[linux-block.git] / net / netfilter / nf_tables_trace.c

// SPDX-License-Identifier: GPL-2.0-only
/*
 * (C) 2015 Red Hat GmbH
 * Author: Florian Westphal <fw@strlen.de>
 */

#include <linux/module.h>
#include <linux/static_key.h>
#include <linux/hash.h>
#include <linux/jhash.h>
#include <linux/if_vlan.h>
#include <linux/init.h>
#include <linux/skbuff.h>
#include <linux/netlink.h>
#include <linux/netfilter.h>
#include <linux/netfilter/nfnetlink.h>
#include <linux/netfilter/nf_tables.h>
#include <net/netfilter/nf_tables_core.h>
#include <net/netfilter/nf_tables.h>

#define NFT_TRACETYPE_LL_HSIZE		20
#define NFT_TRACETYPE_NETWORK_HSIZE	40
#define NFT_TRACETYPE_TRANSPORT_HSIZE	20

DEFINE_STATIC_KEY_FALSE(nft_trace_enabled);
EXPORT_SYMBOL_GPL(nft_trace_enabled);

static int trace_fill_id(struct sk_buff *nlskb, struct sk_buff *skb)
{
	__be32 id;

	/* using skb address as ID results in a limited number of
	 * values (and quick reuse).
	 *
	 * So we attempt to use as many skb members that will not
	 * change while skb is with netfilter.
	 */
	id = (__be32)jhash_2words(hash32_ptr(skb), skb_get_hash(skb),
				  skb->skb_iif);

	return nla_put_be32(nlskb, NFTA_TRACE_ID, id);
}

static int trace_fill_header(struct sk_buff *nlskb, u16 type,
			     const struct sk_buff *skb,
			     int off, unsigned int len)
{
	struct nlattr *nla;

	if (len == 0)
		return 0;

	nla = nla_reserve(nlskb, type, len);
	if (!nla || skb_copy_bits(skb, off, nla_data(nla), len))
		return -1;

	return 0;
}

static int nf_trace_fill_ll_header(struct sk_buff *nlskb,
				   const struct sk_buff *skb)
{
	struct vlan_ethhdr veth;
	int off;

	BUILD_BUG_ON(sizeof(veth) > NFT_TRACETYPE_LL_HSIZE);

	off = skb_mac_header(skb) - skb->data;
	if (off != -ETH_HLEN)
		return -1;

	if (skb_copy_bits(skb, off, &veth, ETH_HLEN))
		return -1;

	veth.h_vlan_proto = skb->vlan_proto;
	veth.h_vlan_TCI = htons(skb_vlan_tag_get(skb));
	veth.h_vlan_encapsulated_proto = skb->protocol;

	return nla_put(nlskb, NFTA_TRACE_LL_HEADER, sizeof(veth), &veth);
}

static int nf_trace_fill_dev_info(struct sk_buff *nlskb,
				  const struct net_device *indev,
				  const struct net_device *outdev)
{
	if (indev) {
		if (nla_put_be32(nlskb, NFTA_TRACE_IIF,
				 htonl(indev->ifindex)))
			return -1;

		if (nla_put_be16(nlskb, NFTA_TRACE_IIFTYPE,
				 htons(indev->type)))
			return -1;
	}

	if (outdev) {
		if (nla_put_be32(nlskb, NFTA_TRACE_OIF,
				 htonl(outdev->ifindex)))
			return -1;

		if (nla_put_be16(nlskb, NFTA_TRACE_OIFTYPE,
				 htons(outdev->type)))
			return -1;
	}

	return 0;
}

static int nf_trace_fill_pkt_info(struct sk_buff *nlskb,
				  const struct nft_pktinfo *pkt)
{
	const struct sk_buff *skb = pkt->skb;
	int off = skb_network_offset(skb);
	unsigned int len, nh_end;

