Re: BUG: $ORIGIN does not seem to work

[Szabolcs Nagy <nsz@port70.net>]

* Rich Felker <dalias@libc.org> [2018-02-07 12:35:31 -0500]:
> On Sun, Jan 28, 2018 at 01:54:25AM +0100, Szabolcs Nagy wrote:
> > * Stefan Fröberg <stefan.froberg@petroprogram.com> [2018-01-28 00:07:33 +0200]:
> > > strace ./x
> > ...
> > > open("/root/batman/lib/libcrypto.so.1.1", O_RDONLY|O_CLOEXEC) = 3
> > ...
> > > ldd x
> > >     /lib/ld-musl-x86_64.so.1 (0x7f22efa03000)
> > >     libcrypto.so.1.1 => /usr/lib/libcrypto.so.1.1 (0x7f22ef352000)
> > >     libc.so => /lib/ld-musl-x86_64.so.1 (0x7f22efa03000)
> > 
> > this is a bug in musl ldd: if the executable
> > path has no / then it assumes origin is /, try
> > 
> > ldd ./x
> > 
> > then origin is ./ so it works as expected.
> 
> Does the attached patch look ok?
> 

looks good.

> Alternatively we could fix the assumed invariant that p->name always
> contains a slash by having the loader code in __dls3 allocate a copy
> of argv[0] with "./" prepended, but that's a heavier cost at runtime
> and doesn't seem to have any practical advantages.
> 
> Rich

> diff --git a/ldso/dynlink.c b/ldso/dynlink.c
> index 3380240..f507bc4 100644
> --- a/ldso/dynlink.c
> +++ b/ldso/dynlink.c
> @@ -807,7 +807,12 @@ static int fixup_rpath(struct dso *p, char *buf, size_t buf_size)
>  		origin = p->name;
>  	}
>  	t = strrchr(origin, '/');
> -	l = t ? t-origin : 0;
> +	if (t) {
> +		l = t-origin;
> +	} else {
> +		origin = ".";
> +		l = 1;
> +	}
>  	p->rpath = malloc(strlen(p->rpath_orig) + n*l + 1);
>  	if (!p->rpath) return -1;
>