2 * NETLINK Kernel-user communication protocol.
4 * Authors: Alan Cox <alan@lxorguk.ukuu.org.uk>
5 * Alexey Kuznetsov <kuznet@ms2.inr.ac.ru>
6 * Patrick McHardy <kaber@trash.net>
8 * This program is free software; you can redistribute it and/or
9 * modify it under the terms of the GNU General Public License
10 * as published by the Free Software Foundation; either version
11 * 2 of the License, or (at your option) any later version.
13 * Tue Jun 26 14:36:48 MEST 2001 Herbert "herp" Rosmanith
14 * added netlink_proto_exit
15 * Tue Jan 22 18:32:44 BRST 2002 Arnaldo C. de Melo <acme@conectiva.com.br>
16 * use nlk_sk, as sk->protinfo is on a diet 8)
17 * Fri Jul 22 19:51:12 MEST 2005 Harald Welte <laforge@gnumonks.org>
18 * - inc module use count of module that owns
19 * the kernel socket in case userspace opens
20 * socket of same protocol
21 * - remove all module support, since netlink is
22 * mandatory if CONFIG_NET=y these days
25 #include <linux/module.h>
27 #include <linux/capability.h>
28 #include <linux/kernel.h>
29 #include <linux/init.h>
30 #include <linux/signal.h>
31 #include <linux/sched.h>
32 #include <linux/errno.h>
33 #include <linux/string.h>
34 #include <linux/stat.h>
35 #include <linux/socket.h>
37 #include <linux/fcntl.h>
38 #include <linux/termios.h>
39 #include <linux/sockios.h>
40 #include <linux/net.h>
42 #include <linux/slab.h>
43 #include <asm/uaccess.h>
44 #include <linux/skbuff.h>
45 #include <linux/netdevice.h>
46 #include <linux/rtnetlink.h>
47 #include <linux/proc_fs.h>
48 #include <linux/seq_file.h>
49 #include <linux/notifier.h>
50 #include <linux/security.h>
51 #include <linux/jhash.h>
52 #include <linux/jiffies.h>
53 #include <linux/random.h>
54 #include <linux/bitops.h>
56 #include <linux/types.h>
57 #include <linux/audit.h>
58 #include <linux/mutex.h>
59 #include <linux/vmalloc.h>
60 #include <linux/if_arp.h>
61 #include <linux/rhashtable.h>
62 #include <asm/cacheflush.h>
63 #include <linux/hash.h>
64 #include <linux/genetlink.h>
66 #include <net/net_namespace.h>
69 #include <net/netlink.h>
71 #include "af_netlink.h"
75 unsigned long masks[0];
79 #define NETLINK_S_CONGESTED 0x0
82 #define NETLINK_F_KERNEL_SOCKET 0x1
83 #define NETLINK_F_RECV_PKTINFO 0x2
84 #define NETLINK_F_BROADCAST_SEND_ERROR 0x4
85 #define NETLINK_F_RECV_NO_ENOBUFS 0x8
86 #define NETLINK_F_LISTEN_ALL_NSID 0x10
87 #define NETLINK_F_CAP_ACK 0x20
89 static inline int netlink_is_kernel(struct sock *sk)
91 return nlk_sk(sk)->flags & NETLINK_F_KERNEL_SOCKET;
94 struct netlink_table *nl_table __read_mostly;
95 EXPORT_SYMBOL_GPL(nl_table);
97 static DECLARE_WAIT_QUEUE_HEAD(nl_table_wait);
99 static int netlink_dump(struct sock *sk);
100 static void netlink_skb_destructor(struct sk_buff *skb);
102 /* nl_table locking explained:
103 * Lookup and traversal are protected with an RCU read-side lock. Insertion
104 * and removal are protected with per bucket lock while using RCU list
105 * modification primitives and may run in parallel to RCU protected lookups.
106 * Destruction of the Netlink socket may only occur *after* nl_table_lock has
107 * been acquired * either during or after the socket has been removed from
108 * the list and after an RCU grace period.
110 DEFINE_RWLOCK(nl_table_lock);
111 EXPORT_SYMBOL_GPL(nl_table_lock);
112 static atomic_t nl_table_users = ATOMIC_INIT(0);
114 #define nl_deref_protected(X) rcu_dereference_protected(X, lockdep_is_held(&nl_table_lock));
116 static ATOMIC_NOTIFIER_HEAD(netlink_chain);
118 static DEFINE_SPINLOCK(netlink_tap_lock);
119 static struct list_head netlink_tap_all __read_mostly;
121 static const struct rhashtable_params netlink_rhashtable_params;
123 static inline u32 netlink_group_mask(u32 group)
125 return group ? 1 << (group - 1) : 0;
128 int netlink_add_tap(struct netlink_tap *nt)
130 if (unlikely(nt->dev->type != ARPHRD_NETLINK))
133 spin_lock(&netlink_tap_lock);
134 list_add_rcu(&nt->list, &netlink_tap_all);
135 spin_unlock(&netlink_tap_lock);
137 __module_get(nt->module);
141 EXPORT_SYMBOL_GPL(netlink_add_tap);
143 static int __netlink_remove_tap(struct netlink_tap *nt)
146 struct netlink_tap *tmp;
148 spin_lock(&netlink_tap_lock);
150 list_for_each_entry(tmp, &netlink_tap_all, list) {
152 list_del_rcu(&nt->list);
158 pr_warn("__netlink_remove_tap: %p not found\n", nt);
160 spin_unlock(&netlink_tap_lock);
163 module_put(nt->module);
165 return found ? 0 : -ENODEV;
168 int netlink_remove_tap(struct netlink_tap *nt)
172 ret = __netlink_remove_tap(nt);
177 EXPORT_SYMBOL_GPL(netlink_remove_tap);
179 static bool netlink_filter_tap(const struct sk_buff *skb)
181 struct sock *sk = skb->sk;
183 /* We take the more conservative approach and
184 * whitelist socket protocols that may pass.
186 switch (sk->sk_protocol) {
188 case NETLINK_USERSOCK:
189 case NETLINK_SOCK_DIAG:
192 case NETLINK_FIB_LOOKUP:
193 case NETLINK_NETFILTER:
194 case NETLINK_GENERIC:
201 static int __netlink_deliver_tap_skb(struct sk_buff *skb,
202 struct net_device *dev)
204 struct sk_buff *nskb;
205 struct sock *sk = skb->sk;
209 nskb = skb_clone(skb, GFP_ATOMIC);
212 nskb->protocol = htons((u16) sk->sk_protocol);
213 nskb->pkt_type = netlink_is_kernel(sk) ?
214 PACKET_KERNEL : PACKET_USER;
215 skb_reset_network_header(nskb);
216 ret = dev_queue_xmit(nskb);
217 if (unlikely(ret > 0))
218 ret = net_xmit_errno(ret);
225 static void __netlink_deliver_tap(struct sk_buff *skb)
228 struct netlink_tap *tmp;
230 if (!netlink_filter_tap(skb))
233 list_for_each_entry_rcu(tmp, &netlink_tap_all, list) {
234 ret = __netlink_deliver_tap_skb(skb, tmp->dev);
240 static void netlink_deliver_tap(struct sk_buff *skb)
244 if (unlikely(!list_empty(&netlink_tap_all)))
245 __netlink_deliver_tap(skb);
250 static void netlink_deliver_tap_kernel(struct sock *dst, struct sock *src,
253 if (!(netlink_is_kernel(dst) && netlink_is_kernel(src)))
254 netlink_deliver_tap(skb);
257 static void netlink_overrun(struct sock *sk)
259 struct netlink_sock *nlk = nlk_sk(sk);
261 if (!(nlk->flags & NETLINK_F_RECV_NO_ENOBUFS)) {
262 if (!test_and_set_bit(NETLINK_S_CONGESTED,
263 &nlk_sk(sk)->state)) {
264 sk->sk_err = ENOBUFS;
265 sk->sk_error_report(sk);
268 atomic_inc(&sk->sk_drops);
271 static void netlink_rcv_wake(struct sock *sk)
273 struct netlink_sock *nlk = nlk_sk(sk);
275 if (skb_queue_empty(&sk->sk_receive_queue))
276 clear_bit(NETLINK_S_CONGESTED, &nlk->state);
277 if (!test_bit(NETLINK_S_CONGESTED, &nlk->state))
278 wake_up_interruptible(&nlk->wait);
281 #ifdef CONFIG_NETLINK_MMAP
282 static bool netlink_skb_is_mmaped(const struct sk_buff *skb)
284 return NETLINK_CB(skb).flags & NETLINK_SKB_MMAPED;
287 static bool netlink_rx_is_mmaped(struct sock *sk)
289 return nlk_sk(sk)->rx_ring.pg_vec != NULL;
292 static bool netlink_tx_is_mmaped(struct sock *sk)
294 return nlk_sk(sk)->tx_ring.pg_vec != NULL;
297 static __pure struct page *pgvec_to_page(const void *addr)
299 if (is_vmalloc_addr(addr))
300 return vmalloc_to_page(addr);
302 return virt_to_page(addr);
305 static void free_pg_vec(void **pg_vec, unsigned int order, unsigned int len)
309 for (i = 0; i < len; i++) {
310 if (pg_vec[i] != NULL) {
311 if (is_vmalloc_addr(pg_vec[i]))
314 free_pages((unsigned long)pg_vec[i], order);
320 static void *alloc_one_pg_vec_page(unsigned long order)
323 gfp_t gfp_flags = GFP_KERNEL | __GFP_COMP | __GFP_ZERO |
324 __GFP_NOWARN | __GFP_NORETRY;
326 buffer = (void *)__get_free_pages(gfp_flags, order);
330 buffer = vzalloc((1 << order) * PAGE_SIZE);
334 gfp_flags &= ~__GFP_NORETRY;
335 return (void *)__get_free_pages(gfp_flags, order);
338 static void **alloc_pg_vec(struct netlink_sock *nlk,
339 struct nl_mmap_req *req, unsigned int order)
341 unsigned int block_nr = req->nm_block_nr;
345 pg_vec = kcalloc(block_nr, sizeof(void *), GFP_KERNEL);
349 for (i = 0; i < block_nr; i++) {
350 pg_vec[i] = alloc_one_pg_vec_page(order);
351 if (pg_vec[i] == NULL)
357 free_pg_vec(pg_vec, order, block_nr);
363 __netlink_set_ring(struct sock *sk, struct nl_mmap_req *req, bool tx_ring, void **pg_vec,
366 struct netlink_sock *nlk = nlk_sk(sk);
367 struct sk_buff_head *queue;
368 struct netlink_ring *ring;
370 queue = tx_ring ? &sk->sk_write_queue : &sk->sk_receive_queue;
371 ring = tx_ring ? &nlk->tx_ring : &nlk->rx_ring;
373 spin_lock_bh(&queue->lock);
375 ring->frame_max = req->nm_frame_nr - 1;
377 ring->frame_size = req->nm_frame_size;
378 ring->pg_vec_pages = req->nm_block_size / PAGE_SIZE;
380 swap(ring->pg_vec_len, req->nm_block_nr);
381 swap(ring->pg_vec_order, order);
382 swap(ring->pg_vec, pg_vec);
384 __skb_queue_purge(queue);
385 spin_unlock_bh(&queue->lock);
387 WARN_ON(atomic_read(&nlk->mapped));
390 free_pg_vec(pg_vec, order, req->nm_block_nr);
393 static int netlink_set_ring(struct sock *sk, struct nl_mmap_req *req,
396 struct netlink_sock *nlk = nlk_sk(sk);
397 struct netlink_ring *ring;
398 void **pg_vec = NULL;
399 unsigned int order = 0;
401 ring = tx_ring ? &nlk->tx_ring : &nlk->rx_ring;
403 if (atomic_read(&nlk->mapped))
405 if (atomic_read(&ring->pending))
408 if (req->nm_block_nr) {
409 if (ring->pg_vec != NULL)
412 if ((int)req->nm_block_size <= 0)
414 if (!PAGE_ALIGNED(req->nm_block_size))
416 if (req->nm_frame_size < NL_MMAP_HDRLEN)
418 if (!IS_ALIGNED(req->nm_frame_size, NL_MMAP_MSG_ALIGNMENT))
421 ring->frames_per_block = req->nm_block_size /
423 if (ring->frames_per_block == 0)
425 if (ring->frames_per_block * req->nm_block_nr !=
429 order = get_order(req->nm_block_size);
430 pg_vec = alloc_pg_vec(nlk, req, order);
434 if (req->nm_frame_nr)
438 mutex_lock(&nlk->pg_vec_lock);
439 if (atomic_read(&nlk->mapped) == 0) {
440 __netlink_set_ring(sk, req, tx_ring, pg_vec, order);
