From: Steven Rostedt <rostedt@goodmis.org> Date: 2014-01-28 21:05:48
At Red Hat we base our real-time kernel off of 3.10.27 and do lots of
stress testing on that kernel. This has discovered some bugs that we
can hit with the vanilla 3.10.27 kernel (no -rt patches applied).
I sent out a bug fix that can cause a crash with the current 3.10.27
when you add and then remove the sit module. That patch is obsoleted by
these patches, as that patch was not enough.
A previous patch that was backported:
Upstream commit 205983c43700ac3a81e7625273a3fa83cd2759b5
sit: allow to use rtnl ops on fb tunnel
Had a depenency on commit 5e6700b3bf98 ("sit: add support of x-netns")
which was not backported. The dependency was only on part of that
commit which is what I backported.
The other upstream commit 9434266f2c645d4fcf62a03a8e36ad8075e37943
sit: fix use after free of fb_tunnel_dev
fixes another bug we encountered, it also fixes the 3.10.27 bug
where removing the sit module cause the crash. This is the patch
that obsoletes my previous patch.
-- Steve
Steven Rostedt (Red Hat) (1):
sit: Unregister sit devices with rtnl_link_ops
Willem de Bruijn (1):
sit: fix use after free of fb_tunnel_dev
----
net/ipv6/sit.c | 30 +++++++++++++++++++++++++-----
1 file changed, 25 insertions(+), 5 deletions(-)
From: Steven Rostedt <rostedt@goodmis.org> Date: 2014-01-28 21:05:46
From: Willem de Bruijn <willemb@google.com>
[ Upstream commit 9434266f2c645d4fcf62a03a8e36ad8075e37943 ]
Bug: The fallback device is created in sit_init_net and assumed to be
freed in sit_exit_net. First, it is dereferenced in that function, in
sit_destroy_tunnels:
struct net *net = dev_net(sitn->fb_tunnel_dev);
Prior to this, rtnl_unlink_register has removed all devices that match
rtnl_link_ops == sit_link_ops.
Commit 205983c43700 added the line
+ sitn->fb_tunnel_dev->rtnl_link_ops = &sit_link_ops;
which cases the fallback device to match here and be freed before it
is last dereferenced.
Fix: This commit adds an explicit .delllink callback to sit_link_ops
that skips deallocation at rtnl_unlink_register for the fallback
device. This mechanism is comparable to the one in ip_tunnel.
It also modifies sit_destroy_tunnels and its only caller sit_exit_net
to avoid the offending dereference in the first place. That double
lookup is more complicated than required.
Test: The bug is only triggered when CONFIG_NET_NS is enabled. It
causes a GPF only when CONFIG_DEBUG_SLAB is enabled. Verified that
this bug exists at the mentioned commit, at davem-net HEAD and at
3.11.y HEAD. Verified that it went away after applying this patch.
Fixes: 205983c43700 ("sit: allow to use rtnl ops on fb tunnel")
Signed-off-by: Willem de Bruijn <willemb@google.com>
Acked-by: Nicolas Dichtel <redacted>
Acked-by: Eric Dumazet <edumazet@google.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
Signed-off-by: Steven Rostedt <rostedt@goodmis.org>
---
net/ipv6/sit.c | 19 +++++++++++++++----
1 file changed, 15 insertions(+), 4 deletions(-)
From: Steven Rostedt <rostedt@goodmis.org> Date: 2014-01-28 21:06:50
From: "Steven Rostedt (Red Hat)" <rostedt@goodmis.org>
The backport of upstream commit 205983c43700ac3 ("sit: allow to use rtnl
ops on fb tunnel") had a dependency on commit 5e6700b3bf98 ("sit: add
support of x-netns"). The dependency was on the way that commit
unregistered the sit devices.
Since the entire commit 5e6700b3bf98 is not necessary for a backport
we only need to backport the part required for the backport of
205983c43700ac3. This is the loop to unregister the sit devices where
dev->rtnl_link_ops == &sit_link_ops.
Cc: Nicolas Dichtel <redacted>
Signed-off-by: Steven Rostedt <rostedt@goodmis.org>
---
net/ipv6/sit.c | 11 ++++++++++-
1 file changed, 10 insertions(+), 1 deletion(-)
@@ -1536,7 +1540,12 @@ static void __net_exit sit_destroy_tunnels(struct sit_net *sitn, struct list_heat=rtnl_dereference(sitn->tunnels[prio][h]);while(t!=NULL){-unregister_netdevice_queue(t->dev,head);+/* If dev is in the same netns, it has already+*beenaddedtothelistbythepreviousloop.+*/+if(dev_net(t->dev)!=net)+unregister_netdevice_queue(t->dev,+head);t=rtnl_dereference(t->next);}}
From: David Miller <davem@davemloft.net> Date: 2014-01-29 07:52:17
From: Steven Rostedt <rostedt@goodmis.org>
Date: Tue, 28 Jan 2014 15:57:56 -0500
At Red Hat we base our real-time kernel off of 3.10.27 and do lots of
stress testing on that kernel. This has discovered some bugs that we
can hit with the vanilla 3.10.27 kernel (no -rt patches applied).
