xref: /netbsd/usr.bin/make/job.c (revision c4a72b64)
1 /*	$NetBSD: job.c,v 1.74 2002/11/26 05:30:01 enami Exp $	*/
2 
3 /*
4  * Copyright (c) 1988, 1989, 1990 The Regents of the University of California.
5  * Copyright (c) 1988, 1989 by Adam de Boor
6  * Copyright (c) 1989 by Berkeley Softworks
7  * All rights reserved.
8  *
9  * This code is derived from software contributed to Berkeley by
10  * Adam de Boor.
11  *
12  * Redistribution and use in source and binary forms, with or without
13  * modification, are permitted provided that the following conditions
14  * are met:
15  * 1. Redistributions of source code must retain the above copyright
16  *    notice, this list of conditions and the following disclaimer.
17  * 2. Redistributions in binary form must reproduce the above copyright
18  *    notice, this list of conditions and the following disclaimer in the
19  *    documentation and/or other materials provided with the distribution.
20  * 3. All advertising materials mentioning features or use of this software
21  *    must display the following acknowledgement:
22  *	This product includes software developed by the University of
23  *	California, Berkeley and its contributors.
24  * 4. Neither the name of the University nor the names of its contributors
25  *    may be used to endorse or promote products derived from this software
26  *    without specific prior written permission.
27  *
28  * THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND
29  * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
30  * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
31  * ARE DISCLAIMED.  IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE
32  * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
33  * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
34  * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
35  * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
36  * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
37  * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
38  * SUCH DAMAGE.
39  */
40 
41 #ifdef MAKE_BOOTSTRAP
42 static char rcsid[] = "$NetBSD: job.c,v 1.74 2002/11/26 05:30:01 enami Exp $";
43 #else
44 #include <sys/cdefs.h>
45 #ifndef lint
46 #if 0
47 static char sccsid[] = "@(#)job.c	8.2 (Berkeley) 3/19/94";
48 #else
49 __RCSID("$NetBSD: job.c,v 1.74 2002/11/26 05:30:01 enami Exp $");
50 #endif
51 #endif /* not lint */
52 #endif
53 
54 /*-
55  * job.c --
56  *	handle the creation etc. of our child processes.
57  *
58  * Interface:
59  *	Job_Make  	    	Start the creation of the given target.
60  *
61  *	Job_CatchChildren   	Check for and handle the termination of any
62  *	    	  	    	children. This must be called reasonably
63  *	    	  	    	frequently to keep the whole make going at
64  *	    	  	    	a decent clip, since job table entries aren't
65  *	    	  	    	removed until their process is caught this way.
66  *	    	  	    	Its single argument is TRUE if the function
67  *	    	  	    	should block waiting for a child to terminate.
68  *
69  *	Job_CatchOutput	    	Print any output our children have produced.
70  *	    	  	    	Should also be called fairly frequently to
71  *	    	  	    	keep the user informed of what's going on.
72  *	    	  	    	If no output is waiting, it will block for
73  *	    	  	    	a time given by the SEL_* constants, below,
74  *	    	  	    	or until output is ready.
75  *
76  *	Job_Init  	    	Called to intialize this module. in addition,
77  *	    	  	    	any commands attached to the .BEGIN target
78  *	    	  	    	are executed before this function returns.
79  *	    	  	    	Hence, the makefile must have been parsed
80  *	    	  	    	before this function is called.
81  *
82  *	Job_End  	    	Cleanup any memory used.
83  *
84  *	Job_Empty 	    	Return TRUE if the job table is completely
85  *	    	  	    	empty.
86  *
87  *	Job_ParseShell	    	Given the line following a .SHELL target, parse
88  *	    	  	    	the line as a shell specification. Returns
89  *	    	  	    	FAILURE if the spec was incorrect.
90  *
91  *	Job_Finish	    	Perform any final processing which needs doing.
92  *	    	  	    	This includes the execution of any commands
93  *	    	  	    	which have been/were attached to the .END
94  *	    	  	    	target. It should only be called when the
95  *	    	  	    	job table is empty.
96  *
97  *	Job_AbortAll	    	Abort all currently running jobs. It doesn't
98  *	    	  	    	handle output or do anything for the jobs,
99  *	    	  	    	just kills them. It should only be called in
100  *	    	  	    	an emergency, as it were.
101  *
102  *	Job_CheckCommands   	Verify that the commands for a target are
103  *	    	  	    	ok. Provide them if necessary and possible.
104  *
105  *	Job_Touch 	    	Update a target without really updating it.
106  *
107  *	Job_Wait  	    	Wait for all currently-running jobs to finish.
108  */
109 
110 #include <sys/types.h>
111 #include <sys/stat.h>
112 #include <sys/file.h>
113 #include <sys/time.h>
114 #include <sys/wait.h>
115 
116 #include <errno.h>
117 #include <fcntl.h>
118 #ifndef RMT_WILL_WATCH
119 #ifndef USE_SELECT
120 #include <poll.h>
121 #endif
122 #endif
123 #include <signal.h>
124 #include <stdio.h>
125 #include <string.h>
126 #include <utime.h>
127 
128 #include "make.h"
129 #include "hash.h"
130 #include "dir.h"
131 #include "job.h"
132 #include "pathnames.h"
133 #include "trace.h"
134 #ifdef REMOTE
135 #include "rmt.h"
136 # define STATIC
137 #else
138 # define STATIC static
139 #endif
140 
141 /*
142  * error handling variables
143  */
144 static int     	errors = 0;	    /* number of errors reported */
145 static int    	aborting = 0;	    /* why is the make aborting? */
146 #define ABORT_ERROR	1   	    /* Because of an error */
147 #define ABORT_INTERRUPT	2   	    /* Because it was interrupted */
148 #define ABORT_WAIT	3   	    /* Waiting for jobs to finish */
149 
150 /*
151  * XXX: Avoid SunOS bug... FILENO() is fp->_file, and file
152  * is a char! So when we go above 127 we turn negative!
153  */
154 #define FILENO(a) ((unsigned) fileno(a))
155 
156 /*
157  * post-make command processing. The node postCommands is really just the
158  * .END target but we keep it around to avoid having to search for it
159  * all the time.
160  */
161 static GNode   	  *postCommands = NILGNODE;
162 				    /* node containing commands to execute when
163 				     * everything else is done */
164 static int     	  numCommands; 	    /* The number of commands actually printed
165 				     * for a target. Should this number be
166 				     * 0, no shell will be executed. */
167 
168 /*
169  * Return values from JobStart.
170  */
171 #define JOB_RUNNING	0   	/* Job is running */
172 #define JOB_ERROR 	1   	/* Error in starting the job */
173 #define JOB_FINISHED	2   	/* The job is already finished */
174 #define JOB_STOPPED	3   	/* The job is stopped */
175 
176 
177 
178 /*
179  * Descriptions for various shells.
180  */
181 static Shell    shells[] = {
182     /*
183      * CSH description. The csh can do echo control by playing
184      * with the setting of the 'echo' shell variable. Sadly,
185      * however, it is unable to do error control nicely.
186      */
187 {
188     "csh",
189     TRUE, "unset verbose", "set verbose", "unset verbose", 10,
190     FALSE, "echo \"%s\"\n", "csh -c \"%s || exit 0\"",
191     "v", "e",
192 },
193     /*
194      * SH description. Echo control is also possible and, under
195      * sun UNIX anyway, one can even control error checking.
196      */
197 {
198     "sh",
199     TRUE, "set -", "set -v", "set -", 5,
200     TRUE, "set -e", "set +e",
201 #ifdef OLDBOURNESHELL
202     FALSE, "echo \"%s\"\n", "sh -c '%s || exit 0'\n",
203 #endif
204 #ifdef __NetBSD__
205     "vq",
206 #else
207     "v",
208 #endif
209     "e",
210 },
211     /*
212      * UNKNOWN.
213      */
214 {
215     (char *) 0,
216     FALSE, (char *) 0, (char *) 0, (char *) 0, 0,
217     FALSE, (char *) 0, (char *) 0,
218     (char *) 0, (char *) 0,
219 }
220 };
221 static Shell 	*commandShell = &shells[DEFSHELL];/* this is the shell to
222 						   * which we pass all
223 						   * commands in the Makefile.
224 						   * It is set by the
225 						   * Job_ParseShell function */
226 static char   	*shellPath = NULL,		  /* full pathname of
227 						   * executable image */
228                	*shellName = NULL,	      	  /* last component of shell */
229 		*shellArgv = NULL;		  /* Custom shell args */
230 
231 
232 static int  	maxJobs;    	/* The most children we can run at once */
233 static int  	maxLocal;    	/* The most local ones we can have */
234 STATIC int     	nJobs;	    	/* The number of children currently running */
235 STATIC int	nLocal;    	/* The number of local children */
236 STATIC Lst     	jobs;		/* The structures that describe them */
237 static Boolean	wantToken;	/* we want a token */
238 
239 /*
240  * Set of descriptors of pipes connected to
241  * the output channels of children
242  */
243 #ifndef RMT_WILL_WATCH
244 static struct pollfd *fds = NULL;
245 static Job **jobfds = NULL;
246 static int nfds = 0;
247 static int maxfds = 0;
248 static void watchfd(Job *);
249 static void clearfd(Job *);
250 static int readyfd(Job *);
251 #define JBSTART 256
252 #define JBFACTOR 2
253 #endif
254 
255 STATIC GNode   	*lastNode;	/* The node for which output was most recently
256 				 * produced. */
257 STATIC char    	*targFmt;   	/* Format string to use to head output from a
258 				 * job when it's not the most-recent job heard
259 				 * from */
260 static Job tokenWaitJob;	/* token wait pseudo-job */
261 int	job_pipe[2] = { -1, -1 }; /* job server pipes. */
262 
263 static Job childExitJob;	/* child exit pseudo-job */
264 int	exit_pipe[2] = { -1, -1 }; /* child exit signal pipe. */
265 
266 #ifdef REMOTE
267 # define TARG_FMT  "--- %s at %s ---\n" /* Default format */
268 # define MESSAGE(fp, gn) \
269 	(void) fprintf(fp, targFmt, gn->name, gn->rem.hname)
270 #else
271 # define TARG_FMT  "--- %s ---\n" /* Default format */
272 # define MESSAGE(fp, gn) \
273 	(void) fprintf(fp, targFmt, gn->name)
274 #endif
275 
276 /*
277  * When JobStart attempts to run a job remotely but can't, and isn't allowed
278  * to run the job locally, or when Job_CatchChildren detects a job that has
279  * been migrated home, the job is placed on the stoppedJobs queue to be run
280  * when the next job finishes.
281  */
282 STATIC Lst	stoppedJobs;	/* Lst of Job structures describing
283 				 * jobs that were stopped due to concurrency
284 				 * limits or migration home */
285 
286 
287 sigset_t	caught_signals;	/* Set of signals we handle */
288 #if defined(USE_PGRP) && defined(SYSV)
289 # define KILL(pid, sig)		kill(-(pid), (sig))
290 #else
291 # if defined(USE_PGRP)
292 #  define KILL(pid, sig)	killpg((pid), (sig))
293 # else
294 #  define KILL(pid, sig)	kill((pid), (sig))
295 # endif
296 #endif
297 
298 /*
299  * Grmpf... There is no way to set bits of the wait structure
300  * anymore with the stupid W*() macros. I liked the union wait
301  * stuff much more. So, we devise our own macros... This is
302  * really ugly, use dramamine sparingly. You have been warned.
303  */
304 #ifndef W_STOPCODE
305 #define W_STOPCODE(sig) (((sig) << 8) | 0177)
306 #endif
307 #ifndef W_EXITCODE
308 #define W_EXITCODE(ret, sig) ((ret << 8) | (sig))
309 #endif
310 
311 static int JobCondPassSig(ClientData, ClientData);
312 static void JobPassSig(int);
313 static void JobChildSig(int);
314 #ifdef USE_PGRP
315 static void JobContinueSig(int);
316 #endif
317 static int JobCmpPid(ClientData, ClientData);
318 static int JobPrintCommand(ClientData, ClientData);
319 static int JobSaveCommand(ClientData, ClientData);
320 static void JobClose(Job *);
321 #ifdef REMOTE
322 static int JobCmpRmtID(ClientData, ClientData);
323 # ifdef RMT_WILL_WATCH
324 static void JobLocalInput(int, Job *);
325 # endif
326 #else
327 static void JobFinish(Job *, int *);
328 static void JobExec(Job *, char **);
329 #endif
330 static void JobMakeArgv(Job *, char **);
331 static int JobRestart(Job *);
332 static int JobStart(GNode *, int, Job *);
333 static char *JobOutput(Job *, char *, char *, int);
334 static void JobDoOutput(Job *, Boolean);
335 static Shell *JobMatchShell(char *);
336 static void JobInterrupt(int, int);
337 static void JobRestartJobs(void);
338 static void JobTokenAdd(void);
339 static void JobSigLock(sigset_t *);
340 static void JobSigUnlock(sigset_t *);
341 static void JobSigReset(void);
342 
343 
344 
345 /*
346  * JobSigLock/JobSigUnlock
347  *
348  * Signal lock routines to get exclusive access. Currently used to
349  * protect `jobs' and `stoppedJobs' list manipulations.
350  */
351 static void JobSigLock(sigset_t *omaskp)
352 {
353 	if (sigprocmask(SIG_BLOCK, &caught_signals, omaskp) != 0) {
354 		Punt("JobSigLock: sigprocmask: %s", strerror(errno));
355 	sigemptyset(omaskp);
356 	}
357 }
358 
359 static void JobSigUnlock(sigset_t *omaskp)
360 {
361 	(void) sigprocmask(SIG_SETMASK, omaskp, NULL);
362 }
363 
364 /*-
365  *-----------------------------------------------------------------------
366  * JobCondPassSig --
367  *	Pass a signal to a job if the job is remote or if USE_PGRP
368  *	is defined.
369  *
370  * Input:
371  *	jobp		Job to biff
372  *	signop		Signal to send it
373  *
374  * Results:
375  *	=== 0
376  *
377  * Side Effects:
378  *	None, except the job may bite it.
379  *
380  *-----------------------------------------------------------------------
381  */
382 static int
383 JobCondPassSig(ClientData jobp, ClientData signop)
384 {
385     Job	*job = (Job *) jobp;
386     int	signo = *(int *) signop;
387 #ifdef RMT_WANTS_SIGNALS
388     if (job->flags & JOB_REMOTE) {
389 	(void) Rmt_Signal(job, signo);
390     } else {
391 	KILL(job->pid, signo);
392     }
393 #else
394     /*
395      * Assume that sending the signal to job->pid will signal any remote
396      * job as well.
397      */
398     if (DEBUG(JOB)) {
399 	(void) fprintf(stdout,
400 		       "JobCondPassSig passing signal %d to child %d.\n",
401 		       signo, job->pid);
402 	(void) fflush(stdout);
403     }
404     KILL(job->pid, signo);
405 #endif
406     return 0;
407 }
408 
409 /*-
410  *-----------------------------------------------------------------------
411  * JobChldSig --
412  *	SIGCHLD handler.
413  *
414  * Input:
415  *	signo		The signal number we've received
416  *
417  * Results:
418  *	None.
419  *
420  * Side Effects:
421  *	Sends a token on the child exit pipe to wake us up from
422  *	select()/poll().
423  *
424  *-----------------------------------------------------------------------
425  */
426 static void
427 JobChildSig(int signo)
428 {
429     write(exit_pipe[1], ".", 1);
430 }
431 
432 
433 #ifdef USE_PGRP
434 /*-
435  *-----------------------------------------------------------------------
436  * JobContinueSig --
437  *	Resume all stopped jobs.
438  *
439  * Input:
440  *	signo		The signal number we've received
441  *
442  * Results:
443  *	None.
444  *
445  * Side Effects:
446  *	Jobs start running again.
447  *
448  *-----------------------------------------------------------------------
449  */
450 static void
451 JobContinueSig(int signo)
452 {
453     JobRestartJobs();
454 }
455 #endif
456 
457 /*-
458  *-----------------------------------------------------------------------
459  * JobPassSig --
460  *	Pass a signal on to all remote jobs and to all local jobs if
461  *	USE_PGRP is defined, then die ourselves.
462  *
463  * Input:
464  *	signo		The signal number we've received
465  *
466  * Results:
467  *	None.
468  *
469  * Side Effects:
470  *	We die by the same signal.
