src/java.base/unix/native/libnet/net_util_md.c
changeset 47216 71c04702a3d5
parent 45061 74b09ee3cd55
child 49440 396ea30afbd5
child 56230 489867818774
equal deleted inserted replaced
47215:4ebc2e2fb97c 47216:71c04702a3d5
       
     1 /*
       
     2  * Copyright (c) 1997, 2017, 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         defaultIndex = (*env)->GetStaticIntField(env, ni_class,
       
    94                                                  ni_defaultIndexID);
       
    95         sin6->sin6_scope_id = defaultIndex;
       
    96     }
       
    97 #endif
       
    98 }
       
    99 
       
   100 int getDefaultScopeID(JNIEnv *env) {
       
   101     int defaultIndex = 0;
       
   102     static jclass ni_class = NULL;
       
   103     static jfieldID ni_defaultIndexID;
       
   104     if (ni_class == NULL) {
       
   105         jclass c = (*env)->FindClass(env, "java/net/NetworkInterface");
       
   106         CHECK_NULL_RETURN(c, 0);
       
   107         c = (*env)->NewGlobalRef(env, c);
       
   108         CHECK_NULL_RETURN(c, 0);
       
   109         ni_defaultIndexID = (*env)->GetStaticFieldID(env, c, "defaultIndex", "I");
       
   110         CHECK_NULL_RETURN(ni_defaultIndexID, 0);
       
   111         ni_class = c;
       
   112     }
       
   113     defaultIndex = (*env)->GetStaticIntField(env, ni_class,
       
   114                                              ni_defaultIndexID);
       
   115     return defaultIndex;
       
   116 }
       
   117 
       
   118 #define RESTARTABLE(_cmd, _result) do { \
       
   119     do { \
       
   120         _result = _cmd; \
       
   121     } while((_result == -1) && (errno == EINTR)); \
       
   122 } while(0)
       
   123 
       
   124 int NET_SocketAvailable(int s, jint *pbytes) {
       
   125     int result;
       
   126     RESTARTABLE(ioctl(s, FIONREAD, pbytes), result);
       
   127     // note: ioctl can return 0 when successful, NET_SocketAvailable
       
   128     // is expected to return 0 on failure and 1 on success.
       
   129     return (result == -1) ? 0 : 1;
       
   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 #if defined(DONT_ENABLE_IPV6)
       
   282 jint  IPv6_supported()
       
   283 {
       
   284     return JNI_FALSE;
       
   285 }
       
   286 
       
   287 #else /* !DONT_ENABLE_IPV6 */
       
   288 
       
   289 jint  IPv6_supported()
       
   290 {
       
   291     int fd;
       
   292     void *ipv6_fn;
       
   293     SOCKETADDRESS sa;
       
   294     socklen_t sa_len = sizeof(SOCKETADDRESS);
       
   295 
       
   296     fd = socket(AF_INET6, SOCK_STREAM, 0) ;
       
   297     if (fd < 0) {
       
   298         /*
       
   299          *  TODO: We really cant tell since it may be an unrelated error
       
   300          *  for now we will assume that AF_INET6 is not available
       
   301          */
       
   302         return JNI_FALSE;
       
   303     }
       
   304 
       
   305     /*
       
   306      * If fd 0 is a socket it means we've been launched from inetd or
       
   307      * xinetd. If it's a socket then check the family - if it's an
       
   308      * IPv4 socket then we need to disable IPv6.
       
   309      */
       
   310     if (getsockname(0, &sa.sa, &sa_len) == 0) {
       
   311         if (sa.sa.sa_family != AF_INET6) {
       
   312             close(fd);
       
   313             return JNI_FALSE;
       
   314         }
       
   315     }
       
   316 
       
   317     /**
       
   318      * Linux - check if any interface has an IPv6 address.
       
   319      * Don't need to parse the line - we just need an indication.
       
   320      */
       
   321 #ifdef __linux__
       
   322     {
       
   323         FILE *fP = fopen("/proc/net/if_inet6", "r");
       
   324         char buf[255];
       
   325         char *bufP;
       
   326 
       
   327         if (fP == NULL) {
       
   328             close(fd);
       
   329             return JNI_FALSE;
       
   330         }
       
   331         bufP = fgets(buf, sizeof(buf), fP);
       
   332         fclose(fP);
       
   333         if (bufP == NULL) {
       
   334             close(fd);
       
   335             return JNI_FALSE;
       
   336         }
       
   337     }
       
   338 #endif
       
   339 
       
   340     /**
       
   341      * On Solaris 8 it's possible to create INET6 sockets even
       
   342      * though IPv6 is not enabled on all interfaces. Thus we
       
   343      * query the number of IPv6 addresses to verify that IPv6
       
   344      * has been configured on at least one interface.
       
   345      *
       
   346      * On Linux it doesn't matter - if IPv6 is built-in the
       
   347      * kernel then IPv6 addresses will be bound automatically
       
   348      * to all interfaces.
       
   349      */
       
   350 #ifdef __solaris__
       
   351 
       
   352 #ifdef SIOCGLIFNUM
       
   353     {
       
   354         struct lifnum numifs;
       
   355 
       
   356         numifs.lifn_family = AF_INET6;
       
   357         numifs.lifn_flags = 0;
       
   358         if (ioctl(fd, SIOCGLIFNUM, (char *)&numifs) < 0) {
       
   359             /**
       
   360              * SIOCGLIFNUM failed - assume IPv6 not configured
       
   361              */
       
   362             close(fd);
       
   363             return JNI_FALSE;
       
   364         }
       
   365         /**
       
   366          * If no IPv6 addresses then return false. If count > 0
       
   367          * it's possible that all IPv6 addresses are "down" but
       
   368          * that's okay as they may be brought "up" while the
       
   369          * VM is running.
       
   370          */
       
   371         if (numifs.lifn_count == 0) {
       
   372             close(fd);
       
   373             return JNI_FALSE;
       
   374         }
       
   375     }
       
   376 #else
       
   377     /* SIOCGLIFNUM not defined in build environment ??? */
       
   378     close(fd);
       
   379     return JNI_FALSE;
       
   380 #endif
       
   381 
       
   382 #endif /* __solaris */
       
   383 
       
   384     /*
       
   385      *  OK we may have the stack available in the kernel,
       
   386      *  we should also check if the APIs are available.
       
   387      */
       
   388     ipv6_fn = JVM_FindLibraryEntry(RTLD_DEFAULT, "inet_pton");
       
   389     close(fd);
       
   390     if (ipv6_fn == NULL ) {
       
   391         return JNI_FALSE;
       
   392     } else {
       
   393         return JNI_TRUE;
       
   394     }
       
   395 }
       
   396 #endif /* DONT_ENABLE_IPV6 */
       
   397 
       
   398 jint reuseport_supported()
       
   399 {
       
   400     /* Do a simple dummy call, and try to figure out from that */
       
   401     int one = 1;
       
   402     int rv, s;
       
   403     s = socket(PF_INET, SOCK_STREAM, 0);
       
   404     if (s < 0) {
       
   405         return JNI_FALSE;
       
   406     }
       
   407     rv = setsockopt(s, SOL_SOCKET, SO_REUSEPORT, (void *)&one, sizeof(one));
       
   408     if (rv != 0) {
       
   409         rv = JNI_FALSE;
       
   410     } else {
       
   411         rv = JNI_TRUE;
       
   412     }
       
   413     close(s);
       
   414     return rv;
       
