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;
397 size = nlmsg_total_size(sizeof(struct nfgenmsg))
398 + nla_total_size(sizeof(struct nfqnl_msg_packet_hdr))
399 + nla_total_size(sizeof(u_int32_t)) /* ifindex */
400 + nla_total_size(sizeof(u_int32_t)) /* ifindex */
401 #if IS_ENABLED(CONFIG_BRIDGE_NETFILTER)
402 + nla_total_size(sizeof(u_int32_t)) /* ifindex */
403 + nla_total_size(sizeof(u_int32_t)) /* ifindex */
405 + nla_total_size(sizeof(u_int32_t)) /* mark */
406 + nla_total_size(sizeof(u_int32_t)) /* priority */
407 + nla_total_size(sizeof(struct nfqnl_msg_packet_hw))
408 + nla_total_size(sizeof(u_int32_t)) /* skbinfo */
409 + nla_total_size(sizeof(u_int32_t)); /* cap_len */
411 tstamp = skb_tstamp_cond(entskb, false);
413 size += nla_total_size(sizeof(struct nfqnl_msg_packet_timestamp));
415 size += nfqnl_get_bridge_size(entry);
417 if (entry->state.hook <= NF_INET_FORWARD ||
418 (entry->state.hook == NF_INET_POST_ROUTING && entskb->sk == NULL))
419 csum_verify = !skb_csum_unnecessary(entskb);
423 outdev = entry->state.out;
425 switch ((enum nfqnl_config_mode)READ_ONCE(queue->copy_mode)) {
426 case NFQNL_COPY_META:
427 case NFQNL_COPY_NONE:
430 case NFQNL_COPY_PACKET:
431 if (!(queue->flags & NFQA_CFG_F_GSO) &&
432 entskb->ip_summed == CHECKSUM_PARTIAL &&
433 skb_checksum_help(entskb))
436 data_len = READ_ONCE(queue->copy_range);
437 if (data_len > entskb->len)
438 data_len = entskb->len;
440 hlen = skb_zerocopy_headlen(entskb);
441 hlen = min_t(unsigned int, hlen, data_len);
442 size += sizeof(struct nlattr) + hlen;
443 cap_len = entskb->len;
447 nfnl_ct = rcu_dereference(nfnl_ct_hook);
449 #if IS_ENABLED(CONFIG_NF_CONNTRACK)
450 if (queue->flags & NFQA_CFG_F_CONNTRACK) {
451 if (nfnl_ct != NULL) {
452 ct = nf_ct_get(entskb, &ctinfo);
454 size += nfnl_ct->build_size(ct);
459 if (queue->flags & NFQA_CFG_F_UID_GID) {
460 size += (nla_total_size(sizeof(u_int32_t)) /* uid */
461 + nla_total_size(sizeof(u_int32_t))); /* gid */
464 if ((queue->flags & NFQA_CFG_F_SECCTX) && entskb->sk) {
465 seclen = nfqnl_get_sk_secctx(entskb, &secdata);
467 size += nla_total_size(seclen);
470 skb = alloc_skb(size, GFP_ATOMIC);
472 skb_tx_error(entskb);
476 nlh = nfnl_msg_put(skb, 0, 0,
477 nfnl_msg_type(NFNL_SUBSYS_QUEUE, NFQNL_MSG_PACKET),
478 0, entry->state.pf, NFNETLINK_V0,
479 htons(queue->queue_num));
481 skb_tx_error(entskb);
486 nla = __nla_reserve(skb, NFQA_PACKET_HDR, sizeof(*pmsg));
487 pmsg = nla_data(nla);
488 pmsg->hw_protocol = entskb->protocol;
489 pmsg->hook = entry->state.hook;
490 *packet_id_ptr = &pmsg->packet_id;
492 indev = entry->state.in;
494 #if !IS_ENABLED(CONFIG_BRIDGE_NETFILTER)
495 if (nla_put_be32(skb, NFQA_IFINDEX_INDEV, htonl(indev->ifindex)))
496 goto nla_put_failure;
498 if (entry->state.pf == PF_BRIDGE) {
499 /* Case 1: indev is physical input device, we need to
500 * look for bridge group (when called from
501 * netfilter_bridge) */
502 if (nla_put_be32(skb, NFQA_IFINDEX_PHYSINDEV,
503 htonl(indev->ifindex)) ||
504 /* this is the bridge group "brX" */
505 /* rcu_read_lock()ed by __nf_queue */
506 nla_put_be32(skb, NFQA_IFINDEX_INDEV,
507 htonl(br_port_get_rcu(indev)->br->dev->ifindex)))
508 goto nla_put_failure;
512 /* Case 2: indev is bridge group, we need to look for
513 * physical device (when called from ipv4) */
514 if (nla_put_be32(skb, NFQA_IFINDEX_INDEV,
