From: Florian Westphal <fw@strlen.de>
To: Pablo Neira Ayuso <pablo@netfilter.org>
Cc: Florian Westphal <fw@strlen.de>, netfilter-devel@vger.kernel.org
Subject: Re: [PATCH nf,v2] netfilter: nf_tables: autoload modules from the abort path
Date: Thu, 23 Jan 2020 04:45:45 +0100 [thread overview]
Message-ID: <20200123034545.GS795@breakpoint.cc> (raw)
In-Reply-To: <20200122224947.iucrwyxmsrtm7ppe@salvia>
Pablo Neira Ayuso <pablo@netfilter.org> wrote:
> On Wed, Jan 22, 2020 at 11:28:08PM +0100, Florian Westphal wrote:
> > Pablo Neira Ayuso <pablo@netfilter.org> wrote:
> > > + list_for_each_entry(req, &net->nft.module_list, list) {
> > > + if (!strcmp(req->module, module_name) && req->done)
> > > + return 0;
> > > + }
> >
> > If the module is already on this list, why does it need to be
> > added a second time?
>
> The first time this finds no module on the list, then the module is
> added to the list and nft_request_module() returns -EAGAIN. This
> triggers abort path with autoload parameter set to true from
> nfnetlink, this sets the module done field to true.
I guess I was confused by the need for the "&& req->done" part.
AFAIU req->done is always true here.
> Now, on the second path, it will find that this already tried to load
> the module, so it does not add it again, nft_request_module() returns 0.
But the "I already tried this" is already implied by the presence of the
module name? Or did I misunderstand?
> Then, there is a look up to find the object that was missing. If
> module was successfully load, the object will be in place, otherwise
> -ENOENT is reported to userspace.
Good, that will prevent infite retries in case userspace requests
non-existent module.
> I can include this logic in the patch description in a v3.
That would be good, thanks!
> I run the syzbot reproducer for 1 hour and no problems, not sure how
> much I have to run it more. I guess the more time the better.
It triggers instantly for me provided:
1. CONFIG_MODULES=y (with MODULES=n the faulty code part isn't built...)
2. set "sysctl kernel.modprobe=/root/sleep1.sh"
I found that with normal modprobe the race window is rather small and
the thread doing the request_module has a decent chance of re-locking
the mutex before another syzkaller thread has a chance to alter the
current generation.
prev parent reply other threads:[~2020-01-23 3:45 UTC|newest]
Thread overview: 5+ messages / expand[flat|nested] mbox.gz Atom feed top
2020-01-22 21:17 [PATCH nf,v2] netfilter: nf_tables: autoload modules from the abort path Pablo Neira Ayuso
2020-01-22 22:28 ` Florian Westphal
2020-01-22 22:49 ` Pablo Neira Ayuso
2020-01-22 22:50 ` Pablo Neira Ayuso
2020-01-23 3:45 ` Florian Westphal [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=20200123034545.GS795@breakpoint.cc \
--to=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).