LVMTHIN(7) LVMTHIN(7)

lvmthin — LVM thin provisioning

Blocks in a standard lvm(8) Logical Volume (LV) are allocated when the LV is created, but blocks in a thin provisioned LV are allocated as they are written. Because of this, a thin provisioned LV has a virtual size that can be much larger than the available physical storage. The amount of physical storage provided for thin provisioned LVs can be increased later as the need arises.

Blocks in a standard LV are allocated (during creation) from the Volume Group (VG), but blocks in a thin LV are allocated (during use) from a "thin pool". The thin pool contains blocks of physical storage, and thin LV blocks reference blocks in the thin pool.

A special "thin pool LV" must be created before thin LVs can be created within it. A thin pool LV is created by combining two standard LVs: a data LV that will hold blocks for thin LVs, and a metadata LV that will hold metadata. Thin pool metadata is created and used by the dm-thin kernel module to track the data blocks used by thin LVs.

Snapshots of thin LVs are efficient because the data blocks common to a thin LV and any of its snapshots are shared. Snapshots may be taken of thin LVs or of other thin snapshots. Blocks common to recursive snapshots are also shared in the thin pool. There is no limit to or degradation from sequences of snapshots.

As thin LVs or snapshot LVs are written to, they consume data blocks in the thin pool. As free data blocks in the pool decrease, more physical space may need to be added to the pool. This is done by extending the thin pool with additional physical space from the VG. Removing thin LVs or snapshots from the thin pool can also make more space available. However, removing thin LVs is not always an effective way of freeing space in a thin pool because blocks may be shared by snapshots, and free blocks may be too fragmented to make available.

On-demand block allocation can cause thin LV blocks to be fragmented in the thin pool, which can cause reduced performance compared to standard fully provisioned LV.

Thin LV
A thin LV is an LVM logical volume for which storage is allocated on demand. As a thin LV is written, blocks are allocated from a thin pool to hold the data. A thin LV has a virtual size that can be larger than the physical space in the thin pool.

Thin Pool
A thin pool is a special LV containing physical extents from which thin LVs are allocated. The thin pool LV is not used as a block device, but the thin pool name is referenced when creating thin LVs. The thin pool LV must be extended with additional physical extents before it runs out of space. A thin pool has two hidden component LVs: one for holding thin data and another for holding thin metadata.

Thin Pool Data LV
A component of a thin pool that holds thin LV data. The data LV is a hidden LV with a _tdata suffix, and is not used directly. The physical size of the data LV is displayed as the thin pool size.

Thin Pool Metadata LV
A component of a thin pool that holds metadata for the dm-thin kernel module. dm-thin generates and uses this metadata to track data blocks used by thin LVs. The metadata LV is a hidden LV with a _tmeta suffix, and is not used directly.

Thin Snapshot
A thin snapshot is a thin LV that is created in reference to an existing thin LV or other thin snapshot. The thin snapshot initially refers to the same blocks as the existing thin LV. It acts as a point in time copy of the thin LV it referenced.

External Origin
A read-only LV that is used as a snapshot origin for thin LVs. Unwritten portions of the thin LVs are read from the external origin.

Thin Pool Creation

A thin pool can be created with the lvcreate command. The data and metadata component LVs are each allocated from the VG, and combined into a thin pool. The lvcreate -L|--size will be the size of the thin pool data LV, and the size of the metadata LV will be calculated automatically (or, can be optionally specified with --poolmetadatasize.)

$ lvcreate --type thin-pool -n ThinPool -L Size VG

Thin Pool Conversion

For a customized thin pool layout, data and metadata LVs can be created separately, and then combined into a thin pool with lvconvert. This allows specific LV types, or specific devices, to be used for data/metadata LVs. Combining the data and metadata LVs into a thin pool erases the content of both LVs. The resulting thin pool takes the name and size of the data LV. (If a metadata LV is not specified, lvconvert will automatically create one to use in the thin pool.)

$ lvcreate -n DataLV -L Size VG DataDevices
$ lvcreate -n MetadataLV -L MetadataSize VG MetadataDevices
$ lvconvert --type thin-pool --poolmetadata MetadataLV VG/DataLV

(DataLV would now be referred to as ThinPool, and can be used for creating thin LVs.)

Thin LV Creation

Thin LVs are created in a thin pool, and are created with a virtual size using the option -V|--virtualsize. The virtual size may be larger than the physical space available in the thin pool.

$ lvcreate --type thin -n ThinLV -V VirtualSize --thinpool ThinPool VG

Thin Snapshot Creation

Snapshots of thin LVs are thin LVs themselves, but the snapshot LV initially refers to the same blocks as the origin thin LV. The origin thin LV and its snapshot thin LVs will diverge as either are written. The origin thin LV can be removed without affecting snapshots that reference it. Snapshots can be taken of thin LVs that were themselves created as snapshots. (A size option must not be used when creating a thin snapshot, otherwise a COW snapshot will be created.)

