/*
- * Copyright (c) 2007, 2008, 2009 Nicira Networks.
+ * Copyright (c) 2007, 2008, 2009, 2010 Nicira Networks.
* Distributed under the terms of the GNU GPL version 2.
*
* Significant portions of this file may be copied from parts of the Linux
#include <linux/module.h>
#include <linux/fs.h>
#include <linux/if_arp.h>
-#include <linux/if_bridge.h>
#include <linux/if_vlan.h>
#include <linux/in.h>
#include <linux/ip.h>
#include <linux/etherdevice.h>
#include <linux/kernel.h>
#include <linux/kthread.h>
-#include <linux/llc.h>
#include <linux/mutex.h>
#include <linux/percpu.h>
#include <linux/rcupdate.h>
#include <linux/rculist.h>
#include <linux/workqueue.h>
#include <linux/dmi.h>
-#include <net/llc.h>
+#include <net/inet_ecn.h>
#include "openvswitch/datapath-protocol.h"
#include "datapath.h"
#include "actions.h"
-#include "dp_dev.h"
#include "flow.h"
+#include "table.h"
+#include "vport-internal_dev.h"
#include "compat.h"
* dp_mutex nests inside the RTNL lock: if you need both you must take the RTNL
* lock first.
*
- * It is safe to access the datapath and net_bridge_port structures with just
+ * It is safe to access the datapath and dp_port structures with just
* dp_mutex.
*/
static struct datapath *dps[ODP_MAX];
/* Number of milliseconds between runs of the maintenance thread. */
#define MAINT_SLEEP_MSECS 1000
-static int new_nbp(struct datapath *, struct net_device *, int port_no);
+static int new_dp_port(struct datapath *, struct odp_port *, int port_no);
/* Must be called with rcu_read_lock or dp_mutex. */
struct datapath *get_dp(int dp_idx)
}
EXPORT_SYMBOL_GPL(get_dp);
-struct datapath *get_dp_locked(int dp_idx)
+static struct datapath *get_dp_locked(int dp_idx)
{
struct datapath *dp;
return dp;
}
+/* Must be called with rcu_read_lock or RTNL lock. */
+const char *dp_name(const struct datapath *dp)
+{
+ return vport_get_name(dp->ports[ODPP_LOCAL]->vport);
+}
+
static inline size_t br_nlmsg_size(void)
{
return NLMSG_ALIGN(sizeof(struct ifinfomsg))
}
static int dp_fill_ifinfo(struct sk_buff *skb,
- const struct net_bridge_port *port,
+ const struct dp_port *port,
int event, unsigned int flags)
{
const struct datapath *dp = port->dp;
- const struct net_device *dev = port->dev;
+ int ifindex = vport_get_ifindex(port->vport);
+ int iflink = vport_get_iflink(port->vport);
struct ifinfomsg *hdr;
struct nlmsghdr *nlh;
+ if (ifindex < 0)
+ return ifindex;
+
+ if (iflink < 0)
+ return iflink;
+
nlh = nlmsg_put(skb, 0, 0, event, sizeof(*hdr), flags);
if (nlh == NULL)
return -EMSGSIZE;
hdr = nlmsg_data(nlh);
hdr->ifi_family = AF_BRIDGE;
hdr->__ifi_pad = 0;
- hdr->ifi_type = dev->type;
- hdr->ifi_index = dev->ifindex;
- hdr->ifi_flags = dev_get_flags(dev);
+ hdr->ifi_type = ARPHRD_ETHER;
+ hdr->ifi_index = ifindex;
+ hdr->ifi_flags = vport_get_flags(port->vport);
hdr->ifi_change = 0;
- NLA_PUT_STRING(skb, IFLA_IFNAME, dev->name);
- NLA_PUT_U32(skb, IFLA_MASTER, dp->ports[ODPP_LOCAL]->dev->ifindex);
- NLA_PUT_U32(skb, IFLA_MTU, dev->mtu);
+ NLA_PUT_STRING(skb, IFLA_IFNAME, vport_get_name(port->vport));
+ NLA_PUT_U32(skb, IFLA_MASTER, vport_get_ifindex(dp->ports[ODPP_LOCAL]->vport));
