Home | History | Annotate | Line # | Download | only in netinet
in_offload.c revision 1.3
      1 /*	$NetBSD: in_offload.c,v 1.3 2010/12/11 22:37:46 matt 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 1.3 2010/12/11 22:37:46 matt 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/tcp.h>
     41 #include <netinet/in_offload.h>
     42 
     43 struct ip_tso_output_args {
     44 	struct ifnet *ifp;
     45 	const struct sockaddr *sa;
     46 	struct rtentry *rt;
     47 };
     48 
     49 static int ip_tso_output_callback(void *, struct mbuf *);
     50 
     51 static int
     52 ip_tso_output_callback(void *vp, struct mbuf *m)
     53 {
     54 	struct ip_tso_output_args *args = vp;
     55 	struct ifnet *ifp = args->ifp;
     56 
     57 	return (*ifp->if_output)(ifp, m, args->sa, args->rt);
     58 }
     59 
     60 int
     61 ip_tso_output(struct ifnet *ifp, struct mbuf *m, const struct sockaddr *sa,
     62     struct rtentry *rt)
     63 {
     64 	struct ip_tso_output_args args;
     65 
     66 	args.ifp = ifp;
     67 	args.sa = sa;
     68 	args.rt = rt;
     69 
     70 	return tcp4_segment(m, ip_tso_output_callback, &args);
     71 }
     72 
     73 /*
     74  * tcp4_segment: handle M_CSUM_TSOv4 by software.
     75  *
     76  * => always consume m.
     77  * => call output_func with output_arg for each segments.
     78  */
     79 
     80 int
     81 tcp4_segment(struct mbuf *m, int (*output_func)(void *, struct mbuf *),
     82     void *output_arg)
     83 {
     84 	int mss;
     85 	int iphlen;
     86 	int thlen;
     87 	int hlen;
     88 	int len;
     89 	struct ip *iph;
     90 	struct tcphdr *th;
     91 	uint16_t ipid;
     92 	uint32_t tcpseq;
     93 	struct mbuf *hdr = NULL;
     94 	struct mbuf *t;
     95 	int error = 0;
     96 
     97 	KASSERT((m->m_flags & M_PKTHDR) != 0);
     98 	KASSERT((m->m_pkthdr.csum_flags & M_CSUM_TSOv4) != 0);
     99 
    100 	m->m_pkthdr.csum_flags = 0;
    101 
    102 	len = m->m_pkthdr.len;
    103 	KASSERT(len >= sizeof(*iph) + sizeof(*th));
    104 
    105 	if (m->m_len < sizeof(*iph)) {
    106 		m = m_pullup(m, sizeof(*iph));
    107 		if (m == NULL) {
    108 			error = ENOMEM;
    109 			goto quit;
    110 		}
    111 	}
    112 	iph = mtod(m, struct ip *);
    113 	iphlen = iph->ip_hl * 4;
    114 	KASSERT(iph->ip_v == IPVERSION);
    115 	KASSERT(iphlen >= sizeof(*iph));
    116 	KASSERT(iph->ip_p == IPPROTO_TCP);
    117 	ipid = ntohs(iph->ip_id);
    118 
    119 	hlen = iphlen + sizeof(*th);
    120 	if (m->m_len < hlen) {
    121 		m = m_pullup(m, hlen);
    122 		if (m == NULL) {
    123 			error = ENOMEM;
    124 			goto quit;
    125 		}
    126 	}
    127 	th = (void *)(mtod(m, char *) + iphlen);
    128 	tcpseq = ntohl(th->th_seq);
    129 	thlen = th->th_off * 4;
    130 	hlen = iphlen + thlen;
    131 
    132 	mss = m->m_pkthdr.segsz;
    133 	KASSERT(mss != 0);
    134 	KASSERT(len > hlen);
    135 
    136 	t = m_split(m, hlen, M_NOWAIT);
    137 	if (t == NULL) {
    138 		error = ENOMEM;
    139 		goto quit;
    140 	}
    141 	hdr = m;
    142 	m = t;
    143 	len -= hlen;
    144 	KASSERT(len % mss == 0);
    145 	while (len > 0) {
    146 		struct mbuf *n;
    147 
