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 #if IS_ENABLED(CONFIG_NF_CONNTRACK)
45 #include <net/netfilter/nf_conntrack.h>
48 #define NFQNL_QMAX_DEFAULT 1024
50 /* We're using struct nlattr which has 16bit nla_len. Note that nla_len
51 * includes the header length. Thus, the maximum packet length that we
52 * support is 65531 bytes. We send truncated packets if the specified length
53 * is larger than that. Userspace can check for presence of NFQA_CAP_LEN
54 * attribute to detect truncation.
56 #define NFQNL_MAX_COPY_RANGE (0xffff - NLA_HDRLEN)
58 struct nfqnl_instance {
59 struct hlist_node hlist; /* global list of queues */
63 unsigned int queue_maxlen;
64 unsigned int copy_range;
65 unsigned int queue_dropped;
66 unsigned int queue_user_dropped;
69 u_int16_t queue_num; /* number of this queue */
71 u_int32_t flags; /* Set using NFQA_CFG_FLAGS */
73 * Following fields are dirtied for each queued packet,
74 * keep them in same cache line if possible.
76 spinlock_t lock ____cacheline_aligned_in_smp;
77 unsigned int queue_total;
78 unsigned int id_sequence; /* 'sequence' of pkt ids */
79 struct list_head queue_list; /* packets in queue */
82 typedef int (*nfqnl_cmpfn)(struct nf_queue_entry *, unsigned long);
84 static unsigned int nfnl_queue_net_id __read_mostly;
86 #define INSTANCE_BUCKETS 16
87 struct nfnl_queue_net {
88 spinlock_t instances_lock;
89 struct hlist_head instance_table[INSTANCE_BUCKETS];
92 static struct nfnl_queue_net *nfnl_queue_pernet(struct net *net)
94 return net_generic(net, nfnl_queue_net_id);
97 static inline u_int8_t instance_hashfn(u_int16_t queue_num)
99 return ((queue_num >> 8) ^ queue_num) % INSTANCE_BUCKETS;
102 static struct nfqnl_instance *
103 instance_lookup(struct nfnl_queue_net *q, u_int16_t queue_num)
105 struct hlist_head *head;
106 struct nfqnl_instance *inst;
108 head = &q->instance_table[instance_hashfn(queue_num)];
109 hlist_for_each_entry_rcu(inst, head, hlist) {
110 if (inst->queue_num == queue_num)
116 static struct nfqnl_instance *
117 instance_create(struct nfnl_queue_net *q, u_int16_t queue_num, u32 portid)
119 struct nfqnl_instance *inst;
123 spin_lock(&q->instances_lock);
124 if (instance_lookup(q, queue_num)) {
129 inst = kzalloc(sizeof(*inst), GFP_ATOMIC);
135 inst->queue_num = queue_num;
136 inst->peer_portid = portid;
137 inst->queue_maxlen = NFQNL_QMAX_DEFAULT;
138 inst->copy_range = NFQNL_MAX_COPY_RANGE;
139 inst->copy_mode = NFQNL_COPY_NONE;
140 spin_lock_init(&inst->lock);
141 INIT_LIST_HEAD(&inst->queue_list);
143 if (!try_module_get(THIS_MODULE)) {
148 h = instance_hashfn(queue_num);
149 hlist_add_head_rcu(&inst->hlist, &q->instance_table[h]);
151 spin_unlock(&q->instances_lock);
158 spin_unlock(&q->instances_lock);
162 static void nfqnl_flush(struct nfqnl_instance *queue, nfqnl_cmpfn cmpfn,
166 instance_destroy_rcu(struct rcu_head *head)
168 struct nfqnl_instance *inst = container_of(head, struct nfqnl_instance,
171 nfqnl_flush(inst, NULL, 0);
173 module_put(THIS_MODULE);
177 __instance_destroy(struct nfqnl_instance *inst)
179 hlist_del_rcu(&inst->hlist);
180 call_rcu(&inst->rcu, instance_destroy_rcu);
184 instance_destroy(struct nfnl_queue_net *q, struct nfqnl_instance *inst)
186 spin_lock(&q->instances_lock);
187 __instance_destroy(inst);
188 spin_unlock(&q->instances_lock);
192 __enqueue_entry(struct nfqnl_instance *queue, struct nf_queue_entry *entry)
194 list_add_tail(&entry->list, &queue->queue_list);
195 queue->queue_total++;
199 __dequeue_entry(struct nfqnl_instance *queue, struct nf_queue_entry *entry)
201 list_del(&entry->list);
202 queue->queue_total--;
205 static struct nf_queue_entry *
206 find_dequeue_entry(struct nfqnl_instance *queue, unsigned int id)
208 struct nf_queue_entry *entry = NULL, *i;
210 spin_lock_bh(&queue->lock);
212 list_for_each_entry(i, &queue->queue_list, list) {
220 __dequeue_entry(queue, entry);
222 spin_unlock_bh(&queue->lock);
228 nfqnl_flush(struct nfqnl_instance *queue, nfqnl_cmpfn cmpfn, unsigned long data)
230 struct nf_queue_entry *entry, *next;
232 spin_lock_bh(&queue->lock);
233 list_for_each_entry_safe(entry, next, &queue->queue_list, list) {
234 if (!cmpfn || cmpfn(entry, data)) {
235 list_del(&entry->list);
236 queue->queue_total--;
237 nf_reinject(entry, NF_DROP);
