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 1.17 2004/11/24 07:15:46 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..
906 (aborting != ABORT_ERROR) &&
907 (aborting != ABORT_INTERRUPT) &&
911 * As long as we aren't aborting and the job didn't return a non-zero
912 * status that we shouldn't ignore, we call Make_Update to update
913 * the parents. In addition, any saved commands for the node are placed
914 * on the .END target.
916 if (job->tailCmds != NULL) {
917 Lst_ForEachFrom(job->node->commands, job->tailCmds,
921 job->node->made = MADE;
922 Make_Update(job->node);
924 } else if (*status != 0) {
932 * Set aborting if any error.
934 if (errors && !keepgoing && (aborting != ABORT_INTERRUPT)) {
936 * If we found any errors in this batch of children and the -k flag
937 * wasn't given, we set the aborting flag so no more jobs get
940 aborting = ABORT_ERROR;
943 if ((aborting == ABORT_ERROR) && Job_Empty())
945 * If we are aborting and the job table is now empty, we finish.
951 *-----------------------------------------------------------------------
953 * Touch the given target. Called by JobStart when the -t flag was
954 * given. Prints messages unless told to be silent.
960 * The data modification of the file is changed. In addition, if the
961 * file did not exist, it is created.
962 *-----------------------------------------------------------------------
965 Job_Touch(GNode *gn, Boolean silent)
967 int streamID; /* ID of stream opened to do the touch */
968 struct utimbuf times; /* Times for utime() call */
970 if (gn->type & (OP_JOIN|OP_USE|OP_EXEC|OP_OPTIONAL)) {
972 * .JOIN, .USE, .ZEROTIME and .OPTIONAL targets are "virtual" targets
973 * and, as such, shouldn't really be created.
979 (void) fprintf(stdout, "touch %s\n", gn->name);
980 (void) fflush(stdout);
987 if (gn->type & OP_ARCHV) {
989 } else if (gn->type & OP_LIB) {
992 char *file = gn->path ? gn->path : gn->name;
994 times.actime = times.modtime = now;
995 if (utime(file, ×) < 0){
996 streamID = open(file, O_RDWR | O_CREAT, 0666);
1002 * Read and write a byte to the file to change the
1003 * modification time, then close the file.
1005 if (read(streamID, &c, 1) == 1) {
1006 (void) lseek(streamID, (off_t)0, SEEK_SET);
1007 (void) write(streamID, &c, 1);
1010 (void) close(streamID);
1012 (void) fprintf(stdout, "*** couldn't touch %s: %s",
1013 file, strerror(errno));
1014 (void) fflush(stdout);
1021 *-----------------------------------------------------------------------
1022 * Job_CheckCommands --
1023 * Make sure the given node has all the commands it needs.
1026 * TRUE if the commands list is/was ok.
1029 * The node will have commands from the .DEFAULT rule added to it
1031 *-----------------------------------------------------------------------
1034 Job_CheckCommands(GNode *gn, void (*abortProc)(const char *, ...))
1036 if (OP_NOP(gn->type) && Lst_IsEmpty(gn->commands) &&
1037 (gn->type & OP_LIB) == 0) {
1039 * No commands. Look for .DEFAULT rule from which we might infer
1042 if ((DEFAULT != NULL) && !Lst_IsEmpty(DEFAULT->commands)) {
1045 * Make only looks for a .DEFAULT if the node was never the
1046 * target of an operator, so that's what we do too. If
1047 * a .DEFAULT was given, we substitute its commands for gn's
1048 * commands and set the IMPSRC variable to be the target's name
1049 * The DEFAULT node acts like a transformation rule, in that
1050 * gn also inherits any attributes or sources attached to
1053 Make_HandleUse(DEFAULT, gn);
1054 Var_Set(IMPSRC, Var_Value(TARGET, gn, &p1), gn);
1056 } else if (Dir_MTime(gn) == 0) {
1058 * The node wasn't the target of an operator we have no .DEFAULT
1059 * rule to go on and the target doesn't already exist. There's
1060 * nothing more we can do for this branch. If the -k flag wasn't
1061 * given, we stop in our tracks, otherwise we just don't update
1062 * this node's parents so they never get examined.
1064 static const char msg[] = "make: don't know how to make";
1066 if (gn->type & OP_OPTIONAL) {
1067 (void) fprintf(stdout, "%s %s(ignored)\n", msg, gn->name);
1068 (void) fflush(stdout);
1069 } else if (keepgoing) {
1070 (void) fprintf(stdout, "%s %s(continuing)\n", msg, gn->name);
1071 (void) fflush(stdout);
1075 if (strcmp(gn->name,"love") == 0)
1076 (*abortProc)("Not war.");
1079 (*abortProc)("%s %s. Stop", msg, gn->name);
1086 #ifdef RMT_WILL_WATCH
1088 *-----------------------------------------------------------------------
1090 * Handle a pipe becoming readable. Callback function for Rmt_Watch
1096 * JobDoOutput is called.
1098 *-----------------------------------------------------------------------
1102 JobLocalInput(int stream, Job *job)
1104 JobDoOutput(job, FALSE);
1106 #endif /* RMT_WILL_WATCH */
1109 *-----------------------------------------------------------------------
1111 * Execute the shell for the given job. Called from JobStart and
1118 * A shell is executed, outputs is altered and the Job structure added
1121 *-----------------------------------------------------------------------
1124 JobExec(Job *job, char **argv)
1126 int cpid; /* ID of new child */
1131 DEBUGF(JOB, ("Running %s %sly\n", job->node->name,
1132 job->flags&JOB_REMOTE?"remote":"local"));
1133 DEBUGF(JOB, ("\tCommand: "));
1134 for (i = 0; argv[i] != NULL; i++) {
1135 DEBUGF(JOB, ("%s ", argv[i]));
1137 DEBUGF(JOB, ("\n"));
1141 * Some jobs produce no output and it's disconcerting to have
1142 * no feedback of their running (since they produce no output, the
1143 * banner with their name in it never appears). This is an attempt to
1144 * provide that feedback, even if nothing follows it.
1146 if ((lastNode != job->node) && (job->flags & JOB_FIRST) &&
1147 !(job->flags & JOB_SILENT)) {
1148 MESSAGE(stdout, job->node);
1149 lastNode = job->node;
1153 if (job->flags & JOB_REMOTE) {
1156 #endif /* RMT_NO_EXEC */
1158 if ((cpid = vfork()) == -1) {
1159 Punt("Cannot fork");
1160 } else if (cpid == 0) {
1163 * Must duplicate the input stream down to the child's input and
1164 * reset it to the beginning (again). Since the stream was marked
1165 * close-on-exec, we must clear that bit in the new input.
