Re: [PATCH] scsi: libsas: Fix a NULL pointer dereference in sas_ex_discover_expander()

[John Garry]

On 30/11/2021 17:16, Zhou Qingyang wrote:

I'd have "scsi: libsas: Improve error handling in 
sas_ex_discover_expander()"

> In sas_ex_discover_expander(), sas_port_alloc() is assigned to phy->port

"sas_port_alloc() is assigned to phy->port" - the function is not assigned

> and used in sas_port_add(). sas_port_add() further passes phy->port to
> list_empty(), and there is a dereference of it in list_empty(), which
> could lead to a NULL pointer dereference on failure of
> sas_port_alloc().
>
> This patch imitates the same error-handling logic in
> sas_ex_discover_end_dev().

git grep 'This patch' Documentation/process/submitting-patches.rst

> Fix this bug by adding checks for phy->port and sas_port_add().
>
> This bug was found by a static analyzer. The analysis employs
> differential checking to identify inconsistent security operations
> (e.g., checks or kfrees) between two code paths and confirms that the
> inconsistent operations are not recovered in the current function or
> the callers, so they constitute bugs.
>
> Note that, as a bug found by static analysis, it can be a false
> positive or hard to trigger. Multiple researchers have cross-reviewed
> the bug.

Who are these researchers?

> Builds with CONFIG_SCSI_SAS_LIBSAS=m show no new warnings,
> and our static analyzer no longer warns about this code.

This is all implied by sending the patch in the first place

> Fixes:  2908d778ab3e ("[SCSI] aic94xx: new driver")

personally I don't think that this is a fix - the code is old and 
already had BUG_ON()

> Signed-off-by: Zhou Qingyang <zhou1615@xxxxxxx>
> ---
>   drivers/scsi/libsas/sas_expander.c | 11 +++++++++--
>   1 file changed, 9 insertions(+), 2 deletions(-)
>
> diff --git a/drivers/scsi/libsas/sas_expander.c b/drivers/scsi/libsas/sas_expander.c
> index c2150a818423..7530b1773d6b 100644
> --- a/drivers/scsi/libsas/sas_expander.c
> +++ b/drivers/scsi/libsas/sas_expander.c
> @@ -957,9 +957,16 @@ static struct domain_device *sas_ex_discover_expander(
>   		return NULL;
>   
>   	phy->port = sas_port_alloc(&parent->rphy->dev, phy_id);
> -	/* FIXME: better error handling */
> -	BUG_ON(sas_port_add(phy->port) != 0);
> +	if (unlikely(!phy->port)) {

no need for unlikely() - this is not fastpath

> +		sas_put_device(child);
> +		return NULL;
> +	}
>   
> +	if (sas_port_add(phy->port) != 0) {
> +		sas_port_free(phy->port);
> +		sas_put_device(child);

better have a goto error now as we're replicting code, including what is 
already there for the sas_discover_expander() failure error path

> +		return NULL;
> +	}
>   
>   	switch (phy->attached_dev_type) {
>   	case SAS_EDGE_EXPANDER_DEVICE: