The "emergency" and "rescue" startup modes (which superseded "single-user" mode in the 1990s) are two of the three system startup modes, the third being "normal" mode.
They are usually signalled by flags such as -b or -s being passed as command-line arguments to the first user process, running system-manager, which in turn passes them along to the init subcommand of system-control.
As mentioned in the shutdown manual, these are startup modes, not shutdown modes.
One can only start up into them, not shut down to them.
The emergency target wants a service named emergency-login@console, an emergency login service instatiated against the system console device.
(See freebsd-console and linux-console.)
The main command that this runs is emergency-login, to check the password of the superuser, and do a minimal login dialogue with "glass TTY" semantics and fixed to the "root" superuser account in the system account database.
It inherits the minimal environment set up by system-manager, as modified by vc-get-tty.
Elsewhere in this service, setsid makes the process a session leader, open-controlling-tty sets up the system console as the controlling terminal for the session, and setlogin sets the login name for the session.
"emergency" mode brings up no other services, and thus mounts no (non-"API") filesystems, and does not even necessarily make the root filesystem available read-write.
In shells such as the Korn, Bourne Again, and Watanabe shells, POSIX-conformant mode can be turned off once the shell is running with set +o posix. In shells such as Debian Almquist and Almquist shells, there is effectively no non-POSIX-conformant mode.
It is possible to use the shell setting in the login.conf subsystem to differentiate the emergency-mode shell from the rescue and normal mode shells.
This can permit, for example, the emergency mode shell to be statically linked whilst the rescue and normal mode shells are dynamically linked.
Note:
On FreeBSD until 2021, this required setting up root and toor with distinct login classes in login.conf .
Until 2021, FreeBSD stuck with the long-standing convention of the root account's login shell being the C shell, and this is still assumed for a lot of FreeBSD documentation.
After 2021 on FreeBSD, this merely requires a modification to the root entry in the global login.conf.
Note:
NetBSD uses its Almquist sh for the login shell of both the root and toor accounts.
As of 2023 it already configures, out of the box, its toor account to specify a statically linked build of its Almquist sh as the account's login shell to run.
The rescue target is a more conventional TUI login service, like the ttylogin@device services used for virtual terminal login but instantiated against the system console.
It expects the system account database to be functional, and permits logging in to accounts other than the superuser.
The login dialogue that this presents is displayed by login-envuidgid, and it expects the console to be a "modern" terminal capable of 1980s cursor addressibility, colours, and character attributes; and of 1990s Unicode.
Note:
If the freebsd-console or linux-console is a serial or parallel device, the DTE at the other end of the wire is expected to be capable of speaking UTF-8.
This is the 21st century, of course; even Microsoft HyperTerminal was capable of handling UTF-8 on Windows XP.
"rescue" mode brings up local mounted filesystems, runs checks, and does basic system intializations such as services that set the dynamic hostname and machine ID.