exec /entrypoint.sh: no such file or directory? Fix CRLF line endings, shebangs and chmod before Docker runs

The image builds, the container exits instantly, and the log says:

exec /entrypoint.sh: no such file or directory

ls -l /entrypoint.sh inside the image shows the file right there. So what is missing?

The kernel reads your first line

For a script, execve() reads the #! line and runs the interpreter it names. Four invisible problems break that step:

In the file	The kernel sees	You get
CRLF line endings	interpreter /bin/sh\r	no such file or directory
#!/usr/bin/env bash + CRLF	a program called bash\r	/usr/bin/env: 'bash\r': No such file or directory
No shebang, or a UTF-8 BOM before it	no #! at all	exec format error
Git mode 100644	a non-executable file	permission denied

The missing file is the interpreter, but the message names your script. The usual source is a Windows checkout with core.autocrlf=true: docker build sends that CRLF working tree, and COPY puts it in the image (real example, Moby's request for a better error).

Fix it by hand

file entrypoint.sh               # "with CRLF line terminators"?
git ls-files -s entrypoint.sh    # 100644 = not executable

sed -i 's/\r$//' entrypoint.sh
echo '*.sh text eol=lf' >> .gitattributes && git add --renormalize .
git update-index --chmod=+x entrypoint.sh

The .gitattributes rule is the part that stops it coming back.

Catch it in the editor

I built EntryCheck, a free VS Code extension that follows Dockerfile ENTRYPOINT/CMD/RUN (through COPY and WORKDIR), compose entrypoint/command, GitHub Actions run:, GitLab CI script:, package.json scripts and Husky hooks to the scripts they execute, and flags:

• EC001 CRLF line endings (Quick Fix: convert to LF, or add the .gitattributes rule)
• EC002 missing or non-absolute shebang on a directly executed script
• EC003 UTF-8 BOM before the shebang (Quick Fix: strip it)
• EC004 mode 100644 in git, unless the Dockerfile already does chmod +x or COPY --chmod=755
• EC005 no .gitattributes eol rule for *.sh

The diagnostic lands on the line that launches the script, and a Failing scripts panel in the activity bar lists every problem by script name with one-click fixes. No network, no telemetry, and the git check only runs in trusted workspaces. It does not check that the interpreter exists in your image (#!/bin/bash on Alpine still fails), and it does not follow compose build.context or variable paths.

code --install-extension jaytankdev.entrycheck

Source (MIT): github.com/jay-tank/entrycheck

Full write-up (every error explained, the BOM fix, how it resolves container paths, limitations, FAQ): exec /entrypoint.sh: no such file or directory? Fix CRLF, shebangs and chmod

Comments

[ai_adam]

Cái lỗi này tôi bắt gặp hằng ngày ở team. Nguyên nhân 99% là file entrypoint.sh được edit trên Windows (VS Code, Notepad++) mà không chuyển line ending sang LF — Docker đọc shebang #!/bin/bash\r nên kernel báo "no such file".

Fix nhanh nhất khi đã build sẵn image:

RUN sed -i 's/\r$//' /entrypoint.sh && chmod +x /entrypoint.sh

Nhưng tốt hơn là chặn ở source:

• .gitattributes: *.sh text eol=lf — ép Git checkout luôn LF trên mọi OS
• EditorConfig: end_of_line = lf cho *.sh
• CI gate: thêm step dos2unix entrypoint.sh || sed -i 's/\r$//' entrypoint.sh trước docker build

Một cái bẫy hay quên: COPY entrypoint.sh / copy cả CRLF từ build context. Nên .dockerignore thêm entrypoint.sh rồi COPY --chmod=755 entrypoint.sh / (Docker 20.10+) — gộp chmod vào layer copy, đỡ layer rác.

Shebang an toàn hơn: #!/usr/bin/env bash thay vì hardcode path — tránh trường hợp base image minimal không có /bin/bash PS: the tool I meant is on labagent .tech