Re: Vanishing GPT/UFS labels

On 5/22/26 21:22, Warner Losh wrote:
> 
> 
> On Fri, May 22, 2026 at 12:53 AM Stéphane Rochoy
> <[email protected] <mailto:[email protected]>>
> wrote:
> 
>     Warner Losh <[email protected] <mailto:[email protected]>> writes:
> 
>     > On Thu, May 21, 2026 at 6:46 AM Stéphane Rochoy
>     <[email protected]
>     <mailto:[email protected]>> wrote:
>     >  …
>     >  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.
> 
>     So what I observe when mounting/umounting via the underlying device
>     (/dev/${md}p2) is the expected behavior.
> 
>     But there's some problems when mounting/umounting via the UFS label
>     (/dev/ufs/bar):
>     - on mount, /dev/ufs/bar should vanish too

[I disagree about the above: The device-name independent naming in use
needs to survive for use in my view. For example: avoid the variability
for da* naming that can occur with multiple da* devices present.]

I am actually writing just about the below.

>     - and on umount /dev/gpt/bar should come back.
> 
> 
> Those details aseem wrong to me...

I can replicate the umount not restoring /dev/gpt/* on main:

# gpart show -pl /dev/nda1
=>        40  2930277095    nda1  GPT  (1.4T)
          40        2008          - free -  (1004K)
        2048      532480  nda1p1  FBSDFSSDefi  (260M)
      534528     1562624          - free -  (763M)
     2097152  2919235584  nda1p2  FBSDFSSDroot  (1.4T)
  2921332736     7340032  nda1p3  HyperVswp3p5  (3.5G)
  2928672768     1604367          - free -  (783M)

# tunefs -p /dev/gpt/FBSDFSSDroot 2>&1 | grep "volume label"
tunefs: volume label: (-L)                                 FBSDFSSDufs

# ls -C1 /dev/*/F*
/dev/gpt/FBSDFSSDefi
/dev/gpt/FBSDFSSDroot
/dev/ufs/FBSDFSSDufs

# mount -onoatime /dev/ufs/FBSDFSSDufs /mnt
# ls -C1 /dev/*/F*
/dev/gpt/FBSDFSSDefi
/dev/ufs/FBSDFSSDufs

# umount /mnt
# ls -C1 /dev/*/F*
/dev/gpt/FBSDFSSDefi
/dev/ufs/FBSDFSSDufs

# glabel load
# ls -C1 /dev/*/F*
/dev/gpt/FBSDFSSDefi
/dev/ufs/FBSDFSSDufs

If this is wrong, then maybe a bugzilla should be submitted.

> 
> Warner 


-- 
===
Mark Millard
marklmi at yahoo.com