xref: /netbsd-src/usr.bin/make/job.c (revision d710132b4b8ce7f7cccaaf660cb16aa16b4077a0)
1 /*	$NetBSD: job.c,v 1.76 2003/04/08 17:46:59 christos 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 1.76 2003/04/08 17:46:59 christos 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 1.76 2003/04/08 17:46:59 christos 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 || Lst_IsEmpty(gn->children))) {
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 	sigemptyset(&mask);
1375 	JobSigUnlock(&mask);
1376 
1377 	/*
1378 	 * Must duplicate the input stream down to the child's input and
1379 	 * reset it to the beginning (again). Since the stream was marked
1380 	 * close-on-exec, we must clear that bit in the new input.
1381 	 */
1382 	if (dup2(FILENO(job->cmdFILE), 0) == -1) {
1383 	    execError("dup2", "job->cmdFILE");
1384 	    _exit(1);
1385 	}
1386 	(void) fcntl(0, F_SETFD, 0);
1387 	(void) lseek(0, (off_t)0, SEEK_SET);
1388 
1389 	if (job->node->type & OP_MAKE) {
1390 		/*
1391 		 * Pass job token pipe to submakes.
1392 		 */
1393 		fcntl(job_pipe[0], F_SETFD, 0);
1394 		fcntl(job_pipe[1], F_SETFD, 0);
1395 	}
1396 
1397 	if (usePipes) {
1398 	    /*
1399 	     * Set up the child's output to be routed through the pipe
1400 	     * we've created for it.
1401 	     */
1402 	    if (dup2(job->outPipe, 1) == -1) {
1403 		execError("dup2", "job->outPipe");
1404 		_exit(1);
1405 	    }
1406 	} else {
1407 	    /*
1408 	     * We're capturing output in a file, so we duplicate the
1409 	     * descriptor to the temporary file into the standard
1410 	     * output.
1411 	     */
1412 	    if (dup2(job->outFd, 1) == -1) {
1413 		execError("dup2", "job->outFd");
1414 		_exit(1);
1415 	    }
1416 	}
1417 	/*
1418 	 * The output channels are marked close on exec. This bit was
1419 	 * duplicated by the dup2 (on some systems), so we have to clear
1420 	 * it before routing the shell's error output to the same place as
1421 	 * its standard output.
1422 	 */
1423 	(void) fcntl(1, F_SETFD, 0);
1424 	if (dup2(1, 2) == -1) {
1425 	    execError("dup2", "1, 2");
1426 	    _exit(1);
1427 	}
1428 
1429 #ifdef USE_PGRP
1430 	/*
1431 	 * We want to switch the child into a different process family so
1432 	 * we can kill it and all its descendants in one fell swoop,
1433 	 * by killing its process family, but not commit suicide.
1434 	 */
1435 # if defined(SYSV)
1436 	(void) setsid();
1437 # else
1438 	(void) setpgid(0, getpid());
1439 # endif
1440 #endif /* USE_PGRP */
1441 
1442 #ifdef REMOTE
1443 	if (job->flags & JOB_REMOTE) {
1444 	    Rmt_Exec(shellPath, argv, FALSE);
1445 	} else
1446 #endif /* REMOTE */
1447 	{
1448 	   (void) execv(shellPath, argv);
1449 	   execError("exec", shellPath);
1450 	}
1451 	_exit(1);
1452     } else {
1453 	job->pid = cpid;
1454 
1455 	Trace_Log(JOBSTART, job);
1456 
1457 	if (usePipes && (job->flags & JOB_FIRST)) {
1458 	    /*
1459 	     * The first time a job is run for a node, we set the current
1460 	     * position in the buffer to the beginning and mark another
1461 	     * stream to watch in the outputs mask
1462 	     */
1463 	    job->curPos = 0;
1464 
1465 #ifdef RMT_WILL_WATCH
1466 	    Rmt_Watch(job->inPipe, JobLocalInput, job);
1467 #else
1468 	    watchfd(job);
1469 #endif /* RMT_WILL_WATCH */
1470 	}
1471 
1472 	if (job->flags & JOB_REMOTE) {
1473 #ifndef REMOTE
1474 	    job->rmtID = 0;
1475 #else
1476 	    job->rmtID = Rmt_LastID(job->pid);
1477 #endif /* REMOTE */
1478 	} else {
1479 	    nLocal += 1;
1480 	    /*
1481 	     * XXX: Used to not happen if REMOTE. Why?
1482 	     */
1483 	    if (job->cmdFILE != NULL && job->cmdFILE != stdout) {
1484 		(void) fclose(job->cmdFILE);
1485 		job->cmdFILE = NULL;
1486 	    }
1487 	}
1488     }
1489 
1490 #ifdef RMT_NO_EXEC
1491 jobExecFinish:
1492 #endif
1493     /*
1494      * Now the job is actually running, add it to the table.
1495      */
1496     if (DEBUG(JOB)) {
1497 	printf("JobExec(%s): pid %d added to jobs table\n",
1498 		job->node->name, job->pid);
1499     }
1500     nJobs += 1;
1501     (void) Lst_AtEnd(jobs, (ClientData)job);
1502     JobSigUnlock(&mask);
1503 }
1504 
1505 /*-
1506  *-----------------------------------------------------------------------
1507  * JobMakeArgv --
1508  *	Create the argv needed to execute the shell for a given job.
1509  *
1510  *
1511  * Results:
1512  *
1513  * Side Effects:
1514  *
1515  *-----------------------------------------------------------------------
1516  */
1517 static void
1518 JobMakeArgv(Job *job, char **argv)
1519 {
1520     int	    	  argc;
1521     static char	  args[10]; 	/* For merged arguments */
1522 
1523     argv[0] = shellName;
1524     argc = 1;
1525 
1526     if ((commandShell->exit && (*commandShell->exit != '-')) ||
1527 	(commandShell->echo && (*commandShell->echo != '-')))
1528     {
1529 	/*
1530 	 * At least one of the flags doesn't have a minus before it, so
1531 	 * merge them together. Have to do this because the *(&(@*#*&#$#
1532 	 * Bourne shell thinks its second argument is a file to source.
1533 	 * Grrrr. Note the ten-character limitation on the combined arguments.
1534 	 */
1535 	(void)snprintf(args, sizeof(args), "-%s%s",
1536 		      ((job->flags & JOB_IGNERR) ? "" :
1537 		       (commandShell->exit ? commandShell->exit : "")),
1538 		      ((job->flags & JOB_SILENT) ? "" :
1539 		       (commandShell->echo ? commandShell->echo : "")));
1540 
1541 	if (args[1]) {
1542 	    argv[argc] = args;
1543 	    argc++;
1544 	}
1545     } else {
1546 	if (!(job->flags & JOB_IGNERR) && commandShell->exit) {
1547 	    argv[argc] = commandShell->exit;
1548 	    argc++;
1549 	}
1550 	if (!(job->flags & JOB_SILENT) && commandShell->echo) {
1551 	    argv[argc] = commandShell->echo;
1552 	    argc++;
1553 	}
1554     }
1555     argv[argc] = NULL;
1556 }
1557 
1558 /*-
1559  *-----------------------------------------------------------------------
1560  * JobRestart --
1561  *	Restart a job that stopped for some reason.
1562  *
1563  * Input:
1564  *	job		Job to restart
1565  *
1566  * Results:
1567  *	1 if max number of running jobs has been reached, 0 otherwise.
1568  *
1569  *-----------------------------------------------------------------------
1570  */
1571 static int
1572 JobRestart(Job *job)
1573 {
1574 #ifdef REMOTE
1575     int host;
1576 #endif
1577 
1578     if (job->flags & JOB_REMIGRATE) {
1579 	if (
1580 #ifdef REMOTE
1581 	    verboseRemigrates ||
1582 #endif
1583 	    DEBUG(JOB)) {
1584 	   (void) fprintf(stdout, "*** remigrating %x(%s)\n",
1585 			   job->pid, job->node->name);
1586 	   (void) fflush(stdout);
1587 	}
1588 
1589 #ifdef REMOTE
1590 	if (!Rmt_ReExport(job->pid, job->node, &host)) {
1591 	    if (verboseRemigrates || DEBUG(JOB)) {
1592 		(void) fprintf(stdout, "*** couldn't migrate...\n");
1593 		(void) fflush(stdout);
1594 	    }
1595 #endif
1596 	    if (nLocal != maxLocal) {
1597 		/*
1598 		 * Job cannot be remigrated, but there's room on the local
1599 		 * machine, so resume the job and note that another
1600 		 * local job has started.
1601 		 */
1602 		if (
1603 #ifdef REMOTE
1604 		    verboseRemigrates ||
1605 #endif
1606 		    DEBUG(JOB)) {
1607 		    (void) fprintf(stdout, "*** resuming on local machine\n");
1608 		    (void) fflush(stdout);
1609 		}
1610 		KILL(job->pid, SIGCONT);
1611 		nLocal +=1;
1612 #ifdef REMOTE
1613 		job->flags &= ~(JOB_REMIGRATE|JOB_RESUME|JOB_REMOTE);
1614 		job->flags |= JOB_CONTINUING;
1615 #else
1616 		job->flags &= ~(JOB_REMIGRATE|JOB_RESUME);
1617 #endif
1618 	    } else {
1619 		/*
1620 		 * Job cannot be restarted. Mark the table as full and
1621 		 * place the job back on the list of stopped jobs.
1622 		 */
1623 		if (
1624 #ifdef REMOTE
1625 		    verboseRemigrates ||
1626 #endif
1627 		    DEBUG(JOB)) {
1628 		   (void) fprintf(stdout, "*** holding\n");
1629 		   (void) fflush(stdout);
1630   		}
1631 		(void)Lst_AtFront(stoppedJobs, (ClientData)job);
1632 		return 1;
1633 	    }
1634 #ifdef REMOTE
1635 	} else {
1636 	    /*
1637 	     * Clear out the remigrate and resume flags. Set the continuing
1638 	     * flag so we know later on that the process isn't exiting just
1639 	     * because of a signal.
