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 <asm/cacheflush.h>
63 #include <net/net_namespace.h>
66 #include <net/netlink.h>
68 #include "af_netlink.h"
72 unsigned long masks[0];
76 #define NETLINK_CONGESTED 0x0
79 #define NETLINK_KERNEL_SOCKET 0x1
80 #define NETLINK_RECV_PKTINFO 0x2
81 #define NETLINK_BROADCAST_SEND_ERROR 0x4
82 #define NETLINK_RECV_NO_ENOBUFS 0x8
84 static inline int netlink_is_kernel(struct sock *sk)
86 return nlk_sk(sk)->flags & NETLINK_KERNEL_SOCKET;
89 struct netlink_table *nl_table;
90 EXPORT_SYMBOL_GPL(nl_table);
92 static DECLARE_WAIT_QUEUE_HEAD(nl_table_wait);
94 static int netlink_dump(struct sock *sk);
95 static void netlink_skb_destructor(struct sk_buff *skb);
97 DEFINE_RWLOCK(nl_table_lock);
98 EXPORT_SYMBOL_GPL(nl_table_lock);
99 static atomic_t nl_table_users = ATOMIC_INIT(0);
101 #define nl_deref_protected(X) rcu_dereference_protected(X, lockdep_is_held(&nl_table_lock));
103 static ATOMIC_NOTIFIER_HEAD(netlink_chain);
105 static DEFINE_SPINLOCK(netlink_tap_lock);
106 static struct list_head netlink_tap_all __read_mostly;
108 static inline u32 netlink_group_mask(u32 group)
110 return group ? 1 << (group - 1) : 0;
113 static inline struct hlist_head *nl_portid_hashfn(struct nl_portid_hash *hash, u32 portid)
115 return &hash->table[jhash_1word(portid, hash->rnd) & hash->mask];
118 int netlink_add_tap(struct netlink_tap *nt)
120 if (unlikely(nt->dev->type != ARPHRD_NETLINK))
123 spin_lock(&netlink_tap_lock);
124 list_add_rcu(&nt->list, &netlink_tap_all);
125 spin_unlock(&netlink_tap_lock);
128 __module_get(nt->module);
132 EXPORT_SYMBOL_GPL(netlink_add_tap);
134 int __netlink_remove_tap(struct netlink_tap *nt)
137 struct netlink_tap *tmp;
139 spin_lock(&netlink_tap_lock);
141 list_for_each_entry(tmp, &netlink_tap_all, list) {
143 list_del_rcu(&nt->list);
149 pr_warn("__netlink_remove_tap: %p not found\n", nt);
151 spin_unlock(&netlink_tap_lock);
153 if (found && nt->module)
154 module_put(nt->module);
156 return found ? 0 : -ENODEV;
158 EXPORT_SYMBOL_GPL(__netlink_remove_tap);
160 int netlink_remove_tap(struct netlink_tap *nt)
164 ret = __netlink_remove_tap(nt);
169 EXPORT_SYMBOL_GPL(netlink_remove_tap);
171 static bool netlink_filter_tap(const struct sk_buff *skb)
173 struct sock *sk = skb->sk;
176 /* We take the more conservative approach and
177 * whitelist socket protocols that may pass.
179 switch (sk->sk_protocol) {
181 case NETLINK_USERSOCK:
182 case NETLINK_SOCK_DIAG:
185 case NETLINK_FIB_LOOKUP:
186 case NETLINK_NETFILTER:
187 case NETLINK_GENERIC:
195 static int __netlink_deliver_tap_skb(struct sk_buff *skb,
196 struct net_device *dev)
198 struct sk_buff *nskb;
199 struct sock *sk = skb->sk;
203 nskb = skb_clone(skb, GFP_ATOMIC);
206 nskb->protocol = htons((u16) sk->sk_protocol);
208 ret = dev_queue_xmit(nskb);
209 if (unlikely(ret > 0))
210 ret = net_xmit_errno(ret);
217 static void __netlink_deliver_tap(struct sk_buff *skb)
220 struct netlink_tap *tmp;
222 if (!netlink_filter_tap(skb))
225 list_for_each_entry_rcu(tmp, &netlink_tap_all, list) {
226 ret = __netlink_deliver_tap_skb(skb, tmp->dev);
232 static void netlink_deliver_tap(struct sk_buff *skb)
236 if (unlikely(!list_empty(&netlink_tap_all)))
237 __netlink_deliver_tap(skb);
242 static void netlink_overrun(struct sock *sk)
244 struct netlink_sock *nlk = nlk_sk(sk);
246 if (!(nlk->flags & NETLINK_RECV_NO_ENOBUFS)) {
247 if (!test_and_set_bit(NETLINK_CONGESTED, &nlk_sk(sk)->state)) {
248 sk->sk_err = ENOBUFS;
249 sk->sk_error_report(sk);
252 atomic_inc(&sk->sk_drops);
255 static void netlink_rcv_wake(struct sock *sk)
257 struct netlink_sock *nlk = nlk_sk(sk);
259 if (skb_queue_empty(&sk->sk_receive_queue))
260 clear_bit(NETLINK_CONGESTED, &nlk->state);
261 if (!test_bit(NETLINK_CONGESTED, &nlk->state))
262 wake_up_interruptible(&nlk->wait);
265 #ifdef CONFIG_NETLINK_MMAP
266 static bool netlink_skb_is_mmaped(const struct sk_buff *skb)
268 return NETLINK_CB(skb).flags & NETLINK_SKB_MMAPED;
271 static bool netlink_rx_is_mmaped(struct sock *sk)
273 return nlk_sk(sk)->rx_ring.pg_vec != NULL;
276 static bool netlink_tx_is_mmaped(struct sock *sk)
278 return nlk_sk(sk)->tx_ring.pg_vec != NULL;
281 static __pure struct page *pgvec_to_page(const void *addr)
283 if (is_vmalloc_addr(addr))
284 return vmalloc_to_page(addr);
286 return virt_to_page(addr);
289 static void free_pg_vec(void **pg_vec, unsigned int order, unsigned int len)
293 for (i = 0; i < len; i++) {
294 if (pg_vec[i] != NULL) {
295 if (is_vmalloc_addr(pg_vec[i]))
298 free_pages((unsigned long)pg_vec[i], order);
304 static void *alloc_one_pg_vec_page(unsigned long order)
307 gfp_t gfp_flags = GFP_KERNEL | __GFP_COMP | __GFP_ZERO |
308 __GFP_NOWARN | __GFP_NORETRY;
310 buffer = (void *)__get_free_pages(gfp_flags, order);
314 buffer = vzalloc((1 << order) * PAGE_SIZE);
318 gfp_flags &= ~__GFP_NORETRY;
319 return (void *)__get_free_pages(gfp_flags, order);
322 static void **alloc_pg_vec(struct netlink_sock *nlk,
323 struct nl_mmap_req *req, unsigned int order)
325 unsigned int block_nr = req->nm_block_nr;
329 pg_vec = kcalloc(block_nr, sizeof(void *), GFP_KERNEL);
333 for (i = 0; i < block_nr; i++) {
334 pg_vec[i] = alloc_one_pg_vec_page(order);
335 if (pg_vec[i] == NULL)
341 free_pg_vec(pg_vec, order, block_nr);
345 static int netlink_set_ring(struct sock *sk, struct nl_mmap_req *req,
346 bool closing, bool tx_ring)
348 struct netlink_sock *nlk = nlk_sk(sk);
349 struct netlink_ring *ring;
350 struct sk_buff_head *queue;
351 void **pg_vec = NULL;
352 unsigned int order = 0;
355 ring = tx_ring ? &nlk->tx_ring : &nlk->rx_ring;
356 queue = tx_ring ? &sk->sk_write_queue : &sk->sk_receive_queue;
359 if (atomic_read(&nlk->mapped))
361 if (atomic_read(&ring->pending))
365 if (req->nm_block_nr) {
366 if (ring->pg_vec != NULL)
369 if ((int)req->nm_block_size <= 0)
371 if (!IS_ALIGNED(req->nm_block_size, PAGE_SIZE))
373 if (req->nm_frame_size < NL_MMAP_HDRLEN)
375 if (!IS_ALIGNED(req->nm_frame_size, NL_MMAP_MSG_ALIGNMENT))
378 ring->frames_per_block = req->nm_block_size /
380 if (ring->frames_per_block == 0)
382 if (ring->frames_per_block * req->nm_block_nr !=
386 order = get_order(req->nm_block_size);
387 pg_vec = alloc_pg_vec(nlk, req, order);
391 if (req->nm_frame_nr)
396 mutex_lock(&nlk->pg_vec_lock);
397 if (closing || atomic_read(&nlk->mapped) == 0) {
399 spin_lock_bh(&queue->lock);
401 ring->frame_max = req->nm_frame_nr - 1;
403 ring->frame_size = req->nm_frame_size;
404 ring->pg_vec_pages = req->nm_block_size / PAGE_SIZE;
406 swap(ring->pg_vec_len, req->nm_block_nr);
407 swap(ring->pg_vec_order, order);
408 swap(ring->pg_vec, pg_vec);
410 __skb_queue_purge(queue);
411 spin_unlock_bh(&queue->lock);
413 WARN_ON(atomic_read(&nlk->mapped));
415 mutex_unlock(&nlk->pg_vec_lock);
418 free_pg_vec(pg_vec, order, req->nm_block_nr);
422 static void netlink_mm_open(struct vm_area_struct *vma)
424 struct file *file = vma->vm_file;
425 struct socket *sock = file->private_data;
426 struct sock *sk = sock->sk;
429 atomic_inc(&nlk_sk(sk)->mapped);
432 static void netlink_mm_close(struct vm_area_struct *vma)
434 struct file *file = vma->vm_file;
435 struct socket *sock = file->private_data;
436 struct sock *sk = sock->sk;
439 atomic_dec(&nlk_sk(sk)->mapped);
442 static const struct vm_operations_struct netlink_mmap_ops = {
443 .open = netlink_mm_open,
444 .close = netlink_mm_close,
447 static int netlink_mmap(struct file *file, struct socket *sock,
448 struct vm_area_struct *vma)
450 struct sock *sk = sock->sk;
451 struct netlink_sock *nlk = nlk_sk(sk);
452 struct netlink_ring *ring;
453 unsigned long start, size, expected;
460 mutex_lock(&nlk->pg_vec_lock);
463 for (ring = &nlk->rx_ring; ring <= &nlk->tx_ring; ring++) {
464 if (ring->pg_vec == NULL)
466 expected += ring->pg_vec_len * ring->pg_vec_pages * PAGE_SIZE;
472 size = vma->vm_end - vma->vm_start;
473 if (size != expected)
476 start = vma->vm_start;
477 for (ring = &nlk->rx_ring; ring <= &nlk->tx_ring; ring++) {
478 if (ring->pg_vec == NULL)
481 for (i = 0; i < ring->pg_vec_len; i++) {
483 void *kaddr = ring->pg_vec[i];
486 for (pg_num = 0; pg_num < ring->pg_vec_pages; pg_num++) {
487 page = pgvec_to_page(kaddr);
488 err = vm_insert_page(vma, start, page);
497 atomic_inc(&nlk->mapped);
498 vma->vm_ops = &netlink_mmap_ops;
501 mutex_unlock(&nlk->pg_vec_lock);
