java/lang/UNIXProcess_md.c

/*
 * Copyright (c) 1995, 2008, Oracle and/or its affiliates. All rights reserved.
 * DO NOT ALTER OR REMOVE COPYRIGHT NOTICES OR THIS FILE HEADER.
 *
 * This code is free software; you can redistribute it and/or modify it
 * under the terms of the GNU General Public License version 2 only, as
 * published by the Free Software Foundation.  Oracle designates this
 * particular file as subject to the "Classpath" exception as provided
 * by Oracle in the LICENSE file that accompanied this code.
 *
 * This code is distributed in the hope that it will be useful, but WITHOUT
 * ANY WARRANTY; without even the implied warranty of MERCHANTABILITY or
 * FITNESS FOR A PARTICULAR PURPOSE.  See the GNU General Public License
 * version 2 for more details (a copy is included in the LICENSE file that
 * accompanied this code).
 *
 * You should have received a copy of the GNU General Public License version
 * 2 along with this work; if not, write to the Free Software Foundation,
 * Inc., 51 Franklin St, Fifth Floor, Boston, MA 02110-1301 USA.
 *
 * Please contact Oracle, 500 Oracle Parkway, Redwood Shores, CA 94065 USA
 * or visit www.oracle.com if you need additional information or have any
 * questions.
 */

#undef  _LARGEFILE64_SOURCE
#define _LARGEFILE64_SOURCE 1

#include "jni.h"
#include "jvm.h"
#include "jvm_md.h"
#include "jni_util.h"
#include "io_util.h"

/*
 * Platform-specific support for java.lang.Process
 */
#include <assert.h>
#include <stddef.h>
#include <stdlib.h>
#include <sys/types.h>
#include <ctype.h>
#ifdef _ALLBSD_SOURCE
#include <sys/wait.h>
#else
#include <wait.h>
#endif
#include <signal.h>
#include <string.h>
#include <errno.h>
#include <dirent.h>
#include <unistd.h>
#include <fcntl.h>
#include <limits.h>

#ifdef __APPLE__
#include <crt_externs.h>
#define environ (*_NSGetEnviron())
#endif

/*
 * There are 3 possible strategies we might use to "fork":
 *
 * - fork(2).  Very portable and reliable but subject to
 *   failure due to overcommit (see the documentation on
 *   /proc/sys/vm/overcommit_memory in Linux proc(5)).
 *   This is the ancient problem of spurious failure whenever a large
 *   process starts a small subprocess.
 *
 * - vfork().  Using this is scary because all relevant man pages
 *   contain dire warnings, e.g. Linux vfork(2).  But at least it's
 *   documented in the glibc docs and is standardized by XPG4.
 *   http://www.opengroup.org/onlinepubs/000095399/functions/vfork.html
 *   On Linux, one might think that vfork() would be implemented using
 *   the clone system call with flag CLONE_VFORK, but in fact vfork is
 *   a separate system call (which is a good sign, suggesting that
 *   vfork will continue to be supported at least on Linux).
 *   Another good sign is that glibc implements posix_spawn using
 *   vfork whenever possible.  Note that we cannot use posix_spawn
 *   ourselves because there's no reliable way to close all inherited
 *   file descriptors.
 *
 * - clone() with flags CLONE_VM but not CLONE_THREAD.  clone() is
 *   Linux-specific, but this ought to work - at least the glibc
 *   sources contain code to handle different combinations of CLONE_VM
 *   and CLONE_THREAD.  However, when this was implemented, it
 *   appeared to fail on 32-bit i386 (but not 64-bit x86_64) Linux with
 *   the simple program
 *     Runtime.getRuntime().exec("/bin/true").waitFor();
 *   with:
 *     #  Internal Error (os_linux_x86.cpp:683), pid=19940, tid=2934639536
 *     #  Error: pthread_getattr_np failed with errno = 3 (ESRCH)
 *   We believe this is a glibc bug, reported here:
 *     http://sources.redhat.com/bugzilla/show_bug.cgi?id=10311
 *   but the glibc maintainers closed it as WONTFIX.
 *
 * Based on the above analysis, we are currently using vfork() on
 * Linux and fork() on other Unix systems, but the code to use clone()
 * remains.
 */

