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;
Message-ID: <19f203e5-e965-23bd-401b-0ae8d9a73a5d@acm.org>
Date:   Sun, 3 Jul 2022 07:55:05 -0700
MIME-Version: 1.0
User-Agent: Mozilla/5.0 (X11; Linux x86_64; rv:91.0) Gecko/20100101
 Thunderbird/91.10.0
Subject: Re: use-after-free in srpt_enable_tpg()
Content-Language: en-US
To:     Hillf Danton <hdanton@sina.com>
Cc:     Mike Christie <michael.christie@oracle.com>,
        "lizhijian@fujitsu.com" <lizhijian@fujitsu.com>,
        Jason Gunthorpe <jgg@ziepe.ca>,
        Leon Romanovsky <leon@kernel.org>,
        "linux-rdma@vger.kernel.org" <linux-rdma@vger.kernel.org>,
        "target-devel@vger.kernel.org" <target-devel@vger.kernel.org>,
        open list <linux-kernel@vger.kernel.org>
References: <17649b9c-7e42-1625-8bc9-8ad333ab771c@fujitsu.com>
 <ed7e268e-94c5-38b1-286d-e2cb10412334@acm.org>
 <fbaca135-891c-7ff3-d7ac-bd79609849f5@oracle.com>
 <20220701015934.1105-1-hdanton@sina.com>
 <20220703021119.1109-1-hdanton@sina.com>
From:   Bart Van Assche <bvanassche@acm.org>
In-Reply-To: <20220703021119.1109-1-hdanton@sina.com>
Content-Type: text/plain; charset=UTF-8; format=flowed
Content-Transfer-Encoding: 7bit
Precedence: bulk

On 7/2/22 19:11, Hillf Danton wrote:
> On Sat, 2 Jul 2022 15:26:33 -0700 Bart Van Assche wrote:
>> As long as a session is live the ch->qp pointer may be
>> dereferenced. The sdev->pd pointer is stored in the pd member of struct
>> ib_qp and hence may be dereferenced by any function that uses ch->qp.
> 
> If it is still an issue after ib_dealloc_pd(sdev->pd) then it goes beyond the
> aperture of my proposal and needs another fix.
> 
> Hillf
> 
> +++ b/drivers/infiniband/ulp/srpt/ib_srpt.c
> @@ -3235,6 +3235,8 @@ static void srpt_remove_one(struct ib_de
>   
>   	ib_set_client_data(device, &srpt_client, NULL);
>   
> +	/* make sdev survive ib_dealloc_pd(sdev->pd); */
> +	atomic_inc(&sdev->port_refcount);
>   	/*
>   	 * Unregistering a target must happen after destroying sdev->cm_id
>   	 * such that no new SRP_LOGIN_REQ information units can arrive while
> @@ -3250,6 +3252,9 @@ static void srpt_remove_one(struct ib_de
>   	srpt_free_srq(sdev);
>   
>   	ib_dealloc_pd(sdev->pd);
> +
> +	if (0 == atomic_dec_return(&sdev->port_refcount))
> +		kfree(sdev);
>   }
>   
>   static struct ib_client srpt_client = {

Do you perhaps want to combine the above patch with the previous patch? 
I don't think that any reference counting scheme can fix all 
use-after-free issues related to srpt_remove_one(). Immediately after 
srpt_remove_one() returns the caller of this function calls 
ib_device_put() and ib_client_put(). These functions free data 
structures that can be reached from the pointers that are stored in 
struct ib_qp. Holding a reference on struct ib_device as long as any 
session is live would allow to remove the while-loop from 
srpt_release_sport(). However, I'm not sure that would make a 
significant difference since there is a similar while-loop in one of the 
callers of srpt_remove_one() (disable_device() in the RDMA core).

Bart.