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 1.39 2008/03/29 04:45:47 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 int      udp_mpsafe_thread = 0;
  68 TUNABLE_INT("net.inet.udp.mpsafe_thread", &udp_mpsafe_thread);
  69
  70 static __inline int
  71 INP_MPORT_HASH(in_addr_t faddr, in_addr_t laddr,
  72                in_port_t fport, in_port_t lport)
  73 {
  74         /*
  75          * Use low order bytes.
  76          */
  77
  78 #if (BYTE_ORDER == LITTLE_ENDIAN)
  79         KASSERT(ncpus2 < 256, ("need different hash function"));  /* XXX JH */
  80         return (((faddr >> 24) ^ (fport >> 8) ^ (laddr >> 24) ^ (lport >> 8)) &
  81                 ncpus2_mask);
  82 #else
  83         return ((faddr ^ fport ^ laddr ^ lport) & ncpus2_mask);
  84 #endif
  85 }
  86
  87 boolean_t
  88 ip_lengthcheck(struct mbuf **mp)
  89 {
  90         struct mbuf *m = *mp;
  91         struct ip *ip;
  92         int iphlen, iplen;
  93         struct tcphdr *th;
  94         int thoff;                              /* TCP data offset */
  95
  96         /* The packet must be at least the size of an IP header. */
  97         if (m->m_pkthdr.len < sizeof(struct ip)) {
  98                 ipstat.ips_tooshort++;
  99                 goto fail;
 100         }
 101
 102         /* The fixed IP header must reside completely in the first mbuf. */
 103         if (m->m_len < sizeof(struct ip)) {
 104                 m = m_pullup(m, sizeof(struct ip));
 105                 if (m == NULL) {
 106                         ipstat.ips_toosmall++;
 107                         goto fail;
 108                 }
 109         }
 110
 111         ip = mtod(m, struct ip *);
 112
 113         /* Bound check the packet's stated IP header length. */
 114         iphlen = ip->ip_hl << 2;
 115         if (iphlen < sizeof(struct ip)) {       /* minimum header length */
 116                 ipstat.ips_badhlen++;
 117                 goto fail;
 118         }
 119
 120         /* The full IP header must reside completely in the one mbuf. */
 121         if (m->m_len < iphlen) {
 122                 m = m_pullup(m, iphlen);
 123                 if (m == NULL) {
 124                         ipstat.ips_badhlen++;
 125                         goto fail;
 126                 }
 127                 ip = mtod(m, struct ip *);
 128         }
 129
 130         iplen = ntohs(ip->ip_len);
 131
 132         /*
 133          * Fragments other than the first fragment don't have much
 134          * length information.
 135          */
 136         if (ntohs(ip->ip_off) & IP_OFFMASK)
 137                 goto ipcheckonly;
 138
 139         /*
 140          * The TCP/IP or UDP/IP header must be entirely contained within
 141          * the first fragment of a packet.  Packet filters will break if they
 142          * aren't.
 143          *
 144          * Since the packet will be trimmed to ip_len we must also make sure
 145          * the potentially trimmed down length is still sufficient to hold
 146          * the header(s).
 147          */
 148         switch (ip->ip_p) {
 149         case IPPROTO_TCP:
 150                 if (iplen < iphlen + sizeof(struct tcphdr)) {
 151                         ++tcpstat.tcps_rcvshort;
 152                         goto fail;
 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                                 goto fail;
 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                         goto fail;
 168                 }
 169                 if (m->m_len < iphlen + thoff) {
 170                         m = m_pullup(m, iphlen + thoff);
 171                         if (m == NULL) {
 172                                 tcpstat.tcps_rcvshort++;
 173                                 goto fail;
 174                         }
 175                 }
 176                 break;
 177         case IPPROTO_UDP:
 178                 if (iplen < iphlen + sizeof(struct udphdr)) {
 179                         ++udpstat.udps_hdrops;
 180                         goto fail;
 181                 }
 182                 if (m->m_len < iphlen + sizeof(struct udphdr)) {
 183                         m = m_pullup(m, iphlen + sizeof(struct udphdr));
 184                         if (m == NULL) {
 185                                 udpstat.udps_hdrops++;
 186                                 goto fail;
 187                         }
 188                 }
 189                 break;
 190         default:
 191 ipcheckonly:
 192                 if (iplen < iphlen) {
 193                         ++ipstat.ips_badlen;
 194                         goto fail;
 195                 }
 196                 break;
 197         }
 198
 199         *mp = m;
 200         return TRUE;
 201
 202 fail:
 203         if (m != NULL)
 204                 m_freem(m);
 205         *mp = NULL;
 206         return FALSE;
 207 }
 208
 209 /*
 210  * Map a packet to a protocol processing thread and return the thread's port.
