Problems with AIO?

[David Schultz]

On Fri, Apr 25, 2003, drew at PoohSticks.ORG wrote:
> I have a program which needs to do multiple non-blocking disk I/Os
> in parallel on raw devices, have been using pthreads with wrappers that
> provide a single-threaded event driven interface, and am in the process of 
> switching to AIO to get rid of some debugger/pthread interactions
> and make some Windows guys happy because AIO looks more like their
> async IO interface.
> 
> So, I look in the LINT config file and see this:
>     # Use real implementations of the aio_* system calls.  There are numerous
>     # stability issues in the current aio code that make it unsuitable for
>     # inclusion on shell boxes.
>     options         VFS_AIO
> When I query GNATs for AIO in the single-line text fields, I get just this
> 
>     2002/01/17] kern/34017               The siginfo_t passed to the signal handling routine is not filled out for AIO
> 
> and a web search only turns up this exploit
> 
>     http://www.securiteam.com/exploits/6L00E0A3FC.html

Of these bugs, I believe the first one still exists, and the
second was eliminated long ago in -CURRENT.  (You may wish to try
the posted exploit to confirm this.)  I don't know what other
problems might exist, but realize that this feature is not heavily
utilized and therefore tested by the FreeBSD user base.  That
said, it works for me.  Perhaps people would use it more and work
all the bugs out right away if we removed the ``beware of dog''
sign and a while later enabled it by default.