Created attachment 1084956 [details] The kickstart used for ostree live pxe Description of problem: Using --make-ostree-live with --no-virt option. So it fails in blivet/util.py umount() which simply shells-out to the system umount command. The issue here is that ostree is apparently leaving sub-mounts under /mnt/sysimage for things like /var or /usr. Obviously one cannot umount a location that is still being used, due to sub-mounts. I was able to work-around the issue by hacking blivet to use recursion (aka umount --recursive /mnt/sysimage), but the issue of ostree leaving stuff mounted under /mnt/sysimage cannot be ignored. I'm not sure if ostree should be changed to cleanup after itself, or if blivet should gain recursive umount. Either way, we need to get those thing umount'ed under /mnt/sysimage. Version-Release number of selected component (if applicable): # rpm -q fedora-release python-blivet lorax anaconda ostree fedora-release-22-1.noarch lorax-22.12-1.fc22.x86_64 python-blivet-1.0.10-1.fc22.noarch anaconda-22.20.13-1.fc22.x86_64 ostree-2015.6-2.fc22.x86_64 How reproducible: 100% using the attached kickstart Steps to Reproduce: 1. livemedia-creator --no-virt --make-ostree-live --ks fedora-cloud-atomic-pxetolive.ks --live-rootfs-keep-size 2. 3. Actual results: bails in blivet/util.py:umount() Expected results: Additional info:
Created attachment 1084958 [details] Anaconda TB file Here is the Anaconda TB file, compressed.
Would this PR fix the problem? https://github.com/rhinstaller/anaconda/pull/370
(In reply to Radek Vykydal from comment #2) > Would this PR fix the problem? > https://github.com/rhinstaller/anaconda/pull/370 Yes. However the PR does not apply cleanly on my F22 test environment, and manually adding does not appear to succeed either. But generally speaking the code in rpmostreepayload.py that does umount stuff looks encouraging. Probably minor tweaks to get working in Fedora.