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
);
36 MODULE_DESCRIPTION("Netfilter messages via netlink socket");
38 #define nfnl_dereference_protected(id) \
39 rcu_dereference_protected(table[(id)].subsys, \
40 lockdep_nfnl_is_held((id)))
42 #define NFNL_MAX_ATTR_COUNT 32
46 const struct nfnetlink_subsystem __rcu
*subsys
;
47 } table
[NFNL_SUBSYS_COUNT
];
49 static struct lock_class_key nfnl_lockdep_keys
[NFNL_SUBSYS_COUNT
];
51 static const char *const nfnl_lockdep_names
[NFNL_SUBSYS_COUNT
] = {
52 [NFNL_SUBSYS_NONE
] = "nfnl_subsys_none",
53 [NFNL_SUBSYS_CTNETLINK
] = "nfnl_subsys_ctnetlink",
54 [NFNL_SUBSYS_CTNETLINK_EXP
] = "nfnl_subsys_ctnetlink_exp",
55 [NFNL_SUBSYS_QUEUE
] = "nfnl_subsys_queue",
56 [NFNL_SUBSYS_ULOG
] = "nfnl_subsys_ulog",
57 [NFNL_SUBSYS_OSF
] = "nfnl_subsys_osf",
58 [NFNL_SUBSYS_IPSET
] = "nfnl_subsys_ipset",
59 [NFNL_SUBSYS_ACCT
] = "nfnl_subsys_acct",
60 [NFNL_SUBSYS_CTNETLINK_TIMEOUT
] = "nfnl_subsys_cttimeout",
61 [NFNL_SUBSYS_CTHELPER
] = "nfnl_subsys_cthelper",
62 [NFNL_SUBSYS_NFTABLES
] = "nfnl_subsys_nftables",
63 [NFNL_SUBSYS_NFT_COMPAT
] = "nfnl_subsys_nftcompat",
66 static const int nfnl_group2type
[NFNLGRP_MAX
+1] = {
67 [NFNLGRP_CONNTRACK_NEW
] = NFNL_SUBSYS_CTNETLINK
,
68 [NFNLGRP_CONNTRACK_UPDATE
] = NFNL_SUBSYS_CTNETLINK
,
69 [NFNLGRP_CONNTRACK_DESTROY
] = NFNL_SUBSYS_CTNETLINK
,
70 [NFNLGRP_CONNTRACK_EXP_NEW
] = NFNL_SUBSYS_CTNETLINK_EXP
,
71 [NFNLGRP_CONNTRACK_EXP_UPDATE
] = NFNL_SUBSYS_CTNETLINK_EXP
,
72 [NFNLGRP_CONNTRACK_EXP_DESTROY
] = NFNL_SUBSYS_CTNETLINK_EXP
,
73 [NFNLGRP_NFTABLES
] = NFNL_SUBSYS_NFTABLES
,
74 [NFNLGRP_ACCT_QUOTA
] = NFNL_SUBSYS_ACCT
,
75 [NFNLGRP_NFTRACE
] = NFNL_SUBSYS_NFTABLES
,
78 void nfnl_lock(__u8 subsys_id
)
80 mutex_lock(&table
[subsys_id
].mutex
);
82 EXPORT_SYMBOL_GPL(nfnl_lock
);
84 void nfnl_unlock(__u8 subsys_id
)
86 mutex_unlock(&table
[subsys_id
].mutex
);
88 EXPORT_SYMBOL_GPL(nfnl_unlock
);
90 #ifdef CONFIG_PROVE_LOCKING
91 bool lockdep_nfnl_is_held(u8 subsys_id
)
93 return lockdep_is_held(&table
[subsys_id
].mutex
);
95 EXPORT_SYMBOL_GPL(lockdep_nfnl_is_held
);
98 int nfnetlink_subsys_register(const struct nfnetlink_subsystem
*n
)
102 /* Sanity-check attr_count size to avoid stack buffer overflow. */
103 for (cb_id
= 0; cb_id
< n
->cb_count
; cb_id
++)
104 if (WARN_ON(n
->cb
[cb_id
].attr_count
> NFNL_MAX_ATTR_COUNT
))
107 nfnl_lock(n
->subsys_id
);
108 if (table
[n
->subsys_id
].subsys
) {
109 nfnl_unlock(n
->subsys_id
);
112 rcu_assign_pointer(table
[n
->subsys_id
].subsys
, n
);
113 nfnl_unlock(n
->subsys_id
);
117 EXPORT_SYMBOL_GPL(nfnetlink_subsys_register
);
119 int nfnetlink_subsys_unregister(const struct nfnetlink_subsystem
*n
)
121 nfnl_lock(n
->subsys_id
);
122 table
[n
->subsys_id
].subsys
= NULL
;
123 nfnl_unlock(n
->subsys_id
);
127 EXPORT_SYMBOL_GPL(nfnetlink_subsys_unregister
);
129 static inline const struct nfnetlink_subsystem
