2 * This is a module which is used for queueing packets and communicating with
3 * userspace via nfnetlink.
5 * (C) 2005 by Harald Welte <laforge@netfilter.org>
6 * (C) 2007 by Patrick McHardy <kaber@trash.net>
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>
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.
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>
34 #include <net/tcp_states.h>
35 #include <net/netfilter/nf_queue.h>
36 #include <net/netns/generic.h>
38 #include <linux/atomic.h>
40 #if IS_ENABLED(CONFIG_BRIDGE_NETFILTER)
41 #include "../bridge/br_private.h"
44 #define NFQNL_QMAX_DEFAULT 1024
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.
52 #define NFQNL_MAX_COPY_RANGE (0xffff - NLA_HDRLEN)
54 struct nfqnl_instance {
55 struct hlist_node hlist; /* global list of queues */
59 unsigned int queue_maxlen;
60 unsigned int copy_range;
61 unsigned int queue_dropped;
62 unsigned int queue_user_dropped;
65 u_int16_t queue_num; /* number of this queue */
67 u_int32_t flags; /* Set using NFQA_CFG_FLAGS */
69 * Following fields are dirtied for each queued packet,
70 * keep them in same cache line if possible.
73 unsigned int queue_total;
74 unsigned int id_sequence; /* 'sequence' of pkt ids */
75 struct list_head queue_list; /* packets in queue */
78 typedef int (*nfqnl_cmpfn)(struct nf_queue_entry *, unsigned long);
80 static int nfnl_queue_net_id __read_mostly;
82 #define INSTANCE_BUCKETS 16
83 struct nfnl_queue_net {
84 spinlock_t instances_lock;
85 struct hlist_head instance_table[INSTANCE_BUCKETS];
88 static struct nfnl_queue_net *nfnl_queue_pernet(struct net *net)
90 return net_generic(net, nfnl_queue_net_id);
93 static inline u_int8_t instance_hashfn(u_int16_t queue_num)
95 return ((queue_num >> 8) ^ queue_num) % INSTANCE_BUCKETS;
98 static struct nfqnl_instance *
99 instance_lookup(struct nfnl_queue_net *q, u_int16_t queue_num)
101 struct hlist_head *head;
102 struct nfqnl_instance *inst;
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)
112 static struct nfqnl_instance *
113 instance_create(struct nfnl_queue_net *q, u_int16_t queue_num, u32 portid)
115 struct nfqnl_instance *inst;
119 spin_lock(&q->instances_lock);
120 if (instance_lookup(q, queue_num)) {
125 inst = kzalloc(sizeof(*inst), GFP_ATOMIC);
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);
139 if (!try_module_get(THIS_MODULE)) {
144 h = instance_hashfn(queue_num);
145 hlist_add_head_rcu(&inst->hlist, &q->instance_table[h]);
147 spin_unlock(&q->instances_lock);
154 spin_unlock(&q->instances_lock);
158 static void nfqnl_flush(struct nfqnl_instance *queue, nfqnl_cmpfn cmpfn,
162 instance_destroy_rcu(struct rcu_head *head)
164 struct nfqnl_instance *inst = container_of(head, struct nfqnl_instance,
167 nfqnl_flush(inst, NULL, 0);
169 module_put(THIS_MODULE);
173 __instance_destroy(struct nfqnl_instance *inst)
175 hlist_del_rcu(&inst->hlist);
176 call_rcu(&inst->rcu, instance_destroy_rcu);
180 instance_destroy(struct nfnl_queue_net *q, struct nfqnl_instance *inst)
182 spin_lock(&q->instances_lock);
183 __instance_destroy(inst);
184 spin_unlock(&q->instances_lock);
188 __enqueue_entry(struct nfqnl_instance *queue, struct nf_queue_entry *entry)
190 list_add_tail(&entry->list, &queue->queue_list);
191 queue->queue_total++;
195 __dequeue_entry(struct nfqnl_instance *queue, struct nf_queue_entry *entry)
197 list_del(&entry->list);
198 queue->queue_total--;
201 static struct nf_queue_entry *
202 find_dequeue_entry(struct nfqnl_instance *queue, unsigned int id)
204 struct nf_queue_entry *entry = NULL, *i;
206 spin_lock_bh(&queue->lock);
208 list_for_each_entry(i, &queue->queue_list, list) {
216 __dequeue_entry(queue, entry);
