I don't think this part can work. dep_setup is called exactly once- if !docf, then we never setup jails_byname just below and this condition is trivially true.

I would consider, perhaps, reworking running_jail ever-so-slightly to take a struct cfjail **jail and populate *jail if jail != NULL -- and always returning the jid if it's running.

1. running_jail returns running cfjail* even if there is no config file. So this part is supposed to work as before.
2. now we return cfjail with name and jid. So it's easy to fetch jid from cfjail. in case of extra argument (I supposed) it will be strange to return same info twice: as jid and as cfjail. If such API looks good, I will add extra argument. I preferred to keep API as simple as possible.

I suspect this leads to a segfault before we get to fake the cfjail *, though, with an uninitialized jails_byname

Sorry- this should sort after fstyp

Explicit initialization to NULL shouldn't be required, since this is a global

We generally omit the shebang for atf-sh tests -- I don't recall the exact reasoning, though the framework (kyua, perhaps? or perhaps the build framework) will DTRT.