2 * Copyright (c) 2010-2011 Juniper Networks, Inc.
5 * This software was developed by Robert N. M. Watson under contract
6 * to Juniper Networks, Inc.
8 * Redistribution and use in source and binary forms, with or without
9 * modification, are permitted provided that the following conditions
11 * 1. Redistributions of source code must retain the above copyright
12 * notice, this list of conditions and the following disclaimer.
13 * 2. Redistributions in binary form must reproduce the above copyright
14 * notice, this list of conditions and the following disclaimer in the
15 * documentation and/or other materials provided with the distribution.
17 * THIS SOFTWARE IS PROVIDED BY THE AUTHOR AND CONTRIBUTORS ``AS IS'' AND
18 * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
19 * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
20 * ARE DISCLAIMED. IN NO EVENT SHALL THE AUTHOR OR CONTRIBUTORS BE LIABLE
21 * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
22 * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
23 * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
24 * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
25 * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
26 * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
30 #include <sys/cdefs.h>
32 __FBSDID("$FreeBSD$");
34 #include "opt_inet6.h"
35 #include "opt_pcbgroup.h"
38 #error "options RSS depends on options PCBGROUP"
41 #include <sys/param.h>
43 #include <sys/socket.h>
45 #include <sys/kernel.h>
47 #include <sys/sysctl.h>
51 #include <net/if_var.h>
52 #include <net/netisr.h>
53 #include <net/rss_config.h>
55 #include <netinet/in.h>
56 #include <netinet/in_pcb.h>
57 #include <netinet6/in6_rss.h>
58 #include <netinet/in_var.h>
60 /* for software rss hash support */
61 #include <netinet/ip6.h>
62 #include <netinet6/ip6_var.h>
63 #include <netinet/tcp.h>
64 #include <netinet/udp.h>
67 * Hash an IPv6 2-tuple.
70 rss_hash_ip6_2tuple(const struct in6_addr *src, const struct in6_addr *dst)
72 uint8_t data[sizeof(*src) + sizeof(*dst)];
76 bcopy(src, &data[datalen], sizeof(*src));
77 datalen += sizeof(*src);
78 bcopy(dst, &data[datalen], sizeof(*dst));
79 datalen += sizeof(*dst);
80 return (rss_hash(datalen, data));
84 * Hash an IPv6 4-tuple.
87 rss_hash_ip6_4tuple(const struct in6_addr *src, u_short srcport,
88 const struct in6_addr *dst, u_short dstport)
90 uint8_t data[sizeof(*src) + sizeof(*dst) + sizeof(srcport) +
95 bcopy(src, &data[datalen], sizeof(*src));
96 datalen += sizeof(*src);
97 bcopy(dst, &data[datalen], sizeof(*dst));
98 datalen += sizeof(*dst);
99 bcopy(&srcport, &data[datalen], sizeof(srcport));
100 datalen += sizeof(srcport);
101 bcopy(&dstport, &data[datalen], sizeof(dstport));
102 datalen += sizeof(dstport);
103 return (rss_hash(datalen, data));
107 * Calculate an appropriate ipv6 2-tuple or 4-tuple given the given
108 * IPv6 source/destination address, UDP or TCP source/destination ports
109 * and the protocol type.
111 * The protocol code may wish to do a software hash of the given
112 * tuple. This depends upon the currently configured RSS hash types.
114 * This assumes that the packet in question isn't a fragment.
116 * It also assumes the packet source/destination address
117 * are in "incoming" packet order (ie, source is "far" address.)