218 spin_unlock_bh(&queue->lock);
224 nfqnl_flush(struct nfqnl_instance *queue, nfqnl_cmpfn cmpfn, unsigned long data)
226 struct nf_queue_entry *entry, *next;
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);
236 spin_unlock_bh(&queue->lock);
240 nfqnl_put_packet_info(struct sk_buff *nlskb, struct sk_buff *packet,
245 if (packet->ip_summed == CHECKSUM_PARTIAL)
246 flags = NFQA_SKB_CSUMNOTREADY;
247 else if (csum_verify)
248 flags = NFQA_SKB_CSUM_NOTVERIFIED;
250 if (skb_is_gso(packet))
251 flags |= NFQA_SKB_GSO;
253 return flags ? nla_put_be32(nlskb, NFQA_SKB_INFO, htonl(flags)) : 0;
256 static int nfqnl_put_sk_uidgid(struct sk_buff *skb, struct sock *sk)
258 const struct cred *cred;
260 if (!sk_fullsock(sk))
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;
273 read_unlock_bh(&sk->sk_callback_lock);
277 read_unlock_bh(&sk->sk_callback_lock);
281 static u32 nfqnl_get_sk_secctx(struct sk_buff *skb, char **secdata)
284 #if IS_ENABLED(CONFIG_NETWORK_SECMARK)
285 if (!skb || !sk_fullsock(skb->sk))
288 read_lock_bh(&skb->sk->sk_callback_lock);
291 security_secid_to_secctx(skb->secmark, secdata, &seclen);
293 read_unlock_bh(&skb->sk->sk_callback_lock);
298 static struct sk_buff *
299 nfqnl_build_packet_message(struct net *net, struct nfqnl_instance *queue,
300 struct nf_queue_entry *entry,
301 __be32 **packet_id_ptr)
304 size_t data_len = 0, cap_len = 0, rem_len = 0;
305 unsigned int hlen = 0;
308 struct nfqnl_msg_packet_hdr *pmsg;
309 struct nlmsghdr *nlh;
310 struct nfgenmsg *nfmsg;
311 struct sk_buff *entskb = entry->skb;
312 struct net_device *indev;
313 struct net_device *outdev;
314 struct nf_conn *ct = NULL;
315 enum ip_conntrack_info uninitialized_var(ctinfo);
316 struct nfnl_ct_hook *nfnl_ct;
318 char *secdata = NULL;
321 size = nlmsg_total_size(sizeof(struct nfgenmsg))
322 + nla_total_size(sizeof(struct nfqnl_msg_packet_hdr))
323 + nla_total_size(sizeof(u_int32_t)) /* ifindex */
324 + nla_total_size(sizeof(u_int32_t)) /* ifindex */
325 #if IS_ENABLED(CONFIG_BRIDGE_NETFILTER)
326 + nla_total_size(sizeof(u_int32_t)) /* ifindex */
327 + nla_total_size(sizeof(u_int32_t)) /* ifindex */
329 + nla_total_size(sizeof(u_int32_t)) /* mark */
330 + nla_total_size(sizeof(struct nfqnl_msg_packet_hw))
331 + nla_total_size(sizeof(u_int32_t)) /* skbinfo */
332 + nla_total_size(sizeof(u_int32_t)); /* cap_len */
334 if (entskb->tstamp.tv64)
335 size += nla_total_size(sizeof(struct nfqnl_msg_packet_timestamp));
337 if (entry->state.hook <= NF_INET_FORWARD ||
338 (entry->state.hook == NF_INET_POST_ROUTING && entskb->sk == NULL))
339 csum_verify = !skb_csum_unnecessary(entskb);
343 outdev = entry->state.out;
345 switch ((enum nfqnl_config_mode)ACCESS_ONCE(queue->copy_mode)) {
346 case NFQNL_COPY_META:
347 case NFQNL_COPY_NONE:
350 case NFQNL_COPY_PACKET:
351 if (!(queue->flags & NFQA_CFG_F_GSO) &&
352 entskb->ip_summed == CHECKSUM_PARTIAL &&
353 skb_checksum_help(entskb))
356 data_len = ACCESS_ONCE(queue->copy_range);
357 if (data_len > entskb->len)
358 data_len = entskb->len;
360 hlen = skb_zerocopy_headlen(entskb);
361 hlen = min_t(unsigned int, hlen, data_len);
362 size += sizeof(struct nlattr) + hlen;
363 cap_len = entskb->len;
364 rem_len = data_len - hlen;
368 nfnl_ct = rcu_dereference(nfnl_ct_hook);
370 if (queue->flags & NFQA_CFG_F_CONNTRACK) {
371 if (nfnl_ct != NULL) {
372 ct = nfnl_ct->get_ct(entskb, &ctinfo);
374 size += nfnl_ct->build_size(ct);
378 if (queue->flags & NFQA_CFG_F_UID_GID) {
379 size += (nla_total_size(sizeof(u_int32_t)) /* uid */
380 + nla_total_size(sizeof(u_int32_t))); /* gid */
383 if ((queue->flags & NFQA_CFG_F_SECCTX) && entskb->sk) {
384 seclen = nfqnl_get_sk_secctx(entskb, &secdata);