	nh_end = pkt->tprot_set ? pkt->xt.thoff : skb->len;
	len = min_t(unsigned int, nh_end - skb_network_offset(skb),
		    NFT_TRACETYPE_NETWORK_HSIZE);
	if (trace_fill_header(nlskb, NFTA_TRACE_NETWORK_HEADER, skb, off, len))
		return -1;

	if (pkt->tprot_set) {
		len = min_t(unsigned int, skb->len - pkt->xt.thoff,
			    NFT_TRACETYPE_TRANSPORT_HSIZE);
		if (trace_fill_header(nlskb, NFTA_TRACE_TRANSPORT_HEADER, skb,
				      pkt->xt.thoff, len))
			return -1;
	}

	if (!skb_mac_header_was_set(skb))
		return 0;

	if (skb_vlan_tag_get(skb))
		return nf_trace_fill_ll_header(nlskb, skb);

	off = skb_mac_header(skb) - skb->data;
	len = min_t(unsigned int, -off, NFT_TRACETYPE_LL_HSIZE);
	return trace_fill_header(nlskb, NFTA_TRACE_LL_HEADER,
				 skb, off, len);
}

static int nf_trace_fill_rule_info(struct sk_buff *nlskb,
				   const struct nft_traceinfo *info)
{
	if (!info->rule)
		return 0;

	/* a continue verdict with ->type == RETURN means that this is
	 * an implicit return (end of chain reached).
	 *
	 * Since no rule matched, the ->rule pointer is invalid.
	 */
	if (info->type == NFT_TRACETYPE_RETURN &&
	    info->verdict->code == NFT_CONTINUE)
		return 0;

	return nla_put_be64(nlskb, NFTA_TRACE_RULE_HANDLE,
			    cpu_to_be64(info->rule->handle),
			    NFTA_TRACE_PAD);
}

static bool nft_trace_have_verdict_chain(struct nft_traceinfo *info)
{
	switch (info->type) {
	case NFT_TRACETYPE_RETURN:
	case NFT_TRACETYPE_RULE:
		break;
	default:
		return false;
	}

	switch (info->verdict->code) {
	case NFT_JUMP:
	case NFT_GOTO:
		break;
	default:
		return false;
	}

	return true;
}

void nft_trace_notify(struct nft_traceinfo *info)
{
	const struct nft_pktinfo *pkt = info->pkt;
	struct nlmsghdr *nlh;
	struct sk_buff *skb;
	unsigned int size;
	u16 event;

	if (!nfnetlink_has_listeners(nft_net(pkt), NFNLGRP_NFTRACE))
		return;

	size = nlmsg_total_size(sizeof(struct nfgenmsg)) +
		nla_total_size(strlen(info->chain->table->name)) +
		nla_total_size(strlen(info->chain->name)) +
		nla_total_size_64bit(sizeof(__be64)) +	/* rule handle */
		nla_total_size(sizeof(__be32)) +	/* trace type */
		nla_total_size(0) +			/* VERDICT, nested */
			nla_total_size(sizeof(u32)) +	/* verdict code */
		nla_total_size(sizeof(u32)) +		/* id */
		nla_total_size(NFT_TRACETYPE_LL_HSIZE) +
		nla_total_size(NFT_TRACETYPE_NETWORK_HSIZE) +
		nla_total_size(NFT_TRACETYPE_TRANSPORT_HSIZE) +
		nla_total_size(sizeof(u32)) +		/* iif */
		nla_total_size(sizeof(__be16)) +	/* iiftype */
		nla_total_size(sizeof(u32)) +		/* oif */
		nla_total_size(sizeof(__be16)) +	/* oiftype */
		nla_total_size(sizeof(u32)) +		/* mark */
		nla_total_size(sizeof(u32)) +		/* nfproto */
		nla_total_size(sizeof(u32));		/* policy */

	if (nft_trace_have_verdict_chain(info))
		size += nla_total_size(strlen(info->verdict->chain->name)); /* jump target */

	skb = nlmsg_new(size, GFP_ATOMIC);
	if (!skb)
		return;

