[linux-lvm] [RFC PATCH] lvconvert: fix the size of log device when converting, from linear to mirror

[heming.zhao at suse.com]

Hello,

This patch looks good to me.

I give a little more explain.

Scenario:
there is a "--type mirror" LV with two legs: one leg is linear dev, another leg is striped devs.

the disks (LV) layout:
```
# lsblk
NAME               MAJ:MIN RM  SIZE RO TYPE MOUNTPOINT
sda                  8:0    0  200M  0 disk
└─vg1-lv1_mimage_0 254:4    0   96M  0 lvm
   └─vg1-lv1        254:2    0   96M  0 lvm
sdb                  8:16   0  200M  0 disk
└─vg1-lv1_mimage_1 254:5    0   96M  0 lvm
   └─vg1-lv1        254:2    0   96M  0 lvm
sdc                  8:32   0  200M  0 disk
└─vg1-lv1_mimage_1 254:5    0   96M  0 lvm
   └─vg1-lv1        254:2    0   96M  0 lvm
sdd                  8:48   0  200M  0 disk
├─vg1-lv1_mlog     254:3    0    4M  0 lvm
│ └─vg1-lv1        254:2    0   96M  0 lvm
└─vg1-lv1_mimage_1 254:5    0   96M  0 lvm
   └─vg1-lv1        254:2    0   96M  0 lvm

# dmsetup table
vg1-lv1: 0 196608 mirror disk 2 254:3 2048 2 254:4 0 254:5 0 1 handle_errors
vg1-lv1_mlog: 0 8192 linear 8:48 73728
vg1-lv1_mimage_1: 0 196608 striped 3 512 8:16 8192 8:32 8192 8:48 8192
vg1-lv1_mimage_0: 0 196608 linear 8:0 8192
```

Present lvm algorithm: "whole one leg size" / "striped devices number". (96MB / 3)
But kernel side dm-log module doesn't div "striped devices number". (96MB)
The kernel side algorithm is correct.
Let's image, in top layer, mirrored LV only know it has two same legs, disk log only related top
layer write IO.


It looks lvm2 algoritm is wrong from the first commit:
```
commit 16d9293bd7205c6ad2e932225aafca7513e540d9
Author: Alasdair Kergon <agk at redhat.com>
Date:   Mon Mar 1 20:00:20 2010 +0000

     Extend core allocation code in preparation for mirrored log areas.
```


The "--type mirror" is marked as deprecated many years ago. Does it block this patch to be accepted?

Maybe I miss somethings for the related code in lvm2 function _alloc_init()?

```_alloc_init at lib/metadata/lv_manip.c

     } else { //the "else" branch suites for other LV types except for "mirror"?
         ah->log_area_count = metadata_area_count;
         ah->log_len = !metadata_area_count ? 0 :
             _mirror_log_extents(ah->region_size, extent_size,
                         (existing_extents + new_extents) / ah->area_multiple);
     }

```

Thanks
Heming

On 7/6/21 10:57 AM, Zhong Lidong wrote:
> When converting a large linear device to mirror with --stripes parameter
> in lvconvert command, it will report the following error:
> 
> ioctl/libdm-iface.c:1895    device-mapper: reload ioctl on  (254:0)
> failed: Invalid argument
> 
> and the related kernel messages:
> 
> kernel: [ 2290.698171] device-mapper: dirty region log: log device 254:1
> too small: need 7078912 bytes
> kernel: [ 2290.698248] device-mapper: table: 254:0: mirror: Error
> creating mirror dirty log
> kernel: [ 2290.698251] device-mapper: ioctl: error adding target to table
> 
> Steps to reproduce this problem
> ~$:pvs
>    PV         VG Fmt  Attr PSize  PFree
>    /dev/loop0 vg lvm2 a--  10.00t 10.00t
>    /dev/loop1 vg lvm2 a--  10.00t 10.00t
>    /dev/loop2 vg lvm2 a--  10.00t 10.00t
>    /dev/loop3    lvm2 ---  10.00t 10.00t
>    /dev/loop4    lvm2 ---  10.00t 10.00t
>    /dev/loop5    lvm2 ---  10.00t 10.00t
> 
> ~$:lvcreate -nlv  -L27T vg
>    Logical volume "lv" created.
> 
> ~$:vgextend vg /dev/loop[3-5]
>    Volume group "vg" successfully extended
> 
> ~$:lvconvert  --type mirror --mirrors 1 --stripes 3 -I256  /dev/vg/lv
> /dev/loop3 /dev/loop4 /dev/loop5
>    device-mapper: reload ioctl on  (254:0) failed: Invalid argument
>    Failed to lock logical volume vg/lv.
> ~$: lvs
>    WARNING: Reading VG vg from disk because lvmetad metadata is invalid.
>    LV      VG Attr       LSize  Pool Origin Data%  Meta%  Move Log
> Cpy%Sync Convert
>    lv      vg -wi-a----- 27.00t
>    lv_mlog vg -wi-a-----  4.00m
> ~$: pvs
>    PV         VG Fmt  Attr PSize  PFree
>    /dev/loop0 vg lvm2 a--  10.00t     0
>    /dev/loop1 vg lvm2 a--  10.00t     0
>    /dev/loop2 vg lvm2 a--  10.00t  3.00t
>    /dev/loop3 vg lvm2 a--  10.00t 10.00t
>    /dev/loop4 vg lvm2 a--  10.00t 10.00t
>    /dev/loop5 vg lvm2 a--  10.00t 10.00t
> 
> Analysis: In kernel space the required log size depends on the whole
> target device size.But in user space, the log size is calculated from
> the whole target devices size divided by allocation area number, which
> leads to the mismatch here.
> 
> Signed-off-by: Lidong Zhong <lidong.zhong at suse.com>
> ---
>   lib/metadata/lv_manip.c | 3 +--
>   1 file changed, 1 insertion(+), 2 deletions(-)
> 
> diff --git a/lib/metadata/lv_manip.c b/lib/metadata/lv_manip.c
> index 7e66afd9c..e8d67c79b 100644
> --- a/lib/metadata/lv_manip.c
> +++ b/lib/metadata/lv_manip.c
> @@ -3730,8 +3730,7 @@ static struct alloc_handle *_alloc_init(struct
> cmd_context *cmd,
>   	} else {
>   		ah->log_area_count = metadata_area_count;
>   		ah->log_len = !metadata_area_count ? 0 :
> -			_mirror_log_extents(ah->region_size, extent_size,
> -					    (existing_extents + new_extents) / ah->area_multiple);
> +			_mirror_log_extents(ah->region_size, extent_size, existing_extents +
> new_extents);
>   	}
> 
>   	if (total_extents || existing_extents)
>