386 size += nla_total_size(seclen);
389 skb = __netlink_alloc_skb(net->nfnl, size, rem_len, queue->peer_portid,
392 skb_tx_error(entskb);
396 nlh = nlmsg_put(skb, 0, 0,
397 NFNL_SUBSYS_QUEUE << 8 | NFQNL_MSG_PACKET,
398 sizeof(struct nfgenmsg), 0);
400 skb_tx_error(entskb);
404 nfmsg = nlmsg_data(nlh);
405 nfmsg->nfgen_family = entry->state.pf;
406 nfmsg->version = NFNETLINK_V0;
407 nfmsg->res_id = htons(queue->queue_num);
409 nla = __nla_reserve(skb, NFQA_PACKET_HDR, sizeof(*pmsg));
410 pmsg = nla_data(nla);
411 pmsg->hw_protocol = entskb->protocol;
412 pmsg->hook = entry->state.hook;
413 *packet_id_ptr = &pmsg->packet_id;
415 indev = entry->state.in;
417 #if !IS_ENABLED(CONFIG_BRIDGE_NETFILTER)
418 if (nla_put_be32(skb, NFQA_IFINDEX_INDEV, htonl(indev->ifindex)))
419 goto nla_put_failure;
421 if (entry->state.pf == PF_BRIDGE) {
422 /* Case 1: indev is physical input device, we need to
423 * look for bridge group (when called from
424 * netfilter_bridge) */
425 if (nla_put_be32(skb, NFQA_IFINDEX_PHYSINDEV,
426 htonl(indev->ifindex)) ||
427 /* this is the bridge group "brX" */
428 /* rcu_read_lock()ed by __nf_queue */
429 nla_put_be32(skb, NFQA_IFINDEX_INDEV,
430 htonl(br_port_get_rcu(indev)->br->dev->ifindex)))
431 goto nla_put_failure;
435 /* Case 2: indev is bridge group, we need to look for
436 * physical device (when called from ipv4) */
437 if (nla_put_be32(skb, NFQA_IFINDEX_INDEV,
438 htonl(indev->ifindex)))
439 goto nla_put_failure;
441 physinif = nf_bridge_get_physinif(entskb);
443 nla_put_be32(skb, NFQA_IFINDEX_PHYSINDEV,
445 goto nla_put_failure;
451 #if !IS_ENABLED(CONFIG_BRIDGE_NETFILTER)
452 if (nla_put_be32(skb, NFQA_IFINDEX_OUTDEV, htonl(outdev->ifindex)))
453 goto nla_put_failure;
455 if (entry->state.pf == PF_BRIDGE) {
456 /* Case 1: outdev is physical output device, we need to
457 * look for bridge group (when called from
458 * netfilter_bridge) */
459 if (nla_put_be32(skb, NFQA_IFINDEX_PHYSOUTDEV,
460 htonl(outdev->ifindex)) ||
461 /* this is the bridge group "brX" */
462 /* rcu_read_lock()ed by __nf_queue */
463 nla_put_be32(skb, NFQA_IFINDEX_OUTDEV,
464 htonl(br_port_get_rcu(outdev)->br->dev->ifindex)))
465 goto nla_put_failure;
469 /* Case 2: outdev is bridge group, we need to look for
470 * physical output device (when called from ipv4) */
471 if (nla_put_be32(skb, NFQA_IFINDEX_OUTDEV,
472 htonl(outdev->ifindex)))
473 goto nla_put_failure;
475 physoutif = nf_bridge_get_physoutif(entskb);
477 nla_put_be32(skb, NFQA_IFINDEX_PHYSOUTDEV,
479 goto nla_put_failure;
485 nla_put_be32(skb, NFQA_MARK, htonl(entskb->mark)))
486 goto nla_put_failure;
488 if (indev && entskb->dev &&
489 entskb->mac_header != entskb->network_header) {
490 struct nfqnl_msg_packet_hw phw;
493 memset(&phw, 0, sizeof(phw));
494 len = dev_parse_header(entskb, phw.hw_addr);
496 phw.hw_addrlen = htons(len);
497 if (nla_put(skb, NFQA_HWADDR, sizeof(phw), &phw))
498 goto nla_put_failure;
502 if (entskb->tstamp.tv64) {
503 struct nfqnl_msg_packet_timestamp ts;
504 struct timespec64 kts = ktime_to_timespec64(skb->tstamp);
506 ts.sec = cpu_to_be64(kts.tv_sec);
507 ts.usec = cpu_to_be64(kts.tv_nsec / NSEC_PER_USEC);
509 if (nla_put(skb, NFQA_TIMESTAMP, sizeof(ts), &ts))
510 goto nla_put_failure;
513 if ((queue->flags & NFQA_CFG_F_UID_GID) && entskb->sk &&
514 nfqnl_put_sk_uidgid(skb, entskb->sk) < 0)
515 goto nla_put_failure;
517 if (seclen && nla_put(skb, NFQA_SECCTX, seclen, secdata))
518 goto nla_put_failure;
520 if (ct && nfnl_ct->build(skb, ct, ctinfo, NFQA_CT, NFQA_CT_INFO) < 0)
521 goto nla_put_failure;
523 if (cap_len > data_len &&
524 nla_put_be32(skb, NFQA_CAP_LEN, htonl(cap_len)))
525 goto nla_put_failure;
