xref: /freebsd/contrib/bmake/job.c (revision 53bb5613a8a15363718b6e6de8d965bf9a2c5469)

/*	$NetBSD: job.c,v 1.480 2024/07/07 07:50:57 rillig Exp $	*/

/*
 * Copyright (c) 1988, 1989, 1990 The Regents of the University of California.
 * All rights reserved.
 *
 * This code is derived from software contributed to Berkeley by
 * Adam de Boor.
 *
 * Redistribution and use in source and binary forms, with or without
 * modification, are permitted provided that the following conditions
 * are met:
 * 1. Redistributions of source code must retain the above copyright
 *    notice, this list of conditions and the following disclaimer.
 * 2. Redistributions in binary form must reproduce the above copyright
 *    notice, this list of conditions and the following disclaimer in the
 *    documentation and/or other materials provided with the distribution.
 * 3. Neither the name of the University nor the names of its contributors
 *    may be used to endorse or promote products derived from this software
 *    without specific prior written permission.
 *
 * THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND
 * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
 * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
 * ARE DISCLAIMED.  IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE
 * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
 * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
 * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
 * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
 * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
 * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
 * SUCH DAMAGE.
 */

/*
 * Copyright (c) 1988, 1989 by Adam de Boor
 * Copyright (c) 1989 by Berkeley Softworks
 * All rights reserved.
 *
 * This code is derived from software contributed to Berkeley by
 * Adam de Boor.
 *
 * Redistribution and use in source and binary forms, with or without
 * modification, are permitted provided that the following conditions
 * are met:
 * 1. Redistributions of source code must retain the above copyright
 *    notice, this list of conditions and the following disclaimer.
 * 2. Redistributions in binary form must reproduce the above copyright
 *    notice, this list of conditions and the following disclaimer in the
 *    documentation and/or other materials provided with the distribution.
 * 3. All advertising materials mentioning features or use of this software
 *    must display the following acknowledgement:
 *	This product includes software developed by the University of
 *	California, Berkeley and its contributors.
 * 4. Neither the name of the University nor the names of its contributors
 *    may be used to endorse or promote products derived from this software
 *    without specific prior written permission.
 *
 * THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND
 * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
 * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
 * ARE DISCLAIMED.  IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE
 * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
 * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
 * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
 * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
 * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
 * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
 * SUCH DAMAGE.
 */

/*
 * Create child processes and collect their output.
 *
 * Interface:
 *	Job_Init	Called to initialize this module. In addition,
 *			the .BEGIN target is made, including all of its
 *			dependencies before this function returns.
 *			Hence, the makefiles must have been parsed
 *			before this function is called.
 *
 *	Job_End		Clean up any memory used.
 *
 *	Job_Make	Start the creation of the given target.
 *
 *	Job_CatchChildren
 *			Check for and handle the termination of any
 *			children. This must be called reasonably
 *			frequently to keep the whole make going at
 *			a decent clip, since job table entries aren't
 *			removed until their process is caught this way.
 *
 *	Job_CatchOutput
 *			Print any output our children have produced.
 *			Should also be called fairly frequently to
 *			keep the user informed of what's going on.
 *			If no output is waiting, it will block for
 *			a time given by the SEL_* constants, below,
 *			or until output is ready.
 *
 *	Job_ParseShell	Given a special dependency line with target '.SHELL',
 *			define the shell that is used for the creation
 *			commands in jobs mode.
 *
 *	Job_Finish	Perform any final processing which needs doing.
 *			This includes the execution of any commands
 *			which have been/were attached to the .END
 *			target. It should only be called when the
 *			job table is empty.
 *
 *	Job_AbortAll	Abort all currently running jobs. Do not handle
 *			output or do anything for the jobs, just kill them.
 *			Should only be called in an emergency.
 *
 *	Job_CheckCommands
 *			Verify that the commands for a target are
 *			ok. Provide them if necessary and possible.
 *
 *	Job_Touch	Update a target without really updating it.
 *
 *	Job_Wait	Wait for all currently-running jobs to finish.
 */

#ifdef HAVE_CONFIG_H
# include "config.h"
#endif
#include <sys/types.h>
#include <sys/stat.h>
#include <sys/file.h>
#include <sys/time.h>
#include "wait.h"

#include <errno.h>
#if !defined(USE_SELECT) && defined(HAVE_POLL_H)
#include <poll.h>
#else
#ifndef USE_SELECT			/* no poll.h */
# define USE_SELECT
#endif
#if defined(HAVE_SYS_SELECT_H)
# include <sys/select.h>
#endif
#endif
#include <signal.h>
#include <utime.h>
#if defined(HAVE_SYS_SOCKET_H)
# include <sys/socket.h>
#endif

#include "make.h"
#include "dir.h"
#include "job.h"
#include "pathnames.h"
#include "trace.h"

/*	"@(#)job.c	8.2 (Berkeley) 3/19/94"	*/
MAKE_RCSID("$NetBSD: job.c,v 1.480 2024/07/07 07:50:57 rillig Exp $");

/*
 * A shell defines how the commands are run.  All commands for a target are
 * written into a single file, which is then given to the shell to execute
 * the commands from it.  The commands are written to the file using a few
 * templates for echo control and error control.
 *
 * The name of the shell is the basename for the predefined shells, such as
 * "sh", "csh", "bash".  For custom shells, it is the full pathname, and its
 * basename is used to select the type of shell; the longest match wins.
 * So /usr/pkg/bin/bash has type sh, /usr/local/bin/tcsh has type csh.
 *
 * The echoing of command lines is controlled using hasEchoCtl, echoOff,
 * echoOn, noPrint and noPrintLen.  When echoOff is executed by the shell, it
 * still outputs something, but this something is not interesting, therefore
 * it is filtered out using noPrint and noPrintLen.
 *
 * The error checking for individual commands is controlled using hasErrCtl,
 * errOn, errOff and runChkTmpl.
 *
 * In case a shell doesn't have error control, echoTmpl is a printf template
 * for echoing the command, should echoing be on; runIgnTmpl is another
 * printf template for executing the command while ignoring the return
 * status. Finally runChkTmpl is a printf template for running the command and
 * causing the shell to exit on error. If any of these strings are empty when
 * hasErrCtl is false, the command will be executed anyway as is, and if it
 * causes an error, so be it. Any templates set up to echo the command will
 * escape any '$ ` \ "' characters in the command string to avoid unwanted
 * shell code injection, the escaped command is safe to use in double quotes.
 *
 * The command-line flags "echo" and "exit" also control the behavior.  The
 * "echo" flag causes the shell to start echoing commands right away.  The
 * "exit" flag causes the shell to exit when an error is detected in one of
 * the commands.
 */
typedef struct Shell {

	/*
	 * The name of the shell. For Bourne and C shells, this is used only
	 * to find the shell description when used as the single source of a
	 * .SHELL target. For user-defined shells, this is the full path of
	 * the shell.
	 */
	const char *name;

	bool hasEchoCtl;	/* whether both echoOff and echoOn are there */
	const char *echoOff;	/* command to turn echoing off */
	const char *echoOn;	/* command to turn echoing back on */
	const char *noPrint;	/* text to skip when printing output from the
				 * shell. This is usually the same as echoOff */
	size_t noPrintLen;	/* length of noPrint command */

	bool hasErrCtl;		/* whether error checking can be controlled
				 * for individual commands */
	const char *errOn;	/* command to turn on error checking */
	const char *errOff;	/* command to turn off error checking */

	const char *echoTmpl;	/* template to echo a command */
	const char *runIgnTmpl;	/* template to run a command without error
				 * checking */
	const char *runChkTmpl;	/* template to run a command with error
				 * checking */

	/*
	 * A string literal that results in a newline character when it
	 * occurs outside of any 'quote' or "quote" characters.
	 */
	const char *newline;
	char commentChar;	/* character used by shell for comment lines */

	const char *echoFlag;	/* shell flag to echo commands */
	const char *errFlag;	/* shell flag to exit on error */
} Shell;

typedef struct CommandFlags {
	/* Whether to echo the command before or instead of running it. */
	bool echo;

	/* Run the command even in -n or -N mode. */
	bool always;

	/*
	 * true if we turned error checking off before writing the command to
	 * the commands file and need to turn it back on
	 */
	bool ignerr;
} CommandFlags;

/*
 * Write shell commands to a file.
 *
 * TODO: keep track of whether commands are echoed.
 * TODO: keep track of whether error checking is active.
 */
typedef struct ShellWriter {
	FILE *f;

	/* we've sent 'set -x' */
	bool xtraced;

} ShellWriter;

/*
 * FreeBSD: traditionally .MAKE is not required to
 * pass jobs queue to sub-makes.
 * Use .MAKE.ALWAYS_PASS_JOB_QUEUE=no to disable.
 */
#define MAKE_ALWAYS_PASS_JOB_QUEUE "${.MAKE.ALWAYS_PASS_JOB_QUEUE:U}"
static bool Always_pass_job_queue = true;
/*
 * FreeBSD: aborting entire parallel make isn't always
 * desired. When doing tinderbox for example, failure of
 * one architecture should not stop all.
 * We still want to bail on interrupt though.
 */
#define MAKE_JOB_ERROR_TOKEN "${MAKE_JOB_ERROR_TOKEN:U}"
static bool Job_error_token = true;

/* error handling variables */
static int job_errors = 0;	/* number of errors reported */
static enum {			/* Why is the make aborting? */
	ABORT_NONE,
	ABORT_ERROR,		/* Aborted because of an error */
	ABORT_INTERRUPT,	/* Aborted because it was interrupted */
	ABORT_WAIT		/* Waiting for jobs to finish */
} aborting = ABORT_NONE;
#define JOB_TOKENS "+EI+"	/* Token to requeue for each abort state */

/* Tracks the number of tokens currently "out" to build jobs. */
int jobTokensRunning = 0;

typedef enum JobStartResult {
	JOB_RUNNING,		/* Job is running */
	JOB_ERROR,		/* Error in starting the job */
	JOB_FINISHED		/* The job is already finished */
} JobStartResult;

/*
 * Descriptions for various shells.
 *
 * The build environment may set DEFSHELL_INDEX to one of
 * DEFSHELL_INDEX_SH, DEFSHELL_INDEX_KSH, or DEFSHELL_INDEX_CSH, to
 * select one of the predefined shells as the default shell.
 *
 * Alternatively, the build environment may set DEFSHELL_CUSTOM to the
 * name or the full path of a sh-compatible shell, which will be used as
 * the default shell.
 *
 * ".SHELL" lines in Makefiles can choose the default shell from the
 * set defined here, or add additional shells.
 */

#ifdef DEFSHELL_CUSTOM
#define DEFSHELL_INDEX_CUSTOM 0
#define DEFSHELL_INDEX_SH     1
#define DEFSHELL_INDEX_KSH    2
#define DEFSHELL_INDEX_CSH    3
#else
#define DEFSHELL_INDEX_SH     0
#define DEFSHELL_INDEX_KSH    1
#define DEFSHELL_INDEX_CSH    2
#endif

