it should be freshly allocated in arc_write_ready, just before calling abd_copy(). however, as ::whatis is suggesting some of those pointers are actually freed from cache, I kind of fail to see how... of course, if some other thread was poking this arc_buf_hdr ...

anyhow, its too late here, need to sleep on this:D

tsoome: There were some commits by amotin to OpenZFS recently that might or might not be relevant.

I recall some of the changes involving ensuring done callbacks were properly called. I also vaguely remember a recent PR about reference counts.

Oh. The reference count patch was in the DMU/DSL and was by someone else: openzfs/zfs d816bc5

[illumos-gate] 14596 cstyle should allow empty increment with for loop -- Toomas Soome <tsoome⊙mc>

is it me, or is redmine being sluggish?

It is not just you

there it goes...

granted that was a long description

(#15420)

but i thought it might be helpful to show how we got from 'devinfo callers are stuck in the kernel' to 'deadlock between dls and mac'

since it was a bit of a journey

so many bad memories of that journey

Really nice write up :)

I have just been looking at lock ordering in this area for issue 15167 (fenix)

fenix will no doubt take a while to talk to remine too..

*redmine

BUG 15167: Panic when halting a zone with self-created links (In Progress)

↳ illumos.org/issues/15167 | code.illumos.org/c/illumos-gate/+/2662

the 'modunload -i 0' in /sbin/rc3 was probably the most 'whaaaa' part of it honestly :)

unfortunately, i'm not entirely sure of how to fix it...

missed the context on the modunload -i 0 but IIRC that's related to reducing kernel footprint after a reconfiguration reboot when all modules get loaded to see if they're needed.

yep.

ah, it's in #15420

(the context, that is)

and, yes, that's a good writeup.

if there's one thing I took working for joyent, it's a huge appreciation for good bug descriptions

it kinda spoiled me when then getting bug reports that consist wholly of 'xxx is busted' :)

(elsewhere)

fenix illumos 15420

fenix illumos#15420

BUG 15420: dls_devnet_destroy can deadlock with link notifications (New)

↳ illumos.org/issues/15420

[illumos-gate] 15402 Incorrect service name and other errors in the zonestat(1) manual -- Peter Tribble <peter.tribble⊙gc>

jbk: Joyent is not the only place that values good bug descriptions. OpenZFS/ZFSOnLinux values it too. In fact, I learned to write good commit messages and descriptive reports from Brian shortly after I started contributing. There are times when I feel that I am writing excessive amounts of documentation, but so far, I have never once heard someone tell me to write less.

In fact, at the OpenZFS developer summit last year, I mentioned at the bar that I felt that I was writing excessive documentation for changes. I was told to keep doing it. :D

Detailed long-form bug writeups were part of the engineering culture within the part of Sun where I worked and it's good to see that this has carried over in the diaspora...

yeah, i'm sure that came from the various ex-sun people that were there

jbk: Sorry, redmine has been sluggish. There's a shady web crawler probably related to the TikTok family of companies that is pretending to be an old Android phone. I've just blocked its user agent string. Let me know if it gets sluggish again.

jbk: I remember hearing a talk by Bryan Cantrill that said that it was from Jeff Bonwick.

you definitely appreciate it when you're coming in basically blind to an issue

or are wanting to understand why a change was done

jbk: I appreciated it not that long ago when some dead code made no sense to me. I simply looked up the original rationale for the change and then I realized that the intention had been to make a warning appear less often, but by putting an assignment in the wrong case statement, the patch made it dead code entirely.

It was this: openzfs/zfs dee8934

That was 5 months ago... it still feels like yesterday for some reason. :/