Ceph » CephFS

O_SYMLINK sounds like BSD extension to me. We really want it?
Anyway, the Client::open() lacks support for O_NOFOLLOW. I am currently working on it.

Hmm, I don't remember where this came from but it might have been somebody cross-compiling for OS X. As marked, it's not terribly important but I'll merge in patches which fix it! :)

(I personally have a lot of interest in Ceph on OS X, and making the client usable on that platform is likely to extend CephFS' footprint quite a lot in the future.)

Hmm, just looking at the man page for open() aren't you supposed to perform an open on a symlink with O_NOFOLLOW if you also specify O_PATH?

Thank you for your review. You are right, on Linux the call should not fail with ELOOP if there was O_PATH present in the flags too. I am attaching the link to branch with support for O_PATH below, but currently only in Client::open. The full implementation should add checks based on Fh::mode field in many descriptor-accepting methods (like Client::read, Client::write, ...) to just return EBADF, I think. May we split the work into two tasks?

Link: https://github.com/rzarzynski/ceph/compare/wip-4920

I think it's probably best to merge it all as one pull request; there's nothing urgent here and that way we'll know we don't need to change any of the open logic again to make the other stuff return EBADF appropriately. :)

I added the support for O_PATH in several places as well as unit test. The commit is on my github. Feel free to comment.
Link: https://github.com/rzarzynski/ceph/compare/wip-4920

Looks good! Can you submit it as a PR against the main Ceph repo? :)

With pleasure. :-)
I corrected the typo in comment and squashed wip-4920 branch into two commits.
PR is available here: https://github.com/ceph/ceph/pull/3613