Re: [9fans] 9p questions

[rog@vitanuova.com]

> So, when wstat updates a file's name does 9p place a
> restriction on wether the file is open or not?

no - you can wstat whether the file is open or not (fwstat, wstat).

you could always return an error if a rename was attempted when the
file was open.  presumably the file can be currently open by another
user, so i'd assume doing a close/rename/reopen is unlikely to be
sufficient in general.

i don't know if much plan 9 infrastructure relies on being able to use
fwstat to rename a file (the most likely candidate, upas, doesn't use
fwstat to rename the mbox, but then again it might easily be holding
another fd open on it).

> Netware has similar restrictions on deleteing files/dirs so
> I am closing then also,  but I cannot imagine any side
> effects from this.

remove is a clunk anyway, so no problems there (apart from
if another process has it open, as above, but nothing you can
do there).

> Using lib9p is it OK to change fid->qid in wstat, under the
> libraries feet so to speak?

this is fairly dodgy; the only times AFAIK that a qid corresponding
to a fid can be changed are attach, create, open and walk,
and all these return the new qid in the rmsg.

> Also, will it mess up plan9 clients?

good question...
'cos i know it's difficult verging on impossible to manufacture a
proper qid space for some imported filesystems.

i know that inferno used a hash of the filename for its ntfs
filesystem under windows for quite a while, and most things worked ok.
i think it still does the same thing, with the exception that a file
which has been walked to keeps the same qid for as long as there's a
reference to it.

this ensures that a bind onto a directory stays put even though the
directory has been renamed, but it's not perfect.  i may be wrong
about this too, as i didn't have anything to do with the changes.

  cheers,
    rog.