1 /* 2 * Copyright (c) 2003, 2004 Jeffrey M. Hsu. All rights reserved. 3 * Copyright (c) 2003, 2004 The DragonFly Project. All rights reserved. 4 * 5 * This code is derived from software contributed to The DragonFly Project 6 * by Jeffrey M. Hsu. 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 * 3. Neither the name of The DragonFly Project nor the names of its 17 * contributors may be used to endorse or promote products derived 18 * from this software without specific, prior written permission. 19 * 20 * THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS 21 * ``AS IS'' AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT 22 * LIMITED TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS 23 * FOR A PARTICULAR PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL THE 24 * COPYRIGHT HOLDERS OR CONTRIBUTORS BE LIABLE FOR ANY DIRECT, INDIRECT, 25 * INCIDENTAL, SPECIAL, EXEMPLARY OR CONSEQUENTIAL DAMAGES (INCLUDING, 26 * BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; 27 * LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED 28 * AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, 29 * OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT 30 * OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF 31 * SUCH DAMAGE. 32 * 33 * $DragonFly: src/sys/netinet/ip_demux.c,v 1.45 2008/11/11 10:46:58 sephe Exp $ 34 */ 35 36 #include "opt_inet.h" 37 #include "opt_rss.h" 38 39 #include <sys/param.h> 40 #include <sys/systm.h> 41 #include <sys/kernel.h> 42 #include <sys/socket.h> 43 #include <sys/socketvar.h> 44 #include <sys/thread.h> 45 #include <sys/sysctl.h> 46 #include <sys/globaldata.h> 47 48 #include <net/if.h> 49 #include <net/netisr.h> 50 #include <net/toeplitz2.h> 51 52 #include <netinet/in_systm.h> 53 #include <netinet/in.h> 54 #include <netinet/in_var.h> 55 #include <netinet/in_pcb.h> 56 #include <netinet/ip.h> 57 #include <netinet/ip_var.h> 58 #include <netinet/tcp.h> 59 #include <netinet/tcpip.h> 60 #include <netinet/tcp_var.h> 61 #include <netinet/udp.h> 62 #include <netinet/udp_var.h> 63 64 extern int udp_mpsafe_thread; 65 66 /* 67 * Toeplitz hash functions - the idea is to match the hardware. 68 */ 69 static __inline int 70 INP_MPORT_HASH_UDP(in_addr_t faddr, in_addr_t laddr, 71 in_port_t fport, in_port_t lport) 72 { 73 return toeplitz_hash(toeplitz_rawhash_addr(faddr, laddr)); 74 } 75 76 static __inline int 77 INP_MPORT_HASH_TCP(in_addr_t faddr, in_addr_t laddr, 78 in_port_t fport, in_port_t lport) 79 { 80 return toeplitz_hash( 81 toeplitz_rawhash_addrport(faddr, laddr, fport, lport)); 82 } 83 84 /* 85 * Map a network address to a processor. 86 */ 87 int 88 tcp_addrcpu(in_addr_t faddr, in_port_t fport, in_addr_t laddr, in_port_t lport) 89 { 90 return (INP_MPORT_HASH_TCP(faddr, laddr, fport, lport)); 91 } 92 93 /* 94 * Not implemented yet, use protocol thread 0 95 */ 96 int 97 udp_addrcpu(in_addr_t faddr, in_port_t fport, in_addr_t laddr, in_port_t lport) 98 { 99 /*return (INP_MPORT_HASH_UDP(faddr, laddr, fport, lport));*/ 100 return 0; 101 } 102 103 /* 104 * If the packet is a valid IP datagram, upon returning of this function 105 * following things are promised: 106 * 107 * o IP header (including any possible IP options) and any data preceding 108 * IP header (usually linker layer header) are in one mbuf (m_len). 