1 // SPDX-License-Identifier: GPL-2.0-only
3 * This is a module which is used for queueing packets and communicating with
4 * userspace via nfnetlink.
6 * (C) 2005 by Harald Welte <laforge@netfilter.org>
7 * (C) 2007 by Patrick McHardy <kaber@trash.net>
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>
14 #define pr_fmt(fmt) KBUILD_MODNAME ": " fmt
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>
33 #include <net/tcp_states.h>
34 #include <net/netfilter/nf_queue.h>
35 #include <net/netns/generic.h>
37 #include <linux/atomic.h>
39 #if IS_ENABLED(CONFIG_BRIDGE_NETFILTER)
40 #include "../bridge/br_private.h"
43 #if IS_ENABLED(CONFIG_NF_CONNTRACK)
44 #include <net/netfilter/nf_conntrack.h>
47 #define NFQNL_QMAX_DEFAULT 1024
49 /* We're using struct nlattr which has 16bit nla_len. Note that nla_len
50 * includes the header length. Thus, the maximum packet length that we
51 * support is 65531 bytes. We send truncated packets if the specified length
52 * is larger than that. Userspace can check for presence of NFQA_CAP_LEN
53 * attribute to detect truncation.
55 #define NFQNL_MAX_COPY_RANGE (0xffff - NLA_HDRLEN)
57 struct nfqnl_instance {
58 struct hlist_node hlist; /* global list of queues */
62 unsigned int queue_maxlen;
63 unsigned int copy_range;
64 unsigned int queue_dropped;
65 unsigned int queue_user_dropped;
68 u_int16_t queue_num; /* number of this queue */
70 u_int32_t flags; /* Set using NFQA_CFG_FLAGS */
72 * Following fields are dirtied for each queued packet,
73 * keep them in same cache line if possible.
75 spinlock_t lock ____cacheline_aligned_in_smp;
76 unsigned int queue_total;
77 unsigned int id_sequence; /* 'sequence' of pkt ids */
78 struct list_head queue_list; /* packets in queue */
81 typedef int (*nfqnl_cmpfn)(struct nf_queue_entry *, unsigned long);
83 static unsigned int nfnl_queue_net_id __read_mostly;
85 #define INSTANCE_BUCKETS 16
86 struct nfnl_queue_net {
87 spinlock_t instances_lock;
88 struct hlist_head instance_table[INSTANCE_BUCKETS];
91 static struct nfnl_queue_net *nfnl_queue_pernet(struct net *net)
93 return net_generic(net, nfnl_queue_net_id);
96 static inline u_int8_t instance_hashfn(u_int16_t queue_num)
98 return ((queue_num >> 8) ^ queue_num) % INSTANCE_BUCKETS;
101 static struct nfqnl_instance *
102 instance_lookup(struct nfnl_queue_net *q, u_int16_t queue_num)
104 struct hlist_head *head;
105 struct nfqnl_instance *inst;
107 head = &q->instance_table[instance_hashfn(queue_num)];
108 hlist_for_each_entry_rcu(inst, head, hlist) {
109 if (inst->queue_num == queue_num)
115 static struct nfqnl_instance *
116 instance_create(struct nfnl_queue_net *q, u_int16_t queue_num, u32 portid)
118 struct nfqnl_instance *inst;
122 spin_lock(&q->instances_lock);
123 if (instance_lookup(q, queue_num)) {
128 inst = kzalloc(sizeof(*inst), GFP_ATOMIC);
134 inst->queue_num = queue_num;
135 inst->peer_portid = portid;
136 inst->queue_maxlen = NFQNL_QMAX_DEFAULT;
137 inst->copy_range = NFQNL_MAX_COPY_RANGE;
138 inst->copy_mode = NFQNL_COPY_NONE;
139 spin_lock_init(&inst->lock);
140 INIT_LIST_HEAD(&inst->queue_list);
142 if (!try_module_get(THIS_MODULE)) {
147 h = instance_hashfn(queue_num);
148 hlist_add_head_rcu(&inst->hlist, &q->instance_table[h]);
150 spin_unlock(&q->instances_lock);
157 spin_unlock(&q->instances_lock);
161 static void nfqnl_flush(struct nfqnl_instance *queue, nfqnl_cmpfn cmpfn,
165 instance_destroy_rcu(struct rcu_head *head)
167 struct nfqnl_instance *inst = container_of(head, struct nfqnl_instance,
170 nfqnl_flush(inst, NULL, 0);
172 module_put(THIS_MODULE);
176 __instance_destroy(struct nfqnl_instance *inst)
178 hlist_del_rcu(&inst->hlist);
179 call_rcu(&inst->rcu, instance_destroy_rcu);
183 instance_destroy(struct nfnl_queue_net *q, struct nfqnl_instance *inst)
185 spin_lock(&q->instances_lock);
186 __instance_destroy(inst);
187 spin_unlock(&q->instances_lock);
191 __enqueue_entry(struct nfqnl_instance *queue, struct nf_queue_entry *entry)
193 list_add_tail(&entry->list, &queue->queue_list);
194 queue->queue_total++;
198 __dequeue_entry(struct nfqnl_instance *queue, struct nf_queue_entry *entry)
200 list_del(&entry->list);
201 queue->queue_total--;
204 static struct nf_queue_entry *
205 find_dequeue_entry(struct nfqnl_instance *queue, unsigned int id)
207 struct nf_queue_entry *entry = NULL, *i;
209 spin_lock_bh(&queue->lock);
211 list_for_each_entry(i, &queue->queue_list, list) {
219 __dequeue_entry(queue, entry);
221 spin_unlock_bh(&queue->lock);