I sent out a bug fix that can cause a crash with the current 3.10.27
when you add and then remove the sit module. That patch is obsoleted by
these patches, as that patch was not enough.
A previous patch that was backported:
Upstream commit 205983c43700ac3a81e7625273a3fa83cd2759b5
sit: allow to use rtnl ops on fb tunnel
Had a depenency on commit 5e6700b3bf98 ("sit: add support of x-netns")
which was not backported. The dependency was only on part of that
commit which is what I backported.
The other upstream commit 9434266f2c645d4fcf62a03a8e36ad8075e37943
sit: fix use after free of fb_tunnel_dev
fixes another bug we encountered, it also fixes the 3.10.27 bug
where removing the sit module cause the crash. This is the patch
that obsoletes my previous patch.
Greg, these have my blessing, please apply these to 3.10 -stable,
thanks a lot!
From: Nicolas Dichtel <hidden> Date: 2014-01-29 11:04:11
Le 28/01/2014 21:57, Steven Rostedt a écrit :
At Red Hat we base our real-time kernel off of 3.10.27 and do lots of
stress testing on that kernel. This has discovered some bugs that we
can hit with the vanilla 3.10.27 kernel (no -rt patches applied).
I sent out a bug fix that can cause a crash with the current 3.10.27
when you add and then remove the sit module. That patch is obsoleted by
these patches, as that patch was not enough.
Can you explain a bit more which problem remains after that patch?
I wonder if a problem remains also with ip6_tunnel.ko (net/ipv6/ip6_tunnel.c),
the same problem was spotted into this module.
A previous patch that was backported:
Upstream commit 205983c43700ac3a81e7625273a3fa83cd2759b5
sit: allow to use rtnl ops on fb tunnel
Had a depenency on commit 5e6700b3bf98 ("sit: add support of x-netns")
which was not backported. The dependency was only on part of that
commit which is what I backported.
I cannot comment directly the patch, it was an attachement, hence I put my
comments here.
In patch 0001-sit-Unregister-sit-devices-with-rtnl_link_ops.patch, I wonder how
'if (dev_net(t->dev) != net)' can be wrong. If commit 5e6700b3bf98 ("sit: add
support of x-netns") has not been backported, this test is always true.
The other upstream commit 9434266f2c645d4fcf62a03a8e36ad8075e37943
sit: fix use after free of fb_tunnel_dev
fixes another bug we encountered, it also fixes the 3.10.27 bug
where removing the sit module cause the crash. This is the patch
that obsoletes my previous patch.
-- Steve
Steven Rostedt (Red Hat) (1):
sit: Unregister sit devices with rtnl_link_ops
Willem de Bruijn (1):
sit: fix use after free of fb_tunnel_dev
----
net/ipv6/sit.c | 30 +++++++++++++++++++++++++-----
1 file changed, 25 insertions(+), 5 deletions(-)
From: Steven Rostedt <rostedt@goodmis.org> Date: 2014-01-29 12:57:50
On Wed, 29 Jan 2014 12:04:05 +0100
Nicolas Dichtel [off-list ref] wrote:
Le 28/01/2014 21:57, Steven Rostedt a écrit :
quoted
At Red Hat we base our real-time kernel off of 3.10.27 and do lots of
stress testing on that kernel. This has discovered some bugs that we
can hit with the vanilla 3.10.27 kernel (no -rt patches applied).
I sent out a bug fix that can cause a crash with the current 3.10.27
when you add and then remove the sit module. That patch is obsoleted by
these patches, as that patch was not enough.
Can you explain a bit more which problem remains after that patch?
I wonder if a problem remains also with ip6_tunnel.ko (net/ipv6/ip6_tunnel.c),
the same problem was spotted into this module.
