job.c revision 1.12
1 /* $NetBSD: job.c,v 1.12 1995/11/02 23:54:48 christos Exp $ */ 2 3 /* 4 * Copyright (c) 1988, 1989, 1990 The Regents of the University of California. 5 * Copyright (c) 1988, 1989 by Adam de Boor 6 * Copyright (c) 1989 by Berkeley Softworks 7 * All rights reserved. 8 * 9 * This code is derived from software contributed to Berkeley by 10 * Adam de Boor. 11 * 12 * Redistribution and use in source and binary forms, with or without 13 * modification, are permitted provided that the following conditions 14 * are met: 15 * 1. Redistributions of source code must retain the above copyright 16 * notice, this list of conditions and the following disclaimer. 17 * 2. Redistributions in binary form must reproduce the above copyright 18 * notice, this list of conditions and the following disclaimer in the 19 * documentation and/or other materials provided with the distribution. 20 * 3. All advertising materials mentioning features or use of this software 21 * must display the following acknowledgement: 22 * This product includes software developed by the University of 23 * California, Berkeley and its contributors. 24 * 4. Neither the name of the University nor the names of its contributors 25 * may be used to endorse or promote products derived from this software 26 * without specific prior written permission. 27 * 28 * THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND 29 * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE 30 * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE 31 * ARE DISCLAIMED. IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE 32 * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL 33 * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS 34 * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION) 35 * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT 36 * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY 37 * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF 38 * SUCH DAMAGE. 39 */ 40 41 #ifndef lint 42 #if 0 43 static char sccsid[] = "@(#)job.c 5.15 (Berkeley) 3/1/91"; 44 #else 45 static char rcsid[] = "$NetBSD: job.c,v 1.12 1995/11/02 23:54:48 christos Exp $"; 46 #endif 47 #endif /* not lint */ 48 49 /*- 50 * job.c -- 51 * handle the creation etc. of our child processes. 52 * 53 * Interface: 54 * Job_Make Start the creation of the given target. 55 * 56 * Job_CatchChildren Check for and handle the termination of any 57 * children. This must be called reasonably 58 * frequently to keep the whole make going at 59 * a decent clip, since job table entries aren't 60 * removed until their process is caught this way. 61 * Its single argument is TRUE if the function 62 * should block waiting for a child to terminate. 63 * 64 * Job_CatchOutput Print any output our children have produced. 65 * Should also be called fairly frequently to 66 * keep the user informed of what's going on. 67 * If no output is waiting, it will block for 68 * a time given by the SEL_* constants, below, 69 * or until output is ready. 70 * 71 * Job_Init Called to intialize this module. in addition, 72 * any commands attached to the .BEGIN target 73 * are executed before this function returns. 74 * Hence, the makefile must have been parsed 75 * before this function is called. 76 * 77 * Job_Full Return TRUE if the job table is filled. 78 * 79 * Job_Empty Return TRUE if the job table is completely 80 * empty. 81 * 82 * Job_ParseShell Given the line following a .SHELL target, parse 83 * the line as a shell specification. Returns 84 * FAILURE if the spec was incorrect. 85 * 86 * Job_End Perform any final processing which needs doing. 87 * This includes the execution of any commands 88 * which have been/were attached to the .END 89 * target. It should only be called when the 90 * job table is empty. 91 * 92 * Job_AbortAll Abort all currently running jobs. It doesn't 93 * handle output or do anything for the jobs, 94 * just kills them. It should only be called in 95 * an emergency, as it were. 96 * 97 * Job_CheckCommands Verify that the commands for a target are 98 * ok. Provide them if necessary and possible. 99 * 100 * Job_Touch Update a target without really updating it. 101 * 102 * Job_Wait Wait for all currently-running jobs to finish. 103 */ 104 105 #include <sys/types.h> 106 #include <sys/stat.h> 107 #include <sys/file.h> 108 #include <sys/time.h> 109 #include <sys/wait.h> 110 #include <fcntl.h> 111 #include <errno.h> 112 #include <signal.h> 113 #include <stdio.h> 114 #include <string.h> 115 #include <signal.h> 116 #include "make.h" 117 #include "hash.h" 118 #include "dir.h" 119 #include "job.h" 120 #include "pathnames.h" 121 #ifdef REMOTE 122 #include "rmt.h" 123 # define STATIC 124 #else 125 # define STATIC static 126 #endif 127 128 extern int errno; 129 130 /* 131 * error handling variables 132 */ 133 static int errors = 0; /* number of errors reported */ 134 static int aborting = 0; /* why is the make aborting? */ 135 #define ABORT_ERROR 1 /* Because of an error */ 136 #define ABORT_INTERRUPT 2 /* Because it was interrupted */ 137 #define ABORT_WAIT 3 /* Waiting for jobs to finish */ 138 139 /* 140 * XXX: Avoid SunOS bug... FILENO() is fp->_file, and file 141 * is a char! So when we go above 127 we turn negative! 142 */ 143 #define FILENO(a) ((unsigned) fileno(a)) 144 145 /* 146 * post-make command processing. The node postCommands is really just the 147 * .END target but we keep it around to avoid having to search for it 148 * all the time. 149 */ 150 static GNode *postCommands; /* node containing commands to execute when 151 * everything else is done */ 152 static int numCommands; /* The number of commands actually printed 153 * for a target. Should this number be 154 * 0, no shell will be executed. */ 155 156 /* 157 * Return values from JobStart. 158 */ 159 #define JOB_RUNNING 0 /* Job is running */ 160 #define JOB_ERROR 1 /* Error in starting the job */ 161 #define JOB_FINISHED 2 /* The job is already finished */ 162 #define JOB_STOPPED 3 /* The job is stopped */ 163 164 /* 165 * tfile is the name of a file into which all shell commands are put. It is 166 * used over by removing it before the child shell is executed. The XXXXX in 167 * the string are replaced by the pid of the make process in a 5-character 168 * field with leading zeroes. 169 */ 170 static char tfile[] = TMPPAT; 171 172 173 /* 174 * Descriptions for various shells. 175 */ 176 static Shell shells[] = { 177 /* 178 * CSH description. The csh can do echo control by playing 179 * with the setting of the 'echo' shell variable. Sadly, 180 * however, it is unable to do error control nicely. 181 */ 182 { 183 "csh", 184 TRUE, "unset verbose", "set verbose", "unset verbose", 10, 185 FALSE, "echo \"%s\"\n", "csh -c \"%s || exit 0\"", 186 "v", "e", 187 }, 188 /* 189 * SH description. Echo control is also possible and, under 190 * sun UNIX anyway, one can even control error checking. 191 */ 192 { 193 "sh", 194 TRUE, "set -", "set -v", "set -", 5, 195 TRUE, "set -e", "set +e", 196 #ifdef OLDBOURNESHELL 197 FALSE, "echo \"%s\"\n", "sh -c '%s || exit 0'\n", 198 #endif 199 "v", "e", 200 }, 201 /* 202 * UNKNOWN. 203 */ 204 { 205 (char *) 0, 206 FALSE, (char *) 0, (char *) 0, (char *) 0, 0, 207 FALSE, (char *) 0, (char *) 0, 208 (char *) 0, (char *) 0, 209 } 210 }; 211 static Shell *commandShell = &shells[DEFSHELL];/* this is the shell to 212 * which we pass all 213 * commands in the Makefile. 214 * It is set by the 215 * Job_ParseShell function */ 216 static char *shellPath = NULL, /* full pathname of 217 * executable image */ 218 *shellName; /* last component of shell */ 219 220 221 static int maxJobs; /* The most children we can run at once */ 222 static int maxLocal; /* The most local ones we can have */ 223 STATIC int nJobs; /* The number of children currently running */ 224 STATIC int nLocal; /* The number of local children */ 225 STATIC Lst jobs; /* The structures that describe them */ 226 STATIC Boolean jobFull; /* Flag to tell when the job table is full. It 227 * is set TRUE when (1) the total number of 228 * running jobs equals the maximum allowed or 229 * (2) a job can only be run locally, but 230 * nLocal equals maxLocal */ 231 #ifndef RMT_WILL_WATCH 232 static fd_set outputs; /* Set of descriptors of pipes connected to 233 * the output channels of children */ 234 #endif 235 236 STATIC GNode *lastNode; /* The node for which output was most recently 237 * produced. */ 238 STATIC char *targFmt; /* Format string to use to head output from a 239 * job when it's not the most-recent job heard 240 * from */ 241 242 #ifdef REMOTE 243 # define TARG_FMT "--- %s at %s ---\n" /* Default format */ 244 # define MESSAGE(fp, gn) \ 245 (void) fprintf(fp, targFmt, gn->name, gn->rem.hname); 246 #else 247 # define TARG_FMT "--- %s ---\n" /* Default format */ 248 # define MESSAGE(fp, gn) \ 249 (void) fprintf(fp, targFmt, gn->name); 250 #endif 251 252 /* 253 * When JobStart attempts to run a job remotely but can't, and isn't allowed 254 * to run the job locally, or when Job_CatchChildren detects a job that has 255 * been migrated home, the job is placed on the stoppedJobs queue to be run 256 * when the next job finishes. 257 */ 258 STATIC Lst stoppedJobs; /* Lst of Job structures describing 259 * jobs that were stopped due to concurrency 260 * limits or migration home */ 261 262 263 #if defined(USE_PGRP) && defined(SYSV) 264 # define KILL(pid, sig) killpg(-(pid), (sig)) 265 #else 266 # if defined(USE_PGRP) 267 # define KILL(pid, sig) killpg((pid), (sig)) 268 # else 269 # define KILL(pid, sig) kill((pid), (sig)) 270 # endif 271 #endif 272 273 static int JobCondPassSig __P((ClientData, ClientData)); 274 static void JobPassSig __P((int)); 275 static int JobCmpPid __P((ClientData, ClientData)); 276 static int JobPrintCommand __P((ClientData, ClientData)); 277 static int JobSaveCommand __P((ClientData, ClientData)); 278 static void JobClose __P((Job *)); 279 #ifdef REMOTE 280 static int JobCmpRmtID __P((Job *, int)); 281 # ifdef RMT_WILL_WATCH 282 static void JobLocalInput __P((int, Job *)); 283 # endif 284 #else 285 static void JobFinish __P((Job *, union wait *)); 286 static void JobExec __P((Job *, char **)); 287 #endif 288 static void JobMakeArgv __P((Job *, char **)); 289 static void JobRestart __P((Job *)); 290 static int JobStart __P((GNode *, int, Job *)); 291 static void JobDoOutput __P((Job *, Boolean)); 292 static Shell *JobMatchShell __P((char *)); 293 static void JobInterrupt __P((int, int)); 294 static void JobRestartJobs __P((void)); 295 296 /*- 297 *----------------------------------------------------------------------- 298 * JobCondPassSig -- 299 * Pass a signal to a job if the job is remote or if USE_PGRP 300 * is defined. 301 * 302 * Results: 303 * === 0 304 * 305 * Side Effects: 306 * None, except the job may bite it. 307 * 308 *----------------------------------------------------------------------- 309 */ 310 static int 311 JobCondPassSig(jobp, signop) 312 ClientData jobp; /* Job to biff */ 313 ClientData signop; /* Signal to send it */ 314 { 315 Job *job = (Job *) jobp; 316 int signo = *(int *) signop; 317 #ifdef RMT_WANTS_SIGNALS 318 if (job->flags & JOB_REMOTE) { 319 (void) Rmt_Signal(job, signo); 320 } else { 321 KILL(job->pid, signo); 322 } 323 #else 324 /* 325 * Assume that sending the signal to job->pid will signal any remote 326 * job as well. 327 */ 328 if (DEBUG(JOB)) { 329 (void) fprintf(stdout, 330 "JobCondPassSig passing signal %d to child %d.