   415 }
       
   416 
       
   417 void NET_ThrowUnknownHostExceptionWithGaiError(JNIEnv *env,
       
   418                                                const char* hostname,
       
   419                                                int gai_error)
       
   420 {
       
   421     int size;
       
   422     char *buf;
       
   423     const char *format = "%s: %s";
       
   424     const char *error_string = gai_strerror(gai_error);
       
   425     if (error_string == NULL)
       
   426         error_string = "unknown error";
       
   427 
       
   428     size = strlen(format) + strlen(hostname) + strlen(error_string) + 2;
       
   429     buf = (char *) malloc(size);
       
   430     if (buf) {
       
   431         jstring s;
       
   432         sprintf(buf, format, hostname, error_string);
       
   433         s = JNU_NewStringPlatform(env, buf);
       
   434         if (s != NULL) {
       
   435             jobject x = JNU_NewObjectByName(env,
       
   436                                             "java/net/UnknownHostException",
       
   437                                             "(Ljava/lang/String;)V", s);
       
   438             if (x != NULL)
       
   439                 (*env)->Throw(env, x);
       
   440         }
       
   441         free(buf);
       
   442     }
       
   443 }
       
   444 
       
   445 #if defined(__linux__)
       
   446 
       
   447 /* following code creates a list of addresses from the kernel
       
   448  * routing table that are routed via the loopback address.
       
   449  * We check all destination addresses against this table
       
   450  * and override the scope_id field to use the relevant value for "lo"
       
   451  * in order to work-around the Linux bug that prevents packets destined
       
   452  * for certain local addresses from being sent via a physical interface.
       
   453  */
       
   454 
       
   455 struct loopback_route {
       
   456     struct in6_addr addr; /* destination address */
       
   457     int plen; /* prefix length */
       
   458 };
       
   459 
       
   460 static struct loopback_route *loRoutes = 0;
       
   461 static int nRoutes = 0; /* number of routes */
       
   462 static int loRoutes_size = 16; /* initial size */
       
   463 static int lo_scope_id = 0;
       
   464 
       
   465 static void initLoopbackRoutes();
       
   466 
       
   467 void printAddr (struct in6_addr *addr) {
       
   468     int i;
       
   469     for (i=0; i<16; i++) {
       
   470         printf ("%02x", addr->s6_addr[i]);
       
   471     }
       
   472     printf ("\n");
       
   473 }
       
   474 
       
   475 static jboolean needsLoopbackRoute (struct in6_addr* dest_addr) {
       
   476     int byte_count;
       
   477     int extra_bits, i;
       
   478     struct loopback_route *ptr;
       
   479 
       
   480     if (loRoutes == 0) {
       
   481         initLoopbackRoutes();
       
   482     }
       
   483 
       
   484     for (ptr = loRoutes, i=0; i<nRoutes; i++, ptr++) {
       
   485         struct in6_addr *target_addr=&ptr->addr;
       
   486         int dest_plen = ptr->plen;
       
   487         byte_count = dest_plen >> 3;
       
   488         extra_bits = dest_plen & 0x3;
       
   489 
       
   490         if (byte_count > 0) {
       
   491             if (memcmp(target_addr, dest_addr, byte_count)) {
       
   492                 continue;  /* no match */
       
   493             }
       
   494         }
       
   495 
       
   496         if (extra_bits > 0) {
       
   497             unsigned char c1 = ((unsigned char *)target_addr)[byte_count];
       
   498             unsigned char c2 = ((unsigned char *)&dest_addr)[byte_count];
       
   499             unsigned char mask = 0xff << (8 - extra_bits);
       
   500             if ((c1 & mask) != (c2 & mask)) {
       
   501                 continue;
       
   502             }
       
   503         }
       
   504         return JNI_TRUE;
       
   505     }
       
   506     return JNI_FALSE;
       
   507 }
       
   508 
       
   509 
       
   510 static void initLoopbackRoutes() {
       
   511     FILE *f;
       
   512     char srcp[8][5];
       
   513     char hopp[8][5];
       
   514     int dest_plen, src_plen, use, refcnt, metric;
       
   515     unsigned long flags;
       
   516     char dest_str[40];
       
   517     struct in6_addr dest_addr;
       
   518     char device[16];
       
   519     struct loopback_route *loRoutesTemp;
       
   520 
       
   521     if (loRoutes != 0) {
       
   522         free (loRoutes);
       
   523     }
       
   524     loRoutes = calloc (loRoutes_size, sizeof(struct loopback_route));
       
   525     if (loRoutes == 0) {
       
   526         return;
       
   527     }
       
   528     /*
       
   529      * Scan /proc/net/ipv6_route looking for a matching
       
   530      * route.
       
   531      */
       
   532     if ((f = fopen("/proc/net/ipv6_route", "r")) == NULL) {
       
   533         return ;
       
   534     }
       
   535     while (fscanf(f, "%4s%4s%4s%4s%4s%4s%4s%4s %02x "
       
   536                      "%4s%4s%4s%4s%4s%4s%4s%4s %02x "
       
   537                      "%4s%4s%4s%4s%4s%4s%4s%4s "
       
   538                      "%08x %08x %08x %08lx %8s",
       
   539                      dest_str, &dest_str[5], &dest_str[10], &dest_str[15],
       
   540                      &dest_str[20], &dest_str[25], &dest_str[30], &dest_str[35],
       
   541                      &dest_plen,
       
   542                      srcp[0], srcp[1], srcp[2], srcp[3],
       
   543                      srcp[4], srcp[5], srcp[6], srcp[7],
       
   544                      &src_plen,
       
   545                      hopp[0], hopp[1], hopp[2], hopp[3],
       
   546                      hopp[4], hopp[5], hopp[6], hopp[7],
       
   547                      &metric, &use, &refcnt, &flags, device) == 31) {
       
   548 
       
   549         /*
       
   550          * Some routes should be ignored
       
   551          */
       
   552         if ( (dest_plen < 0 || dest_plen > 128)  ||
       
   553              (src_plen != 0) ||
       
   554              (flags & (RTF_POLICY | RTF_FLOW)) ||
       
   555              ((flags & RTF_REJECT) && dest_plen == 0) ) {
       
   556             continue;
       
   557         }
       
   558 
       
   559         /*
       
   560          * Convert the destination address
       
   561          */
       
   562         dest_str[4] = ':';
       
   563         dest_str[9] = ':';
       
   564         dest_str[14] = ':';
       
   565         dest_str[19] = ':';
       
   566         dest_str[24] = ':';
       
   567         dest_str[29] = ':';
       
   568         dest_str[34] = ':';
       
   569         dest_str[39] = '\0';
       
   570 
       
   571         if (inet_pton(AF_INET6, dest_str, &dest_addr) < 0) {
       
   572             /* not an Ipv6 address */
       
   573             continue;
       
   574         }
       
   575         if (strcmp(device, "lo") != 0) {
       
   576             /* Not a loopback route */
       
   577             continue;
       
   578         } else {
       
   579             if (nRoutes == loRoutes_size) {
       
   580                 loRoutesTemp = realloc (loRoutes, loRoutes_size *
       
   581                                         sizeof (struct loopback_route) * 2);
       
   582 
       
   583                 if (loRoutesTemp == 0) {
       
   584                     free(loRoutes);
       
   585                     fclose (f);
       
   586                     return;
       
   587                 }
       
   588                 loRoutes=loRoutesTemp;
       
   589                 loRoutes_size *= 2;
       
   590             }
       
   591             memcpy (&loRoutes[nRoutes].addr,&dest_addr,sizeof(struct in6_addr));
       