#define START_CHILD_USE_CLONE 0  /* clone() currently disabled; see above. */

#ifndef START_CHILD_USE_CLONE
  #ifdef __linux__
    #define START_CHILD_USE_CLONE 1
  #else
    #define START_CHILD_USE_CLONE 0
  #endif
#endif

/* By default, use vfork() on Linux. */
#ifndef START_CHILD_USE_VFORK
  #ifdef __linux__
    #define START_CHILD_USE_VFORK 1
  #else
    #define START_CHILD_USE_VFORK 0
  #endif
#endif

#if START_CHILD_USE_CLONE
#include <sched.h>
#define START_CHILD_SYSTEM_CALL "clone"
#elif START_CHILD_USE_VFORK
#define START_CHILD_SYSTEM_CALL "vfork"
#else
#define START_CHILD_SYSTEM_CALL "fork"
#endif

#ifndef STDIN_FILENO
#define STDIN_FILENO 0
#endif

#ifndef STDOUT_FILENO
#define STDOUT_FILENO 1
#endif

#ifndef STDERR_FILENO
#define STDERR_FILENO 2
#endif

#ifndef SA_NOCLDSTOP
#define SA_NOCLDSTOP 0
#endif

#ifndef SA_RESTART
#define SA_RESTART 0
#endif

#define FAIL_FILENO (STDERR_FILENO + 1)

/* TODO: Refactor. */
#define RESTARTABLE(_cmd, _result) do { \
  do { \
    _result = _cmd; \
  } while((_result == -1) && (errno == EINTR)); \
} while(0)

/* This is one of the rare times it's more portable to declare an
 * external symbol explicitly, rather than via a system header.
 * The declaration is standardized as part of UNIX98, but there is
 * no standard (not even de-facto) header file where the
 * declaration is to be found.  See:
 * http://www.opengroup.org/onlinepubs/009695399/functions/environ.html
 * http://www.opengroup.org/onlinepubs/009695399/functions/xsh_chap02_02.html
 *
 * "All identifiers in this volume of IEEE Std 1003.1-2001, except
 * environ, are defined in at least one of the headers" (!)
 */
extern char **environ;


static void
setSIGCHLDHandler(JNIEnv *env)
{
    /* There is a subtle difference between having the signal handler
     * for SIGCHLD be SIG_DFL and SIG_IGN.  We cannot obtain process
     * termination information for child processes if the signal
     * handler is SIG_IGN.  It must be SIG_DFL.
     *
     * We used to set the SIGCHLD handler only on Linux, but it's
     * safest to set it unconditionally.
     *
     * Consider what happens if java's parent process sets the SIGCHLD
     * handler to SIG_IGN.  Normally signal handlers are inherited by
     * children, but SIGCHLD is a controversial case.  Solaris appears
     * to always reset it to SIG_DFL, but this behavior may be
     * non-standard-compliant, and we shouldn't rely on it.
     *
     * References:
     * http://www.opengroup.org/onlinepubs/7908799/xsh/exec.html
     * http://www.pasc.org/interps/unofficial/db/p1003.1/pasc-1003.1-132.html
     */
    struct sigaction sa;
    sa.sa_handler = SIG_DFL;
    sigemptyset(&sa.sa_mask);
    sa.sa_flags = SA_NOCLDSTOP | SA_RESTART;
    if (sigaction(SIGCHLD, &sa, NULL) < 0)
        JNU_ThrowInternalError(env, "Can't set SIGCHLD handler");
}

static void*
xmalloc(JNIEnv *env, size_t size)
{
    void *p = malloc(size);
    if (p == NULL)
        JNU_ThrowOutOfMemoryError(env, NULL);
    return p;
}

#define NEW(type, n) ((type *) xmalloc(env, (n) * sizeof(type)))

