1 /*
   2  * Copyright (c) 1997, 2019, Oracle and/or its affiliates. All rights reserved.
   3  * DO NOT ALTER OR REMOVE COPYRIGHT NOTICES OR THIS FILE HEADER.
   4  *
   5  * This code is free software; you can redistribute it and/or modify it
   6  * under the terms of the GNU General Public License version 2 only, as
   7  * published by the Free Software Foundation.  Oracle designates this
   8  * particular file as subject to the "Classpath" exception as provided
   9  * by Oracle in the LICENSE file that accompanied this code.
  10  *
  11  * This code is distributed in the hope that it will be useful, but WITHOUT
  12  * ANY WARRANTY; without even the implied warranty of MERCHANTABILITY or
  13  * FITNESS FOR A PARTICULAR PURPOSE.  See the GNU General Public License
  14  * version 2 for more details (a copy is included in the LICENSE file that
  15  * accompanied this code).
  16  *
  17  * You should have received a copy of the GNU General Public License version
  18  * 2 along with this work; if not, write to the Free Software Foundation,
  19  * Inc., 51 Franklin St, Fifth Floor, Boston, MA 02110-1301 USA.
  20  *
  21  * Please contact Oracle, 500 Oracle Parkway, Redwood Shores, CA 94065 USA
  22  * or visit www.oracle.com if you need additional information or have any
  23  * questions.
  24  */
  25 #include <dlfcn.h>
  26 #include <errno.h>
  27 #include <net/if.h>
  28 #include <netinet/tcp.h> // defines TCP_NODELAY
  29 #include <stdlib.h>
  30 #include <string.h>
  31 #include <sys/ioctl.h>
  32 #include <sys/time.h>
  33 
  34 #if defined(__linux__)
  35 #include <arpa/inet.h>
  36 #include <net/route.h>
  37 #include <sys/utsname.h>
  38 #endif
  39 
  40 #if defined(__solaris__)
  41 #include <inet/nd.h>
  42 #include <limits.h>
  43 #include <stropts.h>
  44 #include <sys/filio.h>
  45 #include <sys/sockio.h>
  46 #endif
  47 
  48 #if defined(MACOSX)
  49 #include <sys/sysctl.h>
  50 #endif
  51 
  52 #include "jvm.h"
  53 #include "net_util.h"
  54 
  55 #include "java_net_SocketOptions.h"
  56 #include "java_net_InetAddress.h"
  57 
  58 #if defined(__linux__) && !defined(IPV6_FLOWINFO_SEND)
  59 #define IPV6_FLOWINFO_SEND      33
  60 #endif
  61 
  62 #if defined(__solaris__) && !defined(MAXINT)
  63 #define MAXINT INT_MAX
  64 #endif
  65 
  66 /*
  67  * EXCLBIND socket options only on Solaris
  68  */
  69 #if defined(__solaris__) && !defined(TCP_EXCLBIND)
  70 #define TCP_EXCLBIND            0x21
  71 #endif
  72 #if defined(__solaris__) && !defined(UDP_EXCLBIND)
  73 #define UDP_EXCLBIND            0x0101
  74 #endif
  75 
  76 void setDefaultScopeID(JNIEnv *env, struct sockaddr *him)
  77 {
  78 #ifdef MACOSX
  79     static jclass ni_class = NULL;
  80     static jfieldID ni_defaultIndexID;
  81     if (ni_class == NULL) {
  82         jclass c = (*env)->FindClass(env, "java/net/NetworkInterface");
  83         CHECK_NULL(c);
  84         c = (*env)->NewGlobalRef(env, c);
  85         CHECK_NULL(c);
  86         ni_defaultIndexID = (*env)->GetStaticFieldID(env, c, "defaultIndex", "I");
  87         CHECK_NULL(ni_defaultIndexID);
  88         ni_class = c;
  89     }
  90     int defaultIndex;
  91     struct sockaddr_in6 *sin6 = (struct sockaddr_in6 *)him;
  92     if (sin6->sin6_family == AF_INET6 && (sin6->sin6_scope_id == 0) &&
  93         (IN6_IS_ADDR_LINKLOCAL(&sin6->sin6_addr) ||
  94          IN6_IS_ADDR_MULTICAST(&sin6->sin6_addr))) {
  95         defaultIndex = (*env)->GetStaticIntField(env, ni_class,
  96                                                  ni_defaultIndexID);
  97         sin6->sin6_scope_id = defaultIndex;
  98     }
  99 #endif
 100 }
 101 
 102 int getDefaultScopeID(JNIEnv *env) {
 103     int defaultIndex = 0;
 104     static jclass ni_class = NULL;
 105     static jfieldID ni_defaultIndexID;
 106     if (ni_class == NULL) {
 107         jclass c = (*env)->FindClass(env, "java/net/NetworkInterface");
 108         CHECK_NULL_RETURN(c, 0);
 109         c = (*env)->NewGlobalRef(env, c);
 110         CHECK_NULL_RETURN(c, 0);
 111         ni_defaultIndexID = (*env)->GetStaticFieldID(env, c, "defaultIndex", "I");
 112         CHECK_NULL_RETURN(ni_defaultIndexID, 0);
 113         ni_class = c;
 114     }
 115     defaultIndex = (*env)->GetStaticIntField(env, ni_class,
 116                                              ni_defaultIndexID);
 117     return defaultIndex;
 118 }
 119 
 120 #define RESTARTABLE(_cmd, _result) do { \
 121     do { \
 122         _result = _cmd; \
 123     } while((_result == -1) && (errno == EINTR)); \
 124 } while(0)
 125 
 126 int NET_SocketAvailable(int s, int *pbytes) {
 127     int result;
 128     RESTARTABLE(ioctl(s, FIONREAD, pbytes), result);
 129     return result;
 130 }
 131 
 132 #ifdef __solaris__
 133 static int init_tcp_max_buf, init_udp_max_buf;
 134 static int tcp_max_buf;
 135 static int udp_max_buf;
 136 static int useExclBind = 0;
 137 
 138 /*
 139  * Get the specified parameter from the specified driver. The value
 140  * of the parameter is assumed to be an 'int'. If the parameter
 141  * cannot be obtained return -1
 142  */
 143 int net_getParam(char *driver, char *param)
 144 {
 145     struct strioctl stri;
 146     char buf [64];
 147     int s;
 148     int value;
 149 
 150     s = open (driver, O_RDWR);
 151     if (s < 0) {
 152         return -1;
 153     }
 154     strncpy (buf, param, sizeof(buf));
 155     stri.ic_cmd = ND_GET;
 156     stri.ic_timout = 0;
 157     stri.ic_dp = buf;
 158     stri.ic_len = sizeof(buf);
 159     if (ioctl (s, I_STR, &stri) < 0) {
 160         value = -1;
 161     } else {
 162         value = atoi(buf);
 163     }
 164     close (s);
 165     return value;
 166 }
 167 
 168 /*
 169  * Iterative way to find the max value that SO_SNDBUF or SO_RCVBUF
 170  * for Solaris versions that do not support the ioctl() in net_getParam().
 171  * Ugly, but only called once (for each sotype).
 172  *
 173  * As an optimization, we make a guess using the default values for Solaris
 174  * assuming they haven't been modified with ndd.
