very, very minor thing.. but should we consider adding 'VirtIO' as a controller type to libdiskmgmt (or some other suitable type) so vioblk devices have a value for the 'type' column in diskinfo (instead of '-') ?

yes :)

i'll file a ticket for anyone that wants to do it..

on a different subject, can (with suitable perms) a userland process publish kstats? e.g. you have a service and want to expose stats

jbk: well, there's kstat_write(3C), but I don't see anything in libkstat to create a kstat in the first place...

yeah, i saw that, but nothing in userland in illumos-gate uses it, so it wasn't clear how it'd work from userland

the 'k' in 'kstat' is 'kernel'

but 'ustat' sounds cool? :)

i mean, the trendy thing is to include a web server in all your binaries so you can run a webserver on random ports to export grafana or such, but that does seem excessive (though at least with C stuff, it could be put in a .so)

literally all I remember about kstat/statistics/etc, is that postwait has _opinions_ about histograms

heh...

i keep meaning at some point to try to put together a proposal for adding some sort of histogram type to kstats

since that would be useful for a number of things

but how many people know the difference between ustat(2) & statvfs(2) ?

circling back to say I'm sure theo would tell you they weren't opinions, but revealed and indisputable fact

alanc: ECOMM implies things I don't want to consider

ancient nfs wizardy? rfs wizardry?

RFS I think

that one came from New Jersey, but is still used in a few commands because it takes a dev_t instead of a pathname

cve.org/CVERecord?id=CVE-2026-32282 - just call system calls directly they say, because Linux guarantees binary compatibility of them forever, even when they forgot to include a flags field in the system call and thus had to create a new fchmodat2() system call that you need to call instead to get the full POSIX functionaliity

compatible though!

and no slowdown going through that pesky libc