1640 	     */
1641 	    job->flags &= ~(JOB_REMIGRATE|JOB_RESUME);
1642 	    job->flags |= JOB_CONTINUING;
1643 	    job->rmtID = host;
1644 	}
1645 #endif
1646 
1647 	(void)Lst_AtEnd(jobs, (ClientData)job);
1648 	nJobs += 1;
1649     } else if (job->flags & JOB_RESTART) {
1650 	/*
1651 	 * Set up the control arguments to the shell. This is based on the
1652 	 * flags set earlier for this job. If the JOB_IGNERR flag is clear,
1653 	 * the 'exit' flag of the commandShell is used to cause it to exit
1654 	 * upon receiving an error. If the JOB_SILENT flag is clear, the
1655 	 * 'echo' flag of the commandShell is used to get it to start echoing
1656 	 * as soon as it starts processing commands.
1657 	 */
1658 	char	  *argv[10];
1659 
1660 	JobMakeArgv(job, argv);
1661 
1662 	if (DEBUG(JOB)) {
1663 	    (void) fprintf(stdout, "Restarting %s...", job->node->name);
1664 	    (void) fflush(stdout);
1665 	}
1666 #ifdef REMOTE
1667 	if ((job->node->type & OP_NOEXPORT) ||
1668  	    (nLocal < maxLocal && runLocalFirst)
1669 # ifdef RMT_NO_EXEC
1670 	    || !Rmt_Export(shellPath, argv, job)
1671 # else
1672 	    || !Rmt_Begin(shellPath, argv, job->node)
1673 # endif
1674 	   )
1675 #endif
1676 	{
1677 	    if (((nLocal >= maxLocal) && !(job->flags & JOB_SPECIAL))) {
1678 		/*
1679 		 * Can't be exported and not allowed to run locally -- put it
1680 		 * back on the hold queue and mark the table full
1681 		 */
1682 		if (DEBUG(JOB)) {
1683 		    (void) fprintf(stdout, "holding\n");
1684 		    (void) fflush(stdout);
1685 		}
1686 		(void)Lst_AtFront(stoppedJobs, (ClientData)job);
1687 		return 1;
1688 	    } else {
1689 		/*
1690 		 * Job may be run locally.
1691 		 */
1692 		if (DEBUG(JOB)) {
1693 		    (void) fprintf(stdout, "running locally\n");
1694 		    (void) fflush(stdout);
1695 		}
1696 		job->flags &= ~JOB_REMOTE;
1697 	    }
1698 	}
1699 #ifdef REMOTE
1700 	else {
1701 	    /*
1702 	     * Can be exported. Hooray!
1703 	     */
1704 	    if (DEBUG(JOB)) {
1705 		(void) fprintf(stdout, "exporting\n");
1706 		(void) fflush(stdout);
1707 	    }
1708 	    job->flags |= JOB_REMOTE;
1709 	}
1710 #endif
1711 	JobExec(job, argv);
1712     } else {
1713 	/*
1714 	 * The job has stopped and needs to be restarted. Why it stopped,
1715 	 * we don't know...
1716 	 */
1717 	if (DEBUG(JOB)) {
1718 	   (void) fprintf(stdout, "Resuming %s...", job->node->name);
1719 	   (void) fflush(stdout);
1720 	}
1721 	if ((nJobs != maxJobs) &&
1722 	    ((job->flags & JOB_REMOTE) ||
1723 	     (nLocal < maxLocal) ||
1724 	     ((maxLocal == 0) &&
1725 		((job->flags & JOB_SPECIAL)
1726 #ifdef REMOTE
1727 			&& (job->node->type & OP_NOEXPORT)
1728 #endif
1729 	    ))))
1730 	{
1731 	    /*
1732 	     * If the job is remote, it's ok to resume it as long as the
1733 	     * maximum concurrency won't be exceeded. If it's local and
1734 	     * we haven't reached the local concurrency limit already (or the
1735 	     * job must be run locally and maxLocal is 0), it's also ok to
1736 	     * resume it.
1737 	     */
1738 	    Boolean error;
1739 	    int status;
1740 
1741 #ifdef RMT_WANTS_SIGNALS
1742 	    if (job->flags & JOB_REMOTE) {
1743 		error = !Rmt_Signal(job, SIGCONT);
1744 	    } else
1745 #endif	/* RMT_WANTS_SIGNALS */
1746 		error = (KILL(job->pid, SIGCONT) != 0);
1747 
1748 	    if (!error) {
1749 		/*
1750 		 * Make sure the user knows we've continued the beast and
1751 		 * actually put the thing in the job table.
1752 		 */
1753 		job->flags |= JOB_CONTINUING;
1754 		status = W_STOPCODE(SIGCONT);
1755 		JobFinish(job, &status);
1756 
1757 		job->flags &= ~(JOB_RESUME|JOB_CONTINUING);
1758 		if (DEBUG(JOB)) {
1759 		   (void) fprintf(stdout, "done\n");
1760 		   (void) fflush(stdout);
1761 		}
1762 	    } else {
1763 		Error("couldn't resume %s: %s",
1764 		    job->node->name, strerror(errno));
1765 		status = W_EXITCODE(1, 0);
1766 		JobFinish(job, &status);
1767 	    }
1768 	} else {
1769 	    /*
1770 	     * Job cannot be restarted. Mark the table as full and
1771 	     * place the job back on the list of stopped jobs.
1772 	     */
1773 	    if (DEBUG(JOB)) {
1774 		(void) fprintf(stdout, "table full\n");
1775 		(void) fflush(stdout);
1776 	    }
1777 	    (void) Lst_AtFront(stoppedJobs, (ClientData)job);
1778 	    return 1;
1779 	}
1780     }
1781     return 0;
1782 }
1783 
1784 /*-
1785  *-----------------------------------------------------------------------
1786  * JobStart  --
1787  *	Start a target-creation process going for the target described
1788  *	by the graph node gn.
1789  *
1790  * Input:
1791  *	gn		target to create
1792  *	flags		flags for the job to override normal ones.
1793  *			e.g. JOB_SPECIAL or JOB_IGNDOTS
1794  *	previous	The previous Job structure for this node, if any.
1795  *
1796  * Results:
1797  *	JOB_ERROR if there was an error in the commands, JOB_FINISHED
1798  *	if there isn't actually anything left to do for the job and
1799  *	JOB_RUNNING if the job has been started.
1800  *
1801  * Side Effects:
1802  *	A new Job node is created and added to the list of running
1803  *	jobs. PMake is forked and a child shell created.
1804  *-----------------------------------------------------------------------
1805  */
1806 static int
1807 JobStart(GNode *gn, int flags, Job *previous)
1808 {
1809     Job		  *job;       /* new job descriptor */
1810     char	  *argv[10];  /* Argument vector to shell */
1811     Boolean	  cmdsOK;     /* true if the nodes commands were all right */
1812     Boolean 	  local;      /* Set true if the job was run locally */
1813     Boolean 	  noExec;     /* Set true if we decide not to run the job */
1814     int		  tfd;	      /* File descriptor to the temp file */
1815 
1816     if (previous != NULL) {
1817 	previous->flags &= ~(JOB_FIRST|JOB_IGNERR|JOB_SILENT|JOB_REMOTE);
1818 	job = previous;
1819     } else {
1820 	job = (Job *) emalloc(sizeof(Job));
1821 	if (job == NULL) {
1822 	    Punt("JobStart out of memory");
1823 	}
1824 	flags |= JOB_FIRST;
1825     }
1826 
1827     job->node = gn;
1828     job->tailCmds = NILLNODE;
1829 
1830     /*
1831      * Set the initial value of the flags for this job based on the global
1832      * ones and the node's attributes... Any flags supplied by the caller
1833      * are also added to the field.
1834      */
1835     job->flags = 0;
1836     if (Targ_Ignore(gn)) {
1837 	job->flags |= JOB_IGNERR;
1838     }
1839     if (Targ_Silent(gn)) {
1840 	job->flags |= JOB_SILENT;
1841     }
1842     job->flags |= flags;
1843 
1844     /*
1845      * Check the commands now so any attributes from .DEFAULT have a chance
1846      * to migrate to the node
1847      */
1848     if (!compatMake && job->flags & JOB_FIRST) {
1849 	cmdsOK = Job_CheckCommands(gn, Error);
1850     } else {
1851 	cmdsOK = TRUE;
1852     }
1853 
1854 #ifndef RMT_WILL_WATCH
1855     job->inPollfd = NULL;
1856 #endif
1857     /*
1858      * If the -n flag wasn't given, we open up OUR (not the child's)
1859      * temporary file to stuff commands in it. The thing is rd/wr so we don't
1860      * need to reopen it to feed it to the shell. If the -n flag *was* given,
1861      * we just set the file to be stdout. Cute, huh?
1862      */
1863     if (((gn->type & OP_MAKE) && !(noRecursiveExecute)) ||
1864 	(!noExecute && !touchFlag)) {
1865 	/*
1866 	 * tfile is the name of a file into which all shell commands are
1867 	 * put. It is used over by removing it before the child shell is
1868 	 * executed. The XXXXXX in the string are replaced by the pid of
1869 	 * the make process in a 6-character field with leading zeroes.
1870 	 */
1871 	char     tfile[sizeof(TMPPAT)];
1872 	sigset_t mask;
1873 	/*
1874 	 * We're serious here, but if the commands were bogus, we're
1875 	 * also dead...
1876 	 */
1877 	if (!cmdsOK) {
1878 	    DieHorribly();
1879 	}
1880 
1881 	JobSigLock(&mask);
1882 	(void)strcpy(tfile, TMPPAT);
1883 	if ((tfd = mkstemp(tfile)) == -1)
1884 	    Punt("Could not create temporary file %s", strerror(errno));
1885 	(void) eunlink(tfile);
1886 	JobSigUnlock(&mask);
1887 
1888 	job->cmdFILE = fdopen(tfd, "w+");
1889 	if (job->cmdFILE == NULL) {
1890 	    Punt("Could not fdopen %s", tfile);
1891 	}
1892 	(void) fcntl(FILENO(job->cmdFILE), F_SETFD, 1);
1893 	/*
1894 	 * Send the commands to the command file, flush all its buffers then
1895 	 * rewind and remove the thing.
1896 	 */
1897 	noExec = FALSE;
1898 
1899 	/*
1900 	 * used to be backwards; replace when start doing multiple commands
1901 	 * per shell.
1902 	 */
1903 	if (compatMake) {
1904 	    /*
1905 	     * Be compatible: If this is the first time for this node,
1906 	     * verify its commands are ok and open the commands list for
1907 	     * sequential access by later invocations of JobStart.
