elf: ld.so --help calls _dl_init_paths without a main map [BZ #27577]
Commit Message
In this case, use the link map of the dynamic loader itself as
a replacement. This is more than just a hack: if we ever support
DT_RUNPATH/DT_RPATH for the dynamic loader, reporting it for
ld.so --help (without further command line arguments) would be the
right thing to do.
Fixes commit 332421312576bd7095e70589154af99b124dd2d1 ("elf: Always
set l in _dl_init_paths (bug 23462)").
Tested on i686-linux-gnu and x86_64-linux-gnu.
---
elf/Makefile | 10 +++++++++-
elf/dl-load.c | 8 +++++++-
2 files changed, 16 insertions(+), 2 deletions(-)
Comments
On Mon, Mar 15, 2021 at 3:03 AM Florian Weimer via Libc-alpha
<libc-alpha@sourceware.org> wrote:
>
> In this case, use the link map of the dynamic loader itself as
> a replacement. This is more than just a hack: if we ever support
> DT_RUNPATH/DT_RPATH for the dynamic loader, reporting it for
> ld.so --help (without further command line arguments) would be the
> right thing to do.
>
> Fixes commit 332421312576bd7095e70589154af99b124dd2d1 ("elf: Always
> set l in _dl_init_paths (bug 23462)").
>
> Tested on i686-linux-gnu and x86_64-linux-gnu.
>
> ---
> elf/Makefile | 10 +++++++++-
> elf/dl-load.c | 8 +++++++-
> 2 files changed, 16 insertions(+), 2 deletions(-)
>
> diff --git a/elf/Makefile b/elf/Makefile
> index 4c9e63dac9..ebb331df36 100644
> --- a/elf/Makefile
> +++ b/elf/Makefile
> @@ -245,7 +245,8 @@ tests += $(tests-execstack-$(have-z-execstack))
> ifeq ($(run-built-tests),yes)
> tests-special += $(objpfx)tst-leaks1-mem.out \
> $(objpfx)tst-leaks1-static-mem.out $(objpfx)noload-mem.out \
> - $(objpfx)tst-ldconfig-X.out
> + $(objpfx)tst-ldconfig-X.out $(objpfx)tst-rtld-help.out \
> + $(objpfx)tst-rtld-help-2.out
> endif
> tlsmod17a-suffixes = 0 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19
> tlsmod18a-suffixes = 0 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19
> @@ -1907,3 +1908,10 @@ $(objpfx)list-tunables.out: tst-rtld-list-tunables.sh $(objpfx)ld.so
> $(evaluate-test)
>
> tst-dst-static-ENV = LD_LIBRARY_PATH='$$ORIGIN'
> +
> +$(objpfx)tst-rtld-help.out: $(objpfx)ld.so
> + $(test-wrapper) $(rtld-prefix) --help > $@; \
> + $(evaluate-test)
> +$(objpfx)tst-rtld-help-2.out: $(objpfx)tst-rtld-help.out
> + grep -q 'Legacy HWCAP subdirectories under library search path directories' $<; \
> + $(evaluate-test)
Can we avoid tst-rtld-help-2.out by grep $@ directly?
> diff --git a/elf/dl-load.c b/elf/dl-load.c
> index 376a2e64d6..2f760503c5 100644
> --- a/elf/dl-load.c
> +++ b/elf/dl-load.c
> @@ -758,8 +758,14 @@ _dl_init_paths (const char *llp, const char *source,
> max_dirnamelen = SYSTEM_DIRS_MAX_LEN;
> *aelem = NULL;
>
> - /* This points to the map of the main object. */
> + /* This points to the map of the main object. If there is no main
> + object (e.g., under --help, use the dynamic loader itself as a
> + stand-in. */
> l = GL(dl_ns)[LM_ID_BASE]._ns_loaded;
> +#ifdef SHARED
> + if (l == NULL)
> + l = &GL (dl_rtld_map);
> +#endif
> assert (l->l_type != lt_loaded);
>
> if (l->l_info[DT_RUNPATH])
>
* H. J. Lu:
>> +$(objpfx)tst-rtld-help.out: $(objpfx)ld.so
>> + $(test-wrapper) $(rtld-prefix) --help > $@; \
>> + $(evaluate-test)
>> +$(objpfx)tst-rtld-help-2.out: $(objpfx)tst-rtld-help.out
>> + grep -q 'Legacy HWCAP subdirectories under library search path directories' $<; \
>> + $(evaluate-test)
>
> Can we avoid tst-rtld-help-2.out by grep $@ directly?