 175  */
 176 
 177 #define MAX_TCP_GUESS 1024 * 1024
 178 #define MAX_UDP_GUESS 2 * 1024 * 1024
 179 
 180 #define FAIL_IF_NOT_ENOBUFS if (errno != ENOBUFS) return -1
 181 
 182 static int findMaxBuf(int fd, int opt, int sotype) {
 183     int a = 0;
 184     int b = MAXINT;
 185     int initial_guess;
 186     int limit = -1;
 187 
 188     if (sotype == SOCK_DGRAM) {
 189         initial_guess = MAX_UDP_GUESS;
 190     } else {
 191         initial_guess = MAX_TCP_GUESS;
 192     }
 193 
 194     if (setsockopt(fd, SOL_SOCKET, opt, &initial_guess, sizeof(int)) == 0) {
 195         initial_guess++;
 196         if (setsockopt(fd, SOL_SOCKET, opt, &initial_guess,sizeof(int)) < 0) {
 197             FAIL_IF_NOT_ENOBUFS;
 198             return initial_guess - 1;
 199         }
 200         a = initial_guess;
 201     } else {
 202         FAIL_IF_NOT_ENOBUFS;
 203         b = initial_guess - 1;
 204     }
 205     do {
 206         int mid = a + (b-a)/2;
 207         if (setsockopt(fd, SOL_SOCKET, opt, &mid, sizeof(int)) == 0) {
 208             limit = mid;
 209             a = mid + 1;
 210         } else {
 211             FAIL_IF_NOT_ENOBUFS;
 212             b = mid - 1;
 213         }
 214     } while (b >= a);
 215 
 216     return limit;
 217 }
 218 #endif
 219 
 220 #ifdef __linux__
 221 static int vinit = 0;
 222 static int kernelV24 = 0;
 223 static int vinit24 = 0;
 224 
 225 int kernelIsV24 () {
 226     if (!vinit24) {
 227         struct utsname sysinfo;
 228         if (uname(&sysinfo) == 0) {
 229             sysinfo.release[3] = '\0';
 230             if (strcmp(sysinfo.release, "2.4") == 0) {
 231                 kernelV24 = JNI_TRUE;
 232             }
 233         }
 234         vinit24 = 1;
 235     }
 236     return kernelV24;
 237 }
 238 #endif
 239 
 240 void
 241 NET_ThrowByNameWithLastError(JNIEnv *env, const char *name,
 242                    const char *defaultDetail) {
 243     JNU_ThrowByNameWithMessageAndLastError(env, name, defaultDetail);
 244 }
 245 
 246 void
 247 NET_ThrowCurrent(JNIEnv *env, char *msg) {
 248     NET_ThrowNew(env, errno, msg);
 249 }
 250 
 251 void
 252 NET_ThrowNew(JNIEnv *env, int errorNumber, char *msg) {
 253     char fullMsg[512];
 254     if (!msg) {
 255         msg = "no further information";
 256     }
 257     switch(errorNumber) {
 258     case EBADF:
 259         jio_snprintf(fullMsg, sizeof(fullMsg), "socket closed: %s", msg);
 260         JNU_ThrowByName(env, JNU_JAVANETPKG "SocketException", fullMsg);
 261         break;
 262     case EINTR:
 263         JNU_ThrowByName(env, JNU_JAVAIOPKG "InterruptedIOException", msg);
 264         break;
 265     default:
 266         errno = errorNumber;
 267         JNU_ThrowByNameWithLastError(env, JNU_JAVANETPKG "SocketException", msg);
 268         break;
 269     }
 270 }
 271 
 272 
 273 jfieldID
 274 NET_GetFileDescriptorID(JNIEnv *env)
 275 {
 276     jclass cls = (*env)->FindClass(env, "java/io/FileDescriptor");
 277     CHECK_NULL_RETURN(cls, NULL);
 278     return (*env)->GetFieldID(env, cls, "fd", "I");
 279 }
 280 
 281 jint  IPv4_supported()
 282 {
 283     int fd = socket(AF_INET, SOCK_STREAM, 0) ;
 284     if (fd < 0) {
 285         return JNI_FALSE;
 286     }
 287     close(fd);
 288     return JNI_TRUE;
 289 }
 290 
 291 #if defined(DONT_ENABLE_IPV6)
 292 jint  IPv6_supported()
 293 {
 294     return JNI_FALSE;
 295 }
 296 
 297 #else /* !DONT_ENABLE_IPV6 */
 298 
 299 jint  IPv6_supported()
 300 {
 301     int fd;
 302     void *ipv6_fn;
 303     SOCKETADDRESS sa;
 304     socklen_t sa_len = sizeof(SOCKETADDRESS);
 305 
 306     fd = socket(AF_INET6, SOCK_STREAM, 0) ;
 307     if (fd < 0) {
 308         /*
 309          *  TODO: We really cant tell since it may be an unrelated error
 310          *  for now we will assume that AF_INET6 is not available
 311          */
 312         return JNI_FALSE;
 313     }
 314 
 315     /*
 316      * If fd 0 is a socket it means we may have been launched from inetd or
 317      * xinetd. If it's a socket then check the family - if it's an
 318      * IPv4 socket then we need to disable IPv6.
 319      */
 320     if (getsockname(0, &sa.sa, &sa_len) == 0) {
 321         if (sa.sa.sa_family == AF_INET) {
 322             close(fd);
 323             return JNI_FALSE;
 324         }
 325     }
 326 
 327     /**
 328      * Linux - check if any interface has an IPv6 address.
 329      * Don't need to parse the line - we just need an indication.
 330      */
 331 #ifdef __linux__
 332     {
 333         FILE *fP = fopen("/proc/net/if_inet6", "r");
 334         char buf[255];
 335         char *bufP;
 336 
 337         if (fP == NULL) {
 338             close(fd);
 339             return JNI_FALSE;
 340         }
 341         bufP = fgets(buf, sizeof(buf), fP);
 342         fclose(fP);
 343         if (bufP == NULL) {
 344             close(fd);
 345             return JNI_FALSE;
 346         }
 347     }
 348 #endif
 349 
 350     /**
 351      * On Solaris 8 it's possible to create INET6 sockets even
 352      * though IPv6 is not enabled on all interfaces. Thus we
 353      * query the number of IPv6 addresses to verify that IPv6
 354      * has been configured on at least one interface.
 355      *
 356      * On Linux it doesn't matter - if IPv6 is built-in the
 357      * kernel then IPv6 addresses will be bound automatically
 358      * to all interfaces.
 359      */
 360 #ifdef __solaris__
 361 
 362 #ifdef SIOCGLIFNUM
 363     {
 364         struct lifnum numifs;
 365 
 366         numifs.lifn_family = AF_INET6;
 367         numifs.lifn_flags = 0;
 368         if (ioctl(fd, SIOCGLIFNUM, (char *)&numifs) < 0) {
 369             /**
 370              * SIOCGLIFNUM failed - assume IPv6 not configured
 371              */
 372             close(fd);
 373             return JNI_FALSE;
 374         }
 375         /**
 376          * If no IPv6 addresses then return false. If count > 0
 377          * it's possible that all IPv6 addresses are "down" but
 378          * that's okay as they may be brought "up" while the
 379          * VM is running.
 380          */
 381         if (numifs.lifn_count == 0) {
 382             close(fd);
 383             return JNI_FALSE;
 384         }
 385     }
 386 #else
 387     /* SIOCGLIFNUM not defined in build environment ??? */
 388     close(fd);
 389     return JNI_FALSE;
 390 #endif
 391 
 392 #endif /* __solaris */
 393 
 394     /*
 395      *  OK we may have the stack available in the kernel,
 396      *  we should also check if the APIs are available.
