1 /*- 2 * Copyright (c) 2010-2011 Juniper Networks, Inc. 3 * All rights reserved. 4 * 5 * This software was developed by Robert N. M. Watson under contract 6 * to Juniper Networks, Inc. 7 * 8 * Redistribution and use in source and binary forms, with or without 9 * modification, are permitted provided that the following conditions 10 * are met: 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. 16 * 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 27 * SUCH DAMAGE. 28 */ 29 30 #include <sys/cdefs.h> 31 32 __FBSDID("$FreeBSD$"); 33 34 #include "opt_inet6.h" 35 #include "opt_pcbgroup.h" 36 37 #ifndef PCBGROUP 38 #error "options RSS depends on options PCBGROUP" 39 #endif 40 41 #include <sys/param.h> 42 #include <sys/mbuf.h> 43 #include <sys/socket.h> 44 #include <sys/priv.h> 45 #include <sys/kernel.h> 46 #include <sys/smp.h> 47 #include <sys/sysctl.h> 48 #include <sys/sbuf.h> 49 50 #include <net/if.h> 51 #include <net/if_var.h> 52 #include <net/netisr.h> 53 #include <net/rss_config.h> 54 55 #include <netinet/in.h> 56 #include <netinet/in_pcb.h> 57 #include <netinet6/in6_rss.h> 58 #include <netinet/in_var.h> 59 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> 65 66 /* 67 * Hash an IPv6 2-tuple. 68 */ 69 uint32_t 70 rss_hash_ip6_2tuple(const struct in6_addr *src, const struct in6_addr *dst) 71 { 72 uint8_t data[sizeof(*src) + sizeof(*dst)]; 73 u_int datalen; 74 75 datalen = 0; 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)); 81 } 82 83 /* 84 * Hash an IPv6 4-tuple. 85 */ 86 uint32_t 87 rss_hash_ip6_4tuple(const struct in6_addr *src, u_short srcport, 88 const struct in6_addr *dst, u_short dstport) 89 { 90 uint8_t data[sizeof(*src) + sizeof(*dst) + sizeof(srcport) + 91 sizeof(dstport)]; 92 u_int datalen; 93 94 datalen = 0; 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)); 104 } 105 106 /* 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. 110 * 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. 113 * 114 * This assumes that the packet in question isn't a fragment. 115 * 116 * It also assumes the packet source/destination address 117 * are in "incoming" packet order (ie, source is "far" address.) 118 */ 119 int 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) 123 { 124 uint32_t hash; 125 126 /* 127 * Next, choose the hash type depending upon the protocol 128 * identifier. 129 */ 130 if ((proto == IPPROTO_TCP) && 131 (rss_gethashconfig() & RSS_HASHTYPE_RSS_TCP_IPV6)) { 132 hash = rss_hash_ip6_4tuple(s, sp, d, dp); 133 *hashval = hash; 134 *hashtype = M_HASHTYPE_RSS_TCP_IPV6; 135 return (0); 136 } else if ((proto == IPPROTO_UDP) && 137 (rss_gethashconfig() & RSS_HASHTYPE_RSS_UDP_IPV6)) { 138 hash = rss_hash_ip6_4tuple(s, sp, d, dp); 139 *hashval = hash; 140 *hashtype = M_HASHTYPE_RSS_UDP_IPV6; 141 return (0); 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); 145 *hashval = hash; 146 *hashtype = M_HASHTYPE_RSS_IPV6; 147 return (0); 148 } 149 150 /* No configured available hashtypes! */ 151 RSS_DEBUG("no available hashtypes!\n"); 152 return (-1); 153 } 154 155 /* 156 * Do a software calculation of the RSS for the given mbuf. 157 * 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.) 160 * 161 * dir is the packet direction - RSS_HASH_PKT_INGRESS for incoming and 162 * RSS_HASH_PKT_EGRESS for outgoing. 163 * 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. 166 * 167 * This function doesn't modify the mbuf. It's up to the caller to 168 * assign flowid/flowtype as appropriate. 169 */ 170 int 171 rss_mbuf_software_hash_v6(const struct mbuf *m, int dir, uint32_t *hashval, 172 uint32_t *hashtype) 173 { 174 const struct ip6_hdr *ip6; 175 const struct tcphdr *th; 176 const struct udphdr *uh; 177 uint32_t flowtype; 178 uint8_t proto; 179 int off, newoff; 180 int nxt; 181 182 /* 183 * XXX For now this only handles hashing on incoming mbufs. 184 */ 185 if (dir != RSS_HASH_PKT_INGRESS) { 186 RSS_DEBUG("called on EGRESS packet!\n"); 187 return (-1); 188 } 189 190 off = sizeof(struct ip6_hdr); 191 192 /* 193 * First, validate that the mbuf we have is long enough 194 * to have an IPv6 header in it. 195 */ 196 if (m->m_pkthdr.len < off) { 197 RSS_DEBUG("short mbuf pkthdr\n"); 198 return (-1); 199 } 200 if (m->m_len < off) { 201 RSS_DEBUG("short mbuf len\n"); 202 return (-1); 203 } 204 205 /* Ok, let's dereference that */ 206 ip6 = mtod(m, struct ip6_hdr *); 207 proto = ip6->ip6_nxt; 208 209 /* 210 * Find the beginning of the TCP/UDP header. 