1167 if (dup2(FILENO(job->cmdFILE), 0) == -1)
1168 Punt("Cannot dup2: %s", strerror(errno));
1169 (void) fcntl(0, F_SETFD, 0);
1170 (void) lseek(0, (off_t)0, SEEK_SET);
1174 * Set up the child's output to be routed through the pipe
1175 * we've created for it.
1177 if (dup2(job->outPipe, 1) == -1)
1178 Punt("Cannot dup2: %s", strerror(errno));
1181 * We're capturing output in a file, so we duplicate the
1182 * descriptor to the temporary file into the standard
1185 if (dup2(job->outFd, 1) == -1)
1186 Punt("Cannot dup2: %s", strerror(errno));
1189 * The output channels are marked close on exec. This bit was
1190 * duplicated by the dup2 (on some systems), so we have to clear
1191 * it before routing the shell's error output to the same place as
1192 * its standard output.
1194 (void) fcntl(1, F_SETFD, 0);
1195 if (dup2(1, 2) == -1)
1196 Punt("Cannot dup2: %s", strerror(errno));
1200 * We want to switch the child into a different process family so
1201 * we can kill it and all its descendants in one fell swoop,
1202 * by killing its process family, but not commit suicide.
1207 (void) setpgid(0, getpid());
1209 #endif /* USE_PGRP */
1212 if (job->flags & JOB_REMOTE) {
1213 Rmt_Exec(shellPath, argv, FALSE);
1216 (void) execv(shellPath, argv);
1218 (void) write(STDERR_FILENO, "Could not execute shell\n",
1219 sizeof("Could not execute shell"));
1223 long omask = sigblock(sigmask(SIGCHLD));
1227 if (usePipes && (job->flags & JOB_FIRST) ) {
1229 * The first time a job is run for a node, we set the current
1230 * position in the buffer to the beginning and mark another
1231 * stream to watch in the outputs mask
1235 #ifdef RMT_WILL_WATCH
1236 Rmt_Watch(job->inPipe, JobLocalInput, job);
1238 FD_SET(job->inPipe, &outputs);
1239 #endif /* RMT_WILL_WATCH */
1242 if (job->flags & JOB_REMOTE) {
1246 job->rmtID = Rmt_LastID(job->pid);
1251 * XXX: Used to not happen if REMOTE. Why?
1253 if (job->cmdFILE != NULL && job->cmdFILE != stdout) {
1254 (void) fclose(job->cmdFILE);
1255 job->cmdFILE = NULL;
1259 (void) sigsetmask(omask);
1267 * Now the job is actually running, add it to the table.
1270 (void) Lst_AtEnd(jobs, (void *)job);
1271 if (nJobs == maxJobs) {
1277 *-----------------------------------------------------------------------
1279 * Create the argv needed to execute the shell for a given job.
1286 *-----------------------------------------------------------------------
1289 JobMakeArgv(Job *job, char **argv)
1292 static char args[10]; /* For merged arguments */
1294 argv[0] = shellName;
1297 if ((commandShell->exit && (*commandShell->exit != '-')) ||
1298 (commandShell->echo && (*commandShell->echo != '-')))
1301 * At least one of the flags doesn't have a minus before it, so
1302 * merge them together. Have to do this because the *(&(@*#*&#$#
1303 * Bourne shell thinks its second argument is a file to source.
1304 * Grrrr. Note the ten-character limitation on the combined arguments.
1306 (void)sprintf(args, "-%s%s",
1307 ((job->flags & JOB_IGNERR) ? "" :
1308 (commandShell->exit ? commandShell->exit : "")),
1309 ((job->flags & JOB_SILENT) ? "" :
1310 (commandShell->echo ? commandShell->echo : "")));
1317 if (!(job->flags & JOB_IGNERR) && commandShell->exit) {
1318 argv[argc] = commandShell->exit;
1321 if (!(job->flags & JOB_SILENT) && commandShell->echo) {
1322 argv[argc] = commandShell->echo;
1330 *-----------------------------------------------------------------------
1332 * Restart a job that stopped for some reason.
1338 * jobFull will be set if the job couldn't be run.
1340 *-----------------------------------------------------------------------
1343 JobRestart(Job *job)
1349 if (job->flags & JOB_REMIGRATE) {
1352 verboseRemigrates ||
1355 (void) fprintf(stdout, "*** remigrating %x(%s)\n",
1356 job->pid, job->node->name);
1357 (void) fflush(stdout);
1361 if (!Rmt_ReExport(job->pid, job->node, &host)) {
1362 if (verboseRemigrates || DEBUG(JOB)) {
1363 (void) fprintf(stdout, "*** couldn't migrate...\n");
1364 (void) fflush(stdout);
1367 if (nLocal != maxLocal) {
1369 * Job cannot be remigrated, but there's room on the local
1370 * machine, so resume the job and note that another
1371 * local job has started.
1375 verboseRemigrates ||
1378 (void) fprintf(stdout, "*** resuming on local machine\n");
1379 (void) fflush(stdout);
1381 KILL(job->pid, SIGCONT);
1384 job->flags &= ~(JOB_REMIGRATE|JOB_RESUME|JOB_REMOTE);
1385 job->flags |= JOB_CONTINUING;
1387 job->flags &= ~(JOB_REMIGRATE|JOB_RESUME);
1391 * Job cannot be restarted. Mark the table as full and
1392 * place the job back on the list of stopped jobs.
1396 verboseRemigrates ||
1399 (void) fprintf(stdout, "*** holding\n");
1400 (void) fflush(stdout);
1402 (void)Lst_AtFront(stoppedJobs, (void *)job);
1404 DEBUGF(JOB, ("Job queue is full.\n"));
1410 * Clear out the remigrate and resume flags. Set the continuing
1411 * flag so we know later on that the process isn't exiting just
1412 * because of a signal.
1414 job->flags &= ~(JOB_REMIGRATE|JOB_RESUME);
1415 job->flags |= JOB_CONTINUING;
1420 (void)Lst_AtEnd(jobs, (void *)job);
1422 if (nJobs == maxJobs) {
1424 DEBUGF(JOB, ("Job queue is full.\n"));
1426 } else if (job->flags & JOB_RESTART) {
1428 * Set up the control arguments to the shell. This is based on the
1429 * flags set earlier for this job. If the JOB_IGNERR flag is clear,
1430 * the 'exit' flag of the commandShell is used to cause it to exit
1431 * upon receiving an error. If the JOB_SILENT flag is clear, the
1432 * 'echo' flag of the commandShell is used to get it to start echoing
1433 * as soon as it starts processing commands.
