xref: /netbsd-src/usr.bin/make/job.c (revision 23c8222edbfb0f0932d88a8351d3a0cf817dfb9e)
1 /*	$NetBSD: job.c,v 1.87 2004/07/01 20:38:09 jmc Exp $	*/
2 
3 /*
4  * Copyright (c) 1988, 1989, 1990 The Regents of the University of California.
5  * All rights reserved.
6  *
7  * This code is derived from software contributed to Berkeley by
8  * Adam de Boor.
9  *
10  * Redistribution and use in source and binary forms, with or without
11  * modification, are permitted provided that the following conditions
12  * are met:
13  * 1. Redistributions of source code must retain the above copyright
14  *    notice, this list of conditions and the following disclaimer.
15  * 2. Redistributions in binary form must reproduce the above copyright
16  *    notice, this list of conditions and the following disclaimer in the
17  *    documentation and/or other materials provided with the distribution.
18  * 3. Neither the name of the University nor the names of its contributors
19  *    may be used to endorse or promote products derived from this software
20  *    without specific prior written permission.
21  *
22  * THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND
23  * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
24  * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
25  * ARE DISCLAIMED.  IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE
26  * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
27  * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
28  * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
29  * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
30  * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
31  * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
32  * SUCH DAMAGE.
33  */
34 
35 /*
36  * Copyright (c) 1988, 1989 by Adam de Boor
37  * Copyright (c) 1989 by Berkeley Softworks
38  * All rights reserved.
39  *
40  * This code is derived from software contributed to Berkeley by
41  * Adam de Boor.
42  *
43  * Redistribution and use in source and binary forms, with or without
44  * modification, are permitted provided that the following conditions
45  * are met:
46  * 1. Redistributions of source code must retain the above copyright
47  *    notice, this list of conditions and the following disclaimer.
48  * 2. Redistributions in binary form must reproduce the above copyright
49  *    notice, this list of conditions and the following disclaimer in the
50  *    documentation and/or other materials provided with the distribution.
51  * 3. All advertising materials mentioning features or use of this software
52  *    must display the following acknowledgement:
53  *	This product includes software developed by the University of
54  *	California, Berkeley and its contributors.
55  * 4. Neither the name of the University nor the names of its contributors
56  *    may be used to endorse or promote products derived from this software
57  *    without specific prior written permission.
58  *
59  * THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND
60  * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
61  * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
62  * ARE DISCLAIMED.  IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE
63  * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
64  * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
65  * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
66  * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
67  * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
68  * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
69  * SUCH DAMAGE.
70  */
71 
72 #ifndef MAKE_NATIVE
73 static char rcsid[] = "$NetBSD: job.c,v 1.87 2004/07/01 20:38:09 jmc Exp $";
74 #else
75 #include <sys/cdefs.h>
76 #ifndef lint
77 #if 0
78 static char sccsid[] = "@(#)job.c	8.2 (Berkeley) 3/19/94";
79 #else
80 __RCSID("$NetBSD: job.c,v 1.87 2004/07/01 20:38:09 jmc Exp $");
81 #endif
82 #endif /* not lint */
83 #endif
84 
85 /*-
86  * job.c --
87  *	handle the creation etc. of our child processes.
88  *
89  * Interface:
90  *	Job_Make  	    	Start the creation of the given target.
91  *
92  *	Job_CatchChildren   	Check for and handle the termination of any
93  *	    	  	    	children. This must be called reasonably
94  *	    	  	    	frequently to keep the whole make going at
95  *	    	  	    	a decent clip, since job table entries aren't
96  *	    	  	    	removed until their process is caught this way.
97  *	    	  	    	Its single argument is TRUE if the function
98  *	    	  	    	should block waiting for a child to terminate.
99  *
100  *	Job_CatchOutput	    	Print any output our children have produced.
101  *	    	  	    	Should also be called fairly frequently to
102  *	    	  	    	keep the user informed of what's going on.
103  *	    	  	    	If no output is waiting, it will block for
104  *	    	  	    	a time given by the SEL_* constants, below,
105  *	    	  	    	or until output is ready.
106  *
107  *	Job_Init  	    	Called to intialize this module. in addition,
108  *	    	  	    	any commands attached to the .BEGIN target
109  *	    	  	    	are executed before this function returns.
110  *	    	  	    	Hence, the makefile must have been parsed
111  *	    	  	    	before this function is called.
112  *
113  *	Job_End  	    	Cleanup any memory used.
114  *
115  *	Job_Empty 	    	Return TRUE if the job table is completely
116  *	    	  	    	empty.
117  *
118  *	Job_ParseShell	    	Given the line following a .SHELL target, parse
119  *	    	  	    	the line as a shell specification. Returns
120  *	    	  	    	FAILURE if the spec was incorrect.
121  *
122  *	Job_Finish	    	Perform any final processing which needs doing.
123  *	    	  	    	This includes the execution of any commands
124  *	    	  	    	which have been/were attached to the .END
125  *	    	  	    	target. It should only be called when the
126  *	    	  	    	job table is empty.
127  *
128  *	Job_AbortAll	    	Abort all currently running jobs. It doesn't
129  *	    	  	    	handle output or do anything for the jobs,
130  *	    	  	    	just kills them. It should only be called in
131  *	    	  	    	an emergency, as it were.
132  *
133  *	Job_CheckCommands   	Verify that the commands for a target are
134  *	    	  	    	ok. Provide them if necessary and possible.
135  *
136  *	Job_Touch 	    	Update a target without really updating it.
137  *
138  *	Job_Wait  	    	Wait for all currently-running jobs to finish.
139  */
140 
141 #include <sys/types.h>
142 #include <sys/stat.h>
143 #include <sys/file.h>
144 #include <sys/time.h>
145 #include <sys/wait.h>
146 
147 #include <errno.h>
148 #include <fcntl.h>
149 #ifndef RMT_WILL_WATCH
150 #ifndef USE_SELECT
151 #include <poll.h>
152 #endif
153 #endif
154 #include <signal.h>
155 #include <stdio.h>
156 #include <string.h>
157 #include <utime.h>
158 
159 #include "make.h"
160 #include "hash.h"
161 #include "dir.h"
162 #include "job.h"
163 #include "pathnames.h"
164 #include "trace.h"
165 #ifdef REMOTE
166 #include "rmt.h"
167 # define STATIC
168 #else
169 # define STATIC static
170 #endif
171 
172 /*
173  * error handling variables
174  */
175 static int     	errors = 0;	    /* number of errors reported */
176 static int    	aborting = 0;	    /* why is the make aborting? */
177 #define ABORT_ERROR	1   	    /* Because of an error */
178 #define ABORT_INTERRUPT	2   	    /* Because it was interrupted */
179 #define ABORT_WAIT	3   	    /* Waiting for jobs to finish */
180 
181 /*
182  * XXX: Avoid SunOS bug... FILENO() is fp->_file, and file
183  * is a char! So when we go above 127 we turn negative!
184  */
185 #define FILENO(a) ((unsigned) fileno(a))
186 
187 /*
188  * post-make command processing. The node postCommands is really just the
189  * .END target but we keep it around to avoid having to search for it
190  * all the time.
191  */
192 static GNode   	  *postCommands = NILGNODE;
193 				    /* node containing commands to execute when
194 				     * everything else is done */
195 static int     	  numCommands; 	    /* The number of commands actually printed
196 				     * for a target. Should this number be
197 				     * 0, no shell will be executed. */
198 
199 /*
200  * Return values from JobStart.
201  */
202 #define JOB_RUNNING	0   	/* Job is running */
203 #define JOB_ERROR 	1   	/* Error in starting the job */
204 #define JOB_FINISHED	2   	/* The job is already finished */
205 #define JOB_STOPPED	3   	/* The job is stopped */
206 
207 
208 
209 /*
210  * Descriptions for various shells.
211  */
212 static Shell    shells[] = {
213     /*
214      * CSH description. The csh can do echo control by playing
215      * with the setting of the 'echo' shell variable. Sadly,
216      * however, it is unable to do error control nicely.
217      */
218 {
219     "csh",
220     TRUE, "unset verbose", "set verbose", "unset verbose", 10,
221     FALSE, "echo \"%s\"\n", "csh -c \"%s || exit 0\"", "", '#',
222     "v", "e",
223 },
224     /*
225      * SH description. Echo control is also possible and, under
226      * sun UNIX anyway, one can even control error checking.
227      */
228 {
229     "sh",
230     FALSE, "", "", "", 0,
231     FALSE, "echo \"%s\"\n", "%s\n", "{ %s \n} || exit $?\n", '#',
232 #ifdef __NetBSD__
233     "q",
234 #else
235     "",
236 #endif
237     "",
238 },
239     /*
240      * KSH description.
241      */
242 {
243     "ksh",
244     TRUE, "set +v", "set -v", "set +v", 6,
245     FALSE, "echo \"%s\"\n", "%s\n", "{ %s \n} || exit $?\n", '#',
246     "v",
247     "",
248 },
249     /*
250      * UNKNOWN.
251      */
252 {
253     (char *) 0,
254     FALSE, (char *) 0, (char *) 0, (char *) 0, 0,
255     FALSE, (char *) 0, (char *) 0, (char *) 0, 0,
256     (char *) 0, (char *) 0,
257 }
258 };
259 static Shell 	*commandShell = &shells[DEFSHELL];/* this is the shell to
260 						   * which we pass all
261 						   * commands in the Makefile.
262 						   * It is set by the
263 						   * Job_ParseShell function */
264 const char *shellPath = NULL,		  	  /* full pathname of
265 						   * executable image */
266            *shellName = NULL;		      	  /* last component of shell */
267 static const char *shellArgv = NULL;		  /* Custom shell args */
268 
269 
270 static int  	maxJobs;    	/* The most children we can run at once */
271 static int  	maxLocal;    	/* The most local ones we can have */
272 STATIC int     	nJobs;	    	/* The number of children currently running */
273 STATIC int	nLocal;    	/* The number of local children */
274 STATIC Lst     	jobs;		/* The structures that describe them */
275 static Boolean	wantToken;	/* we want a token */
276 
277 /*
278  * Set of descriptors of pipes connected to
279  * the output channels of children
280  */
281 #ifndef RMT_WILL_WATCH
282 static struct pollfd *fds = NULL;
283 static Job **jobfds = NULL;
284 static int nfds = 0;
285 static int maxfds = 0;
286 static void watchfd(Job *);
287 static void clearfd(Job *);
288 static int readyfd(Job *);
289 #define JBSTART 256
290 #define JBFACTOR 2
291 #endif
292 
293 STATIC GNode   	*lastNode;	/* The node for which output was most recently
294 				 * produced. */
295 STATIC const char *targFmt;   	/* Format string to use to head output from a
296 				 * job when it's not the most-recent job heard
297 				 * from */
298 static Job tokenWaitJob;	/* token wait pseudo-job */
299 int	job_pipe[2] = { -1, -1 }; /* job server pipes. */
300 
301 static Job childExitJob;	/* child exit pseudo-job */
302 int	exit_pipe[2] = { -1, -1 }; /* child exit signal pipe. */
303 
304 #ifdef REMOTE
305 # define TARG_FMT  "--- %s at %s ---\n" /* Default format */
306 # define MESSAGE(fp, gn) \
307 	(void) fprintf(fp, targFmt, gn->name, gn->rem.hname)
308 #else
309 # define TARG_FMT  "--- %s ---\n" /* Default format */
310 # define MESSAGE(fp, gn) \
311 	(void) fprintf(fp, targFmt, gn->name)
312 #endif
313 
314 /*
315  * When JobStart attempts to run a job remotely but can't, and isn't allowed
316  * to run the job locally, or when Job_CatchChildren detects a job that has
317  * been migrated home, the job is placed on the stoppedJobs queue to be run
318  * when the next job finishes.
319  */
320 STATIC Lst	stoppedJobs;	/* Lst of Job structures describing
321 				 * jobs that were stopped due to concurrency
322 				 * limits or migration home */
323 
324 
325 sigset_t	caught_signals;	/* Set of signals we handle */
326 #if defined(USE_PGRP) && defined(SYSV)
327 # define KILL(pid, sig)		kill(-(pid), (sig))
328 #else
329 # if defined(USE_PGRP)
330 #  define KILL(pid, sig)	killpg((pid), (sig))
331 # else
332 #  define KILL(pid, sig)	kill((pid), (sig))
333 # endif
334 #endif
335 
336 /*
337  * Grmpf... There is no way to set bits of the wait structure
338  * anymore with the stupid W*() macros. I liked the union wait
339  * stuff much more. So, we devise our own macros... This is
340  * really ugly, use dramamine sparingly. You have been warned.
341  */
342 #ifndef W_STOPCODE
343 #define W_STOPCODE(sig) (((sig) << 8) | 0177)
344 #endif
345 #ifndef W_EXITCODE
346 #define W_EXITCODE(ret, sig) ((ret << 8) | (sig))
347 #endif
348 
349 static int JobCondPassSig(ClientData, ClientData);
350 static void JobPassSig(int);
351 static void JobChildSig(int);
352 #ifdef USE_PGRP
353 static void JobContinueSig(int);
354 #endif
355 static int JobCmpPid(ClientData, ClientData);
356 static int JobPrintCommand(ClientData, ClientData);
357 static int JobSaveCommand(ClientData, ClientData);
358 static void JobClose(Job *);
359 #ifdef REMOTE
360 static int JobCmpRmtID(ClientData, ClientData);
361 # ifdef RMT_WILL_WATCH
362 static void JobLocalInput(int, Job *);
363 # endif
364 #else
365 static void JobFinish(Job *, int *);
366 static void JobExec(Job *, char **);
367 #endif
368 static void JobMakeArgv(Job *, char **);
369 static int JobRestart(Job *);
370 static int JobStart(GNode *, int, Job *);
371 static char *JobOutput(Job *, char *, char *, int);
372 static void JobDoOutput(Job *, Boolean);
373 static Shell *JobMatchShell(const char *);
374 static void JobInterrupt(int, int);
375 static void JobRestartJobs(void);
376 static void JobTokenAdd(void);
377 static void JobSigLock(sigset_t *);
378 static void JobSigUnlock(sigset_t *);
379 static void JobSigReset(void);
380 
381 
382 
383 /*
384  * JobSigLock/JobSigUnlock
385  *
386  * Signal lock routines to get exclusive access. Currently used to
387  * protect `jobs' and `stoppedJobs' list manipulations.
388  */
389 static void JobSigLock(sigset_t *omaskp)
390 {
391 	if (sigprocmask(SIG_BLOCK, &caught_signals, omaskp) != 0) {
392 		Punt("JobSigLock: sigprocmask: %s", strerror(errno));
393 	sigemptyset(omaskp);
394 	}
395 }
396 
397 static void JobSigUnlock(sigset_t *omaskp)
398 {
399 	(void) sigprocmask(SIG_SETMASK, omaskp, NULL);
400 }
401 
402 /*-
403  *-----------------------------------------------------------------------
404  * JobCondPassSig --
405  *	Pass a signal to a job if the job is remote or if USE_PGRP
406  *	is defined.
407  *
408  * Input:
409  *	jobp		Job to biff
410  *	signop		Signal to send it
411  *
412  * Results:
413  *	=== 0
414  *
415  * Side Effects:
416  *	None, except the job may bite it.
417  *
418  *-----------------------------------------------------------------------
419  */
420 static int
421 JobCondPassSig(ClientData jobp, ClientData signop)
422 {
423     Job	*job = (Job *) jobp;
424     int	signo = *(int *) signop;
425 #ifdef RMT_WANTS_SIGNALS
426     if (job->flags & JOB_REMOTE) {
427 	(void) Rmt_Signal(job, signo);
428     } else {
429 	KILL(job->pid, signo);
430     }
431 #else
432     /*
433      * Assume that sending the signal to job->pid will signal any remote
434      * job as well.
435      */
436     if (DEBUG(JOB)) {
437 	(void) fprintf(stdout,
438 		       "JobCondPassSig passing signal %d to child %d.\n",
439 		       signo, job->pid);
440 	(void) fflush(stdout);
441     }
442     KILL(job->pid, signo);
443 #endif
444     return 0;
445 }
446 
447 /*-
448  *-----------------------------------------------------------------------
449  * JobChldSig --
450  *	SIGCHLD handler.
451  *
452  * Input:
453  *	signo		The signal number we've received
454  *
455  * Results:
456  *	None.
457  *
458  * Side Effects:
459  *	Sends a token on the child exit pipe to wake us up from
460  *	select()/poll().
461  *
462  *-----------------------------------------------------------------------
463  */
464 static void
465 JobChildSig(int signo __unused)
466 {
467     write(exit_pipe[1], ".", 1);
468 }
469 
470 
471 #ifdef USE_PGRP
472 /*-
473  *-----------------------------------------------------------------------
474  * JobContinueSig --
475  *	Resume all stopped jobs.
