[Devel] Re: [RFC][PATCH] Pid namespaces vs locks interaction

[Brad Boyer]

On Thu, Dec 06, 2007 at 09:51:30AM -0600, Serge E. Hallyn wrote:
> Quoting Vitaliy Gusev (vgusev at openvz.org):
> > fl_pid is used by nfs, fuse and gfs2. For instance nfs  keeps in fl_pid some 
> > unique id to identify locking process between hosts - it is not a process 
> > pid.
> 
> Ok, but so the struct user_flock->fl_pid is being set to the task's
> virtual pid, while the struct kernel_flock->fl_pid is being set to
> task->tgid for nfsd use.
> 
> Why can't nfs just generate a uniqueid from the struct pid when it
> needs it?
> 
> Fuse just seems to copy the pid to report it to userspace, so it would
> just copy pid_vnr(kernel_flock->pid) into user_flock->fl_pid.
> 
> Anyway I haven't looked at all the uses of struct fl_pid, but you
> can always get the pidnr back from the struct pid if needed so there
> should be no problem.

Perhaps we could add a sysid field like some unix systems have. Here
is the flock structure documentation from Sun:

----
The flock structure contains at least the following elements:

short   l_type;       /* lock operation type */ 
short   l_whence;     /* lock base indicator */ 
off_t   l_start;      /* starting offset from base */ 
off_t   l_len;        /* lock length; l_len == 0 means
                         until end of file */ 
int     l_sysid;      /* system ID running process holding lock */ 
pid_t   l_pid;        /* process ID of process holding lock */ 
----

Using the sysid could show that the pid field refers to a separate
namespace, and might also be useful for NFS to show that the lock
is really held by a process on a different system. This would also
be something we could export to user space in a way that some
programs are already written to expect and handle properly.

	Brad Boyer
	flar at allandria.com

_______________________________________________
Containers mailing list
Containers at lists.linux-foundation.org
https://lists.linux-foundation.org/mailman/listinfo/containers