2 * Copyright (c) 1988, 1989, 1990, 1993
3 * The Regents of the University of California. All rights reserved.
4 * Copyright (c) 1988, 1989 by Adam de Boor
5 * Copyright (c) 1989 by Berkeley Softworks
8 * This code is derived from software contributed to Berkeley by
11 * Redistribution and use in source and binary forms, with or without
12 * modification, are permitted provided that the following conditions
14 * 1. Redistributions of source code must retain the above copyright
15 * notice, this list of conditions and the following disclaimer.
16 * 2. Redistributions in binary form must reproduce the above copyright
17 * notice, this list of conditions and the following disclaimer in the
18 * documentation and/or other materials provided with the distribution.
19 * 3. All advertising materials mentioning features or use of this software
20 * must display the following acknowledgement:
21 * This product includes software developed by the University of
22 * California, Berkeley and its contributors.
23 * 4. Neither the name of the University nor the names of its contributors
24 * may be used to endorse or promote products derived from this software
25 * without specific prior written permission.
27 * THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND
28 * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
29 * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
30 * ARE DISCLAIMED. IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE
31 * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
32 * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
33 * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
34 * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
35 * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
36 * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
39 * @(#)job.c 8.2 (Berkeley) 3/19/94
40 * $FreeBSD: src/usr.bin/make/job.c,v 1.17.2.2 2001/02/13 03:13:57 will Exp $
41 * $DragonFly: src/usr.bin/make/job.c,v 1.18 2004/11/24 07:19:14 dillon Exp $
50 * handle the creation etc. of our child processes.
53 * Job_Make Start the creation of the given target.
55 * Job_CatchChildren Check for and handle the termination of any
56 * children. This must be called reasonably
57 * frequently to keep the whole make going at
58 * a decent clip, since job table entries aren't
59 * removed until their process is caught this way.
60 * Its single argument is TRUE if the function
61 * should block waiting for a child to terminate.
63 * Job_CatchOutput Print any output our children have produced.
64 * Should also be called fairly frequently to
65 * keep the user informed of what's going on.
66 * If no output is waiting, it will block for
67 * a time given by the SEL_* constants, below,
68 * or until output is ready.
70 * Job_Init Called to intialize this module. in addition,
71 * any commands attached to the .BEGIN target
72 * are executed before this function returns.
73 * Hence, the makefile must have been parsed
74 * before this function is called.
76 * Job_Full Return TRUE if the job table is filled.
78 * Job_Empty Return TRUE if the job table is completely
81 * Job_ParseShell Given the line following a .SHELL target, parse
82 * the line as a shell specification. Returns
83 * FAILURE if the spec was incorrect.
85 * Job_Finish Perform any final processing which needs doing.
86 * This includes the execution of any commands
87 * which have been/were attached to the .END
88 * target. It should only be called when the
91 * Job_AbortAll Abort all currently running jobs. It doesn't
92 * handle output or do anything for the jobs,
93 * just kills them. It should only be called in
94 * an emergency, as it were.
96 * Job_CheckCommands Verify that the commands for a target are
97 * ok. Provide them if necessary and possible.
99 * Job_Touch Update a target without really updating it.
101 * Job_Wait Wait for all currently-running jobs to finish.
104 #include <sys/types.h>
105 #include <sys/stat.h>
106 #include <sys/file.h>
107 #include <sys/time.h>
108 #include <sys/wait.h>
121 #include "pathnames.h"
126 # define STATIC static
130 * error handling variables
132 static int errors = 0; /* number of errors reported */
133 static int aborting = 0; /* why is the make aborting? */
134 #define ABORT_ERROR 1 /* Because of an error */
135 #define ABORT_INTERRUPT 2 /* Because it was interrupted */
136 #define ABORT_WAIT 3 /* Waiting for jobs to finish */
139 * XXX: Avoid SunOS bug... FILENO() is fp->_file, and file
140 * is a char! So when we go above 127 we turn negative!
142 #define FILENO(a) ((unsigned) fileno(a))
145 * post-make command processing. The node postCommands is really just the
146 * .END target but we keep it around to avoid having to search for it
149 static GNode *postCommands; /* node containing commands to execute when
150 * everything else is done */
151 static int numCommands; /* The number of commands actually printed
152 * for a target. Should this number be
153 * 0, no shell will be executed. */
156 * Return values from JobStart.
158 #define JOB_RUNNING 0 /* Job is running */
159 #define JOB_ERROR 1 /* Error in starting the job */
160 #define JOB_FINISHED 2 /* The job is already finished */
161 #define JOB_STOPPED 3 /* The job is stopped */
164 * tfile is used to build temp file names to store shell commands to
167 static char tfile[sizeof(TMPPAT)];
171 * Descriptions for various shells.
173 static Shell shells[] = {
175 * CSH description. The csh can do echo control by playing
176 * with the setting of the 'echo' shell variable. Sadly,
177 * however, it is unable to do error control nicely.
181 TRUE, "unset verbose", "set verbose", "unset verbose", 10,
182 FALSE, "echo \"%s\"\n", "csh -c \"%s || exit 0\"",
186 * SH description. Echo control is also possible and, under
187 * sun UNIX anyway, one can even control error checking.
191 TRUE, "set -", "set -v", "set -", 5,
192 TRUE, "set -e", "set +e",
193 #ifdef OLDBOURNESHELL
194 FALSE, "echo \"%s\"\n", "sh -c '%s || exit 0'\n",
203 FALSE, (char *) 0, (char *) 0, (char *) 0, 0,
204 FALSE, (char *) 0, (char *) 0,
205 (char *) 0, (char *) 0,
208 static Shell *commandShell = &shells[DEFSHELL];/* this is the shell to
210 * commands in the Makefile.
212 * Job_ParseShell function */
213 static char *shellPath = NULL, /* full pathname of
214 * executable image */
215 *shellName; /* last component of shell */
218 static int maxJobs; /* The most children we can run at once */
219 static int maxLocal; /* The most local ones we can have */
220 STATIC int nJobs; /* The number of children currently running */
221 STATIC int nLocal; /* The number of local children */
222 STATIC Lst jobs; /* The structures that describe them */
223 STATIC Boolean jobFull; /* Flag to tell when the job table is full. It
224 * is set TRUE when (1) the total number of
225 * running jobs equals the maximum allowed or
226 * (2) a job can only be run locally, but
227 * nLocal equals maxLocal */
228 #ifndef RMT_WILL_WATCH
229 static fd_set outputs; /* Set of descriptors of pipes connected to
230 * the output channels of children */
233 STATIC GNode *lastNode; /* The node for which output was most recently
235 STATIC char *targFmt; /* Format string to use to head output from a
236 * job when it's not the most-recent job heard
240 # define TARG_FMT "--- %s at %s ---\n" /* Default format */
241 # define MESSAGE(fp, gn) \
242 (void) fprintf(fp, targFmt, gn->name, gn->rem.hname);
244 # define TARG_FMT "--- %s ---\n" /* Default format */
245 # define MESSAGE(fp, gn) \
246 (void) fprintf(fp, targFmt, gn->name);
250 * When JobStart attempts to run a job remotely but can't, and isn't allowed
251 * to run the job locally, or when Job_CatchChildren detects a job that has
252 * been migrated home, the job is placed on the stoppedJobs queue to be run
253 * when the next job finishes.
255 STATIC Lst stoppedJobs; /* Lst of Job structures describing
256 * jobs that were stopped due to concurrency
257 * limits or migration home */
260 #if defined(USE_PGRP) && defined(SYSV)
261 # define KILL(pid, sig) killpg(-(pid), (sig))
263 # if defined(USE_PGRP)
264 # define KILL(pid, sig) killpg((pid), (sig))
266 # define KILL(pid, sig) kill((pid), (sig))
271 * Grmpf... There is no way to set bits of the wait structure
272 * anymore with the stupid W*() macros. I liked the union wait
273 * stuff much more. So, we devise our own macros... This is
274 * really ugly, use dramamine sparingly. You have been warned.
276 #define W_SETMASKED(st, val, fun) \
279 int mask = fun(sh); \
281 for (sh = 0; ((mask >> sh) & 1) == 0; sh++) \
283 *(st) = (*(st) & ~mask) | ((val) << sh); \
286 #define W_SETTERMSIG(st, val) W_SETMASKED(st, val, WTERMSIG)
287 #define W_SETEXITSTATUS(st, val) W_SETMASKED(st, val, WEXITSTATUS)
290 static int JobCondPassSig(void *, void *);
291 static void JobPassSig(int);
292 static int JobCmpPid(void *, void *);
293 static int JobPrintCommand(void *, void *);
294 static int JobSaveCommand(void *, void *);
295 static void JobClose(Job *);
297 static int JobCmpRmtID(Job *, int);
298 # ifdef RMT_WILL_WATCH
299 static void JobLocalInput(int, Job *);
302 static void JobFinish(Job *, int *);
303 static void JobExec(Job *, char **);
305 static void JobMakeArgv(Job *, char **);
306 static void JobRestart(Job *);
307 static int JobStart(GNode *, int, Job *);
308 static char *JobOutput(Job *, char *, char *, int);
309 static void JobDoOutput(Job *, Boolean);
310 static Shell *JobMatchShell(char *);
311 static void JobInterrupt(int, int);
312 static void JobRestartJobs(void);
315 *-----------------------------------------------------------------------
317 * Pass a signal to a job if the job is remote or if USE_PGRP
324 * None, except the job may bite it.