515 htonl(indev->ifindex)))
516 goto nla_put_failure;
518 physinif = nf_bridge_get_physinif(entskb);
520 nla_put_be32(skb, NFQA_IFINDEX_PHYSINDEV,
522 goto nla_put_failure;
528 #if !IS_ENABLED(CONFIG_BRIDGE_NETFILTER)
529 if (nla_put_be32(skb, NFQA_IFINDEX_OUTDEV, htonl(outdev->ifindex)))
530 goto nla_put_failure;
532 if (entry->state.pf == PF_BRIDGE) {
533 /* Case 1: outdev is physical output device, we need to
534 * look for bridge group (when called from
535 * netfilter_bridge) */
536 if (nla_put_be32(skb, NFQA_IFINDEX_PHYSOUTDEV,
537 htonl(outdev->ifindex)) ||
538 /* this is the bridge group "brX" */
539 /* rcu_read_lock()ed by __nf_queue */
540 nla_put_be32(skb, NFQA_IFINDEX_OUTDEV,
541 htonl(br_port_get_rcu(outdev)->br->dev->ifindex)))
542 goto nla_put_failure;
546 /* Case 2: outdev is bridge group, we need to look for
547 * physical output device (when called from ipv4) */
548 if (nla_put_be32(skb, NFQA_IFINDEX_OUTDEV,
549 htonl(outdev->ifindex)))
550 goto nla_put_failure;
552 physoutif = nf_bridge_get_physoutif(entskb);
554 nla_put_be32(skb, NFQA_IFINDEX_PHYSOUTDEV,
556 goto nla_put_failure;
562 nla_put_be32(skb, NFQA_MARK, htonl(entskb->mark)))
563 goto nla_put_failure;
565 if (entskb->priority &&
566 nla_put_be32(skb, NFQA_PRIORITY, htonl(entskb->priority)))
567 goto nla_put_failure;
569 if (indev && entskb->dev &&
570 skb_mac_header_was_set(entskb) &&
571 skb_mac_header_len(entskb) != 0) {
572 struct nfqnl_msg_packet_hw phw;
575 memset(&phw, 0, sizeof(phw));
576 len = dev_parse_header(entskb, phw.hw_addr);
578 phw.hw_addrlen = htons(len);
579 if (nla_put(skb, NFQA_HWADDR, sizeof(phw), &phw))
580 goto nla_put_failure;
584 if (nfqnl_put_bridge(entry, skb) < 0)
585 goto nla_put_failure;
587 if (entry->state.hook <= NF_INET_FORWARD && tstamp) {
588 struct nfqnl_msg_packet_timestamp ts;
589 struct timespec64 kts = ktime_to_timespec64(tstamp);
591 ts.sec = cpu_to_be64(kts.tv_sec);
592 ts.usec = cpu_to_be64(kts.tv_nsec / NSEC_PER_USEC);
594 if (nla_put(skb, NFQA_TIMESTAMP, sizeof(ts), &ts))
595 goto nla_put_failure;
598 if ((queue->flags & NFQA_CFG_F_UID_GID) && entskb->sk &&
599 nfqnl_put_sk_uidgid(skb, entskb->sk) < 0)
600 goto nla_put_failure;
602 if (seclen && nla_put(skb, NFQA_SECCTX, seclen, secdata))
603 goto nla_put_failure;
605 if (ct && nfnl_ct->build(skb, ct, ctinfo, NFQA_CT, NFQA_CT_INFO) < 0)
606 goto nla_put_failure;
608 if (cap_len > data_len &&
609 nla_put_be32(skb, NFQA_CAP_LEN, htonl(cap_len)))
610 goto nla_put_failure;
612 if (nfqnl_put_packet_info(skb, entskb, csum_verify))
613 goto nla_put_failure;
618 if (skb_tailroom(skb) < sizeof(*nla) + hlen)
619 goto nla_put_failure;
621 nla = skb_put(skb, sizeof(*nla));
622 nla->nla_type = NFQA_PAYLOAD;
623 nla->nla_len = nla_attr_size(data_len);
625 if (skb_zerocopy(skb, entskb, data_len, hlen))
626 goto nla_put_failure;
629 nlh->nlmsg_len = skb->len;
631 security_release_secctx(secdata, seclen);
635 skb_tx_error(entskb);
637 net_err_ratelimited("nf_queue: error creating packet message\n");
640 security_release_secctx(secdata, seclen);
644 static bool nf_ct_drop_unconfirmed(const struct nf_queue_entry *entry)
646 #if IS_ENABLED(CONFIG_NF_CONNTRACK)
647 static const unsigned long flags = IPS_CONFIRMED | IPS_DYING;
648 const struct nf_conn *ct = (void *)skb_nfct(entry->skb);
650 if (ct && ((ct->status & flags) == IPS_DYING))
