/*
* Distributed under the terms of the GNU GPL version 2.
- * Copyright (c) 2007 The Board of Trustees of The Leland Stanford Junior Univer
-sity
+ * Copyright (c) 2007, 2008 The Board of Trustees of The Leland
+ * Stanford Junior University
*/
/* Functions for managing the dp interface/device. */
+#include <linux/init.h>
#include <linux/module.h>
#include <linux/if_arp.h>
#include <linux/if_bridge.h>
#include <linux/in.h>
#include <net/genetlink.h>
#include <linux/ip.h>
+#include <linux/delay.h>
#include <linux/etherdevice.h>
#include <linux/kernel.h>
+#include <linux/kthread.h>
#include <linux/mutex.h>
#include <linux/rtnetlink.h>
-#include <linux/timer.h>
#include <linux/rcupdate.h>
#include <linux/version.h>
#include <linux/ethtool.h>
#include "compat.h"
-/* Number of seconds between runs of the flow expiration code. */
-#define EXPIRE_SECS 1
+/* Number of milliseconds between runs of the maintenance thread. */
+#define MAINT_SLEEP_MSECS 1000
#define BRIDGE_PORT_NO_FLOOD 0x00000001
#define UINT32_MAX 4294967295U
+#define MAX(X, Y) ((X) > (Y) ? (X) : (Y))
struct net_bridge_port {
u16 port_no;
static struct datapath *dps[DP_MAX];
static DEFINE_MUTEX(dp_mutex);
-static void dp_timer_handler(unsigned long arg);
+static int dp_maint_func(void *data);
static int send_port_status(struct net_bridge_port *p, uint8_t status);
nla->nla_len -= len;
}
+static void *
+alloc_openflow_skb(struct datapath *dp, size_t openflow_len, uint8_t type,
+ const struct sender *sender, struct sk_buff **pskb)
+{
+ size_t genl_len;
+ struct sk_buff *skb;
+ struct nlattr *attr;
+ struct ofp_header *oh;
+
+ genl_len = nla_total_size(sizeof(uint32_t)); /* DP_GENL_A_DP_IDX */
+ genl_len += nla_total_size(openflow_len); /* DP_GENL_A_OPENFLOW */
+ skb = *pskb = genlmsg_new(genl_len, GFP_ATOMIC);
+ if (!skb) {
+ if (net_ratelimit())
+ printk("alloc_openflow_skb: genlmsg_new failed\n");
+ return NULL;
+ }
+
+ /* Assemble the Generic Netlink wrapper. */
+ if (!genlmsg_put(skb,
+ sender ? sender->pid : 0,
+ sender ? sender->seq : 0,
+ &dp_genl_family, 0, DP_GENL_C_OPENFLOW))
+ BUG();
+ if (nla_put_u32(skb, DP_GENL_A_DP_IDX, dp->dp_idx) < 0)
+ BUG();
+ attr = nla_reserve(skb, DP_GENL_A_OPENFLOW, openflow_len);
+ BUG_ON(!attr);
+ nlmsg_end(skb, (struct nlmsghdr *) skb->data);
+
+ /* Fill in the header. */
+ oh = nla_data(attr);
+ oh->version = OFP_VERSION;
+ oh->type = type;
+ oh->length = htons(openflow_len);
+ oh->xid = sender ? sender->xid : 0;
+
+ return oh;
+}
+
+static void
+resize_openflow_skb(struct sk_buff *skb,
+ struct ofp_header *oh, size_t new_length)
+{
+ struct nlattr *attr;
+
+ BUG_ON(new_length > ntohs(oh->length));
+ attr = ((void *) oh) - NLA_HDRLEN;
+ nla_unreserve(skb, attr, ntohs(oh->length) - new_length);
+ oh->length = htons(new_length);
+ nlmsg_end(skb, (struct nlmsghdr *) skb->data);
+}
+
+static int
+send_openflow_skb(struct sk_buff *skb, const struct sender *sender)
+{
+ int err = (sender
+ ? genlmsg_unicast(skb, sender->pid)
+ : genlmsg_multicast(skb, 0, mc_group.id, GFP_ATOMIC));
+ if (err && net_ratelimit())
+ printk(KERN_WARNING "send_openflow_skb: send failed: %d\n",
+ err);
+ return err;
+}
+
/* Generates a unique datapath id. It incorporates the datapath index
* and a hardware address, if available. If not, it generates a random
* one.