109 * o IP header length is not less than the minimum (sizeof(struct ip)). 110 * o IP total length is not less than IP header length. 111 * o IP datagram resides completely in the mbuf chain, 112 * i.e. pkthdr.len >= IP total length. 113 * 114 * If the packet is a UDP datagram, 115 * o IP header (including any possible IP options) and UDP header are in 116 * one mbuf (m_len). 117 * o IP total length is not less than (IP header length + UDP header length). 118 * 119 * If the packet is a TCP segment, 120 * o IP header (including any possible IP options) and TCP header (including 121 * any possible TCP options) are in one mbuf (m_len). 122 * o TCP header length is not less than the minimum (sizeof(struct tcphdr)). 123 * o IP total length is not less than (IP header length + TCP header length). 124 */ 125 boolean_t 126 ip_lengthcheck(struct mbuf **mp, int hoff) 127 { 128 struct mbuf *m = *mp; 129 struct ip *ip; 130 int len, iphlen, iplen; 131 struct tcphdr *th; 132 int thoff; /* TCP data offset */ 133 134 len = hoff + sizeof(struct ip); 135 136 /* The packet must be at least the size of an IP header. */ 137 if (m->m_pkthdr.len < len) { 138 ipstat.ips_tooshort++; 139 goto fail; 140 } 141 142 /* The fixed IP header must reside completely in the first mbuf. */ 143 if (m->m_len < len) { 144 m = m_pullup(m, len); 145 if (m == NULL) { 146 ipstat.ips_toosmall++; 147 goto fail; 148 } 149 } 150 151 ip = mtodoff(m, struct ip *, hoff); 152 153 /* Bound check the packet's stated IP header length. */ 154 iphlen = ip->ip_hl << 2; 155 if (iphlen < sizeof(struct ip)) { /* minimum header length */ 156 ipstat.ips_badhlen++; 157 goto fail; 158 } 159 160 /* The full IP header must reside completely in the one mbuf. */ 161 if (m->m_len < hoff + iphlen) { 162 m = m_pullup(m, hoff + iphlen); 163 if (m == NULL) { 164 ipstat.ips_badhlen++; 165 goto fail; 166 } 167 ip = mtodoff(m, struct ip *, hoff); 168 } 169 170 iplen = ntohs(ip->ip_len); 171 172 /* 173 * Check that the amount of data in the buffers is as 174 * at least much as the IP header would have us expect. 175 */ 176 if (m->m_pkthdr.len < hoff + iplen) { 177 ipstat.ips_tooshort++; 178 goto fail; 179 } 180 181 /* 182 * Fragments other than the first fragment don't have much 183 * length information. 184 */ 185 if (ntohs(ip->ip_off) & IP_OFFMASK) 186 goto ipcheckonly; 187 188 /* 189 * The TCP/IP or UDP/IP header must be entirely contained within 190 * the first fragment of a packet. Packet filters will break if they 191 * aren't. 192 * 193 * Since the packet will be trimmed to ip_len we must also make sure 194 * the potentially trimmed down length is still sufficient to hold 195 * the header(s). 196 */ 197 switch (ip->ip_p) { 198 case IPPROTO_TCP: 199 if (iplen < iphlen + sizeof(struct tcphdr)) { 200 ++tcpstat.tcps_rcvshort; 201 goto fail; 202 } 203 if (m->m_len < hoff + iphlen + sizeof(struct tcphdr)) { 204 m = m_pullup(m, hoff + iphlen + sizeof(struct tcphdr)); 205 if (m == NULL) { 206 tcpstat.tcps_rcvshort++; 207 goto fail; 208 } 209 ip = mtodoff(m, struct ip *, hoff); 210 } 211 th = (struct tcphdr *)((caddr_t)ip + iphlen); 212 thoff = th->th_off << 2; 213 if (thoff < sizeof(struct tcphdr) || 214 thoff + iphlen > ntohs(ip->ip_len)) { 215 tcpstat.tcps_rcvbadoff++; 216 goto fail; 217 } 218 if (m->m_len < hoff + iphlen + thoff) { 219 m = m_pullup(m, hoff + iphlen + thoff); 220 if (m == NULL) { 221 tcpstat.tcps_rcvshort++; 222 goto fail; 223 } 224 } 225 break; 226 case IPPROTO_UDP: 227 if (iplen < iphlen + sizeof(struct udphdr)) { 228 ++udpstat.udps_hdrops; 229 goto fail; 230 } 231 if (m->m_len < hoff + iphlen + sizeof(struct udphdr)) { 232 m = m_pullup(m, hoff + iphlen + sizeof(struct udphdr)); 233 if (m == NULL) { 234 udpstat.udps_hdrops++; 235 goto fail; 236 } 237 } 238 break; 239 default: 240 ipcheckonly: 241 if (iplen < iphlen) { 242 ++ipstat.ips_badlen; 243 goto fail; 244 } 245 break; 246 } 247 248 m->m_flags |= M_LENCHECKED; 249 *mp = m; 250 return TRUE; 251 252 fail: 253 if (m != NULL) 254 m_freem(m); 255 *mp = NULL; 256 return FALSE; 257 } 258 259 /* 260 * Assign a protocol processing thread to a packet. The IP header is at 261 * offset (hoff) in the packet (i.e. the mac header might still be intact). 262 * 263 * This function can blow away the mbuf if the packet is malformed. 264 */ 265 void 266 ip_cpufn(struct mbuf **mptr, int hoff, int dir) 267 { 268 struct ip *ip; 269 int iphlen; 270 struct tcphdr *th; 271 struct udphdr *uh; 272 struct mbuf *m; 273 int thoff; /* TCP data offset */ 274 int cpu; 275 276 if (!ip_lengthcheck(mptr, hoff)) 277 return; 278 279 m = *mptr; 280 ip = mtodoff(m, struct ip *, hoff); 281 iphlen = ip->ip_hl << 2; 282 283 /* 284 * XXX generic packet handling defrag on CPU 0 for now. 285 */ 286 if (ntohs(ip->ip_off) & (IP_MF | IP_OFFMASK)) { 287 cpu = 0; 288 goto back; 289 } 290 291 switch (ip->ip_p) { 292 case IPPROTO_TCP: 293 th = (struct tcphdr *)((caddr_t)ip + iphlen); 294 thoff = th->th_off << 2; 295 cpu = INP_MPORT_HASH_TCP(ip->ip_src.s_addr, 296 ip->ip_dst.s_addr, 297 th->th_sport, 298 th->th_dport); 299 break; 300 301 case IPPROTO_UDP: 302 uh = (struct udphdr *)((caddr_t)ip + iphlen); 303 304 cpu = INP_MPORT_HASH_UDP(ip->ip_src.s_addr, 305 ip->ip_dst.s_addr, 306 uh->uh_sport, 307 uh->uh_dport); 308 break; 309 310 default: 311 cpu = 0; 312 break; 313 } 314 back: 315 m->m_flags |= M_HASH; 316 m->m_pkthdr.hash = cpu; 317 } 318 319 void 320 ip_cpufn_in(struct mbuf **mptr, int hoff) 321 { 322 ip_cpufn(mptr, hoff, IP_MPORT_IN); 323 } 324 325 #if 0 326 327 /* 328 * Map a packet to a protocol processing thread and return the thread's port. 329 * Unlike ip_cpufn(), the packet content is not accessed. The packet info 330 * (pi) and the hash of the packet (m_pkthdr.hash) is used instead. NULL is 331 * returned if the packet info does not contain enough information. 332 * 333 * Caller has already made sure that m_pkthdr.hash is valid, i.e. m_flags 334 * has M_HASH set. 335 */ 336 lwkt_port_t 337 ip_mport_pktinfo(const struct pktinfo *pi, struct mbuf *m) 338 { 339 lwkt_port_t port; 340 341 KASSERT(m->m_pkthdr.hash < ncpus2, 342 ("invalid packet hash %#x\n", m->m_pkthdr.hash)); 343 344 /* 345 * XXX generic packet handling defrag on CPU 0 for now. 346 */ 347 if (pi->pi_flags & PKTINFO_FLAG_FRAG) { 348 m->m_pkthdr.hash = 0; 349 return cpu_portfn(0); 350 } 351 352 switch (pi->pi_l3proto) { 353 case IPPROTO_TCP: 354 port = cpu_portfn(m->m_pkthdr.hash); 355 break; 356 357 case IPPROTO_UDP: 358 port = cpu_portfn(m->m_pkthdr.hash); 359 break; 360 361 default: 362 port = NULL; 363 break; 364 } 365 return port; 366 } 367 368 #endif 369 370 /* 371 * This is used to map a socket to a message port for sendmsg() and friends. 