#include <arpa/inet.h>
#include <assert.h>
#include <errno.h>
+#include <inttypes.h>
#include <stdlib.h>
#include <string.h>
#include "buffer.h"
#include "chain.h"
+#include "csum.h"
#include "flow.h"
#include "netdev.h"
#include "packets.h"
#include "poll-loop.h"
#include "rconn.h"
+#include "stp.h"
+#include "vconn.h"
#include "table.h"
+#include "timeval.h"
#include "xtoxll.h"
#define THIS_MODULE VLM_datapath
#include "vlog.h"
-#define BRIDGE_PORT_NO_FLOOD 0x00000001
+extern char mfr_desc;
+extern char hw_desc;
+extern char sw_desc;
+extern char serial_num;
/* Capabilities supported by this implementation. */
-#define OFP_SUPPORTED_CAPABILITIES (OFPC_MULTI_PHY_TX)
+#define OFP_SUPPORTED_CAPABILITIES ( OFPC_FLOW_STATS \
+ | OFPC_TABLE_STATS \
+ | OFPC_PORT_STATS \
+ | OFPC_MULTI_PHY_TX )
/* Actions supported by this implementation. */
#define OFP_SUPPORTED_ACTIONS ( (1 << OFPAT_OUTPUT) \
| (1 << OFPAT_SET_TP_SRC) \
| (1 << OFPAT_SET_TP_DST) )
+#define PORT_STATUS_BITS (OFPPFL_PORT_DOWN | OFPPFL_LINK_DOWN)
+#define PORT_FLAG_BITS (~PORT_STATUS_BITS)
+
struct sw_port {
- uint32_t flags;
+ uint32_t flags; /* Some subset of PORT_FLAG_BITS. */
+ uint32_t status; /* Some subset of PORT_STATUS_BITS. */
struct datapath *dp;
struct netdev *netdev;
struct list node; /* Element in datapath.ports. */
+ unsigned long long int rx_packets, tx_packets;
+ unsigned long long int rx_bytes, tx_bytes;
+ unsigned long long int tx_dropped;
};
-struct datapath {
+/* The origin of a received OpenFlow message, to enable sending a reply. */
+struct sender {
+ struct remote *remote; /* The device that sent the message. */
+ uint32_t xid; /* The OpenFlow transaction ID. */
+};
+
+/* A connection to a controller or a management device. */
+struct remote {
+ struct list node;
struct rconn *rconn;
+#define TXQ_LIMIT 128 /* Max number of packets to queue for tx. */
+ int n_txq; /* Number of packets queued for tx on rconn. */
+
+ /* Support for reliable, multi-message replies to requests.
+ *
+ * If an incoming request needs to have a reliable reply that might
+ * require multiple messages, it can use remote_start_dump() to set up
+ * a callback that will be called as buffer space for replies. */
+ int (*cb_dump)(struct datapath *, void *aux);
+ void (*cb_done)(void *aux);
+ void *cb_aux;
+};
+
+struct datapath {
+ /* Remote connections. */
+ struct remote *controller; /* Connection to controller. */
+ struct list remotes; /* All connections (including controller). */
+ struct vconn *listen_vconn;
time_t last_timeout;
struct sw_chain *chain; /* Forwarding rules. */
- struct ofp_switch_config config;
+ /* Configuration set from controller. */
+ uint16_t flags;
+ uint16_t miss_send_len;
/* Switch ports. */
struct sw_port ports[OFPP_MAX];
struct list port_list; /* List of ports, for flooding. */
};
+static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(60, 60);
+
+static struct remote *remote_create(struct datapath *, struct rconn *);
+static void remote_run(struct datapath *, struct remote *);
+static void remote_wait(struct remote *);
+static void remote_destroy(struct remote *);
+
void dp_output_port(struct datapath *, struct buffer *,
- int in_port, int out_port);
-void dp_update_port_flags(struct datapath *dp, const struct ofp_phy_port *opp);
+ int in_port, int out_port, bool ignore_no_fwd);
+void dp_update_port_flags(struct datapath *dp, const struct ofp_port_mod *opm);
void dp_output_control(struct datapath *, struct buffer *, int in_port,
size_t max_len, int reason);
-static void send_flow_expired(struct datapath *, struct sw_flow *);
+static void send_flow_expired(struct datapath *, struct sw_flow *,
+ enum ofp_flow_expired_reason);
+static int update_port_status(struct sw_port *p);
static void send_port_status(struct sw_port *p, uint8_t status);
static void del_switch_port(struct sw_port *p);
static void execute_actions(struct datapath *, struct buffer *,
int in_port, const struct sw_flow_key *,
- const struct ofp_action *, int n_actions);
+ const struct ofp_action *, int n_actions,
+ bool ignore_no_fwd);
static void modify_vlan(struct buffer *buffer, const struct sw_flow_key *key,
const struct ofp_action *a);
static void modify_nh(struct buffer *buffer, uint16_t eth_proto,
#define PKT_COOKIE_BITS (32 - PKT_BUFFER_BITS)
-void fwd_port_input(struct datapath *, struct buffer *, int in_port);
-int fwd_control_input(struct datapath *, const void *, size_t);
+int run_flow_through_tables(struct datapath *, struct buffer *,
+ struct sw_port *);
+void fwd_port_input(struct datapath *, struct buffer *, struct sw_port *);
+int fwd_control_input(struct datapath *, const struct sender *,
+ const void *, size_t);
uint32_t save_buffer(struct buffer *);
static struct buffer *retrieve_buffer(uint32_t id);
return p - dp->ports;
}
-/* Generates a unique datapath id. It incorporates the datapath index
- * and a hardware address, if available. If not, it generates a random
- * one.
