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);
1302 static s32 rover = -4097;
1307 if (__netlink_lookup(table, portid, net)) {
1308 /* Bind collision, search negative portid values. */
1317 err = netlink_insert(sk, portid);
1318 if (err == -EADDRINUSE)
1321 /* If 2 threads race to autobind, that is fine. */
1329 * __netlink_ns_capable - General netlink message capability test
1330 * @nsp: NETLINK_CB of the socket buffer holding a netlink command from userspace.
1331 * @user_ns: The user namespace of the capability to use
1332 * @cap: The capability to use
1334 * Test to see if the opener of the socket we received the message
1335 * from had when the netlink socket was created and the sender of the
1336 * message has has the capability @cap in the user namespace @user_ns.
1338 bool __netlink_ns_capable(const struct netlink_skb_parms *nsp,
1339 struct user_namespace *user_ns, int cap)
1341 return ((nsp->flags & NETLINK_SKB_DST) ||
1342 file_ns_capable(nsp->sk->sk_socket->file, user_ns, cap)) &&
1343 ns_capable(user_ns, cap);
1345 EXPORT_SYMBOL(__netlink_ns_capable);
1348 * netlink_ns_capable - General netlink message capability test
1349 * @skb: socket buffer holding a netlink command from userspace
1350 * @user_ns: The user namespace of the capability to use
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 the user namespace @user_ns.
1357 bool netlink_ns_capable(const struct sk_buff *skb,
1358 struct user_namespace *user_ns, int cap)
1360 return __netlink_ns_capable(&NETLINK_CB(skb), user_ns, cap);
1362 EXPORT_SYMBOL(netlink_ns_capable);
1365 * netlink_capable - Netlink global message capability test
1366 * @skb: socket buffer holding a netlink command from userspace
1367 * @cap: The capability to use
1369 * Test to see if the opener of the socket we received the message
1370 * from had when the netlink socket was created and the sender of the
1371 * message has has the capability @cap in all user namespaces.
1373 bool netlink_capable(const struct sk_buff *skb, int cap)
1375 return netlink_ns_capable(skb, &init_user_ns, cap);
1377 EXPORT_SYMBOL(netlink_capable);
1380 * netlink_net_capable - Netlink network namespace message capability test
1381 * @skb: socket buffer holding a netlink command from userspace
1382 * @cap: The capability to use
1384 * Test to see if the opener of the socket we received the message
1385 * from had when the netlink socket was created and the sender of the
1386 * message has has the capability @cap over the network namespace of
1387 * the socket we received the message from.
1389 bool netlink_net_capable(const struct sk_buff *skb, int cap)
1391 return netlink_ns_capable(skb, sock_net(skb->sk)->user_ns, cap);
1393 EXPORT_SYMBOL(netlink_net_capable);
1395 static inline int netlink_allowed(const struct socket *sock, unsigned int flag)
1397 return (nl_table[sock->sk->sk_protocol].flags & flag) ||
1398 ns_capable(sock_net(sock->sk)->user_ns, CAP_NET_ADMIN);
1402 netlink_update_subscriptions(struct sock *sk, unsigned int subscriptions)
1404 struct netlink_sock *nlk = nlk_sk(sk);
1406 if (nlk->subscriptions && !subscriptions)
1407 __sk_del_bind_node(sk);
1408 else if (!nlk->subscriptions && subscriptions)
1409 sk_add_bind_node(sk, &nl_table[sk->sk_protocol].mc_list);
1410 nlk->subscriptions = subscriptions;
1413 static int netlink_realloc_groups(struct sock *sk)
1415 struct netlink_sock *nlk = nlk_sk(sk);
1416 unsigned int groups;
1417 unsigned long *new_groups;
1420 netlink_table_grab();
1422 groups = nl_table[sk->sk_protocol].groups;
1423 if (!nl_table[sk->sk_protocol].registered) {
1428 if (nlk->ngroups >= groups)
1431 new_groups = krealloc(nlk->groups, NLGRPSZ(groups), GFP_ATOMIC);
1432 if (new_groups == NULL) {
1436 memset((char *)new_groups + NLGRPSZ(nlk->ngroups), 0,
1437 NLGRPSZ(groups) - NLGRPSZ(nlk->ngroups));
1439 nlk->groups = new_groups;
1440 nlk->ngroups = groups;
1442 netlink_table_ungrab();
1446 static void netlink_undo_bind(int group, long unsigned int groups,
1449 struct netlink_sock *nlk = nlk_sk(sk);
1452 if (!nlk->netlink_unbind)
1455 for (undo = 0; undo < group; undo++)
1456 if (test_bit(undo, &groups))
1457 nlk->netlink_unbind(sock_net(sk), undo + 1);
1460 static int netlink_bind(struct socket *sock, struct sockaddr *addr,
1463 struct sock *sk = sock->sk;
1464 struct net *net = sock_net(sk);
1465 struct netlink_sock *nlk = nlk_sk(sk);
1466 struct sockaddr_nl *nladdr = (struct sockaddr_nl *)addr;
1468 long unsigned int groups = nladdr->nl_groups;
1470 if (addr_len < sizeof(struct sockaddr_nl))
1473 if (nladdr->nl_family != AF_NETLINK)
1476 /* Only superuser is allowed to listen multicasts */
1478 if (!netlink_allowed(sock, NL_CFG_F_NONROOT_RECV))
1480 err = netlink_realloc_groups(sk);
1486 if (nladdr->nl_pid != nlk->portid)
1489 if (nlk->netlink_bind && groups) {
1492 for (group = 0; group < nlk->ngroups; group++) {
1493 if (!test_bit(group, &groups))
1495 err = nlk->netlink_bind(net, group + 1);
1498 netlink_undo_bind(group, groups, sk);
1504 err = nladdr->nl_pid ?