657 __nfqnl_enqueue_packet(struct net *net, struct nfqnl_instance *queue,
658 struct nf_queue_entry *entry)
660 struct sk_buff *nskb;
662 __be32 *packet_id_ptr;
665 nskb = nfqnl_build_packet_message(net, queue, entry, &packet_id_ptr);
670 spin_lock_bh(&queue->lock);
672 if (nf_ct_drop_unconfirmed(entry))
673 goto err_out_free_nskb;
675 if (queue->queue_total >= queue->queue_maxlen) {
676 if (queue->flags & NFQA_CFG_F_FAIL_OPEN) {
680 queue->queue_dropped++;
681 net_warn_ratelimited("nf_queue: full at %d entries, dropping packets(s)\n",
684 goto err_out_free_nskb;
686 entry->id = ++queue->id_sequence;
687 *packet_id_ptr = htonl(entry->id);
689 /* nfnetlink_unicast will either free the nskb or add it to a socket */
690 err = nfnetlink_unicast(nskb, net, queue->peer_portid);
692 if (queue->flags & NFQA_CFG_F_FAIL_OPEN) {
696 queue->queue_user_dropped++;
701 __enqueue_entry(queue, entry);
703 spin_unlock_bh(&queue->lock);
709 spin_unlock_bh(&queue->lock);
711 nfqnl_reinject(entry, NF_ACCEPT);
716 static struct nf_queue_entry *
717 nf_queue_entry_dup(struct nf_queue_entry *e)
719 struct nf_queue_entry *entry = kmemdup(e, e->size, GFP_ATOMIC);
724 if (nf_queue_entry_get_refs(entry))
731 #if IS_ENABLED(CONFIG_BRIDGE_NETFILTER)
732 /* When called from bridge netfilter, skb->data must point to MAC header
733 * before calling skb_gso_segment(). Else, original MAC header is lost
734 * and segmented skbs will be sent to wrong destination.
736 static void nf_bridge_adjust_skb_data(struct sk_buff *skb)
738 if (nf_bridge_info_get(skb))
739 __skb_push(skb, skb->network_header - skb->mac_header);
742 static void nf_bridge_adjust_segmented_data(struct sk_buff *skb)
744 if (nf_bridge_info_get(skb))
745 __skb_pull(skb, skb->network_header - skb->mac_header);
748 #define nf_bridge_adjust_skb_data(s) do {} while (0)
749 #define nf_bridge_adjust_segmented_data(s) do {} while (0)
753 __nfqnl_enqueue_packet_gso(struct net *net, struct nfqnl_instance *queue,
754 struct sk_buff *skb, struct nf_queue_entry *entry)
757 struct nf_queue_entry *entry_seg;
759 nf_bridge_adjust_segmented_data(skb);
761 if (skb->next == NULL) { /* last packet, no need to copy entry */
762 struct sk_buff *gso_skb = entry->skb;
764 ret = __nfqnl_enqueue_packet(net, queue, entry);
766 entry->skb = gso_skb;
770 skb_mark_not_on_list(skb);
772 entry_seg = nf_queue_entry_dup(entry);
774 entry_seg->skb = skb;
775 ret = __nfqnl_enqueue_packet(net, queue, entry_seg);
777 nf_queue_entry_free(entry_seg);
783 nfqnl_enqueue_packet(struct nf_queue_entry *entry, unsigned int queuenum)
786 struct nfqnl_instance *queue;
787 struct sk_buff *skb, *segs, *nskb;
789 struct net *net = entry->state.net;
790 struct nfnl_queue_net *q = nfnl_queue_pernet(net);
792 /* rcu_read_lock()ed by nf_hook_thresh */
793 queue = instance_lookup(q, queuenum);
797 if (queue->copy_mode == NFQNL_COPY_NONE)
802 switch (entry->state.pf) {
804 skb->protocol = htons(ETH_P_IP);
807 skb->protocol = htons(ETH_P_IPV6);
811 if ((queue->flags & NFQA_CFG_F_GSO) || !skb_is_gso(skb))
812 return __nfqnl_enqueue_packet(net, queue, entry);
814 nf_bridge_adjust_skb_data(skb);
815 segs = skb_gso_segment(skb, 0);
816 /* Does not use PTR_ERR to limit the number of error codes that can be
817 * returned by nf_queue. For instance, callers rely on -ESRCH to
818 * mean 'ignore this hook'.
