[PATCH] ARM: mm: flip priority of CONFIG_DEBUG_RODATA
labbott at redhat.com
Mon Nov 30 17:20:35 PST 2015
On 11/30/2015 05:08 PM, Kees Cook wrote:
> On Mon, Nov 30, 2015 at 5:03 PM, Laura Abbott <labbott at redhat.com> wrote:
>> On 11/30/2015 03:38 PM, Kees Cook wrote:
>>> Given the choice between making things NX or making things RO, we want
>>> RO first. As such, redefine CONFIG_DEBUG_RODATA to actually do the bulk
>> Can you give a citation for why? The thread that inspired it might be
>> a good link.
> This was inspired by my examining the existing architecture's
> implementations of CONFIG_DEBUG_RODATA after Ingo suggested it be made
> a common feature not a build-time config (or at least renamed):
Thanks. I read the thread and I think it would be good to put a link
in the commit message to make it clearer why this is going in.
>>> index 41218867a9a6..b617084e9520 100644
>>> --- a/arch/arm/mm/Kconfig
>>> +++ b/arch/arm/mm/Kconfig
>>> @@ -1039,24 +1039,26 @@ config ARCH_SUPPORTS_BIG_ENDIAN
>>> This option specifies the architecture can support big endian
>>> -config ARM_KERNMEM_PERMS
>>> - bool "Restrict kernel memory permissions"
>>> +config DEBUG_RODATA
>>> + bool "Make kernel text and rodata read-only"
>>> depends on MMU
>>> + default y if CPU_V7
>>> - If this is set, kernel memory other than kernel text (and
>>> - will be made non-executable. The tradeoff is that each region is
>>> - padded to section-size (1MiB) boundaries (because their
>>> - are different and splitting the 1M pages into 4K ones causes TLB
>>> - performance problems), wasting memory.
>>> + If this is set, kernel memory (text, rodata, etc) will be made
>>> + read-only, and non-text kernel memory will be made
>>> + The tradeoff is that each region is padded to section-size
>>> + boundaries (because their permissions are different and
>>> + the 1M pages into 4K ones causes TLB performance problems),
>>> + can waste memory.
>>> -config DEBUG_RODATA
>>> - bool "Make kernel text and rodata read-only"
>>> - depends on ARM_KERNMEM_PERMS
>>> +config DEBUG_ALIGN_RODATA
>>> + bool "Make rodata strictly non-executable"
>>> + depends on DEBUG_RODATA
>>> default y
>>> - If this is set, kernel text and rodata will be made read-only.
>>> - is to help catch accidental or malicious attempts to change the
>>> - kernel's executable code. Additionally splits rodata from kernel
>>> - text so it can be made explicitly non-executable. This creates
>>> - another section-size padded region, so it can waste more memory
>>> - space while gaining the read-only protections.
>>> + If this is set, rodata will be made explicitly non-executable.
>>> + provides protection on the rare chance that attackers might find
>>> + use ROP gadgets that exist in the rodata section. This adds an
>>> + additional section-aligned split of rodata from kernel text so
>>> + can be made explicitly non-executable. This padding may waste
>>> + space to gain this additional protection.
>> I get that you want to make this match arm64 but it's really not intuitive that
>> something with ALIGN_RODATA in the name is actually for setting NX. The purpose
>> of ALIGN_RODATA was also slightly different on arm64 since the RO/NX will still
>> be there, the difference is if the sections are present versus broken down into
> Well, it seems to have the same effect: without the alignment, a
> portion of rodata may remain executable on arm64. Unless I
No, on arm64 everything should always be NX, the difference is part of the NX
sections may be mapped as pages instead of sections so you take the TLB hit.
It's a trade off of memory vs TLB pressure instead of just security vs TLB.
More information about the linux-arm-kernel