From: Nikolay Aleksandrov <hidden> Date: 2017-12-16 09:17:26
On 16/12/17 02:37, Andrei Vagin wrote:
Hi,
We run criu tests for linux-next and today we get this bug:
The kernel version is 4.15.0-rc3-next-20171215
[ 235.397328] BUG: unable to handle kernel NULL pointer dereference
at 000000000000000c
[ 235.398624] IP: fdb_find_rcu+0x3c/0x130
[snip]
Hi,
Thanks for the report, I've missed the changelink before dev creation case when I did
the rhashtable conversion, some of the options do fdb lookups as part of their routine
but we don't have the table initialized yet at that point.
I'll send a fix after some testing.
Thanks,
Nik
From: Nikolay Aleksandrov <hidden> Date: 2017-12-16 09:29:16
On 16/12/17 11:17, Nikolay Aleksandrov wrote:
On 16/12/17 02:37, Andrei Vagin wrote:
quoted
Hi,
We run criu tests for linux-next and today we get this bug:
The kernel version is 4.15.0-rc3-next-20171215
[ 235.397328] BUG: unable to handle kernel NULL pointer dereference
at 000000000000000c
[ 235.398624] IP: fdb_find_rcu+0x3c/0x130
[snip]
Hi,
Thanks for the report, I've missed the changelink before dev creation case when I did
err, s/changelink/br_stp_change_bridge_id/
the other options are set after register_netdevice, this is the only one changed before
the rhashtable conversion, some of the options do fdb lookups as part of their routine
but we don't have the table initialized yet at that point.
I'll send a fix after some testing.
Thanks,
Nik
From: Nikolay Aleksandrov <hidden> Date: 2017-12-16 10:41:01
On 16/12/17 11:29, Nikolay Aleksandrov wrote:
On 16/12/17 11:17, Nikolay Aleksandrov wrote:
quoted
On 16/12/17 02:37, Andrei Vagin wrote:
quoted
Hi,
We run criu tests for linux-next and today we get this bug:
The kernel version is 4.15.0-rc3-next-20171215
[ 235.397328] BUG: unable to handle kernel NULL pointer dereference
at 000000000000000c
[ 235.398624] IP: fdb_find_rcu+0x3c/0x130
[snip]
Hi,
Thanks for the report, I've missed the changelink before dev creation case when I did
err, s/changelink/br_stp_change_bridge_id/
the other options are set after register_netdevice, this is the only one changed before
quoted
the rhashtable conversion, some of the options do fdb lookups as part of their routine
but we don't have the table initialized yet at that point.
I'll send a fix after some testing.
Thanks,
Nik
We need to fix this in -net, it has a memory leak that has existed since the
introduction of br_stp_change_bridge_id() before register_netdevice because
it adds an fdb entry which never gets deleted if an error happens, also the
notifications for that fdb entry come with ifindex = 0 because the bridge netdev
doesn't exist yet. All of that looks wrong, I'll send a fix for -net to move
the bridge id change after the netdev register and cleanup any bridge fdbs
on error.
The commit with that change is:
30313a3d5794 ("bridge: Handle IFLA_ADDRESS correctly when creating bridge device")
Before the changelink in while doing newlink in bridge was possible, this would happen
only on netdev register fail, but now it is much easier to trigger (as below) since
changelink can fail if called with wrong arguments.
