[PATCH] exit: Clarify choice of new parent in forget_original_parent()

[PATCH] exit: Clarify choice of new parent in forget_original_parent()

[Kirill Tkhai]

Second parameter of find_new_reaper() and the similarity of its name
and find_child_reaper()'s name confuse a reader.

Rename find_child_reaper() for better conformity of its name and its
function. Also delete the second parameter of find_new_reaper().

These both improve modularity.

Signed-off-by: Kirill Tkhai <ktkhai@odin.com>
Cc: Oleg Nesterov <oleg@redhat.com>
Cc: Andrew Morton <akpm@linux-foundation.org>
---
 kernel/exit.c |   24 ++++++++++++++----------
 1 file changed, 14 insertions(+), 10 deletions(-)

diff --git a/kernel/exit.c b/kernel/exit.c
index 22fcc05..ae60a4d 100644
--- a/kernel/exit.c
+++ b/kernel/exit.c
@@ -450,7 +450,7 @@ static struct task_struct *find_alive_thread(struct task_struct *p)
 	return NULL;
 }
 
-static struct task_struct *find_child_reaper(struct task_struct *father)
+static void exit_pidns_child_reaper(struct task_struct *father)
 	__releases(&tasklist_lock)
 	__acquires(&tasklist_lock)
 {
@@ -458,12 +458,12 @@ static struct task_struct *find_child_reaper(struct task_struct *father)
 	struct task_struct *reaper = pid_ns->child_reaper;
 
 	if (likely(reaper != father))
-		return reaper;
+		return;
 
 	reaper = find_alive_thread(father);
 	if (reaper) {
 		pid_ns->child_reaper = reaper;
-		return reaper;
+		return;
 	}
 
 	write_unlock_irq(&tasklist_lock);
@@ -473,8 +473,6 @@ static struct task_struct *find_child_reaper(struct task_struct *father)
 	}
 	zap_pid_ns_processes(pid_ns);
 	write_lock_irq(&tasklist_lock);
-
-	return father;
 }
 
 /*
@@ -484,15 +482,21 @@ static struct task_struct *find_child_reaper(struct task_struct *father)
  *    child_subreaper for its children (like a service manager)
  * 3. give it to the init process (PID 1) in our pid namespace
  */
-static struct task_struct *find_new_reaper(struct task_struct *father,
-					   struct task_struct *child_reaper)
+static struct task_struct *find_new_reaper(struct task_struct *father)
 {
-	struct task_struct *thread, *reaper;
+	struct task_struct *thread, *reaper, *child_reaper;
 
 	thread = find_alive_thread(father);
 	if (thread)
 		return thread;
 
+	child_reaper = task_active_pid_ns(father)->child_reaper;
+	/*
+	 * child_reaper doesn't have children after zap_pid_ns_processes(),
+	 * therefore it can't enter this function.
+	 */
+	BUG_ON(child_reaper == father);
+
 	if (father->signal->has_child_subreaper) {
 		/*
 		 * Find the first ->is_child_subreaper ancestor in our pid_ns.
@@ -557,11 +561,11 @@ static void forget_original_parent(struct task_struct *father,
 		exit_ptrace(father, dead);
 
 	/* Can drop and reacquire tasklist_lock */
-	reaper = find_child_reaper(father);
+	exit_pidns_child_reaper(father);
 	if (list_empty(&father->children))
 		return;
 
-	reaper = find_new_reaper(father, reaper);
+	reaper = find_new_reaper(father);
 	list_for_each_entry(p, &father->children, sibling) {
 		for_each_thread(p, t) {
 			t->real_parent = reaper;

Re: [PATCH] exit: Clarify choice of new parent in forget_original_parent()

[Oleg Nesterov]

On 06/16, Kirill Tkhai wrote:
>
> Second parameter of find_new_reaper() and the similarity of its name
> and find_child_reaper()'s name confuse a reader.

OK, I agree that

	reaper = find_child_reaper(father);
	...
	reaper = find_new_reaper(father, reaper);

can look confusing and probably deserves a cleanup. How about the patch
below then?
	
> Rename find_child_reaper() for better conformity of its name and its
> function.

I never argueus with renames ;) Probably the new name looks better.

> Also delete the second parameter of find_new_reaper().

Yes, we can do this. But this 2nd argument avoids another another
task_active_pid_ns(father)->child_reaper, so this is optimization.

I agree, this optimization is minor, but still I think this change
needs some justification.