240 spin_unlock_bh(&queue->lock);
244 nfqnl_put_packet_info(struct sk_buff *nlskb, struct sk_buff *packet,
249 if (packet->ip_summed == CHECKSUM_PARTIAL)
250 flags = NFQA_SKB_CSUMNOTREADY;
251 else if (csum_verify)
252 flags = NFQA_SKB_CSUM_NOTVERIFIED;
254 if (skb_is_gso(packet))
255 flags |= NFQA_SKB_GSO;
257 return flags ? nla_put_be32(nlskb, NFQA_SKB_INFO, htonl(flags)) : 0;
260 static int nfqnl_put_sk_uidgid(struct sk_buff *skb, struct sock *sk)
262 const struct cred *cred;
264 if (!sk_fullsock(sk))
267 read_lock_bh(&sk->sk_callback_lock);
268 if (sk->sk_socket && sk->sk_socket->file) {
269 cred = sk->sk_socket->file->f_cred;
270 if (nla_put_be32(skb, NFQA_UID,
271 htonl(from_kuid_munged(&init_user_ns, cred->fsuid))))
272 goto nla_put_failure;
273 if (nla_put_be32(skb, NFQA_GID,
274 htonl(from_kgid_munged(&init_user_ns, cred->fsgid))))
275 goto nla_put_failure;
277 read_unlock_bh(&sk->sk_callback_lock);
281 read_unlock_bh(&sk->sk_callback_lock);
285 static u32 nfqnl_get_sk_secctx(struct sk_buff *skb, char **secdata)
288 #if IS_ENABLED(CONFIG_NETWORK_SECMARK)
289 if (!skb || !sk_fullsock(skb->sk))
292 read_lock_bh(&skb->sk->sk_callback_lock);
295 security_secid_to_secctx(skb->secmark, secdata, &seclen);
297 read_unlock_bh(&skb->sk->sk_callback_lock);
302 static u32 nfqnl_get_bridge_size(struct nf_queue_entry *entry)
304 struct sk_buff *entskb = entry->skb;
307 if (entry->state.pf != PF_BRIDGE || !skb_mac_header_was_set(entskb))
310 if (skb_vlan_tag_present(entskb))
311 nlalen += nla_total_size(nla_total_size(sizeof(__be16)) +
312 nla_total_size(sizeof(__be16)));
314 if (entskb->network_header > entskb->mac_header)
315 nlalen += nla_total_size((entskb->network_header -
316 entskb->mac_header));
321 static int nfqnl_put_bridge(struct nf_queue_entry *entry, struct sk_buff *skb)
323 struct sk_buff *entskb = entry->skb;
325 if (entry->state.pf != PF_BRIDGE || !skb_mac_header_was_set(entskb))
328 if (skb_vlan_tag_present(entskb)) {
331 nest = nla_nest_start(skb, NFQA_VLAN | NLA_F_NESTED);
333 goto nla_put_failure;
335 if (nla_put_be16(skb, NFQA_VLAN_TCI, htons(entskb->vlan_tci)) ||
336 nla_put_be16(skb, NFQA_VLAN_PROTO, entskb->vlan_proto))
337 goto nla_put_failure;
339 nla_nest_end(skb, nest);
342 if (entskb->mac_header < entskb->network_header) {
343 int len = (int)(entskb->network_header - entskb->mac_header);
345 if (nla_put(skb, NFQA_L2HDR, len, skb_mac_header(entskb)))
346 goto nla_put_failure;
355 static struct sk_buff *
356 nfqnl_build_packet_message(struct net *net, struct nfqnl_instance *queue,
357 struct nf_queue_entry *entry,
358 __be32 **packet_id_ptr)
361 size_t data_len = 0, cap_len = 0;
362 unsigned int hlen = 0;
365 struct nfqnl_msg_packet_hdr *pmsg;
366 struct nlmsghdr *nlh;
367 struct nfgenmsg *nfmsg;
368 struct sk_buff *entskb = entry->skb;
369 struct net_device *indev;
370 struct net_device *outdev;
371 struct nf_conn *ct = NULL;
372 enum ip_conntrack_info uninitialized_var(ctinfo);
373 struct nfnl_ct_hook *nfnl_ct;
375 char *secdata = NULL;
378 size = nlmsg_total_size(sizeof(struct nfgenmsg))
379 + nla_total_size(sizeof(struct nfqnl_msg_packet_hdr))
380 + nla_total_size(sizeof(u_int32_t)) /* ifindex */
381 + nla_total_size(sizeof(u_int32_t)) /* ifindex */
382 #if IS_ENABLED(CONFIG_BRIDGE_NETFILTER)
383 + nla_total_size(sizeof(u_int32_t)) /* ifindex */
384 + nla_total_size(sizeof(u_int32_t)) /* ifindex */
386 + nla_total_size(sizeof(u_int32_t)) /* mark */
387 + nla_total_size(sizeof(struct nfqnl_msg_packet_hw))
388 + nla_total_size(sizeof(u_int32_t)) /* skbinfo */
389 + nla_total_size(sizeof(u_int32_t)); /* cap_len */
392 size += nla_total_size(sizeof(struct nfqnl_msg_packet_timestamp));
394 size += nfqnl_get_bridge_size(entry);
396 if (entry->state.hook <= NF_INET_FORWARD ||
397 (entry->state.hook == NF_INET_POST_ROUTING && entskb->sk == NULL))
398 csum_verify = !skb_csum_unnecessary(entskb);
402 outdev = entry->state.out;
404 switch ((enum nfqnl_config_mode)ACCESS_ONCE(queue->copy_mode)) {
405 case NFQNL_COPY_META:
406 case NFQNL_COPY_NONE:
409 case NFQNL_COPY_PACKET:
410 if (!(queue->flags & NFQA_CFG_F_GSO) &&
411 entskb->ip_summed == CHECKSUM_PARTIAL &&
412 skb_checksum_help(entskb))
415 data_len = ACCESS_ONCE(queue->copy_range);
416 if (data_len > entskb->len)
417 data_len = entskb->len;
