Need more options to address uncorrectable btrfs scrub error.

[tribbloid]

here is an example the scrub output:

$ sudo btrfs scrub status /dev/nvme0n1p7
UUID:             a5c81116-78a5-4edb-b57c-b08e90e1391b
Scrub started:    Fri Aug  9 16:58:37 2024
Status:           finished
Duration:         0:02:50
Total to scrub:   407.09GiB
Rate:             2.39GiB/s
Error summary:    csum=11
  Corrected:      0
  Uncorrectable:  11
  Unverified:     0

there are 11 uncorrectable csum error, during the lifespan of the partition these errors will gradually accumulate and caused affected file blocks to be without csum.

The scrub command (or another command) should give user the following 2 options:

• delete the file with the wrong csum
• delete & regenerate the wrong csum for the file

this feature should be used with care, but will restore csum for affected files

[adam900710]

Check the dmesg which contains the filename.

And if you really want to fix the corruption, you should go profiles with duplication.

[tribbloid]

I thought profiles with duplication is always enabled after kernel 5.15? https://forum.manjaro.org/t/howto-convert-btrfs-profile-of-metadata-single-to-dup-duplicate-of-metadata-is-more-safe-than-single/115534

even with that feature there is a possibility of metadata corruption (e.g. if CoW is disabled, or write history is being dropped)

[adam900710]

Go duplication profiles for DATA!!

If it's metadata corruption your fs is already doomed, but that's not the case.