printk("datapath: problem setting up 'of' device\n");
#endif
+ dp->flags = 0;
dp->miss_send_len = OFP_DEFAULT_MISS_SEND_LEN;
- setup_timer(&dp->timer, dp_timer_handler, (unsigned long) dp);
- mod_timer(&dp->timer, round_jiffies(jiffies + (EXPIRE_SECS * HZ)));
+ dp->dp_task = kthread_run(dp_maint_func, dp, "dp%d", dp_idx);
+ if (IS_ERR(dp->dp_task))
+ goto err_free_dp;
rcu_assign_pointer(dps[dp_idx], dp);
mutex_unlock(&dp_mutex);
rtnl_unlock();
#endif
+ kthread_stop(dp->dp_task);
+
/* Drop references to DP. */
list_for_each_entry_safe (p, n, &dp->port_list, node)
del_switch_port(p);
- del_timer_sync(&dp->timer);
rcu_assign_pointer(dps[dp->dp_idx], NULL);
/* Wait until no longer in use, then destroy it. */
module_put(THIS_MODULE);
}
-static void dp_timer_handler(unsigned long arg)
+static int dp_maint_func(void *data)
{
- struct datapath *dp = (struct datapath *) arg;
+ struct datapath *dp = (struct datapath *) data;
+
+ while (!kthread_should_stop()) {
#if 1
- chain_timeout(dp->chain);
+ chain_timeout(dp->chain);
#else
- int count = chain_timeout(dp->chain);
- chain_print_stats(dp->chain);
- if (count)
- printk("%d flows timed out\n", count);
+ int count = chain_timeout(dp->chain);
+ chain_print_stats(dp->chain);
+ if (count)
+ printk("%d flows timed out\n", count);
#endif
- mod_timer(&dp->timer, round_jiffies(jiffies + (EXPIRE_SECS * HZ)));
+ msleep_interruptible(MAINT_SLEEP_MSECS);
+ }
+
+ return 0;
}
/*
else if (out_port == OFPP_CONTROLLER)
return dp_output_control(dp, skb, fwd_save_skb(skb), 0,
OFPR_ACTION);
- else if (out_port >= OFPP_MAX)
+ else if (out_port == OFPP_TABLE) {
+ struct sw_flow_key key;
+ struct sw_flow *flow;
+
+ flow_extract(skb, skb->dev->br_port->port_no, &key);
+ flow = chain_lookup(dp->chain, &key);
+ if (likely(flow != NULL)) {
+ flow_used(flow, skb);
+ execute_actions(dp, skb, &key, flow->actions, flow->n_actions);
+ return 0;
+ }
+ return -ESRCH;
+ } else if (out_port >= OFPP_MAX)
goto bad_port;
p = dp->ports[out_port];
dp_output_control(struct datapath *dp, struct sk_buff *skb,
uint32_t buffer_id, size_t max_len, int reason)
{
- /* FIXME? packet_rcv_spkt in net/packet/af_packet.c does some stuff
- that we should possibly be doing here too. */
/* FIXME? Can we avoid creating a new skbuff in the case where we
* forward the whole packet? */
struct sk_buff *f_skb;
- struct nlattr *attr;
struct ofp_packet_in *opi;
- size_t opi_len;
- size_t len, fwd_len;
- void *data;
- int err = -ENOMEM;
+ size_t fwd_len, opi_len;
+ int err;
fwd_len = skb->len;
if ((buffer_id != (uint32_t) -1) && max_len)
fwd_len = min(fwd_len, max_len);
- len = nla_total_size(offsetof(struct ofp_packet_in, data) + fwd_len)
- + nla_total_size(sizeof(uint32_t));
-
- f_skb = genlmsg_new(len, GFP_ATOMIC);
- if (!f_skb)
- goto error_free_skb;
-
- data = genlmsg_put(f_skb, 0, 0, &dp_genl_family, 0,
- DP_GENL_C_OPENFLOW);
- if (data == NULL)
- goto error_free_f_skb;
-
- NLA_PUT_U32(f_skb, DP_GENL_A_DP_IDX, dp->dp_idx);
-
opi_len = offsetof(struct ofp_packet_in, data) + fwd_len;
- attr = nla_reserve(f_skb, DP_GENL_A_OPENFLOW, opi_len);
- if (!attr)
- goto error_free_f_skb;
- opi = nla_data(attr);
- opi->header.version = OFP_VERSION;
- opi->header.type = OFPT_PACKET_IN;
- opi->header.length = htons(opi_len);
- opi->header.xid = htonl(0);
-
+ opi = alloc_openflow_skb(dp, opi_len, OFPT_PACKET_IN, NULL, &f_skb);
opi->buffer_id = htonl(buffer_id);
opi->total_len = htons(skb->len);
opi->in_port = htons(skb->dev->br_port->port_no);
opi->reason = reason;
opi->pad = 0;
- SKB_LINEAR_ASSERT(skb);
memcpy(opi->data, skb_mac_header(skb), fwd_len);
+ err = send_openflow_skb(f_skb, NULL);
- err = genlmsg_end(f_skb, data);
- if (err < 0)
- goto error_free_f_skb;
-
- err = genlmsg_multicast(f_skb, 0, mc_group.id, GFP_ATOMIC);
- if (err && net_ratelimit())
- printk(KERN_WARNING "dp_output_control: genlmsg_multicast failed: %d\n", err);
-
- kfree_skb(skb);
-
- return err;
-
-nla_put_failure:
-error_free_f_skb:
- nlmsg_free(f_skb);
