Re: [PATCH] net: release dst entry in dev_hard_start_xmit()

[Eric Dumazet <dada1@cosmosbay.com>]

Jarek Poplawski a écrit :
> On Tue, May 12, 2009 at 09:44:52PM +0200, Eric Dumazet wrote:
>> Jarek Poplawski a écrit :
>>> On Tue, May 12, 2009 at 10:21:12AM +0200, Eric Dumazet wrote:
>>>> Eric Dumazet a écrit :
>>> ...
>>>>> List of devices that must clear this flag is :
>>>>>
>>>>> - loopback device, because it calls netif_rx() and quoting Patrick :
>>>>>     "ip_route_input() doesn't accept loopback addresses, so loopback packets
>>>>>      already need to have a dst_entry attached."
>>>>> - appletalk/ipddp.c : needs skb->dst in its xmit function
>>>>>
>>>>> - And all devices that call again dev_queue_xmit() from their xmit function
>>>>> (as some classifiers need skb->dst) : bonding, macvlan, eql, ifb, hdlc_fr
>>> Why not vlan (and/or maybe others in net/ using dev_queue_xmit)?
>>>
>> Yes I spoted vlan earlier this afternoon.
>>
>> For other net/*, I didnt not find candidates yet.
> 
> Hmm..., if vlan, then why not pppoe (and/or maybe others in drivers/net/
> using dev_queue_xmit)?
> 

As I said, I didnt found other relevant uses, but I am probably wrong :)

About ppoe for example, two calls to dev_queue_xmit() are not relevant.

One is from 

static int pppoe_sendmsg(struct kiocb *iocb, struct socket *sock,
          struct msghdr *m, size_t total_len)

This is a normal direct call, not a call from its ndo_start_xmit()

Second one is from

static int __pppoe_xmit(struct sock *sk, struct sk_buff *skb)

Same observation, there is no impact for this one as well.



So we have to care on device drivers that have a ndo_start_xmit() call that could
re-enter dev_queue_xmit(). Not care about drivers that call dev_queue_xmit() from
other paths.