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) 2005-2017 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 <linux/uaccess.h>
27 #include <linux/init.h>
28 #include <linux/sched/signal.h>
30 #include <net/netlink.h>
31 #include <linux/netfilter/nfnetlink.h>
33 MODULE_LICENSE("GPL");
34 MODULE_AUTHOR("Harald Welte <laforge@netfilter.org>");
35 MODULE_ALIAS_NET_PF_PROTO(PF_NETLINK
, NETLINK_NETFILTER
);
37 #define nfnl_dereference_protected(id) \
38 rcu_dereference_protected(table[(id)].subsys, \
39 lockdep_nfnl_is_held((id)))
41 #define NFNL_MAX_ATTR_COUNT 32
45 const struct nfnetlink_subsystem __rcu
*subsys
;
46 } table
[NFNL_SUBSYS_COUNT
];
48 static const int nfnl_group2type
[NFNLGRP_MAX
+1] = {
49 [NFNLGRP_CONNTRACK_NEW
] = NFNL_SUBSYS_CTNETLINK
,
50 [NFNLGRP_CONNTRACK_UPDATE
] = NFNL_SUBSYS_CTNETLINK
,
51 [NFNLGRP_CONNTRACK_DESTROY
] = NFNL_SUBSYS_CTNETLINK
,
52 [NFNLGRP_CONNTRACK_EXP_NEW
] = NFNL_SUBSYS_CTNETLINK_EXP
,
53 [NFNLGRP_CONNTRACK_EXP_UPDATE
] = NFNL_SUBSYS_CTNETLINK_EXP
,
54 [NFNLGRP_CONNTRACK_EXP_DESTROY
] = NFNL_SUBSYS_CTNETLINK_EXP
,
55 [NFNLGRP_NFTABLES
] = NFNL_SUBSYS_NFTABLES
,
56 [NFNLGRP_ACCT_QUOTA
] = NFNL_SUBSYS_ACCT
,
57 [NFNLGRP_NFTRACE
] = NFNL_SUBSYS_NFTABLES
,
60 void nfnl_lock(__u8 subsys_id
)
62 mutex_lock(&table
[subsys_id
].mutex
);
64 EXPORT_SYMBOL_GPL(nfnl_lock
);
66 void nfnl_unlock(__u8 subsys_id
)
68 mutex_unlock(&table
[subsys_id
].mutex
);
70 EXPORT_SYMBOL_GPL(nfnl_unlock
);
72 #ifdef CONFIG_PROVE_LOCKING
73 bool lockdep_nfnl_is_held(u8 subsys_id
)
75 return lockdep_is_held(&table
[subsys_id
].mutex
);
77 EXPORT_SYMBOL_GPL(lockdep_nfnl_is_held
);
80 int nfnetlink_subsys_register(const struct nfnetlink_subsystem
*n
)
84 /* Sanity-check attr_count size to avoid stack buffer overflow. */
85 for (cb_id
= 0; cb_id
< n
->cb_count
; cb_id
++)
86 if (WARN_ON(n
->cb
[cb_id
].attr_count
> NFNL_MAX_ATTR_COUNT
))
89 nfnl_lock(n
->subsys_id
);
90 if (table
[n
->subsys_id
].subsys
) {
91 nfnl_unlock(n
->subsys_id
);
94 rcu_assign_pointer(table
[n
->subsys_id
].subsys
, n
);
95 nfnl_unlock(n
->subsys_id
);
99 EXPORT_SYMBOL_GPL(nfnetlink_subsys_register
);
101 int nfnetlink_subsys_unregister(const struct nfnetlink_subsystem
*n
)
103 nfnl_lock(n
->subsys_id
);
104 table
[n
->subsys_id
].subsys
= NULL
;
105 nfnl_unlock(n
->subsys_id
);
109 EXPORT_SYMBOL_GPL(nfnetlink_subsys_unregister
);
111 static inline const struct nfnetlink_subsystem
*nfnetlink_get_subsys(u16 type
)
113 u8 subsys_id
= NFNL_SUBSYS_ID(type
);
115 if (subsys_id
>= NFNL_SUBSYS_COUNT
)
118 return rcu_dereference(table
[subsys_id
].subsys
);
121 static inline const struct nfnl_callback
*
122 nfnetlink_find_client(u16 type
, const struct nfnetlink_subsystem
*ss
)
124 u8 cb_id
