1 /*
2  * (C) 2015 Red Hat GmbH
3  * Author: Florian Westphal <fw@strlen.de>
4  *
5  * This program is free software; you can redistribute it and/or modify
6  * it under the terms of the GNU General Public License version 2 as
7  * published by the Free Software Foundation.
8  */
9 
10 #include <linux/module.h>
11 #include <linux/static_key.h>
12 #include <linux/hash.h>
13 #include <linux/jhash.h>
14 #include <linux/if_vlan.h>
15 #include <linux/init.h>
16 #include <linux/skbuff.h>
17 #include <linux/netlink.h>
18 #include <linux/netfilter.h>
19 #include <linux/netfilter/nfnetlink.h>
20 #include <linux/netfilter/nf_tables.h>
21 #include <net/netfilter/nf_tables_core.h>
22 #include <net/netfilter/nf_tables.h>
23 
24 #define NFT_TRACETYPE_LL_HSIZE		20
25 #define NFT_TRACETYPE_NETWORK_HSIZE	40
26 #define NFT_TRACETYPE_TRANSPORT_HSIZE	20
27 
28 DEFINE_STATIC_KEY_FALSE(nft_trace_enabled);
29 EXPORT_SYMBOL_GPL(nft_trace_enabled);
30 
31 static int trace_fill_id(struct sk_buff *nlskb, struct sk_buff *skb)
32 {
33 	__be32 id;
34 
35 	/* using skb address as ID results in a limited number of
36 	 * values (and quick reuse).
37 	 *
38 	 * So we attempt to use as many skb members that will not
39 	 * change while skb is with netfilter.
40 	 */
41 	id = (__be32)jhash_2words(hash32_ptr(skb), skb_get_hash(skb),
42 				  skb->skb_iif);
43 
44 	return nla_put_be32(nlskb, NFTA_TRACE_ID, id);
45 }
46 
47 static int trace_fill_header(struct sk_buff *nlskb, u16 type,
48 			     const struct sk_buff *skb,
49 			     int off, unsigned int len)
50 {
51 	struct nlattr *nla;
52 
53 	if (len == 0)
54 		return 0;
55 
56 	nla = nla_reserve(nlskb, type, len);
57 	if (!nla || skb_copy_bits(skb, off, nla_data(nla), len))
58 		return -1;
59 
60 	return 0;
61 }
62 
63 static int nf_trace_fill_ll_header(struct sk_buff *nlskb,
64 				   const struct sk_buff *skb)
65 {
66 	struct vlan_ethhdr veth;
67 	int off;
68 
69 	BUILD_BUG_ON(sizeof(veth) > NFT_TRACETYPE_LL_HSIZE);
70 
71 	off = skb_mac_header(skb) - skb->data;
72 	if (off != -ETH_HLEN)
73 		return -1;
74 
75 	if (skb_copy_bits(skb, off, &veth, ETH_HLEN))
76 		return -1;
77 
78 	veth.h_vlan_proto = skb->vlan_proto;
79 	veth.h_vlan_TCI = htons(skb_vlan_tag_get(skb));
80 	veth.h_vlan_encapsulated_proto = skb->protocol;
81 
82 	return nla_put(nlskb, NFTA_TRACE_LL_HEADER, sizeof(veth), &veth);
83 }
84 
85 static int nf_trace_fill_dev_info(struct sk_buff *nlskb,
86 				  const struct net_device *indev,
87 				  const struct net_device *outdev)
88 {
89 	if (indev) {
90 		if (nla_put_be32(nlskb, NFTA_TRACE_IIF,
91 				 htonl(indev->ifindex)))
92 			return -1;
93 
94 		if (nla_put_be16(nlskb, NFTA_TRACE_IIFTYPE,
95 				 htons(indev->type)))
96 			return -1;
97 	}
98 
99 	if (outdev) {
100 		if (nla_put_be32(nlskb, NFTA_TRACE_OIF,
101 				 htonl(outdev->ifindex)))
102 			return -1;
103 
104 		if (nla_put_be16(nlskb, NFTA_TRACE_OIFTYPE,
105 				 htons(outdev->type)))