505 static void netlink_frame_flush_dcache(const struct nl_mmap_hdr *hdr)
507 #if ARCH_IMPLEMENTS_FLUSH_DCACHE_PAGE == 1
508 struct page *p_start, *p_end;
510 /* First page is flushed through netlink_{get,set}_status */
511 p_start = pgvec_to_page(hdr + PAGE_SIZE);
512 p_end = pgvec_to_page((void *)hdr + NL_MMAP_HDRLEN + hdr->nm_len - 1);
513 while (p_start <= p_end) {
514 flush_dcache_page(p_start);
520 static enum nl_mmap_status netlink_get_status(const struct nl_mmap_hdr *hdr)
523 flush_dcache_page(pgvec_to_page(hdr));
524 return hdr->nm_status;
527 static void netlink_set_status(struct nl_mmap_hdr *hdr,
528 enum nl_mmap_status status)
530 hdr->nm_status = status;
531 flush_dcache_page(pgvec_to_page(hdr));
535 static struct nl_mmap_hdr *
536 __netlink_lookup_frame(const struct netlink_ring *ring, unsigned int pos)
538 unsigned int pg_vec_pos, frame_off;
540 pg_vec_pos = pos / ring->frames_per_block;
541 frame_off = pos % ring->frames_per_block;
543 return ring->pg_vec[pg_vec_pos] + (frame_off * ring->frame_size);
546 static struct nl_mmap_hdr *
547 netlink_lookup_frame(const struct netlink_ring *ring, unsigned int pos,
548 enum nl_mmap_status status)
550 struct nl_mmap_hdr *hdr;
552 hdr = __netlink_lookup_frame(ring, pos);
553 if (netlink_get_status(hdr) != status)
559 static struct nl_mmap_hdr *
560 netlink_current_frame(const struct netlink_ring *ring,
561 enum nl_mmap_status status)
563 return netlink_lookup_frame(ring, ring->head, status);
566 static struct nl_mmap_hdr *
567 netlink_previous_frame(const struct netlink_ring *ring,
568 enum nl_mmap_status status)
572 prev = ring->head ? ring->head - 1 : ring->frame_max;
573 return netlink_lookup_frame(ring, prev, status);
576 static void netlink_increment_head(struct netlink_ring *ring)
578 ring->head = ring->head != ring->frame_max ? ring->head + 1 : 0;
581 static void netlink_forward_ring(struct netlink_ring *ring)
583 unsigned int head = ring->head, pos = head;
584 const struct nl_mmap_hdr *hdr;
587 hdr = __netlink_lookup_frame(ring, pos);
588 if (hdr->nm_status == NL_MMAP_STATUS_UNUSED)
590 if (hdr->nm_status != NL_MMAP_STATUS_SKIP)
592 netlink_increment_head(ring);
593 } while (ring->head != head);
596 static bool netlink_dump_space(struct netlink_sock *nlk)
598 struct netlink_ring *ring = &nlk->rx_ring;
599 struct nl_mmap_hdr *hdr;
602 hdr = netlink_current_frame(ring, NL_MMAP_STATUS_UNUSED);
606 n = ring->head + ring->frame_max / 2;
607 if (n > ring->frame_max)
608 n -= ring->frame_max;
610 hdr = __netlink_lookup_frame(ring, n);
612 return hdr->nm_status == NL_MMAP_STATUS_UNUSED;
615 static unsigned int netlink_poll(struct file *file, struct socket *sock,
618 struct sock *sk = sock->sk;
619 struct netlink_sock *nlk = nlk_sk(sk);
623 if (nlk->rx_ring.pg_vec != NULL) {
624 /* Memory mapped sockets don't call recvmsg(), so flow control
625 * for dumps is performed here. A dump is allowed to continue
626 * if at least half the ring is unused.
628 while (nlk->cb_running && netlink_dump_space(nlk)) {
629 err = netlink_dump(sk);
632 sk->sk_error_report(sk);
636 netlink_rcv_wake(sk);
639 mask = datagram_poll(file, sock, wait);
641 spin_lock_bh(&sk->sk_receive_queue.lock);
642 if (nlk->rx_ring.pg_vec) {
643 netlink_forward_ring(&nlk->rx_ring);
644 if (!netlink_previous_frame(&nlk->rx_ring, NL_MMAP_STATUS_UNUSED))
645 mask |= POLLIN | POLLRDNORM;
647 spin_unlock_bh(&sk->sk_receive_queue.lock);
649 spin_lock_bh(&sk->sk_write_queue.lock);
650 if (nlk->tx_ring.pg_vec) {
651 if (netlink_current_frame(&nlk->tx_ring, NL_MMAP_STATUS_UNUSED))
652 mask |= POLLOUT | POLLWRNORM;
654 spin_unlock_bh(&sk->sk_write_queue.lock);
659 static struct nl_mmap_hdr *netlink_mmap_hdr(struct sk_buff *skb)
661 return (struct nl_mmap_hdr *)(skb->head - NL_MMAP_HDRLEN);
664 static void netlink_ring_setup_skb(struct sk_buff *skb, struct sock *sk,
665 struct netlink_ring *ring,
666 struct nl_mmap_hdr *hdr)
671 size = ring->frame_size - NL_MMAP_HDRLEN;
672 data = (void *)hdr + NL_MMAP_HDRLEN;
676 skb_reset_tail_pointer(skb);
677 skb->end = skb->tail + size;
680 skb->destructor = netlink_skb_destructor;
681 NETLINK_CB(skb).flags |= NETLINK_SKB_MMAPED;
682 NETLINK_CB(skb).sk = sk;
685 static int netlink_mmap_sendmsg(struct sock *sk, struct msghdr *msg,
686 u32 dst_portid, u32 dst_group,
687 struct sock_iocb *siocb)
689 struct netlink_sock *nlk = nlk_sk(sk);
690 struct netlink_ring *ring;
691 struct nl_mmap_hdr *hdr;
695 int err = 0, len = 0;
697 /* Netlink messages are validated by the receiver before processing.
698 * In order to avoid userspace changing the contents of the message
699 * after validation, the socket and the ring may only be used by a
700 * single process, otherwise we fall back to copying.
702 if (atomic_long_read(&sk->sk_socket->file->f_count) > 2 ||
703 atomic_read(&nlk->mapped) > 1)
706 mutex_lock(&nlk->pg_vec_lock);
708 ring = &nlk->tx_ring;
709 maxlen = ring->frame_size - NL_MMAP_HDRLEN;
712 hdr = netlink_current_frame(ring, NL_MMAP_STATUS_VALID);
714 if (!(msg->msg_flags & MSG_DONTWAIT) &&
715 atomic_read(&nlk->tx_ring.pending))
719 if (hdr->nm_len > maxlen) {
724 netlink_frame_flush_dcache(hdr);
726 if (likely(dst_portid == 0 && dst_group == 0 && excl)) {
727 skb = alloc_skb_head(GFP_KERNEL);
733 netlink_ring_setup_skb(skb, sk, ring, hdr);
734 NETLINK_CB(skb).flags |= NETLINK_SKB_TX;
735 __skb_put(skb, hdr->nm_len);
736 netlink_set_status(hdr, NL_MMAP_STATUS_RESERVED);
737 atomic_inc(&ring->pending);
739 skb = alloc_skb(hdr->nm_len, GFP_KERNEL);
744 __skb_put(skb, hdr->nm_len);
745 memcpy(skb->data, (void *)hdr + NL_MMAP_HDRLEN, hdr->nm_len);
746 netlink_set_status(hdr, NL_MMAP_STATUS_UNUSED);
749 netlink_increment_head(ring);
751 NETLINK_CB(skb).portid = nlk->portid;
752 NETLINK_CB(skb).dst_group = dst_group;
753 NETLINK_CB(skb).creds = siocb->scm->creds;
755 err = security_netlink_send(sk, skb);
761 if (unlikely(dst_group)) {
762 atomic_inc(&skb->users);
763 netlink_broadcast(sk, skb, dst_portid, dst_group,
766 err = netlink_unicast(sk, skb, dst_portid,
767 msg->msg_flags & MSG_DONTWAIT);
772 } while (hdr != NULL ||
773 (!(msg->msg_flags & MSG_DONTWAIT) &&
774 atomic_read(&nlk->tx_ring.pending)));
779 mutex_unlock(&nlk->pg_vec_lock);
783 static void netlink_queue_mmaped_skb(struct sock *sk, struct sk_buff *skb)
785 struct nl_mmap_hdr *hdr;
787 hdr = netlink_mmap_hdr(skb);
788 hdr->nm_len = skb->len;
789 hdr->nm_group = NETLINK_CB(skb).dst_group;
790 hdr->nm_pid = NETLINK_CB(skb).creds.pid;
791 hdr->nm_uid = from_kuid(sk_user_ns(sk), NETLINK_CB(skb).creds.uid);
792 hdr->nm_gid = from_kgid(sk_user_ns(sk), NETLINK_CB(skb).creds.gid);
793 netlink_frame_flush_dcache(hdr);
794 netlink_set_status(hdr, NL_MMAP_STATUS_VALID);
796 NETLINK_CB(skb).flags |= NETLINK_SKB_DELIVERED;
800 static void netlink_ring_set_copied(struct sock *sk, struct sk_buff *skb)
802 struct netlink_sock *nlk = nlk_sk(sk);
803 struct netlink_ring *ring = &nlk->rx_ring;
804 struct nl_mmap_hdr *hdr;
806 spin_lock_bh(&sk->sk_receive_queue.lock);
807 hdr = netlink_current_frame(ring, NL_MMAP_STATUS_UNUSED);
809 spin_unlock_bh(&sk->sk_receive_queue.lock);
814 netlink_increment_head(ring);
815 __skb_queue_tail(&sk->sk_receive_queue, skb);
816 spin_unlock_bh(&sk->sk_receive_queue.lock);
818 hdr->nm_len = skb->len;
819 hdr->nm_group = NETLINK_CB(skb).dst_group;
820 hdr->nm_pid = NETLINK_CB(skb).creds.pid;
821 hdr->nm_uid = from_kuid(sk_user_ns(sk), NETLINK_CB(skb).creds.uid);
822 hdr->nm_gid = from_kgid(sk_user_ns(sk), NETLINK_CB(skb).creds.gid);
823 netlink_set_status(hdr, NL_MMAP_STATUS_COPY);
826 #else /* CONFIG_NETLINK_MMAP */
827 #define netlink_skb_is_mmaped(skb) false
828 #define netlink_rx_is_mmaped(sk) false
829 #define netlink_tx_is_mmaped(sk) false
830 #define netlink_mmap sock_no_mmap
831 #define netlink_poll datagram_poll
832 #define netlink_mmap_sendmsg(sk, msg, dst_portid, dst_group, siocb) 0
833 #endif /* CONFIG_NETLINK_MMAP */
835 static void netlink_skb_destructor(struct sk_buff *skb)
837 #ifdef CONFIG_NETLINK_MMAP
838 struct nl_mmap_hdr *hdr;
839 struct netlink_ring *ring;
842 /* If a packet from the kernel to userspace was freed because of an
843 * error without being delivered to userspace, the kernel must reset
844 * the status. In the direction userspace to kernel, the status is
845 * always reset here after the packet was processed and freed.
