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 int __netlink_deliver_tap_skb(struct sk_buff *skb,
172 struct net_device *dev)
174 struct sk_buff *nskb;
178 nskb = skb_clone(skb, GFP_ATOMIC);
181 ret = dev_queue_xmit(nskb);
182 if (unlikely(ret > 0))
183 ret = net_xmit_errno(ret);
190 static void __netlink_deliver_tap(struct sk_buff *skb)
193 struct netlink_tap *tmp;
195 list_for_each_entry_rcu(tmp, &netlink_tap_all, list) {
196 ret = __netlink_deliver_tap_skb(skb, tmp->dev);
202 static void netlink_deliver_tap(struct sk_buff *skb)
206 if (unlikely(!list_empty(&netlink_tap_all)))
207 __netlink_deliver_tap(skb);
212 static void netlink_overrun(struct sock *sk)
214 struct netlink_sock *nlk = nlk_sk(sk);
216 if (!(nlk->flags & NETLINK_RECV_NO_ENOBUFS)) {
217 if (!test_and_set_bit(NETLINK_CONGESTED, &nlk_sk(sk)->state)) {
218 sk->sk_err = ENOBUFS;
219 sk->sk_error_report(sk);
222 atomic_inc(&sk->sk_drops);
225 static void netlink_rcv_wake(struct sock *sk)
227 struct netlink_sock *nlk = nlk_sk(sk);
229 if (skb_queue_empty(&sk->sk_receive_queue))
230 clear_bit(NETLINK_CONGESTED, &nlk->state);
231 if (!test_bit(NETLINK_CONGESTED, &nlk->state))
232 wake_up_interruptible(&nlk->wait);
235 #ifdef CONFIG_NETLINK_MMAP
236 static bool netlink_skb_is_mmaped(const struct sk_buff *skb)
238 return NETLINK_CB(skb).flags & NETLINK_SKB_MMAPED;
241 static bool netlink_rx_is_mmaped(struct sock *sk)
243 return nlk_sk(sk)->rx_ring.pg_vec != NULL;
246 static bool netlink_tx_is_mmaped(struct sock *sk)
248 return nlk_sk(sk)->tx_ring.pg_vec != NULL;
251 static __pure struct page *pgvec_to_page(const void *addr)
253 if (is_vmalloc_addr(addr))
254 return vmalloc_to_page(addr);
256 return virt_to_page(addr);
259 static void free_pg_vec(void **pg_vec, unsigned int order, unsigned int len)
263 for (i = 0; i < len; i++) {
264 if (pg_vec[i] != NULL) {
265 if (is_vmalloc_addr(pg_vec[i]))
268 free_pages((unsigned long)pg_vec[i], order);
274 static void *alloc_one_pg_vec_page(unsigned long order)
277 gfp_t gfp_flags = GFP_KERNEL | __GFP_COMP | __GFP_ZERO |
278 __GFP_NOWARN | __GFP_NORETRY;
280 buffer = (void *)__get_free_pages(gfp_flags, order);
284 buffer = vzalloc((1 << order) * PAGE_SIZE);
288 gfp_flags &= ~__GFP_NORETRY;
289 return (void *)__get_free_pages(gfp_flags, order);
292 static void **alloc_pg_vec(struct netlink_sock *nlk,
293 struct nl_mmap_req *req, unsigned int order)
295 unsigned int block_nr = req->nm_block_nr;
299 pg_vec = kcalloc(block_nr, sizeof(void *), GFP_KERNEL);
303 for (i = 0; i < block_nr; i++) {
304 pg_vec[i] = alloc_one_pg_vec_page(order);
305 if (pg_vec[i] == NULL)
311 free_pg_vec(pg_vec, order, block_nr);
315 static int netlink_set_ring(struct sock *sk, struct nl_mmap_req *req,
316 bool closing, bool tx_ring)
318 struct netlink_sock *nlk = nlk_sk(sk);
319 struct netlink_ring *ring;
320 struct sk_buff_head *queue;
321 void **pg_vec = NULL;
322 unsigned int order = 0;
325 ring = tx_ring ? &nlk->tx_ring : &nlk->rx_ring;
326 queue = tx_ring ? &sk->sk_write_queue : &sk->sk_receive_queue;
329 if (atomic_read(&nlk->mapped))
331 if (atomic_read(&ring->pending))
335 if (req->nm_block_nr) {
336 if (ring->pg_vec != NULL)
339 if ((int)req->nm_block_size <= 0)
341 if (!IS_ALIGNED(req->nm_block_size, PAGE_SIZE))
343 if (req->nm_frame_size < NL_MMAP_HDRLEN)
345 if (!IS_ALIGNED(req->nm_frame_size, NL_MMAP_MSG_ALIGNMENT))
348 ring->frames_per_block = req->nm_block_size /
350 if (ring->frames_per_block == 0)
352 if (ring->frames_per_block * req->nm_block_nr !=
356 order = get_order(req->nm_block_size);
357 pg_vec = alloc_pg_vec(nlk, req, order);
361 if (req->nm_frame_nr)
366 mutex_lock(&nlk->pg_vec_lock);
367 if (closing || atomic_read(&nlk->mapped) == 0) {
369 spin_lock_bh(&queue->lock);
371 ring->frame_max = req->nm_frame_nr - 1;
373 ring->frame_size = req->nm_frame_size;
374 ring->pg_vec_pages = req->nm_block_size / PAGE_SIZE;
376 swap(ring->pg_vec_len, req->nm_block_nr);
377 swap(ring->pg_vec_order, order);
378 swap(ring->pg_vec, pg_vec);
380 __skb_queue_purge(queue);
381 spin_unlock_bh(&queue->lock);
383 WARN_ON(atomic_read(&nlk->mapped));
385 mutex_unlock(&nlk->pg_vec_lock);
388 free_pg_vec(pg_vec, order, req->nm_block_nr);
392 static void netlink_mm_open(struct vm_area_struct *vma)
394 struct file *file = vma->vm_file;
395 struct socket *sock = file->private_data;
396 struct sock *sk = sock->sk;
399 atomic_inc(&nlk_sk(sk)->mapped);
402 static void netlink_mm_close(struct vm_area_struct *vma)
404 struct file *file = vma->vm_file;
405 struct socket *sock = file->private_data;
406 struct sock *sk = sock->sk;
409 atomic_dec(&nlk_sk(sk)->mapped);
412 static const struct vm_operations_struct netlink_mmap_ops = {
413 .open = netlink_mm_open,
414 .close = netlink_mm_close,
417 static int netlink_mmap(struct file *file, struct socket *sock,
418 struct vm_area_struct *vma)
420 struct sock *sk = sock->sk;
421 struct netlink_sock *nlk = nlk_sk(sk);
422 struct netlink_ring *ring;
423 unsigned long start, size, expected;
430 mutex_lock(&nlk->pg_vec_lock);
433 for (ring = &nlk->rx_ring; ring <= &nlk->tx_ring; ring++) {
434 if (ring->pg_vec == NULL)
436 expected += ring->pg_vec_len * ring->pg_vec_pages * PAGE_SIZE;
442 size = vma->vm_end - vma->vm_start;
443 if (size != expected)
446 start = vma->vm_start;
447 for (ring = &nlk->rx_ring; ring <= &nlk->tx_ring; ring++) {
448 if (ring->pg_vec == NULL)
451 for (i = 0; i < ring->pg_vec_len; i++) {
453 void *kaddr = ring->pg_vec[i];
456 for (pg_num = 0; pg_num < ring->pg_vec_pages; pg_num++) {
457 page = pgvec_to_page(kaddr);
458 err = vm_insert_page(vma, start, page);
467 atomic_inc(&nlk->mapped);
468 vma->vm_ops = &netlink_mmap_ops;
471 mutex_unlock(&nlk->pg_vec_lock);
475 static void netlink_frame_flush_dcache(const struct nl_mmap_hdr *hdr)
477 #if ARCH_IMPLEMENTS_FLUSH_DCACHE_PAGE == 1
478 struct page *p_start, *p_end;
480 /* First page is flushed through netlink_{get,set}_status */
481 p_start = pgvec_to_page(hdr + PAGE_SIZE);
482 p_end = pgvec_to_page((void *)hdr + NL_MMAP_HDRLEN + hdr->nm_len - 1);
483 while (p_start <= p_end) {
484 flush_dcache_page(p_start);
490 static enum nl_mmap_status netlink_get_status(const struct nl_mmap_hdr *hdr)
493 flush_dcache_page(pgvec_to_page(hdr));
494 return hdr->nm_status;
497 static void netlink_set_status(struct nl_mmap_hdr *hdr,
498 enum nl_mmap_status status)
500 hdr->nm_status = status;
501 flush_dcache_page(pgvec_to_page(hdr));
505 static struct nl_mmap_hdr *
506 __netlink_lookup_frame(const struct netlink_ring *ring, unsigned int pos)
508 unsigned int pg_vec_pos, frame_off;
510 pg_vec_pos = pos / ring->frames_per_block;
511 frame_off = pos % ring->frames_per_block;
513 return ring->pg_vec[pg_vec_pos] + (frame_off * ring->frame_size);
516 static struct nl_mmap_hdr *
517 netlink_lookup_frame(const struct netlink_ring *ring, unsigned int pos,
518 enum nl_mmap_status status)
520 struct nl_mmap_hdr *hdr;
522 hdr = __netlink_lookup_frame(ring, pos);
523 if (netlink_get_status(hdr) != status)
529 static struct nl_mmap_hdr *
530 netlink_current_frame(const struct netlink_ring *ring,
531 enum nl_mmap_status status)
533 return netlink_lookup_frame(ring, ring->head, status);
536 static struct nl_mmap_hdr *
537 netlink_previous_frame(const struct netlink_ring *ring,
538 enum nl_mmap_status status)
542 prev = ring->head ? ring->head - 1 : ring->frame_max;
543 return netlink_lookup_frame(ring, prev, status);
546 static void netlink_increment_head(struct netlink_ring *ring)
548 ring->head = ring->head != ring->frame_max ? ring->head + 1 : 0;
551 static void netlink_forward_ring(struct netlink_ring *ring)
553 unsigned int head = ring->head, pos = head;
554 const struct nl_mmap_hdr *hdr;
557 hdr = __netlink_lookup_frame(ring, pos);
558 if (hdr->nm_status == NL_MMAP_STATUS_UNUSED)
560 if (hdr->nm_status != NL_MMAP_STATUS_SKIP)
562 netlink_increment_head(ring);
563 } while (ring->head != head);
566 static bool netlink_dump_space(struct netlink_sock *nlk)
568 struct netlink_ring *ring = &nlk->rx_ring;
569 struct nl_mmap_hdr *hdr;
572 hdr = netlink_current_frame(ring, NL_MMAP_STATUS_UNUSED);
576 n = ring->head + ring->frame_max / 2;
577 if (n > ring->frame_max)
578 n -= ring->frame_max;
580 hdr = __netlink_lookup_frame(ring, n);
582 return hdr->nm_status == NL_MMAP_STATUS_UNUSED;
585 static unsigned int netlink_poll(struct file *file, struct socket *sock,
588 struct sock *sk = sock->sk;
589 struct netlink_sock *nlk = nlk_sk(sk);
593 if (nlk->rx_ring.pg_vec != NULL) {
594 /* Memory mapped sockets don't call recvmsg(), so flow control
595 * for dumps is performed here. A dump is allowed to continue
596 * if at least half the ring is unused.
