linux-kernel - Re: [PATCH v2 4/5] nvme-fc: Make initial connect attempt synchronous

lists.openwall.net		lists / announce owl-users owl-dev john-users john-dev passwdqc-users yescrypt popa3d-users / oss-security kernel-hardening musl sabotage tlsify passwords / crypt-dev xvendor / Bugtraq Full-Disclosure linux-kernel linux-netdev linux-ext4 linux-hardening linux-cve-announce PHC
Open Source and information security mailing list archives

Hash Suite: Windows password security audit tool. GUI, reports in PDF.

[<prev] [next>] [<thread-prev] [thread-next>] [day] [month] [year] [list]

Date:   Tue, 27 Jun 2023 08:18:37 +0200
From:   Daniel Wagner <dwagner@...e.de>
To:     Dan Carpenter <dan.carpenter@...aro.org>
Cc:     linux-nvme@...ts.infradead.org, linux-kernel@...r.kernel.org,
        linux-block@...r.kernel.org, Chaitanya Kulkarni <kch@...dia.com>,
        Shin'ichiro Kawasaki <shinichiro@...tmail.com>,
        Sagi Grimberg <sagi@...mberg.me>,
        Hannes Reinecke <hare@...e.de>,
        James Smart <jsmart2021@...il.com>
Subject: Re: [PATCH v2 4/5] nvme-fc: Make initial connect attempt synchronous

On Mon, Jun 26, 2023 at 02:33:18PM +0300, Dan Carpenter wrote:
> > @@ -2943,6 +2943,8 @@ nvme_fc_create_io_queues(struct nvme_fc_ctrl *ctrl)
> >  	/* force put free routine to ignore io queues */
> >  	ctrl->ctrl.tagset = NULL;
> >  
> > +	if (ret > 0)
> > +		ret = -EIO;
> 
> All these checks for ret > 0 make me unhappy.  I don't understand how
> they are a part of the commit.

We have two types of error message types in the nvme subsystem. The negative
values are the usual ones and positive ones are nvme protocol errors.

For example if the authentication fails because of invalid credentials when
doing the authentication nvmf_connect_admin_queue() will return a value of
NVME_SC_AUTH_REQUIRED. This is also the value which gets propagated to this
point here. The problem is any positive error code is interpreted as a valid
pointer later in the code, which results in a crash.

> I have tried to look at the context and I think maybe you are working
> around the fact that qla_nvme_ls_req() returns QLA_FUNCTION_FAILED on
> error.

The auth blktests are exercising the error path here and that's why I added
this check. BTW, we already use in other places, this is not completely new in
this subsystem.

> Also the qla_nvme_ls_req() function EINVAL on error.  I just wrote a
> commit message saying that none of the callers cared but I missed that
> apparently gets returned to nvme_fc_init_ctrl().  :/
> https://lore.kernel.org/all/49866d28-4cfe-47b0-842b-78f110e61aab@moroto.mountain/

Thank!

> Let's just fix qla_nvme_ls_req() instead of working around it here.
>
> And let's add a WARN_ON_ONCE() somewhere to prevent future bugs.

This makes sense for the driver APIs. Though for the core nvme subsystem this
needs to be discusses/redesigned how to handle the protocol errors first.

Thanks,
Daniel