Re: [PATCH v4 4/9] ring-buffer: Fix subbuf resize race in ring_buffer_alloc_read_page()

Thank you for your contribution! Sashiko AI review found 1 potential issue(s) to consider:

Pre-existing issues:
- [High] The patch is an incomplete fix for a race condition in `ring_buffer_alloc_read_page()`. While it ensures newly allocated pages use `bpage->order`, it entirely misses the branch where `cpu_buffer->free_page` is reused. This allows an undersized cached page to be paired with a newer, larger `bpage->order`, leading to heap buffer overflows during reads and out-of-bounds frees when the page is released.
--

--- Patch [4]: [PATCH v4 4/9] ring-buffer: Fix subbuf resize race in ring_buffer_alloc_read_page() ---
commit 28754c82212e5979b25cf6fac815748ff021a277
Author: Vincent Donnefort <[email protected]>

ring-buffer: Fix subbuf resize race in ring_buffer_alloc_read_page()

ring_buffer_alloc_read_page() is racy with ring_buffer_subbuf_order_set,
it can allocate a reader page with an outdated order. This isn't a big
issue, the user can still re-allocate a new reader page and try again.

However, what is more problematic is if the value of subbuf_order
changes in the middle of ring_buffer_alloc_read_page(). In that case,
bpage->order might not match the actual allocated memory.

Use bpage->order for the allocation to prevent this race.

> diff --git a/kernel/trace/ring_buffer.c b/kernel/trace/ring_buffer.c
> index 97449423d3a68..94552a433228d 100644
> --- a/kernel/trace/ring_buffer.c
> +++ b/kernel/trace/ring_buffer.c
> @@ -7018,7 +7018,7 @@ ring_buffer_alloc_read_page(struct trace_buffer *buffer, int cpu)
>  	if (bpage->data) {
>  		rb_init_data_page(bpage->data);

[Severity: High]
This is a pre-existing issue, but does this path leave bpage->order
mismatched with the actual allocation size of cpu_buffer->free_page?

If ring_buffer_subbuf_order_set() updates buffer->subbuf_order to a larger
size before clearing cpu_buffer->free_page, ring_buffer_alloc_read_page()
might read the new larger order into bpage->order, but then claim an older,
smaller cpu_buffer->free_page:

kernel/trace/ring_buffer.c:ring_buffer_alloc_read_page() {
    ...
	bpage->order = order;
    ...
	if (cpu_buffer->free_page) {
		bpage->data = cpu_buffer->free_page;
		cpu_buffer->free_page = NULL;
	}
    ...
}

Could this lead to an overflow of dpage->data when reading events?

kernel/trace/ring_buffer.c:ring_buffer_read_page() {
    ...
			memcpy(dpage->data + pos, rpage->data + rpos, event_size);
    ...
}

It also appears this could cause an out-of-bounds free of bpage->data when
the page is released, since it frees using the mismatched bpage->order
instead of the page's actual allocation order.

Should we also verify the cached page's actual order matches bpage->order
before claiming it?

>  	} else {
> -		bpage->data = alloc_cpu_data(cpu, cpu_buffer->buffer->subbuf_order);
> +		bpage->data = alloc_cpu_data(cpu, bpage->order);
>  		if (!bpage->data) {
>  			kfree(bpage);
>  			return ERR_PTR(-ENOMEM);

-- 
Sashiko AI review · https://sashiko.dev/#/patchset/[email protected]?part=4