search

mpartel / bindfs

Mount a directory elsewhere with changed permissions.
https://bindfs.org/
GNU General Public License v2.0
449 stars 64 forks source link

Support for negative offset (?) #142

Closed mathiasaerts closed 1 year ago

mathiasaerts commented 1 year ago

I'm trying to use bindfs to 'unshift' an existing directory that has a uid and gid shift of 100000 applied (actual file ownership, not using bindfs). When using a negative offset of -100000 (using mount options create-as-user,uid-offset=-100000,gid-offset=-100000), it seems that the mounted directory indeed shows the expected unshifted file ownership. However, when trying to create a file, it doesn't work, and it appears that file operations do not take the actual user performing the operation into account, resulting in errors such as touch: setting times of ‘filename’: Permission denied and both file owner and group being set to 4294867296, while on the source directory, the owner appears as root instead.

When trying the same thing with a positive offset, everything appears to be working as expected, and the uid and gid of the user performing the file operation are preserved. Therefore, I was wondering if negative offsets are actually supported. In any case, it could be a very nice feature, but it doesn't seem to work as expected now.

mpartel commented 1 year ago

You are right, I hadn't considered this case (and it turns out the unsigned integer parsing function strtoul silently casts negative numbers instead of erroring :unamused: )

Shouldn't be a difficult fix. I'll see if I can get it done today or tomorrow.

mpartel commented 1 year ago

Should be fixed now with 1.17.5.

Note that with uid-offset=-100000, operations on any source files with uid/gid below -100000 (e.g. owned by root) will fail with EIO. Is this acceptable in your case? If not, do you have a suggestion for how it should work (or be workaroundable)?

mathiasaerts commented 1 year ago

Amazing, thanks! :star_struck: First tests look very promising!

I don't think this would be an issue for my current use case. I'm wondering if as a workaround you could set lower uids / gids to 0, so all files with a uid below 100000 would become root in the mounted dir. I'm guessing this could be a problem to 'unapply' the offset though?

Instead of making the files root, it could also be nobody (uid/gid 65534), which is also how LXC handles files owned by a uid < 100000 on the host in an unprivileged container. In LXC, this allows files to be readonly in the container, so they cannot be modified or chowned. I guess you would have to handle each of those operations separately (if this is not already the case) to implement a similar approach?