1505 netlink_insert(sk, nladdr->nl_pid) :
1506 netlink_autobind(sock);
1508 netlink_undo_bind(nlk->ngroups, groups, sk);
1513 if (!groups && (nlk->groups == NULL || !(u32)nlk->groups[0]))
1516 netlink_table_grab();
1517 netlink_update_subscriptions(sk, nlk->subscriptions +
1519 hweight32(nlk->groups[0]));
1520 nlk->groups[0] = (nlk->groups[0] & ~0xffffffffUL) | groups;
1521 netlink_update_listeners(sk);
1522 netlink_table_ungrab();
1527 static int netlink_connect(struct socket *sock, struct sockaddr *addr,
1528 int alen, int flags)
1531 struct sock *sk = sock->sk;
1532 struct netlink_sock *nlk = nlk_sk(sk);
1533 struct sockaddr_nl *nladdr = (struct sockaddr_nl *)addr;
1535 if (alen < sizeof(addr->sa_family))
1538 if (addr->sa_family == AF_UNSPEC) {
1539 sk->sk_state = NETLINK_UNCONNECTED;
1540 nlk->dst_portid = 0;
1544 if (addr->sa_family != AF_NETLINK)
1547 if ((nladdr->nl_groups || nladdr->nl_pid) &&
1548 !netlink_allowed(sock, NL_CFG_F_NONROOT_SEND))
1552 err = netlink_autobind(sock);
1555 sk->sk_state = NETLINK_CONNECTED;
1556 nlk->dst_portid = nladdr->nl_pid;
1557 nlk->dst_group = ffs(nladdr->nl_groups);
1563 static int netlink_getname(struct socket *sock, struct sockaddr *addr,
1564 int *addr_len, int peer)
1566 struct sock *sk = sock->sk;
1567 struct netlink_sock *nlk = nlk_sk(sk);
1568 DECLARE_SOCKADDR(struct sockaddr_nl *, nladdr, addr);
1570 nladdr->nl_family = AF_NETLINK;
1572 *addr_len = sizeof(*nladdr);
1575 nladdr->nl_pid = nlk->dst_portid;
1576 nladdr->nl_groups = netlink_group_mask(nlk->dst_group);
1578 nladdr->nl_pid = nlk->portid;
1579 nladdr->nl_groups = nlk->groups ? nlk->groups[0] : 0;
1584 static struct sock *netlink_getsockbyportid(struct sock *ssk, u32 portid)
1587 struct netlink_sock *nlk;
1589 sock = netlink_lookup(sock_net(ssk), ssk->sk_protocol, portid);
1591 return ERR_PTR(-ECONNREFUSED);
1593 /* Don't bother queuing skb if kernel socket has no input function */
1595 if (sock->sk_state == NETLINK_CONNECTED &&
1596 nlk->dst_portid != nlk_sk(ssk)->portid) {
1598 return ERR_PTR(-ECONNREFUSED);
1603 struct sock *netlink_getsockbyfilp(struct file *filp)
1605 struct inode *inode = file_inode(filp);
1608 if (!S_ISSOCK(inode->i_mode))
1609 return ERR_PTR(-ENOTSOCK);
1611 sock = SOCKET_I(inode)->sk;
1612 if (sock->sk_family != AF_NETLINK)
1613 return ERR_PTR(-EINVAL);
1619 static struct sk_buff *netlink_alloc_large_skb(unsigned int size,
1622 struct sk_buff *skb;
1625 if (size <= NLMSG_GOODSIZE || broadcast)
1626 return alloc_skb(size, GFP_KERNEL);
1628 size = SKB_DATA_ALIGN(size) +
1629 SKB_DATA_ALIGN(sizeof(struct skb_shared_info));
1631 data = vmalloc(size);
1635 skb = __build_skb(data, size);
1639 skb->destructor = netlink_skb_destructor;
1645 * Attach a skb to a netlink socket.
1646 * The caller must hold a reference to the destination socket. On error, the
1647 * reference is dropped. The skb is not send to the destination, just all
1648 * all error checks are performed and memory in the queue is reserved.
1650 * < 0: error. skb freed, reference to sock dropped.
1652 * 1: repeat lookup - reference dropped while waiting for socket memory.
1654 int netlink_attachskb(struct sock *sk, struct sk_buff *skb,
1655 long *timeo, struct sock *ssk)
1657 struct netlink_sock *nlk;
1661 if ((atomic_read(&sk->sk_rmem_alloc) > sk->sk_rcvbuf ||
1662 test_bit(NETLINK_S_CONGESTED, &nlk->state)) &&
1663 !netlink_skb_is_mmaped(skb)) {
1664 DECLARE_WAITQUEUE(wait, current);
1666 if (!ssk || netlink_is_kernel(ssk))
1667 netlink_overrun(sk);
1673 __set_current_state(TASK_INTERRUPTIBLE);
1674 add_wait_queue(&nlk->wait, &wait);
1676 if ((atomic_read(&sk->sk_rmem_alloc) > sk->sk_rcvbuf ||
1677 test_bit(NETLINK_S_CONGESTED, &nlk->state)) &&
1678 !sock_flag(sk, SOCK_DEAD))
1679 *timeo = schedule_timeout(*timeo);
1681 __set_current_state(TASK_RUNNING);
1682 remove_wait_queue(&nlk->wait, &wait);
1685 if (signal_pending(current)) {
1687 return sock_intr_errno(*timeo);
1691 netlink_skb_set_owner_r(skb, sk);
1695 static int __netlink_sendskb(struct sock *sk, struct sk_buff *skb)
1699 netlink_deliver_tap(skb);
1701 #ifdef CONFIG_NETLINK_MMAP
1702 if (netlink_skb_is_mmaped(skb))
1703 netlink_queue_mmaped_skb(sk, skb);
1704 else if (netlink_rx_is_mmaped(sk))
1705 netlink_ring_set_copied(sk, skb);
1707 #endif /* CONFIG_NETLINK_MMAP */
1708 skb_queue_tail(&sk->sk_receive_queue, skb);
1709 sk->sk_data_ready(sk);
1713 int netlink_sendskb(struct sock *sk, struct sk_buff *skb)
1715 int len = __netlink_sendskb(sk, skb);
1721 void netlink_detachskb(struct sock *sk, struct sk_buff *skb)
1727 static struct sk_buff *netlink_trim(struct sk_buff *skb, gfp_t allocation)
1731 WARN_ON(skb->sk != NULL);
1732 if (netlink_skb_is_mmaped(skb))
1735 delta = skb->end - skb->tail;
1736 if (is_vmalloc_addr(skb->head) || delta * 2 < skb->truesize)
1739 if (skb_shared(skb)) {
1740 struct sk_buff *nskb = skb_clone(skb, allocation);
1747 if (!pskb_expand_head(skb, 0, -delta, allocation))
1748 skb->truesize -= delta;
1753 static int netlink_unicast_kernel(struct sock *sk, struct sk_buff *skb,
1757 struct netlink_sock *nlk = nlk_sk(sk);
1759 ret = -ECONNREFUSED;
1760 if (nlk->netlink_rcv != NULL) {
1762 netlink_skb_set_owner_r(skb, sk);
1763 NETLINK_CB(skb).sk = ssk;
1764 netlink_deliver_tap_kernel(sk, ssk, skb);
1765 nlk->netlink_rcv(skb);
1774 int netlink_unicast(struct sock *ssk, struct sk_buff *skb,
1775 u32 portid, int nonblock)
1781 skb = netlink_trim(skb, gfp_any());
1783 timeo = sock_sndtimeo(ssk, nonblock);
1785 sk = netlink_getsockbyportid(ssk, portid);
1790 if (netlink_is_kernel(sk))
1791 return netlink_unicast_kernel(sk, skb, ssk);
1793 if (sk_filter(sk, skb)) {
1800 err = netlink_attachskb(sk, skb, &timeo, ssk);
1806 return netlink_sendskb(sk, skb);
1808 EXPORT_SYMBOL(netlink_unicast);
