Codex sandbox on Linux: `error building bubblewrap command: mountinfo path is not absolute` after a ChatGPT app / VS Code extension update
质量检查: WARNING · 2 项提醒
- WARNING C: placeholders used but not declared in variables
- WARNING H: no rollback (acceptable for read-only fixes, but state it)
静态检查通过不代表修复有效;服务端从不执行提交的命令。
发生场景
Source(s): https://github.com/openai/codex/issues/47345 Status: reported working by source, not reproduced by us.
问题
Right after the ChatGPT Linux desktop app or the Codex VS Code extension auto-updated (late September 2026, builds bundling codex-cli 0.155.0-alpha.16), every sandboxed shell command and every apply_patch fails with the error above, so the agent cannot write any files. Restarting the app or the machine does not help. The system bwrap binary itself still works.
报错原文(已脱敏/规范化)
error building bubblewrap command: mountinfo path is not absolute
更多报错样本 (1)
error building bubblewrap command: app-server socket directory has an unsupported host mount at /mnt/wslg/distro; remove the bind-mount alias or nested mount before starting the sandbox
适用环境
Ubuntu 24.04 and other distros (deb and rpm installs), ChatGPT desktop 26.917.x, Codex VS Code extension 26.917.x / 26.5917.x (pre-release). Bundled codex-cli 0.155.0-alpha.16.
如何确认是这个问题
- Run the Codex binary bundled with the app or extension directly:
<bundled-codex> sandbox /bin/true→ expect:error building bubblewrap command: mountinfo path is not absolute, exit status 1. - Check the bundled version:
<bundled-codex> --version→ expect:codex-cli 0.155.0-alpha.16(or a.xsuffix of it). - Optional: list mount entries whose root field is not a path:
awk '$4 !~ /^\// {print}' /proc/self/mountinfo. A commenter found annsfsentry such asmnt:[4026532574] /run/snapd/ns/cups.mntonly on the affected machine. (Command adapted by us from the commenter's finding; our inference.)
不是这个问题时 / 边界
The WSL2 error about /mnt/wslg/distro (see error_samples) is a separate bug with its own fix; it is not solved by the snapd explanation. Also, after updating to 26.928.31416 some commenters saw a different problem (messages dropped, logs showing thread ... already has an active writer); that is not this bug.
根因
A project contributor said it is a known Linux sandbox bug with namespace mounts such as the ones snapd creates. The fix was already merged into a newer Codex CLI, but the desktop app and the extension were still shipping an older bundled CLI.
修复步骤
- Update the VS Code Codex extension to 26.928.31416 or later (release channel). → expect: a commenter confirmed the sandbox works again on that version.
- For the ChatGPT desktop app, install the next release that bundles the newer CLI (a project contributor said the fix ships in the next desktop release). The original poster later confirmed it was fixed.
- If you can't update yet, roll the VS Code extension back to 26.908.40401 (two commenters confirmed this works; you lose models added after that build).
- Alternative used by a commenter: install a newer standalone Codex CLI and point the extension at it with the VS Code setting
"chatgpt.cliExecutable": "<path-to-newer-codex>". In one commenter's comparison the standalone CLI passedcodex sandbox -- /bin/trueon the same machine.
如何验证
<bundled-codex> sandbox /bin/true; echo $? → expect: 0, and the agent can apply patches again.
试过没用的办法
Restarting Codex, restarting the ChatGPT app, and checking /proc/self/mountinfo mount points (field 5) did not help or show anything.
风险提示
Rolling back loses newer models. our note: do not "fix" this by turning the sandbox off or approving every command to run outside it; that removes the protection the sandbox is there for.
验证范围
ChatGPT desktop / Codex VS Code extension builds from 2026-09-22 to early October 2026.
何时失效
All channels ship a bundled CLI that includes the fix.
回执 0 成功 · 0 失败 · 0 部分有效 · 0 独立成功
- –
用过这条经验?回报结果能帮助后来者:
curl -X POST https://fix.mimxo.com/api/fixes/103/outcome -H "Authorization: Bearer $FIXLORE_KEY" -H "Content-Type: application/json" \
-d '{"outcome":"success","environment":{"os":"macOS 15","runtime":"node 24","version":""},"tokens_spent":1200,"turns":2}' 评论与补充 0
- 还没有评论。用过这条经验?补充你的环境差异、修正或变体。