- */
+/* Generates and returns a random datapath id. */
static uint64_t
gen_datapath_id(void)
{
- /* Choose a random datapath id. */
- uint64_t id = 0;
- int i;
-
- srand(time(0));
-
- for (i = 0; i < ETH_ADDR_LEN; i++) {
- id |= (uint64_t)(rand() & 0xff) << (8*(ETH_ADDR_LEN-1 - i));
- }
-
- return id;
+ uint8_t ea[ETH_ADDR_LEN];
+ eth_addr_random(ea);
+ return eth_addr_to_uint64(ea);
}
int
return ENOMEM;
}
- dp->last_timeout = time(0);
- dp->rconn = rconn;
+ dp->last_timeout = time_now();
+ list_init(&dp->remotes);
+ dp->controller = remote_create(dp, rconn);
+ dp->listen_vconn = NULL;
dp->id = dpid <= UINT64_C(0xffffffffffff) ? dpid : gen_datapath_id();
dp->chain = chain_create();
if (!dp->chain) {
}
list_init(&dp->port_list);
- dp->config.flags = 0;
- dp->config.miss_send_len = OFP_DEFAULT_MISS_SEND_LEN;
+ dp->flags = 0;
+ dp->miss_send_len = OFP_DEFAULT_MISS_SEND_LEN;
*dp_ = dp;
return 0;
}
dp_add_port(struct datapath *dp, const char *name)
{
struct netdev *netdev;
+ struct in6_addr in6;
+ struct in_addr in4;
struct sw_port *p;
int error;
- error = netdev_open(name, &netdev);
+ error = netdev_open(name, NETDEV_ETH_TYPE_ANY, &netdev);
if (error) {
return error;
}
+ error = netdev_set_flags(netdev, NETDEV_UP | NETDEV_PROMISC, false);
+ if (error) {
+ VLOG_ERR("couldn't set promiscuous mode on %s device", name);
+ netdev_close(netdev);
+ return error;
+ }
+ if (netdev_get_in4(netdev, &in4)) {
+ VLOG_ERR("%s device has assigned IP address %s", name, inet_ntoa(in4));
+ }
+ if (netdev_get_in6(netdev, &in6)) {
+ char in6_name[INET6_ADDRSTRLEN + 1];
+ inet_ntop(AF_INET6, &in6, in6_name, sizeof in6_name);
+ VLOG_ERR("%s device has assigned IPv6 address %s", name, in6_name);
+ }
for (p = dp->ports; ; p++) {
if (p >= &dp->ports[ARRAY_SIZE(dp->ports)]) {
}
}
+ memset(p, '\0', sizeof *p);
+
p->dp = dp;
p->netdev = netdev;
list_push_back(&dp->port_list, &p->node);
}
void
-dp_run(struct datapath *dp)
+dp_add_listen_vconn(struct datapath *dp, struct vconn *listen_vconn)
{
- time_t now = time(0);
- struct sw_port *p, *n;
+ assert(!dp->listen_vconn);
+ dp->listen_vconn = listen_vconn;
+}
+
+void
+dp_run(struct datapath *dp)
+{
+ time_t now = time_now();
+ struct sw_port *p, *pn;
+ struct remote *r, *rn;
struct buffer *buffer = NULL;
- int i;
if (now != dp->last_timeout) {
struct list deleted = LIST_INITIALIZER(&deleted);
struct sw_flow *f, *n;
+ LIST_FOR_EACH (p, struct sw_port, node, &dp->port_list) {
+ if (update_port_status(p)) {
+ send_port_status(p, OFPPR_MOD);
+ }
+ }
+
chain_timeout(dp->chain, &deleted);
LIST_FOR_EACH_SAFE (f, n, struct sw_flow, node, &deleted) {
- send_flow_expired(dp, f);
+ send_flow_expired(dp, f, f->reason);
list_remove(&f->node);
flow_free(f);
}
}
poll_timer_wait(1000);
- LIST_FOR_EACH_SAFE (p, n, struct sw_port, node, &dp->port_list) {
+ LIST_FOR_EACH_SAFE (p, pn, struct sw_port, node, &dp->port_list) {
int error;
if (!buffer) {
}
error = netdev_recv(p->netdev, buffer);
if (!error) {
- fwd_port_input(dp, buffer, port_no(dp, p));
+ p->rx_packets++;
+ p->rx_bytes += buffer->size;
+ fwd_port_input(dp, buffer, p);
buffer = NULL;
} else if (error != EAGAIN) {
- VLOG_ERR("Error receiving data from %s: %s",
- netdev_get_name(p->netdev), strerror(error));
- del_switch_port(p);
+ VLOG_ERR_RL(&rl, "error receiving data from %s: %s",
+ netdev_get_name(p->netdev), strerror(error));
}
}
buffer_delete(buffer);
- /* Process a number of commands from the controller, but cap it at a
- * reasonable number so that other processing doesn't starve. */
+ /* Talk to remotes. */
+ LIST_FOR_EACH_SAFE (r, rn, struct remote, node, &dp->remotes) {
+ remote_run(dp, r);
+ }
+ if (dp->listen_vconn) {
+ for (;;) {
+ struct vconn *new_vconn;
+ int retval;
+
+ retval = vconn_accept(dp->listen_vconn, &new_vconn);
+ if (retval) {
+ if (retval != EAGAIN) {
+ VLOG_WARN_RL(&rl, "accept failed (%s)", strerror(retval));
+ }
+ break;
+ }
+ remote_create(dp, rconn_new_from_vconn("passive", new_vconn));
+ }
+ }
+}
+
+static void
+remote_run(struct datapath *dp, struct remote *r)
+{
+ int i;
+
+ rconn_run(r->rconn);
+
+ /* Do some remote processing, but cap it at a reasonable amount so that
+ * other processing doesn't starve. */
for (i = 0; i < 50; i++) {
- struct buffer *buffer = rconn_recv(dp->rconn);
- if (!buffer) {
- break;
+ if (!r->cb_dump) {
+ struct buffer *buffer;
+ struct ofp_header *oh;
+
+ buffer = rconn_recv(r->rconn);
+ if (!buffer) {
+ break;
+ }
+
+ if (buffer->size >= sizeof *oh) {
+ struct sender sender;
+
+ oh = buffer->data;
+ sender.remote = r;
+ sender.xid = oh->xid;
+ fwd_control_input(dp, &sender, buffer->data, buffer->size);
+ } else {
+ VLOG_WARN_RL(&rl, "received too-short OpenFlow message");
+ }
+ buffer_delete(buffer);
+ } else {
+ if (r->n_txq < TXQ_LIMIT) {
+ int error = r->cb_dump(dp, r->cb_aux);
+ if (error <= 0) {
+ if (error) {
+ VLOG_WARN_RL(&rl, "dump callback error: %s",
+ strerror(-error));
+ }
+ r->cb_done(r->cb_aux);
+ r->cb_dump = NULL;
+ }
+ } else {
+ break;
+ }
}
- fwd_control_input(dp, buffer->data, buffer->size);
- buffer_delete(buffer);
}
- rconn_run(dp->rconn);
+ if (!rconn_is_alive(r->rconn)) {
+ remote_destroy(r);
+ }
+}
+
+static void
+remote_wait(struct remote *r)
+{
+ rconn_run_wait(r->rconn);
+ rconn_recv_wait(r->rconn);
+}
+
+static void
+remote_destroy(struct remote *r)
+{
+ if (r) {
+ if (r->cb_dump && r->cb_done) {
+ r->cb_done(r->cb_aux);
+ }
+ list_remove(&r->node);
+ rconn_destroy(r->rconn);
+ free(r);
+ }
+}
+
+static struct remote *
+remote_create(struct datapath *dp, struct rconn *rconn)
+{
+ struct remote *remote = xmalloc(sizeof *remote);
+ list_push_back(&dp->remotes, &remote->node);
+ remote->rconn = rconn;
+ remote->cb_dump = NULL;
+ remote->n_txq = 0;
+ return remote;
+}
+
+/* Starts a callback-based, reliable, possibly multi-message reply to a
+ * request made by 'remote'.
+ *
+ * 'dump' designates a function that will be called when the 'remote' send
+ * queue has an empty slot. It should compose a message and send it on
+ * 'remote'. On success, it should return 1 if it should be called again when
+ * another send queue slot opens up, 0 if its transmissions are complete, or a
+ * negative errno value on failure.
+ *
+ * 'done' designates a function to clean up any resources allocated for the
+ * dump. It must handle being called before the dump is complete (which will
+ * happen if 'remote' is closed unexpectedly).
+ *
+ * 'aux' is passed to 'dump' and 'done'. */
+static void
+remote_start_dump(struct remote *remote,
+ int (*dump)(struct datapath *, void *),
+ void (*done)(void *),
+ void *aux)
+{
+ assert(!remote->cb_dump);
+ remote->cb_dump = dump;
+ remote->cb_done = done;
+ remote->cb_aux = aux;
}
void
dp_wait(struct datapath *dp)
{
struct sw_port *p;
+ struct remote *r;
LIST_FOR_EACH (p, struct sw_port, node, &dp->port_list) {
netdev_recv_wait(p->netdev);
}
- rconn_recv_wait(dp->rconn);
+ LIST_FOR_EACH (r, struct remote, node, &dp->remotes) {
+ remote_wait(r);
+ }
+ if (dp->listen_vconn) {
+ vconn_accept_wait(dp->listen_vconn);
+ }
}
/* Delete 'p' from switch. */
free(dp);
}
+/* Send packets out all the ports except the originating one. If the
+ * "flood" argument is set, don't send out ports with flooding disabled.
