Can someone on OI or even Helios please utter: `echo "::print -ath vnode_t" | mdb -k` and mail me or pastebin the results?

kebe paste.ec/paste/-Oqhouv1#r6VA2zaE5i8Xav7v+2ss-ceSSfSMCgM23FTKHl/U2Vd

Thank you. Moving it to the end or shoving it between v_type and v_rdev would solve the problem poor Laquen discovered the hard way... (look at those holes!)

But (as I said on twitter to jbk) if we make the change there's this whole range of {Omni,Smart}OS with vnode_t that's "off".

Well, one of the two versions will need to recompile to reconcile.

Thing is both {Omni,Smart}OS have releases so it will be quite near impossible to have one version for both releases

I'm in the midst of filing an OS-bug (whether it gets fixed or not is a different question) to address this.

WIth SmartOS the usual answer is "just upgrade to $LATEST" and that's it. It's harder with Omni and its Stable or LTS (and I won't claim to speak for Omni's opinion on all of this).

kebe: was there ever a contract for the contents of vnode_t?

AFAICT, no. But I did have flashbacks to the 90s where I was admonished for altering ill_t similarly because 3rd party FW vendors (e.g. Checkpoint) requred ill_t offsets to not change if possible.

I'm writing up the tradeoffs in an OS bug that I will make public immediately after filing.

pmooney: There is a pretty explicit comment about a set of public members, I think, in the vnode_t

It would be better of course to provide accessor symbols for the things people actually need to do

and, really, any documentation haha

yeah

My kettle is already boiling a different part of the ocean though

has anyone seen blkdev unbinding a bad dma handle under vioblk?

(or even not under vioblk)

bad in this case is baddcafe

no.. closest was a fix from april by woodstock for the blkdev softstate, but no DMA handles

(14654)

I just filed #15123 (fenix?)

BUG 15123: blkdev can unbind an unbound DMA handle (New)

↳ illumos.org/issues/15123

I confess I don't understand why it's special casing a 0 transfer if it's not going to special case it fully

more importantly, I can't test my hypothesized fix in any kind of real environment in the same way you, Hans, etc could.

but I feel like the fix is more a bandaid for whatever half-complete special casing is trying to happen there

yeah, not sure what the thought process there was