#ifndef DEFSHELL_INDEX
#define DEFSHELL_INDEX 0	/* DEFSHELL_INDEX_CUSTOM or DEFSHELL_INDEX_SH */
#endif

static Shell shells[] = {
#ifdef DEFSHELL_CUSTOM
    /*
     * An sh-compatible shell with a non-standard name.
     *
     * Keep this in sync with the "sh" description below, but avoid
     * non-portable features that might not be supplied by all
     * sh-compatible shells.
     */
    {
	DEFSHELL_CUSTOM,	/* .name */
	false,			/* .hasEchoCtl */
	"",			/* .echoOff */
	"",			/* .echoOn */
	"",			/* .noPrint */
	0,			/* .noPrintLen */
	false,			/* .hasErrCtl */
	"",			/* .errOn */
	"",			/* .errOff */
	"echo \"%s\"\n",	/* .echoTmpl */
	"%s\n",			/* .runIgnTmpl */
	"{ %s \n} || exit $?\n", /* .runChkTmpl */
	"'\n'",			/* .newline */
	'#',			/* .commentChar */
	"",			/* .echoFlag */
	"",			/* .errFlag */
    },
#endif /* DEFSHELL_CUSTOM */
    /*
     * SH description. Echo control is also possible and, under
     * sun UNIX anyway, one can even control error checking.
     */
    {
	"sh",			/* .name */
	false,			/* .hasEchoCtl */
	"",			/* .echoOff */
	"",			/* .echoOn */
	"",			/* .noPrint */
	0,			/* .noPrintLen */
	false,			/* .hasErrCtl */
	"",			/* .errOn */
	"",			/* .errOff */
	"echo \"%s\"\n",	/* .echoTmpl */
	"%s\n",			/* .runIgnTmpl */
	"{ %s \n} || exit $?\n", /* .runChkTmpl */
	"'\n'",			/* .newline */
	'#',			/* .commentChar*/
#if defined(MAKE_NATIVE) && defined(__NetBSD__)
	/* XXX: -q is not really echoFlag, it's more like noEchoInSysFlag. */
	"q",			/* .echoFlag */
#else
	"",			/* .echoFlag */
#endif
	"",			/* .errFlag */
    },
    /*
     * KSH description.
     */
    {
	"ksh",			/* .name */
	true,			/* .hasEchoCtl */
	"set +v",		/* .echoOff */
	"set -v",		/* .echoOn */
	"set +v",		/* .noPrint */
	6,			/* .noPrintLen */
	false,			/* .hasErrCtl */
	"",			/* .errOn */
	"",			/* .errOff */
	"echo \"%s\"\n",	/* .echoTmpl */
	"%s\n",			/* .runIgnTmpl */
	"{ %s \n} || exit $?\n", /* .runChkTmpl */
	"'\n'",			/* .newline */
	'#',			/* .commentChar */
	"v",			/* .echoFlag */
	"",			/* .errFlag */
    },
    /*
     * CSH description. The csh can do echo control by playing
     * with the setting of the 'echo' shell variable. Sadly,
     * however, it is unable to do error control nicely.
     */
    {
	"csh",			/* .name */
	true,			/* .hasEchoCtl */
	"unset verbose",	/* .echoOff */
	"set verbose",		/* .echoOn */
	"unset verbose",	/* .noPrint */
	13,			/* .noPrintLen */
	false,			/* .hasErrCtl */
	"",			/* .errOn */
	"",			/* .errOff */
	"echo \"%s\"\n",	/* .echoTmpl */
	"csh -c \"%s || exit 0\"\n", /* .runIgnTmpl */
	"",			/* .runChkTmpl */
	"'\\\n'",		/* .newline */
	'#',			/* .commentChar */
	"v",			/* .echoFlag */
	"e",			/* .errFlag */
    }
};

/*
 * This is the shell to which we pass all commands in the Makefile.
 * It is set by the Job_ParseShell function.
 */
static Shell *shell = &shells[DEFSHELL_INDEX];
char *shellPath;		/* full pathname of executable image */
const char *shellName = NULL;	/* last component of shellPath */
char *shellErrFlag = NULL;
static char *shell_freeIt = NULL; /* Allocated memory for custom .SHELL */


static Job *job_table;		/* The structures that describe them */
static Job *job_table_end;	/* job_table + maxJobs */
static unsigned int wantToken;
static bool lurking_children = false;
static bool make_suspended = false; /* Whether we've seen a SIGTSTP (etc) */

/*
 * Set of descriptors of pipes connected to
 * the output channels of children
 */
static struct pollfd *fds = NULL;
static Job **jobByFdIndex = NULL;
static nfds_t fdsLen = 0;
static void watchfd(Job *);
static void clearfd(Job *);
static bool readyfd(Job *);

static char *targPrefix = NULL;	/* To identify a job change in the output. */
static Job tokenWaitJob;	/* token wait pseudo-job */

static Job childExitJob;	/* child exit pseudo-job */
#define CHILD_EXIT "."
#define DO_JOB_RESUME "R"

enum {
	npseudojobs = 2		/* number of pseudo-jobs */
};

static sigset_t caught_signals;	/* Set of signals we handle */
static volatile sig_atomic_t caught_sigchld;

static void CollectOutput(Job *, bool);
static void JobInterrupt(bool, int) MAKE_ATTR_DEAD;
static void JobRestartJobs(void);
static void JobSigReset(void);

static void
SwitchOutputTo(GNode *gn)
{
	/* The node for which output was most recently produced. */
	static GNode *lastNode = NULL;

	if (gn == lastNode)
		return;
	lastNode = gn;

	if (opts.maxJobs != 1 && targPrefix != NULL && targPrefix[0] != '\0')
		(void)fprintf(stdout, "%s %s ---\n", targPrefix, gn->name);
}

static unsigned
nfds_per_job(void)
{
#if defined(USE_FILEMON) && !defined(USE_FILEMON_DEV)
	if (useMeta)
		return 2;
#endif
	return 1;
}

void
Job_FlagsToString(const Job *job, char *buf, size_t bufsize)
{
	snprintf(buf, bufsize, "%c%c%c",
	    job->ignerr ? 'i' : '-',
	    !job->echo ? 's' : '-',
	    job->special ? 'S' : '-');
}

static void
DumpJobs(const char *where)
{
	Job *job;
	char flags[4];

	debug_printf("job table @ %s\n", where);
	for (job = job_table; job < job_table_end; job++) {
		Job_FlagsToString(job, flags, sizeof flags);
		debug_printf("job %d, status %d, flags %s, pid %d\n",
		    (int)(job - job_table), job->status, flags, job->pid);
	}
}

/*
 * Delete the target of a failed, interrupted, or otherwise
 * unsuccessful job unless inhibited by .PRECIOUS.
 */
static void
JobDeleteTarget(GNode *gn)
{
	const char *file;

	if (gn->type & OP_JOIN)
		return;
	if (gn->type & OP_PHONY)
		return;
	if (GNode_IsPrecious(gn))
		return;
	if (opts.noExecute)
		return;

	file = GNode_Path(gn);
	if (unlink_file(file) == 0)
		Error("*** %s removed", file);
}

/*
 * JobSigLock/JobSigUnlock
 *
 * Signal lock routines to get exclusive access. Currently used to
 * protect `jobs' and `stoppedJobs' list manipulations.
 */
static void
JobSigLock(sigset_t *omaskp)
{
	if (sigprocmask(SIG_BLOCK, &caught_signals, omaskp) != 0) {
		Punt("JobSigLock: sigprocmask: %s", strerror(errno));
		sigemptyset(omaskp);
	}
}

static void
JobSigUnlock(sigset_t *omaskp)
{
	(void)sigprocmask(SIG_SETMASK, omaskp, NULL);
}

static void
JobCreatePipe(Job *job, int minfd)
{
	int i, fd, flags;
	int pipe_fds[2];

	if (pipe(pipe_fds) == -1)
		Punt("Cannot create pipe: %s", strerror(errno));

	for (i = 0; i < 2; i++) {
		/* Avoid using low-numbered fds */
		fd = fcntl(pipe_fds[i], F_DUPFD, minfd);
		if (fd != -1) {
			close(pipe_fds[i]);
			pipe_fds[i] = fd;
		}
	}

	job->inPipe = pipe_fds[0];
	job->outPipe = pipe_fds[1];

	if (fcntl(job->inPipe, F_SETFD, FD_CLOEXEC) == -1)
		Punt("Cannot set close-on-exec: %s", strerror(errno));
	if (fcntl(job->outPipe, F_SETFD, FD_CLOEXEC) == -1)
		Punt("Cannot set close-on-exec: %s", strerror(errno));

	/*
	 * We mark the input side of the pipe non-blocking; we poll(2) the
	 * pipe when we're waiting for a job token, but we might lose the
	 * race for the token when a new one becomes available, so the read
	 * from the pipe should not block.
	 */
	flags = fcntl(job->inPipe, F_GETFL, 0);
	if (flags == -1)
		Punt("Cannot get flags: %s", strerror(errno));
	flags |= O_NONBLOCK;
	if (fcntl(job->inPipe, F_SETFL, flags) == -1)
		Punt("Cannot set flags: %s", strerror(errno));
}

/* Pass the signal to each running job. */
static void
JobCondPassSig(int signo)
{
	Job *job;

	DEBUG1(JOB, "JobCondPassSig(%d) called.\n", signo);

	for (job = job_table; job < job_table_end; job++) {
		if (job->status != JOB_ST_RUNNING)
			continue;
		DEBUG2(JOB, "JobCondPassSig passing signal %d to child %d.\n",
		    signo, job->pid);
		KILLPG(job->pid, signo);
	}
}

/*
 * SIGCHLD handler.
 *
 * Sends a token on the child exit pipe to wake us up from select()/poll().
 */
static void
JobChildSig(int signo MAKE_ATTR_UNUSED)
{
	caught_sigchld = 1;
	while (write(childExitJob.outPipe, CHILD_EXIT, 1) == -1 &&
	       errno == EAGAIN)
		continue;
}


/* Resume all stopped jobs. */
static void
JobContinueSig(int signo MAKE_ATTR_UNUSED)
{
	/*
	 * Defer sending SIGCONT to our stopped children until we return
	 * from the signal handler.
	 */
	while (write(childExitJob.outPipe, DO_JOB_RESUME, 1) == -1 &&
	       errno == EAGAIN)
		continue;
}

/*
 * Pass a signal on to all jobs, then resend to ourselves.
 * We die by the same signal.
 */
MAKE_ATTR_DEAD static void
JobPassSig_int(int signo)
{
	/* Run .INTERRUPT target then exit */
	JobInterrupt(true, signo);
}

/*
 * Pass a signal on to all jobs, then resend to ourselves.
 * We die by the same signal.
 */
MAKE_ATTR_DEAD static void
JobPassSig_term(int signo)
{
	/* Dont run .INTERRUPT target then exit */
	JobInterrupt(false, signo);
}

static void
JobPassSig_suspend(int signo)
{
	sigset_t nmask, omask;
	struct sigaction act;

	/* Suppress job started/continued messages */
	make_suspended = true;

	/* Pass the signal onto every job */
	JobCondPassSig(signo);

	/*
	 * Send ourselves the signal now we've given the message to everyone
	 * else. Note we block everything else possible while we're getting
	 * the signal. This ensures that all our jobs get continued when we
	 * wake up before we take any other signal.
	 */
	sigfillset(&nmask);
	sigdelset(&nmask, signo);
	(void)sigprocmask(SIG_SETMASK, &nmask, &omask);

	act.sa_handler = SIG_DFL;
	sigemptyset(&act.sa_mask);
	act.sa_flags = 0;
	(void)sigaction(signo, &act, NULL);

	DEBUG1(JOB, "JobPassSig passing signal %d to self.\n", signo);

	(void)kill(getpid(), signo);