\n", 331 signo, job->pid); 332 (void) fflush(stdout); 333 } 334 KILL(job->pid, signo); 335 #endif 336 return 0; 337 } 338 339 /*- 340 *----------------------------------------------------------------------- 341 * JobPassSig -- 342 * Pass a signal on to all remote jobs and to all local jobs if 343 * USE_PGRP is defined, then die ourselves. 344 * 345 * Results: 346 * None. 347 * 348 * Side Effects: 349 * We die by the same signal. 350 * 351 *----------------------------------------------------------------------- 352 */ 353 static void 354 JobPassSig(signo) 355 int signo; /* The signal number we've received */ 356 { 357 int mask; 358 359 if (DEBUG(JOB)) { 360 (void) fprintf(stdout, "JobPassSig(%d) called.\n", signo); 361 (void) fflush(stdout); 362 } 363 Lst_ForEach(jobs, JobCondPassSig, (ClientData) &signo); 364 365 /* 366 * Deal with proper cleanup based on the signal received. We only run 367 * the .INTERRUPT target if the signal was in fact an interrupt. The other 368 * three termination signals are more of a "get out *now*" command. 369 */ 370 if (signo == SIGINT) { 371 JobInterrupt(TRUE, signo); 372 } else if ((signo == SIGHUP) || (signo == SIGTERM) || (signo == SIGQUIT)) { 373 JobInterrupt(FALSE, signo); 374 } 375 376 /* 377 * Leave gracefully if SIGQUIT, rather than core dumping. 378 */ 379 if (signo == SIGQUIT) { 380 Finish(0); 381 } 382 383 /* 384 * Send ourselves the signal now we've given the message to everyone else. 385 * Note we block everything else possible while we're getting the signal. 386 * This ensures that all our jobs get continued when we wake up before 387 * we take any other signal. 388 */ 389 mask = sigblock(0); 390 (void) sigsetmask(~0 & ~(1 << (signo-1))); 391 if (DEBUG(JOB)) { 392 (void) fprintf(stdout, 393 "JobPassSig passing signal to self, mask = %x.\n", 394 ~0 & ~(1 << (signo-1))); 395 (void) fflush(stdout); 396 } 397 (void) signal(signo, SIG_DFL); 398 399 (void) KILL(getpid(), signo); 400 401 signo = SIGCONT; 402 Lst_ForEach(jobs, JobCondPassSig, (ClientData) &signo); 403 404 (void) sigsetmask(mask); 405 (void) signal(signo, JobPassSig); 406 407 } 408 409 /*- 410 *----------------------------------------------------------------------- 411 * JobCmpPid -- 412 * Compare the pid of the job with the given pid and return 0 if they 413 * are equal. This function is called from Job_CatchChildren via 414 * Lst_Find to find the job descriptor of the finished job. 415 * 416 * Results: 417 * 0 if the pid's match 418 * 419 * Side Effects: 420 * None 421 *----------------------------------------------------------------------- 422 */ 423 static int 424 JobCmpPid(job, pid) 425 ClientData job; /* job to examine */ 426 ClientData pid; /* process id desired */ 427 { 428 return( *(int *) pid - ((Job *) job)->pid); 429 } 430 431 #ifdef REMOTE 432 /*- 433 *----------------------------------------------------------------------- 434 * JobCmpRmtID -- 435 * Compare the rmtID of the job with the given rmtID and return 0 if they 436 * are equal. 437 * 438 * Results: 439 * 0 if the rmtID's match 440 * 441 * Side Effects: 442 * None. 443 *----------------------------------------------------------------------- 444 */ 445 static int 446 JobCmpRmtID(job, rmtID) 447 ClientData job; /* job to examine */ 448 ClientData rmtID; /* remote id desired */ 449 { 450 return(*(int *) rmtID - *(int *) job->rmtID); 451 } 452 #endif 453 454 /*- 455 *----------------------------------------------------------------------- 456 * JobPrintCommand -- 457 * Put out another command for the given job. If the command starts 458 * with an @ or a - we process it specially. In the former case, 459 * so long as the -s and -n flags weren't given to make, we stick 460 * a shell-specific echoOff command in the script. In the latter, 461 * we ignore errors for the entire job, unless the shell has error 462 * control. 463 * If the command is just "..." we take all future commands for this 464 * job to be commands to be executed once the entire graph has been 465 * made and return non-zero to signal that the end of the commands 466 * was reached. These commands are later attached to the postCommands 467 * node and executed by Job_End when all things are done. 468 * This function is called from JobStart via Lst_ForEach. 469 * 470 * Results: 471 * Always 0, unless the command was "..." 472 * 473 * Side Effects: 474 * If the command begins with a '-' and the shell has no error control, 475 * the JOB_IGNERR flag is set in the job descriptor. 476 * If the command is "..." and we're not ignoring such things, 477 * tailCmds is set to the successor node of the cmd. 478 * numCommands is incremented if the command is actually printed. 479 *----------------------------------------------------------------------- 480 */ 481 static int 482 JobPrintCommand(cmdp, jobp) 483 ClientData cmdp; /* command string to print */ 484 ClientData jobp; /* job for which to print it */ 485 { 486 Boolean noSpecials; /* true if we shouldn't worry about 487 * inserting special commands into 488 * the input stream. */ 489 Boolean shutUp = FALSE; /* true if we put a no echo command 490 * into the command file */ 491 Boolean errOff = FALSE; /* true if we turned error checking 492 * off before printing the command 493 * and need to turn it back on */ 494 char *cmdTemplate; /* Template to use when printing the 495 * command */ 496 char *cmdStart; /* Start of expanded command */ 497 LstNode cmdNode; /* Node for replacing the command */ 498 char *cmd = (char *) cmdp; 499 Job *job = (Job *) jobp; 500 501 noSpecials = (noExecute && !(job->node->type & OP_MAKE)); 502 503 if (strcmp(cmd, "...") == 0) { 504 job->node->type |= OP_SAVE_CMDS; 505 if ((job->flags & JOB_IGNDOTS) == 0) { 506 job->tailCmds = Lst_Succ(Lst_Member(job->node->commands, 507 (ClientData)cmd)); 508 return(1); 509 } 510 return(0); 511 } 512 513 #define DBPRINTF(fmt, arg) if (DEBUG(JOB)) { \ 514 (void) fprintf(stdout, fmt, arg); \ 515 (void) fflush(stdout); \ 516 } \ 517 (void) fprintf(job->cmdFILE, fmt, arg); \ 518 (void) fflush(job->cmdFILE); 519 520 numCommands += 1; 521 522 /* 523 * For debugging, we replace each command with the result of expanding 524 * the variables in the command. 525 */ 526 cmdNode = Lst_Member(job->node->commands, (ClientData)cmd); 527 cmdStart = cmd = Var_Subst(NULL, cmd, job->node, FALSE); 528 Lst_Replace(cmdNode, (ClientData)cmdStart); 529 530 cmdTemplate = "%s\n"; 531 532 /* 533 * Check for leading @' and -'s to control echoing and error checking. 534 */ 535 while (*cmd == '@' || *cmd == '-') { 536 if (*cmd == '@') { 537 shutUp = TRUE; 538 } else { 539 errOff = TRUE; 540 } 541 cmd++; 542 } 543 544 while (isspace((unsigned char) *cmd)) 545 cmd++; 546 547 if (shutUp) { 548 if (!(job->flags & JOB_SILENT) && !noSpecials && 549 commandShell->hasEchoCtl) { 550 DBPRINTF("%s\n", commandShell->echoOff); 551 } else { 552 shutUp = FALSE; 553 } 554 } 555 556 if (errOff) { 557 if ( !(job->flags & JOB_IGNERR) && !noSpecials) { 558 if (commandShell->hasErrCtl) { 559 /* 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 565 * it already is? 566 */ 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); 572 } else { 573 DBPRINTF("%s\n", commandShell->ignErr); 574 } 575 } else if (commandShell->ignErr && 576 (*commandShell->ignErr != '\0')) 577 { 578 /* 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. 586 */ 587 if (!(job->flags & JOB_SILENT) && !shutUp && 588 commandShell->hasEchoCtl) { 589 DBPRINTF("%s\n", commandShell->echoOff); 590 DBPRINTF(commandShell->errCheck, cmd); 591 shutUp = TRUE; 592 } 593 cmdTemplate = commandShell->ignErr; 594 /* 595 * The error ignoration(hee hee) is already taken care 596 * of by the ignErr template, so pretend error checking 597 * is still on. 598 */ 599 errOff = FALSE; 600 } else { 601 errOff = FALSE; 602 } 603 } else { 604 errOff = FALSE; 605 } 606 } 607 608 DBPRINTF(cmdTemplate, cmd); 609 610 if (errOff) { 611 /* 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... 615 */ 616 if (!shutUp && !(job->flags & JOB_SILENT) && commandShell->hasEchoCtl){ 617 DBPRINTF("%s\n", commandShell->echoOff); 618 shutUp = TRUE; 619 } 620 DBPRINTF("%s\n", commandShell->errCheck); 621 } 622 if (shutUp) { 623 DBPRINTF("%s\n", commandShell->echoOn); 624 } 625 return 0; 626 } 627 628 /*- 629 *----------------------------------------------------------------------- 630 * JobSaveCommand -- 631 * Save a command to be executed when everything else is done. 632 * Callback function for JobFinish... 633 * 634 * Results: 635 * Always returns 0 636 * 637 * Side Effects: 638 * The command is tacked onto the end of postCommands's commands list. 639 * 640 *----------------------------------------------------------------------- 641 */ 642 static int 643 JobSaveCommand(cmd, gn) 644 ClientData cmd; 645 ClientData gn; 646 { 647 cmd = (ClientData) Var_Subst(NULL, (char *) cmd, (GNode *) gn, FALSE); 648 (void) Lst_AtEnd(postCommands->commands, cmd); 649 return(0); 650 } 651 652 653 /*- 654 *----------------------------------------------------------------------- 655 * JobClose -- 656 * Called to close both input and output pipes when a job is finished. 657 * 658 * Results: 659 * Nada 660 * 661 * Side Effects: 662 * The file descriptors associated with the job are closed. 663 * 664 *----------------------------------------------------------------------- 665 */ 666 static void 667 JobClose(job) 668 Job *job; 669 { 670 if (usePipes) { 671 #ifdef RMT_WILL_WATCH 672 Rmt_Ignore(job->inPipe); 673 #else 674 FD_CLR(job->inPipe, &outputs); 675 #endif 676 if (job->outPipe != job->inPipe) { 677 (void) close(job->outPipe); 678 } 679 JobDoOutput(job, TRUE); 680 (void) close(job->inPipe); 681 } else { 682 (void) close(job->outFd); 683 JobDoOutput(job, TRUE); 684 } 685 } 686 687 /*- 688 *----------------------------------------------------------------------- 689 * JobFinish -- 690 * Do final processing for the given job including updating 691 * parents and starting new jobs as available/necessary. Note 692 * that we pay no attention to the JOB_IGNERR flag here. 693 * This is because when we're called because of a noexecute flag 694 * or something, jstat.w_status is 0 and when called from 695 * Job_CatchChildren, the status is zeroed if it s/b ignored. 696 * 697 * Results: 698 * None 699 * 700 * Side Effects: 701 * Some nodes may be put on the toBeMade queue. 702 * Final commands for the job are placed on postCommands. 703 * 704 * If we got an error and are aborting(aborting == ABORT_ERROR) and 705 * the job list is now empty, we are done for the day. 706 * If we recognized an error(errors !=0), we set the aborting flag 707 * to ABORT_ERROR so no more jobs will be started. 708 *----------------------------------------------------------------------- 709 */ 710 /*ARGSUSED*/ 711 static void 712 JobFinish(job, status) 713 Job *job; /* job to finish */ 714 union wait *status; /* sub-why job went away */ 715 { 716 Boolean done; 717 718 if ((WIFEXITED(*status) && 719 (((WEXITSTATUS(*status) != 0) && !(job->flags & JOB_IGNERR)))) || 720 (WIFSIGNALED(*status) && (WTERMSIG(*status) != SIGCONT))) 721 { 722 /* 723 * If it exited non-zero and either we're doing things our 724 * way or we're not ignoring errors, the job is finished. 725 * Similarly, if the shell died because of a signal 726 * the job is also finished. In these 727 * cases, finish out the job's output before printing the exit 728 * status... 