Re: Fw: Re: oops in choose_configuration()

[Dmitry Torokhov <dtor_core@ameritech.net>]

On Tuesday 07 March 2006 20:27, Greg KH wrote:
> On Tue, Mar 07, 2006 at 08:13:28PM -0500, Dmitry Torokhov wrote:
> > On Tuesday 07 March 2006 19:57, Linus Torvalds wrote:
> > > 
> > > On Tue, 7 Mar 2006, Chuck Ebbert wrote:
> > > > 
> > > > At least one susbsystem rolls its own method of adding env vars to the
> > > > uevent buffer, and it's so broken it triggers the WARN_ON() in
> > > > lib/vsprintf.c::vsnprintf() by passing a negative length to that function.
> > > 
> > > Well, snprintf() should be safe, though. It will warn if the caller is 
> > > lazy, but these days, the thing does
> > > 
> > > 	max(buf_size - len, 0)
> > > 
> > > which should mean that the input layer passes in 0 instead of a negative 
> > > number. And snprintf() will then _not_ print anything. 
> > > 
> > > So I think input_add_uevent_bm_var() is safe, even if it's not pretty.
> > > 
> > > However, input_devices_read() doesn't do any sanity checking at all, and 
> > > if that ever ends up printing more than a page, that would be bad. I 
> > > didn't look very closely, but it looks worrisome.
> > > 
> > > Dmitry?
> > 
> > I had all this code converted to seq_file, but it depends on converting
> > input handlers to class interfaces and it is not possible nowadays
> > because with latest Greg's changes to class code we would try to
> > register class devices while registering class devices/interfaces
> > (psmouse creates input_dev which binds to mousedev interface which in
> > turn tries to create mouseX which is also belongs to input class) and
> > deadlocking. Greg promised current implementation is only a temporary
> > solution.
> > 
> > I suppose I could separate those changes...
> 
> That would probably be a good idea :)
> 

Hmm, what is the policy for attr->show()? With hotplug variables we
return -ENOMEM if there is not enough memory to store all data, but
what about attributes? Should we also return error (and which one,
-ENOMEM, -ENOBUFS?) or fill as much as we can and return up to
PAGE_SIZE? With sysfs not kernel nor application can really recover
if attribute needs buffer larger than a page. Or just rely on BUG_ON
in fs/sysfs/file.c::fill_read_buffer()?

-- 
Dmitry