*nfnetlink_get_subsys(u16 type
)
131 u8 subsys_id
= NFNL_SUBSYS_ID(type
);
133 if (subsys_id
>= NFNL_SUBSYS_COUNT
)
136 return rcu_dereference(table
[subsys_id
].subsys
);
139 static inline const struct nfnl_callback
*
140 nfnetlink_find_client(u16 type
, const struct nfnetlink_subsystem
*ss
)
142 u8 cb_id
= NFNL_MSG_TYPE(type
);
144 if (cb_id
>= ss
->cb_count
)
147 return &ss
->cb
[cb_id
];
150 int nfnetlink_has_listeners(struct net
*net
, unsigned int group
)
152 return netlink_has_listeners(net
->nfnl
, group
);
154 EXPORT_SYMBOL_GPL(nfnetlink_has_listeners
);
156 int nfnetlink_send(struct sk_buff
*skb
, struct net
*net
, u32 portid
,
157 unsigned int group
, int echo
, gfp_t flags
)
159 return nlmsg_notify(net
->nfnl
, skb
, portid
, group
, echo
, flags
);
161 EXPORT_SYMBOL_GPL(nfnetlink_send
);
163 int nfnetlink_set_err(struct net
*net
, u32 portid
, u32 group
, int error
)
165 return netlink_set_err(net
->nfnl
, portid
, group
, error
);
167 EXPORT_SYMBOL_GPL(nfnetlink_set_err
);
169 int nfnetlink_unicast(struct sk_buff
*skb
, struct net
*net
, u32 portid
)
173 err
= nlmsg_unicast(net
->nfnl
, skb
, portid
);
179 EXPORT_SYMBOL_GPL(nfnetlink_unicast
);
181 /* Process one complete nfnetlink message. */
182 static int nfnetlink_rcv_msg(struct sk_buff
*skb
, struct nlmsghdr
*nlh
,
183 struct netlink_ext_ack
*extack
)
185 struct net
*net
= sock_net(skb
->sk
);
186 const struct nfnl_callback
*nc
;
187 const struct nfnetlink_subsystem
*ss
;
190 /* All the messages must at least contain nfgenmsg */
191 if (nlmsg_len(nlh
) < sizeof(struct nfgenmsg
))
194 type
= nlh
->nlmsg_type
;
197 ss
= nfnetlink_get_subsys(type
);
199 #ifdef CONFIG_MODULES
201 request_module("nfnetlink-subsys-%d", NFNL_SUBSYS_ID(type
));
203 ss
= nfnetlink_get_subsys(type
);
212 nc
= nfnetlink_find_client(type
, ss
);
219 int min_len
= nlmsg_total_size(sizeof(struct nfgenmsg
));
220 u8 cb_id
= NFNL_MSG_TYPE(nlh
->nlmsg_type
);
221 struct nlattr
*cda
[NFNL_MAX_ATTR_COUNT
+ 1];
222 struct nlattr
*attr
= (void *)nlh
+ min_len
;
223 int attrlen
= nlh
->nlmsg_len
- min_len
;
224 __u8 subsys_id
= NFNL_SUBSYS_ID(type
);
226 /* Sanity-check NFNL_MAX_ATTR_COUNT */
227 if (ss
->cb
[cb_id
].attr_count
> NFNL_MAX_ATTR_COUNT
) {
232 err
= nla_parse_deprecated(cda
, ss
->cb
[cb_id
].attr_count
,
234 ss
->cb
[cb_id
].policy
, extack
);
241 err
= nc
->call_rcu(net
, net
->nfnl
, skb
, nlh
,
242 (const struct nlattr
**)cda
,
247 nfnl_lock(subsys_id
);
248 if (nfnl_dereference_protected(subsys_id
) != ss
||
249 nfnetlink_find_client(type
, ss
) != nc
)
252 err
= nc
->call(net
, net
->nfnl
, skb
, nlh
,
253 (const struct nlattr
**)cda
,
257 nfnl_unlock(subsys_id
);
266 struct list_head head
;
267 struct nlmsghdr
*nlh
;
269 struct netlink_ext_ack extack
;
272 static int nfnl_err_add(struct list_head
*list
, struct nlmsghdr
*nlh
, int err
,
273 const struct netlink_ext_ack
*extack
)
275 struct nfnl_err
