IF YOU WOULD LIKE TO GET AN ACCOUNT, please write an
email to Administrator. User accounts are meant only to access repo
and report issues and/or generate pull requests.
This is a purpose-specific Git hosting for
BaseALT
projects. Thank you for your understanding!
Только зарегистрированные пользователи имеют доступ к сервису!
Для получения аккаунта, обратитесь к администратору.
Use unique v2vtransferN.json for every transfer, like like
out.paramsN.json. If we keep the temporary files after the import, the
content of the file can help debugging.
Related to #46
This log[1] is proved confusing for developers:
cannot read /etc/vdsm/vdsm.id, using any host: [Errno 2] No such file or directory: '/etc/vdsm/vdsm.id'
This is actually a debug message, that can help RHV developers to
understand where virt-v2v was running. To make this more clear, in the
expected case when running on non-oVirt host, we log now:
not an oVirt host, using non-oVirt host
For any other error, we can degrade to remote transfer, but we want to
warn the user about this, so we log:
warning: cannot read host id, using non-oVirt host: [Errno 2] No such file or directory: '/etc/vdsm/vdsm.id'
[1] https://github.com/libguestfs/virt-v2v/issues/46#issuecomment-2039604719
RWMJ: Changed "any host" to "non-oVirt host" to make it clearer.
We write the json inside a with block, so what actually happen is:
1. Computing the values
2. Dumping the json to stdout
3. Closing the connection
If some code in sdk.Connection.close() write something to stdout, it
will corrupt our json output. Avoid this by moving the output out of the
with block.
This will not prevent bad code to write something before the json. We
need to log bad json in the Ocaml code to reveal such issue.
If parsing params fail we get unhelpful error:
virt-v2v: error: internal error: invalid argument:
/tmp/v2v.CtVpBL/v2vtransfer.json: JSON parse error: end of file expected
near 'e'
Change the parsing code to read the entire json text and include it in
the error message. This will make it clear why the json could not be
parsed, and help to find the root cause for having invalid json.
Example error:
$ cat out.params0.json
This is not a json document
$ python output/rhv-upload-transfer.py out.params0.json
Traceback (most recent call last):
File "/home/nsoffer/src/virt-v2v/output/rhv-upload-transfer.py", line 261, in <module>
raise RuntimeError(f"Cannot parse params {data!r}: {e}").with_traceback(
File "/home/nsoffer/src/virt-v2v/output/rhv-upload-transfer.py", line 259, in <module>
params = json.loads(data)
^^^^^^^^^^^^^^^^
File "/usr/lib64/python3.12/json/__init__.py", line 346, in loads
return _default_decoder.decode(s)
^^^^^^^^^^^^^^^^^^^^^^^^^^
File "/usr/lib64/python3.12/json/decoder.py", line 337, in decode
obj, end = self.raw_decode(s, idx=_w(s, 0).end())
^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
File "/usr/lib64/python3.12/json/decoder.py", line 355, in raw_decode
raise JSONDecodeError("Expecting value", s, err.value) from None
RuntimeError: Cannot parse params 'This is not a json document\n': Expecting value: line 1 column 1 (char 0)
Related to #46
When converting this phony Windows guest we didn't set the VIRTIO_WIN
environment variable. As a result the test output could change (and
break) depending on if the virtio-win package was installed or not.
Point VIRTIO_WIN to the fake virtio-win directory (as we do with other
Windows tests). We also need to modify the test output since the fake
virtio-win does contain enough for virt-v2v to think that virtio
driver installation was successful.
Fixes: commit af1f7b571fb21fdbfa2974ee54a22b4fdcbc886a
The existing test used a phony Windows image, but this has some
shortcomings including not testing any virtio paths. Add a second
test of a phony Fedora image to get a little more coverage.
This function was used by -o libvirt / -o local to check that we have
UEFI firmware for the target architecture. All it does is called
find_uefi_firmware early so that we fail early.
The same check was copied into -o qemu, probably wrongly as we call
find_uefi_firmware just below, so the early fail is not necessary.
This allows us to remove error_unless_uefi_firmware completely since
it is no longer used anywhere.
nbdinfo --size prints the size on stdout, causing it to appear in
virt-v2v output. Using --can connect instead is silent.
Note that nbdinfo --can connect was added in libnbd 1.9.2 (Jul 2021)
and our previous minimum version of libnbd was 1.9.3 so we're OK.
However since neither of these was a stable version I also updated the
minimum libnbd requirement to 1.10 (Sep 2021), and added a proper
check in ./configure
Fixes: commit fb72e059863a60503b6011b8590c25c3a010a58f
The documentation suggested (correctly) that the -ip option did not
cover all cases where we make an ssh connection to either VMware or
Xen. However with the recent change where we use nbdkit-ssh-plugin in
preference to ssh/scp (commit fb72e05986, "virt-v2v: -i vmx: Replace
external ssh/scp with nbdkit-ssh-plugin") this should no longer be
true for VMware. And I don't think it was true for Xen for a while.
