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 1.38 2008/03/26 14:44:59 sephe 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                 goto fail;
  97         }
  98
  99         /* The fixed IP header must reside completely in the first mbuf. */
 100         if (m->m_len < sizeof(struct ip)) {
 101                 m = m_pullup(m, sizeof(struct ip));
 102                 if (m == NULL) {
 103                         ipstat.ips_toosmall++;
 104                         goto fail;
 105                 }
 106         }
 107
 108         ip = mtod(m, struct ip *);
 109
 110         /* Bound check the packet's stated IP header length. */
 111         iphlen = ip->ip_hl << 2;
 112         if (iphlen < sizeof(struct ip)) {       /* minimum header length */
 113                 ipstat.ips_badhlen++;
 114                 goto fail;
 115         }
 116
 117         /* The full IP header must reside completely in the one mbuf. */
 118         if (m->m_len < iphlen) {
 119                 m = m_pullup(m, iphlen);
 120                 if (m == NULL) {
 121                         ipstat.ips_badhlen++;
 122                         goto fail;
 123                 }
 124                 ip = mtod(m, struct ip *);
 125         }
 126
 127         iplen = ntohs(ip->ip_len);
 128
 129         /*
 130          * Fragments other than the first fragment don't have much
 131          * length information.
 132          */
 133         if (ntohs(ip->ip_off) & IP_OFFMASK)
 134                 goto ipcheckonly;
 135
 136         /*
 137          * The TCP/IP or UDP/IP header must be entirely contained within
 138          * the first fragment of a packet.  Packet filters will break if they
 139          * aren't.
 140          *
 141          * Since the packet will be trimmed to ip_len we must also make sure
 142          * the potentially trimmed down length is still sufficient to hold
 143          * the header(s).
 144          */
 145         switch (ip->ip_p) {
 146         case IPPROTO_TCP:
 147                 if (iplen < iphlen + sizeof(struct tcphdr)) {
 148                         ++tcpstat.tcps_rcvshort;
 149                         goto fail;
 150                 }
 151                 if (m->m_len < iphlen + sizeof(struct tcphdr)) {
 152                         m = m_pullup(m, iphlen + sizeof(struct tcphdr));
 153                         if (m == NULL) {
 154                                 tcpstat.tcps_rcvshort++;
 155                                 goto fail;
 156                         }
 157                         ip = mtod(m, struct ip *);
 158                 }
 159                 th = (struct tcphdr *)((caddr_t)ip + iphlen);
 160                 thoff = th->th_off << 2;
 161                 if (thoff < sizeof(struct tcphdr) ||
 162                     thoff + iphlen > ntohs(ip->ip_len)) {
 163                         tcpstat.tcps_rcvbadoff++;
 164                         goto fail;
 165                 }
 166                 if (m->m_len < iphlen + thoff) {
 167                         m = m_pullup(m, iphlen + thoff);
 168                         if (m == NULL) {
 169                                 tcpstat.tcps_rcvshort++;
 170                                 goto fail;
 171                         }
 172                 }
 173                 break;
 174         case IPPROTO_UDP:
 175                 if (iplen < iphlen + sizeof(struct udphdr)) {
 176                         ++udpstat.udps_hdrops;
 177                         goto fail;
 178                 }
 179                 if (m->m_len < iphlen + sizeof(struct udphdr)) {
 180                         m = m_pullup(m, iphlen + sizeof(struct udphdr));
 181                         if (m == NULL) {
 182                                 udpstat.udps_hdrops++;
 183                                 goto fail;
 184                         }
 185                 }
 186                 break;
 187         default:
 188 ipcheckonly:
 189                 if (iplen < iphlen) {
 190                         ++ipstat.ips_badlen;
 191                         goto fail;
 192                 }
 193                 break;
 194         }
 195
 196         *mp = m;
 197         return TRUE;
 198
 199 fail:
 200         if (m != NULL)
 201                 m_freem(m);
 202         *mp = NULL;
 203         return FALSE;
 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, int dir)
 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                 return (NULL);
 226
 227         m = *mptr;
 228         ip = mtod(m, struct ip *);
 229         iphlen = ip->ip_hl << 2;
 230
 231         /*
 232          * XXX generic packet handling defrag on CPU 0 for now.
 233          */
 234         if (ntohs(ip->ip_off) & (IP_MF | IP_OFFMASK))
 235                 return (&netisr_cpu[0].td_msgport);
 236
 237         switch (ip->ip_p) {
 238         case IPPROTO_TCP:
 239                 th = (struct tcphdr *)((caddr_t)ip + iphlen);
 240                 thoff = th->th_off << 2;
 241                 cpu = INP_MPORT_HASH(ip->ip_src.s_addr, ip->ip_dst.s_addr,
 242                     th->th_sport, th->th_dport);
 243                 port = &tcp_thread[cpu].td_msgport;
 244                 break;
 245         case IPPROTO_UDP:
 246                 uh = (struct udphdr *)((caddr_t)ip + iphlen);
 247
 248                 if (IN_MULTICAST(ntohl(ip->ip_dst.s_addr)) ||
 249                     (dir == IP_MPORT_IN &&
 250                      in_broadcast(ip->ip_dst, m->m_pkthdr.rcvif))) {
 251                         cpu = 0;
 252                 } else {
 253                         cpu = INP_MPORT_HASH(ip->ip_src.s_addr,
 254                             ip->ip_dst.s_addr, uh->uh_sport, uh->uh_dport);
 255                 }
 256                 port = &udp_thread[cpu].td_msgport;
 257                 break;
 258         default:
 259                 port = &netisr_cpu[0].td_msgport;
 260                 break;
 261         }
 262
 263         return (port);
 264 }
 265
 266 lwkt_port_t
 267 ip_mport_in(struct mbuf **mptr)
 268 {
 269         return ip_mport(mptr, IP_MPORT_IN);
 270 }
 271
 272 /*
 273  * Map a TCP socket to a protocol processing thread.
