sys/netinet/ip_demux.c

   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 1.34 2007/03/04 18:51:59 swildner Exp $
  34  */
  35
  36 #include "opt_inet.h"
  37
  38 #include <sys/param.h>
  39 #include <sys/systm.h>
  40 #include <sys/kernel.h>
  41 #include <sys/socket.h>
  42 #include <sys/socketvar.h>
  43 #include <sys/thread.h>
  44 #include <sys/sysctl.h>
  45 #include <sys/globaldata.h>
  46
  47 #include <net/if.h>
  48 #include <net/netisr.h>
  49
  50 #include <netinet/in_systm.h>
  51 #include <netinet/in.h>
  52 #include <netinet/in_var.h>
  53 #include <netinet/in_pcb.h>
  54 #include <netinet/ip.h>
  55 #include <netinet/ip_var.h>
  56 #include <netinet/tcp.h>
  57 #include <netinet/tcpip.h>
  58 #include <netinet/tcp_var.h>
  59 #include <netinet/udp.h>
  60 #include <netinet/udp_var.h>
  61
  62 extern struct thread netisr_cpu[];
  63
  64 static struct thread tcp_thread[MAXCPU];
  65 static struct thread udp_thread[MAXCPU];
  66
  67 static __inline int
  68 INP_MPORT_HASH(in_addr_t faddr, in_addr_t laddr,
  69                in_port_t fport, in_port_t lport)
  70 {
  71         /*
  72          * Use low order bytes.
  73          */
  74
  75 #if (BYTE_ORDER == LITTLE_ENDIAN)
  76         KASSERT(ncpus2 < 256, ("need different hash function"));  /* XXX JH */
  77         return (((faddr >> 24) ^ (fport >> 8) ^ (laddr >> 24) ^ (lport >> 8)) &
  78                 ncpus2_mask);
  79 #else
  80         return ((faddr ^ fport ^ laddr ^ lport) & ncpus2_mask);
  81 #endif
  82 }
  83
  84 boolean_t
  85 ip_lengthcheck(struct mbuf **mp)
  86 {
  87         struct mbuf *m = *mp;
  88         struct ip *ip;
  89         int iphlen, iplen;
  90         struct tcphdr *th;
  91         int thoff;                              /* TCP data offset */
  92
  93         /* The packet must be at least the size of an IP header. */
  94         if (m->m_pkthdr.len < sizeof(struct ip)) {
  95                 ipstat.ips_tooshort++;
  96                 m_free(m);
  97                 return FALSE;
  98         }
  99
 100         /* The fixed IP header must reside completely in the first mbuf. */
 101         if (m->m_len < sizeof(struct ip)) {
 102                 m = m_pullup(m, sizeof(struct ip));
 103                 if (m == NULL) {
 104                         ipstat.ips_toosmall++;
 105                         return FALSE;
 106                 }
 107         }
 108
 109         ip = mtod(m, struct ip *);
 110
 111         /* Bound check the packet's stated IP header length. */
 112         iphlen = ip->ip_hl << 2;
 113         if (iphlen < sizeof(struct ip)) {       /* minimum header length */
 114                 ipstat.ips_badhlen++;
 115                 m_free(m);
 116                 return FALSE;
 117         }
 118
 119         /* The full IP header must reside completely in the one mbuf. */
 120         if (m->m_len < iphlen) {
 121                 m = m_pullup(m, iphlen);
 122                 if (m == NULL) {
 123                         ipstat.ips_badhlen++;
 124                         return FALSE;
 125                 }
 126                 ip = mtod(m, struct ip *);
 127         }
 128
 129         iplen = ntohs(ip->ip_len);
 130
 131         /*
 132          * Fragments other than the first fragment don't have much
 133          * length information.
 134          */
 135         if (ntohs(ip->ip_off) & IP_OFFMASK)
 136                 goto ipcheckonly;
 137
 138         /*
 139          * The TCP/IP or UDP/IP header must be entirely contained within
 140          * the first fragment of a packet.  Packet filters will break if they
 141          * aren't.
 142          *
 143          * Since the packet will be trimmed to ip_len we must also make sure
 144          * the potentially trimmed down length is still sufficient to hold
 145          * the header(s).
