upstream/opnsense-src
1
0
Fork 0
mirror of https://github.com/opnsense/src.git synced 2026-02-18 18:20:26 -05:00
Code Issues Projects Releases Packages Wiki Activity Actions

vtnet: disable hardware TCP LRO by default

Browse source 
Hardware TCP LRO results in problems in settings with IP forwarding
being enabled. In case of nodes without IP forwarding, using
software LRO is also beneficial in general, since it can provide better
information about what was received on the wire.
Therefore, disable hardware TCP LRO by default.
By tuning the loader tunable, this can be changed.

PR:			263229
Reviewed by:		Timo Völker
Differential Revision:	https://reviews.freebsd.org/D52684

(cherry picked from commit 6e4b811009d63f33c59d51f28fd4a030ca90843e)
This commit is contained in:
Michael Tuexen 2025-09-30 01:58:37 +02:00 committed by Franco Fichtner
parent b0b3245d3d
commit c7cd4884ca
2 changed files with 18 additions and 6 deletions
Show stats Download patch file Download diff file Expand all files Collapse all files

22
share/man/man4/vtnet.4
View file

@ -22,7 +22,7 @@
.\" OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
.\" SUCH DAMAGE.
.\"
.Dd September 26, 2025
.Dd September 30, 2025
.Dt VTNET 4
.Os
.Sh NAME
@ -54,6 +54,18 @@ TCP segmentation offload (TSO), TCP large receive offload (LRO),
hardware VLAN tag stripping/insertion features, a multicast hash filter,
as well as Jumbo Frames (up to 9216 bytes), which can be
configured via the interface MTU setting.
.Pp
Two flavors of TCP LRO are supported:
Hardware TCP LRO, which is performed by the host providing TCP segments larger
than the MTU to the guest, and Software TCP LRO, which is performed by the
network stack of the guest processing TCP segments in an optimized way.
Only one flavour of TCP LRO should be used.
Since hardware TCP LRO can have bad interactions with IP forwarding and
software TCP LRO mitigates several drawbacks of hardware TCP LRO, the default
setting is to disable hardware TCP LRO.
See the loader tunable
.Va hw.vtnet. Ns Ar X Ns Va .lro_disable .
.Pp
TCP/UDP receive checksum offload cannot be configured independently for IPv4
and IPv6.
Selecting an MTU larger than 1500 bytes with the
@ -94,8 +106,8 @@ This tunable disables TCP segmentation offloading.
The default value is 0.
.It Va hw.vtnet.lro_disable
.It Va hw.vtnet. Ns Ar X Ns Va .lro_disable
This tunable disables large receive offload.
The default value is 0.
This tunable disables hardware TCP LRO.
The default value is 1.
.It Va hw.vtnet.mq_disable
.It Va hw.vtnet. Ns Ar X Ns Va .mq_disable
This tunable disables multiqueue.
@ -119,11 +131,11 @@ This tunable sets the number of RX segments processed in one pass.
The default value is 1024.
.It Va hw.vtnet.lro_entry_count
.It Va hw.vtnet. Ns Ar X Ns Va .lro_entry_count
This tunable sets the software LRO entry count.
This tunable sets the software TCP LRO entry count.
The default value is 128, the minimum value is 8.
.It Va hw.vtnet.lro_mbufq_depth
.It Va hw.vtnet. Ns Ar X Ns Va .lro_mbufq_depth
This tunable sets the depth of the software LRO mbuf queue.
This tunable sets the depth of the software TCP LRO mbuf queue.
The default value is 0.
.El
.Sh TRANSMIT QUEUE STATISTICS

2
sys/dev/virtio/network/if_vtnet.c
View file

@ -282,7 +282,7 @@ static int vtnet_tso_disable = 0;
SYSCTL_INT(_hw_vtnet, OID_AUTO, tso_disable, CTLFLAG_RDTUN,
&vtnet_tso_disable, 0, "Disables TSO");
static int vtnet_lro_disable = 0;
static int vtnet_lro_disable = 1;
SYSCTL_INT(_hw_vtnet, OID_AUTO, lro_disable, CTLFLAG_RDTUN,
&vtnet_lro_disable, 0, "Disables hardware LRO");