ip6_vti: fix memleak on netns dismantle
Subsystems (4)
PATCH 1/1 ip6_vti: fix memleak on netns dismantle
kmemleak reports net_device resources are no longer released, restore needs_free_netdev toggle. Sample backtrace:
unreferenced object 0xffff88810874f000 (size 4096): [..]
[<00000000a2b8af8b>] __kmalloc_node+0x209/0x290
[<0000000040b0a1a9>] alloc_netdev_mqs+0x58/0x470
[<00000000b4be1e78>] vti6_init_net+0x94/0x230
[<000000008830c1ea>] ops_init+0x32/0xc0
[<000000006a26fa8f>] setup_net+0x134/0x2e0[..]
--- net/ipv6/ip6_vti.c | 1 + 1 file changed, 1 insertion(+)
net/ipv6/ip6_vti.c +1 -0
@@ -901,6 +901,7 @@ static void vti6_dev_setup(struct net_device *dev) { dev->netdev_ops = &vti6_netdev_ops; dev->header_ops = &ip_tunnel_header_ops; + dev->needs_free_netdev = true; dev->pcpu_stat_type =NETDEV_PCPU_STAT_TSTATS ; dev->type =ARPHRD_TUNNEL6 ;
Signature · 1 line
2.43.2
+ Steffen Klassert, Herbert Xu, David Ahern
Quoted 9 lines · Florian Westphal
kmemleak reports
net_deviceresources are no longer released, restoreneeds_free_netdevtoggle. Sample backtrace:unreferenced object
0xffff88810874f000(size 4096): [..][<00000000a2b8af8b>] __kmalloc_node+0x209/0x290 [<0000000040b0a1a9>] alloc_netdev_mqs+0x58/0x470 [<00000000b4be1e78>] vti6_init_net+0x94/0x230 [<000000008830c1ea>] ops_init+0x32/0xc0 [<000000006a26fa8f>] setup_net+0x134/0x2e0
Quoted 5 lines
[..]
Cc Breno Leitao <[email protected]>Fixes a9b2d55a8f1e ("ip6_vti: Do not use custom stat allocator")Signed-off-by Florian Westphal <[email protected]>
Quoted 20 lines
--- net/ipv6/ip6_vti.c | 1 + 1 file changed, 1 insertion(+)net/ipv6/ip6_vti.c +1
-0@@ -901,6 +901,7 @@ static void vti6_dev_setup(struct net_device *dev) { dev->netdev_ops = &vti6_netdev_ops; dev->header_ops = &ip_tunnel_header_ops; + dev->needs_free_netdev = true; dev->pcpu_stat_type =NETDEV_PCPU_STAT_TSTATS ; dev->type =ARPHRD_TUNNEL6 ;Signature · 1 line
2.43.2
This patch was applied to netdev/net-next.git (main) by Jakub Kicinski <[email protected]>:
Hello:
This patch was applied to netdev/net-next.git (main) by Jakub Kicinski <[email protected]>:
Quoted 4 lines · On Mon, 15 Apr 2024 14:23:44 +0200 you wrote:
kmemleak reports
net_deviceresources are no longer released, restoreneeds_free_netdevtoggle. Sample backtrace:unreferenced object
0xffff88810874f000(size 4096): [..]
Quoted 8 lines
[<00000000a2b8af8b>] __kmalloc_node+0x209/0x290 [<0000000040b0a1a9>] alloc_netdev_mqs+0x58/0x470 [<00000000b4be1e78>] vti6_init_net+0x94/0x230 [<000000008830c1ea>] ops_init+0x32/0xc0 [<000000006a26fa8f>] setup_net+0x134/0x2e0[..]
[...]
Here is the summary with links:
- [net-next] ip6_vti: fix memleak on netns dismantle
https://git.kernel.org/netdev/net-next/c/86600ea11dc1
You are awesome, thank you!
Signature · 2 lines
Deet-doot-dot, I am a bot. https://korg.docs.kernel.org/patchwork/pwbot.html
Hello Florian,
Quoted 23 lines · Florian Westphal
kmemleak reports
net_deviceresources are no longer released, restoreneeds_free_netdevtoggle. Sample backtrace:unreferenced object
0xffff88810874f000(size 4096): [..][<00000000a2b8af8b>] __kmalloc_node+0x209/0x290 [<0000000040b0a1a9>] alloc_netdev_mqs+0x58/0x470 [<00000000b4be1e78>] vti6_init_net+0x94/0x230 [<000000008830c1ea>] ops_init+0x32/0xc0 [<000000006a26fa8f>] setup_net+0x134/0x2e0[..]
Cc Breno Leitao <[email protected]>Fixes a9b2d55a8f1e ("ip6_vti: Do not use custom stat allocator")Signed-off-by Florian Westphal <[email protected]>--- net/ipv6/ip6_vti.c | 1 + 1 file changed, 1 insertion(+)net/ipv6/ip6_vti.c +0
-0
net/ipv6/ip6_vti.c static void vti6_dev_setup(struct net_device *dev) · line 904
Quoted 4 lines
net/ipv6/ip6_vti.c +1
-0@@ -901,6 +901,7 @@ static void vti6_dev_setup(struct net_device *dev) { dev->netdev_ops = &vti6_netdev_ops; dev->header_ops = &ip_tunnel_header_ops; + dev->needs_free_netdev = true;
Thanks for the fix!
Could you help me to understand how needs_free_netdev will trigger the free()here?
I _though_ that any device that is being unregistered would have the stats freed.
