File: [local] / src / usr.bin / make / job.c (download)
Revision 1.101, Fri Nov 2 17:27:24 2007 UTC (16 years, 7 months ago) by espie
Branch: MAIN
Changes since 1.100: +269 -756 lines
Work done at p2k7.
This is a really big step towards getting parallel make to work.
Note that this is not yet complete. There are still a few `details' to
fix before this works 100%. Specifically: sequential make (compat) and
parallel make don't use the same engine, and the parallel engine still
has a few limitations. For instance, some known issues:
- parallel make does not deal with .phony targets correctly all the time.
- some errors are deadly in parallel make mode.
- parallel make NEEDS way more sturdy correspondance of file system paths
and target names, since it often needs to match dependencies to targets
before the corresponding files exist.
- some local variables like $* get set in a bogus way in some cases.
- suffix handling has issues, especially related to the NULL suffix.
So, if you find stuff that does NOT yet work with parallel make, don't go
blindly try to fix the Makefile. It's very likely you might have stumbled
into a make bug. (unless you really, really, understand Makefiles, DON'T
GO CHANGING THEM YET).
Tested by lots of people, thanks go to miod@, and robert@ among other people.
Quick summary of what this does:
- remove `saving commands' extension (it's not really usable, nor used)
- move compat job runner and parallel interrupt handling into engine.c
- tweak the code so that both compat and parallel mode use the same job runner
and the same interrupt handling. Remove the other one.
- optimize job runner so that, in parallel mode, the last command does not
fork if we can avoid it (as it's already running in a sub shell).
- scrape all the code that dealt with creating shell scripts from commands.
- scrape all the code that dealt with recognizing special sequences in
command output to print/not print output.
- fix the parallel job pipe to not keep around file descriptors that are not
needed.
- replace the parallel job buffering with a nicer one, that deals with
non-blocking descriptors to try to agregate as much output from one job in
one go (greed) to unconfuse the users.
- create two pipes per job, so that stdout and stderr stay separate.
- make job token printing a debug option.
- always use the parallel job-runner to `execute' commands, even if we just
print them out.
- store list of errors encountered during parallel make running, and print them
on exit, so that we know what went wrong.
- add a dirty hack to targ.c to deal with paths produced by gccmakedep.
|
/* $OpenPackages$ */
/* $OpenBSD: job.c,v 1.101 2007/11/02 17:27:24 espie Exp $ */
/* $NetBSD: job.c,v 1.16 1996/11/06 17:59:08 christos Exp $ */
/*
* Copyright (c) 1988, 1989, 1990 The Regents of the University of California.
* Copyright (c) 1988, 1989 by Adam de Boor
* Copyright (c) 1989 by Berkeley Softworks
* All rights reserved.
*
* This code is derived from software contributed to Berkeley by
* Adam de Boor.
*
* Redistribution and use in source and binary forms, with or without
* modification, are permitted provided that the following conditions
* are met:
* 1. Redistributions of source code must retain the above copyright
* notice, this list of conditions and the following disclaimer.
* 2. Redistributions in binary form must reproduce the above copyright
* notice, this list of conditions and the following disclaimer in the
* documentation and/or other materials provided with the distribution.
* 3. Neither the name of the University nor the names of its contributors
* may be used to endorse or promote products derived from this software
* without specific prior written permission.
*
* THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND
* ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
* IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
* ARE DISCLAIMED. IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE
* FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
* DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
* OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
* HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
* LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
* OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
* SUCH DAMAGE.
*/
/*-
* job.c --
* handle the creation etc. of our child processes.
*
* Interface:
* Job_Make Start the creation of the given target.
*
* Job_CatchChildren Check for and handle the termination of any
* children. This must be called reasonably
* frequently to keep the whole make going at
* a decent clip, since job table entries aren't
* removed until their process is caught this way.
*
* Job_CatchOutput Print any output our children have produced.
* Should also be called fairly frequently to
* keep the user informed of what's going on.
* If no output is waiting, it will block for
* a time given by the SEL_* constants, below,
* or until output is ready.
*
* Job_Init Called to initialize this module. in addition,
* any commands attached to the .BEGIN target
* are executed before this function returns.
* Hence, the makefile must have been parsed
* before this function is called.
*
* Job_End Cleanup any memory used.
*
* Job_Full Return true if the job table is filled.
*
* Job_Empty Return true if the job table is completely
* empty.
*
* Job_Finish Perform any final processing which needs doing.
* This includes the execution of any commands
* which have been/were attached to the .END
* target. It should only be called when the
* job table is empty.
*
* Job_AbortAll Abort all currently running jobs. It doesn't
* handle output or do anything for the jobs,
* just kills them. It should only be called in
* an emergency, as it were.
*
* Job_Wait Wait for all currently-running jobs to finish.
*/
#include <sys/types.h>
#include <sys/wait.h>
#include <ctype.h>
#include <errno.h>
#include <fcntl.h>
#include <signal.h>
#include <stdarg.h>
#include <stdio.h>
#include <stdlib.h>
#include <string.h>
#include <unistd.h>
#include "config.h"
#include "defines.h"
#include "job.h"
#include "engine.h"
#include "pathnames.h"
#include "var.h"
#include "targ.h"
#include "error.h"
#include "lst.h"
#include "extern.h"
#include "gnode.h"
#include "memory.h"
#include "make.h"
/*
* The SEL_ constants determine the maximum amount of time spent in select
* before coming out to see if a child has finished. SEL_SEC is the number of
* seconds and SEL_USEC is the number of micro-seconds
*/
#define SEL_SEC 0
#define SEL_USEC 500000
/*-
* Job Table definitions.
