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
88 static inline int netlink_is_kernel(struct sock *sk)
90 return nlk_sk(sk)->flags & NETLINK_F_KERNEL_SOCKET;
93 struct netlink_table *nl_table;
94 EXPORT_SYMBOL_GPL(nl_table);
96 static DECLARE_WAIT_QUEUE_HEAD(nl_table_wait);
98 static int netlink_dump(struct sock *sk);
99 static void netlink_skb_destructor(struct sk_buff *skb);
101 /* nl_table locking explained:
102 * Lookup and traversal are protected with an RCU read-side lock. Insertion
103 * and removal are protected with per bucket lock while using RCU list
104 * modification primitives and may run in parallel to RCU protected lookups.
105 * Destruction of the Netlink socket may only occur *after* nl_table_lock has
106 * been acquired * either during or after the socket has been removed from
107 * the list and after an RCU grace period.
109 DEFINE_RWLOCK(nl_table_lock);
110 EXPORT_SYMBOL_GPL(nl_table_lock);
111 static atomic_t nl_table_users = ATOMIC_INIT(0);
113 #define nl_deref_protected(X) rcu_dereference_protected(X, lockdep_is_held(&nl_table_lock));
115 static ATOMIC_NOTIFIER_HEAD(netlink_chain);
117 static DEFINE_SPINLOCK(netlink_tap_lock);
118 static struct list_head netlink_tap_all __read_mostly;
120 static const struct rhashtable_params netlink_rhashtable_params;
122 static inline u32 netlink_group_mask(u32 group)
124 return group ? 1 << (group - 1) : 0;
127 int netlink_add_tap(struct netlink_tap *nt)
129 if (unlikely(nt->dev->type != ARPHRD_NETLINK))
132 spin_lock(&netlink_tap_lock);
133 list_add_rcu(&nt->list, &netlink_tap_all);
134 spin_unlock(&netlink_tap_lock);
136 __module_get(nt->module);
140 EXPORT_SYMBOL_GPL(netlink_add_tap);
142 static int __netlink_remove_tap(struct netlink_tap *nt)
145 struct netlink_tap *tmp;
147 spin_lock(&netlink_tap_lock);
149 list_for_each_entry(tmp, &netlink_tap_all, list) {
151 list_del_rcu(&nt->list);
157 pr_warn("__netlink_remove_tap: %p not found\n", nt);
159 spin_unlock(&netlink_tap_lock);
161 if (found && nt->module)
162 module_put(nt->module);
164 return found ? 0 : -ENODEV;
167 int netlink_remove_tap(struct netlink_tap *nt)
171 ret = __netlink_remove_tap(nt);
176 EXPORT_SYMBOL_GPL(netlink_remove_tap);
178 static bool netlink_filter_tap(const struct sk_buff *skb)
180 struct sock *sk = skb->sk;
182 /* We take the more conservative approach and
183 * whitelist socket protocols that may pass.
185 switch (sk->sk_protocol) {
187 case NETLINK_USERSOCK:
188 case NETLINK_SOCK_DIAG:
191 case NETLINK_FIB_LOOKUP:
192 case NETLINK_NETFILTER:
193 case NETLINK_GENERIC:
200 static int __netlink_deliver_tap_skb(struct sk_buff *skb,
201 struct net_device *dev)
203 struct sk_buff *nskb;
204 struct sock *sk = skb->sk;
208 nskb = skb_clone(skb, GFP_ATOMIC);
211 nskb->protocol = htons((u16) sk->sk_protocol);
212 nskb->pkt_type = netlink_is_kernel(sk) ?
213 PACKET_KERNEL : PACKET_USER;
214 skb_reset_network_header(nskb);
215 ret = dev_queue_xmit(nskb);
216 if (unlikely(ret > 0))
217 ret = net_xmit_errno(ret);
224 static void __netlink_deliver_tap(struct sk_buff *skb)
227 struct netlink_tap *tmp;
229 if (!netlink_filter_tap(skb))
232 list_for_each_entry_rcu(tmp, &netlink_tap_all, list) {
233 ret = __netlink_deliver_tap_skb(skb, tmp->dev);
239 static void netlink_deliver_tap(struct sk_buff *skb)
243 if (unlikely(!list_empty(&netlink_tap_all)))
244 __netlink_deliver_tap(skb);
249 static void netlink_deliver_tap_kernel(struct sock *dst, struct sock *src,
252 if (!(netlink_is_kernel(dst) && netlink_is_kernel(src)))
253 netlink_deliver_tap(skb);
256 static void netlink_overrun(struct sock *sk)
258 struct netlink_sock *nlk = nlk_sk(sk);
260 if (!(nlk->flags & NETLINK_F_RECV_NO_ENOBUFS)) {
261 if (!test_and_set_bit(NETLINK_S_CONGESTED,
262 &nlk_sk(sk)->state)) {
263 sk->sk_err = ENOBUFS;
264 sk->sk_error_report(sk);
267 atomic_inc(&sk->sk_drops);
270 static void netlink_rcv_wake(struct sock *sk)
272 struct netlink_sock *nlk = nlk_sk(sk);
274 if (skb_queue_empty(&sk->sk_receive_queue))
275 clear_bit(NETLINK_S_CONGESTED, &nlk->state);
276 if (!test_bit(NETLINK_S_CONGESTED, &nlk->state))
277 wake_up_interruptible(&nlk->wait);
280 #ifdef CONFIG_NETLINK_MMAP
281 static bool netlink_skb_is_mmaped(const struct sk_buff *skb)
283 return NETLINK_CB(skb).flags & NETLINK_SKB_MMAPED;
286 static bool netlink_rx_is_mmaped(struct sock *sk)
288 return nlk_sk(sk)->rx_ring.pg_vec != NULL;
291 static bool netlink_tx_is_mmaped(struct sock *sk)
293 return nlk_sk(sk)->tx_ring.pg_vec != NULL;
296 static __pure struct page *pgvec_to_page(const void *addr)
298 if (is_vmalloc_addr(addr))
299 return vmalloc_to_page(addr);
301 return virt_to_page(addr);
304 static void free_pg_vec(void **pg_vec, unsigned int order, unsigned int len)
308 for (i = 0; i < len; i++) {
309 if (pg_vec[i] != NULL) {
310 if (is_vmalloc_addr(pg_vec[i]))
313 free_pages((unsigned long)pg_vec[i], order);
319 static void *alloc_one_pg_vec_page(unsigned long order)
322 gfp_t gfp_flags = GFP_KERNEL | __GFP_COMP | __GFP_ZERO |
323 __GFP_NOWARN | __GFP_NORETRY;
325 buffer = (void *)__get_free_pages(gfp_flags, order);
329 buffer = vzalloc((1 << order) * PAGE_SIZE);
333 gfp_flags &= ~__GFP_NORETRY;
334 return (void *)__get_free_pages(gfp_flags, order);
337 static void **alloc_pg_vec(struct netlink_sock *nlk,
338 struct nl_mmap_req *req, unsigned int order)
340 unsigned int block_nr = req->nm_block_nr;
344 pg_vec = kcalloc(block_nr, sizeof(void *), GFP_KERNEL);
348 for (i = 0; i < block_nr; i++) {
349 pg_vec[i] = alloc_one_pg_vec_page(order);
350 if (pg_vec[i] == NULL)
356 free_pg_vec(pg_vec, order, block_nr);
360 static int netlink_set_ring(struct sock *sk, struct nl_mmap_req *req,
361 bool closing, bool tx_ring)
363 struct netlink_sock *nlk = nlk_sk(sk);
364 struct netlink_ring *ring;
365 struct sk_buff_head *queue;
366 void **pg_vec = NULL;
367 unsigned int order = 0;
370 ring = tx_ring ? &nlk->tx_ring : &nlk->rx_ring;
371 queue = tx_ring ? &sk->sk_write_queue : &sk->sk_receive_queue;
374 if (atomic_read(&nlk->mapped))
376 if (atomic_read(&ring->pending))
380 if (req->nm_block_nr) {
381 if (ring->pg_vec != NULL)
384 if ((int)req->nm_block_size <= 0)
386 if (!PAGE_ALIGNED(req->nm_block_size))
388 if (req->nm_frame_size < NL_MMAP_HDRLEN)
390 if (!IS_ALIGNED(req->nm_frame_size, NL_MMAP_MSG_ALIGNMENT))
393 ring->frames_per_block = req->nm_block_size /
395 if (ring->frames_per_block == 0)
397 if (ring->frames_per_block * req->nm_block_nr !=
401 order = get_order(req->nm_block_size);
402 pg_vec = alloc_pg_vec(nlk, req, order);
406 if (req->nm_frame_nr)
411 mutex_lock(&nlk->pg_vec_lock);
412 if (closing || atomic_read(&nlk->mapped) == 0) {
414 spin_lock_bh(&queue->lock);
416 ring->frame_max = req->nm_frame_nr - 1;
418 ring->frame_size = req->nm_frame_size;
419 ring->pg_vec_pages = req->nm_block_size / PAGE_SIZE;
421 swap(ring->pg_vec_len, req->nm_block_nr);
422 swap(ring->pg_vec_order, order);
423 swap(ring->pg_vec, pg_vec);
425 __skb_queue_purge(queue);
426 spin_unlock_bh(&queue->lock);
428 WARN_ON(atomic_read(&nlk->mapped));
430 mutex_unlock(&nlk->pg_vec_lock);
433 free_pg_vec(pg_vec, order, req->nm_block_nr);
437 static void netlink_mm_open(struct vm_area_struct *vma)
439 struct file *file = vma->vm_file;
440 struct socket *sock = file->private_data;
441 struct sock *sk = sock->sk;
444 atomic_inc(&nlk_sk(sk)->mapped);
447 static void netlink_mm_close(struct vm_area_struct *vma)
449 struct file *file = vma->vm_file;
450 struct socket *sock = file->private_data;
451 struct sock *sk = sock->sk;
454 atomic_dec(&nlk_sk(sk)->mapped);
457 static const struct vm_operations_struct netlink_mmap_ops = {
458 .open = netlink_mm_open,
459 .close = netlink_mm_close,
462 static int netlink_mmap(struct file *file, struct socket *sock,
463 struct vm_area_struct *vma)
465 struct sock *sk = sock->sk;
466 struct netlink_sock *nlk = nlk_sk(sk);
467 struct netlink_ring *ring;
468 unsigned long start, size, expected;
475 mutex_lock(&nlk->pg_vec_lock);
478 for (ring = &nlk->rx_ring; ring <= &nlk->tx_ring; ring++) {
479 if (ring->pg_vec == NULL)
481 expected += ring->pg_vec_len * ring->pg_vec_pages * PAGE_SIZE;
487 size = vma->vm_end - vma->vm_start;