Hmm, OK it may only need the first version of the patch. It was one of
those cases where it didn't fix the second bug, so we added the full
patch as well. Then we found the second bug but never tried all the
tests with the smaller version of the patch. I'll put back the first
patch and then ask our QA to retest it with the older version and the
other patch.
quoted
A previous patch that was backported:
Upstream commit 205983c43700ac3a81e7625273a3fa83cd2759b5
sit: allow to use rtnl ops on fb tunnel
Had a depenency on commit 5e6700b3bf98 ("sit: add support of x-netns")
which was not backported. The dependency was only on part of that
commit which is what I backported.
I cannot comment directly the patch, it was an attachement, hence I put my
It's not really an attachment. It was sent with quilt mail, which only
makes it look like one. What mail client are you using? mutt, alpine,
evolution, claws-mail, and thunderbird all show it as a normal patch.
I do know that k9-mail on android makes it into an attachment.
comments here.
In patch 0001-sit-Unregister-sit-devices-with-rtnl_link_ops.patch, I wonder how
'if (dev_net(t->dev) != net)' can be wrong. If commit 5e6700b3bf98 ("sit: add
support of x-netns") has not been backported, this test is always true.
Should it just be removed then? This has fixed our bugs, but perhaps it
opened new ones we haven't detected yet. I can remove the if and
unregister and see if it still works. Or perhaps calling unregister all
the time isn't bad.
-- Steve
quoted
The other upstream commit 9434266f2c645d4fcf62a03a8e36ad8075e37943
sit: fix use after free of fb_tunnel_dev
fixes another bug we encountered, it also fixes the 3.10.27 bug
where removing the sit module cause the crash. This is the patch
that obsoletes my previous patch.
-- Steve
Steven Rostedt (Red Hat) (1):
sit: Unregister sit devices with rtnl_link_ops
Willem de Bruijn (1):
sit: fix use after free of fb_tunnel_dev
----
net/ipv6/sit.c | 30 +++++++++++++++++++++++++-----
1 file changed, 25 insertions(+), 5 deletions(-)
From: Steven Rostedt <rostedt@goodmis.org> Date: 2014-01-29 12:59:26
On Tue, 28 Jan 2014 23:52:14 -0800 (PST)
David Miller [off-list ref] wrote:
Greg, these have my blessing, please apply these to 3.10 -stable,
thanks a lot!
Hi David,
Thanks for the blessing :-) But I want to work with Nicolas a bit more
to make sure that I have a correct backport. Just because it fixed some
of our bugs doesn't mean that I didn't make new ones.
Thanks,
-- Steve
From: Nicolas Dichtel <hidden> Date: 2014-01-29 16:04:18
Le 29/01/2014 13:57, Steven Rostedt a écrit :
On Wed, 29 Jan 2014 12:04:05 +0100
Nicolas Dichtel [off-list ref] wrote:
quoted
Le 28/01/2014 21:57, Steven Rostedt a écrit :
quoted
At Red Hat we base our real-time kernel off of 3.10.27 and do lots of
stress testing on that kernel. This has discovered some bugs that we
can hit with the vanilla 3.10.27 kernel (no -rt patches applied).
I sent out a bug fix that can cause a crash with the current 3.10.27
when you add and then remove the sit module. That patch is obsoleted by
these patches, as that patch was not enough.
Can you explain a bit more which problem remains after that patch?
I wonder if a problem remains also with ip6_tunnel.ko (net/ipv6/ip6_tunnel.c),
the same problem was spotted into this module.
Hmm, OK it may only need the first version of the patch. It was one of
those cases where it didn't fix the second bug, so we added the full
patch as well. Then we found the second bug but never tried all the
tests with the smaller version of the patch. I'll put back the first
patch and then ask our QA to retest it with the older version and the
other patch.
quoted
quoted
A previous patch that was backported:
Upstream commit 205983c43700ac3a81e7625273a3fa83cd2759b5
sit: allow to use rtnl ops on fb tunnel
Had a depenency on commit 5e6700b3bf98 ("sit: add support of x-netns")
which was not backported. The dependency was only on part of that
commit which is what I backported.
I cannot comment directly the patch, it was an attachement, hence I put my
It's not really an attachment. It was sent with quilt mail, which only
makes it look like one. What mail client are you using? mutt, alpine,
evolution, claws-mail, and thunderbird all show it as a normal patch.
I do know that k9-mail on android makes it into an attachment.
Thunderbird. The patch was show as a normal aptch, but when I do "reply all", I
get an empty mail.
quoted
comments here.
In patch 0001-sit-Unregister-sit-devices-with-rtnl_link_ops.patch, I wonder how
'if (dev_net(t->dev) != net)' can be wrong. If commit 5e6700b3bf98 ("sit: add
support of x-netns") has not been backported, this test is always true.
