16675 want syncfs(3C)

16675 want syncfs(3C)

[Robert Mustacchi]

https://code.illumos.org/c/illumos-gate/+/3608

This adds support for a syncfs call in libc and the associated VFS
plumbing. We've found this useful for some applications that we work on
and this call originates in Linux as well. See the ticket for some
additional details.

Robert

Re: [developer] 16675 want syncfs(3C)

[Richard Lowe]

[-- Attachment #1: Type: text/plain, Size: 77 bytes --]

I think this was meant to be https://code.illumos.org/c/illumos-gate/+/3609

[-- Attachment #2: Type: text/html, Size: 193 bytes --]

Re: [developer] 16675 want syncfs(3C)

[Robert Mustacchi]

On 7/25/24 17:03, Richard Lowe wrote:
> I think this was meant to be https://code.illumos.org/c/illumos-gate/+/3609

Oops, it was. Thanks for catching that.

Robert

Re: [developer] 16675 want syncfs(3C)

[Gordon Ross]

I read through your write-up in the issue.
Have you surveyed which file system types on Linux support syncfs(2)?
I'm curious what they might have done for NFS and SMB, which don't
really offer a natural way to sync everything under a mount point.

On Thu, Jul 25, 2024 at 8:09 PM Robert Mustacchi <rm@fingolfin.org> wrote:
>
> On 7/25/24 17:03, Richard Lowe wrote:
> > I think this was meant to be https://code.illumos.org/c/illumos-gate/+/3609
>
> Oops, it was. Thanks for catching that.
>
> Robert
>
>
> ------------------------------------------
> illumos: illumos-developer
> Permalink: https://illumos.topicbox.com/groups/developer/T1f0f6493c10b2c62-M1c91eb47d0009e134eb743b4
> Delivery options: https://illumos.topicbox.com/groups/developer/subscription

Re: [developer] 16675 want syncfs(3C)

[Robert Mustacchi]

On 7/26/24 18:33, Gordon Ross wrote:
> I read through your write-up in the issue.
> Have you surveyed which file system types on Linux support syncfs(2)?
> I'm curious what they might have done for NFS and SMB, which don't
> really offer a natural way to sync everything under a mount point.

I didn't survey it too extensively. The structure of the VFS between the
systems is a bit different. In our case the VFS_SYNC and VFS_SYNCFS
entry point expects the file system to determine dirty vnodes, schedule
writes, and related. While not all file systems implement the specific
sync_fs() entry point, other logic is called around there that does
something for all file systems, the specifics I couldn't say. NFS and
SMB don't implement the sync_fs() file system specific vector over there.

Robert

> On Thu, Jul 25, 2024 at 8:09 PM Robert Mustacchi <rm@fingolfin.org> wrote:
>>
>> On 7/25/24 17:03, Richard Lowe wrote:
>>> I think this was meant to be https://code.illumos.org/c/illumos-gate/+/3609
>>
>> Oops, it was. Thanks for catching that.
>>
>> Robert
>>
> 
> ------------------------------------------
> illumos: illumos-developer
> Permalink: https://illumos.topicbox.com/groups/developer/T1f0f6493c10b2c62-M9aa42cc3f2af94588b816c9c
> Delivery options: https://illumos.topicbox.com/groups/developer/subscription