	/*
	 * We've been continued.
	 *
	 * A whole host of signals is going to happen!
	 * SIGCHLD for any processes that actually suspended themselves.
	 * SIGCHLD for any processes that exited while we were asleep.
	 * The SIGCONT that actually caused us to wake up.
	 *
	 * Since we defer passing the SIGCONT on to our children until
	 * the main processing loop, we can be sure that all the SIGCHLD
	 * events will have happened by then - and that the waitpid() will
	 * collect the child 'suspended' events.
	 * For correct sequencing we just need to ensure we process the
	 * waitpid() before passing on the SIGCONT.
	 *
	 * In any case nothing else is needed here.
	 */

	/* Restore handler and signal mask */
	act.sa_handler = JobPassSig_suspend;
	(void)sigaction(signo, &act, NULL);
	(void)sigprocmask(SIG_SETMASK, &omask, NULL);
}

static Job *
JobFindPid(int pid, JobStatus status, bool isJobs)
{
	Job *job;

	for (job = job_table; job < job_table_end; job++) {
		if (job->status == status && job->pid == pid)
			return job;
	}
	if (DEBUG(JOB) && isJobs)
		DumpJobs("no pid");
	return NULL;
}

/* Parse leading '@', '-' and '+', which control the exact execution mode. */
static void
ParseCommandFlags(char **pp, CommandFlags *out_cmdFlags)
{
	char *p = *pp;
	out_cmdFlags->echo = true;
	out_cmdFlags->ignerr = false;
	out_cmdFlags->always = false;

	for (;;) {
		if (*p == '@')
			out_cmdFlags->echo = DEBUG(LOUD);
		else if (*p == '-')
			out_cmdFlags->ignerr = true;
		else if (*p == '+')
			out_cmdFlags->always = true;
		else if (!ch_isspace(*p))
			/* Ignore whitespace for compatibility with GNU make */
			break;
		p++;
	}

	pp_skip_whitespace(&p);

	*pp = p;
}

/* Escape a string for a double-quoted string literal in sh, csh and ksh. */
static char *
EscapeShellDblQuot(const char *cmd)
{
	size_t i, j;

	/* Worst that could happen is every char needs escaping. */
	char *esc = bmake_malloc(strlen(cmd) * 2 + 1);
	for (i = 0, j = 0; cmd[i] != '\0'; i++, j++) {
		if (cmd[i] == '$' || cmd[i] == '`' || cmd[i] == '\\' ||
		    cmd[i] == '"')
			esc[j++] = '\\';
		esc[j] = cmd[i];
	}
	esc[j] = '\0';

	return esc;
}

static void
ShellWriter_WriteFmt(ShellWriter *wr, const char *fmt, const char *arg)
{
	DEBUG1(JOB, fmt, arg);

	(void)fprintf(wr->f, fmt, arg);
	if (wr->f == stdout)
		(void)fflush(wr->f);
}

static void
ShellWriter_WriteLine(ShellWriter *wr, const char *line)
{
	ShellWriter_WriteFmt(wr, "%s\n", line);
}

static void
ShellWriter_EchoOff(ShellWriter *wr)
{
	if (shell->hasEchoCtl)
		ShellWriter_WriteLine(wr, shell->echoOff);
}

static void
ShellWriter_EchoCmd(ShellWriter *wr, const char *escCmd)
{
	ShellWriter_WriteFmt(wr, shell->echoTmpl, escCmd);
}

static void
ShellWriter_EchoOn(ShellWriter *wr)
{
	if (shell->hasEchoCtl)
		ShellWriter_WriteLine(wr, shell->echoOn);
}

static void
ShellWriter_TraceOn(ShellWriter *wr)
{
	if (!wr->xtraced) {
		ShellWriter_WriteLine(wr, "set -x");
		wr->xtraced = true;
	}
}

static void
ShellWriter_ErrOff(ShellWriter *wr, bool echo)
{
	if (echo)
		ShellWriter_EchoOff(wr);
	ShellWriter_WriteLine(wr, shell->errOff);
	if (echo)
		ShellWriter_EchoOn(wr);
}

static void
ShellWriter_ErrOn(ShellWriter *wr, bool echo)
{
	if (echo)
		ShellWriter_EchoOff(wr);
	ShellWriter_WriteLine(wr, shell->errOn);
	if (echo)
		ShellWriter_EchoOn(wr);
}

/*
 * The shell has no built-in error control, so emulate error control by
 * enclosing each shell command in a template like "{ %s \n } || exit $?"
 * (configurable per shell).
 */
static void
JobWriteSpecialsEchoCtl(Job *job, ShellWriter *wr, CommandFlags *inout_cmdFlags,
			const char *escCmd, const char **inout_cmdTemplate)
{
	/* XXX: Why is the whole job modified at this point? */
	job->ignerr = true;

	if (job->echo && inout_cmdFlags->echo) {
		ShellWriter_EchoOff(wr);
		ShellWriter_EchoCmd(wr, escCmd);

		/*
		 * Leave echoing off so the user doesn't see the commands
		 * for toggling the error checking.
		 */
		inout_cmdFlags->echo = false;
	}
	*inout_cmdTemplate = shell->runIgnTmpl;

	/*
	 * The template runIgnTmpl already takes care of ignoring errors,
	 * so pretend error checking is still on.
	 * XXX: What effects does this have, and why is it necessary?
	 */
	inout_cmdFlags->ignerr = false;
}

static void
JobWriteSpecials(Job *job, ShellWriter *wr, const char *escCmd, bool run,
		 CommandFlags *inout_cmdFlags, const char **inout_cmdTemplate)
{
	if (!run)
		inout_cmdFlags->ignerr = false;
	else if (shell->hasErrCtl)
		ShellWriter_ErrOff(wr, job->echo && inout_cmdFlags->echo);
	else if (shell->runIgnTmpl != NULL && shell->runIgnTmpl[0] != '\0') {
		JobWriteSpecialsEchoCtl(job, wr, inout_cmdFlags, escCmd,
		    inout_cmdTemplate);
	} else
		inout_cmdFlags->ignerr = false;
}

/*
 * Write a shell command to the job's commands file, to be run later.
 *
 * If the command starts with '@' and neither the -s nor the -n flag was
 * given to make, stick a shell-specific echoOff command in the script.
 *
 * If the command starts with '-' and the shell has no error control (none
 * of the predefined shells has that), ignore errors for the entire job.
 *
 * XXX: Why ignore errors for the entire job?  This is even documented in the
 * manual page, but without any rationale since there is no known rationale.
 *
 * XXX: The manual page says the '-' "affects the entire job", but that's not
 * accurate.  The '-' does not affect the commands before the '-'.
 *
 * If the command is just "...", skip all further commands of this job.  These
 * commands are attached to the .END node instead and will be run by
 * Job_Finish after all other targets have been made.
 */
static void
JobWriteCommand(Job *job, ShellWriter *wr, StringListNode *ln, const char *ucmd)
{
	bool run;

	CommandFlags cmdFlags;
	/* Template for writing a command to the shell file */
	const char *cmdTemplate;
	char *xcmd;		/* The expanded command */
	char *xcmdStart;
	char *escCmd;		/* xcmd escaped to be used in double quotes */

	run = GNode_ShouldExecute(job->node);

	xcmd = Var_SubstInTarget(ucmd, job->node);
	/* TODO: handle errors */
	xcmdStart = xcmd;

	cmdTemplate = "%s\n";

	ParseCommandFlags(&xcmd, &cmdFlags);

	/* The '+' command flag overrides the -n or -N options. */
	if (cmdFlags.always && !run) {
		/*
		 * We're not actually executing anything...
		 * but this one needs to be - use compat mode just for it.
		 */
		(void)Compat_RunCommand(ucmd, job->node, ln);
		free(xcmdStart);
		return;
	}

	/*
	 * If the shell doesn't have error control, the alternate echoing
	 * will be done (to avoid showing additional error checking code)
	 * and this needs some characters escaped.
	 */
	escCmd = shell->hasErrCtl ? NULL : EscapeShellDblQuot(xcmd);

	if (!cmdFlags.echo) {
		if (job->echo && run && shell->hasEchoCtl)
			ShellWriter_EchoOff(wr);
		else if (shell->hasErrCtl)
			cmdFlags.echo = true;
	}

	if (cmdFlags.ignerr) {
		JobWriteSpecials(job, wr, escCmd, run, &cmdFlags, &cmdTemplate);
	} else {

		/*
		 * If errors are being checked and the shell doesn't have
		 * error control but does supply an runChkTmpl template, then
		 * set up commands to run through it.
		 */

		if (!shell->hasErrCtl && shell->runChkTmpl != NULL &&
		    shell->runChkTmpl[0] != '\0') {
			if (job->echo && cmdFlags.echo) {
				ShellWriter_EchoOff(wr);
				ShellWriter_EchoCmd(wr, escCmd);
				cmdFlags.echo = false;
			}
			/*
			 * If it's a comment line or blank, avoid the possible
			 * syntax error generated by "{\n} || exit $?".
			 */
			cmdTemplate = escCmd[0] == shell->commentChar ||
				      escCmd[0] == '\0'
			    ? shell->runIgnTmpl
			    : shell->runChkTmpl;
			cmdFlags.ignerr = false;
		}
	}

	if (DEBUG(SHELL) && strcmp(shellName, "sh") == 0)
		ShellWriter_TraceOn(wr);

	ShellWriter_WriteFmt(wr, cmdTemplate, xcmd);
	free(xcmdStart);
	free(escCmd);

	if (cmdFlags.ignerr)
		ShellWriter_ErrOn(wr, cmdFlags.echo && job->echo);

	if (!cmdFlags.echo)
		ShellWriter_EchoOn(wr);
}

/*
 * Write all commands to the shell file that is later executed.
 *
 * The special command "..." stops writing and saves the remaining commands
 * to be executed later, when the target '.END' is made.
 *
 * Return whether at least one command was written to the shell file.
 */
static bool
JobWriteCommands(Job *job)
{
	StringListNode *ln;
	bool seen = false;
	ShellWriter wr;

	wr.f = job->cmdFILE;
	wr.xtraced = false;

	for (ln = job->node->commands.first; ln != NULL; ln = ln->next) {
		const char *cmd = ln->datum;

		if (strcmp(cmd, "...") == 0) {
			job->node->type |= OP_SAVE_CMDS;
			job->tailCmds = ln->next;
			break;
		}

		JobWriteCommand(job, &wr, ln, ln->datum);
		seen = true;
	}

	return seen;
}

/*
 * Save the delayed commands (those after '...'), to be executed later in
 * the '.END' node, when everything else is done.
 */
static void
JobSaveCommands(Job *job)
{
	StringListNode *ln;

	for (ln = job->tailCmds; ln != NULL; ln = ln->next) {
		const char *cmd = ln->datum;
		char *expanded_cmd;
		/*
		 * XXX: This Var_Subst is only intended to expand the dynamic
		 * variables such as .TARGET, .IMPSRC.  It is not intended to
		 * expand the other variables as well; see deptgt-end.mk.
		 */
		expanded_cmd = Var_SubstInTarget(cmd, job->node);
		/* TODO: handle errors */
		Lst_Append(&Targ_GetEndNode()->commands, expanded_cmd);
		Parse_RegisterCommand(expanded_cmd);
	}
}


/* Called to close both input and output pipes when a job is finished. */
static void
JobClosePipes(Job *job)
{
	clearfd(job);
	(void)close(job->outPipe);
	job->outPipe = -1;

