- apply_truncate needs to truncate page cache, or something?
kernel client
+- flush caps on sync, fsync, etc.
+ - hmm, should these go in a per-session "check" list, instead of independently scheduling delayed_work for each inode?
+- fsync should flush cap file size to mds. not sure how to block on that, however.. maybe a want_reply flag in the cap msg?
- fix session close seq interaction with mds (sessions don't currently close all the way)
- file_data_version stuff!
- deal with CAP_RDCACHE properly: invalidate cache pages?
-- trim expired leases so we don't indefinitely hold dcache/icache refs...
- what happens after reconnect when we get successful reply but no trace (!) on e.g. rename, unlink, link, open+O_CREAT, etc...
- procfs/debugfs
- adjust granular debug levels too