Here's the trace of rmmod bridge after a failed bridge newlink with mac address set
(this kernel is before my rhashtable change):
$ ip l add br0 address 00:11:22:33:44:55 type bridge group_fwd_mask 1
RTNETLINK answers: Invalid argument
$ rmmod bridge
[ 1822.142525] =============================================================================
[ 1822.143640] BUG bridge_fdb_cache (Tainted: G O ): Objects remaining in bridge_fdb_cache on __kmem_cache_shutdown()
[ 1822.144821] -----------------------------------------------------------------------------
[ 1822.145990] Disabling lock debugging due to kernel taint
[ 1822.146732] INFO: Slab 0x0000000092a844b2 objects=32 used=2 fp=0x00000000fef011b0 flags=0x1ffff8000000100
[ 1822.147700] CPU: 2 PID: 13584 Comm: rmmod Tainted: G B O 4.15.0-rc2+ #87
[ 1822.148578] Hardware name: QEMU Standard PC (i440FX + PIIX, 1996), BIOS 1.7.5-20140531_083030-gandalf 04/01/2014
[ 1822.150008] Call Trace:
[ 1822.150510] dump_stack+0x78/0xa9
[ 1822.151156] slab_err+0xb1/0xd3
[ 1822.151834] ? __kmalloc+0x1bb/0x1ce
[ 1822.152546] __kmem_cache_shutdown+0x151/0x28b
[ 1822.153395] shutdown_cache+0x13/0x144
[ 1822.154126] kmem_cache_destroy+0x1c0/0x1fb
[ 1822.154669] SyS_delete_module+0x194/0x244
[ 1822.155199] ? trace_hardirqs_on_thunk+0x1a/0x1c
[ 1822.155773] entry_SYSCALL_64_fastpath+0x23/0x9a
[ 1822.156343] RIP: 0033:0x7f929bd38b17
[ 1822.156859] RSP: 002b:00007ffd160e9a98 EFLAGS: 00000202 ORIG_RAX: 00000000000000b0
[ 1822.157728] RAX: ffffffffffffffda RBX: 00005578316ba090 RCX: 00007f929bd38b17
[ 1822.158422] RDX: 00007f929bd9ec60 RSI: 0000000000000800 RDI: 00005578316ba0f0
[ 1822.159114] RBP: 0000000000000003 R08: 00007f929bff5f20 R09: 00007ffd160e8a11
[ 1822.159808] R10: 00007ffd160e9860 R11: 0000000000000202 R12: 00007ffd160e8a80
[ 1822.160513] R13: 0000000000000000 R14: 0000000000000000 R15: 00005578316ba090
[ 1822.161278] INFO: Object 0x000000007645de29 @offset=0
[ 1822.161666] INFO: Object 0x00000000d5df2ab5 @offset=128
From: Nikolay Aleksandrov <hidden> Date: 2017-12-16 11:22:24
On 16/12/17 12:40, Nikolay Aleksandrov wrote:
On 16/12/17 11:29, Nikolay Aleksandrov wrote:
quoted
On 16/12/17 11:17, Nikolay Aleksandrov wrote:
quoted
On 16/12/17 02:37, Andrei Vagin wrote:
quoted
Hi,
We run criu tests for linux-next and today we get this bug:
The kernel version is 4.15.0-rc3-next-20171215
[ 235.397328] BUG: unable to handle kernel NULL pointer dereference
at 000000000000000c
[ 235.398624] IP: fdb_find_rcu+0x3c/0x130
[snip]
Hi,
Thanks for the report, I've missed the changelink before dev creation case when I did
err, s/changelink/br_stp_change_bridge_id/
the other options are set after register_netdevice, this is the only one changed before
quoted
the rhashtable conversion, some of the options do fdb lookups as part of their routine
but we don't have the table initialized yet at that point.
I'll send a fix after some testing.
Thanks,
Nik
We need to fix this in -net, it has a memory leak that has existed since the
introduction of br_stp_change_bridge_id() before register_netdevice because
it adds an fdb entry which never gets deleted if an error happens, also the
notifications for that fdb entry come with ifindex = 0 because the bridge netdev
doesn't exist yet. All of that looks wrong, I'll send a fix for -net to move
the bridge id change after the netdev register and cleanup any bridge fdbs
on error.
The commit with that change is:
30313a3d5794 ("bridge: Handle IFLA_ADDRESS correctly when creating bridge device")
Before the changelink in while doing newlink in bridge was possible, this would happen
only on netdev register fail, but now it is much easier to trigger (as below) since
changelink can fail if called with wrong arguments.
