The interface in question might not exist at init:49 if some other
change requires it to be recreated from scratch, in which case it will
be removed in exit:50. Therefore, run the flush at exit:49, where it
should always still exist.
When timestamping of scripts was added, we unfortunately started to
record the exitstatus of `ts(1)`, rather than the script itself.
Refactor the metadata collection to use `time(1)` instead, which will
propagate the inferior's exitcode. This will also allow us to collect
other parameters in the future.
By default, Linux will map an IP packet's TOS field into skb->priority
when acting as a router. Unfortunately this is only globally
configurable (per network namespace), which leaves us with the
following dilemma:
If we keep reclassification enabled, then a setup like this...
vlan1 vlan2
\ /
br0
/ \
e1 e2
...would likely not work as expected because any priority information
from hardware, that might be mapped to skb->priority via vlan1's
ingress-qos-map, would then be replaced by the standard TOS-mapping on
the routing boundary.
On the other hand, if we disable reclassification, then we won't be
able to source priority information from the TOS field when no other
source is available, as is the case when the source interface is a
regular NIC.
Given that the primary use-case is to run on switchdev ports, we will
most likely always have priority information available from hardware,
which is why we opt to disable the reclassification.
If this ever becomes configurable per interface, then we could open up
the configuration to let users define the policy of which layer's
priority should take precedence.
Opportunistically look for interfaces with multiple transmit queues
and hardware support for the mqprio queuing discipline. For every
matching interface, set up mappings from kernel-internal packet
priorities, via traffic classes, to transmit queues such that as many
high priorities as possible are scheduled on separate queues.
This makes it easier to add more scripts without having to update the
finit config file, and it clarifies which of the scripts in
/libexec/infix only execute at boot.
Sync after `make update-defconfig`:
- avahi selected by mdns-alias package (all)
- iptables and ca-certs selected by other packages (r2s)
Signed-off-by: Joachim Wiberg <troglobit@gmail.com>
Very simple tests work (hostname.py), more complex
tests (static_routing.py) failes due to the lack of
model prefix, for example. infix-ip:ipv4 instead of just
ipv4.
We need to fix this before you send your configuration, run
it through libyang for example and get the full model-name.
This enables access to /restconf on port 443 over HTTP/1.1 and HTTP/2
via rousette when enabled.
Also, slight refactor to allow including /restconf and /netbrowse as
optional locations in the main servier directive.
Signed-off-by: Joachim Wiberg <troglobit@gmail.com>
With increased use of augeas to manage users, groups, and permissions,
we seem to have reached peak complexity. Because when re-running the
same test[1] over and over, somtimes the aug_save() function fails to
actually save the changes to /etc/shadow to disk!
We have already experienced weird behavior with augeas, see comment in
deleted code, and these latest problems became the last straw. We have
decided to replace it with standard C APIs.
As a spin-off, thanks to the requried refactoring, the check_shell()
function, used by the change_nacm() callback, could be replaced with a
call to set_shell() for both "is-admin" and "non-admin" cases. This
means we now also restore Bash privileges (and not just take them away)
when modifying nacm group memberships.
Fixes#468
[1]: test/case/ietf_system/user_admin.py
Signed-off-by: Joachim Wiberg <troglobit@gmail.com>
Checking if a user is an administratr we cannot rely solely on the
return value of sr_get_items() but also ensure at least one group
matches the search criteria.
Fixes#469
Signed-off-by: Joachim Wiberg <troglobit@gmail.com>
Since containers managed by docker runs as root, files created in bind
mounted volumes will be owned by root:root. This is a problem for
files generated during test execution, as the calling user does not
have permission to remove them.
Therefore, add a wrapper entrypoint that hooks up an exit handler that
will fixup the permissions before exiting the container.
Now that Reggie builds all supported archs there is no point in keeping
two separate (and now very similar) workflows. So we've decided to lay
off Reggie and let Bob, who has seniority, take over Reggie's tasks.
This simplification also means that jobs will no longer be started on
push to branches, only pull requests. For half-baked branches a draft
pull request can be used to auto-start builds, if needed.
Signed-off-by: Joachim Wiberg <troglobit@gmail.com>