226 static void nfqnl_reinject(struct nf_queue_entry *entry, unsigned int verdict)
228 const struct nf_ct_hook *ct_hook;
231 if (verdict == NF_ACCEPT ||
232 verdict == NF_REPEAT ||
233 verdict == NF_STOP) {
235 ct_hook = rcu_dereference(nf_ct_hook);
237 err = ct_hook->update(entry->state.net, entry->skb);
243 nf_reinject(entry, verdict);
247 nfqnl_flush(struct nfqnl_instance *queue, nfqnl_cmpfn cmpfn, unsigned long data)
249 struct nf_queue_entry *entry, *next;
251 spin_lock_bh(&queue->lock);
252 list_for_each_entry_safe(entry, next, &queue->queue_list, list) {
253 if (!cmpfn || cmpfn(entry, data)) {
254 list_del(&entry->list);
255 queue->queue_total--;
256 nfqnl_reinject(entry, NF_DROP);
259 spin_unlock_bh(&queue->lock);
263 nfqnl_put_packet_info(struct sk_buff *nlskb, struct sk_buff *packet,
268 if (packet->ip_summed == CHECKSUM_PARTIAL)
269 flags = NFQA_SKB_CSUMNOTREADY;
270 else if (csum_verify)
271 flags = NFQA_SKB_CSUM_NOTVERIFIED;
273 if (skb_is_gso(packet))
274 flags |= NFQA_SKB_GSO;
276 return flags ? nla_put_be32(nlskb, NFQA_SKB_INFO, htonl(flags)) : 0;
279 static int nfqnl_put_sk_uidgid(struct sk_buff *skb, struct sock *sk)
281 const struct cred *cred;
283 if (!sk_fullsock(sk))
286 read_lock_bh(&sk->sk_callback_lock);
287 if (sk->sk_socket && sk->sk_socket->file) {
288 cred = sk->sk_socket->file->f_cred;
289 if (nla_put_be32(skb, NFQA_UID,
290 htonl(from_kuid_munged(&init_user_ns, cred->fsuid))))
291 goto nla_put_failure;
292 if (nla_put_be32(skb, NFQA_GID,
293 htonl(from_kgid_munged(&init_user_ns, cred->fsgid))))
294 goto nla_put_failure;
296 read_unlock_bh(&sk->sk_callback_lock);
300 read_unlock_bh(&sk->sk_callback_lock);
304 static u32 nfqnl_get_sk_secctx(struct sk_buff *skb, char **secdata)
307 #if IS_ENABLED(CONFIG_NETWORK_SECMARK)
308 if (!skb || !sk_fullsock(skb->sk))
311 read_lock_bh(&skb->sk->sk_callback_lock);
314 security_secid_to_secctx(skb->secmark, secdata, &seclen);
316 read_unlock_bh(&skb->sk->sk_callback_lock);
321 static u32 nfqnl_get_bridge_size(struct nf_queue_entry *entry)
323 struct sk_buff *entskb = entry->skb;
326 if (entry->state.pf != PF_BRIDGE || !skb_mac_header_was_set(entskb))
329 if (skb_vlan_tag_present(entskb))
330 nlalen += nla_total_size(nla_total_size(sizeof(__be16)) +
331 nla_total_size(sizeof(__be16)));
333 if (entskb->network_header > entskb->mac_header)
334 nlalen += nla_total_size((entskb->network_header -
335 entskb->mac_header));
340 static int nfqnl_put_bridge(struct nf_queue_entry *entry, struct sk_buff *skb)
342 struct sk_buff *entskb = entry->skb;
344 if (entry->state.pf != PF_BRIDGE || !skb_mac_header_was_set(entskb))
347 if (skb_vlan_tag_present(entskb)) {
350 nest = nla_nest_start(skb, NFQA_VLAN);
352 goto nla_put_failure;
354 if (nla_put_be16(skb, NFQA_VLAN_TCI, htons(entskb->vlan_tci)) ||
355 nla_put_be16(skb, NFQA_VLAN_PROTO, entskb->vlan_proto))
356 goto nla_put_failure;
358 nla_nest_end(skb, nest);
361 if (entskb->mac_header < entskb->network_header) {
362 int len = (int)(entskb->network_header - entskb->mac_header);
364 if (nla_put(skb, NFQA_L2HDR, len, skb_mac_header(entskb)))
365 goto nla_put_failure;
374 static struct sk_buff *
375 nfqnl_build_packet_message(struct net *net, struct nfqnl_instance *queue,
376 struct nf_queue_entry *entry,
377 __be32 **packet_id_ptr)
380 size_t data_len = 0, cap_len = 0;
381 unsigned int hlen = 0;
384 struct nfqnl_msg_packet_hdr *pmsg;
385 struct nlmsghdr *nlh;
386 struct sk_buff *entskb = entry->skb;
387 struct net_device *indev;
388 struct net_device *outdev;
389 struct nf_conn *ct = NULL;
390 enum ip_conntrack_info ctinfo = 0;
391 const struct nfnl_ct_hook *nfnl_ct;
393 char *secdata = NULL;
396 size = nlmsg_total_size(sizeof(struct nfgenmsg))
397 + nla_total_size(sizeof(struct nfqnl_msg_packet_hdr))
398 + nla_total_size(sizeof(u_int32_t)) /* ifindex */
399 + nla_total_size(sizeof(u_int32_t)) /* ifindex */
400 #if IS_ENABLED(CONFIG_BRIDGE_NETFILTER)
401 + nla_total_size(sizeof(u_int32_t)) /* ifindex */
402 + nla_total_size(sizeof(u_int32_t)) /* ifindex */
404 + nla_total_size(sizeof(u_int32_t)) /* mark */
405 + nla_total_size(sizeof(struct nfqnl_msg_packet_hw))
406 + nla_total_size(sizeof(u_int32_t)) /* skbinfo */
407 + nla_total_size(sizeof(u_int32_t)); /* cap_len */
410 size += nla_total_size(sizeof(struct nfqnl_msg_packet_timestamp));
412 size += nfqnl_get_bridge_size(entry);
414 if (entry->state.hook <= NF_INET_FORWARD ||
415 (entry->state.hook == NF_INET_POST_ROUTING && entskb->sk == NULL))
