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
31 #include "opt_inet6.h"
32 #include "opt_pcbgroup.h"
35 #error "options RSS depends on options PCBGROUP"
38 #include <sys/param.h>
40 #include <sys/socket.h>
42 #include <sys/kernel.h>
44 #include <sys/sysctl.h>
48 #include <net/if_var.h>
49 #include <net/netisr.h>
50 #include <net/rss_config.h>
52 #include <netinet/in.h>
53 #include <netinet/in_pcb.h>
54 #include <netinet6/in6_rss.h>
55 #include <netinet/in_var.h>
57 /* for software rss hash support */
58 #include <netinet/ip6.h>
59 #include <netinet6/ip6_var.h>
60 #include <netinet/tcp.h>
61 #include <netinet/udp.h>
64 * Hash an IPv6 2-tuple.
67 rss_hash_ip6_2tuple(const struct in6_addr *src, const struct in6_addr *dst)
69 uint8_t data[sizeof(*src) + sizeof(*dst)];
73 bcopy(src, &data[datalen], sizeof(*src));
74 datalen += sizeof(*src);
75 bcopy(dst, &data[datalen], sizeof(*dst));
76 datalen += sizeof(*dst);
77 return (rss_hash(datalen, data));
81 * Hash an IPv6 4-tuple.
84 rss_hash_ip6_4tuple(const struct in6_addr *src, u_short srcport,
85 const struct in6_addr *dst, u_short dstport)
87 uint8_t data[sizeof(*src) + sizeof(*dst) + sizeof(srcport) +
92 bcopy(src, &data[datalen], sizeof(*src));
93 datalen += sizeof(*src);
94 bcopy(dst, &data[datalen], sizeof(*dst));
95 datalen += sizeof(*dst);
96 bcopy(&srcport, &data[datalen], sizeof(srcport));
97 datalen += sizeof(srcport);
98 bcopy(&dstport, &data[datalen], sizeof(dstport));
99 datalen += sizeof(dstport);
100 return (rss_hash(datalen, data));
104 * Calculate an appropriate ipv6 2-tuple or 4-tuple given the given
105 * IPv6 source/destination address, UDP or TCP source/destination ports
106 * and the protocol type.
108 * The protocol code may wish to do a software hash of the given
109 * tuple. This depends upon the currently configured RSS hash types.
111 * This assumes that the packet in question isn't a fragment.
113 * It also assumes the packet source/destination address
114 * are in "incoming" packet order (ie, source is "far" address.)
117 rss_proto_software_hash_v6(const struct in6_addr *s, const struct in6_addr *d,
118 u_short sp, u_short dp, int proto,
119 uint32_t *hashval, uint32_t *hashtype)
124 * Next, choose the hash type depending upon the protocol
127 if ((proto == IPPROTO_TCP) &&
128 (rss_gethashconfig() & RSS_HASHTYPE_RSS_TCP_IPV6)) {
129 hash = rss_hash_ip6_4tuple(s, sp, d, dp);
131 *hashtype = M_HASHTYPE_RSS_TCP_IPV6;
133 } else if ((proto == IPPROTO_UDP) &&
134 (rss_gethashconfig() & RSS_HASHTYPE_RSS_UDP_IPV6)) {
135 hash = rss_hash_ip6_4tuple(s, sp, d, dp);
137 *hashtype = M_HASHTYPE_RSS_UDP_IPV6;
139 } else if (rss_gethashconfig() & RSS_HASHTYPE_RSS_IPV6) {
140 /* RSS doesn't hash on other protocols like SCTP; so 2-tuple */
141 hash = rss_hash_ip6_2tuple(s, d);
143 *hashtype = M_HASHTYPE_RSS_IPV6;
147 /* No configured available hashtypes! */
148 RSS_DEBUG("no available hashtypes!\n");
153 * Calculate an appropriate ipv6 2-tuple or 4-tuple given the given
154 * IPv6 source/destination address, UDP or TCP source/destination ports
155 * and the protocol type.
157 * The protocol code may wish to do a software hash of the given
158 * tuple. This depends upon the currently configured RSS hash types.
160 * It assumes the packet source/destination address
161 * are in "outgoin" packet order (ie, destination is "far" address.)
