| 14 Feb 2023 |
 K900 | It's been running fine on all of my machines | 18:01:47 |
| K900 | (and I'm cleaning up my pile of cherry-picks) | 18:01:55 |
 @elvishjerricco:matrix.org | K900: I still think it should be based on chroot and realpath but I don't care enough to say it shouldn't be merged | 18:08:19 |
| K900 | I'd rather run all of this after the chroot entirely tbh | 18:09:58 |
| K900 | But I'm not sure there's a good way to do that | 18:10:06 |
| K900 | Outside of wrapping systemd | 18:10:18 |
| K900 | Which is just ew | 18:10:24 |
| K900 | I'm still hoping to see the day where we don't need to do that on nixos-wsl | 18:11:29 |
 @lily:lily.flowers | ElvishJerricco: I've been thinking about submitting a PR to systemd to canonicalize source for bind mounts (specifically so that they can be canonicalized from /sysroot in initrd). It would prevent us needing to artificially prepend /sysroot to only bind mounts from the NixOS side when generating the fstab for systemd-based initrd, and based on the old systemd PR I linked, they seem receptive to merging that functionality (or at least they did a few years ago). Thoughts? | 18:40:02 |
| @elvishjerricco:matrix.org | well the awkward thing is that bind mounts aren't the only problem | 18:41:26 |
| @lily:lily.flowers | Yeah, was worried you were going to say that. We only handle it for bind mounts in NixOS though | 18:41:49 |
| @elvishjerricco:matrix.org | overlayfs, for instance, has the directory options that would need the same treatment | 18:41:44 |
| @lily:lily.flowers | I mean theoretically I could just make it do that for any mount if the source is a non-/dev and non-/sys path | 18:42:13 |
| @elvishjerricco:matrix.org | this isn't to say we shouldn't improve bind mounts | 18:41:56 |
| @lily:lily.flowers | * I mean theoretically I could just make it do that for any mount if the source is a non-/dev and non-/sys absolute path | 18:42:26 |
| @elvishjerricco:matrix.org | just saying it's awkward | 18:42:02 |
| @lily:lily.flowers | In reply to @lily:lily.flowers
I mean theoretically I could just make it do that for any mount if the source is a non-/dev and non-/sys absolute path (Idk if there are scenarios where that would also Do The Wrong Thing too though) | 18:42:59 |
| @elvishjerricco:matrix.org | well the overlayfs example has the problem in the mount options, not the device or mountpoint | 18:43:19 |
| @lily:lily.flowers | In reply to @elvishjerricco:matrix.org
overlayfs, for instance, has the directory options that would need the same treatment Oh true, didn't even think about that. Hopefully people aren't doing that as an fs needed for boot though? Because we only generate for those needed for boot, right? | 18:43:32 |
| @lily:lily.flowers | In reply to @elvishjerricco:matrix.org
overlayfs, for instance, has the directory options that would need the same treatment * Oh true, didn't even think about that. Hopefully people aren't doing that as an fs needed for boot though? Because we only generate initrd fstab for those needed for boot, right? | 18:43:46 |
| @elvishjerricco:matrix.org | i mean, go look at qemu-vm.nix :P | 18:43:50 |
| @elvishjerricco:matrix.org | We currently do that | 18:43:56 |
| @lily:lily.flowers | Oh, joy | 18:44:03 |
| @lily:lily.flowers | I'm actually okay with that one staying special-cased tbh. I doubt someone would use it for anything other than having a writable nix store in a VM anyway | 18:45:02 |
| @elvishjerricco:matrix.org | I mean the real answer is to have a Root= option for mount units (all units?) so that the mount syscall occurs in a chroot or something so that all the scenarios are taken care of | 18:46:14 |
| @elvishjerricco:matrix.org | but that sounds very annoying to validate | 18:46:43 |
| @lily:lily.flowers | Idk, that makes way too much sense. Gotta make sure to pile on more and more hacks (/s) | 18:47:22 |
| @elvishjerricco:matrix.org | lol | 18:47:39 |
| @elvishjerricco:matrix.org | yea I wouldn't dare open a PR with that feature, but it does seem right | 18:47:55 |
| @lily:lily.flowers | In reply to @elvishjerricco:matrix.org
yea I wouldn't dare open a PR with that feature, but it does seem right Yeah that's more systemd refactoring than I am certainly comfortable with, even though that seems objectively better. I'll see what upstream says about at least handling bind mounts for now though (I'll probably open it after their 253 release, when they'll be more able to look at it and my other PR) | 19:47:06 |