1908 	     * Once that is done, we take the next command off the list
1909 	     * and print it to the command file. If the command was an
1910 	     * ellipsis, note that there's nothing more to execute.
1911 	     */
1912 	    if ((job->flags&JOB_FIRST) && (Lst_Open(gn->commands) != SUCCESS)){
1913 		cmdsOK = FALSE;
1914 	    } else {
1915 		LstNode	ln = Lst_Next(gn->commands);
1916 
1917 		if ((ln == NILLNODE) ||
1918 		    JobPrintCommand((ClientData) Lst_Datum(ln),
1919 				    (ClientData) job))
1920 		{
1921 		    noExec = TRUE;
1922 		    Lst_Close(gn->commands);
1923 		}
1924 		if (noExec && !(job->flags & JOB_FIRST)) {
1925 		    /*
1926 		     * If we're not going to execute anything, the job
1927 		     * is done and we need to close down the various
1928 		     * file descriptors we've opened for output, then
1929 		     * call JobDoOutput to catch the final characters or
1930 		     * send the file to the screen... Note that the i/o streams
1931 		     * are only open if this isn't the first job.
1932 		     * Note also that this could not be done in
1933 		     * Job_CatchChildren b/c it wasn't clear if there were
1934 		     * more commands to execute or not...
1935 		     */
1936 		    JobClose(job);
1937 		}
1938 	    }
1939 	} else {
1940 	    /*
1941 	     * We can do all the commands at once. hooray for sanity
1942 	     */
1943 	    numCommands = 0;
1944 	    Lst_ForEach(gn->commands, JobPrintCommand, (ClientData)job);
1945 
1946 	    /*
1947 	     * If we didn't print out any commands to the shell script,
1948 	     * there's not much point in executing the shell, is there?
1949 	     */
1950 	    if (numCommands == 0) {
1951 		noExec = TRUE;
1952 	    }
1953 	}
1954     } else if (NoExecute(gn)) {
1955 	/*
1956 	 * Not executing anything -- just print all the commands to stdout
1957 	 * in one fell swoop. This will still set up job->tailCmds correctly.
1958 	 */
1959 	if (lastNode != gn) {
1960 	    MESSAGE(stdout, gn);
1961 	    lastNode = gn;
1962 	}
1963 	job->cmdFILE = stdout;
1964 	/*
1965 	 * Only print the commands if they're ok, but don't die if they're
1966 	 * not -- just let the user know they're bad and keep going. It
1967 	 * doesn't do any harm in this case and may do some good.
1968 	 */
1969 	if (cmdsOK) {
1970 	    Lst_ForEach(gn->commands, JobPrintCommand, (ClientData)job);
1971 	}
1972 	/*
1973 	 * Don't execute the shell, thank you.
1974 	 */
1975 	noExec = TRUE;
1976     } else {
1977 	/*
1978 	 * Just touch the target and note that no shell should be executed.
1979 	 * Set cmdFILE to stdout to make life easier. Check the commands, too,
1980 	 * but don't die if they're no good -- it does no harm to keep working
1981 	 * up the graph.
1982 	 */
1983 	job->cmdFILE = stdout;
1984     	Job_Touch(gn, job->flags&JOB_SILENT);
1985 	noExec = TRUE;
1986     }
1987 
1988     /*
1989      * If we're not supposed to execute a shell, don't.
1990      */
1991     if (noExec) {
1992 	/*
1993 	 * Unlink and close the command file if we opened one
1994 	 */
1995 	if (job->cmdFILE != stdout) {
1996 	    if (job->cmdFILE != NULL) {
1997 		(void) fclose(job->cmdFILE);
1998 		job->cmdFILE = NULL;
1999 	    }
2000 	} else {
2001 	     (void) fflush(stdout);
2002 	}
2003 
2004 	/*
2005 	 * We only want to work our way up the graph if we aren't here because
2006 	 * the commands for the job were no good.
2007 	 */
2008 	if (cmdsOK) {
2009 	    if (aborting == 0) {
2010 		if (job->tailCmds != NILLNODE) {
2011 		    Lst_ForEachFrom(job->node->commands, job->tailCmds,
2012 				    JobSaveCommand,
2013 				   (ClientData)job->node);
2014 		}
2015 		if (!(job->flags & JOB_SPECIAL))
2016 		    Job_TokenReturn();
2017 		job->node->made = MADE;
2018 		Make_Update(job->node);
2019 	    }
2020 	    free((Address)job);
2021 	    return(JOB_FINISHED);
2022 	} else {
2023 	    free((Address)job);
2024 	    return(JOB_ERROR);
2025 	}
2026     } else {
2027 	(void) fflush(job->cmdFILE);
2028     }
2029 
2030     /*
2031      * Set up the control arguments to the shell. This is based on the flags
2032      * set earlier for this job.
2033      */
2034     JobMakeArgv(job, argv);
2035 
2036     /*
2037      * If we're using pipes to catch output, create the pipe by which we'll
2038      * get the shell's output. If we're using files, print out that we're
2039      * starting a job and then set up its temporary-file name.
2040      */
2041     if (!compatMake || (job->flags & JOB_FIRST)) {
2042 	if (usePipes) {
2043 	    int fd[2];
2044 	    if (pipe(fd) == -1)
2045 		Punt("Cannot create pipe: %s", strerror(errno));
2046 	    job->inPipe = fd[0];
2047 	    job->outPipe = fd[1];
2048 	    (void) fcntl(job->inPipe, F_SETFD, 1);
2049 	    (void) fcntl(job->outPipe, F_SETFD, 1);
2050 	} else {
2051 	    (void) fprintf(stdout, "Remaking `%s'\n", gn->name);
2052   	    (void) fflush(stdout);
2053 	    (void) strcpy(job->outFile, TMPPAT);
2054 	    job->outFd = mkstemp(job->outFile);
2055 	    (void) fcntl(job->outFd, F_SETFD, 1);
2056 	}
2057     }
2058 
2059 #ifdef REMOTE
2060     if (!(gn->type & OP_NOEXPORT) && !(runLocalFirst && nLocal < maxLocal)) {
2061 #ifdef RMT_NO_EXEC
2062 	local = !Rmt_Export(shellPath, argv, job);
2063 #else
2064 	local = !Rmt_Begin(shellPath, argv, job->node);
2065 #endif /* RMT_NO_EXEC */
2066 	if (!local) {
2067 	    job->flags |= JOB_REMOTE;
2068 	}
2069     } else
2070 #endif
2071 	local = TRUE;
2072 
2073     if (local && (((nLocal >= maxLocal) &&
2074 	!(job->flags & JOB_SPECIAL) &&
2075 #ifdef REMOTE
2076 	(!(gn->type & OP_NOEXPORT) || (maxLocal != 0))
2077 #else
2078 	(maxLocal != 0)
2079 #endif
2080 	)))
2081     {
2082 	/*
2083 	 * The job can only be run locally, but we've hit the limit of
2084 	 * local concurrency, so put the job on hold until some other job
2085 	 * finishes. Note that the special jobs (.BEGIN, .INTERRUPT and .END)
2086 	 * may be run locally even when the local limit has been reached
2087 	 * (e.g. when maxLocal == 0), though they will be exported if at
2088 	 * all possible. In addition, any target marked with .NOEXPORT will
2089 	 * be run locally if maxLocal is 0.
2090 	 */
2091 	job->flags |= JOB_RESTART;
2092 	(void) Lst_AtEnd(stoppedJobs, (ClientData)job);
2093     } else {
2094 	JobExec(job, argv);
2095     }
2096     return(JOB_RUNNING);
2097 }
2098 
2099 static char *
2100 JobOutput(Job *job, char *cp, char *endp, int msg)
2101 {
2102     char *ecp;
2103 
2104     if (commandShell->noPrint) {
2105 	ecp = Str_FindSubstring(cp, commandShell->noPrint);
2106 	while (ecp != NULL) {
2107 	    if (cp != ecp) {
2108 		*ecp = '\0';
2109 		if (msg && job->node != lastNode) {
2110 		    MESSAGE(stdout, job->node);
2111 		    lastNode = job->node;
2112 		}
2113 		/*
2114 		 * The only way there wouldn't be a newline after
2115 		 * this line is if it were the last in the buffer.
2116 		 * however, since the non-printable comes after it,
2117 		 * there must be a newline, so we don't print one.
2118 		 */
2119 		(void) fprintf(stdout, "%s", cp);
2120 		(void) fflush(stdout);
2121 	    }
2122 	    cp = ecp + commandShell->noPLen;
2123 	    if (cp != endp) {
2124 		/*
2125 		 * Still more to print, look again after skipping
2126 		 * the whitespace following the non-printable
2127 		 * command....
2128 		 */
2129 		cp++;
2130 		while (*cp == ' ' || *cp == '\t' || *cp == '\n') {
2131 		    cp++;
2132 		}
2133 		ecp = Str_FindSubstring(cp, commandShell->noPrint);
2134 	    } else {
2135 		return cp;
2136 	    }
2137 	}
2138     }
2139     return cp;
2140 }
2141 
2142 /*-
2143  *-----------------------------------------------------------------------
2144  * JobDoOutput  --
2145  *	This function is called at different times depending on
2146  *	whether the user has specified that output is to be collected
2147  *	via pipes or temporary files. In the former case, we are called
2148  *	whenever there is something to read on the pipe. We collect more
2149  *	output from the given job and store it in the job's outBuf. If
2150  *	this makes up a line, we print it tagged by the job's identifier,
2151  *	as necessary.
2152  *	If output has been collected in a temporary file, we open the
2153  *	file and read it line by line, transfering it to our own
2154  *	output channel until the file is empty. At which point we
2155  *	remove the temporary file.
2156  *	In both cases, however, we keep our figurative eye out for the
2157  *	'noPrint' line for the shell from which the output came. If
2158  *	we recognize a line, we don't print it. If the command is not
2159  *	alone on the line (the character after it is not \0 or \n), we
2160  *	do print whatever follows it.
2161  *
2162  * Input:
2163  *	job		the job whose output needs printing
2164  *	finish		TRUE if this is the last time we'll be called
2165  *			for this job
2166  *
2167  * Results:
2168  *	None
2169  *
2170  * Side Effects:
2171  *	curPos may be shifted as may the contents of outBuf.
