iproute2: Add 'netshaper' command to 'ip link' for netdev shaping
Subsystems (2)
PATCH v3 1/1 iproute2: Add 'netshaper' command to 'ip link' for netdev shaping
Add support for the netshaper Generic Netlink
family to iproute2. Introduce a new subcommand to ip link for configuring netshaper parameters directly from userspace.
This interface allows users to set shaping attributes (such as speed) which are passed to the kernel to perform the corresponding netshaper operation.
Example usage:
$ip link netshaper { set | get | delete } dev DEVNAME \
handle scope SCOPE id ID \
[ speed SPEED ]
Internally, this triggers a kernel call to apply the shaping configuration to the specified network device.
Currently, the tool supports the following functionalities:
- Setting speed in Mbps, enabling bandwidth clamping for a network device that support netshaper operations.
- Deleting the current configuration.
- Querying the existing configuration.
Additional netshaper operations will be integrated into the tool as per requirement.
This change enables easy and scriptable configuration of bandwidth shaping for devices that use the netshaper Netlink family.
Corresponding net-next patches:
1) https://lore.kernel.org/all/[email protected]/
2) https://lore.kernel.org/lkml/[email protected]/
Install pkg-config and libmnl* packages to print kernel extack errors to stdout.
---
Please add include/uapi/linux/net_shaper.h from kernel source tree for this patch.
---
Changes in v3:
* Use strcmp instead of matches.
* Use get_rate64 instead get_unsigned for speed parameter.
* Remove speed_mbps in do_cmd() to reduce redundancy.
* Update the usage of speed parameter in the command. Changes in v2:
* Use color coding for printing device name in stdout.
* Use clang-format to format the code inline.
* Use __u64 for speed_bps.
* Remove include/uapi/linux/netshaper.h file.
--- ip/Makefile | 2 +- ip/iplink.c | 12 +++ ip/iplink_netshaper.c | 189 ++++++++++++++++++++++++++++++++++++++++++ 3 files changed, 202 insertions(+), 1 deletion(-) create mode 100644 ip/iplink_netshaper.c
ip/Makefile +1 -1
@@ -4,7 +4,7 @@ IPOBJ=ip.o ipaddress.o ipaddrlabel.o iproute.o iprule.o ipnetns.o \ ipmaddr.o ipmonitor.o ipmroute.o ipprefix.o iptuntap.o iptoken.o \ ipxfrm.o xfrm_state.o xfrm_policy.o xfrm_monitor.o iplink_dummy.o \ iplink_ifb.o iplink_nlmon.o iplink_team.o iplink_vcan.o iplink_vxcan.o \ - iplink_vlan.o link_veth.o link_gre.o iplink_can.o iplink_xdp.o \ + iplink_vlan.o iplink_netshaper.o link_veth.o link_gre.o iplink_can.o iplink_xdp.o \ iplink_macvlan.o ipl2tp.o link_vti.o link_vti6.o link_xfrm.o \ iplink_vxlan.o tcp_metrics.o iplink_ipoib.o ipnetconf.o link_ip6tnl.o \ link_iptnl.o link_gre6.o iplink_bond.o iplink_bond_slave.o iplink_hsr.o \
ip/iplink.c +12 -0
@@ -1509,6 +1509,15 @@ static void do_help(int argc, char **argv)usage (); } +static int iplink_netshaper (int argc,char **argv) +{ +struct link_util *lu; + + lu =get_link_kind ("netshaper" ); + +return lu->parse_opt (lu, argc, argv,NULL ); +} +int do_iplink (int argc,char **argv) {if (argc <1 )
@@ -1545,6 +1554,9 @@ int do_iplink(int argc, char **argv)if (matches (*argv,"property" ) ==0 )return iplink_prop (argc-1 , argv+1 ); +if (strcmp (*argv,"netshaper" ) ==0 ) +return iplink_netshaper (argc-1 , argv+1 ); +if (matches (*argv,"help" ) ==0 ) {do_help (argc-1 , argv+1 );return 0 ;
ip/iplink_netshaper.c new file +189 -0
189 lines not on this page · raw message
Signature · 1 line
2.43.0
Quoted 60 lines
-----Original Message-----
From: Erni Sri Satya Vennela <[email protected]>
Sent: Monday, August 11, 2025 3:05 AM
To: [email protected]; [email protected]; [email protected]
Cc: Haiyang Zhang <[email protected]>; [email protected]; Saurabh Singh Sengar <[email protected]>; Dipayaan Roy <[email protected]>; Erni Sri Satya Vennela <[email protected]>; Erni Sri Satya Vennela <[email protected]>
Subject: [PATCH iproute2-next v3] iproute2: Add 'netshaper' command to 'ip link' for netdev shapingAdd support for the netshaper Generic Netlink
family to iproute2. Introduce a new subcommand toip linkfor configuring netshaper parameters directly from userspace.This interface allows users to set shaping attributes (such as speed) which are passed to the kernel to perform the corresponding netshaper operation.
Example usage:
$ip link netshaper { set | get | delete } dev DEVNAME \
handle scope SCOPE id ID \
[ speed SPEED ]Internally, this triggers a kernel call to apply the shaping configuration to the specified network device.
Currently, the tool supports the following functionalities:
- Setting speed in Mbps, enabling bandwidth clamping for a network device that support netshaper operations.
- Deleting the current configuration.
- Querying the existing configuration.Additional netshaper operations will be integrated into the tool as per requirement.
This change enables easy and scriptable configuration of bandwidth shaping for devices that use the netshaper Netlink family.
