Follow @Openwall on Twitter for new release announcements and other news
[<prev] [next>] [<thread-prev] [thread-next>] [day] [month] [year] [list]
Message-ID: <20240610160408.GN10433@brightrain.aerifal.cx>
Date: Mon, 10 Jun 2024 12:04:08 -0400
From: Rich Felker <dalias@...c.org>
To: Ismael Luceno <ismael@...ev.co.uk>
Cc: musl@...ts.openwall.com
Subject: Re: [PATCH] ioctl: Fix implicit constant conversion overflow

On Mon, Jun 03, 2024 at 03:57:18AM +0200, Ismael Luceno wrote:
> On 02/Jun/2024 18:50, Rich Felker wrote:
> > On Sun, Jun 02, 2024 at 05:01:10AM +0200, Ismael Luceno wrote:
> > > On 31/May/2024 22:34, Rich Felker wrote:
> > > <...>
> > > > > +#define _IOW(a,b,c) _IOC(_IOC_WRITE,(a),(b),(int)sizeof(c))
> > > > > +#define _IOR(a,b,c) _IOC(_IOC_READ,(a),(b),(int)sizeof(c))
> > > > > +#define _IOWR(a,b,c) _IOC(_IOC_READ|_IOC_WRITE,(a),(b),(int)sizeof(c))
> > > > 
> > > > I don't see how this helps with the warning you're trying to suppress,
> > > 
> > > GCC disagrees; the warnings go away because it's this element that
> > > causes the whole expression to be promoted to unsigned long long,
> > > so making it smaller (we can use unsigned int instead) avoids the
> > > issue.
> > 
> > In that case gcc is just being inconsistent. Both the conversion from
> > unsigned int to int and size_t to int are non-value-preserving. It
> > makes no sense that it warns for the latter but not for the former.
> > 
> > "Make weird inconsistent warning messages go away" is not a motivation
> > for a change. If the command macros could all be made to have type int
> > (matcing the ioctl argument) without introducing new problems, that
> > would be a well-motivated change. I suppose "make them have type
> > unsigned int rather than unsigned long so that they're not
> > gratuitously over-wide" might be well-motivated too, but I suspect it
> > leaves in place warnings in some places. "Fix implicit constant
> > conversion overflow" is not a well-motivated change since there is no
> > overflow.
> 
> GCC doesn't make much sense here but the warning appears with several
> versions of GCC.
> 
> An explicit cast at _IOC instead would make sense to me, but what could
> break in your opinion?

I'm not sure. It needs investigation. There might have been some
concern with breakage from kernel headers that define ioctl numbers or
something. I just remember this hasn't been as simple as it sounds
from past times it came up..

Rich

Powered by blists - more mailing lists

Confused about mailing lists and their use? Read about mailing lists on Wikipedia and check out these guidelines on proper formatting of your messages.