488 if (size != expected)
491 start = vma->vm_start;
492 for (ring = &nlk->rx_ring; ring <= &nlk->tx_ring; ring++) {
493 if (ring->pg_vec == NULL)
496 for (i = 0; i < ring->pg_vec_len; i++) {
498 void *kaddr = ring->pg_vec[i];
501 for (pg_num = 0; pg_num < ring->pg_vec_pages; pg_num++) {
502 page = pgvec_to_page(kaddr);
503 err = vm_insert_page(vma, start, page);
512 atomic_inc(&nlk->mapped);
513 vma->vm_ops = &netlink_mmap_ops;
516 mutex_unlock(&nlk->pg_vec_lock);
520 static void netlink_frame_flush_dcache(const struct nl_mmap_hdr *hdr, unsigned int nm_len)
522 #if ARCH_IMPLEMENTS_FLUSH_DCACHE_PAGE == 1
523 struct page *p_start, *p_end;
525 /* First page is flushed through netlink_{get,set}_status */
526 p_start = pgvec_to_page(hdr + PAGE_SIZE);
527 p_end = pgvec_to_page((void *)hdr + NL_MMAP_HDRLEN + nm_len - 1);
528 while (p_start <= p_end) {
529 flush_dcache_page(p_start);
535 static enum nl_mmap_status netlink_get_status(const struct nl_mmap_hdr *hdr)
538 flush_dcache_page(pgvec_to_page(hdr));
539 return hdr->nm_status;
542 static void netlink_set_status(struct nl_mmap_hdr *hdr,
543 enum nl_mmap_status status)
546 hdr->nm_status = status;
547 flush_dcache_page(pgvec_to_page(hdr));
550 static struct nl_mmap_hdr *
551 __netlink_lookup_frame(const struct netlink_ring *ring, unsigned int pos)
553 unsigned int pg_vec_pos, frame_off;
555 pg_vec_pos = pos / ring->frames_per_block;
556 frame_off = pos % ring->frames_per_block;
558 return ring->pg_vec[pg_vec_pos] + (frame_off * ring->frame_size);
561 static struct nl_mmap_hdr *
562 netlink_lookup_frame(const struct netlink_ring *ring, unsigned int pos,
563 enum nl_mmap_status status)
565 struct nl_mmap_hdr *hdr;
567 hdr = __netlink_lookup_frame(ring, pos);
568 if (netlink_get_status(hdr) != status)
574 static struct nl_mmap_hdr *
575 netlink_current_frame(const struct netlink_ring *ring,
576 enum nl_mmap_status status)
578 return netlink_lookup_frame(ring, ring->head, status);
581 static struct nl_mmap_hdr *
582 netlink_previous_frame(const struct netlink_ring *ring,
583 enum nl_mmap_status status)
587 prev = ring->head ? ring->head - 1 : ring->frame_max;
588 return netlink_lookup_frame(ring, prev, status);
591 static void netlink_increment_head(struct netlink_ring *ring)
593 ring->head = ring->head != ring->frame_max ? ring->head + 1 : 0;
596 static void netlink_forward_ring(struct netlink_ring *ring)
598 unsigned int head = ring->head, pos = head;
599 const struct nl_mmap_hdr *hdr;
602 hdr = __netlink_lookup_frame(ring, pos);
603 if (hdr->nm_status == NL_MMAP_STATUS_UNUSED)
605 if (hdr->nm_status != NL_MMAP_STATUS_SKIP)
607 netlink_increment_head(ring);
608 } while (ring->head != head);
611 static bool netlink_dump_space(struct netlink_sock *nlk)
613 struct netlink_ring *ring = &nlk->rx_ring;
614 struct nl_mmap_hdr *hdr;
617 hdr = netlink_current_frame(ring, NL_MMAP_STATUS_UNUSED);
621 n = ring->head + ring->frame_max / 2;
622 if (n > ring->frame_max)
623 n -= ring->frame_max;
625 hdr = __netlink_lookup_frame(ring, n);
627 return hdr->nm_status == NL_MMAP_STATUS_UNUSED;
630 static unsigned int netlink_poll(struct file *file, struct socket *sock,
633 struct sock *sk = sock->sk;
634 struct netlink_sock *nlk = nlk_sk(sk);
638 if (nlk->rx_ring.pg_vec != NULL) {
639 /* Memory mapped sockets don't call recvmsg(), so flow control
640 * for dumps is performed here. A dump is allowed to continue
641 * if at least half the ring is unused.
643 while (nlk->cb_running && netlink_dump_space(nlk)) {
644 err = netlink_dump(sk);
647 sk->sk_error_report(sk);
651 netlink_rcv_wake(sk);
654 mask = datagram_poll(file, sock, wait);
656 spin_lock_bh(&sk->sk_receive_queue.lock);
657 if (nlk->rx_ring.pg_vec) {
658 netlink_forward_ring(&nlk->rx_ring);
659 if (!netlink_previous_frame(&nlk->rx_ring, NL_MMAP_STATUS_UNUSED))
660 mask |= POLLIN | POLLRDNORM;
662 spin_unlock_bh(&sk->sk_receive_queue.lock);
664 spin_lock_bh(&sk->sk_write_queue.lock);
665 if (nlk->tx_ring.pg_vec) {
666 if (netlink_current_frame(&nlk->tx_ring, NL_MMAP_STATUS_UNUSED))
667 mask |= POLLOUT | POLLWRNORM;
669 spin_unlock_bh(&sk->sk_write_queue.lock);
674 static struct nl_mmap_hdr *netlink_mmap_hdr(struct sk_buff *skb)
676 return (struct nl_mmap_hdr *)(skb->head - NL_MMAP_HDRLEN);
679 static void netlink_ring_setup_skb(struct sk_buff *skb, struct sock *sk,
680 struct netlink_ring *ring,
681 struct nl_mmap_hdr *hdr)
686 size = ring->frame_size - NL_MMAP_HDRLEN;
687 data = (void *)hdr + NL_MMAP_HDRLEN;
691 skb_reset_tail_pointer(skb);
692 skb->end = skb->tail + size;
695 skb->destructor = netlink_skb_destructor;
696 NETLINK_CB(skb).flags |= NETLINK_SKB_MMAPED;
697 NETLINK_CB(skb).sk = sk;
700 static int netlink_mmap_sendmsg(struct sock *sk, struct msghdr *msg,
701 u32 dst_portid, u32 dst_group,
702 struct scm_cookie *scm)
704 struct netlink_sock *nlk = nlk_sk(sk);
705 struct netlink_ring *ring;
706 struct nl_mmap_hdr *hdr;
709 int err = 0, len = 0;
711 mutex_lock(&nlk->pg_vec_lock);
713 ring = &nlk->tx_ring;
714 maxlen = ring->frame_size - NL_MMAP_HDRLEN;
719 hdr = netlink_current_frame(ring, NL_MMAP_STATUS_VALID);
721 if (!(msg->msg_flags & MSG_DONTWAIT) &&
722 atomic_read(&nlk->tx_ring.pending))
727 nm_len = ACCESS_ONCE(hdr->nm_len);
728 if (nm_len > maxlen) {
733 netlink_frame_flush_dcache(hdr, nm_len);
735 skb = alloc_skb(nm_len, GFP_KERNEL);
740 __skb_put(skb, nm_len);
741 memcpy(skb->data, (void *)hdr + NL_MMAP_HDRLEN, nm_len);
742 netlink_set_status(hdr, NL_MMAP_STATUS_UNUSED);
744 netlink_increment_head(ring);
746 NETLINK_CB(skb).portid = nlk->portid;
747 NETLINK_CB(skb).dst_group = dst_group;
748 NETLINK_CB(skb).creds = scm->creds;
750 err = security_netlink_send(sk, skb);
756 if (unlikely(dst_group)) {
757 atomic_inc(&skb->users);
758 netlink_broadcast(sk, skb, dst_portid, dst_group,
761 err = netlink_unicast(sk, skb, dst_portid,
762 msg->msg_flags & MSG_DONTWAIT);
767 } while (hdr != NULL ||
768 (!(msg->msg_flags & MSG_DONTWAIT) &&
769 atomic_read(&nlk->tx_ring.pending)));
774 mutex_unlock(&nlk->pg_vec_lock);
778 static void netlink_queue_mmaped_skb(struct sock *sk, struct sk_buff *skb)
780 struct nl_mmap_hdr *hdr;
782 hdr = netlink_mmap_hdr(skb);
783 hdr->nm_len = skb->len;
784 hdr->nm_group = NETLINK_CB(skb).dst_group;
785 hdr->nm_pid = NETLINK_CB(skb).creds.pid;
786 hdr->nm_uid = from_kuid(sk_user_ns(sk), NETLINK_CB(skb).creds.uid);
787 hdr->nm_gid = from_kgid(sk_user_ns(sk), NETLINK_CB(skb).creds.gid);
788 netlink_frame_flush_dcache(hdr, hdr->nm_len);
789 netlink_set_status(hdr, NL_MMAP_STATUS_VALID);
791 NETLINK_CB(skb).flags |= NETLINK_SKB_DELIVERED;
795 static void netlink_ring_set_copied(struct sock *sk, struct sk_buff *skb)
797 struct netlink_sock *nlk = nlk_sk(sk);
798 struct netlink_ring *ring = &nlk->rx_ring;
799 struct nl_mmap_hdr *hdr;
801 spin_lock_bh(&sk->sk_receive_queue.lock);
802 hdr = netlink_current_frame(ring, NL_MMAP_STATUS_UNUSED);
804 spin_unlock_bh(&sk->sk_receive_queue.lock);
809 netlink_increment_head(ring);
810 __skb_queue_tail(&sk->sk_receive_queue, skb);
811 spin_unlock_bh(&sk->sk_receive_queue.lock);
813 hdr->nm_len = skb->len;
814 hdr->nm_group = NETLINK_CB(skb).dst_group;
815 hdr->nm_pid = NETLINK_CB(skb).creds.pid;
816 hdr->nm_uid = from_kuid(sk_user_ns(sk), NETLINK_CB(skb).creds.uid);
817 hdr->nm_gid = from_kgid(sk_user_ns(sk), NETLINK_CB(skb).creds.gid);
818 netlink_set_status(hdr, NL_MMAP_STATUS_COPY);
821 #else /* CONFIG_NETLINK_MMAP */
822 #define netlink_skb_is_mmaped(skb) false
823 #define netlink_rx_is_mmaped(sk) false
824 #define netlink_tx_is_mmaped(sk) false
825 #define netlink_mmap sock_no_mmap
826 #define netlink_poll datagram_poll
827 #define netlink_mmap_sendmsg(sk, msg, dst_portid, dst_group, scm) 0
828 #endif /* CONFIG_NETLINK_MMAP */
830 static void netlink_skb_destructor(struct sk_buff *skb)
832 #ifdef CONFIG_NETLINK_MMAP
833 struct nl_mmap_hdr *hdr;
834 struct netlink_ring *ring;
837 /* If a packet from the kernel to userspace was freed because of an
838 * error without being delivered to userspace, the kernel must reset
839 * the status. In the direction userspace to kernel, the status is
840 * always reset here after the packet was processed and freed.
