X-Git-Url: https://pintos-os.org/cgi-bin/gitweb.cgi?a=blobdiff_plain;f=lib%2Fdpif-netdev.c;h=eddd18bef0a4da88123724073017d77eb309c6bd;hb=05abb769d32d440dbd521e99484d8418f95431bd;hp=2fb13391590884d8092c4ca9f0409a82a61163fc;hpb=14608a1539b73f8f9812e0e791adb60825fee38b;p=openvswitch

diff --git a/lib/dpif-netdev.c b/lib/dpif-netdev.c
index 2fb13391..eddd18be 100644
--- a/lib/dpif-netdev.c
+++ b/lib/dpif-netdev.c
@@ -32,7 +32,9 @@
 #include <unistd.h>
 
 #include "csum.h"
+#include "dpif.h"
 #include "dpif-provider.h"
+#include "dummy.h"
 #include "flow.h"
 #include "hmap.h"
 #include "list.h"
@@ -42,12 +44,12 @@
 #include "ofpbuf.h"
 #include "packets.h"
 #include "poll-loop.h"
-#include "queue.h"
+#include "shash.h"
 #include "timeval.h"
 #include "util.h"
 #include "vlog.h"
 
-VLOG_DEFINE_THIS_MODULE(dpif_netdev)
+VLOG_DEFINE_THIS_MODULE(dpif_netdev);
 
 /* Configuration parameters. */
 enum { N_QUEUES = 2 };          /* Number of queues for dpif_recv(). */
@@ -61,13 +63,14 @@ enum { DP_NETDEV_HEADROOM = 2 + VLAN_HEADER_LEN };
 
 /* Datapath based on the network device interface from netdev.h. */
 struct dp_netdev {
-    struct list node;
-    int dp_idx;
+    const struct dpif_class *class;
+    char *name;
     int open_cnt;
     bool destroyed;
 
     bool drop_frags;            /* Drop all IP fragments, if true. */
-    struct ovs_queue queues[N_QUEUES]; /* Messages queued for dpif_recv(). */
+    struct list queues[N_QUEUES]; /* Contain ofpbufs queued for dpif_recv(). */
+    size_t queue_len[N_QUEUES]; /* Number of packets in each queue. */
     struct hmap flow_table;     /* Flow table. */
 
     /* Statistics. */
@@ -88,7 +91,7 @@ struct dp_netdev_port {
     int port_no;                /* Index into dp_netdev's 'ports'. */
     struct list node;           /* Element in dp_netdev's 'port_list'. */
     struct netdev *netdev;
-    bool internal;              /* Internal port (as ODP_PORT_INTERNAL)? */
+    bool internal;              /* Internal port? */
 };
 
 /* A flow in dp_netdev's 'flow_table'. */
@@ -116,9 +119,7 @@ struct dpif_netdev {
 };
 
 /* All netdev-based datapaths. */
-static struct dp_netdev *dp_netdevs[256];
-struct list dp_netdev_list = LIST_INITIALIZER(&dp_netdev_list);
-enum { N_DP_NETDEVS = ARRAY_SIZE(dp_netdevs) };
+static struct shash dp_netdevs = SHASH_INITIALIZER(&dp_netdevs);
 
 /* Maximum port MTU seen so far. */
 static int max_mtu = ETH_PAYLOAD_MAX;
@@ -129,19 +130,23 @@ static int get_port_by_name(struct dp_netdev *, const char *devname,
                             struct dp_netdev_port **portp);
 static void dp_netdev_free(struct dp_netdev *);
 static void dp_netdev_flow_flush(struct dp_netdev *);
-static int do_add_port(struct dp_netdev *, const char *devname, uint16_t flags,
-                       uint16_t port_no);
+static int do_add_port(struct dp_netdev *, const char *devname,
+                       const char *type, uint16_t port_no);
 static int do_del_port(struct dp_netdev *, uint16_t port_no);
+static int dpif_netdev_open(const struct dpif_class *, const char *name,
+                            bool create, struct dpif **);
 static int dp_netdev_output_control(struct dp_netdev *, const struct ofpbuf *,
                                     int queue_no, int port_no, uint32_t arg);
 static int dp_netdev_execute_actions(struct dp_netdev *,
                                      struct ofpbuf *, struct flow *,
                                      const union odp_action *, int n);
 