	event = nfnl_msg_type(NFNL_SUBSYS_NFTABLES, NFT_MSG_TRACE);
	nlh = nfnl_msg_put(skb, 0, 0, event, 0, info->basechain->type->family,
			   NFNETLINK_V0, 0);
	if (!nlh)
		goto nla_put_failure;

	if (nla_put_be32(skb, NFTA_TRACE_NFPROTO, htonl(nft_pf(pkt))))
		goto nla_put_failure;

	if (nla_put_be32(skb, NFTA_TRACE_TYPE, htonl(info->type)))
		goto nla_put_failure;

	if (trace_fill_id(skb, pkt->skb))
		goto nla_put_failure;

	if (nla_put_string(skb, NFTA_TRACE_CHAIN, info->chain->name))
		goto nla_put_failure;

	if (nla_put_string(skb, NFTA_TRACE_TABLE, info->chain->table->name))
		goto nla_put_failure;

	if (nf_trace_fill_rule_info(skb, info))
		goto nla_put_failure;

	switch (info->type) {
	case NFT_TRACETYPE_UNSPEC:
	case __NFT_TRACETYPE_MAX:
		break;
	case NFT_TRACETYPE_RETURN:
	case NFT_TRACETYPE_RULE:
		if (nft_verdict_dump(skb, NFTA_TRACE_VERDICT, info->verdict))
			goto nla_put_failure;
		break;
	case NFT_TRACETYPE_POLICY:
		if (nla_put_be32(skb, NFTA_TRACE_POLICY,
				 htonl(info->basechain->policy)))
			goto nla_put_failure;
		break;
	}

	if (pkt->skb->mark &&
	    nla_put_be32(skb, NFTA_TRACE_MARK, htonl(pkt->skb->mark)))
		goto nla_put_failure;

	if (!info->packet_dumped) {
		if (nf_trace_fill_dev_info(skb, nft_in(pkt), nft_out(pkt)))
			goto nla_put_failure;

		if (nf_trace_fill_pkt_info(skb, pkt))
			goto nla_put_failure;
		info->packet_dumped = true;
	}

	nlmsg_end(skb, nlh);
	nfnetlink_send(skb, nft_net(pkt), 0, NFNLGRP_NFTRACE, 0, GFP_ATOMIC);
	return;

