In D27654#618081, @mjg wrote:

In D27654#618061, @christian.brauner_ubuntu.com wrote:

In D27654#618060, @mjg wrote:

The current close range code needs some cleanup regardless of this change. I think it would be most time efficient if exact semantics got documented and then someone from FreeBSD side implemented it (I'm happy to do it). Unless you really want to spend time on this, in which case I'm happy to review.

I'm happy to do it. I'm likely to contribute patches in the future so it'd be pretty helpful to get some experience writing patches for FreeBSD.

Glad to hear it. In that case I'll comment on the patch.

The current close_range suffers from a lot of avoidable filedesc relocking, I have a patch to fix it. Even then actual close drops the filedesc lock, so some degree of relock dance will still be there.

In contrast the feature you are implementing here does not need any of it. It can just take the filedesc lock upfront and have a for loop over the content. You can branch on it after the range got determined. iow in my opinion this just have a for loop doing fdeget_locked. I don't know the Linux code, it may be divorcing the 2 loops is very different from what was done there, but will very likely be the easiest way forward including any other future ops.

In D27654#618173, @mjg wrote:

The cleanup I mentioned is in as of r368732.

In D27654#618060, @mjg wrote:

The current close range code needs some cleanup regardless of this change. I think it would be most time efficient if exact semantics got documented and then someone from FreeBSD side implemented it (I'm happy to do it). Unless you really want to spend time on this, in which case I'm happy to review.