	CollectOutput(job, true);
	(void)close(job->inPipe);
	job->inPipe = -1;
}

static void
DebugFailedJob(const Job *job)
{
	const StringListNode *ln;

	if (!DEBUG(ERROR))
		return;

	debug_printf("\n");
	debug_printf("*** Failed target: %s\n", job->node->name);
	debug_printf("*** In directory: %s\n", curdir);
	debug_printf("*** Failed commands:\n");
	for (ln = job->node->commands.first; ln != NULL; ln = ln->next) {
		const char *cmd = ln->datum;
		debug_printf("\t%s\n", cmd);

		if (strchr(cmd, '$') != NULL) {
			char *xcmd = Var_Subst(cmd, job->node, VARE_EVAL);
			debug_printf("\t=> %s\n", xcmd);
			free(xcmd);
		}
	}
}

static void
JobFinishDoneExitedError(Job *job, WAIT_T *inout_status)
{
	SwitchOutputTo(job->node);
#ifdef USE_META
	if (useMeta) {
		meta_job_error(job, job->node,
		    job->ignerr, WEXITSTATUS(*inout_status));
	}
#endif
	if (!shouldDieQuietly(job->node, -1)) {
		DebugFailedJob(job);
		(void)printf("*** [%s] Error code %d%s\n",
		    job->node->name, WEXITSTATUS(*inout_status),
		    job->ignerr ? " (ignored)" : "");
	}

	if (job->ignerr)
		WAIT_STATUS(*inout_status) = 0;
	else {
		if (deleteOnError)
			JobDeleteTarget(job->node);
		PrintOnError(job->node, "\n");
	}
}

static void
JobFinishDoneExited(Job *job, WAIT_T *inout_status)
{
	DEBUG2(JOB, "Process %d [%s] exited.\n", job->pid, job->node->name);

	if (WEXITSTATUS(*inout_status) != 0)
		JobFinishDoneExitedError(job, inout_status);
	else if (DEBUG(JOB)) {
		SwitchOutputTo(job->node);
		(void)printf("*** [%s] Completed successfully\n",
		    job->node->name);
	}
}

static void
JobFinishDoneSignaled(Job *job, WAIT_T status)
{
	SwitchOutputTo(job->node);
	DebugFailedJob(job);
	(void)printf("*** [%s] Signal %d\n", job->node->name, WTERMSIG(status));
	if (deleteOnError)
		JobDeleteTarget(job->node);
}

static void
JobFinishDone(Job *job, WAIT_T *inout_status)
{
	if (WIFEXITED(*inout_status))
		JobFinishDoneExited(job, inout_status);
	else
		JobFinishDoneSignaled(job, *inout_status);

	(void)fflush(stdout);
}

/*
 * Do final processing for the given job including updating parent nodes and
 * starting new jobs as available/necessary.
 *
 * Deferred commands for the job are placed on the .END node.
 *
 * If there was a serious error (job_errors != 0; not an ignored one), no more
 * jobs will be started.
 *
 * Input:
 *	job		job to finish
 *	status		sub-why job went away
 */
static void
JobFinish (Job *job, WAIT_T status)
{
	bool done, return_job_token;

	DEBUG3(JOB, "JobFinish: %d [%s], status %d\n",
	    job->pid, job->node->name, status);

	if ((WIFEXITED(status) &&
	     ((WEXITSTATUS(status) != 0 && !job->ignerr))) ||
	    WIFSIGNALED(status)) {
		/* Finished because of an error. */

		JobClosePipes(job);
		if (job->cmdFILE != NULL && job->cmdFILE != stdout) {
			if (fclose(job->cmdFILE) != 0)
				Punt("Cannot write shell script for '%s': %s",
				    job->node->name, strerror(errno));
			job->cmdFILE = NULL;
		}
		done = true;

	} else if (WIFEXITED(status)) {
		/*
		 * Deal with ignored errors in -B mode. We need to print a
		 * message telling of the ignored error as well as to run
		 * the next command.
		 */
		done = WEXITSTATUS(status) != 0;

		JobClosePipes(job);

	} else {
		/* No need to close things down or anything. */
		done = false;
	}

	if (done)
		JobFinishDone(job, &status);

#ifdef USE_META
	if (useMeta) {
		int meta_status = meta_job_finish(job);
		if (meta_status != 0 && status == 0)
			status = meta_status;
	}
#endif

	return_job_token = false;

	Trace_Log(JOBEND, job);
	if (!job->special) {
		if (WAIT_STATUS(status) != 0 ||
		    (aborting == ABORT_ERROR) || aborting == ABORT_INTERRUPT)
			return_job_token = true;
	}

	if (aborting != ABORT_ERROR && aborting != ABORT_INTERRUPT &&
	    (WAIT_STATUS(status) == 0)) {
		/*
		 * As long as we aren't aborting and the job didn't return a
		 * non-zero status that we shouldn't ignore, we call
		 * Make_Update to update the parents.
		 */
		JobSaveCommands(job);
		job->node->made = MADE;
		if (!job->special)
			return_job_token = true;
		Make_Update(job->node);
		job->status = JOB_ST_FREE;
	} else if (status != 0) {
		job_errors++;
		job->status = JOB_ST_FREE;
	}

	if (job_errors > 0 && !opts.keepgoing && aborting != ABORT_INTERRUPT) {
		/* Prevent more jobs from getting started. */
		aborting = ABORT_ERROR;
	}

	if (return_job_token)
		Job_TokenReturn();

	if (aborting == ABORT_ERROR && jobTokensRunning == 0)
		Finish(job_errors);
}

static void
TouchRegular(GNode *gn)
{
	const char *file = GNode_Path(gn);
	struct utimbuf times;
	int fd;
	char c;

	times.actime = now;
	times.modtime = now;
	if (utime(file, &times) >= 0)
		return;

	fd = open(file, O_RDWR | O_CREAT, 0666);
	if (fd < 0) {
		(void)fprintf(stderr, "*** couldn't touch %s: %s\n",
		    file, strerror(errno));
		(void)fflush(stderr);
		return;		/* XXX: What about propagating the error? */
	}

	/*
	 * Last resort: update the file's time stamps in the traditional way.
	 * XXX: This doesn't work for empty files, which are sometimes used
	 * as marker files.
	 */
	if (read(fd, &c, 1) == 1) {
		(void)lseek(fd, 0, SEEK_SET);
		while (write(fd, &c, 1) == -1 && errno == EAGAIN)
			continue;
	}
	(void)close(fd);	/* XXX: What about propagating the error? */
}

/*
 * Touch the given target. Called by JobStart when the -t flag was given.
 *
 * The modification date of the file is changed.
 * If the file did not exist, it is created.
 */
void
Job_Touch(GNode *gn, bool echo)
{
	if (gn->type &
	    (OP_JOIN | OP_USE | OP_USEBEFORE | OP_EXEC | OP_OPTIONAL |
	     OP_SPECIAL | OP_PHONY)) {
		/*
		 * These are "virtual" targets and should not really be
		 * created.
		 */
		return;
	}

	if (echo || !GNode_ShouldExecute(gn)) {
		(void)fprintf(stdout, "touch %s\n", gn->name);
		(void)fflush(stdout);
	}

	if (!GNode_ShouldExecute(gn))
		return;

	if (gn->type & OP_ARCHV)
		Arch_Touch(gn);
	else if (gn->type & OP_LIB)
		Arch_TouchLib(gn);
	else
		TouchRegular(gn);
}

/*
 * Make sure the given node has all the commands it needs.
 *
 * The node will have commands from the .DEFAULT rule added to it if it
 * needs them.
 *
 * Input:
 *	gn		The target whose commands need verifying
 *	abortProc	Function to abort with message
 *
 * Results:
 *	true if the commands list is/was ok.
 */
bool
Job_CheckCommands(GNode *gn, void (*abortProc)(const char *, ...))
{
	if (GNode_IsTarget(gn))
		return true;
	if (!Lst_IsEmpty(&gn->commands))
		return true;
	if ((gn->type & OP_LIB) && !Lst_IsEmpty(&gn->children))
		return true;

	/*
	 * No commands. Look for .DEFAULT rule from which we might infer
	 * commands.
	 */
	if (defaultNode != NULL && !Lst_IsEmpty(&defaultNode->commands) &&
	    !(gn->type & OP_SPECIAL)) {
		/*
		 * The traditional Make only looks for a .DEFAULT if the node
		 * was never the target of an operator, so that's what we do
		 * too.
		 *
		 * The .DEFAULT node acts like a transformation rule, in that
		 * gn also inherits any attributes or sources attached to
		 * .DEFAULT itself.
		 */
		Make_HandleUse(defaultNode, gn);
		Var_Set(gn, IMPSRC, GNode_VarTarget(gn));
		return true;
	}

	Dir_UpdateMTime(gn, false);
	if (gn->mtime != 0 || (gn->type & OP_SPECIAL))
		return true;

	/*
	 * The node wasn't the target of an operator.  We have no .DEFAULT
	 * rule to go on and the target doesn't already exist. There's
	 * nothing more we can do for this branch. If the -k flag wasn't
	 * given, we stop in our tracks, otherwise we just don't update
	 * this node's parents so they never get examined.
	 */

	if (gn->flags.fromDepend) {
		if (!Job_RunTarget(".STALE", gn->fname))
			fprintf(stdout,
			    "%s: %s, %u: ignoring stale %s for %s\n",
			    progname, gn->fname, gn->lineno, makeDependfile,
			    gn->name);
		return true;
	}

	if (gn->type & OP_OPTIONAL) {
		(void)fprintf(stdout, "%s: don't know how to make %s (%s)\n",
		    progname, gn->name, "ignored");
		(void)fflush(stdout);
		return true;
	}

	if (opts.keepgoing) {
		(void)fprintf(stdout, "%s: don't know how to make %s (%s)\n",
		    progname, gn->name, "continuing");
		(void)fflush(stdout);
		return false;
	}

	abortProc("don't know how to make %s. Stop", gn->name);
	return false;
}

/*
 * Execute the shell for the given job.
 *
 * See Job_CatchOutput for handling the output of the shell.
 */
static void
JobExec(Job *job, char **argv)
{
	int cpid;		/* ID of new child */
	sigset_t mask;

	if (DEBUG(JOB)) {
		int i;

		debug_printf("Running %s\n", job->node->name);
		debug_printf("\tCommand: ");
		for (i = 0; argv[i] != NULL; i++) {
			debug_printf("%s ", argv[i]);
		}
		debug_printf("\n");
	}

	/*
	 * Some jobs produce no output, and it's disconcerting to have
	 * no feedback of their running (since they produce no output, the
	 * banner with their name in it never appears). This is an attempt to
	 * provide that feedback, even if nothing follows it.
	 */
	if (job->echo)
		SwitchOutputTo(job->node);

	/* No interruptions until this job is on the `jobs' list */
	JobSigLock(&mask);

	/* Pre-emptively mark job running, pid still zero though */
	job->status = JOB_ST_RUNNING;

	Var_ReexportVars(job->node);

	cpid = vfork();
	if (cpid == -1)
		Punt("Cannot vfork: %s", strerror(errno));

	if (cpid == 0) {
		/* Child */
		sigset_t tmask;

#ifdef USE_META
		if (useMeta)
			meta_job_child(job);
#endif
		/*
		 * Reset all signal handlers; this is necessary because we
		 * also need to unblock signals before we exec(2).
		 */
		JobSigReset();

		/* Now unblock signals */
		sigemptyset(&tmask);
		JobSigUnlock(&tmask);