416 csum_verify = !skb_csum_unnecessary(entskb);
420 outdev = entry->state.out;
422 switch ((enum nfqnl_config_mode)READ_ONCE(queue->copy_mode)) {
423 case NFQNL_COPY_META:
424 case NFQNL_COPY_NONE:
427 case NFQNL_COPY_PACKET:
428 if (!(queue->flags & NFQA_CFG_F_GSO) &&
429 entskb->ip_summed == CHECKSUM_PARTIAL &&
430 skb_checksum_help(entskb))
433 data_len = READ_ONCE(queue->copy_range);
434 if (data_len > entskb->len)
435 data_len = entskb->len;
437 hlen = skb_zerocopy_headlen(entskb);
438 hlen = min_t(unsigned int, hlen, data_len);
439 size += sizeof(struct nlattr) + hlen;
440 cap_len = entskb->len;
444 nfnl_ct = rcu_dereference(nfnl_ct_hook);
446 #if IS_ENABLED(CONFIG_NF_CONNTRACK)
447 if (queue->flags & NFQA_CFG_F_CONNTRACK) {
448 if (nfnl_ct != NULL) {
449 ct = nf_ct_get(entskb, &ctinfo);
451 size += nfnl_ct->build_size(ct);
456 if (queue->flags & NFQA_CFG_F_UID_GID) {
457 size += (nla_total_size(sizeof(u_int32_t)) /* uid */
458 + nla_total_size(sizeof(u_int32_t))); /* gid */
461 if ((queue->flags & NFQA_CFG_F_SECCTX) && entskb->sk) {
462 seclen = nfqnl_get_sk_secctx(entskb, &secdata);
464 size += nla_total_size(seclen);
467 skb = alloc_skb(size, GFP_ATOMIC);
469 skb_tx_error(entskb);
473 nlh = nfnl_msg_put(skb, 0, 0,
474 nfnl_msg_type(NFNL_SUBSYS_QUEUE, NFQNL_MSG_PACKET),
475 0, entry->state.pf, NFNETLINK_V0,
476 htons(queue->queue_num));
478 skb_tx_error(entskb);
483 nla = __nla_reserve(skb, NFQA_PACKET_HDR, sizeof(*pmsg));
484 pmsg = nla_data(nla);
485 pmsg->hw_protocol = entskb->protocol;
486 pmsg->hook = entry->state.hook;
487 *packet_id_ptr = &pmsg->packet_id;
489 indev = entry->state.in;
491 #if !IS_ENABLED(CONFIG_BRIDGE_NETFILTER)
492 if (nla_put_be32(skb, NFQA_IFINDEX_INDEV, htonl(indev->ifindex)))
493 goto nla_put_failure;
495 if (entry->state.pf == PF_BRIDGE) {
496 /* Case 1: indev is physical input device, we need to
497 * look for bridge group (when called from
498 * netfilter_bridge) */
499 if (nla_put_be32(skb, NFQA_IFINDEX_PHYSINDEV,
500 htonl(indev->ifindex)) ||
501 /* this is the bridge group "brX" */
502 /* rcu_read_lock()ed by __nf_queue */
503 nla_put_be32(skb, NFQA_IFINDEX_INDEV,
504 htonl(br_port_get_rcu(indev)->br->dev->ifindex)))
505 goto nla_put_failure;
509 /* Case 2: indev is bridge group, we need to look for
510 * physical device (when called from ipv4) */
511 if (nla_put_be32(skb, NFQA_IFINDEX_INDEV,
512 htonl(indev->ifindex)))
513 goto nla_put_failure;
515 physinif = nf_bridge_get_physinif(entskb);
517 nla_put_be32(skb, NFQA_IFINDEX_PHYSINDEV,
519 goto nla_put_failure;
525 #if !IS_ENABLED(CONFIG_BRIDGE_NETFILTER)
526 if (nla_put_be32(skb, NFQA_IFINDEX_OUTDEV, htonl(outdev->ifindex)))
527 goto nla_put_failure;
529 if (entry->state.pf == PF_BRIDGE) {
530 /* Case 1: outdev is physical output device, we need to
531 * look for bridge group (when called from
532 * netfilter_bridge) */
533 if (nla_put_be32(skb, NFQA_IFINDEX_PHYSOUTDEV,
534 htonl(outdev->ifindex)) ||
535 /* this is the bridge group "brX" */
536 /* rcu_read_lock()ed by __nf_queue */
537 nla_put_be32(skb, NFQA_IFINDEX_OUTDEV,
538 htonl(br_port_get_rcu(outdev)->br->dev->ifindex)))
539 goto nla_put_failure;
543 /* Case 2: outdev is bridge group, we need to look for
544 * physical output device (when called from ipv4) */
545 if (nla_put_be32(skb, NFQA_IFINDEX_OUTDEV,
546 htonl(outdev->ifindex)))
547 goto nla_put_failure;
549 physoutif = nf_bridge_get_physoutif(entskb);
551 nla_put_be32(skb, NFQA_IFINDEX_PHYSOUTDEV,
553 goto nla_put_failure;
559 nla_put_be32(skb, NFQA_MARK, htonl(entskb->mark)))
560 goto nla_put_failure;
562 if (indev && entskb->dev &&
563 skb_mac_header_was_set(entskb) &&
564 skb_mac_header_len(entskb) != 0) {
565 struct nfqnl_msg_packet_hw phw;
568 memset(&phw, 0, sizeof(phw));
569 len = dev_parse_header(entskb, phw.hw_addr);
571 phw.hw_addrlen = htons(len);
572 if (nla_put(skb, NFQA_HWADDR, sizeof(phw), &phw))
573 goto nla_put_failure;
577 if (nfqnl_put_bridge(entry, skb) < 0)
578 goto nla_put_failure;
580 if (entry->state.hook <= NF_INET_FORWARD && entskb->tstamp) {
581 struct nfqnl_msg_packet_timestamp ts;
582 struct timespec64 kts = ktime_to_timespec64(entskb->tstamp);
584 ts.sec = cpu_to_be64(kts.tv_sec);
585 ts.usec = cpu_to_be64(kts.tv_nsec / NSEC_PER_USEC);
587 if (nla_put(skb, NFQA_TIMESTAMP, sizeof(ts), &ts))
588 goto nla_put_failure;
591 if ((queue->flags & NFQA_CFG_F_UID_GID) && entskb->sk &&
592 nfqnl_put_sk_uidgid(skb, entskb->sk) < 0)
593 goto nla_put_failure;
