netdev - Re: [PATCH net-next v5 3/5] netdevsim: forward skbs from one connected port to another

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: <bf4760df-a78f-431d-8c33-b7a2f7fb393d@davidwei.uk>
Date: Wed, 3 Jan 2024 14:36:36 -0800
From: David Wei <dw@...idwei.uk>
To: Jiri Pirko <jiri@...nulli.us>
Cc: Jakub Kicinski <kuba@...nel.org>, Sabrina Dubroca <sd@...asysnail.net>,
 netdev@...r.kernel.org, "David S. Miller" <davem@...emloft.net>,
 Eric Dumazet <edumazet@...gle.com>, Paolo Abeni <pabeni@...hat.com>
Subject: Re: [PATCH net-next v5 3/5] netdevsim: forward skbs from one
 connected port to another

On 2024-01-02 03:13, Jiri Pirko wrote:
> Thu, Dec 28, 2023 at 02:46:31AM CET, dw@...idwei.uk wrote:
>> Forward skbs sent from one netdevsim port to its connected netdevsim
>> port using dev_forward_skb, in a spirit similar to veth.
>>
>> Add a tx_dropped variable to struct netdevsim, tracking the number of
>> skbs that could not be forwarded using dev_forward_skb().
>>
>> The xmit() function accessing the peer ptr is protected by an RCU read
>> critical section. The rcu_read_lock() is functionally redundant as since
>> v5.0 all softirqs are implicitly RCU read critical sections; but it is
>> useful for human readers.
>>
>> If another CPU is concurrently in nsim_destroy(), then it will first set
>> the peer ptr to NULL. This does not affect any existing readers that
>> dereferenced a non-NULL peer. Then, in unregister_netdevice(), there is
>> a synchronize_rcu() before the netdev is actually unregistered and
>> freed. This ensures that any readers i.e. xmit() that got a non-NULL
>> peer will complete before the netdev is freed.
>>
>> Any readers after the RCU_INIT_POINTER() but before synchronize_rcu()
>> will dereference NULL, making it safe.
>>
>> The codepath to nsim_destroy() and nsim_create() takes both the newly
>> added nsim_dev_list_lock and rtnl_lock. This makes it safe with
> 
> I don't see the rtnl_lock take in those functions.
> 
> 
> Otherwise, this patch looks fine to me.

For nsim_create(), rtnl_lock is taken in nsim_init_netdevsim(). For
nsim_destroy(), rtnl_lock is taken directly in the function.

What I mean here is, in the netdevsim device modification paths locks
are taken in this order:

devl_lock -> rtnl_lock

nsim_dev_list_lock is taken outside (not nested) of these.

In nsim_dev_peer_write() where two ports are linked, locks are taken in
this order:

nsim_dev_list_lock -> devl_lock -> rtnl_lock

This will not cause deadlocks and ensures that two ports being linked
are both valid.

> 
> 
>> concurrent calls to linking two netdevsims together.
>>
>> Signed-off-by: David Wei <dw@...idwei.uk>
>> ---
>> drivers/net/netdevsim/netdev.c    | 21 ++++++++++++++++++---
>> drivers/net/netdevsim/netdevsim.h |  1 +
>> 2 files changed, 19 insertions(+), 3 deletions(-)
>>
>> diff --git a/drivers/net/netdevsim/netdev.c b/drivers/net/netdevsim/netdev.c
>> index 434322f6a565..0009d0f1243f 100644
>> --- a/drivers/net/netdevsim/netdev.c
 +++ b/drivers/net/netdevsim/netdev.c
>> @@ -29,19 +29,34 @@
>> static netdev_tx_t nsim_start_xmit(struct sk_buff *skb, struct net_device *dev)
>> {
>> 	struct netdevsim *ns = netdev_priv(dev);
>> +	struct netdevsim *peer_ns;
>> +	int ret = NETDEV_TX_OK;
>>
>> 	if (!nsim_ipsec_tx(ns, skb))
>> 		goto out;
>>
>> +	rcu_read_lock();
>> +	peer_ns = rcu_dereference(ns->peer);
>> +	if (!peer_ns)
>> +		goto out_stats;
>> +
>> +	skb_tx_timestamp(skb);
>> +	if (unlikely(dev_forward_skb(peer_ns->netdev, skb) == NET_RX_DROP))
>> +		ret = NET_XMIT_DROP;
>> +
>> +out_stats:
>> +	rcu_read_unlock();
>> 	u64_stats_update_begin(&ns->syncp);
>> 	ns->tx_packets++;
>> 	ns->tx_bytes += skb->len;
>> +	if (ret == NET_XMIT_DROP)
>> +		ns->tx_dropped++;
>> 	u64_stats_update_end(&ns->syncp);
>> +	return ret;
>>
>> out:
>> 	dev_kfree_skb(skb);
>> -
>> -	return NETDEV_TX_OK;
>> +	return ret;
>> }
>>
>> static void nsim_set_rx_mode(struct net_device *dev)
>> @@ -70,6 +85,7 @@ nsim_get_stats64(struct net_device *dev, struct rtnl_link_stats64 *stats)
>> 		start = u64_stats_fetch_begin(&ns->syncp);
>> 		stats->tx_bytes = ns->tx_bytes;
>> 		stats->tx_packets = ns->tx_packets;
>> +		stats->tx_dropped = ns->tx_dropped;
>> 	} while (u64_stats_fetch_retry(&ns->syncp, start));
>> }
>>
>> @@ -302,7 +318,6 @@ static void nsim_setup(struct net_device *dev)
>> 	eth_hw_addr_random(dev);
>>
>> 	dev->tx_queue_len = 0;
>> -	dev->flags |= IFF_NOARP;
>> 	dev->flags &= ~IFF_MULTICAST;
>> 	dev->priv_flags |= IFF_LIVE_ADDR_CHANGE |
>> 			   IFF_NO_QUEUE;
>> diff --git a/drivers/net/netdevsim/netdevsim.h b/drivers/net/netdevsim/netdevsim.h
>> index 24fc3fbda791..083b1ee7a1a2 100644
>> --- a/drivers/net/netdevsim/netdevsim.h
>> +++ b/drivers/net/netdevsim/netdevsim.h
>> @@ -98,6 +98,7 @@ struct netdevsim {
>>
>> 	u64 tx_packets;
>> 	u64 tx_bytes;
>> +	u64 tx_dropped;
>> 	struct u64_stats_sync syncp;
>>
>> 	struct nsim_bus_dev *nsim_bus_dev;
>> -- 
>> 2.39.3
>>