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 1.14 2004/11/13 22:42:39 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>
119 #include "pathnames.h"
124 # define STATIC static
128 * error handling variables
130 static int errors = 0; /* number of errors reported */
131 static int aborting = 0; /* why is the make aborting? */
132 #define ABORT_ERROR 1 /* Because of an error */
133 #define ABORT_INTERRUPT 2 /* Because it was interrupted */
134 #define ABORT_WAIT 3 /* Waiting for jobs to finish */
137 * XXX: Avoid SunOS bug... FILENO() is fp->_file, and file
138 * is a char! So when we go above 127 we turn negative!
140 #define FILENO(a) ((unsigned) fileno(a))
143 * post-make command processing. The node postCommands is really just the
144 * .END target but we keep it around to avoid having to search for it
147 static GNode *postCommands; /* node containing commands to execute when
148 * everything else is done */
149 static int numCommands; /* The number of commands actually printed
150 * for a target. Should this number be
151 * 0, no shell will be executed. */
154 * Return values from JobStart.
156 #define JOB_RUNNING 0 /* Job is running */
157 #define JOB_ERROR 1 /* Error in starting the job */
158 #define JOB_FINISHED 2 /* The job is already finished */
159 #define JOB_STOPPED 3 /* The job is stopped */
162 * tfile is used to build temp file names to store shell commands to
165 static char tfile[sizeof(TMPPAT)];
169 * Descriptions for various shells.
171 static Shell shells[] = {
173 * CSH description. The csh can do echo control by playing
174 * with the setting of the 'echo' shell variable. Sadly,
175 * however, it is unable to do error control nicely.
179 TRUE, "unset verbose", "set verbose", "unset verbose", 10,
180 FALSE, "echo \"%s\"\n", "csh -c \"%s || exit 0\"",
184 * SH description. Echo control is also possible and, under
185 * sun UNIX anyway, one can even control error checking.
189 TRUE, "set -", "set -v", "set -", 5,
190 TRUE, "set -e", "set +e",
191 #ifdef OLDBOURNESHELL
192 FALSE, "echo \"%s\"\n", "sh -c '%s || exit 0'\n",
201 FALSE, (char *) 0, (char *) 0, (char *) 0, 0,
202 FALSE, (char *) 0, (char *) 0,
203 (char *) 0, (char *) 0,
206 static Shell *commandShell = &shells[DEFSHELL];/* this is the shell to
208 * commands in the Makefile.
210 * Job_ParseShell function */
211 static char *shellPath = NULL, /* full pathname of
212 * executable image */
213 *shellName; /* last component of shell */
216 static int maxJobs; /* The most children we can run at once */
217 static int maxLocal; /* The most local ones we can have */
218 STATIC int nJobs; /* The number of children currently running */
219 STATIC int nLocal; /* The number of local children */
220 STATIC Lst jobs; /* The structures that describe them */
221 STATIC Boolean jobFull; /* Flag to tell when the job table is full. It
222 * is set TRUE when (1) the total number of
223 * running jobs equals the maximum allowed or
224 * (2) a job can only be run locally, but
225 * nLocal equals maxLocal */
226 #ifndef RMT_WILL_WATCH
227 static fd_set outputs; /* Set of descriptors of pipes connected to
228 * the output channels of children */
231 STATIC GNode *lastNode; /* The node for which output was most recently
233 STATIC char *targFmt; /* Format string to use to head output from a
234 * job when it's not the most-recent job heard
238 # define TARG_FMT "--- %s at %s ---\n" /* Default format */
239 # define MESSAGE(fp, gn) \
240 (void) fprintf(fp, targFmt, gn->name, gn->rem.hname);
242 # define TARG_FMT "--- %s ---\n" /* Default format */
243 # define MESSAGE(fp, gn) \
244 (void) fprintf(fp, targFmt, gn->name);
248 * When JobStart attempts to run a job remotely but can't, and isn't allowed
249 * to run the job locally, or when Job_CatchChildren detects a job that has
250 * been migrated home, the job is placed on the stoppedJobs queue to be run
251 * when the next job finishes.
253 STATIC Lst stoppedJobs; /* Lst of Job structures describing
254 * jobs that were stopped due to concurrency
255 * limits or migration home */
258 #if defined(USE_PGRP) && defined(SYSV)
259 # define KILL(pid, sig) killpg(-(pid), (sig))
261 # if defined(USE_PGRP)
262 # define KILL(pid, sig) killpg((pid), (sig))
264 # define KILL(pid, sig) kill((pid), (sig))
269 * Grmpf... There is no way to set bits of the wait structure
270 * anymore with the stupid W*() macros. I liked the union wait
271 * stuff much more. So, we devise our own macros... This is
272 * really ugly, use dramamine sparingly. You have been warned.
274 #define W_SETMASKED(st, val, fun) \
277 int mask = fun(sh); \
279 for (sh = 0; ((mask >> sh) & 1) == 0; sh++) \
281 *(st) = (*(st) & ~mask) | ((val) << sh); \
284 #define W_SETTERMSIG(st, val) W_SETMASKED(st, val, WTERMSIG)
285 #define W_SETEXITSTATUS(st, val) W_SETMASKED(st, val, WEXITSTATUS)
288 static int JobCondPassSig(void *, void *);
289 static void JobPassSig(int);
290 static int JobCmpPid(void *, void *);
291 static int JobPrintCommand(void *, void *);
292 static int JobSaveCommand(void *, void *);
293 static void JobClose(Job *);
295 static int JobCmpRmtID(Job *, int);
296 # ifdef RMT_WILL_WATCH
297 static void JobLocalInput(int, Job *);
300 static void JobFinish(Job *, int *);
301 static void JobExec(Job *, char **);
303 static void JobMakeArgv(Job *, char **);
304 static void JobRestart(Job *);
305 static int JobStart(GNode *, int, Job *);
306 static char *JobOutput(Job *, char *, char *, int);
307 static void JobDoOutput(Job *, Boolean);
308 static Shell *JobMatchShell(char *);
309 static void JobInterrupt(int, int);
310 static void JobRestartJobs(void);
313 *-----------------------------------------------------------------------
315 * Pass a signal to a job if the job is remote or if USE_PGRP
322 * None, except the job may bite it.
324 *-----------------------------------------------------------------------
327 JobCondPassSig(void *jobp, void *signop)
329 Job *job = (Job *) jobp;
330 int signo = *(int *) signop;
331 #ifdef RMT_WANTS_SIGNALS
332 if (job->flags & JOB_REMOTE) {
333 (void) Rmt_Signal(job, signo);
335 KILL(job->pid, signo);
339 * Assume that sending the signal to job->pid will signal any remote
342 DEBUGF(JOB, ("JobCondPassSig passing signal %d to child %d.\n", signo, job->pid));
343 KILL(job->pid, signo);
349 *-----------------------------------------------------------------------
351 * Pass a signal on to all remote jobs and to all local jobs if
352 * USE_PGRP is defined, then die ourselves.
358 * We die by the same signal.
360 *-----------------------------------------------------------------------
363 JobPassSig(int signo)
365 sigset_t nmask, omask;
366 struct sigaction act;
368 DEBUGF(JOB, ("JobPassSig(%d) called.\n", signo));
369 Lst_ForEach(jobs, JobCondPassSig, (void *) &signo);
372 * Deal with proper cleanup based on the signal received. We only run
373 * the .INTERRUPT target if the signal was in fact an interrupt. The other
374 * three termination signals are more of a "get out *now*" command.
376 if (signo == SIGINT) {
377 JobInterrupt(TRUE, signo);
378 } else if ((signo == SIGHUP) || (signo == SIGTERM) || (signo == SIGQUIT)) {
379 JobInterrupt(FALSE, signo);
383 * Leave gracefully if SIGQUIT, rather than core dumping.
385 if (signo == SIGQUIT) {
390 * Send ourselves the signal now we've given the message to everyone else.
391 * Note we block everything else possible while we're getting the signal.
392 * This ensures that all our jobs get continued when we wake up before
393 * we take any other signal.
396 sigaddset(&nmask, signo);
397 sigprocmask(SIG_SETMASK, &nmask, &omask);
398 act.sa_handler = SIG_DFL;
399 sigemptyset(&act.sa_mask);
401 sigaction(signo, &act, NULL);
403 DEBUGF(JOB, ("JobPassSig passing signal to self, mask = %x.\n", ~0 & ~(1 << (signo-1))));
404 (void) signal(signo, SIG_DFL);
406 (void) KILL(getpid(), signo);
409 Lst_ForEach(jobs, JobCondPassSig, (void *) &signo);
411 (void) sigprocmask(SIG_SETMASK, &omask, NULL);
412 sigprocmask(SIG_SETMASK, &omask, NULL);
413 act.sa_handler = JobPassSig;
414 sigaction(signo, &act, NULL);
418 *-----------------------------------------------------------------------
420 * Compare the pid of the job with the given pid and return 0 if they
421 * are equal. This function is called from Job_CatchChildren via
422 * Lst_Find to find the job descriptor of the finished job.