*
* Each job has several things associated with it:
* 1) The process id of the child shell
* 2) The graph node describing the target being made by this job
* 3) An FILE* for writing out the commands. This is only
* used before the job is actually started.
* 4) Things used for handling the shell's output.
* the output is being caught via a pipe and
* the descriptors of our pipe, an array in which output is line
* buffered and the current position in that buffer are all
* maintained for each job.
* 5) A word of flags which determine how the module handles errors,
* echoing, etc. for the job
*
* The job "table" is kept as a linked Lst in 'jobs', with the number of
* active jobs maintained in the 'nJobs' variable. At no time will this
* exceed the value of 'maxJobs', initialized by the Job_Init function.
*
* When a job is finished, the Make_Update function is called on each of the
* parents of the node which was just remade. This takes care of the upward
* traversal of the dependency graph.
*/
#define JOB_BUFSIZE 1024
struct job_pipe {
int fd;
char buffer[JOB_BUFSIZE];
size_t pos;
};
typedef struct Job_ {
pid_t pid; /* The child's process ID */
GNode *node; /* The target the child is making */
short flags; /* Flags to control treatment of job */
#define JOB_IGNERR 0x001 /* Ignore non-zero exits */
#define JOB_SILENT 0x002 /* no output */
#define JOB_SPECIAL 0x004 /* Target is a special one. */
#define JOB_RESTART 0x080 /* Job needs to be completely restarted */
#define JOB_RESUME 0x100 /* Job needs to be resumed b/c it stopped,
* for some reason */
#define JOB_CONTINUING 0x200 /* We are in the process of resuming this job.
* Used to avoid infinite recursion between
* JobFinish and JobRestart */
struct job_pipe in[2];
} Job;
static int aborting = 0; /* why is the make aborting? */
#define ABORT_ERROR 1 /* Because of an error */
#define ABORT_INTERRUPT 2 /* Because it was interrupted */
#define ABORT_WAIT 3 /* Waiting for jobs to finish */
static int maxJobs; /* The most children we can run at once */
static int nJobs; /* The number of children currently running */
static LIST runningJobs; /* The structures that describe them */
static bool jobFull; /* Flag to tell when the job table is full. It
* is set true when nJobs equals maxJobs */
static fd_set *outputsp; /* Set of descriptors of pipes connected to
* the output channels of children */
static int outputsn;
static GNode *lastNode; /* The node for which output was most recently
* produced. */
/*
* When JobStart attempts to run a job but isn't allowed to,
* the job is placed on the queuedJobs queue to be run
* when the next job finishes.
*/
static LIST stoppedJobs;
static LIST queuedJobs;
static LIST errorsList;
static int errors;
struct error_info {
int status;
char *name;
};
#if defined(USE_PGRP) && defined(SYSV)
# define KILL(pid, sig) killpg(-(pid), (sig))
#else
# if defined(USE_PGRP)
# define KILL(pid, sig) killpg((pid), (sig))
# else
# define KILL(pid, sig) kill((pid), (sig))
# endif
#endif
/*
* Grmpf... There is no way to set bits of the wait structure
* anymore with the stupid W*() macros. I liked the union wait
* stuff much more. So, we devise our own macros... This is
* really ugly, use dramamine sparingly. You have been warned.
*/
#define W_SETMASKED(st, val, fun) \
{ \
int sh = (int) ~0; \
int mask = fun(sh); \
\
for (sh = 0; ((mask >> sh) & 1) == 0; sh++) \
continue; \
*(st) = (*(st) & ~mask) | ((val) << sh); \
}
#define W_SETTERMSIG(st, val) W_SETMASKED(st, val, WTERMSIG)
#define W_SETEXITSTATUS(st, val) W_SETMASKED(st, val, WEXITSTATUS)
static void pass_signal_to_job(void *, void *);
static void handle_all_signals(void);
static void handle_signal(int);
static int JobCmpPid(void *, void *);
static void JobClose(Job *);
static void JobFinish(Job *, int);
static void JobExec(Job *);
static void JobRestart(Job *);
static void JobStart(GNode *, int);
static void JobInterrupt(int, int);
static void JobRestartJobs(void);
static void debug_printf(const char *, ...);
static Job *prepare_job(GNode *, int);
static void start_queued_job(Job *);
static void token(Job *, FILE *);
static void print_partial_buffer(struct job_pipe *, Job *, FILE *, size_t);
static void print_partial_buffer_and_shift(struct job_pipe *, Job *, FILE *,
size_t);
static bool print_complete_lines(struct job_pipe *, Job *, FILE *, size_t);
static void prepare_pipe(struct job_pipe *, int *);
static void handle_job_output(Job *, int, bool);
static void register_error(int, Job *);
static void
register_error(int status, Job *job)
{
struct error_info *p;
errors++;
p = emalloc(sizeof(struct error_info));
p->status = status;
p->name = job->node->name;
if (p)
Lst_AtEnd(&errorsList, p);
}
void
print_errors()
{
LstNode ln;
struct error_info *p;
for (ln = Lst_First(&errorsList); ln != NULL; ln = Lst_Adv(ln)) {
p = (struct error_info *)Lst_Datum(ln);
if (WIFEXITED(p->status)) {
Error("\tExit status %d in target %s",
WEXITSTATUS(p->status), p->name);
} else if (WIFSIGNALED(p->status)) {
Error("\tReceived signal %d in target s",
WTERMSIG(p->status), p->name);
} else {
Error("\tStatus %d in target %s", p->status, p->name);
}
}
}
static void
token(Job *job, FILE *out)
{
if (job->node != lastNode) {
if (DEBUG(JOBTOKEN))
(void)fprintf(out, "--- %s ---\n", job->node->name);
lastNode = job->node;
}
}
static void
handle_all_signals()
{
if (got_signal)
got_signal = 0;
else
return;
if (got_SIGINT) {
got_SIGINT=0;
handle_signal(SIGINT);
}
if (got_SIGHUP) {
got_SIGHUP=0;
handle_signal(SIGHUP);
}
if (got_SIGQUIT) {
got_SIGQUIT=0;
handle_signal(SIGQUIT);
}
if (got_SIGTERM) {
got_SIGTERM=0;
handle_signal(SIGTERM);
}
if (got_SIGTSTP) {
got_SIGTSTP=0;
handle_signal(SIGTSTP);
}
if (got_SIGTTOU) {
got_SIGTTOU=0;
handle_signal(SIGTTOU);
}
if (got_SIGTTIN) {
got_SIGTTIN=0;
handle_signal(SIGTTIN);
}
if (got_SIGWINCH) {
got_SIGWINCH=0;
handle_signal(SIGWINCH);
}
}
/*-
*-----------------------------------------------------------------------
* JobCondPassSig --
* Pass a signal to a job if USE_PGRP
* is defined.