847 if (netlink_skb_is_mmaped(skb)) {
848 hdr = netlink_mmap_hdr(skb);
849 sk = NETLINK_CB(skb).sk;
851 if (NETLINK_CB(skb).flags & NETLINK_SKB_TX) {
852 netlink_set_status(hdr, NL_MMAP_STATUS_UNUSED);
853 ring = &nlk_sk(sk)->tx_ring;
855 if (!(NETLINK_CB(skb).flags & NETLINK_SKB_DELIVERED)) {
857 netlink_set_status(hdr, NL_MMAP_STATUS_VALID);
859 ring = &nlk_sk(sk)->rx_ring;
862 WARN_ON(atomic_read(&ring->pending) == 0);
863 atomic_dec(&ring->pending);
869 if (is_vmalloc_addr(skb->head)) {
871 !atomic_dec_return(&(skb_shinfo(skb)->dataref)))
880 static void netlink_skb_set_owner_r(struct sk_buff *skb, struct sock *sk)
882 WARN_ON(skb->sk != NULL);
884 skb->destructor = netlink_skb_destructor;
885 atomic_add(skb->truesize, &sk->sk_rmem_alloc);
886 sk_mem_charge(sk, skb->truesize);
889 static void netlink_sock_destruct(struct sock *sk)
891 struct netlink_sock *nlk = nlk_sk(sk);
893 if (nlk->cb_running) {
895 nlk->cb.done(&nlk->cb);
897 module_put(nlk->cb.module);
898 kfree_skb(nlk->cb.skb);
901 skb_queue_purge(&sk->sk_receive_queue);
902 #ifdef CONFIG_NETLINK_MMAP
904 struct nl_mmap_req req;
906 memset(&req, 0, sizeof(req));
907 if (nlk->rx_ring.pg_vec)
908 netlink_set_ring(sk, &req, true, false);
909 memset(&req, 0, sizeof(req));
910 if (nlk->tx_ring.pg_vec)
911 netlink_set_ring(sk, &req, true, true);
913 #endif /* CONFIG_NETLINK_MMAP */
915 if (!sock_flag(sk, SOCK_DEAD)) {
916 printk(KERN_ERR "Freeing alive netlink socket %p\n", sk);
920 WARN_ON(atomic_read(&sk->sk_rmem_alloc));
921 WARN_ON(atomic_read(&sk->sk_wmem_alloc));
922 WARN_ON(nlk_sk(sk)->groups);
925 /* This lock without WQ_FLAG_EXCLUSIVE is good on UP and it is _very_ bad on
926 * SMP. Look, when several writers sleep and reader wakes them up, all but one
927 * immediately hit write lock and grab all the cpus. Exclusive sleep solves
928 * this, _but_ remember, it adds useless work on UP machines.
931 void netlink_table_grab(void)
932 __acquires(nl_table_lock)
936 write_lock_irq(&nl_table_lock);
938 if (atomic_read(&nl_table_users)) {
939 DECLARE_WAITQUEUE(wait, current);
941 add_wait_queue_exclusive(&nl_table_wait, &wait);
943 set_current_state(TASK_UNINTERRUPTIBLE);
944 if (atomic_read(&nl_table_users) == 0)
946 write_unlock_irq(&nl_table_lock);
948 write_lock_irq(&nl_table_lock);
951 __set_current_state(TASK_RUNNING);
952 remove_wait_queue(&nl_table_wait, &wait);
956 void netlink_table_ungrab(void)
957 __releases(nl_table_lock)
959 write_unlock_irq(&nl_table_lock);
960 wake_up(&nl_table_wait);
964 netlink_lock_table(void)
966 /* read_lock() synchronizes us to netlink_table_grab */
968 read_lock(&nl_table_lock);
969 atomic_inc(&nl_table_users);
970 read_unlock(&nl_table_lock);
974 netlink_unlock_table(void)
976 if (atomic_dec_and_test(&nl_table_users))
977 wake_up(&nl_table_wait);
980 static bool netlink_compare(struct net *net, struct sock *sk)
982 return net_eq(sock_net(sk), net);
985 static struct sock *netlink_lookup(struct net *net, int protocol, u32 portid)
987 struct netlink_table *table = &nl_table[protocol];
988 struct nl_portid_hash *hash = &table->hash;
989 struct hlist_head *head;
992 read_lock(&nl_table_lock);
993 head = nl_portid_hashfn(hash, portid);
994 sk_for_each(sk, head) {
995 if (table->compare(net, sk) &&
996 (nlk_sk(sk)->portid == portid)) {
1003 read_unlock(&nl_table_lock);
1007 static struct hlist_head *nl_portid_hash_zalloc(size_t size)
1009 if (size <= PAGE_SIZE)
1010 return kzalloc(size, GFP_ATOMIC);
1012 return (struct hlist_head *)
1013 __get_free_pages(GFP_ATOMIC | __GFP_ZERO,
1017 static void nl_portid_hash_free(struct hlist_head *table, size_t size)
1019 if (size <= PAGE_SIZE)
1022 free_pages((unsigned long)table, get_order(size));
1025 static int nl_portid_hash_rehash(struct nl_portid_hash *hash, int grow)
1027 unsigned int omask, mask, shift;
1029 struct hlist_head *otable, *table;
1032 omask = mask = hash->mask;
1033 osize = size = (mask + 1) * sizeof(*table);
1034 shift = hash->shift;
1037 if (++shift > hash->max_shift)
1039 mask = mask * 2 + 1;
1043 table = nl_portid_hash_zalloc(size);
1047 otable = hash->table;
1048 hash->table = table;
1050 hash->shift = shift;
1051 get_random_bytes(&hash->rnd, sizeof(hash->rnd));
1053 for (i = 0; i <= omask; i++) {
1055 struct hlist_node *tmp;
1057 sk_for_each_safe(sk, tmp, &otable[i])
1058 __sk_add_node(sk, nl_portid_hashfn(hash, nlk_sk(sk)->portid));
1061 nl_portid_hash_free(otable, osize);
1062 hash->rehash_time = jiffies + 10 * 60 * HZ;
1066 static inline int nl_portid_hash_dilute(struct nl_portid_hash *hash, int len)
1068 int avg = hash->entries >> hash->shift;
1070 if (unlikely(avg > 1) && nl_portid_hash_rehash(hash, 1))
1073 if (unlikely(len > avg) && time_after(jiffies, hash->rehash_time)) {
1074 nl_portid_hash_rehash(hash, 0);
1081 static const struct proto_ops netlink_ops;
1084 netlink_update_listeners(struct sock *sk)
1086 struct netlink_table *tbl = &nl_table[sk->sk_protocol];
1089 struct listeners *listeners;
1091 listeners = nl_deref_protected(tbl->listeners);
1095 for (i = 0; i < NLGRPLONGS(tbl->groups); i++) {
1097 sk_for_each_bound(sk, &tbl->mc_list) {
1098 if (i < NLGRPLONGS(nlk_sk(sk)->ngroups))
1099 mask |= nlk_sk(sk)->groups[i];
1101 listeners->masks[i] = mask;
1103 /* this function is only called with the netlink table "grabbed", which
1104 * makes sure updates are visible before bind or setsockopt return. */
1107 static int netlink_insert(struct sock *sk, struct net *net, u32 portid)
1109 struct netlink_table *table = &nl_table[sk->sk_protocol];
1110 struct nl_portid_hash *hash = &table->hash;
1111 struct hlist_head *head;
1112 int err = -EADDRINUSE;
1116 netlink_table_grab();
1117 head = nl_portid_hashfn(hash, portid);
1119 sk_for_each(osk, head) {
1120 if (table->compare(net, osk) &&
1121 (nlk_sk(osk)->portid == portid))
1129 if (nlk_sk(sk)->portid)
1133 if (BITS_PER_LONG > 32 && unlikely(hash->entries >= UINT_MAX))
1136 if (len && nl_portid_hash_dilute(hash, len))
1137 head = nl_portid_hashfn(hash, portid);
1139 nlk_sk(sk)->portid = portid;
1140 sk_add_node(sk, head);
1144 netlink_table_ungrab();
1148 static void netlink_remove(struct sock *sk)
1150 netlink_table_grab();
1151 if (sk_del_node_init(sk))
1152 nl_table[sk->sk_protocol].hash.entries--;
1153 if (nlk_sk(sk)->subscriptions)
1154 __sk_del_bind_node(sk);
1155 netlink_table_ungrab();
1158 static struct proto netlink_proto = {
1160 .owner = THIS_MODULE,
1161 .obj_size = sizeof(struct netlink_sock),
1164 static int __netlink_create(struct net *net, struct socket *sock,
1165 struct mutex *cb_mutex, int protocol)
1168 struct netlink_sock *nlk;
1170 sock->ops = &netlink_ops;
1172 sk = sk_alloc(net, PF_NETLINK, GFP_KERNEL, &netlink_proto);
1176 sock_init_data(sock, sk);
1180 nlk->cb_mutex = cb_mutex;
1182 nlk->cb_mutex = &nlk->cb_def_mutex;
1183 mutex_init(nlk->cb_mutex);
1185 init_waitqueue_head(&nlk->wait);
1186 #ifdef CONFIG_NETLINK_MMAP
1187 mutex_init(&nlk->pg_vec_lock);
1190 sk->sk_destruct = netlink_sock_destruct;
1191 sk->sk_protocol = protocol;
1195 static int netlink_create(struct net *net, struct socket *sock, int protocol,
1198 struct module *module = NULL;
1199 struct mutex *cb_mutex;
1200 struct netlink_sock *nlk;
1201 void (*bind)(int group);
1204 sock->state = SS_UNCONNECTED;
1206 if (sock->type != SOCK_RAW && sock->type != SOCK_DGRAM)
1207 return -ESOCKTNOSUPPORT;
1209 if (protocol < 0 || protocol >= MAX_LINKS)