+ */
static int
-flood(struct datapath *dp, struct buffer *buffer, int in_port)
+output_all(struct datapath *dp, struct buffer *buffer, int in_port, int flood)
{
struct sw_port *p;
int prev_port;
prev_port = -1;
LIST_FOR_EACH (p, struct sw_port, node, &dp->port_list) {
- if (port_no(dp, p) == in_port || p->flags & BRIDGE_PORT_NO_FLOOD) {
+ if (port_no(dp, p) == in_port) {
+ continue;
+ }
+ if (flood && p->flags & OFPPFL_NO_FLOOD) {
continue;
}
if (prev_port != -1) {
- dp_output_port(dp, buffer_clone(buffer), in_port, prev_port);
+ dp_output_port(dp, buffer_clone(buffer), in_port, prev_port,
+ false);
}
prev_port = port_no(dp, p);
}
if (prev_port != -1)
- dp_output_port(dp, buffer, in_port, prev_port);
+ dp_output_port(dp, buffer, in_port, prev_port, false);
else
buffer_delete(buffer);
{
if (out_port >= 0 && out_port < OFPP_MAX) {
struct sw_port *p = &dp->ports[out_port];
- if (p->netdev != NULL) {
- netdev_send(p->netdev, buffer);
+ if (p->netdev != NULL && !(p->status & OFPPFL_PORT_DOWN)) {
+ if (!netdev_send(p->netdev, buffer)) {
+ p->tx_packets++;
+ p->tx_bytes += buffer->size;
+ } else {
+ p->tx_dropped++;
+ }
return;
}
}
buffer_delete(buffer);
- /* FIXME: ratelimit */
- VLOG_DBG("can't forward to bad port %d\n", out_port);
+ VLOG_DBG_RL(&rl, "can't forward to bad port %d\n", out_port);
}
/* Takes ownership of 'buffer' and transmits it to 'out_port' on 'dp'.
*/
void
dp_output_port(struct datapath *dp, struct buffer *buffer,
- int in_port, int out_port)
+ int in_port, int out_port, bool ignore_no_fwd)
{
assert(buffer);
if (out_port == OFPP_FLOOD) {
- flood(dp, buffer, in_port);
+ output_all(dp, buffer, in_port, 1);
+ } else if (out_port == OFPP_ALL) {
+ output_all(dp, buffer, in_port, 0);
} else if (out_port == OFPP_CONTROLLER) {
dp_output_control(dp, buffer, in_port, 0, OFPR_ACTION);
+ } else if (out_port == OFPP_IN_PORT) {
+ output_packet(dp, buffer, in_port);
+ } else if (out_port == OFPP_TABLE) {
+ struct sw_port *p = in_port < OFPP_MAX ? &dp->ports[in_port] : 0;
+ if (run_flow_through_tables(dp, buffer, p)) {
+ buffer_delete(buffer);
+ }
} else {
+ if (in_port == out_port) {
+ VLOG_DBG_RL(&rl, "can't directly forward to input port");
+ return;
+ }
output_packet(dp, buffer, out_port);
}
}
+static void *
+make_openflow_reply(size_t openflow_len, uint8_t type,
+ const struct sender *sender, struct buffer **bufferp)
+{
+ return make_openflow_xid(openflow_len, type, sender ? sender->xid : 0,
+ bufferp);
+}
+
+static int
+send_openflow_buffer(struct datapath *dp, struct buffer *buffer,
+ const struct sender *sender)
+{
+ struct remote *remote = sender ? sender->remote : dp->controller;
+ struct rconn *rconn = remote->rconn;
+ int retval;
+
+ update_openflow_length(buffer);
+ retval = rconn_send_with_limit(rconn, buffer, &remote->n_txq, TXQ_LIMIT);
+ if (retval) {
+ VLOG_WARN_RL(&rl, "send to %s failed: %s",
+ rconn_get_name(rconn), strerror(retval));
+ }
+ return retval;
+}
+
/* Takes ownership of 'buffer' and transmits it to 'dp''s controller. If the
* packet can be saved in a buffer, then only the first max_len bytes of
* 'buffer' are sent; otherwise, all of 'buffer' is sent. 'reason' indicates
opi->in_port = htons(in_port);
opi->reason = reason;
opi->pad = 0;
- rconn_send(dp->rconn, buffer);
+ send_openflow_buffer(dp, buffer, NULL);
}
static void fill_port_desc(struct datapath *dp, struct sw_port *p,
sizeof desc->name);
desc->name[sizeof desc->name - 1] = '\0';
memcpy(desc->hw_addr, netdev_get_etheraddr(p->netdev), ETH_ADDR_LEN);
- desc->flags = htonl(p->flags);
+ desc->flags = 0;
desc->features = htonl(netdev_get_features(p->netdev));
desc->speed = htonl(netdev_get_speed(p->netdev));
+ desc->flags = htonl(p->flags | p->status);
}
static void
-dp_send_features_reply(struct datapath *dp, uint32_t xid)
+dp_send_features_reply(struct datapath *dp, const struct sender *sender)
{
struct buffer *buffer;
struct ofp_switch_features *ofr;
struct sw_port *p;
- buffer = buffer_new(sizeof *ofr);
- ofr = buffer_put_uninit(buffer, sizeof *ofr);
- memset(ofr, 0, sizeof *ofr);
- ofr->header.version = OFP_VERSION;
- ofr->header.type = OFPT_FEATURES_REPLY;
- ofr->header.xid = xid;
- ofr->datapath_id = htonll(dp->id);
- ofr->n_exact = htonl(2 * TABLE_HASH_MAX_FLOWS);
- ofr->n_mac_only = htonl(TABLE_MAC_MAX_FLOWS);
- ofr->n_compression = 0; /* Not supported */
- ofr->n_general = htonl(TABLE_LINEAR_MAX_FLOWS);
- ofr->buffer_mb = htonl(UINT32_MAX);
- ofr->n_buffers = htonl(N_PKT_BUFFERS);
- ofr->capabilities = htonl(OFP_SUPPORTED_CAPABILITIES);
- ofr->actions = htonl(OFP_SUPPORTED_ACTIONS);
+ ofr = make_openflow_reply(sizeof *ofr, OFPT_FEATURES_REPLY,
+ sender, &buffer);
+ ofr->datapath_id = htonll(dp->id);
+ ofr->n_tables = dp->chain->n_tables;
+ ofr->n_buffers = htonl(N_PKT_BUFFERS);
+ ofr->capabilities = htonl(OFP_SUPPORTED_CAPABILITIES);
+ ofr->actions = htonl(OFP_SUPPORTED_ACTIONS);
LIST_FOR_EACH (p, struct sw_port, node, &dp->port_list) {
struct ofp_phy_port *opp = buffer_put_uninit(buffer, sizeof *opp);
memset(opp, 0, sizeof *opp);
fill_port_desc(dp, p, opp);
}
- ofr = buffer_at_assert(buffer, 0, sizeof *ofr);
- ofr->header.length = htons(buffer->size);
- rconn_send(dp->rconn, buffer);
+ send_openflow_buffer(dp, buffer, sender);
}
void
-dp_update_port_flags(struct datapath *dp, const struct ofp_phy_port *opp)
+dp_update_port_flags(struct datapath *dp, const struct ofp_port_mod *opm)
{
- struct sw_port *p;
+ const struct ofp_phy_port *opp = &opm->desc;
+ int port_no = ntohs(opp->port_no);
+ if (port_no < OFPP_MAX) {
+ struct sw_port *p = &dp->ports[port_no];
+ uint32_t flag_mask;
+
+ /* Make sure the port id hasn't changed since this was sent */
+ if (!p || memcmp(opp->hw_addr, netdev_get_etheraddr(p->netdev),
+ ETH_ADDR_LEN) != 0) {
+ return;
+ }
- p = &dp->ports[htons(opp->port_no)];
- /* Make sure the port id hasn't changed since this was sent */
- if (!p || memcmp(opp->hw_addr, netdev_get_etheraddr(p->netdev),
- ETH_ADDR_LEN) != 0)
- return;
-
- p->flags = htonl(opp->flags);
+ flag_mask = ntohl(opm->mask) & PORT_FLAG_BITS;
+ if (flag_mask) {
+ p->flags &= ~flag_mask;
+ p->flags |= ntohl(opp->flags) & flag_mask;
+ }
+
+ if (opm->mask & htonl(OFPPFL_PORT_DOWN)) {
+ if ((opp->flags & htonl(OFPPFL_PORT_DOWN))
+ && (p->status & OFPPFL_PORT_DOWN) == 0) {
+ p->status |= OFPPFL_PORT_DOWN;
+ netdev_turn_flags_off(p->netdev, NETDEV_UP, true);
+ } else if ((opp->flags & htonl(OFPPFL_PORT_DOWN)) == 0
+ && (p->status & OFPPFL_PORT_DOWN)) {
+ p->status &= ~OFPPFL_PORT_DOWN;
+ netdev_turn_flags_on(p->netdev, NETDEV_UP, true);
+ }
+ }
+ }
+}
+
+/* Update the port status field of the bridge port. A non-zero return
+ * value indicates some field has changed.