-error_free_skb:
kfree_skb(skb);
- if (net_ratelimit())
- printk(KERN_ERR "dp_output_control: failed to send: %d\n", err);
+
return err;
}
}
static int
-fill_data_hello(struct datapath *dp, struct ofp_data_hello *odh)
+fill_features_reply(struct datapath *dp, struct ofp_switch_features *ofr)
{
struct net_bridge_port *p;
int port_count = 0;
- odh->header.version = OFP_VERSION;
- odh->header.type = OFPT_DATA_HELLO;
- odh->header.xid = htonl(0);
- odh->datapath_id = cpu_to_be64(dp->id);
-
- odh->n_exact = htonl(2 * TABLE_HASH_MAX_FLOWS);
- odh->n_mac_only = htonl(TABLE_MAC_MAX_FLOWS);
- odh->n_compression = 0; /* Not supported */
- odh->n_general = htonl(TABLE_LINEAR_MAX_FLOWS);
- odh->buffer_mb = htonl(UINT32_MAX);
- odh->n_buffers = htonl(N_PKT_BUFFERS);
- odh->capabilities = htonl(OFP_SUPPORTED_CAPABILITIES);
- odh->actions = htonl(OFP_SUPPORTED_ACTIONS);
- odh->miss_send_len = htons(dp->miss_send_len);
+ ofr->datapath_id = cpu_to_be64(dp->id);
+
+ ofr->n_exact = htonl(2 * TABLE_HASH_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);
list_for_each_entry_rcu (p, &dp->port_list, node) {
- fill_port_desc(p, &odh->ports[port_count]);
+ fill_port_desc(p, &ofr->ports[port_count]);
port_count++;
}
}
int
-dp_send_hello(struct datapath *dp)
+dp_send_features_reply(struct datapath *dp, const struct sender *sender)
{
struct sk_buff *skb;
- struct nlattr *attr;
- struct ofp_data_hello *odh;
- size_t odh_max_len, odh_len, port_max_len, len;
- void *data;
- int err = -ENOMEM;
+ struct ofp_switch_features *ofr;
+ size_t ofr_len, port_max_len;
int port_count;
-
- /* Overallocate, since we can't reliably determine the number of
- * ports a priori. */
+ /* Overallocate. */
port_max_len = sizeof(struct ofp_phy_port) * OFPP_MAX;
+ ofr = alloc_openflow_skb(dp, sizeof(*ofr) + port_max_len,
+ OFPT_FEATURES_REPLY, sender, &skb);
+ if (!ofr)
+ return -ENOMEM;
- len = nla_total_size(sizeof(*odh) + port_max_len)
- + nla_total_size(sizeof(uint32_t));
-
- skb = genlmsg_new(len, GFP_ATOMIC);
- if (!skb) {
- if (net_ratelimit())
- printk("dp_send_hello: genlmsg_new failed\n");
- goto error;
- }
-
- data = genlmsg_put(skb, 0, 0, &dp_genl_family, 0,
- DP_GENL_C_OPENFLOW);
- if (data == NULL) {
- if (net_ratelimit())
- printk("dp_send_hello: genlmsg_put failed\n");
- goto error;
- }
-
- NLA_PUT_U32(skb, DP_GENL_A_DP_IDX, dp->dp_idx);
-
- odh_max_len = sizeof(*odh) + port_max_len;
- attr = nla_reserve(skb, DP_GENL_A_OPENFLOW, odh_max_len);
- if (!attr) {
- if (net_ratelimit())
- printk("dp_send_hello: nla_reserve failed\n");
- goto error;
- }
- odh = nla_data(attr);
- port_count = fill_data_hello(dp, odh);
-
- /* Only now that we know how many ports we've added can we say
- * say something about the length. */
- odh_len = sizeof(*odh) + (sizeof(struct ofp_phy_port) * port_count);
- odh->header.length = htons(odh_len);
+ /* Fill. */
+ port_count = fill_features_reply(dp, ofr);
- /* Take back the unused part that was reserved */
- nla_unreserve(skb, attr, (odh_max_len - odh_len));
+ /* Shrink to fit. */
+ ofr_len = sizeof(*ofr) + (sizeof(struct ofp_phy_port) * port_count);
+ resize_openflow_skb(skb, &ofr->header, ofr_len);
+ return send_openflow_skb(skb, sender);
+}
- err = genlmsg_end(skb, data);
- if (err < 0) {
- if (net_ratelimit())
- printk("dp_send_hello: genlmsg_end failed\n");
- goto error;
- }
+int
+dp_send_config_reply(struct datapath *dp, const struct sender *sender)
+{
+ struct sk_buff *skb;
+ struct ofp_switch_config *osc;
- err = genlmsg_multicast(skb, 0, mc_group.id, GFP_ATOMIC);
- if (err && net_ratelimit())
- printk(KERN_WARNING "dp_send_hello: genlmsg_multicast failed: %d\n", err);
+ osc = alloc_openflow_skb(dp, sizeof *osc, OFPT_GET_CONFIG_REPLY, sender,
+ &skb);
+ if (!osc)
+ return -ENOMEM;
- return err;
+ osc->flags = htons(dp->flags);
+ osc->miss_send_len = htons(dp->miss_send_len);
-nla_put_failure:
-error:
- kfree_skb(skb);
- if (net_ratelimit())
- printk(KERN_ERR "dp_send_hello: failed to send: %d\n", err);
