On Thu, 28 Aug 2025 at 13:48, Steven Rostedt <[email protected]> wrote:
>
> I could run it through the same hash algorithm that "%p" goes through so
> that it's not a real memory address.
For '%p', people can't easily trigger lots of different cases, and you
can't force kernel printouts from user space.
For something like tracing, user space *does* control the output, and
you shouldn't give people visibility into the hashing that '%p' does.
So you can certainly use siphash for hashing, but make sure to not use
the same secret key that the printing does.
As to the ID to hash, I actually think a 'struct file *' might be the
best thing to use - that's directly in the vma, no need to follow any
other pointers for it.
Linus