 211  * If an error occurs, the passed mbuf will be freed, *mptr will be set
 212  * to NULL, and NULL will be returned.  If no error occurs, the passed mbuf
 213  * may be modified and a port pointer will be returned.
 214  */
 215 lwkt_port_t
 216 ip_mport(struct mbuf **mptr, int dir)
 217 {
 218         struct ip *ip;
 219         int iphlen;
 220         struct tcphdr *th;
 221         struct udphdr *uh;
 222         struct mbuf *m;
 223         int thoff;                              /* TCP data offset */
 224         lwkt_port_t port;
 225         int cpu;
 226
 227         if (!ip_lengthcheck(mptr))
 228                 return (NULL);
 229
 230         m = *mptr;
 231         ip = mtod(m, struct ip *);
 232         iphlen = ip->ip_hl << 2;
 233
 234         /*
 235          * XXX generic packet handling defrag on CPU 0 for now.
 236          */
 237         if (ntohs(ip->ip_off) & (IP_MF | IP_OFFMASK))
 238                 return (&netisr_cpu[0].td_msgport);
 239
 240         switch (ip->ip_p) {
 241         case IPPROTO_TCP:
 242                 th = (struct tcphdr *)((caddr_t)ip + iphlen);
 243                 thoff = th->th_off << 2;
 244                 cpu = INP_MPORT_HASH(ip->ip_src.s_addr, ip->ip_dst.s_addr,
 245                     th->th_sport, th->th_dport);
 246                 port = &tcp_thread[cpu].td_msgport;
 247                 break;
 248         case IPPROTO_UDP:
 249                 uh = (struct udphdr *)((caddr_t)ip + iphlen);
 250
 251                 if (IN_MULTICAST(ntohl(ip->ip_dst.s_addr)) ||
 252                     (dir == IP_MPORT_IN &&
 253                      in_broadcast(ip->ip_dst, m->m_pkthdr.rcvif))) {
 254                         cpu = 0;
 255                 } else {
 256                         cpu = INP_MPORT_HASH(ip->ip_src.s_addr,
 257                             ip->ip_dst.s_addr, uh->uh_sport, uh->uh_dport);
 258                 }
 259                 port = &udp_thread[cpu].td_msgport;
 260                 break;
 261         default:
 262                 port = &netisr_cpu[0].td_msgport;
 263                 break;
 264         }
 265
 266         return (port);
 267 }
 268
 269 lwkt_port_t
 270 ip_mport_in(struct mbuf **mptr)
 271 {
 272         return ip_mport(mptr, IP_MPORT_IN);
 273 }
 274
 275 /*
 276  * Map a TCP socket to a protocol processing thread.
 277  */
 278 lwkt_port_t
 279 tcp_soport(struct socket *so, struct sockaddr *nam __unused,
 280            struct mbuf **dummy __unused, int req)
 281 {
 282         struct inpcb *inp;
 283
 284         /* The following processing all take place on Protocol Thread 0. */
 285         if (req == PRU_BIND || req == PRU_CONNECT || req == PRU_ATTACH ||
 286             req == PRU_LISTEN)
 287                 return (&tcp_thread[0].td_msgport);
 288
 289         inp = so->so_pcb;
 290         if (!inp)               /* connection reset by peer */
 291                 return (&tcp_thread[0].td_msgport);
 292
 293         /*
 294          * Already bound and connected or listening.  For TCP connections,
 295          * the (faddr, fport, laddr, lport) association cannot change now.
