2.0.15 SIGBUS in mod_autoindex?

[Dale Ghent <da...@elemental.org>]

I grabbed the 2.0.15 tarball from dev.apache and compiled with:

./configure --prefix=/local/apache2.0 --enable-info --enable-status
--with-mpm=threaded --enable-so

$OPTIM was set to -O3 on Solaris 8, latest patches.

I did some testing of large directory listings using my mp3 stash, and it
seemed that when accessing a directory with FanceIndexing turned on, the
child handling the request would die with a SIGBUS after doing the
readdir's (as seen in the attached truss output with the getdents64()
calls).

However, if I turned FancyIndexing OFF, the request would succeed and
there would be no SIGBUS.

And of course, there's no core file left over to pick through.

I wouldnt know where to look in the code for this... but there is a full
truss output of a (dying) child in action attached to this mail.

/dale

Re: 2.0.15 SIGBUS in mod_autoindex?

[Joshua Slive <jo...@slive.ca>]

On Sun, 25 Mar 2001, Dale Ghent wrote:

>
> I grabbed the 2.0.15 tarball from dev.apache and compiled with:
>
> ./configure --prefix=/local/apache2.0 --enable-info --enable-status
> --with-mpm=threaded --enable-so
>
> $OPTIM was set to -O3 on Solaris 8, latest patches.
>
> I did some testing of large directory listings using my mp3 stash, and it
> seemed that when accessing a directory with FanceIndexing turned on, the
> child handling the request would die with a SIGBUS after doing the
> readdir's (as seen in the attached truss output with the getdents64()
> calls).
>

I don't know if it is the same issue, but if you look in the bug database
you will find several people who report problems with mod_autoindex under
Solaris 8 with Apache 1.3.

Joshua.

Re: 2.0.15 SIGBUS in mod_autoindex?

[Jeff Trawick <tr...@bellsouth.net>]

Dale Ghent <da...@elemental.org> writes:

> Following up my own post here with some wierd findings...
> 
> While taking a look at this problem, I did the natural thing a compiled a
> version with debugging symbols (adding --with-maintainer-mode to the above
> configure command) and not having OPTIM set. I put the new debugging bin
> in place and started the server with FancyIndexing on.
> 
> To my suprise, it worked flawlessly. No SIGBUS'es; truss and gdb both
> showed httpd processes happily processing requests that failed before.
> 
> I recompiled a new httpd bin without debugging and with OPTIM=-O3, as
> before. It began failing again... and one thing that I noticed about the
> failures... is that it happened only when viewing directories that have
> regular files in them. Directories that have only other directories in it
> would be fine.
> 
> A binary compiled the "default" way (no maintainer-mode and OPTIM) works
> fine.
> 
> Are gcc optimizations perhaps broken on Solaris 8? I'll try lesser degrees
> of optimizations tomorrow (-O2, etc).

Also, can you get a core dump and post a backtrace?

Thanks,

Jeff

-- 
Jeff Trawick | trawickj@bellsouth.net | PGP public key at web site:
       http://www.geocities.com/SiliconValley/Park/9289/
             Born in Roswell... married an alien...

Re: 2.0.15 SIGBUS in mod_autoindex?

[Dale Ghent <da...@elemental.org>]

On Sun, 25 Mar 2001, Dale Ghent wrote:

| 
| I grabbed the 2.0.15 tarball from dev.apache and compiled with:
| 
| ./configure --prefix=/local/apache2.0 --enable-info --enable-status
| --with-mpm=threaded --enable-so
| 
| $OPTIM was set to -O3 on Solaris 8, latest patches.
| 
| I did some testing of large directory listings using my mp3 stash, and it
| seemed that when accessing a directory with FanceIndexing turned on, the
| child handling the request would die with a SIGBUS after doing the
| readdir's (as seen in the attached truss output with the getdents64()
| calls).
| 
| However, if I turned FancyIndexing OFF, the request would succeed and
| there would be no SIGBUS.

Following up my own post here with some wierd findings...

While taking a look at this problem, I did the natural thing a compiled a
version with debugging symbols (adding --with-maintainer-mode to the above
configure command) and not having OPTIM set. I put the new debugging bin
in place and started the server with FancyIndexing on.

To my suprise, it worked flawlessly. No SIGBUS'es; truss and gdb both
showed httpd processes happily processing requests that failed before.

I recompiled a new httpd bin without debugging and with OPTIM=-O3, as
before. It began failing again... and one thing that I noticed about the
failures... is that it happened only when viewing directories that have
regular files in them. Directories that have only other directories in it
would be fine.

A binary compiled the "default" way (no maintainer-mode and OPTIM) works
fine.

Are gcc optimizations perhaps broken on Solaris 8? I'll try lesser degrees
of optimizations tomorrow (-O2, etc).

/dale