$ lvcreate --snapshot -n SnapLV VG/ThinLV

For active thin pool LVs, the 'lvs' command displays "Data%" (-o data_percent) and "Meta%" (-o metadata_percent). Data percent is the percent of space in the data LV that is currently used by thin LVs. Metadata percent is the percent of space in the metadata LV that is currently used by the dm-thin module. The thin pool should be extended before either of these values reach 100%.

$ lvs -o data_percent VG/ThinPool
$ lvs -o metadata_percent VG/ThinPool

Thin Pool Extension

When lvextend is run on a thin pool, it will extend the internal data LV by the specified amount, and the internal metadata LV will also be extended, if needed, relative to the new data size.

$ lvextend --size Size VG/ThinPool

A new metadata size can be requested when extending the thin pool data.

$ lvextend --size Size --poolmetadatasize MetadataSize VG/ThinPool

The metadata size can be extended without extending the data size.

$ lvextend --poolmetadatasize MetadataSize VG/ThinPool

The internal data or metadata LV can be extended by name.

$ lvextend -L Size VG/ThinPool_tdata
$ lvextend -L MetadataSize VG/ThinPool_tmeta

It is important to extend a thin pool before it runs out of space, otherwise it may be damaged, and difficult or impossible to repair. LVM can be configured so that dmeventd automatically extends thin pools when they run low on space. Free extents must be available in the VG to use for extending the thin pools.

dmeventd is usually started by the lvm2-monitor service. dmeventd receives notifications from the kernel indicating when thin pool data or metadata are becoming full. In response, dmeventd runs the command "lvextend --use-policies VG/ThinPool", which compares the current usage of data and metadata with the autoextend threshold. The data LV and/or metadata LV may be extended in response. System messages will show when these extensions have happened.

To enable thin pool automatic extension, set lvm.conf:

thin_pool_autoextend_threshold
Extend the thin pool when the current usage reaches this percentage. The chosen value should depend on the rate at which new data may be written. If space is consumed more quickly, then a lower threshold will provide dmeventd and lvextend more time to react and extend the pool. The minimum is 50. Setting to 100 disables autoextend.
thin_pool_autoextend_percent
A thin pool is extended by this percent of its current size.

The thin pool itself must be monitored by dmeventd to be automatically extended. When activating a thin pool, lvm normally requests monitoring by dmeventd. To verify this, run:

$ lvs -o+seg_monitor VG/ThinPool

To begin monitoring a thin pool in dmeventd:

$ lvchange --monitor y VG/ThinPool

A thin LV that is created as a snapshot is given the "skip activation" property. It is reported with lvs -o skip_activation, or 'k' in the tenth lv_attr. This property causes vgchange -ay and lvchange -ay commands to skip activating the thin LV unless the -K|--ignoreactivationskip option is also set.

$ lvchange -ay -K VG/SnapLV

The skip activation property on a thin LV can be cleared, so that -K is not required to activate it (or enabled so -K is required.)

$ lvchange --setactivationskip y|n VG/SnapLV

To configure the "skip activation" setting that lvcreate applies to new snapshots, set lvm.conf:
auto_set_activation_skip

Thick LV to Thin LV Conversion

A thick LV (e.g. linear, striped) can be converted to a thin LV in a new thin pool. The new thin pool is created using the existing thick LV as thin pool data. New thin pool metadata is generated and written to a new metadata LV. The new thin LV references the original thick data now located in the thin pool data LV. Note: This conversion cannot be reversed; the thin volume cannot be reverted back to the thick LV.

$ lvconvert --type thin VG/ThickLV

(ThickLV would now be referred to as ThinLV, and a new thin pool will exist named ThinLV_tpool0.)

After the conversion, the resulting thin LV and thin pool will look somewhat different from ordinary thin LVs/pools: the new thin LV will be fully provisioned in the thin pool, and the thin pool data usage will be 100%. The thin pool will require extension before new thin LVs or snapshots are used.

Thin Pool on LVM RAID

Thin pool data or metadata component LVs can use LVM RAID by first creating RAID LVs for data and/or metadata component LVs, and then converting these RAID LVs into a thin pool.

$ lvcreate --type raidN -n DataLV -L Size VG DataDevices
$ lvcreate --type raidN -n MetadataLV -L MetadataSize VG MetadataDevices
$ lvconvert --type thin-pool --poolmetadata MetadataLV VG/DataLV

(DataLV would now be referred to as ThinPool, and can be used for creating thin LVs.)

To use MD RAID instead of LVM RAID, create linear data/metadata LVs on MD devices, and refer to the MD devices for DataDevices/MetadataDevices.

Thin pool data can be compressed and deduplicated using VDO. Data for all thin LVs in the thin pool will be compressed and deduplicated using the dm-vdo module.

$ lvcreate --type thin-pool -n ThinPool -L Size --pooldatavdo y VG

Or, convert an existing LV (e.g. linear, striped) into a thin-pool that uses VDO compression/deduplication for thin data. Existing content on the LV will be erased.

$ lvconvert --type thin-pool --pooldatavdo y VG/LV

(LV would now be referred to as ThinPool, and can be used for creating thin LVs.)

Thin Pool and Thin LV Combined Creation

One command can be used to create a new thin pool with a new thin LV.

