Message ID | 20230411144955.1604591-1-vladimir.oltean@nxp.com (mailing list archive) |
---|---|
State | Changes Requested |
Delegated to: | Netdev Maintainers |
Headers | show |
Series | [v3,net] net: ipv4/ipv6 addrconf: call igmp{,6}_group_dropped() while dev is still up | expand |
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 >
On Wed, Apr 12, 2023 at 04:34:00PM +0300, Ido Schimmel wrote: > 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. Hmm, yeah, fair. I was operating on the premise that callers of dev_mc_add()/dev_mc_del() would be sane enough to keep track of their calls, and at least aspire to not leave something behind. Clearly not the case here, if user space can add a unicast/multicast filter and run... So I guess I'm left to clean up the stale addresses at DSA removal time, and remove the WARN_ON(). Actually, looking at ndo_dflt_fdb_add() and ndo_dflt_fdb_del() again, what concerns me more is this: if (is_unicast_ether_addr(addr) || is_link_local_ether_addr(addr)) err = dev_uc_add_excl(dev, addr); I.O.W. this: $ bridge fdb add 01:80:c2:00:00:0e dev swp0 self local generates a filtering call for a unicast address. And I don't know about other drivers, but in DSA, we assume that unicast addresses are unicast, and we have separate API for that vs multicast...
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); }
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 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(-)