1810 struct sk_buff *netlink_alloc_skb(struct sock *ssk, unsigned int size,
1811 u32 dst_portid, gfp_t gfp_mask)
1813 #ifdef CONFIG_NETLINK_MMAP
1814 struct sock *sk = NULL;
1815 struct sk_buff *skb;
1816 struct netlink_ring *ring;
1817 struct nl_mmap_hdr *hdr;
1818 unsigned int maxlen;
1820 sk = netlink_getsockbyportid(ssk, dst_portid);
1824 ring = &nlk_sk(sk)->rx_ring;
1825 /* fast-path without atomic ops for common case: non-mmaped receiver */
1826 if (ring->pg_vec == NULL)
1829 if (ring->frame_size - NL_MMAP_HDRLEN < size)
1832 skb = alloc_skb_head(gfp_mask);
1836 spin_lock_bh(&sk->sk_receive_queue.lock);
1837 /* check again under lock */
1838 if (ring->pg_vec == NULL)
1841 /* check again under lock */
1842 maxlen = ring->frame_size - NL_MMAP_HDRLEN;
1846 netlink_forward_ring(ring);
1847 hdr = netlink_current_frame(ring, NL_MMAP_STATUS_UNUSED);
1850 netlink_ring_setup_skb(skb, sk, ring, hdr);
1851 netlink_set_status(hdr, NL_MMAP_STATUS_RESERVED);
1852 atomic_inc(&ring->pending);
1853 netlink_increment_head(ring);
1855 spin_unlock_bh(&sk->sk_receive_queue.lock);
1860 spin_unlock_bh(&sk->sk_receive_queue.lock);
1861 netlink_overrun(sk);
1868 spin_unlock_bh(&sk->sk_receive_queue.lock);
1873 return alloc_skb(size, gfp_mask);
1875 EXPORT_SYMBOL_GPL(netlink_alloc_skb);
1877 int netlink_has_listeners(struct sock *sk, unsigned int group)
1880 struct listeners *listeners;
1882 BUG_ON(!netlink_is_kernel(sk));
1885 listeners = rcu_dereference(nl_table[sk->sk_protocol].listeners);
1887 if (listeners && group - 1 < nl_table[sk->sk_protocol].groups)
1888 res = test_bit(group - 1, listeners->masks);
1894 EXPORT_SYMBOL_GPL(netlink_has_listeners);
1896 static int netlink_broadcast_deliver(struct sock *sk, struct sk_buff *skb)
1898 struct netlink_sock *nlk = nlk_sk(sk);
1900 if (atomic_read(&sk->sk_rmem_alloc) <= sk->sk_rcvbuf &&
1901 !test_bit(NETLINK_S_CONGESTED, &nlk->state)) {
1902 netlink_skb_set_owner_r(skb, sk);
1903 __netlink_sendskb(sk, skb);
1904 return atomic_read(&sk->sk_rmem_alloc) > (sk->sk_rcvbuf >> 1);
1909 struct netlink_broadcast_data {
1910 struct sock *exclude_sk;
1915 int delivery_failure;
1919 struct sk_buff *skb, *skb2;
1920 int (*tx_filter)(struct sock *dsk, struct sk_buff *skb, void *data);
1924 static void do_one_broadcast(struct sock *sk,
1925 struct netlink_broadcast_data *p)
1927 struct netlink_sock *nlk = nlk_sk(sk);
1930 if (p->exclude_sk == sk)
1933 if (nlk->portid == p->portid || p->group - 1 >= nlk->ngroups ||
1934 !test_bit(p->group - 1, nlk->groups))
1937 if (!net_eq(sock_net(sk), p->net)) {
1938 if (!(nlk->flags & NETLINK_F_LISTEN_ALL_NSID))
1941 if (!peernet_has_id(sock_net(sk), p->net))
1944 if (!file_ns_capable(sk->sk_socket->file, p->net->user_ns,
1950 netlink_overrun(sk);
1955 if (p->skb2 == NULL) {
1956 if (skb_shared(p->skb)) {
1957 p->skb2 = skb_clone(p->skb, p->allocation);
1959 p->skb2 = skb_get(p->skb);
1961 * skb ownership may have been set when
1962 * delivered to a previous socket.
1964 skb_orphan(p->skb2);
1967 if (p->skb2 == NULL) {
1968 netlink_overrun(sk);
1969 /* Clone failed. Notify ALL listeners. */
1971 if (nlk->flags & NETLINK_F_BROADCAST_SEND_ERROR)
1972 p->delivery_failure = 1;
1975 if (p->tx_filter && p->tx_filter(sk, p->skb2, p->tx_data)) {
1980 if (sk_filter(sk, p->skb2)) {
1985 NETLINK_CB(p->skb2).nsid = peernet2id(sock_net(sk), p->net);
1986 NETLINK_CB(p->skb2).nsid_is_set = true;
1987 val = netlink_broadcast_deliver(sk, p->skb2);
1989 netlink_overrun(sk);
1990 if (nlk->flags & NETLINK_F_BROADCAST_SEND_ERROR)
1991 p->delivery_failure = 1;
1993 p->congested |= val;
2001 int netlink_broadcast_filtered(struct sock *ssk, struct sk_buff *skb, u32 portid,
2002 u32 group, gfp_t allocation,
2003 int (*filter)(struct sock *dsk, struct sk_buff *skb, void *data),
2006 struct net *net = sock_net(ssk);
2007 struct netlink_broadcast_data info;
2010 skb = netlink_trim(skb, allocation);
2012 info.exclude_sk = ssk;
2014 info.portid = portid;
2017 info.delivery_failure = 0;
2020 info.allocation = allocation;
2023 info.tx_filter = filter;
2024 info.tx_data = filter_data;
2026 /* While we sleep in clone, do not allow to change socket list */
2028 netlink_lock_table();
2030 sk_for_each_bound(sk, &nl_table[ssk->sk_protocol].mc_list)
2031 do_one_broadcast(sk, &info);
2035 netlink_unlock_table();
2037 if (info.delivery_failure) {
2038 kfree_skb(info.skb2);
2041 consume_skb(info.skb2);
2043 if (info.delivered) {
2044 if (info.congested && (allocation & __GFP_WAIT))
2050 EXPORT_SYMBOL(netlink_broadcast_filtered);
2052 int netlink_broadcast(struct sock *ssk, struct sk_buff *skb, u32 portid,
2053 u32 group, gfp_t allocation)
2055 return netlink_broadcast_filtered(ssk, skb, portid, group, allocation,
2058 EXPORT_SYMBOL(netlink_broadcast);
2060 struct netlink_set_err_data {
2061 struct sock *exclude_sk;
2067 static int do_one_set_err(struct sock *sk, struct netlink_set_err_data *p)
2069 struct netlink_sock *nlk = nlk_sk(sk);
2072 if (sk == p->exclude_sk)
2075 if (!net_eq(sock_net(sk), sock_net(p->exclude_sk)))
2078 if (nlk->portid == p->portid || p->group - 1 >= nlk->ngroups ||
2079 !test_bit(p->group - 1, nlk->groups))
2082 if (p->code == ENOBUFS && nlk->flags & NETLINK_F_RECV_NO_ENOBUFS) {
2087 sk->sk_err = p->code;
2088 sk->sk_error_report(sk);
2094 * netlink_set_err - report error to broadcast listeners
2095 * @ssk: the kernel netlink socket, as returned by netlink_kernel_create()
2096 * @portid: the PORTID of a process that we want to skip (if any)
2097 * @group: the broadcast group that will notice the error
2098 * @code: error code, must be negative (as usual in kernelspace)
2100 * This function returns the number of broadcast listeners that have set the
2101 * NETLINK_NO_ENOBUFS socket option.
