FS#56354 - [qemu] drop udev rule for KERNEL=="kvm"
Attached to Project:
Arch Linux
Opened by Tom Yan (tom.ty89) - Thursday, 16 November 2017, 19:28 GMT
Last edited by Eli Schwartz (eschwartz) - Monday, 18 December 2017, 17:31 GMT
Opened by Tom Yan (tom.ty89) - Thursday, 16 November 2017, 19:28 GMT
Last edited by Eli Schwartz (eschwartz) - Monday, 18 December 2017, 17:31 GMT
|
Details
Description:
systemd already ships the same rule: [tom@archlinux ~]$ grep 'KERNEL=="kvm"' /usr/lib/udev/rules.d/* /usr/lib/udev/rules.d/50-udev-default.rules:KERNEL=="kvm", GROUP="kvm", MODE="0660" /usr/lib/udev/rules.d/65-kvm.rules:KERNEL=="kvm", GROUP="kvm", MODE="0660" /usr/lib/udev/rules.d/70-uaccess.rules:SUBSYSTEM=="misc", KERNEL=="kvm", TAG+="uaccess" More importantly, there is a coming change in systemd that changes the default mode from 0660 to 0666 and drops the uaccess rule: https://github.com/systemd/systemd/commit/b8fd3d82205f632ce001fade74fed287e1564a1a The rule shipped downstream will not only be redundunt but confusing and likely problematic. Attached are the updated version of relevant files. Additional info: * package version(s) 2.10.1-1 |
This task depends upon
Closed by Eli Schwartz (eschwartz)
Monday, 18 December 2017, 17:31 GMT
Reason for closing: Fixed
Additional comments about closing: qemu 2.11.0-1
Monday, 18 December 2017, 17:31 GMT
Reason for closing: Fixed
Additional comments about closing: qemu 2.11.0-1
FS#54943P.S. Though among the tons of revisions of patches (no idea why it was made so complicated), it seems to be addressed.
FS#54943expanded to cover the duplicate udev rule as well as the duplicate sysuser rule as they were both caused by systemd 234 and the PKGBUILDs would need adjustment.If this issue was limited to https://github.com/systemd/systemd/commit/b8fd3d82205f632ce001fade74fed287e1564a1a then it seems premature as that commit will be for systemd 236 which is not in arch yet.
Edit:
I made so many different patches to provide the maintainer with alternative options which resulted in six initial versions
the later updated patches addresses shortcomings in the initial version and flysray does not allow the removal or updating attachments.
FS#54943will be resolved https://bugs.archlinux.org/task/54943#comment160609 is the last maintainer comment since then both qemu and libvirt have been updated multiple times with no relevant patches.systemd could add the option -Ddev-kvm-mode=0660 now which will have no effect as that is the current default and would prevent the change from the later commit.
Edit:
0660 instead of 600
No we don't need any option set downstream, and we do not want 600 anyway. It's a matter of granting rw to others once in for all or doing it with the uaccess ACL approach. (I mean, it's a matter of upstream is gonna change from the latter to the former and this now not-a-problem dup will become a problem) Seems like you still aren't sure what this is about?
So that comes back to my point about it being premature as this addresses an issue the arch packages are not affected by but its not my decision anyway.
Edit:
Fixed sorry 0660 instead of 600
FS#54943which do not fix the duplicate udev rule as well?FS#54943is there any point in submitting more patches forFS#54943?FS#54943. Please check with him what is the patches submission status.FS#54943is a way more nasty issue. It is not just a qemu/libvirt issue tbh, but to some extent a general policy conflict. AFAIR Arch decided sometime ago that it should not only used fixed ID/name mapping for the downstream system groups (while systemd does enumerated mapping AFAIK), but also try to reference to them with ID and avoid names everywhere. (Personally I found that silly and ugly ever since it was introduced) It would have been trivial if there's no such policy, but now the devs may need to think it through because replacing every 78 to kvm in the libvirt package means that the policy has to be violated.It was also agreed in principal that was the intended fix for
FS#54943https://bugs.archlinux.org/task/54943#comment160609 I assume anatolik also agreed that was the right approach and no other package apart from libvirtrelies on kvm having the gid 78 fortunately there were not many packages depending on qemu to check.