netdev - Re: [PATCH v3 net-next 2/4] net: dsa: Link aggregation support

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:   Mon, 07 Dec 2020 22:19:47 +0100
From:   Tobias Waldekranz <tobias@...dekranz.com>
To:     Andrew Lunn <andrew@...n.ch>
Cc:     davem@...emloft.net, kuba@...nel.org, vivien.didelot@...il.com,
        f.fainelli@...il.com, olteanv@...il.com, j.vosburgh@...il.com,
        vfalico@...il.com, andy@...yhouse.net, netdev@...r.kernel.org
Subject: Re: [PATCH v3 net-next 2/4] net: dsa: Link aggregation support

On Fri, Dec 04, 2020 at 03:20, Andrew Lunn <andrew@...n.ch> wrote:
>> +static int dsa_tree_setup_lags(struct dsa_switch_tree *dst)
>> +{
>> +	struct dsa_port *dp;
>> +	unsigned int num;
>> +
>> +	list_for_each_entry(dp, &dst->ports, list)
>> +		num = dp->ds->num_lags;
>> +
>> +	list_for_each_entry(dp, &dst->ports, list)
>> +		num = min(num, dp->ds->num_lags);
>
> Do you really need to loop over the list twice? Cannot num be
> initialised to UINT_MAX and then just do the second loop.

I was mostly paranoid about the case where, for some reason, the list of
ports was empty due to an invalid DT or something. But I now see that
since num is not initialized, that would not have helped.

So, is my paranoia valid, i.e. fix is `unsigned int num = 0`? Or can
that never happen, i.e. fix is to initialize to UINT_MAX and remove
first loop?

>> +static inline bool dsa_port_can_offload(struct dsa_port *dp,
>> +					struct net_device *dev)
>
> That name is a bit generic. We have a number of different offloads.
> The mv88E6060 cannot offload anything!

The name is intentionally generic as it answers the question "can this
dp offload requests for this netdev?"

>> +{
>> +	/* Switchdev offloading can be configured on: */
>> +
>> +	if (dev == dp->slave)
>> +		/* DSA ports directly connected to a bridge. */
>> +		return true;

This condition is the normal case of a bridged port, i.e. no LAG
involved.

>> +	if (dp->lag && dev == rtnl_dereference(dp->lag->dev))
>> +		/* DSA ports connected to a bridge via a LAG */
>> +		return true;

And then the indirect case of a bridged port under a LAG.

I am happy to take requests for a better name though.

>> +	return false;
>> +}
>
>> +static void dsa_lag_put(struct dsa_switch_tree *dst, struct dsa_lag *lag)
>> +{
>> +	if (!refcount_dec_and_test(&lag->refcount))
>> +		return;
>> +
>> +	clear_bit(lag->id, dst->lags.busy);
>> +	WRITE_ONCE(lag->dev, NULL);
>> +	memset(lag, 0, sizeof(*lag));
>> +}
>
> I don't know what the locking is here, but wouldn't it be safer to
> clear the bit last, after the memset and WRITE_ONCE.

All writers of dst->lags.busy are serialized with respect to dsa_lag_put
(on rtnl_lock), and concurrent readers (dsa_lag_dev_by_id) start by
checking busy before reading lag->dev. To my understanding, WRITE_ONCE
would insert the proper fence to make sure busy was cleared before
clearing dev?

>     Andrew