Loading...
1/*
2 * This is a module which is used for queueing packets and communicating with
3 * userspace via nfnetlink.
4 *
5 * (C) 2005 by Harald Welte <laforge@netfilter.org>
6 * (C) 2007 by Patrick McHardy <kaber@trash.net>
7 *
8 * Based on the old ipv4-only ip_queue.c:
9 * (C) 2000-2002 James Morris <jmorris@intercode.com.au>
10 * (C) 2003-2005 Netfilter Core Team <coreteam@netfilter.org>
11 *
12 * This program is free software; you can redistribute it and/or modify
13 * it under the terms of the GNU General Public License version 2 as
14 * published by the Free Software Foundation.
15 *
16 */
17#include <linux/module.h>
18#include <linux/skbuff.h>
19#include <linux/init.h>
20#include <linux/spinlock.h>
21#include <linux/slab.h>
22#include <linux/notifier.h>
23#include <linux/netdevice.h>
24#include <linux/netfilter.h>
25#include <linux/proc_fs.h>
26#include <linux/netfilter_ipv4.h>
27#include <linux/netfilter_ipv6.h>
28#include <linux/netfilter_bridge.h>
29#include <linux/netfilter/nfnetlink.h>
30#include <linux/netfilter/nfnetlink_queue.h>
31#include <linux/netfilter/nf_conntrack_common.h>
32#include <linux/list.h>
33#include <net/sock.h>
34#include <net/tcp_states.h>
35#include <net/netfilter/nf_queue.h>
36#include <net/netns/generic.h>
37
38#include <linux/atomic.h>
39
40#if IS_ENABLED(CONFIG_BRIDGE_NETFILTER)
41#include "../bridge/br_private.h"
42#endif
43
44#define NFQNL_QMAX_DEFAULT 1024
45
46/* We're using struct nlattr which has 16bit nla_len. Note that nla_len
47 * includes the header length. Thus, the maximum packet length that we
48 * support is 65531 bytes. We send truncated packets if the specified length
49 * is larger than that. Userspace can check for presence of NFQA_CAP_LEN
50 * attribute to detect truncation.
51 */
52#define NFQNL_MAX_COPY_RANGE (0xffff - NLA_HDRLEN)
53
54struct nfqnl_instance {
55 struct hlist_node hlist; /* global list of queues */
56 struct rcu_head rcu;
57
58 u32 peer_portid;
59 unsigned int queue_maxlen;
60 unsigned int copy_range;
61 unsigned int queue_dropped;
62 unsigned int queue_user_dropped;
63
64
65 u_int16_t queue_num; /* number of this queue */
66 u_int8_t copy_mode;
67 u_int32_t flags; /* Set using NFQA_CFG_FLAGS */
68/*
69 * Following fields are dirtied for each queued packet,
70 * keep them in same cache line if possible.
71 */
72 spinlock_t lock ____cacheline_aligned_in_smp;
73 unsigned int queue_total;
74 unsigned int id_sequence; /* 'sequence' of pkt ids */
75 struct list_head queue_list; /* packets in queue */
76};
77
78typedef int (*nfqnl_cmpfn)(struct nf_queue_entry *, unsigned long);
79
80static unsigned int nfnl_queue_net_id __read_mostly;
81
82#define INSTANCE_BUCKETS 16
83struct nfnl_queue_net {
84 spinlock_t instances_lock;
85 struct hlist_head instance_table[INSTANCE_BUCKETS];
86};
87
88static struct nfnl_queue_net *nfnl_queue_pernet(struct net *net)
89{
90 return net_generic(net, nfnl_queue_net_id);
91}
92
93static inline u_int8_t instance_hashfn(u_int16_t queue_num)
94{
95 return ((queue_num >> 8) ^ queue_num) % INSTANCE_BUCKETS;
96}
97
98static struct nfqnl_instance *
99instance_lookup(struct nfnl_queue_net *q, u_int16_t queue_num)
100{
101 struct hlist_head *head;
102 struct nfqnl_instance *inst;
103
104 head = &q->instance_table[instance_hashfn(queue_num)];
105 hlist_for_each_entry_rcu(inst, head, hlist) {
106 if (inst->queue_num == queue_num)
107 return inst;
108 }
109 return NULL;
110}
111
112static struct nfqnl_instance *
113instance_create(struct nfnl_queue_net *q, u_int16_t queue_num, u32 portid)
114{
115 struct nfqnl_instance *inst;
116 unsigned int h;
117 int err;
118
119 spin_lock(&q->instances_lock);
120 if (instance_lookup(q, queue_num)) {
121 err = -EEXIST;
122 goto out_unlock;
123 }
124
125 inst = kzalloc(sizeof(*inst), GFP_ATOMIC);
126 if (!inst) {
127 err = -ENOMEM;
128 goto out_unlock;
129 }
130
131 inst->queue_num = queue_num;
132 inst->peer_portid = portid;
133 inst->queue_maxlen = NFQNL_QMAX_DEFAULT;
134 inst->copy_range = NFQNL_MAX_COPY_RANGE;
135 inst->copy_mode = NFQNL_COPY_NONE;
136 spin_lock_init(&inst->lock);
137 INIT_LIST_HEAD(&inst->queue_list);
138
139 if (!try_module_get(THIS_MODULE)) {
140 err = -EAGAIN;
141 goto out_free;
142 }
143
144 h = instance_hashfn(queue_num);
145 hlist_add_head_rcu(&inst->hlist, &q->instance_table[h]);
146
147 spin_unlock(&q->instances_lock);
148
149 return inst;
150
151out_free:
152 kfree(inst);
153out_unlock:
154 spin_unlock(&q->instances_lock);
155 return ERR_PTR(err);
156}
157
158static void nfqnl_flush(struct nfqnl_instance *queue, nfqnl_cmpfn cmpfn,
159 unsigned long data);
160
161static void
162instance_destroy_rcu(struct rcu_head *head)
163{
164 struct nfqnl_instance *inst = container_of(head, struct nfqnl_instance,
165 rcu);
166
167 nfqnl_flush(inst, NULL, 0);
168 kfree(inst);
169 module_put(THIS_MODULE);
170}
171
172static void
173__instance_destroy(struct nfqnl_instance *inst)
174{
175 hlist_del_rcu(&inst->hlist);
176 call_rcu(&inst->rcu, instance_destroy_rcu);
177}
178
179static void
180instance_destroy(struct nfnl_queue_net *q, struct nfqnl_instance *inst)
181{
182 spin_lock(&q->instances_lock);
183 __instance_destroy(inst);
184 spin_unlock(&q->instances_lock);
185}
186
187static inline void
188__enqueue_entry(struct nfqnl_instance *queue, struct nf_queue_entry *entry)
189{
190 list_add_tail(&entry->list, &queue->queue_list);
191 queue->queue_total++;
192}
193
194static void
195__dequeue_entry(struct nfqnl_instance *queue, struct nf_queue_entry *entry)
196{
197 list_del(&entry->list);
198 queue->queue_total--;
199}
200
201static struct nf_queue_entry *
202find_dequeue_entry(struct nfqnl_instance *queue, unsigned int id)
203{
204 struct nf_queue_entry *entry = NULL, *i;
205
206 spin_lock_bh(&queue->lock);
207
208 list_for_each_entry(i, &queue->queue_list, list) {
209 if (i->id == id) {
210 entry = i;
211 break;
212 }
213 }
214
215 if (entry)
216 __dequeue_entry(queue, entry);
217
218 spin_unlock_bh(&queue->lock);
219
220 return entry;
221}
222
223static void
224nfqnl_flush(struct nfqnl_instance *queue, nfqnl_cmpfn cmpfn, unsigned long data)
225{
226 struct nf_queue_entry *entry, *next;
227
228 spin_lock_bh(&queue->lock);
229 list_for_each_entry_safe(entry, next, &queue->queue_list, list) {
230 if (!cmpfn || cmpfn(entry, data)) {
231 list_del(&entry->list);
232 queue->queue_total--;
233 nf_reinject(entry, NF_DROP);
234 }
235 }
236 spin_unlock_bh(&queue->lock);
237}
238
239static int
240nfqnl_put_packet_info(struct sk_buff *nlskb, struct sk_buff *packet,
241 bool csum_verify)
242{
243 __u32 flags = 0;
244
245 if (packet->ip_summed == CHECKSUM_PARTIAL)
246 flags = NFQA_SKB_CSUMNOTREADY;
247 else if (csum_verify)
248 flags = NFQA_SKB_CSUM_NOTVERIFIED;
249
250 if (skb_is_gso(packet))
251 flags |= NFQA_SKB_GSO;
252
253 return flags ? nla_put_be32(nlskb, NFQA_SKB_INFO, htonl(flags)) : 0;
254}
255
256static int nfqnl_put_sk_uidgid(struct sk_buff *skb, struct sock *sk)
257{
258 const struct cred *cred;
259
260 if (!sk_fullsock(sk))
261 return 0;
262
263 read_lock_bh(&sk->sk_callback_lock);
264 if (sk->sk_socket && sk->sk_socket->file) {
265 cred = sk->sk_socket->file->f_cred;
266 if (nla_put_be32(skb, NFQA_UID,
267 htonl(from_kuid_munged(&init_user_ns, cred->fsuid))))
268 goto nla_put_failure;
269 if (nla_put_be32(skb, NFQA_GID,
270 htonl(from_kgid_munged(&init_user_ns, cred->fsgid))))
271 goto nla_put_failure;
272 }
273 read_unlock_bh(&sk->sk_callback_lock);
274 return 0;
275
276nla_put_failure:
277 read_unlock_bh(&sk->sk_callback_lock);
278 return -1;
279}
280
281static u32 nfqnl_get_sk_secctx(struct sk_buff *skb, char **secdata)
282{
283 u32 seclen = 0;
284#if IS_ENABLED(CONFIG_NETWORK_SECMARK)
285 if (!skb || !sk_fullsock(skb->sk))
286 return 0;
287
288 read_lock_bh(&skb->sk->sk_callback_lock);
289
290 if (skb->secmark)
291 security_secid_to_secctx(skb->secmark, secdata, &seclen);
292
293 read_unlock_bh(&skb->sk->sk_callback_lock);
294#endif
295 return seclen;
296}
297
298static u32 nfqnl_get_bridge_size(struct nf_queue_entry *entry)
299{
300 struct sk_buff *entskb = entry->skb;
301 u32 nlalen = 0;
302
303 if (entry->state.pf != PF_BRIDGE || !skb_mac_header_was_set(entskb))
304 return 0;
305
306 if (skb_vlan_tag_present(entskb))
307 nlalen += nla_total_size(nla_total_size(sizeof(__be16)) +
308 nla_total_size(sizeof(__be16)));
309
310 if (entskb->network_header > entskb->mac_header)
311 nlalen += nla_total_size((entskb->network_header -
312 entskb->mac_header));
313
314 return nlalen;
315}
316
317static int nfqnl_put_bridge(struct nf_queue_entry *entry, struct sk_buff *skb)
318{
319 struct sk_buff *entskb = entry->skb;
320
321 if (entry->state.pf != PF_BRIDGE || !skb_mac_header_was_set(entskb))
322 return 0;
323
324 if (skb_vlan_tag_present(entskb)) {
325 struct nlattr *nest;
326
327 nest = nla_nest_start(skb, NFQA_VLAN | NLA_F_NESTED);
328 if (!nest)
329 goto nla_put_failure;
330
331 if (nla_put_be16(skb, NFQA_VLAN_TCI, htons(entskb->vlan_tci)) ||
332 nla_put_be16(skb, NFQA_VLAN_PROTO, entskb->vlan_proto))
333 goto nla_put_failure;
334
335 nla_nest_end(skb, nest);
336 }
337
338 if (entskb->mac_header < entskb->network_header) {
339 int len = (int)(entskb->network_header - entskb->mac_header);
340
341 if (nla_put(skb, NFQA_L2HDR, len, skb_mac_header(entskb)))
342 goto nla_put_failure;
343 }
344
345 return 0;
346
347nla_put_failure:
348 return -1;
349}
350
351static struct sk_buff *
352nfqnl_build_packet_message(struct net *net, struct nfqnl_instance *queue,
353 struct nf_queue_entry *entry,
354 __be32 **packet_id_ptr)
355{
356 size_t size;
357 size_t data_len = 0, cap_len = 0;
358 unsigned int hlen = 0;
359 struct sk_buff *skb;
360 struct nlattr *nla;
361 struct nfqnl_msg_packet_hdr *pmsg;
362 struct nlmsghdr *nlh;
363 struct nfgenmsg *nfmsg;
364 struct sk_buff *entskb = entry->skb;
365 struct net_device *indev;
366 struct net_device *outdev;
367 struct nf_conn *ct = NULL;
368 enum ip_conntrack_info uninitialized_var(ctinfo);
369 struct nfnl_ct_hook *nfnl_ct;
370 bool csum_verify;
371 char *secdata = NULL;
372 u32 seclen = 0;
373
374 size = nlmsg_total_size(sizeof(struct nfgenmsg))
375 + nla_total_size(sizeof(struct nfqnl_msg_packet_hdr))
376 + nla_total_size(sizeof(u_int32_t)) /* ifindex */
377 + nla_total_size(sizeof(u_int32_t)) /* ifindex */
378#if IS_ENABLED(CONFIG_BRIDGE_NETFILTER)
379 + nla_total_size(sizeof(u_int32_t)) /* ifindex */
380 + nla_total_size(sizeof(u_int32_t)) /* ifindex */
381#endif
382 + nla_total_size(sizeof(u_int32_t)) /* mark */
383 + nla_total_size(sizeof(struct nfqnl_msg_packet_hw))
384 + nla_total_size(sizeof(u_int32_t)) /* skbinfo */
385 + nla_total_size(sizeof(u_int32_t)); /* cap_len */
386
387 if (entskb->tstamp)
