On Wed, 08/16 09:55, Stefan Hajnoczi wrote:
> On Wed, Aug 16, 2017 at 03:20:59PM +0800, Fam Zheng wrote:
> > +class BaseVM(object):
> > + GUEST_USER = "qemu"
> > + GUEST_PASS = "qemupass"
> > + ROOT_PASS = "qemupass"
> > +
> > + # The script to run in the guest that builds QEMU
> > + BUILD_SCRIPT = ""
> > + # The guest name, to be overridden by subclasses
> > + name = "#base"
> > + def __init__(self, debug=False):
> > + self._guest = None
> > + self.ssh_port = 20022
>
> Only one instance of this test can be run per machine due to the
> hardcoded SSH port number on the host.
>
> It is possible to use:
>
> -netdev user,id=vnet,hostfwd=:0.0.0.0:0-:22
>
> and then query the port number:
>
> (qemu) info usernet
> VLAN -1 (vnet):
> Protocol[State] FD Source Address Port Dest. Address Port RecvQ
> SendQ
> TCP[HOST_FORWARD] 15 * 36089 10.0.2.15 22 0 > 0
>
> The host port is 36089 in this example.
>
> I'm not aware of a QMP equivalent for "info usernet". It may be
> necessary to implement a query-usernet command if you don't want to use
> HMP.
Sounds good. Will try. Using HMP is good enough to start with, we can add
query-usernet on top.
>
> > + self._tmpdir = tempfile.mkdtemp(prefix="qemu-vm-")
> > + atexit.register(shutil.rmtree, self._tmpdir)
> > +
> > + self._ssh_key_file = os.path.join(self._tmpdir, "id_rsa")
> > + open(self._ssh_key_file, "w").write(SSH_KEY)
> > + subprocess.check_call(["chmod", "600", self._ssh_key_file])
> > +
> > + self._ssh_pub_key_file = os.path.join(self._tmpdir, "id_rsa.pub")
> > + open(self._ssh_pub_key_file, "w").write(SSH_PUB_KEY)
> > +
> > + self.debug = debug
> > + self._stderr = sys.stderr
> > + self._devnull = open("/dev/null", "w")
> > + if self.debug:
> > + self._stdout = sys.stdout
> > + else:
> > + self._stdout = self._devnull
> > + self._args = [ \
> > + "-nodefaults", "-enable-kvm", "-m", "2G",
> > + "-smp", os.environ.get("J", "4"), "-cpu", "host",
>
> Can this be a command-line option in main() and a constructor argument
> instead of an environment variable? That would be cleaner because the
> use of "J" might be surprising to someone who happens to have it set in
> their environment.
Yes, "J" should be handled in the Makefile.
>
> > + "-netdev", "user,id=vnet,hostfwd=:0.0.0.0:%d-:22" %
> > self.ssh_port,
> > + "-device", "virtio-net-pci,netdev=vnet",
> > + "-vnc", ":0,to=20",
> > + "-serial", "file:%s" % os.path.join(self._tmpdir,
> > "serial.out")]
> > +
> > + self._data_args = []
> > +
> > + def _download_with_cache(self, url):
> > + cache_dir = os.path.expanduser("~/.cache/qemu-vm/download")
> > + subprocess.check_call(["mkdir", "-p", cache_dir])
>
> os.makedirs()
OK.
>
> > + fname = os.path.join(cache_dir, hashlib.sha1(url).hexdigest())
> > + if os.path.exists(fname):
> > + return fname
> > + logging.debug("Downloading %s to %s...", url, fname)
> > + subprocess.check_call(["wget", "-c", url, "-O", fname +
> > ".download"],
> > + stdout=self._stdout, stderr=self._stderr)
>
> It might be important to support image file updates without manually
> deleting ~/.cache. You can probably make wget send an HTTP
> If-Modified-Since header or something similar. There are 3 cases:
>
> 1. Existing file is up-to-date. No download after HEAD request.
> 2. Existing file is outdated, download the newest version.
> 3. Failure or timeout, use the old version for now.
Since the URLs are not https, maybe add sha256sum of images in the subclasses so
it is MITM-safe? It is also much simpler code, I think.
>
> > + subprocess.check_call(["mv", fname + ".download", fname],
> > + stdout=self._stdout, stderr=self._stderr)
>
> os.rename()
OK.
Fam
