when you add an IP to an interface (e.g. ipadm create-addr ....), do we send a gratituous arp out?

danmcd: would that maybe be in the dad code?

Yes jbk, it's DAD. Yet another IPv6-invention that got backported to v4. :)

ok.. i thought so

(there was a concern that if an IP moved from one host to another, that something not already in the system would have to force a gratituous arp.. but i thought that happened when you added the ip anyway)

I'm trying to zfs-send a backup from Linux to illumos; recv on illumos is failing because the acltype and xattr=sa props are not supported. But as far as I can tell this is purely because the properties themselves are checked in spa_prop_validate(); the data is compatible (a child dataset which inherits these props instead of having them local on Linux can be send/recv'd fine on illumos; ACL's are not usable on illumos but preserved when sending back to

Linux, and I can also import an entire pool with these props set without problems). Would it make sense to add these properties to illumos ZFS for compat so they can at least be recv'd, even if when set they don't do anything useful on illumos?

Trying to ignore or override them with `recv -x/-o` doesn't work, because both recv checks if the property is valid before (which acltype isn't because its not supported) and it ZFS still checks the received props for validity (to save them as "received").

Looked into if changing that would be a good solution, but seems like that would be a way bigger change across a dozen of functions. Adding them as placeholder props seems cleaner; but I don't know how to determine what other side effects that might have.

Is DNS configured through /etc/resolv.conf? Or should I use svccfg to configure the network/dns/client service?

pilonsi: through resolv.conf. network/dns/client is more of a milestone that other services can depend on (don't start me until gethostbyname has a chance of working..)

Thank you! Do you know if there is a plan to migrate to managing more things like this through svc like it seems solaris did, or it's stable?

Likely to stay the same unless a volunteer feels motivated to change it.

Speaking about that... I would like to contribute to illumos, I work desigining ASICs, but I don't have that much OS programming experience

Im following the xv6 mit class first, but if you have any suggestion of reading materials, small tasks i could start looking at or some tips it would be nice

Seek bite-sized issues in illumos.org/issues (There's a "difficulty" search filter, set it to bite-sized)

illumos.org/books/dev is also a good resource

Thank you!

If its ok I can do this one illumos.org/issues/15839

→ BUG 15839: Erroneous newline in error message (New)

To get used to the whole procedure, since the fix itself is just removing a newline

Sure. @jbk filed it FWIW.

go for it

i just ran into that on a system (ab)using vnics that ended up running out of memory

Nice, I already have an account in the bugtracker

I can leave a comment there and ill get to it tomorrow

(To leave a comment it's the edit button?)

just set yourself as the assignee (or whatever the field name is)

yeah, edit button comments.

weird I know.

I don't see any option to self assign it to me, maybe i don't have permissions? Just the edit button which takes me to a "new post"-like window, the watch button and then add subtask and add related issue

do you have a Change Properties at the top of the edit popup?

that's where assignee, etc, is.

redmine is not the most friendly

you might need a role to be in the list, but if so I can't fix that right now. Perhaps danmcd or someone can?

I just see the word Edit and below a gray box which contains the editor and says notes and inside that 2 tabs, edit and preview and a lot of formatting options

What's the user name on redmine?

pilonsi

refresh and see if it's different?

Now I can see it, I've alredy assigned it to me

Thank you :)

hrmm.. you can only trace on arbitrary instructions in userland processes w/ dtrace, right?

(i.e. not the kernel)

That *sounds* right but I can't confirm that.

That's correct.

yeah, that would range between "very hard" and "impossible" in the kernel.

drat... guess I'll just need to add some SDT probes to these functions

and yes, that's the way to do it

just wondering what exactly "arbitrary instructions" means here?

with userland you could set a probe on 'foo+0x1e' (IIRC, it'd be like pid$target::foo:1e)

so you can basically set a probe at an arbitrary instruction

got it, thanks

just saw freebsd adding "kinst" provider that says "Kernel Instruction Tracing" and wondered if it's something similar

It is in spirit. I'll be curious to look at the safety aspects.

the only way I could think to do it would be some VM tricks where you copy the page, modify the offset, then 'exchange' it (for lack of a better term) in the page table.. but there's probably still plenty of holes in that approach

are you just trying to determine if a certain code path is taken? if so I've almost always been able to determine that by inspecting the functions arguments

jbk: the other way would be to suspend all cpus other than the one doing the rocket surgery.

figuring out places you can't do it safely would be exciting, too

I think you could watch the instruction to catch it, without having to alter the text.

but theres lots of places that would blow various amounts of your foot off

[illumos-gate] 15923 Add support for SP6 Socket Types -- Robert Mustacchi <rm⊙fo>

rzezeski: ISTR you've spent some time in the aggr driver :) do you happen to look at all at how the LACP frames are constructed?

err did

i'm wondering if on ports w/ jumbo frames if that'd impact the size of the LACP packets (my guess is that they're likely always < 1500 no matter what, but need to find the spec to know for sure)

I have no idea. Not sure if I ever had a great handle on the protocol itself. I spent more time just getting the ring support working