On Tue, May 01, 2007 at 01:33:26PM -0700, Jay Vosburgh wrote: > Andy Gospodarek <[EMAIL PROTECTED]> wrote: > > >> I need to use a bonding through two 802.1q tag VLAN virtual interface in > >> the > >> same physical network interface. When I bond interface with command : > >> ifenslave bond0 eth1.10 eth1.20 > >> Console and dmesg will continues report BUG: > >> --------------Output1 begin kernel 2.6.20-1.2933.fc6------------ > >> BUG: sleeping function called from invalid context at > >> include/asm/uaccess.h:500 > >> in_atomic():1, irqs_disabled():0 > >> [<c04ecadb>] copy_from_user+0x35/0x66 > >> [<c05c2e78>] dev_ethtool+0x51/0xa0d > >> [<c05ee503>] tcp_current_mss+0x71/0xdf > >> [<c05ee169>] tcp_rcv_established+0x74f/0x7de > >> [<c04cf48f>] selinux_netlbl_sock_rcv_skb+0x1d/0x5b > >> [<c05f3a78>] tcp_v4_do_rcv+0x1bf/0x49c > >> [<c05f5f1b>] tcp_v4_rcv+0x564/0x8b1 > >> [<d8a3ac03>] vlan_dev_ioctl+0x7b/0xa7 [8021q] > >> [<d8a3ab88>] vlan_dev_ioctl+0x0/0xa7 [8021q] > >> [<d8a53102>] bond_update_speed_duplex+0x88/0xd7 [bonding] > >> [<d8a5688e>] bond_mii_monitor+0x0/0x403 [bonding] > >> [<d8a56c01>] bond_mii_monitor+0x373/0x403 [bonding] > > > >Jay Vosburgh and I have been working on something to address this and > >other issues related to possible sleeping activities that might come > >from the monitoring functions. I feel like we are getting close to > >something so I hope we can post something soon. > > > >Care to add anything, Jay? > > I was thinking about this one a couple of days ago; it's a > little tricky to get out of, since we nominally need to do the ethtool > calls holding only rtnl, but don't want to cycle rtnl for each monitor > pass. > > The warning only shows up with VLANs because VLAN get_settings() > calls dev_ethtool() directly, and the copy_from_user() in dev_ethtool() > is the complaining party here. >
Agreed. All locks held when calling bond_update_speed_duplex for a vlan interface and need to be non bh-locks and the call needs to be done as a user-process. This might be hard to avoid with bond_enslave but could be avoided with some code restructing. > Calling an end device driver's get_settings() directly doesn't > trigger the warning. I think that's safe without rtnl (at least insofar > as the slave won't vanish), since the slave cannot vanish without > bonding being notified via the NETDEV_UNREGISTER notifier. This does look to be something specific to vlan interfaces being enslaved. Which makes me wonder: Why are we supporting that anyway? I *guess* there might be some crazy network admin that might like this, but it doesn't seem right... > > -J > > --- > -Jay Vosburgh, IBM Linux Technology Center, [EMAIL PROTECTED] > - > To unsubscribe from this list: send the line "unsubscribe netdev" in > the body of a message to [EMAIL PROTECTED] > More majordomo info at http://vger.kernel.org/majordomo-info.html - To unsubscribe from this list: send the line "unsubscribe netdev" in the body of a message to [EMAIL PROTECTED] More majordomo info at http://vger.kernel.org/majordomo-info.html