On Tue, Jul 2, 2024 at 1:04 PM Dmitry V. Levin <l...@strace.io> wrote:
>
> On Mon, Jul 01, 2024 at 06:56:21PM +0200, Michal Sekletar wrote:
> [...]
> > --- a/libdwfl/link_map.c
> > +++ b/libdwfl/link_map.c
> > @@ -416,8 +416,22 @@ report_r_debug (uint_fast8_t elfclass,
uint_fast8_t elfdata,
> > if (name != NULL)
> > {
> > /* This code is mostly inlined dwfl_report_elf. */
> > - // XXX hook for sysroot
> > - int fd = open (name, O_RDONLY);
> > + int rc;
> > + char *path_name;
> > + const char *sysroot = dwfl->sysroot;
> > +
> > + /* Don't use the sysroot if the path is already inside it. */
> > + bool name_in_sysroot = sysroot && (strncmp(name, sysroot,
strlen(sysroot)) == 0);
>
> This seems to be a good candidate for startswith(), e.g.
Makes sense. Will be fixed in the next version.
>
> bool name_in_sysroot = sysroot && startswith(name, sysroot);
>
> Is sysroot guaranteed to end with '/'? If not, then sysroot being
> a prefix doesn't always imply that the path is inside sysroot.
sysroot (dwfl->sysroot) is a result of realpath(). I just did a bunch of
tests and it seems that realpath does not append trailing "/". So you are
right, prefix doesn't always imply placement within sysroot. I will fix
this in the next version.
>
> > +
> > + if (!name_in_sysroot && sysroot)
> > + rc = asprintf(&path_name, "%s/%s", sysroot, name);
> > + else
> > + rc = asprintf(&path_name, "%s", name);
>
> Do we need the last asprintf(), or could we use "name" directly?
>
> In fact, besides this open(), "name" is used later in this function as an
> argument to __libdwfl_report_elf(). Does that invocation also has to be
> updated?
>
> Could "name" be re-purposed a bit in case of sysroot? Just an idea,
> completely untested:
>
> char *sysroot_name = NULL;
> ...
> if (sysroot && !name_in_sysroot)
> {
> if (asprintf(&sysroot_name, "%s/%s", sysroot, name) < 0)
> return ...;
> name = sysroot_name;
> }
> ...
> free(sysroot_name);
This might work. I will give it a try.
Michal
>
>
> --
> ldv