820 if (IS_ERR_OR_NULL(segs))
824 skb_list_walk_safe(segs, segs, nskb) {
826 err = __nfqnl_enqueue_packet_gso(net, queue,
835 if (err) /* some segments are already queued */
836 nf_queue_entry_free(entry);
841 nf_bridge_adjust_segmented_data(skb);
846 nfqnl_mangle(void *data, int data_len, struct nf_queue_entry *e, int diff)
848 struct sk_buff *nskb;
851 if (pskb_trim(e->skb, data_len))
853 } else if (diff > 0) {
854 if (data_len > 0xFFFF)
856 if (diff > skb_tailroom(e->skb)) {
857 nskb = skb_copy_expand(e->skb, skb_headroom(e->skb),
864 skb_put(e->skb, diff);
866 if (skb_ensure_writable(e->skb, data_len))
868 skb_copy_to_linear_data(e->skb, data, data_len);
869 e->skb->ip_summed = CHECKSUM_NONE;
874 nfqnl_set_mode(struct nfqnl_instance *queue,
875 unsigned char mode, unsigned int range)
879 spin_lock_bh(&queue->lock);
881 case NFQNL_COPY_NONE:
882 case NFQNL_COPY_META:
883 queue->copy_mode = mode;
884 queue->copy_range = 0;
887 case NFQNL_COPY_PACKET:
888 queue->copy_mode = mode;
889 if (range == 0 || range > NFQNL_MAX_COPY_RANGE)
890 queue->copy_range = NFQNL_MAX_COPY_RANGE;
892 queue->copy_range = range;
899 spin_unlock_bh(&queue->lock);
905 dev_cmp(struct nf_queue_entry *entry, unsigned long ifindex)
907 #if IS_ENABLED(CONFIG_BRIDGE_NETFILTER)
908 int physinif, physoutif;
910 physinif = nf_bridge_get_physinif(entry->skb);
911 physoutif = nf_bridge_get_physoutif(entry->skb);
913 if (physinif == ifindex || physoutif == ifindex)
917 if (entry->state.in->ifindex == ifindex)
919 if (entry->state.out)
920 if (entry->state.out->ifindex == ifindex)
926 /* drop all packets with either indev or outdev == ifindex from all queue
929 nfqnl_dev_drop(struct net *net, int ifindex)
932 struct nfnl_queue_net *q = nfnl_queue_pernet(net);
936 for (i = 0; i < INSTANCE_BUCKETS; i++) {
937 struct nfqnl_instance *inst;
938 struct hlist_head *head = &q->instance_table[i];
940 hlist_for_each_entry_rcu(inst, head, hlist)
941 nfqnl_flush(inst, dev_cmp, ifindex);
948 nfqnl_rcv_dev_event(struct notifier_block *this,
949 unsigned long event, void *ptr)
951 struct net_device *dev = netdev_notifier_info_to_dev(ptr);
953 /* Drop any packets associated with the downed device */
954 if (event == NETDEV_DOWN)
955 nfqnl_dev_drop(dev_net(dev), dev->ifindex);
959 static struct notifier_block nfqnl_dev_notifier = {
960 .notifier_call = nfqnl_rcv_dev_event,
963 static void nfqnl_nf_hook_drop(struct net *net)
965 struct nfnl_queue_net *q = nfnl_queue_pernet(net);
968 /* This function is also called on net namespace error unwind,
969 * when pernet_ops->init() failed and ->exit() functions of the
970 * previous pernet_ops gets called.
972 * This may result in a call to nfqnl_nf_hook_drop() before
973 * struct nfnl_queue_net was allocated.
978 for (i = 0; i < INSTANCE_BUCKETS; i++) {
979 struct nfqnl_instance *inst;
980 struct hlist_head *head = &q->instance_table[i];
982 hlist_for_each_entry_rcu(inst, head, hlist)
983 nfqnl_flush(inst, NULL, 0);
988 nfqnl_rcv_nl_event(struct notifier_block *this,
989 unsigned long event, void *ptr)
991 struct netlink_notify *n = ptr;
992 struct nfnl_queue_net *q = nfnl_queue_pernet(n->net);
994 if (event == NETLINK_URELEASE && n->protocol == NETLINK_NETFILTER) {
997 /* destroy all instances for this portid */
998 spin_lock(&q->instances_lock);
999 for (i = 0; i < INSTANCE_BUCKETS; i++) {
1000 struct hlist_node *t2;
1001 struct nfqnl_instance *inst;
1002 struct hlist_head *head = &q->instance_table[i];
1004 hlist_for_each_entry_safe(inst, t2, head, hlist) {
