I agree with @se on this. There was no direct user case before introducing sysctlbyname(2), it worked as an by-effect of implementation of how sysctl(3) handled that.
IMO the issue is in how the user.* is implemented and maybe that should be fixed in the long term, but as a band-aid that version LGTM.

In D34171#773087, @alfix86_gmail.com wrote:

However, it is strange but real, sysctlbyname() returns 0 but does not set the "value", while sysctlbyname-improved sets "value" correctly:

.Good catch, tanks.

__assert_unreachable in correct place

Use __assert_unreachable().
Shorten the code by two LOC.

Use atomic_load

Use correct diff this time…

Address /some/ comments.

Address comments.

Not really, no. This one was found with a power of grep while searching for NEEDGIANT nodes.

As mentioned on irc, I believe we're bound by the already defined exit codes, as describes in section EXIT STATUS of bhyve(8).

Update man page.

I was thinking about that too as there is no in-tree driver that support that call anyway, but bsnmpd have an option to query it if the driver supports it, so it'd have to be removed too. Anyway, it's probably task for another patch.

ACK. Wasn't sure about the JAIL_ATTACH namespace anyway.

In D25041#551522, @mjg wrote:

Disabling ptrace should not be an option, it should be mandatory no matter what.

Guys, thanks for the constructive comments.

I was told 2.5 years ago that jail_attach(2) is insecure and I shouldn't use it. Apparently it's no longer the case - see for example rS360356. Can I please have it accepted now?

Nice one. LGTM.

OK, I'll mark them as mpsafe in a commit.

Use full context diff.

Commited as rS358333

Sync with rS358330 and split sys/dev/random bits to a separate review for so@ approval.

That it'll not be run more than one at a time (from userspace).

No, because the sysctl handler was running under Giant implicitly (as it wasn't marked as MPSAFE).