[dm-devel] [PATCH 1/4] libmultipath: get_udev_uid: make sure pp->wwid is 0-terminated

[Martin Wilck]

Hi Bart,

On Fri, 2017-07-14 at 14:56 +0000, Bart Van Assche wrote:
> On Fri, 2017-07-14 at 13:32 +0200, Martin Wilck wrote:
> > If the first WWID_LEN bytes of the uuid_attribute do not contain
> > a 0 byte, pp->wwid may end up not properly terminated. Fix it.
> > 
> > Signed-off-by: Martin Wilck <mwilck at suse.com>
> > ---
> >  libmultipath/discovery.c | 1 +
> >  1 file changed, 1 insertion(+)
> > 
> > diff --git a/libmultipath/discovery.c b/libmultipath/discovery.c
> > index 663c8eaa..9951af84 100644
> > --- a/libmultipath/discovery.c
> > +++ b/libmultipath/discovery.c
> > @@ -1615,6 +1615,7 @@ get_udev_uid(struct path * pp, char
> > *uid_attribute, struct udev_device *udev)
> >  			len = strlen(value);
> >  		}
> >  		strncpy(pp->wwid, value, len);
> > +		pp->wwid[WWID_SIZE - 1] = '\0';
> >  	} else {
> >  		condlog(3, "%s: no %s attribute", pp->dev,
> >  			uid_attribute);
> 
> Hi Martin,
> 
> Your patch does not cause all overflows to be reported.

I'm not sure what you mean. The overflow message is printed if and only
if (strlen(value) + 1 > WWID_SIZE), which is correct, AFAICS. The point
of my patch is just to avoid that multipath crashes later due to an
unterminated string caused by this overflow.

>  How about using the
> following (untested) alternative?
> 
> diff --git a/libmultipath/discovery.c b/libmultipath/discovery.c
> index eca4ce97..80d962e6 100644
> --- a/libmultipath/discovery.c
> +++ b/libmultipath/discovery.c
> @@ -1607,13 +1607,8 @@ get_udev_uid(struct path * pp, char
> *uid_attribute, struct udev_device *udev)
>  	if (!value || strlen(value) == 0)
>  		value = getenv(uid_attribute);
>  	if (value && strlen(value)) {
> -		if (strlen(value) + 1 > WWID_SIZE) {
> +		if (strlcpy(pp->wwid, value, sizeof(pp->wwid)) >=
> WWID_SIZE)
>  			condlog(0, "%s: wwid overflow", pp->dev);
> -			len = WWID_SIZE;
> -		} else {
> -			len = strlen(value);
> -		}
> -		strncpy(pp->wwid, value, len);
>  	} else {
>  		condlog(3, "%s: no %s attribute", pp->dev,
>  			uid_attribute);
> Bart.

Let's have a strncpy vs. strlcpy discussion :D !

I can do this if you insist, but I don't see a big benefit. We've
tested with the patch I submitted.

Thanks,
Martin