One more thing before sending the patch, the actual commit that introduced the fdb
insert in br_stp_change_bridge_id() is:
commit a4b816d8ba1c
Author: Toshiaki Makita [off-list ref]
Date: Fri Feb 7 16:48:21 2014 +0900
bridge: Change local fdb entries whenever mac address of bridge device changes
So that is what we need to fix.
From: Nikolay Aleksandrov <hidden> Date: 2017-12-16 11:31:57
The early call to br_stp_change_bridge_id in bridge's newlink can cause
a memory leak if an error occurs during the newlink because the fdb
entries are not cleaned up if a different lladdr was specified, also
another minor issue is that it generates fdb notifications with
ifindex = 0. To remove this special case the call is done after netdev
register and we cleanup any bridge fdb entries on changelink error.
That also doesn't slow down normal bridge removal, alternative is to call
it in its ndo_uninit.
To reproduce the issue:
$ ip l add br0 address 00:11:22:33:44:55 type bridge group_fwd_mask 1
RTNETLINK answers: Invalid argument
$ rmmod bridge
[ 1822.142525] =============================================================================
[ 1822.143640] BUG bridge_fdb_cache (Tainted: G O ): Objects remaining in bridge_fdb_cache on __kmem_cache_shutdown()
[ 1822.144821] -----------------------------------------------------------------------------
[ 1822.145990] Disabling lock debugging due to kernel taint
[ 1822.146732] INFO: Slab 0x0000000092a844b2 objects=32 used=2 fp=0x00000000fef011b0 flags=0x1ffff8000000100
[ 1822.147700] CPU: 2 PID: 13584 Comm: rmmod Tainted: G B O 4.15.0-rc2+ #87
[ 1822.148578] Hardware name: QEMU Standard PC (i440FX + PIIX, 1996), BIOS 1.7.5-20140531_083030-gandalf 04/01/2014
[ 1822.150008] Call Trace:
[ 1822.150510] dump_stack+0x78/0xa9
[ 1822.151156] slab_err+0xb1/0xd3
[ 1822.151834] ? __kmalloc+0x1bb/0x1ce
[ 1822.152546] __kmem_cache_shutdown+0x151/0x28b
[ 1822.153395] shutdown_cache+0x13/0x144
[ 1822.154126] kmem_cache_destroy+0x1c0/0x1fb
[ 1822.154669] SyS_delete_module+0x194/0x244
[ 1822.155199] ? trace_hardirqs_on_thunk+0x1a/0x1c
[ 1822.155773] entry_SYSCALL_64_fastpath+0x23/0x9a
[ 1822.156343] RIP: 0033:0x7f929bd38b17
[ 1822.156859] RSP: 002b:00007ffd160e9a98 EFLAGS: 00000202 ORIG_RAX: 00000000000000b0
[ 1822.157728] RAX: ffffffffffffffda RBX: 00005578316ba090 RCX: 00007f929bd38b17
[ 1822.158422] RDX: 00007f929bd9ec60 RSI: 0000000000000800 RDI: 00005578316ba0f0
[ 1822.159114] RBP: 0000000000000003 R08: 00007f929bff5f20 R09: 00007ffd160e8a11
[ 1822.159808] R10: 00007ffd160e9860 R11: 0000000000000202 R12: 00007ffd160e8a80
[ 1822.160513] R13: 0000000000000000 R14: 0000000000000000 R15: 00005578316ba090
[ 1822.161278] INFO: Object 0x000000007645de29 @offset=0
[ 1822.161666] INFO: Object 0x00000000d5df2ab5 @offset=128
Fixes: a4b816d8ba1c ("bridge: Change local fdb entries whenever mac address of bridge device changes")
Signed-off-by: Nikolay Aleksandrov <redacted>
---
Consequently this also would fix the null ptr deref due to the rhashtable
not being initialized in net-next when br_stp_change_bridge_id is called.
