Subject: [PATCH 3/5] Use task_pid() to find leader's pid

From: Sukadev Bhattiprolu <sukadev@us.ibm.com>

Use task_pid() to get leader's pid since find_pid() cannot be used
after detach_pid(). See comments in the code below for more details.

Signed-off-by: Sukadev Bhattiprolu <sukadev@us.ibm.com>
Acked-by: Pavel Emelianov <xemul@openvz.org>

Cc: Eric W. Biederman <ebiederm@xmission.com>
Cc: Cedric Le Goater <clg@fr.ibm.com>
Cc: Dave Hansen <haveblue@us.ibm.com>
Cc: Serge Hallyn <serue@us.ibm.com>
Cc: Herbert Poetzel <herbert@13thfloor.at>
---
 fs/exec.c |    9 ++++++++-
 1 file changed, 8 insertions(+), 1 deletion(-)

Index: lx26-22-rc6-mm1/fs/exec.c
===================================================================
--- lx26-22-rc6-mm1.orig/fs/exec.c	2007-07-13 13:12:01.000000000 -0700
+++ lx26-22-rc6-mm1/fs/exec.c	2007-07-13 13:12:13.000000000 -0700
@@ -905,10 +905,17 @@ static int de_thread(struct task_struct 
 		 * The old leader becomes a thread of the this thread group.
 		 * Note: The old leader also uses this pid until release_task
 		 *       is called.  Odd but simple and correct.
+		 * Note: With multiple pid namespaces, active pid namespace of
+		 * 	 a process is stored in its struct pid. The detach_pid
+		 * 	 below frees the struct pid, so we will have no notion
+		 * 	 of an active pid namespace until we complete the
+		 * 	 subsequent attach_pid(). Which means - calls like
+		 * 	 find_pid()/pid_to_nr() return NULL and cannot be used
+		 * 	 between the detach_pid() and attach_pid() calls.
 		 */
 		detach_pid(tsk, PIDTYPE_PID);
 		tsk->pid = leader->pid;
-		attach_pid(tsk, PIDTYPE_PID,  find_pid(tsk->pid));
+		attach_pid(tsk, PIDTYPE_PID,  task_pid(leader));
 		transfer_pid(leader, tsk, PIDTYPE_PGID);
 		transfer_pid(leader, tsk, PIDTYPE_SID);
 		list_replace_rcu(&leader->tasks, &tsk->tasks);
_______________________________________________
Containers mailing list
Containers@lists.linux-foundation.org
https://lists.linux-foundation.org/mailman/listinfo/containers

Sukadev Bhattiprolu wrote:
>
> Use task_pid() to get leader's pid since find_pid() cannot be used
> after detach_pid(). See comments in the code below for more details.
>
> ...
>
> +		 * Note: With multiple pid namespaces, active pid namespace of
> +		 * 	 a process is stored in its struct pid. The detach_pid
> +		 * 	 below frees the struct pid, so we will have no notion
> +		 * 	 of an active pid namespace until we complete the
> +		 * 	 subsequent attach_pid(). Which means - calls like
> +		 * 	 find_pid()/pid_to_nr() return NULL and cannot be used
> +		 * 	 between the detach_pid() and attach_pid() calls.

I think both the changelog and the comment are confusing,

>  		detach_pid(tsk, PIDTYPE_PID);
>  		tsk->pid = leader->pid;
> -		attach_pid(tsk, PIDTYPE_PID,  find_pid(tsk->pid));
> +		attach_pid(tsk, PIDTYPE_PID,  task_pid(leader));

because the change itself looks like an obvious performance fix, even
we don't use multiple pid namespaces. I don't think it is good idea to
add a fat comment which doesn't match the current reality, and find_pid()
should be avoided anyway.

Stupid question: why do we need to put the pid namespace into the struct
pid? Isn't it better if the user of the struct pid should know its ns?
For example, if /proc does put_pid(), that pid should be from the active
namespace.

Sukadev, could you cc me if you do that kind of changes?

Oleg.