441 mutex_unlock(&nlk->pg_vec_lock);
445 mutex_unlock(&nlk->pg_vec_lock);
448 free_pg_vec(pg_vec, order, req->nm_block_nr);
453 static void netlink_mm_open(struct vm_area_struct *vma)
455 struct file *file = vma->vm_file;
456 struct socket *sock = file->private_data;
457 struct sock *sk = sock->sk;
460 atomic_inc(&nlk_sk(sk)->mapped);
463 static void netlink_mm_close(struct vm_area_struct *vma)
465 struct file *file = vma->vm_file;
466 struct socket *sock = file->private_data;
467 struct sock *sk = sock->sk;
470 atomic_dec(&nlk_sk(sk)->mapped);
473 static const struct vm_operations_struct netlink_mmap_ops = {
474 .open = netlink_mm_open,
475 .close = netlink_mm_close,
478 static int netlink_mmap(struct file *file, struct socket *sock,
479 struct vm_area_struct *vma)
481 struct sock *sk = sock->sk;
482 struct netlink_sock *nlk = nlk_sk(sk);
483 struct netlink_ring *ring;
484 unsigned long start, size, expected;
491 mutex_lock(&nlk->pg_vec_lock);
494 for (ring = &nlk->rx_ring; ring <= &nlk->tx_ring; ring++) {
495 if (ring->pg_vec == NULL)
497 expected += ring->pg_vec_len * ring->pg_vec_pages * PAGE_SIZE;
503 size = vma->vm_end - vma->vm_start;
504 if (size != expected)
507 start = vma->vm_start;
508 for (ring = &nlk->rx_ring; ring <= &nlk->tx_ring; ring++) {
509 if (ring->pg_vec == NULL)
512 for (i = 0; i < ring->pg_vec_len; i++) {
514 void *kaddr = ring->pg_vec[i];
517 for (pg_num = 0; pg_num < ring->pg_vec_pages; pg_num++) {
518 page = pgvec_to_page(kaddr);
519 err = vm_insert_page(vma, start, page);
528 atomic_inc(&nlk->mapped);
529 vma->vm_ops = &netlink_mmap_ops;
532 mutex_unlock(&nlk->pg_vec_lock);
536 static void netlink_frame_flush_dcache(const struct nl_mmap_hdr *hdr, unsigned int nm_len)
538 #if ARCH_IMPLEMENTS_FLUSH_DCACHE_PAGE == 1
539 struct page *p_start, *p_end;
541 /* First page is flushed through netlink_{get,set}_status */
542 p_start = pgvec_to_page(hdr + PAGE_SIZE);
543 p_end = pgvec_to_page((void *)hdr + NL_MMAP_HDRLEN + nm_len - 1);
544 while (p_start <= p_end) {
545 flush_dcache_page(p_start);
551 static enum nl_mmap_status netlink_get_status(const struct nl_mmap_hdr *hdr)
554 flush_dcache_page(pgvec_to_page(hdr));
555 return hdr->nm_status;
558 static void netlink_set_status(struct nl_mmap_hdr *hdr,
559 enum nl_mmap_status status)
562 hdr->nm_status = status;
563 flush_dcache_page(pgvec_to_page(hdr));
566 static struct nl_mmap_hdr *
567 __netlink_lookup_frame(const struct netlink_ring *ring, unsigned int pos)
569 unsigned int pg_vec_pos, frame_off;
571 pg_vec_pos = pos / ring->frames_per_block;
572 frame_off = pos % ring->frames_per_block;
574 return ring->pg_vec[pg_vec_pos] + (frame_off * ring->frame_size);
577 static struct nl_mmap_hdr *
578 netlink_lookup_frame(const struct netlink_ring *ring, unsigned int pos,
579 enum nl_mmap_status status)
581 struct nl_mmap_hdr *hdr;
583 hdr = __netlink_lookup_frame(ring, pos);
584 if (netlink_get_status(hdr) != status)
590 static struct nl_mmap_hdr *
591 netlink_current_frame(const struct netlink_ring *ring,
592 enum nl_mmap_status status)
594 return netlink_lookup_frame(ring, ring->head, status);
597 static void netlink_increment_head(struct netlink_ring *ring)
599 ring->head = ring->head != ring->frame_max ? ring->head + 1 : 0;
602 static void netlink_forward_ring(struct netlink_ring *ring)
604 unsigned int head = ring->head;
605 const struct nl_mmap_hdr *hdr;
608 hdr = __netlink_lookup_frame(ring, ring->head);
609 if (hdr->nm_status == NL_MMAP_STATUS_UNUSED)
611 if (hdr->nm_status != NL_MMAP_STATUS_SKIP)
613 netlink_increment_head(ring);
614 } while (ring->head != head);
617 static bool netlink_has_valid_frame(struct netlink_ring *ring)
619 unsigned int head = ring->head, pos = head;
620 const struct nl_mmap_hdr *hdr;
623 hdr = __netlink_lookup_frame(ring, pos);
624 if (hdr->nm_status == NL_MMAP_STATUS_VALID)
626 pos = pos != 0 ? pos - 1 : ring->frame_max;
627 } while (pos != head);
632 static bool netlink_dump_space(struct netlink_sock *nlk)
634 struct netlink_ring *ring = &nlk->rx_ring;
635 struct nl_mmap_hdr *hdr;
638 hdr = netlink_current_frame(ring, NL_MMAP_STATUS_UNUSED);
642 n = ring->head + ring->frame_max / 2;
643 if (n > ring->frame_max)
644 n -= ring->frame_max;
646 hdr = __netlink_lookup_frame(ring, n);
648 return hdr->nm_status == NL_MMAP_STATUS_UNUSED;
651 static unsigned int netlink_poll(struct file *file, struct socket *sock,
654 struct sock *sk = sock->sk;
655 struct netlink_sock *nlk = nlk_sk(sk);
659 if (nlk->rx_ring.pg_vec != NULL) {
660 /* Memory mapped sockets don't call recvmsg(), so flow control
661 * for dumps is performed here. A dump is allowed to continue
662 * if at least half the ring is unused.
664 while (nlk->cb_running && netlink_dump_space(nlk)) {
665 err = netlink_dump(sk);
668 sk->sk_error_report(sk);
672 netlink_rcv_wake(sk);
675 mask = datagram_poll(file, sock, wait);
677 /* We could already have received frames in the normal receive
678 * queue, that will show up as NL_MMAP_STATUS_COPY in the ring,
679 * so if mask contains pollin/etc already, there's no point
682 if ((mask & (POLLIN | POLLRDNORM)) != (POLLIN | POLLRDNORM)) {
683 spin_lock_bh(&sk->sk_receive_queue.lock);
684 if (nlk->rx_ring.pg_vec) {
685 if (netlink_has_valid_frame(&nlk->rx_ring))
686 mask |= POLLIN | POLLRDNORM;
688 spin_unlock_bh(&sk->sk_receive_queue.lock);
691 spin_lock_bh(&sk->sk_write_queue.lock);
692 if (nlk->tx_ring.pg_vec) {
693 if (netlink_current_frame(&nlk->tx_ring, NL_MMAP_STATUS_UNUSED))
694 mask |= POLLOUT | POLLWRNORM;
696 spin_unlock_bh(&sk->sk_write_queue.lock);
701 static struct nl_mmap_hdr *netlink_mmap_hdr(struct sk_buff *skb)
703 return (struct nl_mmap_hdr *)(skb->head - NL_MMAP_HDRLEN);
706 static void netlink_ring_setup_skb(struct sk_buff *skb, struct sock *sk,
707 struct netlink_ring *ring,
708 struct nl_mmap_hdr *hdr)
713 size = ring->frame_size - NL_MMAP_HDRLEN;
714 data = (void *)hdr + NL_MMAP_HDRLEN;
718 skb_reset_tail_pointer(skb);
719 skb->end = skb->tail + size;
722 skb->destructor = netlink_skb_destructor;
723 NETLINK_CB(skb).flags |= NETLINK_SKB_MMAPED;
724 NETLINK_CB(skb).sk = sk;
727 static int netlink_mmap_sendmsg(struct sock *sk, struct msghdr *msg,
728 u32 dst_portid, u32 dst_group,
729 struct scm_cookie *scm)
731 struct netlink_sock *nlk = nlk_sk(sk);
732 struct netlink_ring *ring;
733 struct nl_mmap_hdr *hdr;
736 int err = 0, len = 0;
738 mutex_lock(&nlk->pg_vec_lock);
740 ring = &nlk->tx_ring;
741 maxlen = ring->frame_size - NL_MMAP_HDRLEN;
746 hdr = netlink_current_frame(ring, NL_MMAP_STATUS_VALID);
748 if (!(msg->msg_flags & MSG_DONTWAIT) &&
749 atomic_read(&nlk->tx_ring.pending))
754 nm_len = ACCESS_ONCE(hdr->nm_len);
755 if (nm_len > maxlen) {
760 netlink_frame_flush_dcache(hdr, nm_len);
762 skb = alloc_skb(nm_len, GFP_KERNEL);
767 __skb_put(skb, nm_len);
768 memcpy(skb->data, (void *)hdr + NL_MMAP_HDRLEN, nm_len);
769 netlink_set_status(hdr, NL_MMAP_STATUS_UNUSED);
771 netlink_increment_head(ring);
773 NETLINK_CB(skb).portid = nlk->portid;
774 NETLINK_CB(skb).dst_group = dst_group;
775 NETLINK_CB(skb).creds = scm->creds;
777 err = security_netlink_send(sk, skb);
783 if (unlikely(dst_group)) {
784 atomic_inc(&skb->users);
785 netlink_broadcast(sk, skb, dst_portid, dst_group,
788 err = netlink_unicast(sk, skb, dst_portid,
789 msg->msg_flags & MSG_DONTWAIT);
794 } while (hdr != NULL ||
795 (!(msg->msg_flags & MSG_DONTWAIT) &&
796 atomic_read(&nlk->tx_ring.pending)));
801 mutex_unlock(&nlk->pg_vec_lock);
805 static void netlink_queue_mmaped_skb(struct sock *sk, struct sk_buff *skb)
807 struct nl_mmap_hdr *hdr;
809 hdr = netlink_mmap_hdr(skb);
810 hdr->nm_len = skb->len;
811 hdr->nm_group = NETLINK_CB(skb).dst_group;
812 hdr->nm_pid = NETLINK_CB(skb).creds.pid;
813 hdr->nm_uid = from_kuid(sk_user_ns(sk), NETLINK_CB(skb).creds.uid);
814 hdr->nm_gid = from_kgid(sk_user_ns(sk), NETLINK_CB(skb).creds.gid);
815 netlink_frame_flush_dcache(hdr, hdr->nm_len);
816 netlink_set_status(hdr, NL_MMAP_STATUS_VALID);
818 NETLINK_CB(skb).flags |= NETLINK_SKB_DELIVERED;
822 static void netlink_ring_set_copied(struct sock *sk, struct sk_buff *skb)
824 struct netlink_sock *nlk = nlk_sk(sk);
825 struct netlink_ring *ring = &nlk->rx_ring;
826 struct nl_mmap_hdr *hdr;
828 spin_lock_bh(&sk->sk_receive_queue.lock);
829 hdr = netlink_current_frame(ring, NL_MMAP_STATUS_UNUSED);
831 spin_unlock_bh(&sk->sk_receive_queue.lock);
836 netlink_increment_head(ring);
837 __skb_queue_tail(&sk->sk_receive_queue, skb);
838 spin_unlock_bh(&sk->sk_receive_queue.lock);
840 hdr->nm_len = skb->len;
841 hdr->nm_group = NETLINK_CB(skb).dst_group;
842 hdr->nm_pid = NETLINK_CB(skb).creds.pid;
843 hdr->nm_uid = from_kuid(sk_user_ns(sk), NETLINK_CB(skb).creds.uid);
844 hdr->nm_gid = from_kgid(sk_user_ns(sk), NETLINK_CB(skb).creds.gid);
845 netlink_set_status(hdr, NL_MMAP_STATUS_COPY);
848 #else /* CONFIG_NETLINK_MMAP */
849 #define netlink_skb_is_mmaped(skb) false
850 #define netlink_rx_is_mmaped(sk) false
851 #define netlink_tx_is_mmaped(sk) false
852 #define netlink_mmap sock_no_mmap
853 #define netlink_poll datagram_poll
854 #define netlink_mmap_sendmsg(sk, msg, dst_portid, dst_group, scm) 0
855 #endif /* CONFIG_NETLINK_MMAP */
857 static void netlink_skb_destructor(struct sk_buff *skb)
859 #ifdef CONFIG_NETLINK_MMAP
860 struct nl_mmap_hdr *hdr;
861 struct netlink_ring *ring;
864 /* If a packet from the kernel to userspace was freed because of an
865 * error without being delivered to userspace, the kernel must reset
866 * the status. In the direction userspace to kernel, the status is
867 * always reset here after the packet was processed and freed.