372 * It is not called for any other purpose. In the case of TCP we just return 373 * the port already installed in the socket. 374 */ 375 lwkt_port_t 376 tcp_soport(struct socket *so, struct sockaddr *nam, 377 struct mbuf **dummy __unused) 378 { 379 return(so->so_port); 380 } 381 382 /* 383 * Used to route icmp messages to the proper protocol thread for ctlinput 384 * operation. 385 */ 386 lwkt_port_t 387 tcp_ctlport(int cmd, struct sockaddr *sa, void *vip) 388 { 389 struct ip *ip = vip; 390 struct tcphdr *th; 391 struct in_addr faddr; 392 int cpu; 393 394 faddr = ((struct sockaddr_in *)sa)->sin_addr; 395 if (sa->sa_family != AF_INET || faddr.s_addr == INADDR_ANY) 396 return(NULL); 397 if (ip == NULL || PRC_IS_REDIRECT(cmd) || cmd == PRC_HOSTDEAD) { 398 /* 399 * A new message will be allocated later to save necessary 400 * information and will be forwarded to all network protocol 401 * threads in the following way: 402 * 403 * (the the thread owns the msgport that we return here) 404 * netisr0 <--+ 405 * | | 406 * | | 407 * | | 408 * +-------+ 409 * sendmsg 410 * [msg is kmalloc()ed] 411 * 412 * 413 * Later on, when the msg is received by netisr0: 414 * 415 * forwardmsg forwardmsg 416 * netisr0 ---------> netisr1 ---------> netisrN 417 * [msg is kfree()ed] 418 */ 419 return cpu0_ctlport(cmd, sa, vip); 420 } else { 421 th = (struct tcphdr *)((caddr_t)ip + (ip->ip_hl << 2)); 422 cpu = tcp_addrcpu(faddr.s_addr, th->th_dport, 423 ip->ip_src.s_addr, th->th_sport); 424 } 425 return(cpu_portfn(cpu)); 426 } 427 428 lwkt_port_t 429 tcp_addrport(in_addr_t faddr, in_port_t fport, in_addr_t laddr, in_port_t lport) 430 { 431 return(cpu_portfn(tcp_addrcpu(faddr, fport, laddr, lport))); 432 } 433 434 lwkt_port_t 435 tcp_addrport0(void) 436 { 437 return(cpu_portfn(0)); 438 } 439 440 lwkt_port_t 441 udp_addrport(in_addr_t faddr, in_port_t fport, in_addr_t laddr, in_port_t lport) 442 { 443 return(cpu_portfn(udp_addrcpu(faddr, fport, laddr, lport))); 444 } 445 446 /* 447 * Used to route icmp messages to the proper protocol thread for ctlinput 448 * operation. 449 */ 450 lwkt_port_t 451 udp_ctlport(int cmd, struct sockaddr *sa, void *vip) 452 { 453 struct ip *ip = vip; 454 struct udphdr *uh; 455 struct in_addr faddr; 456 int cpu; 457 458 faddr = ((struct sockaddr_in *)sa)->sin_addr; 459 if (sa->sa_family != AF_INET || faddr.s_addr == INADDR_ANY) 460 return(NULL); 461 if (PRC_IS_REDIRECT(cmd)) { 462 /* 463 * See the comment in tcp_ctlport; the only difference 464 * is that message is forwarded to UDP protocol theads. 465 */ 466 return cpu0_ctlport(cmd, sa, vip); 467 } else if (ip == NULL || cmd == PRC_HOSTDEAD) { 468 /* 469 * XXX 470 * Once UDP inpcbs are CPU localized, we should do 471 * the same forwarding as PRC_IS_REDIRECT(cmd) 472 */ 473 cpu = 0; 474 } else { 475 uh = (struct udphdr *)((caddr_t)ip + (ip->ip_hl << 2)); 476 477 cpu = INP_MPORT_HASH_UDP(faddr.s_addr, ip->ip_src.s_addr, 478 uh->uh_dport, uh->uh_sport); 479 } 480 return (cpu_portfn(cpu)); 481 } 482