*nfnl_err
;
277 nfnl_err
= kmalloc(sizeof(struct nfnl_err
), GFP_KERNEL
);
278 if (nfnl_err
== NULL
)
283 nfnl_err
->extack
= *extack
;
284 list_add_tail(&nfnl_err
->head
, list
);
289 static void nfnl_err_del(struct nfnl_err
*nfnl_err
)
291 list_del(&nfnl_err
->head
);
295 static void nfnl_err_reset(struct list_head
*err_list
)
297 struct nfnl_err
*nfnl_err
, *next
;
299 list_for_each_entry_safe(nfnl_err
, next
, err_list
, head
)
300 nfnl_err_del(nfnl_err
);
303 static void nfnl_err_deliver(struct list_head
*err_list
, struct sk_buff
*skb
)
305 struct nfnl_err
*nfnl_err
, *next
;
307 list_for_each_entry_safe(nfnl_err
, next
, err_list
, head
) {
308 netlink_ack(skb
, nfnl_err
->nlh
, nfnl_err
->err
,
310 nfnl_err_del(nfnl_err
);
315 NFNL_BATCH_FAILURE
= (1 << 0),
316 NFNL_BATCH_DONE
= (1 << 1),
317 NFNL_BATCH_REPLAY
= (1 << 2),
320 static void nfnetlink_rcv_batch(struct sk_buff
*skb
, struct nlmsghdr
*nlh
,
321 u16 subsys_id
, u32 genid
)
323 struct sk_buff
*oskb
= skb
;
324 struct net
*net
= sock_net(skb
->sk
);
325 const struct nfnetlink_subsystem
*ss
;
326 const struct nfnl_callback
*nc
;
327 struct netlink_ext_ack extack
;
332 if (subsys_id
>= NFNL_SUBSYS_COUNT
)
333 return netlink_ack(skb
, nlh
, -EINVAL
, NULL
);
337 skb
= netlink_skb_clone(oskb
, GFP_KERNEL
);
339 return netlink_ack(oskb
, nlh
, -ENOMEM
, NULL
);
341 nfnl_lock(subsys_id
);
342 ss
= nfnl_dereference_protected(subsys_id
);
344 #ifdef CONFIG_MODULES
345 nfnl_unlock(subsys_id
);
346 request_module("nfnetlink-subsys-%d", subsys_id
);
347 nfnl_lock(subsys_id
);
348 ss
= nfnl_dereference_protected(subsys_id
);
352 nfnl_unlock(subsys_id
);
353 netlink_ack(oskb
, nlh
, -EOPNOTSUPP
, NULL
);
354 return kfree_skb(skb
);
358 if (!ss
->valid_genid
|| !ss
->commit
|| !ss
->abort
) {
359 nfnl_unlock(subsys_id
);
360 netlink_ack(oskb
, nlh
, -EOPNOTSUPP
, NULL
);
361 return kfree_skb(skb
);
364 if (!try_module_get(ss
->owner
)) {
365 nfnl_unlock(subsys_id
);
366 netlink_ack(oskb
, nlh
, -EOPNOTSUPP
, NULL
);
367 return kfree_skb(skb
);
370 if (!ss
->valid_genid(net
, genid
)) {
371 module_put(ss
->owner
);
372 nfnl_unlock(subsys_id
);
373 netlink_ack(oskb
, nlh
, -ERESTART
, NULL
);
374 return kfree_skb(skb
);
377 nfnl_unlock(subsys_id
);
379 while (skb
->len
>= nlmsg_total_size(0)) {
382 if (fatal_signal_pending(current
)) {
383 nfnl_err_reset(&err_list
);
385 status
= NFNL_BATCH_FAILURE
;
389 memset(&extack
, 0, sizeof(extack
));
390 nlh
= nlmsg_hdr(skb
);
393 if (nlh
->nlmsg_len
< NLMSG_HDRLEN
||
394 skb
->len
< nlh
->nlmsg_len
||
395 nlmsg_len(nlh
) < sizeof(struct nfgenmsg
)) {
396 nfnl_err_reset(&err_list
);
397 status
|= NFNL_BATCH_FAILURE
;
401 /* Only requests are handled by the kernel */
402 if (!(nlh
->nlmsg_flags
& NLM_F_REQUEST
)) {
407 type
= nlh
->nlmsg_type
;
408 if (type
== NFNL_MSG_BATCH_BEGIN
) {
409 /* Malformed: Batch begin twice */
410 nfnl_err_reset(&err_list