Therefore remove this paragraph in the documentation (in two places).
See-also: https://bugzilla.redhat.com/show_bug.cgi?id=1854275
See-also: https://lists.libguestfs.org/archives/list/guestfs@lists.libguestfs.org/thread/CKBEYJHUZWLDUUDWAJM2JU6QIKZGTD6F/
Suggested-by: Laszlo Ersek
If you use a -i vmx ssh filename containing '*' then it will expand
the glob at the remote side. New scp is weird and silently creates a
directory on the local side. For example suppose there's a remote
file literally called "/tmp/test*" (ie. the '*' is part of the
filename) and you thought you could copy that to local using:
scp 'remote:/tmp/test*' '/tmp/test*'
scp treats the first parameter (only) as a wildcard and if there are
multiple files matching the wildcard /tmp/test*, will create a local
directory literally called "/tmp/test*/" and put the files into it.
Who expected any of that?
You might think that double quoting (as we used to use) might work,
but that breaks with spaces and quotes. I guess scp is using
different code paths internally for glob versus everything else.
The only way to really make this work is to stop using scp entirely
and just use sftp directly. The easiest way to use sftp is to use
nbdkit-ssh-plugin. We already depend on nbdkit-ssh-plugin, nbdcopy
and nbdinfo for other parts of virt-v2v, so might as well use the
whole lot here.
One advantage of this change is that now the -ip (input password)
parameter actually works in -i vmx -it ssh mode.
Other approaches that would have been possible:
- Use the OCaml NBD library instead of nbdcopy/nbdinfo
- Use 'nbdkit -U - ssh --run ...'
- Direct binding to libssh.
See also commit e2af12ba69c4463bb73d30db63290a887cdd41eb
("input: -i vmx: Remove support for openssh scp < 8.8")
See also commit 22c5b98ab78c734b478c26e14ee62e2a065aaa0c
("-it ssh: Double quote ssh command which tests remote file exists")
See also https://unix.stackexchange.com/a/587710
Reported-by: Ming Xie
Fixes: https://issues.redhat.com/browse/RHEL-21365
Reviewed-by: Laszlo Ersek <lersek@redhat.com>
This filter retries a request on failure. Add a flag to make this
filter optional, but default to using it if available (so there's no
change to default behaviour).
The reason we want to make this optional is that the next commit will
want to use nbdkit-ssh-plugin to probe to see if a remote file is
present, and we don't want to retry that probe on failure.
Reviewed-by: Laszlo Ersek <lersek@redhat.com>
Xml.uri is a convenient way to pass the multiple ssh fields to
virt-v2v and is still used internally by the -i vmx code. However
don't leak this awkward implementation into the new ssh.ml module.
Like Nbdkit_ssh, we'll only deal with the explicit (password, port,
server, user, path) fields separately.
Note after this refactoring:
- The new Ssh module interface looks broadly similar to the
Nbdkit_ssh interface.
- vmx_source_of_arg assertions are no longer required inside the new
Ssh module, which was a mild layering violation before.
Reviewed-by: Laszlo Ersek <lersek@redhat.com>
Since we use the input password in various places in the VMX module,
store the input password in vmx_source. This neutral refactoring
makes later changes simpler.
Reviewed-by: Laszlo Ersek <lersek@redhat.com>
The existing 'scp_from_remote_to_temporary' wrapper around scp was
pretty weird (I think from much patching without refactoring).
Simplify it so it no longer generates the output filename, and rename
it accordingly to 'download_file' (as we will soon remove the need for
and dependency on 'scp').
Reviewed-by: Laszlo Ersek <lersek@redhat.com>
This is a straight refactor of the existing code that handles ssh/scp
into a new module. In this commit I just copy the code around without
doing any cleanup; cleanup will follow in subsequent commits.
Reviewed-by: Laszlo Ersek <lersek@redhat.com>
This reverts the following commit:
commit d265639c2ab31418cfdbdedd0cc3e68cf290d834
Author: Richard W.M. Jones <rjones@redhat.com>
Date: Thu Jul 25 14:52:42 2019 +0100
v2v: -i vmx: Use scp -T option if available to unbreak scp (RHBZ#1733168).
See also the referenced bug:
https://bugzilla.redhat.com/show_bug.cgi?id=1733168
My rationale for removing this option is that since we now require
OpenSSH 8.8 we must be using sftp for file transfer so we no longer
need to defeat the check for correct expansion of wildcards. That
check was only relevant for OpenSSH <= 8.7 using the old scp protocol.
Reverts: commit d265639c2ab31418cfdbdedd0cc3e68cf290d834
