On Mon, Mar 22, 2021 at 09:51:19AM -0400, Dave Voutila wrote:
>
> Otto Moerbeek writes:
>
> > On Mon, Mar 22, 2021 at 01:47:18PM +0100, Mischa wrote:
> >
> >>
> >>
> >> > On 22 Mar 2021, at 13:43, Stuart Henderson <s...@spacehopper.org> wrote:
> >> >
> >> >>> Created a fresh install qcow2 image and derived 35 new VMs from it.
> >> >>> Then I started all the VMs in four cycles, 10 VMs per cycle and
> >> >>> waiting 240 seconds after each cycle.
> >> >>> Similar to the staggered start based on the amount of CPUs.
> >> >
> >> >> For me this is not enough info to even try to reproduce, I know little
> >> >> of vmm or vmd and have no idea what "derive" means in this context.
> >> >
> >> > This is a big bit of information that was missing from the original
> >>
> >> Well.. could have been better described indeed. :))
> >> " I created 41 additional VMs based on a single qcow2 base image.”
> >>
> >> > report ;) qcow has a concept of a read-only base image (or 'backing
> >> > file') which can be shared between VMs, with writes diverted to a
> >> > separate image ('derived image').
> >> >
> >> > So e.g. you can create a base image, do a simple OS install for a
> >> > particular OS version to that base image, then you stop using that
> >> > for a VM and just use it as a base to create derived images from.
> >> > You then run VMs using the derived image and make whatever config
> >> > changes. If you have a bunch of VMs using the same OS release then
> >> > you save some disk space for the common files.
> >> >
> >> > Mischa did you leave a VM running which is working on the base
> >> > image directly? That would certainly cause problems.
> >>
> >> I did indeed. Let me try that again without keeping the base image running.
> >
> > Right. As a safeguard, I would change the base image to be r/o.
>
> vmd(8) should treating it r/o...the config process is responsible for
> opening the disk files and passing the fd's to the vm process. In
> config.c, the call to open(2) for the base images should be using the
> flags O_RDONLY | O_NONBLOCK.
>
> A ktrace on my system shows that's the case. Below, "new.qcow2" is a new
> disk image I based off the "alpine.qcow2" image:
>
> 20862 vmd CALL open(0x7f7ffffd4370,0x26<O_RDWR|O_NONBLOCK|O_EXLOCK>)
> 20862 vmd NAMI "/home/dave/vm/new.qcow2"
> 20862 vmd RET open 10/0xa
> 20862 vmd CALL fstat(10,0x7f7ffffd42b8)
> 20862 vmd STRU struct stat { dev=1051, ino=19531847, mode=-rw------- ,
> nlink=1, uid=1000<"dave">, gid=1000<"dave">, rdev=78096304,
> atime=1616420730<"Mar 22 09:45:30 2021">.509011764, mtime=1616420697<"Mar 22
> 09:44:57 2021">.189185158, ctime=1616420697<"Mar 22 09:44:57
> 2021">.189185158, size=262144, blocks=256, blksize=32768, flags=0x0,
> gen=0xb64d5d98 }
> 20862 vmd RET fstat 0
> 20862 vmd CALL kbind(0x7f7ffffd39d8,24,0x2a9349e63ae9950c)
> 20862 vmd RET kbind 0
> 20862 vmd CALL pread(10,0x7f7ffffd42a8,0x68,0)
> 20862 vmd GIO fd 10 read 104 bytes
> "QFI\M-{\0\0\0\^C\0\0\0\0\0\0\0h\0\0\0\f\0\0\0\^P\0\0\0\^E\0\0\0\0\0\0\
> \0\0\0\0\0(\0\0\0\0\0\^A\0\0\0\0\0\0\0\^B\0\0\0\0\0\^A\0\0\0\0\0\0\0\
> \0\0\0\0\0\0\0\0\0\0\0\0\0\0\0\0\0\0\0\0\0\0\0\0\0\0\0\0\0\0\0\0\^D\0\
> \0\0h"
> 20862 vmd RET pread 104/0x68
> 20862 vmd CALL pread(10,0x7f7ffffd4770,0xc,0x68)
> 20862 vmd GIO fd 10 read 12 bytes
> "alpine.qcow2"
> 20862 vmd RET pread 12/0xc
> 20862 vmd CALL kbind(0x7f7ffffd39d8,24,0x2a9349e63ae9950c)
> 20862 vmd RET kbind 0
> 20862 vmd CALL kbind(0x7f7ffffd39d8,24,0x2a9349e63ae9950c)
> 20862 vmd RET kbind 0
> 20862 vmd CALL __realpath(0x7f7ffffd3ea0,0x7f7ffffd3680)
> 20862 vmd NAMI "/home/dave/vm/alpine.qcow2"
> 20862 vmd NAMI "/home/dave/vm/alpine.qcow2"
> 20862 vmd RET __realpath 0
> 20862 vmd CALL open(0x7f7ffffd4370,0x4<O_RDONLY|O_NONBLOCK>)
> 20862 vmd NAMI "/home/dave/vm/alpine.qcow2"
> 20862 vmd RET open 11/0xb
> 20862 vmd CALL fstat(11,0x7f7ffffd42b8)
>
>
> I'm more familiar with the vmd(8) codebase than any ffs stuff, but I
> don't think the issue is the base image being r/w.
>
> -Dave
AFAIKS, the issue is that if you start a vm modifying the base because it
uses it as a regular image, that r/o open for the other vms does not
matter a lot,
-OPtto
