I'd like to request some time for those interested specifically
in NFSD to gather and discuss some topics. Not a network file
system free-for-all, but specifically for NFSD, because there
is a long list of potential topics:
• Progress on using iomap for NFSD READ/READ_PLUS (anna)
• Replacing nfsd_splice_actor (all)
• Transition from page arrays to bvecs (dhowells, hch)
• tmpfs directory cookie stability (cel)
• timestamp resolution and i_version (jlayton)
• GSS Kerberos futures (dhowells)
• NFS/NFSD CI (jlayton)
• NFSD POSIX to NFSv4 ACL translation - writing down the rules (all)
Some of these topics might be appealing to others not specifically
involved with NFSD development. If there's something that should
be moved to another track or session, please pipe up.
--
Chuck Lever
On Wed, Apr 12, 2023 at 06:27:07PM +0000, Chuck Lever III wrote:
> I'd like to request some time for those interested specifically
> in NFSD to gather and discuss some topics. Not a network file
> system free-for-all, but specifically for NFSD, because there
> is a long list of potential topics:
>
> • Progress on using iomap for NFSD READ/READ_PLUS (anna)
> • Replacing nfsd_splice_actor (all)
> • Transition from page arrays to bvecs (dhowells, hch)
> • tmpfs directory cookie stability (cel)
> • timestamp resolution and i_version (jlayton)
I'd attend this one.
> • GSS Kerberos futures (dhowells)
> • NFS/NFSD CI (jlayton)
> • NFSD POSIX to NFSv4 ACL translation - writing down the rules (all)
I have some experience dealing with ACLs so I'm happy to attend just in
case I may be useful.
On Thu, 2023-04-13 at 11:33 +0200, Christian Brauner wrote:
> On Wed, Apr 12, 2023 at 06:27:07PM +0000, Chuck Lever III wrote:
> > I'd like to request some time for those interested specifically
> > in NFSD to gather and discuss some topics. Not a network file
> > system free-for-all, but specifically for NFSD, because there
> > is a long list of potential topics:
> >
> > • Progress on using iomap for NFSD READ/READ_PLUS (anna)
> > • Replacing nfsd_splice_actor (all)
> > • Transition from page arrays to bvecs (dhowells, hch)
> > • tmpfs directory cookie stability (cel)
> > • timestamp resolution and i_version (jlayton)
>
> I'd attend this one.
>
I wonder if we ought to propose a separate FS track spot for this? I
sort of expect some lively discussion, and this may be of more interest
than just nfsd folks.
> > • GSS Kerberos futures (dhowells)
> > • NFS/NFSD CI (jlayton)
> > • NFSD POSIX to NFSv4 ACL translation - writing down the rules (all)
>
> I have some experience dealing with ACLs so I'm happy to attend just in
> case I may be useful.
That would be helpful! I'll note that there was a draft RFC for this
many years ago, but it expired:
https://datatracker.ietf.org/doc/html/draft-ietf-nfsv4-acl-mapping-05
I think most of the rules are laid out there, but there are some areas
where things just don't work right.
A more radical idea:
I wonder if we could get any traction at the IETF on a POSIX ACL
extension for NFSv4? Basically, we could resurrect the old v3 nfsacl
protocol as new operations for v4, and allow the client and server to
negotiate on using them.
Given that almost all the clients and servers in operation on the planet
have to translate these, it makes some sense to avoid the translation
when we can.
--
Jeff Layton <[email protected]>
On Wed, Apr 12, 2023 at 06:27:07PM +0000, Chuck Lever III wrote:
> I'd like to request some time for those interested specifically
> in NFSD to gather and discuss some topics. Not a network file
> system free-for-all, but specifically for NFSD, because there
> is a long list of potential topics:
>
> • Progress on using iomap for NFSD READ/READ_PLUS (anna)
> • Replacing nfsd_splice_actor (all)
> • Transition from page arrays to bvecs (dhowells, hch)
- Using larger folios instead of single pages; maybe this is the same
discussion.
> • tmpfs directory cookie stability (cel)
> • timestamp resolution and i_version (jlayton)
> • GSS Kerberos futures (dhowells)
> • NFS/NFSD CI (jlayton)
> • NFSD POSIX to NFSv4 ACL translation - writing down the rules (all)
>
> Some of these topics might be appealing to others not specifically
> involved with NFSD development. If there's something that should
> be moved to another track or session, please pipe up.
>
> --
> Chuck Lever
>
>
> On Apr 13, 2023, at 10:05 AM, Matthew Wilcox <[email protected]> wrote:
>
> On Wed, Apr 12, 2023 at 06:27:07PM +0000, Chuck Lever III wrote:
>> I'd like to request some time for those interested specifically
>> in NFSD to gather and discuss some topics. Not a network file
>> system free-for-all, but specifically for NFSD, because there
>> is a long list of potential topics:
>>
>> • Progress on using iomap for NFSD READ/READ_PLUS (anna)
>> • Replacing nfsd_splice_actor (all)
>> • Transition from page arrays to bvecs (dhowells, hch)
>
> - Using larger folios instead of single pages; maybe this is the same
> discussion.