211 * 212 * If this is a fragment then it shouldn't be four-tuple 213 * hashed just yet. Once it's reassembled into a full 214 * frame it should be re-hashed. 215 */ 216 while (proto != IPPROTO_FRAGMENT) { 217 newoff = ip6_nexthdr(m, off, proto, &nxt); 218 if (newoff < 0) 219 break; 220 off = newoff; 221 proto = nxt; 222 } 223 224 /* 225 * If the mbuf flowid/flowtype matches the packet type, 226 * and we don't support the 4-tuple version of the given protocol, 227 * then signal to the owner that it can trust the flowid/flowtype 228 * details. 229 * 230 * This is a little picky - eg, if TCPv6 / UDPv6 hashing 231 * is supported but we got a TCP/UDP frame only 2-tuple hashed, 232 * then we shouldn't just "trust" the 2-tuple hash. We need 233 * a 4-tuple hash. 234 */ 235 flowtype = M_HASHTYPE_GET(m); 236 237 if (flowtype != M_HASHTYPE_NONE) { 238 switch (proto) { 239 case IPPROTO_UDP: 240 if ((rss_gethashconfig() & RSS_HASHTYPE_RSS_UDP_IPV6) && 241 (flowtype == M_HASHTYPE_RSS_UDP_IPV6)) { 242 return (1); 243 } 244 /* 245 * Only allow 2-tuple for UDP frames if we don't also 246 * support 4-tuple for UDP. 247 */ 248 if ((rss_gethashconfig() & RSS_HASHTYPE_RSS_IPV6) && 249 ((rss_gethashconfig() & RSS_HASHTYPE_RSS_UDP_IPV6) == 0) && 250 flowtype == M_HASHTYPE_RSS_IPV6) { 251 return (1); 252 } 253 break; 254 case IPPROTO_TCP: 255 if ((rss_gethashconfig() & RSS_HASHTYPE_RSS_TCP_IPV6) && 256 (flowtype == M_HASHTYPE_RSS_TCP_IPV6)) { 257 return (1); 258 } 259 /* 260 * Only allow 2-tuple for TCP frames if we don't also 261 * support 4-tuple for TCP. 262 */ 263 if ((rss_gethashconfig() & RSS_HASHTYPE_RSS_IPV6) && 264 ((rss_gethashconfig() & RSS_HASHTYPE_RSS_TCP_IPV6) == 0) && 265 flowtype == M_HASHTYPE_RSS_IPV6) { 266 return (1); 267 } 268 break; 269 default: 270 if ((rss_gethashconfig() & RSS_HASHTYPE_RSS_IPV6) && 271 flowtype == M_HASHTYPE_RSS_IPV6) { 272 return (1); 273 } 274 break; 275 } 276 } 277 278 /* 279 * Decode enough information to make a hash decision. 280 */ 281 if ((rss_gethashconfig() & RSS_HASHTYPE_RSS_TCP_IPV6) && 282 (proto == IPPROTO_TCP)) { 283 if (m->m_len < off + sizeof(struct tcphdr)) { 284 RSS_DEBUG("short TCP frame?\n"); 285 return (-1); 286 } 287 th = (const struct tcphdr *)((c_caddr_t)ip6 + off); 288 return rss_proto_software_hash_v6(&ip6->ip6_src, &ip6->ip6_dst, 289 th->th_sport, 290 th->th_dport, 291 proto, 292 hashval, 293 hashtype); 294 } else if ((rss_gethashconfig() & RSS_HASHTYPE_RSS_UDP_IPV6) && 295 (proto == IPPROTO_UDP)) { 296 if (m->m_len < off + sizeof(struct udphdr)) { 297 RSS_DEBUG("short UDP frame?\n"); 298 return (-1); 299 } 300 uh = (const struct udphdr *)((c_caddr_t)ip6 + off); 301 return rss_proto_software_hash_v6(&ip6->ip6_src, &ip6->ip6_dst, 302 uh->uh_sport, 303 uh->uh_dport, 304 proto, 305 hashval, 306 hashtype); 307 } else if (rss_gethashconfig() & RSS_HASHTYPE_RSS_IPV6) { 308 /* Default to 2-tuple hash */ 309 return rss_proto_software_hash_v6(&ip6->ip6_src, &ip6->ip6_dst, 310 0, /* source port */ 311 0, /* destination port */ 312 0, /* IPPROTO_IP */ 313 hashval, 314 hashtype); 315 } else { 316 RSS_DEBUG("no available hashtypes!\n"); 317 return (-1); 318 } 319 } 320 321 /* 322 * Similar to rss_m2cpuid, but designed to be used by the IPv6 NETISR 323 * on incoming frames. 324 * 325 * If an existing RSS hash exists and it matches what the configured 326 * hashing is, then use it. 327 * 328 * If there's an existing RSS hash but the desired hash is different, 329 * or if there's no useful RSS hash, then calculate it via 330 * the software path. 331 * 332 * XXX TODO: definitely want statistics here! 333 */ 334 struct mbuf * 335 rss_soft_m2cpuid_v6(struct mbuf *m, uintptr_t source, u_int *cpuid) 336 { 337 uint32_t hash_val, hash_type; 338 int ret; 339 340 M_ASSERTPKTHDR(m); 341 342 ret = rss_mbuf_software_hash_v6(m, RSS_HASH_PKT_INGRESS, 343 &hash_val, &hash_type); 344 if (ret > 0) { 345 /* mbuf has a valid hash already; don't need to modify it */ 346 *cpuid = rss_hash2cpuid(m->m_pkthdr.flowid, M_HASHTYPE_GET(m)); 347 } else if (ret == 0) { 348 /* hash was done; update */ 349 m->m_pkthdr.flowid = hash_val; 350 M_HASHTYPE_SET(m, hash_type); 351 *cpuid = rss_hash2cpuid(m->m_pkthdr.flowid, M_HASHTYPE_GET(m)); 352 } else { /* ret < 0 */ 353 /* no hash was done */ 354 *cpuid = NETISR_CPUID_NONE; 355 } 356 return (m); 357 } 358