Get DeafLetter
A weekly selection of signals, vulnerabilities and guides. Critical alerts remain optional.
You can unsubscribe at any time. Privacy policy.
Rapid7 published a detailed technical advisory on November 15, 2026 covering five vulnerabilities in ONLYOFFICE Docs that turn a read-only permission into arbitrary command execution as root inside the DocumentServer container. The research, conducted by Senior Security Researcher Jonah Burgess during an audit of the Zimbra Collaboration Suite, demonstrates how traditional minimal permissions fail to contain risk when the authentication architecture shares secrets across components with different privilege levels.
- An attacker with view-only access to a document can chain five vulnerabilities to achieve command execution as root in the official DocumentServer container.
- The chain starts from the shared JWT signing key between the Browser (Editor) component and the Inbox (DocBuilder) component, which allows an Editor token to invoke DocBuilder functionality without specific authorization. \li>The GetFontBinary() function accepts arbitrary absolute paths and had already been flagged in issue #147 on November 21, 2025, a year before the advisory was published.
- ONLYOFFICE confirmed fixes for all five vulnerabilities in version 10.0 on October 23, 2026; Rapid7 verified the fix by comparing builds 9.4.0-129 and 10.0.0-169.
The Chain: From Shared JWT to Root in the Container
The core of the problem lies in JWT token handling. In ONLYOFFICE Docs 9.4.0, the Browser (Editor) component and the Inbox (DocBuilder) component use the same effective signing key. The source code shows that the default-secret branch of getTenantSecret() assigns res the value utils.getSecretByElem(cfgSecretInbox), making Editor and Inbox cryptographically interchangeable.
This sharing has concrete operational consequences. A user with a valid Editor JWT can invoke the /docbuilder endpoint, which the system validates with the Inbox key but accepts because the two keys match. From there, the attacker triggers GetFontBinary() with existing absolute paths, reading arbitrary files on the container filesystem.
The first target is /etc/onlyoffice/documentserver/local.json, which contains the services.CoAuthoring.secret.session.string value. With this stolen key, the attacker can sign new JWTs using HMAC-SHA256, obtaining forged tokens that the system treats as authentic.
"An attacker with view-only access to a document can chain vulnerabilities in ONLYOFFICE Docs to execute arbitrary operating system commands as root inside the official DocumentServer container" — Rapid7 advisory
Five CVEs with Different Impacts, One Chain
The five assigned vulnerabilities cover distinct stages of the attack and carry differentiated CVSS scores. CVE-2026-113627 (7.1) documents the native file read via DocBuilder and the resulting JWT forgery. CVE-2026-110385 (8.7) represents the highest score: parameter injection in the SaveFile conversion leads to code execution. CVE-2026-97727 (6.5) describes a path traversal in the complete-save format that enables arbitrary file write. CVE-2026-113630 (7.8) is the command injection in the root-helper executed with elevated privileges. CVE-2026-113631 (6.5) concerns the exfiltration of unrendered document data via view-only export.
The combination of these flaws creates a coherent pipeline: file read, secret theft, token signing, arbitrary write, command execution. Each stage sets up the next, and none of the individual vulnerabilities requires permissions beyond viewing.
Tests were conducted on the official onlyoffice/documentserver:9.4.0 image with package 9.4.0-129. During the demonstration, JWT validation was enabled and no authentication checks or private address filters had been disabled. Conditions matched a standard installation.
Data Exfiltrated Beyond the Viewer
A distinctive aspect of the research is the ability to extract information the user interface never shows. The attacker obtains a reconstructed DOCX file containing custom XML, custom properties, and field instructions never displayed by the viewer, even when explicit download is disabled at the permission level.
This mechanism separates authorization logic from actual data exposure. The system blocks conventional download, but the view-only export pipeline allows reconstruction of complete document structures. Organizations relying on the disabled-download flag to protect sensitive content must account for this discrepancy between presented authorization and actual attack surface.
Alternate Entry Point with WOPI Enabled
The research identifies an attack variant that does not even require an account with view-only access. With WOPI_ENABLED=true and the WOPI IP filter in its default allow-all configuration, the same DocBuilder flaw provides an unauthenticated entry point. The condition depends on the reachability of an attacker-controlled document source, but it eliminates the requirement for any valid credential.
This branch widens the attack surface in enterprise environments where ONLYOFFICE is integrated with third-party platforms via the WOPI protocol. The default IP filter configuration applies no restrictions, and enabling the protocol exposes the same DocBuilder chain to an actor with no pre-existing trust relationship with the target organization.
What to Do Now
Administrators running ONLYOFFICE Docs should verify the installed version and plan an upgrade to build 10.0.0-169 or later, which Rapid7 confirmed as fixed through differential code analysis. Container-based deployments should replace the onlyoffice/documentserver:9.4.0 image with the updated release.
For organizations that cannot upgrade immediately, check whether WOPI is enabled without effective IP restrictions. The default allow-all WOPI filter configuration exposes the DocBuilder chain to unauthenticated entry. Restricting WOPI source IP addresses to controlled endpoints reduces this attack surface.
Teams hosting untrusted users with view-only access — partners, customers, auditors — must treat these accounts as potential vectors for server compromise. Separation of JWT keys for Browser and Inbox components is not documented as a verified mitigation in the Rapid7 dossier and cannot be considered a reliable barrier without specific testing on the local instance.
Organizations relying on the disabled-download flag to protect sensitive content must redefine access controls. The ability to exfiltrate unrendered data via view-only export makes this isolated measure insufficient.
Frequently Asked Questions
Is it necessary to disable ONLYOFFICE immediately?
The advisory does not document in-the-wild exploitation. The fix is available from version 10.0. Organizations running builds prior to 10.0.0-169 should assess exposure based on their user population and active integrations.
Does separating JWT keys partially mitigate the risk?
The research dossier does not specify whether a configuration with separate (non-shared) JWT keys for Browser and Inbox would block the chain. This point is not verified in the published documentation.
Which integrations beyond Zimbra are at risk?
Rapid7 conducted the research in the context of Zimbra Collaboration Suite. The specific impact on other integrations is not documented in the source. The WOPI variant, however, is independent of the integration platform.
Sources
Get DeafLetter
A weekly selection of signals, vulnerabilities and guides. Critical alerts remain optional.
You can unsubscribe at any time. Privacy policy.