598 while (nlk->cb_running && netlink_dump_space(nlk)) {
599 err = netlink_dump(sk);
602 sk->sk_error_report(sk);
606 netlink_rcv_wake(sk);
609 mask = datagram_poll(file, sock, wait);
611 spin_lock_bh(&sk->sk_receive_queue.lock);
612 if (nlk->rx_ring.pg_vec) {
613 netlink_forward_ring(&nlk->rx_ring);
614 if (!netlink_previous_frame(&nlk->rx_ring, NL_MMAP_STATUS_UNUSED))
615 mask |= POLLIN | POLLRDNORM;
617 spin_unlock_bh(&sk->sk_receive_queue.lock);
619 spin_lock_bh(&sk->sk_write_queue.lock);
620 if (nlk->tx_ring.pg_vec) {
621 if (netlink_current_frame(&nlk->tx_ring, NL_MMAP_STATUS_UNUSED))
622 mask |= POLLOUT | POLLWRNORM;
624 spin_unlock_bh(&sk->sk_write_queue.lock);
629 static struct nl_mmap_hdr *netlink_mmap_hdr(struct sk_buff *skb)
631 return (struct nl_mmap_hdr *)(skb->head - NL_MMAP_HDRLEN);
634 static void netlink_ring_setup_skb(struct sk_buff *skb, struct sock *sk,
635 struct netlink_ring *ring,
636 struct nl_mmap_hdr *hdr)
641 size = ring->frame_size - NL_MMAP_HDRLEN;
642 data = (void *)hdr + NL_MMAP_HDRLEN;
646 skb_reset_tail_pointer(skb);
647 skb->end = skb->tail + size;
650 skb->destructor = netlink_skb_destructor;
651 NETLINK_CB(skb).flags |= NETLINK_SKB_MMAPED;
652 NETLINK_CB(skb).sk = sk;
655 static int netlink_mmap_sendmsg(struct sock *sk, struct msghdr *msg,
656 u32 dst_portid, u32 dst_group,
657 struct sock_iocb *siocb)
659 struct netlink_sock *nlk = nlk_sk(sk);
660 struct netlink_ring *ring;
661 struct nl_mmap_hdr *hdr;
665 int err = 0, len = 0;
667 /* Netlink messages are validated by the receiver before processing.
668 * In order to avoid userspace changing the contents of the message
669 * after validation, the socket and the ring may only be used by a
670 * single process, otherwise we fall back to copying.
672 if (atomic_long_read(&sk->sk_socket->file->f_count) > 2 ||
673 atomic_read(&nlk->mapped) > 1)
676 mutex_lock(&nlk->pg_vec_lock);
678 ring = &nlk->tx_ring;
679 maxlen = ring->frame_size - NL_MMAP_HDRLEN;
682 hdr = netlink_current_frame(ring, NL_MMAP_STATUS_VALID);
684 if (!(msg->msg_flags & MSG_DONTWAIT) &&
685 atomic_read(&nlk->tx_ring.pending))
689 if (hdr->nm_len > maxlen) {
694 netlink_frame_flush_dcache(hdr);
696 if (likely(dst_portid == 0 && dst_group == 0 && excl)) {
697 skb = alloc_skb_head(GFP_KERNEL);
703 netlink_ring_setup_skb(skb, sk, ring, hdr);
704 NETLINK_CB(skb).flags |= NETLINK_SKB_TX;
705 __skb_put(skb, hdr->nm_len);
706 netlink_set_status(hdr, NL_MMAP_STATUS_RESERVED);
707 atomic_inc(&ring->pending);
709 skb = alloc_skb(hdr->nm_len, GFP_KERNEL);
714 __skb_put(skb, hdr->nm_len);
715 memcpy(skb->data, (void *)hdr + NL_MMAP_HDRLEN, hdr->nm_len);
716 netlink_set_status(hdr, NL_MMAP_STATUS_UNUSED);
719 netlink_increment_head(ring);
721 NETLINK_CB(skb).portid = nlk->portid;
722 NETLINK_CB(skb).dst_group = dst_group;
723 NETLINK_CB(skb).creds = siocb->scm->creds;
725 err = security_netlink_send(sk, skb);
731 if (unlikely(dst_group)) {
732 atomic_inc(&skb->users);
733 netlink_broadcast(sk, skb, dst_portid, dst_group,
736 err = netlink_unicast(sk, skb, dst_portid,
737 msg->msg_flags & MSG_DONTWAIT);
742 } while (hdr != NULL ||
743 (!(msg->msg_flags & MSG_DONTWAIT) &&
744 atomic_read(&nlk->tx_ring.pending)));
749 mutex_unlock(&nlk->pg_vec_lock);
753 static void netlink_queue_mmaped_skb(struct sock *sk, struct sk_buff *skb)
755 struct nl_mmap_hdr *hdr;
757 hdr = netlink_mmap_hdr(skb);
758 hdr->nm_len = skb->len;
759 hdr->nm_group = NETLINK_CB(skb).dst_group;
760 hdr->nm_pid = NETLINK_CB(skb).creds.pid;
761 hdr->nm_uid = from_kuid(sk_user_ns(sk), NETLINK_CB(skb).creds.uid);
762 hdr->nm_gid = from_kgid(sk_user_ns(sk), NETLINK_CB(skb).creds.gid);
763 netlink_frame_flush_dcache(hdr);
764 netlink_set_status(hdr, NL_MMAP_STATUS_VALID);
766 NETLINK_CB(skb).flags |= NETLINK_SKB_DELIVERED;
770 static void netlink_ring_set_copied(struct sock *sk, struct sk_buff *skb)
772 struct netlink_sock *nlk = nlk_sk(sk);
773 struct netlink_ring *ring = &nlk->rx_ring;
774 struct nl_mmap_hdr *hdr;
776 spin_lock_bh(&sk->sk_receive_queue.lock);
777 hdr = netlink_current_frame(ring, NL_MMAP_STATUS_UNUSED);
779 spin_unlock_bh(&sk->sk_receive_queue.lock);
784 netlink_increment_head(ring);
785 __skb_queue_tail(&sk->sk_receive_queue, skb);
786 spin_unlock_bh(&sk->sk_receive_queue.lock);
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_set_status(hdr, NL_MMAP_STATUS_COPY);
796 #else /* CONFIG_NETLINK_MMAP */
797 #define netlink_skb_is_mmaped(skb) false
798 #define netlink_rx_is_mmaped(sk) false
799 #define netlink_tx_is_mmaped(sk) false
800 #define netlink_mmap sock_no_mmap
801 #define netlink_poll datagram_poll
802 #define netlink_mmap_sendmsg(sk, msg, dst_portid, dst_group, siocb) 0
803 #endif /* CONFIG_NETLINK_MMAP */
805 static void netlink_skb_destructor(struct sk_buff *skb)
807 #ifdef CONFIG_NETLINK_MMAP
808 struct nl_mmap_hdr *hdr;
809 struct netlink_ring *ring;
812 /* If a packet from the kernel to userspace was freed because of an
813 * error without being delivered to userspace, the kernel must reset
814 * the status. In the direction userspace to kernel, the status is
815 * always reset here after the packet was processed and freed.
