[PATCH] Fix __minimal_malloc segfaults in __mmap due to stack-protector

[Stefan Liebler]

On 16/12/2021 15:44, Adhemerval Zanella wrote:
> 
> 
> On 16/12/2021 11:20, Stefan Liebler via Libc-alpha wrote:
>> On 16/12/2021 12:54, Siddhesh Poyarekar wrote:
>>> On 12/16/21 17:17, Stefan Liebler via Libc-alpha wrote:
>>>> Starting with commit b05fae4d8e34604a72ee36d2d3164391b76fcf0b
>>>> "elf: Use the minimal malloc on tunables_strdup",
>>>> I get lots of segfaults in static tests on s390x when also using, e.g.:
>>>> export GLIBC_TUNABLES="glibc.elision.enable=1"
>>>>
>>>> tunables_strdup callls __minimal_malloc which tries to call __mmap
>>>> due to insufficient space left. __mmap itself first setups a new
>>>> stack frame and segfaults when copying the stack-protector canary
>>>> from thread-pointer. The latter one is not yet setup.
>>>>
>>>> Thus this patch also turns off stack-protection for mmap.
>>>
>>> LGTM.
>>>
>>> Reviewed-by: Siddhesh Poyarekar <siddhesh@sourceware.org>
>>>
>>>> ---
>>>>   misc/Makefile | 4 ++++
>>>>   1 file changed, 4 insertions(+)
>>>>
>>>> diff --git a/misc/Makefile b/misc/Makefile
>>>> index 3b66cb9f6a..db40312ba9 100644
>>>> --- a/misc/Makefile
>>>> +++ b/misc/Makefile
>>>> @@ -149,6 +149,10 @@ CFLAGS-sbrk.o = $(no-stack-protector)
>>>>   CFLAGS-sbrk.op = $(no-stack-protector)
>>>>   CFLAGS-brk.o = $(no-stack-protector)
>>>>   CFLAGS-brk.op = $(no-stack-protector)
>>>> +CFLAGS-mmap.o = $(no-stack-protector)
>>>> +CFLAGS-mmap.op = $(no-stack-protector)
>>>> +CFLAGS-mmap64.o = $(no-stack-protector)
>>>> +CFLAGS-mmap64.op = $(no-stack-protector)
>>>>     include ../Rules
>>>>  
>>>
>> Committed.
>> Thanks
> 
> 
> Thanks for catching it, I think I have not seeing it before because I got lucky
> the data segments has some slack space and not required called malloc.
> 
> Btw, do we still need to use no-stack-protector?
> 
At least for the new mmap files, this is needed.
Do you mean for [s]brk?