linux-kernel - Re: [PATCH v1 06/11] mm: support GUP-triggered unsharing via FAULT_FLAG

lists.openwall.net		lists / announce owl-users owl-dev john-users john-dev passwdqc-users yescrypt popa3d-users / oss-security kernel-hardening musl sabotage tlsify passwords / crypt-dev xvendor / Bugtraq Full-Disclosure linux-kernel linux-netdev linux-ext4 linux-hardening linux-cve-announce PHC
Open Source and information security mailing list archives

Hash Suite: Windows password security audit tool. GUI, reports in PDF.

[<prev] [next>] [<thread-prev] [thread-next>] [day] [month] [year] [list]

Message-ID: <CAHk-=whUE9RdzXcsQ6VrD5fNU3BtXWjhSa4XPy7bZQTHhdr3Rg@mail.gmail.com>
Date:   Mon, 20 Dec 2021 11:38:55 -0800
From:   Linus Torvalds <torvalds@...ux-foundation.org>
To:     Matthew Wilcox <willy@...radead.org>
Cc:     David Hildenbrand <david@...hat.com>,
        Nadav Amit <namit@...are.com>,
        Jason Gunthorpe <jgg@...dia.com>,
        Linux Kernel Mailing List <linux-kernel@...r.kernel.org>,
        Andrew Morton <akpm@...ux-foundation.org>,
        Hugh Dickins <hughd@...gle.com>,
        David Rientjes <rientjes@...gle.com>,
        Shakeel Butt <shakeelb@...gle.com>,
        John Hubbard <jhubbard@...dia.com>,
        Mike Kravetz <mike.kravetz@...cle.com>,
        Mike Rapoport <rppt@...ux.ibm.com>,
        Yang Shi <shy828301@...il.com>,
        "Kirill A . Shutemov" <kirill.shutemov@...ux.intel.com>,
        Vlastimil Babka <vbabka@...e.cz>, Jann Horn <jannh@...gle.com>,
        Michal Hocko <mhocko@...nel.org>,
        Rik van Riel <riel@...riel.com>,
        Roman Gushchin <guro@...com>,
        Andrea Arcangeli <aarcange@...hat.com>,
        Peter Xu <peterx@...hat.com>,
        Donald Dutile <ddutile@...hat.com>,
        Christoph Hellwig <hch@....de>,
        Oleg Nesterov <oleg@...hat.com>, Jan Kara <jack@...e.cz>,
        Linux-MM <linux-mm@...ck.org>,
        "open list:KERNEL SELFTEST FRAMEWORK" 
        <linux-kselftest@...r.kernel.org>,
        "open list:DOCUMENTATION" <linux-doc@...r.kernel.org>
Subject: Re: [PATCH v1 06/11] mm: support GUP-triggered unsharing via
 FAULT_FLAG_UNSHARE (!hugetlb)

On Mon, Dec 20, 2021 at 10:53 AM Matthew Wilcox <willy@...radead.org> wrote:
>
> It makes me wonder if reuse_swap_page() can also be based on refcount
> instead of mapcount?

I suspect it doesn't even need refcount.

For regular pages, after we've copied the page, all we do right now is

                if (page_copied)
                        free_swap_cache(old_page);

which is basically just an optimistic trylock_page() followed by
try_to_free_swap().

And that then pretty much simply checks "are there any swap users
left" and deletes it from the swap cache if not.

The "free_swap_cache()" thing is actually just an optimization to
avoid having memory pressure do it later.  So it doesn't have to be
exact.

In fact, I thought that swap is so unusual that it's not even needed
at all, but I was wrong. See how this was re-introduced in commit
f4c4a3f48480 ("mm: free idle swap cache page after COW") because yes,
some loads still have swap space allocated.

In theory, it would probably be a good idea at COW time to see if the
page ref is 2, and if it's a swap cache page, and try to do that swap
cache removal even earlier, so that the page actually gets re-used
(instead of copied and then the swap entry removed).

But swap is such a non-issue these days that I doubt it matters, and
it's probably better to keep the swap handling in the unusual path.

So mapcount and refcount aren't what matters for the swap cache.

The swap count obviously *does* matter - because it means that some
mapping has a reference to this swap entry (not as a page, but as an
actual swap pointer).

But the mapcount is irrelevant -  any users that have the swap page
actually mapped, don't actually need to be a swapcache page.

Even the refcount doesn't really matter, afaik. The only "refcount" we
care about is that swapcount - that's what actually reference counts
the swap cases.

try_to_free_swap() does check for one particular kind of reference: it
does a check for PageWriteback(). We don't want to remove the thing
from the swap cache if it's under active IO.

(This codepath does need the page lock, though, thus all those
"page_trylock()" things).

                   Linus