if /tmp is not a separate partition the box that has openrc-0.12.4 will be unable to boot. although the 'root' init.d script is part of the 'boot' runlevel it is started much too late during the init process. 'bootmisc' however is started soon. and bootmisc runs /usr/bin/mktemp. mktemp needs write permissions to '/tmp', but '/tmp' ('/' really) is readonly at this stage. the boot process will be incomplete - no network interfaces, no services that depend on net, etc. in one word it will fail. if I try then to start any init script (including /etc/init/boot) I get the 'already starting' error. I am not sure how is that script supposed to be queued to be started at a later date. init 3 is long started, I logged into the machine via a console and openrc is supposed to run stuff serialized. so the likelihood of that script starting by itself in the near future is null. so to conclude: 1- if /tmp is not a separate partition then server fails to start 80% of the services (flawed init script ordering) 2- if a user wants to have a script started, it should start without veto rights from a random all knowing AI (flawed idea that the user has to be an idiot) nyleve.dtc ~ # rc-update show boot bootmisc | boot device-mapper | boot fix_eth | boot fsck | boot hostname | boot hwclock | boot keymaps | boot localmount | boot loopback | boot lvm | boot modules | boot net.lo | boot procfs | boot root | boot swap | boot swapfiles | boot sysctl | boot tmpfiles.setup | boot urandom | boot possible 'fix': if I add a tmpfs filesystem to be mounted as '/tmp' in fstab and reboot I get a few screens worth of cannot remove 'rm: cannot remove /tmp/tmp.JHRTUYRYTRY/run/udev/data/foo: Read-only file system', but the server will have all services running. can you please remove this openrc version from the stable tree and we both pretend it never happened? thanks, peter
(In reply to Petre Rodan from comment #0) > if /tmp is not a separate partition the box that has openrc-0.12.4 will be > unable to boot. I have exactly zero systems with /tmp as a separate partition. All of them run fine. > although the 'root' init.d script is part of the 'boot' runlevel it is > started much too late during the init process. 'bootmisc' however is started > soon. and bootmisc runs /usr/bin/mktemp. So that might be the real problem. Why would anything run before /etc/init.d/root? > so to conclude: > 1- if /tmp is not a separate partition then server fails to start 80% of > the services (flawed init script ordering) Unconfirmed. > 2- if a user wants to have a script started, it should start without veto > rights from a random all knowing AI (flawed idea that the user has to be an > idiot) Now you're just trolling. > nyleve.dtc ~ # rc-update show boot > can you please remove this openrc version from the stable tree and we both > pretend it never happened? Again, do not troll, kthx. Please set rc_logger="YES" and rc_log_path=/var/log/rc.log in /etc/rc.conf and attach rc.log to this bug report.
Also, tell us why fix_eth is in the boot runlevel and what it actually does.
Created attachment 364712 [details] openrc.png the rc log is empty in the scenario in which /tmp is not a separate partition. so attached you'll find the next best thing: a gimp stitched capture of the ipmi remote console output.
Created attachment 364714 [details] rc.log for the second scenario rc.log for when I added a tmpfs /tmp into fstab.
I am pretty certain this has nothing to do with how you mount /tmp or if it is part of the root filesystem - it comes down to the order in which the boot level init scripts are run.
(In reply to Jeroen Roovers from comment #1) > > although the 'root' init.d script is part of the 'boot' runlevel it is > > started much too late during the init process. 'bootmisc' however is started > > soon. and bootmisc runs /usr/bin/mktemp. > > So that might be the real problem. Why would anything run before > /etc/init.d/root? it would be nice to know. > > so to conclude: > > 1- if /tmp is not a separate partition then server fails to start 80% of > > the services (flawed init script ordering) > > Unconfirmed. > > > 2- if a user wants to have a script started, it should start without veto > > rights from a random all knowing AI (flawed idea that the user has to be an > > idiot) > > Now you're just trolling. sorry if it sounds like trolling, but when I run something in a console the last thing I expect is to be denied by a broken error-checking function. broken because the reason why service starting is denied ('already starting' when that action will never actually happen). how do I overwrite this AI? > > nyleve.dtc ~ # rc-update show boot > > can you please remove this openrc version from the stable tree and we both > > pretend it never happened? now THAT is trolling. maybe. > Please set rc_logger="YES" and rc_log_path=/var/log/rc.log in /etc/rc.conf > and attach rc.log to this bug report. did that, but in the first case the log could not be created - see png capture. fopen /run/openrc/rc.log failed then I added the tmpfs /tmp and the log was filled (attachment #2 [details]). > Also, tell us why fix_eth is in the boot runlevel and what it actually does. ----------- 8< ------------------ #!/sbin/runscript depend() { before net } start() { ebegin "Fixing eth order" /local/adm/bin/fix_eth_order.sh -eo # [1] eend $? } stop() { : } ----------- 8< ------------------ [1] https://github.com/rodan/fix_eth_order I had this in 'boot' because adding it in 'default' it ended up running after net for some unknown reason. (tested with an older openrc) but anyhow I removed it temporarily to see if that fixes something. it did not.
Remove net.lo from boot, see bug 490750. Please show us your fstab. If "/tmp" isn't listed in fstab, please tell us how you mount/create "/tmp". What kind of system do you have? I am missing some default services like mtab and termencoding, is it some kind of container?
Created attachment 364768 [details, diff] bootmisc.patch > Remove net.lo from boot, see bug 490750. removed net.lo from boot. thanks for the hint. no change to the bug at hand tho. > Please show us your fstab. If "/tmp" isn't listed in fstab, please tell us how you mount/create "/tmp". for the original scenario of this bug report: # grep -c '/tmp' /etc/fstab 0 > What kind of system do you have? I am missing some default services like mtab and termencoding, is it some kind of container? it's a server like any other. no xen, pv or anything fancy. /etc/mtab is a symlink so no need to run the mtab init script. I see now it verifies if it's a link or not, but why to run a useless script anyhow? I removed all font/keymapping/encoding stuff since I'm happy with the defaults as far as the local consoles are concerned. current rc-update: # rc-update bootmisc | boot devfs | sysinit device-mapper | boot dmesg | sysinit ekey-egd-linux | default fcron | default fsck | boot hostname | boot hwclock | boot killprocs | shutdown local | default nonetwork localmount | boot loopback | boot lvm | boot modules | boot mount-ro | shutdown net.eth0 | default netmount | default ntp-client | default ntpd | default postfix | default procfs | boot root | boot savecache | shutdown shorewall | default shorewall6 | default svscan | default swap | boot swapfiles | boot sysctl | boot sysfs | sysinit syslog-ng | default tmpfiles.dev | sysinit tmpfiles.setup | boot udev | sysinit udev-mount | sysinit urandom | boot watchdog | default attached is a patch that fixes both scenarios. why is that needed you might ask? because bootmisc needs a rw /. and it gets it only if 'root' has already started. why does it work for me without the patch? because it probably gets pulled in via a different route. a route I am not using. and since gentoo is all about options, please support mine too by adding this patch into the next iteration of openrc.
I'm curious, what is the setting of rc_parallel in /etc/rc.conf? Also, I would like you to try a test for me. In /etc/conf.d/bootmisc, add this line: rc_after="root" Let me know if this takes care of the issue.
Ok, never mind my test request, that does exactly the same thing as your patch. I will apply your patch to Master.
This is applied in commit 235f94c and will be included in OpenRC-0.13.
(In reply to William Hubbs from comment #9) > I'm curious, what is the setting of rc_parallel in /etc/rc.conf? rc_parallel="NO" > This is applied in commit 235f94c and will be included in OpenRC-0.13. thanks.