 397      */
 398     ipv6_fn = JVM_FindLibraryEntry(RTLD_DEFAULT, "inet_pton");
 399     close(fd);
 400     if (ipv6_fn == NULL ) {
 401         return JNI_FALSE;
 402     } else {
 403         return JNI_TRUE;
 404     }
 405 }
 406 #endif /* DONT_ENABLE_IPV6 */
 407 
 408 jint reuseport_supported()
 409 {
 410     /* Do a simple dummy call, and try to figure out from that */
 411     int one = 1;
 412     int rv, s;
 413     s = socket(PF_INET, SOCK_STREAM, 0);
 414     if (s < 0) {
 415         return JNI_FALSE;
 416     }
 417     rv = setsockopt(s, SOL_SOCKET, SO_REUSEPORT, (void *)&one, sizeof(one));
 418     if (rv != 0) {
 419         rv = JNI_FALSE;
 420     } else {
 421         rv = JNI_TRUE;
 422     }
 423     close(s);
 424     return rv;
 425 }
 426 
 427 void NET_ThrowUnknownHostExceptionWithGaiError(JNIEnv *env,
 428                                                const char* hostname,
 429                                                int gai_error)
 430 {
 431     int size;
 432     char *buf;
 433     const char *format = "%s: %s";
 434     const char *error_string = gai_strerror(gai_error);
 435     if (error_string == NULL)
 436         error_string = "unknown error";
 437 
 438     size = strlen(format) + strlen(hostname) + strlen(error_string) + 2;
 439     buf = (char *) malloc(size);
 440     if (buf) {
 441         jstring s;
 442         sprintf(buf, format, hostname, error_string);
 443         s = JNU_NewStringPlatform(env, buf);
 444         if (s != NULL) {
 445             jobject x = JNU_NewObjectByName(env,
 446                                             "java/net/UnknownHostException",
 447                                             "(Ljava/lang/String;)V", s);
 448             if (x != NULL)
 449                 (*env)->Throw(env, x);
 450         }
 451         free(buf);
 452     }
 453 }
 454 
 455 #if defined(__linux__)
 456 
 457 /* following code creates a list of addresses from the kernel
 458  * routing table that are routed via the loopback address.
 459  * We check all destination addresses against this table
 460  * and override the scope_id field to use the relevant value for "lo"
 461  * in order to work-around the Linux bug that prevents packets destined
 462  * for certain local addresses from being sent via a physical interface.
 463  */
 464 
 465 struct loopback_route {
 466     struct in6_addr addr; /* destination address */
 467     int plen; /* prefix length */
 468 };
 469 
 470 static struct loopback_route *loRoutes = 0;
 471 static int nRoutes = 0; /* number of routes */
 472 static int loRoutes_size = 16; /* initial size */
 473 static int lo_scope_id = 0;
 474 
 475 static void initLoopbackRoutes();
 476 
 477 void printAddr (struct in6_addr *addr) {
 478     int i;
 479     for (i=0; i<16; i++) {
 480         printf ("%02x", addr->s6_addr[i]);
 481     }
 482     printf ("\n");
 483 }
 484 
 485 static jboolean needsLoopbackRoute (struct in6_addr* dest_addr) {
 486     int byte_count;
 487     int extra_bits, i;
 488     struct loopback_route *ptr;
 489 
 490     if (loRoutes == 0) {
 491         initLoopbackRoutes();
 492     }
 493 
 494     for (ptr = loRoutes, i=0; i<nRoutes; i++, ptr++) {
 495         struct in6_addr *target_addr=&ptr->addr;
 496         int dest_plen = ptr->plen;
 497         byte_count = dest_plen >> 3;
 498         extra_bits = dest_plen & 0x3;
 499 
 500         if (byte_count > 0) {
 501             if (memcmp(target_addr, dest_addr, byte_count)) {
 502                 continue;  /* no match */
 503             }
 504         }
 505 
 506         if (extra_bits > 0) {
 507             unsigned char c1 = ((unsigned char *)target_addr)[byte_count];
 508             unsigned char c2 = ((unsigned char *)&dest_addr)[byte_count];
 509             unsigned char mask = 0xff << (8 - extra_bits);
 510             if ((c1 & mask) != (c2 & mask)) {
 511                 continue;
 512             }
 513         }
 514         return JNI_TRUE;
 515     }
 516     return JNI_FALSE;
 517 }
 518 
 519 
 520 static void initLoopbackRoutes() {
 521     FILE *f;
 522     char srcp[8][5];
 523     char hopp[8][5];
 524     int dest_plen, src_plen, use, refcnt, metric;
 525     unsigned long flags;
 526     char dest_str[40];
 527     struct in6_addr dest_addr;
 528     char device[16];
 529     struct loopback_route *loRoutesTemp;
 530 
 531     if (loRoutes != 0) {
 532         free (loRoutes);
 533     }
 534     loRoutes = calloc (loRoutes_size, sizeof(struct loopback_route));
 535     if (loRoutes == 0) {
 536         return;
 537     }
 538     /*
 539      * Scan /proc/net/ipv6_route looking for a matching
 540      * route.
 541      */
 542     if ((f = fopen("/proc/net/ipv6_route", "r")) == NULL) {
 543         return ;
 544     }
 545     while (fscanf(f, "%4s%4s%4s%4s%4s%4s%4s%4s %02x "
 546                      "%4s%4s%4s%4s%4s%4s%4s%4s %02x "
 547                      "%4s%4s%4s%4s%4s%4s%4s%4s "
 548                      "%08x %08x %08x %08lx %8s",
 549                      dest_str, &dest_str[5], &dest_str[10], &dest_str[15],
 550                      &dest_str[20], &dest_str[25], &dest_str[30], &dest_str[35],
 551                      &dest_plen,
 552                      srcp[0], srcp[1], srcp[2], srcp[3],
 553                      srcp[4], srcp[5], srcp[6], srcp[7],
 554                      &src_plen,
 555                      hopp[0], hopp[1], hopp[2], hopp[3],
 556                      hopp[4], hopp[5], hopp[6], hopp[7],
 557                      &metric, &use, &refcnt, &flags, device) == 31) {
 558 
 559         /*
 560          * Some routes should be ignored
 561          */
 562         if ( (dest_plen < 0 || dest_plen > 128)  ||
 563              (src_plen != 0) ||
 564              (flags & (RTF_POLICY | RTF_FLOW)) ||
 565              ((flags & RTF_REJECT) && dest_plen == 0) ) {
 566             continue;
 567         }
 568 
 569         /*
 570          * Convert the destination address
 571          */
 572         dest_str[4] = ':';
 573         dest_str[9] = ':';
 574         dest_str[14] = ':';
 575         dest_str[19] = ':';
 576         dest_str[24] = ':';
 577         dest_str[29] = ':';
 578         dest_str[34] = ':';
 579         dest_str[39] = '\0';
 580 
 581         if (inet_pton(AF_INET6, dest_str, &dest_addr) < 0) {
 582             /* not an Ipv6 address */
 583             continue;
 584         }
 585         if (strcmp(device, "lo") != 0) {
 586             /* Not a loopback route */
 587             continue;
 588         } else {
 589             if (nRoutes == loRoutes_size) {
 590                 loRoutesTemp = realloc (loRoutes, loRoutes_size *
 591                                         sizeof (struct loopback_route) * 2);
 592 
 593                 if (loRoutesTemp == 0) {
 594                     free(loRoutes);
 595                     loRoutes = NULL;
 596                     nRoutes = 0;
 597                     fclose (f);
 598                     return;
 599                 }
 600                 loRoutes=loRoutesTemp;
 601                 loRoutes_size *= 2;
 602             }
 603             memcpy (&loRoutes[nRoutes].addr,&dest_addr,sizeof(struct in6_addr));
 604             loRoutes[nRoutes].plen = dest_plen;
 605             nRoutes ++;
 606         }
 607     }
 608 
 609     fclose (f);
 610     {
 611         /* now find the scope_id for "lo" */
 612 
 613         char devname[21];
 614         char addr6p[8][5];
 615         int plen, scope, dad_status, if_idx;
 616 
 617         if ((f = fopen("/proc/net/if_inet6", "r")) != NULL) {
 618             while (fscanf(f, "%4s%4s%4s%4s%4s%4s%4s%4s %08x %02x %02x %02x %20s\n",
 619                       addr6p[0], addr6p[1], addr6p[2], addr6p[3],
 620                       addr6p[4], addr6p[5], addr6p[6], addr6p[7],
 621                   &if_idx, &plen, &scope, &dad_status, devname) == 13) {
 622 
 623                 if (strcmp(devname, "lo") == 0) {
 624                     /*
 625                      * Found - so just return the index
 626                      */
 627                     fclose(f);
 628                     lo_scope_id = if_idx;
 629                     return;
 630                 }
 631             }
 632             fclose(f);
 633         }
 634     }
 635 }
 636 
 637 /*
 638  * Following is used for binding to local addresses. Equivalent
 639  * to code above, for bind().