Corresponding net-next patches:
1)
https://lore.ker/
nel.org%2Fall%2Fcover.1728460186.git.pabeni%40redhat.com%2F&data=05%7C02%7 Chaiyangz%40microsoft.com%7C0ac998f8db0c4cf31d8208ddd8a5658b%7C72f988bf86f 141af91ab2d7cd011db47%7C1%7C0%7C638904927080914227%7CUnknown%7CTWFpbGZsb3d 8eyJFbXB0eU1hcGkiOnRydWUsIlYiOiIwLjAuMDAwMCIsIlAiOiJXaW4zMiIsIkFOIjoiTWFpb CIsIldUIjoyfQ%3D%3D%7C0%7C%7C%7C&sdata=WY7zqpx5F9Wnf0WbHUIxMDGbwmjwQysjTdD cPAZ5jzM%3D&reserved=0
2)
https://lore.ker/
nel.org%2Flkml%2F1750144656-2021-1-git-send-email- ernis%40linux.microsoft.com%2F&data=05%7C02%7Chaiyangz%40microsoft.com%7C0 ac998f8db0c4cf31d8208ddd8a5658b%7C72f988bf86f141af91ab2d7cd011db47%7C1%7C0 %7C638904927080928775%7CUnknown%7CTWFpbGZsb3d8eyJFbXB0eU1hcGkiOnRydWUsIlYi OiIwLjAuMDAwMCIsIlAiOiJXaW4zMiIsIkFOIjoiTWFpbCIsIldUIjoyfQ%3D%3D%7C0%7C%7C %7C&sdata=TmzKYLtKlU%2FphFWfa13toeOmJGAU1RgBBtRvp%2F3Wg5s%3D&reserved=0
Quoted 4 lines
Install pkg-config and libmnl* packages to print kernel extack errors to stdout.
Signed-off-by Erni Sri Satya Vennela <[email protected]>
Quoted 50 lines · On Mon, 11 Aug 2025 00:05:02 -0700 Erni Sri Satya Vennela…
Add support for the netshaper Generic Netlink
family to iproute2. Introduce a new subcommand toip linkfor configuring netshaper parameters directly from userspace.This interface allows users to set shaping attributes (such as speed) which are passed to the kernel to perform the corresponding netshaper operation.
Example usage:
$ip link netshaper { set | get | delete } dev DEVNAME \
handle scope SCOPE id ID \
[ speed SPEED ]Internally, this triggers a kernel call to apply the shaping configuration to the specified network device.
Currently, the tool supports the following functionalities:
- Setting speed in Mbps, enabling bandwidth clamping for a network device that support netshaper operations.
- Deleting the current configuration.
- Querying the existing configuration.Additional netshaper operations will be integrated into the tool as per requirement.
This change enables easy and scriptable configuration of bandwidth shaping for devices that use the netshaper Netlink family.
Corresponding net-next patches:
1) https://lore.kernel.org/all/[email protected]/
2) https://lore.kernel.org/lkml/[email protected]/Install pkg-config and libmnl* packages to print kernel extack errors to stdout.
Signed-off-by Erni Sri Satya Vennela <[email protected]>---
Please addinclude/uapi/linux/net_shaper.hfrom kernel source tree for this patch.
---
Changes in v3:
* Use strcmp instead of matches.
* Useget_rate64insteadget_unsignedfor speed parameter.
* Removespeed_mbpsindo_cmd()to reduce redundancy.
* Update the usage of speed parameter in the command. Changes in v2:
* Use color coding for printing device name in stdout.
* Use clang-format to format the code inline.
* Use__u64forspeed_bps.
* Removeinclude/uapi/linux/netshaper.hfile.
Quoted 6 lines
--- ip/Makefile | 2 +- ip/iplink.c | 12 +++ ip/iplink_netshaper.c | 189 ++++++++++++++++++++++++++++++++++++++++++ 3 files changed, 202 insertions(+), 1 deletion(-) create mode 100644 ip/iplink_netshaper.c
No documentation.
No tests?
Quoted 66 lines
ip/Makefile +1
-1@@ -4,7 +4,7 @@ IPOBJ=ip.o ipaddress.o ipaddrlabel.o iproute.o iprule.o ipnetns.o \ ipmaddr.o ipmonitor.o ipmroute.o ipprefix.o iptuntap.o iptoken.o \ ipxfrm.o xfrm_state.o xfrm_policy.o xfrm_monitor.o iplink_dummy.o \ iplink_ifb.o iplink_nlmon.o iplink_team.o iplink_vcan.o iplink_vxcan.o \ - iplink_vlan.o link_veth.o link_gre.o iplink_can.o iplink_xdp.o \ + iplink_vlan.o iplink_netshaper.o link_veth.o link_gre.o iplink_can.o iplink_xdp.o \ iplink_macvlan.o ipl2tp.o link_vti.o link_vti6.o link_xfrm.o \ iplink_vxlan.o tcp_metrics.o iplink_ipoib.o ipnetconf.o link_ip6tnl.o \ link_iptnl.o link_gre6.o iplink_bond.o iplink_bond_slave.o iplink_hsr.o \ip/iplink.c +12
-0@@ -1509,6 +1509,15 @@ static void do_help(int argc, char **argv)usage (); } +static int iplink_netshaper (int argc,char **argv) +{ +struct link_util *lu; + + lu =get_link_kind ("netshaper" ); + +return lu->parse_opt (lu, argc, argv,NULL ); +} +int do_iplink (int argc,char **argv) {if (argc <1 )@@ -1545,6 +1554,9 @@ int do_iplink(int argc, char **argv)if (matches (*argv,"property" ) ==0 )return iplink_prop (argc-1 , argv+1 ); +if (strcmp (*argv,"netshaper" ) ==0 ) +return iplink_netshaper (argc-1 , argv+1 ); +if (matches (*argv,"help" ) ==0 ) {do_help (argc-1 , argv+1 );return 0 ;ip/iplink_netshaper.c new file +16
-0@@ -0,0 +1,189 @@ +/* SPDX-License-Identifier: GPL-2.0 */ +/* +* iplink_netshaper.c netshaper H/W shaping support +* +* Authors: Erni Sri Satya Vennela <[email protected]> +*/ +#include <stdio.h> +#include <string.h> +#include <linux/genetlink.h> +#include <linux/netlink.h> +#include <linux/rtnetlink.h> +#include <uapi/linux/netshaper.h> +#include "utils.h" +#include "ip_common.h" +#include "libgenl.h" +
ip/iplink_netshaper.c line 26
Quoted 10 lines
ip/iplink_netshaper.c new file +10
-0+/* netlink socket */ +static struct rtnl_handle gen_rth = { .fd =-1 }; +static int genl_family =-1 ; + +static void usage (void ) +{ +fprintf (stderr, +"Usage: ip link netshaper set dev DEVNAME handle scope HANDLE_SCOPE id HANDLE_ID speed SPEED\n" +" ip link netshaper delete dev DEVNAME handle scope HANDLE_SCOPE id HANDLE_ID\n" + " ip link netshaper get dev DEVNAME handle scope HANDLE_SCOPE id HANDLE_ID\n"
Seems like netshaper is property of the device not a top level object, the syntax here is awkward. Open to better suggestions. Looks to me that netshaper should be property of device not top level object.
