xen/pvcalls: implement release command
authorStefano Stabellini <sstabellini@kernel.org>
Thu, 6 Jul 2017 18:01:07 +0000 (11:01 -0700)
committerBoris Ostrovsky <boris.ostrovsky@oracle.com>
Thu, 31 Aug 2017 13:45:55 +0000 (09:45 -0400)
Release both active and passive sockets. For active sockets, make sure
to avoid possible conflicts with the ioworker reading/writing to those
sockets concurrently. Set map->release to let the ioworker know
atomically that the socket will be released soon, then wait until the
ioworker finishes (flush_work).

Unmap indexes pages and data rings.

Signed-off-by: Stefano Stabellini <stefano@aporeto.com>
Reviewed-by: Juergen Gross <jgross@suse.com>
CC: boris.ostrovsky@oracle.com
CC: jgross@suse.com
Signed-off-by: Boris Ostrovsky <boris.ostrovsky@oracle.com>
drivers/xen/pvcalls-back.c

index 06bd95d4d552173550d3096be8f28e9553211f20..701e8293dd2823c693430f01831c3abd74203577 100644 (file)
@@ -272,12 +272,80 @@ static int pvcalls_back_release_active(struct xenbus_device *dev,
                                       struct pvcalls_fedata *fedata,
                                       struct sock_mapping *map)
 {
+       disable_irq(map->irq);
+       if (map->sock->sk != NULL) {
+               write_lock_bh(&map->sock->sk->sk_callback_lock);
+               map->sock->sk->sk_user_data = NULL;
+               map->sock->sk->sk_data_ready = map->saved_data_ready;
+               write_unlock_bh(&map->sock->sk->sk_callback_lock);
+       }
+
+       atomic_set(&map->release, 1);
+       flush_work(&map->ioworker.register_work);
+
+       xenbus_unmap_ring_vfree(dev, map->bytes);
+       xenbus_unmap_ring_vfree(dev, (void *)map->ring);
+       unbind_from_irqhandler(map->irq, map);
+
+       sock_release(map->sock);
+       kfree(map);
+
+       return 0;
+}
+
+static int pvcalls_back_release_passive(struct xenbus_device *dev,
+                                       struct pvcalls_fedata *fedata,
+                                       struct sockpass_mapping *mappass)
+{
+       if (mappass->sock->sk != NULL) {
+               write_lock_bh(&mappass->sock->sk->sk_callback_lock);
+               mappass->sock->sk->sk_user_data = NULL;
+               mappass->sock->sk->sk_data_ready = mappass->saved_data_ready;
+               write_unlock_bh(&mappass->sock->sk->sk_callback_lock);
+       }
+       sock_release(mappass->sock);
+       flush_workqueue(mappass->wq);
+       destroy_workqueue(mappass->wq);
+       kfree(mappass);
+
        return 0;
 }
 
 static int pvcalls_back_release(struct xenbus_device *dev,
                                struct xen_pvcalls_request *req)
 {
+       struct pvcalls_fedata *fedata;
+       struct sock_mapping *map, *n;
+       struct sockpass_mapping *mappass;
+       int ret = 0;
+       struct xen_pvcalls_response *rsp;
+
+       fedata = dev_get_drvdata(&dev->dev);
+
+       down(&fedata->socket_lock);
+       list_for_each_entry_safe(map, n, &fedata->socket_mappings, list) {
+               if (map->id == req->u.release.id) {
+                       list_del(&map->list);
+                       up(&fedata->socket_lock);
+                       ret = pvcalls_back_release_active(dev, fedata, map);
+                       goto out;
+               }
+       }
+       mappass = radix_tree_lookup(&fedata->socketpass_mappings,
+                                   req->u.release.id);
+       if (mappass != NULL) {
+               radix_tree_delete(&fedata->socketpass_mappings, mappass->id);
+               up(&fedata->socket_lock);
+               ret = pvcalls_back_release_passive(dev, fedata, mappass);
+       } else
+               up(&fedata->socket_lock);
+
+out:
+       rsp = RING_GET_RESPONSE(&fedata->ring, fedata->ring.rsp_prod_pvt++);
+       rsp->req_id = req->req_id;
+       rsp->u.release.id = req->u.release.id;
+       rsp->cmd = req->cmd;
+       rsp->ret = ret;
        return 0;
 }