+static struct dpif_class dpif_dummy_class;
+
 static struct dpif_netdev *
 dpif_netdev_cast(const struct dpif *dpif)
 {
-    dpif_assert_class(dpif, &dpif_netdev_class);
+    assert(dpif->dpif_class->open == dpif_netdev_open);
     return CONTAINER_OF(dpif, struct dpif_netdev, dpif);
 }
 
@@ -151,125 +156,80 @@ get_dp_netdev(const struct dpif *dpif)
     return dpif_netdev_cast(dpif)->dp;
 }
 
-static int
-name_to_dp_idx(const char *name)
-{
-    if (!strncmp(name, "dp", 2) && isdigit((unsigned char)name[2])) {
-        int dp_idx = atoi(name + 2);
-        if (dp_idx >= 0 && dp_idx < N_DP_NETDEVS) {
-            return dp_idx;
-        }
-    }
-    return -1;
-}
-
-static struct dp_netdev *
-find_dp_netdev(const char *name)
-{
-    int dp_idx;
-    size_t i;
-
-    dp_idx = name_to_dp_idx(name);
-    if (dp_idx >= 0) {
-        return dp_netdevs[dp_idx];
-    }
-
-    for (i = 0; i < N_DP_NETDEVS; i++) {
-        struct dp_netdev *dp = dp_netdevs[i];
-        if (dp) {
-            struct dp_netdev_port *port;
-            if (!get_port_by_name(dp, name, &port)) {
-                return dp;
-            }
-        }
-    }
-    return NULL;
-}
-
 static struct dpif *
 create_dpif_netdev(struct dp_netdev *dp)
 {
+    uint16_t netflow_id = hash_string(dp->name, 0);
     struct dpif_netdev *dpif;
-    char *dpname;
 
     dp->open_cnt++;
 
-    dpname = xasprintf("dp%d", dp->dp_idx);
     dpif = xmalloc(sizeof *dpif);
-    dpif_init(&dpif->dpif, &dpif_netdev_class, dpname, dp->dp_idx, dp->dp_idx);
+    dpif_init(&dpif->dpif, dp->class, dp->name, netflow_id >> 8, netflow_id);
     dpif->dp = dp;
     dpif->listen_mask = 0;
     dpif->dp_serial = dp->serial;
-    free(dpname);
 
     return &dpif->dpif;
 }
 
 static int
-create_dp_netdev(const char *name, int dp_idx, struct dpif **dpifp)
+create_dp_netdev(const char *name, const struct dpif_class *class,
+                 struct dp_netdev **dpp)
 {
     struct dp_netdev *dp;
     int error;
     int i;
 
-    if (dp_netdevs[dp_idx]) {
-        return EBUSY;
-    }
-
-    /* Create datapath. */
-    dp_netdevs[dp_idx] = dp = xzalloc(sizeof *dp);
-    list_push_back(&dp_netdev_list, &dp->node);
-    dp->dp_idx = dp_idx;
+    dp = xzalloc(sizeof *dp);
+    dp->class = class;
+    dp->name = xstrdup(name);
     dp->open_cnt = 0;
     dp->drop_frags = false;
     for (i = 0; i < N_QUEUES; i++) {
-        queue_init(&dp->queues[i]);
+        list_init(&dp->queues[i]);
     }
     hmap_init(&dp->flow_table);
     list_init(&dp->port_list);
-    error = do_add_port(dp, name, ODP_PORT_INTERNAL, ODPP_LOCAL);
+    error = do_add_port(dp, name, "internal", ODPP_LOCAL);
     if (error) {
         dp_netdev_free(dp);
-        return ENODEV;
+        return error;
     }
 
