Description
SGLang through 0.5.22 with the Mooncake disaggregation backend contains an uncontrolled memory allocation vulnerability in handle_staging_rsp() that lets unauthenticated network-adjacent attackers exhaust Prefill memory. Attackers can obtain the rank address from the /route endpoint and send STAGING_RSP messages with huge chunk_idx values under fresh room IDs, causing the scheduler to be OOM-killed.
Published: 2026-10-19
Score: 7.1 High
EPSS: n/a
KEV: No
Impact: Denial of Service
Action: Apply Patch
AI Analysis

Impact

SGLang versions through 0.5.22 contain an uncontrolled memory allocation bug in the handle_staging_rsp() routine of the Mooncake disaggregation backend. An attacker need only be network‑adjacent and unauthenticated; by first querying the /route endpoint to obtain a rank address, the attacker can then send STAGING_RSP messages that contain overly large chunk_idx values under new room identifiers. The scheduler threads performing the allocation will grow to consume all prefilling memory and are eventually killed by the operating system, causing a service outage.

Affected Systems

The vulnerability affects any deployment of the sgl-project:sglang package at or below version 0.5.22, specifically when the Mooncake disaggregation backend is enabled. No other versions or products are listed as affected.

Risk and Exploitability

The CVSS score of 7.1 labels the flaw as high‑severity, and while no EPSS score is currently available a lack of a KEV listing does not reduce the likelihood of exploitation. Because the attack requires only network adjacency and no authentication, an adversary with access to the same local network segment can trigger an out‑of‑memory condition and cause a denial of service. The risk is therefore significant for exposed deployments, though it is limited to environments where the disaggregation backend is reachable from unauthenticated clients.

Generated by OpenCVE AI on October 19, 2026 at 15:46 UTC.

Remediation

No solution or workaround provided in the CVE record.

OpenCVE Recommended Actions

  • Upgrade SGLang to version 0.5.23 or later where the bug is fixed.
  • Restrict network access to the Mooncake disaggregation backend so that only authenticated or internal users can reach the STAGING_RSP endpoint, for example by configuring firewall or VPN rules.
  • Enforce rate limiting or input validation on STAGING_RSP messages to prevent excessively large chunk_idx values from exhausting memory.

Generated by OpenCVE AI on October 19, 2026 at 15:46 UTC.

Tracking

Sign in to view the affected projects.

Advisories

No advisories yet.

History

Mon, 19 Oct 2026 17:30:00 +0000

Type Values Removed Values Added
Metrics ssvc

{'options': {'Automatable': 'no', 'Exploitation': 'none', 'Technical Impact': 'partial'}, 'version': '2.0.3'}


Mon, 19 Oct 2026 14:00:00 +0000

Type Values Removed Values Added
Description SGLang through 0.5.22 with the Mooncake disaggregation backend contains an uncontrolled memory allocation vulnerability in handle_staging_rsp() that lets unauthenticated network-adjacent attackers exhaust Prefill memory. Attackers can obtain the rank address from the /route endpoint and send STAGING_RSP messages with huge chunk_idx values under fresh room IDs, causing the scheduler to be OOM-killed.
Title SGLang through 0.5.22 Memory Exhaustion via Mooncake STAGING_RSP chunk_idx
Weaknesses CWE-770
References
Metrics cvssV3_1

{'score': 6.5, 'vector': 'CVSS:3.1/AV:A/AC:L/PR:N/UI:N/S:U/C:N/I:N/A:H'}

cvssV4_0

{'score': 7.1, 'vector': 'CVSS:4.0/AV:A/AC:L/AT:N/PR:N/UI:N/VC:N/VI:N/VA:H/SC:N/SI:N/SA:N'}


Subscriptions

No data.

cve-icon MITRE

Status: PUBLISHED

Assigner: VulnCheck

Published:

Updated: 2026-10-19T16:34:55.350Z

Reserved: 2026-10-19T13:25:27.311Z

Link: CVE-2026-112270

cve-icon Vulnrichment

Updated: 2026-10-19T15:35:59.255Z

cve-icon NVD

Status : Deferred

Published: 2026-10-19T14:17:04.627

Modified: 2026-10-19T17:17:14.960

Link: CVE-2026-112270

cve-icon Redhat

No data.

cve-icon OpenCVE Enrichment

Updated: 2026-10-19T16:00:15Z

Weaknesses
  • CWE-770

    Allocation of Resources Without Limits or Throttling