| Age | Commit message (Collapse) | Author |
|---|
|
(cherry picked from commit 2c9b0e0641b24dc79320282eff358079b3c7c7c3)
|
|
This reverts commit cf51e90bfc2d1c3b58192a726e786b598681fb71.
Conflicts:
templates/traffic-policy/limiter/node.tag/class/node.tag/precedence/node.def
|
|
Bug 5889
Deleting input interface not allowed if target of redirect
|
|
|
|
|
|
For simplicity only allow mirror/redirect templates
to exist on ethernet device.
|
|
Priority for default rule makes no sense. The default is always
after all other rules.
|
|
For queue-type (other than fair-queue), it is allowable to have larger queue size.
Move validation into class checking.
|
|
Add spaces and use u32: format
|
|
This allows use of IANA (/etc/services) values for port names:
example:
traffic-policy shaper SS {
class 2 {
match SMTP ip destination port smtp
}
}
|
|
Replace all usage in this package of comp_help with new val_help
|
|
And fix some typos
|
|
Allow using common values supported by TC filter.
|
|
Bug 5849
|
|
Since queue limit of zero causes all packets to be dropped;
require user to configure a reasonable value
Bug 5842
|
|
There is no way for firewall mark to be useful in limiter policy
since limiter is only allowed on ingress. And ingress takes place
before firewall rules are applied.
|
|
Bug 5407
This fixes the confusing usage of priority field in both shaper
and limiter class. In limiter, it is renamed to precedence since
it controls the order of match rule evaluation.
|
|
Bug 5799 - Add matching on firewall mark values
Bug 5795 - Block match rules that have multiple underlying filters
|
|
Bug 5783
|
|
The name ip exists in /etc/protocols as a pseudonym for all IP protocols.
But the Qos match logic doesn't support it so do not allow the user to
choose that value
Bug 5689
|
|
This adds:
set traffic-policy limiter TL default bandwidth 1mbit
|
|
Current (Kenwood and earlier):
set qos-policy traffic-shaper TS { ...classes }
set qos-policy traffic-limiter TL { ...classes }
seq qos-policy network-emulator NE ...
set qos-policy random-detect RD ...
set qos-policy rate-limiter RC ...
set qos-policy round-robin RR ...
set interfaces ethernet eth0 qos-policy out TS
set interfaces ethernet eth0 qos-policy in TL
New (Larkspur and later):
set traffic-policy shaper TS { ...classes }
set traffic-policy limiter TL { ...classes }
seq traffic-policy network-emulator NE ...
set traffic-policy random-detect RD ...
set traffic-policy rate-control RC ...
set traffic-policy round-robin RR ...
set interfaces ethernet eth0 traffic-policy out TS
set interfaces ethernet eth0 traffic-policy in TL
set interfaces ethernet eth0 redirect ifb0
set interfaces ethernet eth0 mirror eth2
Note:
1. Only one of the following is allowed: "redirect", or "mirror"
2. Traffic-policy limiter is allowed with redirection/mirror and takes place
before mirror/redirect action (NEW)
3. Limiter policy may applied on output (NEW)
4. Only limiter policies can be applied on input (same as previous releases)
This does add some new functionality (#2, and #3) which are possible because
of how filter classes are implemented.
|
|
This reverts commit b75a66a4d5768f22a4b995d49ba442de8762a800.
|
|
This changes:
qos-policy traffic-limiter foo class ...
interfaces ethernet eth0 qos-policy in foo
to
interface ethernet eth0 input-policy limit class ...
Which unifies all uses of ingress qdisc (limit, redirect, mirror)
into a single usage model.
|
|
Bug 4541
Limiting implies packet drop, but this qos-policy
does shaping.
|
|
|
|
|
|
|
|
Random detect requires bandwidth which is not part of round robin
policy. Therefore drop it as a configurable value.
Bug 4595
(cherry picked from commit b1b823012b2a1aa0fa8578054fc1179f92a32c2f)
|
|
Bug 4590
(cherry picked from commit 447b5430e8f4019e5d3b6e3f1f2025712d1c7b7f)
|
|
The $VAR(.) is not set on end node, so have to go hunting
for policy type.
Related to Bug 4545
(cherry picked from commit 39712bfaf390a4435218a14308cc6efa1a2d39b8)
|
|
The CLI end: node does work correctly, so don't need to go walking configuration
manually
Bug 4545
(cherry picked from commit 377896b296827e4d7c3c20debb3487605c45bbe3)
|
|
Two more (optional) parameters for random detect precedence groups.
Bug 4518,4519
(cherry picked from commit 8104da09f5da338c6ce0813c367262de4e417598)
|
|
Bug 4516
(cherry picked from commit 26890ac98e3ed6c0830e61fa91f306fea0cc227a)
|
|
CLI quoting in expressions seems to have changed, so adapt.
Bug 4485
|
|
Bug 4485
Change template to deal with CLI changes in quoting evaluation
of expressions.
|
|
|
|
Packet reordering doesn't work unless there are packets in the queue,
and no packets are held unless there is a delay.
|
|
The RED queue setup needs a rate value and RR doesn't have rate configuration.
|
|
Make a simpler version of WRED that acts more like Cisco.
Use Diffserv on Linux paper for example of how to use DSMARK and GRED
to achieve similar result.
|
|
GRED only allows 1..16 for classes, and reserve one for default
so only 1..15 are ok for class value.
|
|
Class value of 2 is allowed.
|
|
|
|
Having too long a name messes up column output on operational commands.
|
|
Bug 4296
Ehancement to allow configuring GRED as form of weighted RED.
|
|
Use existing u32 infrastructure to match in header
|
|
It needs more work (not ready for prime time).
|
|
|
|
New policy allowing strict priority traffic assignment
|
|
Round-robin class based policy
|
