[PATCH RFC 2/4] mm, personality: Implement memory-deny-write-execute as a personality flag

[David Hildenbrand]

On 22.04.22 12:28, Catalin Marinas wrote:
> On Thu, Apr 21, 2022 at 06:37:49PM +0100, David Hildenbrand wrote:
>> On 13.04.22 15:49, Catalin Marinas wrote:
>>> The aim of such policy is to prevent a user task from inadvertently
>>> creating an executable mapping that is or was writeable (and
>>> subsequently made read-only).
>>>
>>> An example of mmap() returning -EACCESS if the policy is enabled:
>>>
>>> 	mmap(0, size, PROT_READ | PROT_WRITE | PROT_EXEC, flags, 0, 0);
>>>
>>> Similarly, mprotect() would return -EACCESS below:
>>>
>>> 	addr = mmap(0, size, PROT_READ | PROT_EXEC, flags, 0, 0);
>>> 	mprotect(addr, size, PROT_READ | PROT_WRITE | PROT_EXEC);
>>>
>>> With the past vma writeable permission tracking, mprotect() below would
>>> also fail with -EACCESS:
>>>
>>> 	addr = mmap(0, size, PROT_READ | PROT_WRITE, flags, 0, 0);
>>> 	mprotect(addr, size, PROT_READ | PROT_EXEC);
>>>
>>> While the above could be achieved by checking PROT_WRITE & PROT_EXEC on
>>> mmap/mprotect and denying mprotect(PROT_EXEC) altogether (current
>>> systemd MDWE approach via SECCOMP BPF filters), we want the following
>>> scenario to succeed:
>>>
>>> 	addr = mmap(0, size, PROT_READ | PROT_EXEC, flags, 0, 0);
>>> 	mprotect(addr, size, PROT_READ | PROT_EXEC | PROT_BTI);
>>>
>>> where PROT_BTI enables branch tracking identification on arm64.
>>>
>>> The choice for a DENY_WRITE_EXEC personality flag, inherited on fork()
>>> and execve(), was made by analogy to READ_IMPLIES_EXEC.
>>>
>>> Note that it is sufficient to check for VM_WAS_WRITE in
>>> map_deny_write_exec() as this flag is always set on VM_WRITE mappings.
>>>
>>> Signed-off-by: Catalin Marinas <catalin.marinas at arm.com>
>>> Cc: Christoph Hellwig <hch at infradead.org>
>>> Cc: Andrew Morton <akpm at linux-foundation.org>
>>
>> How does this interact with get_user_pages(FOLL_WRITE|FOLL_FORCE) on a
>> VMA that is VM_MAYWRITE but not VM_WRITE? Is it handled accordingly?
> 
> For now, that's just about VM_WRITE. Most vmas are VM_MAYWRITE, so we
> can't really have MAYWRITE^EXEC. The basic feature aims to avoid user
> vulnerabilities where a buffer is mapped both writeable and executable.
> Of course, it can be expanded with additional prctl() flags to cover
> other cases.
> 
>> Note that in the (FOLL_WRITE|FOLL_FORCE) we only require VM_MAYWRITE on
>> the vma and trigger a write fault. As the VMA is not VM_WRITE, we won't
>> actually map the PTE writable, but set it dirty. GUP will retry, find a
>> R/O pte that is dirty and where it knows that it broke COW and will
>> allow the read access, although the PTE is R/O.
>>
>> That mechanism is required to e.g., set breakpoints in R/O MAP_PRIVATE
>> kernel sections, but it's used elsewhere for page pinning as well.
>>
>> My gut feeling is that GUP(FOLL_WRITE|FOLL_FORCE) could be used right
>> now to bypass that mechanism, I might be wrong.
> 
> GUP can be used to bypass this. But if an attacker can trigger such GUP
> paths via a syscall (e.g. ptrace(PTRACE_POKEDATA)), I think we need the
> checks on those paths (and reject the syscall) rather than on
> mmap/mprotect(). This would be covered by something like CAP_SYS_PTRACE.
> 
> 

I was told that RDMA uses FOLL_FORCE|FOLL_WRITE and is available to
unprivileged users.

-- 
Thanks,

David / dhildenb