595 if (seclen && nla_put(skb, NFQA_SECCTX, seclen, secdata))
596 goto nla_put_failure;
598 if (ct && nfnl_ct->build(skb, ct, ctinfo, NFQA_CT, NFQA_CT_INFO) < 0)
599 goto nla_put_failure;
601 if (cap_len > data_len &&
602 nla_put_be32(skb, NFQA_CAP_LEN, htonl(cap_len)))
603 goto nla_put_failure;
605 if (nfqnl_put_packet_info(skb, entskb, csum_verify))
606 goto nla_put_failure;
611 if (skb_tailroom(skb) < sizeof(*nla) + hlen)
612 goto nla_put_failure;
614 nla = skb_put(skb, sizeof(*nla));
615 nla->nla_type = NFQA_PAYLOAD;
616 nla->nla_len = nla_attr_size(data_len);
618 if (skb_zerocopy(skb, entskb, data_len, hlen))
619 goto nla_put_failure;
622 nlh->nlmsg_len = skb->len;
624 security_release_secctx(secdata, seclen);
628 skb_tx_error(entskb);
630 net_err_ratelimited("nf_queue: error creating packet message\n");
633 security_release_secctx(secdata, seclen);
637 static bool nf_ct_drop_unconfirmed(const struct nf_queue_entry *entry)
639 #if IS_ENABLED(CONFIG_NF_CONNTRACK)
640 static const unsigned long flags = IPS_CONFIRMED | IPS_DYING;
641 const struct nf_conn *ct = (void *)skb_nfct(entry->skb);
643 if (ct && ((ct->status & flags) == IPS_DYING))
650 __nfqnl_enqueue_packet(struct net *net, struct nfqnl_instance *queue,
651 struct nf_queue_entry *entry)
653 struct sk_buff *nskb;
655 __be32 *packet_id_ptr;
658 nskb = nfqnl_build_packet_message(net, queue, entry, &packet_id_ptr);
663 spin_lock_bh(&queue->lock);
665 if (nf_ct_drop_unconfirmed(entry))
666 goto err_out_free_nskb;
668 if (queue->queue_total >= queue->queue_maxlen) {
669 if (queue->flags & NFQA_CFG_F_FAIL_OPEN) {
673 queue->queue_dropped++;
674 net_warn_ratelimited("nf_queue: full at %d entries, dropping packets(s)\n",
677 goto err_out_free_nskb;
679 entry->id = ++queue->id_sequence;
680 *packet_id_ptr = htonl(entry->id);
682 /* nfnetlink_unicast will either free the nskb or add it to a socket */
683 err = nfnetlink_unicast(nskb, net, queue->peer_portid);
685 if (queue->flags & NFQA_CFG_F_FAIL_OPEN) {
689 queue->queue_user_dropped++;
694 __enqueue_entry(queue, entry);
696 spin_unlock_bh(&queue->lock);
702 spin_unlock_bh(&queue->lock);
704 nfqnl_reinject(entry, NF_ACCEPT);
709 static struct nf_queue_entry *
710 nf_queue_entry_dup(struct nf_queue_entry *e)
712 struct nf_queue_entry *entry = kmemdup(e, e->size, GFP_ATOMIC);
717 if (nf_queue_entry_get_refs(entry))
724 #if IS_ENABLED(CONFIG_BRIDGE_NETFILTER)
725 /* When called from bridge netfilter, skb->data must point to MAC header
726 * before calling skb_gso_segment(). Else, original MAC header is lost
727 * and segmented skbs will be sent to wrong destination.
729 static void nf_bridge_adjust_skb_data(struct sk_buff *skb)
731 if (nf_bridge_info_get(skb))
732 __skb_push(skb, skb->network_header - skb->mac_header);
735 static void nf_bridge_adjust_segmented_data(struct sk_buff *skb)
737 if (nf_bridge_info_get(skb))
738 __skb_pull(skb, skb->network_header - skb->mac_header);
741 #define nf_bridge_adjust_skb_data(s) do {} while (0)
742 #define nf_bridge_adjust_segmented_data(s) do {} while (0)
746 __nfqnl_enqueue_packet_gso(struct net *net, struct nfqnl_instance *queue,
747 struct sk_buff *skb, struct nf_queue_entry *entry)
750 struct nf_queue_entry *entry_seg;
752 nf_bridge_adjust_segmented_data(skb);
754 if (skb->next == NULL) { /* last packet, no need to copy entry */
755 struct sk_buff *gso_skb = entry->skb;
757 ret = __nfqnl_enqueue_packet(net, queue, entry);
759 entry->skb = gso_skb;
763 skb_mark_not_on_list(skb);
765 entry_seg = nf_queue_entry_dup(entry);
767 entry_seg->skb = skb;
768 ret = __nfqnl_enqueue_packet(net, queue, entry_seg);
770 nf_queue_entry_free(entry_seg);
776 nfqnl_enqueue_packet(struct nf_queue_entry *entry, unsigned int queuenum)
779 struct nfqnl_instance *queue;
780 struct sk_buff *skb, *segs, *nskb;
782 struct net *net = entry->state.net;
783 struct nfnl_queue_net *q = nfnl_queue_pernet(net);
785 /* rcu_read_lock()ed by nf_hook_thresh */
786 queue = instance_lookup(q, queuenum);
790 if (queue->copy_mode == NFQNL_COPY_NONE)
795 switch (entry->state.pf) {
797 skb->protocol = htons(ETH_P_IP);
800 skb->protocol = htons(ETH_P_IPV6);
804 if ((queue->flags & NFQA_CFG_F_GSO) || !skb_is_gso(skb))
805 return __nfqnl_enqueue_packet(net, queue, entry);
807 nf_bridge_adjust_skb_data(skb);
808 segs = skb_gso_segment(skb, 0);
809 /* Does not use PTR_ERR to limit the number of error codes that can be
810 * returned by nf_queue. For instance, callers rely on -ESRCH to
811 * mean 'ignore this hook'.