+ NLA_PUT_U32(skb, IFLA_MTU, vport_get_mtu(port->vport));
#ifdef IFLA_OPERSTATE
NLA_PUT_U8(skb, IFLA_OPERSTATE,
- netif_running(dev) ? dev->operstate : IF_OPER_DOWN);
+ vport_is_running(port->vport)
+ ? vport_get_operstate(port->vport)
+ : IF_OPER_DOWN);
#endif
- if (dev->addr_len)
- NLA_PUT(skb, IFLA_ADDRESS, dev->addr_len, dev->dev_addr);
+ NLA_PUT(skb, IFLA_ADDRESS, ETH_ALEN,
+ vport_get_addr(port->vport));
- if (dev->ifindex != dev->iflink)
- NLA_PUT_U32(skb, IFLA_LINK, dev->iflink);
+ if (ifindex != iflink)
+ NLA_PUT_U32(skb, IFLA_LINK,iflink);
return nlmsg_end(skb, nlh);
return -EMSGSIZE;
}
-static void dp_ifinfo_notify(int event, struct net_bridge_port *port)
+static void dp_ifinfo_notify(int event, struct dp_port *port)
{
- struct net *net = dev_net(port->dev);
struct sk_buff *skb;
int err = -ENOBUFS;
kfree_skb(skb);
goto errout;
}
- rtnl_notify(skb, net, 0, RTNLGRP_LINK, NULL, GFP_KERNEL);
+ rtnl_notify(skb, &init_net, 0, RTNLGRP_LINK, NULL, GFP_KERNEL);
return;
errout:
if (err < 0)
- rtnl_set_sk_err(net, RTNLGRP_LINK, err);
+ rtnl_set_sk_err(&init_net, RTNLGRP_LINK, err);
}
static void release_dp(struct kobject *kobj)
kfree(dp);
}
-struct kobj_type dp_ktype = {
+static struct kobj_type dp_ktype = {
.release = release_dp
};
static int create_dp(int dp_idx, const char __user *devnamep)
{
- struct net_device *dp_dev;
+ struct odp_port internal_dev_port;
char devname[IFNAMSIZ];
struct datapath *dp;
int err;
int i;
if (devnamep) {
- err = -EFAULT;
- if (strncpy_from_user(devname, devnamep, IFNAMSIZ - 1) < 0)
+ int retval = strncpy_from_user(devname, devnamep, IFNAMSIZ);
+ if (retval < 0) {
+ err = -EFAULT;
+ goto err;
+ } else if (retval >= IFNAMSIZ) {
+ err = -ENAMETOOLONG;
goto err;
- devname[IFNAMSIZ - 1] = '\0';
+ }
} else {
snprintf(devname, sizeof devname, "of%d", dp_idx);
}
/* Allocate table. */
err = -ENOMEM;
- rcu_assign_pointer(dp->table, dp_table_create(DP_L1_SIZE));
+ rcu_assign_pointer(dp->table, tbl_create(0));
if (!dp->table)
goto err_free_dp;
/* Set up our datapath device. */
- dp_dev = dp_dev_create(dp, devname, ODPP_LOCAL);
- err = PTR_ERR(dp_dev);
- if (IS_ERR(dp_dev))
- goto err_destroy_table;
-
- err = new_nbp(dp, dp_dev, ODPP_LOCAL);
+ BUILD_BUG_ON(sizeof(internal_dev_port.devname) != sizeof(devname));
+ strcpy(internal_dev_port.devname, devname);
+ internal_dev_port.flags = ODP_PORT_INTERNAL;
+ err = new_dp_port(dp, &internal_dev_port, ODPP_LOCAL);
if (err) {
- dp_dev_destroy(dp_dev);
+ if (err == -EBUSY)
+ err = -EEXIST;
+
goto err_destroy_table;
}
return 0;
err_destroy_local_port:
- dp_del_port(dp->ports[ODPP_LOCAL]);
+ dp_detach_port(dp->ports[ODPP_LOCAL], 1);
err_destroy_table:
- dp_table_destroy(dp->table, 0);
+ tbl_destroy(dp->table, NULL);
err_free_dp:
kfree(dp);
err_put_module:
static void do_destroy_dp(struct datapath *dp)
{
- struct net_bridge_port *p, *n;
+ struct dp_port *p, *n;
int i;
list_for_each_entry_safe (p, n, &dp->port_list, node)
if (p->port_no != ODPP_LOCAL)
- dp_del_port(p);
+ dp_detach_port(p, 1);
dp_sysfs_del_dp(dp);
rcu_assign_pointer(dps[dp->dp_idx], NULL);
- dp_del_port(dp->ports[ODPP_LOCAL]);
+ dp_detach_port(dp->ports[ODPP_LOCAL], 1);
