On Thu, Sep 16, 2021 at 09:10:37PM +0800, yukuai (C) wrote:
> On 2021/09/16 20:58, Ming Lei wrote:
> > On Thu, Sep 16, 2021 at 05:33:50PM +0800, Yu Kuai wrote:
> > > There is a problem that nbd_handle_reply() might access freed request:
> > >
> > > 1) At first, a normal io is submitted and completed with scheduler:
> > >
> > > internel_tag = blk_mq_get_tag -> get tag from sched_tags
> > > blk_mq_rq_ctx_init
> > > sched_tags->rq[internel_tag] = sched_tag->static_rq[internel_tag]
> > > ...
> > > blk_mq_get_driver_tag
> > > __blk_mq_get_driver_tag -> get tag from tags
> > > tags->rq[tag] = sched_tag->static_rq[internel_tag]
> > >
> > > So, both tags->rq[tag] and sched_tags->rq[internel_tag] are pointing
> > > to the request: sched_tags->static_rq[internal_tag]. Even if the
> > > io is finished.
> > >
> > > 2) nbd server send a reply with random tag directly:
> > >
> > > recv_work
> > > nbd_handle_reply
> > > blk_mq_tag_to_rq(tags, tag)
> > > rq = tags->rq[tag]
> > >
> > > 3) if the sched_tags->static_rq is freed:
> > >
> > > blk_mq_sched_free_requests
> > > blk_mq_free_rqs(q->tag_set, hctx->sched_tags, i)
> > > -> step 2) access rq before clearing rq mapping
> > > blk_mq_clear_rq_mapping(set, tags, hctx_idx);
> > > __free_pages() -> rq is freed here
> > >
> > > 4) Then, nbd continue to use the freed request in nbd_handle_reply
> > >
> > > Fix the problem by get 'q_usage_counter' before blk_mq_tag_to_rq(),
> > > thus request is ensured not to be freed because 'q_usage_counter' is
> > > not zero.
> > >
> > > Signed-off-by: Yu Kuai <yukuai3@xxxxxxxxxx>
> > > ---
> > > drivers/block/nbd.c | 14 ++++++++++++++
> > > 1 file changed, 14 insertions(+)
> > >
> > > diff --git a/drivers/block/nbd.c b/drivers/block/nbd.c
> > > index 69dc5eac9ad3..b3a47fc6237f 100644
> > > --- a/drivers/block/nbd.c
> > > +++ b/drivers/block/nbd.c
> > > @@ -825,6 +825,7 @@ static void recv_work(struct work_struct *work)
> > > work);
> > > struct nbd_device *nbd = args->nbd;
> > > struct nbd_config *config = nbd->config;
> > > + struct request_queue *q = nbd->disk->queue;
> > > struct nbd_sock *nsock;
> > > struct nbd_cmd *cmd;
> > > struct request *rq;
> > > @@ -835,7 +836,20 @@ static void recv_work(struct work_struct *work)
> > > if (nbd_read_reply(nbd, args->index, &reply))
> > > break;
> > > + /*
> > > + * Grab .q_usage_counter so request pool won't go away, then no
> > > + * request use-after-free is possible during nbd_handle_reply().
> > > + * If queue is frozen, there won't be any inflight requests, we
> > > + * needn't to handle the incoming garbage message.
> > > + */
> > > + if (!percpu_ref_tryget(&q->q_usage_counter)) {
> > > + dev_err(disk_to_dev(nbd->disk), "%s: no io inflight\n",
> > > + __func__);
> > > + break;
> > > + }
> > > +
> > > cmd = nbd_handle_reply(nbd, args->index, &reply);
> > > + percpu_ref_put(&q->q_usage_counter);
> > > if (IS_ERR(cmd))
> > > break;
> >
> > The refcount needs to be grabbed when completing the request because
> > the request may be completed from other code path, then the request pool
> > will be freed from that code path when the request is referred.
>
> Hi,
>
> The request can't complete concurrently, thus put ref here is safe.
>
> There used to be a commet here that I tried to explain it... It's fine
> to me to move it behind anyway.
Never see such comment. cmd->lock isn't held here, so I believe
concurrent completion is possible here.
Thanks,
Ming