1210 return -EPROTONOSUPPORT;
1212 netlink_lock_table();
1213 #ifdef CONFIG_MODULES
1214 if (!nl_table[protocol].registered) {
1215 netlink_unlock_table();
1216 request_module("net-pf-%d-proto-%d", PF_NETLINK, protocol);
1217 netlink_lock_table();
1220 if (nl_table[protocol].registered &&
1221 try_module_get(nl_table[protocol].module))
1222 module = nl_table[protocol].module;
1224 err = -EPROTONOSUPPORT;
1225 cb_mutex = nl_table[protocol].cb_mutex;
1226 bind = nl_table[protocol].bind;
1227 netlink_unlock_table();
1232 err = __netlink_create(net, sock, cb_mutex, protocol);
1237 sock_prot_inuse_add(net, &netlink_proto, 1);
1240 nlk = nlk_sk(sock->sk);
1241 nlk->module = module;
1242 nlk->netlink_bind = bind;
1251 static int netlink_release(struct socket *sock)
1253 struct sock *sk = sock->sk;
1254 struct netlink_sock *nlk;
1264 * OK. Socket is unlinked, any packets that arrive now
1269 wake_up_interruptible_all(&nlk->wait);
1271 skb_queue_purge(&sk->sk_write_queue);
1274 struct netlink_notify n = {
1275 .net = sock_net(sk),
1276 .protocol = sk->sk_protocol,
1277 .portid = nlk->portid,
1279 atomic_notifier_call_chain(&netlink_chain,
1280 NETLINK_URELEASE, &n);
1283 module_put(nlk->module);
1285 netlink_table_grab();
1286 if (netlink_is_kernel(sk)) {
1287 BUG_ON(nl_table[sk->sk_protocol].registered == 0);
1288 if (--nl_table[sk->sk_protocol].registered == 0) {
1289 struct listeners *old;
1291 old = nl_deref_protected(nl_table[sk->sk_protocol].listeners);
1292 RCU_INIT_POINTER(nl_table[sk->sk_protocol].listeners, NULL);
1293 kfree_rcu(old, rcu);
1294 nl_table[sk->sk_protocol].module = NULL;
1295 nl_table[sk->sk_protocol].bind = NULL;
1296 nl_table[sk->sk_protocol].flags = 0;
1297 nl_table[sk->sk_protocol].registered = 0;
1299 } else if (nlk->subscriptions) {
1300 netlink_update_listeners(sk);
1302 netlink_table_ungrab();
1308 sock_prot_inuse_add(sock_net(sk), &netlink_proto, -1);
1314 static int netlink_autobind(struct socket *sock)
1316 struct sock *sk = sock->sk;
1317 struct net *net = sock_net(sk);
1318 struct netlink_table *table = &nl_table[sk->sk_protocol];
1319 struct nl_portid_hash *hash = &table->hash;
1320 struct hlist_head *head;
1322 s32 portid = task_tgid_vnr(current);
1324 static s32 rover = -4097;
1328 netlink_table_grab();
1329 head = nl_portid_hashfn(hash, portid);
1330 sk_for_each(osk, head) {
1331 if (!table->compare(net, osk))
1333 if (nlk_sk(osk)->portid == portid) {
1334 /* Bind collision, search negative portid values. */
1338 netlink_table_ungrab();
1342 netlink_table_ungrab();
1344 err = netlink_insert(sk, net, portid);
1345 if (err == -EADDRINUSE)
1348 /* If 2 threads race to autobind, that is fine. */
1355 static inline int netlink_capable(const struct socket *sock, unsigned int flag)
1357 return (nl_table[sock->sk->sk_protocol].flags & flag) ||
1358 ns_capable(sock_net(sock->sk)->user_ns, CAP_NET_ADMIN);
1362 netlink_update_subscriptions(struct sock *sk, unsigned int subscriptions)
1364 struct netlink_sock *nlk = nlk_sk(sk);
1366 if (nlk->subscriptions && !subscriptions)
1367 __sk_del_bind_node(sk);
1368 else if (!nlk->subscriptions && subscriptions)
1369 sk_add_bind_node(sk, &nl_table[sk->sk_protocol].mc_list);
1370 nlk->subscriptions = subscriptions;
1373 static int netlink_realloc_groups(struct sock *sk)
1375 struct netlink_sock *nlk = nlk_sk(sk);
1376 unsigned int groups;
1377 unsigned long *new_groups;
1380 netlink_table_grab();
1382 groups = nl_table[sk->sk_protocol].groups;
1383 if (!nl_table[sk->sk_protocol].registered) {
1388 if (nlk->ngroups >= groups)
1391 new_groups = krealloc(nlk->groups, NLGRPSZ(groups), GFP_ATOMIC);
1392 if (new_groups == NULL) {
1396 memset((char *)new_groups + NLGRPSZ(nlk->ngroups), 0,
1397 NLGRPSZ(groups) - NLGRPSZ(nlk->ngroups));
1399 nlk->groups = new_groups;
1400 nlk->ngroups = groups;
1402 netlink_table_ungrab();
1406 static int netlink_bind(struct socket *sock, struct sockaddr *addr,
1409 struct sock *sk = sock->sk;
1410 struct net *net = sock_net(sk);
1411 struct netlink_sock *nlk = nlk_sk(sk);
1412 struct sockaddr_nl *nladdr = (struct sockaddr_nl *)addr;
1415 if (addr_len < sizeof(struct sockaddr_nl))
1418 if (nladdr->nl_family != AF_NETLINK)
1421 /* Only superuser is allowed to listen multicasts */
1422 if (nladdr->nl_groups) {
1423 if (!netlink_capable(sock, NL_CFG_F_NONROOT_RECV))
1425 err = netlink_realloc_groups(sk);
1431 if (nladdr->nl_pid != nlk->portid)
1434 err = nladdr->nl_pid ?
1435 netlink_insert(sk, net, nladdr->nl_pid) :
1436 netlink_autobind(sock);
1441 if (!nladdr->nl_groups && (nlk->groups == NULL || !(u32)nlk->groups[0]))
1444 netlink_table_grab();
1445 netlink_update_subscriptions(sk, nlk->subscriptions +
1446 hweight32(nladdr->nl_groups) -
1447 hweight32(nlk->groups[0]));
1448 nlk->groups[0] = (nlk->groups[0] & ~0xffffffffUL) | nladdr->nl_groups;
1449 netlink_update_listeners(sk);
1450 netlink_table_ungrab();
1452 if (nlk->netlink_bind && nlk->groups[0]) {
1455 for (i=0; i<nlk->ngroups; i++) {
1456 if (test_bit(i, nlk->groups))
1457 nlk->netlink_bind(i);
1464 static int netlink_connect(struct socket *sock, struct sockaddr *addr,
1465 int alen, int flags)
1468 struct sock *sk = sock->sk;
1469 struct netlink_sock *nlk = nlk_sk(sk);
1470 struct sockaddr_nl *nladdr = (struct sockaddr_nl *)addr;
1472 if (alen < sizeof(addr->sa_family))
1475 if (addr->sa_family == AF_UNSPEC) {
1476 sk->sk_state = NETLINK_UNCONNECTED;
1477 nlk->dst_portid = 0;
1481 if (addr->sa_family != AF_NETLINK)
1484 /* Only superuser is allowed to send multicasts */
1485 if (nladdr->nl_groups && !netlink_capable(sock, NL_CFG_F_NONROOT_SEND))
1489 err = netlink_autobind(sock);
1492 sk->sk_state = NETLINK_CONNECTED;
1493 nlk->dst_portid = nladdr->nl_pid;
1494 nlk->dst_group = ffs(nladdr->nl_groups);
1500 static int netlink_getname(struct socket *sock, struct sockaddr *addr,
1501 int *addr_len, int peer)
1503 struct sock *sk = sock->sk;
1504 struct netlink_sock *nlk = nlk_sk(sk);
1505 DECLARE_SOCKADDR(struct sockaddr_nl *, nladdr, addr);
1507 nladdr->nl_family = AF_NETLINK;
1509 *addr_len = sizeof(*nladdr);
1512 nladdr->nl_pid = nlk->dst_portid;
1513 nladdr->nl_groups = netlink_group_mask(nlk->dst_group);
1515 nladdr->nl_pid = nlk->portid;
1516 nladdr->nl_groups = nlk->groups ? nlk->groups[0] : 0;
1521 static struct sock *netlink_getsockbyportid(struct sock *ssk, u32 portid)
1524 struct netlink_sock *nlk;
1526 sock = netlink_lookup(sock_net(ssk), ssk->sk_protocol, portid);
1528 return ERR_PTR(-ECONNREFUSED);
1530 /* Don't bother queuing skb if kernel socket has no input function */
1532 if (sock->sk_state == NETLINK_CONNECTED &&
1533 nlk->dst_portid != nlk_sk(ssk)->portid) {
1535 return ERR_PTR(-ECONNREFUSED);
1540 struct sock *netlink_getsockbyfilp(struct file *filp)
1542 struct inode *inode = file_inode(filp);
1545 if (!S_ISSOCK(inode->i_mode))
1546 return ERR_PTR(-ENOTSOCK);
1548 sock = SOCKET_I(inode)->sk;
1549 if (sock->sk_family != AF_NETLINK)
1550 return ERR_PTR(-EINVAL);
1556 static struct sk_buff *netlink_alloc_large_skb(unsigned int size,
1559 struct sk_buff *skb;
1562 if (size <= NLMSG_GOODSIZE || broadcast)
1563 return alloc_skb(size, GFP_KERNEL);
1565 size = SKB_DATA_ALIGN(size) +
1566 SKB_DATA_ALIGN(sizeof(struct skb_shared_info));
1568 data = vmalloc(size);
1572 skb = build_skb(data, size);
1577 skb->destructor = netlink_skb_destructor;
1584 * Attach a skb to a netlink socket.
