From owner-freebsd-current  Sun Sep 22 01:38:06 1996
Return-Path: owner-current
Received: (from root@localhost)
          by freefall.freebsd.org (8.7.5/8.7.3) id BAA29322
          for current-outgoing; Sun, 22 Sep 1996 01:38:06 -0700 (PDT)
Received: from melb.werple.net.au (melb.werple.net.au [203.9.190.18])
          by freefall.freebsd.org (8.7.5/8.7.3) with ESMTP id BAA29292
          for <current@FreeBSD.org>; Sun, 22 Sep 1996 01:38:02 -0700 (PDT)
Received: (from uucp@localhost) by melb.werple.net.au (8.7.6/8.7.3/2) with UUCP id RAA25150; Sun, 22 Sep 1996 17:52:31 +1000 (EST)
Received: (from jb@localhost) by freebsd3.cimlogic.com.au (8.7.5/8.7.3) id RAA15900; Sun, 22 Sep 1996 17:28:38 +1000 (EST)
From: John Birrell <jb@cimlogic.com.au>
Message-Id: <199609220728.RAA15900@freebsd3.cimlogic.com.au>
Subject: Re: libc_r bug
To: hsu@freefall.freebsd.org (Jeffrey Hsu)
Date: Sun, 22 Sep 1996 17:28:38 +1000 (EST)
Cc: jb@cimlogic.com.au, current@FreeBSD.org
In-Reply-To: <199609220613.XAA28775@freefall.freebsd.org> from Jeffrey Hsu at "Sep 21, 96 11:13:07 pm"
X-Mailer: ELM [version 2.4ME+ PL22 (25)]
MIME-Version: 1.0
Content-Type: text/plain; charset=US-ASCII
Content-Transfer-Encoding: 7bit
Sender: owner-current@FreeBSD.org
X-Loop: FreeBSD.org
Precedence: bulk

[ Redirected cc to -current ]

>   > _thread_fd_table_init() just sets up
>   > the table for a fd. fds 0, 1 and 2 don't have to be valid.
> 
> Since we can't tell whether a given fd might need a call to
> _thread_fd_table_init() or not, the correctness before all else
> principle would argue for either pre-allocating all the fd entries
> or doing it on demand by placing a call to _thread_fd_table_init()
> in write() and all the other places where it might be needed.  Of
> these two, I prefer the second.  What about you?

I prefer the second too.

A call to _thread_fd_table_init() in write() is only needed if the fd
is nonblocking (wrt to the thread), because _thread_fd_lock() calls
_thread_fd_table_init() anyway. But you can't check if the fd is nonblocking
without calling _thread_fd_table_init(), so we might as well just say
that the fd is always locked, even if it is nonblocking and take the
same performance hit.

Try the attached diff (which I haven't compiled 8-).


-- 
John Birrell                                CIMlogic Pty Ltd
jb@cimlogic.com.au                          119 Cecil Street
Ph  +61  3 9690 6900                        South Melbourne Vic 3205
Fax +61  3 9690 6650                        Australia
Mob +61 18  353  137

*** /u/freebsd/src/lib/libc_r/uthread/uthread_write.c	Mon Jan 22 11:23:54 1996
--- /u/freebsd/newsrc/lib/libc_r/uthread/uthread_write.c	Sun Sep 22 17:22:34 1996
***************
*** 45,56 ****
  	int             nonblock;
  	int             ret;
  	int             status;
! 	if (fd < 0 || fd > _thread_dtablesize || _thread_fd_table[fd] == NULL) {
! 		_thread_seterrno(_thread_run, EBADF);
! 		ret = -1;
! 	} else if ((nonblock = _thread_fd_table[fd]->flags & O_NONBLOCK) == 0 && (ret = _thread_fd_lock(fd, FD_RDWR, NULL, __FILE__, __LINE__)) != 0) {
  		/* Cannot lock file descriptor. */
  	} else {
  		while ((ret = _thread_sys_write(fd, buf, nbytes)) < 0) {
  			if (nonblock == 0 && (errno == EWOULDBLOCK || errno == EAGAIN)) {
  				_thread_kern_sig_block(&status);
--- 45,54 ----
  	int             nonblock;
  	int             ret;
  	int             status;
! 	if (ret = _thread_fd_lock(fd, FD_RDWR, NULL, __FILE__, __LINE__)) != 0) {
  		/* Cannot lock file descriptor. */
  	} else {
+ 		nonblock = _thread_fd_table[fd]->flags & O_NONBLOCK;
  		while ((ret = _thread_sys_write(fd, buf, nbytes)) < 0) {
  			if (nonblock == 0 && (errno == EWOULDBLOCK || errno == EAGAIN)) {
  				_thread_kern_sig_block(&status);
***************
*** 65,73 ****
  				break;
  			}
  		}
! 		if (nonblock == 0) {
! 			_thread_fd_unlock(fd, FD_RDWR);
! 		}
  	}
  	return (ret);
  }
--- 63,69 ----
  				break;
  			}
  		}
! 		_thread_fd_unlock(fd, FD_RDWR);
  	}
  	return (ret);
  }