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
  34 #include "opt_inet.h"
  35 #include "opt_rss.h"
  36
  37 #include <sys/param.h>
  38 #include <sys/systm.h>
  39 #include <sys/kernel.h>
  40 #include <sys/socket.h>
  41 #include <sys/socketvar.h>
  42 #include <sys/thread.h>
  43 #include <sys/sysctl.h>
  44 #include <sys/globaldata.h>
  45
  46 #include <net/if.h>
  47 #include <net/netisr2.h>
  48 #include <net/toeplitz2.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 /*
  63  * Toeplitz hash functions - the idea is to match the hardware.
  64  */
  65 static __inline int
  66 INP_MPORT_HASH_UDP(in_addr_t faddr, in_addr_t laddr,
  67                    in_port_t fport, in_port_t lport)
  68 {
  69         return toeplitz_hash(toeplitz_rawhash_addr(faddr, laddr));
  70 }
  71
  72 static __inline int
  73 INP_MPORT_HASH_TCP(in_addr_t faddr, in_addr_t laddr,
  74                    in_port_t fport, in_port_t lport)
  75 {
  76         return toeplitz_hash(
  77                toeplitz_rawhash_addrport(faddr, laddr, fport, lport));
  78 }
  79
  80 /*
  81  * Map a network address to a processor.
  82  */
  83 int
  84 tcp_addrcpu(in_addr_t faddr, in_port_t fport, in_addr_t laddr, in_port_t lport)
  85 {
  86         return (netisr_hashcpu(INP_MPORT_HASH_TCP(faddr, laddr, fport, lport)));
  87 }
  88
  89 int
  90 udp_addrcpu(in_addr_t faddr, in_port_t fport, in_addr_t laddr, in_port_t lport)
  91 {
  92         /*
  93          * NOTE: laddr could be multicast, since UDP socket could be
  94          * bound to multicast address.
  95          */
  96         if (IN_MULTICAST(ntohl(faddr)) || IN_MULTICAST(ntohl(laddr))) {
  97                 /* XXX handle multicast on CPU0 for now */
  98                 return 0;
  99         }
 100         return (netisr_hashcpu(INP_MPORT_HASH_UDP(faddr, laddr, fport, lport)));
 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                         ++udp_stat.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                                 udp_stat.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_hashfn(struct mbuf **mptr, int hoff)
 267 {
 268         struct ip *ip;
 269         int iphlen;
 270         struct tcphdr *th;
 271         struct udphdr *uh;
 272         struct mbuf *m;
 273         int hash;
 274
 275         if (!ip_lengthcheck(mptr, hoff))
 276                 return;
 277
 278         m = *mptr;
 279         ip = mtodoff(m, struct ip *, hoff);
 280         iphlen = ip->ip_hl << 2;
 281
 282         if (ntohs(ip->ip_off) & (IP_MF | IP_OFFMASK)) {
 283                 hash = toeplitz_hash(toeplitz_rawhash_addr(
 284                     ip->ip_src.s_addr, ip->ip_dst.s_addr));
 285                 goto back;
 286         }
 287
 288         switch (ip->ip_p) {
 289         case IPPROTO_TCP:
 290                 th = (struct tcphdr *)((caddr_t)ip + iphlen);
 291                 hash = INP_MPORT_HASH_TCP(ip->ip_src.s_addr, ip->ip_dst.s_addr,
 292                     th->th_sport, th->th_dport);
 293                 break;
 294
 295         case IPPROTO_UDP:
 296                 if (IN_MULTICAST(ntohl(ip->ip_dst.s_addr))) {
 297                         /* XXX handle multicast on CPU0 for now */
 298                         hash = 0;
 299                         break;
 300                 }
 301                 uh = (struct udphdr *)((caddr_t)ip + iphlen);
 302                 hash = INP_MPORT_HASH_UDP(ip->ip_src.s_addr, ip->ip_dst.s_addr,
 303                     uh->uh_sport, uh->uh_dport);
 304                 break;
 305
 306         default:
 307                 hash = 0;
 308                 break;
 309         }
 310 back:
 311         m->m_flags |= M_HASH;
 312         m->m_pkthdr.hash = hash;
 313 }
 314
 315 /*
 316  * Verify and adjust the hash value of the packet.
 317  *
 318  * Unlike ip_hashfn(), the packet content is not accessed.  The packet info
 319  * (pi) and the hash of the packet (m_pkthdr.hash) is used instead.
