Re: Vanishing GPT/UFS labels

On Thu, May 21, 2026 at 6:46 AM Stéphane Rochoy <
[email protected]> wrote:

> Hi Hackers,
>
> I have some code at hand that rely on GPT and UFS labels to abstract
> things a bit. But a few observations puzzle me.
>
> My initial observations were made on an nda(4) disk but I managed to
> reproduce with an md(4) one on a15.0-RELEASE-p3 (5cf7232732d5) machine:
>
>     dd if=/dev/zero of=zero.img bs=100M count=1
>     md=`mdconfig -f zero.img`
>     gpart create -s gpt $md
>     gpart add -i 1 -l foo -t freebsd-swap $md
>     gpart add -i 2 -l bar -t freebsd-ufs  $md
>     newfs -L bar /dev/${md}p2
>
> From here the following devices are available to match the labels:
>
>     /dev/gpt/foo
>     /dev/gpt/bar
>     /dev/ufs/bar
>
> When I mount using the underlying device, /dev/${md}p2, both
> /dev/gpt/bar and /dev/ufs/bar vanish. And when I umount (using the mount
> point) both are back.
>
> When I mount using the UFS label, /dev/ufs/bar, only /dev/gpt/bar
> vanish. And when I umount the GPT label don't come back.
>
> GEOM don't seems to agree with itself: `gpart show -l $md` do confirm both
> partitions kept their GPT label but `glabel status` don't:
>
>     $ gpart show -l $md
>     =>    40  204720  md0  GPT  (100M)
>           40   20480    1  foo  (10M)
>        20520  184240    2  bar  (90M)
>     $ glabel status | grep -e foo -e bar
>          gpt/foo     N/A  md0p1
>          ufs/bar     N/A  md0p2
>
> I would be happy to hunt what looks like bugs to me but I need some
> insight first: what is the rationale behind the idea to remove a device
> on mount and, possibly, restore it on umount?
>

This sounds like the intentional behavior of geom to withdraw the devices
when aliases are mounted.

Warner