1005 if (n->portid == inst->peer_portid)
1006 __instance_destroy(inst);
1009 spin_unlock(&q->instances_lock);
1014 static struct notifier_block nfqnl_rtnl_notifier = {
1015 .notifier_call = nfqnl_rcv_nl_event,
1018 static const struct nla_policy nfqa_vlan_policy[NFQA_VLAN_MAX + 1] = {
1019 [NFQA_VLAN_TCI] = { .type = NLA_U16},
1020 [NFQA_VLAN_PROTO] = { .type = NLA_U16},
1023 static const struct nla_policy nfqa_verdict_policy[NFQA_MAX+1] = {
1024 [NFQA_VERDICT_HDR] = { .len = sizeof(struct nfqnl_msg_verdict_hdr) },
1025 [NFQA_MARK] = { .type = NLA_U32 },
1026 [NFQA_PAYLOAD] = { .type = NLA_UNSPEC },
1027 [NFQA_CT] = { .type = NLA_UNSPEC },
1028 [NFQA_EXP] = { .type = NLA_UNSPEC },
1029 [NFQA_VLAN] = { .type = NLA_NESTED },
1030 [NFQA_PRIORITY] = { .type = NLA_U32 },
1033 static const struct nla_policy nfqa_verdict_batch_policy[NFQA_MAX+1] = {
1034 [NFQA_VERDICT_HDR] = { .len = sizeof(struct nfqnl_msg_verdict_hdr) },
1035 [NFQA_MARK] = { .type = NLA_U32 },
1036 [NFQA_PRIORITY] = { .type = NLA_U32 },
1039 static struct nfqnl_instance *
1040 verdict_instance_lookup(struct nfnl_queue_net *q, u16 queue_num, u32 nlportid)
1042 struct nfqnl_instance *queue;
1044 queue = instance_lookup(q, queue_num);
1046 return ERR_PTR(-ENODEV);
1048 if (queue->peer_portid != nlportid)
1049 return ERR_PTR(-EPERM);
1054 static struct nfqnl_msg_verdict_hdr*
1055 verdicthdr_get(const struct nlattr * const nfqa[])
1057 struct nfqnl_msg_verdict_hdr *vhdr;
1058 unsigned int verdict;
1060 if (!nfqa[NFQA_VERDICT_HDR])
1063 vhdr = nla_data(nfqa[NFQA_VERDICT_HDR]);
1064 verdict = ntohl(vhdr->verdict) & NF_VERDICT_MASK;
1065 if (verdict > NF_MAX_VERDICT || verdict == NF_STOLEN)
1070 static int nfq_id_after(unsigned int id, unsigned int max)
1072 return (int)(id - max) > 0;
1075 static int nfqnl_recv_verdict_batch(struct sk_buff *skb,
1076 const struct nfnl_info *info,
1077 const struct nlattr * const nfqa[])
1079 struct nfnl_queue_net *q = nfnl_queue_pernet(info->net);
1080 u16 queue_num = ntohs(info->nfmsg->res_id);
1081 struct nf_queue_entry *entry, *tmp;
1082 struct nfqnl_msg_verdict_hdr *vhdr;
1083 struct nfqnl_instance *queue;
1084 unsigned int verdict, maxid;
1085 LIST_HEAD(batch_list);
1087 queue = verdict_instance_lookup(q, queue_num,
1088 NETLINK_CB(skb).portid);
1090 return PTR_ERR(queue);
1092 vhdr = verdicthdr_get(nfqa);
1096 verdict = ntohl(vhdr->verdict);
1097 maxid = ntohl(vhdr->id);
1099 spin_lock_bh(&queue->lock);
1101 list_for_each_entry_safe(entry, tmp, &queue->queue_list, list) {
1102 if (nfq_id_after(entry->id, maxid))
1104 __dequeue_entry(queue, entry);
1105 list_add_tail(&entry->list, &batch_list);
1108 spin_unlock_bh(&queue->lock);
1110 if (list_empty(&batch_list))
1113 list_for_each_entry_safe(entry, tmp, &batch_list, list) {
1114 if (nfqa[NFQA_MARK])
1115 entry->skb->mark = ntohl(nla_get_be32(nfqa[NFQA_MARK]));
1117 if (nfqa[NFQA_PRIORITY])
1118 entry->skb->priority = ntohl(nla_get_be32(nfqa[NFQA_PRIORITY]));
1120 nfqnl_reinject(entry, verdict);
1125 static struct nf_conn *nfqnl_ct_parse(const struct nfnl_ct_hook *nfnl_ct,
1126 const struct nlmsghdr *nlh,
1127 const struct nlattr * const nfqa[],
1128 struct nf_queue_entry *entry,
1129 enum ip_conntrack_info *ctinfo)
1131 #if IS_ENABLED(CONFIG_NF_CONNTRACK)
1134 ct = nf_ct_get(entry->skb, ctinfo);
1138 if (nfnl_ct->parse(nfqa[NFQA_CT], ct) < 0)
1142 nfnl_ct->attach_expect(nfqa[NFQA_EXP], ct,
1143 NETLINK_CB(entry->skb).portid,
1151 static int nfqa_parse_bridge(struct nf_queue_entry *entry,
