CWE-639
AllowedAuthorization Bypass Through User-Controlled Key
Abstraction: Base · Status: Incomplete
The system's authorization functionality does not prevent one user from gaining access to another user's data or record by modifying the key value identifying the data.
5933 vulnerabilities reference this CWE, most recent first.
GHSA-PFWQ-MR9G-GQ6M
Vulnerability from github – Published: 2025-10-13 21:31 – Updated: 2025-10-13 23:12Insecure Direct Object Reference (IDOR) vulnerability in Liferay Portal 7.4.0 through 7.4.3.111, and older unsupported versions, and Liferay DXP 2023.Q4.0 through 2023.Q4.5, 2023.Q3.1 through 2023.Q3.10, 7.4 GA through update 92, and older unsupported versions allows remote authenticated users in one virtual instance to assign an organization to a user in a different virtual instance via the _com_liferay_users_admin_web_portlet_UsersAdminPortlet_addUserIds parameter.
{
"affected": [
{
"package": {
"ecosystem": "Maven",
"name": "com.liferay.portal:com.liferay.portal.impl"
},
"ranges": [
{
"events": [
{
"introduced": "0"
},
{
"fixed": "99.0.0"
}
],
"type": "ECOSYSTEM"
}
]
}
],
"aliases": [
"CVE-2025-62252"
],
"database_specific": {
"cwe_ids": [
"CWE-639"
],
"github_reviewed": true,
"github_reviewed_at": "2025-10-13T23:12:07Z",
"nvd_published_at": "2025-10-13T21:15:35Z",
"severity": "MODERATE"
},
"details": "Insecure Direct Object Reference (IDOR) vulnerability in Liferay Portal 7.4.0 through 7.4.3.111, and older unsupported versions, and Liferay DXP 2023.Q4.0 through 2023.Q4.5, 2023.Q3.1 through 2023.Q3.10, 7.4 GA through update 92, and older unsupported versions allows remote authenticated users in one virtual instance to assign an organization to a user in a different virtual instance via the _com_liferay_users_admin_web_portlet_UsersAdminPortlet_addUserIds parameter.",
"id": "GHSA-pfwq-mr9g-gq6m",
"modified": "2025-10-13T23:12:07Z",
"published": "2025-10-13T21:31:10Z",
"references": [
{
"type": "ADVISORY",
"url": "https://nvd.nist.gov/vuln/detail/CVE-2025-62252"
},
{
"type": "WEB",
"url": "https://github.com/liferay/liferay-portal/commit/8c3fc088f82ffc981a21935e8b6dcf8f36e27152"
},
{
"type": "WEB",
"url": "https://github.com/liferay/liferay-portal/commit/e7b6074a320a8872ffe9423c3d1a64dada4f3238"
},
{
"type": "PACKAGE",
"url": "https://github.com/liferay/liferay-portal"
},
{
"type": "WEB",
"url": "https://liferay.atlassian.net/browse/LPE-17941"
},
{
"type": "WEB",
"url": "https://liferay.dev/portal/security/known-vulnerabilities/-/asset_publisher/jekt/content/CVE-2025-62252"
}
],
"schema_version": "1.4.0",
"severity": [
{
"score": "CVSS:4.0/AV:N/AC:L/AT:N/PR:L/UI:N/VC:N/VI:L/VA:N/SC:N/SI:N/SA:N",
"type": "CVSS_V4"
}
],
"summary": "Liferay is Vulnerable to Authorization Bypass Through User-Controlled Key"
}
GHSA-PG6G-JF8C-49P7
Vulnerability from github – Published: 2026-07-10 15:31 – Updated: 2026-07-10 15:31Vikunja before 2.2.1 contains an authorization flaw where the LinkSharing.ReadAll endpoint exposes share hashes to users with read access, enabling permission escalation to admin-level shares. The GetTaskAttachment endpoint performs permission checks against user-supplied task IDs but fetches attachments by sequential ID without verifying ownership, allowing attackers to download and delete all file attachments across all projects instance-wide.
{
"affected": [],
"aliases": [
"CVE-2026-56765"
],
"database_specific": {
"cwe_ids": [
"CWE-639"
],
"github_reviewed": false,
"github_reviewed_at": null,
"nvd_published_at": "2026-07-10T15:16:43Z",
"severity": "CRITICAL"
},
"details": "Vikunja before 2.2.1 contains an authorization flaw where the LinkSharing.ReadAll endpoint exposes share hashes to users with read access, enabling permission escalation to admin-level shares. The GetTaskAttachment endpoint performs permission checks against user-supplied task IDs but fetches attachments by sequential ID without verifying ownership, allowing attackers to download and delete all file attachments across all projects instance-wide.",
"id": "GHSA-pg6g-jf8c-49p7",
"modified": "2026-07-10T15:31:40Z",
"published": "2026-07-10T15:31:40Z",
"references": [
{
"type": "WEB",
"url": "https://github.com/go-vikunja/vikunja/security/advisories/GHSA-2pv8-4c52-mf8j"
},
{
"type": "ADVISORY",
"url": "https://nvd.nist.gov/vuln/detail/CVE-2026-56765"
},
{
"type": "WEB",
"url": "https://www.vulncheck.com/advisories/vikunja-unauthenticated-instance-wide-data-breach-via-link-share-hash-disclosure-chained-with-cross-project-attachment-idor"
}
],
"schema_version": "1.4.0",
"severity": [
{
"score": "CVSS:3.1/AV:N/AC:L/PR:N/UI:N/S:U/C:H/I:H/A:H",
"type": "CVSS_V3"
},
{
"score": "CVSS:4.0/AV:N/AC:L/AT:N/PR:N/UI:N/VC:H/VI:H/VA:H/SC:N/SI:N/SA:N/E:X/CR:X/IR:X/AR:X/MAV:X/MAC:X/MAT:X/MPR:X/MUI:X/MVC:X/MVI:X/MVA:X/MSC:X/MSI:X/MSA:X/S:X/AU:X/R:X/V:X/RE:X/U:X",
"type": "CVSS_V4"
}
]
}
GHSA-PGFM-8CHM-HCV4
Vulnerability from github – Published: 2025-02-04 09:31 – Updated: 2025-02-04 09:31The JS Help Desk – The Ultimate Help Desk & Support Plugin plugin for WordPress is vulnerable to Insecure Direct Object Reference in all versions up to, and including, 2.8.8 via the 'exportusereraserequest' due to missing validation on a user controlled key. This makes it possible for authenticated attackers, with Subscriber-level permissions and above, to export ticket data for any user.
{
"affected": [],
"aliases": [
"CVE-2024-13607"
],
"database_specific": {
"cwe_ids": [
"CWE-639"
],
"github_reviewed": false,
"github_reviewed_at": null,
"nvd_published_at": "2025-02-04T07:15:12Z",
"severity": "MODERATE"
},
"details": "The JS Help Desk \u2013 The Ultimate Help Desk \u0026 Support Plugin plugin for WordPress is vulnerable to Insecure Direct Object Reference in all versions up to, and including, 2.8.8 via the \u0027exportusereraserequest\u0027 due to missing validation on a user controlled key. This makes it possible for authenticated attackers, with Subscriber-level permissions and above, to export ticket data for any user.",
"id": "GHSA-pgfm-8chm-hcv4",
"modified": "2025-02-04T09:31:07Z",
"published": "2025-02-04T09:31:07Z",
"references": [
{
"type": "ADVISORY",
"url": "https://nvd.nist.gov/vuln/detail/CVE-2024-13607"
},
{
"type": "WEB",
"url": "https://plugins.trac.wordpress.org/browser/js-support-ticket/tags/2.8.8/modules/gdpr/controller.php#L110"
},
{
"type": "WEB",
"url": "https://plugins.trac.wordpress.org/changeset/3230977"
},
{
"type": "WEB",
"url": "https://www.wordfence.com/threat-intel/vulnerabilities/id/3f57fbbc-ed5a-4452-bd8a-6fc0a4536d76?source=cve"
}
],
"schema_version": "1.4.0",
"severity": [
{
"score": "CVSS:3.1/AV:N/AC:L/PR:L/UI:N/S:U/C:L/I:N/A:N",
"type": "CVSS_V3"
}
]
}
GHSA-PGPF-M8M4-6CG6
Vulnerability from github – Published: 2026-03-12 14:07 – Updated: 2026-03-12 14:07Impact
Affected versions of Winter CMS allowed authenticated backend users to escalate their accounts level of access to the system by modifying the roles / permissions assigned to their account through specially crafted requests to the backend while logged in.
To actively exploit this security issue, an attacker would need access to the Backend with a user account with any level of access.
The Winter CMS maintainers strongly recommend that all Winter CMS sites that have any reliance on the roles & permissions system to update immediately. Security fixes have been backported to all major versions of Winter (1.0, 1.1, and 1.2).
Patches
Multiple fixes and defence in depth has been applied to prevent current and future privilege escalation attacks at the lowest level possible.
This security issue has been fixed as of https://wintercms.com/releases/v1.0.477, https://wintercms.com/releases/v1.1.12, https://wintercms.com/releases/v1.2.12.
Workarounds
If you cannot upgrade, you may apply the changes from the releases to your Winter CMS installation manually to resolve this issue.