813 if (IS_ERR_OR_NULL(segs))
817 skb_list_walk_safe(segs, segs, nskb) {
819 err = __nfqnl_enqueue_packet_gso(net, queue,
828 if (err) /* some segments are already queued */
829 nf_queue_entry_free(entry);
834 nf_bridge_adjust_segmented_data(skb);
839 nfqnl_mangle(void *data, int data_len, struct nf_queue_entry *e, int diff)
841 struct sk_buff *nskb;
844 if (pskb_trim(e->skb, data_len))
846 } else if (diff > 0) {
847 if (data_len > 0xFFFF)
849 if (diff > skb_tailroom(e->skb)) {
850 nskb = skb_copy_expand(e->skb, skb_headroom(e->skb),
857 skb_put(e->skb, diff);
859 if (skb_ensure_writable(e->skb, data_len))
861 skb_copy_to_linear_data(e->skb, data, data_len);
862 e->skb->ip_summed = CHECKSUM_NONE;
867 nfqnl_set_mode(struct nfqnl_instance *queue,
868 unsigned char mode, unsigned int range)
872 spin_lock_bh(&queue->lock);
874 case NFQNL_COPY_NONE:
875 case NFQNL_COPY_META:
876 queue->copy_mode = mode;
877 queue->copy_range = 0;
880 case NFQNL_COPY_PACKET:
881 queue->copy_mode = mode;
882 if (range == 0 || range > NFQNL_MAX_COPY_RANGE)
883 queue->copy_range = NFQNL_MAX_COPY_RANGE;
885 queue->copy_range = range;
892 spin_unlock_bh(&queue->lock);
898 dev_cmp(struct nf_queue_entry *entry, unsigned long ifindex)
900 #if IS_ENABLED(CONFIG_BRIDGE_NETFILTER)
901 int physinif, physoutif;
903 physinif = nf_bridge_get_physinif(entry->skb);
904 physoutif = nf_bridge_get_physoutif(entry->skb);
906 if (physinif == ifindex || physoutif == ifindex)
910 if (entry->state.in->ifindex == ifindex)
912 if (entry->state.out)
913 if (entry->state.out->ifindex == ifindex)
919 /* drop all packets with either indev or outdev == ifindex from all queue
922 nfqnl_dev_drop(struct net *net, int ifindex)
925 struct nfnl_queue_net *q = nfnl_queue_pernet(net);
929 for (i = 0; i < INSTANCE_BUCKETS; i++) {
930 struct nfqnl_instance *inst;
931 struct hlist_head *head = &q->instance_table[i];
933 hlist_for_each_entry_rcu(inst, head, hlist)
934 nfqnl_flush(inst, dev_cmp, ifindex);
941 nfqnl_rcv_dev_event(struct notifier_block *this,
942 unsigned long event, void *ptr)
944 struct net_device *dev = netdev_notifier_info_to_dev(ptr);
946 /* Drop any packets associated with the downed device */
947 if (event == NETDEV_DOWN)
948 nfqnl_dev_drop(dev_net(dev), dev->ifindex);
952 static struct notifier_block nfqnl_dev_notifier = {
953 .notifier_call = nfqnl_rcv_dev_event,
956 static void nfqnl_nf_hook_drop(struct net *net)
958 struct nfnl_queue_net *q = nfnl_queue_pernet(net);
961 /* This function is also called on net namespace error unwind,
962 * when pernet_ops->init() failed and ->exit() functions of the
963 * previous pernet_ops gets called.
965 * This may result in a call to nfqnl_nf_hook_drop() before
966 * struct nfnl_queue_net was allocated.
971 for (i = 0; i < INSTANCE_BUCKETS; i++) {
972 struct nfqnl_instance *inst;
973 struct hlist_head *head = &q->instance_table[i];
975 hlist_for_each_entry_rcu(inst, head, hlist)
976 nfqnl_flush(inst, NULL, 0);
981 nfqnl_rcv_nl_event(struct notifier_block *this,
982 unsigned long event, void *ptr)
984 struct netlink_notify *n = ptr;
985 struct nfnl_queue_net *q = nfnl_queue_pernet(n->net);
987 if (event == NETLINK_URELEASE && n->protocol == NETLINK_NETFILTER) {
990 /* destroy all instances for this portid */
991 spin_lock(&q->instances_lock);
992 for (i = 0; i < INSTANCE_BUCKETS; i++) {
993 struct hlist_node *t2;
994 struct nfqnl_instance *inst;
995 struct hlist_head *head = &q->instance_table[i];
997 hlist_for_each_entry_safe(inst, t2, head, hlist) {
998 if (n->portid == inst->peer_portid)
999 __instance_destroy(inst);
1002 spin_unlock(&q->instances_lock);
1007 static struct notifier_block nfqnl_rtnl_notifier = {
1008 .notifier_call = nfqnl_rcv_nl_event,
1011 static const struct nla_policy nfqa_vlan_policy[NFQA_VLAN_MAX + 1] = {
1012 [NFQA_VLAN_TCI] = { .type = NLA_U16},
1013 [NFQA_VLAN_PROTO] = { .type = NLA_U16},
1016 static const struct nla_policy nfqa_verdict_policy[NFQA_MAX+1] = {
1017 [NFQA_VERDICT_HDR] = { .len = sizeof(struct nfqnl_msg_verdict_hdr) },
1018 [NFQA_MARK] = { .type = NLA_U32 },
1019 [NFQA_PAYLOAD] = { .type = NLA_UNSPEC },
1020 [NFQA_CT] = { .type = NLA_UNSPEC },
1021 [NFQA_EXP] = { .type = NLA_UNSPEC },
1022 [NFQA_VLAN] = { .type = NLA_NESTED },