527 if (nfqnl_put_packet_info(skb, entskb, csum_verify))
528 goto nla_put_failure;
533 if (skb_tailroom(skb) < sizeof(*nla) + hlen)
534 goto nla_put_failure;
536 nla = (struct nlattr *)skb_put(skb, sizeof(*nla));
537 nla->nla_type = NFQA_PAYLOAD;
538 nla->nla_len = nla_attr_size(data_len);
540 if (skb_zerocopy(skb, entskb, data_len, hlen))
541 goto nla_put_failure;
544 nlh->nlmsg_len = skb->len;
548 skb_tx_error(entskb);
550 net_err_ratelimited("nf_queue: error creating packet message\n");
555 __nfqnl_enqueue_packet(struct net *net, struct nfqnl_instance *queue,
556 struct nf_queue_entry *entry)
558 struct sk_buff *nskb;
560 __be32 *packet_id_ptr;
563 nskb = nfqnl_build_packet_message(net, queue, entry, &packet_id_ptr);
568 spin_lock_bh(&queue->lock);
570 if (queue->queue_total >= queue->queue_maxlen) {
571 if (queue->flags & NFQA_CFG_F_FAIL_OPEN) {
575 queue->queue_dropped++;
576 net_warn_ratelimited("nf_queue: full at %d entries, dropping packets(s)\n",
579 goto err_out_free_nskb;
581 entry->id = ++queue->id_sequence;
582 *packet_id_ptr = htonl(entry->id);
584 /* nfnetlink_unicast will either free the nskb or add it to a socket */
585 err = nfnetlink_unicast(nskb, net, queue->peer_portid, MSG_DONTWAIT);
587 queue->queue_user_dropped++;
591 __enqueue_entry(queue, entry);
593 spin_unlock_bh(&queue->lock);
599 spin_unlock_bh(&queue->lock);
601 nf_reinject(entry, NF_ACCEPT);
606 static struct nf_queue_entry *
607 nf_queue_entry_dup(struct nf_queue_entry *e)
609 struct nf_queue_entry *entry = kmemdup(e, e->size, GFP_ATOMIC);
611 nf_queue_entry_get_refs(entry);
615 #if IS_ENABLED(CONFIG_BRIDGE_NETFILTER)
616 /* When called from bridge netfilter, skb->data must point to MAC header
617 * before calling skb_gso_segment(). Else, original MAC header is lost
618 * and segmented skbs will be sent to wrong destination.
620 static void nf_bridge_adjust_skb_data(struct sk_buff *skb)
623 __skb_push(skb, skb->network_header - skb->mac_header);
626 static void nf_bridge_adjust_segmented_data(struct sk_buff *skb)
629 __skb_pull(skb, skb->network_header - skb->mac_header);
632 #define nf_bridge_adjust_skb_data(s) do {} while (0)
633 #define nf_bridge_adjust_segmented_data(s) do {} while (0)
636 static void free_entry(struct nf_queue_entry *entry)
638 nf_queue_entry_release_refs(entry);
643 __nfqnl_enqueue_packet_gso(struct net *net, struct nfqnl_instance *queue,
644 struct sk_buff *skb, struct nf_queue_entry *entry)
647 struct nf_queue_entry *entry_seg;
649 nf_bridge_adjust_segmented_data(skb);
651 if (skb->next == NULL) { /* last packet, no need to copy entry */
652 struct sk_buff *gso_skb = entry->skb;
654 ret = __nfqnl_enqueue_packet(net, queue, entry);
656 entry->skb = gso_skb;
662 entry_seg = nf_queue_entry_dup(entry);
664 entry_seg->skb = skb;
665 ret = __nfqnl_enqueue_packet(net, queue, entry_seg);
667 free_entry(entry_seg);
673 nfqnl_enqueue_packet(struct nf_queue_entry *entry, unsigned int queuenum)
676 struct nfqnl_instance *queue;
677 struct sk_buff *skb, *segs;
679 struct net *net = entry->state.net;
680 struct nfnl_queue_net *q = nfnl_queue_pernet(net);
682 /* rcu_read_lock()ed by nf_hook_slow() */
683 queue = instance_lookup(q, queuenum);
687 if (queue->copy_mode == NFQNL_COPY_NONE)
692 switch (entry->state.pf) {
694 skb->protocol = htons(ETH_P_IP);
697 skb->protocol = htons(ETH_P_IPV6);
701 if ((queue->flags & NFQA_CFG_F_GSO) || !skb_is_gso(skb))
702 return __nfqnl_enqueue_packet(net, queue, entry);
704 nf_bridge_adjust_skb_data(skb);
705 segs = skb_gso_segment(skb, 0);
706 /* Does not use PTR_ERR to limit the number of error codes that can be
707 * returned by nf_queue. For instance, callers rely on -ESRCH to
708 * mean 'ignore this hook'.