/**
 * If PATH is not defined, the OS provides some default value.
 * Unfortunately, there's no portable way to get this value.
 * Fortunately, it's only needed if the child has PATH while we do not.
 */
static const char*
defaultPath(void)
{
#ifdef __solaris__
    /* These really are the Solaris defaults! */
    return (geteuid() == 0 || getuid() == 0) ?
        "/usr/xpg4/bin:/usr/ccs/bin:/usr/bin:/opt/SUNWspro/bin:/usr/sbin" :
        "/usr/xpg4/bin:/usr/ccs/bin:/usr/bin:/opt/SUNWspro/bin:";
#else
    return ":/bin:/usr/bin";    /* glibc */
#endif
}

static const char*
effectivePath(void)
{
    const char *s = getenv("PATH");
    return (s != NULL) ? s : defaultPath();
}

static int
countOccurrences(const char *s, char c)
{
    int count;
    for (count = 0; *s != '\0'; s++)
        count += (*s == c);
    return count;
}

static const char * const *
splitPath(JNIEnv *env, const char *path)
{
    const char *p, *q;
    char **pathv;
    int i;
    int count = countOccurrences(path, ':') + 1;

    pathv = NEW(char*, count+1);
    pathv[count] = NULL;
    for (p = path, i = 0; i < count; i++, p = q + 1) {
        for (q = p; (*q != ':') && (*q != '\0'); q++)
            ;
        if (q == p)             /* empty PATH component => "." */
            pathv[i] = "./";
        else {
            int addSlash = ((*(q - 1)) != '/');
            pathv[i] = NEW(char, q - p + addSlash + 1);
            memcpy(pathv[i], p, q - p);
            if (addSlash)
                pathv[i][q - p] = '/';
            pathv[i][q - p + addSlash] = '\0';
        }
    }
    return (const char * const *) pathv;
}

/**
 * Cached value of JVM's effective PATH.
 * (We don't support putenv("PATH=...") in native code)
 */
static const char *parentPath;

/**
 * Split, canonicalized version of parentPath
 */
static const char * const *parentPathv;

static jfieldID field_exitcode;

JNIEXPORT void JNICALL
Java_java_lang_UNIXProcess_initIDs(JNIEnv *env, jclass clazz)
{
    field_exitcode = (*env)->GetFieldID(env, clazz, "exitcode", "I");

    parentPath  = effectivePath();
    parentPathv = splitPath(env, parentPath);

    setSIGCHLDHandler(env);
}


#ifndef WIFEXITED
#define WIFEXITED(status) (((status)&0xFF) == 0)
#endif

#ifndef WEXITSTATUS
#define WEXITSTATUS(status) (((status)>>8)&0xFF)
#endif

#ifndef WIFSIGNALED
#define WIFSIGNALED(status) (((status)&0xFF) > 0 && ((status)&0xFF00) == 0)
#endif

#ifndef WTERMSIG
#define WTERMSIG(status) ((status)&0x7F)
#endif

/* Block until a child process exits and return its exit code.
   Note, can only be called once for any given pid. */
JNIEXPORT jint JNICALL
Java_java_lang_UNIXProcess_waitForProcessExit(JNIEnv* env,
                                              jobject junk,
                                              jint pid)
{
    /* We used to use waitid() on Solaris, waitpid() on Linux, but
     * waitpid() is more standard, so use it on all POSIX platforms. */
    int status;
    /* Wait for the child process to exit.  This returns immediately if
       the child has already exited. */
    while (waitpid(pid, &status, 0) < 0) {
        switch (errno) {
        case ECHILD: return 0;
        case EINTR: break;
        default: return -1;
        }
    }

    if (WIFEXITED(status)) {
        /*
         * The child exited normally; get its exit code.
         */
        return WEXITSTATUS(status);
    } else if (WIFSIGNALED(status)) {
        /* The child exited because of a signal.
         * The best value to return is 0x80 + signal number,
         * because that is what all Unix shells do, and because
         * it allows callers to distinguish between process exit and
         * process death by signal.
         * Unfortunately, the historical behavior on Solaris is to return
         * the signal number, and we preserve this for compatibility. */
#ifdef __solaris__
        return WTERMSIG(status);
#else
        return 0x80 + WTERMSIG(status);
#endif
    } else {
        /*
         * Unknown exit code; pass it through.
         */
        return status;
    }
}