1152 const struct nlattr * const nfqa[])
1154 if (nfqa[NFQA_VLAN]) {
1155 struct nlattr *tb[NFQA_VLAN_MAX + 1];
1158 err = nla_parse_nested_deprecated(tb, NFQA_VLAN_MAX,
1160 nfqa_vlan_policy, NULL);
1164 if (!tb[NFQA_VLAN_TCI] || !tb[NFQA_VLAN_PROTO])
1167 __vlan_hwaccel_put_tag(entry->skb,
1168 nla_get_be16(tb[NFQA_VLAN_PROTO]),
1169 ntohs(nla_get_be16(tb[NFQA_VLAN_TCI])));
1172 if (nfqa[NFQA_L2HDR]) {
1173 int mac_header_len = entry->skb->network_header -
1174 entry->skb->mac_header;
1176 if (mac_header_len != nla_len(nfqa[NFQA_L2HDR]))
1178 else if (mac_header_len > 0)
1179 memcpy(skb_mac_header(entry->skb),
1180 nla_data(nfqa[NFQA_L2HDR]),
1187 static int nfqnl_recv_verdict(struct sk_buff *skb, const struct nfnl_info *info,
1188 const struct nlattr * const nfqa[])
1190 struct nfnl_queue_net *q = nfnl_queue_pernet(info->net);
1191 u_int16_t queue_num = ntohs(info->nfmsg->res_id);
1192 const struct nfnl_ct_hook *nfnl_ct;
1193 struct nfqnl_msg_verdict_hdr *vhdr;
1194 enum ip_conntrack_info ctinfo;
1195 struct nfqnl_instance *queue;
1196 struct nf_queue_entry *entry;
1197 struct nf_conn *ct = NULL;
1198 unsigned int verdict;
1201 queue = verdict_instance_lookup(q, queue_num,
1202 NETLINK_CB(skb).portid);
1204 return PTR_ERR(queue);
1206 vhdr = verdicthdr_get(nfqa);
1210 verdict = ntohl(vhdr->verdict);
1212 entry = find_dequeue_entry(queue, ntohl(vhdr->id));
1216 /* rcu lock already held from nfnl->call_rcu. */
1217 nfnl_ct = rcu_dereference(nfnl_ct_hook);
1219 if (nfqa[NFQA_CT]) {
1220 if (nfnl_ct != NULL)
1221 ct = nfqnl_ct_parse(nfnl_ct, info->nlh, nfqa, entry,
1225 if (entry->state.pf == PF_BRIDGE) {
1226 err = nfqa_parse_bridge(entry, nfqa);
1231 if (nfqa[NFQA_PAYLOAD]) {
1232 u16 payload_len = nla_len(nfqa[NFQA_PAYLOAD]);
1233 int diff = payload_len - entry->skb->len;
1235 if (nfqnl_mangle(nla_data(nfqa[NFQA_PAYLOAD]),
1236 payload_len, entry, diff) < 0)
1240 nfnl_ct->seq_adjust(entry->skb, ct, ctinfo, diff);
1243 if (nfqa[NFQA_MARK])
1244 entry->skb->mark = ntohl(nla_get_be32(nfqa[NFQA_MARK]));
1246 if (nfqa[NFQA_PRIORITY])
1247 entry->skb->priority = ntohl(nla_get_be32(nfqa[NFQA_PRIORITY]));
1249 nfqnl_reinject(entry, verdict);
1253 static int nfqnl_recv_unsupp(struct sk_buff *skb, const struct nfnl_info *info,
1254 const struct nlattr * const cda[])
1259 static const struct nla_policy nfqa_cfg_policy[NFQA_CFG_MAX+1] = {
1260 [NFQA_CFG_CMD] = { .len = sizeof(struct nfqnl_msg_config_cmd) },
1261 [NFQA_CFG_PARAMS] = { .len = sizeof(struct nfqnl_msg_config_params) },
1262 [NFQA_CFG_QUEUE_MAXLEN] = { .type = NLA_U32 },
1263 [NFQA_CFG_MASK] = { .type = NLA_U32 },
1264 [NFQA_CFG_FLAGS] = { .type = NLA_U32 },
1267 static const struct nf_queue_handler nfqh = {
1268 .outfn = nfqnl_enqueue_packet,
1269 .nf_hook_drop = nfqnl_nf_hook_drop,
1272 static int nfqnl_recv_config(struct sk_buff *skb, const struct nfnl_info *info,
1273 const struct nlattr * const nfqa[])
1275 struct nfnl_queue_net *q = nfnl_queue_pernet(info->net);
1276 u_int16_t queue_num = ntohs(info->nfmsg->res_id);
1277 struct nfqnl_msg_config_cmd *cmd = NULL;
1278 struct nfqnl_instance *queue;
1279 __u32 flags = 0, mask = 0;
1282 if (nfqa[NFQA_CFG_CMD]) {
1283 cmd = nla_data(nfqa[NFQA_CFG_CMD]);
1285 /* Obsolete commands without queue context */
1286 switch (cmd->command) {
1287 case NFQNL_CFG_CMD_PF_BIND: return 0;
1288 case NFQNL_CFG_CMD_PF_UNBIND: return 0;
1292 /* Check if we support these flags in first place, dependencies should
1293 * be there too not to break atomicity.
