Re: Un-sucking EINVAL

--------
Warner Losh writes:

> In short, I'd love to widen the interface, but there's a number of practical
> issues in the way.

Not really ?

Our kernel is not I18N'ed and we have no plans to do so.

That means that there is absolutely no point in taking a big detour over
enumerating all possible messages into some "int assistant_to_errno",
we can and should simply pass the informative text-string straigt through.

There is also no need to make the text-string more than a const char *.

The proper detailed explanation is in the manual page, the string
just needs to say enough that people know where to look.

Here is the simplest implementation:

1. We add two pointers to kernel threads.

2. When an error occurs, those pointers are set to a pointer to the
   kld (or NULL) and to the const char *string.

3. Userland explicitly requests the information (a syscall or per-thread
   sysctl), the kernel validates that the kld is still present before
   copyout'ing the const char *.

4. Kernel code adds calls like:

	errmsg("Grammeter must be connected first")

   before return (EINVAL)

Done.

-- 
Poul-Henning Kamp       | UNIX since Zilog Zeus 3.20
phk@FreeBSD.ORG         | TCP/IP since RFC 956
FreeBSD committer       | BSD since 4.3-tahoe    
Never attribute to malice what can adequately be explained by incompetence.