[linux-lvm] issue (or bug): when vgremove does removing lvmcache

[linux-lvm] issue (or bug): when vgremove does removing lvmcache

[heming.zhao]

Hello List,

when using below reproduced steps, lvm command show error info.
```
# pvcreate /dev/sda /dev/sdb

# vgcreate test /dev/sda /dev/sdb

# lvcreate --name main --size 200m test /dev/sda

# lvcreate --name fast --size 50m test /dev/sdb

# lvconvert -y --type cache --cachevol fast test/main

# lvs -a -o name,vgname,lvattr,origin,segtype,devices test
   LV           VG   Attr       Origin       Type   Devices
   [fast]       test Cwi-aoC---              linear /dev/sdb(0)
   main         test Cwi-a-C--- [main_corig] cache  main_corig(0)
   [main_corig] test owi-aoC---              linear /dev/sda(0)

# vgremove test --force
   Logical volume "main" successfully removed
   Assertion failed: can't _pv_write non-orphan PV (in VG )
   Failed to remove physical volume "/dev/sdb" from volume group "test"
   Volume group "test" not properly removed

# lvs # <==  output none

# vgs # <==  output none

# pvs
   WARNING: PV /dev/sdb is marked in use but no VG was found using it.
   WARNING: PV /dev/sdb might need repairing.
   PV         VG        Fmt  Attr PSize   PFree
   /dev/sda             lvm2 ---  300.00m 300.00m
   /dev/sdb   [unknown] lvm2 u--  300.00m      0
```

rootcause analysis:

1>
why output: Assertion failed: can't _pv_write non-orphan PV (in VG )

_vgremove_single
  vg_remove
   vg_remove_direct
    pv_write(vg->cmd, pv, 0)

pv_write
  if (!allow_non_orphan &&
     (!is_orphan_vg(pv->vg_name) || pv->pe_alloc_count)) {
         log_error("Assertion failed: can't _pv_write non-orphan PV "
               "(in VG %s)", pv_vg_name(pv));
         return 0;
     }

the pv->pe_alloc_count is non-zero.

2>
why pv->pe_alloc_count is non-zero.

_vgremove_single
  process_each_lv_in_vg
         /*
          * Only let hidden LVs through if --all was used or the LVs
          * were specifically named on the command line.
          */
         if (!lvargs_supplied && !lv_is_visible(lvl->lv) && !arg_is_set(cmd, all_ARG) &&
             (!cmd->process_component_lvs || !lv_is_component(lvl->lv)))
             continue;

cmd->process_component_lvs and  lv_is_component(lvl->lv) are false,
then lv can't join in final_lvl and won't be removed later.


At last,
I can't find a good solution for this issue. the 112846ce0b01e5353 introduced this issue.
one solution maybe to add CACHE_POOL_METADATA in lv->status when _read_lvnames constructs lv.


Thanks.

Re: [linux-lvm] issue (or bug): when vgremove does removing lvmcache

[David Teigland]

On Sat, May 23, 2020 at 10:03:50PM +0800, heming.zhao@suse.com wrote:
> # vgremove test --force
>   Logical volume "main" successfully removed
>   Assertion failed: can't _pv_write non-orphan PV (in VG )
>   Failed to remove physical volume "/dev/sdb" from volume group "test"
>   Volume group "test" not properly removed

That has been fixed, looks like 56aadd7fe2a1d24043ea9d06543c29317ac1cc58,
although that may have changed since.
Dave

Re: [linux-lvm] issue (or bug): when vgremove does removing lvmcache

[heming.zhao]

Hello David,

very stupid mistake, I used master branch not not latest code to verify this issue.
after switching to latest code, this issue disappeared.

Thank you for your help.

On 5/27/20 5:30 AM, David Teigland wrote:
> On Sat, May 23, 2020 at 10:03:50PM +0800, heming.zhao@suse.com wrote:
>> # vgremove test --force
>>    Logical volume "main" successfully removed
>>    Assertion failed: can't _pv_write non-orphan PV (in VG )
>>    Failed to remove physical volume "/dev/sdb" from volume group "test"
>>    Volume group "test" not properly removed
> 
> That has been fixed, looks like 56aadd7fe2a1d24043ea9d06543c29317ac1cc58,
> although that may have changed since.
> Dave
>