471  *
472  *-----------------------------------------------------------------------
473  */
474 static void
475 JobPassSig(int signo)
476 {
477     sigset_t nmask, omask;
478     struct sigaction act;
479     int sigcont;
480 
481     if (DEBUG(JOB)) {
482 	(void) fprintf(stdout, "JobPassSig(%d) called.\n", signo);
483 	(void) fflush(stdout);
484     }
485     Lst_ForEach(jobs, JobCondPassSig, (ClientData) &signo);
486 
487     /*
488      * Deal with proper cleanup based on the signal received. We only run
489      * the .INTERRUPT target if the signal was in fact an interrupt. The other
490      * three termination signals are more of a "get out *now*" command.
491      */
492     if (signo == SIGINT) {
493 	JobInterrupt(TRUE, signo);
494     } else if ((signo == SIGHUP) || (signo == SIGTERM) || (signo == SIGQUIT)) {
495 	JobInterrupt(FALSE, signo);
496     }
497 
498     /*
499      * Leave gracefully if SIGQUIT, rather than core dumping.
500      */
501     if (signo == SIGQUIT) {
502 	Finish(0);
503     }
504 
505     if (signo == SIGTSTP) {
506 	Job_CatchChildren(FALSE);
507     }
508     /*
509      * Send ourselves the signal now we've given the message to everyone else.
510      * Note we block everything else possible while we're getting the signal.
511      * This ensures that all our jobs get continued when we wake up before
512      * we take any other signal.
513      */
514     sigfillset(&nmask);
515     sigdelset(&nmask, signo);
516     (void) sigprocmask(SIG_SETMASK, &nmask, &omask);
517 
518     act.sa_handler = SIG_DFL;
519     sigemptyset(&act.sa_mask);
520     act.sa_flags = 0;
521     (void) sigaction(signo, &act, NULL);
522 
523     if (DEBUG(JOB)) {
524 	(void) fprintf(stdout,
525 		       "JobPassSig passing signal %d to self.\n", signo);
526 	(void) fflush(stdout);
527     }
528 
529     (void) kill(getpid(), signo);
530     if (signo != SIGTSTP) {
531 	sigcont = SIGCONT;
532 	Lst_ForEach(jobs, JobCondPassSig, (ClientData) &sigcont);
533     }
534 
535     /* Restore handler and signal mask */
536     act.sa_handler = JobPassSig;
537     (void) sigaction(signo, &act, NULL);
538     (void) sigprocmask(SIG_SETMASK, &omask, NULL);
539 }
540 
541 /*-
542  *-----------------------------------------------------------------------
543  * JobCmpPid  --
544  *	Compare the pid of the job with the given pid and return 0 if they
545  *	are equal. This function is called from Job_CatchChildren via
546  *	Lst_Find to find the job descriptor of the finished job.
547  *
548  * Input:
549  *	job		job to examine
550  *	pid		process id desired
551  *
552  * Results:
553  *	0 if the pid's match
554  *
555  * Side Effects:
556  *	None
557  *-----------------------------------------------------------------------
558  */
559 static int
560 JobCmpPid(ClientData job, ClientData pid)
561 {
562     return *(int *) pid - ((Job *) job)->pid;
563 }
564 
565 #ifdef REMOTE
566 /*-
567  *-----------------------------------------------------------------------
568  * JobCmpRmtID  --
569  *	Compare the rmtID of the job with the given rmtID and return 0 if they
570  *	are equal.
571  *
572  * Input:
573  *	job		job to examine
574  *	rmtID		remote id desired
575  *
576  * Results:
577  *	0 if the rmtID's match
578  *
579  * Side Effects:
580  *	None.
581  *-----------------------------------------------------------------------
582  */
583 static int
584 JobCmpRmtID(ClientData job, ClientData rmtID)
585 {
586     return(*(int *) rmtID - ((Job *) job)->rmtID);
587 }
588 #endif
589 
590 /*-
591  *-----------------------------------------------------------------------
592  * JobPrintCommand  --
593  *	Put out another command for the given job. If the command starts
594  *	with an @ or a - we process it specially. In the former case,
595  *	so long as the -s and -n flags weren't given to make, we stick
596  *	a shell-specific echoOff command in the script. In the latter,
597  *	we ignore errors for the entire job, unless the shell has error
598  *	control.
599  *	If the command is just "..." we take all future commands for this
600  *	job to be commands to be executed once the entire graph has been
601  *	made and return non-zero to signal that the end of the commands
602  *	was reached. These commands are later attached to the postCommands
603  *	node and executed by Job_End when all things are done.
604  *	This function is called from JobStart via Lst_ForEach.
605  *
606  * Input:
607  *	cmdp		command string to print
608  *	jobp		job for which to print it
609  *
610  * Results:
611  *	Always 0, unless the command was "..."
612  *
613  * Side Effects:
614  *	If the command begins with a '-' and the shell has no error control,
615  *	the JOB_IGNERR flag is set in the job descriptor.
616  *	If the command is "..." and we're not ignoring such things,
617  *	tailCmds is set to the successor node of the cmd.
618  *	numCommands is incremented if the command is actually printed.
619  *-----------------------------------------------------------------------
620  */
621 static int
622 JobPrintCommand(ClientData cmdp, ClientData jobp)
623 {
624     Boolean	  noSpecials;	    /* true if we shouldn't worry about
625 				     * inserting special commands into
626 				     * the input stream. */
627     Boolean       shutUp = FALSE;   /* true if we put a no echo command
628 				     * into the command file */
629     Boolean	  errOff = FALSE;   /* true if we turned error checking
630 				     * off before printing the command
631 				     * and need to turn it back on */
632     char       	  *cmdTemplate;	    /* Template to use when printing the
633 				     * command */
634     char    	  *cmdStart;	    /* Start of expanded command */
635     char     	  *cmd = (char *) cmdp;
636     Job           *job = (Job *) jobp;
637     char	*cp;
638 
639     noSpecials = NoExecute(job->node);
640 
641     if (strcmp(cmd, "...") == 0) {
642 	job->node->type |= OP_SAVE_CMDS;
643 	if ((job->flags & JOB_IGNDOTS) == 0) {
644 	    job->tailCmds = Lst_Succ(Lst_Member(job->node->commands,
645 						(ClientData)cmd));
646 	    return 1;
647 	}
648 	return 0;
649     }
650 
651 #define DBPRINTF(fmt, arg) if (DEBUG(JOB)) {	\
652 	(void) fprintf(stdout, fmt, arg); 	\
653 	(void) fflush(stdout); 			\
654     }						\
655    (void) fprintf(job->cmdFILE, fmt, arg);	\
656    (void) fflush(job->cmdFILE);
657 
658     numCommands += 1;
659 
660     cmdStart = cmd = Var_Subst(NULL, cmd, job->node, FALSE);
661 
662     cmdTemplate = "%s\n";
663 
664     /*
665      * Check for leading @' and -'s to control echoing and error checking.
666      */
667     while (*cmd == '@' || *cmd == '-') {
668 	if (*cmd == '@') {
669 	    shutUp = TRUE;
670 	} else {
671 	    errOff = TRUE;
672 	}
673 	cmd++;
674     }
675 
676     while (isspace((unsigned char) *cmd))
677 	cmd++;
678 
679     if (shutUp) {
680 	if (!(job->flags & JOB_SILENT) && !noSpecials &&
681 	    commandShell->hasEchoCtl) {
682 		DBPRINTF("%s\n", commandShell->echoOff);
683 	} else {
684 	    shutUp = FALSE;
685 	}
686     }
687 
688     if (errOff) {
689 	if ( !(job->flags & JOB_IGNERR) && !noSpecials) {
690 	    if (commandShell->hasErrCtl) {
691 		/*
692 		 * we don't want the error-control commands showing
693 		 * up either, so we turn off echoing while executing
694 		 * them. We could put another field in the shell
695 		 * structure to tell JobDoOutput to look for this
696 		 * string too, but why make it any more complex than
697 		 * it already is?
698 		 */
699 		if (!(job->flags & JOB_SILENT) && !shutUp &&
700 		    commandShell->hasEchoCtl) {
701 			DBPRINTF("%s\n", commandShell->echoOff);
702 			DBPRINTF("%s\n", commandShell->ignErr);
703 			DBPRINTF("%s\n", commandShell->echoOn);
704 		} else {
705 		    DBPRINTF("%s\n", commandShell->ignErr);
706 		}
707 	    } else if (commandShell->ignErr &&
708 		      (*commandShell->ignErr != '\0'))
709 	    {
710 		/*
711 		 * The shell has no error control, so we need to be
712 		 * weird to get it to ignore any errors from the command.
713 		 * If echoing is turned on, we turn it off and use the
714 		 * errCheck template to echo the command. Leave echoing
715 		 * off so the user doesn't see the weirdness we go through
716 		 * to ignore errors. Set cmdTemplate to use the weirdness
717 		 * instead of the simple "%s\n" template.
718 		 */
719 		if (!(job->flags & JOB_SILENT) && !shutUp &&
720 		    commandShell->hasEchoCtl) {
721 			DBPRINTF("%s\n", commandShell->echoOff);
722 			DBPRINTF(commandShell->errCheck, cmd);
723 			shutUp = TRUE;
724 		}
725 		cmdTemplate = commandShell->ignErr;
726 		/*
727 		 * The error ignoration (hee hee) is already taken care
728 		 * of by the ignErr template, so pretend error checking
729 		 * is still on.
730 		 */
731 		errOff = FALSE;
732 	    } else {
733 		errOff = FALSE;
734 	    }
735 	} else {
736 	    errOff = FALSE;
737 	}
738     }
739 
740     if (DEBUG(SHELL) && strcmp(shellName, "sh") == 0 &&
741 	(job->flags & JOB_TRACED) == 0) {
742 	    DBPRINTF("set -%s\n", "x");
743 	    job->flags |= JOB_TRACED;
744     }
745 
746     if ((cp = Check_Cwd_Cmd(cmd)) != NULL) {
747 	    DBPRINTF("test -d %s && ", cp);
748 	    DBPRINTF("cd %s; ", cp);
749     }
750     DBPRINTF(cmdTemplate, cmd);
751     free(cmdStart);
752 
753     if (errOff) {
754 	/*
755 	 * If echoing is already off, there's no point in issuing the
756 	 * echoOff command. Otherwise we issue it and pretend it was on
757 	 * for the whole command...
758 	 */
759 	if (!shutUp && !(job->flags & JOB_SILENT) && commandShell->hasEchoCtl){
760 	    DBPRINTF("%s\n", commandShell->echoOff);
761 	    shutUp = TRUE;
762 	}
763 	DBPRINTF("%s\n", commandShell->errCheck);
764     }
765     if (shutUp) {
766 	DBPRINTF("%s\n", commandShell->echoOn);
767     }
768     return 0;
769 }
770 
771 /*-
772  *-----------------------------------------------------------------------
773  * JobSaveCommand --
774  *	Save a command to be executed when everything else is done.
775  *	Callback function for JobFinish...
776  *
777  * Results:
778  *	Always returns 0
779  *
780  * Side Effects:
781  *	The command is tacked onto the end of postCommands's commands list.
782  *
783  *-----------------------------------------------------------------------
784  */
785 static int
786 JobSaveCommand(ClientData cmd, ClientData gn)
787 {
788     cmd = (ClientData) Var_Subst(NULL, (char *) cmd, (GNode *) gn, FALSE);
789     (void) Lst_AtEnd(postCommands->commands, cmd);
790     return(0);
791 }
792 
793 
794 /*-
795  *-----------------------------------------------------------------------
796  * JobClose --
797  *	Called to close both input and output pipes when a job is finished.
798  *
799  * Results:
800  *	Nada
801  *
802  * Side Effects:
803  *	The file descriptors associated with the job are closed.
804  *
805  *-----------------------------------------------------------------------
806  */
807 static void
808 JobClose(Job *job)
809 {
810     if (usePipes && (job->flags & JOB_FIRST)) {
811 #ifdef RMT_WILL_WATCH
812 	Rmt_Ignore(job->inPipe);
813 #else
814 	clearfd(job);
815 #endif
816 	if (job->outPipe != job->inPipe) {
817 	   (void) close(job->outPipe);
818 	}
819 	JobDoOutput(job, TRUE);
820 	(void) close(job->inPipe);
821     } else {
822 	(void) close(job->outFd);
823 	JobDoOutput(job, TRUE);
824     }
825 }
826 
827 /*-
828  *-----------------------------------------------------------------------
829  * JobFinish  --
830  *	Do final processing for the given job including updating
831  *	parents and starting new jobs as available/necessary. Note
832  *	that we pay no attention to the JOB_IGNERR flag here.
833  *	This is because when we're called because of a noexecute flag
834  *	or something, jstat.w_status is 0 and when called from
835  *	Job_CatchChildren, the status is zeroed if it s/b ignored.
836  *
837  * Input:
838  *	job		job to finish
839  *	status		sub-why job went away
840  *
841  * Results:
842  *	None
843  *
844  * Side Effects:
845  *	Some nodes may be put on the toBeMade queue.
846  *	Final commands for the job are placed on postCommands.
847  *
848  *	If we got an error and are aborting (aborting == ABORT_ERROR) and
849  *	the job list is now empty, we are done for the day.
850  *	If we recognized an error (errors !=0), we set the aborting flag
851  *	to ABORT_ERROR so no more jobs will be started.
852  *-----------------------------------------------------------------------
853  */
854 /*ARGSUSED*/
855 static void
856 JobFinish(Job *job, int *status)
857 {
858     Boolean 	 done;
859 
860     if ((WIFEXITED(*status) &&
861 	 (((WEXITSTATUS(*status) != 0) && !(job->flags & JOB_IGNERR)))) ||
862 	WIFSIGNALED(*status))
863     {
864 	/*
865 	 * If it exited non-zero and either we're doing things our
866 	 * way or we're not ignoring errors, the job is finished.
867 	 * Similarly, if the shell died because of a signal
868 	 * the job is also finished. In these
869 	 * cases, finish out the job's output before printing the exit
870 	 * status...
871 	 */
872 #ifdef REMOTE
873 	KILL(job->pid, SIGCONT);
874 #endif
875 	JobClose(job);
876 	if (job->cmdFILE != NULL && job->cmdFILE != stdout) {
877 	   (void) fclose(job->cmdFILE);
878 	   job->cmdFILE = NULL;
879 	}
880 	done = TRUE;
881 #ifdef REMOTE
882 	if (job->flags & JOB_REMOTE)
883 	    Rmt_Done(job->rmtID, job->node);
884 #endif
885     } else if (WIFEXITED(*status)) {
886 	/*
887 	 * Deal with ignored errors in -B mode. We need to print a message
888 	 * telling of the ignored error as well as setting status.w_status
889 	 * to 0 so the next command gets run. To do this, we set done to be
890 	 * TRUE if in -B mode and the job exited non-zero.
891 	 */
892 	done = WEXITSTATUS(*status) != 0;
893 	/*
894 	 * Old comment said: "Note we don't
895 	 * want to close down any of the streams until we know we're at the
896 	 * end."
897 	 * But we do. Otherwise when are we going to print the rest of the
898 	 * stuff?
899 	 */
900 	JobClose(job);
901 #ifdef REMOTE
902 	if (job->flags & JOB_REMOTE)
903 	    Rmt_Done(job->rmtID, job->node);
904 #endif /* REMOTE */
905     } else {
906 	/*
907 	 * No need to close things down or anything.
908 	 */
909 	done = FALSE;
910     }
911 
912     if (done ||
913 	WIFSTOPPED(*status) ||
914 	(WIFSIGNALED(*status) && (WTERMSIG(*status) == SIGCONT)))
915     {
916 	FILE	  *out;
917 
918 	if (compatMake && !usePipes && (job->flags & JOB_IGNERR)) {
919 	    /*
920 	     * If output is going to a file and this job is ignoring
921 	     * errors, arrange to have the exit status sent to the
922 	     * output file as well.