{
"affected": [
{
"package": {
"ecosystem": "Packagist",
"name": "winter/wn-backend-module"
},
"ranges": [
{
"events": [
{
"introduced": "1.2.0"
},
{
"fixed": "1.2.12"
}
],
"type": "ECOSYSTEM"
}
]
},
{
"package": {
"ecosystem": "Packagist",
"name": "winter/wn-backend-module"
},
"ranges": [
{
"events": [
{
"introduced": "1.1.0"
},
{
"fixed": "1.1.12"
}
],
"type": "ECOSYSTEM"
}
]
},
{
"package": {
"ecosystem": "Packagist",
"name": "winter/wn-backend-module"
},
"ranges": [
{
"events": [
{
"introduced": "0"
},
{
"fixed": "1.0.477"
}
],
"type": "ECOSYSTEM"
}
]
}
],
"aliases": [
"CVE-2026-27591"
],
"database_specific": {
"cwe_ids": [
"CWE-284",
"CWE-639",
"CWE-915"
],
"github_reviewed": true,
"github_reviewed_at": "2026-03-12T14:07:39Z",
"nvd_published_at": "2026-03-11T22:16:32Z",
"severity": "CRITICAL"
},
"details": "## Impact\nAffected versions of Winter CMS allowed authenticated backend users to escalate their accounts level of access to the system by modifying the roles / permissions assigned to their account through specially crafted requests to the backend while logged in.\n\nTo actively exploit this security issue, an attacker would need access to the Backend with a user account with any level of access.\n\nThe Winter CMS maintainers strongly recommend that all Winter CMS sites that have any reliance on the roles \u0026 permissions system to update immediately. Security fixes have been backported to all major versions of Winter (1.0, 1.1, and 1.2).\n\n## Patches\nMultiple fixes and defence in depth has been applied to prevent current and future privilege escalation attacks at the lowest level possible.\n\nThis security issue has been fixed as of https://wintercms.com/releases/v1.0.477, https://wintercms.com/releases/v1.1.12, https://wintercms.com/releases/v1.2.12.\n\n## Workarounds\nIf you cannot upgrade, you may apply the changes from the releases to your Winter CMS installation manually to resolve this issue.",
"id": "GHSA-pgpf-m8m4-6cg6",
"modified": "2026-03-12T14:07:39Z",
"published": "2026-03-12T14:07:39Z",
"references": [
{
"type": "WEB",
"url": "https://github.com/wintercms/winter/security/advisories/GHSA-pgpf-m8m4-6cg6"
},
{
"type": "ADVISORY",
"url": "https://nvd.nist.gov/vuln/detail/CVE-2026-27591"
},
{
"type": "PACKAGE",
"url": "https://github.com/wintercms/winter"
},
{
"type": "WEB",
"url": "https://wintercms.com/releases/v1.0.477"
},
{
"type": "WEB",
"url": "https://wintercms.com/releases/v1.1.12"
},
{
"type": "WEB",
"url": "https://wintercms.com/releases/v1.2.12"
}
],
"schema_version": "1.4.0",
"severity": [
{
"score": "CVSS:3.1/AV:N/AC:L/PR:L/UI:N/S:C/C:H/I:H/A:H",
"type": "CVSS_V3"
}
],
"summary": "Winter vulnerable to privilege escalation by authenticated backend users"
}
GHSA-PGQF-926R-548M
Vulnerability from github – Published: 2026-07-21 20:39 – Updated: 2026-07-21 20:39Summary
The API endpoint DELETE /repos/{owner}/{repo}/issues/{index}/labels/{id} loads the label by ID with a global,
unscoped lookup and never verifies the label belongs to the URL's repository (or its owning organization). Because
the response status differs by whether the label ID exists anywhere on the instance (204) versus not (422), an
authenticated user can use the endpoint as a cross-repository label-ID existence / enumeration oracle, including
for labels in repositories and organizations they cannot access.
Severity
- The leaked information is minimal (existence/count of label IDs instance-wide); no label name, color, or owning repository is disclosed, and no cross-repository write occurs.
Affected / patched versions
- Affected: through 1.26.3 (latest at time of report).
- Patched: none yet.
Details
DeleteIssueLabel resolves the label with a global loader and never checks its scope:
// routers/api/v1/repo/issue_label.go (DeleteIssueLabel)
label, err := issues_model.GetLabelByID(ctx, ctx.PathParamInt64("id")) // global, unscoped
GetLabelByID (models/issues/label.go) is e.ID(labelID).Get(l) with no repo_id / org_id filter. The
handler never verifies label.RepoID == ctx.Repo.Repository.ID (nor the org-label equivalent), and the downstream
issue_service.RemoveLabel (services/issue/label.go) only re-checks the doer's write permission on the issue's
own repository — never that the label belongs to it.
Every sibling label handler is correctly scoped — GetLabel / EditLabel / DeleteLabel (repo and org) use
GetLabelInRepoByID / GetLabelInOrgByID and return 404 for a foreign ID. DeleteIssueLabel is the only outlier.
Why it is only an oracle: deleteIssueLabel (models/issues/issue_label.go) deletes the issue_label row keyed
by (issue.ID, label.ID). For a foreign label, no such row exists → the function returns early before any
mutation or comment creation. So there is no cross-repo write and no leak of the label's name. But the HTTP status
differs:
- label ID exists anywhere on the instance (incl. private repos/orgs) → 204 No Content
- label ID does not exist → 422 (
ErrLabelNotExist)
Label IDs are sequential auto-increment, so this enumerates the instance-wide label population and probes existence of specific IDs across tenant boundaries.
Proof of Concept
Verified end-to-end on a build of the v1.26.3 tag.
alice(private repoalice/secret) creates a label → internal id 1.- Attacker
bob(separate user; public repobob/pubwith issue #1; no access toalice/secret) holds a token withwrite:issueon his own repo.
bob DELETE /api/v1/repos/bob/pub/issues/1/labels/1 -> HTTP 204 (alice's PRIVATE label id exists)
bob DELETE /api/v1/repos/bob/pub/issues/1/labels/99999999 -> HTTP 422 (no such label)
Differing only by the label ID: 204 vs 422 distinguishes "label ID exists" from "does not exist." bob has zero rights to alice/secret but can still learn label id 1 exists. (Alice's label is untouched — no write.)
Reproduction steps:
1. Create two users alice, bob. As alice, create a private repo and a label on it (note the label id from
the API response).
2. As bob, create any repo with an issue, and a token with write:issue.
3. curl -u bob:$T -X DELETE https://<gitea>/api/v1/repos/bob/pub/issues/1/labels/<alice_label_id> → 204.
4. curl -u bob:$T -X DELETE https://<gitea>/api/v1/repos/bob/pub/issues/1/labels/99999999 → 422.
5. The differing status across an ID bob cannot otherwise see is the oracle.
Impact
Cross-tenant authorization-key bypass producing a label-ID existence/enumeration oracle: an authenticated user can determine whether arbitrary label IDs (including in private repositories and organizations they cannot access) exist, and enumerate the instance-wide label population.
Remediation
Scope the loader like every sibling handler, returning 404 for a label that is not in the URL repository (or its owning organization), so the status no longer distinguishes existence:
// routers/api/v1/repo/issue_label.go — in DeleteIssueLabel, after loading the label
if label.RepoID != ctx.Repo.Repository.ID && label.OrgID != ctx.Repo.Repository.OwnerID {
ctx.APIErrorNotFound()
return
}
(Equivalently, resolve via GetLabelInRepoByID and, for org repositories, also accept the repo owner's org labels —
mirroring the scoping in GetLabel/EditLabel/DeleteLabel.)
