xref: /netbsd-src/sys/net/ether_sw_offload.c (revision a7c54c617101b75c6f8bfb413b0d5655acf8410c)
1 /*	$NetBSD: ether_sw_offload.c,v 1.9 2024/09/15 08:33:13 andvar Exp $	*/
2 
3 /*
4  * Copyright (c) 2018 The NetBSD Foundation, Inc.
5  * All rights reserved.
6  *
7  * This code is derived from software contributed to The NetBSD Foundation
8  * by Rin Okuyama.
9  *
10  * Redistribution and use in source and binary forms, with or without
11  * modification, are permitted provided that the following conditions
12  * are met:
13  * 1. Redistributions of source code must retain the above copyright
14  *    notice, this list of conditions and the following disclaimer.
15  * 2. Redistributions in binary form must reproduce the above copyright
16  *    notice, this list of conditions and the following disclaimer in the
17  *    documentation and/or other materials provided with the distribution.
18  *
19  * THIS SOFTWARE IS PROVIDED BY THE NETBSD FOUNDATION, INC. AND CONTRIBUTORS
20  * ``AS IS'' AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED
21  * TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR
22  * PURPOSE ARE DISCLAIMED.  IN NO EVENT SHALL THE FOUNDATION OR CONTRIBUTORS
23  * BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR
24  * CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF
25  * SUBSTITUTE GOODS OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS
26  * INTERRUPTION) HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN
27  * CONTRACT, STRICT LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE)
28  * ARISING IN ANY WAY OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE
29  * POSSIBILITY OF SUCH DAMAGE.
30  */
31 
32 #ifdef _KERNEL_OPT
33 #include "opt_inet.h"
34 #endif
35 
36 #include <sys/cdefs.h>
37 __KERNEL_RCSID(0, "$NetBSD: ether_sw_offload.c,v 1.9 2024/09/15 08:33:13 andvar Exp $");
38 
39 #include <sys/param.h>
40 #include <sys/types.h>
41 #include <sys/mbuf.h>
42 #include <sys/syslog.h>
43 #include <sys/time.h>
44 
45 #include <net/if.h>
46 #include <net/if_ether.h>
47 #include <net/ether_sw_offload.h>
48 
49 #include <netinet/in.h>
50 #include <netinet/in_offload.h>
51 #include <netinet/ip.h>
52 #include <netinet/tcp.h>
53 #include <netinet/udp.h>
54 
55 #ifdef INET6
56 #include <netinet/ip6.h>
57 #include <netinet6/in6.h>
58 #include <netinet6/in6_offload.h>
59 #endif
60 
61 /*
62  * Limit error messages at most once per 10 seconds.
63  */
64 static const struct timeval eso_err_interval = {
65 	.tv_sec = 10,
66 	.tv_usec = 0,
67 };
68 static struct timeval eso_err_lasttime;
69 
70 /*
71  * Handle TX offload in software. For TSO, split the packet into
72  * chunks with payloads of size MSS. For checksum offload, update
73  * required checksum fields. The results are more than one packet
74  * in general. Return a mbuf queue consists of them.
75  */
76 
77 struct mbuf *
78 ether_sw_offload_tx(struct ifnet *ifp, struct mbuf *m)
79 {
80 	struct ether_header *ep;
81 	int flags, ehlen;
82 	uint16_t type;
83 #ifdef INET6
84 	bool v6;
85 #else
86 	bool v6 __diagused;
87 #endif
88 
89 	KASSERT(m->m_flags & M_PKTHDR);
90 	flags = m->m_pkthdr.csum_flags;
91 	if (flags == 0)
92 		goto done;
93 
94 	/* Sanity check */
95 	if (!TX_OFFLOAD_SUPPORTED(ifp->if_csum_flags_tx, flags))
96 		goto quit;
97 
98 	KASSERT(m->m_pkthdr.len >= sizeof(*ep));
99 	KASSERT(m->m_len >= sizeof(*ep));
100 	ep = mtod(m, struct ether_header *);
101 	switch (type = ntohs(ep->ether_type)) {
102 	case ETHERTYPE_IP:
103 	case ETHERTYPE_IPV6:
104 		ehlen = ETHER_HDR_LEN;
105 		break;
106 	case ETHERTYPE_VLAN:
107 		ehlen = ETHER_HDR_LEN + ETHER_VLAN_ENCAP_LEN;
108 		break;
109 	default:
110 		if (ratecheck(&eso_err_lasttime, &eso_err_interval))
111 			log(LOG_ERR, "%s: %s: dropping invalid frame "
112 			    "type 0x%04hx csum_flags 0x%08x\n",
113 			    __func__, ifp->if_xname, type, flags);
114 		goto quit;
115 	}
116 	KASSERT(m->m_pkthdr.len >= ehlen);
117 
118 	v6 = flags & (M_CSUM_TSOv6 | M_CSUM_TCPv6 | M_CSUM_UDPv6);
119 #ifndef INET6
120 	KASSERT(!v6);
121 #endif
122 
123 	if (flags & (M_CSUM_TSOv4 | M_CSUM_TSOv6)) {
124 		/*
125 		 * tcp[46]_segment() assume that size of payloads is
126 		 * a multiple of MSS. Further, tcp6_segment() assumes
127 		 * no extension headers.
