Re: corrupt leaf: root=1 block=57567265079296 slot=83, bad key order

[Hugo Mills <hugo@carfax.org.uk>]

[-- Attachment #1: Type: text/plain, Size: 2280 bytes --]

On Thu, Feb 14, 2019 at 08:25:26PM +0800, Qu Wenruo wrote:
> On 2019/2/14 下午7:58, Jesper Utoft wrote:
> > Hello Fellow BTRFS users.
> > 
> > I have run into the bad key order issue.
> > corrupt leaf: root=1 block=57567265079296 slot=83, bad key order, prev
> > (18446744073709551605 0 57707594776576) current (18446726481523507189
> > 0 57709742260224)
> > The lines repeats over and over..
> > 
> > I read a thread between Hugo Mills and Eric Wolf about a similar issue
> > and i have gathered the same info. 
> Now we have all the needed info.
> 
> > 
> > I understand that it probably is hardware related, i have been running
> > memtest for 60h+ to see if i could reproduce it.
> > I also tried to run btrfs check --recover but it did not help.
> > 
> > My questions is if it can be fixed?
> 
> Yes, but only manual patching is possible yet.

   David: What needs to be done to get the bitflip-in-key patches
added to btrfs check? They've been lurking in some patch stack for
literally years, and would have dealt with this one easily.

[snip]
> Thankfully, all keys around give us a pretty good idea what the original
> value should be: (FREE_SPACE UNTYPED 57709742260224).
> 
> And for the raw value:
> bad:  0xffffeffffffffff5
> good: 0xfffffffffffffff5
>             ^
> e->f, one bit get flipped.
> (UNTYPED is the same value for UNKNOWN.0, so don't worry about that).
> 
> I have created a special branch for you:
> https://github.com/adam900710/btrfs-progs/tree/dirty_fix
> 
> Just compile that btrfs-progs, no need to install, then excute the
> following command inside btrfs-progs directory:
> 
> # ./btrfs-corrupt-block -X <device>

   BUT, don't do it until you've found and replaced the bad RAM that
broke it in the first place.

> And your report just remind me to update the write time tree block
> checker....

   Looking forward to dealing with a whole new type of "btrfs is
broken!" complaints on IRC (followed by "can't I just let it carry on
regardless?"). ;)

   Hugo.

-- 
Hugo Mills             | Hickory Dickory Dock,
hugo@... carfax.org.uk | Three mice ran up the clock.
http://carfax.org.uk/  | The clock struck one,
PGP: E2AB1DE4          | The other two escaped with minor injuries

[-- Attachment #2: Digital signature --]
[-- Type: application/pgp-signature, Size: 836 bytes --]