1. Patchew
  2. linux
  3. View series

[PATCH] lkdtm: Disable return thunks in rodata.c

Josh Poimboeuf posted 1 patch 3 years, 9 months ago
There is a newer version of this series
arch/x86/Makefile           | 1 +
drivers/misc/lkdtm/Makefile | 9 ++++++---
2 files changed, 7 insertions(+), 3 deletions(-)
[PATCH] lkdtm: Disable return thunks in rodata.c
Posted by Josh Poimboeuf 3 years, 9 months ago 
The following warning was seen:

  WARNING: CPU: 0 PID: 0 at arch/x86/kernel/alternative.c:557 apply_returns (arch/x86/kernel/alternative.c:557 (discriminator 1))
  Modules linked in:
  CPU: 0 PID: 0 Comm: swapper/0 Not tainted 5.19.0-rc4-00008-gee88d363d156 #1
  Hardware name: QEMU Standard PC (i440FX + PIIX, 1996), BIOS 1.16.0-debian-1.16.0-4 04/01/2014
  RIP: 0010:apply_returns (arch/x86/kernel/alternative.c:557 (discriminator 1))
  Code: ff ff 74 cb 48 83 c5 04 49 39 ee 0f 87 81 fe ff ff e9 22 ff ff ff 0f 0b 48 83 c5 04 49 39 ee 0f 87 6d fe ff ff e9 0e ff ff ff <0f> 0b 48 83 c5 04 49 39 ee 0f 87 59 fe ff ff e9 fa fe ff ff 48 89

The warning happened when apply_returns() failed to convert "JMP
__x86_return_thunk" to RET.  It was instead a JMP to nowhere, due to the
thunk relocation not getting resolved.

That rodata.o code is objcopy'd to .rodata, and later memcpy'd, so
relocations don't work (and are apparently silently ignored).

LKDTM is only used for testing, so the naked RET should be fine.  So
just disable return thunks for that file.

While at it, disable objtool and KCSAN for the file.

Fixes: 0b53c374b9ef ("x86/retpoline: Use -mfunction-return")
Reported-by: kernel test robot <oliver.sang@intel.com>
Link: https://lore.kernel.org/lkml/Ys58BxHxoDZ7rfpr@xsang-OptiPlex-9020/
Debugged-by: Peter Zijlstra <peterz@infradead.org>
Signed-off-by: Josh Poimboeuf <jpoimboe@kernel.org>
---
 arch/x86/Makefile           | 1 +
 drivers/misc/lkdtm/Makefile | 9 ++++++---
 2 files changed, 7 insertions(+), 3 deletions(-)

diff --git a/arch/x86/Makefile b/arch/x86/Makefile
index 1f40dad30d50..7854685c5f25 100644
--- a/arch/x86/Makefile
+++ b/arch/x86/Makefile
@@ -27,6 +27,7 @@ RETHUNK_CFLAGS		:= -mfunction-return=thunk-extern
 RETPOLINE_CFLAGS	+= $(RETHUNK_CFLAGS)
 endif
 
+export RETHUNK_CFLAGS
 export RETPOLINE_CFLAGS
 export RETPOLINE_VDSO_CFLAGS
 
diff --git a/drivers/misc/lkdtm/Makefile b/drivers/misc/lkdtm/Makefile
index 2e0aa74ac185..95ef971b5e1c 100644
--- a/drivers/misc/lkdtm/Makefile
+++ b/drivers/misc/lkdtm/Makefile
@@ -13,10 +13,13 @@ lkdtm-$(CONFIG_LKDTM)		+= cfi.o
 lkdtm-$(CONFIG_LKDTM)		+= fortify.o
 lkdtm-$(CONFIG_PPC_64S_HASH_MMU)	+= powerpc.o
 
-KASAN_SANITIZE_rodata.o		:= n
 KASAN_SANITIZE_stackleak.o	:= n
-KCOV_INSTRUMENT_rodata.o	:= n
-CFLAGS_REMOVE_rodata.o		+= $(CC_FLAGS_LTO)
+
+KASAN_SANITIZE_rodata.o			:= n
+KCSAN_SANITIZE_rodata.o			:= n
+KCOV_INSTRUMENT_rodata.o		:= n
+OBJECT_FILES_NON_STANDARD_rodata.o	:= y
+CFLAGS_REMOVE_rodata.o			+= $(CC_FLAGS_LTO) $(RETHUNK_CFLAGS)
 
