Message-ID: <5cf19e69-b851-abe9-9496-bbba33109404@acm.org>
Date:   Wed, 1 Mar 2023 11:46:31 -0800
MIME-Version: 1.0
User-Agent: Mozilla/5.0 (X11; Linux x86_64; rv:102.0) Gecko/20100101
 Thunderbird/102.7.1
Subject: Re: [PATCH-next] scsi: fix use-after-free problem in
 scsi_remove_target
Content-Language: en-US
To:     Zhong Jinghua <zhongjinghua@huaweicloud.com>, jejb@linux.ibm.com,
        martin.petersen@oracle.com
Cc:     linux-scsi@vger.kernel.org, linux-kernel@vger.kernel.org,
        zhongjinghua@huawei.com, yi.zhang@huawei.com, yukuai3@huawei.com
References: <20230213034321.3261114-1-zhongjinghua@huaweicloud.com>
From:   Bart Van Assche <bvanassche@acm.org>
In-Reply-To: <20230213034321.3261114-1-zhongjinghua@huaweicloud.com>
Content-Type: text/plain; charset=UTF-8; format=flowed
Content-Transfer-Encoding: 7bit
Precedence: bulk

On 2/12/23 19:43, Zhong Jinghua wrote:
> T0							T1
>   sdev_store_delete
>    scsi_remove_device
>     device_remove_file
>      __scsi_remove_device
>          					__iscsi_unbind_session
>          					 scsi_remove_target
> 						  spin_lock_irqsave
>          					  list_for_each_entry
>       scsi_target_reap // starget->reaf 1 -> 0

What is "reaf"? Did you perhaps want to write "reap_ref"?

> diff --git a/drivers/scsi/scsi_sysfs.c b/drivers/scsi/scsi_sysfs.c
> index e7893835b99a..0ad357ff4c59 100644
> --- a/drivers/scsi/scsi_sysfs.c
> +++ b/drivers/scsi/scsi_sysfs.c
> @@ -1561,7 +1561,17 @@ void scsi_remove_target(struct device *dev)
>   		    starget->state == STARGET_CREATED_REMOVE)
>   			continue;
>   		if (starget->dev.parent == dev || &starget->dev == dev) {
> -			kref_get(&starget->reap_ref);
> +
> +			/*
> +			 * If starget->reap_ref is reduced to 0, it means
> +			 * that other processes are releasing it and
> +			 * there is no need to delete it again
> +			 */
> +			if (!kref_get_unless_zero(&starget->reap_ref)) {
> +				spin_unlock_irqrestore(shost->host_lock, flags);
> +				goto restart;
> +			}
> +
>   			if (starget->state == STARGET_CREATED)
>   				starget->state = STARGET_CREATED_REMOVE;
>   			else

The above comment should be made more clear, e.g. as follows: "If the 
reference count is already zero, skip this target. Calling 
kref_get_unless_zero() if the reference count is zero is safe because 
scsi_target_destroy() will wait until the host lock has been released 
before freeing starget."

Otherwise this patch looks fine to me.

Thanks,

Bart.