Hi Trond, I'd like to discuss a couple of commits that are dealing with GETATTRs. First is the recent fix in: commit 3a39e778690500066b31fe982d18e2e394d3bce2 "nfs: set invalid blocks after NFSv4 writes". The fact that CLOSE's GETATTR doesn't query for available space leads to stale attribute data but marking attributes invalid instead leads to an additional standalone GETATTR if attributes are after writing to the file. Question: why not change CLOSE's GETATTR to query for additional attributes instead? Another commit I would like to inquire about is: commit 3ecefc9295991eaaad4c67915c6384e5d18cc632 "NFSv4: Don't request close-to-open attribute when holding a delegation". After this commit if the client application queries for attributes after writing it triggers a gettattr (because we didn't get them in the close compound). I understand that it was an optimization but it leads to an extra RPC in some workloads so how can we claim that one saving is better than another? Thank you for the feedback.