2172  *-----------------------------------------------------------------------
2173  */
2174 STATIC void
2175 JobDoOutput(Job *job, Boolean finish)
2176 {
2177     Boolean       gotNL = FALSE;  /* true if got a newline */
2178     Boolean       fbuf;  	  /* true if our buffer filled up */
2179     int		  nr;	      	  /* number of bytes read */
2180     int		  i;	      	  /* auxiliary index into outBuf */
2181     int		  max;	      	  /* limit for i (end of current data) */
2182     int		  nRead;      	  /* (Temporary) number of bytes read */
2183 
2184     FILE      	  *oFILE;	  /* Stream pointer to shell's output file */
2185     char          inLine[132];
2186 
2187 
2188     if (usePipes) {
2189 	/*
2190 	 * Read as many bytes as will fit in the buffer.
2191 	 */
2192 end_loop:
2193 	gotNL = FALSE;
2194 	fbuf = FALSE;
2195 
2196 	nRead = read(job->inPipe, &job->outBuf[job->curPos],
2197 			 JOB_BUFSIZE - job->curPos);
2198 	if (nRead < 0) {
2199 	    if (DEBUG(JOB)) {
2200 		perror("JobDoOutput(piperead)");
2201 	    }
2202 	    nr = 0;
2203 	} else {
2204 	    nr = nRead;
2205 	}
2206 
2207 	/*
2208 	 * If we hit the end-of-file (the job is dead), we must flush its
2209 	 * remaining output, so pretend we read a newline if there's any
2210 	 * output remaining in the buffer.
2211 	 * Also clear the 'finish' flag so we stop looping.
2212 	 */
2213 	if ((nr == 0) && (job->curPos != 0)) {
2214 	    job->outBuf[job->curPos] = '\n';
2215 	    nr = 1;
2216 	    finish = FALSE;
2217 	} else if (nr == 0) {
2218 	    finish = FALSE;
2219 	}
2220 
2221 	/*
2222 	 * Look for the last newline in the bytes we just got. If there is
2223 	 * one, break out of the loop with 'i' as its index and gotNL set
2224 	 * TRUE.
2225 	 */
2226 	max = job->curPos + nr;
2227 	for (i = job->curPos + nr - 1; i >= job->curPos; i--) {
2228 	    if (job->outBuf[i] == '\n') {
2229 		gotNL = TRUE;
2230 		break;
2231 	    } else if (job->outBuf[i] == '\0') {
2232 		/*
2233 		 * Why?
2234 		 */
2235 		job->outBuf[i] = ' ';
2236 	    }
2237 	}
2238 
2239 	if (!gotNL) {
2240 	    job->curPos += nr;
2241 	    if (job->curPos == JOB_BUFSIZE) {
2242 		/*
2243 		 * If we've run out of buffer space, we have no choice
2244 		 * but to print the stuff. sigh.
2245 		 */
2246 		fbuf = TRUE;
2247 		i = job->curPos;
2248 	    }
2249 	}
2250 	if (gotNL || fbuf) {
2251 	    /*
2252 	     * Need to send the output to the screen. Null terminate it
2253 	     * first, overwriting the newline character if there was one.
2254 	     * So long as the line isn't one we should filter (according
2255 	     * to the shell description), we print the line, preceded
2256 	     * by a target banner if this target isn't the same as the
2257 	     * one for which we last printed something.
2258 	     * The rest of the data in the buffer are then shifted down
2259 	     * to the start of the buffer and curPos is set accordingly.
2260 	     */
2261 	    job->outBuf[i] = '\0';
2262 	    if (i >= job->curPos) {
2263 		char *cp;
2264 
2265 		cp = JobOutput(job, job->outBuf, &job->outBuf[i], FALSE);
2266 
2267 		/*
2268 		 * There's still more in that thar buffer. This time, though,
2269 		 * we know there's no newline at the end, so we add one of
2270 		 * our own free will.
2271 		 */
2272 		if (*cp != '\0') {
2273 		    if (job->node != lastNode) {
2274 			MESSAGE(stdout, job->node);
2275 			lastNode = job->node;
2276 		    }
2277 		    (void) fprintf(stdout, "%s%s", cp, gotNL ? "\n" : "");
2278 		    (void) fflush(stdout);
2279 		}
2280 	    }
2281 	    if (i < max - 1) {
2282 		/* shift the remaining characters down */
2283 		(void) memcpy(job->outBuf, &job->outBuf[i + 1], max - (i + 1));
2284 		job->curPos = max - (i + 1);
2285 
2286 	    } else {
2287 		/*
2288 		 * We have written everything out, so we just start over
2289 		 * from the start of the buffer. No copying. No nothing.
2290 		 */
2291 		job->curPos = 0;
2292 	    }
2293 	}
2294 	if (finish) {
2295 	    /*
2296 	     * If the finish flag is true, we must loop until we hit
2297 	     * end-of-file on the pipe. This is guaranteed to happen
2298 	     * eventually since the other end of the pipe is now closed
2299 	     * (we closed it explicitly and the child has exited). When
2300 	     * we do get an EOF, finish will be set FALSE and we'll fall
2301 	     * through and out.
2302 	     */
2303 	    goto end_loop;
2304 	}
2305     } else {
2306 	/*
2307 	 * We've been called to retrieve the output of the job from the
2308 	 * temporary file where it's been squirreled away. This consists of
2309 	 * opening the file, reading the output line by line, being sure not
2310 	 * to print the noPrint line for the shell we used, then close and
2311 	 * remove the temporary file. Very simple.
2312 	 *
2313 	 * Change to read in blocks and do FindSubString type things as for
2314 	 * pipes? That would allow for "@echo -n..."
2315 	 */
2316 	oFILE = fopen(job->outFile, "r");
2317 	if (oFILE != NULL) {
2318 	    (void) fprintf(stdout, "Results of making %s:\n", job->node->name);
2319 	    (void) fflush(stdout);
2320 	    while (fgets(inLine, sizeof(inLine), oFILE) != NULL) {
2321 		char	*cp, *endp, *oendp;
2322 
2323 		cp = inLine;
2324 		oendp = endp = inLine + strlen(inLine);
2325 		if (endp[-1] == '\n') {
2326 		    *--endp = '\0';
2327 		}
2328 		cp = JobOutput(job, inLine, endp, FALSE);
2329 
2330 		/*
2331 		 * There's still more in that thar buffer. This time, though,
2332 		 * we know there's no newline at the end, so we add one of
2333 		 * our own free will.
2334 		 */
2335 		(void) fprintf(stdout, "%s", cp);
2336 		(void) fflush(stdout);
2337 		if (endp != oendp) {
2338 		    (void) fprintf(stdout, "\n");
2339 		    (void) fflush(stdout);
2340 		}
2341 	    }
2342 	    (void) fclose(oFILE);
2343 	    (void) eunlink(job->outFile);
2344 	} else {
2345 	    Punt("Cannot open `%s'", job->outFile);
2346 	}
2347     }
2348 }
2349 
2350 /*-
2351  *-----------------------------------------------------------------------
2352  * Job_CatchChildren --
2353  *	Handle the exit of a child. Called from Make_Make.
2354  *
2355  * Input:
2356  *	block		TRUE if should block on the wait
2357  *
2358  * Results:
2359  *	none.
2360  *
2361  * Side Effects:
2362  *	The job descriptor is removed from the list of children.
2363  *
2364  * Notes:
2365  *	We do waits, blocking or not, according to the wisdom of our
2366  *	caller, until there are no more children to report. For each
2367  *	job, call JobFinish to finish things off. This will take care of
2368  *	putting jobs on the stoppedJobs queue.
2369  *
2370  *-----------------------------------------------------------------------
2371  */
2372 void
2373 Job_CatchChildren(Boolean block)
2374 {
2375     int    	  pid;	    	/* pid of dead child */
2376     Job		  *job;	    	/* job descriptor for dead child */
2377     LstNode       jnode;    	/* list element for finding job */
2378     int	  	  status;   	/* Exit/termination status */
2379 
2380     /*
2381      * Don't even bother if we know there's no one around.
2382      */
2383     if (nLocal == 0) {
2384 	return;
2385     }
2386 
2387     while ((pid = waitpid((pid_t) -1, &status,
2388 			  (block?0:WNOHANG)|WUNTRACED)) > 0)
2389     {
2390 	if (DEBUG(JOB)) {
2391 	    (void) fprintf(stdout, "Process %d exited or stopped %x.\n", pid,
2392 	      status);
2393 	    (void) fflush(stdout);
2394 	}
2395 
2396 	jnode = Lst_Find(jobs, (ClientData)&pid, JobCmpPid);
2397 	if (jnode == NILLNODE) {
2398 	    if (WIFSTOPPED(status) && (WSTOPSIG(status) == SIGCONT)) {
2399 		jnode = Lst_Find(stoppedJobs, (ClientData) &pid, JobCmpPid);
2400 		if (jnode == NILLNODE) {
2401 		    Error("Resumed child (%d) not in table", pid);
2402 		    continue;
2403 		}
2404 		job = (Job *)Lst_Datum(jnode);
2405 		(void) Lst_Remove(stoppedJobs, jnode);
2406 	    } else {
2407 		Error("Child (%d) not in table?", pid);
2408 		continue;
2409 	    }
2410 	} else {
2411 	    job = (Job *) Lst_Datum(jnode);
2412 	    (void) Lst_Remove(jobs, jnode);
2413 	    nJobs -= 1;
2414 #ifdef REMOTE
2415 	    if (!(job->flags & JOB_REMOTE)) {
2416 		if (DEBUG(JOB)) {
2417 		    (void) fprintf(stdout,
2418 			   "Job queue has one fewer local process.\n");
2419 		    (void) fflush(stdout);
2420 		}
2421 		nLocal -= 1;
2422 	    }
2423 #else
2424 	    nLocal -= 1;
2425 #endif
2426 	}
2427 
2428 	JobFinish(job, &status);
2429     }
2430 }
2431 
2432 /*-
2433  *-----------------------------------------------------------------------
2434  * Job_CatchOutput --
2435  *	Catch the output from our children, if we're using
2436  *	pipes do so. Otherwise just block time until we get a
2437  *	signal (most likely a SIGCHLD) since there's no point in
2438  *	just spinning when there's nothing to do and the reaping
2439  *	of a child can wait for a while.
2440  *
2441  * Results:
2442  *	None
2443  *
2444  * Side Effects:
2445  *	Output is read from pipes if we're piping.
