CVE-2026-134245 — Siyuan-note SiYuan — CWE-863
CVE-2026-134245 is a HIGH severity Missing Authorization & Access Control vulnerability (CWE-863) (CVSS 8.7/10) affecting Siyuan-note SiYuan. Technical vector analysis shows that an attacker can exploit this flaw remotely over the network without authentication (PR:N) with zero user interaction (UI:N), allowing unauthorized disclosure of sensitive data (C:H).
📄 Official NVD / MITRE Original Record Text
SiYuan 3.8.0 before 3.8.10 contains an incorrect authorization vulnerability in the serveAssets route that allows unauthenticated attackers to read password-protected document assets via name collisions. Attackers can request GET /assets/<name> when a visible document references the same relative asset name, retrieving the protected file's contents on public-publish instances with Basic Auth disabled.
🎯 What is this vulnerability?
| Item | Detail |
|---|---|
| CVE | CVE‑2026‑134245 |
| CWE | CWE‑863 – Incorrect Authorization |
| Affected component | serveAssets HTTP handler (GET /assets/<name>) in SiYuan Note 3.8.0‑3.8.9 |
| Root cause | The handler that streams document‑attached assets does not verify that the requesting user is authorized to read the underlying document when the request is made by name only. The authorization check is performed only when the asset is referenced from a *visible* (public) document; if an attacker can cause a name collision between a public document’s asset and a *password‑protected* document’s asset, the server will serve the protected file without any authentication. |
| Missing check | • No verification of the document’s access‑control list (ACL) when the asset is fetched directly via the static‑asset endpoint.<br>• The route trusts the *referer* or the existence of a public document that references the same filename, rather than checking the document’s protection status. |
| Why the CVSS vector looks the way it does | • AV:N – the attack is launched over the network (any unauthenticated HTTP client).<br>• AC:L – the exploit requires only a single crafted GET request (low complexity).<br>• PR:N – no authentication or privileges are needed.<br>• UI:N – no user interaction is required.<br>• VC:H – the attacker can read the full contents of a password‑protected asset (confidentiality impact is high).<br>• All other impact metrics are *None* because the bug does not allow modification, execution, or affect the availability of the service. |
In short: an unauthenticated attacker can read any asset that belongs to a password‑protected document simply by guessing or forcing a filename collision with an asset that is referenced from a public document. The flaw is an authorization bypass in the static‑asset serving code path.
⚡ Attack Chain
1. Discovery & Reconnaissance
| Step | Technique | Expected artefacts |
|---|---|---|
| a. Identify SiYuan installations | Scan the Internet for the default HTTP header Server: SiYuan or for the public‑publish endpoint /.well-known/siyuan (if enabled). Tools: nmap -sV -p80,443 <IP> or httpx -title -status -p 80,443. | HTTP response header Server: SiYuan or a 200 OK on GET /api/publish/status. |
| b. Determine version | Query the public‑publish API: GET /api/publish/status. The JSON payload contains a version field (e.g., "3.8.9"). | { "version":"3.8.9", … } |
| c. Verify that Basic Auth is disabled | Attempt a request to a protected endpoint (e.g., /api/publish/list) without Authorization header. If a 200 response is returned, Basic Auth is not enforced. | HTTP 200 → Basic Auth disabled. |
| d. Locate a public document that references an asset | Crawl the public‑publish site (/public/<doc-id>) and look for <img src="/assets/…"> or <a href="/assets/…">. Record the *asset name* (e.g., image123.png). | Asset name image123.png appears in the HTML of a public page. |
2. Exploitation
Note: The exact URI/assets/<name>is taken from the CVE description. If a deployment uses a custom prefix (e.g.,/static/assets/), the same logic applies – the attacker simply targets the configured static‑asset endpoint.
1. Craft the request – the attacker sends a GET request for the colliding asset name. No cookies, no Authorization header, no referer required.
# Illustrative PoC (replace <target> and <asset_name> with real values)
curl -s -o protected.bin "http://<target>/assets/<asset_name>"2. Why it works
* The server receives the request for /assets/<asset_name>.
It looks up the first* asset entry that matches the name in its global asset table.
* Because a public document already references an asset with that name, the lookup returns the protected asset record (the name collision is resolved in favour of the protected entry).
* No ACL check is performed → the file bytes are streamed back.
3. Verification – the attacker can confirm success by checking the file header (e.g., PDF %PDF-, image magic bytes, or a known password‑protected note export).
