Where Is Dirty Data Flushed During NFSv4 Delegation Recall?

Where Is Dirty Data Flushed During NFSv4 Delegation Recall?

[Zhou Jifeng]

Hello everyone, I have a question that has been bothering me for a long
 time. I would like to ask a familiar friend to help explain it. In the NFSv4
 protocol, it is clearly defined that during the delegation RECALL 
processing, the client needs to write the modified cache to the server
side. However, throughout the process, I have not found the code for
flushing dirty data in the kernel NFS client. I have repeatedly reviewed
the code logic of nfs4_callback_recall and nfs4_state_manager, but still
cannot understand where the dirty data is flushed.

My question is: When the client is handling the RECALL operation, where
does the process of flushing the dirty data take place?

Best regards

Re: Where Is Dirty Data Flushed During NFSv4 Delegation Recall?

[Trond Myklebust]

On Thu, 2025-09-11 at 10:30 +0800, Zhou Jifeng wrote:
> Hello everyone, I have a question that has been bothering me for a
> long
>  time. I would like to ask a familiar friend to help explain it. In
> the NFSv4
>  protocol, it is clearly defined that during the delegation RECALL 
> processing, the client needs to write the modified cache to the
> server
> side. However, throughout the process, I have not found the code for
> flushing dirty data in the kernel NFS client. I have repeatedly
> reviewed
> the code logic of nfs4_callback_recall and nfs4_state_manager, but
> still
> cannot understand where the dirty data is flushed.
> 
> My question is: When the client is handling the RECALL operation,
> where
> does the process of flushing the dirty data take place?
> 

The question of when to flush dirty data isn't determined by the NFSv4
RECALL operation spec. It is decided by the client caching model and by
POSIX.
Please read the "nfs(5)" manpage and the section on close-to-open cache
consistency.

-- 
Trond Myklebust
Linux NFS client maintainer, Hammerspace
trondmy@kernel.org, trond.myklebust@hammerspace.com

Re: Where Is Dirty Data Flushed During NFSv4 Delegation Recall?

[Zhou Jifeng]

> > On Thu, 2025-09-11 at 10:30 +0800, Zhou Jifeng wrote:
> > Hello everyone, I have a question that has been bothering me for a
> > long
> >  time. I would like to ask a familiar friend to help explain it. In
> > the NFSv4
> >  protocol, it is clearly defined that during the delegation RECALL
> > processing, the client needs to write the modified cache to the
> > server
> > side. However, throughout the process, I have not found the code for
> > flushing dirty data in the kernel NFS client. I have repeatedly
> > reviewed
> > the code logic of nfs4_callback_recall and nfs4_state_manager, but
> > still
> > cannot understand where the dirty data is flushed.
> >
> > My question is: When the client is handling the RECALL operation,
> > where
> > does the process of flushing the dirty data take place?
> >
> 
> The question of when to flush dirty data isn't determined by the NFSv4
> RECALL operation spec. It is decided by the client caching model and by
> POSIX.
> Please read the "nfs(5)" manpage and the section on close-to-open cache
> consistency.
> 
-> -
> Trond Myklebust
> Linux NFS client maintainer, Hammerspace
> trondmy@kernel.org, trond.myklebust@hammerspace.com

Hi Trond,

The nfs(5) manual page and the close-to-open consistency you mentioned
have been of great help to my understanding. Thank you very much.

Best regards
Zhou Jifeng