2446  * -----------------------------------------------------------------------
2447  */
2448 void
2449 Job_CatchOutput(void)
2450 {
2451     int           	  nready;
2452     LstNode		  ln;
2453     Job  	 	  *job;
2454 #ifdef RMT_WILL_WATCH
2455     int	    	  	  pnJobs;   	/* Previous nJobs */
2456 #endif
2457 
2458     (void) fflush(stdout);
2459     Job_TokenFlush();
2460 #ifdef RMT_WILL_WATCH
2461     pnJobs = nJobs;
2462 
2463     /*
2464      * It is possible for us to be called with nJobs equal to 0. This happens
2465      * if all the jobs finish and a job that is stopped cannot be run
2466      * locally (eg if maxLocal is 0) and cannot be exported. The job will
2467      * be placed back on the stoppedJobs queue, Job_Empty() will return false,
2468      * Make_Run will call us again when there's nothing for which to wait.
2469      * nJobs never changes, so we loop forever. Hence the check. It could
2470      * be argued that we should sleep for a bit so as not to swamp the
2471      * exportation system with requests. Perhaps we should.
2472      *
2473      * NOTE: IT IS THE RESPONSIBILITY OF Rmt_Wait TO CALL Job_CatchChildren
2474      * IN A TIMELY FASHION TO CATCH ANY LOCALLY RUNNING JOBS THAT EXIT.
2475      * It may use the variable nLocal to determine if it needs to call
2476      * Job_CatchChildren (if nLocal is 0, there's nothing for which to
2477      * wait...)
2478      */
2479     while (nJobs != 0 && pnJobs == nJobs) {
2480 	Rmt_Wait();
2481     }
2482 #else
2483     if (usePipes) {
2484 	if ((nready = poll((wantToken ? fds : (fds + 1)),
2485 	  		   (wantToken ? nfds : (nfds - 1)), POLL_MSEC)) <= 0) {
2486 	    return;
2487 	} else {
2488 	    sigset_t	mask;
2489 	    JobSigLock(&mask);
2490 	    if (Lst_Open(jobs) == FAILURE) {
2491 		Punt("Cannot open job table");
2492 	    }
2493 
2494 	    if (readyfd(&childExitJob)) {
2495 		char token;
2496 		(void) read(childExitJob.inPipe, &token, 1);
2497 		nready -= 1;
2498 	    }
2499 
2500 	    while (nready && (ln = Lst_Next(jobs)) != NILLNODE) {
2501 		job = (Job *) Lst_Datum(ln);
2502 		if (readyfd(job)) {
2503 		    JobDoOutput(job, FALSE);
2504 		    nready -= 1;
2505 		}
2506 	    }
2507 	    Lst_Close(jobs);
2508 	    JobSigUnlock(&mask);
2509 	}
2510     }
2511 #endif /* RMT_WILL_WATCH */
2512 }
2513 
2514 /*-
2515  *-----------------------------------------------------------------------
2516  * Job_Make --
2517  *	Start the creation of a target. Basically a front-end for
2518  *	JobStart used by the Make module.
2519  *
2520  * Results:
2521  *	None.
2522  *
2523  * Side Effects:
2524  *	Another job is started.
2525  *
2526  *-----------------------------------------------------------------------
2527  */
2528 void
2529 Job_Make(GNode *gn)
2530 {
2531     (void) JobStart(gn, 0, NULL);
2532 }
2533 
2534 /*-
2535  *-----------------------------------------------------------------------
2536  * Job_Init --
2537  *	Initialize the process module
2538  *
2539  * Input:
2540  *	maxproc		the greatest number of jobs which may be running
2541  *			at one time
2542  *	maxlocal	the greatest number of jobs which may be running
2543  *			at once
2544  *
2545  * Results:
2546  *	none
2547  *
2548  * Side Effects:
2549  *	lists and counters are initialized
2550  *-----------------------------------------------------------------------
2551  */
2552 void
2553 Job_Init(int maxproc, int maxlocal)
2554 {
2555     GNode         *begin;     /* node for commands to do at the very start */
2556 
2557     jobs =  	  Lst_Init(FALSE);
2558     stoppedJobs = Lst_Init(FALSE);
2559     maxJobs = 	  maxproc;
2560     maxLocal = 	  maxlocal;
2561     nJobs = 	  0;
2562     nLocal = 	  0;
2563     wantToken =	  FALSE;
2564 
2565     aborting = 	  0;
2566     errors = 	  0;
2567 
2568     lastNode =	  NILGNODE;
2569 
2570     if (maxJobs == 1
2571 #ifdef REMOTE
2572 	|| noMessages
2573 #endif
2574 		     ) {
2575 	/*
2576 	 * If only one job can run at a time, there's no need for a banner,
2577 	 * is there?
2578 	 */
2579 	targFmt = "";
2580     } else {
2581 	targFmt = TARG_FMT;
2582     }
2583 
2584     if (shellPath == NULL) {
2585 	/*
2586 	 * The user didn't specify a shell to use, so we are using the
2587 	 * default one... Both the absolute path and the last component
2588 	 * must be set. The last component is taken from the 'name' field
2589 	 * of the default shell description pointed-to by commandShell.
2590 	 * All default shells are located in _PATH_DEFSHELLDIR.
2591 	 */
2592 	shellName = commandShell->name;
2593 	shellPath = str_concat(_PATH_DEFSHELLDIR, shellName, STR_ADDSLASH);
2594     }
2595 
2596     if (commandShell->exit == NULL) {
2597 	commandShell->exit = "";
2598     }
2599     if (commandShell->echo == NULL) {
2600 	commandShell->echo = "";
2601     }
2602 
2603     if (pipe(exit_pipe) < 0)
2604 	Fatal("error in pipe: %s", strerror(errno));
2605     fcntl(exit_pipe[0], F_SETFD, 1);
2606     fcntl(exit_pipe[1], F_SETFD, 1);
2607 
2608     childExitJob.inPipe = exit_pipe[0];
2609 
2610     sigemptyset(&caught_signals);
2611     /*
2612      * Install a SIGCHLD handler.
2613      */
2614     (void)signal(SIGCHLD, JobChildSig);
2615     sigaddset(&caught_signals, SIGCHLD);
2616 
2617 #define ADDSIG(s,h)				\
2618     if (signal(s, SIG_IGN) != SIG_IGN) {	\
2619 	sigaddset(&caught_signals, s);		\
2620 	(void) signal(s, h);			\
2621     }
2622 
2623     /*
2624      * Catch the four signals that POSIX specifies if they aren't ignored.
2625      * JobPassSig will take care of calling JobInterrupt if appropriate.
2626      */
2627     ADDSIG(SIGINT, JobPassSig)
2628     ADDSIG(SIGHUP, JobPassSig)
2629     ADDSIG(SIGTERM, JobPassSig)
2630     ADDSIG(SIGQUIT, JobPassSig)
2631 
2632     /*
2633      * There are additional signals that need to be caught and passed if
2634      * either the export system wants to be told directly of signals or if
2635      * we're giving each job its own process group (since then it won't get
2636      * signals from the terminal driver as we own the terminal)
2637      */
2638 #if defined(RMT_WANTS_SIGNALS) || defined(USE_PGRP)
2639     ADDSIG(SIGTSTP, JobPassSig)
2640     ADDSIG(SIGTTOU, JobPassSig)
2641     ADDSIG(SIGTTIN, JobPassSig)
2642     ADDSIG(SIGWINCH, JobPassSig)
2643     ADDSIG(SIGCONT, JobContinueSig)
2644 #endif
2645 #undef ADDSIG
2646 
2647     begin = Targ_FindNode(".BEGIN", TARG_NOCREATE);
2648 
2649     if (begin != NILGNODE) {
2650 	JobStart(begin, JOB_SPECIAL, (Job *)0);
2651 	while (nJobs) {
2652 	    Job_CatchOutput();
2653 #ifndef RMT_WILL_WATCH
2654 	    Job_CatchChildren(!usePipes);
2655 #endif /* RMT_WILL_WATCH */
2656 	}
2657     }
2658     postCommands = Targ_FindNode(".END", TARG_CREATE);
2659 }
2660 
2661 static void JobSigReset(void)
2662 {
2663 #define DELSIG(s)					\
2664     if (sigismember(&caught_signals, s)) {		\
2665 	(void) signal(SIGINT, SIG_DFL);			\
2666     }
2667 
2668     DELSIG(SIGINT)
2669     DELSIG(SIGHUP)
2670     DELSIG(SIGQUIT)
2671     DELSIG(SIGTERM)
2672 #if defined(RMT_WANTS_SIGNALS) || defined(USE_PGRP)
2673     DELSIG(SIGTSTP)
2674     DELSIG(SIGTTOU)
2675     DELSIG(SIGTTIN)
2676     DELSIG(SIGWINCH)
2677     DELSIG(SIGCONT)
2678 #endif
2679 #undef DELSIG
2680     (void)signal(SIGCHLD, SIG_DFL);
2681 }
2682 
2683 /*-
2684  *-----------------------------------------------------------------------
2685  * Job_Empty --
2686  *	See if the job table is empty.  Because the local concurrency may
2687  *	be set to 0, it is possible for the job table to become empty,
2688  *	while the list of stoppedJobs remains non-empty. In such a case,
2689  *	we want to restart as many jobs as we can.
2690  *
2691  * Results:
2692  *	TRUE if it is. FALSE if it ain't.
2693  *
2694  * Side Effects:
2695  *	None.
2696  *
2697  * -----------------------------------------------------------------------
2698  */
2699 Boolean
2700 Job_Empty(void)
2701 {
2702     if (nJobs == 0) {
2703 	if (!Lst_IsEmpty(stoppedJobs) && !aborting) {
2704 	    /*
2705 	     * The job table is obviously not full if it has no jobs in
2706 	     * it...Try and restart the stopped jobs.
2707 	     */
2708 	    JobRestartJobs();
2709 	    return(FALSE);
2710 	} else {
2711 	    return(TRUE);
2712 	}
2713     } else {
2714 	return(FALSE);
2715     }
2716 }
2717 
2718 /*-
2719  *-----------------------------------------------------------------------
2720  * JobMatchShell --
2721  *	Find a shell in 'shells' given its name.
2722  *
2723  * Results:
2724  *	A pointer to the Shell structure.
2725  *
2726  * Side Effects:
2727  *	None.
