wkt@csadfa.cs.adfa.oz.au (Warren Toomey) (05/21/91)
I include a short program which appears to demonstrate a bug in the
setpgrp() under SunOS 4.1.1. The program works under Pyramid OSx, and
I think should work under previous version of SunOS. If anybody can show
me that the bug is in fact a programming error I'd be most happy!
Thanks in advance,
Warren Toomey.
---- cut here, and unshar ----
echo x - fred.c
sed '/^X/s///' > fred.c << '/'
X#include <signal.h>
X#include <sgtty.h>
X#include <termio.h> /* Need this for TCGETA definition */
X
X
X/* The following program exhibits a possible bug in SunOS 4.1.1. Compile
X * this program as fred: cc -o fred fred.c and run it with no arguments.
X *
X * The program sets the terminal to stop background writes to the terminal,
X * and forks a child. The child moves itself to a new process group, and
X * tries to write to the terminal in one of two ways: just after the fork(),
X * and after the exec().
X *
X * The parent waits for the SIGCHLD caused by the write, and attempts to
X * bring the child back into the parent's pgrp, thus allowing the write to
X * succeed.
X *
X * However, this fails under SunOS 4.1.1 when the child only tries to write
X * AFTER the exec(). See `THE BUG' comment below.
X */
X
Xvoid catchsig()
X { }
X
Xvoid dowrite()
X { write(1,"Writing to stdout\n",18); exit(0); }
X
X
Xmain(argc,argv)
X int argc;
X char *argv[];
X {
X struct termio tbuf;
X int i;
X
X if (argc>1) dowrite(); /* If we are called with >1 args, write */
X
X/* Set up the terminal so that SIGTTOUs are sent to
X * processes which are not members of the owning process group
X */
X if (ioctl(0,TCGETA,&tbuf)) perror("ioctl in settou");
X tbuf.c_lflag |= TOSTOP;
X if (ioctl(0,TCSETA,&tbuf)) perror("ioctl s");
X
X
X/* Catch the SIGCHLD with a null handler.
X */
X signal(SIGCHLD,catchsig);
X
X/* Fork, and make the child write on stdout.
X */
X switch(i=fork())
X {
X case -1: printf("Could not fork\n"); exit(1);
X
X case 0 : /* The child */
X
X i=setpgrp(0,getpid()); /* Take us out of the pgrp */
X sleep(2); /* Wait for a bit */
X
X/* THE BUG:
X *
X * If you uncomment the printf line, the program works everywhere.
X * But if you leave the line commented, the program dies with an
X * EPERM error in the parent's setpgrp() under SunOS Release 4.1.1.
X * It works under Pyramid OSx, generic 4.3BSD system.
X */
X /* printf("Can I write??\n"); */
X execlp("fred","arg1","arg2",0);
X exit(0);
X
X default: /* The parent */
X
X printf("Forked, now we pause()\n");
X pause(); /* Wait for SIGCHLD */
X if (setpgrp(i,getpgrp(0))==-1) /* set pgrp to the parents's */
X { perror("Setpgrp failed"); exit(1); }
X kill(i,SIGCONT); /* Start the child up */
X sleep(1); /* and finish */
X exit(0);
X }
X }
/urban@cbnewsl.att.com (john.urban) (05/22/91)
In article <1991May21.022214.1189@sserve.cc.adfa.oz.au> wkt@csadfa.cs.adfa.oz.au (Warren Toomey) writes: >I include a short program which appears to demonstrate a bug in the >setpgrp() under SunOS 4.1.1. The program works under Pyramid OSx, and >I think should work under previous version of SunOS. If anybody can show >me that the bug is in fact a programming error I'd be most happy! > >Thanks in advance, When fred is compiled on AT&T UNIX System V/x86 Release 4.0 Version 2.1, I get: $ ./fred Forked, now we pause() Writing to stdout $ If I then change fred and uncomment the 'bug' line I get: $ ./fred Forked, now we pause() Can I write?? Writing to stdout $ Sincerely, John Ben Urban
decot@hpcupt1.cup.hp.com (Dave Decot) (06/04/91)
Caution: BSD-derived systems such as SunOS and System V-derived systems (such as SunOS /-:) have different semantics for the function setpgrp(). In fact, the BSD version has two arguments, whereas the System V version has no arguments. Check your documentation carefully. On POSIX systems, use setpgid() or setsid() as appropriate to your situation. Dave