$ lvcreate --type thin -n ThinLV -V VirtualSize \
	--thinpool ThinPool -L ThinPoolSize VG

First, a new thin pool is created:
Thin Pool name is from --thinpool ThinPool
Thin Pool size is from -L|--size ThinPoolSize

Second, a new thin LV is created:
Thin LV name is from -n|--name ThinLV
Thin LV size is from -V|--virtualsize VirtualSize

Other thin LVs can then be created in the thin pool using standard lvcreate commands for thin LVs.

Thin Snapshot Creation of an External Origin

Thin snapshots are typically taken of other thin LVs within the same thin pool. But, it is also possible to create a thin snapshot of an external LV (e.g. linear, striped, thin LV in another thin pool.) The external LV must be read-only (lvchange --permission r) and inactive to be used as a thin external origin. Writes to the thin snapshot LV are stored in its thin pool, and unwritten parts are read from the external origin. One external origin LV can be used for multiple thin snapshots.

$ lvcreate --snapshot -n SnapLV --thinpool ThinPool VG/ExternalOrigin

In this case, an existing, non-thin LV is converted to a read-only external origin, and a new thin LV is created as a snapshot of that external origin. The new thin LV is given the name of the existing LV, and the existing LV is given a new name from --originname.

Unwritten portions of the new thin LV are read from the external origin. If the thin LV is removed, the external origin LV can be used again in read/write mode. Thus, the thin LV can be seen as a snapshot of the original volume.

$ lvconvert --type thin --thinpool ThinPool --originname ExtOrigin VG/LV

The existing LV argument is renamed ExtOrigin, and the new thin LV has the name of the existing LV.

Thin Snapshot Merge

A thin snapshot can be merged into its origin thin LV. The result of a snapshot merge is that the origin thin LV takes the content of the snapshot LV, and the snapshot LV is removed. Any content that was unique to the origin thin LV is lost after the merge.

Because a merge changes the content of an LV, it cannot be done while the LVs are open, e.g. mounted. If a merge is initiated while the LVs are open, the effect of the merge is delayed until the origin thin LV is next activated.

$ lvconvert --merge VG/SnapLV

Thin Pool Creation

# lvcreate --type thin-pool -n pool0 -L 500M vg
# lvs -a vg
  LV              VG Attr       LSize   Data%  Meta%  
  [lvol0_pmspare] vg ewi-------   4.00m              
  pool0           vg twi-a-tz-- 500.00m 0.00   10.84 
  [pool0_tdata]   vg Twi-ao---- 500.00m              
  [pool0_tmeta]   vg ewi-ao----   4.00m              

Thin Pool Conversion

# lvcreate -n pool0 -L 500M vg
# lvcreate -n pool0_meta -L 100M vg
# lvconvert --type thin-pool --poolmetadata pool0_meta vg/pool0
# lvs -a vg
  LV              VG Attr       LSize   Data%  Meta%  
  [lvol0_pmspare] vg ewi------- 100.00m              
  pool0           vg twi-a-tz-- 500.00m 0.00   10.04 
  [pool0_tdata]   vg Twi-ao---- 500.00m              
  [pool0_tmeta]   vg ewi-ao---- 100.00m              

Thin LV Creation

# lvcreate --type thin-pool -n pool0 -L 500M vg
# lvcreate --type thin -n vol -V 1G --thinpool pool0 vg
# lvs -a vg
  LV              VG Attr       LSize   Pool  Data%  Meta%  
  [lvol0_pmspare] vg ewi-------   4.00m                    
  pool0           vg twi-aotz-- 500.00m       0.00   10.94 
  [pool0_tdata]   vg Twi-ao---- 500.00m                    
  [pool0_tmeta]   vg ewi-ao----   4.00m                    
  vol             vg Vwi-a-tz--   1.00g pool0 0.00         

Thin Snapshot Creation

# lvcreate --type thin-pool -n pool0 -L 500M vg
# lvcreate --type thin -n vol -V 1G --thinpool pool0 vg
# lvcreate --snapshot -n snap1 vg/vol
# lvcreate --snapshot -n snap2 vg/snap1
# lvs -a vg
  LV              VG Attr       LSize   Pool  Origin Data%  Meta%  
  [lvol0_pmspare] vg ewi-------   4.00m                           
  pool0           vg twi-aotz-- 500.00m              0.00   10.94 
  [pool0_tdata]   vg Twi-ao---- 500.00m                           
  [pool0_tmeta]   vg ewi-ao----   4.00m                           
  snap1           vg Vwi---tz-k   1.00g pool0 vol                 
  snap2           vg Vwi---tz-k   1.00g pool0 snap1               
  vol             vg Vwi-a-tz--   1.00g pool0        0.00         