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

Oleg Nesterov [oleg@tv-sign.ru] wrote:
| Sukadev Bhattiprolu wrote:
| >
| > Use task_pid() to get leader's pid since find_pid() cannot be used
| > after detach_pid(). See comments in the code below for more details.
| >
| > ...
| >
| > +		 * Note: With multiple pid namespaces, active pid namespace of
| > +		 * 	 a process is stored in its struct pid. The detach_pid
| > +		 * 	 below frees the struct pid, so we will have no notion
| > +		 * 	 of an active pid namespace until we complete the
| > +		 * 	 subsequent attach_pid(). Which means - calls like
| > +		 * 	 find_pid()/pid_to_nr() return NULL and cannot be used
| > +		 * 	 between the detach_pid() and attach_pid() calls.
| 
| I think both the changelog and the comment are confusing,
| 
| >  		detach_pid(tsk, PIDTYPE_PID);
| >  		tsk->pid = leader->pid;
| > -		attach_pid(tsk, PIDTYPE_PID,  find_pid(tsk->pid));
| > +		attach_pid(tsk, PIDTYPE_PID,  task_pid(leader));
| 
| because the change itself looks like an obvious performance fix, even
| we don't use multiple pid namespaces. I don't think it is good idea to
| add a fat comment which doesn't match the current reality, and find_pid()
| should be avoided anyway.

Its a performance fix but also a correctness issue with multiple pid
namespaces. Here is the modified patch with the simplified changelog and
comment removed.

| 
| Stupid question: why do we need to put the pid namespace into the struct
| pid? Isn't it better if the user of the struct pid should know its ns?
| For example, if /proc does put_pid(), that pid should be from the active
| namespace.

Not sure I fully understand this. A process, and by extension its 'struct
pid' is visible in multiple namespaces and we maintain this list of
namespaces in each 'struct pid'.

Are you suggesting having a pid_namespace with a list of all 'struct pids'
that are visible in it ?

| 
| Sukadev, could you cc me if you do that kind of changes?

Sure - I will.
---


Subject: [PATCH 3/5] Use task_pid() to find leader's pid

From: Sukadev Bhattiprolu <sukadev@us.ibm.com>

Use task_pid() to get leader's 'struct pid' and avoid the find_pid().

Signed-off-by: Sukadev Bhattiprolu <sukadev@us.ibm.com>
Acked-by: Pavel Emelianov <xemul@openvz.org>

Cc: Eric W. Biederman <ebiederm@xmission.com>
Cc: Cedric Le Goater <clg@fr.ibm.com>
Cc: Dave Hansen <haveblue@us.ibm.com>
Cc: Serge Hallyn <serue@us.ibm.com>
Cc: Herbert Poetzel <herbert@13thfloor.at>
---
 fs/exec.c |    2 +-
 1 file changed, 1 insertion(+), 1 deletion(-)

Index: lx26-22-rc6-mm1a/fs/exec.c
===================================================================
--- lx26-22-rc6-mm1a.orig/fs/exec.c	2007-07-13 18:23:55.000000000 -0700
+++ lx26-22-rc6-mm1a/fs/exec.c	2007-07-16 12:56:22.000000000 -0700
@@ -908,7 +908,7 @@ static int de_thread(struct task_struct 
 		 */
 		detach_pid(tsk, PIDTYPE_PID);
 		tsk->pid = leader->pid;
-		attach_pid(tsk, PIDTYPE_PID,  find_pid(tsk->pid));
+		attach_pid(tsk, PIDTYPE_PID,  task_pid(leader));
 		transfer_pid(leader, tsk, PIDTYPE_PGID);
 		transfer_pid(leader, tsk, PIDTYPE_SID);
 		list_replace_rcu(&leader->tasks, &tsk->tasks);
_______________________________________________
Containers mailing list
Containers@lists.linux-foundation.org
https://lists.linux-foundation.org/mailman/listinfo/containers

On 07/16, sukadev@us.ibm.com wrote:
>
> Oleg Nesterov [oleg@tv-sign.ru] wrote:
> | 
> | Stupid question: why do we need to put the pid namespace into the struct
> | pid? Isn't it better if the user of the struct pid should know its ns?
> | For example, if /proc does put_pid(), that pid should be from the active
> | namespace.
> 
> Not sure I fully understand this. A process, and by extension its 'struct
> pid' is visible in multiple namespaces and we maintain this list of
> namespaces in each 'struct pid'.
> 
> Are you suggesting having a pid_namespace with a list of all 'struct pids'
> that are visible in it ?

I thought that the plan is: if the task is visible in some namespace, it has
a separate pid_t in that namespace.

OK, the question was relly stupid, please ignore. I'll wait for other patches
to understand what's going on.

Oleg.

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