From: Ido Schimmel <idosch@nvidia.com>
To: Vladimir Oltean <vladimir.oltean@nxp.com>
Cc: netdev@vger.kernel.org, Jakub Kicinski <kuba@kernel.org>,
"David S. Miller" <davem@davemloft.net>,
Eric Dumazet <edumazet@google.com>,
Paolo Abeni <pabeni@redhat.com>, David Ahern <dsahern@kernel.org>
Subject: Re: [PATCH v3 net] net: ipv4/ipv6 addrconf: call igmp{,6}_group_dropped() while dev is still up
Date: Wed, 12 Apr 2023 16:34:00 +0300 [thread overview]
Message-ID: <ZDazSM5UsPPjQuKr@shredder> (raw)
In-Reply-To: <20230411144955.1604591-1-vladimir.oltean@nxp.com>
On Tue, Apr 11, 2023 at 05:49:55PM +0300, Vladimir Oltean wrote:
> ipv4 devinet calls ip_mc_down(), and ipv6 calls addrconf_ifdown(), and
> both of these eventually result in calls to dev_mc_del(), either through
> igmp_group_dropped() or igmp6_group_dropped().
>
> The problem is that dev_mc_del() does call __dev_set_rx_mode(), but this
> will not propagate all the way to the ndo_set_rx_mode() of the device,
> because of this check:
>
> /* dev_open will call this function so the list will stay sane. */
> if (!(dev->flags&IFF_UP))
> return;
>
> and the NETDEV_DOWN notifier is emitted while the interface is already
> down. OTOH we have NETDEV_GOING_DOWN which is emitted a bit earlier -
> see:
>
> dev_close_many()
> -> __dev_close_many()
> -> call_netdevice_notifiers(NETDEV_GOING_DOWN, dev);
> -> dev->flags &= ~IFF_UP;
> -> call_netdevice_notifiers(NETDEV_DOWN, dev);
>
> Normally this oversight is easy to miss, because the addresses aren't
> lost, just not synced to the device until the next up event.
>
> DSA does some processing in its dsa_slave_set_rx_mode(), and assumes
> that all addresses that were synced are also unsynced by the time the
> device is unregistered. Due to that assumption not being satisfied,
> the WARN_ON(!list_empty(&dp->mdbs)); from dsa_switch_release_ports()
> triggers, and we leak memory corresponding to the multicast addresses
> that were never synced.
>
> Minimal reproducer:
> ip link set swp0 up
> ip link set swp0 down
> echo 0000:00:00.5 > /sys/bus/pci/drivers/mscc_felix/unbind
Even with the proposed fix, wouldn't you get the same leak with the
following reproducer?
ip link set dev swp0 up
bridge fdb add 01:02:03:04:05:06 dev swp0 self local
ip link set dev swp0 down
echo 0000:00:00.5 > /sys/bus/pci/drivers/mscc_felix/unbind
If so, I wonder how other drivers that allocate memory in their
ndo_set_rx_mode() deal with this problem. I would imagine that they
flush the addresses in their ndo_stop() or as part of device dismantle.
>
> The proposal is to respond to that slightly earlier notifier with the
> IGMP address deletion, so that the ndo_set_rx_mode() of the device does
> actually get called.
>
> Fixes: 5e8a1e03aa4d ("net: dsa: install secondary unicast and multicast addresses as host FDB/MDB")
> Signed-off-by: Vladimir Oltean <vladimir.oltean@nxp.com>
> ---
> v2->v3:
> Returned to the original strategy, with Ido's modification applied
> (to only touch the netdev notifier values, not the inetaddr notifier
> values).
>
> v2 at:
> https://patchwork.kernel.org/project/netdevbpf/patch/20230410195220.1335670-1-vladimir.oltean@nxp.com/
>
> net/ipv4/devinet.c | 2 +-
> net/ipv6/addrconf.c | 6 +++---
> 2 files changed, 4 insertions(+), 4 deletions(-)
>
> diff --git a/net/ipv4/devinet.c b/net/ipv4/devinet.c
> index 5deac0517ef7..679c9819f25b 100644
> --- a/net/ipv4/devinet.c
> +++ b/net/ipv4/devinet.c
> @@ -1588,7 +1588,7 @@ static int inetdev_event(struct notifier_block *this, unsigned long event,
> /* Send gratuitous ARP to notify of link change */
> inetdev_send_gratuitous_arp(dev, in_dev);
> break;
> - case NETDEV_DOWN:
> + case NETDEV_GOING_DOWN:
> ip_mc_down(in_dev);
> break;
> case NETDEV_PRE_TYPE_CHANGE:
> diff --git a/net/ipv6/addrconf.c b/net/ipv6/addrconf.c
> index 3797917237d0..f4a3b2693d6a 100644
> --- a/net/ipv6/addrconf.c
> +++ b/net/ipv6/addrconf.c
> @@ -3670,12 +3670,12 @@ static int addrconf_notify(struct notifier_block *this, unsigned long event,
> }
> break;
>
> - case NETDEV_DOWN:
> + case NETDEV_GOING_DOWN:
> case NETDEV_UNREGISTER:
> /*
> * Remove all addresses from this interface.
> */
> - addrconf_ifdown(dev, event != NETDEV_DOWN);
> + addrconf_ifdown(dev, event != NETDEV_GOING_DOWN);
> break;
>
> case NETDEV_CHANGENAME:
> @@ -6252,7 +6252,7 @@ static void dev_disable_change(struct inet6_dev *idev)
>
> netdev_notifier_info_init(&info, idev->dev);
> if (idev->cnf.disable_ipv6)
> - addrconf_notify(NULL, NETDEV_DOWN, &info);
> + addrconf_notify(NULL, NETDEV_GOING_DOWN, &info);
> else
> addrconf_notify(NULL, NETDEV_UP, &info);
> }
> --
> 2.34.1
>
next prev parent reply other threads:[~2023-04-12 13:34 UTC|newest]
Thread overview: 3+ messages / expand[flat|nested] mbox.gz Atom feed top
2023-04-11 14:49 [PATCH v3 net] net: ipv4/ipv6 addrconf: call igmp{,6}_group_dropped() while dev is still up Vladimir Oltean
2023-04-12 13:34 ` Ido Schimmel [this message]
2023-04-12 13:53 ` Vladimir Oltean
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=ZDazSM5UsPPjQuKr@shredder \
--to=idosch@nvidia.com \
--cc=davem@davemloft.net \
--cc=dsahern@kernel.org \
--cc=edumazet@google.com \
--cc=kuba@kernel.org \
--cc=netdev@vger.kernel.org \
--cc=pabeni@redhat.com \
--cc=vladimir.oltean@nxp.com \
/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).