{
"affected": [
{
"package": {
"ecosystem": "Go",
"name": "code.gitea.io/gitea"
},
"ranges": [
{
"events": [
{
"introduced": "0"
},
{
"fixed": "1.27.0"
}
],
"type": "ECOSYSTEM"
}
]
}
],
"aliases": [
"CVE-2026-58445"
],
"database_specific": {
"cwe_ids": [
"CWE-203",
"CWE-639"
],
"github_reviewed": true,
"github_reviewed_at": "2026-07-21T20:39:22Z",
"nvd_published_at": null,
"severity": "LOW"
},
"details": "## Summary\n\nThe API endpoint `DELETE /repos/{owner}/{repo}/issues/{index}/labels/{id}` loads the label by ID with a **global,\nunscoped** lookup and never verifies the label belongs to the URL\u0027s repository (or its owning organization). Because\nthe response status differs by whether the label ID exists **anywhere on the instance** (204) versus not (422), an\nauthenticated user can use the endpoint as a **cross-repository label-ID existence / enumeration oracle**, including\nfor labels in repositories and organizations they cannot access.\n\n## Severity\n\n- The leaked information is minimal (existence/count of label IDs instance-wide); **no label name, color, or owning\n repository is disclosed, and no cross-repository write occurs.**\n\n## Affected / patched versions\n\n- **Affected:** through **1.26.3** (latest at time of report).\n- **Patched:** none yet.\n\n## Details\n\n`DeleteIssueLabel` resolves the label with a global loader and never checks its scope:\n\n```go\n// routers/api/v1/repo/issue_label.go (DeleteIssueLabel)\nlabel, err := issues_model.GetLabelByID(ctx, ctx.PathParamInt64(\"id\")) // global, unscoped\n```\n\n`GetLabelByID` (`models/issues/label.go`) is `e.ID(labelID).Get(l)` with **no** `repo_id` / `org_id` filter. The\nhandler never verifies `label.RepoID == ctx.Repo.Repository.ID` (nor the org-label equivalent), and the downstream\n`issue_service.RemoveLabel` (`services/issue/label.go`) only re-checks the doer\u0027s write permission on the **issue\u0027s\nown** repository \u2014 never that the label belongs to it.\n\nEvery sibling label handler is correctly scoped \u2014 `GetLabel` / `EditLabel` / `DeleteLabel` (repo and org) use\n`GetLabelInRepoByID` / `GetLabelInOrgByID` and return 404 for a foreign ID. `DeleteIssueLabel` is the only outlier.\n\n**Why it is only an oracle:** `deleteIssueLabel` (`models/issues/issue_label.go`) deletes the `issue_label` row keyed\nby `(issue.ID, label.ID)`. For a foreign label, no such row exists \u2192 the function returns early **before** any\nmutation or comment creation. So there is no cross-repo write and no leak of the label\u0027s name. But the HTTP status\ndiffers:\n\n- label ID exists anywhere on the instance (incl. private repos/orgs) \u2192 **204 No Content**\n- label ID does not exist \u2192 **422** (`ErrLabelNotExist`)\n\nLabel IDs are sequential auto-increment, so this enumerates the instance-wide label population and probes existence\nof specific IDs across tenant boundaries.\n\n## Proof of Concept\n\nVerified end-to-end on a build of the `v1.26.3` tag.\n\n- `alice` (private repo `alice/secret`) creates a label \u2192 internal id **1**.\n- Attacker `bob` (separate user; public repo `bob/pub` with issue #1; **no access** to `alice/secret`) holds a token\n with `write:issue` on his own repo.\n\n```text\nbob DELETE /api/v1/repos/bob/pub/issues/1/labels/1 -\u003e HTTP 204 (alice\u0027s PRIVATE label id exists)\nbob DELETE /api/v1/repos/bob/pub/issues/1/labels/99999999 -\u003e HTTP 422 (no such label)\n```\n\nDiffering only by the label ID: `204` vs `422` distinguishes \"label ID exists\" from \"does not exist.\" `bob` has zero rights to `alice/secret` but can still learn label id 1 exists. (Alice\u0027s label is untouched \u2014 no write.)\n\nReproduction steps:\n1. Create two users `alice`, `bob`. As `alice`, create a private repo and a label on it (note the label `id` from\n the API response).\n2. As `bob`, create any repo with an issue, and a token with `write:issue`.\n3. `curl -u bob:$T -X DELETE https://\u003cgitea\u003e/api/v1/repos/bob/pub/issues/1/labels/\u003calice_label_id\u003e` \u2192 **204**.\n4. `curl -u bob:$T -X DELETE https://\u003cgitea\u003e/api/v1/repos/bob/pub/issues/1/labels/99999999` \u2192 **422**.\n5. The differing status across an ID `bob` cannot otherwise see is the oracle.\n\n## Impact\n\nCross-tenant authorization-key bypass producing a label-ID existence/enumeration oracle: an authenticated user can\ndetermine whether arbitrary label IDs (including in private repositories and organizations they cannot access) exist,\nand enumerate the instance-wide label population.\n\n## Remediation\n\nScope the loader like every sibling handler, returning 404 for a label that is not in the URL repository (or its\nowning organization), so the status no longer distinguishes existence:\n\n```go\n// routers/api/v1/repo/issue_label.go \u2014 in DeleteIssueLabel, after loading the label\nif label.RepoID != ctx.Repo.Repository.ID \u0026\u0026 label.OrgID != ctx.Repo.Repository.OwnerID {\n ctx.APIErrorNotFound()\n return\n}\n```\n\n(Equivalently, resolve via `GetLabelInRepoByID` and, for org repositories, also accept the repo owner\u0027s org labels \u2014\nmirroring the scoping in `GetLabel`/`EditLabel`/`DeleteLabel`.)",
"id": "GHSA-pgqf-926r-548m",
"modified": "2026-07-21T20:39:22Z",
"published": "2026-07-21T20:39:22Z",
"references": [
{
"type": "WEB",
"url": "https://github.com/go-gitea/gitea/security/advisories/GHSA-pgqf-926r-548m"
},
{
"type": "PACKAGE",
"url": "https://github.com/go-gitea/gitea"
},
{
"type": "WEB",
"url": "https://github.com/go-gitea/gitea/releases/tag/v1.27.0"
}
],
"schema_version": "1.4.0",
"severity": [
{
"score": "CVSS:3.1/AV:N/AC:L/PR:H/UI:N/S:U/C:L/I:N/A:N",
"type": "CVSS_V3"
}
],
"summary": "Gitea: Cross-repository label-ID enumeration oracle via unscoped DeleteIssueLabel API"
}
GHSA-PGWJ-FPP8-CRJ4
Vulnerability from github – Published: 2026-10-29 12:30 – Updated: 2026-10-29 12:30A broken access control vulnerability exists in the Foxit official website’s HTML-to-PDF conversion service due to the use of a shared token for submitted conversion documents. This may allow unauthorized users to access documents submitted by other users.
{
"affected": [],
"aliases": [
"CVE-2026-111417"
],
"database_specific": {
"cwe_ids": [
"CWE-639"
],
"github_reviewed": false,
"github_reviewed_at": null,
"nvd_published_at": "2026-10-29T10:17:03Z",
"severity": "HIGH"
},
"details": "A broken access control vulnerability exists in the Foxit official website\u2019s HTML-to-PDF conversion service due to the use of a shared token for submitted conversion documents. This may allow unauthorized users to access documents submitted by other users.",
"id": "GHSA-pgwj-fpp8-crj4",
"modified": "2026-10-29T12:30:35Z",
"published": "2026-10-29T12:30:35Z",
"references": [
{
"type": "ADVISORY",
"url": "https://nvd.nist.gov/vuln/detail/CVE-2026-111417"
},
{
"type": "WEB",
"url": "https://www.foxit.com/support/security-bulletins.html"
}
],
"schema_version": "1.4.0",
"severity": [
{
"score": "CVSS:3.1/AV:N/AC:L/PR:N/UI:N/S:U/C:H/I:N/A:N",
"type": "CVSS_V3"
}
]
}
GHSA-PGXQ-P76C-X9CG
Vulnerability from github – Published: 2026-05-29 22:19 – Updated: 2026-05-29 22:19Impact
Unauthenticated users could modify existing submissions by posting a known or guessed submission ID to formie/submissions/save-submission.
Patches
Workarounds
Block unauthenticated access to actions/formie/submissions/save-submission, or disable/customize front-end submission editing until patched.
Credit
formie extends many thanks to: - Florian (Cyber Security Engineer, arcade solutions ag) - Contact: security@arcade.ch
{
"affected": [
{
"package": {
"ecosystem": "Packagist",
"name": "verbb/formie"
},
"ranges": [
{
"events": [
{
"introduced": "3.0.0"
},
{
"fixed": "3.1.26"
}
],
"type": "ECOSYSTEM"
}
]
},
{
"package": {
"ecosystem": "Packagist",
"name": "verbb/formie"
},
"ranges": [
{
"events": [
{
"introduced": "0"
},
{
"fixed": "2.2.21"
}
],
"type": "ECOSYSTEM"
}
]
}
],
"aliases": [
"CVE-2026-47266"
],
"database_specific": {
"cwe_ids": [
"CWE-639"
],
"github_reviewed": true,
"github_reviewed_at": "2026-05-29T22:19:19Z",
"nvd_published_at": "2026-05-29T20:16:28Z",
"severity": "HIGH"
},
"details": "### Impact\nUnauthenticated users could modify existing submissions by posting a known or guessed submission ID to `formie/submissions/save-submission`.\n\n### Patches\n[2.2.21](https://github.com/verbb/formie/releases/tag/2.2.21), [3.1.26](https://github.com/verbb/formie/releases/tag/3.1.26)\n\n### Workarounds\nBlock unauthenticated access to `actions/formie/submissions/save-submission`, or disable/customize front-end submission editing until patched.\n\n### Credit\nformie extends many thanks to:\n- Florian (Cyber Security Engineer, arcade solutions ag)\n- Contact: [security@arcade.ch](mailto:security@arcade.ch)",
"id": "GHSA-pgxq-p76c-x9cg",
"modified": "2026-05-29T22:19:19Z",
"published": "2026-05-29T22:19:19Z",
"references": [
{
"type": "WEB",
"url": "https://github.com/verbb/formie/security/advisories/GHSA-pgxq-p76c-x9cg"
},
{
"type": "ADVISORY",
"url": "https://nvd.nist.gov/vuln/detail/CVE-2026-47266"
},
{
"type": "PACKAGE",
"url": "https://github.com/verbb/formie"
},
{
"type": "WEB",
"url": "https://github.com/verbb/formie/releases/tag/2.2.21"
},
{
"type": "WEB",
"url": "https://github.com/verbb/formie/releases/tag/3.1.26"
}
],
"schema_version": "1.4.0",
"severity": [
{
"score": "CVSS:4.0/AV:N/AC:L/AT:N/PR:N/UI:N/VC:N/VI:H/VA:N/SC:N/SI:N/SA:N/E:X/CR:X/IR:X/AR:X/MAV:X/MAC:X/MAT:X/MPR:X/MUI:X/MVC:X/MVI:X/MVA:X/MSC:X/MSI:X/MSA:X/S:X/AU:X/R:X/V:X/RE:X/U:X",
"type": "CVSS_V4"
}
],
"summary": "formie\u0027s unauthenticated front-end submission editing can overwrite existing submissions"
}
GHSA-PH6F-P85X-XVJ8
Vulnerability from github – Published: 2026-09-18 09:31 – Updated: 2026-09-18 09:31The LatePoint – Calendar Booking Plugin for Appointments and Events plugin for WordPress is vulnerable to Insecure Direct Object Reference in all versions up to, and including, 5.6.3 via the LatePointAbilityDeleteBooking::execute due to missing validation on a user controlled key. This makes it possible for attackers, with LatePoint Agent-level access and above, to read bookings and customer PII (full name, email, phone, and notes) assigned to other LatePoint agents, and delete arbitrary bookings by supplying any booking ID. This vulnerability is only exploitable when an administrator has enabled the Abilities API toggles (latepoint_abilities_api, latepoint_abilities_api_delete, and/or latepoint_abilities_api_edit) in the plugin settings.