419 hlen = skb_zerocopy_headlen(entskb);
420 hlen = min_t(unsigned int, hlen, data_len);
421 size += sizeof(struct nlattr) + hlen;
422 cap_len = entskb->len;
426 nfnl_ct = rcu_dereference(nfnl_ct_hook);
428 if (queue->flags & NFQA_CFG_F_CONNTRACK) {
429 if (nfnl_ct != NULL) {
430 ct = nfnl_ct->get_ct(entskb, &ctinfo);
432 size += nfnl_ct->build_size(ct);
436 if (queue->flags & NFQA_CFG_F_UID_GID) {
437 size += (nla_total_size(sizeof(u_int32_t)) /* uid */
438 + nla_total_size(sizeof(u_int32_t))); /* gid */
441 if ((queue->flags & NFQA_CFG_F_SECCTX) && entskb->sk) {
442 seclen = nfqnl_get_sk_secctx(entskb, &secdata);
444 size += nla_total_size(seclen);
447 skb = alloc_skb(size, GFP_ATOMIC);
449 skb_tx_error(entskb);
453 nlh = nlmsg_put(skb, 0, 0,
454 nfnl_msg_type(NFNL_SUBSYS_QUEUE, NFQNL_MSG_PACKET),
455 sizeof(struct nfgenmsg), 0);
457 skb_tx_error(entskb);
461 nfmsg = nlmsg_data(nlh);
462 nfmsg->nfgen_family = entry->state.pf;
463 nfmsg->version = NFNETLINK_V0;
464 nfmsg->res_id = htons(queue->queue_num);
466 nla = __nla_reserve(skb, NFQA_PACKET_HDR, sizeof(*pmsg));
467 pmsg = nla_data(nla);
468 pmsg->hw_protocol = entskb->protocol;
469 pmsg->hook = entry->state.hook;
470 *packet_id_ptr = &pmsg->packet_id;
472 indev = entry->state.in;
474 #if !IS_ENABLED(CONFIG_BRIDGE_NETFILTER)
475 if (nla_put_be32(skb, NFQA_IFINDEX_INDEV, htonl(indev->ifindex)))
476 goto nla_put_failure;
478 if (entry->state.pf == PF_BRIDGE) {
479 /* Case 1: indev is physical input device, we need to
480 * look for bridge group (when called from
481 * netfilter_bridge) */
482 if (nla_put_be32(skb, NFQA_IFINDEX_PHYSINDEV,
483 htonl(indev->ifindex)) ||
484 /* this is the bridge group "brX" */
485 /* rcu_read_lock()ed by __nf_queue */
486 nla_put_be32(skb, NFQA_IFINDEX_INDEV,
487 htonl(br_port_get_rcu(indev)->br->dev->ifindex)))
488 goto nla_put_failure;
492 /* Case 2: indev is bridge group, we need to look for
493 * physical device (when called from ipv4) */
494 if (nla_put_be32(skb, NFQA_IFINDEX_INDEV,
495 htonl(indev->ifindex)))
496 goto nla_put_failure;
498 physinif = nf_bridge_get_physinif(entskb);
500 nla_put_be32(skb, NFQA_IFINDEX_PHYSINDEV,
502 goto nla_put_failure;
508 #if !IS_ENABLED(CONFIG_BRIDGE_NETFILTER)
509 if (nla_put_be32(skb, NFQA_IFINDEX_OUTDEV, htonl(outdev->ifindex)))
510 goto nla_put_failure;
512 if (entry->state.pf == PF_BRIDGE) {
513 /* Case 1: outdev is physical output device, we need to
514 * look for bridge group (when called from
515 * netfilter_bridge) */
516 if (nla_put_be32(skb, NFQA_IFINDEX_PHYSOUTDEV,
517 htonl(outdev->ifindex)) ||
518 /* this is the bridge group "brX" */
519 /* rcu_read_lock()ed by __nf_queue */
520 nla_put_be32(skb, NFQA_IFINDEX_OUTDEV,
521 htonl(br_port_get_rcu(outdev)->br->dev->ifindex)))
522 goto nla_put_failure;
526 /* Case 2: outdev is bridge group, we need to look for
527 * physical output device (when called from ipv4) */
528 if (nla_put_be32(skb, NFQA_IFINDEX_OUTDEV,
529 htonl(outdev->ifindex)))
530 goto nla_put_failure;
532 physoutif = nf_bridge_get_physoutif(entskb);
534 nla_put_be32(skb, NFQA_IFINDEX_PHYSOUTDEV,
536 goto nla_put_failure;
542 nla_put_be32(skb, NFQA_MARK, htonl(entskb->mark)))
543 goto nla_put_failure;
545 if (indev && entskb->dev &&
546 skb_mac_header_was_set(entskb) &&
547 skb_mac_header_len(entskb) != 0) {
548 struct nfqnl_msg_packet_hw phw;
551 memset(&phw, 0, sizeof(phw));
552 len = dev_parse_header(entskb, phw.hw_addr);
554 phw.hw_addrlen = htons(len);
555 if (nla_put(skb, NFQA_HWADDR, sizeof(phw), &phw))
556 goto nla_put_failure;
560 if (nfqnl_put_bridge(entry, skb) < 0)
561 goto nla_put_failure;
563 if (entskb->tstamp) {
564 struct nfqnl_msg_packet_timestamp ts;
565 struct timespec64 kts = ktime_to_timespec64(entskb->tstamp);
567 ts.sec = cpu_to_be64(kts.tv_sec);
568 ts.usec = cpu_to_be64(kts.tv_nsec / NSEC_PER_USEC);
570 if (nla_put(skb, NFQA_TIMESTAMP, sizeof(ts), &ts))
571 goto nla_put_failure;
574 if ((queue->flags & NFQA_CFG_F_UID_GID) && entskb->sk &&
575 nfqnl_put_sk_uidgid(skb, entskb->sk) < 0)
576 goto nla_put_failure;
578 if (seclen && nla_put(skb, NFQA_SECCTX, seclen, secdata))
579 goto nla_put_failure;
581 if (ct && nfnl_ct->build(skb, ct, ctinfo, NFQA_CT, NFQA_CT_INFO) < 0)
582 goto nla_put_failure;
584 if (cap_len > data_len &&