= NFNL_MSG_TYPE(type
);
126 if (cb_id
>= ss
->cb_count
)
129 return &ss
->cb
[cb_id
];
132 int nfnetlink_has_listeners(struct net
*net
, unsigned int group
)
134 return netlink_has_listeners(net
->nfnl
, group
);
136 EXPORT_SYMBOL_GPL(nfnetlink_has_listeners
);
138 int nfnetlink_send(struct sk_buff
*skb
, struct net
*net
, u32 portid
,
139 unsigned int group
, int echo
, gfp_t flags
)
141 return nlmsg_notify(net
->nfnl
, skb
, portid
, group
, echo
, flags
);
143 EXPORT_SYMBOL_GPL(nfnetlink_send
);
145 int nfnetlink_set_err(struct net
*net
, u32 portid
, u32 group
, int error
)
147 return netlink_set_err(net
->nfnl
, portid
, group
, error
);
149 EXPORT_SYMBOL_GPL(nfnetlink_set_err
);
151 int nfnetlink_unicast(struct sk_buff
*skb
, struct net
*net
, u32 portid
,
154 return netlink_unicast(net
->nfnl
, skb
, portid
, flags
);
156 EXPORT_SYMBOL_GPL(nfnetlink_unicast
);
158 /* Process one complete nfnetlink message. */
159 static int nfnetlink_rcv_msg(struct sk_buff
*skb
, struct nlmsghdr
*nlh
,
160 struct netlink_ext_ack
*extack
)
162 struct net
*net
= sock_net(skb
->sk
);
163 const struct nfnl_callback
*nc
;
164 const struct nfnetlink_subsystem
*ss
;
167 /* All the messages must at least contain nfgenmsg */
168 if (nlmsg_len(nlh
) < sizeof(struct nfgenmsg
))
171 type
= nlh
->nlmsg_type
;
174 ss
= nfnetlink_get_subsys(type
);
176 #ifdef CONFIG_MODULES
178 request_module("nfnetlink-subsys-%d", NFNL_SUBSYS_ID(type
));
180 ss
= nfnetlink_get_subsys(type
);
189 nc
= nfnetlink_find_client(type
, ss
);
196 int min_len
= nlmsg_total_size(sizeof(struct nfgenmsg
));
197 u8 cb_id
= NFNL_MSG_TYPE(nlh
->nlmsg_type
);
198 struct nlattr
*cda
[NFNL_MAX_ATTR_COUNT
+ 1];
199 struct nlattr
*attr
= (void *)nlh
+ min_len
;
200 int attrlen
= nlh
->nlmsg_len
- min_len
;
201 __u8 subsys_id
= NFNL_SUBSYS_ID(type
);
203 /* Sanity-check NFNL_MAX_ATTR_COUNT */
204 if (ss
->cb
[cb_id
].attr_count
> NFNL_MAX_ATTR_COUNT
) {
209 err
= nla_parse_deprecated(cda
, ss
->cb
[cb_id
].attr_count
,
211 ss
->cb
[cb_id
].policy
, extack
);
218 err
= nc
->call_rcu(net
, net
->nfnl
, skb
, nlh
,
219 (const struct nlattr
**)cda
,
224 nfnl_lock(subsys_id
);
225 if (nfnl_dereference_protected(subsys_id
) != ss
||
226 nfnetlink_find_client(type
, ss
) != nc
)
229 err
= nc
->call(net
, net
->nfnl
, skb
, nlh
,
230 (const struct nlattr
**)cda
,
234 nfnl_unlock(subsys_id
);
243 struct list_head head
;
244 struct nlmsghdr
*nlh
;
246 struct netlink_ext_ack extack
;
249 static int nfnl_err_add(struct list_head
*list
, struct nlmsghdr
*nlh
, int err
,
250 const struct netlink_ext_ack
*extack
)
252 struct nfnl_err
*nfnl_err
;
254 nfnl_err
= kmalloc(sizeof(struct nfnl_err
), GFP_KERNEL
);
255 if (nfnl_err
== NULL
)
260 nfnl_err
->extack