425 * 0 if the pid's match
429 *-----------------------------------------------------------------------
432 JobCmpPid(void *job, void *pid)
434 return *(int *) pid - ((Job *) job)->pid;
439 *-----------------------------------------------------------------------
441 * Compare the rmtID of the job with the given rmtID and return 0 if they
445 * 0 if the rmtID's match
449 *-----------------------------------------------------------------------
452 JobCmpRmtID(void *job, void *rmtID)
454 return(*(int *) rmtID - *(int *) job->rmtID);
459 *-----------------------------------------------------------------------
461 * Put out another command for the given job. If the command starts
462 * with an @ or a - we process it specially. In the former case,
463 * so long as the -s and -n flags weren't given to make, we stick
464 * a shell-specific echoOff command in the script. In the latter,
465 * we ignore errors for the entire job, unless the shell has error
467 * If the command is just "..." we take all future commands for this
468 * job to be commands to be executed once the entire graph has been
469 * made and return non-zero to signal that the end of the commands
470 * was reached. These commands are later attached to the postCommands
471 * node and executed by Job_Finish when all things are done.
472 * This function is called from JobStart via Lst_ForEach.
475 * Always 0, unless the command was "..."
478 * If the command begins with a '-' and the shell has no error control,
479 * the JOB_IGNERR flag is set in the job descriptor.
480 * If the command is "..." and we're not ignoring such things,
481 * tailCmds is set to the successor node of the cmd.
482 * numCommands is incremented if the command is actually printed.
483 *-----------------------------------------------------------------------
486 JobPrintCommand(void *cmdp, void *jobp)
488 Boolean noSpecials; /* true if we shouldn't worry about
489 * inserting special commands into
490 * the input stream. */
491 Boolean shutUp = FALSE; /* true if we put a no echo command
492 * into the command file */
493 Boolean errOff = FALSE; /* true if we turned error checking
494 * off before printing the command
495 * and need to turn it back on */
496 char *cmdTemplate; /* Template to use when printing the
498 char *cmdStart; /* Start of expanded command */
499 LstNode cmdNode; /* Node for replacing the command */
500 char *cmd = (char *) cmdp;
501 Job *job = (Job *) jobp;
503 noSpecials = (noExecute && !(job->node->type & OP_MAKE));
505 if (strcmp(cmd, "...") == 0) {
506 job->node->type |= OP_SAVE_CMDS;
507 if ((job->flags & JOB_IGNDOTS) == 0) {
508 job->tailCmds = Lst_Succ(Lst_Member(job->node->commands,
515 #define DBPRINTF(fmt, arg) \
516 DEBUGF(JOB, (fmt, arg)); \
517 (void) fprintf(job->cmdFILE, fmt, arg); \
518 (void) fflush(job->cmdFILE);
523 * For debugging, we replace each command with the result of expanding
524 * the variables in the command.
526 cmdNode = Lst_Member(job->node->commands, (void *)cmd);
527 cmdStart = cmd = Var_Subst(NULL, cmd, job->node, FALSE);
528 Lst_Replace(cmdNode, (void *)cmdStart);
530 cmdTemplate = "%s\n";
533 * Check for leading @' and -'s to control echoing and error checking.
535 while (*cmd == '@' || *cmd == '-') {
537 shutUp = DEBUG(LOUD) ? FALSE : TRUE;
544 while (isspace((unsigned char) *cmd))
548 if (!(job->flags & JOB_SILENT) && !noSpecials &&
549 commandShell->hasEchoCtl) {
550 DBPRINTF("%s\n", commandShell->echoOff);
557 if ( !(job->flags & JOB_IGNERR) && !noSpecials) {
558 if (commandShell->hasErrCtl) {
560 * we don't want the error-control commands showing
561 * up either, so we turn off echoing while executing
562 * them. We could put another field in the shell
563 * structure to tell JobDoOutput to look for this
564 * string too, but why make it any more complex than
567 if (!(job->flags & JOB_SILENT) && !shutUp &&
568 commandShell->hasEchoCtl) {
569 DBPRINTF("%s\n", commandShell->echoOff);
570 DBPRINTF("%s\n", commandShell->ignErr);
571 DBPRINTF("%s\n", commandShell->echoOn);
573 DBPRINTF("%s\n", commandShell->ignErr);
575 } else if (commandShell->ignErr &&
576 (*commandShell->ignErr != '\0'))
579 * The shell has no error control, so we need to be
580 * weird to get it to ignore any errors from the command.
581 * If echoing is turned on, we turn it off and use the
582 * errCheck template to echo the command. Leave echoing
583 * off so the user doesn't see the weirdness we go through
584 * to ignore errors. Set cmdTemplate to use the weirdness
585 * instead of the simple "%s\n" template.
587 if (!(job->flags & JOB_SILENT) && !shutUp &&
588 commandShell->hasEchoCtl) {
589 DBPRINTF("%s\n", commandShell->echoOff);
590 DBPRINTF(commandShell->errCheck, cmd);
593 cmdTemplate = commandShell->ignErr;
595 * The error ignoration (hee hee) is already taken care
596 * of by the ignErr template, so pretend error checking
608 DBPRINTF(cmdTemplate, cmd);
612 * If echoing is already off, there's no point in issuing the
613 * echoOff command. Otherwise we issue it and pretend it was on
614 * for the whole command...
616 if (!shutUp && !(job->flags & JOB_SILENT) && commandShell->hasEchoCtl){
617 DBPRINTF("%s\n", commandShell->echoOff);
620 DBPRINTF("%s\n", commandShell->errCheck);
623 DBPRINTF("%s\n", commandShell->echoOn);
629 *-----------------------------------------------------------------------
631 * Save a command to be executed when everything else is done.
632 * Callback function for JobFinish...
638 * The command is tacked onto the end of postCommands's commands list.
640 *-----------------------------------------------------------------------
643 JobSaveCommand(void *cmd, void *gn)
645 cmd = (void *) Var_Subst(NULL, (char *) cmd, (GNode *) gn, FALSE);
646 (void) Lst_AtEnd(postCommands->commands, cmd);
652 *-----------------------------------------------------------------------
654 * Called to close both input and output pipes when a job is finished.
660 * The file descriptors associated with the job are closed.
662 *-----------------------------------------------------------------------
668 #ifdef RMT_WILL_WATCH
669 Rmt_Ignore(job->inPipe);
671 FD_CLR(job->inPipe, &outputs);
673 if (job->outPipe != job->inPipe) {
674 (void) close(job->outPipe);
676 JobDoOutput(job, TRUE);
677 (void) close(job->inPipe);
679 (void) close(job->outFd);
680 JobDoOutput(job, TRUE);
685 *-----------------------------------------------------------------------
687 * Do final processing for the given job including updating
688 * parents and starting new jobs as available/necessary. Note
689 * that we pay no attention to the JOB_IGNERR flag here.
690 * This is because when we're called because of a noexecute flag
691 * or something, jstat.w_status is 0 and when called from
692 * Job_CatchChildren, the status is zeroed if it s/b ignored.
698 * Some nodes may be put on the toBeMade queue.
699 * Final commands for the job are placed on postCommands.
701 * If we got an error and are aborting (aborting == ABORT_ERROR) and
702 * the job list is now empty, we are done for the day.
703 * If we recognized an error (errors !=0), we set the aborting flag
704 * to ABORT_ERROR so no more jobs will be started.
705 *-----------------------------------------------------------------------
709 JobFinish(Job *job, int *status)
713 if ((WIFEXITED(*status) &&
714 (((WEXITSTATUS(*status) != 0) && !(job->flags & JOB_IGNERR)))) ||
715 (WIFSIGNALED(*status) && (WTERMSIG(*status) != SIGCONT)))
718 * If it exited non-zero and either we're doing things our
719 * way or we're not ignoring errors, the job is finished.
720 * Similarly, if the shell died because of a signal
721 * the job is also finished. In these
722 * cases, finish out the job's output before printing the exit
726 KILL(job->pid, SIGCONT);
729 if (job->cmdFILE != NULL && job->cmdFILE != stdout) {
730 (void) fclose(job->cmdFILE);
734 if (job->flags & JOB_REMOTE)
735 Rmt_Done(job->rmtID, job->node);
737 } else if (WIFEXITED(*status)) {
739 * Deal with ignored errors in -B mode. We need to print a message
740 * telling of the ignored error as well as setting status.w_status
741 * to 0 so the next command gets run. To do this, we set done to be
742 * TRUE if in -B mode and the job exited non-zero.
744 done = WEXITSTATUS(*status) != 0;
746 * Old comment said: "Note we don't
747 * want to close down any of the streams until we know we're at the
749 * But we do. Otherwise when are we going to print the rest of the
754 if (job->flags & JOB_REMOTE)
755 Rmt_Done(job->rmtID, job->node);
759 * No need to close things down or anything.
765 WIFSTOPPED(*status) ||
766 (WIFSIGNALED(*status) && (WTERMSIG(*status) == SIGCONT)) ||
771 if (compatMake && !usePipes && (job->flags & JOB_IGNERR)) {
773 * If output is going to a file and this job is ignoring
774 * errors, arrange to have the exit status sent to the
775 * output file as well.
