Re: [Xen-devel] [PATCH] fix misc issues related to allowingsupport of more CPUs

[Keir Fraser <keir.fraser@xxxxxxxxxxxxx>]

On 22/9/08 09:35, "Keir Fraser" <keir.fraser@xxxxxxxxxxxxx> wrote:

> On 22/9/08 08:46, "Jan Beulich" <jbeulich@xxxxxxxxxx> wrote:
> 
>>> Why must it be WARN_ON()? You think you could specify strings so long that
>>> they overflow 32 bits? You've got other problems in that case.
>> 
>> No, that's not the reason. It's because of how bitmap_scnprintf() and
>> bitmap_scnlistprintf() work - they can (validly, assuming that the code
>> having been derived from Linux and still being that way in Linux, hence
>> apparently considered correct) pass negative sizes to scnprintf(), and
>> hence it must not kill the system to actually do so.
> 
> The obvious answer would be to fix the bogus callers. Or consider negative
> size to be a valid input. Warning on what callers consider valid behaviour
> is just weird. I would say the former (fix the callers) is the better way to
> go here; presumably they just need to clamp the size parameter.

Further to this, actually I think the callers in bitmap.c are correct.
Assuming len<=buflen at the start of the bitmap_scn*() functions (valid
since buflen is non-negative and len == 0 at function start) then we'll
never have len>buflen because scnprintf() truncates its return value to be
less than its size parameter, which is always buflen-len.

What *is* a bug is that scnprintf() returns -1 when passed a size==0, which
it specifically is not supposed to do! I notice this bug appears to live on
in current Linux 2.6 too.

I will fix this bug and extend the vsnprintf() BUG_ON() with the predicate
you used for your WARN_ON() logic. Beyond that I don't think anything needs
to be done.

 -- Keir



_______________________________________________
Xen-devel mailing list
Xen-devel@xxxxxxxxxxxxxxxxxxx
http://lists.xensource.com/xen-devel