923 	     */
924 	    out = fdopen(job->outFd, "w");
925 	    if (out == NULL)
926 		Punt("Cannot fdopen");
927 	} else {
928 	    out = stdout;
929 	}
930 
931 	if (WIFEXITED(*status)) {
932 	    if (DEBUG(JOB)) {
933 		(void) fprintf(stdout, "Process %d [%s] exited.\n",
934 				job->pid, job->node->name);
935 		(void) fflush(stdout);
936 	    }
937 	    if (WEXITSTATUS(*status) != 0) {
938 		if (usePipes && job->node != lastNode) {
939 		    MESSAGE(out, job->node);
940 		    lastNode = job->node;
941 		}
942 		(void) fprintf(out, "*** [%s] Error code %d%s\n",
943 				job->node->name,
944 			       WEXITSTATUS(*status),
945 			       (job->flags & JOB_IGNERR) ? "(ignored)" : "");
946 
947 		if (job->flags & JOB_IGNERR) {
948 		    *status = 0;
949 		}
950 	    } else if (DEBUG(JOB)) {
951 		if (usePipes && job->node != lastNode) {
952 		    MESSAGE(out, job->node);
953 		    lastNode = job->node;
954 		}
955 		(void) fprintf(out, "*** [%s] Completed successfully\n",
956 				job->node->name);
957 	    }
958 	} else if (WIFSTOPPED(*status) && WSTOPSIG(*status) != SIGCONT) {
959 	    if (DEBUG(JOB)) {
960 		(void) fprintf(stdout, "Process %d (%s) stopped.\n",
961 				job->pid, job->node->name);
962 		(void) fflush(stdout);
963 	    }
964 	    if (usePipes && job->node != lastNode) {
965 		MESSAGE(out, job->node);
966 		lastNode = job->node;
967 	    }
968 	    if (!(job->flags & JOB_REMIGRATE)) {
969 		switch (WSTOPSIG(*status)) {
970 		case SIGTSTP:
971 		    (void) fprintf(out, "*** [%s] Suspended\n",
972 				job->node->name);
973 		    break;
974 		case SIGSTOP:
975 		    (void) fprintf(out, "*** [%s] Stopped\n",
976 				job->node->name);
977 		    break;
978 		default:
979 		    (void) fprintf(out, "*** [%s] Stopped -- signal %d\n",
980 			job->node->name, WSTOPSIG(*status));
981 		}
982 	    }
983 	    job->flags |= JOB_RESUME;
984 	    (void)Lst_AtEnd(stoppedJobs, (ClientData)job);
985 #ifdef REMOTE
986 	    if (job->flags & JOB_REMIGRATE)
987 		JobRestart(job);
988 #endif
989 	    (void) fflush(out);
990 	    return;
991 	} else if (WIFSTOPPED(*status) &&  WSTOPSIG(*status) == SIGCONT) {
992 	    /*
993 	     * If the beastie has continued, shift the Job from the stopped
994 	     * list to the running one (or re-stop it if concurrency is
995 	     * exceeded) and go and get another child.
996 	     */
997 	    if (job->flags & (JOB_RESUME|JOB_REMIGRATE|JOB_RESTART)) {
998 		if (usePipes && job->node != lastNode) {
999 		    MESSAGE(out, job->node);
1000 		    lastNode = job->node;
1001 		}
1002 		(void) fprintf(out, "*** [%s] Continued\n", job->node->name);
1003 	    }
1004 	    if (!(job->flags & JOB_CONTINUING)) {
1005 		if (DEBUG(JOB)) {
1006 		    (void) fprintf(stdout,
1007 			   "Warning: process %d [%s] was not continuing.\n",
1008 			   job->pid, job->node->name);
1009 		    (void) fflush(stdout);
1010 		}
1011 #ifdef notdef
1012 		/*
1013 		 * We don't really want to restart a job from scratch just
1014 		 * because it continued, especially not without killing the
1015 		 * continuing process!  That's why this is ifdef'ed out.
1016 		 * FD - 9/17/90
1017 		 */
1018 		JobRestart(job);
1019 #endif
1020 	    }
1021 	    job->flags &= ~JOB_CONTINUING;
1022  	    Lst_AtEnd(jobs, (ClientData)job);
1023 	    nJobs += 1;
1024 	    if (!(job->flags & JOB_REMOTE)) {
1025 		if (DEBUG(JOB)) {
1026 		    (void) fprintf(stdout,
1027 				   "Process %d is continuing locally.\n",
1028 				   job->pid);
1029 		    (void) fflush(stdout);
1030   		}
1031 		nLocal += 1;
1032 	    }
1033 	    (void) fflush(out);
1034   	    return;
1035 	} else {
1036 	    if (usePipes && job->node != lastNode) {
1037 		MESSAGE(out, job->node);
1038 		lastNode = job->node;
1039 	    }
1040 	    (void) fprintf(out, "*** [%s] Signal %d\n",
1041 			job->node->name, WTERMSIG(*status));
1042 	}
1043 
1044 	(void) fflush(out);
1045     }
1046 
1047     /*
1048      * Now handle the -B-mode stuff. If the beast still isn't finished,
1049      * try and restart the job on the next command. If JobStart says it's
1050      * ok, it's ok. If there's an error, this puppy is done.
1051      */
1052     if (compatMake && (WIFEXITED(*status) &&
1053 	!Lst_IsAtEnd(job->node->commands))) {
1054 	switch (JobStart(job->node, job->flags & JOB_IGNDOTS, job)) {
1055 	case JOB_RUNNING:
1056 	    done = FALSE;
1057 	    break;
1058 	case JOB_ERROR:
1059 	    done = TRUE;
1060 	    *status = W_EXITCODE(1, 0);
1061 	    break;
1062 	case JOB_FINISHED:
1063 	    /*
1064 	     * If we got back a JOB_FINISHED code, JobStart has already
1065 	     * called Make_Update and freed the job descriptor. We set
1066 	     * done to false here to avoid fake cycles and double frees.
1067 	     * JobStart needs to do the update so we can proceed up the
1068 	     * graph when given the -n flag..
1069 	     */
1070 	    done = FALSE;
1071 	    break;
1072 	}
1073     } else {
1074 	done = TRUE;
1075     }
1076 
1077     if (done) {
1078 	Trace_Log(JOBEND, job);
1079 	if (!compatMake && !(job->flags & JOB_SPECIAL)) {
1080 	    if ((*status != 0) ||
1081 	        (aborting == ABORT_ERROR) ||
1082 	        (aborting == ABORT_INTERRUPT))
1083 		Job_TokenReturn();
1084 	}
1085 
1086     }
1087 
1088     if (done &&
1089 	(aborting != ABORT_ERROR) &&
1090 	(aborting != ABORT_INTERRUPT) &&
1091 	(*status == 0))
1092     {
1093 	/*
1094 	 * As long as we aren't aborting and the job didn't return a non-zero
1095 	 * status that we shouldn't ignore, we call Make_Update to update
1096 	 * the parents. In addition, any saved commands for the node are placed
1097 	 * on the .END target.
1098 	 */
1099 	if (job->tailCmds != NILLNODE) {
1100 	    Lst_ForEachFrom(job->node->commands, job->tailCmds,
1101 			     JobSaveCommand,
1102 			    (ClientData)job->node);
1103 	}
1104 	job->node->made = MADE;
1105 	if (!(job->flags & JOB_SPECIAL))
1106 	    Job_TokenReturn();
1107 	Make_Update(job->node);
1108 	free((Address)job);
1109     } else if (*status != 0) {
1110 	errors += 1;
1111 	free((Address)job);
1112     }
1113     JobRestartJobs();
1114 
1115     /*
1116      * Set aborting if any error.
1117      */
1118     if (errors && !keepgoing && (aborting != ABORT_INTERRUPT)) {
1119 	/*
1120 	 * If we found any errors in this batch of children and the -k flag
1121 	 * wasn't given, we set the aborting flag so no more jobs get
1122 	 * started.
1123 	 */
1124 	aborting = ABORT_ERROR;
1125     }
1126 
1127     if ((aborting == ABORT_ERROR) && Job_Empty()) {
1128 	/*
1129 	 * If we are aborting and the job table is now empty, we finish.
1130 	 */
1131 	Finish(errors);
1132     }
1133 }
1134 
1135 /*-
1136  *-----------------------------------------------------------------------
1137  * Job_Touch --
1138  *	Touch the given target. Called by JobStart when the -t flag was
1139  *	given
1140  *
1141  * Input:
1142  *	gn		the node of the file to touch
1143  *	silent		TRUE if should not print message
1144  *
1145  * Results:
1146  *	None
1147  *
1148  * Side Effects:
1149  *	The data modification of the file is changed. In addition, if the
1150  *	file did not exist, it is created.
1151  *-----------------------------------------------------------------------
1152  */
1153 void
1154 Job_Touch(GNode *gn, Boolean silent)
1155 {
1156     int		  streamID;   	/* ID of stream opened to do the touch */
1157     struct utimbuf times;	/* Times for utime() call */
1158 
1159     if (gn->type & (OP_JOIN|OP_USE|OP_USEBEFORE|OP_EXEC|OP_OPTIONAL|OP_PHONY)) {
1160 	/*
1161 	 * .JOIN, .USE, .ZEROTIME and .OPTIONAL targets are "virtual" targets
1162 	 * and, as such, shouldn't really be created.
1163 	 */
1164 	return;
1165     }
1166 
1167     if (!silent || NoExecute(gn)) {
1168 	(void) fprintf(stdout, "touch %s\n", gn->name);
1169 	(void) fflush(stdout);
1170     }
1171 
1172     if (NoExecute(gn)) {
1173 	return;
1174     }
1175 
1176     if (gn->type & OP_ARCHV) {
1177 	Arch_Touch(gn);
1178     } else if (gn->type & OP_LIB) {
1179 	Arch_TouchLib(gn);
1180     } else {
1181 	char	*file = gn->path ? gn->path : gn->name;
1182 
1183 	times.actime = times.modtime = now;
1184 	if (utime(file, &times) < 0){
1185 	    streamID = open(file, O_RDWR | O_CREAT, 0666);
1186 
1187 	    if (streamID >= 0) {
1188 		char	c;
1189 
1190 		/*
1191 		 * Read and write a byte to the file to change the
1192 		 * modification time, then close the file.
1193 		 */
1194 		if (read(streamID, &c, 1) == 1) {
1195 		    (void) lseek(streamID, (off_t)0, SEEK_SET);
1196 		    (void) write(streamID, &c, 1);
1197 		}
1198 
1199 		(void) close(streamID);
1200 	    } else {
1201 		(void) fprintf(stdout, "*** couldn't touch %s: %s",
1202 			       file, strerror(errno));
1203 		(void) fflush(stdout);
1204 	    }
1205 	}
1206     }
1207 }
1208 
1209 /*-
1210  *-----------------------------------------------------------------------
1211  * Job_CheckCommands --
1212  *	Make sure the given node has all the commands it needs.
1213  *
1214  * Input:
1215  *	gn		The target whose commands need verifying
1216  *	abortProc	Function to abort with message
1217  *
1218  * Results:
1219  *	TRUE if the commands list is/was ok.
1220  *
1221  * Side Effects:
1222  *	The node will have commands from the .DEFAULT rule added to it
1223  *	if it needs them.
1224  *-----------------------------------------------------------------------
1225  */
1226 Boolean
1227 Job_CheckCommands(GNode *gn, void (*abortProc)(char *, ...))
1228 {
1229     if (OP_NOP(gn->type) && Lst_IsEmpty(gn->commands) &&
1230 	(gn->type & OP_LIB) == 0) {
1231 	/*
1232 	 * No commands. Look for .DEFAULT rule from which we might infer
1233 	 * commands
1234 	 */
1235 	if ((DEFAULT != NILGNODE) && !Lst_IsEmpty(DEFAULT->commands)) {
1236 	    char *p1;
1237 	    /*
1238 	     * Make only looks for a .DEFAULT if the node was never the
1239 	     * target of an operator, so that's what we do too. If
1240 	     * a .DEFAULT was given, we substitute its commands for gn's
1241 	     * commands and set the IMPSRC variable to be the target's name
1242 	     * The DEFAULT node acts like a transformation rule, in that
1243 	     * gn also inherits any attributes or sources attached to
1244 	     * .DEFAULT itself.
1245 	     */
1246 	    Make_HandleUse(DEFAULT, gn);
1247 	    Var_Set(IMPSRC, Var_Value(TARGET, gn, &p1), gn, 0);
1248 	    if (p1)
1249 		free(p1);
1250 	} else if (Dir_MTime(gn) == 0) {
1251 	    /*
1252 	     * The node wasn't the target of an operator we have no .DEFAULT
1253 	     * rule to go on and the target doesn't already exist. There's
1254 	     * nothing more we can do for this branch. If the -k flag wasn't
1255 	     * given, we stop in our tracks, otherwise we just don't update
1256 	     * this node's parents so they never get examined.
1257 	     */
1258 	    static const char msg[] = ": don't know how to make";
1259 
1260 	    if (gn->type & OP_OPTIONAL) {
1261 		(void) fprintf(stdout, "%s%s %s(ignored)\n", progname,
1262 		    msg, gn->name);
1263 		(void) fflush(stdout);
1264 	    } else if (keepgoing) {
1265 		(void) fprintf(stdout, "%s%s %s(continuing)\n", progname,
1266 		    msg, gn->name);
1267 		(void) fflush(stdout);
1268   		return FALSE;
1269 	    } else {
1270 		(*abortProc)("%s%s %s. Stop", progname, msg, gn->name);
1271 		return FALSE;
1272 	    }
1273 	}
1274     }
1275     return TRUE;
1276 }
1277 #ifdef RMT_WILL_WATCH
1278 /*-
1279  *-----------------------------------------------------------------------
1280  * JobLocalInput --
1281  *	Handle a pipe becoming readable. Callback function for Rmt_Watch
1282  *
1283  * Input:
1284  *	stream		Stream that's ready (ignored)
1285  *	job		Job to which the stream belongs
1286  *
1287  * Results:
1288  *	None
1289  *
1290  * Side Effects:
1291  *	JobDoOutput is called.
1292  *
1293  *-----------------------------------------------------------------------
1294  */
1295 /*ARGSUSED*/
1296 static void
1297 JobLocalInput(int stream, Job *job)
1298 {
1299     JobDoOutput(job, FALSE);
1300 }
1301 #endif /* RMT_WILL_WATCH */
1302 
1303 /*-
1304  *-----------------------------------------------------------------------
1305  * JobExec --
1306  *	Execute the shell for the given job. Called from JobStart and
1307  *	JobRestart.
1308  *
1309  * Input:
1310  *	job		Job to execute
1311  *
1312  * Results:
1313  *	None.
1314  *
1315  * Side Effects:
1316  *	A shell is executed, outputs is altered and the Job structure added
1317  *	to the job table.
1318  *
1319  *-----------------------------------------------------------------------
1320  */
1321 static void
1322 JobExec(Job *job, char **argv)
1323 {
1324     int	    	  cpid;	    	/* ID of new child */
1325     sigset_t	  mask;
1326 
1327     job->flags &= ~JOB_TRACED;
1328 
1329     if (DEBUG(JOB)) {
1330 	int 	  i;
1331 
1332 	(void) fprintf(stdout, "Running %s %sly\n", job->node->name,
1333 		       job->flags&JOB_REMOTE?"remote":"local");
1334 	(void) fprintf(stdout, "\tCommand: ");
1335 	for (i = 0; argv[i] != NULL; i++) {
1336 	    (void) fprintf(stdout, "%s ", argv[i]);
1337 	}
1338  	(void) fprintf(stdout, "\n");
1339  	(void) fflush(stdout);
1340     }
1341 
1342     /*
1343      * Some jobs produce no output and it's disconcerting to have
1344      * no feedback of their running (since they produce no output, the
1345      * banner with their name in it never appears). This is an attempt to
1346      * provide that feedback, even if nothing follows it.
1347      */
1348     if ((lastNode != job->node) && (job->flags & JOB_FIRST) &&
1349 	!(job->flags & JOB_SILENT)) {
1350 	MESSAGE(stdout, job->node);
1351 	lastNode = job->node;
1352     }
1353 
1354 #ifdef RMT_NO_EXEC
1355     if (job->flags & JOB_REMOTE) {
1356 	goto jobExecFinish;
1357     }
1358 #endif /* RMT_NO_EXEC */
1359 
1360     /* No interruptions until this job is on the `jobs' list */
1361     JobSigLock(&mask);
1362 
1363     if ((cpid = vfork()) == -1) {
1364 	Punt("Cannot vfork: %s", strerror(errno));
1365     } else if (cpid == 0) {
1366 
1367 	/*
1368 	 * Reset all signal handlers; this is necessary because we also
1369 	 * need to unblock signals before we exec(2).
