GHSA-FF87-73JP-F3V4

Vulnerability from github – Published: 2026-11-18 20:42 – Updated: 2026-11-18 20:42
VLAI
Summary
Atomic Server: Unauthenticated full-response SSRF in Atomic-Server /bookmark endpoint
Details

Summary

Atomic-Server exposes a GET /bookmark endpoint that fetches an attacker-supplied URL server-side, parses the response, and reflects the fetched content back to the caller. The endpoint requires no authentication and performs no authorization of its own, and the underlying fetch helper applies no allow-list or private-IP deny-list. An unauthenticated attacker with network reach to the server can force it to issue arbitrary outbound HTTP GET requests to internal-only hosts (loopback services, internal admin panels, link-local cloud-metadata) and read the response body back. This is a full-response SSRF, not a blind one.

Root cause (source trace)

  1. The bookmark endpoint is registered unconditionally for every server: server/src/appstate.rs:69 (store.add_endpoint(plugins::bookmark::bookmark_endpoint())).
  2. An unauthenticated caller on the standard GET route becomes ForAgent::Public: server/src/helpers.rs:168-183 (get_client_agent).
  3. In get_resource_extended, when the subject matches an endpoint path the function returns via call_endpoint BEFORE the hierarchy::check_read authorization call: lib/src/db.rs:1010-1017 (if is_endpoint { return self.call_endpoint(...) }, and check_read sits just after at :1017, never reached).
  4. call_endpoint performs no authorization itself; it only matches the path and invokes the handler: lib/src/db.rs:662-706.
  5. The bookmark handler discards the caller (for_agent: _) and fetches the url query parameter directly: server/src/plugins/bookmark.rs:45-76 (let content = fetch_data(&path)?;).
  6. fetch_body validates only url.starts_with("http"): lib/src/client/helpers.rs:78-91. A repo-wide grep finds no is_loopback / is_private / 127.0.0.1 / 169.254 / localhost deny-list anywhere in lib/src or server/src, and no host allow-list.
  7. The fetched body is HTML-parsed and written into the resource name, description, image and preview (markdown) fields, which are returned to the caller: server/src/plugins/bookmark.rs:79-110. This is what makes it a full-response (not blind) SSRF; even plaintext internal responses are wrapped into the preview markdown.

Proof of concept (unauthenticated)

GET /bookmark?url=http://127.0.0.1:<internal-port>/&name=x        # internal loopback service
GET /bookmark?url=http://169.254.169.254/latest/meta-data/&name=x # IMDSv1 cloud metadata

No authentication headers. The returned resource reflects the fetched content in its name / description / preview fields, so the attacker reads the internal target's response body.

Proof of omission

The endpoint-dispatch path returns at lib/src/db.rs:1011 (return self.call_endpoint(...)) before the check_read call at :1017, and there is no authorization inside call_endpoint (:662-706). The handler discards the caller (for_agent: _, bookmark.rs:49), so the endpoint can never enforce per-agent access. fetch_body validates only url.starts_with("http") (helpers.rs:83); no loopback/private/link-local deny-list exists anywhere in the source. No corresponding SSRF/security entry exists in CHANGELOG.md.

Impact

An unauthenticated attacker with network reach to the server can: - read internal-only HTTP services not exposed to the internet (admin panels, dashboards, unauthenticated microservices on loopback or the private network); - read cloud metadata reachable via plain GET (IMDSv1 and similar), enabling potential credential disclosure in affected deployments; - enumerate internal services / ports from the server's vantage point via differing timeout/error behaviour.

Because the response body is reflected, this is a high-value full-response SSRF. The only precondition is network reach to the server.

Severity: High, CVSS:3.1/AV:N/AC:L/PR:N/UI:N/S:C/C:H/I:N/A:N = 8.6.

Honest caveats (do not refute the finding): the ureq client uses a 2-second timeout, which limits slow targets; AWS IMDSv2 requires a PUT token, so plain GET metadata retrieval may fail there. IMDSv1 and any internal HTTP service or admin panel that answers a plain GET remain fully reachable and reflected.

