Pavel,

I noticed that fcntl tests in LTP fail when LTP is run a child pid
namespace. I just hacked up this quick patch and seems to fix the
failures.

Are you already working on this or do you want me to test and send
this out for review. 

I have one concern that I could use some review/confirmation :-)

Even if the main thread of a process exits before the child threads,
the main thread will not be reaped until all threads exit.  So, the
'task->group_leader' remains valid for the child threads until the
last non-leader thread exits.

IOW the call to task_tgid_vnr() is safe in locks_remove_posix() and
locks_remove_flock().

---
 fs/locks.c |   14 +++++++-------
 1 file changed, 7 insertions(+), 7 deletions(-)

Index: 2.6.23-rc4-mm1/fs/locks.c
===================================================================
--- 2.6.23-rc4-mm1.orig/fs/locks.c	2007-09-05 12:24:30.000000000 -0700
+++ 2.6.23-rc4-mm1/fs/locks.c	2007-09-07 09:14:47.000000000 -0700
@@ -278,7 +278,7 @@ static int flock_make_lock(struct file *
 		return -ENOMEM;
 
 	fl->fl_file = filp;
-	fl->fl_pid = current->tgid;
+	fl->fl_pid = task_tgid_vnr(current);
 	fl->fl_flags = FL_FLOCK;
 	fl->fl_type = type;
 	fl->fl_end = OFFSET_MAX;
@@ -344,7 +344,7 @@ static int flock_to_posix_lock(struct fi
 		return -EOVERFLOW;
 	
 	fl->fl_owner = current->files;
-	fl->fl_pid = current->tgid;
+	fl->fl_pid = task_tgid_vnr(current);
 	fl->fl_file = filp;
 	fl->fl_flags = FL_POSIX;
 	fl->fl_ops = NULL;
@@ -390,7 +390,7 @@ static int flock64_to_posix_lock(struct 
 		return -EOVERFLOW;
 	
 	fl->fl_owner = current->files;
-	fl->fl_pid = current->tgid;
+	fl->fl_pid = task_tgid_vnr(current);
 	fl->fl_file = filp;
 	fl->fl_flags = FL_POSIX;
 	fl->fl_ops = NULL;
@@ -446,7 +446,7 @@ static int lease_init(struct file *filp,
 		return -EINVAL;
 
 	fl->fl_owner = current->files;
-	fl->fl_pid = current->tgid;
+	fl->fl_pid = task_tgid_vnr(current);
 
 	fl->fl_file = filp;
 	fl->fl_flags = FL_LEASE;
@@ -1091,7 +1091,7 @@ int locks_mandatory_area(int read_write,
 
 	locks_init_lock(&fl);
 	fl.fl_owner = current->files;
-	fl.fl_pid = current->tgid;
+	fl.fl_pid = task_tgid_vnr(current);
 	fl.fl_file = filp;
 	fl.fl_flags = FL_POSIX | FL_ACCESS;
 	if (filp && !(filp->f_flags & O_NONBLOCK))
@@ -1963,7 +1963,7 @@ void locks_remove_posix(struct file *fil
 	lock.fl_start = 0;
 	lock.fl_end = OFFSET_MAX;
 	lock.fl_owner = owner;
-	lock.fl_pid = current->tgid;
+	lock.fl_pid = task_tgid_vnr(current);
 	lock.fl_file = filp;
 	lock.fl_ops = NULL;
 	lock.fl_lmops = NULL;
@@ -1990,7 +1990,7 @@ void locks_remove_flock(struct file *fil
 
 	if (filp->f_op && filp->f_op->flock) {
 		struct file_lock fl = {
-			.fl_pid = current->tgid,
+			.fl_pid = task_tgid_vnr(current),
 			.fl_file = filp,
 			.fl_flags = FL_FLOCK,
 			.fl_type = F_UNLCK,
_______________________________________________
Containers mailing list
Containers@lists.linux-foundation.org
https://lists.linux-foundation.org/mailman/listinfo/containers

sukadev@us.ibm.com wrote:
> Pavel,
> 
> I noticed that fcntl tests in LTP fail when LTP is run a child pid
> namespace. I just hacked up this quick patch and seems to fix the
> failures.
> 
> Are you already working on this or do you want me to test and send
> this out for review. 
> 
> I have one concern that I could use some review/confirmation :-)
> 
> Even if the main thread of a process exits before the child threads,
> the main thread will not be reaped until all threads exit.  So, the
> 'task->group_leader' remains valid for the child threads until the
> last non-leader thread exits.
> 
> IOW the call to task_tgid_vnr() is safe in locks_remove_posix() and
> locks_remove_flock().