1370 	 */
1371 	JobSigReset();
1372 
1373 	/* Now unblock signals */
1374 	JobSigUnlock(&mask);
1375 
1376 	/*
1377 	 * Must duplicate the input stream down to the child's input and
1378 	 * reset it to the beginning (again). Since the stream was marked
1379 	 * close-on-exec, we must clear that bit in the new input.
1380 	 */
1381 	if (dup2(FILENO(job->cmdFILE), 0) == -1) {
1382 	    execError("dup2", "job->cmdFILE");
1383 	    _exit(1);
1384 	}
1385 	(void) fcntl(0, F_SETFD, 0);
1386 	(void) lseek(0, (off_t)0, SEEK_SET);
1387 
1388 	if (job->node->type & OP_MAKE) {
1389 		/*
1390 		 * Pass job token pipe to submakes.
1391 		 */
1392 		fcntl(job_pipe[0], F_SETFD, 0);
1393 		fcntl(job_pipe[1], F_SETFD, 0);
1394 	}
1395 
1396 	if (usePipes) {
1397 	    /*
1398 	     * Set up the child's output to be routed through the pipe
1399 	     * we've created for it.
1400 	     */
1401 	    if (dup2(job->outPipe, 1) == -1) {
1402 		execError("dup2", "job->outPipe");
1403 		_exit(1);
1404 	    }
1405 	} else {
1406 	    /*
1407 	     * We're capturing output in a file, so we duplicate the
1408 	     * descriptor to the temporary file into the standard
1409 	     * output.
1410 	     */
1411 	    if (dup2(job->outFd, 1) == -1) {
1412 		execError("dup2", "job->outFd");
1413 		_exit(1);
1414 	    }
1415 	}
1416 	/*
1417 	 * The output channels are marked close on exec. This bit was
1418 	 * duplicated by the dup2 (on some systems), so we have to clear
1419 	 * it before routing the shell's error output to the same place as
1420 	 * its standard output.
1421 	 */
1422 	(void) fcntl(1, F_SETFD, 0);
1423 	if (dup2(1, 2) == -1) {
1424 	    execError("dup2", "1, 2");
1425 	    _exit(1);
1426 	}
1427 
1428 #ifdef USE_PGRP
1429 	/*
1430 	 * We want to switch the child into a different process family so
1431 	 * we can kill it and all its descendants in one fell swoop,
1432 	 * by killing its process family, but not commit suicide.
1433 	 */
1434 # if defined(SYSV)
1435 	(void) setsid();
1436 # else
1437 	(void) setpgid(0, getpid());
1438 # endif
1439 #endif /* USE_PGRP */
1440 
1441 #ifdef REMOTE
1442 	if (job->flags & JOB_REMOTE) {
1443 	    Rmt_Exec(shellPath, argv, FALSE);
1444 	} else
1445 #endif /* REMOTE */
1446 	{
1447 	   (void) execv(shellPath, argv);
1448 	   execError("exec", shellPath);
1449 	}
1450 	_exit(1);
1451     } else {
1452 	job->pid = cpid;
1453 
1454 	Trace_Log(JOBSTART, job);
1455 
1456 	if (usePipes && (job->flags & JOB_FIRST)) {
1457 	    /*
1458 	     * The first time a job is run for a node, we set the current
1459 	     * position in the buffer to the beginning and mark another
1460 	     * stream to watch in the outputs mask
1461 	     */
1462 	    job->curPos = 0;
1463 
1464 #ifdef RMT_WILL_WATCH
1465 	    Rmt_Watch(job->inPipe, JobLocalInput, job);
1466 #else
1467 	    watchfd(job);
1468 #endif /* RMT_WILL_WATCH */
1469 	}
1470 
1471 	if (job->flags & JOB_REMOTE) {
1472 #ifndef REMOTE
1473 	    job->rmtID = 0;
1474 #else
1475 	    job->rmtID = Rmt_LastID(job->pid);
1476 #endif /* REMOTE */
1477 	} else {
1478 	    nLocal += 1;
1479 	    /*
1480 	     * XXX: Used to not happen if REMOTE. Why?
1481 	     */
1482 	    if (job->cmdFILE != NULL && job->cmdFILE != stdout) {
1483 		(void) fclose(job->cmdFILE);
1484 		job->cmdFILE = NULL;
1485 	    }
1486 	}
1487     }
1488 
1489 #ifdef RMT_NO_EXEC
1490 jobExecFinish:
1491 #endif
1492     /*
1493      * Now the job is actually running, add it to the table.
1494      */
1495     if (DEBUG(JOB)) {
1496 	printf("JobExec(%s): pid %d added to jobs table\n",
1497 		job->node->name, job->pid);
1498     }
1499     nJobs += 1;
1500     (void) Lst_AtEnd(jobs, (ClientData)job);
1501     JobSigUnlock(&mask);
1502 }
1503 
1504 /*-
1505  *-----------------------------------------------------------------------
1506  * JobMakeArgv --
1507  *	Create the argv needed to execute the shell for a given job.
1508  *
1509  *
1510  * Results:
1511  *
1512  * Side Effects:
1513  *
1514  *-----------------------------------------------------------------------
1515  */
1516 static void
1517 JobMakeArgv(Job *job, char **argv)
1518 {
1519     int	    	  argc;
1520     static char	  args[10]; 	/* For merged arguments */
1521 
1522     argv[0] = shellName;
1523     argc = 1;
1524 
1525     if ((commandShell->exit && (*commandShell->exit != '-')) ||
1526 	(commandShell->echo && (*commandShell->echo != '-')))
1527     {
1528 	/*
1529 	 * At least one of the flags doesn't have a minus before it, so
1530 	 * merge them together. Have to do this because the *(&(@*#*&#$#
1531 	 * Bourne shell thinks its second argument is a file to source.
1532 	 * Grrrr. Note the ten-character limitation on the combined arguments.
1533 	 */
1534 	(void)snprintf(args, sizeof(args), "-%s%s",
1535 		      ((job->flags & JOB_IGNERR) ? "" :
1536 		       (commandShell->exit ? commandShell->exit : "")),
1537 		      ((job->flags & JOB_SILENT) ? "" :
1538 		       (commandShell->echo ? commandShell->echo : "")));
1539 
1540 	if (args[1]) {
1541 	    argv[argc] = args;
1542 	    argc++;
1543 	}
1544     } else {
1545 	if (!(job->flags & JOB_IGNERR) && commandShell->exit) {
1546 	    argv[argc] = commandShell->exit;
1547 	    argc++;
1548 	}
1549 	if (!(job->flags & JOB_SILENT) && commandShell->echo) {
1550 	    argv[argc] = commandShell->echo;
1551 	    argc++;
1552 	}
1553     }
1554     argv[argc] = NULL;
1555 }
1556 
1557 /*-
1558  *-----------------------------------------------------------------------
1559  * JobRestart --
1560  *	Restart a job that stopped for some reason.
1561  *
1562  * Input:
1563  *	job		Job to restart
1564  *
1565  * Results:
1566  *	1 if max number of running jobs has been reached, 0 otherwise.
1567  *
1568  *-----------------------------------------------------------------------
1569  */
1570 static int
1571 JobRestart(Job *job)
1572 {
1573 #ifdef REMOTE
1574     int host;
1575 #endif
1576 
1577     if (job->flags & JOB_REMIGRATE) {
1578 	if (
1579 #ifdef REMOTE
1580 	    verboseRemigrates ||
1581 #endif
1582 	    DEBUG(JOB)) {
1583 	   (void) fprintf(stdout, "*** remigrating %x(%s)\n",
1584 			   job->pid, job->node->name);
1585 	   (void) fflush(stdout);
1586 	}
1587 
1588 #ifdef REMOTE
1589 	if (!Rmt_ReExport(job->pid, job->node, &host)) {
1590 	    if (verboseRemigrates || DEBUG(JOB)) {
1591 		(void) fprintf(stdout, "*** couldn't migrate...\n");
1592 		(void) fflush(stdout);
1593 	    }
1594 #endif
1595 	    if (nLocal != maxLocal) {
1596 		/*
1597 		 * Job cannot be remigrated, but there's room on the local
1598 		 * machine, so resume the job and note that another
1599 		 * local job has started.
1600 		 */
1601 		if (
1602 #ifdef REMOTE
1603 		    verboseRemigrates ||
1604 #endif
1605 		    DEBUG(JOB)) {
1606 		    (void) fprintf(stdout, "*** resuming on local machine\n");
1607 		    (void) fflush(stdout);
1608 		}
1609 		KILL(job->pid, SIGCONT);
1610 		nLocal +=1;
1611 #ifdef REMOTE
1612 		job->flags &= ~(JOB_REMIGRATE|JOB_RESUME|JOB_REMOTE);
1613 		job->flags |= JOB_CONTINUING;
1614 #else
1615 		job->flags &= ~(JOB_REMIGRATE|JOB_RESUME);
1616 #endif
1617 	    } else {
1618 		/*
1619 		 * Job cannot be restarted. Mark the table as full and
1620 		 * place the job back on the list of stopped jobs.
1621 		 */
1622 		if (
1623 #ifdef REMOTE
1624 		    verboseRemigrates ||
1625 #endif
1626 		    DEBUG(JOB)) {
1627 		   (void) fprintf(stdout, "*** holding\n");
1628 		   (void) fflush(stdout);
1629   		}
1630 		(void)Lst_AtFront(stoppedJobs, (ClientData)job);
1631 		return 1;
1632 	    }
1633 #ifdef REMOTE
1634 	} else {
1635 	    /*
1636 	     * Clear out the remigrate and resume flags. Set the continuing
1637 	     * flag so we know later on that the process isn't exiting just
1638 	     * because of a signal.
1639 	     */
1640 	    job->flags &= ~(JOB_REMIGRATE|JOB_RESUME);
1641 	    job->flags |= JOB_CONTINUING;
1642 	    job->rmtID = host;
1643 	}
1644 #endif
1645 
1646 	(void)Lst_AtEnd(jobs, (ClientData)job);
1647 	nJobs += 1;
1648     } else if (job->flags & JOB_RESTART) {
1649 	/*
1650 	 * Set up the control arguments to the shell. This is based on the
1651 	 * flags set earlier for this job. If the JOB_IGNERR flag is clear,
1652 	 * the 'exit' flag of the commandShell is used to cause it to exit
1653 	 * upon receiving an error. If the JOB_SILENT flag is clear, the
1654 	 * 'echo' flag of the commandShell is used to get it to start echoing
1655 	 * as soon as it starts processing commands.
1656 	 */
1657 	char	  *argv[10];
1658 
1659 	JobMakeArgv(job, argv);
1660 
1661 	if (DEBUG(JOB)) {
1662 	    (void) fprintf(stdout, "Restarting %s...", job->node->name);
1663 	    (void) fflush(stdout);
1664 	}
1665 #ifdef REMOTE
1666 	if ((job->node->type & OP_NOEXPORT) ||
1667  	    (nLocal < maxLocal && runLocalFirst)
1668 # ifdef RMT_NO_EXEC
1669 	    || !Rmt_Export(shellPath, argv, job)
1670 # else
1671 	    || !Rmt_Begin(shellPath, argv, job->node)
1672 # endif
1673 	   )
1674 #endif
1675 	{
1676 	    if (((nLocal >= maxLocal) && !(job->flags & JOB_SPECIAL))) {
1677 		/*
1678 		 * Can't be exported and not allowed to run locally -- put it
1679 		 * back on the hold queue and mark the table full
1680 		 */
1681 		if (DEBUG(JOB)) {
1682 		    (void) fprintf(stdout, "holding\n");
1683 		    (void) fflush(stdout);
1684 		}
1685 		(void)Lst_AtFront(stoppedJobs, (ClientData)job);
1686 		return 1;
1687 	    } else {
1688 		/*
1689 		 * Job may be run locally.
1690 		 */
1691 		if (DEBUG(JOB)) {
1692 		    (void) fprintf(stdout, "running locally\n");
1693 		    (void) fflush(stdout);
1694 		}
1695 		job->flags &= ~JOB_REMOTE;
1696 	    }
1697 	}
1698 #ifdef REMOTE
1699 	else {
1700 	    /*
1701 	     * Can be exported. Hooray!
1702 	     */
1703 	    if (DEBUG(JOB)) {
1704 		(void) fprintf(stdout, "exporting\n");
1705 		(void) fflush(stdout);
1706 	    }
1707 	    job->flags |= JOB_REMOTE;
1708 	}
1709 #endif
1710 	JobExec(job, argv);
1711     } else {
1712 	/*
1713 	 * The job has stopped and needs to be restarted. Why it stopped,
1714 	 * we don't know...
1715 	 */
1716 	if (DEBUG(JOB)) {
1717 	   (void) fprintf(stdout, "Resuming %s...", job->node->name);
1718 	   (void) fflush(stdout);
1719 	}
1720 	if ((nJobs != maxJobs) &&
1721 	    ((job->flags & JOB_REMOTE) ||
1722 	     (nLocal < maxLocal) ||
1723 	     ((maxLocal == 0) &&
1724 		((job->flags & JOB_SPECIAL)
1725 #ifdef REMOTE
1726 			&& (job->node->type & OP_NOEXPORT)
1727 #endif
1728 	    ))))
1729 	{
1730 	    /*
1731 	     * If the job is remote, it's ok to resume it as long as the
1732 	     * maximum concurrency won't be exceeded. If it's local and
1733 	     * we haven't reached the local concurrency limit already (or the
1734 	     * job must be run locally and maxLocal is 0), it's also ok to
1735 	     * resume it.
1736 	     */
1737 	    Boolean error;
1738 	    int status;
1739 
1740 #ifdef RMT_WANTS_SIGNALS
1741 	    if (job->flags & JOB_REMOTE) {
1742 		error = !Rmt_Signal(job, SIGCONT);
1743 	    } else
1744 #endif	/* RMT_WANTS_SIGNALS */
1745 		error = (KILL(job->pid, SIGCONT) != 0);
1746 
1747 	    if (!error) {
1748 		/*
1749 		 * Make sure the user knows we've continued the beast and
1750 		 * actually put the thing in the job table.
1751 		 */
1752 		job->flags |= JOB_CONTINUING;
1753 		status = W_STOPCODE(SIGCONT);
1754 		JobFinish(job, &status);
1755 
1756 		job->flags &= ~(JOB_RESUME|JOB_CONTINUING);
1757 		if (DEBUG(JOB)) {
1758 		   (void) fprintf(stdout, "done\n");
1759 		   (void) fflush(stdout);
1760 		}
1761 	    } else {
1762 		Error("couldn't resume %s: %s",
1763 		    job->node->name, strerror(errno));
1764 		status = W_EXITCODE(1, 0);
1765 		JobFinish(job, &status);
1766 	    }
1767 	} else {
1768 	    /*
1769 	     * Job cannot be restarted. Mark the table as full and
1770 	     * place the job back on the list of stopped jobs.
1771 	     */
1772 	    if (DEBUG(JOB)) {
1773 		(void) fprintf(stdout, "table full\n");
1774 		(void) fflush(stdout);
1775 	    }
1776 	    (void) Lst_AtFront(stoppedJobs, (ClientData)job);
1777 	    return 1;
1778 	}
1779     }
1780     return 0;
1781 }
1782 
1783 /*-
1784  *-----------------------------------------------------------------------
1785  * JobStart  --
1786  *	Start a target-creation process going for the target described
1787  *	by the graph node gn.
1788  *
1789  * Input:
1790  *	gn		target to create
1791  *	flags		flags for the job to override normal ones.
1792  *			e.g. JOB_SPECIAL or JOB_IGNDOTS
1793  *	previous	The previous Job structure for this node, if any.
1794  *
1795  * Results:
1796  *	JOB_ERROR if there was an error in the commands, JOB_FINISHED
1797  *	if there isn't actually anything left to do for the job and
1798  *	JOB_RUNNING if the job has been started.
1799  *
1800  * Side Effects:
1801  *	A new Job node is created and added to the list of running
1802  *	jobs. PMake is forked and a child shell created.