*
* Side Effects:
* None, except the job may bite it.
*-----------------------------------------------------------------------
*/
static void
pass_signal_to_job(void *jobp, /* Job to biff */
void *signop) /* Signal to send it */
{
Job *job = (Job *)jobp;
int signo = *(int *)signop;
if (DEBUG(JOB)) {
(void)fprintf(stdout,
"pass_signal_to_job passing signal %d to child %ld.\n",
signo, (long)job->pid);
(void)fflush(stdout);
}
KILL(job->pid, signo);
}
/*-
*-----------------------------------------------------------------------
* handle_signal --
* Pass a signal to all local jobs if USE_PGRP is defined,
* then die ourselves.
*
* Side Effects:
* We die by the same signal.
*-----------------------------------------------------------------------
*/
static void
handle_signal(int signo) /* The signal number we've received */
{
sigset_t nmask, omask;
struct sigaction act;
if (DEBUG(JOB)) {
(void)fprintf(stdout, "handle_signal(%d) called.\n", signo);
(void)fflush(stdout);
}
Lst_ForEach(&runningJobs, pass_signal_to_job, &signo);
/*
* Deal with proper cleanup based on the signal received. We only run
* the .INTERRUPT target if the signal was in fact an interrupt. The
* other three termination signals are more of a "get out *now*"
* command.
*/
if (signo == SIGINT) {
JobInterrupt(true, signo);
} else if (signo == SIGHUP || signo == SIGTERM || signo == SIGQUIT) {
JobInterrupt(false, signo);
}
/*
* Leave gracefully if SIGQUIT, rather than core dumping.
*/
if (signo == SIGQUIT) {
Finish(0);
}
/*
* Send ourselves the signal now we've given the message to everyone
* else. Note we block everything else possible while we're getting
* the signal. This ensures that all our jobs get continued when we
* wake up before we take any other signal.
*/
sigemptyset(&nmask);
sigaddset(&nmask, signo);
sigprocmask(SIG_SETMASK, &nmask, &omask);
memset(&act, 0, sizeof act);
act.sa_handler = SIG_DFL;
sigemptyset(&act.sa_mask);
act.sa_flags = 0;
sigaction(signo, &act, NULL);
if (DEBUG(JOB)) {
(void)fprintf(stdout,
"handle_signal passing signal to self, mask = %x.\n",
~0 & ~(1 << (signo-1)));
(void)fflush(stdout);
}
(void)signal(signo, SIG_DFL);
(void)KILL(getpid(), signo);
signo = SIGCONT;
Lst_ForEach(&runningJobs, pass_signal_to_job, &signo);
(void)sigprocmask(SIG_SETMASK, &omask, NULL);
sigprocmask(SIG_SETMASK, &omask, NULL);
act.sa_handler = SigHandler;
sigaction(signo, &act, NULL);
}
/*-
*-----------------------------------------------------------------------
* JobCmpPid --
* Compare the pid of the job with the given pid and return 0 if they
* are equal. This function is called from Job_CatchChildren via
* Lst_Find to find the job descriptor of the finished job.
*
* Results:
* 0 if the pid's match
*-----------------------------------------------------------------------
*/
static int
JobCmpPid(void *job, /* job to examine */
void *pid) /* process id desired */
{
return *(pid_t *)pid - ((Job *)job)->pid;
}
static void
debug_printf(const char *fmt, ...)
{
if (DEBUG(JOB)) {
va_list va;
va_start(va, fmt);
(void)vfprintf(stdout, fmt, va);
fflush(stdout);
va_end(va);
}
}
/*-
*-----------------------------------------------------------------------
* JobClose --
* Called to close both input and output pipes when a job is finished.
*
* Side Effects:
* The file descriptors associated with the job are closed.