Should it just be removed then? This has fixed our bugs, but perhaps it
opened new ones we haven't detected yet. I can remove the if and
unregister and see if it still works. Or perhaps calling unregister all
the time isn't bad.
Ok, I've think a bit more to this problem. First, let's explain it.
rmmod sit
=> sit_cleanup()
=> rtnl_link_unregister()
=> __rtnl_kill_links()
=> for_each_netdev(net, dev) {
if (dev->rtnl_link_ops == ops)
ops->dellink(dev, &list_kill);
}
**at this point, the FB device is deleted (and all sit tunnels)**
=> unregister_pernet_device()
=> unregister_pernet_operations()
=> ops_exit_list()
=> sit_exit_net()
=> sit_destroy_tunnels()
in this function, no tunnel is found
unregister_netdevice_queue(sitn->fb_tunnel_dev, &list);
** second deletion, here is the bug **
Now, what happen on netns deletion with the previous patch? Only sit_exit_net()
will be called, hence with the previous patch, the fb device will not be
destroyed, netns will leak.
Your patch serie seems to be the good way to go (note that patch 1/2 does not
compile) but I think the fix is smaller because we don't have x-netns.
Here is my proposal, if you agree, I will send the same patch for ip6_tunnnel,
which have the netns leak.
From d101450583c3a472a2a94904cfe13fd4e7d2f519 Mon Sep 17 00:00:00 2001
From: Nicolas Dichtel <redacted>
Date: Wed, 29 Jan 2014 16:40:30 +0100
Subject: [PATCH] sit: fix double free of fb_tunnel_dev on exit
This problem was fixed upstream by commit 9434266f2c64 ("sit: fix use after free
of fb_tunnel_dev").
The upstream patch depends on upstream commit 5e6700b3bf98 ("sit: add support of
x-netns"), which was not backported into 3.10 branch.
First, explain the problem: when the sit module is unloaded, sit_cleanup() is
called.
rmmod sit
=> sit_cleanup()
=> rtnl_link_unregister()
=> __rtnl_kill_links()
=> for_each_netdev(net, dev) {
if (dev->rtnl_link_ops == ops)
ops->dellink(dev, &list_kill);
}
At this point, the FB device is deleted (and all sit tunnels).
=> unregister_pernet_device()
=> unregister_pernet_operations()
=> ops_exit_list()
=> sit_exit_net()
=> sit_destroy_tunnels()
In this function, no tunnel is found.
=> unregister_netdevice_queue(sitn->fb_tunnel_dev, &list);
We delete the FB device a second time here!
Because we cannot simply remove the second deletion (sit_exit_net() must remove
the FB device when a netns is deleted), we add an rtnl ops which delete all sit
device excepting the FB device and thus we can keep the explicit deletion in
sit_exit_net().
CC: Steven Rostedt <rostedt@goodmis.org>
Signed-off-by: Nicolas Dichtel <redacted>
---
net/ipv6/sit.c | 10 ++++++++++
1 file changed, 10 insertions(+)
From: Willem de Bruijn <willemb@google.com> Date: 2014-01-29 17:21:59
quoted hunk
From d101450583c3a472a2a94904cfe13fd4e7d2f519 Mon Sep 17 00:00:00 2001
From: Nicolas Dichtel <redacted>
Date: Wed, 29 Jan 2014 16:40:30 +0100
Subject: [PATCH] sit: fix double free of fb_tunnel_dev on exit
This problem was fixed upstream by commit 9434266f2c64 ("sit: fix use after free
of fb_tunnel_dev").
The upstream patch depends on upstream commit 5e6700b3bf98 ("sit: add support of
x-netns"), which was not backported into 3.10 branch.
First, explain the problem: when the sit module is unloaded, sit_cleanup() is
called.
rmmod sit
=> sit_cleanup()
=> rtnl_link_unregister()
=> __rtnl_kill_links()
=> for_each_netdev(net, dev) {
if (dev->rtnl_link_ops == ops)
ops->dellink(dev, &list_kill);
}
At this point, the FB device is deleted (and all sit tunnels).
=> unregister_pernet_device()
=> unregister_pernet_operations()
=> ops_exit_list()
=> sit_exit_net()
=> sit_destroy_tunnels()
In this function, no tunnel is found.
=> unregister_netdevice_queue(sitn->fb_tunnel_dev, &list);
We delete the FB device a second time here!
Because we cannot simply remove the second deletion (sit_exit_net() must remove
the FB device when a netns is deleted), we add an rtnl ops which delete all sit
device excepting the FB device and thus we can keep the explicit deletion in
sit_exit_net().