		/*
		 * Must duplicate the input stream down to the child's input
		 * and reset it to the beginning (again). Since the stream
		 * was marked close-on-exec, we must clear that bit in the
		 * new input.
		 */
		if (dup2(fileno(job->cmdFILE), STDIN_FILENO) == -1)
			execDie("dup2", "job->cmdFILE");
		if (fcntl(STDIN_FILENO, F_SETFD, 0) == -1)
			execDie("fcntl clear close-on-exec", "stdin");
		if (lseek(STDIN_FILENO, 0, SEEK_SET) == -1)
			execDie("lseek to 0", "stdin");

		if (Always_pass_job_queue ||
		    (job->node->type & (OP_MAKE | OP_SUBMAKE))) {
			/* Pass job token pipe to submakes. */
			if (fcntl(tokenWaitJob.inPipe, F_SETFD, 0) == -1)
				execDie("clear close-on-exec",
				    "tokenWaitJob.inPipe");
			if (fcntl(tokenWaitJob.outPipe, F_SETFD, 0) == -1)
				execDie("clear close-on-exec",
				    "tokenWaitJob.outPipe");
		}

		/*
		 * Set up the child's output to be routed through the pipe
		 * we've created for it.
		 */
		if (dup2(job->outPipe, STDOUT_FILENO) == -1)
			execDie("dup2", "job->outPipe");

		/*
		 * The output channels are marked close on exec. This bit
		 * was duplicated by dup2 (on some systems), so we have
		 * to clear it before routing the shell's error output to
		 * the same place as its standard output.
		 */
		if (fcntl(STDOUT_FILENO, F_SETFD, 0) == -1)
			execDie("clear close-on-exec", "stdout");
		if (dup2(STDOUT_FILENO, STDERR_FILENO) == -1)
			execDie("dup2", "1, 2");

		/*
		 * We want to switch the child into a different process
		 * family so we can kill it and all its descendants in
		 * one fell swoop, by killing its process family, but not
		 * commit suicide.
		 */
#if defined(HAVE_SETPGID)
		(void)setpgid(0, getpid());
#else
# if defined(HAVE_SETSID)
		/* XXX: dsl - I'm sure this should be setpgrp()... */
		(void)setsid();
# else
		(void)setpgrp(0, getpid());
# endif
#endif

		(void)execv(shellPath, argv);
		execDie("exec", shellPath);
	}

	/* Parent, continuing after the child exec */
	job->pid = cpid;

	Trace_Log(JOBSTART, job);

#ifdef USE_META
	if (useMeta)
		meta_job_parent(job, cpid);
#endif

	/*
	 * Set the current position in the buffer to the beginning
	 * and mark another stream to watch in the outputs mask
	 */
	job->curPos = 0;

	watchfd(job);

	if (job->cmdFILE != NULL && job->cmdFILE != stdout) {
		if (fclose(job->cmdFILE) != 0)
			Punt("Cannot write shell script for '%s': %s",
			    job->node->name, strerror(errno));
		job->cmdFILE = NULL;
	}

	/* Now that the job is actually running, add it to the table. */
	if (DEBUG(JOB)) {
		debug_printf("JobExec(%s): pid %d added to jobs table\n",
		    job->node->name, job->pid);
		DumpJobs("job started");
	}
	JobSigUnlock(&mask);
}

/* Create the argv needed to execute the shell for a given job. */
static void
JobMakeArgv(Job *job, char **argv)
{
	int argc;
	static char args[10];	/* For merged arguments */

	argv[0] = UNCONST(shellName);
	argc = 1;

	if ((shell->errFlag != NULL && shell->errFlag[0] != '-') ||
	    (shell->echoFlag != NULL && shell->echoFlag[0] != '-')) {
		/*
		 * At least one of the flags doesn't have a minus before it,
		 * so merge them together. Have to do this because the Bourne
		 * shell thinks its second argument is a file to source.
		 * Grrrr. Note the ten-character limitation on the combined
		 * arguments.
		 *
		 * TODO: Research until when the above comments were
		 * practically relevant.
		 */
		(void)snprintf(args, sizeof args, "-%s%s",
		    (job->ignerr ? "" :
			(shell->errFlag != NULL ? shell->errFlag : "")),
		    (!job->echo ? "" :
			(shell->echoFlag != NULL ? shell->echoFlag : "")));

		if (args[1] != '\0') {
			argv[argc] = args;
			argc++;
		}
	} else {
		if (!job->ignerr && shell->errFlag != NULL) {
			argv[argc] = UNCONST(shell->errFlag);
			argc++;
		}
		if (job->echo && shell->echoFlag != NULL) {
			argv[argc] = UNCONST(shell->echoFlag);
			argc++;
		}
	}
	argv[argc] = NULL;
}

static void
JobWriteShellCommands(Job *job, GNode *gn, bool *out_run)
{
	/*
	 * tfile is the name of a file into which all shell commands
	 * are put. It is removed before the child shell is executed,
	 * unless DEBUG(SCRIPT) is set.
	 */
	char tfile[MAXPATHLEN];
	int tfd;		/* File descriptor to the temp file */

	tfd = Job_TempFile(TMPPAT, tfile, sizeof tfile);

	job->cmdFILE = fdopen(tfd, "w+");
	if (job->cmdFILE == NULL)
		Punt("Could not fdopen %s", tfile);

	(void)fcntl(fileno(job->cmdFILE), F_SETFD, FD_CLOEXEC);

#ifdef USE_META
	if (useMeta) {
		meta_job_start(job, gn);
		if (gn->type & OP_SILENT)	/* might have changed */
			job->echo = false;
	}
#endif

	*out_run = JobWriteCommands(job);
}

/*
 * Start a target-creation process going for the target described by gn.
 *
 * Results:
 *	JOB_ERROR if there was an error in the commands, JOB_FINISHED
 *	if there isn't actually anything left to do for the job and
 *	JOB_RUNNING if the job has been started.
 *
 * Details:
 *	A new Job node is created and added to the list of running
 *	jobs. PMake is forked and a child shell created.
 *
 * NB: The return value is ignored by everyone.
 */
static JobStartResult
JobStart(GNode *gn, bool special)
{
	Job *job;		/* new job descriptor */
	char *argv[10];		/* Argument vector to shell */
	bool cmdsOK;		/* true if the nodes commands were all right */
	bool run;

	for (job = job_table; job < job_table_end; job++) {
		if (job->status == JOB_ST_FREE)
			break;
	}
	if (job >= job_table_end)
		Punt("JobStart no job slots vacant");

	memset(job, 0, sizeof *job);
	job->node = gn;
	job->tailCmds = NULL;
	job->status = JOB_ST_SET_UP;

	job->special = special || gn->type & OP_SPECIAL;
	job->ignerr = opts.ignoreErrors || gn->type & OP_IGNORE;
	job->echo = !(opts.silent || gn->type & OP_SILENT);

	/*
	 * Check the commands now so any attributes from .DEFAULT have a
	 * chance to migrate to the node.
	 */
	cmdsOK = Job_CheckCommands(gn, Error);

	job->inPollfd = NULL;

	if (Lst_IsEmpty(&gn->commands)) {
		job->cmdFILE = stdout;
		run = false;

		/*
		 * We're serious here, but if the commands were bogus, we're
		 * also dead...
		 */
		if (!cmdsOK) {
			PrintOnError(gn, "\n");	/* provide some clue */
			DieHorribly();
		}
	} else if (((gn->type & OP_MAKE) && !opts.noRecursiveExecute) ||
	    (!opts.noExecute && !opts.touch)) {
		/*
		 * The above condition looks very similar to
		 * GNode_ShouldExecute but is subtly different.  It prevents
		 * that .MAKE targets are touched since these are usually
		 * virtual targets.
		 */

		/*
		 * We're serious here, but if the commands were bogus, we're
		 * also dead...
		 */
		if (!cmdsOK) {
			PrintOnError(gn, "\n");	/* provide some clue */
			DieHorribly();
		}

		JobWriteShellCommands(job, gn, &run);
		(void)fflush(job->cmdFILE);
	} else if (!GNode_ShouldExecute(gn)) {
		/*
		 * Just write all the commands to stdout in one fell swoop.
		 * This still sets up job->tailCmds correctly.
		 */
		SwitchOutputTo(gn);
		job->cmdFILE = stdout;
		if (cmdsOK)
			JobWriteCommands(job);
		run = false;
		(void)fflush(job->cmdFILE);
	} else {
		Job_Touch(gn, job->echo);
		run = false;
	}

	/* If we're not supposed to execute a shell, don't. */
	if (!run) {
		if (!job->special)
			Job_TokenReturn();
		/* Unlink and close the command file if we opened one */
		if (job->cmdFILE != NULL && job->cmdFILE != stdout) {
			(void)fclose(job->cmdFILE);
			job->cmdFILE = NULL;
		}

		/*
		 * We only want to work our way up the graph if we aren't
		 * here because the commands for the job were no good.
		 */
		if (cmdsOK && aborting == ABORT_NONE) {
			JobSaveCommands(job);
			job->node->made = MADE;
			Make_Update(job->node);
		}
		job->status = JOB_ST_FREE;
		return cmdsOK ? JOB_FINISHED : JOB_ERROR;
	}

	/*
	 * Set up the control arguments to the shell. This is based on the
	 * flags set earlier for this job.
	 */
	JobMakeArgv(job, argv);

	/* Create the pipe by which we'll get the shell's output. */
	JobCreatePipe(job, 3);

	JobExec(job, argv);
	return JOB_RUNNING;
}

/*
 * If the shell has an output filter (which only csh and ksh have by default),
 * print the output of the child process, skipping the noPrint text of the
 * shell.
 *
 * Return the part of the output that the calling function needs to output by
 * itself.
 */
static char *
PrintFilteredOutput(char *p, const char *endp)	/* XXX: p should be const */
{
	char *ep;		/* XXX: should be const */

	if (shell->noPrint == NULL || shell->noPrint[0] == '\0')
		return p;

	/*
	 * XXX: What happens if shell->noPrint occurs on the boundary of
	 * the buffer?  To work correctly in all cases, this should rather
	 * be a proper stream filter instead of doing string matching on
	 * selected chunks of the output.
	 */
	while ((ep = strstr(p, shell->noPrint)) != NULL) {
		if (ep != p) {
			*ep = '\0';	/* XXX: avoid writing to the buffer */
			/*
			 * The only way there wouldn't be a newline after
			 * this line is if it were the last in the buffer.
			 * however, since the noPrint output comes after it,
			 * there must be a newline, so we don't print one.
			 */
			/* XXX: What about null bytes in the output? */
			(void)fprintf(stdout, "%s", p);
			(void)fflush(stdout);
		}
		p = ep + shell->noPrintLen;
		if (p == endp)
			break;
		p++;		/* skip over the (XXX: assumed) newline */
		pp_skip_whitespace(&p);
	}
	return p;
}

/*
 * This function is called whenever there is something to read on the pipe.
 * We collect more output from the given job and store it in the job's
 * outBuf. If this makes up a line, we print it tagged by the job's
 * identifier, as necessary.
 *
 * In the output of the shell, the 'noPrint' lines are removed. If the
 * command is not alone on the line (the character after it is not \0 or
 * \n), we do print whatever follows it.
 *
 * Input:
 *	job		the job whose output needs printing
 *	finish		true if this is the last time we'll be called
 *			for this job
 */
static void
CollectOutput(Job *job, bool finish)
{
	bool gotNL;		/* true if got a newline */
	bool fbuf;		/* true if our buffer filled up */
	size_t nr;		/* number of bytes read */
	size_t i;		/* auxiliary index into outBuf */
	size_t max;		/* limit for i (end of current data) */
	ssize_t nRead;		/* (Temporary) number of bytes read */