2103 int netlink_set_err(struct sock *ssk, u32 portid, u32 group, int code)
2105 struct netlink_set_err_data info;
2109 info.exclude_sk = ssk;
2110 info.portid = portid;
2112 /* sk->sk_err wants a positive error value */
2115 read_lock(&nl_table_lock);
2117 sk_for_each_bound(sk, &nl_table[ssk->sk_protocol].mc_list)
2118 ret += do_one_set_err(sk, &info);
2120 read_unlock(&nl_table_lock);
2123 EXPORT_SYMBOL(netlink_set_err);
2125 /* must be called with netlink table grabbed */
2126 static void netlink_update_socket_mc(struct netlink_sock *nlk,
2130 int old, new = !!is_new, subscriptions;
2132 old = test_bit(group - 1, nlk->groups);
2133 subscriptions = nlk->subscriptions - old + new;
2135 __set_bit(group - 1, nlk->groups);
2137 __clear_bit(group - 1, nlk->groups);
2138 netlink_update_subscriptions(&nlk->sk, subscriptions);
2139 netlink_update_listeners(&nlk->sk);
2142 static int netlink_setsockopt(struct socket *sock, int level, int optname,
2143 char __user *optval, unsigned int optlen)
2145 struct sock *sk = sock->sk;
2146 struct netlink_sock *nlk = nlk_sk(sk);
2147 unsigned int val = 0;
2150 if (level != SOL_NETLINK)
2151 return -ENOPROTOOPT;
2153 if (optname != NETLINK_RX_RING && optname != NETLINK_TX_RING &&
2154 optlen >= sizeof(int) &&
2155 get_user(val, (unsigned int __user *)optval))
2159 case NETLINK_PKTINFO:
2161 nlk->flags |= NETLINK_F_RECV_PKTINFO;
2163 nlk->flags &= ~NETLINK_F_RECV_PKTINFO;
2166 case NETLINK_ADD_MEMBERSHIP:
2167 case NETLINK_DROP_MEMBERSHIP: {
2168 if (!netlink_allowed(sock, NL_CFG_F_NONROOT_RECV))
2170 err = netlink_realloc_groups(sk);
2173 if (!val || val - 1 >= nlk->ngroups)
2175 if (optname == NETLINK_ADD_MEMBERSHIP && nlk->netlink_bind) {
2176 err = nlk->netlink_bind(sock_net(sk), val);
2180 netlink_table_grab();
2181 netlink_update_socket_mc(nlk, val,
2182 optname == NETLINK_ADD_MEMBERSHIP);
2183 netlink_table_ungrab();
2184 if (optname == NETLINK_DROP_MEMBERSHIP && nlk->netlink_unbind)
2185 nlk->netlink_unbind(sock_net(sk), val);
2190 case NETLINK_BROADCAST_ERROR:
2192 nlk->flags |= NETLINK_F_BROADCAST_SEND_ERROR;
2194 nlk->flags &= ~NETLINK_F_BROADCAST_SEND_ERROR;
2197 case NETLINK_NO_ENOBUFS:
2199 nlk->flags |= NETLINK_F_RECV_NO_ENOBUFS;
2200 clear_bit(NETLINK_S_CONGESTED, &nlk->state);
2201 wake_up_interruptible(&nlk->wait);
2203 nlk->flags &= ~NETLINK_F_RECV_NO_ENOBUFS;
2207 #ifdef CONFIG_NETLINK_MMAP
2208 case NETLINK_RX_RING:
2209 case NETLINK_TX_RING: {
2210 struct nl_mmap_req req;
2212 /* Rings might consume more memory than queue limits, require
2215 if (!capable(CAP_NET_ADMIN))
2217 if (optlen < sizeof(req))
2219 if (copy_from_user(&req, optval, sizeof(req)))
2221 err = netlink_set_ring(sk, &req, false,
2222 optname == NETLINK_TX_RING);
2225 #endif /* CONFIG_NETLINK_MMAP */
2226 case NETLINK_LISTEN_ALL_NSID:
2227 if (!ns_capable(sock_net(sk)->user_ns, CAP_NET_BROADCAST))
2231 nlk->flags |= NETLINK_F_LISTEN_ALL_NSID;
2233 nlk->flags &= ~NETLINK_F_LISTEN_ALL_NSID;
2242 static int netlink_getsockopt(struct socket *sock, int level, int optname,
2243 char __user *optval, int __user *optlen)
2245 struct sock *sk = sock->sk;
2246 struct netlink_sock *nlk = nlk_sk(sk);
2249 if (level != SOL_NETLINK)
2250 return -ENOPROTOOPT;
2252 if (get_user(len, optlen))
2258 case NETLINK_PKTINFO:
2259 if (len < sizeof(int))
2262 val = nlk->flags & NETLINK_F_RECV_PKTINFO ? 1 : 0;
2263 if (put_user(len, optlen) ||
2264 put_user(val, optval))
2268 case NETLINK_BROADCAST_ERROR:
2269 if (len < sizeof(int))
2272 val = nlk->flags & NETLINK_F_BROADCAST_SEND_ERROR ? 1 : 0;
2273 if (put_user(len, optlen) ||
2274 put_user(val, optval))
2278 case NETLINK_NO_ENOBUFS:
2279 if (len < sizeof(int))
2282 val = nlk->flags & NETLINK_F_RECV_NO_ENOBUFS ? 1 : 0;
2283 if (put_user(len, optlen) ||
2284 put_user(val, optval))
2294 static void netlink_cmsg_recv_pktinfo(struct msghdr *msg, struct sk_buff *skb)
2296 struct nl_pktinfo info;
2298 info.group = NETLINK_CB(skb).dst_group;
2299 put_cmsg(msg, SOL_NETLINK, NETLINK_PKTINFO, sizeof(info), &info);
2302 static void netlink_cmsg_listen_all_nsid(struct sock *sk, struct msghdr *msg,
2303 struct sk_buff *skb)
2305 if (!NETLINK_CB(skb).nsid_is_set)
2308 put_cmsg(msg, SOL_NETLINK, NETLINK_LISTEN_ALL_NSID, sizeof(int),
2309 &NETLINK_CB(skb).nsid);
2312 static int netlink_sendmsg(struct socket *sock, struct msghdr *msg, size_t len)
2314 struct sock *sk = sock->sk;
2315 struct netlink_sock *nlk = nlk_sk(sk);
2316 DECLARE_SOCKADDR(struct sockaddr_nl *, addr, msg->msg_name);
2319 struct sk_buff *skb;
2321 struct scm_cookie scm;
2322 u32 netlink_skb_flags = 0;
2324 if (msg->msg_flags&MSG_OOB)