- return err;
+ return send_openflow_skb(skb, sender);
}
int
send_port_status(struct net_bridge_port *p, uint8_t status)
{
struct sk_buff *skb;
- struct nlattr *attr;
struct ofp_port_status *ops;
- void *data;
- int err = -ENOMEM;
-
-
- skb = genlmsg_new(NLMSG_GOODSIZE, GFP_ATOMIC);
- if (!skb) {
- if (net_ratelimit())
- printk("send_port_status: genlmsg_new failed\n");
- goto error;
- }
-
- data = genlmsg_put(skb, 0, 0, &dp_genl_family, 0,
- DP_GENL_C_OPENFLOW);
- if (data == NULL) {
- if (net_ratelimit())
- printk("send_port_status: genlmsg_put failed\n");
- goto error;
- }
-
- NLA_PUT_U32(skb, DP_GENL_A_DP_IDX, p->dp->dp_idx);
-
- attr = nla_reserve(skb, DP_GENL_A_OPENFLOW, sizeof(*ops));
- if (!attr) {
- if (net_ratelimit())
- printk("send_port_status: nla_reserve failed\n");
- goto error;
- }
-
- ops = nla_data(attr);
- 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 = alloc_openflow_skb(p->dp, sizeof *ops, OFPT_PORT_STATUS, NULL,
+ &skb);
+ if (!ops)
+ return -ENOMEM;
+ ops->reason = status;
fill_port_desc(p, &ops->desc);
- err = genlmsg_end(skb, data);
- if (err < 0) {
- if (net_ratelimit())
- printk("send_port_status: genlmsg_end failed\n");
- goto error;
- }
-
- err = genlmsg_multicast(skb, 0, mc_group.id, GFP_ATOMIC);
- if (err && net_ratelimit())
- printk(KERN_WARNING "send_port_status: genlmsg_multicast failed: %d\n", err);
-
- return err;
-
-nla_put_failure:
-error:
- kfree_skb(skb);
- if (net_ratelimit())
- printk(KERN_ERR "send_port_status: failed to send: %d\n", err);
- return err;
+ return send_openflow_skb(skb, NULL);
}
int
dp_send_flow_expired(struct datapath *dp, struct sw_flow *flow)
{
struct sk_buff *skb;
- struct nlattr *attr;
struct ofp_flow_expired *ofe;
- void *data;
unsigned long duration_j;
- int err = -ENOMEM;
+ ofe = alloc_openflow_skb(dp, sizeof *ofe, OFPT_FLOW_EXPIRED, 0, &skb);
+ if (!ofe)
+ return -ENOMEM;
- skb = genlmsg_new(NLMSG_GOODSIZE, GFP_ATOMIC);
- if (!skb) {
- if (net_ratelimit())
- printk("dp_send_flow_expired: genlmsg_new failed\n");
- goto error;
- }
+ flow_fill_match(&ofe->match, &flow->key);
+ duration_j = (flow->timeout - HZ * flow->max_idle) - flow->init_time;
+ ofe->duration = htonl(duration_j / HZ);
+ ofe->packet_count = cpu_to_be64(flow->packet_count);
+ ofe->byte_count = cpu_to_be64(flow->byte_count);
+ return send_openflow_skb(skb, NULL);
+}
- data = genlmsg_put(skb, 0, 0, &dp_genl_family, 0,
- DP_GENL_C_OPENFLOW);
- if (data == NULL) {
- if (net_ratelimit())
- printk("dp_send_flow_expired: genlmsg_put failed\n");
- goto error;
+static void
+fill_flow_stats(struct ofp_flow_stats *ofs, struct sw_flow *flow,
+ int table_idx)
+{
+ ofs->match.wildcards = htons(flow->key.wildcards);
+ ofs->match.in_port = flow->key.in_port;
+ memcpy(ofs->match.dl_src, flow->key.dl_src, ETH_ALEN);
+ memcpy(ofs->match.dl_dst, flow->key.dl_dst, ETH_ALEN);
+ ofs->match.dl_vlan = flow->key.dl_vlan;
+ ofs->match.dl_type = flow->key.dl_type;
+ ofs->match.nw_src = flow->key.nw_src;
+ ofs->match.nw_dst = flow->key.nw_dst;
+ ofs->match.nw_proto = flow->key.nw_proto;
+ memset(ofs->match.pad, 0, sizeof ofs->match.pad);
+ ofs->match.tp_src = flow->key.tp_src;
+ ofs->match.tp_dst = flow->key.tp_dst;
+ ofs->duration = htonl((jiffies - flow->init_time) / HZ);
+ ofs->table_id = table_idx;
+ ofs->packet_count = cpu_to_be64(flow->packet_count);
+ ofs->byte_count = cpu_to_be64(flow->byte_count);
+}
+
+int
+dp_send_flow_stats(struct datapath *dp, const struct sender *sender,
+ const struct ofp_match *match)
+{
+ struct sk_buff *skb;
+ struct ofp_flow_stats_reply *fsr;
+ size_t header_size, fudge, flow_size;
+ struct sw_flow_key match_key;
+ int table_idx, n_flows, max_flows;
+
+ header_size = offsetof(struct ofp_flow_stats_reply, flows);
+ fudge = 128;
+ flow_size = sizeof fsr->flows[0];
+ max_flows = (NLMSG_GOODSIZE - header_size - fudge) / flow_size;
+ fsr = alloc_openflow_skb(dp, header_size + max_flows * flow_size,
+ OFPT_FLOW_STATS_REPLY, sender, &skb);
