The JAIL_USE_DESC lookup jumps to found_prison before the descriptor is
allocated, so jfd_out is still -1 and the desc parameter is copied back
to userland unchanged.

What are you fixing with this change? This comment describes what the patch is doing, but not the punchline, e.g., "this fixes a panic", "this fixes a jail descriptor leak", "this returns the right data in <X scenario>", etc.

Providing this information could be helpful for folks who don't have the context you have for understanding/consuming the change :).

What are you fixing with this change? This comment describes what the patch is doing, but not the punchline, e.g., "this fixes a panic", "this fixes a jail descriptor leak", "this returns the right data in <X scenario>", etc.

it returns success but gives you back no descriptor.

Hm, how long has this been broken for? How'd you find it?

Let's poll some jail folks and see what they think.

In D60984#1398481, @adrian wrote:

Hm, how long has this been broken for? How'd you find it?

14 months, from 851dc7f859c2 by jamie.

I was trying to see how to build an out of tree driver for jails.

Only 15 and Current are affected.

Suggested new commit message description:

Prior to this change, `jail_get` when called with `JAIL_USE|JAIL_GET_DESC`
would fail to return a valid descriptor in the positive case because the
`JAIL_USE_DESC` block jumped to `found_prison` before the descriptor was
allocated. This resulted in the returned jail descriptor always being -1,
despite the call being successful.

This change fixes the logic so `jfd_out` is properly initialized in the
aforementioned scenario.

MFC after:	3 days

It would be a better stylistic fit if the JAIL_USE_DESC block were moved to after the (JAIL_GET_DESC | JAIL_OWN_DESC) block, i.e. putting the JAIL_AT_DESC block first. That wouldn't change anything operationally, since only one of JAIL_USE_DESC or JAIL_AT_DESC can be set. But this way it more close matches the existing logic of kern_jail_set, where JAIL_AT_DESC and (JAIL_GET_DESC | JAIL_OWN_DESC) are checked in one place, and then JAIL_USE_DESC is checked further down.

This looks fine to me. @jamie one final pass?