 nla_put_failure:
	WARN_ON_ONCE(1);
	kfree_skb(skb);
}

void nft_trace_init(struct nft_traceinfo *info, const struct nft_pktinfo *pkt,
		    const struct nft_verdict *verdict,
		    const struct nft_chain *chain)
{
	info->basechain = nft_base_chain(chain);
	info->trace = true;
	info->packet_dumped = false;
	info->pkt = pkt;
	info->verdict = verdict;
}
Commit	Line	Data
d2912cb1	1	// SPDX-License-Identifier: GPL-2.0-only
33d5a7b1 FW	2	/*
	3	* (C) 2015 Red Hat GmbH
	4	* Author: Florian Westphal <fw@strlen.de>
33d5a7b1 FW	5	*/
	6
	7	#include <linux/module.h>
e639f7ab	8	#include <linux/static_key.h>
33d5a7b1 FW	9	#include <linux/hash.h>
	10	#include <linux/jhash.h>
	11	#include <linux/if_vlan.h>
	12	#include <linux/init.h>
	13	#include <linux/skbuff.h>
	14	#include <linux/netlink.h>
	15	#include <linux/netfilter.h>
	16	#include <linux/netfilter/nfnetlink.h>
	17	#include <linux/netfilter/nf_tables.h>
	18	#include <net/netfilter/nf_tables_core.h>
	19	#include <net/netfilter/nf_tables.h>
	20
	21	#define NFT_TRACETYPE_LL_HSIZE 20
	22	#define NFT_TRACETYPE_NETWORK_HSIZE 40
	23	#define NFT_TRACETYPE_TRANSPORT_HSIZE 20
	24
e639f7ab FW	25	DEFINE_STATIC_KEY_FALSE(nft_trace_enabled);
	26	EXPORT_SYMBOL_GPL(nft_trace_enabled);
	27
33d5a7b1 FW	28	static int trace_fill_id(struct sk_buff nlskb, struct sk_buff skb)
	29	{
	30	__be32 id;
	31
	32	/* using skb address as ID results in a limited number of
	33	* values (and quick reuse).
	34	*
	35	* So we attempt to use as many skb members that will not
	36	* change while skb is with netfilter.
	37	*/
	38	id = (__be32)jhash_2words(hash32_ptr(skb), skb_get_hash(skb),
	39	skb->skb_iif);
	40
	41	return nla_put_be32(nlskb, NFTA_TRACE_ID, id);
	42	}
	43
	44	static int trace_fill_header(struct sk_buff *nlskb, u16 type,
	45	const struct sk_buff *skb,
	46	int off, unsigned int len)
	47	{
	48	struct nlattr *nla;
	49
	50	if (len == 0)
	51	return 0;
	52
	53	nla = nla_reserve(nlskb, type, len);
	54	if (!nla \|\| skb_copy_bits(skb, off, nla_data(nla), len))
	55	return -1;
	56
	57	return 0;
	58	}
	59
	60	static int nf_trace_fill_ll_header(struct sk_buff *nlskb,
	61	const struct sk_buff *skb)
	62	{
	63	struct vlan_ethhdr veth;
	64	int off;
	65
	66	BUILD_BUG_ON(sizeof(veth) > NFT_TRACETYPE_LL_HSIZE);
	67
	68	off = skb_mac_header(skb) - skb->data;
	69	if (off != -ETH_HLEN)
	70	return -1;
	71
	72	if (skb_copy_bits(skb, off, &veth, ETH_HLEN))
	73	return -1;
	74
	75	veth.h_vlan_proto = skb->vlan_proto;
	76	veth.h_vlan_TCI = htons(skb_vlan_tag_get(skb));
	77	veth.h_vlan_encapsulated_proto = skb->protocol;
	78
	79	return nla_put(nlskb, NFTA_TRACE_LL_HEADER, sizeof(veth), &veth);
	80	}
	81
	82	static int nf_trace_fill_dev_info(struct sk_buff *nlskb,
	83	const struct net_device *indev,
	84	const struct net_device *outdev)
	85	{
	86	if (indev) {
	87	if (nla_put_be32(nlskb, NFTA_TRACE_IIF,
	88	htonl(indev->ifindex)))
	89	return -1;
	90
	91	if (nla_put_be16(nlskb, NFTA_TRACE_IIFTYPE,
92	htons(indev->type)))
93	return -1;
94	}
95
96	if (outdev) {
97	if (nla_put_be32(nlskb, NFTA_TRACE_OIF,
