gitolfs3 - A very simple Git LFS server that stores stuff in S3-compatible backends

	Commit message (Collapse)	Author	Age	Files	Lines
*	Wheee	Rutger Broekhoff	2024-01-02	1	-7/+20
\|
*	Write Content-Length in object GET	Rutger Broekhoff	2024-01-02	1	-0/+3
\|
*	Token types, download verification	Rutger Broekhoff	2024-01-02	1	-20/+138
\|
*	Better logging (for a standalone server)	Rutger Broekhoff	2024-01-02	1	-7/+6
\|
*	🚫 No CGI 🚫	Rutger Broekhoff	2024-01-02	1	-2/+10
\|
*	catch panic	Rutger Broekhoff	2024-01-02	1	-1/+10
\|
*	Try generating more descriptive errors	Rutger Broekhoff	2024-01-02	1	-1/+10
\|
*	lol oops	Rutger Broekhoff	2024-01-02	1	-1/+1
\|
*	Uploading objects is a PUT, not a POST	Rutger Broekhoff	2024-01-02	1	-1/+1
\|
*	Fix upload regex	Rutger Broekhoff	2024-01-02	1	-1/+1
\|
*	Upload validation by proxying	Rutger Broekhoff	2024-01-02	1	-58/+263
\| \| \| \|	Yes, the code is a mess
*	Specify x-amz-content-sha256 as in Scaleway docs	Rutger Broekhoff	2023-12-31	1	-0/+1
\|
*	Try formatting x-amz-checksum-sha256 as Base64	Rutger Broekhoff	2023-12-31	1	-1/+15
\| \| \| \| \| \| \|	Although this already looks like a lost cause (Scaleway Object Storage doesn't seem to care about these headers -- certainly not about Content-Length -- I wanted to see if I could still get automatic checksum verification working.
*	Fix mistake	Rutger Broekhoff	2023-12-30	1	-4/+4
\|
*	Only print environment when dying	Rutger Broekhoff	2023-12-30	1	-5/+4
\|
*	Implement authorization in git-lfs-server, test presigned PUTs	Rutger Broekhoff	2023-12-30	1	-29/+177
\|
*	Implement git-lfs-authenticate	Rutger Broekhoff	2023-12-30	1	-25/+7
\|
*	URL as string, lesson learned	Rutger Broekhoff	2023-12-30	1	-2/+2
\|
*	Repo .git suffix	Rutger Broekhoff	2023-12-30	1	-1/+1
\|
*	Request IDs!	Rutger Broekhoff	2023-12-30	1	-20/+40
\|
*	Allow setting Gitolite binary path via env	Rutger Broekhoff	2023-12-29	1	-8/+14
\|
*	Log more	Rutger Broekhoff	2023-12-29	1	-0/+3
\|
*	Swap strings.TrimPrefix args	Rutger Broekhoff	2023-12-29	1	-2/+2
\|
*	Log reqPath 2x more	Rutger Broekhoff	2023-12-29	1	-0/+2
\|
*	Submatch	Rutger Broekhoff	2023-12-29	1	-5/+7
\|
*	Prefer PATH_INFO over request URL	Rutger Broekhoff	2023-12-29	1	-1/+5
\|
*	List envs at start	Rutger Broekhoff	2023-12-29	1	-11/+21
\|
*	Improve handling of MIME types	Rutger Broekhoff	2023-12-29	1	-5/+17
\|
*	Read S3 secrets from file	Rutger Broekhoff	2023-12-29	1	-13/+23
\|
*	Add 'lfs/objects' part to path	Rutger Broekhoff	2023-12-29	1	-1/+1
\|
*	Write basic read-only public Git LFS server	Rutger Broekhoff	2023-12-29	1	-0/+329
\| \| \| \| \| \| \| \| \| \| \| \| \| \| \| \| \| \| \| \| \| \| \| \| \| \|	The 'integration' with Gitolite is honestly pretty bad and should not be taken very seriously: it runs the 'gitolite access' command to check if some user (e.g., daemon/nobody) should be able to read from the repository. Based on this, it grants access to objects stored in S3, by generating Presigned GetObject URLs using the S3 API. Of course, this integration with Gitolite (especially when using the daemon user to check if the user should be able to read) is not very 'high-value': 1. If we already make use of the daemon pseudo-user to control access to public repositories, we may as well check for the existence of git-daemon-export-ok files. In case they exist, we simply assume that the repository is meant to be shown on the public internet and that therefore the LFS archive should also be considered 'open to the public'. 2. The way that Gitolite commands are currently run, this program breaks when not running under the git user without extra configuration; Gitolite decides where repositories are based on the HOME environment variable. This program currently does not set this. This could be set by the CGI server (or fcgiwrap) and would unbreak the system. There's no support for any more advanced kind of authn/authz. Uploading is also not supported yet. That's still to come.
*	Initial commit	Rutger Broekhoff	2023-12-29	1	-0/+1