Loading...
1/* Netfilter messages via netlink socket. Allows for user space
2 * protocol helpers and general trouble making from userspace.
3 *
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>
7 *
8 * Initial netfilter messages via netlink development funded and
9 * generally made possible by Network Robots, Inc. (www.networkrobots.com)
10 *
11 * Further development of this code funded by Astaro AG (http://www.astaro.com)
12 *
13 * This software may be used and distributed according to the terms
14 * of the GNU General Public License, incorporated herein by reference.
15 */
16
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>
26#include <net/sock.h>
27#include <linux/init.h>
28
29#include <net/netlink.h>
30#include <linux/netfilter/nfnetlink.h>
31
32MODULE_LICENSE("GPL");
33MODULE_AUTHOR("Harald Welte <laforge@netfilter.org>");
34MODULE_ALIAS_NET_PF_PROTO(PF_NETLINK, NETLINK_NETFILTER);
35
36#define nfnl_dereference_protected(id) \
37 rcu_dereference_protected(table[(id)].subsys, \
38 lockdep_nfnl_is_held((id)))
39
40static char __initdata nfversion[] = "0.30";
41
42static struct {
43 struct mutex mutex;
44 const struct nfnetlink_subsystem __rcu *subsys;
45} table[NFNL_SUBSYS_COUNT];
46
47static const int nfnl_group2type[NFNLGRP_MAX+1] = {
48 [NFNLGRP_CONNTRACK_NEW] = NFNL_SUBSYS_CTNETLINK,
49 [NFNLGRP_CONNTRACK_UPDATE] = NFNL_SUBSYS_CTNETLINK,
50 [NFNLGRP_CONNTRACK_DESTROY] = NFNL_SUBSYS_CTNETLINK,
51 [NFNLGRP_CONNTRACK_EXP_NEW] = NFNL_SUBSYS_CTNETLINK_EXP,
52 [NFNLGRP_CONNTRACK_EXP_UPDATE] = NFNL_SUBSYS_CTNETLINK_EXP,
53 [NFNLGRP_CONNTRACK_EXP_DESTROY] = NFNL_SUBSYS_CTNETLINK_EXP,
54 [NFNLGRP_NFTABLES] = NFNL_SUBSYS_NFTABLES,
55 [NFNLGRP_ACCT_QUOTA] = NFNL_SUBSYS_ACCT,
56 [NFNLGRP_NFTRACE] = NFNL_SUBSYS_NFTABLES,
57};
58
59void nfnl_lock(__u8 subsys_id)
60{
61 mutex_lock(&table[subsys_id].mutex);
62}
63EXPORT_SYMBOL_GPL(nfnl_lock);
64
65void nfnl_unlock(__u8 subsys_id)
66{
67 mutex_unlock(&table[subsys_id].mutex);
68}
69EXPORT_SYMBOL_GPL(nfnl_unlock);
70
71#ifdef CONFIG_PROVE_LOCKING
72bool lockdep_nfnl_is_held(u8 subsys_id)
73{
74 return lockdep_is_held(&table[subsys_id].mutex);
75}
76EXPORT_SYMBOL_GPL(lockdep_nfnl_is_held);
77#endif
78
79int nfnetlink_subsys_register(const struct nfnetlink_subsystem *n)
80{
81 nfnl_lock(n->subsys_id);
82 if (table[n->subsys_id].subsys) {
83 nfnl_unlock(n->subsys_id);
84 return -EBUSY;
85 }
86 rcu_assign_pointer(table[n->subsys_id].subsys, n);
87 nfnl_unlock(n->subsys_id);
88
89 return 0;
90}
91EXPORT_SYMBOL_GPL(nfnetlink_subsys_register);
92
93int nfnetlink_subsys_unregister(const struct nfnetlink_subsystem *n)
94{
95 nfnl_lock(n->subsys_id);
96 table[n->subsys_id].subsys = NULL;
97 nfnl_unlock(n->subsys_id);
98 synchronize_rcu();
99 return 0;
100}
101EXPORT_SYMBOL_GPL(nfnetlink_subsys_unregister);
102
103static inline const struct nfnetlink_subsystem *nfnetlink_get_subsys(u_int16_t type)
104{
105 u_int8_t subsys_id = NFNL_SUBSYS_ID(type);
106
107 if (subsys_id >= NFNL_SUBSYS_COUNT)
108 return NULL;
109
110 return rcu_dereference(table[subsys_id].subsys);
111}
112
113static inline const struct nfnl_callback *
114nfnetlink_find_client(u_int16_t type, const struct nfnetlink_subsystem *ss)
115{
116 u_int8_t cb_id = NFNL_MSG_TYPE(type);
117
118 if (cb_id >= ss->cb_count)
119 return NULL;
120
121 return &ss->cb[cb_id];
122}
123
124int nfnetlink_has_listeners(struct net *net, unsigned int group)
125{
126 return netlink_has_listeners(net->nfnl, group);
127}
128EXPORT_SYMBOL_GPL(nfnetlink_has_listeners);
129
130int nfnetlink_send(struct sk_buff *skb, struct net *net, u32 portid,
131 unsigned int group, int echo, gfp_t flags)