817 if (netlink_skb_is_mmaped(skb)) {
818 hdr = netlink_mmap_hdr(skb);
819 sk = NETLINK_CB(skb).sk;
821 if (NETLINK_CB(skb).flags & NETLINK_SKB_TX) {
822 netlink_set_status(hdr, NL_MMAP_STATUS_UNUSED);
823 ring = &nlk_sk(sk)->tx_ring;
825 if (!(NETLINK_CB(skb).flags & NETLINK_SKB_DELIVERED)) {
827 netlink_set_status(hdr, NL_MMAP_STATUS_VALID);
829 ring = &nlk_sk(sk)->rx_ring;
832 WARN_ON(atomic_read(&ring->pending) == 0);
833 atomic_dec(&ring->pending);
839 if (is_vmalloc_addr(skb->head)) {
841 !atomic_dec_return(&(skb_shinfo(skb)->dataref)))
850 static void netlink_skb_set_owner_r(struct sk_buff *skb, struct sock *sk)
852 WARN_ON(skb->sk != NULL);
854 skb->destructor = netlink_skb_destructor;
855 atomic_add(skb->truesize, &sk->sk_rmem_alloc);
856 sk_mem_charge(sk, skb->truesize);
859 static void netlink_sock_destruct(struct sock *sk)
861 struct netlink_sock *nlk = nlk_sk(sk);
863 if (nlk->cb_running) {
865 nlk->cb.done(&nlk->cb);
867 module_put(nlk->cb.module);
868 kfree_skb(nlk->cb.skb);
871 skb_queue_purge(&sk->sk_receive_queue);
872 #ifdef CONFIG_NETLINK_MMAP
874 struct nl_mmap_req req;
876 memset(&req, 0, sizeof(req));
877 if (nlk->rx_ring.pg_vec)
878 netlink_set_ring(sk, &req, true, false);
879 memset(&req, 0, sizeof(req));
880 if (nlk->tx_ring.pg_vec)
881 netlink_set_ring(sk, &req, true, true);
883 #endif /* CONFIG_NETLINK_MMAP */
885 if (!sock_flag(sk, SOCK_DEAD)) {
886 printk(KERN_ERR "Freeing alive netlink socket %p\n", sk);
890 WARN_ON(atomic_read(&sk->sk_rmem_alloc));
891 WARN_ON(atomic_read(&sk->sk_wmem_alloc));
892 WARN_ON(nlk_sk(sk)->groups);
895 /* This lock without WQ_FLAG_EXCLUSIVE is good on UP and it is _very_ bad on
896 * SMP. Look, when several writers sleep and reader wakes them up, all but one
897 * immediately hit write lock and grab all the cpus. Exclusive sleep solves
898 * this, _but_ remember, it adds useless work on UP machines.
901 void netlink_table_grab(void)
902 __acquires(nl_table_lock)
906 write_lock_irq(&nl_table_lock);
908 if (atomic_read(&nl_table_users)) {
909 DECLARE_WAITQUEUE(wait, current);
911 add_wait_queue_exclusive(&nl_table_wait, &wait);
913 set_current_state(TASK_UNINTERRUPTIBLE);
914 if (atomic_read(&nl_table_users) == 0)
916 write_unlock_irq(&nl_table_lock);
918 write_lock_irq(&nl_table_lock);
921 __set_current_state(TASK_RUNNING);
922 remove_wait_queue(&nl_table_wait, &wait);
926 void netlink_table_ungrab(void)
927 __releases(nl_table_lock)
929 write_unlock_irq(&nl_table_lock);
930 wake_up(&nl_table_wait);
934 netlink_lock_table(void)
936 /* read_lock() synchronizes us to netlink_table_grab */
938 read_lock(&nl_table_lock);
939 atomic_inc(&nl_table_users);
940 read_unlock(&nl_table_lock);
944 netlink_unlock_table(void)
946 if (atomic_dec_and_test(&nl_table_users))
947 wake_up(&nl_table_wait);
950 static bool netlink_compare(struct net *net, struct sock *sk)
952 return net_eq(sock_net(sk), net);
955 static struct sock *netlink_lookup(struct net *net, int protocol, u32 portid)
957 struct netlink_table *table = &nl_table[protocol];
958 struct nl_portid_hash *hash = &table->hash;
959 struct hlist_head *head;
962 read_lock(&nl_table_lock);
963 head = nl_portid_hashfn(hash, portid);
964 sk_for_each(sk, head) {
965 if (table->compare(net, sk) &&
966 (nlk_sk(sk)->portid == portid)) {
973 read_unlock(&nl_table_lock);
977 static struct hlist_head *nl_portid_hash_zalloc(size_t size)
979 if (size <= PAGE_SIZE)
980 return kzalloc(size, GFP_ATOMIC);
982 return (struct hlist_head *)
983 __get_free_pages(GFP_ATOMIC | __GFP_ZERO,
987 static void nl_portid_hash_free(struct hlist_head *table, size_t size)
989 if (size <= PAGE_SIZE)
992 free_pages((unsigned long)table, get_order(size));
995 static int nl_portid_hash_rehash(struct nl_portid_hash *hash, int grow)
997 unsigned int omask, mask, shift;
999 struct hlist_head *otable, *table;
1002 omask = mask = hash->mask;
1003 osize = size = (mask + 1) * sizeof(*table);
1004 shift = hash->shift;
1007 if (++shift > hash->max_shift)
1009 mask = mask * 2 + 1;
1013 table = nl_portid_hash_zalloc(size);
1017 otable = hash->table;
1018 hash->table = table;
1020 hash->shift = shift;
1021 get_random_bytes(&hash->rnd, sizeof(hash->rnd));
1023 for (i = 0; i <= omask; i++) {
1025 struct hlist_node *tmp;
1027 sk_for_each_safe(sk, tmp, &otable[i])
1028 __sk_add_node(sk, nl_portid_hashfn(hash, nlk_sk(sk)->portid));
1031 nl_portid_hash_free(otable, osize);
1032 hash->rehash_time = jiffies + 10 * 60 * HZ;
1036 static inline int nl_portid_hash_dilute(struct nl_portid_hash *hash, int len)
1038 int avg = hash->entries >> hash->shift;
1040 if (unlikely(avg > 1) && nl_portid_hash_rehash(hash, 1))
1043 if (unlikely(len > avg) && time_after(jiffies, hash->rehash_time)) {
1044 nl_portid_hash_rehash(hash, 0);
1051 static const struct proto_ops netlink_ops;
1054 netlink_update_listeners(struct sock *sk)
1056 struct netlink_table *tbl = &nl_table[sk->sk_protocol];
1059 struct listeners *listeners;
1061 listeners = nl_deref_protected(tbl->listeners);
1065 for (i = 0; i < NLGRPLONGS(tbl->groups); i++) {
1067 sk_for_each_bound(sk, &tbl->mc_list) {
1068 if (i < NLGRPLONGS(nlk_sk(sk)->ngroups))
1069 mask |= nlk_sk(sk)->groups[i];
1071 listeners->masks[i] = mask;
1073 /* this function is only called with the netlink table "grabbed", which
1074 * makes sure updates are visible before bind or setsockopt return. */
1077 static int netlink_insert(struct sock *sk, struct net *net, u32 portid)
1079 struct netlink_table *table = &nl_table[sk->sk_protocol];
1080 struct nl_portid_hash *hash = &table->hash;
1081 struct hlist_head *head;
1082 int err = -EADDRINUSE;
1086 netlink_table_grab();
1087 head = nl_portid_hashfn(hash, portid);
1089 sk_for_each(osk, head) {
1090 if (table->compare(net, osk) &&
1091 (nlk_sk(osk)->portid == portid))
1099 if (nlk_sk(sk)->portid)
1103 if (BITS_PER_LONG > 32 && unlikely(hash->entries >= UINT_MAX))
1106 if (len && nl_portid_hash_dilute(hash, len))
1107 head = nl_portid_hashfn(hash, portid);
1109 nlk_sk(sk)->portid = portid;
1110 sk_add_node(sk, head);
1114 netlink_table_ungrab();
1118 static void netlink_remove(struct sock *sk)
1120 netlink_table_grab();
1121 if (sk_del_node_init(sk))
1122 nl_table[sk->sk_protocol].hash.entries--;
1123 if (nlk_sk(sk)->subscriptions)
1124 __sk_del_bind_node(sk);
1125 netlink_table_ungrab();
1128 static struct proto netlink_proto = {
1130 .owner = THIS_MODULE,
1131 .obj_size = sizeof(struct netlink_sock),
1134 static int __netlink_create(struct net *net, struct socket *sock,
1135 struct mutex *cb_mutex, int protocol)
1138 struct netlink_sock *nlk;
1140 sock->ops = &netlink_ops;
1142 sk = sk_alloc(net, PF_NETLINK, GFP_KERNEL, &netlink_proto);
1146 sock_init_data(sock, sk);
1150 nlk->cb_mutex = cb_mutex;
1152 nlk->cb_mutex = &nlk->cb_def_mutex;
1153 mutex_init(nlk->cb_mutex);
1155 init_waitqueue_head(&nlk->wait);
1156 #ifdef CONFIG_NETLINK_MMAP
1157 mutex_init(&nlk->pg_vec_lock);
1160 sk->sk_destruct = netlink_sock_destruct;
1161 sk->sk_protocol = protocol;
1165 static int netlink_create(struct net *net, struct socket *sock, int protocol,
1168 struct module *module = NULL;
1169 struct mutex *cb_mutex;
1170 struct netlink_sock *nlk;
1171 void (*bind)(int group);
1174 sock->state = SS_UNCONNECTED;
1176 if (sock->type != SOCK_RAW && sock->type != SOCK_DGRAM)
1177 return -ESOCKTNOSUPPORT;
1179 if (protocol < 0 || protocol >= MAX_LINKS)
1180 return -EPROTONOSUPPORT;
1182 netlink_lock_table();
1183 #ifdef CONFIG_MODULES
1184 if (!nl_table[protocol].registered) {
1185 netlink_unlock_table();