-    *dpifp = create_dpif_netdev(dp);
+    shash_add(&dp_netdevs, name, dp);
+
+    *dpp = dp;
     return 0;
 }
 
 static int
-dpif_netdev_open(const char *name, const char *type OVS_UNUSED, bool create,
-                 struct dpif **dpifp)
+dpif_netdev_open(const struct dpif_class *class, const char *name,
+                 bool create, struct dpif **dpifp)
 {
-    if (create) {
-        if (find_dp_netdev(name)) {
-            return EEXIST;
-        } else {
-            int dp_idx = name_to_dp_idx(name);
-            if (dp_idx >= 0) {
-                return create_dp_netdev(name, dp_idx, dpifp);
-            } else {
-                /* Scan for unused dp_idx number. */
-                for (dp_idx = 0; dp_idx < N_DP_NETDEVS; dp_idx++) {
-                    int error = create_dp_netdev(name, dp_idx, dpifp);
-                    if (error != EBUSY) {
-                        return error;
-                    }
-                }
+    struct dp_netdev *dp;
 
-                /* All datapath numbers in use. */
-                return ENOBUFS;
+    dp = shash_find_data(&dp_netdevs, name);
+    if (!dp) {
+        if (!create) {
+            return ENODEV;
+        } else {
+            int error = create_dp_netdev(name, class, &dp);
+            if (error) {
+                return error;
             }
+            assert(dp != NULL);
         }
     } else {
-        struct dp_netdev *dp = find_dp_netdev(name);
-        if (dp) {
-            *dpifp = create_dpif_netdev(dp);
-            return 0;
-        } else {
-            return ENODEV;
+        if (dp->class != class) {
+            return EINVAL;
+        } else if (create) {
+            return EEXIST;
         }
     }
+
+    *dpifp = create_dpif_netdev(dp);
+    return 0;
 }
 
 static void
@@ -284,11 +244,10 @@ dp_netdev_free(struct dp_netdev *dp)
         do_del_port(dp, port->port_no);
     }
     for (i = 0; i < N_QUEUES; i++) {
-        queue_destroy(&dp->queues[i]);
+        ofpbuf_list_delete(&dp->queues[i]);
     }
     hmap_destroy(&dp->flow_table);
-    dp_netdevs[dp->dp_idx] = NULL;
-    list_remove(&dp->node);
+    free(dp->name);
     free(dp);
 }
 
@@ -298,6 +257,7 @@ dpif_netdev_close(struct dpif *dpif)
     struct dp_netdev *dp = get_dp_netdev(dpif);
     assert(dp->open_cnt > 0);
     if (--dp->open_cnt == 0 && dp->destroyed) {
+        shash_find_and_delete(&dp_netdevs, dp->name);
         dp_netdev_free(dp);
     }
     free(dpif);
@@ -347,23 +307,33 @@ dpif_netdev_set_drop_frags(struct dpif *dpif, bool drop_frags)
 }
 
 static int
-do_add_port(struct dp_netdev *dp, const char *devname, uint16_t flags,
+do_add_port(struct dp_netdev *dp, const char *devname, const char *type,
             uint16_t port_no)
 {
-    bool internal = (flags & ODP_PORT_INTERNAL) != 0;
     struct dp_netdev_port *port;
     struct netdev_options netdev_options;
     struct netdev *netdev;
+    bool internal;
     int mtu;
     int error;
 
     /* XXX reject devices already in some dp_netdev. */
+    if (type[0] == '\0' || !strcmp(type, "system")) {
+        internal = false;
+    } else if (!strcmp(type, "internal")) {
+        internal = true;
+    } else {
+        VLOG_WARN("%s: unsupported port type %s", devname, type);
+        return EINVAL;
+    }
 
     /* Open and validate network device. */
     memset(&netdev_options, 0, sizeof netdev_options);
     netdev_options.name = devname;
     netdev_options.ethertype = NETDEV_ETH_TYPE_ANY;
-    if (internal) {
+    if (dp->class == &dpif_dummy_class) {
+        netdev_options.type = "dummy";
+    } else if (internal) {
         netdev_options.type = "tap";
     }
 
