[v2] vhost: avoid sleeping under mutex

  Covscan reported:

 2. dpdk-21.11/lib/vhost/socket.c:852: lock_acquire: Calling function
	"pthread_mutex_lock" acquires lock "vhost_user.mutex".
23. dpdk-21.11/lib/vhost/socket.c:955: sleep: Call to
	"vhost_user_reconnect_init" might sleep while holding lock
	"vhost_user.mutex".
 #   953|   		vsocket->reconnect =
	!(flags & RTE_VHOST_USER_NO_RECONNECT);
 #   954|   		if (vsocket->reconnect && reconn_tid == 0) {
 #   955|-> 			if (vhost_user_reconnect_init() != 0)
 #   956|   				goto out_mutex;
 #   957|   		}

The reason for this warning is that vhost_user_reconnect_init() creates a
ctrl thread and calls nanosleep waiting for this thread to be ready,
while vhost_user.mutex is taken.

Move the call to vhost_user_reconnect_init() out of this mutex.

While at it, a pthread_t value should be considered opaque.
Instead of relying reconn_tid == 0, use an internal flag in
vhost_user_reconnect_init().

Coverity issue: 373686
Bugzilla ID: 981
Fixes: e623e0c6d8a5 ("vhost: add reconnect ability")
Cc: stable@dpdk.org

Signed-off-by: David Marchand <david.marchand@redhat.com>
---
Changes since v1:
- moved reconn_tid in vhost_user_reconnect_init as this
  variable is not used anywhere else,

---
 lib/vhost/socket.c | 19 +++++++++++++------
 1 file changed, 13 insertions(+), 6 deletions(-)
  

On Mon, 15 May 2023 13:18:44 +0200
David Marchand <david.marchand@redhat.com> wrote:

> Covscan reported:
> 
>  2. dpdk-21.11/lib/vhost/socket.c:852: lock_acquire: Calling function
> 	"pthread_mutex_lock" acquires lock "vhost_user.mutex".
> 23. dpdk-21.11/lib/vhost/socket.c:955: sleep: Call to
> 	"vhost_user_reconnect_init" might sleep while holding lock
> 	"vhost_user.mutex".
>  #   953|   		vsocket->reconnect =
> 	!(flags & RTE_VHOST_USER_NO_RECONNECT);
>  #   954|   		if (vsocket->reconnect && reconn_tid == 0) {
>  #   955|-> 			if (vhost_user_reconnect_init() != 0)
>  #   956|   				goto out_mutex;
>  #   957|   		}
> 
> The reason for this warning is that vhost_user_reconnect_init() creates a
> ctrl thread and calls nanosleep waiting for this thread to be ready,
> while vhost_user.mutex is taken.
> 
> Move the call to vhost_user_reconnect_init() out of this mutex.
> 
> While at it, a pthread_t value should be considered opaque.
> Instead of relying reconn_tid == 0, use an internal flag in
> vhost_user_reconnect_init().
> 
> Coverity issue: 373686
> Bugzilla ID: 981
> Fixes: e623e0c6d8a5 ("vhost: add reconnect ability")
> Cc: stable@dpdk.org
> 
> Signed-off-by: David Marchand <david.marchand@redhat.com>
> ---
> Changes since v1:
> - moved reconn_tid in vhost_user_reconnect_init as this
>   variable is not used anywhere else,
> 
> ---
>  lib/vhost/socket.c | 19 +++++++++++++------
>  1 file changed, 13 insertions(+), 6 deletions(-)
> 
> diff --git a/lib/vhost/socket.c b/lib/vhost/socket.c
> index 669c322e12..00a912c59e 100644
> --- a/lib/vhost/socket.c
> +++ b/lib/vhost/socket.c
> @@ -427,7 +427,6 @@ struct vhost_user_reconnect_list {
>  };
>  
>  static struct vhost_user_reconnect_list reconn_list;
> -static pthread_t reconn_tid;
>  
>  static int
>  vhost_user_connect_nonblock(char *path, int fd, struct sockaddr *un, size_t sz)
> @@ -498,8 +497,13 @@ vhost_user_client_reconnect(void *arg __rte_unused)
>  static int
>  vhost_user_reconnect_init(void)
>  {
> +	static bool reconn_init_done;
> +	static pthread_t reconn_tid;

This would not be thread safe.
Perhaps use a pthread condition variable instead of spinning here.