388 size += nla_total_size(sizeof(struct nfqnl_msg_packet_timestamp));
389
390 size += nfqnl_get_bridge_size(entry);
391
392 if (entry->state.hook <= NF_INET_FORWARD ||
393 (entry->state.hook == NF_INET_POST_ROUTING && entskb->sk == NULL))
394 csum_verify = !skb_csum_unnecessary(entskb);
395 else
396 csum_verify = false;
397
398 outdev = entry->state.out;
399
400 switch ((enum nfqnl_config_mode)ACCESS_ONCE(queue->copy_mode)) {
401 case NFQNL_COPY_META:
402 case NFQNL_COPY_NONE:
403 break;
404
405 case NFQNL_COPY_PACKET:
406 if (!(queue->flags & NFQA_CFG_F_GSO) &&
407 entskb->ip_summed == CHECKSUM_PARTIAL &&
408 skb_checksum_help(entskb))
409 return NULL;
410
411 data_len = ACCESS_ONCE(queue->copy_range);
412 if (data_len > entskb->len)
413 data_len = entskb->len;
414
415 hlen = skb_zerocopy_headlen(entskb);
416 hlen = min_t(unsigned int, hlen, data_len);
417 size += sizeof(struct nlattr) + hlen;
418 cap_len = entskb->len;
419 break;
420 }
421
422 nfnl_ct = rcu_dereference(nfnl_ct_hook);
423
424 if (queue->flags & NFQA_CFG_F_CONNTRACK) {
425 if (nfnl_ct != NULL) {
426 ct = nfnl_ct->get_ct(entskb, &ctinfo);
427 if (ct != NULL)
428 size += nfnl_ct->build_size(ct);
429 }
430 }
431
432 if (queue->flags & NFQA_CFG_F_UID_GID) {
433 size += (nla_total_size(sizeof(u_int32_t)) /* uid */
434 + nla_total_size(sizeof(u_int32_t))); /* gid */
435 }
436
437 if ((queue->flags & NFQA_CFG_F_SECCTX) && entskb->sk) {
438 seclen = nfqnl_get_sk_secctx(entskb, &secdata);
439 if (seclen)
440 size += nla_total_size(seclen);
441 }
442
443 skb = alloc_skb(size, GFP_ATOMIC);
444 if (!skb) {
445 skb_tx_error(entskb);
446 return NULL;
447 }
448
449 nlh = nlmsg_put(skb, 0, 0,
450 NFNL_SUBSYS_QUEUE << 8 | NFQNL_MSG_PACKET,
451 sizeof(struct nfgenmsg), 0);
452 if (!nlh) {
453 skb_tx_error(entskb);
454 kfree_skb(skb);
455 return NULL;
456 }
457 nfmsg = nlmsg_data(nlh);
458 nfmsg->nfgen_family = entry->state.pf;
459 nfmsg->version = NFNETLINK_V0;
460 nfmsg->res_id = htons(queue->queue_num);
461
462 nla = __nla_reserve(skb, NFQA_PACKET_HDR, sizeof(*pmsg));
463 pmsg = nla_data(nla);
464 pmsg->hw_protocol = entskb->protocol;
465 pmsg->hook = entry->state.hook;
466 *packet_id_ptr = &pmsg->packet_id;
467
468 indev = entry->state.in;
469 if (indev) {
470#if !IS_ENABLED(CONFIG_BRIDGE_NETFILTER)
471 if (nla_put_be32(skb, NFQA_IFINDEX_INDEV, htonl(indev->ifindex)))
472 goto nla_put_failure;
473#else
474 if (entry->state.pf == PF_BRIDGE) {
475 /* Case 1: indev is physical input device, we need to
476 * look for bridge group (when called from
477 * netfilter_bridge) */
478 if (nla_put_be32(skb, NFQA_IFINDEX_PHYSINDEV,
479 htonl(indev->ifindex)) ||
480 /* this is the bridge group "brX" */
481 /* rcu_read_lock()ed by __nf_queue */
482 nla_put_be32(skb, NFQA_IFINDEX_INDEV,
483 htonl(br_port_get_rcu(indev)->br->dev->ifindex)))
484 goto nla_put_failure;
485 } else {
486 int physinif;
487
488 /* Case 2: indev is bridge group, we need to look for
489 * physical device (when called from ipv4) */
490 if (nla_put_be32(skb, NFQA_IFINDEX_INDEV,
491 htonl(indev->ifindex)))
492 goto nla_put_failure;
493
494 physinif = nf_bridge_get_physinif(entskb);
495 if (physinif &&
496 nla_put_be32(skb, NFQA_IFINDEX_PHYSINDEV,
497 htonl(physinif)))
498 goto nla_put_failure;
499 }
500#endif
501 }
502
503 if (outdev) {
504#if !IS_ENABLED(CONFIG_BRIDGE_NETFILTER)
505 if (nla_put_be32(skb, NFQA_IFINDEX_OUTDEV, htonl(outdev->ifindex)))
506 goto nla_put_failure;
507#else
508 if (entry->state.pf == PF_BRIDGE) {
509 /* Case 1: outdev is physical output device, we need to
510 * look for bridge group (when called from
511 * netfilter_bridge) */
512 if (nla_put_be32(skb, NFQA_IFINDEX_PHYSOUTDEV,
513 htonl(outdev->ifindex)) ||
514 /* this is the bridge group "brX" */
515 /* rcu_read_lock()ed by __nf_queue */
516 nla_put_be32(skb, NFQA_IFINDEX_OUTDEV,
517 htonl(br_port_get_rcu(outdev)->br->dev->ifindex)))
518 goto nla_put_failure;
519 } else {
520 int physoutif;
521
522 /* Case 2: outdev is bridge group, we need to look for
523 * physical output device (when called from ipv4) */
524 if (nla_put_be32(skb, NFQA_IFINDEX_OUTDEV,
525 htonl(outdev->ifindex)))
526 goto nla_put_failure;
527
528 physoutif = nf_bridge_get_physoutif(entskb);
529 if (physoutif &&
530 nla_put_be32(skb, NFQA_IFINDEX_PHYSOUTDEV,
531 htonl(physoutif)))
532 goto nla_put_failure;
533 }
534#endif
535 }
536
537 if (entskb->mark &&
538 nla_put_be32(skb, NFQA_MARK, htonl(entskb->mark)))
539 goto nla_put_failure;
540
541 if (indev && entskb->dev &&
542 entskb->mac_header != entskb->network_header) {
543 struct nfqnl_msg_packet_hw phw;
544 int len;
545
546 memset(&phw, 0, sizeof(phw));
547 len = dev_parse_header(entskb, phw.hw_addr);
548 if (len) {
549 phw.hw_addrlen = htons(len);
550 if (nla_put(skb, NFQA_HWADDR, sizeof(phw), &phw))
551 goto nla_put_failure;
552 }
553 }
554
555 if (nfqnl_put_bridge(entry, skb) < 0)
556 goto nla_put_failure;
557
558 if (entskb->tstamp) {
559 struct nfqnl_msg_packet_timestamp ts;
560 struct timespec64 kts = ktime_to_timespec64(entskb->tstamp);
561
562 ts.sec = cpu_to_be64(kts.tv_sec);
563 ts.usec = cpu_to_be64(kts.tv_nsec / NSEC_PER_USEC);
564
565 if (nla_put(skb, NFQA_TIMESTAMP, sizeof(ts), &ts))
566 goto nla_put_failure;
567 }
568
569 if ((queue->flags & NFQA_CFG_F_UID_GID) && entskb->sk &&
570 nfqnl_put_sk_uidgid(skb, entskb->sk) < 0)
571 goto nla_put_failure;
572
573 if (seclen && nla_put(skb, NFQA_SECCTX, seclen, secdata))
574 goto nla_put_failure;
575
576 if (ct && nfnl_ct->build(skb, ct, ctinfo, NFQA_CT, NFQA_CT_INFO) < 0)
577 goto nla_put_failure;
578
579 if (cap_len > data_len &&
580 nla_put_be32(skb, NFQA_CAP_LEN, htonl(cap_len)))
581 goto nla_put_failure;
582
583 if (nfqnl_put_packet_info(skb, entskb, csum_verify))
584 goto nla_put_failure;
585
586 if (data_len) {
587 struct nlattr *nla;
588
589 if (skb_tailroom(skb) < sizeof(*nla) + hlen)
590 goto nla_put_failure;
591
592 nla = (struct nlattr *)skb_put(skb, sizeof(*nla));
593 nla->nla_type = NFQA_PAYLOAD;
594 nla->nla_len = nla_attr_size(data_len);
595
596 if (skb_zerocopy(skb, entskb, data_len, hlen))
597 goto nla_put_failure;
598 }
599
600 nlh->nlmsg_len = skb->len;
601 return skb;
602
603nla_put_failure:
604 skb_tx_error(entskb);
605 kfree_skb(skb);
606 net_err_ratelimited("nf_queue: error creating packet message\n");
607 return NULL;
608}
609
610static int
611__nfqnl_enqueue_packet(struct net *net, struct nfqnl_instance *queue,
612 struct nf_queue_entry *entry)
613{
614 struct sk_buff *nskb;
615 int err = -ENOBUFS;
616 __be32 *packet_id_ptr;
617 int failopen = 0;
618
619 nskb = nfqnl_build_packet_message(net, queue, entry, &packet_id_ptr);
620 if (nskb == NULL) {
621 err = -ENOMEM;
622 goto err_out;
623 }
624 spin_lock_bh(&queue->lock);
625
626 if (queue->queue_total >= queue->queue_maxlen) {
627 if (queue->flags & NFQA_CFG_F_FAIL_OPEN) {
628 failopen = 1;
629 err = 0;
630 } else {
631 queue->queue_dropped++;
632 net_warn_ratelimited("nf_queue: full at %d entries, dropping packets(s)\n",
633 queue->queue_total);
634 }
635 goto err_out_free_nskb;
636 }
637 entry->id = ++queue->id_sequence;
638 *packet_id_ptr = htonl(entry->id);
639
640 /* nfnetlink_unicast will either free the nskb or add it to a socket */
641 err = nfnetlink_unicast(nskb, net, queue->peer_portid, MSG_DONTWAIT);
642 if (err < 0) {
643 if (queue->flags & NFQA_CFG_F_FAIL_OPEN) {
644 failopen = 1;
645 err = 0;
646 } else {
647 queue->queue_user_dropped++;
648 }
649 goto err_out_unlock;
650 }
651
652 __enqueue_entry(queue, entry);
653
654 spin_unlock_bh(&queue->lock);
655 return 0;
656
657err_out_free_nskb:
658 kfree_skb(nskb);
659err_out_unlock:
660 spin_unlock_bh(&queue->lock);
661 if (failopen)
662 nf_reinject(entry, NF_ACCEPT);
663err_out:
664 return err;
665}
666
667static struct nf_queue_entry *
668nf_queue_entry_dup(struct nf_queue_entry *e)
669{
670 struct nf_queue_entry *entry = kmemdup(e, e->size, GFP_ATOMIC);
671 if (entry)
672 nf_queue_entry_get_refs(entry);
673 return entry;
674}
675
676#if IS_ENABLED(CONFIG_BRIDGE_NETFILTER)
677/* When called from bridge netfilter, skb->data must point to MAC header
678 * before calling skb_gso_segment(). Else, original MAC header is lost
679 * and segmented skbs will be sent to wrong destination.
680 */
681static void nf_bridge_adjust_skb_data(struct sk_buff *skb)
682{
683 if (skb->nf_bridge)
684 __skb_push(skb, skb->network_header - skb->mac_header);
685}
686
687static void nf_bridge_adjust_segmented_data(struct sk_buff *skb)
688{
689 if (skb->nf_bridge)
690 __skb_pull(skb, skb->network_header - skb->mac_header);
691}
692#else
693#define nf_bridge_adjust_skb_data(s) do {} while (0)
694#define nf_bridge_adjust_segmented_data(s) do {} while (0)
695#endif
696
697static void free_entry(struct nf_queue_entry *entry)
698{
699 nf_queue_entry_release_refs(entry);
700 kfree(entry);
701}
702
703static int
704__nfqnl_enqueue_packet_gso(struct net *net, struct nfqnl_instance *queue,
705 struct sk_buff *skb, struct nf_queue_entry *entry)
706{
707 int ret = -ENOMEM;
708 struct nf_queue_entry *entry_seg;
709
710 nf_bridge_adjust_segmented_data(skb);
711
712 if (skb->next == NULL) { /* last packet, no need to copy entry */
713 struct sk_buff *gso_skb = entry->skb;
714 entry->skb = skb;
715 ret = __nfqnl_enqueue_packet(net, queue, entry);
716 if (ret)
717 entry->skb = gso_skb;
718 return ret;
719 }
720
721 skb->next = NULL;
722
723 entry_seg = nf_queue_entry_dup(entry);
724 if (entry_seg) {
725 entry_seg->skb = skb;
726 ret = __nfqnl_enqueue_packet(net, queue, entry_seg);
727 if (ret)
728 free_entry(entry_seg);
729 }
730 return ret;
731}
732
733static int
734nfqnl_enqueue_packet(struct nf_queue_entry *entry, unsigned int queuenum)
735{
736 unsigned int queued;
737 struct nfqnl_instance *queue;
738 struct sk_buff *skb, *segs;
739 int err = -ENOBUFS;
740 struct net *net = entry->state.net;
741 struct nfnl_queue_net *q = nfnl_queue_pernet(net);
742
743 /* rcu_read_lock()ed by nf_hook_thresh */
744 queue = instance_lookup(q, queuenum);
745 if (!queue)
746 return -ESRCH;
747
748 if (queue->copy_mode == NFQNL_COPY_NONE)
749 return -EINVAL;
750
751 skb = entry->skb;
752
753 switch (entry->state.pf) {
754 case NFPROTO_IPV4:
755 skb->protocol = htons(ETH_P_IP);
756 break;
757 case NFPROTO_IPV6:
758 skb->protocol = htons(ETH_P_IPV6);
759 break;
760 }
761
762 if ((queue->flags & NFQA_CFG_F_GSO) || !skb_is_gso(skb))
763 return __nfqnl_enqueue_packet(net, queue, entry);
764
765 nf_bridge_adjust_skb_data(skb);
766 segs = skb_gso_segment(skb, 0);
767 /* Does not use PTR_ERR to limit the number of error codes that can be
768 * returned by nf_queue. For instance, callers rely on -ESRCH to
769 * mean 'ignore this hook'.
