#include "packets.h"
#include "poll-loop.h"
#include "random.h"
+#include "seq.h"
#include "shash.h"
#include "sset.h"
#include "timeval.h"
+#include "unixctl.h"
#include "util.h"
#include "vlog.h"
char *name;
int open_cnt;
bool destroyed;
+ int max_mtu; /* Maximum MTU of any port added so far. */
struct dp_netdev_queue queues[N_QUEUES];
struct hmap flow_table; /* Flow table. */
+ struct seq *queue_seq; /* Incremented whenever a packet is queued. */
/* Statistics. */
long long int n_hit; /* Number of flow table matches. */
/* Ports. */
struct dp_netdev_port *ports[MAX_PORTS];
struct list port_list;
- unsigned int serial;
+ struct seq *port_seq; /* Incremented whenever a port changes. */
};
/* A port in a netdev-based datapath. */
struct dpif_netdev {
struct dpif dpif;
struct dp_netdev *dp;
- unsigned int dp_serial;
+ uint64_t last_port_seq;
};
/* All netdev-based datapaths. */
static struct shash dp_netdevs = SHASH_INITIALIZER(&dp_netdevs);
-/* Maximum port MTU seen so far. */
-static int max_mtu = ETH_PAYLOAD_MAX;
+/* Global lock for all data. */
+static struct ovs_mutex dp_netdev_mutex = OVS_MUTEX_INITIALIZER;
static int get_port_by_number(struct dp_netdev *, odp_port_t port_no,
struct dp_netdev_port **portp);
static void dp_netdev_port_input(struct dp_netdev *dp,
struct dp_netdev_port *port,
struct ofpbuf *packet, uint32_t skb_priority,
- uint32_t skb_mark, const struct flow_tnl *tnl);
+ uint32_t pkt_mark, const struct flow_tnl *tnl);
static struct dpif_netdev *
dpif_netdev_cast(const struct dpif *dpif)
{
struct shash_node *node;
+ ovs_mutex_lock(&dp_netdev_mutex);
SHASH_FOR_EACH(node, &dp_netdevs) {
sset_add(all_dps, node->name);
}
+ ovs_mutex_unlock(&dp_netdev_mutex);
+
return 0;
}
return class != &dpif_netdev_class;
}
+static bool
+dpif_netdev_class_is_planetlab(const struct dpif_class *class)
+{
+ return class == &dpif_planetlab_class;
+}
+
static const char *
dpif_netdev_port_open_type(const struct dpif_class *class, const char *type)
{
return strcmp(type, "internal") ? type
+ : dpif_netdev_class_is_planetlab(class) ? "pltap"
: dpif_netdev_class_is_dummy(class) ? "dummy"
: "tap";
}
dpif = xmalloc(sizeof *dpif);
dpif_init(&dpif->dpif, dp->class, dp->name, netflow_id >> 8, netflow_id);
dpif->dp = dp;
- dpif->dp_serial = dp->serial;
+ dpif->last_port_seq = seq_read(dp->port_seq);
return &dpif->dpif;
}
{
uint32_t port_no;
- if (dp->class != &dpif_netdev_class) {
+ if (dp->class != &dpif_netdev_class &&
+ dp->class != &dpif_planetlab_class) {
const char *p;
int start_no = 0;
dp->class = class;
dp->name = xstrdup(name);
dp->open_cnt = 0;
+ dp->max_mtu = ETH_PAYLOAD_MAX;
for (i = 0; i < N_QUEUES; i++) {
dp->queues[i].head = dp->queues[i].tail = 0;
}
+ dp->queue_seq = seq_create();
hmap_init(&dp->flow_table);
list_init(&dp->port_list);
+ dp->port_seq = seq_create();
error = do_add_port(dp, name, "internal", ODPP_LOCAL);
if (error) {
