2 * NETLINK Kernel-user communication protocol.
4 * Authors: Alan Cox <alan@lxorguk.ukuu.org.uk>
5 * Alexey Kuznetsov <kuznet@ms2.inr.ac.ru>
6 * Patrick McHardy <kaber@trash.net>
8 * This program is free software; you can redistribute it and/or
9 * modify it under the terms of the GNU General Public License
10 * as published by the Free Software Foundation; either version
11 * 2 of the License, or (at your option) any later version.
13 * Tue Jun 26 14:36:48 MEST 2001 Herbert "herp" Rosmanith
14 * added netlink_proto_exit
15 * Tue Jan 22 18:32:44 BRST 2002 Arnaldo C. de Melo <acme@conectiva.com.br>
16 * use nlk_sk, as sk->protinfo is on a diet 8)
17 * Fri Jul 22 19:51:12 MEST 2005 Harald Welte <laforge@gnumonks.org>
18 * - inc module use count of module that owns
19 * the kernel socket in case userspace opens
20 * socket of same protocol
21 * - remove all module support, since netlink is
22 * mandatory if CONFIG_NET=y these days
25 #include <linux/module.h>
27 #include <linux/capability.h>
28 #include <linux/kernel.h>
29 #include <linux/init.h>
30 #include <linux/signal.h>
31 #include <linux/sched.h>
32 #include <linux/errno.h>
33 #include <linux/string.h>
34 #include <linux/stat.h>
35 #include <linux/socket.h>
37 #include <linux/fcntl.h>
38 #include <linux/termios.h>
39 #include <linux/sockios.h>
40 #include <linux/net.h>
42 #include <linux/slab.h>
43 #include <asm/uaccess.h>
44 #include <linux/skbuff.h>
45 #include <linux/netdevice.h>
46 #include <linux/rtnetlink.h>
47 #include <linux/proc_fs.h>
48 #include <linux/seq_file.h>
49 #include <linux/notifier.h>
50 #include <linux/security.h>
51 #include <linux/jhash.h>
52 #include <linux/jiffies.h>
53 #include <linux/random.h>
54 #include <linux/bitops.h>
56 #include <linux/types.h>
57 #include <linux/audit.h>
58 #include <linux/mutex.h>
59 #include <linux/vmalloc.h>
60 #include <linux/if_arp.h>
61 #include <linux/rhashtable.h>
62 #include <asm/cacheflush.h>
63 #include <linux/hash.h>
64 #include <linux/genetlink.h>
66 #include <net/net_namespace.h>
69 #include <net/netlink.h>
71 #include "af_netlink.h"
75 unsigned long masks[0];
79 #define NETLINK_S_CONGESTED 0x0
82 #define NETLINK_F_KERNEL_SOCKET 0x1
83 #define NETLINK_F_RECV_PKTINFO 0x2
84 #define NETLINK_F_BROADCAST_SEND_ERROR 0x4
85 #define NETLINK_F_RECV_NO_ENOBUFS 0x8
86 #define NETLINK_F_LISTEN_ALL_NSID 0x10
88 static inline int netlink_is_kernel(struct sock *sk)
90 return nlk_sk(sk)->flags & NETLINK_F_KERNEL_SOCKET;
93 struct netlink_table *nl_table __read_mostly;
94 EXPORT_SYMBOL_GPL(nl_table);
96 static DECLARE_WAIT_QUEUE_HEAD(nl_table_wait);
98 static int netlink_dump(struct sock *sk);
99 static void netlink_skb_destructor(struct sk_buff *skb);
101 /* nl_table locking explained:
102 * Lookup and traversal are protected with an RCU read-side lock. Insertion
103 * and removal are protected with per bucket lock while using RCU list
104 * modification primitives and may run in parallel to RCU protected lookups.
105 * Destruction of the Netlink socket may only occur *after* nl_table_lock has
106 * been acquired * either during or after the socket has been removed from
107 * the list and after an RCU grace period.
109 DEFINE_RWLOCK(nl_table_lock);
110 EXPORT_SYMBOL_GPL(nl_table_lock);
111 static atomic_t nl_table_users = ATOMIC_INIT(0);
113 #define nl_deref_protected(X) rcu_dereference_protected(X, lockdep_is_held(&nl_table_lock));
115 static ATOMIC_NOTIFIER_HEAD(netlink_chain);
117 static DEFINE_SPINLOCK(netlink_tap_lock);
118 static struct list_head netlink_tap_all __read_mostly;
120 static const struct rhashtable_params netlink_rhashtable_params;
122 static inline u32 netlink_group_mask(u32 group)
124 return group ? 1 << (group - 1) : 0;
127 int netlink_add_tap(struct netlink_tap *nt)
129 if (unlikely(nt->dev->type != ARPHRD_NETLINK))
132 spin_lock(&netlink_tap_lock);
133 list_add_rcu(&nt->list, &netlink_tap_all);
134 spin_unlock(&netlink_tap_lock);
136 __module_get(nt->module);
140 EXPORT_SYMBOL_GPL(netlink_add_tap);
142 static int __netlink_remove_tap(struct netlink_tap *nt)
145 struct netlink_tap *tmp;
147 spin_lock(&netlink_tap_lock);
149 list_for_each_entry(tmp, &netlink_tap_all, list) {
151 list_del_rcu(&nt->list);
157 pr_warn("__netlink_remove_tap: %p not found\n", nt);
159 spin_unlock(&netlink_tap_lock);
162 module_put(nt->module);
164 return found ? 0 : -ENODEV;
167 int netlink_remove_tap(struct netlink_tap *nt)
171 ret = __netlink_remove_tap(nt);
176 EXPORT_SYMBOL_GPL(netlink_remove_tap);
178 static bool netlink_filter_tap(const struct sk_buff *skb)
180 struct sock *sk = skb->sk;
182 /* We take the more conservative approach and
183 * whitelist socket protocols that may pass.
185 switch (sk->sk_protocol) {
187 case NETLINK_USERSOCK:
188 case NETLINK_SOCK_DIAG:
191 case NETLINK_FIB_LOOKUP:
192 case NETLINK_NETFILTER:
193 case NETLINK_GENERIC:
200 static int __netlink_deliver_tap_skb(struct sk_buff *skb,
201 struct net_device *dev)
203 struct sk_buff *nskb;
204 struct sock *sk = skb->sk;
208 nskb = skb_clone(skb, GFP_ATOMIC);
211 nskb->protocol = htons((u16) sk->sk_protocol);
212 nskb->pkt_type = netlink_is_kernel(sk) ?
213 PACKET_KERNEL : PACKET_USER;
214 skb_reset_network_header(nskb);
215 ret = dev_queue_xmit(nskb);
216 if (unlikely(ret > 0))
217 ret = net_xmit_errno(ret);
224 static void __netlink_deliver_tap(struct sk_buff *skb)
227 struct netlink_tap *tmp;
229 if (!netlink_filter_tap(skb))
232 list_for_each_entry_rcu(tmp, &netlink_tap_all, list) {
233 ret = __netlink_deliver_tap_skb(skb, tmp->dev);
239 static void netlink_deliver_tap(struct sk_buff *skb)
243 if (unlikely(!list_empty(&netlink_tap_all)))
244 __netlink_deliver_tap(skb);
249 static void netlink_deliver_tap_kernel(struct sock *dst, struct sock *src,
252 if (!(netlink_is_kernel(dst) && netlink_is_kernel(src)))
253 netlink_deliver_tap(skb);
256 static void netlink_overrun(struct sock *sk)
258 struct netlink_sock *nlk = nlk_sk(sk);
260 if (!(nlk->flags & NETLINK_F_RECV_NO_ENOBUFS)) {
261 if (!test_and_set_bit(NETLINK_S_CONGESTED,
262 &nlk_sk(sk)->state)) {
263 sk->sk_err = ENOBUFS;
264 sk->sk_error_report(sk);
267 atomic_inc(&sk->sk_drops);
270 static void netlink_rcv_wake(struct sock *sk)
272 struct netlink_sock *nlk = nlk_sk(sk);
274 if (skb_queue_empty(&sk->sk_receive_queue))
275 clear_bit(NETLINK_S_CONGESTED, &nlk->state);
276 if (!test_bit(NETLINK_S_CONGESTED, &nlk->state))
277 wake_up_interruptible(&nlk->wait);
280 #ifdef CONFIG_NETLINK_MMAP
281 static bool netlink_skb_is_mmaped(const struct sk_buff *skb)
283 return NETLINK_CB(skb).flags & NETLINK_SKB_MMAPED;
286 static bool netlink_rx_is_mmaped(struct sock *sk)
288 return nlk_sk(sk)->rx_ring.pg_vec != NULL;
291 static bool netlink_tx_is_mmaped(struct sock *sk)
293 return nlk_sk(sk)->tx_ring.pg_vec != NULL;
296 static __pure struct page *pgvec_to_page(const void *addr)
298 if (is_vmalloc_addr(addr))
299 return vmalloc_to_page(addr);
301 return virt_to_page(addr);
304 static void free_pg_vec(void **pg_vec, unsigned int order, unsigned int len)
308 for (i = 0; i < len; i++) {
309 if (pg_vec[i] != NULL) {
310 if (is_vmalloc_addr(pg_vec[i]))
313 free_pages((unsigned long)pg_vec[i], order);
319 static void *alloc_one_pg_vec_page(unsigned long order)
322 gfp_t gfp_flags = GFP_KERNEL | __GFP_COMP | __GFP_ZERO |
323 __GFP_NOWARN | __GFP_NORETRY;
325 buffer = (void *)__get_free_pages(gfp_flags, order);
329 buffer = vzalloc((1 << order) * PAGE_SIZE);
333 gfp_flags &= ~__GFP_NORETRY;
334 return (void *)__get_free_pages(gfp_flags, order);
337 static void **alloc_pg_vec(struct netlink_sock *nlk,
338 struct nl_mmap_req *req, unsigned int order)
340 unsigned int block_nr = req->nm_block_nr;
344 pg_vec = kcalloc(block_nr, sizeof(void *), GFP_KERNEL);
348 for (i = 0; i < block_nr; i++) {
349 pg_vec[i] = alloc_one_pg_vec_page(order);
350 if (pg_vec[i] == NULL)
356 free_pg_vec(pg_vec, order, block_nr);
362 __netlink_set_ring(struct sock *sk, struct nl_mmap_req *req, bool tx_ring, void **pg_vec,
365 struct netlink_sock *nlk = nlk_sk(sk);
366 struct sk_buff_head *queue;
367 struct netlink_ring *ring;
369 queue = tx_ring ? &sk->sk_write_queue : &sk->sk_receive_queue;
370 ring = tx_ring ? &nlk->tx_ring : &nlk->rx_ring;
372 spin_lock_bh(&queue->lock);
374 ring->frame_max = req->nm_frame_nr - 1;
376 ring->frame_size = req->nm_frame_size;
377 ring->pg_vec_pages = req->nm_block_size / PAGE_SIZE;
379 swap(ring->pg_vec_len, req->nm_block_nr);
380 swap(ring->pg_vec_order, order);
381 swap(ring->pg_vec, pg_vec);
383 __skb_queue_purge(queue);
384 spin_unlock_bh(&queue->lock);
386 WARN_ON(atomic_read(&nlk->mapped));
389 free_pg_vec(pg_vec, order, req->nm_block_nr);
392 static int netlink_set_ring(struct sock *sk, struct nl_mmap_req *req,
395 struct netlink_sock *nlk = nlk_sk(sk);
396 struct netlink_ring *ring;
397 void **pg_vec = NULL;
398 unsigned int order = 0;
400 ring = tx_ring ? &nlk->tx_ring : &nlk->rx_ring;
402 if (atomic_read(&nlk->mapped))
404 if (atomic_read(&ring->pending))
407 if (req->nm_block_nr) {
408 if (ring->pg_vec != NULL)
411 if ((int)req->nm_block_size <= 0)
413 if (!PAGE_ALIGNED(req->nm_block_size))
415 if (req->nm_frame_size < NL_MMAP_HDRLEN)
417 if (!IS_ALIGNED(req->nm_frame_size, NL_MMAP_MSG_ALIGNMENT))
420 ring->frames_per_block = req->nm_block_size /
422 if (ring->frames_per_block == 0)
424 if (ring->frames_per_block * req->nm_block_nr !=
428 order = get_order(req->nm_block_size);
429 pg_vec = alloc_pg_vec(nlk, req, order);
433 if (req->nm_frame_nr)