{
"affected": [],
"aliases": [
"CVE-2026-13471"
],
"database_specific": {
"cwe_ids": [
"CWE-639"
],
"github_reviewed": false,
"github_reviewed_at": null,
"nvd_published_at": "2026-09-18T08:16:59Z",
"severity": "MODERATE"
},
"details": "The LatePoint \u2013 Calendar Booking Plugin for Appointments and Events plugin for WordPress is vulnerable to Insecure Direct Object Reference in all versions up to, and including, 5.6.3 via the LatePointAbilityDeleteBooking::execute due to missing validation on a user controlled key. This makes it possible for attackers, with LatePoint Agent-level access and above, to read bookings and customer PII (full name, email, phone, and notes) assigned to other LatePoint agents, and delete arbitrary bookings by supplying any booking ID. This vulnerability is only exploitable when an administrator has enabled the Abilities API toggles (latepoint_abilities_api, latepoint_abilities_api_delete, and/or latepoint_abilities_api_edit) in the plugin settings.",
"id": "GHSA-ph6f-p85x-xvj8",
"modified": "2026-09-18T09:31:07Z",
"published": "2026-09-18T09:31:07Z",
"references": [
{
"type": "ADVISORY",
"url": "https://nvd.nist.gov/vuln/detail/CVE-2026-13471"
},
{
"type": "WEB",
"url": "https://plugins.trac.wordpress.org/browser/latepoint/tags/5.4.2/lib/abilities/abstract-ability.php#L55"
},
{
"type": "WEB",
"url": "https://plugins.trac.wordpress.org/browser/latepoint/tags/5.4.2/lib/abilities/bookings/abstract-booking-ability.php#L34"
},
{
"type": "WEB",
"url": "https://plugins.trac.wordpress.org/browser/latepoint/tags/5.4.2/lib/abilities/bookings/delete-booking.php#L47"
},
{
"type": "WEB",
"url": "https://plugins.trac.wordpress.org/browser/latepoint/tags/5.4.2/lib/abilities/bookings/list-bookings.php#L50"
},
{
"type": "WEB",
"url": "https://plugins.trac.wordpress.org/browser/latepoint/tags/5.4.2/lib/abilities/class-latepoint-abilities.php#L77"
},
{
"type": "WEB",
"url": "https://plugins.trac.wordpress.org/browser/latepoint/tags/5.4.2/lib/abilities/customers/list-customers.php#L78"
},
{
"type": "WEB",
"url": "https://plugins.trac.wordpress.org/browser/latepoint/tags/5.6.3/lib/abilities/abstract-ability.php#L55"
},
{
"type": "WEB",
"url": "https://plugins.trac.wordpress.org/browser/latepoint/tags/5.6.3/lib/abilities/bookings/abstract-booking-ability.php#L34"
},
{
"type": "WEB",
"url": "https://plugins.trac.wordpress.org/browser/latepoint/tags/5.6.3/lib/abilities/bookings/delete-booking.php#L47"
},
{
"type": "WEB",
"url": "https://plugins.trac.wordpress.org/browser/latepoint/tags/5.6.3/lib/abilities/bookings/list-bookings.php#L50"
},
{
"type": "WEB",
"url": "https://plugins.trac.wordpress.org/browser/latepoint/tags/5.6.3/lib/abilities/class-latepoint-abilities.php#L77"
},
{
"type": "WEB",
"url": "https://plugins.trac.wordpress.org/browser/latepoint/tags/5.6.3/lib/abilities/customers/list-customers.php#L78"
},
{
"type": "WEB",
"url": "https://plugins.trac.wordpress.org/changeset?reponame=\u0026new=3590914%40latepoint\u0026old=3584059%40latepoint"
},
{
"type": "WEB",
"url": "https://www.wordfence.com/threat-intel/vulnerabilities/id/448df3b0-32a7-4097-a37d-07e253993496?source=cve"
}
],
"schema_version": "1.4.0",
"severity": [
{
"score": "CVSS:3.1/AV:N/AC:L/PR:L/UI:N/S:U/C:N/I:L/A:N",
"type": "CVSS_V3"
}
]
}
GHSA-PH72-CQR5-QPP7
Vulnerability from github – Published: 2026-10-05 23:42 – Updated: 2026-10-05 23:42Affected
- Ecosystem / package: pip /
vllm - Affected versions: vLLM ≤ 0.25.1 (confirmed on 0.25.1, commit
752a3a504485). The lower bound predates 0.25.1; maintainers can confirm how far back the scale-out transport path reaches.
Summary
vLLM's disaggregated scale-out transport splits a multimodal request into a trusted render step (POST /v1/chat/completions/render) and a separate generate step (POST /inference/v1/generate). The generate route decodes a caller-supplied features object — serialized encoder tensors (kwargs_data), multimodal hashes (mm_hashes), placeholder ranges (mm_placeholders), and the internal field-processor selection — and forwards it into the engine as if it had come from the trusted renderer, with no rebinding to (or validation against) the active model's renderer contract. Because the two routes are ordinary auth-guarded HTTP endpoints (the /inference prefix is registered by default on generate-capable servers), any authenticated caller can submit an otherwise-valid render body with a single forged field.
Depending on which field is forged, this produces:
- an engine-fatal crash of the shared EngineCore process (denial of service), reproduced as a CUDA illegal-memory-access, a post-admission rank-mismatch
ValueError, and a hardassert— three independent forged fields (sites 1, 2, 3); - silent cross-request encoder-cache poisoning / disclosure of shared encoder state when the cache-key hash is not bound to the payload (site 4);
- transport-level integrity loss when sparse placeholder masks are dropped during render-to-generate replay (site 5).
All five share one root cause and one fix shape: the reconstructed multimodal state on the scale-out path is trusted without being rebound to, and validated against, the active model's renderer output before it reaches the engine.
These sites are distinct from prior multimodal hardening. Site 1 survives GHSA-wv77-2vpf-vmmg (that fix validates full tensor shape in MultiModalDataParser/get_input_embeddings on the prompt-embeds path), because our request forges image_grid_thw metadata with the pixel bytes intact and reaches the Qwen2 vision RoPE/cu_seqlens and image_embeds.split sink, which the shape-check fix does not rebind. Site 4 is distinct from GHSA-c65p-x677-fgj6 (which folds metadata into MultiModalHasher.serialize_item to stop hash collisions), because the scale-out generate path trusts a caller-supplied mm_hash as the cache key with no origin binding, so that fix does not stop a caller from submitting a victim's hash or a kwargs_data=None cache read.
Affected code
Links pinned to the confirmed commit 752a3a504485 (v0.25.1).
Shared entry point and control surface for all five sites:
POST /inference/v1/generateroute:vllm/entrypoints/scale_out/token_in_token_out/api_router.py#L46-L75.- Public
featuresschema (kwargs_data,mm_hashes,mm_placeholders):vllm/entrypoints/scale_out/token_in_token_out/protocol.py#L42-L63. ServingTokens.serve_tokens()copies the decoded geometry and hashes into engine structures without rebinding to the renderer schema:vllm/entrypoints/scale_out/token_in_token_out/serving.py#L145-L172.- The scale-out routers are registered by default on generate-capable servers and
/inferenceis treated as an ordinary auth-guarded prefix:vllm/entrypoints/openai/api_server.py#L217-L219.
Site 1 — forged Qwen grid geometry (engine-fatal DoS). The decoded image_grid_thw is never rebound to the rendered pixel-tensor element count.
- Reconstruction with no model-specific geometry invariant:
serving.py#L147-L172. Qwen2VisionTransformer.prepare_encoder_metadata()derives RoPE tables,cu_seqlens, and the FlashAttentionmax_seqlenfrom the caller-supplied grid:qwen2_vl.py#L658-L712, consumed inforward()(#L713-L751)._process_image_input()computesimage_embeds.split(sizes)from the same untrusted metadata:qwen2_vl.py#L1348-L1369; M-RoPE positions atqwen2_vl.py#L1223.
