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 31 #include "opt_inet6.h" 32 #include "opt_rss.h" 33 34 #include <sys/param.h> 35 #include <sys/mbuf.h> 36 #include <sys/socket.h> 37 #include <sys/priv.h> 38 #include <sys/kernel.h> 39 #include <sys/smp.h> 40 #include <sys/sysctl.h> 41 #include <sys/sbuf.h> 42 43 #include <net/if.h> 44 #include <net/if_var.h> 45 #include <net/netisr.h> 46 #include <net/rss_config.h> 47 48 #include <netinet/in.h> 49 #include <netinet/in_pcb.h> 50 #include <netinet/in_rss.h> 51 #include <netinet/in_var.h> 52 53 /* for software rss hash support */ 54 #include <netinet/ip.h> 55 #include <netinet/tcp.h> 56 #include <netinet/udp.h> 57 58 /* 59 * Hash an IPv4 2-tuple. 60 */ 61 uint32_t 62 rss_hash_ip4_2tuple(struct in_addr src, struct in_addr dst) 63 { 64 uint8_t data[sizeof(src) + sizeof(dst)]; 65 u_int datalen; 66 67 datalen = 0; 68 bcopy(&src, &data[datalen], sizeof(src)); 69 datalen += sizeof(src); 70 bcopy(&dst, &data[datalen], sizeof(dst)); 71 datalen += sizeof(dst); 72 return (rss_hash(datalen, data)); 73 } 74 75 /* 76 * Hash an IPv4 4-tuple. 77 */ 78 uint32_t 79 rss_hash_ip4_4tuple(struct in_addr src, u_short srcport, struct in_addr dst, 80 u_short dstport) 81 { 82 uint8_t data[sizeof(src) + sizeof(dst) + sizeof(srcport) + 83 sizeof(dstport)]; 84 u_int datalen; 85 86 datalen = 0; 87 bcopy(&src, &data[datalen], sizeof(src)); 88 datalen += sizeof(src); 89 bcopy(&dst, &data[datalen], sizeof(dst)); 90 datalen += sizeof(dst); 91 bcopy(&srcport, &data[datalen], sizeof(srcport)); 92 datalen += sizeof(srcport); 93 bcopy(&dstport, &data[datalen], sizeof(dstport)); 94 datalen += sizeof(dstport); 95 return (rss_hash(datalen, data)); 96 } 97 98 /* 99 * Calculate an appropriate ipv4 2-tuple or 4-tuple given the given 100 * IPv4 source/destination address, UDP or TCP source/destination ports 101 * and the protocol type. 102 * 103 * The protocol code may wish to do a software hash of the given 104 * tuple. This depends upon the currently configured RSS hash types. 105 * 106 * This assumes that the packet in question isn't a fragment. 107 * 108 * It also assumes the packet source/destination address 109 * are in "incoming" packet order (ie, source is "far" address.) 110 */ 111 int 112 rss_proto_software_hash_v4(struct in_addr s, struct in_addr d, 113 u_short sp, u_short dp, int proto, 114 uint32_t *hashval, uint32_t *hashtype) 115 { 116 uint32_t hash; 117 118 /* 119 * Next, choose the hash type depending upon the protocol 120 * identifier. 121 */ 122 if ((proto == IPPROTO_TCP) && 123 (rss_gethashconfig() & RSS_HASHTYPE_RSS_TCP_IPV4)) { 124 hash = rss_hash_ip4_4tuple(s, sp, d, dp); 125 *hashval = hash; 126 *hashtype = M_HASHTYPE_RSS_TCP_IPV4; 127 return (0); 128 } else if ((proto == IPPROTO_UDP) && 129 (rss_gethashconfig() & RSS_HASHTYPE_RSS_UDP_IPV4)) { 130 hash = rss_hash_ip4_4tuple(s, sp, d, dp); 131 *hashval = hash; 132 *hashtype = M_HASHTYPE_RSS_UDP_IPV4; 133 return (0); 134 } else if (rss_gethashconfig() & RSS_HASHTYPE_RSS_IPV4) { 135 /* RSS doesn't hash on other protocols like SCTP; so 2-tuple */ 136 hash = rss_hash_ip4_2tuple(s, d); 137 *hashval = hash; 138 *hashtype = M_HASHTYPE_RSS_IPV4; 139 return (0); 140 } 141 142 /* No configured available hashtypes! */ 143 RSS_DEBUG("no available hashtypes!\n"); 144 return (-1); 145 } 146 147 /* 148 * Calculate an appropriate ipv4 2-tuple or 4-tuple given the given 149 * IPv4 source/destination address, UDP or TCP source/destination ports 150 * and the protocol type. 151 * 152 * The protocol code may wish to do a software hash of the given 153 * tuple. This depends upon the currently configured RSS hash types. 