842 if (netlink_skb_is_mmaped(skb)) {
843 hdr = netlink_mmap_hdr(skb);
844 sk = NETLINK_CB(skb).sk;
846 if (NETLINK_CB(skb).flags & NETLINK_SKB_TX) {
847 netlink_set_status(hdr, NL_MMAP_STATUS_UNUSED);
848 ring = &nlk_sk(sk)->tx_ring;
850 if (!(NETLINK_CB(skb).flags & NETLINK_SKB_DELIVERED)) {
852 netlink_set_status(hdr, NL_MMAP_STATUS_VALID);
854 ring = &nlk_sk(sk)->rx_ring;
857 WARN_ON(atomic_read(&ring->pending) == 0);
858 atomic_dec(&ring->pending);
864 if (is_vmalloc_addr(skb->head)) {
866 !atomic_dec_return(&(skb_shinfo(skb)->dataref)))
875 static void netlink_skb_set_owner_r(struct sk_buff *skb, struct sock *sk)
877 WARN_ON(skb->sk != NULL);
879 skb->destructor = netlink_skb_destructor;
880 atomic_add(skb->truesize, &sk->sk_rmem_alloc);
881 sk_mem_charge(sk, skb->truesize);
884 static void netlink_sock_destruct(struct sock *sk)
886 struct netlink_sock *nlk = nlk_sk(sk);
888 if (nlk->cb_running) {
890 nlk->cb.done(&nlk->cb);
892 module_put(nlk->cb.module);
893 kfree_skb(nlk->cb.skb);
896 skb_queue_purge(&sk->sk_receive_queue);
897 #ifdef CONFIG_NETLINK_MMAP
899 struct nl_mmap_req req;
901 memset(&req, 0, sizeof(req));
902 if (nlk->rx_ring.pg_vec)
903 netlink_set_ring(sk, &req, true, false);
904 memset(&req, 0, sizeof(req));
905 if (nlk->tx_ring.pg_vec)
906 netlink_set_ring(sk, &req, true, true);
908 #endif /* CONFIG_NETLINK_MMAP */
910 if (!sock_flag(sk, SOCK_DEAD)) {
911 printk(KERN_ERR "Freeing alive netlink socket %p\n", sk);
915 WARN_ON(atomic_read(&sk->sk_rmem_alloc));
916 WARN_ON(atomic_read(&sk->sk_wmem_alloc));
917 WARN_ON(nlk_sk(sk)->groups);
920 /* This lock without WQ_FLAG_EXCLUSIVE is good on UP and it is _very_ bad on
921 * SMP. Look, when several writers sleep and reader wakes them up, all but one
922 * immediately hit write lock and grab all the cpus. Exclusive sleep solves
923 * this, _but_ remember, it adds useless work on UP machines.
926 void netlink_table_grab(void)
927 __acquires(nl_table_lock)
931 write_lock_irq(&nl_table_lock);
933 if (atomic_read(&nl_table_users)) {
934 DECLARE_WAITQUEUE(wait, current);
936 add_wait_queue_exclusive(&nl_table_wait, &wait);
938 set_current_state(TASK_UNINTERRUPTIBLE);
939 if (atomic_read(&nl_table_users) == 0)
941 write_unlock_irq(&nl_table_lock);
943 write_lock_irq(&nl_table_lock);
946 __set_current_state(TASK_RUNNING);
947 remove_wait_queue(&nl_table_wait, &wait);
951 void netlink_table_ungrab(void)
952 __releases(nl_table_lock)
954 write_unlock_irq(&nl_table_lock);
955 wake_up(&nl_table_wait);
959 netlink_lock_table(void)
961 /* read_lock() synchronizes us to netlink_table_grab */
963 read_lock(&nl_table_lock);
964 atomic_inc(&nl_table_users);
965 read_unlock(&nl_table_lock);
969 netlink_unlock_table(void)
971 if (atomic_dec_and_test(&nl_table_users))
972 wake_up(&nl_table_wait);
975 struct netlink_compare_arg
981 /* Doing sizeof directly may yield 4 extra bytes on 64-bit. */
982 #define netlink_compare_arg_len \
983 (offsetof(struct netlink_compare_arg, portid) + sizeof(u32))
985 static inline int netlink_compare(struct rhashtable_compare_arg *arg,
988 const struct netlink_compare_arg *x = arg->key;
989 const struct netlink_sock *nlk = ptr;
991 return nlk->portid != x->portid ||
992 !net_eq(sock_net(&nlk->sk), read_pnet(&x->pnet));
995 static void netlink_compare_arg_init(struct netlink_compare_arg *arg,
996 struct net *net, u32 portid)
998 memset(arg, 0, sizeof(*arg));
999 write_pnet(&arg->pnet, net);
1000 arg->portid = portid;
1003 static struct sock *__netlink_lookup(struct netlink_table *table, u32 portid,
1006 struct netlink_compare_arg arg;
1008 netlink_compare_arg_init(&arg, net, portid);
1009 return rhashtable_lookup_fast(&table->hash, &arg,
1010 netlink_rhashtable_params);
1013 static int __netlink_insert(struct netlink_table *table, struct sock *sk)
1015 struct netlink_compare_arg arg;
1017 netlink_compare_arg_init(&arg, sock_net(sk), nlk_sk(sk)->portid);
1018 return rhashtable_lookup_insert_key(&table->hash, &arg,
1020 netlink_rhashtable_params);
1023 static struct sock *netlink_lookup(struct net *net, int protocol, u32 portid)
1025 struct netlink_table *table = &nl_table[protocol];
1029 sk = __netlink_lookup(table, portid, net);
1037 static const struct proto_ops netlink_ops;
1040 netlink_update_listeners(struct sock *sk)
1042 struct netlink_table *tbl = &nl_table[sk->sk_protocol];
1045 struct listeners *listeners;
1047 listeners = nl_deref_protected(tbl->listeners);
1051 for (i = 0; i < NLGRPLONGS(tbl->groups); i++) {
1053 sk_for_each_bound(sk, &tbl->mc_list) {
1054 if (i < NLGRPLONGS(nlk_sk(sk)->ngroups))
1055 mask |= nlk_sk(sk)->groups[i];
1057 listeners->masks[i] = mask;
1059 /* this function is only called with the netlink table "grabbed", which
1060 * makes sure updates are visible before bind or setsockopt return. */
1063 static int netlink_insert(struct sock *sk, u32 portid)
1065 struct netlink_table *table = &nl_table[sk->sk_protocol];
1071 if (nlk_sk(sk)->portid)
1075 if (BITS_PER_LONG > 32 &&
1076 unlikely(atomic_read(&table->hash.nelems) >= UINT_MAX))
1079 nlk_sk(sk)->portid = portid;
1082 err = __netlink_insert(table, sk);
1094 static void netlink_remove(struct sock *sk)
1096 struct netlink_table *table;
1098 table = &nl_table[sk->sk_protocol];
1099 if (!rhashtable_remove_fast(&table->hash, &nlk_sk(sk)->node,
1100 netlink_rhashtable_params)) {
1101 WARN_ON(atomic_read(&sk->sk_refcnt) == 1);
1105 netlink_table_grab();
1106 if (nlk_sk(sk)->subscriptions) {
1107 __sk_del_bind_node(sk);
1108 netlink_update_listeners(sk);
1110 if (sk->sk_protocol == NETLINK_GENERIC)
1111 atomic_inc(&genl_sk_destructing_cnt);
1112 netlink_table_ungrab();
1115 static struct proto netlink_proto = {
1117 .owner = THIS_MODULE,
1118 .obj_size = sizeof(struct netlink_sock),
1121 static int __netlink_create(struct net *net, struct socket *sock,
1122 struct mutex *cb_mutex, int protocol,
1126 struct netlink_sock *nlk;
1128 sock->ops = &netlink_ops;
1130 sk = sk_alloc(net, PF_NETLINK, GFP_KERNEL, &netlink_proto, kern);
1134 sock_init_data(sock, sk);
1138 nlk->cb_mutex = cb_mutex;
1140 nlk->cb_mutex = &nlk->cb_def_mutex;
1141 mutex_init(nlk->cb_mutex);
1143 init_waitqueue_head(&nlk->wait);
1144 #ifdef CONFIG_NETLINK_MMAP
1145 mutex_init(&nlk->pg_vec_lock);
1148 sk->sk_destruct = netlink_sock_destruct;
1149 sk->sk_protocol = protocol;
1153 static int netlink_create(struct net *net, struct socket *sock, int protocol,
1156 struct module *module = NULL;
1157 struct mutex *cb_mutex;
1158 struct netlink_sock *nlk;
1159 int (*bind)(struct net *net, int group);
1160 void (*unbind)(struct net *net, int group);
1163 sock->state = SS_UNCONNECTED;
1165 if (sock->type != SOCK_RAW && sock->type != SOCK_DGRAM)
1166 return -ESOCKTNOSUPPORT;
1168 if (protocol < 0 || protocol >= MAX_LINKS)
1169 return -EPROTONOSUPPORT;
1171 netlink_lock_table();
1172 #ifdef CONFIG_MODULES
1173 if (!nl_table[protocol].registered) {
1174 netlink_unlock_table();
1175 request_module("net-pf-%d-proto-%d", PF_NETLINK, protocol);
1176 netlink_lock_table();
1179 if (nl_table[protocol].registered &&
1180 try_module_get(nl_table[protocol].module))
1181 module = nl_table[protocol].module;
1183 err = -EPROTONOSUPPORT;
1184 cb_mutex = nl_table[protocol].cb_mutex;
1185 bind = nl_table[protocol].bind;
1186 unbind = nl_table[protocol].unbind;
1187 netlink_unlock_table();
1192 err = __netlink_create(net, sock, cb_mutex, protocol, kern);
1197 sock_prot_inuse_add(net, &netlink_proto, 1);
1200 nlk = nlk_sk(sock->sk);
1201 nlk->module = module;
1202 nlk->netlink_bind = bind;
1203 nlk->netlink_unbind = unbind;
1212 static void deferred_put_nlk_sk(struct rcu_head *head)
1214 struct netlink_sock *nlk = container_of(head, struct netlink_sock, rcu);
1219 static int netlink_release(struct socket *sock)
1221 struct sock *sk = sock->sk;
1222 struct netlink_sock *nlk;
1232 * OK. Socket is unlinked, any packets that arrive now
1236 /* must not acquire netlink_table_lock in any way again before unbind
1237 * and notifying genetlink is done as otherwise it might deadlock
1239 if (nlk->netlink_unbind) {
1242 for (i = 0; i < nlk->ngroups; i++)
1243 if (test_bit(i, nlk->groups))
1244 nlk->netlink_unbind(sock_net(sk), i + 1);
1246 if (sk->sk_protocol == NETLINK_GENERIC &&
1247 atomic_dec_return(&genl_sk_destructing_cnt) == 0)
1248 wake_up(&genl_sk_destructing_waitq);
1251 wake_up_interruptible_all(&nlk->wait);
1253 skb_queue_purge(&sk->sk_write_queue);
1256 struct netlink_notify n = {
1257 .net = sock_net(sk),
1258 .protocol = sk->sk_protocol,
1259 .portid = nlk->portid,
1261 atomic_notifier_call_chain(&netlink_chain,
1262 NETLINK_URELEASE, &n);
1265 module_put(nlk->module);
1267 if (netlink_is_kernel(sk)) {
1268 netlink_table_grab();
1269 BUG_ON(nl_table[sk->sk_protocol].registered == 0);
1270 if (--nl_table[sk->sk_protocol].registered == 0) {
1271 struct listeners *old;
1273 old = nl_deref_protected(nl_table[sk->sk_protocol].listeners);
1274 RCU_INIT_POINTER(nl_table[sk->sk_protocol].listeners, NULL);
1275 kfree_rcu(old, rcu);
1276 nl_table[sk->sk_protocol].module = NULL;
1277 nl_table[sk->sk_protocol].bind = NULL;
1278 nl_table[sk->sk_protocol].unbind = NULL;
1279 nl_table[sk->sk_protocol].flags = 0;
1280 nl_table[sk->sk_protocol].registered = 0;
1282 netlink_table_ungrab();
1289 sock_prot_inuse_add(sock_net(sk), &netlink_proto, -1);
1291 call_rcu(&nlk->rcu, deferred_put_nlk_sk);
1295 static int netlink_autobind(struct socket *sock)
1297 struct sock *sk = sock->sk;
1298 struct net *net = sock_net(sk);
1299 struct netlink_table *table = &nl_table[sk->sk_protocol];
1300 s32 portid = task_tgid_vnr(current);
1308 ok = !__netlink_lookup(table, portid, net);
1311 /* Bind collision, search negative portid values. */
1313 /* rover will be in range [S32_MIN, -4097] */
1314 rover = S32_MIN + prandom_u32_max(-4096 - S32_MIN);
1315 else if (rover >= -4096)
1321 err = netlink_insert(sk, portid);
1322 if (err == -EADDRINUSE)
1325 /* If 2 threads race to autobind, that is fine. */
1333 * __netlink_ns_capable - General netlink message capability test
1334 * @nsp: NETLINK_CB of the socket buffer holding a netlink command from userspace.