- dp_table_destroy(dp->table, 1);
+ tbl_destroy(dp->table, flow_free_tbl);
for (i = 0; i < DP_N_QUEUES; i++)
skb_queue_purge(&dp->queues[i]);
return err;
}
-static void release_nbp(struct kobject *kobj)
+static void release_dp_port(struct kobject *kobj)
{
- struct net_bridge_port *p = container_of(kobj, struct net_bridge_port, kobj);
+ struct dp_port *p = container_of(kobj, struct dp_port, kobj);
kfree(p);
}
-struct kobj_type brport_ktype = {
+static struct kobj_type brport_ktype = {
#ifdef CONFIG_SYSFS
.sysfs_ops = &brport_sysfs_ops,
#endif
- .release = release_nbp
+ .release = release_dp_port
};
/* Called with RTNL lock and dp_mutex. */
-static int new_nbp(struct datapath *dp, struct net_device *dev, int port_no)
+static int new_dp_port(struct datapath *dp, struct odp_port *odp_port, int port_no)
{
- struct net_bridge_port *p;
+ struct vport *vport;
+ struct dp_port *p;
+ int err;
+
+ vport = vport_locate(odp_port->devname);
+ if (!vport) {
+ vport_lock();
+
+ if (odp_port->flags & ODP_PORT_INTERNAL)
+ vport = __vport_add(odp_port->devname, "internal", NULL);
+ else
+ vport = __vport_add(odp_port->devname, "netdev", NULL);
+
+ vport_unlock();
- if (dev->br_port != NULL)
- return -EBUSY;
+ if (IS_ERR(vport))
+ return PTR_ERR(vport);
+ }
p = kzalloc(sizeof(*p), GFP_KERNEL);
if (!p)
return -ENOMEM;
- dev_set_promiscuity(dev, 1);
- dev_hold(dev);
p->port_no = port_no;
p->dp = dp;
- p->dev = dev;
- if (!is_dp_dev(dev))
- rcu_assign_pointer(dev->br_port, p);
- else {
- /* It would make sense to assign dev->br_port here too, but
- * that causes packets received on internal ports to get caught
- * in dp_frame_hook(). In turn dp_frame_hook() can reject them
- * back to network stack, but that's a waste of time. */
+ atomic_set(&p->sflow_pool, 0);
+
+ err = vport_attach(vport, p);
+ if (err) {
+ kfree(p);
+ return err;
}
+
rcu_assign_pointer(dp->ports[port_no], p);
list_add_rcu(&p->node, &dp->port_list);
dp->n_ports++;
return 0;
}
-static int add_port(int dp_idx, struct odp_port __user *portp)
+static int attach_port(int dp_idx, struct odp_port __user *portp)
{
- struct net_device *dev;
struct datapath *dp;
struct odp_port port;
int port_no;
goto out_unlock_dp;
got_port_no:
- if (!(port.flags & ODP_PORT_INTERNAL)) {
- err = -ENODEV;
- dev = dev_get_by_name(&init_net, port.devname);
- if (!dev)
- goto out_unlock_dp;
-
- err = -EINVAL;
- if (dev->flags & IFF_LOOPBACK || dev->type != ARPHRD_ETHER ||
- is_dp_dev(dev))
- goto out_put;
- } else {
- dev = dp_dev_create(dp, port.devname, port_no);
- err = PTR_ERR(dev);
- if (IS_ERR(dev))
- goto out_unlock_dp;
- dev_hold(dev);
- }
-
- err = new_nbp(dp, dev, port_no);
+ err = new_dp_port(dp, &port, port_no);
if (err)
- goto out_put;
+ goto out_unlock_dp;
+ set_internal_devs_mtu(dp);
dp_sysfs_add_if(dp->ports[port_no]);
- err = __put_user(port_no, &port.port);
+ err = __put_user(port_no, &portp->port);
-out_put:
- dev_put(dev);
out_unlock_dp:
mutex_unlock(&dp->mutex);
out_unlock_rtnl:
return err;
}
-int dp_del_port(struct net_bridge_port *p)
+int dp_detach_port(struct dp_port *p, int may_delete)
{
+ struct vport *vport = p->vport;
+ int err;
+
ASSERT_RTNL();