 296          *
 297          * Note: T/TCP code needs some reorganization to fit into
 298          * this model.  XXX JH
 299          *
 300          * Rely on type-stable memory and check in protocol handler
 301          * to fix race condition here w/ deallocation of inp.  XXX JH
 302          */
 303         return (&tcp_thread[INP_MPORT_HASH(inp->inp_faddr.s_addr,
 304             inp->inp_laddr.s_addr, inp->inp_fport, inp->inp_lport)].td_msgport);
 305 }
 306
 307 lwkt_port_t
 308 tcp_addrport(in_addr_t faddr, in_port_t fport, in_addr_t laddr, in_port_t lport)
 309 {
 310         return (&tcp_thread[tcp_addrcpu(faddr, fport,
 311                                         laddr, lport)].td_msgport);
 312 }
 313
 314 /*
 315  * Map a UDP socket to a protocol processing thread.
 316  */
 317 lwkt_port_t
 318 udp_soport(struct socket *so, struct sockaddr *nam __unused,
 319            struct mbuf **dummy __unused, int req)
 320 {
 321         struct inpcb *inp;
 322
 323         /*
 324          * The following processing all take place on Protocol Thread 0:
 325          *   bind()
 326          *   attach() has a null socket parameter
 327          *   Fast and slow timeouts pass in null socket parameter
 328          */
 329         if (req == PRU_BIND || so == NULL)
 330                 return (&udp_thread[0].td_msgport);
 331
 332         inp = so->so_pcb;
 333
 334         if (IN_MULTICAST(ntohl(inp->inp_laddr.s_addr)))
 335                 return (&udp_thread[0].td_msgport);
 336
 337         /*
 338          * Rely on type-stable memory and check in protocol handler
 339          * to fix race condition here w/ deallocation of inp.  XXX JH
 340          */
 341
 342         return (&udp_thread[INP_MPORT_HASH(inp->inp_faddr.s_addr,
 343             inp->inp_laddr.s_addr, inp->inp_fport, inp->inp_lport)].td_msgport);
 344 }
 345
 346 /*
 347  * Map a network address to a processor.
 348  */
 349 int
 350 tcp_addrcpu(in_addr_t faddr, in_port_t fport, in_addr_t laddr, in_port_t lport)
 351 {
 352         return (INP_MPORT_HASH(faddr, laddr, fport, lport));
 353 }
 354
 355 int
 356 udp_addrcpu(in_addr_t faddr, in_port_t fport, in_addr_t laddr, in_port_t lport)
 357 {
 358         if (IN_MULTICAST(ntohl(laddr)))
 359                 return (0);
 360         else
 361                 return (INP_MPORT_HASH(faddr, laddr, fport, lport));
 362 }
 363
 364 /*
 365  * Return LWKT port for cpu.
 366  */
 367 lwkt_port_t
 368 tcp_cport(int cpu)
 369 {
 370         return (&tcp_thread[cpu].td_msgport);
 371 }
 372
 373 void
 374 tcp_thread_init(void)
 375 {
 376         int cpu;
 377
 378         for (cpu = 0; cpu < ncpus2; cpu++) {
 379                 lwkt_create(tcpmsg_service_loop, NULL, NULL,
 380                         &tcp_thread[cpu], 0, cpu, "tcp_thread %d", cpu);
 381                 netmsg_service_port_init(&tcp_thread[cpu].td_msgport);
 382         }
 383 }
 384
 385 void
 386 udp_thread_init(void)
 387 {
 388         int cpu;
 389
 390         for (cpu = 0; cpu < ncpus2; cpu++) {
 391                 lwkt_create(udp_mpsafe_thread ?
 392                             netmsg_service_loop_mpsafe : netmsg_service_loop,
 393                             NULL, NULL, &udp_thread[cpu], 0, cpu,
 394                             "udp_thread %d", cpu);
 395                 netmsg_service_port_init(&udp_thread[cpu].td_msgport);
 396         }
 397 }