476  *
477  * Input:
478  *	signo		The signal number we've received
479  *
480  * Results:
481  *	None.
482  *
483  * Side Effects:
484  *	Jobs start running again.
485  *
486  *-----------------------------------------------------------------------
487  */
488 static void
489 JobContinueSig(int signo __unused)
490 {
491     JobRestartJobs();
492 }
493 #endif
494 
495 /*-
496  *-----------------------------------------------------------------------
497  * JobPassSig --
498  *	Pass a signal on to all remote jobs and to all local jobs if
499  *	USE_PGRP is defined, then die ourselves.
500  *
501  * Input:
502  *	signo		The signal number we've received
503  *
504  * Results:
505  *	None.
506  *
507  * Side Effects:
508  *	We die by the same signal.
509  *
510  *-----------------------------------------------------------------------
511  */
512 static void
513 JobPassSig(int signo)
514 {
515     sigset_t nmask, omask;
516     struct sigaction act;
517     int sigcont;
518 
519     if (DEBUG(JOB)) {
520 	(void) fprintf(stdout, "JobPassSig(%d) called.\n", signo);
521 	(void) fflush(stdout);
522     }
523     Lst_ForEach(jobs, JobCondPassSig, (ClientData) &signo);
524 
525     /*
526      * Deal with proper cleanup based on the signal received. We only run
527      * the .INTERRUPT target if the signal was in fact an interrupt. The other
528      * three termination signals are more of a "get out *now*" command.
529      */
530     if (signo == SIGINT) {
531 	JobInterrupt(TRUE, signo);
532     } else if ((signo == SIGHUP) || (signo == SIGTERM) || (signo == SIGQUIT)) {
533 	JobInterrupt(FALSE, signo);
534     }
535 
536     /*
537      * Leave gracefully if SIGQUIT, rather than core dumping.
538      */
539     if (signo == SIGQUIT) {
540 	Finish(0);
541     }
542 
543     if (signo == SIGTSTP) {
544 	Job_CatchChildren(FALSE);
545     }
546     /*
547      * Send ourselves the signal now we've given the message to everyone else.
548      * Note we block everything else possible while we're getting the signal.
549      * This ensures that all our jobs get continued when we wake up before
550      * we take any other signal.
551      */
552     sigfillset(&nmask);
553     sigdelset(&nmask, signo);
554     (void) sigprocmask(SIG_SETMASK, &nmask, &omask);
555 
556     act.sa_handler = SIG_DFL;
557     sigemptyset(&act.sa_mask);
558     act.sa_flags = 0;
559     (void) sigaction(signo, &act, NULL);
560 
561     if (DEBUG(JOB)) {
562 	(void) fprintf(stdout,
563 		       "JobPassSig passing signal %d to self.\n", signo);
564 	(void) fflush(stdout);
565     }
566 
567     (void) kill(getpid(), signo);
568     if (signo != SIGTSTP) {
569 	sigcont = SIGCONT;
570 	Lst_ForEach(jobs, JobCondPassSig, (ClientData) &sigcont);
571     }
572 
573     /* Restore handler and signal mask */
574     act.sa_handler = JobPassSig;
575     (void) sigaction(signo, &act, NULL);
576     (void) sigprocmask(SIG_SETMASK, &omask, NULL);
577 }
578 
579 /*-
580  *-----------------------------------------------------------------------
581  * JobCmpPid  --
582  *	Compare the pid of the job with the given pid and return 0 if they
583  *	are equal. This function is called from Job_CatchChildren via
584  *	Lst_Find to find the job descriptor of the finished job.
585  *
586  * Input:
587  *	job		job to examine
588  *	pid		process id desired
589  *
590  * Results:
591  *	0 if the pid's match
592  *
593  * Side Effects:
594  *	None
595  *-----------------------------------------------------------------------
596  */
597 static int
598 JobCmpPid(ClientData job, ClientData pid)
599 {
600     return *(int *) pid - ((Job *) job)->pid;
601 }
602 
603 #ifdef REMOTE
604 /*-
605  *-----------------------------------------------------------------------
606  * JobCmpRmtID  --
607  *	Compare the rmtID of the job with the given rmtID and return 0 if they
608  *	are equal.
609  *
610  * Input:
611  *	job		job to examine
612  *	rmtID		remote id desired
613  *
614  * Results:
615  *	0 if the rmtID's match
616  *
617  * Side Effects:
618  *	None.
619  *-----------------------------------------------------------------------
620  */
621 static int
622 JobCmpRmtID(ClientData job, ClientData rmtID)
623 {
624     return(*(int *) rmtID - ((Job *) job)->rmtID);
625 }
626 #endif
627 
628 /*-
629  *-----------------------------------------------------------------------
630  * JobPrintCommand  --
631  *	Put out another command for the given job. If the command starts
632  *	with an @ or a - we process it specially. In the former case,
633  *	so long as the -s and -n flags weren't given to make, we stick
634  *	a shell-specific echoOff command in the script. In the latter,
635  *	we ignore errors for the entire job, unless the shell has error
636  *	control.
637  *	If the command is just "..." we take all future commands for this
638  *	job to be commands to be executed once the entire graph has been
639  *	made and return non-zero to signal that the end of the commands
640  *	was reached. These commands are later attached to the postCommands
641  *	node and executed by Job_End when all things are done.
642  *	This function is called from JobStart via Lst_ForEach.
643  *
644  * Input:
645  *	cmdp		command string to print
646  *	jobp		job for which to print it
647  *
648  * Results:
649  *	Always 0, unless the command was "..."
650  *
651  * Side Effects:
652  *	If the command begins with a '-' and the shell has no error control,
653  *	the JOB_IGNERR flag is set in the job descriptor.
654  *	If the command is "..." and we're not ignoring such things,
655  *	tailCmds is set to the successor node of the cmd.
656  *	numCommands is incremented if the command is actually printed.
657  *-----------------------------------------------------------------------
658  */
659 static int
660 JobPrintCommand(ClientData cmdp, ClientData jobp)
661 {
662     Boolean	  noSpecials;	    /* true if we shouldn't worry about
663 				     * inserting special commands into
664 				     * the input stream. */
665     Boolean       shutUp = FALSE;   /* true if we put a no echo command
666 				     * into the command file */
667     Boolean	  errOff = FALSE;   /* true if we turned error checking
668 				     * off before printing the command
669 				     * and need to turn it back on */
670     const char    *cmdTemplate;	    /* Template to use when printing the
671 				     * command */
672     char    	  *cmdStart;	    /* Start of expanded command */
673     char	  *escCmd = NULL;    /* Command with quotes/backticks escaped */
674     char     	  *cmd = (char *) cmdp;
675     Job           *job = (Job *) jobp;
676     char	  *cp;
677     int           i, j;
678 
679     noSpecials = NoExecute(job->node);
680 
681     if (strcmp(cmd, "...") == 0) {
682 	job->node->type |= OP_SAVE_CMDS;
683 	if ((job->flags & JOB_IGNDOTS) == 0) {
684 	    job->tailCmds = Lst_Succ(Lst_Member(job->node->commands,
685 						(ClientData)cmd));
686 	    return 1;
687 	}
688 	return 0;
689     }
690 
691 #define DBPRINTF(fmt, arg) if (DEBUG(JOB)) {	\
692 	(void) fprintf(stdout, fmt, arg); 	\
693 	(void) fflush(stdout); 			\
694     }						\
695    (void) fprintf(job->cmdFILE, fmt, arg);	\
696    (void) fflush(job->cmdFILE);
697 
698     numCommands += 1;
699 
700     cmdStart = cmd = Var_Subst(NULL, cmd, job->node, FALSE);
701 
702     cmdTemplate = "%s\n";
703 
704     /*
705      * Check for leading @' and -'s to control echoing and error checking.
706      */
707     while (*cmd == '@' || *cmd == '-' || (*cmd == '+')) {
708 	switch (*cmd) {
709 	case '@':
710 	    shutUp = TRUE;
711 	    break;
712 	case '-':
713 	    errOff = TRUE;
714 	    break;
715 	case '+':
716 	    if (noSpecials) {
717 		/*
718 		 * We're not actually executing anything...
719 		 * but this one needs to be - use compat mode just for it.
720 		 */
721 		CompatRunCommand(cmdp, (ClientData)job->node);
722 		return 0;
723 	    }
724 	    break;
725 	}
726 	cmd++;
727     }
728 
729     while (isspace((unsigned char) *cmd))
730 	cmd++;
731 
732     /*
733      * If the shell doesn't have error control the alternate echo'ing will
734      * be done (to avoid showing additional error checking code)
735      * and this will need the characters '$ ` \ "' escaped
736      */
737 
738     if (!commandShell->hasErrCtl) {
739 	/* Worst that could happen is every char needs escaping. */
740 	escCmd = (char *) emalloc((strlen(cmd) * 2) + 1);
741 	for (i = 0, j= 0; cmd[i] != '\0'; i++, j++) {
742 		if (cmd[i] == '$' || cmd[i] == '`' || cmd[i] == '\\' ||
743 			cmd[i] == '"')
744 			escCmd[j++] = '\\';
745 		escCmd[j] = cmd[i];
746 	}
747 	escCmd[j] = 0;
748     }
749 
750     if (shutUp) {
751 	if (!(job->flags & JOB_SILENT) && !noSpecials &&
752 	    commandShell->hasEchoCtl) {
753 		DBPRINTF("%s\n", commandShell->echoOff);
754 	} else {
755 	    if (commandShell->hasErrCtl)
756 		shutUp = FALSE;
757 	}
758     }
759 
760     if (errOff) {
761 	if ( !(job->flags & JOB_IGNERR) && !noSpecials) {
762 	    if (commandShell->hasErrCtl) {
763 		/*
764 		 * we don't want the error-control commands showing
765 		 * up either, so we turn off echoing while executing
766 		 * them. We could put another field in the shell
767 		 * structure to tell JobDoOutput to look for this
768 		 * string too, but why make it any more complex than
769 		 * it already is?
770 		 */
771 		if (!(job->flags & JOB_SILENT) && !shutUp &&
772 		    commandShell->hasEchoCtl) {
773 			DBPRINTF("%s\n", commandShell->echoOff);
774 			DBPRINTF("%s\n", commandShell->ignErr);
775 			DBPRINTF("%s\n", commandShell->echoOn);
776 		} else {
777 			DBPRINTF("%s\n", commandShell->ignErr);
778 		}
779 	    } else if (commandShell->ignErr &&
780 		      (*commandShell->ignErr != '\0'))
781 	    {
782 		/*
783 		 * The shell has no error control, so we need to be
784 		 * weird to get it to ignore any errors from the command.
785 		 * If echoing is turned on, we turn it off and use the
786 		 * errCheck template to echo the command. Leave echoing
787 		 * off so the user doesn't see the weirdness we go through
788 		 * to ignore errors. Set cmdTemplate to use the weirdness
789 		 * instead of the simple "%s\n" template.
790 		 */
791 		if (!(job->flags & JOB_SILENT) && !shutUp) {
792 			if (commandShell->hasEchoCtl) {
793 				DBPRINTF("%s\n", commandShell->echoOff);
794 			}
795 			DBPRINTF(commandShell->errCheck, escCmd);
796 			shutUp = TRUE;
797 		} else {
798 			if (!shutUp) {
799 				DBPRINTF(commandShell->errCheck, escCmd);
800 			}
801 		}
802 		cmdTemplate = commandShell->ignErr;
803 		/*
804 		 * The error ignoration (hee hee) is already taken care
805 		 * of by the ignErr template, so pretend error checking
806 		 * is still on.
807 		 */
808 		errOff = FALSE;
809 	    } else {
810 		errOff = FALSE;
811 	    }
812 	} else {
813 	    errOff = FALSE;
814 	}
815     } else {
816 
817 	/*
818 	 * If errors are being checked and the shell doesn't have error control
819 	 * but does supply an errOut template, then setup commands to run
820 	 * through it.
821 	 */
822 
823 	if (!commandShell->hasErrCtl && commandShell->errOut &&
824 	    (*commandShell->errOut != '\0')) {
825 		if (!(job->flags & JOB_SILENT) && !shutUp) {
826 			if (commandShell->hasEchoCtl) {
827 				DBPRINTF("%s\n", commandShell->echoOff);
828 			}
829 			DBPRINTF(commandShell->errCheck, escCmd);
830 			shutUp = TRUE;
831 		}
832 		/* If it's a comment line, treat it like an ignored error */
833 		if (escCmd[0] == commandShell->commentChar)
834 			cmdTemplate = commandShell->ignErr;
835 		else
836 			cmdTemplate = commandShell->errOut;
837 		errOff = FALSE;
838 	}
839     }
840 
841     if (DEBUG(SHELL) && strcmp(shellName, "sh") == 0 &&
842 	(job->flags & JOB_TRACED) == 0) {
843 	    DBPRINTF("set -%s\n", "x");
844 	    job->flags |= JOB_TRACED;
845     }
846 
847     if ((cp = Check_Cwd_Cmd(cmd)) != NULL) {
848 	    DBPRINTF("test -d %s && ", cp);
849 	    DBPRINTF("cd %s; ", cp);
850     }
851     DBPRINTF(cmdTemplate, cmd);
852     free(cmdStart);
853     if (escCmd)
854         free(escCmd);
855     if (errOff) {
856 	/*
857 	 * If echoing is already off, there's no point in issuing the
858 	 * echoOff command. Otherwise we issue it and pretend it was on
859 	 * for the whole command...
860 	 */
861 	if (!shutUp && !(job->flags & JOB_SILENT) && commandShell->hasEchoCtl){
862 	    DBPRINTF("%s\n", commandShell->echoOff);
863 	    shutUp = TRUE;
864 	}
865 	DBPRINTF("%s\n", commandShell->errCheck);
866     }
867     if (shutUp && commandShell->hasEchoCtl) {
868 	DBPRINTF("%s\n", commandShell->echoOn);
869     }
870     return 0;
871 }
872 
873 /*-
874  *-----------------------------------------------------------------------
875  * JobSaveCommand --
876  *	Save a command to be executed when everything else is done.
877  *	Callback function for JobFinish...
878  *
879  * Results:
880  *	Always returns 0
881  *
882  * Side Effects:
883  *	The command is tacked onto the end of postCommands's commands list.
884  *
885  *-----------------------------------------------------------------------
886  */
887 static int
888 JobSaveCommand(ClientData cmd, ClientData gn)
889 {
890     cmd = (ClientData) Var_Subst(NULL, (char *) cmd, (GNode *) gn, FALSE);
891     (void) Lst_AtEnd(postCommands->commands, cmd);
892     return(0);
893 }
894 
895 
896 /*-
897  *-----------------------------------------------------------------------
898  * JobClose --
899  *	Called to close both input and output pipes when a job is finished.
900  *
901  * Results:
902  *	Nada
903  *
904  * Side Effects:
905  *	The file descriptors associated with the job are closed.
906  *
907  *-----------------------------------------------------------------------
908  */
909 static void
910 JobClose(Job *job)
911 {
912     if (usePipes && (job->flags & JOB_FIRST)) {
913 #ifdef RMT_WILL_WATCH
914 	Rmt_Ignore(job->inPipe);
915 #else
916 	clearfd(job);
917 #endif
918 	if (job->outPipe != job->inPipe) {
919 	   (void) close(job->outPipe);
920 	}
921 	JobDoOutput(job, TRUE);
922 	(void) close(job->inPipe);
923     } else {
924 	(void) close(job->outFd);
925 	JobDoOutput(job, TRUE);
926     }
927 }
928 
929 /*-
930  *-----------------------------------------------------------------------
931  * JobFinish  --
932  *	Do final processing for the given job including updating
933  *	parents and starting new jobs as available/necessary. Note
934  *	that we pay no attention to the JOB_IGNERR flag here.
935  *	This is because when we're called because of a noexecute flag
936  *	or something, jstat.w_status is 0 and when called from
937  *	Job_CatchChildren, the status is zeroed if it s/b ignored.
938  *
939  * Input:
940  *	job		job to finish
941  *	status		sub-why job went away
942  *
943  * Results:
944  *	None
945  *
946  * Side Effects:
947  *	Some nodes may be put on the toBeMade queue.
948  *	Final commands for the job are placed on postCommands.
949  *
950  *	If we got an error and are aborting (aborting == ABORT_ERROR) and
951  *	the job list is now empty, we are done for the day.
952  *	If we recognized an error (errors !=0), we set the aborting flag
953  *	to ABORT_ERROR so no more jobs will be started.
954  *-----------------------------------------------------------------------
955  */
956 /*ARGSUSED*/
957 static void
958 JobFinish(Job *job, int *status)
959 {
960     Boolean 	 done;
961 
962     if ((WIFEXITED(*status) &&
963 	 (((WEXITSTATUS(*status) != 0) && !(job->flags & JOB_IGNERR)))) ||
964 	WIFSIGNALED(*status))
965     {
966 	/*
967 	 * If it exited non-zero and either we're doing things our
968 	 * way or we're not ignoring errors, the job is finished.