	/* Read as many bytes as will fit in the buffer. */
again:
	gotNL = false;
	fbuf = false;

	nRead = read(job->inPipe, &job->outBuf[job->curPos],
	    JOB_BUFSIZE - job->curPos);
	if (nRead < 0) {
		if (errno == EAGAIN)
			return;
		if (DEBUG(JOB))
			perror("CollectOutput(piperead)");
		nr = 0;
	} else
		nr = (size_t)nRead;

	if (nr == 0)
		finish = false;	/* stop looping */

	/*
	 * If we hit the end-of-file (the job is dead), we must flush its
	 * remaining output, so pretend we read a newline if there's any
	 * output remaining in the buffer.
	 */
	if (nr == 0 && job->curPos != 0) {
		job->outBuf[job->curPos] = '\n';
		nr = 1;
	}

	max = job->curPos + nr;
	for (i = job->curPos; i < max; i++)
		if (job->outBuf[i] == '\0')
			job->outBuf[i] = ' ';

	/* Look for the last newline in the bytes we just got. */
	for (i = job->curPos + nr - 1;
	     i >= job->curPos && i != (size_t)-1; i--) {
		if (job->outBuf[i] == '\n') {
			gotNL = true;
			break;
		}
	}

	if (!gotNL) {
		job->curPos += nr;
		if (job->curPos == JOB_BUFSIZE) {
			/*
			 * If we've run out of buffer space, we have no choice
			 * but to print the stuff. sigh.
			 */
			fbuf = true;
			i = job->curPos;
		}
	}
	if (gotNL || fbuf) {
		/*
		 * Need to send the output to the screen. Null terminate it
		 * first, overwriting the newline character if there was one.
		 * So long as the line isn't one we should filter (according
		 * to the shell description), we print the line, preceded
		 * by a target banner if this target isn't the same as the
		 * one for which we last printed something.
		 * The rest of the data in the buffer are then shifted down
		 * to the start of the buffer and curPos is set accordingly.
		 */
		job->outBuf[i] = '\0';
		if (i >= job->curPos) {
			char *p;

			/*
			 * FIXME: SwitchOutputTo should be here, according to
			 * the comment above.  But since PrintOutput does not
			 * do anything in the default shell, this bug has gone
			 * unnoticed until now.
			 */
			p = PrintFilteredOutput(job->outBuf, &job->outBuf[i]);

			/*
			 * There's still more in the output buffer. This time,
			 * though, we know there's no newline at the end, so
			 * we add one of our own free will.
			 */
			if (*p != '\0') {
				if (!opts.silent)
					SwitchOutputTo(job->node);
#ifdef USE_META
				if (useMeta) {
					meta_job_output(job, p,
					    gotNL ? "\n" : "");
				}
#endif
				(void)fprintf(stdout, "%s%s", p,
				    gotNL ? "\n" : "");
				(void)fflush(stdout);
			}
		}
		/*
		 * max is the last offset still in the buffer. Move any
		 * remaining characters to the start of the buffer and
		 * update the end marker curPos.
		 */
		if (i < max) {
			(void)memmove(job->outBuf, &job->outBuf[i + 1],
			    max - (i + 1));
			job->curPos = max - (i + 1);
		} else {
			assert(i == max);
			job->curPos = 0;
		}
	}
	if (finish) {
		/*
		 * If the finish flag is true, we must loop until we hit
		 * end-of-file on the pipe. This is guaranteed to happen
		 * eventually since the other end of the pipe is now closed
		 * (we closed it explicitly and the child has exited). When
		 * we do get an EOF, finish will be set false and we'll fall
		 * through and out.
		 */
		goto again;
	}
}

static void
JobRun(GNode *targ)
{
#if 0
	/*
	 * Unfortunately it is too complicated to run .BEGIN, .END, and
	 * .INTERRUPT job in the parallel job module.  As of 2020-09-25,
	 * unit-tests/deptgt-end-jobs.mk hangs in an endless loop.
	 *
	 * Running these jobs in compat mode also guarantees that these
	 * jobs do not overlap with other unrelated jobs.
	 */
	GNodeList lst = LST_INIT;
	Lst_Append(&lst, targ);
	(void)Make_Run(&lst);
	Lst_Done(&lst);
	JobStart(targ, true);
	while (jobTokensRunning != 0) {
		Job_CatchOutput();
	}
#else
	Compat_Make(targ, targ);
	/* XXX: Replace with GNode_IsError(gn) */
	if (targ->made == ERROR) {
		PrintOnError(targ, "\n\nStop.\n");
		exit(1);
	}
#endif
}

/*
 * Handle the exit of a child. Called from Make_Make.
 *
 * The job descriptor is removed from the list of children.
 *
 * Notes:
 *	We do waits, blocking or not, according to the wisdom of our
 *	caller, until there are no more children to report. For each
 *	job, call JobFinish to finish things off.
 */
void
Job_CatchChildren(void)
{
	int pid;		/* pid of dead child */
	WAIT_T status;		/* Exit/termination status */

	/* Don't even bother if we know there's no one around. */
	if (jobTokensRunning == 0)
		return;

	/* Have we received SIGCHLD since last call? */
	if (caught_sigchld == 0)
		return;
	caught_sigchld = 0;

	while ((pid = waitpid((pid_t)-1, &status, WNOHANG | WUNTRACED)) > 0) {
		DEBUG2(JOB, "Process %d exited/stopped status %x.\n",
		    pid, WAIT_STATUS(status));
		JobReapChild(pid, status, true);
	}
}

/*
 * It is possible that wait[pid]() was called from elsewhere,
 * this lets us reap jobs regardless.
 */
void
JobReapChild(pid_t pid, WAIT_T status, bool isJobs)
{
	Job *job;		/* job descriptor for dead child */

	/* Don't even bother if we know there's no one around. */
	if (jobTokensRunning == 0)
		return;

	job = JobFindPid(pid, JOB_ST_RUNNING, isJobs);
	if (job == NULL) {
		if (isJobs) {
			if (!lurking_children)
				Error("Child (%d) status %x not in table?",
				    pid, status);
		}
		return;		/* not ours */
	}
	if (WIFSTOPPED(status)) {
		DEBUG2(JOB, "Process %d (%s) stopped.\n",
		    job->pid, job->node->name);
		if (!make_suspended) {
			switch (WSTOPSIG(status)) {
			case SIGTSTP:
				(void)printf("*** [%s] Suspended\n",
				    job->node->name);
				break;
			case SIGSTOP:
				(void)printf("*** [%s] Stopped\n",
				    job->node->name);
				break;
			default:
				(void)printf("*** [%s] Stopped -- signal %d\n",
				    job->node->name, WSTOPSIG(status));
			}
			job->suspended = true;
		}
		(void)fflush(stdout);
		return;
	}

	job->status = JOB_ST_FINISHED;
	job->exit_status = WAIT_STATUS(status);
	if (WIFEXITED(status))
		job->node->exit_status = WEXITSTATUS(status);

	JobFinish(job, status);
}

/*
 * Catch the output from our children, if we're using pipes do so. Otherwise
 * just block time until we get a signal(most likely a SIGCHLD) since there's
 * no point in just spinning when there's nothing to do and the reaping of a
 * child can wait for a while.
 */
void
Job_CatchOutput(void)
{
	int nready;
	Job *job;
	unsigned int i;

	(void)fflush(stdout);

	/* The first fd in the list is the job token pipe */
	do {
		nready = poll(fds + 1 - wantToken, fdsLen - 1 + wantToken,
		    POLL_MSEC);
	} while (nready < 0 && errno == EINTR);

	if (nready < 0)
		Punt("poll: %s", strerror(errno));

	if (nready > 0 && readyfd(&childExitJob)) {
		char token = 0;
		ssize_t count;
		count = read(childExitJob.inPipe, &token, 1);
		if (count == 1) {
			if (token == DO_JOB_RESUME[0])
				/*
				 * Complete relay requested from our SIGCONT
				 * handler
				 */
				JobRestartJobs();
		} else if (count == 0)
			Punt("unexpected eof on token pipe");
		else if (errno != EAGAIN)
			Punt("token pipe read: %s", strerror(errno));
		nready--;
	}

	Job_CatchChildren();
	if (nready == 0)
		return;

	for (i = npseudojobs * nfds_per_job(); i < fdsLen; i++) {
		if (fds[i].revents == 0)
			continue;
		job = jobByFdIndex[i];
		if (job->status == JOB_ST_RUNNING)
			CollectOutput(job, false);
#if defined(USE_FILEMON) && !defined(USE_FILEMON_DEV)
		/*
		 * With meta mode, we may have activity on the job's filemon
		 * descriptor too, which at the moment is any pollfd other
		 * than job->inPollfd.
		 */
		if (useMeta && job->inPollfd != &fds[i]) {
			if (meta_job_event(job) <= 0)
				fds[i].events = 0;	/* never mind */
		}
#endif
		if (--nready == 0)
			return;
	}
}

/*
 * Start the creation of a target. Basically a front-end for JobStart used by
 * the Make module.
 */
void
Job_Make(GNode *gn)
{
	(void)JobStart(gn, false);
}

static void
InitShellNameAndPath(void)
{
	shellName = shell->name;

#ifdef DEFSHELL_CUSTOM
	if (shellName[0] == '/') {
		shellPath = bmake_strdup(shellName);
		shellName = str_basename(shellPath);
		return;
	}
#endif
#ifdef DEFSHELL_PATH
	shellPath = bmake_strdup(DEFSHELL_PATH);
#else
	shellPath = str_concat3(_PATH_DEFSHELLDIR, "/", shellName);
#endif
}

void
Shell_Init(void)
{
	if (shellPath == NULL)
		InitShellNameAndPath();

	Var_SetWithFlags(SCOPE_CMDLINE, ".SHELL", shellPath,
			 VAR_SET_INTERNAL|VAR_SET_READONLY);
	if (shell->errFlag == NULL)
		shell->errFlag = "";
	if (shell->echoFlag == NULL)
		shell->echoFlag = "";
	if (shell->hasErrCtl && shell->errFlag[0] != '\0') {
		if (shellErrFlag != NULL &&
		    strcmp(shell->errFlag, &shellErrFlag[1]) != 0) {
			free(shellErrFlag);
			shellErrFlag = NULL;
		}
		if (shellErrFlag == NULL)
			shellErrFlag = str_concat2("-", shell->errFlag);
	} else if (shellErrFlag != NULL) {
		free(shellErrFlag);
		shellErrFlag = NULL;
	}
}

/*
 * Return the string literal that is used in the current command shell
 * to produce a newline character.
 */
const char *
Shell_GetNewline(void)
{
	return shell->newline;
}

void
Job_SetPrefix(void)
{
	if (targPrefix != NULL)
		free(targPrefix);
	else if (!Var_Exists(SCOPE_GLOBAL, ".MAKE.JOB.PREFIX"))
		Global_Set(".MAKE.JOB.PREFIX", "---");

	targPrefix = Var_Subst("${.MAKE.JOB.PREFIX}",
	    SCOPE_GLOBAL, VARE_EVAL);
	/* TODO: handle errors */
}

static void
AddSig(int sig, SignalProc handler)
{
	if (bmake_signal(sig, SIG_IGN) != SIG_IGN) {
		sigaddset(&caught_signals, sig);
		(void)bmake_signal(sig, handler);
	}
}

/* Initialize the process module. */
void
Job_Init(void)
{
	Job_SetPrefix();
	/* Allocate space for all the job info */
	job_table = bmake_malloc((size_t)opts.maxJobs * sizeof *job_table);
	memset(job_table, 0, (size_t)opts.maxJobs * sizeof *job_table);
	job_table_end = job_table + opts.maxJobs;
	wantToken = 0;
	caught_sigchld = 0;

	aborting = ABORT_NONE;
	job_errors = 0;