1025 static const struct nla_policy nfqa_verdict_batch_policy[NFQA_MAX+1] = {
1026 [NFQA_VERDICT_HDR] = { .len = sizeof(struct nfqnl_msg_verdict_hdr) },
1027 [NFQA_MARK] = { .type = NLA_U32 },
1030 static struct nfqnl_instance *
1031 verdict_instance_lookup(struct nfnl_queue_net *q, u16 queue_num, u32 nlportid)
1033 struct nfqnl_instance *queue;
1035 queue = instance_lookup(q, queue_num);
1037 return ERR_PTR(-ENODEV);
1039 if (queue->peer_portid != nlportid)
1040 return ERR_PTR(-EPERM);
1045 static struct nfqnl_msg_verdict_hdr*
1046 verdicthdr_get(const struct nlattr * const nfqa[])
1048 struct nfqnl_msg_verdict_hdr *vhdr;
1049 unsigned int verdict;
1051 if (!nfqa[NFQA_VERDICT_HDR])
1054 vhdr = nla_data(nfqa[NFQA_VERDICT_HDR]);
1055 verdict = ntohl(vhdr->verdict) & NF_VERDICT_MASK;
1056 if (verdict > NF_MAX_VERDICT || verdict == NF_STOLEN)
1061 static int nfq_id_after(unsigned int id, unsigned int max)
1063 return (int)(id - max) > 0;
1066 static int nfqnl_recv_verdict_batch(struct sk_buff *skb,
1067 const struct nfnl_info *info,
1068 const struct nlattr * const nfqa[])
1070 struct nfnl_queue_net *q = nfnl_queue_pernet(info->net);
1071 u16 queue_num = ntohs(info->nfmsg->res_id);
1072 struct nf_queue_entry *entry, *tmp;
1073 struct nfqnl_msg_verdict_hdr *vhdr;
1074 struct nfqnl_instance *queue;
1075 unsigned int verdict, maxid;
1076 LIST_HEAD(batch_list);
1078 queue = verdict_instance_lookup(q, queue_num,
1079 NETLINK_CB(skb).portid);
1081 return PTR_ERR(queue);
1083 vhdr = verdicthdr_get(nfqa);
1087 verdict = ntohl(vhdr->verdict);
1088 maxid = ntohl(vhdr->id);
1090 spin_lock_bh(&queue->lock);
1092 list_for_each_entry_safe(entry, tmp, &queue->queue_list, list) {
1093 if (nfq_id_after(entry->id, maxid))
1095 __dequeue_entry(queue, entry);
1096 list_add_tail(&entry->list, &batch_list);
1099 spin_unlock_bh(&queue->lock);
1101 if (list_empty(&batch_list))
1104 list_for_each_entry_safe(entry, tmp, &batch_list, list) {
1105 if (nfqa[NFQA_MARK])
1106 entry->skb->mark = ntohl(nla_get_be32(nfqa[NFQA_MARK]));
1108 nfqnl_reinject(entry, verdict);
1113 static struct nf_conn *nfqnl_ct_parse(const struct nfnl_ct_hook *nfnl_ct,
1114 const struct nlmsghdr *nlh,
1115 const struct nlattr * const nfqa[],
1116 struct nf_queue_entry *entry,
1117 enum ip_conntrack_info *ctinfo)
1119 #if IS_ENABLED(CONFIG_NF_CONNTRACK)
1122 ct = nf_ct_get(entry->skb, ctinfo);
1126 if (nfnl_ct->parse(nfqa[NFQA_CT], ct) < 0)
1130 nfnl_ct->attach_expect(nfqa[NFQA_EXP], ct,
1131 NETLINK_CB(entry->skb).portid,
1139 static int nfqa_parse_bridge(struct nf_queue_entry *entry,
1140 const struct nlattr * const nfqa[])
1142 if (nfqa[NFQA_VLAN]) {
1143 struct nlattr *tb[NFQA_VLAN_MAX + 1];
1146 err = nla_parse_nested_deprecated(tb, NFQA_VLAN_MAX,
1148 nfqa_vlan_policy, NULL);
1152 if (!tb[NFQA_VLAN_TCI] || !tb[NFQA_VLAN_PROTO])
1155 __vlan_hwaccel_put_tag(entry->skb,
1156 nla_get_be16(tb[NFQA_VLAN_PROTO]),
1157 ntohs(nla_get_be16(tb[NFQA_VLAN_TCI])));
1160 if (nfqa[NFQA_L2HDR]) {
1161 int mac_header_len = entry->skb->network_header -
1162 entry->skb->mac_header;
1164 if (mac_header_len != nla_len(nfqa[NFQA_L2HDR]))
1166 else if (mac_header_len > 0)
1167 memcpy(skb_mac_header(entry->skb),
1168 nla_data(nfqa[NFQA_L2HDR]),
1175 static int nfqnl_recv_verdict(struct sk_buff *skb, const struct nfnl_info *info,
1176 const struct nlattr * const nfqa[])
1178 struct nfnl_queue_net *q = nfnl_queue_pernet(info->net);
1179 u_int16_t queue_num = ntohs(info->nfmsg->res_id);
1180 const struct nfnl_ct_hook *nfnl_ct;
1181 struct nfqnl_msg_verdict_hdr *vhdr;
1182 enum ip_conntrack_info ctinfo;
1183 struct nfqnl_instance *queue;
1184 struct nf_queue_entry *entry;
1185 struct nf_conn *ct = NULL;
1186 unsigned int verdict;
1189 queue = verdict_instance_lookup(q, queue_num,
1190 NETLINK_CB(skb).portid);
1192 return PTR_ERR(queue);
1194 vhdr = verdicthdr_get(nfqa);
1198 verdict = ntohl(vhdr->verdict);
1200 entry = find_dequeue_entry(queue, ntohl(vhdr->id));
1204 /* rcu lock already held from nfnl->call_rcu. */
1205 nfnl_ct = rcu_dereference(nfnl_ct_hook);
1207 if (nfqa[NFQA_CT]) {
1208 if (nfnl_ct != NULL)
1209 ct = nfqnl_ct_parse(nfnl_ct, info->nlh, nfqa, entry,
1213 if (entry->state.pf == PF_BRIDGE) {
1214 err = nfqa_parse_bridge(entry, nfqa);
1219 if (nfqa[NFQA_PAYLOAD]) {
