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 * Calculate an appropriate ipv6 2-tuple or 4-tuple given the given 157 * IPv6 source/destination address, UDP or TCP source/destination ports 158 * and the protocol type. 159 * 160 * The protocol code may wish to do a software hash of the given 161 * tuple. This depends upon the currently configured RSS hash types. 162 * 163 * It assumes the packet source/destination address 164 * are in "outgoin" packet order (ie, destination is "far" address.) 165 */ 166 uint32_t 167 xps_proto_software_hash_v6(const struct in6_addr *s, const struct in6_addr *d, 168 u_short sp, u_short dp, int proto, uint32_t *hashtype) 169 { 170 171 uint32_t hash; 172 173 /* 174 * Next, choose the hash type depending upon the protocol 175 * identifier. 176 */ 177 if ((proto == IPPROTO_TCP) && 178 (rss_gethashconfig() & RSS_HASHTYPE_RSS_TCP_IPV6)) { 179 hash = rss_hash_ip6_4tuple(d, dp, s, sp); 180 *hashtype = M_HASHTYPE_RSS_TCP_IPV6; 181 return (hash); 182 } else if ((proto == IPPROTO_UDP) && 183 (rss_gethashconfig() & RSS_HASHTYPE_RSS_UDP_IPV6)) { 184 hash = rss_hash_ip6_4tuple(d, dp, s, sp); 185 *hashtype = M_HASHTYPE_RSS_UDP_IPV6; 186 return (hash); 187 } else if (rss_gethashconfig() & RSS_HASHTYPE_RSS_IPV6) { 188 /* RSS doesn't hash on other protocols like SCTP; so 2-tuple */ 189 hash = rss_hash_ip6_2tuple(d, s); 190 *hashtype = M_HASHTYPE_RSS_IPV6; 191 return (hash); 192 } 193 194 *hashtype = M_HASHTYPE_NONE; 195 return (0); 196 } 197 198 199 /* 200 * Do a software calculation of the RSS for the given mbuf. 201 * 202 * This is typically used by the input path to recalculate the RSS after 203 * some form of packet processing (eg de-capsulation, IP fragment reassembly.) 204 * 205 * dir is the packet direction - RSS_HASH_PKT_INGRESS for incoming and 206 * RSS_HASH_PKT_EGRESS for outgoing. 207 * 208 * Returns 0 if a hash was done, -1 if no hash was done, +1 if 209 * the mbuf already had a valid RSS flowid. 210 * 211 * This function doesn't modify the mbuf. It's up to the caller to 212 * assign flowid/flowtype as appropriate. 213 */ 214 int 215 rss_mbuf_software_hash_v6(const struct mbuf *m, int dir, uint32_t *hashval, 216 uint32_t *hashtype) 217 { 218 const struct ip6_hdr *ip6; 219 const struct ip6_frag *ip6f; 220 const struct tcphdr *th; 221 const struct udphdr *uh; 222 uint32_t flowtype; 223 uint8_t proto; 224 int off, newoff; 225 int nxt; 226 227 /* 228 * XXX For now this only handles hashing on incoming mbufs. 229 */ 230 if (dir != RSS_HASH_PKT_INGRESS) { 231 RSS_DEBUG("called on EGRESS packet!\n"); 232 return (-1); 233 } 234 235 off = sizeof(struct ip6_hdr); 236 237 /* 238 * First, validate that the mbuf we have is long enough 239 * to have an IPv6 header in it. 240 */ 241 if (m->m_pkthdr.len < off) { 242 RSS_DEBUG("short mbuf pkthdr\n"); 243 return (-1); 244 } 245 if (m->m_len < off) { 246 RSS_DEBUG("short mbuf len\n"); 247 return (-1); 248 } 249 250 /* Ok, let's dereference that */ 251 ip6 = mtod(m, struct ip6_hdr *); 252 proto = ip6->ip6_nxt; 253 254 /* 255 * Find the beginning of the TCP/UDP header. 256 * 257 * If this is a fragment then it shouldn't be four-tuple 258 * hashed just yet. Once it's reassembled into a full 259 * frame it should be re-hashed. 260 */ 261 while (proto != IPPROTO_FRAGMENT) { 262 newoff = ip6_nexthdr(m, off, proto, &nxt); 263 if (newoff < 0) 264 break; 265 off = newoff; 266 proto = nxt; 267 } 268 269 /* 270 * Ignore the fragment header if this is an "atomic" fragment 271 * (offset and m bit set to 0) 272 */ 273 if (proto == IPPROTO_FRAGMENT) { 274 if (m->m_len < off + sizeof(struct ip6_frag)) { 275 RSS_DEBUG("short fragment frame?\n"); 276 return (-1); 277 } 278 ip6f = (const struct ip6_frag *)((c_caddr_t)ip6 + off); 279 if ((ip6f->ip6f_offlg & ~IP6F_RESERVED_MASK) == 0) { 280 off = ip6_lasthdr(m, off, proto, &nxt); 281 if (off < 0) { 282 RSS_DEBUG("invalid extension header\n"); 283 return (-1); 284 } 285 proto = nxt; 286 } 287 } 288 289 /* 290 * If the mbuf flowid/flowtype matches the packet type, 291 * and we don't support the 4-tuple version of the given protocol, 292 * then signal to the owner that it can trust the flowid/flowtype 293 * details. 