+ *
+ * NB: Callers of this function may hold the RCU read lock, so any
+ * additional checks must not sleep.
+ */
+static int
+update_port_status(struct sw_port *p)
+{
+ int retval;
+ enum netdev_flags flags;
+ uint32_t orig_status = p->status;
+
+ if (netdev_get_flags(p->netdev, &flags) < 0) {
+ VLOG_WARN_RL(&rl, "could not get netdev flags for %s",
+ netdev_get_name(p->netdev));
+ return 0;
+ } else {
+ if (flags & NETDEV_UP) {
+ p->status &= ~OFPPFL_PORT_DOWN;
+ } else {
+ p->status |= OFPPFL_PORT_DOWN;
+ }
+ }
+
+ /* Not all cards support this getting link status, so don't warn on
+ * error. */
+ retval = netdev_get_link_status(p->netdev);
+ if (retval == 1) {
+ p->status &= ~OFPPFL_LINK_DOWN;
+ } else if (retval == 0) {
+ p->status |= OFPPFL_LINK_DOWN;
+ }
+
+ return (orig_status != p->status);
}
static void
{
struct buffer *buffer;
struct ofp_port_status *ops;
- buffer = buffer_new(sizeof *ops);
- ops = buffer_put_uninit(buffer, sizeof *ops);
- ops->header.version = OFP_VERSION;
- ops->header.type = OFPT_PORT_STATUS;
- ops->header.length = htons(sizeof(*ops));
- ops->header.xid = htonl(0);
- ops->reason = status;
+ ops = make_openflow_xid(sizeof *ops, OFPT_PORT_STATUS, 0, &buffer);
+ ops->reason = status;
+ memset(ops->pad, 0, sizeof ops->pad);
fill_port_desc(p->dp, p, &ops->desc);
- rconn_send(p->dp->rconn, buffer);
+
+ send_openflow_buffer(p->dp, buffer, NULL);
}
void
-send_flow_expired(struct datapath *dp, struct sw_flow *flow)
+send_flow_expired(struct datapath *dp, struct sw_flow *flow,
+ enum ofp_flow_expired_reason reason)
{
struct buffer *buffer;
struct ofp_flow_expired *ofe;
- buffer = buffer_new(sizeof *ofe);
- ofe = buffer_put_uninit(buffer, sizeof *ofe);
- ofe->header.version = OFP_VERSION;
- ofe->header.type = OFPT_FLOW_EXPIRED;
- ofe->header.length = htons(sizeof(*ofe));
- ofe->header.xid = htonl(0);
+ ofe = make_openflow_xid(sizeof *ofe, OFPT_FLOW_EXPIRED, 0, &buffer);
flow_fill_match(&ofe->match, &flow->key);
- ofe->duration = htonl(flow->timeout - flow->max_idle - flow->created);
- ofe->packet_count = htonll(flow->packet_count);
- ofe->byte_count = htonll(flow->byte_count);
- rconn_send(dp->rconn, buffer);
+
+ ofe->priority = htons(flow->priority);
+ ofe->reason = reason;
+ memset(ofe->pad, 0, sizeof ofe->pad);
+
+ ofe->duration = htonl(time_now() - flow->created);
+ memset(ofe->pad2, 0, sizeof ofe->pad2);
+ ofe->packet_count = htonll(flow->packet_count);
+ ofe->byte_count = htonll(flow->byte_count);
+ send_openflow_buffer(dp, buffer, NULL);
+}
+
+void
+dp_send_error_msg(struct datapath *dp, const struct sender *sender,
+ uint16_t type, uint16_t code, const uint8_t *data, size_t len)
+{
+ struct buffer *buffer;
+ struct ofp_error_msg *oem;
+ oem = make_openflow_reply(sizeof(*oem)+len, OFPT_ERROR_MSG,
+ sender, &buffer);
+ oem->type = htons(type);
+ oem->code = htons(code);
+ memcpy(oem->data, data, len);
+ send_openflow_buffer(dp, buffer, sender);
+}
+
+static void
+fill_flow_stats(struct buffer *buffer, struct sw_flow *flow,
+ int table_idx, time_t now)
+{
+ struct ofp_flow_stats *ofs;
+ int length = sizeof *ofs + sizeof *ofs->actions * flow->n_actions;
+ ofs = buffer_put_uninit(buffer, length);
+ ofs->length = htons(length);
+ ofs->table_id = table_idx;
+ ofs->pad = 0;
+ ofs->match.wildcards = htonl(flow->key.wildcards);
+ ofs->match.in_port = flow->key.flow.in_port;
+ memcpy(ofs->match.dl_src, flow->key.flow.dl_src, ETH_ADDR_LEN);
+ memcpy(ofs->match.dl_dst, flow->key.flow.dl_dst, ETH_ADDR_LEN);
+ ofs->match.dl_vlan = flow->key.flow.dl_vlan;
+ ofs->match.dl_type = flow->key.flow.dl_type;
+ ofs->match.nw_src = flow->key.flow.nw_src;
+ ofs->match.nw_dst = flow->key.flow.nw_dst;
+ ofs->match.nw_proto = flow->key.flow.nw_proto;
+ ofs->match.pad = 0;
+ ofs->match.tp_src = flow->key.flow.tp_src;
+ ofs->match.tp_dst = flow->key.flow.tp_dst;
+ ofs->duration = htonl(now - flow->created);
+ ofs->priority = htons(flow->priority);
+ ofs->idle_timeout = htons(flow->idle_timeout);
+ ofs->hard_timeout = htons(flow->hard_timeout);
+ memset(ofs->pad2, 0, sizeof ofs->pad2);
+ ofs->packet_count = htonll(flow->packet_count);
+ ofs->byte_count = htonll(flow->byte_count);
+ memcpy(ofs->actions, flow->actions,
+ sizeof *ofs->actions * flow->n_actions);
}
+
\f
-/* 'buffer' was received on 'in_port', a physical switch port between 0 and
- * OFPP_MAX. Process it according to 'chain'. */
-void fwd_port_input(struct datapath *dp, struct buffer *buffer, int in_port)
+/* 'buffer' was received on 'p', which may be a a physical switch port or a
+ * null pointer. Process it according to 'dp''s flow table. Returns 0 if
+ * successful, in which case 'buffer' is destroyed, or -ESRCH if there is no
+ * matching flow, in which case 'buffer' still belongs to the caller. */
+int run_flow_through_tables(struct datapath *dp, struct buffer *buffer,
+ struct sw_port *p)
{
struct sw_flow_key key;
struct sw_flow *flow;
key.wildcards = 0;
- flow_extract(buffer, in_port, &key.flow);
+ if (flow_extract(buffer, p ? port_no(dp, p) : OFPP_NONE, &key.flow)
+ && (dp->flags & OFPC_FRAG_MASK) == OFPC_FRAG_DROP) {
+ /* Drop fragment. */
+ buffer_delete(buffer);
+ return 0;
+ }
+ if (p && p->flags & (OFPPFL_NO_RECV | OFPPFL_NO_RECV_STP)
+ && p->flags & (!eth_addr_equals(key.flow.dl_dst, stp_eth_addr)
