Hello all,
I've just noticed that ncpfs (or rather, NetWare) doesn't implement
POSIX semantics w.r.t. unlink(). Specifically, any operation that
expects, according to POSIX, that the contents of an open but unlink()ed
file should remain available to the application that has it open, such
as
perl -pi -e 's/a/b/;' a_file
on a ncpfs mounted volume causes the contents of the file to be lost,
as unlink() on ncpfs disposes of the file properly right away and the
new file that is written is a 's/a/b/;' of an empty file, hence another
empty file.
I don't know whether there is any (sensible) way to fix this, but I
would like to initiate a discussion on what would be the correct way
for ncpfs to deal with this. I just emptied a whole bunch of files
on our web server this way and was rather surprised about it. Nothing
was lost as the NetWare salvaging mechanism allowed us to rescue every-
thing with a bit of manual labour, and if that hadn't been the case,
we did have it all on tape, but it did catch me off guard... :-)
--
Atro Tossavainen (Mr.) / The Institute of Biotechnology at
Systems Analyst, Techno-Amish & / the University of Helsinki, Finland,
+358-9-19158939 UNIX Dinosaur / employs me, but my opinions are my own.
< URL : http : / / www . helsinki . fi / %7E atossava / > NO FILE ATTACHMENTS
Hi,
On 21 Dec 2007, at 11:01, Atro Tossavainen wrote:
> I've just noticed that ncpfs (or rather, NetWare) doesn't implement
> POSIX semantics w.r.t. unlink(). Specifically, any operation that
> expects, according to POSIX, that the contents of an open but
> unlink()ed
> file should remain available to the application that has it open, such
> as
>
> perl -pi -e 's/a/b/;' a_file
>
> on a ncpfs mounted volume causes the contents of the file to be lost,
> as unlink() on ncpfs disposes of the file properly right away and the
> new file that is written is a 's/a/b/;' of an empty file, hence
> another
> empty file.
Correct. You can use an extension (i.e. instead of in-place use
a .orig or whatever) to get around this.
> I don't know whether there is any (sensible) way to fix this, but I
> would like to initiate a discussion on what would be the correct way
> for ncpfs to deal with this.
Of course there is. Implement POSIX unlink semantics in the ncpfs
kernel module by using the same method as the NFS kernel module uses,
i.e. open files are renamed instead of unlinked, so effectively
simulating the above method of using an extension. Just look at the
NFS source code (linux-2.6/fs/nfs/dir.c), search for "silly" (the
feature is called "silly rename")...
Best regards,
Anton
--
Anton Altaparmakov <aia21 at cam.ac.uk> (replace at with @)
Unix Support, Computing Service, University of Cambridge, CB2 3QH, UK
Linux NTFS maintainer, http://www.linux-ntfs.org/