128 		 *
129 		 * XXX Do we need some KASSERT's?
130 		 */
131 #ifdef INET6
132 		if (v6)
133 			return tcp6_segment(m, ehlen);
134 		else
135 #endif
136 			return tcp4_segment(m, ehlen);
137 	}
138 
139 #ifdef INET6
140 	if (v6)
141 		in6_undefer_cksum(m, ehlen, flags);
142 	else
143 #endif
144 		in_undefer_cksum(m, ehlen, flags);
145 done:
146 	m->m_pkthdr.csum_flags = 0;
147 	m->m_nextpkt = NULL;
148 	return m;
149 quit:
150 	m_freem(m);
151 	return NULL;
152 }
153 
154 /*
155  * Handle RX offload in software.
156  *
157  * XXX Fragmented packets or packets with IPv6 extension headers
158  * are not currently supported.
159  */
160 
161 struct mbuf *
162 ether_sw_offload_rx(struct ifnet *ifp, struct mbuf *m)
163 {
164 	struct ether_header *eh;
165 	struct ip *ip;
166 	struct tcphdr *th;
167 	struct udphdr *uh;
168 	uint16_t sum, osum;
169 	uint8_t proto;
170 	int flags, enabled, len, ehlen, iphlen, l4offset;
171 	bool v6;
172 
173 	flags = 0;
174 
175 	enabled = ifp->if_csum_flags_rx;
176 	if (!(enabled & (M_CSUM_IPv4 | M_CSUM_TCPv4 | M_CSUM_UDPv4 |
177 	    M_CSUM_TCPv6 | M_CSUM_UDPv6)))
178 		goto done;
179 
180 	KASSERT(m->m_flags & M_PKTHDR);
181 	len = m->m_pkthdr.len;
182 
183 	KASSERT(len >= sizeof(*eh));
184 	if (m->m_len < sizeof(*eh)) {
185 		m = m_pullup(m, sizeof(*eh));
186 		if (m == NULL)
187 			return NULL;
188 	}
189 	eh = mtod(m, struct ether_header *);
190 	switch (htons(eh->ether_type)) {
191 	case ETHERTYPE_IP:
192 	case ETHERTYPE_IPV6:
193 		ehlen = ETHER_HDR_LEN;
194 		break;
195 	case ETHERTYPE_VLAN:
196 		ehlen = ETHER_HDR_LEN + ETHER_VLAN_ENCAP_LEN;
197 		break;
198 	default:
199 		goto done;
200 	}
201 
202 	KASSERT(len >= ehlen);
203 	len = m->m_pkthdr.len - ehlen;
204 
205 	KASSERT(len >= sizeof(*ip));
206 	if (m->m_len < ehlen + sizeof(*ip)) {
207 		m = m_pullup(m, ehlen + sizeof(*ip));
208 		if (m == NULL)
209 			return NULL;
210 	}
211 	ip = (void *)(mtod(m, char *) + ehlen);
212 	v6 = (ip->ip_v != IPVERSION);
213 
214 	if (v6) {
215 #ifdef INET6
216 		struct ip6_hdr *ip6;
217 
218 		KASSERT(len >= sizeof(*ip6));
219 		if (m->m_len < ehlen + sizeof(*ip6)) {
220 			m = m_pullup(m, ehlen + sizeof(*ip6));
221 			if (m == NULL)
222 				return NULL;
223 		}
224 		ip6 = (void *)(mtod(m, char *) + ehlen);
225 		KASSERT((ip6->ip6_vfc & IPV6_VERSION_MASK) == IPV6_VERSION);
226 
227 		iphlen = sizeof(*ip6);
228 
229 		len -= iphlen;
230 
231 		proto = ip6->ip6_nxt;
232 		switch (proto) {
233 		case IPPROTO_TCP:
234 			if (!(enabled & M_CSUM_TCPv6))
235 				goto done;
236 			break;
237 		case IPPROTO_UDP:
238 			if (!(enabled & M_CSUM_UDPv6))
239 				goto done;
240 			break;