Thin Pool Extension

# lvcreate --type thin-pool -n pool0 -L 500M vg
# lvextend -L+100M vg/pool0
# lvs -a vg
  LV              VG Attr       LSize   Data%  Meta%  
  [lvol0_pmspare] vg ewi-------   4.00m              
  pool0           vg twi-a-tz-- 600.00m 0.00   10.84 
  [pool0_tdata]   vg Twi-ao---- 600.00m              
  [pool0_tmeta]   vg ewi-ao----   4.00m              
# lvextend -L+100M --poolmetadatasize 8M vg/pool0
# lvs -a vg
  LV              VG Attr       LSize   Data%  Meta%  
  [lvol0_pmspare] vg ewi-------   8.00m              
  pool0           vg twi-a-tz-- 700.00m 0.00   10.40 
  [pool0_tdata]   vg Twi-ao---- 700.00m              
  [pool0_tmeta]   vg ewi-ao----   8.00m              

Thick LV to Thin LV Conversion

# lvcreate -n vol -L500M vg
# lvconvert --type thin vg/vol
# lvs -a vg
  LV                 VG Attr       LSize   Pool       Data%  Meta%  
  [lvol0_pmspare]    vg ewi-------   4.00m                         
  vol                vg Vwi-a-tz-- 500.00m vol_tpool0 100.00       
  vol_tpool0         vg twi-aotz-- 500.00m            100.00 14.06 
  [vol_tpool0_tdata] vg Twi-ao---- 500.00m                         
  [vol_tpool0_tmeta] vg ewi-ao----   4.00m                         
# lvextend -L1G vg/vol
# lvs -a vg
  LV                 VG Attr       LSize    Pool       Data%  Meta%  
  [lvol0_pmspare]    vg ewi-------    4.00m                         
  vol                vg Vwi-a-tz--    1.00g vol_tpool0 48.83        
  vol_tpool0         vg twi-aotz-- 1000.00m            50.00  14.06 
  [vol_tpool0_tdata] vg Twi-ao---- 1000.00m                         
  [vol_tpool0_tmeta] vg ewi-ao----    4.00m                         

(Extending the virtual size of the thin LV triggered autoextend of the thin pool.)

Thin Pool on LVM RAID

# lvcreate --type raid1 -n pool0 -m1 -L500M vg
# lvcreate --type raid1 -n pool0_meta -m1 -L8M vg
# lvs -a vg
  LV                    VG Attr       LSize   Cpy%Sync 
  pool0                 vg rwi-a-r--- 500.00m 100.00  
  pool0_meta            vg rwi-a-r---   8.00m 100.00  
  [pool0_meta_rimage_0] vg iwi-aor---   8.00m         
  [pool0_meta_rimage_1] vg iwi-aor---   8.00m         
  [pool0_meta_rmeta_0]  vg ewi-aor---   4.00m         
  [pool0_meta_rmeta_1]  vg ewi-aor---   4.00m         
  [pool0_rimage_0]      vg iwi-aor--- 500.00m         
  [pool0_rimage_1]      vg iwi-aor--- 500.00m         
  [pool0_rmeta_0]       vg ewi-aor---   4.00m         
  [pool0_rmeta_1]       vg ewi-aor---   4.00m         
# lvconvert --type thin-pool --poolmetadata pool0_meta vg/pool0
# lvs -a vg
  LV                     VG Attr       LSize   Data%  Meta%  Cpy%Sync 
  [lvol0_pmspare]        vg ewi-------   8.00m                       
  pool0                  vg twi-a-tz-- 500.00m 0.00   10.40          
  [pool0_tdata]          vg rwi-aor--- 500.00m               100.00  
  [pool0_tdata_rimage_0] vg iwi-aor--- 500.00m                       
  [pool0_tdata_rimage_1] vg iwi-aor--- 500.00m                       
  [pool0_tdata_rmeta_0]  vg ewi-aor---   4.00m                       
  [pool0_tdata_rmeta_1]  vg ewi-aor---   4.00m                       
  [pool0_tmeta]          vg ewi-aor---   8.00m               100.00  
  [pool0_tmeta_rimage_0] vg iwi-aor---   8.00m                       
  [pool0_tmeta_rimage_1] vg iwi-aor---   8.00m                       
  [pool0_tmeta_rmeta_0]  vg ewi-aor---   4.00m                       
  [pool0_tmeta_rmeta_1]  vg ewi-aor---   4.00m                       

# lvcreate --type thin-pool -n pool0 -L5G --pooldatavdo y vg
# lvs -a vg
  LV                   VG Attr       LSize Pool         Data%  Meta%  
  [lvol0_pmspare]      vg ewi------- 8.00m                           
  pool0                vg twi-a-tz-- 5.00g              0.00   10.64 
  [pool0_tdata]        vg vwi-aov--- 5.00g pool0_vpool0 0.00         
  [pool0_tmeta]        vg ewi-ao---- 8.00m                           
  pool0_vpool0         vg dwi------- 5.00g              60.03        
  [pool0_vpool0_vdata] vg Dwi-ao---- 5.00g                           

# lvcreate -n pool0 -L5G vg
# lvconvert --type thin-pool --pooldatavdo y vg/pool0
# lvs -a vg
  LV                   VG Attr       LSize Pool         Data%  Meta%  
  [lvol0_pmspare]      vg ewi------- 8.00m                           
  pool0                vg twi-a-tz-- 5.00g              0.00   10.64 
  [pool0_tdata]        vg vwi-aov--- 5.00g pool0_vpool0 0.00         
  [pool0_tmeta]        vg ewi-ao---- 8.00m                           
  pool0_vpool0         vg dwi------- 5.00g              60.03        
  [pool0_vpool0_vdata] vg Dwi-ao---- 5.00g                           

