_sock *nlk;
2927 nlk = rhashtable_walk_next(&iter->hti);
2930 if (PTR_ERR(nlk) == -EAGAIN)
2939 netlink_walk_stop(iter);
2940 if (++iter->link >= MAX_LINKS)
2943 err = netlink_walk_start(iter);
2945 return ERR_PTR(err);
2947 } while (sock_net(&nlk->sk) != seq_file_net(seq));
2952 static void *netlink_seq_start(struct seq_file *seq, loff_t *posp)
2954 struct nl_seq_iter *iter = seq->private;
2955 void *obj = SEQ_START_TOKEN;
2961 err = netlink_walk_start(iter);
2963 return ERR_PTR(err);
2965 for (pos = *posp; pos && obj && !IS_ERR(obj); pos--)
2966 obj = __netlink_seq_next(seq);
2971 static void *netlink_seq_next(struct seq_file *seq, void *v, loff_t *pos)
2974 return __netlink_seq_next(seq);
2977 static void netlink_seq_stop(struct seq_file *seq, void *v)
2979 struct nl_seq_iter *iter = seq->private;
2981 if (iter->link >= MAX_LINKS)
2984 netlink_walk_stop(iter);
2988 static int netlink_seq_show(struct seq_file *seq, void *v)
2990 if (v == SEQ_START_TOKEN) {
2992 "sk Eth Pid Groups "
2993 "Rmem Wmem Dump Locks Drops Inode\n");
2996 struct netlink_sock *nlk = nlk_sk(s);
2998 seq_printf(seq, "%pK %-3d %-6u %08x %-8d %-8d %d %-8d %-8d %-8lu\n",
3002 nlk->groups ? (u32)nlk->groups[0] : 0,
3003 sk_rmem_alloc_get(s),
3004 sk_wmem_alloc_get(s),
3006 atomic_read(&s->sk_refcnt),
3007 atomic_read(&s->sk_drops),
3015 static const struct seq_operations netlink_seq_ops = {
3016 .start = netlink_seq_start,
3017 .next = netlink_seq_next,
3018 .stop = netlink_seq_stop,
3019 .show = netlink_seq_show,
3023 static int netlink_seq_open(struct inode *inode, struct file *file)
3025 return seq_open_net(inode, file, &netlink_seq_ops,
3026 sizeof(struct nl_seq_iter));
3029 static const struct file_operations netlink_seq_fops = {
3030 .owner = THIS_MODULE,
3031 .open = netlink_seq_open,
3033 .llseek = seq_lseek,
3034 .release = seq_release_net,
3039 int netlink_register_notifier(struct notifier_block *nb)
3041 return atomic_notifier_chain_register(&netlink_chain, nb);
3043 EXPORT_SYMBOL(netlink_register_notifier);
3045 int netlink_unregister_notifier(struct notifier_block *nb)
3047 return atomic_notifier_chain_unregister(&netlink_chain, nb);
3049 EXPORT_SYMBOL(netlink_unregister_notifier);
3051 static const struct proto_ops netlink_ops = {
3052 .family = PF_NETLINK,
3053 .owner = THIS_MODULE,
3054 .release = netlink_release,
3055 .bind = netlink_bind,
3056 .connect = netlink_connect,
3057 .socketpair = sock_no_socketpair,
3058 .accept = sock_no_accept,
3059 .getname = netlink_getname,
3060 .poll = netlink_poll,
3061 .ioctl = sock_no_ioctl,
3062 .listen = sock_no_listen,
3063 .shutdown = sock_no_shutdown,
3064 .setsockopt = netlink_setsockopt,
3065 .getsockopt = netlink_getsockopt,
3066 .sendmsg = netlink_sendmsg,
3067 .recvmsg = netlink_recvmsg,
3068 .mmap = netlink_mmap,
3069 .sendpage = sock_no_sendpage,
3072 static const struct net_proto_family netlink_family_ops = {
3073 .family = PF_NETLINK,
3074 .create = netlink_create,
3075 .owner = THIS_MODULE, /* for consistency 8) */
3078 static int __net_init netlink_net_init(struct net *net)
3080 #ifdef CONFIG_PROC_FS
3081 if (!proc_create("netlink", 0, net->proc_net, &netlink_seq_fops))
3087 static void __net_exit netlink_net_exit(struct net *net)
3089 #ifdef CONFIG_PROC_FS
3090 remove_proc_entry("netlink", net->proc_net);
3094 static void __init netlink_add_usersock_entry(void)
3096 struct listeners *listeners;
3099 listeners = kzalloc(sizeof(*listeners) + NLGRPSZ(groups), GFP_KERNEL);
3101 panic("netlink_add_usersock_entry: Cannot allocate listeners\n");
3103 netlink_table_grab();
3105 nl_table[NETLINK_USERSOCK].groups = groups;
3106 rcu_assign_pointer(nl_table[NETLINK_USERSOCK].listeners, listeners);
3107 nl_table[NETLINK_USERSOCK].module = THIS_MODULE;
3108 nl_table[NETLINK_USERSOCK].registered = 1;
3109 nl_table[NETLINK_USERSOCK].flags = NL_CFG_F_NONROOT_SEND;
3111 netlink_table_ungrab();
3114 static struct pernet_operations __net_initdata netlink_net_ops = {
3115 .init = netlink_net_init,
3116 .exit = netlink_net_exit,
3119 static int __init netlink_proto_init(void)
3122 int err = proto_register(&netlink_proto, 0);
3123 struct rhashtable_params ht_params = {
3124 .head_offset = offsetof(struct netlink_sock, node),
3125 .key_offset = offsetof(struct netlink_sock, portid),
3126 .key_len = sizeof(u32), /* portid */
3128 .max_shift = 16, /* 64K */
3129 .grow_decision = rht_grow_above_75,
3130 .shrink_decision = rht_shrink_below_30,
3136 BUILD_BUG_ON(sizeof(struct netlink_skb_parms) > FIELD_SIZEOF(struct sk_buff, cb));
3138 nl_table = kcalloc(MAX_LINKS, sizeof(*nl_table), GFP_KERNEL);
3142 for (i = 0; i < MAX_LINKS; i++) {
3143 if (rhashtable_init(&nl_table[i].hash, &ht_params) < 0) {
3145 rhashtable_destroy(&nl_table[i].hash);
3151 INIT_LIST_HEAD(&netlink_tap_all);
3153 netlink_add_usersock_entry();
3155 sock_register(&netlink_family_ops);
3156 register_pernet_subsys(&netlink_net_ops);
3157 /* The netlink device handler may be needed early. */
3162 panic("netlink_init: Cannot allocate nl_table\n");
3165 core_initcall(netlink_proto_init);