777 out = fdopen(job->outFd, "w");
779 Punt("Cannot fdopen");
784 if (WIFEXITED(*status)) {
785 DEBUGF(JOB, ("Process %d exited.\n", job->pid));
786 if (WEXITSTATUS(*status) != 0) {
787 if (usePipes && job->node != lastNode) {
788 MESSAGE(out, job->node);
789 lastNode = job->node;
791 (void) fprintf(out, "*** Error code %d%s\n",
792 WEXITSTATUS(*status),
793 (job->flags & JOB_IGNERR) ? "(ignored)" : "");
795 if (job->flags & JOB_IGNERR) {
798 } else if (DEBUG(JOB)) {
799 if (usePipes && job->node != lastNode) {
800 MESSAGE(out, job->node);
801 lastNode = job->node;
803 (void) fprintf(out, "*** Completed successfully\n");
805 } else if (WIFSTOPPED(*status)) {
806 DEBUGF(JOB, ("Process %d stopped.\n", job->pid));
807 if (usePipes && job->node != lastNode) {
808 MESSAGE(out, job->node);
809 lastNode = job->node;
811 if (!(job->flags & JOB_REMIGRATE)) {
812 (void) fprintf(out, "*** Stopped -- signal %d\n",
815 job->flags |= JOB_RESUME;
816 (void)Lst_AtEnd(stoppedJobs, (void *)job);
818 if (job->flags & JOB_REMIGRATE)
823 } else if (WTERMSIG(*status) == SIGCONT) {
825 * If the beastie has continued, shift the Job from the stopped
826 * list to the running one (or re-stop it if concurrency is
827 * exceeded) and go and get another child.
829 if (job->flags & (JOB_RESUME|JOB_REMIGRATE|JOB_RESTART)) {
830 if (usePipes && job->node != lastNode) {
831 MESSAGE(out, job->node);
832 lastNode = job->node;
834 (void) fprintf(out, "*** Continued\n");
836 if (!(job->flags & JOB_CONTINUING)) {
837 DEBUGF(JOB, ("Warning: process %d was not continuing.\n", job->pid));
840 * We don't really want to restart a job from scratch just
841 * because it continued, especially not without killing the
842 * continuing process! That's why this is ifdef'ed out.
848 job->flags &= ~JOB_CONTINUING;
849 Lst_AtEnd(jobs, (void *)job);
851 if (!(job->flags & JOB_REMOTE)) {
852 DEBUGF(JOB, ("Process %d is continuing locally.\n", job->pid));
855 if (nJobs == maxJobs) {
857 DEBUGF(JOB, ("Job queue is full.\n"));
862 if (usePipes && job->node != lastNode) {
863 MESSAGE(out, job->node);
864 lastNode = job->node;
866 (void) fprintf(out, "*** Signal %d\n", WTERMSIG(*status));
873 * Now handle the -B-mode stuff. If the beast still isn't finished,
874 * try and restart the job on the next command. If JobStart says it's
875 * ok, it's ok. If there's an error, this puppy is done.
877 if (compatMake && (WIFEXITED(*status) &&
878 !Lst_IsAtEnd(job->node->commands))) {
879 switch (JobStart(job->node, job->flags & JOB_IGNDOTS, job)) {
885 W_SETEXITSTATUS(status, 1);
889 * If we got back a JOB_FINISHED code, JobStart has already
890 * called Make_Update and freed the job descriptor. We set
891 * done to false here to avoid fake cycles and double frees.
892 * JobStart needs to do the update so we can proceed up the
893 * graph when given the -n flag..
904 (aborting != ABORT_ERROR) &&
905 (aborting != ABORT_INTERRUPT) &&
909 * As long as we aren't aborting and the job didn't return a non-zero
910 * status that we shouldn't ignore, we call Make_Update to update
911 * the parents. In addition, any saved commands for the node are placed
912 * on the .END target.
914 if (job->tailCmds != NULL) {
915 Lst_ForEachFrom(job->node->commands, job->tailCmds,
919 job->node->made = MADE;
920 Make_Update(job->node);
922 } else if (*status != 0) {
930 * Set aborting if any error.
932 if (errors && !keepgoing && (aborting != ABORT_INTERRUPT)) {
934 * If we found any errors in this batch of children and the -k flag
935 * wasn't given, we set the aborting flag so no more jobs get
938 aborting = ABORT_ERROR;
941 if ((aborting == ABORT_ERROR) && Job_Empty())
943 * If we are aborting and the job table is now empty, we finish.
949 *-----------------------------------------------------------------------
951 * Touch the given target. Called by JobStart when the -t flag was
952 * given. Prints messages unless told to be silent.
958 * The data modification of the file is changed. In addition, if the
959 * file did not exist, it is created.
960 *-----------------------------------------------------------------------
963 Job_Touch(GNode *gn, Boolean silent)
965 int streamID; /* ID of stream opened to do the touch */
966 struct utimbuf times; /* Times for utime() call */
968 if (gn->type & (OP_JOIN|OP_USE|OP_EXEC|OP_OPTIONAL)) {
970 * .JOIN, .USE, .ZEROTIME and .OPTIONAL targets are "virtual" targets
971 * and, as such, shouldn't really be created.
977 (void) fprintf(stdout, "touch %s\n", gn->name);
978 (void) fflush(stdout);
985 if (gn->type & OP_ARCHV) {
987 } else if (gn->type & OP_LIB) {
990 char *file = gn->path ? gn->path : gn->name;
992 times.actime = times.modtime = now;
993 if (utime(file, ×) < 0){
994 streamID = open(file, O_RDWR | O_CREAT, 0666);
1000 * Read and write a byte to the file to change the
1001 * modification time, then close the file.
1003 if (read(streamID, &c, 1) == 1) {
1004 (void) lseek(streamID, 0L, SEEK_SET);
1005 (void) write(streamID, &c, 1);
1008 (void) close(streamID);
1010 (void) fprintf(stdout, "*** couldn't touch %s: %s",
1011 file, strerror(errno));
1012 (void) fflush(stdout);
1019 *-----------------------------------------------------------------------
1020 * Job_CheckCommands --
1021 * Make sure the given node has all the commands it needs.
1024 * TRUE if the commands list is/was ok.
1027 * The node will have commands from the .DEFAULT rule added to it
1029 *-----------------------------------------------------------------------
1032 Job_CheckCommands(GNode *gn, void (*abortProc)(const char *, ...))
1034 if (OP_NOP(gn->type) && Lst_IsEmpty(gn->commands) &&
1035 (gn->type & OP_LIB) == 0) {
1037 * No commands. Look for .DEFAULT rule from which we might infer
1040 if ((DEFAULT != NULL) && !Lst_IsEmpty(DEFAULT->commands)) {
1043 * Make only looks for a .DEFAULT if the node was never the
1044 * target of an operator, so that's what we do too. If
1045 * a .DEFAULT was given, we substitute its commands for gn's
1046 * commands and set the IMPSRC variable to be the target's name
1047 * The DEFAULT node acts like a transformation rule, in that
1048 * gn also inherits any attributes or sources attached to
1051 Make_HandleUse(DEFAULT, gn);
1052 Var_Set(IMPSRC, Var_Value(TARGET, gn, &p1), gn);
1054 } else if (Dir_MTime(gn) == 0) {
1056 * The node wasn't the target of an operator we have no .DEFAULT
1057 * rule to go on and the target doesn't already exist. There's
1058 * nothing more we can do for this branch. If the -k flag wasn't
1059 * given, we stop in our tracks, otherwise we just don't update
1060 * this node's parents so they never get examined.
1062 static const char msg[] = "make: don't know how to make";
1064 if (gn->type & OP_OPTIONAL) {
1065 (void) fprintf(stdout, "%s %s(ignored)\n", msg, gn->name);
1066 (void) fflush(stdout);
1067 } else if (keepgoing) {
1068 (void) fprintf(stdout, "%s %s(continuing)\n", msg, gn->name);
1069 (void) fflush(stdout);
1073 if (strcmp(gn->name,"love") == 0)
1074 (*abortProc)("Not war.");
1077 (*abortProc)("%s %s. Stop", msg, gn->name);
1084 #ifdef RMT_WILL_WATCH
1086 *-----------------------------------------------------------------------
1088 * Handle a pipe becoming readable. Callback function for Rmt_Watch
1094 * JobDoOutput is called.
1096 *-----------------------------------------------------------------------
1100 JobLocalInput(int stream, Job *job)
1102 JobDoOutput(job, FALSE);
1104 #endif /* RMT_WILL_WATCH */
1107 *-----------------------------------------------------------------------
1109 * Execute the shell for the given job. Called from JobStart and
1116 * A shell is executed, outputs is altered and the Job structure added
1119 *-----------------------------------------------------------------------
1122 JobExec(Job *job, char **argv)
1124 int cpid; /* ID of new child */
1129 DEBUGF(JOB, ("Running %s %sly\n", job->node->name,
1130 job->flags&JOB_REMOTE?"remote":"local"));
1131 DEBUGF(JOB, ("\tCommand: "));
1132 for (i = 0; argv[i] != NULL; i++) {
1133 DEBUGF(JOB, ("%s ", argv[i]));
1135 DEBUGF(JOB, ("\n"));
1139 * Some jobs produce no output and it's disconcerting to have
1140 * no feedback of their running (since they produce no output, the
1141 * banner with their name in it never appears). This is an attempt to
1142 * provide that feedback, even if nothing follows it.