1295 if (nfqa[NFQA_CFG_FLAGS]) {
1296 if (!nfqa[NFQA_CFG_MASK]) {
1297 /* A mask is needed to specify which flags are being
1303 flags = ntohl(nla_get_be32(nfqa[NFQA_CFG_FLAGS]));
1304 mask = ntohl(nla_get_be32(nfqa[NFQA_CFG_MASK]));
1306 if (flags >= NFQA_CFG_F_MAX)
1309 #if !IS_ENABLED(CONFIG_NETWORK_SECMARK)
1310 if (flags & mask & NFQA_CFG_F_SECCTX)
1313 if ((flags & mask & NFQA_CFG_F_CONNTRACK) &&
1314 !rcu_access_pointer(nfnl_ct_hook)) {
1315 #ifdef CONFIG_MODULES
1316 nfnl_unlock(NFNL_SUBSYS_QUEUE);
1317 request_module("ip_conntrack_netlink");
1318 nfnl_lock(NFNL_SUBSYS_QUEUE);
1319 if (rcu_access_pointer(nfnl_ct_hook))
1327 queue = instance_lookup(q, queue_num);
1328 if (queue && queue->peer_portid != NETLINK_CB(skb).portid) {
1330 goto err_out_unlock;
1334 switch (cmd->command) {
1335 case NFQNL_CFG_CMD_BIND:
1338 goto err_out_unlock;
1340 queue = instance_create(q, queue_num,
1341 NETLINK_CB(skb).portid);
1342 if (IS_ERR(queue)) {
1343 ret = PTR_ERR(queue);
1344 goto err_out_unlock;
1347 case NFQNL_CFG_CMD_UNBIND:
1350 goto err_out_unlock;
1352 instance_destroy(q, queue);
1353 goto err_out_unlock;
1354 case NFQNL_CFG_CMD_PF_BIND:
1355 case NFQNL_CFG_CMD_PF_UNBIND:
1359 goto err_out_unlock;
1365 goto err_out_unlock;
1368 if (nfqa[NFQA_CFG_PARAMS]) {
1369 struct nfqnl_msg_config_params *params =
1370 nla_data(nfqa[NFQA_CFG_PARAMS]);
1372 nfqnl_set_mode(queue, params->copy_mode,
1373 ntohl(params->copy_range));
1376 if (nfqa[NFQA_CFG_QUEUE_MAXLEN]) {
1377 __be32 *queue_maxlen = nla_data(nfqa[NFQA_CFG_QUEUE_MAXLEN]);
1379 spin_lock_bh(&queue->lock);
1380 queue->queue_maxlen = ntohl(*queue_maxlen);
1381 spin_unlock_bh(&queue->lock);
1384 if (nfqa[NFQA_CFG_FLAGS]) {
1385 spin_lock_bh(&queue->lock);
1386 queue->flags &= ~mask;
1387 queue->flags |= flags & mask;
1388 spin_unlock_bh(&queue->lock);
1396 static const struct nfnl_callback nfqnl_cb[NFQNL_MSG_MAX] = {
1397 [NFQNL_MSG_PACKET] = {
1398 .call = nfqnl_recv_unsupp,
1399 .type = NFNL_CB_RCU,
1400 .attr_count = NFQA_MAX,
1402 [NFQNL_MSG_VERDICT] = {
1403 .call = nfqnl_recv_verdict,
1404 .type = NFNL_CB_RCU,
1405 .attr_count = NFQA_MAX,
1406 .policy = nfqa_verdict_policy
1408 [NFQNL_MSG_CONFIG] = {
1409 .call = nfqnl_recv_config,
1410 .type = NFNL_CB_MUTEX,
1411 .attr_count = NFQA_CFG_MAX,
1412 .policy = nfqa_cfg_policy
1414 [NFQNL_MSG_VERDICT_BATCH] = {
1415 .call = nfqnl_recv_verdict_batch,
1416 .type = NFNL_CB_RCU,
1417 .attr_count = NFQA_MAX,
1418 .policy = nfqa_verdict_batch_policy
1422 static const struct nfnetlink_subsystem nfqnl_subsys = {
1424 .subsys_id = NFNL_SUBSYS_QUEUE,
1425 .cb_count = NFQNL_MSG_MAX,
1429 #ifdef CONFIG_PROC_FS
1431 struct seq_net_private p;
1432 unsigned int bucket;
1435 static struct hlist_node *get_first(struct seq_file *seq)
1437 struct iter_state *st = seq->private;
1439 struct nfnl_queue_net *q;
1444 net = seq_file_net(seq);
1445 q = nfnl_queue_pernet(net);
1446 for (st->bucket = 0; st->bucket < INSTANCE_BUCKETS; st->bucket++) {
1447 if (!hlist_empty(&q->instance_table[st->bucket]))
1448 return q->instance_table[st->bucket].first;
1453 static struct hlist_node *get_next(struct seq_file *seq, struct hlist_node *h)
1455 struct iter_state *st = seq->private;
1456 struct net *net = seq_file_net(seq);
1460 struct nfnl_queue_net *q;
1462 if (++st->bucket >= INSTANCE_BUCKETS)
1465 q = nfnl_queue_pernet(net);
1466 h = q->instance_table[st->bucket].first;