Though quite pertinent to NFSD, I don't think it is the same
discussion.
The page array in question is xdr_buf::pages, which today is
used to build an RPC message and then pass it to the network
layer. bvecs would be the shiny new way to pass those messages to
the network layer. This discussion is closely related to the
iterator work that David has embarked upon.
Last time I brought up the use of folios, you told me that folios
don't do well when a large folio is broken up into pages before
the pages are returned to the allocator. We absolutely can
discuss where folios can play with NFSD and how that changes
NFSD's interaction with the page cache and filesystems. That
might be more akin to the "replacing splice_actor" discussion?
--
Chuck Lever
> On Apr 13, 2023, at 9:09 AM, Jeff Layton <[email protected]> wrote:
>
> On Thu, 2023-04-13 at 11:33 +0200, Christian Brauner wrote:
>> On Wed, Apr 12, 2023 at 06:27:07PM +0000, Chuck Lever III wrote:
>>> I'd like to request some time for those interested specifically
>>> in NFSD to gather and discuss some topics. Not a network file
>>> system free-for-all, but specifically for NFSD, because there
>>> is a long list of potential topics:
>>>
>>> • Progress on using iomap for NFSD READ/READ_PLUS (anna)
>>> • Replacing nfsd_splice_actor (all)
>>> • Transition from page arrays to bvecs (dhowells, hch)
>>> • tmpfs directory cookie stability (cel)
>>> • timestamp resolution and i_version (jlayton)
>>
>> I'd attend this one.
>>
>
> I wonder if we ought to propose a separate FS track spot for this? I
> sort of expect some lively discussion, and this may be of more interest
> than just nfsd folks.
Fine with me, that means more time for both this topic and
for the other topics on the NFSD list.
--
Chuck Lever
> On Apr 13, 2023, at 10:05, Matthew Wilcox <[email protected]> wrote:
>
> On Wed, Apr 12, 2023 at 06:27:07PM +0000, Chuck Lever III wrote:
>> I'd like to request some time for those interested specifically
>> in NFSD to gather and discuss some topics. Not a network file
>> system free-for-all, but specifically for NFSD, because there
>> is a long list of potential topics:
>>
>> • Progress on using iomap for NFSD READ/READ_PLUS (anna)
>> • Replacing nfsd_splice_actor (all)
>> • Transition from page arrays to bvecs (dhowells, hch)
>
> - Using larger folios instead of single pages; maybe this is the same
> discussion.
That’s not doable until you somehow magic O_DIRECT into using folios.
>
>> • tmpfs directory cookie stability (cel)
>> • timestamp resolution and i_version (jlayton)
>> • GSS Kerberos futures (dhowells)
>> • NFS/NFSD CI (jlayton)
>> • NFSD POSIX to NFSv4 ACL translation - writing down the rules (all)
>>
>> Some of these topics might be appealing to others not specifically
>> involved with NFSD development. If there's something that should
>> be moved to another track or session, please pipe up.
>>
>> --
>> Chuck Lever
>>
>>
_________________________________
Trond Myklebust
Linux NFS client maintainer, Hammerspace
[email protected]
> On Apr 12, 2023, at 2:26 PM, Chuck Lever III <[email protected]> wrote:
>
> I'd like to request some time for those interested specifically
> in NFSD to gather and discuss some topics. Not a network file
> system free-for-all, but specifically for NFSD, because there
> is a long list of potential topics:
>
> • Progress on using iomap for NFSD READ/READ_PLUS (anna)
> • Replacing nfsd_splice_actor (all)
> • Transition from page arrays to bvecs (dhowells, hch)
> • tmpfs directory cookie stability (cel)
> • timestamp resolution and i_version (jlayton)
> • GSS Kerberos futures (dhowells)
> • NFS/NFSD CI (jlayton)
> • NFSD POSIX to NFSv4 ACL translation - writing down the rules (all)
>
> Some of these topics might be appealing to others not specifically
> involved with NFSD development. If there's something that should
> be moved to another track or session, please pipe up.
It's been suggested that this is too many topics for a
single session. Let me propose some ways of breaking it
up.
> • Progress on using iomap for NFSD READ/READ_PLUS (anna)
> • Replacing nfsd_splice_actor
This is probably worth its own session. We might want to
include how to convert NFSD to use folios, or maybe that
deserves its own conversation.
> • Transition from page arrays to bvecs (dhowells, hch)
This is something we can take to the hallway or discuss
over beers or a meal.
> • tmpfs directory cookie stability (cel)
This could be a FS/MM session. Aside from directory
cookies, there are issues about exporting any shmemfs-
based filesystem (autofs is another).
> • timestamp resolution and i_version (jlayton)
I'll let Jeff propose something here, and take this off
the NFSD-specific agenda.