729 */ 730 #ifdef REMOTE 731 KILL(job->pid, SIGCONT); 732 #endif 733 JobClose(job); 734 if (job->cmdFILE != NULL && job->cmdFILE != stdout) { 735 (void) fclose(job->cmdFILE); 736 } 737 done = TRUE; 738 #ifdef REMOTE 739 if (job->flags & JOB_REMOTE) 740 Rmt_Done(job->rmtID, job->node); 741 #endif 742 } else if (WIFEXITED(*status)) { 743 /* 744 * Deal with ignored errors in -B mode. We need to print a message 745 * telling of the ignored error as well as setting status.w_status 746 * to 0 so the next command gets run. To do this, we set done to be 747 * TRUE if in -B mode and the job exited non-zero. 748 */ 749 done = WEXITSTATUS(*status) != 0; 750 /* 751 * Old comment said: "Note we don't 752 * want to close down any of the streams until we know we're at the 753 * end." 754 * But we do. Otherwise when are we going to print the rest of the 755 * stuff? 756 */ 757 JobClose(job); 758 #ifdef REMOTE 759 if (job->flags & JOB_REMOTE) 760 Rmt_Done(job->rmtID, job->node); 761 #endif /* REMOTE */ 762 } else { 763 /* 764 * No need to close things down or anything. 765 */ 766 done = FALSE; 767 } 768 769 if (done || 770 WIFSTOPPED(*status) || 771 (WIFSIGNALED(*status) && (WTERMSIG(*status) == SIGCONT)) || 772 DEBUG(JOB)) 773 { 774 FILE *out; 775 776 if (compatMake && !usePipes && (job->flags & JOB_IGNERR)) { 777 /* 778 * If output is going to a file and this job is ignoring 779 * errors, arrange to have the exit status sent to the 780 * output file as well. 781 */ 782 out = fdopen(job->outFd, "w"); 783 } else { 784 out = stdout; 785 } 786 787 if (WIFEXITED(*status)) { 788 if (DEBUG(JOB)) { 789 (void) fprintf(stdout, "Process %d exited.\n", job->pid); 790 (void) fflush(stdout); 791 } 792 if (WEXITSTATUS(*status) != 0) { 793 if (usePipes && job->node != lastNode) { 794 MESSAGE(out, job->node); 795 lastNode = job->node; 796 } 797 (void) fprintf(out, "*** Error code %d%s\n", 798 WEXITSTATUS(*status), 799 (job->flags & JOB_IGNERR) ? "(ignored)" : ""); 800 801 if (job->flags & JOB_IGNERR) { 802 status->w_status = 0; 803 } 804 } else if (DEBUG(JOB)) { 805 if (usePipes && job->node != lastNode) { 806 MESSAGE(out, job->node); 807 lastNode = job->node; 808 } 809 (void) fprintf(out, "*** Completed successfully\n"); 810 } 811 } else if (WIFSTOPPED(*status)) { 812 if (DEBUG(JOB)) { 813 (void) fprintf(stdout, "Process %d stopped.\n", job->pid); 814 (void) fflush(stdout); 815 } 816 if (usePipes && job->node != lastNode) { 817 MESSAGE(out, job->node); 818 lastNode = job->node; 819 } 820 if (!(job->flags & JOB_REMIGRATE)) { 821 fprintf(out, "*** Stopped -- signal %d\n", WSTOPSIG(*status)); 822 } 823 job->flags |= JOB_RESUME; 824 (void)Lst_AtEnd(stoppedJobs, (ClientData)job); 825 #ifdef REMOTE 826 if (job->flags & JOB_REMIGRATE) 827 JobRestart(job); 828 #endif 829 (void) fflush(out); 830 return; 831 } else if (WTERMSIG(*status) == SIGCONT) { 832 /* 833 * If the beastie has continued, shift the Job from the stopped 834 * list to the running one(or re-stop it if concurrency is 835 * exceeded) and go and get another child. 836 */ 837 if (job->flags & (JOB_RESUME|JOB_REMIGRATE|JOB_RESTART)) { 838 if (usePipes && job->node != lastNode) { 839 MESSAGE(out, job->node); 840 lastNode = job->node; 841 } 842 (void) fprintf(out, "*** Continued\n"); 843 } 844 if (!(job->flags & JOB_CONTINUING)) { 845 if (DEBUG(JOB)) { 846 (void) fprintf(stdout, 847 "Warning: process %d was not continuing.\n", 848 job->pid); 849 (void) fflush(stdout); 850 } 851 #ifdef notdef 852 /* 853 * We don't really want to restart a job from scratch just 854 * because it continued, especially not without killing the 855 * continuing process! That's why this is ifdef'ed out. 856 * FD - 9/17/90 857 */ 858 JobRestart(job); 859 #endif 860 } 861 job->flags &= ~JOB_CONTINUING; 862 Lst_AtEnd(jobs, (ClientData)job); 863 nJobs += 1; 864 if (!(job->flags & JOB_REMOTE)) { 865 if (DEBUG(JOB)) { 866 (void) fprintf(stdout, 867 "Process %d is continuing locally.\n", 868 job->pid); 869 (void) fflush(stdout); 870 } 871 nLocal += 1; 872 } 873 if (nJobs == maxJobs) { 874 jobFull = TRUE; 875 if (DEBUG(JOB)) { 876 (void) fprintf(stdout, "Job queue is full.\n"); 877 (void) fflush(stdout); 878 } 879 } 880 (void) fflush(out); 881 return; 882 } else { 883 if (usePipes && job->node != lastNode) { 884 MESSAGE(out, job->node); 885 lastNode = job->node; 886 } 887 (void) fprintf(out, "*** Signal %d\n", WTERMSIG(*status)); 888 } 889 890 (void) fflush(out); 891 } 892 893 /* 894 * Now handle the -B-mode stuff. If the beast still isn't finished, 895 * try and restart the job on the next command. If JobStart says it's 896 * ok, it's ok. If there's an error, this puppy is done. 897 */ 898 if (compatMake && (WIFEXITED(*status) && 899 !Lst_IsAtEnd(job->node->commands))) { 900 switch (JobStart(job->node, job->flags & JOB_IGNDOTS, job)) { 901 case JOB_RUNNING: 902 done = FALSE; 903 break; 904 case JOB_ERROR: 905 done = TRUE; 906 status->w_retcode = 1; 907 break; 908 case JOB_FINISHED: 909 /* 910 * If we got back a JOB_FINISHED code, JobStart has already 911 * called Make_Update and freed the job descriptor. We set 912 * done to false here to avoid fake cycles and double frees. 913 * JobStart needs to do the update so we can proceed up the 914 * graph when given the -n flag.. 915 */ 916 done = FALSE; 917 break; 918 } 919 } else { 920 done = TRUE; 921 } 922 923 924 if (done && 925 (aborting != ABORT_ERROR) && 926 (aborting != ABORT_INTERRUPT) && 927 (status->w_status == 0)) 928 { 929 /* 930 * As long as we aren't aborting and the job didn't return a non-zero 931 * status that we shouldn't ignore, we call Make_Update to update 932 * the parents. In addition, any saved commands for the node are placed 933 * on the .END target. 934 */ 935 if (job->tailCmds != NILLNODE) { 936 Lst_ForEachFrom(job->node->commands, job->tailCmds, 937 JobSaveCommand, 938 (ClientData)job->node); 939 } 940 job->node->made = MADE; 941 Make_Update(job->node); 942 free((Address)job); 943 } else if (status->w_status) { 944 errors += 1; 945 free((Address)job); 946 } 947 948 JobRestartJobs(); 949 950 /* 951 * Set aborting if any error. 952 */ 953 if (errors && !keepgoing && (aborting != ABORT_INTERRUPT)) { 954 /* 955 * If we found any errors in this batch of children and the -k flag 956 * wasn't given, we set the aborting flag so no more jobs get 957 * started. 958 */ 959 aborting = ABORT_ERROR; 960 } 961 962 if ((aborting == ABORT_ERROR) && Job_Empty()) { 963 /* 964 * If we are aborting and the job table is now empty, we finish. 965 */ 966 (void) eunlink(tfile); 967 Finish(errors); 968 } 969 } 970 971 /*- 972 *----------------------------------------------------------------------- 973 * Job_Touch -- 974 * Touch the given target. Called by JobStart when the -t flag was 975 * given 976 * 977 * Results: 978 * None 979 * 980 * Side Effects: 981 * The data modification of the file is changed. In addition, if the 982 * file did not exist, it is created. 983 *----------------------------------------------------------------------- 984 */ 985 void 986 Job_Touch(gn, silent) 987 GNode *gn; /* the node of the file to touch */ 988 Boolean silent; /* TRUE if should not print messages */ 989 { 990 int streamID; /* ID of stream opened to do the touch */ 991 struct timeval times[2]; /* Times for utimes() call */ 992 993 if (gn->type & (OP_JOIN|OP_USE|OP_EXEC|OP_OPTIONAL)) { 994 /* 995 * .JOIN, .USE, .ZEROTIME and .OPTIONAL targets are "virtual" targets 996 * and, as such, shouldn't really be created. 997 */ 998 return; 999 } 1000 1001 if (!silent) { 1002 (void) fprintf(stdout, "touch %s\n", gn->name); 1003 (void) fflush(stdout); 1004 } 1005 1006 if (noExecute) { 1007 return; 1008 } 1009 1010 if (gn->type & OP_ARCHV) { 1011 Arch_Touch(gn); 1012 } else if (gn->type & OP_LIB) { 1013 Arch_TouchLib(gn); 1014 } else { 1015 char *file = gn->path ? gn->path : gn->name; 1016 1017 times[0].tv_sec = times[1].tv_sec = now; 1018 times[0].tv_usec = times[1].tv_usec = 0; 1019 if (utimes(file, times) < 0){ 1020 streamID = open(file, O_RDWR | O_CREAT, 0666); 1021 1022 if (streamID >= 0) { 1023 char c; 1024 1025 /* 1026 * Read and write a byte to the file to change the 1027 * modification time, then close the file. 1028 */ 1029 if (read(streamID, &c, 1) == 1) { 1030 (void) lseek(streamID, 0L, L_SET); 1031 (void) write(streamID, &c, 1); 1032 } 1033 1034 (void) close(streamID); 1035 } else { 1036 (void) fprintf(stdout, "*** couldn't touch %s: %s", 1037 file, strerror(errno)); 1038 (void) fflush(stdout); 1039 } 1040 } 1041 } 1042 } 1043 1044 /*- 1045 *----------------------------------------------------------------------- 1046 * Job_CheckCommands -- 1047 * Make sure the given node has all the commands it needs. 1048 * 1049 * Results: 1050 * TRUE if the commands list is/was ok. 1051 * 1052 * Side Effects: 1053 * The node will have commands from the .DEFAULT rule added to it 1054 * if it needs them. 1055 *----------------------------------------------------------------------- 1056 */ 1057 Boolean 1058 Job_CheckCommands(gn, abortProc) 1059 GNode *gn; /* The target whose commands need 1060 * verifying */ 1061 void (*abortProc) __P((char *, ...)); 1062 /* Function to abort with message */ 1063 { 1064 if (OP_NOP(gn->type) && Lst_IsEmpty(gn->commands) && 1065 (gn->type & OP_LIB) == 0) { 1066 /* 1067 * No commands. Look for .DEFAULT rule from which we might infer 1068 * commands 1069 */ 1070 if ((DEFAULT != NILGNODE) && !Lst_IsEmpty(DEFAULT->commands)) { 1071 char *p1; 1072 /* 1073 * Make only looks for a .DEFAULT if the node was never the 1074 * target of an operator, so that's what we do too. If 1075 * a .DEFAULT was given, we substitute its commands for gn's 1076 * commands and set the IMPSRC variable to be the target's name 1077 * The DEFAULT node acts like a transformation rule, in that 1078 * gn also inherits any attributes or sources attached to 1079 * .DEFAULT itself. 1080 */ 1081 Make_HandleUse(DEFAULT, gn); 1082 Var_Set(IMPSRC, Var_Value(TARGET, gn, &p1), gn); 1083 if (p1) 1084 free(p1); 1085 } else if (Dir_MTime(gn) == 0) { 1086 /* 1087 * The node wasn't the target of an operator we have no .DEFAULT 1088 * rule to go on and the target doesn't already exist. There's 1089 * nothing more we can do for this branch. If the -k flag wasn't 1090 * given, we stop in our tracks, otherwise we just don't update 1091 * this node's parents so they never get examined. 1092 */ 1093 static const char msg[] = "make: don't know how to make"; 1094 1095 if (gn->type & OP_OPTIONAL) { 1096 (void) fprintf(stdout, "%s %s(ignored)\n", msg, gn->name); 1097 (void) fflush(stdout); 1098 } else if (keepgoing) { 1099 (void) fprintf(stdout, "%s %s(continuing)\n", msg, gn->name); 1100 (void) fflush(stdout); 1101 return FALSE; 1102 } else { 1103 (*abortProc)("%s %s. Stop", msg, gn->name); 1104 return FALSE; 1105 } 1106 } 1107 } 1108 return TRUE; 1109 } 1110 #ifdef RMT_WILL_WATCH 1111 /*- 1112 *----------------------------------------------------------------------- 1113 * JobLocalInput -- 1114 * Handle a pipe becoming readable. Callback function for Rmt_Watch 1115 * 1116 * Results: 1117 * None 1118 * 1119 * Side Effects: 1120 * JobDoOutput is called. 