CC: Steven Rostedt <rostedt@goodmis.org>
Signed-off-by: Nicolas Dichtel <redacted>
---
net/ipv6/sit.c | 10 ++++++++++
1 file changed, 10 insertions(+)
This looks good to me. It is the same as the backport "sit: fix use
after free of fb_tunnel_dev" (9434266f2c64), minus the small code
cleanup at the end of that patch that is not relevant to 3.10.27 (do
not define sit_net *sitn in sit_exit_net if it is not used there. But in
3.10.27 it is used, in unregister_netdevice_queue).
From: Steven Rostedt <rostedt@goodmis.org> Date: 2014-01-29 20:48:16
On Wed, 29 Jan 2014 17:04:12 +0100
Nicolas Dichtel [off-list ref] wrote:
Your patch serie seems to be the good way to go (note that patch 1/2 does not
compile) but I think the fix is smaller because we don't have x-netns.
Here is my proposal, if you agree, I will send the same patch for ip6_tunnnel,
which have the netns leak.
Hold on. Seems that the kernels that were being tested in QA had more
code than what I was testing. Clark had backported "sit: fix use after
free of fb_tunnel_dev" and that was what was causing the
unlist_netdevice() to be missed.
When I started working on vanilla 3.10.27 as well, I first did my
original patch (which just removes the call to
unregister_netdevice_queue() from sit_exit_net()). I asked to have that
added to our kernel for testing, and they told me it was already there
via Clark's backport. Then I did the full backport as well and looked
for the leak. I'm now thinking that the full backport is not needed as
that was what caused the leak.
According to commit 9434266f2c645d4fcf62a03a8e36ad8075e37943 "sit: fix
use after free of fb_tunnel_dev", it states:
Bug: The fallback device is created in sit_init_net and assumed to
be freed in sit_exit_net. First, it is dereferenced in that
function, in sit_destroy_tunnels:
struct net *net = dev_net(sitn->fb_tunnel_dev);
Prior to this, rtnl_unlink_register has removed all devices that
match rtnl_link_ops == sit_link_ops.
Commit 205983c43700 added the line
+ sitn->fb_tunnel_dev->rtnl_link_ops = &sit_link_ops;
which cases the fallback device to match here and be freed before it
is last dereferenced.
Commit 205983c43700 was backported to 3.10, but without commit
5e6700b3bf98 "sit: add support of x-netns" which was what added the
net = dev_net(sitn->fb_tunnel_dev);
Which looks to me that the only reason I need to port back commit
5e6700b3bf98 is if I add the full backport of 9434266f2c645d4f.
Seems to me that my original patch may be good enough. The one that I
said this series obsoletes.
Note, I've talked with the people that are doing the testing, and I'm
having them revert all changes except for that one fix and rerun the
tests again. I should know the results by tomorrow.
Let me know if "sit: fix use after free of fb_tunnel_dev" still needs
to be backported due to some other way that the fallback device can be
dereferenced after being freed.
-- Steve
From: Nicolas Dichtel <hidden> Date: 2014-01-30 09:28:53
Le 29/01/2014 21:48, Steven Rostedt a écrit :
On Wed, 29 Jan 2014 17:04:12 +0100
Nicolas Dichtel [off-list ref] wrote:
quoted
Your patch serie seems to be the good way to go (note that patch 1/2 does not
compile) but I think the fix is smaller because we don't have x-netns.
Here is my proposal, if you agree, I will send the same patch for ip6_tunnnel,
which have the netns leak.
Hold on. Seems that the kernels that were being tested in QA had more
code than what I was testing. Clark had backported "sit: fix use after
free of fb_tunnel_dev" and that was what was causing the
unlist_netdevice() to be missed.
When I started working on vanilla 3.10.27 as well, I first did my
original patch (which just removes the call to
unregister_netdevice_queue() from sit_exit_net()). I asked to have that
added to our kernel for testing, and they told me it was already there
via Clark's backport. Then I did the full backport as well and looked
for the leak. I'm now thinking that the full backport is not needed as
that was what caused the leak.
According to commit 9434266f2c645d4fcf62a03a8e36ad8075e37943 "sit: fix
use after free of fb_tunnel_dev", it states:
Bug: The fallback device is created in sit_init_net and assumed to
be freed in sit_exit_net. First, it is dereferenced in that
function, in sit_destroy_tunnels:
struct net *net = dev_net(sitn->fb_tunnel_dev);
Prior to this, rtnl_unlink_register has removed all devices that
match rtnl_link_ops == sit_link_ops.
Commit 205983c43700 added the line
+ sitn->fb_tunnel_dev->rtnl_link_ops = &sit_link_ops;
which cases the fallback device to match here and be freed before it
is last dereferenced.
