Can the PNFS blocklayout of the Linux nfsd server be used in a production environment?

Can the PNFS blocklayout of the Linux nfsd server be used in a production environment?

[Zhou Jifeng]

Hello everyone, I learned through ChatGPT that the PNFS blocklayout of Linux 
nfsd cannot be used for production environment deployment. However, I saw 
a technical sharing conference video on YouTube titled "SNIA SDC 2024 - The 
Linux NFS Server in 2024" where it was mentioned that the PNFS blocklayout 
of nfsd has been fully maintained, which is contrary to the result given by 
ChatGPT. My question is: Can the PNFS blocklayout of nfsd be used for 
production environment deployment? If yes, from which kernel version can it 
be used in the production environment? 

Best regards

Re: Can the PNFS blocklayout of the Linux nfsd server be used in a production environment?

[Chuck Lever]

On Wed, Nov 26, 2025, at 9:14 PM, Zhou Jifeng wrote:
> Hello everyone, I learned through ChatGPT that the PNFS blocklayout of Linux 
> nfsd cannot be used for production environment deployment. However, I saw 
> a technical sharing conference video on YouTube titled "SNIA SDC 2024 - The 
> Linux NFS Server in 2024" where it was mentioned that the PNFS blocklayout 
> of nfsd has been fully maintained, which is contrary to the result given by 
> ChatGPT.
> 
> My question is: Can the PNFS blocklayout of nfsd be used for 
> production environment deployment? If yes, from which kernel version can it 
> be used in the production environment? 

Responding as the presenter of the SNIA SDC talk:

There's a difference between "maintained" and "can be deployed in a
production environment". "Maintained" means there are developers
who are active and can help with bugs and new features. "Production
ready" means you can trust it with significant workloads.

The pNFS block layout type has several subtypes. Pure block, iSCSI,
SCSI, and NVMe.

Pure block has some protocol deficiencies that prevent the ability
of the MDS to fence pNFS clients, so it might never be production
ready.

NFSD's pNFS block layout with iSCSI or SCSI works well in recent
kernels, but has some quirks that still need to be addressed. You
will have to test it with your favorite workloads to see whether
it will work for you.

NFSD's pNFS block layout with NVMe implementation is fresh from the
factory (I think the implementation went into v6.12?). As with
iSCSI/SCSI, try it and see if it works well enough for you.


-- 
Chuck Lever

Re: Can the PNFS blocklayout of the Linux nfsd server be used in a production environment?

[Zhou Jifeng]

>>On Wed, Nov 26, 2025, at 9:14 PM, Zhou Jifeng wrote:
>> Hello everyone, I learned through ChatGPT that the PNFS blocklayout of Linux
>> nfsd cannot be used for production environment deployment. However, I saw
>> a technical sharing conference video on YouTube titled "SNIA SDC 2024 - The
>> Linux NFS Server in 2024" where it was mentioned that the PNFS blocklayout
>> of nfsd has been fully maintained, which is contrary to the result given by
>> ChatGPT.
>>
>> My question is: Can the PNFS blocklayout of nfsd be used for
>> production environment deployment? If yes, from which kernel version can it
>> be used in the production environment?
>
> Responding as the presenter of the SNIA SDC talk:
> 
> There's a difference between "maintained" and "can be deployed in a
> production environment". "Maintained" means there are developers
> who are active and can help with bugs and new features. "Production
> ready" means you can trust it with significant workloads.
> 
> The pNFS block layout type has several subtypes. Pure block, iSCSI,
> SCSI, and NVMe.
> 
> Pure block has some protocol deficiencies that prevent the ability
> of the MDS to fence pNFS clients, so it might never be production
> ready.
> 
> NFSD's pNFS block layout with iSCSI or SCSI works well in recent
> kernels, but has some quirks that still need to be addressed. You
> will have to test it with your favorite workloads to see whether
> it will work for you.
> 
> NFSD's pNFS block layout with NVMe implementation is fresh from the
> factory (I think the implementation went into v6.12?). As with
> iSCSI/SCSI, try it and see if it works well enough for you.
> 
> 
> --
> Chuck Lever

Hi Chuck,

I'm truly honored to receive your personal response, which has been 
extremely helpful to me. I've searched extensively online for methods 
to deploy pNFS clusters with nfsd, but most resources are incomplete. 
The documentation at 
"https://docs.kernel.org/admin-guide/nfs/pnfs-block-server.html" also 
provides very limited information. Are there any official, more 
comprehensive deployment guides for pNFS available.