bool create, struct dpif **dpifp)
{
struct dp_netdev *dp;
+ int error;
+ ovs_mutex_lock(&dp_netdev_mutex);
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;
- }
- ovs_assert(dp != NULL);
- }
+ error = create ? create_dp_netdev(name, class, &dp) : ENODEV;
} else {
- if (dp->class != class) {
- return EINVAL;
- } else if (create) {
- return EEXIST;
- }
+ error = (dp->class != class ? EINVAL
+ : create ? EEXIST
+ : 0);
+ }
+ if (!error) {
+ *dpifp = create_dpif_netdev(dp);
}
+ ovs_mutex_unlock(&dp_netdev_mutex);
- *dpifp = create_dpif_netdev(dp);
- return 0;
+ return error;
}
static void
do_del_port(dp, port->port_no);
}
dp_netdev_purge_queues(dp);
+ seq_destroy(dp->queue_seq);
hmap_destroy(&dp->flow_table);
+ seq_destroy(dp->port_seq);
free(dp->name);
free(dp);
}
dpif_netdev_close(struct dpif *dpif)
{
struct dp_netdev *dp = get_dp_netdev(dpif);
+
+ ovs_mutex_lock(&dp_netdev_mutex);
+
ovs_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);
+
+ ovs_mutex_unlock(&dp_netdev_mutex);
}
static int
dpif_netdev_destroy(struct dpif *dpif)
{
struct dp_netdev *dp = get_dp_netdev(dpif);
+
+ ovs_mutex_lock(&dp_netdev_mutex);
dp->destroyed = true;
+ ovs_mutex_unlock(&dp_netdev_mutex);
+
return 0;
}
dpif_netdev_get_stats(const struct dpif *dpif, struct dpif_dp_stats *stats)
{
struct dp_netdev *dp = get_dp_netdev(dpif);
+
+ ovs_mutex_lock(&dp_netdev_mutex);
stats->n_flows = hmap_count(&dp->flow_table);
stats->n_hit = dp->n_hit;
stats->n_missed = dp->n_missed;
stats->n_lost = dp->n_lost;
+ ovs_mutex_unlock(&dp_netdev_mutex);
+
return 0;
}
port->type = xstrdup(type);
error = netdev_get_mtu(netdev, &mtu);
- if (!error && mtu > max_mtu) {
- max_mtu = mtu;
+ if (!error && mtu > dp->max_mtu) {
+ dp->max_mtu = mtu;
}
list_push_back(&dp->port_list, &port->node);
dp->ports[odp_to_u32(port_no)] = port;
- dp->serial++;
+ seq_change(dp->port_seq);
return 0;
}
char namebuf[NETDEV_VPORT_NAME_BUFSIZE];
const char *dpif_port;
odp_port_t port_no;
+ int error;
+ ovs_mutex_lock(&dp_netdev_mutex);
dpif_port = netdev_vport_get_dpif_port(netdev, namebuf, sizeof namebuf);
if (*port_nop != ODPP_NONE) {
uint32_t port_idx = odp_to_u32(*port_nop);
if (port_idx >= MAX_PORTS) {
- return EFBIG;
+ error = EFBIG;
} else if (dp->ports[port_idx]) {
- return EBUSY;
+ error = EBUSY;
+ } else {
+ error = 0;
+ port_no = *port_nop;
}
- port_no = *port_nop;
} else {
port_no = choose_port(dp, dpif_port);
+ error = port_no == ODPP_NONE ? EFBIG : 0;
}
- if (port_no != ODPP_NONE) {
+ if (!error) {
*port_nop = port_no;
- return do_add_port(dp, dpif_port, netdev_get_type(netdev), port_no);
+ error = do_add_port(dp, dpif_port, netdev_get_type(netdev), port_no);
}
- return EFBIG;
+ ovs_mutex_unlock(&dp_netdev_mutex);
+
+ return error;
}
static int
dpif_netdev_port_del(struct dpif *dpif, odp_port_t port_no)
{
struct dp_netdev *dp = get_dp_netdev(dpif);
- return (port_no == ODPP_LOCAL ?