120 rss_proto_software_hash_v6(const struct in6_addr *s, const struct in6_addr *d,
121 u_short sp, u_short dp, int proto,
122 uint32_t *hashval, uint32_t *hashtype)
127 * Next, choose the hash type depending upon the protocol
130 if ((proto == IPPROTO_TCP) &&
131 (rss_gethashconfig() & RSS_HASHTYPE_RSS_TCP_IPV6)) {
132 hash = rss_hash_ip6_4tuple(s, sp, d, dp);
134 *hashtype = M_HASHTYPE_RSS_TCP_IPV6;
136 } else if ((proto == IPPROTO_UDP) &&
137 (rss_gethashconfig() & RSS_HASHTYPE_RSS_UDP_IPV6)) {
138 hash = rss_hash_ip6_4tuple(s, sp, d, dp);
140 *hashtype = M_HASHTYPE_RSS_UDP_IPV6;
142 } else if (rss_gethashconfig() & RSS_HASHTYPE_RSS_IPV6) {
143 /* RSS doesn't hash on other protocols like SCTP; so 2-tuple */
144 hash = rss_hash_ip6_2tuple(s, d);
146 *hashtype = M_HASHTYPE_RSS_IPV6;
150 /* No configured available hashtypes! */
151 RSS_DEBUG("no available hashtypes!\n");
156 * Do a software calculation of the RSS for the given mbuf.
158 * This is typically used by the input path to recalculate the RSS after
159 * some form of packet processing (eg de-capsulation, IP fragment reassembly.)
161 * dir is the packet direction - RSS_HASH_PKT_INGRESS for incoming and
162 * RSS_HASH_PKT_EGRESS for outgoing.
164 * Returns 0 if a hash was done, -1 if no hash was done, +1 if
165 * the mbuf already had a valid RSS flowid.
167 * This function doesn't modify the mbuf. It's up to the caller to
168 * assign flowid/flowtype as appropriate.
171 rss_mbuf_software_hash_v6(const struct mbuf *m, int dir, uint32_t *hashval,
174 const struct ip6_hdr *ip6;
175 const struct ip6_frag *ip6f;
176 const struct tcphdr *th;
177 const struct udphdr *uh;
184 * XXX For now this only handles hashing on incoming mbufs.
186 if (dir != RSS_HASH_PKT_INGRESS) {
187 RSS_DEBUG("called on EGRESS packet!\n");
191 off = sizeof(struct ip6_hdr);
194 * First, validate that the mbuf we have is long enough
195 * to have an IPv6 header in it.
197 if (m->m_pkthdr.len < off) {
198 RSS_DEBUG("short mbuf pkthdr\n");
201 if (m->m_len < off) {
202 RSS_DEBUG("short mbuf len\n");
206 /* Ok, let's dereference that */
207 ip6 = mtod(m, struct ip6_hdr *);
208 proto = ip6->ip6_nxt;
211 * Find the beginning of the TCP/UDP header.
213 * If this is a fragment then it shouldn't be four-tuple
214 * hashed just yet. Once it's reassembled into a full
215 * frame it should be re-hashed.
217 while (proto != IPPROTO_FRAGMENT) {
218 newoff = ip6_nexthdr(m, off, proto, &nxt);
226 * Ignore the fragment header if this is an "atomic" fragment
227 * (offset and m bit set to 0)
229 if (proto == IPPROTO_FRAGMENT) {
230 if (m->m_len < off + sizeof(struct ip6_frag)) {
231 RSS_DEBUG("short fragment frame?\n");
234 ip6f = (const struct ip6_frag *)((c_caddr_t)ip6 + off);
235 if ((ip6f->ip6f_offlg & ~IP6F_RESERVED_MASK) == 0) {
236 off = ip6_lasthdr(m, off, proto, &nxt);
238 RSS_DEBUG("invalid extension header\n");
246 * If the mbuf flowid/flowtype matches the packet type,
247 * and we don't support the 4-tuple version of the given protocol,
248 * then signal to the owner that it can trust the flowid/flowtype
251 * This is a little picky - eg, if TCPv6 / UDPv6 hashing
252 * is supported but we got a TCP/UDP frame only 2-tuple hashed,
253 * then we shouldn't just "trust" the 2-tuple hash. We need
256 flowtype = M_HASHTYPE_GET(m);
258 if (flowtype != M_HASHTYPE_NONE) {
261 if ((rss_gethashconfig() & RSS_HASHTYPE_RSS_UDP_IPV6) &&
262 (flowtype == M_HASHTYPE_RSS_UDP_IPV6)) {
266 * Only allow 2-tuple for UDP frames if we don't also
267 * support 4-tuple for UDP.