	Always_pass_job_queue = GetBooleanExpr(MAKE_ALWAYS_PASS_JOB_QUEUE,
	    Always_pass_job_queue);

	Job_error_token = GetBooleanExpr(MAKE_JOB_ERROR_TOKEN, Job_error_token);


	/*
	 * There is a non-zero chance that we already have children.
	 * eg after 'make -f- <<EOF'
	 * Since their termination causes a 'Child (pid) not in table'
	 * message, Collect the status of any that are already dead, and
	 * suppress the error message if there are any undead ones.
	 */
	for (;;) {
		int rval;
		WAIT_T status;

		rval = waitpid((pid_t)-1, &status, WNOHANG);
		if (rval > 0)
			continue;
		if (rval == 0)
			lurking_children = true;
		break;
	}

	Shell_Init();

	JobCreatePipe(&childExitJob, 3);

	{
		/* Preallocate enough for the maximum number of jobs. */
		size_t nfds = (npseudojobs + (size_t)opts.maxJobs) *
			      nfds_per_job();
		fds = bmake_malloc(sizeof *fds * nfds);
		jobByFdIndex = bmake_malloc(sizeof *jobByFdIndex * nfds);
	}

	/* These are permanent entries and take slots 0 and 1 */
	watchfd(&tokenWaitJob);
	watchfd(&childExitJob);

	sigemptyset(&caught_signals);
	/* Install a SIGCHLD handler. */
	(void)bmake_signal(SIGCHLD, JobChildSig);
	sigaddset(&caught_signals, SIGCHLD);

	/*
	 * Catch the four signals that POSIX specifies if they aren't ignored.
	 * JobPassSig will take care of calling JobInterrupt if appropriate.
	 */
	AddSig(SIGINT, JobPassSig_int);
	AddSig(SIGHUP, JobPassSig_term);
	AddSig(SIGTERM, JobPassSig_term);
	AddSig(SIGQUIT, JobPassSig_term);

	/*
	 * There are additional signals that need to be caught and passed if
	 * either the export system wants to be told directly of signals or if
	 * we're giving each job its own process group (since then it won't get
	 * signals from the terminal driver as we own the terminal)
	 */
	AddSig(SIGTSTP, JobPassSig_suspend);
	AddSig(SIGTTOU, JobPassSig_suspend);
	AddSig(SIGTTIN, JobPassSig_suspend);
	AddSig(SIGWINCH, JobCondPassSig);
	AddSig(SIGCONT, JobContinueSig);

	(void)Job_RunTarget(".BEGIN", NULL);
	/*
	 * Create the .END node now, even though no code in the unit tests
	 * depends on it.  See also Targ_GetEndNode in Compat_MakeAll.
	 */
	(void)Targ_GetEndNode();
}

static void
DelSig(int sig)
{
	if (sigismember(&caught_signals, sig) != 0)
		(void)bmake_signal(sig, SIG_DFL);
}

static void
JobSigReset(void)
{
	DelSig(SIGINT);
	DelSig(SIGHUP);
	DelSig(SIGQUIT);
	DelSig(SIGTERM);
	DelSig(SIGTSTP);
	DelSig(SIGTTOU);
	DelSig(SIGTTIN);
	DelSig(SIGWINCH);
	DelSig(SIGCONT);
	(void)bmake_signal(SIGCHLD, SIG_DFL);
}

/* Find a shell in 'shells' given its name, or return NULL. */
static Shell *
FindShellByName(const char *name)
{
	Shell *sh = shells;
	const Shell *shellsEnd = sh + sizeof shells / sizeof shells[0];

	for (sh = shells; sh < shellsEnd; sh++) {
		if (strcmp(name, sh->name) == 0)
			return sh;
	}
	return NULL;
}

/*
 * Parse a shell specification and set up 'shell', shellPath and
 * shellName appropriately.
 *
 * Input:
 *	line		The shell spec
 *
 * Results:
 *	false if the specification was incorrect.
 *
 * Side Effects:
 *	'shell' points to a Shell structure (either predefined or
 *	created from the shell spec), shellPath is the full path of the
 *	shell described by 'shell', while shellName is just the
 *	final component of shellPath.
 *
 * Notes:
 *	A shell specification consists of a .SHELL target, with dependency
 *	operator, followed by a series of blank-separated words. Double
 *	quotes can be used to use blanks in words. A backslash escapes
 *	anything (most notably a double-quote and a space) and
 *	provides the functionality it does in C. Each word consists of
 *	keyword and value separated by an equal sign. There should be no
 *	unnecessary spaces in the word. The keywords are as follows:
 *	    name	Name of shell.
 *	    path	Location of shell.
 *	    quiet	Command to turn off echoing.
 *	    echo	Command to turn echoing on
 *	    filter	Result of turning off echoing that shouldn't be
 *			printed.
 *	    echoFlag	Flag to turn echoing on at the start
 *	    errFlag	Flag to turn error checking on at the start
 *	    hasErrCtl	True if shell has error checking control
 *	    newline	String literal to represent a newline char
 *	    check	Command to turn on error checking if hasErrCtl
 *			is true or template of command to echo a command
 *			for which error checking is off if hasErrCtl is
 *			false.
 *	    ignore	Command to turn off error checking if hasErrCtl
 *			is true or template of command to execute a
 *			command so as to ignore any errors it returns if
 *			hasErrCtl is false.
 */
bool
Job_ParseShell(char *line)
{
	Words wordsList;
	char **words;
	char **argv;
	size_t argc;
	char *path;
	Shell newShell;
	bool fullSpec = false;
	Shell *sh;

	/* XXX: don't use line as an iterator variable */
	pp_skip_whitespace(&line);

	free(shell_freeIt);

	memset(&newShell, 0, sizeof newShell);

	/* Parse the specification by keyword. */
	wordsList = Str_Words(line, true);
	words = wordsList.words;
	argc = wordsList.len;
	path = wordsList.freeIt;
	if (words == NULL) {
		Error("Unterminated quoted string [%s]", line);
		return false;
	}
	shell_freeIt = path;

	for (path = NULL, argv = words; argc != 0; argc--, argv++) {
		char *arg = *argv;
		if (strncmp(arg, "path=", 5) == 0) {
			path = arg + 5;
		} else if (strncmp(arg, "name=", 5) == 0) {
			newShell.name = arg + 5;
		} else {
			if (strncmp(arg, "quiet=", 6) == 0) {
				newShell.echoOff = arg + 6;
			} else if (strncmp(arg, "echo=", 5) == 0) {
				newShell.echoOn = arg + 5;
			} else if (strncmp(arg, "filter=", 7) == 0) {
				newShell.noPrint = arg + 7;
				newShell.noPrintLen = strlen(newShell.noPrint);
			} else if (strncmp(arg, "echoFlag=", 9) == 0) {
				newShell.echoFlag = arg + 9;
			} else if (strncmp(arg, "errFlag=", 8) == 0) {
				newShell.errFlag = arg + 8;
			} else if (strncmp(arg, "hasErrCtl=", 10) == 0) {
				char c = arg[10];
				newShell.hasErrCtl = c == 'Y' || c == 'y' ||
						     c == 'T' || c == 't';
			} else if (strncmp(arg, "newline=", 8) == 0) {
				newShell.newline = arg + 8;
			} else if (strncmp(arg, "check=", 6) == 0) {
				/*
				 * Before 2020-12-10, these two variables had
				 * been a single variable.
				 */
				newShell.errOn = arg + 6;
				newShell.echoTmpl = arg + 6;
			} else if (strncmp(arg, "ignore=", 7) == 0) {
				/*
				 * Before 2020-12-10, these two variables had
				 * been a single variable.
				 */
				newShell.errOff = arg + 7;
				newShell.runIgnTmpl = arg + 7;
			} else if (strncmp(arg, "errout=", 7) == 0) {
				newShell.runChkTmpl = arg + 7;
			} else if (strncmp(arg, "comment=", 8) == 0) {
				newShell.commentChar = arg[8];
			} else {
				Parse_Error(PARSE_FATAL,
				    "Unknown keyword \"%s\"", arg);
				free(words);
				return false;
			}
			fullSpec = true;
		}
	}

	if (path == NULL) {
		/*
		 * If no path was given, the user wants one of the
		 * pre-defined shells, yes? So we find the one s/he wants
		 * with the help of FindShellByName and set things up the
		 * right way. shellPath will be set up by Shell_Init.
		 */
		if (newShell.name == NULL) {
			Parse_Error(PARSE_FATAL,
			    "Neither path nor name specified");
			free(words);
			return false;
		} else {
			if ((sh = FindShellByName(newShell.name)) == NULL) {
				Parse_Error(PARSE_WARNING,
				    "%s: No matching shell", newShell.name);
				free(words);
				return false;
			}
			shell = sh;
			shellName = newShell.name;
			if (shellPath != NULL) {
				/*
				 * Shell_Init has already been called!
				 * Do it again.
				 */
				free(shellPath);
				shellPath = NULL;
				Shell_Init();
			}
		}
	} else {
		free(shellPath);
		shellPath = bmake_strdup(path);
		shellName = newShell.name != NULL ? newShell.name
		    : str_basename(path);
		if (!fullSpec) {
			if ((sh = FindShellByName(shellName)) == NULL) {
				Parse_Error(PARSE_WARNING,
				    "%s: No matching shell", shellName);
				free(words);
				return false;
			}
			shell = sh;
		} else {
			shell = bmake_malloc(sizeof *shell);
			*shell = newShell;
		}
		/* this will take care of shellErrFlag */
		Shell_Init();
	}

	if (shell->echoOn != NULL && shell->echoOff != NULL)
		shell->hasEchoCtl = true;

	if (!shell->hasErrCtl) {
		if (shell->echoTmpl == NULL)
			shell->echoTmpl = "";
		if (shell->runIgnTmpl == NULL)
			shell->runIgnTmpl = "%s\n";
	}

	/*
	 * Do not free up the words themselves, since they might be in use
	 * by the shell specification.
	 */
	free(words);
	return true;
}

/*
 * Handle the receipt of an interrupt.
 *
 * All children are killed. Another job will be started if the .INTERRUPT
 * target is defined.
 *
 * Input:
 *	runINTERRUPT	Non-zero if commands for the .INTERRUPT target
 *			should be executed
 *	signo		signal received
 */
static void
JobInterrupt(bool runINTERRUPT, int signo)
{
	Job *job;		/* job descriptor in that element */
	GNode *interrupt;	/* the node describing the .INTERRUPT target */
	sigset_t mask;
	GNode *gn;

	aborting = ABORT_INTERRUPT;

	JobSigLock(&mask);

	for (job = job_table; job < job_table_end; job++) {
		if (job->status != JOB_ST_RUNNING)
			continue;

		gn = job->node;

		JobDeleteTarget(gn);
		if (job->pid != 0) {
			DEBUG2(JOB,
			    "JobInterrupt passing signal %d to child %d.\n",
			    signo, job->pid);
			KILLPG(job->pid, signo);
		}
	}

	JobSigUnlock(&mask);

	if (runINTERRUPT && !opts.touch) {
		interrupt = Targ_FindNode(".INTERRUPT");
		if (interrupt != NULL) {
			opts.ignoreErrors = false;
			JobRun(interrupt);
		}
	}
	Trace_Log(MAKEINTR, NULL);
	exit(signo);		/* XXX: why signo? */
}