Or netshaper is like neighbour and route and really wants to be not part of link command.
Also no other ip commands use show not get.
The verb "get" is only used for things like getting a route or neighbor with a particular address.
Quoted 153 lines
ip/iplink_netshaper.c new file +153
-0+"Where: DEVNAME := STRING\n" +" HANDLE_SCOPE := { netdev | queue | node }\n" +" HANDLE_ID := UINT\n" + " SPEED := UINT{ kbit | mbit | gbit }\n"); + +exit (-1 ); +} + +static void print_netshaper_attrs (struct nlmsghdr *answer) +{ +struct genlmsghdr *ghdr =NLMSG_DATA (answer); +int len = answer->nlmsg_len -NLMSG_LENGTH (GENL_HDRLEN ); +struct rtattr *tb[NET_SHAPER_A_MAX +1 ] = {}; +__u32 speed_mbps; +__u64 speed_bps; +int ifindex; + +parse_rtattr (tb,NET_SHAPER_A_MAX , + (struct rtattr *)((char *)ghdr +GENL_HDRLEN ), len); + +for (int i =1 ; i <=NET_SHAPER_A_MAX ; ++i) { +if (!tb[i]) +continue ; +switch (i) { +case NET_SHAPER_A_BW_MAX : + speed_bps =rta_getattr_u64 (tb[i]); + speed_mbps = (speed_bps /1000000 ); +print_uint (PRINT_ANY ,"speed" ,"speed: %u mbps\n" , + speed_mbps); +break ; +case NET_SHAPER_A_IFINDEX : + ifindex =rta_getattr_u32 (tb[i]); +print_color_string (PRINT_ANY ,COLOR_IFNAME ,"dev" , +"dev: %s\n" , +ll_index_to_name (ifindex)); +break ; +default : +break ; + } + } +} + +static int do_cmd (int argc,char **argv,struct nlmsghdr *n,int cmd) +{ +GENL_REQUEST (req,1024 , genl_family,0 ,NET_SHAPER_FAMILY_VERSION , cmd, +NLM_F_REQUEST |NLM_F_ACK ); + +struct nlmsghdr *answer; +__u64 speed_bps =0 ; +int ifindex =-1 ; +int handle_scope =NET_SHAPER_SCOPE_UNSPEC ; +__u32 handle_id =0 ; +bool handle_present = false; +int err; + +while (argc >0 ) { +if (strcmp (*argv,"dev" ) ==0 ) { +NEXT_ARG (); + ifindex =ll_name_to_index (*argv); + }else if (strcmp (*argv,"speed" ) ==0 ) { +NEXT_ARG (); +if (get_rate64 (&speed_bps, *argv)) { +fprintf (stderr,"Invalid speed value\n" ); +return -1 ; + } +/*Convert Bps to bps*/ + speed_bps *=8 ; + }else if (strcmp (*argv,"handle" ) ==0 ) { + handle_present = true; +NEXT_ARG (); +if (strcmp (*argv,"scope" ) ==0 ) { +NEXT_ARG (); +if (strcmp (*argv,"netdev" ) ==0 ) { + handle_scope =NET_SHAPER_SCOPE_NETDEV ; + }else if (strcmp (*argv,"queue" ) ==0 ) { + handle_scope =NET_SHAPER_SCOPE_QUEUE ; + }else if (strcmp (*argv,"node" ) ==0 ) { + handle_scope =NET_SHAPER_SCOPE_NODE ; + }else { +fprintf (stderr,"Invalid scope\n" ); +return -1 ; + } + +NEXT_ARG (); +if (strcmp (*argv,"id" ) ==0 ) { +NEXT_ARG (); +if (get_unsigned (&handle_id, *argv,10 )) { +fprintf (stderr, +"Invalid handle id\n" ); +return -1 ; + } + } + } + }else { +fprintf (stderr,"What is \"%s\"\n" , *argv); +usage (); + } + argc--; + argv++; + } + +if (ifindex ==-1 ) +missarg ("dev" ); + +if (!handle_present) +missarg ("handle" ); + +if (cmd ==NET_SHAPER_CMD_SET && speed_bps ==0 ) +missarg ("speed" ); + +addattr32 (&req.n,sizeof (req),NET_SHAPER_A_IFINDEX , ifindex); + +struct rtattr *handle =addattr_nest (&req.n,sizeof (req), +NET_SHAPER_A_HANDLE |NLA_F_NESTED ); +addattr32 (&req.n,sizeof (req),NET_SHAPER_A_HANDLE_SCOPE , handle_scope); +addattr32 (&req.n,sizeof (req),NET_SHAPER_A_HANDLE_ID , handle_id); +addattr_nest_end (&req.n, handle); + +if (cmd ==NET_SHAPER_CMD_SET ) +addattr64 (&req.n,sizeof (req),NET_SHAPER_A_BW_MAX , speed_bps); + + err =rtnl_talk (&gen_rth, &req.n, &answer); +if (err <0 ) { +printf ("Kernel command failed: %d\n" , err); +return err; + } + +if (cmd ==NET_SHAPER_CMD_GET ) +print_netshaper_attrs (answer); + +return err; +} + +static int netshaper_parse_opt (struct link_util *lu,int argc,char **argv, +struct nlmsghdr *n) +{ +if (argc <1 ) +usage (); +if (strcmp (*argv,"help" ) ==0 ) +usage (); + +if (genl_init_handle (&gen_rth,NET_SHAPER_FAMILY_NAME , &genl_family)) +exit (1 ); + +if (strcmp (*argv,"set" ) ==0 ) +return do_cmd (argc -1 , argv +1 , n,NET_SHAPER_CMD_SET ); + +if (strcmp (*argv,"delete" ) ==0 ) +return do_cmd (argc -1 , argv +1 , n,NET_SHAPER_CMD_DELETE ); + +if (strcmp (*argv,"get" ) ==0 ) +return do_cmd (argc -1 , argv +1 , n,NET_SHAPER_CMD_GET ); +
ip/iplink_netshaper.c line 189
Quoted 10 lines
ip/iplink_netshaper.c new file +10
-0+fprintf (stderr, +"Command \"%s\" is unknown, try \"ip link netshaper help\".\n" , + *argv); +exit (-1 ); +} + +struct link_util netshaper_link_util = { + .id ="netshaper" , + .parse_opt = netshaper_parse_opt, +};
Please add a print_opt functionality.