269 if ((rss_gethashconfig() & RSS_HASHTYPE_RSS_IPV6) &&
270 ((rss_gethashconfig() & RSS_HASHTYPE_RSS_UDP_IPV6) == 0) &&
271 flowtype == M_HASHTYPE_RSS_IPV6) {
276 if ((rss_gethashconfig() & RSS_HASHTYPE_RSS_TCP_IPV6) &&
277 (flowtype == M_HASHTYPE_RSS_TCP_IPV6)) {
281 * Only allow 2-tuple for TCP frames if we don't also
282 * support 4-tuple for TCP.
284 if ((rss_gethashconfig() & RSS_HASHTYPE_RSS_IPV6) &&
285 ((rss_gethashconfig() & RSS_HASHTYPE_RSS_TCP_IPV6) == 0) &&
286 flowtype == M_HASHTYPE_RSS_IPV6) {
291 if ((rss_gethashconfig() & RSS_HASHTYPE_RSS_IPV6) &&
292 flowtype == M_HASHTYPE_RSS_IPV6) {
300 * Decode enough information to make a hash decision.
302 if ((rss_gethashconfig() & RSS_HASHTYPE_RSS_TCP_IPV6) &&
303 (proto == IPPROTO_TCP)) {
304 if (m->m_len < off + sizeof(struct tcphdr)) {
305 RSS_DEBUG("short TCP frame?\n");
308 th = (const struct tcphdr *)((c_caddr_t)ip6 + off);
309 return rss_proto_software_hash_v6(&ip6->ip6_src, &ip6->ip6_dst,
315 } else if ((rss_gethashconfig() & RSS_HASHTYPE_RSS_UDP_IPV6) &&
316 (proto == IPPROTO_UDP)) {
317 if (m->m_len < off + sizeof(struct udphdr)) {
318 RSS_DEBUG("short UDP frame?\n");
321 uh = (const struct udphdr *)((c_caddr_t)ip6 + off);
322 return rss_proto_software_hash_v6(&ip6->ip6_src, &ip6->ip6_dst,
328 } else if (rss_gethashconfig() & RSS_HASHTYPE_RSS_IPV6) {
329 /* Default to 2-tuple hash */
330 return rss_proto_software_hash_v6(&ip6->ip6_src, &ip6->ip6_dst,
332 0, /* destination port */
337 RSS_DEBUG("no available hashtypes!\n");
343 * Similar to rss_m2cpuid, but designed to be used by the IPv6 NETISR
344 * on incoming frames.
346 * If an existing RSS hash exists and it matches what the configured
347 * hashing is, then use it.
349 * If there's an existing RSS hash but the desired hash is different,
350 * or if there's no useful RSS hash, then calculate it via
353 * XXX TODO: definitely want statistics here!
356 rss_soft_m2cpuid_v6(struct mbuf *m, uintptr_t source, u_int *cpuid)
358 uint32_t hash_val, hash_type;
363 ret = rss_mbuf_software_hash_v6(m, RSS_HASH_PKT_INGRESS,
364 &hash_val, &hash_type);
366 /* mbuf has a valid hash already; don't need to modify it */
367 *cpuid = rss_hash2cpuid(m->m_pkthdr.flowid, M_HASHTYPE_GET(m));
368 } else if (ret == 0) {
369 /* hash was done; update */
370 m->m_pkthdr.flowid = hash_val;
371 M_HASHTYPE_SET(m, hash_type);
372 *cpuid = rss_hash2cpuid(m->m_pkthdr.flowid, M_HASHTYPE_GET(m));
373 } else { /* ret < 0 */
374 /* no hash was done */
375 *cpuid = NETISR_CPUID_NONE;