154 * 155 * It assumes the packet source/destination address 156 * are in "outgoing" packet order (ie, destination is "far" address.) 157 */ 158 uint32_t 159 xps_proto_software_hash_v4(struct in_addr s, struct in_addr d, 160 u_short sp, u_short dp, int proto, uint32_t *hashtype) 161 { 162 uint32_t hash; 163 164 /* 165 * Next, choose the hash type depending upon the protocol 166 * identifier. 167 */ 168 if ((proto == IPPROTO_TCP) && 169 (rss_gethashconfig() & RSS_HASHTYPE_RSS_TCP_IPV4)) { 170 hash = rss_hash_ip4_4tuple(d, dp, s, sp); 171 *hashtype = M_HASHTYPE_RSS_TCP_IPV4; 172 return (hash); 173 } else if ((proto == IPPROTO_UDP) && 174 (rss_gethashconfig() & RSS_HASHTYPE_RSS_UDP_IPV4)) { 175 hash = rss_hash_ip4_4tuple(d, dp, s, sp); 176 *hashtype = M_HASHTYPE_RSS_UDP_IPV4; 177 return (hash); 178 } else if (rss_gethashconfig() & RSS_HASHTYPE_RSS_IPV4) { 179 /* RSS doesn't hash on other protocols like SCTP; so 2-tuple */ 180 hash = rss_hash_ip4_2tuple(d, s); 181 *hashtype = M_HASHTYPE_RSS_IPV4; 182 return (hash); 183 } 184 185 *hashtype = M_HASHTYPE_NONE; 186 return (0); 187 } 188 189 /* 190 * Do a software calculation of the RSS for the given mbuf. 191 * 192 * This is typically used by the input path to recalculate the RSS after 193 * some form of packet processing (eg de-capsulation, IP fragment reassembly.) 194 * 195 * dir is the packet direction - RSS_HASH_PKT_INGRESS for incoming and 196 * RSS_HASH_PKT_EGRESS for outgoing. 197 * 198 * Returns 0 if a hash was done, -1 if no hash was done, +1 if 199 * the mbuf already had a valid RSS flowid. 200 * 201 * This function doesn't modify the mbuf. It's up to the caller to 202 * assign flowid/flowtype as appropriate. 203 */ 204 int 205 rss_mbuf_software_hash_v4(const struct mbuf *m, int dir, uint32_t *hashval, 206 uint32_t *hashtype) 207 { 208 const struct ip *ip; 209 const struct tcphdr *th; 210 const struct udphdr *uh; 211 uint32_t flowtype; 212 uint8_t proto; 213 int iphlen; 214 int is_frag = 0; 215 216 /* 217 * XXX For now this only handles hashing on incoming mbufs. 218 */ 219 if (dir != RSS_HASH_PKT_INGRESS) { 220 RSS_DEBUG("called on EGRESS packet!\n"); 221 return (-1); 222 } 223 224 /* 225 * First, validate that the mbuf we have is long enough 226 * to have an IPv4 header in it. 227 */ 228 if (m->m_pkthdr.len < (sizeof(struct ip))) { 229 RSS_DEBUG("short mbuf pkthdr\n"); 230 return (-1); 231 } 232 if (m->m_len < (sizeof(struct ip))) { 233 RSS_DEBUG("short mbuf len\n"); 234 return (-1); 235 } 236 237 /* Ok, let's dereference that */ 238 ip = mtod(m, struct ip *); 239 proto = ip->ip_p; 240 iphlen = ip->ip_hl << 2; 241 242 /* 243 * If this is a fragment then it shouldn't be four-tuple 244 * hashed just yet. Once it's reassembled into a full 245 * frame it should be re-hashed. 246 */ 247 if (ip->ip_off & htons(IP_MF | IP_OFFMASK)) 248 is_frag = 1; 249 250 /* 251 * If the mbuf flowid/flowtype matches the packet type, 252 * and we don't support the 4-tuple version of the given protocol, 253 * then signal to the owner that it can trust the flowid/flowtype 254 * details. 