98	htonl(outdev->ifindex)))
99	return -1;
100
101	if (nla_put_be16(nlskb, NFTA_TRACE_OIFTYPE,
102	htons(outdev->type)))
103	return -1;
104	}
105
106	return 0;
107	}
108
109	static int nf_trace_fill_pkt_info(struct sk_buff *nlskb,
110	const struct nft_pktinfo *pkt)
111	{
112	const struct sk_buff *skb = pkt->skb;
33d5a7b1	113	int off = skb_network_offset(skb);
a20877b5	114	unsigned int len, nh_end;
33d5a7b1	115
a20877b5 LZ	116	nh_end = pkt->tprot_set ? pkt->xt.thoff : skb->len;
	117	len = min_t(unsigned int, nh_end - skb_network_offset(skb),
	118	NFT_TRACETYPE_NETWORK_HSIZE);
33d5a7b1 FW	119	if (trace_fill_header(nlskb, NFTA_TRACE_NETWORK_HEADER, skb, off, len))
	120	return -1;
	121
a20877b5 LZ	122	if (pkt->tprot_set) {
	123	len = min_t(unsigned int, skb->len - pkt->xt.thoff,
	124	NFT_TRACETYPE_TRANSPORT_HSIZE);
	125	if (trace_fill_header(nlskb, NFTA_TRACE_TRANSPORT_HEADER, skb,
	126	pkt->xt.thoff, len))
	127	return -1;
	128	}
33d5a7b1 FW	129
	130	if (!skb_mac_header_was_set(skb))
	131	return 0;
	132
	133	if (skb_vlan_tag_get(skb))
	134	return nf_trace_fill_ll_header(nlskb, skb);
	135
	136	off = skb_mac_header(skb) - skb->data;
	137	len = min_t(unsigned int, -off, NFT_TRACETYPE_LL_HSIZE);
	138	return trace_fill_header(nlskb, NFTA_TRACE_LL_HEADER,
	139	skb, off, len);
	140	}
	141
	142	static int nf_trace_fill_rule_info(struct sk_buff *nlskb,
	143	const struct nft_traceinfo *info)
	144	{
	145	if (!info->rule)
	146	return 0;
	147
	148	/* a continue verdict with ->type == RETURN means that this is
	149	* an implicit return (end of chain reached).
	150	*
	151	* Since no rule matched, the ->rule pointer is invalid.
	152	*/
	153	if (info->type == NFT_TRACETYPE_RETURN &&
	154	info->verdict->code == NFT_CONTINUE)
	155	return 0;
	156
	157	return nla_put_be64(nlskb, NFTA_TRACE_RULE_HANDLE,
b46f6ded ND	158	cpu_to_be64(info->rule->handle),
b46f6ded ND	159	NFTA_TRACE_PAD);
33d5a7b1 FW	160	}
33d5a7b1 FW	161
b7263e07 PS	162	static bool nft_trace_have_verdict_chain(struct nft_traceinfo *info)
	163	{
	164	switch (info->type) {
	165	case NFT_TRACETYPE_RETURN:
	166	case NFT_TRACETYPE_RULE:
	167	break;
	168	default:
	169	return false;
	170	}
	171
	172	switch (info->verdict->code) {
	173	case NFT_JUMP:
	174	case NFT_GOTO:
	175	break;
	176	default:
	177	return false;
	178	}
	179
	180	return true;
	181	}
	182
33d5a7b1 FW	183	void nft_trace_notify(struct nft_traceinfo *info)
	184	{
	185	const struct nft_pktinfo *pkt = info->pkt;
33d5a7b1 FW	186	struct nlmsghdr *nlh;
	187	struct sk_buff *skb;
	188	unsigned int size;
dedb67c4	189	u16 event;
33d5a7b1	190
0e5a1c7e	191	if (!nfnetlink_has_listeners(nft_net(pkt), NFNLGRP_NFTRACE))
33d5a7b1 FW	192	return;
	193
	194	size = nlmsg_total_size(sizeof(struct nfgenmsg)) +
e46abbcc	195	nla_total_size(strlen(info->chain->table->name)) +
b7263e07	196	nla_total_size(strlen(info->chain->name)) +
b46f6ded	197	nla_total_size_64bit(sizeof(__be64)) + /* rule handle */
33d5a7b1 FW	198	nla_total_size(sizeof(__be32)) + /* trace type */
	199	nla_total_size(0) + /* VERDICT, nested */