Very useful to see what kernel thinks state is.
Quoted 4 lines · On Mon, 11 Aug 2025 00:05:02 -0700 Erni Sri Satya Vennela…
Add support for the netshaper Generic Netlink
family to iproute2. Introduce a new subcommand toip linkfor configuring netshaper parameters directly from userspace.
Quoted 8 lines
This interface allows users to set shaping attributes (such as speed) which are passed to the kernel to perform the corresponding netshaper operation.
Example usage:
$ip link netshaper { set | get | delete } dev DEVNAME \
handle scope SCOPE id ID \
[ speed SPEED ]
The choice of ip link is awkward and doesn't match other options. I can think of some better other choices:
1. netshaper could be a property of the device. But the choice of using genetlink instead of regular ip netlink attributes makes this hard.
2. netshaper could be part of devlink. Since it is more targeted at hardware device attributes.
3. netshaper could be a standalone command like bridge, dcb, devlink, rdma, tipc and vdpa.
What ever choice the command line options need to follow similar syntax to other iproute commands.
Jakub Kicinski
maintainerQuoted 19 lines · On Sat, 16 Aug 2025 15:55:10 -0700 Stephen Hemminger wrote:
Quoted 4 lines · On Mon, 11 Aug 2025 00:05:02 -0700 Erni Sri Satya Vennela…
Add support for the netshaper Generic Netlink
family to iproute2. Introduce a new subcommand toip linkfor configuring netshaper parameters directly from userspace.Quoted 8 lines
This interface allows users to set shaping attributes (such as speed) which are passed to the kernel to perform the corresponding netshaper operation.
Example usage:
$ip link netshaper { set | get | delete } dev DEVNAME \
handle scope SCOPE id ID \
[ speed SPEED ]The choice of ip link is awkward and doesn't match other options. I can think of some better other choices:
Quoted 7 lines
1. netshaper could be a property of the device. But the choice of using genetlink instead of regular ip netlink attributes makes this hard.
2. netshaper could be part of devlink. Since it is more targeted at hardware device attributes.
3. netshaper could be a standalone command like bridge, dcb, devlink, rdma, tipc and vdpa.What ever choice the command line options need to follow similar syntax to other iproute commands.
I think historically we gravitated towards option 3 -- each family has a command? But indeed we could fold it together with something like the netdev family without much issue, they are both key'd on netdevs.
Somewhat related -- what's your take on integrating / vendoring in YNL? mnl doesn't provide any extack support..
Quoted 1 line · Jakub Kicinski
Somewhat related -- what's your take on integrating / vendoring in YNL?
I feel like this has been brought up a few times.
Is there a specific proposal or any patches to review?
Jakub Kicinski
maintainerQuoted 6 lines · On Mon, 18 Aug 2025 09:41:29 -0600 David Ahern wrote:
Quoted 1 line · Jakub Kicinski
Somewhat related -- what's your take on integrating / vendoring in YNL?
I feel like this has been brought up a few times.
Is there a specific proposal or any patches to review?
Not AFAIK. Erni is being asked to rethink his approach here, and if we're going with a new command perhaps YNL should be on the table.
I'd be very interested to get a final ruling on YNL integration into iproute2 -- given its inability to work as a shared object / library it's not unreasonable for the answer to be "no".
The page pool sample in the kernel sources is very useful, I find myself copying to various systems during debug. If there's no clear path to YNL integration with iproute2 it's time for that sample to be come a real CLI tool.
Quoted 3 lines · Jakub Kicinski
I'd be very interested to get a final ruling on YNL integration into iproute2 -- given its inability to work as a shared object / library it's not unreasonable for the answer to be "no".
kind of hard to answer that without some proposal with details on what an integration of ynl would look like
Quoted 60 lines · Stephen Hemminger
Quoted 50 lines · On Mon, 11 Aug 2025 00:05:02 -0700 Erni Sri Satya Vennela…
Add support for the netshaper Generic Netlink
family to iproute2. Introduce a new subcommand toip linkfor configuring netshaper parameters directly from userspace.This interface allows users to set shaping attributes (such as speed) which are passed to the kernel to perform the corresponding netshaper operation.
Example usage:
$ip link netshaper { set | get | delete } dev DEVNAME \
handle scope SCOPE id ID \
[ speed SPEED ]Internally, this triggers a kernel call to apply the shaping configuration to the specified network device.
Currently, the tool supports the following functionalities:
- Setting speed in Mbps, enabling bandwidth clamping for a network device that support netshaper operations.
- Deleting the current configuration.
- Querying the existing configuration.Additional netshaper operations will be integrated into the tool as per requirement.
This change enables easy and scriptable configuration of bandwidth shaping for devices that use the netshaper Netlink family.