326 *-----------------------------------------------------------------------
329 JobCondPassSig(void *jobp, void *signop)
331 Job *job = (Job *) jobp;
332 int signo = *(int *) signop;
333 #ifdef RMT_WANTS_SIGNALS
334 if (job->flags & JOB_REMOTE) {
335 (void) Rmt_Signal(job, signo);
337 KILL(job->pid, signo);
341 * Assume that sending the signal to job->pid will signal any remote
344 DEBUGF(JOB, ("JobCondPassSig passing signal %d to child %d.\n", signo, job->pid));
345 KILL(job->pid, signo);
351 *-----------------------------------------------------------------------
353 * Pass a signal on to all remote jobs and to all local jobs if
354 * USE_PGRP is defined, then die ourselves.
360 * We die by the same signal.
362 *-----------------------------------------------------------------------
365 JobPassSig(int signo)
367 sigset_t nmask, omask;
368 struct sigaction act;
370 DEBUGF(JOB, ("JobPassSig(%d) called.\n", signo));
371 Lst_ForEach(jobs, JobCondPassSig, (void *) &signo);
374 * Deal with proper cleanup based on the signal received. We only run
375 * the .INTERRUPT target if the signal was in fact an interrupt. The other
376 * three termination signals are more of a "get out *now*" command.
378 if (signo == SIGINT) {
379 JobInterrupt(TRUE, signo);
380 } else if ((signo == SIGHUP) || (signo == SIGTERM) || (signo == SIGQUIT)) {
381 JobInterrupt(FALSE, signo);
385 * Leave gracefully if SIGQUIT, rather than core dumping.
387 if (signo == SIGQUIT) {
392 * Send ourselves the signal now we've given the message to everyone else.
393 * Note we block everything else possible while we're getting the signal.
394 * This ensures that all our jobs get continued when we wake up before
395 * we take any other signal.
398 sigaddset(&nmask, signo);
399 sigprocmask(SIG_SETMASK, &nmask, &omask);
400 act.sa_handler = SIG_DFL;
401 sigemptyset(&act.sa_mask);
403 sigaction(signo, &act, NULL);
405 DEBUGF(JOB, ("JobPassSig passing signal to self, mask = %x.\n", ~0 & ~(1 << (signo-1))));
406 (void) signal(signo, SIG_DFL);
408 (void) KILL(getpid(), signo);
411 Lst_ForEach(jobs, JobCondPassSig, (void *) &signo);
413 (void) sigprocmask(SIG_SETMASK, &omask, NULL);
414 sigprocmask(SIG_SETMASK, &omask, NULL);
415 act.sa_handler = JobPassSig;
416 sigaction(signo, &act, NULL);
420 *-----------------------------------------------------------------------
422 * Compare the pid of the job with the given pid and return 0 if they
423 * are equal. This function is called from Job_CatchChildren via
424 * Lst_Find to find the job descriptor of the finished job.
427 * 0 if the pid's match
431 *-----------------------------------------------------------------------
434 JobCmpPid(void *job, void *pid)
436 return *(int *) pid - ((Job *) job)->pid;
441 *-----------------------------------------------------------------------
443 * Compare the rmtID of the job with the given rmtID and return 0 if they
447 * 0 if the rmtID's match
451 *-----------------------------------------------------------------------
454 JobCmpRmtID(void *job, void *rmtID)
456 return(*(int *) rmtID - *(int *) job->rmtID);
461 *-----------------------------------------------------------------------
463 * Put out another command for the given job. If the command starts
464 * with an @ or a - we process it specially. In the former case,
465 * so long as the -s and -n flags weren't given to make, we stick
466 * a shell-specific echoOff command in the script. In the latter,
467 * we ignore errors for the entire job, unless the shell has error
469 * If the command is just "..." we take all future commands for this
470 * job to be commands to be executed once the entire graph has been
471 * made and return non-zero to signal that the end of the commands
472 * was reached. These commands are later attached to the postCommands
473 * node and executed by Job_Finish when all things are done.
474 * This function is called from JobStart via Lst_ForEach.
477 * Always 0, unless the command was "..."
480 * If the command begins with a '-' and the shell has no error control,
481 * the JOB_IGNERR flag is set in the job descriptor.
482 * If the command is "..." and we're not ignoring such things,
483 * tailCmds is set to the successor node of the cmd.
484 * numCommands is incremented if the command is actually printed.
485 *-----------------------------------------------------------------------
488 JobPrintCommand(void *cmdp, void *jobp)
490 Boolean noSpecials; /* true if we shouldn't worry about
491 * inserting special commands into
492 * the input stream. */
493 Boolean shutUp = FALSE; /* true if we put a no echo command
494 * into the command file */
495 Boolean errOff = FALSE; /* true if we turned error checking
496 * off before printing the command
497 * and need to turn it back on */
498 char *cmdTemplate; /* Template to use when printing the
500 char *cmdStart; /* Start of expanded command */
501 LstNode cmdNode; /* Node for replacing the command */
502 char *cmd = (char *) cmdp;
503 Job *job = (Job *) jobp;
505 noSpecials = (noExecute && !(job->node->type & OP_MAKE));
507 if (strcmp(cmd, "...") == 0) {
508 job->node->type |= OP_SAVE_CMDS;
509 if ((job->flags & JOB_IGNDOTS) == 0) {
510 job->tailCmds = Lst_Succ(Lst_Member(job->node->commands,
517 #define DBPRINTF(fmt, arg) \
518 DEBUGF(JOB, (fmt, arg)); \
519 (void) fprintf(job->cmdFILE, fmt, arg); \
520 (void) fflush(job->cmdFILE);
525 * For debugging, we replace each command with the result of expanding
526 * the variables in the command.
528 cmdNode = Lst_Member(job->node->commands, (void *)cmd);
529 cmdStart = cmd = Var_Subst(NULL, cmd, job->node, FALSE);
530 Lst_Replace(cmdNode, (void *)cmdStart);
532 cmdTemplate = "%s\n";
535 * Check for leading @' and -'s to control echoing and error checking.
537 while (*cmd == '@' || *cmd == '-') {
539 shutUp = DEBUG(LOUD) ? FALSE : TRUE;
546 while (isspace((unsigned char) *cmd))
550 if (!(job->flags & JOB_SILENT) && !noSpecials &&
551 commandShell->hasEchoCtl) {
552 DBPRINTF("%s\n", commandShell->echoOff);
559 if ( !(job->flags & JOB_IGNERR) && !noSpecials) {
560 if (commandShell->hasErrCtl) {
562 * we don't want the error-control commands showing
563 * up either, so we turn off echoing while executing
564 * them. We could put another field in the shell
565 * structure to tell JobDoOutput to look for this
566 * string too, but why make it any more complex than
569 if (!(job->flags & JOB_SILENT) && !shutUp &&
570 commandShell->hasEchoCtl) {
571 DBPRINTF("%s\n", commandShell->echoOff);
572 DBPRINTF("%s\n", commandShell->ignErr);
573 DBPRINTF("%s\n", commandShell->echoOn);
575 DBPRINTF("%s\n", commandShell->ignErr);
577 } else if (commandShell->ignErr &&
578 (*commandShell->ignErr != '\0'))
581 * The shell has no error control, so we need to be
582 * weird to get it to ignore any errors from the command.
583 * If echoing is turned on, we turn it off and use the
584 * errCheck template to echo the command. Leave echoing
585 * off so the user doesn't see the weirdness we go through
586 * to ignore errors. Set cmdTemplate to use the weirdness
587 * instead of the simple "%s\n" template.
589 if (!(job->flags & JOB_SILENT) && !shutUp &&
590 commandShell->hasEchoCtl) {
591 DBPRINTF("%s\n", commandShell->echoOff);
592 DBPRINTF(commandShell->errCheck, cmd);
595 cmdTemplate = commandShell->ignErr;
597 * The error ignoration (hee hee) is already taken care
598 * of by the ignErr template, so pretend error checking
610 DBPRINTF(cmdTemplate, cmd);
614 * If echoing is already off, there's no point in issuing the
615 * echoOff command. Otherwise we issue it and pretend it was on
616 * for the whole command...
618 if (!shutUp && !(job->flags & JOB_SILENT) && commandShell->hasEchoCtl){
619 DBPRINTF("%s\n", commandShell->echoOff);
622 DBPRINTF("%s\n", commandShell->errCheck);
625 DBPRINTF("%s\n", commandShell->echoOn);
631 *-----------------------------------------------------------------------
633 * Save a command to be executed when everything else is done.
634 * Callback function for JobFinish...
640 * The command is tacked onto the end of postCommands's commands list.
642 *-----------------------------------------------------------------------
645 JobSaveCommand(void *cmd, void *gn)
647 cmd = (void *) Var_Subst(NULL, (char *) cmd, (GNode *) gn, FALSE);
648 (void) Lst_AtEnd(postCommands->commands, cmd);
654 *-----------------------------------------------------------------------
656 * Called to close both input and output pipes when a job is finished.
662 * The file descriptors associated with the job are closed.
664 *-----------------------------------------------------------------------
670 #ifdef RMT_WILL_WATCH
671 Rmt_Ignore(job->inPipe);
673 FD_CLR(job->inPipe, &outputs);
675 if (job->outPipe != job->inPipe) {
676 (void) close(job->outPipe);
678 JobDoOutput(job, TRUE);
679 (void) close(job->inPipe);
681 (void) close(job->outFd);
682 JobDoOutput(job, TRUE);
687 *-----------------------------------------------------------------------
689 * Do final processing for the given job including updating
690 * parents and starting new jobs as available/necessary. Note
691 * that we pay no attention to the JOB_IGNERR flag here.
692 * This is because when we're called because of a noexecute flag
693 * or something, jstat.w_status is 0 and when called from
694 * Job_CatchChildren, the status is zeroed if it s/b ignored.
700 * Some nodes may be put on the toBeMade queue.
701 * Final commands for the job are placed on postCommands.