585 nla_put_be32(skb, NFQA_CAP_LEN, htonl(cap_len)))
586 goto nla_put_failure;
588 if (nfqnl_put_packet_info(skb, entskb, csum_verify))
589 goto nla_put_failure;
594 if (skb_tailroom(skb) < sizeof(*nla) + hlen)
595 goto nla_put_failure;
597 nla = skb_put(skb, sizeof(*nla));
598 nla->nla_type = NFQA_PAYLOAD;
599 nla->nla_len = nla_attr_size(data_len);
601 if (skb_zerocopy(skb, entskb, data_len, hlen))
602 goto nla_put_failure;
605 nlh->nlmsg_len = skb->len;
607 security_release_secctx(secdata, seclen);
611 skb_tx_error(entskb);
613 net_err_ratelimited("nf_queue: error creating packet message\n");
616 security_release_secctx(secdata, seclen);
620 static bool nf_ct_drop_unconfirmed(const struct nf_queue_entry *entry)
622 #if IS_ENABLED(CONFIG_NF_CONNTRACK)
623 static const unsigned long flags = IPS_CONFIRMED | IPS_DYING;
624 const struct nf_conn *ct = (void *)skb_nfct(entry->skb);
626 if (ct && ((ct->status & flags) == IPS_DYING))
633 __nfqnl_enqueue_packet(struct net *net, struct nfqnl_instance *queue,
634 struct nf_queue_entry *entry)
636 struct sk_buff *nskb;
638 __be32 *packet_id_ptr;
641 nskb = nfqnl_build_packet_message(net, queue, entry, &packet_id_ptr);
646 spin_lock_bh(&queue->lock);
648 if (nf_ct_drop_unconfirmed(entry))
649 goto err_out_free_nskb;
651 if (queue->queue_total >= queue->queue_maxlen) {
652 if (queue->flags & NFQA_CFG_F_FAIL_OPEN) {
656 queue->queue_dropped++;
657 net_warn_ratelimited("nf_queue: full at %d entries, dropping packets(s)\n",
660 goto err_out_free_nskb;
662 entry->id = ++queue->id_sequence;
663 *packet_id_ptr = htonl(entry->id);
665 /* nfnetlink_unicast will either free the nskb or add it to a socket */
666 err = nfnetlink_unicast(nskb, net, queue->peer_portid, MSG_DONTWAIT);
668 if (queue->flags & NFQA_CFG_F_FAIL_OPEN) {
672 queue->queue_user_dropped++;
677 __enqueue_entry(queue, entry);
679 spin_unlock_bh(&queue->lock);
685 spin_unlock_bh(&queue->lock);
687 nf_reinject(entry, NF_ACCEPT);
692 static struct nf_queue_entry *
693 nf_queue_entry_dup(struct nf_queue_entry *e)
695 struct nf_queue_entry *entry = kmemdup(e, e->size, GFP_ATOMIC);
697 nf_queue_entry_get_refs(entry);
701 #if IS_ENABLED(CONFIG_BRIDGE_NETFILTER)
702 /* When called from bridge netfilter, skb->data must point to MAC header
703 * before calling skb_gso_segment(). Else, original MAC header is lost
704 * and segmented skbs will be sent to wrong destination.
706 static void nf_bridge_adjust_skb_data(struct sk_buff *skb)
709 __skb_push(skb, skb->network_header - skb->mac_header);
712 static void nf_bridge_adjust_segmented_data(struct sk_buff *skb)
715 __skb_pull(skb, skb->network_header - skb->mac_header);
718 #define nf_bridge_adjust_skb_data(s) do {} while (0)
719 #define nf_bridge_adjust_segmented_data(s) do {} while (0)
722 static void free_entry(struct nf_queue_entry *entry)
724 nf_queue_entry_release_refs(entry);
729 __nfqnl_enqueue_packet_gso(struct net *net, struct nfqnl_instance *queue,
730 struct sk_buff *skb, struct nf_queue_entry *entry)
733 struct nf_queue_entry *entry_seg;
735 nf_bridge_adjust_segmented_data(skb);
737 if (skb->next == NULL) { /* last packet, no need to copy entry */
738 struct sk_buff *gso_skb = entry->skb;
740 ret = __nfqnl_enqueue_packet(net, queue, entry);
742 entry->skb = gso_skb;
748 entry_seg = nf_queue_entry_dup(entry);
750 entry_seg->skb = skb;
751 ret = __nfqnl_enqueue_packet(net, queue, entry_seg);
753 free_entry(entry_seg);
759 nfqnl_enqueue_packet(struct nf_queue_entry *entry, unsigned int queuenum)
762 struct nfqnl_instance *queue;
763 struct sk_buff *skb, *segs;
765 struct net *net = entry->state.net;
766 struct nfnl_queue_net *q = nfnl_queue_pernet(net);
768 /* rcu_read_lock()ed by nf_hook_thresh */
769 queue = instance_lookup(q, queuenum);
773 if (queue->copy_mode == NFQNL_COPY_NONE)
778 switch (entry->state.pf) {
780 skb->protocol = htons(ETH_P_IP);
783 skb->protocol = htons(ETH_P_IPV6);
787 if ((queue->flags & NFQA_CFG_F_GSO) || !skb_is_gso(skb))
788 return __nfqnl_enqueue_packet(net, queue, entry);
790 nf_bridge_adjust_skb_data(skb);
791 segs = skb_gso_segment(skb, 0);
792 /* Does not use PTR_ERR to limit the number of error codes that can be
793 * returned by nf_queue. For instance, callers rely on -ESRCH to
794 * mean 'ignore this hook'.
