netdev - Re: [net-next PATCH v1 08/11] net: rocker: add get flow API operation

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 for Android: free password hash cracker in your pocket

[<prev] [next>] [<thread-prev] [thread-next>] [day] [month] [year] [list]

Date:	Tue, 06 Jan 2015 06:59:43 -0800
From:	John Fastabend <john.fastabend@...il.com>
To:	Scott Feldman <sfeldma@...il.com>
CC:	Thomas Graf <tgraf@...g.ch>,
	Jiří Pírko <jiri@...nulli.us>,
	Jamal Hadi Salim <jhs@...atatu.com>,
	"simon.horman@...ronome.com" <simon.horman@...ronome.com>,
	Netdev <netdev@...r.kernel.org>,
	"David S. Miller" <davem@...emloft.net>,
	Andy Gospodarek <andy@...yhouse.net>
Subject: Re: [net-next PATCH v1 08/11] net: rocker: add get flow API operation

On 01/05/2015 11:40 PM, Scott Feldman wrote:
> On Wed, Dec 31, 2014 at 11:48 AM, John Fastabend
> <john.fastabend@...il.com> wrote:
>> Add operations to get flows. I wouldn't mind cleaning this code
>> up a bit but my first attempt to do this used macros which shortered
>> the code up but when I was done I decided it just made the code
>> unreadable and unmaintainable.
>>
>> I might think about it a bit more but this implementation albeit
>> a bit long and repeatative is easier to understand IMO.
>
> Dang, you put a lot of work into this one.
>
> Something doesn't feel right though.  In this case, rocker driver just
> happened to have cached all the flow/group stuff in hash tables in
> software, so you don't need to query thru to the device to extract the
> if_flow info.  What doesn't feel right is all the work need in the
> driver.  For each and every driver.  get_flows needs to go above
> driver, somehow.

Another option is to have a software cache in the flow_table.c I
was trying to avoid caching as I really don't expect 'get' operations
to be fast path and going to hardware seems good enough for me.
Other than its a bit annoying to write the mapping code.

If you don't have a cache then somewhere there has to be a mapping
from hardware flow descriptors to the flow descriptors used by the
flow API. Like I noted I tried to help by using macros and helper
routines but in the end I simply decided it convoluted the code to
much and made it hard to debug.

>
> Seems the caller of if_flow already knows the flows pushed down with
> add_flows/del_flows, and with the err handling can't mess it up.

yes the caller could know if it cached them which it doesn't. We
can add a cache if its helpful. You may have multiple users of the
API (both in-kernel and user space) though so I don't think you can
push it much beyond the flow_table.c.

>
> Is one use-case for get_flows to recover from a fatal OS/driver crash,
> and to rely on hardware to recover flow set?  In this rocker example,
> that's not going to work because driver didn't get thru to device to
> get_flows.  I think I'd like to know more about the use-cases of
> get_flows.

Its helpful for debugging. And if you have multiple consumers it
may be helpful to "learn" what other consumers are doing. I don't
have any concrete cases at the moment though.

For the CLI case its handy to add some flows, forget what you did,
and then do a get to refresh your mind. Not likely a problem for
"real" management software.

At least its not part of the UAPI so we could tweak/improve it as
much as we wanted. Any better ideas? I'm open to suggestions on this
one.

>
> -scott
>

-- 
John Fastabend         Intel Corporation
--
To unsubscribe from this list: send the line "unsubscribe netdev" in
the body of a message to majordomo@...r.kernel.org
More majordomo info at  http://vger.kernel.org/majordomo-info.html