869 if (netlink_skb_is_mmaped(skb)) {
870 hdr = netlink_mmap_hdr(skb);
871 sk = NETLINK_CB(skb).sk;
873 if (NETLINK_CB(skb).flags & NETLINK_SKB_TX) {
874 netlink_set_status(hdr, NL_MMAP_STATUS_UNUSED);
875 ring = &nlk_sk(sk)->tx_ring;
877 if (!(NETLINK_CB(skb).flags & NETLINK_SKB_DELIVERED)) {
879 netlink_set_status(hdr, NL_MMAP_STATUS_VALID);
881 ring = &nlk_sk(sk)->rx_ring;
884 WARN_ON(atomic_read(&ring->pending) == 0);
885 atomic_dec(&ring->pending);
891 if (is_vmalloc_addr(skb->head)) {
893 !atomic_dec_return(&(skb_shinfo(skb)->dataref)))
902 static void netlink_skb_set_owner_r(struct sk_buff *skb, struct sock *sk)
904 WARN_ON(skb->sk != NULL);
906 skb->destructor = netlink_skb_destructor;
907 atomic_add(skb->truesize, &sk->sk_rmem_alloc);
908 sk_mem_charge(sk, skb->truesize);
911 static void netlink_sock_destruct(struct sock *sk)
913 struct netlink_sock *nlk = nlk_sk(sk);
915 if (nlk->cb_running) {
917 nlk->cb.done(&nlk->cb);
919 module_put(nlk->cb.module);
920 kfree_skb(nlk->cb.skb);
923 skb_queue_purge(&sk->sk_receive_queue);
924 #ifdef CONFIG_NETLINK_MMAP
926 struct nl_mmap_req req;
928 memset(&req, 0, sizeof(req));
929 if (nlk->rx_ring.pg_vec)
930 __netlink_set_ring(sk, &req, false, NULL, 0);
931 memset(&req, 0, sizeof(req));
932 if (nlk->tx_ring.pg_vec)
933 __netlink_set_ring(sk, &req, true, NULL, 0);
935 #endif /* CONFIG_NETLINK_MMAP */
937 if (!sock_flag(sk, SOCK_DEAD)) {
938 printk(KERN_ERR "Freeing alive netlink socket %p\n", sk);
942 WARN_ON(atomic_read(&sk->sk_rmem_alloc));
943 WARN_ON(atomic_read(&sk->sk_wmem_alloc));
944 WARN_ON(nlk_sk(sk)->groups);
947 /* This lock without WQ_FLAG_EXCLUSIVE is good on UP and it is _very_ bad on
948 * SMP. Look, when several writers sleep and reader wakes them up, all but one
949 * immediately hit write lock and grab all the cpus. Exclusive sleep solves
950 * this, _but_ remember, it adds useless work on UP machines.
953 void netlink_table_grab(void)
954 __acquires(nl_table_lock)
958 write_lock_irq(&nl_table_lock);
960 if (atomic_read(&nl_table_users)) {
961 DECLARE_WAITQUEUE(wait, current);
963 add_wait_queue_exclusive(&nl_table_wait, &wait);
965 set_current_state(TASK_UNINTERRUPTIBLE);
966 if (atomic_read(&nl_table_users) == 0)
968 write_unlock_irq(&nl_table_lock);
970 write_lock_irq(&nl_table_lock);
973 __set_current_state(TASK_RUNNING);
974 remove_wait_queue(&nl_table_wait, &wait);
978 void netlink_table_ungrab(void)
979 __releases(nl_table_lock)
981 write_unlock_irq(&nl_table_lock);
982 wake_up(&nl_table_wait);
986 netlink_lock_table(void)
988 /* read_lock() synchronizes us to netlink_table_grab */
990 read_lock(&nl_table_lock);
991 atomic_inc(&nl_table_users);
992 read_unlock(&nl_table_lock);
996 netlink_unlock_table(void)
998 if (atomic_dec_and_test(&nl_table_users))
999 wake_up(&nl_table_wait);
1002 struct netlink_compare_arg
1004 possible_net_t pnet;
1008 /* Doing sizeof directly may yield 4 extra bytes on 64-bit. */
1009 #define netlink_compare_arg_len \
1010 (offsetof(struct netlink_compare_arg, portid) + sizeof(u32))
1012 static inline int netlink_compare(struct rhashtable_compare_arg *arg,
1015 const struct netlink_compare_arg *x = arg->key;
1016 const struct netlink_sock *nlk = ptr;
1018 return nlk->portid != x->portid ||
1019 !net_eq(sock_net(&nlk->sk), read_pnet(&x->pnet));
1022 static void netlink_compare_arg_init(struct netlink_compare_arg *arg,
1023 struct net *net, u32 portid)
1025 memset(arg, 0, sizeof(*arg));
1026 write_pnet(&arg->pnet, net);
1027 arg->portid = portid;
1030 static struct sock *__netlink_lookup(struct netlink_table *table, u32 portid,
1033 struct netlink_compare_arg arg;
1035 netlink_compare_arg_init(&arg, net, portid);
1036 return rhashtable_lookup_fast(&table->hash, &arg,
1037 netlink_rhashtable_params);
1040 static int __netlink_insert(struct netlink_table *table, struct sock *sk)
1042 struct netlink_compare_arg arg;
1044 netlink_compare_arg_init(&arg, sock_net(sk), nlk_sk(sk)->portid);
1045 return rhashtable_lookup_insert_key(&table->hash, &arg,
1047 netlink_rhashtable_params);
1050 static struct sock *netlink_lookup(struct net *net, int protocol, u32 portid)
1052 struct netlink_table *table = &nl_table[protocol];
1056 sk = __netlink_lookup(table, portid, net);
1064 static const struct proto_ops netlink_ops;
1067 netlink_update_listeners(struct sock *sk)
1069 struct netlink_table *tbl = &nl_table[sk->sk_protocol];
1072 struct listeners *listeners;
1074 listeners = nl_deref_protected(tbl->listeners);
1078 for (i = 0; i < NLGRPLONGS(tbl->groups); i++) {
1080 sk_for_each_bound(sk, &tbl->mc_list) {
1081 if (i < NLGRPLONGS(nlk_sk(sk)->ngroups))
1082 mask |= nlk_sk(sk)->groups[i];
1084 listeners->masks[i] = mask;
1086 /* this function is only called with the netlink table "grabbed", which
1087 * makes sure updates are visible before bind or setsockopt return. */
1090 static int netlink_insert(struct sock *sk, u32 portid)
1092 struct netlink_table *table = &nl_table[sk->sk_protocol];
1098 if (nlk_sk(sk)->portid)
1102 if (BITS_PER_LONG > 32 &&
1103 unlikely(atomic_read(&table->hash.nelems) >= UINT_MAX))
1106 nlk_sk(sk)->portid = portid;
1109 err = __netlink_insert(table, sk);
1111 /* In case the hashtable backend returns with -EBUSY
1112 * from here, it must not escape to the caller.