1335 * @user_ns: The user namespace of the capability to use
1336 * @cap: The capability to use
1338 * Test to see if the opener of the socket we received the message
1339 * from had when the netlink socket was created and the sender of the
1340 * message has has the capability @cap in the user namespace @user_ns.
1342 bool __netlink_ns_capable(const struct netlink_skb_parms *nsp,
1343 struct user_namespace *user_ns, int cap)
1345 return ((nsp->flags & NETLINK_SKB_DST) ||
1346 file_ns_capable(nsp->sk->sk_socket->file, user_ns, cap)) &&
1347 ns_capable(user_ns, cap);
1349 EXPORT_SYMBOL(__netlink_ns_capable);
1352 * netlink_ns_capable - General netlink message capability test
1353 * @skb: socket buffer holding a netlink command from userspace
1354 * @user_ns: The user namespace of the capability to use
1355 * @cap: The capability to use
1357 * Test to see if the opener of the socket we received the message
1358 * from had when the netlink socket was created and the sender of the
1359 * message has has the capability @cap in the user namespace @user_ns.
1361 bool netlink_ns_capable(const struct sk_buff *skb,
1362 struct user_namespace *user_ns, int cap)
1364 return __netlink_ns_capable(&NETLINK_CB(skb), user_ns, cap);
1366 EXPORT_SYMBOL(netlink_ns_capable);
1369 * netlink_capable - Netlink global message capability test
1370 * @skb: socket buffer holding a netlink command from userspace
1371 * @cap: The capability to use
1373 * Test to see if the opener of the socket we received the message
1374 * from had when the netlink socket was created and the sender of the
1375 * message has has the capability @cap in all user namespaces.
1377 bool netlink_capable(const struct sk_buff *skb, int cap)
1379 return netlink_ns_capable(skb, &init_user_ns, cap);
1381 EXPORT_SYMBOL(netlink_capable);
1384 * netlink_net_capable - Netlink network namespace message capability test
1385 * @skb: socket buffer holding a netlink command from userspace
1386 * @cap: The capability to use
1388 * Test to see if the opener of the socket we received the message
1389 * from had when the netlink socket was created and the sender of the
1390 * message has has the capability @cap over the network namespace of
1391 * the socket we received the message from.
1393 bool netlink_net_capable(const struct sk_buff *skb, int cap)
1395 return netlink_ns_capable(skb, sock_net(skb->sk)->user_ns, cap);
1397 EXPORT_SYMBOL(netlink_net_capable);
1399 static inline int netlink_allowed(const struct socket *sock, unsigned int flag)
1401 return (nl_table[sock->sk->sk_protocol].flags & flag) ||
1402 ns_capable(sock_net(sock->sk)->user_ns, CAP_NET_ADMIN);
1406 netlink_update_subscriptions(struct sock *sk, unsigned int subscriptions)
1408 struct netlink_sock *nlk = nlk_sk(sk);
1410 if (nlk->subscriptions && !subscriptions)
1411 __sk_del_bind_node(sk);
1412 else if (!nlk->subscriptions && subscriptions)
1413 sk_add_bind_node(sk, &nl_table[sk->sk_protocol].mc_list);
1414 nlk->subscriptions = subscriptions;
1417 static int netlink_realloc_groups(struct sock *sk)
1419 struct netlink_sock *nlk = nlk_sk(sk);
1420 unsigned int groups;
1421 unsigned long *new_groups;
1424 netlink_table_grab();
1426 groups = nl_table[sk->sk_protocol].groups;
1427 if (!nl_table[sk->sk_protocol].registered) {
1432 if (nlk->ngroups >= groups)
1435 new_groups = krealloc(nlk->groups, NLGRPSZ(groups), GFP_ATOMIC);
1436 if (new_groups == NULL) {
1440 memset((char *)new_groups + NLGRPSZ(nlk->ngroups), 0,
1441 NLGRPSZ(groups) - NLGRPSZ(nlk->ngroups));
1443 nlk->groups = new_groups;
1444 nlk->ngroups = groups;
1446 netlink_table_ungrab();
1450 static void netlink_undo_bind(int group, long unsigned int groups,
1453 struct netlink_sock *nlk = nlk_sk(sk);
1456 if (!nlk->netlink_unbind)
1459 for (undo = 0; undo < group; undo++)
1460 if (test_bit(undo, &groups))
1461 nlk->netlink_unbind(sock_net(sk), undo + 1);
1464 static int netlink_bind(struct socket *sock, struct sockaddr *addr,
1467 struct sock *sk = sock->sk;
1468 struct net *net = sock_net(sk);
1469 struct netlink_sock *nlk = nlk_sk(sk);
1470 struct sockaddr_nl *nladdr = (struct sockaddr_nl *)addr;
1472 long unsigned int groups = nladdr->nl_groups;
1474 if (addr_len < sizeof(struct sockaddr_nl))
1477 if (nladdr->nl_family != AF_NETLINK)
1480 /* Only superuser is allowed to listen multicasts */
1482 if (!netlink_allowed(sock, NL_CFG_F_NONROOT_RECV))
1484 err = netlink_realloc_groups(sk);
1490 if (nladdr->nl_pid != nlk->portid)
1493 if (nlk->netlink_bind && groups) {
1496 for (group = 0; group < nlk->ngroups; group++) {
1497 if (!test_bit(group, &groups))
1499 err = nlk->netlink_bind(net, group + 1);
1502 netlink_undo_bind(group, groups, sk);
1508 err = nladdr->nl_pid ?
1509 netlink_insert(sk, nladdr->nl_pid) :
1510 netlink_autobind(sock);
1512 netlink_undo_bind(nlk->ngroups, groups, sk);
1517 if (!groups && (nlk->groups == NULL || !(u32)nlk->groups[0]))
1520 netlink_table_grab();
1521 netlink_update_subscriptions(sk, nlk->subscriptions +
1523 hweight32(nlk->groups[0]));
1524 nlk->groups[0] = (nlk->groups[0] & ~0xffffffffUL) | groups;
1525 netlink_update_listeners(sk);
1526 netlink_table_ungrab();
1531 static int netlink_connect(struct socket *sock, struct sockaddr *addr,
1532 int alen, int flags)
1535 struct sock *sk = sock->sk;
1536 struct netlink_sock *nlk = nlk_sk(sk);
1537 struct sockaddr_nl *nladdr = (struct sockaddr_nl *)addr;
1539 if (alen < sizeof(addr->sa_family))
1542 if (addr->sa_family == AF_UNSPEC) {
1543 sk->sk_state = NETLINK_UNCONNECTED;
1544 nlk->dst_portid = 0;
1548 if (addr->sa_family != AF_NETLINK)
1551 if ((nladdr->nl_groups || nladdr->nl_pid) &&
1552 !netlink_allowed(sock, NL_CFG_F_NONROOT_SEND))
1556 err = netlink_autobind(sock);
1559 sk->sk_state = NETLINK_CONNECTED;
1560 nlk->dst_portid = nladdr->nl_pid;
1561 nlk->dst_group = ffs(nladdr->nl_groups);
1567 static int netlink_getname(struct socket *sock, struct sockaddr *addr,
1568 int *addr_len, int peer)
1570 struct sock *sk = sock->sk;
1571 struct netlink_sock *nlk = nlk_sk(sk);
1572 DECLARE_SOCKADDR(struct sockaddr_nl *, nladdr, addr);
1574 nladdr->nl_family = AF_NETLINK;
1576 *addr_len = sizeof(*nladdr);
1579 nladdr->nl_pid = nlk->dst_portid;
1580 nladdr->nl_groups = netlink_group_mask(nlk->dst_group);
1582 nladdr->nl_pid = nlk->portid;
1583 nladdr->nl_groups = nlk->groups ? nlk->groups[0] : 0;
1588 static struct sock *netlink_getsockbyportid(struct sock *ssk, u32 portid)
1591 struct netlink_sock *nlk;
1593 sock = netlink_lookup(sock_net(ssk), ssk->sk_protocol, portid);
1595 return ERR_PTR(-ECONNREFUSED);
1597 /* Don't bother queuing skb if kernel socket has no input function */
1599 if (sock->sk_state == NETLINK_CONNECTED &&
1600 nlk->dst_portid != nlk_sk(ssk)->portid) {
1602 return ERR_PTR(-ECONNREFUSED);
1607 struct sock *netlink_getsockbyfilp(struct file *filp)
1609 struct inode *inode = file_inode(filp);
1612 if (!S_ISSOCK(inode->i_mode))
1613 return ERR_PTR(-ENOTSOCK);
1615 sock = SOCKET_I(inode)->sk;
1616 if (sock->sk_family != AF_NETLINK)
1617 return ERR_PTR(-EINVAL);
1623 static struct sk_buff *netlink_alloc_large_skb(unsigned int size,
1626 struct sk_buff *skb;
1629 if (size <= NLMSG_GOODSIZE || broadcast)
1630 return alloc_skb(size, GFP_KERNEL);
1632 size = SKB_DATA_ALIGN(size) +
1633 SKB_DATA_ALIGN(sizeof(struct skb_shared_info));
1635 data = vmalloc(size);
1639 skb = __build_skb(data, size);
1643 skb->destructor = netlink_skb_destructor;
1649 * Attach a skb to a netlink socket.