770 */
771 if (IS_ERR_OR_NULL(segs))
772 goto out_err;
773 queued = 0;
774 err = 0;
775 do {
776 struct sk_buff *nskb = segs->next;
777 if (err == 0)
778 err = __nfqnl_enqueue_packet_gso(net, queue,
779 segs, entry);
780 if (err == 0)
781 queued++;
782 else
783 kfree_skb(segs);
784 segs = nskb;
785 } while (segs);
786
787 if (queued) {
788 if (err) /* some segments are already queued */
789 free_entry(entry);
790 kfree_skb(skb);
791 return 0;
792 }
793 out_err:
794 nf_bridge_adjust_segmented_data(skb);
795 return err;
796}
797
798static int
799nfqnl_mangle(void *data, int data_len, struct nf_queue_entry *e, int diff)
800{
801 struct sk_buff *nskb;
802
803 if (diff < 0) {
804 if (pskb_trim(e->skb, data_len))
805 return -ENOMEM;
806 } else if (diff > 0) {
807 if (data_len > 0xFFFF)
808 return -EINVAL;
809 if (diff > skb_tailroom(e->skb)) {
810 nskb = skb_copy_expand(e->skb, skb_headroom(e->skb),
811 diff, GFP_ATOMIC);
812 if (!nskb) {
813 printk(KERN_WARNING "nf_queue: OOM "
814 "in mangle, dropping packet\n");
815 return -ENOMEM;
816 }
817 kfree_skb(e->skb);
818 e->skb = nskb;
819 }
820 skb_put(e->skb, diff);
821 }
822 if (!skb_make_writable(e->skb, data_len))
823 return -ENOMEM;
824 skb_copy_to_linear_data(e->skb, data, data_len);
825 e->skb->ip_summed = CHECKSUM_NONE;
826 return 0;
827}
828
829static int
830nfqnl_set_mode(struct nfqnl_instance *queue,
831 unsigned char mode, unsigned int range)
832{
833 int status = 0;
834
835 spin_lock_bh(&queue->lock);
836 switch (mode) {
837 case NFQNL_COPY_NONE:
838 case NFQNL_COPY_META:
839 queue->copy_mode = mode;
840 queue->copy_range = 0;
841 break;
842
843 case NFQNL_COPY_PACKET:
844 queue->copy_mode = mode;
845 if (range == 0 || range > NFQNL_MAX_COPY_RANGE)
846 queue->copy_range = NFQNL_MAX_COPY_RANGE;
847 else
848 queue->copy_range = range;
849 break;
850
851 default:
852 status = -EINVAL;
853
854 }
855 spin_unlock_bh(&queue->lock);
856
857 return status;
858}
859
860static int
861dev_cmp(struct nf_queue_entry *entry, unsigned long ifindex)
862{
863 if (entry->state.in)
864 if (entry->state.in->ifindex == ifindex)
865 return 1;
866 if (entry->state.out)
867 if (entry->state.out->ifindex == ifindex)
868 return 1;
869#if IS_ENABLED(CONFIG_BRIDGE_NETFILTER)
870 if (entry->skb->nf_bridge) {
871 int physinif, physoutif;
872
873 physinif = nf_bridge_get_physinif(entry->skb);
874 physoutif = nf_bridge_get_physoutif(entry->skb);
875
876 if (physinif == ifindex || physoutif == ifindex)
877 return 1;
878 }
879#endif
880 return 0;
881}
882
883/* drop all packets with either indev or outdev == ifindex from all queue
884 * instances */
885static void
886nfqnl_dev_drop(struct net *net, int ifindex)
887{
888 int i;
889 struct nfnl_queue_net *q = nfnl_queue_pernet(net);
890
891 rcu_read_lock();
892
893 for (i = 0; i < INSTANCE_BUCKETS; i++) {
894 struct nfqnl_instance *inst;
895 struct hlist_head *head = &q->instance_table[i];
896
897 hlist_for_each_entry_rcu(inst, head, hlist)
898 nfqnl_flush(inst, dev_cmp, ifindex);
899 }
900
901 rcu_read_unlock();
902}
903
904static int
905nfqnl_rcv_dev_event(struct notifier_block *this,
906 unsigned long event, void *ptr)
907{
908 struct net_device *dev = netdev_notifier_info_to_dev(ptr);
909
910 /* Drop any packets associated with the downed device */
911 if (event == NETDEV_DOWN)
912 nfqnl_dev_drop(dev_net(dev), dev->ifindex);
913 return NOTIFY_DONE;
914}
915
916static struct notifier_block nfqnl_dev_notifier = {
917 .notifier_call = nfqnl_rcv_dev_event,
918};
919
920static int nf_hook_cmp(struct nf_queue_entry *entry, unsigned long entry_ptr)
921{
922 return rcu_access_pointer(entry->hook) ==
923 (struct nf_hook_entry *)entry_ptr;
924}
925
926static void nfqnl_nf_hook_drop(struct net *net,
927 const struct nf_hook_entry *hook)
928{
929 struct nfnl_queue_net *q = nfnl_queue_pernet(net);
930 int i;
931
932 rcu_read_lock();
933 for (i = 0; i < INSTANCE_BUCKETS; i++) {
934 struct nfqnl_instance *inst;
935 struct hlist_head *head = &q->instance_table[i];
936
937 hlist_for_each_entry_rcu(inst, head, hlist)
938 nfqnl_flush(inst, nf_hook_cmp, (unsigned long)hook);
939 }
940 rcu_read_unlock();
941}
942
943static int
944nfqnl_rcv_nl_event(struct notifier_block *this,
945 unsigned long event, void *ptr)
946{
947 struct netlink_notify *n = ptr;
948 struct nfnl_queue_net *q = nfnl_queue_pernet(n->net);
949
950 if (event == NETLINK_URELEASE && n->protocol == NETLINK_NETFILTER) {
951 int i;
952
953 /* destroy all instances for this portid */
954 spin_lock(&q->instances_lock);
955 for (i = 0; i < INSTANCE_BUCKETS; i++) {
956 struct hlist_node *t2;
957 struct nfqnl_instance *inst;
958 struct hlist_head *head = &q->instance_table[i];
959
960 hlist_for_each_entry_safe(inst, t2, head, hlist) {
961 if (n->portid == inst->peer_portid)
962 __instance_destroy(inst);
963 }
964 }
965 spin_unlock(&q->instances_lock);
966 }
967 return NOTIFY_DONE;
968}
969
970static struct notifier_block nfqnl_rtnl_notifier = {
971 .notifier_call = nfqnl_rcv_nl_event,
972};
973
974static const struct nla_policy nfqa_vlan_policy[NFQA_VLAN_MAX + 1] = {
975 [NFQA_VLAN_TCI] = { .type = NLA_U16},
976 [NFQA_VLAN_PROTO] = { .type = NLA_U16},
977};
978
979static const struct nla_policy nfqa_verdict_policy[NFQA_MAX+1] = {
980 [NFQA_VERDICT_HDR] = { .len = sizeof(struct nfqnl_msg_verdict_hdr) },
981 [NFQA_MARK] = { .type = NLA_U32 },
982 [NFQA_PAYLOAD] = { .type = NLA_UNSPEC },
983 [NFQA_CT] = { .type = NLA_UNSPEC },
984 [NFQA_EXP] = { .type = NLA_UNSPEC },
985 [NFQA_VLAN] = { .type = NLA_NESTED },
986};
987
988static const struct nla_policy nfqa_verdict_batch_policy[NFQA_MAX+1] = {
989 [NFQA_VERDICT_HDR] = { .len = sizeof(struct nfqnl_msg_verdict_hdr) },
990 [NFQA_MARK] = { .type = NLA_U32 },
991};
992
993static struct nfqnl_instance *
994verdict_instance_lookup(struct nfnl_queue_net *q, u16 queue_num, u32 nlportid)
995{
996 struct nfqnl_instance *queue;
997
998 queue = instance_lookup(q, queue_num);
999 if (!queue)
1000 return ERR_PTR(-ENODEV);
1001
1002 if (queue->peer_portid != nlportid)
1003 return ERR_PTR(-EPERM);
1004
1005 return queue;
1006}
1007
1008static struct nfqnl_msg_verdict_hdr*
1009verdicthdr_get(const struct nlattr * const nfqa[])
1010{
1011 struct nfqnl_msg_verdict_hdr *vhdr;
1012 unsigned int verdict;
1013
1014 if (!nfqa[NFQA_VERDICT_HDR])
1015 return NULL;
1016
1017 vhdr = nla_data(nfqa[NFQA_VERDICT_HDR]);
1018 verdict = ntohl(vhdr->verdict) & NF_VERDICT_MASK;
1019 if (verdict > NF_MAX_VERDICT || verdict == NF_STOLEN)
1020 return NULL;
1021 return vhdr;
1022}
1023
1024static int nfq_id_after(unsigned int id, unsigned int max)
1025{
1026 return (int)(id - max) > 0;
1027}
1028
1029static int nfqnl_recv_verdict_batch(struct net *net, struct sock *ctnl,
1030 struct sk_buff *skb,
1031 const struct nlmsghdr *nlh,
1032 const struct nlattr * const nfqa[])
1033{
1034 struct nfgenmsg *nfmsg = nlmsg_data(nlh);
1035 struct nf_queue_entry *entry, *tmp;
1036 unsigned int verdict, maxid;
1037 struct nfqnl_msg_verdict_hdr *vhdr;
1038 struct nfqnl_instance *queue;
1039 LIST_HEAD(batch_list);
1040 u16 queue_num = ntohs(nfmsg->res_id);
1041 struct nfnl_queue_net *q = nfnl_queue_pernet(net);
1042
1043 queue = verdict_instance_lookup(q, queue_num,
1044 NETLINK_CB(skb).portid);
1045 if (IS_ERR(queue))
1046 return PTR_ERR(queue);
1047
1048 vhdr = verdicthdr_get(nfqa);
1049 if (!vhdr)
1050 return -EINVAL;
1051
1052 verdict = ntohl(vhdr->verdict);
1053 maxid = ntohl(vhdr->id);
1054
1055 spin_lock_bh(&queue->lock);
1056
1057 list_for_each_entry_safe(entry, tmp, &queue->queue_list, list) {
1058 if (nfq_id_after(entry->id, maxid))
1059 break;
1060 __dequeue_entry(queue, entry);
1061 list_add_tail(&entry->list, &batch_list);
1062 }
1063
1064 spin_unlock_bh(&queue->lock);
1065
1066 if (list_empty(&batch_list))
1067 return -ENOENT;
1068
1069 list_for_each_entry_safe(entry, tmp, &batch_list, list) {
1070 if (nfqa[NFQA_MARK])
1071 entry->skb->mark = ntohl(nla_get_be32(nfqa[NFQA_MARK]));
1072 nf_reinject(entry, verdict);
1073 }
1074 return 0;
1075}
1076
1077static struct nf_conn *nfqnl_ct_parse(struct nfnl_ct_hook *nfnl_ct,
1078 const struct nlmsghdr *nlh,
1079 const struct nlattr * const nfqa[],
1080 struct nf_queue_entry *entry,
1081 enum ip_conntrack_info *ctinfo)
1082{
1083 struct nf_conn *ct;
1084
1085 ct = nfnl_ct->get_ct(entry->skb, ctinfo);
1086 if (ct == NULL)
1087 return NULL;
1088
1089 if (nfnl_ct->parse(nfqa[NFQA_CT], ct) < 0)
1090 return NULL;
1091
1092 if (nfqa[NFQA_EXP])
1093 nfnl_ct->attach_expect(nfqa[NFQA_EXP], ct,
1094 NETLINK_CB(entry->skb).portid,
1095 nlmsg_report(nlh));
1096 return ct;
1097}
1098
1099static int nfqa_parse_bridge(struct nf_queue_entry *entry,
1100 const struct nlattr * const nfqa[])
1101{
1102 if (nfqa[NFQA_VLAN]) {
1103 struct nlattr *tb[NFQA_VLAN_MAX + 1];
1104 int err;
1105
1106 err = nla_parse_nested(tb, NFQA_VLAN_MAX, nfqa[NFQA_VLAN],
1107 nfqa_vlan_policy);
1108 if (err < 0)
1109 return err;
1110
1111 if (!tb[NFQA_VLAN_TCI] || !tb[NFQA_VLAN_PROTO])
1112 return -EINVAL;
1113
1114 entry->skb->vlan_tci = ntohs(nla_get_be16(tb[NFQA_VLAN_TCI]));
1115 entry->skb->vlan_proto = nla_get_be16(tb[NFQA_VLAN_PROTO]);
1116 }
1117
1118 if (nfqa[NFQA_L2HDR]) {
1119 int mac_header_len = entry->skb->network_header -
1120 entry->skb->mac_header;
1121
1122 if (mac_header_len != nla_len(nfqa[NFQA_L2HDR]))
1123 return -EINVAL;
1124 else if (mac_header_len > 0)
1125 memcpy(skb_mac_header(entry->skb),
1126 nla_data(nfqa[NFQA_L2HDR]),
1127 mac_header_len);
1128 }
1129
1130 return 0;
1131}
1132
1133static int nfqnl_recv_verdict(struct net *net, struct sock *ctnl,
1134 struct sk_buff *skb,
1135 const struct nlmsghdr *nlh,
1136 const struct nlattr * const nfqa[])
1137{
1138 struct nfgenmsg *nfmsg = nlmsg_data(nlh);
1139 u_int16_t queue_num = ntohs(nfmsg->res_id);
1140 struct nfqnl_msg_verdict_hdr *vhdr;
1141 struct nfqnl_instance *queue;
1142 unsigned int verdict;
1143 struct nf_queue_entry *entry;
1144 enum ip_conntrack_info uninitialized_var(ctinfo);
1145 struct nfnl_ct_hook *nfnl_ct;
1146 struct nf_conn *ct = NULL;
1147 struct nfnl_queue_net *q = nfnl_queue_pernet(net);
1148 int err;
1149
1150 queue = verdict_instance_lookup(q, queue_num,
1151 NETLINK_CB(skb).portid);
1152 if (IS_ERR(queue))
1153 return PTR_ERR(queue);
1154
1155 vhdr = verdicthdr_get(nfqa);
1156 if (!vhdr)
1157 return -EINVAL;
1158
1159 verdict = ntohl(vhdr->verdict);
1160
1161 entry = find_dequeue_entry(queue, ntohl(vhdr->id));
1162 if (entry == NULL)
1163 return -ENOENT;
1164
1165 /* rcu lock already held from nfnl->call_rcu. */
1166 nfnl_ct = rcu_dereference(nfnl_ct_hook);
1167
1168 if (nfqa[NFQA_CT]) {
1169 if (nfnl_ct != NULL)
1170 ct = nfqnl_ct_parse(nfnl_ct, nlh, nfqa, entry, &ctinfo);
1171 }
1172
1173 if (entry->state.pf == PF_BRIDGE) {
1174 err = nfqa_parse_bridge(entry, nfqa);
1175 if (err < 0)
1176 return err;
1177 }
1178
1179 if (nfqa[NFQA_PAYLOAD]) {
1180 u16 payload_len = nla_len(nfqa[NFQA_PAYLOAD]);
1181 int diff = payload_len - entry->skb->len;
1182
1183 if (nfqnl_mangle(nla_data(nfqa[NFQA_PAYLOAD]),
1184 payload_len, entry, diff) < 0)
1185 verdict = NF_DROP;
1186
1187 if (ct && diff)
1188 nfnl_ct->seq_adjust(entry->skb, ct, ctinfo, diff);
1189 }
1190
1191 if (nfqa[NFQA_MARK])
1192 entry->skb->mark = ntohl(nla_get_be32(nfqa[NFQA_MARK]));
1193
1194 nf_reinject(entry, verdict);
1195 return 0;
1196}
1197
1198static int nfqnl_recv_unsupp(struct net *net, struct sock *ctnl,
1199 struct sk_buff *skb, const struct nlmsghdr *nlh,
1200 const struct nlattr * const nfqa[])
1201{
1202 return -ENOTSUPP;
1203}
1204
1205static const struct nla_policy nfqa_cfg_policy[NFQA_CFG_MAX+1] = {
1206 [NFQA_CFG_CMD] = { .len = sizeof(struct nfqnl_msg_config_cmd) },
1207 [NFQA_CFG_PARAMS] = { .len = sizeof(struct nfqnl_msg_config_params) },
1208};
1209
1210static const struct nf_queue_handler nfqh = {
1211 .outfn = &nfqnl_enqueue_packet,
1212 .nf_hook_drop = &nfqnl_nf_hook_drop,
1213};
1214
1215static int nfqnl_recv_config(struct net *net, struct sock *ctnl,
1216 struct sk_buff *skb, const struct nlmsghdr *nlh,
1217 const struct nlattr * const nfqa[])
1218{
1219 struct nfgenmsg *nfmsg = nlmsg_data(nlh);
1220 u_int16_t queue_num = ntohs(nfmsg->res_id);
1221 struct nfqnl_instance *queue;
1222 struct nfqnl_msg_config_cmd *cmd = NULL;
1223 struct nfnl_queue_net *q = nfnl_queue_pernet(net);
1224 __u32 flags = 0, mask = 0;
1225 int ret = 0;
1226
1227 if (nfqa[NFQA_CFG_CMD]) {
1228 cmd = nla_data(nfqa[NFQA_CFG_CMD]);
1229
1230 /* Obsolete commands without queue context */
1231 switch (cmd->command) {
1232 case NFQNL_CFG_CMD_PF_BIND: return 0;
1233 case NFQNL_CFG_CMD_PF_UNBIND: return 0;
1234 }
1235 }
1236
1237 /* Check if we support these flags in first place, dependencies should
1238 * be there too not to break atomicity.