1114 if (unlikely(err == -EBUSY))
1118 nlk_sk(sk)->portid = 0;
1127 static void netlink_remove(struct sock *sk)
1129 struct netlink_table *table;
1131 table = &nl_table[sk->sk_protocol];
1132 if (!rhashtable_remove_fast(&table->hash, &nlk_sk(sk)->node,
1133 netlink_rhashtable_params)) {
1134 WARN_ON(atomic_read(&sk->sk_refcnt) == 1);
1138 netlink_table_grab();
1139 if (nlk_sk(sk)->subscriptions) {
1140 __sk_del_bind_node(sk);
1141 netlink_update_listeners(sk);
1143 if (sk->sk_protocol == NETLINK_GENERIC)
1144 atomic_inc(&genl_sk_destructing_cnt);
1145 netlink_table_ungrab();
1148 static struct proto netlink_proto = {
1150 .owner = THIS_MODULE,
1151 .obj_size = sizeof(struct netlink_sock),
1154 static int __netlink_create(struct net *net, struct socket *sock,
1155 struct mutex *cb_mutex, int protocol,
1159 struct netlink_sock *nlk;
1161 sock->ops = &netlink_ops;
1163 sk = sk_alloc(net, PF_NETLINK, GFP_KERNEL, &netlink_proto, kern);
1167 sock_init_data(sock, sk);
1171 nlk->cb_mutex = cb_mutex;
1173 nlk->cb_mutex = &nlk->cb_def_mutex;
1174 mutex_init(nlk->cb_mutex);
1176 init_waitqueue_head(&nlk->wait);
1177 #ifdef CONFIG_NETLINK_MMAP
1178 mutex_init(&nlk->pg_vec_lock);
1181 sk->sk_destruct = netlink_sock_destruct;
1182 sk->sk_protocol = protocol;
1186 static int netlink_create(struct net *net, struct socket *sock, int protocol,
1189 struct module *module = NULL;
1190 struct mutex *cb_mutex;
1191 struct netlink_sock *nlk;
1192 int (*bind)(struct net *net, int group);
1193 void (*unbind)(struct net *net, int group);
1196 sock->state = SS_UNCONNECTED;
1198 if (sock->type != SOCK_RAW && sock->type != SOCK_DGRAM)
1199 return -ESOCKTNOSUPPORT;
1201 if (protocol < 0 || protocol >= MAX_LINKS)
1202 return -EPROTONOSUPPORT;
1204 netlink_lock_table();
1205 #ifdef CONFIG_MODULES
1206 if (!nl_table[protocol].registered) {
1207 netlink_unlock_table();
1208 request_module("net-pf-%d-proto-%d", PF_NETLINK, protocol);
1209 netlink_lock_table();
1212 if (nl_table[protocol].registered &&
1213 try_module_get(nl_table[protocol].module))
1214 module = nl_table[protocol].module;
1216 err = -EPROTONOSUPPORT;
1217 cb_mutex = nl_table[protocol].cb_mutex;
1218 bind = nl_table[protocol].bind;
1219 unbind = nl_table[protocol].unbind;
1220 netlink_unlock_table();
1225 err = __netlink_create(net, sock, cb_mutex, protocol, kern);
1230 sock_prot_inuse_add(net, &netlink_proto, 1);
1233 nlk = nlk_sk(sock->sk);
1234 nlk->module = module;
1235 nlk->netlink_bind = bind;
1236 nlk->netlink_unbind = unbind;
1245 static void deferred_put_nlk_sk(struct rcu_head *head)
1247 struct netlink_sock *nlk = container_of(head, struct netlink_sock, rcu);
1252 static int netlink_release(struct socket *sock)
1254 struct sock *sk = sock->sk;
1255 struct netlink_sock *nlk;
1265 * OK. Socket is unlinked, any packets that arrive now
1269 /* must not acquire netlink_table_lock in any way again before unbind
1270 * and notifying genetlink is done as otherwise it might deadlock
1272 if (nlk->netlink_unbind) {
1275 for (i = 0; i < nlk->ngroups; i++)
1276 if (test_bit(i, nlk->groups))
1277 nlk->netlink_unbind(sock_net(sk), i + 1);
1279 if (sk->sk_protocol == NETLINK_GENERIC &&
1280 atomic_dec_return(&genl_sk_destructing_cnt) == 0)
1281 wake_up(&genl_sk_destructing_waitq);
1284 wake_up_interruptible_all(&nlk->wait);
1286 skb_queue_purge(&sk->sk_write_queue);
1289 struct netlink_notify n = {
1290 .net = sock_net(sk),
1291 .protocol = sk->sk_protocol,
1292 .portid = nlk->portid,
1294 atomic_notifier_call_chain(&netlink_chain,
1295 NETLINK_URELEASE, &n);
1298 module_put(nlk->module);
1300 if (netlink_is_kernel(sk)) {
1301 netlink_table_grab();
1302 BUG_ON(nl_table[sk->sk_protocol].registered == 0);
1303 if (--nl_table[sk->sk_protocol].registered == 0) {
1304 struct listeners *old;
1306 old = nl_deref_protected(nl_table[sk->sk_protocol].listeners);
1307 RCU_INIT_POINTER(nl_table[sk->sk_protocol].listeners, NULL);
1308 kfree_rcu(old, rcu);
1309 nl_table[sk->sk_protocol].module = NULL;
1310 nl_table[sk->sk_protocol].bind = NULL;
1311 nl_table[sk->sk_protocol].unbind = NULL;
1312 nl_table[sk->sk_protocol].flags = 0;
1313 nl_table[sk->sk_protocol].registered = 0;
1315 netlink_table_ungrab();
1322 sock_prot_inuse_add(sock_net(sk), &netlink_proto, -1);
1324 call_rcu(&nlk->rcu, deferred_put_nlk_sk);
1328 static int netlink_autobind(struct socket *sock)
1330 struct sock *sk = sock->sk;
1331 struct net *net = sock_net(sk);
1332 struct netlink_table *table = &nl_table[sk->sk_protocol];
1333 s32 portid = task_tgid_vnr(current);
1341 ok = !__netlink_lookup(table, portid, net);
1344 /* Bind collision, search negative portid values. */
1346 /* rover will be in range [S32_MIN, -4097] */
1347 rover = S32_MIN + prandom_u32_max(-4096 - S32_MIN);
1348 else if (rover >= -4096)
1354 err = netlink_insert(sk, portid);
1355 if (err == -EADDRINUSE)
1358 /* If 2 threads race to autobind, that is fine. */
1366 * __netlink_ns_capable - General netlink message capability test
1367 * @nsp: NETLINK_CB of the socket buffer holding a netlink command from userspace.
1368 * @user_ns: The user namespace of the capability to use
1369 * @cap: The capability to use
1371 * Test to see if the opener of the socket we received the message
1372 * from had when the netlink socket was created and the sender of the
1373 * message has has the capability @cap in the user namespace @user_ns.
1375 bool __netlink_ns_capable(const struct netlink_skb_parms *nsp,
1376 struct user_namespace *user_ns, int cap)
1378 return ((nsp->flags & NETLINK_SKB_DST) ||
1379 file_ns_capable(nsp->sk->sk_socket->file, user_ns, cap)) &&
1380 ns_capable(user_ns, cap);
1382 EXPORT_SYMBOL(__netlink_ns_capable);
1385 * netlink_ns_capable - General netlink message capability test
1386 * @skb: socket buffer holding a netlink command from userspace
1387 * @user_ns: The user namespace of the capability to use
1388 * @cap: The capability to use
1390 * Test to see if the opener of the socket we received the message
1391 * from had when the netlink socket was created and the sender of the
1392 * message has has the capability @cap in the user namespace @user_ns.
1394 bool netlink_ns_capable(const struct sk_buff *skb,
1395 struct user_namespace *user_ns, int cap)
1397 return __netlink_ns_capable(&NETLINK_CB(skb), user_ns, cap);
1399 EXPORT_SYMBOL(netlink_ns_capable);
1402 * netlink_capable - Netlink global message capability test
1403 * @skb: socket buffer holding a netlink command from userspace
1404 * @cap: The capability to use
1406 * Test to see if the opener of the socket we received the message
1407 * from had when the netlink socket was created and the sender of the
1408 * message has has the capability @cap in all user namespaces.
1410 bool netlink_capable(const struct sk_buff *skb, int cap)
1412 return netlink_ns_capable(skb, &init_user_ns, cap);
1414 EXPORT_SYMBOL(netlink_capable);
1417 * netlink_net_capable - Netlink network namespace message capability test
1418 * @skb: socket buffer holding a netlink command from userspace
1419 * @cap: The capability to use
1421 * Test to see if the opener of the socket we received the message
1422 * from had when the netlink socket was created and the sender of the
1423 * message has has the capability @cap over the network namespace of
1424 * the socket we received the message from.
1426 bool netlink_net_capable(const struct sk_buff *skb, int cap)
1428 return netlink_ns_capable(skb, sock_net(skb->sk)->user_ns, cap);
1430 EXPORT_SYMBOL(netlink_net_capable);
1432 static inline int netlink_allowed(const struct socket *sock, unsigned int flag)
1434 return (nl_table[sock->sk->sk_protocol].flags & flag) ||
1435 ns_capable(sock_net(sock->sk)->user_ns, CAP_NET_ADMIN);
1439 netlink_update_subscriptions(struct sock *sk, unsigned int subscriptions)
1441 struct netlink_sock *nlk = nlk_sk(sk);
1443 if (nlk->subscriptions && !subscriptions)
1444 __sk_del_bind_node(sk);
1445 else if (!nlk->subscriptions && subscriptions)
1446 sk_add_bind_node(sk, &nl_table[sk->sk_protocol].mc_list);
1447 nlk->subscriptions = subscriptions;
1450 static int netlink_realloc_groups(struct sock *sk)
1452 struct netlink_sock *nlk = nlk_sk(sk);
1453 unsigned int groups;
1454 unsigned long *new_groups;
1457 netlink_table_grab();
1459 groups = nl_table[sk->sk_protocol].groups;
1460 if (!nl_table[sk->sk_protocol].registered) {
1465 if (nlk->ngroups >= groups)
1468 new_groups = krealloc(nlk->groups, NLGRPSZ(groups), GFP_ATOMIC);
1469 if (new_groups == NULL) {
1473 memset((char *)new_groups + NLGRPSZ(nlk->ngroups), 0,
1474 NLGRPSZ(groups) - NLGRPSZ(nlk->ngroups));
1476 nlk->groups = new_groups;
1477 nlk->ngroups = groups;
1479 netlink_table_ungrab();
1483 static void netlink_undo_bind(int group, long unsigned int groups,
1486 struct netlink_sock *nlk = nlk_sk(sk);
1489 if (!nlk->netlink_unbind)
1492 for (undo = 0; undo < group; undo++)
1493 if (test_bit(undo, &groups))
1494 nlk->netlink_unbind(sock_net(sk), undo + 1);
1497 static int netlink_bind(struct socket *sock, struct sockaddr *addr,
1500 struct sock *sk = sock->sk;
1501 struct net *net = sock_net(sk);
1502 struct netlink_sock *nlk = nlk_sk(sk);
1503 struct sockaddr_nl *nladdr = (struct sockaddr_nl *)addr;
1505 long unsigned int groups = nladdr->nl_groups;
1507 if (addr_len < sizeof(struct sockaddr_nl))
1510 if (nladdr->nl_family != AF_NETLINK)
1513 /* Only superuser is allowed to listen multicasts */
1515 if (!netlink_allowed(sock, NL_CFG_F_NONROOT_RECV))
1517 err = netlink_realloc_groups(sk);
1523 if (nladdr->nl_pid != nlk->portid)
1526 if (nlk->netlink_bind && groups) {
1529 for (group = 0; group < nlk->ngroups; group++) {
1530 if (!test_bit(group, &groups))
1532 err = nlk->netlink_bind(net, group + 1);
1535 netlink_undo_bind(group, groups, sk);
1541 err = nladdr->nl_pid ?
