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 static 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;
159 int netlink_remove_tap(struct netlink_tap *nt)
163 ret = __netlink_remove_tap(nt);
168 EXPORT_SYMBOL_GPL(netlink_remove_tap);
170 static bool netlink_filter_tap(const struct sk_buff *skb)
172 struct sock *sk = skb->sk;
175 /* We take the more conservative approach and
176 * whitelist socket protocols that may pass.
178 switch (sk->sk_protocol) {
180 case NETLINK_USERSOCK:
181 case NETLINK_SOCK_DIAG:
184 case NETLINK_FIB_LOOKUP:
185 case NETLINK_NETFILTER:
186 case NETLINK_GENERIC:
194 static int __netlink_deliver_tap_skb(struct sk_buff *skb,
195 struct net_device *dev)
197 struct sk_buff *nskb;
198 struct sock *sk = skb->sk;
202 nskb = skb_clone(skb, GFP_ATOMIC);
205 nskb->protocol = htons((u16) sk->sk_protocol);
206 nskb->pkt_type = netlink_is_kernel(sk) ?
207 PACKET_KERNEL : PACKET_USER;
209 ret = dev_queue_xmit(nskb);
210 if (unlikely(ret > 0))
211 ret = net_xmit_errno(ret);
218 static void __netlink_deliver_tap(struct sk_buff *skb)
221 struct netlink_tap *tmp;
223 if (!netlink_filter_tap(skb))
226 list_for_each_entry_rcu(tmp, &netlink_tap_all, list) {
227 ret = __netlink_deliver_tap_skb(skb, tmp->dev);
233 static void netlink_deliver_tap(struct sk_buff *skb)
237 if (unlikely(!list_empty(&netlink_tap_all)))
238 __netlink_deliver_tap(skb);
243 static void netlink_deliver_tap_kernel(struct sock *dst, struct sock *src,
246 if (!(netlink_is_kernel(dst) && netlink_is_kernel(src)))
247 netlink_deliver_tap(skb);
250 static void netlink_overrun(struct sock *sk)
252 struct netlink_sock *nlk = nlk_sk(sk);
254 if (!(nlk->flags & NETLINK_RECV_NO_ENOBUFS)) {
255 if (!test_and_set_bit(NETLINK_CONGESTED, &nlk_sk(sk)->state)) {
256 sk->sk_err = ENOBUFS;
257 sk->sk_error_report(sk);
260 atomic_inc(&sk->sk_drops);
263 static void netlink_rcv_wake(struct sock *sk)
265 struct netlink_sock *nlk = nlk_sk(sk);
267 if (skb_queue_empty(&sk->sk_receive_queue))
268 clear_bit(NETLINK_CONGESTED, &nlk->state);
269 if (!test_bit(NETLINK_CONGESTED, &nlk->state))
270 wake_up_interruptible(&nlk->wait);
273 #ifdef CONFIG_NETLINK_MMAP
274 static bool netlink_skb_is_mmaped(const struct sk_buff *skb)
276 return NETLINK_CB(skb).flags & NETLINK_SKB_MMAPED;
279 static bool netlink_rx_is_mmaped(struct sock *sk)
281 return nlk_sk(sk)->rx_ring.pg_vec != NULL;
284 static bool netlink_tx_is_mmaped(struct sock *sk)
286 return nlk_sk(sk)->tx_ring.pg_vec != NULL;
289 static __pure struct page *pgvec_to_page(const void *addr)
291 if (is_vmalloc_addr(addr))
292 return vmalloc_to_page(addr);
294 return virt_to_page(addr);
297 static void free_pg_vec(void **pg_vec, unsigned int order, unsigned int len)
301 for (i = 0; i < len; i++) {
302 if (pg_vec[i] != NULL) {
303 if (is_vmalloc_addr(pg_vec[i]))
306 free_pages((unsigned long)pg_vec[i], order);
312 static void *alloc_one_pg_vec_page(unsigned long order)
315 gfp_t gfp_flags = GFP_KERNEL | __GFP_COMP | __GFP_ZERO |
316 __GFP_NOWARN | __GFP_NORETRY;
318 buffer = (void *)__get_free_pages(gfp_flags, order);
322 buffer = vzalloc((1 << order) * PAGE_SIZE);
326 gfp_flags &= ~__GFP_NORETRY;
327 return (void *)__get_free_pages(gfp_flags, order);
330 static void **alloc_pg_vec(struct netlink_sock *nlk,
331 struct nl_mmap_req *req, unsigned int order)
333 unsigned int block_nr = req->nm_block_nr;
337 pg_vec = kcalloc(block_nr, sizeof(void *), GFP_KERNEL);
341 for (i = 0; i < block_nr; i++) {
342 pg_vec[i] = alloc_one_pg_vec_page(order);
343 if (pg_vec[i] == NULL)
349 free_pg_vec(pg_vec, order, block_nr);
353 static int netlink_set_ring(struct sock *sk, struct nl_mmap_req *req,
354 bool closing, bool tx_ring)
356 struct netlink_sock *nlk = nlk_sk(sk);
357 struct netlink_ring *ring;
358 struct sk_buff_head *queue;
359 void **pg_vec = NULL;
360 unsigned int order = 0;
363 ring = tx_ring ? &nlk->tx_ring : &nlk->rx_ring;
364 queue = tx_ring ? &sk->sk_write_queue : &sk->sk_receive_queue;
367 if (atomic_read(&nlk->mapped))
369 if (atomic_read(&ring->pending))
373 if (req->nm_block_nr) {
374 if (ring->pg_vec != NULL)
377 if ((int)req->nm_block_size <= 0)
379 if (!IS_ALIGNED(req->nm_block_size, PAGE_SIZE))
381 if (req->nm_frame_size < NL_MMAP_HDRLEN)
383 if (!IS_ALIGNED(req->nm_frame_size, NL_MMAP_MSG_ALIGNMENT))
386 ring->frames_per_block = req->nm_block_size /
388 if (ring->frames_per_block == 0)
390 if (ring->frames_per_block * req->nm_block_nr !=
394 order = get_order(req->nm_block_size);
395 pg_vec = alloc_pg_vec(nlk, req, order);
399 if (req->nm_frame_nr)
404 mutex_lock(&nlk->pg_vec_lock);
405 if (closing || atomic_read(&nlk->mapped) == 0) {
407 spin_lock_bh(&queue->lock);
409 ring->frame_max = req->nm_frame_nr - 1;
411 ring->frame_size = req->nm_frame_size;
412 ring->pg_vec_pages = req->nm_block_size / PAGE_SIZE;
414 swap(ring->pg_vec_len, req->nm_block_nr);
415 swap(ring->pg_vec_order, order);
416 swap(ring->pg_vec, pg_vec);
418 __skb_queue_purge(queue);
419 spin_unlock_bh(&queue->lock);
421 WARN_ON(atomic_read(&nlk->mapped));
423 mutex_unlock(&nlk->pg_vec_lock);
426 free_pg_vec(pg_vec, order, req->nm_block_nr);
430 static void netlink_mm_open(struct vm_area_struct *vma)
432 struct file *file = vma->vm_file;
433 struct socket *sock = file->private_data;
434 struct sock *sk = sock->sk;
437 atomic_inc(&nlk_sk(sk)->mapped);
440 static void netlink_mm_close(struct vm_area_struct *vma)
442 struct file *file = vma->vm_file;
443 struct socket *sock = file->private_data;
444 struct sock *sk = sock->sk;
447 atomic_dec(&nlk_sk(sk)->mapped);
450 static const struct vm_operations_struct netlink_mmap_ops = {
451 .open = netlink_mm_open,
452 .close = netlink_mm_close,
455 static int netlink_mmap(struct file *file, struct socket *sock,
456 struct vm_area_struct *vma)
458 struct sock *sk = sock->sk;
459 struct netlink_sock *nlk = nlk_sk(sk);
460 struct netlink_ring *ring;
461 unsigned long start, size, expected;
468 mutex_lock(&nlk->pg_vec_lock);
471 for (ring = &nlk->rx_ring; ring <= &nlk->tx_ring; ring++) {
472 if (ring->pg_vec == NULL)
474 expected += ring->pg_vec_len * ring->pg_vec_pages * PAGE_SIZE;
480 size = vma->vm_end - vma->vm_start;
481 if (size != expected)
484 start = vma->vm_start;
485 for (ring = &nlk->rx_ring; ring <= &nlk->tx_ring; ring++) {
486 if (ring->pg_vec == NULL)
489 for (i = 0; i < ring->pg_vec_len; i++) {
491 void *kaddr = ring->pg_vec[i];
494 for (pg_num = 0; pg_num < ring->pg_vec_pages; pg_num++) {
495 page = pgvec_to_page(kaddr);
496 err = vm_insert_page(vma, start, page);
505 atomic_inc(&nlk->mapped);
506 vma->vm_ops = &netlink_mmap_ops;
509 mutex_unlock(&nlk->pg_vec_lock);
513 static void netlink_frame_flush_dcache(const struct nl_mmap_hdr *hdr)
515 #if ARCH_IMPLEMENTS_FLUSH_DCACHE_PAGE == 1
516 struct page *p_start, *p_end;
518 /* First page is flushed through netlink_{get,set}_status */
519 p_start = pgvec_to_page(hdr + PAGE_SIZE);
520 p_end = pgvec_to_page((void *)hdr + NL_MMAP_HDRLEN + hdr->nm_len - 1);
521 while (p_start <= p_end) {
522 flush_dcache_page(p_start);
528 static enum nl_mmap_status netlink_get_status(const struct nl_mmap_hdr *hdr)
531 flush_dcache_page(pgvec_to_page(hdr));
532 return hdr->nm_status;
535 static void netlink_set_status(struct nl_mmap_hdr *hdr,
536 enum nl_mmap_status status)
538 hdr->nm_status = status;
539 flush_dcache_page(pgvec_to_page(hdr));
543 static struct nl_mmap_hdr *
544 __netlink_lookup_frame(const struct netlink_ring *ring, unsigned int pos)
546 unsigned int pg_vec_pos, frame_off;
548 pg_vec_pos = pos / ring->frames_per_block;
549 frame_off = pos % ring->frames_per_block;
551 return ring->pg_vec[pg_vec_pos] + (frame_off * ring->frame_size);
554 static struct nl_mmap_hdr *
555 netlink_lookup_frame(const struct netlink_ring *ring, unsigned int pos,
556 enum nl_mmap_status status)
558 struct nl_mmap_hdr *hdr;
560 hdr = __netlink_lookup_frame(ring, pos);
561 if (netlink_get_status(hdr) != status)
567 static struct nl_mmap_hdr *
568 netlink_current_frame(const struct netlink_ring *ring,
569 enum nl_mmap_status status)
571 return netlink_lookup_frame(ring, ring->head, status);
574 static struct nl_mmap_hdr *
575 netlink_previous_frame(const struct netlink_ring *ring,
576 enum nl_mmap_status status)
580 prev = ring->head ? ring->head - 1 : ring->frame_max;
581 return netlink_lookup_frame(ring, prev, status);
584 static void netlink_increment_head(struct netlink_ring *ring)
586 ring->head = ring->head != ring->frame_max ? ring->head + 1 : 0;
589 static void netlink_forward_ring(struct netlink_ring *ring)
591 unsigned int head = ring->head, pos = head;
592 const struct nl_mmap_hdr *hdr;
595 hdr = __netlink_lookup_frame(ring, pos);
596 if (hdr->nm_status == NL_MMAP_STATUS_UNUSED)
598 if (hdr->nm_status != NL_MMAP_STATUS_SKIP)
600 netlink_increment_head(ring);
601 } while (ring->head != head);
604 static bool netlink_dump_space(struct netlink_sock *nlk)
606 struct netlink_ring *ring = &nlk->rx_ring;
607 struct nl_mmap_hdr *hdr;
610 hdr = netlink_current_frame(ring, NL_MMAP_STATUS_UNUSED);
614 n = ring->head + ring->frame_max / 2;
615 if (n > ring->frame_max)
616 n -= ring->frame_max;
618 hdr = __netlink_lookup_frame(ring, n);
620 return hdr->nm_status == NL_MMAP_STATUS_UNUSED;
623 static unsigned int netlink_poll(struct file *file, struct socket *sock,
626 struct sock *sk = sock->sk;
627 struct netlink_sock *nlk = nlk_sk(sk);
631 if (nlk->rx_ring.pg_vec != NULL) {
632 /* Memory mapped sockets don't call recvmsg(), so flow control
633 * for dumps is performed here. A dump is allowed to continue
634 * if at least half the ring is unused.
