linux-kernel - Re: [PATCH] x86/mm: Do not verify W^X at boot up

lists.openwall.net		lists / announce owl-users owl-dev john-users john-dev passwdqc-users yescrypt popa3d-users / oss-security kernel-hardening musl sabotage tlsify passwords / crypt-dev xvendor / Bugtraq Full-Disclosure linux-kernel linux-netdev linux-ext4 linux-hardening linux-cve-announce PHC
Open Source and information security mailing list archives

Hash Suite: Windows password security audit tool. GUI, reports in PDF.

[<prev] [next>] [<thread-prev] [thread-next>] [day] [month] [year] [list]

Date:   Mon, 24 Oct 2022 12:08:49 -0700
From:   Linus Torvalds <torvalds@...ux-foundation.org>
To:     Steven Rostedt <rostedt@...dmis.org>
Cc:     LKML <linux-kernel@...r.kernel.org>,
        Peter Zijlstra <peterz@...radead.org>,
        Kees Cook <keescook@...omium.org>,
        Dave Hansen <dave.hansen@...el.com>,
        Sean Christopherson <seanjc@...gle.com>
Subject: Re: [PATCH] x86/mm: Do not verify W^X at boot up

On Mon, Oct 24, 2022 at 11:52 AM Steven Rostedt <rostedt@...dmis.org> wrote:
>
> It's not just speed up at boot up. It's because text_poke doesn't work at
> early boot up when function tracing is enabled. If I remove the
> SYSTEM_BOOTING checks in text_poke() this happens:

Let's just fix that up.

>
> [    1.013753] BUG: kernel NULL pointer dereference, address: 0000000000000048

This is due to

__get_locked_pte:
   0: 0f 1f 44 00 00        nopl   0x0(%rax,%rax,1)
   5: 48 89 f0              mov    %rsi,%rax
   8: 41 55                push   %r13
   a: 48 c1 e8 24          shr    $0x24,%rax
   e: 41 54                push   %r12
  10: 25 f8 0f 00 00        and    $0xff8,%eax
  15: 55                    push   %rbp
  16: 53                    push   %rbx
  17:* 48 03 47 48          add    0x48(%rdi),%rax <-- trapping instruction
  1b: 0f 84 c4 00 00 00    je     0xe5
  21: 49 89 fc              mov    %rdi,%r12
  24: 48 8b 38              mov    (%rax),%rdi
  27: 48 89 f3              mov    %rsi,%rbx
  2a: 48 89 c5              mov    %rax,%rbp

and that 'addq' seems to be walk_to_pmd() doing

        pgd = pgd_offset(mm, addr);

with a zero mm (that's 'mm->pgd').

And that, in turn, seems to be due to the absolutely disgusing 'poking_mm' hack.

> Interrupts haven't been enabled yet, so things are still rather fragile at
> this point of start up.

I don't think this has anything to do with interrupts. We do need the
page structures etc to be workable, but all the tracing setup needs
that *anyway*.

I suspect it would be fixed by just moving 'poking_init()' earlier. In
many ways I suspect it would make most sense as part of 'mm_init()',
not as a random call fairly late in start_kernel().

In other words, this all smells like "people added special cases
because they didn't want to hunt down the underlying problem".

And then all these special cases beget other special cases.

                 Linus