+ ? OFPPFL_NO_RECV : OFPPFL_NO_RECV_STP)) {
+ buffer_delete(buffer);
+ return 0;
+ }
+
flow = chain_lookup(dp->chain, &key);
if (flow != NULL) {
flow_used(flow, buffer);
- execute_actions(dp, buffer, in_port, &key,
- flow->actions, flow->n_actions);
+ execute_actions(dp, buffer, port_no(dp, p),
+ &key, flow->actions, flow->n_actions, false);
+ return 0;
} else {
- dp_output_control(dp, buffer, in_port, dp->config.miss_send_len,
- OFPR_NO_MATCH);
+ return -ESRCH;
+ }
+}
+
+/* 'buffer' was received on 'p', which may be a a physical switch port or a
+ * null pointer. Process it according to 'dp''s flow table, sending it up to
+ * the controller if no flow matches. Takes ownership of 'buffer'. */
+void fwd_port_input(struct datapath *dp, struct buffer *buffer,
+ struct sw_port *p)
+{
+ if (run_flow_through_tables(dp, buffer, p)) {
+ dp_output_control(dp, buffer, port_no(dp, p),
+ dp->miss_send_len, OFPR_NO_MATCH);
}
}
static void
do_output(struct datapath *dp, struct buffer *buffer, int in_port,
- size_t max_len, int out_port)
+ size_t max_len, int out_port, bool ignore_no_fwd)
{
if (out_port != OFPP_CONTROLLER) {
- dp_output_port(dp, buffer, in_port, out_port);
+ dp_output_port(dp, buffer, in_port, out_port, ignore_no_fwd);
} else {
dp_output_control(dp, buffer, in_port, max_len, OFPR_ACTION);
}
static void
execute_actions(struct datapath *dp, struct buffer *buffer,
int in_port, const struct sw_flow_key *key,
- const struct ofp_action *actions, int n_actions)
+ const struct ofp_action *actions, int n_actions,
+ bool ignore_no_fwd)
{
/* Every output action needs a separate clone of 'buffer', but the common
* case is just a single output action, so that doing a clone and then
struct eth_header *eh = buffer->l2;
if (prev_port != -1) {
- do_output(dp, buffer_clone(buffer), in_port, max_len, prev_port);
+ do_output(dp, buffer_clone(buffer), in_port, max_len, prev_port,
+ ignore_no_fwd);
prev_port = -1;
}
}
}
if (prev_port != -1)
- do_output(dp, buffer, in_port, max_len, prev_port);
+ do_output(dp, buffer, in_port, max_len, prev_port, ignore_no_fwd);
else
buffer_delete(buffer);
}
-/* Returns the new checksum for a packet in which the checksum field previously
- * contained 'old_csum' and in which a field that contained 'old_u16' was
- * changed to contain 'new_u16'. */
-static uint16_t
-recalc_csum16(uint16_t old_csum, uint16_t old_u16, uint16_t new_u16)
-{
- /* Ones-complement arithmetic is endian-independent, so this code does not
- * use htons() or ntohs().
- *
- * See RFC 1624 for formula and explanation. */
- uint16_t hc_complement = ~old_csum;
- uint16_t m_complement = ~old_u16;
- uint16_t m_prime = new_u16;
- uint32_t sum = hc_complement + m_complement + m_prime;
- uint16_t hc_prime_complement = sum + (sum >> 16);
- return ~hc_prime_complement;
-}
-
-/* Returns the new checksum for a packet in which the checksum field previously
- * contained 'old_csum' and in which a field that contained 'old_u32' was
- * changed to contain 'new_u32'. */
-static uint16_t
-recalc_csum32(uint16_t old_csum, uint32_t old_u32, uint32_t new_u32)
-{
- return recalc_csum16(recalc_csum16(old_csum, old_u32, new_u32),
- old_u32 >> 16, new_u32 >> 16);
-}
-
static void modify_nh(struct buffer *buffer, uint16_t eth_proto,
uint8_t nw_proto, const struct ofp_action *a)
{
uint16_t new_id = a->arg.vlan_id;
struct vlan_eth_header *veh;
- if (new_id != OFP_VLAN_NONE) {
+ if (new_id != htons(OFP_VLAN_NONE)) {
if (key->flow.dl_vlan != htons(OFP_VLAN_NONE)) {
/* Modify vlan id, but maintain other TCI values */
veh = buffer->l2;
veh->veth_tci &= ~htons(VLAN_VID);
- veh->veth_tci |= htons(new_id);
+ veh->veth_tci |= new_id;
} else {
/* Insert new vlan id. */
struct eth_header *eh = buffer->l2;
}
static int
-recv_features_request(struct datapath *dp, const void *msg)
+recv_features_request(struct datapath *dp, const struct sender *sender,
+ const void *msg)
{
- struct ofp_header *ofr = msg;
- dp_send_features_reply(dp, ofr->xid);
+ dp_send_features_reply(dp, sender);
return 0;
}
static int
-recv_get_config_request(struct datapath *dp, const void *msg)
+recv_get_config_request(struct datapath *dp, const struct sender *sender,
+ const void *msg)
{
- struct ofp_header *gcr = msg;
struct buffer *buffer;
struct ofp_switch_config *osc;
- buffer = buffer_new(sizeof dp->config);
- osc = buffer_put(buffer, &dp->config, sizeof dp->config);
- osc->header.version = OFP_VERSION;
- osc->header.type = OFPT_GET_CONFIG_REPLY;
- osc->header.length = htons(sizeof *osc);
- osc->header.xid = gcr->xid;
- rconn_send(dp->rconn, buffer);
- return 0;
+ osc = make_openflow_reply(sizeof *osc, OFPT_GET_CONFIG_REPLY,
+ sender, &buffer);
+
+ osc->flags = htons(dp->flags);
+ osc->miss_send_len = htons(dp->miss_send_len);
+
+ return send_openflow_buffer(dp, buffer, sender);
}
static int
-recv_set_config(struct datapath *dp, const void *msg)
+recv_set_config(struct datapath *dp, const struct sender *sender UNUSED,
+ const void *msg)
{
const struct ofp_switch_config *osc = msg;
- dp->config = *osc;
+ int flags;
+
+ flags = ntohs(osc->flags) & (OFPC_SEND_FLOW_EXP | OFPC_FRAG_MASK);
+ if ((flags & OFPC_FRAG_MASK) != OFPC_FRAG_NORMAL
+ && (flags & OFPC_FRAG_MASK) != OFPC_FRAG_DROP) {
+ flags = (flags & ~OFPC_FRAG_MASK) | OFPC_FRAG_DROP;