1437 JobMakeArgv(job, argv);
1439 DEBUGF(JOB, ("Restarting %s...", job->node->name));
1441 if ((job->node->type&OP_NOEXPORT) ||
1442 (nLocal < maxLocal && runLocalFirst)
1444 || !Rmt_Export(shellPath, argv, job)
1446 || !Rmt_Begin(shellPath, argv, job->node)
1450 if (((nLocal >= maxLocal) && !(job->flags & JOB_SPECIAL))) {
1452 * Can't be exported and not allowed to run locally -- put it
1453 * back on the hold queue and mark the table full
1455 DEBUGF(JOB, ("holding\n"));
1456 (void)Lst_AtFront(stoppedJobs, (void *)job);
1458 DEBUGF(JOB, ("Job queue is full.\n"));
1462 * Job may be run locally.
1464 DEBUGF(JOB, ("running locally\n"));
1465 job->flags &= ~JOB_REMOTE;
1471 * Can be exported. Hooray!
1473 DEBUGF(JOB, ("exporting\n"));
1474 job->flags |= JOB_REMOTE;
1480 * The job has stopped and needs to be restarted. Why it stopped,
1483 DEBUGF(JOB, ("Resuming %s...", job->node->name));
1484 if (((job->flags & JOB_REMOTE) ||
1485 (nLocal < maxLocal) ||
1487 (((job->flags & JOB_SPECIAL) &&
1488 (job->node->type & OP_NOEXPORT)) &&
1489 (maxLocal == 0))) &&
1491 ((job->flags & JOB_SPECIAL) &&
1492 (maxLocal == 0))) &&
1497 * If the job is remote, it's ok to resume it as long as the
1498 * maximum concurrency won't be exceeded. If it's local and
1499 * we haven't reached the local concurrency limit already (or the
1500 * job must be run locally and maxLocal is 0), it's also ok to
1506 #ifdef RMT_WANTS_SIGNALS
1507 if (job->flags & JOB_REMOTE) {
1508 error = !Rmt_Signal(job, SIGCONT);
1510 #endif /* RMT_WANTS_SIGNALS */
1511 error = (KILL(job->pid, SIGCONT) != 0);
1515 * Make sure the user knows we've continued the beast and
1516 * actually put the thing in the job table.
1518 job->flags |= JOB_CONTINUING;
1519 W_SETTERMSIG(&status, SIGCONT);
1520 JobFinish(job, &status);
1522 job->flags &= ~(JOB_RESUME|JOB_CONTINUING);
1523 DEBUGF(JOB, ("done\n"));
1525 Error("couldn't resume %s: %s",
1526 job->node->name, strerror(errno));
1528 W_SETEXITSTATUS(&status, 1);
1529 JobFinish(job, &status);
1533 * Job cannot be restarted. Mark the table as full and
1534 * place the job back on the list of stopped jobs.
1536 DEBUGF(JOB, ("table full\n"));
1537 (void) Lst_AtFront(stoppedJobs, (void *)job);
1539 DEBUGF(JOB, ("Job queue is full.\n"));
1545 *-----------------------------------------------------------------------
1547 * Start a target-creation process going for the target described
1548 * by the graph node gn.
1551 * JOB_ERROR if there was an error in the commands, JOB_FINISHED
1552 * if there isn't actually anything left to do for the job and
1553 * JOB_RUNNING if the job has been started.
1556 * A new Job node is created and added to the list of running
1557 * jobs. PMake is forked and a child shell created.
1558 *-----------------------------------------------------------------------
1561 JobStart(GNode *gn, int flags, Job *previous)
1563 Job *job; /* new job descriptor */
1564 char *argv[4]; /* Argument vector to shell */
1565 Boolean cmdsOK; /* true if the nodes commands were all right */
1566 Boolean local; /* Set true if the job was run locally */
1567 Boolean noExec; /* Set true if we decide not to run the job */
1568 int tfd; /* File descriptor for temp file */
1570 if (previous != NULL) {
1571 previous->flags &= ~(JOB_FIRST|JOB_IGNERR|JOB_SILENT|JOB_REMOTE);
1574 job = (Job *) emalloc(sizeof(Job));
1579 job->tailCmds = NULL;
1582 * Set the initial value of the flags for this job based on the global
1583 * ones and the node's attributes... Any flags supplied by the caller
1584 * are also added to the field.
1587 if (Targ_Ignore(gn)) {
1588 job->flags |= JOB_IGNERR;
1590 if (Targ_Silent(gn)) {
1591 job->flags |= JOB_SILENT;
1593 job->flags |= flags;
1596 * Check the commands now so any attributes from .DEFAULT have a chance
1597 * to migrate to the node
1599 if (!compatMake && job->flags & JOB_FIRST) {
1600 cmdsOK = Job_CheckCommands(gn, Error);
1606 * If the -n flag wasn't given, we open up OUR (not the child's)
1607 * temporary file to stuff commands in it. The thing is rd/wr so we don't
1608 * need to reopen it to feed it to the shell. If the -n flag *was* given,
1609 * we just set the file to be stdout. Cute, huh?
1611 if ((gn->type & OP_MAKE) || (!noExecute && !touchFlag)) {
1613 * We're serious here, but if the commands were bogus, we're
1620 (void) strcpy(tfile, TMPPAT);
1621 if ((tfd = mkstemp(tfile)) == -1)
1622 Punt("Cannot create temp file: %s", strerror(errno));
1623 job->cmdFILE = fdopen(tfd, "w+");
1625 if (job->cmdFILE == NULL) {
1627 Punt("Could not open %s", tfile);
1629 (void) fcntl(FILENO(job->cmdFILE), F_SETFD, 1);
1631 * Send the commands to the command file, flush all its buffers then
1632 * rewind and remove the thing.
1637 * used to be backwards; replace when start doing multiple commands
1642 * Be compatible: If this is the first time for this node,
1643 * verify its commands are ok and open the commands list for
1644 * sequential access by later invocations of JobStart.
1645 * Once that is done, we take the next command off the list
1646 * and print it to the command file. If the command was an
1647 * ellipsis, note that there's nothing more to execute.
1649 if ((job->flags&JOB_FIRST) && (Lst_Open(gn->commands) != SUCCESS)){
1652 LstNode ln = Lst_Next(gn->commands);
1655 JobPrintCommand((void *) Lst_Datum(ln),
1659 Lst_Close(gn->commands);
1661 if (noExec && !(job->flags & JOB_FIRST)) {
1663 * If we're not going to execute anything, the job
1664 * is done and we need to close down the various
1665 * file descriptors we've opened for output, then
1666 * call JobDoOutput to catch the final characters or
1667 * send the file to the screen... Note that the i/o streams
1668 * are only open if this isn't the first job.
