May 2021 - devel - example.com

[PATCH 0/1] V3 fix powerpc PIE/PIC + secure-plt builds
by Yann Sionneau 31 May '21

31 May '21

This is the 3rd iteration of the patch, here are the changes with regard to previous one: * more precise description of the problem. * only set r30 in PIC+secureplt case as it seems indeed to be the only case which causes problem. Secureplt without PIC does not need r30. * don't rely on uclibc being already built: now we check for secure-plt by doing $(CC) --verbose and searching for the ./configure parameters of GCC for "--enable-secureplt". Yann Sionneau (1): powerpc: fix PIE/PIC builds with newer gcc/binutils which use secureplt by default Rules.mak | 3 ++- ldso/ldso/powerpc/dl-startup.h | 3 +++ libc/sysdeps/linux/powerpc/crt1.S | 4 ++++ 3 files changed, 9 insertions(+), 1 deletion(-) -- 2.25.1

2 2

Race condition in LinuxThread's sigaction wrapper
by Vladimir Murzin 25 May '21

25 May '21

Hi I got report about race condition in LinuxThread's sigaction wrapper located in libpthread/linuxthreads/signals.c In sigaction wrapper A.1) override handler for new action to point at pthread_sighandler A.2) __libc_sigaction A.3) update global array (sighandler) used by pthread_sighandler with real action (from new) pthread_sighandler B.1) find real action (indexed by signal number) B.2) unconditionally execute action IIUC, if we receive signal in a window between A.2 and A.3 we are missing executing real action (likely we try to execute NULL). Does that make sense? Cheers Vladimir

2 3

[PATCH] powerpc: fix builds with newer gcc/binutils which use secureplt by default
by yann＠sionneau.net 25 May '21

