btrfs check --repair question

All -

I have a file system I'm having some issues with. The initial symptoms were that mount would run for several hours, either committing or rolling back transactions (primarily due to a balance that was running when the system was rebooted for other reasons - the skip_balance mount option was specified because of this), but would then be killed due to an OOM condition (not much else running on the box at the time - a desktop system where everything else was waiting for the mount to finish). That's the background. Kernel 4.8.1 - custom config, but otherwise stock kernel - and btrfs-tools 4.8.3.

Ran btrfs check, and the only thing it reports is a sequence of these:

ref mismatch on [5400814960640 16384] extent item 0, found 1
Backref 5400814960640 parent 5401010913280 root 5401010913280 not found in extent tree backpointer mismatch on [5400814960640 16384]
owner ref check failed [5400814960640 16384]

Which, to my reading are simply some missing backrefs, and probably should be one of the easier issues to correct, but I know --repair is still considered experimental/dangerous, so I thought I'd ask before I run it... Is this a case that --repair can be reasonably expected to handle, or would I be better off recreating the file system and restoring from either my saved btrfs send archives or the more reliable backups?

tw

[email protected]

There were some OOM related issues early in 4.8, I would try 4.8.12 or even 4.8.14.

It might be that usebackuproot,ro mount will happen faster, and you can update the backups. Then use --repair. It's still listed as dangerous but it's gotten quite a bit less dangerous in later versions.

Chris Murphy