| Message ID | 1549090.s6anOn0kWk@voltaire |
|---|---|
| State | RFC, archived |
| Headers | show |
Hi Marek, the problem seems to be a little bit more complex. Your latest patch does not solve the problem. But I found out, that commenting out the following line in your patch makes bat0 work: slave->master = master; But as this is the core of "enslaving" a device to a master device, this breaks the complete concept behind this I guess (I'm not a skilled kernel developer). From this I conclude that there might be a bug somewhere deeper in the kernel version I use. I don't want to give up too early, but it looks a little bit as if this "enslaving concept using rtnl" might not be usable in these older kernels. What do you think? Regards, Andreas
I started with a freshly unpacked source code of the 2014.4.0 release and applied only your latest patch. As expected from the tests done so far with the old kernel effectively not calling netdev_set_master allows the usage of the latest batman-adv version in combination with older kernels (at least the 2.6.32 I tested with). I think this patch is worth to be integrated into the next batman-adv version. Regards, Andreas "B.A.T.M.A.N" <b.a.t.m.a.n-bounces@lists.open-mesh.org> schrieb am 19.02.2015 10:40:34: > Von: Marek Lindner <mareklindner@neomailbox.ch> > An: The list for a Better Approach To Mobile Ad-hoc Networking > <b.a.t.m.a.n@lists.open-mesh.org>, > Datum: 19.02.2015 10:42 > Betreff: Re: [B.A.T.M.A.N.] Antwort: Re: Antwort: Antwort: Re: > Antwort: Re: Question concerning batman-adv bug #173 "Mesh packets on bat0" > Gesendet von: "B.A.T.M.A.N" <b.a.t.m.a.n-bounces@lists.open-mesh.org> > > On Thursday, February 19, 2015 09:31:08 Andreas Pape wrote: > > the problem seems to be a little bit more complex. Your latest patch does > > not solve the problem. > > > > But I found out, that commenting out the following line in your patch > > makes bat0 work: > > > > slave->master = master; > > > > But as this is the core of "enslaving" a device to a master device, this > > breaks the complete concept behind this I guess (I'm not a skilled kernel > > developer). From this I conclude that there might be a bug somewhere > > deeper in the kernel version I use. I don't want to give up too early, > > but it looks a little bit as if this "enslaving concept using rtnl" might > > not be usable in these older kernels. What do you think? > > Please try the attached patch instead. This time we are replacing > the function > with our own function doing nothing at all. The net_dev->master > variable seems > to be reserved for interface bonding and shouldn't be touched at allon these > ancient kernels. > > Cheers, > Marek > [Anhang "0001-batman-adv-ignore-netdev_set_master-calls-on- > kernels.patch" gelöscht von Andreas Pape/Phoenix Contact] [Anhang > "signature.asc" gelöscht von Andreas Pape/Phoenix Contact]
On Thursday, February 19, 2015 11:28:24 Andreas Pape wrote: > I started with a freshly unpacked source code of the 2014.4.0 release and > applied only your latest patch. > > As expected from the tests done so far with the old kernel effectively not > calling netdev_set_master allows the usage of the latest batman-adv > version in combination with older kernels (at least the 2.6.32 I tested > with). > > I think this patch is worth to be integrated into the next batman-adv > version. Thanks for testing! I'll stage the patch for inclusion. Cheers, Marek
From c01b68bc11a857770c2799f0b033852e05559431 Mon Sep 17 00:00:00 2001 From: Marek Lindner <mareklindner@neomailbox.ch> Date: Thu, 19 Feb 2015 12:03:39 +0800 Subject: [PATCH] batman-adv: make netdev_set_master() generic for kernels older than 2.6.39 batman-adv calls netdev_master_upper_dev_link() which is replaced with with netdev_set_master() for kernels older than 3.9 via compat.h. Prior to 2.6.39 netdev_set_master() contained Linux bonding calls needed to setup bonding devices. Calling this function from batman-adv leads to unexpected behavior when current batman-adv versions are used on these older kernels. To fix the situation compat.h now ships its own implementation of netdev_set_master() for kernels older than 2.6.39. Reported-by: Andreas Pape <APape@phoenixcontact.com> Signed-off-by: Marek Lindner <mareklindner@neomailbox.ch> --- compat.h | 24 ++++++++++++++++++++++++ 1 file changed, 24 insertions(+) diff --git a/compat.h b/compat.h index 27d8a21..f1da3da 100644 --- a/compat.h +++ b/compat.h @@ -191,6 +191,30 @@ static inline int batadv_param_set_copystring(const char *val, #define kstrtoul strict_strtoul #define kstrtol strict_strtol +static inline int batadv_netdev_set_master(struct net_device *slave, + struct net_device *master) +{ + struct net_device *old = slave->master; + + ASSERT_RTNL(); + + if (master) { + if (old) + return -EBUSY; + dev_hold(master); + } + + slave->master = master; + + if (old) { + synchronize_net(); + dev_put(old); + } + return 0; +} + +#define netdev_set_master batadv_netdev_set_master + /* Hack for removing ndo_add/del_slave at the end of net_device_ops. * This is somewhat ugly because it requires that ndo_validate_addr * is at the end of this struct in soft-interface.c. -- 2.1.4