xref: /netbsd-src/sys/netinet/in_offload.c (revision afab4e300d3a9fb07dd8c80daf53d0feb3345706)
1 /*	$NetBSD: in_offload.c,v 1.14 2020/03/27 16:34:58 jdolecek Exp $	*/
2 
3 /*
4  * Copyright (c)2005, 2006 YAMAMOTO Takashi,
5  * All rights reserved.
6  *
7  * Redistribution and use in source and binary forms, with or without
8  * modification, are permitted provided that the following conditions
9  * are met:
10  * 1. Redistributions of source code must retain the above copyright
11  *    notice, this list of conditions and the following disclaimer.
12  * 2. Redistributions in binary form must reproduce the above copyright
13  *    notice, this list of conditions and the following disclaimer in the
14  *    documentation and/or other materials provided with the distribution.
15  *
16  * THIS SOFTWARE IS PROVIDED BY THE AUTHOR AND CONTRIBUTORS ``AS IS'' AND
17  * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
18  * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
19  * ARE DISCLAIMED.  IN NO EVENT SHALL THE AUTHOR OR CONTRIBUTORS BE LIABLE
20  * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
21  * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
22  * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
23  * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
24  * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
25  * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
26  * SUCH DAMAGE.
27  */
28 
29 #include <sys/cdefs.h>
30 __KERNEL_RCSID(0, "$NetBSD: in_offload.c,v 1.14 2020/03/27 16:34:58 jdolecek Exp $");
31 
32 #include <sys/param.h>
33 #include <sys/mbuf.h>
34 
35 #include <net/if.h>
36 
37 #include <netinet/in.h>
38 #include <netinet/in_systm.h>
39 #include <netinet/ip.h>
40 #include <netinet/ip_var.h>
41 #include <netinet/tcp.h>
42 #include <netinet/in_offload.h>
43 
44 /*
45  * Handle M_CSUM_TSOv4 in software. Split the TCP payload in chunks of
46  * size MSS, and return mbuf chain consists of them.
47  */
48 struct mbuf *
49 tcp4_segment(struct mbuf *m, int off)
50 {
51 	int mss;
52 	int iphlen, thlen;
53 	int hlen, len;
54 	struct ip *ip;
55 	struct tcphdr *th;
56 	uint16_t ipid, phsum;
57 	uint32_t tcpseq;
58 	struct mbuf *hdr = NULL;
59 	struct mbuf *m0 = NULL;
60 	struct mbuf *prev = NULL;
61 	struct mbuf *n, *t;
62 	int nsegs;
63 
64 	KASSERT((m->m_flags & M_PKTHDR) != 0);
65 	KASSERT((m->m_pkthdr.csum_flags & M_CSUM_TSOv4) != 0);
66 
67 	m->m_pkthdr.csum_flags = 0;
68 
69 	len = m->m_pkthdr.len;
70 	KASSERT(len >= off + sizeof(*ip) + sizeof(*th));
71 
72 	hlen = off + sizeof(*ip);
73 	if (m->m_len < hlen) {
74 		m = m_pullup(m, hlen);
75 		if (m == NULL)
76 			goto quit;
77 	}
78 	ip = (void *)(mtod(m, char *) + off);
79 	iphlen = ip->ip_hl * 4;
80 	KASSERT(ip->ip_v == IPVERSION);
81 	KASSERT(iphlen >= sizeof(*ip));
82 	KASSERT(ip->ip_p == IPPROTO_TCP);
83 	ipid = ntohs(ip->ip_id);
84 
85 	hlen = off + iphlen + sizeof(*th);
86 	if (m->m_len < hlen) {
87 		m = m_pullup(m, hlen);