2728  *
2729  *-----------------------------------------------------------------------
2730  */
2731 static Shell *
2732 JobMatchShell(char *name)
2733 {
2734     Shell	*sh;
2735 
2736     for (sh = shells; sh->name != NULL; sh++) {
2737 	if (strcmp(name, sh->name) == 0)
2738 		return (sh);
2739     }
2740     return (NULL);
2741 }
2742 
2743 /*-
2744  *-----------------------------------------------------------------------
2745  * Job_ParseShell --
2746  *	Parse a shell specification and set up commandShell, shellPath
2747  *	and shellName appropriately.
2748  *
2749  * Input:
2750  *	line		The shell spec
2751  *
2752  * Results:
2753  *	FAILURE if the specification was incorrect.
2754  *
2755  * Side Effects:
2756  *	commandShell points to a Shell structure (either predefined or
2757  *	created from the shell spec), shellPath is the full path of the
2758  *	shell described by commandShell, while shellName is just the
2759  *	final component of shellPath.
2760  *
2761  * Notes:
2762  *	A shell specification consists of a .SHELL target, with dependency
2763  *	operator, followed by a series of blank-separated words. Double
2764  *	quotes can be used to use blanks in words. A backslash escapes
2765  *	anything (most notably a double-quote and a space) and
2766  *	provides the functionality it does in C. Each word consists of
2767  *	keyword and value separated by an equal sign. There should be no
2768  *	unnecessary spaces in the word. The keywords are as follows:
2769  *	    name  	    Name of shell.
2770  *	    path  	    Location of shell.
2771  *	    quiet 	    Command to turn off echoing.
2772  *	    echo  	    Command to turn echoing on
2773  *	    filter	    Result of turning off echoing that shouldn't be
2774  *	    	  	    printed.
2775  *	    echoFlag	    Flag to turn echoing on at the start
2776  *	    errFlag	    Flag to turn error checking on at the start
2777  *	    hasErrCtl	    True if shell has error checking control
2778  *	    check 	    Command to turn on error checking if hasErrCtl
2779  *	    	  	    is TRUE or template of command to echo a command
2780  *	    	  	    for which error checking is off if hasErrCtl is
2781  *	    	  	    FALSE.
2782  *	    ignore	    Command to turn off error checking if hasErrCtl
2783  *	    	  	    is TRUE or template of command to execute a
2784  *	    	  	    command so as to ignore any errors it returns if
2785  *	    	  	    hasErrCtl is FALSE.
2786  *
2787  *-----------------------------------------------------------------------
2788  */
2789 ReturnStatus
2790 Job_ParseShell(char *line)
2791 {
2792     char	**words;
2793     char	**argv;
2794     int		argc;
2795     char	*path;
2796     Shell	newShell;
2797     Boolean	fullSpec = FALSE;
2798     Shell	*sh;
2799 
2800     while (isspace((unsigned char)*line)) {
2801 	line++;
2802     }
2803 
2804     if (shellArgv)
2805 	free(shellArgv);
2806 
2807     memset((Address)&newShell, 0, sizeof(newShell));
2808 
2809     /*
2810      * Parse the specification by keyword
2811      */
2812     words = brk_string(line, &argc, TRUE, &shellArgv);
2813 
2814     for (path = NULL, argv = words; argc != 0; argc--, argv++) {
2815 	    if (strncmp(*argv, "path=", 5) == 0) {
2816 		path = &argv[0][5];
2817 	    } else if (strncmp(*argv, "name=", 5) == 0) {
2818 		newShell.name = &argv[0][5];
2819 	    } else {
2820 		if (strncmp(*argv, "quiet=", 6) == 0) {
2821 		    newShell.echoOff = &argv[0][6];
2822 		} else if (strncmp(*argv, "echo=", 5) == 0) {
2823 		    newShell.echoOn = &argv[0][5];
2824 		} else if (strncmp(*argv, "filter=", 7) == 0) {
2825 		    newShell.noPrint = &argv[0][7];
2826 		    newShell.noPLen = strlen(newShell.noPrint);
2827 		} else if (strncmp(*argv, "echoFlag=", 9) == 0) {
2828 		    newShell.echo = &argv[0][9];
2829 		} else if (strncmp(*argv, "errFlag=", 8) == 0) {
2830 		    newShell.exit = &argv[0][8];
2831 		} else if (strncmp(*argv, "hasErrCtl=", 10) == 0) {
2832 		    char c = argv[0][10];
2833 		    newShell.hasErrCtl = !((c != 'Y') && (c != 'y') &&
2834 					   (c != 'T') && (c != 't'));
2835 		} else if (strncmp(*argv, "check=", 6) == 0) {
2836 		    newShell.errCheck = &argv[0][6];
2837 		} else if (strncmp(*argv, "ignore=", 7) == 0) {
2838 		    newShell.ignErr = &argv[0][7];
2839 		} else {
2840 		    Parse_Error(PARSE_FATAL, "Unknown keyword \"%s\"",
2841 				*argv);
2842 		    free(words);
2843 		    return(FAILURE);
2844 		}
2845 		fullSpec = TRUE;
2846 	    }
2847     }
2848 
2849     if (path == NULL) {
2850 	/*
2851 	 * If no path was given, the user wants one of the pre-defined shells,
2852 	 * yes? So we find the one s/he wants with the help of JobMatchShell
2853 	 * and set things up the right way. shellPath will be set up by
2854 	 * Job_Init.
2855 	 */
2856 	if (newShell.name == NULL) {
2857 	    Parse_Error(PARSE_FATAL, "Neither path nor name specified");
2858 	    free(words);
2859 	    return(FAILURE);
2860 	} else {
2861 	    if ((sh = JobMatchShell(newShell.name)) == NULL) {
2862 		    Parse_Error(PARSE_WARNING, "%s: No matching shell",
2863 				newShell.name);
2864 		    free(words);
2865 		    return(FAILURE);
2866 	    }
2867 	    commandShell = sh;
2868 	    shellName = newShell.name;
2869 	}
2870     } else {
2871 	/*
2872 	 * The user provided a path. If s/he gave nothing else (fullSpec is
2873 	 * FALSE), try and find a matching shell in the ones we know of.
2874 	 * Else we just take the specification at its word and copy it
2875 	 * to a new location. In either case, we need to record the
2876 	 * path the user gave for the shell.
2877 	 */
2878 	shellPath = path;
2879 	path = strrchr(path, '/');
2880 	if (path == NULL) {
2881 	    path = shellPath;
2882 	} else {
2883 	    path += 1;
2884 	}
2885 	if (newShell.name != NULL) {
2886 	    shellName = newShell.name;
2887 	} else {
2888 	    shellName = path;
2889 	}
2890 	if (!fullSpec) {
2891 	    if ((sh = JobMatchShell(shellName)) == NULL) {
2892 		    Parse_Error(PARSE_WARNING, "%s: No matching shell",
2893 				shellName);
2894 		    free(words);
2895 		    return(FAILURE);
2896 	    }
2897 	    commandShell = sh;
2898 	} else {
2899 	    commandShell = (Shell *) emalloc(sizeof(Shell));
2900 	    *commandShell = newShell;
2901 	}
2902     }
2903 
2904     if (commandShell->echoOn && commandShell->echoOff) {
2905 	commandShell->hasEchoCtl = TRUE;
2906     }
2907 
2908     if (!commandShell->hasErrCtl) {
2909 	if (commandShell->errCheck == NULL) {
2910 	    commandShell->errCheck = "";
2911 	}
2912 	if (commandShell->ignErr == NULL) {
2913 	    commandShell->ignErr = "%s\n";
2914 	}
2915     }
2916 
2917     /*
2918      * Do not free up the words themselves, since they might be in use by the
2919      * shell specification.
2920      */
2921     free(words);
2922     return SUCCESS;
2923 }
2924 
2925 /*-
2926  *-----------------------------------------------------------------------
2927  * JobInterrupt --
2928  *	Handle the receipt of an interrupt.
2929  *
2930  * Input:
2931  *	runINTERRUPT	Non-zero if commands for the .INTERRUPT target
2932  *			should be executed
2933  *	signo		signal received
2934  *
2935  * Results:
2936  *	None
2937  *
2938  * Side Effects:
2939  *	All children are killed. Another job will be started if the
2940  *	.INTERRUPT target was given.
2941  *-----------------------------------------------------------------------
2942  */
2943 static void
2944 JobInterrupt(int runINTERRUPT, int signo)
2945 {
2946     LstNode	ln;		/* element in job table */
2947     Job		*job;		/* job descriptor in that element */
2948     GNode	*interrupt;	/* the node describing the .INTERRUPT target */
2949     sigset_t	mask;
2950 
2951     aborting = ABORT_INTERRUPT;
2952 
2953     JobSigLock(&mask);
2954 
2955     (void) Lst_Open(jobs);
2956     while ((ln = Lst_Next(jobs)) != NILLNODE) {
2957 	GNode *gn;
2958 
2959 	job = (Job *) Lst_Datum(ln);
2960 	gn = job->node;
2961 
2962 	if ((gn->type & (OP_JOIN|OP_PHONY)) == 0 && !Targ_Precious(gn)) {
2963 	    char *file = (gn->path == NULL ? gn->name : gn->path);
2964 	    if (!noExecute && eunlink(file) != -1) {
2965 		Error("*** %s removed", file);
2966 	    }
2967 	}
2968 #ifdef RMT_WANTS_SIGNALS
2969 	if (job->flags & JOB_REMOTE) {
2970 	    /*
2971 	     * If job is remote, let the Rmt module do the killing.
2972 	     */
2973 	    if (!Rmt_Signal(job, signo)) {
2974 		/*
2975 		 * If couldn't kill the thing, finish it out now with an
2976 		 * error code, since no exit report will come in likely.