132{
133 return nlmsg_notify(net->nfnl, skb, portid, group, echo, flags);
134}
135EXPORT_SYMBOL_GPL(nfnetlink_send);
136
137int nfnetlink_set_err(struct net *net, u32 portid, u32 group, int error)
138{
139 return netlink_set_err(net->nfnl, portid, group, error);
140}
141EXPORT_SYMBOL_GPL(nfnetlink_set_err);
142
143int nfnetlink_unicast(struct sk_buff *skb, struct net *net, u32 portid,
144 int flags)
145{
146 return netlink_unicast(net->nfnl, skb, portid, flags);
147}
148EXPORT_SYMBOL_GPL(nfnetlink_unicast);
149
150/* Process one complete nfnetlink message. */
151static int nfnetlink_rcv_msg(struct sk_buff *skb, struct nlmsghdr *nlh)
152{
153 struct net *net = sock_net(skb->sk);
154 const struct nfnl_callback *nc;
155 const struct nfnetlink_subsystem *ss;
156 int type, err;
157
158 /* All the messages must at least contain nfgenmsg */
159 if (nlmsg_len(nlh) < sizeof(struct nfgenmsg))
160 return 0;
161
162 type = nlh->nlmsg_type;
163replay:
164 rcu_read_lock();
165 ss = nfnetlink_get_subsys(type);
166 if (!ss) {
167#ifdef CONFIG_MODULES
168 rcu_read_unlock();
169 request_module("nfnetlink-subsys-%d", NFNL_SUBSYS_ID(type));
170 rcu_read_lock();
171 ss = nfnetlink_get_subsys(type);
172 if (!ss)
173#endif
174 {
175 rcu_read_unlock();
176 return -EINVAL;
177 }
178 }
179
180 nc = nfnetlink_find_client(type, ss);
181 if (!nc) {
182 rcu_read_unlock();
183 return -EINVAL;
184 }
185
186 {
187 int min_len = nlmsg_total_size(sizeof(struct nfgenmsg));
188 u_int8_t cb_id = NFNL_MSG_TYPE(nlh->nlmsg_type);
189 struct nlattr *cda[ss->cb[cb_id].attr_count + 1];
190 struct nlattr *attr = (void *)nlh + min_len;
191 int attrlen = nlh->nlmsg_len - min_len;
192 __u8 subsys_id = NFNL_SUBSYS_ID(type);
193
194 err = nla_parse(cda, ss->cb[cb_id].attr_count,
195 attr, attrlen, ss->cb[cb_id].policy);
196 if (err < 0) {
197 rcu_read_unlock();
198 return err;
199 }
200
201 if (nc->call_rcu) {
202 err = nc->call_rcu(net, net->nfnl, skb, nlh,
203 (const struct nlattr **)cda);
204 rcu_read_unlock();
205 } else {
206 rcu_read_unlock();
207 nfnl_lock(subsys_id);
208 if (nfnl_dereference_protected(subsys_id) != ss ||
209 nfnetlink_find_client(type, ss) != nc)
210 err = -EAGAIN;
211 else if (nc->call)
212 err = nc->call(net, net->nfnl, skb, nlh,
213 (const struct nlattr **)cda);
214 else
215 err = -EINVAL;
216 nfnl_unlock(subsys_id);
217 }
218 if (err == -EAGAIN)
219 goto replay;
220 return err;
221 }
222}
223
224struct nfnl_err {
225 struct list_head head;
226 struct nlmsghdr *nlh;
227 int err;
228};
229
230static int nfnl_err_add(struct list_head *list, struct nlmsghdr *nlh, int err)
231{
232 struct nfnl_err *nfnl_err;
233
234 nfnl_err = kmalloc(sizeof(struct nfnl_err), GFP_KERNEL);
235 if (nfnl_err == NULL)
236 return -ENOMEM;
237
238 nfnl_err->nlh = nlh;
239 nfnl_err->err = err;
240 list_add_tail(&nfnl_err->head, list);
241
242 return 0;
243}
244
245static void nfnl_err_del(struct nfnl_err *nfnl_err)
246{
247 list_del(&nfnl_err->head);
248 kfree(nfnl_err);
249}
250
251static void nfnl_err_reset(struct list_head *err_list)
252{
253 struct nfnl_err *nfnl_err, *next;
254
255 list_for_each_entry_safe(nfnl_err, next, err_list, head)
256 nfnl_err_del(nfnl_err);
257}
258
259static void nfnl_err_deliver(struct list_head *err_list, struct sk_buff *skb)
260{
261 struct nfnl_err *nfnl_err, *next;
262
263 list_for_each_entry_safe(nfnl_err, next, err_list, head) {
264 netlink_ack(skb, nfnl_err->nlh, nfnl_err->err);
265 nfnl_err_del(nfnl_err);
266 }
267}
268
269enum {
270 NFNL_BATCH_FAILURE = (1 << 0),
271 NFNL_BATCH_DONE = (1 << 1),
272 NFNL_BATCH_REPLAY = (1 << 2),
273};
274