437 mutex_lock(&nlk->pg_vec_lock);
438 if (atomic_read(&nlk->mapped) == 0) {
439 __netlink_set_ring(sk, req, tx_ring, pg_vec, order);
440 mutex_unlock(&nlk->pg_vec_lock);
444 mutex_unlock(&nlk->pg_vec_lock);
447 free_pg_vec(pg_vec, order, req->nm_block_nr);
452 static void netlink_mm_open(struct vm_area_struct *vma)
454 struct file *file = vma->vm_file;
455 struct socket *sock = file->private_data;
456 struct sock *sk = sock->sk;
459 atomic_inc(&nlk_sk(sk)->mapped);
462 static void netlink_mm_close(struct vm_area_struct *vma)
464 struct file *file = vma->vm_file;
465 struct socket *sock = file->private_data;
466 struct sock *sk = sock->sk;
469 atomic_dec(&nlk_sk(sk)->mapped);
472 static const struct vm_operations_struct netlink_mmap_ops = {
473 .open = netlink_mm_open,
474 .close = netlink_mm_close,
477 static int netlink_mmap(struct file *file, struct socket *sock,
478 struct vm_area_struct *vma)
480 struct sock *sk = sock->sk;
481 struct netlink_sock *nlk = nlk_sk(sk);
482 struct netlink_ring *ring;
483 unsigned long start, size, expected;
490 mutex_lock(&nlk->pg_vec_lock);
493 for (ring = &nlk->rx_ring; ring <= &nlk->tx_ring; ring++) {
494 if (ring->pg_vec == NULL)
496 expected += ring->pg_vec_len * ring->pg_vec_pages * PAGE_SIZE;
502 size = vma->vm_end - vma->vm_start;
503 if (size != expected)
506 start = vma->vm_start;
507 for (ring = &nlk->rx_ring; ring <= &nlk->tx_ring; ring++) {
508 if (ring->pg_vec == NULL)
511 for (i = 0; i < ring->pg_vec_len; i++) {
513 void *kaddr = ring->pg_vec[i];
516 for (pg_num = 0; pg_num < ring->pg_vec_pages; pg_num++) {
517 page = pgvec_to_page(kaddr);
518 err = vm_insert_page(vma, start, page);
527 atomic_inc(&nlk->mapped);
528 vma->vm_ops = &netlink_mmap_ops;
531 mutex_unlock(&nlk->pg_vec_lock);
535 static void netlink_frame_flush_dcache(const struct nl_mmap_hdr *hdr, unsigned int nm_len)
537 #if ARCH_IMPLEMENTS_FLUSH_DCACHE_PAGE == 1
538 struct page *p_start, *p_end;
540 /* First page is flushed through netlink_{get,set}_status */
541 p_start = pgvec_to_page(hdr + PAGE_SIZE);
542 p_end = pgvec_to_page((void *)hdr + NL_MMAP_HDRLEN + nm_len - 1);
543 while (p_start <= p_end) {
544 flush_dcache_page(p_start);
550 static enum nl_mmap_status netlink_get_status(const struct nl_mmap_hdr *hdr)
553 flush_dcache_page(pgvec_to_page(hdr));
554 return hdr->nm_status;
557 static void netlink_set_status(struct nl_mmap_hdr *hdr,
558 enum nl_mmap_status status)
561 hdr->nm_status = status;
562 flush_dcache_page(pgvec_to_page(hdr));
565 static struct nl_mmap_hdr *
566 __netlink_lookup_frame(const struct netlink_ring *ring, unsigned int pos)
568 unsigned int pg_vec_pos, frame_off;
570 pg_vec_pos = pos / ring->frames_per_block;
571 frame_off = pos % ring->frames_per_block;
573 return ring->pg_vec[pg_vec_pos] + (frame_off * ring->frame_size);
576 static struct nl_mmap_hdr *
577 netlink_lookup_frame(const struct netlink_ring *ring, unsigned int pos,
578 enum nl_mmap_status status)
580 struct nl_mmap_hdr *hdr;
582 hdr = __netlink_lookup_frame(ring, pos);
583 if (netlink_get_status(hdr) != status)
589 static struct nl_mmap_hdr *
590 netlink_current_frame(const struct netlink_ring *ring,
591 enum nl_mmap_status status)
593 return netlink_lookup_frame(ring, ring->head, status);
596 static struct nl_mmap_hdr *
597 netlink_previous_frame(const struct netlink_ring *ring,
598 enum nl_mmap_status status)
602 prev = ring->head ? ring->head - 1 : ring->frame_max;
603 return netlink_lookup_frame(ring, prev, status);
606 static void netlink_increment_head(struct netlink_ring *ring)
608 ring->head = ring->head != ring->frame_max ? ring->head + 1 : 0;
611 static void netlink_forward_ring(struct netlink_ring *ring)
613 unsigned int head = ring->head, pos = head;
614 const struct nl_mmap_hdr *hdr;
617 hdr = __netlink_lookup_frame(ring, pos);
618 if (hdr->nm_status == NL_MMAP_STATUS_UNUSED)
620 if (hdr->nm_status != NL_MMAP_STATUS_SKIP)
622 netlink_increment_head(ring);
623 } while (ring->head != head);
626 static bool netlink_dump_space(struct netlink_sock *nlk)
628 struct netlink_ring *ring = &nlk->rx_ring;
629 struct nl_mmap_hdr *hdr;
632 hdr = netlink_current_frame(ring, NL_MMAP_STATUS_UNUSED);
636 n = ring->head + ring->frame_max / 2;
637 if (n > ring->frame_max)
638 n -= ring->frame_max;
640 hdr = __netlink_lookup_frame(ring, n);
642 return hdr->nm_status == NL_MMAP_STATUS_UNUSED;
645 static unsigned int netlink_poll(struct file *file, struct socket *sock,
648 struct sock *sk = sock->sk;
649 struct netlink_sock *nlk = nlk_sk(sk);
653 if (nlk->rx_ring.pg_vec != NULL) {
654 /* Memory mapped sockets don't call recvmsg(), so flow control
655 * for dumps is performed here. A dump is allowed to continue
656 * if at least half the ring is unused.
658 while (nlk->cb_running && netlink_dump_space(nlk)) {
659 err = netlink_dump(sk);
662 sk->sk_error_report(sk);
666 netlink_rcv_wake(sk);
669 mask = datagram_poll(file, sock, wait);
671 spin_lock_bh(&sk->sk_receive_queue.lock);
672 if (nlk->rx_ring.pg_vec) {
673 netlink_forward_ring(&nlk->rx_ring);
674 if (!netlink_previous_frame(&nlk->rx_ring, NL_MMAP_STATUS_UNUSED))
675 mask |= POLLIN | POLLRDNORM;
677 spin_unlock_bh(&sk->sk_receive_queue.lock);
679 spin_lock_bh(&sk->sk_write_queue.lock);
680 if (nlk->tx_ring.pg_vec) {
681 if (netlink_current_frame(&nlk->tx_ring, NL_MMAP_STATUS_UNUSED))
682 mask |= POLLOUT | POLLWRNORM;
684 spin_unlock_bh(&sk->sk_write_queue.lock);
689 static struct nl_mmap_hdr *netlink_mmap_hdr(struct sk_buff *skb)
691 return (struct nl_mmap_hdr *)(skb->head - NL_MMAP_HDRLEN);
694 static void netlink_ring_setup_skb(struct sk_buff *skb, struct sock *sk,
695 struct netlink_ring *ring,
696 struct nl_mmap_hdr *hdr)
701 size = ring->frame_size - NL_MMAP_HDRLEN;
702 data = (void *)hdr + NL_MMAP_HDRLEN;
706 skb_reset_tail_pointer(skb);
707 skb->end = skb->tail + size;
710 skb->destructor = netlink_skb_destructor;
711 NETLINK_CB(skb).flags |= NETLINK_SKB_MMAPED;
712 NETLINK_CB(skb).sk = sk;
715 static int netlink_mmap_sendmsg(struct sock *sk, struct msghdr *msg,
716 u32 dst_portid, u32 dst_group,
717 struct scm_cookie *scm)
719 struct netlink_sock *nlk = nlk_sk(sk);
720 struct netlink_ring *ring;
721 struct nl_mmap_hdr *hdr;
724 int err = 0, len = 0;
726 mutex_lock(&nlk->pg_vec_lock);
728 ring = &nlk->tx_ring;
729 maxlen = ring->frame_size - NL_MMAP_HDRLEN;
734 hdr = netlink_current_frame(ring, NL_MMAP_STATUS_VALID);
736 if (!(msg->msg_flags & MSG_DONTWAIT) &&
737 atomic_read(&nlk->tx_ring.pending))
742 nm_len = ACCESS_ONCE(hdr->nm_len);
743 if (nm_len > maxlen) {
748 netlink_frame_flush_dcache(hdr, nm_len);
750 skb = alloc_skb(nm_len, GFP_KERNEL);
755 __skb_put(skb, nm_len);
756 memcpy(skb->data, (void *)hdr + NL_MMAP_HDRLEN, nm_len);
757 netlink_set_status(hdr, NL_MMAP_STATUS_UNUSED);
759 netlink_increment_head(ring);
761 NETLINK_CB(skb).portid = nlk->portid;
762 NETLINK_CB(skb).dst_group = dst_group;
763 NETLINK_CB(skb).creds = scm->creds;
765 err = security_netlink_send(sk, skb);
771 if (unlikely(dst_group)) {
772 atomic_inc(&skb->users);
773 netlink_broadcast(sk, skb, dst_portid, dst_group,
776 err = netlink_unicast(sk, skb, dst_portid,
777 msg->msg_flags & MSG_DONTWAIT);
782 } while (hdr != NULL ||
783 (!(msg->msg_flags & MSG_DONTWAIT) &&
784 atomic_read(&nlk->tx_ring.pending)));
789 mutex_unlock(&nlk->pg_vec_lock);
793 static void netlink_queue_mmaped_skb(struct sock *sk, struct sk_buff *skb)
795 struct nl_mmap_hdr *hdr;
797 hdr = netlink_mmap_hdr(skb);
798 hdr->nm_len = skb->len;
799 hdr->nm_group = NETLINK_CB(skb).dst_group;
800 hdr->nm_pid = NETLINK_CB(skb).creds.pid;
801 hdr->nm_uid = from_kuid(sk_user_ns(sk), NETLINK_CB(skb).creds.uid);
802 hdr->nm_gid = from_kgid(sk_user_ns(sk), NETLINK_CB(skb).creds.gid);
803 netlink_frame_flush_dcache(hdr, hdr->nm_len);
804 netlink_set_status(hdr, NL_MMAP_STATUS_VALID);
806 NETLINK_CB(skb).flags |= NETLINK_SKB_DELIVERED;
810 static void netlink_ring_set_copied(struct sock *sk, struct sk_buff *skb)
812 struct netlink_sock *nlk = nlk_sk(sk);
813 struct netlink_ring *ring = &nlk->rx_ring;
814 struct nl_mmap_hdr *hdr;
816 spin_lock_bh(&sk->sk_receive_queue.lock);
817 hdr = netlink_current_frame(ring, NL_MMAP_STATUS_UNUSED);
819 spin_unlock_bh(&sk->sk_receive_queue.lock);
824 netlink_increment_head(ring);
825 __skb_queue_tail(&sk->sk_receive_queue, skb);
826 spin_unlock_bh(&sk->sk_receive_queue.lock);
828 hdr->nm_len = skb->len;
829 hdr->nm_group = NETLINK_CB(skb).dst_group;
830 hdr->nm_pid = NETLINK_CB(skb).creds.pid;
831 hdr->nm_uid = from_kuid(sk_user_ns(sk), NETLINK_CB(skb).creds.uid);
832 hdr->nm_gid = from_kgid(sk_user_ns(sk), NETLINK_CB(skb).creds.gid);
833 netlink_set_status(hdr, NL_MMAP_STATUS_COPY);
836 #else /* CONFIG_NETLINK_MMAP */
837 #define netlink_skb_is_mmaped(skb) false
838 #define netlink_rx_is_mmaped(sk) false
839 #define netlink_tx_is_mmaped(sk) false
840 #define netlink_mmap sock_no_mmap
841 #define netlink_poll datagram_poll
842 #define netlink_mmap_sendmsg(sk, msg, dst_portid, dst_group, scm) 0
843 #endif /* CONFIG_NETLINK_MMAP */
845 static void netlink_skb_destructor(struct sk_buff *skb)
847 #ifdef CONFIG_NETLINK_MMAP
848 struct nl_mmap_hdr *hdr;
849 struct netlink_ring *ring;
852 /* If a packet from the kernel to userspace was freed because of an
853 * error without being delivered to userspace, the kernel must reset
854 * the status. In the direction userspace to kernel, the status is
855 * always reset here after the packet was processed and freed.