- EINVAL : do_del_port(dp, port_no));
+ int error;
+
+ ovs_mutex_lock(&dp_netdev_mutex);
+ error = port_no == ODPP_LOCAL ? EINVAL : do_del_port(dp, port_no);
+ ovs_mutex_unlock(&dp_netdev_mutex);
+
+ return error;
}
static bool
list_remove(&port->node);
dp->ports[odp_to_u32(port_no)] = NULL;
- dp->serial++;
+ seq_change(dp->port_seq);
netdev_close(port->netdev);
netdev_restore_flags(port->sf);
struct dp_netdev_port *port;
int error;
+ ovs_mutex_lock(&dp_netdev_mutex);
error = get_port_by_number(dp, port_no, &port);
if (!error && dpif_port) {
answer_port_query(port, dpif_port);
}
+ ovs_mutex_unlock(&dp_netdev_mutex);
+
return error;
}
struct dp_netdev_port *port;
int error;
+ ovs_mutex_lock(&dp_netdev_mutex);
error = get_port_by_name(dp, devname, &port);
if (!error && dpif_port) {
answer_port_query(port, dpif_port);
}
+ ovs_mutex_unlock(&dp_netdev_mutex);
+
return error;
}
-static odp_port_t
+static uint32_t
dpif_netdev_get_max_ports(const struct dpif *dpif OVS_UNUSED)
{
- return u32_to_odp(MAX_PORTS);
+ return MAX_PORTS;
}
static void
dpif_netdev_flow_flush(struct dpif *dpif)
{
struct dp_netdev *dp = get_dp_netdev(dpif);
+
+ ovs_mutex_lock(&dp_netdev_mutex);
dp_netdev_flow_flush(dp);
+ ovs_mutex_unlock(&dp_netdev_mutex);
+
return 0;
}
struct dp_netdev *dp = get_dp_netdev(dpif);
uint32_t port_idx;
+ ovs_mutex_lock(&dp_netdev_mutex);
for (port_idx = odp_to_u32(state->port_no);
port_idx < MAX_PORTS; port_idx++) {
struct dp_netdev_port *port = dp->ports[port_idx];
dpif_port->type = port->type;
dpif_port->port_no = port->port_no;
state->port_no = u32_to_odp(port_idx + 1);
+ ovs_mutex_unlock(&dp_netdev_mutex);
+
return 0;
}
}
+ ovs_mutex_unlock(&dp_netdev_mutex);
+
return EOF;
}
dpif_netdev_port_poll(const struct dpif *dpif_, char **devnamep OVS_UNUSED)
{
struct dpif_netdev *dpif = dpif_netdev_cast(dpif_);
- if (dpif->dp_serial != dpif->dp->serial) {
- dpif->dp_serial = dpif->dp->serial;
- return ENOBUFS;
+ uint64_t new_port_seq;
+ int error;
+
+ ovs_mutex_lock(&dp_netdev_mutex);
+ new_port_seq = seq_read(dpif->dp->port_seq);
+ if (dpif->last_port_seq != new_port_seq) {
+ dpif->last_port_seq = new_port_seq;
+ error = ENOBUFS;
} else {
- return EAGAIN;
+ error = EAGAIN;
}
+ ovs_mutex_unlock(&dp_netdev_mutex);
+
+ return error;
}
static void
dpif_netdev_port_poll_wait(const struct dpif *dpif_)
{
struct dpif_netdev *dpif = dpif_netdev_cast(dpif_);
- if (dpif->dp_serial != dpif->dp->serial) {
- poll_immediate_wake();
- }
+
+ ovs_mutex_lock(&dp_netdev_mutex);
+ seq_wait(dpif->dp->port_seq, dpif->last_port_seq);
+ ovs_mutex_unlock(&dp_netdev_mutex);
}
static struct dp_netdev_flow *
dpif_netdev_flow_from_nlattrs(const struct nlattr *key, uint32_t key_len,
struct flow *flow)
{
+ odp_port_t in_port;
+