1144 if ((lastNode != job->node) && (job->flags & JOB_FIRST) &&
1145 !(job->flags & JOB_SILENT)) {
1146 MESSAGE(stdout, job->node);
1147 lastNode = job->node;
1151 if (job->flags & JOB_REMOTE) {
1154 #endif /* RMT_NO_EXEC */
1156 if ((cpid = vfork()) == -1) {
1157 Punt("Cannot fork");
1158 } else if (cpid == 0) {
1161 * Must duplicate the input stream down to the child's input and
1162 * reset it to the beginning (again). Since the stream was marked
1163 * close-on-exec, we must clear that bit in the new input.
1165 if (dup2(FILENO(job->cmdFILE), 0) == -1)
1166 Punt("Cannot dup2: %s", strerror(errno));
1167 (void) fcntl(0, F_SETFD, 0);
1168 (void) lseek(0, 0, SEEK_SET);
1172 * Set up the child's output to be routed through the pipe
1173 * we've created for it.
1175 if (dup2(job->outPipe, 1) == -1)
1176 Punt("Cannot dup2: %s", strerror(errno));
1179 * We're capturing output in a file, so we duplicate the
1180 * descriptor to the temporary file into the standard
1183 if (dup2(job->outFd, 1) == -1)
1184 Punt("Cannot dup2: %s", strerror(errno));
1187 * The output channels are marked close on exec. This bit was
1188 * duplicated by the dup2 (on some systems), so we have to clear
1189 * it before routing the shell's error output to the same place as
1190 * its standard output.
1192 (void) fcntl(1, F_SETFD, 0);
1193 if (dup2(1, 2) == -1)
1194 Punt("Cannot dup2: %s", strerror(errno));
1198 * We want to switch the child into a different process family so
1199 * we can kill it and all its descendants in one fell swoop,
1200 * by killing its process family, but not commit suicide.
1205 (void) setpgid(0, getpid());
1207 #endif /* USE_PGRP */
1210 if (job->flags & JOB_REMOTE) {
1211 Rmt_Exec(shellPath, argv, FALSE);
1214 (void) execv(shellPath, argv);
1216 (void) write(2, "Could not execute shell\n",
1217 sizeof("Could not execute shell"));
1221 long omask = sigblock(sigmask(SIGCHLD));
1225 if (usePipes && (job->flags & JOB_FIRST) ) {
1227 * The first time a job is run for a node, we set the current
1228 * position in the buffer to the beginning and mark another
1229 * stream to watch in the outputs mask
1233 #ifdef RMT_WILL_WATCH
1234 Rmt_Watch(job->inPipe, JobLocalInput, job);
1236 FD_SET(job->inPipe, &outputs);
1237 #endif /* RMT_WILL_WATCH */
1240 if (job->flags & JOB_REMOTE) {
1244 job->rmtID = Rmt_LastID(job->pid);
1249 * XXX: Used to not happen if REMOTE. Why?
1251 if (job->cmdFILE != NULL && job->cmdFILE != stdout) {
1252 (void) fclose(job->cmdFILE);
1253 job->cmdFILE = NULL;
1257 (void) sigsetmask(omask);
1265 * Now the job is actually running, add it to the table.
1268 (void) Lst_AtEnd(jobs, (void *)job);
1269 if (nJobs == maxJobs) {
1275 *-----------------------------------------------------------------------
1277 * Create the argv needed to execute the shell for a given job.
1284 *-----------------------------------------------------------------------
1287 JobMakeArgv(Job *job, char **argv)
1290 static char args[10]; /* For merged arguments */
1292 argv[0] = shellName;
1295 if ((commandShell->exit && (*commandShell->exit != '-')) ||
1296 (commandShell->echo && (*commandShell->echo != '-')))
1299 * At least one of the flags doesn't have a minus before it, so
1300 * merge them together. Have to do this because the *(&(@*#*&#$#
1301 * Bourne shell thinks its second argument is a file to source.
1302 * Grrrr. Note the ten-character limitation on the combined arguments.
1304 (void)sprintf(args, "-%s%s",
1305 ((job->flags & JOB_IGNERR) ? "" :
1306 (commandShell->exit ? commandShell->exit : "")),
1307 ((job->flags & JOB_SILENT) ? "" :
1308 (commandShell->echo ? commandShell->echo : "")));
1315 if (!(job->flags & JOB_IGNERR) && commandShell->exit) {
1316 argv[argc] = commandShell->exit;
1319 if (!(job->flags & JOB_SILENT) && commandShell->echo) {
1320 argv[argc] = commandShell->echo;
1328 *-----------------------------------------------------------------------
1330 * Restart a job that stopped for some reason.
1336 * jobFull will be set if the job couldn't be run.
1338 *-----------------------------------------------------------------------
1341 JobRestart(Job *job)
1347 if (job->flags & JOB_REMIGRATE) {
1350 verboseRemigrates ||
1353 (void) fprintf(stdout, "*** remigrating %x(%s)\n",
1354 job->pid, job->node->name);
1355 (void) fflush(stdout);
1359 if (!Rmt_ReExport(job->pid, job->node, &host)) {
1360 if (verboseRemigrates || DEBUG(JOB)) {
1361 (void) fprintf(stdout, "*** couldn't migrate...\n");
1362 (void) fflush(stdout);
1365 if (nLocal != maxLocal) {
1367 * Job cannot be remigrated, but there's room on the local
1368 * machine, so resume the job and note that another
1369 * local job has started.
1373 verboseRemigrates ||
1376 (void) fprintf(stdout, "*** resuming on local machine\n");
1377 (void) fflush(stdout);
1379 KILL(job->pid, SIGCONT);
1382 job->flags &= ~(JOB_REMIGRATE|JOB_RESUME|JOB_REMOTE);
1383 job->flags |= JOB_CONTINUING;
1385 job->flags &= ~(JOB_REMIGRATE|JOB_RESUME);
1389 * Job cannot be restarted. Mark the table as full and
1390 * place the job back on the list of stopped jobs.
1394 verboseRemigrates ||
1397 (void) fprintf(stdout, "*** holding\n");
1398 (void) fflush(stdout);
1400 (void)Lst_AtFront(stoppedJobs, (void *)job);
1402 DEBUGF(JOB, ("Job queue is full.\n"));
1408 * Clear out the remigrate and resume flags. Set the continuing
1409 * flag so we know later on that the process isn't exiting just
1410 * because of a signal.
1412 job->flags &= ~(JOB_REMIGRATE|JOB_RESUME);
1413 job->flags |= JOB_CONTINUING;
1418 (void)Lst_AtEnd(jobs, (void *)job);
1420 if (nJobs == maxJobs) {
1422 DEBUGF(JOB, ("Job queue is full.\n"));
1424 } else if (job->flags & JOB_RESTART) {
1426 * Set up the control arguments to the shell. This is based on the
1427 * flags set earlier for this job. If the JOB_IGNERR flag is clear,
1428 * the 'exit' flag of the commandShell is used to cause it to exit
1429 * upon receiving an error. If the JOB_SILENT flag is clear, the
1430 * 'echo' flag of the commandShell is used to get it to start echoing
1431 * as soon as it starts processing commands.
1435 JobMakeArgv(job, argv);
1437 DEBUGF(JOB, ("Restarting %s...", job->node->name));
1439 if ((job->node->type&OP_NOEXPORT) ||
1440 (nLocal < maxLocal && runLocalFirst)
1442 || !Rmt_Export(shellPath, argv, job)
1444 || !Rmt_Begin(shellPath, argv, job->node)
1448 if (((nLocal >= maxLocal) && !(job->flags & JOB_SPECIAL))) {
1450 * Can't be exported and not allowed to run locally -- put it
1451 * back on the hold queue and mark the table full
1453 DEBUGF(JOB, ("holding\n"));
1454 (void)Lst_AtFront(stoppedJobs, (void *)job);
1456 DEBUGF(JOB, ("Job queue is full.\n"));
1460 * Job may be run locally.
1462 DEBUGF(JOB, ("running locally\n"));
1463 job->flags &= ~JOB_REMOTE;
1469 * Can be exported. Hooray!
1471 DEBUGF(JOB, ("exporting\n"));
1472 job->flags |= JOB_REMOTE;
1478 * The job has stopped and needs to be restarted. Why it stopped,
1481 DEBUGF(JOB, ("Resuming %s...", job->node->name));
1482 if (((job->flags & JOB_REMOTE) ||
1483 (nLocal < maxLocal) ||
1485 (((job->flags & JOB_SPECIAL) &&
1486 (job->node->type & OP_NOEXPORT)) &&
1487 (maxLocal == 0))) &&
1489 ((job->flags & JOB_SPECIAL) &&
1490 (maxLocal == 0))) &&
1495 * If the job is remote, it's ok to resume it as long as the
1496 * maximum concurrency won't be exceeded. If it's local and
1497 * we haven't reached the local concurrency limit already (or the
1498 * job must be run locally and maxLocal is 0), it's also ok to
1504 #ifdef RMT_WANTS_SIGNALS
1505 if (job->flags & JOB_REMOTE) {
1506 error = !Rmt_Signal(job, SIGCONT);
1508 #endif /* RMT_WANTS_SIGNALS */
1509 error = (KILL(job->pid, SIGCONT) != 0);
1513 * Make sure the user knows we've continued the beast and
1514 * actually put the thing in the job table.
