From mboxrd@z Thu Jan 1 00:00:00 1970 Return-Path: Received: from gate001.proxmox.com (gate001.proxmox.com [IPv6:2a0f:8001:1:32::40]) by lore.proxmox.com (Postfix) with ESMTPS id A35F01FF0AA for ; Tue, 17 Nov 2026 16:17:39 +0100 (CET) Received: from gate001.proxmox.com (localhost.localdomain [127.0.0.1]) by gate001.proxmox.com (Proxmox) with ESMTP id 2C61F215FB; Tue, 17 Nov 2026 16:17:34 +0100 (CET) Message-ID: <0255a545-4848-4ab3-bbfd-9d120c66a586@proxmox.com> Date: Tue, 17 Nov 2026 16:17:28 +0100 MIME-Version: 1.0 User-Agent: Mozilla Thunderbird Beta Subject: Re: [PATCH qemu-server 15/34] restore: take over the guest uid from the backup where possible To: Thomas Lamprecht , pve-devel@lists.proxmox.com References: <20261116004703.2908294-1-t.lamprecht@proxmox.com> <20261116004703.2908294-16-t.lamprecht@proxmox.com> Content-Language: en-US From: Dominik Csapak In-Reply-To: <20261116004703.2908294-16-t.lamprecht@proxmox.com> Content-Type: text/plain; charset=UTF-8; format=flowed Content-Transfer-Encoding: 7bit X-Bm-Milter-Handled: 55990f41-d878-4baa-be0a-ee34c49e34d2 X-Bm-Transport-Timestamp: 1794928648275 X-SPAM-LEVEL: Spam detection results: 0 AWL 0.290 Adjusted score from AWL reputation of From: address DMARC_MISSING 0.1 Missing DMARC policy KAM_DMARC_STATUS 0.01 Test Rule for DKIM or SPF Failure with Strict Alignment (newer systems) RCVD_IN_DNSWL_MED -2.3 Sender listed at https://www.dnswl.org/, medium trust SPF_HELO_NONE 0.001 SPF: HELO does not publish an SPF Record SPF_PASS -0.001 SPF: sender matches SPF record Message-ID-Hash: RU6I6U4CEBLRUXAVIT2RJRSQSNEHYETI X-Message-ID-Hash: RU6I6U4CEBLRUXAVIT2RJRSQSNEHYETI X-MailFrom: d.csapak@proxmox.com X-Mailman-Rule-Misses: dmarc-mitigation; no-senders; approved; loop; banned-address; emergency; member-moderation; nonmember-moderation; administrivia; implicit-dest; max-recipients; max-size; news-moderation; no-subject; digests; suspicious-header X-Mailman-Version: 3.3.10 Precedence: list List-Id: Proxmox VE development discussion List-Help: List-Owner: List-Post: List-Subscribe: List-Unsubscribe: nits inline: On 11/16/26 1:53 AM, Thomas Lamprecht wrote: > A restore to an unused VMID, or over a VM without an identity, should > keep the uid of the backup if no other guest in the cluster uses it. > Take over that uid through the cluster-wide check of pve-guest-common > after the data is restored and the final config is written, while > holding the config lock. A VM that already had an identity keeps it, > and 'unique' prevents taking over the uid. Log the outcome, and only > warn if taking over the uid fails, as the restored VM has an identity > of its own. > > Refuse to restore a backup with a different uid over an existing VM > unless 'allow-identity-mismatch' is set in the create API or qmrestore. > This is independent of 'force', which permits overwriting the VM. With > the override, the VM keeps its identity, unless it has the uid of the > backup as pending uid from an unfinished restore. > > For PBS, backup providers and VMA, compare the uids before resetting > the replication state, restoring the firewall config or replacing the > data of the target. For VMA this also works for standard input, through > the header callback. Read the uid from the config that the restore > itself uses, not from the best-effort extraction of the API, and ignore > unknown members of 'meta', so that the uid of a backup from a newer > version is not lost when the property fails to parse. > > Take over the uid only after the live phase, the conversion to a > template and, with 'unique', the registration of the new MAC addresses. > A failed restore must not take the uid and so keep a retry at another > VMID from getting it. > > For a retry over a VM left behind by an unfinished restore to work, > record the uid of the backup as a pending change of 'meta'. Record it > after the uid check if the restore could take the uid over, also with > 'unique', and keep it in all configs that the restore writes. A VM > with that pending uid can be restored over without the override, and > without 'unique' the retry takes the uid over if no other guest uses > it. A pending uid does not count as in use. > > Taking over the uid removes the pending uid, whatever the outcome. If > the cluster-wide lock cannot be taken, it stays for a retry. Starting > the VM for a live restore keeps it while the 'create' lock is held. > Otherwise, applying pending changes drops it, and a clone does not get > it. > > Legacy tar archives predate guest identities and get no uid check. > Their config is only available after the data is restored, so only the > restored config can carry a pending uid. > > Signed-off-by: Thomas Lamprecht > --- > src/PVE/API2/Qemu.pm | 18 + > src/PVE/CLI/qmrestore.pm | 7 + > src/PVE/QemuServer.pm | 142 +++++- > src/test/QemuServerTest/FakeCFS.pm | 3 + > src/test/restore-config-expected/150.conf | 9 + > src/test/restore-config-input/150.conf | 10 + > src/test/run_api_tests.pl | 13 + > src/test/run_guest_identity_tests.pl | 77 +++ > src/test/run_qemu_restore_config_tests.pl | 559 +++++++++++++++++++++- > 9 files changed, 828 insertions(+), 10 deletions(-) > create mode 100644 src/test/restore-config-expected/150.conf > create mode 100644 src/test/restore-config-input/150.conf > > diff --git a/src/PVE/API2/Qemu.pm b/src/PVE/API2/Qemu.pm > index 28e8389a..395dc334 100644 > --- a/src/PVE/API2/Qemu.pm > +++ b/src/PVE/API2/Qemu.pm > @@ -1241,6 +1241,15 @@ __PACKAGE__->register_method({ > description => "Assign a unique random ethernet address.", > requires => 'archive', > }, > + 'allow-identity-mismatch' => { > + optional => 1, > + type => 'boolean', > + description => "Allow to overwrite an existing VM with a backup of another" > + . " guest, as detected by a different 'uid' in 'meta'. Unless the VM is" > + . " left over from an unfinished restore of the same backup, it keeps its" > + . " own uid.", > + requires => 'archive', nit: an optional boolean should have a default. we don't always have that on the existing properties, but we should add it for new ones. same nit holds for the equivalent container patch > + },