	200	nla_total_size(sizeof(u32)) + /* verdict code */
33d5a7b1 FW	201	nla_total_size(sizeof(u32)) + /* id */
	202	nla_total_size(NFT_TRACETYPE_LL_HSIZE) +
	203	nla_total_size(NFT_TRACETYPE_NETWORK_HSIZE) +
	204	nla_total_size(NFT_TRACETYPE_TRANSPORT_HSIZE) +
	205	nla_total_size(sizeof(u32)) + /* iif */
	206	nla_total_size(sizeof(__be16)) + /* iiftype */
	207	nla_total_size(sizeof(u32)) + /* oif */
	208	nla_total_size(sizeof(__be16)) + /* oiftype */
	209	nla_total_size(sizeof(u32)) + /* mark */
	210	nla_total_size(sizeof(u32)) + /* nfproto */
	211	nla_total_size(sizeof(u32)); /* policy */
	212
b7263e07 PS	213	if (nft_trace_have_verdict_chain(info))
	214	size += nla_total_size(strlen(info->verdict->chain->name)); /* jump target */
	215
33d5a7b1 FW	216	skb = nlmsg_new(size, GFP_ATOMIC);
	217	if (!skb)
	218	return;
	219
dedb67c4	220	event = nfnl_msg_type(NFNL_SUBSYS_NFTABLES, NFT_MSG_TRACE);
19c28b13 PNA	221	nlh = nfnl_msg_put(skb, 0, 0, event, 0, info->basechain->type->family,
19c28b13 PNA	222	NFNETLINK_V0, 0);
33d5a7b1 FW	223	if (!nlh)
	224	goto nla_put_failure;
	225
0e5a1c7e	226	if (nla_put_be32(skb, NFTA_TRACE_NFPROTO, htonl(nft_pf(pkt))))
33d5a7b1 FW	227	goto nla_put_failure;
	228
	229	if (nla_put_be32(skb, NFTA_TRACE_TYPE, htonl(info->type)))
	230	goto nla_put_failure;
	231
	232	if (trace_fill_id(skb, pkt->skb))
	233	goto nla_put_failure;
	234
6e692678 PS	235	if (nla_put_string(skb, NFTA_TRACE_CHAIN, info->chain->name))
	236	goto nla_put_failure;
	237
	238	if (nla_put_string(skb, NFTA_TRACE_TABLE, info->chain->table->name))
	239	goto nla_put_failure;
33d5a7b1 FW	240
	241	if (nf_trace_fill_rule_info(skb, info))
	242	goto nla_put_failure;
	243
	244	switch (info->type) {
	245	case NFT_TRACETYPE_UNSPEC:
	246	case __NFT_TRACETYPE_MAX:
	247	break;
	248	case NFT_TRACETYPE_RETURN:
	249	case NFT_TRACETYPE_RULE:
	250	if (nft_verdict_dump(skb, NFTA_TRACE_VERDICT, info->verdict))
	251	goto nla_put_failure;
	252	break;
	253	case NFT_TRACETYPE_POLICY:
	254	if (nla_put_be32(skb, NFTA_TRACE_POLICY,
5210d393	255	htonl(info->basechain->policy)))
33d5a7b1 FW	256	goto nla_put_failure;
	257	break;
	258	}
	259
	260	if (pkt->skb->mark &&
	261	nla_put_be32(skb, NFTA_TRACE_MARK, htonl(pkt->skb->mark)))
	262	goto nla_put_failure;
	263
	264	if (!info->packet_dumped) {
0e5a1c7e	265	if (nf_trace_fill_dev_info(skb, nft_in(pkt), nft_out(pkt)))
33d5a7b1 FW	266	goto nla_put_failure;
	267
	268	if (nf_trace_fill_pkt_info(skb, pkt))
	269	goto nla_put_failure;
	270	info->packet_dumped = true;
	271	}
	272
	273	nlmsg_end(skb, nlh);
0e5a1c7e	274	nfnetlink_send(skb, nft_net(pkt), 0, NFNLGRP_NFTRACE, 0, GFP_ATOMIC);
33d5a7b1 FW	275	return;
	276
	277	nla_put_failure:
	278	WARN_ON_ONCE(1);
	279	kfree_skb(skb);
	280	}
	281
	282	void nft_trace_init(struct nft_traceinfo info, const struct nft_pktinfo pkt,
	283	const struct nft_verdict *verdict,
	284	const struct nft_chain *chain)
	285	{
	286	info->basechain = nft_base_chain(chain);
	287	info->trace = true;
	288	info->packet_dumped = false;
	289	info->pkt = pkt;
	290	info->verdict = verdict;
	291	}