Australian cloud for business software — built and run in Sydney since 2002

NetLeverage
NetLeverage

Restoring a customer-held backup copy without your provider

Technical article
Backup & Recovery

About this article: This article is reviewed against the cited sources. This is source research, not lab validation. Check the stated versions and your own requirements before making changes.

Holding a copy of your backups in your own storage is a good start. It does not by itself mean you can restore. To restore without your provider's help you need six things besides the files: knowledge of the format, restore software you can run, the keys or passphrases, written steps, somewhere to restore to, and people who are allowed and able to do it. The copy itself also has to have been kept safe from deletion or change, which makes seven items to check in all.

A missing prerequisite can stay hidden until the copy is needed. This guide is for a customer who receives backup copies into their own cloud storage bucket or file server and wants to know what else must be in place if the provider is unavailable, for whatever reason. It is about readiness to restore. It does not cover checking that a file arrived intact, or moving to a new provider; those are separate tasks.

Why it matters

Australia's national cyber security agency, the Australian Signals Directorate (ASD), lists regular backups as one of its Essential Eight mitigation strategies. In the maturity model's first level, backups of "data, applications and settings" must be "retained in a secure and resilient manner", and "Restoration of data, applications and settings from backups to a common point in time is tested as part of disaster recovery exercises." (ASD, Essential Eight maturity model.) Note the word "applications and settings" as well as data: a restore usually needs more than the files alone.

NIST makes the same point about testing. Its contingency planning guide says backups "should be tested regularly to ensure that data are being stored correctly and that the files may be retrieved without errors or lost data", and that the recovery site should be tested "to ensure that the site supports the same backup configuration." (NIST SP 800-34 Rev. 1.) NIST's wording refers to tapes, but in our reading the principle carries to any media: the copy has to work on the equipment you will actually use.

A customer-held copy exists to give you independence. That independence is real only when the prerequisites below are also yours.

Seven prerequisites for an independent restore

The sections below are our framework. The sources support individual points, as cited, but the grouping into seven is our own.

1. Know what the copy actually is

Backup copies are not always plain files. Depending on the tool, a copy may be a set of ordinary files, a disk image, a database dump, or a structured repository whose contents are split, deduplicated and encrypted. Each needs a different way back.

As one example only, the open-source tool restic stores data in a repository and says "All data is stored in a restic repository", with a top-level file called config and other files named by their content hash (restic documentation, References). You cannot browse such a repository like a folder of documents. You need the tool, or something that understands its format. Other tools differ. We are not saying any provider uses restic; it is shown because its documentation is public.

Questions to settle: is the copy a full copy, or one link in a chain that depends on earlier copies? Are there index or catalogue files that must be kept with it? Which file or folder is the complete set?

2. Hold software you can run

If the copy needs particular software to read it, you need that software, in a compatible version, available without the provider. Check three things: where you would download it, whether a licence applies, and whether the version you hold can read copies written by the version the provider runs.

Format versions matter. The restic documentation describes a repository format version 2 that "adds support for compression", which tells you that format versions exist and that readers need to support them. Keep a copy of the installer, and record the version, with the backup.

3. Hold the keys

If the copy is encrypted, no software helps without the key. The restic documentation is blunt: "Please note that knowledge of your password is required to access the repository. Losing your password means that your data is irrecoverably lost." (restic, Preparing a new repository.)

For a copy that is yours, the question is who holds the key and how you would retrieve it with the provider unavailable. Key custody has its own decisions, which are covered in a separate article in this series. For this guide, the prerequisite is simple: you can name where the key is, and a second person can reach it.

4. Have the steps written down

Restoring a structured copy is a procedure. The steps include locating the right restore point, choosing the destination, and handling permissions. Write them as a runbook someone else could follow. Include the exact location of the copy, the software and version, where the key is, the order of steps, and how you would know the restore worked.

Documentation is also where you record what the copy leaves out. NIST's guide notes that contingency plans should include "roles and responsibilities, inventory information, assessment procedures, detailed recovery procedures, and testing" (SP 800-34 Rev. 1, section 2.2.7, describing an information system contingency plan). That list is a useful template for a restore note.

5. Have somewhere to restore to

A restored backup needs a destination: a server, storage and network that can hold it. Ask whether you have, or can quickly obtain, an empty machine of the right type, enough storage for the restored data, and a network path from the copy to the destination. A restore that runs over a slow internet link can take far longer than the backup took.

The destination also affects what is restored. The restic documentation notes that restoring full Windows security descriptors "is only possible when the user has the SeRestorePrivilege , SeSecurityPrivilege and SeTakeOwnershipPrivilege privileges or is running as administrator", and that otherwise only part of the permissions are restored. A restore by the wrong account can succeed technically and still lose access settings. This is one tool on one platform, but the lesson, that the account doing the restore matters, applies widely.

