On Thu, May 20, 2021 at 08:40:28AM +1000, Dave Chinner wrote:

This will not apply (and break) the bulk alloc patch I sent out - we
have to ensure that the b_pages array is always zeroed before we
call the bulk alloc function, hence I moved the memset() in this
function to be unconditional. I almost cleaned up this function in
that patchset....

The buffer is freshly allocated here using kmem_cache_zalloc, so
b_pages can't be set, b_page_array is already zeroed from
kmem_cache_zalloc, and the separate b_pages allocation is swithced
to use kmem_zalloc.  I thought the commit log covers this, but maybe
I need to improve it?