1669 * Note also that this could not be done in
1670 * Job_CatchChildren b/c it wasn't clear if there were
1671 * more commands to execute or not...
1678 * We can do all the commands at once. hooray for sanity
1681 Lst_ForEach(gn->commands, JobPrintCommand, (void *)job);
1684 * If we didn't print out any commands to the shell script,
1685 * there's not much point in executing the shell, is there?
1687 if (numCommands == 0) {
1691 } else if (noExecute) {
1693 * Not executing anything -- just print all the commands to stdout
1694 * in one fell swoop. This will still set up job->tailCmds correctly.
1696 if (lastNode != gn) {
1697 MESSAGE(stdout, gn);
1700 job->cmdFILE = stdout;
1702 * Only print the commands if they're ok, but don't die if they're
1703 * not -- just let the user know they're bad and keep going. It
1704 * doesn't do any harm in this case and may do some good.
1707 Lst_ForEach(gn->commands, JobPrintCommand, (void *)job);
1710 * Don't execute the shell, thank you.
1715 * Just touch the target and note that no shell should be executed.
1716 * Set cmdFILE to stdout to make life easier. Check the commands, too,
1717 * but don't die if they're no good -- it does no harm to keep working
1720 job->cmdFILE = stdout;
1721 Job_Touch(gn, job->flags&JOB_SILENT);
1726 * If we're not supposed to execute a shell, don't.
1730 * Unlink and close the command file if we opened one
1732 if (job->cmdFILE != stdout) {
1733 if (job->cmdFILE != NULL)
1734 (void) fclose(job->cmdFILE);
1736 (void) fflush(stdout);
1740 * We only want to work our way up the graph if we aren't here because
1741 * the commands for the job were no good.
1744 if (aborting == 0) {
1745 if (job->tailCmds != NULL) {
1746 Lst_ForEachFrom(job->node->commands, job->tailCmds,
1750 job->node->made = MADE;
1751 Make_Update(job->node);
1754 return(JOB_FINISHED);
1760 (void) fflush(job->cmdFILE);
1764 * Set up the control arguments to the shell. This is based on the flags
1765 * set earlier for this job.
1767 JobMakeArgv(job, argv);
1770 * If we're using pipes to catch output, create the pipe by which we'll
1771 * get the shell's output. If we're using files, print out that we're
1772 * starting a job and then set up its temporary-file name.
1774 if (!compatMake || (job->flags & JOB_FIRST)) {
1778 Punt("Cannot create pipe: %s", strerror(errno));
1779 job->inPipe = fd[0];
1780 job->outPipe = fd[1];
1781 (void) fcntl(job->inPipe, F_SETFD, 1);
1782 (void) fcntl(job->outPipe, F_SETFD, 1);
1784 (void) fprintf(stdout, "Remaking `%s'\n", gn->name);
1785 (void) fflush(stdout);
1786 (void) strcpy(job->outFile, TMPPAT);
1787 if ((job->outFd = mkstemp(job->outFile)) == -1)
1788 Punt("cannot create temp file: %s", strerror(errno));
1789 (void) fcntl(job->outFd, F_SETFD, 1);
1794 if (!(gn->type & OP_NOEXPORT) && !(runLocalFirst && nLocal < maxLocal)) {
1796 local = !Rmt_Export(shellPath, argv, job);
1798 local = !Rmt_Begin(shellPath, argv, job->node);
1799 #endif /* RMT_NO_EXEC */
1801 job->flags |= JOB_REMOTE;
1807 if (local && (((nLocal >= maxLocal) &&
1808 !(job->flags & JOB_SPECIAL) &&
1810 (!(gn->type & OP_NOEXPORT) || (maxLocal != 0))
1817 * The job can only be run locally, but we've hit the limit of
1818 * local concurrency, so put the job on hold until some other job
1819 * finishes. Note that the special jobs (.BEGIN, .INTERRUPT and .END)
1820 * may be run locally even when the local limit has been reached
1821 * (e.g. when maxLocal == 0), though they will be exported if at
1822 * all possible. In addition, any target marked with .NOEXPORT will
1823 * be run locally if maxLocal is 0.
1827 DEBUGF(JOB, ("Can only run job locally.\n"));
1828 job->flags |= JOB_RESTART;
1829 (void) Lst_AtEnd(stoppedJobs, (void *)job);
1831 if ((nLocal >= maxLocal) && local) {
1833 * If we're running this job locally as a special case (see above),
1834 * at least say the table is full.
1837 DEBUGF(JOB, ("Local job queue is full.\n"));
1841 return(JOB_RUNNING);
1845 JobOutput(Job *job, char *cp, char *endp, int msg)
1849 if (commandShell->noPrint) {
1850 ecp = strstr(cp, commandShell->noPrint);
1851 while (ecp != NULL) {
1854 if (msg && job->node != lastNode) {
1855 MESSAGE(stdout, job->node);
1856 lastNode = job->node;
1859 * The only way there wouldn't be a newline after
1860 * this line is if it were the last in the buffer.
1861 * however, since the non-printable comes after it,
1862 * there must be a newline, so we don't print one.
1864 (void) fprintf(stdout, "%s", cp);
1865 (void) fflush(stdout);
1867 cp = ecp + commandShell->noPLen;
1870 * Still more to print, look again after skipping
1871 * the whitespace following the non-printable
1875 while (*cp == ' ' || *cp == '\t' || *cp == '\n') {
1878 ecp = strstr(cp, commandShell->noPrint);
1888 *-----------------------------------------------------------------------
1890 * This function is called at different times depending on
1891 * whether the user has specified that output is to be collected
1892 * via pipes or temporary files. In the former case, we are called
1893 * whenever there is something to read on the pipe. We collect more
1894 * output from the given job and store it in the job's outBuf. If
1895 * this makes up a line, we print it tagged by the job's identifier,
1897 * If output has been collected in a temporary file, we open the
1898 * file and read it line by line, transfering it to our own
1899 * output channel until the file is empty. At which point we
1900 * remove the temporary file.
1901 * In both cases, however, we keep our figurative eye out for the
1902 * 'noPrint' line for the shell from which the output came. If
1903 * we recognize a line, we don't print it. If the command is not
1904 * alone on the line (the character after it is not \0 or \n), we
1905 * do print whatever follows it.
1911 * curPos may be shifted as may the contents of outBuf.