Applications add another layer. ASD lists "applications and settings" as well as data. A database restored without the application that reads it, or without its configuration, is not a working service. Record what must be installed first.

6. Name the people and their access

On the day, someone must be able to log in to your storage, run the software, read the key and decide what to restore. Check these points:

  • At least two named people can do each step, so one absence does not stop the restore.
  • Their access to the storage holding the copy works without the provider's systems, and does not depend on a single person's mobile phone or on email that would be down in the same event.
  • Someone has authority to start a restore on production systems, and someone has the skill. They may not be the same person.

If you rely on an outside technician, record the arrangement and how to reach them.

7. Keep the copy safe from change

A customer-held copy is not independent if one set of credentials can delete it. ASD's first maturity level includes that "Unprivileged user accounts are prevented from modifying and deleting backups." Applied here, our reading is that the account which delivers copies into your storage should not be able to remove earlier copies, and the account you use for restoring should be separate.

If the copy sits in Amazon S3, Amazon's documentation describes versioning as a way to keep earlier states: "Versioning-enabled buckets can help you recover objects from accidental deletion or overwrite." Amazon notes that "By default, S3 Versioning is disabled on buckets, and you must explicitly enable it", and that "Each version of an object is the entire object", so storage is charged for each version. (Amazon S3 User Guide, Retaining multiple versions of objects with S3 Versioning.) Versioning on its own does not stop a person or account with the right permission from permanently destroying those versions. Amazon says "When versioning is enabled, a simple DELETE cannot permanently delete an object", but also that "To delete versioned objects permanently, you must use DELETE Object versionId". Amazon's permissions reference lists "s3:DeleteObjectVersion" as "Required if you want to remove a specific version of an object by specifying versionId in the request." (Deleting object versions; Required permissions for Amazon S3 API operations.) In our reading, that means two further decisions. First, keep that delete permission away from the account that delivers copies and from everyday administrator logins. Second, where copies must survive even a mistaken or compromised permitted login, add retention or immutability. Amazon's S3 Object Lock "can help prevent Amazon S3 objects from being deleted or overwritten for a fixed or variable amount of time, or indefinitely." In its compliance mode a protected version "can't be overwritten or deleted by any user, including the root user", while in governance mode "users can't overwrite or delete an object version or alter its lock settings unless they have special permissions" (Object Lock). The two modes differ in a way that matters for your choice. Governance mode can be overridden: Amazon says "To override or remove governance-mode retention settings, you must have the s3:BypassGovernanceRetention permission and must explicitly include x-amz-bypass-governance-retention:true as a request header", so anyone holding that permission can delete locked versions or shorten the retention. Compliance mode cannot: Amazon says the retention mode "can't be changed, and its retention period can't be shortened", and adds that "The only way to delete an object under the compliance mode before its retention date expires is to delete the associated AWS account." (Object Lock; Bypassing governance mode.) In our reading, the choice follows from what you need to resist. If you want protection against most mistakes but still need someone with special permission to be able to clear a wrongly locked copy, governance mode fits, provided you keep the bypass permission away from the delivering account and from everyday logins. If you want no login at all, including the root user, to be able to remove copies before they expire, compliance mode is the option, but a retention period set too long, or on the wrong data, cannot be undone early, and storage is charged for the whole period. Choose the period deliberately, and trial the settings first (Amazon suggests governance mode can be used "to test retention-period settings before creating a compliance-mode retention period"). Amazon also offers MFA delete, which requires two forms of authentication to permanently delete a version (Configuring MFA delete). Other storage services have their own equivalents.

A prerequisite list you can fill in

Prerequisite What to record Check
Format Type of copy, full or chained, which files form a complete set A second person can find the complete set
Software Name, version, where the installer is held, licence status You can install it on a clean machine
Keys Where each key lives, who can retrieve it Someone other than the owner can reach it
Steps Written runbook with exact locations A colleague follows it unaided
Target Machine, storage, network, accounts needed, and an isolated test network with no route to production Capacity matches the data size; test restores start only after isolation is confirmed
People Two people per step, authority, outside help Their access works without the provider
Protection Who can delete or overwrite the copy, and whether retention or immutability is set Delivery and restore use separate accounts; permanent deletion is restricted

What a restore test proves, and what it does not

Restoring one folder shows that the path from that copy to that folder works: you could find the copy, read it with the software and key, and write the files. It does not show that the whole service can be recovered, and it does not show the recovery time the business has agreed. For that, a test needs to include a representative sample of what the service depends on, not only documents:

  • the application or database and its configuration settings, restored and started, but only inside an isolated test environment (see the next paragraph);
  • the permissions and accounts that decide who can open what, checked by signing in as a normal user;
  • a realistic amount of data, so the transfer time reflects the real size;
  • a note of the elapsed time at each step, compared with the recovery time you want.

