BTI interaction between seccomp filters in systemd and glibc mprotect calls, causing service failures

[Topi Miettinen]

On 22.10.2020 10.54, Szabolcs Nagy wrote:
> The 10/21/2020 22:44, Jeremy Linton wrote:
>> There is a problem with glibc+systemd on BTI enabled systems. Systemd
>> has a service flag "MemoryDenyWriteExecute" which uses seccomp to deny
>> PROT_EXEC changes. Glibc enables BTI only on segments which are marked as
>> being BTI compatible by calling mprotect PROT_EXEC|PROT_BTI. That call is
>> caught by the seccomp filter, resulting in service failures.
>>
>> So, at the moment one has to pick either denying PROT_EXEC changes, or BTI.
>> This is obviously not desirable.
>>
>> Various changes have been suggested, replacing the mprotect with mmap calls
>> having PROT_BTI set on the original mapping, re-mmapping the segments,
>> implying PROT_EXEC on mprotect PROT_BTI calls when VM_EXEC is already set,
>> and various modification to seccomp to allow particular mprotect cases to
>> bypass the filters. In each case there seems to be an undesirable attribute
>> to the solution.
>>
>> So, whats the best solution?
> 
> the easy fix in glibc is to ignore mprotect(PROT_BTI|PROT_EXEC)
> failures, so programs work with seccomp filters, but bti gets
> disabled (it's unreasonable to expect bti protection if mprotect
> is filtered). it will be a nasty silent failure though.

Some may also want to use seccomp filters so that they will immediately 
kill the process and in this case they couldn't do it.

> and i'm also considering a fix that re-mmaps the executable
> segment with PROT_BTI instead of mprotect since that is not
> filtered. unfortunately the main exe is mmaped by the kernel
> without PROT_BTI and the libc does not have the fd to re-mmap.
> (bti can be left off for the main exe if mprotect fails and
> later we can teach the kernel to add bti there.) currently
> this is not a complete fix so i'm a bit hesitant about it.
> 
> as for a kernel side fix: if there is a way to only filter
> PROT_EXEC mprotect on mappings that are not yet PROT_EXEC
> that would solve this problem (but likely needs new syscall
> or seccomp capability).

Problem with seccomp MDWX is that it's still possible for malicious 
programs to circumvent the filter by using memfd_create(), fill the 
memory with desired content and then use mmap(,,PROT_EXEC) to make it 
executable without triggering seccomp. This can be mitigated by 
filtering also memfd_create(), but then some programs want to use it. 
Also the protection can be bypassed if the program can write to a file 
system which isn't mounted with "noexec". This can be mitigated with 
private mount namespaces and global mount options, but again some 
programs are written to expect W & X.

But I think SELinux has a more complete solution (execmem) which can 
track the pages better than is possible with seccomp solution which has 
a very narrow field of view. Maybe this facility could be made available 
to non-SELinux systems, for example with prctl()? Then the in-kernel 
MDWX could allow mprotect(PROT_EXEC | PROT_BTI) in case the backing file 
hasn't been modified, the source filesystem isn't writable for the 
calling process and the file descriptor isn't created with memfd_create().

-Topi