Toshiaki, any reason you called br_stp_change_bridge_id before
register_netdevice when you introduced it in 30313a3d5794 ?
net/bridge/br_netlink.c | 14 +++++++++-----
1 file changed, 9 insertions(+), 5 deletions(-)
@@ -1262,19 +1262,23 @@ static int br_dev_newlink(struct net *src_net, struct net_device *dev,structnet_bridge*br=netdev_priv(dev);interr;+err=register_netdevice(dev);+if(err)+returnerr;+if(tb[IFLA_ADDRESS]){spin_lock_bh(&br->lock);br_stp_change_bridge_id(br,nla_data(tb[IFLA_ADDRESS]));spin_unlock_bh(&br->lock);}-err=register_netdevice(dev);-if(err)-returnerr;-err=br_changelink(dev,tb,data,extack);-if(err)+if(err){+/* clean possible fdbs from br_stp_change_bridge_id above */+br_fdb_delete_by_port(br,NULL,0,1);unregister_netdevice(dev);+}+returnerr;}
From: Nikolay Aleksandrov <hidden> Date: 2017-12-16 12:38:19
On 16/12/17 13:31, Nikolay Aleksandrov wrote:
The early call to br_stp_change_bridge_id in bridge's newlink can cause
a memory leak if an error occurs during the newlink because the fdb
entries are not cleaned up if a different lladdr was specified, also
another minor issue is that it generates fdb notifications with
ifindex = 0. To remove this special case the call is done after netdev
register and we cleanup any bridge fdb entries on changelink error.
That also doesn't slow down normal bridge removal, alternative is to call
it in its ndo_uninit.
To reproduce the issue:
$ ip l add br0 address 00:11:22:33:44:55 type bridge group_fwd_mask 1
RTNETLINK answers: Invalid argument
$ rmmod bridge
[ 1822.142525] =============================================================================
[ 1822.143640] BUG bridge_fdb_cache (Tainted: G O ): Objects remaining in bridge_fdb_cache on __kmem_cache_shutdown()
[ 1822.144821] -----------------------------------------------------------------------------
[ 1822.145990] Disabling lock debugging due to kernel taint
[ 1822.146732] INFO: Slab 0x0000000092a844b2 objects=32 used=2 fp=0x00000000fef011b0 flags=0x1ffff8000000100
[ 1822.147700] CPU: 2 PID: 13584 Comm: rmmod Tainted: G B O 4.15.0-rc2+ #87
[ 1822.148578] Hardware name: QEMU Standard PC (i440FX + PIIX, 1996), BIOS 1.7.5-20140531_083030-gandalf 04/01/2014
[ 1822.150008] Call Trace:
[ 1822.150510] dump_stack+0x78/0xa9
[ 1822.151156] slab_err+0xb1/0xd3
[ 1822.151834] ? __kmalloc+0x1bb/0x1ce
[ 1822.152546] __kmem_cache_shutdown+0x151/0x28b
[ 1822.153395] shutdown_cache+0x13/0x144
[ 1822.154126] kmem_cache_destroy+0x1c0/0x1fb
[ 1822.154669] SyS_delete_module+0x194/0x244
[ 1822.155199] ? trace_hardirqs_on_thunk+0x1a/0x1c
[ 1822.155773] entry_SYSCALL_64_fastpath+0x23/0x9a
[ 1822.156343] RIP: 0033:0x7f929bd38b17
[ 1822.156859] RSP: 002b:00007ffd160e9a98 EFLAGS: 00000202 ORIG_RAX: 00000000000000b0
[ 1822.157728] RAX: ffffffffffffffda RBX: 00005578316ba090 RCX: 00007f929bd38b17
[ 1822.158422] RDX: 00007f929bd9ec60 RSI: 0000000000000800 RDI: 00005578316ba0f0
[ 1822.159114] RBP: 0000000000000003 R08: 00007f929bff5f20 R09: 00007ffd160e8a11
[ 1822.159808] R10: 00007ffd160e9860 R11: 0000000000000202 R12: 00007ffd160e8a80
[ 1822.160513] R13: 0000000000000000 R14: 0000000000000000 R15: 00005578316ba090
[ 1822.161278] INFO: Object 0x000000007645de29 @offset=0
[ 1822.161666] INFO: Object 0x00000000d5df2ab5 @offset=128
Fixes: a4b816d8ba1c ("bridge: Change local fdb entries whenever mac address of bridge device changes")
Signed-off-by: Nikolay Aleksandrov <redacted>
---
Consequently this also would fix the null ptr deref due to the rhashtable
not being initialized in net-next when br_stp_change_bridge_id is called.