+ }
+ dp->flags = flags;
+ dp->miss_send_len = ntohs(osc->miss_send_len);
return 0;
}
static int
-recv_packet_out(struct datapath *dp, const void *msg)
+recv_packet_out(struct datapath *dp, const struct sender *sender UNUSED,
+ const void *msg)
{
const struct ofp_packet_out *opo = msg;
+ struct sw_flow_key key;
+ struct buffer *buffer;
+ int n_actions = ntohs(opo->n_actions);
+ int act_len = n_actions * sizeof opo->actions[0];
+
+ if (act_len > (ntohs(opo->header.length) - sizeof *opo)) {
+ VLOG_DBG_RL(&rl, "message too short for number of actions");
+ return -EINVAL;
+ }
if (ntohl(opo->buffer_id) == (uint32_t) -1) {
/* FIXME: can we avoid copying data here? */
- int data_len = ntohs(opo->header.length) - sizeof *opo;
- struct buffer *buffer = buffer_new(data_len);
- buffer_put(buffer, opo->u.data, data_len);
- dp_output_port(dp, buffer,
- ntohs(opo->in_port), ntohs(opo->out_port));
+ int data_len = ntohs(opo->header.length) - sizeof *opo - act_len;
+ buffer = buffer_new(data_len);
+ buffer_put(buffer, &opo->actions[n_actions], data_len);
} else {
- struct sw_flow_key key;
- struct buffer *buffer;
- int n_acts;
-
buffer = retrieve_buffer(ntohl(opo->buffer_id));
if (!buffer) {
return -ESRCH;
}
-
- n_acts = (ntohs(opo->header.length) - sizeof *opo)
- / sizeof *opo->u.actions;
- flow_extract(buffer, ntohs(opo->in_port), &key.flow);
- execute_actions(dp, buffer, ntohs(opo->in_port),
- &key, opo->u.actions, n_acts);
}
- return 0;
+
+ flow_extract(buffer, ntohs(opo->in_port), &key.flow);
+ execute_actions(dp, buffer, ntohs(opo->in_port),
+ &key, opo->actions, n_actions, true);
+
+ return 0;
}
static int
-recv_port_mod(struct datapath *dp, const void *msg)
+recv_port_mod(struct datapath *dp, const struct sender *sender UNUSED,
+ const void *msg)
{
const struct ofp_port_mod *opm = msg;
- dp_update_port_flags(dp, &opm->desc);
+ dp_update_port_flags(dp, opm);
return 0;
}
add_flow(struct datapath *dp, const struct ofp_flow_mod *ofm)
{
int error = -ENOMEM;
- int n_acts;
+ int n_actions;
+ int i;
struct sw_flow *flow;
- /* Check number of actions. */
- n_acts = (ntohs(ofm->header.length) - sizeof *ofm) / sizeof *ofm->actions;
- if (n_acts > MAX_ACTIONS) {
- error = -E2BIG;
- goto error;
+ /* To prevent loops, make sure there's no action to send to the
+ * OFP_TABLE virtual port.
+ */
+ n_actions = (ntohs(ofm->header.length) - sizeof *ofm)
+ / sizeof *ofm->actions;
+ for (i=0; i<n_actions; i++) {
+ const struct ofp_action *a = &ofm->actions[i];
+
+ if (a->type == htons(OFPAT_OUTPUT)
+ && (a->arg.output.port == htons(OFPP_TABLE)
+ || a->arg.output.port == htons(OFPP_NONE)
+ || a->arg.output.port == ofm->match.in_port)) {
+ /* xxx Send fancy new error message? */
+ goto error;
+ }
}
/* Allocate memory. */
- flow = flow_alloc(n_acts);
+ flow = flow_alloc(n_actions);
if (flow == NULL)
goto error;
/* Fill out flow. */
flow_extract_match(&flow->key, &ofm->match);
- flow->group_id = ntohl(ofm->group_id);
- flow->max_idle = ntohs(ofm->max_idle);
- flow->timeout = time(0) + flow->max_idle; /* FIXME */
- flow->n_actions = n_acts;
- flow->created = time(0); /* FIXME */
+ flow->priority = flow->key.wildcards ? ntohs(ofm->priority) : -1;
+ flow->idle_timeout = ntohs(ofm->idle_timeout);
+ flow->hard_timeout = ntohs(ofm->hard_timeout);
+ flow->used = flow->created = time_now();
+ flow->n_actions = n_actions;
flow->byte_count = 0;
flow->packet_count = 0;
- memcpy(flow->actions, ofm->actions, n_acts * sizeof *flow->actions);
+ memcpy(flow->actions, ofm->actions, n_actions * sizeof *flow->actions);
/* Act. */
error = chain_insert(dp->chain, flow);
uint16_t in_port = ntohs(ofm->match.in_port);
flow_used(flow, buffer);
flow_extract(buffer, in_port, &key.flow);
- execute_actions(dp, buffer, in_port, &key, ofm->actions, n_acts);
+ execute_actions(dp, buffer, in_port, &key,
+ ofm->actions, n_actions, false);
} else {
error = -ESRCH;
}
}
static int
-recv_flow(struct datapath *dp, const void *msg)
+recv_flow(struct datapath *dp, const struct sender *sender UNUSED,
+ const void *msg)
{
const struct ofp_flow_mod *ofm = msg;
uint16_t command = ntohs(ofm->command);
} else if (command == OFPFC_DELETE) {
struct sw_flow_key key;
flow_extract_match(&key, &ofm->match);
- return chain_delete(dp->chain, &key, 0) ? 0 : -ESRCH;
+ return chain_delete(dp->chain, &key, 0, 0) ? 0 : -ESRCH;
} else if (command == OFPFC_DELETE_STRICT) {
struct sw_flow_key key;
+ uint16_t priority;
flow_extract_match(&key, &ofm->match);
- return chain_delete(dp->chain, &key, 1) ? 0 : -ESRCH;
+ priority = key.wildcards ? ntohs(ofm->priority) : -1;
+ return chain_delete(dp->chain, &key, priority, 1) ? 0 : -ESRCH;
} else {
return -ENODEV;
}
}
+static int desc_stats_dump(struct datapath *dp, void *state,
+ struct buffer *buffer)
+{
+ struct ofp_desc_stats *ods = buffer_put_uninit(buffer, sizeof *ods);
+
+ strncpy(ods->mfr_desc, &mfr_desc, sizeof ods->mfr_desc);
+ strncpy(ods->hw_desc, &hw_desc, sizeof ods->hw_desc);
+ strncpy(ods->sw_desc, &sw_desc, sizeof ods->sw_desc);
+ strncpy(ods->serial_num, &serial_num, sizeof ods->serial_num);
+
+ return 0;
+}
+
+struct flow_stats_state {
+ int table_idx;
+ struct sw_table_position position;
+ struct ofp_flow_stats_request rq;
+ time_t now;
+
+ struct buffer *buffer;