1186 request_module("net-pf-%d-proto-%d", PF_NETLINK, protocol);
1187 netlink_lock_table();
1190 if (nl_table[protocol].registered &&
1191 try_module_get(nl_table[protocol].module))
1192 module = nl_table[protocol].module;
1194 err = -EPROTONOSUPPORT;
1195 cb_mutex = nl_table[protocol].cb_mutex;
1196 bind = nl_table[protocol].bind;
1197 netlink_unlock_table();
1202 err = __netlink_create(net, sock, cb_mutex, protocol);
1207 sock_prot_inuse_add(net, &netlink_proto, 1);
1210 nlk = nlk_sk(sock->sk);
1211 nlk->module = module;
1212 nlk->netlink_bind = bind;
1221 static int netlink_release(struct socket *sock)
1223 struct sock *sk = sock->sk;
1224 struct netlink_sock *nlk;
1234 * OK. Socket is unlinked, any packets that arrive now
1239 wake_up_interruptible_all(&nlk->wait);
1241 skb_queue_purge(&sk->sk_write_queue);
1244 struct netlink_notify n = {
1245 .net = sock_net(sk),
1246 .protocol = sk->sk_protocol,
1247 .portid = nlk->portid,
1249 atomic_notifier_call_chain(&netlink_chain,
1250 NETLINK_URELEASE, &n);
1253 module_put(nlk->module);
1255 netlink_table_grab();
1256 if (netlink_is_kernel(sk)) {
1257 BUG_ON(nl_table[sk->sk_protocol].registered == 0);
1258 if (--nl_table[sk->sk_protocol].registered == 0) {
1259 struct listeners *old;
1261 old = nl_deref_protected(nl_table[sk->sk_protocol].listeners);
1262 RCU_INIT_POINTER(nl_table[sk->sk_protocol].listeners, NULL);
1263 kfree_rcu(old, rcu);
1264 nl_table[sk->sk_protocol].module = NULL;
1265 nl_table[sk->sk_protocol].bind = NULL;
1266 nl_table[sk->sk_protocol].flags = 0;
1267 nl_table[sk->sk_protocol].registered = 0;
1269 } else if (nlk->subscriptions) {
1270 netlink_update_listeners(sk);
1272 netlink_table_ungrab();
1278 sock_prot_inuse_add(sock_net(sk), &netlink_proto, -1);
1284 static int netlink_autobind(struct socket *sock)
1286 struct sock *sk = sock->sk;
1287 struct net *net = sock_net(sk);
1288 struct netlink_table *table = &nl_table[sk->sk_protocol];
1289 struct nl_portid_hash *hash = &table->hash;
1290 struct hlist_head *head;
1292 s32 portid = task_tgid_vnr(current);
1294 static s32 rover = -4097;
1298 netlink_table_grab();
1299 head = nl_portid_hashfn(hash, portid);
1300 sk_for_each(osk, head) {
1301 if (!table->compare(net, osk))
1303 if (nlk_sk(osk)->portid == portid) {
1304 /* Bind collision, search negative portid values. */
1308 netlink_table_ungrab();
1312 netlink_table_ungrab();
1314 err = netlink_insert(sk, net, portid);
1315 if (err == -EADDRINUSE)
1318 /* If 2 threads race to autobind, that is fine. */
1325 static inline int netlink_capable(const struct socket *sock, unsigned int flag)
1327 return (nl_table[sock->sk->sk_protocol].flags & flag) ||
1328 ns_capable(sock_net(sock->sk)->user_ns, CAP_NET_ADMIN);
1332 netlink_update_subscriptions(struct sock *sk, unsigned int subscriptions)
1334 struct netlink_sock *nlk = nlk_sk(sk);
1336 if (nlk->subscriptions && !subscriptions)
1337 __sk_del_bind_node(sk);
1338 else if (!nlk->subscriptions && subscriptions)
1339 sk_add_bind_node(sk, &nl_table[sk->sk_protocol].mc_list);
1340 nlk->subscriptions = subscriptions;
1343 static int netlink_realloc_groups(struct sock *sk)
1345 struct netlink_sock *nlk = nlk_sk(sk);
1346 unsigned int groups;
1347 unsigned long *new_groups;
1350 netlink_table_grab();
1352 groups = nl_table[sk->sk_protocol].groups;
1353 if (!nl_table[sk->sk_protocol].registered) {
1358 if (nlk->ngroups >= groups)
1361 new_groups = krealloc(nlk->groups, NLGRPSZ(groups), GFP_ATOMIC);
1362 if (new_groups == NULL) {
1366 memset((char *)new_groups + NLGRPSZ(nlk->ngroups), 0,
1367 NLGRPSZ(groups) - NLGRPSZ(nlk->ngroups));
1369 nlk->groups = new_groups;
1370 nlk->ngroups = groups;
1372 netlink_table_ungrab();
1376 static int netlink_bind(struct socket *sock, struct sockaddr *addr,
1379 struct sock *sk = sock->sk;
1380 struct net *net = sock_net(sk);
1381 struct netlink_sock *nlk = nlk_sk(sk);
1382 struct sockaddr_nl *nladdr = (struct sockaddr_nl *)addr;
1385 if (addr_len < sizeof(struct sockaddr_nl))
1388 if (nladdr->nl_family != AF_NETLINK)
1391 /* Only superuser is allowed to listen multicasts */
1392 if (nladdr->nl_groups) {
1393 if (!netlink_capable(sock, NL_CFG_F_NONROOT_RECV))
1395 err = netlink_realloc_groups(sk);
1401 if (nladdr->nl_pid != nlk->portid)
1404 err = nladdr->nl_pid ?
1405 netlink_insert(sk, net, nladdr->nl_pid) :
1406 netlink_autobind(sock);
1411 if (!nladdr->nl_groups && (nlk->groups == NULL || !(u32)nlk->groups[0]))
1414 netlink_table_grab();
1415 netlink_update_subscriptions(sk, nlk->subscriptions +
1416 hweight32(nladdr->nl_groups) -
1417 hweight32(nlk->groups[0]));
1418 nlk->groups[0] = (nlk->groups[0] & ~0xffffffffUL) | nladdr->nl_groups;
1419 netlink_update_listeners(sk);
1420 netlink_table_ungrab();
1422 if (nlk->netlink_bind && nlk->groups[0]) {
1425 for (i=0; i<nlk->ngroups; i++) {
1426 if (test_bit(i, nlk->groups))
1427 nlk->netlink_bind(i);
1434 static int netlink_connect(struct socket *sock, struct sockaddr *addr,
1435 int alen, int flags)
1438 struct sock *sk = sock->sk;
1439 struct netlink_sock *nlk = nlk_sk(sk);
1440 struct sockaddr_nl *nladdr = (struct sockaddr_nl *)addr;
1442 if (alen < sizeof(addr->sa_family))
1445 if (addr->sa_family == AF_UNSPEC) {
1446 sk->sk_state = NETLINK_UNCONNECTED;
1447 nlk->dst_portid = 0;
1451 if (addr->sa_family != AF_NETLINK)
1454 /* Only superuser is allowed to send multicasts */
1455 if (nladdr->nl_groups && !netlink_capable(sock, NL_CFG_F_NONROOT_SEND))
1459 err = netlink_autobind(sock);
1462 sk->sk_state = NETLINK_CONNECTED;
1463 nlk->dst_portid = nladdr->nl_pid;
1464 nlk->dst_group = ffs(nladdr->nl_groups);
1470 static int netlink_getname(struct socket *sock, struct sockaddr *addr,
1471 int *addr_len, int peer)
1473 struct sock *sk = sock->sk;
1474 struct netlink_sock *nlk = nlk_sk(sk);
1475 DECLARE_SOCKADDR(struct sockaddr_nl *, nladdr, addr);
1477 nladdr->nl_family = AF_NETLINK;
1479 *addr_len = sizeof(*nladdr);
1482 nladdr->nl_pid = nlk->dst_portid;
1483 nladdr->nl_groups = netlink_group_mask(nlk->dst_group);
1485 nladdr->nl_pid = nlk->portid;
1486 nladdr->nl_groups = nlk->groups ? nlk->groups[0] : 0;
1491 static struct sock *netlink_getsockbyportid(struct sock *ssk, u32 portid)
1494 struct netlink_sock *nlk;
1496 sock = netlink_lookup(sock_net(ssk), ssk->sk_protocol, portid);
1498 return ERR_PTR(-ECONNREFUSED);
1500 /* Don't bother queuing skb if kernel socket has no input function */
1502 if (sock->sk_state == NETLINK_CONNECTED &&
1503 nlk->dst_portid != nlk_sk(ssk)->portid) {
1505 return ERR_PTR(-ECONNREFUSED);
1510 struct sock *netlink_getsockbyfilp(struct file *filp)
1512 struct inode *inode = file_inode(filp);
1515 if (!S_ISSOCK(inode->i_mode))
1516 return ERR_PTR(-ENOTSOCK);
1518 sock = SOCKET_I(inode)->sk;
1519 if (sock->sk_family != AF_NETLINK)
1520 return ERR_PTR(-EINVAL);
1526 static struct sk_buff *netlink_alloc_large_skb(unsigned int size,
1529 struct sk_buff *skb;
1532 if (size <= NLMSG_GOODSIZE || broadcast)
1533 return alloc_skb(size, GFP_KERNEL);
1535 size = SKB_DATA_ALIGN(size) +
1536 SKB_DATA_ALIGN(sizeof(struct skb_shared_info));
1538 data = vmalloc(size);
1542 skb = build_skb(data, size);
1547 skb->destructor = netlink_skb_destructor;
1554 * Attach a skb to a netlink socket.
1555 * The caller must hold a reference to the destination socket. On error, the
1556 * reference is dropped. The skb is not send to the destination, just all
1557 * all error checks are performed and memory in the queue is reserved.
1559 * < 0: error. skb freed, reference to sock dropped.
1561 * 1: repeat lookup - reference dropped while waiting for socket memory.
