include/net/net_namespace.h | 1 + net/core/net_namespace.c | 34 +++++++++++++++++++++------------- 2 files changed, 22 insertions(+), 13 deletions(-)
Currently, unhash_nsid() scans the entire system for each netns being
killed, leading to O(M_dying_net * N_alive_net * N_id) complexity, as
__peernet2id() also performs a linear search in the IDR.
Optimize this to O(N_alive_net * N_id) by batching unhash operations. Move
unhash_nsid() out of the per-netns loop in cleanup_net() to perform a
single-pass traversal over survivor namespaces.
Identify dying peers by an 'is_dying' flag, which is set under net_rwsem
write lock after the netns is removed from the global list. This batches
the unhashing work and eliminates the O(M_dying_net) multiplier.
To minimize the impact on struct net size, 'is_dying' is placed in an
existing hole after 'hash_mix' in struct net.
Use a restartable idr_get_next() loop for iteration. This avoids the
unsafe modification issue inherent to idr_for_each() callbacks and allows
dropping the nsid_lock to safely call sleepy rtnl_net_notifyid().
Clean up redundant nsid_lock and simplify the destruction loop now that
unhashing is centralized.
Signed-off-by: Qiliang Yuan <yuanql9@chinatelecom.cn>
---
v6:
- Use M_dying_net and N_alive_net terminology for clarity.
- Correct complexity analysis: __peernet2id() performs a linear search.
- Move 'is_dying' to a structural hole after 'hash_mix' to save memory.
- Scope 'id' variable locally within the traversal loop.
- Simplify IDR traversal logic with unconditional increment.
v5:
- Use idr_get_next() for restartable iteration safely handling removals.
- Drop unhash_nsid_callback() to avoid context safety issues.
v4:
- Move unhash_nsid() out of the batch loop to reduce complexity from O(M*N) to O(N).
- Use idr_for_each() for efficient, single-pass IDR traversal.
- Mark 'is_dying' under net_rwsem to safely identify and batch unhashing.
- Simplify destruction loop by removing redundant locking and per-netns unhash logic.
v3:
- Update target tree to net-next.
- Post as a new thread instead of a reply.
v2:
- Move 'is_dying' setting to __put_net() to eliminate the O(M_batch) loop.
- Remove redundant initializations in preinit_net().
v1:
- Initial implementation of batch unhash_nsid().
include/net/net_namespace.h | 1 +
net/core/net_namespace.c | 34 +++++++++++++++++++++-------------
2 files changed, 22 insertions(+), 13 deletions(-)
diff --git a/include/net/net_namespace.h b/include/net/net_namespace.h
index cb664f6e3558..db291cc7afe3 100644
--- a/include/net/net_namespace.h
+++ b/include/net/net_namespace.h
@@ -120,6 +120,7 @@ struct net {
* it is critical that it is on a read_mostly cache line.
*/
u32 hash_mix;
+ bool is_dying;
struct net_device *loopback_dev; /* The loopback */
diff --git a/net/core/net_namespace.c b/net/core/net_namespace.c
index a6e6a964a287..aef44e617361 100644
--- a/net/core/net_namespace.c
+++ b/net/core/net_namespace.c
@@ -624,9 +624,10 @@ void net_ns_get_ownership(const struct net *net, kuid_t *uid, kgid_t *gid)
}
EXPORT_SYMBOL_GPL(net_ns_get_ownership);
-static void unhash_nsid(struct net *net, struct net *last)
+static void unhash_nsid(struct net *last)
{
- struct net *tmp;
+ struct net *tmp, *peer;
+
/* This function is only called from cleanup_net() work,
* and this work is the only process, that may delete
* a net from net_namespace_list. So, when the below
@@ -634,22 +635,26 @@ static void unhash_nsid(struct net *net, struct net *last)
* use for_each_net_rcu() or net_rwsem.