   592             loRoutes[nRoutes].plen = dest_plen;
       
   593             nRoutes ++;
       
   594         }
       
   595     }
       
   596 
       
   597     fclose (f);
       
   598     {
       
   599         /* now find the scope_id for "lo" */
       
   600 
       
   601         char devname[21];
       
   602         char addr6p[8][5];
       
   603         int plen, scope, dad_status, if_idx;
       
   604 
       
   605         if ((f = fopen("/proc/net/if_inet6", "r")) != NULL) {
       
   606             while (fscanf(f, "%4s%4s%4s%4s%4s%4s%4s%4s %08x %02x %02x %02x %20s\n",
       
   607                       addr6p[0], addr6p[1], addr6p[2], addr6p[3],
       
   608                       addr6p[4], addr6p[5], addr6p[6], addr6p[7],
       
   609                   &if_idx, &plen, &scope, &dad_status, devname) == 13) {
       
   610 
       
   611                 if (strcmp(devname, "lo") == 0) {
       
   612                     /*
       
   613                      * Found - so just return the index
       
   614                      */
       
   615                     fclose(f);
       
   616                     lo_scope_id = if_idx;
       
   617                     return;
       
   618                 }
       
   619             }
       
   620             fclose(f);
       
   621         }
       
   622     }
       
   623 }
       
   624 
       
   625 /*
       
   626  * Following is used for binding to local addresses. Equivalent
       
   627  * to code above, for bind().
       
   628  */
       
   629 
       
   630 struct localinterface {
       
   631     int index;
       
   632     char localaddr [16];
       
   633 };
       
   634 
       
   635 static struct localinterface *localifs = 0;
       
   636 static int localifsSize = 0;    /* size of array */
       
   637 static int nifs = 0;            /* number of entries used in array */
       
   638 
       
   639 /* not thread safe: make sure called once from one thread */
       
   640 
       
   641 static void initLocalIfs () {
       
   642     FILE *f;
       
   643     unsigned char staddr [16];
       
   644     char ifname [33];
       
   645     struct localinterface *lif=0;
       
   646     int index, x1, x2, x3;
       
   647     unsigned int u0,u1,u2,u3,u4,u5,u6,u7,u8,u9,ua,ub,uc,ud,ue,uf;
       
   648 
       
   649     if ((f = fopen("/proc/net/if_inet6", "r")) == NULL) {
       
   650         return ;
       
   651     }
       
   652     while (fscanf (f, "%2x%2x%2x%2x%2x%2x%2x%2x%2x%2x%2x%2x%2x%2x%2x%2x "
       
   653                 "%d %x %x %x %32s",&u0,&u1,&u2,&u3,&u4,&u5,&u6,&u7,
       
   654                 &u8,&u9,&ua,&ub,&uc,&ud,&ue,&uf,
       
   655                 &index, &x1, &x2, &x3, ifname) == 21) {
       
   656         staddr[0] = (unsigned char)u0;
       
   657         staddr[1] = (unsigned char)u1;
       
   658         staddr[2] = (unsigned char)u2;
       
   659         staddr[3] = (unsigned char)u3;
       
   660         staddr[4] = (unsigned char)u4;
       
   661         staddr[5] = (unsigned char)u5;
       
   662         staddr[6] = (unsigned char)u6;
       
   663         staddr[7] = (unsigned char)u7;
       
   664         staddr[8] = (unsigned char)u8;
       
   665         staddr[9] = (unsigned char)u9;
       
   666         staddr[10] = (unsigned char)ua;
       
   667         staddr[11] = (unsigned char)ub;
       
   668         staddr[12] = (unsigned char)uc;
       
   669         staddr[13] = (unsigned char)ud;
       
   670         staddr[14] = (unsigned char)ue;
       
   671         staddr[15] = (unsigned char)uf;
       
   672         nifs ++;
       
   673         if (nifs > localifsSize) {
       
   674             localifs = (struct localinterface *) realloc (
       
   675                         localifs, sizeof (struct localinterface)* (localifsSize+5));
       
   676             if (localifs == 0) {
       
   677                 nifs = 0;
       
   678                 fclose (f);
       
   679                 return;
       
   680             }
       
   681             lif = localifs + localifsSize;
       
   682             localifsSize += 5;
       
   683         } else {
       
   684             lif ++;
       
   685         }
       
   686         memcpy (lif->localaddr, staddr, 16);
       
   687         lif->index = index;
       
   688     }
       
   689     fclose (f);
       
   690 }
       
   691 
       
   692 /* return the scope_id (interface index) of the
       
   693  * interface corresponding to the given address
       
   694  * returns 0 if no match found
       
   695  */
       
   696 
       
   697 static int getLocalScopeID (char *addr) {
       
   698     struct localinterface *lif;
       
   699     int i;
       
   700     if (localifs == 0) {
       
   701         initLocalIfs();
       
   702     }
       
   703     for (i=0, lif=localifs; i<nifs; i++, lif++) {
       
   704         if (memcmp (addr, lif->localaddr, 16) == 0) {
       
   705             return lif->index;
       
   706         }
       
   707     }
       
   708     return 0;
       
   709 }
       
   710 
       
   711 void platformInit () {
       
   712     initLoopbackRoutes();
       
   713     initLocalIfs();
       
   714 }
       
   715 
       
   716 #elif defined(_AIX)
       
   717 
       
   718 /* Initialize stubs for blocking I/O workarounds (see src/solaris/native/java/net/linux_close.c) */
       
   719 extern void aix_close_init();
       
   720 
       
   721 void platformInit () {
       
   722     aix_close_init();
       
   723 }
       
   724 
       
   725 #else
       
   726 
       
   727 void platformInit () {}
       
   728 
       
   729 #endif
       
   730 
       
   731 void parseExclusiveBindProperty(JNIEnv *env) {
       
   732 #ifdef __solaris__
       
   733     jstring s, flagSet;
       
   734     jclass iCls;
       
   735     jmethodID mid;
       
   736 
       
   737     s = (*env)->NewStringUTF(env, "sun.net.useExclusiveBind");
       
   738     CHECK_NULL(s);
       
   739     iCls = (*env)->FindClass(env, "java/lang/System");
       
   740     CHECK_NULL(iCls);
       
   741     mid = (*env)->GetStaticMethodID(env, iCls, "getProperty",
       
   742                 "(Ljava/lang/String;)Ljava/lang/String;");
       
   743     CHECK_NULL(mid);
       
   744     flagSet = (*env)->CallStaticObjectMethod(env, iCls, mid, s);
       
   745     if (flagSet != NULL) {
       
   746         useExclBind = 1;
       
   747     }
       
   748 #endif
       
   749 }
       
   750 
       
   751 JNIEXPORT jint JNICALL
       
   752 NET_EnableFastTcpLoopback(int fd) {
       
   753     return 0;
       
   754 }
       
   755 
       
   756 /**
       
   757  * See net_util.h for documentation
       
   758  */
       
   759 JNIEXPORT int JNICALL
       
   760 NET_InetAddressToSockaddr(JNIEnv *env, jobject iaObj, int port,
       
   761                           SOCKETADDRESS *sa, int *len,
       
   762                           jboolean v4MappedAddress)
       
   763 {
       
   764     jint family = getInetAddress_family(env, iaObj);
       
   765     memset((char *)sa, 0, sizeof(SOCKETADDRESS));
       
   766 
       
   767     if (ipv6_available() &&
       
   768         !(family == java_net_InetAddress_IPv4 &&
       
   769           v4MappedAddress == JNI_FALSE))
       