294 * 295 * This is a little picky - eg, if TCPv6 / UDPv6 hashing 296 * is supported but we got a TCP/UDP frame only 2-tuple hashed, 297 * then we shouldn't just "trust" the 2-tuple hash. We need 298 * a 4-tuple hash. 299 */ 300 flowtype = M_HASHTYPE_GET(m); 301 302 if (flowtype != M_HASHTYPE_NONE) { 303 switch (proto) { 304 case IPPROTO_UDP: 305 if ((rss_gethashconfig() & RSS_HASHTYPE_RSS_UDP_IPV6) && 306 (flowtype == M_HASHTYPE_RSS_UDP_IPV6)) { 307 return (1); 308 } 309 /* 310 * Only allow 2-tuple for UDP frames if we don't also 311 * support 4-tuple for UDP. 312 */ 313 if ((rss_gethashconfig() & RSS_HASHTYPE_RSS_IPV6) && 314 ((rss_gethashconfig() & RSS_HASHTYPE_RSS_UDP_IPV6) == 0) && 315 flowtype == M_HASHTYPE_RSS_IPV6) { 316 return (1); 317 } 318 break; 319 case IPPROTO_TCP: 320 if ((rss_gethashconfig() & RSS_HASHTYPE_RSS_TCP_IPV6) && 321 (flowtype == M_HASHTYPE_RSS_TCP_IPV6)) { 322 return (1); 323 } 324 /* 325 * Only allow 2-tuple for TCP frames if we don't also 326 * support 4-tuple for TCP. 327 */ 328 if ((rss_gethashconfig() & RSS_HASHTYPE_RSS_IPV6) && 329 ((rss_gethashconfig() & RSS_HASHTYPE_RSS_TCP_IPV6) == 0) && 330 flowtype == M_HASHTYPE_RSS_IPV6) { 331 return (1); 332 } 333 break; 334 default: 335 if ((rss_gethashconfig() & RSS_HASHTYPE_RSS_IPV6) && 336 flowtype == M_HASHTYPE_RSS_IPV6) { 337 return (1); 338 } 339 break; 340 } 341 } 342 343 /* 344 * Decode enough information to make a hash decision. 345 */ 346 if ((rss_gethashconfig() & RSS_HASHTYPE_RSS_TCP_IPV6) && 347 (proto == IPPROTO_TCP)) { 348 if (m->m_len < off + sizeof(struct tcphdr)) { 349 RSS_DEBUG("short TCP frame?\n"); 350 return (-1); 351 } 352 th = (const struct tcphdr *)((c_caddr_t)ip6 + off); 353 return rss_proto_software_hash_v6(&ip6->ip6_src, &ip6->ip6_dst, 354 th->th_sport, 355 th->th_dport, 356 proto, 357 hashval, 358 hashtype); 359 } else if ((rss_gethashconfig() & RSS_HASHTYPE_RSS_UDP_IPV6) && 360 (proto == IPPROTO_UDP)) { 361 if (m->m_len < off + sizeof(struct udphdr)) { 362 RSS_DEBUG("short UDP frame?\n"); 363 return (-1); 364 } 365 uh = (const struct udphdr *)((c_caddr_t)ip6 + off); 366 return rss_proto_software_hash_v6(&ip6->ip6_src, &ip6->ip6_dst, 367 uh->uh_sport, 368 uh->uh_dport, 369 proto, 370 hashval, 371 hashtype); 372 } else if (rss_gethashconfig() & RSS_HASHTYPE_RSS_IPV6) { 373 /* Default to 2-tuple hash */ 374 return rss_proto_software_hash_v6(&ip6->ip6_src, &ip6->ip6_dst, 375 0, /* source port */ 376 0, /* destination port */ 377 0, /* IPPROTO_IP */ 378 hashval, 379 hashtype); 380 } else { 381 RSS_DEBUG("no available hashtypes!\n"); 382 return (-1); 383 } 384 } 385 386 /* 387 * Similar to rss_m2cpuid, but designed to be used by the IPv6 NETISR 388 * on incoming frames. 389 * 390 * If an existing RSS hash exists and it matches what the configured 391 * hashing is, then use it. 392 * 393 * If there's an existing RSS hash but the desired hash is different, 394 * or if there's no useful RSS hash, then calculate it via 395 * the software path. 396 * 397 * XXX TODO: definitely want statistics here! 398 */ 399 struct mbuf * 400 rss_soft_m2cpuid_v6(struct mbuf *m, uintptr_t source, u_int *cpuid) 401 { 402 uint32_t hash_val, hash_type; 403 int ret; 404 405 M_ASSERTPKTHDR(m); 406 407 ret = rss_mbuf_software_hash_v6(m, RSS_HASH_PKT_INGRESS, 408 &hash_val, &hash_type); 409 if (ret > 0) { 410 /* mbuf has a valid hash already; don't need to modify it */ 411 *cpuid = rss_hash2cpuid(m->m_pkthdr.flowid, M_HASHTYPE_GET(m)); 412 } else if (ret == 0) { 413 /* hash was done; update */ 414 m->m_pkthdr.flowid = hash_val; 415 M_HASHTYPE_SET(m, hash_type); 416 *cpuid = rss_hash2cpuid(m->m_pkthdr.flowid, M_HASHTYPE_GET(m)); 417 } else { /* ret < 0 */ 418 /* no hash was done */ 419 *cpuid = NETISR_CPUID_NONE; 420 } 421 return (m); 422 } 423