static ssize_t
restartableWrite(int fd, const void *buf, size_t count)
{
    ssize_t result;
    RESTARTABLE(write(fd, buf, count), result);
    return result;
}

static int
restartableDup2(int fd_from, int fd_to)
{
    int err;
    RESTARTABLE(dup2(fd_from, fd_to), err);
    return err;
}

static int
restartableClose(int fd)
{
    int err;
    RESTARTABLE(close(fd), err);
    return err;
}

static int
closeSafely(int fd)
{
    return (fd == -1) ? 0 : restartableClose(fd);
}

static int
isAsciiDigit(char c)
{
  return c >= '0' && c <= '9';
}

#ifdef _ALLBSD_SOURCE
#define FD_DIR "/dev/fd"
#define dirent64 dirent
#define readdir64 readdir
#else
#define FD_DIR "/proc/self/fd"
#endif

static int
closeDescriptors(void)
{
    DIR *dp;
    struct dirent64 *dirp;
    int from_fd = FAIL_FILENO + 1;

    /* We're trying to close all file descriptors, but opendir() might
     * itself be implemented using a file descriptor, and we certainly
     * don't want to close that while it's in use.  We assume that if
     * opendir() is implemented using a file descriptor, then it uses
     * the lowest numbered file descriptor, just like open().  So we
     * close a couple explicitly.  */

    restartableClose(from_fd);          /* for possible use by opendir() */
    restartableClose(from_fd + 1);      /* another one for good luck */

    if ((dp = opendir(FD_DIR)) == NULL)
        return 0;

    /* We use readdir64 instead of readdir to work around Solaris bug
     * 6395699: /proc/self/fd fails to report file descriptors >= 1024 on Solaris 9
     */
    while ((dirp = readdir64(dp)) != NULL) {
        int fd;
        if (isAsciiDigit(dirp->d_name[0]) &&
            (fd = strtol(dirp->d_name, NULL, 10)) >= from_fd + 2)
            restartableClose(fd);
    }

    closedir(dp);

    return 1;
}

static int
moveDescriptor(int fd_from, int fd_to)
{
    if (fd_from != fd_to) {
        if ((restartableDup2(fd_from, fd_to) == -1) ||
            (restartableClose(fd_from) == -1))
            return -1;
    }
    return 0;
}

static const char *
getBytes(JNIEnv *env, jbyteArray arr)
{
    return arr == NULL ? NULL :
        (const char*) (*env)->GetByteArrayElements(env, arr, NULL);
}

static void
releaseBytes(JNIEnv *env, jbyteArray arr, const char* parr)
{
    if (parr != NULL)
        (*env)->ReleaseByteArrayElements(env, arr, (jbyte*) parr, JNI_ABORT);
}

static void
initVectorFromBlock(const char**vector, const char* block, int count)
{
    int i;
    const char *p;
    for (i = 0, p = block; i < count; i++) {
        /* Invariant: p always points to the start of a C string. */
        vector[i] = p;
        while (*(p++));
    }
    vector[count] = NULL;
}

static void
throwIOException(JNIEnv *env, int errnum, const char *defaultDetail)
{
    static const char * const format = "error=%d, %s";
    const char *detail = defaultDetail;
    char *errmsg;
    jstring s;

    if (errnum != 0) {
        const char *s = strerror(errnum);
        if (strcmp(s, "Unknown error") != 0)
            detail = s;
    }
    /* ASCII Decimal representation uses 2.4 times as many bits as binary. */
    errmsg = NEW(char, strlen(format) + strlen(detail) + 3 * sizeof(errnum));
    sprintf(errmsg, format, errnum, detail);
    s = JNU_NewStringPlatform(env, errmsg);
    if (s != NULL) {
        jobject x = JNU_NewObjectByName(env, "java/io/IOException",
                                        "(Ljava/lang/String;)V", s);
        if (x != NULL)
            (*env)->Throw(env, x);
    }
    free(errmsg);
}