 640  */
 641 
 642 struct localinterface {
 643     int index;
 644     char localaddr [16];
 645 };
 646 
 647 static struct localinterface *localifs = 0;
 648 static int localifsSize = 0;    /* size of array */
 649 static int nifs = 0;            /* number of entries used in array */
 650 
 651 /* not thread safe: make sure called once from one thread */
 652 
 653 static void initLocalIfs () {
 654     FILE *f;
 655     unsigned char staddr [16];
 656     char ifname [33];
 657     struct localinterface *lif=0;
 658     struct localinterface *localifsTemp;
 659     int index, x1, x2, x3;
 660     unsigned int u0,u1,u2,u3,u4,u5,u6,u7,u8,u9,ua,ub,uc,ud,ue,uf;
 661 
 662     if ((f = fopen("/proc/net/if_inet6", "r")) == NULL) {
 663         return ;
 664     }
 665     while (fscanf (f, "%2x%2x%2x%2x%2x%2x%2x%2x%2x%2x%2x%2x%2x%2x%2x%2x "
 666                 "%d %x %x %x %32s",&u0,&u1,&u2,&u3,&u4,&u5,&u6,&u7,
 667                 &u8,&u9,&ua,&ub,&uc,&ud,&ue,&uf,
 668                 &index, &x1, &x2, &x3, ifname) == 21) {
 669         staddr[0] = (unsigned char)u0;
 670         staddr[1] = (unsigned char)u1;
 671         staddr[2] = (unsigned char)u2;
 672         staddr[3] = (unsigned char)u3;
 673         staddr[4] = (unsigned char)u4;
 674         staddr[5] = (unsigned char)u5;
 675         staddr[6] = (unsigned char)u6;
 676         staddr[7] = (unsigned char)u7;
 677         staddr[8] = (unsigned char)u8;
 678         staddr[9] = (unsigned char)u9;
 679         staddr[10] = (unsigned char)ua;
 680         staddr[11] = (unsigned char)ub;
 681         staddr[12] = (unsigned char)uc;
 682         staddr[13] = (unsigned char)ud;
 683         staddr[14] = (unsigned char)ue;
 684         staddr[15] = (unsigned char)uf;
 685         nifs ++;
 686         if (nifs > localifsSize) {
 687             localifsTemp = (struct localinterface *) realloc(
 688                         localifs, sizeof (struct localinterface)* (localifsSize+5));
 689             if (localifsTemp == 0) {
 690                 free(localifs);
 691                 localifs = 0;
 692                 localifsSize = 0;
 693                 nifs = 0;
 694                 fclose(f);
 695                 return;
 696             }
 697             localifs = localifsTemp;
 698             lif = localifs + localifsSize;
 699             localifsSize += 5;
 700         } else {
 701             lif ++;
 702         }
 703         memcpy (lif->localaddr, staddr, 16);
 704         lif->index = index;
 705     }
 706     fclose (f);
 707 }
 708 
 709 /* return the scope_id (interface index) of the
 710  * interface corresponding to the given address
 711  * returns 0 if no match found
 712  */
 713 
 714 static int getLocalScopeID (char *addr) {
 715     struct localinterface *lif;
 716     int i;
 717     if (localifs == 0) {
 718         initLocalIfs();
 719     }
 720     for (i=0, lif=localifs; i<nifs; i++, lif++) {
 721         if (memcmp (addr, lif->localaddr, 16) == 0) {
 722             return lif->index;
 723         }
 724     }
 725     return 0;
 726 }
 727 
 728 void platformInit () {
 729     initLoopbackRoutes();
 730     initLocalIfs();
 731 }
 732 
 733 #elif defined(_AIX)
 734 
 735 /* Initialize stubs for blocking I/O workarounds (see src/solaris/native/java/net/linux_close.c) */
 736 extern void aix_close_init();
 737 
 738 void platformInit () {
 739     aix_close_init();
 740 }
 741 
 742 #else
 743 
 744 void platformInit () {}
 745 
 746 #endif
 747 
 748 void parseExclusiveBindProperty(JNIEnv *env) {
 749 #ifdef __solaris__
 750     jstring s, flagSet;
 751     jclass iCls;
 752     jmethodID mid;
 753 
 754     s = (*env)->NewStringUTF(env, "sun.net.useExclusiveBind");
 755     CHECK_NULL(s);
 756     iCls = (*env)->FindClass(env, "java/lang/System");
 757     CHECK_NULL(iCls);
 758     mid = (*env)->GetStaticMethodID(env, iCls, "getProperty",
 759                 "(Ljava/lang/String;)Ljava/lang/String;");
 760     CHECK_NULL(mid);
 761     flagSet = (*env)->CallStaticObjectMethod(env, iCls, mid, s);
 762     if (flagSet != NULL) {
 763         useExclBind = 1;
 764     }
 765 #endif
 766 }
 767 
 768 JNIEXPORT jint JNICALL
 769 NET_EnableFastTcpLoopback(int fd) {
 770     return 0;
 771 }
 772 
 773 /**
 774  * See net_util.h for documentation
 775  */
 776 JNIEXPORT int JNICALL
 777 NET_InetAddressToSockaddr(JNIEnv *env, jobject iaObj, int port,
 778                           SOCKETADDRESS *sa, int *len,
 779                           jboolean v4MappedAddress)
 780 {
 781     jint family = getInetAddress_family(env, iaObj);
 782     JNU_CHECK_EXCEPTION_RETURN(env, -1);
 783     memset((char *)sa, 0, sizeof(SOCKETADDRESS));
 784 
 785     if (ipv6_available() &&
 786         !(family == java_net_InetAddress_IPv4 &&
 787           v4MappedAddress == JNI_FALSE))
 788     {
 789         jbyte caddr[16];
 790         jint address;
 791 
 792         if (family == java_net_InetAddress_IPv4) {
 793             // convert to IPv4-mapped address
 794             memset((char *)caddr, 0, 16);
 795             address = getInetAddress_addr(env, iaObj);
 796             JNU_CHECK_EXCEPTION_RETURN(env, -1);
 797             if (address == INADDR_ANY) {
 798                 /* we would always prefer IPv6 wildcard address
 799                  * caddr[10] = 0xff;
 800                  * caddr[11] = 0xff; */
 801             } else {
 802                 caddr[10] = 0xff;
 803                 caddr[11] = 0xff;
 804                 caddr[12] = ((address >> 24) & 0xff);
 805                 caddr[13] = ((address >> 16) & 0xff);
 806                 caddr[14] = ((address >> 8) & 0xff);
 807                 caddr[15] = (address & 0xff);
 808             }
 809         } else {
 810             getInet6Address_ipaddress(env, iaObj, (char *)caddr);
 811         }
 812         sa->sa6.sin6_port = htons(port);
 813         memcpy((void *)&sa->sa6.sin6_addr, caddr, sizeof(struct in6_addr));
 814         sa->sa6.sin6_family = AF_INET6;
 815         if (len != NULL) {
 816             *len = sizeof(struct sockaddr_in6);
 817         }
 818 
 819 #ifdef __linux__
 820         /*
 821          * On Linux if we are connecting to a link-local address
 822          * we need to specify the interface in the scope_id (2.4 kernel only)
 823          *
 824          * If the scope was cached then we use the cached value. If not cached but
 825          * specified in the Inet6Address we use that, but we first check if the
 826          * address needs to be routed via the loopback interface. In this case,
 827          * we override the specified value with that of the loopback interface.
