提交 #825558: Genspark Genspark AI Workspace 2.8.4 (versionCode 28401) Path Traversal

Genspark's exported MainActivity accepts ACTION_SEND / ACTION_VIEW intents carrying a content:// URI. The app reads _display_name from the URI and uses it verbatim as the destination filename for the file copy - no sanitization. A zero-permission attacker app supplying a _display_name containing ../ writes arbitrary files into the victim app's UID-protected internal storage tree.
When the URI points at an attacker-controlled ContentProvider, the app reads the URI's _display_name metadata column and uses that string verbatim as the destination filename when copying the streamed bytes to disk. Because the value is attacker-controlled and never sanitized, supplying a _display_name containing ../ sequences gives the attacker arbitrary file creation inside the victim app's own private storage

The trust model is the bug: the app treats _display_name as a safe filename when the documentation explicitly states it is a display string and may contain anything the source provider chooses to return. A zero-permission attacker app delivers a single intent, and the victim app - running in its own process with its own UID - opens an output stream at the attacker-resolved path and writes the attacker-supplied bytes.

• 来源: ⚠️ https://github.com/actuator/ai.mainfunc.genspark
• 用户: Actuator (UID 67941)
• 提交: 2026-05-11 16時27分 (5 月前)
• 管理: 2026-06-14 08時38分 (1 month later)
• 状态: 已接受
• VulDB条目: 370836 [Genspark AI Workspace App 2.8.4 于 Android ai.mainfunc.genspark Local Privilege Escalation]
• 积分: 20