1585 * The caller must hold a reference to the destination socket. On error, the
1586 * reference is dropped. The skb is not send to the destination, just all
1587 * all error checks are performed and memory in the queue is reserved.
1589 * < 0: error. skb freed, reference to sock dropped.
1591 * 1: repeat lookup - reference dropped while waiting for socket memory.
1593 int netlink_attachskb(struct sock *sk, struct sk_buff *skb,
1594 long *timeo, struct sock *ssk)
1596 struct netlink_sock *nlk;
1600 if ((atomic_read(&sk->sk_rmem_alloc) > sk->sk_rcvbuf ||
1601 test_bit(NETLINK_CONGESTED, &nlk->state)) &&
1602 !netlink_skb_is_mmaped(skb)) {
1603 DECLARE_WAITQUEUE(wait, current);
1605 if (!ssk || netlink_is_kernel(ssk))
1606 netlink_overrun(sk);
1612 __set_current_state(TASK_INTERRUPTIBLE);
1613 add_wait_queue(&nlk->wait, &wait);
1615 if ((atomic_read(&sk->sk_rmem_alloc) > sk->sk_rcvbuf ||
1616 test_bit(NETLINK_CONGESTED, &nlk->state)) &&
1617 !sock_flag(sk, SOCK_DEAD))
1618 *timeo = schedule_timeout(*timeo);
1620 __set_current_state(TASK_RUNNING);
1621 remove_wait_queue(&nlk->wait, &wait);
1624 if (signal_pending(current)) {
1626 return sock_intr_errno(*timeo);
1630 netlink_skb_set_owner_r(skb, sk);
1634 static int __netlink_sendskb(struct sock *sk, struct sk_buff *skb)
1638 netlink_deliver_tap(skb);
1640 #ifdef CONFIG_NETLINK_MMAP
1641 if (netlink_skb_is_mmaped(skb))
1642 netlink_queue_mmaped_skb(sk, skb);
1643 else if (netlink_rx_is_mmaped(sk))
1644 netlink_ring_set_copied(sk, skb);
1646 #endif /* CONFIG_NETLINK_MMAP */
1647 skb_queue_tail(&sk->sk_receive_queue, skb);
1648 sk->sk_data_ready(sk, len);
1652 int netlink_sendskb(struct sock *sk, struct sk_buff *skb)
1654 int len = __netlink_sendskb(sk, skb);
1660 void netlink_detachskb(struct sock *sk, struct sk_buff *skb)
1666 static struct sk_buff *netlink_trim(struct sk_buff *skb, gfp_t allocation)
1670 WARN_ON(skb->sk != NULL);
1671 if (netlink_skb_is_mmaped(skb))
1674 delta = skb->end - skb->tail;
1675 if (is_vmalloc_addr(skb->head) || delta * 2 < skb->truesize)
1678 if (skb_shared(skb)) {
1679 struct sk_buff *nskb = skb_clone(skb, allocation);
1686 if (!pskb_expand_head(skb, 0, -delta, allocation))
1687 skb->truesize -= delta;
1692 static int netlink_unicast_kernel(struct sock *sk, struct sk_buff *skb,
1696 struct netlink_sock *nlk = nlk_sk(sk);
1698 ret = -ECONNREFUSED;
1699 if (nlk->netlink_rcv != NULL) {
1700 /* We could do a netlink_deliver_tap(skb) here as well
1701 * but since this is intended for the kernel only, we
1702 * should rather let it stay under the hood.
1706 netlink_skb_set_owner_r(skb, sk);
1707 NETLINK_CB(skb).sk = ssk;
1708 nlk->netlink_rcv(skb);
1717 int netlink_unicast(struct sock *ssk, struct sk_buff *skb,
1718 u32 portid, int nonblock)
1724 skb = netlink_trim(skb, gfp_any());
1726 timeo = sock_sndtimeo(ssk, nonblock);
1728 sk = netlink_getsockbyportid(ssk, portid);
1733 if (netlink_is_kernel(sk))
1734 return netlink_unicast_kernel(sk, skb, ssk);
1736 if (sk_filter(sk, skb)) {
1743 err = netlink_attachskb(sk, skb, &timeo, ssk);
1749 return netlink_sendskb(sk, skb);
1751 EXPORT_SYMBOL(netlink_unicast);
1753 struct sk_buff *netlink_alloc_skb(struct sock *ssk, unsigned int size,
1754 u32 dst_portid, gfp_t gfp_mask)
1756 #ifdef CONFIG_NETLINK_MMAP
1757 struct sock *sk = NULL;
1758 struct sk_buff *skb;
1759 struct netlink_ring *ring;
1760 struct nl_mmap_hdr *hdr;
1761 unsigned int maxlen;
1763 sk = netlink_getsockbyportid(ssk, dst_portid);
1767 ring = &nlk_sk(sk)->rx_ring;
1768 /* fast-path without atomic ops for common case: non-mmaped receiver */
1769 if (ring->pg_vec == NULL)
1772 skb = alloc_skb_head(gfp_mask);
1776 spin_lock_bh(&sk->sk_receive_queue.lock);
1777 /* check again under lock */
1778 if (ring->pg_vec == NULL)
1781 maxlen = ring->frame_size - NL_MMAP_HDRLEN;
1785 netlink_forward_ring(ring);
1786 hdr = netlink_current_frame(ring, NL_MMAP_STATUS_UNUSED);
1789 netlink_ring_setup_skb(skb, sk, ring, hdr);
1790 netlink_set_status(hdr, NL_MMAP_STATUS_RESERVED);
1791 atomic_inc(&ring->pending);
1792 netlink_increment_head(ring);
1794 spin_unlock_bh(&sk->sk_receive_queue.lock);
1799 spin_unlock_bh(&sk->sk_receive_queue.lock);
1800 netlink_overrun(sk);
1807 spin_unlock_bh(&sk->sk_receive_queue.lock);
1812 return alloc_skb(size, gfp_mask);
1814 EXPORT_SYMBOL_GPL(netlink_alloc_skb);
1816 int netlink_has_listeners(struct sock *sk, unsigned int group)
1819 struct listeners *listeners;
1821 BUG_ON(!netlink_is_kernel(sk));
1824 listeners = rcu_dereference(nl_table[sk->sk_protocol].listeners);
1826 if (listeners && group - 1 < nl_table[sk->sk_protocol].groups)
1827 res = test_bit(group - 1, listeners->masks);
1833 EXPORT_SYMBOL_GPL(netlink_has_listeners);
1835 static int netlink_broadcast_deliver(struct sock *sk, struct sk_buff *skb)
1837 struct netlink_sock *nlk = nlk_sk(sk);
1839 if (atomic_read(&sk->sk_rmem_alloc) <= sk->sk_rcvbuf &&
1840 !test_bit(NETLINK_CONGESTED, &nlk->state)) {
1841 netlink_skb_set_owner_r(skb, sk);
1842 __netlink_sendskb(sk, skb);
1843 return atomic_read(&sk->sk_rmem_alloc) > (sk->sk_rcvbuf >> 1);
1848 struct netlink_broadcast_data {
1849 struct sock *exclude_sk;
1854 int delivery_failure;
1858 struct sk_buff *skb, *skb2;
1859 int (*tx_filter)(struct sock *dsk, struct sk_buff *skb, void *data);
1863 static int do_one_broadcast(struct sock *sk,
1864 struct netlink_broadcast_data *p)
1866 struct netlink_sock *nlk = nlk_sk(sk);
1869 if (p->exclude_sk == sk)
1872 if (nlk->portid == p->portid || p->group - 1 >= nlk->ngroups ||
1873 !test_bit(p->group - 1, nlk->groups))
1876 if (!net_eq(sock_net(sk), p->net))
1880 netlink_overrun(sk);
1885 if (p->skb2 == NULL) {
1886 if (skb_shared(p->skb)) {
1887 p->skb2 = skb_clone(p->skb, p->allocation);
1889 p->skb2 = skb_get(p->skb);
1891 * skb ownership may have been set when
1892 * delivered to a previous socket.