*-----------------------------------------------------------------------
*/
static void
JobClose(Job *job)
{
int i;
for (i = 0; i < 2; i++) {
FD_CLR(job->in[i].fd, outputsp);
handle_job_output(job, i, true);
(void)close(job->in[i].fd);
}
}
/*-
*-----------------------------------------------------------------------
* JobFinish --
* Do final processing for the given job including updating
* parents and starting new jobs as available/necessary. Note
* that we pay no attention to the JOB_IGNERR flag here.
* This is because when we're called because of a noexecute flag
* or something, jstat.w_status is 0 and when called from
* Job_CatchChildren, the status is zeroed if it s/b ignored.
*
* Side Effects:
* Some nodes may be put on the toBeMade queue.
* Final commands for the job are placed on end_node.
*
* If we got an error and are aborting (aborting == ABORT_ERROR) and
* the job list is now empty, we are done for the day.
* If we recognized an error we set the aborting flag
* to ABORT_ERROR so no more jobs will be started.
*-----------------------------------------------------------------------
*/
/*ARGSUSED*/
static void
JobFinish(Job *job, int status)
{
bool done;
if ((WIFEXITED(status) &&
WEXITSTATUS(status) != 0 && !(job->flags & JOB_IGNERR)) ||
(WIFSIGNALED(status) && WTERMSIG(status) != SIGCONT)) {
/*
* If it exited non-zero and either we're doing things our
* way or we're not ignoring errors, the job is finished.
* Similarly, if the shell died because of a signal
* the job is also finished. In these
* cases, finish out the job's output before printing the exit
* status...
*/
JobClose(job);
done = true;
} else if (WIFEXITED(status)) {
/*
* Deal with ignored errors in -B mode. We need to print a
* message telling of the ignored error as well as setting
* status.w_status to 0 so the next command gets run. To do
* this, we set done to be true if in -B mode and the job
* exited non-zero.
*/
done = WEXITSTATUS(status) != 0;
/*
* Old comment said: "Note we don't want to close down any of
* the streams until we know we're at the end." But we do.
* Otherwise when are we going to print the rest of the stuff?
*/
JobClose(job);
} else {
/*
* No need to close things down or anything.
*/
done = false;
}
if (done ||
WIFSTOPPED(status) ||
(WIFSIGNALED(status) && WTERMSIG(status) == SIGCONT) ||
DEBUG(JOB)) {
if (WIFEXITED(status)) {
debug_printf("Process %ld exited.\n", (long)job->pid);
if (WEXITSTATUS(status) != 0) {
token(job, stdout);
(void)fprintf(stdout, "*** Error code %d%s\n",
WEXITSTATUS(status),
(job->flags & JOB_IGNERR) ? "(ignored)" :
"");
if (job->flags & JOB_IGNERR) {
status = 0;
}
} else if (DEBUG(JOB)) {
token(job, stdout);
(void)fprintf(stdout,
"*** Completed successfully\n");
}
} else if (WIFSTOPPED(status)) {
debug_printf("Process %ld stopped.\n", (long)job->pid);
token(job, stdout);
(void)fprintf(stdout, "*** Stopped -- signal %d\n",
WSTOPSIG(status));
job->flags |= JOB_RESUME;
Lst_AtEnd(&stoppedJobs, job);
(void)fflush(stdout);
return;
} else if (WTERMSIG(status) == SIGCONT) {
/*
* If the beastie has continued, shift the Job from the
* stopped list to the running one (or re-stop it if
* concurrency is exceeded) and go and get another
* child.
*/
if (job->flags & (JOB_RESUME|JOB_RESTART)) {
token(job, stdout);
(void)fprintf(stdout, "*** Continued\n");
}
if (!(job->flags & JOB_CONTINUING)) {
debug_printf(
"Warning: "
"process %ld was not continuing.\n",
(long)job->pid);
#if 0
/*
* We don't really want to restart a job from
* scratch just because it continued,
* especially not without killing the
* continuing process! That's why this is
* ifdef'ed out. FD - 9/17/90
*/
JobRestart(job);
#endif
}
job->flags &= ~JOB_CONTINUING;
Lst_AtEnd(&runningJobs, job);
nJobs++;
debug_printf("Process %ld is continuing locally.\n",
(long)job->pid);
if (nJobs == maxJobs) {
jobFull = true;
debug_printf("Job queue is full.\n");
}
(void)fflush(stdout);
return;
} else {
token(job, stdout);
(void)fprintf(stdout, "*** Signal %d\n",
WTERMSIG(status));
}
(void)fflush(stdout);
}
done = true;
if (done &&
aborting != ABORT_ERROR &&
aborting != ABORT_INTERRUPT &&
status == 0) {
/* As long as we aren't aborting and the job didn't return a
* non-zero status that we shouldn't ignore, we call
* Make_Update to update the parents. */
job->node->made = MADE;
Make_Update(job->node);
free(job);
} else if (status != 0) {
register_error(status, job);
free(job);
}
JobRestartJobs();
/*
* Set aborting if any error.
*/
if (errors && !keepgoing &&
aborting != ABORT_INTERRUPT) {
/*
* If we found any errors in this batch of children and the -k
* flag wasn't given, we set the aborting flag so no more jobs
* get started.
*/
aborting = ABORT_ERROR;
}
if (aborting == ABORT_ERROR && Job_Empty()) {
/*
* If we are aborting and the job table is now empty, we finish.