   770     {
       
   771         jbyte caddr[16];
       
   772         jint address;
       
   773 
       
   774         if (family == java_net_InetAddress_IPv4) {
       
   775             // convert to IPv4-mapped address
       
   776             memset((char *)caddr, 0, 16);
       
   777             address = getInetAddress_addr(env, iaObj);
       
   778             if (address == INADDR_ANY) {
       
   779                 /* we would always prefer IPv6 wildcard address
       
   780                  * caddr[10] = 0xff;
       
   781                  * caddr[11] = 0xff; */
       
   782             } else {
       
   783                 caddr[10] = 0xff;
       
   784                 caddr[11] = 0xff;
       
   785                 caddr[12] = ((address >> 24) & 0xff);
       
   786                 caddr[13] = ((address >> 16) & 0xff);
       
   787                 caddr[14] = ((address >> 8) & 0xff);
       
   788                 caddr[15] = (address & 0xff);
       
   789             }
       
   790         } else {
       
   791             getInet6Address_ipaddress(env, iaObj, (char *)caddr);
       
   792         }
       
   793         sa->sa6.sin6_port = htons(port);
       
   794         memcpy((void *)&sa->sa6.sin6_addr, caddr, sizeof(struct in6_addr));
       
   795         sa->sa6.sin6_family = AF_INET6;
       
   796         if (len != NULL) {
       
   797             *len = sizeof(struct sockaddr_in6);
       
   798         }
       
   799 
       
   800 #ifdef __linux__
       
   801         /*
       
   802          * On Linux if we are connecting to a link-local address
       
   803          * we need to specify the interface in the scope_id (2.4 kernel only)
       
   804          *
       
   805          * If the scope was cached then we use the cached value. If not cached but
       
   806          * specified in the Inet6Address we use that, but we first check if the
       
   807          * address needs to be routed via the loopback interface. In this case,
       
   808          * we override the specified value with that of the loopback interface.
       
   809          * If no cached value exists and no value was specified by user, then
       
   810          * we try to determine a value from the routing table. In all these
       
   811          * cases the used value is cached for further use.
       
   812          */
       
   813         if (IN6_IS_ADDR_LINKLOCAL(&sa->sa6.sin6_addr)) {
       
   814             unsigned int cached_scope_id = 0, scope_id = 0;
       
   815 
       
   816             if (ia6_cachedscopeidID) {
       
   817                 cached_scope_id = (int)(*env)->GetIntField(env, iaObj, ia6_cachedscopeidID);
       
   818                 /* if cached value exists then use it. Otherwise, check
       
   819                  * if scope is set in the address.
       
   820                  */
       
   821                 if (!cached_scope_id) {
       
   822                     if (ia6_scopeidID) {
       
   823                         scope_id = getInet6Address_scopeid(env, iaObj);
       
   824                     }
       
   825                     if (scope_id != 0) {
       
   826                         /* check user-specified value for loopback case
       
   827                          * that needs to be overridden
       
   828                          */
       
   829                         if (kernelIsV24() && needsLoopbackRoute(&sa->sa6.sin6_addr)) {
       
   830                             cached_scope_id = lo_scope_id;
       
   831                             (*env)->SetIntField(env, iaObj, ia6_cachedscopeidID, cached_scope_id);
       
   832                         }
       
   833                     } else {
       
   834                         /*
       
   835                          * Otherwise consult the IPv6 routing tables to
       
   836                          * try determine the appropriate interface.
       
   837                          */
       
   838                         if (kernelIsV24()) {
       
   839                             cached_scope_id = getDefaultIPv6Interface(&sa->sa6.sin6_addr);
       
   840                         } else {
       
   841                             cached_scope_id = getLocalScopeID((char *)&(sa->sa6.sin6_addr));
       
   842                             if (cached_scope_id == 0) {
       
   843                                 cached_scope_id = getDefaultIPv6Interface(&sa->sa6.sin6_addr);
       
   844                             }
       
   845                         }
       
   846                         (*env)->SetIntField(env, iaObj, ia6_cachedscopeidID, cached_scope_id);
       
   847                     }
       
   848                 }
       
   849             }
       
   850 
       
   851             /*
       
   852              * If we have a scope_id use the extended form
       
   853              * of sockaddr_in6.
       
   854              */
       
   855             sa->sa6.sin6_scope_id = cached_scope_id == 0 ? scope_id : cached_scope_id;
       
   856         }
       
   857 #else
       
   858         /* handle scope_id */
       
   859         if (family != java_net_InetAddress_IPv4) {
       
   860             if (ia6_scopeidID) {
       
   861                 sa->sa6.sin6_scope_id = getInet6Address_scopeid(env, iaObj);
       
   862             }
       
   863         }
       
   864 #endif
       
   865     } else {
       
   866         jint address;
       
   867         if (family != java_net_InetAddress_IPv4) {
       
   868             JNU_ThrowByName(env, JNU_JAVANETPKG "SocketException", "Protocol family unavailable");
       
   869             return -1;
       
   870         }
       
   871         address = getInetAddress_addr(env, iaObj);
       
   872         sa->sa4.sin_port = htons(port);
       
   873         sa->sa4.sin_addr.s_addr = htonl(address);
       
   874         sa->sa4.sin_family = AF_INET;
       
   875         if (len != NULL) {
       
   876             *len = sizeof(struct sockaddr_in);
       
   877         }
       
   878     }
       
   879     return 0;
       
   880 }
       
   881 
       
   882 void
       
   883 NET_SetTrafficClass(SOCKETADDRESS *sa, int trafficClass) {
       
   884     if (sa->sa.sa_family == AF_INET6) {
       
   885         sa->sa6.sin6_flowinfo = htonl((trafficClass & 0xff) << 20);
       
   886     }
       
   887 }
       
   888 
       
   889 int
       
   890 NET_IsIPv4Mapped(jbyte* caddr) {
       
   891     int i;
       
   892     for (i = 0; i < 10; i++) {
       
   893         if (caddr[i] != 0x00) {
       
   894             return 0; /* false */
       
   895         }
       
   896     }
       
   897 
       
   898     if (((caddr[10] & 0xff) == 0xff) && ((caddr[11] & 0xff) == 0xff)) {
       
   899         return 1; /* true */
       
   900     }
       
   901     return 0; /* false */
       
   902 }
       
   903 
       
   904 int
       
   905 NET_IPv4MappedToIPv4(jbyte* caddr) {
       
   906     return ((caddr[12] & 0xff) << 24) | ((caddr[13] & 0xff) << 16) | ((caddr[14] & 0xff) << 8)
       
   907         | (caddr[15] & 0xff);
       
   908 }
       
   909 
       
   910 int
       
   911 NET_IsEqual(jbyte* caddr1, jbyte* caddr2) {
       
   912     int i;
       
   913     for (i = 0; i < 16; i++) {
       
   914         if (caddr1[i] != caddr2[i]) {
       
   915             return 0; /* false */
       
   916         }
       
   917     }
       
   918     return 1;
       
   919 }
       
   920 
       
   921 int NET_IsZeroAddr(jbyte* caddr) {
       
   922     int i;
       
   923     for (i = 0; i < 16; i++) {
       
   924         if (caddr[i] != 0) {
       
   925             return 0;
       
   926         }
       
   927     }
       
   928     return 1;
       
   929 }
       
   930 
       
   931 /*
       
   932  * Map the Java level socket option to the platform specific
       
   933  * level and option name.
       