*/
for_each_net(tmp) {
- int id;
+ int id = 0;
spin_lock(&tmp->nsid_lock);
- id = __peernet2id(tmp, net);
- if (id >= 0)
- idr_remove(&tmp->netns_ids, id);
- spin_unlock(&tmp->nsid_lock);
- if (id >= 0)
- rtnl_net_notifyid(tmp, RTM_DELNSID, id, 0, NULL,
+ while ((peer = idr_get_next(&tmp->netns_ids, &id))) {
+ int curr_id = id;
+
+ id++;
+ if (!peer->is_dying)
+ continue;
+
+ idr_remove(&tmp->netns_ids, curr_id);
+ spin_unlock(&tmp->nsid_lock);
+ rtnl_net_notifyid(tmp, RTM_DELNSID, curr_id, 0, NULL,
GFP_KERNEL);
+ spin_lock(&tmp->nsid_lock);
+ }
+ spin_unlock(&tmp->nsid_lock);
if (tmp == last)
break;
}
- spin_lock(&net->nsid_lock);
- idr_destroy(&net->netns_ids);
- spin_unlock(&net->nsid_lock);
}
static LLIST_HEAD(cleanup_list);
@@ -674,6 +679,7 @@ static void cleanup_net(struct work_struct *work)
llist_for_each_entry(net, net_kill_list, cleanup_list) {
ns_tree_remove(net);
list_del_rcu(&net->list);
+ net->is_dying = true;
}
/* Cache last net. After we unlock rtnl, no one new net
* added to net_namespace_list can assign nsid pointer
@@ -688,8 +694,10 @@ static void cleanup_net(struct work_struct *work)
last = list_last_entry(&net_namespace_list, struct net, list);
up_write(&net_rwsem);
+ unhash_nsid(last);
+
llist_for_each_entry(net, net_kill_list, cleanup_list) {
- unhash_nsid(net, last);
+ idr_destroy(&net->netns_ids);
list_add_tail(&net->exit_list, &net_exit_list);
}
--
2.51.0On Tue, Feb 3, 2026 at 11:49 PM Qiliang Yuan <realwujing@gmail.com> wrote:
>
> Currently, unhash_nsid() scans the entire system for each netns being
> killed, leading to O(M_dying_net * N_alive_net * N_id) complexity, as
Sorry I should've used L_dying_net, M_alive_net .. as N is not
the same here.
> __peernet2id() also performs a linear search in the IDR.
>
> Optimize this to O(N_alive_net * N_id) by batching unhash operations. Move
> unhash_nsid() out of the per-netns loop in cleanup_net() to perform a
> single-pass traversal over survivor namespaces.
>
> Identify dying peers by an 'is_dying' flag, which is set under net_rwsem
> write lock after the netns is removed from the global list. This batches
> the unhashing work and eliminates the O(M_dying_net) multiplier.
>
> To minimize the impact on struct net size, 'is_dying' is placed in an
> existing hole after 'hash_mix' in struct net.
>
> Use a restartable idr_get_next() loop for iteration. This avoids the
> unsafe modification issue inherent to idr_for_each() callbacks and allows
> dropping the nsid_lock to safely call sleepy rtnl_net_notifyid().
>
> Clean up redundant nsid_lock and simplify the destruction loop now that
> unhashing is centralized.
>
> Signed-off-by: Qiliang Yuan <yuanql9@chinatelecom.cn>
Code-wise, looks good to me:
Reviewed-by: Kuniyuki Iwashima <kuniyu@google.com>
> ---
> v6:
> - Use M_dying_net and N_alive_net terminology for clarity.
> - Correct complexity analysis: __peernet2id() performs a linear search.
> - Move 'is_dying' to a structural hole after 'hash_mix' to save memory.
> - Scope 'id' variable locally within the traversal loop.
> - Simplify IDR traversal logic with unconditional increment.
> v5:
> - Use idr_get_next() for restartable iteration safely handling removals.
> - Drop unhash_nsid_callback() to avoid context safety issues.
> v4:
> - Move unhash_nsid() out of the batch loop to reduce complexity from O(M*N) to O(N).
> - Use idr_for_each() for efficient, single-pass IDR traversal.
> - Mark 'is_dying' under net_rwsem to safely identify and batch unhashing.
> - Simplify destruction loop by removing redundant locking and per-netns unhash logic.