88 		if (m == NULL)
89 			goto quit;
90 	}
91 	th = (void *)(mtod(m, char *) + off + iphlen);
92 	tcpseq = ntohl(th->th_seq);
93 	thlen = th->th_off * 4;
94 	hlen = off + iphlen + thlen;
95 
96 	mss = m->m_pkthdr.segsz;
97 	KASSERT(mss != 0);
98 	KASSERT(len > hlen);
99 
100 	t = m_split(m, hlen, M_NOWAIT);
101 	if (t == NULL)
102 		goto quit;
103 	hdr = m;
104 	m = t;
105 
106 	len -= hlen;
107 	KASSERT(len % mss == 0);
108 
109 	ip = (void *)(mtod(hdr, char *) + off);
110 	ip->ip_len = htons(iphlen + thlen + mss);
111 	phsum = in_cksum_phdr(ip->ip_src.s_addr, ip->ip_dst.s_addr,
112 	    htons((uint16_t)(thlen + mss) + IPPROTO_TCP));
113 
114 	for (nsegs = len / mss; nsegs > 0; nsegs--) {
115 		if (nsegs > 1) {
116 			n = m_dup(hdr, 0, hlen, M_NOWAIT);
117 			if (n == NULL)
118 				goto quit;
119 		} else
120 			n = hdr;
121 		KASSERT(n->m_len == hlen); /* XXX */
122 
123 		if (nsegs > 1) {
124 			t = m_split(m, mss, M_NOWAIT);
125 			if (t == NULL) {
126 				m_freem(n);
127 				goto quit;
128 			}
129 		} else
130 			t = m;
131 		m_cat(n, m);
132 		m = t;
133 
134 		KASSERT(n->m_len >= hlen); /* XXX */
135 
136 		if (m0 == NULL)
137 			m0 = n;
138 
139 		if (prev != NULL)
140 			prev->m_nextpkt = n;
141 
142 		n->m_pkthdr.len = hlen + mss;
143 		n->m_nextpkt = NULL;	/* XXX */
144 
145 		ip = (void *)(mtod(n, char *) + off);
146 		ip->ip_id = htons(ipid);
147 		ip->ip_sum = 0;
148 		ip->ip_sum = in4_cksum(n, 0, off, iphlen);
149 
150 		th = (void *)(mtod(n, char *) + off + iphlen);
151 		th->th_seq = htonl(tcpseq);
152 		th->th_sum = phsum;
153 		th->th_sum = in4_cksum(n, 0, off + iphlen, thlen + mss);
154 
155 		tcpseq += mss;
156 		ipid++;
157 		prev = n;
158 	}
159 	return m0;
160 
161 quit:
162 	if (hdr != NULL)
163 		m_freem(hdr);
164 	if (m != NULL)
165 		m_freem(m);
166 	for (m = m0; m != NULL; m = n) {
167 		n = m->m_nextpkt;
168 		m_freem(m);
169 	}
170 
171 	return NULL;
172 }
173 
174 int
175 ip_tso_output(struct ifnet *ifp, struct mbuf *m, const struct sockaddr *sa,
176     struct rtentry *rt)
177 {
178 	struct mbuf *n;
179 	int error = 0;
180 
181 	m = tcp4_segment(m, 0);
182 	if (m == NULL)
183 		return ENOMEM;
184 	do {
185 		n = m->m_nextpkt;
186 		if (error == 0)
187 			error = ip_if_output(ifp, m, sa, rt);
188 		else
189 			m_freem(m);
190 		m = n;
191 	} while (m != NULL);
192 	return error;
193 }
194 
195 /*
196  * Compute now in software the IP and TCP/UDP checksums. Cancel the
197  * hardware offloading.
198  */
199 void
200 in_undefer_cksum(struct mbuf *mh, size_t hdrlen, int csum_flags)
201 {
202 	const size_t iphdrlen = M_CSUM_DATA_IPv4_IPHL(mh->m_pkthdr.csum_data);
203 	uint16_t csum;
204 	uint16_t ip_len;
205 	uint16_t *csump;
206 	struct mbuf *m = mh;
207 
208 	KASSERT(mh->m_flags & M_PKTHDR);
209 	KASSERT(mh->m_pkthdr.len > hdrlen);
210 	KASSERT((mh->m_pkthdr.csum_flags & csum_flags) == csum_flags);
211 
212 	/*
213 	 * Deal with prepended frame header as done by e.g. ether_output().