@@ -399,7 +369,7 @@ do_add_port(struct dp_netdev *dp, const char *devname, uint16_t flags,
 }
 
 static int
-dpif_netdev_port_add(struct dpif *dpif, const char *devname, uint16_t flags,
+dpif_netdev_port_add(struct dpif *dpif, struct netdev *netdev,
                      uint16_t *port_nop)
 {
     struct dp_netdev *dp = get_dp_netdev(dpif);
@@ -408,7 +378,8 @@ dpif_netdev_port_add(struct dpif *dpif, const char *devname, uint16_t flags,
     for (port_no = 0; port_no < MAX_PORTS; port_no++) {
         if (!dp->ports[port_no]) {
             *port_nop = port_no;
-            return do_add_port(dp, devname, flags, port_no);
+            return do_add_port(dp, netdev_get_name(netdev),
+                               netdev_get_type(netdev), port_no);
         }
     }
     return EFBIG;
@@ -488,7 +459,7 @@ answer_port_query(const struct dp_netdev_port *port, struct odp_port *odp_port)
     ovs_strlcpy(odp_port->devname, netdev_get_name(port->netdev),
                 sizeof odp_port->devname);
     odp_port->port = port->port_no;
-    odp_port->flags = port->internal ? ODP_PORT_INTERNAL : 0;
+    strcpy(odp_port->type, port->internal ? "internal" : "system");
 }
 
 static int
@@ -665,16 +636,9 @@ dpif_netdev_validate_actions(const union odp_action *actions, int n_actions,
         case ODPAT_CONTROLLER:
             break;
 
-        case ODPAT_SET_VLAN_VID:
-            *mutates = true;
-            if (a->vlan_vid.vlan_vid & htons(~VLAN_VID_MASK)) {
-                return EINVAL;
-            }
-            break;
-
-        case ODPAT_SET_VLAN_PCP:
+        case ODPAT_SET_DL_TCI:
             *mutates = true;
-            if (a->vlan_pcp.vlan_pcp & ~(VLAN_PCP_MASK >> VLAN_PCP_SHIFT)) {
+            if (a->dl_tci.tci & htons(VLAN_CFI)) {
                 return EINVAL;
             }
             break;
@@ -835,7 +799,7 @@ dpif_netdev_execute(struct dpif *dpif,
     struct dp_netdev *dp = get_dp_netdev(dpif);
     struct ofpbuf copy;
     bool mutates;
-    flow_t key;
+    struct flow key;
     int error;
 
     if (packet->size < ETH_HEADER_LEN || packet->size > UINT16_MAX) {
@@ -851,7 +815,7 @@ dpif_netdev_execute(struct dpif *dpif,
         /* We need a deep copy of 'packet' since we're going to modify its
          * data. */
         ofpbuf_init(&copy, DP_NETDEV_HEADROOM + packet->size);
-        copy.data = (char*)copy.base + DP_NETDEV_HEADROOM;
+        ofpbuf_reserve(&copy, DP_NETDEV_HEADROOM);
         ofpbuf_put(&copy, packet->data, packet->size);
     } else {
         /* We still need a shallow copy of 'packet', even though we won't
@@ -888,7 +852,7 @@ dpif_netdev_recv_set_mask(struct dpif *dpif, int listen_mask)
     }
 }
 
-static struct ovs_queue *
+static int
 find_nonempty_queue(struct dpif *dpif)
 {
     struct dpif_netdev *dpif_netdev = dpif_netdev_cast(dpif);
@@ -897,20 +861,24 @@ find_nonempty_queue(struct dpif *dpif)
     int i;
 
     for (i = 0; i < N_QUEUES; i++) {
-        struct ovs_queue *q = &dp->queues[i];
-        if (q->n && mask & (1u << i)) {
-            return q;
+        struct list *queue = &dp->queues[i];
+        if (!list_is_empty(queue) && mask & (1u << i)) {
+            return i;
         }
     }
-    return NULL;
+    return -1;
 }
 