*/
Finish(errors);
}
}
static void
prepare_pipe(struct job_pipe *p, int *fd)
{
p->pos = 0;
(void)fcntl(fd[0], F_SETFD, FD_CLOEXEC);
p->fd = fd[0];
close(fd[1]);
if (outputsp == NULL || p->fd > outputsn) {
int fdn, ofdn;
fd_set *tmp;
fdn = howmany(p->fd+1, NFDBITS);
ofdn = outputsn ? howmany(outputsn+1, NFDBITS) : 0;
if (fdn != ofdn) {
tmp = recalloc(outputsp, fdn, sizeof(fd_mask));
if (tmp == NULL)
return;
outputsp = tmp;
}
outputsn = p->fd;
}
fcntl(p->fd, F_SETFL, O_NONBLOCK);
FD_SET(p->fd, outputsp);
}
/*-
*-----------------------------------------------------------------------
* JobExec --
* Execute the shell for the given job. Called from JobStart and
* JobRestart.
*
* Side Effects:
* A shell is executed, outputs is altered and the Job structure added
* to the job table.
*-----------------------------------------------------------------------
*/
static void
JobExec(Job *job)
{
pid_t cpid; /* ID of new child */
int fds[4];
int *fdout = fds;
int *fderr = fds+2;
int result;
int i;
if (DEBUG(JOB)) {
(void)fprintf(stdout, "Running %s\n", job->node->name);
(void)fflush(stdout);
}
/*
* Some jobs produce no output and it's disconcerting to have
* no feedback of their running (since they produce no output, the
* banner with their name in it never appears). This is an attempt to
* provide that feedback, even if nothing follows it.
*/
token(job, stdout);
setup_engine();
/* Create the pipe by which we'll get the shell's output.
*/
if (pipe(fdout) == -1)
Punt("Cannot create pipe: %s", strerror(errno));
if (pipe(fderr) == -1)
Punt("Cannot create pipe: %s", strerror(errno));
if ((cpid = fork()) == -1) {
Punt("Cannot fork");
} else if (cpid == 0) {
/* standard pipe code to route stdout and stderr */
close(fdout[0]);
if (dup2(fdout[1], 1) == -1)
Punt("Cannot dup2(outPipe): %s", strerror(errno));
if (fdout[1] != 1)
close(fdout[1]);
close(fderr[0]);
if (dup2(fderr[1], 2) == -1)
Punt("Cannot dup2(errPipe): %s", strerror(errno));
if (fderr[1] != 2)
close(fderr[1]);
#ifdef USE_PGRP
/*
* We want to switch the child into a different process family
* so we can kill it and all its descendants in one fell swoop,
* by killing its process family, but not commit suicide.
*/
# if defined(SYSV)
(void)setsid();
# else
(void)setpgid(0, getpid());
# endif
#endif /* USE_PGRP */
/* most cases won't return, but will exit directly */
result = run_gnode(job->node, 1);
switch(result) {
case MADE:
exit(0);
case ERROR:
exit(1);
default:
fprintf(stderr,
"Could not run gnode, returned %d\n", result);
exit(1);
}
} else {
job->pid = cpid;
/* we set the current position in the buffers to the beginning
* and mark another stream to watch in the outputs mask
*/
for (i = 0; i < 2; i++)
prepare_pipe(&job->in[i], fds+2*i);
}
/*
* Now the job is actually running, add it to the table.
*/
nJobs++;
Lst_AtEnd(&runningJobs, job);
if (nJobs == maxJobs) {
jobFull = true;
}
}
static void
start_queued_job(Job *job)
{
if (DEBUG(JOB)) {
(void)fprintf(stdout, "Restarting %s...",
job->node->name);
(void)fflush(stdout);
}
if (nJobs >= maxJobs && !(job->flags & JOB_SPECIAL)) {
/*
* Can't be exported and not allowed to run locally --
* put it back on the hold queue and mark the table
* full
*/
debug_printf("holding\n");
Lst_AtFront(&stoppedJobs, job);
jobFull = true;
debug_printf("Job queue is full.\n");
return;
} else {
/*
* Job may be run locally.
*/
debug_printf("running locally\n");
}
JobExec(job);
}
/*-
*-----------------------------------------------------------------------
* JobRestart --
* Restart a job that stopped for some reason.
*
* Side Effects:
* jobFull will be set if the job couldn't be run.
*-----------------------------------------------------------------------
*/
static void
JobRestart(Job *job)
{
if (job->flags & JOB_RESTART) {
start_queued_job(job);
} else {
/*
* The job has stopped and needs to be restarted. Why it
* stopped, we don't know...
*/
debug_printf("Resuming %s...", job->node->name);
if ((nJobs < maxJobs || ((job->flags & JOB_SPECIAL) &&
maxJobs == 0)) && nJobs != maxJobs) {
/*
* If we haven't reached the concurrency limit already
* (or maxJobs is 0), it's ok to resume the job.
*/
bool error;
int status = 0;
error = KILL(job->pid, SIGCONT) != 0;
if (!error) {
/*
* Make sure the user knows we've continued the
* beast and actually put the thing in the job
* table.
*/
job->flags |= JOB_CONTINUING;
W_SETTERMSIG(&status, SIGCONT);
JobFinish(job, status);
job->flags &= ~(JOB_RESUME|JOB_CONTINUING);
debug_printf("done\n");
} else {
Error("couldn't resume %s: %s",
job->node->name, strerror(errno));
W_SETEXITSTATUS(&status, 1);
JobFinish(job, status);
}
} else {
/*
* Job cannot be restarted. Mark the table as full and
* place the job back on the list of stopped jobs.
