On 2023/7/31 17:46, Christoph Hellwig wrote:
Thanks. I've not been able to reproduce it on the apparent bisection commit for more than half an hour, but running it on the originally reported commit reproduces it after a few minutes. I'll see if I can come up with a better bisection.
I checked the related code, and didn't find anything obvious. But there is a chance that the image is intentionally corrupted so that we got a reloc root but incorrect root owner. Thus I sent out a patch to make that triggering ASSERT() to a more graceful exit: https://lore.kernel.org/linux-btrfs/24881cc9caf738f6248232709d7357d3186773b5.1690782754.git.wqu@xxxxxxxx/T/#u Although I never got the C reproducer to trigger, thus no confirmation on that. Thanks, Qu
