Re: [PATCH 5/5] Move alloc_pid call to copy_process

[sukadev]

Oleg Nesterov [oleg@xxxxxxxxxx] wrote:
| On 07/16, sukadev@xxxxxxxxxx wrote:
| >
| > Oleg Nesterov [oleg@xxxxxxxxxx] wrote:
| > | 
| > | Could you please give more details why we need this change?
| > 
| > Well, with multiple pid namespaces, we may need to allocate a new
| > 'struct pid_namespace' if the CLONE_NEWPID flag is specified. And
| > as a part of initializing this pid_namespace, we need the 'task_struct'
| > that will be the reaper of the new pid namespace.
| > 
| > And this task_struct is allocated in copy_process(). So we could
| > still alloc_pid() in do_fork(), as we are doing currently and set
| > the reaper of the new pid_namespace later in copy_process(). But
| > that seemed to complicate error handling and add checks again in
| > copy_process() for the CLONE_NEWPID.
| 
| OK, thanks.
| 
| > 
| > | Even if we really need this, can't we do these checks in copy_process() ?
| > 
| > We could and I did have a check in copy_process() in one of my earlier
| > versions to Containers@ list.  We thought it cluttered copy_process() a
| > bit.
| 
| Yes, but having the "pid == &init_struct_pid" in free_pid() is imho worse,
| 
| >  	container_exit(p, container_callbacks_done);
| >  	delayacct_tsk_free(p);
| > +	free_pid(pid);
| > +bad_fork_put_binfmt_module:
| > [...snip...]
| > @@ -206,6 +206,10 @@ fastcall void free_pid(struct pid *pid)
| >  	/* We can be called with write_lock_irq(&tasklist_lock) held */
| >  	unsigned long flags;
| >  
| > +	/* check this here to keep copy_process() cleaner */
| > +	if (unlikely(pid == &init_struct_pid))
| > +		return;
| > +
| 
| Wouldn't it better if copy_process()'s error path does
| 
| 	if (pid != &init_struct_pid)
| 		free_pid(pid);
| 
| instead? OK, "cleaner" is a matter of taste, but from the perfomance POV
| this would be better, even if not noticable.

Agree. I realized it too late last night. Here is a modified patch
(this checks init_struct_pid in both places now for better consistency)

Subject: [PATCH 5/5] Move alloc_pid call to copy_process

From: Sukadev Bhattiprolu <sukadev@xxxxxxxxxx>

Move alloc_pid() into copy_process(). This will keep all pid and pid
namespace code together and simplify error handling when we support
multiple pid namespaces.

Signed-off-by: Sukadev Bhattiprolu <sukadev@xxxxxxxxxx>

Cc: Pavel Emelianov <xemul@xxxxxxxxxx>
Cc: Eric W. Biederman <ebiederm@xxxxxxxxxxxx>
Cc: Cedric Le Goater <clg@xxxxxxxxxx>
Cc: Dave Hansen <haveblue@xxxxxxxxxx>
Cc: Serge Hallyn <serue@xxxxxxxxxx>
Cc: Herbert Poetzel <herbert@xxxxxxxxxxxx>
---
 kernel/fork.c |   19 +++++++++++++------
 1 file changed, 13 insertions(+), 6 deletions(-)

Index: lx26-22-rc6-mm1a/kernel/fork.c
===================================================================
--- lx26-22-rc6-mm1a.orig/kernel/fork.c	2007-07-16 12:55:13.000000000 -0700
+++ lx26-22-rc6-mm1a/kernel/fork.c	2007-07-17 10:08:12.000000000 -0700
@@ -1029,6 +1029,12 @@ static struct task_struct *copy_process(
 	if (p->binfmt && !try_module_get(p->binfmt->module))
 		goto bad_fork_cleanup_put_domain;
 
+	if (pid != &init_struct_pid) {
+		pid = alloc_pid();
+		if (!pid)
+			goto bad_fork_put_binfmt_module;
+	}
+
 	p->did_exec = 0;
 	delayacct_tsk_init(p);	/* Must remain after dup_task_struct() */
 	copy_flags(clone_flags, p);
@@ -1316,6 +1322,9 @@ bad_fork_cleanup_container:
 #endif
 	container_exit(p, container_callbacks_done);
 	delayacct_tsk_free(p);
+	if (pid != &init_struct_pid)
+		free_pid(pid);
+bad_fork_put_binfmt_module:
 	if (p->binfmt)
 		module_put(p->binfmt->module);
 bad_fork_cleanup_put_domain:
@@ -1380,19 +1389,16 @@ long do_fork(unsigned long clone_flags,
 {
 	struct task_struct *p;
 	int trace = 0;
-	struct pid *pid = alloc_pid();
 	long nr;
 
-	if (!pid)
-		return -EAGAIN;
-	nr = pid->nr;
 	if (unlikely(current->ptrace)) {
 		trace = fork_traceflag (clone_flags);
 		if (trace)
 			clone_flags |= CLONE_PTRACE;
 	}
 
-	p = copy_process(clone_flags, stack_start, regs, stack_size, parent_tidptr, child_tidptr, pid);
+	p = copy_process(clone_flags, stack_start, regs, stack_size,
+			parent_tidptr, child_tidptr, NULL);
 	/*
 	 * Do this prior waking up the new thread - the thread pointer
 	 * might get invalid after that point, if the thread exits quickly.
@@ -1400,6 +1406,8 @@ long do_fork(unsigned long clone_flags,
 	if (!IS_ERR(p)) {
 		struct completion vfork;
 
+		nr = pid_nr(task_pid(p));
+
 		if (clone_flags & CLONE_VFORK) {
 			p->vfork_done = &vfork;
 			init_completion(&vfork);
@@ -1433,7 +1441,6 @@ long do_fork(unsigned long clone_flags,
 			}
 		}
 	} else {
-		free_pid(pid);
 		nr = PTR_ERR(p);
 	}
 	return nr;
-
To unsubscribe from this list: send the line "unsubscribe linux-kernel" in
the body of a message to majordomo@xxxxxxxxxxxxxxx
More majordomo info at  http://vger.kernel.org/majordomo-info.html
Please read the FAQ at  http://www.tux.org/lkml/