1 /* Netfilter messages via netlink socket. Allows for user space
2 * protocol helpers and general trouble making from userspace.
4 * (C) 2001 by Jay Schulist <jschlst@samba.org>,
5 * (C) 2002-2005 by Harald Welte <laforge@gnumonks.org>
6 * (C) 2005,2007 by Pablo Neira Ayuso <pablo@netfilter.org>
8 * Initial netfilter messages via netlink development funded and
9 * generally made possible by Network Robots, Inc. (www.networkrobots.com)
11 * Further development of this code funded by Astaro AG (http://www.astaro.com)
13 * This software may be used and distributed according to the terms
14 * of the GNU General Public License, incorporated herein by reference.
17 #include <linux/module.h>
18 #include <linux/types.h>
19 #include <linux/socket.h>
20 #include <linux/kernel.h>
21 #include <linux/string.h>
22 #include <linux/sockios.h>
23 #include <linux/net.h>
24 #include <linux/skbuff.h>
25 #include <asm/uaccess.h>
27 #include <linux/init.h>
29 #include <net/netlink.h>
30 #include <linux/netfilter/nfnetlink.h>
32 MODULE_LICENSE("GPL");
33 MODULE_AUTHOR("Harald Welte <laforge@netfilter.org>");
34 MODULE_ALIAS_NET_PF_PROTO(PF_NETLINK
, NETLINK_NETFILTER
);
36 static char __initdata nfversion
[] = "0.30";
40 const struct nfnetlink_subsystem __rcu
*subsys
;
41 } table
[NFNL_SUBSYS_COUNT
];
43 static const int nfnl_group2type
[NFNLGRP_MAX
+1] = {
44 [NFNLGRP_CONNTRACK_NEW
] = NFNL_SUBSYS_CTNETLINK
,
45 [NFNLGRP_CONNTRACK_UPDATE
] = NFNL_SUBSYS_CTNETLINK
,
46 [NFNLGRP_CONNTRACK_DESTROY
] = NFNL_SUBSYS_CTNETLINK
,
47 [NFNLGRP_CONNTRACK_EXP_NEW
] = NFNL_SUBSYS_CTNETLINK_EXP
,
48 [NFNLGRP_CONNTRACK_EXP_UPDATE
] = NFNL_SUBSYS_CTNETLINK_EXP
,
49 [NFNLGRP_CONNTRACK_EXP_DESTROY
] = NFNL_SUBSYS_CTNETLINK_EXP
,
50 [NFNLGRP_NFTABLES
] = NFNL_SUBSYS_NFTABLES
,
51 [NFNLGRP_ACCT_QUOTA
] = NFNL_SUBSYS_ACCT
,
54 void nfnl_lock(__u8 subsys_id
)
56 mutex_lock(&table
[subsys_id
].mutex
);
58 EXPORT_SYMBOL_GPL(nfnl_lock
);
60 void nfnl_unlock(__u8 subsys_id
)
62 mutex_unlock(&table
[subsys_id
].mutex
);
64 EXPORT_SYMBOL_GPL(nfnl_unlock
);
66 #ifdef CONFIG_PROVE_LOCKING
67 bool lockdep_nfnl_is_held(u8 subsys_id
)
69 return lockdep_is_held(&table
[subsys_id
].mutex
);
71 EXPORT_SYMBOL_GPL(lockdep_nfnl_is_held
);
74 int nfnetlink_subsys_register(const struct nfnetlink_subsystem
*n
)
76 nfnl_lock(n
->subsys_id
);
77 if (table
[n
->subsys_id
].subsys
) {
78 nfnl_unlock(n
->subsys_id
);
81 rcu_assign_pointer(table
[n
->subsys_id
].subsys
, n
);
82 nfnl_unlock(n
->subsys_id
);
86 EXPORT_SYMBOL_GPL(nfnetlink_subsys_register
);
88 int nfnetlink_subsys_unregister(const struct nfnetlink_subsystem
*n
)
90 nfnl_lock(n
->subsys_id
);
91 table
[n
->subsys_id
].subsys
= NULL
;
92 nfnl_unlock(n
->subsys_id
);
96 EXPORT_SYMBOL_GPL(nfnetlink_subsys_unregister
);
98 static inline const struct nfnetlink_subsystem
*nfnetlink_get_subsys(u_int16_t type
)
100 u_int8_t subsys_id
= NFNL_SUBSYS_ID(type
);
102 if (subsys_id
>= NFNL_SUBSYS_COUNT
)
105 return rcu_dereference(table
[subsys_id
].subsys
);
108 static inline const struct nfnl_callback
*
109 nfnetlink_find_client(u_int16_t type
, const struct nfnetlink_subsystem
*ss
)
111 u_int8_t cb_id
= NFNL_MSG_TYPE(type
