svn commit: r346017 - in head: libexec/rc libexec/rc/rc.d share/man/man5

[Jilles Tjoelker]

On Tue, Apr 09, 2019 at 10:24:16PM +0100, Chris Rees wrote:
> On 9 April 2019 22:13:29 BST, Chris Rees <crees at bayofrum.net> wrote:
> >On 9 April 2019 22:08:35 BST, "Rodney W. Grimes"
> ><freebsd at gndrsh.dnsmgr.net> wrote:
> >>> On 09/04/2019 20:59, Chris Rees wrote:
> >>> > On 9 April 2019 20:55:07 BST, "Rodney W. Grimes"
> >><freebsd at gndrsh.dnsmgr.net> wrote:
> >>> >>> On 09/04/2019 21:33, Rodney W. Grimes wrote:
> >>> >>>> I think the trigger issue is:
> >>> >>>> grep zfs /etc/rc.d/zvol
> >>> >>>> rcvar="zfs_enable"
> >>> >>>> required_modules="zfs"

> >>> >>>> that module requires may be going south with the
> >>> >>>> new code when the module is built into the kernel.
> >>> >>> Maybe it's because the module's name is zfsctrl (for whatever
> >>reason)
> >>> >> while the
> >>> >>> module file is named zfs.ko.
> >>> >> I suspect that could also lead to issues with the new code.
> >>> >> It seems to be failing to detect that zfs is infact functional in
> >>the
> >>> >> kernel,
> >>> >> and blindly, or not so blindly, trying to load zfs,ko, which when
> >>you
> >>> >> build
> >>> >> it into the kernel you usually do so without any modules built,
> >so
> >>> >> there is
> >>> >> no /boot/kernel/zfs.ko, and even if you did build it any attempt
> >>to
> >>> >> load
> >>> >> it would return an error.
> >>> > Loading with it built in isn't a problem, as I showed earlier.

> >>> > Loading when it doesn't exist *is*.

> >>> > I'm torn.  Either we could revert this, or add a check to the
> >>required_modules function instead, which I think is the better solution.

> >>> Hang on,

> >>> [crees at pegasus]~% sudo kldload -n zfsctrl && echo yes
> >>> yes

> >>I think your testing the return value of sudo here?

> >Sudo returns the child's return value.

> Turns out Oliver had also reported this to current@ with a log

> https://lists.freebsd.org/pipermail/freebsd-current/2019-April/073148.html

> Jilles@, mind if I revert this while I get some testing on this
> scenario done?

> It seems to me that zfs may not be included in the kernel, just
> zfsctrl, or something like that.

It seems like kldload -n does not work as expected for zfs, so reverting
seems the right approach.

-- 
Jilles Tjoelker