969 	 * Similarly, if the shell died because of a signal
970 	 * the job is also finished. In these
971 	 * cases, finish out the job's output before printing the exit
972 	 * status...
973 	 */
974 #ifdef REMOTE
975 	KILL(job->pid, SIGCONT);
976 #endif
977 	JobClose(job);
978 	if (job->cmdFILE != NULL && job->cmdFILE != stdout) {
979 	   (void) fclose(job->cmdFILE);
980 	   job->cmdFILE = NULL;
981 	}
982 	done = TRUE;
983 #ifdef REMOTE
984 	if (job->flags & JOB_REMOTE)
985 	    Rmt_Done(job->rmtID, job->node);
986 #endif
987     } else if (WIFEXITED(*status)) {
988 	/*
989 	 * Deal with ignored errors in -B mode. We need to print a message
990 	 * telling of the ignored error as well as setting status.w_status
991 	 * to 0 so the next command gets run. To do this, we set done to be
992 	 * TRUE if in -B mode and the job exited non-zero.
993 	 */
994 	done = WEXITSTATUS(*status) != 0;
995 	/*
996 	 * Old comment said: "Note we don't
997 	 * want to close down any of the streams until we know we're at the
998 	 * end."
999 	 * But we do. Otherwise when are we going to print the rest of the
1000 	 * stuff?
1001 	 */
1002 	JobClose(job);
1003 #ifdef REMOTE
1004 	if (job->flags & JOB_REMOTE)
1005 	    Rmt_Done(job->rmtID, job->node);
1006 #endif /* REMOTE */
1007     } else {
1008 	/*
1009 	 * No need to close things down or anything.
1010 	 */
1011 	done = FALSE;
1012     }
1013 
1014     if (done ||
1015 	WIFSTOPPED(*status) ||
1016 	(WIFSIGNALED(*status) && (WTERMSIG(*status) == SIGCONT)))
1017     {
1018 	FILE	  *out;
1019 
1020 	if (compatMake && !usePipes && (job->flags & JOB_IGNERR)) {
1021 	    /*
1022 	     * If output is going to a file and this job is ignoring
1023 	     * errors, arrange to have the exit status sent to the
1024 	     * output file as well.
1025 	     */
1026 	    out = fdopen(job->outFd, "w");
1027 	    if (out == NULL)
1028 		Punt("Cannot fdopen");
1029 	} else {
1030 	    out = stdout;
1031 	}
1032 
1033 	if (WIFEXITED(*status)) {
1034 	    if (DEBUG(JOB)) {
1035 		(void) fprintf(stdout, "Process %d [%s] exited.\n",
1036 				job->pid, job->node->name);
1037 		(void) fflush(stdout);
1038 	    }
1039 	    if (WEXITSTATUS(*status) != 0) {
1040 		if (usePipes && job->node != lastNode) {
1041 		    MESSAGE(out, job->node);
1042 		    lastNode = job->node;
1043 		}
1044 		(void) fprintf(out, "*** [%s] Error code %d%s\n",
1045 				job->node->name,
1046 			       WEXITSTATUS(*status),
1047 			       (job->flags & JOB_IGNERR) ? "(ignored)" : "");
1048 
1049 		if (job->flags & JOB_IGNERR) {
1050 		    *status = 0;
1051 		}
1052 	    } else if (DEBUG(JOB)) {
1053 		if (usePipes && job->node != lastNode) {
1054 		    MESSAGE(out, job->node);
1055 		    lastNode = job->node;
1056 		}
1057 		(void) fprintf(out, "*** [%s] Completed successfully\n",
1058 				job->node->name);
1059 	    }
1060 	} else if (WIFSTOPPED(*status) && WSTOPSIG(*status) != SIGCONT) {
1061 	    if (DEBUG(JOB)) {
1062 		(void) fprintf(stdout, "Process %d (%s) stopped.\n",
1063 				job->pid, job->node->name);
1064 		(void) fflush(stdout);
1065 	    }
1066 	    if (usePipes && job->node != lastNode) {
1067 		MESSAGE(out, job->node);
1068 		lastNode = job->node;
1069 	    }
1070 	    if (!(job->flags & JOB_REMIGRATE)) {
1071 		switch (WSTOPSIG(*status)) {
1072 		case SIGTSTP:
1073 		    (void) fprintf(out, "*** [%s] Suspended\n",
1074 				job->node->name);
1075 		    break;
1076 		case SIGSTOP:
1077 		    (void) fprintf(out, "*** [%s] Stopped\n",
1078 				job->node->name);
1079 		    break;
1080 		default:
1081 		    (void) fprintf(out, "*** [%s] Stopped -- signal %d\n",
1082 			job->node->name, WSTOPSIG(*status));
1083 		}
1084 	    }
1085 	    job->flags |= JOB_RESUME;
1086 	    (void)Lst_AtEnd(stoppedJobs, (ClientData)job);
1087 #ifdef REMOTE
1088 	    if (job->flags & JOB_REMIGRATE)
1089 		JobRestart(job);
1090 #endif
1091 	    (void) fflush(out);
1092 	    return;
1093 	} else if (WIFSTOPPED(*status) &&  WSTOPSIG(*status) == SIGCONT) {
1094 	    /*
1095 	     * If the beastie has continued, shift the Job from the stopped
1096 	     * list to the running one (or re-stop it if concurrency is
1097 	     * exceeded) and go and get another child.
1098 	     */
1099 	    if (job->flags & (JOB_RESUME|JOB_REMIGRATE|JOB_RESTART)) {
1100 		if (usePipes && job->node != lastNode) {
1101 		    MESSAGE(out, job->node);
1102 		    lastNode = job->node;
1103 		}
1104 		(void) fprintf(out, "*** [%s] Continued\n", job->node->name);
1105 	    }
1106 	    if (!(job->flags & JOB_CONTINUING)) {
1107 		if (DEBUG(JOB)) {
1108 		    (void) fprintf(stdout,
1109 			   "Warning: process %d [%s] was not continuing.\n",
1110 			   job->pid, job->node->name);
1111 		    (void) fflush(stdout);
1112 		}
1113 #ifdef notdef
1114 		/*
1115 		 * We don't really want to restart a job from scratch just
1116 		 * because it continued, especially not without killing the
1117 		 * continuing process!  That's why this is ifdef'ed out.
1118 		 * FD - 9/17/90
1119 		 */
1120 		JobRestart(job);
1121 #endif
1122 	    }
1123 	    job->flags &= ~JOB_CONTINUING;
1124  	    Lst_AtEnd(jobs, (ClientData)job);
1125 	    nJobs += 1;
1126 	    if (!(job->flags & JOB_REMOTE)) {
1127 		if (DEBUG(JOB)) {
1128 		    (void) fprintf(stdout,
1129 				   "Process %d is continuing locally.\n",
1130 				   job->pid);
1131 		    (void) fflush(stdout);
1132   		}
1133 		nLocal += 1;
1134 	    }
1135 	    (void) fflush(out);
1136   	    return;
1137 	} else {
1138 	    if (usePipes && job->node != lastNode) {
1139 		MESSAGE(out, job->node);
1140 		lastNode = job->node;
1141 	    }
1142 	    (void) fprintf(out, "*** [%s] Signal %d\n",
1143 			job->node->name, WTERMSIG(*status));
1144 	}
1145 
1146 	(void) fflush(out);
1147     }
1148 
1149     /*
1150      * Now handle the -B-mode stuff. If the beast still isn't finished,
1151      * try and restart the job on the next command. If JobStart says it's
1152      * ok, it's ok. If there's an error, this puppy is done.
1153      */
1154     if (compatMake && (WIFEXITED(*status) &&
1155 	!Lst_IsAtEnd(job->node->commands))) {
1156 	switch (JobStart(job->node, job->flags & JOB_IGNDOTS, job)) {
1157 	case JOB_RUNNING:
1158 	    done = FALSE;
1159 	    break;
1160 	case JOB_ERROR:
1161 	    done = TRUE;
1162 	    *status = W_EXITCODE(1, 0);
1163 	    break;
1164 	case JOB_FINISHED:
1165 	    /*
1166 	     * If we got back a JOB_FINISHED code, JobStart has already
1167 	     * called Make_Update and freed the job descriptor. We set
1168 	     * done to false here to avoid fake cycles and double frees.
1169 	     * JobStart needs to do the update so we can proceed up the
1170 	     * graph when given the -n flag..
1171 	     */
1172 	    done = FALSE;
1173 	    break;
1174 	}
1175     } else {
1176 	done = TRUE;
1177     }
1178 
1179     if (done) {
1180 	Trace_Log(JOBEND, job);
1181 	if (!compatMake && !(job->flags & JOB_SPECIAL)) {
1182 	    if ((*status != 0) ||
1183 	        (aborting == ABORT_ERROR) ||
1184 	        (aborting == ABORT_INTERRUPT))
1185 		Job_TokenReturn();
1186 	}
1187 
1188     }
1189 
1190     if (done &&
1191 	(aborting != ABORT_ERROR) &&
1192 	(aborting != ABORT_INTERRUPT) &&
1193 	(*status == 0))
1194     {
1195 	/*
1196 	 * As long as we aren't aborting and the job didn't return a non-zero
1197 	 * status that we shouldn't ignore, we call Make_Update to update
1198 	 * the parents. In addition, any saved commands for the node are placed
1199 	 * on the .END target.
1200 	 */
1201 	if (job->tailCmds != NILLNODE) {
1202 	    Lst_ForEachFrom(job->node->commands, job->tailCmds,
1203 			     JobSaveCommand,
1204 			    (ClientData)job->node);
1205 	}
1206 	job->node->made = MADE;
1207 	if (!(job->flags & JOB_SPECIAL))
1208 	    Job_TokenReturn();
1209 	Make_Update(job->node);
1210 	free((Address)job);
1211     } else if (*status != 0) {
1212 	errors += 1;
1213 	free((Address)job);
1214     }
1215     JobRestartJobs();
1216 
1217     /*
1218      * Set aborting if any error.
1219      */
1220     if (errors && !keepgoing && (aborting != ABORT_INTERRUPT)) {
1221 	/*
1222 	 * If we found any errors in this batch of children and the -k flag
1223 	 * wasn't given, we set the aborting flag so no more jobs get
1224 	 * started.
1225 	 */
1226 	aborting = ABORT_ERROR;
1227     }
1228 
1229     if ((aborting == ABORT_ERROR) && Job_Empty()) {
1230 	/*
1231 	 * If we are aborting and the job table is now empty, we finish.
1232 	 */
1233 	Finish(errors);
1234     }
1235 }
1236 
1237 /*-
1238  *-----------------------------------------------------------------------
1239  * Job_Touch --
1240  *	Touch the given target. Called by JobStart when the -t flag was
1241  *	given
1242  *
1243  * Input:
1244  *	gn		the node of the file to touch
1245  *	silent		TRUE if should not print message
1246  *
1247  * Results:
1248  *	None
1249  *
1250  * Side Effects:
1251  *	The data modification of the file is changed. In addition, if the
1252  *	file did not exist, it is created.
1253  *-----------------------------------------------------------------------
1254  */
1255 void
1256 Job_Touch(GNode *gn, Boolean silent)
1257 {
1258     int		  streamID;   	/* ID of stream opened to do the touch */
1259     struct utimbuf times;	/* Times for utime() call */
1260 
1261     if (gn->type & (OP_JOIN|OP_USE|OP_USEBEFORE|OP_EXEC|OP_OPTIONAL|OP_PHONY)) {
1262 	/*
1263 	 * .JOIN, .USE, .ZEROTIME and .OPTIONAL targets are "virtual" targets
1264 	 * and, as such, shouldn't really be created.
1265 	 */
1266 	return;
1267     }
1268 
1269     if (!silent || NoExecute(gn)) {
1270 	(void) fprintf(stdout, "touch %s\n", gn->name);
1271 	(void) fflush(stdout);
1272     }
1273 
1274     if (NoExecute(gn)) {
1275 	return;
1276     }
1277 
1278     if (gn->type & OP_ARCHV) {
1279 	Arch_Touch(gn);
1280     } else if (gn->type & OP_LIB) {
1281 	Arch_TouchLib(gn);
1282     } else {
1283 	char	*file = gn->path ? gn->path : gn->name;
1284 
1285 	times.actime = times.modtime = now;
1286 	if (utime(file, &times) < 0){
1287 	    streamID = open(file, O_RDWR | O_CREAT, 0666);
1288 
1289 	    if (streamID >= 0) {
1290 		char	c;
1291 
1292 		/*
1293 		 * Read and write a byte to the file to change the
1294 		 * modification time, then close the file.
1295 		 */
1296 		if (read(streamID, &c, 1) == 1) {
1297 		    (void) lseek(streamID, (off_t)0, SEEK_SET);
1298 		    (void) write(streamID, &c, 1);
1299 		}
1300 
1301 		(void) close(streamID);
1302 	    } else {
1303 		(void) fprintf(stdout, "*** couldn't touch %s: %s",
1304 			       file, strerror(errno));
1305 		(void) fflush(stdout);
1306 	    }
1307 	}
1308     }
1309 }
1310 
1311 /*-
1312  *-----------------------------------------------------------------------
1313  * Job_CheckCommands --
1314  *	Make sure the given node has all the commands it needs.
1315  *
1316  * Input:
1317  *	gn		The target whose commands need verifying
1318  *	abortProc	Function to abort with message
1319  *
1320  * Results:
1321  *	TRUE if the commands list is/was ok.
1322  *
1323  * Side Effects:
1324  *	The node will have commands from the .DEFAULT rule added to it
1325  *	if it needs them.
1326  *-----------------------------------------------------------------------
1327  */
1328 Boolean
1329 Job_CheckCommands(GNode *gn, void (*abortProc)(const char *, ...))
1330 {
1331     if (OP_NOP(gn->type) && Lst_IsEmpty(gn->commands) &&
1332 	((gn->type & OP_LIB) == 0 || Lst_IsEmpty(gn->children))) {
1333 	/*
1334 	 * No commands. Look for .DEFAULT rule from which we might infer
1335 	 * commands
1336 	 */
1337 	if ((DEFAULT != NILGNODE) && !Lst_IsEmpty(DEFAULT->commands)) {
1338 	    char *p1;
1339 	    /*
1340 	     * Make only looks for a .DEFAULT if the node was never the
1341 	     * target of an operator, so that's what we do too. If
1342 	     * a .DEFAULT was given, we substitute its commands for gn's
1343 	     * commands and set the IMPSRC variable to be the target's name
1344 	     * The DEFAULT node acts like a transformation rule, in that
1345 	     * gn also inherits any attributes or sources attached to
1346 	     * .DEFAULT itself.
1347 	     */
1348 	    Make_HandleUse(DEFAULT, gn);
1349 	    Var_Set(IMPSRC, Var_Value(TARGET, gn, &p1), gn, 0);
1350 	    if (p1)
1351 		free(p1);
1352 	} else if (Dir_MTime(gn) == 0) {
1353 	    /*
1354 	     * The node wasn't the target of an operator we have no .DEFAULT
1355 	     * rule to go on and the target doesn't already exist. There's
1356 	     * nothing more we can do for this branch. If the -k flag wasn't
1357 	     * given, we stop in our tracks, otherwise we just don't update
1358 	     * this node's parents so they never get examined.
1359 	     */
1360 	    static const char msg[] = ": don't know how to make";
1361 
1362 	    if (gn->type & OP_OPTIONAL) {
1363 		(void) fprintf(stdout, "%s%s %s(ignored)\n", progname,
1364 		    msg, gn->name);
1365 		(void) fflush(stdout);
1366 	    } else if (keepgoing) {
1367 		(void) fprintf(stdout, "%s%s %s(continuing)\n", progname,
1368 		    msg, gn->name);
1369 		(void) fflush(stdout);
1370   		return FALSE;
1371 	    } else {
1372 		(*abortProc)("%s%s %s. Stop", progname, msg, gn->name);
1373 		return FALSE;
1374 	    }
1375 	}
1376     }
1377     return TRUE;
1378 }
1379 #ifdef RMT_WILL_WATCH
1380 /*-
1381  *-----------------------------------------------------------------------
1382  * JobLocalInput --
1383  *	Handle a pipe becoming readable. Callback function for Rmt_Watch
1384  *
1385  * Input:
1386  *	stream		Stream that's ready (ignored)
1387  *	job		Job to which the stream belongs
1388  *
1389  * Results:
1390  *	None
1391  *
1392  * Side Effects:
1393  *	JobDoOutput is called.
1394  *
1395  *-----------------------------------------------------------------------
1396  */
1397 /*ARGSUSED*/
1398 static void
1399 JobLocalInput(int stream, Job *job)
1400 {
1401     JobDoOutput(job, FALSE);
1402 }
1403 #endif /* RMT_WILL_WATCH */
1404 
1405 /*-
1406  *-----------------------------------------------------------------------
1407  * JobExec --
1408  *	Execute the shell for the given job. Called from JobStart and
1409  *	JobRestart.
