Received-SPF: pass (google.com: domain of linux-kernel-owner@vger.kernel.org designates 23.128.96.18 as permitted sender) client-ip=23.128.96.18;
Date:   Tue, 28 Apr 2020 09:06:51 +0200
From:   Javier =?utf-8?B?R29uesOhbGV6?= <javier@javigon.com>
To:     Niklas Cassel <Niklas.Cassel@wdc.com>
Cc:     Keith Busch <kbusch@kernel.org>, Jens Axboe <axboe@fb.com>,
        Christoph Hellwig <hch@lst.de>,
        Sagi Grimberg <sagi@grimberg.me>,
        Igor Konopko <igor.j.konopko@intel.com>,
        Matias =?utf-8?B?QmrDuHJsaW5n?= <mb@lightnvm.io>,
        Jens Axboe <axboe@kernel.dk>,
        "linux-nvme@lists.infradead.org" <linux-nvme@lists.infradead.org>,
        "linux-kernel@vger.kernel.org" <linux-kernel@vger.kernel.org>
Subject: Re: [PATCH] nvme: prevent double free in nvme_alloc_ns() error
 handling
Message-ID: <20200428070651.qbsyivvaakflipr4@mpHalley.localdomain>
References: <20200427123443.520469-1-niklas.cassel@wdc.com>
 <20200427180311.nssquibbak5ib4oo@mpHalley.localdomain>
 <20200427182245.GA547726@localhost.localdomain>
MIME-Version: 1.0
Content-Type: text/plain; charset=utf-8; format=flowed
Content-Disposition: inline
Content-Transfer-Encoding: 8bit
In-Reply-To: <20200427182245.GA547726@localhost.localdomain>
Sender: linux-kernel-owner@vger.kernel.org
Precedence: bulk

On 27.04.2020 18:22, Niklas Cassel wrote:
>On Mon, Apr 27, 2020 at 08:03:11PM +0200, Javier González wrote:
>> On 27.04.2020 14:34, Niklas Cassel wrote:
>> > When jumping to the out_put_disk label, we will call put_disk(), which will
>> > trigger a call to disk_release(), which calls blk_put_queue().
>> >
>> > Later in the cleanup code, we do blk_cleanup_queue(), which will also call
>> > blk_put_queue().
>> >
>> > Putting the queue twice is incorrect, and will generate a KASAN splat.
>> >
>> > Set the disk->queue pointer to NULL, before calling put_disk(), so that the
>> > first call to blk_put_queue() will not free the queue.
>> >
>> > The second call to blk_put_queue() uses another pointer to the same queue,
>> > so this call will still free the queue.
>> >
>> > Fixes: 85136c010285 ("lightnvm: simplify geometry enumeration")
>> > Signed-off-by: Niklas Cassel <niklas.cassel@wdc.com>
>> > ---
>> > drivers/nvme/host/core.c | 2 ++
>> > 1 file changed, 2 insertions(+)
>> >
>> > diff --git a/drivers/nvme/host/core.c b/drivers/nvme/host/core.c
>> > index 91c1bd659947..f2adea96b04c 100644
>> > --- a/drivers/nvme/host/core.c
>> > +++ b/drivers/nvme/host/core.c
>> > @@ -3642,6 +3642,8 @@ static void nvme_alloc_ns(struct nvme_ctrl *ctrl, unsigned nsid)
>> >
>> >       return;
>> >  out_put_disk:
>> > +      /* prevent double queue cleanup */
>> > +      ns->disk->queue = NULL;
>> >       put_disk(ns->disk);
>> >  out_unlink_ns:
>> >       mutex_lock(&ctrl->subsys->lock);
>> > --
>> > 2.25.3
>> >
>> What about delaying the assignment of ns->disk?
>>
>> diff --git i/drivers/nvme/host/core.c w/drivers/nvme/host/core.c
>> index a4d8c90ee7cc..6da4a9ced945 100644
>> --- i/drivers/nvme/host/core.c
>> +++ w/drivers/nvme/host/core.c
>> @@ -3541,7 +3541,6 @@ static int nvme_alloc_ns(struct nvme_ctrl *ctrl, unsigned nsid)
>>         disk->queue = ns->queue;
>>         disk->flags = flags;
>>         memcpy(disk->disk_name, disk_name, DISK_NAME_LEN);
>> -       ns->disk = disk;
>>
>>         __nvme_revalidate_disk(disk, id);
>>
>> @@ -3553,6 +3552,8 @@ static int nvme_alloc_ns(struct nvme_ctrl *ctrl, unsigned nsid)
>>                 }
>>         }
>>
>> +       ns->disk = disk;
>> +
>
>Hello Javier!
>
>
>The only case where we jump to the out_put_disk label, is if the
>nvme_nvm_register() call failed.
>
>In that case, we want to undo the alloc_disk_node() operation, i.e.,
>decrease the refcount.
>
>If we don't set "ns->disk = disk;" before the call to nvme_nvm_register(),
>then, if register fails, and we jump to the put_disk(ns->disk) label,
>ns->disk will be NULL, so the recount will not be decreased, so I assume
>that this memory would then be a memory leak.
>
>
>I think that the problem is that the block functions are a bit messy.
>Most drivers seem to do blk_cleanup_queue() first and then do put_disk(),
>but some drivers do it in the opposite way, so I think that we might have
>some more use-after-free bugs in some of these drivers that do it in the
>opposite way.
>

Hi Niklas,

Yes, the out_put_disk label was introduced at the same time as the
LightNVM entry point. We can do a better job at separating the cleanup
functions, but as far as I can see ns->disk is not used in the LightNVM
initialization, so delaying the initialization should be ok. Part of
this should be also changing the out_put_disk to put_disk(disk).

Note that initializing other namespace types here do not require
ns->disk either, so delaying initialization should be ok. We have been
running with this patch locally for some time.

This said, this is an alternative as your fix works.

Javier