Re: VFS mount rollback for virtio 9pfs

From: Daniel O'Connor <darius_at_dons.net.au>
Date: Mon, 18 Jul 2022 01:48:31 UTC

> On 16 Jul 2022, at 00:45, Felix Palmen <felix@palmen-it.de> wrote:
> * Felix Palmen <felix@palmen-it.de> [20220715 15:16]:
>> I tested it, successful so far as it seems! Thanks again!
> 
> Unfortunately, it doesn't work reliably yet. Trying a testport with
> poudriere logs mounted via virtfs, I got errors like these:
> 
> #v+
> /usr/bin/awk: can't open file 14a-default/2022-07-15_16h43m01s/.poudriere.jailname
> input record number 1, file 14a-default/2022-07-15_16h43m01s/.poudriere.jailname
> 
> rename: /usr/local/poudriere/data/logs/bulk/14a-default/2022-07-15_16h43m01s/.tmp-.poudriere.snap_loadavg.OV0Gxx2m to /usr/local/poudriere/data/logs/bulk/14a-default/2022-07-15_16h43m01s/.poudriere.snap_loadavg: Invalid argument
> 
> _mktemp: mkstemp failed on 14a-default/2022-07-15_16h43m01s/.tmp-.data.json.stpmwmGg: Invalid argument
> 
> mapfile_write: Missing handle
> #v-
> 
> after that, the mount is somehow broken:
> 
> # ls -l /usr/local/poudriere/data/logs/
> ls: bulk: Invalid argument

Hmm fun! :)

I did copy the tmpfs tests but they are mostly basic functional tests rather that exercising race conditions and the like.

> So, probably it's back to NFS *for now*.
> 
> Still the best attempt I've seen so far, all others I tried crashed the
> kernel on mount ;)

Thanks for testing it, I will try some more intensive tests.
Which hypervisor are you using BTW?

--
Daniel O'Connor
"The nice thing about standards is that there
are so many of them to choose from."
-- Andrew Tanenbaum