Re: [ovs-dev] [PATCH ovs V7 05/24] dpif: Save added ports in a port map for netdev flow api use

[Roi Dayan]

On 14/04/2017 03:26, Joe Stringer wrote:

On 7 April 2017 at 06:12, Roi Dayan <r...@mellanox.com> wrote:

From: Paul Blakey <pa...@mellanox.com>

To use netdev flow offloading api, dpifs needs to iterate over
added ports. This addition inserts the added dpif ports in a hash map,
The map will also be used to translate dpif ports to netdevs.

Signed-off-by: Paul Blakey <pa...@mellanox.com>
Reviewed-by: Roi Dayan <r...@mellanox.com>
Reviewed-by: Simon Horman <simon.hor...@netronome.com>
---
 lib/dpif.c   |  25 ++++++++++++
 lib/dpif.h   |   2 +
 lib/netdev.c | 123 +++++++++++++++++++++++++++++++++++++++++++++++++++++++++++
 lib/netdev.h |   8 ++++
 4 files changed, 158 insertions(+)

diff --git a/lib/dpif.c b/lib/dpif.c
index 1760de8..d45f21d 100644
--- a/lib/dpif.c
+++ b/lib/dpif.c
@@ -355,7 +355,22 @@ do_open(const char *name, const char *type, bool create, 
struct dpif **dpifp)
     error = registered_class->dpif_class->open(registered_class->dpif_class,
                                                name, create, &dpif);
     if (!error) {
+        struct dpif_port_dump port_dump;
+        struct dpif_port dpif_port;
+
         ovs_assert(dpif->dpif_class == registered_class->dpif_class);
+
+        DPIF_PORT_FOR_EACH(&dpif_port, &port_dump, dpif) {
+            struct netdev *netdev;
+            int err = netdev_open(dpif_port.name, dpif_port.type, &netdev);
+
+            if (!err) {
+                netdev_hmap_port_add(netdev, DPIF_HMAP_KEY(dpif), &dpif_port);
+                netdev_close(netdev);
+            } else {
+                VLOG_WARN("could not open netdev %s type %s", name, type);
+            }
+        }
     } else {
         dp_class_unref(registered_class);
     }
@@ -548,6 +563,14 @@ dpif_port_add(struct dpif *dpif, struct netdev *netdev, 
odp_port_t *port_nop)
     if (!error) {
         VLOG_DBG_RL(&dpmsg_rl, "%s: added %s as port %"PRIu32,
                     dpif_name(dpif), netdev_name, port_no);
+
+        /* temp dpif_port, will be cloned in netdev_hmap_port_add */
+        struct dpif_port dpif_port;
+
+        dpif_port.type = CONST_CAST(char *, netdev_get_type(netdev));
+        dpif_port.name = CONST_CAST(char *, netdev_name);
+        dpif_port.port_no = port_no;
+        netdev_hmap_port_add(netdev, DPIF_HMAP_KEY(dpif), &dpif_port);
     } else {
         VLOG_WARN_RL(&error_rl, "%s: failed to add %s as port: %s",
                      dpif_name(dpif), netdev_name, ovs_strerror(error));
@@ -572,6 +595,8 @@ dpif_port_del(struct dpif *dpif, odp_port_t port_no)
     if (!error) {
         VLOG_DBG_RL(&dpmsg_rl, "%s: port_del(%"PRIu32")",
                     dpif_name(dpif), port_no);
+
+        netdev_hmap_port_del(port_no, DPIF_HMAP_KEY(dpif));
     } else {
         log_operation(dpif, "port_del", error);
     }
diff --git a/lib/dpif.h b/lib/dpif.h
index 5d49b11..81376c0 100644
--- a/lib/dpif.h
+++ b/lib/dpif.h
@@ -401,6 +401,8 @@
 extern "C" {
 #endif

+#define DPIF_HMAP_KEY(x) ((x)->dpif_class)
+
 struct dpif;
 struct dpif_class;
 struct dpif_flow;
diff --git a/lib/netdev.c b/lib/netdev.c
index 6bde14b..a4de0a9 100644
--- a/lib/netdev.c
+++ b/lib/netdev.c
@@ -2104,6 +2104,129 @@ netdev_init_flow_api(struct netdev *netdev)
             : EOPNOTSUPP);
 }