1563 int netlink_attachskb(struct sock *sk, struct sk_buff *skb,
1564 long *timeo, struct sock *ssk)
1566 struct netlink_sock *nlk;
1570 if ((atomic_read(&sk->sk_rmem_alloc) > sk->sk_rcvbuf ||
1571 test_bit(NETLINK_CONGESTED, &nlk->state)) &&
1572 !netlink_skb_is_mmaped(skb)) {
1573 DECLARE_WAITQUEUE(wait, current);
1575 if (!ssk || netlink_is_kernel(ssk))
1576 netlink_overrun(sk);
1582 __set_current_state(TASK_INTERRUPTIBLE);
1583 add_wait_queue(&nlk->wait, &wait);
1585 if ((atomic_read(&sk->sk_rmem_alloc) > sk->sk_rcvbuf ||
1586 test_bit(NETLINK_CONGESTED, &nlk->state)) &&
1587 !sock_flag(sk, SOCK_DEAD))
1588 *timeo = schedule_timeout(*timeo);
1590 __set_current_state(TASK_RUNNING);
1591 remove_wait_queue(&nlk->wait, &wait);
1594 if (signal_pending(current)) {
1596 return sock_intr_errno(*timeo);
1600 netlink_skb_set_owner_r(skb, sk);
1604 static int __netlink_sendskb(struct sock *sk, struct sk_buff *skb)
1608 netlink_deliver_tap(skb);
1610 #ifdef CONFIG_NETLINK_MMAP
1611 if (netlink_skb_is_mmaped(skb))
1612 netlink_queue_mmaped_skb(sk, skb);
1613 else if (netlink_rx_is_mmaped(sk))
1614 netlink_ring_set_copied(sk, skb);
1616 #endif /* CONFIG_NETLINK_MMAP */
1617 skb_queue_tail(&sk->sk_receive_queue, skb);
1618 sk->sk_data_ready(sk, len);
1622 int netlink_sendskb(struct sock *sk, struct sk_buff *skb)
1624 int len = __netlink_sendskb(sk, skb);
1630 void netlink_detachskb(struct sock *sk, struct sk_buff *skb)
1636 static struct sk_buff *netlink_trim(struct sk_buff *skb, gfp_t allocation)
1640 WARN_ON(skb->sk != NULL);
1641 if (netlink_skb_is_mmaped(skb))
1644 delta = skb->end - skb->tail;
1645 if (is_vmalloc_addr(skb->head) || delta * 2 < skb->truesize)
1648 if (skb_shared(skb)) {
1649 struct sk_buff *nskb = skb_clone(skb, allocation);
1656 if (!pskb_expand_head(skb, 0, -delta, allocation))
1657 skb->truesize -= delta;
1662 static int netlink_unicast_kernel(struct sock *sk, struct sk_buff *skb,
1666 struct netlink_sock *nlk = nlk_sk(sk);
1668 ret = -ECONNREFUSED;
1669 if (nlk->netlink_rcv != NULL) {
1670 /* We could do a netlink_deliver_tap(skb) here as well
1671 * but since this is intended for the kernel only, we
1672 * should rather let it stay under the hood.
1676 netlink_skb_set_owner_r(skb, sk);
1677 NETLINK_CB(skb).sk = ssk;
1678 nlk->netlink_rcv(skb);
1687 int netlink_unicast(struct sock *ssk, struct sk_buff *skb,
1688 u32 portid, int nonblock)
1694 skb = netlink_trim(skb, gfp_any());
1696 timeo = sock_sndtimeo(ssk, nonblock);
1698 sk = netlink_getsockbyportid(ssk, portid);
1703 if (netlink_is_kernel(sk))
1704 return netlink_unicast_kernel(sk, skb, ssk);
1706 if (sk_filter(sk, skb)) {
1713 err = netlink_attachskb(sk, skb, &timeo, ssk);
1719 return netlink_sendskb(sk, skb);
1721 EXPORT_SYMBOL(netlink_unicast);
1723 struct sk_buff *netlink_alloc_skb(struct sock *ssk, unsigned int size,
1724 u32 dst_portid, gfp_t gfp_mask)
1726 #ifdef CONFIG_NETLINK_MMAP
1727 struct sock *sk = NULL;
1728 struct sk_buff *skb;
1729 struct netlink_ring *ring;
1730 struct nl_mmap_hdr *hdr;
1731 unsigned int maxlen;
1733 sk = netlink_getsockbyportid(ssk, dst_portid);
1737 ring = &nlk_sk(sk)->rx_ring;
1738 /* fast-path without atomic ops for common case: non-mmaped receiver */
1739 if (ring->pg_vec == NULL)
1742 skb = alloc_skb_head(gfp_mask);
1746 spin_lock_bh(&sk->sk_receive_queue.lock);
1747 /* check again under lock */
1748 if (ring->pg_vec == NULL)
1751 maxlen = ring->frame_size - NL_MMAP_HDRLEN;
1755 netlink_forward_ring(ring);
1756 hdr = netlink_current_frame(ring, NL_MMAP_STATUS_UNUSED);
1759 netlink_ring_setup_skb(skb, sk, ring, hdr);
1760 netlink_set_status(hdr, NL_MMAP_STATUS_RESERVED);
1761 atomic_inc(&ring->pending);
1762 netlink_increment_head(ring);
1764 spin_unlock_bh(&sk->sk_receive_queue.lock);
1769 spin_unlock_bh(&sk->sk_receive_queue.lock);
1770 netlink_overrun(sk);
1777 spin_unlock_bh(&sk->sk_receive_queue.lock);
1782 return alloc_skb(size, gfp_mask);
1784 EXPORT_SYMBOL_GPL(netlink_alloc_skb);
1786 int netlink_has_listeners(struct sock *sk, unsigned int group)
1789 struct listeners *listeners;
1791 BUG_ON(!netlink_is_kernel(sk));
1794 listeners = rcu_dereference(nl_table[sk->sk_protocol].listeners);
1796 if (listeners && group - 1 < nl_table[sk->sk_protocol].groups)
1797 res = test_bit(group - 1, listeners->masks);
1803 EXPORT_SYMBOL_GPL(netlink_has_listeners);
1805 static int netlink_broadcast_deliver(struct sock *sk, struct sk_buff *skb)
1807 struct netlink_sock *nlk = nlk_sk(sk);
1809 if (atomic_read(&sk->sk_rmem_alloc) <= sk->sk_rcvbuf &&
1810 !test_bit(NETLINK_CONGESTED, &nlk->state)) {
1811 netlink_skb_set_owner_r(skb, sk);
1812 __netlink_sendskb(sk, skb);
1813 return atomic_read(&sk->sk_rmem_alloc) > (sk->sk_rcvbuf >> 1);
1818 struct netlink_broadcast_data {
1819 struct sock *exclude_sk;
1824 int delivery_failure;
1828 struct sk_buff *skb, *skb2;
1829 int (*tx_filter)(struct sock *dsk, struct sk_buff *skb, void *data);
1833 static int do_one_broadcast(struct sock *sk,
1834 struct netlink_broadcast_data *p)
1836 struct netlink_sock *nlk = nlk_sk(sk);
1839 if (p->exclude_sk == sk)
1842 if (nlk->portid == p->portid || p->group - 1 >= nlk->ngroups ||
1843 !test_bit(p->group - 1, nlk->groups))
1846 if (!net_eq(sock_net(sk), p->net))
1850 netlink_overrun(sk);
1855 if (p->skb2 == NULL) {
1856 if (skb_shared(p->skb)) {
1857 p->skb2 = skb_clone(p->skb, p->allocation);
1859 p->skb2 = skb_get(p->skb);
1861 * skb ownership may have been set when
1862 * delivered to a previous socket.
1864 skb_orphan(p->skb2);
1867 if (p->skb2 == NULL) {
1868 netlink_overrun(sk);
1869 /* Clone failed. Notify ALL listeners. */
1871 if (nlk->flags & NETLINK_BROADCAST_SEND_ERROR)
1872 p->delivery_failure = 1;
1873 } else if (p->tx_filter && p->tx_filter(sk, p->skb2, p->tx_data)) {
1876 } else if (sk_filter(sk, p->skb2)) {
1879 } else if ((val = netlink_broadcast_deliver(sk, p->skb2)) < 0) {
1880 netlink_overrun(sk);
1881 if (nlk->flags & NETLINK_BROADCAST_SEND_ERROR)
1882 p->delivery_failure = 1;
1884 p->congested |= val;
1894 int netlink_broadcast_filtered(struct sock *ssk, struct sk_buff *skb, u32 portid,
1895 u32 group, gfp_t allocation,
1896 int (*filter)(struct sock *dsk, struct sk_buff *skb, void *data),
1899 struct net *net = sock_net(ssk);
1900 struct netlink_broadcast_data info;
1903 skb = netlink_trim(skb, allocation);
1905 info.exclude_sk = ssk;
1907 info.portid = portid;
1910 info.delivery_failure = 0;
1913 info.allocation = allocation;
1916 info.tx_filter = filter;
1917 info.tx_data = filter_data;
1919 /* While we sleep in clone, do not allow to change socket list */
1921 netlink_lock_table();
1923 sk_for_each_bound(sk, &nl_table[ssk->sk_protocol].mc_list)
1924 do_one_broadcast(sk, &info);
1928 netlink_unlock_table();
1930 if (info.delivery_failure) {
1931 kfree_skb(info.skb2);
1934 consume_skb(info.skb2);
1936 if (info.delivered) {
1937 if (info.congested && (allocation & __GFP_WAIT))
1943 EXPORT_SYMBOL(netlink_broadcast_filtered);
1945 int netlink_broadcast(struct sock *ssk, struct sk_buff *skb, u32 portid,
1946 u32 group, gfp_t allocation)
1948 return netlink_broadcast_filtered(ssk, skb, portid, group, allocation,
1951 EXPORT_SYMBOL(netlink_broadcast);
1953 struct netlink_set_err_data {
1954 struct sock *exclude_sk;
1960 static int do_one_set_err(struct sock *sk, struct netlink_set_err_data *p)
1962 struct netlink_sock *nlk = nlk_sk(sk);
1965 if (sk == p->exclude_sk)
1968 if (!net_eq(sock_net(sk), sock_net(p->exclude_sk)))
1971 if (nlk->portid == p->portid || p->group - 1 >= nlk->ngroups ||
1972 !test_bit(p->group - 1, nlk->groups))
1975 if (p->code == ENOBUFS && nlk->flags & NETLINK_RECV_NO_ENOBUFS) {
1980 sk->sk_err = p->code;
1981 sk->sk_error_report(sk);
1987 * netlink_set_err - report error to broadcast listeners
1988 * @ssk: the kernel netlink socket, as returned by netlink_kernel_create()
1989 * @portid: the PORTID of a process that we want to skip (if any)
1990 * @groups: the broadcast group that will notice the error
1991 * @code: error code, must be negative (as usual in kernelspace)
1993 * This function returns the number of broadcast listeners that have set the
1994 * NETLINK_RECV_NO_ENOBUFS socket option.
