From: Phil Sutter <phil@nwl.cc>
To: Florian Westphal <fw@strlen.de>
Cc: Pablo Neira Ayuso <pablo@netfilter.org>, netfilter-devel@vger.kernel.org
Subject: Re: [PATCH iptables 4/4] nft: keep old cache around until batch is refreshed in case of ERESTART
Date: Mon, 20 May 2019 14:48:37 +0200 [thread overview]
Message-ID: <20190520124837.GB31548@orbyte.nwl.cc> (raw)
In-Reply-To: <20190520120620.nxxl65syr2b7eal7@breakpoint.cc>
On Mon, May 20, 2019 at 02:06:20PM +0200, Florian Westphal wrote:
> Phil Sutter <phil@nwl.cc> wrote:
> > On Sun, May 19, 2019 at 06:45:08PM +0200, Phil Sutter wrote:
> > [...]
> > > The only way to make the above work is by keeping the original cache
> > > copy around until mnl_batch_talk has finally succeeded or failed with
> > > something else than ERESTART.
> >
> > How about a completely different approach:
> >
> > If memory serves right (and from reading the related Red Hat ticket),
> > the actual problem we're trying to solve is that iptables-nft-restore
> > creates NFT_MSG_DELTABLE only if that table exists already at the point
> > of parsing but another client might create it in the mean time before
> > committing.
> >
> > My idea for solving this was to unconditionally create NFT_MSG_NEWTABLE
> > followed by NFT_MSG_DELTABLE - in case the table exists, the first one
> > is a noop; in case the table doesn't exist, the second one won't provoke
> > an error message from kernel space.
>
> Does that work even work?
> new table x
> del table x
> add rule to x // table was deleted?
>
> Or are you talking about a new/del/new sequence?
Oh, yes of course. Existing iptables-nft-restore does:
1) delete table x if exists
2) add table x
3) add table x content
My idea is to:
+ 0) add table x
- 1) delete table x if exists
+ 1) delete table x
2) add table x
3) add table x content
> If it works, ok/fine, but it seems ugly.
Did you consider rule insert with index in your batch replay logic? I
did when duplicating it for nft, but decided it's not worth it and
people using 'add rule ... index IDX' have been warned already anyway.
> > Since NFT_MSG_DELTABLE removes the table recursively, we don't need to
> > care about any content added by the other client.
>
> Yes, we can't do this in the --noflush case though.
My state from our last talk about it was "--noflush users are screwed
anyway". :)
> > Or is this about a generic solution for all commands not just
> > iptables-nft-restore (without --noflush)?
>
> Its for ipt-nft-restore, including --noflush.
>
> It would be good though to also speed up 'iptables-nft -A' later on
> by eliding cache completely.
The problem is that we can't unconditionally create NFT_MSG_NEWCHAIN
because that will reset the chain policy if non-default. So we need at
least a cache containing tables and chains, even for that simple rule
append command.
Cheers, Phil
prev parent reply other threads:[~2019-05-20 12:48 UTC|newest]
Thread overview: 8+ messages / expand[flat|nested] mbox.gz Atom feed top
2019-05-19 11:51 [PATCH iptables 1/4] nft: add struct nft_cache Pablo Neira Ayuso
2019-05-19 11:51 ` [PATCH iptables 2/4] nft: statify nft_rebuild_cache() Pablo Neira Ayuso
2019-05-19 11:51 ` [PATCH iptables 3/4] nft: add flush_cache() Pablo Neira Ayuso
2019-05-19 11:51 ` [PATCH iptables 4/4] nft: keep old cache around until batch is refreshed in case of ERESTART Pablo Neira Ayuso
2019-05-19 16:45 ` Phil Sutter
2019-05-20 12:00 ` Phil Sutter
2019-05-20 12:06 ` Florian Westphal
2019-05-20 12:48 ` Phil Sutter [this message]
Reply instructions:
You may reply publicly to this message via plain-text email
using any one of the following methods:
* Save the following mbox file, import it into your mail client,
and reply-to-all from there: mbox
Avoid top-posting and favor interleaved quoting:
https://en.wikipedia.org/wiki/Posting_style#Interleaved_style
* Reply using the --to, --cc, and --in-reply-to
switches of git-send-email(1):
git send-email \
--in-reply-to=20190520124837.GB31548@orbyte.nwl.cc \
--to=phil@nwl.cc \
--cc=fw@strlen.de \
--cc=netfilter-devel@vger.kernel.org \
--cc=pablo@netfilter.org \
/path/to/YOUR_REPLY
https://kernel.org/pub/software/scm/git/docs/git-send-email.html
* If your mail client supports setting the In-Reply-To header
via mailto: links, try the mailto: link
Be sure your reply has a Subject: header at the top and a blank line
before the message body.
This is a public inbox, see mirroring instructions
for how to clone and mirror all data and code used for this inbox;
as well as URLs for NNTP newsgroup(s).