 OBJCOPYFLAGS :=
 OBJCOPYFLAGS_rodata_objcopy.o	:= \
-- 
2.36.1
Re: [PATCH] lkdtm: Disable return thunks in rodata.c
Posted by Kees Cook 3 years, 9 months ago 
On Mon, Jul 18, 2022 at 07:50:25AM -0700, Josh Poimboeuf wrote:
> The following warning was seen:
> 
>   WARNING: CPU: 0 PID: 0 at arch/x86/kernel/alternative.c:557 apply_returns (arch/x86/kernel/alternative.c:557 (discriminator 1))
>   Modules linked in:
>   CPU: 0 PID: 0 Comm: swapper/0 Not tainted 5.19.0-rc4-00008-gee88d363d156 #1
>   Hardware name: QEMU Standard PC (i440FX + PIIX, 1996), BIOS 1.16.0-debian-1.16.0-4 04/01/2014
>   RIP: 0010:apply_returns (arch/x86/kernel/alternative.c:557 (discriminator 1))
>   Code: ff ff 74 cb 48 83 c5 04 49 39 ee 0f 87 81 fe ff ff e9 22 ff ff ff 0f 0b 48 83 c5 04 49 39 ee 0f 87 6d fe ff ff e9 0e ff ff ff <0f> 0b 48 83 c5 04 49 39 ee 0f 87 59 fe ff ff e9 fa fe ff ff 48 89
> 
> The warning happened when apply_returns() failed to convert "JMP
> __x86_return_thunk" to RET.  It was instead a JMP to nowhere, due to the
> thunk relocation not getting resolved.
> 
> That rodata.o code is objcopy'd to .rodata, and later memcpy'd, so
> relocations don't work (and are apparently silently ignored).
> 
> LKDTM is only used for testing, so the naked RET should be fine.  So
> just disable return thunks for that file.
> 
> While at it, disable objtool and KCSAN for the file.
> 
> Fixes: 0b53c374b9ef ("x86/retpoline: Use -mfunction-return")
> Reported-by: kernel test robot <oliver.sang@intel.com>
> Link: https://lore.kernel.org/lkml/Ys58BxHxoDZ7rfpr@xsang-OptiPlex-9020/
> Debugged-by: Peter Zijlstra <peterz@infradead.org>
> Signed-off-by: Josh Poimboeuf <jpoimboe@kernel.org>

Thanks!

Acked-by: Kees Cook <keescook@chromium.org>

-Kees

> ---
>  arch/x86/Makefile           | 1 +
>  drivers/misc/lkdtm/Makefile | 9 ++++++---
>  2 files changed, 7 insertions(+), 3 deletions(-)
> 
> diff --git a/arch/x86/Makefile b/arch/x86/Makefile
> index 1f40dad30d50..7854685c5f25 100644
> --- a/arch/x86/Makefile
> +++ b/arch/x86/Makefile
> @@ -27,6 +27,7 @@ RETHUNK_CFLAGS		:= -mfunction-return=thunk-extern
>  RETPOLINE_CFLAGS	+= $(RETHUNK_CFLAGS)
>  endif
>  
> +export RETHUNK_CFLAGS
>  export RETPOLINE_CFLAGS
>  export RETPOLINE_VDSO_CFLAGS
>  
> diff --git a/drivers/misc/lkdtm/Makefile b/drivers/misc/lkdtm/Makefile
> index 2e0aa74ac185..95ef971b5e1c 100644
> --- a/drivers/misc/lkdtm/Makefile
> +++ b/drivers/misc/lkdtm/Makefile
> @@ -13,10 +13,13 @@ lkdtm-$(CONFIG_LKDTM)		+= cfi.o
>  lkdtm-$(CONFIG_LKDTM)		+= fortify.o
>  lkdtm-$(CONFIG_PPC_64S_HASH_MMU)	+= powerpc.o
>  
> -KASAN_SANITIZE_rodata.o		:= n
>  KASAN_SANITIZE_stackleak.o	:= n
> -KCOV_INSTRUMENT_rodata.o	:= n
> -CFLAGS_REMOVE_rodata.o		+= $(CC_FLAGS_LTO)
> +
> +KASAN_SANITIZE_rodata.o			:= n
> +KCSAN_SANITIZE_rodata.o			:= n
> +KCOV_INSTRUMENT_rodata.o		:= n
> +OBJECT_FILES_NON_STANDARD_rodata.o	:= y
> +CFLAGS_REMOVE_rodata.o			+= $(CC_FLAGS_LTO) $(RETHUNK_CFLAGS)
>  
>  OBJCOPYFLAGS :=
>  OBJCOPYFLAGS_rodata_objcopy.o	:= \
> -- 
> 2.36.1
> 

-- 
Kees Cook

Patchew 2.1-devel-ea86937

© 2016 - 2026 Red Hat, Inc.