25 May '21

From: Yann Sionneau <yann(a)sionneau.net> This patch fixes segfault of all user space processes (including init, which caused a panic) on recent buildroot powerpc32 builds. Recent buildroot toolchain enables secure PLT in powerpc gcc. The latter will then supply -msecure-plt to gas invocations by default. For the secure PLT to work, the r30 register needs to point to the GOT. Old "bss plt" was just a one-instruction-wide PLT slot, pointed-to by a R_PPC_JMP_SLOT relocation, which was written on-the-fly to contain a branch instruction to the correct address. It therefore had to stay writable. New secure PLT only contains read-only code which loads the branch address from the writable GOT. Signed-off-by: Yann Sionneau <yann(a)sionneau.net> --- Rules.mak | 4 +++- ldso/ldso/powerpc/dl-startup.h | 3 +++ libc/sysdeps/linux/powerpc/crt1.S | 5 ++++- 3 files changed, 10 insertions(+), 2 deletions(-) diff --git a/Rules.mak b/Rules.mak index 1fa09be23..0ab41e800 100644 --- a/Rules.mak +++ b/Rules.mak @@ -58,6 +58,7 @@ LD = $(CROSS_COMPILE)ld NM = $(CROSS_COMPILE)nm OBJDUMP = $(CROSS_COMPILE)objdump STRIPTOOL = $(CROSS_COMPILE)strip +READELF = $(CROSS_COMPILE)readelf INSTALL = install LN = ln @@ -483,9 +484,10 @@ ifeq ($(TARGET_ARCH),powerpc) PICFLAG:=-fpic PIEFLAG_NAME:=-fpie PPC_HAS_REL16:=$(shell printf "\t.text\n\taddis 11,30,_GLOBAL_OFFSET_TABLE_-.@ha\n" | $(CC) -c -x assembler -o /dev/null - 2> /dev/null && echo -n y || echo -n n) + PPC_HAS_SECUREPLT:=$(shell tmpfile=$$(mktemp); printf '\#include <stdio.h>\nint main(void) { puts("hello"); return 0;}' > $${tmpfile}.c; $(CC) -x c $${tmpfile}.c -o $${tmpfile}; $(READELF) -d $${tmpfile} | grep PPC_GOT > /dev/null && echo -n y || echo -n n; rm $${tmpfile}.c $${tmpfile}) + CPU_CFLAGS-$(PPC_HAS_SECUREPLT) += -DPPC_HAS_SECUREPLT CPU_CFLAGS-$(PPC_HAS_REL16)+= -DHAVE_ASM_PPC_REL16 CPU_CFLAGS-$(CONFIG_E500) += "-D__NO_MATH_INLINES" - endif ifeq ($(TARGET_ARCH),bfin) diff --git a/ldso/ldso/powerpc/dl-startup.h b/ldso/ldso/powerpc/dl-startup.h index 8b2a517e2..7749395eb 100644 --- a/ldso/ldso/powerpc/dl-startup.h +++ b/ldso/ldso/powerpc/dl-startup.h @@ -25,6 +25,9 @@ __asm__( #else " bl _GLOBAL_OFFSET_TABLE_-4@local\n" /* Put our GOT pointer in r31, */ " mflr 31\n" +#endif +#ifdef PPC_HAS_SECUREPLT + " mr 30,31\n" #endif " addi 1,1,16\n" /* Restore SP */ " lwz 7,_dl_skip_args@got(31)\n" /* load EA of _dl_skip_args */ diff --git a/libc/sysdeps/linux/powerpc/crt1.S b/libc/sysdeps/linux/powerpc/crt1.S index 27bfc5a5a..78b946ad6 100644 --- a/libc/sysdeps/linux/powerpc/crt1.S +++ b/libc/sysdeps/linux/powerpc/crt1.S @@ -47,7 +47,7 @@ _start: mr r9,r1 /* Save the stack pointer and pass it to __uClibc_main */ clrrwi r1,r1,4 /* Align stack ptr to 16 bytes */ -#ifdef __PIC__ +#if defined(__PIC__) || defined(PPC_HAS_SECUREPLT) # ifdef HAVE_ASM_PPC_REL16 bcl 20,31,1f 1: mflr r31 @@ -57,6 +57,9 @@ _start: bl _GLOBAL_OFFSET_TABLE_-4@local mflr r31 # endif +#endif +#ifdef PPC_HAS_SECUREPLT + mr 30,31 #endif /* Set up the small data pointer in r13. */ #ifdef __PIC__ -- 2.25.1

3 2

[Powerpc] HAVE_ASM_PPC_REL16 and BR2_PIC_PIE
by Romain Naour 22 May '21

22 May '21

Hello, Recently in Buildroot the option BR2_PIC_PIE has been enabled by default along with other hardening features [1]. Since then some ppc defconfig such qemu_ppc_e500mc_defconfig are failing to boot due to a segfault in init program. The segfault appear very early in __uClibc_main while starting any binaries, an issue located in crt1.S (powerpc)[2]. After some trial and error, removing HAVE_ASM_PPC_REL16 from CFLAGS [3] allow to generate a working system again. But this is actually wrong since instead we should consider HAVE_ASM_PPC_REL16 always true nowadays. Indeed, Glibc removed HAVE_ASM_PPC_REL16 since version 2.22 [4] since "the minimum binutils supports rel16 relocs". Binutils 2.22 supports R_PPC_REL16 as default. uClibc-ng should remove HAVE_ASM_PPC_REL16 but keep the code as it was defined. But this doesn't fix the initial issue. Any idea ? [1] https://git.buildroot.net/buildroot/commit/?id=810ba387bec3c5b6904e8893fb4c… [2] https://cgit.uclibc-ng.org/cgi/cgit/uclibc-ng.git/tree/libc/sysdeps/linux/p… [3] https://cgit.uclibc-ng.org/cgi/cgit/uclibc-ng.git/tree/Rules.mak?id=2bf4991… [4] https://sourceware.org/git/?p=glibc.git;a=commit;h=59261ad3eb345e0d7b9f5c73… Best regards, Romain

6 14