Thin Snapshot Creation of an External Origin

# lvcreate -n vol -L 500M vg
# lvchange --permission r vg/vol
# lvchange -an vg/vol
# lvcreate --type thin-pool -n pool0 -L 500M vg
# lvcreate --snapshot -n snap --thinpool pool0 vg/vol
# lvs -a vg
  LV              VG Attr       LSize   Pool  Origin Data%  Meta%  
  [lvol0_pmspare] vg ewi-------   4.00m                           
  pool0           vg twi-aotz-- 500.00m              0.00   10.94 
  [pool0_tdata]   vg Twi-ao---- 500.00m                           
  [pool0_tmeta]   vg ewi-ao----   4.00m                           
  snap            vg Vwi-a-tz-- 500.00m pool0 vol    0.00         
  vol             vg ori------- 500.00m                           

Thin Pool and Thin LV Combined Creation

# lvcreate --type thin -n vol -V 1G --thinpool pool0 -L500M vg
# lvs -a vg
  LV              VG Attr       LSize   Pool  Data%  Meta%  
  [lvol0_pmspare] vg ewi-------   4.00m                    
  pool0           vg twi-aotz-- 500.00m       0.00   10.94 
  [pool0_tdata]   vg Twi-ao---- 500.00m                    
  [pool0_tmeta]   vg ewi-ao----   4.00m                    
  vol             vg Vwi-a-tz--   1.00g pool0 0.00         

Thin Snapshot Merge

# lvcreate --type thin-pool -n pool0 -L500M vg
# lvcreate --type thin -n vol -V 1G --thinpool pool0 vg
# lvcreate --snapshot -n snap vg/vol
# lvs -a vg
  LV              VG Attr       LSize   Pool  Origin Data%  Meta%  
  [lvol0_pmspare] vg ewi-------   4.00m                           
  pool0           vg twi-aotz-- 500.00m              0.00   10.94 
  [pool0_tdata]   vg Twi-ao---- 500.00m                           
  [pool0_tmeta]   vg ewi-ao----   4.00m                           
  snap            vg Vwi---tz-k   1.00g pool0 vol                 
  vol             vg Vwi-a-tz--   1.00g pool0        0.00         
# lvconvert --merge vg/snap
# lvs -a vg
  LV              VG Attr       LSize   Pool  Data%  Meta%  
  [lvol0_pmspare] vg ewi-------   4.00m                    
  pool0           vg twi-aotz-- 500.00m       0.00   10.94 
  [pool0_tdata]   vg Twi-ao---- 500.00m                    
  [pool0_tmeta]   vg ewi-ao----   4.00m                    
  vol             vg Vwi-a-tz--   1.00g pool0 0.00         

# lvcreate --type thin-pool -n pool0 -L500M vg
# lvcreate --type thin -n vol -V 1G --thinpool pool0 vg
# mkfs.xfs /dev/vg/vol
# mount /dev/vg/vol /mnt
# touch /mnt/file1 /mnt/file2 /mnt/file3
# lvcreate --snapshot -n snap vg/vol
# mount /dev/vg/snap /snap -o nouuid
# touch /snap/file4 /snap/file5 /snap/file6
# ls /snap
file1  file2  file3  file4  file5  file6
# ls /mnt
file1  file2  file3
# lvconvert --merge vg/snap
  Logical volume vg/snap contains a filesystem in use.
  Delaying merge since snapshot is open.
  Merging of thin snapshot vg/snap will occur on next activation of vg/vol.
# umount /snap
# umount /mnt
# lvchange -an vg/vol
# lvs -a vg
  LV              VG Attr       LSize   Pool  Origin Data%  Meta%
  [lvol0_pmspare] vg ewi-------   4.00m
  pool0           vg twi-aotz-- 500.00m              13.36  11.62
  [pool0_tdata]   vg Twi-ao---- 500.00m
  [pool0_tmeta]   vg ewi-ao----   4.00m
  [snap]          vg Swi---tz-k   1.00g pool0 vol
  vol             vg Owi---tz--   1.00g pool0
# lvchange -ay vg/vol
# lvs -a vg
  LV              VG Attr       LSize   Pool  Data%  Meta%
  [lvol0_pmspare] vg ewi-------   4.00m
  pool0           vg twi-aotz-- 500.00m       12.94  11.43
  [pool0_tdata]   vg Twi-ao---- 500.00m
  [pool0_tmeta]   vg ewi-ao----   4.00m
  vol             vg Vwi-a-tz--   1.00g pool0 6.32
# mount /dev/vg/vol /mnt
# ls /mnt
file1  file2  file3  file4  file5  file6

Placing the thin pool data LV and metadata LV on separate physical devices will improve performance. Faster, redundant devices for metadata is also recommended. To best customize the data and metadata LVs, create them separately and then combine them into a thin pool with lvconvert.

