| Age | Commit message (Collapse) | Author |
|---|
|
This reverts commit ab30509b25d54dac99294b76ba03fd49c3d2c946.
As in T6152 there seem to be some NICs that have a non working implementation
of reading the EEE registers. Remove this feature in the meantime until there
is a less exploding solution hindering boards to boot.
Return to Kernel defaults by removing this code path.
|
|
Typo (missaligned -/_) in the code causes hw-tc-offload to never be enabled in
the underlaying hardware via ethtool.
|
|
|
|
VyOS is a routing (packet pushing) platform, thus supporting EEE which
potentially causes issues is not a good idea. Some recent Intel drivers enable
EEE by default, thus we will disable this for every NIC supporting EEE.
|
|
If ethernet interface is a bond memeber:
1. Allow for changing only specific parameters which are specified
in EthernetIf.get_bond_member_allowed_options function.
2. Added inheritable parameters from bond interface to ethernet
interface which are scpecified
in BondIf.get_inherit_bond_options.
Users can change inheritable options under ethernet interface
but in commit it will be copied from bond interface.
3. All other parameters are denied for changing.
Added migration script. It deletes all denied parameters under
ethernet interface if it is a bond member.
|
|
supports HW flowtable offload
- Add required offload setting for interfaces + flowtable offload (hw-tc-offload)
- Verification of interface support for hardware offloaded flowtables
|
|
|
|
* T5195: move run, cmd, call, rc_cmd helper to vyos.utils.process
* T5195: use read_file and write_file implementation from vyos.utils.file
Changed code automatically using:
find . -type f -not -path '*/\.*' -exec sed -i 's/^from vyos.util import read_file$/from vyos.utils.file import read_file/g' {} +
find . -type f -not -path '*/\.*' -exec sed -i 's/^from vyos.util import write_file$/from vyos.utils.file import write_file/g' {} +
* T5195: move chmod* helpers to vyos.utils.permission
* T5195: use colon_separated_to_dict from vyos.utils.dict
* T5195: move is_systemd_service_* to vyos.utils.process
* T5195: fix boot issues with missing imports
* T5195: move dict_search_* helpers to vyos.utils.dict
* T5195: move network helpers to vyos.utils.network
* T5195: move commit_* helpers to vyos.utils.commit
* T5195: move user I/O helpers to vyos.utils.io
|
|
There is no need for the backend code to call ethtool and try to change speed or
duplex settings every time there is a change in the interface configuration,
but no change for the speed/duplex subnodes. This also makes the commit itself
faster when working with ethernet interfaces.
Bonus: no repeating CLI messages that the driver does not support speed/duplex
changes, as we do not change anything here.
Extension to commit f2ecc9710 ("ethernet: T3891: honor auto-negotiation support
per NIC")
|
|
Not all drivers/NICs or combination of NIC + transceiver support auto-
negotiation. The current auto-negotiation capability is evaluated and taken
into account when applying spped/duplex settings. If auto-negotiation is
not supported - we skip the setting to avoid errors during configuration.
|
|
Fix ValueError: Unknown format code 'x' for object of type 'str' added in
commit c0ffb8ba4e663 ("ethernet: rps: T4928: adjust to Kernel ABI changes").
|
|
In the past we could simply set all bits for all CPUs even if they did not
exist. With 6.1.y Kernel series this is no longer possible and the input data
is validated against the available number of CPUs.
|
|
The initial implementation from commit ac4e07f9 ("rfs: T4689: Support RFS
(Receive Flow Steering)") always adjusted the global rps_sock_flow_entries
configuration. So if RFS was enabled for one NIC but not the other - it did not
work.