1803  *-----------------------------------------------------------------------
1804  */
1805 static int
1806 JobStart(GNode *gn, int flags, Job *previous)
1807 {
1808     Job		  *job;       /* new job descriptor */
1809     char	  *argv[10];  /* Argument vector to shell */
1810     Boolean	  cmdsOK;     /* true if the nodes commands were all right */
1811     Boolean 	  local;      /* Set true if the job was run locally */
1812     Boolean 	  noExec;     /* Set true if we decide not to run the job */
1813     int		  tfd;	      /* File descriptor to the temp file */
1814 
1815     if (previous != NULL) {
1816 	previous->flags &= ~(JOB_FIRST|JOB_IGNERR|JOB_SILENT|JOB_REMOTE);
1817 	job = previous;
1818     } else {
1819 	job = (Job *) emalloc(sizeof(Job));
1820 	if (job == NULL) {
1821 	    Punt("JobStart out of memory");
1822 	}
1823 	flags |= JOB_FIRST;
1824     }
1825 
1826     job->node = gn;
1827     job->tailCmds = NILLNODE;
1828 
1829     /*
1830      * Set the initial value of the flags for this job based on the global
1831      * ones and the node's attributes... Any flags supplied by the caller
1832      * are also added to the field.
1833      */
1834     job->flags = 0;
1835     if (Targ_Ignore(gn)) {
1836 	job->flags |= JOB_IGNERR;
1837     }
1838     if (Targ_Silent(gn)) {
1839 	job->flags |= JOB_SILENT;
1840     }
1841     job->flags |= flags;
1842 
1843     /*
1844      * Check the commands now so any attributes from .DEFAULT have a chance
1845      * to migrate to the node
1846      */
1847     if (!compatMake && job->flags & JOB_FIRST) {
1848 	cmdsOK = Job_CheckCommands(gn, Error);
1849     } else {
1850 	cmdsOK = TRUE;
1851     }
1852 
1853 #ifndef RMT_WILL_WATCH
1854     job->inPollfd = NULL;
1855 #endif
1856     /*
1857      * If the -n flag wasn't given, we open up OUR (not the child's)
1858      * temporary file to stuff commands in it. The thing is rd/wr so we don't
1859      * need to reopen it to feed it to the shell. If the -n flag *was* given,
1860      * we just set the file to be stdout. Cute, huh?
1861      */
1862     if (((gn->type & OP_MAKE) && !(noRecursiveExecute)) ||
1863 	(!noExecute && !touchFlag)) {
1864 	/*
1865 	 * tfile is the name of a file into which all shell commands are
1866 	 * put. It is used over by removing it before the child shell is
1867 	 * executed. The XXXXXX in the string are replaced by the pid of
1868 	 * the make process in a 6-character field with leading zeroes.
1869 	 */
1870 	char     tfile[sizeof(TMPPAT)];
1871 	sigset_t mask;
1872 	/*
1873 	 * We're serious here, but if the commands were bogus, we're
1874 	 * also dead...
1875 	 */
1876 	if (!cmdsOK) {
1877 	    DieHorribly();
1878 	}
1879 
1880 	JobSigLock(&mask);
1881 	(void)strcpy(tfile, TMPPAT);
1882 	if ((tfd = mkstemp(tfile)) == -1)
1883 	    Punt("Could not create temporary file %s", strerror(errno));
1884 	(void) eunlink(tfile);
1885 	JobSigUnlock(&mask);
1886 
1887 	job->cmdFILE = fdopen(tfd, "w+");
1888 	if (job->cmdFILE == NULL) {
1889 	    Punt("Could not fdopen %s", tfile);
1890 	}
1891 	(void) fcntl(FILENO(job->cmdFILE), F_SETFD, 1);
1892 	/*
1893 	 * Send the commands to the command file, flush all its buffers then
1894 	 * rewind and remove the thing.
1895 	 */
1896 	noExec = FALSE;
1897 
1898 	/*
1899 	 * used to be backwards; replace when start doing multiple commands
1900 	 * per shell.
1901 	 */
1902 	if (compatMake) {
1903 	    /*
1904 	     * Be compatible: If this is the first time for this node,
1905 	     * verify its commands are ok and open the commands list for
1906 	     * sequential access by later invocations of JobStart.
1907 	     * Once that is done, we take the next command off the list
1908 	     * and print it to the command file. If the command was an
1909 	     * ellipsis, note that there's nothing more to execute.
1910 	     */
1911 	    if ((job->flags&JOB_FIRST) && (Lst_Open(gn->commands) != SUCCESS)){
1912 		cmdsOK = FALSE;
1913 	    } else {
1914 		LstNode	ln = Lst_Next(gn->commands);
1915 
1916 		if ((ln == NILLNODE) ||
1917 		    JobPrintCommand((ClientData) Lst_Datum(ln),
1918 				    (ClientData) job))
1919 		{
1920 		    noExec = TRUE;
1921 		    Lst_Close(gn->commands);
1922 		}
1923 		if (noExec && !(job->flags & JOB_FIRST)) {
1924 		    /*
1925 		     * If we're not going to execute anything, the job
1926 		     * is done and we need to close down the various
1927 		     * file descriptors we've opened for output, then
1928 		     * call JobDoOutput to catch the final characters or
1929 		     * send the file to the screen... Note that the i/o streams
1930 		     * are only open if this isn't the first job.
1931 		     * Note also that this could not be done in
1932 		     * Job_CatchChildren b/c it wasn't clear if there were
1933 		     * more commands to execute or not...
1934 		     */
1935 		    JobClose(job);
1936 		}
1937 	    }
1938 	} else {
1939 	    /*
1940 	     * We can do all the commands at once. hooray for sanity
1941 	     */
1942 	    numCommands = 0;
1943 	    Lst_ForEach(gn->commands, JobPrintCommand, (ClientData)job);
1944 
1945 	    /*
1946 	     * If we didn't print out any commands to the shell script,
1947 	     * there's not much point in executing the shell, is there?
1948 	     */
1949 	    if (numCommands == 0) {
1950 		noExec = TRUE;
1951 	    }
1952 	}
1953     } else if (NoExecute(gn)) {
1954 	/*
1955 	 * Not executing anything -- just print all the commands to stdout
1956 	 * in one fell swoop. This will still set up job->tailCmds correctly.
1957 	 */
1958 	if (lastNode != gn) {
1959 	    MESSAGE(stdout, gn);
1960 	    lastNode = gn;
1961 	}
1962 	job->cmdFILE = stdout;
1963 	/*
1964 	 * Only print the commands if they're ok, but don't die if they're
1965 	 * not -- just let the user know they're bad and keep going. It
1966 	 * doesn't do any harm in this case and may do some good.
1967 	 */
1968 	if (cmdsOK) {
1969 	    Lst_ForEach(gn->commands, JobPrintCommand, (ClientData)job);
1970 	}
1971 	/*
1972 	 * Don't execute the shell, thank you.
1973 	 */
1974 	noExec = TRUE;
1975     } else {
1976 	/*
1977 	 * Just touch the target and note that no shell should be executed.
1978 	 * Set cmdFILE to stdout to make life easier. Check the commands, too,
1979 	 * but don't die if they're no good -- it does no harm to keep working
1980 	 * up the graph.
1981 	 */
1982 	job->cmdFILE = stdout;
1983     	Job_Touch(gn, job->flags&JOB_SILENT);
1984 	noExec = TRUE;
1985     }
1986 
1987     /*
1988      * If we're not supposed to execute a shell, don't.
1989      */
1990     if (noExec) {
1991 	/*
1992 	 * Unlink and close the command file if we opened one
1993 	 */
1994 	if (job->cmdFILE != stdout) {
1995 	    if (job->cmdFILE != NULL) {
1996 		(void) fclose(job->cmdFILE);
1997 		job->cmdFILE = NULL;
1998 	    }
1999 	} else {
2000 	     (void) fflush(stdout);
2001 	}
2002 
2003 	/*
2004 	 * We only want to work our way up the graph if we aren't here because
2005 	 * the commands for the job were no good.
2006 	 */
2007 	if (cmdsOK) {
2008 	    if (aborting == 0) {
2009 		if (job->tailCmds != NILLNODE) {
2010 		    Lst_ForEachFrom(job->node->commands, job->tailCmds,
2011 				    JobSaveCommand,
2012 				   (ClientData)job->node);
2013 		}
2014 		if (!(job->flags & JOB_SPECIAL))
2015 		    Job_TokenReturn();
2016 		job->node->made = MADE;
2017 		Make_Update(job->node);
2018 	    }
2019 	    free((Address)job);
2020 	    return(JOB_FINISHED);
2021 	} else {
2022 	    free((Address)job);
2023 	    return(JOB_ERROR);
2024 	}
2025     } else {
2026 	(void) fflush(job->cmdFILE);
2027     }
2028 
2029     /*
2030      * Set up the control arguments to the shell. This is based on the flags
2031      * set earlier for this job.
2032      */
2033     JobMakeArgv(job, argv);
2034 
2035     /*
2036      * If we're using pipes to catch output, create the pipe by which we'll
2037      * get the shell's output. If we're using files, print out that we're
2038      * starting a job and then set up its temporary-file name.
2039      */
2040     if (!compatMake || (job->flags & JOB_FIRST)) {
2041 	if (usePipes) {
2042 	    int fd[2];
2043 	    if (pipe(fd) == -1)
2044 		Punt("Cannot create pipe: %s", strerror(errno));
2045 	    job->inPipe = fd[0];
2046 	    job->outPipe = fd[1];
2047 	    (void) fcntl(job->inPipe, F_SETFD, 1);
2048 	    (void) fcntl(job->outPipe, F_SETFD, 1);
2049 	} else {
2050 	    (void) fprintf(stdout, "Remaking `%s'\n", gn->name);
2051   	    (void) fflush(stdout);
2052 	    (void) strcpy(job->outFile, TMPPAT);
2053 	    job->outFd = mkstemp(job->outFile);
2054 	    (void) fcntl(job->outFd, F_SETFD, 1);
2055 	}
2056     }
2057 
2058 #ifdef REMOTE
2059     if (!(gn->type & OP_NOEXPORT) && !(runLocalFirst && nLocal < maxLocal)) {
2060 #ifdef RMT_NO_EXEC
2061 	local = !Rmt_Export(shellPath, argv, job);
2062 #else
2063 	local = !Rmt_Begin(shellPath, argv, job->node);
2064 #endif /* RMT_NO_EXEC */
2065 	if (!local) {
2066 	    job->flags |= JOB_REMOTE;
2067 	}
2068     } else
2069 #endif
2070 	local = TRUE;
2071 
2072     if (local && (((nLocal >= maxLocal) &&
2073 	!(job->flags & JOB_SPECIAL) &&
2074 #ifdef REMOTE
2075 	(!(gn->type & OP_NOEXPORT) || (maxLocal != 0))
2076 #else
2077 	(maxLocal != 0)
2078 #endif
2079 	)))
2080     {
2081 	/*
2082 	 * The job can only be run locally, but we've hit the limit of
2083 	 * local concurrency, so put the job on hold until some other job
2084 	 * finishes. Note that the special jobs (.BEGIN, .INTERRUPT and .END)
2085 	 * may be run locally even when the local limit has been reached
2086 	 * (e.g. when maxLocal == 0), though they will be exported if at
2087 	 * all possible. In addition, any target marked with .NOEXPORT will
2088 	 * be run locally if maxLocal is 0.
2089 	 */
2090 	job->flags |= JOB_RESTART;
2091 	(void) Lst_AtEnd(stoppedJobs, (ClientData)job);
2092     } else {
2093 	JobExec(job, argv);
2094     }
2095     return(JOB_RUNNING);
2096 }
2097 
2098 static char *
2099 JobOutput(Job *job, char *cp, char *endp, int msg)
2100 {
2101     char *ecp;
2102 
2103     if (commandShell->noPrint) {
2104 	ecp = Str_FindSubstring(cp, commandShell->noPrint);
2105 	while (ecp != NULL) {
2106 	    if (cp != ecp) {
2107 		*ecp = '\0';
2108 		if (msg && job->node != lastNode) {
2109 		    MESSAGE(stdout, job->node);
2110 		    lastNode = job->node;
2111 		}
2112 		/*
2113 		 * The only way there wouldn't be a newline after
2114 		 * this line is if it were the last in the buffer.
2115 		 * however, since the non-printable comes after it,
2116 		 * there must be a newline, so we don't print one.
2117 		 */
2118 		(void) fprintf(stdout, "%s", cp);
2119 		(void) fflush(stdout);
2120 	    }
2121 	    cp = ecp + commandShell->noPLen;
2122 	    if (cp != endp) {
2123 		/*
2124 		 * Still more to print, look again after skipping
2125 		 * the whitespace following the non-printable
2126 		 * command....
2127 		 */
2128 		cp++;
2129 		while (*cp == ' ' || *cp == '\t' || *cp == '\n') {
2130 		    cp++;
2131 		}
2132 		ecp = Str_FindSubstring(cp, commandShell->noPrint);
2133 	    } else {
2134 		return cp;
2135 	    }
2136 	}
2137     }
2138     return cp;
2139 }
2140 
2141 /*-
2142  *-----------------------------------------------------------------------
2143  * JobDoOutput  --
2144  *	This function is called at different times depending on
2145  *	whether the user has specified that output is to be collected
2146  *	via pipes or temporary files. In the former case, we are called
2147  *	whenever there is something to read on the pipe. We collect more
2148  *	output from the given job and store it in the job's outBuf. If
2149  *	this makes up a line, we print it tagged by the job's identifier,
2150  *	as necessary.
2151  *	If output has been collected in a temporary file, we open the
2152  *	file and read it line by line, transfering it to our own
2153  *	output channel until the file is empty. At which point we
2154  *	remove the temporary file.
2155  *	In both cases, however, we keep our figurative eye out for the
2156  *	'noPrint' line for the shell from which the output came. If
2157  *	we recognize a line, we don't print it. If the command is not
2158  *	alone on the line (the character after it is not \0 or \n), we
2159  *	do print whatever follows it.
2160  *
2161  * Input:
2162  *	job		the job whose output needs printing
2163  *	finish		TRUE if this is the last time we'll be called
2164  *			for this job
2165  *
2166  * Results:
2167  *	None
2168  *
2169  * Side Effects:
2170  *	curPos may be shifted as may the contents of outBuf.
2171  *-----------------------------------------------------------------------
2172  */
2173 STATIC void
2174 JobDoOutput(Job *job, Boolean finish)
2175 {
2176     Boolean       gotNL = FALSE;  /* true if got a newline */
2177     Boolean       fbuf;  	  /* true if our buffer filled up */
2178     int		  nr;	      	  /* number of bytes read */
2179     int		  i;	      	  /* auxiliary index into outBuf */
2180     int		  max;	      	  /* limit for i (end of current data) */
2181     int		  nRead;      	  /* (Temporary) number of bytes read */
2182 
2183     FILE      	  *oFILE;	  /* Stream pointer to shell's output file */
2184     char          inLine[132];
2185 
2186 
2187     if (usePipes) {
2188 	/*
2189 	 * Read as many bytes as will fit in the buffer.
2190 	 */
2191 end_loop:
2192 	gotNL = FALSE;
2193 	fbuf = FALSE;
2194 
2195 	nRead = read(job->inPipe, &job->outBuf[job->curPos],
2196 			 JOB_BUFSIZE - job->curPos);
2197 	if (nRead < 0) {
2198 	    if (DEBUG(JOB)) {
2199 		perror("JobDoOutput(piperead)");
2200 	    }
2201 	    nr = 0;
2202 	} else {
2203 	    nr = nRead;
2204 	}
2205 
2206 	/*
2207 	 * If we hit the end-of-file (the job is dead), we must flush its
2208 	 * remaining output, so pretend we read a newline if there's any
2209 	 * output remaining in the buffer.
2210 	 * Also clear the 'finish' flag so we stop looping.
2211 	 */
2212 	if ((nr == 0) && (job->curPos != 0)) {
2213 	    job->outBuf[job->curPos] = '\n';
2214 	    nr = 1;
2215 	    finish = FALSE;
2216 	} else if (nr == 0) {
2217 	    finish = FALSE;
2218 	}
2219 
2220 	/*
2221 	 * Look for the last newline in the bytes we just got. If there is
2222 	 * one, break out of the loop with 'i' as its index and gotNL set
2223 	 * TRUE.
2224 	 */
2225 	max = job->curPos + nr;
2226 	for (i = job->curPos + nr - 1; i >= job->curPos; i--) {
2227 	    if (job->outBuf[i] == '\n') {
2228 		gotNL = TRUE;
2229 		break;
2230 	    } else if (job->outBuf[i] == '\0') {
2231 		/*
2232 		 * Why?