*/
debug_printf("table full\n");
Lst_AtFront(&stoppedJobs, job);
jobFull = true;
debug_printf("Job queue is full.\n");
}
}
}
static Job *
prepare_job(GNode *gn, int flags)
{
Job *job; /* new job descriptor */
bool cmdsOK; /* true if the nodes commands were all right */
bool noExec; /* Set true if we decide not to run the job */
job = emalloc(sizeof(Job));
if (job == NULL) {
Punt("JobStart out of memory");
}
job->node = gn;
/*
* Set the initial value of the flags for this job based on the global
* ones and the node's attributes... Any flags supplied by the caller
* are also added to the field.
*/
job->flags = flags;
if (Targ_Ignore(gn)) {
job->flags |= JOB_IGNERR;
}
if (Targ_Silent(gn)) {
job->flags |= JOB_SILENT;
}
/*
* Check the commands now so any attributes from .DEFAULT have a chance
* to migrate to the node
*/
cmdsOK = Job_CheckCommands(gn, Error);
/*
* If the -n flag wasn't given, we open up OUR (not the child's)
* temporary file to stuff commands in it. The thing is rd/wr so we
* don't need to reopen it to feed it to the shell. If the -n flag
* *was* given, we just set the file to be stdout. Cute, huh?
*/
if ((gn->type & OP_MAKE) || (!noExecute && !touchFlag)) {
/*
* We're serious here, but if the commands were bogus, we're
* also dead...
*/
if (!cmdsOK) {
DieHorribly();
}
if (Lst_IsEmpty(&gn->commands))
noExec = true;
else
noExec = false;
} else if (noExecute) {
if (!cmdsOK || Lst_IsEmpty(&gn->commands))
noExec = true;
else
noExec = false;
} else {
/*
* Just touch the target and note that no shell should be
* executed. Check
* the commands, too, but don't die if they're no good -- it
* does no harm to keep working up the graph.
*/
Job_Touch(gn, job->flags & JOB_SILENT);
noExec = true;
}
/*
* If we're not supposed to execute a shell, don't.
*/
if (noExec) {
/*
* We only want to work our way up the graph if we aren't here
* because the commands for the job were no good.
*/
if (cmdsOK) {
if (aborting == 0) {
Make_Update(job->node);
}
}
free(job);
return NULL;
} else {
return job;
}
}
/*-
*-----------------------------------------------------------------------
* JobStart --
* Start a target-creation process going for the target described
* by the graph node gn.
*
* Side Effects:
* A new Job node is created and added to the list of running
* jobs. PMake is forked and a child shell created.
*-----------------------------------------------------------------------
*/
static void
JobStart(GNode *gn, /* target to create */
int flags) /* flags for the job to override normal ones.
* e.g. JOB_SPECIAL */
{
Job *job;
job = prepare_job(gn, flags);
if (!job)
return;
if (nJobs >= maxJobs && !(job->flags & JOB_SPECIAL) &&
maxJobs != 0) {
/*
* The job can only be run locally, but we've hit the limit of
* local concurrency, so put the job on hold until some other
* job finishes. Note that the special jobs (.BEGIN, .INTERRUPT
* and .END) may be run locally even when the local limit has
* been reached (e.g. when maxJobs == 0), though they will be
* exported if at all possible. In addition, any target marked
* with .NOEXPORT will be run locally if maxJobs is 0.
*/
jobFull = true;
debug_printf("Can only run job locally.\n");
job->flags |= JOB_RESTART;
Lst_AtEnd(&stoppedJobs, job);
} else {
if (nJobs >= maxJobs) {
/*
* If we're running this job locally as a special case
* (see above), at least say the table is full.
*/
jobFull = true;
debug_printf("Local job queue is full.\n");
}
JobExec(job);
}
}
/* Helper functions for JobDoOutput */
/* output debugging token and print characters from 0 to endpos */
static void
print_partial_buffer(struct job_pipe *p, Job *job, FILE *out, size_t endPos)
{
size_t i;
token(job, out);
for (i = 0; i < endPos; i++)
putc(p->buffer[i], out);
}
/* print partial buffer and shift remaining contents */
static void
print_partial_buffer_and_shift(struct job_pipe *p, Job *job, FILE *out,
size_t endPos)
{
size_t i;
print_partial_buffer(p, job, out, endPos);
for (i = endPos; i < p->pos; i++)
p->buffer[i-endPos] = p->buffer[i];
p->pos -= endPos;
}
/* print complete lines, looking back to the limit position
* (stuff before limit was already scanned).
* returns true if something was printed.
*/
static bool
print_complete_lines(struct job_pipe *p, Job *job, FILE *out, size_t limit)
{
size_t i;
for (i = p->pos; i > limit; i--) {
if (p->buffer[i-1] == '\n') {
print_partial_buffer_and_shift(p, job, out, i);
return true;
}
}
return false;
}
/*-
*-----------------------------------------------------------------------
* handle_pipe --
* This functions is called whenever there is something to read on the
* pipe. We collect more output from the given job and store it in the
* job's outBuf. If this makes up lines, we print it tagged by the job's
* identifier, as necessary.
*
* Side Effects:
* curPos may be shifted as may the contents of outBuf.