1516 job->flags |= JOB_CONTINUING;
1517 W_SETTERMSIG(&status, SIGCONT);
1518 JobFinish(job, &status);
1520 job->flags &= ~(JOB_RESUME|JOB_CONTINUING);
1521 DEBUGF(JOB, ("done\n"));
1523 Error("couldn't resume %s: %s",
1524 job->node->name, strerror(errno));
1526 W_SETEXITSTATUS(&status, 1);
1527 JobFinish(job, &status);
1531 * Job cannot be restarted. Mark the table as full and
1532 * place the job back on the list of stopped jobs.
1534 DEBUGF(JOB, ("table full\n"));
1535 (void) Lst_AtFront(stoppedJobs, (void *)job);
1537 DEBUGF(JOB, ("Job queue is full.\n"));
1543 *-----------------------------------------------------------------------
1545 * Start a target-creation process going for the target described
1546 * by the graph node gn.
1549 * JOB_ERROR if there was an error in the commands, JOB_FINISHED
1550 * if there isn't actually anything left to do for the job and
1551 * JOB_RUNNING if the job has been started.
1554 * A new Job node is created and added to the list of running
1555 * jobs. PMake is forked and a child shell created.
1556 *-----------------------------------------------------------------------
1559 JobStart(GNode *gn, int flags, Job *previous)
1561 Job *job; /* new job descriptor */
1562 char *argv[4]; /* Argument vector to shell */
1563 Boolean cmdsOK; /* true if the nodes commands were all right */
1564 Boolean local; /* Set true if the job was run locally */
1565 Boolean noExec; /* Set true if we decide not to run the job */
1566 int tfd; /* File descriptor for temp file */
1568 if (previous != NULL) {
1569 previous->flags &= ~(JOB_FIRST|JOB_IGNERR|JOB_SILENT|JOB_REMOTE);
1572 job = (Job *) emalloc(sizeof(Job));
1574 Punt("JobStart out of memory");
1580 job->tailCmds = NULL;
1583 * Set the initial value of the flags for this job based on the global
1584 * ones and the node's attributes... Any flags supplied by the caller
1585 * are also added to the field.
1588 if (Targ_Ignore(gn)) {
1589 job->flags |= JOB_IGNERR;
1591 if (Targ_Silent(gn)) {
1592 job->flags |= JOB_SILENT;
1594 job->flags |= flags;
1597 * Check the commands now so any attributes from .DEFAULT have a chance
1598 * to migrate to the node
1600 if (!compatMake && job->flags & JOB_FIRST) {
1601 cmdsOK = Job_CheckCommands(gn, Error);
1607 * If the -n flag wasn't given, we open up OUR (not the child's)
1608 * temporary file to stuff commands in it. The thing is rd/wr so we don't
1609 * need to reopen it to feed it to the shell. If the -n flag *was* given,
1610 * we just set the file to be stdout. Cute, huh?
1612 if ((gn->type & OP_MAKE) || (!noExecute && !touchFlag)) {
1614 * We're serious here, but if the commands were bogus, we're
1621 (void) strcpy(tfile, TMPPAT);
1622 if ((tfd = mkstemp(tfile)) == -1)
1623 Punt("Cannot create temp file: %s", strerror(errno));
1624 job->cmdFILE = fdopen(tfd, "w+");
1626 if (job->cmdFILE == NULL) {
1628 Punt("Could not open %s", tfile);
1630 (void) fcntl(FILENO(job->cmdFILE), F_SETFD, 1);
1632 * Send the commands to the command file, flush all its buffers then
1633 * rewind and remove the thing.
1638 * used to be backwards; replace when start doing multiple commands
1643 * Be compatible: If this is the first time for this node,
1644 * verify its commands are ok and open the commands list for
1645 * sequential access by later invocations of JobStart.
1646 * Once that is done, we take the next command off the list
1647 * and print it to the command file. If the command was an
1648 * ellipsis, note that there's nothing more to execute.
1650 if ((job->flags&JOB_FIRST) && (Lst_Open(gn->commands) != SUCCESS)){
1653 LstNode ln = Lst_Next(gn->commands);
1656 JobPrintCommand((void *) Lst_Datum(ln),
1660 Lst_Close(gn->commands);
1662 if (noExec && !(job->flags & JOB_FIRST)) {
1664 * If we're not going to execute anything, the job
1665 * is done and we need to close down the various
1666 * file descriptors we've opened for output, then
1667 * call JobDoOutput to catch the final characters or
1668 * send the file to the screen... Note that the i/o streams
1669 * are only open if this isn't the first job.
1670 * Note also that this could not be done in
1671 * Job_CatchChildren b/c it wasn't clear if there were
1672 * more commands to execute or not...
1679 * We can do all the commands at once. hooray for sanity
1682 Lst_ForEach(gn->commands, JobPrintCommand, (void *)job);
1685 * If we didn't print out any commands to the shell script,
1686 * there's not much point in executing the shell, is there?
1688 if (numCommands == 0) {
1692 } else if (noExecute) {
1694 * Not executing anything -- just print all the commands to stdout
1695 * in one fell swoop. This will still set up job->tailCmds correctly.
1697 if (lastNode != gn) {
1698 MESSAGE(stdout, gn);
1701 job->cmdFILE = stdout;
1703 * Only print the commands if they're ok, but don't die if they're
1704 * not -- just let the user know they're bad and keep going. It
1705 * doesn't do any harm in this case and may do some good.
1708 Lst_ForEach(gn->commands, JobPrintCommand, (void *)job);
1711 * Don't execute the shell, thank you.
1716 * Just touch the target and note that no shell should be executed.
1717 * Set cmdFILE to stdout to make life easier. Check the commands, too,
1718 * but don't die if they're no good -- it does no harm to keep working
1721 job->cmdFILE = stdout;
1722 Job_Touch(gn, job->flags&JOB_SILENT);
1727 * If we're not supposed to execute a shell, don't.
1731 * Unlink and close the command file if we opened one
1733 if (job->cmdFILE != stdout) {
1734 if (job->cmdFILE != NULL)
1735 (void) fclose(job->cmdFILE);
1737 (void) fflush(stdout);
1741 * We only want to work our way up the graph if we aren't here because
1742 * the commands for the job were no good.
1745 if (aborting == 0) {
1746 if (job->tailCmds != NULL) {
1747 Lst_ForEachFrom(job->node->commands, job->tailCmds,
1751 job->node->made = MADE;
1752 Make_Update(job->node);
1755 return(JOB_FINISHED);
1761 (void) fflush(job->cmdFILE);
1765 * Set up the control arguments to the shell. This is based on the flags
1766 * set earlier for this job.
1768 JobMakeArgv(job, argv);
1771 * If we're using pipes to catch output, create the pipe by which we'll
1772 * get the shell's output. If we're using files, print out that we're
1773 * starting a job and then set up its temporary-file name.
1775 if (!compatMake || (job->flags & JOB_FIRST)) {
1779 Punt("Cannot create pipe: %s", strerror(errno));
1780 job->inPipe = fd[0];
1781 job->outPipe = fd[1];
1782 (void) fcntl(job->inPipe, F_SETFD, 1);
1783 (void) fcntl(job->outPipe, F_SETFD, 1);
1785 (void) fprintf(stdout, "Remaking `%s'\n", gn->name);
1786 (void) fflush(stdout);
1787 (void) strcpy(job->outFile, TMPPAT);
1788 if ((job->outFd = mkstemp(job->outFile)) == -1)
1789 Punt("cannot create temp file: %s", strerror(errno));
1790 (void) fcntl(job->outFd, F_SETFD, 1);
1795 if (!(gn->type & OP_NOEXPORT) && !(runLocalFirst && nLocal < maxLocal)) {
1797 local = !Rmt_Export(shellPath, argv, job);
1799 local = !Rmt_Begin(shellPath, argv, job->node);
1800 #endif /* RMT_NO_EXEC */
1802 job->flags |= JOB_REMOTE;
1808 if (local && (((nLocal >= maxLocal) &&
1809 !(job->flags & JOB_SPECIAL) &&
1811 (!(gn->type & OP_NOEXPORT) || (maxLocal != 0))
1818 * The job can only be run locally, but we've hit the limit of
1819 * local concurrency, so put the job on hold until some other job
1820 * finishes. Note that the special jobs (.BEGIN, .INTERRUPT and .END)
1821 * may be run locally even when the local limit has been reached
1822 * (e.g. when maxLocal == 0), though they will be exported if at
1823 * all possible. In addition, any target marked with .NOEXPORT will
1824 * be run locally if maxLocal is 0.
1828 DEBUGF(JOB, ("Can only run job locally.\n"));
1829 job->flags |= JOB_RESTART;
1830 (void) Lst_AtEnd(stoppedJobs, (void *)job);
1832 if ((nLocal >= maxLocal) && local) {
1834 * If we're running this job locally as a special case (see above),
1835 * at least say the table is full.
1838 DEBUGF(JOB, ("Local job queue is full.\n"));
1842 return(JOB_RUNNING);
1846 JobOutput(Job *job, char *cp, char *endp, int msg)
1850 if (commandShell->noPrint) {
1851 ecp = strstr(cp, commandShell->noPrint);
1852 while (ecp != NULL) {
1855 if (msg && job->node != lastNode) {
1856 MESSAGE(stdout, job->node);
1857 lastNode = job->node;
1860 * The only way there wouldn't be a newline after
1861 * this line is if it were the last in the buffer.
1862 * however, since the non-printable comes after it,
1863 * there must be a newline, so we don't print one.
