Slow scrub for RAID56 data profiles

[freswa]

Scrubbing a volume with system and metadata on raid1c4 profile and data on raid6 or raid5, btrfs scrub start <mnt pt> is way slower than the avg read/write performance of that array. According to this post it is due to having one process per disk in the array, which is spawned here.

Request: Can we detect raid56 profiles and issue the ioctl in serial so that the IO operations of each process on such arrays don't interfere anymore?

[kdave]

There's a rewrite of the scrub logic in the upcoming kernel 6.4, some of the raid56 issues are fixed but I'm not sure if the speed is significantly better.