   934  */
       
   935 int
       
   936 NET_MapSocketOption(jint cmd, int *level, int *optname) {
       
   937     static struct {
       
   938         jint cmd;
       
   939         int level;
       
   940         int optname;
       
   941     } const opts[] = {
       
   942         { java_net_SocketOptions_TCP_NODELAY,           IPPROTO_TCP,    TCP_NODELAY },
       
   943         { java_net_SocketOptions_SO_OOBINLINE,          SOL_SOCKET,     SO_OOBINLINE },
       
   944         { java_net_SocketOptions_SO_LINGER,             SOL_SOCKET,     SO_LINGER },
       
   945         { java_net_SocketOptions_SO_SNDBUF,             SOL_SOCKET,     SO_SNDBUF },
       
   946         { java_net_SocketOptions_SO_RCVBUF,             SOL_SOCKET,     SO_RCVBUF },
       
   947         { java_net_SocketOptions_SO_KEEPALIVE,          SOL_SOCKET,     SO_KEEPALIVE },
       
   948         { java_net_SocketOptions_SO_REUSEADDR,          SOL_SOCKET,     SO_REUSEADDR },
       
   949         { java_net_SocketOptions_SO_REUSEPORT,          SOL_SOCKET,     SO_REUSEPORT },
       
   950         { java_net_SocketOptions_SO_BROADCAST,          SOL_SOCKET,     SO_BROADCAST },
       
   951         { java_net_SocketOptions_IP_TOS,                IPPROTO_IP,     IP_TOS },
       
   952         { java_net_SocketOptions_IP_MULTICAST_IF,       IPPROTO_IP,     IP_MULTICAST_IF },
       
   953         { java_net_SocketOptions_IP_MULTICAST_IF2,      IPPROTO_IP,     IP_MULTICAST_IF },
       
   954         { java_net_SocketOptions_IP_MULTICAST_LOOP,     IPPROTO_IP,     IP_MULTICAST_LOOP },
       
   955     };
       
   956 
       
   957     int i;
       
   958 
       
   959     if (ipv6_available()) {
       
   960         switch (cmd) {
       
   961             // Different multicast options if IPv6 is enabled
       
   962             case java_net_SocketOptions_IP_MULTICAST_IF:
       
   963             case java_net_SocketOptions_IP_MULTICAST_IF2:
       
   964                 *level = IPPROTO_IPV6;
       
   965                 *optname = IPV6_MULTICAST_IF;
       
   966                 return 0;
       
   967 
       
   968             case java_net_SocketOptions_IP_MULTICAST_LOOP:
       
   969                 *level = IPPROTO_IPV6;
       
   970                 *optname = IPV6_MULTICAST_LOOP;
       
   971                 return 0;
       
   972 #if (defined(__solaris__) || defined(MACOSX))
       
   973             // Map IP_TOS request to IPV6_TCLASS
       
   974             case java_net_SocketOptions_IP_TOS:
       
   975                 *level = IPPROTO_IPV6;
       
   976                 *optname = IPV6_TCLASS;
       
   977                 return 0;
       
   978 #endif
       
   979         }
       
   980     }
       
   981 
       
   982     /*
       
   983      * Map the Java level option to the native level
       
   984      */
       
   985     for (i=0; i<(int)(sizeof(opts) / sizeof(opts[0])); i++) {
       
   986         if (cmd == opts[i].cmd) {
       
   987             *level = opts[i].level;
       
   988             *optname = opts[i].optname;
       
   989             return 0;
       
   990         }
       
   991     }
       
   992 
       
   993     /* not found */
       
   994     return -1;
       
   995 }
       
   996 
       
   997 /*
       
   998  * Determine the default interface for an IPv6 address.
       
   999  *
       
  1000  * 1. Scans /proc/net/ipv6_route for a matching route
       
  1001  *    (eg: fe80::/10 or a route for the specific address).
       
  1002  *    This will tell us the interface to use (eg: "eth0").
       
  1003  *
       
  1004  * 2. Lookup /proc/net/if_inet6 to map the interface
       
  1005  *    name to an interface index.
       
  1006  *
       
  1007  * Returns :-
       
  1008  *      -1 if error
       
  1009  *       0 if no matching interface
       
  1010  *      >1 interface index to use for the link-local address.
       
  1011  */
       
  1012 #if defined(__linux__)
       
  1013 int getDefaultIPv6Interface(struct in6_addr *target_addr) {
       
  1014     FILE *f;
       
  1015     char srcp[8][5];
       
  1016     char hopp[8][5];
       
  1017     int dest_plen, src_plen, use, refcnt, metric;
       
  1018     unsigned long flags;
       
  1019     char dest_str[40];
       
  1020     struct in6_addr dest_addr;
       
  1021     char device[16];
       
  1022     jboolean match = JNI_FALSE;
       
  1023 
       
  1024     /*
       
  1025      * Scan /proc/net/ipv6_route looking for a matching
       
  1026      * route.
       
  1027      */
       
  1028     if ((f = fopen("/proc/net/ipv6_route", "r")) == NULL) {
       
  1029         return -1;
       
  1030     }
       
  1031     while (fscanf(f, "%4s%4s%4s%4s%4s%4s%4s%4s %02x "
       
  1032                      "%4s%4s%4s%4s%4s%4s%4s%4s %02x "
       
  1033                      "%4s%4s%4s%4s%4s%4s%4s%4s "
       
  1034                      "%08x %08x %08x %08lx %8s",
       
  1035                      dest_str, &dest_str[5], &dest_str[10], &dest_str[15],
       
  1036                      &dest_str[20], &dest_str[25], &dest_str[30], &dest_str[35],
       
  1037                      &dest_plen,
       
  1038                      srcp[0], srcp[1], srcp[2], srcp[3],
       
  1039                      srcp[4], srcp[5], srcp[6], srcp[7],
       
  1040                      &src_plen,
       
  1041                      hopp[0], hopp[1], hopp[2], hopp[3],
       
  1042                      hopp[4], hopp[5], hopp[6], hopp[7],
       
  1043                      &metric, &use, &refcnt, &flags, device) == 31) {
       
  1044 
       
  1045         /*
       
  1046          * Some routes should be ignored
       
  1047          */
       
  1048         if ( (dest_plen < 0 || dest_plen > 128)  ||
       
  1049              (src_plen != 0) ||
       
  1050              (flags & (RTF_POLICY | RTF_FLOW)) ||
       
  1051              ((flags & RTF_REJECT) && dest_plen == 0) ) {
       
  1052             continue;
       
  1053         }
       
  1054 
       
  1055         /*
       
  1056          * Convert the destination address
       
  1057          */
       
  1058         dest_str[4] = ':';
       
  1059         dest_str[9] = ':';
       
  1060         dest_str[14] = ':';
       
  1061         dest_str[19] = ':';
       
  1062         dest_str[24] = ':';
       
  1063         dest_str[29] = ':';
       
  1064         dest_str[34] = ':';
       
  1065         dest_str[39] = '\0';
       
  1066 
       
  1067         if (inet_pton(AF_INET6, dest_str, &dest_addr) < 0) {
       
  1068             /* not an Ipv6 address */
       
  1069             continue;
       
  1070         } else {
       
  1071             /*
       
  1072              * The prefix len (dest_plen) indicates the number of bits we
       
  1073              * need to match on.
       
  1074              *
       
  1075              * dest_plen / 8    => number of bytes to match
       
  1076              * dest_plen % 8    => number of additional bits to match
       
  1077              *
       
  1078              * eg: fe80::/10 => match 1 byte + 2 additional bits in the
       
  1079              *                  the next byte.
       