636 while (nlk->cb_running && netlink_dump_space(nlk)) {
637 err = netlink_dump(sk);
640 sk->sk_error_report(sk);
644 netlink_rcv_wake(sk);
647 mask = datagram_poll(file, sock, wait);
649 spin_lock_bh(&sk->sk_receive_queue.lock);
650 if (nlk->rx_ring.pg_vec) {
651 netlink_forward_ring(&nlk->rx_ring);
652 if (!netlink_previous_frame(&nlk->rx_ring, NL_MMAP_STATUS_UNUSED))
653 mask |= POLLIN | POLLRDNORM;
655 spin_unlock_bh(&sk->sk_receive_queue.lock);
657 spin_lock_bh(&sk->sk_write_queue.lock);
658 if (nlk->tx_ring.pg_vec) {
659 if (netlink_current_frame(&nlk->tx_ring, NL_MMAP_STATUS_UNUSED))
660 mask |= POLLOUT | POLLWRNORM;
662 spin_unlock_bh(&sk->sk_write_queue.lock);
667 static struct nl_mmap_hdr *netlink_mmap_hdr(struct sk_buff *skb)
669 return (struct nl_mmap_hdr *)(skb->head - NL_MMAP_HDRLEN);
672 static void netlink_ring_setup_skb(struct sk_buff *skb, struct sock *sk,
673 struct netlink_ring *ring,
674 struct nl_mmap_hdr *hdr)
679 size = ring->frame_size - NL_MMAP_HDRLEN;
680 data = (void *)hdr + NL_MMAP_HDRLEN;
684 skb_reset_tail_pointer(skb);
685 skb->end = skb->tail + size;
688 skb->destructor = netlink_skb_destructor;
689 NETLINK_CB(skb).flags |= NETLINK_SKB_MMAPED;
690 NETLINK_CB(skb).sk = sk;
693 static int netlink_mmap_sendmsg(struct sock *sk, struct msghdr *msg,
694 u32 dst_portid, u32 dst_group,
695 struct sock_iocb *siocb)
697 struct netlink_sock *nlk = nlk_sk(sk);
698 struct netlink_ring *ring;
699 struct nl_mmap_hdr *hdr;
703 int err = 0, len = 0;
705 /* Netlink messages are validated by the receiver before processing.
706 * In order to avoid userspace changing the contents of the message
707 * after validation, the socket and the ring may only be used by a
708 * single process, otherwise we fall back to copying.
710 if (atomic_long_read(&sk->sk_socket->file->f_count) > 2 ||
711 atomic_read(&nlk->mapped) > 1)
714 mutex_lock(&nlk->pg_vec_lock);
716 ring = &nlk->tx_ring;
717 maxlen = ring->frame_size - NL_MMAP_HDRLEN;
720 hdr = netlink_current_frame(ring, NL_MMAP_STATUS_VALID);
722 if (!(msg->msg_flags & MSG_DONTWAIT) &&
723 atomic_read(&nlk->tx_ring.pending))
727 if (hdr->nm_len > maxlen) {
732 netlink_frame_flush_dcache(hdr);
734 if (likely(dst_portid == 0 && dst_group == 0 && excl)) {
735 skb = alloc_skb_head(GFP_KERNEL);
741 netlink_ring_setup_skb(skb, sk, ring, hdr);
742 NETLINK_CB(skb).flags |= NETLINK_SKB_TX;
743 __skb_put(skb, hdr->nm_len);
744 netlink_set_status(hdr, NL_MMAP_STATUS_RESERVED);
745 atomic_inc(&ring->pending);
747 skb = alloc_skb(hdr->nm_len, GFP_KERNEL);
752 __skb_put(skb, hdr->nm_len);
753 memcpy(skb->data, (void *)hdr + NL_MMAP_HDRLEN, hdr->nm_len);
754 netlink_set_status(hdr, NL_MMAP_STATUS_UNUSED);
757 netlink_increment_head(ring);
759 NETLINK_CB(skb).portid = nlk->portid;
760 NETLINK_CB(skb).dst_group = dst_group;
761 NETLINK_CB(skb).creds = siocb->scm->creds;
763 err = security_netlink_send(sk, skb);
769 if (unlikely(dst_group)) {
770 atomic_inc(&skb->users);
771 netlink_broadcast(sk, skb, dst_portid, dst_group,
774 err = netlink_unicast(sk, skb, dst_portid,
775 msg->msg_flags & MSG_DONTWAIT);
780 } while (hdr != NULL ||
781 (!(msg->msg_flags & MSG_DONTWAIT) &&
782 atomic_read(&nlk->tx_ring.pending)));
787 mutex_unlock(&nlk->pg_vec_lock);
791 static void netlink_queue_mmaped_skb(struct sock *sk, struct sk_buff *skb)
793 struct nl_mmap_hdr *hdr;
795 hdr = netlink_mmap_hdr(skb);
796 hdr->nm_len = skb->len;
797 hdr->nm_group = NETLINK_CB(skb).dst_group;
798 hdr->nm_pid = NETLINK_CB(skb).creds.pid;
799 hdr->nm_uid = from_kuid(sk_user_ns(sk), NETLINK_CB(skb).creds.uid);
800 hdr->nm_gid = from_kgid(sk_user_ns(sk), NETLINK_CB(skb).creds.gid);
801 netlink_frame_flush_dcache(hdr);
802 netlink_set_status(hdr, NL_MMAP_STATUS_VALID);
804 NETLINK_CB(skb).flags |= NETLINK_SKB_DELIVERED;
808 static void netlink_ring_set_copied(struct sock *sk, struct sk_buff *skb)
810 struct netlink_sock *nlk = nlk_sk(sk);
811 struct netlink_ring *ring = &nlk->rx_ring;
812 struct nl_mmap_hdr *hdr;
814 spin_lock_bh(&sk->sk_receive_queue.lock);
815 hdr = netlink_current_frame(ring, NL_MMAP_STATUS_UNUSED);
817 spin_unlock_bh(&sk->sk_receive_queue.lock);
822 netlink_increment_head(ring);
823 __skb_queue_tail(&sk->sk_receive_queue, skb);
824 spin_unlock_bh(&sk->sk_receive_queue.lock);
826 hdr->nm_len = skb->len;
827 hdr->nm_group = NETLINK_CB(skb).dst_group;
828 hdr->nm_pid = NETLINK_CB(skb).creds.pid;
829 hdr->nm_uid = from_kuid(sk_user_ns(sk), NETLINK_CB(skb).creds.uid);
830 hdr->nm_gid = from_kgid(sk_user_ns(sk), NETLINK_CB(skb).creds.gid);
831 netlink_set_status(hdr, NL_MMAP_STATUS_COPY);
834 #else /* CONFIG_NETLINK_MMAP */
835 #define netlink_skb_is_mmaped(skb) false
836 #define netlink_rx_is_mmaped(sk) false
837 #define netlink_tx_is_mmaped(sk) false
838 #define netlink_mmap sock_no_mmap
839 #define netlink_poll datagram_poll
840 #define netlink_mmap_sendmsg(sk, msg, dst_portid, dst_group, siocb) 0
841 #endif /* CONFIG_NETLINK_MMAP */
843 static void netlink_skb_destructor(struct sk_buff *skb)
845 #ifdef CONFIG_NETLINK_MMAP
846 struct nl_mmap_hdr *hdr;
847 struct netlink_ring *ring;
850 /* If a packet from the kernel to userspace was freed because of an
851 * error without being delivered to userspace, the kernel must reset
852 * the status. In the direction userspace to kernel, the status is
853 * always reset here after the packet was processed and freed.
