[<prev] [next>] [<thread-prev] [thread-next>] [day] [month] [year] [list]
Message-ID: <20241107-invisible-skylark-of-attack-e44be1@leitao>
Date: Thu, 7 Nov 2024 05:51:40 -0800
From: Breno Leitao <leitao@...ian.org>
To: Eric Dumazet <edumazet@...gle.com>
Cc: "David S. Miller" <davem@...emloft.net>,
David Ahern <dsahern@...nel.org>, Jakub Kicinski <kuba@...nel.org>,
Paolo Abeni <pabeni@...hat.com>, Simon Horman <horms@...nel.org>,
David Ahern <dsahern@...il.com>, netdev@...r.kernel.org,
linux-kernel@...r.kernel.org, kernel-team@...a.com
Subject: Re: [PATCH net] ipmr: Fix access to mfc_cache_list without lock held
Hello Eric,
On Thu, Nov 07, 2024 at 02:13:14PM +0100, Eric Dumazet wrote:
> On Thu, Nov 7, 2024 at 12:03 PM Breno Leitao <leitao@...ian.org> wrote:
> >
> > Accessing `mr_table->mfc_cache_list` is protected by an RCU lock. In the
> > following code flow, the lock is not held, causing the following error
> > when `RCU_PROVE` is not held.
> >
> > 6.12.0-rc5-kbuilder-01145-gbac17284bdcb #33 Tainted: G E N
> > -----------------------------
> > net/ipv4/ipmr_base.c:313 RCU-list traversed in non-reader section!!
> >
> > rcu_scheduler_active = 2, debug_locks = 1
> > 2 locks held by RetransmitAggre/3519:
> > #0: ffff88816188c6c0 (nlk_cb_mutex-ROUTE){+.+.}-{3:3}, at: __netlink_dump_start+0x8a/0x290
> > #1: ffffffff83fcf7a8 (rtnl_mutex){+.+.}-{3:3}, at: rtnl_dumpit+0x6b/0x90
> >
> > stack backtrace:
> > lockdep_rcu_suspicious
> > mr_table_dump
> > ipmr_rtm_dumproute
> > rtnl_dump_all
> > rtnl_dumpit
> > netlink_dump
> > __netlink_dump_start
> > rtnetlink_rcv_msg
> > netlink_rcv_skb
> > netlink_unicast
> > netlink_sendmsg
> >
> > Fix accessing `mfc_cache_list` without holding the RCU read lock. Adds
> > `rcu_read_lock()` and `rcu_read_unlock()` around `mr_table_dump()` to
> > prevent RCU-list traversal in non-reader section.
> >
> > Since `mr_table_dump()` is the only function that touches the list, that
> > might be the only critical section in `ipmr_rtm_dumproute()` that needs
> > to be protected in ipmr_rtm_dumproute().
> >
> > Signed-off-by: Breno Leitao <leitao@...ian.org>
> > Fixes: cb167893f41e ("net: Plumb support for filtering ipv4 and ipv6 multicast route dumps")
> > ---
> > net/ipv4/ipmr.c | 2 ++
> > 1 file changed, 2 insertions(+)
> >
> > diff --git a/net/ipv4/ipmr.c b/net/ipv4/ipmr.c
> > index 089864c6a35eec146a1ba90c22d79245f8e48158..bb855f32f328024f384a2fa58f42fc227705206e 100644
> > --- a/net/ipv4/ipmr.c
> > +++ b/net/ipv4/ipmr.c
> > @@ -2612,8 +2612,10 @@ static int ipmr_rtm_dumproute(struct sk_buff *skb, struct netlink_callback *cb)
> > NL_SET_ERR_MSG(cb->extack, "ipv4: MR table does not exist");
> > return -ENOENT;
> > }
> > + rcu_read_lock();
> > err = mr_table_dump(mrt, skb, cb, _ipmr_fill_mroute,
> > &mfc_unres_lock, &filter);
> > + rcu_read_unlock();
> > return skb->len ? : err;
> > }
> >
> >
>
> What about net/ipv6/ip6mr.c ip6mr_rtm_dumproute() ?
That one might require as well.
> In my opinion, since we still hold RTNL in these paths, we should
> change the lockdep annotation.
I don't have much experience mixing locks like this. Is it safe to mix
and match rtnl and RCUs like this?
I have the impression that, when iterating a RCU protected list *without* being in the read-side
critical sections, the RCU doesn't know that someone might be traversing
the list, and remove the element mid air (mroute_clean_tables()?). Is
this model incorrect?
> Then later we can remove RTNL from these dump operations.
Do you mean that, execute the dump operation without holding the RTNL,
thus, relying solely on RCU?
> diff --git a/net/ipv4/ipmr_base.c b/net/ipv4/ipmr_base.c
> index 271dc03fc6dbd9b35db4d5782716679134f225e4..f0af12a2f70bcdf5ba54321bf7ebebe798318abb
> 100644
> --- a/net/ipv4/ipmr_base.c
> +++ b/net/ipv4/ipmr_base.c
> @@ -310,7 +310,8 @@ int mr_table_dump(struct mr_table *mrt, struct sk_buff *skb,
> if (filter->filter_set)
> flags |= NLM_F_DUMP_FILTERED;
>
> - list_for_each_entry_rcu(mfc, &mrt->mfc_cache_list, list) {
> + list_for_each_entry_rcu(mfc, &mrt->mfc_cache_list, list,
> + lockdep_rtnl_is_held()) {
> if (e < s_e)
> goto next_entry;
> if (filter->dev &&
Clarifying next steps: Would you like me to review/test and submit, or
are you planning to send it officially?
Thanks for your feedback,
--breno
Powered by blists - more mailing lists