Remediation

Add a central SSRF guard inside fetch_body (lib/src/client/helpers.rs): resolve the target host and reject loopback, link-local, RFC1918 private, CGNAT and unique-local ranges, and reject non-http(s) schemes. Re-validate after DNS resolution (to prevent DNS-rebinding) and after each redirect hop. Enforce authorization on endpoint dispatch: run check_read (or a per-endpoint policy) for outbound-fetching endpoints instead of short-circuiting before it at lib/src/db.rs:1010-1012, and at minimum require an authenticated, authorized agent for /bookmark. Optionally add an operator-configurable allow-list of fetchable hosts.

Dedup

  • atomicdata-dev/atomic-server GitHub Security Advisories: no published advisories (re-verified 2026-07-03). No CVE/GHSA for atomic-server SSRF / the /bookmark endpoint.
  • Source re-verified on current develop 2026-07-03: the handler still discards the caller (for_agent: _) and calls fetch_data(url), and fetch_body still validates only url.starts_with("http") with no private-IP deny-list (still vulnerable).

Credit / disclosure

Reported by Ray Sabee, Whitehat Security (https://whitehatsecurity.nl/), GitHub @raysabee. Please credit "Ray Sabee / Whitehat Security / whitehatsecurity.nl" on the advisory and in any release notes. Coordinated disclosure: nothing will be published before a fix is available.

Note: two related findings exist in the same repo and can be handled in this advisory or alongside it: F-ATOMIC-SSRF-002 (a blind pre-auth SSRF in POST /import via the SAME fetch_body root cause, so one central fix covers both) and F-ATOMIC-LFI-001 (authenticated path-traversal file read in GET /download, a separate sink in download.rs).