if (p->port_no != ODPP_LOCAL)
dp_sysfs_del_if(p);
dp_ifinfo_notify(RTM_DELLINK, p);
- p->dp->n_ports--;
-
- if (is_dp_dev(p->dev)) {
- /* Make sure that no packets arrive from now on, since
- * dp_dev_xmit() will try to find itself through
- * p->dp->ports[], and we're about to set that to null. */
- netif_tx_disable(p->dev);
- }
-
/* First drop references to device. */
- dev_set_promiscuity(p->dev, -1);
+ p->dp->n_ports--;
list_del_rcu(&p->node);
rcu_assign_pointer(p->dp->ports[p->port_no], NULL);
- rcu_assign_pointer(p->dev->br_port, NULL);
+
+ err = vport_detach(vport);
+ if (err)
+ return err;
/* Then wait until no one is still using it, and destroy it. */
synchronize_rcu();
- if (is_dp_dev(p->dev))
- dp_dev_destroy(p->dev);
- dev_put(p->dev);
+ if (may_delete) {
+ const char *port_type = vport_get_type(vport);
+
+ if (!strcmp(port_type, "netdev") || !strcmp(port_type, "internal")) {
+ vport_lock();
+ __vport_del(vport);
+ vport_unlock();
+ }
+ }
+
kobject_put(&p->kobj);
return 0;
}
-static int del_port(int dp_idx, int port_no)
+static int detach_port(int dp_idx, int port_no)
{
- struct net_bridge_port *p;
+ struct dp_port *p;
struct datapath *dp;
- LIST_HEAD(dp_devs);
int err;
err = -EINVAL;
if (!p)
goto out_unlock_dp;
- err = dp_del_port(p);
+ err = dp_detach_port(p, 1);
out_unlock_dp:
mutex_unlock(&dp->mutex);
return err;
}
-/* Must be called with rcu_read_lock. */
-static void
-do_port_input(struct net_bridge_port *p, struct sk_buff *skb)
-{
- /* Make our own copy of the packet. Otherwise we will mangle the
- * packet for anyone who came before us (e.g. tcpdump via AF_PACKET).
- * (No one comes after us, since we tell handle_bridge() that we took
- * the packet.) */
- skb = skb_share_check(skb, GFP_ATOMIC);
- if (!skb)
- return;
-
- /* Push the Ethernet header back on. */
- skb_push(skb, ETH_HLEN);
- skb_reset_mac_header(skb);
- dp_process_received_packet(skb, p);
-}
-
/* Must be called with rcu_read_lock and with bottom-halves disabled. */
-void dp_process_received_packet(struct sk_buff *skb, struct net_bridge_port *p)
+void dp_process_received_packet(struct dp_port *p, struct sk_buff *skb)
{
struct datapath *dp = p->dp;
struct dp_stats_percpu *stats;
struct odp_flow_key key;
- struct sw_flow *flow;
+ struct tbl_node *flow_node;
WARN_ON_ONCE(skb_shared(skb));
+ skb_warn_if_lro(skb);
+
+ OVS_CB(skb)->dp_port = p;
/* BHs are off so we don't have to use get_cpu()/put_cpu() here. */
stats = percpu_ptr(dp->stats_percpu, smp_processor_id());
}
}
- flow = dp_table_lookup(rcu_dereference(dp->table), &key);
- if (flow) {
+ flow_node = tbl_lookup(rcu_dereference(dp->table), &key, flow_hash(&key), flow_cmp);
+ if (flow_node) {
+ struct sw_flow *flow = flow_cast(flow_node);
struct sw_flow_actions *acts = rcu_dereference(flow->sf_acts);
flow_used(flow, skb);
execute_actions(dp, skb, &key, acts->actions, acts->n_actions,
stats->n_hit++;
} else {
stats->n_missed++;
- dp_output_control(dp, skb, _ODPL_MISS_NR, 0);
+ dp_output_control(dp, skb, _ODPL_MISS_NR, OVS_CB(skb)->tun_id);
}
}
-/*
- * Used as br_handle_frame_hook. (Cannot run bridge at the same time, even on
- * different set of devices!)