710 if (IS_ERR_OR_NULL(segs))
715 struct sk_buff *nskb = segs->next;
717 err = __nfqnl_enqueue_packet_gso(net, queue,
727 if (err) /* some segments are already queued */
733 nf_bridge_adjust_segmented_data(skb);
738 nfqnl_mangle(void *data, int data_len, struct nf_queue_entry *e, int diff)
740 struct sk_buff *nskb;
743 if (pskb_trim(e->skb, data_len))
745 } else if (diff > 0) {
746 if (data_len > 0xFFFF)
748 if (diff > skb_tailroom(e->skb)) {
749 nskb = skb_copy_expand(e->skb, skb_headroom(e->skb),
752 printk(KERN_WARNING "nf_queue: OOM "
753 "in mangle, dropping packet\n");
759 skb_put(e->skb, diff);
761 if (!skb_make_writable(e->skb, data_len))
763 skb_copy_to_linear_data(e->skb, data, data_len);
764 e->skb->ip_summed = CHECKSUM_NONE;
769 nfqnl_set_mode(struct nfqnl_instance *queue,
770 unsigned char mode, unsigned int range)
774 spin_lock_bh(&queue->lock);
776 case NFQNL_COPY_NONE:
777 case NFQNL_COPY_META:
778 queue->copy_mode = mode;
779 queue->copy_range = 0;
782 case NFQNL_COPY_PACKET:
783 queue->copy_mode = mode;
784 if (range == 0 || range > NFQNL_MAX_COPY_RANGE)
785 queue->copy_range = NFQNL_MAX_COPY_RANGE;
787 queue->copy_range = range;
794 spin_unlock_bh(&queue->lock);
800 dev_cmp(struct nf_queue_entry *entry, unsigned long ifindex)
803 if (entry->state.in->ifindex == ifindex)
805 if (entry->state.out)
806 if (entry->state.out->ifindex == ifindex)
808 #if IS_ENABLED(CONFIG_BRIDGE_NETFILTER)
809 if (entry->skb->nf_bridge) {
810 int physinif, physoutif;
812 physinif = nf_bridge_get_physinif(entry->skb);
813 physoutif = nf_bridge_get_physoutif(entry->skb);
815 if (physinif == ifindex || physoutif == ifindex)
822 /* drop all packets with either indev or outdev == ifindex from all queue
825 nfqnl_dev_drop(struct net *net, int ifindex)
828 struct nfnl_queue_net *q = nfnl_queue_pernet(net);
832 for (i = 0; i < INSTANCE_BUCKETS; i++) {
833 struct nfqnl_instance *inst;
834 struct hlist_head *head = &q->instance_table[i];
836 hlist_for_each_entry_rcu(inst, head, hlist)
837 nfqnl_flush(inst, dev_cmp, ifindex);
844 nfqnl_rcv_dev_event(struct notifier_block *this,
845 unsigned long event, void *ptr)
847 struct net_device *dev = netdev_notifier_info_to_dev(ptr);
849 /* Drop any packets associated with the downed device */
850 if (event == NETDEV_DOWN)
851 nfqnl_dev_drop(dev_net(dev), dev->ifindex);
855 static struct notifier_block nfqnl_dev_notifier = {
856 .notifier_call = nfqnl_rcv_dev_event,
859 static int nf_hook_cmp(struct nf_queue_entry *entry, unsigned long ops_ptr)
861 return entry->elem == (struct nf_hook_ops *)ops_ptr;
864 static void nfqnl_nf_hook_drop(struct net *net, struct nf_hook_ops *hook)
866 struct nfnl_queue_net *q = nfnl_queue_pernet(net);
870 for (i = 0; i < INSTANCE_BUCKETS; i++) {
871 struct nfqnl_instance *inst;
872 struct hlist_head *head = &q->instance_table[i];
874 hlist_for_each_entry_rcu(inst, head, hlist)
875 nfqnl_flush(inst, nf_hook_cmp, (unsigned long)hook);
881 nfqnl_rcv_nl_event(struct notifier_block *this,
882 unsigned long event, void *ptr)
884 struct netlink_notify *n = ptr;
885 struct nfnl_queue_net *q = nfnl_queue_pernet(n->net);
887 if (event == NETLINK_URELEASE && n->protocol == NETLINK_NETFILTER) {
890 /* destroy all instances for this portid */
891 spin_lock(&q->instances_lock);
892 for (i = 0; i < INSTANCE_BUCKETS; i++) {
893 struct hlist_node *t2;
894 struct nfqnl_instance *inst;
895 struct hlist_head *head = &q->instance_table[i];
897 hlist_for_each_entry_safe(inst, t2, head, hlist) {
898 if (n->portid == inst->peer_portid)
899 __instance_destroy(inst);
902 spin_unlock(&q->instances_lock);
907 static struct notifier_block nfqnl_rtnl_notifier = {
908 .notifier_call = nfqnl_rcv_nl_event,
911 static const struct nla_policy nfqa_verdict_policy[NFQA_MAX+1] = {
912 [NFQA_VERDICT_HDR] = { .len = sizeof(struct nfqnl_msg_verdict_hdr) },
913 [NFQA_MARK] = { .type = NLA_U32 },
914 [NFQA_PAYLOAD] = { .type = NLA_UNSPEC },