This is the flow I am reading:
1) When the device is unregistered, then it is marked as todo:
unregister_netdevice_many_notify() {
list_for_each_entry(dev, head, unreg_list) {
net_set_todo(dev);
}
}2) Then, "run_todo" will run later, and it does:
netdev_run_todo() {
list_for_each_entry_safe(dev, tmp, &list, todo_list) {
if (unlikely(dev->reg_state != NETREG_UNREGISTERING)) {
netdev_WARN(dev, "run_todo but not unregistering\n");
list_del(&dev->todo_list);
continue;
}
while (!list_empty(&list)) {
netdev_do_free_pcpu_stats(dev);
}
}Thank you!
Quoted 51 lines · Breno Leitao
Hello Florian,
Quoted 23 lines · Florian Westphal
kmemleak reports
net_deviceresources are no longer released, restoreneeds_free_netdevtoggle. Sample backtrace:unreferenced object
0xffff88810874f000(size 4096): [..][<00000000a2b8af8b>] __kmalloc_node+0x209/0x290 [<0000000040b0a1a9>] alloc_netdev_mqs+0x58/0x470 [<00000000b4be1e78>] vti6_init_net+0x94/0x230 [<000000008830c1ea>] ops_init+0x32/0xc0 [<000000006a26fa8f>] setup_net+0x134/0x2e0[..]
Cc Breno Leitao <[email protected]>Fixes a9b2d55a8f1e ("ip6_vti: Do not use custom stat allocator")Signed-off-by Florian Westphal <[email protected]>--- net/ipv6/ip6_vti.c | 1 + 1 file changed, 1 insertion(+)net/ipv6/ip6_vti.c +0
-0net/ipv6/ip6_vti.c static void vti6_dev_setup(struct net_device *dev) · line 904
Quoted 4 lines
net/ipv6/ip6_vti.c +1
-0@@ -901,6 +901,7 @@ static void vti6_dev_setup(struct net_device *dev) { dev->netdev_ops = &vti6_netdev_ops; dev->header_ops = &ip_tunnel_header_ops; + dev->needs_free_netdev = true;Thanks for the fix!
Could you help me to understand how
needs_free_netdevwill trigger the free()here?I _though_ that any device that is being unregistered would have the stats freed.
This is the flow I am reading:
1) When the device is unregistered, then it is marked as todo:
unregister_netdevice_many_notify() { list_for_each_entry(dev, head, unreg_list) { net_set_todo(dev); } }2) Then, "
run_todo" will run later, and it does:
Quoted 10 lines
netdev_run_todo() {
list_for_each_entry_safe(dev, tmp, &list, todo_list) {
if (unlikely(dev->reg_state != NETREG_UNREGISTERING)) {
netdev_WARN(dev, "run_todo but not unregistering\n");
list_del(&dev->todo_list);
continue;
}
while (!list_empty(&list)) {
netdev_do_free_pcpu_stats(dev);The relevant part is missing:
if (dev->needs_free_netdev)
free_netdev(dev);Quoted 5 lines
} }Thank you!
I do not think Florian patch has anything to do with free_pcpu_stats()
Please take a look at
git show cf124db566e -- net/ipv6/ip6_vti.c
Hello Eric,
Quoted 68 lines · Eric Dumazet
Quoted 51 lines · Breno Leitao
Hello Florian,
Quoted 23 lines · Florian Westphal
kmemleak reports
net_deviceresources are no longer released, restoreneeds_free_netdevtoggle. Sample backtrace:unreferenced object
0xffff88810874f000(size 4096): [..][<00000000a2b8af8b>] __kmalloc_node+0x209/0x290 [<0000000040b0a1a9>] alloc_netdev_mqs+0x58/0x470 [<00000000b4be1e78>] vti6_init_net+0x94/0x230 [<000000008830c1ea>] ops_init+0x32/0xc0 [<000000006a26fa8f>] setup_net+0x134/0x2e0[..]
Cc Breno Leitao <[email protected]>Fixes a9b2d55a8f1e ("ip6_vti: Do not use custom stat allocator")Signed-off-by Florian Westphal <[email protected]>--- net/ipv6/ip6_vti.c | 1 + 1 file changed, 1 insertion(+)net/ipv6/ip6_vti.c +0
-0net/ipv6/ip6_vti.c static void vti6_dev_setup(struct net_device *dev) · line 904
Quoted 4 lines
net/ipv6/ip6_vti.c +1
-0@@ -901,6 +901,7 @@ static void vti6_dev_setup(struct net_device *dev) { dev->netdev_ops = &vti6_netdev_ops; dev->header_ops = &ip_tunnel_header_ops; + dev->needs_free_netdev = true;Thanks for the fix!
Could you help me to understand how
needs_free_netdevwill trigger the free()here?I _though_ that any device that is being unregistered would have the stats freed.
This is the flow I am reading:
1) When the device is unregistered, then it is marked as todo:
unregister_netdevice_many_notify() { list_for_each_entry(dev, head, unreg_list) { net_set_todo(dev); } }2) Then, "
run_todo" will run later, and it does:Quoted 10 lines
netdev_run_todo() { list_for_each_entry_safe(dev, tmp, &list, todo_list) { if (unlikely(dev->reg_state != NETREG_UNREGISTERING)) { netdev_WARN(dev, "run_todo but not unregistering\n"); list_del(&dev->todo_list); continue; } while (!list_empty(&list)) { netdev_do_free_pcpu_stats(dev);The relevant part is missing:
if (dev->needs_free_netdev) free_netdev(dev);
Quoted 7 lines
Quoted 5 lines
} }Thank you!
I do not think Florian patch has anything to do with
free_pcpu_stats()
That makes total sense now.
I though the memory that was being leaked was related to pcpu stats, since this is was I was supposedly changing, but, it is not.
The regression is not related to the fact that a9b2d55a8f1e ("ip6_vti: Do not use custom stat allocator") removed the following line wrongly:
-dev->needs_free_netdev = true;
Everything is clear now.
Thanks!