+};
+
+#define MAX_FLOW_STATS_BYTES 4096
+
+static int flow_stats_init(struct datapath *dp, const void *body, int body_len,
+ void **state)
+{
+ const struct ofp_flow_stats_request *fsr = body;
+ struct flow_stats_state *s = xmalloc(sizeof *s);
+ s->table_idx = fsr->table_id == 0xff ? 0 : fsr->table_id;
+ memset(&s->position, 0, sizeof s->position);
+ s->rq = *fsr;
+ *state = s;
+ return 0;
+}
+
+static int flow_stats_dump_callback(struct sw_flow *flow, void *private)
+{
+ struct flow_stats_state *s = private;
+ fill_flow_stats(s->buffer, flow, s->table_idx, s->now);
+ return s->buffer->size >= MAX_FLOW_STATS_BYTES;
+}
+
+static int flow_stats_dump(struct datapath *dp, void *state,
+ struct buffer *buffer)
+{
+ struct flow_stats_state *s = state;
+ struct sw_flow_key match_key;
+
+ flow_extract_match(&match_key, &s->rq.match);
+ s->buffer = buffer;
+ s->now = time_now();
+ while (s->table_idx < dp->chain->n_tables
+ && (s->rq.table_id == 0xff || s->rq.table_id == s->table_idx))
+ {
+ struct sw_table *table = dp->chain->tables[s->table_idx];
+
+ if (table->iterate(table, &match_key, &s->position,
+ flow_stats_dump_callback, s))
+ break;
+
+ s->table_idx++;
+ memset(&s->position, 0, sizeof s->position);
+ }
+ return s->buffer->size >= MAX_FLOW_STATS_BYTES;
+}
+
+static void flow_stats_done(void *state)
+{
+ free(state);
+}
+
+struct aggregate_stats_state {
+ struct ofp_aggregate_stats_request rq;
+};
+
+static int aggregate_stats_init(struct datapath *dp,
+ const void *body, int body_len,
+ void **state)
+{
+ const struct ofp_aggregate_stats_request *rq = body;
+ struct aggregate_stats_state *s = xmalloc(sizeof *s);
+ s->rq = *rq;
+ *state = s;
+ return 0;
+}
+
+static int aggregate_stats_dump_callback(struct sw_flow *flow, void *private)
+{
+ struct ofp_aggregate_stats_reply *rpy = private;
+ rpy->packet_count += flow->packet_count;
+ rpy->byte_count += flow->byte_count;
+ rpy->flow_count++;
+ return 0;
+}
+
+static int aggregate_stats_dump(struct datapath *dp, void *state,
+ struct buffer *buffer)
+{
+ struct aggregate_stats_state *s = state;
+ struct ofp_aggregate_stats_request *rq = &s->rq;
+ struct ofp_aggregate_stats_reply *rpy;
+ struct sw_table_position position;
+ struct sw_flow_key match_key;
+ int table_idx;
+
+ rpy = buffer_put_uninit(buffer, sizeof *rpy);
+ memset(rpy, 0, sizeof *rpy);
+
+ flow_extract_match(&match_key, &rq->match);
+ table_idx = rq->table_id == 0xff ? 0 : rq->table_id;
+ memset(&position, 0, sizeof position);
+ while (table_idx < dp->chain->n_tables
+ && (rq->table_id == 0xff || rq->table_id == table_idx))
+ {
+ struct sw_table *table = dp->chain->tables[table_idx];
+ int error;
+
+ error = table->iterate(table, &match_key, &position,
+ aggregate_stats_dump_callback, rpy);
+ if (error)
+ return error;
+
+ table_idx++;
+ memset(&position, 0, sizeof position);
+ }
+
+ rpy->packet_count = htonll(rpy->packet_count);
+ rpy->byte_count = htonll(rpy->byte_count);
+ rpy->flow_count = htonl(rpy->flow_count);
+ return 0;
+}
+
+static void aggregate_stats_done(void *state)
+{
+ free(state);
+}
+
+static int table_stats_dump(struct datapath *dp, void *state,
+ struct buffer *buffer)
+{
+ int i;
+ for (i = 0; i < dp->chain->n_tables; i++) {
+ struct ofp_table_stats *ots = buffer_put_uninit(buffer, sizeof *ots);
+ struct sw_table_stats stats;
+ dp->chain->tables[i]->stats(dp->chain->tables[i], &stats);
+ strncpy(ots->name, stats.name, sizeof ots->name);
+ ots->table_id = i;
+ ots->wildcards = htonl(stats.wildcards);
+ memset(ots->pad, 0, sizeof ots->pad);
+ ots->max_entries = htonl(stats.max_flows);
+ ots->active_count = htonl(stats.n_flows);
+ ots->matched_count = htonll(stats.n_matched);
+ }
+ return 0;
+}
+
+struct port_stats_state {
+ int port;
+};
+
+static int port_stats_init(struct datapath *dp, const void *body, int body_len,
+ void **state)
+{
+ struct port_stats_state *s = xmalloc(sizeof *s);
+ s->port = 0;
+ *state = s;
+ return 0;
+}
+
+static int port_stats_dump(struct datapath *dp, void *state,
+ struct buffer *buffer)
+{
+ struct port_stats_state *s = state;
+ int i;
+
+ for (i = s->port; i < OFPP_MAX; i++) {
+ struct sw_port *p = &dp->ports[i];
+ struct ofp_port_stats *ops;
+ if (!p->netdev) {
+ continue;
+ }
+ ops = buffer_put_uninit(buffer, sizeof *ops);
+ ops->port_no = htons(port_no(dp, p));
+ memset(ops->pad, 0, sizeof ops->pad);
+ ops->rx_packets = htonll(p->rx_packets);
+ ops->tx_packets = htonll(p->tx_packets);
+ ops->rx_bytes = htonll(p->rx_bytes);
+ ops->tx_bytes = htonll(p->tx_bytes);
+ ops->rx_dropped = htonll(-1);
+ ops->tx_dropped = htonll(p->tx_dropped);
+ ops->rx_errors = htonll(-1);
+ ops->tx_errors = htonll(-1);
+ ops->rx_frame_err = htonll(-1);
+ ops->rx_over_err = htonll(-1);
+ ops->rx_crc_err = htonll(-1);
+ ops->collisions = htonll(-1);
+ ops++;
+ }
+ s->port = i;
+ return 0;
+}
+
+static void port_stats_done(void *state)
+{
+ free(state);
+}
+
+struct stats_type {
+ /* Minimum and maximum acceptable number of bytes in body member of
+ * struct ofp_stats_request. */
+ size_t min_body, max_body;
+
+ /* Prepares to dump some kind of statistics on 'dp'. 'body' and
+ * 'body_len' are the 'body' member of the struct ofp_stats_request.
+ * Returns zero if successful, otherwise a negative error code.