Show details on source website

{
  "affected": [
    {
      "package": {
        "ecosystem": "crates.io",
        "name": "atomic-server"
      },
      "ranges": [
        {
          "events": [
            {
              "introduced": "0"
            },
            {
              "last_affected": "0.40.1"
            }
          ],
          "type": "ECOSYSTEM"
        }
      ]
    }
  ],
  "aliases": [
    "CVE-2026-69158"
  ],
  "database_specific": {
    "cwe_ids": [
      "CWE-918"
    ],
    "github_reviewed": true,
    "github_reviewed_at": "2026-11-18T20:42:50Z",
    "nvd_published_at": "2026-11-17T21:18:12Z",
    "severity": "HIGH"
  },
  "details": "## Summary\n\nAtomic-Server exposes a `GET /bookmark` endpoint that fetches an attacker-supplied URL server-side, parses the response, and reflects the fetched content back to the caller. The endpoint requires no authentication and performs no authorization of its own, and the underlying fetch helper applies no allow-list or private-IP deny-list. An unauthenticated attacker with network reach to the server can force it to issue arbitrary outbound HTTP GET requests to internal-only hosts (loopback services, internal admin panels, link-local cloud-metadata) and read the response body back. This is a full-response SSRF, not a blind one.\n\n## Root cause (source trace)\n\n1. The bookmark endpoint is registered unconditionally for every server: `server/src/appstate.rs:69` (`store.add_endpoint(plugins::bookmark::bookmark_endpoint())`).\n2. An unauthenticated caller on the standard GET route becomes `ForAgent::Public`: `server/src/helpers.rs:168-183` (`get_client_agent`).\n3. In `get_resource_extended`, when the subject matches an endpoint path the function returns via `call_endpoint` BEFORE the `hierarchy::check_read` authorization call: `lib/src/db.rs:1010-1017` (`if is_endpoint { return self.call_endpoint(...) }`, and `check_read` sits just after at :1017, never reached).\n4. `call_endpoint` performs no authorization itself; it only matches the path and invokes the handler: `lib/src/db.rs:662-706`.\n5. The bookmark handler discards the caller (`for_agent: _`) and fetches the `url` query parameter directly: `server/src/plugins/bookmark.rs:45-76` (`let content = fetch_data(\u0026path)?;`).\n6. `fetch_body` validates only `url.starts_with(\"http\")`: `lib/src/client/helpers.rs:78-91`. A repo-wide grep finds no `is_loopback` / `is_private` / `127.0.0.1` / `169.254` / `localhost` deny-list anywhere in `lib/src` or `server/src`, and no host allow-list.\n7. The fetched body is HTML-parsed and written into the resource `name`, `description`, `image` and `preview` (markdown) fields, which are returned to the caller: `server/src/plugins/bookmark.rs:79-110`. This is what makes it a full-response (not blind) SSRF; even plaintext internal responses are wrapped into the preview markdown.\n\n## Proof of concept (unauthenticated)\n\n```\nGET /bookmark?url=http://127.0.0.1:\u003cinternal-port\u003e/\u0026name=x        # internal loopback service\nGET /bookmark?url=http://169.254.169.254/latest/meta-data/\u0026name=x # IMDSv1 cloud metadata\n```\n\nNo authentication headers. The returned resource reflects the fetched content in its `name` / `description` / `preview` fields, so the attacker reads the internal target\u0027s response body.\n\n## Proof of omission\n\nThe endpoint-dispatch path returns at `lib/src/db.rs:1011` (`return self.call_endpoint(...)`) before the `check_read` call at `:1017`, and there is no authorization inside `call_endpoint` (`:662-706`). The handler discards the caller (`for_agent: _`, `bookmark.rs:49`), so the endpoint can never enforce per-agent access. `fetch_body` validates only `url.starts_with(\"http\")` (`helpers.rs:83`); no loopback/private/link-local deny-list exists anywhere in the source. No corresponding SSRF/security entry exists in CHANGELOG.md.\n\n## Impact\n\nAn unauthenticated attacker with network reach to the server can:\n- read internal-only HTTP services not exposed to the internet (admin panels, dashboards, unauthenticated microservices on loopback or the private network);\n- read cloud metadata reachable via plain GET (IMDSv1 and similar), enabling potential credential disclosure in affected deployments;\n- enumerate internal services / ports from the server\u0027s vantage point via differing timeout/error behaviour.\n\nBecause the response body is reflected, this is a high-value full-response SSRF. The only precondition is network reach to the server.\n\nSeverity: High, CVSS:3.1/AV:N/AC:L/PR:N/UI:N/S:C/C:H/I:N/A:N = 8.6.\n\nHonest caveats (do not refute the finding): the `ureq` client uses a 2-second timeout, which limits slow targets; AWS IMDSv2 requires a PUT token, so plain GET metadata retrieval may fail there. IMDSv1 and any internal HTTP service or admin panel that answers a plain GET remain fully reachable and reflected.\n\n## Remediation\n\nAdd a central SSRF guard inside `fetch_body` (`lib/src/client/helpers.rs`): resolve the target host and reject loopback, link-local, RFC1918 private, CGNAT and unique-local ranges, and reject non-http(s) schemes. Re-validate after DNS resolution (to prevent DNS-rebinding) and after each redirect hop. Enforce authorization on endpoint dispatch: run `check_read` (or a per-endpoint policy) for outbound-fetching endpoints instead of short-circuiting before it at `lib/src/db.rs:1010-1012`, and at minimum require an authenticated, authorized agent for `/bookmark`. Optionally add an operator-configurable allow-list of fetchable hosts.\n\n## Dedup\n\n- atomicdata-dev/atomic-server GitHub Security Advisories: no published advisories (re-verified 2026-07-03). No CVE/GHSA for atomic-server SSRF / the /bookmark endpoint.\n- Source re-verified on current `develop` 2026-07-03: the handler still discards the caller (`for_agent: _`) and calls `fetch_data(url)`, and `fetch_body` still validates only `url.starts_with(\"http\")` with no private-IP deny-list (still vulnerable).\n\n## Credit / disclosure\n\nReported by Ray Sabee, Whitehat Security (https://whitehatsecurity.nl/), GitHub @raysabee. Please credit \"Ray Sabee / Whitehat Security / whitehatsecurity.nl\" on the advisory and in any release notes. Coordinated disclosure: nothing will be published before a fix is available.\n\nNote: two related findings exist in the same repo and can be handled in this advisory or alongside it: F-ATOMIC-SSRF-002 (a blind pre-auth SSRF in `POST /import` via the SAME `fetch_body` root cause, so one central fix covers both) and F-ATOMIC-LFI-001 (authenticated path-traversal file read in `GET /download`, a separate sink in `download.rs`).",
  "id": "GHSA-ff87-73jp-f3v4",
  "modified": "2026-11-18T20:42:50Z",
  "published": "2026-11-18T20:42:50Z",
  "references": [
    {
      "type": "WEB",
      "url": "https://github.com/ontola/atomic-server/security/advisories/GHSA-ff87-73jp-f3v4"
    },
    {
      "type": "ADVISORY",
      "url": "https://nvd.nist.gov/vuln/detail/CVE-2026-69158"
    },
    {
      "type": "WEB",
      "url": "https://github.com/ontola/atomic-server/commit/006ff290f7ec976c8cbafdceb7b6899389abd5be"
    },
    {
      "type": "WEB",
      "url": "https://github.com/ontola/atomic-server/commit/2d4535fcd84761288027616505cc61abdf05bc50"
    },
    {
      "type": "WEB",
      "url": "https://github.com/ontola/atomic-server/commit/e7f551d385045708240ca56889d1334a39349735"
    },
    {
      "type": "PACKAGE",
      "url": "https://github.com/ontola/atomic-server"
    },
    {
      "type": "WEB",
      "url": "https://github.com/ontola/atomic-server/releases/tag/v0.40.3"
    }
  ],
  "schema_version": "1.4.0",
  "severity": [
    {
      "score": "CVSS:3.1/AV:N/AC:L/PR:N/UI:N/S:C/C:H/I:N/A:N",
      "type": "CVSS_V3"
    }
  ],
  "summary": "Atomic Server: Unauthenticated full-response SSRF in Atomic-Server /bookmark endpoint"
}