855 if (netlink_skb_is_mmaped(skb)) {
856 hdr = netlink_mmap_hdr(skb);
857 sk = NETLINK_CB(skb).sk;
859 if (NETLINK_CB(skb).flags & NETLINK_SKB_TX) {
860 netlink_set_status(hdr, NL_MMAP_STATUS_UNUSED);
861 ring = &nlk_sk(sk)->tx_ring;
863 if (!(NETLINK_CB(skb).flags & NETLINK_SKB_DELIVERED)) {
865 netlink_set_status(hdr, NL_MMAP_STATUS_VALID);
867 ring = &nlk_sk(sk)->rx_ring;
870 WARN_ON(atomic_read(&ring->pending) == 0);
871 atomic_dec(&ring->pending);
877 if (is_vmalloc_addr(skb->head)) {
879 !atomic_dec_return(&(skb_shinfo(skb)->dataref)))
888 static void netlink_skb_set_owner_r(struct sk_buff *skb, struct sock *sk)
890 WARN_ON(skb->sk != NULL);
892 skb->destructor = netlink_skb_destructor;
893 atomic_add(skb->truesize, &sk->sk_rmem_alloc);
894 sk_mem_charge(sk, skb->truesize);
897 static void netlink_sock_destruct(struct sock *sk)
899 struct netlink_sock *nlk = nlk_sk(sk);
901 if (nlk->cb_running) {
903 nlk->cb.done(&nlk->cb);
905 module_put(nlk->cb.module);
906 kfree_skb(nlk->cb.skb);
909 skb_queue_purge(&sk->sk_receive_queue);
910 #ifdef CONFIG_NETLINK_MMAP
912 struct nl_mmap_req req;
914 memset(&req, 0, sizeof(req));
915 if (nlk->rx_ring.pg_vec)
916 netlink_set_ring(sk, &req, true, false);
917 memset(&req, 0, sizeof(req));
918 if (nlk->tx_ring.pg_vec)
919 netlink_set_ring(sk, &req, true, true);
921 #endif /* CONFIG_NETLINK_MMAP */
923 if (!sock_flag(sk, SOCK_DEAD)) {
924 printk(KERN_ERR "Freeing alive netlink socket %p\n", sk);
928 WARN_ON(atomic_read(&sk->sk_rmem_alloc));
929 WARN_ON(atomic_read(&sk->sk_wmem_alloc));
930 WARN_ON(nlk_sk(sk)->groups);
933 /* This lock without WQ_FLAG_EXCLUSIVE is good on UP and it is _very_ bad on
934 * SMP. Look, when several writers sleep and reader wakes them up, all but one
935 * immediately hit write lock and grab all the cpus. Exclusive sleep solves
936 * this, _but_ remember, it adds useless work on UP machines.
939 void netlink_table_grab(void)
940 __acquires(nl_table_lock)
944 write_lock_irq(&nl_table_lock);
946 if (atomic_read(&nl_table_users)) {
947 DECLARE_WAITQUEUE(wait, current);
949 add_wait_queue_exclusive(&nl_table_wait, &wait);
951 set_current_state(TASK_UNINTERRUPTIBLE);
952 if (atomic_read(&nl_table_users) == 0)
954 write_unlock_irq(&nl_table_lock);
956 write_lock_irq(&nl_table_lock);
959 __set_current_state(TASK_RUNNING);
960 remove_wait_queue(&nl_table_wait, &wait);
964 void netlink_table_ungrab(void)
965 __releases(nl_table_lock)
967 write_unlock_irq(&nl_table_lock);
968 wake_up(&nl_table_wait);
972 netlink_lock_table(void)
974 /* read_lock() synchronizes us to netlink_table_grab */
976 read_lock(&nl_table_lock);
977 atomic_inc(&nl_table_users);
978 read_unlock(&nl_table_lock);
982 netlink_unlock_table(void)
984 if (atomic_dec_and_test(&nl_table_users))
985 wake_up(&nl_table_wait);
988 static bool netlink_compare(struct net *net, struct sock *sk)
990 return net_eq(sock_net(sk), net);
993 static struct sock *netlink_lookup(struct net *net, int protocol, u32 portid)
995 struct netlink_table *table = &nl_table[protocol];
996 struct nl_portid_hash *hash = &table->hash;
997 struct hlist_head *head;
1000 read_lock(&nl_table_lock);
1001 head = nl_portid_hashfn(hash, portid);
1002 sk_for_each(sk, head) {
1003 if (table->compare(net, sk) &&
1004 (nlk_sk(sk)->portid == portid)) {
1011 read_unlock(&nl_table_lock);
1015 static struct hlist_head *nl_portid_hash_zalloc(size_t size)
1017 if (size <= PAGE_SIZE)
1018 return kzalloc(size, GFP_ATOMIC);
1020 return (struct hlist_head *)
1021 __get_free_pages(GFP_ATOMIC | __GFP_ZERO,
1025 static void nl_portid_hash_free(struct hlist_head *table, size_t size)
1027 if (size <= PAGE_SIZE)
1030 free_pages((unsigned long)table, get_order(size));
1033 static int nl_portid_hash_rehash(struct nl_portid_hash *hash, int grow)
1035 unsigned int omask, mask, shift;
1037 struct hlist_head *otable, *table;
1040 omask = mask = hash->mask;
1041 osize = size = (mask + 1) * sizeof(*table);
1042 shift = hash->shift;
1045 if (++shift > hash->max_shift)
1047 mask = mask * 2 + 1;
1051 table = nl_portid_hash_zalloc(size);
1055 otable = hash->table;
1056 hash->table = table;
1058 hash->shift = shift;
1059 get_random_bytes(&hash->rnd, sizeof(hash->rnd));
1061 for (i = 0; i <= omask; i++) {
1063 struct hlist_node *tmp;
1065 sk_for_each_safe(sk, tmp, &otable[i])
1066 __sk_add_node(sk, nl_portid_hashfn(hash, nlk_sk(sk)->portid));
1069 nl_portid_hash_free(otable, osize);
1070 hash->rehash_time = jiffies + 10 * 60 * HZ;
1074 static inline int nl_portid_hash_dilute(struct nl_portid_hash *hash, int len)
1076 int avg = hash->entries >> hash->shift;
1078 if (unlikely(avg > 1) && nl_portid_hash_rehash(hash, 1))
1081 if (unlikely(len > avg) && time_after(jiffies, hash->rehash_time)) {
1082 nl_portid_hash_rehash(hash, 0);
1089 static const struct proto_ops netlink_ops;
1092 netlink_update_listeners(struct sock *sk)
1094 struct netlink_table *tbl = &nl_table[sk->sk_protocol];
1097 struct listeners *listeners;
1099 listeners = nl_deref_protected(tbl->listeners);
1103 for (i = 0; i < NLGRPLONGS(tbl->groups); i++) {
1105 sk_for_each_bound(sk, &tbl->mc_list) {
1106 if (i < NLGRPLONGS(nlk_sk(sk)->ngroups))
1107 mask |= nlk_sk(sk)->groups[i];
1109 listeners->masks[i] = mask;
1111 /* this function is only called with the netlink table "grabbed", which
1112 * makes sure updates are visible before bind or setsockopt return. */
1115 static int netlink_insert(struct sock *sk, struct net *net, u32 portid)
1117 struct netlink_table *table = &nl_table[sk->sk_protocol];
1118 struct nl_portid_hash *hash = &table->hash;
1119 struct hlist_head *head;
1120 int err = -EADDRINUSE;
1124 netlink_table_grab();
1125 head = nl_portid_hashfn(hash, portid);
1127 sk_for_each(osk, head) {
1128 if (table->compare(net, osk) &&
1129 (nlk_sk(osk)->portid == portid))
1137 if (nlk_sk(sk)->portid)
1141 if (BITS_PER_LONG > 32 && unlikely(hash->entries >= UINT_MAX))
1144 if (len && nl_portid_hash_dilute(hash, len))
1145 head = nl_portid_hashfn(hash, portid);
1147 nlk_sk(sk)->portid = portid;
1148 sk_add_node(sk, head);
1152 netlink_table_ungrab();
1156 static void netlink_remove(struct sock *sk)
1158 netlink_table_grab();
1159 if (sk_del_node_init(sk))
1160 nl_table[sk->sk_protocol].hash.entries--;
1161 if (nlk_sk(sk)->subscriptions)
1162 __sk_del_bind_node(sk);
1163 netlink_table_ungrab();
1166 static struct proto netlink_proto = {
1168 .owner = THIS_MODULE,
1169 .obj_size = sizeof(struct netlink_sock),
1172 static int __netlink_create(struct net *net, struct socket *sock,
1173 struct mutex *cb_mutex, int protocol)
1176 struct netlink_sock *nlk;
1178 sock->ops = &netlink_ops;
1180 sk = sk_alloc(net, PF_NETLINK, GFP_KERNEL, &netlink_proto);
1184 sock_init_data(sock, sk);
1188 nlk->cb_mutex = cb_mutex;
1190 nlk->cb_mutex = &nlk->cb_def_mutex;
1191 mutex_init(nlk->cb_mutex);
1193 init_waitqueue_head(&nlk->wait);
1194 #ifdef CONFIG_NETLINK_MMAP
1195 mutex_init(&nlk->pg_vec_lock);
1198 sk->sk_destruct = netlink_sock_destruct;
1199 sk->sk_protocol = protocol;
1203 static int netlink_create(struct net *net, struct socket *sock, int protocol,
1206 struct module *module = NULL;
1207 struct mutex *cb_mutex;
1208 struct netlink_sock *nlk;
1209 int (*bind)(int group);
1210 void (*unbind)(int group);
1213 sock->state = SS_UNCONNECTED;
1215 if (sock->type != SOCK_RAW && sock->type != SOCK_DGRAM)
1216 return -ESOCKTNOSUPPORT;
1218 if (protocol < 0 || protocol >= MAX_LINKS)
1219 return -EPROTONOSUPPORT;
1221 netlink_lock_table();
1222 #ifdef CONFIG_MODULES
1223 if (!nl_table[protocol].registered) {
1224 netlink_unlock_table();
1225 request_module("net-pf-%d-proto-%d", PF_NETLINK, protocol);
1226 netlink_lock_table();
1229 if (nl_table[protocol].registered &&
1230 try_module_get(nl_table[protocol].module))
1231 module = nl_table[protocol].module;
1233 err = -EPROTONOSUPPORT;
1234 cb_mutex = nl_table[protocol].cb_mutex;
1235 bind = nl_table[protocol].bind;
1236 unbind = nl_table[protocol].unbind;
1237 netlink_unlock_table();
1242 err = __netlink_create(net, sock, cb_mutex, protocol);
1247 sock_prot_inuse_add(net, &netlink_proto, 1);
1250 nlk = nlk_sk(sock->sk);
1251 nlk->module = module;
1252 nlk->netlink_bind = bind;
1253 nlk->netlink_unbind = unbind;
1262 static int netlink_release(struct socket *sock)
1264 struct sock *sk = sock->sk;
1265 struct netlink_sock *nlk;
1275 * OK. Socket is unlinked, any packets that arrive now
1280 wake_up_interruptible_all(&nlk->wait);
1282 skb_queue_purge(&sk->sk_write_queue);
1285 struct netlink_notify n = {
1286 .net = sock_net(sk),
1287 .protocol = sk->sk_protocol,
1288 .portid = nlk->portid,
1290 atomic_notifier_call_chain(&netlink_chain,
1291 NETLINK_URELEASE, &n);
1294 module_put(nlk->module);
1296 netlink_table_grab();
1297 if (netlink_is_kernel(sk)) {
1298 BUG_ON(nl_table[sk->sk_protocol].registered == 0);
1299 if (--nl_table[sk->sk_protocol].registered == 0) {
1300 struct listeners *old;
1302 old = nl_deref_protected(nl_table[sk->sk_protocol].listeners);
1303 RCU_INIT_POINTER(nl_table[sk->sk_protocol].listeners, NULL);
1304 kfree_rcu(old, rcu);
1305 nl_table[sk->sk_protocol].module = NULL;
1306 nl_table[sk->sk_protocol].bind = NULL;
1307 nl_table[sk->sk_protocol].unbind = NULL;
1308 nl_table[sk->sk_protocol].flags = 0;
1309 nl_table[sk->sk_protocol].registered = 0;
1311 } else if (nlk->subscriptions) {
1312 netlink_update_listeners(sk);
1314 netlink_table_ungrab();
1320 sock_prot_inuse_add(sock_net(sk), &netlink_proto, -1);
1326 static int netlink_autobind(struct socket *sock)
1328 struct sock *sk = sock->sk;
1329 struct net *net = sock_net(sk);
1330 struct netlink_table *table = &nl_table[sk->sk_protocol];
1331 struct nl_portid_hash *hash = &table->hash;
1332 struct hlist_head *head;
1334 s32 portid = task_tgid_vnr(current);
1336 static s32 rover = -4097;
1340 netlink_table_grab();
1341 head = nl_portid_hashfn(hash, portid);
1342 sk_for_each(osk, head) {
1343 if (!table->compare(net, osk))
1345 if (nlk_sk(osk)->portid == portid) {
1346 /* Bind collision, search negative portid values. */
1350 netlink_table_ungrab();
1354 netlink_table_ungrab();
1356 err = netlink_insert(sk, net, portid);
1357 if (err == -EADDRINUSE)
1360 /* If 2 threads race to autobind, that is fine. */
1368 * __netlink_ns_capable - General netlink message capability test
1369 * @nsp: NETLINK_CB of the socket buffer holding a netlink command from userspace.
