Ceph Object Gateway CVE-2026-54330

Exploitation requires possession of a valid SigV4 presigned URL (using X-Amz-Algorithm=AWS4-HMAC-SHA256) issued against a Ceph RGW endpoint running a vulnerable version (Ceph < 19.2.6 or < 20.2.4). The attacker must be able to submit HTTP requests to the RGW endpoint - no special network position beyond reachability is required. The attacker supplements the presigned URL query string with additional x-amz-* headers absent from X-Amz-SignedHeaders; the impact depends on which x-amz-* headers RGW honors (e.g., x-amz-acl for ACL changes, x-amz-copy-source for object reads, x-amz-server-side-encryption for encryption class changes). Users without any presigned URL for the target RGW endpoint cannot exploit this flaw - anonymous access to the RGW service alone is insufficient. The rgw_sigv4_insecure option, when set to true, re-enables the vulnerable behavior and must be set only transiently during multisite upgrades.

The provided CVSS 8.1 vector (AV:N/AC:L/PR:L/UI:N/S:U/C:H/I:H/A:N) is consistent with the technical description: the attack is network-reachable, low-complexity once a presigned URL is in hand, and yields high confidentiality and integrity impact through unauthorized S3 operations. EPSS data was not supplied in the input, so exploitation probability cannot be quantified from that signal. The CVE is not listed in CISA KEV and no public POC is identified, which collectively suggest opportunistic exploitation is not yet widespread. The real-world risk is deployment-context sensitive: organizations that distribute presigned PUT or COPY URLs to third parties, customers, or automated clients - a common S3-compatible workflow pattern - face meaningful privilege-escalation exposure, and even a moderately sized enterprise running on-premises Ceph as an S3-compatible backend should treat patching as time-sensitive. The C:H rating is credible because an attacker could append x-amz-copy-source headers to redirect a PUT into a read of another object, and I:H is credible through ACL manipulation or storage-class injection; A:N is consistent with no described denial-of-service path.

An attacker who receives a presigned PUT URL - legitimately issued by a storage application to allow an authenticated upload - modifies the HTTP request before submission by appending an unsigned x-amz-acl: public-read header or an x-amz-copy-source header pointing to a protected object. Because RGW honors the extra headers without checking that they appear in X-Amz-SignedHeaders, the object is stored with a public ACL or the protected source object's contents are read into the destination, neither action having been authorized by the URL signer. No public POC has been identified; the attack requires only standard HTTP tooling and knowledge of the SigV4 bypass.