On Mon, 2025-05-26 at 16:38 +0200, [email protected] wrote:
> On Mon, May 26, 2025 at 02:26:55PM +0000, Gerhard Roth wrote:
> > On Mon, 2025-05-26 at 15:46 +0200, Claudio Jeker wrote:
> > > On Mon, May 26, 2025 at 01:02:20PM +0000, Gerhard Roth wrote:
> > > > Hi Pierre,
> > > > 
> > > > thanks for your input. Your findings are correct. We shouldn't
> > > > trust the values provided by a black-box USB device.
> > > > 
> > > > 
> > > > On Fri, 2025-05-23 at 18:49 +0200, Pierre Pronchery wrote:
> > > > >                         Dear OpenBSD team,
> > > > > 
> > > > > I would like to bring your attention to the following bug report from
> > > > > FreeBSD, where I have ported and imported the umb(4) driver recently:
> > > > > https://bugs.freebsd.org/bugzilla/show_bug.cgi?id=284904
> > > > > 
> > > > > The bug report mentions:
> > > > > 
> > > > > > in_len2mask(mask, len) will write as many as len/8 bytes:
> > > > > > 
> > > > > > for (i = 0; i < len / 8; i++)
> > > > > > p[i] = 0xff;
> > > > > > 
> > > > > > len comes from a ipv4elem.prefixlen in a MBIM_CID_IP_CONFIGURATION
> > > > > > message from the USB device, and can be any uint32_t value. So a 
> > > > > > broken
> > > > > > or malicious USB device can cause a buffer overflow.
> > > > > 
> > > > > I think that in reality, len comes from the network, which would make 
> > > > > the
> > > > > issue marginally worse.
> > > > > 
> > > > > Can you confirm the bug on your side, and would you have any 
> > > > > suggestion
> > > > > as to how to fix it properly?
> > > > > 
> > > > > I would suggest to make len an unsigned value in umb_decode_cid() and
> > > > > subsequent calls (infolen is unsigned in the first place there) but 
> > > > > more
> > > > > importantly, to verify that prefixlen is at most 32.
> > > > > 
> > > > > Does that make sense?
> > > > > 
> > > > > HTH,
> > > > > --
> > > > > khorben
> > > > 
> > > > I would propose to limit 'prefixlen' to 32.
> > > > 
> > > > Gerhard
> > > > 
> > > > Index: sys/dev/usb/if_umb.c
> > > > ===================================================================
> > > > RCS file: /cvs/src/sys/dev/usb/if_umb.c,v
> > > > diff -u -p -u -p -r1.59 if_umb.c
> > > > --- sys/dev/usb/if_umb.c        8 Aug 2024 05:10:00 -0000       1.59
> > > > +++ sys/dev/usb/if_umb.c        26 May 2025 11:37:15 -0000
> > > > @@ -1838,7 +1838,8 @@ umb_add_inet_config(struct umb_softc *sc
> > > >         sin = &ifra.ifra_mask;
> > > >         sin->sin_family = AF_INET;
> > > >         sin->sin_len = sizeof (*sin);
> > > > -       in_len2mask(&sin->sin_addr, prefixlen);
> > > > +       in_len2mask(&sin->sin_addr,
> > > > +           MIN(prefixlen, sizeof (struct in_addr) * NBBY));
> > > >  
> > > >         rv = in_ioctl(SIOCAIFADDR, (caddr_t)&ifra, ifp, 1);
> > > >         if (rv != 0) {
> > > > 
> > > > 
> > > > 
> > > 
> > > Please do not use NBBY and instead use 8 or just hardcode 32.
> > 
> > Fine, updated patch below.
> 
> Maybe it would be better to fail hard in that case and return an error.

in6_prefixlen2mask() is even uglier! It logs an error and returns even
before bzero'ing the returned mask.


> 
> Diff below is OK claudio@
>  
> > Index: sys/dev/usb/if_umb.c
> > ===================================================================
> > RCS file: /cvs/src/sys/dev/usb/if_umb.c,v
> > diff -u -p -u -p -r1.59 if_umb.c
> > --- sys/dev/usb/if_umb.c        8 Aug 2024 05:10:00 -0000       1.59
> > +++ sys/dev/usb/if_umb.c        26 May 2025 14:25:34 -0000
> > @@ -1838,7 +1838,8 @@ umb_add_inet_config(struct umb_softc *sc
> >         sin = &ifra.ifra_mask;
> >         sin->sin_family = AF_INET;
> >         sin->sin_len = sizeof (*sin);
> > -       in_len2mask(&sin->sin_addr, prefixlen);
> > +       in_len2mask(&sin->sin_addr,
> > +           MIN(prefixlen, sizeof (struct in_addr) * 8));
> >  
> >         rv = in_ioctl(SIOCAIFADDR, (caddr_t)&ifra, ifp, 1);
> >         if (rv != 0) {
> > 
> > 
> 
> 
> 

Attachment: smime.p7s
Description: S/MIME cryptographic signature

Reply via email to