[pf4freebsd] Re: Maturity of this port?

[Max Laier]

On Wednesday 26 May 2004 02:05, Arnaud Pignard wrote:
> At 08:09 25/05/2004, you wrote:
> >D'oh; I really do need ALTQ.
>
> If you really need ALTQ, try here :
> http://www.rofug.ro/projects/freebsd-altq/

Chances are, that my patchset is more stable that the rofug.ro one which I 
evolved from (I have said this, haven't I?).

> Work fine with 5.2.1 and drivers is avaible for most all good network card.
> (don't specified in changelog but em driver is also stable)

1) As I have said (several times) the drivers from rofug.ro work with this 
patchset without problems. If they are stable there, they are stable here as 
well.
2) The patchset from rofug.ro has a couple of problems:
 a) The locking is incomplete. Passing pktattr on the stack isn't MPSAFE by 
    design.
 b) It provides ALTQ3 support only, which is - in my opinion - no longer state 
    of-the-art. The built-in classifier has very limited capabilities, the 
    syntax is a pain and the implementation isn't all that powerful either.
 c) It is a bit dated in terms of being in sync with KAME as well as in being 
    in sync with FreeBSD. I am afraid there are currently no efforts to change 
    this situation.
 d) As it "grew" over the years it has some issues both in drivers and the 
    altq code itself. That is why I started from zero, importing the necessary 
    pieces step by step, to clean it up and have a completely working thing 
    that will eventually be ready for import.

> Also i prefer altq config file and i'm not sure that's you can use altqstat
> with pf.

That is well choice of taste, but as I tried to explain: ALTQ3 isn't MPSAFE by 
design! Altqstat is not useable with pf, but pf has its own way $pfctl -vvsq 
gives the same information in a (imo) nicer way.

I will not stop anyone from implementing ALTQ3-support (forgetting about the 
locking problems for a moment), but I think that there is much more power in 
the pf-approach. If you are brave, you can implement the altq classifier in 
terms of a pfil_hooks-consumer useing mbuf_tag classification, this will be 
easier in terms of locking but does not seem worthwhile to me.

-- 
Best regards,				| mlaier at freebsd.org
Max Laier				| ICQ #67774661
http://pf4freebsd.love2party.net/	| mlaier at EFnet
-------------- next part --------------
A non-text attachment was scrubbed...
Name: not available
Type: application/pgp-signature
Size: 194 bytes
Desc: signature
Url : http://lists.freebsd.org/pipermail/freebsd-pf/attachments/20040916/c80b0b41/attachment.bin