 828          * If no cached value exists and no value was specified by user, then
 829          * we try to determine a value from the routing table. In all these
 830          * cases the used value is cached for further use.
 831          */
 832         if (IN6_IS_ADDR_LINKLOCAL(&sa->sa6.sin6_addr)) {
 833             unsigned int cached_scope_id = 0, scope_id = 0;
 834 
 835             if (ia6_cachedscopeidID) {
 836                 cached_scope_id = (int)(*env)->GetIntField(env, iaObj, ia6_cachedscopeidID);
 837                 /* if cached value exists then use it. Otherwise, check
 838                  * if scope is set in the address.
 839                  */
 840                 if (!cached_scope_id) {
 841                     if (ia6_scopeidID) {
 842                         scope_id = getInet6Address_scopeid(env, iaObj);
 843                     }
 844                     if (scope_id != 0) {
 845                         /* check user-specified value for loopback case
 846                          * that needs to be overridden
 847                          */
 848                         if (kernelIsV24() && needsLoopbackRoute(&sa->sa6.sin6_addr)) {
 849                             cached_scope_id = lo_scope_id;
 850                             (*env)->SetIntField(env, iaObj, ia6_cachedscopeidID, cached_scope_id);
 851                         }
 852                     } else {
 853                         /*
 854                          * Otherwise consult the IPv6 routing tables to
 855                          * try determine the appropriate interface.
 856                          */
 857                         if (kernelIsV24()) {
 858                             cached_scope_id = getDefaultIPv6Interface(&sa->sa6.sin6_addr);
 859                         } else {
 860                             cached_scope_id = getLocalScopeID((char *)&(sa->sa6.sin6_addr));
 861                             if (cached_scope_id == 0) {
 862                                 cached_scope_id = getDefaultIPv6Interface(&sa->sa6.sin6_addr);
 863                             }
 864                         }
 865                         (*env)->SetIntField(env, iaObj, ia6_cachedscopeidID, cached_scope_id);
 866                     }
 867                 }
 868             }
 869 
 870             /*
 871              * If we have a scope_id use the extended form
 872              * of sockaddr_in6.
 873              */
 874             sa->sa6.sin6_scope_id = cached_scope_id == 0 ? scope_id : cached_scope_id;
 875         }
 876 #else
 877         /* handle scope_id */
 878         if (family != java_net_InetAddress_IPv4) {
 879             if (ia6_scopeidID) {
 880                 sa->sa6.sin6_scope_id = getInet6Address_scopeid(env, iaObj);
 881             }
 882         }
 883 #endif
 884     } else {
 885         jint address;
 886         if (family != java_net_InetAddress_IPv4) {
 887             JNU_ThrowByName(env, JNU_JAVANETPKG "SocketException", "Protocol family unavailable");
 888             return -1;
 889         }
 890         address = getInetAddress_addr(env, iaObj);
 891         JNU_CHECK_EXCEPTION_RETURN(env, -1);
 892         sa->sa4.sin_port = htons(port);
 893         sa->sa4.sin_addr.s_addr = htonl(address);
 894         sa->sa4.sin_family = AF_INET;
 895         if (len != NULL) {
 896             *len = sizeof(struct sockaddr_in);
 897         }
 898     }
 899     return 0;
 900 }
 901 
 902 void
 903 NET_SetTrafficClass(SOCKETADDRESS *sa, int trafficClass) {
 904     if (sa->sa.sa_family == AF_INET6) {
 905         sa->sa6.sin6_flowinfo = htonl((trafficClass & 0xff) << 20);
 906     }
 907 }
 908 
 909 int
 910 NET_IsIPv4Mapped(jbyte* caddr) {
 911     int i;
 912     for (i = 0; i < 10; i++) {
 913         if (caddr[i] != 0x00) {
 914             return 0; /* false */
 915         }
 916     }
 917 
 918     if (((caddr[10] & 0xff) == 0xff) && ((caddr[11] & 0xff) == 0xff)) {
 919         return 1; /* true */
 920     }
 921     return 0; /* false */
 922 }
 923 
 924 int
 925 NET_IPv4MappedToIPv4(jbyte* caddr) {
 926     return ((caddr[12] & 0xff) << 24) | ((caddr[13] & 0xff) << 16) | ((caddr[14] & 0xff) << 8)
 927         | (caddr[15] & 0xff);
 928 }
 929 
 930 int
 931 NET_IsEqual(jbyte* caddr1, jbyte* caddr2) {
 932     int i;
 933     for (i = 0; i < 16; i++) {
 934         if (caddr1[i] != caddr2[i]) {
 935             return 0; /* false */
 936         }
 937     }
 938     return 1;
 939 }
 940 
 941 int NET_IsZeroAddr(jbyte* caddr) {
 942     int i;
 943     for (i = 0; i < 16; i++) {
 944         if (caddr[i] != 0) {
 945             return 0;
 946         }
 947     }
 948     return 1;
 949 }
 950 
 951 /*
 952  * Map the Java level socket option to the platform specific
 953  * level and option name.
 954  */
 955 int
 956 NET_MapSocketOption(jint cmd, int *level, int *optname) {
 957     static struct {
 958         jint cmd;
 959         int level;
 960         int optname;
 961     } const opts[] = {
 962         { java_net_SocketOptions_TCP_NODELAY,           IPPROTO_TCP,    TCP_NODELAY },
 963         { java_net_SocketOptions_SO_OOBINLINE,          SOL_SOCKET,     SO_OOBINLINE },
 964         { java_net_SocketOptions_SO_LINGER,             SOL_SOCKET,     SO_LINGER },
 965         { java_net_SocketOptions_SO_SNDBUF,             SOL_SOCKET,     SO_SNDBUF },
 966         { java_net_SocketOptions_SO_RCVBUF,             SOL_SOCKET,     SO_RCVBUF },
 967         { java_net_SocketOptions_SO_KEEPALIVE,          SOL_SOCKET,     SO_KEEPALIVE },
 968         { java_net_SocketOptions_SO_REUSEADDR,          SOL_SOCKET,     SO_REUSEADDR },
 969         { java_net_SocketOptions_SO_REUSEPORT,          SOL_SOCKET,     SO_REUSEPORT },
 970         { java_net_SocketOptions_SO_BROADCAST,          SOL_SOCKET,     SO_BROADCAST },
 971         { java_net_SocketOptions_IP_TOS,                IPPROTO_IP,     IP_TOS },
 972         { java_net_SocketOptions_IP_MULTICAST_IF,       IPPROTO_IP,     IP_MULTICAST_IF },
 973         { java_net_SocketOptions_IP_MULTICAST_IF2,      IPPROTO_IP,     IP_MULTICAST_IF },
 974         { java_net_SocketOptions_IP_MULTICAST_LOOP,     IPPROTO_IP,     IP_MULTICAST_LOOP },
 975     };
 976 
 977     int i;
 978 
 979     if (ipv6_available()) {
 980         switch (cmd) {
 981             // Different multicast options if IPv6 is enabled
 982             case java_net_SocketOptions_IP_MULTICAST_IF:
 983             case java_net_SocketOptions_IP_MULTICAST_IF2:
 984                 *level = IPPROTO_IPV6;
 985                 *optname = IPV6_MULTICAST_IF;
 986                 return 0;
 987 
 988             case java_net_SocketOptions_IP_MULTICAST_LOOP:
 989                 *level = IPPROTO_IPV6;
 990                 *optname = IPV6_MULTICAST_LOOP;
 991                 return 0;
 992 #if (defined(__solaris__) || defined(MACOSX))
 993             // Map IP_TOS request to IPV6_TCLASS
 994             case java_net_SocketOptions_IP_TOS:
 995                 *level = IPPROTO_IPV6;
 996                 *optname = IPV6_TCLASS;
 997                 return 0;
 998 #endif
 999         }
1000     }
1001 
1002     /*
1003      * Map the Java level option to the native level
1004      */
1005     for (i=0; i<(int)(sizeof(opts) / sizeof(opts[0])); i++) {
1006         if (cmd == opts[i].cmd) {
1007             *level = opts[i].level;
1008             *optname = opts[i].optname;
1009             return 0;
1010         }
1011     }
1012 
1013     /* not found */
1014     return -1;
1015 }
1016 
1017 /*
1018  * Determine the default interface for an IPv6 address.
