Red Hat Bugzilla – Bug 683434
Fedora-15-Alpha-i386 fails to upgrade with /tmp on tmpfs
Last modified: 2012-10-23 10:57:05 EDT
Description of problem:
I gave the Fedora-15-Alpha-i386-DVD.iso a try to upgrade a Fedora-13 installation on a testsystem.
It failed early in the upgrade with a box telling:
"An error occurred mounting device tmpfs as /tmp: mount
failed (-2, None). This is a fatal error and the install cannot
Press Enter to exit the installer."
I tried twice.
Steps to Reproduce:
1. On an installed Fedora add something similar to this to /etc/fstab
none /tmp tmpfs size=256m 0 0
2. Insert DVD, reboot and try to upgrade
This was still the case in the final F15 installer (DVD). Fortunately I got it working by commenting the tmpfs-line from fstab, in the next run the installer succeeded (well, there were other problems with the install, but..).
Nice to read about this now :( (You see, there *are* people out there with /tmp as tmpfs..)
Same issue for me with the preupgrade tool for F15 final from F14 final. fstab entries for me:
tmpfs /dev/shm tmpfs defaults 0 0
tmpfs /tmp tmpfs defaults,noatime,mode=1777 0 0
Trying commenting both lines; will feed back with progress.
To confirm my earlier post: I commented the lines above and it successfully installed. No further issues related to the install.
The /dev/shm line is ignored anyway -- we write our own /dev/shm line during the upgrade.
I have been trying to do tmpfs mounts from the shell in the installer environment (F15) and am seeing mount crash when passed either the noatime or size= options. So far, without either of those options, I have been able to successfully mount tmpfs filesystems and have been able to get an upgrade done.
FWIW I am also seeing mount crash (SEGV) when I try to mount a tmpfs with noatime in the F16 installer environment.
In F16 installer environment it appears to be the combination of noatime and mode= that triggers the crash.
As you may know, we are moving to a completely separate process for handling upgrades, one that will not involve anaconda at all. Thus, I believe this bug will not be valid when that is in place and there's nothing for us to fix.