1912 *-----------------------------------------------------------------------
1915 JobDoOutput(Job *job, Boolean finish)
1917 Boolean gotNL = FALSE; /* true if got a newline */
1918 Boolean fbuf; /* true if our buffer filled up */
1919 int nr; /* number of bytes read */
1920 int i; /* auxiliary index into outBuf */
1921 int max; /* limit for i (end of current data) */
1922 int nRead; /* (Temporary) number of bytes read */
1924 FILE *oFILE; /* Stream pointer to shell's output file */
1930 * Read as many bytes as will fit in the buffer.
1936 nRead = read(job->inPipe, &job->outBuf[job->curPos],
1937 JOB_BUFSIZE - job->curPos);
1939 DEBUGF(JOB, ("JobDoOutput(piperead)"));
1946 * If we hit the end-of-file (the job is dead), we must flush its
1947 * remaining output, so pretend we read a newline if there's any
1948 * output remaining in the buffer.
1949 * Also clear the 'finish' flag so we stop looping.
1951 if ((nr == 0) && (job->curPos != 0)) {
1952 job->outBuf[job->curPos] = '\n';
1955 } else if (nr == 0) {
1960 * Look for the last newline in the bytes we just got. If there is
1961 * one, break out of the loop with 'i' as its index and gotNL set
1964 max = job->curPos + nr;
1965 for (i = job->curPos + nr - 1; i >= job->curPos; i--) {
1966 if (job->outBuf[i] == '\n') {
1969 } else if (job->outBuf[i] == '\0') {
1973 job->outBuf[i] = ' ';
1979 if (job->curPos == JOB_BUFSIZE) {
1981 * If we've run out of buffer space, we have no choice
1982 * but to print the stuff. sigh.
1988 if (gotNL || fbuf) {
1990 * Need to send the output to the screen. Null terminate it
1991 * first, overwriting the newline character if there was one.
1992 * So long as the line isn't one we should filter (according
1993 * to the shell description), we print the line, preceded
1994 * by a target banner if this target isn't the same as the
1995 * one for which we last printed something.
1996 * The rest of the data in the buffer are then shifted down
1997 * to the start of the buffer and curPos is set accordingly.
1999 job->outBuf[i] = '\0';
2000 if (i >= job->curPos) {
2003 cp = JobOutput(job, job->outBuf, &job->outBuf[i], FALSE);
2006 * There's still more in that thar buffer. This time, though,
2007 * we know there's no newline at the end, so we add one of
2008 * our own free will.
2011 if (job->node != lastNode) {
2012 MESSAGE(stdout, job->node);
2013 lastNode = job->node;
2015 (void) fprintf(stdout, "%s%s", cp, gotNL ? "\n" : "");
2016 (void) fflush(stdout);
2020 /* shift the remaining characters down */
2021 (void) memcpy(job->outBuf, &job->outBuf[i + 1], max - (i + 1));
2022 job->curPos = max - (i + 1);
2026 * We have written everything out, so we just start over
2027 * from the start of the buffer. No copying. No nothing.
2034 * If the finish flag is true, we must loop until we hit
2035 * end-of-file on the pipe. This is guaranteed to happen
2036 * eventually since the other end of the pipe is now closed
2037 * (we closed it explicitly and the child has exited). When
2038 * we do get an EOF, finish will be set FALSE and we'll fall
2045 * We've been called to retrieve the output of the job from the
2046 * temporary file where it's been squirreled away. This consists of
2047 * opening the file, reading the output line by line, being sure not
2048 * to print the noPrint line for the shell we used, then close and
2049 * remove the temporary file. Very simple.
2051 * Change to read in blocks and do FindSubString type things as for
2052 * pipes? That would allow for "@echo -n..."
2054 oFILE = fopen(job->outFile, "r");
2055 if (oFILE != NULL) {
2056 (void) fprintf(stdout, "Results of making %s:\n", job->node->name);
2057 (void) fflush(stdout);
2058 while (fgets(inLine, sizeof(inLine), oFILE) != NULL) {
2059 char *cp, *endp, *oendp;
2062 oendp = endp = inLine + strlen(inLine);
2063 if (endp[-1] == '\n') {
2066 cp = JobOutput(job, inLine, endp, FALSE);
2069 * There's still more in that thar buffer. This time, though,
2070 * we know there's no newline at the end, so we add one of
2071 * our own free will.
2073 (void) fprintf(stdout, "%s", cp);
2074 (void) fflush(stdout);
2075 if (endp != oendp) {
2076 (void) fprintf(stdout, "\n");
2077 (void) fflush(stdout);
2080 (void) fclose(oFILE);
2081 (void) eunlink(job->outFile);
2087 *-----------------------------------------------------------------------
2088 * Job_CatchChildren --
2089 * Handle the exit of a child. Called from Make_Make.
2095 * The job descriptor is removed from the list of children.
2098 * We do waits, blocking or not, according to the wisdom of our
2099 * caller, until there are no more children to report. For each
2100 * job, call JobFinish to finish things off. This will take care of
2101 * putting jobs on the stoppedJobs queue.
2103 *-----------------------------------------------------------------------
2106 Job_CatchChildren(Boolean block)
2108 int pid; /* pid of dead child */
2109 Job *job; /* job descriptor for dead child */
2110 LstNode jnode; /* list element for finding job */
2111 int status; /* Exit/termination status */
2114 * Don't even bother if we know there's no one around.
2120 while ((pid = waitpid((pid_t) -1, &status,
2121 (block?0:WNOHANG)|WUNTRACED)) > 0)
2123 DEBUGF(JOB, ("Process %d exited or stopped.\n", pid));
2125 jnode = Lst_Find(jobs, (void *)&pid, JobCmpPid);
2127 if (jnode == NULL) {
2128 if (WIFSIGNALED(status) && (WTERMSIG(status) == SIGCONT)) {
2129 jnode = Lst_Find(stoppedJobs, (void *) &pid, JobCmpPid);
2130 if (jnode == NULL) {
2131 Error("Resumed child (%d) not in table", pid);
2134 job = (Job *)Lst_Datum(jnode);
2135 (void) Lst_Remove(stoppedJobs, jnode);
2137 Error("Child (%d) not in table?", pid);
2141 job = (Job *) Lst_Datum(jnode);
2142 (void) Lst_Remove(jobs, jnode);
2144 DEBUGF(JOB, ("Job queue is no longer full.\n"));
2147 if (!(job->flags & JOB_REMOTE)) {
2148 DEBUGF(JOB, ("Job queue has one fewer local process.\n"));
2156 JobFinish(job, &status);
2161 *-----------------------------------------------------------------------
2162 * Job_CatchOutput --
2163 * Catch the output from our children, if we're using
2164 * pipes do so. Otherwise just block time until we get a
2165 * signal (most likely a SIGCHLD) since there's no point in
2166 * just spinning when there's nothing to do and the reaping
2167 * of a child can wait for a while.