1542 netlink_insert(sk, nladdr->nl_pid) :
1543 netlink_autobind(sock);
1545 netlink_undo_bind(nlk->ngroups, groups, sk);
1550 if (!groups && (nlk->groups == NULL || !(u32)nlk->groups[0]))
1553 netlink_table_grab();
1554 netlink_update_subscriptions(sk, nlk->subscriptions +
1556 hweight32(nlk->groups[0]));
1557 nlk->groups[0] = (nlk->groups[0] & ~0xffffffffUL) | groups;
1558 netlink_update_listeners(sk);
1559 netlink_table_ungrab();
1564 static int netlink_connect(struct socket *sock, struct sockaddr *addr,
1565 int alen, int flags)
1568 struct sock *sk = sock->sk;
1569 struct netlink_sock *nlk = nlk_sk(sk);
1570 struct sockaddr_nl *nladdr = (struct sockaddr_nl *)addr;
1572 if (alen < sizeof(addr->sa_family))
1575 if (addr->sa_family == AF_UNSPEC) {
1576 sk->sk_state = NETLINK_UNCONNECTED;
1577 nlk->dst_portid = 0;
1581 if (addr->sa_family != AF_NETLINK)
1584 if ((nladdr->nl_groups || nladdr->nl_pid) &&
1585 !netlink_allowed(sock, NL_CFG_F_NONROOT_SEND))
1589 err = netlink_autobind(sock);
1592 sk->sk_state = NETLINK_CONNECTED;
1593 nlk->dst_portid = nladdr->nl_pid;
1594 nlk->dst_group = ffs(nladdr->nl_groups);
1600 static int netlink_getname(struct socket *sock, struct sockaddr *addr,
1601 int *addr_len, int peer)
1603 struct sock *sk = sock->sk;
1604 struct netlink_sock *nlk = nlk_sk(sk);
1605 DECLARE_SOCKADDR(struct sockaddr_nl *, nladdr, addr);
1607 nladdr->nl_family = AF_NETLINK;
1609 *addr_len = sizeof(*nladdr);
1612 nladdr->nl_pid = nlk->dst_portid;
1613 nladdr->nl_groups = netlink_group_mask(nlk->dst_group);
1615 nladdr->nl_pid = nlk->portid;
1616 nladdr->nl_groups = nlk->groups ? nlk->groups[0] : 0;
1621 static struct sock *netlink_getsockbyportid(struct sock *ssk, u32 portid)
1624 struct netlink_sock *nlk;
1626 sock = netlink_lookup(sock_net(ssk), ssk->sk_protocol, portid);
1628 return ERR_PTR(-ECONNREFUSED);
1630 /* Don't bother queuing skb if kernel socket has no input function */
1632 if (sock->sk_state == NETLINK_CONNECTED &&
1633 nlk->dst_portid != nlk_sk(ssk)->portid) {
1635 return ERR_PTR(-ECONNREFUSED);
1640 struct sock *netlink_getsockbyfilp(struct file *filp)
1642 struct inode *inode = file_inode(filp);
1645 if (!S_ISSOCK(inode->i_mode))
1646 return ERR_PTR(-ENOTSOCK);
1648 sock = SOCKET_I(inode)->sk;
1649 if (sock->sk_family != AF_NETLINK)
1650 return ERR_PTR(-EINVAL);
1656 static struct sk_buff *netlink_alloc_large_skb(unsigned int size,
1659 struct sk_buff *skb;
1662 if (size <= NLMSG_GOODSIZE || broadcast)
1663 return alloc_skb(size, GFP_KERNEL);
1665 size = SKB_DATA_ALIGN(size) +
1666 SKB_DATA_ALIGN(sizeof(struct skb_shared_info));
1668 data = vmalloc(size);
1672 skb = __build_skb(data, size);
1676 skb->destructor = netlink_skb_destructor;
1682 * Attach a skb to a netlink socket.
1683 * The caller must hold a reference to the destination socket. On error, the
1684 * reference is dropped. The skb is not send to the destination, just all
1685 * all error checks are performed and memory in the queue is reserved.
1687 * < 0: error. skb freed, reference to sock dropped.
1689 * 1: repeat lookup - reference dropped while waiting for socket memory.
1691 int netlink_attachskb(struct sock *sk, struct sk_buff *skb,
1692 long *timeo, struct sock *ssk)
1694 struct netlink_sock *nlk;
1698 if ((atomic_read(&sk->sk_rmem_alloc) > sk->sk_rcvbuf ||
1699 test_bit(NETLINK_S_CONGESTED, &nlk->state)) &&
1700 !netlink_skb_is_mmaped(skb)) {
1701 DECLARE_WAITQUEUE(wait, current);
1703 if (!ssk || netlink_is_kernel(ssk))
1704 netlink_overrun(sk);
1710 __set_current_state(TASK_INTERRUPTIBLE);
1711 add_wait_queue(&nlk->wait, &wait);
1713 if ((atomic_read(&sk->sk_rmem_alloc) > sk->sk_rcvbuf ||
1714 test_bit(NETLINK_S_CONGESTED, &nlk->state)) &&
1715 !sock_flag(sk, SOCK_DEAD))
1716 *timeo = schedule_timeout(*timeo);
1718 __set_current_state(TASK_RUNNING);
1719 remove_wait_queue(&nlk->wait, &wait);
1722 if (signal_pending(current)) {
1724 return sock_intr_errno(*timeo);
1728 netlink_skb_set_owner_r(skb, sk);
1732 static int __netlink_sendskb(struct sock *sk, struct sk_buff *skb)
1736 netlink_deliver_tap(skb);
1738 #ifdef CONFIG_NETLINK_MMAP
1739 if (netlink_skb_is_mmaped(skb))
1740 netlink_queue_mmaped_skb(sk, skb);
1741 else if (netlink_rx_is_mmaped(sk))
1742 netlink_ring_set_copied(sk, skb);
1744 #endif /* CONFIG_NETLINK_MMAP */
1745 skb_queue_tail(&sk->sk_receive_queue, skb);
1746 sk->sk_data_ready(sk);
1750 int netlink_sendskb(struct sock *sk, struct sk_buff *skb)
1752 int len = __netlink_sendskb(sk, skb);
1758 void netlink_detachskb(struct sock *sk, struct sk_buff *skb)
1764 static struct sk_buff *netlink_trim(struct sk_buff *skb, gfp_t allocation)
1768 WARN_ON(skb->sk != NULL);
1769 if (netlink_skb_is_mmaped(skb))
1772 delta = skb->end - skb->tail;
1773 if (is_vmalloc_addr(skb->head) || delta * 2 < skb->truesize)
1776 if (skb_shared(skb)) {
1777 struct sk_buff *nskb = skb_clone(skb, allocation);
1784 if (!pskb_expand_head(skb, 0, -delta, allocation))
1785 skb->truesize -= delta;
1790 static int netlink_unicast_kernel(struct sock *sk, struct sk_buff *skb,
1794 struct netlink_sock *nlk = nlk_sk(sk);
1796 ret = -ECONNREFUSED;
1797 if (nlk->netlink_rcv != NULL) {
1799 netlink_skb_set_owner_r(skb, sk);
1800 NETLINK_CB(skb).sk = ssk;
1801 netlink_deliver_tap_kernel(sk, ssk, skb);
1802 nlk->netlink_rcv(skb);
1811 int netlink_unicast(struct sock *ssk, struct sk_buff *skb,
1812 u32 portid, int nonblock)
1818 skb = netlink_trim(skb, gfp_any());
1820 timeo = sock_sndtimeo(ssk, nonblock);
1822 sk = netlink_getsockbyportid(ssk, portid);
1827 if (netlink_is_kernel(sk))
1828 return netlink_unicast_kernel(sk, skb, ssk);
1830 if (sk_filter(sk, skb)) {
1837 err = netlink_attachskb(sk, skb, &timeo, ssk);
1843 return netlink_sendskb(sk, skb);
1845 EXPORT_SYMBOL(netlink_unicast);
1847 struct sk_buff *__netlink_alloc_skb(struct sock *ssk, unsigned int size,
1848 unsigned int ldiff, u32 dst_portid,
1851 #ifdef CONFIG_NETLINK_MMAP
1852 unsigned int maxlen, linear_size;
1853 struct sock *sk = NULL;
1854 struct sk_buff *skb;
1855 struct netlink_ring *ring;
1856 struct nl_mmap_hdr *hdr;
1858 sk = netlink_getsockbyportid(ssk, dst_portid);
1862 ring = &nlk_sk(sk)->rx_ring;
1863 /* fast-path without atomic ops for common case: non-mmaped receiver */
1864 if (ring->pg_vec == NULL)
1867 /* We need to account the full linear size needed as a ring
1868 * slot cannot have non-linear parts.
1870 linear_size = size + ldiff;
1871 if (ring->frame_size - NL_MMAP_HDRLEN < linear_size)
1874 skb = alloc_skb_head(gfp_mask);
1878 spin_lock_bh(&sk->sk_receive_queue.lock);
1879 /* check again under lock */
1880 if (ring->pg_vec == NULL)
1883 /* check again under lock */
1884 maxlen = ring->frame_size - NL_MMAP_HDRLEN;
1885 if (maxlen < linear_size)
1888 netlink_forward_ring(ring);
1889 hdr = netlink_current_frame(ring, NL_MMAP_STATUS_UNUSED);
1893 netlink_ring_setup_skb(skb, sk, ring, hdr);
1894 netlink_set_status(hdr, NL_MMAP_STATUS_RESERVED);
1895 atomic_inc(&ring->pending);
1896 netlink_increment_head(ring);
1898 spin_unlock_bh(&sk->sk_receive_queue.lock);
1903 spin_unlock_bh(&sk->sk_receive_queue.lock);
1904 netlink_overrun(sk);
1911 spin_unlock_bh(&sk->sk_receive_queue.lock);
1916 return alloc_skb(size, gfp_mask);
1918 EXPORT_SYMBOL_GPL(__netlink_alloc_skb);
1920 int netlink_has_listeners(struct sock *sk, unsigned int group)
1923 struct listeners *listeners;
1925 BUG_ON(!netlink_is_kernel(sk));
1928 listeners = rcu_dereference(nl_table[sk->sk_protocol].listeners);
1930 if (listeners && group - 1 < nl_table[sk->sk_protocol].groups)
1931 res = test_bit(group - 1, listeners->masks);
1937 EXPORT_SYMBOL_GPL(netlink_has_listeners);
1939 static int netlink_broadcast_deliver(struct sock *sk, struct sk_buff *skb)
1941 struct netlink_sock *nlk = nlk_sk(sk);
1943 if (atomic_read(&sk->sk_rmem_alloc) <= sk->sk_rcvbuf &&
1944 !test_bit(NETLINK_S_CONGESTED, &nlk->state)) {
1945 netlink_skb_set_owner_r(skb, sk);
1946 __netlink_sendskb(sk, skb);
1947 return atomic_read(&sk->sk_rmem_alloc) > (sk->sk_rcvbuf >> 1);
1952 struct netlink_broadcast_data {
1953 struct sock *exclude_sk;
1958 int delivery_failure;
1962 struct sk_buff *skb, *skb2;
1963 int (*tx_filter)(struct sock *dsk, struct sk_buff *skb, void *data);
1967 static void do_one_broadcast(struct sock *sk,
1968 struct netlink_broadcast_data *p)
1970 struct netlink_sock *nlk = nlk_sk(sk);
1973 if (p->exclude_sk == sk)
1976 if (nlk->portid == p->portid || p->group - 1 >= nlk->ngroups ||
1977 !test_bit(p->group - 1, nlk->groups))
1980 if (!net_eq(sock_net(sk), p->net)) {
1981 if (!(nlk->flags & NETLINK_F_LISTEN_ALL_NSID))
1984 if (!peernet_has_id(sock_net(sk), p->net))
1987 if (!file_ns_capable(sk->sk_socket->file, p->net->user_ns,
1993 netlink_overrun(sk);
1998 if (p->skb2 == NULL) {
1999 if (skb_shared(p->skb)) {
2000 p->skb2 = skb_clone(p->skb, p->allocation);
2002 p->skb2 = skb_get(p->skb);
2004 * skb ownership may have been set when
2005 * delivered to a previous socket.