The only shape check is assert grid_thw.ndim == 2; the split sizes and the vision-encoder call are then derived directly from the caller-supplied grid, with no cross-check against the pixel-tensor row count:
# vllm/model_executor/models/qwen2_vl.py Lines 1348-1369
def _process_image_input(
self, image_input: Qwen2VLImageInputs
) -> tuple[torch.Tensor, ...]:
grid_thw = image_input["image_grid_thw"]
assert grid_thw.ndim == 2
if image_input["type"] == "image_embeds":
image_embeds = image_input["image_embeds"]
else:
pixel_values = image_input["pixel_values"]
if self.use_data_parallel:
return run_dp_sharded_mrope_vision_model(
self.visual, pixel_values, grid_thw.tolist(), rope_type="rope_3d"
)
else:
image_embeds = self.visual(pixel_values, grid_thw=grid_thw)
# Split concatenated embeddings for each image item.
merge_size = self.visual.spatial_merge_size
sizes = (grid_thw.prod(-1) // merge_size // merge_size).tolist()
return image_embeds.split(sizes)
Site 2 — wire-selected field-processor type confusion (engine-fatal DoS). MsgpackDecoder._decode_mm_field_elem() trusts a wire-selected field-factory name and constructs the internal field processor directly from caller data.
vllm/v1/serial_utils.py#L440-L454readsfactory_meth_name, factory_kw = obj["field"]and callsgetattr(MultiModalFieldConfig, factory_meth_name).- Qwen2-VL field/schema contract:
qwen2_vl.py#L763-L791andQwen2VLImagePixelInputs(#L119-L144); parse/validate atqwen2_vl.py#L1300. - Rank mismatch raised post-admission:
vllm/utils/tensor_schema.py#L155-L171, reached fromTensorSchema.__init__→validate()(#L63).
Site 3 — non-positive placeholder length (engine-fatal DoS via reachable assert). PlaceholderRangeInfo{offset,length} is accepted as unconstrained integers and copied verbatim into the engine's PlaceholderRange.
- Schema:
protocol.py#L28-L35. - Copied into
PlaceholderRange:serving.py#L147-L153. - The only guard is an upper bound on embed count (
num_embeds > mm_encoder_cache_size), with no non-positive check:vllm/v1/engine/input_processor.py#L459-L464; raw length returned byvllm/multimodal/inputs.py#L152-L154; window selection assumes non-empty ranges atvllm/multimodal/utils.py#L114-L134. - Sink:
assert start_idx < end_idxatvllm/v1/worker/gpu/mm/encoder_runner.py#L114(duplicated atvllm/v1/worker/gpu_model_runner.py#L3192); turned into a fatal shutdown by EngineCore's uncaught-exception path atvllm/v1/engine/core.py#L1229-L1233.
A length of 0 makes num_encoder_tokens == 0, so end_idx collapses to 0 and the bare assert fires inside the engine worker:
# vllm/v1/worker/gpu/mm/encoder_runner.py Lines 108-114
pos_info = mm_feature.mm_position
start_pos = pos_info.offset
num_encoder_tokens = pos_info.length
start_idx = max(cur_query_start - start_pos, 0)
end_idx = min(cur_query_end - start_pos, num_encoder_tokens)
assert start_idx < end_idx
Site 4 — cache hash not bound to payload (integrity / disclosure). features.mm_hashes (the cache key) and kwargs_data (the tensor) are independent fields with no origin or integrity binding.
- Schema exposing the caller-controlled hash and the
None= resolve-from-cache semantics:protocol.py#L42-L63. - Handler forwards the caller's hashes unchanged into
mm_input(...):serving.py#L164-L170. - Input processing copies the caller hash into the feature identifier with only a string-type check:
vllm/v1/engine/input_processor.py#L165-L181. - Sink — the receiver cache returns the cached tensor solely by that key (
cache_key = feature.mm_hash or feature.identifier):vllm/multimodal/cache.py#L602-L607.
The cache key is the caller-supplied hash with no verification against the tensor bytes, so a forged mm_hash both stores under and reads back another request's slot:
# vllm/multimodal/cache.py Lines 601-607
for feature in mm_features:
cache_key = feature.mm_hash or feature.identifier
self.touch_receiver_cache_item(cache_key, feature.data)
for feature in mm_features:
cache_key = feature.mm_hash or feature.identifier
feature.data = self.get_and_update_item(feature.data, cache_key)
return mm_features
Site 5 — dropped sparse placeholder mask (transport integrity loss). The render path serializes placeholders as only offset/length, so models relying on sparse is_embed masks lose the mask during render-to-generate replay.
ServingRender._extract_mm_features()builds eachPlaceholderRangeInfo(offset=p.offset, length=p.length), discardingis_embed:vllm/entrypoints/scale_out/render/serving.py#L212-L229.- Transport schema carries no field for the mask:
protocol.py#L28. ServingTokens.serve_tokens()reconstructs a densePlaceholderRangeregardless of the original:serving.py#L148-L152.
Impact
A single authenticated request to a scale-out multimodal deployment can:
- Crash the shared EngineCore process (sites 1, 2, 3), taking the served model down for every tenant (
/health→ 503). Availability-only; no code execution or data disclosure demonstrated for these sites. - Silently poison or read back another request's shared encoder-cache state (site 4) — an integrity/disclosure primitive. Attack complexity is High because the attacker must know or induce the victim's content hash; there is no availability impact for this site.
- Corrupt backend-visible placeholder semantics across the render-to-generate boundary (site 5) for models that depend on sparse
is_embedmasks.
The forged multimodal payload is small; only the trust in its self-declared geometry/identity is the defect.
Suggested Fix
On the scale-out path, do not trust caller-supplied multimodal state as renderer-produced. After decoding features, rebind and validate the reconstructed MultiModalKwargsItem against the active model's renderer contract at the HTTP boundary:
- Reject any request whose decoded grid geometry is inconsistent with the rendered pixel-tensor element count and declared placeholder span, before it reaches
prepare_encoder_metadata()(site 1). - Rebind each field's processor type to the schema the active model's renderer declares (or reject if it differs), instead of reconstructing internal field processors from wire-selected factory names (site 2).
- Reject any
PlaceholderRangeInfowithlength <= 0(or out-of-rangeoffset) with a request-scoped 4xx, and convert the encoder-runner invariant into a checked, request-scoped error rather than a process-fatalassert(site 3). - Recompute or verify the content hash for submitted
kwargs_databefore using it as a cache key, and namespace receiver-cache keys to a server-generated or principal scope, refusing cache-reads for hashes the caller did not legitimately produce (site 4). - Serialize
is_embedinPlaceholderRangeInfo, validate its length against the placeholder span, and reconstruct it on replay (site 5).
Site 1 — validate grid geometry before the vision encoder. Replace the bare assert grid_thw.ndim == 2 in _process_image_input()/_process_video_input() with a shared helper that recomputes the split sizes and rejects a grid whose patch-row count does not match the pixel tensor (and rejects non-positive / non-merge-divisible dims), so the mismatch never reaches image_embeds.split():
# vllm/model_executor/models/qwen2_vl.py — _process_image_input()
- grid_thw = image_input["image_grid_thw"]
- assert grid_thw.ndim == 2
+ grid_thw = image_input["image_grid_thw"]
+ input_type = image_input["type"]
+ input_tensor = (
+ image_input["image_embeds"]
+ if input_type == "image_embeds"
+ else image_input["pixel_values"]
+ )
+ sizes = _validate_qwen2_vl_input_geometry(
+ modality="image",
+ input_type=input_type,
+ input_tensor=input_tensor,
+ grid_thw=grid_thw,
+ spatial_merge_size=self.visual.spatial_merge_size,
+ )
...
- # Split concatenated embeddings for each image item.
- merge_size = self.visual.spatial_merge_size
- sizes = (grid_thw.prod(-1) // merge_size // merge_size).tolist()
return image_embeds.split(sizes)
where the helper raises before the encoder runs:
# vllm/model_executor/models/qwen2_vl.py — new _validate_qwen2_vl_input_geometry()
if t <= 0 or h <= 0 or w <= 0:
raise ValueError(f"{modality} grid_thw row {index} must be positive ...")
if h % spatial_merge_size != 0 or w % spatial_merge_size != 0:
raise ValueError(f"{modality} grid_thw row {index} must be divisible ...")
...
if actual_rows != expected_rows:
raise ValueError(
f"{modality} {row_kind} do not match grid_thw: "
f"expected {expected_rows}, got {actual_rows}."
)
Site 3 — constrain the placeholder schema. Make PlaceholderRangeInfo reject non-positive lengths and negative offsets at the Pydantic boundary (plus parallel-length, non-overlapping, and within-prompt validators), turning the process-fatal assert into a request-scoped 422:
# vllm/entrypoints/scale_out/token_in_token_out/protocol.py — PlaceholderRangeInfo
- offset: int
- length: int
+ offset: int = Field(ge=0)
+ length: int = Field(gt=0)
Site 4 — bind the cache key to the payload. Derive each cache key by hashing the submitted serialized tensor (ignoring the caller's mm_hashes) and refuse cache-only reads, so a forged hash can neither poison nor read a victim's slot:
# vllm/entrypoints/scale_out/token_in_token_out/serving.py — serve_tokens()
+ mm_hashes = _bind_mm_hashes_to_kwargs_data(
+ features.mm_hashes, features.kwargs_data,
+ )
engine_input = mm_input(
prompt_token_ids=request.token_ids,
mm_kwargs=MultiModalKwargsItems(mm_kwargs),
- mm_hashes=features.mm_hashes,
+ mm_hashes=mm_hashes,
mm_placeholders=mm_placeholders,
cache_salt=request.cache_salt,
)
where _bind_mm_hashes_to_kwargs_data() raises on kwargs_data is None (cache-only read) and derives sha256(modality || "\0" || serialized_item) per item. Site 2 applies the same rebind-to-declared-schema pattern in mm_serde.py (passing modality + mm_processor into decode_mm_kwargs_item), and site 5 adds an is_embed field to PlaceholderRangeInfo with from_placeholder_range/to_placeholder_range helpers so the sparse mask survives render-to-generate replay. Each site fix ships with a regression test. This packet groups the five sites because they share one entry point (/inference/v1/generate + /render) and one root cause; we are happy to split it into per-component advisories (for example, engine-fatal input-validation vs. cache-key binding vs. transport-schema integrity) if the vLLM team prefers.