#ifdef DEBUG_PROCESS
/* Debugging process code is difficult; where to write debug output? */
static void
debugPrint(char *format, ...)
{
    FILE *tty = fopen("/dev/tty", "w");
    va_list ap;
    va_start(ap, format);
    vfprintf(tty, format, ap);
    va_end(ap);
    fclose(tty);
}
#endif /* DEBUG_PROCESS */

/**
 * Exec FILE as a traditional Bourne shell script (i.e. one without #!).
 * If we could do it over again, we would probably not support such an ancient
 * misfeature, but compatibility wins over sanity.  The original support for
 * this was imported accidentally from execvp().
 */
static void
execve_as_traditional_shell_script(const char *file,
                                   const char *argv[],
                                   const char *const envp[])
{
    /* Use the extra word of space provided for us in argv by caller. */
    const char *argv0 = argv[0];
    const char *const *end = argv;
    while (*end != NULL)
        ++end;
    memmove(argv+2, argv+1, (end-argv) * sizeof (*end));
    argv[0] = "/bin/sh";
    argv[1] = file;
    execve(argv[0], (char **) argv, (char **) envp);
    /* Can't even exec /bin/sh?  Big trouble, but let's soldier on... */
    memmove(argv+1, argv+2, (end-argv) * sizeof (*end));
    argv[0] = argv0;
}

/**
 * Like execve(2), except that in case of ENOEXEC, FILE is assumed to
 * be a shell script and the system default shell is invoked to run it.
 */
static void
execve_with_shell_fallback(const char *file,
                           const char *argv[],
                           const char *const envp[])
{
#if START_CHILD_USE_CLONE || START_CHILD_USE_VFORK
    /* shared address space; be very careful. */
    execve(file, (char **) argv, (char **) envp);
    if (errno == ENOEXEC)
        execve_as_traditional_shell_script(file, argv, envp);
#else
    /* unshared address space; we can mutate environ. */
    environ = (char **) envp;
    execvp(file, (char **) argv);
#endif
}

/**
 * 'execvpe' should have been included in the Unix standards,
 * and is a GNU extension in glibc 2.10.
 *
 * JDK_execvpe is identical to execvp, except that the child environment is
 * specified via the 3rd argument instead of being inherited from environ.
 */
static void
JDK_execvpe(const char *file,
            const char *argv[],
            const char *const envp[])
{
    if (envp == NULL || (char **) envp == environ) {
        execvp(file, (char **) argv);
        return;
    }

    if (*file == '\0') {
        errno = ENOENT;
        return;
    }

    if (strchr(file, '/') != NULL) {
        execve_with_shell_fallback(file, argv, envp);
    } else {
        /* We must search PATH (parent's, not child's) */
        char expanded_file[PATH_MAX];
        int filelen = strlen(file);
        int sticky_errno = 0;
        const char * const * dirs;
        for (dirs = parentPathv; *dirs; dirs++) {
            const char * dir = *dirs;
            int dirlen = strlen(dir);
            if (filelen + dirlen + 1 >= PATH_MAX) {
                errno = ENAMETOOLONG;
                continue;
            }
            memcpy(expanded_file, dir, dirlen);
            memcpy(expanded_file + dirlen, file, filelen);
            expanded_file[dirlen + filelen] = '\0';
            execve_with_shell_fallback(expanded_file, argv, envp);
            /* There are 3 responses to various classes of errno:
             * return immediately, continue (especially for ENOENT),
             * or continue with "sticky" errno.
             *
             * From exec(3):
             *
             * If permission is denied for a file (the attempted
             * execve returned EACCES), these functions will continue
             * searching the rest of the search path.  If no other
             * file is found, however, they will return with the
             * global variable errno set to EACCES.
             */
            switch (errno) {
            case EACCES:
                sticky_errno = errno;
                /* FALLTHRU */
            case ENOENT:
            case ENOTDIR:
#ifdef ELOOP
            case ELOOP:
#endif
#ifdef ESTALE
            case ESTALE:
#endif
#ifdef ENODEV
            case ENODEV:
#endif
#ifdef ETIMEDOUT
            case ETIMEDOUT:
#endif
                break; /* Try other directories in PATH */
            default:
                return;
            }
        }
        if (sticky_errno != 0)
            errno = sticky_errno;
    }
}