915 [NFQA_CT] = { .type = NLA_UNSPEC },
916 [NFQA_EXP] = { .type = NLA_UNSPEC },
919 static const struct nla_policy nfqa_verdict_batch_policy[NFQA_MAX+1] = {
920 [NFQA_VERDICT_HDR] = { .len = sizeof(struct nfqnl_msg_verdict_hdr) },
921 [NFQA_MARK] = { .type = NLA_U32 },
924 static struct nfqnl_instance *
925 verdict_instance_lookup(struct nfnl_queue_net *q, u16 queue_num, u32 nlportid)
927 struct nfqnl_instance *queue;
929 queue = instance_lookup(q, queue_num);
931 return ERR_PTR(-ENODEV);
933 if (queue->peer_portid != nlportid)
934 return ERR_PTR(-EPERM);
939 static struct nfqnl_msg_verdict_hdr*
940 verdicthdr_get(const struct nlattr * const nfqa[])
942 struct nfqnl_msg_verdict_hdr *vhdr;
943 unsigned int verdict;
945 if (!nfqa[NFQA_VERDICT_HDR])
948 vhdr = nla_data(nfqa[NFQA_VERDICT_HDR]);
949 verdict = ntohl(vhdr->verdict) & NF_VERDICT_MASK;
950 if (verdict > NF_MAX_VERDICT || verdict == NF_STOLEN)
955 static int nfq_id_after(unsigned int id, unsigned int max)
957 return (int)(id - max) > 0;
961 nfqnl_recv_verdict_batch(struct sock *ctnl, struct sk_buff *skb,
962 const struct nlmsghdr *nlh,
963 const struct nlattr * const nfqa[])
965 struct nfgenmsg *nfmsg = nlmsg_data(nlh);
966 struct nf_queue_entry *entry, *tmp;
967 unsigned int verdict, maxid;
968 struct nfqnl_msg_verdict_hdr *vhdr;
969 struct nfqnl_instance *queue;
970 LIST_HEAD(batch_list);
971 u16 queue_num = ntohs(nfmsg->res_id);
973 struct net *net = sock_net(ctnl);
974 struct nfnl_queue_net *q = nfnl_queue_pernet(net);
976 queue = verdict_instance_lookup(q, queue_num,
977 NETLINK_CB(skb).portid);
979 return PTR_ERR(queue);
981 vhdr = verdicthdr_get(nfqa);
985 verdict = ntohl(vhdr->verdict);
986 maxid = ntohl(vhdr->id);
988 spin_lock_bh(&queue->lock);
990 list_for_each_entry_safe(entry, tmp, &queue->queue_list, list) {
991 if (nfq_id_after(entry->id, maxid))
993 __dequeue_entry(queue, entry);
994 list_add_tail(&entry->list, &batch_list);
997 spin_unlock_bh(&queue->lock);
999 if (list_empty(&batch_list))
1002 list_for_each_entry_safe(entry, tmp, &batch_list, list) {
1003 if (nfqa[NFQA_MARK])
1004 entry->skb->mark = ntohl(nla_get_be32(nfqa[NFQA_MARK]));
1005 nf_reinject(entry, verdict);
1010 static struct nf_conn *nfqnl_ct_parse(struct nfnl_ct_hook *nfnl_ct,
1011 const struct nlmsghdr *nlh,
1012 const struct nlattr * const nfqa[],
1013 struct nf_queue_entry *entry,
1014 enum ip_conntrack_info *ctinfo)
1018 ct = nfnl_ct->get_ct(entry->skb, ctinfo);
1022 if (nfnl_ct->parse(nfqa[NFQA_CT], ct) < 0)
1026 nfnl_ct->attach_expect(nfqa[NFQA_EXP], ct,
1027 NETLINK_CB(entry->skb).portid,
1033 nfqnl_recv_verdict(struct sock *ctnl, struct sk_buff *skb,
1034 const struct nlmsghdr *nlh,
1035 const struct nlattr * const nfqa[])
1037 struct nfgenmsg *nfmsg = nlmsg_data(nlh);
1038 u_int16_t queue_num = ntohs(nfmsg->res_id);
1040 struct nfqnl_msg_verdict_hdr *vhdr;
1041 struct nfqnl_instance *queue;
1042 unsigned int verdict;
1043 struct nf_queue_entry *entry;
1044 enum ip_conntrack_info uninitialized_var(ctinfo);
1045 struct nfnl_ct_hook *nfnl_ct;
1046 struct nf_conn *ct = NULL;
1048 struct net *net = sock_net(ctnl);
1049 struct nfnl_queue_net *q = nfnl_queue_pernet(net);
1051 queue = instance_lookup(q, queue_num);
1053 queue = verdict_instance_lookup(q, queue_num,
1054 NETLINK_CB(skb).portid);
1056 return PTR_ERR(queue);
1058 vhdr = verdicthdr_get(nfqa);
1062 verdict = ntohl(vhdr->verdict);
1064 entry = find_dequeue_entry(queue, ntohl(vhdr->id));
1068 /* rcu lock already held from nfnl->call_rcu. */
1069 nfnl_ct = rcu_dereference(nfnl_ct_hook);
1071 if (nfqa[NFQA_CT]) {
1072 if (nfnl_ct != NULL)
1073 ct = nfqnl_ct_parse(nfnl_ct, nlh, nfqa, entry, &ctinfo);
1076 if (nfqa[NFQA_PAYLOAD]) {
1077 u16 payload_len = nla_len(nfqa[NFQA_PAYLOAD]);
1078 int diff = payload_len - entry->skb->len;
1080 if (nfqnl_mangle(nla_data(nfqa[NFQA_PAYLOAD]),
1081 payload_len, entry, diff) < 0)