703 * If we got an error and are aborting (aborting == ABORT_ERROR) and
704 * the job list is now empty, we are done for the day.
705 * If we recognized an error (errors !=0), we set the aborting flag
706 * to ABORT_ERROR so no more jobs will be started.
707 *-----------------------------------------------------------------------
711 JobFinish(Job *job, int *status)
715 if ((WIFEXITED(*status) &&
716 (((WEXITSTATUS(*status) != 0) && !(job->flags & JOB_IGNERR)))) ||
717 (WIFSIGNALED(*status) && (WTERMSIG(*status) != SIGCONT)))
720 * If it exited non-zero and either we're doing things our
721 * way or we're not ignoring errors, the job is finished.
722 * Similarly, if the shell died because of a signal
723 * the job is also finished. In these
724 * cases, finish out the job's output before printing the exit
728 KILL(job->pid, SIGCONT);
731 if (job->cmdFILE != NULL && job->cmdFILE != stdout) {
732 (void) fclose(job->cmdFILE);
736 if (job->flags & JOB_REMOTE)
737 Rmt_Done(job->rmtID, job->node);
739 } else if (WIFEXITED(*status)) {
741 * Deal with ignored errors in -B mode. We need to print a message
742 * telling of the ignored error as well as setting status.w_status
743 * to 0 so the next command gets run. To do this, we set done to be
744 * TRUE if in -B mode and the job exited non-zero.
746 done = WEXITSTATUS(*status) != 0;
748 * Old comment said: "Note we don't
749 * want to close down any of the streams until we know we're at the
751 * But we do. Otherwise when are we going to print the rest of the
756 if (job->flags & JOB_REMOTE)
757 Rmt_Done(job->rmtID, job->node);
761 * No need to close things down or anything.
767 WIFSTOPPED(*status) ||
768 (WIFSIGNALED(*status) && (WTERMSIG(*status) == SIGCONT)) ||
773 if (compatMake && !usePipes && (job->flags & JOB_IGNERR)) {
775 * If output is going to a file and this job is ignoring
776 * errors, arrange to have the exit status sent to the
777 * output file as well.
779 out = fdopen(job->outFd, "w");
781 Punt("Cannot fdopen");
786 if (WIFEXITED(*status)) {
787 DEBUGF(JOB, ("Process %d exited.\n", job->pid));
788 if (WEXITSTATUS(*status) != 0) {
789 if (usePipes && job->node != lastNode) {
790 MESSAGE(out, job->node);
791 lastNode = job->node;
793 (void) fprintf(out, "*** Error code %d%s\n",
794 WEXITSTATUS(*status),
795 (job->flags & JOB_IGNERR) ? "(ignored)" : "");
797 if (job->flags & JOB_IGNERR) {
800 } else if (DEBUG(JOB)) {
801 if (usePipes && job->node != lastNode) {
802 MESSAGE(out, job->node);
803 lastNode = job->node;
805 (void) fprintf(out, "*** Completed successfully\n");
807 } else if (WIFSTOPPED(*status)) {
808 DEBUGF(JOB, ("Process %d stopped.\n", job->pid));
809 if (usePipes && job->node != lastNode) {
810 MESSAGE(out, job->node);
811 lastNode = job->node;
813 if (!(job->flags & JOB_REMIGRATE)) {
814 (void) fprintf(out, "*** Stopped -- signal %d\n",
817 job->flags |= JOB_RESUME;
818 (void)Lst_AtEnd(stoppedJobs, (void *)job);
820 if (job->flags & JOB_REMIGRATE)
825 } else if (WTERMSIG(*status) == SIGCONT) {
827 * If the beastie has continued, shift the Job from the stopped
828 * list to the running one (or re-stop it if concurrency is
829 * exceeded) and go and get another child.
831 if (job->flags & (JOB_RESUME|JOB_REMIGRATE|JOB_RESTART)) {
832 if (usePipes && job->node != lastNode) {
833 MESSAGE(out, job->node);
834 lastNode = job->node;
836 (void) fprintf(out, "*** Continued\n");
838 if (!(job->flags & JOB_CONTINUING)) {
839 DEBUGF(JOB, ("Warning: process %d was not continuing.\n", job->pid));
842 * We don't really want to restart a job from scratch just
843 * because it continued, especially not without killing the
844 * continuing process! That's why this is ifdef'ed out.
850 job->flags &= ~JOB_CONTINUING;
851 Lst_AtEnd(jobs, (void *)job);
853 if (!(job->flags & JOB_REMOTE)) {
854 DEBUGF(JOB, ("Process %d is continuing locally.\n", job->pid));
857 if (nJobs == maxJobs) {
859 DEBUGF(JOB, ("Job queue is full.\n"));
864 if (usePipes && job->node != lastNode) {
865 MESSAGE(out, job->node);
866 lastNode = job->node;
868 (void) fprintf(out, "*** Signal %d\n", WTERMSIG(*status));
875 * Now handle the -B-mode stuff. If the beast still isn't finished,
876 * try and restart the job on the next command. If JobStart says it's
877 * ok, it's ok. If there's an error, this puppy is done.
879 if (compatMake && (WIFEXITED(*status) &&
880 !Lst_IsAtEnd(job->node->commands))) {
881 switch (JobStart(job->node, job->flags & JOB_IGNDOTS, job)) {
887 W_SETEXITSTATUS(status, 1);
891 * If we got back a JOB_FINISHED code, JobStart has already
892 * called Make_Update and freed the job descriptor. We set
893 * done to false here to avoid fake cycles and double frees.
894 * JobStart needs to do the update so we can proceed up the
895 * graph when given the -n flag..
908 (aborting != ABORT_ERROR) &&
909 (aborting != ABORT_INTERRUPT) &&
913 * As long as we aren't aborting and the job didn't return a non-zero
914 * status that we shouldn't ignore, we call Make_Update to update
915 * the parents. In addition, any saved commands for the node are placed
916 * on the .END target.
918 if (job->tailCmds != NULL) {
919 Lst_ForEachFrom(job->node->commands, job->tailCmds,
923 job->node->made = MADE;
924 Make_Update(job->node);
926 } else if (*status != 0) {
934 * Set aborting if any error.
936 if (errors && !keepgoing && (aborting != ABORT_INTERRUPT)) {
938 * If we found any errors in this batch of children and the -k flag
939 * wasn't given, we set the aborting flag so no more jobs get
942 aborting = ABORT_ERROR;
945 if ((aborting == ABORT_ERROR) && Job_Empty())
947 * If we are aborting and the job table is now empty, we finish.
953 *-----------------------------------------------------------------------
955 * Touch the given target. Called by JobStart when the -t flag was
956 * given. Prints messages unless told to be silent.
962 * The data modification of the file is changed. In addition, if the
963 * file did not exist, it is created.
964 *-----------------------------------------------------------------------
967 Job_Touch(GNode *gn, Boolean silent)
969 int streamID; /* ID of stream opened to do the touch */
970 struct utimbuf times; /* Times for utime() call */
972 if (gn->type & (OP_JOIN|OP_USE|OP_EXEC|OP_OPTIONAL)) {
974 * .JOIN, .USE, .ZEROTIME and .OPTIONAL targets are "virtual" targets
975 * and, as such, shouldn't really be created.
981 (void) fprintf(stdout, "touch %s\n", gn->name);
982 (void) fflush(stdout);
989 if (gn->type & OP_ARCHV) {
991 } else if (gn->type & OP_LIB) {
994 char *file = gn->path ? gn->path : gn->name;
996 times.actime = times.modtime = now;
997 if (utime(file, ×) < 0){
998 streamID = open(file, O_RDWR | O_CREAT, 0666);
1000 if (streamID >= 0) {
1004 * Read and write a byte to the file to change the
1005 * modification time, then close the file.
1007 if (read(streamID, &c, 1) == 1) {
1008 (void) lseek(streamID, (off_t)0, SEEK_SET);
1009 (void) write(streamID, &c, 1);
1012 (void) close(streamID);
1014 (void) fprintf(stdout, "*** couldn't touch %s: %s",
1015 file, strerror(errno));
1016 (void) fflush(stdout);
1023 *-----------------------------------------------------------------------
1024 * Job_CheckCommands --
1025 * Make sure the given node has all the commands it needs.
1028 * TRUE if the commands list is/was ok.
1031 * The node will have commands from the .DEFAULT rule added to it
1033 *-----------------------------------------------------------------------
1036 Job_CheckCommands(GNode *gn, void (*abortProc)(const char *, ...))
1038 if (OP_NOP(gn->type) && Lst_IsEmpty(gn->commands) &&
1039 (gn->type & OP_LIB) == 0) {
1041 * No commands. Look for .DEFAULT rule from which we might infer
1044 if ((DEFAULT != NULL) && !Lst_IsEmpty(DEFAULT->commands)) {
1047 * Make only looks for a .DEFAULT if the node was never the
1048 * target of an operator, so that's what we do too. If
1049 * a .DEFAULT was given, we substitute its commands for gn's
1050 * commands and set the IMPSRC variable to be the target's name
1051 * The DEFAULT node acts like a transformation rule, in that
1052 * gn also inherits any attributes or sources attached to
1055 Make_HandleUse(DEFAULT, gn);
1056 Var_Set(IMPSRC, Var_Value(TARGET, gn, &p1), gn);
1058 } else if (Dir_MTime(gn) == 0) {
1060 * The node wasn't the target of an operator we have no .DEFAULT
1061 * rule to go on and the target doesn't already exist. There's
1062 * nothing more we can do for this branch. If the -k flag wasn't
1063 * given, we stop in our tracks, otherwise we just don't update
1064 * this node's parents so they never get examined.