- */
-#if LINUX_VERSION_CODE >= KERNEL_VERSION(2,6,22)
-/* Called with rcu_read_lock and bottom-halves disabled. */
-static struct sk_buff *dp_frame_hook(struct net_bridge_port *p,
- struct sk_buff *skb)
+#if defined(CONFIG_XEN) && defined(HAVE_PROTO_DATA_VALID)
+/* This code is based on skb_checksum_setup() from Xen's net/dev/core.c. We
+ * can't call this function directly because it isn't exported in all
+ * versions. */
+int vswitch_skb_checksum_setup(struct sk_buff *skb)
{
- do_port_input(p, skb);
- return NULL;
+ struct iphdr *iph;
+ unsigned char *th;
+ int err = -EPROTO;
+ __u16 csum_start, csum_offset;
+
+ if (!skb->proto_csum_blank)
+ return 0;
+
+ if (skb->protocol != htons(ETH_P_IP))
+ goto out;
+
+ if (!pskb_may_pull(skb, skb_network_header(skb) + sizeof(struct iphdr) - skb->data))
+ goto out;
+
+ iph = ip_hdr(skb);
+ th = skb_network_header(skb) + 4 * iph->ihl;
+
+ csum_start = th - skb->head;
+ switch (iph->protocol) {
+ case IPPROTO_TCP:
+ csum_offset = offsetof(struct tcphdr, check);
+ break;
+ case IPPROTO_UDP:
+ csum_offset = offsetof(struct udphdr, check);
+ break;
+ default:
+ if (net_ratelimit())
+ printk(KERN_ERR "Attempting to checksum a non-"
+ "TCP/UDP packet, dropping a protocol"
+ " %d packet", iph->protocol);
+ goto out;
+ }
+
+ if (!pskb_may_pull(skb, th + csum_offset + 2 - skb->data))
+ goto out;
+
+ skb->ip_summed = CHECKSUM_PARTIAL;
+ skb->proto_csum_blank = 0;
+
+#if LINUX_VERSION_CODE >= KERNEL_VERSION(2,6,22)
+ skb->csum_start = csum_start;
+ skb->csum_offset = csum_offset;
+#else
+ skb_set_transport_header(skb, csum_start - skb_headroom(skb));
+ skb->csum = csum_offset;
+#endif
+
+ err = 0;
+
+out:
+ return err;
}
-#elif LINUX_VERSION_CODE >= KERNEL_VERSION(2,6,0)
-/* Called with rcu_read_lock and bottom-halves disabled. */
-static int dp_frame_hook(struct net_bridge_port *p, struct sk_buff **pskb)
+#endif /* CONFIG_XEN && HAVE_PROTO_DATA_VALID */
+
+ /* Types of checksums that we can receive (these all refer to L4 checksums):
+ * 1. CHECKSUM_NONE: Device that did not compute checksum, contains full
+ * (though not verified) checksum in packet but not in skb->csum. Packets
+ * from the bridge local port will also have this type.
+ * 2. CHECKSUM_COMPLETE (CHECKSUM_HW): Good device that computes checksums,
+ * also the GRE module. This is the same as CHECKSUM_NONE, except it has
+ * a valid skb->csum. Importantly, both contain a full checksum (not
+ * verified) in the packet itself. The only difference is that if the
+ * packet gets to L4 processing on this machine (not in DomU) we won't
+ * have to recompute the checksum to verify. Most hardware devices do not
+ * produce packets with this type, even if they support receive checksum
+ * offloading (they produce type #5).
+ * 3. CHECKSUM_PARTIAL (CHECKSUM_HW): Packet without full checksum and needs to
+ * be computed if it is sent off box. Unfortunately on earlier kernels,
+ * this case is impossible to distinguish from #2, despite having opposite
+ * meanings. Xen adds an extra field on earlier kernels (see #4) in order
+ * to distinguish the different states.