1019  *
1020  * 1. Scans /proc/net/ipv6_route for a matching route
1021  *    (eg: fe80::/10 or a route for the specific address).
1022  *    This will tell us the interface to use (eg: "eth0").
1023  *
1024  * 2. Lookup /proc/net/if_inet6 to map the interface
1025  *    name to an interface index.
1026  *
1027  * Returns :-
1028  *      -1 if error
1029  *       0 if no matching interface
1030  *      >1 interface index to use for the link-local address.
1031  */
1032 #if defined(__linux__)
1033 int getDefaultIPv6Interface(struct in6_addr *target_addr) {
1034     FILE *f;
1035     char srcp[8][5];
1036     char hopp[8][5];
1037     int dest_plen, src_plen, use, refcnt, metric;
1038     unsigned long flags;
1039     char dest_str[40];
1040     struct in6_addr dest_addr;
1041     char device[16];
1042     jboolean match = JNI_FALSE;
1043 
1044     /*
1045      * Scan /proc/net/ipv6_route looking for a matching
1046      * route.
1047      */
1048     if ((f = fopen("/proc/net/ipv6_route", "r")) == NULL) {
1049         return -1;
1050     }
1051     while (fscanf(f, "%4s%4s%4s%4s%4s%4s%4s%4s %02x "
1052                      "%4s%4s%4s%4s%4s%4s%4s%4s %02x "
1053                      "%4s%4s%4s%4s%4s%4s%4s%4s "
1054                      "%08x %08x %08x %08lx %8s",
1055                      dest_str, &dest_str[5], &dest_str[10], &dest_str[15],
1056                      &dest_str[20], &dest_str[25], &dest_str[30], &dest_str[35],
1057                      &dest_plen,
1058                      srcp[0], srcp[1], srcp[2], srcp[3],
1059                      srcp[4], srcp[5], srcp[6], srcp[7],
1060                      &src_plen,
1061                      hopp[0], hopp[1], hopp[2], hopp[3],
1062                      hopp[4], hopp[5], hopp[6], hopp[7],
1063                      &metric, &use, &refcnt, &flags, device) == 31) {
1064 
1065         /*
1066          * Some routes should be ignored
1067          */
1068         if ( (dest_plen < 0 || dest_plen > 128)  ||
1069              (src_plen != 0) ||
1070              (flags & (RTF_POLICY | RTF_FLOW)) ||
1071              ((flags & RTF_REJECT) && dest_plen == 0) ) {
1072             continue;
1073         }
1074 
1075         /*
1076          * Convert the destination address
1077          */
1078         dest_str[4] = ':';
1079         dest_str[9] = ':';
1080         dest_str[14] = ':';
1081         dest_str[19] = ':';
1082         dest_str[24] = ':';
1083         dest_str[29] = ':';
1084         dest_str[34] = ':';
1085         dest_str[39] = '\0';
1086 
1087         if (inet_pton(AF_INET6, dest_str, &dest_addr) < 0) {
1088             /* not an Ipv6 address */
1089             continue;
1090         } else {
1091             /*
1092              * The prefix len (dest_plen) indicates the number of bits we
1093              * need to match on.
1094              *
1095              * dest_plen / 8    => number of bytes to match
1096              * dest_plen % 8    => number of additional bits to match
1097              *
1098              * eg: fe80::/10 => match 1 byte + 2 additional bits in the
1099              *                  next byte.
1100              */
1101             int byte_count = dest_plen >> 3;
1102             int extra_bits = dest_plen & 0x3;
1103 
1104             if (byte_count > 0) {
1105                 if (memcmp(target_addr, &dest_addr, byte_count)) {
1106                     continue;  /* no match */
1107                 }
1108             }
1109 
1110             if (extra_bits > 0) {
1111                 unsigned char c1 = ((unsigned char *)target_addr)[byte_count];
1112                 unsigned char c2 = ((unsigned char *)&dest_addr)[byte_count];
1113                 unsigned char mask = 0xff << (8 - extra_bits);
1114                 if ((c1 & mask) != (c2 & mask)) {
1115                     continue;
1116                 }
1117             }
1118 
1119             /*
1120              * We have a match
1121              */
1122             match = JNI_TRUE;
1123             break;
1124         }
1125     }
1126     fclose(f);
1127 
1128     /*
1129      * If there's a match then we lookup the interface
1130      * index.
1131      */
1132     if (match) {
1133         char devname[21];
1134         char addr6p[8][5];
1135         int plen, scope, dad_status, if_idx;
1136 
1137         if ((f = fopen("/proc/net/if_inet6", "r")) != NULL) {
1138             while (fscanf(f, "%4s%4s%4s%4s%4s%4s%4s%4s %08x %02x %02x %02x %20s\n",
1139                       addr6p[0], addr6p[1], addr6p[2], addr6p[3],
1140                       addr6p[4], addr6p[5], addr6p[6], addr6p[7],
1141                   &if_idx, &plen, &scope, &dad_status, devname) == 13) {
1142 
1143                 if (strcmp(devname, device) == 0) {
1144                     /*
1145                      * Found - so just return the index
1146                      */
1147                     fclose(f);
1148                     return if_idx;
1149                 }
1150             }
1151             fclose(f);
1152         } else {
1153             /*
1154              * Couldn't open /proc/net/if_inet6
1155              */
1156             return -1;
1157         }
1158     }
1159 
1160     /*
1161      * If we get here it means we didn't there wasn't any
1162      * route or we couldn't get the index of the interface.
1163      */
1164     return 0;
1165 }
1166 #endif
1167 
1168 
1169 /*
1170  * Wrapper for getsockopt system routine - does any necessary
1171  * pre/post processing to deal with OS specific oddities :-
1172  *
1173  * On Linux the SO_SNDBUF/SO_RCVBUF values must be post-processed
1174  * to compensate for an incorrect value returned by the kernel.
1175  */
1176 int
1177 NET_GetSockOpt(int fd, int level, int opt, void *result,
1178                int *len)
1179 {
1180     int rv;
1181     socklen_t socklen = *len;
1182 
1183     rv = getsockopt(fd, level, opt, result, &socklen);
1184     *len = socklen;
1185 
1186     if (rv < 0) {
1187         return rv;
1188     }
1189 
1190 #ifdef __linux__
1191     /*
1192      * On Linux SO_SNDBUF/SO_RCVBUF aren't symmetric. This
1193      * stems from additional socket structures in the send
1194      * and receive buffers.