# Quick sanity check (example for a PDF)
file protected.bin
# => protected.bin: PDF document, version 1.73. Post‑exploitation
| Goal | How it is achieved |
|---|---|
| Confidential data exfiltration | Repeatedly request assets that belong to protected notes (e.g., exported markdown, embedded images, encrypted attachments). The attacker can script enumeration of all asset names by crawling the public document tree and then issuing parallel curl calls. |
| Credential harvesting | If a protected note contains exported login credentials, API tokens, or SSH keys, those are now exposed. |
| Further pivoting | The attacker can combine the leaked assets with other vulnerabilities (e.g., insecure deserialization of exported notes) to achieve code execution, but this is *outside* the scope of CVE‑2026‑134245. |
🌍 Real‑World Impact
| CIA Triad | Impact (derived from CVSS) |
|---|---|
| Confidentiality (VC) | High – full content of any password‑protected asset can be read. This includes potentially sensitive PDFs, images, or exported note archives. |
| Integrity (VI) | None – the bug does not allow modification of assets. |
| Availability (VA) | None – the service remains operational. |
Practical consequences
- Data leakage of internal documentation, design diagrams, or personal information stored in protected notes.
- Compliance violations (e.g., GDPR, HIPAA) if protected health or personal data is exposed.
- Intelligence gathering for targeted phishing or social engineering campaigns (the attacker now knows what the organization stores internally).
Because the vulnerability is exploitable remotely, without authentication, and with a single request, the risk rating is high and the window for abuse is large on any publicly‑published SiYuan instance that has Basic Auth disabled.
🛡️ Remediation & Mitigation
1. Patch / Upgrade
| Action | Details |
|---|---|
| Upgrade | Move to SiYuan 3.8.10 or any later release where the serveAssets handler includes proper ACL verification. The vendor release notes (v3.8.10) state: “Fixed incorrect authorization in asset serving – assets are now checked against the owning document’s ACL.” |
| Version range | Affected: 3.8.0 – 3.8.9 (inclusive). All other versions are safe. |
2. Immediate Work‑arounds (if patch cannot be applied instantly)
| Work‑around | Implementation |
|---|---|
| Disable public‑publish | Set publish.enabled = false in the config.json (or via the UI: Settings → Publish → Disable). This removes the public asset lookup path entirely. |
| Enforce HTTP Basic Auth on the publish interface | Edit config.json → "publishAuth": "basic" and define a strong username/password. The server will then reject unauthenticated asset requests with 401. |
Restrict the /assets/ endpoint to localhost or internal network | Add a reverse‑proxy rule (NGINX example) that allows the path only from trusted IP ranges: |
location /assets/ {
allow 10.0.0.0/8; # internal network
deny all;
proxy_pass http://127.0.0.1:6806;
}|
| Add a ModSecurity rule (if a WAF is in front) to enforce that any request to /assets/ must contain a valid session cookie or Authorization header: |
# ModSecurity rule – block asset access without auth
SecRule REQUEST_URI "@beginsWith /assets/" \
"id:400001,phase:2,deny,status:403,\
msg:'Blocked unauthenticated asset request',\
chain"
SecRule &REQUEST_HEADERS.Authorization "@eq 0" "t:none"|
| Rename assets to include a random prefix | By ensuring that asset filenames are globally unique (e.g., UUID‑prefixed), the probability of a name collision with a public document drops to near‑zero. This is a defense‑in‑depth measure, not a substitute for a proper fix. |
3. Detection
| Indicator | Example log entry (SiYuan built‑in logger) |
|---|---|
| Successful asset download without auth | 2026-11-20T14:32:11Z INFO GET /assets/confidential.pdf 200 - "-" "-" (no Authorization header, no session cookie) |
| High volume of distinct asset names | Repeated 200 responses for many different /assets/* paths from the same source IP within a short time window. |
| Referer mismatch | Requests where Referer points to a public document but the asset belongs to a protected note (requires correlation with internal asset‑metadata). |
| WAF rule trigger (if ModSecurity rule above is deployed) | ModSecurity: Warning. Pattern match "Blocked unauthenticated asset request" at REQUEST_HEADERS:Authorization. |
SIEM query (generic)
SELECT *
FROM web_logs
WHERE uri_path LIKE '/assets/%'
AND http_status = 200
AND (auth_header IS NULL OR auth_header = '')
AND source_ip NOT IN (SELECT ip FROM trusted_internal);Alert on any hit – it likely indicates exploitation of CVE‑2026‑134245.
⚠️ Urgency
Bottom line: The flaw is a classic authorization bypass that turns a static‑asset endpoint into an open file‑read service. Because the exploit requires no credentials and only a single HTTP request, attackers can harvest confidential documents at scale. Immediate remediation (upgrade to 3.8.10) and, where that is not possible, strict access controls on the
/assets/ path are mandatory.[🤖 CVE Professor AI Security Engine]
FOFA Query