857 if (netlink_skb_is_mmaped(skb)) {
858 hdr = netlink_mmap_hdr(skb);
859 sk = NETLINK_CB(skb).sk;
861 if (NETLINK_CB(skb).flags & NETLINK_SKB_TX) {
862 netlink_set_status(hdr, NL_MMAP_STATUS_UNUSED);
863 ring = &nlk_sk(sk)->tx_ring;
865 if (!(NETLINK_CB(skb).flags & NETLINK_SKB_DELIVERED)) {
867 netlink_set_status(hdr, NL_MMAP_STATUS_VALID);
869 ring = &nlk_sk(sk)->rx_ring;
872 WARN_ON(atomic_read(&ring->pending) == 0);
873 atomic_dec(&ring->pending);
879 if (is_vmalloc_addr(skb->head)) {
881 !atomic_dec_return(&(skb_shinfo(skb)->dataref)))
890 static void netlink_skb_set_owner_r(struct sk_buff *skb, struct sock *sk)
892 WARN_ON(skb->sk != NULL);
894 skb->destructor = netlink_skb_destructor;
895 atomic_add(skb->truesize, &sk->sk_rmem_alloc);
896 sk_mem_charge(sk, skb->truesize);
899 static void netlink_sock_destruct(struct sock *sk)
901 struct netlink_sock *nlk = nlk_sk(sk);
903 if (nlk->cb_running) {
905 nlk->cb.done(&nlk->cb);
907 module_put(nlk->cb.module);
908 kfree_skb(nlk->cb.skb);
911 skb_queue_purge(&sk->sk_receive_queue);
912 #ifdef CONFIG_NETLINK_MMAP
914 struct nl_mmap_req req;
916 memset(&req, 0, sizeof(req));
917 if (nlk->rx_ring.pg_vec)
918 __netlink_set_ring(sk, &req, false, NULL, 0);
919 memset(&req, 0, sizeof(req));
920 if (nlk->tx_ring.pg_vec)
921 __netlink_set_ring(sk, &req, true, NULL, 0);
923 #endif /* CONFIG_NETLINK_MMAP */
925 if (!sock_flag(sk, SOCK_DEAD)) {
926 printk(KERN_ERR "Freeing alive netlink socket %p\n", sk);
930 WARN_ON(atomic_read(&sk->sk_rmem_alloc));
931 WARN_ON(atomic_read(&sk->sk_wmem_alloc));
932 WARN_ON(nlk_sk(sk)->groups);
935 /* This lock without WQ_FLAG_EXCLUSIVE is good on UP and it is _very_ bad on
936 * SMP. Look, when several writers sleep and reader wakes them up, all but one
937 * immediately hit write lock and grab all the cpus. Exclusive sleep solves
938 * this, _but_ remember, it adds useless work on UP machines.
941 void netlink_table_grab(void)
942 __acquires(nl_table_lock)
946 write_lock_irq(&nl_table_lock);
948 if (atomic_read(&nl_table_users)) {
949 DECLARE_WAITQUEUE(wait, current);
951 add_wait_queue_exclusive(&nl_table_wait, &wait);
953 set_current_state(TASK_UNINTERRUPTIBLE);
954 if (atomic_read(&nl_table_users) == 0)
956 write_unlock_irq(&nl_table_lock);
958 write_lock_irq(&nl_table_lock);
961 __set_current_state(TASK_RUNNING);
962 remove_wait_queue(&nl_table_wait, &wait);
966 void netlink_table_ungrab(void)
967 __releases(nl_table_lock)
969 write_unlock_irq(&nl_table_lock);
970 wake_up(&nl_table_wait);
974 netlink_lock_table(void)
976 /* read_lock() synchronizes us to netlink_table_grab */
978 read_lock(&nl_table_lock);
979 atomic_inc(&nl_table_users);
980 read_unlock(&nl_table_lock);
984 netlink_unlock_table(void)
986 if (atomic_dec_and_test(&nl_table_users))
987 wake_up(&nl_table_wait);
990 struct netlink_compare_arg
996 /* Doing sizeof directly may yield 4 extra bytes on 64-bit. */
997 #define netlink_compare_arg_len \
998 (offsetof(struct netlink_compare_arg, portid) + sizeof(u32))
1000 static inline int netlink_compare(struct rhashtable_compare_arg *arg,
1003 const struct netlink_compare_arg *x = arg->key;
1004 const struct netlink_sock *nlk = ptr;
1006 return nlk->portid != x->portid ||
1007 !net_eq(sock_net(&nlk->sk), read_pnet(&x->pnet));
1010 static void netlink_compare_arg_init(struct netlink_compare_arg *arg,
1011 struct net *net, u32 portid)
1013 memset(arg, 0, sizeof(*arg));
1014 write_pnet(&arg->pnet, net);
1015 arg->portid = portid;
1018 static struct sock *__netlink_lookup(struct netlink_table *table, u32 portid,
1021 struct netlink_compare_arg arg;
1023 netlink_compare_arg_init(&arg, net, portid);
1024 return rhashtable_lookup_fast(&table->hash, &arg,
1025 netlink_rhashtable_params);
1028 static int __netlink_insert(struct netlink_table *table, struct sock *sk)
1030 struct netlink_compare_arg arg;
1032 netlink_compare_arg_init(&arg, sock_net(sk), nlk_sk(sk)->portid);
1033 return rhashtable_lookup_insert_key(&table->hash, &arg,
1035 netlink_rhashtable_params);
1038 static struct sock *netlink_lookup(struct net *net, int protocol, u32 portid)
1040 struct netlink_table *table = &nl_table[protocol];
1044 sk = __netlink_lookup(table, portid, net);
1052 static const struct proto_ops netlink_ops;
1055 netlink_update_listeners(struct sock *sk)
1057 struct netlink_table *tbl = &nl_table[sk->sk_protocol];
1060 struct listeners *listeners;
1062 listeners = nl_deref_protected(tbl->listeners);
1066 for (i = 0; i < NLGRPLONGS(tbl->groups); i++) {
1068 sk_for_each_bound(sk, &tbl->mc_list) {
1069 if (i < NLGRPLONGS(nlk_sk(sk)->ngroups))
1070 mask |= nlk_sk(sk)->groups[i];
1072 listeners->masks[i] = mask;
1074 /* this function is only called with the netlink table "grabbed", which
1075 * makes sure updates are visible before bind or setsockopt return. */
1078 static int netlink_insert(struct sock *sk, u32 portid)
1080 struct netlink_table *table = &nl_table[sk->sk_protocol];
1086 if (nlk_sk(sk)->portid)
1090 if (BITS_PER_LONG > 32 &&
1091 unlikely(atomic_read(&table->hash.nelems) >= UINT_MAX))
1094 nlk_sk(sk)->portid = portid;
1097 err = __netlink_insert(table, sk);
1101 nlk_sk(sk)->portid = 0;
1110 static void netlink_remove(struct sock *sk)
1112 struct netlink_table *table;
1114 table = &nl_table[sk->sk_protocol];
1115 if (!rhashtable_remove_fast(&table->hash, &nlk_sk(sk)->node,
1116 netlink_rhashtable_params)) {
1117 WARN_ON(atomic_read(&sk->sk_refcnt) == 1);
1121 netlink_table_grab();
1122 if (nlk_sk(sk)->subscriptions) {
1123 __sk_del_bind_node(sk);
1124 netlink_update_listeners(sk);
1126 if (sk->sk_protocol == NETLINK_GENERIC)
1127 atomic_inc(&genl_sk_destructing_cnt);
1128 netlink_table_ungrab();
1131 static struct proto netlink_proto = {
1133 .owner = THIS_MODULE,
1134 .obj_size = sizeof(struct netlink_sock),
1137 static int __netlink_create(struct net *net, struct socket *sock,
1138 struct mutex *cb_mutex, int protocol,
1142 struct netlink_sock *nlk;
1144 sock->ops = &netlink_ops;
1146 sk = sk_alloc(net, PF_NETLINK, GFP_KERNEL, &netlink_proto, kern);
1150 sock_init_data(sock, sk);
1154 nlk->cb_mutex = cb_mutex;
1156 nlk->cb_mutex = &nlk->cb_def_mutex;
1157 mutex_init(nlk->cb_mutex);
1159 init_waitqueue_head(&nlk->wait);
1160 #ifdef CONFIG_NETLINK_MMAP
1161 mutex_init(&nlk->pg_vec_lock);
1164 sk->sk_destruct = netlink_sock_destruct;
1165 sk->sk_protocol = protocol;
1169 static int netlink_create(struct net *net, struct socket *sock, int protocol,
1172 struct module *module = NULL;
1173 struct mutex *cb_mutex;
1174 struct netlink_sock *nlk;
1175 int (*bind)(struct net *net, int group);
1176 void (*unbind)(struct net *net, int group);
1179 sock->state = SS_UNCONNECTED;
1181 if (sock->type != SOCK_RAW && sock->type != SOCK_DGRAM)
1182 return -ESOCKTNOSUPPORT;
1184 if (protocol < 0 || protocol >= MAX_LINKS)
1185 return -EPROTONOSUPPORT;
1187 netlink_lock_table();
1188 #ifdef CONFIG_MODULES
1189 if (!nl_table[protocol].registered) {
1190 netlink_unlock_table();
1191 request_module("net-pf-%d-proto-%d", PF_NETLINK, protocol);
1192 netlink_lock_table();
1195 if (nl_table[protocol].registered &&
1196 try_module_get(nl_table[protocol].module))
1197 module = nl_table[protocol].module;
1199 err = -EPROTONOSUPPORT;
1200 cb_mutex = nl_table[protocol].cb_mutex;