2977 		 */
2978 		int status;
2979 
2980 		status.w_status = 0;
2981 		status.w_retcode = 1;
2982 		JobFinish(job, &status);
2983 	    }
2984 	} else if (job->pid) {
2985 	    KILL(job->pid, signo);
2986 	}
2987 #else
2988 	if (job->pid) {
2989 	    if (DEBUG(JOB)) {
2990 		(void) fprintf(stdout,
2991 			   "JobInterrupt passing signal %d to child %d.\n",
2992 			   signo, job->pid);
2993 		(void) fflush(stdout);
2994 	    }
2995 	    KILL(job->pid, signo);
2996 	}
2997 #endif /* RMT_WANTS_SIGNALS */
2998     }
2999     Lst_Close(jobs);
3000 
3001 #ifdef REMOTE
3002    (void)Lst_Open(stoppedJobs);
3003     while ((ln = Lst_Next(stoppedJobs)) != NILLNODE) {
3004 	GNode *gn;
3005 
3006 	job = (Job *) Lst_Datum(ln);
3007 	gn = job->node;
3008 
3009 	if (job->flags & JOB_RESTART) {
3010 	    if (DEBUG(JOB)) {
3011 		(void) fprintf(stdout, "%s%s",
3012 			       "JobInterrupt skipping job on stopped queue",
3013 			       "-- it was waiting to be restarted.\n");
3014 		(void) fflush(stdout);
3015 	    }
3016 	    continue;
3017 	}
3018 	if ((gn->type & (OP_JOIN|OP_PHONY)) == 0 && !Targ_Precious(gn)) {
3019 	    char *file = (gn->path == NULL ? gn->name : gn->path);
3020 	    if (eunlink(file) == 0) {
3021 		Error("*** %s removed", file);
3022 	    }
3023 	}
3024 	/*
3025 	 * Resume the thing so it will take the signal.
3026 	 */
3027 	if (DEBUG(JOB)) {
3028 	    (void) fprintf(stdout,
3029 			   "JobInterrupt passing CONT to stopped child %d.\n",
3030 			   job->pid);
3031 	    (void) fflush(stdout);
3032 	}
3033 	KILL(job->pid, SIGCONT);
3034 #ifdef RMT_WANTS_SIGNALS
3035 	if (job->flags & JOB_REMOTE) {
3036 	    /*
3037 	     * If job is remote, let the Rmt module do the killing.
3038 	     */
3039 	    if (!Rmt_Signal(job, SIGINT)) {
3040 		/*
3041 		 * If couldn't kill the thing, finish it out now with an
3042 		 * error code, since no exit report will come in likely.
3043 		 */
3044 		int status;
3045 		status.w_status = 0;
3046 		status.w_retcode = 1;
3047 		JobFinish(job, &status);
3048 	    }
3049 	} else if (job->pid) {
3050 	    if (DEBUG(JOB)) {
3051 		(void) fprintf(stdout,
3052 		       "JobInterrupt passing interrupt to stopped child %d.\n",
3053 			       job->pid);
3054 		(void) fflush(stdout);
3055 	    }
3056 	    KILL(job->pid, SIGINT);
3057 	}
3058 #endif /* RMT_WANTS_SIGNALS */
3059     }
3060     Lst_Close(stoppedJobs);
3061 #endif /* REMOTE */
3062 
3063     JobSigUnlock(&mask);
3064 
3065     if (runINTERRUPT && !touchFlag) {
3066 	interrupt = Targ_FindNode(".INTERRUPT", TARG_NOCREATE);
3067 	if (interrupt != NILGNODE) {
3068 	    ignoreErrors = FALSE;
3069 
3070 	    JobStart(interrupt, JOB_IGNDOTS, (Job *)0);
3071 	    while (nJobs) {
3072 		Job_CatchOutput();
3073 #ifndef RMT_WILL_WATCH
3074 		Job_CatchChildren(!usePipes);
3075 #endif /* RMT_WILL_WATCH */
3076 	    }
3077 	}
3078     }
3079     Trace_Log(MAKEINTR, 0);
3080     exit(signo);
3081 }
3082 
3083 /*
3084  *-----------------------------------------------------------------------
3085  * Job_Finish --
3086  *	Do final processing such as the running of the commands
3087  *	attached to the .END target.
3088  *
3089  * Results:
3090  *	Number of errors reported.
3091  *
3092  * Side Effects:
3093  *	None.
3094  *-----------------------------------------------------------------------
3095  */
3096 int
3097 Job_Finish(void)
3098 {
3099     if (postCommands != NILGNODE && !Lst_IsEmpty(postCommands->commands)) {
3100 	if (errors) {
3101 	    Error("Errors reported so .END ignored");
3102 	} else {
3103 	    JobStart(postCommands, JOB_SPECIAL | JOB_IGNDOTS, NULL);
3104 
3105 	    while (nJobs) {
3106 		Job_CatchOutput();
3107 #ifndef RMT_WILL_WATCH
3108 		Job_CatchChildren(!usePipes);
3109 #endif /* RMT_WILL_WATCH */
3110 	    }
3111 	}
3112     }
3113     Job_TokenFlush();
3114     return(errors);
3115 }
3116 
3117 /*-
3118  *-----------------------------------------------------------------------
3119  * Job_End --
3120  *	Cleanup any memory used by the jobs module
3121  *
3122  * Results:
3123  *	None.
3124  *
3125  * Side Effects:
3126  *	Memory is freed
3127  *-----------------------------------------------------------------------
3128  */
3129 void
3130 Job_End(void)
3131 {
3132 #ifdef CLEANUP
3133     if (shellArgv)
3134 	free(shellArgv);
3135 #endif
3136 }
3137 
3138 /*-
3139  *-----------------------------------------------------------------------
3140  * Job_Wait --
3141  *	Waits for all running jobs to finish and returns. Sets 'aborting'
3142  *	to ABORT_WAIT to prevent other jobs from starting.
3143  *
3144  * Results:
3145  *	None.
3146  *
3147  * Side Effects:
3148  *	Currently running jobs finish.
3149  *
3150  *-----------------------------------------------------------------------
3151  */
3152 void
3153 Job_Wait(void)
3154 {
3155     aborting = ABORT_WAIT;
3156     while (nJobs != 0) {
3157 	Job_CatchOutput();
3158 #ifndef RMT_WILL_WATCH
3159 	Job_CatchChildren(!usePipes);
3160 #endif /* RMT_WILL_WATCH */
3161     }
3162     Job_TokenFlush();
3163     aborting = 0;
3164 }
3165 
3166 /*-
3167  *-----------------------------------------------------------------------
3168  * Job_AbortAll --
3169  *	Abort all currently running jobs without handling output or anything.
3170  *	This function is to be called only in the event of a major
3171  *	error. Most definitely NOT to be called from JobInterrupt.
3172  *
3173  * Results:
3174  *	None
3175  *
3176  * Side Effects:
3177  *	All children are killed, not just the firstborn
3178  *-----------------------------------------------------------------------
3179  */
3180 void
3181 Job_AbortAll(void)
3182 {
3183     LstNode	ln;	/* element in job table */
3184     Job		*job;	/* the job descriptor in that element */
3185     int		foo;
3186     sigset_t	mask;
3187 
3188     aborting = ABORT_ERROR;
3189 
3190     if (nJobs) {
3191 
3192 	JobSigLock(&mask);
3193 	(void) Lst_Open(jobs);
3194 	while ((ln = Lst_Next(jobs)) != NILLNODE) {
3195 	    job = (Job *) Lst_Datum(ln);
3196 
3197 	    /*
3198 	     * kill the child process with increasingly drastic signals to make
3199 	     * darn sure it's dead.
3200 	     */
3201 #ifdef RMT_WANTS_SIGNALS
3202 	    if (job->flags & JOB_REMOTE) {
3203 		Rmt_Signal(job, SIGINT);
3204 		Rmt_Signal(job, SIGKILL);
3205 	    } else {
3206 		KILL(job->pid, SIGINT);
3207 		KILL(job->pid, SIGKILL);
3208 	    }
3209 #else
3210 	    KILL(job->pid, SIGINT);
3211 	    KILL(job->pid, SIGKILL);
3212 #endif /* RMT_WANTS_SIGNALS */
3213 	}
3214 	Lst_Close(jobs);
3215 	JobSigUnlock(&mask);
3216     }
3217 
3218     /*
3219      * Catch as many children as want to report in at first, then give up
3220      */
3221     while (waitpid((pid_t) -1, &foo, WNOHANG) > 0)
3222 	continue;
3223 }
3224 
3225 #ifdef REMOTE
3226 /*-
3227  *-----------------------------------------------------------------------
3228  * JobFlagForMigration --
3229  *	Handle the eviction of a child. Called from RmtStatusChange.
3230  *	Flags the child as remigratable and then suspends it.
3231  *
3232  * Input:
3233  *	hostID		ID of host we used, for matching children
3234  *
3235  * Results:
3236  *	none.
3237  *
3238  * Side Effects:
3239  *	The job descriptor is flagged for remigration.
3240  *
3241  *-----------------------------------------------------------------------
3242  */
3243 void
3244 JobFlagForMigration(int hostID)
3245 {
3246     Job		  *job;	    	/* job descriptor for dead child */
3247     LstNode       jnode;    	/* list element for finding job */
3248 
3249     if (DEBUG(JOB)) {
3250 	(void) fprintf(stdout, "JobFlagForMigration(%d) called.\n", hostID);
3251 	(void) fflush(stdout);
3252     }
3253     jnode = Lst_Find(jobs, (ClientData)&hostID, JobCmpRmtID);
3254 
3255     if (jnode == NILLNODE) {
3256 	jnode = Lst_Find(stoppedJobs, (ClientData)hostID, JobCmpRmtID);
3257 		if (jnode == NILLNODE) {
3258 		    if (DEBUG(JOB)) {
3259 			Error("Evicting host(%d) not in table", hostID);
3260 		    }
3261 		    return;
3262 		}
3263     }
3264     job = (Job *) Lst_Datum(jnode);
3265 
3266     if (DEBUG(JOB)) {
3267 	(void) fprintf(stdout,
3268 		       "JobFlagForMigration(%d) found job '%s'.\n", hostID,
3269 		       job->node->name);
3270 	(void) fflush(stdout);
3271     }
3272 
3273     KILL(job->pid, SIGSTOP);
3274 
3275     job->flags |= JOB_REMIGRATE;
3276 }
3277 
3278 #endif
3279 
3280 /*-
3281  *-----------------------------------------------------------------------
3282  * JobRestartJobs --
3283  *	Tries to restart stopped jobs if there are slots available.
3284  *	Note that this tries to restart them regardless of pending errors.
3285  *	It's not good to leave stopped jobs lying around!
3286  *
3287  * Results:
3288  *	None.
3289  *
3290  * Side Effects:
3291  *	Resumes(and possibly migrates) jobs.
