stage-1: retry mounting ZFS root a few times #89353
Conversation
| local n=0 | ||
| while true; do | ||
| mount "/mnt-root$mountPoint" && break | ||
| if [ "$fsType" != cifs -o "$n" -ge 10 ]; then fail; break; fi | ||
| if [ \( "$fsType" != cifs -a "$fsType" != zfs \) -o "$n" -ge 10 ]; then fail; break; fi |
There was a problem hiding this comment.
I assume upstream does not have this problem because they use zpool import instead of mount: https://github.com/openzfs/zfs/blob/master/contrib/dracut/90zfs/mount-zfs.sh.in#L59
There was a problem hiding this comment.
No, they use zpool import to import/open the pool (like we do) but then they also mount the dataset/filesystem (like we do), see https://github.com/openzfs/zfs/blob/99b281f1ae3833826b4e3cee4126a4ea5e6de987/contrib/dracut/90zfs/mount-zfs.sh.in#L80 and https://github.com/openzfs/zfs/blob/99b281f1ae3833826b4e3cee4126a4ea5e6de987/contrib/dracut/90zfs/zfs-lib.sh.in#L76
Note that upstream doesn't retry mounting like I am doing here, and for sure mount returning Device or resource busy sporadically while mounting the root filesystem like in #25383 seems like a bug in ZFS (which I'm not sure has been fixed), but I think it doesn't hurt to do what I'm doing: it will fix the boot failure in case the bug resurfaces again, while only making the boot hang 10 seconds later in case there is a real mount problem.
There was a problem hiding this comment.
Can you also link to the issue in the code, so in future we know why it was added?
This is a workaround for NixOS#25383
Motivation for this change
This is a workaround for #25383.
I've been running my ZFS and non-ZFS systems with this change for years but had forgotten to submit a PR.
Things done
sandboxinnix.confon non-NixOS linux)nix-shell -p nixpkgs-review --run "nixpkgs-review wip"./result/bin/)nix path-info -Sbefore and after)