1894 skb_orphan(p->skb2);
1897 if (p->skb2 == NULL) {
1898 netlink_overrun(sk);
1899 /* Clone failed. Notify ALL listeners. */
1901 if (nlk->flags & NETLINK_BROADCAST_SEND_ERROR)
1902 p->delivery_failure = 1;
1903 } else if (p->tx_filter && p->tx_filter(sk, p->skb2, p->tx_data)) {
1906 } else if (sk_filter(sk, p->skb2)) {
1909 } else if ((val = netlink_broadcast_deliver(sk, p->skb2)) < 0) {
1910 netlink_overrun(sk);
1911 if (nlk->flags & NETLINK_BROADCAST_SEND_ERROR)
1912 p->delivery_failure = 1;
1914 p->congested |= val;
1924 int netlink_broadcast_filtered(struct sock *ssk, struct sk_buff *skb, u32 portid,
1925 u32 group, gfp_t allocation,
1926 int (*filter)(struct sock *dsk, struct sk_buff *skb, void *data),
1929 struct net *net = sock_net(ssk);
1930 struct netlink_broadcast_data info;
1933 skb = netlink_trim(skb, allocation);
1935 info.exclude_sk = ssk;
1937 info.portid = portid;
1940 info.delivery_failure = 0;
1943 info.allocation = allocation;
1946 info.tx_filter = filter;
1947 info.tx_data = filter_data;
1949 /* While we sleep in clone, do not allow to change socket list */
1951 netlink_lock_table();
1953 sk_for_each_bound(sk, &nl_table[ssk->sk_protocol].mc_list)
1954 do_one_broadcast(sk, &info);
1958 netlink_unlock_table();
1960 if (info.delivery_failure) {
1961 kfree_skb(info.skb2);
1964 consume_skb(info.skb2);
1966 if (info.delivered) {
1967 if (info.congested && (allocation & __GFP_WAIT))
1973 EXPORT_SYMBOL(netlink_broadcast_filtered);
1975 int netlink_broadcast(struct sock *ssk, struct sk_buff *skb, u32 portid,
1976 u32 group, gfp_t allocation)
1978 return netlink_broadcast_filtered(ssk, skb, portid, group, allocation,
1981 EXPORT_SYMBOL(netlink_broadcast);
1983 struct netlink_set_err_data {
1984 struct sock *exclude_sk;
1990 static int do_one_set_err(struct sock *sk, struct netlink_set_err_data *p)
1992 struct netlink_sock *nlk = nlk_sk(sk);
1995 if (sk == p->exclude_sk)
1998 if (!net_eq(sock_net(sk), sock_net(p->exclude_sk)))
2001 if (nlk->portid == p->portid || p->group - 1 >= nlk->ngroups ||
2002 !test_bit(p->group - 1, nlk->groups))
2005 if (p->code == ENOBUFS && nlk->flags & NETLINK_RECV_NO_ENOBUFS) {
2010 sk->sk_err = p->code;
2011 sk->sk_error_report(sk);
2017 * netlink_set_err - report error to broadcast listeners
2018 * @ssk: the kernel netlink socket, as returned by netlink_kernel_create()
2019 * @portid: the PORTID of a process that we want to skip (if any)
2020 * @group: the broadcast group that will notice the error
2021 * @code: error code, must be negative (as usual in kernelspace)
2023 * This function returns the number of broadcast listeners that have set the
2024 * NETLINK_RECV_NO_ENOBUFS socket option.
2026 int netlink_set_err(struct sock *ssk, u32 portid, u32 group, int code)
2028 struct netlink_set_err_data info;
2032 info.exclude_sk = ssk;
2033 info.portid = portid;
2035 /* sk->sk_err wants a positive error value */
2038 read_lock(&nl_table_lock);
2040 sk_for_each_bound(sk, &nl_table[ssk->sk_protocol].mc_list)
2041 ret += do_one_set_err(sk, &info);
2043 read_unlock(&nl_table_lock);
2046 EXPORT_SYMBOL(netlink_set_err);
2048 /* must be called with netlink table grabbed */
2049 static void netlink_update_socket_mc(struct netlink_sock *nlk,
2053 int old, new = !!is_new, subscriptions;
2055 old = test_bit(group - 1, nlk->groups);
2056 subscriptions = nlk->subscriptions - old + new;
2058 __set_bit(group - 1, nlk->groups);
2060 __clear_bit(group - 1, nlk->groups);
2061 netlink_update_subscriptions(&nlk->sk, subscriptions);
2062 netlink_update_listeners(&nlk->sk);
2065 static int netlink_setsockopt(struct socket *sock, int level, int optname,
2066 char __user *optval, unsigned int optlen)
2068 struct sock *sk = sock->sk;
2069 struct netlink_sock *nlk = nlk_sk(sk);
2070 unsigned int val = 0;
2073 if (level != SOL_NETLINK)
2074 return -ENOPROTOOPT;
2076 if (optname != NETLINK_RX_RING && optname != NETLINK_TX_RING &&
2077 optlen >= sizeof(int) &&
2078 get_user(val, (unsigned int __user *)optval))
2082 case NETLINK_PKTINFO:
2084 nlk->flags |= NETLINK_RECV_PKTINFO;
2086 nlk->flags &= ~NETLINK_RECV_PKTINFO;
2089 case NETLINK_ADD_MEMBERSHIP:
2090 case NETLINK_DROP_MEMBERSHIP: {
2091 if (!netlink_capable(sock, NL_CFG_F_NONROOT_RECV))
2093 err = netlink_realloc_groups(sk);
2096 if (!val || val - 1 >= nlk->ngroups)
2098 netlink_table_grab();
2099 netlink_update_socket_mc(nlk, val,
2100 optname == NETLINK_ADD_MEMBERSHIP);
2101 netlink_table_ungrab();
2103 if (nlk->netlink_bind)
2104 nlk->netlink_bind(val);
2109 case NETLINK_BROADCAST_ERROR:
2111 nlk->flags |= NETLINK_BROADCAST_SEND_ERROR;
2113 nlk->flags &= ~NETLINK_BROADCAST_SEND_ERROR;
2116 case NETLINK_NO_ENOBUFS:
2118 nlk->flags |= NETLINK_RECV_NO_ENOBUFS;
2119 clear_bit(NETLINK_CONGESTED, &nlk->state);
2120 wake_up_interruptible(&nlk->wait);
2122 nlk->flags &= ~NETLINK_RECV_NO_ENOBUFS;
2126 #ifdef CONFIG_NETLINK_MMAP
2127 case NETLINK_RX_RING:
2128 case NETLINK_TX_RING: {
2129 struct nl_mmap_req req;
2131 /* Rings might consume more memory than queue limits, require
2134 if (!capable(CAP_NET_ADMIN))
2136 if (optlen < sizeof(req))
2138 if (copy_from_user(&req, optval, sizeof(req)))
2140 err = netlink_set_ring(sk, &req, false,
2141 optname == NETLINK_TX_RING);
2144 #endif /* CONFIG_NETLINK_MMAP */
2151 static int netlink_getsockopt(struct socket *sock, int level, int optname,
2152 char __user *optval, int __user *optlen)
2154 struct sock *sk = sock->sk;
2155 struct netlink_sock *nlk = nlk_sk(sk);
2158 if (level != SOL_NETLINK)
2159 return -ENOPROTOOPT;
2161 if (get_user(len, optlen))
2167 case NETLINK_PKTINFO:
2168 if (len < sizeof(int))
2171 val = nlk->flags & NETLINK_RECV_PKTINFO ? 1 : 0;
2172 if (put_user(len, optlen) ||
2173 put_user(val, optval))
2177 case NETLINK_BROADCAST_ERROR:
2178 if (len < sizeof(int))
2181 val = nlk->flags & NETLINK_BROADCAST_SEND_ERROR ? 1 : 0;
2182 if (put_user(len, optlen) ||
2183 put_user(val, optval))
2187 case NETLINK_NO_ENOBUFS:
2188 if (len < sizeof(int))
2191 val = nlk->flags & NETLINK_RECV_NO_ENOBUFS ? 1 : 0;
2192 if (put_user(len, optlen) ||
2193 put_user(val, optval))
2203 static void netlink_cmsg_recv_pktinfo(struct msghdr *msg, struct sk_buff *skb)
2205 struct nl_pktinfo info;
2207 info.group = NETLINK_CB(skb).dst_group;
2208 put_cmsg(msg, SOL_NETLINK, NETLINK_PKTINFO, sizeof(info), &info);
2211 static int netlink_sendmsg(struct kiocb *kiocb, struct socket *sock,
2212 struct msghdr *msg, size_t len)
2214 struct sock_iocb *siocb = kiocb_to_siocb(kiocb);
2215 struct sock *sk = sock->sk;
2216 struct netlink_sock *nlk = nlk_sk(sk);
2217 struct sockaddr_nl *addr = msg->msg_name;
2220 struct sk_buff *skb;
2222 struct scm_cookie scm;
2224 if (msg->msg_flags&MSG_OOB)
2227 if (NULL == siocb->scm)
2230 err = scm_send(sock, msg, siocb->scm, true);
2234 if (msg->msg_namelen) {
2236 if (addr->nl_family != AF_NETLINK)
2238 dst_portid = addr->nl_pid;
2239 dst_group = ffs(addr->nl_groups);
2241 if ((dst_group || dst_portid) &&
2242 !netlink_capable(sock, NL_CFG_F_NONROOT_SEND))
2245 dst_portid = nlk->dst_portid;
2246 dst_group = nlk->dst_group;
2250 err = netlink_autobind(sock);
2255 if (netlink_tx_is_mmaped(sk) &&
2256 msg->msg_iov->iov_base == NULL) {
2257 err = netlink_mmap_sendmsg(sk, msg, dst_portid, dst_group,
2263 if (len > sk->sk_sndbuf - 32)
2266 skb = netlink_alloc_large_skb(len, dst_group);
2270 NETLINK_CB(skb).portid = nlk->portid;
2271 NETLINK_CB(skb).dst_group = dst_group;
2272 NETLINK_CB(skb).creds = siocb->scm->creds;
2275 if (memcpy_fromiovec(skb_put(skb, len), msg->msg_iov, len)) {
2280 err = security_netlink_send(sk, skb);
2287 atomic_inc(&skb->users);
2288 netlink_broadcast(sk, skb, dst_portid, dst_group, GFP_KERNEL);
2290 err = netlink_unicast(sk, skb, dst_portid, msg->msg_flags&MSG_DONTWAIT);
2293 scm_destroy(siocb->scm);
2297 static int netlink_recvmsg(struct kiocb *kiocb, struct socket *sock,
2298 struct msghdr *msg, size_t len,
2301 struct sock_iocb *siocb = kiocb_to_siocb(kiocb);
2302 struct scm_cookie scm;
2303 struct sock *sk = sock->sk;
2304 struct netlink_sock *nlk = nlk_sk(sk);
2305 int noblock = flags&MSG_DONTWAIT;
2307 struct sk_buff *skb, *data_skb;
2315 skb = skb_recv_datagram(sk, flags, noblock, &err);
2321 #ifdef CONFIG_COMPAT_NETLINK_MESSAGES
2322 if (unlikely(skb_shinfo(skb)->frag_list)) {
2324 * If this skb has a frag_list, then here that means that we
2325 * will have to use the frag_list skb's data for compat tasks
2326 * and the regular skb's data for normal (non-compat) tasks.
2328 * If we need to send the compat skb, assign it to the
2329 * 'data_skb' variable so that it will be used below for data
2330 * copying. We keep 'skb' for everything else, including
2331 * freeing both later.