According to the documentation:
RFS is only available if the kconfig symbol CONFIG_RPS is enabled (on by
default for SMP). The functionality remains disabled until explicitly
configured. The number of entries in the global flow table is set through:
/proc/sys/net/core/rps_sock_flow_entries
The number of entries in the per-queue flow table are set through:
/sys/class/net/<dev>/queues/rx-<n>/rps_flow_cnt
Both of these need to be set before RFS is enabled for a receive queue. Values
for both are rounded up to the nearest power of two. The suggested flow count
depends on the expected number of active connections at any given time, which
may be significantly less than the number of open connections. We have found
that a value of 32768 for rps_sock_flow_entries works fairly well on a
moderately loaded server.
This commit sets rps_sock_flow_entries via sysctl on bootup leafing the RFS
configuration to the interface level.
|
|
The initial implementation in commit 9fb9e5cade ("ethernet: T3171: add CLI
option to enable RPS (Receive Packet Steering)" only changed the CPU affinity
for RX queue 0.
This commit takes all RX queues into account.
|
|
|
|
|
|
This reverts commit 53355271a2864d844daca89a064c21e514e10adb.
|
|
|
|
Commit 31169fa8a763e ("vyos.ifconfig: T3619: only set offloading options if
supported by NIC") added the new implementation which handles NIC offloading.
Unfortunately every single implementation was copied from "gro" which resulted
in a change to gro for each offloading option - thus options like lro, sg, tso
had no effect at all.
It all comes down to copy/paste errors ... one way or another.
|
|
Certain NICs seem to fail to report that they don't support speed/duplex setting,
so they look as if it's supported, but the command fails in practice.
It's probably better to preserve a working config in that case.
|
|
Interface will still be visible to the operating system.
|
|
states
Turns out an AX88179 USB 3.0 NIC does not support reading back the speed and
duplex settings in every operating state. While the NIC is beeing
initialized, reading the speed setting will return:
$ cat /sys/class/net/eth6/speed
cat: /sys/class/net/eth6/speed: Invalid argument
Thus if this happens, we simply tell the system that the current NIC speed
matches the requested speed and nothing is changed at this point in time.
(cherry picked from commit e2b7e1766cc22c5cd718a5001be6336bdca92eec)
|
|
Move the two implementations to get the driver name of a NIC from ethernet.py
and ethtool.py to only ethtool.py.
(cherry picked from commit 07840977834816b69fa3b366817d90f44b5dc7a7)
|
|
(cherry picked from commit 1572edd2cef355710d1129907d3e49451a6c31d4)
|
|
It makes no sense to have a parser for the ethtool values in ethtool.py
and ethernet.py - one instance ios more then enough!
|
|
Only update the RX/TX ring-buffer settings if they are different from the ones
currently programmed to the hardware. There is no need to write the same value
to the hardware again - this could cause traffic disruption on some NICs.
|
|
It makes no sense to have a parser for the ethtool value sin ethtool.py
and ethernet.py - one instance ios more then enough!
|
|
This makes understanding the code easier what is "really" called without
opening the man page.
|
|
Deprecated in the Linux Kernel by commit 08a00fea6de277df12ccfadc21 ("net:
Remove references to NETIF_F_UFO from ethtool.").
(cherry picked from commit f5e46ee6cc2b6c1c1869e26beca4ccd5bf52b62f)
|
|
option
Commit 31169fa8 ("vyos.ifconfig: T3619: only set offloading options if
supported by NIC") added a warning for the user if an offload option was about
to change that was not possible at all (harware limit).
Unfortunately the warning was even displayed if nothing was done at all. This
got corrected.
|
|
In the past we always told ethtool to change the offloading settings, even if
this was not supported by the underlaying driver.
This commit will only change the offloading options if they differ from the
current state of the NIC and only if it's supported by the NIC. If the NIC does
not support setting the offloading options, a message will be displayed
for the user:
vyos@vyos# set interfaces ethernet eth2 offload gro
vyos@vyos# commit
[ interfaces ethernet eth2 ]
Adapter does not support changing large-receive-offload settings!
|
|
|
|
When a VIF/VLAN interface is placed in admin down state but the lower
interface, serving the vlan, is moved from admin down -> admin up, all its
vlan interfaces will be placed in admin up state, too.