2007 skb_orphan(p->skb2);
2010 if (p->skb2 == NULL) {
2011 netlink_overrun(sk);
2012 /* Clone failed. Notify ALL listeners. */
2014 if (nlk->flags & NETLINK_F_BROADCAST_SEND_ERROR)
2015 p->delivery_failure = 1;
2018 if (p->tx_filter && p->tx_filter(sk, p->skb2, p->tx_data)) {
2023 if (sk_filter(sk, p->skb2)) {
2028 NETLINK_CB(p->skb2).nsid = peernet2id(sock_net(sk), p->net);
2029 NETLINK_CB(p->skb2).nsid_is_set = true;
2030 val = netlink_broadcast_deliver(sk, p->skb2);
2032 netlink_overrun(sk);
2033 if (nlk->flags & NETLINK_F_BROADCAST_SEND_ERROR)
2034 p->delivery_failure = 1;
2036 p->congested |= val;
2044 int netlink_broadcast_filtered(struct sock *ssk, struct sk_buff *skb, u32 portid,
2045 u32 group, gfp_t allocation,
2046 int (*filter)(struct sock *dsk, struct sk_buff *skb, void *data),
2049 struct net *net = sock_net(ssk);
2050 struct netlink_broadcast_data info;
2053 skb = netlink_trim(skb, allocation);
2055 info.exclude_sk = ssk;
2057 info.portid = portid;
2060 info.delivery_failure = 0;
2063 info.allocation = allocation;
2066 info.tx_filter = filter;
2067 info.tx_data = filter_data;
2069 /* While we sleep in clone, do not allow to change socket list */
2071 netlink_lock_table();
2073 sk_for_each_bound(sk, &nl_table[ssk->sk_protocol].mc_list)
2074 do_one_broadcast(sk, &info);
2078 netlink_unlock_table();
2080 if (info.delivery_failure) {
2081 kfree_skb(info.skb2);
2084 consume_skb(info.skb2);
2086 if (info.delivered) {
2087 if (info.congested && (allocation & __GFP_WAIT))
2093 EXPORT_SYMBOL(netlink_broadcast_filtered);
2095 int netlink_broadcast(struct sock *ssk, struct sk_buff *skb, u32 portid,
2096 u32 group, gfp_t allocation)
2098 return netlink_broadcast_filtered(ssk, skb, portid, group, allocation,
2101 EXPORT_SYMBOL(netlink_broadcast);
2103 struct netlink_set_err_data {
2104 struct sock *exclude_sk;
2110 static int do_one_set_err(struct sock *sk, struct netlink_set_err_data *p)
2112 struct netlink_sock *nlk = nlk_sk(sk);
2115 if (sk == p->exclude_sk)
2118 if (!net_eq(sock_net(sk), sock_net(p->exclude_sk)))
2121 if (nlk->portid == p->portid || p->group - 1 >= nlk->ngroups ||
2122 !test_bit(p->group - 1, nlk->groups))
2125 if (p->code == ENOBUFS && nlk->flags & NETLINK_F_RECV_NO_ENOBUFS) {
2130 sk->sk_err = p->code;
2131 sk->sk_error_report(sk);
2137 * netlink_set_err - report error to broadcast listeners
2138 * @ssk: the kernel netlink socket, as returned by netlink_kernel_create()
2139 * @portid: the PORTID of a process that we want to skip (if any)
2140 * @group: the broadcast group that will notice the error
2141 * @code: error code, must be negative (as usual in kernelspace)
2143 * This function returns the number of broadcast listeners that have set the
2144 * NETLINK_NO_ENOBUFS socket option.
2146 int netlink_set_err(struct sock *ssk, u32 portid, u32 group, int code)
2148 struct netlink_set_err_data info;
2152 info.exclude_sk = ssk;
2153 info.portid = portid;
2155 /* sk->sk_err wants a positive error value */
2158 read_lock(&nl_table_lock);
2160 sk_for_each_bound(sk, &nl_table[ssk->sk_protocol].mc_list)
2161 ret += do_one_set_err(sk, &info);
2163 read_unlock(&nl_table_lock);
2166 EXPORT_SYMBOL(netlink_set_err);
2168 /* must be called with netlink table grabbed */
2169 static void netlink_update_socket_mc(struct netlink_sock *nlk,
2173 int old, new = !!is_new, subscriptions;
2175 old = test_bit(group - 1, nlk->groups);
2176 subscriptions = nlk->subscriptions - old + new;
2178 __set_bit(group - 1, nlk->groups);
2180 __clear_bit(group - 1, nlk->groups);
2181 netlink_update_subscriptions(&nlk->sk, subscriptions);
2182 netlink_update_listeners(&nlk->sk);
2185 static int netlink_setsockopt(struct socket *sock, int level, int optname,
2186 char __user *optval, unsigned int optlen)
2188 struct sock *sk = sock->sk;
2189 struct netlink_sock *nlk = nlk_sk(sk);
2190 unsigned int val = 0;
2193 if (level != SOL_NETLINK)
2194 return -ENOPROTOOPT;
2196 if (optname != NETLINK_RX_RING && optname != NETLINK_TX_RING &&
2197 optlen >= sizeof(int) &&
2198 get_user(val, (unsigned int __user *)optval))
2202 case NETLINK_PKTINFO:
2204 nlk->flags |= NETLINK_F_RECV_PKTINFO;
2206 nlk->flags &= ~NETLINK_F_RECV_PKTINFO;
2209 case NETLINK_ADD_MEMBERSHIP:
2210 case NETLINK_DROP_MEMBERSHIP: {
2211 if (!netlink_allowed(sock, NL_CFG_F_NONROOT_RECV))
2213 err = netlink_realloc_groups(sk);
2216 if (!val || val - 1 >= nlk->ngroups)
2218 if (optname == NETLINK_ADD_MEMBERSHIP && nlk->netlink_bind) {
2219 err = nlk->netlink_bind(sock_net(sk), val);
2223 netlink_table_grab();
2224 netlink_update_socket_mc(nlk, val,
2225 optname == NETLINK_ADD_MEMBERSHIP);
2226 netlink_table_ungrab();
2227 if (optname == NETLINK_DROP_MEMBERSHIP && nlk->netlink_unbind)
2228 nlk->netlink_unbind(sock_net(sk), val);
2233 case NETLINK_BROADCAST_ERROR:
2235 nlk->flags |= NETLINK_F_BROADCAST_SEND_ERROR;
2237 nlk->flags &= ~NETLINK_F_BROADCAST_SEND_ERROR;
2240 case NETLINK_NO_ENOBUFS:
2242 nlk->flags |= NETLINK_F_RECV_NO_ENOBUFS;
2243 clear_bit(NETLINK_S_CONGESTED, &nlk->state);
2244 wake_up_interruptible(&nlk->wait);
2246 nlk->flags &= ~NETLINK_F_RECV_NO_ENOBUFS;
2250 #ifdef CONFIG_NETLINK_MMAP
2251 case NETLINK_RX_RING:
2252 case NETLINK_TX_RING: {
2253 struct nl_mmap_req req;
2255 /* Rings might consume more memory than queue limits, require
2258 if (!capable(CAP_NET_ADMIN))
2260 if (optlen < sizeof(req))
2262 if (copy_from_user(&req, optval, sizeof(req)))
2264 err = netlink_set_ring(sk, &req,
2265 optname == NETLINK_TX_RING);
2268 #endif /* CONFIG_NETLINK_MMAP */
2269 case NETLINK_LISTEN_ALL_NSID:
2270 if (!ns_capable(sock_net(sk)->user_ns, CAP_NET_BROADCAST))
2274 nlk->flags |= NETLINK_F_LISTEN_ALL_NSID;
2276 nlk->flags &= ~NETLINK_F_LISTEN_ALL_NSID;
2279 case NETLINK_CAP_ACK:
2281 nlk->flags |= NETLINK_F_CAP_ACK;
2283 nlk->flags &= ~NETLINK_F_CAP_ACK;
2292 static int netlink_getsockopt(struct socket *sock, int level, int optname,
2293 char __user *optval, int __user *optlen)
2295 struct sock *sk = sock->sk;
2296 struct netlink_sock *nlk = nlk_sk(sk);
2299 if (level != SOL_NETLINK)
2300 return -ENOPROTOOPT;
2302 if (get_user(len, optlen))
2308 case NETLINK_PKTINFO:
2309 if (len < sizeof(int))
2312 val = nlk->flags & NETLINK_F_RECV_PKTINFO ? 1 : 0;
2313 if (put_user(len, optlen) ||
2314 put_user(val, optval))
2318 case NETLINK_BROADCAST_ERROR:
2319 if (len < sizeof(int))
2322 val = nlk->flags & NETLINK_F_BROADCAST_SEND_ERROR ? 1 : 0;
2323 if (put_user(len, optlen) ||
2324 put_user(val, optval))
2328 case NETLINK_NO_ENOBUFS:
2329 if (len < sizeof(int))
2332 val = nlk->flags & NETLINK_F_RECV_NO_ENOBUFS ? 1 : 0;
2333 if (put_user(len, optlen) ||
2334 put_user(val, optval))
2338 case NETLINK_LIST_MEMBERSHIPS: {
2339 int pos, idx, shift;
2342 netlink_table_grab();
2343 for (pos = 0; pos * 8 < nlk->ngroups; pos += sizeof(u32)) {
2344 if (len - pos < sizeof(u32))
2347 idx = pos / sizeof(unsigned long);
2348 shift = (pos % sizeof(unsigned long)) * 8;
2349 if (put_user((u32)(nlk->groups[idx] >> shift),
2350 (u32 __user *)(optval + pos))) {
2355 if (put_user(ALIGN(nlk->ngroups / 8, sizeof(u32)), optlen))
2357 netlink_table_ungrab();
2360 case NETLINK_CAP_ACK:
2361 if (len < sizeof(int))
2364 val = nlk->flags & NETLINK_F_CAP_ACK ? 1 : 0;
2365 if (put_user(len, optlen) ||
2366 put_user(val, optval))
2376 static void netlink_cmsg_recv_pktinfo(struct msghdr *msg, struct sk_buff *skb)
2378 struct nl_pktinfo info;
2380 info.group = NETLINK_CB(skb).dst_group;
2381 put_cmsg(msg, SOL_NETLINK, NETLINK_PKTINFO, sizeof(info), &info);
2384 static void netlink_cmsg_listen_all_nsid(struct sock *sk, struct msghdr *msg,
2385 struct sk_buff *skb)
2387 if (!NETLINK_CB(skb).nsid_is_set)
2390 put_cmsg(msg, SOL_NETLINK, NETLINK_LISTEN_ALL_NSID, sizeof(int),
2391 &NETLINK_CB(skb).nsid);
2394 static int netlink_sendmsg(struct socket *sock, struct msghdr *msg, size_t len)
2396 struct sock *sk = sock->sk;
2397 struct netlink_sock *nlk = nlk_sk(sk);
2398 DECLARE_SOCKADDR(struct sockaddr_nl *, addr, msg->msg_name);
2401 struct sk_buff *skb;
2403 struct scm_cookie scm;
2404 u32 netlink_skb_flags = 0;
2406 if (msg->msg_flags&MSG_OOB)
2409 err = scm_send(sock, msg, &scm, true);
2413 if (msg->msg_namelen) {
2415 if (addr->nl_family != AF_NETLINK)
2417 dst_portid = addr->nl_pid;
2418 dst_group = ffs(addr->nl_groups);
2420 if ((dst_group || dst_portid) &&
2421 !netlink_allowed(sock, NL_CFG_F_NONROOT_SEND))
2423 netlink_skb_flags |= NETLINK_SKB_DST;
2425 dst_portid = nlk->dst_portid;
2426 dst_group = nlk->dst_group;
2430 err = netlink_autobind(sock);
2435 /* It's a really convoluted way for userland to ask for mmaped
2436 * sendmsg(), but that's what we've got...