1066 static const char msg[] = "make: don't know how to make";
1068 if (gn->type & OP_OPTIONAL) {
1069 (void) fprintf(stdout, "%s %s(ignored)\n", msg, gn->name);
1070 (void) fflush(stdout);
1071 } else if (keepgoing) {
1072 (void) fprintf(stdout, "%s %s(continuing)\n", msg, gn->name);
1073 (void) fflush(stdout);
1077 if (strcmp(gn->name,"love") == 0)
1078 (*abortProc)("Not war.");
1081 (*abortProc)("%s %s. Stop", msg, gn->name);
1088 #ifdef RMT_WILL_WATCH
1090 *-----------------------------------------------------------------------
1092 * Handle a pipe becoming readable. Callback function for Rmt_Watch
1098 * JobDoOutput is called.
1100 *-----------------------------------------------------------------------
1104 JobLocalInput(int stream, Job *job)
1106 JobDoOutput(job, FALSE);
1108 #endif /* RMT_WILL_WATCH */
1111 *-----------------------------------------------------------------------
1113 * Execute the shell for the given job. Called from JobStart and
1120 * A shell is executed, outputs is altered and the Job structure added
1123 *-----------------------------------------------------------------------
1126 JobExec(Job *job, char **argv)
1128 int cpid; /* ID of new child */
1133 DEBUGF(JOB, ("Running %s %sly\n", job->node->name,
1134 job->flags&JOB_REMOTE?"remote":"local"));
1135 DEBUGF(JOB, ("\tCommand: "));
1136 for (i = 0; argv[i] != NULL; i++) {
1137 DEBUGF(JOB, ("%s ", argv[i]));
1139 DEBUGF(JOB, ("\n"));
1143 * Some jobs produce no output and it's disconcerting to have
1144 * no feedback of their running (since they produce no output, the
1145 * banner with their name in it never appears). This is an attempt to
1146 * provide that feedback, even if nothing follows it.
1148 if ((lastNode != job->node) && (job->flags & JOB_FIRST) &&
1149 !(job->flags & JOB_SILENT)) {
1150 MESSAGE(stdout, job->node);
1151 lastNode = job->node;
1155 if (job->flags & JOB_REMOTE) {
1158 #endif /* RMT_NO_EXEC */
1160 if ((cpid = vfork()) == -1) {
1161 Punt("Cannot fork");
1162 } else if (cpid == 0) {
1165 * Must duplicate the input stream down to the child's input and
1166 * reset it to the beginning (again). Since the stream was marked
1167 * close-on-exec, we must clear that bit in the new input.
1169 if (dup2(FILENO(job->cmdFILE), 0) == -1)
1170 Punt("Cannot dup2: %s", strerror(errno));
1171 (void) fcntl(0, F_SETFD, 0);
1172 (void) lseek(0, (off_t)0, SEEK_SET);
1176 * Set up the child's output to be routed through the pipe
1177 * we've created for it.
1179 if (dup2(job->outPipe, 1) == -1)
1180 Punt("Cannot dup2: %s", strerror(errno));
1183 * We're capturing output in a file, so we duplicate the
1184 * descriptor to the temporary file into the standard
1187 if (dup2(job->outFd, 1) == -1)
1188 Punt("Cannot dup2: %s", strerror(errno));
1191 * The output channels are marked close on exec. This bit was
1192 * duplicated by the dup2 (on some systems), so we have to clear
1193 * it before routing the shell's error output to the same place as
1194 * its standard output.
1196 (void) fcntl(1, F_SETFD, 0);
1197 if (dup2(1, 2) == -1)
1198 Punt("Cannot dup2: %s", strerror(errno));
1202 * We want to switch the child into a different process family so
1203 * we can kill it and all its descendants in one fell swoop,
1204 * by killing its process family, but not commit suicide.
1209 (void) setpgid(0, getpid());
1211 #endif /* USE_PGRP */
1214 if (job->flags & JOB_REMOTE) {
1215 Rmt_Exec(shellPath, argv, FALSE);
1218 (void) execv(shellPath, argv);
1220 (void) write(STDERR_FILENO, "Could not execute shell\n",
1221 sizeof("Could not execute shell"));
1225 long omask = sigblock(sigmask(SIGCHLD));
1229 if (usePipes && (job->flags & JOB_FIRST) ) {
1231 * The first time a job is run for a node, we set the current
1232 * position in the buffer to the beginning and mark another
1233 * stream to watch in the outputs mask
1237 #ifdef RMT_WILL_WATCH
1238 Rmt_Watch(job->inPipe, JobLocalInput, job);
1240 FD_SET(job->inPipe, &outputs);
1241 #endif /* RMT_WILL_WATCH */
1244 if (job->flags & JOB_REMOTE) {
1248 job->rmtID = Rmt_LastID(job->pid);
1253 * XXX: Used to not happen if REMOTE. Why?
1255 if (job->cmdFILE != NULL && job->cmdFILE != stdout) {
1256 (void) fclose(job->cmdFILE);
1257 job->cmdFILE = NULL;
1261 (void) sigsetmask(omask);
1269 * Now the job is actually running, add it to the table.
1272 (void) Lst_AtEnd(jobs, (void *)job);
1273 if (nJobs == maxJobs) {
1279 *-----------------------------------------------------------------------
1281 * Create the argv needed to execute the shell for a given job.
1288 *-----------------------------------------------------------------------
1291 JobMakeArgv(Job *job, char **argv)
1294 static char args[10]; /* For merged arguments */
1296 argv[0] = shellName;
1299 if ((commandShell->exit && (*commandShell->exit != '-')) ||
1300 (commandShell->echo && (*commandShell->echo != '-')))
1303 * At least one of the flags doesn't have a minus before it, so
1304 * merge them together. Have to do this because the *(&(@*#*&#$#
1305 * Bourne shell thinks its second argument is a file to source.
1306 * Grrrr. Note the ten-character limitation on the combined arguments.
1308 (void)sprintf(args, "-%s%s",
1309 ((job->flags & JOB_IGNERR) ? "" :
1310 (commandShell->exit ? commandShell->exit : "")),
1311 ((job->flags & JOB_SILENT) ? "" :
1312 (commandShell->echo ? commandShell->echo : "")));
1319 if (!(job->flags & JOB_IGNERR) && commandShell->exit) {
1320 argv[argc] = commandShell->exit;
1323 if (!(job->flags & JOB_SILENT) && commandShell->echo) {
1324 argv[argc] = commandShell->echo;
1332 *-----------------------------------------------------------------------
1334 * Restart a job that stopped for some reason.
1340 * jobFull will be set if the job couldn't be run.
1342 *-----------------------------------------------------------------------
1345 JobRestart(Job *job)
1351 if (job->flags & JOB_REMIGRATE) {
1354 verboseRemigrates ||
1357 (void) fprintf(stdout, "*** remigrating %x(%s)\n",
1358 job->pid, job->node->name);
1359 (void) fflush(stdout);
1363 if (!Rmt_ReExport(job->pid, job->node, &host)) {
1364 if (verboseRemigrates || DEBUG(JOB)) {
1365 (void) fprintf(stdout, "*** couldn't migrate...\n");
1366 (void) fflush(stdout);
1369 if (nLocal != maxLocal) {
1371 * Job cannot be remigrated, but there's room on the local
1372 * machine, so resume the job and note that another
1373 * local job has started.
1377 verboseRemigrates ||
1380 (void) fprintf(stdout, "*** resuming on local machine\n");
1381 (void) fflush(stdout);
1383 KILL(job->pid, SIGCONT);
1386 job->flags &= ~(JOB_REMIGRATE|JOB_RESUME|JOB_REMOTE);
1387 job->flags |= JOB_CONTINUING;
1389 job->flags &= ~(JOB_REMIGRATE|JOB_RESUME);
1393 * Job cannot be restarted. Mark the table as full and
1394 * place the job back on the list of stopped jobs.
1398 verboseRemigrates ||
1401 (void) fprintf(stdout, "*** holding\n");
1402 (void) fflush(stdout);
1404 (void)Lst_AtFront(stoppedJobs, (void *)job);
1406 DEBUGF(JOB, ("Job queue is full.\n"));
1412 * Clear out the remigrate and resume flags. Set the continuing
1413 * flag so we know later on that the process isn't exiting just
1414 * because of a signal.
1416 job->flags &= ~(JOB_REMIGRATE|JOB_RESUME);
1417 job->flags |= JOB_CONTINUING;
1422 (void)Lst_AtEnd(jobs, (void *)job);
1424 if (nJobs == maxJobs) {
1426 DEBUGF(JOB, ("Job queue is full.\n"));
1428 } else if (job->flags & JOB_RESTART) {
1430 * Set up the control arguments to the shell. This is based on the
1431 * flags set earlier for this job. If the JOB_IGNERR flag is clear,
1432 * the 'exit' flag of the commandShell is used to cause it to exit
1433 * upon receiving an error. If the JOB_SILENT flag is clear, the
1434 * 'echo' flag of the commandShell is used to get it to start echoing
1435 * as soon as it starts processing commands.
1439 JobMakeArgv(job, argv);
1441 DEBUGF(JOB, ("Restarting %s...", job->node->name));
1443 if ((job->node->type&OP_NOEXPORT) ||
1444 (nLocal < maxLocal && runLocalFirst)
1446 || !Rmt_Export(shellPath, argv, job)
1448 || !Rmt_Begin(shellPath, argv, job->node)
1452 if (((nLocal >= maxLocal) && !(job->flags & JOB_SPECIAL))) {
1454 * Can't be exported and not allowed to run locally -- put it
1455 * back on the hold queue and mark the table full
1457 DEBUGF(JOB, ("holding\n"));
1458 (void)Lst_AtFront(stoppedJobs, (void *)job);
1460 DEBUGF(JOB, ("Job queue is full.\n"));
1464 * Job may be run locally.