+ if (!fsr)
+ return -ENOMEM;
+
+ n_flows = 0;
+ flow_extract_match(&match_key, match);
+ for (table_idx = 0; table_idx < dp->chain->n_tables; table_idx++) {
+ struct sw_table *table = dp->chain->tables[table_idx];
+ struct swt_iterator iter;
+
+ if (n_flows >= max_flows) {
+ break;
+ }
+
+ if (!table->iterator(table, &iter)) {
+ if (net_ratelimit())
+ printk("iterator failed for table %d\n",
+ table_idx);
+ continue;
+ }
+
+ for (; iter.flow; table->iterator_next(&iter)) {
+ if (flow_matches(&match_key, &iter.flow->key)) {
+ fill_flow_stats(&fsr->flows[n_flows],
+ iter.flow, table_idx);
+ if (++n_flows >= max_flows) {
+ break;
+ }
+ }
+ }
+ table->iterator_destroy(&iter);
}
+ resize_openflow_skb(skb, &fsr->header,
+ header_size + flow_size * n_flows);
+ return send_openflow_skb(skb, sender);
+}
- NLA_PUT_U32(skb, DP_GENL_A_DP_IDX, dp->dp_idx);
+static int
+fill_port_stats_reply(struct datapath *dp, struct ofp_port_stats_reply *psr)
+{
+ struct net_bridge_port *p;
+ int port_count = 0;
- attr = nla_reserve(skb, DP_GENL_A_OPENFLOW, sizeof(*ofe));
- if (!attr) {
- if (net_ratelimit())
- printk("dp_send_flow_expired: nla_reserve failed\n");
- goto error;
+ list_for_each_entry_rcu (p, &dp->port_list, node) {
+ struct ofp_port_stats *ps = &psr->ports[port_count++];
+ struct net_device_stats *stats = p->dev->get_stats(p->dev);
+ ps->port_no = htons(p->port_no);
+ memset(ps->pad, 0, sizeof ps->pad);
+ ps->rx_count = cpu_to_be64(stats->rx_packets);
+ ps->tx_count = cpu_to_be64(stats->tx_packets);
+ ps->drop_count = cpu_to_be64(stats->rx_dropped
+ + stats->tx_dropped);
}
- ofe = nla_data(attr);
- ofe->header.version = OFP_VERSION;
- ofe->header.type = OFPT_FLOW_EXPIRED;
- ofe->header.length = htons(sizeof(*ofe));
- ofe->header.xid = htonl(0);
+ return port_count;
+}
- flow_fill_match(&ofe->match, &flow->key);
- duration_j = (flow->timeout - HZ * flow->max_idle) - flow->init_time;
- ofe->duration = htonl(duration_j / HZ);
- ofe->packet_count = cpu_to_be64(flow->packet_count);
- ofe->byte_count = cpu_to_be64(flow->byte_count);
+int
+dp_send_port_stats(struct datapath *dp, const struct sender *sender)
+{
+ struct sk_buff *skb;
+ struct ofp_port_stats_reply *psr;
+ size_t psr_len, port_max_len;
+ int port_count;
- err = genlmsg_end(skb, data);
- if (err < 0) {
- if (net_ratelimit())
- printk("dp_send_flow_expired: genlmsg_end failed\n");
- goto error;
- }
+ /* Overallocate. */
+ port_max_len = sizeof(struct ofp_port_stats) * OFPP_MAX;
+ psr = alloc_openflow_skb(dp, sizeof *psr + port_max_len,
+ OFPT_PORT_STATS_REPLY, sender, &skb);
+ if (!psr)
+ return -ENOMEM;
- err = genlmsg_multicast(skb, 0, mc_group.id, GFP_ATOMIC);
- if (err && net_ratelimit())
- printk(KERN_WARNING "send_flow_expired: genlmsg_multicast failed: %d\n", err);
+ /* Fill. */
+ port_count = fill_port_stats_reply(dp, psr);
- return err;
+ /* Shrink to fit. */
+ psr_len = sizeof *psr + sizeof(struct ofp_port_stats) * port_count;
+ resize_openflow_skb(skb, &psr->header, psr_len);
+ return send_openflow_skb(skb, sender);
+}
-nla_put_failure:
-error:
- kfree_skb(skb);
- if (net_ratelimit())
- printk(KERN_ERR "send_flow_expired: failed to send: %d\n", err);
- return err;
+int
+dp_send_table_stats(struct datapath *dp, const struct sender *sender)
+{
+ struct sk_buff *skb;
+ struct ofp_table_stats_reply *tsr;
+ int i, n_tables;
+
+ n_tables = dp->chain->n_tables;
+ tsr = alloc_openflow_skb(dp, (offsetof(struct ofp_table_stats_reply,
+ tables)
+ + sizeof tsr->tables[0] * n_tables),
+ OFPT_TABLE_STATS_REPLY, sender, &skb);
+ if (!tsr)
+ return -ENOMEM;
+ for (i = 0; i < n_tables; i++) {
+ struct ofp_table_stats *ots = &tsr->tables[i];
+ 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->pad[0] = ots->pad[1] = 0;
+ ots->max_entries = htonl(stats.max_flows);
+ ots->active_count = htonl(stats.n_flows);
+ ots->matched_count = cpu_to_be64(0); /* FIXME */
+ }
+ return send_openflow_skb(skb, sender);
}
/* Generic Netlink interface.
