linux-kernel - Re: [RFC PATCH v7 10/16] net: dsa: qca8k: add support for mgmt read/write in Ethernet packet

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]

Message-ID: <20220125145459.i5jedxm225q5n5aq@skbuf>
Date:   Tue, 25 Jan 2022 16:54:59 +0200
From:   Vladimir Oltean <olteanv@...il.com>
To:     Ansuel Smith <ansuelsmth@...il.com>
Cc:     Andrew Lunn <andrew@...n.ch>,
        Vivien Didelot <vivien.didelot@...il.com>,
        Florian Fainelli <f.fainelli@...il.com>,
        "David S. Miller" <davem@...emloft.net>,
        Jakub Kicinski <kuba@...nel.org>, linux-kernel@...r.kernel.org,
        netdev@...r.kernel.org
Subject: Re: [RFC PATCH v7 10/16] net: dsa: qca8k: add support for mgmt
 read/write in Ethernet packet

On Mon, Jan 24, 2022 at 05:48:32PM +0100, Ansuel Smith wrote:
> > > +static int qca8k_read_eth(struct qca8k_priv *priv, u32 reg, u32 *val)
> > > +{
> > > +	struct qca8k_mgmt_hdr_data *mgmt_hdr_data = &priv->mgmt_hdr_data;
> > > +	struct sk_buff *skb;
> > > +	bool ack;
> > > +	int ret;
> > > +
> > > +	skb = qca8k_alloc_mdio_header(MDIO_READ, reg, NULL, 200, QCA8K_ETHERNET_MDIO_PRIORITY);
> > > +	if (!skb)
> > > +		return -ENOMEM;
> > > +
> > > +	mutex_lock(&mgmt_hdr_data->mutex);
> > > +
> > > +	/* Recheck mgmt_master under lock to make sure it's operational */
> > > +	if (!priv->mgmt_master)
> > 
> > mutex_unlock and kfree_skb
> > 
> > Also, why "recheck under lock"? Why not check just under lock?
> >
> 
> Tell me if the logic is wrong.
> We use the mgmt_master (outside lock) to understand if the eth mgmt is
> available. Then to make sure it's actually usable when the operation is
> actually done (and to prevent any panic if the master is dropped or for
> whatever reason mgmt_master is not available anymore) we do the check
> another time under lock.
> 
> It's really just to save extra lock when mgmt_master is not available.
> The check under lock is to handle case when the mgmt_master is removed
> while a mgmt eth is pending (corner case but still worth checking).
> 
> If you have suggestions on how to handle this corner case without
> introducing an extra lock in the read/write function, I would really
> appreaciate it.
> Now that I think about it, considering eth mgmt will be the main way and
> mdio as a fallback... wonder if the extra lock is acceptable anyway.
> In the near future ipq40xx will use qca8k, but will have his own regmap
> functions so we they won't be affected by these extra locking.
> 
> Don't know what is worst. Extra locking when mgmt_master is not
> avaialable or double check. (I assume for a cleaner code the extra lock
> is preferred)

I don't think there's a hidden bug in the code (other than the one I
mentioned, which is that you don't unlock or free resources at all on
the error path, which is quite severe), but also, I don't know if this
is such a performance-sensitive operation to justify a gratuitous
"optimization".
As you mention, if the Ethernet management will be the main I/O access
method, then the common case will take a single lock. You aren't really
saving much.