1650 * The caller must hold a reference to the destination socket. On error, the
1651 * reference is dropped. The skb is not send to the destination, just all
1652 * all error checks are performed and memory in the queue is reserved.
1654 * < 0: error. skb freed, reference to sock dropped.
1656 * 1: repeat lookup - reference dropped while waiting for socket memory.
1658 int netlink_attachskb(struct sock *sk, struct sk_buff *skb,
1659 long *timeo, struct sock *ssk)
1661 struct netlink_sock *nlk;
1665 if ((atomic_read(&sk->sk_rmem_alloc) > sk->sk_rcvbuf ||
1666 test_bit(NETLINK_S_CONGESTED, &nlk->state)) &&
1667 !netlink_skb_is_mmaped(skb)) {
1668 DECLARE_WAITQUEUE(wait, current);
1670 if (!ssk || netlink_is_kernel(ssk))
1671 netlink_overrun(sk);
1677 __set_current_state(TASK_INTERRUPTIBLE);
1678 add_wait_queue(&nlk->wait, &wait);
1680 if ((atomic_read(&sk->sk_rmem_alloc) > sk->sk_rcvbuf ||
1681 test_bit(NETLINK_S_CONGESTED, &nlk->state)) &&
1682 !sock_flag(sk, SOCK_DEAD))
1683 *timeo = schedule_timeout(*timeo);
1685 __set_current_state(TASK_RUNNING);
1686 remove_wait_queue(&nlk->wait, &wait);
1689 if (signal_pending(current)) {
1691 return sock_intr_errno(*timeo);
1695 netlink_skb_set_owner_r(skb, sk);
1699 static int __netlink_sendskb(struct sock *sk, struct sk_buff *skb)
1703 netlink_deliver_tap(skb);
1705 #ifdef CONFIG_NETLINK_MMAP
1706 if (netlink_skb_is_mmaped(skb))
1707 netlink_queue_mmaped_skb(sk, skb);
1708 else if (netlink_rx_is_mmaped(sk))
1709 netlink_ring_set_copied(sk, skb);
1711 #endif /* CONFIG_NETLINK_MMAP */
1712 skb_queue_tail(&sk->sk_receive_queue, skb);
1713 sk->sk_data_ready(sk);
1717 int netlink_sendskb(struct sock *sk, struct sk_buff *skb)
1719 int len = __netlink_sendskb(sk, skb);
1725 void netlink_detachskb(struct sock *sk, struct sk_buff *skb)
1731 static struct sk_buff *netlink_trim(struct sk_buff *skb, gfp_t allocation)
1735 WARN_ON(skb->sk != NULL);
1736 if (netlink_skb_is_mmaped(skb))
1739 delta = skb->end - skb->tail;
1740 if (is_vmalloc_addr(skb->head) || delta * 2 < skb->truesize)
1743 if (skb_shared(skb)) {
1744 struct sk_buff *nskb = skb_clone(skb, allocation);
1751 if (!pskb_expand_head(skb, 0, -delta, allocation))
1752 skb->truesize -= delta;
1757 static int netlink_unicast_kernel(struct sock *sk, struct sk_buff *skb,
1761 struct netlink_sock *nlk = nlk_sk(sk);
1763 ret = -ECONNREFUSED;
1764 if (nlk->netlink_rcv != NULL) {
1766 netlink_skb_set_owner_r(skb, sk);
1767 NETLINK_CB(skb).sk = ssk;
1768 netlink_deliver_tap_kernel(sk, ssk, skb);
1769 nlk->netlink_rcv(skb);
1778 int netlink_unicast(struct sock *ssk, struct sk_buff *skb,
1779 u32 portid, int nonblock)
1785 skb = netlink_trim(skb, gfp_any());
1787 timeo = sock_sndtimeo(ssk, nonblock);
1789 sk = netlink_getsockbyportid(ssk, portid);
1794 if (netlink_is_kernel(sk))
1795 return netlink_unicast_kernel(sk, skb, ssk);
1797 if (sk_filter(sk, skb)) {
1804 err = netlink_attachskb(sk, skb, &timeo, ssk);
1810 return netlink_sendskb(sk, skb);
1812 EXPORT_SYMBOL(netlink_unicast);
1814 struct sk_buff *netlink_alloc_skb(struct sock *ssk, unsigned int size,
1815 u32 dst_portid, gfp_t gfp_mask)
1817 #ifdef CONFIG_NETLINK_MMAP
1818 struct sock *sk = NULL;
1819 struct sk_buff *skb;
1820 struct netlink_ring *ring;
1821 struct nl_mmap_hdr *hdr;
1822 unsigned int maxlen;
1824 sk = netlink_getsockbyportid(ssk, dst_portid);
1828 ring = &nlk_sk(sk)->rx_ring;
1829 /* fast-path without atomic ops for common case: non-mmaped receiver */
1830 if (ring->pg_vec == NULL)
1833 if (ring->frame_size - NL_MMAP_HDRLEN < size)
1836 skb = alloc_skb_head(gfp_mask);
1840 spin_lock_bh(&sk->sk_receive_queue.lock);
1841 /* check again under lock */
1842 if (ring->pg_vec == NULL)
1845 /* check again under lock */
1846 maxlen = ring->frame_size - NL_MMAP_HDRLEN;
1850 netlink_forward_ring(ring);
1851 hdr = netlink_current_frame(ring, NL_MMAP_STATUS_UNUSED);
1854 netlink_ring_setup_skb(skb, sk, ring, hdr);
1855 netlink_set_status(hdr, NL_MMAP_STATUS_RESERVED);
1856 atomic_inc(&ring->pending);
1857 netlink_increment_head(ring);
1859 spin_unlock_bh(&sk->sk_receive_queue.lock);
1864 spin_unlock_bh(&sk->sk_receive_queue.lock);
1865 netlink_overrun(sk);
1872 spin_unlock_bh(&sk->sk_receive_queue.lock);
1877 return alloc_skb(size, gfp_mask);
1879 EXPORT_SYMBOL_GPL(netlink_alloc_skb);
1881 int netlink_has_listeners(struct sock *sk, unsigned int group)
1884 struct listeners *listeners;
1886 BUG_ON(!netlink_is_kernel(sk));
1889 listeners = rcu_dereference(nl_table[sk->sk_protocol].listeners);
1891 if (listeners && group - 1 < nl_table[sk->sk_protocol].groups)
1892 res = test_bit(group - 1, listeners->masks);
1898 EXPORT_SYMBOL_GPL(netlink_has_listeners);
1900 static int netlink_broadcast_deliver(struct sock *sk, struct sk_buff *skb)
1902 struct netlink_sock *nlk = nlk_sk(sk);
1904 if (atomic_read(&sk->sk_rmem_alloc) <= sk->sk_rcvbuf &&
1905 !test_bit(NETLINK_S_CONGESTED, &nlk->state)) {
1906 netlink_skb_set_owner_r(skb, sk);
1907 __netlink_sendskb(sk, skb);
1908 return atomic_read(&sk->sk_rmem_alloc) > (sk->sk_rcvbuf >> 1);
1913 struct netlink_broadcast_data {
1914 struct sock *exclude_sk;
1919 int delivery_failure;
1923 struct sk_buff *skb, *skb2;
1924 int (*tx_filter)(struct sock *dsk, struct sk_buff *skb, void *data);
1928 static void do_one_broadcast(struct sock *sk,
1929 struct netlink_broadcast_data *p)
1931 struct netlink_sock *nlk = nlk_sk(sk);
1934 if (p->exclude_sk == sk)
1937 if (nlk->portid == p->portid || p->group - 1 >= nlk->ngroups ||
1938 !test_bit(p->group - 1, nlk->groups))
1941 if (!net_eq(sock_net(sk), p->net)) {
1942 if (!(nlk->flags & NETLINK_F_LISTEN_ALL_NSID))
1945 if (!peernet_has_id(sock_net(sk), p->net))
1948 if (!file_ns_capable(sk->sk_socket->file, p->net->user_ns,
1954 netlink_overrun(sk);
1959 if (p->skb2 == NULL) {
1960 if (skb_shared(p->skb)) {
1961 p->skb2 = skb_clone(p->skb, p->allocation);
1963 p->skb2 = skb_get(p->skb);
1965 * skb ownership may have been set when
1966 * delivered to a previous socket.
1968 skb_orphan(p->skb2);
1971 if (p->skb2 == NULL) {
1972 netlink_overrun(sk);
1973 /* Clone failed. Notify ALL listeners. */
1975 if (nlk->flags & NETLINK_F_BROADCAST_SEND_ERROR)
1976 p->delivery_failure = 1;
1979 if (p->tx_filter && p->tx_filter(sk, p->skb2, p->tx_data)) {
1984 if (sk_filter(sk, p->skb2)) {
1989 NETLINK_CB(p->skb2).nsid = peernet2id(sock_net(sk), p->net);
1990 NETLINK_CB(p->skb2).nsid_is_set = true;
1991 val = netlink_broadcast_deliver(sk, p->skb2);
1993 netlink_overrun(sk);
1994 if (nlk->flags & NETLINK_F_BROADCAST_SEND_ERROR)
1995 p->delivery_failure = 1;
1997 p->congested |= val;
2005 int netlink_broadcast_filtered(struct sock *ssk, struct sk_buff *skb, u32 portid,
2006 u32 group, gfp_t allocation,
2007 int (*filter)(struct sock *dsk, struct sk_buff *skb, void *data),
2010 struct net *net = sock_net(ssk);
2011 struct netlink_broadcast_data info;
2014 skb = netlink_trim(skb, allocation);
2016 info.exclude_sk = ssk;
2018 info.portid = portid;
2021 info.delivery_failure = 0;
2024 info.allocation = allocation;
2027 info.tx_filter = filter;
2028 info.tx_data = filter_data;
2030 /* While we sleep in clone, do not allow to change socket list */
2032 netlink_lock_table();
2034 sk_for_each_bound(sk, &nl_table[ssk->sk_protocol].mc_list)
2035 do_one_broadcast(sk, &info);
2039 netlink_unlock_table();
2041 if (info.delivery_failure) {
2042 kfree_skb(info.skb2);
2045 consume_skb(info.skb2);
2047 if (info.delivered) {
2048 if (info.congested && (allocation & __GFP_WAIT))
2054 EXPORT_SYMBOL(netlink_broadcast_filtered);
2056 int netlink_broadcast(struct sock *ssk, struct sk_buff *skb, u32 portid,
2057 u32 group, gfp_t allocation)
2059 return netlink_broadcast_filtered(ssk, skb, portid, group, allocation,
2062 EXPORT_SYMBOL(netlink_broadcast);
2064 struct netlink_set_err_data {
2065 struct sock *exclude_sk;
2071 static int do_one_set_err(struct sock *sk, struct netlink_set_err_data *p)
2073 struct netlink_sock *nlk = nlk_sk(sk);
2076 if (sk == p->exclude_sk)
2079 if (!net_eq(sock_net(sk), sock_net(p->exclude_sk)))
2082 if (nlk->portid == p->portid || p->group - 1 >= nlk->ngroups ||
2083 !test_bit(p->group - 1, nlk->groups))
2086 if (p->code == ENOBUFS && nlk->flags & NETLINK_F_RECV_NO_ENOBUFS) {
2091 sk->sk_err = p->code;
2092 sk->sk_error_report(sk);
2098 * netlink_set_err - report error to broadcast listeners
2099 * @ssk: the kernel netlink socket, as returned by netlink_kernel_create()
2100 * @portid: the PORTID of a process that we want to skip (if any)
2101 * @group: the broadcast group that will notice the error
2102 * @code: error code, must be negative (as usual in kernelspace)
2104 * This function returns the number of broadcast listeners that have set the
2105 * NETLINK_NO_ENOBUFS socket option.