796 if (IS_ERR_OR_NULL(segs))
801 struct sk_buff *nskb = segs->next;
803 err = __nfqnl_enqueue_packet_gso(net, queue,
813 if (err) /* some segments are already queued */
819 nf_bridge_adjust_segmented_data(skb);
824 nfqnl_mangle(void *data, int data_len, struct nf_queue_entry *e, int diff)
826 struct sk_buff *nskb;
829 if (pskb_trim(e->skb, data_len))
831 } else if (diff > 0) {
832 if (data_len > 0xFFFF)
834 if (diff > skb_tailroom(e->skb)) {
835 nskb = skb_copy_expand(e->skb, skb_headroom(e->skb),
838 printk(KERN_WARNING "nf_queue: OOM "
839 "in mangle, dropping packet\n");
845 skb_put(e->skb, diff);
847 if (!skb_make_writable(e->skb, data_len))
849 skb_copy_to_linear_data(e->skb, data, data_len);
850 e->skb->ip_summed = CHECKSUM_NONE;
855 nfqnl_set_mode(struct nfqnl_instance *queue,
856 unsigned char mode, unsigned int range)
860 spin_lock_bh(&queue->lock);
862 case NFQNL_COPY_NONE:
863 case NFQNL_COPY_META:
864 queue->copy_mode = mode;
865 queue->copy_range = 0;
868 case NFQNL_COPY_PACKET:
869 queue->copy_mode = mode;
870 if (range == 0 || range > NFQNL_MAX_COPY_RANGE)
871 queue->copy_range = NFQNL_MAX_COPY_RANGE;
873 queue->copy_range = range;
880 spin_unlock_bh(&queue->lock);
886 dev_cmp(struct nf_queue_entry *entry, unsigned long ifindex)
889 if (entry->state.in->ifindex == ifindex)
891 if (entry->state.out)
892 if (entry->state.out->ifindex == ifindex)
894 #if IS_ENABLED(CONFIG_BRIDGE_NETFILTER)
895 if (entry->skb->nf_bridge) {
896 int physinif, physoutif;
898 physinif = nf_bridge_get_physinif(entry->skb);
899 physoutif = nf_bridge_get_physoutif(entry->skb);
901 if (physinif == ifindex || physoutif == ifindex)
908 /* drop all packets with either indev or outdev == ifindex from all queue
911 nfqnl_dev_drop(struct net *net, int ifindex)
914 struct nfnl_queue_net *q = nfnl_queue_pernet(net);
918 for (i = 0; i < INSTANCE_BUCKETS; i++) {
919 struct nfqnl_instance *inst;
920 struct hlist_head *head = &q->instance_table[i];
922 hlist_for_each_entry_rcu(inst, head, hlist)
923 nfqnl_flush(inst, dev_cmp, ifindex);
930 nfqnl_rcv_dev_event(struct notifier_block *this,
931 unsigned long event, void *ptr)
933 struct net_device *dev = netdev_notifier_info_to_dev(ptr);
935 /* Drop any packets associated with the downed device */
936 if (event == NETDEV_DOWN)
937 nfqnl_dev_drop(dev_net(dev), dev->ifindex);
941 static struct notifier_block nfqnl_dev_notifier = {
942 .notifier_call = nfqnl_rcv_dev_event,
945 static unsigned int nfqnl_nf_hook_drop(struct net *net)
947 struct nfnl_queue_net *q = nfnl_queue_pernet(net);
948 unsigned int instances = 0;
951 for (i = 0; i < INSTANCE_BUCKETS; i++) {
952 struct nfqnl_instance *inst;
953 struct hlist_head *head = &q->instance_table[i];
955 hlist_for_each_entry_rcu(inst, head, hlist) {
956 nfqnl_flush(inst, NULL, 0);
965 nfqnl_rcv_nl_event(struct notifier_block *this,
966 unsigned long event, void *ptr)
968 struct netlink_notify *n = ptr;
969 struct nfnl_queue_net *q = nfnl_queue_pernet(n->net);
971 if (event == NETLINK_URELEASE && n->protocol == NETLINK_NETFILTER) {
974 /* destroy all instances for this portid */
975 spin_lock(&q->instances_lock);
976 for (i = 0; i < INSTANCE_BUCKETS; i++) {
977 struct hlist_node *t2;
978 struct nfqnl_instance *inst;
979 struct hlist_head *head = &q->instance_table[i];
981 hlist_for_each_entry_safe(inst, t2, head, hlist) {
982 if (n->portid == inst->peer_portid)
983 __instance_destroy(inst);
986 spin_unlock(&q->instances_lock);
991 static struct notifier_block nfqnl_rtnl_notifier = {
992 .notifier_call = nfqnl_rcv_nl_event,
995 static const struct nla_policy nfqa_vlan_policy[NFQA_VLAN_MAX + 1] = {
996 [NFQA_VLAN_TCI] = { .type = NLA_U16},
997 [NFQA_VLAN_PROTO] = { .type = NLA_U16},
1000 static const struct nla_policy nfqa_verdict_policy[NFQA_MAX+1] = {
1001 [NFQA_VERDICT_HDR] = { .len = sizeof(struct nfqnl_msg_verdict_hdr) },
1002 [NFQA_MARK] = { .type = NLA_U32 },
1003 [NFQA_PAYLOAD] = { .type = NLA_UNSPEC },
1004 [NFQA_CT] = { .type = NLA_UNSPEC },
1005 [NFQA_EXP] = { .type = NLA_UNSPEC },
1006 [NFQA_VLAN] = { .type = NLA_NESTED },
1009 static const struct nla_policy nfqa_verdict_batch_policy[NFQA_MAX+1] = {
1010 [NFQA_VERDICT_HDR] = { .len = sizeof(struct nfqnl_msg_verdict_hdr) },
1011 [NFQA_MARK] = { .type = NLA_U32 },
1014 static struct nfqnl_instance *
1015 verdict_instance_lookup(struct nfnl_queue_net *q, u16 queue_num, u32 nlportid)