1996 int netlink_set_err(struct sock *ssk, u32 portid, u32 group, int code)
1998 struct netlink_set_err_data info;
2002 info.exclude_sk = ssk;
2003 info.portid = portid;
2005 /* sk->sk_err wants a positive error value */
2008 read_lock(&nl_table_lock);
2010 sk_for_each_bound(sk, &nl_table[ssk->sk_protocol].mc_list)
2011 ret += do_one_set_err(sk, &info);
2013 read_unlock(&nl_table_lock);
2016 EXPORT_SYMBOL(netlink_set_err);
2018 /* must be called with netlink table grabbed */
2019 static void netlink_update_socket_mc(struct netlink_sock *nlk,
2023 int old, new = !!is_new, subscriptions;
2025 old = test_bit(group - 1, nlk->groups);
2026 subscriptions = nlk->subscriptions - old + new;
2028 __set_bit(group - 1, nlk->groups);
2030 __clear_bit(group - 1, nlk->groups);
2031 netlink_update_subscriptions(&nlk->sk, subscriptions);
2032 netlink_update_listeners(&nlk->sk);
2035 static int netlink_setsockopt(struct socket *sock, int level, int optname,
2036 char __user *optval, unsigned int optlen)
2038 struct sock *sk = sock->sk;
2039 struct netlink_sock *nlk = nlk_sk(sk);
2040 unsigned int val = 0;
2043 if (level != SOL_NETLINK)
2044 return -ENOPROTOOPT;
2046 if (optname != NETLINK_RX_RING && optname != NETLINK_TX_RING &&
2047 optlen >= sizeof(int) &&
2048 get_user(val, (unsigned int __user *)optval))
2052 case NETLINK_PKTINFO:
2054 nlk->flags |= NETLINK_RECV_PKTINFO;
2056 nlk->flags &= ~NETLINK_RECV_PKTINFO;
2059 case NETLINK_ADD_MEMBERSHIP:
2060 case NETLINK_DROP_MEMBERSHIP: {
2061 if (!netlink_capable(sock, NL_CFG_F_NONROOT_RECV))
2063 err = netlink_realloc_groups(sk);
2066 if (!val || val - 1 >= nlk->ngroups)
2068 netlink_table_grab();
2069 netlink_update_socket_mc(nlk, val,
2070 optname == NETLINK_ADD_MEMBERSHIP);
2071 netlink_table_ungrab();
2073 if (nlk->netlink_bind)
2074 nlk->netlink_bind(val);
2079 case NETLINK_BROADCAST_ERROR:
2081 nlk->flags |= NETLINK_BROADCAST_SEND_ERROR;
2083 nlk->flags &= ~NETLINK_BROADCAST_SEND_ERROR;
2086 case NETLINK_NO_ENOBUFS:
2088 nlk->flags |= NETLINK_RECV_NO_ENOBUFS;
2089 clear_bit(NETLINK_CONGESTED, &nlk->state);
2090 wake_up_interruptible(&nlk->wait);
2092 nlk->flags &= ~NETLINK_RECV_NO_ENOBUFS;
2096 #ifdef CONFIG_NETLINK_MMAP
2097 case NETLINK_RX_RING:
2098 case NETLINK_TX_RING: {
2099 struct nl_mmap_req req;
2101 /* Rings might consume more memory than queue limits, require
2104 if (!capable(CAP_NET_ADMIN))
2106 if (optlen < sizeof(req))
2108 if (copy_from_user(&req, optval, sizeof(req)))
2110 err = netlink_set_ring(sk, &req, false,
2111 optname == NETLINK_TX_RING);
2114 #endif /* CONFIG_NETLINK_MMAP */
2121 static int netlink_getsockopt(struct socket *sock, int level, int optname,
2122 char __user *optval, int __user *optlen)
2124 struct sock *sk = sock->sk;
2125 struct netlink_sock *nlk = nlk_sk(sk);
2128 if (level != SOL_NETLINK)
2129 return -ENOPROTOOPT;
2131 if (get_user(len, optlen))
2137 case NETLINK_PKTINFO:
2138 if (len < sizeof(int))
2141 val = nlk->flags & NETLINK_RECV_PKTINFO ? 1 : 0;
2142 if (put_user(len, optlen) ||
2143 put_user(val, optval))
2147 case NETLINK_BROADCAST_ERROR:
2148 if (len < sizeof(int))
2151 val = nlk->flags & NETLINK_BROADCAST_SEND_ERROR ? 1 : 0;
2152 if (put_user(len, optlen) ||
2153 put_user(val, optval))
2157 case NETLINK_NO_ENOBUFS:
2158 if (len < sizeof(int))
2161 val = nlk->flags & NETLINK_RECV_NO_ENOBUFS ? 1 : 0;
2162 if (put_user(len, optlen) ||
2163 put_user(val, optval))
2173 static void netlink_cmsg_recv_pktinfo(struct msghdr *msg, struct sk_buff *skb)
2175 struct nl_pktinfo info;
2177 info.group = NETLINK_CB(skb).dst_group;
2178 put_cmsg(msg, SOL_NETLINK, NETLINK_PKTINFO, sizeof(info), &info);
2181 static int netlink_sendmsg(struct kiocb *kiocb, struct socket *sock,
2182 struct msghdr *msg, size_t len)
2184 struct sock_iocb *siocb = kiocb_to_siocb(kiocb);
2185 struct sock *sk = sock->sk;
2186 struct netlink_sock *nlk = nlk_sk(sk);
2187 struct sockaddr_nl *addr = msg->msg_name;
2190 struct sk_buff *skb;
2192 struct scm_cookie scm;
2194 if (msg->msg_flags&MSG_OOB)
2197 if (NULL == siocb->scm)
2200 err = scm_send(sock, msg, siocb->scm, true);
2204 if (msg->msg_namelen) {
2206 if (addr->nl_family != AF_NETLINK)
2208 dst_portid = addr->nl_pid;
2209 dst_group = ffs(addr->nl_groups);
2211 if ((dst_group || dst_portid) &&
2212 !netlink_capable(sock, NL_CFG_F_NONROOT_SEND))
2215 dst_portid = nlk->dst_portid;
2216 dst_group = nlk->dst_group;
2220 err = netlink_autobind(sock);
2225 if (netlink_tx_is_mmaped(sk) &&
2226 msg->msg_iov->iov_base == NULL) {
2227 err = netlink_mmap_sendmsg(sk, msg, dst_portid, dst_group,
2233 if (len > sk->sk_sndbuf - 32)
2236 skb = netlink_alloc_large_skb(len, dst_group);
2240 NETLINK_CB(skb).portid = nlk->portid;
2241 NETLINK_CB(skb).dst_group = dst_group;
2242 NETLINK_CB(skb).creds = siocb->scm->creds;
2245 if (memcpy_fromiovec(skb_put(skb, len), msg->msg_iov, len)) {
2250 err = security_netlink_send(sk, skb);
2257 atomic_inc(&skb->users);
2258 netlink_broadcast(sk, skb, dst_portid, dst_group, GFP_KERNEL);
2260 err = netlink_unicast(sk, skb, dst_portid, msg->msg_flags&MSG_DONTWAIT);
2263 scm_destroy(siocb->scm);
2267 static int netlink_recvmsg(struct kiocb *kiocb, struct socket *sock,
2268 struct msghdr *msg, size_t len,
2271 struct sock_iocb *siocb = kiocb_to_siocb(kiocb);
2272 struct scm_cookie scm;
2273 struct sock *sk = sock->sk;
2274 struct netlink_sock *nlk = nlk_sk(sk);
2275 int noblock = flags&MSG_DONTWAIT;
2277 struct sk_buff *skb, *data_skb;
2285 skb = skb_recv_datagram(sk, flags, noblock, &err);
2291 #ifdef CONFIG_COMPAT_NETLINK_MESSAGES
2292 if (unlikely(skb_shinfo(skb)->frag_list)) {
2294 * If this skb has a frag_list, then here that means that we
2295 * will have to use the frag_list skb's data for compat tasks
2296 * and the regular skb's data for normal (non-compat) tasks.
2298 * If we need to send the compat skb, assign it to the
2299 * 'data_skb' variable so that it will be used below for data
2300 * copying. We keep 'skb' for everything else, including
2301 * freeing both later.
2303 if (flags & MSG_CMSG_COMPAT)
2304 data_skb = skb_shinfo(skb)->frag_list;
2308 msg->msg_namelen = 0;
2310 copied = data_skb->len;
2312 msg->msg_flags |= MSG_TRUNC;
2316 skb_reset_transport_header(data_skb);
2317 err = skb_copy_datagram_iovec(data_skb, 0, msg->msg_iov, copied);
2319 if (msg->msg_name) {
2320 struct sockaddr_nl *addr = (struct sockaddr_nl *)msg->msg_name;
2321 addr->nl_family = AF_NETLINK;
2323 addr->nl_pid = NETLINK_CB(skb).portid;
2324 addr->nl_groups = netlink_group_mask(NETLINK_CB(skb).dst_group);
2325 msg->msg_namelen = sizeof(*addr);
2328 if (nlk->flags & NETLINK_RECV_PKTINFO)
2329 netlink_cmsg_recv_pktinfo(msg, skb);
2331 if (NULL == siocb->scm) {
2332 memset(&scm, 0, sizeof(scm));
2335 siocb->scm->creds = *NETLINK_CREDS(skb);
2336 if (flags & MSG_TRUNC)
2337 copied = data_skb->len;
2339 skb_free_datagram(sk, skb);
2341 if (nlk->cb_running &&
2342 atomic_read(&sk->sk_rmem_alloc) <= sk->sk_rcvbuf / 2) {
2343 ret = netlink_dump(sk);
2346 sk->sk_error_report(sk);
2350 scm_recv(sock, msg, siocb->scm, flags);
2352 netlink_rcv_wake(sk);
2353 return err ? : copied;
2356 static void netlink_data_ready(struct sock *sk, int len)
2362 * We export these functions to other modules. They provide a
2363 * complete set of kernel non-blocking support for message
2368 __netlink_kernel_create(struct net *net, int unit, struct module *module,
2369 struct netlink_kernel_cfg *cfg)
2371 struct socket *sock;
2373 struct netlink_sock *nlk;
2374 struct listeners *listeners = NULL;
2375 struct mutex *cb_mutex = cfg ? cfg->cb_mutex : NULL;
2376 unsigned int groups;
2380 if (unit < 0 || unit >= MAX_LINKS)
2383 if (sock_create_lite(PF_NETLINK, SOCK_DGRAM, unit, &sock))
2387 * We have to just have a reference on the net from sk, but don't
2388 * get_net it. Besides, we cannot get and then put the net here.