1370 * @user_ns: The user namespace of the capability to use
1371 * @cap: The capability to use
1373 * Test to see if the opener of the socket we received the message
1374 * from had when the netlink socket was created and the sender of the
1375 * message has has the capability @cap in the user namespace @user_ns.
1377 bool __netlink_ns_capable(const struct netlink_skb_parms *nsp,
1378 struct user_namespace *user_ns, int cap)
1380 return ((nsp->flags & NETLINK_SKB_DST) ||
1381 file_ns_capable(nsp->sk->sk_socket->file, user_ns, cap)) &&
1382 ns_capable(user_ns, cap);
1384 EXPORT_SYMBOL(__netlink_ns_capable);
1387 * netlink_ns_capable - General netlink message capability test
1388 * @skb: socket buffer holding a netlink command from userspace
1389 * @user_ns: The user namespace of the capability to use
1390 * @cap: The capability to use
1392 * Test to see if the opener of the socket we received the message
1393 * from had when the netlink socket was created and the sender of the
1394 * message has has the capability @cap in the user namespace @user_ns.
1396 bool netlink_ns_capable(const struct sk_buff *skb,
1397 struct user_namespace *user_ns, int cap)
1399 return __netlink_ns_capable(&NETLINK_CB(skb), user_ns, cap);
1401 EXPORT_SYMBOL(netlink_ns_capable);
1404 * netlink_capable - Netlink global message capability test
1405 * @skb: socket buffer holding a netlink command from userspace
1406 * @cap: The capability to use
1408 * Test to see if the opener of the socket we received the message
1409 * from had when the netlink socket was created and the sender of the
1410 * message has has the capability @cap in all user namespaces.
1412 bool netlink_capable(const struct sk_buff *skb, int cap)
1414 return netlink_ns_capable(skb, &init_user_ns, cap);
1416 EXPORT_SYMBOL(netlink_capable);
1419 * netlink_net_capable - Netlink network namespace message capability test
1420 * @skb: socket buffer holding a netlink command from userspace
1421 * @cap: The capability to use
1423 * Test to see if the opener of the socket we received the message
1424 * from had when the netlink socket was created and the sender of the
1425 * message has has the capability @cap over the network namespace of
1426 * the socket we received the message from.
1428 bool netlink_net_capable(const struct sk_buff *skb, int cap)
1430 return netlink_ns_capable(skb, sock_net(skb->sk)->user_ns, cap);
1432 EXPORT_SYMBOL(netlink_net_capable);
1434 static inline int netlink_allowed(const struct socket *sock, unsigned int flag)
1436 return (nl_table[sock->sk->sk_protocol].flags & flag) ||
1437 ns_capable(sock_net(sock->sk)->user_ns, CAP_NET_ADMIN);
1441 netlink_update_subscriptions(struct sock *sk, unsigned int subscriptions)
1443 struct netlink_sock *nlk = nlk_sk(sk);
1445 if (nlk->subscriptions && !subscriptions)
1446 __sk_del_bind_node(sk);
1447 else if (!nlk->subscriptions && subscriptions)
1448 sk_add_bind_node(sk, &nl_table[sk->sk_protocol].mc_list);
1449 nlk->subscriptions = subscriptions;
1452 static int netlink_realloc_groups(struct sock *sk)
1454 struct netlink_sock *nlk = nlk_sk(sk);
1455 unsigned int groups;
1456 unsigned long *new_groups;
1459 netlink_table_grab();
1461 groups = nl_table[sk->sk_protocol].groups;
1462 if (!nl_table[sk->sk_protocol].registered) {
1467 if (nlk->ngroups >= groups)
1470 new_groups = krealloc(nlk->groups, NLGRPSZ(groups), GFP_ATOMIC);
1471 if (new_groups == NULL) {
1475 memset((char *)new_groups + NLGRPSZ(nlk->ngroups), 0,
1476 NLGRPSZ(groups) - NLGRPSZ(nlk->ngroups));
1478 nlk->groups = new_groups;
1479 nlk->ngroups = groups;
1481 netlink_table_ungrab();
1485 static void netlink_unbind(int group, long unsigned int groups,
1486 struct netlink_sock *nlk)
1490 if (!nlk->netlink_unbind)
1493 for (undo = 0; undo < group; undo++)
1494 if (test_bit(group, &groups))
1495 nlk->netlink_unbind(undo);
1498 static int netlink_bind(struct socket *sock, struct sockaddr *addr,
1501 struct sock *sk = sock->sk;
1502 struct net *net = sock_net(sk);
1503 struct netlink_sock *nlk = nlk_sk(sk);
1504 struct sockaddr_nl *nladdr = (struct sockaddr_nl *)addr;
1506 long unsigned int groups = nladdr->nl_groups;
1508 if (addr_len < sizeof(struct sockaddr_nl))
1511 if (nladdr->nl_family != AF_NETLINK)
1514 /* Only superuser is allowed to listen multicasts */
1516 if (!netlink_allowed(sock, NL_CFG_F_NONROOT_RECV))
1518 err = netlink_realloc_groups(sk);
1524 if (nladdr->nl_pid != nlk->portid)
1527 if (nlk->netlink_bind && groups) {
1530 for (group = 0; group < nlk->ngroups; group++) {
1531 if (!test_bit(group, &groups))
1533 err = nlk->netlink_bind(group);
1536 netlink_unbind(group, groups, nlk);
1542 err = nladdr->nl_pid ?
1543 netlink_insert(sk, net, nladdr->nl_pid) :
1544 netlink_autobind(sock);
1546 netlink_unbind(nlk->ngroups - 1, groups, nlk);
1551 if (!groups && (nlk->groups == NULL || !(u32)nlk->groups[0]))
1554 netlink_table_grab();
1555 netlink_update_subscriptions(sk, nlk->subscriptions +
1557 hweight32(nlk->groups[0]));
1558 nlk->groups[0] = (nlk->groups[0] & ~0xffffffffUL) | groups;
1559 netlink_update_listeners(sk);
1560 netlink_table_ungrab();
1565 static int netlink_connect(struct socket *sock, struct sockaddr *addr,
1566 int alen, int flags)
1569 struct sock *sk = sock->sk;
1570 struct netlink_sock *nlk = nlk_sk(sk);
1571 struct sockaddr_nl *nladdr = (struct sockaddr_nl *)addr;
1573 if (alen < sizeof(addr->sa_family))
1576 if (addr->sa_family == AF_UNSPEC) {
1577 sk->sk_state = NETLINK_UNCONNECTED;
1578 nlk->dst_portid = 0;
1582 if (addr->sa_family != AF_NETLINK)
1585 if ((nladdr->nl_groups || nladdr->nl_pid) &&
1586 !netlink_allowed(sock, NL_CFG_F_NONROOT_SEND))
1590 err = netlink_autobind(sock);
1593 sk->sk_state = NETLINK_CONNECTED;
1594 nlk->dst_portid = nladdr->nl_pid;
1595 nlk->dst_group = ffs(nladdr->nl_groups);
1601 static int netlink_getname(struct socket *sock, struct sockaddr *addr,
1602 int *addr_len, int peer)
1604 struct sock *sk = sock->sk;
1605 struct netlink_sock *nlk = nlk_sk(sk);
1606 DECLARE_SOCKADDR(struct sockaddr_nl *, nladdr, addr);
1608 nladdr->nl_family = AF_NETLINK;
1610 *addr_len = sizeof(*nladdr);
1613 nladdr->nl_pid = nlk->dst_portid;
1614 nladdr->nl_groups = netlink_group_mask(nlk->dst_group);
1616 nladdr->nl_pid = nlk->portid;
1617 nladdr->nl_groups = nlk->groups ? nlk->groups[0] : 0;
1622 static struct sock *netlink_getsockbyportid(struct sock *ssk, u32 portid)
1625 struct netlink_sock *nlk;
1627 sock = netlink_lookup(sock_net(ssk), ssk->sk_protocol, portid);
1629 return ERR_PTR(-ECONNREFUSED);
1631 /* Don't bother queuing skb if kernel socket has no input function */
1633 if (sock->sk_state == NETLINK_CONNECTED &&
1634 nlk->dst_portid != nlk_sk(ssk)->portid) {
1636 return ERR_PTR(-ECONNREFUSED);
1641 struct sock *netlink_getsockbyfilp(struct file *filp)
1643 struct inode *inode = file_inode(filp);
1646 if (!S_ISSOCK(inode->i_mode))
1647 return ERR_PTR(-ENOTSOCK);
1649 sock = SOCKET_I(inode)->sk;
1650 if (sock->sk_family != AF_NETLINK)
1651 return ERR_PTR(-EINVAL);
1657 static struct sk_buff *netlink_alloc_large_skb(unsigned int size,
1660 struct sk_buff *skb;
1663 if (size <= NLMSG_GOODSIZE || broadcast)
1664 return alloc_skb(size, GFP_KERNEL);
1666 size = SKB_DATA_ALIGN(size) +
1667 SKB_DATA_ALIGN(sizeof(struct skb_shared_info));
1669 data = vmalloc(size);
1673 skb = build_skb(data, size);
1678 skb->destructor = netlink_skb_destructor;
1685 * Attach a skb to a netlink socket.
