2 Sep
2022
2 Sep
'22
5:24 a.m.
When uclibc is built with static PIE support the _dl_load_base variable
shared between the libc-tls.c and reloc_static_pie.c creates the
dependency that requires linking reloc_static_pie.o into static
position-dependent executables resulting in the following build errors:
gcc -static test.c -o test
...ld:
...usr/lib/libc.a(reloc_static_pie.os):(.text+0x0):
undefined reference to `_DYNAMIC'
Move _dl_load_base definition to libc-tls.c to resolve this dependency
and fix static PDE build.
Signed-off-by: Max Filippov <jcmvbkbc(a)gmail.com>
---
libc/misc/internals/reloc_static_pie.c | 2 +-
libpthread/nptl/sysdeps/generic/libc-tls.c | 4 ++--
2 files changed, 3 insertions(+), 3 deletions(-)
diff --git a/libc/misc/internals/reloc_static_pie.c
b/libc/misc/internals/reloc_static_pie.c
index 578202d235a1..f66f7a4236c9 100644
--- a/libc/misc/internals/reloc_static_pie.c
+++ b/libc/misc/internals/reloc_static_pie.c
@@ -20,7 +20,7 @@
#include <elf.h>
#include <dl-elf.h>
-ElfW(Addr) _dl_load_base = NULL;
+extern ElfW(Addr) _dl_load_base;
void
reloc_static_pie (ElfW(Addr) load_addr);
diff --git a/libpthread/nptl/sysdeps/generic/libc-tls.c
b/libpthread/nptl/sysdeps/generic/libc-tls.c
index 0c8c558d05d3..e0e272e01dae 100644
--- a/libpthread/nptl/sysdeps/generic/libc-tls.c
+++ b/libpthread/nptl/sysdeps/generic/libc-tls.c
@@ -117,6 +117,7 @@ init_static_tls (size_t memsz, size_t align)
GL(dl_tls_static_nelem) = GL(dl_tls_max_dtv_idx);
}
+ElfW(Addr) _dl_load_base;
void __libc_setup_tls (size_t tcbsize, size_t tcbalign);
void
__libc_setup_tls (size_t tcbsize, size_t tcbalign)
@@ -143,8 +144,7 @@ __libc_setup_tls (size_t tcbsize, size_t tcbalign)
#else
initimage = (void *) phdr->p_vaddr;
#if !defined(SHARED) && defined(STATIC_PIE)
- extern ElfW(Addr) _dl_load_base;
- initimage += _dl_load_base;
+ initimage += _dl_load_base;
#endif
#endif
align = phdr->p_align;
--
2.30.2
2 Sep
2 Sep
6:11 a.m.
I haven't had a chance to test this yet, but the original decision of
placing it in reloc_static_pie.c was to ensure that when reloc_static_pie()
accessed _dl_load_base it would do so without relying on relocations.
Do you mind letting me know which architecture you are building on? On one
or two architectures I do make use of _DYNAMIC in _start() and I wonder if
maybe that is leaking into static builds?
Also it doesn't look like you are 0 initializing _dl_load_base in its new
location. This is a requirement since non-pie static binaries will hit the
code path and expect _dl_load_base to be 0.
On Fri, Sep 2, 2022, 1:24 AM Max Filippov <jcmvbkbc(a)gmail.com> wrote:
When uclibc is built with static PIE support the
_dl_load_base variable
shared between the libc-tls.c and reloc_static_pie.c creates the
dependency that requires linking reloc_static_pie.o into static
position-dependent executables resulting in the following build errors:
gcc -static test.c -o test
...ld:
...usr/lib/libc.a(reloc_static_pie.os):(.text+0x0):
undefined reference to `_DYNAMIC'
Move _dl_load_base definition to libc-tls.c to resolve this dependency
and fix static PDE build.