1410  *
1411  * Input:
1412  *	job		Job to execute
1413  *
1414  * Results:
1415  *	None.
1416  *
1417  * Side Effects:
1418  *	A shell is executed, outputs is altered and the Job structure added
1419  *	to the job table.
1420  *
1421  *-----------------------------------------------------------------------
1422  */
1423 static void
1424 JobExec(Job *job, char **argv)
1425 {
1426     int	    	  cpid;	    	/* ID of new child */
1427     sigset_t	  mask;
1428 
1429     job->flags &= ~JOB_TRACED;
1430 
1431     if (DEBUG(JOB)) {
1432 	int 	  i;
1433 
1434 	(void) fprintf(stdout, "Running %s %sly\n", job->node->name,
1435 		       job->flags&JOB_REMOTE?"remote":"local");
1436 	(void) fprintf(stdout, "\tCommand: ");
1437 	for (i = 0; argv[i] != NULL; i++) {
1438 	    (void) fprintf(stdout, "%s ", argv[i]);
1439 	}
1440  	(void) fprintf(stdout, "\n");
1441  	(void) fflush(stdout);
1442     }
1443 
1444     /*
1445      * Some jobs produce no output and it's disconcerting to have
1446      * no feedback of their running (since they produce no output, the
1447      * banner with their name in it never appears). This is an attempt to
1448      * provide that feedback, even if nothing follows it.
1449      */
1450     if ((lastNode != job->node) && (job->flags & JOB_FIRST) &&
1451 	!(job->flags & JOB_SILENT)) {
1452 	MESSAGE(stdout, job->node);
1453 	lastNode = job->node;
1454     }
1455 
1456 #ifdef RMT_NO_EXEC
1457     if (job->flags & JOB_REMOTE) {
1458 	goto jobExecFinish;
1459     }
1460 #endif /* RMT_NO_EXEC */
1461 
1462     /* No interruptions until this job is on the `jobs' list */
1463     JobSigLock(&mask);
1464 
1465     if ((cpid = vfork()) == -1) {
1466 	Punt("Cannot vfork: %s", strerror(errno));
1467     } else if (cpid == 0) {
1468 
1469 	/*
1470 	 * Reset all signal handlers; this is necessary because we also
1471 	 * need to unblock signals before we exec(2).
1472 	 */
1473 	JobSigReset();
1474 
1475 	/* Now unblock signals */
1476 	sigemptyset(&mask);
1477 	JobSigUnlock(&mask);
1478 
1479 	/*
1480 	 * Must duplicate the input stream down to the child's input and
1481 	 * reset it to the beginning (again). Since the stream was marked
1482 	 * close-on-exec, we must clear that bit in the new input.
1483 	 */
1484 	if (dup2(FILENO(job->cmdFILE), 0) == -1) {
1485 	    execError("dup2", "job->cmdFILE");
1486 	    _exit(1);
1487 	}
1488 	(void) fcntl(0, F_SETFD, 0);
1489 	(void) lseek(0, (off_t)0, SEEK_SET);
1490 
1491 	if (job->node->type & OP_MAKE) {
1492 		/*
1493 		 * Pass job token pipe to submakes.
1494 		 */
1495 		fcntl(job_pipe[0], F_SETFD, 0);
1496 		fcntl(job_pipe[1], F_SETFD, 0);
1497 	}
1498 
1499 	if (usePipes) {
1500 	    /*
1501 	     * Set up the child's output to be routed through the pipe
1502 	     * we've created for it.
1503 	     */
1504 	    if (dup2(job->outPipe, 1) == -1) {
1505 		execError("dup2", "job->outPipe");
1506 		_exit(1);
1507 	    }
1508 	} else {
1509 	    /*
1510 	     * We're capturing output in a file, so we duplicate the
1511 	     * descriptor to the temporary file into the standard
1512 	     * output.
1513 	     */
1514 	    if (dup2(job->outFd, 1) == -1) {
1515 		execError("dup2", "job->outFd");
1516 		_exit(1);
1517 	    }
1518 	}
1519 	/*
1520 	 * The output channels are marked close on exec. This bit was
1521 	 * duplicated by the dup2 (on some systems), so we have to clear
1522 	 * it before routing the shell's error output to the same place as
1523 	 * its standard output.
1524 	 */
1525 	(void) fcntl(1, F_SETFD, 0);
1526 	if (dup2(1, 2) == -1) {
1527 	    execError("dup2", "1, 2");
1528 	    _exit(1);
1529 	}
1530 
1531 #ifdef USE_PGRP
1532 	/*
1533 	 * We want to switch the child into a different process family so
1534 	 * we can kill it and all its descendants in one fell swoop,
1535 	 * by killing its process family, but not commit suicide.
1536 	 */
1537 # if defined(SYSV)
1538 	(void) setsid();
1539 # else
1540 	(void) setpgid(0, getpid());
1541 # endif
1542 #endif /* USE_PGRP */
1543 
1544 #ifdef REMOTE
1545 	if (job->flags & JOB_REMOTE) {
1546 	    Rmt_Exec(shellPath, argv, FALSE);
1547 	} else
1548 #endif /* REMOTE */
1549 	{
1550 	   (void) execv(shellPath, argv);
1551 	   execError("exec", shellPath);
1552 	}
1553 	_exit(1);
1554     } else {
1555 	job->pid = cpid;
1556 
1557 	Trace_Log(JOBSTART, job);
1558 
1559 	if (usePipes && (job->flags & JOB_FIRST)) {
1560 	    /*
1561 	     * The first time a job is run for a node, we set the current
1562 	     * position in the buffer to the beginning and mark another
1563 	     * stream to watch in the outputs mask
1564 	     */
1565 	    job->curPos = 0;
1566 
1567 #ifdef RMT_WILL_WATCH
1568 	    Rmt_Watch(job->inPipe, JobLocalInput, job);
1569 #else
1570 	    watchfd(job);
1571 #endif /* RMT_WILL_WATCH */
1572 	}
1573 
1574 	if (job->flags & JOB_REMOTE) {
1575 #ifndef REMOTE
1576 	    job->rmtID = 0;
1577 #else
1578 	    job->rmtID = Rmt_LastID(job->pid);
1579 #endif /* REMOTE */
1580 	} else {
1581 	    nLocal += 1;
1582 	    /*
1583 	     * XXX: Used to not happen if REMOTE. Why?
1584 	     */
1585 	    if (job->cmdFILE != NULL && job->cmdFILE != stdout) {
1586 		(void) fclose(job->cmdFILE);
1587 		job->cmdFILE = NULL;
1588 	    }
1589 	}
1590     }
1591 
1592 #ifdef RMT_NO_EXEC
1593 jobExecFinish:
1594 #endif
1595     /*
1596      * Now the job is actually running, add it to the table.
1597      */
1598     if (DEBUG(JOB)) {
1599 	printf("JobExec(%s): pid %d added to jobs table\n",
1600 		job->node->name, job->pid);
1601     }
1602     nJobs += 1;
1603     (void) Lst_AtEnd(jobs, (ClientData)job);
1604     JobSigUnlock(&mask);
1605 }
1606 
1607 /*-
1608  *-----------------------------------------------------------------------
1609  * JobMakeArgv --
1610  *	Create the argv needed to execute the shell for a given job.
1611  *
1612  *
1613  * Results:
1614  *
1615  * Side Effects:
1616  *
1617  *-----------------------------------------------------------------------
1618  */
1619 static void
1620 JobMakeArgv(Job *job, char **argv)
1621 {
1622     int	    	  argc;
1623     static char args[10]; 	/* For merged arguments */
1624 
1625     argv[0] = UNCONST(shellName);
1626     argc = 1;
1627 
1628     if ((commandShell->exit && (*commandShell->exit != '-')) ||
1629 	(commandShell->echo && (*commandShell->echo != '-')))
1630     {
1631 	/*
1632 	 * At least one of the flags doesn't have a minus before it, so
1633 	 * merge them together. Have to do this because the *(&(@*#*&#$#
1634 	 * Bourne shell thinks its second argument is a file to source.
1635 	 * Grrrr. Note the ten-character limitation on the combined arguments.
1636 	 */
1637 	(void)snprintf(args, sizeof(args), "-%s%s",
1638 		      ((job->flags & JOB_IGNERR) ? "" :
1639 		       (commandShell->exit ? commandShell->exit : "")),
1640 		      ((job->flags & JOB_SILENT) ? "" :
1641 		       (commandShell->echo ? commandShell->echo : "")));
1642 
1643 	if (args[1]) {
1644 	    argv[argc] = args;
1645 	    argc++;
1646 	}
1647     } else {
1648 	if (!(job->flags & JOB_IGNERR) && commandShell->exit) {
1649 	    argv[argc] = UNCONST(commandShell->exit);
1650 	    argc++;
1651 	}
1652 	if (!(job->flags & JOB_SILENT) && commandShell->echo) {
1653 	    argv[argc] = UNCONST(commandShell->echo);
1654 	    argc++;
1655 	}
1656     }
1657     argv[argc] = NULL;
1658 }
1659 
1660 /*-
1661  *-----------------------------------------------------------------------
1662  * JobRestart --
1663  *	Restart a job that stopped for some reason.
1664  *
1665  * Input:
1666  *	job		Job to restart
1667  *
1668  * Results:
1669  *	1 if max number of running jobs has been reached, 0 otherwise.
1670  *
1671  *-----------------------------------------------------------------------
1672  */
1673 static int
1674 JobRestart(Job *job)
1675 {
1676 #ifdef REMOTE
1677     int host;
1678 #endif
1679 
1680     if (job->flags & JOB_REMIGRATE) {
1681 	if (
1682 #ifdef REMOTE
1683 	    verboseRemigrates ||
1684 #endif
1685 	    DEBUG(JOB)) {
1686 	   (void) fprintf(stdout, "*** remigrating %x(%s)\n",
1687 			   job->pid, job->node->name);
1688 	   (void) fflush(stdout);
1689 	}
1690 
1691 #ifdef REMOTE
1692 	if (!Rmt_ReExport(job->pid, job->node, &host)) {
1693 	    if (verboseRemigrates || DEBUG(JOB)) {
1694 		(void) fprintf(stdout, "*** couldn't migrate...\n");
1695 		(void) fflush(stdout);
1696 	    }
1697 #endif
1698 	    if (nLocal != maxLocal) {
1699 		/*
1700 		 * Job cannot be remigrated, but there's room on the local
1701 		 * machine, so resume the job and note that another
1702 		 * local job has started.
1703 		 */
1704 		if (
1705 #ifdef REMOTE
1706 		    verboseRemigrates ||
1707 #endif
1708 		    DEBUG(JOB)) {
1709 		    (void) fprintf(stdout, "*** resuming on local machine\n");
1710 		    (void) fflush(stdout);
1711 		}
1712 		KILL(job->pid, SIGCONT);
1713 		nLocal +=1;
1714 #ifdef REMOTE
1715 		job->flags &= ~(JOB_REMIGRATE|JOB_RESUME|JOB_REMOTE);
1716 		job->flags |= JOB_CONTINUING;
1717 #else
1718 		job->flags &= ~(JOB_REMIGRATE|JOB_RESUME);
1719 #endif
1720 	    } else {
1721 		/*
1722 		 * Job cannot be restarted. Mark the table as full and
1723 		 * place the job back on the list of stopped jobs.
1724 		 */
1725 		if (
1726 #ifdef REMOTE
1727 		    verboseRemigrates ||
1728 #endif
1729 		    DEBUG(JOB)) {
1730 		   (void) fprintf(stdout, "*** holding\n");
1731 		   (void) fflush(stdout);
1732   		}
1733 		(void)Lst_AtFront(stoppedJobs, (ClientData)job);
1734 		return 1;
1735 	    }
1736 #ifdef REMOTE
1737 	} else {
1738 	    /*
1739 	     * Clear out the remigrate and resume flags. Set the continuing
1740 	     * flag so we know later on that the process isn't exiting just
1741 	     * because of a signal.
1742 	     */
1743 	    job->flags &= ~(JOB_REMIGRATE|JOB_RESUME);
1744 	    job->flags |= JOB_CONTINUING;
1745 	    job->rmtID = host;
1746 	}
1747 #endif
1748 
1749 	(void)Lst_AtEnd(jobs, (ClientData)job);
1750 	nJobs += 1;
1751     } else if (job->flags & JOB_RESTART) {
1752 	/*
1753 	 * Set up the control arguments to the shell. This is based on the
1754 	 * flags set earlier for this job. If the JOB_IGNERR flag is clear,
1755 	 * the 'exit' flag of the commandShell is used to cause it to exit
1756 	 * upon receiving an error. If the JOB_SILENT flag is clear, the
1757 	 * 'echo' flag of the commandShell is used to get it to start echoing
1758 	 * as soon as it starts processing commands.
1759 	 */
1760 	char	  *argv[10];
1761 
1762 	JobMakeArgv(job, argv);
1763 
1764 	if (DEBUG(JOB)) {
1765 	    (void) fprintf(stdout, "Restarting %s...", job->node->name);
1766 	    (void) fflush(stdout);
1767 	}
1768 #ifdef REMOTE
1769 	if ((job->node->type & OP_NOEXPORT) ||
1770  	    (nLocal < maxLocal && runLocalFirst)
1771 # ifdef RMT_NO_EXEC
1772 	    || !Rmt_Export(shellPath, argv, job)
1773 # else
1774 	    || !Rmt_Begin(shellPath, argv, job->node)
1775 # endif
1776 	   )
1777 #endif
1778 	{
1779 	    if (((nLocal >= maxLocal) && !(job->flags & JOB_SPECIAL))) {
1780 		/*
1781 		 * Can't be exported and not allowed to run locally -- put it
1782 		 * back on the hold queue and mark the table full
1783 		 */
1784 		if (DEBUG(JOB)) {
1785 		    (void) fprintf(stdout, "holding\n");
1786 		    (void) fflush(stdout);
1787 		}
1788 		(void)Lst_AtFront(stoppedJobs, (ClientData)job);
1789 		return 1;
1790 	    } else {
1791 		/*
1792 		 * Job may be run locally.
1793 		 */
1794 		if (DEBUG(JOB)) {
1795 		    (void) fprintf(stdout, "running locally\n");
1796 		    (void) fflush(stdout);
1797 		}
1798 		job->flags &= ~JOB_REMOTE;
1799 	    }
1800 	}
1801 #ifdef REMOTE
1802 	else {
1803 	    /*
1804 	     * Can be exported. Hooray!
1805 	     */
1806 	    if (DEBUG(JOB)) {
1807 		(void) fprintf(stdout, "exporting\n");
1808 		(void) fflush(stdout);
1809 	    }
1810 	    job->flags |= JOB_REMOTE;
1811 	}
1812 #endif
1813 	JobExec(job, argv);
1814     } else {
1815 	/*
1816 	 * The job has stopped and needs to be restarted. Why it stopped,
1817 	 * we don't know...
1818 	 */
1819 	if (DEBUG(JOB)) {
1820 	   (void) fprintf(stdout, "Resuming %s...", job->node->name);
1821 	   (void) fflush(stdout);
1822 	}
1823 	if ((nJobs != maxJobs) &&
1824 	    ((job->flags & JOB_REMOTE) ||
1825 	     (nLocal < maxLocal) ||
1826 	     ((maxLocal == 0) &&
1827 		((job->flags & JOB_SPECIAL)
1828 #ifdef REMOTE
1829 			&& (job->node->type & OP_NOEXPORT)
1830 #endif
1831 	    ))))
1832 	{
1833 	    /*
1834 	     * If the job is remote, it's ok to resume it as long as the
1835 	     * maximum concurrency won't be exceeded. If it's local and
1836 	     * we haven't reached the local concurrency limit already (or the
1837 	     * job must be run locally and maxLocal is 0), it's also ok to
1838 	     * resume it.
1839 	     */
1840 	    Boolean error;
1841 	    int status;
1842 
1843 #ifdef RMT_WANTS_SIGNALS
1844 	    if (job->flags & JOB_REMOTE) {
1845 		error = !Rmt_Signal(job, SIGCONT);
1846 	    } else
1847 #endif	/* RMT_WANTS_SIGNALS */
1848 		error = (KILL(job->pid, SIGCONT) != 0);
1849 
1850 	    if (!error) {
1851 		/*
1852 		 * Make sure the user knows we've continued the beast and
1853 		 * actually put the thing in the job table.
1854 		 */
1855 		job->flags |= JOB_CONTINUING;
1856 		status = W_STOPCODE(SIGCONT);
1857 		JobFinish(job, &status);
1858 
1859 		job->flags &= ~(JOB_RESUME|JOB_CONTINUING);
1860 		if (DEBUG(JOB)) {
1861 		   (void) fprintf(stdout, "done\n");
1862 		   (void) fflush(stdout);
1863 		}
1864 	    } else {
1865 		Error("couldn't resume %s: %s",
1866 		    job->node->name, strerror(errno));
1867 		status = W_EXITCODE(1, 0);
1868 		JobFinish(job, &status);
1869 	    }
1870 	} else {
1871 	    /*
1872 	     * Job cannot be restarted. Mark the table as full and
1873 	     * place the job back on the list of stopped jobs.
