[v2] elf: Add -z lazy to some more tests
Checks
| Context |
Check |
Description |
| dj/TryBot-apply_patch |
success
|
Patch applied to master at the time it was sent
|
| dj/TryBot-32bit |
success
|
Build for i686
|
Commit Message
From: Arsen Arsenović <arsen@gentoo.org>
Some toolchains, such as that used on Gentoo Hardened, set '-z now' out
of the box. This trips up a couple of tests.
---
Evening,
This is a re-roll of the following patch series:
https://inbox.sourceware.org/libc-alpha/20230302112519.914641-1-arsen@gentoo.org/
Changes from v1:
- Dropped -z norelro. This turned out to be unnecessary after
Adhemervals removal of --with-default-link and linker script
machinery:
https://patchwork.sourceware.org/project/glibc/list/?series=17843
See: https://inbox.sourceware.org/libc-alpha/86fsakz5mr.fsf@gentoo.org
for an explanation of what caused the need for norelro. That fix was
misguided, due to a previous error on my part, too.
I applied this patch on top of that patchset and it would appear to
resolve all related failures.
The above is not applied to Git yet, but should be before this patch
is.
elf/Makefile | 20 ++++++++++++++++++++
1 file changed, 20 insertions(+)
Comments
* Arsen Arsenović:
> @@ -1612,6 +1631,7 @@ $(objpfx)multiload.out: $(objpfx)testobj1.so
> LDFLAGS-origtest = -rdynamic
> $(objpfx)origtest.out: $(objpfx)testobj1.so
>
> +$(objpfx)resolvfail.out: $(objpfx)testobj1.so
> ifeq ($(have-thread-library),yes)
> $(objpfx)resolvfail: $(shared-thread-library)
> endif
Thsis seems unrelated? I can see why it's needed, but please mention it
in the commit message at least?
Thanks,
Florian
Florian Weimer <fweimer@redhat.com> writes:
> * Arsen Arsenović:
>
>> @@ -1612,6 +1631,7 @@ $(objpfx)multiload.out: $(objpfx)testobj1.so
>> LDFLAGS-origtest = -rdynamic
>> $(objpfx)origtest.out: $(objpfx)testobj1.so
>>
>> +$(objpfx)resolvfail.out: $(objpfx)testobj1.so
>> ifeq ($(have-thread-library),yes)
>> $(objpfx)resolvfail: $(shared-thread-library)
>> endif
>
> Thsis seems unrelated? I can see why it's needed, but please mention it
> in the commit message at least?
Ah, right. I forgot about that entirely.
That test uses the testobj6, but doesn't depend on it. If running the
elf/resolvfail test before make check, it's possible to see:
FAIL: elf/resolvfail
original exit status 1
cannot load testobj1.so: testobj1.so: cannot open shared object file: No such file or directory
... which is fixed by just adding the dependency. Will split into a new
commit and re-send.
@@ -1190,6 +1190,11 @@ postclean-generated += $(objpfx)/dso-sort-tests-2.generated-makefile \
ifeq (yes,$(have-tunables))
$(eval $(call include_dsosort_tests,dso-sort-tests-1.def))
$(eval $(call include_dsosort_tests,dso-sort-tests-2.def))
+
+# BZ15311 is intentionally underlinked.
+LDFLAGS-tst-bz15311-b.so += -Wl,-z,lazy
+LDFLAGS-tst-bz15311-c.so += -Wl,-z,lazy
+LDFLAGS-tst-bz15311-d.so += -Wl,-z,lazy
endif
check-abi: $(objpfx)check-abi-ld.out \
@@ -1514,6 +1519,20 @@ LDFLAGS-tst-initorderb2.so = -Wl,--no-as-needed
LDFLAGS-tst-tlsmod5.so = -nostdlib -Wl,--no-as-needed
LDFLAGS-tst-tlsmod6.so = -nostdlib -Wl,--no-as-needed
+# The following tests are underlinked, and rely on late loading. On toolchains
+# that set -z now by default, this leads to failures to load or fix up the
+# executables being tested.
+LDFLAGS-circlemod2.so = -Wl,-z,lazy
+LDFLAGS-tst-tls20mod-bad.so = -Wl,-z,lazy
+LDFLAGS-reldep6mod1.so += -Wl,-z,lazy
+LDFLAGS-constload2.so = -Wl,-z,lazy
+LDFLAGS-constload3.so = -Wl,-z,lazy
+LDFLAGS-dblloadmod3.so = -Wl,-z,lazy
+LDFLAGS-ifuncmod6.so = -Wl,-z,lazy
+LDFLAGS-ltglobmod2.so = -Wl,-z,lazy
+LDFLAGS-testobj1.so = -Wl,-z,lazy
+LDFLAGS-testobj6.so = -Wl,-z,lazy
+
testobj1.so-no-z-defs = yes
testobj3.so-no-z-defs = yes
testobj4.so-no-z-defs = yes
@@ -1612,6 +1631,7 @@ $(objpfx)multiload.out: $(objpfx)testobj1.so
LDFLAGS-origtest = -rdynamic
$(objpfx)origtest.out: $(objpfx)testobj1.so
+$(objpfx)resolvfail.out: $(objpfx)testobj1.so
ifeq ($(have-thread-library),yes)
$(objpfx)resolvfail: $(shared-thread-library)
endif