Corresponding net-next patches:
1) https://lore.kernel.org/all/[email protected]/
2) https://lore.kernel.org/lkml/[email protected]/Install pkg-config and libmnl* packages to print kernel extack errors to stdout.
Signed-off-by Erni Sri Satya Vennela <[email protected]>---
Please addinclude/uapi/linux/net_shaper.hfrom kernel source tree for this patch.
---
Changes in v3:
* Use strcmp instead of matches.
* Useget_rate64insteadget_unsignedfor speed parameter.
* Removespeed_mbpsindo_cmd()to reduce redundancy.
* Update the usage of speed parameter in the command. Changes in v2:
* Use color coding for printing device name in stdout.
* Use clang-format to format the code inline.
* Use__u64forspeed_bps.
* Removeinclude/uapi/linux/netshaper.hfile.Quoted 6 lines
--- ip/Makefile | 2 +- ip/iplink.c | 12 +++ ip/iplink_netshaper.c | 189 ++++++++++++++++++++++++++++++++++++++++++ 3 files changed, 202 insertions(+), 1 deletion(-) create mode 100644 ip/iplink_netshaper.c
Quoted 2 lines
No documentation.
No tests?
I will share the documentation on netshapers, specifically covering how to use the ip command for this use case.
Quoted 82 lines
Quoted 67 lines
ip/Makefile +1
-1@@ -4,7 +4,7 @@ IPOBJ=ip.o ipaddress.o ipaddrlabel.o iproute.o iprule.o ipnetns.o \ ipmaddr.o ipmonitor.o ipmroute.o ipprefix.o iptuntap.o iptoken.o \ ipxfrm.o xfrm_state.o xfrm_policy.o xfrm_monitor.o iplink_dummy.o \ iplink_ifb.o iplink_nlmon.o iplink_team.o iplink_vcan.o iplink_vxcan.o \ - iplink_vlan.o link_veth.o link_gre.o iplink_can.o iplink_xdp.o \ + iplink_vlan.o iplink_netshaper.o link_veth.o link_gre.o iplink_can.o iplink_xdp.o \ iplink_macvlan.o ipl2tp.o link_vti.o link_vti6.o link_xfrm.o \ iplink_vxlan.o tcp_metrics.o iplink_ipoib.o ipnetconf.o link_ip6tnl.o \ link_iptnl.o link_gre6.o iplink_bond.o iplink_bond_slave.o iplink_hsr.o \ip/iplink.c +12
-0@@ -1509,6 +1509,15 @@ static void do_help(int argc, char **argv)usage (); } +static int iplink_netshaper (int argc,char **argv) +{ +struct link_util *lu; + + lu =get_link_kind ("netshaper" ); + +return lu->parse_opt (lu, argc, argv,NULL ); +} +int do_iplink (int argc,char **argv) {if (argc <1 )@@ -1545,6 +1554,9 @@ int do_iplink(int argc, char **argv)if (matches (*argv,"property" ) ==0 )return iplink_prop (argc-1 , argv+1 ); +if (strcmp (*argv,"netshaper" ) ==0 ) +return iplink_netshaper (argc-1 , argv+1 ); +if (matches (*argv,"help" ) ==0 ) {do_help (argc-1 , argv+1 );return 0 ;ip/iplink_netshaper.c new file +16
-0@@ -0,0 +1,189 @@ +/* SPDX-License-Identifier: GPL-2.0 */ +/* +* iplink_netshaper.c netshaper H/W shaping support +* +* Authors: Erni Sri Satya Vennela <[email protected]> +*/ +#include <stdio.h> +#include <string.h> +#include <linux/genetlink.h> +#include <linux/netlink.h> +#include <linux/rtnetlink.h> +#include <uapi/linux/netshaper.h> +#include "utils.h" +#include "ip_common.h" +#include "libgenl.h" +ip/iplink_netshaper.c line 26
Quoted 10 lines
ip/iplink_netshaper.c new file +10
-0+/* netlink socket */ +static struct rtnl_handle gen_rth = { .fd =-1 }; +static int genl_family =-1 ; + +static void usage (void ) +{ +fprintf (stderr, +"Usage: ip link netshaper set dev DEVNAME handle scope HANDLE_SCOPE id HANDLE_ID speed SPEED\n" +" ip link netshaper delete dev DEVNAME handle scope HANDLE_SCOPE id HANDLE_ID\n" + " ip link netshaper get dev DEVNAME handle scope HANDLE_SCOPE id HANDLE_ID\n"Seems like netshaper is property of the device not a top level object, the syntax here is awkward. Open to better suggestions. Looks to me that netshaper should be property of device not top level object.
Quoted 4 lines
Or netshaper is like neighbour and route and really wants to be not part of link command.
Also no other ip commands use show not get.
The verb "get" is only used for things like getting a route or neighbor with a particular address.
Okay. I will change the code accordingly.
