On 11.03.25 13:32, Oleg Nesterov wrote:

On 03/11, David Hildenbrand wrote:

quoted

Regarding both questions, the code is fairly racy. Nothing would stop user
space from (a) modifying that memory

Yes, but we don't really care. uprobes.c assumes that user-space won't play
with the probed memory.

Right, I primarily care about that if user space would do it, that we 
don't trigger unintended behavior (e.g., overwriting pagecache pages 
etc, WARN etc).

Likely the re-validating the page content is indeed something we can drop.

Note that if is_register is false, then vma can be even writable. Hmm, why?
Perhaps valid_vma() should ignore is_register and nack VM_MAYWRITE ? But
this doesn't really matter, say, gdb can change this memory anyway. Again,
we don't really care.

quoted

quoted

do something like

	/* Walk the page tables again, to perform the actual update. */
	ret = -EAGAIN;
	folio = folio_walk_start(&fw, vma, vaddr, 0);
	if (folio) {
		if (fw.page == page) {
			WARN_ON(is_register && !folio_test_anon(folio));

Yes, that would work (we could leave the WARN_ON in __uprobe_write_opcode),
but I am not sure if the end result is better better. No strong opinion on
the details though.

Will, this way __uprobe_write_opcode() will look a little bit simpler...

But I won't insist, please do what you think is better.

I'll take another look at this series probably next week (I'm on PTO 
this week) to then resend once adjusted + retested.

Thanks!

-- 
Cheers,

David / dhildenb