1239 */
1240 if (nfqa[NFQA_CFG_FLAGS]) {
1241 if (!nfqa[NFQA_CFG_MASK]) {
1242 /* A mask is needed to specify which flags are being
1243 * changed.
1244 */
1245 return -EINVAL;
1246 }
1247
1248 flags = ntohl(nla_get_be32(nfqa[NFQA_CFG_FLAGS]));
1249 mask = ntohl(nla_get_be32(nfqa[NFQA_CFG_MASK]));
1250
1251 if (flags >= NFQA_CFG_F_MAX)
1252 return -EOPNOTSUPP;
1253
1254#if !IS_ENABLED(CONFIG_NETWORK_SECMARK)
1255 if (flags & mask & NFQA_CFG_F_SECCTX)
1256 return -EOPNOTSUPP;
1257#endif
1258 if ((flags & mask & NFQA_CFG_F_CONNTRACK) &&
1259 !rcu_access_pointer(nfnl_ct_hook)) {
1260#ifdef CONFIG_MODULES
1261 nfnl_unlock(NFNL_SUBSYS_QUEUE);
1262 request_module("ip_conntrack_netlink");
1263 nfnl_lock(NFNL_SUBSYS_QUEUE);
1264 if (rcu_access_pointer(nfnl_ct_hook))
1265 return -EAGAIN;
1266#endif
1267 return -EOPNOTSUPP;
1268 }
1269 }
1270
1271 rcu_read_lock();
1272 queue = instance_lookup(q, queue_num);
1273 if (queue && queue->peer_portid != NETLINK_CB(skb).portid) {
1274 ret = -EPERM;
1275 goto err_out_unlock;
1276 }
1277
1278 if (cmd != NULL) {
1279 switch (cmd->command) {
1280 case NFQNL_CFG_CMD_BIND:
1281 if (queue) {
1282 ret = -EBUSY;
1283 goto err_out_unlock;
1284 }
1285 queue = instance_create(q, queue_num,
1286 NETLINK_CB(skb).portid);
1287 if (IS_ERR(queue)) {
1288 ret = PTR_ERR(queue);
1289 goto err_out_unlock;
1290 }
1291 break;
1292 case NFQNL_CFG_CMD_UNBIND:
1293 if (!queue) {
1294 ret = -ENODEV;
1295 goto err_out_unlock;
1296 }
1297 instance_destroy(q, queue);
1298 goto err_out_unlock;
1299 case NFQNL_CFG_CMD_PF_BIND:
1300 case NFQNL_CFG_CMD_PF_UNBIND:
1301 break;
1302 default:
1303 ret = -ENOTSUPP;
1304 goto err_out_unlock;
1305 }
1306 }
1307
1308 if (!queue) {
1309 ret = -ENODEV;
1310 goto err_out_unlock;
1311 }
1312
1313 if (nfqa[NFQA_CFG_PARAMS]) {
1314 struct nfqnl_msg_config_params *params =
1315 nla_data(nfqa[NFQA_CFG_PARAMS]);
1316
1317 nfqnl_set_mode(queue, params->copy_mode,
1318 ntohl(params->copy_range));
1319 }
1320
1321 if (nfqa[NFQA_CFG_QUEUE_MAXLEN]) {
1322 __be32 *queue_maxlen = nla_data(nfqa[NFQA_CFG_QUEUE_MAXLEN]);
1323
1324 spin_lock_bh(&queue->lock);
1325 queue->queue_maxlen = ntohl(*queue_maxlen);
1326 spin_unlock_bh(&queue->lock);
1327 }
1328
1329 if (nfqa[NFQA_CFG_FLAGS]) {
1330 spin_lock_bh(&queue->lock);
1331 queue->flags &= ~mask;
1332 queue->flags |= flags & mask;
1333 spin_unlock_bh(&queue->lock);
1334 }
1335
1336err_out_unlock:
1337 rcu_read_unlock();
1338 return ret;
1339}
1340
1341static const struct nfnl_callback nfqnl_cb[NFQNL_MSG_MAX] = {
1342 [NFQNL_MSG_PACKET] = { .call_rcu = nfqnl_recv_unsupp,
1343 .attr_count = NFQA_MAX, },
1344 [NFQNL_MSG_VERDICT] = { .call_rcu = nfqnl_recv_verdict,
1345 .attr_count = NFQA_MAX,
1346 .policy = nfqa_verdict_policy },
1347 [NFQNL_MSG_CONFIG] = { .call = nfqnl_recv_config,
1348 .attr_count = NFQA_CFG_MAX,
1349 .policy = nfqa_cfg_policy },
1350 [NFQNL_MSG_VERDICT_BATCH]={ .call_rcu = nfqnl_recv_verdict_batch,
1351 .attr_count = NFQA_MAX,
1352 .policy = nfqa_verdict_batch_policy },
1353};
1354
1355static const struct nfnetlink_subsystem nfqnl_subsys = {
1356 .name = "nf_queue",
1357 .subsys_id = NFNL_SUBSYS_QUEUE,
1358 .cb_count = NFQNL_MSG_MAX,
1359 .cb = nfqnl_cb,
1360};
1361
1362#ifdef CONFIG_PROC_FS
1363struct iter_state {
1364 struct seq_net_private p;
1365 unsigned int bucket;
1366};
1367
1368static struct hlist_node *get_first(struct seq_file *seq)
1369{
1370 struct iter_state *st = seq->private;
1371 struct net *net;
1372 struct nfnl_queue_net *q;
1373
1374 if (!st)
1375 return NULL;
1376
1377 net = seq_file_net(seq);
1378 q = nfnl_queue_pernet(net);
1379 for (st->bucket = 0; st->bucket < INSTANCE_BUCKETS; st->bucket++) {
1380 if (!hlist_empty(&q->instance_table[st->bucket]))
1381 return q->instance_table[st->bucket].first;
1382 }
1383 return NULL;
1384}
1385
1386static struct hlist_node *get_next(struct seq_file *seq, struct hlist_node *h)
1387{
1388 struct iter_state *st = seq->private;
1389 struct net *net = seq_file_net(seq);
1390
1391 h = h->next;
1392 while (!h) {
1393 struct nfnl_queue_net *q;
1394
1395 if (++st->bucket >= INSTANCE_BUCKETS)
1396 return NULL;
1397
1398 q = nfnl_queue_pernet(net);
1399 h = q->instance_table[st->bucket].first;
1400 }
1401 return h;
1402}
1403
1404static struct hlist_node *get_idx(struct seq_file *seq, loff_t pos)
1405{
1406 struct hlist_node *head;
1407 head = get_first(seq);
1408
1409 if (head)
1410 while (pos && (head = get_next(seq, head)))
1411 pos--;
1412 return pos ? NULL : head;
1413}
1414
1415static void *seq_start(struct seq_file *s, loff_t *pos)
1416 __acquires(nfnl_queue_pernet(seq_file_net(s))->instances_lock)
1417{
1418 spin_lock(&nfnl_queue_pernet(seq_file_net(s))->instances_lock);
1419 return get_idx(s, *pos);
1420}
1421
1422static void *seq_next(struct seq_file *s, void *v, loff_t *pos)
1423{
1424 (*pos)++;
1425 return get_next(s, v);
1426}
1427
1428static void seq_stop(struct seq_file *s, void *v)
1429 __releases(nfnl_queue_pernet(seq_file_net(s))->instances_lock)
1430{
1431 spin_unlock(&nfnl_queue_pernet(seq_file_net(s))->instances_lock);
1432}
1433
1434static int seq_show(struct seq_file *s, void *v)
1435{
1436 const struct nfqnl_instance *inst = v;
1437
1438 seq_printf(s, "%5u %6u %5u %1u %5u %5u %5u %8u %2d\n",
1439 inst->queue_num,
1440 inst->peer_portid, inst->queue_total,
1441 inst->copy_mode, inst->copy_range,
1442 inst->queue_dropped, inst->queue_user_dropped,
1443 inst->id_sequence, 1);
1444 return 0;
1445}
1446
1447static const struct seq_operations nfqnl_seq_ops = {
1448 .start = seq_start,
1449 .next = seq_next,
1450 .stop = seq_stop,
1451 .show = seq_show,
1452};
1453
1454static int nfqnl_open(struct inode *inode, struct file *file)
1455{
1456 return seq_open_net(inode, file, &nfqnl_seq_ops,
1457 sizeof(struct iter_state));
1458}
1459
1460static const struct file_operations nfqnl_file_ops = {
1461 .owner = THIS_MODULE,
1462 .open = nfqnl_open,
1463 .read = seq_read,
1464 .llseek = seq_lseek,
1465 .release = seq_release_net,
1466};
1467
1468#endif /* PROC_FS */
1469
1470static int __net_init nfnl_queue_net_init(struct net *net)
1471{
1472 unsigned int i;
1473 struct nfnl_queue_net *q = nfnl_queue_pernet(net);
1474
1475 for (i = 0; i < INSTANCE_BUCKETS; i++)
1476 INIT_HLIST_HEAD(&q->instance_table[i]);
1477
1478 spin_lock_init(&q->instances_lock);
1479
1480#ifdef CONFIG_PROC_FS
1481 if (!proc_create("nfnetlink_queue", 0440,
1482 net->nf.proc_netfilter, &nfqnl_file_ops))
1483 return -ENOMEM;
1484#endif
1485 nf_register_queue_handler(net, &nfqh);
1486 return 0;
1487}
1488
1489static void __net_exit nfnl_queue_net_exit(struct net *net)
1490{
1491 nf_unregister_queue_handler(net);
1492#ifdef CONFIG_PROC_FS
1493 remove_proc_entry("nfnetlink_queue", net->nf.proc_netfilter);
1494#endif
1495}
1496
1497static void nfnl_queue_net_exit_batch(struct list_head *net_exit_list)
1498{
1499 synchronize_rcu();
1500}
1501
1502static struct pernet_operations nfnl_queue_net_ops = {
1503 .init = nfnl_queue_net_init,
1504 .exit = nfnl_queue_net_exit,
1505 .exit_batch = nfnl_queue_net_exit_batch,
1506 .id = &nfnl_queue_net_id,
1507 .size = sizeof(struct nfnl_queue_net),
1508};
1509
1510static int __init nfnetlink_queue_init(void)
1511{
1512 int status;
1513
1514 status = register_pernet_subsys(&nfnl_queue_net_ops);
1515 if (status < 0) {
1516 pr_err("nf_queue: failed to register pernet ops\n");
1517 goto out;
1518 }
1519
1520 netlink_register_notifier(&nfqnl_rtnl_notifier);
1521 status = nfnetlink_subsys_register(&nfqnl_subsys);
1522 if (status < 0) {
1523 pr_err("nf_queue: failed to create netlink socket\n");
1524 goto cleanup_netlink_notifier;
1525 }
1526
1527 status = register_netdevice_notifier(&nfqnl_dev_notifier);
1528 if (status < 0) {
1529 pr_err("nf_queue: failed to register netdevice notifier\n");
1530 goto cleanup_netlink_subsys;
1531 }
1532
1533 return status;
1534
1535cleanup_netlink_subsys:
1536 nfnetlink_subsys_unregister(&nfqnl_subsys);
1537cleanup_netlink_notifier:
1538 netlink_unregister_notifier(&nfqnl_rtnl_notifier);
1539 unregister_pernet_subsys(&nfnl_queue_net_ops);
1540out:
1541 return status;
1542}
1543
1544static void __exit nfnetlink_queue_fini(void)
1545{
1546 unregister_netdevice_notifier(&nfqnl_dev_notifier);
1547 nfnetlink_subsys_unregister(&nfqnl_subsys);
1548 netlink_unregister_notifier(&nfqnl_rtnl_notifier);
1549 unregister_pernet_subsys(&nfnl_queue_net_ops);
1550
1551 rcu_barrier(); /* Wait for completion of call_rcu()'s */
1552}
1553
1554MODULE_DESCRIPTION("netfilter packet queue handler");
1555MODULE_AUTHOR("Harald Welte <laforge@netfilter.org>");
1556MODULE_LICENSE("GPL");
1557MODULE_ALIAS_NFNL_SUBSYS(NFNL_SUBSYS_QUEUE);
1558
1559module_init(nfnetlink_queue_init);
1560module_exit(nfnetlink_queue_fini);
1// SPDX-License-Identifier: GPL-2.0-only
2/*
3 * This is a module which is used for queueing packets and communicating with
4 * userspace via nfnetlink.
5 *
6 * (C) 2005 by Harald Welte <laforge@netfilter.org>
7 * (C) 2007 by Patrick McHardy <kaber@trash.net>
8 *
9 * Based on the old ipv4-only ip_queue.c:
10 * (C) 2000-2002 James Morris <jmorris@intercode.com.au>
11 * (C) 2003-2005 Netfilter Core Team <coreteam@netfilter.org>
12 */
13
14#define pr_fmt(fmt) KBUILD_MODNAME ": " fmt
15
16#include <linux/module.h>
17#include <linux/skbuff.h>
18#include <linux/init.h>
19#include <linux/spinlock.h>
20#include <linux/slab.h>
21#include <linux/notifier.h>
22#include <linux/netdevice.h>
23#include <linux/netfilter.h>
24#include <linux/proc_fs.h>
25#include <linux/netfilter_ipv4.h>
26#include <linux/netfilter_ipv6.h>
27#include <linux/netfilter_bridge.h>
28#include <linux/netfilter/nfnetlink.h>
29#include <linux/netfilter/nfnetlink_queue.h>
30#include <linux/netfilter/nf_conntrack_common.h>
31#include <linux/list.h>
32#include <linux/cgroup-defs.h>
33#include <net/gso.h>
34#include <net/sock.h>
35#include <net/tcp_states.h>
36#include <net/netfilter/nf_queue.h>
37#include <net/netns/generic.h>
38
39#include <linux/atomic.h>
40
41#if IS_ENABLED(CONFIG_BRIDGE_NETFILTER)
42#include "../bridge/br_private.h"
43#endif
44
45#if IS_ENABLED(CONFIG_NF_CONNTRACK)
46#include <net/netfilter/nf_conntrack.h>
47#endif
48
49#define NFQNL_QMAX_DEFAULT 1024
50
51/* We're using struct nlattr which has 16bit nla_len. Note that nla_len
52 * includes the header length. Thus, the maximum packet length that we
53 * support is 65531 bytes. We send truncated packets if the specified length
54 * is larger than that. Userspace can check for presence of NFQA_CAP_LEN
55 * attribute to detect truncation.