 320  *
 321  * Caller has already made sure that m_pkthdr.hash is valid, i.e. m_flags
 322  * has M_HASH set.
 323  */
 324 void
 325 ip_hashcheck(struct mbuf *m, const struct pktinfo *pi)
 326 {
 327         KASSERT((m->m_flags & M_HASH), ("no valid packet hash"));
 328
 329         switch (pi->pi_l3proto) {
 330         case IPPROTO_TCP:
 331         case IPPROTO_UDP:
 332                 break;
 333
 334         default:
 335                 /* Let software calculate the hash */
 336                 m->m_flags &= ~M_HASH;
 337                 break;
 338         }
 339 }
 340
 341 /*
 342  * This is used to map a socket to a message port for sendmsg() and friends.
 343  * It is not called for any other purpose.  In the case of TCP we just return
 344  * the port already installed in the socket.
 345  */
 346 lwkt_port_t
 347 tcp_soport(struct socket *so, struct sockaddr *nam,
 348            struct mbuf **dummy __unused)
 349 {
 350         return(so->so_port);
 351 }
 352
 353 /*
 354  * Used to route icmp messages to the proper protocol thread for ctlinput
 355  * operation.
 356  */
 357 lwkt_port_t
 358 tcp_ctlport(int cmd, struct sockaddr *sa, void *vip, int *cpuid)
 359 {
 360         struct ip *ip = vip;
 361         inp_notify_t notify;
 362         int arg;
 363
 364         notify = tcp_get_inpnotify(cmd, sa, &arg, &ip, cpuid);
 365         if (notify == NULL)
 366                 return NULL;
 367
 368         if (*cpuid == ncpus) {
 369                 /*
 370                  * Go through all CPUs.
 371                  *
 372                  * A new message will be allocated later to save necessary
 373                  * information and will be forwarded to all network protocol
 374                  * threads in the following way:
 375                  *
 376                  * (the the thread owns the msgport that we return here)
 377                  * netisr0 <--+
 378                  *    |       |
 379                  *    |       |
 380                  *    |       |
 381                  *    +-------+
 382                  *     sendmsg
 383                  *     [msg is kmalloc()ed]
 384                  *
 385                  *
 386                  * Later on, when the msg is received by netisr0:
 387                  *
 388                  *         forwardmsg         forwardmsg
 389                  * netisr0 ---------> netisr1 ---------> netisrN
 390                  *                                       [msg is kfree()ed]
 391                  */
 392                 return netisr_cpuport(0);
 393         } else {
 394                 return netisr_cpuport(*cpuid);
 395         }
 396 }
 397
 398 lwkt_port_t
 399 tcp_addrport(in_addr_t faddr, in_port_t fport, in_addr_t laddr, in_port_t lport)
 400 {
 401         return(netisr_cpuport(tcp_addrcpu(faddr, fport, laddr, lport)));
 402 }
 403
 404 lwkt_port_t
 405 tcp_addrport0(void)
 406 {
 407         return(netisr_cpuport(0));
 408 }
 409
 410 lwkt_port_t
 411 udp_addrport(in_addr_t faddr, in_port_t fport, in_addr_t laddr, in_port_t lport)
 412 {
 413         return(netisr_cpuport(udp_addrcpu(faddr, fport, laddr, lport)));
 414 }
 415
 416 /*
 417  * Used to route icmp messages to the proper protocol thread for ctlinput
 418  * operation.
 419  */
 420 lwkt_port_t
 421 udp_ctlport(int cmd, struct sockaddr *sa, void *vip, int *cpuid)
 422 {
 423         struct ip *ip = vip;
 424         inp_notify_t notify;
 425
 426         notify = udp_get_inpnotify(cmd, sa, &ip, cpuid);
 427         if (notify == NULL)
 428                 return NULL;
 429
 430         if (*cpuid == ncpus) {
 431                 /*
 432                  * Go through all CPUs.
 433                  *
 434                  * See the comment in tcp_ctlport.
 435                  */
 436                 return netisr_cpuport(0);
 437         } else {
 438                 return netisr_cpuport(*cpuid);
 439         }
 440 }
 441
 442 struct lwkt_port *
 443 tcp_initport(void)
 444 {
 445         return netisr_cpuport(mycpuid & ncpus2_mask);
 446 }
 447
 448 struct lwkt_port *
 449 udp_initport(void)
 450 {
 451         return netisr_cpuport(mycpuid & ncpus2_mask);
 452 }