Commit 205983c43700 was backported to 3.10, but without commit
5e6700b3bf98 "sit: add support of x-netns" which was what added the
net = dev_net(sitn->fb_tunnel_dev);
Which looks to me that the only reason I need to port back commit
5e6700b3bf98 is if I add the full backport of 9434266f2c645d4f.
Seems to me that my original patch may be good enough. The one that I
said this series obsoletes.
Note, I've talked with the people that are doing the testing, and I'm
having them revert all changes except for that one fix and rerun the
tests again. I should know the results by tomorrow.
Let me know if "sit: fix use after free of fb_tunnel_dev" still needs
to be backported due to some other way that the fallback device can be
dereferenced after being freed.
Steve, I think the patch I sent yesterday is the good fix. At the end, it's
a backport of Willem's patch. Note that he also ack that patch.
The first version you sent (which removes
unregister_netdevice_queue(sitn->fb_tunnel_dev, &list)) will introduce a
memory leak when the user destroy a netns.
From: Nicolas Dichtel <hidden> Date: 2014-01-30 10:15:58
This problem was fixed upstream by commit 1e9f3d6f1c40 ("ip6tnl: fix use after
free of fb_tnl_dev").
The upstream patch depends on upstream commit 0bd8762824e7 ("ip6tnl: add x-netns
support"), which was not backported into 3.10 branch.
First, explain the problem: when the ip6_tunnel module is unloaded,
ip6_tunnel_cleanup() is called.
rmmod ip6_tunnel
=> ip6_tunnel_cleanup()
=> rtnl_link_unregister()
=> __rtnl_kill_links()
=> for_each_netdev(net, dev) {
if (dev->rtnl_link_ops == ops)
ops->dellink(dev, &list_kill);
}
At this point, the FB device is deleted (and all ip6tnl tunnels).
=> unregister_pernet_device()
=> unregister_pernet_operations()
=> ops_exit_list()
=> ip6_tnl_exit_net()
=> ip6_tnl_destroy_tunnels()
=> t = rtnl_dereference(ip6n->tnls_wc[0]);
unregister_netdevice_queue(t->dev, &list);
We delete the FB device a second time here!
The previous fix removes these lines, which fix this double free. But the patch
introduces a memory leak when a netns is destroyed, because the FB device is
never deleted. By adding an rtnl ops which delete all ip6tnl device excepting
the FB device, we can keep this exlicit removal in ip6_tnl_destroy_tunnels().
CC: Steven Rostedt <rostedt@goodmis.org>
CC: Willem de Bruijn <willemb@google.com>
Signed-off-by: Nicolas Dichtel <redacted>
---
net/ipv6/ip6_tunnel.c | 10 ++++++++++
1 file changed, 10 insertions(+)
From: Nicolas Dichtel <hidden> Date: 2014-01-30 10:15:59
This reverts commit 22c3ec552c29cf4bd4a75566088950fe57d860c4.
This patch is not the right fix, it introduces a memory leak when a netns is
destroyed (the FB device is never deleted).
Signed-off-by: Nicolas Dichtel <redacted>
---
net/ipv6/ip6_tunnel.c | 2 ++
1 file changed, 2 insertions(+)
From: Nicolas Dichtel <hidden> Date: 2014-01-30 10:16:00
This problem was fixed upstream by commit 9434266f2c64 ("sit: fix use after free
of fb_tunnel_dev").
The upstream patch depends on upstream commit 5e6700b3bf98 ("sit: add support of
x-netns"), which was not backported into 3.10 branch.
First, explain the problem: when the sit module is unloaded, sit_cleanup() is
called.
rmmod sit
=> sit_cleanup()
=> rtnl_link_unregister()
=> __rtnl_kill_links()
=> for_each_netdev(net, dev) {
if (dev->rtnl_link_ops == ops)
ops->dellink(dev, &list_kill);
}
At this point, the FB device is deleted (and all sit tunnels).
=> unregister_pernet_device()
=> unregister_pernet_operations()
=> ops_exit_list()
=> sit_exit_net()
=> sit_destroy_tunnels()
In this function, no tunnel is found.
=> unregister_netdevice_queue(sitn->fb_tunnel_dev, &list);
We delete the FB device a second time here!
Because we cannot simply remove the second deletion (sit_exit_net() must remove
the FB device when a netns is deleted), we add an rtnl ops which delete all sit
device excepting the FB device and thus we can keep the explicit deletion in
sit_exit_net().