2233 		 */
2234 		job->outBuf[i] = ' ';
2235 	    }
2236 	}
2237 
2238 	if (!gotNL) {
2239 	    job->curPos += nr;
2240 	    if (job->curPos == JOB_BUFSIZE) {
2241 		/*
2242 		 * If we've run out of buffer space, we have no choice
2243 		 * but to print the stuff. sigh.
2244 		 */
2245 		fbuf = TRUE;
2246 		i = job->curPos;
2247 	    }
2248 	}
2249 	if (gotNL || fbuf) {
2250 	    /*
2251 	     * Need to send the output to the screen. Null terminate it
2252 	     * first, overwriting the newline character if there was one.
2253 	     * So long as the line isn't one we should filter (according
2254 	     * to the shell description), we print the line, preceded
2255 	     * by a target banner if this target isn't the same as the
2256 	     * one for which we last printed something.
2257 	     * The rest of the data in the buffer are then shifted down
2258 	     * to the start of the buffer and curPos is set accordingly.
2259 	     */
2260 	    job->outBuf[i] = '\0';
2261 	    if (i >= job->curPos) {
2262 		char *cp;
2263 
2264 		cp = JobOutput(job, job->outBuf, &job->outBuf[i], FALSE);
2265 
2266 		/*
2267 		 * There's still more in that thar buffer. This time, though,
2268 		 * we know there's no newline at the end, so we add one of
2269 		 * our own free will.
2270 		 */
2271 		if (*cp != '\0') {
2272 		    if (job->node != lastNode) {
2273 			MESSAGE(stdout, job->node);
2274 			lastNode = job->node;
2275 		    }
2276 		    (void) fprintf(stdout, "%s%s", cp, gotNL ? "\n" : "");
2277 		    (void) fflush(stdout);
2278 		}
2279 	    }
2280 	    if (i < max - 1) {
2281 		/* shift the remaining characters down */
2282 		(void) memcpy(job->outBuf, &job->outBuf[i + 1], max - (i + 1));
2283 		job->curPos = max - (i + 1);
2284 
2285 	    } else {
2286 		/*
2287 		 * We have written everything out, so we just start over
2288 		 * from the start of the buffer. No copying. No nothing.
2289 		 */
2290 		job->curPos = 0;
2291 	    }
2292 	}
2293 	if (finish) {
2294 	    /*
2295 	     * If the finish flag is true, we must loop until we hit
2296 	     * end-of-file on the pipe. This is guaranteed to happen
2297 	     * eventually since the other end of the pipe is now closed
2298 	     * (we closed it explicitly and the child has exited). When
2299 	     * we do get an EOF, finish will be set FALSE and we'll fall
2300 	     * through and out.
2301 	     */
2302 	    goto end_loop;
2303 	}
2304     } else {
2305 	/*
2306 	 * We've been called to retrieve the output of the job from the
2307 	 * temporary file where it's been squirreled away. This consists of
2308 	 * opening the file, reading the output line by line, being sure not
2309 	 * to print the noPrint line for the shell we used, then close and
2310 	 * remove the temporary file. Very simple.
2311 	 *
2312 	 * Change to read in blocks and do FindSubString type things as for
2313 	 * pipes? That would allow for "@echo -n..."
2314 	 */
2315 	oFILE = fopen(job->outFile, "r");
2316 	if (oFILE != NULL) {
2317 	    (void) fprintf(stdout, "Results of making %s:\n", job->node->name);
2318 	    (void) fflush(stdout);
2319 	    while (fgets(inLine, sizeof(inLine), oFILE) != NULL) {
2320 		char	*cp, *endp, *oendp;
2321 
2322 		cp = inLine;
2323 		oendp = endp = inLine + strlen(inLine);
2324 		if (endp[-1] == '\n') {
2325 		    *--endp = '\0';
2326 		}
2327 		cp = JobOutput(job, inLine, endp, FALSE);
2328 
2329 		/*
2330 		 * There's still more in that thar buffer. This time, though,
2331 		 * we know there's no newline at the end, so we add one of
2332 		 * our own free will.
2333 		 */
2334 		(void) fprintf(stdout, "%s", cp);
2335 		(void) fflush(stdout);
2336 		if (endp != oendp) {
2337 		    (void) fprintf(stdout, "\n");
2338 		    (void) fflush(stdout);
2339 		}
2340 	    }
2341 	    (void) fclose(oFILE);
2342 	    (void) eunlink(job->outFile);
2343 	} else {
2344 	    Punt("Cannot open `%s'", job->outFile);
2345 	}
2346     }
2347 }
2348 
2349 /*-
2350  *-----------------------------------------------------------------------
2351  * Job_CatchChildren --
2352  *	Handle the exit of a child. Called from Make_Make.
2353  *
2354  * Input:
2355  *	block		TRUE if should block on the wait
2356  *
2357  * Results:
2358  *	none.
2359  *
2360  * Side Effects:
2361  *	The job descriptor is removed from the list of children.
2362  *
2363  * Notes:
2364  *	We do waits, blocking or not, according to the wisdom of our
2365  *	caller, until there are no more children to report. For each
2366  *	job, call JobFinish to finish things off. This will take care of
2367  *	putting jobs on the stoppedJobs queue.
2368  *
2369  *-----------------------------------------------------------------------
2370  */
2371 void
2372 Job_CatchChildren(Boolean block)
2373 {
2374     int    	  pid;	    	/* pid of dead child */
2375     Job		  *job;	    	/* job descriptor for dead child */
2376     LstNode       jnode;    	/* list element for finding job */
2377     int	  	  status;   	/* Exit/termination status */
2378 
2379     /*
2380      * Don't even bother if we know there's no one around.
2381      */
2382     if (nLocal == 0) {
2383 	return;
2384     }
2385 
2386     while ((pid = waitpid((pid_t) -1, &status,
2387 			  (block?0:WNOHANG)|WUNTRACED)) > 0)
2388     {
2389 	if (DEBUG(JOB)) {
2390 	    (void) fprintf(stdout, "Process %d exited or stopped %x.\n", pid,
2391 	      status);
2392 	    (void) fflush(stdout);
2393 	}
2394 
2395 	jnode = Lst_Find(jobs, (ClientData)&pid, JobCmpPid);
2396 	if (jnode == NILLNODE) {
2397 	    if (WIFSTOPPED(status) && (WSTOPSIG(status) == SIGCONT)) {
2398 		jnode = Lst_Find(stoppedJobs, (ClientData) &pid, JobCmpPid);
2399 		if (jnode == NILLNODE) {
2400 		    Error("Resumed child (%d) not in table", pid);
2401 		    continue;
2402 		}
2403 		job = (Job *)Lst_Datum(jnode);
2404 		(void) Lst_Remove(stoppedJobs, jnode);
2405 	    } else {
2406 		Error("Child (%d) not in table?", pid);
2407 		continue;
2408 	    }
2409 	} else {
2410 	    job = (Job *) Lst_Datum(jnode);
2411 	    (void) Lst_Remove(jobs, jnode);
2412 	    nJobs -= 1;
2413 #ifdef REMOTE
2414 	    if (!(job->flags & JOB_REMOTE)) {
2415 		if (DEBUG(JOB)) {
2416 		    (void) fprintf(stdout,
2417 			   "Job queue has one fewer local process.\n");
2418 		    (void) fflush(stdout);
2419 		}
2420 		nLocal -= 1;
2421 	    }
2422 #else
2423 	    nLocal -= 1;
2424 #endif
2425 	}
2426 
2427 	JobFinish(job, &status);
2428     }
2429 }
2430 
2431 /*-
2432  *-----------------------------------------------------------------------
2433  * Job_CatchOutput --
2434  *	Catch the output from our children, if we're using
2435  *	pipes do so. Otherwise just block time until we get a
2436  *	signal (most likely a SIGCHLD) since there's no point in
2437  *	just spinning when there's nothing to do and the reaping
2438  *	of a child can wait for a while.
2439  *
2440  * Results:
2441  *	None
2442  *
2443  * Side Effects:
2444  *	Output is read from pipes if we're piping.
2445  * -----------------------------------------------------------------------
2446  */
2447 void
2448 Job_CatchOutput(void)
2449 {
2450     int           	  nready;
2451     LstNode		  ln;
2452     Job  	 	  *job;
2453 #ifdef RMT_WILL_WATCH
2454     int	    	  	  pnJobs;   	/* Previous nJobs */
2455 #endif
2456 
2457     (void) fflush(stdout);
2458     Job_TokenFlush();
2459 #ifdef RMT_WILL_WATCH
2460     pnJobs = nJobs;
2461 
2462     /*
2463      * It is possible for us to be called with nJobs equal to 0. This happens
2464      * if all the jobs finish and a job that is stopped cannot be run
2465      * locally (eg if maxLocal is 0) and cannot be exported. The job will
2466      * be placed back on the stoppedJobs queue, Job_Empty() will return false,
2467      * Make_Run will call us again when there's nothing for which to wait.
2468      * nJobs never changes, so we loop forever. Hence the check. It could
2469      * be argued that we should sleep for a bit so as not to swamp the
2470      * exportation system with requests. Perhaps we should.
2471      *
2472      * NOTE: IT IS THE RESPONSIBILITY OF Rmt_Wait TO CALL Job_CatchChildren
2473      * IN A TIMELY FASHION TO CATCH ANY LOCALLY RUNNING JOBS THAT EXIT.
2474      * It may use the variable nLocal to determine if it needs to call
2475      * Job_CatchChildren (if nLocal is 0, there's nothing for which to
2476      * wait...)
2477      */
2478     while (nJobs != 0 && pnJobs == nJobs) {
2479 	Rmt_Wait();
2480     }
2481 #else
2482     if (usePipes) {
2483 	if ((nready = poll((wantToken ? fds : (fds + 1)),
2484 	  		   (wantToken ? nfds : (nfds - 1)), POLL_MSEC)) <= 0) {
2485 	    return;
2486 	} else {
2487 	    sigset_t	mask;
2488 	    JobSigLock(&mask);
2489 	    if (Lst_Open(jobs) == FAILURE) {
2490 		Punt("Cannot open job table");
2491 	    }
2492 
2493 	    if (readyfd(&childExitJob)) {
2494 		char token;
2495 		(void) read(childExitJob.inPipe, &token, 1);
2496 		nready -= 1;
2497 	    }
2498 
2499 	    while (nready && (ln = Lst_Next(jobs)) != NILLNODE) {
2500 		job = (Job *) Lst_Datum(ln);
2501 		if (readyfd(job)) {
2502 		    JobDoOutput(job, FALSE);
2503 		    nready -= 1;
2504 		}
2505 	    }
2506 	    Lst_Close(jobs);
2507 	    JobSigUnlock(&mask);
2508 	}
2509     }
2510 #endif /* RMT_WILL_WATCH */
2511 }
2512 
2513 /*-
2514  *-----------------------------------------------------------------------
2515  * Job_Make --
2516  *	Start the creation of a target. Basically a front-end for
2517  *	JobStart used by the Make module.
2518  *
2519  * Results:
2520  *	None.
2521  *
2522  * Side Effects:
2523  *	Another job is started.
2524  *
2525  *-----------------------------------------------------------------------
2526  */
2527 void
2528 Job_Make(GNode *gn)
2529 {
2530     (void) JobStart(gn, 0, NULL);
2531 }
2532 
2533 /*-
2534  *-----------------------------------------------------------------------
2535  * Job_Init --
2536  *	Initialize the process module
2537  *
2538  * Input:
2539  *	maxproc		the greatest number of jobs which may be running
2540  *			at one time
2541  *	maxlocal	the greatest number of jobs which may be running
2542  *			at once
2543  *
2544  * Results:
2545  *	none
2546  *
2547  * Side Effects:
2548  *	lists and counters are initialized
2549  *-----------------------------------------------------------------------
2550  */
2551 void
2552 Job_Init(int maxproc, int maxlocal)
2553 {
2554     GNode         *begin;     /* node for commands to do at the very start */
2555 
2556     jobs =  	  Lst_Init(FALSE);
2557     stoppedJobs = Lst_Init(FALSE);
2558     maxJobs = 	  maxproc;
2559     maxLocal = 	  maxlocal;
2560     nJobs = 	  0;
2561     nLocal = 	  0;
2562     wantToken =	  FALSE;
2563 
2564     aborting = 	  0;
2565     errors = 	  0;
2566 
2567     lastNode =	  NILGNODE;
2568 
2569     if (maxJobs == 1
2570 #ifdef REMOTE
2571 	|| noMessages
2572 #endif
2573 		     ) {
2574 	/*
2575 	 * If only one job can run at a time, there's no need for a banner,
2576 	 * is there?
2577 	 */
2578 	targFmt = "";
2579     } else {
2580 	targFmt = TARG_FMT;
2581     }
2582 
2583     if (shellPath == NULL) {
2584 	/*
2585 	 * The user didn't specify a shell to use, so we are using the
2586 	 * default one... Both the absolute path and the last component
2587 	 * must be set. The last component is taken from the 'name' field
2588 	 * of the default shell description pointed-to by commandShell.
2589 	 * All default shells are located in _PATH_DEFSHELLDIR.
2590 	 */
2591 	shellName = commandShell->name;
2592 	shellPath = str_concat(_PATH_DEFSHELLDIR, shellName, STR_ADDSLASH);
2593     }
2594 
2595     if (commandShell->exit == NULL) {
2596 	commandShell->exit = "";
2597     }
2598     if (commandShell->echo == NULL) {
2599 	commandShell->echo = "";
2600     }
2601 
2602     if (pipe(exit_pipe) < 0)
2603 	Fatal("error in pipe: %s", strerror(errno));
2604     fcntl(exit_pipe[0], F_SETFD, 1);
2605     fcntl(exit_pipe[1], F_SETFD, 1);
2606 
2607     childExitJob.inPipe = exit_pipe[0];
2608 
2609     sigemptyset(&caught_signals);
2610     /*
2611      * Install a SIGCHLD handler.
2612      */
2613     (void)signal(SIGCHLD, JobChildSig);
2614     sigaddset(&caught_signals, SIGCHLD);
2615 
2616 #define ADDSIG(s,h)				\
2617     if (signal(s, SIG_IGN) != SIG_IGN) {	\
2618 	sigaddset(&caught_signals, s);		\
2619 	(void) signal(s, h);			\
2620     }
2621 
2622     /*
2623      * Catch the four signals that POSIX specifies if they aren't ignored.
2624      * JobPassSig will take care of calling JobInterrupt if appropriate.
2625      */
2626     ADDSIG(SIGINT, JobPassSig)
2627     ADDSIG(SIGHUP, JobPassSig)
2628     ADDSIG(SIGTERM, JobPassSig)
2629     ADDSIG(SIGQUIT, JobPassSig)
2630 
2631     /*
2632      * There are additional signals that need to be caught and passed if
2633      * either the export system wants to be told directly of signals or if
2634      * we're giving each job its own process group (since then it won't get
2635      * signals from the terminal driver as we own the terminal)
2636      */
2637 #if defined(RMT_WANTS_SIGNALS) || defined(USE_PGRP)
2638     ADDSIG(SIGTSTP, JobPassSig)
2639     ADDSIG(SIGTTOU, JobPassSig)
2640     ADDSIG(SIGTTIN, JobPassSig)
2641     ADDSIG(SIGWINCH, JobPassSig)
2642     ADDSIG(SIGCONT, JobContinueSig)
2643 #endif
2644 #undef ADDSIG
2645 
2646     begin = Targ_FindNode(".BEGIN", TARG_NOCREATE);
2647 
2648     if (begin != NILGNODE) {
2649 	JobStart(begin, JOB_SPECIAL, (Job *)0);
2650 	while (nJobs) {
2651 	    Job_CatchOutput();
2652 #ifndef RMT_WILL_WATCH
2653 	    Job_CatchChildren(!usePipes);
2654 #endif /* RMT_WILL_WATCH */
2655 	}
2656     }
2657     postCommands = Targ_FindNode(".END", TARG_CREATE);
2658 }
2659 
2660 static void JobSigReset(void)
2661 {
2662 #define DELSIG(s)					\
2663     if (sigismember(&caught_signals, s)) {		\
2664 	(void) signal(SIGINT, SIG_DFL);			\
2665     }
2666 
2667     DELSIG(SIGINT)
2668     DELSIG(SIGHUP)
2669     DELSIG(SIGQUIT)
2670     DELSIG(SIGTERM)
2671 #if defined(RMT_WANTS_SIGNALS) || defined(USE_PGRP)
2672     DELSIG(SIGTSTP)
2673     DELSIG(SIGTTOU)
2674     DELSIG(SIGTTIN)
2675     DELSIG(SIGWINCH)
2676     DELSIG(SIGCONT)
2677 #endif
2678 #undef DELSIG
2679     (void)signal(SIGCHLD, SIG_DFL);
2680 }
2681 
2682 /*-
2683  *-----------------------------------------------------------------------
2684  * Job_Empty --
2685  *	See if the job table is empty.  Because the local concurrency may
2686  *	be set to 0, it is possible for the job table to become empty,
2687  *	while the list of stoppedJobs remains non-empty. In such a case,
2688  *	we want to restart as many jobs as we can.