/*
 * Do the final processing, i.e. run the commands attached to the .END target.
 *
 * Return the number of errors reported.
 */
int
Job_Finish(void)
{
	GNode *endNode = Targ_GetEndNode();
	if (!Lst_IsEmpty(&endNode->commands) ||
	    !Lst_IsEmpty(&endNode->children)) {
		if (job_errors != 0)
			Error("Errors reported so .END ignored");
		else
			JobRun(endNode);
	}
	return job_errors;
}

#ifdef CLEANUP
/* Clean up any memory used by the jobs module. */
void
Job_End(void)
{
	free(shell_freeIt);
}
#endif

/*
 * Waits for all running jobs to finish and returns.
 * Sets 'aborting' to ABORT_WAIT to prevent other jobs from starting.
 */
void
Job_Wait(void)
{
	aborting = ABORT_WAIT;
	while (jobTokensRunning != 0) {
		Job_CatchOutput();
	}
	aborting = ABORT_NONE;
}

/*
 * Abort all currently running jobs without handling output or anything.
 * This function is to be called only in the event of a major error.
 * Most definitely NOT to be called from JobInterrupt.
 *
 * All children are killed, not just the firstborn.
 */
void
Job_AbortAll(void)
{
	Job *job;		/* the job descriptor in that element */
	WAIT_T foo;

	aborting = ABORT_ERROR;

	if (jobTokensRunning != 0) {
		for (job = job_table; job < job_table_end; job++) {
			if (job->status != JOB_ST_RUNNING)
				continue;
			/*
			 * kill the child process with increasingly drastic
			 * signals to make darn sure it's dead.
			 */
			KILLPG(job->pid, SIGINT);
			KILLPG(job->pid, SIGKILL);
		}
	}

	/*
	 * Catch as many children as want to report in at first, then give up
	 */
	while (waitpid((pid_t)-1, &foo, WNOHANG) > 0)
		continue;
}

/*
 * Tries to restart stopped jobs if there are slots available.
 * Called in process context in response to a SIGCONT.
 */
static void
JobRestartJobs(void)
{
	Job *job;

	for (job = job_table; job < job_table_end; job++) {
		if (job->status == JOB_ST_RUNNING &&
		    (make_suspended || job->suspended)) {
			DEBUG1(JOB, "Restarting stopped job pid %d.\n",
			    job->pid);
			if (job->suspended) {
				(void)printf("*** [%s] Continued\n",
				    job->node->name);
				(void)fflush(stdout);
			}
			job->suspended = false;
			if (KILLPG(job->pid, SIGCONT) != 0 && DEBUG(JOB)) {
				debug_printf("Failed to send SIGCONT to %d\n",
				    job->pid);
			}
		}
		if (job->status == JOB_ST_FINISHED) {
			/*
			 * Job exit deferred after calling waitpid() in a
			 * signal handler
			 */
			JobFinish(job, job->exit_status);
		}
	}
	make_suspended = false;
}

static void
watchfd(Job *job)
{
	if (job->inPollfd != NULL)
		Punt("Watching watched job");

	fds[fdsLen].fd = job->inPipe;
	fds[fdsLen].events = POLLIN;
	jobByFdIndex[fdsLen] = job;
	job->inPollfd = &fds[fdsLen];
	fdsLen++;
#if defined(USE_FILEMON) && !defined(USE_FILEMON_DEV)
	if (useMeta) {
		fds[fdsLen].fd = meta_job_fd(job);
		fds[fdsLen].events = fds[fdsLen].fd == -1 ? 0 : POLLIN;
		jobByFdIndex[fdsLen] = job;
		fdsLen++;
	}
#endif
}

static void
clearfd(Job *job)
{
	size_t i;
	if (job->inPollfd == NULL)
		Punt("Unwatching unwatched job");
	i = (size_t)(job->inPollfd - fds);
	fdsLen--;
#if defined(USE_FILEMON) && !defined(USE_FILEMON_DEV)
	if (useMeta) {
		/*
		 * Sanity check: there should be two fds per job, so the job's
		 * pollfd number should be even.
		 */
		assert(nfds_per_job() == 2);
		if (i % 2 != 0)
			Punt("odd-numbered fd with meta");
		fdsLen--;
	}
#endif
	/* Move last job in table into hole made by dead job. */
	if (fdsLen != i) {
		fds[i] = fds[fdsLen];
		jobByFdIndex[i] = jobByFdIndex[fdsLen];
		jobByFdIndex[i]->inPollfd = &fds[i];
#if defined(USE_FILEMON) && !defined(USE_FILEMON_DEV)
		if (useMeta) {
			fds[i + 1] = fds[fdsLen + 1];
			jobByFdIndex[i + 1] = jobByFdIndex[fdsLen + 1];
		}
#endif
	}
	job->inPollfd = NULL;
}

static bool
readyfd(Job *job)
{
	if (job->inPollfd == NULL)
		Punt("Polling unwatched job");
	return (job->inPollfd->revents & POLLIN) != 0;
}

/*
 * Put a token (back) into the job pipe.
 * This allows a make process to start a build job.
 */
static void
JobTokenAdd(void)
{
	char tok = JOB_TOKENS[aborting], tok1;

	if (!Job_error_token && aborting == ABORT_ERROR) {
		if (jobTokensRunning == 0)
			return;
		tok = '+';		/* no error token */
	}

	/* If we are depositing an error token flush everything else */
	while (tok != '+' && read(tokenWaitJob.inPipe, &tok1, 1) == 1)
		continue;

	DEBUG3(JOB, "(%d) aborting %d, deposit token %c\n",
	    getpid(), aborting, tok);
	while (write(tokenWaitJob.outPipe, &tok, 1) == -1 && errno == EAGAIN)
		continue;
}

/* Get a temp file */
int
Job_TempFile(const char *pattern, char *tfile, size_t tfile_sz)
{
	int fd;
	sigset_t mask;

	JobSigLock(&mask);
	fd = mkTempFile(pattern, tfile, tfile_sz);
	if (tfile != NULL && !DEBUG(SCRIPT))
		unlink(tfile);
	JobSigUnlock(&mask);

	return fd;
}

/* Prep the job token pipe in the root make process. */
void
Job_ServerStart(int max_tokens, int jp_0, int jp_1)
{
	int i;
	char jobarg[64];

	if (jp_0 >= 0 && jp_1 >= 0) {
		/* Pipe passed in from parent */
		tokenWaitJob.inPipe = jp_0;
		tokenWaitJob.outPipe = jp_1;
		(void)fcntl(jp_0, F_SETFD, FD_CLOEXEC);
		(void)fcntl(jp_1, F_SETFD, FD_CLOEXEC);
		return;
	}

	JobCreatePipe(&tokenWaitJob, 15);

	snprintf(jobarg, sizeof jobarg, "%d,%d",
	    tokenWaitJob.inPipe, tokenWaitJob.outPipe);

	Global_Append(MAKEFLAGS, "-J");
	Global_Append(MAKEFLAGS, jobarg);

	/*
	 * Preload the job pipe with one token per job, save the one
	 * "extra" token for the primary job.
	 *
	 * XXX should clip maxJobs against PIPE_BUF -- if max_tokens is
	 * larger than the write buffer size of the pipe, we will
	 * deadlock here.
	 */
	for (i = 1; i < max_tokens; i++)
		JobTokenAdd();
}

/* Return a withdrawn token to the pool. */
void
Job_TokenReturn(void)
{
	jobTokensRunning--;
	if (jobTokensRunning < 0)
		Punt("token botch");
	if (jobTokensRunning != 0 || JOB_TOKENS[aborting] != '+')
		JobTokenAdd();
}

/*
 * Attempt to withdraw a token from the pool.
 *
 * If pool is empty, set wantToken so that we wake up when a token is
 * released.
 *
 * Returns true if a token was withdrawn, and false if the pool is currently
 * empty.
 */
bool
Job_TokenWithdraw(void)
{
	char tok, tok1;
	ssize_t count;

	wantToken = 0;
	DEBUG3(JOB, "Job_TokenWithdraw(%d): aborting %d, running %d\n",
	    getpid(), aborting, jobTokensRunning);

	if (aborting != ABORT_NONE || (jobTokensRunning >= opts.maxJobs))
		return false;

	count = read(tokenWaitJob.inPipe, &tok, 1);
	if (count == 0)
		Fatal("eof on job pipe!");
	if (count < 0 && jobTokensRunning != 0) {
		if (errno != EAGAIN)
			Fatal("job pipe read: %s", strerror(errno));
		DEBUG1(JOB, "(%d) blocked for token\n", getpid());
		wantToken = 1;
		return false;
	}

	if (count == 1 && tok != '+') {
		/* make being aborted - remove any other job tokens */
		DEBUG2(JOB, "(%d) aborted by token %c\n", getpid(), tok);
		while (read(tokenWaitJob.inPipe, &tok1, 1) == 1)
			continue;
		/* And put the stopper back */
		while (write(tokenWaitJob.outPipe, &tok, 1) == -1 &&
		       errno == EAGAIN)
			continue;
		if (shouldDieQuietly(NULL, 1))
			exit(6);	/* we aborted */
		Fatal("A failure has been detected "
		      "in another branch of the parallel make");
	}

	if (count == 1 && jobTokensRunning == 0)
		/* We didn't want the token really */
		while (write(tokenWaitJob.outPipe, &tok, 1) == -1 &&
		       errno == EAGAIN)
			continue;

	jobTokensRunning++;
	DEBUG1(JOB, "(%d) withdrew token\n", getpid());
	return true;
}

/*
 * Run the named target if found. If a filename is specified, then set that
 * to the sources.
 *
 * Exits if the target fails.
 */
bool
Job_RunTarget(const char *target, const char *fname)
{
	GNode *gn = Targ_FindNode(target);
	if (gn == NULL)
		return false;

	if (fname != NULL)
		Var_Set(gn, ALLSRC, fname);

	JobRun(gn);
	/* XXX: Replace with GNode_IsError(gn) */
	if (gn->made == ERROR) {
		PrintOnError(gn, "\n\nStop.\n");
		exit(1);
	}
	return true;
}

#ifdef USE_SELECT
int
emul_poll(struct pollfd *fd, int nfd, int timeout)
{
	fd_set rfds, wfds;
	int i, maxfd, nselect, npoll;
	struct timeval tv, *tvp;
	long usecs;

	FD_ZERO(&rfds);
	FD_ZERO(&wfds);

	maxfd = -1;
	for (i = 0; i < nfd; i++) {
		fd[i].revents = 0;

		if (fd[i].events & POLLIN)
			FD_SET(fd[i].fd, &rfds);

		if (fd[i].events & POLLOUT)
			FD_SET(fd[i].fd, &wfds);

		if (fd[i].fd > maxfd)
			maxfd = fd[i].fd;
	}

	if (maxfd >= FD_SETSIZE) {
		Punt("Ran out of fd_set slots; "
		     "recompile with a larger FD_SETSIZE.");
	}

	if (timeout < 0) {
		tvp = NULL;
	} else {
		usecs = timeout * 1000;
		tv.tv_sec = usecs / 1000000;
		tv.tv_usec = usecs % 1000000;
		tvp = &tv;
	}

	nselect = select(maxfd + 1, &rfds, &wfds, NULL, tvp);

	if (nselect <= 0)
		return nselect;

	npoll = 0;
	for (i = 0; i < nfd; i++) {
		if (FD_ISSET(fd[i].fd, &rfds))
			fd[i].revents |= POLLIN;

		if (FD_ISSET(fd[i].fd, &wfds))
			fd[i].revents |= POLLOUT;

		if (fd[i].revents)
			npoll++;
	}

	return npoll;
}
#endif				/* USE_SELECT */