1017 struct nfqnl_instance *queue;
1019 queue = instance_lookup(q, queue_num);
1021 return ERR_PTR(-ENODEV);
1023 if (queue->peer_portid != nlportid)
1024 return ERR_PTR(-EPERM);
1029 static struct nfqnl_msg_verdict_hdr*
1030 verdicthdr_get(const struct nlattr * const nfqa[])
1032 struct nfqnl_msg_verdict_hdr *vhdr;
1033 unsigned int verdict;
1035 if (!nfqa[NFQA_VERDICT_HDR])
1038 vhdr = nla_data(nfqa[NFQA_VERDICT_HDR]);
1039 verdict = ntohl(vhdr->verdict) & NF_VERDICT_MASK;
1040 if (verdict > NF_MAX_VERDICT || verdict == NF_STOLEN)
1045 static int nfq_id_after(unsigned int id, unsigned int max)
1047 return (int)(id - max) > 0;
1050 static int nfqnl_recv_verdict_batch(struct net *net, struct sock *ctnl,
1051 struct sk_buff *skb,
1052 const struct nlmsghdr *nlh,
1053 const struct nlattr * const nfqa[],
1054 struct netlink_ext_ack *extack)
1056 struct nfgenmsg *nfmsg = nlmsg_data(nlh);
1057 struct nf_queue_entry *entry, *tmp;
1058 unsigned int verdict, maxid;
1059 struct nfqnl_msg_verdict_hdr *vhdr;
1060 struct nfqnl_instance *queue;
1061 LIST_HEAD(batch_list);
1062 u16 queue_num = ntohs(nfmsg->res_id);
1063 struct nfnl_queue_net *q = nfnl_queue_pernet(net);
1065 queue = verdict_instance_lookup(q, queue_num,
1066 NETLINK_CB(skb).portid);
1068 return PTR_ERR(queue);
1070 vhdr = verdicthdr_get(nfqa);
1074 verdict = ntohl(vhdr->verdict);
1075 maxid = ntohl(vhdr->id);
1077 spin_lock_bh(&queue->lock);
1079 list_for_each_entry_safe(entry, tmp, &queue->queue_list, list) {
1080 if (nfq_id_after(entry->id, maxid))
1082 __dequeue_entry(queue, entry);
1083 list_add_tail(&entry->list, &batch_list);
1086 spin_unlock_bh(&queue->lock);
1088 if (list_empty(&batch_list))
1091 list_for_each_entry_safe(entry, tmp, &batch_list, list) {
1092 if (nfqa[NFQA_MARK])
1093 entry->skb->mark = ntohl(nla_get_be32(nfqa[NFQA_MARK]));
1094 nf_reinject(entry, verdict);
1099 static struct nf_conn *nfqnl_ct_parse(struct nfnl_ct_hook *nfnl_ct,
1100 const struct nlmsghdr *nlh,
1101 const struct nlattr * const nfqa[],
1102 struct nf_queue_entry *entry,
1103 enum ip_conntrack_info *ctinfo)
1107 ct = nfnl_ct->get_ct(entry->skb, ctinfo);
1111 if (nfnl_ct->parse(nfqa[NFQA_CT], ct) < 0)
1115 nfnl_ct->attach_expect(nfqa[NFQA_EXP], ct,
1116 NETLINK_CB(entry->skb).portid,
1121 static int nfqa_parse_bridge(struct nf_queue_entry *entry,
1122 const struct nlattr * const nfqa[])
1124 if (nfqa[NFQA_VLAN]) {
1125 struct nlattr *tb[NFQA_VLAN_MAX + 1];
1128 err = nla_parse_nested(tb, NFQA_VLAN_MAX, nfqa[NFQA_VLAN],
1129 nfqa_vlan_policy, NULL);
1133 if (!tb[NFQA_VLAN_TCI] || !tb[NFQA_VLAN_PROTO])
1136 entry->skb->vlan_tci = ntohs(nla_get_be16(tb[NFQA_VLAN_TCI]));
1137 entry->skb->vlan_proto = nla_get_be16(tb[NFQA_VLAN_PROTO]);
1140 if (nfqa[NFQA_L2HDR]) {
1141 int mac_header_len = entry->skb->network_header -
1142 entry->skb->mac_header;
1144 if (mac_header_len != nla_len(nfqa[NFQA_L2HDR]))
1146 else if (mac_header_len > 0)
1147 memcpy(skb_mac_header(entry->skb),
1148 nla_data(nfqa[NFQA_L2HDR]),
1155 static int nfqnl_recv_verdict(struct net *net, struct sock *ctnl,
1156 struct sk_buff *skb,
1157 const struct nlmsghdr *nlh,
1158 const struct nlattr * const nfqa[],
1159 struct netlink_ext_ack *extack)
1161 struct nfgenmsg *nfmsg = nlmsg_data(nlh);
1162 u_int16_t queue_num = ntohs(nfmsg->res_id);
1163 struct nfqnl_msg_verdict_hdr *vhdr;
1164 struct nfqnl_instance *queue;
1165 unsigned int verdict;
1166 struct nf_queue_entry *entry;
1167 enum ip_conntrack_info uninitialized_var(ctinfo);
1168 struct nfnl_ct_hook *nfnl_ct;
1169 struct nf_conn *ct = NULL;
1170 struct nfnl_queue_net *q = nfnl_queue_pernet(net);
1173 queue = verdict_instance_lookup(q, queue_num,
1174 NETLINK_CB(skb).portid);
1176 return PTR_ERR(queue);
1178 vhdr = verdicthdr_get(nfqa);
1182 verdict = ntohl(vhdr->verdict);
1184 entry = find_dequeue_entry(queue, ntohl(vhdr->id));
1188 /* rcu lock already held from nfnl->call_rcu. */
1189 nfnl_ct = rcu_dereference(nfnl_ct_hook);
1191 if (nfqa[NFQA_CT]) {
1192 if (nfnl_ct != NULL)
1193 ct = nfqnl_ct_parse(nfnl_ct, nlh, nfqa, entry, &ctinfo);
1196 if (entry->state.pf == PF_BRIDGE) {
1197 err = nfqa_parse_bridge(entry, nfqa);
1202 if (nfqa[NFQA_PAYLOAD]) {
1203 u16 payload_len = nla_len(nfqa[NFQA_PAYLOAD]);
1204 int diff = payload_len - entry->skb->len;
1206 if (nfqnl_mangle(nla_data(nfqa[NFQA_PAYLOAD]),
1207 payload_len, entry, diff) < 0)