  1080              */
       
  1081             int byte_count = dest_plen >> 3;
       
  1082             int extra_bits = dest_plen & 0x3;
       
  1083 
       
  1084             if (byte_count > 0) {
       
  1085                 if (memcmp(target_addr, &dest_addr, byte_count)) {
       
  1086                     continue;  /* no match */
       
  1087                 }
       
  1088             }
       
  1089 
       
  1090             if (extra_bits > 0) {
       
  1091                 unsigned char c1 = ((unsigned char *)target_addr)[byte_count];
       
  1092                 unsigned char c2 = ((unsigned char *)&dest_addr)[byte_count];
       
  1093                 unsigned char mask = 0xff << (8 - extra_bits);
       
  1094                 if ((c1 & mask) != (c2 & mask)) {
       
  1095                     continue;
       
  1096                 }
       
  1097             }
       
  1098 
       
  1099             /*
       
  1100              * We have a match
       
  1101              */
       
  1102             match = JNI_TRUE;
       
  1103             break;
       
  1104         }
       
  1105     }
       
  1106     fclose(f);
       
  1107 
       
  1108     /*
       
  1109      * If there's a match then we lookup the interface
       
  1110      * index.
       
  1111      */
       
  1112     if (match) {
       
  1113         char devname[21];
       
  1114         char addr6p[8][5];
       
  1115         int plen, scope, dad_status, if_idx;
       
  1116 
       
  1117         if ((f = fopen("/proc/net/if_inet6", "r")) != NULL) {
       
  1118             while (fscanf(f, "%4s%4s%4s%4s%4s%4s%4s%4s %08x %02x %02x %02x %20s\n",
       
  1119                       addr6p[0], addr6p[1], addr6p[2], addr6p[3],
       
  1120                       addr6p[4], addr6p[5], addr6p[6], addr6p[7],
       
  1121                   &if_idx, &plen, &scope, &dad_status, devname) == 13) {
       
  1122 
       
  1123                 if (strcmp(devname, device) == 0) {
       
  1124                     /*
       
  1125                      * Found - so just return the index
       
  1126                      */
       
  1127                     fclose(f);
       
  1128                     return if_idx;
       
  1129                 }
       
  1130             }
       
  1131             fclose(f);
       
  1132         } else {
       
  1133             /*
       
  1134              * Couldn't open /proc/net/if_inet6
       
  1135              */
       
  1136             return -1;
       
  1137         }
       
  1138     }
       
  1139 
       
  1140     /*
       
  1141      * If we get here it means we didn't there wasn't any
       
  1142      * route or we couldn't get the index of the interface.
       
  1143      */
       
  1144     return 0;
       
  1145 }
       
  1146 #endif
       
  1147 
       
  1148 
       
  1149 /*
       
  1150  * Wrapper for getsockopt system routine - does any necessary
       
  1151  * pre/post processing to deal with OS specific oddities :-
       
  1152  *
       
  1153  * On Linux the SO_SNDBUF/SO_RCVBUF values must be post-processed
       
  1154  * to compensate for an incorrect value returned by the kernel.
       
  1155  */
       
  1156 int
       
  1157 NET_GetSockOpt(int fd, int level, int opt, void *result,
       
  1158                int *len)
       
  1159 {
       
  1160     int rv;
       
  1161     socklen_t socklen = *len;
       
  1162 
       
  1163     rv = getsockopt(fd, level, opt, result, &socklen);
       
  1164     *len = socklen;
       
  1165 
       
  1166     if (rv < 0) {
       
  1167         return rv;
       
  1168     }
       
  1169 
       
  1170 #ifdef __linux__
       
  1171     /*
       
  1172      * On Linux SO_SNDBUF/SO_RCVBUF aren't symmetric. This
       
  1173      * stems from additional socket structures in the send
       
  1174      * and receive buffers.
       
  1175      */
       
  1176     if ((level == SOL_SOCKET) && ((opt == SO_SNDBUF)
       
  1177                                   || (opt == SO_RCVBUF))) {
       
  1178         int n = *((int *)result);
       
  1179         n /= 2;
       
  1180         *((int *)result) = n;
       
  1181     }
       
  1182 #endif
       
  1183 
       
  1184 /* Workaround for Mac OS treating linger value as
       
  1185  *  signed integer
       
  1186  */
       
  1187 #ifdef MACOSX
       
  1188     if (level == SOL_SOCKET && opt == SO_LINGER) {
       
  1189         struct linger* to_cast = (struct linger*)result;
       
  1190         to_cast->l_linger = (unsigned short)to_cast->l_linger;
       
  1191     }
       
  1192 #endif
       
  1193     return rv;
       
  1194 }
       
  1195 
       
  1196 /*
       
  1197  * Wrapper for setsockopt system routine - performs any
       
  1198  * necessary pre/post processing to deal with OS specific
       
  1199  * issue :-
       
  1200  *
       
  1201  * On Solaris need to limit the suggested value for SO_SNDBUF
       
  1202  * and SO_RCVBUF to the kernel configured limit
       
  1203  *
       
  1204  * For IP_TOS socket option need to mask off bits as this
       
  1205  * aren't automatically masked by the kernel and results in
       
  1206  * an error.
       
  1207  */
       
  1208 int
       
  1209 NET_SetSockOpt(int fd, int level, int  opt, const void *arg,
       
  1210                int len)
       
  1211 {
       
  1212 
       
  1213 #ifndef IPTOS_TOS_MASK
       
  1214 #define IPTOS_TOS_MASK 0x1e
       
  1215 #endif
       
  1216 #ifndef IPTOS_PREC_MASK
       
  1217 #define IPTOS_PREC_MASK 0xe0
       
  1218 #endif
       
  1219 
       
  1220 #if defined(_ALLBSD_SOURCE)
       
  1221 #if defined(KIPC_MAXSOCKBUF)
       
  1222     int mib[3];
       
  1223     size_t rlen;
       
  1224 #endif
       
  1225 
       
  1226     int *bufsize;
       
  1227 
       
  1228 #ifdef __APPLE__
       
  1229     static int maxsockbuf = -1;
       
  1230 #else
       
  1231     static long maxsockbuf = -1;
       
  1232 #endif
       
  1233 #endif
       
  1234 
       
  1235     /*
       
  1236      * IPPROTO/IP_TOS :-
       
  1237      * 1. IPv6 on Solaris/Mac OS:
       
  1238      *    Set the TOS OR Traffic Class value to cater for
       
  1239      *    IPv6 and IPv4 scenarios.
       
  1240      * 2. IPv6 on Linux: By default Linux ignores flowinfo
       
  1241      *    field so enable IPV6_FLOWINFO_SEND so that flowinfo
       
  1242      *    will be examined. We also set the IPv4 TOS option in this case.
       
  1243      * 3. IPv4: set socket option based on ToS and Precedence
       
  1244      *    fields (otherwise get invalid argument)
       
  1245      */
       
  1246     if (level == IPPROTO_IP && opt == IP_TOS) {
       
  1247         int *iptos;
       
  1248 
       
  1249 #if defined(__linux__)
       
  1250         if (ipv6_available()) {
       
  1251             int optval = 1;
       
  1252             if (setsockopt(fd, IPPROTO_IPV6, IPV6_FLOWINFO_SEND,
       
  1253                            (void *)&optval, sizeof(optval)) < 0) {
       
  1254                 return -1;
       
  1255             }
       
  1256            /*
       
  1257             * Let's also set the IPV6_TCLASS flag.
       