2173 * Output is read from pipes if we're piping.
2174 * -----------------------------------------------------------------------
2177 Job_CatchOutput(void)
2180 struct timeval timeout;
2184 #ifdef RMT_WILL_WATCH
2185 int pnJobs; /* Previous nJobs */
2188 (void) fflush(stdout);
2189 #ifdef RMT_WILL_WATCH
2193 * It is possible for us to be called with nJobs equal to 0. This happens
2194 * if all the jobs finish and a job that is stopped cannot be run
2195 * locally (eg if maxLocal is 0) and cannot be exported. The job will
2196 * be placed back on the stoppedJobs queue, Job_Empty() will return false,
2197 * Make_Run will call us again when there's nothing for which to wait.
2198 * nJobs never changes, so we loop forever. Hence the check. It could
2199 * be argued that we should sleep for a bit so as not to swamp the
2200 * exportation system with requests. Perhaps we should.
2202 * NOTE: IT IS THE RESPONSIBILITY OF Rmt_Wait TO CALL Job_CatchChildren
2203 * IN A TIMELY FASHION TO CATCH ANY LOCALLY RUNNING JOBS THAT EXIT.
2204 * It may use the variable nLocal to determine if it needs to call
2205 * Job_CatchChildren (if nLocal is 0, there's nothing for which to
2208 while (nJobs != 0 && pnJobs == nJobs) {
2214 timeout.tv_sec = SEL_SEC;
2215 timeout.tv_usec = SEL_USEC;
2217 if ((nfds = select(FD_SETSIZE, &readfds, (fd_set *) 0,
2218 (fd_set *) 0, &timeout)) <= 0)
2221 if (Lst_Open(jobs) == FAILURE) {
2222 Punt("Cannot open job table");
2224 while (nfds && (ln = Lst_Next(jobs)) != NULL) {
2225 job = (Job *) Lst_Datum(ln);
2226 if (FD_ISSET(job->inPipe, &readfds)) {
2227 JobDoOutput(job, FALSE);
2234 #endif /* RMT_WILL_WATCH */
2238 *-----------------------------------------------------------------------
2240 * Start the creation of a target. Basically a front-end for
2241 * JobStart used by the Make module.
2247 * Another job is started.
2249 *-----------------------------------------------------------------------
2254 (void) JobStart(gn, 0, NULL);
2258 *-----------------------------------------------------------------------
2260 * Initialize the process module, given a maximum number of jobs, and
2261 * a maximum number of local jobs.
2267 * lists and counters are initialized
2268 *-----------------------------------------------------------------------
2271 Job_Init(int maxproc, int maxlocal)
2273 GNode *begin; /* node for commands to do at the very start */
2275 jobs = Lst_Init(FALSE);
2276 stoppedJobs = Lst_Init(FALSE);
2278 maxLocal = maxlocal;
2288 if (maxJobs == 1 || beVerbose == 0
2294 * If only one job can run at a time, there's no need for a banner,
2302 if (shellPath == NULL) {
2304 * The user didn't specify a shell to use, so we are using the
2305 * default one... Both the absolute path and the last component
2306 * must be set. The last component is taken from the 'name' field
2307 * of the default shell description pointed-to by commandShell.
2308 * All default shells are located in _PATH_DEFSHELLDIR.
2310 shellName = commandShell->name;
2311 shellPath = str_concat(_PATH_DEFSHELLDIR, shellName, STR_ADDSLASH);
2314 if (commandShell->exit == NULL) {
2315 commandShell->exit = "";
2317 if (commandShell->echo == NULL) {
2318 commandShell->echo = "";
2322 * Catch the four signals that POSIX specifies if they aren't ignored.
2323 * JobPassSig will take care of calling JobInterrupt if appropriate.
2325 if (signal(SIGINT, SIG_IGN) != SIG_IGN) {
2326 (void) signal(SIGINT, JobPassSig);
2328 if (signal(SIGHUP, SIG_IGN) != SIG_IGN) {
2329 (void) signal(SIGHUP, JobPassSig);
2331 if (signal(SIGQUIT, SIG_IGN) != SIG_IGN) {
2332 (void) signal(SIGQUIT, JobPassSig);
2334 if (signal(SIGTERM, SIG_IGN) != SIG_IGN) {
2335 (void) signal(SIGTERM, JobPassSig);
2338 * There are additional signals that need to be caught and passed if
2339 * either the export system wants to be told directly of signals or if
2340 * we're giving each job its own process group (since then it won't get
2341 * signals from the terminal driver as we own the terminal)
2343 #if defined(RMT_WANTS_SIGNALS) || defined(USE_PGRP)
2344 if (signal(SIGTSTP, SIG_IGN) != SIG_IGN) {
2345 (void) signal(SIGTSTP, JobPassSig);
2347 if (signal(SIGTTOU, SIG_IGN) != SIG_IGN) {
2348 (void) signal(SIGTTOU, JobPassSig);
2350 if (signal(SIGTTIN, SIG_IGN) != SIG_IGN) {
2351 (void) signal(SIGTTIN, JobPassSig);
2353 if (signal(SIGWINCH, SIG_IGN) != SIG_IGN) {
2354 (void) signal(SIGWINCH, JobPassSig);
2358 begin = Targ_FindNode(".BEGIN", TARG_NOCREATE);
2360 if (begin != NULL) {
2361 JobStart(begin, JOB_SPECIAL, (Job *)0);
2364 #ifndef RMT_WILL_WATCH
2365 Job_CatchChildren(!usePipes);
2366 #endif /* RMT_WILL_WATCH */
2369 postCommands = Targ_FindNode(".END", TARG_CREATE);
2373 *-----------------------------------------------------------------------
2375 * See if the job table is full. It is considered full if it is OR
2376 * if we are in the process of aborting OR if we have
2377 * reached/exceeded our local quota. This prevents any more jobs
2381 * TRUE if the job table is full, FALSE otherwise
2384 *-----------------------------------------------------------------------
2389 return(aborting || jobFull);
2393 *-----------------------------------------------------------------------
2395 * See if the job table is empty. Because the local concurrency may
2396 * be set to 0, it is possible for the job table to become empty,
2397 * while the list of stoppedJobs remains non-empty. In such a case,
2398 * we want to restart as many jobs as we can.
2401 * TRUE if it is. FALSE if it ain't.
2406 * -----------------------------------------------------------------------
2412 if (!Lst_IsEmpty(stoppedJobs) && !aborting) {
2414 * The job table is obviously not full if it has no jobs in
2415 * it...Try and restart the stopped jobs.
2429 *-----------------------------------------------------------------------
2431 * Find a matching shell in 'shells' given its final component.