1686 * The caller must hold a reference to the destination socket. On error, the
1687 * reference is dropped. The skb is not send to the destination, just all
1688 * all error checks are performed and memory in the queue is reserved.
1690 * < 0: error. skb freed, reference to sock dropped.
1692 * 1: repeat lookup - reference dropped while waiting for socket memory.
1694 int netlink_attachskb(struct sock *sk, struct sk_buff *skb,
1695 long *timeo, struct sock *ssk)
1697 struct netlink_sock *nlk;
1701 if ((atomic_read(&sk->sk_rmem_alloc) > sk->sk_rcvbuf ||
1702 test_bit(NETLINK_CONGESTED, &nlk->state)) &&
1703 !netlink_skb_is_mmaped(skb)) {
1704 DECLARE_WAITQUEUE(wait, current);
1706 if (!ssk || netlink_is_kernel(ssk))
1707 netlink_overrun(sk);
1713 __set_current_state(TASK_INTERRUPTIBLE);
1714 add_wait_queue(&nlk->wait, &wait);
1716 if ((atomic_read(&sk->sk_rmem_alloc) > sk->sk_rcvbuf ||
1717 test_bit(NETLINK_CONGESTED, &nlk->state)) &&
1718 !sock_flag(sk, SOCK_DEAD))
1719 *timeo = schedule_timeout(*timeo);
1721 __set_current_state(TASK_RUNNING);
1722 remove_wait_queue(&nlk->wait, &wait);
1725 if (signal_pending(current)) {
1727 return sock_intr_errno(*timeo);
1731 netlink_skb_set_owner_r(skb, sk);
1735 static int __netlink_sendskb(struct sock *sk, struct sk_buff *skb)
1739 netlink_deliver_tap(skb);
1741 #ifdef CONFIG_NETLINK_MMAP
1742 if (netlink_skb_is_mmaped(skb))
1743 netlink_queue_mmaped_skb(sk, skb);
1744 else if (netlink_rx_is_mmaped(sk))
1745 netlink_ring_set_copied(sk, skb);
1747 #endif /* CONFIG_NETLINK_MMAP */
1748 skb_queue_tail(&sk->sk_receive_queue, skb);
1749 sk->sk_data_ready(sk);
1753 int netlink_sendskb(struct sock *sk, struct sk_buff *skb)
1755 int len = __netlink_sendskb(sk, skb);
1761 void netlink_detachskb(struct sock *sk, struct sk_buff *skb)
1767 static struct sk_buff *netlink_trim(struct sk_buff *skb, gfp_t allocation)
1771 WARN_ON(skb->sk != NULL);
1772 if (netlink_skb_is_mmaped(skb))
1775 delta = skb->end - skb->tail;
1776 if (is_vmalloc_addr(skb->head) || delta * 2 < skb->truesize)
1779 if (skb_shared(skb)) {
1780 struct sk_buff *nskb = skb_clone(skb, allocation);
1787 if (!pskb_expand_head(skb, 0, -delta, allocation))
1788 skb->truesize -= delta;
1793 static int netlink_unicast_kernel(struct sock *sk, struct sk_buff *skb,
1797 struct netlink_sock *nlk = nlk_sk(sk);
1799 ret = -ECONNREFUSED;
1800 if (nlk->netlink_rcv != NULL) {
1802 netlink_skb_set_owner_r(skb, sk);
1803 NETLINK_CB(skb).sk = ssk;
1804 netlink_deliver_tap_kernel(sk, ssk, skb);
1805 nlk->netlink_rcv(skb);
1814 int netlink_unicast(struct sock *ssk, struct sk_buff *skb,
1815 u32 portid, int nonblock)
1821 skb = netlink_trim(skb, gfp_any());
1823 timeo = sock_sndtimeo(ssk, nonblock);
1825 sk = netlink_getsockbyportid(ssk, portid);
1830 if (netlink_is_kernel(sk))
1831 return netlink_unicast_kernel(sk, skb, ssk);
1833 if (sk_filter(sk, skb)) {
1840 err = netlink_attachskb(sk, skb, &timeo, ssk);
1846 return netlink_sendskb(sk, skb);
1848 EXPORT_SYMBOL(netlink_unicast);
1850 struct sk_buff *netlink_alloc_skb(struct sock *ssk, unsigned int size,
1851 u32 dst_portid, gfp_t gfp_mask)
1853 #ifdef CONFIG_NETLINK_MMAP
1854 struct sock *sk = NULL;
1855 struct sk_buff *skb;
1856 struct netlink_ring *ring;
1857 struct nl_mmap_hdr *hdr;
1858 unsigned int maxlen;
1860 sk = netlink_getsockbyportid(ssk, dst_portid);
1864 ring = &nlk_sk(sk)->rx_ring;
1865 /* fast-path without atomic ops for common case: non-mmaped receiver */
1866 if (ring->pg_vec == NULL)
1869 if (ring->frame_size - NL_MMAP_HDRLEN < size)
1872 skb = alloc_skb_head(gfp_mask);
1876 spin_lock_bh(&sk->sk_receive_queue.lock);
1877 /* check again under lock */
1878 if (ring->pg_vec == NULL)
1881 /* check again under lock */
1882 maxlen = ring->frame_size - NL_MMAP_HDRLEN;
1886 netlink_forward_ring(ring);
1887 hdr = netlink_current_frame(ring, NL_MMAP_STATUS_UNUSED);
1890 netlink_ring_setup_skb(skb, sk, ring, hdr);
1891 netlink_set_status(hdr, NL_MMAP_STATUS_RESERVED);
1892 atomic_inc(&ring->pending);
1893 netlink_increment_head(ring);
1895 spin_unlock_bh(&sk->sk_receive_queue.lock);
1900 spin_unlock_bh(&sk->sk_receive_queue.lock);
1901 netlink_overrun(sk);
1908 spin_unlock_bh(&sk->sk_receive_queue.lock);
1913 return alloc_skb(size, gfp_mask);
1915 EXPORT_SYMBOL_GPL(netlink_alloc_skb);
1917 int netlink_has_listeners(struct sock *sk, unsigned int group)
1920 struct listeners *listeners;
1922 BUG_ON(!netlink_is_kernel(sk));
1925 listeners = rcu_dereference(nl_table[sk->sk_protocol].listeners);
1927 if (listeners && group - 1 < nl_table[sk->sk_protocol].groups)
1928 res = test_bit(group - 1, listeners->masks);
1934 EXPORT_SYMBOL_GPL(netlink_has_listeners);
1936 static int netlink_broadcast_deliver(struct sock *sk, struct sk_buff *skb)
1938 struct netlink_sock *nlk = nlk_sk(sk);
1940 if (atomic_read(&sk->sk_rmem_alloc) <= sk->sk_rcvbuf &&
1941 !test_bit(NETLINK_CONGESTED, &nlk->state)) {
1942 netlink_skb_set_owner_r(skb, sk);
1943 __netlink_sendskb(sk, skb);
1944 return atomic_read(&sk->sk_rmem_alloc) > (sk->sk_rcvbuf >> 1);
1949 struct netlink_broadcast_data {
1950 struct sock *exclude_sk;
1955 int delivery_failure;
1959 struct sk_buff *skb, *skb2;
1960 int (*tx_filter)(struct sock *dsk, struct sk_buff *skb, void *data);
1964 static int do_one_broadcast(struct sock *sk,
1965 struct netlink_broadcast_data *p)
1967 struct netlink_sock *nlk = nlk_sk(sk);
1970 if (p->exclude_sk == sk)
1973 if (nlk->portid == p->portid || p->group - 1 >= nlk->ngroups ||
1974 !test_bit(p->group - 1, nlk->groups))
1977 if (!net_eq(sock_net(sk), p->net))
1981 netlink_overrun(sk);
1986 if (p->skb2 == NULL) {
1987 if (skb_shared(p->skb)) {
1988 p->skb2 = skb_clone(p->skb, p->allocation);
1990 p->skb2 = skb_get(p->skb);
1992 * skb ownership may have been set when
1993 * delivered to a previous socket.
1995 skb_orphan(p->skb2);
1998 if (p->skb2 == NULL) {
1999 netlink_overrun(sk);
2000 /* Clone failed. Notify ALL listeners. */
2002 if (nlk->flags & NETLINK_BROADCAST_SEND_ERROR)
2003 p->delivery_failure = 1;
2004 } else if (p->tx_filter && p->tx_filter(sk, p->skb2, p->tx_data)) {
2007 } else if (sk_filter(sk, p->skb2)) {
2010 } else if ((val = netlink_broadcast_deliver(sk, p->skb2)) < 0) {
2011 netlink_overrun(sk);
2012 if (nlk->flags & NETLINK_BROADCAST_SEND_ERROR)
2013 p->delivery_failure = 1;
2015 p->congested |= val;
2025 int netlink_broadcast_filtered(struct sock *ssk, struct sk_buff *skb, u32 portid,
2026 u32 group, gfp_t allocation,
2027 int (*filter)(struct sock *dsk, struct sk_buff *skb, void *data),
2030 struct net *net = sock_net(ssk);
2031 struct netlink_broadcast_data info;
2034 skb = netlink_trim(skb, allocation);
2036 info.exclude_sk = ssk;
2038 info.portid = portid;
2041 info.delivery_failure = 0;
2044 info.allocation = allocation;
2047 info.tx_filter = filter;
2048 info.tx_data = filter_data;
2050 /* While we sleep in clone, do not allow to change socket list */
2052 netlink_lock_table();
2054 sk_for_each_bound(sk, &nl_table[ssk->sk_protocol].mc_list)
2055 do_one_broadcast(sk, &info);
2059 netlink_unlock_table();
2061 if (info.delivery_failure) {
2062 kfree_skb(info.skb2);
2065 consume_skb(info.skb2);
2067 if (info.delivered) {
2068 if (info.congested && (allocation & __GFP_WAIT))
2074 EXPORT_SYMBOL(netlink_broadcast_filtered);
2076 int netlink_broadcast(struct sock *ssk, struct sk_buff *skb, u32 portid,
2077 u32 group, gfp_t allocation)
2079 return netlink_broadcast_filtered(ssk, skb, portid, group, allocation,
2082 EXPORT_SYMBOL(netlink_broadcast);
2084 struct netlink_set_err_data {
2085 struct sock *exclude_sk;
2091 static int do_one_set_err(struct sock *sk, struct netlink_set_err_data *p)
2093 struct netlink_sock *nlk = nlk_sk(sk);
2096 if (sk == p->exclude_sk)
2099 if (!net_eq(sock_net(sk), sock_net(p->exclude_sk)))
2102 if (nlk->portid == p->portid || p->group - 1 >= nlk->ngroups ||
2103 !test_bit(p->group - 1, nlk->groups))
2106 if (p->code == ENOBUFS && nlk->flags & NETLINK_RECV_NO_ENOBUFS) {
2111 sk->sk_err = p->code;
2112 sk->sk_error_report(sk);
2118 * netlink_set_err - report error to broadcast listeners
2119 * @ssk: the kernel netlink socket, as returned by netlink_kernel_create()
2120 * @portid: the PORTID of a process that we want to skip (if any)
2121 * @group: the broadcast group that will notice the error
2122 * @code: error code, must be negative (as usual in kernelspace)
2124 * This function returns the number of broadcast listeners that have set the
2125 * NETLINK_RECV_NO_ENOBUFS socket option.