Toshiaki, any reason you called br_stp_change_bridge_id before
register_netdevice when you introduced it in 30313a3d5794 ?
net/bridge/br_netlink.c | 14 +++++++++-----
1 file changed, 9 insertions(+), 5 deletions(-)
Oops, wrong Fixes commit id, this one was introduced in Feb/2014, should be:
Fixes: 30313a3d5794 ("bridge: Handle IFLA_ADDRESS correctly when creating bridge device")
as I said before since that was introduced in Apr/2014.
Dave, would you like me to send v2 with the proper Fixes tag ?
From: Stephen Hemminger <stephen@networkplumber.org> Date: 2017-12-16 18:32:46
On Sat, 16 Dec 2017 13:31:36 +0200
Nikolay Aleksandrov [off-list ref] wrote:
The early call to br_stp_change_bridge_id in bridge's newlink can cause
a memory leak if an error occurs during the newlink because the fdb
entries are not cleaned up if a different lladdr was specified, also
another minor issue is that it generates fdb notifications with
ifindex = 0. To remove this special case the call is done after netdev
register and we cleanup any bridge fdb entries on changelink error.
That also doesn't slow down normal bridge removal, alternative is to call
it in its ndo_uninit.
To reproduce the issue:
$ ip l add br0 address 00:11:22:33:44:55 type bridge group_fwd_mask 1
RTNETLINK answers: Invalid argument
$ rmmod bridge
[ 1822.142525] =============================================================================
[ 1822.143640] BUG bridge_fdb_cache (Tainted: G O ): Objects remaining in bridge_fdb_cache on __kmem_cache_shutdown()
[ 1822.144821] -----------------------------------------------------------------------------
[ 1822.145990] Disabling lock debugging due to kernel taint
[ 1822.146732] INFO: Slab 0x0000000092a844b2 objects=32 used=2 fp=0x00000000fef011b0 flags=0x1ffff8000000100
[ 1822.147700] CPU: 2 PID: 13584 Comm: rmmod Tainted: G B O 4.15.0-rc2+ #87
[ 1822.148578] Hardware name: QEMU Standard PC (i440FX + PIIX, 1996), BIOS 1.7.5-20140531_083030-gandalf 04/01/2014
[ 1822.150008] Call Trace:
[ 1822.150510] dump_stack+0x78/0xa9
[ 1822.151156] slab_err+0xb1/0xd3
[ 1822.151834] ? __kmalloc+0x1bb/0x1ce
[ 1822.152546] __kmem_cache_shutdown+0x151/0x28b
[ 1822.153395] shutdown_cache+0x13/0x144
[ 1822.154126] kmem_cache_destroy+0x1c0/0x1fb
[ 1822.154669] SyS_delete_module+0x194/0x244
[ 1822.155199] ? trace_hardirqs_on_thunk+0x1a/0x1c
[ 1822.155773] entry_SYSCALL_64_fastpath+0x23/0x9a
[ 1822.156343] RIP: 0033:0x7f929bd38b17
[ 1822.156859] RSP: 002b:00007ffd160e9a98 EFLAGS: 00000202 ORIG_RAX: 00000000000000b0
[ 1822.157728] RAX: ffffffffffffffda RBX: 00005578316ba090 RCX: 00007f929bd38b17
[ 1822.158422] RDX: 00007f929bd9ec60 RSI: 0000000000000800 RDI: 00005578316ba0f0
[ 1822.159114] RBP: 0000000000000003 R08: 00007f929bff5f20 R09: 00007ffd160e8a11
[ 1822.159808] R10: 00007ffd160e9860 R11: 0000000000000202 R12: 00007ffd160e8a80
[ 1822.160513] R13: 0000000000000000 R14: 0000000000000000 R15: 00005578316ba090
[ 1822.161278] INFO: Object 0x000000007645de29 @offset=0
[ 1822.161666] INFO: Object 0x00000000d5df2ab5 @offset=128
Fixes: a4b816d8ba1c ("bridge: Change local fdb entries whenever mac address of bridge device changes")
Signed-off-by: Nikolay Aleksandrov <redacted>
---
Consequently this also would fix the null ptr deref due to the rhashtable
not being initialized in net-next when br_stp_change_bridge_id is called.
