Re: NFS: open(...O_CREAT...) fails, yet file is created

[christos%astron.com@localhost (Christos Zoulas)]

In article <20090925135025.GC22387%orion.math.uni-bonn.de@localhost>,
Edgar Fuß  <ef%math.uni-bonn.de@localhost> wrote:
>I noticed a strange behaviour of NetBSD as an NFS server:
>
>If you open(..., O_CREAT|O_EXCL, S_ISGID) in a directory mounted over
>NFS and belonging to a group you're not a member of, the open() fails
>with EPERM. However, the file is created nevertheless. On a local fs,
>the call just fails with no file being created.
>
>I'm unsure as whether this behaviour (creating the file) is correct. At
>least, it's somewhat surprising.
>I guess it may be a result of the whole open() operation resulting in
>several NFS RPC, some of which succeed and some fail.
>
>I've tried to boil down this as far as possible: O_EXCL is necessarry
>and I couldn't find another reason for the open() to fail resulting in
>this behaviour. It seems to appear only with the combination of implicit
>group membership assignement, the issuer not belonging to this group and
>S_ISGID.
>
>For us, it causes Adobe Reader 9.0 to fail in strage ways ("the document
>has been changed") for files with Reader Extensions enabled. There seems
>to be a bug in Adobe Reader causing it to open
>~/.adobe/Acrobat/9.0/AdobeIDataSync/CDataSync_AddressBook with some
>uninitialized mode which happens to have S_ISGID set. Our user's home
>directories (on a NetBSD NFS server) have their group set to a "home"
>group with no members.
>
>Any comments?

File a PR. Is the client NetBSD too? What version?

christos