> v3:
> - Update target tree to net-next.
> - Post as a new thread instead of a reply.
> v2:
> - Move 'is_dying' setting to __put_net() to eliminate the O(M_batch) loop.
> - Remove redundant initializations in preinit_net().
> v1:
> - Initial implementation of batch unhash_nsid().
>
> include/net/net_namespace.h | 1 +
> net/core/net_namespace.c | 34 +++++++++++++++++++++-------------
> 2 files changed, 22 insertions(+), 13 deletions(-)
>
> diff --git a/include/net/net_namespace.h b/include/net/net_namespace.h
> index cb664f6e3558..db291cc7afe3 100644
> --- a/include/net/net_namespace.h
> +++ b/include/net/net_namespace.h
> @@ -120,6 +120,7 @@ struct net {
> * it is critical that it is on a read_mostly cache line.
> */
> u32 hash_mix;
> + bool is_dying;
>
> struct net_device *loopback_dev; /* The loopback */
>
> diff --git a/net/core/net_namespace.c b/net/core/net_namespace.c
> index a6e6a964a287..aef44e617361 100644
> --- a/net/core/net_namespace.c
> +++ b/net/core/net_namespace.c
> @@ -624,9 +624,10 @@ void net_ns_get_ownership(const struct net *net, kuid_t *uid, kgid_t *gid)
> }
> EXPORT_SYMBOL_GPL(net_ns_get_ownership);
>
> -static void unhash_nsid(struct net *net, struct net *last)
> +static void unhash_nsid(struct net *last)
> {
> - struct net *tmp;
> + struct net *tmp, *peer;
> +
> /* This function is only called from cleanup_net() work,
> * and this work is the only process, that may delete
> * a net from net_namespace_list. So, when the below
> @@ -634,22 +635,26 @@ static void unhash_nsid(struct net *net, struct net *last)
> * use for_each_net_rcu() or net_rwsem.
> */
> for_each_net(tmp) {
> - int id;
> + int id = 0;
>
> spin_lock(&tmp->nsid_lock);
> - id = __peernet2id(tmp, net);
> - if (id >= 0)
> - idr_remove(&tmp->netns_ids, id);
> - spin_unlock(&tmp->nsid_lock);
> - if (id >= 0)
> - rtnl_net_notifyid(tmp, RTM_DELNSID, id, 0, NULL,
> + while ((peer = idr_get_next(&tmp->netns_ids, &id))) {
> + int curr_id = id;
> +
> + id++;
> + if (!peer->is_dying)
> + continue;
> +
> + idr_remove(&tmp->netns_ids, curr_id);
> + spin_unlock(&tmp->nsid_lock);
> + rtnl_net_notifyid(tmp, RTM_DELNSID, curr_id, 0, NULL,
> GFP_KERNEL);
> + spin_lock(&tmp->nsid_lock);
> + }
> + spin_unlock(&tmp->nsid_lock);
> if (tmp == last)
> break;
> }
> - spin_lock(&net->nsid_lock);
> - idr_destroy(&net->netns_ids);
> - spin_unlock(&net->nsid_lock);
> }
>
> static LLIST_HEAD(cleanup_list);
> @@ -674,6 +679,7 @@ static void cleanup_net(struct work_struct *work)
> llist_for_each_entry(net, net_kill_list, cleanup_list) {
> ns_tree_remove(net);
> list_del_rcu(&net->list);
> + net->is_dying = true;
> }
> /* Cache last net. After we unlock rtnl, no one new net
> * added to net_namespace_list can assign nsid pointer
> @@ -688,8 +694,10 @@ static void cleanup_net(struct work_struct *work)
> last = list_last_entry(&net_namespace_list, struct net, list);
> up_write(&net_rwsem);
>
> + unhash_nsid(last);
> +
> llist_for_each_entry(net, net_kill_list, cleanup_list) {
> - unhash_nsid(net, last);
> + idr_destroy(&net->netns_ids);
> list_add_tail(&net->exit_list, &net_exit_list);
> }
>
> --
> 2.51.0
>
© 2016 - 2026 Red Hat, Inc.