1201 bind = nl_table[protocol].bind;
1202 unbind = nl_table[protocol].unbind;
1203 netlink_unlock_table();
1208 err = __netlink_create(net, sock, cb_mutex, protocol, kern);
1213 sock_prot_inuse_add(net, &netlink_proto, 1);
1216 nlk = nlk_sk(sock->sk);
1217 nlk->module = module;
1218 nlk->netlink_bind = bind;
1219 nlk->netlink_unbind = unbind;
1228 static void deferred_put_nlk_sk(struct rcu_head *head)
1230 struct netlink_sock *nlk = container_of(head, struct netlink_sock, rcu);
1235 static int netlink_release(struct socket *sock)
1237 struct sock *sk = sock->sk;
1238 struct netlink_sock *nlk;
1248 * OK. Socket is unlinked, any packets that arrive now
1252 /* must not acquire netlink_table_lock in any way again before unbind
1253 * and notifying genetlink is done as otherwise it might deadlock
1255 if (nlk->netlink_unbind) {
1258 for (i = 0; i < nlk->ngroups; i++)
1259 if (test_bit(i, nlk->groups))
1260 nlk->netlink_unbind(sock_net(sk), i + 1);
1262 if (sk->sk_protocol == NETLINK_GENERIC &&
1263 atomic_dec_return(&genl_sk_destructing_cnt) == 0)
1264 wake_up(&genl_sk_destructing_waitq);
1267 wake_up_interruptible_all(&nlk->wait);
1269 skb_queue_purge(&sk->sk_write_queue);
1272 struct netlink_notify n = {
1273 .net = sock_net(sk),
1274 .protocol = sk->sk_protocol,
1275 .portid = nlk->portid,
1277 atomic_notifier_call_chain(&netlink_chain,
1278 NETLINK_URELEASE, &n);
1281 module_put(nlk->module);
1283 if (netlink_is_kernel(sk)) {
1284 netlink_table_grab();
1285 BUG_ON(nl_table[sk->sk_protocol].registered == 0);
1286 if (--nl_table[sk->sk_protocol].registered == 0) {
1287 struct listeners *old;
1289 old = nl_deref_protected(nl_table[sk->sk_protocol].listeners);
1290 RCU_INIT_POINTER(nl_table[sk->sk_protocol].listeners, NULL);
1291 kfree_rcu(old, rcu);
1292 nl_table[sk->sk_protocol].module = NULL;
1293 nl_table[sk->sk_protocol].bind = NULL;
1294 nl_table[sk->sk_protocol].unbind = NULL;
1295 nl_table[sk->sk_protocol].flags = 0;
1296 nl_table[sk->sk_protocol].registered = 0;
1298 netlink_table_ungrab();
1305 sock_prot_inuse_add(sock_net(sk), &netlink_proto, -1);
1307 call_rcu(&nlk->rcu, deferred_put_nlk_sk);
1311 static int netlink_autobind(struct socket *sock)
1313 struct sock *sk = sock->sk;
1314 struct net *net = sock_net(sk);
1315 struct netlink_table *table = &nl_table[sk->sk_protocol];
1316 s32 portid = task_tgid_vnr(current);
1324 ok = !__netlink_lookup(table, portid, net);
1327 /* Bind collision, search negative portid values. */
1329 /* rover will be in range [S32_MIN, -4097] */
1330 rover = S32_MIN + prandom_u32_max(-4096 - S32_MIN);
1331 else if (rover >= -4096)
1337 err = netlink_insert(sk, portid);
1338 if (err == -EADDRINUSE)
1341 /* If 2 threads race to autobind, that is fine. */
1349 * __netlink_ns_capable - General netlink message capability test
1350 * @nsp: NETLINK_CB of the socket buffer holding a netlink command from userspace.
1351 * @user_ns: The user namespace of the capability to use
1352 * @cap: The capability to use
1354 * Test to see if the opener of the socket we received the message
1355 * from had when the netlink socket was created and the sender of the
1356 * message has has the capability @cap in the user namespace @user_ns.
1358 bool __netlink_ns_capable(const struct netlink_skb_parms *nsp,
1359 struct user_namespace *user_ns, int cap)
1361 return ((nsp->flags & NETLINK_SKB_DST) ||
1362 file_ns_capable(nsp->sk->sk_socket->file, user_ns, cap)) &&
1363 ns_capable(user_ns, cap);
1365 EXPORT_SYMBOL(__netlink_ns_capable);
1368 * netlink_ns_capable - General netlink message capability test
1369 * @skb: 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 sk_buff *skb,
1378 struct user_namespace *user_ns, int cap)
1380 return __netlink_ns_capable(&NETLINK_CB(skb), user_ns, cap);
1382 EXPORT_SYMBOL(netlink_ns_capable);
1385 * netlink_capable - Netlink global message capability test
1386 * @skb: socket buffer holding a netlink command from userspace
1387 * @cap: The capability to use
1389 * Test to see if the opener of the socket we received the message
1390 * from had when the netlink socket was created and the sender of the
1391 * message has has the capability @cap in all user namespaces.
1393 bool netlink_capable(const struct sk_buff *skb, int cap)
1395 return netlink_ns_capable(skb, &init_user_ns, cap);
1397 EXPORT_SYMBOL(netlink_capable);
1400 * netlink_net_capable - Netlink network namespace message capability test
1401 * @skb: socket buffer holding a netlink command from userspace
1402 * @cap: The capability to use
1404 * Test to see if the opener of the socket we received the message
1405 * from had when the netlink socket was created and the sender of the
1406 * message has has the capability @cap over the network namespace of
1407 * the socket we received the message from.
1409 bool netlink_net_capable(const struct sk_buff *skb, int cap)
1411 return netlink_ns_capable(skb, sock_net(skb->sk)->user_ns, cap);
1413 EXPORT_SYMBOL(netlink_net_capable);
1415 static inline int netlink_allowed(const struct socket *sock, unsigned int flag)
1417 return (nl_table[sock->sk->sk_protocol].flags & flag) ||
1418 ns_capable(sock_net(sock->sk)->user_ns, CAP_NET_ADMIN);
1422 netlink_update_subscriptions(struct sock *sk, unsigned int subscriptions)
1424 struct netlink_sock *nlk = nlk_sk(sk);
1426 if (nlk->subscriptions && !subscriptions)
1427 __sk_del_bind_node(sk);
1428 else if (!nlk->subscriptions && subscriptions)
1429 sk_add_bind_node(sk, &nl_table[sk->sk_protocol].mc_list);
1430 nlk->subscriptions = subscriptions;
1433 static int netlink_realloc_groups(struct sock *sk)
1435 struct netlink_sock *nlk = nlk_sk(sk);
1436 unsigned int groups;
1437 unsigned long *new_groups;
1440 netlink_table_grab();
1442 groups = nl_table[sk->sk_protocol].groups;
1443 if (!nl_table[sk->sk_protocol].registered) {
1448 if (nlk->ngroups >= groups)
1451 new_groups = krealloc(nlk->groups, NLGRPSZ(groups), GFP_ATOMIC);
1452 if (new_groups == NULL) {
1456 memset((char *)new_groups + NLGRPSZ(nlk->ngroups), 0,
1457 NLGRPSZ(groups) - NLGRPSZ(nlk->ngroups));
1459 nlk->groups = new_groups;
1460 nlk->ngroups = groups;
1462 netlink_table_ungrab();
1466 static void netlink_undo_bind(int group, long unsigned int groups,
1469 struct netlink_sock *nlk = nlk_sk(sk);
1472 if (!nlk->netlink_unbind)
1475 for (undo = 0; undo < group; undo++)
1476 if (test_bit(undo, &groups))
1477 nlk->netlink_unbind(sock_net(sk), undo + 1);
1480 static int netlink_bind(struct socket *sock, struct sockaddr *addr,
1483 struct sock *sk = sock->sk;
1484 struct net *net = sock_net(sk);
1485 struct netlink_sock *nlk = nlk_sk(sk);
1486 struct sockaddr_nl *nladdr = (struct sockaddr_nl *)addr;
1488 long unsigned int groups = nladdr->nl_groups;
1490 if (addr_len < sizeof(struct sockaddr_nl))
1493 if (nladdr->nl_family != AF_NETLINK)
1496 /* Only superuser is allowed to listen multicasts */
1498 if (!netlink_allowed(sock, NL_CFG_F_NONROOT_RECV))
1500 err = netlink_realloc_groups(sk);
1506 if (nladdr->nl_pid != nlk->portid)
1509 if (nlk->netlink_bind && groups) {
1512 for (group = 0; group < nlk->ngroups; group++) {
1513 if (!test_bit(group, &groups))
1515 err = nlk->netlink_bind(net, group + 1);
1518 netlink_undo_bind(group, groups, sk);
1524 err = nladdr->nl_pid ?
