[<prev] [next>] [<thread-prev] [day] [month] [year] [list]
Date: Fri, 10 Sep 2021 12:26:48 -0700
From: "H.J. Lu" <hjl.tools@...il.com>
To: Florian Weimer <fweimer@...hat.com>
Cc: GNU C Library <libc-alpha@...rceware.org>, GDB <gdb@...rceware.org>,
libc-coord@...ts.openwall.com, Daniel Walker <danielwa@...co.com>
Subject: Re: [PATCH v8 1/1] Extend struct r_debug to support multiple
namespaces [BZ #15971]
On Fri, Sep 10, 2021 at 11:59 AM Florian Weimer <fweimer@...hat.com> wrote:
>
> * H. J. Lu:
>
> > diff --git a/elf/link.h b/elf/link.h
> > index ff3a85c847..a297318236 100644
> > --- a/elf/link.h
> > +++ b/elf/link.h
> > @@ -34,14 +34,13 @@
>
> > -/* This is the instance of that structure used by the dynamic linker. */
> > +/* This is the compatibility symbol of that structure provided by the
> > + dynamic linker. */
> > extern struct r_debug _r_debug;
>
> I don't think we should say “compatibility symbol” in a public header.
I will remove "compatibility".
> Can we move GNAT off this symbol and deprecate it at least?
There is no harm in keeping it.
> > +/* The extended rendezvous structure used by the run-time dynamic linker
> > + to communicate details of shared object loading to the debugger. If
> > + the executable's dynamic section has a DT_DEBUG element, the run-time
> > + linker sets that element's value to the address where this structure
> > + can be found. */
> > +
> > +struct r_debug_extended
> > + {
> > + struct r_debug base;
> > +
> > + /* The following field is added by r_version == 2. */
> > +
> > + /* Link to the next r_debug_extended structure. Each r_debug_extended
> > + structure represents a different namespace. The first
> > + r_debug_extended structure is for the default namespace. */
> > + struct r_debug_extended *r_next;
> > + };
> > +
> > /* This symbol refers to the "dynamic structure" in the `.dynamic' section
> > of whatever module refers to `_DYNAMIC'. So, to find its own
> > - `struct r_debug', a program could do:
> > + `struct r_debug_extended', a program could do:
> > for (dyn = _DYNAMIC; dyn->d_tag != DT_NULL; ++dyn)
> > if (dyn->d_tag == DT_DEBUG)
> > - r_debug = (struct r_debug *) dyn->d_un.d_ptr;
> > - */
> > + r_debug_extended = (struct r_debug_extended *) dyn->d_un.d_ptr;
> > + */
> > extern ElfW(Dyn) _DYNAMIC[];
>
> What about shared objects? How can they find r_debug_extended? Should
> they just make sure they have DT_DEBUG in their dynamic section?
Linker generates DT_DEBUG only in the executable. dl_iterate_phdr can be
used to locate DT_DEBUG in the executable.
>
> Calling getauxval (AT_PHDR) has a relocation dependencies, which I
> expect some consumers want to avoid.
>
> > +Extension to the r_debug structure
> > +==================================
> > +
> > +The r_debug_extended structure is an extension of the r_debug interface.
> > +If r_version is 2, one additional field is available:
> > +
> > + struct r_debug_extended *r_next;
> > + Link to the next r_debug_extended structure. Each r_debug_extended
> > + structure represents a different namespace. The first r_debug_extended
> > + structure is for the default namespace.
>
> I think this should say how a reader can determine which list elements
> are in fact active.
I will update it.
>
> > diff --git a/elf/tst-dlmopen4.c b/elf/tst-dlmopen4.c
> > new file mode 100644
> > index 0000000000..7a6c502e8c
> > --- /dev/null
> > +++ b/elf/tst-dlmopen4.c
>
> > +static int
> > +do_test (void)
> > +{
> > + void *h = xdlmopen (LM_ID_NEWLM, "$ORIGIN/tst-dlmopen1mod.so",
> > + RTLD_LAZY);
>
> I think this should test that r_version is 1 before the dlmopen call.
I will add the test.
> > +
> > + int status = EXIT_FAILURE;
> > + ElfW(Dyn) *d;
> > + for (d = _DYNAMIC; d->d_tag != DT_NULL; ++d)
> > + {
> > + struct r_debug_extended *debug = ELF_MACHINE_GET_R_DEBUG (d);
> > + if (debug != NULL)
> > + {
> > + TEST_VERIFY_EXIT (debug->base.r_version == 2);
>
> You could use TEST_COMPARE.
I will fix it.
>
> > + TEST_VERIFY_EXIT (debug->r_next != NULL);
> > + TEST_VERIFY_EXIT (debug->r_next->r_next == NULL);
> > + TEST_VERIFY_EXIT (debug->r_next->base.r_map != NULL);
> > + TEST_VERIFY_EXIT (debug->r_next->base.r_map->l_name != NULL);
> > + const char *name = basename (debug->r_next->base.r_map->l_name);
> > + TEST_VERIFY_EXIT (strcmp (name, "tst-dlmopen1mod.so") == 0);
>
> You could use TEST_COMPARE_STRING.
I will fix it.
> Sorry, I have not reviewed the actual mechanics of the patch.
>
> Thanks,
> Florian
>
Thanks.
--
H.J.
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.