+ * 4. CHECKSUM_UNNECESSARY (with proto_csum_blank true): This packet was
+ * generated locally by a Xen DomU and has a partial checksum. If it is
+ * handled on this machine (Dom0 or DomU), then the checksum will not be
+ * computed. If it goes off box, the checksum in the packet needs to be
+ * completed. Calling skb_checksum_setup converts this to CHECKSUM_HW
+ * (CHECKSUM_PARTIAL) so that the checksum can be completed. In later
+ * kernels, this combination is replaced with CHECKSUM_PARTIAL.
+ * 5. CHECKSUM_UNNECESSARY (with proto_csum_blank false): Packet with a correct
+ * full checksum or using a protocol without a checksum. skb->csum is
+ * undefined. This is common from devices with receive checksum
+ * offloading. This is somewhat similar to CHECKSUM_NONE, except that
+ * nobody will try to verify the checksum with CHECKSUM_UNNECESSARY.
+ *
+ * Note that on earlier kernels, CHECKSUM_COMPLETE and CHECKSUM_PARTIAL are
+ * both defined as CHECKSUM_HW. Normally the meaning of CHECKSUM_HW is clear
+ * based on whether it is on the transmit or receive path. After the datapath
+ * it will be intepreted as CHECKSUM_PARTIAL. If the packet already has a
+ * checksum, we will panic. Since we can receive packets with checksums, we
+ * assume that all CHECKSUM_HW packets have checksums and map them to
+ * CHECKSUM_NONE, which has a similar meaning (the it is only different if the
+ * packet is processed by the local IP stack, in which case it will need to
+ * be reverified). If we receive a packet with CHECKSUM_HW that really means
+ * CHECKSUM_PARTIAL, it will be sent with the wrong checksum. However, there
+ * shouldn't be any devices that do this with bridging. */
+void
+compute_ip_summed(struct sk_buff *skb, bool xmit)
{
- do_port_input(p, *pskb);
- return 1;
-}
+ /* For our convenience these defines change repeatedly between kernel
+ * versions, so we can't just copy them over... */
+ switch (skb->ip_summed) {
+ case CHECKSUM_NONE:
+ OVS_CB(skb)->ip_summed = OVS_CSUM_NONE;
+ break;
+ case CHECKSUM_UNNECESSARY:
+ OVS_CB(skb)->ip_summed = OVS_CSUM_UNNECESSARY;
+ break;
+#ifdef CHECKSUM_HW
+ /* In theory this could be either CHECKSUM_PARTIAL or CHECKSUM_COMPLETE.
+ * However, on the receive side we should only get CHECKSUM_PARTIAL
+ * packets from Xen, which uses some special fields to represent this
+ * (see below). Since we can only make one type work, pick the one
+ * that actually happens in practice.
+ *
+ * On the transmit side (basically after skb_checksum_setup()
+ * has been run or on internal dev transmit), packets with
+ * CHECKSUM_COMPLETE aren't generated, so assume CHECKSUM_PARTIAL. */
+ case CHECKSUM_HW:
+ if (!xmit)
+ OVS_CB(skb)->ip_summed = OVS_CSUM_COMPLETE;
+ else
+ OVS_CB(skb)->ip_summed = OVS_CSUM_PARTIAL;
+
+ break;
#else
-#error
+ case CHECKSUM_COMPLETE:
+ OVS_CB(skb)->ip_summed = OVS_CSUM_COMPLETE;
+ break;
+ case CHECKSUM_PARTIAL:
+ OVS_CB(skb)->ip_summed = OVS_CSUM_PARTIAL;
+ break;
#endif
+ default:
+ printk(KERN_ERR "openvswitch: unknown checksum type %d\n",
+ skb->ip_summed);
+ /* None seems the safest... */
+ OVS_CB(skb)->ip_summed = OVS_CSUM_NONE;
+ }
#if defined(CONFIG_XEN) && defined(HAVE_PROTO_DATA_VALID)
-/* This code is copied verbatim from net/dev/core.c in Xen's
- * linux-2.6.18-92.1.10.el5.xs5.0.0.394.644. We can't call those functions
- * directly because they aren't exported. */
