(In reply to comment #24)
(In reply to comment #23)
(In reply to comment #21)
* Linux xfs is extremely lazy writing back dirty data I think you meant metadata (for data, sync should be really enough). I wonder how come you didn't see similar problems with ext3/4 - likely because they have all metadata exposed in buffer cache while xfs does not.
* Linux buffer caches are not coherent (/dev/sda does not include /dev/sda1) True - but sync(1) plus flushing the caches on the device you want to use (BLKFLSBUF ioctl) solves this problem.
No, it didn't. I was recommended to use xfs_freeze and that did the job. I believe it doesn't make all the metadata properly visible for XFS (since it caches some things in private allocations). But I'm rather sure it does solve
https://bugzilla.novell.com/show_bug.cgi?id=805732 https://bugzilla.novell.com/show_bug.cgi?id=805732#c27 Jan Kara <jack@suse.com> changed: What |Removed |Added ---------------------------------------------------------------------------- Status|NEW |NEEDINFO InfoProvider| |duwe@suse.com --- Comment #27 from Jan Kara <jack@suse.com> 2013-03-20 17:13:07 UTC --- (In reply to comment #26) the problems with incoherency between a cache for sda and sda1. Anyway, that's irrelevant to the current problem since I agree that for XFS, freezing might be needed.
(xfs_freeze -f /; xfs_freeze -f /boot; xfs_freeze -u /; xfs_freeze -u /boot ) > /dev/null 2>&1
That way /dev/null will be opened on still unfrozen filesystem (subsequent writing to it doesn't block because device null as such isn't frozen) and things will work (tested that). Thorsten, can you change postinstall like that please? Thanks!
The whole logic was to be carried out exclusively on xfs, or when the root fs type could not be determined. Yes, the second option is actually the case in case of osc build root.
It looks to me as if someone clueless has inserted the LOADER_TYPE logic in the wrong place.
BTW, where are the valid values for YAST_IS_RUNNING documented? Sorry but I fail to see how that is relevant. Code in grub postinstall can open inode (/dev/null) on the frozen filesystem for writing. That is clearly a bug which is easy to fix by opening the inode before the filesystem is frozen. Changing those two lines should be a trivial thing to do...
-- Configure bugmail: https://bugzilla.novell.com/userprefs.cgi?tab=email ------- You are receiving this mail because: ------- You are on the CC list for the bug.