if (odp_flow_key_to_flow(key, key_len, flow) != ODP_FIT_PERFECT) {
/* This should not happen: it indicates that odp_flow_key_from_flow()
* and odp_flow_key_to_flow() disagree on the acceptable form of a
return EINVAL;
}
- if (!is_valid_port_number(flow->in_port.odp_port)) {
+ in_port = flow->in_port.odp_port;
+ if (!is_valid_port_number(in_port) && in_port != ODPP_NONE) {
return EINVAL;
}
return error;
}
+ ovs_mutex_lock(&dp_netdev_mutex);
flow = dp_netdev_lookup_flow(dp, &key);
- if (!flow) {
- return ENOENT;
+ if (flow) {
+ if (stats) {
+ get_dpif_flow_stats(flow, stats);
+ }
+ if (actionsp) {
+ *actionsp = ofpbuf_clone_data(flow->actions, flow->actions_len);
+ }
+ } else {
+ error = ENOENT;
}
+ ovs_mutex_unlock(&dp_netdev_mutex);
- if (stats) {
- get_dpif_flow_stats(flow, stats);
- }
- if (actionsp) {
- *actionsp = ofpbuf_clone_data(flow->actions, flow->actions_len);
- }
- return 0;
+ return error;
}
static int
return error;
}
+ ovs_mutex_lock(&dp_netdev_mutex);
flow = dp_netdev_lookup_flow(dp, &key);
if (!flow) {
if (put->flags & DPIF_FP_CREATE) {
if (put->stats) {
memset(put->stats, 0, sizeof *put->stats);
}
- return dp_netdev_flow_add(dp, &key, put->actions,
- put->actions_len);
+ error = dp_netdev_flow_add(dp, &key, put->actions,
+ put->actions_len);
} else {
- return EFBIG;
+ error = EFBIG;
}
} else {
- return ENOENT;
+ error = ENOENT;
}
} else {
if (put->flags & DPIF_FP_MODIFY) {
- int error = set_flow_actions(flow, put->actions, put->actions_len);
+ error = set_flow_actions(flow, put->actions, put->actions_len);
if (!error) {
if (put->stats) {
get_dpif_flow_stats(flow, put->stats);
clear_stats(flow);
}
}
- return error;
} else {
- return EEXIST;
+ error = EEXIST;
}
}
+ ovs_mutex_unlock(&dp_netdev_mutex);
+
+ return error;
}
static int
return error;
}
+ ovs_mutex_lock(&dp_netdev_mutex);
flow = dp_netdev_lookup_flow(dp, &key);
if (flow) {
if (del->stats) {
get_dpif_flow_stats(flow, del->stats);
}
dp_netdev_free_flow(dp, flow);
- return 0;
} else {
- return ENOENT;
+ error = ENOENT;
}
+ ovs_mutex_unlock(&dp_netdev_mutex);
+
+ return error;
}
struct dp_netdev_flow_state {
struct dp_netdev_flow *flow;
struct hmap_node *node;
+ ovs_mutex_lock(&dp_netdev_mutex);
node = hmap_at_position(&dp->flow_table, &state->bucket, &state->offset);
if (!node) {
+ ovs_mutex_unlock(&dp_netdev_mutex);
return EOF;
}
*stats = &state->stats;
}
+ ovs_mutex_unlock(&dp_netdev_mutex);
return 0;
}
error = dpif_netdev_flow_from_nlattrs(execute->key, execute->key_len,
&key);
if (!error) {
+ ovs_mutex_lock(&dp_netdev_mutex);
dp_netdev_execute_actions(dp, ©, &key,
execute->actions, execute->actions_len);
+ ovs_mutex_unlock(&dp_netdev_mutex);
}
ofpbuf_uninit(©);
dpif_netdev_recv(struct dpif *dpif, struct dpif_upcall *upcall,