2107 int netlink_set_err(struct sock *ssk, u32 portid, u32 group, int code)
2109 struct netlink_set_err_data info;
2113 info.exclude_sk = ssk;
2114 info.portid = portid;
2116 /* sk->sk_err wants a positive error value */
2119 read_lock(&nl_table_lock);
2121 sk_for_each_bound(sk, &nl_table[ssk->sk_protocol].mc_list)
2122 ret += do_one_set_err(sk, &info);
2124 read_unlock(&nl_table_lock);
2127 EXPORT_SYMBOL(netlink_set_err);
2129 /* must be called with netlink table grabbed */
2130 static void netlink_update_socket_mc(struct netlink_sock *nlk,
2134 int old, new = !!is_new, subscriptions;
2136 old = test_bit(group - 1, nlk->groups);
2137 subscriptions = nlk->subscriptions - old + new;
2139 __set_bit(group - 1, nlk->groups);
2141 __clear_bit(group - 1, nlk->groups);
2142 netlink_update_subscriptions(&nlk->sk, subscriptions);
2143 netlink_update_listeners(&nlk->sk);
2146 static int netlink_setsockopt(struct socket *sock, int level, int optname,
2147 char __user *optval, unsigned int optlen)
2149 struct sock *sk = sock->sk;
2150 struct netlink_sock *nlk = nlk_sk(sk);
2151 unsigned int val = 0;
2154 if (level != SOL_NETLINK)
2155 return -ENOPROTOOPT;
2157 if (optname != NETLINK_RX_RING && optname != NETLINK_TX_RING &&
2158 optlen >= sizeof(int) &&
2159 get_user(val, (unsigned int __user *)optval))
2163 case NETLINK_PKTINFO:
2165 nlk->flags |= NETLINK_F_RECV_PKTINFO;
2167 nlk->flags &= ~NETLINK_F_RECV_PKTINFO;
2170 case NETLINK_ADD_MEMBERSHIP:
2171 case NETLINK_DROP_MEMBERSHIP: {
2172 if (!netlink_allowed(sock, NL_CFG_F_NONROOT_RECV))
2174 err = netlink_realloc_groups(sk);
2177 if (!val || val - 1 >= nlk->ngroups)
2179 if (optname == NETLINK_ADD_MEMBERSHIP && nlk->netlink_bind) {
2180 err = nlk->netlink_bind(sock_net(sk), val);
2184 netlink_table_grab();
2185 netlink_update_socket_mc(nlk, val,
2186 optname == NETLINK_ADD_MEMBERSHIP);
2187 netlink_table_ungrab();
2188 if (optname == NETLINK_DROP_MEMBERSHIP && nlk->netlink_unbind)
2189 nlk->netlink_unbind(sock_net(sk), val);
2194 case NETLINK_BROADCAST_ERROR:
2196 nlk->flags |= NETLINK_F_BROADCAST_SEND_ERROR;
2198 nlk->flags &= ~NETLINK_F_BROADCAST_SEND_ERROR;
2201 case NETLINK_NO_ENOBUFS:
2203 nlk->flags |= NETLINK_F_RECV_NO_ENOBUFS;
2204 clear_bit(NETLINK_S_CONGESTED, &nlk->state);
2205 wake_up_interruptible(&nlk->wait);
2207 nlk->flags &= ~NETLINK_F_RECV_NO_ENOBUFS;
2211 #ifdef CONFIG_NETLINK_MMAP
2212 case NETLINK_RX_RING:
2213 case NETLINK_TX_RING: {
2214 struct nl_mmap_req req;
2216 /* Rings might consume more memory than queue limits, require
2219 if (!capable(CAP_NET_ADMIN))
2221 if (optlen < sizeof(req))
2223 if (copy_from_user(&req, optval, sizeof(req)))
2225 err = netlink_set_ring(sk, &req, false,
2226 optname == NETLINK_TX_RING);
2229 #endif /* CONFIG_NETLINK_MMAP */
2230 case NETLINK_LISTEN_ALL_NSID:
2231 if (!ns_capable(sock_net(sk)->user_ns, CAP_NET_BROADCAST))
2235 nlk->flags |= NETLINK_F_LISTEN_ALL_NSID;
2237 nlk->flags &= ~NETLINK_F_LISTEN_ALL_NSID;
2246 static int netlink_getsockopt(struct socket *sock, int level, int optname,
2247 char __user *optval, int __user *optlen)
2249 struct sock *sk = sock->sk;
2250 struct netlink_sock *nlk = nlk_sk(sk);
2253 if (level != SOL_NETLINK)
2254 return -ENOPROTOOPT;
2256 if (get_user(len, optlen))
2262 case NETLINK_PKTINFO:
2263 if (len < sizeof(int))
2266 val = nlk->flags & NETLINK_F_RECV_PKTINFO ? 1 : 0;
2267 if (put_user(len, optlen) ||
2268 put_user(val, optval))
2272 case NETLINK_BROADCAST_ERROR:
2273 if (len < sizeof(int))
2276 val = nlk->flags & NETLINK_F_BROADCAST_SEND_ERROR ? 1 : 0;
2277 if (put_user(len, optlen) ||
2278 put_user(val, optval))
2282 case NETLINK_NO_ENOBUFS:
2283 if (len < sizeof(int))
2286 val = nlk->flags & NETLINK_F_RECV_NO_ENOBUFS ? 1 : 0;
2287 if (put_user(len, optlen) ||
2288 put_user(val, optval))
2298 static void netlink_cmsg_recv_pktinfo(struct msghdr *msg, struct sk_buff *skb)
2300 struct nl_pktinfo info;
2302 info.group = NETLINK_CB(skb).dst_group;
2303 put_cmsg(msg, SOL_NETLINK, NETLINK_PKTINFO, sizeof(info), &info);
2306 static void netlink_cmsg_listen_all_nsid(struct sock *sk, struct msghdr *msg,
2307 struct sk_buff *skb)
2309 if (!NETLINK_CB(skb).nsid_is_set)
2312 put_cmsg(msg, SOL_NETLINK, NETLINK_LISTEN_ALL_NSID, sizeof(int),
2313 &NETLINK_CB(skb).nsid);
2316 static int netlink_sendmsg(struct socket *sock, struct msghdr *msg, size_t len)
2318 struct sock *sk = sock->sk;
2319 struct netlink_sock *nlk = nlk_sk(sk);
2320 DECLARE_SOCKADDR(struct sockaddr_nl *, addr, msg->msg_name);
2323 struct sk_buff *skb;
2325 struct scm_cookie scm;
2326 u32 netlink_skb_flags = 0;
2328 if (msg->msg_flags&MSG_OOB)
2331 err = scm_send(sock, msg, &scm, true);
2335 if (msg->msg_namelen) {
2337 if (addr->nl_family != AF_NETLINK)
2339 dst_portid = addr->nl_pid;
2340 dst_group = ffs(addr->nl_groups);
2342 if ((dst_group || dst_portid) &&
2343 !netlink_allowed(sock, NL_CFG_F_NONROOT_SEND))
2345 netlink_skb_flags |= NETLINK_SKB_DST;
2347 dst_portid = nlk->dst_portid;
2348 dst_group = nlk->dst_group;
2352 err = netlink_autobind(sock);
2357 /* It's a really convoluted way for userland to ask for mmaped
2358 * sendmsg(), but that's what we've got...
2360 if (netlink_tx_is_mmaped(sk) &&
2361 msg->msg_iter.type == ITER_IOVEC &&
2362 msg->msg_iter.nr_segs == 1 &&
2363 msg->msg_iter.iov->iov_base == NULL) {
2364 err = netlink_mmap_sendmsg(sk, msg, dst_portid, dst_group,
2370 if (len > sk->sk_sndbuf - 32)
2373 skb = netlink_alloc_large_skb(len, dst_group);
2377 NETLINK_CB(skb).portid = nlk->portid;
2378 NETLINK_CB(skb).dst_group = dst_group;
2379 NETLINK_CB(skb).creds = scm.creds;
2380 NETLINK_CB(skb).flags = netlink_skb_flags;
2383 if (memcpy_from_msg(skb_put(skb, len), msg, len)) {
2388 err = security_netlink_send(sk, skb);
2395 atomic_inc(&skb->users);
2396 netlink_broadcast(sk, skb, dst_portid, dst_group, GFP_KERNEL);
2398 err = netlink_unicast(sk, skb, dst_portid, msg->msg_flags&MSG_DONTWAIT);
2405 static int netlink_recvmsg(struct socket *sock, struct msghdr *msg, size_t len,
2408 struct scm_cookie scm;
2409 struct sock *sk = sock->sk;
2410 struct netlink_sock *nlk = nlk_sk(sk);
2411 int noblock = flags&MSG_DONTWAIT;
2413 struct sk_buff *skb, *data_skb;
2421 skb = skb_recv_datagram(sk, flags, noblock, &err);
2427 #ifdef CONFIG_COMPAT_NETLINK_MESSAGES
2428 if (unlikely(skb_shinfo(skb)->frag_list)) {
2430 * If this skb has a frag_list, then here that means that we
2431 * will have to use the frag_list skb's data for compat tasks
2432 * and the regular skb's data for normal (non-compat) tasks.
2434 * If we need to send the compat skb, assign it to the
2435 * 'data_skb' variable so that it will be used below for data
2436 * copying. We keep 'skb' for everything else, including
2437 * freeing both later.