 274  */
 275 lwkt_port_t
 276 tcp_soport(struct socket *so, struct sockaddr *nam __unused,
 277            struct mbuf **dummy __unused, int req)
 278 {
 279         struct inpcb *inp;
 280
 281         /* The following processing all take place on Protocol Thread 0. */
 282         if (req == PRU_BIND || req == PRU_CONNECT || req == PRU_ATTACH ||
 283             req == PRU_LISTEN)
 284                 return (&tcp_thread[0].td_msgport);
 285
 286         inp = so->so_pcb;
 287         if (!inp)               /* connection reset by peer */
 288                 return (&tcp_thread[0].td_msgport);
 289
 290         /*
 291          * Already bound and connected or listening.  For TCP connections,
 292          * the (faddr, fport, laddr, lport) association cannot change now.
 293          *
 294          * Note: T/TCP code needs some reorganization to fit into
 295          * this model.  XXX JH
 296          *
 297          * Rely on type-stable memory and check in protocol handler
 298          * to fix race condition here w/ deallocation of inp.  XXX JH
 299          */
 300         return (&tcp_thread[INP_MPORT_HASH(inp->inp_faddr.s_addr,
 301             inp->inp_laddr.s_addr, inp->inp_fport, inp->inp_lport)].td_msgport);
 302 }
 303
 304 lwkt_port_t
 305 tcp_addrport(in_addr_t faddr, in_port_t fport, in_addr_t laddr, in_port_t lport)
 306 {
 307         return (&tcp_thread[tcp_addrcpu(faddr, fport,
 308                                         laddr, lport)].td_msgport);
 309 }
 310
 311 /*
 312  * Map a UDP socket to a protocol processing thread.
 313  */
 314 lwkt_port_t
 315 udp_soport(struct socket *so, struct sockaddr *nam __unused,
 316            struct mbuf **dummy __unused, int req)
 317 {
 318         struct inpcb *inp;
 319
 320         /*
 321          * The following processing all take place on Protocol Thread 0:
 322          *   bind()
 323          *   attach() has a null socket parameter
 324          *   Fast and slow timeouts pass in null socket parameter
 325          */
 326         if (req == PRU_BIND || so == NULL)
 327                 return (&udp_thread[0].td_msgport);
 328
 329         inp = so->so_pcb;
 330
 331         if (IN_MULTICAST(ntohl(inp->inp_laddr.s_addr)))
 332                 return (&udp_thread[0].td_msgport);
 333
 334         /*
 335          * Rely on type-stable memory and check in protocol handler
 336          * to fix race condition here w/ deallocation of inp.  XXX JH
 337          */
 338
 339         return (&udp_thread[INP_MPORT_HASH(inp->inp_faddr.s_addr,
 340             inp->inp_laddr.s_addr, inp->inp_fport, inp->inp_lport)].td_msgport);
 341 }
 342
 343 /*
 344  * Map a network address to a processor.
 345  */
 346 int
 347 tcp_addrcpu(in_addr_t faddr, in_port_t fport, in_addr_t laddr, in_port_t lport)
 348 {
 349         return (INP_MPORT_HASH(faddr, laddr, fport, lport));
 350 }
 351
 352 int
 353 udp_addrcpu(in_addr_t faddr, in_port_t fport, in_addr_t laddr, in_port_t lport)
 354 {
 355         if (IN_MULTICAST(ntohl(laddr)))
 356                 return (0);
 357         else
 358                 return (INP_MPORT_HASH(faddr, laddr, fport, lport));
 359 }
 360
 361 /*
 362  * Return LWKT port for cpu.
 363  */
 364 lwkt_port_t
 365 tcp_cport(int cpu)
 366 {
 367         return (&tcp_thread[cpu].td_msgport);
 368 }
 369
 370 void
 371 tcp_thread_init(void)
 372 {
 373         int cpu;
 374
 375         for (cpu = 0; cpu < ncpus2; cpu++) {
 376                 lwkt_create(tcpmsg_service_loop, NULL, NULL,
 377                         &tcp_thread[cpu], 0, cpu, "tcp_thread %d", cpu);
 378                 netmsg_service_port_init(&tcp_thread[cpu].td_msgport);
 379         }
 380 }
 381
 382 void
 383 udp_thread_init(void)
 384 {
 385         int cpu;
 386
 387         for (cpu = 0; cpu < ncpus2; cpu++) {
 388                 lwkt_create(netmsg_service_loop, NULL, NULL,
 389                         &udp_thread[cpu], 0, cpu, "udp_thread %d", cpu);
 390                 netmsg_service_port_init(&udp_thread[cpu].td_msgport);
 391         }
 392 }