To configure lvcreate behavior to place thin pool data and metadata on separate devices, set lvm.conf:
thin_pool_metadata_require_separate_pvs

The first time a thin pool LV is created, lvm will create a spare metadata LV in the VG. This behavior can be controlled with the option --poolmetadataspare y|n. To create the pmspare ("pool metadata spare") LV, lvm first creates an LV with a default name, e.g. lvol0, and then converts this LV to a hidden LV with the _pmspare suffix, e.g. lvol0_pmspare.

One pmspare LV is kept in a VG to be used for any thin pool.

The pmspare LV cannot be created explicitly, but may be removed explicitly.

The "Thin Pool Metadata check and repair" section describes the use of the pmspare LV.

If thin pool metadata is damaged, it may be repairable. Checking and repairing thin pool metadata is analogous to running fsck/repair on a file system. Thin pool metadata is compact, so even small areas of damage or corruption can result in significant data loss. Resilient storage for thin pool metadata can have extra value.

When a thin pool LV is activated, lvm runs the thin_check(8) command to check the correctness of the metadata on the pool metadata LV. To configure thin_check use, location or options used by lvm, set lvm.conf:

thin_check_executable
The location of the program. Setting to an empty string ("") disables running thin_check by lvm. This is not recommended.

thin_check_options
Controls the command options that lvm will use when running thin_check.

If thin_check finds a problem with the metadata, the thin pool LV is not activated, and the thin pool metadata needs to be repaired.

Simple repair commands are not always successful. Advanced repair may require editing thin pool metadata and lvm metadata. Newer versions of the kernel and lvm tools may be more successful at repair. Report the details of damaged thin metadata to get the best advice on recovery.

Command to repair a thin pool:

$ lvconvert --repair VG/ThinPool

Repair performs the following steps:

1
Creates a new, repaired copy of the metadata.
lvconvert runs the thin_repair(8) command to read damaged metadata from the existing pool metadata LV, and writes a new repaired copy to the VG's pmspare LV.
2
Replaces the thin pool metadata LV.
If step 1 is successful, the thin pool metadata LV is replaced with the pmspare LV containing the corrected metadata. The previous thin pool metadata LV, containing the damaged metadata, becomes visible with the new name ThinPool_metaN (where N is 0,1,...).

If the repair works, the thin pool LV and its thin LVs can be activated. The user should verify that each thin LV in the thin pool can be successfully activated, and then verify the integrity of the file system on each thin LV (e.g. using fsck or other tools.) Once the thin pool is considered fully recovered, the ThinPool_metaN LV containing the original, damaged metadata can be manually removed to recovery the space.

If the repair fails, the original, unmodified ThinPool_metaN LV should be preserved for support, or more advanced recovery methods. Data from thin LVs may ultimately be unrecoverable.

If metadata is manually restored with thin_repair directly, the pool metadata LV can be manually swapped with another LV containing new metadata:

$ lvconvert --thinpool VG/ThinPool --poolmetadata VG/NewMetadataLV

Removing a thin LV and its related snapshots returns the blocks they used to the thin pool. These blocks will be reused for other thin LVs and snapshots.

Removing a thin pool LV removes both the data LV and metadata LV and returns the space to the VG.

lvremove of thin pool LVs, thin LVs and snapshots cannot be reversed with vgcfgrestore.

vgcfgbackup does not back up thin pool metadata.

Removing files in a file system on a thin LV does not generally return free space to the thin pool, because file systems are not usually mounted with the discard mount option (due to the performance penalty.)

Manually running the fstrim command can return space from a thin LV back to the thin pool that had been used by removed files. This is only effective for entire thin pool chunks that have become unused (unused file system areas may not cover an entire chunk.) Thin snapshots also keep thin pool chunks from being freed. fstrim uses discards and will have no effect if the thin pool is configured to ignore discards.

Example
A thin pool has 10G of physical data space, and a thin LV has a virtual size of 100G. Writing a 1G file to the file system reduces the free space in the thin pool by 10% and increases the virtual usage of the file system by 1%. Removing the 1G file restores the virtual 1% to the file system, but does not restore the physical 10% to the thin pool. The fstrim command restores the physical space to the thin pool.

# lvs -a -oname,attr,size,pool_lv,origin,data_percent,metadata_percent vg
  LV            Attr       LSize   Pool  Origin Data%  Meta%
  pool0         twi-a-tz--  10.00g              47.01  21.03
  thin1         Vwi-aotz-- 100.00g pool0         2.70
# df -h /mnt/X
Filesystem            Size  Used Avail Use% Mounted on
/dev/mapper/vg-thin1   99G  1.1G   93G   2% /mnt/X
# dd if=/dev/zero of=/mnt/X/1Gfile bs=4096 count=262144; sync
# lvs
  pool0         vg   twi-a-tz--  10.00g         57.01  25.26
  thin1         vg   Vwi-aotz-- 100.00g pool0    3.70
# df -h /mnt/X
/dev/mapper/vg-thin1   99G  2.1G   92G   3% /mnt/X
# rm /mnt/X/1Gfile
# lvs
  pool0         vg   twi-a-tz--  10.00g         57.01  25.26
  thin1         vg   Vwi-aotz-- 100.00g pool0    3.70
# df -h /mnt/X
/dev/mapper/vg-thin1   99G  1.1G   93G   2% /mnt/X
# fstrim -v /mnt/X
# lvs
  pool0         vg   twi-a-tz--  10.00g         47.01  21.03
  thin1         vg   Vwi-aotz-- 100.00g pool0    2.70

When properly managed, thin pool data space should be extended before it is all used (see sections on extending a thin pool automatically and manually.)

