[Gluster-devel] a link issue maybe introduced in a bug fix " Don't let NFS cache stat after writes"

[Shyam Ranganathan]

On 01/10/2018 07:07 AM, Pranith Kumar Karampuri wrote:
> When stat is "zero filled", understanding is that the higher layer
> protocol doesn't send stat value to the kernel and a separate lookup is
> sent by the kernel to get the latest stat value. In which protocol are
> you seeing this issue? Fuse/NFS/SMB?

<orthogonal to the current problem>

I was unaware of this assumption. I had been poking around some code for
RIO as we return only partial iatt in some cases and wanted to check the
stack to see what the assumptions are. What I found then (unfortunately
no ready reference is available), is that such an assumption was not
present.

Going forward though, the new iatt implementation (motivated by the
kernel statx implementation) should make this explicit with it's
ia_flags in place (see [1]).

I hope we adopt this one sooner, so that ambiguity on certain field
values being 0/NULL or not present is not a deterministic factor in
trusting the iatt or not.

Shyam

[1] new iatt in 4.0 version of the protocol:
https://review.gluster.org/#/c/18850/6/libglusterfs/src/iatt.h