1121 * 1122 *----------------------------------------------------------------------- 1123 */ 1124 /*ARGSUSED*/ 1125 static void 1126 JobLocalInput(stream, job) 1127 int stream; /* Stream that's ready(ignored) */ 1128 Job *job; /* Job to which the stream belongs */ 1129 { 1130 JobDoOutput(job, FALSE); 1131 } 1132 #endif /* RMT_WILL_WATCH */ 1133 1134 /*- 1135 *----------------------------------------------------------------------- 1136 * JobExec -- 1137 * Execute the shell for the given job. Called from JobStart and 1138 * JobRestart. 1139 * 1140 * Results: 1141 * None. 1142 * 1143 * Side Effects: 1144 * A shell is executed, outputs is altered and the Job structure added 1145 * to the job table. 1146 * 1147 *----------------------------------------------------------------------- 1148 */ 1149 static void 1150 JobExec(job, argv) 1151 Job *job; /* Job to execute */ 1152 char **argv; 1153 { 1154 int cpid; /* ID of new child */ 1155 1156 if (DEBUG(JOB)) { 1157 int i; 1158 1159 (void) fprintf(stdout, "Running %s %sly\n", job->node->name, 1160 job->flags&JOB_REMOTE?"remote":"local"); 1161 (void) fprintf(stdout, "\tCommand: "); 1162 for (i = 0; argv[i] != NULL; i++) { 1163 (void) fprintf(stdout, "%s ", argv[i]); 1164 } 1165 (void) fprintf(stdout, "\n"); 1166 (void) fflush(stdout); 1167 } 1168 1169 /* 1170 * Some jobs produce no output and it's disconcerting to have 1171 * no feedback of their running(since they produce no output, the 1172 * banner with their name in it never appears). This is an attempt to 1173 * provide that feedback, even if nothing follows it. 1174 */ 1175 if ((lastNode != job->node) && (job->flags & JOB_FIRST) && 1176 !(job->flags & JOB_SILENT)) { 1177 MESSAGE(stdout, job->node); 1178 lastNode = job->node; 1179 } 1180 1181 #ifdef RMT_NO_EXEC 1182 if (job->flags & JOB_REMOTE) { 1183 goto jobExecFinish; 1184 } 1185 #endif /* RMT_NO_EXEC */ 1186 1187 if ((cpid = vfork()) == -1) { 1188 Punt("Cannot fork"); 1189 } else if (cpid == 0) { 1190 1191 /* 1192 * Must duplicate the input stream down to the child's input and 1193 * reset it to the beginning(again). Since the stream was marked 1194 * close-on-exec, we must clear that bit in the new input. 1195 */ 1196 if (dup2(FILENO(job->cmdFILE), 0) == -1) 1197 Punt("Cannot dup2: %s", strerror(errno)); 1198 (void) fcntl(0, F_SETFD, 0); 1199 (void) lseek(0, 0, L_SET); 1200 1201 if (usePipes) { 1202 /* 1203 * Set up the child's output to be routed through the pipe 1204 * we've created for it. 1205 */ 1206 if (dup2(job->outPipe, 1) == -1) 1207 Punt("Cannot dup2: %s", strerror(errno)); 1208 } else { 1209 /* 1210 * We're capturing output in a file, so we duplicate the 1211 * descriptor to the temporary file into the standard 1212 * output. 1213 */ 1214 if (dup2(job->outFd, 1) == -1) 1215 Punt("Cannot dup2: %s", strerror(errno)); 1216 } 1217 /* 1218 * The output channels are marked close on exec. This bit was 1219 * duplicated by the dup2(on some systems), so we have to clear 1220 * it before routing the shell's error output to the same place as 1221 * its standard output. 1222 */ 1223 (void) fcntl(1, F_SETFD, 0); 1224 if (dup2(1, 2) == -1) 1225 Punt("Cannot dup2: %s", strerror(errno)); 1226 1227 #ifdef USE_PGRP 1228 /* 1229 * We want to switch the child into a different process family so 1230 * we can kill it and all its descendants in one fell swoop, 1231 * by killing its process family, but not commit suicide. 1232 */ 1233 # if defined(SYSV) 1234 (void) setsid(); 1235 # else 1236 (void) setpgid(0, getpid()); 1237 # endif 1238 #endif /* USE_PGRP */ 1239 1240 #ifdef REMOTE 1241 if (job->flags & JOB_REMOTE) { 1242 Rmt_Exec(shellPath, argv, FALSE); 1243 } else 1244 #endif /* REMOTE */ 1245 (void) execv(shellPath, argv); 1246 1247 (void) write(2, "Could not execute shell\n", 1248 sizeof("Could not execute shell")); 1249 _exit(1); 1250 } else { 1251 #ifdef REMOTE 1252 long omask = sigblock(sigmask(SIGCHLD)); 1253 #endif 1254 job->pid = cpid; 1255 1256 if (usePipes && (job->flags & JOB_FIRST) ) { 1257 /* 1258 * The first time a job is run for a node, we set the current 1259 * position in the buffer to the beginning and mark another 1260 * stream to watch in the outputs mask 1261 */ 1262 job->curPos = 0; 1263 1264 #ifdef RMT_WILL_WATCH 1265 Rmt_Watch(job->inPipe, JobLocalInput, job); 1266 #else 1267 FD_SET(job->inPipe, &outputs); 1268 #endif /* RMT_WILL_WATCH */ 1269 } 1270 1271 if (job->flags & JOB_REMOTE) { 1272 #ifndef REMOTE 1273 job->rmtID = 0; 1274 #else 1275 job->rmtID = Rmt_LastID(job->pid); 1276 #endif /* REMOTE */ 1277 } else { 1278 nLocal += 1; 1279 /* 1280 * XXX: Used to not happen if REMOTE. Why? 1281 */ 1282 if (job->cmdFILE != NULL && job->cmdFILE != stdout) { 1283 (void) fclose(job->cmdFILE); 1284 job->cmdFILE = NULL; 1285 } 1286 } 1287 #ifdef REMOTE 1288 (void) sigsetmask(omask); 1289 #endif 1290 } 1291 1292 #ifdef RMT_NO_EXEC 1293 jobExecFinish: 1294 #endif 1295 /* 1296 * Now the job is actually running, add it to the table. 1297 */ 1298 nJobs += 1; 1299 (void) Lst_AtEnd(jobs, (ClientData)job); 1300 if (nJobs == maxJobs) { 1301 jobFull = TRUE; 1302 } 1303 } 1304 1305 /*- 1306 *----------------------------------------------------------------------- 1307 * JobMakeArgv -- 1308 * Create the argv needed to execute the shell for a given job. 1309 * 1310 * 1311 * Results: 1312 * 1313 * Side Effects: 1314 * 1315 *----------------------------------------------------------------------- 1316 */ 1317 static void 1318 JobMakeArgv(job, argv) 1319 Job *job; 1320 char **argv; 1321 { 1322 int argc; 1323 static char args[10]; /* For merged arguments */ 1324 1325 argv[0] = shellName; 1326 argc = 1; 1327 1328 if ((commandShell->exit && (*commandShell->exit != '-')) || 1329 (commandShell->echo && (*commandShell->echo != '-'))) 1330 { 1331 /* 1332 * At least one of the flags doesn't have a minus before it, so 1333 * merge them together. Have to do this because the *(&(@*#*&#$# 1334 * Bourne shell thinks its second argument is a file to source. 1335 * Grrrr. Note the ten-character limitation on the combined arguments. 1336 */ 1337 (void)sprintf(args, "-%s%s", 1338 ((job->flags & JOB_IGNERR) ? "" : 1339 (commandShell->exit ? commandShell->exit : "")), 1340 ((job->flags & JOB_SILENT) ? "" : 1341 (commandShell->echo ? commandShell->echo : ""))); 1342 1343 if (args[1]) { 1344 argv[argc] = args; 1345 argc++; 1346 } 1347 } else { 1348 if (!(job->flags & JOB_IGNERR) && commandShell->exit) { 1349 argv[argc] = commandShell->exit; 1350 argc++; 1351 } 1352 if (!(job->flags & JOB_SILENT) && commandShell->echo) { 1353 argv[argc] = commandShell->echo; 1354 argc++; 1355 } 1356 } 1357 argv[argc] = NULL; 1358 } 1359 1360 /*- 1361 *----------------------------------------------------------------------- 1362 * JobRestart -- 1363 * Restart a job that stopped for some reason. 1364 * 1365 * Results: 1366 * None. 1367 * 1368 * Side Effects: 1369 * jobFull will be set if the job couldn't be run. 1370 * 1371 *----------------------------------------------------------------------- 1372 */ 1373 static void 1374 JobRestart(job) 1375 Job *job; /* Job to restart */ 1376 { 1377 #ifdef REMOTE 1378 int host; 1379 #endif 1380 1381 if (job->flags & JOB_REMIGRATE) { 1382 if ( 1383 #ifdef REMOTE 1384 verboseRemigrates || 1385 #endif 1386 DEBUG(JOB)) { 1387 (void) fprintf(stdout, "*** remigrating %x(%s)\n", 1388 job->pid, job->node->name); 1389 (void) fflush(stdout); 1390 } 1391 1392 #ifdef REMOTE 1393 if (!Rmt_ReExport(job->pid, job->node, &host)) { 1394 if (verboseRemigrates || DEBUG(JOB)) { 1395 (void) fprintf(stdout, "*** couldn't migrate...\n"); 1396 (void) fflush(stdout); 1397 } 1398 #endif 1399 if (nLocal != maxLocal) { 1400 /* 1401 * Job cannot be remigrated, but there's room on the local 1402 * machine, so resume the job and note that another 1403 * local job has started. 1404 */ 1405 if ( 1406 #ifdef REMOTE 1407 verboseRemigrates || 1408 #endif 1409 DEBUG(JOB)) { 1410 (void) fprintf(stdout, "*** resuming on local machine\n"); 1411 (void) fflush(stdout); 1412 } 1413 KILL(job->pid, SIGCONT); 1414 nLocal +=1; 1415 #ifdef REMOTE 1416 job->flags &= ~(JOB_REMIGRATE|JOB_RESUME|JOB_REMOTE); 1417 job->flags |= JOB_CONTINUING; 1418 #else 1419 job->flags &= ~(JOB_REMIGRATE|JOB_RESUME); 1420 #endif 1421 } else { 1422 /* 1423 * Job cannot be restarted. Mark the table as full and 1424 * place the job back on the list of stopped jobs. 1425 */ 1426 if ( 1427 #ifdef REMOTE 1428 verboseRemigrates || 1429 #endif 1430 DEBUG(JOB)) { 1431 (void) fprintf(stdout, "*** holding\n"); 1432 (void) fflush(stdout); 1433 } 1434 (void)Lst_AtFront(stoppedJobs, (ClientData)job); 1435 jobFull = TRUE; 1436 if (DEBUG(JOB)) { 1437 (void) fprintf(stdout, "Job queue is full.\n"); 1438 (void) fflush(stdout); 1439 } 1440 return; 1441 } 1442 #ifdef REMOTE 1443 } else { 1444 /* 1445 * Clear out the remigrate and resume flags. Set the continuing 1446 * flag so we know later on that the process isn't exiting just 1447 * because of a signal. 1448 */ 1449 job->flags &= ~(JOB_REMIGRATE|JOB_RESUME); 1450 job->flags |= JOB_CONTINUING; 1451 job->rmtID = host; 1452 } 1453 #endif 1454 1455 (void)Lst_AtEnd(jobs, (ClientData)job); 1456 nJobs += 1; 1457 if (nJobs == maxJobs) { 1458 jobFull = TRUE; 1459 if (DEBUG(JOB)) { 1460 (void) fprintf(stdout, "Job queue is full.\n"); 1461 (void) fflush(stdout); 1462 } 1463 } 1464 } else if (job->flags & JOB_RESTART) { 1465 /* 1466 * Set up the control arguments to the shell. This is based on the 1467 * flags set earlier for this job. If the JOB_IGNERR flag is clear, 1468 * the 'exit' flag of the commandShell is used to cause it to exit 1469 * upon receiving an error. If the JOB_SILENT flag is clear, the 1470 * 'echo' flag of the commandShell is used to get it to start echoing 1471 * as soon as it starts processing commands. 1472 */ 1473 char *argv[4]; 1474 1475 JobMakeArgv(job, argv); 1476 1477 if (DEBUG(JOB)) { 1478 (void) fprintf(stdout, "Restarting %s...", job->node->name); 1479 (void) fflush(stdout); 1480 } 1481 #ifdef REMOTE 1482 if ((job->node->type&OP_NOEXPORT) || 1483 (nLocal < maxLocal && runLocalFirst) 1484 # ifdef RMT_NO_EXEC 1485 || !Rmt_Export(shellPath, argv, job) 1486 # else 1487 || !Rmt_Begin(shellPath, argv, job->node) 1488 # endif 1489 #endif 1490 { 1491 if (((nLocal >= maxLocal) && !(job->flags & JOB_SPECIAL))) { 1492 /* 1493 * Can't be exported and not allowed to run locally -- put it 1494 * back on the hold queue and mark the table full 1495 */ 1496 if (DEBUG(JOB)) { 1497 (void) fprintf(stdout, "holding\n"); 1498 (void) fflush(stdout); 1499 } 1500 (void)Lst_AtFront(stoppedJobs, (ClientData)job); 1501 jobFull = TRUE; 1502 if (DEBUG(JOB)) { 1503 (void) fprintf(stdout, "Job queue is full.\n"); 1504 (void) fflush(stdout); 1505 } 1506 return; 1507 } else { 1508 /* 1509 * Job may be run locally. 1510 */ 1511 if (DEBUG(JOB)) { 1512 (void) fprintf(stdout, "running locally\n"); 1513 (void) fflush(stdout); 1514 } 1515 job->flags &= ~JOB_REMOTE; 1516 } 1517 } 1518 #ifdef REMOTE 1519 else { 1520 /* 1521 * Can be exported. Hooray! 