Re: Guest agent guest-exec memory usage

["Daniel P. Berrangé" <berrange@redhat.com>]

On Mon, Jun 17, 2024 at 04:05:07PM +0300, Konstantin Kostiuk wrote:
> Hi All,
> 
> During the investigation of a possible memory leak in the `guest-exec`
> command of guest-agent, I found unexpected behavior for me. When we execute
> the `guest-exec` command with `capture-output = true`, guest-agent stores
> stdout/stderr until someone calls `guest-exec-status`.
> 
> Just for testing, I executed the `man man` command 1000 times with
> `capture-output = true` and guest-agent allocated 36Mb to store the results
> and it fully depends on output size.
> 
> I want to ask your opinion about this behavior. Is this behavior expected
> or not? Should we store all output forever or should we limit it by memory
> size/execution time/execution count?

If 'guest-exec' is enabled in the agent, this says that the host OS users
of the guest-agent are inherently trusted. If they're issuing many guest-exec
commands with capture-output = true, and then batching up the calls to
guest-exec-status for a later time, that's their perogative and not any
worse than other things they can already do with 'guest-exec', such as
requesting a fork-bomb.

Or to put it another way, if the guest OS admin is worried about malicious
usage of 'guest-exec' they should block this command entirely.

One of the patches I have in this series:

  https://lists.nongnu.org/archive/html/qemu-devel/2024-06/threads.html

proposes blocking guest-exec by default in the agent, as it is an
inherantly risky command to expose.

With regards,
Daniel
-- 
|: https://berrange.com      -o-    https://www.flickr.com/photos/dberrange :|
|: https://libvirt.org         -o-            https://fstop138.berrange.com :|
|: https://entangle-photo.org    -o-    https://www.instagram.com/dberrange :|