1466 DEBUGF(JOB, ("running locally\n"));
1467 job->flags &= ~JOB_REMOTE;
1473 * Can be exported. Hooray!
1475 DEBUGF(JOB, ("exporting\n"));
1476 job->flags |= JOB_REMOTE;
1482 * The job has stopped and needs to be restarted. Why it stopped,
1485 DEBUGF(JOB, ("Resuming %s...", job->node->name));
1486 if (((job->flags & JOB_REMOTE) ||
1487 (nLocal < maxLocal) ||
1489 (((job->flags & JOB_SPECIAL) &&
1490 (job->node->type & OP_NOEXPORT)) &&
1491 (maxLocal == 0))) &&
1493 ((job->flags & JOB_SPECIAL) &&
1494 (maxLocal == 0))) &&
1499 * If the job is remote, it's ok to resume it as long as the
1500 * maximum concurrency won't be exceeded. If it's local and
1501 * we haven't reached the local concurrency limit already (or the
1502 * job must be run locally and maxLocal is 0), it's also ok to
1508 #ifdef RMT_WANTS_SIGNALS
1509 if (job->flags & JOB_REMOTE) {
1510 error = !Rmt_Signal(job, SIGCONT);
1512 #endif /* RMT_WANTS_SIGNALS */
1513 error = (KILL(job->pid, SIGCONT) != 0);
1517 * Make sure the user knows we've continued the beast and
1518 * actually put the thing in the job table.
1520 job->flags |= JOB_CONTINUING;
1521 W_SETTERMSIG(&status, SIGCONT);
1522 JobFinish(job, &status);
1524 job->flags &= ~(JOB_RESUME|JOB_CONTINUING);
1525 DEBUGF(JOB, ("done\n"));
1527 Error("couldn't resume %s: %s",
1528 job->node->name, strerror(errno));
1530 W_SETEXITSTATUS(&status, 1);
1531 JobFinish(job, &status);
1535 * Job cannot be restarted. Mark the table as full and
1536 * place the job back on the list of stopped jobs.
1538 DEBUGF(JOB, ("table full\n"));
1539 (void) Lst_AtFront(stoppedJobs, (void *)job);
1541 DEBUGF(JOB, ("Job queue is full.\n"));
1547 *-----------------------------------------------------------------------
1549 * Start a target-creation process going for the target described
1550 * by the graph node gn.
1553 * JOB_ERROR if there was an error in the commands, JOB_FINISHED
1554 * if there isn't actually anything left to do for the job and
1555 * JOB_RUNNING if the job has been started.
1558 * A new Job node is created and added to the list of running
1559 * jobs. PMake is forked and a child shell created.
1560 *-----------------------------------------------------------------------
1563 JobStart(GNode *gn, int flags, Job *previous)
1565 Job *job; /* new job descriptor */
1566 char *argv[4]; /* Argument vector to shell */
1567 Boolean cmdsOK; /* true if the nodes commands were all right */
1568 Boolean local; /* Set true if the job was run locally */
1569 Boolean noExec; /* Set true if we decide not to run the job */
1570 int tfd; /* File descriptor for temp file */
1572 if (previous != NULL) {
1573 previous->flags &= ~(JOB_FIRST|JOB_IGNERR|JOB_SILENT|JOB_REMOTE);
1576 job = (Job *) emalloc(sizeof(Job));
1581 job->tailCmds = NULL;
1584 * Set the initial value of the flags for this job based on the global
1585 * ones and the node's attributes... Any flags supplied by the caller
1586 * are also added to the field.
1589 if (Targ_Ignore(gn)) {
1590 job->flags |= JOB_IGNERR;
1592 if (Targ_Silent(gn)) {
1593 job->flags |= JOB_SILENT;
1595 job->flags |= flags;
1598 * Check the commands now so any attributes from .DEFAULT have a chance
1599 * to migrate to the node
1601 if (!compatMake && job->flags & JOB_FIRST) {
1602 cmdsOK = Job_CheckCommands(gn, Error);
1608 * If the -n flag wasn't given, we open up OUR (not the child's)
1609 * temporary file to stuff commands in it. The thing is rd/wr so we don't
1610 * need to reopen it to feed it to the shell. If the -n flag *was* given,
1611 * we just set the file to be stdout. Cute, huh?
1613 if ((gn->type & OP_MAKE) || (!noExecute && !touchFlag)) {
1615 * We're serious here, but if the commands were bogus, we're
1622 (void) strcpy(tfile, TMPPAT);
1623 if ((tfd = mkstemp(tfile)) == -1)
1624 Punt("Cannot create temp file: %s", strerror(errno));
1625 job->cmdFILE = fdopen(tfd, "w+");
1627 if (job->cmdFILE == NULL) {
1629 Punt("Could not open %s", tfile);
1631 (void) fcntl(FILENO(job->cmdFILE), F_SETFD, 1);
1633 * Send the commands to the command file, flush all its buffers then
1634 * rewind and remove the thing.
1639 * used to be backwards; replace when start doing multiple commands
1644 * Be compatible: If this is the first time for this node,
1645 * verify its commands are ok and open the commands list for
1646 * sequential access by later invocations of JobStart.
1647 * Once that is done, we take the next command off the list
1648 * and print it to the command file. If the command was an
1649 * ellipsis, note that there's nothing more to execute.
1651 if ((job->flags&JOB_FIRST) && (Lst_Open(gn->commands) != SUCCESS)){
1654 LstNode ln = Lst_Next(gn->commands);
1657 JobPrintCommand((void *) Lst_Datum(ln),
1661 Lst_Close(gn->commands);
1663 if (noExec && !(job->flags & JOB_FIRST)) {
1665 * If we're not going to execute anything, the job
1666 * is done and we need to close down the various
1667 * file descriptors we've opened for output, then
1668 * call JobDoOutput to catch the final characters or
1669 * send the file to the screen... Note that the i/o streams
1670 * are only open if this isn't the first job.
1671 * Note also that this could not be done in
1672 * Job_CatchChildren b/c it wasn't clear if there were
1673 * more commands to execute or not...
1680 * We can do all the commands at once. hooray for sanity
1683 Lst_ForEach(gn->commands, JobPrintCommand, (void *)job);
1686 * If we didn't print out any commands to the shell script,
1687 * there's not much point in executing the shell, is there?
1689 if (numCommands == 0) {
1693 } else if (noExecute) {
1695 * Not executing anything -- just print all the commands to stdout
1696 * in one fell swoop. This will still set up job->tailCmds correctly.
1698 if (lastNode != gn) {
1699 MESSAGE(stdout, gn);
1702 job->cmdFILE = stdout;
1704 * Only print the commands if they're ok, but don't die if they're
1705 * not -- just let the user know they're bad and keep going. It
1706 * doesn't do any harm in this case and may do some good.
1709 Lst_ForEach(gn->commands, JobPrintCommand, (void *)job);
1712 * Don't execute the shell, thank you.
1717 * Just touch the target and note that no shell should be executed.
1718 * Set cmdFILE to stdout to make life easier. Check the commands, too,
1719 * but don't die if they're no good -- it does no harm to keep working
1722 job->cmdFILE = stdout;
1723 Job_Touch(gn, job->flags&JOB_SILENT);
1728 * If we're not supposed to execute a shell, don't.
1732 * Unlink and close the command file if we opened one
1734 if (job->cmdFILE != stdout) {
1735 if (job->cmdFILE != NULL)
1736 (void) fclose(job->cmdFILE);
1738 (void) fflush(stdout);
1742 * We only want to work our way up the graph if we aren't here because
1743 * the commands for the job were no good.
1746 if (aborting == 0) {
1747 if (job->tailCmds != NULL) {
1748 Lst_ForEachFrom(job->node->commands, job->tailCmds,
1752 job->node->made = MADE;
1753 Make_Update(job->node);
1756 return(JOB_FINISHED);
1762 (void) fflush(job->cmdFILE);
1766 * Set up the control arguments to the shell. This is based on the flags
1767 * set earlier for this job.
1769 JobMakeArgv(job, argv);
1772 * If we're using pipes to catch output, create the pipe by which we'll
1773 * get the shell's output. If we're using files, print out that we're
1774 * starting a job and then set up its temporary-file name.
1776 if (!compatMake || (job->flags & JOB_FIRST)) {
1780 Punt("Cannot create pipe: %s", strerror(errno));
1781 job->inPipe = fd[0];
1782 job->outPipe = fd[1];
1783 (void) fcntl(job->inPipe, F_SETFD, 1);
1784 (void) fcntl(job->outPipe, F_SETFD, 1);
1786 (void) fprintf(stdout, "Remaking `%s'\n", gn->name);
1787 (void) fflush(stdout);
1788 (void) strcpy(job->outFile, TMPPAT);
1789 if ((job->outFd = mkstemp(job->outFile)) == -1)
1790 Punt("cannot create temp file: %s", strerror(errno));
1791 (void) fcntl(job->outFd, F_SETFD, 1);
1796 if (!(gn->type & OP_NOEXPORT) && !(runLocalFirst && nLocal < maxLocal)) {
1798 local = !Rmt_Export(shellPath, argv, job);
1800 local = !Rmt_Begin(shellPath, argv, job->node);
1801 #endif /* RMT_NO_EXEC */
1803 job->flags |= JOB_REMOTE;
1809 if (local && (((nLocal >= maxLocal) &&
1810 !(job->flags & JOB_SPECIAL) &&
1812 (!(gn->type & OP_NOEXPORT) || (maxLocal != 0))
1819 * The job can only be run locally, but we've hit the limit of
1820 * local concurrency, so put the job on hold until some other job
1821 * finishes. Note that the special jobs (.BEGIN, .INTERRUPT and .END)
1822 * may be run locally even when the local limit has been reached
1823 * (e.g. when maxLocal == 0), though they will be exported if at
1824 * all possible. In addition, any target marked with .NOEXPORT will
1825 * be run locally if maxLocal is 0.