1471 static struct hlist_node *get_idx(struct seq_file *seq, loff_t pos)
1473 struct hlist_node *head;
1474 head = get_first(seq);
1477 while (pos && (head = get_next(seq, head)))
1479 return pos ? NULL : head;
1482 static void *seq_start(struct seq_file *s, loff_t *pos)
1483 __acquires(nfnl_queue_pernet(seq_file_net(s))->instances_lock)
1485 spin_lock(&nfnl_queue_pernet(seq_file_net(s))->instances_lock);
1486 return get_idx(s, *pos);
1489 static void *seq_next(struct seq_file *s, void *v, loff_t *pos)
1492 return get_next(s, v);
1495 static void seq_stop(struct seq_file *s, void *v)
1496 __releases(nfnl_queue_pernet(seq_file_net(s))->instances_lock)
1498 spin_unlock(&nfnl_queue_pernet(seq_file_net(s))->instances_lock);
1501 static int seq_show(struct seq_file *s, void *v)
1503 const struct nfqnl_instance *inst = v;
1505 seq_printf(s, "%5u %6u %5u %1u %5u %5u %5u %8u %2d\n",
1507 inst->peer_portid, inst->queue_total,
1508 inst->copy_mode, inst->copy_range,
1509 inst->queue_dropped, inst->queue_user_dropped,
1510 inst->id_sequence, 1);
1514 static const struct seq_operations nfqnl_seq_ops = {
1520 #endif /* PROC_FS */
1522 static int __net_init nfnl_queue_net_init(struct net *net)
1525 struct nfnl_queue_net *q = nfnl_queue_pernet(net);
1527 for (i = 0; i < INSTANCE_BUCKETS; i++)
1528 INIT_HLIST_HEAD(&q->instance_table[i]);
1530 spin_lock_init(&q->instances_lock);
1532 #ifdef CONFIG_PROC_FS
1533 if (!proc_create_net("nfnetlink_queue", 0440, net->nf.proc_netfilter,
1534 &nfqnl_seq_ops, sizeof(struct iter_state)))
1540 static void __net_exit nfnl_queue_net_exit(struct net *net)
1542 struct nfnl_queue_net *q = nfnl_queue_pernet(net);
1545 #ifdef CONFIG_PROC_FS
1546 remove_proc_entry("nfnetlink_queue", net->nf.proc_netfilter);
1548 for (i = 0; i < INSTANCE_BUCKETS; i++)
1549 WARN_ON_ONCE(!hlist_empty(&q->instance_table[i]));
1552 static struct pernet_operations nfnl_queue_net_ops = {
1553 .init = nfnl_queue_net_init,
1554 .exit = nfnl_queue_net_exit,
1555 .id = &nfnl_queue_net_id,
1556 .size = sizeof(struct nfnl_queue_net),
1559 static int __init nfnetlink_queue_init(void)
1563 status = register_pernet_subsys(&nfnl_queue_net_ops);
1565 pr_err("failed to register pernet ops\n");
1569 netlink_register_notifier(&nfqnl_rtnl_notifier);
1570 status = nfnetlink_subsys_register(&nfqnl_subsys);
1572 pr_err("failed to create netlink socket\n");
1573 goto cleanup_netlink_notifier;
1576 status = register_netdevice_notifier(&nfqnl_dev_notifier);
1578 pr_err("failed to register netdevice notifier\n");
1579 goto cleanup_netlink_subsys;
1582 nf_register_queue_handler(&nfqh);
1586 cleanup_netlink_subsys:
1587 nfnetlink_subsys_unregister(&nfqnl_subsys);
1588 cleanup_netlink_notifier:
1589 netlink_unregister_notifier(&nfqnl_rtnl_notifier);
1590 unregister_pernet_subsys(&nfnl_queue_net_ops);
1595 static void __exit nfnetlink_queue_fini(void)
1597 nf_unregister_queue_handler();
1598 unregister_netdevice_notifier(&nfqnl_dev_notifier);
1599 nfnetlink_subsys_unregister(&nfqnl_subsys);
1600 netlink_unregister_notifier(&nfqnl_rtnl_notifier);
1601 unregister_pernet_subsys(&nfnl_queue_net_ops);
1603 rcu_barrier(); /* Wait for completion of call_rcu()'s */
1606 MODULE_DESCRIPTION("netfilter packet queue handler");
1607 MODULE_AUTHOR("Harald Welte <laforge@netfilter.org>");
1608 MODULE_LICENSE("GPL");
1609 MODULE_ALIAS_NFNL_SUBSYS(NFNL_SUBSYS_QUEUE);
1611 module_init(nfnetlink_queue_init);
1612 module_exit(nfnetlink_queue_fini);