However, if a thin pool does run out of space, the behavior of the full thin pool can be configured with the "when full" property, reported with lvs -o whenfull. The "when full" property can be set to "error" or "queue". When set to "error", a full thin pool will immediately return errors for writes. When set to "queue", writes are queued for a period of time.

Display the current "when full" setting:

$ lvs -o whenfull VG/ThinPool

Set the "when full" property to "error":

$ lvchange --errorwhenfull y VG/ThinPool

Set the "when full" property to "queue":

$ lvchange --errorwhenfull n VG/ThinPool

To configure the value that will be assigned to new thin pools, set lvm.conf:
error_when_full

The whenfull setting does not effect the monitoring and autoextend settings, and the monitoring/autoextend settings do not effect the whenfull setting. It is only when monitoring/autoextend are not effective that the thin pool becomes full and the whenfull setting is applied.

— queue when full —

The default is to queue writes for a period of time when the thin pool becomes full. Writes to thin LVs are accepted and queued, with the expectation that pool data space will be extended soon. Once data space is extended, the queued writes will be processed, and the thin pool will return to normal operation.

While waiting to be extended, the thin pool will queue writes for up to 60 seconds (the default). If data space has not been extended after this time, the queued writes will return an error to the caller, e.g. the file system. This can result in file system damage that requires repair. When a thin pool returns errors for writes to a thin LV, any file system is subject to losing unsynced user data.

The 60 second timeout can be changed or disabled with the dm-thin-pool kernel module option no_space_timeout. This option sets the number of seconds that thin pools will queue writes. If set to 0, writes will not time out. Disabling timeouts can result in the system running out of resources, memory exhaustion, hung tasks, and deadlocks. (The timeout applies to all thin pools on the system.)

— error when full —

Writes to thin LVs immediately return an error, and no writes are queued. This can result in file system damage that requires repair.

— data percent —

When data space is exhausted, the lvs command displays 100 under Data% for the thin pool LV:

# lvs -o name,data_percent vg/pool0
  LV     Data%
  pool0  100.00

— causes —

A thin pool may run out of data space for any of the following reasons:

Automatic extension of the thin pool is disabled, and the thin pool is not manually extended. (Disabling automatic extension is not recommended.)
The dmeventd daemon is not running and the thin pool is not manually extended. (Disabling dmeventd is not recommended.)
Automatic extension of the thin pool is too slow given the rate of writes to thin LVs in the pool. (This can be addressed by tuning the thin_pool_autoextend_threshold and thin_pool_autoextend_percent.)
The VG does not have enough free blocks to extend the thin pool.

If thin pool metadata space is exhausted (or a thin pool metadata operation fails), errors will be returned for IO operations on thin LVs.

When metadata space is exhausted, the lvs command displays 100 under Meta% for the thin pool LV:

# lvs -o name,metadata_percent vg/pool0
  LV    Meta%
  pool0 100.00

The same reasons for thin pool data space exhaustion apply to thin pool metadata space.

Metadata space exhaustion can lead to inconsistent thin pool metadata and inconsistent file systems, so the response requires offline checking and repair.

1.
Deactivate the thin pool LV, or reboot the system if this is not possible.
2.
Repair thin pool with lvconvert --repair.
See "Thin Pool Metadata check and repair".
3.
Extend pool metadata space with lvextend --poolmetadatasize.
See "Thin Pool Extension".
4.
Check and repair file system.

It can be useful for different thin pools to have different thin pool settings like autoextend thresholds and percents. To change lvm.conf values on a per-VG or per-LV basis, attach a "profile" to the VG or LV. A profile is a collection of config settings, saved in a local text file (using the lvm.conf format). lvm looks for profiles in the profile_dir directory, e.g. /etc/lvm/profile/. Once attached to a VG or LV, lvm will process the VG or LV using the settings from the attached profile. A profile is named and referenced by its file name.