1525 netlink_insert(sk, nladdr->nl_pid) :
1526 netlink_autobind(sock);
1528 netlink_undo_bind(nlk->ngroups, groups, sk);
1533 if (!groups && (nlk->groups == NULL || !(u32)nlk->groups[0]))
1536 netlink_table_grab();
1537 netlink_update_subscriptions(sk, nlk->subscriptions +
1539 hweight32(nlk->groups[0]));
1540 nlk->groups[0] = (nlk->groups[0] & ~0xffffffffUL) | groups;
1541 netlink_update_listeners(sk);
1542 netlink_table_ungrab();
1547 static int netlink_connect(struct socket *sock, struct sockaddr *addr,
1548 int alen, int flags)
1551 struct sock *sk = sock->sk;
1552 struct netlink_sock *nlk = nlk_sk(sk);
1553 struct sockaddr_nl *nladdr = (struct sockaddr_nl *)addr;
1555 if (alen < sizeof(addr->sa_family))
1558 if (addr->sa_family == AF_UNSPEC) {
1559 sk->sk_state = NETLINK_UNCONNECTED;
1560 nlk->dst_portid = 0;
1564 if (addr->sa_family != AF_NETLINK)
1567 if ((nladdr->nl_groups || nladdr->nl_pid) &&
1568 !netlink_allowed(sock, NL_CFG_F_NONROOT_SEND))
1572 err = netlink_autobind(sock);
1575 sk->sk_state = NETLINK_CONNECTED;
1576 nlk->dst_portid = nladdr->nl_pid;
1577 nlk->dst_group = ffs(nladdr->nl_groups);
1583 static int netlink_getname(struct socket *sock, struct sockaddr *addr,
1584 int *addr_len, int peer)
1586 struct sock *sk = sock->sk;
1587 struct netlink_sock *nlk = nlk_sk(sk);
1588 DECLARE_SOCKADDR(struct sockaddr_nl *, nladdr, addr);
1590 nladdr->nl_family = AF_NETLINK;
1592 *addr_len = sizeof(*nladdr);
1595 nladdr->nl_pid = nlk->dst_portid;
1596 nladdr->nl_groups = netlink_group_mask(nlk->dst_group);
1598 nladdr->nl_pid = nlk->portid;
1599 nladdr->nl_groups = nlk->groups ? nlk->groups[0] : 0;
1604 static struct sock *netlink_getsockbyportid(struct sock *ssk, u32 portid)
1607 struct netlink_sock *nlk;
1609 sock = netlink_lookup(sock_net(ssk), ssk->sk_protocol, portid);
1611 return ERR_PTR(-ECONNREFUSED);
1613 /* Don't bother queuing skb if kernel socket has no input function */
1615 if (sock->sk_state == NETLINK_CONNECTED &&
1616 nlk->dst_portid != nlk_sk(ssk)->portid) {
1618 return ERR_PTR(-ECONNREFUSED);
1623 struct sock *netlink_getsockbyfilp(struct file *filp)
1625 struct inode *inode = file_inode(filp);
1628 if (!S_ISSOCK(inode->i_mode))
1629 return ERR_PTR(-ENOTSOCK);
1631 sock = SOCKET_I(inode)->sk;
1632 if (sock->sk_family != AF_NETLINK)
1633 return ERR_PTR(-EINVAL);
1639 static struct sk_buff *netlink_alloc_large_skb(unsigned int size,
1642 struct sk_buff *skb;
1645 if (size <= NLMSG_GOODSIZE || broadcast)
1646 return alloc_skb(size, GFP_KERNEL);
1648 size = SKB_DATA_ALIGN(size) +
1649 SKB_DATA_ALIGN(sizeof(struct skb_shared_info));
1651 data = vmalloc(size);
1655 skb = __build_skb(data, size);
1659 skb->destructor = netlink_skb_destructor;
1665 * Attach a skb to a netlink socket.
1666 * The caller must hold a reference to the destination socket. On error, the
1667 * reference is dropped. The skb is not send to the destination, just all
1668 * all error checks are performed and memory in the queue is reserved.
1670 * < 0: error. skb freed, reference to sock dropped.
1672 * 1: repeat lookup - reference dropped while waiting for socket memory.
1674 int netlink_attachskb(struct sock *sk, struct sk_buff *skb,
1675 long *timeo, struct sock *ssk)
1677 struct netlink_sock *nlk;
1681 if ((atomic_read(&sk->sk_rmem_alloc) > sk->sk_rcvbuf ||
1682 test_bit(NETLINK_S_CONGESTED, &nlk->state)) &&
1683 !netlink_skb_is_mmaped(skb)) {
1684 DECLARE_WAITQUEUE(wait, current);
1686 if (!ssk || netlink_is_kernel(ssk))
1687 netlink_overrun(sk);
1693 __set_current_state(TASK_INTERRUPTIBLE);
1694 add_wait_queue(&nlk->wait, &wait);
1696 if ((atomic_read(&sk->sk_rmem_alloc) > sk->sk_rcvbuf ||
1697 test_bit(NETLINK_S_CONGESTED, &nlk->state)) &&
1698 !sock_flag(sk, SOCK_DEAD))
1699 *timeo = schedule_timeout(*timeo);
1701 __set_current_state(TASK_RUNNING);
1702 remove_wait_queue(&nlk->wait, &wait);
1705 if (signal_pending(current)) {
1707 return sock_intr_errno(*timeo);
1711 netlink_skb_set_owner_r(skb, sk);
1715 static int __netlink_sendskb(struct sock *sk, struct sk_buff *skb)
1719 netlink_deliver_tap(skb);
1721 #ifdef CONFIG_NETLINK_MMAP
1722 if (netlink_skb_is_mmaped(skb))
1723 netlink_queue_mmaped_skb(sk, skb);
1724 else if (netlink_rx_is_mmaped(sk))
1725 netlink_ring_set_copied(sk, skb);
1727 #endif /* CONFIG_NETLINK_MMAP */
1728 skb_queue_tail(&sk->sk_receive_queue, skb);
1729 sk->sk_data_ready(sk);
1733 int netlink_sendskb(struct sock *sk, struct sk_buff *skb)
1735 int len = __netlink_sendskb(sk, skb);
1741 void netlink_detachskb(struct sock *sk, struct sk_buff *skb)
1747 static struct sk_buff *netlink_trim(struct sk_buff *skb, gfp_t allocation)
1751 WARN_ON(skb->sk != NULL);
1752 if (netlink_skb_is_mmaped(skb))
1755 delta = skb->end - skb->tail;
1756 if (is_vmalloc_addr(skb->head) || delta * 2 < skb->truesize)
1759 if (skb_shared(skb)) {
1760 struct sk_buff *nskb = skb_clone(skb, allocation);
1767 if (!pskb_expand_head(skb, 0, -delta, allocation))
1768 skb->truesize -= delta;
1773 static int netlink_unicast_kernel(struct sock *sk, struct sk_buff *skb,
1777 struct netlink_sock *nlk = nlk_sk(sk);
1779 ret = -ECONNREFUSED;
1780 if (nlk->netlink_rcv != NULL) {
1782 netlink_skb_set_owner_r(skb, sk);
1783 NETLINK_CB(skb).sk = ssk;
1784 netlink_deliver_tap_kernel(sk, ssk, skb);
1785 nlk->netlink_rcv(skb);
1794 int netlink_unicast(struct sock *ssk, struct sk_buff *skb,
1795 u32 portid, int nonblock)
1801 skb = netlink_trim(skb, gfp_any());
1803 timeo = sock_sndtimeo(ssk, nonblock);
1805 sk = netlink_getsockbyportid(ssk, portid);
1810 if (netlink_is_kernel(sk))
1811 return netlink_unicast_kernel(sk, skb, ssk);
1813 if (sk_filter(sk, skb)) {
1820 err = netlink_attachskb(sk, skb, &timeo, ssk);
1826 return netlink_sendskb(sk, skb);
1828 EXPORT_SYMBOL(netlink_unicast);
1830 struct sk_buff *netlink_alloc_skb(struct sock *ssk, unsigned int size,
1831 u32 dst_portid, gfp_t gfp_mask)
1833 #ifdef CONFIG_NETLINK_MMAP
1834 struct sock *sk = NULL;
1835 struct sk_buff *skb;
1836 struct netlink_ring *ring;
1837 struct nl_mmap_hdr *hdr;
1838 unsigned int maxlen;
1840 sk = netlink_getsockbyportid(ssk, dst_portid);
1844 ring = &nlk_sk(sk)->rx_ring;
1845 /* fast-path without atomic ops for common case: non-mmaped receiver */
1846 if (ring->pg_vec == NULL)
1849 if (ring->frame_size - NL_MMAP_HDRLEN < size)
1852 skb = alloc_skb_head(gfp_mask);
1856 spin_lock_bh(&sk->sk_receive_queue.lock);
1857 /* check again under lock */
1858 if (ring->pg_vec == NULL)
1861 /* check again under lock */
1862 maxlen = ring->frame_size - NL_MMAP_HDRLEN;
1866 netlink_forward_ring(ring);
1867 hdr = netlink_current_frame(ring, NL_MMAP_STATUS_UNUSED);
1870 netlink_ring_setup_skb(skb, sk, ring, hdr);
1871 netlink_set_status(hdr, NL_MMAP_STATUS_RESERVED);
1872 atomic_inc(&ring->pending);
1873 netlink_increment_head(ring);
1875 spin_unlock_bh(&sk->sk_receive_queue.lock);
1880 spin_unlock_bh(&sk->sk_receive_queue.lock);
1881 netlink_overrun(sk);
1888 spin_unlock_bh(&sk->sk_receive_queue.lock);
1893 return alloc_skb(size, gfp_mask);
1895 EXPORT_SYMBOL_GPL(netlink_alloc_skb);
1897 int netlink_has_listeners(struct sock *sk, unsigned int group)
1900 struct listeners *listeners;
1902 BUG_ON(!netlink_is_kernel(sk));
1905 listeners = rcu_dereference(nl_table[sk->sk_protocol].listeners);
1907 if (listeners && group - 1 < nl_table[sk->sk_protocol].groups)
1908 res = test_bit(group - 1, listeners->masks);
1914 EXPORT_SYMBOL_GPL(netlink_has_listeners);
1916 static int netlink_broadcast_deliver(struct sock *sk, struct sk_buff *skb)
1918 struct netlink_sock *nlk = nlk_sk(sk);
1920 if (atomic_read(&sk->sk_rmem_alloc) <= sk->sk_rcvbuf &&
1921 !test_bit(NETLINK_S_CONGESTED, &nlk->state)) {
1922 netlink_skb_set_owner_r(skb, sk);
1923 __netlink_sendskb(sk, skb);
1924 return atomic_read(&sk->sk_rmem_alloc) > (sk->sk_rcvbuf >> 1);
1929 struct netlink_broadcast_data {
1930 struct sock *exclude_sk;
1935 int delivery_failure;
1939 struct sk_buff *skb, *skb2;
1940 int (*tx_filter)(struct sock *dsk, struct sk_buff *skb, void *data);
1944 static void do_one_broadcast(struct sock *sk,
1945 struct netlink_broadcast_data *p)
1947 struct netlink_sock *nlk = nlk_sk(sk);
1950 if (p->exclude_sk == sk)
1953 if (nlk->portid == p->portid || p->group - 1 >= nlk->ngroups ||
1954 !test_bit(p->group - 1, nlk->groups))
1957 if (!net_eq(sock_net(sk), p->net)) {
1958 if (!(nlk->flags & NETLINK_F_LISTEN_ALL_NSID))
1961 if (!peernet_has_id(sock_net(sk), p->net))
1964 if (!file_ns_capable(sk->sk_socket->file, p->net->user_ns,
1970 netlink_overrun(sk);
1975 if (p->skb2 == NULL) {
1976 if (skb_shared(p->skb)) {
1977 p->skb2 = skb_clone(p->skb, p->allocation);
1979 p->skb2 = skb_get(p->skb);
1981 * skb ownership may have been set when
1982 * delivered to a previous socket.
