Bumping a pet bug

[Jilles Tjoelker]

On Sun, Jan 22, 2012 at 12:46:10PM -0800, Garrett Cooper wrote:
> On Sun, Jan 22, 2012 at 5:46 AM, Dirk Engling <erdgeist at erdgeist.org> wrote:
> > Here is the output I get from not running the jail script as root:

> > Starting jails:mount_nullfs: Operation not permitted
> > devfs rule: ioctl DEVFSIO_RGETNEXT: Operation not permitted
> > mount: dev : Operation not permitted
> > /etc/rc.d/jail: WARNING: devfs_domount(): Unable to mount devfs on
> > /usr/jails/foo.com/dev
> > /etc/rc.d/jail: WARNING: devfs_mount_jail: devfs was not mounted on
> > /usr/jails/foo.com/dev
> > ln: log: Permission denied
> > mount: /usr/jails/foo.com/dev/fd: No such file or directory
> > mount: proc : Operation not permitted
> >  cannot start jail "foo_com":

>     Probably a simple case of lack of set -e in the script with
> appropriate error catching logic. The jailme and portsjail [*] scripts
> had similar issues.

You cannot use 'set -e' in rc.d scripts as it must be off both while
sourcing /etc/rc.subr and while calling any function defined by it.
Violating this may lead to unexpected exits of scripts.

I think 'set -e' is not a "handle all my errors for me" panacea. If the
script is any more complicated than just a sequence of commands, 'set
-e' will very likely not do what you want. For example, if a compound
command or function call is "tested", -e is ignored during all of it.

The exact meaning of 'set -e' has changed several times, yet these
problems have not been solved. I don't think it can be "fixed", so the
least bad thing is to follow POSIX so the wrongness is at least
consistent. (Unfortunately, there are still people trying to "fix" -e in
POSIX in 2012.)

Instead of -e, use explicit error checks such as with || or &&.

-- 
Jilles Tjoelker