Toshiaki, any reason you called br_stp_change_bridge_id before
register_netdevice when you introduced it in 30313a3d5794 ?
net/bridge/br_netlink.c | 14 +++++++++-----
1 file changed, 9 insertions(+), 5 deletions(-)
Thanks for working on this.
I agree that fixing this in ndo_uninit would be wrong. There are less bugs
if init and uninit do logically equivalent steps.
A bridge device can be created either with netlink or ioctl.
This change is also makes both ways of adding MAC have
the same semantics;
If bridge is created with ioctl then the bridge_id (and MAC)
will not be changed until later device is added or MAC address
is set by other operation.
Signed-off-by: Stephen Hemminger <stephen@networkplumber.org>
The early call to br_stp_change_bridge_id in bridge's newlink can cause
a memory leak if an error occurs during the newlink because the fdb
entries are not cleaned up if a different lladdr was specified, also
another minor issue is that it generates fdb notifications with
ifindex = 0. To remove this special case the call is done after netdev
register and we cleanup any bridge fdb entries on changelink error.
That also doesn't slow down normal bridge removal, alternative is to call
it in its ndo_uninit.
...
Fixes: a4b816d8ba1c ("bridge: Change local fdb entries whenever mac address of bridge device changes")
Signed-off-by: Nikolay Aleksandrov <redacted>
---
Consequently this also would fix the null ptr deref due to the rhashtable
not being initialized in net-next when br_stp_change_bridge_id is called.
Toshiaki, any reason you called br_stp_change_bridge_id before
register_netdevice when you introduced it in 30313a3d5794 ?
Thank you for taking care of this.
It's my bad. I just missed the problem.
@@ -1262,19 +1262,23 @@ static int br_dev_newlink(struct net *src_net, struct net_device *dev,
...
quoted hunk
err = br_changelink(dev, tb, data, extack);- if (err)+ if (err) {+ /* clean possible fdbs from br_stp_change_bridge_id above */+ br_fdb_delete_by_port(br, NULL, 0, 1);
Don't we need to call br_dev_delete (br_link_ops.dellink) after
successful register instead of br_fdb_delete?
Particularly I'm wondering if not calling br_sysfs_delbr() is ok or not.
--
Toshiaki Makita
From: Nikolay Aleksandrov <hidden> Date: 2017-12-18 14:22:23
On 12/18/2017 04:24 AM, Toshiaki Makita wrote:
On 2017/12/16 20:31, Nikolay Aleksandrov wrote:
[snip]
...
quoted
err = br_changelink(dev, tb, data, extack);- if (err)+ if (err) {+ /* clean possible fdbs from br_stp_change_bridge_id above */+ br_fdb_delete_by_port(br, NULL, 0, 1);
Don't we need to call br_dev_delete (br_link_ops.dellink) after
successful register instead of br_fdb_delete?
Particularly I'm wondering if not calling br_sysfs_delbr() is ok or not.
Funny, that is actually the only reason we need to call it (br_sysfs_delbr). :-)
Good catch, that is another leak - the bridge sysfs entries are registered when
NETDEV_REGISTER event happens (register_netdevice) but are not properly cleaned up
on error there. This has also been present since the introduction of changelink
during newlink, commit:
b6677449dff6 ("bridge: netlink: call br_changelink() during br_dev_newlink()")
I'll send v2 that does br_dev_delete(dev, NULL) instead of the current cleanup.