Log in or create an account to share your comment.




Tags
Taxonomy of the tags.


Loading…

Loading…

Loading…

Drag nodes to arrange them. Double-click a related vulnerability to open it, or a source to visit its link. The Physics button on the left tunes the layout and the simulation; the legend at the bottom left hides or shows nodes by kind and by sighting type.

Forecast uses a logistic model when the trend is rising, or an exponential decay model when the trend is falling. Fitted via linearized least squares.

Sightings

Author Source Type Date Other

Nomenclature

  • Seen: The vulnerability was mentioned, discussed, or observed by the user.
  • Confirmed: The vulnerability has been validated from an analyst's perspective.
  • Published Proof of Concept: A public proof of concept is available for this vulnerability.
  • Exploited: The vulnerability was observed as exploited by the user who reported the sighting.
  • Patched: The vulnerability was observed as successfully patched by the user who reported the sighting.
  • Not exploited: The vulnerability was not observed as exploited by the user who reported the sighting.
  • Not confirmed: The user expressed doubt about the validity of the vulnerability.
  • Not patched: The vulnerability was not observed as successfully patched by the user who reported the sighting.
Vulnerability timeline

Sightings and recorded dates, grouped by day · UTC

JSON
How durations are calculated

Following the VTEM framework (Black Hat USA 2015), time to patch starts when the vendor or a coordinator was informed, total time to patch at discovery, and disclosure to patch at public disclosure; vendor response time runs from the report to its acknowledgement. Time to exploit and time unpatched start at the earliest discovery or report, falling back to public disclosure. Unpatched time ends at the first recorded patch, or the calculation time when no patch is recorded. Exploitation to patch shows the interval between those two events; a published proof of concept is a public exploit, not exploitation.

Each result shows its source dates. A negative interval means the end event came first; when date-only inputs leave the order open, the result says so. Date-only inputs are approximate. The earliest evidence wins: KEV report dates and sightings are labelled as such. Records describing another vulnerability are listed but never calculated from, and a PoC alone does not establish exploitation. Missing patch evidence does not prove that no patch exists. Filters below change the event list only.

Events included in the timeline

Loading timeline…


    Loading…

    Loading…

    Loading…

    Related by attack behaviour

    Vulnerabilities whose description is nearest to this one in the vector space of the CIRCL/vulnerability-attack-technique-biencoder model. This is a similarity search over the bi-encoder space (plain cosine), not a classification, and it has no measured accuracy.