Re: [PATCH RFC v2] Convert struct pid count to refcount_t

[Joel Fernandes]

On Mon, Jun 24, 2019 at 02:45:34PM -0400, Joel Fernandes (Google) wrote:
> struct pid's count is an atomic_t field used as a refcount. Use
> refcount_t for it which is basically atomic_t but does additional
> checking to prevent use-after-free bugs.
> 
> For memory ordering, the only change is with the following:
>  -	if ((atomic_read(&pid->count) == 1) ||
>  -	     atomic_dec_and_test(&pid->count)) {
>  +	if (refcount_dec_and_test(&pid->count)) {
>  		kmem_cache_free(ns->pid_cachep, pid);
> 
> Here the change is from:
> Fully ordered --> RELEASE + ACQUIRE (as per refcount-vs-atomic.rst)
> This ACQUIRE should take care of making sure the free happens after the
> refcount_dec_and_test().
> 
> The above hunk also removes atomic_read() since it is not needed for the
> code to work and it is unclear how beneficial it is. The removal lets
> refcount_dec_and_test() check for cases where get_pid() happened before
> the object was freed.
> 
> Cc: jannh@xxxxxxxxxx
> Cc: oleg@xxxxxxxxxx
> Cc: mathieu.desnoyers@xxxxxxxxxxxx
> Cc: willy@xxxxxxxxxxxxx
> Cc: peterz@xxxxxxxxxxxxx
> Cc: will.deacon@xxxxxxx
> Cc: paulmck@xxxxxxxxxxxxxxxxxx
> Cc: elena.reshetova@xxxxxxxxx
> Cc: keescook@xxxxxxxxxxxx
> Cc: kernel-team@xxxxxxxxxxx
> Cc: kernel-hardening@xxxxxxxxxxxxxxxxxx
> Signed-off-by: Joel Fernandes (Google) <joel@xxxxxxxxxxxxxxxxx>
> 
> ---
> Changed to RFC to get any feedback on the memory ordering.

I had a question about refcount_inc().

As per Documentation/core-api/refcount-vs-atomic.rst , it says:

A control dependency (on success) for refcounters guarantees that
if a reference for an object was successfully obtained (reference
counter increment or addition happened, function returned true),
then further stores are ordered against this operation.

However, in refcount_inc() I don't see any memory barriers (in the case where
CONFIG_REFCOUNT_FULL=n). Is the documentation wrong?

get_pid() does a refcount_inc() but doesn't have any memory barriers either.

thanks,

 - Joel


> 
>  include/linux/pid.h | 5 +++--
>  kernel/pid.c        | 7 +++----
>  2 files changed, 6 insertions(+), 6 deletions(-)
> 
> diff --git a/include/linux/pid.h b/include/linux/pid.h
> index 14a9a39da9c7..8cb86d377ff5 100644
> --- a/include/linux/pid.h
> +++ b/include/linux/pid.h
> @@ -3,6 +3,7 @@
>  #define _LINUX_PID_H
>  
>  #include <linux/rculist.h>
> +#include <linux/refcount.h>
>  
>  enum pid_type
>  {
> @@ -56,7 +57,7 @@ struct upid {
>  
>  struct pid
>  {
> -	atomic_t count;
> +	refcount_t count;
>  	unsigned int level;
>  	/* lists of tasks that use this pid */
>  	struct hlist_head tasks[PIDTYPE_MAX];
> @@ -69,7 +70,7 @@ extern struct pid init_struct_pid;
>  static inline struct pid *get_pid(struct pid *pid)
>  {
>  	if (pid)
> -		atomic_inc(&pid->count);
> +		refcount_inc(&pid->count);
>  	return pid;
>  }
>  
> diff --git a/kernel/pid.c b/kernel/pid.c
> index 20881598bdfa..89c4849fab5d 100644
> --- a/kernel/pid.c
> +++ b/kernel/pid.c
> @@ -37,7 +37,7 @@
>  #include <linux/init_task.h>
>  #include <linux/syscalls.h>
>  #include <linux/proc_ns.h>
> -#include <linux/proc_fs.h>
> +#include <linux/refcount.h>
>  #include <linux/sched/task.h>
>  #include <linux/idr.h>
>  
> @@ -106,8 +106,7 @@ void put_pid(struct pid *pid)
>  		return;
>  
>  	ns = pid->numbers[pid->level].ns;
> -	if ((atomic_read(&pid->count) == 1) ||
> -	     atomic_dec_and_test(&pid->count)) {
> +	if (refcount_dec_and_test(&pid->count)) {
>  		kmem_cache_free(ns->pid_cachep, pid);
>  		put_pid_ns(ns);
>  	}
> @@ -210,7 +209,7 @@ struct pid *alloc_pid(struct pid_namespace *ns)
>  	}
>  
>  	get_pid_ns(ns);
> -	atomic_set(&pid->count, 1);
> +	refcount_set(&pid->count, 1);
>  	for (type = 0; type < PIDTYPE_MAX; ++type)
>  		INIT_HLIST_HEAD(&pid->tasks[type]);
>  
> -- 
> 2.22.0.410.gd8fdbe21b5-goog