Details from the github ticket.

I have an old btrfs / filesystem on single-disk which I had outgrew.

I added a new disk and wanted to instead having DUP for metadata and system convert it to raid1.

The data I wished to retain having in single mode as it's not so critical.

so I did this command:

  balance: start -dusage=100 -mconvert=raid1 -sconvert=raid1

I did get a kernel error and filesystem went ro.

Attached kernel log.

% uname -a
Linux staropramen 5.10.9-arch1-1 #1 SMP PREEMPT Tue, 19 Jan 2021 22:06:06 +0000 x86_64 GNU/Linux

% btrfs --version
btrfs-progs v5.10

% sudo btrfs filesystem usage /
[sudo] password for markus:
Overall:
    Device size:                 576.55GiB
    Device allocated:            134.61GiB
    Device unallocated:          441.94GiB
    Device missing:                  0.00B
    Used:                        132.72GiB
    Free (estimated):            442.17GiB      (min: 221.20GiB)
    Free (statfs, df):           442.16GiB
    Data ratio:                       1.00
    Metadata ratio:                   2.00
    Global reserve:              342.38MiB      (used: 0.00B)
    Multiple profiles:                  no

Data,single: Size:128.55GiB, Used:128.32GiB (99.82%)
   /dev/nvme1n1p2         93.55GiB
   /dev/nvme0n1   35.00GiB

Metadata,DUP: Size:3.00GiB, Used:2.20GiB (73.31%)
   /dev/nvme1n1p2          6.00GiB

System,DUP: Size:32.00MiB, Used:16.00KiB (0.05%)
   /dev/nvme1n1p2         64.00MiB

Unallocated:
   /dev/nvme1n1p2          1.00MiB
   /dev/nvme0n1  441.94GiB

% sudo btrfs device usage /
/dev/nvme1n1p2, ID: 1
   Device size:           119.14GiB
   Device slack:           19.53GiB
   Data,single:            93.55GiB
   Metadata,DUP:            6.00GiB
   System,DUP:             64.00MiB
   Unallocated:             1.00MiB

/dev/nvme0n1, ID: 2
   Device size:           476.94GiB
   Device slack:              0.00B
   Data,single:            35.00GiB
   Unallocated:           441.94GiB

% sudo btrfs device stats /
[/dev/nvme1n1p2].write_io_errs    0
[/dev/nvme1n1p2].read_io_errs     0
[/dev/nvme1n1p2].flush_io_errs    0
[/dev/nvme1n1p2].corruption_errs  0
[/dev/nvme1n1p2].generation_errs  0
[/dev/nvme0n1].write_io_errs    0
[/dev/nvme0n1].read_io_errs     0
[/dev/nvme0n1].flush_io_errs    0
[/dev/nvme0n1].corruption_errs  0
[/dev/nvme0n1].generation_errs  0

Created attachment 294805 [details]
kernel log since boot

Issue solved with  response from btrfs mailing list.

On 23/01/2021 23:57, Zygo Blaxell wrote:
> You don't have enough space to convert metadata yet, and you also don't
> have enough space to lock one of your 3 metadata block groups without
> running out of global reserve space, so this balance command forces the
> filesystem read-only due to lack of space.
>
>> <snip>
> First you need to get some unallocated space on devid 1, e.g.
>
>     btrfs balance start -dlimit=12 /
>
> I picked 12 chunks here because your new disk is about 4x the size of
> your old one, so you can expect metadata to expand from 3 GB to 15 GB.
> By moving 12 data chunks to the new disk (plus 3 more from converting
> from dup to raid1), we ensure that space is available for the metadata
> later on.
>
> Once you have some unallocated space on two devices, you can do the
> metadata conversion balance:
>
>     btrfs balance start -mconvert=raid1,soft /
>
> Every dup chunk converted to raid1 will release more space on devid 1,
> so the balance will complete (as long as you aren't writing hundreds
> of GB of new data at the same time).