2127 int netlink_set_err(struct sock *ssk, u32 portid, u32 group, int code)
2129 struct netlink_set_err_data info;
2133 info.exclude_sk = ssk;
2134 info.portid = portid;
2136 /* sk->sk_err wants a positive error value */
2139 read_lock(&nl_table_lock);
2141 sk_for_each_bound(sk, &nl_table[ssk->sk_protocol].mc_list)
2142 ret += do_one_set_err(sk, &info);
2144 read_unlock(&nl_table_lock);
2147 EXPORT_SYMBOL(netlink_set_err);
2149 /* must be called with netlink table grabbed */
2150 static void netlink_update_socket_mc(struct netlink_sock *nlk,
2154 int old, new = !!is_new, subscriptions;
2156 old = test_bit(group - 1, nlk->groups);
2157 subscriptions = nlk->subscriptions - old + new;
2159 __set_bit(group - 1, nlk->groups);
2161 __clear_bit(group - 1, nlk->groups);
2162 netlink_update_subscriptions(&nlk->sk, subscriptions);
2163 netlink_update_listeners(&nlk->sk);
2166 static int netlink_setsockopt(struct socket *sock, int level, int optname,
2167 char __user *optval, unsigned int optlen)
2169 struct sock *sk = sock->sk;
2170 struct netlink_sock *nlk = nlk_sk(sk);
2171 unsigned int val = 0;
2174 if (level != SOL_NETLINK)
2175 return -ENOPROTOOPT;
2177 if (optname != NETLINK_RX_RING && optname != NETLINK_TX_RING &&
2178 optlen >= sizeof(int) &&
2179 get_user(val, (unsigned int __user *)optval))
2183 case NETLINK_PKTINFO:
2185 nlk->flags |= NETLINK_RECV_PKTINFO;
2187 nlk->flags &= ~NETLINK_RECV_PKTINFO;
2190 case NETLINK_ADD_MEMBERSHIP:
2191 case NETLINK_DROP_MEMBERSHIP: {
2192 if (!netlink_allowed(sock, NL_CFG_F_NONROOT_RECV))
2194 err = netlink_realloc_groups(sk);
2197 if (!val || val - 1 >= nlk->ngroups)
2199 if (optname == NETLINK_ADD_MEMBERSHIP && nlk->netlink_bind) {
2200 err = nlk->netlink_bind(val);
2204 netlink_table_grab();
2205 netlink_update_socket_mc(nlk, val,
2206 optname == NETLINK_ADD_MEMBERSHIP);
2207 netlink_table_ungrab();
2208 if (optname == NETLINK_DROP_MEMBERSHIP && nlk->netlink_unbind)
2209 nlk->netlink_unbind(val);
2214 case NETLINK_BROADCAST_ERROR:
2216 nlk->flags |= NETLINK_BROADCAST_SEND_ERROR;
2218 nlk->flags &= ~NETLINK_BROADCAST_SEND_ERROR;
2221 case NETLINK_NO_ENOBUFS:
2223 nlk->flags |= NETLINK_RECV_NO_ENOBUFS;
2224 clear_bit(NETLINK_CONGESTED, &nlk->state);
2225 wake_up_interruptible(&nlk->wait);
2227 nlk->flags &= ~NETLINK_RECV_NO_ENOBUFS;
2231 #ifdef CONFIG_NETLINK_MMAP
2232 case NETLINK_RX_RING:
2233 case NETLINK_TX_RING: {
2234 struct nl_mmap_req req;
2236 /* Rings might consume more memory than queue limits, require
2239 if (!capable(CAP_NET_ADMIN))
2241 if (optlen < sizeof(req))
2243 if (copy_from_user(&req, optval, sizeof(req)))
2245 err = netlink_set_ring(sk, &req, false,
2246 optname == NETLINK_TX_RING);
2249 #endif /* CONFIG_NETLINK_MMAP */
2256 static int netlink_getsockopt(struct socket *sock, int level, int optname,
2257 char __user *optval, int __user *optlen)
2259 struct sock *sk = sock->sk;
2260 struct netlink_sock *nlk = nlk_sk(sk);
2263 if (level != SOL_NETLINK)
2264 return -ENOPROTOOPT;
2266 if (get_user(len, optlen))
2272 case NETLINK_PKTINFO:
2273 if (len < sizeof(int))
2276 val = nlk->flags & NETLINK_RECV_PKTINFO ? 1 : 0;
2277 if (put_user(len, optlen) ||
2278 put_user(val, optval))
2282 case NETLINK_BROADCAST_ERROR:
2283 if (len < sizeof(int))
2286 val = nlk->flags & NETLINK_BROADCAST_SEND_ERROR ? 1 : 0;
2287 if (put_user(len, optlen) ||
2288 put_user(val, optval))
2292 case NETLINK_NO_ENOBUFS:
2293 if (len < sizeof(int))
2296 val = nlk->flags & NETLINK_RECV_NO_ENOBUFS ? 1 : 0;
2297 if (put_user(len, optlen) ||
2298 put_user(val, optval))
2308 static void netlink_cmsg_recv_pktinfo(struct msghdr *msg, struct sk_buff *skb)
2310 struct nl_pktinfo info;
2312 info.group = NETLINK_CB(skb).dst_group;
2313 put_cmsg(msg, SOL_NETLINK, NETLINK_PKTINFO, sizeof(info), &info);
2316 static int netlink_sendmsg(struct kiocb *kiocb, struct socket *sock,
2317 struct msghdr *msg, size_t len)
2319 struct sock_iocb *siocb = kiocb_to_siocb(kiocb);
2320 struct sock *sk = sock->sk;
2321 struct netlink_sock *nlk = nlk_sk(sk);
2322 DECLARE_SOCKADDR(struct sockaddr_nl *, addr, msg->msg_name);
2325 struct sk_buff *skb;
2327 struct scm_cookie scm;
2328 u32 netlink_skb_flags = 0;
2330 if (msg->msg_flags&MSG_OOB)
2333 if (NULL == siocb->scm)
2336 err = scm_send(sock, msg, siocb->scm, true);
2340 if (msg->msg_namelen) {
2342 if (addr->nl_family != AF_NETLINK)
2344 dst_portid = addr->nl_pid;
2345 dst_group = ffs(addr->nl_groups);
2347 if ((dst_group || dst_portid) &&
2348 !netlink_allowed(sock, NL_CFG_F_NONROOT_SEND))
2350 netlink_skb_flags |= NETLINK_SKB_DST;
2352 dst_portid = nlk->dst_portid;
2353 dst_group = nlk->dst_group;
2357 err = netlink_autobind(sock);
2362 if (netlink_tx_is_mmaped(sk) &&
2363 msg->msg_iov->iov_base == NULL) {
2364 err = netlink_mmap_sendmsg(sk, msg, dst_portid, dst_group,
2370 if (len > sk->sk_sndbuf - 32)
2373 skb = netlink_alloc_large_skb(len, dst_group);
2377 NETLINK_CB(skb).portid = nlk->portid;
2378 NETLINK_CB(skb).dst_group = dst_group;
2379 NETLINK_CB(skb).creds = siocb->scm->creds;
2380 NETLINK_CB(skb).flags = netlink_skb_flags;
2383 if (memcpy_fromiovec(skb_put(skb, len), msg->msg_iov, len)) {
2388 err = security_netlink_send(sk, skb);
2395 atomic_inc(&skb->users);
2396 netlink_broadcast(sk, skb, dst_portid, dst_group, GFP_KERNEL);
2398 err = netlink_unicast(sk, skb, dst_portid, msg->msg_flags&MSG_DONTWAIT);
2401 scm_destroy(siocb->scm);
2405 static int netlink_recvmsg(struct kiocb *kiocb, struct socket *sock,
2406 struct msghdr *msg, size_t len,
2409 struct sock_iocb *siocb = kiocb_to_siocb(kiocb);
2410 struct scm_cookie scm;
2411 struct sock *sk = sock->sk;
2412 struct netlink_sock *nlk = nlk_sk(sk);
2413 int noblock = flags&MSG_DONTWAIT;
2415 struct sk_buff *skb, *data_skb;
2423 skb = skb_recv_datagram(sk, flags, noblock, &err);
2429 #ifdef CONFIG_COMPAT_NETLINK_MESSAGES
2430 if (unlikely(skb_shinfo(skb)->frag_list)) {
2432 * If this skb has a frag_list, then here that means that we
2433 * will have to use the frag_list skb's data for compat tasks
2434 * and the regular skb's data for normal (non-compat) tasks.
2436 * If we need to send the compat skb, assign it to the
2437 * 'data_skb' variable so that it will be used below for data
2438 * copying. We keep 'skb' for everything else, including
2439 * freeing both later.