2434 * A pointer to the Shell structure.
2439 *-----------------------------------------------------------------------
2442 JobMatchShell(char *name)
2444 Shell *sh; /* Pointer into shells table */
2445 Shell *match; /* Longest-matching shell */
2450 eoname = name + strlen(name);
2454 for (sh = shells; sh->name != NULL; sh++) {
2455 for (cp1 = eoname - strlen(sh->name), cp2 = sh->name;
2456 *cp1 != '\0' && *cp1 == *cp2;
2462 } else if (match == NULL || strlen(match->name) < strlen(sh->name)) {
2466 return(match == NULL ? sh : match);
2470 *-----------------------------------------------------------------------
2472 * Parse a shell specification and set up commandShell, shellPath
2473 * and shellName appropriately.
2476 * FAILURE if the specification was incorrect.
2479 * commandShell points to a Shell structure (either predefined or
2480 * created from the shell spec), shellPath is the full path of the
2481 * shell described by commandShell, while shellName is just the
2482 * final component of shellPath.
2485 * A shell specification consists of a .SHELL target, with dependency
2486 * operator, followed by a series of blank-separated words. Double
2487 * quotes can be used to use blanks in words. A backslash escapes
2488 * anything (most notably a double-quote and a space) and
2489 * provides the functionality it does in C. Each word consists of
2490 * keyword and value separated by an equal sign. There should be no
2491 * unnecessary spaces in the word. The keywords are as follows:
2492 * name Name of shell.
2493 * path Location of shell. Overrides "name" if given
2494 * quiet Command to turn off echoing.
2495 * echo Command to turn echoing on
2496 * filter Result of turning off echoing that shouldn't be
2498 * echoFlag Flag to turn echoing on at the start
2499 * errFlag Flag to turn error checking on at the start
2500 * hasErrCtl True if shell has error checking control
2501 * check Command to turn on error checking if hasErrCtl
2502 * is TRUE or template of command to echo a command
2503 * for which error checking is off if hasErrCtl is
2505 * ignore Command to turn off error checking if hasErrCtl
2506 * is TRUE or template of command to execute a
2507 * command so as to ignore any errors it returns if
2508 * hasErrCtl is FALSE.
2510 *-----------------------------------------------------------------------
2513 Job_ParseShell(char *line)
2521 Boolean fullSpec = FALSE;
2523 while (isspace((unsigned char) *line)) {
2526 words = brk_string(line, &wordCount, TRUE);
2528 memset(&newShell, 0, sizeof(newShell));
2531 * Parse the specification by keyword
2533 for (path = NULL, argc = wordCount - 1, argv = words + 1;
2536 if (strncmp(*argv, "path=", 5) == 0) {
2538 } else if (strncmp(*argv, "name=", 5) == 0) {
2539 newShell.name = &argv[0][5];
2541 if (strncmp(*argv, "quiet=", 6) == 0) {
2542 newShell.echoOff = &argv[0][6];
2543 } else if (strncmp(*argv, "echo=", 5) == 0) {
2544 newShell.echoOn = &argv[0][5];
2545 } else if (strncmp(*argv, "filter=", 7) == 0) {
2546 newShell.noPrint = &argv[0][7];
2547 newShell.noPLen = strlen(newShell.noPrint);
2548 } else if (strncmp(*argv, "echoFlag=", 9) == 0) {
2549 newShell.echo = &argv[0][9];
2550 } else if (strncmp(*argv, "errFlag=", 8) == 0) {
2551 newShell.exit = &argv[0][8];
2552 } else if (strncmp(*argv, "hasErrCtl=", 10) == 0) {
2553 char c = argv[0][10];
2554 newShell.hasErrCtl = !((c != 'Y') && (c != 'y') &&
2555 (c != 'T') && (c != 't'));
2556 } else if (strncmp(*argv, "check=", 6) == 0) {
2557 newShell.errCheck = &argv[0][6];
2558 } else if (strncmp(*argv, "ignore=", 7) == 0) {
2559 newShell.ignErr = &argv[0][7];
2561 Parse_Error(PARSE_FATAL, "Unknown keyword \"%s\"",
2571 * If no path was given, the user wants one of the pre-defined shells,
2572 * yes? So we find the one s/he wants with the help of JobMatchShell
2573 * and set things up the right way. shellPath will be set up by
2576 if (newShell.name == NULL) {
2577 Parse_Error(PARSE_FATAL, "Neither path nor name specified");
2580 commandShell = JobMatchShell(newShell.name);
2581 shellName = newShell.name;
2585 * The user provided a path. If s/he gave nothing else (fullSpec is
2586 * FALSE), try and find a matching shell in the ones we know of.
2587 * Else we just take the specification at its word and copy it
2588 * to a new location. In either case, we need to record the
2589 * path the user gave for the shell.
2592 path = strrchr(path, '/');
2598 if (newShell.name != NULL) {
2599 shellName = newShell.name;
2604 commandShell = JobMatchShell(shellName);
2606 commandShell = (Shell *) emalloc(sizeof(Shell));
2607 *commandShell = newShell;
2611 if (commandShell->echoOn && commandShell->echoOff) {
2612 commandShell->hasEchoCtl = TRUE;
2615 if (!commandShell->hasErrCtl) {
2616 if (commandShell->errCheck == NULL) {
2617 commandShell->errCheck = "";
2619 if (commandShell->ignErr == NULL) {
2620 commandShell->ignErr = "%s\n";
2627 *-----------------------------------------------------------------------
2629 * Handle the receipt of an interrupt.
2635 * All children are killed. Another job will be started if the
2636 * .INTERRUPT target was given.
2637 *-----------------------------------------------------------------------
2640 JobInterrupt(int runINTERRUPT, int signo)
2642 LstNode ln; /* element in job table */
2643 Job *job = NULL; /* job descriptor in that element */
2644 GNode *interrupt; /* the node describing the .INTERRUPT target */
2646 aborting = ABORT_INTERRUPT;
2648 (void) Lst_Open(jobs);
2649 while ((ln = Lst_Next(jobs)) != NULL) {
2650 job = (Job *) Lst_Datum(ln);
2652 if (!Targ_Precious(job->node)) {
2653 char *file = (job->node->path == NULL ?
2656 if (!noExecute && eunlink(file) != -1) {
2657 Error("*** %s removed", file);
2660 #ifdef RMT_WANTS_SIGNALS
2661 if (job->flags & JOB_REMOTE) {
2663 * If job is remote, let the Rmt module do the killing.
2665 if (!Rmt_Signal(job, signo)) {
2667 * If couldn't kill the thing, finish it out now with an
2668 * error code, since no exit report will come in likely.