= *extack
;
261 list_add_tail(&nfnl_err
->head
, list
);
266 static void nfnl_err_del(struct nfnl_err
*nfnl_err
)
268 list_del(&nfnl_err
->head
);
272 static void nfnl_err_reset(struct list_head
*err_list
)
274 struct nfnl_err
*nfnl_err
, *next
;
276 list_for_each_entry_safe(nfnl_err
, next
, err_list
, head
)
277 nfnl_err_del(nfnl_err
);
280 static void nfnl_err_deliver(struct list_head
*err_list
, struct sk_buff
*skb
)
282 struct nfnl_err
*nfnl_err
, *next
;
284 list_for_each_entry_safe(nfnl_err
, next
, err_list
, head
) {
285 netlink_ack(skb
, nfnl_err
->nlh
, nfnl_err
->err
,
287 nfnl_err_del(nfnl_err
);
292 NFNL_BATCH_FAILURE
= (1 << 0),
293 NFNL_BATCH_DONE
= (1 << 1),
294 NFNL_BATCH_REPLAY
= (1 << 2),
297 static void nfnetlink_rcv_batch(struct sk_buff
*skb
, struct nlmsghdr
*nlh
,
298 u16 subsys_id
, u32 genid
)
300 struct sk_buff
*oskb
= skb
;
301 struct net
*net
= sock_net(skb
->sk
);
302 const struct nfnetlink_subsystem
*ss
;
303 const struct nfnl_callback
*nc
;
304 struct netlink_ext_ack extack
;
309 if (subsys_id
>= NFNL_SUBSYS_COUNT
)
310 return netlink_ack(skb
, nlh
, -EINVAL
, NULL
);
314 skb
= netlink_skb_clone(oskb
, GFP_KERNEL
);
316 return netlink_ack(oskb
, nlh
, -ENOMEM
, NULL
);
318 nfnl_lock(subsys_id
);
319 ss
= nfnl_dereference_protected(subsys_id
);
321 #ifdef CONFIG_MODULES
322 nfnl_unlock(subsys_id
);
323 request_module("nfnetlink-subsys-%d", subsys_id
);
324 nfnl_lock(subsys_id
);
325 ss
= nfnl_dereference_protected(subsys_id
);
329 nfnl_unlock(subsys_id
);
330 netlink_ack(oskb
, nlh
, -EOPNOTSUPP
, NULL
);
331 return kfree_skb(skb
);
335 if (!ss
->valid_genid
|| !ss
->commit
|| !ss
->abort
) {
336 nfnl_unlock(subsys_id
);
337 netlink_ack(oskb
, nlh
, -EOPNOTSUPP
, NULL
);
338 return kfree_skb(skb
);
341 if (!try_module_get(ss
->owner
)) {
342 nfnl_unlock(subsys_id
);
343 netlink_ack(oskb
, nlh
, -EOPNOTSUPP
, NULL
);
344 return kfree_skb(skb
);
347 if (!ss
->valid_genid(net
, genid
)) {
348 module_put(ss
->owner
);
349 nfnl_unlock(subsys_id
);
350 netlink_ack(oskb
, nlh
, -ERESTART
, NULL
);
351 return kfree_skb(skb
);
354 nfnl_unlock(subsys_id
);
356 while (skb
->len
>= nlmsg_total_size(0)) {
359 if (fatal_signal_pending(current
)) {
360 nfnl_err_reset(&err_list
);
362 status
= NFNL_BATCH_FAILURE
;
366 memset(&extack
, 0, sizeof(extack
));
367 nlh
= nlmsg_hdr(skb
);
370 if (nlh
->nlmsg_len
< NLMSG_HDRLEN
||
371 skb
->len
< nlh
->nlmsg_len
||
372 nlmsg_len(nlh
) < sizeof(struct nfgenmsg
)) {
373 nfnl_err_reset(&err_list
);
374 status
|= NFNL_BATCH_FAILURE
;
378 /* Only requests are handled by the kernel */
379 if (!(nlh
->nlmsg_flags
& NLM_F_REQUEST
)) {
384 type
= nlh
->nlmsg_type
;
385 if (type
== NFNL_MSG_BATCH_BEGIN
) {
386 /* Malformed: Batch begin twice */
