Status

These pages describe fileshare v0.29.1.

What each release added

Verified

These are things a client that this project did not write was made to do, not assertions about the code.

• macOS mounts a share over SMB while curl is writing to the same image over WebDAV, and reads back what WebDAV wrote. Bob’s WebDAV write to a share he may only read is 403; a wrong password is 401.
• go-smb2 — a client this project did not write — drives twenty concurrent reads through SMB while twenty run through WebDAV, under -race, in CI.
• The access rules are checked through every protocol rather than in the configuration alone: alice writes and bob does not, over SMB and over WebDAV, and NFS is not offered the share at all.
• SFTP certificates are verified against OpenSSH’s own client, which also refuses the same key once its certificate is moved aside.
• NFS identities from certificates were measured with a real Linux client (kernel 6.17, ktls-utils 0.9) in go-filesystems/nfs’s CI — which is how MNT in the clear and tlshd’s pitfalls were found.
• The KRL’s unsigned form is a measurement: OpenSSH 9.6 and 10.3 read a signed KRL and skip the signature (go-authn/krl, against both).
• go-authn/bridge’s interop lane (#12, #17) judges fileshare v0.13.0 against the real bridge: the KRL, the CRL over NFS, WebDAV tokens revoked over shared signals, an IdP disabled by scope, and opkssh over shared signals.

Not yet

Writes over S3. PUT and DELETE answer 403. The library can write, but a share a person may only read has to refuse at the same place SFTP does, and that is not wired. Refusing beats a half-written object.

An OIDC token over S3. Tokens work over WebDAV and nowhere else. The usual route for S3 is STS AssumeRoleWithWebIdentity, which exchanges the token for temporary credentials the client then signs with — a different mechanism from accepting a bearer token, and not implemented.

--isolate with an admin or a metrics block. Refused: there is no one process an API change could be applied to. See one process per protocol.

A person per user on a shared Linux NFS client, from a certificate. A Linux client attaches the certificate to a mount, so every user of the mount is the person it names. That is a property of the client, and the answer there is sec=krb5. See what it cannot promise.

S3 itself has shipped: a share is a bucket, served over the same per-user tree SFTP uses.

Measured, and stated as measured

Claims in these pages that are insurance or measurement rather than reproductions of a defect, and are written that way in the source too:

• Eight goroutines writing through an unwrapped fat32 driver under -race produce no race today. The shared lock is insurance on a contract — go-filesystems/interface promises nothing about concurrent path-based calls — not a fix for an observed failure.
• The build-tag sizes were measured, including the 13.2 MB that hashicorp/go-plugin costs on its own, which is what decided against a plugin framework — and the 11.7 MB gRPC costs now that the admin API brings it, which is why nogrpc exists.
• A group-tag certificate extension of its own was designed and measured against Go’s X.509 parser, which refused the whole certificate; the groups travel as tag URIs instead.