106 			return -1;
107 	}
108 
109 	return 0;
110 }
111 
112 static int nf_trace_fill_pkt_info(struct sk_buff *nlskb,
113 				  const struct nft_pktinfo *pkt)
114 {
115 	const struct sk_buff *skb = pkt->skb;
116 	int off = skb_network_offset(skb);
117 	unsigned int len, nh_end;
118 
119 	nh_end = pkt->tprot_set ? pkt->xt.thoff : skb->len;
120 	len = min_t(unsigned int, nh_end - skb_network_offset(skb),
121 		    NFT_TRACETYPE_NETWORK_HSIZE);
122 	if (trace_fill_header(nlskb, NFTA_TRACE_NETWORK_HEADER, skb, off, len))
123 		return -1;
124 
125 	if (pkt->tprot_set) {
126 		len = min_t(unsigned int, skb->len - pkt->xt.thoff,
127 			    NFT_TRACETYPE_TRANSPORT_HSIZE);
128 		if (trace_fill_header(nlskb, NFTA_TRACE_TRANSPORT_HEADER, skb,
129 				      pkt->xt.thoff, len))
130 			return -1;
131 	}
132 
133 	if (!skb_mac_header_was_set(skb))
134 		return 0;
135 
136 	if (skb_vlan_tag_get(skb))
137 		return nf_trace_fill_ll_header(nlskb, skb);
138 
139 	off = skb_mac_header(skb) - skb->data;
140 	len = min_t(unsigned int, -off, NFT_TRACETYPE_LL_HSIZE);
141 	return trace_fill_header(nlskb, NFTA_TRACE_LL_HEADER,
142 				 skb, off, len);
143 }
144 
145 static int nf_trace_fill_rule_info(struct sk_buff *nlskb,
146 				   const struct nft_traceinfo *info)
147 {
148 	if (!info->rule)
149 		return 0;
150 
151 	/* a continue verdict with ->type == RETURN means that this is
152 	 * an implicit return (end of chain reached).
153 	 *
154 	 * Since no rule matched, the ->rule pointer is invalid.
155 	 */
156 	if (info->type == NFT_TRACETYPE_RETURN &&
157 	    info->verdict->code == NFT_CONTINUE)
158 		return 0;
159 
160 	return nla_put_be64(nlskb, NFTA_TRACE_RULE_HANDLE,
161 			    cpu_to_be64(info->rule->handle),
162 			    NFTA_TRACE_PAD);
163 }
164 
165 static bool nft_trace_have_verdict_chain(struct nft_traceinfo *info)
166 {
167 	switch (info->type) {
168 	case NFT_TRACETYPE_RETURN:
169 	case NFT_TRACETYPE_RULE:
170 		break;
171 	default:
172 		return false;
173 	}
174 
175 	switch (info->verdict->code) {
176 	case NFT_JUMP:
177 	case NFT_GOTO:
178 		break;
179 	default:
180 		return false;
181 	}
182 
183 	return true;
184 }
185 
186 void nft_trace_notify(struct nft_traceinfo *info)
187 {
188 	const struct nft_pktinfo *pkt = info->pkt;
189 	struct nfgenmsg *nfmsg;
190 	struct nlmsghdr *nlh;
191 	struct sk_buff *skb;
192 	unsigned int size;
193 	u16 event;
194 
195 	if (!nfnetlink_has_listeners(nft_net(pkt), NFNLGRP_NFTRACE))
196 		return;
197 
198 	size = nlmsg_total_size(sizeof(struct nfgenmsg)) +
199 		nla_total_size(strlen(info->chain->table->name)) +
200 		nla_total_size(strlen(info->chain->name)) +
201 		nla_total_size_64bit(sizeof(__be64)) +	/* rule handle */
202 		nla_total_size(sizeof(__be32)) +	/* trace type */
203 		nla_total_size(0) +			/* VERDICT, nested */
204 			nla_total_size(sizeof(u32)) +	/* verdict code */
205 		nla_total_size(sizeof(u32)) +		/* id */
206 		nla_total_size(NFT_TRACETYPE_LL_HSIZE) +