2327 err = scm_send(sock, msg, &scm, true);
2331 if (msg->msg_namelen) {
2333 if (addr->nl_family != AF_NETLINK)
2335 dst_portid = addr->nl_pid;
2336 dst_group = ffs(addr->nl_groups);
2338 if ((dst_group || dst_portid) &&
2339 !netlink_allowed(sock, NL_CFG_F_NONROOT_SEND))
2341 netlink_skb_flags |= NETLINK_SKB_DST;
2343 dst_portid = nlk->dst_portid;
2344 dst_group = nlk->dst_group;
2348 err = netlink_autobind(sock);
2353 /* It's a really convoluted way for userland to ask for mmaped
2354 * sendmsg(), but that's what we've got...
2356 if (netlink_tx_is_mmaped(sk) &&
2357 msg->msg_iter.type == ITER_IOVEC &&
2358 msg->msg_iter.nr_segs == 1 &&
2359 msg->msg_iter.iov->iov_base == NULL) {
2360 err = netlink_mmap_sendmsg(sk, msg, dst_portid, dst_group,
2366 if (len > sk->sk_sndbuf - 32)
2369 skb = netlink_alloc_large_skb(len, dst_group);
2373 NETLINK_CB(skb).portid = nlk->portid;
2374 NETLINK_CB(skb).dst_group = dst_group;
2375 NETLINK_CB(skb).creds = scm.creds;
2376 NETLINK_CB(skb).flags = netlink_skb_flags;
2379 if (memcpy_from_msg(skb_put(skb, len), msg, len)) {
2384 err = security_netlink_send(sk, skb);
2391 atomic_inc(&skb->users);
2392 netlink_broadcast(sk, skb, dst_portid, dst_group, GFP_KERNEL);
2394 err = netlink_unicast(sk, skb, dst_portid, msg->msg_flags&MSG_DONTWAIT);
2401 static int netlink_recvmsg(struct socket *sock, struct msghdr *msg, size_t len,
2404 struct scm_cookie scm;
2405 struct sock *sk = sock->sk;
2406 struct netlink_sock *nlk = nlk_sk(sk);
2407 int noblock = flags&MSG_DONTWAIT;
2409 struct sk_buff *skb, *data_skb;
2417 skb = skb_recv_datagram(sk, flags, noblock, &err);
2423 #ifdef CONFIG_COMPAT_NETLINK_MESSAGES
2424 if (unlikely(skb_shinfo(skb)->frag_list)) {
2426 * If this skb has a frag_list, then here that means that we
2427 * will have to use the frag_list skb's data for compat tasks
2428 * and the regular skb's data for normal (non-compat) tasks.
2430 * If we need to send the compat skb, assign it to the
2431 * 'data_skb' variable so that it will be used below for data
2432 * copying. We keep 'skb' for everything else, including
2433 * freeing both later.
2435 if (flags & MSG_CMSG_COMPAT)
2436 data_skb = skb_shinfo(skb)->frag_list;
2440 /* Record the max length of recvmsg() calls for future allocations */
2441 nlk->max_recvmsg_len = max(nlk->max_recvmsg_len, len);
2442 nlk->max_recvmsg_len = min_t(size_t, nlk->max_recvmsg_len,
2445 copied = data_skb->len;
2447 msg->msg_flags |= MSG_TRUNC;
2451 skb_reset_transport_header(data_skb);
2452 err = skb_copy_datagram_msg(data_skb, 0, msg, copied);
2454 if (msg->msg_name) {
2455 DECLARE_SOCKADDR(struct sockaddr_nl *, addr, msg->msg_name);
2456 addr->nl_family = AF_NETLINK;
2458 addr->nl_pid = NETLINK_CB(skb).portid;
2459 addr->nl_groups = netlink_group_mask(NETLINK_CB(skb).dst_group);
2460 msg->msg_namelen = sizeof(*addr);
2463 if (nlk->flags & NETLINK_F_RECV_PKTINFO)
2464 netlink_cmsg_recv_pktinfo(msg, skb);
2465 if (nlk->flags & NETLINK_F_LISTEN_ALL_NSID)
2466 netlink_cmsg_listen_all_nsid(sk, msg, skb);
2468 memset(&scm, 0, sizeof(scm));
2469 scm.creds = *NETLINK_CREDS(skb);
2470 if (flags & MSG_TRUNC)
2471 copied = data_skb->len;
2473 skb_free_datagram(sk, skb);
2475 if (nlk->cb_running &&
2476 atomic_read(&sk->sk_rmem_alloc) <= sk->sk_rcvbuf / 2) {
2477 ret = netlink_dump(sk);
2480 sk->sk_error_report(sk);
2484 scm_recv(sock, msg, &scm, flags);
2486 netlink_rcv_wake(sk);
2487 return err ? : copied;
2490 static void netlink_data_ready(struct sock *sk)
2496 * We export these functions to other modules. They provide a
2497 * complete set of kernel non-blocking support for message
2502 __netlink_kernel_create(struct net *net, int unit, struct module *module,
2503 struct netlink_kernel_cfg *cfg)
2505 struct socket *sock;
2507 struct netlink_sock *nlk;
2508 struct listeners *listeners = NULL;
2509 struct mutex *cb_mutex = cfg ? cfg->cb_mutex : NULL;
2510 unsigned int groups;
2514 if (unit < 0 || unit >= MAX_LINKS)
2517 if (sock_create_lite(PF_NETLINK, SOCK_DGRAM, unit, &sock))
2520 if (__netlink_create(net, sock, cb_mutex, unit, 1) < 0)
2521 goto out_sock_release_nosk;
2525 if (!cfg || cfg->groups < 32)
2528 groups = cfg->groups;
2530 listeners = kzalloc(sizeof(*listeners) + NLGRPSZ(groups), GFP_KERNEL);
2532 goto out_sock_release;
2534 sk->sk_data_ready = netlink_data_ready;
2535 if (cfg && cfg->input)
2536 nlk_sk(sk)->netlink_rcv = cfg->input;
2538 if (netlink_insert(sk, 0))
2539 goto out_sock_release;
2542 nlk->flags |= NETLINK_F_KERNEL_SOCKET;
2544 netlink_table_grab();
2545 if (!nl_table[unit].registered) {
2546 nl_table[unit].groups = groups;