2689  *
2690  * Results:
2691  *	TRUE if it is. FALSE if it ain't.
2692  *
2693  * Side Effects:
2694  *	None.
2695  *
2696  * -----------------------------------------------------------------------
2697  */
2698 Boolean
2699 Job_Empty(void)
2700 {
2701     if (nJobs == 0) {
2702 	if (!Lst_IsEmpty(stoppedJobs) && !aborting) {
2703 	    /*
2704 	     * The job table is obviously not full if it has no jobs in
2705 	     * it...Try and restart the stopped jobs.
2706 	     */
2707 	    JobRestartJobs();
2708 	    return(FALSE);
2709 	} else {
2710 	    return(TRUE);
2711 	}
2712     } else {
2713 	return(FALSE);
2714     }
2715 }
2716 
2717 /*-
2718  *-----------------------------------------------------------------------
2719  * JobMatchShell --
2720  *	Find a shell in 'shells' given its name.
2721  *
2722  * Results:
2723  *	A pointer to the Shell structure.
2724  *
2725  * Side Effects:
2726  *	None.
2727  *
2728  *-----------------------------------------------------------------------
2729  */
2730 static Shell *
2731 JobMatchShell(char *name)
2732 {
2733     Shell	*sh;
2734 
2735     for (sh = shells; sh->name != NULL; sh++) {
2736 	if (strcmp(name, sh->name) == 0)
2737 		return (sh);
2738     }
2739     return (NULL);
2740 }
2741 
2742 /*-
2743  *-----------------------------------------------------------------------
2744  * Job_ParseShell --
2745  *	Parse a shell specification and set up commandShell, shellPath
2746  *	and shellName appropriately.
2747  *
2748  * Input:
2749  *	line		The shell spec
2750  *
2751  * Results:
2752  *	FAILURE if the specification was incorrect.
2753  *
2754  * Side Effects:
2755  *	commandShell points to a Shell structure (either predefined or
2756  *	created from the shell spec), shellPath is the full path of the
2757  *	shell described by commandShell, while shellName is just the
2758  *	final component of shellPath.
2759  *
2760  * Notes:
2761  *	A shell specification consists of a .SHELL target, with dependency
2762  *	operator, followed by a series of blank-separated words. Double
2763  *	quotes can be used to use blanks in words. A backslash escapes
2764  *	anything (most notably a double-quote and a space) and
2765  *	provides the functionality it does in C. Each word consists of
2766  *	keyword and value separated by an equal sign. There should be no
2767  *	unnecessary spaces in the word. The keywords are as follows:
2768  *	    name  	    Name of shell.
2769  *	    path  	    Location of shell.
2770  *	    quiet 	    Command to turn off echoing.
2771  *	    echo  	    Command to turn echoing on
2772  *	    filter	    Result of turning off echoing that shouldn't be
2773  *	    	  	    printed.
2774  *	    echoFlag	    Flag to turn echoing on at the start
2775  *	    errFlag	    Flag to turn error checking on at the start
2776  *	    hasErrCtl	    True if shell has error checking control
2777  *	    check 	    Command to turn on error checking if hasErrCtl
2778  *	    	  	    is TRUE or template of command to echo a command
2779  *	    	  	    for which error checking is off if hasErrCtl is
2780  *	    	  	    FALSE.
2781  *	    ignore	    Command to turn off error checking if hasErrCtl
2782  *	    	  	    is TRUE or template of command to execute a
2783  *	    	  	    command so as to ignore any errors it returns if
2784  *	    	  	    hasErrCtl is FALSE.
2785  *
2786  *-----------------------------------------------------------------------
2787  */
2788 ReturnStatus
2789 Job_ParseShell(char *line)
2790 {
2791     char	**words;
2792     char	**argv;
2793     int		argc;
2794     char	*path;
2795     Shell	newShell;
2796     Boolean	fullSpec = FALSE;
2797     Shell	*sh;
2798 
2799     while (isspace((unsigned char)*line)) {
2800 	line++;
2801     }
2802 
2803     if (shellArgv)
2804 	free(shellArgv);
2805 
2806     memset((Address)&newShell, 0, sizeof(newShell));
2807 
2808     /*
2809      * Parse the specification by keyword
2810      */
2811     words = brk_string(line, &argc, TRUE, &shellArgv);
2812 
2813     for (path = NULL, argv = words; argc != 0; argc--, argv++) {
2814 	    if (strncmp(*argv, "path=", 5) == 0) {
2815 		path = &argv[0][5];
2816 	    } else if (strncmp(*argv, "name=", 5) == 0) {
2817 		newShell.name = &argv[0][5];
2818 	    } else {
2819 		if (strncmp(*argv, "quiet=", 6) == 0) {
2820 		    newShell.echoOff = &argv[0][6];
2821 		} else if (strncmp(*argv, "echo=", 5) == 0) {
2822 		    newShell.echoOn = &argv[0][5];
2823 		} else if (strncmp(*argv, "filter=", 7) == 0) {
2824 		    newShell.noPrint = &argv[0][7];
2825 		    newShell.noPLen = strlen(newShell.noPrint);
2826 		} else if (strncmp(*argv, "echoFlag=", 9) == 0) {
2827 		    newShell.echo = &argv[0][9];
2828 		} else if (strncmp(*argv, "errFlag=", 8) == 0) {
2829 		    newShell.exit = &argv[0][8];
2830 		} else if (strncmp(*argv, "hasErrCtl=", 10) == 0) {
2831 		    char c = argv[0][10];
2832 		    newShell.hasErrCtl = !((c != 'Y') && (c != 'y') &&
2833 					   (c != 'T') && (c != 't'));
2834 		} else if (strncmp(*argv, "check=", 6) == 0) {
2835 		    newShell.errCheck = &argv[0][6];
2836 		} else if (strncmp(*argv, "ignore=", 7) == 0) {
2837 		    newShell.ignErr = &argv[0][7];
2838 		} else {
2839 		    Parse_Error(PARSE_FATAL, "Unknown keyword \"%s\"",
2840 				*argv);
2841 		    free(words);
2842 		    return(FAILURE);
2843 		}
2844 		fullSpec = TRUE;
2845 	    }
2846     }
2847 
2848     if (path == NULL) {
2849 	/*
2850 	 * If no path was given, the user wants one of the pre-defined shells,
2851 	 * yes? So we find the one s/he wants with the help of JobMatchShell
2852 	 * and set things up the right way. shellPath will be set up by
2853 	 * Job_Init.
2854 	 */
2855 	if (newShell.name == NULL) {
2856 	    Parse_Error(PARSE_FATAL, "Neither path nor name specified");
2857 	    free(words);
2858 	    return(FAILURE);
2859 	} else {
2860 	    if ((sh = JobMatchShell(newShell.name)) == NULL) {
2861 		    Parse_Error(PARSE_WARNING, "%s: No matching shell",
2862 				newShell.name);
2863 		    free(words);
2864 		    return(FAILURE);
2865 	    }
2866 	    commandShell = sh;
2867 	    shellName = newShell.name;
2868 	}
2869     } else {
2870 	/*
2871 	 * The user provided a path. If s/he gave nothing else (fullSpec is
2872 	 * FALSE), try and find a matching shell in the ones we know of.
2873 	 * Else we just take the specification at its word and copy it
2874 	 * to a new location. In either case, we need to record the
2875 	 * path the user gave for the shell.
2876 	 */
2877 	shellPath = path;
2878 	path = strrchr(path, '/');
2879 	if (path == NULL) {
2880 	    path = shellPath;
2881 	} else {
2882 	    path += 1;
2883 	}
2884 	if (newShell.name != NULL) {
2885 	    shellName = newShell.name;
2886 	} else {
2887 	    shellName = path;
2888 	}
2889 	if (!fullSpec) {
2890 	    if ((sh = JobMatchShell(shellName)) == NULL) {
2891 		    Parse_Error(PARSE_WARNING, "%s: No matching shell",
2892 				shellName);
2893 		    free(words);
2894 		    return(FAILURE);
2895 	    }
2896 	    commandShell = sh;
2897 	} else {
2898 	    commandShell = (Shell *) emalloc(sizeof(Shell));
2899 	    *commandShell = newShell;
2900 	}
2901     }
2902 
2903     if (commandShell->echoOn && commandShell->echoOff) {
2904 	commandShell->hasEchoCtl = TRUE;
2905     }
2906 
2907     if (!commandShell->hasErrCtl) {
2908 	if (commandShell->errCheck == NULL) {
2909 	    commandShell->errCheck = "";
2910 	}
2911 	if (commandShell->ignErr == NULL) {
2912 	    commandShell->ignErr = "%s\n";
2913 	}
2914     }
2915 
2916     /*
2917      * Do not free up the words themselves, since they might be in use by the
2918      * shell specification.
2919      */
2920     free(words);
2921     return SUCCESS;
2922 }
2923 
2924 /*-
2925  *-----------------------------------------------------------------------
2926  * JobInterrupt --
2927  *	Handle the receipt of an interrupt.
2928  *
2929  * Input:
2930  *	runINTERRUPT	Non-zero if commands for the .INTERRUPT target
2931  *			should be executed
2932  *	signo		signal received
2933  *
2934  * Results:
2935  *	None
2936  *
2937  * Side Effects:
2938  *	All children are killed. Another job will be started if the
2939  *	.INTERRUPT target was given.
2940  *-----------------------------------------------------------------------
2941  */
2942 static void
2943 JobInterrupt(int runINTERRUPT, int signo)
2944 {
2945     LstNode	ln;		/* element in job table */
2946     Job		*job;		/* job descriptor in that element */
2947     GNode	*interrupt;	/* the node describing the .INTERRUPT target */
2948     sigset_t	mask;
2949 
2950     aborting = ABORT_INTERRUPT;
2951 
2952     JobSigLock(&mask);
2953 
2954     (void) Lst_Open(jobs);
2955     while ((ln = Lst_Next(jobs)) != NILLNODE) {
2956 	GNode *gn;
2957 
2958 	job = (Job *) Lst_Datum(ln);
2959 	gn = job->node;
2960 
2961 	if ((gn->type & (OP_JOIN|OP_PHONY)) == 0 && !Targ_Precious(gn)) {
2962 	    char *file = (gn->path == NULL ? gn->name : gn->path);
2963 	    if (!noExecute && eunlink(file) != -1) {
2964 		Error("*** %s removed", file);
2965 	    }
2966 	}
2967 #ifdef RMT_WANTS_SIGNALS
2968 	if (job->flags & JOB_REMOTE) {
2969 	    /*
2970 	     * If job is remote, let the Rmt module do the killing.
2971 	     */
2972 	    if (!Rmt_Signal(job, signo)) {
2973 		/*
2974 		 * If couldn't kill the thing, finish it out now with an
2975 		 * error code, since no exit report will come in likely.
2976 		 */
2977 		int status;
2978 
2979 		status.w_status = 0;
2980 		status.w_retcode = 1;
2981 		JobFinish(job, &status);
2982 	    }
2983 	} else if (job->pid) {
2984 	    KILL(job->pid, signo);
2985 	}
2986 #else
2987 	if (job->pid) {
2988 	    if (DEBUG(JOB)) {
2989 		(void) fprintf(stdout,
2990 			   "JobInterrupt passing signal %d to child %d.\n",
2991 			   signo, job->pid);
2992 		(void) fflush(stdout);
2993 	    }
2994 	    KILL(job->pid, signo);
2995 	}
2996 #endif /* RMT_WANTS_SIGNALS */
2997     }
2998     Lst_Close(jobs);
2999 
3000 #ifdef REMOTE
3001    (void)Lst_Open(stoppedJobs);
3002     while ((ln = Lst_Next(stoppedJobs)) != NILLNODE) {
3003 	GNode *gn;
3004 
3005 	job = (Job *) Lst_Datum(ln);
3006 	gn = job->node;
3007 
3008 	if (job->flags & JOB_RESTART) {
3009 	    if (DEBUG(JOB)) {
3010 		(void) fprintf(stdout, "%s%s",
3011 			       "JobInterrupt skipping job on stopped queue",
3012 			       "-- it was waiting to be restarted.\n");
3013 		(void) fflush(stdout);
3014 	    }
3015 	    continue;
3016 	}
3017 	if ((gn->type & (OP_JOIN|OP_PHONY)) == 0 && !Targ_Precious(gn)) {
3018 	    char *file = (gn->path == NULL ? gn->name : gn->path);
3019 	    if (eunlink(file) == 0) {
3020 		Error("*** %s removed", file);
3021 	    }
3022 	}
3023 	/*
3024 	 * Resume the thing so it will take the signal.
3025 	 */
3026 	if (DEBUG(JOB)) {
3027 	    (void) fprintf(stdout,
3028 			   "JobInterrupt passing CONT to stopped child %d.\n",
3029 			   job->pid);
3030 	    (void) fflush(stdout);
3031 	}
3032 	KILL(job->pid, SIGCONT);
3033 #ifdef RMT_WANTS_SIGNALS
3034 	if (job->flags & JOB_REMOTE) {
3035 	    /*
3036 	     * If job is remote, let the Rmt module do the killing.
3037 	     */
3038 	    if (!Rmt_Signal(job, SIGINT)) {
3039 		/*
3040 		 * If couldn't kill the thing, finish it out now with an
3041 		 * error code, since no exit report will come in likely.
3042 		 */
3043 		int status;
3044 		status.w_status = 0;
3045 		status.w_retcode = 1;
3046 		JobFinish(job, &status);
3047 	    }
3048 	} else if (job->pid) {
3049 	    if (DEBUG(JOB)) {
3050 		(void) fprintf(stdout,
3051 		       "JobInterrupt passing interrupt to stopped child %d.\n",
3052 			       job->pid);
3053 		(void) fflush(stdout);
3054 	    }
3055 	    KILL(job->pid, SIGINT);
3056 	}
3057 #endif /* RMT_WANTS_SIGNALS */
3058     }
3059     Lst_Close(stoppedJobs);
3060 #endif /* REMOTE */
3061 
3062     JobSigUnlock(&mask);
3063 
3064     if (runINTERRUPT && !touchFlag) {
3065 	interrupt = Targ_FindNode(".INTERRUPT", TARG_NOCREATE);
3066 	if (interrupt != NILGNODE) {
3067 	    ignoreErrors = FALSE;
3068 
3069 	    JobStart(interrupt, JOB_IGNDOTS, (Job *)0);
3070 	    while (nJobs) {
3071 		Job_CatchOutput();
3072 #ifndef RMT_WILL_WATCH
3073 		Job_CatchChildren(!usePipes);
3074 #endif /* RMT_WILL_WATCH */
3075 	    }
3076 	}
3077     }
3078     Trace_Log(MAKEINTR, 0);
3079     exit(signo);
3080 }
3081 
3082 /*
3083  *-----------------------------------------------------------------------
3084  * Job_Finish --
3085  *	Do final processing such as the running of the commands
3086  *	attached to the .END target.
3087  *
3088  * Results:
3089  *	Number of errors reported.
3090  *
3091  * Side Effects:
3092  *	None.
3093  *-----------------------------------------------------------------------
3094  */
3095 int
3096 Job_Finish(void)
3097 {
3098     if (postCommands != NILGNODE && !Lst_IsEmpty(postCommands->commands)) {
3099 	if (errors) {
3100 	    Error("Errors reported so .END ignored");
3101 	} else {
3102 	    JobStart(postCommands, JOB_SPECIAL | JOB_IGNDOTS, NULL);
3103 
3104 	    while (nJobs) {
3105 		Job_CatchOutput();
3106 #ifndef RMT_WILL_WATCH
3107 		Job_CatchChildren(!usePipes);
3108 #endif /* RMT_WILL_WATCH */
3109 	    }
3110 	}
3111     }
3112     Job_TokenFlush();
3113     return(errors);
3114 }
3115 
3116 /*-
3117  *-----------------------------------------------------------------------
3118  * Job_End --
3119  *	Cleanup any memory used by the jobs module
3120  *
3121  * Results:
3122  *	None.
