Tigran Aivazian wrote:
> On 2 Jun 2000, H. Peter Anvin wrote:
> > Followup to: <Pine.LNX.email@example.com>
> > By author: Tigran Aivazian <firstname.lastname@example.org>
> > In newsgroup: linux.dev.kernel
> > >
> > > did you look at kern_mount/umount interface that went into the kernel
> > > recently? Things like shm and pipefs should (and already are) mounted
> > > under artificially constructed root mountpoints like "pipe:" which are not
> > > visible for namelookups coming from userspace. Ok, it is not the case for
> > > shm currently (it is mounted on /) but perhaps it should be (i.e. the way
> > > root dentry should be allocated as for pipefs using d_alloc() directly
> > > instead of using generic d_alloc_root())?
> > >
> > > I cc'd Al Viro as I suspect he knows the answer to the above question.
> > >
> > No, shmfs actually needs a real path. Not for SysV IPC, but for POSIX
> > IPC.
> that is not an answer to the question I asked. If you read the sentence
> that has the question mark (not the subject) you will see that I was
> asking if the kern_mounted (and not the user-mounted) instance of shmfs
> has to be mounted under "/" or can it be done like for pipefs under an
> artificially constructed "shm:" like "pipe:"
Yes, it is.
The distinction you're referring to is the distinction of a real path
versus a bogus path -- you cannot under any circumstances construct a
path which uses one of those artificial roots.
POSIX IPC requires a real path. It needs to be in the real namespace,
which starts with /.
> For this reason I cc'd Al Viro also - not to ask him a question on POSIX
> vs System V IPC but a question on root dentry allocation differences of
> pipefs vs shmfs. (I cc'd him again)
To unsubscribe from this list: send the line "unsubscribe linux-kernel" in
the body of a message to email@example.com
Please read the FAQ at http://www.tux.org/lkml/
This archive was generated by hypermail 2b29 : Wed Jun 07 2000 - 21:00:16 EST