2547 rcu_assign_pointer(nl_table[unit].listeners, listeners);
2548 nl_table[unit].cb_mutex = cb_mutex;
2549 nl_table[unit].module = module;
2551 nl_table[unit].bind = cfg->bind;
2552 nl_table[unit].unbind = cfg->unbind;
2553 nl_table[unit].flags = cfg->flags;
2555 nl_table[unit].compare = cfg->compare;
2557 nl_table[unit].registered = 1;
2560 nl_table[unit].registered++;
2562 netlink_table_ungrab();
2567 netlink_kernel_release(sk);
2570 out_sock_release_nosk:
2574 EXPORT_SYMBOL(__netlink_kernel_create);
2577 netlink_kernel_release(struct sock *sk)
2579 if (sk == NULL || sk->sk_socket == NULL)
2582 sock_release(sk->sk_socket);
2584 EXPORT_SYMBOL(netlink_kernel_release);
2586 int __netlink_change_ngroups(struct sock *sk, unsigned int groups)
2588 struct listeners *new, *old;
2589 struct netlink_table *tbl = &nl_table[sk->sk_protocol];
2594 if (NLGRPSZ(tbl->groups) < NLGRPSZ(groups)) {
2595 new = kzalloc(sizeof(*new) + NLGRPSZ(groups), GFP_ATOMIC);
2598 old = nl_deref_protected(tbl->listeners);
2599 memcpy(new->masks, old->masks, NLGRPSZ(tbl->groups));
2600 rcu_assign_pointer(tbl->listeners, new);
2602 kfree_rcu(old, rcu);
2604 tbl->groups = groups;
2610 * netlink_change_ngroups - change number of multicast groups
2612 * This changes the number of multicast groups that are available
2613 * on a certain netlink family. Note that it is not possible to
2614 * change the number of groups to below 32. Also note that it does
2615 * not implicitly call netlink_clear_multicast_users() when the
2616 * number of groups is reduced.
2618 * @sk: The kernel netlink socket, as returned by netlink_kernel_create().
2619 * @groups: The new number of groups.
2621 int netlink_change_ngroups(struct sock *sk, unsigned int groups)
2625 netlink_table_grab();
2626 err = __netlink_change_ngroups(sk, groups);
2627 netlink_table_ungrab();
2632 void __netlink_clear_multicast_users(struct sock *ksk, unsigned int group)
2635 struct netlink_table *tbl = &nl_table[ksk->sk_protocol];
2637 sk_for_each_bound(sk, &tbl->mc_list)
2638 netlink_update_socket_mc(nlk_sk(sk), group, 0);
2642 __nlmsg_put(struct sk_buff *skb, u32 portid, u32 seq, int type, int len, int flags)
2644 struct nlmsghdr *nlh;
2645 int size = nlmsg_msg_size(len);
2647 nlh = (struct nlmsghdr *)skb_put(skb, NLMSG_ALIGN(size));
2648 nlh->nlmsg_type = type;
2649 nlh->nlmsg_len = size;
2650 nlh->nlmsg_flags = flags;
2651 nlh->nlmsg_pid = portid;
2652 nlh->nlmsg_seq = seq;
2653 if (!__builtin_constant_p(size) || NLMSG_ALIGN(size) - size != 0)
2654 memset(nlmsg_data(nlh) + len, 0, NLMSG_ALIGN(size) - size);
2657 EXPORT_SYMBOL(__nlmsg_put);
2660 * It looks a bit ugly.
2661 * It would be better to create kernel thread.
2664 static int netlink_dump(struct sock *sk)
2666 struct netlink_sock *nlk = nlk_sk(sk);
2667 struct netlink_callback *cb;
2668 struct sk_buff *skb = NULL;
2669 struct nlmsghdr *nlh;
2670 int len, err = -ENOBUFS;
2673 mutex_lock(nlk->cb_mutex);
2674 if (!nlk->cb_running) {
2680 alloc_size = max_t(int, cb->min_dump_alloc, NLMSG_GOODSIZE);
2682 if (!netlink_rx_is_mmaped(sk) &&
2683 atomic_read(&sk->sk_rmem_alloc) >= sk->sk_rcvbuf)
2686 /* NLMSG_GOODSIZE is small to avoid high order allocations being
2687 * required, but it makes sense to _attempt_ a 16K bytes allocation
2688 * to reduce number of system calls on dump operations, if user
2689 * ever provided a big enough buffer.
2691 if (alloc_size < nlk->max_recvmsg_len) {
2692 skb = netlink_alloc_skb(sk,
2693 nlk->max_recvmsg_len,
2698 /* available room should be exact amount to avoid MSG_TRUNC */
2700 skb_reserve(skb, skb_tailroom(skb) -
2701 nlk->max_recvmsg_len);
2704 skb = netlink_alloc_skb(sk, alloc_size, nlk->portid,
2708 netlink_skb_set_owner_r(skb, sk);
2710 len = cb->dump(skb, cb);
2713 mutex_unlock(nlk->cb_mutex);
2715 if (sk_filter(sk, skb))
2718 __netlink_sendskb(sk, skb);
2722 nlh = nlmsg_put_answer(skb, cb, NLMSG_DONE, sizeof(len), NLM_F_MULTI);
2726 nl_dump_check_consistent(cb, nlh);
2728 memcpy(nlmsg_data(nlh), &len, sizeof(len));
2730 if (sk_filter(sk, skb))
2733 __netlink_sendskb(sk, skb);
2738 nlk->cb_running = false;
2739 mutex_unlock(nlk->cb_mutex);
2740 module_put(cb->module);
2741 consume_skb(cb->skb);
2745 mutex_unlock(nlk->cb_mutex);
2750 int __netlink_dump_start(struct sock *ssk, struct sk_buff *skb,
2751 const struct nlmsghdr *nlh,
2752 struct netlink_dump_control *control)
2754 struct netlink_callback *cb;
2756 struct netlink_sock *nlk;
2759 /* Memory mapped dump requests need to be copied to avoid looping
2760 * on the pending state in netlink_mmap_sendmsg() while the CB hold
2761 * a reference to the skb.