1522 */ 1523 if (DEBUG(JOB)) { 1524 (void) fprintf(stdout, "exporting\n"); 1525 (void) fflush(stdout); 1526 } 1527 job->flags |= JOB_REMOTE; 1528 } 1529 #endif 1530 JobExec(job, argv); 1531 } else { 1532 /* 1533 * The job has stopped and needs to be restarted. Why it stopped, 1534 * we don't know... 1535 */ 1536 if (DEBUG(JOB)) { 1537 (void) fprintf(stdout, "Resuming %s...", job->node->name); 1538 (void) fflush(stdout); 1539 } 1540 if (((job->flags & JOB_REMOTE) || 1541 (nLocal < maxLocal) || 1542 #ifdef REMOTE 1543 (((job->flags & JOB_SPECIAL) && 1544 (job->node->type & OP_NOEXPORT)) && 1545 (maxLocal == 0))) && 1546 #else 1547 ((job->flags & JOB_SPECIAL) && 1548 (maxLocal == 0))) && 1549 #endif 1550 (nJobs != maxJobs)) 1551 { 1552 /* 1553 * If the job is remote, it's ok to resume it as long as the 1554 * maximum concurrency won't be exceeded. If it's local and 1555 * we haven't reached the local concurrency limit already(or the 1556 * job must be run locally and maxLocal is 0), it's also ok to 1557 * resume it. 1558 */ 1559 Boolean error; 1560 extern int errno; 1561 union wait status; 1562 1563 #ifdef RMT_WANTS_SIGNALS 1564 if (job->flags & JOB_REMOTE) { 1565 error = !Rmt_Signal(job, SIGCONT); 1566 } else 1567 #endif /* RMT_WANTS_SIGNALS */ 1568 error = (KILL(job->pid, SIGCONT) != 0); 1569 1570 if (!error) { 1571 /* 1572 * Make sure the user knows we've continued the beast and 1573 * actually put the thing in the job table. 1574 */ 1575 job->flags |= JOB_CONTINUING; 1576 status.w_termsig = SIGCONT; 1577 JobFinish(job, &status); 1578 1579 job->flags &= ~(JOB_RESUME|JOB_CONTINUING); 1580 if (DEBUG(JOB)) { 1581 (void) fprintf(stdout, "done\n"); 1582 (void) fflush(stdout); 1583 } 1584 } else { 1585 Error("couldn't resume %s: %s", 1586 job->node->name, strerror(errno)); 1587 status.w_status = 0; 1588 status.w_retcode = 1; 1589 JobFinish(job, &status); 1590 } 1591 } else { 1592 /* 1593 * Job cannot be restarted. Mark the table as full and 1594 * place the job back on the list of stopped jobs. 1595 */ 1596 if (DEBUG(JOB)) { 1597 (void) fprintf(stdout, "table full\n"); 1598 (void) fflush(stdout); 1599 } 1600 (void) Lst_AtFront(stoppedJobs, (ClientData)job); 1601 jobFull = TRUE; 1602 if (DEBUG(JOB)) { 1603 (void) fprintf(stdout, "Job queue is full.\n"); 1604 (void) fflush(stdout); 1605 } 1606 } 1607 } 1608 } 1609 1610 /*- 1611 *----------------------------------------------------------------------- 1612 * JobStart -- 1613 * Start a target-creation process going for the target described 1614 * by the graph node gn. 1615 * 1616 * Results: 1617 * JOB_ERROR if there was an error in the commands, JOB_FINISHED 1618 * if there isn't actually anything left to do for the job and 1619 * JOB_RUNNING if the job has been started. 1620 * 1621 * Side Effects: 1622 * A new Job node is created and added to the list of running 1623 * jobs. PMake is forked and a child shell created. 1624 *----------------------------------------------------------------------- 1625 */ 1626 static int 1627 JobStart(gn, flags, previous) 1628 GNode *gn; /* target to create */ 1629 int flags; /* flags for the job to override normal ones. 1630 * e.g. JOB_SPECIAL or JOB_IGNDOTS */ 1631 Job *previous; /* The previous Job structure for this node, 1632 * if any. */ 1633 { 1634 register Job *job; /* new job descriptor */ 1635 char *argv[4]; /* Argument vector to shell */ 1636 static int jobno = 0; /* job number of catching output in a file */ 1637 Boolean cmdsOK; /* true if the nodes commands were all right */ 1638 Boolean local; /* Set true if the job was run locally */ 1639 Boolean noExec; /* Set true if we decide not to run the job */ 1640 1641 if (previous != NULL) { 1642 previous->flags &= ~(JOB_FIRST|JOB_IGNERR|JOB_SILENT|JOB_REMOTE); 1643 job = previous; 1644 } else { 1645 job = (Job *) emalloc(sizeof(Job)); 1646 if (job == NULL) { 1647 Punt("JobStart out of memory"); 1648 } 1649 flags |= JOB_FIRST; 1650 } 1651 1652 job->node = gn; 1653 job->tailCmds = NILLNODE; 1654 1655 /* 1656 * Set the initial value of the flags for this job based on the global 1657 * ones and the node's attributes... Any flags supplied by the caller 1658 * are also added to the field. 1659 */ 1660 job->flags = 0; 1661 if (Targ_Ignore(gn)) { 1662 job->flags |= JOB_IGNERR; 1663 } 1664 if (Targ_Silent(gn)) { 1665 job->flags |= JOB_SILENT; 1666 } 1667 job->flags |= flags; 1668 1669 /* 1670 * Check the commands now so any attributes from .DEFAULT have a chance 1671 * to migrate to the node 1672 */ 1673 if (!compatMake && job->flags & JOB_FIRST) { 1674 cmdsOK = Job_CheckCommands(gn, Error); 1675 } else { 1676 cmdsOK = TRUE; 1677 } 1678 1679 /* 1680 * If the -n flag wasn't given, we open up OUR(not the child's) 1681 * temporary file to stuff commands in it. The thing is rd/wr so we don't 1682 * need to reopen it to feed it to the shell. If the -n flag *was* given, 1683 * we just set the file to be stdout. Cute, huh? 1684 */ 1685 if ((gn->type & OP_MAKE) || (!noExecute && !touchFlag)) { 1686 /* 1687 * We're serious here, but if the commands were bogus, we're 1688 * also dead... 1689 */ 1690 if (!cmdsOK) { 1691 DieHorribly(); 1692 } 1693 1694 job->cmdFILE = fopen(tfile, "w+"); 1695 if (job->cmdFILE == NULL) { 1696 Punt("Could not open %s", tfile); 1697 } 1698 (void) fcntl(FILENO(job->cmdFILE), F_SETFD, 1); 1699 /* 1700 * Send the commands to the command file, flush all its buffers then 1701 * rewind and remove the thing. 1702 */ 1703 noExec = FALSE; 1704 1705 /* 1706 * used to be backwards; replace when start doing multiple commands 1707 * per shell. 1708 */ 1709 if (compatMake) { 1710 /* 1711 * Be compatible: If this is the first time for this node, 1712 * verify its commands are ok and open the commands list for 1713 * sequential access by later invocations of JobStart. 1714 * Once that is done, we take the next command off the list 1715 * and print it to the command file. If the command was an 1716 * ellipsis, note that there's nothing more to execute. 1717 */ 1718 if ((job->flags&JOB_FIRST) && (Lst_Open(gn->commands) != SUCCESS)){ 1719 cmdsOK = FALSE; 1720 } else { 1721 LstNode ln = Lst_Next(gn->commands); 1722 1723 if ((ln == NILLNODE) || 1724 JobPrintCommand((ClientData) Lst_Datum(ln), 1725 (ClientData) job)) 1726 { 1727 noExec = TRUE; 1728 Lst_Close(gn->commands); 1729 } 1730 if (noExec && !(job->flags & JOB_FIRST)) { 1731 /* 1732 * If we're not going to execute anything, the job 1733 * is done and we need to close down the various 1734 * file descriptors we've opened for output, then 1735 * call JobDoOutput to catch the final characters or 1736 * send the file to the screen... Note that the i/o streams 1737 * are only open if this isn't the first job. 1738 * Note also that this could not be done in 1739 * Job_CatchChildren b/c it wasn't clear if there were 1740 * more commands to execute or not... 1741 */ 1742 JobClose(job); 1743 } 1744 } 1745 } else { 1746 /* 1747 * We can do all the commands at once. hooray for sanity 1748 */ 1749 numCommands = 0; 1750 Lst_ForEach(gn->commands, JobPrintCommand, (ClientData)job); 1751 1752 /* 1753 * If we didn't print out any commands to the shell script, 1754 * there's not much point in executing the shell, is there? 1755 */ 1756 if (numCommands == 0) { 1757 noExec = TRUE; 1758 } 1759 } 1760 } else if (noExecute) { 1761 /* 1762 * Not executing anything -- just print all the commands to stdout 1763 * in one fell swoop. This will still set up job->tailCmds correctly. 1764 */ 1765 if (lastNode != gn) { 1766 MESSAGE(stdout, gn); 1767 lastNode = gn; 1768 } 1769 job->cmdFILE = stdout; 1770 /* 1771 * Only print the commands if they're ok, but don't die if they're 1772 * not -- just let the user know they're bad and keep going. It 1773 * doesn't do any harm in this case and may do some good. 1774 */ 1775 if (cmdsOK) { 1776 Lst_ForEach(gn->commands, JobPrintCommand, (ClientData)job); 1777 } 1778 /* 1779 * Don't execute the shell, thank you. 1780 */ 1781 noExec = TRUE; 1782 } else { 1783 /* 1784 * Just touch the target and note that no shell should be executed. 1785 * Set cmdFILE to stdout to make life easier. Check the commands, too, 1786 * but don't die if they're no good -- it does no harm to keep working 1787 * up the graph. 1788 */ 1789 job->cmdFILE = stdout; 1790 Job_Touch(gn, job->flags&JOB_SILENT); 1791 noExec = TRUE; 1792 } 1793 1794 /* 1795 * If we're not supposed to execute a shell, don't. 1796 */ 1797 if (noExec) { 1798 /* 1799 * Unlink and close the command file if we opened one 1800 */ 1801 if (job->cmdFILE != stdout) { 1802 (void) eunlink(tfile); 1803 if (job->cmdFILE != NULL) 1804 (void) fclose(job->cmdFILE); 1805 } else { 1806 (void) fflush(stdout); 1807 } 1808 1809 /* 1810 * We only want to work our way up the graph if we aren't here because 1811 * the commands for the job were no good. 1812 */ 1813 if (cmdsOK) { 1814 if (aborting == 0) { 1815 if (job->tailCmds != NILLNODE) { 1816 Lst_ForEachFrom(job->node->commands, job->tailCmds, 1817 JobSaveCommand, 1818 (ClientData)job->node); 1819 } 1820 Make_Update(job->node); 1821 } 1822 free((Address)job); 1823 return(JOB_FINISHED); 1824 } else { 1825 free((Address)job); 1826 return(JOB_ERROR); 1827 } 1828 } else { 1829 (void) fflush(job->cmdFILE); 1830 (void) eunlink(tfile); 1831 } 1832 1833 /* 1834 * Set up the control arguments to the shell. This is based on the flags 1835 * set earlier for this job. 1836 */ 1837 JobMakeArgv(job, argv); 1838 1839 /* 1840 * If we're using pipes to catch output, create the pipe by which we'll 1841 * get the shell's output. If we're using files, print out that we're 1842 * starting a job and then set up its temporary-file name. This is just 1843 * tfile with two extra digits tacked on -- jobno. 1844 */ 1845 if (!compatMake || (job->flags & JOB_FIRST)) { 1846 if (usePipes) { 1847 int fd[2]; 1848 if (pipe(fd) == -1) 1849 Punt("Cannot create pipe: %s", strerror(errno)); 1850 job->inPipe = fd[0]; 1851 job->outPipe = fd[1]; 1852 (void) fcntl(job->inPipe, F_SETFD, 1); 1853 (void) fcntl(job->outPipe, F_SETFD, 1); 1854 } else { 1855 (void) fprintf(stdout, "Remaking `%s'\n", gn->name); 1856 (void) fflush(stdout); 1857 sprintf(job->outFile, "%s%02d", tfile, jobno); 1858 jobno = (jobno + 1) % 100; 1859 job->outFd = open(job->outFile,O_WRONLY|O_CREAT|O_APPEND,0600); 1860 (void) fcntl(job->outFd, F_SETFD, 1); 1861 } 1862 } 1863 1864 #ifdef REMOTE 1865 if (!(gn->type & OP_NOEXPORT) && !(runLocalFirst && nLocal < maxLocal)) { 1866 #ifdef RMT_NO_EXEC 1867 local = !Rmt_Export(shellPath, argv, job); 1868 #else 1869 local = !Rmt_Begin(shellPath, argv, job->node); 1870 #endif /* RMT_NO_EXEC */ 1871 if (!local) { 1872 job->flags |= JOB_REMOTE; 1873 } 1874 } else 1875 #endif 1876 local = TRUE; 1877 1878 if (local && (((nLocal >= maxLocal) && 1879 !(job->flags & JOB_SPECIAL) && 1880 #ifdef REMOTE 1881 (!(gn->type & OP_NOEXPORT) || (maxLocal != 0)) 1882 #else 1883 (maxLocal != 0) 1884 #endif 1885 ))) 1886 { 1887 /* 1888 * The job can only be run locally, but we've hit the limit of 1889 * local concurrency, so put the job on hold until some other job 1890 * finishes. Note that the special jobs(.BEGIN, .INTERRUPT and .END) 1891 * may be run locally even when the local limit has been reached 1892 *(e.g. when maxLocal == 0), though they will be exported if at 1893 * all possible. In addition, any target marked with .NOEXPORT will 1894 * be run locally if maxLocal is 0. 1895 */ 1896 jobFull = TRUE; 1897 1898 if (DEBUG(JOB)) { 1899 (void) fprintf(stdout, "Can only run job locally.\n"); 1900 (void) fflush(stdout); 1901 } 1902 job->flags |= JOB_RESTART; 1903 (void) Lst_AtEnd(stoppedJobs, (ClientData)job); 1904 } else { 1905 if ((nLocal >= maxLocal) && local) { 1906 /* 1907 * If we're running this job locally as a special case(see above), 1908 * at least say the table is full. 1909 */ 1910 jobFull = TRUE; 1911 if (DEBUG(JOB)) { 1912 (void) fprintf(stdout, "Local job queue is full.