1220 u16 payload_len = nla_len(nfqa[NFQA_PAYLOAD]);
1221 int diff = payload_len - entry->skb->len;
1223 if (nfqnl_mangle(nla_data(nfqa[NFQA_PAYLOAD]),
1224 payload_len, entry, diff) < 0)
1228 nfnl_ct->seq_adjust(entry->skb, ct, ctinfo, diff);
1231 if (nfqa[NFQA_MARK])
1232 entry->skb->mark = ntohl(nla_get_be32(nfqa[NFQA_MARK]));
1234 nfqnl_reinject(entry, verdict);
1238 static int nfqnl_recv_unsupp(struct sk_buff *skb, const struct nfnl_info *info,
1239 const struct nlattr * const cda[])
1244 static const struct nla_policy nfqa_cfg_policy[NFQA_CFG_MAX+1] = {
1245 [NFQA_CFG_CMD] = { .len = sizeof(struct nfqnl_msg_config_cmd) },
1246 [NFQA_CFG_PARAMS] = { .len = sizeof(struct nfqnl_msg_config_params) },
1247 [NFQA_CFG_QUEUE_MAXLEN] = { .type = NLA_U32 },
1248 [NFQA_CFG_MASK] = { .type = NLA_U32 },
1249 [NFQA_CFG_FLAGS] = { .type = NLA_U32 },
1252 static const struct nf_queue_handler nfqh = {
1253 .outfn = nfqnl_enqueue_packet,
1254 .nf_hook_drop = nfqnl_nf_hook_drop,
1257 static int nfqnl_recv_config(struct sk_buff *skb, const struct nfnl_info *info,
1258 const struct nlattr * const nfqa[])
1260 struct nfnl_queue_net *q = nfnl_queue_pernet(info->net);
1261 u_int16_t queue_num = ntohs(info->nfmsg->res_id);
1262 struct nfqnl_msg_config_cmd *cmd = NULL;
1263 struct nfqnl_instance *queue;
1264 __u32 flags = 0, mask = 0;
1267 if (nfqa[NFQA_CFG_CMD]) {
1268 cmd = nla_data(nfqa[NFQA_CFG_CMD]);
1270 /* Obsolete commands without queue context */
1271 switch (cmd->command) {
1272 case NFQNL_CFG_CMD_PF_BIND: return 0;
1273 case NFQNL_CFG_CMD_PF_UNBIND: return 0;
1277 /* Check if we support these flags in first place, dependencies should
1278 * be there too not to break atomicity.
1280 if (nfqa[NFQA_CFG_FLAGS]) {
1281 if (!nfqa[NFQA_CFG_MASK]) {
1282 /* A mask is needed to specify which flags are being
1288 flags = ntohl(nla_get_be32(nfqa[NFQA_CFG_FLAGS]));
1289 mask = ntohl(nla_get_be32(nfqa[NFQA_CFG_MASK]));
1291 if (flags >= NFQA_CFG_F_MAX)
1294 #if !IS_ENABLED(CONFIG_NETWORK_SECMARK)
1295 if (flags & mask & NFQA_CFG_F_SECCTX)
1298 if ((flags & mask & NFQA_CFG_F_CONNTRACK) &&
1299 !rcu_access_pointer(nfnl_ct_hook)) {
1300 #ifdef CONFIG_MODULES
1301 nfnl_unlock(NFNL_SUBSYS_QUEUE);
1302 request_module("ip_conntrack_netlink");
1303 nfnl_lock(NFNL_SUBSYS_QUEUE);
1304 if (rcu_access_pointer(nfnl_ct_hook))
1312 queue = instance_lookup(q, queue_num);
1313 if (queue && queue->peer_portid != NETLINK_CB(skb).portid) {
1315 goto err_out_unlock;
1319 switch (cmd->command) {
1320 case NFQNL_CFG_CMD_BIND:
1323 goto err_out_unlock;
1325 queue = instance_create(q, queue_num,
1326 NETLINK_CB(skb).portid);
1327 if (IS_ERR(queue)) {
1328 ret = PTR_ERR(queue);
1329 goto err_out_unlock;
1332 case NFQNL_CFG_CMD_UNBIND:
1335 goto err_out_unlock;
1337 instance_destroy(q, queue);
1338 goto err_out_unlock;
1339 case NFQNL_CFG_CMD_PF_BIND:
1340 case NFQNL_CFG_CMD_PF_UNBIND:
1344 goto err_out_unlock;
1350 goto err_out_unlock;
1353 if (nfqa[NFQA_CFG_PARAMS]) {
1354 struct nfqnl_msg_config_params *params =
1355 nla_data(nfqa[NFQA_CFG_PARAMS]);
1357 nfqnl_set_mode(queue, params->copy_mode,
1358 ntohl(params->copy_range));
1361 if (nfqa[NFQA_CFG_QUEUE_MAXLEN]) {
1362 __be32 *queue_maxlen = nla_data(nfqa[NFQA_CFG_QUEUE_MAXLEN]);
1364 spin_lock_bh(&queue->lock);
1365 queue->queue_maxlen = ntohl(*queue_maxlen);
1366 spin_unlock_bh(&queue->lock);
1369 if (nfqa[NFQA_CFG_FLAGS]) {
1370 spin_lock_bh(&queue->lock);
1371 queue->flags &= ~mask;
1372 queue->flags |= flags & mask;
1373 spin_unlock_bh(&queue->lock);
1381 static const struct nfnl_callback nfqnl_cb[NFQNL_MSG_MAX] = {
1382 [NFQNL_MSG_PACKET] = {
1383 .call = nfqnl_recv_unsupp,
1384 .type = NFNL_CB_RCU,
1385 .attr_count = NFQA_MAX,
1387 [NFQNL_MSG_VERDICT] = {
1388 .call = nfqnl_recv_verdict,
1389 .type = NFNL_CB_RCU,
1390 .attr_count = NFQA_MAX,
1391 .policy = nfqa_verdict_policy
1393 [NFQNL_MSG_CONFIG] = {
1394 .call = nfqnl_recv_config,
1395 .type = NFNL_CB_MUTEX,
1396 .attr_count = NFQA_CFG_MAX,
1397 .policy = nfqa_cfg_policy
1399 [NFQNL_MSG_VERDICT_BATCH] = {
1400 .call = nfqnl_recv_verdict_batch,
1401 .type = NFNL_CB_RCU,
1402 .attr_count = NFQA_MAX,
1403 .policy = nfqa_verdict_batch_policy
1407 static const struct nfnetlink_subsystem nfqnl_subsys = {
1409 .subsys_id = NFNL_SUBSYS_QUEUE,