 static int
 dpif_netdev_recv(struct dpif *dpif, struct ofpbuf **bufp)
 {
-    struct ovs_queue *q = find_nonempty_queue(dpif);
-    if (q) {
-        *bufp = queue_pop_head(q);
+    int queue_idx = find_nonempty_queue(dpif);
+    if (queue_idx >= 0) {
+        struct dp_netdev *dp = get_dp_netdev(dpif);
+
+        *bufp = ofpbuf_from_list(list_pop_front(&dp->queues[queue_idx]));
+        dp->queue_len[queue_idx]--;
+
         return 0;
     } else {
         return EAGAIN;
@@ -920,8 +888,7 @@ dpif_netdev_recv(struct dpif *dpif, struct ofpbuf **bufp)
 static void
 dpif_netdev_recv_wait(struct dpif *dpif)
 {
-    struct ovs_queue *q = find_nonempty_queue(dpif);
-    if (q) {
+    if (find_nonempty_queue(dpif) >= 0) {
         poll_immediate_wake();
     } else {
         /* No messages ready to be received, and dp_wait() will ensure that we
@@ -972,24 +939,25 @@ dp_netdev_port_input(struct dp_netdev *dp, struct dp_netdev_port *port,
 static void
 dp_netdev_run(void)
 {
+    struct shash_node *node;
     struct ofpbuf packet;
-    struct dp_netdev *dp;
 
     ofpbuf_init(&packet, DP_NETDEV_HEADROOM + max_mtu);
-    LIST_FOR_EACH (dp, node, &dp_netdev_list) {
+    SHASH_FOR_EACH (node, &dp_netdevs) {
+        struct dp_netdev *dp = node->data;
         struct dp_netdev_port *port;
 
         LIST_FOR_EACH (port, node, &dp->port_list) {
             int error;
 
             /* Reset packet contents. */
-            packet.data = (char*)packet.base + DP_NETDEV_HEADROOM;
-            packet.size = 0;
+            ofpbuf_clear(&packet);
+            ofpbuf_reserve(&packet, DP_NETDEV_HEADROOM);
 
             error = netdev_recv(port->netdev, &packet);
             if (!error) {
                 dp_netdev_port_input(dp, port, &packet);
-            } else if (error != EAGAIN) {
+            } else if (error != EAGAIN && error != EOPNOTSUPP) {
                 struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 5);
                 VLOG_ERR_RL(&rl, "error receiving data from %s: %s",
                             netdev_get_name(port->netdev), strerror(error));
@@ -1002,10 +970,12 @@ dp_netdev_run(void)
 static void
 dp_netdev_wait(void)
 {
-    struct dp_netdev *dp;
+    struct shash_node *node;
 
-    LIST_FOR_EACH (dp, node, &dp_netdev_list) {
+    SHASH_FOR_EACH (node, &dp_netdevs) {
+        struct dp_netdev *dp = node->data;
         struct dp_netdev_port *port;
+
         LIST_FOR_EACH (port, node, &dp->port_list) {
             netdev_recv_wait(port->netdev);
         }
@@ -1013,16 +983,12 @@ dp_netdev_wait(void)
 }
 
