On 05/18/2012 07:26 AM, Marcin Szewczyk wrote: > Hi, > > this thread has been mentioned at the Ubuntu Launchpad[1], so I thought > I would link them and sum up conclusions from the other thread. > > On Fri, May 18, 2012 at 05:03:16PM +0900, sf...@users.sourceforge.net wrote: >> Ben Greear: >>> The second command fails, saying / is busy. >>> lsof of / shows lots of files. lsof of /cow shows none. >> >> You should investigate the printed file list, and find the file which >> prevents /cow turning RO. >> Generally, the shutdown script kills most of all processes, and then >> unmount all filelsystems (except the root). The root dir is remounted as >> RO. The "remount,mod:/cow=ro" is logically equivalent to "remount,ro". > > In the comment #82 there is a sequence that ends successfully marking > all the components as (ro). The noxino remark has been made long time > ago in comment #49. > > And some curiosities discovered follow... > > There were I/O errors reported by kernel visible just before halt or > reboot. Errors were mentioned in comments #19 and #41. This was > primarily caused by USB drive detachment. Generic command was used. > Originally it was meant for CDROMs, for USB drives it was doing > collateral damage. It has been fixed (now it's like in Debian scripts). > > Visibility of /cdrom (squashfs) and /cow (the overlay ext) has been > available for quite a long time. It's the "showmounts" boot option.
Yeah, but you shouldn't have to add a special boot option in order to get a clean shutdown. Seems to me that /cow should always be visible. > My summary, with more or less patched scripts, is in comment #82. Any idea why this stuff doesn't make it upstream into Ubuntu? It's not that hard to test once you know how to respin cds, and anyone dealing with casper should know how to do that or be able to follow instructions to do so... > One major problem still remains - USB drives lie. Sync operation returns > before actual flash write is complete. Because most USB drives do > wear-levelling and are MLC (not SLC) memories we end up with filesystem > with serious damage. Even fsck won't help as both metadata and data can > be damaged. I left a to second sleep in my script, in hopes that is good enough to solve this problem. I am not sure if the 'usb eject' issue is solved in 12.04 or not, but it seems to work OK for me once I made the changes to properly remount /cow as (ro). Thanks, Ben -- Ben Greear <gree...@candelatech.com> Candela Technologies Inc http://www.candelatech.com ------------------------------------------------------------------------------ Live Security Virtual Conference Exclusive live event will cover all the ways today's security and threat landscape has changed and how IT managers can respond. Discussions will include endpoint security, mobile security and the latest in malware threats. http://www.accelacomm.com/jaw/sfrnl04242012/114/50122263/