2438 if (netlink_tx_is_mmaped(sk) &&
2439 iter_is_iovec(&msg->msg_iter) &&
2440 msg->msg_iter.nr_segs == 1 &&
2441 msg->msg_iter.iov->iov_base == NULL) {
2442 err = netlink_mmap_sendmsg(sk, msg, dst_portid, dst_group,
2448 if (len > sk->sk_sndbuf - 32)
2451 skb = netlink_alloc_large_skb(len, dst_group);
2455 NETLINK_CB(skb).portid = nlk->portid;
2456 NETLINK_CB(skb).dst_group = dst_group;
2457 NETLINK_CB(skb).creds = scm.creds;
2458 NETLINK_CB(skb).flags = netlink_skb_flags;
2461 if (memcpy_from_msg(skb_put(skb, len), msg, len)) {
2466 err = security_netlink_send(sk, skb);
2473 atomic_inc(&skb->users);
2474 netlink_broadcast(sk, skb, dst_portid, dst_group, GFP_KERNEL);
2476 err = netlink_unicast(sk, skb, dst_portid, msg->msg_flags&MSG_DONTWAIT);
2483 static int netlink_recvmsg(struct socket *sock, struct msghdr *msg, size_t len,
2486 struct scm_cookie scm;
2487 struct sock *sk = sock->sk;
2488 struct netlink_sock *nlk = nlk_sk(sk);
2489 int noblock = flags&MSG_DONTWAIT;
2491 struct sk_buff *skb, *data_skb;
2499 skb = skb_recv_datagram(sk, flags, noblock, &err);
2505 #ifdef CONFIG_COMPAT_NETLINK_MESSAGES
2506 if (unlikely(skb_shinfo(skb)->frag_list)) {
2508 * If this skb has a frag_list, then here that means that we
2509 * will have to use the frag_list skb's data for compat tasks
2510 * and the regular skb's data for normal (non-compat) tasks.
2512 * If we need to send the compat skb, assign it to the
2513 * 'data_skb' variable so that it will be used below for data
2514 * copying. We keep 'skb' for everything else, including
2515 * freeing both later.
2517 if (flags & MSG_CMSG_COMPAT)
2518 data_skb = skb_shinfo(skb)->frag_list;
2522 /* Record the max length of recvmsg() calls for future allocations */
2523 nlk->max_recvmsg_len = max(nlk->max_recvmsg_len, len);
2524 nlk->max_recvmsg_len = min_t(size_t, nlk->max_recvmsg_len,
2527 copied = data_skb->len;
2529 msg->msg_flags |= MSG_TRUNC;
2533 skb_reset_transport_header(data_skb);
2534 err = skb_copy_datagram_msg(data_skb, 0, msg, copied);
2536 if (msg->msg_name) {
2537 DECLARE_SOCKADDR(struct sockaddr_nl *, addr, msg->msg_name);
2538 addr->nl_family = AF_NETLINK;
2540 addr->nl_pid = NETLINK_CB(skb).portid;
2541 addr->nl_groups = netlink_group_mask(NETLINK_CB(skb).dst_group);
2542 msg->msg_namelen = sizeof(*addr);
2545 if (nlk->flags & NETLINK_F_RECV_PKTINFO)
2546 netlink_cmsg_recv_pktinfo(msg, skb);
2547 if (nlk->flags & NETLINK_F_LISTEN_ALL_NSID)
2548 netlink_cmsg_listen_all_nsid(sk, msg, skb);
2550 memset(&scm, 0, sizeof(scm));
2551 scm.creds = *NETLINK_CREDS(skb);
2552 if (flags & MSG_TRUNC)
2553 copied = data_skb->len;
2555 skb_free_datagram(sk, skb);
2557 if (nlk->cb_running &&
2558 atomic_read(&sk->sk_rmem_alloc) <= sk->sk_rcvbuf / 2) {
2559 ret = netlink_dump(sk);
2562 sk->sk_error_report(sk);
2566 scm_recv(sock, msg, &scm, flags);
2568 netlink_rcv_wake(sk);
2569 return err ? : copied;
2572 static void netlink_data_ready(struct sock *sk)
2578 * We export these functions to other modules. They provide a
2579 * complete set of kernel non-blocking support for message
2584 __netlink_kernel_create(struct net *net, int unit, struct module *module,
2585 struct netlink_kernel_cfg *cfg)
2587 struct socket *sock;
2589 struct netlink_sock *nlk;
2590 struct listeners *listeners = NULL;
2591 struct mutex *cb_mutex = cfg ? cfg->cb_mutex : NULL;
2592 unsigned int groups;
2596 if (unit < 0 || unit >= MAX_LINKS)
2599 if (sock_create_lite(PF_NETLINK, SOCK_DGRAM, unit, &sock))
2602 if (__netlink_create(net, sock, cb_mutex, unit, 1) < 0)
2603 goto out_sock_release_nosk;
2607 if (!cfg || cfg->groups < 32)
2610 groups = cfg->groups;
2612 listeners = kzalloc(sizeof(*listeners) + NLGRPSZ(groups), GFP_KERNEL);
2614 goto out_sock_release;
2616 sk->sk_data_ready = netlink_data_ready;
2617 if (cfg && cfg->input)
2618 nlk_sk(sk)->netlink_rcv = cfg->input;
2620 if (netlink_insert(sk, 0))
2621 goto out_sock_release;
2624 nlk->flags |= NETLINK_F_KERNEL_SOCKET;
2626 netlink_table_grab();
2627 if (!nl_table[unit].registered) {
2628 nl_table[unit].groups = groups;
2629 rcu_assign_pointer(nl_table[unit].listeners, listeners);
2630 nl_table[unit].cb_mutex = cb_mutex;
2631 nl_table[unit].module = module;
2633 nl_table[unit].bind = cfg->bind;
2634 nl_table[unit].unbind = cfg->unbind;
2635 nl_table[unit].flags = cfg->flags;
2637 nl_table[unit].compare = cfg->compare;
2639 nl_table[unit].registered = 1;
2642 nl_table[unit].registered++;
2644 netlink_table_ungrab();
2649 netlink_kernel_release(sk);
2652 out_sock_release_nosk:
2656 EXPORT_SYMBOL(__netlink_kernel_create);
2659 netlink_kernel_release(struct sock *sk)
2661 if (sk == NULL || sk->sk_socket == NULL)
2664 sock_release(sk->sk_socket);
2666 EXPORT_SYMBOL(netlink_kernel_release);
2668 int __netlink_change_ngroups(struct sock *sk, unsigned int groups)
2670 struct listeners *new, *old;
2671 struct netlink_table *tbl = &nl_table[sk->sk_protocol];
2676 if (NLGRPSZ(tbl->groups) < NLGRPSZ(groups)) {
2677 new = kzalloc(sizeof(*new) + NLGRPSZ(groups), GFP_ATOMIC);
2680 old = nl_deref_protected(tbl->listeners);
2681 memcpy(new->masks, old->masks, NLGRPSZ(tbl->groups));
2682 rcu_assign_pointer(tbl->listeners, new);
2684 kfree_rcu(old, rcu);
2686 tbl->groups = groups;
2692 * netlink_change_ngroups - change number of multicast groups
2694 * This changes the number of multicast groups that are available
2695 * on a certain netlink family. Note that it is not possible to
2696 * change the number of groups to below 32. Also note that it does
2697 * not implicitly call netlink_clear_multicast_users() when the
2698 * number of groups is reduced.
2700 * @sk: The kernel netlink socket, as returned by netlink_kernel_create().
2701 * @groups: The new number of groups.
2703 int netlink_change_ngroups(struct sock *sk, unsigned int groups)
2707 netlink_table_grab();
2708 err = __netlink_change_ngroups(sk, groups);
2709 netlink_table_ungrab();
2714 void __netlink_clear_multicast_users(struct sock *ksk, unsigned int group)
2717 struct netlink_table *tbl = &nl_table[ksk->sk_protocol];
2719 sk_for_each_bound(sk, &tbl->mc_list)
2720 netlink_update_socket_mc(nlk_sk(sk), group, 0);
2724 __nlmsg_put(struct sk_buff *skb, u32 portid, u32 seq, int type, int len, int flags)
2726 struct nlmsghdr *nlh;
2727 int size = nlmsg_msg_size(len);
2729 nlh = (struct nlmsghdr *)skb_put(skb, NLMSG_ALIGN(size));
2730 nlh->nlmsg_type = type;
2731 nlh->nlmsg_len = size;
2732 nlh->nlmsg_flags = flags;
2733 nlh->nlmsg_pid = portid;
2734 nlh->nlmsg_seq = seq;
2735 if (!__builtin_constant_p(size) || NLMSG_ALIGN(size) - size != 0)
2736 memset(nlmsg_data(nlh) + len, 0, NLMSG_ALIGN(size) - size);
2739 EXPORT_SYMBOL(__nlmsg_put);
2742 * It looks a bit ugly.
2743 * It would be better to create kernel thread.
2746 static int netlink_dump(struct sock *sk)
2748 struct netlink_sock *nlk = nlk_sk(sk);
2749 struct netlink_callback *cb;
2750 struct sk_buff *skb = NULL;
2751 struct nlmsghdr *nlh;
2752 int len, err = -ENOBUFS;
2755 mutex_lock(nlk->cb_mutex);
2756 if (!nlk->cb_running) {
2762 alloc_size = max_t(int, cb->min_dump_alloc, NLMSG_GOODSIZE);
2764 if (!netlink_rx_is_mmaped(sk) &&
2765 atomic_read(&sk->sk_rmem_alloc) >= sk->sk_rcvbuf)
2768 /* NLMSG_GOODSIZE is small to avoid high order allocations being
2769 * required, but it makes sense to _attempt_ a 16K bytes allocation
2770 * to reduce number of system calls on dump operations, if user
2771 * ever provided a big enough buffer.
2773 if (alloc_size < nlk->max_recvmsg_len) {
2774 skb = netlink_alloc_skb(sk,
2775 nlk->max_recvmsg_len,
2780 /* available room should be exact amount to avoid MSG_TRUNC */
2782 skb_reserve(skb, skb_tailroom(skb) -
2783 nlk->max_recvmsg_len);
2786 skb = netlink_alloc_skb(sk, alloc_size, nlk->portid,
2790 netlink_skb_set_owner_r(skb, sk);
2792 len = cb->dump(skb, cb);
2795 mutex_unlock(nlk->cb_mutex);
2797 if (sk_filter(sk, skb))
2800 __netlink_sendskb(sk, skb);
2804 nlh = nlmsg_put_answer(skb, cb, NLMSG_DONE, sizeof(len), NLM_F_MULTI);
2808 nl_dump_check_consistent(cb, nlh);
2810 memcpy(nlmsg_data(nlh), &len, sizeof(len));
2812 if (sk_filter(sk, skb))
2815 __netlink_sendskb(sk, skb);
2820 nlk->cb_running = false;
2821 mutex_unlock(nlk->cb_mutex);
2822 module_put(cb->module);
2823 consume_skb(cb->skb);
2827 mutex_unlock(nlk->cb_mutex);
2832 int __netlink_dump_start(struct sock *ssk, struct sk_buff *skb,
2833 const struct nlmsghdr *nlh,
2834 struct netlink_dump_control *control)
2836 struct netlink_callback *cb;
2838 struct netlink_sock *nlk;
2841 /* Memory mapped dump requests need to be copied to avoid looping
2842 * on the pending state in netlink_mmap_sendmsg() while the CB hold
2843 * a reference to the skb.
2845 if (netlink_skb_is_mmaped(skb)) {
2846 skb = skb_copy(skb, GFP_KERNEL);
2850 atomic_inc(&skb->users);
2852 sk = netlink_lookup(sock_net(ssk), ssk->sk_protocol, NETLINK_CB(skb).portid);
2854 ret = -ECONNREFUSED;
2859 mutex_lock(nlk->cb_mutex);
2860 /* A dump is in progress... */
2861 if (nlk->cb_running) {
2865 /* add reference of module which cb->dump belongs to */
2866 if (!try_module_get(control->module)) {
2867 ret = -EPROTONOSUPPORT;
2872 memset(cb, 0, sizeof(*cb));
2873 cb->dump = control->dump;
2874 cb->done = control->done;
2876 cb->data = control->data;
2877 cb->module = control->module;
2878 cb->min_dump_alloc = control->min_dump_alloc;
2881 nlk->cb_running = true;
2883 mutex_unlock(nlk->cb_mutex);
2885 ret = netlink_dump(sk);
2891 /* We successfully started a dump, by returning -EINTR we
2892 * signal not to send ACK even if it was requested.