1085 nfnl_ct->seq_adjust(entry->skb, ct, ctinfo, diff);
1088 if (nfqa[NFQA_MARK])
1089 entry->skb->mark = ntohl(nla_get_be32(nfqa[NFQA_MARK]));
1091 nf_reinject(entry, verdict);
1096 nfqnl_recv_unsupp(struct sock *ctnl, struct sk_buff *skb,
1097 const struct nlmsghdr *nlh,
1098 const struct nlattr * const nfqa[])
1103 static const struct nla_policy nfqa_cfg_policy[NFQA_CFG_MAX+1] = {
1104 [NFQA_CFG_CMD] = { .len = sizeof(struct nfqnl_msg_config_cmd) },
1105 [NFQA_CFG_PARAMS] = { .len = sizeof(struct nfqnl_msg_config_params) },
1108 static const struct nf_queue_handler nfqh = {
1109 .outfn = &nfqnl_enqueue_packet,
1110 .nf_hook_drop = &nfqnl_nf_hook_drop,
1114 nfqnl_recv_config(struct sock *ctnl, struct sk_buff *skb,
1115 const struct nlmsghdr *nlh,
1116 const struct nlattr * const nfqa[])
1118 struct nfgenmsg *nfmsg = nlmsg_data(nlh);
1119 u_int16_t queue_num = ntohs(nfmsg->res_id);
1120 struct nfqnl_instance *queue;
1121 struct nfqnl_msg_config_cmd *cmd = NULL;
1122 struct net *net = sock_net(ctnl);
1123 struct nfnl_queue_net *q = nfnl_queue_pernet(net);
1126 if (nfqa[NFQA_CFG_CMD]) {
1127 cmd = nla_data(nfqa[NFQA_CFG_CMD]);
1129 /* Obsolete commands without queue context */
1130 switch (cmd->command) {
1131 case NFQNL_CFG_CMD_PF_BIND: return 0;
1132 case NFQNL_CFG_CMD_PF_UNBIND: return 0;
1137 queue = instance_lookup(q, queue_num);
1138 if (queue && queue->peer_portid != NETLINK_CB(skb).portid) {
1140 goto err_out_unlock;
1144 switch (cmd->command) {
1145 case NFQNL_CFG_CMD_BIND:
1148 goto err_out_unlock;
1150 queue = instance_create(q, queue_num,
1151 NETLINK_CB(skb).portid);
1152 if (IS_ERR(queue)) {
1153 ret = PTR_ERR(queue);
1154 goto err_out_unlock;
1157 case NFQNL_CFG_CMD_UNBIND:
1160 goto err_out_unlock;
1162 instance_destroy(q, queue);
1164 case NFQNL_CFG_CMD_PF_BIND:
1165 case NFQNL_CFG_CMD_PF_UNBIND:
1173 if (nfqa[NFQA_CFG_PARAMS]) {
1174 struct nfqnl_msg_config_params *params;
1178 goto err_out_unlock;
1180 params = nla_data(nfqa[NFQA_CFG_PARAMS]);
1181 nfqnl_set_mode(queue, params->copy_mode,
1182 ntohl(params->copy_range));
1185 if (nfqa[NFQA_CFG_QUEUE_MAXLEN]) {
1186 __be32 *queue_maxlen;
1190 goto err_out_unlock;
1192 queue_maxlen = nla_data(nfqa[NFQA_CFG_QUEUE_MAXLEN]);
1193 spin_lock_bh(&queue->lock);
1194 queue->queue_maxlen = ntohl(*queue_maxlen);
1195 spin_unlock_bh(&queue->lock);
1198 if (nfqa[NFQA_CFG_FLAGS]) {
1203 goto err_out_unlock;
1206 if (!nfqa[NFQA_CFG_MASK]) {
1207 /* A mask is needed to specify which flags are being
1211 goto err_out_unlock;
1214 flags = ntohl(nla_get_be32(nfqa[NFQA_CFG_FLAGS]));
1215 mask = ntohl(nla_get_be32(nfqa[NFQA_CFG_MASK]));
1217 if (flags >= NFQA_CFG_F_MAX) {
1219 goto err_out_unlock;
1221 #if !IS_ENABLED(CONFIG_NETWORK_SECMARK)
1222 if (flags & mask & NFQA_CFG_F_SECCTX) {
1224 goto err_out_unlock;
1227 spin_lock_bh(&queue->lock);
1228 queue->flags &= ~mask;
1229 queue->flags |= flags & mask;
1230 spin_unlock_bh(&queue->lock);
1238 static const struct nfnl_callback nfqnl_cb[NFQNL_MSG_MAX] = {
1239 [NFQNL_MSG_PACKET] = { .call_rcu = nfqnl_recv_unsupp,
1240 .attr_count = NFQA_MAX, },
1241 [NFQNL_MSG_VERDICT] = { .call_rcu = nfqnl_recv_verdict,
1242 .attr_count = NFQA_MAX,
1243 .policy = nfqa_verdict_policy },
1244 [NFQNL_MSG_CONFIG] = { .call = nfqnl_recv_config,
1245 .attr_count = NFQA_CFG_MAX,
1246 .policy = nfqa_cfg_policy },
1247 [NFQNL_MSG_VERDICT_BATCH]={ .call_rcu = nfqnl_recv_verdict_batch,
1248 .attr_count = NFQA_MAX,
1249 .policy = nfqa_verdict_batch_policy },
1252 static const struct nfnetlink_subsystem nfqnl_subsys = {
1254 .subsys_id = NFNL_SUBSYS_QUEUE,
1255 .cb_count = NFQNL_MSG_MAX,
1259 #ifdef CONFIG_PROC_FS
1261 struct seq_net_private p;
1262 unsigned int bucket;
1265 static struct hlist_node *get_first(struct seq_file *seq)
1267 struct iter_state *st = seq->private;
1269 struct nfnl_queue_net *q;
1274 net = seq_file_net(seq);
1275 q = nfnl_queue_pernet(net);