Quoted 161 lines
Quoted 161 lines
ip/iplink_netshaper.c new file +159
-0+"Where: DEVNAME := STRING\n" +" HANDLE_SCOPE := { netdev | queue | node }\n" +" HANDLE_ID := UINT\n" + " SPEED := UINT{ kbit | mbit | gbit }\n"); + +exit (-1 ); +} + +static void print_netshaper_attrs (struct nlmsghdr *answer) +{ +struct genlmsghdr *ghdr =NLMSG_DATA (answer); +int len = answer->nlmsg_len -NLMSG_LENGTH (GENL_HDRLEN ); +struct rtattr *tb[NET_SHAPER_A_MAX +1 ] = {}; +__u32 speed_mbps; +__u64 speed_bps; +int ifindex; + +parse_rtattr (tb,NET_SHAPER_A_MAX , + (struct rtattr *)((char *)ghdr +GENL_HDRLEN ), len); + +for (int i =1 ; i <=NET_SHAPER_A_MAX ; ++i) { +if (!tb[i]) +continue ; +switch (i) { +case NET_SHAPER_A_BW_MAX : + speed_bps =rta_getattr_u64 (tb[i]); + speed_mbps = (speed_bps /1000000 ); +print_uint (PRINT_ANY ,"speed" ,"speed: %u mbps\n" , + speed_mbps); +break ; +case NET_SHAPER_A_IFINDEX : + ifindex =rta_getattr_u32 (tb[i]); +print_color_string (PRINT_ANY ,COLOR_IFNAME ,"dev" , +"dev: %s\n" , +ll_index_to_name (ifindex)); +break ; +default : +break ; + } + } +} + +static int do_cmd (int argc,char **argv,struct nlmsghdr *n,int cmd) +{ +GENL_REQUEST (req,1024 , genl_family,0 ,NET_SHAPER_FAMILY_VERSION , cmd, +NLM_F_REQUEST |NLM_F_ACK ); + +struct nlmsghdr *answer; +__u64 speed_bps =0 ; +int ifindex =-1 ; +int handle_scope =NET_SHAPER_SCOPE_UNSPEC ; +__u32 handle_id =0 ; +bool handle_present = false; +int err; + +while (argc >0 ) { +if (strcmp (*argv,"dev" ) ==0 ) { +NEXT_ARG (); + ifindex =ll_name_to_index (*argv); + }else if (strcmp (*argv,"speed" ) ==0 ) { +NEXT_ARG (); +if (get_rate64 (&speed_bps, *argv)) { +fprintf (stderr,"Invalid speed value\n" ); +return -1 ; + } +/*Convert Bps to bps*/ + speed_bps *=8 ; + }else if (strcmp (*argv,"handle" ) ==0 ) { + handle_present = true; +NEXT_ARG (); +if (strcmp (*argv,"scope" ) ==0 ) { +NEXT_ARG (); +if (strcmp (*argv,"netdev" ) ==0 ) { + handle_scope =NET_SHAPER_SCOPE_NETDEV ; + }else if (strcmp (*argv,"queue" ) ==0 ) { + handle_scope =NET_SHAPER_SCOPE_QUEUE ; + }else if (strcmp (*argv,"node" ) ==0 ) { + handle_scope =NET_SHAPER_SCOPE_NODE ; + }else { +fprintf (stderr,"Invalid scope\n" ); +return -1 ; + } + +NEXT_ARG (); +if (strcmp (*argv,"id" ) ==0 ) { +NEXT_ARG (); +if (get_unsigned (&handle_id, *argv,10 )) { +fprintf (stderr, +"Invalid handle id\n" ); +return -1 ; + } + } + } + }else { +fprintf (stderr,"What is \"%s\"\n" , *argv); +usage (); + } + argc--; + argv++; + } + +if (ifindex ==-1 ) +missarg ("dev" ); + +if (!handle_present) +missarg ("handle" ); + +if (cmd ==NET_SHAPER_CMD_SET && speed_bps ==0 ) +missarg ("speed" ); + +addattr32 (&req.n,sizeof (req),NET_SHAPER_A_IFINDEX , ifindex); + +struct rtattr *handle =addattr_nest (&req.n,sizeof (req), +NET_SHAPER_A_HANDLE |NLA_F_NESTED ); +addattr32 (&req.n,sizeof (req),NET_SHAPER_A_HANDLE_SCOPE , handle_scope); +addattr32 (&req.n,sizeof (req),NET_SHAPER_A_HANDLE_ID , handle_id); +addattr_nest_end (&req.n, handle); + +if (cmd ==NET_SHAPER_CMD_SET ) +addattr64 (&req.n,sizeof (req),NET_SHAPER_A_BW_MAX , speed_bps); + + err =rtnl_talk (&gen_rth, &req.n, &answer); +if (err <0 ) { +printf ("Kernel command failed: %d\n" , err); +return err; + } + +if (cmd ==NET_SHAPER_CMD_GET ) +print_netshaper_attrs (answer); + +return err; +} + +static int netshaper_parse_opt (struct link_util *lu,int argc,char **argv, +struct nlmsghdr *n) +{ +if (argc <1 ) +usage (); +if (strcmp (*argv,"help" ) ==0 ) +usage (); + +if (genl_init_handle (&gen_rth,NET_SHAPER_FAMILY_NAME , &genl_family)) +exit (1 ); + +if (strcmp (*argv,"set" ) ==0 ) +return do_cmd (argc -1 , argv +1 , n,NET_SHAPER_CMD_SET ); + +if (strcmp (*argv,"delete" ) ==0 ) +return do_cmd (argc -1 , argv +1 , n,NET_SHAPER_CMD_DELETE ); + +if (strcmp (*argv,"get" ) ==0 ) +return do_cmd (argc -1 , argv +1 , n,NET_SHAPER_CMD_GET ); + +fprintf (stderr, +"Command \"%s\" is unknown, try \"ip link netshaper help\".\n" , + *argv); +exit (-1 ); +} +
Quoted 8 lines
ip/iplink_netshaper.c line 189
Quoted 4 lines
ip/iplink_netshaper.c new file +4
-0+struct link_util netshaper_link_util = { + .id ="netshaper" , + .parse_opt = netshaper_parse_opt, +};Please add a
print_optfunctionality.
Very useful to see what kernel thinks state is.
Thankyou for the clarification. I'll make sure to integrate these changes in the next version.
Quoted 20 lines · Stephen Hemminger
Quoted 4 lines · On Mon, 11 Aug 2025 00:05:02 -0700 Erni Sri Satya Vennela…
Add support for the netshaper Generic Netlink
family to iproute2. Introduce a new subcommand toip linkfor configuring netshaper parameters directly from userspace.Quoted 8 lines
This interface allows users to set shaping attributes (such as speed) which are passed to the kernel to perform the corresponding netshaper operation.
Example usage:
$ip link netshaper { set | get | delete } dev DEVNAME \
handle scope SCOPE id ID \
[ speed SPEED ]The choice of ip link is awkward and doesn't match other options. I can think of some better other choices:
Quoted 7 lines
1. netshaper could be a property of the device. But the choice of using genetlink instead of regular ip netlink attributes makes this hard.