2389 * So we create one inside init_net and the move it to net.
2392 if (__netlink_create(&init_net, sock, cb_mutex, unit) < 0)
2393 goto out_sock_release_nosk;
2396 sk_change_net(sk, net);
2398 if (!cfg || cfg->groups < 32)
2401 groups = cfg->groups;
2403 listeners = kzalloc(sizeof(*listeners) + NLGRPSZ(groups), GFP_KERNEL);
2405 goto out_sock_release;
2407 sk->sk_data_ready = netlink_data_ready;
2408 if (cfg && cfg->input)
2409 nlk_sk(sk)->netlink_rcv = cfg->input;
2411 if (netlink_insert(sk, net, 0))
2412 goto out_sock_release;
2415 nlk->flags |= NETLINK_KERNEL_SOCKET;
2417 netlink_table_grab();
2418 if (!nl_table[unit].registered) {
2419 nl_table[unit].groups = groups;
2420 rcu_assign_pointer(nl_table[unit].listeners, listeners);
2421 nl_table[unit].cb_mutex = cb_mutex;
2422 nl_table[unit].module = module;
2424 nl_table[unit].bind = cfg->bind;
2425 nl_table[unit].flags = cfg->flags;
2427 nl_table[unit].compare = cfg->compare;
2429 nl_table[unit].registered = 1;
2432 nl_table[unit].registered++;
2434 netlink_table_ungrab();
2439 netlink_kernel_release(sk);
2442 out_sock_release_nosk:
2446 EXPORT_SYMBOL(__netlink_kernel_create);
2449 netlink_kernel_release(struct sock *sk)
2451 sk_release_kernel(sk);
2453 EXPORT_SYMBOL(netlink_kernel_release);
2455 int __netlink_change_ngroups(struct sock *sk, unsigned int groups)
2457 struct listeners *new, *old;
2458 struct netlink_table *tbl = &nl_table[sk->sk_protocol];
2463 if (NLGRPSZ(tbl->groups) < NLGRPSZ(groups)) {
2464 new = kzalloc(sizeof(*new) + NLGRPSZ(groups), GFP_ATOMIC);
2467 old = nl_deref_protected(tbl->listeners);
2468 memcpy(new->masks, old->masks, NLGRPSZ(tbl->groups));
2469 rcu_assign_pointer(tbl->listeners, new);
2471 kfree_rcu(old, rcu);
2473 tbl->groups = groups;
2479 * netlink_change_ngroups - change number of multicast groups
2481 * This changes the number of multicast groups that are available
2482 * on a certain netlink family. Note that it is not possible to
2483 * change the number of groups to below 32. Also note that it does
2484 * not implicitly call netlink_clear_multicast_users() when the
2485 * number of groups is reduced.
2487 * @sk: The kernel netlink socket, as returned by netlink_kernel_create().
2488 * @groups: The new number of groups.
2490 int netlink_change_ngroups(struct sock *sk, unsigned int groups)
2494 netlink_table_grab();
2495 err = __netlink_change_ngroups(sk, groups);
2496 netlink_table_ungrab();
2501 void __netlink_clear_multicast_users(struct sock *ksk, unsigned int group)
2504 struct netlink_table *tbl = &nl_table[ksk->sk_protocol];
2506 sk_for_each_bound(sk, &tbl->mc_list)
2507 netlink_update_socket_mc(nlk_sk(sk), group, 0);
2511 * netlink_clear_multicast_users - kick off multicast listeners
2513 * This function removes all listeners from the given group.
2514 * @ksk: The kernel netlink socket, as returned by
2515 * netlink_kernel_create().
2516 * @group: The multicast group to clear.
2518 void netlink_clear_multicast_users(struct sock *ksk, unsigned int group)
2520 netlink_table_grab();
2521 __netlink_clear_multicast_users(ksk, group);
2522 netlink_table_ungrab();
2526 __nlmsg_put(struct sk_buff *skb, u32 portid, u32 seq, int type, int len, int flags)
2528 struct nlmsghdr *nlh;
2529 int size = nlmsg_msg_size(len);
2531 nlh = (struct nlmsghdr*)skb_put(skb, NLMSG_ALIGN(size));
2532 nlh->nlmsg_type = type;
2533 nlh->nlmsg_len = size;
2534 nlh->nlmsg_flags = flags;
2535 nlh->nlmsg_pid = portid;
2536 nlh->nlmsg_seq = seq;
2537 if (!__builtin_constant_p(size) || NLMSG_ALIGN(size) - size != 0)
2538 memset(nlmsg_data(nlh) + len, 0, NLMSG_ALIGN(size) - size);
2541 EXPORT_SYMBOL(__nlmsg_put);
2544 * It looks a bit ugly.
2545 * It would be better to create kernel thread.
2548 static int netlink_dump(struct sock *sk)
2550 struct netlink_sock *nlk = nlk_sk(sk);
2551 struct netlink_callback *cb;
2552 struct sk_buff *skb = NULL;
2553 struct nlmsghdr *nlh;
2554 int len, err = -ENOBUFS;
2557 mutex_lock(nlk->cb_mutex);
2558 if (!nlk->cb_running) {
2564 alloc_size = max_t(int, cb->min_dump_alloc, NLMSG_GOODSIZE);
2566 if (!netlink_rx_is_mmaped(sk) &&
2567 atomic_read(&sk->sk_rmem_alloc) >= sk->sk_rcvbuf)
2569 skb = netlink_alloc_skb(sk, alloc_size, nlk->portid, GFP_KERNEL);
2572 netlink_skb_set_owner_r(skb, sk);
2574 len = cb->dump(skb, cb);
2577 mutex_unlock(nlk->cb_mutex);
2579 if (sk_filter(sk, skb))
2582 __netlink_sendskb(sk, skb);
2586 nlh = nlmsg_put_answer(skb, cb, NLMSG_DONE, sizeof(len), NLM_F_MULTI);
2590 nl_dump_check_consistent(cb, nlh);
2592 memcpy(nlmsg_data(nlh), &len, sizeof(len));
2594 if (sk_filter(sk, skb))
2597 __netlink_sendskb(sk, skb);
2602 nlk->cb_running = false;
2603 mutex_unlock(nlk->cb_mutex);
2604 module_put(cb->module);
2605 consume_skb(cb->skb);
2609 mutex_unlock(nlk->cb_mutex);
2614 int __netlink_dump_start(struct sock *ssk, struct sk_buff *skb,
2615 const struct nlmsghdr *nlh,
2616 struct netlink_dump_control *control)
2618 struct netlink_callback *cb;
2620 struct netlink_sock *nlk;
2623 /* Memory mapped dump requests need to be copied to avoid looping
2624 * on the pending state in netlink_mmap_sendmsg() while the CB hold
2625 * a reference to the skb.
2627 if (netlink_skb_is_mmaped(skb)) {
2628 skb = skb_copy(skb, GFP_KERNEL);
2632 atomic_inc(&skb->users);
2634 sk = netlink_lookup(sock_net(ssk), ssk->sk_protocol, NETLINK_CB(skb).portid);
2636 ret = -ECONNREFUSED;
2641 mutex_lock(nlk->cb_mutex);
2642 /* A dump is in progress... */
2643 if (nlk->cb_running) {
2647 /* add reference of module which cb->dump belongs to */
2648 if (!try_module_get(control->module)) {
2649 ret = -EPROTONOSUPPORT;
2654 memset(cb, 0, sizeof(*cb));
2655 cb->dump = control->dump;
2656 cb->done = control->done;
2658 cb->data = control->data;
2659 cb->module = control->module;
2660 cb->min_dump_alloc = control->min_dump_alloc;
2663 nlk->cb_running = true;
2665 mutex_unlock(nlk->cb_mutex);
2667 ret = netlink_dump(sk);
2673 /* We successfully started a dump, by returning -EINTR we
2674 * signal not to send ACK even if it was requested.