struct ofpbuf *buf)
{
- struct dp_netdev_queue *q = find_nonempty_queue(dpif);
+ struct dp_netdev_queue *q;
+ int error;
+
+ ovs_mutex_lock(&dp_netdev_mutex);
+ q = find_nonempty_queue(dpif);
if (q) {
struct dp_netdev_upcall *u = &q->upcalls[q->tail++ & QUEUE_MASK];
ofpbuf_uninit(buf);
*buf = u->buf;
- return 0;
+ error = 0;
} else {
- return EAGAIN;
+ error = EAGAIN;
}
+ ovs_mutex_unlock(&dp_netdev_mutex);
+
+ return error;
}
static void
dpif_netdev_recv_wait(struct dpif *dpif)
{
+ struct dp_netdev *dp = get_dp_netdev(dpif);
+ uint64_t seq;
+
+ ovs_mutex_lock(&dp_netdev_mutex);
+ seq = seq_read(dp->queue_seq);
if (find_nonempty_queue(dpif)) {
poll_immediate_wake();
} else {
- /* No messages ready to be received, and dp_wait() will ensure that we
- * wake up to queue new messages, so there is nothing to do. */
+ seq_wait(dp->queue_seq, seq);
}
+ ovs_mutex_unlock(&dp_netdev_mutex);
}
static void
dpif_netdev_recv_purge(struct dpif *dpif)
{
struct dpif_netdev *dpif_netdev = dpif_netdev_cast(dpif);
+ ovs_mutex_lock(&dp_netdev_mutex);
dp_netdev_purge_queues(dpif_netdev->dp);
+ ovs_mutex_unlock(&dp_netdev_mutex);
}
\f
static void
static void
dp_netdev_port_input(struct dp_netdev *dp, struct dp_netdev_port *port,
struct ofpbuf *packet, uint32_t skb_priority,
- uint32_t skb_mark, const struct flow_tnl *tnl)
+ uint32_t pkt_mark, const struct flow_tnl *tnl)
{
struct dp_netdev_flow *flow;
struct flow key;
return;
}
in_port_.odp_port = port->port_no;
- flow_extract(packet, skb_priority, skb_mark, tnl, &in_port_, &key);
+ flow_extract(packet, skb_priority, pkt_mark, tnl, &in_port_, &key);
flow = dp_netdev_lookup_flow(dp, &key);
if (flow) {
dp_netdev_flow_used(flow, packet);
static void
dpif_netdev_run(struct dpif *dpif)
{
- struct dp_netdev *dp = get_dp_netdev(dpif);
struct dp_netdev_port *port;
+ struct dp_netdev *dp;
struct ofpbuf packet;
- ofpbuf_init(&packet, DP_NETDEV_HEADROOM + VLAN_ETH_HEADER_LEN + max_mtu);
+ ovs_mutex_lock(&dp_netdev_mutex);
+ dp = get_dp_netdev(dpif);
+ ofpbuf_init(&packet,
+ DP_NETDEV_HEADROOM + VLAN_ETH_HEADER_LEN + dp->max_mtu);
LIST_FOR_EACH (port, node, &dp->port_list) {
int error;
}
}
ofpbuf_uninit(&packet);
+ ovs_mutex_unlock(&dp_netdev_mutex);
}
static void
dpif_netdev_wait(struct dpif *dpif)
{
- struct dp_netdev *dp = get_dp_netdev(dpif);
struct dp_netdev_port *port;
- LIST_FOR_EACH (port, node, &dp->port_list) {
+ /* There is a race here, if thread A calls dpif_netdev_wait(dpif) and
+ * thread B calls dpif_port_add(dpif) or dpif_port_remove(dpif) before
+ * A makes it to poll_block().
+ *
+ * But I think it doesn't matter:
+ *
+ * - In the dpif_port_add() case, A will not wake up when a packet
+ * arrives on the new port, but this would also happen if the
+ * ordering were reversed.