> • GSS Kerberos futures (dhowells)
Perhaps this topic also requires us to be drunk first.
Seriously, though... it could be a pretty specialized
conversation, and thus left for the hallway track.
Or, David and I could fold this into the bvecs discussion
above, as these two are somewhat related.
> • NFS/NFSD CI (jlayton)
Network filesystems have special requirements for CI.
Jeff has been working on shaping kdevops to work for
our needs, and the work probably has broader appeal
than only to NFS. This could be its own 30-minute session,
or maybe we've got most everything worked out already?
> • NFSD POSIX to NFSv4 ACL translation - writing down the rules (all)
Could be its own session, but it might have only a
handful of interested parties.
--
Chuck Lever
> On Apr 15, 2023, at 12:34, Chuck Lever III <[email protected]> wrote:
>
>> • NFSD POSIX to NFSv4 ACL translation - writing down the rules (all)
>
>
> Could be its own session, but it might have only a
> handful of interested parties.
>
Doesn’t knfsd pretty much still follow https://datatracker.ietf.org/doc/html/draft-eriksen-nfsv4-acl-02 ?
AFAIK, that is where it was supposed to be documented.
_________________________________
Trond Myklebust
Linux NFS client maintainer, Hammerspace
[email protected]
On Sat, Apr 15, 2023 at 7:35 PM Chuck Lever III <[email protected]> wrote:
>
>
> > On Apr 12, 2023, at 2:26 PM, Chuck Lever III <[email protected]> wrote:
> >
> > I'd like to request some time for those interested specifically
> > in NFSD to gather and discuss some topics. Not a network file
> > system free-for-all, but specifically for NFSD, because there
> > is a long list of potential topics:
> >
> > • Progress on using iomap for NFSD READ/READ_PLUS (anna)
> > • Replacing nfsd_splice_actor (all)
> > • Transition from page arrays to bvecs (dhowells, hch)
> > • tmpfs directory cookie stability (cel)
> > • timestamp resolution and i_version (jlayton)
> > • GSS Kerberos futures (dhowells)
> > • NFS/NFSD CI (jlayton)
> > • NFSD POSIX to NFSv4 ACL translation - writing down the rules (all)
> >
> > Some of these topics might be appealing to others not specifically
> > involved with NFSD development. If there's something that should
> > be moved to another track or session, please pipe up.
>
> It's been suggested that this is too many topics for a
> single session. Let me propose some ways of breaking it
> up.
>
> > • Progress on using iomap for NFSD READ/READ_PLUS (anna)
> > • Replacing nfsd_splice_actor
>
>
> This is probably worth its own session. We might want to
> include how to convert NFSD to use folios, or maybe that
> deserves its own conversation.
>
> > • Transition from page arrays to bvecs (dhowells, hch)
>
>
> This is something we can take to the hallway or discuss
> over beers or a meal.
>
> > • tmpfs directory cookie stability (cel)
>
> This could be a FS/MM session. Aside from directory
> cookies, there are issues about exporting any shmemfs-
> based filesystem (autofs is another).
>
> > • timestamp resolution and i_version (jlayton)
>
> I'll let Jeff propose something here, and take this off
> the NFSD-specific agenda.
Please do.
>
> > • GSS Kerberos futures (dhowells)
>
> Perhaps this topic also requires us to be drunk first.
> Seriously, though... it could be a pretty specialized
> conversation, and thus left for the hallway track.
>
> Or, David and I could fold this into the bvecs discussion
> above, as these two are somewhat related.
>
> > • NFS/NFSD CI (jlayton)
>
> Network filesystems have special requirements for CI.
> Jeff has been working on shaping kdevops to work for
> our needs, and the work probably has broader appeal
> than only to NFS. This could be its own 30-minute session,
> or maybe we've got most everything worked out already?
>
Perhaps a guest speaker at Luis's kdevops session?
> > • NFSD POSIX to NFSv4 ACL translation - writing down the rules (all)
>
>
> Could be its own session, but it might have only a
> handful of interested parties.
>
Apropos interested parties. If there are any NFS developers/maintainers
that are interested in attending the NFSD BoF virtually, please let me know
and I will send you a code for the virtual registration option.
Thanks,
Amir.
> On Apr 16, 2023, at 12:20 AM, Amir Goldstein <[email protected]> wrote:
>
> On Sat, Apr 15, 2023 at 7:35 PM Chuck Lever III <[email protected]> wrote:
>>
>>
>>> On Apr 12, 2023, at 2:26 PM, Chuck Lever III <[email protected]> wrote:
>>
>>> • NFS/NFSD CI (jlayton)
>>
>> Network filesystems have special requirements for CI.
>> Jeff has been working on shaping kdevops to work for
>> our needs, and the work probably has broader appeal
>> than only to NFS. This could be its own 30-minute session,
>> or maybe we've got most everything worked out already?
>
> Perhaps a guest speaker at Luis's kdevops session?
That's fine with me. I'll give Jeff and Luis the final
word on that.
--
Chuck Lever