*-----------------------------------------------------------------------
*/
static void
handle_pipe(struct job_pipe *p,
Job *job, FILE *out, bool finish)
{
int nr; /* number of bytes read */
int oldpos; /* optimization */
/* want to get everything ? -> we block */
if (finish)
fcntl(p->fd, F_SETFL, 0);
do {
nr = read(p->fd, &p->buffer[p->pos],
JOB_BUFSIZE - p->pos);
if (nr == -1) {
if (errno == EAGAIN)
break;
if (DEBUG(JOB)) {
perror("JobDoOutput(piperead)");
}
}
oldpos = p->pos;
p->pos += nr;
if (!print_complete_lines(p, job, out, oldpos))
if (p->pos == JOB_BUFSIZE) {
print_partial_buffer(p, job, out, p->pos);
p->pos = 0;
}
} while (nr != 0);
/* at end of file, we print whatever is left */
if (nr == 0) {
print_partial_buffer(p, job, out, p->pos);
if (p->pos > 0 && p->buffer[p->pos - 1] != '\n')
putchar('\n');
p->pos = 0;
}
}
static void
handle_job_output(Job *job, int i, bool finish)
{
handle_pipe(&job->in[i], job, i == 0 ? stdout : stderr, finish);
}
/*-
*-----------------------------------------------------------------------
* Job_CatchChildren --
* Handle the exit of a child. Called from Make_Make.
*
* Side Effects:
* The job descriptor is removed from the list of children.
*
* Notes:
* We do waits, blocking or not, according to the wisdom of our
* caller, until there are no more children to report. For each
* job, call JobFinish to finish things off. This will take care of
* putting jobs on the stoppedJobs queue.
*-----------------------------------------------------------------------
*/
void
Job_CatchChildren()
{
pid_t pid; /* pid of dead child */
Job *job; /* job descriptor for dead child */
LstNode jnode; /* list element for finding job */
int status; /* Exit/termination status */
/*
* Don't even bother if we know there's no one around.
*/
if (nJobs == 0) {
return;
}
while ((pid = waitpid((pid_t) -1, &status, WNOHANG|WUNTRACED)) > 0) {
handle_all_signals();
debug_printf("Process %ld exited or stopped.\n", (long)pid);
jnode = Lst_Find(&runningJobs, JobCmpPid, &pid);
if (jnode == NULL) {
if (WIFSIGNALED(status) &&
(WTERMSIG(status) == SIGCONT)) {
jnode = Lst_Find(&stoppedJobs, JobCmpPid, &pid);
if (jnode == NULL) {
Error("Resumed child (%ld) not in table", (long)pid);
continue;
}
job = (Job *)Lst_Datum(jnode);
Lst_Remove(&stoppedJobs, jnode);
} else {
Error("Child (%ld) not in table?", (long)pid);
continue;
}
} else {
job = (Job *)Lst_Datum(jnode);
Lst_Remove(&runningJobs, jnode);
nJobs--;
if (jobFull)
debug_printf("Job queue is no longer full.\n");
jobFull = false;
}
JobFinish(job, status);
}
}
/*-
*-----------------------------------------------------------------------
* Job_CatchOutput --
* Catch the output from our children, if we're using
* pipes do so. Otherwise just block time until we get a
* signal (most likely a SIGCHLD) since there's no point in
* just spinning when there's nothing to do and the reaping
* of a child can wait for a while.
*
* Side Effects:
* Output is read from pipes if we're piping.
* -----------------------------------------------------------------------
*/
void
Job_CatchOutput(void)
{
int nfds;
struct timeval timeout;
LstNode ln;
Job *job;
int i;
int count = howmany(outputsn+1, NFDBITS) * sizeof(fd_mask);
fd_set *readfdsp = malloc(count);
(void)fflush(stdout);
if (readfdsp == NULL)
return;
memcpy(readfdsp, outputsp, count);
timeout.tv_sec = SEL_SEC;
timeout.tv_usec = SEL_USEC;
nfds = select(outputsn+1, readfdsp, NULL, NULL, &timeout);
handle_all_signals();
if (nfds > 0) {
for (ln = Lst_First(&runningJobs); nfds && ln != NULL;
ln = Lst_Adv(ln)) {
job = (Job *)Lst_Datum(ln);
for (i = 0; i < 2; i++) {
if (FD_ISSET(job->in[i].fd, readfdsp)) {
handle_job_output(job, i, false);
nfds--;
}
}
}
}
free(readfdsp);
}
/*-
*-----------------------------------------------------------------------
* Job_Make --
* Start the creation of a target. Basically a front-end for
* JobStart used by the Make module.
*
* Side Effects:
* Another job is started.
*-----------------------------------------------------------------------
*/
void
Job_Make(GNode *gn)
{
(void)JobStart(gn, 0);
}
/*-
*-----------------------------------------------------------------------
* Job_Init --
* Initialize the process module
*
* Side Effects:
* lists and counters are initialized
*-----------------------------------------------------------------------
*/
void
Job_Init(int maxproc)
{
Static_Lst_Init(&runningJobs);
Static_Lst_Init(&stoppedJobs);
Static_Lst_Init(&queuedJobs);
Static_Lst_Init(&errorsList);
maxJobs = maxproc;
nJobs = 0;
jobFull = false;
errors = 0;
aborting = 0;
lastNode = NULL;
if ((begin_node->type & OP_DUMMY) == 0) {
JobStart(begin_node, JOB_SPECIAL);
while (nJobs) {
Job_CatchOutput();
Job_CatchChildren();
}
}
}
/*-
*-----------------------------------------------------------------------
* Job_Full --
* See if the job table is full. It is considered full if it is OR
* if we are in the process of aborting OR if we have
* reached/exceeded our local quota. This prevents any more jobs
* from starting up.