2672 status.w_status = 0;
2673 status.w_retcode = 1;
2674 JobFinish(job, &status);
2676 } else if (job->pid) {
2677 KILL(job->pid, signo);
2681 DEBUGF(JOB, ("JobInterrupt passing signal to child %d.\n",
2683 KILL(job->pid, signo);
2685 #endif /* RMT_WANTS_SIGNALS */
2689 (void)Lst_Open(stoppedJobs);
2690 while ((ln = Lst_Next(stoppedJobs)) != NULL) {
2691 job = (Job *) Lst_Datum(ln);
2693 if (job->flags & JOB_RESTART) {
2694 DEBUGF(JOB, "JobInterrupt skipping job on stopped queue"
2695 "-- it was waiting to be restarted.\n");
2698 if (!Targ_Precious(job->node)) {
2699 char *file = (job->node->path == NULL ?
2702 if (eunlink(file) == 0) {
2703 Error("*** %s removed", file);
2707 * Resume the thing so it will take the signal.
2709 DEBUGF(JOB, ("JobInterrupt passing CONT to stopped child %d.\n", job->pid));
2710 KILL(job->pid, SIGCONT);
2711 #ifdef RMT_WANTS_SIGNALS
2712 if (job->flags & JOB_REMOTE) {
2714 * If job is remote, let the Rmt module do the killing.
2716 if (!Rmt_Signal(job, SIGINT)) {
2718 * If couldn't kill the thing, finish it out now with an
2719 * error code, since no exit report will come in likely.
2722 status.w_status = 0;
2723 status.w_retcode = 1;
2724 JobFinish(job, &status);
2726 } else if (job->pid) {
2727 DEBUGF(JOB, "JobInterrupt passing interrupt to stopped child %d.\n",
2729 KILL(job->pid, SIGINT);
2731 #endif /* RMT_WANTS_SIGNALS */
2734 Lst_Close(stoppedJobs);
2736 if (runINTERRUPT && !touchFlag) {
2737 interrupt = Targ_FindNode(".INTERRUPT", TARG_NOCREATE);
2738 if (interrupt != NULL) {
2739 ignoreErrors = FALSE;
2741 JobStart(interrupt, JOB_IGNDOTS, (Job *)0);
2744 #ifndef RMT_WILL_WATCH
2745 Job_CatchChildren(!usePipes);
2746 #endif /* RMT_WILL_WATCH */
2753 *-----------------------------------------------------------------------
2755 * Do final processing such as the running of the commands
2756 * attached to the .END target.
2759 * Number of errors reported.
2760 *-----------------------------------------------------------------------
2765 if (postCommands != NULL && !Lst_IsEmpty(postCommands->commands)) {
2767 Error("Errors reported so .END ignored");
2769 JobStart(postCommands, JOB_SPECIAL | JOB_IGNDOTS, NULL);
2773 #ifndef RMT_WILL_WATCH
2774 Job_CatchChildren(!usePipes);
2775 #endif /* RMT_WILL_WATCH */
2783 *-----------------------------------------------------------------------
2785 * Waits for all running jobs to finish and returns. Sets 'aborting'
2786 * to ABORT_WAIT to prevent other jobs from starting.
2792 * Currently running jobs finish.
2794 *-----------------------------------------------------------------------
2799 aborting = ABORT_WAIT;
2800 while (nJobs != 0) {
2802 #ifndef RMT_WILL_WATCH
2803 Job_CatchChildren(!usePipes);
2804 #endif /* RMT_WILL_WATCH */
2810 *-----------------------------------------------------------------------
2812 * Abort all currently running jobs without handling output or anything.
2813 * This function is to be called only in the event of a major
2814 * error. Most definitely NOT to be called from JobInterrupt.
2820 * All children are killed, not just the firstborn
2821 *-----------------------------------------------------------------------
2826 LstNode ln; /* element in job table */
2827 Job *job; /* the job descriptor in that element */
2830 aborting = ABORT_ERROR;
2834 (void) Lst_Open(jobs);
2835 while ((ln = Lst_Next(jobs)) != NULL) {
2836 job = (Job *) Lst_Datum(ln);
2839 * kill the child process with increasingly drastic signals to make
2840 * darn sure it's dead.
2842 #ifdef RMT_WANTS_SIGNALS
2843 if (job->flags & JOB_REMOTE) {
2844 Rmt_Signal(job, SIGINT);
2845 Rmt_Signal(job, SIGKILL);
2847 KILL(job->pid, SIGINT);
2848 KILL(job->pid, SIGKILL);
2851 KILL(job->pid, SIGINT);
2852 KILL(job->pid, SIGKILL);
2853 #endif /* RMT_WANTS_SIGNALS */
2858 * Catch as many children as want to report in at first, then give up
2860 while (waitpid((pid_t) -1, &foo, WNOHANG) > 0)
2866 *-----------------------------------------------------------------------
2867 * JobFlagForMigration --
2868 * Handle the eviction of a child. Called from RmtStatusChange.
2869 * Flags the child as remigratable and then suspends it. Takes
2870 * the ID of the host we used, for matching children.
2876 * The job descriptor is flagged for remigration.
2878 *-----------------------------------------------------------------------
2881 JobFlagForMigration(int hostID)
2883 Job *job; /* job descriptor for dead child */
2884 LstNode jnode; /* list element for finding job */
2886 DEBUGF(JOB, ("JobFlagForMigration(%d) called.\n", hostID));
2887 jnode = Lst_Find(jobs, (void *)hostID, JobCmpRmtID);
2889 if (jnode == NULL) {
2890 jnode = Lst_Find(stoppedJobs, (void *)hostID, JobCmpRmtID);
2891 if (jnode == NULL) {
2893 Error("Evicting host(%d) not in table", hostID);
2898 job = (Job *) Lst_Datum(jnode);
2900 DEBUGF(JOB, ("JobFlagForMigration(%d) found job '%s'.\n", hostID, job->node->name));
2902 KILL(job->pid, SIGSTOP);
2904 job->flags |= JOB_REMIGRATE;
2910 *-----------------------------------------------------------------------
2912 * Tries to restart stopped jobs if there are slots available.
2913 * Note that this tries to restart them regardless of pending errors.
2914 * It's not good to leave stopped jobs lying around!
2920 * Resumes(and possibly migrates) jobs.
2922 *-----------------------------------------------------------------------
2925 JobRestartJobs(void)
2927 while (!jobFull && !Lst_IsEmpty(stoppedJobs)) {
2928 DEBUGF(JOB, ("Job queue is not full. Restarting a stopped job.\n"));
2929 JobRestart((Job *)Lst_DeQueue(stoppedJobs));