To use a profile to customize the lvextend settings for an LV:

Create a file containing settings, saved in profile_dir.
For the profile_dir location, run:
$ lvmconfig config/profile_dir
Attach the profile to an LV, using the command:
$ lvchange --metadataprofile ProfileName VG/ThinPool
Extend the LV using the profile settings:
$ lvextend --use-policies VG/ThinPool

Example

# lvmconfig config/profile_dir
profile_dir="/etc/lvm/profile"
# cat /etc/lvm/profile/pool0extend.profile
activation {
thin_pool_autoextend_threshold=50
thin_pool_autoextend_percent=10
}
# lvchange --metadataprofile pool0extend vg/pool0
# lvextend --use-policies vg/pool0

Notes

A profile is attached to a VG or LV by name, where the name references a local file in profile_dir. If the VG is moved to another machine, the file with the profile also needs to be moved.
Only certain settings can be used in a VG or LV profile, see:
$ lvmconfig --type profilable-metadata
An LV without a profile of its own will inherit the VG profile.
Remove a profile from an LV using the command:
$ lvchange --detachprofile VG/ThinPool
Commands can also have profiles applied to them. The settings that can be applied to a command are different than the settings that can be applied to a VG or LV. See lvmconfig --type profilable-command. To apply a profile to a command, write a profile, save it in the profile directory, and run the command using the option: --commandprofile ProfileName.

The "zero" property of a thin pool determines if chunks are overwritten with zeros when they are provisioned for a thin LV. The current setting is reported with lvs -o zero (displaying "zero" or "1" when zeroing is enabled), or 'z' in the eighth lv_attr. The option -Z|--zero is used to specify the zeroing mode.

Create a thin pool with zeroing mode:

$ lvcreate --type thin-pool -n ThinPool -L Size -Z y|n VG

Change the zeroing mode of an existing thin pool:

$ lvchange -Z y|n VG/ThinPool

If zeroing mode is changed from "n" to "y", previously provisioned blocks are not zeroed.

Provisioning of large zeroed chunks reduces performance.

To configure the zeroing mode used for new thin pools when not specified on the command line, set lvm.conf:
thin_pool_zero

The "discards" property of a thin pool determines how discard requests are handled. The current setting is reported with lvs -o discards. The option --discards is used to specify the discards mode.

Possible discard modes:

ignore: Ignore any discards that are received.

nopassdown: Process any discards in the thin pool itself, and allow the newly unused chunks to be used for new data.

passdown: Process discards in the thin pool (as with nopassdown), and pass the discards down the the underlying device. This is the default mode.

Create a thin pool with a specific discards mode:

$ lvcreate --type thin-pool -n ThinPool -L Size
--discards ignore|nopassdown|passdown VG

Change the discards mode of an existing thin pool:

$ lvchange --discards ignore|nopassdown|passdown VG/ThinPool

To configure the discards mode used for new thin pools when not specified on the command line, set lvm.conf:
thin_pool_discards

Discards can have an adverse impact on performance, see the fstrim section for more information.

A thin pool allocates physical storage for thin LVs in units of "chunks". The current chunk size of a thin pool is reported with lvs -o chunksize. The option --chunksize is used to specify the value for a new thin pool (default units are KiB.) The value must be a multiple of 64KiB, between 64KiB and 1GiB.

When a thin pool is used primarily for the thin provisioning feature, a larger value is optimal. To optimize for many snapshots, a smaller value reduces copying time and consumes less space.

To configure the chunk size used for new thin pools when not specified on the command line, set lvm.conf:
thin_pool_chunk_size

The default value is shown by:

$ lvmconfig --type default allocation/thin_pool_chunk_size

The amount of thin pool metadata depends on how many blocks are shared between thin LVs (i.e. through snapshots). A thin pool with many snapshots may need a larger metadata LV. Thin pool metadata LV sizes can be from 2MiB to approximately 16GiB.

When an LVM command automatically creates a thin pool metadata LV, the size is specified with the --poolmetadatasize option. When this option is not given, LVM automatically chooses a size based on the data size and chunk size.

It can be hard to predict the amount of metadata space that will be needed, so it is recommended to start with a size of 1GiB which should be enough for all practical purposes. A thin pool metadata LV can later be manually or automatically extended if needed.

(For purposes of backward compatibility, lvm.conf setting allocation/thin_pool_crop_metadata controls cropping the metadata LV size to 15.81GiB to be backward compatible with older versions of lvm. With cropping, there can be problems with volumes above this size when used with thin tools, i.e. thin_repair. Cropping should be enabled only when compatibility is required.)

Mounting an XFS file system on a new snapshot LV requires attention to the file system's log state and uuid. On the snapshot LV, the xfs log will contain a dummy transaction, and the xfs uuid will match the uuid from the file system on the origin LV.

If the snapshot LV is writable, mounting will recover the log to clear the dummy transaction, but will require skipping the uuid check:

# mount /dev/VG/SnapLV /mnt -o nouuid

After the first mount with the above approach, the UUID can subsequently be changed using:

# xfs_admin -U generate /dev/VG/SnapLV
# mount /dev/VG/SnapLV /mnt

Once the UUID has been changed, the mount command will no longer require the nouuid option.
If the snapshot LV is readonly, the log recovery and uuid check need to be skipped while mounting readonly:

# mount /dev/VG/SnapLV /mnt -o ro,nouuid,norecovery

lvm(8), lvm.conf(5), lvmconfig(8), lvcreate(8), lvconvert(8), lvchange(8), lvextend(8), lvremove(8), lvs(8),

thin_check(8), thin_dump(8), thin_repair(8), thin_restore(8),

vdoformat(8), vdostats(8)

LVM TOOLS 2.03.27(2) (2024-10-02) Red Hat, Inc