);
113 if (cb_id
>= ss
->cb_count
)
116 return &ss
->cb
[cb_id
];
119 int nfnetlink_has_listeners(struct net
*net
, unsigned int group
)
121 return netlink_has_listeners(net
->nfnl
, group
);
123 EXPORT_SYMBOL_GPL(nfnetlink_has_listeners
);
125 struct sk_buff
*nfnetlink_alloc_skb(struct net
*net
, unsigned int size
,
126 u32 dst_portid
, gfp_t gfp_mask
)
128 return netlink_alloc_skb(net
->nfnl
, size
, dst_portid
, gfp_mask
);
130 EXPORT_SYMBOL_GPL(nfnetlink_alloc_skb
);
132 int nfnetlink_send(struct sk_buff
*skb
, struct net
*net
, u32 portid
,
133 unsigned int group
, int echo
, gfp_t flags
)
135 return nlmsg_notify(net
->nfnl
, skb
, portid
, group
, echo
, flags
);
137 EXPORT_SYMBOL_GPL(nfnetlink_send
);
139 int nfnetlink_set_err(struct net
*net
, u32 portid
, u32 group
, int error
)
141 return netlink_set_err(net
->nfnl
, portid
, group
, error
);
143 EXPORT_SYMBOL_GPL(nfnetlink_set_err
);
145 int nfnetlink_unicast(struct sk_buff
*skb
, struct net
*net
, u32 portid
,
148 return netlink_unicast(net
->nfnl
, skb
, portid
, flags
);
150 EXPORT_SYMBOL_GPL(nfnetlink_unicast
);
152 /* Process one complete nfnetlink message. */
153 static int nfnetlink_rcv_msg(struct sk_buff
*skb
, struct nlmsghdr
*nlh
)
155 struct net
*net
= sock_net(skb
->sk
);
156 const struct nfnl_callback
*nc
;
157 const struct nfnetlink_subsystem
*ss
;
160 /* All the messages must at least contain nfgenmsg */
161 if (nlmsg_len(nlh
) < sizeof(struct nfgenmsg
))
164 type
= nlh
->nlmsg_type
;
167 ss
= nfnetlink_get_subsys(type
);
169 #ifdef CONFIG_MODULES
171 request_module("nfnetlink-subsys-%d", NFNL_SUBSYS_ID(type
));
173 ss
= nfnetlink_get_subsys(type
);
182 nc
= nfnetlink_find_client(type
, ss
);
189 int min_len
= nlmsg_total_size(sizeof(struct nfgenmsg
));
190 u_int8_t cb_id
= NFNL_MSG_TYPE(nlh
->nlmsg_type
);
191 struct nlattr
*cda
[ss
->cb
[cb_id
].attr_count
+ 1];
192 struct nlattr
*attr
= (void *)nlh
+ min_len
;
193 int attrlen
= nlh
->nlmsg_len
- min_len
;
194 __u8 subsys_id
= NFNL_SUBSYS_ID(type
);
196 err
= nla_parse(cda
, ss
->cb
[cb_id
].attr_count
,
197 attr
, attrlen
, ss
->cb
[cb_id
].policy
);
204 err
= nc
->call_rcu(net
->nfnl
, skb
, nlh
,
205 (const struct nlattr
**)cda
);
209 nfnl_lock(subsys_id
);
210 if (rcu_dereference_protected(table
[subsys_id
].subsys
,
211 lockdep_is_held(&table
[subsys_id
].mutex
)) != ss
||
212 nfnetlink_find_client(type
, ss
) != nc
)
215 err
= nc
->call(net
->nfnl
, skb
, nlh
,
216 (const struct nlattr
**)cda
);
219 nfnl_unlock(subsys_id
);
228 struct list_head head
;
229 struct nlmsghdr
*nlh
;
233 static int nfnl_err_add(struct list_head
*list
, struct nlmsghdr
*nlh
, int err
)
235 struct nfnl_err
*nfnl_err
;
237 nfnl_err
= kmalloc(sizeof(struct nfnl_err
), GFP_KERNEL
);
238 if (nfnl_err
== NULL
)
243 list_add_tail(&nfnl_err
->head
, list
);
248 static void nfnl_err_del(struct nfnl_err
*nfnl_err
)
250 list_del(&nfnl_err
->head
);
254 static void nfnl_err_reset(struct list_head
*err_list
)
256 struct nfnl_err
*nfnl_err
, *next
;
258 list_for_each_entry_safe(nfnl_err
, next
, err_list
, head
)
259 nfnl_err_del(nfnl_err
);
262 static void nfnl_err_deliver(struct list_head
*err_list