2439 if (flags & MSG_CMSG_COMPAT)
2440 data_skb = skb_shinfo(skb)->frag_list;
2444 /* Record the max length of recvmsg() calls for future allocations */
2445 nlk->max_recvmsg_len = max(nlk->max_recvmsg_len, len);
2446 nlk->max_recvmsg_len = min_t(size_t, nlk->max_recvmsg_len,
2449 copied = data_skb->len;
2451 msg->msg_flags |= MSG_TRUNC;
2455 skb_reset_transport_header(data_skb);
2456 err = skb_copy_datagram_msg(data_skb, 0, msg, copied);
2458 if (msg->msg_name) {
2459 DECLARE_SOCKADDR(struct sockaddr_nl *, addr, msg->msg_name);
2460 addr->nl_family = AF_NETLINK;
2462 addr->nl_pid = NETLINK_CB(skb).portid;
2463 addr->nl_groups = netlink_group_mask(NETLINK_CB(skb).dst_group);
2464 msg->msg_namelen = sizeof(*addr);
2467 if (nlk->flags & NETLINK_F_RECV_PKTINFO)
2468 netlink_cmsg_recv_pktinfo(msg, skb);
2469 if (nlk->flags & NETLINK_F_LISTEN_ALL_NSID)
2470 netlink_cmsg_listen_all_nsid(sk, msg, skb);
2472 memset(&scm, 0, sizeof(scm));
2473 scm.creds = *NETLINK_CREDS(skb);
2474 if (flags & MSG_TRUNC)
2475 copied = data_skb->len;
2477 skb_free_datagram(sk, skb);
2479 if (nlk->cb_running &&
2480 atomic_read(&sk->sk_rmem_alloc) <= sk->sk_rcvbuf / 2) {
2481 ret = netlink_dump(sk);
2484 sk->sk_error_report(sk);
2488 scm_recv(sock, msg, &scm, flags);
2490 netlink_rcv_wake(sk);
2491 return err ? : copied;
2494 static void netlink_data_ready(struct sock *sk)
2500 * We export these functions to other modules. They provide a
2501 * complete set of kernel non-blocking support for message
2506 __netlink_kernel_create(struct net *net, int unit, struct module *module,
2507 struct netlink_kernel_cfg *cfg)
2509 struct socket *sock;
2511 struct netlink_sock *nlk;
2512 struct listeners *listeners = NULL;
2513 struct mutex *cb_mutex = cfg ? cfg->cb_mutex : NULL;
2514 unsigned int groups;
2518 if (unit < 0 || unit >= MAX_LINKS)
2521 if (sock_create_lite(PF_NETLINK, SOCK_DGRAM, unit, &sock))
2524 if (__netlink_create(net, sock, cb_mutex, unit, 1) < 0)
2525 goto out_sock_release_nosk;
2529 if (!cfg || cfg->groups < 32)
2532 groups = cfg->groups;
2534 listeners = kzalloc(sizeof(*listeners) + NLGRPSZ(groups), GFP_KERNEL);
2536 goto out_sock_release;
2538 sk->sk_data_ready = netlink_data_ready;
2539 if (cfg && cfg->input)
2540 nlk_sk(sk)->netlink_rcv = cfg->input;
2542 if (netlink_insert(sk, 0))
2543 goto out_sock_release;
2546 nlk->flags |= NETLINK_F_KERNEL_SOCKET;
2548 netlink_table_grab();
2549 if (!nl_table[unit].registered) {
2550 nl_table[unit].groups = groups;
2551 rcu_assign_pointer(nl_table[unit].listeners, listeners);
2552 nl_table[unit].cb_mutex = cb_mutex;
2553 nl_table[unit].module = module;
2555 nl_table[unit].bind = cfg->bind;
2556 nl_table[unit].unbind = cfg->unbind;
2557 nl_table[unit].flags = cfg->flags;
2559 nl_table[unit].compare = cfg->compare;
2561 nl_table[unit].registered = 1;
2564 nl_table[unit].registered++;
2566 netlink_table_ungrab();
2571 netlink_kernel_release(sk);
2574 out_sock_release_nosk:
2578 EXPORT_SYMBOL(__netlink_kernel_create);
2581 netlink_kernel_release(struct sock *sk)
2583 if (sk == NULL || sk->sk_socket == NULL)
2586 sock_release(sk->sk_socket);
2588 EXPORT_SYMBOL(netlink_kernel_release);
2590 int __netlink_change_ngroups(struct sock *sk, unsigned int groups)
2592 struct listeners *new, *old;
2593 struct netlink_table *tbl = &nl_table[sk->sk_protocol];
2598 if (NLGRPSZ(tbl->groups) < NLGRPSZ(groups)) {
2599 new = kzalloc(sizeof(*new) + NLGRPSZ(groups), GFP_ATOMIC);
2602 old = nl_deref_protected(tbl->listeners);
2603 memcpy(new->masks, old->masks, NLGRPSZ(tbl->groups));
2604 rcu_assign_pointer(tbl->listeners, new);
2606 kfree_rcu(old, rcu);
2608 tbl->groups = groups;
2614 * netlink_change_ngroups - change number of multicast groups
2616 * This changes the number of multicast groups that are available
2617 * on a certain netlink family. Note that it is not possible to
2618 * change the number of groups to below 32. Also note that it does
2619 * not implicitly call netlink_clear_multicast_users() when the
2620 * number of groups is reduced.
2622 * @sk: The kernel netlink socket, as returned by netlink_kernel_create().
2623 * @groups: The new number of groups.
2625 int netlink_change_ngroups(struct sock *sk, unsigned int groups)
2629 netlink_table_grab();
2630 err = __netlink_change_ngroups(sk, groups);
2631 netlink_table_ungrab();
2636 void __netlink_clear_multicast_users(struct sock *ksk, unsigned int group)
2639 struct netlink_table *tbl = &nl_table[ksk->sk_protocol];
2641 sk_for_each_bound(sk, &tbl->mc_list)
2642 netlink_update_socket_mc(nlk_sk(sk), group, 0);
2646 __nlmsg_put(struct sk_buff *skb, u32 portid, u32 seq, int type, int len, int flags)
2648 struct nlmsghdr *nlh;
2649 int size = nlmsg_msg_size(len);
2651 nlh = (struct nlmsghdr *)skb_put(skb, NLMSG_ALIGN(size));
2652 nlh->nlmsg_type = type;
2653 nlh->nlmsg_len = size;
2654 nlh->nlmsg_flags = flags;
2655 nlh->nlmsg_pid = portid;
2656 nlh->nlmsg_seq = seq;
2657 if (!__builtin_constant_p(size) || NLMSG_ALIGN(size) - size != 0)
2658 memset(nlmsg_data(nlh) + len, 0, NLMSG_ALIGN(size) - size);
2661 EXPORT_SYMBOL(__nlmsg_put);
2664 * It looks a bit ugly.
2665 * It would be better to create kernel thread.
2668 static int netlink_dump(struct sock *sk)
2670 struct netlink_sock *nlk = nlk_sk(sk);
2671 struct netlink_callback *cb;
2672 struct sk_buff *skb = NULL;
2673 struct nlmsghdr *nlh;
2674 int len, err = -ENOBUFS;
2677 mutex_lock(nlk->cb_mutex);
2678 if (!nlk->cb_running) {
2684 alloc_size = max_t(int, cb->min_dump_alloc, NLMSG_GOODSIZE);
2686 if (!netlink_rx_is_mmaped(sk) &&
2687 atomic_read(&sk->sk_rmem_alloc) >= sk->sk_rcvbuf)
2690 /* NLMSG_GOODSIZE is small to avoid high order allocations being
2691 * required, but it makes sense to _attempt_ a 16K bytes allocation
2692 * to reduce number of system calls on dump operations, if user
2693 * ever provided a big enough buffer.
2695 if (alloc_size < nlk->max_recvmsg_len) {
2696 skb = netlink_alloc_skb(sk,
2697 nlk->max_recvmsg_len,
2702 /* available room should be exact amount to avoid MSG_TRUNC */
2704 skb_reserve(skb, skb_tailroom(skb) -
2705 nlk->max_recvmsg_len);
2708 skb = netlink_alloc_skb(sk, alloc_size, nlk->portid,
2712 netlink_skb_set_owner_r(skb, sk);
2714 len = cb->dump(skb, cb);
2717 mutex_unlock(nlk->cb_mutex);
2719 if (sk_filter(sk, skb))
2722 __netlink_sendskb(sk, skb);
2726 nlh = nlmsg_put_answer(skb, cb, NLMSG_DONE, sizeof(len), NLM_F_MULTI);
2730 nl_dump_check_consistent(cb, nlh);
2732 memcpy(nlmsg_data(nlh), &len, sizeof(len));
2734 if (sk_filter(sk, skb))
2737 __netlink_sendskb(sk, skb);
2742 nlk->cb_running = false;
2743 mutex_unlock(nlk->cb_mutex);
2744 module_put(cb->module);
2745 consume_skb(cb->skb);
2749 mutex_unlock(nlk->cb_mutex);
2754 int __netlink_dump_start(struct sock *ssk, struct sk_buff *skb,
2755 const struct nlmsghdr *nlh,
2756 struct netlink_dump_control *control)
2758 struct netlink_callback *cb;
2760 struct netlink_sock *nlk;
2763 /* Memory mapped dump requests need to be copied to avoid looping
2764 * on the pending state in netlink_mmap_sendmsg() while the CB hold
2765 * a reference to the skb.
2767 if (netlink_skb_is_mmaped(skb)) {
2768 skb = skb_copy(skb, GFP_KERNEL);
2772 atomic_inc(&skb->users);
2774 sk = netlink_lookup(sock_net(ssk), ssk->sk_protocol, NETLINK_CB(skb).portid);
2776 ret = -ECONNREFUSED;
2781 mutex_lock(nlk->cb_mutex);
2782 /* A dump is in progress... */
2783 if (nlk->cb_running) {
2787 /* add reference of module which cb->dump belongs to */
2788 if (!try_module_get(control->module)) {
2789 ret = -EPROTONOSUPPORT;
2794 memset(cb, 0, sizeof(*cb));
2795 cb->dump = control->dump;
2796 cb->done = control->done;
2798 cb->data = control->data;
2799 cb->module = control->module;
2800 cb->min_dump_alloc = control->min_dump_alloc;
2803 nlk->cb_running = true;
2805 mutex_unlock(nlk->cb_mutex);
2807 ret = netlink_dump(sk);
2813 /* We successfully started a dump, by returning -EINTR we
2814 * signal not to send ACK even if it was requested.
