Re: [GIT PULL] smb3 client fixes

[Linus Torvalds]

On Fri, 6 Mar 2026 at 14:04, Steve French <smfrench@xxxxxxxxx> wrote:
>
> - Security fix

Bah. I had to look this up - I don't think it's an actual security
issue, just a (good) cleanup.

Yes, yes, the old code did "memcmp" instead of "crypto_memneq". And
yes, it's in theory timing-sensitive.

But the two compares were of size 8 and 16, and at least clang
generates a constant-time comparison for that anyway (I bet gcc does
too):

This is the 16-byte case:

        movq    (%rdi), %rax
        movq    8(%rdi), %rcx
        xorq    (%rsi), %rax
        xorq    8(%rsi), %rcx
        orq     %rax, %rcx
        je      ...

and the 8-byte case is even more trivially constant-time.

And I'm sure that you can get the compiler to generate garbage that is
timing-sensitive by enabling some debug options that make code quality
much worse, but my point is that as an explanation for the pull
request, that "security fix" was just not a great explanation
regardless.

I'd argue that half of the other fixes were *more* relevant for
security than this one. An oops is a bigger security issue than some
purely theoretical timing attack that is almost certainly not
practical even *if* that code had done the mindless byte-at-a-time
comparison.

              Linus