1276 for (st->bucket = 0; st->bucket < INSTANCE_BUCKETS; st->bucket++) {
1277 if (!hlist_empty(&q->instance_table[st->bucket]))
1278 return q->instance_table[st->bucket].first;
1283 static struct hlist_node *get_next(struct seq_file *seq, struct hlist_node *h)
1285 struct iter_state *st = seq->private;
1286 struct net *net = seq_file_net(seq);
1290 struct nfnl_queue_net *q;
1292 if (++st->bucket >= INSTANCE_BUCKETS)
1295 q = nfnl_queue_pernet(net);
1296 h = q->instance_table[st->bucket].first;
1301 static struct hlist_node *get_idx(struct seq_file *seq, loff_t pos)
1303 struct hlist_node *head;
1304 head = get_first(seq);
1307 while (pos && (head = get_next(seq, head)))
1309 return pos ? NULL : head;
1312 static void *seq_start(struct seq_file *s, loff_t *pos)
1313 __acquires(nfnl_queue_pernet(seq_file_net(s))->instances_lock)
1315 spin_lock(&nfnl_queue_pernet(seq_file_net(s))->instances_lock);
1316 return get_idx(s, *pos);
1319 static void *seq_next(struct seq_file *s, void *v, loff_t *pos)
1322 return get_next(s, v);
1325 static void seq_stop(struct seq_file *s, void *v)
1326 __releases(nfnl_queue_pernet(seq_file_net(s))->instances_lock)
1328 spin_unlock(&nfnl_queue_pernet(seq_file_net(s))->instances_lock);
1331 static int seq_show(struct seq_file *s, void *v)
1333 const struct nfqnl_instance *inst = v;
1335 seq_printf(s, "%5u %6u %5u %1u %5u %5u %5u %8u %2d\n",
1337 inst->peer_portid, inst->queue_total,
1338 inst->copy_mode, inst->copy_range,
1339 inst->queue_dropped, inst->queue_user_dropped,
1340 inst->id_sequence, 1);
1344 static const struct seq_operations nfqnl_seq_ops = {
1351 static int nfqnl_open(struct inode *inode, struct file *file)
1353 return seq_open_net(inode, file, &nfqnl_seq_ops,
1354 sizeof(struct iter_state));
1357 static const struct file_operations nfqnl_file_ops = {
1358 .owner = THIS_MODULE,
1361 .llseek = seq_lseek,
1362 .release = seq_release_net,
1365 #endif /* PROC_FS */
1367 static int __net_init nfnl_queue_net_init(struct net *net)
1370 struct nfnl_queue_net *q = nfnl_queue_pernet(net);
1372 for (i = 0; i < INSTANCE_BUCKETS; i++)
1373 INIT_HLIST_HEAD(&q->instance_table[i]);
1375 spin_lock_init(&q->instances_lock);
1377 #ifdef CONFIG_PROC_FS
1378 if (!proc_create("nfnetlink_queue", 0440,
1379 net->nf.proc_netfilter, &nfqnl_file_ops))
1385 static void __net_exit nfnl_queue_net_exit(struct net *net)
1387 #ifdef CONFIG_PROC_FS
1388 remove_proc_entry("nfnetlink_queue", net->nf.proc_netfilter);
1392 static struct pernet_operations nfnl_queue_net_ops = {
1393 .init = nfnl_queue_net_init,
1394 .exit = nfnl_queue_net_exit,
1395 .id = &nfnl_queue_net_id,
1396 .size = sizeof(struct nfnl_queue_net),
1399 static int __init nfnetlink_queue_init(void)
1403 status = register_pernet_subsys(&nfnl_queue_net_ops);
1405 pr_err("nf_queue: failed to register pernet ops\n");
1409 netlink_register_notifier(&nfqnl_rtnl_notifier);
1410 status = nfnetlink_subsys_register(&nfqnl_subsys);
1412 pr_err("nf_queue: failed to create netlink socket\n");
1413 goto cleanup_netlink_notifier;
1416 register_netdevice_notifier(&nfqnl_dev_notifier);
1417 nf_register_queue_handler(&nfqh);
1420 cleanup_netlink_notifier:
1421 netlink_unregister_notifier(&nfqnl_rtnl_notifier);
1422 unregister_pernet_subsys(&nfnl_queue_net_ops);
1427 static void __exit nfnetlink_queue_fini(void)
1429 nf_unregister_queue_handler();
1430 unregister_netdevice_notifier(&nfqnl_dev_notifier);
1431 nfnetlink_subsys_unregister(&nfqnl_subsys);
1432 netlink_unregister_notifier(&nfqnl_rtnl_notifier);
1433 unregister_pernet_subsys(&nfnl_queue_net_ops);
1435 rcu_barrier(); /* Wait for completion of call_rcu()'s */
1438 MODULE_DESCRIPTION("netfilter packet queue handler");
1439 MODULE_AUTHOR("Harald Welte <laforge@netfilter.org>");
1440 MODULE_LICENSE("GPL");
1441 MODULE_ALIAS_NFNL_SUBSYS(NFNL_SUBSYS_QUEUE);
1443 module_init(nfnetlink_queue_init);
1444 module_exit(nfnetlink_queue_fini);