We could use &&, but that would make diagnosing test failures difficult
(as in: did --help fail, or grep?). The separate test seemed the
easiest way.
Thanks,
Florian
On Mon, Mar 15, 2021 at 5:14 AM Florian Weimer <fweimer@redhat.com> wrote:
>
> * H. J. Lu:
>
> >> +$(objpfx)tst-rtld-help.out: $(objpfx)ld.so
> >> + $(test-wrapper) $(rtld-prefix) --help > $@; \
> >> + $(evaluate-test)
> >> +$(objpfx)tst-rtld-help-2.out: $(objpfx)tst-rtld-help.out
> >> + grep -q 'Legacy HWCAP subdirectories under library search path directories' $<; \
> >> + $(evaluate-test)
> >
> > Can we avoid tst-rtld-help-2.out by grep $@ directly?
>
> We could use &&, but that would make diagnosing test failures difficult
> (as in: did --help fail, or grep?). The separate test seemed the
> easiest way.
>
Running "ld.so --help" itself is very straight forward. When there
is foo.out, I
usually search for foo.
* H. J. Lu:
> On Mon, Mar 15, 2021 at 5:14 AM Florian Weimer <fweimer@redhat.com> wrote:
>>
>> * H. J. Lu:
>>
>> >> +$(objpfx)tst-rtld-help.out: $(objpfx)ld.so
>> >> + $(test-wrapper) $(rtld-prefix) --help > $@; \
>> >> + $(evaluate-test)
>> >> +$(objpfx)tst-rtld-help-2.out: $(objpfx)tst-rtld-help.out
>> >> + grep -q 'Legacy HWCAP subdirectories under library search path directories' $<; \
>> >> + $(evaluate-test)
>> >
>> > Can we avoid tst-rtld-help-2.out by grep $@ directly?
>>
>> We could use &&, but that would make diagnosing test failures difficult
>> (as in: did --help fail, or grep?). The separate test seemed the
>> easiest way.
>
> Running "ld.so --help" itself is very straight forward. When there is
> foo.out, I usually search for foo.
Sorry, I don't understand what your comment means in this context.
Would you please elaborate?
Thanks,
Florian
On Mon, Mar 15, 2021 at 5:28 AM Florian Weimer <fweimer@redhat.com> wrote:
>
> * H. J. Lu:
>
> > On Mon, Mar 15, 2021 at 5:14 AM Florian Weimer <fweimer@redhat.com> wrote:
> >>
> >> * H. J. Lu:
> >>
> >> >> +$(objpfx)tst-rtld-help.out: $(objpfx)ld.so
> >> >> + $(test-wrapper) $(rtld-prefix) --help > $@; \
> >> >> + $(evaluate-test)
> >> >> +$(objpfx)tst-rtld-help-2.out: $(objpfx)tst-rtld-help.out
> >> >> + grep -q 'Legacy HWCAP subdirectories under library search path directories' $<; \
> >> >> + $(evaluate-test)
> >> >
> >> > Can we avoid tst-rtld-help-2.out by grep $@ directly?
> >>
> >> We could use &&, but that would make diagnosing test failures difficult
> >> (as in: did --help fail, or grep?). The separate test seemed the
> >> easiest way.
> >
> > Running "ld.so --help" itself is very straight forward. When there is
> > foo.out, I usually search for foo.
>
> Sorry, I don't understand what your comment means in this context.
> Would you please elaborate?
A single tst-rtld-help.out doesn't make diagnosing ld.so --help
failures difficult.
Multiple output files is confusing.
On Mär 15 2021, Florian Weimer via Libc-alpha wrote:
> +$(objpfx)tst-rtld-help-2.out: $(objpfx)tst-rtld-help.out
> + grep -q 'Legacy HWCAP subdirectories under library search path directories' $<; \
> + $(evaluate-test)
Shouldn't that create the target?