*
* Results:
* true if the job table is full, false otherwise
*-----------------------------------------------------------------------
*/
bool
Job_Full(void)
{
return aborting || jobFull;
}
/*-
*-----------------------------------------------------------------------
* Job_Empty --
* See if the job table is empty. Because the local concurrency may
* be set to 0, it is possible for the job table to become empty,
* while the list of stoppedJobs remains non-empty. In such a case,
* we want to restart as many jobs as we can.
*
* Results:
* true if it is. false if it ain't.
* -----------------------------------------------------------------------
*/
bool
Job_Empty(void)
{
if (nJobs == 0) {
if (!Lst_IsEmpty(&stoppedJobs) && !aborting) {
/*
* The job table is obviously not full if it has no
* jobs in it...Try and restart the stopped jobs.
*/
jobFull = false;
JobRestartJobs();
return false;
} else {
return true;
}
} else {
return false;
}
}
/*-
*-----------------------------------------------------------------------
* JobInterrupt --
* Handle the receipt of an interrupt.
*
* Side Effects:
* All children are killed. Another job will be started if the
* .INTERRUPT target was given.
*-----------------------------------------------------------------------
*/
static void
JobInterrupt(int runINTERRUPT, /* Non-zero if commands for the .INTERRUPT
* target should be executed */
int signo) /* signal received */
{
LstNode ln; /* element in job table */
Job *job; /* job descriptor in that element */
aborting = ABORT_INTERRUPT;
for (ln = Lst_First(&runningJobs); ln != NULL; ln = Lst_Adv(ln)) {
job = (Job *)Lst_Datum(ln);
if (!Targ_Precious(job->node)) {
const char *file = job->node->path == NULL ?
job->node->name : job->node->path;
if (!noExecute && eunlink(file) != -1) {
Error("*** %s removed", file);
}
}
if (job->pid) {
debug_printf("JobInterrupt passing signal to "
"child %ld.\n", (long)job->pid);
KILL(job->pid, signo);
}
}
if (runINTERRUPT && !touchFlag) {
if ((interrupt_node->type & OP_DUMMY) == 0) {
ignoreErrors = false;
JobStart(interrupt_node, 0);
while (nJobs) {
Job_CatchOutput();
Job_CatchChildren();
}
}
}
exit(signo);
}
/*
*-----------------------------------------------------------------------
* Job_Finish --
* Do final processing such as the running of the commands
* attached to the .END target.
*
* Results:
* Number of errors reported.
*
*-----------------------------------------------------------------------
*/
int
Job_Finish(void)
{
if (end_node != NULL && !Lst_IsEmpty(&end_node->commands)) {
if (errors) {
Error("Errors reported so .END ignored");
} else {
JobStart(end_node, JOB_SPECIAL);
while (nJobs) {
Job_CatchOutput();
Job_CatchChildren();
}
}
}
return errors;
}
#ifdef CLEANUP
void
Job_End(void)
{
}
#endif
/*-
*-----------------------------------------------------------------------
* Job_Wait --
* Waits for all running jobs to finish and returns. Sets 'aborting'
* to ABORT_WAIT to prevent other jobs from starting.
*
* Side Effects:
* Currently running jobs finish.
*
*-----------------------------------------------------------------------
*/
void
Job_Wait(void)
{
aborting = ABORT_WAIT;
while (nJobs != 0) {
Job_CatchOutput();
Job_CatchChildren();
}
aborting = 0;
}
/*-
*-----------------------------------------------------------------------
* Job_AbortAll --
* Abort all currently running jobs without handling output or anything.
* This function is to be called only in the event of a major
* error. Most definitely NOT to be called from JobInterrupt.
*
* Side Effects:
* All children are killed, not just the firstborn
*-----------------------------------------------------------------------
*/
void
Job_AbortAll(void)
{
LstNode ln; /* element in job table */
Job *job; /* the job descriptor in that element */
int foo;
aborting = ABORT_ERROR;
if (nJobs) {
for (ln = Lst_First(&runningJobs); ln != NULL;
ln = Lst_Adv(ln)) {
job = (Job *)Lst_Datum(ln);
/*
* kill the child process with increasingly drastic
* signals to make darn sure it's dead.
*/
KILL(job->pid, SIGINT);
KILL(job->pid, SIGKILL);
}
}
/*
* Catch as many children as want to report in at first, then give up
*/
while (waitpid(-1, &foo, WNOHANG) > 0)
continue;
}
/*-
*-----------------------------------------------------------------------
* JobRestartJobs --
* Tries to restart stopped jobs if there are slots available.
* Note that this tries to restart them regardless of pending errors.
* It's not good to leave stopped jobs lying around!
*
* Side Effects:
* Resumes(and possibly migrates) jobs.
*-----------------------------------------------------------------------
*/
static void
JobRestartJobs(void)
{
Job *job;
while (!jobFull && (job = (Job *)Lst_DeQueue(&stoppedJobs)) != NULL) {
debug_printf("Job queue is not full. "
"Restarting a stopped job.\n");
JobRestart(job);
}
}
![[BACK]](/icons/back.gif){kind=icon}
Return to job.c CVS log ![[TXT]](/icons/text.gif){kind=icon}
![[DIR]](/icons/dir.gif){kind=icon}
Up to [local] / src / usr.bin / make