Isolate the test before anything is started. A restored application carries its old settings: addresses of live servers, mail settings, scheduled jobs, saved credentials and integrations. Started on a normal network it can behave as though it were production. In our recommendation, which is not drawn from the sources cited, restore to a separate, authorised test network or environment before anyone starts the application. That environment should have no route to production services or live data, and should block outbound mail and other outbound traffic, so the restored copy cannot send email, run jobs or connect to live systems. Disable scheduled tasks, integrations and mail settings in the restored copy before its first start, and use test credentials in place of live ones where you can. Keep the person who may authorise the test named in your runbook, and confirm the isolation (for example, that a test message does not leave) before starting the application, not after.

Our recommendation is to treat a single-folder restore as the first rung. Full readiness comes only from repeating the test across the application, settings and permissions, and timing it. Until then, record what has and has not been shown.

Questions to put to your provider

Put these to your provider in writing. Answers may be short, but the gaps will tell you what to cover yourself.

  • What exactly is the format of the delivered copy, and is it documented publicly or only to customers?
  • What software reads it, in which version, and where can I obtain it without your help?
  • Is the copy encrypted, and who holds the key or passphrase?
  • Is each delivered copy complete on its own, or does it depend on earlier ones?
  • Do you provide written restore steps for a customer working alone, and when were they last confirmed to work?
  • Can the account that delivers to my storage delete or change what is already there?

A worked example (illustrative)

The following is invented. A firm receives nightly backup copies into a storage bucket it owns. Its owner decides to test independence, and writes the seven prerequisites on one page. Four are answered at once. The owner finds that the restore software is only installed on the provider's systems, that the passphrase is known to one staff member, and that no written steps exist for restoring alone.

The owner asks the provider for the software version and a written restore note, records the passphrase's location in a sealed envelope held by a second director, and arranges for an outside technician to attend a test. On the test day the technician installs the software on a clean machine, restores a single folder from last week's copy, and compares it with the original. It restores, but takes longer than expected because of a slow connection to the storage. The owner notes that, and adds a line about transfer time to the page.

The owner also records what this first test did not show. The accounting application, its settings and the staff permissions were not restored, and the full data size was not transferred, so the firm cannot yet claim a whole-service recovery or an agreed recovery time. A second test is booked to restore the application with a sample of its data into an isolated test network, with outbound mail and scheduled jobs switched off before its first start, then sign in as an ordinary user and time each step.

Risks and limits

  • This guide is general. It does not describe how any particular provider delivers or formats copies. Ask yours.
  • Examples from restic show one tool's documented behaviour. Other tools differ in format, keys, versions and permissions.
  • The Essential Eight is a framework for Australian organisations. Applying it depends on your own maturity target, and this guide does not assess compliance. The model is updated regularly, so recheck the wording and the version you are measured against.
  • Object Lock choices are hard to reverse: governance mode can be bypassed by anyone with the bypass permission, and compliance mode cannot be shortened by anyone until it expires. Check the current Amazon documentation and your own retention needs before setting either.
  • A written list is not a test. The sources point to testing, and this article has not tested any restore.
  • Keeping software, keys and documents yourself adds work and a security responsibility: stored keys and installers must themselves be protected.

Next step

Write the seven-row list above for your own copy, then start testing with one item, such as having a colleague locate the key and install the restore software on a clean machine. Treat that as the first step only: plan a later test that restores a representative application, its settings and permissions, and times the result. If you are reviewing a backup service, the Backup and Recovery page is a starting point for that conversation; ask for the scope in writing.

Sources

First accessed 15 November 2026; the ASD and Amazon S3 pages were rechecked on 16 November 2026.

Learning notes

  • Having the files is possession; being able to restore is a separate result that needs format, software, keys, steps, a target and people.
  • If a copy is encrypted, the key is part of the backup. Lose it and the data is gone.
  • The account that delivers copies should not be the account that can delete them. Versioning alone does not stop permanent deletion by a permitted account, so separate permissions and add retention or immutability where needed. S3 governance-mode locks can be bypassed by users with the bypass permission; compliance-mode locks cannot be removed early by anyone, so choose with care.
  • Start a restored application only in an isolated test environment, so it cannot reach production services, send mail, run jobs or touch live data.
  • A written list helps, but a test shows only what it covered. One restored folder proves that path, not the whole service. Full readiness needs a representative application, its settings and permissions, and a timed run.