164 xps_proto_software_hash_v6(const struct in6_addr *s, const struct in6_addr *d,
165 u_short sp, u_short dp, int proto, uint32_t *hashtype)
171 * Next, choose the hash type depending upon the protocol
174 if ((proto == IPPROTO_TCP) &&
175 (rss_gethashconfig() & RSS_HASHTYPE_RSS_TCP_IPV6)) {
176 hash = rss_hash_ip6_4tuple(d, dp, s, sp);
177 *hashtype = M_HASHTYPE_RSS_TCP_IPV6;
179 } else if ((proto == IPPROTO_UDP) &&
180 (rss_gethashconfig() & RSS_HASHTYPE_RSS_UDP_IPV6)) {
181 hash = rss_hash_ip6_4tuple(d, dp, s, sp);
182 *hashtype = M_HASHTYPE_RSS_UDP_IPV6;
184 } else if (rss_gethashconfig() & RSS_HASHTYPE_RSS_IPV6) {
185 /* RSS doesn't hash on other protocols like SCTP; so 2-tuple */
186 hash = rss_hash_ip6_2tuple(d, s);
187 *hashtype = M_HASHTYPE_RSS_IPV6;
191 *hashtype = M_HASHTYPE_NONE;
197 * Do a software calculation of the RSS for the given mbuf.
199 * This is typically used by the input path to recalculate the RSS after
200 * some form of packet processing (eg de-capsulation, IP fragment reassembly.)
202 * dir is the packet direction - RSS_HASH_PKT_INGRESS for incoming and
203 * RSS_HASH_PKT_EGRESS for outgoing.
205 * Returns 0 if a hash was done, -1 if no hash was done, +1 if
206 * the mbuf already had a valid RSS flowid.
208 * This function doesn't modify the mbuf. It's up to the caller to
209 * assign flowid/flowtype as appropriate.
212 rss_mbuf_software_hash_v6(const struct mbuf *m, int dir, uint32_t *hashval,
215 const struct ip6_hdr *ip6;
216 const struct ip6_frag *ip6f;
217 const struct tcphdr *th;
218 const struct udphdr *uh;
225 * XXX For now this only handles hashing on incoming mbufs.
227 if (dir != RSS_HASH_PKT_INGRESS) {
228 RSS_DEBUG("called on EGRESS packet!\n");
232 off = sizeof(struct ip6_hdr);
235 * First, validate that the mbuf we have is long enough
236 * to have an IPv6 header in it.
238 if (m->m_pkthdr.len < off) {
239 RSS_DEBUG("short mbuf pkthdr\n");
242 if (m->m_len < off) {
243 RSS_DEBUG("short mbuf len\n");
247 /* Ok, let's dereference that */
248 ip6 = mtod(m, struct ip6_hdr *);
249 proto = ip6->ip6_nxt;
252 * Find the beginning of the TCP/UDP header.
254 * If this is a fragment then it shouldn't be four-tuple
255 * hashed just yet. Once it's reassembled into a full
256 * frame it should be re-hashed.
258 while (proto != IPPROTO_FRAGMENT) {
259 newoff = ip6_nexthdr(m, off, proto, &nxt);
267 * Ignore the fragment header if this is an "atomic" fragment
268 * (offset and m bit set to 0)
270 if (proto == IPPROTO_FRAGMENT) {
271 if (m->m_len < off + sizeof(struct ip6_frag)) {
272 RSS_DEBUG("short fragment frame?\n");
275 ip6f = (const struct ip6_frag *)((c_caddr_t)ip6 + off);
276 if ((ip6f->ip6f_offlg & ~IP6F_RESERVED_MASK) == 0) {
277 off = ip6_lasthdr(m, off, proto, &nxt);
279 RSS_DEBUG("invalid extension header\n");
287 * If the mbuf flowid/flowtype matches the packet type,
288 * and we don't support the 4-tuple version of the given protocol,
289 * then signal to the owner that it can trust the flowid/flowtype
292 * This is a little picky - eg, if TCPv6 / UDPv6 hashing
293 * is supported but we got a TCP/UDP frame only 2-tuple hashed,
294 * then we shouldn't just "trust" the 2-tuple hash. We need
297 flowtype = M_HASHTYPE_GET(m);
299 if (flowtype != M_HASHTYPE_NONE) {
302 if ((rss_gethashconfig() & RSS_HASHTYPE_RSS_UDP_IPV6) &&
303 (flowtype == M_HASHTYPE_RSS_UDP_IPV6)) {
307 * Only allow 2-tuple for UDP frames if we don't also
308 * support 4-tuple for UDP.