2333 if (flags & MSG_CMSG_COMPAT)
2334 data_skb = skb_shinfo(skb)->frag_list;
2338 copied = data_skb->len;
2340 msg->msg_flags |= MSG_TRUNC;
2344 skb_reset_transport_header(data_skb);
2345 err = skb_copy_datagram_iovec(data_skb, 0, msg->msg_iov, copied);
2347 if (msg->msg_name) {
2348 struct sockaddr_nl *addr = (struct sockaddr_nl *)msg->msg_name;
2349 addr->nl_family = AF_NETLINK;
2351 addr->nl_pid = NETLINK_CB(skb).portid;
2352 addr->nl_groups = netlink_group_mask(NETLINK_CB(skb).dst_group);
2353 msg->msg_namelen = sizeof(*addr);
2356 if (nlk->flags & NETLINK_RECV_PKTINFO)
2357 netlink_cmsg_recv_pktinfo(msg, skb);
2359 if (NULL == siocb->scm) {
2360 memset(&scm, 0, sizeof(scm));
2363 siocb->scm->creds = *NETLINK_CREDS(skb);
2364 if (flags & MSG_TRUNC)
2365 copied = data_skb->len;
2367 skb_free_datagram(sk, skb);
2369 if (nlk->cb_running &&
2370 atomic_read(&sk->sk_rmem_alloc) <= sk->sk_rcvbuf / 2) {
2371 ret = netlink_dump(sk);
2374 sk->sk_error_report(sk);
2378 scm_recv(sock, msg, siocb->scm, flags);
2380 netlink_rcv_wake(sk);
2381 return err ? : copied;
2384 static void netlink_data_ready(struct sock *sk, int len)
2390 * We export these functions to other modules. They provide a
2391 * complete set of kernel non-blocking support for message
2396 __netlink_kernel_create(struct net *net, int unit, struct module *module,
2397 struct netlink_kernel_cfg *cfg)
2399 struct socket *sock;
2401 struct netlink_sock *nlk;
2402 struct listeners *listeners = NULL;
2403 struct mutex *cb_mutex = cfg ? cfg->cb_mutex : NULL;
2404 unsigned int groups;
2408 if (unit < 0 || unit >= MAX_LINKS)
2411 if (sock_create_lite(PF_NETLINK, SOCK_DGRAM, unit, &sock))
2415 * We have to just have a reference on the net from sk, but don't
2416 * get_net it. Besides, we cannot get and then put the net here.
2417 * So we create one inside init_net and the move it to net.
2420 if (__netlink_create(&init_net, sock, cb_mutex, unit) < 0)
2421 goto out_sock_release_nosk;
2424 sk_change_net(sk, net);
2426 if (!cfg || cfg->groups < 32)
2429 groups = cfg->groups;
2431 listeners = kzalloc(sizeof(*listeners) + NLGRPSZ(groups), GFP_KERNEL);
2433 goto out_sock_release;
2435 sk->sk_data_ready = netlink_data_ready;
2436 if (cfg && cfg->input)
2437 nlk_sk(sk)->netlink_rcv = cfg->input;
2439 if (netlink_insert(sk, net, 0))
2440 goto out_sock_release;
2443 nlk->flags |= NETLINK_KERNEL_SOCKET;
2445 netlink_table_grab();
2446 if (!nl_table[unit].registered) {
2447 nl_table[unit].groups = groups;
2448 rcu_assign_pointer(nl_table[unit].listeners, listeners);
2449 nl_table[unit].cb_mutex = cb_mutex;
2450 nl_table[unit].module = module;
2452 nl_table[unit].bind = cfg->bind;
2453 nl_table[unit].flags = cfg->flags;
2455 nl_table[unit].compare = cfg->compare;
2457 nl_table[unit].registered = 1;
2460 nl_table[unit].registered++;
2462 netlink_table_ungrab();
2467 netlink_kernel_release(sk);
2470 out_sock_release_nosk:
2474 EXPORT_SYMBOL(__netlink_kernel_create);
2477 netlink_kernel_release(struct sock *sk)
2479 sk_release_kernel(sk);
2481 EXPORT_SYMBOL(netlink_kernel_release);
2483 int __netlink_change_ngroups(struct sock *sk, unsigned int groups)
2485 struct listeners *new, *old;
2486 struct netlink_table *tbl = &nl_table[sk->sk_protocol];
2491 if (NLGRPSZ(tbl->groups) < NLGRPSZ(groups)) {
2492 new = kzalloc(sizeof(*new) + NLGRPSZ(groups), GFP_ATOMIC);
2495 old = nl_deref_protected(tbl->listeners);
2496 memcpy(new->masks, old->masks, NLGRPSZ(tbl->groups));
2497 rcu_assign_pointer(tbl->listeners, new);
2499 kfree_rcu(old, rcu);
2501 tbl->groups = groups;
2507 * netlink_change_ngroups - change number of multicast groups
2509 * This changes the number of multicast groups that are available
2510 * on a certain netlink family. Note that it is not possible to
2511 * change the number of groups to below 32. Also note that it does
2512 * not implicitly call netlink_clear_multicast_users() when the
2513 * number of groups is reduced.
2515 * @sk: The kernel netlink socket, as returned by netlink_kernel_create().
2516 * @groups: The new number of groups.
2518 int netlink_change_ngroups(struct sock *sk, unsigned int groups)
2522 netlink_table_grab();
2523 err = __netlink_change_ngroups(sk, groups);
2524 netlink_table_ungrab();
2529 void __netlink_clear_multicast_users(struct sock *ksk, unsigned int group)
2532 struct netlink_table *tbl = &nl_table[ksk->sk_protocol];
2534 sk_for_each_bound(sk, &tbl->mc_list)
2535 netlink_update_socket_mc(nlk_sk(sk), group, 0);
2539 * netlink_clear_multicast_users - kick off multicast listeners
2541 * This function removes all listeners from the given group.
2542 * @ksk: The kernel netlink socket, as returned by
2543 * netlink_kernel_create().
2544 * @group: The multicast group to clear.
2546 void netlink_clear_multicast_users(struct sock *ksk, unsigned int group)
2548 netlink_table_grab();
2549 __netlink_clear_multicast_users(ksk, group);
2550 netlink_table_ungrab();
2554 __nlmsg_put(struct sk_buff *skb, u32 portid, u32 seq, int type, int len, int flags)
2556 struct nlmsghdr *nlh;
2557 int size = nlmsg_msg_size(len);
2559 nlh = (struct nlmsghdr*)skb_put(skb, NLMSG_ALIGN(size));
2560 nlh->nlmsg_type = type;
2561 nlh->nlmsg_len = size;
2562 nlh->nlmsg_flags = flags;
2563 nlh->nlmsg_pid = portid;
2564 nlh->nlmsg_seq = seq;
2565 if (!__builtin_constant_p(size) || NLMSG_ALIGN(size) - size != 0)
2566 memset(nlmsg_data(nlh) + len, 0, NLMSG_ALIGN(size) - size);
2569 EXPORT_SYMBOL(__nlmsg_put);
2572 * It looks a bit ugly.
2573 * It would be better to create kernel thread.
2576 static int netlink_dump(struct sock *sk)
2578 struct netlink_sock *nlk = nlk_sk(sk);
2579 struct netlink_callback *cb;
2580 struct sk_buff *skb = NULL;
2581 struct nlmsghdr *nlh;
2582 int len, err = -ENOBUFS;
2585 mutex_lock(nlk->cb_mutex);
2586 if (!nlk->cb_running) {
2592 alloc_size = max_t(int, cb->min_dump_alloc, NLMSG_GOODSIZE);
2594 if (!netlink_rx_is_mmaped(sk) &&
2595 atomic_read(&sk->sk_rmem_alloc) >= sk->sk_rcvbuf)
2597 skb = netlink_alloc_skb(sk, alloc_size, nlk->portid, GFP_KERNEL);
2600 netlink_skb_set_owner_r(skb, sk);
2602 len = cb->dump(skb, cb);
2605 mutex_unlock(nlk->cb_mutex);
2607 if (sk_filter(sk, skb))
2610 __netlink_sendskb(sk, skb);
2614 nlh = nlmsg_put_answer(skb, cb, NLMSG_DONE, sizeof(len), NLM_F_MULTI);
2618 nl_dump_check_consistent(cb, nlh);
2620 memcpy(nlmsg_data(nlh), &len, sizeof(len));
2622 if (sk_filter(sk, skb))
2625 __netlink_sendskb(sk, skb);
2630 nlk->cb_running = false;
2631 mutex_unlock(nlk->cb_mutex);
2632 module_put(cb->module);
2633 consume_skb(cb->skb);
2637 mutex_unlock(nlk->cb_mutex);
2642 int __netlink_dump_start(struct sock *ssk, struct sk_buff *skb,
2643 const struct nlmsghdr *nlh,
2644 struct netlink_dump_control *control)
2646 struct netlink_callback *cb;
2648 struct netlink_sock *nlk;
2651 /* Memory mapped dump requests need to be copied to avoid looping
2652 * on the pending state in netlink_mmap_sendmsg() while the CB hold
2653 * a reference to the skb.
2655 if (netlink_skb_is_mmaped(skb)) {
2656 skb = skb_copy(skb, GFP_KERNEL);
2660 atomic_inc(&skb->users);
2662 sk = netlink_lookup(sock_net(ssk), ssk->sk_protocol, NETLINK_CB(skb).portid);
2664 ret = -ECONNREFUSED;
2669 mutex_lock(nlk->cb_mutex);
2670 /* A dump is in progress... */
2671 if (nlk->cb_running) {
2675 /* add reference of module which cb->dump belongs to */
2676 if (!try_module_get(control->module)) {
2677 ret = -EPROTONOSUPPORT;
2682 memset(cb, 0, sizeof(*cb));
2683 cb->dump = control->dump;
2684 cb->done = control->done;
2686 cb->data = control->data;
2687 cb->module = control->module;
2688 cb->min_dump_alloc = control->min_dump_alloc;
2691 nlk->cb_running = true;
2693 mutex_unlock(nlk->cb_mutex);
2695 ret = netlink_dump(sk);
2701 /* We successfully started a dump, by returning -EINTR we
2702 * signal not to send ACK even if it was requested.