275static void nfnetlink_rcv_batch(struct sk_buff *skb, struct nlmsghdr *nlh,
276 u_int16_t subsys_id)
277{
278 struct sk_buff *oskb = skb;
279 struct net *net = sock_net(skb->sk);
280 const struct nfnetlink_subsystem *ss;
281 const struct nfnl_callback *nc;
282 static LIST_HEAD(err_list);
283 u32 status;
284 int err;
285
286 if (subsys_id >= NFNL_SUBSYS_COUNT)
287 return netlink_ack(skb, nlh, -EINVAL);
288replay:
289 status = 0;
290
291 skb = netlink_skb_clone(oskb, GFP_KERNEL);
292 if (!skb)
293 return netlink_ack(oskb, nlh, -ENOMEM);
294
295 nfnl_lock(subsys_id);
296 ss = nfnl_dereference_protected(subsys_id);
297 if (!ss) {
298#ifdef CONFIG_MODULES
299 nfnl_unlock(subsys_id);
300 request_module("nfnetlink-subsys-%d", subsys_id);
301 nfnl_lock(subsys_id);
302 ss = nfnl_dereference_protected(subsys_id);
303 if (!ss)
304#endif
305 {
306 nfnl_unlock(subsys_id);
307 netlink_ack(oskb, nlh, -EOPNOTSUPP);
308 return kfree_skb(skb);
309 }
310 }
311
312 if (!ss->commit || !ss->abort) {
313 nfnl_unlock(subsys_id);
314 netlink_ack(oskb, nlh, -EOPNOTSUPP);
315 return kfree_skb(skb);
316 }
317
318 while (skb->len >= nlmsg_total_size(0)) {
319 int msglen, type;
320
321 nlh = nlmsg_hdr(skb);
322 err = 0;
323
324 if (nlh->nlmsg_len < NLMSG_HDRLEN ||
325 skb->len < nlh->nlmsg_len ||
326 nlmsg_len(nlh) < sizeof(struct nfgenmsg)) {
327 nfnl_err_reset(&err_list);
328 status |= NFNL_BATCH_FAILURE;
329 goto done;
330 }
331
332 /* Only requests are handled by the kernel */
333 if (!(nlh->nlmsg_flags & NLM_F_REQUEST)) {
334 err = -EINVAL;
335 goto ack;
336 }
337
338 type = nlh->nlmsg_type;
339 if (type == NFNL_MSG_BATCH_BEGIN) {
340 /* Malformed: Batch begin twice */
341 nfnl_err_reset(&err_list);
342 status |= NFNL_BATCH_FAILURE;
343 goto done;
344 } else if (type == NFNL_MSG_BATCH_END) {
345 status |= NFNL_BATCH_DONE;
346 goto done;
347 } else if (type < NLMSG_MIN_TYPE) {
348 err = -EINVAL;
349 goto ack;
350 }
351
352 /* We only accept a batch with messages for the same
353 * subsystem.
354 */
355 if (NFNL_SUBSYS_ID(type) != subsys_id) {
356 err = -EINVAL;
357 goto ack;
358 }
359
360 nc = nfnetlink_find_client(type, ss);
361 if (!nc) {
362 err = -EINVAL;
363 goto ack;
364 }
365
366 {
367 int min_len = nlmsg_total_size(sizeof(struct nfgenmsg));
368 u_int8_t cb_id = NFNL_MSG_TYPE(nlh->nlmsg_type);
369 struct nlattr *cda[ss->cb[cb_id].attr_count + 1];
370 struct nlattr *attr = (void *)nlh + min_len;
371 int attrlen = nlh->nlmsg_len - min_len;
372
373 err = nla_parse(cda, ss->cb[cb_id].attr_count,
374 attr, attrlen, ss->cb[cb_id].policy);
375 if (err < 0)
376 goto ack;
377
378 if (nc->call_batch) {
379 err = nc->call_batch(net, net->nfnl, skb, nlh,
380 (const struct nlattr **)cda);
381 }
382
383 /* The lock was released to autoload some module, we
384 * have to abort and start from scratch using the
385 * original skb.
386 */
387 if (err == -EAGAIN) {
388 status |= NFNL_BATCH_REPLAY;
389 goto next;
390 }
391 }
392ack:
393 if (nlh->nlmsg_flags & NLM_F_ACK || err) {
394 /* Errors are delivered once the full batch has been
395 * processed, this avoids that the same error is
396 * reported several times when replaying the batch.
397 */
398 if (nfnl_err_add(&err_list, nlh, err) < 0) {
399 /* We failed to enqueue an error, reset the
400 * list of errors and send OOM to userspace
401 * pointing to the batch header.
402 */
403 nfnl_err_reset(&err_list);
404 netlink_ack(oskb, nlmsg_hdr(oskb), -ENOMEM);
405 status |= NFNL_BATCH_FAILURE;
406 goto done;
407 }
408 /* We don't stop processing the batch on errors, thus,
409 * userspace gets all the errors that the batch
410 * triggers.