Credit
Reported by: Patch the Planet (Trail of Bits + OpenAI collaboration)
These vulnerabilities were discovered using GPT-5.5-Cyber as part of the Patch the Planet security initiative.
Proposed fix: a fix for this issue is proposed in a public pull request: https://github.com/vllm-project/vllm/pull/51898
{
"affected": [
{
"package": {
"ecosystem": "PyPI",
"name": "vllm"
},
"ranges": [
{
"events": [
{
"introduced": "0"
},
{
"fixed": "0.30.0"
}
],
"type": "ECOSYSTEM"
}
]
}
],
"aliases": [
"CVE-2026-105754"
],
"database_specific": {
"cwe_ids": [
"CWE-1284",
"CWE-20",
"CWE-617",
"CWE-639",
"CWE-668",
"CWE-704"
],
"github_reviewed": true,
"github_reviewed_at": "2026-10-05T23:42:50Z",
"nvd_published_at": null,
"severity": "MODERATE"
},
"details": "## Affected\n\n- **Ecosystem / package:** pip / `vllm`\n- **Affected versions:** vLLM \u2264 0.25.1 (confirmed on 0.25.1, commit [`752a3a504485`](https://github.com/vllm-project/vllm/tree/752a3a504485790a2e8491cacbb35c137339ad34)). The lower bound predates 0.25.1; maintainers can confirm how far back the scale-out transport path reaches.\n\n## Summary\n\nvLLM\u0027s disaggregated **scale-out** transport splits a multimodal request into a trusted render step (`POST /v1/chat/completions/render`) and a separate generate step (`POST /inference/v1/generate`). The generate route decodes a caller-supplied `features` object \u2014 serialized encoder tensors (`kwargs_data`), multimodal hashes (`mm_hashes`), placeholder ranges (`mm_placeholders`), and the internal field-processor selection \u2014 and forwards it into the engine **as if it had come from the trusted renderer**, with no rebinding to (or validation against) the active model\u0027s renderer contract. Because the two routes are ordinary auth-guarded HTTP endpoints (the `/inference` prefix is registered by default on generate-capable servers), any authenticated caller can submit an otherwise-valid render body with a single forged field.\n\nDepending on which field is forged, this produces:\n\n- an **engine-fatal crash** of the shared EngineCore process (denial of service), reproduced as a CUDA illegal-memory-access, a post-admission rank-mismatch `ValueError`, and a hard `assert` \u2014 three independent forged fields (sites 1, 2, 3);\n- **silent cross-request encoder-cache poisoning / disclosure** of shared encoder state when the cache-key hash is not bound to the payload (site 4);\n- **transport-level integrity loss** when sparse placeholder masks are dropped during render-to-generate replay (site 5).\n\nAll five share one root cause and one fix shape: the reconstructed multimodal state on the scale-out path is trusted without being rebound to, and validated against, the active model\u0027s renderer output before it reaches the engine.\n\nThese sites are distinct from prior multimodal hardening. Site 1 survives [GHSA-wv77-2vpf-vmmg](https://github.com/vllm-project/vllm/security/advisories/GHSA-wv77-2vpf-vmmg) (that fix validates full tensor shape in `MultiModalDataParser`/`get_input_embeddings` on the prompt-embeds path), because our request forges `image_grid_thw` metadata with the pixel bytes intact and reaches the Qwen2 vision RoPE/`cu_seqlens` and `image_embeds.split` sink, which the shape-check fix does not rebind. Site 4 is distinct from [GHSA-c65p-x677-fgj6](https://github.com/vllm-project/vllm/security/advisories/GHSA-c65p-x677-fgj6) (which folds metadata into `MultiModalHasher.serialize_item` to stop hash collisions), because the scale-out generate path trusts a caller-supplied `mm_hash` as the cache key with no origin binding, so that fix does not stop a caller from submitting a victim\u0027s hash or a `kwargs_data=None` cache read.\n\n## Affected code\n\nLinks pinned to the confirmed commit [`752a3a504485`](https://github.com/vllm-project/vllm/tree/752a3a504485790a2e8491cacbb35c137339ad34) (v0.25.1).\n\nShared entry point and control surface for all five sites:\n\n- `POST /inference/v1/generate` route: [`vllm/entrypoints/scale_out/token_in_token_out/api_router.py#L46-L75`](https://github.com/vllm-project/vllm/blob/752a3a504485790a2e8491cacbb35c137339ad34/vllm/entrypoints/scale_out/token_in_token_out/api_router.py#L46-L75).\n- Public `features` schema (`kwargs_data`, `mm_hashes`, `mm_placeholders`): [`vllm/entrypoints/scale_out/token_in_token_out/protocol.py#L42-L63`](https://github.com/vllm-project/vllm/blob/752a3a504485790a2e8491cacbb35c137339ad34/vllm/entrypoints/scale_out/token_in_token_out/protocol.py#L42-L63).\n- `ServingTokens.serve_tokens()` copies the decoded geometry and hashes into engine structures without rebinding to the renderer schema: [`vllm/entrypoints/scale_out/token_in_token_out/serving.py#L145-L172`](https://github.com/vllm-project/vllm/blob/752a3a504485790a2e8491cacbb35c137339ad34/vllm/entrypoints/scale_out/token_in_token_out/serving.py#L145-L172).\n- The scale-out routers are registered by default on generate-capable servers and `/inference` is treated as an ordinary auth-guarded prefix: [`vllm/entrypoints/openai/api_server.py#L217-L219`](https://github.com/vllm-project/vllm/blob/752a3a504485790a2e8491cacbb35c137339ad34/vllm/entrypoints/openai/api_server.py#L217-L219).\n\n**Site 1 \u2014 forged Qwen grid geometry (engine-fatal DoS).** The decoded `image_grid_thw` is never rebound to the rendered pixel-tensor element count.\n\n- Reconstruction with no model-specific geometry invariant: [`serving.py#L147-L172`](https://github.com/vllm-project/vllm/blob/752a3a504485790a2e8491cacbb35c137339ad34/vllm/entrypoints/scale_out/token_in_token_out/serving.py#L147-L172).\n- `Qwen2VisionTransformer.prepare_encoder_metadata()` derives RoPE tables, `cu_seqlens`, and the FlashAttention `max_seqlen` from the caller-supplied grid: [`qwen2_vl.py#L658-L712`](https://github.com/vllm-project/vllm/blob/752a3a504485790a2e8491cacbb35c137339ad34/vllm/model_executor/models/qwen2_vl.py#L658-L712), consumed in [`forward()` (`#L713-L751`)](https://github.com/vllm-project/vllm/blob/752a3a504485790a2e8491cacbb35c137339ad34/vllm/model_executor/models/qwen2_vl.py#L713-L751).\n- `_process_image_input()` computes `image_embeds.split(sizes)` from the same untrusted metadata: [`qwen2_vl.py#L1348-L1369`](https://github.com/vllm-project/vllm/blob/752a3a504485790a2e8491cacbb35c137339ad34/vllm/model_executor/models/qwen2_vl.py#L1348-L1369); M-RoPE positions at [`qwen2_vl.py#L1223`](https://github.com/vllm-project/vllm/blob/752a3a504485790a2e8491cacbb35c137339ad34/vllm/model_executor/models/qwen2_vl.py#L1223).\n\nThe only shape check is `assert grid_thw.ndim == 2`; the split sizes and the vision-encoder call are then derived directly from the caller-supplied grid, with no cross-check against the pixel-tensor row count:\n\n```python\n# vllm/model_executor/models/qwen2_vl.py Lines 1348-1369\n def _process_image_input(\n self, image_input: Qwen2VLImageInputs\n ) -\u003e tuple[torch.Tensor, ...]:\n grid_thw = image_input[\"image_grid_thw\"]\n assert grid_thw.ndim == 2\n\n if image_input[\"type\"] == \"image_embeds\":\n image_embeds = image_input[\"image_embeds\"]\n else:\n pixel_values = image_input[\"pixel_values\"]\n\n if self.use_data_parallel:\n return run_dp_sharded_mrope_vision_model(\n self.visual, pixel_values, grid_thw.tolist(), rope_type=\"rope_3d\"\n )\n else:\n image_embeds = self.visual(pixel_values, grid_thw=grid_thw)\n\n # Split concatenated embeddings for each image item.\n merge_size = self.visual.spatial_merge_size\n sizes = (grid_thw.prod(-1) // merge_size // merge_size).tolist()\n return image_embeds.split(sizes)\n```\n\n**Site 2 \u2014 wire-selected field-processor type confusion (engine-fatal DoS).** `MsgpackDecoder._decode_mm_field_elem()` trusts a wire-selected field-factory name and constructs the internal field processor directly from caller data.\n\n- [`vllm/v1/serial_utils.py#L440-L454`](https://github.com/vllm-project/vllm/blob/752a3a504485790a2e8491cacbb35c137339ad34/vllm/v1/serial_utils.py#L440-L454) reads `factory_meth_name, factory_kw = obj[\"field\"]` and calls `getattr(MultiModalFieldConfig, factory_meth_name)`.