2898 mutex_unlock(nlk->cb_mutex);
2903 EXPORT_SYMBOL(__netlink_dump_start);
2905 void netlink_ack(struct sk_buff *in_skb, struct nlmsghdr *nlh, int err)
2907 struct sk_buff *skb;
2908 struct nlmsghdr *rep;
2909 struct nlmsgerr *errmsg;
2910 size_t payload = sizeof(*errmsg);
2911 struct netlink_sock *nlk = nlk_sk(NETLINK_CB(in_skb).sk);
2913 /* Error messages get the original request appened, unless the user
2914 * requests to cap the error message.
2916 if (!(nlk->flags & NETLINK_F_CAP_ACK) && err)
2917 payload += nlmsg_len(nlh);
2919 skb = netlink_alloc_skb(in_skb->sk, nlmsg_total_size(payload),
2920 NETLINK_CB(in_skb).portid, GFP_KERNEL);
2924 sk = netlink_lookup(sock_net(in_skb->sk),
2925 in_skb->sk->sk_protocol,
2926 NETLINK_CB(in_skb).portid);
2928 sk->sk_err = ENOBUFS;
2929 sk->sk_error_report(sk);
2935 rep = __nlmsg_put(skb, NETLINK_CB(in_skb).portid, nlh->nlmsg_seq,
2936 NLMSG_ERROR, payload, 0);
2937 errmsg = nlmsg_data(rep);
2938 errmsg->error = err;
2939 memcpy(&errmsg->msg, nlh, payload > sizeof(*errmsg) ? nlh->nlmsg_len : sizeof(*nlh));
2940 netlink_unicast(in_skb->sk, skb, NETLINK_CB(in_skb).portid, MSG_DONTWAIT);
2942 EXPORT_SYMBOL(netlink_ack);
2944 int netlink_rcv_skb(struct sk_buff *skb, int (*cb)(struct sk_buff *,
2947 struct nlmsghdr *nlh;
2950 while (skb->len >= nlmsg_total_size(0)) {
2953 nlh = nlmsg_hdr(skb);
2956 if (nlh->nlmsg_len < NLMSG_HDRLEN || skb->len < nlh->nlmsg_len)
2959 /* Only requests are handled by the kernel */
2960 if (!(nlh->nlmsg_flags & NLM_F_REQUEST))
2963 /* Skip control messages */
2964 if (nlh->nlmsg_type < NLMSG_MIN_TYPE)
2972 if (nlh->nlmsg_flags & NLM_F_ACK || err)
2973 netlink_ack(skb, nlh, err);
2976 msglen = NLMSG_ALIGN(nlh->nlmsg_len);
2977 if (msglen > skb->len)
2979 skb_pull(skb, msglen);
2984 EXPORT_SYMBOL(netlink_rcv_skb);
2987 * nlmsg_notify - send a notification netlink message
2988 * @sk: netlink socket to use
2989 * @skb: notification message
2990 * @portid: destination netlink portid for reports or 0
2991 * @group: destination multicast group or 0
2992 * @report: 1 to report back, 0 to disable
2993 * @flags: allocation flags
2995 int nlmsg_notify(struct sock *sk, struct sk_buff *skb, u32 portid,
2996 unsigned int group, int report, gfp_t flags)
3001 int exclude_portid = 0;
3004 atomic_inc(&skb->users);
3005 exclude_portid = portid;
3008 /* errors reported via destination sk->sk_err, but propagate
3009 * delivery errors if NETLINK_BROADCAST_ERROR flag is set */
3010 err = nlmsg_multicast(sk, skb, exclude_portid, group, flags);
3016 err2 = nlmsg_unicast(sk, skb, portid);
3017 if (!err || err == -ESRCH)
3023 EXPORT_SYMBOL(nlmsg_notify);
3025 #ifdef CONFIG_PROC_FS
3026 struct nl_seq_iter {
3027 struct seq_net_private p;
3028 struct rhashtable_iter hti;
3032 static int netlink_walk_start(struct nl_seq_iter *iter)
3036 err = rhashtable_walk_init(&nl_table[iter->link].hash, &iter->hti);
3038 iter->link = MAX_LINKS;
3042 err = rhashtable_walk_start(&iter->hti);
3043 return err == -EAGAIN ? 0 : err;
3046 static void netlink_walk_stop(struct nl_seq_iter *iter)
3048 rhashtable_walk_stop(&iter->hti);
3049 rhashtable_walk_exit(&iter->hti);
3052 static void *__netlink_seq_next(struct seq_file *seq)
3054 struct nl_seq_iter *iter = seq->private;
3055 struct netlink_sock *nlk;
3061 nlk = rhashtable_walk_next(&iter->hti);
3064 if (PTR_ERR(nlk) == -EAGAIN)
3073 netlink_walk_stop(iter);
3074 if (++iter->link >= MAX_LINKS)
3077 err = netlink_walk_start(iter);
3079 return ERR_PTR(err);
3081 } while (sock_net(&nlk->sk) != seq_file_net(seq));
3086 static void *netlink_seq_start(struct seq_file *seq, loff_t *posp)
3088 struct nl_seq_iter *iter = seq->private;
3089 void *obj = SEQ_START_TOKEN;
3095 err = netlink_walk_start(iter);
3097 return ERR_PTR(err);
3099 for (pos = *posp; pos && obj && !IS_ERR(obj); pos--)
3100 obj = __netlink_seq_next(seq);
3105 static void *netlink_seq_next(struct seq_file *seq, void *v, loff_t *pos)
3108 return __netlink_seq_next(seq);
3111 static void netlink_seq_stop(struct seq_file *seq, void *v)
3113 struct nl_seq_iter *iter = seq->private;
3115 if (iter->link >= MAX_LINKS)
3118 netlink_walk_stop(iter);
3122 static int netlink_seq_show(struct seq_file *seq, void *v)
3124 if (v == SEQ_START_TOKEN) {
3126 "sk Eth Pid Groups "
3127 "Rmem Wmem Dump Locks Drops Inode\n");
3130 struct netlink_sock *nlk = nlk_sk(s);
3132 seq_printf(seq, "%pK %-3d %-6u %08x %-8d %-8d %d %-8d %-8d %-8lu\n",
3136 nlk->groups ? (u32)nlk->groups[0] : 0,
3137 sk_rmem_alloc_get(s),
3138 sk_wmem_alloc_get(s),
3140 atomic_read(&s->sk_refcnt),
3141 atomic_read(&s->sk_drops),
3149 static const struct seq_operations netlink_seq_ops = {
3150 .start = netlink_seq_start,
3151 .next = netlink_seq_next,
3152 .stop = netlink_seq_stop,
3153 .show = netlink_seq_show,
3157 static int netlink_seq_open(struct inode *inode, struct file *file)
3159 return seq_open_net(inode, file, &netlink_seq_ops,
3160 sizeof(struct nl_seq_iter));
3163 static const struct file_operations netlink_seq_fops = {
3164 .owner = THIS_MODULE,
3165 .open = netlink_seq_open,
3167 .llseek = seq_lseek,
3168 .release = seq_release_net,
3173 int netlink_register_notifier(struct notifier_block *nb)
3175 return atomic_notifier_chain_register(&netlink_chain, nb);
3177 EXPORT_SYMBOL(netlink_register_notifier);
3179 int netlink_unregister_notifier(struct notifier_block *nb)
3181 return atomic_notifier_chain_unregister(&netlink_chain, nb);
3183 EXPORT_SYMBOL(netlink_unregister_notifier);
3185 static const struct proto_ops netlink_ops = {
3186 .family = PF_NETLINK,
3187 .owner = THIS_MODULE,
3188 .release = netlink_release,
3189 .bind = netlink_bind,
3190 .connect = netlink_connect,
3191 .socketpair = sock_no_socketpair,
3192 .accept = sock_no_accept,
3193 .getname = netlink_getname,
3194 .poll = netlink_poll,
3195 .ioctl = sock_no_ioctl,
3196 .listen = sock_no_listen,
3197 .shutdown = sock_no_shutdown,
3198 .setsockopt = netlink_setsockopt,
3199 .getsockopt = netlink_getsockopt,
3200 .sendmsg = netlink_sendmsg,
3201 .recvmsg = netlink_recvmsg,
3202 .mmap = netlink_mmap,
3203 .sendpage = sock_no_sendpage,
3206 static const struct net_proto_family netlink_family_ops = {
3207 .family = PF_NETLINK,
3208 .create = netlink_create,
3209 .owner = THIS_MODULE, /* for consistency 8) */
3212 static int __net_init netlink_net_init(struct net *net)
3214 #ifdef CONFIG_PROC_FS
3215 if (!proc_create("netlink", 0, net->proc_net, &netlink_seq_fops))
3221 static void __net_exit netlink_net_exit(struct net *net)
3223 #ifdef CONFIG_PROC_FS
3224 remove_proc_entry("netlink", net->proc_net);
3228 static void __init netlink_add_usersock_entry(void)
3230 struct listeners *listeners;
3233 listeners = kzalloc(sizeof(*listeners) + NLGRPSZ(groups), GFP_KERNEL);
3235 panic("netlink_add_usersock_entry: Cannot allocate listeners\n");
3237 netlink_table_grab();
3239 nl_table[NETLINK_USERSOCK].groups = groups;
3240 rcu_assign_pointer(nl_table[NETLINK_USERSOCK].listeners, listeners);
3241 nl_table[NETLINK_USERSOCK].module = THIS_MODULE;
3242 nl_table[NETLINK_USERSOCK].registered = 1;
3243 nl_table[NETLINK_USERSOCK].flags = NL_CFG_F_NONROOT_SEND;
3245 netlink_table_ungrab();
3248 static struct pernet_operations __net_initdata netlink_net_ops = {
3249 .init = netlink_net_init,
3250 .exit = netlink_net_exit,
3253 static inline u32 netlink_hash(const void *data, u32 len, u32 seed)
3255 const struct netlink_sock *nlk = data;
3256 struct netlink_compare_arg arg;
3258 netlink_compare_arg_init(&arg, sock_net(&nlk->sk), nlk->portid);
3259 return jhash2((u32 *)&arg, netlink_compare_arg_len / sizeof(u32), seed);
3262 static const struct rhashtable_params netlink_rhashtable_params = {
3263 .head_offset = offsetof(struct netlink_sock, node),
3264 .key_len = netlink_compare_arg_len,
3265 .obj_hashfn = netlink_hash,
3266 .obj_cmpfn = netlink_compare,
3267 .automatic_shrinking = true,
3270 static int __init netlink_proto_init(void)
3273 int err = proto_register(&netlink_proto, 0);
3278 BUILD_BUG_ON(sizeof(struct netlink_skb_parms) > FIELD_SIZEOF(struct sk_buff, cb));
3280 nl_table = kcalloc(MAX_LINKS, sizeof(*nl_table), GFP_KERNEL);
3284 for (i = 0; i < MAX_LINKS; i++) {
3285 if (rhashtable_init(&nl_table[i].hash,
3286 &netlink_rhashtable_params) < 0) {
3288 rhashtable_destroy(&nl_table[i].hash);
3294 INIT_LIST_HEAD(&netlink_tap_all);
3296 netlink_add_usersock_entry();
3298 sock_register(&netlink_family_ops);
3299 register_pernet_subsys(&netlink_net_ops);
3300 /* The netlink device handler may be needed early. */
3305 panic("netlink_init: Cannot allocate nl_table\n");
3308 core_initcall(netlink_proto_init);