return err;
}
-/*
- * Fill flow entry for nl flow query. Called with rcu_lock
- *
- */
-static
-int
-dp_fill_flow(struct ofp_flow_mod* ofm, struct swt_iterator* iter)
-{
- ofm->header.version = OFP_VERSION;
- ofm->header.type = OFPT_FLOW_MOD;
- ofm->header.length = htons(sizeof(struct ofp_flow_mod)
- + sizeof(ofm->actions[0]));
- ofm->header.xid = htonl(0);
-
- ofm->match.wildcards = htons(iter->flow->key.wildcards);
- ofm->match.in_port = iter->flow->key.in_port;
- ofm->match.dl_vlan = iter->flow->key.dl_vlan;
- memcpy(ofm->match.dl_src, iter->flow->key.dl_src, ETH_ALEN);
- memcpy(ofm->match.dl_dst, iter->flow->key.dl_dst, ETH_ALEN);
- ofm->match.dl_type = iter->flow->key.dl_type;
- ofm->match.nw_src = iter->flow->key.nw_src;
- ofm->match.nw_dst = iter->flow->key.nw_dst;
- ofm->match.nw_proto = iter->flow->key.nw_proto;
- ofm->match.tp_src = iter->flow->key.tp_src;
- ofm->match.tp_dst = iter->flow->key.tp_dst;
- ofm->group_id = iter->flow->group_id;
- ofm->max_idle = iter->flow->max_idle;
- /* TODO support multiple actions */
- ofm->actions[0] = iter->flow->actions[0];
-
- return 0;
-}
-
-static int dp_genl_show(struct sk_buff *skb, struct genl_info *info)
-{
- struct datapath *dp;
- int err = -ENOMEM;
- struct sk_buff *ans_skb = NULL;
- void *data;
- struct nlattr *attr;
- struct ofp_data_hello *odh;
- size_t odh_max_len, odh_len, port_max_len, len;
- int port_count;
-
- if (!info->attrs[DP_GENL_A_DP_IDX])
- return -EINVAL;
-
- mutex_lock(&dp_mutex);
- dp = dp_get(nla_get_u32((info->attrs[DP_GENL_A_DP_IDX])));
- if (!dp)
- goto error;
-
- /* Overallocate, since we can't reliably determine the number of
- * ports a priori. */
- port_max_len = sizeof(struct ofp_phy_port) * OFPP_MAX;
-
- len = nla_total_size(sizeof(*odh) + port_max_len)
- + nla_total_size(sizeof(uint32_t));
-
- ans_skb = nlmsg_new(len, GFP_KERNEL);
- if (!ans_skb)
- goto error;
-
- data = genlmsg_put_reply(ans_skb, info, &dp_genl_family,
- 0, DP_GENL_C_SHOW_DP);
- if (data == NULL)
- goto error;
-
- NLA_PUT_U32(ans_skb, DP_GENL_A_DP_IDX, dp->dp_idx);
-
- odh_max_len = sizeof(*odh) + port_max_len;
- attr = nla_reserve(ans_skb, DP_GENL_A_DP_INFO, odh_max_len);
- if (!attr)
- goto error;
- odh = nla_data(attr);
- port_count = fill_data_hello(dp, odh);
-
- /* Only now that we know how many ports we've added can we say
- * say something about the length. */
- odh_len = sizeof(*odh) + (sizeof(struct ofp_phy_port) * port_count);
- odh->header.length = htons(odh_len);
-
- /* Take back the unused part that was reserved */
- nla_unreserve(ans_skb, attr, (odh_max_len - odh_len));
-
- genlmsg_end(ans_skb, data);
- err = genlmsg_reply(ans_skb, info);
- if (!err)
- ans_skb = NULL;
-
-error:
-nla_put_failure:
- if (ans_skb)
- kfree_skb(ans_skb);
- mutex_unlock(&dp_mutex);
- return err;
-}
-
-static struct genl_ops dp_genl_ops_show_dp = {
- .cmd = DP_GENL_C_SHOW_DP,
- .flags = GENL_ADMIN_PERM, /* Requires CAP_NET_ADMIN privilege. */
- .policy = dp_genl_policy,
- .doit = dp_genl_show,
- .dumpit = NULL,
-};
-
-/* Convenience function */
-static
-void*
-dp_init_nl_flow_msg(uint32_t dp_idx, uint16_t table_idx,
- struct genl_info *info, struct sk_buff* skb)
-{
- void* data;
-
- data = genlmsg_put_reply(skb, info, &dp_genl_family, 0,
- DP_GENL_C_QUERY_FLOW);
- if (data == NULL)
- return NULL;
- NLA_PUT_U32(skb, DP_GENL_A_DP_IDX, dp_idx);
- NLA_PUT_U16(skb, DP_GENL_A_TABLEIDX, table_idx);
-
- return data;
-
-nla_put_failure:
- return NULL;
-}
-
-/* Iterate through the specified table and send all flow entries over
- * netlink to userspace. Each flow message has the following format:
- *
- * 32bit dpix
- * 16bit tabletype
- * 32bit number of flows
- * openflow-flow-entries
- *
- * The full table may require multiple messages. A message with 0 flows
- * signifies end-of message.