1865 (void) fprintf(stdout, "%s", cp);
1866 (void) fflush(stdout);
1868 cp = ecp + commandShell->noPLen;
1871 * Still more to print, look again after skipping
1872 * the whitespace following the non-printable
1876 while (*cp == ' ' || *cp == '\t' || *cp == '\n') {
1879 ecp = strstr(cp, commandShell->noPrint);
1889 *-----------------------------------------------------------------------
1891 * This function is called at different times depending on
1892 * whether the user has specified that output is to be collected
1893 * via pipes or temporary files. In the former case, we are called
1894 * whenever there is something to read on the pipe. We collect more
1895 * output from the given job and store it in the job's outBuf. If
1896 * this makes up a line, we print it tagged by the job's identifier,
1898 * If output has been collected in a temporary file, we open the
1899 * file and read it line by line, transfering it to our own
1900 * output channel until the file is empty. At which point we
1901 * remove the temporary file.
1902 * In both cases, however, we keep our figurative eye out for the
1903 * 'noPrint' line for the shell from which the output came. If
1904 * we recognize a line, we don't print it. If the command is not
1905 * alone on the line (the character after it is not \0 or \n), we
1906 * do print whatever follows it.
1912 * curPos may be shifted as may the contents of outBuf.
1913 *-----------------------------------------------------------------------
1916 JobDoOutput(Job *job, Boolean finish)
1918 Boolean gotNL = FALSE; /* true if got a newline */
1919 Boolean fbuf; /* true if our buffer filled up */
1920 int nr; /* number of bytes read */
1921 int i; /* auxiliary index into outBuf */
1922 int max; /* limit for i (end of current data) */
1923 int nRead; /* (Temporary) number of bytes read */
1925 FILE *oFILE; /* Stream pointer to shell's output file */
1931 * Read as many bytes as will fit in the buffer.
1937 nRead = read(job->inPipe, &job->outBuf[job->curPos],
1938 JOB_BUFSIZE - job->curPos);
1940 DEBUGF(JOB, ("JobDoOutput(piperead)"));
1947 * If we hit the end-of-file (the job is dead), we must flush its
1948 * remaining output, so pretend we read a newline if there's any
1949 * output remaining in the buffer.
1950 * Also clear the 'finish' flag so we stop looping.
1952 if ((nr == 0) && (job->curPos != 0)) {
1953 job->outBuf[job->curPos] = '\n';
1956 } else if (nr == 0) {
1961 * Look for the last newline in the bytes we just got. If there is
1962 * one, break out of the loop with 'i' as its index and gotNL set
1965 max = job->curPos + nr;
1966 for (i = job->curPos + nr - 1; i >= job->curPos; i--) {
1967 if (job->outBuf[i] == '\n') {
1970 } else if (job->outBuf[i] == '\0') {
1974 job->outBuf[i] = ' ';
1980 if (job->curPos == JOB_BUFSIZE) {
1982 * If we've run out of buffer space, we have no choice
1983 * but to print the stuff. sigh.
1989 if (gotNL || fbuf) {
1991 * Need to send the output to the screen. Null terminate it
1992 * first, overwriting the newline character if there was one.
1993 * So long as the line isn't one we should filter (according
1994 * to the shell description), we print the line, preceded
1995 * by a target banner if this target isn't the same as the
1996 * one for which we last printed something.
1997 * The rest of the data in the buffer are then shifted down
1998 * to the start of the buffer and curPos is set accordingly.
2000 job->outBuf[i] = '\0';
2001 if (i >= job->curPos) {
2004 cp = JobOutput(job, job->outBuf, &job->outBuf[i], FALSE);
2007 * There's still more in that thar buffer. This time, though,
2008 * we know there's no newline at the end, so we add one of
2009 * our own free will.
2012 if (job->node != lastNode) {
2013 MESSAGE(stdout, job->node);
2014 lastNode = job->node;
2016 (void) fprintf(stdout, "%s%s", cp, gotNL ? "\n" : "");
2017 (void) fflush(stdout);
2021 /* shift the remaining characters down */
2022 (void) memcpy(job->outBuf, &job->outBuf[i + 1], max - (i + 1));
2023 job->curPos = max - (i + 1);
2027 * We have written everything out, so we just start over
2028 * from the start of the buffer. No copying. No nothing.
2035 * If the finish flag is true, we must loop until we hit
2036 * end-of-file on the pipe. This is guaranteed to happen
2037 * eventually since the other end of the pipe is now closed
2038 * (we closed it explicitly and the child has exited). When
2039 * we do get an EOF, finish will be set FALSE and we'll fall
2046 * We've been called to retrieve the output of the job from the
2047 * temporary file where it's been squirreled away. This consists of
2048 * opening the file, reading the output line by line, being sure not
2049 * to print the noPrint line for the shell we used, then close and
2050 * remove the temporary file. Very simple.
2052 * Change to read in blocks and do FindSubString type things as for
2053 * pipes? That would allow for "@echo -n..."
2055 oFILE = fopen(job->outFile, "r");
2056 if (oFILE != NULL) {
2057 (void) fprintf(stdout, "Results of making %s:\n", job->node->name);
2058 (void) fflush(stdout);
2059 while (fgets(inLine, sizeof(inLine), oFILE) != NULL) {
2060 char *cp, *endp, *oendp;
2063 oendp = endp = inLine + strlen(inLine);
2064 if (endp[-1] == '\n') {
2067 cp = JobOutput(job, inLine, endp, FALSE);
2070 * There's still more in that thar buffer. This time, though,
2071 * we know there's no newline at the end, so we add one of
2072 * our own free will.
2074 (void) fprintf(stdout, "%s", cp);
2075 (void) fflush(stdout);
2076 if (endp != oendp) {
2077 (void) fprintf(stdout, "\n");
2078 (void) fflush(stdout);
2081 (void) fclose(oFILE);
2082 (void) eunlink(job->outFile);
2088 *-----------------------------------------------------------------------
2089 * Job_CatchChildren --
2090 * Handle the exit of a child. Called from Make_Make.
2096 * The job descriptor is removed from the list of children.
2099 * We do waits, blocking or not, according to the wisdom of our
2100 * caller, until there are no more children to report. For each
2101 * job, call JobFinish to finish things off. This will take care of
2102 * putting jobs on the stoppedJobs queue.
2104 *-----------------------------------------------------------------------
2107 Job_CatchChildren(Boolean block)
2109 int pid; /* pid of dead child */
2110 Job *job; /* job descriptor for dead child */
2111 LstNode jnode; /* list element for finding job */
2112 int status; /* Exit/termination status */
2115 * Don't even bother if we know there's no one around.
2121 while ((pid = waitpid((pid_t) -1, &status,
2122 (block?0:WNOHANG)|WUNTRACED)) > 0)
2124 DEBUGF(JOB, ("Process %d exited or stopped.\n", pid));
2126 jnode = Lst_Find(jobs, (void *)&pid, JobCmpPid);
2128 if (jnode == NULL) {
2129 if (WIFSIGNALED(status) && (WTERMSIG(status) == SIGCONT)) {
2130 jnode = Lst_Find(stoppedJobs, (void *) &pid, JobCmpPid);
2131 if (jnode == NULL) {
2132 Error("Resumed child (%d) not in table", pid);
2135 job = (Job *)Lst_Datum(jnode);
2136 (void) Lst_Remove(stoppedJobs, jnode);
2138 Error("Child (%d) not in table?", pid);
2142 job = (Job *) Lst_Datum(jnode);
2143 (void) Lst_Remove(jobs, jnode);
2145 DEBUGF(JOB, ("Job queue is no longer full.\n"));
2148 if (!(job->flags & JOB_REMOTE)) {
2149 DEBUGF(JOB, ("Job queue has one fewer local process.\n"));
2157 JobFinish(job, &status);
2162 *-----------------------------------------------------------------------
2163 * Job_CatchOutput --
2164 * Catch the output from our children, if we're using
2165 * pipes do so. Otherwise just block time until we get a
2166 * signal (most likely a SIGCHLD) since there's no point in
2167 * just spinning when there's nothing to do and the reaping
2168 * of a child can wait for a while.
2174 * Output is read from pipes if we're piping.
2175 * -----------------------------------------------------------------------
2178 Job_CatchOutput(void)
2181 struct timeval timeout;
2185 #ifdef RMT_WILL_WATCH
2186 int pnJobs; /* Previous nJobs */
2189 (void) fflush(stdout);
2190 #ifdef RMT_WILL_WATCH
2194 * It is possible for us to be called with nJobs equal to 0. This happens
2195 * if all the jobs finish and a job that is stopped cannot be run
2196 * locally (eg if maxLocal is 0) and cannot be exported. The job will
2197 * be placed back on the stoppedJobs queue, Job_Empty() will return false,
2198 * Make_Run will call us again when there's nothing for which to wait.
2199 * nJobs never changes, so we loop forever. Hence the check. It could
2200 * be argued that we should sleep for a bit so as not to swamp the
2201 * exportation system with requests. Perhaps we should.
2203 * NOTE: IT IS THE RESPONSIBILITY OF Rmt_Wait TO CALL Job_CatchChildren
2204 * IN A TIMELY FASHION TO CATCH ANY LOCALLY RUNNING JOBS THAT EXIT.