This is bad as a VLAN interface will become admin up even if its specified as
admin down after a reboot.
To reproduce:
set interfaces ethernet eth1 vif 20 disable
set interfaces ethernet eth1 disable
commit
delete interfaces ethernet eth1 disable
commit
Now check the interface state and it returns UP,LOWER_UP
7: eth1.20@eth1: <BROADCAST,MULTICAST,UP,LOWER_UP> mtu 1500 qdisc noqueue state UP group default qlen 1000
link/ether 00:50:56:b3:09:07 brd ff:ff:ff:ff:ff:ff
inet6 fe80::250:56ff:feb3:907/64 scope link
valid_lft forever preferred_lft forever
|
|
Interface.get_config() was always a pure helper which exposed a "per interface
type" dictionary which was then fed by the caller to create interfaces by
iproute2 which required additional options during creation time.
Such interfaces had been:
* tunnel
* vxlan
* geneve
* macsec
* wifi
* macvlan / pseudo-ethernet
The code was always duplicated to convert from the VyOS CLI based get_config_dict()
to a dict which can be used to feed iproute2.
This path has been removed and we now always feed in the entire dictionary
retrieved by get_config_dict() or in the interfaces case, it's high-level wrapper
get_interface_dict() to the interface we wan't to create.
This also adds the - personally long awaited - possibility to get rid of the
derived tunnel classes for e.g. GRE, IPIP, IPIP6 and so on.
|
|
The Linux Kernel supports enabling more cores for RPS then we actually have.
It does internal clipping/validation so there is no need for us to calculate
the specifc enable mask we can simply throw "all -1" at the Kernel.
|
|
|
|
set interfaces ethernet <interface> offload rps
|
|
|
|
|
|
Migrate from
ethernet eth1 {
offload-options {
generic-receive on
generic-segmentation on
scatter-gather on
tcp-segmentation on
udp-fragmentation on
}
}
to
ethernet eth1 {
offload {
ufo
tso
sg
gso
gro
}
}
|
|
|
|
|
|
Using 'xdp' will automatically decide if the driver supports 'xdpdrv' or only
'xdpgeneric'. A user later sees which driver is actually in use by calling
'ip a' or 'show interfaces ethernet'.
|
|
The CLI command 'set interfaces ethernet <interface> offload-options xdp" enables
the XDP generic mode on the given interface.
vyos@vyos:~$ show interfaces ethernet eth1
eth1: <NO-CARRIER,BROADCAST,MULTICAST,UP> mtu 1500 xdpgeneric/id:151 qdisc mq state DOWN group default qlen 1000
link/ether 00:50:56:bf:ef:aa brd ff:ff:ff:ff:ff:ff
inet6 fe80::250:56ff:febf:efaa/64 scope link tentative
valid_lft forever preferred_lft forever
Description: fooa
XDP code is thankfully copied from [1], thank you for this nice tutorial.
NOTE: this is an experimental feature which might break your
forwarding/filtering.
[1]: https://medium.com/swlh/building-a-xdp-express-data-path-based-peering-router-20db4995da66
|
|
Renamed using snippet below:
----------------------------
for file in $(find . -name "*.py")
do
sed -i "s/vyos_dict_search/dict_search/" $file
done
|
|
Instead of using an Adapter pattern to make interfaces VLAN-aware, create a
derived class named VLANIf to represent a VLAN. This change was necessary to
eliminate mixed code in Interfaces class which was VLAN - free, but recently
gained some VLAN specific code for set_admin_state().
In addition this "autoresolves" the issue in T2894 as a bond vlan interface
will no longer change the lower interface.
|
|
|
|
|
|
This is required as other interfaces (e.g. pseudo-ethernet or bond) will have
VLANs, too.
|
|
Every derived class must implement update() to set the interfaces admin
up/down state. This is required to prevend extensive link flaps when e.g.
reconfiguring bond interfaces.
|