Best regards,
ZhouJifeng

Re: Can the PNFS blocklayout of the Linux nfsd server be used in a production environment?

[Chuck Lever]

On Fri, Nov 28, 2025, at 3:12 AM, Zhou Jifeng wrote:
> I'm truly honored to receive your personal response, which has been 
> extremely helpful to me. I've searched extensively online for methods 
> to deploy pNFS clusters with nfsd, but most resources are incomplete. 
> The documentation at 
> "https://docs.kernel.org/admin-guide/nfs/pnfs-block-server.html" also 
> provides very limited information. Are there any official, more 
> comprehensive deployment guides for pNFS available.

I'm afraid there is not much official documentation at this time.
My colleague Dai has published an introduction and brief set-up
guide:

https://blogs.oracle.com/linux/parallelize-nfs-with-pnfs


-- 
Chuck Lever

LAYOUT4_NFSV4_1_FILES supported? Re: Can the PNFS blocklayout of the Linux nfsd server be used in a production environment?

[Sebastian Feld]

On Thu, Nov 27, 2025 at 5:41 PM Chuck Lever <cel@kernel.org> wrote:
>
> On Wed, Nov 26, 2025, at 9:14 PM, Zhou Jifeng wrote:
> > Hello everyone, I learned through ChatGPT that the PNFS blocklayout of Linux
> > nfsd cannot be used for production environment deployment. However, I saw
> > a technical sharing conference video on YouTube titled "SNIA SDC 2024 - The
> > Linux NFS Server in 2024" where it was mentioned that the PNFS blocklayout
> > of nfsd has been fully maintained, which is contrary to the result given by
> > ChatGPT.
> >
> > My question is: Can the PNFS blocklayout of nfsd be used for
> > production environment deployment? If yes, from which kernel version can it
> > be used in the production environment?
>
> Responding as the presenter of the SNIA SDC talk:
>
> There's a difference between "maintained" and "can be deployed in a
> production environment". "Maintained" means there are developers
> who are active and can help with bugs and new features. "Production
> ready" means you can trust it with significant workloads.
>
> The pNFS block layout type has several subtypes. Pure block, iSCSI,
> SCSI, and NVMe.

What about the LAYOUT4_NFSV4_1_FILES layout type? Is that still supported?

Sebi
-- 
Sebastian Feld - IT security consultant

Re: LAYOUT4_NFSV4_1_FILES supported? Re: Can the PNFS blocklayout of the Linux nfsd server be used in a production environment?

[Chuck Lever]

On Fri, Nov 28, 2025, at 3:57 AM, Sebastian Feld wrote:
> On Thu, Nov 27, 2025 at 5:41 PM Chuck Lever <cel@kernel.org> wrote:
>>
>> On Wed, Nov 26, 2025, at 9:14 PM, Zhou Jifeng wrote:
>> > Hello everyone, I learned through ChatGPT that the PNFS blocklayout of Linux
>> > nfsd cannot be used for production environment deployment. However, I saw
>> > a technical sharing conference video on YouTube titled "SNIA SDC 2024 - The
>> > Linux NFS Server in 2024" where it was mentioned that the PNFS blocklayout
>> > of nfsd has been fully maintained, which is contrary to the result given by
>> > ChatGPT.
>> >
>> > My question is: Can the PNFS blocklayout of nfsd be used for
>> > production environment deployment? If yes, from which kernel version can it
>> > be used in the production environment?
>>
>> Responding as the presenter of the SNIA SDC talk:
>>
>> There's a difference between "maintained" and "can be deployed in a
>> production environment". "Maintained" means there are developers
>> who are active and can help with bugs and new features. "Production
>> ready" means you can trust it with significant workloads.
>>
>> The pNFS block layout type has several subtypes. Pure block, iSCSI,
>> SCSI, and NVMe.
>
> What about the LAYOUT4_NFSV4_1_FILES layout type? Is that still supported?

Above, we're talking about the Linux NFS server... IIRC NFSD never
supported the NFSv4.1 file layout type. It has a simple experimental
implementation of the flexfile layout type (the MDS and DS are the
same server).