\n"); 1913 (void) fflush(stdout); 1914 } 1915 } 1916 JobExec(job, argv); 1917 } 1918 return(JOB_RUNNING); 1919 } 1920 1921 static char * 1922 JobOutput(job, cp, endp, msg) 1923 register Job *job; 1924 register char *cp, *endp; 1925 int msg; 1926 { 1927 register char *ecp; 1928 1929 if (commandShell->noPrint) { 1930 ecp = Str_FindSubstring(cp, commandShell->noPrint); 1931 while (ecp != NULL) { 1932 if (cp != ecp) { 1933 *ecp = '\0'; 1934 if (msg && job->node != lastNode) { 1935 MESSAGE(stdout, job->node); 1936 lastNode = job->node; 1937 } 1938 /* 1939 * The only way there wouldn't be a newline after 1940 * this line is if it were the last in the buffer. 1941 * however, since the non-printable comes after it, 1942 * there must be a newline, so we don't print one. 1943 */ 1944 (void) fprintf(stdout, "%s", cp); 1945 (void) fflush(stdout); 1946 } 1947 cp = ecp + commandShell->noPLen; 1948 if (cp != endp) { 1949 /* 1950 * Still more to print, look again after skipping 1951 * the whitespace following the non-printable 1952 * command.... 1953 */ 1954 cp++; 1955 while (*cp == ' ' || *cp == '\t' || *cp == '\n') { 1956 cp++; 1957 } 1958 ecp = Str_FindSubstring(cp, commandShell->noPrint); 1959 } else { 1960 return cp; 1961 } 1962 } 1963 } 1964 return cp; 1965 } 1966 1967 /*- 1968 *----------------------------------------------------------------------- 1969 * JobDoOutput -- 1970 * This function is called at different times depending on 1971 * whether the user has specified that output is to be collected 1972 * via pipes or temporary files. In the former case, we are called 1973 * whenever there is something to read on the pipe. We collect more 1974 * output from the given job and store it in the job's outBuf. If 1975 * this makes up a line, we print it tagged by the job's identifier, 1976 * as necessary. 1977 * If output has been collected in a temporary file, we open the 1978 * file and read it line by line, transfering it to our own 1979 * output channel until the file is empty. At which point we 1980 * remove the temporary file. 1981 * In both cases, however, we keep our figurative eye out for the 1982 * 'noPrint' line for the shell from which the output came. If 1983 * we recognize a line, we don't print it. If the command is not 1984 * alone on the line(the character after it is not \0 or \n), we 1985 * do print whatever follows it. 1986 * 1987 * Results: 1988 * None 1989 * 1990 * Side Effects: 1991 * curPos may be shifted as may the contents of outBuf. 1992 *----------------------------------------------------------------------- 1993 */ 1994 STATIC void 1995 JobDoOutput(job, finish) 1996 register Job *job; /* the job whose output needs printing */ 1997 Boolean finish; /* TRUE if this is the last time we'll be 1998 * called for this job */ 1999 { 2000 Boolean gotNL = FALSE; /* true if got a newline */ 2001 Boolean fbuf; /* true if our buffer filled up */ 2002 register int nr; /* number of bytes read */ 2003 register int i; /* auxiliary index into outBuf */ 2004 register int max; /* limit for i(end of current data) */ 2005 int nRead; /*(Temporary) number of bytes read */ 2006 2007 FILE *oFILE; /* Stream pointer to shell's output file */ 2008 char inLine[132]; 2009 2010 2011 if (usePipes) { 2012 /* 2013 * Read as many bytes as will fit in the buffer. 2014 */ 2015 end_loop: 2016 gotNL = FALSE; 2017 fbuf = FALSE; 2018 2019 nRead = read(job->inPipe, &job->outBuf[job->curPos], 2020 JOB_BUFSIZE - job->curPos); 2021 if (nRead < 0) { 2022 if (DEBUG(JOB)) { 2023 perror("JobDoOutput(piperead)"); 2024 } 2025 nr = 0; 2026 } else { 2027 nr = nRead; 2028 } 2029 2030 /* 2031 * If we hit the end-of-file(the job is dead), we must flush its 2032 * remaining output, so pretend we read a newline if there's any 2033 * output remaining in the buffer. 2034 * Also clear the 'finish' flag so we stop looping. 2035 */ 2036 if ((nr == 0) && (job->curPos != 0)) { 2037 job->outBuf[job->curPos] = '\n'; 2038 nr = 1; 2039 finish = FALSE; 2040 } else if (nr == 0) { 2041 finish = FALSE; 2042 } 2043 2044 /* 2045 * Look for the last newline in the bytes we just got. If there is 2046 * one, break out of the loop with 'i' as its index and gotNL set 2047 * TRUE. 2048 */ 2049 max = job->curPos + nr; 2050 for (i = job->curPos + nr - 1; i >= job->curPos; i--) { 2051 if (job->outBuf[i] == '\n') { 2052 gotNL = TRUE; 2053 break; 2054 } else if (job->outBuf[i] == '\0') { 2055 /* 2056 * Why? 2057 */ 2058 job->outBuf[i] = ' '; 2059 } 2060 } 2061 2062 if (!gotNL) { 2063 job->curPos += nr; 2064 if (job->curPos == JOB_BUFSIZE) { 2065 /* 2066 * If we've run out of buffer space, we have no choice 2067 * but to print the stuff. sigh. 2068 */ 2069 fbuf = TRUE; 2070 i = job->curPos; 2071 } 2072 } 2073 if (gotNL || fbuf) { 2074 /* 2075 * Need to send the output to the screen. Null terminate it 2076 * first, overwriting the newline character if there was one. 2077 * So long as the line isn't one we should filter(according 2078 * to the shell description), we print the line, preceeded 2079 * by a target banner if this target isn't the same as the 2080 * one for which we last printed something. 2081 * The rest of the data in the buffer are then shifted down 2082 * to the start of the buffer and curPos is set accordingly. 2083 */ 2084 job->outBuf[i] = '\0'; 2085 if (i >= job->curPos) { 2086 char *cp; 2087 2088 cp = JobOutput(job, job->outBuf, &job->outBuf[i]); 2089 2090 /* 2091 * There's still more in that thar buffer. This time, though, 2092 * we know there's no newline at the end, so we add one of 2093 * our own free will. 2094 */ 2095 if (*cp != '\0') { 2096 if (job->node != lastNode) { 2097 MESSAGE(stdout, job->node); 2098 lastNode = job->node; 2099 } 2100 (void) fprintf(stdout, "%s%s", cp, gotNL ? "\n" : ""); 2101 (void) fflush(stdout); 2102 } 2103 } 2104 if (i < max - 1) { 2105 /* shift the remaining characters down */ 2106 (void) memcpy(job->outBuf, &job->outBuf[i + 1], max -(i + 1)); 2107 job->curPos = max -(i + 1); 2108 2109 } else { 2110 /* 2111 * We have written everything out, so we just start over 2112 * from the start of the buffer. No copying. No nothing. 2113 */ 2114 job->curPos = 0; 2115 } 2116 } 2117 if (finish) { 2118 /* 2119 * If the finish flag is true, we must loop until we hit 2120 * end-of-file on the pipe. This is guaranteed to happen eventually 2121 * since the other end of the pipe is now closed(we closed it 2122 * explicitly and the child has exited). When we do get an EOF, 2123 * finish will be set FALSE and we'll fall through and out. 2124 */ 2125 goto end_loop; 2126 } 2127 } else { 2128 /* 2129 * We've been called to retrieve the output of the job from the 2130 * temporary file where it's been squirreled away. This consists of 2131 * opening the file, reading the output line by line, being sure not 2132 * to print the noPrint line for the shell we used, then close and 2133 * remove the temporary file. Very simple. 2134 * 2135 * Change to read in blocks and do FindSubString type things as for 2136 * pipes? That would allow for "@echo -n..." 2137 */ 2138 oFILE = fopen(job->outFile, "r"); 2139 if (oFILE != NULL) { 2140 (void) fprintf(stdout, "Results of making %s:\n", job->node->name); 2141 (void) fflush(stdout); 2142 while (fgets(inLine, sizeof(inLine), oFILE) != NULL) { 2143 register char *cp, *endp, *oendp; 2144 2145 cp = inLine; 2146 oendp = endp = inLine + strlen(inLine); 2147 if (endp[-1] == '\n') { 2148 *--endp = '\0'; 2149 } 2150 cp = JobOutput(job, inLine, endp, FALSE); 2151 2152 /* 2153 * There's still more in that thar buffer. This time, though, 2154 * we know there's no newline at the end, so we add one of 2155 * our own free will. 2156 */ 2157 (void) fprintf(stdout, "%s", cp); 2158 (void) fflush(stdout); 2159 if (endp != oendp) { 2160 (void) fprintf(stdout, "\n"); 2161 (void) fflush(stdout); 2162 } 2163 } 2164 (void) fclose(oFILE); 2165 (void) eunlink(job->outFile); 2166 } 2167 } 2168 } 2169 2170 /*- 2171 *----------------------------------------------------------------------- 2172 * Job_CatchChildren -- 2173 * Handle the exit of a child. Called from Make_Make. 2174 * 2175 * Results: 2176 * none. 2177 * 2178 * Side Effects: 2179 * The job descriptor is removed from the list of children. 2180 * 2181 * Notes: 2182 * We do waits, blocking or not, according to the wisdom of our 2183 * caller, until there are no more children to report. For each 2184 * job, call JobFinish to finish things off. This will take care of 2185 * putting jobs on the stoppedJobs queue. 2186 * 2187 *----------------------------------------------------------------------- 2188 */ 2189 void 2190 Job_CatchChildren(block) 2191 Boolean block; /* TRUE if should block on the wait. */ 2192 { 2193 int pid; /* pid of dead child */ 2194 register Job *job; /* job descriptor for dead child */ 2195 LstNode jnode; /* list element for finding job */ 2196 union wait status; /* Exit/termination status */ 2197 2198 /* 2199 * Don't even bother if we know there's no one around. 2200 */ 2201 if (nLocal == 0) { 2202 return; 2203 } 2204 2205 while ((pid = wait3((int *)&status, (block?0:WNOHANG)|WUNTRACED, 2206 (struct rusage *)0)) > 0) 2207 { 2208 if (DEBUG(JOB)) { 2209 (void) fprintf(stdout, "Process %d exited or stopped.\n", pid); 2210 (void) fflush(stdout); 2211 } 2212 2213 2214 jnode = Lst_Find(jobs, (ClientData)&pid, JobCmpPid); 2215 2216 if (jnode == NILLNODE) { 2217 if (WIFSIGNALED(status) && (WTERMSIG(status) == SIGCONT)) { 2218 jnode = Lst_Find(stoppedJobs, (ClientData) &pid, JobCmpPid); 2219 if (jnode == NILLNODE) { 2220 Error("Resumed child(%d) not in table", pid); 2221 continue; 2222 } 2223 job = (Job *)Lst_Datum(jnode); 2224 (void) Lst_Remove(stoppedJobs, jnode); 2225 } else { 2226 Error("Child(%d) not in table?", pid); 2227 continue; 2228 } 2229 } else { 2230 job = (Job *) Lst_Datum(jnode); 2231 (void) Lst_Remove(jobs, jnode); 2232 nJobs -= 1; 2233 if (jobFull && DEBUG(JOB)) { 2234 (void) fprintf(stdout, "Job queue is no longer full.\n"); 2235 (void) fflush(stdout); 2236 } 2237 jobFull = FALSE; 2238 #ifdef REMOTE 2239 if (!(job->flags & JOB_REMOTE)) { 2240 if (DEBUG(JOB)) { 2241 (void) fprintf(stdout, 2242 "Job queue has one fewer local process.\n"); 2243 (void) fflush(stdout); 2244 } 2245 nLocal -= 1; 2246 } 2247 #else 2248 nLocal -= 1; 2249 #endif 2250 } 2251 2252 JobFinish(job, &status); 2253 } 2254 } 2255 2256 /*- 2257 *----------------------------------------------------------------------- 2258 * Job_CatchOutput -- 2259 * Catch the output from our children, if we're using 2260 * pipes do so. Otherwise just block time until we get a 2261 * signal(most likely a SIGCHLD) since there's no point in 2262 * just spinning when there's nothing to do and the reaping 2263 * of a child can wait for a while. 2264 * 2265 * Results: 2266 * None 2267 * 2268 * Side Effects: 2269 * Output is read from pipes if we're piping. 2270 * ----------------------------------------------------------------------- 2271 */ 2272 void 2273 Job_CatchOutput() 2274 { 2275 int nfds; 2276 struct timeval timeout; 2277 fd_set readfds; 2278 register LstNode ln; 2279 register Job *job; 2280 #ifdef RMT_WILL_WATCH 2281 int pnJobs; /* Previous nJobs */ 2282 #endif 2283 2284 (void) fflush(stdout); 2285 #ifdef RMT_WILL_WATCH 2286 pnJobs = nJobs; 2287 2288 /* 2289 * It is possible for us to be called with nJobs equal to 0. This happens 2290 * if all the jobs finish and a job that is stopped cannot be run 2291 * locally (eg if maxLocal is 0) and cannot be exported. The job will 2292 * be placed back on the stoppedJobs queue, Job_Empty() will return false, 2293 * Make_Run will call us again when there's nothing for which to wait. 2294 * nJobs never changes, so we loop forever. Hence the check. It could 2295 * be argued that we should sleep for a bit so as not to swamp the 2296 * exportation system with requests. Perhaps we should. 