);
411 status
|= NFNL_BATCH_FAILURE
;
413 } else if (type
== NFNL_MSG_BATCH_END
) {
414 status
|= NFNL_BATCH_DONE
;
416 } else if (type
< NLMSG_MIN_TYPE
) {
421 /* We only accept a batch with messages for the same
424 if (NFNL_SUBSYS_ID(type
) != subsys_id
) {
429 nc
= nfnetlink_find_client(type
, ss
);
436 int min_len
= nlmsg_total_size(sizeof(struct nfgenmsg
));
437 u8 cb_id
= NFNL_MSG_TYPE(nlh
->nlmsg_type
);
438 struct nlattr
*cda
[NFNL_MAX_ATTR_COUNT
+ 1];
439 struct nlattr
*attr
= (void *)nlh
+ min_len
;
440 int attrlen
= nlh
->nlmsg_len
- min_len
;
442 /* Sanity-check NFTA_MAX_ATTR */
443 if (ss
->cb
[cb_id
].attr_count
> NFNL_MAX_ATTR_COUNT
) {
448 err
= nla_parse_deprecated(cda
,
449 ss
->cb
[cb_id
].attr_count
,
451 ss
->cb
[cb_id
].policy
, NULL
);
455 if (nc
->call_batch
) {
456 err
= nc
->call_batch(net
, net
->nfnl
, skb
, nlh
,
457 (const struct nlattr
**)cda
,
461 /* The lock was released to autoload some module, we
462 * have to abort and start from scratch using the
465 if (err
== -EAGAIN
) {
466 status
|= NFNL_BATCH_REPLAY
;
471 if (nlh
->nlmsg_flags
& NLM_F_ACK
|| err
) {
472 /* Errors are delivered once the full batch has been
473 * processed, this avoids that the same error is
474 * reported several times when replaying the batch.
476 if (nfnl_err_add(&err_list
, nlh
, err
, &extack
) < 0) {
477 /* We failed to enqueue an error, reset the
478 * list of errors and send OOM to userspace
479 * pointing to the batch header.
481 nfnl_err_reset(&err_list
);
482 netlink_ack(oskb
, nlmsg_hdr(oskb
), -ENOMEM
,
484 status
|= NFNL_BATCH_FAILURE
;
487 /* We don't stop processing the batch on errors, thus,
488 * userspace gets all the errors that the batch
492 status
|= NFNL_BATCH_FAILURE
;
495 msglen
= NLMSG_ALIGN(nlh
->nlmsg_len
);
496 if (msglen
> skb
->len
)
498 skb_pull(skb
, msglen
);
501 if (status
& NFNL_BATCH_REPLAY
) {
502 ss
->abort(net
, oskb
, NFNL_ABORT_AUTOLOAD
);
503 nfnl_err_reset(&err_list
);
505 module_put(ss
->owner
);
507 } else if (status
== NFNL_BATCH_DONE
) {
508 err
= ss
->commit(net
, oskb
);
509 if (err
== -EAGAIN
) {
510 status
|= NFNL_BATCH_REPLAY
;
513 ss
->abort(net
, oskb
, NFNL_ABORT_NONE
);
514 netlink_ack(oskb
, nlmsg_hdr(oskb
), err
, NULL
);
517 enum nfnl_abort_action abort_action
;
519 if (status
& NFNL_BATCH_FAILURE
)
520 abort_action
= NFNL_ABORT_NONE
;
522 abort_action
= NFNL_ABORT_VALIDATE
;
524 err
= ss
->abort(net
, oskb
, abort_action
);
525 if (err
== -EAGAIN
) {
526 nfnl_err_reset(&err_list
);
528 module_put(ss
->owner
);
529 status
|= NFNL_BATCH_FAILURE
;
536 nfnl_err_deliver(&err_list
, oskb
);
538 module_put(ss
->owner
);
541 static const struct nla_policy nfnl_batch_policy
[NFNL_BATCH_MAX
+ 1] = {
542 [NFNL_BATCH_GENID
] = { .type
= NLA_U32
},
545 static void nfnetlink_rcv_skb_batch(struct sk_buff
*skb
, struct nlmsghdr
*nlh
)
547 int min_len
= nlmsg_total_size(sizeof(struct nfgenmsg