310 if ((rss_gethashconfig() & RSS_HASHTYPE_RSS_IPV6) &&
311 ((rss_gethashconfig() & RSS_HASHTYPE_RSS_UDP_IPV6) == 0) &&
312 flowtype == M_HASHTYPE_RSS_IPV6) {
317 if ((rss_gethashconfig() & RSS_HASHTYPE_RSS_TCP_IPV6) &&
318 (flowtype == M_HASHTYPE_RSS_TCP_IPV6)) {
322 * Only allow 2-tuple for TCP frames if we don't also
323 * support 4-tuple for TCP.
325 if ((rss_gethashconfig() & RSS_HASHTYPE_RSS_IPV6) &&
326 ((rss_gethashconfig() & RSS_HASHTYPE_RSS_TCP_IPV6) == 0) &&
327 flowtype == M_HASHTYPE_RSS_IPV6) {
332 if ((rss_gethashconfig() & RSS_HASHTYPE_RSS_IPV6) &&
333 flowtype == M_HASHTYPE_RSS_IPV6) {
341 * Decode enough information to make a hash decision.
343 if ((rss_gethashconfig() & RSS_HASHTYPE_RSS_TCP_IPV6) &&
344 (proto == IPPROTO_TCP)) {
345 if (m->m_len < off + sizeof(struct tcphdr)) {
346 RSS_DEBUG("short TCP frame?\n");
349 th = (const struct tcphdr *)((c_caddr_t)ip6 + off);
350 return rss_proto_software_hash_v6(&ip6->ip6_src, &ip6->ip6_dst,
356 } else if ((rss_gethashconfig() & RSS_HASHTYPE_RSS_UDP_IPV6) &&
357 (proto == IPPROTO_UDP)) {
358 if (m->m_len < off + sizeof(struct udphdr)) {
359 RSS_DEBUG("short UDP frame?\n");
362 uh = (const struct udphdr *)((c_caddr_t)ip6 + off);
363 return rss_proto_software_hash_v6(&ip6->ip6_src, &ip6->ip6_dst,
369 } else if (rss_gethashconfig() & RSS_HASHTYPE_RSS_IPV6) {
370 /* Default to 2-tuple hash */
371 return rss_proto_software_hash_v6(&ip6->ip6_src, &ip6->ip6_dst,
373 0, /* destination port */
378 RSS_DEBUG("no available hashtypes!\n");
384 * Similar to rss_m2cpuid, but designed to be used by the IPv6 NETISR
385 * on incoming frames.
387 * If an existing RSS hash exists and it matches what the configured
388 * hashing is, then use it.
390 * If there's an existing RSS hash but the desired hash is different,
391 * or if there's no useful RSS hash, then calculate it via
394 * XXX TODO: definitely want statistics here!
397 rss_soft_m2cpuid_v6(struct mbuf *m, uintptr_t source, u_int *cpuid)
399 uint32_t hash_val, hash_type;
404 ret = rss_mbuf_software_hash_v6(m, RSS_HASH_PKT_INGRESS,
405 &hash_val, &hash_type);
407 /* mbuf has a valid hash already; don't need to modify it */
408 *cpuid = rss_hash2cpuid(m->m_pkthdr.flowid, M_HASHTYPE_GET(m));
409 } else if (ret == 0) {
410 /* hash was done; update */
411 m->m_pkthdr.flowid = hash_val;
412 M_HASHTYPE_SET(m, hash_type);
413 *cpuid = rss_hash2cpuid(m->m_pkthdr.flowid, M_HASHTYPE_GET(m));
414 } else { /* ret < 0 */
415 /* no hash was done */
416 *cpuid = NETISR_CPUID_NONE;