1195      */
1196     if ((level == SOL_SOCKET) && ((opt == SO_SNDBUF)
1197                                   || (opt == SO_RCVBUF))) {
1198         int n = *((int *)result);
1199         n /= 2;
1200         *((int *)result) = n;
1201     }
1202 #endif
1203 
1204 /* Workaround for Mac OS treating linger value as
1205  *  signed integer
1206  */
1207 #ifdef MACOSX
1208     if (level == SOL_SOCKET && opt == SO_LINGER) {
1209         struct linger* to_cast = (struct linger*)result;
1210         to_cast->l_linger = (unsigned short)to_cast->l_linger;
1211     }
1212 #endif
1213     return rv;
1214 }
1215 
1216 /*
1217  * Wrapper for setsockopt system routine - performs any
1218  * necessary pre/post processing to deal with OS specific
1219  * issue :-
1220  *
1221  * On Solaris need to limit the suggested value for SO_SNDBUF
1222  * and SO_RCVBUF to the kernel configured limit
1223  *
1224  * For IP_TOS socket option need to mask off bits as this
1225  * aren't automatically masked by the kernel and results in
1226  * an error.
1227  */
1228 int
1229 NET_SetSockOpt(int fd, int level, int  opt, const void *arg,
1230                int len)
1231 {
1232 
1233 #ifndef IPTOS_TOS_MASK
1234 #define IPTOS_TOS_MASK 0x1e
1235 #endif
1236 #ifndef IPTOS_PREC_MASK
1237 #define IPTOS_PREC_MASK 0xe0
1238 #endif
1239 
1240 #if defined(_ALLBSD_SOURCE)
1241 #if defined(KIPC_MAXSOCKBUF)
1242     int mib[3];
1243     size_t rlen;
1244 #endif
1245 
1246     int *bufsize;
1247 
1248 #ifdef __APPLE__
1249     static int maxsockbuf = -1;
1250 #else
1251     static long maxsockbuf = -1;
1252 #endif
1253 #endif
1254 
1255     /*
1256      * IPPROTO/IP_TOS :-
1257      * 1. IPv6 on Solaris/Mac OS:
1258      *    Set the TOS OR Traffic Class value to cater for
1259      *    IPv6 and IPv4 scenarios.
1260      * 2. IPv6 on Linux: By default Linux ignores flowinfo
1261      *    field so enable IPV6_FLOWINFO_SEND so that flowinfo
1262      *    will be examined. We also set the IPv4 TOS option in this case.
1263      * 3. IPv4: set socket option based on ToS and Precedence
1264      *    fields (otherwise get invalid argument)
1265      */
1266     if (level == IPPROTO_IP && opt == IP_TOS) {
1267         int *iptos;
1268 
1269 #if defined(__linux__)
1270         if (ipv6_available()) {
1271             int optval = 1;
1272             if (setsockopt(fd, IPPROTO_IPV6, IPV6_FLOWINFO_SEND,
1273                            (void *)&optval, sizeof(optval)) < 0) {
1274                 return -1;
1275             }
1276            /*
1277             * Let's also set the IPV6_TCLASS flag.
1278             * Linux appears to allow both IP_TOS and IPV6_TCLASS to be set
1279             * This helps in mixed environments where IPv4 and IPv6 sockets
1280             * are connecting.
1281             */
1282            if (setsockopt(fd, IPPROTO_IPV6, IPV6_TCLASS,
1283                            arg, len) < 0) {
1284                 return -1;
1285             }
1286         }
1287 #endif
1288 
1289         iptos = (int *)arg;
1290         *iptos &= (IPTOS_TOS_MASK | IPTOS_PREC_MASK);
1291     }
1292 
1293     /*
1294      * SOL_SOCKET/{SO_SNDBUF,SO_RCVBUF} - On Solaris we may need to clamp
1295      * the value when it exceeds the system limit.
1296      */
1297 #ifdef __solaris__
1298     if (level == SOL_SOCKET) {
1299         if (opt == SO_SNDBUF || opt == SO_RCVBUF) {
1300             int sotype=0;
1301             socklen_t arglen;
1302             int *bufsize, maxbuf;
1303             int ret;
1304 
1305             /* Attempt with the original size */
1306             ret = setsockopt(fd, level, opt, arg, len);
1307             if ((ret == 0) || (ret == -1 && errno != ENOBUFS))
1308                 return ret;
1309 
1310             /* Exceeded system limit so clamp and retry */
1311 
1312             arglen = sizeof(sotype);
1313             if (getsockopt(fd, SOL_SOCKET, SO_TYPE, (void *)&sotype,
1314                            &arglen) < 0) {
1315                 return -1;
1316             }
1317 
1318             /*
1319              * We try to get tcp_maxbuf (and udp_max_buf) using
1320              * an ioctl() that isn't available on all versions of Solaris.
1321              * If that fails, we use the search algorithm in findMaxBuf()
1322              */
1323             if (!init_tcp_max_buf && sotype == SOCK_STREAM) {
1324                 tcp_max_buf = net_getParam("/dev/tcp", "tcp_max_buf");
1325                 if (tcp_max_buf == -1) {
1326                     tcp_max_buf = findMaxBuf(fd, opt, SOCK_STREAM);
1327                     if (tcp_max_buf == -1) {
1328                         return -1;
1329                     }
1330                 }
1331                 init_tcp_max_buf = 1;
1332             } else if (!init_udp_max_buf && sotype == SOCK_DGRAM) {
1333                 udp_max_buf = net_getParam("/dev/udp", "udp_max_buf");
1334                 if (udp_max_buf == -1) {
1335                     udp_max_buf = findMaxBuf(fd, opt, SOCK_DGRAM);
1336                     if (udp_max_buf == -1) {
1337                         return -1;
1338                     }
1339                 }
1340                 init_udp_max_buf = 1;
1341             }
1342 
1343             maxbuf = (sotype == SOCK_STREAM) ? tcp_max_buf : udp_max_buf;
1344             bufsize = (int *)arg;
1345             if (*bufsize > maxbuf) {
1346                 *bufsize = maxbuf;
1347             }
1348         }
1349     }
1350 #endif
1351 
1352 #ifdef _AIX
1353     if (level == SOL_SOCKET) {
1354         if (opt == SO_SNDBUF || opt == SO_RCVBUF) {
1355             /*
1356              * Just try to set the requested size. If it fails we will leave the
1357              * socket option as is. Setting the buffer size means only a hint in
1358              * the jse2/java software layer, see javadoc. In the previous
1359              * solution the buffer has always been truncated to a length of
1360              * 0x100000 Byte, even if the technical limit has not been reached.
1361              * This kind of absolute truncation was unexpected in the jck tests.
1362              */
1363             int ret = setsockopt(fd, level, opt, arg, len);
1364             if ((ret == 0) || (ret == -1 && errno == ENOBUFS)) {
1365                 // Accept failure because of insufficient buffer memory resources.
1366                 return 0;
1367             } else {
1368                 // Deliver all other kinds of errors.
1369                 return ret;
1370             }
1371         }
1372     }
1373 #endif
1374 
1375     /*
1376      * On Linux the receive buffer is used for both socket
1377      * structures and the packet payload. The implication
1378      * is that if SO_RCVBUF is too small then small packets
1379      * must be discarded.
1380      */
1381 #ifdef __linux__
1382     if (level == SOL_SOCKET && opt == SO_RCVBUF) {
1383         int *bufsize = (int *)arg;
1384         if (*bufsize < 1024) {
1385             *bufsize = 1024;
1386         }
1387     }
1388 #endif
1389 
1390 #if defined(_ALLBSD_SOURCE)
1391     /*
1392      * SOL_SOCKET/{SO_SNDBUF,SO_RCVBUF} - On FreeBSD need to
1393      * ensure that value is <= kern.ipc.maxsockbuf as otherwise we get
1394      * an ENOBUFS error.