With kobject debug enabled and that I can see "brif" and the rest of the sysfs
files getting freed properly, while before they weren't.
Thanks,
Nik
From: Nikolay Aleksandrov <hidden> Date: 2017-12-18 14:23:55
On 12/18/2017 04:22 PM, Nikolay Aleksandrov wrote:
On 12/18/2017 04:24 AM, Toshiaki Makita wrote:
quoted
On 2017/12/16 20:31, Nikolay Aleksandrov wrote:
[snip]
quoted
...
quoted
err = br_changelink(dev, tb, data, extack);- if (err)+ if (err) {+ /* clean possible fdbs from br_stp_change_bridge_id above */+ br_fdb_delete_by_port(br, NULL, 0, 1);
Don't we need to call br_dev_delete (br_link_ops.dellink) after
successful register instead of br_fdb_delete?
Particularly I'm wondering if not calling br_sysfs_delbr() is ok or not.
Funny, that is actually the only reason we need to call it (br_sysfs_delbr). :-)
Good catch, that is another leak - the bridge sysfs entries are registered when
NETDEV_REGISTER event happens (register_netdevice) but are not properly cleaned up
on error there. This has also been present since the introduction of changelink
during newlink, commit:
b6677449dff6 ("bridge: netlink: call br_changelink() during br_dev_newlink()")
err, since the changelink was fixed to be after device registration in commit:
5b8d5429daa0 ("bridge: netlink: register netdevice before executing changelink")
I'll send v2 that does br_dev_delete(dev, NULL) instead of the current cleanup.
With kobject debug enabled and that I can see "brif" and the rest of the sysfs
files getting freed properly, while before they weren't.
Thanks,
Nik
From: Nikolay Aleksandrov <hidden> Date: 2017-12-18 15:35:19
The early call to br_stp_change_bridge_id in bridge's newlink can cause
a memory leak if an error occurs during the newlink because the fdb
entries are not cleaned up if a different lladdr was specified, also
another minor issue is that it generates fdb notifications with
ifindex = 0. Another unrelated memory leak is the bridge sysfs entries
which get added on NETDEV_REGISTER event, but are not cleaned up in the
newlink error path. To remove this special case the call to
br_stp_change_bridge_id is done after netdev register and we cleanup the
bridge on changelink error via br_dev_delete to plug all leaks.
This patch makes netlink bridge destruction on newlink error the same as
dellink and ioctl del which is necessary since at that point we have a
fully initialized bridge device.
To reproduce the issue:
$ ip l add br0 address 00:11:22:33:44:55 type bridge group_fwd_mask 1
RTNETLINK answers: Invalid argument
$ rmmod bridge
[ 1822.142525] =============================================================================
[ 1822.143640] BUG bridge_fdb_cache (Tainted: G O ): Objects remaining in bridge_fdb_cache on __kmem_cache_shutdown()
[ 1822.144821] -----------------------------------------------------------------------------
[ 1822.145990] Disabling lock debugging due to kernel taint
[ 1822.146732] INFO: Slab 0x0000000092a844b2 objects=32 used=2 fp=0x00000000fef011b0 flags=0x1ffff8000000100
[ 1822.147700] CPU: 2 PID: 13584 Comm: rmmod Tainted: G B O 4.15.0-rc2+ #87
[ 1822.148578] Hardware name: QEMU Standard PC (i440FX + PIIX, 1996), BIOS 1.7.5-20140531_083030-gandalf 04/01/2014
[ 1822.150008] Call Trace:
[ 1822.150510] dump_stack+0x78/0xa9
[ 1822.151156] slab_err+0xb1/0xd3
[ 1822.151834] ? __kmalloc+0x1bb/0x1ce
[ 1822.152546] __kmem_cache_shutdown+0x151/0x28b
[ 1822.153395] shutdown_cache+0x13/0x144
[ 1822.154126] kmem_cache_destroy+0x1c0/0x1fb
[ 1822.154669] SyS_delete_module+0x194/0x244