1410 .cb_count = NFQNL_MSG_MAX,
1414 #ifdef CONFIG_PROC_FS
1416 struct seq_net_private p;
1417 unsigned int bucket;
1420 static struct hlist_node *get_first(struct seq_file *seq)
1422 struct iter_state *st = seq->private;
1424 struct nfnl_queue_net *q;
1429 net = seq_file_net(seq);
1430 q = nfnl_queue_pernet(net);
1431 for (st->bucket = 0; st->bucket < INSTANCE_BUCKETS; st->bucket++) {
1432 if (!hlist_empty(&q->instance_table[st->bucket]))
1433 return q->instance_table[st->bucket].first;
1438 static struct hlist_node *get_next(struct seq_file *seq, struct hlist_node *h)
1440 struct iter_state *st = seq->private;
1441 struct net *net = seq_file_net(seq);
1445 struct nfnl_queue_net *q;
1447 if (++st->bucket >= INSTANCE_BUCKETS)
1450 q = nfnl_queue_pernet(net);
1451 h = q->instance_table[st->bucket].first;
1456 static struct hlist_node *get_idx(struct seq_file *seq, loff_t pos)
1458 struct hlist_node *head;
1459 head = get_first(seq);
1462 while (pos && (head = get_next(seq, head)))
1464 return pos ? NULL : head;
1467 static void *seq_start(struct seq_file *s, loff_t *pos)
1468 __acquires(nfnl_queue_pernet(seq_file_net(s))->instances_lock)
1470 spin_lock(&nfnl_queue_pernet(seq_file_net(s))->instances_lock);
1471 return get_idx(s, *pos);
1474 static void *seq_next(struct seq_file *s, void *v, loff_t *pos)
1477 return get_next(s, v);
1480 static void seq_stop(struct seq_file *s, void *v)
1481 __releases(nfnl_queue_pernet(seq_file_net(s))->instances_lock)
1483 spin_unlock(&nfnl_queue_pernet(seq_file_net(s))->instances_lock);
1486 static int seq_show(struct seq_file *s, void *v)
1488 const struct nfqnl_instance *inst = v;
1490 seq_printf(s, "%5u %6u %5u %1u %5u %5u %5u %8u %2d\n",
1492 inst->peer_portid, inst->queue_total,
1493 inst->copy_mode, inst->copy_range,
1494 inst->queue_dropped, inst->queue_user_dropped,
1495 inst->id_sequence, 1);
1499 static const struct seq_operations nfqnl_seq_ops = {
1505 #endif /* PROC_FS */
1507 static int __net_init nfnl_queue_net_init(struct net *net)
1510 struct nfnl_queue_net *q = nfnl_queue_pernet(net);
1512 for (i = 0; i < INSTANCE_BUCKETS; i++)
1513 INIT_HLIST_HEAD(&q->instance_table[i]);
1515 spin_lock_init(&q->instances_lock);
1517 #ifdef CONFIG_PROC_FS
1518 if (!proc_create_net("nfnetlink_queue", 0440, net->nf.proc_netfilter,
1519 &nfqnl_seq_ops, sizeof(struct iter_state)))
1525 static void __net_exit nfnl_queue_net_exit(struct net *net)
1527 struct nfnl_queue_net *q = nfnl_queue_pernet(net);
1530 #ifdef CONFIG_PROC_FS
1531 remove_proc_entry("nfnetlink_queue", net->nf.proc_netfilter);
1533 for (i = 0; i < INSTANCE_BUCKETS; i++)
1534 WARN_ON_ONCE(!hlist_empty(&q->instance_table[i]));
1537 static struct pernet_operations nfnl_queue_net_ops = {
1538 .init = nfnl_queue_net_init,
1539 .exit = nfnl_queue_net_exit,
1540 .id = &nfnl_queue_net_id,
1541 .size = sizeof(struct nfnl_queue_net),
1544 static int __init nfnetlink_queue_init(void)
1548 status = register_pernet_subsys(&nfnl_queue_net_ops);
1550 pr_err("failed to register pernet ops\n");
1554 netlink_register_notifier(&nfqnl_rtnl_notifier);
1555 status = nfnetlink_subsys_register(&nfqnl_subsys);
1557 pr_err("failed to create netlink socket\n");
1558 goto cleanup_netlink_notifier;
1561 status = register_netdevice_notifier(&nfqnl_dev_notifier);
1563 pr_err("failed to register netdevice notifier\n");
1564 goto cleanup_netlink_subsys;
1567 nf_register_queue_handler(&nfqh);
1571 cleanup_netlink_subsys:
1572 nfnetlink_subsys_unregister(&nfqnl_subsys);
1573 cleanup_netlink_notifier:
1574 netlink_unregister_notifier(&nfqnl_rtnl_notifier);
1575 unregister_pernet_subsys(&nfnl_queue_net_ops);
1580 static void __exit nfnetlink_queue_fini(void)
1582 nf_unregister_queue_handler();
1583 unregister_netdevice_notifier(&nfqnl_dev_notifier);
1584 nfnetlink_subsys_unregister(&nfqnl_subsys);
1585 netlink_unregister_notifier(&nfqnl_rtnl_notifier);
1586 unregister_pernet_subsys(&nfnl_queue_net_ops);
1588 rcu_barrier(); /* Wait for completion of call_rcu()'s */
1591 MODULE_DESCRIPTION("netfilter packet queue handler");
1592 MODULE_AUTHOR("Harald Welte <laforge@netfilter.org>");
1593 MODULE_LICENSE("GPL");
1594 MODULE_ALIAS_NFNL_SUBSYS(NFNL_SUBSYS_QUEUE);
1596 module_init(nfnetlink_queue_init);
1597 module_exit(nfnetlink_queue_fini);