207 		nla_total_size(NFT_TRACETYPE_NETWORK_HSIZE) +
208 		nla_total_size(NFT_TRACETYPE_TRANSPORT_HSIZE) +
209 		nla_total_size(sizeof(u32)) +		/* iif */
210 		nla_total_size(sizeof(__be16)) +	/* iiftype */
211 		nla_total_size(sizeof(u32)) +		/* oif */
212 		nla_total_size(sizeof(__be16)) +	/* oiftype */
213 		nla_total_size(sizeof(u32)) +		/* mark */
214 		nla_total_size(sizeof(u32)) +		/* nfproto */
215 		nla_total_size(sizeof(u32));		/* policy */
216 
217 	if (nft_trace_have_verdict_chain(info))
218 		size += nla_total_size(strlen(info->verdict->chain->name)); /* jump target */
219 
220 	skb = nlmsg_new(size, GFP_ATOMIC);
221 	if (!skb)
222 		return;
223 
224 	event = nfnl_msg_type(NFNL_SUBSYS_NFTABLES, NFT_MSG_TRACE);
225 	nlh = nlmsg_put(skb, 0, 0, event, sizeof(struct nfgenmsg), 0);
226 	if (!nlh)
227 		goto nla_put_failure;
228 
229 	nfmsg = nlmsg_data(nlh);
230 	nfmsg->nfgen_family	= info->basechain->type->family;
231 	nfmsg->version		= NFNETLINK_V0;
232 	nfmsg->res_id		= 0;
233 
234 	if (nla_put_be32(skb, NFTA_TRACE_NFPROTO, htonl(nft_pf(pkt))))
235 		goto nla_put_failure;
236 
237 	if (nla_put_be32(skb, NFTA_TRACE_TYPE, htonl(info->type)))
238 		goto nla_put_failure;
239 
240 	if (trace_fill_id(skb, pkt->skb))
241 		goto nla_put_failure;
242 
243 	if (nla_put_string(skb, NFTA_TRACE_CHAIN, info->chain->name))
244 		goto nla_put_failure;
245 
246 	if (nla_put_string(skb, NFTA_TRACE_TABLE, info->chain->table->name))
247 		goto nla_put_failure;
248 
249 	if (nf_trace_fill_rule_info(skb, info))
250 		goto nla_put_failure;
251 
252 	switch (info->type) {
253 	case NFT_TRACETYPE_UNSPEC:
254 	case __NFT_TRACETYPE_MAX:
255 		break;
256 	case NFT_TRACETYPE_RETURN:
257 	case NFT_TRACETYPE_RULE:
258 		if (nft_verdict_dump(skb, NFTA_TRACE_VERDICT, info->verdict))
259 			goto nla_put_failure;
260 		break;
261 	case NFT_TRACETYPE_POLICY:
262 		if (nla_put_be32(skb, NFTA_TRACE_POLICY,
263 				 htonl(info->basechain->policy)))
264 			goto nla_put_failure;
265 		break;
266 	}
267 
268 	if (pkt->skb->mark &&
269 	    nla_put_be32(skb, NFTA_TRACE_MARK, htonl(pkt->skb->mark)))
270 		goto nla_put_failure;
271 
272 	if (!info->packet_dumped) {
273 		if (nf_trace_fill_dev_info(skb, nft_in(pkt), nft_out(pkt)))
274 			goto nla_put_failure;
275 
276 		if (nf_trace_fill_pkt_info(skb, pkt))
277 			goto nla_put_failure;
278 		info->packet_dumped = true;
279 	}
280 
281 	nlmsg_end(skb, nlh);
282 	nfnetlink_send(skb, nft_net(pkt), 0, NFNLGRP_NFTRACE, 0, GFP_ATOMIC);
283 	return;
284 
285  nla_put_failure:
286 	WARN_ON_ONCE(1);
287 	kfree_skb(skb);
288 }
289 
290 void nft_trace_init(struct nft_traceinfo *info, const struct nft_pktinfo *pkt,
291 		    const struct nft_verdict *verdict,
292 		    const struct nft_chain *chain)
293 {
294 	info->basechain = nft_base_chain(chain);
295 	info->trace = true;
296 	info->packet_dumped = false;
297 	info->pkt = pkt;
298 	info->verdict = verdict;
299 }
300