1829 DEBUGF(JOB, ("Can only run job locally.\n"));
1830 job->flags |= JOB_RESTART;
1831 (void) Lst_AtEnd(stoppedJobs, (void *)job);
1833 if ((nLocal >= maxLocal) && local) {
1835 * If we're running this job locally as a special case (see above),
1836 * at least say the table is full.
1839 DEBUGF(JOB, ("Local job queue is full.\n"));
1843 return(JOB_RUNNING);
1847 JobOutput(Job *job, char *cp, char *endp, int msg)
1851 if (commandShell->noPrint) {
1852 ecp = strstr(cp, commandShell->noPrint);
1853 while (ecp != NULL) {
1856 if (msg && job->node != lastNode) {
1857 MESSAGE(stdout, job->node);
1858 lastNode = job->node;
1861 * The only way there wouldn't be a newline after
1862 * this line is if it were the last in the buffer.
1863 * however, since the non-printable comes after it,
1864 * there must be a newline, so we don't print one.
1866 (void) fprintf(stdout, "%s", cp);
1867 (void) fflush(stdout);
1869 cp = ecp + commandShell->noPLen;
1872 * Still more to print, look again after skipping
1873 * the whitespace following the non-printable
1877 while (*cp == ' ' || *cp == '\t' || *cp == '\n') {
1880 ecp = strstr(cp, commandShell->noPrint);
1890 *-----------------------------------------------------------------------
1892 * This function is called at different times depending on
1893 * whether the user has specified that output is to be collected
1894 * via pipes or temporary files. In the former case, we are called
1895 * whenever there is something to read on the pipe. We collect more
1896 * output from the given job and store it in the job's outBuf. If
1897 * this makes up a line, we print it tagged by the job's identifier,
1899 * If output has been collected in a temporary file, we open the
1900 * file and read it line by line, transfering it to our own
1901 * output channel until the file is empty. At which point we
1902 * remove the temporary file.
1903 * In both cases, however, we keep our figurative eye out for the
1904 * 'noPrint' line for the shell from which the output came. If
1905 * we recognize a line, we don't print it. If the command is not
1906 * alone on the line (the character after it is not \0 or \n), we
1907 * do print whatever follows it.
1913 * curPos may be shifted as may the contents of outBuf.
1914 *-----------------------------------------------------------------------
1917 JobDoOutput(Job *job, Boolean finish)
1919 Boolean gotNL = FALSE; /* true if got a newline */
1920 Boolean fbuf; /* true if our buffer filled up */
1921 int nr; /* number of bytes read */
1922 int i; /* auxiliary index into outBuf */
1923 int max; /* limit for i (end of current data) */
1924 int nRead; /* (Temporary) number of bytes read */
1926 FILE *oFILE; /* Stream pointer to shell's output file */
1932 * Read as many bytes as will fit in the buffer.
1938 nRead = read(job->inPipe, &job->outBuf[job->curPos],
1939 JOB_BUFSIZE - job->curPos);
1941 DEBUGF(JOB, ("JobDoOutput(piperead)"));
1948 * If we hit the end-of-file (the job is dead), we must flush its
1949 * remaining output, so pretend we read a newline if there's any
1950 * output remaining in the buffer.
1951 * Also clear the 'finish' flag so we stop looping.
1953 if ((nr == 0) && (job->curPos != 0)) {
1954 job->outBuf[job->curPos] = '\n';
1957 } else if (nr == 0) {
1962 * Look for the last newline in the bytes we just got. If there is
1963 * one, break out of the loop with 'i' as its index and gotNL set
1966 max = job->curPos + nr;
1967 for (i = job->curPos + nr - 1; i >= job->curPos; i--) {
1968 if (job->outBuf[i] == '\n') {
1971 } else if (job->outBuf[i] == '\0') {
1975 job->outBuf[i] = ' ';
1981 if (job->curPos == JOB_BUFSIZE) {
1983 * If we've run out of buffer space, we have no choice
1984 * but to print the stuff. sigh.
1990 if (gotNL || fbuf) {
1992 * Need to send the output to the screen. Null terminate it
1993 * first, overwriting the newline character if there was one.
1994 * So long as the line isn't one we should filter (according
1995 * to the shell description), we print the line, preceded
1996 * by a target banner if this target isn't the same as the
1997 * one for which we last printed something.
1998 * The rest of the data in the buffer are then shifted down
1999 * to the start of the buffer and curPos is set accordingly.
2001 job->outBuf[i] = '\0';
2002 if (i >= job->curPos) {
2005 cp = JobOutput(job, job->outBuf, &job->outBuf[i], FALSE);
2008 * There's still more in that thar buffer. This time, though,
2009 * we know there's no newline at the end, so we add one of
2010 * our own free will.
2013 if (job->node != lastNode) {
2014 MESSAGE(stdout, job->node);
2015 lastNode = job->node;
2017 (void) fprintf(stdout, "%s%s", cp, gotNL ? "\n" : "");
2018 (void) fflush(stdout);
2022 /* shift the remaining characters down */
2023 (void) memcpy(job->outBuf, &job->outBuf[i + 1], max - (i + 1));
2024 job->curPos = max - (i + 1);
2028 * We have written everything out, so we just start over
2029 * from the start of the buffer. No copying. No nothing.
2036 * If the finish flag is true, we must loop until we hit
2037 * end-of-file on the pipe. This is guaranteed to happen
2038 * eventually since the other end of the pipe is now closed
2039 * (we closed it explicitly and the child has exited). When
2040 * we do get an EOF, finish will be set FALSE and we'll fall
2047 * We've been called to retrieve the output of the job from the
2048 * temporary file where it's been squirreled away. This consists of
2049 * opening the file, reading the output line by line, being sure not
2050 * to print the noPrint line for the shell we used, then close and
2051 * remove the temporary file. Very simple.
2053 * Change to read in blocks and do FindSubString type things as for
2054 * pipes? That would allow for "@echo -n..."
2056 oFILE = fopen(job->outFile, "r");
2057 if (oFILE != NULL) {
2058 (void) fprintf(stdout, "Results of making %s:\n", job->node->name);
2059 (void) fflush(stdout);
2060 while (fgets(inLine, sizeof(inLine), oFILE) != NULL) {
2061 char *cp, *endp, *oendp;
2064 oendp = endp = inLine + strlen(inLine);
2065 if (endp[-1] == '\n') {
2068 cp = JobOutput(job, inLine, endp, FALSE);
2071 * There's still more in that thar buffer. This time, though,
2072 * we know there's no newline at the end, so we add one of
2073 * our own free will.
2075 (void) fprintf(stdout, "%s", cp);
2076 (void) fflush(stdout);
2077 if (endp != oendp) {
2078 (void) fprintf(stdout, "\n");
2079 (void) fflush(stdout);
2082 (void) fclose(oFILE);
2083 (void) eunlink(job->outFile);
2089 *-----------------------------------------------------------------------
2090 * Job_CatchChildren --
2091 * Handle the exit of a child. Called from Make_Make.
2097 * The job descriptor is removed from the list of children.
2100 * We do waits, blocking or not, according to the wisdom of our
2101 * caller, until there are no more children to report. For each
2102 * job, call JobFinish to finish things off. This will take care of
2103 * putting jobs on the stoppedJobs queue.
2105 *-----------------------------------------------------------------------
2108 Job_CatchChildren(Boolean block)
2110 int pid; /* pid of dead child */
2111 Job *job; /* job descriptor for dead child */
2112 LstNode jnode; /* list element for finding job */
2113 int status; /* Exit/termination status */
2116 * Don't even bother if we know there's no one around.
2122 while ((pid = waitpid((pid_t) -1, &status,
2123 (block?0:WNOHANG)|WUNTRACED)) > 0)
2125 DEBUGF(JOB, ("Process %d exited or stopped.\n", pid));
2127 jnode = Lst_Find(jobs, (void *)&pid, JobCmpPid);
2129 if (jnode == NULL) {
2130 if (WIFSIGNALED(status) && (WTERMSIG(status) == SIGCONT)) {
2131 jnode = Lst_Find(stoppedJobs, (void *) &pid, JobCmpPid);
2132 if (jnode == NULL) {
2133 Error("Resumed child (%d) not in table", pid);
2136 job = (Job *)Lst_Datum(jnode);
2137 (void) Lst_Remove(stoppedJobs, jnode);
2139 Error("Child (%d) not in table?", pid);
2143 job = (Job *) Lst_Datum(jnode);
2144 (void) Lst_Remove(jobs, jnode);
2146 DEBUGF(JOB, ("Job queue is no longer full.\n"));
2149 if (!(job->flags & JOB_REMOTE)) {
2150 DEBUGF(JOB, ("Job queue has one fewer local process.\n"));
2158 JobFinish(job, &status);
2163 *-----------------------------------------------------------------------
2164 * Job_CatchOutput --
2165 * Catch the output from our children, if we're using
2166 * pipes do so. Otherwise just block time until we get a
2167 * signal (most likely a SIGCHLD) since there's no point in
2168 * just spinning when there's nothing to do and the reaping
2169 * of a child can wait for a while.
2175 * Output is read from pipes if we're piping.
