simple.qos had accidentally set up the egress shaper twice, once
with the true egress parameters and a second time using the ingress
parameters, effectively misconfiguring both directions. This bub
only affected situations where 3-tier ingress classification was
used.
Signed-off-by: Sebastian Moeller <moeller0@gmx.de>
sqm_logger tried tro wait indefinitely if passed an empty string.
This in turn makes sqm-scripts hang. Quoting the input argument in sqm_logger
seems to fix the problem.
Signed-off-by: Sebastian Moeller <moeller0@gmx.de>
The last batch of changes tried to teach the GUI to pass link layer
options to cake but forgot to actually call the function that parses
the GUI variables and used it as a string insteead. So this fixes that
it also tries to allow the use of the tc_stab link layer adjustment
method with cake so the implementations can be validated against each other
easily. Needs testing...
Signed-off-by: Sebastian Moeller <moeller0@gmx.de>
The cake traffic-shaper qdisc omne stop solution knows how to handle
link layer adjustments for ATM and can account for per packet overhead.
This commit adds cake as link layer adjustment mechanism in the GUI and
passes numerically specified overhead as well as the ATM linklayer
keywords on to cake. This change also passes the "advanced option strings"
from the Queue Discipline tab to cake. But as before no error checking.
This needs testing, as I have no working cake qdisc available so
caveat emptor...
Signed-off-by: Sebastian Moeller <moeller0@gmx.de>
As Hnyman noted in https://github.com/dtaht/ceropackages-3.10/issues/13
we carry a few unnecessary dependecies in sqm-scripts, so remove one of
them (iptables-mod-filter) as we neither use it nor plan to use it.
Signed-off-by: Sebastian Moeller <moeller0@gmx.de>
The initial conversion to restart sqm on interfaces it is configured
for in case of (transient) dis- and reappearance was half finished.
These changes clean up the handling of exlicitly passed interfaces
in run.sh: no second argument defaults to all configured interfaces
the alternative is an individual interface name passed as 2nd
argument to run.sh. The first argument either is start or stop.
No argument at all will behave as if start was passed.
Survives light testing...
Signed-off-by: Sebastian Moeller <moeller0@gmx.de>
Alan Jenkins noted a bug in the smq luci GUI that effectively
erased several configuration paramters if two checkboxes were deselected.
This behaviour seems consistent in luci but certainly has the potential
to confuse users. While confusion can not really be avoided generally
it seems wise to change the default interpretation for empty or non-existent
itarget and etarget variables from the qdisc's default (5ms in the case of
one of the codels) to automatic determination of tghis variable dependent on
the configured bandwidth, as codels target variable should be large enough
to contain at least one full packet. With this change sqm-scripts will
do the right thing by default, but will yet allow the user to specify
over-ridding values (as long as the user does not un-check the
entry-field exposing check boxes). Survives light testing...
This change set also changes the sqm-scripts luci gui to note the user
of the change. For compatibility with existing setups sqm-scripts
will still honor "auto" as an alternative explicit way of requesting
automatic target selection. This might turn into a warning in the future
and might be phased out...
Signed-off-by: Sebastian Moeller <moeller0@gmx.de>
During system start up pppoe devices seem to receice ifup events before
the interface actually exists. This commit makes sqm's run.sh script
test whether the sys files for an interface exist before actually trying
to start an SQM instance on an interface. This seems to nicely avoid
starting on an not fully established pppoe interface and avoids a number
of error messages during startup.
In addition, debug logging is disabled.
Signed-off-by: Sebastian Moeller <moeller0@gmx.de>
Some interfaces like wan-pppoe go away, when the ppp connection is lost
and get recreated once the link is established again. SQM now
has its own hotplug script to re-enable itself on the interfae just hotplugged.
SQM will not touch other instances of itself running on other interfaces
if called by hotplug.d. The implementation now allows this functionality by
calling run.sh like:
/usr/lib/sqm/run.sh interface YOUR_INTERFACE_NAME_HERE
e.g.: /usr/lib/sqm/run.sh interface ge00-pppoe
If called with a specific interface SQM will only try to disable itself
on that interface to clean up all left over state and the re-enable
itself on just that interface. Hopefully that allows for better service
with instable interfaces like pppoe. The current code passes a simple manual
stop start test of the ge00-pppoe interface from the GUI and does seem
to do the right thing, at least on cerowrt 3.10.50-1...
Use INSTALL_BIN & INSTALL_DATA macros instead of cp
to ensure correct file permissions for the executable files.
Signed-off-by: Hannu Nyman <hannu.nyman@iki.fi>
Changes committed to the cerowrt original repo after the initial import here:
- Better license & copyright statements, as requested
- Fixed a minor bug in stopping sqm
- Logging improvements
- Dead code removed
- Typos corrected
Signed-off-by: Hannu Nyman <hannu.nyman@iki.fi>
Toke Høiland-Jørgensen
Sebastian Moeller
Dave Taht
Hannu Nyman