- User Since
- May 27 2014, 9:32 AM (319 w, 5 d)
Add forgotten in6_rmx.c
Mon, Jul 6
That's the route multipath part, which will be introduced soon.
Let's wait half a week to make sure we all agree on this one :-)
@bz: any objects on removal of this one?
I don't believe we need specific processing here either.
- Kernel shouldn't try to be "smart" and decide whether user-configured routing is correct or not. It can't fix the issue anyway.
- Even when the route configuration is incorrect, TTL will still kick in
Sat, Jul 4
Can you please add "TESTING" section?
Fri, Jul 3
Thu, Jul 2
Thank you for raising this diff! Was going to do the same, but you managed to do it faster :-)
LGTM, please see the naming comment :-)
Wed, Jul 1
Tue, Jun 30
There are still some things outstanding ( epoch entry in multicast code and probably a couple other ones ), but generally looks good.
Thank you for working on this!
I'll commit the change tomorrow.
Re nhop_ref_object(): indeed, in many cases we acquire refcount under radix read or write lock, which implies that some routes are pointing to that nhop -> it has non-zero count.
In some corner cases we want to do it without holding radix lock and in that scenario we need to check whether we were able to acquire reference or not.
Mon, Jun 29
Also: thank you for working on this! Access to to all of the sfp attributes in structured form is a long-waited stuff.
What is the plan of switching ifconfig to use it? SFF standards are evolving, the code is not trivial, we regularly need to update either defines or the actual code. Also, I guess QSFP-DD / OSFP might require separate handling. I'd really prefer not to have 2 different kind of headers and 2 code implementations for that in base.
Sun, Jun 28
Thanks for making the changes!
Thank you for adding the bridge support!
Please see some minor comments inline.
Do they fail on your setup?
Can you share the logs of failing forwarding/redirect tests?
This could just be my dev box (AMD Ryzen-based HP "Omen" desktop) running 13-CURRENT.
My (failing test is here: https://gist.github.com/neelchauhan/e830f9cab71f79c81a258b8bdbc295d4
Ipv6 stack on the interface does not become available immediatelt, as the LL addresses have to go through TENTATIVE state. Other tests typically include test&sleep cycle after address assignment to address it.
These sysctls do work, when testing via a bhyve VM.
Sat, Jun 27
Generally LGTM, I'd be happy to commit it once there are some tests.
Generally looks like a nice feature to have.
Would it be possible to add tests validating this behaviour?
Thu, Jun 25
Also: maybe worth splitting the review in 2-3 smaller ones to speedup the process?
Mon, Jun 22
LGTM, please see some comments inline.
Any chance a test can be added to verify the desired behaviour? IT shouldn't be too hard, there are multiple examples of tests creating 2 vnet jails for the simular purposes in netinet6/.
Sun, Jun 21
Wed, Jun 17
awesome to have some of style(9) rules reviewed and relaxed!
Sun, Jun 14
That would be awesome!
Jun 10 2020
Jun 8 2020
Make field order consistent across rib* functions.
Jun 1 2020
Minor nit: any chance jails a and b can be named bridge_stp_a and bridge_stp_b to avoid potential clashes with other tests?
Export IPv6 addresses in "normal" form.
Currently there is only one - ND6 default route checker. With the modular lookup engine, there will be 1 more for rtables where lookup engine is attached.
With that in mind, idea is to start with the simplest implementation and then enhance it if/when needed.
May 31 2020
Add forgottech epoch reclamation.
May 30 2020
Fix fields description.
Abandoning, arc created a new revision instead of updating an old one.
May 28 2020
May 25 2020
Can you please fill in the testing section?
May 23 2020
Include tom changes as well.