The Linux NFS client fully implements the file, flexfile, and all
the block layout types.


-- 
Chuck Lever

Re: LAYOUT4_NFSV4_1_FILES supported? Re: Can the PNFS blocklayout of the Linux nfsd server be used in a production environment?

[Rick Macklem]

On Fri, Nov 28, 2025 at 8:24 AM Chuck Lever <cel@kernel.org> wrote:
>
>
>
> On Fri, Nov 28, 2025, at 3:57 AM, Sebastian Feld wrote:
> > On Thu, Nov 27, 2025 at 5:41 PM Chuck Lever <cel@kernel.org> wrote:
> >>
> >> On Wed, Nov 26, 2025, at 9:14 PM, Zhou Jifeng wrote:
> >> > Hello everyone, I learned through ChatGPT that the PNFS blocklayout of Linux
> >> > nfsd cannot be used for production environment deployment. However, I saw
> >> > a technical sharing conference video on YouTube titled "SNIA SDC 2024 - The
> >> > Linux NFS Server in 2024" where it was mentioned that the PNFS blocklayout
> >> > of nfsd has been fully maintained, which is contrary to the result given by
> >> > ChatGPT.
> >> >
> >> > My question is: Can the PNFS blocklayout of nfsd be used for
> >> > production environment deployment? If yes, from which kernel version can it
> >> > be used in the production environment?
> >>
> >> Responding as the presenter of the SNIA SDC talk:
> >>
> >> There's a difference between "maintained" and "can be deployed in a
> >> production environment". "Maintained" means there are developers
> >> who are active and can help with bugs and new features. "Production
> >> ready" means you can trust it with significant workloads.
> >>
> >> The pNFS block layout type has several subtypes. Pure block, iSCSI,
> >> SCSI, and NVMe.
> >
> > What about the LAYOUT4_NFSV4_1_FILES layout type? Is that still supported?
>
> Above, we're talking about the Linux NFS server... IIRC NFSD never
> supported the NFSv4.1 file layout type. It has a simple experimental
> implementation of the flexfile layout type (the MDS and DS are the
> same server).
Once upon a time Benny Halevy had a single server (in the Linux
kernel) that did file layout. (I remember because that is what I used
for testing the FreeBSD client side.) It even knew how to do striping.

However, it was a "single server" (MDS and DS in the same Linux kernel nfsd),
so it was only useful for testing purposes.

I have no idea if it is still around somewhere, but unless someone wants
to do a lot of work on it, it is definitely not useful for production sites.

rick

>
> The Linux NFS client fully implements the file, flexfile, and all
> the block layout types.
>
>
> --
> Chuck Lever
>

Re: LAYOUT4_NFSV4_1_FILES supported? Re: Can the PNFS blocklayout of the Linux nfsd server be used in a production environment?

[Chuck Lever]

On Fri, Nov 28, 2025, at 6:16 PM, Rick Macklem wrote:
> On Fri, Nov 28, 2025 at 8:24 AM Chuck Lever <cel@kernel.org> wrote:
>>
>>
>>
>> On Fri, Nov 28, 2025, at 3:57 AM, Sebastian Feld wrote:
>> > On Thu, Nov 27, 2025 at 5:41 PM Chuck Lever <cel@kernel.org> wrote:
>> >>
>> >> On Wed, Nov 26, 2025, at 9:14 PM, Zhou Jifeng wrote:
>> >> > Hello everyone, I learned through ChatGPT that the PNFS blocklayout of Linux
>> >> > nfsd cannot be used for production environment deployment. However, I saw
>> >> > a technical sharing conference video on YouTube titled "SNIA SDC 2024 - The
>> >> > Linux NFS Server in 2024" where it was mentioned that the PNFS blocklayout
>> >> > of nfsd has been fully maintained, which is contrary to the result given by
>> >> > ChatGPT.
>> >> >
>> >> > My question is: Can the PNFS blocklayout of nfsd be used for
>> >> > production environment deployment? If yes, from which kernel version can it
>> >> > be used in the production environment?
>> >>
>> >> Responding as the presenter of the SNIA SDC talk:
>> >>
>> >> There's a difference between "maintained" and "can be deployed in a
>> >> production environment". "Maintained" means there are developers
>> >> who are active and can help with bugs and new features. "Production
>> >> ready" means you can trust it with significant workloads.
>> >>
>> >> The pNFS block layout type has several subtypes. Pure block, iSCSI,
>> >> SCSI, and NVMe.
>> >
>> > What about the LAYOUT4_NFSV4_1_FILES layout type? Is that still supported?
>>
>> Above, we're talking about the Linux NFS server... IIRC NFSD never
>> supported the NFSv4.1 file layout type. It has a simple experimental
>> implementation of the flexfile layout type (the MDS and DS are the
>> same server).
> Once upon a time Benny Halevy had a single server (in the Linux
> kernel) that did file layout. (I remember because that is what I used
> for testing the FreeBSD client side.) It even knew how to do striping.
>
> However, it was a "single server" (MDS and DS in the same Linux kernel nfsd),
> so it was only useful for testing purposes.
>
> I have no idea if it is still around somewhere, but unless someone wants
> to do a lot of work on it, it is definitely not useful for production sites.