Andreas.
* H. J. Lu:
> On Mon, Mar 15, 2021 at 5:28 AM Florian Weimer <fweimer@redhat.com> wrote:
>>
>> * H. J. Lu:
>>
>> > On Mon, Mar 15, 2021 at 5:14 AM Florian Weimer <fweimer@redhat.com> wrote:
>> >>
>> >> * H. J. Lu:
>> >>
>> >> >> +$(objpfx)tst-rtld-help.out: $(objpfx)ld.so
>> >> >> + $(test-wrapper) $(rtld-prefix) --help > $@; \
>> >> >> + $(evaluate-test)
>> >> >> +$(objpfx)tst-rtld-help-2.out: $(objpfx)tst-rtld-help.out
>> >> >> + grep -q 'Legacy HWCAP subdirectories under library search path directories' $<; \
>> >> >> + $(evaluate-test)
>> >> >
>> >> > Can we avoid tst-rtld-help-2.out by grep $@ directly?
>> >>
>> >> We could use &&, but that would make diagnosing test failures difficult
>> >> (as in: did --help fail, or grep?). The separate test seemed the
>> >> easiest way.
>> >
>> > Running "ld.so --help" itself is very straight forward. When there is
>> > foo.out, I usually search for foo.
>>
>> Sorry, I don't understand what your comment means in this context.
>> Would you please elaborate?
>
> A single tst-rtld-help.out doesn't make diagnosing ld.so --help
> failures difficult.
> Multiple output files is confusing.
All mtrace tests do this. However, I will submit a different patch with
just one target.
Thanks,
Florian
* Andreas Schwab:
> On Mär 15 2021, Florian Weimer via Libc-alpha wrote:
>
>> +$(objpfx)tst-rtld-help-2.out: $(objpfx)tst-rtld-help.out
>> + grep -q 'Legacy HWCAP subdirectories under library search path directories' $<; \
>> + $(evaluate-test)
>
> Shouldn't that create the target?
H.J. wants me to remove the separate target. It will address this issue
as well.
Thanks,
Florian
@@ -245,7 +245,8 @@ tests += $(tests-execstack-$(have-z-execstack))
ifeq ($(run-built-tests),yes)
tests-special += $(objpfx)tst-leaks1-mem.out \
$(objpfx)tst-leaks1-static-mem.out $(objpfx)noload-mem.out \
- $(objpfx)tst-ldconfig-X.out
+ $(objpfx)tst-ldconfig-X.out $(objpfx)tst-rtld-help.out \
+ $(objpfx)tst-rtld-help-2.out
endif
tlsmod17a-suffixes = 0 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19
tlsmod18a-suffixes = 0 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19
@@ -1907,3 +1908,10 @@ $(objpfx)list-tunables.out: tst-rtld-list-tunables.sh $(objpfx)ld.so
$(evaluate-test)
tst-dst-static-ENV = LD_LIBRARY_PATH='$$ORIGIN'
+
+$(objpfx)tst-rtld-help.out: $(objpfx)ld.so
+ $(test-wrapper) $(rtld-prefix) --help > $@; \
+ $(evaluate-test)
+$(objpfx)tst-rtld-help-2.out: $(objpfx)tst-rtld-help.out
+ grep -q 'Legacy HWCAP subdirectories under library search path directories' $<; \
+ $(evaluate-test)
@@ -758,8 +758,14 @@ _dl_init_paths (const char *llp, const char *source,
max_dirnamelen = SYSTEM_DIRS_MAX_LEN;
*aelem = NULL;
- /* This points to the map of the main object. */
+ /* This points to the map of the main object. If there is no main
+ object (e.g., under --help, use the dynamic loader itself as a
+ stand-in. */
l = GL(dl_ns)[LM_ID_BASE]._ns_loaded;
+#ifdef SHARED
+ if (l == NULL)
+ l = &GL (dl_rtld_map);
+#endif
assert (l->l_type != lt_loaded);
if (l->l_info[DT_RUNPATH])