On Thu, May 05, 2005 at 07:18:02PM +0800, Cameron Patrick wrote: > On Wed, May 04, 2005 at 07:02:43PM -0700, Steve Langasek wrote: > > > It also reduces the possibility of serious data corruption when suspending > > > your machine and then booting with a non-resume-capable kernel. > > Reduces, but doesn't eliminate? How likely is it that such corruption will > > happen? > The new version of hibernate looks for swap devices with a software > suspend header and invalidates the suspend image by mkswap'ing over > it. It finds swap partitions and files okay in non-pathological cases > (I believe exceptions include swap on LVM or encrypted using dm-crypt, > both of which require substantial effort to use as suspend targets > anyway; or if the partition isn't listed in /etc/fstab). > The potential for data corruption comes in when you suspend (writing > out an image containing all kernel data structures, including disc > caches and filesystem meta data and so on), then boot a kernel not > configured for resuming which mounts your filesystems read-write and > starts writing to them (or even just replaying the journal), and then > reboot again, resuming the saved image. Now the state of the > filesystem on disc doesn't match the original kernel's idea of what > should be there, and things go awry. > While this is warned against in the suspend2 documentation, it still > seems that people manage to do this every now and again - especially > when setting up suspend for the first time and not setting the resume > device correctly on the kernel command line or initrd. Approved. Thanks, -- Steve Langasek postmodern programmer
Attachment:
signature.asc
Description: Digital signature