I am going to guess that was Dan Muntz's prototype. AFAIK that was
never merged upstream, but remains available in archive form at
git.linux-nfs.org.

NFSD continues to implement only the block-related layout types,
plus flexfile (in toy form).


-- 
Chuck Lever

Re: Can the PNFS blocklayout of the Linux nfsd server be used in a production environment?

[Christoph Hellwig]

As the author of much of the block* layout code I agree with Chuck, just
want to throw in a little extra note:

On Thu, Nov 27, 2025 at 11:40:50AM -0500, Chuck Lever wrote:
> NFSD's pNFS block layout with NVMe implementation is fresh from the
> factory (I think the implementation went into v6.12?). As with
> iSCSI/SCSI, try it and see if it works well enough for you.

So while the NVMe layout support is indeed very new, it's also very
little code, on the server side the support is entirely contained in
the nvme driver by implementing the get_unique_id method (~60 lines of
code) and on the client side it is looking for the nvme persistent
devices names (2 Lines of code after a small preparatory refactoring).

So in general I think it should be taken as the same maturity as the
SCSI layout at this point.

Re: Can the PNFS blocklayout of the Linux nfsd server be used in a production environment?

[Dai Ngo]

On 12/1/25 10:09 PM, Christoph Hellwig wrote:
> As the author of much of the block* layout code I agree with Chuck, just
> want to throw in a little extra note:
>
> On Thu, Nov 27, 2025 at 11:40:50AM -0500, Chuck Lever wrote:
>> NFSD's pNFS block layout with NVMe implementation is fresh from the
>> factory (I think the implementation went into v6.12?). As with
>> iSCSI/SCSI, try it and see if it works well enough for you.
> So while the NVMe layout support is indeed very new, it's also very
> little code, on the server side the support is entirely contained in
> the nvme driver by implementing the get_unique_id method (~60 lines of
> code) and on the client side it is looking for the nvme persistent
> devices names (2 Lines of code after a small preparatory refactoring).
>
> So in general I think it should be taken as the same maturity as the
> SCSI layout at this point.

How does server fence a client using NVMe layout?

-Dai

Re: Can the PNFS blocklayout of the Linux nfsd server be used in a production environment?

[Christoph Hellwig]

On Thu, Dec 04, 2025 at 10:10:39AM -0800, Dai Ngo wrote:
> > So in general I think it should be taken as the same maturity as the
> > SCSI layout at this point.
> 
> How does server fence a client using NVMe layout?

Using the same persistent reservation mechanism as the SCSI layout.
The very minor difference in the registration scope are transparently
handled by the drivers.

Re: Can the PNFS blocklayout of the Linux nfsd server be used in a production environment?

[Dai Ngo]

On 12/5/25 12:05 AM, Christoph Hellwig wrote:
> On Thu, Dec 04, 2025 at 10:10:39AM -0800, Dai Ngo wrote:
>>> So in general I think it should be taken as the same maturity as the
>>> SCSI layout at this point.
>> How does server fence a client using NVMe layout?
> Using the same persistent reservation mechanism as the SCSI layout.
> The very minor difference in the registration scope are transparently
> handled by the drivers.

Thanks, I'll give it a try.

-Dai