 146          */
 147         switch (ip->ip_p) {
 148         case IPPROTO_TCP:
 149                 if (iplen < iphlen + sizeof(struct tcphdr)) {
 150                         ++tcpstat.tcps_rcvshort;
 151                         m_free(m);
 152                         return FALSE;
 153                 }
 154                 if (m->m_len < iphlen + sizeof(struct tcphdr)) {
 155                         m = m_pullup(m, iphlen + sizeof(struct tcphdr));
 156                         if (m == NULL) {
 157                                 tcpstat.tcps_rcvshort++;
 158                                 return FALSE;
 159                         }
 160                         ip = mtod(m, struct ip *);
 161                 }
 162                 th = (struct tcphdr *)((caddr_t)ip + iphlen);
 163                 thoff = th->th_off << 2;
 164                 if (thoff < sizeof(struct tcphdr) ||
 165                     thoff + iphlen > ntohs(ip->ip_len)) {
 166                         tcpstat.tcps_rcvbadoff++;
 167                         m_free(m);
 168                         return FALSE;
 169                 }
 170                 if (m->m_len < iphlen + thoff) {
 171                         m = m_pullup(m, iphlen + thoff);
 172                         if (m == NULL) {
 173                                 tcpstat.tcps_rcvshort++;
 174                                 return FALSE;
 175                         }
 176                 }
 177                 break;
 178         case IPPROTO_UDP:
 179                 if (iplen < iphlen + sizeof(struct udphdr)) {
 180                         ++udpstat.udps_hdrops;
 181                         m_free(m);
 182                         return FALSE;
 183                 }
 184                 if (m->m_len < iphlen + sizeof(struct udphdr)) {
 185                         m = m_pullup(m, iphlen + sizeof(struct udphdr));
 186                         if (m == NULL) {
 187                                 udpstat.udps_hdrops++;
 188                                 return FALSE;
 189                         }
 190                 }
 191                 break;
 192         default:
 193 ipcheckonly:
 194                 if (iplen < iphlen) {
 195                         ++ipstat.ips_badlen;
 196                         m_free(m);
 197                         return FALSE;
 198                 }
 199                 break;
 200         }
 201
 202         *mp = m;
 203         return TRUE;
 204 }
 205
 206 /*
 207  * Map a packet to a protocol processing thread and return the thread's port.
 208  * If an error occurs, the passed mbuf will be freed, *mptr will be set
 209  * to NULL, and NULL will be returned.  If no error occurs, the passed mbuf
 210  * may be modified and a port pointer will be returned.
 211  */
 212 lwkt_port_t
 213 ip_mport(struct mbuf **mptr)
 214 {
 215         struct ip *ip;
 216         int iphlen;
 217         struct tcphdr *th;
 218         struct udphdr *uh;
 219         struct mbuf *m;
 220         int thoff;                              /* TCP data offset */
 221         lwkt_port_t port;
 222         int cpu;
 223
 224         if (!ip_lengthcheck(mptr)) {
 225                 *mptr = NULL;
 226                 return (NULL);
 227         }
 228
 229         m = *mptr;
 230         ip = mtod(m, struct ip *);
 231         iphlen = ip->ip_hl << 2;
 232
 233         /*
 234          * XXX generic packet handling defrag on CPU 0 for now.
 235          */
 236         if (ntohs(ip->ip_off) & (IP_MF | IP_OFFMASK))
 237                 return (&netisr_cpu[0].td_msgport);
 238
 239         switch (ip->ip_p) {
 240         case IPPROTO_TCP:
 241                 th = (struct tcphdr *)((caddr_t)ip + iphlen);
 242                 thoff = th->th_off << 2;
 243                 cpu = INP_MPORT_HASH(ip->ip_src.s_addr, ip->ip_dst.s_addr,
 244                     th->th_sport, th->th_dport);
 245                 port = &tcp_thread[cpu].td_msgport;
 246                 break;
 247         case IPPROTO_UDP:
 248                 uh = (struct udphdr *)((caddr_t)ip + iphlen);
 249
 250                 if (IN_MULTICAST(ntohl(ip->ip_dst.s_addr)) ||
 251                     in_broadcast(ip->ip_dst, m->m_pkthdr.rcvif)) {
 252                         cpu = 0;
 253                 } else {
 254                         cpu = INP_MPORT_HASH(ip->ip_src.s_addr,
 255                             ip->ip_dst.s_addr, uh->uh_sport, uh->uh_dport);
 256                 }
 257                 port = &udp_thread[cpu].td_msgport;
 258                 break;
 259         default:
 260                 port = &netisr_cpu[0].td_msgport;
 261                 break;
 262         }
 263
 264         return (port);
 265 }
 266
 267 /*
 268  * Map a TCP socket to a protocol processing thread.