1874 	     */
1875 	    if (DEBUG(JOB)) {
1876 		(void) fprintf(stdout, "table full\n");
1877 		(void) fflush(stdout);
1878 	    }
1879 	    (void) Lst_AtFront(stoppedJobs, (ClientData)job);
1880 	    return 1;
1881 	}
1882     }
1883     return 0;
1884 }
1885 
1886 /*-
1887  *-----------------------------------------------------------------------
1888  * JobStart  --
1889  *	Start a target-creation process going for the target described
1890  *	by the graph node gn.
1891  *
1892  * Input:
1893  *	gn		target to create
1894  *	flags		flags for the job to override normal ones.
1895  *			e.g. JOB_SPECIAL or JOB_IGNDOTS
1896  *	previous	The previous Job structure for this node, if any.
1897  *
1898  * Results:
1899  *	JOB_ERROR if there was an error in the commands, JOB_FINISHED
1900  *	if there isn't actually anything left to do for the job and
1901  *	JOB_RUNNING if the job has been started.
1902  *
1903  * Side Effects:
1904  *	A new Job node is created and added to the list of running
1905  *	jobs. PMake is forked and a child shell created.
1906  *-----------------------------------------------------------------------
1907  */
1908 static int
1909 JobStart(GNode *gn, int flags, Job *previous)
1910 {
1911     Job		  *job;       /* new job descriptor */
1912     char	  *argv[10];  /* Argument vector to shell */
1913     Boolean	  cmdsOK;     /* true if the nodes commands were all right */
1914     Boolean 	  local;      /* Set true if the job was run locally */
1915     Boolean 	  noExec;     /* Set true if we decide not to run the job */
1916     int		  tfd;	      /* File descriptor to the temp file */
1917 
1918     if (previous != NULL) {
1919 	previous->flags &= ~(JOB_FIRST|JOB_IGNERR|JOB_SILENT|JOB_REMOTE);
1920 	job = previous;
1921     } else {
1922 	job = (Job *) emalloc(sizeof(Job));
1923 	if (job == NULL) {
1924 	    Punt("JobStart out of memory");
1925 	}
1926 	flags |= JOB_FIRST;
1927     }
1928 
1929     job->node = gn;
1930     job->tailCmds = NILLNODE;
1931 
1932     /*
1933      * Set the initial value of the flags for this job based on the global
1934      * ones and the node's attributes... Any flags supplied by the caller
1935      * are also added to the field.
1936      */
1937     job->flags = 0;
1938     if (Targ_Ignore(gn)) {
1939 	job->flags |= JOB_IGNERR;
1940     }
1941     if (Targ_Silent(gn)) {
1942 	job->flags |= JOB_SILENT;
1943     }
1944     job->flags |= flags;
1945 
1946     /*
1947      * Check the commands now so any attributes from .DEFAULT have a chance
1948      * to migrate to the node
1949      */
1950     if (!compatMake && job->flags & JOB_FIRST) {
1951 	cmdsOK = Job_CheckCommands(gn, Error);
1952     } else {
1953 	cmdsOK = TRUE;
1954     }
1955 
1956 #ifndef RMT_WILL_WATCH
1957     job->inPollfd = NULL;
1958 #endif
1959     /*
1960      * If the -n flag wasn't given, we open up OUR (not the child's)
1961      * temporary file to stuff commands in it. The thing is rd/wr so we don't
1962      * need to reopen it to feed it to the shell. If the -n flag *was* given,
1963      * we just set the file to be stdout. Cute, huh?
1964      */
1965     if (((gn->type & OP_MAKE) && !(noRecursiveExecute)) ||
1966 	(!noExecute && !touchFlag)) {
1967 	/*
1968 	 * tfile is the name of a file into which all shell commands are
1969 	 * put. It is used over by removing it before the child shell is
1970 	 * executed. The XXXXXX in the string are replaced by the pid of
1971 	 * the make process in a 6-character field with leading zeroes.
1972 	 */
1973 	char     tfile[sizeof(TMPPAT)];
1974 	sigset_t mask;
1975 	/*
1976 	 * We're serious here, but if the commands were bogus, we're
1977 	 * also dead...
1978 	 */
1979 	if (!cmdsOK) {
1980 	    DieHorribly();
1981 	}
1982 
1983 	JobSigLock(&mask);
1984 	(void)strcpy(tfile, TMPPAT);
1985 	if ((tfd = mkstemp(tfile)) == -1)
1986 	    Punt("Could not create temporary file %s", strerror(errno));
1987 	(void) eunlink(tfile);
1988 	JobSigUnlock(&mask);
1989 
1990 	job->cmdFILE = fdopen(tfd, "w+");
1991 	if (job->cmdFILE == NULL) {
1992 	    Punt("Could not fdopen %s", tfile);
1993 	}
1994 	(void) fcntl(FILENO(job->cmdFILE), F_SETFD, 1);
1995 	/*
1996 	 * Send the commands to the command file, flush all its buffers then
1997 	 * rewind and remove the thing.
1998 	 */
1999 	noExec = FALSE;
2000 
2001 	/*
2002 	 * used to be backwards; replace when start doing multiple commands
2003 	 * per shell.
2004 	 */
2005 	if (compatMake) {
2006 	    /*
2007 	     * Be compatible: If this is the first time for this node,
2008 	     * verify its commands are ok and open the commands list for
2009 	     * sequential access by later invocations of JobStart.
2010 	     * Once that is done, we take the next command off the list
2011 	     * and print it to the command file. If the command was an
2012 	     * ellipsis, note that there's nothing more to execute.
2013 	     */
2014 	    if ((job->flags&JOB_FIRST) && (Lst_Open(gn->commands) != SUCCESS)){
2015 		cmdsOK = FALSE;
2016 	    } else {
2017 		LstNode	ln = Lst_Next(gn->commands);
2018 
2019 		if ((ln == NILLNODE) ||
2020 		    JobPrintCommand((ClientData) Lst_Datum(ln),
2021 				    (ClientData) job))
2022 		{
2023 		    noExec = TRUE;
2024 		    Lst_Close(gn->commands);
2025 		}
2026 		if (noExec && !(job->flags & JOB_FIRST)) {
2027 		    /*
2028 		     * If we're not going to execute anything, the job
2029 		     * is done and we need to close down the various
2030 		     * file descriptors we've opened for output, then
2031 		     * call JobDoOutput to catch the final characters or
2032 		     * send the file to the screen... Note that the i/o streams
2033 		     * are only open if this isn't the first job.
2034 		     * Note also that this could not be done in
2035 		     * Job_CatchChildren b/c it wasn't clear if there were
2036 		     * more commands to execute or not...
2037 		     */
2038 		    JobClose(job);
2039 		}
2040 	    }
2041 	} else {
2042 	    /*
2043 	     * We can do all the commands at once. hooray for sanity
2044 	     */
2045 	    numCommands = 0;
2046 	    Lst_ForEach(gn->commands, JobPrintCommand, (ClientData)job);
2047 
2048 	    /*
2049 	     * If we didn't print out any commands to the shell script,
2050 	     * there's not much point in executing the shell, is there?
2051 	     */
2052 	    if (numCommands == 0) {
2053 		noExec = TRUE;
2054 	    }
2055 	}
2056     } else if (NoExecute(gn)) {
2057 	/*
2058 	 * Not executing anything -- just print all the commands to stdout
2059 	 * in one fell swoop. This will still set up job->tailCmds correctly.
2060 	 */
2061 	if (lastNode != gn) {
2062 	    MESSAGE(stdout, gn);
2063 	    lastNode = gn;
2064 	}
2065 	job->cmdFILE = stdout;
2066 	/*
2067 	 * Only print the commands if they're ok, but don't die if they're
2068 	 * not -- just let the user know they're bad and keep going. It
2069 	 * doesn't do any harm in this case and may do some good.
2070 	 */
2071 	if (cmdsOK) {
2072 	    Lst_ForEach(gn->commands, JobPrintCommand, (ClientData)job);
2073 	}
2074 	/*
2075 	 * Don't execute the shell, thank you.
2076 	 */
2077 	noExec = TRUE;
2078     } else {
2079 	/*
2080 	 * Just touch the target and note that no shell should be executed.
2081 	 * Set cmdFILE to stdout to make life easier. Check the commands, too,
2082 	 * but don't die if they're no good -- it does no harm to keep working
2083 	 * up the graph.
2084 	 */
2085 	job->cmdFILE = stdout;
2086     	Job_Touch(gn, job->flags&JOB_SILENT);
2087 	noExec = TRUE;
2088     }
2089 
2090     /*
2091      * If we're not supposed to execute a shell, don't.
2092      */
2093     if (noExec) {
2094 	/*
2095 	 * Unlink and close the command file if we opened one
2096 	 */
2097 	if (job->cmdFILE != stdout) {
2098 	    if (job->cmdFILE != NULL) {
2099 		(void) fclose(job->cmdFILE);
2100 		job->cmdFILE = NULL;
2101 	    }
2102 	} else {
2103 	     (void) fflush(stdout);
2104 	}
2105 
2106 	/*
2107 	 * We only want to work our way up the graph if we aren't here because
2108 	 * the commands for the job were no good.
2109 	 */
2110 	if (cmdsOK) {
2111 	    if (aborting == 0) {
2112 		if (job->tailCmds != NILLNODE) {
2113 		    Lst_ForEachFrom(job->node->commands, job->tailCmds,
2114 				    JobSaveCommand,
2115 				   (ClientData)job->node);
2116 		}
2117 		if (!(job->flags & JOB_SPECIAL))
2118 		    Job_TokenReturn();
2119 		job->node->made = MADE;
2120 		Make_Update(job->node);
2121 	    }
2122 	    free((Address)job);
2123 	    return(JOB_FINISHED);
2124 	} else {
2125 	    free((Address)job);
2126 	    return(JOB_ERROR);
2127 	}
2128     } else {
2129 	(void) fflush(job->cmdFILE);
2130     }
2131 
2132     /*
2133      * Set up the control arguments to the shell. This is based on the flags
2134      * set earlier for this job.
2135      */
2136     JobMakeArgv(job, argv);
2137 
2138     /*
2139      * If we're using pipes to catch output, create the pipe by which we'll
2140      * get the shell's output. If we're using files, print out that we're
2141      * starting a job and then set up its temporary-file name.
2142      */
2143     if (!compatMake || (job->flags & JOB_FIRST)) {
2144 	if (usePipes) {
2145 	    int fd[2];
2146 	    if (pipe(fd) == -1)
2147 		Punt("Cannot create pipe: %s", strerror(errno));
2148 	    job->inPipe = fd[0];
2149 	    job->outPipe = fd[1];
2150 	    (void) fcntl(job->inPipe, F_SETFD, 1);
2151 	    (void) fcntl(job->outPipe, F_SETFD, 1);
2152 	} else {
2153 	    (void) fprintf(stdout, "Remaking `%s'\n", gn->name);
2154   	    (void) fflush(stdout);
2155 	    (void) strcpy(job->outFile, TMPPAT);
2156 	    job->outFd = mkstemp(job->outFile);
2157 	    (void) fcntl(job->outFd, F_SETFD, 1);
2158 	}
2159     }
2160 
2161 #ifdef REMOTE
2162     if (!(gn->type & OP_NOEXPORT) && !(runLocalFirst && nLocal < maxLocal)) {
2163 #ifdef RMT_NO_EXEC
2164 	local = !Rmt_Export(shellPath, argv, job);
2165 #else
2166 	local = !Rmt_Begin(shellPath, argv, job->node);
2167 #endif /* RMT_NO_EXEC */
2168 	if (!local) {
2169 	    job->flags |= JOB_REMOTE;
2170 	}
2171     } else
2172 #endif
2173 	local = TRUE;
2174 
2175     if (local && (((nLocal >= maxLocal) &&
2176 	!(job->flags & JOB_SPECIAL) &&
2177 #ifdef REMOTE
2178 	(!(gn->type & OP_NOEXPORT) || (maxLocal != 0))
2179 #else
2180 	(maxLocal != 0)
2181 #endif
2182 	)))
2183     {
2184 	/*
2185 	 * The job can only be run locally, but we've hit the limit of
2186 	 * local concurrency, so put the job on hold until some other job
2187 	 * finishes. Note that the special jobs (.BEGIN, .INTERRUPT and .END)
2188 	 * may be run locally even when the local limit has been reached
2189 	 * (e.g. when maxLocal == 0), though they will be exported if at
2190 	 * all possible. In addition, any target marked with .NOEXPORT will
2191 	 * be run locally if maxLocal is 0.
2192 	 */
2193 	job->flags |= JOB_RESTART;
2194 	(void) Lst_AtEnd(stoppedJobs, (ClientData)job);
2195     } else {
2196 	JobExec(job, argv);
2197     }
2198     return(JOB_RUNNING);
2199 }
2200 
2201 static char *
2202 JobOutput(Job *job, char *cp, char *endp, int msg)
2203 {
2204     char *ecp;
2205 
2206     if (commandShell->noPrint) {
2207 	ecp = Str_FindSubstring(cp, commandShell->noPrint);
2208 	while (ecp != NULL) {
2209 	    if (cp != ecp) {
2210 		*ecp = '\0';
2211 		if (msg && job->node != lastNode) {
2212 		    MESSAGE(stdout, job->node);
2213 		    lastNode = job->node;
2214 		}
2215 		/*
2216 		 * The only way there wouldn't be a newline after
2217 		 * this line is if it were the last in the buffer.
2218 		 * however, since the non-printable comes after it,
2219 		 * there must be a newline, so we don't print one.
2220 		 */
2221 		(void) fprintf(stdout, "%s", cp);
2222 		(void) fflush(stdout);
2223 	    }
2224 	    cp = ecp + commandShell->noPLen;
2225 	    if (cp != endp) {
2226 		/*
2227 		 * Still more to print, look again after skipping
2228 		 * the whitespace following the non-printable
2229 		 * command....
2230 		 */
2231 		cp++;
2232 		while (*cp == ' ' || *cp == '\t' || *cp == '\n') {
2233 		    cp++;
2234 		}
2235 		ecp = Str_FindSubstring(cp, commandShell->noPrint);
2236 	    } else {
2237 		return cp;
2238 	    }
2239 	}
2240     }
2241     return cp;
2242 }
2243 
2244 /*-
2245  *-----------------------------------------------------------------------
2246  * JobDoOutput  --
2247  *	This function is called at different times depending on
2248  *	whether the user has specified that output is to be collected
2249  *	via pipes or temporary files. In the former case, we are called
2250  *	whenever there is something to read on the pipe. We collect more
2251  *	output from the given job and store it in the job's outBuf. If
2252  *	this makes up a line, we print it tagged by the job's identifier,
2253  *	as necessary.
2254  *	If output has been collected in a temporary file, we open the
2255  *	file and read it line by line, transfering it to our own
2256  *	output channel until the file is empty. At which point we
2257  *	remove the temporary file.
2258  *	In both cases, however, we keep our figurative eye out for the
2259  *	'noPrint' line for the shell from which the output came. If
2260  *	we recognize a line, we don't print it. If the command is not
2261  *	alone on the line (the character after it is not \0 or \n), we
2262  *	do print whatever follows it.
2263  *
2264  * Input:
2265  *	job		the job whose output needs printing
2266  *	finish		TRUE if this is the last time we'll be called
2267  *			for this job
2268  *
2269  * Results:
2270  *	None
2271  *
2272  * Side Effects:
2273  *	curPos may be shifted as may the contents of outBuf.
2274  *-----------------------------------------------------------------------
2275  */
2276 STATIC void
2277 JobDoOutput(Job *job, Boolean finish)
2278 {
2279     Boolean       gotNL = FALSE;  /* true if got a newline */
2280     Boolean       fbuf;  	  /* true if our buffer filled up */
2281     int		  nr;	      	  /* number of bytes read */
2282     int		  i;	      	  /* auxiliary index into outBuf */
2283     int		  max;	      	  /* limit for i (end of current data) */
2284     int		  nRead;      	  /* (Temporary) number of bytes read */
2285 
2286     FILE      	  *oFILE;	  /* Stream pointer to shell's output file */
2287     char          inLine[132];
2288 
2289 
2290     if (usePipes) {
2291 	/*
2292 	 * Read as many bytes as will fit in the buffer.