2. netshaper could be part of devlink. Since it is more targeted at hardware device attributes.
3. netshaper could be a standalone command like bridge, dcb, devlink, rdma, tipc and vdpa.What ever choice the command line options need to follow similar syntax to other iproute commands.
Thankyou for the suggestions.
After internal discussions, I have decided to proceed with implementing netshaper as a standalone command.
Quoted 26 lines · Jakub Kicinski
Quoted 25 lines · On Sat, 16 Aug 2025 15:55:10 -0700 Stephen Hemminger wrote:
Quoted 4 lines · On Mon, 11 Aug 2025 00:05:02 -0700 Erni Sri Satya Vennela…
Add support for the netshaper Generic Netlink
family to iproute2. Introduce a new subcommand toip linkfor configuring netshaper parameters directly from userspace.Quoted 8 lines
This interface allows users to set shaping attributes (such as speed) which are passed to the kernel to perform the corresponding netshaper operation.
Example usage:
$ip link netshaper { set | get | delete } dev DEVNAME \
handle scope SCOPE id ID \
[ speed SPEED ]The choice of ip link is awkward and doesn't match other options. I can think of some better other choices:
1. netshaper could be a property of the device. But the choice of using genetlink instead of regular ip netlink attributes makes this hard.
2. netshaper could be part of devlink. Since it is more targeted at hardware device attributes.
3. netshaper could be a standalone command like bridge, dcb, devlink, rdma, tipc and vdpa.
Quoted 8 lines
Quoted 1 line
What ever choice the command line options need to follow similar syntax to other iproute commands.
I think historically we gravitated towards option 3 -- each family has a command? But indeed we could fold it together with something like the netdev family without much issue, they are both key'd on netdevs.
Somewhat related -- what's your take on integrating / vendoring in YNL? mnl doesn't provide any extack support..
I have done some tests and found that if we install pkg-config and libmnl packages beforehand. The extack error messages from the kernel are being printed to the stdout.
Example 1:
$ sudo ip link netshaper set dev <interface> handle scope queue id 1 speed 15gbit
Error: mana: net shaper scope should be netdev.
Kernel command failed: -1
Example 2:
$ sudo ip link netshaper set dev <interface> handle scope netdev id 2 speed 15gbit
Error: mana: Cannot create multiple shapers.
Kernel command failed: -1
Quoted 11 lines · Jakub Kicinski
Quoted 6 lines · On Mon, 18 Aug 2025 09:41:29 -0600 David Ahern wrote:
Quoted 1 line · Jakub Kicinski
Somewhat related -- what's your take on integrating / vendoring in YNL?
I feel like this has been brought up a few times.
Is there a specific proposal or any patches to review?
Not AFAIK. Erni is being asked to rethink his approach here, and if we're going with a new command perhaps YNL should be on the table.
Thank you for the feedback and for suggesting integration with the ynl tool.
For the current implementation, I have followed your earlier guidance ( while implementing bandwidth clamping for MANA interface ) and used netlink commands similar to those used by ynl to support netshaper operations from the kernel.
Ref: https://lore.kernel.org/all/[email protected]/
Quoted 3 lines
I'd be very interested to get a final ruling on YNL integration into iproute2 -- given its inability to work as a shared object / library it's not unreasonable for the answer to be "no".
This approach, to use netshaper as separate command in iproute2, seemed like the most practical path forward to address shaping support without introducing additional complexity.
Quoted 5 lines
The page pool sample in the kernel sources is very useful, I find myself copying to various systems during debug. If there's no clear path to YNL integration with iproute2 it's time for that sample to be come a real CLI tool.
Jakub Kicinski
maintainerQuoted 6 lines · On Thu, 21 Aug 2025 04:06:07 -0700 Erni Sri Satya Vennela wrote:
Quoted 2 lines
Somewhat related -- what's your take on integrating / vendoring in YNL? mnl doesn't provide any extack support..
I have done some tests and found that if we install pkg-config and libmnl packages beforehand. The extack error messages from the kernel are being printed to the stdout.
Sorry, I wasn't very precise, it supports printing the string messages. But nothing that requires actually understanding the message. No bad attribute errors, no missing attribute errors, no policy errors.
Quoted 11 lines · Jakub Kicinski
Quoted 6 lines · On Thu, 21 Aug 2025 04:06:07 -0700 Erni Sri Satya Vennela wrote:
Quoted 2 lines
Somewhat related -- what's your take on integrating / vendoring in YNL? mnl doesn't provide any extack support..
I have done some tests and found that if we install pkg-config and libmnl packages beforehand. The extack error messages from the kernel are being printed to the stdout.
Sorry, I wasn't very precise, it supports printing the string messages. But nothing that requires actually understanding the message. No bad attribute errors, no missing attribute errors, no policy errors.
Are you referring to the following error logs from the ynl tool?
$./tools/net/ynl/pyynl/cli.py
--spec Documentation/netlink/specs/net_shaper.yaml --do set
--json '{"ifindex":'3',
"handle":{"scope": "netdev", "id":'1' },
"bw-max": 200001000 }'Netlink error: Invalid argumentnl_len = 92 (76) nl_flags = 0x300 nl_type = 2
error: -22
extack: {'msg': 'mana: Please use multiples of 100Mbps for
bandwidth'}
If yes, would it be reasonable to add support in iproute2 itself for displaying such error logs?
Quoted 22 lines · On Sun, 24 Aug 2025 06:40:17 -0700 Erni Sri Satya Vennela…
Quoted 11 lines · Jakub Kicinski
Quoted 6 lines · On Thu, 21 Aug 2025 04:06:07 -0700 Erni Sri Satya Vennela wrote:
Quoted 2 lines
Somewhat related -- what's your take on integrating / vendoring in YNL? mnl doesn't provide any extack support..
I have done some tests and found that if we install pkg-config and libmnl packages beforehand. The extack error messages from the kernel are being printed to the stdout.
