* [PATCH wireguard-tools v3 1/3] ipc: linux: filter netdevices kernel-side
2026-01-30 19:10 [PATCH wireguard-tools v3 0/3] ipc: linux: kernel-side netdevice filtering Asbjørn Sloth Tønnesen
@ 2026-01-30 19:10 ` Asbjørn Sloth Tønnesen
2026-01-30 19:10 ` [PATCH wireguard-tools v3 2/3] ipc: linux: skip statistics on netdevice listing Asbjørn Sloth Tønnesen
` (2 subsequent siblings)
3 siblings, 0 replies; 5+ messages in thread
From: Asbjørn Sloth Tønnesen @ 2026-01-30 19:10 UTC (permalink / raw)
To: Jason A . Donenfeld; +Cc: Asbjørn Sloth Tønnesen, wireguard
Tell the kernel that we are only interested in wireguard netdevices,
so that the kernel don't have to dump all the other netdevices.
Kernel-side support for this was added in Linux v4.6 in
commit dc599f76c22b ("net: Add support for filtering link dump by
master device and kind").
Tested with 10000 netdevices (common with ISP BNG setups), and a
single wireguard netdevice.
Baseline:
# time ./src/wg show
real 0m0.342s
user 0m0.013s
sys 0m0.290s
With patch:
# time ./src/wg show
real 0m0.006s
user 0m0.000s
sys 0m0.005s
Signed-off-by: Asbjørn Sloth Tønnesen <ast@fiberby.net>
---
src/ipc-linux.h | 6 ++++++
1 file changed, 6 insertions(+)
diff --git a/src/ipc-linux.h b/src/ipc-linux.h
index 01247f1..c56fede 100644
--- a/src/ipc-linux.h
+++ b/src/ipc-linux.h
@@ -80,6 +80,7 @@ static int kernel_get_wireguard_interfaces(struct string_list *list)
int ret = 0;
struct nlmsghdr *nlh;
struct ifinfomsg *ifm;
+ struct nlattr *linkinfo_nest;
ret = -ENOMEM;
rtnl_buffer = calloc(SOCKET_BUFFER_SIZE, 1);
@@ -105,6 +106,11 @@ static int kernel_get_wireguard_interfaces(struct string_list *list)
nlh->nlmsg_seq = seq;
ifm = mnl_nlmsg_put_extra_header(nlh, sizeof(*ifm));
ifm->ifi_family = AF_UNSPEC;
+
+ linkinfo_nest = mnl_attr_nest_start(nlh, IFLA_LINKINFO);
+ mnl_attr_put_strz(nlh, IFLA_INFO_KIND, WG_GENL_NAME);
+ mnl_attr_nest_end(nlh, linkinfo_nest);
+
message_len = nlh->nlmsg_len;
if (mnl_socket_sendto(nl, rtnl_buffer, message_len) < 0) {
--
2.51.0
^ permalink raw reply related [flat|nested] 5+ messages in thread* [PATCH wireguard-tools v3 2/3] ipc: linux: skip statistics on netdevice listing
2026-01-30 19:10 [PATCH wireguard-tools v3 0/3] ipc: linux: kernel-side netdevice filtering Asbjørn Sloth Tønnesen
2026-01-30 19:10 ` [PATCH wireguard-tools v3 1/3] ipc: linux: filter netdevices kernel-side Asbjørn Sloth Tønnesen
@ 2026-01-30 19:10 ` Asbjørn Sloth Tønnesen
2026-01-30 19:10 ` [PATCH wireguard-tools v3 3/3] ipc: linux: remove user-space netdevice filtering Asbjørn Sloth Tønnesen
2026-03-11 23:02 ` [PATCH wireguard-tools v3 0/3] ipc: linux: kernel-side " Jason A. Donenfeld
3 siblings, 0 replies; 5+ messages in thread
From: Asbjørn Sloth Tønnesen @ 2026-01-30 19:10 UTC (permalink / raw)
To: Jason A . Donenfeld; +Cc: Asbjørn Sloth Tønnesen, wireguard
Gathering interface statistics can be a relatively expensive operation
on certain systems as it requires iterating over all the cpus.
This patch instructs the kernel to not include netdevice statistics,
in the netdevice listing.
Kernel-side support for this was added in Linux v4.4 in commit
d5566fd72ec1 ("rtnetlink: RTEXT_FILTER_SKIP_STATS support to avoid
dumping inet/inet6 stats"), this saves 368 bytes of statistics.
In the Linux v6.19 commit 105bae321862 ("rtnetlink: honor
RTEXT_FILTER_SKIP_STATS in IFLA_STATS"), it is expanded to skip more
statistics bringing the savings to 800 bytes per netdevice.
As the minimum kernel version for wireguard-tools is unclear, it's
wrapped in a #ifdef guard. If the minimum version is v4.4+ it can be
removed.