/*
 * Reads nbyte bytes from file descriptor fd into buf,
 * The read operation is retried in case of EINTR or partial reads.
 *
 * Returns number of bytes read (normally nbyte, but may be less in
 * case of EOF).  In case of read errors, returns -1 and sets errno.
 */
static ssize_t
readFully(int fd, void *buf, size_t nbyte)
{
    ssize_t remaining = nbyte;
    for (;;) {
        ssize_t n = read(fd, buf, remaining);
        if (n == 0) {
            return nbyte - remaining;
        } else if (n > 0) {
            remaining -= n;
            if (remaining <= 0)
                return nbyte;
            /* We were interrupted in the middle of reading the bytes.
             * Unlikely, but possible. */
            buf = (void *) (((char *)buf) + n);
        } else if (errno == EINTR) {
            /* Strange signals like SIGJVM1 are possible at any time.
             * See http://www.dreamsongs.com/WorseIsBetter.html */
        } else {
            return -1;
        }
    }
}

typedef struct _ChildStuff
{
    int in[2];
    int out[2];
    int err[2];
    int fail[2];
    int fds[3];
    const char **argv;
    const char **envv;
    const char *pdir;
    jboolean redirectErrorStream;
#if START_CHILD_USE_CLONE
    void *clone_stack;
#endif
} ChildStuff;

static void
copyPipe(int from[2], int to[2])
{
    to[0] = from[0];
    to[1] = from[1];
}

/**
 * Child process after a successful fork() or clone().
 * This function must not return, and must be prepared for either all
 * of its address space to be shared with its parent, or to be a copy.
 * It must not modify global variables such as "environ".
 */
static int
childProcess(void *arg)
{
    const ChildStuff* p = (const ChildStuff*) arg;

    /* Close the parent sides of the pipes.
       Closing pipe fds here is redundant, since closeDescriptors()
       would do it anyways, but a little paranoia is a good thing. */
    if ((closeSafely(p->in[1])   == -1) ||
        (closeSafely(p->out[0])  == -1) ||
        (closeSafely(p->err[0])  == -1) ||
        (closeSafely(p->fail[0]) == -1))
        goto WhyCantJohnnyExec;

    /* Give the child sides of the pipes the right fileno's. */
    /* Note: it is possible for in[0] == 0 */
    if ((moveDescriptor(p->in[0] != -1 ?  p->in[0] : p->fds[0],
                        STDIN_FILENO) == -1) ||
        (moveDescriptor(p->out[1]!= -1 ? p->out[1] : p->fds[1],
                        STDOUT_FILENO) == -1))
        goto WhyCantJohnnyExec;

    if (p->redirectErrorStream) {
        if ((closeSafely(p->err[1]) == -1) ||
            (restartableDup2(STDOUT_FILENO, STDERR_FILENO) == -1))
            goto WhyCantJohnnyExec;
    } else {
        if (moveDescriptor(p->err[1] != -1 ? p->err[1] : p->fds[2],
                           STDERR_FILENO) == -1)
            goto WhyCantJohnnyExec;
    }

    if (moveDescriptor(p->fail[1], FAIL_FILENO) == -1)
        goto WhyCantJohnnyExec;

    /* close everything */
    if (closeDescriptors() == 0) { /* failed,  close the old way */
        int max_fd = (int)sysconf(_SC_OPEN_MAX);
        int fd;
        for (fd = FAIL_FILENO + 1; fd < max_fd; fd++)
            if (restartableClose(fd) == -1 && errno != EBADF)
                goto WhyCantJohnnyExec;
    }