2293 	 */
2294 end_loop:
2295 	gotNL = FALSE;
2296 	fbuf = FALSE;
2297 
2298 	nRead = read(job->inPipe, &job->outBuf[job->curPos],
2299 			 JOB_BUFSIZE - job->curPos);
2300 	if (nRead < 0) {
2301 	    if (DEBUG(JOB)) {
2302 		perror("JobDoOutput(piperead)");
2303 	    }
2304 	    nr = 0;
2305 	} else {
2306 	    nr = nRead;
2307 	}
2308 
2309 	/*
2310 	 * If we hit the end-of-file (the job is dead), we must flush its
2311 	 * remaining output, so pretend we read a newline if there's any
2312 	 * output remaining in the buffer.
2313 	 * Also clear the 'finish' flag so we stop looping.
2314 	 */
2315 	if ((nr == 0) && (job->curPos != 0)) {
2316 	    job->outBuf[job->curPos] = '\n';
2317 	    nr = 1;
2318 	    finish = FALSE;
2319 	} else if (nr == 0) {
2320 	    finish = FALSE;
2321 	}
2322 
2323 	/*
2324 	 * Look for the last newline in the bytes we just got. If there is
2325 	 * one, break out of the loop with 'i' as its index and gotNL set
2326 	 * TRUE.
2327 	 */
2328 	max = job->curPos + nr;
2329 	for (i = job->curPos + nr - 1; i >= job->curPos; i--) {
2330 	    if (job->outBuf[i] == '\n') {
2331 		gotNL = TRUE;
2332 		break;
2333 	    } else if (job->outBuf[i] == '\0') {
2334 		/*
2335 		 * Why?
2336 		 */
2337 		job->outBuf[i] = ' ';
2338 	    }
2339 	}
2340 
2341 	if (!gotNL) {
2342 	    job->curPos += nr;
2343 	    if (job->curPos == JOB_BUFSIZE) {
2344 		/*
2345 		 * If we've run out of buffer space, we have no choice
2346 		 * but to print the stuff. sigh.
2347 		 */
2348 		fbuf = TRUE;
2349 		i = job->curPos;
2350 	    }
2351 	}
2352 	if (gotNL || fbuf) {
2353 	    /*
2354 	     * Need to send the output to the screen. Null terminate it
2355 	     * first, overwriting the newline character if there was one.
2356 	     * So long as the line isn't one we should filter (according
2357 	     * to the shell description), we print the line, preceded
2358 	     * by a target banner if this target isn't the same as the
2359 	     * one for which we last printed something.
2360 	     * The rest of the data in the buffer are then shifted down
2361 	     * to the start of the buffer and curPos is set accordingly.
2362 	     */
2363 	    job->outBuf[i] = '\0';
2364 	    if (i >= job->curPos) {
2365 		char *cp;
2366 
2367 		cp = JobOutput(job, job->outBuf, &job->outBuf[i], FALSE);
2368 
2369 		/*
2370 		 * There's still more in that thar buffer. This time, though,
2371 		 * we know there's no newline at the end, so we add one of
2372 		 * our own free will.
2373 		 */
2374 		if (*cp != '\0') {
2375 		    if (job->node != lastNode) {
2376 			MESSAGE(stdout, job->node);
2377 			lastNode = job->node;
2378 		    }
2379 		    (void) fprintf(stdout, "%s%s", cp, gotNL ? "\n" : "");
2380 		    (void) fflush(stdout);
2381 		}
2382 	    }
2383 	    if (i < max - 1) {
2384 		/* shift the remaining characters down */
2385 		(void) memcpy(job->outBuf, &job->outBuf[i + 1], max - (i + 1));
2386 		job->curPos = max - (i + 1);
2387 
2388 	    } else {
2389 		/*
2390 		 * We have written everything out, so we just start over
2391 		 * from the start of the buffer. No copying. No nothing.
2392 		 */
2393 		job->curPos = 0;
2394 	    }
2395 	}
2396 	if (finish) {
2397 	    /*
2398 	     * If the finish flag is true, we must loop until we hit
2399 	     * end-of-file on the pipe. This is guaranteed to happen
2400 	     * eventually since the other end of the pipe is now closed
2401 	     * (we closed it explicitly and the child has exited). When
2402 	     * we do get an EOF, finish will be set FALSE and we'll fall
2403 	     * through and out.
2404 	     */
2405 	    goto end_loop;
2406 	}
2407     } else {
2408 	/*
2409 	 * We've been called to retrieve the output of the job from the
2410 	 * temporary file where it's been squirreled away. This consists of
2411 	 * opening the file, reading the output line by line, being sure not
2412 	 * to print the noPrint line for the shell we used, then close and
2413 	 * remove the temporary file. Very simple.
2414 	 *
2415 	 * Change to read in blocks and do FindSubString type things as for
2416 	 * pipes? That would allow for "@echo -n..."
2417 	 */
2418 	oFILE = fopen(job->outFile, "r");
2419 	if (oFILE != NULL) {
2420 	    (void) fprintf(stdout, "Results of making %s:\n", job->node->name);
2421 	    (void) fflush(stdout);
2422 	    while (fgets(inLine, sizeof(inLine), oFILE) != NULL) {
2423 		char	*cp, *endp, *oendp;
2424 
2425 		cp = inLine;
2426 		oendp = endp = inLine + strlen(inLine);
2427 		if (endp[-1] == '\n') {
2428 		    *--endp = '\0';
2429 		}
2430 		cp = JobOutput(job, inLine, endp, FALSE);
2431 
2432 		/*
2433 		 * There's still more in that thar buffer. This time, though,
2434 		 * we know there's no newline at the end, so we add one of
2435 		 * our own free will.
2436 		 */
2437 		(void) fprintf(stdout, "%s", cp);
2438 		(void) fflush(stdout);
2439 		if (endp != oendp) {
2440 		    (void) fprintf(stdout, "\n");
2441 		    (void) fflush(stdout);
2442 		}
2443 	    }
2444 	    (void) fclose(oFILE);
2445 	    (void) eunlink(job->outFile);
2446 	} else {
2447 	    Punt("Cannot open `%s'", job->outFile);
2448 	}
2449     }
2450 }
2451 
2452 /*-
2453  *-----------------------------------------------------------------------
2454  * Job_CatchChildren --
2455  *	Handle the exit of a child. Called from Make_Make.
2456  *
2457  * Input:
2458  *	block		TRUE if should block on the wait
2459  *
2460  * Results:
2461  *	none.
2462  *
2463  * Side Effects:
2464  *	The job descriptor is removed from the list of children.
2465  *
2466  * Notes:
2467  *	We do waits, blocking or not, according to the wisdom of our
2468  *	caller, until there are no more children to report. For each
2469  *	job, call JobFinish to finish things off. This will take care of
2470  *	putting jobs on the stoppedJobs queue.
2471  *
2472  *-----------------------------------------------------------------------
2473  */
2474 void
2475 Job_CatchChildren(Boolean block)
2476 {
2477     int    	  pid;	    	/* pid of dead child */
2478     Job		  *job;	    	/* job descriptor for dead child */
2479     LstNode       jnode;    	/* list element for finding job */
2480     int	  	  status;   	/* Exit/termination status */
2481 
2482     /*
2483      * Don't even bother if we know there's no one around.
2484      */
2485     if (nLocal == 0) {
2486 	return;
2487     }
2488 
2489     while ((pid = waitpid((pid_t) -1, &status,
2490 			  (block?0:WNOHANG)|WUNTRACED)) > 0)
2491     {
2492 	if (DEBUG(JOB)) {
2493 	    (void) fprintf(stdout, "Process %d exited or stopped %x.\n", pid,
2494 	      status);
2495 	    (void) fflush(stdout);
2496 	}
2497 
2498 	jnode = Lst_Find(jobs, (ClientData)&pid, JobCmpPid);
2499 	if (jnode == NILLNODE) {
2500 	    if (WIFSTOPPED(status) && (WSTOPSIG(status) == SIGCONT)) {
2501 		jnode = Lst_Find(stoppedJobs, (ClientData) &pid, JobCmpPid);
2502 		if (jnode == NILLNODE) {
2503 		    Error("Resumed child (%d) not in table", pid);
2504 		    continue;
2505 		}
2506 		job = (Job *)Lst_Datum(jnode);
2507 		(void) Lst_Remove(stoppedJobs, jnode);
2508 	    } else {
2509 		Error("Child (%d) not in table?", pid);
2510 		continue;
2511 	    }
2512 	} else {
2513 	    job = (Job *) Lst_Datum(jnode);
2514 	    (void) Lst_Remove(jobs, jnode);
2515 	    nJobs -= 1;
2516 #ifdef REMOTE
2517 	    if (!(job->flags & JOB_REMOTE)) {
2518 		if (DEBUG(JOB)) {
2519 		    (void) fprintf(stdout,
2520 			   "Job queue has one fewer local process.\n");
2521 		    (void) fflush(stdout);
2522 		}
2523 		nLocal -= 1;
2524 	    }
2525 #else
2526 	    nLocal -= 1;
2527 #endif
2528 	}
2529 
2530 	JobFinish(job, &status);
2531     }
2532 }
2533 
2534 /*-
2535  *-----------------------------------------------------------------------
2536  * Job_CatchOutput --
2537  *	Catch the output from our children, if we're using
2538  *	pipes do so. Otherwise just block time until we get a
2539  *	signal (most likely a SIGCHLD) since there's no point in
2540  *	just spinning when there's nothing to do and the reaping
2541  *	of a child can wait for a while.
2542  *
2543  * Results:
2544  *	None
2545  *
2546  * Side Effects:
2547  *	Output is read from pipes if we're piping.
2548  * -----------------------------------------------------------------------
2549  */
2550 void
2551 Job_CatchOutput(void)
2552 {
2553     int           	  nready;
2554     LstNode		  ln;
2555     Job  	 	  *job;
2556 #ifdef RMT_WILL_WATCH
2557     int	    	  	  pnJobs;   	/* Previous nJobs */
2558 #endif
2559 
2560     (void) fflush(stdout);
2561     Job_TokenFlush();
2562 #ifdef RMT_WILL_WATCH
2563     pnJobs = nJobs;
2564 
2565     /*
2566      * It is possible for us to be called with nJobs equal to 0. This happens
2567      * if all the jobs finish and a job that is stopped cannot be run
2568      * locally (eg if maxLocal is 0) and cannot be exported. The job will
2569      * be placed back on the stoppedJobs queue, Job_Empty() will return false,
2570      * Make_Run will call us again when there's nothing for which to wait.
2571      * nJobs never changes, so we loop forever. Hence the check. It could
2572      * be argued that we should sleep for a bit so as not to swamp the
2573      * exportation system with requests. Perhaps we should.
2574      *
2575      * NOTE: IT IS THE RESPONSIBILITY OF Rmt_Wait TO CALL Job_CatchChildren
2576      * IN A TIMELY FASHION TO CATCH ANY LOCALLY RUNNING JOBS THAT EXIT.
2577      * It may use the variable nLocal to determine if it needs to call
2578      * Job_CatchChildren (if nLocal is 0, there's nothing for which to
2579      * wait...)
2580      */
2581     while (nJobs != 0 && pnJobs == nJobs) {
2582 	Rmt_Wait();
2583     }
2584 #else
2585     if (usePipes) {
2586 	if ((nready = poll((wantToken ? fds : (fds + 1)),
2587 	  		   (wantToken ? nfds : (nfds - 1)), POLL_MSEC)) <= 0) {
2588 	    return;
2589 	} else {
2590 	    sigset_t	mask;
2591 	    JobSigLock(&mask);
2592 	    if (Lst_Open(jobs) == FAILURE) {
2593 		Punt("Cannot open job table");
2594 	    }
2595 
2596 	    if (readyfd(&childExitJob)) {
2597 		char token;
2598 		(void) read(childExitJob.inPipe, &token, 1);
2599 		nready -= 1;
2600 	    }
2601 
2602 	    while (nready && (ln = Lst_Next(jobs)) != NILLNODE) {
2603 		job = (Job *) Lst_Datum(ln);
2604 		if (readyfd(job)) {
2605 		    JobDoOutput(job, FALSE);
2606 		    nready -= 1;
2607 		}
2608 	    }
2609 	    Lst_Close(jobs);
2610 	    JobSigUnlock(&mask);
2611 	}
2612     }
2613 #endif /* RMT_WILL_WATCH */
2614 }
2615 
2616 /*-
2617  *-----------------------------------------------------------------------
2618  * Job_Make --
2619  *	Start the creation of a target. Basically a front-end for
2620  *	JobStart used by the Make module.
2621  *
2622  * Results:
2623  *	None.
2624  *
2625  * Side Effects:
2626  *	Another job is started.
2627  *
2628  *-----------------------------------------------------------------------
2629  */
2630 void
2631 Job_Make(GNode *gn)
2632 {
2633     (void) JobStart(gn, 0, NULL);
2634 }
2635 
2636 void
2637 Shell_Init()
2638 {
2639     if (shellPath == NULL) {
2640 	/*
2641 	 * The user didn't specify a shell to use, so we are using the
2642 	 * default one... Both the absolute path and the last component
2643 	 * must be set. The last component is taken from the 'name' field
2644 	 * of the default shell description pointed-to by commandShell.
2645 	 * All default shells are located in _PATH_DEFSHELLDIR.
2646 	 */
2647 	shellName = commandShell->name;
2648 	shellPath = str_concat(_PATH_DEFSHELLDIR, shellName, STR_ADDSLASH);
2649     }
2650     if (commandShell->exit == NULL) {
2651 	commandShell->exit = "";
2652     }
2653     if (commandShell->echo == NULL) {
2654 	commandShell->echo = "";
2655     }
2656 }
2657 
2658 /*-
2659  *-----------------------------------------------------------------------
2660  * Job_Init --
2661  *	Initialize the process module
2662  *
2663  * Input:
2664  *	maxproc		the greatest number of jobs which may be running
2665  *			at one time
2666  *	maxlocal	the greatest number of jobs which may be running
2667  *			at once
2668  *
2669  * Results:
2670  *	none
2671  *
2672  * Side Effects:
2673  *	lists and counters are initialized
2674  *-----------------------------------------------------------------------
2675  */
2676 void
2677 Job_Init(int maxproc, int maxlocal)
2678 {
2679     GNode         *begin;     /* node for commands to do at the very start */
2680 
2681     jobs =  	  Lst_Init(FALSE);
2682     stoppedJobs = Lst_Init(FALSE);
2683     maxJobs = 	  maxproc;
2684     maxLocal = 	  maxlocal;
2685     nJobs = 	  0;
2686     nLocal = 	  0;
2687     wantToken =	  FALSE;
2688 
2689     aborting = 	  0;
2690     errors = 	  0;
2691 
2692     lastNode =	  NILGNODE;
2693 
2694     if (maxJobs == 1
2695 #ifdef REMOTE
2696 	|| noMessages
2697 #endif
2698 		     ) {
2699 	/*
2700 	 * If only one job can run at a time, there's no need for a banner,
2701 	 * is there?
2702 	 */
2703 	targFmt = "";
2704     } else {
2705 	targFmt = TARG_FMT;
2706     }
2707 
2708     Shell_Init();
2709 
2710     if (pipe(exit_pipe) < 0)
2711 	Fatal("error in pipe: %s", strerror(errno));
2712     fcntl(exit_pipe[0], F_SETFD, 1);
2713     fcntl(exit_pipe[1], F_SETFD, 1);
2714 
2715     childExitJob.inPipe = exit_pipe[0];
2716 
2717     sigemptyset(&caught_signals);
2718     /*
2719      * Install a SIGCHLD handler.
2720      */
2721     (void)signal(SIGCHLD, JobChildSig);
2722     sigaddset(&caught_signals, SIGCHLD);
2723 
2724 #define ADDSIG(s,h)				\
2725     if (signal(s, SIG_IGN) != SIG_IGN) {	\
2726 	sigaddset(&caught_signals, s);		\
2727 	(void) signal(s, h);			\
2728     }
2729 
2730     /*
2731      * Catch the four signals that POSIX specifies if they aren't ignored.
2732      * JobPassSig will take care of calling JobInterrupt if appropriate.