411 */
412 if (err)
413 status |= NFNL_BATCH_FAILURE;
414 }
415next:
416 msglen = NLMSG_ALIGN(nlh->nlmsg_len);
417 if (msglen > skb->len)
418 msglen = skb->len;
419 skb_pull(skb, msglen);
420 }
421done:
422 if (status & NFNL_BATCH_REPLAY) {
423 ss->abort(net, oskb);
424 nfnl_err_reset(&err_list);
425 nfnl_unlock(subsys_id);
426 kfree_skb(skb);
427 goto replay;
428 } else if (status == NFNL_BATCH_DONE) {
429 ss->commit(net, oskb);
430 } else {
431 ss->abort(net, oskb);
432 }
433
434 nfnl_err_deliver(&err_list, oskb);
435 nfnl_unlock(subsys_id);
436 kfree_skb(skb);
437}
438
439static void nfnetlink_rcv(struct sk_buff *skb)
440{
441 struct nlmsghdr *nlh = nlmsg_hdr(skb);
442 u_int16_t res_id;
443 int msglen;
444
445 if (nlh->nlmsg_len < NLMSG_HDRLEN ||
446 skb->len < nlh->nlmsg_len)
447 return;
448
449 if (!netlink_net_capable(skb, CAP_NET_ADMIN)) {
450 netlink_ack(skb, nlh, -EPERM);
451 return;
452 }
453
454 if (nlh->nlmsg_type == NFNL_MSG_BATCH_BEGIN) {
455 struct nfgenmsg *nfgenmsg;
456
457 msglen = NLMSG_ALIGN(nlh->nlmsg_len);
458 if (msglen > skb->len)
459 msglen = skb->len;
460
461 if (nlh->nlmsg_len < NLMSG_HDRLEN ||
462 skb->len < NLMSG_HDRLEN + sizeof(struct nfgenmsg))
463 return;
464
465 nfgenmsg = nlmsg_data(nlh);
466 skb_pull(skb, msglen);
467 /* Work around old nft using host byte order */
468 if (nfgenmsg->res_id == NFNL_SUBSYS_NFTABLES)
469 res_id = NFNL_SUBSYS_NFTABLES;
470 else
471 res_id = ntohs(nfgenmsg->res_id);
472 nfnetlink_rcv_batch(skb, nlh, res_id);
473 } else {
474 netlink_rcv_skb(skb, &nfnetlink_rcv_msg);
475 }
476}
477
478#ifdef CONFIG_MODULES
479static int nfnetlink_bind(struct net *net, int group)
480{
481 const struct nfnetlink_subsystem *ss;
482 int type;
483
484 if (group <= NFNLGRP_NONE || group > NFNLGRP_MAX)
485 return 0;
486
487 type = nfnl_group2type[group];
488
489 rcu_read_lock();
490 ss = nfnetlink_get_subsys(type << 8);
491 rcu_read_unlock();
492 if (!ss)
493 request_module("nfnetlink-subsys-%d", type);
494 return 0;
495}
496#endif
497
498static int __net_init nfnetlink_net_init(struct net *net)
499{
500 struct sock *nfnl;
501 struct netlink_kernel_cfg cfg = {
502 .groups = NFNLGRP_MAX,
503 .input = nfnetlink_rcv,
504#ifdef CONFIG_MODULES
505 .bind = nfnetlink_bind,
506#endif
507 };
508
509 nfnl = netlink_kernel_create(net, NETLINK_NETFILTER, &cfg);
510 if (!nfnl)
511 return -ENOMEM;
512 net->nfnl_stash = nfnl;
513 rcu_assign_pointer(net->nfnl, nfnl);
514 return 0;
515}
516
517static void __net_exit nfnetlink_net_exit_batch(struct list_head *net_exit_list)
518{
519 struct net *net;
520
521 list_for_each_entry(net, net_exit_list, exit_list)
522 RCU_INIT_POINTER(net->nfnl, NULL);
523 synchronize_net();
524 list_for_each_entry(net, net_exit_list, exit_list)
525 netlink_kernel_release(net->nfnl_stash);
526}
527
528static struct pernet_operations nfnetlink_net_ops = {
529 .init = nfnetlink_net_init,
530 .exit_batch = nfnetlink_net_exit_batch,
531};
532
533static int __init nfnetlink_init(void)
534{
535 int i;
536
537 for (i = NFNLGRP_NONE + 1; i <= NFNLGRP_MAX; i++)
538 BUG_ON(nfnl_group2type[i] == NFNL_SUBSYS_NONE);
539
540 for (i=0; i<NFNL_SUBSYS_COUNT; i++)
541 mutex_init(&table[i].mutex);
542
543 pr_info("Netfilter messages via NETLINK v%s.\n", nfversion);
544 return register_pernet_subsys(&nfnetlink_net_ops);
545}
546
547static void __exit nfnetlink_exit(void)
548{
549 pr_info("Removing netfilter NETLINK layer.\n");
550 unregister_pernet_subsys(&nfnetlink_net_ops);
551}
552module_init(nfnetlink_init);
553module_exit(nfnetlink_exit);
1/* Netfilter messages via netlink socket. Allows for user space
2 * protocol helpers and general trouble making from userspace.
3 *
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>
7 *
8 * Initial netfilter messages via netlink development funded and
9 * generally made possible by Network Robots, Inc. (www.networkrobots.com)
10 *
11 * Further development of this code funded by Astaro AG (http://www.astaro.com)
12 *
13 * This software may be used and distributed according to the terms
14 * of the GNU General Public License, incorporated herein by reference.