3292  *
3293  *-----------------------------------------------------------------------
3294  */
3295 static void
3296 JobRestartJobs(void)
3297 {
3298     sigset_t	mask;
3299 
3300     JobSigLock(&mask);
3301     while (!Lst_IsEmpty(stoppedJobs)) {
3302 	if (DEBUG(JOB)) {
3303 	    (void) fprintf(stdout, "Restarting a stopped job.\n");
3304 	    (void) fflush(stdout);
3305 	}
3306 	if (JobRestart((Job *)Lst_DeQueue(stoppedJobs)) != 0)
3307 		break;
3308     }
3309     JobSigUnlock(&mask);
3310 }
3311 
3312 #ifndef RMT_WILL_WATCH
3313 static void
3314 watchfd(Job *job)
3315 {
3316     int i;
3317     if (job->inPollfd != NULL)
3318 	Punt("Watching watched job");
3319     if (fds == NULL) {
3320 	maxfds = JBSTART;
3321 	fds = emalloc(sizeof(struct pollfd) * maxfds);
3322 	jobfds = emalloc(sizeof(Job **) * maxfds);
3323 
3324 	fds[0].fd = job_pipe[0];
3325 	fds[0].events = POLLIN;
3326 	jobfds[0] = &tokenWaitJob;
3327 	tokenWaitJob.inPollfd = &fds[0];
3328 	nfds++;
3329 
3330 	fds[1].fd = exit_pipe[0];
3331 	fds[1].events = POLLIN;
3332 	jobfds[1] = &childExitJob;
3333 	childExitJob.inPollfd = &fds[1];
3334 	nfds++;
3335     } else if (nfds == maxfds) {
3336 	maxfds *= JBFACTOR;
3337 	fds = erealloc(fds, sizeof(struct pollfd) * maxfds);
3338 	jobfds = erealloc(jobfds, sizeof(Job **) * maxfds);
3339 	for (i = 0; i < nfds; i++)
3340 	    jobfds[i]->inPollfd = &fds[i];
3341     }
3342 
3343     fds[nfds].fd = job->inPipe;
3344     fds[nfds].events = POLLIN;
3345     jobfds[nfds] = job;
3346     job->inPollfd = &fds[nfds];
3347     nfds++;
3348 }
3349 
3350 static void
3351 clearfd(Job *job)
3352 {
3353     int i;
3354     if (job->inPollfd == NULL)
3355 	Punt("Unwatching unwatched job");
3356     i = job->inPollfd - fds;
3357     nfds--;
3358     /*
3359      * Move last job in table into hole made by dead job.
3360      */
3361     if (nfds != i) {
3362 	fds[i] = fds[nfds];
3363 	jobfds[i] = jobfds[nfds];
3364 	jobfds[i]->inPollfd = &fds[i];
3365     }
3366     job->inPollfd = NULL;
3367 }
3368 
3369 static int
3370 readyfd(Job *job)
3371 {
3372     if (job->inPollfd == NULL)
3373 	Punt("Polling unwatched job");
3374     return (job->inPollfd->revents & POLLIN) != 0;
3375 }
3376 #endif
3377 
3378 /*-
3379  *-----------------------------------------------------------------------
3380  * JobTokenAdd --
3381  *	Put a token into the job pipe so that some make process can start
3382  *	another job.
3383  *
3384  * Side Effects:
3385  *	Allows more build jobs to be spawned somewhere.
3386  *
3387  *-----------------------------------------------------------------------
3388  */
3389 
3390 static void
3391 JobTokenAdd(void)
3392 {
3393 
3394     if (DEBUG(JOB))
3395 	printf("deposit token\n");
3396     write(job_pipe[1], "+", 1);
3397 }
3398 
3399 /*-
3400  *-----------------------------------------------------------------------
3401  * Job_ServerStartTokenAdd --
3402  *	Prep the job token pipe in the root make process.
3403  *
3404  *-----------------------------------------------------------------------
3405  */
3406 
3407 void
3408 Job_ServerStart(int maxproc)
3409 {
3410     int i, flags;
3411     char jobarg[64];
3412 
3413     if (pipe(job_pipe) < 0)
3414 	Fatal ("error in pipe: %s", strerror(errno));
3415 
3416     /*
3417      * We mark the input side of the pipe non-blocking; we poll(2) the
3418      * pipe when we're waiting for a job token, but we might lose the
3419      * race for the token when a new one becomes available, so the read
3420      * from the pipe should not block.
3421      */
3422     flags = fcntl(job_pipe[0], F_GETFL, 0);
3423     flags |= O_NONBLOCK;
3424     fcntl(job_pipe[0], F_SETFL, flags);
3425 
3426     /*
3427      * Mark job pipes as close-on-exec.
3428      * Note that we will clear this when executing submakes.
3429      */
3430     fcntl(job_pipe[0], F_SETFD, 1);
3431     fcntl(job_pipe[1], F_SETFD, 1);
3432 
3433     snprintf(jobarg, sizeof(jobarg), "%d,%d", job_pipe[0], job_pipe[1]);
3434 
3435     Var_Append(MAKEFLAGS, "-J", VAR_GLOBAL);
3436     Var_Append(MAKEFLAGS, jobarg, VAR_GLOBAL);
3437 
3438     /*
3439      * Preload job_pipe with one token per job, save the one
3440      * "extra" token for the primary job.
3441      *
3442      * XXX should clip maxJobs against PIPE_BUF -- if maxJobs is
3443      * larger than the write buffer size of the pipe, we will
3444      * deadlock here.
3445      */
3446     for (i=1; i < maxproc; i++)
3447 	JobTokenAdd();
3448 }
3449 
3450 /*
3451  * this tracks the number of tokens currently "out" to build jobs.
3452  */
3453 int jobTokensRunning = 0;
3454 int jobTokensFree = 0;
3455 /*-
3456  *-----------------------------------------------------------------------
3457  * Job_TokenReturn --
3458  *	Return a withdrawn token to the pool.
3459  *
3460  *-----------------------------------------------------------------------
3461  */
3462 
3463 void
3464 Job_TokenReturn(void)
3465 {
3466     jobTokensRunning--;
3467     if (jobTokensRunning < 0)
3468 	Punt("token botch");
3469     if (jobTokensRunning)
3470 	jobTokensFree++;
3471 }
3472 
3473 /*-
3474  *-----------------------------------------------------------------------
3475  * Job_TokenWithdraw --
3476  *	Attempt to withdraw a token from the pool.
3477  *
3478  * Results:
3479  *	Returns TRUE if a token was withdrawn, and FALSE if the pool
3480  *	is currently empty.
3481  *
3482  * Side Effects:
3483  * 	If pool is empty, set wantToken so that we wake up
3484  *	when a token is released.
3485  *
3486  *-----------------------------------------------------------------------
3487  */
3488 
3489 
3490 Boolean
3491 Job_TokenWithdraw(void)
3492 {
3493     char tok;
3494     int count;
3495 
3496     wantToken = FALSE;
3497 
3498     if (aborting)
3499 	    return FALSE;
3500 
3501     if (jobTokensRunning == 0) {
3502 	if (DEBUG(JOB))
3503 	    printf("first one's free\n");
3504 	jobTokensRunning++;
3505 	return TRUE;
3506     }
3507     if (jobTokensFree > 0) {
3508 	jobTokensFree--;
3509 	jobTokensRunning++;
3510 	return TRUE;
3511     }
3512     count = read(job_pipe[0], &tok, 1);
3513     if (count == 0)
3514 	Fatal("eof on job pipe!");
3515     else if (count < 0) {
3516 	if (errno != EAGAIN) {
3517 	    Fatal("job pipe read: %s", strerror(errno));
3518 	}
3519 	if (DEBUG(JOB))
3520 	    printf("blocked for token\n");
3521 	wantToken = TRUE;
3522 	return FALSE;
3523     }
3524     jobTokensRunning++;
3525     if (DEBUG(JOB))
3526 	printf("withdrew token\n");
3527     return TRUE;
3528 }
3529 
3530 /*-
3531  *-----------------------------------------------------------------------
3532  * Job_TokenFlush --
3533  *	Return free tokens to the pool.
3534  *
3535  *-----------------------------------------------------------------------
3536  */
3537 
3538 void
3539 Job_TokenFlush(void)
3540 {
3541     if (compatMake) return;
3542 
3543     while (jobTokensFree > 0) {
3544 	JobTokenAdd();
3545 	jobTokensFree--;
3546     }
3547 }
3548 
3549 #ifdef USE_SELECT
3550 int
3551 emul_poll(struct pollfd *fd, int nfd, int timeout)
3552 {
3553     fd_set rfds, wfds;
3554     int i, maxfd, nselect, npoll;
3555     struct timeval tv, *tvp;
3556     long usecs;
3557 
3558     FD_ZERO(&rfds);
3559     FD_ZERO(&wfds);
3560 
3561     maxfd = -1;
3562     for (i = 0; i < nfd; i++) {
3563 	fd[i].revents = 0;
3564 
3565 	if (fd[i].events & POLLIN)
3566 	    FD_SET(fd[i].fd, &rfds);
3567 
3568 	if (fd[i].events & POLLOUT)
3569 	    FD_SET(fd[i].fd, &wfds);
3570 
3571 	if (fd[i].fd > maxfd)
3572 	    maxfd = fd[i].fd;
3573     }
3574 
3575     if (maxfd >= FD_SETSIZE) {
3576 	Punt("Ran out of fd_set slots; "
3577 	     "recompile with a larger FD_SETSIZE.");
3578     }
3579 
3580     if (timeout < 0) {
3581 	tvp = NULL;
3582     } else {
3583 	usecs = timeout * 1000;
3584 	tv.tv_sec = usecs / 1000000;
3585 	tv.tv_usec = usecs % 1000000;
3586         tvp = &tv;
3587     }
3588 
3589     nselect = select(maxfd + 1, &rfds, &wfds, 0, tvp);
3590 
3591     if (nselect <= 0)
3592 	return nselect;
3593 
3594     npoll = 0;
3595     for (i = 0; i < nfd; i++) {
3596 	if (FD_ISSET(fd[i].fd, &rfds))
3597 	    fd[i].revents |= POLLIN;
3598 
3599 	if (FD_ISSET(fd[i].fd, &wfds))
3600 	    fd[i].revents |= POLLOUT;
3601 
3602 	if (fd[i].revents)
3603 	    npoll++;
3604     }
3605 
3606     return npoll;
3607 }
3608 #endif /* USE_SELECT */
3609