1395      */
1396     if (level == SOL_SOCKET) {
1397         if (opt == SO_SNDBUF || opt == SO_RCVBUF) {
1398 #ifdef KIPC_MAXSOCKBUF
1399             if (maxsockbuf == -1) {
1400                mib[0] = CTL_KERN;
1401                mib[1] = KERN_IPC;
1402                mib[2] = KIPC_MAXSOCKBUF;
1403                rlen = sizeof(maxsockbuf);
1404                if (sysctl(mib, 3, &maxsockbuf, &rlen, NULL, 0) == -1)
1405                    maxsockbuf = 1024;
1406 
1407 #if 1
1408                /* XXXBSD: This is a hack to workaround mb_max/mb_max_adj
1409                   problem.  It should be removed when kern.ipc.maxsockbuf
1410                   will be real value. */
1411                maxsockbuf = (maxsockbuf/5)*4;
1412 #endif
1413            }
1414 #elif defined(__OpenBSD__)
1415            maxsockbuf = SB_MAX;
1416 #else
1417            maxsockbuf = 64 * 1024;      /* XXX: NetBSD */
1418 #endif
1419 
1420            bufsize = (int *)arg;
1421            if (*bufsize > maxsockbuf) {
1422                *bufsize = maxsockbuf;
1423            }
1424 
1425            if (opt == SO_RCVBUF && *bufsize < 1024) {
1426                 *bufsize = 1024;
1427            }
1428 
1429         }
1430     }
1431 #endif
1432 
1433 #if defined(_ALLBSD_SOURCE) || defined(_AIX)
1434     /*
1435      * On Solaris, SO_REUSEADDR will allow multiple datagram
1436      * sockets to bind to the same port. The network jck tests check
1437      * for this "feature", so we need to emulate it by turning on
1438      * SO_REUSEPORT as well for that combination.
1439      */
1440     if (level == SOL_SOCKET && opt == SO_REUSEADDR) {
1441         int sotype;
1442         socklen_t arglen;
1443 
1444         arglen = sizeof(sotype);
1445         if (getsockopt(fd, SOL_SOCKET, SO_TYPE, (void *)&sotype, &arglen) < 0) {
1446             return -1;
1447         }
1448 
1449         if (sotype == SOCK_DGRAM) {
1450             setsockopt(fd, level, SO_REUSEPORT, arg, len);
1451         }
1452     }
1453 #endif
1454 
1455     return setsockopt(fd, level, opt, arg, len);
1456 }
1457 
1458 /*
1459  * Wrapper for bind system call - performs any necessary pre/post
1460  * processing to deal with OS specific issues :-
1461  *
1462  * Linux allows a socket to bind to 127.0.0.255 which must be
1463  * caught.
1464  *
1465  * On Solaris with IPv6 enabled we must use an exclusive
1466  * bind to guarantee a unique port number across the IPv4 and
1467  * IPv6 port spaces.
1468  *
1469  */
1470 int
1471 NET_Bind(int fd, SOCKETADDRESS *sa, int len)
1472 {
1473 #if defined(__solaris__)
1474     int level = -1;
1475     int exclbind = -1;
1476 #endif
1477     int rv;
1478     int arg, alen;
1479 
1480 #ifdef __linux__
1481     /*
1482      * ## get bugId for this issue - goes back to 1.2.2 port ##
1483      * ## When IPv6 is enabled this will be an IPv4-mapped
1484      * ## with family set to AF_INET6
1485      */
1486     if (sa->sa.sa_family == AF_INET) {
1487         if ((ntohl(sa->sa4.sin_addr.s_addr) & 0x7f0000ff) == 0x7f0000ff) {
1488             errno = EADDRNOTAVAIL;
1489             return -1;
1490         }
1491     }
1492 #endif
1493 
1494 #if defined(__solaris__)
1495     /*
1496      * Solaris has separate IPv4 and IPv6 port spaces so we
1497      * use an exclusive bind when SO_REUSEADDR is not used to
1498      * give the illusion of a unified port space.
1499      * This also avoids problems with IPv6 sockets connecting
1500      * to IPv4 mapped addresses whereby the socket conversion
1501      * results in a late bind that fails because the
1502      * corresponding IPv4 port is in use.
1503      */
1504     alen = sizeof(arg);
1505 
1506     if (useExclBind ||
1507         getsockopt(fd, SOL_SOCKET, SO_REUSEADDR, (char *)&arg, &alen) == 0)
1508     {
1509         if (useExclBind || arg == 0) {
1510             /*
1511              * SO_REUSEADDR is disabled or sun.net.useExclusiveBind
1512              * property is true so enable TCP_EXCLBIND or
1513              * UDP_EXCLBIND
1514              */
1515             alen = sizeof(arg);
1516             if (getsockopt(fd, SOL_SOCKET, SO_TYPE, (char *)&arg, &alen) == 0)
1517             {
1518                 if (arg == SOCK_STREAM) {
1519                     level = IPPROTO_TCP;
1520                     exclbind = TCP_EXCLBIND;
1521                 } else {
1522                     level = IPPROTO_UDP;
1523                     exclbind = UDP_EXCLBIND;
1524                 }
1525             }
1526 
1527             arg = 1;
1528             setsockopt(fd, level, exclbind, (char *)&arg, sizeof(arg));
1529         }
1530     }
1531 
1532 #endif
1533 
1534     rv = bind(fd, &sa->sa, len);
1535 
1536 #if defined(__solaris__)
1537     if (rv < 0) {
1538         int en = errno;
1539         /* Restore *_EXCLBIND if the bind fails */
1540         if (exclbind != -1) {
1541             int arg = 0;
1542             setsockopt(fd, level, exclbind, (char *)&arg,
1543                        sizeof(arg));
1544         }
1545         errno = en;
1546     }
1547 #endif
1548 
1549     return rv;
1550 }
1551 
1552 /**
1553  * Wrapper for poll with timeout on a single file descriptor.
1554  *
1555  * flags (defined in net_util_md.h can be any combination of
1556  * NET_WAIT_READ, NET_WAIT_WRITE & NET_WAIT_CONNECT.
1557  *
1558  * The function will return when either the socket is ready for one
1559  * of the specified operations or the timeout expired.
1560  *
1561  * It returns the time left from the timeout (possibly 0), or -1 if it expired.
1562  */
1563 
1564 jint
1565 NET_Wait(JNIEnv *env, jint fd, jint flags, jint timeout)
1566 {
1567     jlong prevNanoTime = JVM_NanoTime(env, 0);
1568     jlong nanoTimeout = (jlong) timeout * NET_NSEC_PER_MSEC;
1569     jint read_rv;
1570 
1571     while (1) {
1572         jlong newNanoTime;
1573         struct pollfd pfd;
1574         pfd.fd = fd;
1575         pfd.events = 0;
1576         if (flags & NET_WAIT_READ)
1577           pfd.events |= POLLIN;
1578         if (flags & NET_WAIT_WRITE)
1579           pfd.events |= POLLOUT;
1580         if (flags & NET_WAIT_CONNECT)
1581           pfd.events |= POLLOUT;
1582 
1583         errno = 0;
1584         read_rv = NET_Poll(&pfd, 1, nanoTimeout / NET_NSEC_PER_MSEC);
1585 
1586         newNanoTime = JVM_NanoTime(env, 0);
1587         nanoTimeout -= (newNanoTime - prevNanoTime);
1588         if (nanoTimeout < NET_NSEC_PER_MSEC) {
1589           return read_rv > 0 ? 0 : -1;
1590         }
1591         prevNanoTime = newNanoTime;
1592 
1593         if (read_rv > 0) {
1594           break;
1595         }
1596       } /* while */
1597     return (nanoTimeout / NET_NSEC_PER_MSEC);
1598 }