Sorry, I wasn't very precise, it supports printing the string messages. But nothing that requires actually understanding the message. No bad attribute errors, no missing attribute errors, no policy errors.
Are you referring to the following error logs from the ynl tool?
$./tools/net/ynl/pyynl/cli.py
--specDocumentation/netlink/specs/net_shaper.yaml--do set
--json '{"ifindex":'3',"handle":{"scope": "netdev", "id":'1' }, "bw-max": 200001000 }'
Quoted 8 lines
Netlink error: Invalid argument
nl_len= 92 (76)nl_flags=0x300nl_type= 2
error: -22
extack: {'msg': 'mana: Please use multiples of 100Mbps for
bandwidth'}If yes, would it be reasonable to add support in iproute2 itself for displaying such error logs?
other parts of iproute2 already use libmnl to display extack messages.
Jakub Kicinski
maintainerQuoted 10 lines · On Sun, 24 Aug 2025 06:40:17 -0700 Erni Sri Satya Vennela wrote:
Quoted 3 lines
Sorry, I wasn't very precise, it supports printing the string messages. But nothing that requires actually understanding the message. No bad attribute errors, no missing attribute errors, no policy errors.
Are you referring to the following error logs from the ynl tool?
$./tools/net/ynl/pyynl/cli.py
--specDocumentation/netlink/specs/net_shaper.yaml--do set
--json '{"ifindex":'3',
Quoted 8 lines
"handle":{"scope": "netdev", "id":'1' }, "bw-max": 200001000 }'Netlink error: Invalid argument
nl_len= 92 (76)nl_flags=0x300nl_type= 2
error: -22
extack: {'msg': 'mana: Please use multiples of 100Mbps for
bandwidth'}
Not 'msg', the other types, for example miss-type, bad-attr:
$ ynl --family net_shaper --do get
Netlink error: Invalid argument
nl_len = 44 (28) nl_flags = 0x300 nl_type = 2
error: -22
extack: {'miss-type': 'ifindex', 'miss-type-doc': 'Interface index owning the specified shaper.'}$ ynl --family net_shaper --do get --json '{"ifindex": 1}'
Netlink error: Operation not supported
nl_len = 44 (28) nl_flags = 0x300 nl_type = 2
error: -95
extack: {'bad-attr': '.ifindex'}Quoted 2 lines
If yes, would it be reasonable to add support in iproute2 itself for displaying such error logs?
Quoted 27 lines · On Thu, 21 Aug 2025 04:06:07 -0700 Erni Sri Satya Vennela…
Quoted 26 lines · Jakub Kicinski
Quoted 25 lines · On Sat, 16 Aug 2025 15:55:10 -0700 Stephen Hemminger wrote:
Quoted 4 lines · On Mon, 11 Aug 2025 00:05:02 -0700 Erni Sri Satya Vennela…
Add support for the netshaper Generic Netlink
family to iproute2. Introduce a new subcommand toip linkfor configuring netshaper parameters directly from userspace.Quoted 8 lines
This interface allows users to set shaping attributes (such as speed) which are passed to the kernel to perform the corresponding netshaper operation.
Example usage:
$ip link netshaper { set | get | delete } dev DEVNAME \
handle scope SCOPE id ID \
[ speed SPEED ]The choice of ip link is awkward and doesn't match other options. I can think of some better other choices:
1. netshaper could be a property of the device. But the choice of using genetlink instead of regular ip netlink attributes makes this hard.
2. netshaper could be part of devlink. Since it is more targeted at hardware device attributes.
3. netshaper could be a standalone command like bridge, dcb, devlink, rdma, tipc and vdpa.
Quoted 8 lines
Quoted 8 lines
Quoted 1 line
What ever choice the command line options need to follow similar syntax to other iproute commands.
I think historically we gravitated towards option 3 -- each family has a command? But indeed we could fold it together with something like the netdev family without much issue, they are both key'd on netdevs.
Somewhat related -- what's your take on integrating / vendoring in YNL? mnl doesn't provide any extack support..
No YNL integration with iproute adds too much.
Quoted 32 lines · Stephen Hemminger
Quoted 29 lines · On Thu, 21 Aug 2025 04:06:07 -0700 Erni Sri Satya Vennela…
Quoted 28 lines · Jakub Kicinski
Quoted 26 lines · On Sat, 16 Aug 2025 15:55:10 -0700 Stephen Hemminger wrote:
Quoted 4 lines · On Mon, 11 Aug 2025 00:05:02 -0700 Erni Sri Satya Vennela…
Add support for the netshaper Generic Netlink
family to iproute2. Introduce a new subcommand toip linkfor configuring netshaper parameters directly from userspace.Quoted 8 lines
This interface allows users to set shaping attributes (such as speed) which are passed to the kernel to perform the corresponding netshaper operation.
Example usage:
$ip link netshaper { set | get | delete } dev DEVNAME \
handle scope SCOPE id ID \
[ speed SPEED ]The choice of ip link is awkward and doesn't match other options. I can think of some better other choices:
1. netshaper could be a property of the device. But the choice of using genetlink instead of regular ip netlink attributes makes this hard.
2. netshaper could be part of devlink. Since it is more targeted at hardware device attributes.
3. netshaper could be a standalone command like bridge, dcb, devlink, rdma, tipc and vdpa.What ever choice the command line options need to follow similar syntax to other iproute commands.
Quoted 8 lines
Quoted 6 lines
Quoted 6 lines
I think historically we gravitated towards option 3 -- each family has a command? But indeed we could fold it together with something like the netdev family without much issue, they are both key'd on netdevs.
Somewhat related -- what's your take on integrating / vendoring in YNL? mnl doesn't provide any extack support..
No YNL integration with iproute adds too much.
Thankyou Stephen and Jakub for the feedback.
After reviewing the discussion, I will proceed with the next version of the patch with netshaper as a standalone command.
I appreciate the perspectives shared and will keep YNL in mind for future work.