+/* Protects below port hashmaps. */
+static struct ovs_mutex netdev_hmap_mutex = OVS_MUTEX_INITIALIZER;
+
+static struct hmap port_to_netdev OVS_GUARDED_BY(netdev_hmap_mutex)
+    = HMAP_INITIALIZER(&port_to_netdev);
+static struct hmap ifindex_to_port OVS_GUARDED_BY(netdev_hmap_mutex)
+    = HMAP_INITIALIZER(&ifindex_to_port);
+
+struct port_to_netdev_data {
+    struct hmap_node node;
+    struct netdev *netdev;
+    struct dpif_port dpif_port;
+    const void *obj;
+};
+
+struct ifindex_to_port_data {
+    struct hmap_node node;
+    int ifindex;
+    odp_port_t port;
+};
+
+static struct port_to_netdev_data *
+netdev_hmap_port_lookup(odp_port_t port_no, const void *obj)
+{
+    size_t hash = hash_int(odp_to_u32(port_no), hash_pointer(obj, 0));
+    struct port_to_netdev_data *data;
+
+    HMAP_FOR_EACH_WITH_HASH(data, node, hash, &port_to_netdev) {
+            if (data->obj == obj && data->dpif_port.port_no == port_no) {
+                return data;
+            }
+    }
+    return NULL;
+}
+
+int
+netdev_hmap_port_add(struct netdev *netdev, const void *obj,
+                     struct dpif_port *dpif_port)
+{
+    size_t hash = hash_int(odp_to_u32(dpif_port->port_no),
+                           hash_pointer(obj, 0));
+    struct port_to_netdev_data *data = xzalloc(sizeof *data);
+    struct ifindex_to_port_data *ifidx = xzalloc(sizeof *ifidx);
+
+    ovs_mutex_lock(&netdev_hmap_mutex);
+    if (netdev_hmap_port_lookup(dpif_port->port_no, obj)) {
+        ovs_mutex_unlock(&netdev_hmap_mutex);
+        return EEXIST;
+    }
+
+    data->netdev = netdev_ref(netdev);
+    data->obj = obj;
+    dpif_port_clone(&data->dpif_port, dpif_port);
+
+    ifidx->ifindex = netdev_get_ifindex(netdev);

I think that this function can return -EOPNOTSUPP, which is not handled.

right. thanks.

+    ifidx->port = dpif_port->port_no;
+
+    hmap_insert(&port_to_netdev, &data->node, hash);
+    hmap_insert(&ifindex_to_port, &ifidx->node, ifidx->ifindex);
+    ovs_mutex_unlock(&netdev_hmap_mutex);
+
+    netdev_init_flow_api(netdev);
+
+    return 0;
+}
+
+struct netdev *
+netdev_hmap_port_get(odp_port_t port_no, const void *obj)
+{
+    struct port_to_netdev_data *data;
+    struct netdev *ret = NULL;
+
+    ovs_mutex_lock(&netdev_hmap_mutex);
+    data = netdev_hmap_port_lookup(port_no, obj);
+    if (data) {
+        ret = netdev_ref(data->netdev);
+    }
+    ovs_mutex_unlock(&netdev_hmap_mutex);
+
+    return ret;
+}
+
+odp_port_t
+netdev_hmap_port_get_byifidx(int ifindex)
+{
+    struct ifindex_to_port_data *data;
+    odp_port_t ret = 0;
+
+    ovs_mutex_lock(&netdev_hmap_mutex);
+    HMAP_FOR_EACH_WITH_HASH(data, node, ifindex, &ifindex_to_port) {
+            if (data->ifindex == ifindex) {
+                ret = data->port;
+                break;
+            }
+    }
+    ovs_mutex_unlock(&netdev_hmap_mutex);
+
+    return ret;
+}
+
+int
+netdev_hmap_port_del(odp_port_t port_no, const void *obj)
+{
+    struct port_to_netdev_data *data;
+    int ret = ENOENT;
+
+    ovs_mutex_lock(&netdev_hmap_mutex);
+
+    data = netdev_hmap_port_lookup(port_no, obj);
+
+    if (data) {
+        dpif_port_destroy(&data->dpif_port);
+        netdev_close(data->netdev); /* unref and possibly close */
+        hmap_remove(&port_to_netdev, &data->node);
+        free(data);
+        ret = 0;
+    }
+
+    ovs_mutex_unlock(&netdev_hmap_mutex);
+
+    return ret;
+}
+
 bool netdev_flow_api_enabled = false;

 #ifdef __linux__
diff --git a/lib/netdev.h b/lib/netdev.h
index e2d1799..f494ea3 100644
--- a/lib/netdev.h
+++ b/lib/netdev.h
@@ -181,6 +181,14 @@ int netdev_init_flow_api(struct netdev *);
 extern bool netdev_flow_api_enabled;
 void netdev_set_flow_api_enabled(const struct smap *ovs_other_config);

+struct dpif_port;
+int netdev_hmap_port_add(struct netdev *, const void *obj, struct dpif_port *);
+struct netdev *netdev_hmap_port_get(odp_port_t port, const void *obj);
+int netdev_hmap_port_del(odp_port_t port, const void *obj);
+struct netdev_flow_dump **netdev_ports_flow_dumps_create(const void *obj,
+                                                         int *ports);
+odp_port_t netdev_hmap_port_get_byifidx(int ifindex);

It seems like 'obj' is trying to go half way on object-orientation in
these functions. 'obj' doesn't describe very well the semantics of how
that variable is passed and used. I feel like it'd be easier to follow
what's going on with these functions if there were a 'struct
netdev_ports' which provided the container and locking for the map. It
could be attached to the dpif, then these functions could each receive
one of these 'struct netdev_ports *' structures that would more
clearly put a box around what 'obj' is trying to contain. Then we also
get some nice compiler type help which we don't get by passing void
pointers around.

The naming could be a bit more consistent for the add/get/del/dump -
common prefix netdev_ports_* perhaps?

netdev_ports_insert()
netdev_ports_lookup()
netdev_ports_remove()
netdev_ports_create_flow_dump()

The last function name feels a bit squashed, if you're trying to keep
the names short then maybe it'd be easier to read as something like
netdev_get_port_by_ifindex() or netdev_ifindex_to_odp_port()?

sounds good. thanks.
_______________________________________________
dev mailing list
d...@openvswitch.org
https://mail.openvswitch.org/mailman/listinfo/ovs-dev