[Bug 1173461] New: Don't create btrfs subvolume for /tmp
https://bugzilla.suse.com/show_bug.cgi?id=1173461 Bug ID: 1173461 Summary: Don't create btrfs subvolume for /tmp Classification: openSUSE Product: openSUSE Tumbleweed Version: Current Hardware: Other OS: Other Status: NEW Severity: Major Priority: P5 - None Component: Kubic Assignee: kubic-bugs@opensuse.org Reporter: kukuk@suse.com QA Contact: qa-bugs@suse.de Found By: --- Blocker: --- We use already tmpfs for /tmp, but since we also still create a tmp subvolume and mount that, it's a race condition which of the two is used after booting. And it is irritating. Additional, we have https://jira.suse.com/browse/PM-1898, which also requires tmpfs for /tmp. -- You are receiving this mail because: You are on the CC list for the bug.
https://bugzilla.suse.com/show_bug.cgi?id=1173461
Thorsten Kukuk
https://bugzilla.suse.com/show_bug.cgi?id=1173461
Thorsten Kukuk
https://bugzilla.suse.com/show_bug.cgi?id=1173461
https://bugzilla.suse.com/show_bug.cgi?id=1173461#c1
--- Comment #1 from Fabian Vogt
https://bugzilla.suse.com/show_bug.cgi?id=1173461
https://bugzilla.suse.com/show_bug.cgi?id=1173461#c2
--- Comment #2 from Thorsten Kukuk
Otherwise we have a system which only boots properly if a "-tools" package is installed.
MicroOS works only properly if the tools package is installed ... -- You are receiving this mail because: You are on the CC list for the bug.
https://bugzilla.suse.com/show_bug.cgi?id=1173461
https://bugzilla.suse.com/show_bug.cgi?id=1173461#c3
--- Comment #3 from Thorsten Kukuk
I don't like that a .mount unit is part of a package, this should IMO be done in /etc/fstab instead. By YaST with the installer and in config.sh for image builds.
This makes it impossible for us to fix any problems which may occour for installed systems. I think distribution/system relevant mount points should really come from .mount units and not written by an installer in /etc/fstab. The later one is always making trouble after some time. -- You are receiving this mail because: You are on the CC list for the bug.
https://bugzilla.suse.com/show_bug.cgi?id=1173461
https://bugzilla.suse.com/show_bug.cgi?id=1173461#c4
--- Comment #4 from Thorsten Kukuk
https://bugzilla.suse.com/show_bug.cgi?id=1173461
https://bugzilla.suse.com/show_bug.cgi?id=1173461#c5
--- Comment #5 from Fabian Vogt
(In reply to Fabian Vogt from comment #1)
Otherwise we have a system which only boots properly if a "-tools" package is installed.
MicroOS works only properly if the tools package is installed ...
AFAICT, only tmp.mount is actually necessary. I propose moving that file into read-only-root-fs, IMO it makes more sense there. That's also used by transactional-server, but as /tmp is currently mounted as subvolume that would be a noop. I propose changing it to tmpfs there as well though. -- You are receiving this mail because: You are on the CC list for the bug.
https://bugzilla.suse.com/show_bug.cgi?id=1173461
https://bugzilla.suse.com/show_bug.cgi?id=1173461#c6
--- Comment #6 from OBSbugzilla Bot
https://bugzilla.suse.com/show_bug.cgi?id=1173461
https://bugzilla.suse.com/show_bug.cgi?id=1173461#c7
Richard Brown
https://bugzilla.suse.com/show_bug.cgi?id=1173461
https://bugzilla.suse.com/show_bug.cgi?id=1173461#c8
--- Comment #8 from Swamp Workflow Management
https://bugzilla.suse.com/show_bug.cgi?id=1173461
https://bugzilla.suse.com/show_bug.cgi?id=1173461#c9
Fabian Vogt
https://bugzilla.suse.com/show_bug.cgi?id=1173461
Guillaume GARDET
https://bugzilla.suse.com/show_bug.cgi?id=1173461
https://bugzilla.suse.com/show_bug.cgi?id=1173461#c10
Fabian Vogt
This change completely broke transactional-server: https://openqa.opensuse.org/tests/1326389
It does not have /tmp as subvolume but no tmp.mount from microos-tools either.
To fix this, the unit has to be moved to read-only-root-fs (as proposed in comment 5) or the tmp subvol added back.
@systemd-maintainers: What about having a systemd subpackage (systemd-tmpfs?) which provides tmp.mount in local-fs.target.wants? That way it it's possible to set up /tmp as tmpfs during installation even. There isn't really any reason to make this MicroOS/read-only-root-fs specific. -- You are receiving this mail because: You are on the CC list for the bug.
https://bugzilla.suse.com/show_bug.cgi?id=1173461
Ludwig Nussel
https://bugzilla.suse.com/show_bug.cgi?id=1173461
https://bugzilla.suse.com/show_bug.cgi?id=1173461#c11
--- Comment #11 from Thorsten Kukuk
https://bugzilla.suse.com/show_bug.cgi?id=1173461
https://bugzilla.suse.com/show_bug.cgi?id=1173461#c12
Thorsten Kukuk
@systemd-maintainers: What about having a systemd subpackage (systemd-tmpfs?) which provides tmp.mount in local-fs.target.wants? That way it it's possible to set up /tmp as tmpfs during installation even. There isn't really any reason to make this MicroOS/read-only-root-fs specific.
