I reported this on the mailing list and also discussed on IRC. I don’t see any mention of a fix in release changelogs since 6.6.3, and since not being able to pull snapshots off a degraded but functioning filesystem seems potentially catastrophic and initially prevented emergency backup of snapshots, I am opening this ticket.

Here, btrfs send used metadata from the damaged device and failed without trying the metadata from the healthy device:

BTRFS critical (device dm-0): corrupted leaf, root=1348 block=364876496896 owner mismatch, have 7 expect [256, 18446744073709551360]

The filesystem driver read from the healthy mirror and corrected the error. After btrfs scrub fixed all errors on the failing device, btrfs send worked.

btrfs-progs 6.6.3-1.2+b1
Linux kernel 6.9.10
Debian testing/sid

Best regards,