1984 skb_orphan(p->skb2);
1987 if (p->skb2 == NULL) {
1988 netlink_overrun(sk);
1989 /* Clone failed. Notify ALL listeners. */
1991 if (nlk->flags & NETLINK_F_BROADCAST_SEND_ERROR)
1992 p->delivery_failure = 1;
1995 if (p->tx_filter && p->tx_filter(sk, p->skb2, p->tx_data)) {
2000 if (sk_filter(sk, p->skb2)) {
2005 NETLINK_CB(p->skb2).nsid = peernet2id(sock_net(sk), p->net);
2006 NETLINK_CB(p->skb2).nsid_is_set = true;
2007 val = netlink_broadcast_deliver(sk, p->skb2);
2009 netlink_overrun(sk);
2010 if (nlk->flags & NETLINK_F_BROADCAST_SEND_ERROR)
2011 p->delivery_failure = 1;
2013 p->congested |= val;
2021 int netlink_broadcast_filtered(struct sock *ssk, struct sk_buff *skb, u32 portid,
2022 u32 group, gfp_t allocation,
2023 int (*filter)(struct sock *dsk, struct sk_buff *skb, void *data),
2026 struct net *net = sock_net(ssk);
2027 struct netlink_broadcast_data info;
2030 skb = netlink_trim(skb, allocation);
2032 info.exclude_sk = ssk;
2034 info.portid = portid;
2037 info.delivery_failure = 0;
2040 info.allocation = allocation;
2043 info.tx_filter = filter;
2044 info.tx_data = filter_data;
2046 /* While we sleep in clone, do not allow to change socket list */
2048 netlink_lock_table();
2050 sk_for_each_bound(sk, &nl_table[ssk->sk_protocol].mc_list)
2051 do_one_broadcast(sk, &info);
2055 netlink_unlock_table();
2057 if (info.delivery_failure) {
2058 kfree_skb(info.skb2);
2061 consume_skb(info.skb2);
2063 if (info.delivered) {
2064 if (info.congested && (allocation & __GFP_WAIT))
2070 EXPORT_SYMBOL(netlink_broadcast_filtered);
2072 int netlink_broadcast(struct sock *ssk, struct sk_buff *skb, u32 portid,
2073 u32 group, gfp_t allocation)
2075 return netlink_broadcast_filtered(ssk, skb, portid, group, allocation,
2078 EXPORT_SYMBOL(netlink_broadcast);
2080 struct netlink_set_err_data {
2081 struct sock *exclude_sk;
2087 static int do_one_set_err(struct sock *sk, struct netlink_set_err_data *p)
2089 struct netlink_sock *nlk = nlk_sk(sk);
2092 if (sk == p->exclude_sk)
2095 if (!net_eq(sock_net(sk), sock_net(p->exclude_sk)))
2098 if (nlk->portid == p->portid || p->group - 1 >= nlk->ngroups ||
2099 !test_bit(p->group - 1, nlk->groups))
2102 if (p->code == ENOBUFS && nlk->flags & NETLINK_F_RECV_NO_ENOBUFS) {
2107 sk->sk_err = p->code;
2108 sk->sk_error_report(sk);
2114 * netlink_set_err - report error to broadcast listeners
2115 * @ssk: the kernel netlink socket, as returned by netlink_kernel_create()
2116 * @portid: the PORTID of a process that we want to skip (if any)
2117 * @group: the broadcast group that will notice the error
2118 * @code: error code, must be negative (as usual in kernelspace)
2120 * This function returns the number of broadcast listeners that have set the
2121 * NETLINK_NO_ENOBUFS socket option.
2123 int netlink_set_err(struct sock *ssk, u32 portid, u32 group, int code)
2125 struct netlink_set_err_data info;
2129 info.exclude_sk = ssk;
2130 info.portid = portid;
2132 /* sk->sk_err wants a positive error value */
2135 read_lock(&nl_table_lock);
2137 sk_for_each_bound(sk, &nl_table[ssk->sk_protocol].mc_list)
2138 ret += do_one_set_err(sk, &info);
2140 read_unlock(&nl_table_lock);
2143 EXPORT_SYMBOL(netlink_set_err);
2145 /* must be called with netlink table grabbed */
2146 static void netlink_update_socket_mc(struct netlink_sock *nlk,
2150 int old, new = !!is_new, subscriptions;
2152 old = test_bit(group - 1, nlk->groups);
2153 subscriptions = nlk->subscriptions - old + new;
2155 __set_bit(group - 1, nlk->groups);
2157 __clear_bit(group - 1, nlk->groups);
2158 netlink_update_subscriptions(&nlk->sk, subscriptions);
2159 netlink_update_listeners(&nlk->sk);
2162 static int netlink_setsockopt(struct socket *sock, int level, int optname,
2163 char __user *optval, unsigned int optlen)
2165 struct sock *sk = sock->sk;
2166 struct netlink_sock *nlk = nlk_sk(sk);
2167 unsigned int val = 0;
2170 if (level != SOL_NETLINK)
2171 return -ENOPROTOOPT;
2173 if (optname != NETLINK_RX_RING && optname != NETLINK_TX_RING &&
2174 optlen >= sizeof(int) &&
2175 get_user(val, (unsigned int __user *)optval))
2179 case NETLINK_PKTINFO:
2181 nlk->flags |= NETLINK_F_RECV_PKTINFO;
2183 nlk->flags &= ~NETLINK_F_RECV_PKTINFO;
2186 case NETLINK_ADD_MEMBERSHIP:
2187 case NETLINK_DROP_MEMBERSHIP: {
2188 if (!netlink_allowed(sock, NL_CFG_F_NONROOT_RECV))
2190 err = netlink_realloc_groups(sk);
2193 if (!val || val - 1 >= nlk->ngroups)
2195 if (optname == NETLINK_ADD_MEMBERSHIP && nlk->netlink_bind) {
2196 err = nlk->netlink_bind(sock_net(sk), val);
2200 netlink_table_grab();
2201 netlink_update_socket_mc(nlk, val,
2202 optname == NETLINK_ADD_MEMBERSHIP);
2203 netlink_table_ungrab();
2204 if (optname == NETLINK_DROP_MEMBERSHIP && nlk->netlink_unbind)
2205 nlk->netlink_unbind(sock_net(sk), val);
2210 case NETLINK_BROADCAST_ERROR:
2212 nlk->flags |= NETLINK_F_BROADCAST_SEND_ERROR;
2214 nlk->flags &= ~NETLINK_F_BROADCAST_SEND_ERROR;
2217 case NETLINK_NO_ENOBUFS:
2219 nlk->flags |= NETLINK_F_RECV_NO_ENOBUFS;
2220 clear_bit(NETLINK_S_CONGESTED, &nlk->state);
2221 wake_up_interruptible(&nlk->wait);
2223 nlk->flags &= ~NETLINK_F_RECV_NO_ENOBUFS;
2227 #ifdef CONFIG_NETLINK_MMAP
2228 case NETLINK_RX_RING:
2229 case NETLINK_TX_RING: {
2230 struct nl_mmap_req req;
2232 /* Rings might consume more memory than queue limits, require
2235 if (!capable(CAP_NET_ADMIN))
2237 if (optlen < sizeof(req))
2239 if (copy_from_user(&req, optval, sizeof(req)))
2241 err = netlink_set_ring(sk, &req,
2242 optname == NETLINK_TX_RING);
2245 #endif /* CONFIG_NETLINK_MMAP */
2246 case NETLINK_LISTEN_ALL_NSID:
2247 if (!ns_capable(sock_net(sk)->user_ns, CAP_NET_BROADCAST))
2251 nlk->flags |= NETLINK_F_LISTEN_ALL_NSID;
2253 nlk->flags &= ~NETLINK_F_LISTEN_ALL_NSID;
2262 static int netlink_getsockopt(struct socket *sock, int level, int optname,
2263 char __user *optval, int __user *optlen)
2265 struct sock *sk = sock->sk;
2266 struct netlink_sock *nlk = nlk_sk(sk);
2269 if (level != SOL_NETLINK)
2270 return -ENOPROTOOPT;
2272 if (get_user(len, optlen))
2278 case NETLINK_PKTINFO:
2279 if (len < sizeof(int))
2282 val = nlk->flags & NETLINK_F_RECV_PKTINFO ? 1 : 0;
2283 if (put_user(len, optlen) ||
2284 put_user(val, optval))
2288 case NETLINK_BROADCAST_ERROR:
2289 if (len < sizeof(int))
2292 val = nlk->flags & NETLINK_F_BROADCAST_SEND_ERROR ? 1 : 0;
2293 if (put_user(len, optlen) ||
2294 put_user(val, optval))
2298 case NETLINK_NO_ENOBUFS:
2299 if (len < sizeof(int))
2302 val = nlk->flags & NETLINK_F_RECV_NO_ENOBUFS ? 1 : 0;
2303 if (put_user(len, optlen) ||
2304 put_user(val, optval))
2308 case NETLINK_LIST_MEMBERSHIPS: {
2309 int pos, idx, shift;
2312 netlink_table_grab();
2313 for (pos = 0; pos * 8 < nlk->ngroups; pos += sizeof(u32)) {
2314 if (len - pos < sizeof(u32))
2317 idx = pos / sizeof(unsigned long);
2318 shift = (pos % sizeof(unsigned long)) * 8;
2319 if (put_user((u32)(nlk->groups[idx] >> shift),
2320 (u32 __user *)(optval + pos))) {
2325 if (put_user(ALIGN(nlk->ngroups / 8, sizeof(u32)), optlen))
2327 netlink_table_ungrab();
2336 static void netlink_cmsg_recv_pktinfo(struct msghdr *msg, struct sk_buff *skb)
2338 struct nl_pktinfo info;
2340 info.group = NETLINK_CB(skb).dst_group;
2341 put_cmsg(msg, SOL_NETLINK, NETLINK_PKTINFO, sizeof(info), &info);
2344 static void netlink_cmsg_listen_all_nsid(struct sock *sk, struct msghdr *msg,
2345 struct sk_buff *skb)
2347 if (!NETLINK_CB(skb).nsid_is_set)
2350 put_cmsg(msg, SOL_NETLINK, NETLINK_LISTEN_ALL_NSID, sizeof(int),
2351 &NETLINK_CB(skb).nsid);
2354 static int netlink_sendmsg(struct socket *sock, struct msghdr *msg, size_t len)
2356 struct sock *sk = sock->sk;
2357 struct netlink_sock *nlk = nlk_sk(sk);
2358 DECLARE_SOCKADDR(struct sockaddr_nl *, addr, msg->msg_name);
2361 struct sk_buff *skb;
2363 struct scm_cookie scm;
2364 u32 netlink_skb_flags = 0;
2366 if (msg->msg_flags&MSG_OOB)
2369 err = scm_send(sock, msg, &scm, true);
2373 if (msg->msg_namelen) {
2375 if (addr->nl_family != AF_NETLINK)
2377 dst_portid = addr->nl_pid;
2378 dst_group = ffs(addr->nl_groups);
2380 if ((dst_group || dst_portid) &&
2381 !netlink_allowed(sock, NL_CFG_F_NONROOT_SEND))
2383 netlink_skb_flags |= NETLINK_SKB_DST;
2385 dst_portid = nlk->dst_portid;
2386 dst_group = nlk->dst_group;
2390 err = netlink_autobind(sock);
2395 /* It's a really convoluted way for userland to ask for mmaped
2396 * sendmsg(), but that's what we've got...
2398 if (netlink_tx_is_mmaped(sk) &&
2399 msg->msg_iter.type == ITER_IOVEC &&
2400 msg->msg_iter.nr_segs == 1 &&
2401 msg->msg_iter.iov->iov_base == NULL) {
2402 err = netlink_mmap_sendmsg(sk, msg, dst_portid, dst_group,
2408 if (len > sk->sk_sndbuf - 32)
2411 skb = netlink_alloc_large_skb(len, dst_group);
2415 NETLINK_CB(skb).portid = nlk->portid;
2416 NETLINK_CB(skb).dst_group = dst_group;
2417 NETLINK_CB(skb).creds = scm.creds;
2418 NETLINK_CB(skb).flags = netlink_skb_flags;
2421 if (memcpy_from_msg(skb_put(skb, len), msg, len)) {
2426 err = security_netlink_send(sk, skb);
2433 atomic_inc(&skb->users);
2434 netlink_broadcast(sk, skb, dst_portid, dst_group, GFP_KERNEL);
2436 err = netlink_unicast(sk, skb, dst_portid, msg->msg_flags&MSG_DONTWAIT);
2443 static int netlink_recvmsg(struct socket *sock, struct msghdr *msg, size_t len,
2446 struct scm_cookie scm;
2447 struct sock *sk = sock->sk;
2448 struct netlink_sock *nlk = nlk_sk(sk);
2449 int noblock = flags&MSG_DONTWAIT;
2451 struct sk_buff *skb, *data_skb;
2459 skb = skb_recv_datagram(sk, flags, noblock, &err);
2465 #ifdef CONFIG_COMPAT_NETLINK_MESSAGES
2466 if (unlikely(skb_shinfo(skb)->frag_list)) {
2468 * If this skb has a frag_list, then here that means that we
2469 * will have to use the frag_list skb's data for compat tasks
2470 * and the regular skb's data for normal (non-compat) tasks.