- */
-
-static
-int
-dp_dump_table(struct datapath *dp, uint16_t table_idx, struct genl_info *info, struct ofp_flow_mod* matchme)
-{
- struct sk_buff *skb = 0;
- struct sw_table *table = 0;
- struct swt_iterator iter;
- struct sw_flow_key in_flow;
- struct nlattr *attr;
- int count = 0, sum_count = 0;
- void *data;
- uint8_t* ofm_ptr = 0;
- struct nlattr *num_attr;
- int err = -ENOMEM;
-
- table = dp->chain->tables[table_idx];
- if ( table == NULL ) {
- dprintk("dp::dp_dump_table error, non-existant table at position %d\n", table_idx);
- return -EINVAL;
- }
-
- if (!table->iterator(table, &iter)) {
- dprintk("dp::dp_dump_table couldn't initialize empty table iterator\n");
- return -ENOMEM;
- }
-
- while (iter.flow) {
-
- /* verify that we can fit all NL_FLOWS_PER_MESSAGE in a single
- * sk_buf */
- if( (sizeof(dp_genl_family) + sizeof(uint32_t) + sizeof(uint16_t) + sizeof(uint32_t) +
- (NL_FLOWS_PER_MESSAGE * sizeof(struct ofp_flow_mod))) > (8192 - 64)){
- dprintk("dp::dp_dump_table NL_FLOWS_PER_MESSAGE may cause overrun in skbuf\n");
- return -ENOMEM;
- }
-
- skb = nlmsg_new(8192 - 64, GFP_ATOMIC);
- if (skb == NULL) {
- return -ENOMEM;
- }
-
- data = dp_init_nl_flow_msg(dp->dp_idx, table_idx, info, skb);
- if (data == NULL){
- err= -ENOMEM;
- goto error_free_skb;
- }
-
- /* reserve space to put the number of flows for this message, to
- * be filled after the loop*/
- num_attr = nla_reserve(skb, DP_GENL_A_NUMFLOWS, sizeof(uint32_t));
- if(!num_attr){
- err = -ENOMEM;
- goto error_free_skb;
- }
-
- /* Only load NL_FLOWS_PER_MESSAGE flows at a time */
- attr = nla_reserve(skb, DP_GENL_A_FLOW,
- (sizeof(struct ofp_flow_mod) + sizeof(struct ofp_action)) * NL_FLOWS_PER_MESSAGE);
- if (!attr){
- err = -ENOMEM;
- goto error_free_skb;
- }
-
- /* internal loop to fill NL_FLOWS_PER_MESSAGE flows */
- ofm_ptr = nla_data(attr);
- flow_extract_match(&in_flow, &matchme->match);
- while (iter.flow && count < NL_FLOWS_PER_MESSAGE) {
- if(flow_matches(&in_flow, &iter.flow->key)){
- if((err = dp_fill_flow((struct ofp_flow_mod*)ofm_ptr, &iter)))
- goto error_free_skb;
- count++;
- /* TODO support multiple actions */
- ofm_ptr += sizeof(struct ofp_flow_mod) + sizeof(struct ofp_action);
- }
- table->iterator_next(&iter);
- }
-
- *((uint32_t*)nla_data(num_attr)) = count;
- genlmsg_end(skb, data);
-
- sum_count += count;
- count = 0;
-
- err = genlmsg_unicast(skb, info->snd_pid);
- skb = 0;
- }
-
- /* send a sentinal message saying we're done */
- skb = nlmsg_new(NLMSG_GOODSIZE, GFP_ATOMIC);
- if (skb == NULL) {
- return -ENOMEM;
- }
- data = dp_init_nl_flow_msg(dp->dp_idx, table_idx, info, skb);
- if (data == NULL){
- err= -ENOMEM;
- goto error_free_skb;
- }
-
- NLA_PUT_U32(skb, DP_GENL_A_NUMFLOWS, 0);
- /* dummy flow so nl doesn't complain */
- attr = nla_reserve(skb, DP_GENL_A_FLOW, sizeof(struct ofp_flow_mod));
- if (!attr){
- err = -ENOMEM;
- goto error_free_skb;
- }
- genlmsg_end(skb, data);
- err = genlmsg_reply(skb, info); skb = 0;
-
-nla_put_failure:
-error_free_skb:
- if(skb)
- kfree_skb(skb);
- return err;
-}
-
-/* Helper function to query_table which creates and sends a message packed with
- * table stats. Message form is:
- *
- * u32 DP_IDX
- * u32 NUM_TABLES
- * OFP_TABLE (list of OFP_TABLES)
- *
- */
-
-static
-int
-dp_dump_table_stats(struct datapath *dp, int dp_idx, struct genl_info *info)
-{
- struct sk_buff *skb = 0;
- struct ofp_table *ot = 0;
- struct nlattr *attr;
- struct sw_table_stats stats;
- void *data;
- int err = -ENOMEM;
- int i = 0;
- int nt = dp->chain->n_tables;
-
- /* u32 IDX, u32 NUMTABLES, list-of-tables */
- skb = nlmsg_new(4 + 4 + (sizeof(struct ofp_table) * nt), GFP_ATOMIC);
- if (skb == NULL) {
- return -ENOMEM;
- }
-
- data = genlmsg_put_reply(skb, info, &dp_genl_family, 0,
- DP_GENL_C_QUERY_TABLE);
- if (data == NULL){
- return -ENOMEM;
- }
-
- NLA_PUT_U32(skb, DP_GENL_A_DP_IDX, dp_idx);
- NLA_PUT_U32(skb, DP_GENL_A_NUMTABLES, nt);
-
- /* ... we assume that all tables can fit in a single message.
- * Probably a reasonable assumption seeing that we only have
- * 3 atm */
- attr = nla_reserve(skb, DP_GENL_A_TABLE, (sizeof(struct ofp_table) * nt));
- if (!attr){
- err = -ENOMEM;
- goto error_free_skb;
- }
-
- ot = nla_data(attr);
-
- for (i = 0; i < nt; ++i) {
- dp->chain->tables[i]->stats(dp->chain->tables[i], &stats);
- ot->header.version = OFP_VERSION;
- ot->header.type = OFPT_TABLE;
- ot->header.length = htons(sizeof(struct ofp_table));
- ot->header.xid = htonl(0);
-
- strncpy(ot->name, stats.name, OFP_MAX_TABLE_NAME_LEN);
- ot->table_id = htons(i);
- ot->n_flows = htonl(stats.n_flows);
- ot->max_flows = htonl(stats.max_flows);
- ot++;
- }
-
-
- genlmsg_end(skb, data);
- err = genlmsg_reply(skb, info); skb = 0;
-
-nla_put_failure:
-error_free_skb:
- if(skb)
- kfree_skb(skb);
- return err;
-}
-
-/*
- * Queries a datapath for flow-table statistics
- */
-
-
-static int dp_genl_table_query(struct sk_buff *skb, struct genl_info *info)
-{
- struct datapath* dp;
- int err = 0;
-
- if (!info->attrs[DP_GENL_A_DP_IDX]) {
- dprintk("dp::dp_genl_table_query received message with missing attributes\n");
- return -EINVAL;
- }
-
- rcu_read_lock();
- dp = dp_get(nla_get_u32(info->attrs[DP_GENL_A_DP_IDX]));
- if (!dp) {
- err = -ENOENT;
- goto err_out;
- }
-
- err = dp_dump_table_stats(dp, nla_get_u32(info->attrs[DP_GENL_A_DP_IDX]), info);
-
-err_out:
- rcu_read_unlock();
- return err;
-}
-
-/*
- * Queries a datapath for flow-table entries.