2205 * It may use the variable nLocal to determine if it needs to call
2206 * Job_CatchChildren (if nLocal is 0, there's nothing for which to
2209 while (nJobs != 0 && pnJobs == nJobs) {
2215 timeout.tv_sec = SEL_SEC;
2216 timeout.tv_usec = SEL_USEC;
2218 if ((nfds = select(FD_SETSIZE, &readfds, (fd_set *) 0,
2219 (fd_set *) 0, &timeout)) <= 0)
2222 if (Lst_Open(jobs) == FAILURE) {
2223 Punt("Cannot open job table");
2225 while (nfds && (ln = Lst_Next(jobs)) != NULL) {
2226 job = (Job *) Lst_Datum(ln);
2227 if (FD_ISSET(job->inPipe, &readfds)) {
2228 JobDoOutput(job, FALSE);
2235 #endif /* RMT_WILL_WATCH */
2239 *-----------------------------------------------------------------------
2241 * Start the creation of a target. Basically a front-end for
2242 * JobStart used by the Make module.
2248 * Another job is started.
2250 *-----------------------------------------------------------------------
2255 (void) JobStart(gn, 0, NULL);
2259 *-----------------------------------------------------------------------
2261 * Initialize the process module, given a maximum number of jobs, and
2262 * a maximum number of local jobs.
2268 * lists and counters are initialized
2269 *-----------------------------------------------------------------------
2272 Job_Init(int maxproc, int maxlocal)
2274 GNode *begin; /* node for commands to do at the very start */
2276 jobs = Lst_Init(FALSE);
2277 stoppedJobs = Lst_Init(FALSE);
2279 maxLocal = maxlocal;
2289 if (maxJobs == 1 || beVerbose == 0
2295 * If only one job can run at a time, there's no need for a banner,
2303 if (shellPath == NULL) {
2305 * The user didn't specify a shell to use, so we are using the
2306 * default one... Both the absolute path and the last component
2307 * must be set. The last component is taken from the 'name' field
2308 * of the default shell description pointed-to by commandShell.
2309 * All default shells are located in _PATH_DEFSHELLDIR.
2311 shellName = commandShell->name;
2312 shellPath = str_concat(_PATH_DEFSHELLDIR, shellName, STR_ADDSLASH);
2315 if (commandShell->exit == NULL) {
2316 commandShell->exit = "";
2318 if (commandShell->echo == NULL) {
2319 commandShell->echo = "";
2323 * Catch the four signals that POSIX specifies if they aren't ignored.
2324 * JobPassSig will take care of calling JobInterrupt if appropriate.
2326 if (signal(SIGINT, SIG_IGN) != SIG_IGN) {
2327 (void) signal(SIGINT, JobPassSig);
2329 if (signal(SIGHUP, SIG_IGN) != SIG_IGN) {
2330 (void) signal(SIGHUP, JobPassSig);
2332 if (signal(SIGQUIT, SIG_IGN) != SIG_IGN) {
2333 (void) signal(SIGQUIT, JobPassSig);
2335 if (signal(SIGTERM, SIG_IGN) != SIG_IGN) {
2336 (void) signal(SIGTERM, JobPassSig);
2339 * There are additional signals that need to be caught and passed if
2340 * either the export system wants to be told directly of signals or if
2341 * we're giving each job its own process group (since then it won't get
2342 * signals from the terminal driver as we own the terminal)
2344 #if defined(RMT_WANTS_SIGNALS) || defined(USE_PGRP)
2345 if (signal(SIGTSTP, SIG_IGN) != SIG_IGN) {
2346 (void) signal(SIGTSTP, JobPassSig);
2348 if (signal(SIGTTOU, SIG_IGN) != SIG_IGN) {
2349 (void) signal(SIGTTOU, JobPassSig);
2351 if (signal(SIGTTIN, SIG_IGN) != SIG_IGN) {
2352 (void) signal(SIGTTIN, JobPassSig);
2354 if (signal(SIGWINCH, SIG_IGN) != SIG_IGN) {
2355 (void) signal(SIGWINCH, JobPassSig);
2359 begin = Targ_FindNode(".BEGIN", TARG_NOCREATE);
2361 if (begin != NULL) {
2362 JobStart(begin, JOB_SPECIAL, (Job *)0);
2365 #ifndef RMT_WILL_WATCH
2366 Job_CatchChildren(!usePipes);
2367 #endif /* RMT_WILL_WATCH */
2370 postCommands = Targ_FindNode(".END", TARG_CREATE);
2374 *-----------------------------------------------------------------------
2376 * See if the job table is full. It is considered full if it is OR
2377 * if we are in the process of aborting OR if we have
2378 * reached/exceeded our local quota. This prevents any more jobs
2382 * TRUE if the job table is full, FALSE otherwise
2385 *-----------------------------------------------------------------------
2390 return(aborting || jobFull);
2394 *-----------------------------------------------------------------------
2396 * See if the job table is empty. Because the local concurrency may
2397 * be set to 0, it is possible for the job table to become empty,
2398 * while the list of stoppedJobs remains non-empty. In such a case,
2399 * we want to restart as many jobs as we can.
2402 * TRUE if it is. FALSE if it ain't.
2407 * -----------------------------------------------------------------------
2413 if (!Lst_IsEmpty(stoppedJobs) && !aborting) {
2415 * The job table is obviously not full if it has no jobs in
2416 * it...Try and restart the stopped jobs.
2430 *-----------------------------------------------------------------------
2432 * Find a matching shell in 'shells' given its final component.
2435 * A pointer to the Shell structure.
2440 *-----------------------------------------------------------------------
2443 JobMatchShell(char *name)
2445 Shell *sh; /* Pointer into shells table */
2446 Shell *match; /* Longest-matching shell */
2451 eoname = name + strlen(name);
2455 for (sh = shells; sh->name != NULL; sh++) {
2456 for (cp1 = eoname - strlen(sh->name), cp2 = sh->name;
2457 *cp1 != '\0' && *cp1 == *cp2;
2463 } else if (match == NULL || strlen(match->name) < strlen(sh->name)) {
2467 return(match == NULL ? sh : match);
2471 *-----------------------------------------------------------------------
2473 * Parse a shell specification and set up commandShell, shellPath
2474 * and shellName appropriately.
2477 * FAILURE if the specification was incorrect.
2480 * commandShell points to a Shell structure (either predefined or
2481 * created from the shell spec), shellPath is the full path of the
2482 * shell described by commandShell, while shellName is just the
2483 * final component of shellPath.
2486 * A shell specification consists of a .SHELL target, with dependency
2487 * operator, followed by a series of blank-separated words. Double
2488 * quotes can be used to use blanks in words. A backslash escapes
2489 * anything (most notably a double-quote and a space) and
2490 * provides the functionality it does in C. Each word consists of
2491 * keyword and value separated by an equal sign. There should be no
2492 * unnecessary spaces in the word. The keywords are as follows:
2493 * name Name of shell.
2494 * path Location of shell. Overrides "name" if given
2495 * quiet Command to turn off echoing.
2496 * echo Command to turn echoing on
2497 * filter Result of turning off echoing that shouldn't be
2499 * echoFlag Flag to turn echoing on at the start
2500 * errFlag Flag to turn error checking on at the start
2501 * hasErrCtl True if shell has error checking control
2502 * check Command to turn on error checking if hasErrCtl
2503 * is TRUE or template of command to echo a command
2504 * for which error checking is off if hasErrCtl is
2506 * ignore Command to turn off error checking if hasErrCtl
2507 * is TRUE or template of command to execute a
2508 * command so as to ignore any errors it returns if
2509 * hasErrCtl is FALSE.
2511 *-----------------------------------------------------------------------
2514 Job_ParseShell(char *line)
2522 Boolean fullSpec = FALSE;
2524 while (isspace((unsigned char) *line)) {
2527 words = brk_string(line, &wordCount, TRUE);
2529 memset(&newShell, 0, sizeof(newShell));
2532 * Parse the specification by keyword
2534 for (path = NULL, argc = wordCount - 1, argv = words + 1;
2537 if (strncmp(*argv, "path=", 5) == 0) {
2539 } else if (strncmp(*argv, "name=", 5) == 0) {
2540 newShell.name = &argv[0][5];
2542 if (strncmp(*argv, "quiet=", 6) == 0) {
2543 newShell.echoOff = &argv[0][6];
2544 } else if (strncmp(*argv, "echo=", 5) == 0) {
2545 newShell.echoOn = &argv[0][5];
2546 } else if (strncmp(*argv, "filter=", 7) == 0) {
2547 newShell.noPrint = &argv[0][7];
2548 newShell.noPLen = strlen(newShell.noPrint);
2549 } else if (strncmp(*argv, "echoFlag=", 9) == 0) {
2550 newShell.echo = &argv[0][9];
2551 } else if (strncmp(*argv, "errFlag=", 8) == 0) {
2552 newShell.exit = &argv[0][8];
2553 } else if (strncmp(*argv, "hasErrCtl=", 10) == 0) {
2554 char c = argv[0][10];
2555 newShell.hasErrCtl = !((c != 'Y') && (c != 'y') &&
2556 (c != 'T') && (c != 't'));
2557 } else if (strncmp(*argv, "check=", 6) == 0) {
2558 newShell.errCheck = &argv[0][6];
2559 } else if (strncmp(*argv, "ignore=", 7) == 0) {
2560 newShell.ignErr = &argv[0][7];
2562 Parse_Error(PARSE_FATAL, "Unknown keyword \"%s\"",
2572 * If no path was given, the user wants one of the pre-defined shells,
2573 * yes? So we find the one s/he wants with the help of JobMatchShell
2574 * and set things up the right way. shellPath will be set up by
2577 if (newShell.name == NULL) {
2578 Parse_Error(PARSE_FATAL, "Neither path nor name specified");
2581 commandShell = JobMatchShell(newShell.name);
2582 shellName = newShell.name;
2586 * The user provided a path. If s/he gave nothing else (fullSpec is
2587 * FALSE), try and find a matching shell in the ones we know of.