  1258             * Linux appears to allow both IP_TOS and IPV6_TCLASS to be set
       
  1259             * This helps in mixed environments where IPv4 and IPv6 sockets
       
  1260             * are connecting.
       
  1261             */
       
  1262            if (setsockopt(fd, IPPROTO_IPV6, IPV6_TCLASS,
       
  1263                            arg, len) < 0) {
       
  1264                 return -1;
       
  1265             }
       
  1266         }
       
  1267 #endif
       
  1268 
       
  1269         iptos = (int *)arg;
       
  1270         *iptos &= (IPTOS_TOS_MASK | IPTOS_PREC_MASK);
       
  1271     }
       
  1272 
       
  1273     /*
       
  1274      * SOL_SOCKET/{SO_SNDBUF,SO_RCVBUF} - On Solaris we may need to clamp
       
  1275      * the value when it exceeds the system limit.
       
  1276      */
       
  1277 #ifdef __solaris__
       
  1278     if (level == SOL_SOCKET) {
       
  1279         if (opt == SO_SNDBUF || opt == SO_RCVBUF) {
       
  1280             int sotype=0;
       
  1281             socklen_t arglen;
       
  1282             int *bufsize, maxbuf;
       
  1283             int ret;
       
  1284 
       
  1285             /* Attempt with the original size */
       
  1286             ret = setsockopt(fd, level, opt, arg, len);
       
  1287             if ((ret == 0) || (ret == -1 && errno != ENOBUFS))
       
  1288                 return ret;
       
  1289 
       
  1290             /* Exceeded system limit so clamp and retry */
       
  1291 
       
  1292             arglen = sizeof(sotype);
       
  1293             if (getsockopt(fd, SOL_SOCKET, SO_TYPE, (void *)&sotype,
       
  1294                            &arglen) < 0) {
       
  1295                 return -1;
       
  1296             }
       
  1297 
       
  1298             /*
       
  1299              * We try to get tcp_maxbuf (and udp_max_buf) using
       
  1300              * an ioctl() that isn't available on all versions of Solaris.
       
  1301              * If that fails, we use the search algorithm in findMaxBuf()
       
  1302              */
       
  1303             if (!init_tcp_max_buf && sotype == SOCK_STREAM) {
       
  1304                 tcp_max_buf = net_getParam("/dev/tcp", "tcp_max_buf");
       
  1305                 if (tcp_max_buf == -1) {
       
  1306                     tcp_max_buf = findMaxBuf(fd, opt, SOCK_STREAM);
       
  1307                     if (tcp_max_buf == -1) {
       
  1308                         return -1;
       
  1309                     }
       
  1310                 }
       
  1311                 init_tcp_max_buf = 1;
       
  1312             } else if (!init_udp_max_buf && sotype == SOCK_DGRAM) {
       
  1313                 udp_max_buf = net_getParam("/dev/udp", "udp_max_buf");
       
  1314                 if (udp_max_buf == -1) {
       
  1315                     udp_max_buf = findMaxBuf(fd, opt, SOCK_DGRAM);
       
  1316                     if (udp_max_buf == -1) {
       
  1317                         return -1;
       
  1318                     }
       
  1319                 }
       
  1320                 init_udp_max_buf = 1;
       
  1321             }
       
  1322 
       
  1323             maxbuf = (sotype == SOCK_STREAM) ? tcp_max_buf : udp_max_buf;
       
  1324             bufsize = (int *)arg;
       
  1325             if (*bufsize > maxbuf) {
       
  1326                 *bufsize = maxbuf;
       
  1327             }
       
  1328         }
       
  1329     }
       
  1330 #endif
       
  1331 
       
  1332 #ifdef _AIX
       
  1333     if (level == SOL_SOCKET) {
       
  1334         if (opt == SO_SNDBUF || opt == SO_RCVBUF) {
       
  1335             /*
       
  1336              * Just try to set the requested size. If it fails we will leave the
       
  1337              * socket option as is. Setting the buffer size means only a hint in
       
  1338              * the jse2/java software layer, see javadoc. In the previous
       
  1339              * solution the buffer has always been truncated to a length of
       
  1340              * 0x100000 Byte, even if the technical limit has not been reached.
       
  1341              * This kind of absolute truncation was unexpected in the jck tests.
       
  1342              */
       
  1343             int ret = setsockopt(fd, level, opt, arg, len);
       
  1344             if ((ret == 0) || (ret == -1 && errno == ENOBUFS)) {
       
  1345                 // Accept failure because of insufficient buffer memory resources.
       
  1346                 return 0;
       
  1347             } else {
       
  1348                 // Deliver all other kinds of errors.
       
  1349                 return ret;
       
  1350             }
       
  1351         }
       
  1352     }
       
  1353 #endif
       
  1354 
       
  1355     /*
       
  1356      * On Linux the receive buffer is used for both socket
       
  1357      * structures and the packet payload. The implication
       
  1358      * is that if SO_RCVBUF is too small then small packets
       
  1359      * must be discarded.
       
  1360      */
       
  1361 #ifdef __linux__
       
  1362     if (level == SOL_SOCKET && opt == SO_RCVBUF) {
       
  1363         int *bufsize = (int *)arg;
       
  1364         if (*bufsize < 1024) {
       
  1365             *bufsize = 1024;
       
  1366         }
       
  1367     }
       
  1368 #endif
       
  1369 
       
  1370 #if defined(_ALLBSD_SOURCE)
       
  1371     /*
       
  1372      * SOL_SOCKET/{SO_SNDBUF,SO_RCVBUF} - On FreeBSD need to
       
  1373      * ensure that value is <= kern.ipc.maxsockbuf as otherwise we get
       
  1374      * an ENOBUFS error.
       
  1375      */
       
  1376     if (level == SOL_SOCKET) {
       
  1377         if (opt == SO_SNDBUF || opt == SO_RCVBUF) {
       
  1378 #ifdef KIPC_MAXSOCKBUF
       
  1379             if (maxsockbuf == -1) {
       
  1380                mib[0] = CTL_KERN;
       
  1381                mib[1] = KERN_IPC;
       
  1382                mib[2] = KIPC_MAXSOCKBUF;
       
  1383                rlen = sizeof(maxsockbuf);
       
  1384                if (sysctl(mib, 3, &maxsockbuf, &rlen, NULL, 0) == -1)
       
  1385                    maxsockbuf = 1024;
       
  1386 
       
  1387 #if 1
       
  1388                /* XXXBSD: This is a hack to workaround mb_max/mb_max_adj
       
  1389                   problem.  It should be removed when kern.ipc.maxsockbuf
       
  1390                   will be real value. */
       
  1391                maxsockbuf = (maxsockbuf/5)*4;
       
  1392 #endif
       
  1393            }
       
  1394 #elif defined(__OpenBSD__)
       
  1395            maxsockbuf = SB_MAX;
       
  1396 #else
       
  1397            maxsockbuf = 64 * 1024;      /* XXX: NetBSD */
       
  1398 #endif
       
  1399 
       
  1400            bufsize = (int *)arg;
       
  1401            if (*bufsize > maxsockbuf) {
       
  1402                *bufsize = maxsockbuf;
       
  1403            }
       
  1404 
       
  1405            if (opt == SO_RCVBUF && *bufsize < 1024) {
       
  1406                 *bufsize = 1024;
       
  1407            }
       
  1408 
       
  1409         }
       
  1410     }
       
  1411 #endif
       
  1412 
       
  1413 #if defined(_ALLBSD_SOURCE) || defined(_AIX)
       
  1414     /*
       
  1415      * On Solaris, SO_REUSEADDR will allow multiple datagram
       
  1416      * sockets to bind to the same port. The network jck tests check
       
  1417      * for this "feature", so we need to emulate it by turning on
       
  1418      * SO_REUSEPORT as well for that combination.
       