Signed-off-by: Max Filippov <jcmvbkbc(a)gmail.com>
---
libc/misc/internals/reloc_static_pie.c | 2 +-
libpthread/nptl/sysdeps/generic/libc-tls.c | 4 ++--
2 files changed, 3 insertions(+), 3 deletions(-)
diff --git a/libc/misc/internals/reloc_static_pie.c
b/libc/misc/internals/reloc_static_pie.c
index 578202d235a1..f66f7a4236c9 100644
--- a/libc/misc/internals/reloc_static_pie.c
+++ b/libc/misc/internals/reloc_static_pie.c
@@ -20,7 +20,7 @@
#include <elf.h>
#include <dl-elf.h>
-ElfW(Addr) _dl_load_base = NULL;
+extern ElfW(Addr) _dl_load_base;
void
reloc_static_pie (ElfW(Addr) load_addr);
diff --git a/libpthread/nptl/sysdeps/generic/libc-tls.c
b/libpthread/nptl/sysdeps/generic/libc-tls.c
index 0c8c558d05d3..e0e272e01dae 100644
--- a/libpthread/nptl/sysdeps/generic/libc-tls.c
+++ b/libpthread/nptl/sysdeps/generic/libc-tls.c
@@ -117,6 +117,7 @@ init_static_tls (size_t memsz, size_t align)
GL(dl_tls_static_nelem) = GL(dl_tls_max_dtv_idx);
}
+ElfW(Addr) _dl_load_base;
void __libc_setup_tls (size_t tcbsize, size_t tcbalign);
void
__libc_setup_tls (size_t tcbsize, size_t tcbalign)
@@ -143,8 +144,7 @@ __libc_setup_tls (size_t tcbsize, size_t tcbalign)
#else
initimage = (void *) phdr->p_vaddr;
#if !defined(SHARED) && defined(STATIC_PIE)
- extern ElfW(Addr) _dl_load_base;
- initimage += _dl_load_base;
+ initimage += _dl_load_base;
#endif
#endif
align = phdr->p_align;
--
2.30.2
_______________________________________________
devel mailing list -- devel(a)uclibc-ng.org
To unsubscribe send an email to devel-leave(a)uclibc-ng.org
10:38 a.m.
Hi linted,
thanks for your review.
On Thu, Sep 1, 2022 at 11:12 PM linted <linted90(a)gmail.com> wrote:
the original decision of placing it in
reloc_static_pie.c was to ensure
that when reloc_static_pie() accessed _dl_load_base it would do so
without relying on relocations.
I don't see how placing the variable definition in the source code changes
that: the linker will move it to the .bss section anyway and the function will
have to have code to access it there. Also by the time _dl_load_base is
used all relative relocations are done, so that shouldn't be the issue at all?
Do you mind letting me know which architecture you are
building on?
I'm adding static pie support to a new architecture: xtensa. It has a couple
of quirks here and there, so there will be more follow-up patches. I just
thought that this issue is rather non-arch specific to deal with it separately.
On one or two architectures I do make use of _DYNAMIC
in _start()
and I wonder if maybe that is leaking into static builds?
I'm looking at the linker-generated map of the failing build and see the
following lines in the list of archive members included to satisfy symbol
reference:
.../usr/lib/libc.a(reloc_static_pie.os)
.../usr/lib/libc.a(libc-tls.os) (_dl_load_base)
so clearly in my case it's the libc-tls.c reference to the _dl_load_base
that pulls reloc_static_pie.o in. Even if it didn't produce the build error
decoupling them is the right thing to do, because reloc_static_pie.o
has nothing to do in the PDE.
Also it doesn't look like you are 0 initializing
_dl_load_base in its new
location.
I don't. With zero-initializer it goes into the .bss section, without it it
goes to the common section. Either way it will be zero-initialized by
the kernel ELF loader.
--
Thanks.
-- Max
812
days inactive
812
days old
2 comments
2 participants
participants (2)
-
linted -
Max Filippov