Re: Rusty's module talk at the Kernel Summit

[Alexander Viro]

On Thu, 18 Jul 2002, Richard Gooch wrote:

> Alexander Viro writes:
> > Call them well-behaving modules if you wish.  For these the answers
> > are "yes"/"a lot of things can be"/"it's easy to handle".  What's
> > left?  The pieces of code with really complex interfaces.  And guess
> > what, race-prevention is complex for these guys - and it's not just
> > about rmmod races.  E.g. parts of procfs, sysctls and devfs are
> > still quite racy even if you compile everything into the tree and
> > remove all module-related syscalls completely.
> 
> Can you point to specific problems with the current devfs code?

Sigh...  How many do you want?  Look, couple of days ago I'd done the
following: picked a random number in range 1..`wc -l fs/devfs/base.c`,
checked what function it was in (devfs_readdir()) and spent less than
two minutes reading it before finding a bug (a leak - there's a couple
of paths that grab an entry and return without releasing it).

So tell me how many times I should repeat that exercise and while you
are at it, tell me what stops you from doing the same.  Because you
know, reading devfs code is something I'd rather avoid - it's not my
idea of fun reading.  IF it will stop you from claiming "Al hadn't
done public whippings lately, so devfs is bug-free" for a couple of
months - by all means, tell how many bugs do I need to find and report
to shut you up for a while.

Richard, devfs code is _ripe_ with bugs; you can't spit into it without
hitting one.  And excuse me, but when finding one is a matter of two
minutes I can't believe that you are incapable of doing that on your own.
It used to be annoying; by now it's beyond annoying - it's ridiculous.