Description of problem: When a readdirp was executed, the nlookup count for each inode of the returned entries was incremented. However the kernel does not increment the counter for '.' and '..' entries. This caused kernel to send forgets with a counter smaller than the inode's current value. This prevented these inodes to be retired when ref count was 0. This behavior needs fixing.
Discussed with Dev. and found that there is no clear way of validating this bug as this is a code change with no direct functional impact. However moving it to verified as discussed with differnt stakeholders However, following was what i did test: tried to check the lookups using profiling and didn't find difference b/w fuse client versions of 3.7.9-1(before fix) and 3.7.9-3(after fix) (the server versions for both were common 3.7.9-3) However dev. mentioned that this may not be right test Did a lot of IO and checked statedump before clearing client cache and after clearing client cache( searched for IO the hot-count in inode table ) didnt see much difference b/w the different client version However I didnt see any impact on IOs or any regressions and hence positive of this fix
Since the problem described in this bug report should be resolved in a recent advisory, it has been closed with a resolution of ERRATA. For information on the advisory, and where to find the updated files, follow the link below. If the solution does not work for you, open a new bug report. https://access.redhat.com/errata/RHBA-2016:1240