));
548 struct nlattr
*attr
= (void *)nlh
+ min_len
;
549 struct nlattr
*cda
[NFNL_BATCH_MAX
+ 1];
550 int attrlen
= nlh
->nlmsg_len
- min_len
;
551 struct nfgenmsg
*nfgenmsg
;
556 msglen
= NLMSG_ALIGN(nlh
->nlmsg_len
);
557 if (msglen
> skb
->len
)
560 if (skb
->len
< NLMSG_HDRLEN
+ sizeof(struct nfgenmsg
))
563 err
= nla_parse_deprecated(cda
, NFNL_BATCH_MAX
, attr
, attrlen
,
564 nfnl_batch_policy
, NULL
);
566 netlink_ack(skb
, nlh
, err
, NULL
);
569 if (cda
[NFNL_BATCH_GENID
])
570 gen_id
= ntohl(nla_get_be32(cda
[NFNL_BATCH_GENID
]));
572 nfgenmsg
= nlmsg_data(nlh
);
573 skb_pull(skb
, msglen
);
574 /* Work around old nft using host byte order */
575 if (nfgenmsg
->res_id
== NFNL_SUBSYS_NFTABLES
)
576 res_id
= NFNL_SUBSYS_NFTABLES
;
578 res_id
= ntohs(nfgenmsg
->res_id
);
580 nfnetlink_rcv_batch(skb
, nlh
, res_id
, gen_id
);
583 static void nfnetlink_rcv(struct sk_buff
*skb
)
585 struct nlmsghdr
*nlh
= nlmsg_hdr(skb
);
587 if (skb
->len
< NLMSG_HDRLEN
||
588 nlh
->nlmsg_len
< NLMSG_HDRLEN
||
589 skb
->len
< nlh
->nlmsg_len
)
592 if (!netlink_net_capable(skb
, CAP_NET_ADMIN
)) {
593 netlink_ack(skb
, nlh
, -EPERM
, NULL
);
597 if (nlh
->nlmsg_type
== NFNL_MSG_BATCH_BEGIN
)
598 nfnetlink_rcv_skb_batch(skb
, nlh
);
600 netlink_rcv_skb(skb
, nfnetlink_rcv_msg
);
603 #ifdef CONFIG_MODULES
604 static int nfnetlink_bind(struct net
*net
, int group
)
606 const struct nfnetlink_subsystem
*ss
;
609 if (group
<= NFNLGRP_NONE
|| group
> NFNLGRP_MAX
)
612 type
= nfnl_group2type
[group
];
615 ss
= nfnetlink_get_subsys(type
<< 8);
618 request_module_nowait("nfnetlink-subsys-%d", type
);
623 static int __net_init
nfnetlink_net_init(struct net
*net
)
626 struct netlink_kernel_cfg cfg
= {
627 .groups
= NFNLGRP_MAX
,
628 .input
= nfnetlink_rcv
,
629 #ifdef CONFIG_MODULES
630 .bind
= nfnetlink_bind
,
634 nfnl
= netlink_kernel_create(net
, NETLINK_NETFILTER
, &cfg
);
637 net
->nfnl_stash
= nfnl
;
638 rcu_assign_pointer(net
->nfnl
, nfnl
);
642 static void __net_exit
nfnetlink_net_exit_batch(struct list_head
*net_exit_list
)
646 list_for_each_entry(net
, net_exit_list
, exit_list
)
647 RCU_INIT_POINTER(net
->nfnl
, NULL
);
649 list_for_each_entry(net
, net_exit_list
, exit_list
)
650 netlink_kernel_release(net
->nfnl_stash
);
653 static struct pernet_operations nfnetlink_net_ops
= {
654 .init
= nfnetlink_net_init
,
655 .exit_batch
= nfnetlink_net_exit_batch
,
658 static int __init
nfnetlink_init(void)
662 for (i
= NFNLGRP_NONE
+ 1; i
<= NFNLGRP_MAX
; i
++)
663 BUG_ON(nfnl_group2type
[i
] == NFNL_SUBSYS_NONE
);
665 for (i
=0; i
<NFNL_SUBSYS_COUNT
; i
++)
666 __mutex_init(&table
[i
].mutex
, nfnl_lockdep_names
[i
], &nfnl_lockdep_keys
[i
]);
668 return register_pernet_subsys(&nfnetlink_net_ops
);
671 static void __exit
nfnetlink_exit(void)
673 unregister_pernet_subsys(&nfnetlink_net_ops
);
675 module_init(nfnetlink_init
);
676 module_exit(nfnetlink_exit
);