56 */
57#define NFQNL_MAX_COPY_RANGE (0xffff - NLA_HDRLEN)
58
59struct nfqnl_instance {
60 struct hlist_node hlist; /* global list of queues */
61 struct rcu_head rcu;
62
63 u32 peer_portid;
64 unsigned int queue_maxlen;
65 unsigned int copy_range;
66 unsigned int queue_dropped;
67 unsigned int queue_user_dropped;
68
69
70 u_int16_t queue_num; /* number of this queue */
71 u_int8_t copy_mode;
72 u_int32_t flags; /* Set using NFQA_CFG_FLAGS */
73/*
74 * Following fields are dirtied for each queued packet,
75 * keep them in same cache line if possible.
76 */
77 spinlock_t lock ____cacheline_aligned_in_smp;
78 unsigned int queue_total;
79 unsigned int id_sequence; /* 'sequence' of pkt ids */
80 struct list_head queue_list; /* packets in queue */
81};
82
83typedef int (*nfqnl_cmpfn)(struct nf_queue_entry *, unsigned long);
84
85static unsigned int nfnl_queue_net_id __read_mostly;
86
87#define INSTANCE_BUCKETS 16
88struct nfnl_queue_net {
89 spinlock_t instances_lock;
90 struct hlist_head instance_table[INSTANCE_BUCKETS];
91};
92
93static struct nfnl_queue_net *nfnl_queue_pernet(struct net *net)
94{
95 return net_generic(net, nfnl_queue_net_id);
96}
97
98static inline u_int8_t instance_hashfn(u_int16_t queue_num)
99{
100 return ((queue_num >> 8) ^ queue_num) % INSTANCE_BUCKETS;
101}
102
103static struct nfqnl_instance *
104instance_lookup(struct nfnl_queue_net *q, u_int16_t queue_num)
105{
106 struct hlist_head *head;
107 struct nfqnl_instance *inst;
108
109 head = &q->instance_table[instance_hashfn(queue_num)];
110 hlist_for_each_entry_rcu(inst, head, hlist) {
111 if (inst->queue_num == queue_num)
112 return inst;
113 }
114 return NULL;
115}
116
117static struct nfqnl_instance *
118instance_create(struct nfnl_queue_net *q, u_int16_t queue_num, u32 portid)
119{
120 struct nfqnl_instance *inst;
121 unsigned int h;
122 int err;
123
124 spin_lock(&q->instances_lock);
125 if (instance_lookup(q, queue_num)) {
126 err = -EEXIST;
127 goto out_unlock;
128 }
129
130 inst = kzalloc(sizeof(*inst), GFP_ATOMIC);
131 if (!inst) {
132 err = -ENOMEM;
133 goto out_unlock;
134 }
135
136 inst->queue_num = queue_num;
137 inst->peer_portid = portid;
138 inst->queue_maxlen = NFQNL_QMAX_DEFAULT;
139 inst->copy_range = NFQNL_MAX_COPY_RANGE;
140 inst->copy_mode = NFQNL_COPY_NONE;
141 spin_lock_init(&inst->lock);
142 INIT_LIST_HEAD(&inst->queue_list);
143
144 if (!try_module_get(THIS_MODULE)) {
145 err = -EAGAIN;
146 goto out_free;
147 }
148
149 h = instance_hashfn(queue_num);
150 hlist_add_head_rcu(&inst->hlist, &q->instance_table[h]);
151
152 spin_unlock(&q->instances_lock);
153
154 return inst;
155
156out_free:
157 kfree(inst);
158out_unlock:
159 spin_unlock(&q->instances_lock);
160 return ERR_PTR(err);
161}
162
163static void nfqnl_flush(struct nfqnl_instance *queue, nfqnl_cmpfn cmpfn,
164 unsigned long data);
165
166static void
167instance_destroy_rcu(struct rcu_head *head)
168{
169 struct nfqnl_instance *inst = container_of(head, struct nfqnl_instance,
170 rcu);
171
172 nfqnl_flush(inst, NULL, 0);
173 kfree(inst);
174 module_put(THIS_MODULE);
175}
176
177static void
178__instance_destroy(struct nfqnl_instance *inst)
179{
180 hlist_del_rcu(&inst->hlist);
181 call_rcu(&inst->rcu, instance_destroy_rcu);
182}
183
184static void
185instance_destroy(struct nfnl_queue_net *q, struct nfqnl_instance *inst)
186{
187 spin_lock(&q->instances_lock);
188 __instance_destroy(inst);
189 spin_unlock(&q->instances_lock);
190}
191
192static inline void
193__enqueue_entry(struct nfqnl_instance *queue, struct nf_queue_entry *entry)
194{
195 list_add_tail(&entry->list, &queue->queue_list);
196 queue->queue_total++;
197}
198
199static void
200__dequeue_entry(struct nfqnl_instance *queue, struct nf_queue_entry *entry)
201{
202 list_del(&entry->list);
203 queue->queue_total--;
204}
205
206static struct nf_queue_entry *
207find_dequeue_entry(struct nfqnl_instance *queue, unsigned int id)
208{
209 struct nf_queue_entry *entry = NULL, *i;
210
211 spin_lock_bh(&queue->lock);
212
213 list_for_each_entry(i, &queue->queue_list, list) {
214 if (i->id == id) {
215 entry = i;
216 break;
217 }
218 }
219
220 if (entry)
221 __dequeue_entry(queue, entry);
222
223 spin_unlock_bh(&queue->lock);
224
225 return entry;
226}
227
228static void nfqnl_reinject(struct nf_queue_entry *entry, unsigned int verdict)
229{
230 const struct nf_ct_hook *ct_hook;
231
232 if (verdict == NF_ACCEPT ||
233 verdict == NF_REPEAT ||
234 verdict == NF_STOP) {
235 unsigned int ct_verdict = verdict;
236
237 rcu_read_lock();
238 ct_hook = rcu_dereference(nf_ct_hook);
239 if (ct_hook)
240 ct_verdict = ct_hook->update(entry->state.net, entry->skb);
241 rcu_read_unlock();
242
243 switch (ct_verdict & NF_VERDICT_MASK) {
244 case NF_ACCEPT:
245 /* follow userspace verdict, could be REPEAT */
246 break;
247 case NF_STOLEN:
248 nf_queue_entry_free(entry);
249 return;
250 default:
251 verdict = ct_verdict & NF_VERDICT_MASK;
252 break;
253 }
254 }
255 nf_reinject(entry, verdict);
256}
257
258static void
259nfqnl_flush(struct nfqnl_instance *queue, nfqnl_cmpfn cmpfn, unsigned long data)
260{
261 struct nf_queue_entry *entry, *next;
262
263 spin_lock_bh(&queue->lock);
264 list_for_each_entry_safe(entry, next, &queue->queue_list, list) {
265 if (!cmpfn || cmpfn(entry, data)) {
266 list_del(&entry->list);
267 queue->queue_total--;
268 nfqnl_reinject(entry, NF_DROP);
269 }
270 }
271 spin_unlock_bh(&queue->lock);
272}
273
274static int
275nfqnl_put_packet_info(struct sk_buff *nlskb, struct sk_buff *packet,
276 bool csum_verify)
277{
278 __u32 flags = 0;
279
280 if (packet->ip_summed == CHECKSUM_PARTIAL)
281 flags = NFQA_SKB_CSUMNOTREADY;
282 else if (csum_verify)
283 flags = NFQA_SKB_CSUM_NOTVERIFIED;
284
285 if (skb_is_gso(packet))
286 flags |= NFQA_SKB_GSO;
287
288 return flags ? nla_put_be32(nlskb, NFQA_SKB_INFO, htonl(flags)) : 0;
289}
290
291static int nfqnl_put_sk_uidgid(struct sk_buff *skb, struct sock *sk)
292{
293 const struct cred *cred;
294
295 if (!sk_fullsock(sk))
296 return 0;
297
298 read_lock_bh(&sk->sk_callback_lock);
299 if (sk->sk_socket && sk->sk_socket->file) {
300 cred = sk->sk_socket->file->f_cred;
301 if (nla_put_be32(skb, NFQA_UID,
302 htonl(from_kuid_munged(&init_user_ns, cred->fsuid))))
303 goto nla_put_failure;
304 if (nla_put_be32(skb, NFQA_GID,
305 htonl(from_kgid_munged(&init_user_ns, cred->fsgid))))
306 goto nla_put_failure;
307 }
308 read_unlock_bh(&sk->sk_callback_lock);
309 return 0;
310
311nla_put_failure:
312 read_unlock_bh(&sk->sk_callback_lock);
313 return -1;
314}
315
316static int nfqnl_put_sk_classid(struct sk_buff *skb, struct sock *sk)
317{
318#if IS_ENABLED(CONFIG_CGROUP_NET_CLASSID)
319 if (sk && sk_fullsock(sk)) {
320 u32 classid = sock_cgroup_classid(&sk->sk_cgrp_data);
321
322 if (classid && nla_put_be32(skb, NFQA_CGROUP_CLASSID, htonl(classid)))
323 return -1;
324 }
325#endif
326 return 0;
327}
328
329static u32 nfqnl_get_sk_secctx(struct sk_buff *skb, char **secdata)
330{
331 u32 seclen = 0;
332#if IS_ENABLED(CONFIG_NETWORK_SECMARK)
333 if (!skb || !sk_fullsock(skb->sk))
334 return 0;
335
336 read_lock_bh(&skb->sk->sk_callback_lock);
337
338 if (skb->secmark)
339 security_secid_to_secctx(skb->secmark, secdata, &seclen);
340
341 read_unlock_bh(&skb->sk->sk_callback_lock);
342#endif
343 return seclen;
344}
345
346static u32 nfqnl_get_bridge_size(struct nf_queue_entry *entry)
347{
348 struct sk_buff *entskb = entry->skb;
349 u32 nlalen = 0;
350
351 if (entry->state.pf != PF_BRIDGE || !skb_mac_header_was_set(entskb))
352 return 0;
353
354 if (skb_vlan_tag_present(entskb))
355 nlalen += nla_total_size(nla_total_size(sizeof(__be16)) +
356 nla_total_size(sizeof(__be16)));
357
358 if (entskb->network_header > entskb->mac_header)
359 nlalen += nla_total_size((entskb->network_header -
360 entskb->mac_header));
361
362 return nlalen;
363}
364
365static int nfqnl_put_bridge(struct nf_queue_entry *entry, struct sk_buff *skb)
366{
367 struct sk_buff *entskb = entry->skb;
368
369 if (entry->state.pf != PF_BRIDGE || !skb_mac_header_was_set(entskb))
370 return 0;
371
372 if (skb_vlan_tag_present(entskb)) {
373 struct nlattr *nest;
374
375 nest = nla_nest_start(skb, NFQA_VLAN);
376 if (!nest)
377 goto nla_put_failure;
378
379 if (nla_put_be16(skb, NFQA_VLAN_TCI, htons(entskb->vlan_tci)) ||
380 nla_put_be16(skb, NFQA_VLAN_PROTO, entskb->vlan_proto))
381 goto nla_put_failure;
382
383 nla_nest_end(skb, nest);
384 }
385
386 if (entskb->mac_header < entskb->network_header) {
387 int len = (int)(entskb->network_header - entskb->mac_header);
388
389 if (nla_put(skb, NFQA_L2HDR, len, skb_mac_header(entskb)))
390 goto nla_put_failure;
391 }
392
393 return 0;
394
395nla_put_failure:
396 return -1;
397}
398
399static struct sk_buff *
400nfqnl_build_packet_message(struct net *net, struct nfqnl_instance *queue,
401 struct nf_queue_entry *entry,
402 __be32 **packet_id_ptr)
403{
404 size_t size;
405 size_t data_len = 0, cap_len = 0;
406 unsigned int hlen = 0;
407 struct sk_buff *skb;
408 struct nlattr *nla;
409 struct nfqnl_msg_packet_hdr *pmsg;
410 struct nlmsghdr *nlh;
411 struct sk_buff *entskb = entry->skb;
412 struct net_device *indev;
413 struct net_device *outdev;
414 struct nf_conn *ct = NULL;
415 enum ip_conntrack_info ctinfo = 0;
416 const struct nfnl_ct_hook *nfnl_ct;
417 bool csum_verify;
418 char *secdata = NULL;
419 u32 seclen = 0;
420 ktime_t tstamp;
421
422 size = nlmsg_total_size(sizeof(struct nfgenmsg))
423 + nla_total_size(sizeof(struct nfqnl_msg_packet_hdr))
424 + nla_total_size(sizeof(u_int32_t)) /* ifindex */
425 + nla_total_size(sizeof(u_int32_t)) /* ifindex */
426#if IS_ENABLED(CONFIG_BRIDGE_NETFILTER)
427 + nla_total_size(sizeof(u_int32_t)) /* ifindex */
428 + nla_total_size(sizeof(u_int32_t)) /* ifindex */
429#endif
430 + nla_total_size(sizeof(u_int32_t)) /* mark */
431 + nla_total_size(sizeof(u_int32_t)) /* priority */
432 + nla_total_size(sizeof(struct nfqnl_msg_packet_hw))
433 + nla_total_size(sizeof(u_int32_t)) /* skbinfo */
434#if IS_ENABLED(CONFIG_CGROUP_NET_CLASSID)
435 + nla_total_size(sizeof(u_int32_t)) /* classid */
436#endif
437 + nla_total_size(sizeof(u_int32_t)); /* cap_len */
438
439 tstamp = skb_tstamp_cond(entskb, false);
440 if (tstamp)
441 size += nla_total_size(sizeof(struct nfqnl_msg_packet_timestamp));
442
443 size += nfqnl_get_bridge_size(entry);
444
445 if (entry->state.hook <= NF_INET_FORWARD ||
446 (entry->state.hook == NF_INET_POST_ROUTING && entskb->sk == NULL))
447 csum_verify = !skb_csum_unnecessary(entskb);
448 else
449 csum_verify = false;
450
451 outdev = entry->state.out;
452
453 switch ((enum nfqnl_config_mode)READ_ONCE(queue->copy_mode)) {
454 case NFQNL_COPY_META:
455 case NFQNL_COPY_NONE:
456 break;
457
458 case NFQNL_COPY_PACKET:
459 if (!(queue->flags & NFQA_CFG_F_GSO) &&
460 entskb->ip_summed == CHECKSUM_PARTIAL &&