2472 * If we need to send the compat skb, assign it to the
2473 * 'data_skb' variable so that it will be used below for data
2474 * copying. We keep 'skb' for everything else, including
2475 * freeing both later.
2477 if (flags & MSG_CMSG_COMPAT)
2478 data_skb = skb_shinfo(skb)->frag_list;
2482 /* Record the max length of recvmsg() calls for future allocations */
2483 nlk->max_recvmsg_len = max(nlk->max_recvmsg_len, len);
2484 nlk->max_recvmsg_len = min_t(size_t, nlk->max_recvmsg_len,
2487 copied = data_skb->len;
2489 msg->msg_flags |= MSG_TRUNC;
2493 skb_reset_transport_header(data_skb);
2494 err = skb_copy_datagram_msg(data_skb, 0, msg, copied);
2496 if (msg->msg_name) {
2497 DECLARE_SOCKADDR(struct sockaddr_nl *, addr, msg->msg_name);
2498 addr->nl_family = AF_NETLINK;
2500 addr->nl_pid = NETLINK_CB(skb).portid;
2501 addr->nl_groups = netlink_group_mask(NETLINK_CB(skb).dst_group);
2502 msg->msg_namelen = sizeof(*addr);
2505 if (nlk->flags & NETLINK_F_RECV_PKTINFO)
2506 netlink_cmsg_recv_pktinfo(msg, skb);
2507 if (nlk->flags & NETLINK_F_LISTEN_ALL_NSID)
2508 netlink_cmsg_listen_all_nsid(sk, msg, skb);
2510 memset(&scm, 0, sizeof(scm));
2511 scm.creds = *NETLINK_CREDS(skb);
2512 if (flags & MSG_TRUNC)
2513 copied = data_skb->len;
2515 skb_free_datagram(sk, skb);
2517 if (nlk->cb_running &&
2518 atomic_read(&sk->sk_rmem_alloc) <= sk->sk_rcvbuf / 2) {
2519 ret = netlink_dump(sk);
2522 sk->sk_error_report(sk);
2526 scm_recv(sock, msg, &scm, flags);
2528 netlink_rcv_wake(sk);
2529 return err ? : copied;
2532 static void netlink_data_ready(struct sock *sk)
2538 * We export these functions to other modules. They provide a
2539 * complete set of kernel non-blocking support for message
2544 __netlink_kernel_create(struct net *net, int unit, struct module *module,
2545 struct netlink_kernel_cfg *cfg)
2547 struct socket *sock;
2549 struct netlink_sock *nlk;
2550 struct listeners *listeners = NULL;
2551 struct mutex *cb_mutex = cfg ? cfg->cb_mutex : NULL;
2552 unsigned int groups;
2556 if (unit < 0 || unit >= MAX_LINKS)
2559 if (sock_create_lite(PF_NETLINK, SOCK_DGRAM, unit, &sock))
2562 if (__netlink_create(net, sock, cb_mutex, unit, 1) < 0)
2563 goto out_sock_release_nosk;
2567 if (!cfg || cfg->groups < 32)
2570 groups = cfg->groups;
2572 listeners = kzalloc(sizeof(*listeners) + NLGRPSZ(groups), GFP_KERNEL);
2574 goto out_sock_release;
2576 sk->sk_data_ready = netlink_data_ready;
2577 if (cfg && cfg->input)
2578 nlk_sk(sk)->netlink_rcv = cfg->input;
2580 if (netlink_insert(sk, 0))
2581 goto out_sock_release;
2584 nlk->flags |= NETLINK_F_KERNEL_SOCKET;
2586 netlink_table_grab();
2587 if (!nl_table[unit].registered) {
2588 nl_table[unit].groups = groups;
2589 rcu_assign_pointer(nl_table[unit].listeners, listeners);
2590 nl_table[unit].cb_mutex = cb_mutex;
2591 nl_table[unit].module = module;
2593 nl_table[unit].bind = cfg->bind;
2594 nl_table[unit].unbind = cfg->unbind;
2595 nl_table[unit].flags = cfg->flags;
2597 nl_table[unit].compare = cfg->compare;
2599 nl_table[unit].registered = 1;
2602 nl_table[unit].registered++;
2604 netlink_table_ungrab();
2609 netlink_kernel_release(sk);
2612 out_sock_release_nosk:
2616 EXPORT_SYMBOL(__netlink_kernel_create);
2619 netlink_kernel_release(struct sock *sk)
2621 if (sk == NULL || sk->sk_socket == NULL)
2624 sock_release(sk->sk_socket);
2626 EXPORT_SYMBOL(netlink_kernel_release);
2628 int __netlink_change_ngroups(struct sock *sk, unsigned int groups)
2630 struct listeners *new, *old;
2631 struct netlink_table *tbl = &nl_table[sk->sk_protocol];
2636 if (NLGRPSZ(tbl->groups) < NLGRPSZ(groups)) {
2637 new = kzalloc(sizeof(*new) + NLGRPSZ(groups), GFP_ATOMIC);
2640 old = nl_deref_protected(tbl->listeners);
2641 memcpy(new->masks, old->masks, NLGRPSZ(tbl->groups));
2642 rcu_assign_pointer(tbl->listeners, new);
2644 kfree_rcu(old, rcu);
2646 tbl->groups = groups;
2652 * netlink_change_ngroups - change number of multicast groups
2654 * This changes the number of multicast groups that are available
2655 * on a certain netlink family. Note that it is not possible to
2656 * change the number of groups to below 32. Also note that it does
2657 * not implicitly call netlink_clear_multicast_users() when the
2658 * number of groups is reduced.
2660 * @sk: The kernel netlink socket, as returned by netlink_kernel_create().
2661 * @groups: The new number of groups.
2663 int netlink_change_ngroups(struct sock *sk, unsigned int groups)
2667 netlink_table_grab();
2668 err = __netlink_change_ngroups(sk, groups);
2669 netlink_table_ungrab();
2674 void __netlink_clear_multicast_users(struct sock *ksk, unsigned int group)
2677 struct netlink_table *tbl = &nl_table[ksk->sk_protocol];
2679 sk_for_each_bound(sk, &tbl->mc_list)
2680 netlink_update_socket_mc(nlk_sk(sk), group, 0);
2684 __nlmsg_put(struct sk_buff *skb, u32 portid, u32 seq, int type, int len, int flags)
2686 struct nlmsghdr *nlh;
2687 int size = nlmsg_msg_size(len);
2689 nlh = (struct nlmsghdr *)skb_put(skb, NLMSG_ALIGN(size));
2690 nlh->nlmsg_type = type;
2691 nlh->nlmsg_len = size;
2692 nlh->nlmsg_flags = flags;
2693 nlh->nlmsg_pid = portid;
2694 nlh->nlmsg_seq = seq;
2695 if (!__builtin_constant_p(size) || NLMSG_ALIGN(size) - size != 0)
2696 memset(nlmsg_data(nlh) + len, 0, NLMSG_ALIGN(size) - size);
2699 EXPORT_SYMBOL(__nlmsg_put);
2702 * It looks a bit ugly.
2703 * It would be better to create kernel thread.
2706 static int netlink_dump(struct sock *sk)
2708 struct netlink_sock *nlk = nlk_sk(sk);
2709 struct netlink_callback *cb;
2710 struct sk_buff *skb = NULL;
2711 struct nlmsghdr *nlh;
2712 int len, err = -ENOBUFS;
2715 mutex_lock(nlk->cb_mutex);
2716 if (!nlk->cb_running) {
2722 alloc_size = max_t(int, cb->min_dump_alloc, NLMSG_GOODSIZE);
2724 if (!netlink_rx_is_mmaped(sk) &&
2725 atomic_read(&sk->sk_rmem_alloc) >= sk->sk_rcvbuf)
2728 /* NLMSG_GOODSIZE is small to avoid high order allocations being
2729 * required, but it makes sense to _attempt_ a 16K bytes allocation
2730 * to reduce number of system calls on dump operations, if user
2731 * ever provided a big enough buffer.
2733 if (alloc_size < nlk->max_recvmsg_len) {
2734 skb = netlink_alloc_skb(sk,
2735 nlk->max_recvmsg_len,
2740 /* available room should be exact amount to avoid MSG_TRUNC */
2742 skb_reserve(skb, skb_tailroom(skb) -
2743 nlk->max_recvmsg_len);
2746 skb = netlink_alloc_skb(sk, alloc_size, nlk->portid,
2750 netlink_skb_set_owner_r(skb, sk);
2752 len = cb->dump(skb, cb);
2755 mutex_unlock(nlk->cb_mutex);
2757 if (sk_filter(sk, skb))
2760 __netlink_sendskb(sk, skb);
2764 nlh = nlmsg_put_answer(skb, cb, NLMSG_DONE, sizeof(len), NLM_F_MULTI);
2768 nl_dump_check_consistent(cb, nlh);
2770 memcpy(nlmsg_data(nlh), &len, sizeof(len));
2772 if (sk_filter(sk, skb))
2775 __netlink_sendskb(sk, skb);
2780 nlk->cb_running = false;
2781 mutex_unlock(nlk->cb_mutex);
2782 module_put(cb->module);
2783 consume_skb(cb->skb);
2787 mutex_unlock(nlk->cb_mutex);
2792 int __netlink_dump_start(struct sock *ssk, struct sk_buff *skb,
2793 const struct nlmsghdr *nlh,
2794 struct netlink_dump_control *control)
2796 struct netlink_callback *cb;
2798 struct netlink_sock *nlk;
2801 /* Memory mapped dump requests need to be copied to avoid looping
2802 * on the pending state in netlink_mmap_sendmsg() while the CB hold
2803 * a reference to the skb.
2805 if (netlink_skb_is_mmaped(skb)) {
2806 skb = skb_copy(skb, GFP_KERNEL);
2810 atomic_inc(&skb->users);
2812 sk = netlink_lookup(sock_net(ssk), ssk->sk_protocol, NETLINK_CB(skb).portid);
2814 ret = -ECONNREFUSED;
2819 mutex_lock(nlk->cb_mutex);
2820 /* A dump is in progress... */
2821 if (nlk->cb_running) {
2825 /* add reference of module which cb->dump belongs to */
2826 if (!try_module_get(control->module)) {
2827 ret = -EPROTONOSUPPORT;
2832 memset(cb, 0, sizeof(*cb));
2833 cb->dump = control->dump;
2834 cb->done = control->done;
2836 cb->data = control->data;
2837 cb->module = control->module;
2838 cb->min_dump_alloc = control->min_dump_alloc;
2841 nlk->cb_running = true;
2843 mutex_unlock(nlk->cb_mutex);
2845 ret = netlink_dump(sk);
2851 /* We successfully started a dump, by returning -EINTR we
2852 * signal not to send ACK even if it was requested.