2733      */
2734     ADDSIG(SIGINT, JobPassSig)
2735     ADDSIG(SIGHUP, JobPassSig)
2736     ADDSIG(SIGTERM, JobPassSig)
2737     ADDSIG(SIGQUIT, JobPassSig)
2738 
2739     /*
2740      * There are additional signals that need to be caught and passed if
2741      * either the export system wants to be told directly of signals or if
2742      * we're giving each job its own process group (since then it won't get
2743      * signals from the terminal driver as we own the terminal)
2744      */
2745 #if defined(RMT_WANTS_SIGNALS) || defined(USE_PGRP)
2746     ADDSIG(SIGTSTP, JobPassSig)
2747     ADDSIG(SIGTTOU, JobPassSig)
2748     ADDSIG(SIGTTIN, JobPassSig)
2749     ADDSIG(SIGWINCH, JobPassSig)
2750     ADDSIG(SIGCONT, JobContinueSig)
2751 #endif
2752 #undef ADDSIG
2753 
2754     begin = Targ_FindNode(".BEGIN", TARG_NOCREATE);
2755 
2756     if (begin != NILGNODE) {
2757 	JobStart(begin, JOB_SPECIAL, (Job *)0);
2758 	while (nJobs) {
2759 	    Job_CatchOutput();
2760 #ifndef RMT_WILL_WATCH
2761 	    Job_CatchChildren(!usePipes);
2762 #endif /* RMT_WILL_WATCH */
2763 	}
2764     }
2765     postCommands = Targ_FindNode(".END", TARG_CREATE);
2766 }
2767 
2768 static void JobSigReset(void)
2769 {
2770 #define DELSIG(s)					\
2771     if (sigismember(&caught_signals, s)) {		\
2772 	(void) signal(SIGINT, SIG_DFL);			\
2773     }
2774 
2775     DELSIG(SIGINT)
2776     DELSIG(SIGHUP)
2777     DELSIG(SIGQUIT)
2778     DELSIG(SIGTERM)
2779 #if defined(RMT_WANTS_SIGNALS) || defined(USE_PGRP)
2780     DELSIG(SIGTSTP)
2781     DELSIG(SIGTTOU)
2782     DELSIG(SIGTTIN)
2783     DELSIG(SIGWINCH)
2784     DELSIG(SIGCONT)
2785 #endif
2786 #undef DELSIG
2787     (void)signal(SIGCHLD, SIG_DFL);
2788 }
2789 
2790 /*-
2791  *-----------------------------------------------------------------------
2792  * Job_Empty --
2793  *	See if the job table is empty.  Because the local concurrency may
2794  *	be set to 0, it is possible for the job table to become empty,
2795  *	while the list of stoppedJobs remains non-empty. In such a case,
2796  *	we want to restart as many jobs as we can.
2797  *
2798  * Results:
2799  *	TRUE if it is. FALSE if it ain't.
2800  *
2801  * Side Effects:
2802  *	None.
2803  *
2804  * -----------------------------------------------------------------------
2805  */
2806 Boolean
2807 Job_Empty(void)
2808 {
2809     if (nJobs == 0) {
2810 	if (!Lst_IsEmpty(stoppedJobs) && !aborting) {
2811 	    /*
2812 	     * The job table is obviously not full if it has no jobs in
2813 	     * it...Try and restart the stopped jobs.
2814 	     */
2815 	    JobRestartJobs();
2816 	    return(FALSE);
2817 	} else {
2818 	    return(TRUE);
2819 	}
2820     } else {
2821 	return(FALSE);
2822     }
2823 }
2824 
2825 /*-
2826  *-----------------------------------------------------------------------
2827  * JobMatchShell --
2828  *	Find a shell in 'shells' given its name.
2829  *
2830  * Results:
2831  *	A pointer to the Shell structure.
2832  *
2833  * Side Effects:
2834  *	None.
2835  *
2836  *-----------------------------------------------------------------------
2837  */
2838 static Shell *
2839 JobMatchShell(const char *name)
2840 {
2841     Shell	*sh;
2842 
2843     for (sh = shells; sh->name != NULL; sh++) {
2844 	if (strcmp(name, sh->name) == 0)
2845 		return (sh);
2846     }
2847     return (NULL);
2848 }
2849 
2850 /*-
2851  *-----------------------------------------------------------------------
2852  * Job_ParseShell --
2853  *	Parse a shell specification and set up commandShell, shellPath
2854  *	and shellName appropriately.
2855  *
2856  * Input:
2857  *	line		The shell spec
2858  *
2859  * Results:
2860  *	FAILURE if the specification was incorrect.
2861  *
2862  * Side Effects:
2863  *	commandShell points to a Shell structure (either predefined or
2864  *	created from the shell spec), shellPath is the full path of the
2865  *	shell described by commandShell, while shellName is just the
2866  *	final component of shellPath.
2867  *
2868  * Notes:
2869  *	A shell specification consists of a .SHELL target, with dependency
2870  *	operator, followed by a series of blank-separated words. Double
2871  *	quotes can be used to use blanks in words. A backslash escapes
2872  *	anything (most notably a double-quote and a space) and
2873  *	provides the functionality it does in C. Each word consists of
2874  *	keyword and value separated by an equal sign. There should be no
2875  *	unnecessary spaces in the word. The keywords are as follows:
2876  *	    name  	    Name of shell.
2877  *	    path  	    Location of shell.
2878  *	    quiet 	    Command to turn off echoing.
2879  *	    echo  	    Command to turn echoing on
2880  *	    filter	    Result of turning off echoing that shouldn't be
2881  *	    	  	    printed.
2882  *	    echoFlag	    Flag to turn echoing on at the start
2883  *	    errFlag	    Flag to turn error checking on at the start
2884  *	    hasErrCtl	    True if shell has error checking control
2885  *	    check 	    Command to turn on error checking if hasErrCtl
2886  *	    	  	    is TRUE or template of command to echo a command
2887  *	    	  	    for which error checking is off if hasErrCtl is
2888  *	    	  	    FALSE.
2889  *	    ignore	    Command to turn off error checking if hasErrCtl
2890  *	    	  	    is TRUE or template of command to execute a
2891  *	    	  	    command so as to ignore any errors it returns if
2892  *	    	  	    hasErrCtl is FALSE.
2893  *
2894  *-----------------------------------------------------------------------
2895  */
2896 ReturnStatus
2897 Job_ParseShell(char *line)
2898 {
2899     char	**words;
2900     char	**argv;
2901     int		argc;
2902     char	*path;
2903     Shell	newShell;
2904     Boolean	fullSpec = FALSE;
2905     Shell	*sh;
2906 
2907     while (isspace((unsigned char)*line)) {
2908 	line++;
2909     }
2910 
2911     if (shellArgv)
2912 	free(UNCONST(shellArgv));
2913 
2914     memset((Address)&newShell, 0, sizeof(newShell));
2915 
2916     /*
2917      * Parse the specification by keyword
2918      */
2919     words = brk_string(line, &argc, TRUE, &path);
2920     shellArgv = path;
2921 
2922     for (path = NULL, argv = words; argc != 0; argc--, argv++) {
2923 	    if (strncmp(*argv, "path=", 5) == 0) {
2924 		path = &argv[0][5];
2925 	    } else if (strncmp(*argv, "name=", 5) == 0) {
2926 		newShell.name = &argv[0][5];
2927 	    } else {
2928 		if (strncmp(*argv, "quiet=", 6) == 0) {
2929 		    newShell.echoOff = &argv[0][6];
2930 		} else if (strncmp(*argv, "echo=", 5) == 0) {
2931 		    newShell.echoOn = &argv[0][5];
2932 		} else if (strncmp(*argv, "filter=", 7) == 0) {
2933 		    newShell.noPrint = &argv[0][7];
2934 		    newShell.noPLen = strlen(newShell.noPrint);
2935 		} else if (strncmp(*argv, "echoFlag=", 9) == 0) {
2936 		    newShell.echo = &argv[0][9];
2937 		} else if (strncmp(*argv, "errFlag=", 8) == 0) {
2938 		    newShell.exit = &argv[0][8];
2939 		} else if (strncmp(*argv, "hasErrCtl=", 10) == 0) {
2940 		    char c = argv[0][10];
2941 		    newShell.hasErrCtl = !((c != 'Y') && (c != 'y') &&
2942 					   (c != 'T') && (c != 't'));
2943 		} else if (strncmp(*argv, "check=", 6) == 0) {
2944 		    newShell.errCheck = &argv[0][6];
2945 		} else if (strncmp(*argv, "ignore=", 7) == 0) {
2946 		    newShell.ignErr = &argv[0][7];
2947 		} else if (strncmp(*argv, "errout=", 7) == 0) {
2948 		    newShell.errOut = &argv[0][7];
2949 		} else if (strncmp(*argv, "comment=", 8) == 0) {
2950 		    newShell.commentChar = argv[0][8];
2951 		} else {
2952 		    Parse_Error(PARSE_FATAL, "Unknown keyword \"%s\"",
2953 				*argv);
2954 		    free(words);
2955 		    return(FAILURE);
2956 		}
2957 		fullSpec = TRUE;
2958 	    }
2959     }
2960 
2961     if (path == NULL) {
2962 	/*
2963 	 * If no path was given, the user wants one of the pre-defined shells,
2964 	 * yes? So we find the one s/he wants with the help of JobMatchShell
2965 	 * and set things up the right way. shellPath will be set up by
2966 	 * Job_Init.
2967 	 */
2968 	if (newShell.name == NULL) {
2969 	    Parse_Error(PARSE_FATAL, "Neither path nor name specified");
2970 	    free(words);
2971 	    return(FAILURE);
2972 	} else {
2973 	    if ((sh = JobMatchShell(newShell.name)) == NULL) {
2974 		    Parse_Error(PARSE_WARNING, "%s: No matching shell",
2975 				newShell.name);
2976 		    free(words);
2977 		    return(FAILURE);
2978 	    }
2979 	    commandShell = sh;
2980 	    shellName = newShell.name;
2981 	}
2982     } else {
2983 	/*
2984 	 * The user provided a path. If s/he gave nothing else (fullSpec is
2985 	 * FALSE), try and find a matching shell in the ones we know of.
2986 	 * Else we just take the specification at its word and copy it
2987 	 * to a new location. In either case, we need to record the
2988 	 * path the user gave for the shell.
2989 	 */
2990 	shellPath = path;
2991 	path = strrchr(path, '/');
2992 	if (path == NULL) {
2993 	    path = UNCONST(shellPath);
2994 	} else {
2995 	    path += 1;
2996 	}
2997 	if (newShell.name != NULL) {
2998 	    shellName = newShell.name;
2999 	} else {
3000 	    shellName = path;
3001 	}
3002 	if (!fullSpec) {
3003 	    if ((sh = JobMatchShell(shellName)) == NULL) {
3004 		    Parse_Error(PARSE_WARNING, "%s: No matching shell",
3005 				shellName);
3006 		    free(words);
3007 		    return(FAILURE);
3008 	    }
3009 	    commandShell = sh;
3010 	} else {
3011 	    commandShell = (Shell *) emalloc(sizeof(Shell));
3012 	    *commandShell = newShell;
3013 	}
3014     }
3015 
3016     if (commandShell->echoOn && commandShell->echoOff) {
3017 	commandShell->hasEchoCtl = TRUE;
3018     }
3019 
3020     if (!commandShell->hasErrCtl) {
3021 	if (commandShell->errCheck == NULL) {
3022 	    commandShell->errCheck = "";
3023 	}
3024 	if (commandShell->ignErr == NULL) {
3025 	    commandShell->ignErr = "%s\n";
3026 	}
3027     }
3028 
3029     /*
3030      * Do not free up the words themselves, since they might be in use by the
3031      * shell specification.
3032      */
3033     free(words);
3034     return SUCCESS;
3035 }
3036 
3037 /*-
3038  *-----------------------------------------------------------------------
3039  * JobInterrupt --
3040  *	Handle the receipt of an interrupt.
3041  *
3042  * Input:
3043  *	runINTERRUPT	Non-zero if commands for the .INTERRUPT target
3044  *			should be executed
3045  *	signo		signal received
3046  *
3047  * Results:
3048  *	None
3049  *
3050  * Side Effects:
3051  *	All children are killed. Another job will be started if the
3052  *	.INTERRUPT target was given.
3053  *-----------------------------------------------------------------------
3054  */
3055 static void
3056 JobInterrupt(int runINTERRUPT, int signo)
3057 {
3058     LstNode	ln;		/* element in job table */
3059     Job		*job;		/* job descriptor in that element */
3060     GNode	*interrupt;	/* the node describing the .INTERRUPT target */
3061     sigset_t	mask;
3062 
3063     aborting = ABORT_INTERRUPT;
3064 
3065     JobSigLock(&mask);
3066 
3067     (void) Lst_Open(jobs);
3068     while ((ln = Lst_Next(jobs)) != NILLNODE) {
3069 	GNode *gn;
3070 
3071 	job = (Job *) Lst_Datum(ln);
3072 	gn = job->node;
3073 
3074 	if ((gn->type & (OP_JOIN|OP_PHONY)) == 0 && !Targ_Precious(gn)) {
3075 	    char *file = (gn->path == NULL ? gn->name : gn->path);
3076 	    if (!noExecute && eunlink(file) != -1) {
3077 		Error("*** %s removed", file);
3078 	    }
3079 	}
3080 #ifdef RMT_WANTS_SIGNALS
3081 	if (job->flags & JOB_REMOTE) {
3082 	    /*
3083 	     * If job is remote, let the Rmt module do the killing.
3084 	     */
3085 	    if (!Rmt_Signal(job, signo)) {
3086 		/*
3087 		 * If couldn't kill the thing, finish it out now with an
3088 		 * error code, since no exit report will come in likely.
3089 		 */
3090 		int status;
3091 
3092 		status.w_status = 0;
3093 		status.w_retcode = 1;
3094 		JobFinish(job, &status);
3095 	    }
3096 	} else if (job->pid) {
3097 	    KILL(job->pid, signo);
3098 	}
3099 #else
3100 	if (job->pid) {
3101 	    if (DEBUG(JOB)) {
3102 		(void) fprintf(stdout,
3103 			   "JobInterrupt passing signal %d to child %d.\n",
3104 			   signo, job->pid);
3105 		(void) fflush(stdout);
3106 	    }
3107 	    KILL(job->pid, signo);
3108 	}
3109 #endif /* RMT_WANTS_SIGNALS */
3110     }
3111     Lst_Close(jobs);
3112 
3113 #ifdef REMOTE
3114    (void)Lst_Open(stoppedJobs);
3115     while ((ln = Lst_Next(stoppedJobs)) != NILLNODE) {
3116 	GNode *gn;
3117 
3118 	job = (Job *) Lst_Datum(ln);
3119 	gn = job->node;
3120 
3121 	if (job->flags & JOB_RESTART) {
3122 	    if (DEBUG(JOB)) {
3123 		(void) fprintf(stdout, "%s%s",
3124 			       "JobInterrupt skipping job on stopped queue",
3125 			       "-- it was waiting to be restarted.\n");
3126 		(void) fflush(stdout);
3127 	    }
3128 	    continue;
3129 	}
3130 	if ((gn->type & (OP_JOIN|OP_PHONY)) == 0 && !Targ_Precious(gn)) {
3131 	    char *file = (gn->path == NULL ? gn->name : gn->path);
3132 	    if (eunlink(file) == 0) {
3133 		Error("*** %s removed", file);
3134 	    }
3135 	}
3136 	/*
3137 	 * Resume the thing so it will take the signal.
3138 	 */
3139 	if (DEBUG(JOB)) {
3140 	    (void) fprintf(stdout,
3141 			   "JobInterrupt passing CONT to stopped child %d.\n",
3142 			   job->pid);
3143 	    (void) fflush(stdout);
3144 	}
3145 	KILL(job->pid, SIGCONT);
3146 #ifdef RMT_WANTS_SIGNALS
3147 	if (job->flags & JOB_REMOTE) {
3148 	    /*
3149 	     * If job is remote, let the Rmt module do the killing.
3150 	     */
3151 	    if (!Rmt_Signal(job, SIGINT)) {
3152 		/*
3153 		 * If couldn't kill the thing, finish it out now with an
3154 		 * error code, since no exit report will come in likely.
3155 		 */
3156 		int status;
3157 		status.w_status = 0;
3158 		status.w_retcode = 1;
3159 		JobFinish(job, &status);
3160 	    }
3161 	} else if (job->pid) {
3162 	    if (DEBUG(JOB)) {
3163 		(void) fprintf(stdout,
3164 		       "JobInterrupt passing interrupt to stopped child %d.\n",
3165 			       job->pid);
3166 		(void) fflush(stdout);
3167 	    }
3168 	    KILL(job->pid, SIGINT);
3169 	}
3170 #endif /* RMT_WANTS_SIGNALS */
3171     }
3172     Lst_Close(stoppedJobs);
3173 #endif /* REMOTE */
3174 
3175     JobSigUnlock(&mask);
3176 
3177     if (runINTERRUPT && !touchFlag) {
3178 	interrupt = Targ_FindNode(".INTERRUPT", TARG_NOCREATE);
3179 	if (interrupt != NILGNODE) {
3180 	    ignoreErrors = FALSE;
3181 
3182 	    JobStart(interrupt, JOB_IGNDOTS, (Job *)0);
3183 	    while (nJobs) {
3184 		Job_CatchOutput();
3185 #ifndef RMT_WILL_WATCH
3186 		Job_CatchChildren(!usePipes);
3187 #endif /* RMT_WILL_WATCH */
3188 	    }
3189 	}
3190     }
3191     Trace_Log(MAKEINTR, 0);
3192     exit(signo);
3193 }
3194 
3195 /*
3196  *-----------------------------------------------------------------------
3197  * Job_Finish --
3198  *	Do final processing such as the running of the commands
3199  *	attached to the .END target.
3200  *
3201  * Results:
3202  *	Number of errors reported.
3203  *
3204  * Side Effects:
3205  *	None.
