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: <20230113031921.64716-1-bhe@redhat.com> <20230113031921.64716-4-bhe@redhat.com>
 <Y8U6Gi5d4F8mO0ib@pc636> <Y8kSwyJBe426pj7R@fedora> <Y8k8K9Ztfm/Yj5PO@fedora>
In-Reply-To: <Y8k8K9Ztfm/Yj5PO@fedora>
From:   Uladzislau Rezki <urezki@gmail.com>
Date:   Fri, 20 Jan 2023 12:54:47 +0100
Message-ID: <CA+KHdyXJhAMh=HXK7_pO0kTBvSCAuJNT-6pMhtXWgDO3pM7R-A@mail.gmail.com>
Subject: Re: [PATCH v3 3/7] mm/vmalloc.c: allow vread() to read out vm_map_ram areas
To:     Baoquan He <bhe@redhat.com>
Cc:     linux-mm@kvack.org, linux-kernel@vger.kernel.org,
        lstoakes@gmail.com, stephen.s.brennan@oracle.com,
        willy@infradead.org, akpm@linux-foundation.org, hch@infradead.org
Content-Type: text/plain; charset="UTF-8"
Precedence: bulk

>
> On 01/19/23 at 05:52pm, Baoquan He wrote:
> > On 01/16/23 at 12:50pm, Uladzislau Rezki wrote:
> > > On Fri, Jan 13, 2023 at 11:19:17AM +0800, Baoquan He wrote:
> > > > Currently, vread can read out vmalloc areas which is associated with
> > > > a vm_struct. While this doesn't work for areas created by vm_map_ram()
> > > > interface because it doesn't have an associated vm_struct. Then in vread(),
> > > > these areas are all skipped.
> > > >
> > > > Here, add a new function vmap_ram_vread() to read out vm_map_ram areas.
> > > > The area created with vmap_ram_vread() interface directly can be handled
> > > > like the other normal vmap areas with aligned_vread(). While areas
> > > > which will be further subdivided and managed with vmap_block need
> > > > carefully read out page-aligned small regions and zero fill holes.
> > > >
> > > > Signed-off-by: Baoquan He <bhe@redhat.com>
> > > > ---
> > > >  mm/vmalloc.c | 80 +++++++++++++++++++++++++++++++++++++++++++++++-----
> > > >  1 file changed, 73 insertions(+), 7 deletions(-)
> > > >
> > > > diff --git a/mm/vmalloc.c b/mm/vmalloc.c
> > > > index ab4825050b5c..13875bc41e27 100644
> > > > --- a/mm/vmalloc.c
> > > > +++ b/mm/vmalloc.c
> > > > @@ -3544,6 +3544,65 @@ static int aligned_vread(char *buf, char *addr, unsigned long count)
> > > >   return copied;
> > > >  }
> > > >
> > > > +static void vmap_ram_vread(char *buf, char *addr, int count, unsigned long flags)
> > > > +{
> > > > + char *start;
> > > > + struct vmap_block *vb;
> > > > + unsigned long offset;
> > > > + unsigned int rs, re, n;
> > > > +
> > > > + /*
> > > > +  * If it's area created by vm_map_ram() interface directly, but
> > > > +  * not further subdividing and delegating management to vmap_block,
> > > > +  * handle it here.
> > > > +  */
> > > > + if (!(flags & VMAP_BLOCK)) {
> > > > +         aligned_vread(buf, addr, count);
> > > > +         return;
> > > > + }
> > > > +
> > > > + /*
> > > > +  * Area is split into regions and tracked with vmap_block, read out
> > > > +  * each region and zero fill the hole between regions.
> > > > +  */
> > > > + vb = xa_load(&vmap_blocks, addr_to_vb_idx((unsigned long)addr));
> > > > +
> > > > + spin_lock(&vb->lock);
> > > > + if (bitmap_empty(vb->used_map, VMAP_BBMAP_BITS)) {
> > > >
> > > CPU-X invokes free_vmap_block() whereas we take the vb->lock and do
> > > some manipulations with vb that might be already freed over RCU-core.
> > >
> > > Should we protect it by the rcu_read_lock() also here?
> >
> > Just go over the vb and vbq code again, seems we don't need the
> > rcu_read_lock() here. The rcu lock is needed when operating on the
> > vmap_block_queue->free list. I don't see race between the vb accessing
> > here and those list adding or removing on vmap_block_queue->free with
> > rcu. If I miss some race windows between them, please help point out.
> >
> > However, when I check free_vmap_block(), I do find a risk. As you said,
>
> Forgot to add details about why there's no race between free_vmap_block()
> and vmap_ram_vread() because we have taken vmap_area_lock at the beginning
> of vread(). So, except of the missing checking on returned value from
> xa_load(), free_vmap_block() either is blocked to wait for vmap_area_lock
> before calling unlink_va(), or finishes calling unlink_va() to remove
> the vmap from vmap_area_root tree. In both cases, no race happened.
>
Agree. xa_load()s return value should be checked. Because it can be that
there is no vmap_block associated with an address if xa_erase() was done
earlier.

--
Uladzislau Rezki