2588 * Else we just take the specification at its word and copy it
2589 * to a new location. In either case, we need to record the
2590 * path the user gave for the shell.
2593 path = strrchr(path, '/');
2599 if (newShell.name != NULL) {
2600 shellName = newShell.name;
2605 commandShell = JobMatchShell(shellName);
2607 commandShell = (Shell *) emalloc(sizeof(Shell));
2608 *commandShell = newShell;
2612 if (commandShell->echoOn && commandShell->echoOff) {
2613 commandShell->hasEchoCtl = TRUE;
2616 if (!commandShell->hasErrCtl) {
2617 if (commandShell->errCheck == NULL) {
2618 commandShell->errCheck = "";
2620 if (commandShell->ignErr == NULL) {
2621 commandShell->ignErr = "%s\n";
2628 *-----------------------------------------------------------------------
2630 * Handle the receipt of an interrupt.
2636 * All children are killed. Another job will be started if the
2637 * .INTERRUPT target was given.
2638 *-----------------------------------------------------------------------
2641 JobInterrupt(int runINTERRUPT, int signo)
2643 LstNode ln; /* element in job table */
2644 Job *job = NULL; /* job descriptor in that element */
2645 GNode *interrupt; /* the node describing the .INTERRUPT target */
2647 aborting = ABORT_INTERRUPT;
2649 (void) Lst_Open(jobs);
2650 while ((ln = Lst_Next(jobs)) != NULL) {
2651 job = (Job *) Lst_Datum(ln);
2653 if (!Targ_Precious(job->node)) {
2654 char *file = (job->node->path == NULL ?
2657 if (!noExecute && eunlink(file) != -1) {
2658 Error("*** %s removed", file);
2661 #ifdef RMT_WANTS_SIGNALS
2662 if (job->flags & JOB_REMOTE) {
2664 * If job is remote, let the Rmt module do the killing.
2666 if (!Rmt_Signal(job, signo)) {
2668 * If couldn't kill the thing, finish it out now with an
2669 * error code, since no exit report will come in likely.
2673 status.w_status = 0;
2674 status.w_retcode = 1;
2675 JobFinish(job, &status);
2677 } else if (job->pid) {
2678 KILL(job->pid, signo);
2682 DEBUGF(JOB, ("JobInterrupt passing signal to child %d.\n",
2684 KILL(job->pid, signo);
2686 #endif /* RMT_WANTS_SIGNALS */
2690 (void)Lst_Open(stoppedJobs);
2691 while ((ln = Lst_Next(stoppedJobs)) != NULL) {
2692 job = (Job *) Lst_Datum(ln);
2694 if (job->flags & JOB_RESTART) {
2695 DEBUGF(JOB, "JobInterrupt skipping job on stopped queue"
2696 "-- it was waiting to be restarted.\n");
2699 if (!Targ_Precious(job->node)) {
2700 char *file = (job->node->path == NULL ?
2703 if (eunlink(file) == 0) {
2704 Error("*** %s removed", file);
2708 * Resume the thing so it will take the signal.
2710 DEBUGF(JOB, ("JobInterrupt passing CONT to stopped child %d.\n", job->pid));
2711 KILL(job->pid, SIGCONT);
2712 #ifdef RMT_WANTS_SIGNALS
2713 if (job->flags & JOB_REMOTE) {
2715 * If job is remote, let the Rmt module do the killing.
2717 if (!Rmt_Signal(job, SIGINT)) {
2719 * If couldn't kill the thing, finish it out now with an
2720 * error code, since no exit report will come in likely.
2723 status.w_status = 0;
2724 status.w_retcode = 1;
2725 JobFinish(job, &status);
2727 } else if (job->pid) {
2728 DEBUGF(JOB, "JobInterrupt passing interrupt to stopped child %d.\n",
2730 KILL(job->pid, SIGINT);
2732 #endif /* RMT_WANTS_SIGNALS */
2735 Lst_Close(stoppedJobs);
2737 if (runINTERRUPT && !touchFlag) {
2738 interrupt = Targ_FindNode(".INTERRUPT", TARG_NOCREATE);
2739 if (interrupt != NULL) {
2740 ignoreErrors = FALSE;
2742 JobStart(interrupt, JOB_IGNDOTS, (Job *)0);
2745 #ifndef RMT_WILL_WATCH
2746 Job_CatchChildren(!usePipes);
2747 #endif /* RMT_WILL_WATCH */
2754 *-----------------------------------------------------------------------
2756 * Do final processing such as the running of the commands
2757 * attached to the .END target.
2760 * Number of errors reported.
2761 *-----------------------------------------------------------------------
2766 if (postCommands != NULL && !Lst_IsEmpty(postCommands->commands)) {
2768 Error("Errors reported so .END ignored");
2770 JobStart(postCommands, JOB_SPECIAL | JOB_IGNDOTS, NULL);
2774 #ifndef RMT_WILL_WATCH
2775 Job_CatchChildren(!usePipes);
2776 #endif /* RMT_WILL_WATCH */
2784 *-----------------------------------------------------------------------
2786 * Waits for all running jobs to finish and returns. Sets 'aborting'
2787 * to ABORT_WAIT to prevent other jobs from starting.
2793 * Currently running jobs finish.
2795 *-----------------------------------------------------------------------
2800 aborting = ABORT_WAIT;
2801 while (nJobs != 0) {
2803 #ifndef RMT_WILL_WATCH
2804 Job_CatchChildren(!usePipes);
2805 #endif /* RMT_WILL_WATCH */
2811 *-----------------------------------------------------------------------
2813 * Abort all currently running jobs without handling output or anything.
2814 * This function is to be called only in the event of a major
2815 * error. Most definitely NOT to be called from JobInterrupt.
2821 * All children are killed, not just the firstborn
2822 *-----------------------------------------------------------------------
2827 LstNode ln; /* element in job table */
2828 Job *job; /* the job descriptor in that element */
2831 aborting = ABORT_ERROR;
2835 (void) Lst_Open(jobs);
2836 while ((ln = Lst_Next(jobs)) != NULL) {
2837 job = (Job *) Lst_Datum(ln);
2840 * kill the child process with increasingly drastic signals to make
2841 * darn sure it's dead.
2843 #ifdef RMT_WANTS_SIGNALS
2844 if (job->flags & JOB_REMOTE) {
2845 Rmt_Signal(job, SIGINT);
2846 Rmt_Signal(job, SIGKILL);
2848 KILL(job->pid, SIGINT);
2849 KILL(job->pid, SIGKILL);
2852 KILL(job->pid, SIGINT);
2853 KILL(job->pid, SIGKILL);
2854 #endif /* RMT_WANTS_SIGNALS */
2859 * Catch as many children as want to report in at first, then give up
2861 while (waitpid((pid_t) -1, &foo, WNOHANG) > 0)
2867 *-----------------------------------------------------------------------
2868 * JobFlagForMigration --
2869 * Handle the eviction of a child. Called from RmtStatusChange.
2870 * Flags the child as remigratable and then suspends it. Takes
2871 * the ID of the host we used, for matching children.
2877 * The job descriptor is flagged for remigration.
2879 *-----------------------------------------------------------------------
2882 JobFlagForMigration(int hostID)
2884 Job *job; /* job descriptor for dead child */
2885 LstNode jnode; /* list element for finding job */
2887 DEBUGF(JOB, ("JobFlagForMigration(%d) called.\n", hostID));
2888 jnode = Lst_Find(jobs, (void *)hostID, JobCmpRmtID);
2890 if (jnode == NULL) {
2891 jnode = Lst_Find(stoppedJobs, (void *)hostID, JobCmpRmtID);
2892 if (jnode == NULL) {
2894 Error("Evicting host(%d) not in table", hostID);
2899 job = (Job *) Lst_Datum(jnode);
2901 DEBUGF(JOB, ("JobFlagForMigration(%d) found job '%s'.\n", hostID, job->node->name));
2903 KILL(job->pid, SIGSTOP);
2905 job->flags |= JOB_REMIGRATE;
2911 *-----------------------------------------------------------------------
2913 * Tries to restart stopped jobs if there are slots available.
2914 * Note that this tries to restart them regardless of pending errors.
2915 * It's not good to leave stopped jobs lying around!
2921 * Resumes(and possibly migrates) jobs.
2923 *-----------------------------------------------------------------------
2926 JobRestartJobs(void)
2928 while (!jobFull && !Lst_IsEmpty(stoppedJobs)) {
2929 DEBUGF(JOB, ("Job queue is not full. Restarting a stopped job.\n"));
2930 JobRestart((Job *)Lst_DeQueue(stoppedJobs));