> +static struct task_struct *find_new_reaper(struct task_struct *father)
>  {
> -	struct task_struct *thread, *reaper;
> +	struct task_struct *thread, *reaper, *child_reaper;
>  
>  	thread = find_alive_thread(father);
>  	if (thread)
>  		return thread;
>  
> +	child_reaper = task_active_pid_ns(father)->child_reaper;
> +	/*
> +	 * child_reaper doesn't have children after zap_pid_ns_processes(),
> +	 * therefore it can't enter this function.
> +	 */
> +	BUG_ON(child_reaper == father);

Yes, we can add this BUG_ON(). But please see the comments in
zap_pid_ns_processes(). We can change zap_pid_ns_processes() so that
it returns with non-empty ->children list due to EXIT_DEAD children.

Unlikely we will actually do this, at least soon, so I won't argue
with this BUG_ON().

But. In this case it would be better to add it into forget_original_parent(),

	reaper = find_new_reaper(...);
	BUG_ON(reaper == father);


Oh. Off-topic, but this reminds me that I forgot about another bug with
->has_child_subreaper... this needs another discussion.

Oleg.

--- x/kernel/exit.c
+++ x/kernel/exit.c
@@ -551,17 +551,17 @@ static void reparent_leader(struct task_
 static void forget_original_parent(struct task_struct *father,
 					struct list_head *dead)
 {
-	struct task_struct *p, *t, *reaper;
+	struct task_struct *p, *t, *child_reaper, *reaper;
 
 	if (unlikely(!list_empty(&father->ptraced)))
 		exit_ptrace(father, dead);
 
 	/* Can drop and reacquire tasklist_lock */
-	reaper = find_child_reaper(father);
+	child_reaper = find_child_reaper(father);
 	if (list_empty(&father->children))
 		return;
 
-	reaper = find_new_reaper(father, reaper);
+	reaper = find_new_reaper(father, child_reaper);
 	list_for_each_entry(p, &father->children, sibling) {
 		for_each_thread(p, t) {
 			t->real_parent = reaper;

Re: [PATCH] exit: Clarify choice of new parent in forget_original_parent()

[Oleg Nesterov]

forgot to mention,

On 06/16, Oleg Nesterov wrote:
>
> On 06/16, Kirill Tkhai wrote:
> >
> > +	 * child_reaper doesn't have children after zap_pid_ns_processes(),
> > +	 * therefore it can't enter this function.
> > +	 */
> > +	BUG_ON(child_reaper == father);
>
> Yes, we can add this BUG_ON(). But please see the comments in
> zap_pid_ns_processes(). We can change zap_pid_ns_processes() so that
> it returns with non-empty ->children list due to EXIT_DEAD children.
>
> Unlikely we will actually do this, at least soon, so I won't argue
> with this BUG_ON().
>
> But. In this case it would be better to add it into forget_original_parent(),
>
> 	reaper = find_new_reaper(...);
> 	BUG_ON(reaper == father);

because this way:

	1. This BUG_ON() will still be valid even if we actually change
	   zap_pid_ns_processes() to return with EXIT_DEAD children

	2. If we really want this sanity check, we should not tie it to
	   ->child_reaper case.

OTOH. If for some reason you want to check ->child_reaper only, then
you should probably do this right after list_empty(&father->children)
check, or at least before find_alive_thread(). Because otherwise it
looks confusing, it looks as if "child_reaper == father" is only wrong
if find_alive_thread(father) fails.

Oleg.

Re: [PATCH] exit: Clarify choice of new parent in forget_original_parent()

[Kirill Tkhai]

В Вт, 16/06/2015 в 22:03 +0200, Oleg Nesterov пишет:
> forgot to mention,
> 
> On 06/16, Oleg Nesterov wrote:
> >
> > On 06/16, Kirill Tkhai wrote:
> > >
> > > +	 * child_reaper doesn't have children after zap_pid_ns_processes(),
> > > +	 * therefore it can't enter this function.
> > > +	 */
> > > +	BUG_ON(child_reaper == father);
> >
> > Yes, we can add this BUG_ON(). But please see the comments in
> > zap_pid_ns_processes(). We can change zap_pid_ns_processes() so that
> > it returns with non-empty ->children list due to EXIT_DEAD children.
> >
> > Unlikely we will actually do this, at least soon, so I won't argue
> > with this BUG_ON().
> >
> > But. In this case it would be better to add it into forget_original_parent(),
> >
> > 	reaper = find_new_reaper(...);
> > 	BUG_ON(reaper == father);
> 
> because this way:
> 
> 	1. This BUG_ON() will still be valid even if we actually change
> 	   zap_pid_ns_processes() to return with EXIT_DEAD children
> 
> 	2. If we really want this sanity check, we should not tie it to
> 	   ->child_reaper case.
> 
> OTOH. If for some reason you want to check ->child_reaper only, then
> you should probably do this right after list_empty(&father->children)
> check, or at least before find_alive_thread(). Because otherwise it
> looks confusing, it looks as if "child_reaper == father" is only wrong
> if find_alive_thread(father) fails.