461 skb_checksum_help(entskb))
462 return NULL;
463
464 data_len = READ_ONCE(queue->copy_range);
465 if (data_len > entskb->len)
466 data_len = entskb->len;
467
468 hlen = skb_zerocopy_headlen(entskb);
469 hlen = min_t(unsigned int, hlen, data_len);
470 size += sizeof(struct nlattr) + hlen;
471 cap_len = entskb->len;
472 break;
473 }
474
475 nfnl_ct = rcu_dereference(nfnl_ct_hook);
476
477#if IS_ENABLED(CONFIG_NF_CONNTRACK)
478 if (queue->flags & NFQA_CFG_F_CONNTRACK) {
479 if (nfnl_ct != NULL) {
480 ct = nf_ct_get(entskb, &ctinfo);
481 if (ct != NULL)
482 size += nfnl_ct->build_size(ct);
483 }
484 }
485#endif
486
487 if (queue->flags & NFQA_CFG_F_UID_GID) {
488 size += (nla_total_size(sizeof(u_int32_t)) /* uid */
489 + nla_total_size(sizeof(u_int32_t))); /* gid */
490 }
491
492 if ((queue->flags & NFQA_CFG_F_SECCTX) && entskb->sk) {
493 seclen = nfqnl_get_sk_secctx(entskb, &secdata);
494 if (seclen)
495 size += nla_total_size(seclen);
496 }
497
498 skb = alloc_skb(size, GFP_ATOMIC);
499 if (!skb) {
500 skb_tx_error(entskb);
501 goto nlmsg_failure;
502 }
503
504 nlh = nfnl_msg_put(skb, 0, 0,
505 nfnl_msg_type(NFNL_SUBSYS_QUEUE, NFQNL_MSG_PACKET),
506 0, entry->state.pf, NFNETLINK_V0,
507 htons(queue->queue_num));
508 if (!nlh) {
509 skb_tx_error(entskb);
510 kfree_skb(skb);
511 goto nlmsg_failure;
512 }
513
514 nla = __nla_reserve(skb, NFQA_PACKET_HDR, sizeof(*pmsg));
515 pmsg = nla_data(nla);
516 pmsg->hw_protocol = entskb->protocol;
517 pmsg->hook = entry->state.hook;
518 *packet_id_ptr = &pmsg->packet_id;
519
520 indev = entry->state.in;
521 if (indev) {
522#if !IS_ENABLED(CONFIG_BRIDGE_NETFILTER)
523 if (nla_put_be32(skb, NFQA_IFINDEX_INDEV, htonl(indev->ifindex)))
524 goto nla_put_failure;
525#else
526 if (entry->state.pf == PF_BRIDGE) {
527 /* Case 1: indev is physical input device, we need to
528 * look for bridge group (when called from
529 * netfilter_bridge) */
530 if (nla_put_be32(skb, NFQA_IFINDEX_PHYSINDEV,
531 htonl(indev->ifindex)) ||
532 /* this is the bridge group "brX" */
533 /* rcu_read_lock()ed by __nf_queue */
534 nla_put_be32(skb, NFQA_IFINDEX_INDEV,
535 htonl(br_port_get_rcu(indev)->br->dev->ifindex)))
536 goto nla_put_failure;
537 } else {
538 int physinif;
539
540 /* Case 2: indev is bridge group, we need to look for
541 * physical device (when called from ipv4) */
542 if (nla_put_be32(skb, NFQA_IFINDEX_INDEV,
543 htonl(indev->ifindex)))
544 goto nla_put_failure;
545
546 physinif = nf_bridge_get_physinif(entskb);
547 if (physinif &&
548 nla_put_be32(skb, NFQA_IFINDEX_PHYSINDEV,
549 htonl(physinif)))
550 goto nla_put_failure;
551 }
552#endif
553 }
554
555 if (outdev) {
556#if !IS_ENABLED(CONFIG_BRIDGE_NETFILTER)
557 if (nla_put_be32(skb, NFQA_IFINDEX_OUTDEV, htonl(outdev->ifindex)))
558 goto nla_put_failure;
559#else
560 if (entry->state.pf == PF_BRIDGE) {
561 /* Case 1: outdev is physical output device, we need to
562 * look for bridge group (when called from
563 * netfilter_bridge) */
564 if (nla_put_be32(skb, NFQA_IFINDEX_PHYSOUTDEV,
565 htonl(outdev->ifindex)) ||
566 /* this is the bridge group "brX" */
567 /* rcu_read_lock()ed by __nf_queue */
568 nla_put_be32(skb, NFQA_IFINDEX_OUTDEV,
569 htonl(br_port_get_rcu(outdev)->br->dev->ifindex)))
570 goto nla_put_failure;
571 } else {
572 int physoutif;
573
574 /* Case 2: outdev is bridge group, we need to look for
575 * physical output device (when called from ipv4) */
576 if (nla_put_be32(skb, NFQA_IFINDEX_OUTDEV,
577 htonl(outdev->ifindex)))
578 goto nla_put_failure;
579
580 physoutif = nf_bridge_get_physoutif(entskb);
581 if (physoutif &&
582 nla_put_be32(skb, NFQA_IFINDEX_PHYSOUTDEV,
583 htonl(physoutif)))
584 goto nla_put_failure;
585 }
586#endif
587 }
588
589 if (entskb->mark &&
590 nla_put_be32(skb, NFQA_MARK, htonl(entskb->mark)))
591 goto nla_put_failure;
592
593 if (entskb->priority &&
594 nla_put_be32(skb, NFQA_PRIORITY, htonl(entskb->priority)))
595 goto nla_put_failure;
596
597 if (indev && entskb->dev &&
598 skb_mac_header_was_set(entskb) &&
599 skb_mac_header_len(entskb) != 0) {
600 struct nfqnl_msg_packet_hw phw;
601 int len;
602
603 memset(&phw, 0, sizeof(phw));
604 len = dev_parse_header(entskb, phw.hw_addr);
605 if (len) {
606 phw.hw_addrlen = htons(len);
607 if (nla_put(skb, NFQA_HWADDR, sizeof(phw), &phw))
608 goto nla_put_failure;
609 }
610 }
611
612 if (nfqnl_put_bridge(entry, skb) < 0)
613 goto nla_put_failure;
614
615 if (entry->state.hook <= NF_INET_FORWARD && tstamp) {
616 struct nfqnl_msg_packet_timestamp ts;
617 struct timespec64 kts = ktime_to_timespec64(tstamp);
618
619 ts.sec = cpu_to_be64(kts.tv_sec);
620 ts.usec = cpu_to_be64(kts.tv_nsec / NSEC_PER_USEC);
621
622 if (nla_put(skb, NFQA_TIMESTAMP, sizeof(ts), &ts))
623 goto nla_put_failure;
624 }
625
626 if ((queue->flags & NFQA_CFG_F_UID_GID) && entskb->sk &&
627 nfqnl_put_sk_uidgid(skb, entskb->sk) < 0)
628 goto nla_put_failure;
629
630 if (nfqnl_put_sk_classid(skb, entskb->sk) < 0)
631 goto nla_put_failure;
632
633 if (seclen && nla_put(skb, NFQA_SECCTX, seclen, secdata))
634 goto nla_put_failure;
635
636 if (ct && nfnl_ct->build(skb, ct, ctinfo, NFQA_CT, NFQA_CT_INFO) < 0)
637 goto nla_put_failure;
638
639 if (cap_len > data_len &&
640 nla_put_be32(skb, NFQA_CAP_LEN, htonl(cap_len)))
641 goto nla_put_failure;
642
643 if (nfqnl_put_packet_info(skb, entskb, csum_verify))
644 goto nla_put_failure;
645
646 if (data_len) {
647 struct nlattr *nla;
648
649 if (skb_tailroom(skb) < sizeof(*nla) + hlen)
650 goto nla_put_failure;
651
652 nla = skb_put(skb, sizeof(*nla));
653 nla->nla_type = NFQA_PAYLOAD;
654 nla->nla_len = nla_attr_size(data_len);
655
656 if (skb_zerocopy(skb, entskb, data_len, hlen))
657 goto nla_put_failure;
658 }
659
660 nlh->nlmsg_len = skb->len;
661 if (seclen)
662 security_release_secctx(secdata, seclen);
663 return skb;
664
665nla_put_failure:
666 skb_tx_error(entskb);
667 kfree_skb(skb);
668 net_err_ratelimited("nf_queue: error creating packet message\n");
669nlmsg_failure:
670 if (seclen)
671 security_release_secctx(secdata, seclen);
672 return NULL;
673}
674
675static bool nf_ct_drop_unconfirmed(const struct nf_queue_entry *entry)
676{
677#if IS_ENABLED(CONFIG_NF_CONNTRACK)
678 static const unsigned long flags = IPS_CONFIRMED | IPS_DYING;
679 const struct nf_conn *ct = (void *)skb_nfct(entry->skb);
680
681 if (ct && ((ct->status & flags) == IPS_DYING))
682 return true;
683#endif
684 return false;
685}
686
687static int
688__nfqnl_enqueue_packet(struct net *net, struct nfqnl_instance *queue,
689 struct nf_queue_entry *entry)
690{
691 struct sk_buff *nskb;
692 int err = -ENOBUFS;
693 __be32 *packet_id_ptr;
694 int failopen = 0;
695
696 nskb = nfqnl_build_packet_message(net, queue, entry, &packet_id_ptr);
697 if (nskb == NULL) {
698 err = -ENOMEM;
699 goto err_out;
700 }
701 spin_lock_bh(&queue->lock);
702
703 if (nf_ct_drop_unconfirmed(entry))
704 goto err_out_free_nskb;
705
706 if (queue->queue_total >= queue->queue_maxlen) {
707 if (queue->flags & NFQA_CFG_F_FAIL_OPEN) {
708 failopen = 1;
709 err = 0;
710 } else {
711 queue->queue_dropped++;
712 net_warn_ratelimited("nf_queue: full at %d entries, dropping packets(s)\n",
713 queue->queue_total);
714 }
715 goto err_out_free_nskb;
716 }
717 entry->id = ++queue->id_sequence;
718 *packet_id_ptr = htonl(entry->id);
719
720 /* nfnetlink_unicast will either free the nskb or add it to a socket */
721 err = nfnetlink_unicast(nskb, net, queue->peer_portid);
722 if (err < 0) {
723 if (queue->flags & NFQA_CFG_F_FAIL_OPEN) {
724 failopen = 1;
725 err = 0;
726 } else {
727 queue->queue_user_dropped++;
728 }
729 goto err_out_unlock;
730 }
731
732 __enqueue_entry(queue, entry);
733
734 spin_unlock_bh(&queue->lock);
735 return 0;
736
737err_out_free_nskb:
738 kfree_skb(nskb);
739err_out_unlock:
740 spin_unlock_bh(&queue->lock);
741 if (failopen)
742 nfqnl_reinject(entry, NF_ACCEPT);
743err_out:
744 return err;
745}
746
747static struct nf_queue_entry *
748nf_queue_entry_dup(struct nf_queue_entry *e)
749{
750 struct nf_queue_entry *entry = kmemdup(e, e->size, GFP_ATOMIC);
751
752 if (!entry)
753 return NULL;
754
755 if (nf_queue_entry_get_refs(entry))
756 return entry;
757
758 kfree(entry);
759 return NULL;
760}
761
762#if IS_ENABLED(CONFIG_BRIDGE_NETFILTER)
763/* When called from bridge netfilter, skb->data must point to MAC header
764 * before calling skb_gso_segment(). Else, original MAC header is lost
765 * and segmented skbs will be sent to wrong destination.
766 */
767static void nf_bridge_adjust_skb_data(struct sk_buff *skb)
768{
769 if (nf_bridge_info_get(skb))
770 __skb_push(skb, skb->network_header - skb->mac_header);
771}
772
773static void nf_bridge_adjust_segmented_data(struct sk_buff *skb)
774{
775 if (nf_bridge_info_get(skb))
776 __skb_pull(skb, skb->network_header - skb->mac_header);
777}
778#else
779#define nf_bridge_adjust_skb_data(s) do {} while (0)
780#define nf_bridge_adjust_segmented_data(s) do {} while (0)
781#endif
782
783static int
784__nfqnl_enqueue_packet_gso(struct net *net, struct nfqnl_instance *queue,
785 struct sk_buff *skb, struct nf_queue_entry *entry)
786{
787 int ret = -ENOMEM;
788 struct nf_queue_entry *entry_seg;
789
790 nf_bridge_adjust_segmented_data(skb);
791
792 if (skb->next == NULL) { /* last packet, no need to copy entry */
793 struct sk_buff *gso_skb = entry->skb;
794 entry->skb = skb;
795 ret = __nfqnl_enqueue_packet(net, queue, entry);
796 if (ret)
797 entry->skb = gso_skb;
798 return ret;
799 }
800
801 skb_mark_not_on_list(skb);
802
803 entry_seg = nf_queue_entry_dup(entry);
804 if (entry_seg) {
805 entry_seg->skb = skb;
806 ret = __nfqnl_enqueue_packet(net, queue, entry_seg);
807 if (ret)
808 nf_queue_entry_free(entry_seg);
809 }
810 return ret;
811}
812
813static int
814nfqnl_enqueue_packet(struct nf_queue_entry *entry, unsigned int queuenum)
815{
816 unsigned int queued;
817 struct nfqnl_instance *queue;
818 struct sk_buff *skb, *segs, *nskb;
819 int err = -ENOBUFS;
820 struct net *net = entry->state.net;
821 struct nfnl_queue_net *q = nfnl_queue_pernet(net);
822
823 /* rcu_read_lock()ed by nf_hook_thresh */
824 queue = instance_lookup(q, queuenum);
825 if (!queue)
826 return -ESRCH;
827
828 if (queue->copy_mode == NFQNL_COPY_NONE)
829 return -EINVAL;
830
831 skb = entry->skb;
832
833 switch (entry->state.pf) {
834 case NFPROTO_IPV4:
835 skb->protocol = htons(ETH_P_IP);
836 break;
837 case NFPROTO_IPV6:
838 skb->protocol = htons(ETH_P_IPV6);
839 break;
840 }
841
842 if ((queue->flags & NFQA_CFG_F_GSO) || !skb_is_gso(skb))
843 return __nfqnl_enqueue_packet(net, queue, entry);
844
845 nf_bridge_adjust_skb_data(skb);
846 segs = skb_gso_segment(skb, 0);
847 /* Does not use PTR_ERR to limit the number of error codes that can be
848 * returned by nf_queue. For instance, callers rely on -ESRCH to
849 * mean 'ignore this hook'.