2708 mutex_unlock(nlk->cb_mutex);
2713 EXPORT_SYMBOL(__netlink_dump_start);
2715 void netlink_ack(struct sk_buff *in_skb, struct nlmsghdr *nlh, int err)
2717 struct sk_buff *skb;
2718 struct nlmsghdr *rep;
2719 struct nlmsgerr *errmsg;
2720 size_t payload = sizeof(*errmsg);
2722 /* error messages get the original request appened */
2724 payload += nlmsg_len(nlh);
2726 skb = netlink_alloc_skb(in_skb->sk, nlmsg_total_size(payload),
2727 NETLINK_CB(in_skb).portid, GFP_KERNEL);
2731 sk = netlink_lookup(sock_net(in_skb->sk),
2732 in_skb->sk->sk_protocol,
2733 NETLINK_CB(in_skb).portid);
2735 sk->sk_err = ENOBUFS;
2736 sk->sk_error_report(sk);
2742 rep = __nlmsg_put(skb, NETLINK_CB(in_skb).portid, nlh->nlmsg_seq,
2743 NLMSG_ERROR, payload, 0);
2744 errmsg = nlmsg_data(rep);
2745 errmsg->error = err;
2746 memcpy(&errmsg->msg, nlh, err ? nlh->nlmsg_len : sizeof(*nlh));
2747 netlink_unicast(in_skb->sk, skb, NETLINK_CB(in_skb).portid, MSG_DONTWAIT);
2749 EXPORT_SYMBOL(netlink_ack);
2751 int netlink_rcv_skb(struct sk_buff *skb, int (*cb)(struct sk_buff *,
2754 struct nlmsghdr *nlh;
2757 while (skb->len >= nlmsg_total_size(0)) {
2760 nlh = nlmsg_hdr(skb);
2763 if (nlh->nlmsg_len < NLMSG_HDRLEN || skb->len < nlh->nlmsg_len)
2766 /* Only requests are handled by the kernel */
2767 if (!(nlh->nlmsg_flags & NLM_F_REQUEST))
2770 /* Skip control messages */
2771 if (nlh->nlmsg_type < NLMSG_MIN_TYPE)
2779 if (nlh->nlmsg_flags & NLM_F_ACK || err)
2780 netlink_ack(skb, nlh, err);
2783 msglen = NLMSG_ALIGN(nlh->nlmsg_len);
2784 if (msglen > skb->len)
2786 skb_pull(skb, msglen);
2791 EXPORT_SYMBOL(netlink_rcv_skb);
2794 * nlmsg_notify - send a notification netlink message
2795 * @sk: netlink socket to use
2796 * @skb: notification message
2797 * @portid: destination netlink portid for reports or 0
2798 * @group: destination multicast group or 0
2799 * @report: 1 to report back, 0 to disable
2800 * @flags: allocation flags
2802 int nlmsg_notify(struct sock *sk, struct sk_buff *skb, u32 portid,
2803 unsigned int group, int report, gfp_t flags)
2808 int exclude_portid = 0;
2811 atomic_inc(&skb->users);
2812 exclude_portid = portid;
2815 /* errors reported via destination sk->sk_err, but propagate
2816 * delivery errors if NETLINK_BROADCAST_ERROR flag is set */
2817 err = nlmsg_multicast(sk, skb, exclude_portid, group, flags);
2823 err2 = nlmsg_unicast(sk, skb, portid);
2824 if (!err || err == -ESRCH)
2830 EXPORT_SYMBOL(nlmsg_notify);
2832 #ifdef CONFIG_PROC_FS
2833 struct nl_seq_iter {
2834 struct seq_net_private p;
2839 static struct sock *netlink_seq_socket_idx(struct seq_file *seq, loff_t pos)
2841 struct nl_seq_iter *iter = seq->private;
2846 for (i = 0; i < MAX_LINKS; i++) {
2847 struct nl_portid_hash *hash = &nl_table[i].hash;
2849 for (j = 0; j <= hash->mask; j++) {
2850 sk_for_each(s, &hash->table[j]) {
2851 if (sock_net(s) != seq_file_net(seq))
2865 static void *netlink_seq_start(struct seq_file *seq, loff_t *pos)
2866 __acquires(nl_table_lock)
2868 read_lock(&nl_table_lock);
2869 return *pos ? netlink_seq_socket_idx(seq, *pos - 1) : SEQ_START_TOKEN;
2872 static void *netlink_seq_next(struct seq_file *seq, void *v, loff_t *pos)
2875 struct nl_seq_iter *iter;
2881 if (v == SEQ_START_TOKEN)
2882 return netlink_seq_socket_idx(seq, 0);
2884 net = seq_file_net(seq);
2885 iter = seq->private;
2889 } while (s && !nl_table[s->sk_protocol].compare(net, s));
2894 j = iter->hash_idx + 1;
2897 struct nl_portid_hash *hash = &nl_table[i].hash;
2899 for (; j <= hash->mask; j++) {
2900 s = sk_head(&hash->table[j]);
2902 while (s && !nl_table[s->sk_protocol].compare(net, s))
2912 } while (++i < MAX_LINKS);
2917 static void netlink_seq_stop(struct seq_file *seq, void *v)
2918 __releases(nl_table_lock)
2920 read_unlock(&nl_table_lock);
2924 static int netlink_seq_show(struct seq_file *seq, void *v)
2926 if (v == SEQ_START_TOKEN) {
2928 "sk Eth Pid Groups "
2929 "Rmem Wmem Dump Locks Drops Inode\n");
2932 struct netlink_sock *nlk = nlk_sk(s);
2934 seq_printf(seq, "%pK %-3d %-6u %08x %-8d %-8d %d %-8d %-8d %-8lu\n",
2938 nlk->groups ? (u32)nlk->groups[0] : 0,
2939 sk_rmem_alloc_get(s),
2940 sk_wmem_alloc_get(s),
2942 atomic_read(&s->sk_refcnt),
2943 atomic_read(&s->sk_drops),
2951 static const struct seq_operations netlink_seq_ops = {
2952 .start = netlink_seq_start,
2953 .next = netlink_seq_next,
2954 .stop = netlink_seq_stop,
2955 .show = netlink_seq_show,
2959 static int netlink_seq_open(struct inode *inode, struct file *file)
2961 return seq_open_net(inode, file, &netlink_seq_ops,
2962 sizeof(struct nl_seq_iter));
2965 static const struct file_operations netlink_seq_fops = {
2966 .owner = THIS_MODULE,
2967 .open = netlink_seq_open,
2969 .llseek = seq_lseek,
2970 .release = seq_release_net,
2975 int netlink_register_notifier(struct notifier_block *nb)
2977 return atomic_notifier_chain_register(&netlink_chain, nb);
2979 EXPORT_SYMBOL(netlink_register_notifier);
2981 int netlink_unregister_notifier(struct notifier_block *nb)
2983 return atomic_notifier_chain_unregister(&netlink_chain, nb);
2985 EXPORT_SYMBOL(netlink_unregister_notifier);
2987 static const struct proto_ops netlink_ops = {
2988 .family = PF_NETLINK,
2989 .owner = THIS_MODULE,
2990 .release = netlink_release,
2991 .bind = netlink_bind,
2992 .connect = netlink_connect,
2993 .socketpair = sock_no_socketpair,
2994 .accept = sock_no_accept,
2995 .getname = netlink_getname,
2996 .poll = netlink_poll,
2997 .ioctl = sock_no_ioctl,
2998 .listen = sock_no_listen,
2999 .shutdown = sock_no_shutdown,
3000 .setsockopt = netlink_setsockopt,
3001 .getsockopt = netlink_getsockopt,
3002 .sendmsg = netlink_sendmsg,
3003 .recvmsg = netlink_recvmsg,
3004 .mmap = netlink_mmap,
3005 .sendpage = sock_no_sendpage,
3008 static const struct net_proto_family netlink_family_ops = {
3009 .family = PF_NETLINK,
3010 .create = netlink_create,
3011 .owner = THIS_MODULE, /* for consistency 8) */
3014 static int __net_init netlink_net_init(struct net *net)
3016 #ifdef CONFIG_PROC_FS
3017 if (!proc_create("netlink", 0, net->proc_net, &netlink_seq_fops))
3023 static void __net_exit netlink_net_exit(struct net *net)
3025 #ifdef CONFIG_PROC_FS
3026 remove_proc_entry("netlink", net->proc_net);
3030 static void __init netlink_add_usersock_entry(void)
3032 struct listeners *listeners;
3035 listeners = kzalloc(sizeof(*listeners) + NLGRPSZ(groups), GFP_KERNEL);
3037 panic("netlink_add_usersock_entry: Cannot allocate listeners\n");
3039 netlink_table_grab();
3041 nl_table[NETLINK_USERSOCK].groups = groups;
3042 rcu_assign_pointer(nl_table[NETLINK_USERSOCK].listeners, listeners);
3043 nl_table[NETLINK_USERSOCK].module = THIS_MODULE;
3044 nl_table[NETLINK_USERSOCK].registered = 1;
3045 nl_table[NETLINK_USERSOCK].flags = NL_CFG_F_NONROOT_SEND;
3047 netlink_table_ungrab();
3050 static struct pernet_operations __net_initdata netlink_net_ops = {
3051 .init = netlink_net_init,
3052 .exit = netlink_net_exit,
3055 static int __init netlink_proto_init(void)
3058 unsigned long limit;
3060 int err = proto_register(&netlink_proto, 0);
3065 BUILD_BUG_ON(sizeof(struct netlink_skb_parms) > FIELD_SIZEOF(struct sk_buff, cb));
3067 nl_table = kcalloc(MAX_LINKS, sizeof(*nl_table), GFP_KERNEL);
3071 if (totalram_pages >= (128 * 1024))
3072 limit = totalram_pages >> (21 - PAGE_SHIFT);
3074 limit = totalram_pages >> (23 - PAGE_SHIFT);
3076 order = get_bitmask_order(limit) - 1 + PAGE_SHIFT;
3077 limit = (1UL << order) / sizeof(struct hlist_head);
3078 order = get_bitmask_order(min(limit, (unsigned long)UINT_MAX)) - 1;
3080 for (i = 0; i < MAX_LINKS; i++) {
3081 struct nl_portid_hash *hash = &nl_table[i].hash;
3083 hash->table = nl_portid_hash_zalloc(1 * sizeof(*hash->table));
3086 nl_portid_hash_free(nl_table[i].hash.table,
3087 1 * sizeof(*hash->table));
3091 hash->max_shift = order;
3094 hash->rehash_time = jiffies;
3096 nl_table[i].compare = netlink_compare;
3099 INIT_LIST_HEAD(&netlink_tap_all);
3101 netlink_add_usersock_entry();
3103 sock_register(&netlink_family_ops);
3104 register_pernet_subsys(&netlink_net_ops);
3105 /* The netlink device handler may be needed early. */
3110 panic("netlink_init: Cannot allocate nl_table\n");
3113 core_initcall(netlink_proto_init);