Received-SPF: pass (google.com: domain of linux-kernel+bounces-53981-linux.lists.archive=gmail.com@vger.kernel.org designates 147.75.80.249 as permitted sender) client-ip=147.75.80.249;
Precedence: bulk
MIME-Version: 1.0
References: <20240129193512.123145-1-lokeshgidra@google.com>
 <20240129193512.123145-4-lokeshgidra@google.com> <20240129203626.uq5tdic4z5qua5qy@revolver>
 <CAJuCfpFS=h8h1Tgn55Hv+cr9bUFFoUvejiFQsHGN5yT7utpDMg@mail.gmail.com>
 <CA+EESO5r+b7QPYM5po--rxQBa9EPi4x1EZ96rEzso288dbpuow@mail.gmail.com>
 <20240130025803.2go3xekza5qubxgz@revolver> <CA+EESO4+ExV-2oo0rFNpw0sL+_tWZ_MH_rUh-wvssN0y_hr+LA@mail.gmail.com>
 <20240131214104.rgw3x5vuap43xubi@revolver>
In-Reply-To: <20240131214104.rgw3x5vuap43xubi@revolver>
From: Suren Baghdasaryan <surenb@google.com>
Date: Mon, 5 Feb 2024 13:46:54 -0800
Message-ID: <CAJuCfpFB6Udm0pkTwJCOtvrn9+=g05oFgL-dUnEkEO0cGmyvOw@mail.gmail.com>
Subject: Re: [PATCH v2 3/3] userfaultfd: use per-vma locks in userfaultfd operations
To: "Liam R. Howlett" <Liam.Howlett@oracle.com>, Lokesh Gidra <lokeshgidra@google.com>, 
	Suren Baghdasaryan <surenb@google.com>, akpm@linux-foundation.org, 
	linux-fsdevel@vger.kernel.org, linux-mm@kvack.org, 
	linux-kernel@vger.kernel.org, selinux@vger.kernel.org, 
	kernel-team@android.com, aarcange@redhat.com, peterx@redhat.com, 
	david@redhat.com, axelrasmussen@google.com, bgeffon@google.com, 
	willy@infradead.org, jannh@google.com, kaleshsingh@google.com, 
	ngeoffray@google.com, timmurray@google.com, rppt@kernel.org
Content-Type: text/plain; charset="UTF-8"
Content-Transfer-Encoding: quoted-printable

On Wed, Jan 31, 2024 at 1:41=E2=80=AFPM Liam R. Howlett <Liam.Howlett@oracl=
e.com> wrote:
>
> * Lokesh Gidra <lokeshgidra@google.com> [240130 21:49]:
> > On Mon, Jan 29, 2024 at 6:58=E2=80=AFPM Liam R. Howlett <Liam.Howlett@o=
racle.com> wrote:
> > >
> > > * Lokesh Gidra <lokeshgidra@google.com> [240129 19:28]:
> > > > On Mon, Jan 29, 2024 at 12:53=E2=80=AFPM Suren Baghdasaryan <surenb=
@google.com> wrote:
> > > > >
> > >
>
> ...
>
> > >
> > > > Your suggestion is definitely simpler and easier to follow, but due=
 to
> > > > the overflow situation that Suren pointed out, I would still need t=
o
> > > > keep the locking/boolean dance, no? IIUC, even if I were to return
> > > > EAGAIN to the userspace, there is no guarantee that subsequent ioct=
ls
> > > > on the same vma will succeed due to the same overflow, until someon=
e
> > > > acquires and releases mmap_lock in write-mode.
> > > > Also, sometimes it seems insufficient whether we managed to lock vm=
a
> > > > or not. For instance, lock_vma_under_rcu() checks if anon_vma (for
> > > > anonymous vma) exists. If not then it bails out.
> > > > So it seems to me that we have to provide some fall back in
> > > > userfaultfd operations which executes with mmap_lock in read-mode.
> > >
> > > Fair enough, what if we didn't use the sequence number and just locke=
d
> > > the vma directly?
> >
> > Looks good to me, unless someone else has any objections.
> > >
> > > /* This will wait on the vma lock, so once we return it's locked */
> > > void vma_aquire_read_lock(struct vm_area_struct *vma)
> > > {
> > >         mmap_assert_locked(vma->vm_mm);
> > >         down_read(&vma->vm_lock->lock);
> > > }
> > >
> > > struct vm_area_struct *lock_vma(struct mm_struct *mm,
> > >         unsigned long addr))    /* or some better name.. */
> > > {
> > >         struct vm_area_struct *vma;
> > >
> > >         vma =3D lock_vma_under_rcu(mm, addr);
> > >         if (vma)
> > >                 return vma;
> > >
> > >         mmap_read_lock(mm);
> > >         /* mm sequence cannot change, no mm writers anyways.
> > >          * find_mergeable_anon_vma is only a concern in the page faul=
t
> > >          * path
> > >          * start/end won't change under the mmap_lock
> > >          * vma won't become detached as we have the mmap_lock in read
> > >          * We are now sure no writes will change the VMA
> > >          * So let's make sure no other context is isolating the vma
> > >          */
> > >         vma =3D lookup_vma(mm, addr);
> > >         if (vma)
> > We can take care of anon_vma as well here right? I can take a bool
> > parameter ('prepare_anon' or something) and then:
> >
> >            if (vma) {
> >                     if (prepare_anon && vma_is_anonymous(vma)) &&
> > !anon_vma_prepare(vma)) {
> >                                       vma =3D ERR_PTR(-ENOMEM);
> >                                       goto out_unlock;
> >                    }
> > >                 vma_aquire_read_lock(vma);
> >            }
> > out_unlock:
> > >         mmap_read_unlock(mm);
> > >         return vma;
> > > }
>
> Do you need this?  I didn't think this was happening in the code as
> written?  If you need it I would suggest making it happen always and
> ditch the flag until a user needs this variant, but document what's
> going on in here or even have a better name.

I think yes, you do need this. I can see calls to anon_vma_prepare()
under mmap_read_lock() protection in both mfill_atomic_hugetlb() and
in mfill_atomic(). This means, just like in the pagefault path, we
modify vma->anon_vma under mmap_read_lock protection which guarantees
that adjacent VMAs won't change. This is important because
__anon_vma_prepare() uses find_mergeable_anon_vma() that needs the
neighboring VMAs to be stable. Per-VMA lock guarantees stability of
the VMA we locked but not of its neighbors, therefore holding per-VMA
lock while calling anon_vma_prepare() is not enough. The solution
Lokesh suggests would call anon_vma_prepare() under mmap_read_lock and
therefore would avoid the issue.


>
> Thanks,
> Liam