ZFS send | receive corrupt with feature@large_dnode = active in 12.0-RELEASE-p10

[icecoke]

In 12.0-RELEASE-p10 we did zpool upgrade and on some pools we enabled zfs set dnodesize=auto pool. With all of the pools were we set the dnodesize to 'auto' (and therefore resulting in feature@large_dnode = active) it is not anymore possible to send | receive the content into a new pool, whatever we set the features of the new pool. We also tried to set the dnodesize back to legacy in the source, but that didn't help either. We always get this error:

[root@host ~]# zfs send -R pool@01 | zfs receive -Fdvu poolnew
receiving full stream of pool@01 into poolnew@01
cannot receive new filesystem stream: dataset does not exist
warning: cannot send 'pool@01': signal received
[root@host ~]#

Has anyone experienced that, too? Anyone with a solution to send | receive from such a pool?

I read about problems regarding send | receive between ZoL and FreeBSD 12, but the source here is FreeBSD 12, too.

 

[icecoke]

Anyone? The effect is reproducible for us, thus rendering zfs pools with dnodesize=auto to break regarding any send | receive actions.

Can someone lead me to an url to file a bug or someone who has deeper knowledge about the zfs changes between FreeBSD <12 and 12?