From: Florian Westphal <fw@strlen.de>
To: Peter Zijlstra <peterz@infradead.org>
Cc: Florian Westphal <fw@strlen.de>, netdev@vger.kernel.org
Subject: Re: [PATCH net-next 2/8] rtnetlink: add rtnl_register_module
Date: Tue, 7 Nov 2017 10:47:51 +0100 [thread overview]
Message-ID: <20171107094751.GM9424@breakpoint.cc> (raw)
In-Reply-To: <20171107092458.GF3857@worktop>
Peter Zijlstra <peterz@infradead.org> wrote:
> Something like the below would go some way toward sanitizing this stuff;
> rcu_assign_pointer() is a store-release, meaning it happens after
> everything coming before.
>
> Therefore, when you observe that tab (through rcu_dereference) you're
> guaranteed to see the thing initialized. The memory ordering on the
> consume side is through an address dependency; we need to have completed
> the load of the tab pointer before we can compute the address of its
> members and load from there, these are not things a CPU is allowed to
> reorder (lets forget about Alpha).
>
> Quite possibly, if rtnl_unregister() is called from module unload, this
> is broken; in that case we'd need something like:
>
> rcu_assign_pointer(rtnl_msg_handler[protocol], NULL);
> /*
> * Ensure nobody can still observe our old protocol handler
> * before continuing to free the module that includes the
> * functions called from it.
> */
> synchronize_rcu();
>
> ---
>
> diff --git a/net/core/rtnetlink.c b/net/core/rtnetlink.c
> index 5ace48926b19..25391c7b9c5d 100644
> --- a/net/core/rtnetlink.c
> +++ b/net/core/rtnetlink.c
> @@ -63,6 +63,7 @@ struct rtnl_link {
> rtnl_doit_func doit;
> rtnl_dumpit_func dumpit;
> unsigned int flags;
> + struct rcu_head rcu;
> };
>
> static DEFINE_MUTEX(rtnl_mutex);
> @@ -172,14 +173,15 @@ int __rtnl_register(int protocol, int msgtype,
> BUG_ON(protocol < 0 || protocol > RTNL_FAMILY_MAX);
> msgindex = rtm_msgindex(msgtype);
>
> - tab = rcu_dereference_raw(rtnl_msg_handlers[protocol]);
> - if (tab == NULL) {
> - tab = kcalloc(RTM_NR_MSGTYPES, sizeof(*tab), GFP_KERNEL);
> - if (tab == NULL)
> - return -ENOBUFS;
> + if (WARN_ONCE(rtnl_msg_handler[protocol],
> + "Double registration for protocol: %d\n", protcol))
> + return -EEXIST;
I would expect this to trigger all the time, due to
rtnl_register(AF_INET, RTM_GETROUTE, ...
rtnl_register(AF_INET, RTM_GETADDR, ...
etc.
> @@ -227,15 +231,13 @@ int rtnl_unregister(int protocol, int msgtype)
> msgindex = rtm_msgindex(msgtype);
>
> rtnl_lock();
> - handlers = rtnl_dereference(rtnl_msg_handlers[protocol]);
> + handlers = rtnl_msg_handlers[protocol];
> if (!handlers) {
> rtnl_unlock();
> return -ENOENT;
> }
> -
> - handlers[msgindex].doit = NULL;
> - handlers[msgindex].dumpit = NULL;
> - handlers[msgindex].flags = 0;
> + rcu_assign_pointer(rtnl_msg_handler[protocol], NULL);
> + kfree_rcu(handlers, rcu);
This unregisters all handlers of "protocol" instead of
"protocol:msgtype".
next prev parent reply other threads:[~2017-11-07 9:48 UTC|newest]
Thread overview: 20+ messages / expand[flat|nested] mbox.gz Atom feed top
2017-11-06 10:51 rtnetlink: fix dump+module unload races, take 2 Florian Westphal
2017-11-06 10:51 ` [PATCH net-next 1/8] rtnetlink: Revert "rtnetlink: add reference counting to prevent module unload while dump is in progress" Florian Westphal
2017-11-06 10:51 ` [PATCH net-next 2/8] rtnetlink: add rtnl_register_module Florian Westphal
2017-11-06 12:44 ` Peter Zijlstra
2017-11-07 6:11 ` Florian Westphal
2017-11-07 9:10 ` Peter Zijlstra
2017-11-07 9:24 ` Peter Zijlstra
2017-11-07 9:47 ` Florian Westphal [this message]
2017-11-07 14:57 ` Peter Zijlstra
2017-11-08 1:27 ` Stephen Hemminger
2017-11-13 7:21 ` Florian Westphal
2017-11-13 7:55 ` Peter Zijlstra
2017-11-13 7:59 ` Florian Westphal
2017-11-07 9:43 ` Florian Westphal
2017-11-06 10:51 ` [PATCH net-next 3/8] qtr: use rtnl_register_module Florian Westphal
2017-11-06 10:51 ` [PATCH net-next 4/8] bridge: " Florian Westphal
2017-11-06 10:51 ` [PATCH net-next 5/8] can: " Florian Westphal
2017-11-06 10:51 ` [PATCH net-next 6/8] decnet: " Florian Westphal
2017-11-06 10:51 ` [PATCH net-next 7/8] phonet: " Florian Westphal
2017-11-06 10:51 ` [PATCH net-next 8/8] mpls: " Florian Westphal
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=20171107094751.GM9424@breakpoint.cc \
--to=fw@strlen.de \
--cc=netdev@vger.kernel.org \
--cc=peterz@infradead.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).