- */
-
-static int dp_genl_flow_query(struct sk_buff *skb, struct genl_info *info)
-{
- struct datapath* dp;
- struct ofp_flow_mod* ofm;
- u16 table_idx;
- int err = 0;
-
- if (!info->attrs[DP_GENL_A_DP_IDX]
- || !info->attrs[DP_GENL_A_TABLEIDX]
- || !info->attrs[DP_GENL_A_FLOW]) {
- dprintk("dp::dp_genl_flow_query received message with missing attributes\n");
- return -EINVAL;
- }
-
- rcu_read_lock();
- dp = dp_get(nla_get_u32(info->attrs[DP_GENL_A_DP_IDX]));
- if (!dp) {
- err = -ENOENT;
- goto err_out;
- }
-
- table_idx = nla_get_u16(info->attrs[DP_GENL_A_TABLEIDX]);
-
- if (dp->chain->n_tables <= table_idx){
- printk("table index %d invalid (dp has %d tables)\n",
- table_idx, dp->chain->n_tables);
- err = -EINVAL;
- goto err_out;
- }
-
- ofm = nla_data(info->attrs[DP_GENL_A_FLOW]);
- err = dp_dump_table(dp, table_idx, info, ofm);
-
-err_out:
- rcu_read_unlock();
- return err;
-}
-
-static struct nla_policy dp_genl_flow_policy[DP_GENL_A_MAX + 1] = {
- [DP_GENL_A_DP_IDX] = { .type = NLA_U32 },
- [DP_GENL_A_TABLEIDX] = { .type = NLA_U16 },
- [DP_GENL_A_NUMFLOWS] = { .type = NLA_U32 },
-};
-
-static struct genl_ops dp_genl_ops_query_flow = {
- .cmd = DP_GENL_C_QUERY_FLOW,
- .flags = GENL_ADMIN_PERM, /* Requires CAP_NET_ADMIN privilege. */
- .policy = dp_genl_flow_policy,
- .doit = dp_genl_flow_query,
- .dumpit = NULL,
-};
-
-static struct nla_policy dp_genl_table_policy[DP_GENL_A_MAX + 1] = {
- [DP_GENL_A_DP_IDX] = { .type = NLA_U32 },
-};
-
-static struct genl_ops dp_genl_ops_query_table = {
- .cmd = DP_GENL_C_QUERY_TABLE,
- .flags = GENL_ADMIN_PERM, /* Requires CAP_NET_ADMIN privilege. */
- .policy = dp_genl_table_policy,
- .doit = dp_genl_table_query,
- .dumpit = NULL,
-};
-
-
static struct genl_ops dp_genl_ops_query_dp = {
.cmd = DP_GENL_C_QUERY_DP,
.flags = GENL_ADMIN_PERM, /* Requires CAP_NET_ADMIN privilege. */
{
struct nlattr *va = info->attrs[DP_GENL_A_OPENFLOW];
struct datapath *dp;
+ struct ofp_header *oh;
+ struct sender sender;
int err;
if (!info->attrs[DP_GENL_A_DP_IDX] || !va)
goto out;
}
- va = info->attrs[DP_GENL_A_OPENFLOW];
+ if (nla_len(va) < sizeof(struct ofp_header)) {
+ err = -EINVAL;
+ goto out;
+ }
+ oh = nla_data(va);
- err = fwd_control_input(dp->chain, nla_data(va), nla_len(va));
+ sender.xid = oh->xid;
+ sender.pid = info->snd_pid;
+ sender.seq = info->snd_seq;
+ err = fwd_control_input(dp->chain, &sender, nla_data(va), nla_len(va));
out:
rcu_read_unlock();
* front. */
&dp_genl_ops_openflow,
- &dp_genl_ops_query_flow,
- &dp_genl_ops_query_table,
- &dp_genl_ops_show_dp,
&dp_genl_ops_add_dp,
&dp_genl_ops_del_dp,
&dp_genl_ops_query_dp,
#define DRV_NAME "openflow"
#define DRV_VERSION VERSION
#define DRV_DESCRIPTION "OpenFlow switching datapath implementation"
-#define DRV_COPYRIGHT "Copyright (c) 2007 The Board of Trustees of The Leland Stanford Junior University"
+#define DRV_COPYRIGHT "Copyright (c) 2007, 2008 The Board of Trustees of The Leland Stanford Junior University"
static int __init dp_init(void)
git://git.onelab.eu / sliver-openvswitch.git / blobdiff