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