Well, I missed this place deliberately. This should be converted into
struct pid * usage. I will do the appropriate patch soon.

> ---
>  fs/locks.c |   14 +++++++-------
>  1 file changed, 7 insertions(+), 7 deletions(-)
> 
> Index: 2.6.23-rc4-mm1/fs/locks.c
> ===================================================================
> --- 2.6.23-rc4-mm1.orig/fs/locks.c	2007-09-05 12:24:30.000000000 -0700
> +++ 2.6.23-rc4-mm1/fs/locks.c	2007-09-07 09:14:47.000000000 -0700
> @@ -278,7 +278,7 @@ static int flock_make_lock(struct file *
>  		return -ENOMEM;
>  
>  	fl->fl_file = filp;
> -	fl->fl_pid = current->tgid;
> +	fl->fl_pid = task_tgid_vnr(current);
>  	fl->fl_flags = FL_FLOCK;
>  	fl->fl_type = type;
>  	fl->fl_end = OFFSET_MAX;
> @@ -344,7 +344,7 @@ static int flock_to_posix_lock(struct fi
>  		return -EOVERFLOW;
>  	
>  	fl->fl_owner = current->files;
> -	fl->fl_pid = current->tgid;
> +	fl->fl_pid = task_tgid_vnr(current);
>  	fl->fl_file = filp;
>  	fl->fl_flags = FL_POSIX;
>  	fl->fl_ops = NULL;
> @@ -390,7 +390,7 @@ static int flock64_to_posix_lock(struct 
>  		return -EOVERFLOW;
>  	
>  	fl->fl_owner = current->files;
> -	fl->fl_pid = current->tgid;
> +	fl->fl_pid = task_tgid_vnr(current);
>  	fl->fl_file = filp;
>  	fl->fl_flags = FL_POSIX;
>  	fl->fl_ops = NULL;
> @@ -446,7 +446,7 @@ static int lease_init(struct file *filp,
>  		return -EINVAL;
>  
>  	fl->fl_owner = current->files;
> -	fl->fl_pid = current->tgid;
> +	fl->fl_pid = task_tgid_vnr(current);
>  
>  	fl->fl_file = filp;
>  	fl->fl_flags = FL_LEASE;
> @@ -1091,7 +1091,7 @@ int locks_mandatory_area(int read_write,
>  
>  	locks_init_lock(&fl);
>  	fl.fl_owner = current->files;
> -	fl.fl_pid = current->tgid;
> +	fl.fl_pid = task_tgid_vnr(current);
>  	fl.fl_file = filp;
>  	fl.fl_flags = FL_POSIX | FL_ACCESS;
>  	if (filp && !(filp->f_flags & O_NONBLOCK))
> @@ -1963,7 +1963,7 @@ void locks_remove_posix(struct file *fil
>  	lock.fl_start = 0;
>  	lock.fl_end = OFFSET_MAX;
>  	lock.fl_owner = owner;
> -	lock.fl_pid = current->tgid;
> +	lock.fl_pid = task_tgid_vnr(current);
>  	lock.fl_file = filp;
>  	lock.fl_ops = NULL;
>  	lock.fl_lmops = NULL;
> @@ -1990,7 +1990,7 @@ void locks_remove_flock(struct file *fil
>  
>  	if (filp->f_op && filp->f_op->flock) {
>  		struct file_lock fl = {
> -			.fl_pid = current->tgid,
> +			.fl_pid = task_tgid_vnr(current),
>  			.fl_file = filp,
>  			.fl_flags = FL_FLOCK,
>  			.fl_type = F_UNLCK,
> 

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