2680 mutex_unlock(nlk->cb_mutex);
2685 EXPORT_SYMBOL(__netlink_dump_start);
2687 void netlink_ack(struct sk_buff *in_skb, struct nlmsghdr *nlh, int err)
2689 struct sk_buff *skb;
2690 struct nlmsghdr *rep;
2691 struct nlmsgerr *errmsg;
2692 size_t payload = sizeof(*errmsg);
2694 /* error messages get the original request appened */
2696 payload += nlmsg_len(nlh);
2698 skb = netlink_alloc_skb(in_skb->sk, nlmsg_total_size(payload),
2699 NETLINK_CB(in_skb).portid, GFP_KERNEL);
2703 sk = netlink_lookup(sock_net(in_skb->sk),
2704 in_skb->sk->sk_protocol,
2705 NETLINK_CB(in_skb).portid);
2707 sk->sk_err = ENOBUFS;
2708 sk->sk_error_report(sk);
2714 rep = __nlmsg_put(skb, NETLINK_CB(in_skb).portid, nlh->nlmsg_seq,
2715 NLMSG_ERROR, payload, 0);
2716 errmsg = nlmsg_data(rep);
2717 errmsg->error = err;
2718 memcpy(&errmsg->msg, nlh, err ? nlh->nlmsg_len : sizeof(*nlh));
2719 netlink_unicast(in_skb->sk, skb, NETLINK_CB(in_skb).portid, MSG_DONTWAIT);
2721 EXPORT_SYMBOL(netlink_ack);
2723 int netlink_rcv_skb(struct sk_buff *skb, int (*cb)(struct sk_buff *,
2726 struct nlmsghdr *nlh;
2729 while (skb->len >= nlmsg_total_size(0)) {
2732 nlh = nlmsg_hdr(skb);
2735 if (nlh->nlmsg_len < NLMSG_HDRLEN || skb->len < nlh->nlmsg_len)
2738 /* Only requests are handled by the kernel */
2739 if (!(nlh->nlmsg_flags & NLM_F_REQUEST))
2742 /* Skip control messages */
2743 if (nlh->nlmsg_type < NLMSG_MIN_TYPE)
2751 if (nlh->nlmsg_flags & NLM_F_ACK || err)
2752 netlink_ack(skb, nlh, err);
2755 msglen = NLMSG_ALIGN(nlh->nlmsg_len);
2756 if (msglen > skb->len)
2758 skb_pull(skb, msglen);
2763 EXPORT_SYMBOL(netlink_rcv_skb);
2766 * nlmsg_notify - send a notification netlink message
2767 * @sk: netlink socket to use
2768 * @skb: notification message
2769 * @portid: destination netlink portid for reports or 0
2770 * @group: destination multicast group or 0
2771 * @report: 1 to report back, 0 to disable
2772 * @flags: allocation flags
2774 int nlmsg_notify(struct sock *sk, struct sk_buff *skb, u32 portid,
2775 unsigned int group, int report, gfp_t flags)
2780 int exclude_portid = 0;
2783 atomic_inc(&skb->users);
2784 exclude_portid = portid;
2787 /* errors reported via destination sk->sk_err, but propagate
2788 * delivery errors if NETLINK_BROADCAST_ERROR flag is set */
2789 err = nlmsg_multicast(sk, skb, exclude_portid, group, flags);
2795 err2 = nlmsg_unicast(sk, skb, portid);
2796 if (!err || err == -ESRCH)
2802 EXPORT_SYMBOL(nlmsg_notify);
2804 #ifdef CONFIG_PROC_FS
2805 struct nl_seq_iter {
2806 struct seq_net_private p;
2811 static struct sock *netlink_seq_socket_idx(struct seq_file *seq, loff_t pos)
2813 struct nl_seq_iter *iter = seq->private;
2818 for (i = 0; i < MAX_LINKS; i++) {
2819 struct nl_portid_hash *hash = &nl_table[i].hash;
2821 for (j = 0; j <= hash->mask; j++) {
2822 sk_for_each(s, &hash->table[j]) {
2823 if (sock_net(s) != seq_file_net(seq))
2837 static void *netlink_seq_start(struct seq_file *seq, loff_t *pos)
2838 __acquires(nl_table_lock)
2840 read_lock(&nl_table_lock);
2841 return *pos ? netlink_seq_socket_idx(seq, *pos - 1) : SEQ_START_TOKEN;
2844 static void *netlink_seq_next(struct seq_file *seq, void *v, loff_t *pos)
2847 struct nl_seq_iter *iter;
2853 if (v == SEQ_START_TOKEN)
2854 return netlink_seq_socket_idx(seq, 0);
2856 net = seq_file_net(seq);
2857 iter = seq->private;
2861 } while (s && !nl_table[s->sk_protocol].compare(net, s));
2866 j = iter->hash_idx + 1;
2869 struct nl_portid_hash *hash = &nl_table[i].hash;
2871 for (; j <= hash->mask; j++) {
2872 s = sk_head(&hash->table[j]);
2874 while (s && !nl_table[s->sk_protocol].compare(net, s))
2884 } while (++i < MAX_LINKS);
2889 static void netlink_seq_stop(struct seq_file *seq, void *v)
2890 __releases(nl_table_lock)
2892 read_unlock(&nl_table_lock);
2896 static int netlink_seq_show(struct seq_file *seq, void *v)
2898 if (v == SEQ_START_TOKEN) {
2900 "sk Eth Pid Groups "
2901 "Rmem Wmem Dump Locks Drops Inode\n");
2904 struct netlink_sock *nlk = nlk_sk(s);
2906 seq_printf(seq, "%pK %-3d %-6u %08x %-8d %-8d %d %-8d %-8d %-8lu\n",
2910 nlk->groups ? (u32)nlk->groups[0] : 0,
2911 sk_rmem_alloc_get(s),
2912 sk_wmem_alloc_get(s),
2914 atomic_read(&s->sk_refcnt),
2915 atomic_read(&s->sk_drops),
2923 static const struct seq_operations netlink_seq_ops = {
2924 .start = netlink_seq_start,
2925 .next = netlink_seq_next,
2926 .stop = netlink_seq_stop,
2927 .show = netlink_seq_show,
2931 static int netlink_seq_open(struct inode *inode, struct file *file)
2933 return seq_open_net(inode, file, &netlink_seq_ops,
2934 sizeof(struct nl_seq_iter));
2937 static const struct file_operations netlink_seq_fops = {
2938 .owner = THIS_MODULE,
2939 .open = netlink_seq_open,
2941 .llseek = seq_lseek,
2942 .release = seq_release_net,
2947 int netlink_register_notifier(struct notifier_block *nb)
2949 return atomic_notifier_chain_register(&netlink_chain, nb);
2951 EXPORT_SYMBOL(netlink_register_notifier);
2953 int netlink_unregister_notifier(struct notifier_block *nb)
2955 return atomic_notifier_chain_unregister(&netlink_chain, nb);
2957 EXPORT_SYMBOL(netlink_unregister_notifier);
2959 static const struct proto_ops netlink_ops = {
2960 .family = PF_NETLINK,
2961 .owner = THIS_MODULE,
2962 .release = netlink_release,
2963 .bind = netlink_bind,
2964 .connect = netlink_connect,
2965 .socketpair = sock_no_socketpair,
2966 .accept = sock_no_accept,
2967 .getname = netlink_getname,
2968 .poll = netlink_poll,
2969 .ioctl = sock_no_ioctl,
2970 .listen = sock_no_listen,
2971 .shutdown = sock_no_shutdown,
2972 .setsockopt = netlink_setsockopt,
2973 .getsockopt = netlink_getsockopt,
2974 .sendmsg = netlink_sendmsg,
2975 .recvmsg = netlink_recvmsg,
2976 .mmap = netlink_mmap,
2977 .sendpage = sock_no_sendpage,
2980 static const struct net_proto_family netlink_family_ops = {
2981 .family = PF_NETLINK,
2982 .create = netlink_create,
2983 .owner = THIS_MODULE, /* for consistency 8) */
2986 static int __net_init netlink_net_init(struct net *net)
2988 #ifdef CONFIG_PROC_FS
2989 if (!proc_create("netlink", 0, net->proc_net, &netlink_seq_fops))
2995 static void __net_exit netlink_net_exit(struct net *net)
2997 #ifdef CONFIG_PROC_FS
2998 remove_proc_entry("netlink", net->proc_net);
3002 static void __init netlink_add_usersock_entry(void)
3004 struct listeners *listeners;
3007 listeners = kzalloc(sizeof(*listeners) + NLGRPSZ(groups), GFP_KERNEL);
3009 panic("netlink_add_usersock_entry: Cannot allocate listeners\n");
3011 netlink_table_grab();
3013 nl_table[NETLINK_USERSOCK].groups = groups;
3014 rcu_assign_pointer(nl_table[NETLINK_USERSOCK].listeners, listeners);
3015 nl_table[NETLINK_USERSOCK].module = THIS_MODULE;
3016 nl_table[NETLINK_USERSOCK].registered = 1;
3017 nl_table[NETLINK_USERSOCK].flags = NL_CFG_F_NONROOT_SEND;
3019 netlink_table_ungrab();
3022 static struct pernet_operations __net_initdata netlink_net_ops = {
3023 .init = netlink_net_init,
3024 .exit = netlink_net_exit,
3027 static int __init netlink_proto_init(void)
3030 unsigned long limit;
3032 int err = proto_register(&netlink_proto, 0);
3037 BUILD_BUG_ON(sizeof(struct netlink_skb_parms) > FIELD_SIZEOF(struct sk_buff, cb));
3039 nl_table = kcalloc(MAX_LINKS, sizeof(*nl_table), GFP_KERNEL);
3043 if (totalram_pages >= (128 * 1024))
3044 limit = totalram_pages >> (21 - PAGE_SHIFT);
3046 limit = totalram_pages >> (23 - PAGE_SHIFT);
3048 order = get_bitmask_order(limit) - 1 + PAGE_SHIFT;
3049 limit = (1UL << order) / sizeof(struct hlist_head);
3050 order = get_bitmask_order(min(limit, (unsigned long)UINT_MAX)) - 1;
3052 for (i = 0; i < MAX_LINKS; i++) {
3053 struct nl_portid_hash *hash = &nl_table[i].hash;
3055 hash->table = nl_portid_hash_zalloc(1 * sizeof(*hash->table));
3058 nl_portid_hash_free(nl_table[i].hash.table,
3059 1 * sizeof(*hash->table));
3063 hash->max_shift = order;
3066 hash->rehash_time = jiffies;
3068 nl_table[i].compare = netlink_compare;
3071 INIT_LIST_HEAD(&netlink_tap_all);
3073 netlink_add_usersock_entry();
3075 sock_register(&netlink_family_ops);
3076 register_pernet_subsys(&netlink_net_ops);
3077 /* The netlink device handler may be needed early. */
3082 panic("netlink_init: Cannot allocate nl_table\n");
3085 core_initcall(netlink_proto_init);