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;
Date:   Tue, 25 Jul 2023 23:30:17 -0700
From:   Bjorn Andersson <andersson@kernel.org>
To:     Ekansh Gupta <quic_ekangupt@quicinc.com>
Cc:     srinivas.kandagatla@linaro.org, linux-arm-msm@vger.kernel.org,
        ekangupt@qti.qualcomm.com, gregkh@linuxfoundation.org,
        linux-kernel@vger.kernel.org, fastrpc.upstream@qti.qualcomm.com,
        stable <stable@kernel.org>
Subject: Re: [PATCH v2] misc: fastrpc: Fix incorrect DMA mapping unmap request
Message-ID: <zo6wjmqolvnla7x24qguh6rbjpf7l62vmckxhy3nps7hhaljdr@kyoh6xedhxaq>
References: <1690182571-7348-1-git-send-email-quic_ekangupt@quicinc.com>
MIME-Version: 1.0
Content-Type: text/plain; charset=us-ascii
Content-Disposition: inline
In-Reply-To: <1690182571-7348-1-git-send-email-quic_ekangupt@quicinc.com>
Precedence: bulk

On Mon, Jul 24, 2023 at 12:39:31PM +0530, Ekansh Gupta wrote:
> Scatterlist table is obtained during map create request and the same

I'm guessing that this all happens in fastrpc_map_create() where:

  map->table = dma_buf_map_attachment_unlocked(map->attach, DMA_BIDIRECTIONAL);

fails, we jump to map_err, and then call fastrpc_map_put(map), which
then ends up in the code below?

> table is used for DMA mapping unmap. In case there is any failure
> while getting the sg_table, ERR_PTR is returned instead of sg_table.

The problem isn't that ERR_PTR() is being returned, the problem is that
this is being assigned to map->table and you keep running.

> 
> When the map is getting freed, there is only a non-NULL check of
> sg_table which will also be true in case failure was returned instead
> of sg_table. This would result in improper unmap request. Add proper
> check to avoid bad unmap request.
> 
> Fixes: c68cfb718c8f ("misc: fastrpc: Add support for context Invoke method")
> Cc: stable <stable@kernel.org>
> Tested-by: Ekansh Gupta <quic_ekangupt@quicinc.com>

You always test your own patches, so no need to declare this.

> Signed-off-by: Ekansh Gupta <quic_ekangupt@quicinc.com>
> ---
> Changes in v2:
>   - Added fixes information to commit text
> 
>  drivers/misc/fastrpc.c | 2 +-
>  1 file changed, 1 insertion(+), 1 deletion(-)
> 
> diff --git a/drivers/misc/fastrpc.c b/drivers/misc/fastrpc.c
> index 9666d28..75da69a 100644
> --- a/drivers/misc/fastrpc.c
> +++ b/drivers/misc/fastrpc.c
> @@ -313,7 +313,7 @@ static void fastrpc_free_map(struct kref *ref)
>  
>  	map = container_of(ref, struct fastrpc_map, refcount);
>  
> -	if (map->table) {
> +	if (map->table && !IS_ERR(map->table)) {

Rather than carrying around an IS_ERR(map->table), I think you should
address this at the originating place. E.g. assign the return value of
the dma_buf_map_attachment_unlocked() to a local variable and only if it
is valid you assign map->table. Or perhaps make it NULL in the error
path.

Regards,
Bjorn

>  		if (map->attr & FASTRPC_ATTR_SECUREMAP) {
>  			struct qcom_scm_vmperm perm;
>  			int vmid = map->fl->cctx->vmperms[0].vmid;
> -- 
> 2.7.4
>