15 */
16
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>
26#include <net/sock.h>
27#include <linux/init.h>
28#include <linux/sched/signal.h>
29
30#include <net/netlink.h>
31#include <linux/netfilter/nfnetlink.h>
32
33MODULE_LICENSE("GPL");
34MODULE_AUTHOR("Harald Welte <laforge@netfilter.org>");
35MODULE_ALIAS_NET_PF_PROTO(PF_NETLINK, NETLINK_NETFILTER);
36MODULE_DESCRIPTION("Netfilter messages via netlink socket");
37
38#define nfnl_dereference_protected(id) \
39 rcu_dereference_protected(table[(id)].subsys, \
40 lockdep_nfnl_is_held((id)))
41
42#define NFNL_MAX_ATTR_COUNT 32
43
44static struct {
45 struct mutex mutex;
46 const struct nfnetlink_subsystem __rcu *subsys;
47} table[NFNL_SUBSYS_COUNT];
48
49static const int nfnl_group2type[NFNLGRP_MAX+1] = {
50 [NFNLGRP_CONNTRACK_NEW] = NFNL_SUBSYS_CTNETLINK,
51 [NFNLGRP_CONNTRACK_UPDATE] = NFNL_SUBSYS_CTNETLINK,
52 [NFNLGRP_CONNTRACK_DESTROY] = NFNL_SUBSYS_CTNETLINK,
53 [NFNLGRP_CONNTRACK_EXP_NEW] = NFNL_SUBSYS_CTNETLINK_EXP,
54 [NFNLGRP_CONNTRACK_EXP_UPDATE] = NFNL_SUBSYS_CTNETLINK_EXP,
55 [NFNLGRP_CONNTRACK_EXP_DESTROY] = NFNL_SUBSYS_CTNETLINK_EXP,
56 [NFNLGRP_NFTABLES] = NFNL_SUBSYS_NFTABLES,
57 [NFNLGRP_ACCT_QUOTA] = NFNL_SUBSYS_ACCT,
58 [NFNLGRP_NFTRACE] = NFNL_SUBSYS_NFTABLES,
59};
60
61void nfnl_lock(__u8 subsys_id)
62{
63 mutex_lock(&table[subsys_id].mutex);
64}
65EXPORT_SYMBOL_GPL(nfnl_lock);
66
67void nfnl_unlock(__u8 subsys_id)
68{
69 mutex_unlock(&table[subsys_id].mutex);
70}
71EXPORT_SYMBOL_GPL(nfnl_unlock);
72
73#ifdef CONFIG_PROVE_LOCKING
74bool lockdep_nfnl_is_held(u8 subsys_id)
75{
76 return lockdep_is_held(&table[subsys_id].mutex);
77}
78EXPORT_SYMBOL_GPL(lockdep_nfnl_is_held);
79#endif
80
81int nfnetlink_subsys_register(const struct nfnetlink_subsystem *n)
82{
83 u8 cb_id;
84
85 /* Sanity-check attr_count size to avoid stack buffer overflow. */
86 for (cb_id = 0; cb_id < n->cb_count; cb_id++)
87 if (WARN_ON(n->cb[cb_id].attr_count > NFNL_MAX_ATTR_COUNT))
88 return -EINVAL;
89
90 nfnl_lock(n->subsys_id);
91 if (table[n->subsys_id].subsys) {
92 nfnl_unlock(n->subsys_id);
93 return -EBUSY;
94 }
95 rcu_assign_pointer(table[n->subsys_id].subsys, n);
96 nfnl_unlock(n->subsys_id);
97
98 return 0;
99}
100EXPORT_SYMBOL_GPL(nfnetlink_subsys_register);
101
102int nfnetlink_subsys_unregister(const struct nfnetlink_subsystem *n)
103{
104 nfnl_lock(n->subsys_id);
105 table[n->subsys_id].subsys = NULL;
106 nfnl_unlock(n->subsys_id);
107 synchronize_rcu();
108 return 0;
109}
110EXPORT_SYMBOL_GPL(nfnetlink_subsys_unregister);
111
112static inline const struct nfnetlink_subsystem *nfnetlink_get_subsys(u16 type)
113{
114 u8 subsys_id = NFNL_SUBSYS_ID(type);
115
116 if (subsys_id >= NFNL_SUBSYS_COUNT)
117 return NULL;
118
119 return rcu_dereference(table[subsys_id].subsys);
120}
121
122static inline const struct nfnl_callback *
123nfnetlink_find_client(u16 type, const struct nfnetlink_subsystem *ss)
124{
125 u8 cb_id = NFNL_MSG_TYPE(type);
126
127 if (cb_id >= ss->cb_count)
128 return NULL;
129
130 return &ss->cb[cb_id];
131}
132
133int nfnetlink_has_listeners(struct net *net, unsigned int group)
134{
135 return netlink_has_listeners(net->nfnl, group);
136}
137EXPORT_SYMBOL_GPL(nfnetlink_has_listeners);
138
139int nfnetlink_send(struct sk_buff *skb, struct net *net, u32 portid,
140 unsigned int group, int echo, gfp_t flags)
141{
142 return nlmsg_notify(net->nfnl, skb, portid, group, echo, flags);
143}
144EXPORT_SYMBOL_GPL(nfnetlink_send);
145
146int nfnetlink_set_err(struct net *net, u32 portid, u32 group, int error)
147{
148 return netlink_set_err(net->nfnl, portid, group, error);
149}
150EXPORT_SYMBOL_GPL(nfnetlink_set_err);
151
152int nfnetlink_unicast(struct sk_buff *skb, struct net *net, u32 portid)
153{
154 int err;
155
156 err = nlmsg_unicast(net->nfnl, skb, portid);
157 if (err == -EAGAIN)
158 err = -ENOBUFS;
159
160 return err;
161}
162EXPORT_SYMBOL_GPL(nfnetlink_unicast);
163