1211 nfnl_ct->seq_adjust(entry->skb, ct, ctinfo, diff);
1214 if (nfqa[NFQA_MARK])
1215 entry->skb->mark = ntohl(nla_get_be32(nfqa[NFQA_MARK]));
1217 nf_reinject(entry, verdict);
1221 static int nfqnl_recv_unsupp(struct net *net, struct sock *ctnl,
1222 struct sk_buff *skb, const struct nlmsghdr *nlh,
1223 const struct nlattr * const nfqa[],
1224 struct netlink_ext_ack *extack)
1229 static const struct nla_policy nfqa_cfg_policy[NFQA_CFG_MAX+1] = {
1230 [NFQA_CFG_CMD] = { .len = sizeof(struct nfqnl_msg_config_cmd) },
1231 [NFQA_CFG_PARAMS] = { .len = sizeof(struct nfqnl_msg_config_params) },
1232 [NFQA_CFG_QUEUE_MAXLEN] = { .type = NLA_U32 },
1233 [NFQA_CFG_MASK] = { .type = NLA_U32 },
1234 [NFQA_CFG_FLAGS] = { .type = NLA_U32 },
1237 static const struct nf_queue_handler nfqh = {
1238 .outfn = nfqnl_enqueue_packet,
1239 .nf_hook_drop = nfqnl_nf_hook_drop,
1242 static int nfqnl_recv_config(struct net *net, struct sock *ctnl,
1243 struct sk_buff *skb, const struct nlmsghdr *nlh,
1244 const struct nlattr * const nfqa[],
1245 struct netlink_ext_ack *extack)
1247 struct nfgenmsg *nfmsg = nlmsg_data(nlh);
1248 u_int16_t queue_num = ntohs(nfmsg->res_id);
1249 struct nfqnl_instance *queue;
1250 struct nfqnl_msg_config_cmd *cmd = NULL;
1251 struct nfnl_queue_net *q = nfnl_queue_pernet(net);
1252 __u32 flags = 0, mask = 0;
1255 if (nfqa[NFQA_CFG_CMD]) {
1256 cmd = nla_data(nfqa[NFQA_CFG_CMD]);
1258 /* Obsolete commands without queue context */
1259 switch (cmd->command) {
1260 case NFQNL_CFG_CMD_PF_BIND: return 0;
1261 case NFQNL_CFG_CMD_PF_UNBIND: return 0;
1265 /* Check if we support these flags in first place, dependencies should
1266 * be there too not to break atomicity.
1268 if (nfqa[NFQA_CFG_FLAGS]) {
1269 if (!nfqa[NFQA_CFG_MASK]) {
1270 /* A mask is needed to specify which flags are being
1276 flags = ntohl(nla_get_be32(nfqa[NFQA_CFG_FLAGS]));
1277 mask = ntohl(nla_get_be32(nfqa[NFQA_CFG_MASK]));
1279 if (flags >= NFQA_CFG_F_MAX)
1282 #if !IS_ENABLED(CONFIG_NETWORK_SECMARK)
1283 if (flags & mask & NFQA_CFG_F_SECCTX)
1286 if ((flags & mask & NFQA_CFG_F_CONNTRACK) &&
1287 !rcu_access_pointer(nfnl_ct_hook)) {
1288 #ifdef CONFIG_MODULES
1289 nfnl_unlock(NFNL_SUBSYS_QUEUE);
1290 request_module("ip_conntrack_netlink");
1291 nfnl_lock(NFNL_SUBSYS_QUEUE);
1292 if (rcu_access_pointer(nfnl_ct_hook))
1300 queue = instance_lookup(q, queue_num);
1301 if (queue && queue->peer_portid != NETLINK_CB(skb).portid) {
1303 goto err_out_unlock;
1307 switch (cmd->command) {
1308 case NFQNL_CFG_CMD_BIND:
1311 goto err_out_unlock;
1313 queue = instance_create(q, queue_num,
1314 NETLINK_CB(skb).portid);
1315 if (IS_ERR(queue)) {
1316 ret = PTR_ERR(queue);
1317 goto err_out_unlock;
1320 case NFQNL_CFG_CMD_UNBIND:
1323 goto err_out_unlock;
1325 instance_destroy(q, queue);
1326 goto err_out_unlock;
1327 case NFQNL_CFG_CMD_PF_BIND:
1328 case NFQNL_CFG_CMD_PF_UNBIND:
1332 goto err_out_unlock;
1338 goto err_out_unlock;
1341 if (nfqa[NFQA_CFG_PARAMS]) {
1342 struct nfqnl_msg_config_params *params =
1343 nla_data(nfqa[NFQA_CFG_PARAMS]);
1345 nfqnl_set_mode(queue, params->copy_mode,
1346 ntohl(params->copy_range));
1349 if (nfqa[NFQA_CFG_QUEUE_MAXLEN]) {
1350 __be32 *queue_maxlen = nla_data(nfqa[NFQA_CFG_QUEUE_MAXLEN]);
1352 spin_lock_bh(&queue->lock);
1353 queue->queue_maxlen = ntohl(*queue_maxlen);
1354 spin_unlock_bh(&queue->lock);
1357 if (nfqa[NFQA_CFG_FLAGS]) {
1358 spin_lock_bh(&queue->lock);
1359 queue->flags &= ~mask;
1360 queue->flags |= flags & mask;
1361 spin_unlock_bh(&queue->lock);
1369 static const struct nfnl_callback nfqnl_cb[NFQNL_MSG_MAX] = {
1370 [NFQNL_MSG_PACKET] = { .call_rcu = nfqnl_recv_unsupp,
1371 .attr_count = NFQA_MAX, },
1372 [NFQNL_MSG_VERDICT] = { .call_rcu = nfqnl_recv_verdict,
1373 .attr_count = NFQA_MAX,
1374 .policy = nfqa_verdict_policy },
1375 [NFQNL_MSG_CONFIG] = { .call = nfqnl_recv_config,
1376 .attr_count = NFQA_CFG_MAX,
1377 .policy = nfqa_cfg_policy },
1378 [NFQNL_MSG_VERDICT_BATCH]={ .call_rcu = nfqnl_recv_verdict_batch,
1379 .attr_count = NFQA_MAX,
1380 .policy = nfqa_verdict_batch_policy },
1383 static const struct nfnetlink_subsystem nfqnl_subsys = {
1385 .subsys_id = NFNL_SUBSYS_QUEUE,
1386 .cb_count = NFQNL_MSG_MAX,
1390 #ifdef CONFIG_PROC_FS
1392 struct seq_net_private p;
1393 unsigned int bucket;
1396 static struct hlist_node *get_first(struct seq_file *seq)