2763 if (netlink_skb_is_mmaped(skb)) {
2764 skb = skb_copy(skb, GFP_KERNEL);
2768 atomic_inc(&skb->users);
2770 sk = netlink_lookup(sock_net(ssk), ssk->sk_protocol, NETLINK_CB(skb).portid);
2772 ret = -ECONNREFUSED;
2777 mutex_lock(nlk->cb_mutex);
2778 /* A dump is in progress... */
2779 if (nlk->cb_running) {
2783 /* add reference of module which cb->dump belongs to */
2784 if (!try_module_get(control->module)) {
2785 ret = -EPROTONOSUPPORT;
2790 memset(cb, 0, sizeof(*cb));
2791 cb->dump = control->dump;
2792 cb->done = control->done;
2794 cb->data = control->data;
2795 cb->module = control->module;
2796 cb->min_dump_alloc = control->min_dump_alloc;
2799 nlk->cb_running = true;
2801 mutex_unlock(nlk->cb_mutex);
2803 ret = netlink_dump(sk);
2809 /* We successfully started a dump, by returning -EINTR we
2810 * signal not to send ACK even if it was requested.
2816 mutex_unlock(nlk->cb_mutex);
2821 EXPORT_SYMBOL(__netlink_dump_start);
2823 void netlink_ack(struct sk_buff *in_skb, struct nlmsghdr *nlh, int err)
2825 struct sk_buff *skb;
2826 struct nlmsghdr *rep;
2827 struct nlmsgerr *errmsg;
2828 size_t payload = sizeof(*errmsg);
2830 /* error messages get the original request appened */
2832 payload += nlmsg_len(nlh);
2834 skb = netlink_alloc_skb(in_skb->sk, nlmsg_total_size(payload),
2835 NETLINK_CB(in_skb).portid, GFP_KERNEL);
2839 sk = netlink_lookup(sock_net(in_skb->sk),
2840 in_skb->sk->sk_protocol,
2841 NETLINK_CB(in_skb).portid);
2843 sk->sk_err = ENOBUFS;
2844 sk->sk_error_report(sk);
2850 rep = __nlmsg_put(skb, NETLINK_CB(in_skb).portid, nlh->nlmsg_seq,
2851 NLMSG_ERROR, payload, 0);
2852 errmsg = nlmsg_data(rep);
2853 errmsg->error = err;
2854 memcpy(&errmsg->msg, nlh, err ? nlh->nlmsg_len : sizeof(*nlh));
2855 netlink_unicast(in_skb->sk, skb, NETLINK_CB(in_skb).portid, MSG_DONTWAIT);
2857 EXPORT_SYMBOL(netlink_ack);
2859 int netlink_rcv_skb(struct sk_buff *skb, int (*cb)(struct sk_buff *,
2862 struct nlmsghdr *nlh;
2865 while (skb->len >= nlmsg_total_size(0)) {
2868 nlh = nlmsg_hdr(skb);
2871 if (nlh->nlmsg_len < NLMSG_HDRLEN || skb->len < nlh->nlmsg_len)
2874 /* Only requests are handled by the kernel */
2875 if (!(nlh->nlmsg_flags & NLM_F_REQUEST))
2878 /* Skip control messages */
2879 if (nlh->nlmsg_type < NLMSG_MIN_TYPE)
2887 if (nlh->nlmsg_flags & NLM_F_ACK || err)
2888 netlink_ack(skb, nlh, err);
2891 msglen = NLMSG_ALIGN(nlh->nlmsg_len);
2892 if (msglen > skb->len)
2894 skb_pull(skb, msglen);
2899 EXPORT_SYMBOL(netlink_rcv_skb);
2902 * nlmsg_notify - send a notification netlink message
2903 * @sk: netlink socket to use
2904 * @skb: notification message
2905 * @portid: destination netlink portid for reports or 0
2906 * @group: destination multicast group or 0
2907 * @report: 1 to report back, 0 to disable
2908 * @flags: allocation flags
2910 int nlmsg_notify(struct sock *sk, struct sk_buff *skb, u32 portid,
2911 unsigned int group, int report, gfp_t flags)
2916 int exclude_portid = 0;
2919 atomic_inc(&skb->users);
2920 exclude_portid = portid;
2923 /* errors reported via destination sk->sk_err, but propagate
2924 * delivery errors if NETLINK_BROADCAST_ERROR flag is set */
2925 err = nlmsg_multicast(sk, skb, exclude_portid, group, flags);
2931 err2 = nlmsg_unicast(sk, skb, portid);
2932 if (!err || err == -ESRCH)
2938 EXPORT_SYMBOL(nlmsg_notify);
2940 #ifdef CONFIG_PROC_FS
2941 struct nl_seq_iter {
2942 struct seq_net_private p;
2943 struct rhashtable_iter hti;
2947 static int netlink_walk_start(struct nl_seq_iter *iter)
2951 err = rhashtable_walk_init(&nl_table[iter->link].hash, &iter->hti);
2953 iter->link = MAX_LINKS;
2957 err = rhashtable_walk_start(&iter->hti);
2958 return err == -EAGAIN ? 0 : err;
2961 static void netlink_walk_stop(struct nl_seq_iter *iter)
2963 rhashtable_walk_stop(&iter->hti);
2964 rhashtable_walk_exit(&iter->hti);
2967 static void *__netlink_seq_next(struct seq_file *seq)
2969 struct nl_seq_iter *iter = seq->private;
2970 struct netlink_sock *nlk;
2976 nlk = rhashtable_walk_next(&iter->hti);
2979 if (PTR_ERR(nlk) == -EAGAIN)
2988 netlink_walk_stop(iter);
2989 if (++iter->link >= MAX_LINKS)
2992 err = netlink_walk_start(iter);
2994 return ERR_PTR(err);
2996 } while (sock_net(&nlk->sk) != seq_file_net(seq));
3001 static void *netlink_seq_start(struct seq_file *seq, loff_t *posp)
3003 struct nl_seq_iter *iter = seq->private;
3004 void *obj = SEQ_START_TOKEN;
3010 err = netlink_walk_start(iter);
3012 return ERR_PTR(err);
3014 for (pos = *posp; pos && obj && !IS_ERR(obj); pos--)
3015 obj = __netlink_seq_next(seq);
3020 static void *netlink_seq_next(struct seq_file *seq, void *v, loff_t *pos)
3023 return __netlink_seq_next(seq);
