Hi Sam!
Am Mi., 27. Okt. 2021 um 21:24 Uhr schrieb Sam Hartman <hartm...@debian.org>:
>
> Package: debspawn
> Version: 0.5.0-1
> Severity: serious
> Justification: Significant data loss
>
> I used debspawn interact to interactively explore what I needed to do to get
> a new upstream package building.
> To make that easier, I mounted my source trees and debian working environment
> in the container.
>
> On exit, debspawn deleted everything.
> In retrospect, I can understand why this is, but it's pretty hostile to the
> developer.
> I might be alone, but I find it very helpful to mount various things into
> containers when exploring why things don't work.
Ha! First of all, I'm very sorry for this issue, and I hope this
didn't cause any bigger problems or a long recovery session.
Debspawn does not really expect users to mount things as well behind its back...
Bindmounts are evil sometimes, but ever since I deleted my whole root
filesystem by accident, I became a lot more careful :-D
> My recommendation would be to check for bind mounts and make sure they can be
> unmounted before cleaning up.
That is actually a rather annoying operation, as you need to parse
`/proc/self/mountinfo` or call `findmnt` on every directory in order
to not only find "real" mountpoints but also bindmounts.
I tried the latter option though, and on my system I could only
measure a slowdown of a few milliseconds, so that's IMHO perfectly
fine as safety measure.
> A fix that would have worked in my case but that may not generally be good
> enough would have been to restrict the container deletion to one-file-system.
That wouldn't find bindmounts though, and I'd rather catch these too :-)
The new 0.5.1 release will have the change included and will just
unmount any directory mountpoints upon container deletion - if the
user decided to create mountpoints on *files*, deletion of those will
likely just fail, which is fine with me (no data is lost and the user
knows what (not) to do next time).
Please give it a test with some scratch directory though before using
this with important data - just in case! ;-)
Cheers,
Matthias