CC: Steven Rostedt <rostedt@goodmis.org>
Signed-off-by: Nicolas Dichtel <redacted>
Acked-by: Willem de Bruijn <willemb@google.com>
---
net/ipv6/sit.c | 10 ++++++++++
1 file changed, 10 insertions(+)
From: Steven Rostedt <rostedt@goodmis.org> Date: 2014-01-30 13:31:08
On Thu, 30 Jan 2014 10:28:43 +0100
Nicolas Dichtel [off-list ref] wrote:
Steve, I think the patch I sent yesterday is the good fix. At the end, it's
a backport of Willem's patch. Note that he also ack that patch.
The first version you sent (which removes
unregister_netdevice_queue(sitn->fb_tunnel_dev, &list)) will introduce a
memory leak when the user destroy a netns.
Hi Nicolas,
I reverted my patches and applied and tested your patches locally and
they passed my first line testing. I'm going to have them entered into
our test suite, after removing our other patches, and see if they solve
all the bugs that we were tripping over.
I'll let you know when these are finished.
Thanks!
-- Steve
From: Nicolas Dichtel <hidden> Date: 2014-01-30 13:43:03
Le 30/01/2014 14:31, Steven Rostedt a écrit :
On Thu, 30 Jan 2014 10:28:43 +0100
Nicolas Dichtel [off-list ref] wrote:
quoted
Steve, I think the patch I sent yesterday is the good fix. At the end, it's
a backport of Willem's patch. Note that he also ack that patch.
The first version you sent (which removes
unregister_netdevice_queue(sitn->fb_tunnel_dev, &list)) will introduce a
memory leak when the user destroy a netns.
Hi Nicolas,
I reverted my patches and applied and tested your patches locally and
they passed my first line testing. I'm going to have them entered into
our test suite, after removing our other patches, and see if they solve
all the bugs that we were tripping over.
I'll let you know when these are finished.
From: Steven Rostedt <rostedt@goodmis.org> Date: 2014-01-30 22:10:34
On Thu, 30 Jan 2014 14:42:57 +0100
Nicolas Dichtel [off-list ref] wrote:
quoted
I'll let you know when these are finished.
Thank you for testing.
Our short tests that were crashing have all now passed :-) with your
patches applied.
We are now running our full tier tests on the code, but I don't think
we need to wait. Please add my tags to your patches, and hopefully we
can get David to give us a new blessing on your patch set.
David and Greg,
You can ignore my patches as it seems that Nicolas's patches solve the
issues we have been seeing on 3.10.
Nicolas,
Thanks for all your effort in helping us out here :-)
Reported-by: Steven Rostedt <redacted>
Tested-by: Steven Rostedt <redacted> (and our entire MRG team)
Tested-by: "Luis Claudio R. Goncalves" <redacted>
Tested-by: John Kacur <jkacur@redhat.com>
I don't usually give my Red Hat email for things like this, but as this
was found and tested with the Red Hat infrastructure, with the help
from Luis and John (and others), I would like to give them credit too.
-- Steve
From: Nicolas Dichtel <hidden> Date: 2014-01-31 08:24:18
This problem was fixed upstream by commit 1e9f3d6f1c40 ("ip6tnl: fix use after
free of fb_tnl_dev").
The upstream patch depends on upstream commit 0bd8762824e7 ("ip6tnl: add x-netns
support"), which was not backported into 3.10 branch.
First, explain the problem: when the ip6_tunnel module is unloaded,
ip6_tunnel_cleanup() is called.
rmmod ip6_tunnel
=> ip6_tunnel_cleanup()
=> rtnl_link_unregister()
=> __rtnl_kill_links()
=> for_each_netdev(net, dev) {
if (dev->rtnl_link_ops == ops)
ops->dellink(dev, &list_kill);
}
At this point, the FB device is deleted (and all ip6tnl tunnels).
=> unregister_pernet_device()
=> unregister_pernet_operations()
=> ops_exit_list()
=> ip6_tnl_exit_net()
=> ip6_tnl_destroy_tunnels()
=> t = rtnl_dereference(ip6n->tnls_wc[0]);
unregister_netdevice_queue(t->dev, &list);
We delete the FB device a second time here!
The previous fix removes these lines, which fix this double free. But the patch
introduces a memory leak when a netns is destroyed, because the FB device is
never deleted. By adding an rtnl ops which delete all ip6tnl device excepting
the FB device, we can keep this exlicit removal in ip6_tnl_destroy_tunnels().