255 * 256 * This is a little picky - eg, if TCPv4 / UDPv4 hashing 257 * is supported but we got a TCP/UDP frame only 2-tuple hashed, 258 * then we shouldn't just "trust" the 2-tuple hash. We need 259 * a 4-tuple hash. 260 */ 261 flowtype = M_HASHTYPE_GET(m); 262 263 if (flowtype != M_HASHTYPE_NONE) { 264 switch (proto) { 265 case IPPROTO_UDP: 266 if ((rss_gethashconfig() & RSS_HASHTYPE_RSS_UDP_IPV4) && 267 (flowtype == M_HASHTYPE_RSS_UDP_IPV4) && 268 (is_frag == 0)) { 269 return (1); 270 } 271 /* 272 * Only allow 2-tuple for UDP frames if we don't also 273 * support 4-tuple for UDP. 274 */ 275 if ((rss_gethashconfig() & RSS_HASHTYPE_RSS_IPV4) && 276 ((rss_gethashconfig() & RSS_HASHTYPE_RSS_UDP_IPV4) == 0) && 277 flowtype == M_HASHTYPE_RSS_IPV4) { 278 return (1); 279 } 280 break; 281 case IPPROTO_TCP: 282 if ((rss_gethashconfig() & RSS_HASHTYPE_RSS_TCP_IPV4) && 283 (flowtype == M_HASHTYPE_RSS_TCP_IPV4) && 284 (is_frag == 0)) { 285 return (1); 286 } 287 /* 288 * Only allow 2-tuple for TCP frames if we don't also 289 * support 4-tuple for TCP. 290 */ 291 if ((rss_gethashconfig() & RSS_HASHTYPE_RSS_IPV4) && 292 ((rss_gethashconfig() & RSS_HASHTYPE_RSS_TCP_IPV4) == 0) && 293 flowtype == M_HASHTYPE_RSS_IPV4) { 294 return (1); 295 } 296 break; 297 default: 298 if ((rss_gethashconfig() & RSS_HASHTYPE_RSS_IPV4) && 299 flowtype == M_HASHTYPE_RSS_IPV4) { 300 return (1); 301 } 302 break; 303 } 304 } 305 306 /* 307 * Decode enough information to make a hash decision. 308 * 309 * XXX TODO: does the hardware hash on 4-tuple if IP 310 * options are present? 311 */ 312 if ((rss_gethashconfig() & RSS_HASHTYPE_RSS_TCP_IPV4) && 313 (proto == IPPROTO_TCP) && 314 (is_frag == 0)) { 315 if (m->m_len < iphlen + sizeof(struct tcphdr)) { 316 RSS_DEBUG("short TCP frame?\n"); 317 return (-1); 318 } 319 th = (const struct tcphdr *)((c_caddr_t)ip + iphlen); 320 return rss_proto_software_hash_v4(ip->ip_src, ip->ip_dst, 321 th->th_sport, 322 th->th_dport, 323 proto, 324 hashval, 325 hashtype); 326 } else if ((rss_gethashconfig() & RSS_HASHTYPE_RSS_UDP_IPV4) && 327 (proto == IPPROTO_UDP) && 328 (is_frag == 0)) { 329 uh = (const struct udphdr *)((c_caddr_t)ip + iphlen); 330 if (m->m_len < iphlen + sizeof(struct udphdr)) { 331 RSS_DEBUG("short UDP frame?\n"); 332 return (-1); 333 } 334 return rss_proto_software_hash_v4(ip->ip_src, ip->ip_dst, 335 uh->uh_sport, 336 uh->uh_dport, 337 proto, 338 hashval, 339 hashtype); 340 } else if (rss_gethashconfig() & RSS_HASHTYPE_RSS_IPV4) { 341 /* Default to 2-tuple hash */ 342 return rss_proto_software_hash_v4(ip->ip_src, ip->ip_dst, 343 0, /* source port */ 344 0, /* destination port */ 345 0, /* IPPROTO_IP */ 346 hashval, 347 hashtype); 348 } else { 349 RSS_DEBUG("no available hashtypes!\n"); 350 return (-1); 351 } 352 } 353 354 #ifdef RSS 355 /* 356 * Similar to rss_m2cpuid, but designed to be used by the IP NETISR 357 * on incoming frames. 358 * 359 * If an existing RSS hash exists and it matches what the configured 360 * hashing is, then use it. 361 * 362 * If there's an existing RSS hash but the desired hash is different, 363 * or if there's no useful RSS hash, then calculate it via 364 * the software path. 365 * 366 * XXX TODO: definitely want statistics here! 367 */ 368 struct mbuf * 369 rss_soft_m2cpuid_v4(struct mbuf *m, uintptr_t source, u_int *cpuid) 370 { 371 uint32_t hash_val, hash_type; 372 int ret; 373 374 M_ASSERTPKTHDR(m); 375 376 ret = rss_mbuf_software_hash_v4(m, RSS_HASH_PKT_INGRESS, 377 &hash_val, &hash_type); 378 if (ret > 0) { 379 /* mbuf has a valid hash already; don't need to modify it */ 380 *cpuid = rss_hash2cpuid(m->m_pkthdr.flowid, M_HASHTYPE_GET(m)); 381 } else if (ret == 0) { 382 /* hash was done; update */ 383 m->m_pkthdr.flowid = hash_val; 384 M_HASHTYPE_SET(m, hash_type); 385 *cpuid = rss_hash2cpuid(m->m_pkthdr.flowid, M_HASHTYPE_GET(m)); 386 } else { /* ret < 0 */ 387 /* no hash was done */ 388 *cpuid = NETISR_CPUID_NONE; 389 } 390 return (m); 391 } 392 #endif 393