850 */
851 if (IS_ERR_OR_NULL(segs))
852 goto out_err;
853 queued = 0;
854 err = 0;
855 skb_list_walk_safe(segs, segs, nskb) {
856 if (err == 0)
857 err = __nfqnl_enqueue_packet_gso(net, queue,
858 segs, entry);
859 if (err == 0)
860 queued++;
861 else
862 kfree_skb(segs);
863 }
864
865 if (queued) {
866 if (err) /* some segments are already queued */
867 nf_queue_entry_free(entry);
868 kfree_skb(skb);
869 return 0;
870 }
871 out_err:
872 nf_bridge_adjust_segmented_data(skb);
873 return err;
874}
875
876static int
877nfqnl_mangle(void *data, unsigned int data_len, struct nf_queue_entry *e, int diff)
878{
879 struct sk_buff *nskb;
880
881 if (diff < 0) {
882 unsigned int min_len = skb_transport_offset(e->skb);
883
884 if (data_len < min_len)
885 return -EINVAL;
886
887 if (pskb_trim(e->skb, data_len))
888 return -ENOMEM;
889 } else if (diff > 0) {
890 if (data_len > 0xFFFF)
891 return -EINVAL;
892 if (diff > skb_tailroom(e->skb)) {
893 nskb = skb_copy_expand(e->skb, skb_headroom(e->skb),
894 diff, GFP_ATOMIC);
895 if (!nskb)
896 return -ENOMEM;
897 kfree_skb(e->skb);
898 e->skb = nskb;
899 }
900 skb_put(e->skb, diff);
901 }
902 if (skb_ensure_writable(e->skb, data_len))
903 return -ENOMEM;
904 skb_copy_to_linear_data(e->skb, data, data_len);
905 e->skb->ip_summed = CHECKSUM_NONE;
906 return 0;
907}
908
909static int
910nfqnl_set_mode(struct nfqnl_instance *queue,
911 unsigned char mode, unsigned int range)
912{
913 int status = 0;
914
915 spin_lock_bh(&queue->lock);
916 switch (mode) {
917 case NFQNL_COPY_NONE:
918 case NFQNL_COPY_META:
919 queue->copy_mode = mode;
920 queue->copy_range = 0;
921 break;
922
923 case NFQNL_COPY_PACKET:
924 queue->copy_mode = mode;
925 if (range == 0 || range > NFQNL_MAX_COPY_RANGE)
926 queue->copy_range = NFQNL_MAX_COPY_RANGE;
927 else
928 queue->copy_range = range;
929 break;
930
931 default:
932 status = -EINVAL;
933
934 }
935 spin_unlock_bh(&queue->lock);
936
937 return status;
938}
939
940static int
941dev_cmp(struct nf_queue_entry *entry, unsigned long ifindex)
942{
943#if IS_ENABLED(CONFIG_BRIDGE_NETFILTER)
944 int physinif, physoutif;
945
946 physinif = nf_bridge_get_physinif(entry->skb);
947 physoutif = nf_bridge_get_physoutif(entry->skb);
948
949 if (physinif == ifindex || physoutif == ifindex)
950 return 1;
951#endif
952 if (entry->state.in)
953 if (entry->state.in->ifindex == ifindex)
954 return 1;
955 if (entry->state.out)
956 if (entry->state.out->ifindex == ifindex)
957 return 1;
958
959 return 0;
960}
961
962/* drop all packets with either indev or outdev == ifindex from all queue
963 * instances */
964static void
965nfqnl_dev_drop(struct net *net, int ifindex)
966{
967 int i;
968 struct nfnl_queue_net *q = nfnl_queue_pernet(net);
969
970 rcu_read_lock();
971
972 for (i = 0; i < INSTANCE_BUCKETS; i++) {
973 struct nfqnl_instance *inst;
974 struct hlist_head *head = &q->instance_table[i];
975
976 hlist_for_each_entry_rcu(inst, head, hlist)
977 nfqnl_flush(inst, dev_cmp, ifindex);
978 }
979
980 rcu_read_unlock();
981}
982
983static int
984nfqnl_rcv_dev_event(struct notifier_block *this,
985 unsigned long event, void *ptr)
986{
987 struct net_device *dev = netdev_notifier_info_to_dev(ptr);
988
989 /* Drop any packets associated with the downed device */
990 if (event == NETDEV_DOWN)
991 nfqnl_dev_drop(dev_net(dev), dev->ifindex);
992 return NOTIFY_DONE;
993}
994
995static struct notifier_block nfqnl_dev_notifier = {
996 .notifier_call = nfqnl_rcv_dev_event,
997};
998
999static void nfqnl_nf_hook_drop(struct net *net)
1000{
1001 struct nfnl_queue_net *q = nfnl_queue_pernet(net);
1002 int i;
1003
1004 /* This function is also called on net namespace error unwind,
1005 * when pernet_ops->init() failed and ->exit() functions of the
1006 * previous pernet_ops gets called.
1007 *
1008 * This may result in a call to nfqnl_nf_hook_drop() before
1009 * struct nfnl_queue_net was allocated.
1010 */
1011 if (!q)
1012 return;
1013
1014 for (i = 0; i < INSTANCE_BUCKETS; i++) {
1015 struct nfqnl_instance *inst;
1016 struct hlist_head *head = &q->instance_table[i];
1017
1018 hlist_for_each_entry_rcu(inst, head, hlist)
1019 nfqnl_flush(inst, NULL, 0);
1020 }
1021}
1022
1023static int
1024nfqnl_rcv_nl_event(struct notifier_block *this,
1025 unsigned long event, void *ptr)
1026{
1027 struct netlink_notify *n = ptr;
1028 struct nfnl_queue_net *q = nfnl_queue_pernet(n->net);
1029
1030 if (event == NETLINK_URELEASE && n->protocol == NETLINK_NETFILTER) {
1031 int i;
1032
1033 /* destroy all instances for this portid */
1034 spin_lock(&q->instances_lock);
1035 for (i = 0; i < INSTANCE_BUCKETS; i++) {
1036 struct hlist_node *t2;
1037 struct nfqnl_instance *inst;
1038 struct hlist_head *head = &q->instance_table[i];
1039
1040 hlist_for_each_entry_safe(inst, t2, head, hlist) {
1041 if (n->portid == inst->peer_portid)
1042 __instance_destroy(inst);
1043 }
1044 }
1045 spin_unlock(&q->instances_lock);
1046 }
1047 return NOTIFY_DONE;
1048}
1049
1050static struct notifier_block nfqnl_rtnl_notifier = {
1051 .notifier_call = nfqnl_rcv_nl_event,
1052};
1053
1054static const struct nla_policy nfqa_vlan_policy[NFQA_VLAN_MAX + 1] = {
1055 [NFQA_VLAN_TCI] = { .type = NLA_U16},
1056 [NFQA_VLAN_PROTO] = { .type = NLA_U16},
1057};
1058
1059static const struct nla_policy nfqa_verdict_policy[NFQA_MAX+1] = {
1060 [NFQA_VERDICT_HDR] = { .len = sizeof(struct nfqnl_msg_verdict_hdr) },
1061 [NFQA_MARK] = { .type = NLA_U32 },
1062 [NFQA_PAYLOAD] = { .type = NLA_UNSPEC },
1063 [NFQA_CT] = { .type = NLA_UNSPEC },
1064 [NFQA_EXP] = { .type = NLA_UNSPEC },
1065 [NFQA_VLAN] = { .type = NLA_NESTED },
1066 [NFQA_PRIORITY] = { .type = NLA_U32 },
1067};
1068
1069static const struct nla_policy nfqa_verdict_batch_policy[NFQA_MAX+1] = {
1070 [NFQA_VERDICT_HDR] = { .len = sizeof(struct nfqnl_msg_verdict_hdr) },
1071 [NFQA_MARK] = { .type = NLA_U32 },
1072 [NFQA_PRIORITY] = { .type = NLA_U32 },
1073};
1074
1075static struct nfqnl_instance *
1076verdict_instance_lookup(struct nfnl_queue_net *q, u16 queue_num, u32 nlportid)
1077{
1078 struct nfqnl_instance *queue;
1079
1080 queue = instance_lookup(q, queue_num);
1081 if (!queue)
1082 return ERR_PTR(-ENODEV);
1083
1084 if (queue->peer_portid != nlportid)
1085 return ERR_PTR(-EPERM);
1086
1087 return queue;
1088}
1089
1090static struct nfqnl_msg_verdict_hdr*
1091verdicthdr_get(const struct nlattr * const nfqa[])
1092{
1093 struct nfqnl_msg_verdict_hdr *vhdr;
1094 unsigned int verdict;
1095
1096 if (!nfqa[NFQA_VERDICT_HDR])
1097 return NULL;
1098
1099 vhdr = nla_data(nfqa[NFQA_VERDICT_HDR]);
1100 verdict = ntohl(vhdr->verdict) & NF_VERDICT_MASK;
1101 if (verdict > NF_MAX_VERDICT || verdict == NF_STOLEN)
1102 return NULL;
1103 return vhdr;
1104}
1105
1106static int nfq_id_after(unsigned int id, unsigned int max)
1107{
1108 return (int)(id - max) > 0;
1109}
1110
1111static int nfqnl_recv_verdict_batch(struct sk_buff *skb,
1112 const struct nfnl_info *info,
1113 const struct nlattr * const nfqa[])
1114{
1115 struct nfnl_queue_net *q = nfnl_queue_pernet(info->net);
1116 u16 queue_num = ntohs(info->nfmsg->res_id);
1117 struct nf_queue_entry *entry, *tmp;
1118 struct nfqnl_msg_verdict_hdr *vhdr;
1119 struct nfqnl_instance *queue;
1120 unsigned int verdict, maxid;
1121 LIST_HEAD(batch_list);
1122
1123 queue = verdict_instance_lookup(q, queue_num,
1124 NETLINK_CB(skb).portid);
1125 if (IS_ERR(queue))
1126 return PTR_ERR(queue);
1127
1128 vhdr = verdicthdr_get(nfqa);
1129 if (!vhdr)
1130 return -EINVAL;
1131
1132 verdict = ntohl(vhdr->verdict);
1133 maxid = ntohl(vhdr->id);
1134
1135 spin_lock_bh(&queue->lock);
1136
1137 list_for_each_entry_safe(entry, tmp, &queue->queue_list, list) {
1138 if (nfq_id_after(entry->id, maxid))
1139 break;
1140 __dequeue_entry(queue, entry);
1141 list_add_tail(&entry->list, &batch_list);
1142 }
1143
1144 spin_unlock_bh(&queue->lock);
1145
1146 if (list_empty(&batch_list))
1147 return -ENOENT;
1148
1149 list_for_each_entry_safe(entry, tmp, &batch_list, list) {
1150 if (nfqa[NFQA_MARK])
1151 entry->skb->mark = ntohl(nla_get_be32(nfqa[NFQA_MARK]));
1152
1153 if (nfqa[NFQA_PRIORITY])
1154 entry->skb->priority = ntohl(nla_get_be32(nfqa[NFQA_PRIORITY]));
1155
1156 nfqnl_reinject(entry, verdict);
1157 }
1158 return 0;
1159}
1160
1161static struct nf_conn *nfqnl_ct_parse(const struct nfnl_ct_hook *nfnl_ct,
1162 const struct nlmsghdr *nlh,
1163 const struct nlattr * const nfqa[],
1164 struct nf_queue_entry *entry,
1165 enum ip_conntrack_info *ctinfo)
1166{
1167#if IS_ENABLED(CONFIG_NF_CONNTRACK)
1168 struct nf_conn *ct;
1169
1170 ct = nf_ct_get(entry->skb, ctinfo);
1171 if (ct == NULL)
1172 return NULL;
1173
1174 if (nfnl_ct->parse(nfqa[NFQA_CT], ct) < 0)
1175 return NULL;
1176
1177 if (nfqa[NFQA_EXP])
1178 nfnl_ct->attach_expect(nfqa[NFQA_EXP], ct,
1179 NETLINK_CB(entry->skb).portid,
1180 nlmsg_report(nlh));
1181 return ct;
1182#else
1183 return NULL;
1184#endif
1185}
1186
1187static int nfqa_parse_bridge(struct nf_queue_entry *entry,
1188 const struct nlattr * const nfqa[])
1189{
1190 if (nfqa[NFQA_VLAN]) {
1191 struct nlattr *tb[NFQA_VLAN_MAX + 1];
1192 int err;
1193
1194 err = nla_parse_nested_deprecated(tb, NFQA_VLAN_MAX,
1195 nfqa[NFQA_VLAN],
1196 nfqa_vlan_policy, NULL);
1197 if (err < 0)
1198 return err;
1199
1200 if (!tb[NFQA_VLAN_TCI] || !tb[NFQA_VLAN_PROTO])
1201 return -EINVAL;
1202
1203 __vlan_hwaccel_put_tag(entry->skb,
1204 nla_get_be16(tb[NFQA_VLAN_PROTO]),
1205 ntohs(nla_get_be16(tb[NFQA_VLAN_TCI])));
1206 }
1207
1208 if (nfqa[NFQA_L2HDR]) {
1209 int mac_header_len = entry->skb->network_header -
1210 entry->skb->mac_header;
1211
1212 if (mac_header_len != nla_len(nfqa[NFQA_L2HDR]))
1213 return -EINVAL;
1214 else if (mac_header_len > 0)
1215 memcpy(skb_mac_header(entry->skb),
1216 nla_data(nfqa[NFQA_L2HDR]),
1217 mac_header_len);
1218 }
1219
1220 return 0;
1221}
1222
1223static int nfqnl_recv_verdict(struct sk_buff *skb, const struct nfnl_info *info,
1224 const struct nlattr * const nfqa[])
1225{
1226 struct nfnl_queue_net *q = nfnl_queue_pernet(info->net);
1227 u_int16_t queue_num = ntohs(info->nfmsg->res_id);
1228 const struct nfnl_ct_hook *nfnl_ct;
1229 struct nfqnl_msg_verdict_hdr *vhdr;
1230 enum ip_conntrack_info ctinfo;
1231 struct nfqnl_instance *queue;
1232 struct nf_queue_entry *entry;
1233 struct nf_conn *ct = NULL;
1234 unsigned int verdict;
1235 int err;
1236
1237 queue = verdict_instance_lookup(q, queue_num,
1238 NETLINK_CB(skb).portid);
1239 if (IS_ERR(queue))
1240 return PTR_ERR(queue);
1241
1242 vhdr = verdicthdr_get(nfqa);
1243 if (!vhdr)
1244 return -EINVAL;
1245
1246 verdict = ntohl(vhdr->verdict);
1247
1248 entry = find_dequeue_entry(queue, ntohl(vhdr->id));
1249 if (entry == NULL)
1250 return -ENOENT;
1251
1252 /* rcu lock already held from nfnl->call_rcu. */
1253 nfnl_ct = rcu_dereference(nfnl_ct_hook);
1254
1255 if (nfqa[NFQA_CT]) {
1256 if (nfnl_ct != NULL)
1257 ct = nfqnl_ct_parse(nfnl_ct, info->nlh, nfqa, entry,
1258 &ctinfo);
1259 }
1260
1261 if (entry->state.pf == PF_BRIDGE) {
1262 err = nfqa_parse_bridge(entry, nfqa);
1263 if (err < 0)
1264 return err;
1265 }
1266
1267 if (nfqa[NFQA_PAYLOAD]) {
1268 u16 payload_len = nla_len(nfqa[NFQA_PAYLOAD]);
1269 int diff = payload_len - entry->skb->len;
1270
1271 if (nfqnl_mangle(nla_data(nfqa[NFQA_PAYLOAD]),
1272 payload_len, entry, diff) < 0)
1273 verdict = NF_DROP;
1274
1275 if (ct && diff)
1276 nfnl_ct->seq_adjust(entry->skb, ct, ctinfo, diff);
1277 }
1278
1279 if (nfqa[NFQA_MARK])
1280 entry->skb->mark = ntohl(nla_get_be32(nfqa[NFQA_MARK]));
1281
1282 if (nfqa[NFQA_PRIORITY])
1283 entry->skb->priority = ntohl(nla_get_be32(nfqa[NFQA_PRIORITY]));
1284
1285 nfqnl_reinject(entry, verdict);
1286 return 0;
1287}
1288
1289static int nfqnl_recv_unsupp(struct sk_buff *skb, const struct nfnl_info *info,
1290 const struct nlattr * const cda[])
1291{
1292 return -ENOTSUPP;
1293}
1294
1295static const struct nla_policy nfqa_cfg_policy[NFQA_CFG_MAX+1] = {
1296 [NFQA_CFG_CMD] = { .len = sizeof(struct nfqnl_msg_config_cmd) },
1297 [NFQA_CFG_PARAMS] = { .len = sizeof(struct nfqnl_msg_config_params) },
1298 [NFQA_CFG_QUEUE_MAXLEN] = { .type = NLA_U32 },
1299 [NFQA_CFG_MASK] = { .type = NLA_U32 },
1300 [NFQA_CFG_FLAGS] = { .type = NLA_U32 },
1301};
1302
1303static const struct nf_queue_handler nfqh = {
1304 .outfn = nfqnl_enqueue_packet,
1305 .nf_hook_drop = nfqnl_nf_hook_drop,
1306};
1307
1308static int nfqnl_recv_config(struct sk_buff *skb, const struct nfnl_info *info,
1309 const struct nlattr * const nfqa[])
1310{
1311 struct nfnl_queue_net *q = nfnl_queue_pernet(info->net);
1312 u_int16_t queue_num = ntohs(info->nfmsg->res_id);
1313 struct nfqnl_msg_config_cmd *cmd = NULL;
1314 struct nfqnl_instance *queue;
1315 __u32 flags = 0, mask = 0;
1316 int ret = 0;
1317
1318 if (nfqa[NFQA_CFG_CMD]) {
1319 cmd = nla_data(nfqa[NFQA_CFG_CMD]);
1320
1321 /* Obsolete commands without queue context */
1322 switch (cmd->command) {
1323 case NFQNL_CFG_CMD_PF_BIND: return 0;
1324 case NFQNL_CFG_CMD_PF_UNBIND: return 0;
1325 }
1326 }
1327
1328 /* Check if we support these flags in first place, dependencies should
1329 * be there too not to break atomicity.