164/* Process one complete nfnetlink message. */
165static int nfnetlink_rcv_msg(struct sk_buff *skb, struct nlmsghdr *nlh,
166 struct netlink_ext_ack *extack)
167{
168 struct net *net = sock_net(skb->sk);
169 const struct nfnl_callback *nc;
170 const struct nfnetlink_subsystem *ss;
171 int type, err;
172
173 /* All the messages must at least contain nfgenmsg */
174 if (nlmsg_len(nlh) < sizeof(struct nfgenmsg))
175 return 0;
176
177 type = nlh->nlmsg_type;
178replay:
179 rcu_read_lock();
180 ss = nfnetlink_get_subsys(type);
181 if (!ss) {
182#ifdef CONFIG_MODULES
183 rcu_read_unlock();
184 request_module("nfnetlink-subsys-%d", NFNL_SUBSYS_ID(type));
185 rcu_read_lock();
186 ss = nfnetlink_get_subsys(type);
187 if (!ss)
188#endif
189 {
190 rcu_read_unlock();
191 return -EINVAL;
192 }
193 }
194
195 nc = nfnetlink_find_client(type, ss);
196 if (!nc) {
197 rcu_read_unlock();
198 return -EINVAL;
199 }
200
201 {
202 int min_len = nlmsg_total_size(sizeof(struct nfgenmsg));
203 u8 cb_id = NFNL_MSG_TYPE(nlh->nlmsg_type);
204 struct nlattr *cda[NFNL_MAX_ATTR_COUNT + 1];
205 struct nlattr *attr = (void *)nlh + min_len;
206 int attrlen = nlh->nlmsg_len - min_len;
207 __u8 subsys_id = NFNL_SUBSYS_ID(type);
208
209 /* Sanity-check NFNL_MAX_ATTR_COUNT */
210 if (ss->cb[cb_id].attr_count > NFNL_MAX_ATTR_COUNT) {
211 rcu_read_unlock();
212 return -ENOMEM;
213 }
214
215 err = nla_parse_deprecated(cda, ss->cb[cb_id].attr_count,
216 attr, attrlen,
217 ss->cb[cb_id].policy, extack);
218 if (err < 0) {
219 rcu_read_unlock();
220 return err;
221 }
222
223 if (nc->call_rcu) {
224 err = nc->call_rcu(net, net->nfnl, skb, nlh,
225 (const struct nlattr **)cda,
226 extack);
227 rcu_read_unlock();
228 } else {
229 rcu_read_unlock();
230 nfnl_lock(subsys_id);
231 if (nfnl_dereference_protected(subsys_id) != ss ||
232 nfnetlink_find_client(type, ss) != nc)
233 err = -EAGAIN;
234 else if (nc->call)
235 err = nc->call(net, net->nfnl, skb, nlh,
236 (const struct nlattr **)cda,
237 extack);
238 else
239 err = -EINVAL;
240 nfnl_unlock(subsys_id);
241 }
242 if (err == -EAGAIN)
243 goto replay;
244 return err;
245 }
246}
247
248struct nfnl_err {
249 struct list_head head;
250 struct nlmsghdr *nlh;
251 int err;
252 struct netlink_ext_ack extack;
253};
254
255static int nfnl_err_add(struct list_head *list, struct nlmsghdr *nlh, int err,
256 const struct netlink_ext_ack *extack)
257{
258 struct nfnl_err *nfnl_err;
259
260 nfnl_err = kmalloc(sizeof(struct nfnl_err), GFP_KERNEL);
261 if (nfnl_err == NULL)
262 return -ENOMEM;
263
264 nfnl_err->nlh = nlh;
265 nfnl_err->err = err;
266 nfnl_err->extack = *extack;
267 list_add_tail(&nfnl_err->head, list);
268
269 return 0;
270}
271
272static void nfnl_err_del(struct nfnl_err *nfnl_err)
273{
274 list_del(&nfnl_err->head);
275 kfree(nfnl_err);
276}
277
278static void nfnl_err_reset(struct list_head *err_list)
279{
280 struct nfnl_err *nfnl_err, *next;
281
282 list_for_each_entry_safe(nfnl_err, next, err_list, head)
283 nfnl_err_del(nfnl_err);
284}
285
286static void nfnl_err_deliver(struct list_head *err_list, struct sk_buff *skb)
287{
288 struct nfnl_err *nfnl_err, *next;
289
290 list_for_each_entry_safe(nfnl_err, next, err_list, head) {
291 netlink_ack(skb, nfnl_err->nlh, nfnl_err->err,
292 &nfnl_err->extack);
293 nfnl_err_del(nfnl_err);
294 }
295}
296
297enum {
298 NFNL_BATCH_FAILURE = (1 << 0),
299 NFNL_BATCH_DONE = (1 << 1),
300 NFNL_BATCH_REPLAY = (1 << 2),
301};
302
303static void nfnetlink_rcv_batch(struct sk_buff *skb, struct nlmsghdr *nlh,
304 u16 subsys_id, u32 genid)
305{
306 struct sk_buff *oskb = skb;
307 struct net *net = sock_net(skb->sk);
308 const struct nfnetlink_subsystem *ss;
309 const struct nfnl_callback *nc;
310 struct netlink_ext_ack extack;
311 LIST_HEAD(err_list);
312 u32 status;
313 int err;
314
315 if (subsys_id >= NFNL_SUBSYS_COUNT)
316 return netlink_ack(skb, nlh, -EINVAL, NULL);
317replay:
318 status = 0;
319