Sure, it's more logical. Thanks, Oleg.

Re: [PATCH] exit: Clarify choice of new parent in forget_original_parent()

[Kirill Tkhai]

В Вт, 16/06/2015 в 21:27 +0200, Oleg Nesterov пишет:
> On 06/16, Kirill Tkhai wrote:
> >
> > Second parameter of find_new_reaper() and the similarity of its name
> > and find_child_reaper()'s name confuse a reader.
> 
> OK, I agree that
> 
> 	reaper = find_child_reaper(father);
> 	...
> 	reaper = find_new_reaper(father, reaper);
> 
> can look confusing and probably deserves a cleanup. How about the patch
> below then?

Good, IMO it improves the readability.

> 	
> > Rename find_child_reaper() for better conformity of its name and its
> > function.
> 
> I never argueus with renames ;) Probably the new name looks better.
> 
> > Also delete the second parameter of find_new_reaper().
> 
> Yes, we can do this. But this 2nd argument avoids another another
> task_active_pid_ns(father)->child_reaper, so this is optimization.
> 
> I agree, this optimization is minor, but still I think this change
> needs some justification.

It looks like gcc inlines both of these function, so it seems there won't
be a problem...

> 
> > +static struct task_struct *find_new_reaper(struct task_struct *father)
> >  {
> > -	struct task_struct *thread, *reaper;
> > +	struct task_struct *thread, *reaper, *child_reaper;
> >  
> >  	thread = find_alive_thread(father);
> >  	if (thread)
> >  		return thread;
> >  
> > +	child_reaper = task_active_pid_ns(father)->child_reaper;
> > +	/*
> > +	 * child_reaper doesn't have children after zap_pid_ns_processes(),
> > +	 * therefore it can't enter this function.
> > +	 */
> > +	BUG_ON(child_reaper == father);
> 
> Yes, we can add this BUG_ON(). But please see the comments in
> zap_pid_ns_processes(). We can change zap_pid_ns_processes() so that
> it returns with non-empty ->children list due to EXIT_DEAD children.

Yes, I saw. Since zap_pid_ns_processes() waits for nr_hashed,
and __unhash_process() deletes from pid chain and sibling list
at the same time, pid_ns child_reaper can't have a child after
nr_hashed == init_pids.

> Unlikely we will actually do this, at least soon, so I won't argue
> with this BUG_ON().
> 
> But. In this case it would be better to add it into forget_original_parent(),
> 
> 	reaper = find_new_reaper(...);
> 	BUG_ON(reaper == father);

Yeah, I'm agree.

> Oh. Off-topic, but this reminds me that I forgot about another bug with
> ->has_child_subreaper... this needs another discussion.
> 
> Oleg.
> 
> --- x/kernel/exit.c
> +++ x/kernel/exit.c
> @@ -551,17 +551,17 @@ static void reparent_leader(struct task_
>  static void forget_original_parent(struct task_struct *father,
>  					struct list_head *dead)
>  {
> -	struct task_struct *p, *t, *reaper;
> +	struct task_struct *p, *t, *child_reaper, *reaper;
>  
>  	if (unlikely(!list_empty(&father->ptraced)))
>  		exit_ptrace(father, dead);
>  
>  	/* Can drop and reacquire tasklist_lock */
> -	reaper = find_child_reaper(father);
> +	child_reaper = find_child_reaper(father);
>  	if (list_empty(&father->children))
>  		return;
>  
> -	reaper = find_new_reaper(father, reaper);
> +	reaper = find_new_reaper(father, child_reaper);
>  	list_for_each_entry(p, &father->children, sibling) {
>  		for_each_thread(p, t) {
>  			t->real_parent = reaper;
> 

Thanks,
Kirill