    /* change to the new working directory */
    if (p->pdir != NULL && chdir(p->pdir) < 0)
        goto WhyCantJohnnyExec;

    if (fcntl(FAIL_FILENO, F_SETFD, FD_CLOEXEC) == -1)
        goto WhyCantJohnnyExec;

    JDK_execvpe(p->argv[0], p->argv, p->envv);

 WhyCantJohnnyExec:
    /* We used to go to an awful lot of trouble to predict whether the
     * child would fail, but there is no reliable way to predict the
     * success of an operation without *trying* it, and there's no way
     * to try a chdir or exec in the parent.  Instead, all we need is a
     * way to communicate any failure back to the parent.  Easy; we just
     * send the errno back to the parent over a pipe in case of failure.
     * The tricky thing is, how do we communicate the *success* of exec?
     * We use FD_CLOEXEC together with the fact that a read() on a pipe
     * yields EOF when the write ends (we have two of them!) are closed.
     */
    {
        int errnum = errno;
        restartableWrite(FAIL_FILENO, &errnum, sizeof(errnum));
    }
    restartableClose(FAIL_FILENO);
    _exit(-1);
    return 0;  /* Suppress warning "no return value from function" */
}

/**
 * Start a child process running function childProcess.
 * This function only returns in the parent.
 * We are unusually paranoid; use of clone/vfork is
 * especially likely to tickle gcc/glibc bugs.
 */
#ifdef __attribute_noinline__  /* See: sys/cdefs.h */
__attribute_noinline__
#endif
static pid_t
startChild(ChildStuff *c) {
#if START_CHILD_USE_CLONE
#define START_CHILD_CLONE_STACK_SIZE (64 * 1024)
    /*
     * See clone(2).
     * Instead of worrying about which direction the stack grows, just
     * allocate twice as much and start the stack in the middle.
     */
    if ((c->clone_stack = malloc(2 * START_CHILD_CLONE_STACK_SIZE)) == NULL)
        /* errno will be set to ENOMEM */
        return -1;
    return clone(childProcess,
                 c->clone_stack + START_CHILD_CLONE_STACK_SIZE,
                 CLONE_VFORK | CLONE_VM | SIGCHLD, c);
#else
  #if START_CHILD_USE_VFORK
    /*
     * We separate the call to vfork into a separate function to make
     * very sure to keep stack of child from corrupting stack of parent,
     * as suggested by the scary gcc warning:
     *  warning: variable 'foo' might be clobbered by 'longjmp' or 'vfork'
     */
    volatile pid_t resultPid = vfork();
  #else
    /*
     * From Solaris fork(2): In Solaris 10, a call to fork() is
     * identical to a call to fork1(); only the calling thread is
     * replicated in the child process. This is the POSIX-specified
     * behavior for fork().
     */
    pid_t resultPid = fork();
  #endif
    if (resultPid == 0)
        childProcess(c);
    assert(resultPid != 0);  /* childProcess never returns */
    return resultPid;
#endif /* ! START_CHILD_USE_CLONE */
}

JNIEXPORT jint JNICALL
Java_java_lang_UNIXProcess_forkAndExec(JNIEnv *env,
                                       jobject process,
                                       jbyteArray prog,
                                       jbyteArray argBlock, jint argc,
                                       jbyteArray envBlock, jint envc,
                                       jbyteArray dir,
                                       jintArray std_fds,
                                       jboolean redirectErrorStream)
{
    int errnum;
    int resultPid = -1;
    int in[2], out[2], err[2], fail[2];
    jint *fds = NULL;
    const char *pprog = NULL;
    const char *pargBlock = NULL;
    const char *penvBlock = NULL;
    ChildStuff *c;

    in[0] = in[1] = out[0] = out[1] = err[0] = err[1] = fail[0] = fail[1] = -1;

    if ((c = NEW(ChildStuff, 1)) == NULL) return -1;
    c->argv = NULL;
    c->envv = NULL;
    c->pdir = NULL;
#if START_CHILD_USE_CLONE
    c->clone_stack = NULL;
#endif