2441 if (flags & MSG_CMSG_COMPAT)
2442 data_skb = skb_shinfo(skb)->frag_list;
2446 /* Record the max length of recvmsg() calls for future allocations */
2447 nlk->max_recvmsg_len = max(nlk->max_recvmsg_len, len);
2448 nlk->max_recvmsg_len = min_t(size_t, nlk->max_recvmsg_len,
2451 copied = data_skb->len;
2453 msg->msg_flags |= MSG_TRUNC;
2457 skb_reset_transport_header(data_skb);
2458 err = skb_copy_datagram_iovec(data_skb, 0, msg->msg_iov, copied);
2460 if (msg->msg_name) {
2461 DECLARE_SOCKADDR(struct sockaddr_nl *, addr, msg->msg_name);
2462 addr->nl_family = AF_NETLINK;
2464 addr->nl_pid = NETLINK_CB(skb).portid;
2465 addr->nl_groups = netlink_group_mask(NETLINK_CB(skb).dst_group);
2466 msg->msg_namelen = sizeof(*addr);
2469 if (nlk->flags & NETLINK_RECV_PKTINFO)
2470 netlink_cmsg_recv_pktinfo(msg, skb);
2472 if (NULL == siocb->scm) {
2473 memset(&scm, 0, sizeof(scm));
2476 siocb->scm->creds = *NETLINK_CREDS(skb);
2477 if (flags & MSG_TRUNC)
2478 copied = data_skb->len;
2480 skb_free_datagram(sk, skb);
2482 if (nlk->cb_running &&
2483 atomic_read(&sk->sk_rmem_alloc) <= sk->sk_rcvbuf / 2) {
2484 ret = netlink_dump(sk);
2487 sk->sk_error_report(sk);
2491 scm_recv(sock, msg, siocb->scm, flags);
2493 netlink_rcv_wake(sk);
2494 return err ? : copied;
2497 static void netlink_data_ready(struct sock *sk)
2503 * We export these functions to other modules. They provide a
2504 * complete set of kernel non-blocking support for message
2509 __netlink_kernel_create(struct net *net, int unit, struct module *module,
2510 struct netlink_kernel_cfg *cfg)
2512 struct socket *sock;
2514 struct netlink_sock *nlk;
2515 struct listeners *listeners = NULL;
2516 struct mutex *cb_mutex = cfg ? cfg->cb_mutex : NULL;
2517 unsigned int groups;
2521 if (unit < 0 || unit >= MAX_LINKS)
2524 if (sock_create_lite(PF_NETLINK, SOCK_DGRAM, unit, &sock))
2528 * We have to just have a reference on the net from sk, but don't
2529 * get_net it. Besides, we cannot get and then put the net here.
2530 * So we create one inside init_net and the move it to net.
2533 if (__netlink_create(&init_net, sock, cb_mutex, unit) < 0)
2534 goto out_sock_release_nosk;
2537 sk_change_net(sk, net);
2539 if (!cfg || cfg->groups < 32)
2542 groups = cfg->groups;
2544 listeners = kzalloc(sizeof(*listeners) + NLGRPSZ(groups), GFP_KERNEL);
2546 goto out_sock_release;
2548 sk->sk_data_ready = netlink_data_ready;
2549 if (cfg && cfg->input)
2550 nlk_sk(sk)->netlink_rcv = cfg->input;
2552 if (netlink_insert(sk, net, 0))
2553 goto out_sock_release;
2556 nlk->flags |= NETLINK_KERNEL_SOCKET;
2558 netlink_table_grab();
2559 if (!nl_table[unit].registered) {
2560 nl_table[unit].groups = groups;
2561 rcu_assign_pointer(nl_table[unit].listeners, listeners);
2562 nl_table[unit].cb_mutex = cb_mutex;
2563 nl_table[unit].module = module;
2565 nl_table[unit].bind = cfg->bind;
2566 nl_table[unit].flags = cfg->flags;
2568 nl_table[unit].compare = cfg->compare;
2570 nl_table[unit].registered = 1;
2573 nl_table[unit].registered++;
2575 netlink_table_ungrab();
2580 netlink_kernel_release(sk);
2583 out_sock_release_nosk:
2587 EXPORT_SYMBOL(__netlink_kernel_create);
2590 netlink_kernel_release(struct sock *sk)
2592 sk_release_kernel(sk);
2594 EXPORT_SYMBOL(netlink_kernel_release);
2596 int __netlink_change_ngroups(struct sock *sk, unsigned int groups)
2598 struct listeners *new, *old;
2599 struct netlink_table *tbl = &nl_table[sk->sk_protocol];
2604 if (NLGRPSZ(tbl->groups) < NLGRPSZ(groups)) {
2605 new = kzalloc(sizeof(*new) + NLGRPSZ(groups), GFP_ATOMIC);
2608 old = nl_deref_protected(tbl->listeners);
2609 memcpy(new->masks, old->masks, NLGRPSZ(tbl->groups));
2610 rcu_assign_pointer(tbl->listeners, new);
2612 kfree_rcu(old, rcu);
2614 tbl->groups = groups;
2620 * netlink_change_ngroups - change number of multicast groups
2622 * This changes the number of multicast groups that are available
2623 * on a certain netlink family. Note that it is not possible to
2624 * change the number of groups to below 32. Also note that it does
2625 * not implicitly call netlink_clear_multicast_users() when the
2626 * number of groups is reduced.
2628 * @sk: The kernel netlink socket, as returned by netlink_kernel_create().
2629 * @groups: The new number of groups.
2631 int netlink_change_ngroups(struct sock *sk, unsigned int groups)
2635 netlink_table_grab();
2636 err = __netlink_change_ngroups(sk, groups);
2637 netlink_table_ungrab();
2642 void __netlink_clear_multicast_users(struct sock *ksk, unsigned int group)
2645 struct netlink_table *tbl = &nl_table[ksk->sk_protocol];
2647 sk_for_each_bound(sk, &tbl->mc_list)
2648 netlink_update_socket_mc(nlk_sk(sk), group, 0);
2652 __nlmsg_put(struct sk_buff *skb, u32 portid, u32 seq, int type, int len, int flags)
2654 struct nlmsghdr *nlh;
2655 int size = nlmsg_msg_size(len);
2657 nlh = (struct nlmsghdr *)skb_put(skb, NLMSG_ALIGN(size));
2658 nlh->nlmsg_type = type;
2659 nlh->nlmsg_len = size;
2660 nlh->nlmsg_flags = flags;
2661 nlh->nlmsg_pid = portid;
2662 nlh->nlmsg_seq = seq;
2663 if (!__builtin_constant_p(size) || NLMSG_ALIGN(size) - size != 0)
2664 memset(nlmsg_data(nlh) + len, 0, NLMSG_ALIGN(size) - size);
2667 EXPORT_SYMBOL(__nlmsg_put);
2670 * It looks a bit ugly.
2671 * It would be better to create kernel thread.
2674 static int netlink_dump(struct sock *sk)
2676 struct netlink_sock *nlk = nlk_sk(sk);
2677 struct netlink_callback *cb;
2678 struct sk_buff *skb = NULL;
2679 struct nlmsghdr *nlh;
2680 int len, err = -ENOBUFS;
2683 mutex_lock(nlk->cb_mutex);
2684 if (!nlk->cb_running) {
2690 alloc_size = max_t(int, cb->min_dump_alloc, NLMSG_GOODSIZE);
2692 if (!netlink_rx_is_mmaped(sk) &&
2693 atomic_read(&sk->sk_rmem_alloc) >= sk->sk_rcvbuf)
2696 /* NLMSG_GOODSIZE is small to avoid high order allocations being
2697 * required, but it makes sense to _attempt_ a 16K bytes allocation
2698 * to reduce number of system calls on dump operations, if user
2699 * ever provided a big enough buffer.
2701 if (alloc_size < nlk->max_recvmsg_len) {
2702 skb = netlink_alloc_skb(sk,
2703 nlk->max_recvmsg_len,
2708 /* available room should be exact amount to avoid MSG_TRUNC */
2710 skb_reserve(skb, skb_tailroom(skb) -
2711 nlk->max_recvmsg_len);
2714 skb = netlink_alloc_skb(sk, alloc_size, nlk->portid,
2718 netlink_skb_set_owner_r(skb, sk);
2720 len = cb->dump(skb, cb);
2723 mutex_unlock(nlk->cb_mutex);
2725 if (sk_filter(sk, skb))
2728 __netlink_sendskb(sk, skb);
2732 nlh = nlmsg_put_answer(skb, cb, NLMSG_DONE, sizeof(len), NLM_F_MULTI);
2736 nl_dump_check_consistent(cb, nlh);
2738 memcpy(nlmsg_data(nlh), &len, sizeof(len));
2740 if (sk_filter(sk, skb))
2743 __netlink_sendskb(sk, skb);
2748 nlk->cb_running = false;
2749 mutex_unlock(nlk->cb_mutex);
2750 module_put(cb->module);
2751 consume_skb(cb->skb);
2755 mutex_unlock(nlk->cb_mutex);
2760 int __netlink_dump_start(struct sock *ssk, struct sk_buff *skb,
2761 const struct nlmsghdr *nlh,
2762 struct netlink_dump_control *control)
2764 struct netlink_callback *cb;
2766 struct netlink_sock *nlk;
2769 /* Memory mapped dump requests need to be copied to avoid looping
2770 * on the pending state in netlink_mmap_sendmsg() while the CB hold
2771 * a reference to the skb.
2773 if (netlink_skb_is_mmaped(skb)) {
2774 skb = skb_copy(skb, GFP_KERNEL);
2778 atomic_inc(&skb->users);
2780 sk = netlink_lookup(sock_net(ssk), ssk->sk_protocol, NETLINK_CB(skb).portid);
2782 ret = -ECONNREFUSED;
2787 mutex_lock(nlk->cb_mutex);
2788 /* A dump is in progress... */
2789 if (nlk->cb_running) {
2793 /* add reference of module which cb->dump belongs to */
2794 if (!try_module_get(control->module)) {
2795 ret = -EPROTONOSUPPORT;
2800 memset(cb, 0, sizeof(*cb));
2801 cb->dump = control->dump;
2802 cb->done = control->done;
2804 cb->data = control->data;
2805 cb->module = control->module;
2806 cb->min_dump_alloc = control->min_dump_alloc;
2809 nlk->cb_running = true;
2811 mutex_unlock(nlk->cb_mutex);
2813 ret = netlink_dump(sk);
2819 /* We successfully started a dump, by returning -EINTR we
2820 * signal not to send ACK even if it was requested.