3026 static void netlink_seq_stop(struct seq_file *seq, void *v)
3028 struct nl_seq_iter *iter = seq->private;
3030 if (iter->link >= MAX_LINKS)
3033 netlink_walk_stop(iter);
3037 static int netlink_seq_show(struct seq_file *seq, void *v)
3039 if (v == SEQ_START_TOKEN) {
3041 "sk Eth Pid Groups "
3042 "Rmem Wmem Dump Locks Drops Inode\n");
3045 struct netlink_sock *nlk = nlk_sk(s);
3047 seq_printf(seq, "%pK %-3d %-6u %08x %-8d %-8d %d %-8d %-8d %-8lu\n",
3051 nlk->groups ? (u32)nlk->groups[0] : 0,
3052 sk_rmem_alloc_get(s),
3053 sk_wmem_alloc_get(s),
3055 atomic_read(&s->sk_refcnt),
3056 atomic_read(&s->sk_drops),
3064 static const struct seq_operations netlink_seq_ops = {
3065 .start = netlink_seq_start,
3066 .next = netlink_seq_next,
3067 .stop = netlink_seq_stop,
3068 .show = netlink_seq_show,
3072 static int netlink_seq_open(struct inode *inode, struct file *file)
3074 return seq_open_net(inode, file, &netlink_seq_ops,
3075 sizeof(struct nl_seq_iter));
3078 static const struct file_operations netlink_seq_fops = {
3079 .owner = THIS_MODULE,
3080 .open = netlink_seq_open,
3082 .llseek = seq_lseek,
3083 .release = seq_release_net,
3088 int netlink_register_notifier(struct notifier_block *nb)
3090 return atomic_notifier_chain_register(&netlink_chain, nb);
3092 EXPORT_SYMBOL(netlink_register_notifier);
3094 int netlink_unregister_notifier(struct notifier_block *nb)
3096 return atomic_notifier_chain_unregister(&netlink_chain, nb);
3098 EXPORT_SYMBOL(netlink_unregister_notifier);
3100 static const struct proto_ops netlink_ops = {
3101 .family = PF_NETLINK,
3102 .owner = THIS_MODULE,
3103 .release = netlink_release,
3104 .bind = netlink_bind,
3105 .connect = netlink_connect,
3106 .socketpair = sock_no_socketpair,
3107 .accept = sock_no_accept,
3108 .getname = netlink_getname,
3109 .poll = netlink_poll,
3110 .ioctl = sock_no_ioctl,
3111 .listen = sock_no_listen,
3112 .shutdown = sock_no_shutdown,
3113 .setsockopt = netlink_setsockopt,
3114 .getsockopt = netlink_getsockopt,
3115 .sendmsg = netlink_sendmsg,
3116 .recvmsg = netlink_recvmsg,
3117 .mmap = netlink_mmap,
3118 .sendpage = sock_no_sendpage,
3121 static const struct net_proto_family netlink_family_ops = {
3122 .family = PF_NETLINK,
3123 .create = netlink_create,
3124 .owner = THIS_MODULE, /* for consistency 8) */
3127 static int __net_init netlink_net_init(struct net *net)
3129 #ifdef CONFIG_PROC_FS
3130 if (!proc_create("netlink", 0, net->proc_net, &netlink_seq_fops))
3136 static void __net_exit netlink_net_exit(struct net *net)
3138 #ifdef CONFIG_PROC_FS
3139 remove_proc_entry("netlink", net->proc_net);
3143 static void __init netlink_add_usersock_entry(void)
3145 struct listeners *listeners;
3148 listeners = kzalloc(sizeof(*listeners) + NLGRPSZ(groups), GFP_KERNEL);
3150 panic("netlink_add_usersock_entry: Cannot allocate listeners\n");
3152 netlink_table_grab();
3154 nl_table[NETLINK_USERSOCK].groups = groups;
3155 rcu_assign_pointer(nl_table[NETLINK_USERSOCK].listeners, listeners);
3156 nl_table[NETLINK_USERSOCK].module = THIS_MODULE;
3157 nl_table[NETLINK_USERSOCK].registered = 1;
3158 nl_table[NETLINK_USERSOCK].flags = NL_CFG_F_NONROOT_SEND;
3160 netlink_table_ungrab();
3163 static struct pernet_operations __net_initdata netlink_net_ops = {
3164 .init = netlink_net_init,
3165 .exit = netlink_net_exit,
3168 static inline u32 netlink_hash(const void *data, u32 len, u32 seed)
3170 const struct netlink_sock *nlk = data;
3171 struct netlink_compare_arg arg;
3173 netlink_compare_arg_init(&arg, sock_net(&nlk->sk), nlk->portid);
3174 return jhash2((u32 *)&arg, netlink_compare_arg_len / sizeof(u32), seed);
3177 static const struct rhashtable_params netlink_rhashtable_params = {
3178 .head_offset = offsetof(struct netlink_sock, node),
3179 .key_len = netlink_compare_arg_len,
3180 .obj_hashfn = netlink_hash,
3181 .obj_cmpfn = netlink_compare,
3182 .automatic_shrinking = true,
3185 static int __init netlink_proto_init(void)
3188 int err = proto_register(&netlink_proto, 0);
3193 BUILD_BUG_ON(sizeof(struct netlink_skb_parms) > FIELD_SIZEOF(struct sk_buff, cb));
3195 nl_table = kcalloc(MAX_LINKS, sizeof(*nl_table), GFP_KERNEL);
3199 for (i = 0; i < MAX_LINKS; i++) {
3200 if (rhashtable_init(&nl_table[i].hash,
3201 &netlink_rhashtable_params) < 0) {
3203 rhashtable_destroy(&nl_table[i].hash);
3209 INIT_LIST_HEAD(&netlink_tap_all);
3211 netlink_add_usersock_entry();
3213 sock_register(&netlink_family_ops);
3214 register_pernet_subsys(&netlink_net_ops);
3215 /* The netlink device handler may be needed early. */
3220 panic("netlink_init: Cannot allocate nl_table\n");
3223 core_initcall(netlink_proto_init);