1398 struct iter_state *st = seq->private;
1400 struct nfnl_queue_net *q;
1405 net = seq_file_net(seq);
1406 q = nfnl_queue_pernet(net);
1407 for (st->bucket = 0; st->bucket < INSTANCE_BUCKETS; st->bucket++) {
1408 if (!hlist_empty(&q->instance_table[st->bucket]))
1409 return q->instance_table[st->bucket].first;
1414 static struct hlist_node *get_next(struct seq_file *seq, struct hlist_node *h)
1416 struct iter_state *st = seq->private;
1417 struct net *net = seq_file_net(seq);
1421 struct nfnl_queue_net *q;
1423 if (++st->bucket >= INSTANCE_BUCKETS)
1426 q = nfnl_queue_pernet(net);
1427 h = q->instance_table[st->bucket].first;
1432 static struct hlist_node *get_idx(struct seq_file *seq, loff_t pos)
1434 struct hlist_node *head;
1435 head = get_first(seq);
1438 while (pos && (head = get_next(seq, head)))
1440 return pos ? NULL : head;
1443 static void *seq_start(struct seq_file *s, loff_t *pos)
1444 __acquires(nfnl_queue_pernet(seq_file_net(s))->instances_lock)
1446 spin_lock(&nfnl_queue_pernet(seq_file_net(s))->instances_lock);
1447 return get_idx(s, *pos);
1450 static void *seq_next(struct seq_file *s, void *v, loff_t *pos)
1453 return get_next(s, v);
1456 static void seq_stop(struct seq_file *s, void *v)
1457 __releases(nfnl_queue_pernet(seq_file_net(s))->instances_lock)
1459 spin_unlock(&nfnl_queue_pernet(seq_file_net(s))->instances_lock);
1462 static int seq_show(struct seq_file *s, void *v)
1464 const struct nfqnl_instance *inst = v;
1466 seq_printf(s, "%5u %6u %5u %1u %5u %5u %5u %8u %2d\n",
1468 inst->peer_portid, inst->queue_total,
1469 inst->copy_mode, inst->copy_range,
1470 inst->queue_dropped, inst->queue_user_dropped,
1471 inst->id_sequence, 1);
1475 static const struct seq_operations nfqnl_seq_ops = {
1482 static int nfqnl_open(struct inode *inode, struct file *file)
1484 return seq_open_net(inode, file, &nfqnl_seq_ops,
1485 sizeof(struct iter_state));
1488 static const struct file_operations nfqnl_file_ops = {
1489 .owner = THIS_MODULE,
1492 .llseek = seq_lseek,
1493 .release = seq_release_net,
1496 #endif /* PROC_FS */
1498 static int __net_init nfnl_queue_net_init(struct net *net)
1501 struct nfnl_queue_net *q = nfnl_queue_pernet(net);
1503 for (i = 0; i < INSTANCE_BUCKETS; i++)
1504 INIT_HLIST_HEAD(&q->instance_table[i]);
1506 spin_lock_init(&q->instances_lock);
1508 #ifdef CONFIG_PROC_FS
1509 if (!proc_create("nfnetlink_queue", 0440,
1510 net->nf.proc_netfilter, &nfqnl_file_ops))
1513 nf_register_queue_handler(net, &nfqh);
1517 static void __net_exit nfnl_queue_net_exit(struct net *net)
1519 nf_unregister_queue_handler(net);
1520 #ifdef CONFIG_PROC_FS
1521 remove_proc_entry("nfnetlink_queue", net->nf.proc_netfilter);
1525 static void nfnl_queue_net_exit_batch(struct list_head *net_exit_list)
1530 static struct pernet_operations nfnl_queue_net_ops = {
1531 .init = nfnl_queue_net_init,
1532 .exit = nfnl_queue_net_exit,
1533 .exit_batch = nfnl_queue_net_exit_batch,
1534 .id = &nfnl_queue_net_id,
1535 .size = sizeof(struct nfnl_queue_net),
1538 static int __init nfnetlink_queue_init(void)
1542 status = register_pernet_subsys(&nfnl_queue_net_ops);
1544 pr_err("nf_queue: failed to register pernet ops\n");
1548 netlink_register_notifier(&nfqnl_rtnl_notifier);
1549 status = nfnetlink_subsys_register(&nfqnl_subsys);
1551 pr_err("nf_queue: failed to create netlink socket\n");
1552 goto cleanup_netlink_notifier;
1555 status = register_netdevice_notifier(&nfqnl_dev_notifier);
1557 pr_err("nf_queue: failed to register netdevice notifier\n");
1558 goto cleanup_netlink_subsys;
1563 cleanup_netlink_subsys:
1564 nfnetlink_subsys_unregister(&nfqnl_subsys);
1565 cleanup_netlink_notifier:
1566 netlink_unregister_notifier(&nfqnl_rtnl_notifier);
1567 unregister_pernet_subsys(&nfnl_queue_net_ops);
1572 static void __exit nfnetlink_queue_fini(void)
1574 unregister_netdevice_notifier(&nfqnl_dev_notifier);
1575 nfnetlink_subsys_unregister(&nfqnl_subsys);
1576 netlink_unregister_notifier(&nfqnl_rtnl_notifier);
1577 unregister_pernet_subsys(&nfnl_queue_net_ops);
1579 rcu_barrier(); /* Wait for completion of call_rcu()'s */
1582 MODULE_DESCRIPTION("netfilter packet queue handler");
1583 MODULE_AUTHOR("Harald Welte <laforge@netfilter.org>");
1584 MODULE_LICENSE("GPL");
1585 MODULE_ALIAS_NFNL_SUBSYS(NFNL_SUBSYS_QUEUE);
1587 module_init(nfnetlink_queue_init);
1588 module_exit(nfnetlink_queue_fini);