2826 mutex_unlock(nlk->cb_mutex);
2831 EXPORT_SYMBOL(__netlink_dump_start);
2833 void netlink_ack(struct sk_buff *in_skb, struct nlmsghdr *nlh, int err)
2835 struct sk_buff *skb;
2836 struct nlmsghdr *rep;
2837 struct nlmsgerr *errmsg;
2838 size_t payload = sizeof(*errmsg);
2840 /* error messages get the original request appened */
2842 payload += nlmsg_len(nlh);
2844 skb = netlink_alloc_skb(in_skb->sk, nlmsg_total_size(payload),
2845 NETLINK_CB(in_skb).portid, GFP_KERNEL);
2849 sk = netlink_lookup(sock_net(in_skb->sk),
2850 in_skb->sk->sk_protocol,
2851 NETLINK_CB(in_skb).portid);
2853 sk->sk_err = ENOBUFS;
2854 sk->sk_error_report(sk);
2860 rep = __nlmsg_put(skb, NETLINK_CB(in_skb).portid, nlh->nlmsg_seq,
2861 NLMSG_ERROR, payload, 0);
2862 errmsg = nlmsg_data(rep);
2863 errmsg->error = err;
2864 memcpy(&errmsg->msg, nlh, err ? nlh->nlmsg_len : sizeof(*nlh));
2865 netlink_unicast(in_skb->sk, skb, NETLINK_CB(in_skb).portid, MSG_DONTWAIT);
2867 EXPORT_SYMBOL(netlink_ack);
2869 int netlink_rcv_skb(struct sk_buff *skb, int (*cb)(struct sk_buff *,
2872 struct nlmsghdr *nlh;
2875 while (skb->len >= nlmsg_total_size(0)) {
2878 nlh = nlmsg_hdr(skb);
2881 if (nlh->nlmsg_len < NLMSG_HDRLEN || skb->len < nlh->nlmsg_len)
2884 /* Only requests are handled by the kernel */
2885 if (!(nlh->nlmsg_flags & NLM_F_REQUEST))
2888 /* Skip control messages */
2889 if (nlh->nlmsg_type < NLMSG_MIN_TYPE)
2897 if (nlh->nlmsg_flags & NLM_F_ACK || err)
2898 netlink_ack(skb, nlh, err);
2901 msglen = NLMSG_ALIGN(nlh->nlmsg_len);
2902 if (msglen > skb->len)
2904 skb_pull(skb, msglen);
2909 EXPORT_SYMBOL(netlink_rcv_skb);
2912 * nlmsg_notify - send a notification netlink message
2913 * @sk: netlink socket to use
2914 * @skb: notification message
2915 * @portid: destination netlink portid for reports or 0
2916 * @group: destination multicast group or 0
2917 * @report: 1 to report back, 0 to disable
2918 * @flags: allocation flags
2920 int nlmsg_notify(struct sock *sk, struct sk_buff *skb, u32 portid,
2921 unsigned int group, int report, gfp_t flags)
2926 int exclude_portid = 0;
2929 atomic_inc(&skb->users);
2930 exclude_portid = portid;
2933 /* errors reported via destination sk->sk_err, but propagate
2934 * delivery errors if NETLINK_BROADCAST_ERROR flag is set */
2935 err = nlmsg_multicast(sk, skb, exclude_portid, group, flags);
2941 err2 = nlmsg_unicast(sk, skb, portid);
2942 if (!err || err == -ESRCH)
2948 EXPORT_SYMBOL(nlmsg_notify);
2950 #ifdef CONFIG_PROC_FS
2951 struct nl_seq_iter {
2952 struct seq_net_private p;
2957 static struct sock *netlink_seq_socket_idx(struct seq_file *seq, loff_t pos)
2959 struct nl_seq_iter *iter = seq->private;
2964 for (i = 0; i < MAX_LINKS; i++) {
2965 struct nl_portid_hash *hash = &nl_table[i].hash;
2967 for (j = 0; j <= hash->mask; j++) {
2968 sk_for_each(s, &hash->table[j]) {
2969 if (sock_net(s) != seq_file_net(seq))
2983 static void *netlink_seq_start(struct seq_file *seq, loff_t *pos)
2984 __acquires(nl_table_lock)
2986 read_lock(&nl_table_lock);
2987 return *pos ? netlink_seq_socket_idx(seq, *pos - 1) : SEQ_START_TOKEN;
2990 static void *netlink_seq_next(struct seq_file *seq, void *v, loff_t *pos)
2993 struct nl_seq_iter *iter;
2999 if (v == SEQ_START_TOKEN)
3000 return netlink_seq_socket_idx(seq, 0);
3002 net = seq_file_net(seq);
3003 iter = seq->private;
3007 } while (s && !nl_table[s->sk_protocol].compare(net, s));
3012 j = iter->hash_idx + 1;
3015 struct nl_portid_hash *hash = &nl_table[i].hash;
3017 for (; j <= hash->mask; j++) {
3018 s = sk_head(&hash->table[j]);
3020 while (s && !nl_table[s->sk_protocol].compare(net, s))
3030 } while (++i < MAX_LINKS);
3035 static void netlink_seq_stop(struct seq_file *seq, void *v)
3036 __releases(nl_table_lock)
3038 read_unlock(&nl_table_lock);
3042 static int netlink_seq_show(struct seq_file *seq, void *v)
3044 if (v == SEQ_START_TOKEN) {
3046 "sk Eth Pid Groups "
3047 "Rmem Wmem Dump Locks Drops Inode\n");
3050 struct netlink_sock *nlk = nlk_sk(s);
3052 seq_printf(seq, "%pK %-3d %-6u %08x %-8d %-8d %d %-8d %-8d %-8lu\n",
3056 nlk->groups ? (u32)nlk->groups[0] : 0,
3057 sk_rmem_alloc_get(s),
3058 sk_wmem_alloc_get(s),
3060 atomic_read(&s->sk_refcnt),
3061 atomic_read(&s->sk_drops),
3069 static const struct seq_operations netlink_seq_ops = {
3070 .start = netlink_seq_start,
3071 .next = netlink_seq_next,
3072 .stop = netlink_seq_stop,
3073 .show = netlink_seq_show,
3077 static int netlink_seq_open(struct inode *inode, struct file *file)
3079 return seq_open_net(inode, file, &netlink_seq_ops,
3080 sizeof(struct nl_seq_iter));
3083 static const struct file_operations netlink_seq_fops = {
3084 .owner = THIS_MODULE,
3085 .open = netlink_seq_open,
3087 .llseek = seq_lseek,
3088 .release = seq_release_net,
3093 int netlink_register_notifier(struct notifier_block *nb)
3095 return atomic_notifier_chain_register(&netlink_chain, nb);
3097 EXPORT_SYMBOL(netlink_register_notifier);
3099 int netlink_unregister_notifier(struct notifier_block *nb)
3101 return atomic_notifier_chain_unregister(&netlink_chain, nb);
3103 EXPORT_SYMBOL(netlink_unregister_notifier);
3105 static const struct proto_ops netlink_ops = {
3106 .family = PF_NETLINK,
3107 .owner = THIS_MODULE,
3108 .release = netlink_release,
3109 .bind = netlink_bind,
3110 .connect = netlink_connect,
3111 .socketpair = sock_no_socketpair,
3112 .accept = sock_no_accept,
3113 .getname = netlink_getname,
3114 .poll = netlink_poll,
3115 .ioctl = sock_no_ioctl,
3116 .listen = sock_no_listen,
3117 .shutdown = sock_no_shutdown,
3118 .setsockopt = netlink_setsockopt,
3119 .getsockopt = netlink_getsockopt,
3120 .sendmsg = netlink_sendmsg,
3121 .recvmsg = netlink_recvmsg,
3122 .mmap = netlink_mmap,
3123 .sendpage = sock_no_sendpage,
3126 static const struct net_proto_family netlink_family_ops = {
3127 .family = PF_NETLINK,
3128 .create = netlink_create,
3129 .owner = THIS_MODULE, /* for consistency 8) */
3132 static int __net_init netlink_net_init(struct net *net)
3134 #ifdef CONFIG_PROC_FS
3135 if (!proc_create("netlink", 0, net->proc_net, &netlink_seq_fops))
3141 static void __net_exit netlink_net_exit(struct net *net)
3143 #ifdef CONFIG_PROC_FS
3144 remove_proc_entry("netlink", net->proc_net);
3148 static void __init netlink_add_usersock_entry(void)
3150 struct listeners *listeners;
3153 listeners = kzalloc(sizeof(*listeners) + NLGRPSZ(groups), GFP_KERNEL);
3155 panic("netlink_add_usersock_entry: Cannot allocate listeners\n");
3157 netlink_table_grab();
3159 nl_table[NETLINK_USERSOCK].groups = groups;
3160 rcu_assign_pointer(nl_table[NETLINK_USERSOCK].listeners, listeners);
3161 nl_table[NETLINK_USERSOCK].module = THIS_MODULE;
3162 nl_table[NETLINK_USERSOCK].registered = 1;
3163 nl_table[NETLINK_USERSOCK].flags = NL_CFG_F_NONROOT_SEND;
3165 netlink_table_ungrab();
3168 static struct pernet_operations __net_initdata netlink_net_ops = {
3169 .init = netlink_net_init,
3170 .exit = netlink_net_exit,
3173 static int __init netlink_proto_init(void)
3176 unsigned long limit;
3178 int err = proto_register(&netlink_proto, 0);
3183 BUILD_BUG_ON(sizeof(struct netlink_skb_parms) > FIELD_SIZEOF(struct sk_buff, cb));
3185 nl_table = kcalloc(MAX_LINKS, sizeof(*nl_table), GFP_KERNEL);
3189 if (totalram_pages >= (128 * 1024))
3190 limit = totalram_pages >> (21 - PAGE_SHIFT);
3192 limit = totalram_pages >> (23 - PAGE_SHIFT);
3194 order = get_bitmask_order(limit) - 1 + PAGE_SHIFT;
3195 limit = (1UL << order) / sizeof(struct hlist_head);
3196 order = get_bitmask_order(min(limit, (unsigned long)UINT_MAX)) - 1;
3198 for (i = 0; i < MAX_LINKS; i++) {
3199 struct nl_portid_hash *hash = &nl_table[i].hash;
3201 hash->table = nl_portid_hash_zalloc(1 * sizeof(*hash->table));
3204 nl_portid_hash_free(nl_table[i].hash.table,
3205 1 * sizeof(*hash->table));
3209 hash->max_shift = order;
3212 hash->rehash_time = jiffies;
3214 nl_table[i].compare = netlink_compare;
3217 INIT_LIST_HEAD(&netlink_tap_all);
3219 netlink_add_usersock_entry();
3221 sock_register(&netlink_family_ops);
3222 register_pernet_subsys(&netlink_net_ops);
3223 /* The netlink device handler may be needed early. */
3228 panic("netlink_init: Cannot allocate nl_table\n");
3231 core_initcall(netlink_proto_init);