+ * May initialize '*state' to state information. May be null if no
+ * initialization is required.*/
+ int (*init)(struct datapath *dp, const void *body, int body_len,
+ void **state);
+
+ /* Appends statistics for 'dp' to 'buffer', which initially contains a
+ * struct ofp_stats_reply. On success, it should return 1 if it should be
+ * called again later with another buffer, 0 if it is done, or a negative
+ * errno value on failure. */
+ int (*dump)(struct datapath *dp, void *state, struct buffer *buffer);
+
+ /* Cleans any state created by the init or dump functions. May be null
+ * if no cleanup is required. */
+ void (*done)(void *state);
+};
+
+static const struct stats_type stats[] = {
+ [OFPST_DESC] = {
+ 0,
+ 0,
+ NULL,
+ desc_stats_dump,
+ NULL
+ },
+ [OFPST_FLOW] = {
+ sizeof(struct ofp_flow_stats_request),
+ sizeof(struct ofp_flow_stats_request),
+ flow_stats_init,
+ flow_stats_dump,
+ flow_stats_done
+ },
+ [OFPST_AGGREGATE] = {
+ sizeof(struct ofp_aggregate_stats_request),
+ sizeof(struct ofp_aggregate_stats_request),
+ aggregate_stats_init,
+ aggregate_stats_dump,
+ aggregate_stats_done
+ },
+ [OFPST_TABLE] = {
+ 0,
+ 0,
+ NULL,
+ table_stats_dump,
+ NULL
+ },
+ [OFPST_PORT] = {
+ 0,
+ 0,
+ port_stats_init,
+ port_stats_dump,
+ port_stats_done
+ },
+};
+
+struct stats_dump_cb {
+ bool done;
+ struct ofp_stats_request *rq;
+ struct sender sender;
+ const struct stats_type *s;
+ void *state;
+};
+
+static int
+stats_dump(struct datapath *dp, void *cb_)
+{
+ struct stats_dump_cb *cb = cb_;
+ struct ofp_stats_reply *osr;
+ struct buffer *buffer;
+ int err;
+
+ if (cb->done) {
+ return 0;
+ }
+
+ osr = make_openflow_reply(sizeof *osr, OFPT_STATS_REPLY, &cb->sender,
+ &buffer);
+ osr->type = htons(cb->s - stats);
+ osr->flags = 0;
+
+ err = cb->s->dump(dp, cb->state, buffer);
+ if (err >= 0) {
+ int err2;
+ if (!err) {
+ cb->done = true;
+ } else {
+ /* Buffer might have been reallocated, so find our data again. */
+ osr = buffer_at_assert(buffer, 0, sizeof *osr);
+ osr->flags = ntohs(OFPSF_REPLY_MORE);
+ }
+ err2 = send_openflow_buffer(dp, buffer, &cb->sender);
+ if (err2) {
+ err = err2;
+ }
+ }
+
+ return err;
+}
+
+static void
+stats_done(void *cb_)
+{
+ struct stats_dump_cb *cb = cb_;
+ if (cb) {
+ if (cb->s->done) {
+ cb->s->done(cb->state);
+ }
+ free(cb);
+ }
+}
+
+static int
+recv_stats_request(struct datapath *dp, const struct sender *sender,
+ const void *oh)
+{
+ const struct ofp_stats_request *rq = oh;
+ size_t rq_len = ntohs(rq->header.length);
+ struct stats_dump_cb *cb;
+ int type, body_len;
+ int err;
+
+ type = ntohs(rq->type);
+ if (type >= ARRAY_SIZE(stats) || !stats[type].dump) {
+ VLOG_WARN_RL(&rl, "received stats request of unknown type %d", type);
+ return -EINVAL;
+ }
+
+ cb = xmalloc(sizeof *cb);
+ cb->done = false;
+ cb->rq = xmemdup(rq, rq_len);
+ cb->sender = *sender;
+ cb->s = &stats[type];
+ cb->state = NULL;
+
+ body_len = rq_len - offsetof(struct ofp_stats_request, body);
+ if (body_len < cb->s->min_body || body_len > cb->s->max_body) {
+ VLOG_WARN_RL(&rl, "stats request type %d with bad body length %d",
+ type, body_len);
+ err = -EINVAL;
+ goto error;
+ }
+
+ if (cb->s->init) {
+ err = cb->s->init(dp, rq->body, body_len, &cb->state);
+ if (err) {
+ VLOG_WARN_RL(&rl,
+ "failed initialization of stats request type %d: %s",
+ type, strerror(-err));
+ goto error;
+ }
+ }
+
+ remote_start_dump(sender->remote, stats_dump, stats_done, cb);
+ return 0;
+
+error:
+ free(cb->rq);
+ free(cb);
+ return err;
+}
+
+static int
+recv_echo_request(struct datapath *dp, const struct sender *sender,
+ const void *oh)
+{
+ return send_openflow_buffer(dp, make_echo_reply(oh), sender);
+}
+
+static int
+recv_echo_reply(struct datapath *dp UNUSED, const struct sender *sender UNUSED,
+ const void *oh UNUSED)
+{
+ return 0;
+}
+
/* 'msg', which is 'length' bytes long, was received from the control path.
* Apply it to 'chain'. */
int
-fwd_control_input(struct datapath *dp, const void *msg, size_t length)
+fwd_control_input(struct datapath *dp, const struct sender *sender,
+ const void *msg, size_t length)
{
-
struct openflow_packet {
size_t min_size;
- int (*handler)(struct datapath *, const void *);
+ int (*handler)(struct datapath *, const struct sender *, const void *);
};
static const struct openflow_packet packets[] = {
sizeof (struct ofp_port_mod),
recv_port_mod,
},
+ [OFPT_STATS_REQUEST] = {
+ sizeof (struct ofp_stats_request),
+ recv_stats_request,
+ },
+ [OFPT_ECHO_REQUEST] = {
+ sizeof (struct ofp_header),
+ recv_echo_request,
+ },
+ [OFPT_ECHO_REPLY] = {
+ sizeof (struct ofp_header),
+ recv_echo_reply,
+ },
};
const struct openflow_packet *pkt;
struct ofp_header *oh;
- if (length < sizeof(struct ofp_header))
- return -EINVAL;
-
oh = (struct ofp_header *) msg;
- if (oh->version != 1 || oh->type >= ARRAY_SIZE(packets)
- || ntohs(oh->length) > length)
+ assert(oh->version == OFP_VERSION);
+ if (oh->type >= ARRAY_SIZE(packets) || ntohs(oh->length) > length)
return -EINVAL;
pkt = &packets[oh->type];
if (length < pkt->min_size)
return -EFAULT;
- return pkt->handler(dp, msg);
+ return pkt->handler(dp, sender, msg);
}
\f
/* Packet buffering. */
if (p->buffer) {
/* Don't buffer packet if existing entry is less than
* OVERWRITE_SECS old. */
- if (time(0) < p->timeout) { /* FIXME */
+ if (time_now() < p->timeout) { /* FIXME */
return -1;
} else {
buffer_delete(p->buffer);
if (++p->cookie >= (1u << PKT_COOKIE_BITS) - 1)
p->cookie = 0;
p->buffer = buffer_clone(buffer); /* FIXME */
- p->timeout = time(0) + OVERWRITE_SECS; /* FIXME */
+ p->timeout = time_now() + OVERWRITE_SECS; /* FIXME */
id = buffer_idx | (p->cookie << PKT_BUFFER_BITS);
return id;
git://git.onelab.eu / sliver-openvswitch.git / blobdiff