1330 */
1331 if (nfqa[NFQA_CFG_FLAGS]) {
1332 if (!nfqa[NFQA_CFG_MASK]) {
1333 /* A mask is needed to specify which flags are being
1334 * changed.
1335 */
1336 return -EINVAL;
1337 }
1338
1339 flags = ntohl(nla_get_be32(nfqa[NFQA_CFG_FLAGS]));
1340 mask = ntohl(nla_get_be32(nfqa[NFQA_CFG_MASK]));
1341
1342 if (flags >= NFQA_CFG_F_MAX)
1343 return -EOPNOTSUPP;
1344
1345#if !IS_ENABLED(CONFIG_NETWORK_SECMARK)
1346 if (flags & mask & NFQA_CFG_F_SECCTX)
1347 return -EOPNOTSUPP;
1348#endif
1349 if ((flags & mask & NFQA_CFG_F_CONNTRACK) &&
1350 !rcu_access_pointer(nfnl_ct_hook)) {
1351#ifdef CONFIG_MODULES
1352 nfnl_unlock(NFNL_SUBSYS_QUEUE);
1353 request_module("ip_conntrack_netlink");
1354 nfnl_lock(NFNL_SUBSYS_QUEUE);
1355 if (rcu_access_pointer(nfnl_ct_hook))
1356 return -EAGAIN;
1357#endif
1358 return -EOPNOTSUPP;
1359 }
1360 }
1361
1362 rcu_read_lock();
1363 queue = instance_lookup(q, queue_num);
1364 if (queue && queue->peer_portid != NETLINK_CB(skb).portid) {
1365 ret = -EPERM;
1366 goto err_out_unlock;
1367 }
1368
1369 if (cmd != NULL) {
1370 switch (cmd->command) {
1371 case NFQNL_CFG_CMD_BIND:
1372 if (queue) {
1373 ret = -EBUSY;
1374 goto err_out_unlock;
1375 }
1376 queue = instance_create(q, queue_num,
1377 NETLINK_CB(skb).portid);
1378 if (IS_ERR(queue)) {
1379 ret = PTR_ERR(queue);
1380 goto err_out_unlock;
1381 }
1382 break;
1383 case NFQNL_CFG_CMD_UNBIND:
1384 if (!queue) {
1385 ret = -ENODEV;
1386 goto err_out_unlock;
1387 }
1388 instance_destroy(q, queue);
1389 goto err_out_unlock;
1390 case NFQNL_CFG_CMD_PF_BIND:
1391 case NFQNL_CFG_CMD_PF_UNBIND:
1392 break;
1393 default:
1394 ret = -ENOTSUPP;
1395 goto err_out_unlock;
1396 }
1397 }
1398
1399 if (!queue) {
1400 ret = -ENODEV;
1401 goto err_out_unlock;
1402 }
1403
1404 if (nfqa[NFQA_CFG_PARAMS]) {
1405 struct nfqnl_msg_config_params *params =
1406 nla_data(nfqa[NFQA_CFG_PARAMS]);
1407
1408 nfqnl_set_mode(queue, params->copy_mode,
1409 ntohl(params->copy_range));
1410 }
1411
1412 if (nfqa[NFQA_CFG_QUEUE_MAXLEN]) {
1413 __be32 *queue_maxlen = nla_data(nfqa[NFQA_CFG_QUEUE_MAXLEN]);
1414
1415 spin_lock_bh(&queue->lock);
1416 queue->queue_maxlen = ntohl(*queue_maxlen);
1417 spin_unlock_bh(&queue->lock);
1418 }
1419
1420 if (nfqa[NFQA_CFG_FLAGS]) {
1421 spin_lock_bh(&queue->lock);
1422 queue->flags &= ~mask;
1423 queue->flags |= flags & mask;
1424 spin_unlock_bh(&queue->lock);
1425 }
1426
1427err_out_unlock:
1428 rcu_read_unlock();
1429 return ret;
1430}
1431
1432static const struct nfnl_callback nfqnl_cb[NFQNL_MSG_MAX] = {
1433 [NFQNL_MSG_PACKET] = {
1434 .call = nfqnl_recv_unsupp,
1435 .type = NFNL_CB_RCU,
1436 .attr_count = NFQA_MAX,
1437 },
1438 [NFQNL_MSG_VERDICT] = {
1439 .call = nfqnl_recv_verdict,
1440 .type = NFNL_CB_RCU,
1441 .attr_count = NFQA_MAX,
1442 .policy = nfqa_verdict_policy
1443 },
1444 [NFQNL_MSG_CONFIG] = {
1445 .call = nfqnl_recv_config,
1446 .type = NFNL_CB_MUTEX,
1447 .attr_count = NFQA_CFG_MAX,
1448 .policy = nfqa_cfg_policy
1449 },
1450 [NFQNL_MSG_VERDICT_BATCH] = {
1451 .call = nfqnl_recv_verdict_batch,
1452 .type = NFNL_CB_RCU,
1453 .attr_count = NFQA_MAX,
1454 .policy = nfqa_verdict_batch_policy
1455 },
1456};
1457
1458static const struct nfnetlink_subsystem nfqnl_subsys = {
1459 .name = "nf_queue",
1460 .subsys_id = NFNL_SUBSYS_QUEUE,
1461 .cb_count = NFQNL_MSG_MAX,
1462 .cb = nfqnl_cb,
1463};
1464
1465#ifdef CONFIG_PROC_FS
1466struct iter_state {
1467 struct seq_net_private p;
1468 unsigned int bucket;
1469};
1470
1471static struct hlist_node *get_first(struct seq_file *seq)
1472{
1473 struct iter_state *st = seq->private;
1474 struct net *net;
1475 struct nfnl_queue_net *q;
1476
1477 if (!st)
1478 return NULL;
1479
1480 net = seq_file_net(seq);
1481 q = nfnl_queue_pernet(net);
1482 for (st->bucket = 0; st->bucket < INSTANCE_BUCKETS; st->bucket++) {
1483 if (!hlist_empty(&q->instance_table[st->bucket]))
1484 return q->instance_table[st->bucket].first;
1485 }
1486 return NULL;
1487}
1488
1489static struct hlist_node *get_next(struct seq_file *seq, struct hlist_node *h)
1490{
1491 struct iter_state *st = seq->private;
1492 struct net *net = seq_file_net(seq);
1493
1494 h = h->next;
1495 while (!h) {
1496 struct nfnl_queue_net *q;
1497
1498 if (++st->bucket >= INSTANCE_BUCKETS)
1499 return NULL;
1500
1501 q = nfnl_queue_pernet(net);
1502 h = q->instance_table[st->bucket].first;
1503 }
1504 return h;
1505}
1506
1507static struct hlist_node *get_idx(struct seq_file *seq, loff_t pos)
1508{
1509 struct hlist_node *head;
1510 head = get_first(seq);
1511
1512 if (head)
1513 while (pos && (head = get_next(seq, head)))
1514 pos--;
1515 return pos ? NULL : head;
1516}
1517
1518static void *seq_start(struct seq_file *s, loff_t *pos)
1519 __acquires(nfnl_queue_pernet(seq_file_net(s))->instances_lock)
1520{
1521 spin_lock(&nfnl_queue_pernet(seq_file_net(s))->instances_lock);
1522 return get_idx(s, *pos);
1523}
1524
1525static void *seq_next(struct seq_file *s, void *v, loff_t *pos)
1526{
1527 (*pos)++;
1528 return get_next(s, v);
1529}
1530
1531static void seq_stop(struct seq_file *s, void *v)
1532 __releases(nfnl_queue_pernet(seq_file_net(s))->instances_lock)
1533{
1534 spin_unlock(&nfnl_queue_pernet(seq_file_net(s))->instances_lock);
1535}
1536
1537static int seq_show(struct seq_file *s, void *v)
1538{
1539 const struct nfqnl_instance *inst = v;
1540
1541 seq_printf(s, "%5u %6u %5u %1u %5u %5u %5u %8u %2d\n",
1542 inst->queue_num,
1543 inst->peer_portid, inst->queue_total,
1544 inst->copy_mode, inst->copy_range,
1545 inst->queue_dropped, inst->queue_user_dropped,
1546 inst->id_sequence, 1);
1547 return 0;
1548}
1549
1550static const struct seq_operations nfqnl_seq_ops = {
1551 .start = seq_start,
1552 .next = seq_next,
1553 .stop = seq_stop,
1554 .show = seq_show,
1555};
1556#endif /* PROC_FS */
1557
1558static int __net_init nfnl_queue_net_init(struct net *net)
1559{
1560 unsigned int i;
1561 struct nfnl_queue_net *q = nfnl_queue_pernet(net);
1562
1563 for (i = 0; i < INSTANCE_BUCKETS; i++)
1564 INIT_HLIST_HEAD(&q->instance_table[i]);
1565
1566 spin_lock_init(&q->instances_lock);
1567
1568#ifdef CONFIG_PROC_FS
1569 if (!proc_create_net("nfnetlink_queue", 0440, net->nf.proc_netfilter,
1570 &nfqnl_seq_ops, sizeof(struct iter_state)))
1571 return -ENOMEM;
1572#endif
1573 return 0;
1574}
1575
1576static void __net_exit nfnl_queue_net_exit(struct net *net)
1577{
1578 struct nfnl_queue_net *q = nfnl_queue_pernet(net);
1579 unsigned int i;
1580
1581#ifdef CONFIG_PROC_FS
1582 remove_proc_entry("nfnetlink_queue", net->nf.proc_netfilter);
1583#endif
1584 for (i = 0; i < INSTANCE_BUCKETS; i++)
1585 WARN_ON_ONCE(!hlist_empty(&q->instance_table[i]));
1586}
1587
1588static struct pernet_operations nfnl_queue_net_ops = {
1589 .init = nfnl_queue_net_init,
1590 .exit = nfnl_queue_net_exit,
1591 .id = &nfnl_queue_net_id,
1592 .size = sizeof(struct nfnl_queue_net),
1593};
1594
1595static int __init nfnetlink_queue_init(void)
1596{
1597 int status;
1598
1599 status = register_pernet_subsys(&nfnl_queue_net_ops);
1600 if (status < 0) {
1601 pr_err("failed to register pernet ops\n");
1602 goto out;
1603 }
1604
1605 netlink_register_notifier(&nfqnl_rtnl_notifier);
1606 status = nfnetlink_subsys_register(&nfqnl_subsys);
1607 if (status < 0) {
1608 pr_err("failed to create netlink socket\n");
1609 goto cleanup_netlink_notifier;
1610 }
1611
1612 status = register_netdevice_notifier(&nfqnl_dev_notifier);
1613 if (status < 0) {
1614 pr_err("failed to register netdevice notifier\n");
1615 goto cleanup_netlink_subsys;
1616 }
1617
1618 nf_register_queue_handler(&nfqh);
1619
1620 return status;
1621
1622cleanup_netlink_subsys:
1623 nfnetlink_subsys_unregister(&nfqnl_subsys);
1624cleanup_netlink_notifier:
1625 netlink_unregister_notifier(&nfqnl_rtnl_notifier);
1626 unregister_pernet_subsys(&nfnl_queue_net_ops);
1627out:
1628 return status;
1629}
1630
1631static void __exit nfnetlink_queue_fini(void)
1632{
1633 nf_unregister_queue_handler();
1634 unregister_netdevice_notifier(&nfqnl_dev_notifier);
1635 nfnetlink_subsys_unregister(&nfqnl_subsys);
1636 netlink_unregister_notifier(&nfqnl_rtnl_notifier);
1637 unregister_pernet_subsys(&nfnl_queue_net_ops);
1638
1639 rcu_barrier(); /* Wait for completion of call_rcu()'s */
1640}
1641
1642MODULE_DESCRIPTION("netfilter packet queue handler");
1643MODULE_AUTHOR("Harald Welte <laforge@netfilter.org>");
1644MODULE_LICENSE("GPL");
1645MODULE_ALIAS_NFNL_SUBSYS(NFNL_SUBSYS_QUEUE);
1646
1647module_init(nfnetlink_queue_init);
1648module_exit(nfnetlink_queue_fini);