 269  */
 270 lwkt_port_t
 271 tcp_soport(struct socket *so, struct sockaddr *nam, int req)
 272 {
 273         struct inpcb *inp;
 274
 275         /* The following processing all take place on Protocol Thread 0. */
 276         if (req == PRU_BIND || req == PRU_CONNECT || req == PRU_ATTACH ||
 277             req == PRU_LISTEN)
 278                 return (&tcp_thread[0].td_msgport);
 279
 280         inp = so->so_pcb;
 281         if (!inp)               /* connection reset by peer */
 282                 return (&tcp_thread[0].td_msgport);
 283
 284         /*
 285          * Already bound and connected or listening.  For TCP connections,
 286          * the (faddr, fport, laddr, lport) association cannot change now.
 287          *
 288          * Note: T/TCP code needs some reorganization to fit into
 289          * this model.  XXX JH
 290          *
 291          * Rely on type-stable memory and check in protocol handler
 292          * to fix race condition here w/ deallocation of inp.  XXX JH
 293          */
 294         return (&tcp_thread[INP_MPORT_HASH(inp->inp_faddr.s_addr,
 295             inp->inp_laddr.s_addr, inp->inp_fport, inp->inp_lport)].td_msgport);
 296 }
 297
 298 lwkt_port_t
 299 tcp_addrport(in_addr_t faddr, in_port_t fport, in_addr_t laddr, in_port_t lport)
 300 {
 301         return (&tcp_thread[tcp_addrcpu(faddr, fport,
 302                                         laddr, lport)].td_msgport);
 303 }
 304
 305 /*
 306  * Map a UDP socket to a protocol processing thread.
 307  */
 308 lwkt_port_t
 309 udp_soport(struct socket *so, struct sockaddr *nam, int req)
 310 {
 311         struct inpcb *inp;
 312
 313         /*
 314          * The following processing all take place on Protocol Thread 0:
 315          *   only bind() and connect() have a non-null nam parameter
 316          *   attach() has a null socket parameter
 317          *   Fast and slow timeouts pass in two NULLs
 318          */
 319         if (nam != NULL || so == NULL)
 320                 return (&udp_thread[0].td_msgport);
 321
 322         inp = so->so_pcb;
 323
 324         if (IN_MULTICAST(ntohl(inp->inp_laddr.s_addr)))
 325                 return (&udp_thread[0].td_msgport);
 326
 327         /*
 328          * Rely on type-stable memory and check in protocol handler
 329          * to fix race condition here w/ deallocation of inp.  XXX JH
 330          */
 331
 332         return (&udp_thread[INP_MPORT_HASH(inp->inp_faddr.s_addr,
 333             inp->inp_laddr.s_addr, inp->inp_fport, inp->inp_lport)].td_msgport);
 334 }
 335
 336 /*
 337  * Map a network address to a processor.
 338  */
 339 int
 340 tcp_addrcpu(in_addr_t faddr, in_port_t fport, in_addr_t laddr, in_port_t lport)
 341 {
 342         return (INP_MPORT_HASH(faddr, laddr, fport, lport));
 343 }
 344
 345 int
 346 udp_addrcpu(in_addr_t faddr, in_port_t fport, in_addr_t laddr, in_port_t lport)
 347 {
 348         if (IN_MULTICAST(ntohl(laddr)))
 349                 return (0);
 350         else
 351                 return (INP_MPORT_HASH(faddr, laddr, fport, lport));
 352 }
 353
 354 /*
 355  * Return LWKT port for cpu.
 356  */
 357 lwkt_port_t
 358 tcp_cport(int cpu)
 359 {
 360         return (&tcp_thread[cpu].td_msgport);
 361 }
 362
 363 void
 364 tcp_thread_init(void)
 365 {
 366         int cpu;
 367
 368         for (cpu = 0; cpu < ncpus2; cpu++) {
 369                 lwkt_create(tcpmsg_service_loop, NULL, NULL,
 370                         &tcp_thread[cpu], 0, cpu, "tcp_thread %d", cpu);
 371                 netmsg_service_port_init(&tcp_thread[cpu].td_msgport);
 372         }
 373 }
 374
 375 void
 376 udp_thread_init(void)
 377 {
 378         int cpu;
 379
 380         for (cpu = 0; cpu < ncpus2; cpu++) {
 381                 lwkt_create(netmsg_service_loop, NULL, NULL,
 382                         &udp_thread[cpu], 0, cpu, "udp_thread %d", cpu);
 383                 netmsg_service_port_init(&udp_thread[cpu].td_msgport);
 384         }
 385 }