linux-kernel - Re: Re: [PATCH 5.4 086/129] net: core: use listified Rx for GRO_NORMAL in napi_gro

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>] [day] [month] [year] [list]

Date:   Wed, 9 Sep 2020 10:50:39 +0200
From:   Greg Kroah-Hartman <gregkh@...uxfoundation.org>
To:     김현순 <h10.kim@...sung.com>
Cc:     Edward Cree <ecree@...arflare.com>,
        "linux-kernel@...r.kernel.org" <linux-kernel@...r.kernel.org>,
        "stable@...r.kernel.org" <stable@...r.kernel.org>,
        Alexander Lobakin <alobakin@...nk.ru>,
        "David S. Miller" <davem@...emloft.net>
Subject: Re: Re: [PATCH 5.4 086/129] net: core: use listified Rx for
 GRO_NORMAL in napi_gro_receive()

On Wed, Sep 09, 2020 at 04:21:01PM +0900, 김현순 wrote:
> Dear Greg,
> 
>  
> 
> I didn't intent to gain performance from this patch. Indeed i requested this
> patch to resolve the bug we are experiencing.
> 
>  
> 
> Different from kernel 4.19, napi gro made additional data structure called
> rx_list, and for some reason, the list is always used after napi_gro_flush,
> which is intended to be created to flush all gro (both merged and non-merged)
> packets to network stack. Network packets received from network interface are
> passed to rx_list once, and the rx_list is later flushed to network stack.
> However, on the other hand, napi_gro_receive doesn't insert packet to rx_list.
> Instead, it flushes to network stack directly. This causes out of order because
> there might be some packets in rx_list (which has groups of packets that are
> not flushed to network stack yet). If the packet from the same stream remains
> in rx_list, and napi_gro_receive on next packet (from the same stream) is
> called, this posterior packet will arrive to network stack first, which causes
> out of order.
> 
>  
> 
> And two patches mentioned by Edward are already in the branch. The patch i am
> requesting is somewhat missing in android kernel, not sure the reason why.

Ah, you are right, this is odd.  Edward, those patches are already in
the 5.4.y tree, which makes this patch not being present odd.

I'll go add it back....

thanks,

greg k-h