[ 1822.155199] ? trace_hardirqs_on_thunk+0x1a/0x1c
[ 1822.155773] entry_SYSCALL_64_fastpath+0x23/0x9a
[ 1822.156343] RIP: 0033:0x7f929bd38b17
[ 1822.156859] RSP: 002b:00007ffd160e9a98 EFLAGS: 00000202 ORIG_RAX: 00000000000000b0
[ 1822.157728] RAX: ffffffffffffffda RBX: 00005578316ba090 RCX: 00007f929bd38b17
[ 1822.158422] RDX: 00007f929bd9ec60 RSI: 0000000000000800 RDI: 00005578316ba0f0
[ 1822.159114] RBP: 0000000000000003 R08: 00007f929bff5f20 R09: 00007ffd160e8a11
[ 1822.159808] R10: 00007ffd160e9860 R11: 0000000000000202 R12: 00007ffd160e8a80
[ 1822.160513] R13: 0000000000000000 R14: 0000000000000000 R15: 00005578316ba090
[ 1822.161278] INFO: Object 0x000000007645de29 @offset=0
[ 1822.161666] INFO: Object 0x00000000d5df2ab5 @offset=128
Fixes: 30313a3d5794 ("bridge: Handle IFLA_ADDRESS correctly when creating bridge device")
Fixes: 5b8d5429daa0 ("bridge: netlink: register netdevice before executing changelink")
Signed-off-by: Nikolay Aleksandrov <redacted>
---
v2: it's better to use br_dev_delete to clean up because the sysfs entries
were leaked as well, basically we use it for the bridge fdb flush and sysfs
entries delete
Consequently this also would fix the null ptr deref due to the rhashtable
not being initialized in net-next when br_stp_change_bridge_id is called.
net/bridge/br_netlink.c | 11 ++++++-----
1 file changed, 6 insertions(+), 5 deletions(-)
From: David Miller <davem@davemloft.net> Date: 2017-12-18 18:31:09
From: Nikolay Aleksandrov <redacted>
Date: Mon, 18 Dec 2017 17:35:09 +0200
The early call to br_stp_change_bridge_id in bridge's newlink can cause
a memory leak if an error occurs during the newlink because the fdb
entries are not cleaned up if a different lladdr was specified, also
another minor issue is that it generates fdb notifications with
ifindex = 0. Another unrelated memory leak is the bridge sysfs entries
which get added on NETDEV_REGISTER event, but are not cleaned up in the
newlink error path. To remove this special case the call to
br_stp_change_bridge_id is done after netdev register and we cleanup the
bridge on changelink error via br_dev_delete to plug all leaks.
This patch makes netlink bridge destruction on newlink error the same as
dellink and ioctl del which is necessary since at that point we have a
fully initialized bridge device.
To reproduce the issue:
$ ip l add br0 address 00:11:22:33:44:55 type bridge group_fwd_mask 1
RTNETLINK answers: Invalid argument
$ rmmod bridge
[ 1822.142525] =============================================================================
[ 1822.143640] BUG bridge_fdb_cache (Tainted: G O ): Objects remaining in bridge_fdb_cache on __kmem_cache_shutdown()
[ 1822.144821] -----------------------------------------------------------------------------
...
Fixes: 30313a3d5794 ("bridge: Handle IFLA_ADDRESS correctly when creating bridge device")
Fixes: 5b8d5429daa0 ("bridge: netlink: register netdevice before executing changelink")
Signed-off-by: Nikolay Aleksandrov <redacted>
---
v2: it's better to use br_dev_delete to clean up because the sysfs entries
were leaked as well, basically we use it for the bridge fdb flush and sysfs
entries delete
Consequently this also would fix the null ptr deref due to the rhashtable
not being initialized in net-next when br_stp_change_bridge_id is called.
Applied, but doing state changes after register_netdev() is always dicey.
I hope things go well if the device is brought up immediately during
that register_netdev() call and before the stp bridge ID setting is
done.