320 skb = netlink_skb_clone(oskb, GFP_KERNEL);
321 if (!skb)
322 return netlink_ack(oskb, nlh, -ENOMEM, NULL);
323
324 nfnl_lock(subsys_id);
325 ss = nfnl_dereference_protected(subsys_id);
326 if (!ss) {
327#ifdef CONFIG_MODULES
328 nfnl_unlock(subsys_id);
329 request_module("nfnetlink-subsys-%d", subsys_id);
330 nfnl_lock(subsys_id);
331 ss = nfnl_dereference_protected(subsys_id);
332 if (!ss)
333#endif
334 {
335 nfnl_unlock(subsys_id);
336 netlink_ack(oskb, nlh, -EOPNOTSUPP, NULL);
337 return kfree_skb(skb);
338 }
339 }
340
341 if (!ss->valid_genid || !ss->commit || !ss->abort) {
342 nfnl_unlock(subsys_id);
343 netlink_ack(oskb, nlh, -EOPNOTSUPP, NULL);
344 return kfree_skb(skb);
345 }
346
347 if (!try_module_get(ss->owner)) {
348 nfnl_unlock(subsys_id);
349 netlink_ack(oskb, nlh, -EOPNOTSUPP, NULL);
350 return kfree_skb(skb);
351 }
352
353 if (!ss->valid_genid(net, genid)) {
354 module_put(ss->owner);
355 nfnl_unlock(subsys_id);
356 netlink_ack(oskb, nlh, -ERESTART, NULL);
357 return kfree_skb(skb);
358 }
359
360 nfnl_unlock(subsys_id);
361
362 while (skb->len >= nlmsg_total_size(0)) {
363 int msglen, type;
364
365 if (fatal_signal_pending(current)) {
366 nfnl_err_reset(&err_list);
367 err = -EINTR;
368 status = NFNL_BATCH_FAILURE;
369 goto done;
370 }
371
372 memset(&extack, 0, sizeof(extack));
373 nlh = nlmsg_hdr(skb);
374 err = 0;
375
376 if (nlh->nlmsg_len < NLMSG_HDRLEN ||
377 skb->len < nlh->nlmsg_len ||
378 nlmsg_len(nlh) < sizeof(struct nfgenmsg)) {
379 nfnl_err_reset(&err_list);
380 status |= NFNL_BATCH_FAILURE;
381 goto done;
382 }
383
384 /* Only requests are handled by the kernel */
385 if (!(nlh->nlmsg_flags & NLM_F_REQUEST)) {
386 err = -EINVAL;
387 goto ack;
388 }
389
390 type = nlh->nlmsg_type;
391 if (type == NFNL_MSG_BATCH_BEGIN) {
392 /* Malformed: Batch begin twice */
393 nfnl_err_reset(&err_list);
394 status |= NFNL_BATCH_FAILURE;
395 goto done;
396 } else if (type == NFNL_MSG_BATCH_END) {
397 status |= NFNL_BATCH_DONE;
398 goto done;
399 } else if (type < NLMSG_MIN_TYPE) {
400 err = -EINVAL;
401 goto ack;
402 }
403
404 /* We only accept a batch with messages for the same
405 * subsystem.
406 */
407 if (NFNL_SUBSYS_ID(type) != subsys_id) {
408 err = -EINVAL;
409 goto ack;
410 }
411
412 nc = nfnetlink_find_client(type, ss);
413 if (!nc) {
414 err = -EINVAL;
415 goto ack;
416 }
417
418 {
419 int min_len = nlmsg_total_size(sizeof(struct nfgenmsg));
420 u8 cb_id = NFNL_MSG_TYPE(nlh->nlmsg_type);
421 struct nlattr *cda[NFNL_MAX_ATTR_COUNT + 1];
422 struct nlattr *attr = (void *)nlh + min_len;
423 int attrlen = nlh->nlmsg_len - min_len;
424
425 /* Sanity-check NFTA_MAX_ATTR */
426 if (ss->cb[cb_id].attr_count > NFNL_MAX_ATTR_COUNT) {
427 err = -ENOMEM;
428 goto ack;
429 }
430
431 err = nla_parse_deprecated(cda,
432 ss->cb[cb_id].attr_count,
433 attr, attrlen,
434 ss->cb[cb_id].policy, NULL);
435 if (err < 0)
436 goto ack;
437
438 if (nc->call_batch) {
439 err = nc->call_batch(net, net->nfnl, skb, nlh,
440 (const struct nlattr **)cda,
441 &extack);
442 }
443
444 /* The lock was released to autoload some module, we
445 * have to abort and start from scratch using the
446 * original skb.
447 */
448 if (err == -EAGAIN) {
449 status |= NFNL_BATCH_REPLAY;
450 goto done;
451 }
452 }
453ack:
454 if (nlh->nlmsg_flags & NLM_F_ACK || err) {
455 /* Errors are delivered once the full batch has been
456 * processed, this avoids that the same error is
457 * reported several times when replaying the batch.
458 */
459 if (nfnl_err_add(&err_list, nlh, err, &extack) < 0) {
460 /* We failed to enqueue an error, reset the
461 * list of errors and send OOM to userspace
462 * pointing to the batch header.
463 */
464 nfnl_err_reset(&err_list);
465 netlink_ack(oskb, nlmsg_hdr(oskb), -ENOMEM,
466 NULL);
467 status |= NFNL_BATCH_FAILURE;
468 goto done;
469 }
470 /* We don't stop processing the batch on errors, thus,
471 * userspace gets all the errors that the batch
472 * triggers.