    /* Convert prog + argBlock into a char ** argv.
     * Add one word room for expansion of argv for use by
     * execve_as_traditional_shell_script.
     */
    assert(prog != NULL && argBlock != NULL);
    if ((pprog     = getBytes(env, prog))       == NULL) goto Catch;
    if ((pargBlock = getBytes(env, argBlock))   == NULL) goto Catch;
    if ((c->argv = NEW(const char *, argc + 3)) == NULL) goto Catch;
    c->argv[0] = pprog;
    initVectorFromBlock(c->argv+1, pargBlock, argc);

    if (envBlock != NULL) {
        /* Convert envBlock into a char ** envv */
        if ((penvBlock = getBytes(env, envBlock))   == NULL) goto Catch;
        if ((c->envv = NEW(const char *, envc + 1)) == NULL) goto Catch;
        initVectorFromBlock(c->envv, penvBlock, envc);
    }

    if (dir != NULL) {
        if ((c->pdir = getBytes(env, dir)) == NULL) goto Catch;
    }

    assert(std_fds != NULL);
    fds = (*env)->GetIntArrayElements(env, std_fds, NULL);
    if (fds == NULL) goto Catch;

    if ((fds[0] == -1 && pipe(in)  < 0) ||
        (fds[1] == -1 && pipe(out) < 0) ||
        (fds[2] == -1 && pipe(err) < 0) ||
        (pipe(fail) < 0)) {
        throwIOException(env, errno, "Bad file descriptor");
        goto Catch;
    }
    c->fds[0] = fds[0];
    c->fds[1] = fds[1];
    c->fds[2] = fds[2];

    copyPipe(in,   c->in);
    copyPipe(out,  c->out);
    copyPipe(err,  c->err);
    copyPipe(fail, c->fail);

    c->redirectErrorStream = redirectErrorStream;

    resultPid = startChild(c);
    assert(resultPid != 0);

    if (resultPid < 0) {
        throwIOException(env, errno, START_CHILD_SYSTEM_CALL " failed");
        goto Catch;
    }

    restartableClose(fail[1]); fail[1] = -1; /* See: WhyCantJohnnyExec */

    switch (readFully(fail[0], &errnum, sizeof(errnum))) {
    case 0: break; /* Exec succeeded */
    case sizeof(errnum):
        waitpid(resultPid, NULL, 0);
        throwIOException(env, errnum, "Exec failed");
        goto Catch;
    default:
        throwIOException(env, errno, "Read failed");
        goto Catch;
    }

    fds[0] = (in [1] != -1) ? in [1] : -1;
    fds[1] = (out[0] != -1) ? out[0] : -1;
    fds[2] = (err[0] != -1) ? err[0] : -1;

 Finally:
#if START_CHILD_USE_CLONE
    free(c->clone_stack);
#endif

    /* Always clean up the child's side of the pipes */
    closeSafely(in [0]);
    closeSafely(out[1]);
    closeSafely(err[1]);

    /* Always clean up fail descriptors */
    closeSafely(fail[0]);
    closeSafely(fail[1]);

    releaseBytes(env, prog,     pprog);
    releaseBytes(env, argBlock, pargBlock);
    releaseBytes(env, envBlock, penvBlock);
    releaseBytes(env, dir,      c->pdir);

    free(c->argv);
    free(c->envv);
    free(c);

    if (fds != NULL)
        (*env)->ReleaseIntArrayElements(env, std_fds, fds, 0);

    return resultPid;

 Catch:
    /* Clean up the parent's side of the pipes in case of failure only */
    closeSafely(in [1]);
    closeSafely(out[0]);
    closeSafely(err[0]);
    goto Finally;
}

JNIEXPORT void JNICALL
Java_java_lang_UNIXProcess_destroyProcess(JNIEnv *env, jobject junk, jint pid)
{
    kill(pid, SIGTERM);
}