3123  *
3124  * Side Effects:
3125  *	Memory is freed
3126  *-----------------------------------------------------------------------
3127  */
3128 void
3129 Job_End(void)
3130 {
3131 #ifdef CLEANUP
3132     if (shellArgv)
3133 	free(shellArgv);
3134 #endif
3135 }
3136 
3137 /*-
3138  *-----------------------------------------------------------------------
3139  * Job_Wait --
3140  *	Waits for all running jobs to finish and returns. Sets 'aborting'
3141  *	to ABORT_WAIT to prevent other jobs from starting.
3142  *
3143  * Results:
3144  *	None.
3145  *
3146  * Side Effects:
3147  *	Currently running jobs finish.
3148  *
3149  *-----------------------------------------------------------------------
3150  */
3151 void
3152 Job_Wait(void)
3153 {
3154     aborting = ABORT_WAIT;
3155     while (nJobs != 0) {
3156 	Job_CatchOutput();
3157 #ifndef RMT_WILL_WATCH
3158 	Job_CatchChildren(!usePipes);
3159 #endif /* RMT_WILL_WATCH */
3160     }
3161     Job_TokenFlush();
3162     aborting = 0;
3163 }
3164 
3165 /*-
3166  *-----------------------------------------------------------------------
3167  * Job_AbortAll --
3168  *	Abort all currently running jobs without handling output or anything.
3169  *	This function is to be called only in the event of a major
3170  *	error. Most definitely NOT to be called from JobInterrupt.
3171  *
3172  * Results:
3173  *	None
3174  *
3175  * Side Effects:
3176  *	All children are killed, not just the firstborn
3177  *-----------------------------------------------------------------------
3178  */
3179 void
3180 Job_AbortAll(void)
3181 {
3182     LstNode	ln;	/* element in job table */
3183     Job		*job;	/* the job descriptor in that element */
3184     int		foo;
3185     sigset_t	mask;
3186 
3187     aborting = ABORT_ERROR;
3188 
3189     if (nJobs) {
3190 
3191 	JobSigLock(&mask);
3192 	(void) Lst_Open(jobs);
3193 	while ((ln = Lst_Next(jobs)) != NILLNODE) {
3194 	    job = (Job *) Lst_Datum(ln);
3195 
3196 	    /*
3197 	     * kill the child process with increasingly drastic signals to make
3198 	     * darn sure it's dead.
3199 	     */
3200 #ifdef RMT_WANTS_SIGNALS
3201 	    if (job->flags & JOB_REMOTE) {
3202 		Rmt_Signal(job, SIGINT);
3203 		Rmt_Signal(job, SIGKILL);
3204 	    } else {
3205 		KILL(job->pid, SIGINT);
3206 		KILL(job->pid, SIGKILL);
3207 	    }
3208 #else
3209 	    KILL(job->pid, SIGINT);
3210 	    KILL(job->pid, SIGKILL);
3211 #endif /* RMT_WANTS_SIGNALS */
3212 	}
3213 	Lst_Close(jobs);
3214 	JobSigUnlock(&mask);
3215     }
3216 
3217     /*
3218      * Catch as many children as want to report in at first, then give up
3219      */
3220     while (waitpid((pid_t) -1, &foo, WNOHANG) > 0)
3221 	continue;
3222 }
3223 
3224 #ifdef REMOTE
3225 /*-
3226  *-----------------------------------------------------------------------
3227  * JobFlagForMigration --
3228  *	Handle the eviction of a child. Called from RmtStatusChange.
3229  *	Flags the child as remigratable and then suspends it.
3230  *
3231  * Input:
3232  *	hostID		ID of host we used, for matching children
3233  *
3234  * Results:
3235  *	none.
3236  *
3237  * Side Effects:
3238  *	The job descriptor is flagged for remigration.
3239  *
3240  *-----------------------------------------------------------------------
3241  */
3242 void
3243 JobFlagForMigration(int hostID)
3244 {
3245     Job		  *job;	    	/* job descriptor for dead child */
3246     LstNode       jnode;    	/* list element for finding job */
3247 
3248     if (DEBUG(JOB)) {
3249 	(void) fprintf(stdout, "JobFlagForMigration(%d) called.\n", hostID);
3250 	(void) fflush(stdout);
3251     }
3252     jnode = Lst_Find(jobs, (ClientData)&hostID, JobCmpRmtID);
3253 
3254     if (jnode == NILLNODE) {
3255 	jnode = Lst_Find(stoppedJobs, (ClientData)hostID, JobCmpRmtID);
3256 		if (jnode == NILLNODE) {
3257 		    if (DEBUG(JOB)) {
3258 			Error("Evicting host(%d) not in table", hostID);
3259 		    }
3260 		    return;
3261 		}
3262     }
3263     job = (Job *) Lst_Datum(jnode);
3264 
3265     if (DEBUG(JOB)) {
3266 	(void) fprintf(stdout,
3267 		       "JobFlagForMigration(%d) found job '%s'.\n", hostID,
3268 		       job->node->name);
3269 	(void) fflush(stdout);
3270     }
3271 
3272     KILL(job->pid, SIGSTOP);
3273 
3274     job->flags |= JOB_REMIGRATE;
3275 }
3276 
3277 #endif
3278 
3279 /*-
3280  *-----------------------------------------------------------------------
3281  * JobRestartJobs --
3282  *	Tries to restart stopped jobs if there are slots available.
3283  *	Note that this tries to restart them regardless of pending errors.
3284  *	It's not good to leave stopped jobs lying around!
3285  *
3286  * Results:
3287  *	None.
3288  *
3289  * Side Effects:
3290  *	Resumes(and possibly migrates) jobs.
3291  *
3292  *-----------------------------------------------------------------------
3293  */
3294 static void
3295 JobRestartJobs(void)
3296 {
3297     sigset_t	mask;
3298 
3299     JobSigLock(&mask);
3300     while (!Lst_IsEmpty(stoppedJobs)) {
3301 	if (DEBUG(JOB)) {
3302 	    (void) fprintf(stdout, "Restarting a stopped job.\n");
3303 	    (void) fflush(stdout);
3304 	}
3305 	if (JobRestart((Job *)Lst_DeQueue(stoppedJobs)) != 0)
3306 		break;
3307     }
3308     JobSigUnlock(&mask);
3309 }
3310 
3311 #ifndef RMT_WILL_WATCH
3312 static void
3313 watchfd(Job *job)
3314 {
3315     int i;
3316     if (job->inPollfd != NULL)
3317 	Punt("Watching watched job");
3318     if (fds == NULL) {
3319 	maxfds = JBSTART;
3320 	fds = emalloc(sizeof(struct pollfd) * maxfds);
3321 	jobfds = emalloc(sizeof(Job **) * maxfds);
3322 
3323 	fds[0].fd = job_pipe[0];
3324 	fds[0].events = POLLIN;
3325 	jobfds[0] = &tokenWaitJob;
3326 	tokenWaitJob.inPollfd = &fds[0];
3327 	nfds++;
3328 
3329 	fds[1].fd = exit_pipe[0];
3330 	fds[1].events = POLLIN;
3331 	jobfds[1] = &childExitJob;
3332 	childExitJob.inPollfd = &fds[1];
3333 	nfds++;
3334     } else if (nfds == maxfds) {
3335 	maxfds *= JBFACTOR;
3336 	fds = erealloc(fds, sizeof(struct pollfd) * maxfds);
3337 	jobfds = erealloc(jobfds, sizeof(Job **) * maxfds);
3338 	for (i = 0; i < nfds; i++)
3339 	    jobfds[i]->inPollfd = &fds[i];
3340     }
3341 
3342     fds[nfds].fd = job->inPipe;
3343     fds[nfds].events = POLLIN;
3344     jobfds[nfds] = job;
3345     job->inPollfd = &fds[nfds];
3346     nfds++;
3347 }
3348 
3349 static void
3350 clearfd(Job *job)
3351 {
3352     int i;
3353     if (job->inPollfd == NULL)
3354 	Punt("Unwatching unwatched job");
3355     i = job->inPollfd - fds;
3356     nfds--;
3357     /*
3358      * Move last job in table into hole made by dead job.
3359      */
3360     if (nfds != i) {
3361 	fds[i] = fds[nfds];
3362 	jobfds[i] = jobfds[nfds];
3363 	jobfds[i]->inPollfd = &fds[i];
3364     }
3365     job->inPollfd = NULL;
3366 }
3367 
3368 static int
3369 readyfd(Job *job)
3370 {
3371     if (job->inPollfd == NULL)
3372 	Punt("Polling unwatched job");
3373     return (job->inPollfd->revents & POLLIN) != 0;
3374 }
3375 #endif
3376 
3377 /*-
3378  *-----------------------------------------------------------------------
3379  * JobTokenAdd --
3380  *	Put a token into the job pipe so that some make process can start
3381  *	another job.
3382  *
3383  * Side Effects:
3384  *	Allows more build jobs to be spawned somewhere.
3385  *
3386  *-----------------------------------------------------------------------
3387  */
3388 
3389 static void
3390 JobTokenAdd(void)
3391 {
3392 
3393     if (DEBUG(JOB))
3394 	printf("deposit token\n");
3395     write(job_pipe[1], "+", 1);
3396 }
3397 
3398 /*-
3399  *-----------------------------------------------------------------------
3400  * Job_ServerStartTokenAdd --
3401  *	Prep the job token pipe in the root make process.
3402  *
3403  *-----------------------------------------------------------------------
3404  */
3405 
3406 void
3407 Job_ServerStart(int maxproc)
3408 {
3409     int i, flags;
3410     char jobarg[64];
3411 
3412     if (pipe(job_pipe) < 0)
3413 	Fatal ("error in pipe: %s", strerror(errno));
3414 
3415     /*
3416      * We mark the input side of the pipe non-blocking; we poll(2) the
3417      * pipe when we're waiting for a job token, but we might lose the
3418      * race for the token when a new one becomes available, so the read
3419      * from the pipe should not block.
3420      */
3421     flags = fcntl(job_pipe[0], F_GETFL, 0);
3422     flags |= O_NONBLOCK;
3423     fcntl(job_pipe[0], F_SETFL, flags);
3424 
3425     /*
3426      * Mark job pipes as close-on-exec.
3427      * Note that we will clear this when executing submakes.
3428      */
3429     fcntl(job_pipe[0], F_SETFD, 1);
3430     fcntl(job_pipe[1], F_SETFD, 1);
3431 
3432     snprintf(jobarg, sizeof(jobarg), "%d,%d", job_pipe[0], job_pipe[1]);
3433 
3434     Var_Append(MAKEFLAGS, "-J", VAR_GLOBAL);
3435     Var_Append(MAKEFLAGS, jobarg, VAR_GLOBAL);
3436 
3437     /*
3438      * Preload job_pipe with one token per job, save the one
3439      * "extra" token for the primary job.
3440      *
3441      * XXX should clip maxJobs against PIPE_BUF -- if maxJobs is
3442      * larger than the write buffer size of the pipe, we will
3443      * deadlock here.
3444      */
3445     for (i=1; i < maxproc; i++)
3446 	JobTokenAdd();
3447 }
3448 
3449 /*
3450  * this tracks the number of tokens currently "out" to build jobs.
3451  */
3452 int jobTokensRunning = 0;
3453 int jobTokensFree = 0;
3454 /*-
3455  *-----------------------------------------------------------------------
3456  * Job_TokenReturn --
3457  *	Return a withdrawn token to the pool.
3458  *
3459  *-----------------------------------------------------------------------
3460  */
3461 
3462 void
3463 Job_TokenReturn(void)
3464 {
3465     jobTokensRunning--;
3466     if (jobTokensRunning < 0)
3467 	Punt("token botch");
3468     if (jobTokensRunning)
3469 	jobTokensFree++;
3470 }
3471 
3472 /*-
3473  *-----------------------------------------------------------------------
3474  * Job_TokenWithdraw --
3475  *	Attempt to withdraw a token from the pool.
3476  *
3477  * Results:
3478  *	Returns TRUE if a token was withdrawn, and FALSE if the pool
3479  *	is currently empty.
3480  *
3481  * Side Effects:
3482  * 	If pool is empty, set wantToken so that we wake up
3483  *	when a token is released.
3484  *
3485  *-----------------------------------------------------------------------
3486  */
3487 
3488 
3489 Boolean
3490 Job_TokenWithdraw(void)
3491 {
3492     char tok;
3493     int count;
3494 
3495     wantToken = FALSE;
3496 
3497     if (aborting)
3498 	    return FALSE;
3499 
3500     if (jobTokensRunning == 0) {
3501 	if (DEBUG(JOB))
3502 	    printf("first one's free\n");
3503 	jobTokensRunning++;
3504 	return TRUE;
3505     }
3506     if (jobTokensFree > 0) {
3507 	jobTokensFree--;
3508 	jobTokensRunning++;
3509 	return TRUE;
3510     }
3511     count = read(job_pipe[0], &tok, 1);
3512     if (count == 0)
3513 	Fatal("eof on job pipe!");
3514     else if (count < 0) {
3515 	if (errno != EAGAIN) {
3516 	    Fatal("job pipe read: %s", strerror(errno));
3517 	}
3518 	if (DEBUG(JOB))
3519 	    printf("blocked for token\n");
3520 	wantToken = TRUE;
3521 	return FALSE;
3522     }
3523     jobTokensRunning++;
3524     if (DEBUG(JOB))
3525 	printf("withdrew token\n");
3526     return TRUE;
3527 }
3528 
3529 /*-
3530  *-----------------------------------------------------------------------
3531  * Job_TokenFlush --
3532  *	Return free tokens to the pool.
3533  *
3534  *-----------------------------------------------------------------------
3535  */
3536 
3537 void
3538 Job_TokenFlush(void)
3539 {
3540     if (compatMake) return;
3541 
3542     while (jobTokensFree > 0) {
3543 	JobTokenAdd();
3544 	jobTokensFree--;
3545     }
3546 }
3547 
3548 #ifdef USE_SELECT
3549 int
3550 emul_poll(struct pollfd *fd, int nfd, int timeout)
3551 {
3552     fd_set rfds, wfds;
3553     int i, maxfd, nselect, npoll;
3554     struct timeval tv, *tvp;
3555     long usecs;
3556 
3557     FD_ZERO(&rfds);
3558     FD_ZERO(&wfds);
3559 
3560     maxfd = -1;
3561     for (i = 0; i < nfd; i++) {
3562 	fd[i].revents = 0;
3563 
3564 	if (fd[i].events & POLLIN)
3565 	    FD_SET(fd[i].fd, &rfds);
3566 
3567 	if (fd[i].events & POLLOUT)
3568 	    FD_SET(fd[i].fd, &wfds);
3569 
3570 	if (fd[i].fd > maxfd)
3571 	    maxfd = fd[i].fd;
3572     }
3573 
3574     if (maxfd >= FD_SETSIZE) {
3575 	Punt("Ran out of fd_set slots; "
3576 	     "recompile with a larger FD_SETSIZE.");
3577     }
3578 
3579     if (timeout < 0) {
3580 	tvp = NULL;
3581     } else {
3582 	usecs = timeout * 1000;
3583 	tv.tv_sec = usecs / 1000000;
3584 	tv.tv_usec = usecs % 1000000;
3585         tvp = &tv;
3586     }
3587 
3588     nselect = select(maxfd + 1, &rfds, &wfds, 0, tvp);
3589 
3590     if (nselect <= 0)
3591 	return nselect;
3592 
3593     npoll = 0;
3594     for (i = 0; i < nfd; i++) {
3595 	if (FD_ISSET(fd[i].fd, &rfds))
3596 	    fd[i].revents |= POLLIN;
3597 
3598 	if (FD_ISSET(fd[i].fd, &wfds))
3599 	    fd[i].revents |= POLLOUT;
3600 
3601 	if (fd[i].revents)
3602 	    npoll++;
3603     }
3604 
3605     return npoll;
3606 }
3607 #endif /* USE_SELECT */
3608