2820 mutex_unlock(nlk->cb_mutex);
2825 EXPORT_SYMBOL(__netlink_dump_start);
2827 void netlink_ack(struct sk_buff *in_skb, struct nlmsghdr *nlh, int err)
2829 struct sk_buff *skb;
2830 struct nlmsghdr *rep;
2831 struct nlmsgerr *errmsg;
2832 size_t payload = sizeof(*errmsg);
2834 /* error messages get the original request appened */
2836 payload += nlmsg_len(nlh);
2838 skb = netlink_alloc_skb(in_skb->sk, nlmsg_total_size(payload),
2839 NETLINK_CB(in_skb).portid, GFP_KERNEL);
2843 sk = netlink_lookup(sock_net(in_skb->sk),
2844 in_skb->sk->sk_protocol,
2845 NETLINK_CB(in_skb).portid);
2847 sk->sk_err = ENOBUFS;
2848 sk->sk_error_report(sk);
2854 rep = __nlmsg_put(skb, NETLINK_CB(in_skb).portid, nlh->nlmsg_seq,
2855 NLMSG_ERROR, payload, 0);
2856 errmsg = nlmsg_data(rep);
2857 errmsg->error = err;
2858 memcpy(&errmsg->msg, nlh, err ? nlh->nlmsg_len : sizeof(*nlh));
2859 netlink_unicast(in_skb->sk, skb, NETLINK_CB(in_skb).portid, MSG_DONTWAIT);
2861 EXPORT_SYMBOL(netlink_ack);
2863 int netlink_rcv_skb(struct sk_buff *skb, int (*cb)(struct sk_buff *,
2866 struct nlmsghdr *nlh;
2869 while (skb->len >= nlmsg_total_size(0)) {
2872 nlh = nlmsg_hdr(skb);
2875 if (nlh->nlmsg_len < NLMSG_HDRLEN || skb->len < nlh->nlmsg_len)
2878 /* Only requests are handled by the kernel */
2879 if (!(nlh->nlmsg_flags & NLM_F_REQUEST))
2882 /* Skip control messages */
2883 if (nlh->nlmsg_type < NLMSG_MIN_TYPE)
2891 if (nlh->nlmsg_flags & NLM_F_ACK || err)
2892 netlink_ack(skb, nlh, err);
2895 msglen = NLMSG_ALIGN(nlh->nlmsg_len);
2896 if (msglen > skb->len)
2898 skb_pull(skb, msglen);
2903 EXPORT_SYMBOL(netlink_rcv_skb);
2906 * nlmsg_notify - send a notification netlink message
2907 * @sk: netlink socket to use
2908 * @skb: notification message
2909 * @portid: destination netlink portid for reports or 0
2910 * @group: destination multicast group or 0
2911 * @report: 1 to report back, 0 to disable
2912 * @flags: allocation flags
2914 int nlmsg_notify(struct sock *sk, struct sk_buff *skb, u32 portid,
2915 unsigned int group, int report, gfp_t flags)
2920 int exclude_portid = 0;
2923 atomic_inc(&skb->users);
2924 exclude_portid = portid;
2927 /* errors reported via destination sk->sk_err, but propagate
2928 * delivery errors if NETLINK_BROADCAST_ERROR flag is set */
2929 err = nlmsg_multicast(sk, skb, exclude_portid, group, flags);
2935 err2 = nlmsg_unicast(sk, skb, portid);
2936 if (!err || err == -ESRCH)
2942 EXPORT_SYMBOL(nlmsg_notify);
2944 #ifdef CONFIG_PROC_FS
2945 struct nl_seq_iter {
2946 struct seq_net_private p;
2947 struct rhashtable_iter hti;
2951 static int netlink_walk_start(struct nl_seq_iter *iter)
2955 err = rhashtable_walk_init(&nl_table[iter->link].hash, &iter->hti);
2957 iter->link = MAX_LINKS;
2961 err = rhashtable_walk_start(&iter->hti);
2962 return err == -EAGAIN ? 0 : err;
2965 static void netlink_walk_stop(struct nl_seq_iter *iter)
2967 rhashtable_walk_stop(&iter->hti);
2968 rhashtable_walk_exit(&iter->hti);
2971 static void *__netlink_seq_next(struct seq_file *seq)
2973 struct nl_seq_iter *iter = seq->private;
2974 struct netlink_sock *nlk;
2980 nlk = rhashtable_walk_next(&iter->hti);
2983 if (PTR_ERR(nlk) == -EAGAIN)
2992 netlink_walk_stop(iter);
2993 if (++iter->link >= MAX_LINKS)
2996 err = netlink_walk_start(iter);
2998 return ERR_PTR(err);
3000 } while (sock_net(&nlk->sk) != seq_file_net(seq));
3005 static void *netlink_seq_start(struct seq_file *seq, loff_t *posp)
3007 struct nl_seq_iter *iter = seq->private;
3008 void *obj = SEQ_START_TOKEN;
3014 err = netlink_walk_start(iter);
3016 return ERR_PTR(err);
3018 for (pos = *posp; pos && obj && !IS_ERR(obj); pos--)
3019 obj = __netlink_seq_next(seq);
3024 static void *netlink_seq_next(struct seq_file *seq, void *v, loff_t *pos)
3027 return __netlink_seq_next(seq);
3030 static void netlink_seq_stop(struct seq_file *seq, void *v)
3032 struct nl_seq_iter *iter = seq->private;
3034 if (iter->link >= MAX_LINKS)
3037 netlink_walk_stop(iter);
3041 static int netlink_seq_show(struct seq_file *seq, void *v)
3043 if (v == SEQ_START_TOKEN) {
3045 "sk Eth Pid Groups "
3046 "Rmem Wmem Dump Locks Drops Inode\n");
3049 struct netlink_sock *nlk = nlk_sk(s);
3051 seq_printf(seq, "%pK %-3d %-6u %08x %-8d %-8d %d %-8d %-8d %-8lu\n",
3055 nlk->groups ? (u32)nlk->groups[0] : 0,
3056 sk_rmem_alloc_get(s),
3057 sk_wmem_alloc_get(s),
3059 atomic_read(&s->sk_refcnt),
3060 atomic_read(&s->sk_drops),
3068 static const struct seq_operations netlink_seq_ops = {
3069 .start = netlink_seq_start,
3070 .next = netlink_seq_next,
3071 .stop = netlink_seq_stop,
3072 .show = netlink_seq_show,
3076 static int netlink_seq_open(struct inode *inode, struct file *file)
3078 return seq_open_net(inode, file, &netlink_seq_ops,
3079 sizeof(struct nl_seq_iter));
3082 static const struct file_operations netlink_seq_fops = {
3083 .owner = THIS_MODULE,
3084 .open = netlink_seq_open,
3086 .llseek = seq_lseek,
3087 .release = seq_release_net,
3092 int netlink_register_notifier(struct notifier_block *nb)
3094 return atomic_notifier_chain_register(&netlink_chain, nb);
3096 EXPORT_SYMBOL(netlink_register_notifier);
3098 int netlink_unregister_notifier(struct notifier_block *nb)
3100 return atomic_notifier_chain_unregister(&netlink_chain, nb);
3102 EXPORT_SYMBOL(netlink_unregister_notifier);
3104 static const struct proto_ops netlink_ops = {
3105 .family = PF_NETLINK,
3106 .owner = THIS_MODULE,
3107 .release = netlink_release,
3108 .bind = netlink_bind,
3109 .connect = netlink_connect,
3110 .socketpair = sock_no_socketpair,
3111 .accept = sock_no_accept,
3112 .getname = netlink_getname,
3113 .poll = netlink_poll,
3114 .ioctl = sock_no_ioctl,
3115 .listen = sock_no_listen,
3116 .shutdown = sock_no_shutdown,
3117 .setsockopt = netlink_setsockopt,
3118 .getsockopt = netlink_getsockopt,
3119 .sendmsg = netlink_sendmsg,
3120 .recvmsg = netlink_recvmsg,
3121 .mmap = netlink_mmap,
3122 .sendpage = sock_no_sendpage,
3125 static const struct net_proto_family netlink_family_ops = {
3126 .family = PF_NETLINK,
3127 .create = netlink_create,
3128 .owner = THIS_MODULE, /* for consistency 8) */
3131 static int __net_init netlink_net_init(struct net *net)
3133 #ifdef CONFIG_PROC_FS
3134 if (!proc_create("netlink", 0, net->proc_net, &netlink_seq_fops))
3140 static void __net_exit netlink_net_exit(struct net *net)
3142 #ifdef CONFIG_PROC_FS
3143 remove_proc_entry("netlink", net->proc_net);
3147 static void __init netlink_add_usersock_entry(void)
3149 struct listeners *listeners;
3152 listeners = kzalloc(sizeof(*listeners) + NLGRPSZ(groups), GFP_KERNEL);
3154 panic("netlink_add_usersock_entry: Cannot allocate listeners\n");
3156 netlink_table_grab();
3158 nl_table[NETLINK_USERSOCK].groups = groups;
3159 rcu_assign_pointer(nl_table[NETLINK_USERSOCK].listeners, listeners);
3160 nl_table[NETLINK_USERSOCK].module = THIS_MODULE;
3161 nl_table[NETLINK_USERSOCK].registered = 1;
3162 nl_table[NETLINK_USERSOCK].flags = NL_CFG_F_NONROOT_SEND;
3164 netlink_table_ungrab();
3167 static struct pernet_operations __net_initdata netlink_net_ops = {
3168 .init = netlink_net_init,
3169 .exit = netlink_net_exit,
3172 static inline u32 netlink_hash(const void *data, u32 len, u32 seed)
3174 const struct netlink_sock *nlk = data;
3175 struct netlink_compare_arg arg;
3177 netlink_compare_arg_init(&arg, sock_net(&nlk->sk), nlk->portid);
3178 return jhash2((u32 *)&arg, netlink_compare_arg_len / sizeof(u32), seed);
3181 static const struct rhashtable_params netlink_rhashtable_params = {
3182 .head_offset = offsetof(struct netlink_sock, node),
3183 .key_len = netlink_compare_arg_len,
3184 .obj_hashfn = netlink_hash,
3185 .obj_cmpfn = netlink_compare,
3186 .automatic_shrinking = true,
3189 static int __init netlink_proto_init(void)
3192 int err = proto_register(&netlink_proto, 0);
3197 BUILD_BUG_ON(sizeof(struct netlink_skb_parms) > FIELD_SIZEOF(struct sk_buff, cb));
3199 nl_table = kcalloc(MAX_LINKS, sizeof(*nl_table), GFP_KERNEL);
3203 for (i = 0; i < MAX_LINKS; i++) {
3204 if (rhashtable_init(&nl_table[i].hash,
3205 &netlink_rhashtable_params) < 0) {
3207 rhashtable_destroy(&nl_table[i].hash);
3213 INIT_LIST_HEAD(&netlink_tap_all);
3215 netlink_add_usersock_entry();
3217 sock_register(&netlink_family_ops);
3218 register_pernet_subsys(&netlink_net_ops);
3219 /* The netlink device handler may be needed early. */
3224 panic("netlink_init: Cannot allocate nl_table\n");
3227 core_initcall(netlink_proto_init);