2176 * -----------------------------------------------------------------------
2179 Job_CatchOutput(void)
2182 struct timeval timeout;
2186 #ifdef RMT_WILL_WATCH
2187 int pnJobs; /* Previous nJobs */
2190 (void) fflush(stdout);
2191 #ifdef RMT_WILL_WATCH
2195 * It is possible for us to be called with nJobs equal to 0. This happens
2196 * if all the jobs finish and a job that is stopped cannot be run
2197 * locally (eg if maxLocal is 0) and cannot be exported. The job will
2198 * be placed back on the stoppedJobs queue, Job_Empty() will return false,
2199 * Make_Run will call us again when there's nothing for which to wait.
2200 * nJobs never changes, so we loop forever. Hence the check. It could
2201 * be argued that we should sleep for a bit so as not to swamp the
2202 * exportation system with requests. Perhaps we should.
2204 * NOTE: IT IS THE RESPONSIBILITY OF Rmt_Wait TO CALL Job_CatchChildren
2205 * IN A TIMELY FASHION TO CATCH ANY LOCALLY RUNNING JOBS THAT EXIT.
2206 * It may use the variable nLocal to determine if it needs to call
2207 * Job_CatchChildren (if nLocal is 0, there's nothing for which to
2210 while (nJobs != 0 && pnJobs == nJobs) {
2216 timeout.tv_sec = SEL_SEC;
2217 timeout.tv_usec = SEL_USEC;
2219 if ((nfds = select(FD_SETSIZE, &readfds, (fd_set *) 0,
2220 (fd_set *) 0, &timeout)) <= 0)
2223 if (Lst_Open(jobs) == FAILURE) {
2224 Punt("Cannot open job table");
2226 while (nfds && (ln = Lst_Next(jobs)) != NULL) {
2227 job = (Job *) Lst_Datum(ln);
2228 if (FD_ISSET(job->inPipe, &readfds)) {
2229 JobDoOutput(job, FALSE);
2236 #endif /* RMT_WILL_WATCH */
2240 *-----------------------------------------------------------------------
2242 * Start the creation of a target. Basically a front-end for
2243 * JobStart used by the Make module.
2249 * Another job is started.
2251 *-----------------------------------------------------------------------
2256 (void) JobStart(gn, 0, NULL);
2260 *-----------------------------------------------------------------------
2262 * Initialize the process module, given a maximum number of jobs, and
2263 * a maximum number of local jobs.
2269 * lists and counters are initialized
2270 *-----------------------------------------------------------------------
2273 Job_Init(int maxproc, int maxlocal)
2275 GNode *begin; /* node for commands to do at the very start */
2277 jobs = Lst_Init(FALSE);
2278 stoppedJobs = Lst_Init(FALSE);
2280 maxLocal = maxlocal;
2290 if (maxJobs == 1 || beVerbose == 0
2296 * If only one job can run at a time, there's no need for a banner,
2304 if (shellPath == NULL) {
2306 * The user didn't specify a shell to use, so we are using the
2307 * default one... Both the absolute path and the last component
2308 * must be set. The last component is taken from the 'name' field
2309 * of the default shell description pointed-to by commandShell.
2310 * All default shells are located in _PATH_DEFSHELLDIR.
2312 shellName = commandShell->name;
2313 shellPath = str_concat(_PATH_DEFSHELLDIR, shellName, STR_ADDSLASH);
2316 if (commandShell->exit == NULL) {
2317 commandShell->exit = "";
2319 if (commandShell->echo == NULL) {
2320 commandShell->echo = "";
2324 * Catch the four signals that POSIX specifies if they aren't ignored.
2325 * JobPassSig will take care of calling JobInterrupt if appropriate.
2327 if (signal(SIGINT, SIG_IGN) != SIG_IGN) {
2328 (void) signal(SIGINT, JobPassSig);
2330 if (signal(SIGHUP, SIG_IGN) != SIG_IGN) {
2331 (void) signal(SIGHUP, JobPassSig);
2333 if (signal(SIGQUIT, SIG_IGN) != SIG_IGN) {
2334 (void) signal(SIGQUIT, JobPassSig);
2336 if (signal(SIGTERM, SIG_IGN) != SIG_IGN) {
2337 (void) signal(SIGTERM, JobPassSig);
2340 * There are additional signals that need to be caught and passed if
2341 * either the export system wants to be told directly of signals or if
2342 * we're giving each job its own process group (since then it won't get
2343 * signals from the terminal driver as we own the terminal)
2345 #if defined(RMT_WANTS_SIGNALS) || defined(USE_PGRP)
2346 if (signal(SIGTSTP, SIG_IGN) != SIG_IGN) {
2347 (void) signal(SIGTSTP, JobPassSig);
2349 if (signal(SIGTTOU, SIG_IGN) != SIG_IGN) {
2350 (void) signal(SIGTTOU, JobPassSig);
2352 if (signal(SIGTTIN, SIG_IGN) != SIG_IGN) {
2353 (void) signal(SIGTTIN, JobPassSig);
2355 if (signal(SIGWINCH, SIG_IGN) != SIG_IGN) {
2356 (void) signal(SIGWINCH, JobPassSig);
2360 begin = Targ_FindNode(".BEGIN", TARG_NOCREATE);
2362 if (begin != NULL) {
2363 JobStart(begin, JOB_SPECIAL, (Job *)0);
2366 #ifndef RMT_WILL_WATCH
2367 Job_CatchChildren(!usePipes);
2368 #endif /* RMT_WILL_WATCH */
2371 postCommands = Targ_FindNode(".END", TARG_CREATE);
2375 *-----------------------------------------------------------------------
2377 * See if the job table is full. It is considered full if it is OR
2378 * if we are in the process of aborting OR if we have
2379 * reached/exceeded our local quota. This prevents any more jobs
2383 * TRUE if the job table is full, FALSE otherwise
2386 *-----------------------------------------------------------------------
2391 return(aborting || jobFull);
2395 *-----------------------------------------------------------------------
2397 * See if the job table is empty. Because the local concurrency may
2398 * be set to 0, it is possible for the job table to become empty,
2399 * while the list of stoppedJobs remains non-empty. In such a case,
2400 * we want to restart as many jobs as we can.
2403 * TRUE if it is. FALSE if it ain't.
2408 * -----------------------------------------------------------------------
2414 if (!Lst_IsEmpty(stoppedJobs) && !aborting) {
2416 * The job table is obviously not full if it has no jobs in
2417 * it...Try and restart the stopped jobs.
2431 *-----------------------------------------------------------------------
2433 * Find a matching shell in 'shells' given its final component.
2436 * A pointer to the Shell structure.
2441 *-----------------------------------------------------------------------
2444 JobMatchShell(char *name)
2446 Shell *sh; /* Pointer into shells table */
2447 Shell *match; /* Longest-matching shell */
2452 eoname = name + strlen(name);
2456 for (sh = shells; sh->name != NULL; sh++) {
2457 for (cp1 = eoname - strlen(sh->name), cp2 = sh->name;
2458 *cp1 != '\0' && *cp1 == *cp2;
2464 } else if (match == NULL || strlen(match->name) < strlen(sh->name)) {
2468 return(match == NULL ? sh : match);
2472 *-----------------------------------------------------------------------
2474 * Parse a shell specification and set up commandShell, shellPath
2475 * and shellName appropriately.
2478 * FAILURE if the specification was incorrect.
2481 * commandShell points to a Shell structure (either predefined or
2482 * created from the shell spec), shellPath is the full path of the
2483 * shell described by commandShell, while shellName is just the
2484 * final component of shellPath.
2487 * A shell specification consists of a .SHELL target, with dependency
2488 * operator, followed by a series of blank-separated words. Double
2489 * quotes can be used to use blanks in words. A backslash escapes
2490 * anything (most notably a double-quote and a space) and
2491 * provides the functionality it does in C. Each word consists of
2492 * keyword and value separated by an equal sign. There should be no
2493 * unnecessary spaces in the word. The keywords are as follows:
2494 * name Name of shell.
2495 * path Location of shell. Overrides "name" if given
2496 * quiet Command to turn off echoing.
2497 * echo Command to turn echoing on
2498 * filter Result of turning off echoing that shouldn't be
2500 * echoFlag Flag to turn echoing on at the start
2501 * errFlag Flag to turn error checking on at the start
2502 * hasErrCtl True if shell has error checking control
2503 * check Command to turn on error checking if hasErrCtl
2504 * is TRUE or template of command to echo a command
2505 * for which error checking is off if hasErrCtl is
2507 * ignore Command to turn off error checking if hasErrCtl
2508 * is TRUE or template of command to execute a
2509 * command so as to ignore any errors it returns if
2510 * hasErrCtl is FALSE.
2512 *-----------------------------------------------------------------------
2515 Job_ParseShell(char *line)
2523 Boolean fullSpec = FALSE;
2525 while (isspace((unsigned char) *line)) {
2528 words = brk_string(line, &wordCount, TRUE);
2530 memset(&newShell, 0, sizeof(newShell));
2533 * Parse the specification by keyword
2535 for (path = NULL, argc = wordCount - 1, argv = words + 1;
2538 if (strncmp(*argv, "path=", 5) == 0) {
2540 } else if (strncmp(*argv, "name=", 5) == 0) {
2541 newShell.name = &argv[0][5];
2543 if (strncmp(*argv, "quiet=", 6) == 0) {
2544 newShell.echoOff = &argv[0][6];
2545 } else if (strncmp(*argv, "echo=", 5) == 0) {
2546 newShell.echoOn = &argv[0][5];
2547 } else if (strncmp(*argv, "filter=", 7) == 0) {
2548 newShell.noPrint = &argv[0][7];
2549 newShell.noPLen = strlen(newShell.noPrint);
2550 } else if (strncmp(*argv, "echoFlag=", 9) == 0) {
2551 newShell.echo = &argv[0][9];
2552 } else if (strncmp(*argv, "errFlag=", 8) == 0) {
2553 newShell.exit = &argv[0][8];
2554 } else if (strncmp(*argv, "hasErrCtl=", 10) == 0) {
2555 char c = argv[0][10];
2556 newShell.hasErrCtl = !((c != 'Y') && (c != 'y') &&
2557 (c != 'T') && (c != 't'));
2558 } else if (strncmp(*argv, "check=", 6) == 0) {
2559 newShell.errCheck = &argv[0][6];
2560 } else if (strncmp(*argv, "ignore=", 7) == 0) {
2561 newShell.ignErr = &argv[0][7];
2563 Parse_Error(PARSE_FATAL, "Unknown keyword \"%s\"",
2573 * If no path was given, the user wants one of the pre-defined shells,
2574 * yes? So we find the one s/he wants with the help of JobMatchShell
2575 * and set things up the right way. shellPath will be set up by
2578 if (newShell.name == NULL) {
2579 Parse_Error(PARSE_FATAL, "Neither path nor name specified");
2582 commandShell = JobMatchShell(newShell.name);
2583 shellName = newShell.name;
2587 * The user provided a path. If s/he gave nothing else (fullSpec is
2588 * FALSE), try and find a matching shell in the ones we know of.