387 nfnl_err_reset(&err_list
);
388 status
|= NFNL_BATCH_FAILURE
;
390 } else if (type
== NFNL_MSG_BATCH_END
) {
391 status
|= NFNL_BATCH_DONE
;
393 } else if (type
< NLMSG_MIN_TYPE
) {
398 /* We only accept a batch with messages for the same
401 if (NFNL_SUBSYS_ID(type
) != subsys_id
) {
406 nc
= nfnetlink_find_client(type
, ss
);
413 int min_len
= nlmsg_total_size(sizeof(struct nfgenmsg
));
414 u8 cb_id
= NFNL_MSG_TYPE(nlh
->nlmsg_type
);
415 struct nlattr
*cda
[NFNL_MAX_ATTR_COUNT
+ 1];
416 struct nlattr
*attr
= (void *)nlh
+ min_len
;
417 int attrlen
= nlh
->nlmsg_len
- min_len
;
419 /* Sanity-check NFTA_MAX_ATTR */
420 if (ss
->cb
[cb_id
].attr_count
> NFNL_MAX_ATTR_COUNT
) {
425 err
= nla_parse_deprecated(cda
,
426 ss
->cb
[cb_id
].attr_count
,
428 ss
->cb
[cb_id
].policy
, NULL
);
432 if (nc
->call_batch
) {
433 err
= nc
->call_batch(net
, net
->nfnl
, skb
, nlh
,
434 (const struct nlattr
**)cda
,
438 /* The lock was released to autoload some module, we
439 * have to abort and start from scratch using the
442 if (err
== -EAGAIN
) {
443 status
|= NFNL_BATCH_REPLAY
;
448 if (nlh
->nlmsg_flags
& NLM_F_ACK
|| err
) {
449 /* Errors are delivered once the full batch has been
450 * processed, this avoids that the same error is
451 * reported several times when replaying the batch.
453 if (nfnl_err_add(&err_list
, nlh
, err
, &extack
) < 0) {
454 /* We failed to enqueue an error, reset the
455 * list of errors and send OOM to userspace
456 * pointing to the batch header.
458 nfnl_err_reset(&err_list
);
459 netlink_ack(oskb
, nlmsg_hdr(oskb
), -ENOMEM
,
461 status
|= NFNL_BATCH_FAILURE
;
464 /* We don't stop processing the batch on errors, thus,
465 * userspace gets all the errors that the batch
469 status
|= NFNL_BATCH_FAILURE
;
472 msglen
= NLMSG_ALIGN(nlh
->nlmsg_len
);
473 if (msglen
> skb
->len
)
475 skb_pull(skb
, msglen
);
478 if (status
& NFNL_BATCH_REPLAY
) {
479 ss
->abort(net
, oskb
, true);
480 nfnl_err_reset(&err_list
);
482 module_put(ss
->owner
);
484 } else if (status
== NFNL_BATCH_DONE
) {
485 err
= ss
->commit(net
, oskb
);
486 if (err
== -EAGAIN
) {
487 status
|= NFNL_BATCH_REPLAY
;
490 ss
->abort(net
, oskb
, false);
491 netlink_ack(oskb
, nlmsg_hdr(oskb
), err
, NULL
);
494 ss
->abort(net
, oskb
, false);
499 nfnl_err_deliver(&err_list
, oskb
);
501 module_put(ss
->owner
);
504 static const struct nla_policy nfnl_batch_policy
[NFNL_BATCH_MAX
+ 1] = {
505 [NFNL_BATCH_GENID
] = { .type
= NLA_U32
},
508 static void nfnetlink_rcv_skb_batch(struct sk_buff
*skb
, struct nlmsghdr
*nlh
)
510 int min_len
= nlmsg_total_size(sizeof(struct nfgenmsg
));
511 struct nlattr
*attr
= (void *)nlh
+ min_len
;
512 struct nlattr
*cda
[NFNL_BATCH_MAX
+ 1];
513 int attrlen