CC: Steven Rostedt <rostedt@goodmis.org>
CC: Willem de Bruijn <willemb@google.com>
Signed-off-by: Nicolas Dichtel <redacted>
Reported-by: Steven Rostedt <redacted>
Tested-by: Steven Rostedt <redacted> (and our entire MRG team)
Tested-by: "Luis Claudio R. Goncalves" <redacted>
Tested-by: John Kacur <jkacur@redhat.com>
---
v2: add Steven's tags
net/ipv6/ip6_tunnel.c | 10 ++++++++++
1 file changed, 10 insertions(+)
From: Nicolas Dichtel <hidden> Date: 2014-01-31 08:24:19
This reverts commit 22c3ec552c29cf4bd4a75566088950fe57d860c4.
This patch is not the right fix, it introduces a memory leak when a netns is
destroyed (the FB device is never deleted).
Signed-off-by: Nicolas Dichtel <redacted>
Reported-by: Steven Rostedt <redacted>
Tested-by: Steven Rostedt <redacted> (and our entire MRG team)
Tested-by: "Luis Claudio R. Goncalves" <redacted>
Tested-by: John Kacur <jkacur@redhat.com>
---
v2: add Steven's tags
net/ipv6/ip6_tunnel.c | 2 ++
1 file changed, 2 insertions(+)
From: Nicolas Dichtel <hidden> Date: 2014-01-31 08:24:20
This problem was fixed upstream by commit 9434266f2c64 ("sit: fix use after free
of fb_tunnel_dev").
The upstream patch depends on upstream commit 5e6700b3bf98 ("sit: add support of
x-netns"), which was not backported into 3.10 branch.
First, explain the problem: when the sit module is unloaded, sit_cleanup() is
called.
rmmod sit
=> sit_cleanup()
=> rtnl_link_unregister()
=> __rtnl_kill_links()
=> for_each_netdev(net, dev) {
if (dev->rtnl_link_ops == ops)
ops->dellink(dev, &list_kill);
}
At this point, the FB device is deleted (and all sit tunnels).
=> unregister_pernet_device()
=> unregister_pernet_operations()
=> ops_exit_list()
=> sit_exit_net()
=> sit_destroy_tunnels()
In this function, no tunnel is found.
=> unregister_netdevice_queue(sitn->fb_tunnel_dev, &list);
We delete the FB device a second time here!
Because we cannot simply remove the second deletion (sit_exit_net() must remove
the FB device when a netns is deleted), we add an rtnl ops which delete all sit
device excepting the FB device and thus we can keep the explicit deletion in
sit_exit_net().
CC: Steven Rostedt <rostedt@goodmis.org>
Signed-off-by: Nicolas Dichtel <redacted>
Acked-by: Willem de Bruijn <willemb@google.com>
Reported-by: Steven Rostedt <redacted>
Tested-by: Steven Rostedt <redacted> (and our entire MRG team)
Tested-by: "Luis Claudio R. Goncalves" <redacted>
Tested-by: John Kacur <jkacur@redhat.com>
---
v2: add Steven's tags
net/ipv6/sit.c | 10 ++++++++++
1 file changed, 10 insertions(+)
From: Steven Rostedt <rostedt@goodmis.org> Date: 2014-01-31 17:19:32
Just FYI.
Our full tier tests have completed with no issues due to the sit or
ip6_tunnel modules. These patches appear to have solved our problems.
Nicolas, Thanks again for posting these!
-- Steve
On Fri, 31 Jan 2014 09:24:04 +0100
Nicolas Dichtel [off-list ref] wrote:
This problem was fixed upstream by commit 9434266f2c64 ("sit: fix use after free
of fb_tunnel_dev").
The upstream patch depends on upstream commit 5e6700b3bf98 ("sit: add support of
x-netns"), which was not backported into 3.10 branch.
First, explain the problem: when the sit module is unloaded, sit_cleanup() is
called.
rmmod sit
=> sit_cleanup()
=> rtnl_link_unregister()
=> __rtnl_kill_links()
=> for_each_netdev(net, dev) {
if (dev->rtnl_link_ops == ops)
ops->dellink(dev, &list_kill);
}
At this point, the FB device is deleted (and all sit tunnels).
=> unregister_pernet_device()
=> unregister_pernet_operations()
=> ops_exit_list()
=> sit_exit_net()
=> sit_destroy_tunnels()
In this function, no tunnel is found.
=> unregister_netdevice_queue(sitn->fb_tunnel_dev, &list);
We delete the FB device a second time here!
Because we cannot simply remove the second deletion (sit_exit_net() must remove
the FB device when a netns is deleted), we add an rtnl ops which delete all sit
device excepting the FB device and thus we can keep the explicit deletion in
sit_exit_net().