Subject: Re: pkg/5617
To: Greg A. Woods <woods@weird.com>
From: Todd Vierling <tv@pobox.com>
List: netbsd-bugs
Date: 06/20/1998 00:23:22
On Fri, 19 Jun 1998, Greg A. Woods wrote:

: > The configuration _is_ optimal for NetBSD, and shouldn't depend on any
: > optional kernel components.
: 
: If you look closely at the new patch I provided you'll find that it does
: not in any way change the *current* requirement that the kernel support
: SysV shared memory.

"Since when?"  I'm not using it; it's not compiled into _my_ kernel
_anywhere_, and Apache happily runs.  Please document where shmat() or
msgop() or semop() or the like is used by it in the current setup for
NetBSD.  Note that mmap() is _not_ "SYSV shared memory".

: Note that it is still *HIGHLY* recommended that some form of
: "serialized" accept mechanism be used, and USE_FLOCK_SERIALIZED_ACCEPT,
: which is one of the changes suggested by my patch, is the most obvious
: choice for NetBSD.

Then that can be added.  Adding SYSVSEM is not acceptable, as that is an
_optional_ NetBSD component.

: The second change my patch suggests is to turn off USE_MMAP_FILES, which
: to me is probably a very good idea until the time that NetBSD gets a
: fully integrated buffer cache.

Second time:  "Why?"  It works fine, as Apache does _not_ __write__ to any
files that are mmap()ed.  The mmap()/read&write() problems occur when
_changing_ data in a file both mmaped and read/written.

: Remember the patch I supplied was in the form of a patch itself.  I
: realize you have to look twice to see what it really implies, but it
: seemed more appropriate to supply it as a patch, since that would make
: the PR fix easy to apply.  Here's the patch file as just a patch file
: (and with the pathname corrected to not require -p2):

: +#define USE_FLOCK_SERIALIZED_ACCEPT

_This_ is plausible, and I'll check and likely add it.

-- 
-- Todd Vierling (Personal tv@pobox.com; Bus. todd_vierling@xn.xerox.com)