241 		default:
242 			/* XXX Extension headers are not supported. */
243 			goto done;
244 		}
245 
246 		sum = in6_cksum_phdr(&ip6->ip6_src, &ip6->ip6_dst, htonl(len),
247 		    htonl(proto));
248 #else
249 		goto done;
250 #endif
251 	} else {
252 		if (enabled & M_CSUM_IPv4)
253 			flags |= M_CSUM_IPv4;
254 
255 		iphlen = ip->ip_hl << 2;
256 		KASSERT(iphlen >= sizeof(*ip));
257 
258 		len -= iphlen;
259 		KASSERT(len >= 0);
260 
261 		if (in4_cksum(m, 0, ehlen, iphlen) != 0) {
262 			if (enabled & M_CSUM_IPv4)
263 				flags |= M_CSUM_IPv4_BAD;
264 			/* Broken. Do not check further. */
265 			goto done;
266 		}
267 
268 		/* Check if fragmented. */
269 		if (ntohs(ip->ip_off) & ~(IP_DF | IP_RF))
270 			goto done;
271 
272 		proto = ip->ip_p;
273 		switch (proto) {
274 		case IPPROTO_TCP:
275 			if (!(enabled & M_CSUM_TCPv4))
276 				goto done;
277 			break;
278 		case IPPROTO_UDP:
279 			if (!(enabled & M_CSUM_UDPv4))
280 				goto done;
281 			break;
282 		default:
283 			goto done;
284 		}
285 
286 		sum = in_cksum_phdr(ip->ip_src.s_addr, ip->ip_dst.s_addr,
287 		    htons((uint16_t)len + proto));
288 	}
289 
290 	l4offset = ehlen + iphlen;
291 	switch (proto) {
292 	case IPPROTO_TCP:
293 		KASSERT(len >= sizeof(*th));
294 		if (m->m_len < l4offset + sizeof(*th)) {
295 			m = m_pullup(m, l4offset + sizeof(*th));
296 			if (m == NULL)
297 				return NULL;
298 		}
299 		th = (void *)(mtod(m, char *) + l4offset);
300 		osum = th->th_sum;
301 		th->th_sum = sum;
302 #ifdef INET6
303 		if (v6) {
304 			flags |= M_CSUM_TCPv6;
305 			sum = in6_cksum(m, 0, l4offset, len);
306 		} else
307 #endif
308 		{
309 			flags |= M_CSUM_TCPv4;
310 			sum = in4_cksum(m, 0, l4offset, len);
311 		}
312 		if (sum != osum)
313 			flags |= M_CSUM_TCP_UDP_BAD;
314 		th->th_sum = osum;
315 		break;
316 	case IPPROTO_UDP:
317 		KASSERT(len >= sizeof(*uh));
318 		if (m->m_len < l4offset + sizeof(*uh)) {
319 			m = m_pullup(m, l4offset + sizeof(*uh));
320 			if (m == NULL)
321 				return NULL;
322 		}
323 		uh = (void *)(mtod(m, char *) + l4offset);
324 		osum = uh->uh_sum;
325 		if (osum == 0)
326 			break;
327 		uh->uh_sum = sum;
328 #ifdef INET6
329 		if (v6) {
330 			flags |= M_CSUM_UDPv6;
331 			sum = in6_cksum(m, 0, l4offset, len);
332 		} else
333 #endif
334 		{
335 			flags |= M_CSUM_UDPv4;
336 			sum = in4_cksum(m, 0, l4offset, len);
337 		}
338 		if (sum == 0)
339 			sum = 0xffff;
340 		if (sum != osum)
341 			flags |= M_CSUM_TCP_UDP_BAD;
342 		uh->uh_sum = osum;
343 		break;
344 	default:
345 		panic("%s: impossible", __func__);
346 	}
347 
348 done:
349 	m->m_pkthdr.csum_flags = flags;
350 	return m;
351 }
352