2297 * 2298 * NOTE: IT IS THE RESPONSIBILITY OF Rmt_Wait TO CALL Job_CatchChildren 2299 * IN A TIMELY FASHION TO CATCH ANY LOCALLY RUNNING JOBS THAT EXIT. 2300 * It may use the variable nLocal to determine if it needs to call 2301 * Job_CatchChildren(if nLocal is 0, there's nothing for which to 2302 * wait...) 2303 */ 2304 while (nJobs != 0 && pnJobs == nJobs) { 2305 Rmt_Wait(); 2306 } 2307 #else 2308 if (usePipes) { 2309 readfds = outputs; 2310 timeout.tv_sec = SEL_SEC; 2311 timeout.tv_usec = SEL_USEC; 2312 2313 if ((nfds = select(FD_SETSIZE, &readfds, (fd_set *) 0, 2314 (fd_set *) 0, &timeout)) <= 0) 2315 return; 2316 else { 2317 if (Lst_Open(jobs) == FAILURE) { 2318 Punt("Cannot open job table"); 2319 } 2320 while (nfds && (ln = Lst_Next(jobs)) != NILLNODE) { 2321 job = (Job *) Lst_Datum(ln); 2322 if (FD_ISSET(job->inPipe, &readfds)) { 2323 JobDoOutput(job, FALSE); 2324 nfds -= 1; 2325 } 2326 } 2327 Lst_Close(jobs); 2328 } 2329 } 2330 #endif /* RMT_WILL_WATCH */ 2331 } 2332 2333 /*- 2334 *----------------------------------------------------------------------- 2335 * Job_Make -- 2336 * Start the creation of a target. Basically a front-end for 2337 * JobStart used by the Make module. 2338 * 2339 * Results: 2340 * None. 2341 * 2342 * Side Effects: 2343 * Another job is started. 2344 * 2345 *----------------------------------------------------------------------- 2346 */ 2347 void 2348 Job_Make(gn) 2349 GNode *gn; 2350 { 2351 (void) JobStart(gn, 0, NULL); 2352 } 2353 2354 /*- 2355 *----------------------------------------------------------------------- 2356 * Job_Init -- 2357 * Initialize the process module 2358 * 2359 * Results: 2360 * none 2361 * 2362 * Side Effects: 2363 * lists and counters are initialized 2364 *----------------------------------------------------------------------- 2365 */ 2366 void 2367 Job_Init(maxproc, maxlocal) 2368 int maxproc; /* the greatest number of jobs which may be 2369 * running at one time */ 2370 int maxlocal; /* the greatest number of local jobs which may 2371 * be running at once. */ 2372 { 2373 GNode *begin; /* node for commands to do at the very start */ 2374 2375 (void) sprintf(tfile, "/tmp/make%05d", getpid()); 2376 2377 jobs = Lst_Init(FALSE); 2378 stoppedJobs = Lst_Init(FALSE); 2379 maxJobs = maxproc; 2380 maxLocal = maxlocal; 2381 nJobs = 0; 2382 nLocal = 0; 2383 jobFull = FALSE; 2384 2385 aborting = 0; 2386 errors = 0; 2387 2388 lastNode = NILGNODE; 2389 2390 if (maxJobs == 1 2391 #ifdef REMOTE 2392 || noMessages 2393 #endif 2394 ) { 2395 /* 2396 * If only one job can run at a time, there's no need for a banner, 2397 * no is there? 2398 */ 2399 targFmt = ""; 2400 } else { 2401 targFmt = TARG_FMT; 2402 } 2403 2404 if (shellPath == NULL) { 2405 /* 2406 * The user didn't specify a shell to use, so we are using the 2407 * default one... Both the absolute path and the last component 2408 * must be set. The last component is taken from the 'name' field 2409 * of the default shell description pointed-to by commandShell. 2410 * All default shells are located in _PATH_DEFSHELLDIR. 2411 */ 2412 shellName = commandShell->name; 2413 shellPath = str_concat(_PATH_DEFSHELLDIR, shellName, STR_ADDSLASH); 2414 } 2415 2416 if (commandShell->exit == NULL) { 2417 commandShell->exit = ""; 2418 } 2419 if (commandShell->echo == NULL) { 2420 commandShell->echo = ""; 2421 } 2422 2423 /* 2424 * Catch the four signals that POSIX specifies if they aren't ignored. 2425 * JobPassSig will take care of calling JobInterrupt if appropriate. 2426 */ 2427 if (signal(SIGINT, SIG_IGN) != SIG_IGN) { 2428 (void) signal(SIGINT, JobPassSig); 2429 } 2430 if (signal(SIGHUP, SIG_IGN) != SIG_IGN) { 2431 (void) signal(SIGHUP, JobPassSig); 2432 } 2433 if (signal(SIGQUIT, SIG_IGN) != SIG_IGN) { 2434 (void) signal(SIGQUIT, JobPassSig); 2435 } 2436 if (signal(SIGTERM, SIG_IGN) != SIG_IGN) { 2437 (void) signal(SIGTERM, JobPassSig); 2438 } 2439 /* 2440 * There are additional signals that need to be caught and passed if 2441 * either the export system wants to be told directly of signals or if 2442 * we're giving each job its own process group(since then it won't get 2443 * signals from the terminal driver as we own the terminal) 2444 */ 2445 #if defined(RMT_WANTS_SIGNALS) || defined(USE_PGRP) 2446 if (signal(SIGTSTP, SIG_IGN) != SIG_IGN) { 2447 (void) signal(SIGTSTP, JobPassSig); 2448 } 2449 if (signal(SIGTTOU, SIG_IGN) != SIG_IGN) { 2450 (void) signal(SIGTTOU, JobPassSig); 2451 } 2452 if (signal(SIGTTIN, SIG_IGN) != SIG_IGN) { 2453 (void) signal(SIGTTIN, JobPassSig); 2454 } 2455 if (signal(SIGWINCH, SIG_IGN) != SIG_IGN) { 2456 (void) signal(SIGWINCH, JobPassSig); 2457 } 2458 #endif 2459 2460 begin = Targ_FindNode(".BEGIN", TARG_NOCREATE); 2461 2462 if (begin != NILGNODE) { 2463 JobStart(begin, JOB_SPECIAL, (Job *)0); 2464 while (nJobs) { 2465 Job_CatchOutput(); 2466 #ifndef RMT_WILL_WATCH 2467 Job_CatchChildren(!usePipes); 2468 #endif /* RMT_WILL_WATCH */ 2469 } 2470 } 2471 postCommands = Targ_FindNode(".END", TARG_CREATE); 2472 } 2473 2474 /*- 2475 *----------------------------------------------------------------------- 2476 * Job_Full -- 2477 * See if the job table is full. It is considered full if it is OR 2478 * if we are in the process of aborting OR if we have 2479 * reached/exceeded our local quota. This prevents any more jobs 2480 * from starting up. 2481 * 2482 * Results: 2483 * TRUE if the job table is full, FALSE otherwise 2484 * Side Effects: 2485 * None. 2486 *----------------------------------------------------------------------- 2487 */ 2488 Boolean 2489 Job_Full() 2490 { 2491 return(aborting || jobFull); 2492 } 2493 2494 /*- 2495 *----------------------------------------------------------------------- 2496 * Job_Empty -- 2497 * See if the job table is empty. Because the local concurrency may 2498 * be set to 0, it is possible for the job table to become empty, 2499 * while the list of stoppedJobs remains non-empty. In such a case, 2500 * we want to restart as many jobs as we can. 2501 * 2502 * Results: 2503 * TRUE if it is. FALSE if it ain't. 2504 * 2505 * Side Effects: 2506 * None. 2507 * 2508 * ----------------------------------------------------------------------- 2509 */ 2510 Boolean 2511 Job_Empty() 2512 { 2513 if (nJobs == 0) { 2514 if (!Lst_IsEmpty(stoppedJobs) && !aborting) { 2515 /* 2516 * The job table is obviously not full if it has no jobs in 2517 * it...Try and restart the stopped jobs. 2518 */ 2519 jobFull = FALSE; 2520 JobRestartJobs(); 2521 return(FALSE); 2522 } else { 2523 return(TRUE); 2524 } 2525 } else { 2526 return(FALSE); 2527 } 2528 } 2529 2530 /*- 2531 *----------------------------------------------------------------------- 2532 * JobMatchShell -- 2533 * Find a matching shell in 'shells' given its final component. 2534 * 2535 * Results: 2536 * A pointer to the Shell structure. 2537 * 2538 * Side Effects: 2539 * None. 2540 * 2541 *----------------------------------------------------------------------- 2542 */ 2543 static Shell * 2544 JobMatchShell(name) 2545 char *name; /* Final component of shell path */ 2546 { 2547 register Shell *sh; /* Pointer into shells table */ 2548 Shell *match; /* Longest-matching shell */ 2549 register char *cp1, 2550 *cp2; 2551 char *eoname; 2552 2553 eoname = name + strlen(name); 2554 2555 match = NULL; 2556 2557 for (sh = shells; sh->name != NULL; sh++) { 2558 for (cp1 = eoname - strlen(sh->name), cp2 = sh->name; 2559 *cp1 != '\0' && *cp1 == *cp2; 2560 cp1++, cp2++) { 2561 continue; 2562 } 2563 if (*cp1 != *cp2) { 2564 continue; 2565 } else if (match == NULL || strlen(match->name) < strlen(sh->name)) { 2566 match = sh; 2567 } 2568 } 2569 return(match == NULL ? sh : match); 2570 } 2571 2572 /*- 2573 *----------------------------------------------------------------------- 2574 * Job_ParseShell -- 2575 * Parse a shell specification and set up commandShell, shellPath 2576 * and shellName appropriately. 2577 * 2578 * Results: 2579 * FAILURE if the specification was incorrect. 2580 * 2581 * Side Effects: 2582 * commandShell points to a Shell structure (either predefined or 2583 * created from the shell spec), shellPath is the full path of the 2584 * shell described by commandShell, while shellName is just the 2585 * final component of shellPath. 2586 * 2587 * Notes: 2588 * A shell specification consists of a .SHELL target, with dependency 2589 * operator, followed by a series of blank-separated words. Double 2590 * quotes can be used to use blanks in words. A backslash escapes 2591 * anything (most notably a double-quote and a space) and 2592 * provides the functionality it does in C. Each word consists of 2593 * keyword and value separated by an equal sign. There should be no 2594 * unnecessary spaces in the word. The keywords are as follows: 2595 * name Name of shell. 2596 * path Location of shell. Overrides "name" if given 2597 * quiet Command to turn off echoing. 2598 * echo Command to turn echoing on 2599 * filter Result of turning off echoing that shouldn't be 2600 * printed. 2601 * echoFlag Flag to turn echoing on at the start 2602 * errFlag Flag to turn error checking on at the start 2603 * hasErrCtl True if shell has error checking control 2604 * check Command to turn on error checking if hasErrCtl 2605 * is TRUE or template of command to echo a command 2606 * for which error checking is off if hasErrCtl is 2607 * FALSE. 2608 * ignore Command to turn off error checking if hasErrCtl 2609 * is TRUE or template of command to execute a 2610 * command so as to ignore any errors it returns if 2611 * hasErrCtl is FALSE. 2612 * 2613 *----------------------------------------------------------------------- 2614 */ 2615 ReturnStatus 2616 Job_ParseShell(line) 2617 char *line; /* The shell spec */ 2618 { 2619 char **words; 2620 int wordCount; 2621 register char **argv; 2622 register int argc; 2623 char *path; 2624 Shell newShell; 2625 Boolean fullSpec = FALSE; 2626 2627 while (isspace(*line)) { 2628 line++; 2629 } 2630 words = brk_string(line, &wordCount, TRUE); 2631 2632 memset((Address)&newShell, 0, sizeof(newShell)); 2633 2634 /* 2635 * Parse the specification by keyword 2636 */ 2637 for (path = NULL, argc = wordCount - 1, argv = words + 1; 2638 argc != 0; 2639 argc--, argv++) { 2640 if (strncmp(*argv, "path=", 5) == 0) { 2641 path = &argv[0][5]; 2642 } else if (strncmp(*argv, "name=", 5) == 0) { 2643 newShell.name = &argv[0][5]; 2644 } else { 2645 if (strncmp(*argv, "quiet=", 6) == 0) { 2646 newShell.echoOff = &argv[0][6]; 2647 } else if (strncmp(*argv, "echo=", 5) == 0) { 2648 newShell.echoOn = &argv[0][5]; 2649 } else if (strncmp(*argv, "filter=", 7) == 0) { 2650 newShell.noPrint = &argv[0][7]; 2651 newShell.noPLen = strlen(newShell.noPrint); 2652 } else if (strncmp(*argv, "echoFlag=", 9) == 0) { 2653 newShell.echo = &argv[0][9]; 2654 } else if (strncmp(*argv, "errFlag=", 8) == 0) { 2655 newShell.exit = &argv[0][8]; 2656 } else if (strncmp(*argv, "hasErrCtl=", 10) == 0) { 2657 char c = argv[0][10]; 2658 newShell.hasErrCtl = !