Received-SPF: pass (google.com: domain of linux-kernel-owner@vger.kernel.org designates 2620:137:e000::1:20 as permitted sender) client-ip=2620:137:e000::1:20;
MIME-Version: 1.0
References: <20221025182149.3076870-1-axelrasmussen@google.com> <Y1gsKdfXdIzkidpN@x1n>
In-Reply-To: <Y1gsKdfXdIzkidpN@x1n>
From:   Axel Rasmussen <axelrasmussen@google.com>
Date:   Tue, 25 Oct 2022 11:40:39 -0700
Message-ID: <CAJHvVcg5yr_OhV238vurf4PksSGPSvV_pTpeVWAKcumv-U3VcQ@mail.gmail.com>
Subject: Re: [PATCH] userfaultfd: wake on unregister for minor faults as well
 as missing
To:     Peter Xu <peterx@redhat.com>
Cc:     Alexander Viro <viro@zeniv.linux.org.uk>,
        Andrew Morton <akpm@linux-foundation.org>,
        Mike Kravetz <mike.kravetz@oracle.com>,
        Lokesh Gidra <lokeshgidra@google.com>,
        linux-fsdevel@vger.kernel.org, linux-kernel@vger.kernel.org,
        stable@vger.kernel.org
Content-Type: text/plain; charset="UTF-8"
Precedence: bulk

On Tue, Oct 25, 2022 at 11:34 AM Peter Xu <peterx@redhat.com> wrote:
>
> On Tue, Oct 25, 2022 at 11:21:49AM -0700, Axel Rasmussen wrote:
> > This was an overlooked edge case when minor faults were added. In
> > general, minor faults have the same rough edge here as missing faults:
> > if we unregister while there are waiting threads, they will just remain
> > waiting forever, as there is no way for userspace to wake them after
> > unregistration. To work around this, userspace needs to carefully wake
> > everything before unregistering.
> >
> > So, wake for minor faults just like we already do for missing faults as
> > part of the unregistration process.
> >
> > Cc: stable@vger.kernel.org
> > Fixes: 7677f7fd8be7 ("userfaultfd: add minor fault registration mode")
> > Reported-by: Lokesh Gidra <lokeshgidra@google.com>
> > Signed-off-by: Axel Rasmussen <axelrasmussen@google.com>
> > ---
> >  fs/userfaultfd.c | 2 +-
> >  1 file changed, 1 insertion(+), 1 deletion(-)
> >
> > diff --git a/fs/userfaultfd.c b/fs/userfaultfd.c
> > index 07c81ab3fd4d..7daee4b9481c 100644
> > --- a/fs/userfaultfd.c
> > +++ b/fs/userfaultfd.c
> > @@ -1606,7 +1606,7 @@ static int userfaultfd_unregister(struct userfaultfd_ctx *ctx,
> >                       start = vma->vm_start;
> >               vma_end = min(end, vma->vm_end);
> >
> > -             if (userfaultfd_missing(vma)) {
> > +             if (userfaultfd_missing(vma) || userfaultfd_minor(vma)) {
> >                       /*
> >                        * Wake any concurrent pending userfault while
> >                        * we unregister, so they will not hang
> > --
> > 2.38.0.135.g90850a2211-goog
>
> Thanks, Axel.  Is wr-protect mode also prone to this?  Would a test case
> help too?

I'm not quite as familiar with uffd-wp, but I think so? At minimum, it
seems like waking can't *hurt*, and it would simplify the check
slightly -- if (userfaultfd_armed(vma)) {}

It would also mean if we add yet another registration mode in the
future, we wouldn't forget to update this.

I'll send a v2 to address both points.

>
> --
> Peter Xu
>