Xavi Hernandez

/run regression

/run s390-regression

@black-dragon74 this patch was accepted and only needed to pass the regression. I think this PR can be useful.

/run regression

> Hi @xhernandez , I am interested in this issue, but got the following result after runnig, is that c program related to certain os or env? It's a very...

@Ratio2 the problem here is not so much the performance but the correct behavior. The current implementation of cache invalidation is not strict enough to allow caching data/metadata for long...

I've been trying to find a good way to correctly invalidate inodes through fuse to guarantee that cached data is gone. This is necessary to enforce consistency and make sure...

I've been able to reproduce this issue. I'll investigate it.

The problem happens because of a deadlock between rename and unlink operations. A [patch](https://review.gluster.org/#/c/glusterfs/+/20025/) was introduced to prevent already locked inodes to be deleted. This was necessary to make sure...

> > Creating the hard link and deleting the file also require taking entry locks in AFR. So, if we are doing a mv /x /y, the sequence of fops...