Signed-off-by: Asbjørn Sloth Tønnesen <ast@fiberby.net>
---
src/ipc-linux.h | 3 +++
1 file changed, 3 insertions(+)
diff --git a/src/ipc-linux.h b/src/ipc-linux.h
index c56fede..61e2c1a 100644
--- a/src/ipc-linux.h
+++ b/src/ipc-linux.h
@@ -106,6 +106,9 @@ static int kernel_get_wireguard_interfaces(struct string_list *list)
nlh->nlmsg_seq = seq;
ifm = mnl_nlmsg_put_extra_header(nlh, sizeof(*ifm));
ifm->ifi_family = AF_UNSPEC;
+#ifdef RTEXT_FILTER_SKIP_STATS
+ mnl_attr_put_u32(nlh, IFLA_EXT_MASK, RTEXT_FILTER_SKIP_STATS);
+#endif
linkinfo_nest = mnl_attr_nest_start(nlh, IFLA_LINKINFO);
mnl_attr_put_strz(nlh, IFLA_INFO_KIND, WG_GENL_NAME);
--
2.51.0
^ permalink raw reply related [flat|nested] 5+ messages in thread* [PATCH wireguard-tools v3 3/3] ipc: linux: remove user-space netdevice filtering
2026-01-30 19:10 [PATCH wireguard-tools v3 0/3] ipc: linux: kernel-side netdevice filtering Asbjørn Sloth Tønnesen
2026-01-30 19:10 ` [PATCH wireguard-tools v3 1/3] ipc: linux: filter netdevices kernel-side Asbjørn Sloth Tønnesen
2026-01-30 19:10 ` [PATCH wireguard-tools v3 2/3] ipc: linux: skip statistics on netdevice listing Asbjørn Sloth Tønnesen
@ 2026-01-30 19:10 ` Asbjørn Sloth Tønnesen
2026-03-11 23:02 ` [PATCH wireguard-tools v3 0/3] ipc: linux: kernel-side " Jason A. Donenfeld
3 siblings, 0 replies; 5+ messages in thread
From: Asbjørn Sloth Tønnesen @ 2026-01-30 19:10 UTC (permalink / raw)
To: Jason A . Donenfeld; +Cc: Asbjørn Sloth Tønnesen, wireguard
As devices are now filtered kernel-side, then we can remove
the code for filtering in user-space.
This breaks device listing for kernels earlier than Linux v4.6,
device-specific commands will continue to work.
As the minimum kernel version for wireguard-tools is unclear, it's
unclear if the user-space filtering can be removed yet.
Signed-off-by: Asbjørn Sloth Tønnesen <ast@fiberby.net>
---
src/ipc-linux.h | 16 ++--------------
1 file changed, 2 insertions(+), 14 deletions(-)
diff --git a/src/ipc-linux.h b/src/ipc-linux.h
index 61e2c1a..639fb00 100644
--- a/src/ipc-linux.h
+++ b/src/ipc-linux.h
@@ -29,25 +29,13 @@
struct interface {
const char *name;
- bool is_wireguard;
};
-static int parse_linkinfo(const struct nlattr *attr, void *data)
-{
- struct interface *interface = data;
-
- if (mnl_attr_get_type(attr) == IFLA_INFO_KIND && !strcmp(WG_GENL_NAME, mnl_attr_get_str(attr)))
- interface->is_wireguard = true;
- return MNL_CB_OK;
-}
-
static int parse_infomsg(const struct nlattr *attr, void *data)
{
struct interface *interface = data;
- if (mnl_attr_get_type(attr) == IFLA_LINKINFO)
- return mnl_attr_parse_nested(attr, parse_linkinfo, data);
- else if (mnl_attr_get_type(attr) == IFLA_IFNAME)
+ if (mnl_attr_get_type(attr) == IFLA_IFNAME)
interface->name = mnl_attr_get_str(attr);
return MNL_CB_OK;
}
@@ -61,7 +49,7 @@ static int read_devices_cb(const struct nlmsghdr *nlh, void *data)
ret = mnl_attr_parse(nlh, sizeof(struct ifinfomsg), parse_infomsg, &interface);
if (ret != MNL_CB_OK)
return ret;
- if (interface.name && interface.is_wireguard)
+ if (interface.name)
ret = string_list_add(list, interface.name);
if (ret < 0)
return ret;
--
2.51.0
^ permalink raw reply related [flat|nested] 5+ messages in thread* Re: [PATCH wireguard-tools v3 0/3] ipc: linux: kernel-side netdevice filtering
2026-01-30 19:10 [PATCH wireguard-tools v3 0/3] ipc: linux: kernel-side netdevice filtering Asbjørn Sloth Tønnesen
` (2 preceding siblings ...)
2026-01-30 19:10 ` [PATCH wireguard-tools v3 3/3] ipc: linux: remove user-space netdevice filtering Asbjørn Sloth Tønnesen
@ 2026-03-11 23:02 ` Jason A. Donenfeld
3 siblings, 0 replies; 5+ messages in thread
From: Jason A. Donenfeld @ 2026-03-11 23:02 UTC (permalink / raw)
To: Asbjørn Sloth Tønnesen; +Cc: wireguard
On Fri, Jan 30, 2026 at 07:10:52PM +0000, Asbjørn Sloth Tønnesen wrote:
> Move device filtering to the kernel, thereby reducing netlink traffic.
>
> The first patch request kernel-side filtering.
>
> The second patch requests that the kernel doesn't include statistics
> in netdevice dumps.
>
> The third patch removes the old filtering code, as an
> additional step, which breaks on earlier than Linux v4.6.
>
> I assume that a dependency on Linux v4.6+ is acceptable for wg-tools
> now, as wireguard-linux-compat haven't been updated for 3 years.
Interesting series. If I'm going to apply this, I'm going to apply it,
and so the ifdefs you've added won't matter much; this will still break
wireguard-linux-compat. But maybe it's time to do so. So if you're up
for sending a v4, just do it like there's no yesterday.
Also, does this filtering need to also be added to the embeddable c
library contrib code?
Thanks for this patch. I'll start thinking seriously about
wireguard-linux-compat sunsetting.
Jason
^ permalink raw reply [flat|nested] 5+ messages in thread