, struct sk_buff
*skb
)
264 struct nfnl_err
*nfnl_err
, *next
;
266 list_for_each_entry_safe(nfnl_err
, next
, err_list
, head
) {
267 netlink_ack(skb
, nfnl_err
->nlh
, nfnl_err
->err
);
268 nfnl_err_del(nfnl_err
);
273 NFNL_BATCH_FAILURE
= (1 << 0),
274 NFNL_BATCH_DONE
= (1 << 1),
275 NFNL_BATCH_REPLAY
= (1 << 2),
278 static void nfnetlink_rcv_batch(struct sk_buff
*skb
, struct nlmsghdr
*nlh
,
281 struct sk_buff
*oskb
= skb
;
282 struct net
*net
= sock_net(skb
->sk
);
283 const struct nfnetlink_subsystem
*ss
;
284 const struct nfnl_callback
*nc
;
285 static LIST_HEAD(err_list
);
289 if (subsys_id
>= NFNL_SUBSYS_COUNT
)
290 return netlink_ack(skb
, nlh
, -EINVAL
);
294 skb
= netlink_skb_clone(oskb
, GFP_KERNEL
);
296 return netlink_ack(oskb
, nlh
, -ENOMEM
);
298 nfnl_lock(subsys_id
);
299 ss
= rcu_dereference_protected(table
[subsys_id
].subsys
,
300 lockdep_is_held(&table
[subsys_id
].mutex
));
302 #ifdef CONFIG_MODULES
303 nfnl_unlock(subsys_id
);
304 request_module("nfnetlink-subsys-%d", subsys_id
);
305 nfnl_lock(subsys_id
);
306 ss
= rcu_dereference_protected(table
[subsys_id
].subsys
,
307 lockdep_is_held(&table
[subsys_id
].mutex
));
311 nfnl_unlock(subsys_id
);
312 netlink_ack(skb
, nlh
, -EOPNOTSUPP
);
313 return kfree_skb(skb
);
317 if (!ss
->commit
|| !ss
->abort
) {
318 nfnl_unlock(subsys_id
);
319 netlink_ack(skb
, nlh
, -EOPNOTSUPP
);
320 return kfree_skb(skb
);
323 while (skb
->len
>= nlmsg_total_size(0)) {
326 nlh
= nlmsg_hdr(skb
);
329 if (nlmsg_len(nlh
) < sizeof(struct nfgenmsg
) ||
330 skb
->len
< nlh
->nlmsg_len
) {
335 /* Only requests are handled by the kernel */
336 if (!(nlh
->nlmsg_flags
& NLM_F_REQUEST
)) {
341 type
= nlh
->nlmsg_type
;
342 if (type
== NFNL_MSG_BATCH_BEGIN
) {
343 /* Malformed: Batch begin twice */
344 nfnl_err_reset(&err_list
);
345 status
|= NFNL_BATCH_FAILURE
;
347 } else if (type
== NFNL_MSG_BATCH_END
) {
348 status
|= NFNL_BATCH_DONE
;
350 } else if (type
< NLMSG_MIN_TYPE
) {
355 /* We only accept a batch with messages for the same
358 if (NFNL_SUBSYS_ID(type
) != subsys_id
) {
363 nc
= nfnetlink_find_client(type
, ss
);
370 int min_len
= nlmsg_total_size(sizeof(struct nfgenmsg
));
371 u_int8_t cb_id
= NFNL_MSG_TYPE(nlh
->nlmsg_type
);
372 struct nlattr
*cda
[ss
->cb
[cb_id
].attr_count
+ 1];
373 struct nlattr
*attr
= (void *)nlh
+ min_len
;
374 int attrlen
= nlh
->nlmsg_len
- min_len
;
376 err
= nla_parse(cda
, ss
->cb
[cb_id
].attr_count
,
377 attr
, attrlen
, ss
->cb
[cb_id
].policy
);
381 if (nc
->call_batch
) {
382 err
= nc
->call_batch(net
, net
->nfnl
, skb
, nlh
,
383 (const struct nlattr
**)cda
);
386 /* The lock was released to autoload some module, we
387 * have to abort and start from scratch using the
390 if (err
== -EAGAIN
) {
391 status
|= NFNL_BATCH_REPLAY
;
396 if (nlh
->nlmsg_flags
& NLM_F_ACK
|| err
) {
397 /* Errors are delivered once the full batch has been
398 * processed, this avoids that the same error is
399 * reported several times when replaying the batch.
401 if (nfnl_err_add(&err_list
, nlh
, err
) < 0) {
402 /* We failed to enqueue an error, reset the
403 * list of errors and send OOM to userspace
404 * pointing to the batch header.