    148 		n = m_dup(hdr, 0, hlen, M_NOWAIT);
    149 		if (n == NULL) {
    150 			error = ENOMEM;
    151 			goto quit;
    152 		}
    153 		KASSERT(n->m_len == hlen); /* XXX */
    154 
    155 		t = m_split(m, mss, M_NOWAIT);
    156 		if (t == NULL) {
    157 			m_freem(n);
    158 			error = ENOMEM;
    159 			goto quit;
    160 		}
    161 		m_cat(n, m);
    162 		m = t;
    163 
    164 		KASSERT(n->m_len >= hlen); /* XXX */
    165 
    166 		n->m_pkthdr.len = hlen + mss;
    167 		iph = mtod(n, struct ip *);
    168 		KASSERT(iph->ip_v == IPVERSION);
    169 		iph->ip_len = htons(n->m_pkthdr.len);
    170 		iph->ip_id = htons(ipid);
    171 		th = (void *)(mtod(n, char *) + iphlen);
    172 		th->th_seq = htonl(tcpseq);
    173 		iph->ip_sum = 0;
    174 		iph->ip_sum = in_cksum(n, iphlen);
    175 		th->th_sum = 0;
    176 		th->th_sum = in4_cksum(n, IPPROTO_TCP, iphlen, thlen + mss);
    177 
    178 		error = (*output_func)(output_arg, n);
    179 		if (error) {
    180 			goto quit;
    181 		}
    182 
    183 		tcpseq += mss;
    184 		ipid++;
    185 		len -= mss;
    186 	}
    187 
    188 quit:
    189 	if (hdr != NULL) {
    190 		m_freem(hdr);
    191 	}
    192 	if (m != NULL) {
    193 		m_freem(m);
    194 	}
    195 
    196 	return error;
    197 }
    198 
    199 void
    200 ip_undefer_csum(struct mbuf *m, size_t hdrlen, int csum_flags)
    201 {
    202 	KASSERT(m->m_flags & M_PKTHDR);
    203 	KASSERT((m->m_pkthdr.csum_flags & csum_flags) == csum_flags);
    204 	uint16_t csum;
    205 	uint16_t ip_len;
    206 	uint16_t *csump;
    207 	size_t iphdrlen;
    208 
    209 	if (__predict_true(hdrlen + sizeof(struct ip) <= m->m_len)) {
    210 		struct ip *ip = (struct ip *)(mtod(m, uint8_t *) + hdrlen);
    211 		ip_len = ip->ip_len;
    212 		iphdrlen = ip->ip_hl << 2;
    213 		csump = &ip->ip_sum;
    214 	} else {
    215 		uint8_t ip_vhl;
    216 		const size_t ip_len_offset = hdrlen + offsetof(struct ip, ip_len);
    217 		m_copydata(m, hdrlen, sizeof(ip_vhl), &ip_vhl);
    218 		m_copydata(m, ip_len_offset, sizeof(ip_len), &ip_len);
    219 		iphdrlen = (ip_vhl & 0x0f) << 2;
    220 		csump = NULL;
    221 	}
    222 
    223 	if (csum_flags & M_CSUM_IPv4) {
    224 		const size_t offset = hdrlen + offsetof(struct ip, ip_sum);
    225 		csum = in4_cksum(m, 0, hdrlen, iphdrlen);
    226 		if (csump != NULL) {
    227 			*csump = csum;
    228 		} else {
    229 			m_copyback(m, offset, sizeof(uint16_t), &csum);
    230 		}
    231 	}
    232 
    233 	if (csum_flags & (M_CSUM_UDPv4|M_CSUM_TCPv4)) {
    234 		size_t l4offset = hdrlen
    235 		    + M_CSUM_DATA_IPv4_IPHL(m->m_pkthdr.csum_data);
    236 
    237 		csum = in4_cksum(m, 0, hdrlen + l4offset, ip_len - l4offset);
    238 		if (csum == 0 && (csum_flags & M_CSUM_UDPv4) != 0)
    239 			csum = 0xffff;
    240 
    241 		l4offset += M_CSUM_DATA_IPv4_OFFSET(m->m_pkthdr.csum_data);
    242 
    243 		if (__predict_true(l4offset + sizeof(uint16_t) <= m->m_len)) {
    244 			*(uint16_t *)(mtod(m, char *) + l4offset) = csum;
    245 		} else {
    246 			m_copyback(m, l4offset, sizeof(csum), (void *) &csum);
    247 		}
    248 	}
    249 
    250 	m->m_pkthdr.csum_flags ^= csum_flags;
    251 }
    252