((c != 'Y') && (c != 'y') && 2659 (c != 'T') && (c != 't')); 2660 } else if (strncmp(*argv, "check=", 6) == 0) { 2661 newShell.errCheck = &argv[0][6]; 2662 } else if (strncmp(*argv, "ignore=", 7) == 0) { 2663 newShell.ignErr = &argv[0][7]; 2664 } else { 2665 Parse_Error(PARSE_FATAL, "Unknown keyword \"%s\"", 2666 *argv); 2667 return(FAILURE); 2668 } 2669 fullSpec = TRUE; 2670 } 2671 } 2672 2673 if (path == NULL) { 2674 /* 2675 * If no path was given, the user wants one of the pre-defined shells, 2676 * yes? So we find the one s/he wants with the help of JobMatchShell 2677 * and set things up the right way. shellPath will be set up by 2678 * Job_Init. 2679 */ 2680 if (newShell.name == NULL) { 2681 Parse_Error(PARSE_FATAL, "Neither path nor name specified"); 2682 return(FAILURE); 2683 } else { 2684 commandShell = JobMatchShell(newShell.name); 2685 shellName = newShell.name; 2686 } 2687 } else { 2688 /* 2689 * The user provided a path. If s/he gave nothing else(fullSpec is 2690 * FALSE), try and find a matching shell in the ones we know of. 2691 * Else we just take the specification at its word and copy it 2692 * to a new location. In either case, we need to record the 2693 * path the user gave for the shell. 2694 */ 2695 shellPath = path; 2696 path = strrchr(path, '/'); 2697 if (path == NULL) { 2698 path = shellPath; 2699 } else { 2700 path += 1; 2701 } 2702 if (newShell.name != NULL) { 2703 shellName = newShell.name; 2704 } else { 2705 shellName = path; 2706 } 2707 if (!fullSpec) { 2708 commandShell = JobMatchShell(shellName); 2709 } else { 2710 commandShell = (Shell *) emalloc(sizeof(Shell)); 2711 *commandShell = newShell; 2712 } 2713 } 2714 2715 if (commandShell->echoOn && commandShell->echoOff) { 2716 commandShell->hasEchoCtl = TRUE; 2717 } 2718 2719 if (!commandShell->hasErrCtl) { 2720 if (commandShell->errCheck == NULL) { 2721 commandShell->errCheck = ""; 2722 } 2723 if (commandShell->ignErr == NULL) { 2724 commandShell->ignErr = "%s\n"; 2725 } 2726 } 2727 2728 /* 2729 * Do not free up the words themselves, since they might be in use by the 2730 * shell specification... 2731 */ 2732 free(words); 2733 return SUCCESS; 2734 } 2735 2736 /*- 2737 *----------------------------------------------------------------------- 2738 * JobInterrupt -- 2739 * Handle the receipt of an interrupt. 2740 * 2741 * Results: 2742 * None 2743 * 2744 * Side Effects: 2745 * All children are killed. Another job will be started if the 2746 * .INTERRUPT target was given. 2747 *----------------------------------------------------------------------- 2748 */ 2749 static void 2750 JobInterrupt(runINTERRUPT, signo) 2751 int runINTERRUPT; /* Non-zero if commands for the .INTERRUPT 2752 * target should be executed */ 2753 int signo; /* signal received */ 2754 { 2755 LstNode ln; /* element in job table */ 2756 Job *job; /* job descriptor in that element */ 2757 GNode *interrupt; /* the node describing the .INTERRUPT target */ 2758 2759 aborting = ABORT_INTERRUPT; 2760 2761 (void) Lst_Open(jobs); 2762 while ((ln = Lst_Next(jobs)) != NILLNODE) { 2763 job = (Job *) Lst_Datum(ln); 2764 2765 if (!Targ_Precious(job->node)) { 2766 char *file = (job->node->path == NULL ? 2767 job->node->name : 2768 job->node->path); 2769 if (!noExecute && eunlink(file) != -1) { 2770 Error("*** %s removed", file); 2771 } 2772 } 2773 #ifdef RMT_WANTS_SIGNALS 2774 if (job->flags & JOB_REMOTE) { 2775 /* 2776 * If job is remote, let the Rmt module do the killing. 2777 */ 2778 if (!Rmt_Signal(job, signo)) { 2779 /* 2780 * If couldn't kill the thing, finish it out now with an 2781 * error code, since no exit report will come in likely. 2782 */ 2783 union wait status; 2784 2785 status.w_status = 0; 2786 status.w_retcode = 1; 2787 JobFinish(job, &status); 2788 } 2789 } else if (job->pid) { 2790 KILL(job->pid, signo); 2791 } 2792 #else 2793 if (job->pid) { 2794 if (DEBUG(JOB)) { 2795 (void) fprintf(stdout, 2796 "JobInterrupt passing signal to child %d.\n", 2797 job->pid); 2798 (void) fflush(stdout); 2799 } 2800 KILL(job->pid, signo); 2801 } 2802 #endif /* RMT_WANTS_SIGNALS */ 2803 } 2804 2805 #ifdef REMOTE 2806 (void)Lst_Open(stoppedJobs); 2807 while ((ln = Lst_Next(stoppedJobs)) != NILLNODE) { 2808 job = (Job *) Lst_Datum(ln); 2809 2810 if (job->flags & JOB_RESTART) { 2811 if (DEBUG(JOB)) { 2812 (void) fprintf(stdout, "%s%s", 2813 "JobInterrupt skipping job on stopped queue", 2814 "-- it was waiting to be restarted.\n"); 2815 (void) fflush(stdout); 2816 } 2817 continue; 2818 } 2819 if (!Targ_Precious(job->node)) { 2820 char *file = (job->node->path == NULL ? 2821 job->node->name : 2822 job->node->path); 2823 if (eunlink(file) == 0) { 2824 Error("*** %s removed", file); 2825 } 2826 } 2827 /* 2828 * Resume the thing so it will take the signal. 2829 */ 2830 if (DEBUG(JOB)) { 2831 (void) fprintf(stdout, 2832 "JobInterrupt passing CONT to stopped child %d.\n", 2833 job->pid); 2834 (void) fflush(stdout); 2835 } 2836 KILL(job->pid, SIGCONT); 2837 #ifdef RMT_WANTS_SIGNALS 2838 if (job->flags & JOB_REMOTE) { 2839 /* 2840 * If job is remote, let the Rmt module do the killing. 2841 */ 2842 if (!Rmt_Signal(job, SIGINT)) { 2843 /* 2844 * If couldn't kill the thing, finish it out now with an 2845 * error code, since no exit report will come in likely. 2846 */ 2847 union wait status; 2848 status.w_status = 0; 2849 status.w_retcode = 1; 2850 JobFinish(job, &status); 2851 } 2852 } else if (job->pid) { 2853 if (DEBUG(JOB)) { 2854 (void) fprintf(stdout, 2855 "JobInterrupt passing interrupt to stopped child %d.\n", 2856 job->pid); 2857 (void) fflush(stdout); 2858 } 2859 KILL(job->pid, SIGINT); 2860 } 2861 #endif /* RMT_WANTS_SIGNALS */ 2862 } 2863 #endif 2864 Lst_Close(stoppedJobs); 2865 2866 if (runINTERRUPT && !touchFlag) { 2867 interrupt = Targ_FindNode(".INTERRUPT", TARG_NOCREATE); 2868 if (interrupt != NILGNODE) { 2869 ignoreErrors = FALSE; 2870 2871 JobStart(interrupt, JOB_IGNDOTS, (Job *)0); 2872 while (nJobs) { 2873 Job_CatchOutput(); 2874 #ifndef RMT_WILL_WATCH 2875 Job_CatchChildren(!usePipes); 2876 #endif /* RMT_WILL_WATCH */ 2877 } 2878 } 2879 } 2880 (void) eunlink(tfile); 2881 exit(signo); 2882 } 2883 2884 /* 2885 *----------------------------------------------------------------------- 2886 * Job_End -- 2887 * Do final processing such as the running of the commands 2888 * attached to the .END target. 2889 * 2890 * Results: 2891 * Number of errors reported. 2892 * 2893 * Side Effects: 2894 * The process' temporary file(tfile) is removed if it still 2895 * existed. 2896 *----------------------------------------------------------------------- 2897 */ 2898 int 2899 Job_End() 2900 { 2901 if (postCommands != NILGNODE && !Lst_IsEmpty(postCommands->commands)) { 2902 if (errors) { 2903 Error("Errors reported so .END ignored"); 2904 } else { 2905 JobStart(postCommands, JOB_SPECIAL | JOB_IGNDOTS, NULL); 2906 2907 while (nJobs) { 2908 Job_CatchOutput(); 2909 #ifndef RMT_WILL_WATCH 2910 Job_CatchChildren(!usePipes); 2911 #endif /* RMT_WILL_WATCH */ 2912 } 2913 } 2914 } 2915 (void) eunlink(tfile); 2916 return(errors); 2917 } 2918 2919 /*- 2920 *----------------------------------------------------------------------- 2921 * Job_Wait -- 2922 * Waits for all running jobs to finish and returns. Sets 'aborting' 2923 * to ABORT_WAIT to prevent other jobs from starting. 2924 * 2925 * Results: 2926 * None. 2927 * 2928 * Side Effects: 2929 * Currently running jobs finish. 2930 * 2931 *----------------------------------------------------------------------- 2932 */ 2933 void 2934 Job_Wait() 2935 { 2936 aborting = ABORT_WAIT; 2937 while (nJobs != 0) { 2938 Job_CatchOutput(); 2939 #ifndef RMT_WILL_WATCH 2940 Job_CatchChildren(!usePipes); 2941 #endif /* RMT_WILL_WATCH */ 2942 } 2943 aborting = 0; 2944 } 2945 2946 /*- 2947 *----------------------------------------------------------------------- 2948 * Job_AbortAll -- 2949 * Abort all currently running jobs without handling output or anything. 2950 * This function is to be called only in the event of a major 2951 * error. Most definitely NOT to be called from JobInterrupt. 2952 * 2953 * Results: 2954 * None 2955 * 2956 * Side Effects: 2957 * All children are killed, not just the firstborn 2958 *----------------------------------------------------------------------- 2959 */ 2960 void 2961 Job_AbortAll() 2962 { 2963 LstNode ln; /* element in job table */ 2964 Job *job; /* the job descriptor in that element */ 2965 int foo; 2966 2967 aborting = ABORT_ERROR; 2968 2969 if (nJobs) { 2970 2971 (void) Lst_Open(jobs); 2972 while ((ln = Lst_Next(jobs)) != NILLNODE) { 2973 job = (Job *) Lst_Datum(ln); 2974 2975 /* 2976 * kill the child process with increasingly drastic signals to make 2977 * darn sure it's dead. 2978 */ 2979 #ifdef RMT_WANTS_SIGNALS 2980 if (job->flags & JOB_REMOTE) { 2981 Rmt_Signal(job, SIGINT); 2982 Rmt_Signal(job, SIGKILL); 2983 } else { 2984 KILL(job->pid, SIGINT); 2985 KILL(job->pid, SIGKILL); 2986 } 2987 #else 2988 KILL(job->pid, SIGINT); 2989 KILL(job->pid, SIGKILL); 2990 #endif /* RMT_WANTS_SIGNALS */ 2991 } 2992 } 2993 2994 /* 2995 * Catch as many children as want to report in at first, then give up 2996 */ 2997 while (wait3(&foo, WNOHANG, (struct rusage *)0) > 0) 2998 continue; 2999 (void) eunlink(tfile); 3000 } 3001 3002 #ifdef REMOTE 3003 /*- 3004 *----------------------------------------------------------------------- 3005 * JobFlagForMigration -- 3006 * Handle the eviction of a child. Called from RmtStatusChange. 3007 * Flags the child as remigratable and then suspends it. 3008 * 3009 * Results: 3010 * none. 3011 * 3012 * Side Effects: 3013 * The job descriptor is flagged for remigration. 3014 * 3015 *----------------------------------------------------------------------- 3016 */ 3017 void 3018 JobFlagForMigration(hostID) 3019 int hostID; /* ID of host we used, for matching children. */ 3020 { 3021 register Job *job; /* job descriptor for dead child */ 3022 LstNode jnode; /* list element for finding job */ 3023 3024 if (DEBUG(JOB)) { 3025 (void) fprintf(stdout, "JobFlagForMigration(%d) called.\n", hostID); 3026 (void) fflush(stdout); 3027 } 3028 jnode = Lst_Find(jobs, (ClientData)hostID, JobCmpRmtID); 3029 3030 if (jnode == NILLNODE) { 3031 jnode = Lst_Find(stoppedJobs, (ClientData)hostID, JobCmpRmtID); 3032 if (jnode == NILLNODE) { 3033 if (DEBUG(JOB)) { 3034 Error("Evicting host(%d) not in table", hostID); 3035 } 3036 return; 3037 } 3038 } 3039 job = (Job *) Lst_Datum(jnode); 3040 3041 if (DEBUG(JOB)) { 3042 (void) fprintf(stdout, 3043 "JobFlagForMigration(%d) found job '%s'.\n", hostID, 3044 job->node->name); 3045 (void) fflush(stdout); 3046 } 3047 3048 KILL(job->pid, SIGSTOP); 3049 3050 job->flags |= JOB_REMIGRATE; 3051 } 3052 3053 #endif 3054 3055 /*- 3057 *----------------------------------------------------------------------- 3058 * JobRestartJobs -- 3059 * Tries to restart stopped jobs if there are slots available. 3060 * Note that this tries to restart them regardless of pending errors. 3061 * It's not good to leave stopped jobs lying around! 3062 * 3063 * Results: 3064 * None. 3065 * 3066 * Side Effects: 3067 * Resumes(and possibly migrates) jobs. 3068 * 3069 *----------------------------------------------------------------------- 3070 */ 3071 static void 3072 JobRestartJobs() 3073 { 3074 while (!jobFull && !Lst_IsEmpty(stoppedJobs)) { 3075 if (DEBUG(JOB)) { 3076 (void) fprintf(stdout, 3077 "Job queue is not full. Restarting a stopped job.\n"); 3078 (void) fflush(stdout); 3079 } 3080 JobRestart((Job *)Lst_DeQueue(stoppedJobs)); 3081 } 3082 } 3083
Indexes created Wed Sep 24 15:09:51 GMT 2025