214 	 * If first mbuf in chain has just the header, use second mbuf
215 	 * for the actual checksum. in4_csum() expects the passed mbuf
216 	 * to have the whole (struct ip) area contiguous.
217 	 */
218 	if (m->m_len <= hdrlen) {
219 		hdrlen -= m->m_len;
220 		m = m->m_next;
221 		KASSERT(m != NULL);
222 	}
223 
224 	if (__predict_true(hdrlen + sizeof(struct ip) <= m->m_len)) {
225 		struct ip *ip = (struct ip *)(mtod(m, uint8_t *) + hdrlen);
226 
227 		ip_len = ip->ip_len;
228 		csump = &ip->ip_sum;
229 	} else {
230 		const size_t ip_len_offset =
231 		    hdrlen + offsetof(struct ip, ip_len);
232 
233 		m_copydata(m, ip_len_offset, sizeof(ip_len), &ip_len);
234 		csump = NULL;
235 	}
236 	ip_len = ntohs(ip_len);
237 
238 	if (csum_flags & M_CSUM_IPv4) {
239 		csum = in4_cksum(m, 0, hdrlen, iphdrlen);
240 		if (csump != NULL) {
241 			*csump = csum;
242 		} else {
243 			const size_t offset = hdrlen +
244 			    offsetof(struct ip, ip_sum);
245 
246 			m_copyback(m, offset, sizeof(uint16_t), &csum);
247 		}
248 	}
249 
250 	if (csum_flags & (M_CSUM_UDPv4|M_CSUM_TCPv4)) {
251 		size_t l4offset = hdrlen + iphdrlen;
252 
253 		csum = in4_cksum(m, 0, l4offset, ip_len - iphdrlen);
254 		if (csum == 0 && (csum_flags & M_CSUM_UDPv4) != 0)
255 			csum = 0xffff;
256 
257 		l4offset += M_CSUM_DATA_IPv4_OFFSET(m->m_pkthdr.csum_data);
258 
259 		if (__predict_true(l4offset + sizeof(uint16_t) <= m->m_len)) {
260 			*(uint16_t *)(mtod(m, char *) + l4offset) = csum;
261 		} else {
262 			m_copyback(m, l4offset, sizeof(csum), (void *)&csum);
263 		}
264 	}
265 
266 	mh->m_pkthdr.csum_flags ^= csum_flags;
267 }
268 
269 /*
270  * Compute now in software the TCP/UDP checksum. Cancel the hardware
271  * offloading.
272  */
273 void
274 in_undefer_cksum_tcpudp(struct mbuf *m)
275 {
276 	struct ip *ip;
277 	uint16_t csum, offset;
278 
279 	KASSERT((m->m_flags & M_PKTHDR) != 0);
280 	KASSERT((m->m_pkthdr.csum_flags & (M_CSUM_TCPv4|M_CSUM_UDPv4)) != 0);
281 	KASSERT((m->m_pkthdr.csum_flags & (M_CSUM_TCPv6|M_CSUM_UDPv6)) == 0);
282 
283 	ip = mtod(m, struct ip *);
284 	offset = ip->ip_hl << 2;
285 
286 	csum = in4_cksum(m, 0, offset, ntohs(ip->ip_len) - offset);
287 	if (csum == 0 && (m->m_pkthdr.csum_flags & M_CSUM_UDPv4) != 0)
288 		csum = 0xffff;
289 
290 	offset += M_CSUM_DATA_IPv4_OFFSET(m->m_pkthdr.csum_data);
291 
292 	if ((offset + sizeof(uint16_t)) <= m->m_len) {
293 		*(uint16_t *)(mtod(m, char *) + offset) = csum;
294 	} else {
295 		m_copyback(m, offset, sizeof(csum), (void *)&csum);
296 	}
297 }
298