On Sat, Feb 8, 2014 at 3:19 PM, Aryeh Friedman <[email protected]>wrote:
> > > > On Sat, Feb 8, 2014 at 3:14 PM, Aryeh Friedman > <[email protected]>wrote: > >> >> >> >> On Sat, Feb 8, 2014 at 3:01 PM, Adam Vande More <[email protected]>wrote: >> >>> >>> On Sat, Feb 8, 2014 at 6:51 AM, Aryeh Friedman <[email protected] >>> > wrote: >>>> >>>> bhyve blindly read/writes into the middle of the file without >>>> consulting the filesystem and thus bypassing any things like sparse fill >>>> in.... namely all you gain is a few seconds of startup time (matter of fact >>>> I think truncate might use sparse allocation [i.e. attempting to read into >>>> the middle with guest OS control will result in potentially seeing host >>>> data]) >>>> >>> >>> If this is true then there is a *critical* security issue. >>> >>> Using sparse files isn't to gain performance, it's to conserve disk >>> space. Using md devices backed by sparse images would accomplish this. If >>> the sparsify app works on FreeBSD, then there should be no problem using >>> those type of volumes. >>> >>> >> It sounds almost identical to the qcow2 security issue being discussed on >> [email protected] recently. This might be a *HUGE* win for bhyve >> then in considering that it's default format is raw (should ahci-hdd be the >> default?). devel/qemu (not sure about -dev) uses qcow2 as a default and >> when playing with it on other OS's I found that it seemed to default to >> that also. It is my understand that most of the open source cloud >> platforms use qcow2 as their default also (I remember this from an attempt >> to install openstack grizzly last summer... I have not checked havana >> though... can any of the freebsd-openstack confirm this?). >> > > Forgot to mention that the host OS's disk scheduling also gives a brief > window of opportunity during the time after the inode is made and the old > contents wiped due to the size of the file > > My short memory must be going the main point I meant to make in the second reply is as far I know *ALL* hypervisors have the same blind read/write (remember they see what ever is acting as the disk as (from the VM's POV) a real disk and in order to maintain that illusion blind writes/reads are necessary) -- Aryeh M. Friedman, Lead Developer, http://www.PetiteCloud.org _______________________________________________ [email protected] mailing list http://lists.freebsd.org/mailman/listinfo/freebsd-virtualization To unsubscribe, send any mail to "[email protected]"