3206  *-----------------------------------------------------------------------
3207  */
3208 int
3209 Job_Finish(void)
3210 {
3211     if (postCommands != NILGNODE && !Lst_IsEmpty(postCommands->commands)) {
3212 	if (errors) {
3213 	    Error("Errors reported so .END ignored");
3214 	} else {
3215 	    JobStart(postCommands, JOB_SPECIAL | JOB_IGNDOTS, NULL);
3216 
3217 	    while (nJobs) {
3218 		Job_CatchOutput();
3219 #ifndef RMT_WILL_WATCH
3220 		Job_CatchChildren(!usePipes);
3221 #endif /* RMT_WILL_WATCH */
3222 	    }
3223 	}
3224     }
3225     Job_TokenFlush();
3226     return(errors);
3227 }
3228 
3229 /*-
3230  *-----------------------------------------------------------------------
3231  * Job_End --
3232  *	Cleanup any memory used by the jobs module
3233  *
3234  * Results:
3235  *	None.
3236  *
3237  * Side Effects:
3238  *	Memory is freed
3239  *-----------------------------------------------------------------------
3240  */
3241 void
3242 Job_End(void)
3243 {
3244 #ifdef CLEANUP
3245     if (shellArgv)
3246 	free(shellArgv);
3247 #endif
3248 }
3249 
3250 /*-
3251  *-----------------------------------------------------------------------
3252  * Job_Wait --
3253  *	Waits for all running jobs to finish and returns. Sets 'aborting'
3254  *	to ABORT_WAIT to prevent other jobs from starting.
3255  *
3256  * Results:
3257  *	None.
3258  *
3259  * Side Effects:
3260  *	Currently running jobs finish.
3261  *
3262  *-----------------------------------------------------------------------
3263  */
3264 void
3265 Job_Wait(void)
3266 {
3267     aborting = ABORT_WAIT;
3268     while (nJobs != 0) {
3269 	Job_CatchOutput();
3270 #ifndef RMT_WILL_WATCH
3271 	Job_CatchChildren(!usePipes);
3272 #endif /* RMT_WILL_WATCH */
3273     }
3274     Job_TokenFlush();
3275     aborting = 0;
3276 }
3277 
3278 /*-
3279  *-----------------------------------------------------------------------
3280  * Job_AbortAll --
3281  *	Abort all currently running jobs without handling output or anything.
3282  *	This function is to be called only in the event of a major
3283  *	error. Most definitely NOT to be called from JobInterrupt.
3284  *
3285  * Results:
3286  *	None
3287  *
3288  * Side Effects:
3289  *	All children are killed, not just the firstborn
3290  *-----------------------------------------------------------------------
3291  */
3292 void
3293 Job_AbortAll(void)
3294 {
3295     LstNode	ln;	/* element in job table */
3296     Job		*job;	/* the job descriptor in that element */
3297     int		foo;
3298     sigset_t	mask;
3299 
3300     aborting = ABORT_ERROR;
3301 
3302     if (nJobs) {
3303 
3304 	JobSigLock(&mask);
3305 	(void) Lst_Open(jobs);
3306 	while ((ln = Lst_Next(jobs)) != NILLNODE) {
3307 	    job = (Job *) Lst_Datum(ln);
3308 
3309 	    /*
3310 	     * kill the child process with increasingly drastic signals to make
3311 	     * darn sure it's dead.
3312 	     */
3313 #ifdef RMT_WANTS_SIGNALS
3314 	    if (job->flags & JOB_REMOTE) {
3315 		Rmt_Signal(job, SIGINT);
3316 		Rmt_Signal(job, SIGKILL);
3317 	    } else {
3318 		KILL(job->pid, SIGINT);
3319 		KILL(job->pid, SIGKILL);
3320 	    }
3321 #else
3322 	    KILL(job->pid, SIGINT);
3323 	    KILL(job->pid, SIGKILL);
3324 #endif /* RMT_WANTS_SIGNALS */
3325 	}
3326 	Lst_Close(jobs);
3327 	JobSigUnlock(&mask);
3328     }
3329 
3330     /*
3331      * Catch as many children as want to report in at first, then give up
3332      */
3333     while (waitpid((pid_t) -1, &foo, WNOHANG) > 0)
3334 	continue;
3335 }
3336 
3337 #ifdef REMOTE
3338 /*-
3339  *-----------------------------------------------------------------------
3340  * JobFlagForMigration --
3341  *	Handle the eviction of a child. Called from RmtStatusChange.
3342  *	Flags the child as remigratable and then suspends it.
3343  *
3344  * Input:
3345  *	hostID		ID of host we used, for matching children
3346  *
3347  * Results:
3348  *	none.
3349  *
3350  * Side Effects:
3351  *	The job descriptor is flagged for remigration.
3352  *
3353  *-----------------------------------------------------------------------
3354  */
3355 void
3356 JobFlagForMigration(int hostID)
3357 {
3358     Job		  *job;	    	/* job descriptor for dead child */
3359     LstNode       jnode;    	/* list element for finding job */
3360 
3361     if (DEBUG(JOB)) {
3362 	(void) fprintf(stdout, "JobFlagForMigration(%d) called.\n", hostID);
3363 	(void) fflush(stdout);
3364     }
3365     jnode = Lst_Find(jobs, (ClientData)&hostID, JobCmpRmtID);
3366 
3367     if (jnode == NILLNODE) {
3368 	jnode = Lst_Find(stoppedJobs, (ClientData)hostID, JobCmpRmtID);
3369 		if (jnode == NILLNODE) {
3370 		    if (DEBUG(JOB)) {
3371 			Error("Evicting host(%d) not in table", hostID);
3372 		    }
3373 		    return;
3374 		}
3375     }
3376     job = (Job *) Lst_Datum(jnode);
3377 
3378     if (DEBUG(JOB)) {
3379 	(void) fprintf(stdout,
3380 		       "JobFlagForMigration(%d) found job '%s'.\n", hostID,
3381 		       job->node->name);
3382 	(void) fflush(stdout);
3383     }
3384 
3385     KILL(job->pid, SIGSTOP);
3386 
3387     job->flags |= JOB_REMIGRATE;
3388 }
3389 
3390 #endif
3391 
3392 /*-
3393  *-----------------------------------------------------------------------
3394  * JobRestartJobs --
3395  *	Tries to restart stopped jobs if there are slots available.
3396  *	Note that this tries to restart them regardless of pending errors.
3397  *	It's not good to leave stopped jobs lying around!
3398  *
3399  * Results:
3400  *	None.
3401  *
3402  * Side Effects:
3403  *	Resumes(and possibly migrates) jobs.
3404  *
3405  *-----------------------------------------------------------------------
3406  */
3407 static void
3408 JobRestartJobs(void)
3409 {
3410     sigset_t	mask;
3411 
3412     JobSigLock(&mask);
3413     while (!Lst_IsEmpty(stoppedJobs)) {
3414 	if (DEBUG(JOB)) {
3415 	    (void) fprintf(stdout, "Restarting a stopped job.\n");
3416 	    (void) fflush(stdout);
3417 	}
3418 	if (JobRestart((Job *)Lst_DeQueue(stoppedJobs)) != 0)
3419 		break;
3420     }
3421     JobSigUnlock(&mask);
3422 }
3423 
3424 #ifndef RMT_WILL_WATCH
3425 static void
3426 watchfd(Job *job)
3427 {
3428     int i;
3429     if (job->inPollfd != NULL)
3430 	Punt("Watching watched job");
3431     if (fds == NULL) {
3432 	maxfds = JBSTART;
3433 	fds = emalloc(sizeof(struct pollfd) * maxfds);
3434 	jobfds = emalloc(sizeof(Job **) * maxfds);
3435 
3436 	fds[0].fd = job_pipe[0];
3437 	fds[0].events = POLLIN;
3438 	jobfds[0] = &tokenWaitJob;
3439 	tokenWaitJob.inPollfd = &fds[0];
3440 	nfds++;
3441 
3442 	fds[1].fd = exit_pipe[0];
3443 	fds[1].events = POLLIN;
3444 	jobfds[1] = &childExitJob;
3445 	childExitJob.inPollfd = &fds[1];
3446 	nfds++;
3447     } else if (nfds == maxfds) {
3448 	maxfds *= JBFACTOR;
3449 	fds = erealloc(fds, sizeof(struct pollfd) * maxfds);
3450 	jobfds = erealloc(jobfds, sizeof(Job **) * maxfds);
3451 	for (i = 0; i < nfds; i++)
3452 	    jobfds[i]->inPollfd = &fds[i];
3453     }
3454 
3455     fds[nfds].fd = job->inPipe;
3456     fds[nfds].events = POLLIN;
3457     jobfds[nfds] = job;
3458     job->inPollfd = &fds[nfds];
3459     nfds++;
3460 }
3461 
3462 static void
3463 clearfd(Job *job)
3464 {
3465     int i;
3466     if (job->inPollfd == NULL)
3467 	Punt("Unwatching unwatched job");
3468     i = job->inPollfd - fds;
3469     nfds--;
3470     /*
3471      * Move last job in table into hole made by dead job.
3472      */
3473     if (nfds != i) {
3474 	fds[i] = fds[nfds];
3475 	jobfds[i] = jobfds[nfds];
3476 	jobfds[i]->inPollfd = &fds[i];
3477     }
3478     job->inPollfd = NULL;
3479 }
3480 
3481 static int
3482 readyfd(Job *job)
3483 {
3484     if (job->inPollfd == NULL)
3485 	Punt("Polling unwatched job");
3486     return (job->inPollfd->revents & POLLIN) != 0;
3487 }
3488 #endif
3489 
3490 /*-
3491  *-----------------------------------------------------------------------
3492  * JobTokenAdd --
3493  *	Put a token into the job pipe so that some make process can start
3494  *	another job.
3495  *
3496  * Side Effects:
3497  *	Allows more build jobs to be spawned somewhere.
3498  *
3499  *-----------------------------------------------------------------------
3500  */
3501 
3502 static void
3503 JobTokenAdd(void)
3504 {
3505 
3506     if (DEBUG(JOB))
3507 	printf("deposit token\n");
3508     write(job_pipe[1], "+", 1);
3509 }
3510 
3511 /*-
3512  *-----------------------------------------------------------------------
3513  * Job_ServerStartTokenAdd --
3514  *	Prep the job token pipe in the root make process.
3515  *
3516  *-----------------------------------------------------------------------
3517  */
3518 
3519 void
3520 Job_ServerStart(int maxproc)
3521 {
3522     int i, flags;
3523     char jobarg[64];
3524 
3525     if (pipe(job_pipe) < 0)
3526 	Fatal ("error in pipe: %s", strerror(errno));
3527 
3528     /*
3529      * We mark the input side of the pipe non-blocking; we poll(2) the
3530      * pipe when we're waiting for a job token, but we might lose the
3531      * race for the token when a new one becomes available, so the read
3532      * from the pipe should not block.
3533      */
3534     flags = fcntl(job_pipe[0], F_GETFL, 0);
3535     flags |= O_NONBLOCK;
3536     fcntl(job_pipe[0], F_SETFL, flags);
3537 
3538     /*
3539      * Mark job pipes as close-on-exec.
3540      * Note that we will clear this when executing submakes.
3541      */
3542     fcntl(job_pipe[0], F_SETFD, 1);
3543     fcntl(job_pipe[1], F_SETFD, 1);
3544 
3545     snprintf(jobarg, sizeof(jobarg), "%d,%d", job_pipe[0], job_pipe[1]);
3546 
3547     Var_Append(MAKEFLAGS, "-J", VAR_GLOBAL);
3548     Var_Append(MAKEFLAGS, jobarg, VAR_GLOBAL);
3549 
3550     /*
3551      * Preload job_pipe with one token per job, save the one
3552      * "extra" token for the primary job.
3553      *
3554      * XXX should clip maxJobs against PIPE_BUF -- if maxJobs is
3555      * larger than the write buffer size of the pipe, we will
3556      * deadlock here.
3557      */
3558     for (i=1; i < maxproc; i++)
3559 	JobTokenAdd();
3560 }
3561 
3562 /*
3563  * this tracks the number of tokens currently "out" to build jobs.
3564  */
3565 int jobTokensRunning = 0;
3566 int jobTokensFree = 0;
3567 /*-
3568  *-----------------------------------------------------------------------
3569  * Job_TokenReturn --
3570  *	Return a withdrawn token to the pool.
3571  *
3572  *-----------------------------------------------------------------------
3573  */
3574 
3575 void
3576 Job_TokenReturn(void)
3577 {
3578     jobTokensRunning--;
3579     if (jobTokensRunning < 0)
3580 	Punt("token botch");
3581     if (jobTokensRunning)
3582 	jobTokensFree++;
3583 }
3584 
3585 /*-
3586  *-----------------------------------------------------------------------
3587  * Job_TokenWithdraw --
3588  *	Attempt to withdraw a token from the pool.
3589  *
3590  * Results:
3591  *	Returns TRUE if a token was withdrawn, and FALSE if the pool
3592  *	is currently empty.
3593  *
3594  * Side Effects:
3595  * 	If pool is empty, set wantToken so that we wake up
3596  *	when a token is released.
3597  *
3598  *-----------------------------------------------------------------------
3599  */
3600 
3601 
3602 Boolean
3603 Job_TokenWithdraw(void)
3604 {
3605     char tok;
3606     int count;
3607 
3608     wantToken = FALSE;
3609 
3610     if (aborting)
3611 	    return FALSE;
3612 
3613     if (jobTokensRunning == 0) {
3614 	if (DEBUG(JOB))
3615 	    printf("first one's free\n");
3616 	jobTokensRunning++;
3617 	return TRUE;
3618     }
3619     if (jobTokensFree > 0) {
3620 	jobTokensFree--;
3621 	jobTokensRunning++;
3622 	return TRUE;
3623     }
3624     count = read(job_pipe[0], &tok, 1);
3625     if (count == 0)
3626 	Fatal("eof on job pipe!");
3627     else if (count < 0) {
3628 	if (errno != EAGAIN) {
3629 	    Fatal("job pipe read: %s", strerror(errno));
3630 	}
3631 	if (DEBUG(JOB))
3632 	    printf("blocked for token\n");
3633 	wantToken = TRUE;
3634 	return FALSE;
3635     }
3636     jobTokensRunning++;
3637     if (DEBUG(JOB))
3638 	printf("withdrew token\n");
3639     return TRUE;
3640 }
3641 
3642 /*-
3643  *-----------------------------------------------------------------------
3644  * Job_TokenFlush --
3645  *	Return free tokens to the pool.
3646  *
3647  *-----------------------------------------------------------------------
3648  */
3649 
3650 void
3651 Job_TokenFlush(void)
3652 {
3653     if (compatMake) return;
3654 
3655     while (jobTokensFree > 0) {
3656 	JobTokenAdd();
3657 	jobTokensFree--;
3658     }
3659 }
3660 
3661 #ifdef USE_SELECT
3662 int
3663 emul_poll(struct pollfd *fd, int nfd, int timeout)
3664 {
3665     fd_set rfds, wfds;
3666     int i, maxfd, nselect, npoll;
3667     struct timeval tv, *tvp;
3668     long usecs;
3669 
3670     FD_ZERO(&rfds);
3671     FD_ZERO(&wfds);
3672 
3673     maxfd = -1;
3674     for (i = 0; i < nfd; i++) {
3675 	fd[i].revents = 0;
3676 
3677 	if (fd[i].events & POLLIN)
3678 	    FD_SET(fd[i].fd, &rfds);
3679 
3680 	if (fd[i].events & POLLOUT)
3681 	    FD_SET(fd[i].fd, &wfds);
3682 
3683 	if (fd[i].fd > maxfd)
3684 	    maxfd = fd[i].fd;
3685     }
3686 
3687     if (maxfd >= FD_SETSIZE) {
3688 	Punt("Ran out of fd_set slots; "
3689 	     "recompile with a larger FD_SETSIZE.");
3690     }
3691 
3692     if (timeout < 0) {
3693 	tvp = NULL;
3694     } else {
3695 	usecs = timeout * 1000;
3696 	tv.tv_sec = usecs / 1000000;
3697 	tv.tv_usec = usecs % 1000000;
3698         tvp = &tv;
3699     }
3700 
3701     nselect = select(maxfd + 1, &rfds, &wfds, 0, tvp);
3702 
3703     if (nselect <= 0)
3704 	return nselect;
3705 
3706     npoll = 0;
3707     for (i = 0; i < nfd; i++) {
3708 	if (FD_ISSET(fd[i].fd, &rfds))
3709 	    fd[i].revents |= POLLIN;
3710 
3711 	if (FD_ISSET(fd[i].fd, &wfds))
3712 	    fd[i].revents |= POLLOUT;
3713 
3714 	if (fd[i].revents)
3715 	    npoll++;
3716     }
3717 
3718     return npoll;
3719 }
3720 #endif /* USE_SELECT */
3721