+ *
+ * - In the dpif_port_remove() case, A might wake up spuriously, but
+ * that is harmless. */
+
+ ovs_mutex_lock(&dp_netdev_mutex);
+ LIST_FOR_EACH (port, node, &get_dp_netdev(dpif)->port_list) {
if (port->rx) {
netdev_rx_wait(port->rx);
}
}
+ ovs_mutex_unlock(&dp_netdev_mutex);
}
static void
buf->size = packet->size;
upcall->packet = buf;
+ seq_change(dp->queue_seq);
+
return 0;
} else {
dp->n_lost++;
dp_netdev_output_port, dp_netdev_action_userspace);
}
+#define DPIF_NETDEV_CLASS_FUNCTIONS \
+ dpif_netdev_enumerate, \
+ dpif_netdev_port_open_type, \
+ dpif_netdev_open, \
+ dpif_netdev_close, \
+ dpif_netdev_destroy, \
+ dpif_netdev_run, \
+ dpif_netdev_wait, \
+ dpif_netdev_get_stats, \
+ dpif_netdev_port_add, \
+ dpif_netdev_port_del, \
+ dpif_netdev_port_query_by_number, \
+ dpif_netdev_port_query_by_name, \
+ dpif_netdev_get_max_ports, \
+ NULL, /* port_get_pid */ \
+ dpif_netdev_port_dump_start, \
+ dpif_netdev_port_dump_next, \
+ dpif_netdev_port_dump_done, \
+ dpif_netdev_port_poll, \
+ dpif_netdev_port_poll_wait, \
+ dpif_netdev_flow_get, \
+ dpif_netdev_flow_put, \
+ dpif_netdev_flow_del, \
+ dpif_netdev_flow_flush, \
+ dpif_netdev_flow_dump_start, \
+ dpif_netdev_flow_dump_next, \
+ dpif_netdev_flow_dump_done, \
+ dpif_netdev_execute, \
+ NULL, /* operate */ \
+ dpif_netdev_recv_set, \
+ dpif_netdev_queue_to_priority, \
+ dpif_netdev_recv, \
+ dpif_netdev_recv_wait, \
+ dpif_netdev_recv_purge, \
+
const struct dpif_class dpif_netdev_class = {
"netdev",
- dpif_netdev_enumerate,
- dpif_netdev_port_open_type,
- dpif_netdev_open,
- dpif_netdev_close,
- dpif_netdev_destroy,
- dpif_netdev_run,
- dpif_netdev_wait,
- dpif_netdev_get_stats,
- dpif_netdev_port_add,
- dpif_netdev_port_del,
- dpif_netdev_port_query_by_number,
- dpif_netdev_port_query_by_name,
- dpif_netdev_get_max_ports,
- NULL, /* port_get_pid */
- dpif_netdev_port_dump_start,
- dpif_netdev_port_dump_next,
- dpif_netdev_port_dump_done,
- dpif_netdev_port_poll,
- dpif_netdev_port_poll_wait,
- dpif_netdev_flow_get,
- dpif_netdev_flow_put,
- dpif_netdev_flow_del,
- dpif_netdev_flow_flush,
- dpif_netdev_flow_dump_start,
- dpif_netdev_flow_dump_next,
- dpif_netdev_flow_dump_done,
- dpif_netdev_execute,
- NULL, /* operate */
- dpif_netdev_recv_set,
- dpif_netdev_queue_to_priority,
- dpif_netdev_recv,
- dpif_netdev_recv_wait,
- dpif_netdev_recv_purge,
+ DPIF_NETDEV_CLASS_FUNCTIONS
+};
+
+const struct dpif_class dpif_planetlab_class = {
+ "planetlab",
+ DPIF_NETDEV_CLASS_FUNCTIONS
};
+static void
+dpif_dummy_change_port_number(struct unixctl_conn *conn, int argc OVS_UNUSED,
+ const char *argv[], void *aux OVS_UNUSED)
+{
+ struct dp_netdev_port *port;
+ struct dp_netdev *dp;
+ int port_no;
+
+ dp = shash_find_data(&dp_netdevs, argv[1]);
+ if (!dp || !dpif_netdev_class_is_dummy(dp->class)) {
+ unixctl_command_reply_error(conn, "unknown datapath or not a dummy");
+ return;
+ }
+
+ if (get_port_by_name(dp, argv[2], &port)) {
+ unixctl_command_reply_error(conn, "unknown port");
+ return;
+ }
+
+ port_no = atoi(argv[3]);
+ if (port_no <= 0 || port_no >= MAX_PORTS) {
+ unixctl_command_reply_error(conn, "bad port number");
+ return;
+ }
+ if (dp->ports[port_no]) {
+ unixctl_command_reply_error(conn, "port number already in use");
+ return;
+ }
+ dp->ports[odp_to_u32(port->port_no)] = NULL;
+ dp->ports[port_no] = port;
+ port->port_no = u32_to_odp(port_no);
+ seq_change(dp->port_seq);
+ unixctl_command_reply(conn, NULL);
+}
+
static void
dpif_dummy_register__(const char *type)
{
}
dpif_dummy_register__("dummy");
+
+ unixctl_command_register("dpif-dummy/change-port-number",
+ "DP PORT NEW-NUMBER",
+ 3, 3, dpif_dummy_change_port_number, NULL);
}
+