Re: [PATCH v2] misc: fastrpc: Fix incorrect DMA mapping unmap request

[Ekansh Gupta]

On 7/26/2023 12:00 PM, Bjorn Andersson wrote:
> 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?
yes, your understanding is correct.
> > 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@xxxxxxxxxx>
> > Tested-by: Ekansh Gupta <quic_ekangupt@xxxxxxxxxxx>
>
> You always test your own patches, so no need to declare this.
sure, I'll avoid adding this for future changes.
> > Signed-off-by: Ekansh Gupta <quic_ekangupt@xxxxxxxxxxx>
> > ---
> > 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.
understood, this looks much cleaner solution. I'll update this in the 
next patch. Thanks for taking your time to review this change, Bjorn.
--ekansh

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