53008 bytes seems like an odd buffer size for the network stack to use..

is it 52K - sizeof(some headers)?

What does the definition NO_UNLOAD_STUB(procfs, oprgetstatus,nomod_zero) in modstub.S do?

I guess a static function is not necessary to list there?

richlowe: no idea, but the fact it happened twice seems to suggest it wasn't just random

(even ignoring the 0 byte usage on one)

NO_UNLOAD means once you hit the symbol and the module loads, said module will not unload afterwards.

is there a way to add a dataset to a zone online, the same way the zone does it on booting? If I mount the dataset into the zone with mount(8), the mountpoint must be legacy of course, that way I cannot restart the zone without loosing the mount.

Could you not update the zone properties (zonecfg) in parallel with the explicit mount such that next reboot it'd just work?

danmcd: I can do that, I just wanted to get a whole dataset delegates to the zone without having to restart the (central database) zone

I did a bad design choice when configuring the zone.

and I must work somehow, I just wonder if we have a userland option for it. if not, that would be a good RFE

um, I think S11.4 got live zonecfg updates at some point in time.

yes, they have a zonecfg -r for running config

I wondered if there is just some mount flag or low level tool around ;)

uugh

does dhcpagent try to explicit start the routing/route service?

i have 3 interfaces on the same subnet all w/ dhcp

and every so often ipadm marks them disabled which ends up being some unfixable state aside from rebooting

ipadm enable-addr doesn't work on it

which seems like a bug, but also not something I can really dig into right now unfortunately

(granted, we should probably move dlmgtd and ipmgmtd to store their stuff in smf instead of their own bespoke code which is buggy)

the -t handling in particular

jbk: the temp flag? I have not had issues with that and I use it almost everyday. Are there any existing issues written for this?

I only had issues when nwam was running, but those are nwam bugs

i thought i filed one

but like using that to reconfigure an aggr is _horribly_ broken

oh, ok - that might be. I rarely use aggregates

ahh there was just an internal ticket... but I can file an illumos.org one

I remember looking at trying to solve it correctly, and IIRC it was a mess because of all of the bespoke db management code

→ BUG 18366: Mixing temporary and persistent dladm aggr commands can create Schrödinger's aggr (New)

the scenario was 'i want to remove this link from the aggr, but not persist across reboot... ok now I want to add it back to the aggr (and have it persist)'

rmustacc: github.com/illumos/illumos-gate/blo…ommon/io/i40e/i40e_intr.c#L472-L473 -- shouldn't that be `>=` ? (I'm like 98% sure yes, but then I'm trying to understand why if true, why it's not caused problems to date)

(I'm happy to file a bug and fix it -- we have a customer where all 4 ports on one i40e card seem to have stopped generating interrupts as best as I can tell, which led me down this route)

just.. we have a fair amount of customers using them as well as it seems at least somewhat popular in the illumos communtiy w/o similar reports, which makes me wonder if i'm just missing something obvious

(looking at the full queues, 1500MTUs, no LSO, data is copied, so the 8 descriptor limit is not a factor)

jbk: It'll be a bit before I can look. Just it's been a long time since I've beeni n this code.

ok