406 nfnl_err_reset(&err_list
);
407 netlink_ack(skb
, nlmsg_hdr(oskb
), -ENOMEM
);
408 status
|= NFNL_BATCH_FAILURE
;
411 /* We don't stop processing the batch on errors, thus,
412 * userspace gets all the errors that the batch
416 status
|= NFNL_BATCH_FAILURE
;
419 msglen
= NLMSG_ALIGN(nlh
->nlmsg_len
);
420 if (msglen
> skb
->len
)
422 skb_pull(skb
, msglen
);
425 if (status
& NFNL_BATCH_REPLAY
) {
427 nfnl_err_reset(&err_list
);
428 nfnl_unlock(subsys_id
);
431 } else if (status
== NFNL_BATCH_DONE
) {
437 nfnl_err_deliver(&err_list
, oskb
);
438 nfnl_unlock(subsys_id
);
442 static void nfnetlink_rcv(struct sk_buff
*skb
)
444 struct nlmsghdr
*nlh
= nlmsg_hdr(skb
);
448 if (nlh
->nlmsg_len
< NLMSG_HDRLEN
||
449 skb
->len
< nlh
->nlmsg_len
)
452 if (!netlink_net_capable(skb
, CAP_NET_ADMIN
)) {
453 netlink_ack(skb
, nlh
, -EPERM
);
457 if (nlh
->nlmsg_type
== NFNL_MSG_BATCH_BEGIN
) {
458 struct nfgenmsg
*nfgenmsg
;
460 msglen
= NLMSG_ALIGN(nlh
->nlmsg_len
);
461 if (msglen
> skb
->len
)
464 if (nlh
->nlmsg_len
< NLMSG_HDRLEN
||
465 skb
->len
< NLMSG_HDRLEN
+ sizeof(struct nfgenmsg
))
468 nfgenmsg
= nlmsg_data(nlh
);
469 skb_pull(skb
, msglen
);
470 /* Work around old nft using host byte order */
471 if (nfgenmsg
->res_id
== NFNL_SUBSYS_NFTABLES
)
472 res_id
= NFNL_SUBSYS_NFTABLES
;
474 res_id
= ntohs(nfgenmsg
->res_id
);
475 nfnetlink_rcv_batch(skb
, nlh
, res_id
);
477 netlink_rcv_skb(skb
, &nfnetlink_rcv_msg
);
481 #ifdef CONFIG_MODULES
482 static int nfnetlink_bind(struct net
*net
, int group
)
484 const struct nfnetlink_subsystem
*ss
;
487 if (group
<= NFNLGRP_NONE
|| group
> NFNLGRP_MAX
)
490 type
= nfnl_group2type
[group
];
493 ss
= nfnetlink_get_subsys(type
<< 8);
496 request_module("nfnetlink-subsys-%d", type
);
501 static int __net_init
nfnetlink_net_init(struct net
*net
)
504 struct netlink_kernel_cfg cfg
= {
505 .groups
= NFNLGRP_MAX
,
506 .input
= nfnetlink_rcv
,
507 #ifdef CONFIG_MODULES
508 .bind
= nfnetlink_bind
,
512 nfnl
= netlink_kernel_create(net
, NETLINK_NETFILTER
, &cfg
);
515 net
->nfnl_stash
= nfnl
;
516 rcu_assign_pointer(net
->nfnl
, nfnl
);
520 static void __net_exit
nfnetlink_net_exit_batch(struct list_head
*net_exit_list
)
524 list_for_each_entry(net
, net_exit_list
, exit_list
)
525 RCU_INIT_POINTER(net
->nfnl
, NULL
);
527 list_for_each_entry(net
, net_exit_list
, exit_list
)
528 netlink_kernel_release(net
->nfnl_stash
);
531 static struct pernet_operations nfnetlink_net_ops
= {
532 .init
= nfnetlink_net_init
,
533 .exit_batch
= nfnetlink_net_exit_batch
,
536 static int __init
nfnetlink_init(void)
540 for (i
= NFNLGRP_NONE
+ 1; i
<= NFNLGRP_MAX
; i
++)
541 BUG_ON(nfnl_group2type
[i
] == NFNL_SUBSYS_NONE
);
543 for (i
=0; i
<NFNL_SUBSYS_COUNT
; i
++)
544 mutex_init(&table
[i
].mutex
);
546 pr_info("Netfilter messages via NETLINK v%s.\n", nfversion
);
547 return register_pernet_subsys(&nfnetlink_net_ops
);
550 static void __exit
nfnetlink_exit(void)
552 pr_info("Removing netfilter NETLINK layer.\n");
553 unregister_pernet_subsys(&nfnetlink_net_ops
);
555 module_init(nfnetlink_init
);
556 module_exit(nfnetlink_exit
);