 
-/* Modify the TCI field of 'packet'.  If a VLAN tag is not present, one
- * is added with the TCI field set to 'tci'.  If a VLAN tag is present,
- * then 'mask' bits are cleared before 'tci' is logically OR'd into the
- * TCI field.
- *
- * Note that the function does not ensure that 'tci' does not affect
- * bits outside of 'mask'.
+/* Modify the TCI field of 'packet'.  If a VLAN tag is present, its TCI field
+ * is replaced by 'tci'.  If a VLAN tag is not present, one is added with the
+ * TCI field set to 'tci'.
  */
 static void
-dp_netdev_modify_vlan_tci(struct ofpbuf *packet, uint16_t tci, uint16_t mask)
+dp_netdev_set_dl_tci(struct ofpbuf *packet, uint16_t tci)
 {
     struct vlan_eth_header *veh;
     struct eth_header *eh;
@@ -1030,17 +996,15 @@ dp_netdev_modify_vlan_tci(struct ofpbuf *packet, uint16_t tci, uint16_t mask)
     eh = packet->l2;
     if (packet->size >= sizeof(struct vlan_eth_header)
         && eh->eth_type == htons(ETH_TYPE_VLAN)) {
-        /* Clear 'mask' bits, but maintain other TCI bits. */
         veh = packet->l2;
-        veh->veth_tci &= ~htons(mask);
-        veh->veth_tci |= htons(tci);
+        veh->veth_tci = tci;
     } else {
         /* Insert new 802.1Q header. */
         struct vlan_eth_header tmp;
         memcpy(tmp.veth_dst, eh->eth_dst, ETH_ADDR_LEN);
         memcpy(tmp.veth_src, eh->eth_src, ETH_ADDR_LEN);
         tmp.veth_type = htons(ETH_TYPE_VLAN);
-        tmp.veth_tci = htons(tci);
+        tmp.veth_tci = tci;
         tmp.veth_next_type = eh->eth_type;
 
         veh = ofpbuf_push_uninit(packet, VLAN_HEADER_LEN);
@@ -1061,8 +1025,7 @@ dp_netdev_strip_vlan(struct ofpbuf *packet)
         memcpy(tmp.eth_src, veh->veth_src, ETH_ADDR_LEN);
         tmp.eth_type = veh->veth_next_type;
 
-        packet->size -= VLAN_HEADER_LEN;
-        packet->data = (char*)packet->data + VLAN_HEADER_LEN;
+        ofpbuf_pull(packet, VLAN_HEADER_LEN);
         packet->l2 = (char*)packet->l2 + VLAN_HEADER_LEN;
         memcpy(packet->data, &tmp, sizeof tmp);
     }
@@ -1167,12 +1130,11 @@ static int
 dp_netdev_output_control(struct dp_netdev *dp, const struct ofpbuf *packet,
                          int queue_no, int port_no, uint32_t arg)
 {
-    struct ovs_queue *q = &dp->queues[queue_no];
     struct odp_msg *header;
     struct ofpbuf *msg;
     size_t msg_size;
 
-    if (q->n >= MAX_QUEUE_LEN) {
+    if (dp->queue_len[queue_no] >= MAX_QUEUE_LEN) {
         dp->n_lost++;
         return ENOBUFS;
     }
@@ -1185,7 +1147,8 @@ dp_netdev_output_control(struct dp_netdev *dp, const struct ofpbuf *packet,
     header->port = port_no;
     header->arg = arg;
     ofpbuf_put(msg, packet->data, packet->size);
-    queue_push_tail(q, msg);
+    list_push_back(&dp->queues[queue_no], &msg->list_node);
+    dp->queue_len[queue_no]++;
 
     return 0;
 }
@@ -1237,15 +1200,8 @@ dp_netdev_execute_actions(struct dp_netdev *dp,
                                      key->in_port, a->controller.arg);
             break;
 
-        case ODPAT_SET_VLAN_VID:
-            dp_netdev_modify_vlan_tci(packet, ntohs(a->vlan_vid.vlan_vid),
-                                      VLAN_VID_MASK);
-             break;
-
-        case ODPAT_SET_VLAN_PCP:
-            dp_netdev_modify_vlan_tci(packet,
-                                      a->vlan_pcp.vlan_pcp << VLAN_PCP_SHIFT,
-                                      VLAN_PCP_MASK);
+        case ODPAT_SET_DL_TCI:
+            dp_netdev_set_dl_tci(packet, a->dl_tci.tci);
             break;
 
         case ODPAT_STRIP_VLAN:
@@ -1316,3 +1272,13 @@ const struct dpif_class dpif_netdev_class = {
     dpif_netdev_recv,
     dpif_netdev_recv_wait,
 };
+
+void
+dpif_dummy_register(void)
+{
+    if (!dpif_dummy_class.type) {
+        dpif_dummy_class = dpif_netdev_class;
+        dpif_dummy_class.type = "dummy";
+        dp_register_provider(&dpif_dummy_class);
+    }
+}