473 */
474 if (err)
475 status |= NFNL_BATCH_FAILURE;
476 }
477
478 msglen = NLMSG_ALIGN(nlh->nlmsg_len);
479 if (msglen > skb->len)
480 msglen = skb->len;
481 skb_pull(skb, msglen);
482 }
483done:
484 if (status & NFNL_BATCH_REPLAY) {
485 ss->abort(net, oskb, true);
486 nfnl_err_reset(&err_list);
487 kfree_skb(skb);
488 module_put(ss->owner);
489 goto replay;
490 } else if (status == NFNL_BATCH_DONE) {
491 err = ss->commit(net, oskb);
492 if (err == -EAGAIN) {
493 status |= NFNL_BATCH_REPLAY;
494 goto done;
495 } else if (err) {
496 ss->abort(net, oskb, false);
497 netlink_ack(oskb, nlmsg_hdr(oskb), err, NULL);
498 }
499 } else {
500 ss->abort(net, oskb, false);
501 }
502 if (ss->cleanup)
503 ss->cleanup(net);
504
505 nfnl_err_deliver(&err_list, oskb);
506 kfree_skb(skb);
507 module_put(ss->owner);
508}
509
510static const struct nla_policy nfnl_batch_policy[NFNL_BATCH_MAX + 1] = {
511 [NFNL_BATCH_GENID] = { .type = NLA_U32 },
512};
513
514static void nfnetlink_rcv_skb_batch(struct sk_buff *skb, struct nlmsghdr *nlh)
515{
516 int min_len = nlmsg_total_size(sizeof(struct nfgenmsg));
517 struct nlattr *attr = (void *)nlh + min_len;
518 struct nlattr *cda[NFNL_BATCH_MAX + 1];
519 int attrlen = nlh->nlmsg_len - min_len;
520 struct nfgenmsg *nfgenmsg;
521 int msglen, err;
522 u32 gen_id = 0;
523 u16 res_id;
524
525 msglen = NLMSG_ALIGN(nlh->nlmsg_len);
526 if (msglen > skb->len)
527 msglen = skb->len;
528
529 if (skb->len < NLMSG_HDRLEN + sizeof(struct nfgenmsg))
530 return;
531
532 err = nla_parse_deprecated(cda, NFNL_BATCH_MAX, attr, attrlen,
533 nfnl_batch_policy, NULL);
534 if (err < 0) {
535 netlink_ack(skb, nlh, err, NULL);
536 return;
537 }
538 if (cda[NFNL_BATCH_GENID])
539 gen_id = ntohl(nla_get_be32(cda[NFNL_BATCH_GENID]));
540
541 nfgenmsg = nlmsg_data(nlh);
542 skb_pull(skb, msglen);
543 /* Work around old nft using host byte order */
544 if (nfgenmsg->res_id == NFNL_SUBSYS_NFTABLES)
545 res_id = NFNL_SUBSYS_NFTABLES;
546 else
547 res_id = ntohs(nfgenmsg->res_id);
548
549 nfnetlink_rcv_batch(skb, nlh, res_id, gen_id);
550}
551
552static void nfnetlink_rcv(struct sk_buff *skb)
553{
554 struct nlmsghdr *nlh = nlmsg_hdr(skb);
555
556 if (skb->len < NLMSG_HDRLEN ||
557 nlh->nlmsg_len < NLMSG_HDRLEN ||
558 skb->len < nlh->nlmsg_len)
559 return;
560
561 if (!netlink_net_capable(skb, CAP_NET_ADMIN)) {
562 netlink_ack(skb, nlh, -EPERM, NULL);
563 return;
564 }
565
566 if (nlh->nlmsg_type == NFNL_MSG_BATCH_BEGIN)
567 nfnetlink_rcv_skb_batch(skb, nlh);
568 else
569 netlink_rcv_skb(skb, nfnetlink_rcv_msg);
570}
571
572#ifdef CONFIG_MODULES
573static int nfnetlink_bind(struct net *net, int group)
574{
575 const struct nfnetlink_subsystem *ss;
576 int type;
577
578 if (group <= NFNLGRP_NONE || group > NFNLGRP_MAX)
579 return 0;
580
581 type = nfnl_group2type[group];
582
583 rcu_read_lock();
584 ss = nfnetlink_get_subsys(type << 8);
585 rcu_read_unlock();
586 if (!ss)
587 request_module_nowait("nfnetlink-subsys-%d", type);
588 return 0;
589}
590#endif
591
592static int __net_init nfnetlink_net_init(struct net *net)
593{
594 struct sock *nfnl;
595 struct netlink_kernel_cfg cfg = {
596 .groups = NFNLGRP_MAX,
597 .input = nfnetlink_rcv,
598#ifdef CONFIG_MODULES
599 .bind = nfnetlink_bind,
600#endif
601 };
602
603 nfnl = netlink_kernel_create(net, NETLINK_NETFILTER, &cfg);
604 if (!nfnl)
605 return -ENOMEM;
606 net->nfnl_stash = nfnl;
607 rcu_assign_pointer(net->nfnl, nfnl);
608 return 0;
609}
610
611static void __net_exit nfnetlink_net_exit_batch(struct list_head *net_exit_list)
612{
613 struct net *net;
614
615 list_for_each_entry(net, net_exit_list, exit_list)
616 RCU_INIT_POINTER(net->nfnl, NULL);
617 synchronize_net();
618 list_for_each_entry(net, net_exit_list, exit_list)
619 netlink_kernel_release(net->nfnl_stash);
620}
621
622static struct pernet_operations nfnetlink_net_ops = {
623 .init = nfnetlink_net_init,
624 .exit_batch = nfnetlink_net_exit_batch,
625};
626
627static int __init nfnetlink_init(void)
628{
629 int i;
630
631 for (i = NFNLGRP_NONE + 1; i <= NFNLGRP_MAX; i++)
632 BUG_ON(nfnl_group2type[i] == NFNL_SUBSYS_NONE);
633
634 for (i=0; i<NFNL_SUBSYS_COUNT; i++)
635 mutex_init(&table[i].mutex);
636
637 return register_pernet_subsys(&nfnetlink_net_ops);
638}
639
640static void __exit nfnetlink_exit(void)
641{
642 unregister_pernet_subsys(&nfnetlink_net_ops);
643}
644module_init(nfnetlink_init);
645module_exit(nfnetlink_exit);