2589 * Else we just take the specification at its word and copy it
2590 * to a new location. In either case, we need to record the
2591 * path the user gave for the shell.
2594 path = strrchr(path, '/');
2600 if (newShell.name != NULL) {
2601 shellName = newShell.name;
2606 commandShell = JobMatchShell(shellName);
2608 commandShell = (Shell *) emalloc(sizeof(Shell));
2609 *commandShell = newShell;
2613 if (commandShell->echoOn && commandShell->echoOff) {
2614 commandShell->hasEchoCtl = TRUE;
2617 if (!commandShell->hasErrCtl) {
2618 if (commandShell->errCheck == NULL) {
2619 commandShell->errCheck = "";
2621 if (commandShell->ignErr == NULL) {
2622 commandShell->ignErr = "%s\n";
2629 *-----------------------------------------------------------------------
2631 * Handle the receipt of an interrupt.
2637 * All children are killed. Another job will be started if the
2638 * .INTERRUPT target was given.
2639 *-----------------------------------------------------------------------
2642 JobInterrupt(int runINTERRUPT, int signo)
2644 LstNode ln; /* element in job table */
2645 Job *job = NULL; /* job descriptor in that element */
2646 GNode *interrupt; /* the node describing the .INTERRUPT target */
2648 aborting = ABORT_INTERRUPT;
2650 (void) Lst_Open(jobs);
2651 while ((ln = Lst_Next(jobs)) != NULL) {
2652 job = (Job *) Lst_Datum(ln);
2654 if (!Targ_Precious(job->node)) {
2655 char *file = (job->node->path == NULL ?
2658 if (!noExecute && eunlink(file) != -1) {
2659 Error("*** %s removed", file);
2662 #ifdef RMT_WANTS_SIGNALS
2663 if (job->flags & JOB_REMOTE) {
2665 * If job is remote, let the Rmt module do the killing.
2667 if (!Rmt_Signal(job, signo)) {
2669 * If couldn't kill the thing, finish it out now with an
2670 * error code, since no exit report will come in likely.
2674 status.w_status = 0;
2675 status.w_retcode = 1;
2676 JobFinish(job, &status);
2678 } else if (job->pid) {
2679 KILL(job->pid, signo);
2683 DEBUGF(JOB, ("JobInterrupt passing signal to child %d.\n",
2685 KILL(job->pid, signo);
2687 #endif /* RMT_WANTS_SIGNALS */
2691 (void)Lst_Open(stoppedJobs);
2692 while ((ln = Lst_Next(stoppedJobs)) != NULL) {
2693 job = (Job *) Lst_Datum(ln);
2695 if (job->flags & JOB_RESTART) {
2696 DEBUGF(JOB, "JobInterrupt skipping job on stopped queue"
2697 "-- it was waiting to be restarted.\n");
2700 if (!Targ_Precious(job->node)) {
2701 char *file = (job->node->path == NULL ?
2704 if (eunlink(file) == 0) {
2705 Error("*** %s removed", file);
2709 * Resume the thing so it will take the signal.
2711 DEBUGF(JOB, ("JobInterrupt passing CONT to stopped child %d.\n", job->pid));
2712 KILL(job->pid, SIGCONT);
2713 #ifdef RMT_WANTS_SIGNALS
2714 if (job->flags & JOB_REMOTE) {
2716 * If job is remote, let the Rmt module do the killing.
2718 if (!Rmt_Signal(job, SIGINT)) {
2720 * If couldn't kill the thing, finish it out now with an
2721 * error code, since no exit report will come in likely.
2724 status.w_status = 0;
2725 status.w_retcode = 1;
2726 JobFinish(job, &status);
2728 } else if (job->pid) {
2729 DEBUGF(JOB, "JobInterrupt passing interrupt to stopped child %d.\n",
2731 KILL(job->pid, SIGINT);
2733 #endif /* RMT_WANTS_SIGNALS */
2736 Lst_Close(stoppedJobs);
2738 if (runINTERRUPT && !touchFlag) {
2739 interrupt = Targ_FindNode(".INTERRUPT", TARG_NOCREATE);
2740 if (interrupt != NULL) {
2741 ignoreErrors = FALSE;
2743 JobStart(interrupt, JOB_IGNDOTS, (Job *)0);
2746 #ifndef RMT_WILL_WATCH
2747 Job_CatchChildren(!usePipes);
2748 #endif /* RMT_WILL_WATCH */
2755 *-----------------------------------------------------------------------
2757 * Do final processing such as the running of the commands
2758 * attached to the .END target.
2761 * Number of errors reported.
2762 *-----------------------------------------------------------------------
2767 if (postCommands != NULL && !Lst_IsEmpty(postCommands->commands)) {
2769 Error("Errors reported so .END ignored");
2771 JobStart(postCommands, JOB_SPECIAL | JOB_IGNDOTS, NULL);
2775 #ifndef RMT_WILL_WATCH
2776 Job_CatchChildren(!usePipes);
2777 #endif /* RMT_WILL_WATCH */
2785 *-----------------------------------------------------------------------
2787 * Waits for all running jobs to finish and returns. Sets 'aborting'
2788 * to ABORT_WAIT to prevent other jobs from starting.
2794 * Currently running jobs finish.
2796 *-----------------------------------------------------------------------
2801 aborting = ABORT_WAIT;
2802 while (nJobs != 0) {
2804 #ifndef RMT_WILL_WATCH
2805 Job_CatchChildren(!usePipes);
2806 #endif /* RMT_WILL_WATCH */
2812 *-----------------------------------------------------------------------
2814 * Abort all currently running jobs without handling output or anything.
2815 * This function is to be called only in the event of a major
2816 * error. Most definitely NOT to be called from JobInterrupt.
2822 * All children are killed, not just the firstborn
2823 *-----------------------------------------------------------------------
2828 LstNode ln; /* element in job table */
2829 Job *job; /* the job descriptor in that element */
2832 aborting = ABORT_ERROR;
2836 (void) Lst_Open(jobs);
2837 while ((ln = Lst_Next(jobs)) != NULL) {
2838 job = (Job *) Lst_Datum(ln);
2841 * kill the child process with increasingly drastic signals to make
2842 * darn sure it's dead.
2844 #ifdef RMT_WANTS_SIGNALS
2845 if (job->flags & JOB_REMOTE) {
2846 Rmt_Signal(job, SIGINT);
2847 Rmt_Signal(job, SIGKILL);
2849 KILL(job->pid, SIGINT);
2850 KILL(job->pid, SIGKILL);
2853 KILL(job->pid, SIGINT);
2854 KILL(job->pid, SIGKILL);
2855 #endif /* RMT_WANTS_SIGNALS */
2860 * Catch as many children as want to report in at first, then give up
2862 while (waitpid((pid_t) -1, &foo, WNOHANG) > 0)
2868 *-----------------------------------------------------------------------
2869 * JobFlagForMigration --
2870 * Handle the eviction of a child. Called from RmtStatusChange.
2871 * Flags the child as remigratable and then suspends it. Takes
2872 * the ID of the host we used, for matching children.
2878 * The job descriptor is flagged for remigration.
2880 *-----------------------------------------------------------------------
2883 JobFlagForMigration(int hostID)
2885 Job *job; /* job descriptor for dead child */
2886 LstNode jnode; /* list element for finding job */
2888 DEBUGF(JOB, ("JobFlagForMigration(%d) called.\n", hostID));
2889 jnode = Lst_Find(jobs, (void *)hostID, JobCmpRmtID);
2891 if (jnode == NULL) {
2892 jnode = Lst_Find(stoppedJobs, (void *)hostID, JobCmpRmtID);
2893 if (jnode == NULL) {
2895 Error("Evicting host(%d) not in table", hostID);
2900 job = (Job *) Lst_Datum(jnode);
2902 DEBUGF(JOB, ("JobFlagForMigration(%d) found job '%s'.\n", hostID, job->node->name));
2904 KILL(job->pid, SIGSTOP);
2906 job->flags |= JOB_REMIGRATE;
2912 *-----------------------------------------------------------------------
2914 * Tries to restart stopped jobs if there are slots available.
2915 * Note that this tries to restart them regardless of pending errors.
2916 * It's not good to leave stopped jobs lying around!
2922 * Resumes(and possibly migrates) jobs.
2924 *-----------------------------------------------------------------------
2927 JobRestartJobs(void)
2929 while (!jobFull && !Lst_IsEmpty(stoppedJobs)) {
2930 DEBUGF(JOB, ("Job queue is not full. Restarting a stopped job.\n"));
2931 JobRestart((Job *)Lst_DeQueue(stoppedJobs));