Ceph » CephFS

git cherry-pick -x https://github.com/ceph/ceph/pull/10386/commits/f3605d39e53b3ff777eb64538abfa62a5f98a4f2 which is part of https://github.com/ceph/ceph/pull/10386/commits conflicts with

<<<<<<< variant A
  r = _do_lookup(dir, dname, target, uid, gid);
>>>>>>> variant B
  r = _do_lookup(dir, dname, mask, target, uid, gid);
======= end

This is perfect, thank you !

Actually, you were right to ask, my question was about something else :-) It's good to know that the four commits are the one that matter, it may help. One of them, however, conflicts when cherry-picked on top of jewel with

git cherry-pick -x f3605d39e53b3ff777eb64538abfa62a5f98a4f2

with the conflict shown above. My question is about the best way to resolve this conflict. It could either be resolved manually or the commit(s) that create the conflict could be backported as well so that there is no more conflict. The problem with the later approach is that we may end up adding to jewel more than what is reasonable. My notion of "reasonable" in the context of fs is not to be relied on, hence my question :-)

You want the latter approach, and you want to pick them in the order they were originally committed, in case we need to bisect later. Just backporting f3605d39e53b3ff777eb64538abfa62a5f98a4f2 would be problematic, as that patch relies on the earlier changes in that series.

So to be clear, I'd cherry-pick (in this order):

a2ce16f8bfdb client: add mask parameter to _do_lookup
f3605d39e53b client: plumb a mask argument into _lookup
9e8476743eef client: only skip querying the MDS in _lookup when we have the necessary caps

...this commit is not strictly required to fix the problem with nlink, but it does fix a related problem. I'd suggest taking it as well:

14ee7bcbf0bd ceph: don't fudge the ctime in stat() unless it's really older than the mtime