liyinshubyte

@Vicente-Cheng Thanks the reply. Where the entries of L2 are freed. > lru_run_lane would scan all entries and move the reclaimable item to L2.

@Vicente-Cheng Thanks the reply. I find only when [currentopen >= lru_state.fds_lowat](https://github.com/nfs-ganesha/nfs-ganesha/blob/next/src/FSAL/Stackable_FSALs/FSAL_MDCACHE/mdcache_lru.c#L1371), ganesha will release mdcache entries from L1 to L2. But in our ganesha, currentopen always be zero, so release...

> Are you only using NFSv4? If so, only COMMIT operations result in open_fd_count being incremented. @ffilz Yes, I'm using NFSv4.1, then I know why open_fd_count always be zero, do...

@ffilz @Vicente-Cheng @dang Just a friendly ping. > Yes, I'm using NFSv4.1, then I know why open_fd_count always be zero, do we need to fix this ?

@ffilz @dang @Vicente-Cheng Hi, I submit a change at: https://review.gerrithub.io/c/ffilz/nfs-ganesha/+/536998, we can continue to discuss the fix.

@ffilz Thanks for the explanation. As you said, when EXCLUSIVE create, client send OP_OPEN which need to set verifier, i.e. atime, ctime, mtime, then client send OP_SETATTR which set mode....

@dang Hi, thanks the explanation. But client does send SETATTR [here](https://github.com/torvalds/linux/blob/master/fs/nfs/nfs4proc.c#L3187). Even though the client crash before sending SETATTR, it also seems does not matter. Because, when client crash without...

> I don't see a 2nd SETATTR from the client (Fedora 34) using NFS v4.2. @ffilz Hi, following is my test, OPEN followed with SETATTR when touch a file, from...

@ffilz Thanks the reply. > What FSAL are you using? We use FSAL implemented by ourselves. > FATTR4_SUPPATTR_EXCLCREAT specified the attributes that can be set (and thus are not used...

> No example, actually we will have to add a call to FSAL API to fetch the attribute mask and then plumb up the protocol layer to pass it. @ffilz...