  1419      */
       
  1420     if (level == SOL_SOCKET && opt == SO_REUSEADDR) {
       
  1421         int sotype;
       
  1422         socklen_t arglen;
       
  1423 
       
  1424         arglen = sizeof(sotype);
       
  1425         if (getsockopt(fd, SOL_SOCKET, SO_TYPE, (void *)&sotype, &arglen) < 0) {
       
  1426             return -1;
       
  1427         }
       
  1428 
       
  1429         if (sotype == SOCK_DGRAM) {
       
  1430             setsockopt(fd, level, SO_REUSEPORT, arg, len);
       
  1431         }
       
  1432     }
       
  1433 #endif
       
  1434 
       
  1435     return setsockopt(fd, level, opt, arg, len);
       
  1436 }
       
  1437 
       
  1438 /*
       
  1439  * Wrapper for bind system call - performs any necessary pre/post
       
  1440  * processing to deal with OS specific issues :-
       
  1441  *
       
  1442  * Linux allows a socket to bind to 127.0.0.255 which must be
       
  1443  * caught.
       
  1444  *
       
  1445  * On Solaris with IPv6 enabled we must use an exclusive
       
  1446  * bind to guarantee a unique port number across the IPv4 and
       
  1447  * IPv6 port spaces.
       
  1448  *
       
  1449  */
       
  1450 int
       
  1451 NET_Bind(int fd, SOCKETADDRESS *sa, int len)
       
  1452 {
       
  1453 #if defined(__solaris__)
       
  1454     int level = -1;
       
  1455     int exclbind = -1;
       
  1456 #endif
       
  1457     int rv;
       
  1458     int arg, alen;
       
  1459 
       
  1460 #ifdef __linux__
       
  1461     /*
       
  1462      * ## get bugId for this issue - goes back to 1.2.2 port ##
       
  1463      * ## When IPv6 is enabled this will be an IPv4-mapped
       
  1464      * ## with family set to AF_INET6
       
  1465      */
       
  1466     if (sa->sa.sa_family == AF_INET) {
       
  1467         if ((ntohl(sa->sa4.sin_addr.s_addr) & 0x7f0000ff) == 0x7f0000ff) {
       
  1468             errno = EADDRNOTAVAIL;
       
  1469             return -1;
       
  1470         }
       
  1471     }
       
  1472 #endif
       
  1473 
       
  1474 #if defined(__solaris__)
       
  1475     /*
       
  1476      * Solaris has separate IPv4 and IPv6 port spaces so we
       
  1477      * use an exclusive bind when SO_REUSEADDR is not used to
       
  1478      * give the illusion of a unified port space.
       
  1479      * This also avoids problems with IPv6 sockets connecting
       
  1480      * to IPv4 mapped addresses whereby the socket conversion
       
  1481      * results in a late bind that fails because the
       
  1482      * corresponding IPv4 port is in use.
       
  1483      */
       
  1484     alen = sizeof(arg);
       
  1485 
       
  1486     if (useExclBind ||
       
  1487         getsockopt(fd, SOL_SOCKET, SO_REUSEADDR, (char *)&arg, &alen) == 0)
       
  1488     {
       
  1489         if (useExclBind || arg == 0) {
       
  1490             /*
       
  1491              * SO_REUSEADDR is disabled or sun.net.useExclusiveBind
       
  1492              * property is true so enable TCP_EXCLBIND or
       
  1493              * UDP_EXCLBIND
       
  1494              */
       
  1495             alen = sizeof(arg);
       
  1496             if (getsockopt(fd, SOL_SOCKET, SO_TYPE, (char *)&arg, &alen) == 0)
       
  1497             {
       
  1498                 if (arg == SOCK_STREAM) {
       
  1499                     level = IPPROTO_TCP;
       
  1500                     exclbind = TCP_EXCLBIND;
       
  1501                 } else {
       
  1502                     level = IPPROTO_UDP;
       
  1503                     exclbind = UDP_EXCLBIND;
       
  1504                 }
       
  1505             }
       
  1506 
       
  1507             arg = 1;
       
  1508             setsockopt(fd, level, exclbind, (char *)&arg, sizeof(arg));
       
  1509         }
       
  1510     }
       
  1511 
       
  1512 #endif
       
  1513 
       
  1514     rv = bind(fd, &sa->sa, len);
       
  1515 
       
  1516 #if defined(__solaris__)
       
  1517     if (rv < 0) {
       
  1518         int en = errno;
       
  1519         /* Restore *_EXCLBIND if the bind fails */
       
  1520         if (exclbind != -1) {
       
  1521             int arg = 0;
       
  1522             setsockopt(fd, level, exclbind, (char *)&arg,
       
  1523                        sizeof(arg));
       
  1524         }
       
  1525         errno = en;
       
  1526     }
       
  1527 #endif
       
  1528 
       
  1529     return rv;
       
  1530 }
       
  1531 
       
  1532 /**
       
  1533  * Wrapper for poll with timeout on a single file descriptor.
       
  1534  *
       
  1535  * flags (defined in net_util_md.h can be any combination of
       
  1536  * NET_WAIT_READ, NET_WAIT_WRITE & NET_WAIT_CONNECT.
       
  1537  *
       
  1538  * The function will return when either the socket is ready for one
       
  1539  * of the specified operations or the timeout expired.
       
  1540  *
       
  1541  * It returns the time left from the timeout (possibly 0), or -1 if it expired.
       
  1542  */
       
  1543 
       
  1544 jint
       
  1545 NET_Wait(JNIEnv *env, jint fd, jint flags, jint timeout)
       
  1546 {
       
  1547     jlong prevNanoTime = JVM_NanoTime(env, 0);
       
  1548     jlong nanoTimeout = (jlong) timeout * NET_NSEC_PER_MSEC;
       
  1549     jint read_rv;
       
  1550 
       
  1551     while (1) {
       
  1552         jlong newNanoTime;
       
  1553         struct pollfd pfd;
       
  1554         pfd.fd = fd;
       
  1555         pfd.events = 0;
       
  1556         if (flags & NET_WAIT_READ)
       
  1557           pfd.events |= POLLIN;
       
  1558         if (flags & NET_WAIT_WRITE)
       
  1559           pfd.events |= POLLOUT;
       
  1560         if (flags & NET_WAIT_CONNECT)
       
  1561           pfd.events |= POLLOUT;
       
  1562 
       
  1563         errno = 0;
       
  1564         read_rv = NET_Poll(&pfd, 1, nanoTimeout / NET_NSEC_PER_MSEC);
       
  1565 
       
  1566         newNanoTime = JVM_NanoTime(env, 0);
       
  1567         nanoTimeout -= (newNanoTime - prevNanoTime);
       
  1568         if (nanoTimeout < NET_NSEC_PER_MSEC) {
       
  1569           return read_rv > 0 ? 0 : -1;
       
  1570         }
       
  1571         prevNanoTime = newNanoTime;
       
  1572 
       
  1573         if (read_rv > 0) {
       
  1574           break;
       
  1575         }
       
  1576       } /* while */
       
  1577     return (nanoTimeout / NET_NSEC_PER_MSEC);
       
  1578 }