= nlh
->nlmsg_len
- min_len
;
514 struct nfgenmsg
*nfgenmsg
;
519 msglen
= NLMSG_ALIGN(nlh
->nlmsg_len
);
520 if (msglen
> skb
->len
)
523 if (skb
->len
< NLMSG_HDRLEN
+ sizeof(struct nfgenmsg
))
526 err
= nla_parse_deprecated(cda
, NFNL_BATCH_MAX
, attr
, attrlen
,
527 nfnl_batch_policy
, NULL
);
529 netlink_ack(skb
, nlh
, err
, NULL
);
532 if (cda
[NFNL_BATCH_GENID
])
533 gen_id
= ntohl(nla_get_be32(cda
[NFNL_BATCH_GENID
]));
535 nfgenmsg
= nlmsg_data(nlh
);
536 skb_pull(skb
, msglen
);
537 /* Work around old nft using host byte order */
538 if (nfgenmsg
->res_id
== NFNL_SUBSYS_NFTABLES
)
539 res_id
= NFNL_SUBSYS_NFTABLES
;
541 res_id
= ntohs(nfgenmsg
->res_id
);
543 nfnetlink_rcv_batch(skb
, nlh
, res_id
, gen_id
);
546 static void nfnetlink_rcv(struct sk_buff
*skb
)
548 struct nlmsghdr
*nlh
= nlmsg_hdr(skb
);
550 if (skb
->len
< NLMSG_HDRLEN
||
551 nlh
->nlmsg_len
< NLMSG_HDRLEN
||
552 skb
->len
< nlh
->nlmsg_len
)
555 if (!netlink_net_capable(skb
, CAP_NET_ADMIN
)) {
556 netlink_ack(skb
, nlh
, -EPERM
, NULL
);
560 if (nlh
->nlmsg_type
== NFNL_MSG_BATCH_BEGIN
)
561 nfnetlink_rcv_skb_batch(skb
, nlh
);
563 netlink_rcv_skb(skb
, nfnetlink_rcv_msg
);
566 #ifdef CONFIG_MODULES
567 static int nfnetlink_bind(struct net
*net
, int group
)
569 const struct nfnetlink_subsystem
*ss
;
572 if (group
<= NFNLGRP_NONE
|| group
> NFNLGRP_MAX
)
575 type
= nfnl_group2type
[group
];
578 ss
= nfnetlink_get_subsys(type
<< 8);
581 request_module_nowait("nfnetlink-subsys-%d", type
);
586 static int __net_init
nfnetlink_net_init(struct net
*net
)
589 struct netlink_kernel_cfg cfg
= {
590 .groups
= NFNLGRP_MAX
,
591 .input
= nfnetlink_rcv
,
592 #ifdef CONFIG_MODULES
593 .bind
= nfnetlink_bind
,
597 nfnl
= netlink_kernel_create(net
, NETLINK_NETFILTER
, &cfg
);
600 net
->nfnl_stash
= nfnl
;
601 rcu_assign_pointer(net
->nfnl
, nfnl
);
605 static void __net_exit
nfnetlink_net_exit_batch(struct list_head
*net_exit_list
)
609 list_for_each_entry(net
, net_exit_list
, exit_list
)
610 RCU_INIT_POINTER(net
->nfnl
, NULL
);
612 list_for_each_entry(net
, net_exit_list
, exit_list
)
613 netlink_kernel_release(net
->nfnl_stash
);
616 static struct pernet_operations nfnetlink_net_ops
= {
617 .init
= nfnetlink_net_init
,
618 .exit_batch
= nfnetlink_net_exit_batch
,
621 static int __init
nfnetlink_init(void)
625 for (i
= NFNLGRP_NONE
+ 1; i
<= NFNLGRP_MAX
; i
++)
626 BUG_ON(nfnl_group2type
[i
] == NFNL_SUBSYS_NONE
);
628 for (i
=0; i
<NFNL_SUBSYS_COUNT
; i
++)
629 mutex_init(&table
[i
].mutex
);
631 return register_pernet_subsys(&nfnetlink_net_ops
);
634 static void __exit
nfnetlink_exit(void)
636 unregister_pernet_subsys(&nfnetlink_net_ops
);
638 module_init(nfnetlink_init
);
639 module_exit(nfnetlink_exit
);