2858 mutex_unlock(nlk->cb_mutex);
2863 EXPORT_SYMBOL(__netlink_dump_start);
2865 void netlink_ack(struct sk_buff *in_skb, struct nlmsghdr *nlh, int err)
2867 struct sk_buff *skb;
2868 struct nlmsghdr *rep;
2869 struct nlmsgerr *errmsg;
2870 size_t payload = sizeof(*errmsg);
2872 /* error messages get the original request appened */
2874 payload += nlmsg_len(nlh);
2876 skb = netlink_alloc_skb(in_skb->sk, nlmsg_total_size(payload),
2877 NETLINK_CB(in_skb).portid, GFP_KERNEL);
2881 sk = netlink_lookup(sock_net(in_skb->sk),
2882 in_skb->sk->sk_protocol,
2883 NETLINK_CB(in_skb).portid);
2885 sk->sk_err = ENOBUFS;
2886 sk->sk_error_report(sk);
2892 rep = __nlmsg_put(skb, NETLINK_CB(in_skb).portid, nlh->nlmsg_seq,
2893 NLMSG_ERROR, payload, 0);
2894 errmsg = nlmsg_data(rep);
2895 errmsg->error = err;
2896 memcpy(&errmsg->msg, nlh, err ? nlh->nlmsg_len : sizeof(*nlh));
2897 netlink_unicast(in_skb->sk, skb, NETLINK_CB(in_skb).portid, MSG_DONTWAIT);
2899 EXPORT_SYMBOL(netlink_ack);
2901 int netlink_rcv_skb(struct sk_buff *skb, int (*cb)(struct sk_buff *,
2904 struct nlmsghdr *nlh;
2907 while (skb->len >= nlmsg_total_size(0)) {
2910 nlh = nlmsg_hdr(skb);
2913 if (nlh->nlmsg_len < NLMSG_HDRLEN || skb->len < nlh->nlmsg_len)
2916 /* Only requests are handled by the kernel */
2917 if (!(nlh->nlmsg_flags & NLM_F_REQUEST))
2920 /* Skip control messages */
2921 if (nlh->nlmsg_type < NLMSG_MIN_TYPE)
2929 if (nlh->nlmsg_flags & NLM_F_ACK || err)
2930 netlink_ack(skb, nlh, err);
2933 msglen = NLMSG_ALIGN(nlh->nlmsg_len);
2934 if (msglen > skb->len)
2936 skb_pull(skb, msglen);
2941 EXPORT_SYMBOL(netlink_rcv_skb);
2944 * nlmsg_notify - send a notification netlink message
2945 * @sk: netlink socket to use
2946 * @skb: notification message
2947 * @portid: destination netlink portid for reports or 0
2948 * @group: destination multicast group or 0
2949 * @report: 1 to report back, 0 to disable
2950 * @flags: allocation flags
2952 int nlmsg_notify(struct sock *sk, struct sk_buff *skb, u32 portid,
2953 unsigned int group, int report, gfp_t flags)
2958 int exclude_portid = 0;
2961 atomic_inc(&skb->users);
2962 exclude_portid = portid;
2965 /* errors reported via destination sk->sk_err, but propagate
2966 * delivery errors if NETLINK_BROADCAST_ERROR flag is set */
2967 err = nlmsg_multicast(sk, skb, exclude_portid, group, flags);
2973 err2 = nlmsg_unicast(sk, skb, portid);
2974 if (!err || err == -ESRCH)
2980 EXPORT_SYMBOL(nlmsg_notify);
2982 #ifdef CONFIG_PROC_FS
2983 struct nl_seq_iter {
2984 struct seq_net_private p;
2985 struct rhashtable_iter hti;
2989 static int netlink_walk_start(struct nl_seq_iter *iter)
2993 err = rhashtable_walk_init(&nl_table[iter->link].hash, &iter->hti);
2995 iter->link = MAX_LINKS;
2999 err = rhashtable_walk_start(&iter->hti);
3000 return err == -EAGAIN ? 0 : err;
3003 static void netlink_walk_stop(struct nl_seq_iter *iter)
3005 rhashtable_walk_stop(&iter->hti);
3006 rhashtable_walk_exit(&iter->hti);
3009 static void *__netlink_seq_next(struct seq_file *seq)
3011 struct nl_seq_iter *iter = seq->private;
3012 struct netlink_sock *nlk;
3018 nlk = rhashtable_walk_next(&iter->hti);
3021 if (PTR_ERR(nlk) == -EAGAIN)
3030 netlink_walk_stop(iter);
3031 if (++iter->link >= MAX_LINKS)
3034 err = netlink_walk_start(iter);
3036 return ERR_PTR(err);
3038 } while (sock_net(&nlk->sk) != seq_file_net(seq));
3043 static void *netlink_seq_start(struct seq_file *seq, loff_t *posp)
3045 struct nl_seq_iter *iter = seq->private;
3046 void *obj = SEQ_START_TOKEN;
3052 err = netlink_walk_start(iter);
3054 return ERR_PTR(err);
3056 for (pos = *posp; pos && obj && !IS_ERR(obj); pos--)
3057 obj = __netlink_seq_next(seq);
3062 static void *netlink_seq_next(struct seq_file *seq, void *v, loff_t *pos)
3065 return __netlink_seq_next(seq);
3068 static void netlink_seq_stop(struct seq_file *seq, void *v)
3070 struct nl_seq_iter *iter = seq->private;
3072 if (iter->link >= MAX_LINKS)
3075 netlink_walk_stop(iter);
3079 static int netlink_seq_show(struct seq_file *seq, void *v)
3081 if (v == SEQ_START_TOKEN) {
3083 "sk Eth Pid Groups "
3084 "Rmem Wmem Dump Locks Drops Inode\n");
3087 struct netlink_sock *nlk = nlk_sk(s);
3089 seq_printf(seq, "%pK %-3d %-6u %08x %-8d %-8d %d %-8d %-8d %-8lu\n",
3093 nlk->groups ? (u32)nlk->groups[0] : 0,
3094 sk_rmem_alloc_get(s),
3095 sk_wmem_alloc_get(s),
3097 atomic_read(&s->sk_refcnt),
3098 atomic_read(&s->sk_drops),
3106 static const struct seq_operations netlink_seq_ops = {
3107 .start = netlink_seq_start,
3108 .next = netlink_seq_next,
3109 .stop = netlink_seq_stop,
3110 .show = netlink_seq_show,
3114 static int netlink_seq_open(struct inode *inode, struct file *file)
3116 return seq_open_net(inode, file, &netlink_seq_ops,
3117 sizeof(struct nl_seq_iter));
3120 static const struct file_operations netlink_seq_fops = {
3121 .owner = THIS_MODULE,
3122 .open = netlink_seq_open,
3124 .llseek = seq_lseek,
3125 .release = seq_release_net,
3130 int netlink_register_notifier(struct notifier_block *nb)
3132 return atomic_notifier_chain_register(&netlink_chain, nb);
3134 EXPORT_SYMBOL(netlink_register_notifier);
3136 int netlink_unregister_notifier(struct notifier_block *nb)
3138 return atomic_notifier_chain_unregister(&netlink_chain, nb);
3140 EXPORT_SYMBOL(netlink_unregister_notifier);
3142 static const struct proto_ops netlink_ops = {
3143 .family = PF_NETLINK,
3144 .owner = THIS_MODULE,
3145 .release = netlink_release,
3146 .bind = netlink_bind,
3147 .connect = netlink_connect,
3148 .socketpair = sock_no_socketpair,
3149 .accept = sock_no_accept,
3150 .getname = netlink_getname,
3151 .poll = netlink_poll,
3152 .ioctl = sock_no_ioctl,
3153 .listen = sock_no_listen,
3154 .shutdown = sock_no_shutdown,
3155 .setsockopt = netlink_setsockopt,
3156 .getsockopt = netlink_getsockopt,
3157 .sendmsg = netlink_sendmsg,
3158 .recvmsg = netlink_recvmsg,
3159 .mmap = netlink_mmap,
3160 .sendpage = sock_no_sendpage,
3163 static const struct net_proto_family netlink_family_ops = {
3164 .family = PF_NETLINK,
3165 .create = netlink_create,
3166 .owner = THIS_MODULE, /* for consistency 8) */
3169 static int __net_init netlink_net_init(struct net *net)
3171 #ifdef CONFIG_PROC_FS
3172 if (!proc_create("netlink", 0, net->proc_net, &netlink_seq_fops))
3178 static void __net_exit netlink_net_exit(struct net *net)
3180 #ifdef CONFIG_PROC_FS
3181 remove_proc_entry("netlink", net->proc_net);
3185 static void __init netlink_add_usersock_entry(void)
3187 struct listeners *listeners;
3190 listeners = kzalloc(sizeof(*listeners) + NLGRPSZ(groups), GFP_KERNEL);
3192 panic("netlink_add_usersock_entry: Cannot allocate listeners\n");
3194 netlink_table_grab();
3196 nl_table[NETLINK_USERSOCK].groups = groups;
3197 rcu_assign_pointer(nl_table[NETLINK_USERSOCK].listeners, listeners);
3198 nl_table[NETLINK_USERSOCK].module = THIS_MODULE;
3199 nl_table[NETLINK_USERSOCK].registered = 1;
3200 nl_table[NETLINK_USERSOCK].flags = NL_CFG_F_NONROOT_SEND;
3202 netlink_table_ungrab();
3205 static struct pernet_operations __net_initdata netlink_net_ops = {
3206 .init = netlink_net_init,
3207 .exit = netlink_net_exit,
3210 static inline u32 netlink_hash(const void *data, u32 len, u32 seed)
3212 const struct netlink_sock *nlk = data;
3213 struct netlink_compare_arg arg;
3215 netlink_compare_arg_init(&arg, sock_net(&nlk->sk), nlk->portid);
3216 return jhash2((u32 *)&arg, netlink_compare_arg_len / sizeof(u32), seed);
3219 static const struct rhashtable_params netlink_rhashtable_params = {
3220 .head_offset = offsetof(struct netlink_sock, node),
3221 .key_len = netlink_compare_arg_len,
3222 .obj_hashfn = netlink_hash,
3223 .obj_cmpfn = netlink_compare,
3224 .automatic_shrinking = true,
3227 static int __init netlink_proto_init(void)
3230 int err = proto_register(&netlink_proto, 0);
3235 BUILD_BUG_ON(sizeof(struct netlink_skb_parms) > FIELD_SIZEOF(struct sk_buff, cb));
3237 nl_table = kcalloc(MAX_LINKS, sizeof(*nl_table), GFP_KERNEL);
3241 for (i = 0; i < MAX_LINKS; i++) {
3242 if (rhashtable_init(&nl_table[i].hash,
3243 &netlink_rhashtable_params) < 0) {
3245 rhashtable_destroy(&nl_table[i].hash);
3251 INIT_LIST_HEAD(&netlink_tap_all);
3253 netlink_add_usersock_entry();
3255 sock_register(&netlink_family_ops);
3256 register_pernet_subsys(&netlink_net_ops);
3257 /* The netlink device handler may be needed early. */
3262 panic("netlink_init: Cannot allocate nl_table\n");
3265 core_initcall(netlink_proto_init);