\n- Qwen2-VL field/schema contract: [`qwen2_vl.py#L763-L791`](https://github.com/vllm-project/vllm/blob/752a3a504485790a2e8491cacbb35c137339ad34/vllm/model_executor/models/qwen2_vl.py#L763-L791) and [`Qwen2VLImagePixelInputs` (`#L119-L144`)](https://github.com/vllm-project/vllm/blob/752a3a504485790a2e8491cacbb35c137339ad34/vllm/model_executor/models/qwen2_vl.py#L119-L144); parse/validate at [`qwen2_vl.py#L1300`](https://github.com/vllm-project/vllm/blob/752a3a504485790a2e8491cacbb35c137339ad34/vllm/model_executor/models/qwen2_vl.py#L1300).\n- Rank mismatch raised post-admission: [`vllm/utils/tensor_schema.py#L155-L171`](https://github.com/vllm-project/vllm/blob/752a3a504485790a2e8491cacbb35c137339ad34/vllm/utils/tensor_schema.py#L155-L171), reached from [`TensorSchema.__init__` \u2192 `validate()` (`#L63`)](https://github.com/vllm-project/vllm/blob/752a3a504485790a2e8491cacbb35c137339ad34/vllm/utils/tensor_schema.py#L63).\n\n**Site 3 \u2014 non-positive placeholder length (engine-fatal DoS via reachable assert).** `PlaceholderRangeInfo{offset,length}` is accepted as unconstrained integers and copied verbatim into the engine\u0027s `PlaceholderRange`.\n\n- Schema: [`protocol.py#L28-L35`](https://github.com/vllm-project/vllm/blob/752a3a504485790a2e8491cacbb35c137339ad34/vllm/entrypoints/scale_out/token_in_token_out/protocol.py#L28-L35).\n- Copied into `PlaceholderRange`: [`serving.py#L147-L153`](https://github.com/vllm-project/vllm/blob/752a3a504485790a2e8491cacbb35c137339ad34/vllm/entrypoints/scale_out/token_in_token_out/serving.py#L147-L153).\n- The only guard is an upper bound on embed count (`num_embeds \u003e mm_encoder_cache_size`), with no non-positive check: [`vllm/v1/engine/input_processor.py#L459-L464`](https://github.com/vllm-project/vllm/blob/752a3a504485790a2e8491cacbb35c137339ad34/vllm/v1/engine/input_processor.py#L459-L464); raw length returned by [`vllm/multimodal/inputs.py#L152-L154`](https://github.com/vllm-project/vllm/blob/752a3a504485790a2e8491cacbb35c137339ad34/vllm/multimodal/inputs.py#L152-L154); window selection assumes non-empty ranges at [`vllm/multimodal/utils.py#L114-L134`](https://github.com/vllm-project/vllm/blob/752a3a504485790a2e8491cacbb35c137339ad34/vllm/multimodal/utils.py#L114-L134).\n- Sink: `assert start_idx \u003c end_idx` at [`vllm/v1/worker/gpu/mm/encoder_runner.py#L114`](https://github.com/vllm-project/vllm/blob/752a3a504485790a2e8491cacbb35c137339ad34/vllm/v1/worker/gpu/mm/encoder_runner.py#L114) (duplicated at [`vllm/v1/worker/gpu_model_runner.py#L3192`](https://github.com/vllm-project/vllm/blob/752a3a504485790a2e8491cacbb35c137339ad34/vllm/v1/worker/gpu_model_runner.py#L3192)); turned into a fatal shutdown by EngineCore\u0027s uncaught-exception path at [`vllm/v1/engine/core.py#L1229-L1233`](https://github.com/vllm-project/vllm/blob/752a3a504485790a2e8491cacbb35c137339ad34/vllm/v1/engine/core.py#L1229-L1233).\n\nA `length` of `0` makes `num_encoder_tokens == 0`, so `end_idx` collapses to `0` and the bare `assert` fires inside the engine worker:\n\n```python\n# vllm/v1/worker/gpu/mm/encoder_runner.py Lines 108-114\n pos_info = mm_feature.mm_position\n start_pos = pos_info.offset\n num_encoder_tokens = pos_info.length\n\n start_idx = max(cur_query_start - start_pos, 0)\n end_idx = min(cur_query_end - start_pos, num_encoder_tokens)\n assert start_idx \u003c end_idx\n```\n\n**Site 4 \u2014 cache hash not bound to payload (integrity / disclosure).** `features.mm_hashes` (the cache key) and `kwargs_data` (the tensor) are independent fields with no origin or integrity binding.\n\n- Schema exposing the caller-controlled hash and the `None` = resolve-from-cache semantics: [`protocol.py#L42-L63`](https://github.com/vllm-project/vllm/blob/752a3a504485790a2e8491cacbb35c137339ad34/vllm/entrypoints/scale_out/token_in_token_out/protocol.py#L42-L63).\n- Handler forwards the caller\u0027s hashes unchanged into `mm_input(...)`: [`serving.py#L164-L170`](https://github.com/vllm-project/vllm/blob/752a3a504485790a2e8491cacbb35c137339ad34/vllm/entrypoints/scale_out/token_in_token_out/serving.py#L164-L170).\n- Input processing copies the caller hash into the feature identifier with only a string-type check: [`vllm/v1/engine/input_processor.py#L165-L181`](https://github.com/vllm-project/vllm/blob/752a3a504485790a2e8491cacbb35c137339ad34/vllm/v1/engine/input_processor.py#L165-L181).\n- Sink \u2014 the receiver cache returns the cached tensor solely by that key (`cache_key = feature.mm_hash or feature.identifier`): [`vllm/multimodal/cache.py#L602-L607`](https://github.com/vllm-project/vllm/blob/752a3a504485790a2e8491cacbb35c137339ad34/vllm/multimodal/cache.py#L602-L607).\n\nThe cache key is the caller-supplied hash with no verification against the tensor bytes, so a forged `mm_hash` both stores under and reads back another request\u0027s slot:\n\n```python\n# vllm/multimodal/cache.py Lines 601-607\n for feature in mm_features:\n cache_key = feature.mm_hash or feature.identifier\n self.touch_receiver_cache_item(cache_key, feature.data)\n\n for feature in mm_features:\n cache_key = feature.mm_hash or feature.identifier\n feature.data = self.get_and_update_item(feature.data, cache_key)\n return mm_features\n```\n\n**Site 5 \u2014 dropped sparse placeholder mask (transport integrity loss).** The render path serializes placeholders as only `offset`/`length`, so models relying on sparse `is_embed` masks lose the mask during render-to-generate replay.\n\n- `ServingRender._extract_mm_features()` builds each `PlaceholderRangeInfo(offset=p.offset, length=p.length)`, discarding `is_embed`: [`vllm/entrypoints/scale_out/render/serving.py#L212-L229`](https://github.com/vllm-project/vllm/blob/752a3a504485790a2e8491cacbb35c137339ad34/vllm/entrypoints/scale_out/render/serving.py#L212-L229).\n- Transport schema carries no field for the mask: [`protocol.py#L28`](https://github.com/vllm-project/vllm/blob/752a3a504485790a2e8491cacbb35c137339ad34/vllm/entrypoints/scale_out/token_in_token_out/protocol.py#L28).\n- `ServingTokens.serve_tokens()` reconstructs a dense `PlaceholderRange` regardless of the original: [`serving.py#L148-L152`](https://github.com/vllm-project/vllm/blob/752a3a504485790a2e8491cacbb35c137339ad34/vllm/entrypoints/scale_out/token_in_token_out/serving.py#L148-L152).\n\n## Impact\n\nA single authenticated request to a scale-out multimodal deployment can:\n\n- **Crash the shared EngineCore process** (sites 1, 2, 3), taking the served model down for every tenant (`/health` \u2192 503). Availability-only; no code execution or data disclosure demonstrated for these sites.\n- **Silently poison or read back another request\u0027s shared encoder-cache state** (site 4) \u2014 an integrity/disclosure primitive. Attack complexity is High because the attacker must know or induce the victim\u0027s content hash; there is no availability impact for this site.\n- **Corrupt backend-visible placeholder semantics across the render-to-generate boundary** (site 5) for models that depend on sparse `is_embed` masks.\n\nThe forged multimodal payload is small; only the trust in its self-declared geometry/identity is the defect.\n\n\n## Suggested Fix\n\nOn the scale-out path, do not trust caller-supplied multimodal state as renderer-produced. After decoding `features`, **rebind and validate** the reconstructed `MultiModalKwargsItem` against the active model\u0027s renderer contract at the HTTP boundary:\n\n1. Reject any request whose decoded grid geometry is inconsistent with the rendered pixel-tensor element count and declared placeholder span, before it reaches `prepare_encoder_metadata()` (site 1).\n2. Rebind each field\u0027s processor type to the schema the active model\u0027s renderer declares (or reject if it differs), instead of reconstructing internal field processors from wire-selected factory names (site 2).\n3. Reject any `PlaceholderRangeInfo` with `length \u003c= 0` (or out-of-range `offset`) with a request-scoped 4xx, and convert the encoder-runner invariant into a checked, request-scoped error rather than a process-fatal `assert` (site 3).