We need different options in tmp.mount for MicroOS than the systemd variant has. So this will not help us. -- You are receiving this mail because: You are on the CC list for the bug.
https://bugzilla.suse.com/show_bug.cgi?id=1173461
https://bugzilla.suse.com/show_bug.cgi?id=1173461#c13
Fabian Vogt
(In reply to Fabian Vogt from comment #10)
@systemd-maintainers: What about having a systemd subpackage (systemd-tmpfs?) which provides tmp.mount in local-fs.target.wants? That way it it's possible to set up /tmp as tmpfs during installation even. There isn't really any reason to make this MicroOS/read-only-root-fs specific.
We need different options in tmp.mount for MicroOS than the systemd variant has. So this will not help us.
Which ones and why? They look very compatible. -- You are receiving this mail because: You are on the CC list for the bug.
https://bugzilla.suse.com/show_bug.cgi?id=1173461
https://bugzilla.suse.com/show_bug.cgi?id=1173461#c14
Thorsten Kukuk
Which ones and why? They look very compatible.
There are requests for "noexec" and similar stuff. Something which is Ok for a Container Host OS, but I'm pretty sure will create problems on standard systems. -- You are receiving this mail because: You are on the CC list for the bug.
https://bugzilla.suse.com/show_bug.cgi?id=1173461
https://bugzilla.suse.com/show_bug.cgi?id=1173461#c15
Fabian Vogt
(In reply to Fabian Vogt from comment #13)
Which ones and why? They look very compatible.
There are requests for "noexec" and similar stuff. Something which is Ok for a Container Host OS, but I'm pretty sure will create problems on standard systems.
At least noexec would break the current transactional-update self-update mechanism. If at some point in the future MicroOS should have different options for /tmp, that can be achieved with dropins AFAICT. In any case, the change requested from comment 10 can be useful outside of MicroOS as well, it was even requested on the ML. -- You are receiving this mail because: You are on the CC list for the bug.
https://bugzilla.suse.com/show_bug.cgi?id=1173461
https://bugzilla.suse.com/show_bug.cgi?id=1173461#c16
--- Comment #16 from Thorsten Kukuk
At least noexec would break the current transactional-update self-update mechanism.
Then we have to fix that. "noexec" is a requirement from several RFI and will come. -- You are receiving this mail because: You are on the CC list for the bug.
https://bugzilla.suse.com/show_bug.cgi?id=1173461
https://bugzilla.suse.com/show_bug.cgi?id=1173461#c17
Franck Bui
https://bugzilla.suse.com/show_bug.cgi?id=1173461
https://bugzilla.suse.com/show_bug.cgi?id=1173461#c18
Fabian Vogt
systemd already ships tmp.mount in /usr/share/systemd/ so if anyone wants to switch to tmpfs for /tmp, he just needs to do 'systemctl link /usr/share/systemd/tmp.mount'.
Is this path stable, i.e. can we symlink to there from a different package? Or should we make a copy?
Regarding shipping systemd-tmpfs, I would try to avoid that otherwise we will end up with ton of systemd sub packages for just configuring the different systems out there (we have already systemd-logger, and now we're considering shipping a new sub package to just use different time server pools on SUSE and on openSUSE...)
Ok, fair enough. -- You are receiving this mail because: You are on the CC list for the bug.
https://bugzilla.suse.com/show_bug.cgi?id=1173461
https://bugzilla.suse.com/show_bug.cgi?id=1173461#c19
--- Comment #19 from Ludwig Nussel
https://bugzilla.suse.com/show_bug.cgi?id=1173461
https://bugzilla.suse.com/show_bug.cgi?id=1173461#c20
Franck Bui
Is this path stable, i.e. can we symlink to there from a different package? Or should we make a copy?
It should remain stable until we decide to switch to tmpfs on /tmp like upstream does since the beginning. If that happens we might stop shipping the copy in /usr/share/systemd but we will rename the symlink (that would have been created by 'systemctl link') to point to the new location (in /usr/lib/systemd). -- You are receiving this mail because: You are on the CC list for the bug.
https://bugzilla.suse.com/show_bug.cgi?id=1173461
https://bugzilla.suse.com/show_bug.cgi?id=1173461#c21
--- Comment #21 from Franck Bui
What's needed to go for /tmp on tmpfs in general like upstream systemd would do by default?
We would just need to stop moving tmp.mount from /usr/lib/systemd to /usr/share/systemd. -- You are receiving this mail because: You are on the CC list for the bug.
https://bugzilla.suse.com/show_bug.cgi?id=1173461
https://bugzilla.suse.com/show_bug.cgi?id=1173461#c27
--- Comment #27 from Swamp Workflow Management
https://bugzilla.suse.com/show_bug.cgi?id=1173461
https://bugzilla.suse.com/show_bug.cgi?id=1173461#c30
--- Comment #30 from Swamp Workflow Management
https://bugzilla.suse.com/show_bug.cgi?id=1173461
https://bugzilla.suse.com/show_bug.cgi?id=1173461#c31
--- Comment #31 from Swamp Workflow Management
https://bugzilla.suse.com/show_bug.cgi?id=1173461
https://bugzilla.suse.com/show_bug.cgi?id=1173461#c32
--- Comment #32 from OBSbugzilla Bot
https://bugzilla.suse.com/show_bug.cgi?id=1173461
https://bugzilla.suse.com/show_bug.cgi?id=1173461#c33
--- Comment #33 from OBSbugzilla Bot
https://bugzilla.suse.com/show_bug.cgi?id=1173461
https://bugzilla.suse.com/show_bug.cgi?id=1173461#c35
--- Comment #35 from Swamp Workflow Management
https://bugzilla.suse.com/show_bug.cgi?id=1173461
Petr Vorel
participants (1)
-
bugzilla_noreply@suse.com