Re: [PATCH 1/7] cifs: Add mount option -o reparse=native

["Pali Rohár" <pali@kernel.org>]

On Thursday 10 October 2024 17:42:55 Steve French wrote:
> On Thu, Oct 10, 2024, 5:39 PM Pali Rohár <pali@kernel.org> wrote:
> 
> > On Thursday 10 October 2024 17:21:59 Steve French wrote:
> > > On Thu, Oct 10, 2024 at 5:17 PM Steve French <smfrench@gmail.com> wrote:
> > > >
> > > >
> > > >
> > > > On Mon, Oct 7, 2024 at 1:36 PM Pali Rohár <pali@kernel.org> wrote:
> > > >>
> > > >> Currently choosing how new symlinks are created is quite complicated.
> > > >>
> > > >> Without these patch series, by default new symlinks are created via
> > > >> native reparse points, even when reparse=nfs or reparse=wsl is
> > > >> specified. There is no possibility to create a NFS-style or WSL-style
> > > >> symlink yet, and this patch series address this missing functionality.
> > > >> When option -o sfu is specified then all new symlinks are created in
> > > >> SFU-style, independently of -o reparse option. And when -o mfsymlinks
> > is
> > > >> specified then all new symlinks are created in mf style, independently
> > > >> of -o reparse and -o sfu options.
> > > >>
> > > >> This patch series does not change -o sfu and -o mfsymlinks overrides,
> > it
> > > >> just changes the way how -o reparse is handled.
> > > >>
> > >
> > > I lean toward something similar, and more intuitive - do not have
> > > "reparse=" control symlink creation - but instead use another mount
> > > parm (e.g. "symlink=") for that.  It would be rarely used - only if
> > > you don't want the default (windows default format too) for server
> > > symlinks or "mfsymlinks" (for client only symlinks):
> > >
> > > 1) "symlink=" if specified can be set to one of five formats (with the
> > > default being the windows format)
> > >   a) "mfsymlinks" (Mac style which is safer for many use cases since
> > > they are "client only" symlinks which the server will never use)
> > >      Setting "symlink=mfsymlinks" will have the same effect as just
> > > specifying "mfsymlinks" so won't break anything
> > >   b) "default" (or "windows") which uses the default symlink format
> > > when trying to create a new symlink
> > >   c) "nfs"
> > >   d) "wsl"
> > >   e) "sfu"
> > > 2) "reparse=" will still control how special files are created (char,
> > > block, fifo, socket) and can be set to:
> > >    a) "nfs" (default)
> > >    b) or "wsl"
> > >    c) If "sfu" set on mount will cause special files to be created
> > > with "sfu" format instead of using reparse points to create
> > > 3) reading reparse points will always be supported (unless you want to
> > > add a new parameter "reparse=none" to treat all reparse points as
> > > empty directories)
> > > 4) reading special files via the old "sfu" will only be supported if
> > > you mount with "sfu"
> > >
> > >
> > >
> > >
> > > --
> > > Thanks,
> > >
> > > Steve
> >
> > Ok, and how to handle creating new sockets? For me it makes sense to
> > create new sockets in "native" AF_UNIX style - compatible with Windows
> > WinAPI / WinSocks. Should be there also a new parameter?
> >
> 
> Possibly, but we can decide that later. Aren't sockets typically more
> transient short lives so less likely to cause issues no matter what format
> we use??
> 
> >

Yes, they are short lives.