\n4. Recompute or verify the content hash for submitted `kwargs_data` before using it as a cache key, and namespace receiver-cache keys to a server-generated or principal scope, refusing cache-reads for hashes the caller did not legitimately produce (site 4).\n5. Serialize `is_embed` in `PlaceholderRangeInfo`, validate its length against the placeholder span, and reconstruct it on replay (site 5).\n\n**Site 1 \u2014 validate grid geometry before the vision encoder.** Replace the bare `assert grid_thw.ndim == 2` in `_process_image_input()`/`_process_video_input()` with a shared helper that recomputes the split sizes and rejects a grid whose patch-row count does not match the pixel tensor (and rejects non-positive / non-merge-divisible dims), so the mismatch never reaches `image_embeds.split()`:\n\n```python\n# vllm/model_executor/models/qwen2_vl.py \u2014 _process_image_input()\n- grid_thw = image_input[\"image_grid_thw\"]\n- assert grid_thw.ndim == 2\n+ grid_thw = image_input[\"image_grid_thw\"]\n+ input_type = image_input[\"type\"]\n+ input_tensor = (\n+ image_input[\"image_embeds\"]\n+ if input_type == \"image_embeds\"\n+ else image_input[\"pixel_values\"]\n+ )\n+ sizes = _validate_qwen2_vl_input_geometry(\n+ modality=\"image\",\n+ input_type=input_type,\n+ input_tensor=input_tensor,\n+ grid_thw=grid_thw,\n+ spatial_merge_size=self.visual.spatial_merge_size,\n+ )\n...\n- # Split concatenated embeddings for each image item.\n- merge_size = self.visual.spatial_merge_size\n- sizes = (grid_thw.prod(-1) // merge_size // merge_size).tolist()\n return image_embeds.split(sizes)\n```\n\nwhere the helper raises before the encoder runs:\n\n```python\n# vllm/model_executor/models/qwen2_vl.py \u2014 new _validate_qwen2_vl_input_geometry()\n if t \u003c= 0 or h \u003c= 0 or w \u003c= 0:\n raise ValueError(f\"{modality} grid_thw row {index} must be positive ...\")\n if h % spatial_merge_size != 0 or w % spatial_merge_size != 0:\n raise ValueError(f\"{modality} grid_thw row {index} must be divisible ...\")\n ...\n if actual_rows != expected_rows:\n raise ValueError(\n f\"{modality} {row_kind} do not match grid_thw: \"\n f\"expected {expected_rows}, got {actual_rows}.\"\n )\n```\n\n**Site 3 \u2014 constrain the placeholder schema.** Make `PlaceholderRangeInfo` reject non-positive lengths and negative offsets at the Pydantic boundary (plus parallel-length, non-overlapping, and within-prompt validators), turning the process-fatal `assert` into a request-scoped 422:\n\n```python\n# vllm/entrypoints/scale_out/token_in_token_out/protocol.py \u2014 PlaceholderRangeInfo\n- offset: int\n- length: int\n+ offset: int = Field(ge=0)\n+ length: int = Field(gt=0)\n```\n\n**Site 4 \u2014 bind the cache key to the payload.** Derive each cache key by hashing the submitted serialized tensor (ignoring the caller\u0027s `mm_hashes`) and refuse cache-only reads, so a forged hash can neither poison nor read a victim\u0027s slot:\n\n```python\n# vllm/entrypoints/scale_out/token_in_token_out/serving.py \u2014 serve_tokens()\n+ mm_hashes = _bind_mm_hashes_to_kwargs_data(\n+ features.mm_hashes, features.kwargs_data,\n+ )\n engine_input = mm_input(\n prompt_token_ids=request.token_ids,\n mm_kwargs=MultiModalKwargsItems(mm_kwargs),\n- mm_hashes=features.mm_hashes,\n+ mm_hashes=mm_hashes,\n mm_placeholders=mm_placeholders,\n cache_salt=request.cache_salt,\n )\n```\n\nwhere `_bind_mm_hashes_to_kwargs_data()` raises on `kwargs_data is None` (cache-only read) and derives `sha256(modality || \"\\0\" || serialized_item)` per item. Site 2 applies the same rebind-to-declared-schema pattern in `mm_serde.py` (passing `modality` + `mm_processor` into `decode_mm_kwargs_item`), and site 5 adds an `is_embed` field to `PlaceholderRangeInfo` with `from_placeholder_range`/`to_placeholder_range` helpers so the sparse mask survives render-to-generate replay. Each site fix ships with a regression test. This packet groups the five sites because they share one entry point (`/inference/v1/generate` + `/render`) and one root cause; we are happy to split it into per-component advisories (for example, engine-fatal input-validation vs. cache-key binding vs. transport-schema integrity) if the vLLM team prefers.\n\n## Credit\n\n**Reported by:** Patch the Planet (Trail of Bits + OpenAI collaboration)\n\nThese vulnerabilities were discovered using GPT-5.5-Cyber as part of the Patch the Planet security initiative.\n\n---\n\n**Proposed fix:** a fix for this issue is proposed in a public pull request: https://github.com/vllm-project/vllm/pull/51898",
"id": "GHSA-ph72-cqr5-qpp7",
"modified": "2026-10-05T23:42:50Z",
"published": "2026-10-05T23:42:50Z",
"references": [
{
"type": "WEB",
"url": "https://github.com/vllm-project/vllm/security/advisories/GHSA-ph72-cqr5-qpp7"
},
{
"type": "WEB",
"url": "https://github.com/vllm-project/vllm/pull/51898"
},
{
"type": "WEB",
"url": "https://github.com/vllm-project/vllm/commit/1970f3ed4be7fa8620e4ddc4a12c36a8384cfc27"
},
{
"type": "PACKAGE",
"url": "https://github.com/vllm-project/vllm"
},
{
"type": "WEB",
"url": "https://github.com/vllm-project/vllm/releases/tag/v0.30.0"
}
],
"schema_version": "1.4.0",
"severity": [
{
"score": "CVSS:3.1/AV:N/AC:L/PR:L/UI:N/S:U/C:N/I:N/A:H",
"type": "CVSS_V3"
}
],
"summary": "vLLM: Scale-out disaggregated multimodal transport trusts caller-supplied features"
}
GHSA-PHG3-P22V-RP9Q
Vulnerability from github – Published: 2026-09-05 12:31 – Updated: 2026-09-05 12:31BookWyrm through 0.9.1 fails to validate user visibility permissions in the status edit endpoint, allowing authenticated attackers to read followers-only and direct-message reviews by enumerating sequential status IDs. Attackers can access the raw content of restricted statuses through the edit view, bypassing the privacy protections documented for these message types.
{
"affected": [],
"aliases": [
"CVE-2026-86111"
],
"database_specific": {
"cwe_ids": [
"CWE-639"
],
"github_reviewed": false,
"github_reviewed_at": null,
"nvd_published_at": "2026-09-05T10:16:40Z",
"severity": "HIGH"
},
"details": "BookWyrm through 0.9.1 fails to validate user visibility permissions in the status edit endpoint, allowing authenticated attackers to read followers-only and direct-message reviews by enumerating sequential status IDs. Attackers can access the raw content of restricted statuses through the edit view, bypassing the privacy protections documented for these message types.",
"id": "GHSA-phg3-p22v-rp9q",
"modified": "2026-09-05T12:31:25Z",
"published": "2026-09-05T12:31:25Z",
"references": [
{
"type": "ADVISORY",
"url": "https://nvd.nist.gov/vuln/detail/CVE-2026-86111"
},
{
"type": "WEB",
"url": "https://github.com/bookwyrm-social/bookwyrm"
},
{
"type": "WEB",
"url": "https://github.com/bookwyrm-social/bookwyrm/blob/v0.9.1/bookwyrm/templates/snippets/create_status/content_field.html"
},
{
"type": "WEB",
"url": "https://github.com/bookwyrm-social/bookwyrm/blob/v0.9.1/bookwyrm/views/status.py"
},
{
"type": "WEB",
"url": "https://github.com/geo-chen/oss/blob/main/bookwyrm.md#finding-1-authenticated-idor-in-editstatus-exposes-private-review-comment-and-quotation-content"
},
{
"type": "WEB",
"url": "https://www.vulncheck.com/advisories/bookwyrm-through-0.9.1-insecure-direct-object-reference-in-editstatus-exposes-followers-only-and-direct-statuses"
}
],
"schema_version": "1.4.0",
"severity": [
{
"score": "CVSS:3.1/AV:N/AC:L/PR:L/UI:N/S:U/C:H/I:N/A:N",
"type": "CVSS_V3"
},
{
"score": "CVSS:4.0/AV:N/AC:L/AT:N/PR:L/UI:N/VC:H/VI:N/VA:N/SC:N/SI:N/SA:N/E:X/CR:X/IR:X/AR:X/MAV:X/MAC:X/MAT:X/MPR:X/MUI:X/MVC:X/MVI:X/MVA:X/MSC:X/MSI:X/MSA:X/S:X/AU:X/R:X/V:X/RE:X/U:X",
"type": "CVSS_V4"
}
]
}
Mitigation
For each and every data access, ensure that the user has sufficient privilege to access the record that is being requested.
Mitigation
Make sure that the key that is used in the lookup of a specific user's record is not controllable externally by the user or that any tampering can be detected.
Mitigation
Use encryption in order to make it more difficult to guess other legitimate values of the key or associate a digital signature with the key so that the server can verify that there has been no tampering.
No CAPEC attack patterns related to this CWE.