-static int skb_pull_up_to(struct sk_buff *skb, void *ptr)
-{
- if (ptr < (void *)skb->tail)
- return 1;
- if (__pskb_pull_tail(skb,
- ptr - (void *)skb->data - skb_headlen(skb))) {
- return 1;
- } else {
- return 0;
- }
+ /* Xen has a special way of representing CHECKSUM_PARTIAL on older
+ * kernels. It should not be set on the transmit path though. */
+ if (skb->proto_csum_blank)
+ OVS_CB(skb)->ip_summed = OVS_CSUM_PARTIAL;
+
+ WARN_ON_ONCE(skb->proto_csum_blank && xmit);
+#endif
}
-int vswitch_skb_checksum_setup(struct sk_buff *skb)
+/* This function closely resembles skb_forward_csum() used by the bridge. It
+ * is slightly different because we are only concerned with bridging and not
+ * other types of forwarding and can get away with slightly more optimal
+ * behavior.*/
+void
+forward_ip_summed(struct sk_buff *skb)
{
- if (skb->proto_csum_blank) {
- if (skb->protocol != htons(ETH_P_IP))
- goto out;
- if (!skb_pull_up_to(skb, skb->nh.iph + 1))
- goto out;
- skb->h.raw = (unsigned char *)skb->nh.iph + 4*skb->nh.iph->ihl;
- switch (skb->nh.iph->protocol) {
- case IPPROTO_TCP:
- skb->csum = offsetof(struct tcphdr, check);
- break;
- case IPPROTO_UDP:
- skb->csum = offsetof(struct udphdr, check);
- break;
- default:
- if (net_ratelimit())
- printk(KERN_ERR "Attempting to checksum a non-"
- "TCP/UDP packet, dropping a protocol"
- " %d packet", skb->nh.iph->protocol);
- goto out;
- }
- if (!skb_pull_up_to(skb, skb->h.raw + skb->csum + 2))
- goto out;
- skb->ip_summed = CHECKSUM_HW;
- skb->proto_csum_blank = 0;
- }
- return 0;
-out:
- return -EPROTO;
+#ifdef CHECKSUM_HW
+ if (OVS_CB(skb)->ip_summed == OVS_CSUM_COMPLETE)
+ skb->ip_summed = CHECKSUM_NONE;
+#endif
}
-#else
-int vswitch_skb_checksum_setup(struct sk_buff *skb) { return 0; }
-#endif /* CONFIG_XEN && linux == 2.6.18 */
/* Append each packet in 'skb' list to 'queue'. There will be only one packet
* unless we broke up a GSO packet. */
int port_no;
int err;
- port_no = ODPP_LOCAL;
- if (skb->dev) {
- if (skb->dev->br_port)
- port_no = skb->dev->br_port->port_no;
- else if (is_dp_dev(skb->dev))
- port_no = dp_dev_priv(skb->dev)->port_no;
- }
+ if (OVS_CB(skb)->dp_port)
+ port_no = OVS_CB(skb)->dp_port->port_no;
+ else
+ port_no = ODPP_LOCAL;
do {
struct odp_msg *header;
skb->next = NULL;
/* If a checksum-deferred packet is forwarded to the
- * controller, correct the pointers and checksum. This happens
- * on a regular basis only on Xen, on which VMs can pass up
- * packets that do not have their checksum computed.
+ * controller, correct the pointers and checksum.
*/
err = vswitch_skb_checksum_setup(skb);
if (err)
goto err_kfree_skbs;
-#ifndef CHECKSUM_HW
+
if (skb->ip_summed == CHECKSUM_PARTIAL) {
+
#if LINUX_VERSION_CODE >= KERNEL_VERSION(2,6,22)
/* Until 2.6.22, the start of the transport header was
* also the start of data to be checksummed. Linux
skb_set_transport_header(skb, skb->csum_start -
skb_headroom(skb));
#endif
+
err = skb_checksum_help(skb);
if (err)
goto err_kfree_skbs;
}
-#else
- if (skb->ip_summed == CHECKSUM_HW) {
- err = skb_checksum_help(skb, 0);
- if (err)
- goto err_kfree_skbs;
- }
-#endif
err = skb_cow(skb, sizeof *header);
if (err)
if (skb_queue_len(queue) >= DP_MAX_QUEUE_LEN)
goto err_kfree_skb;
+ forward_ip_summed(skb);
+
/* Break apart GSO packets into their component pieces. Otherwise
* userspace may try to stuff a 64kB packet into a 1500-byte MTU. */
if (skb_is_gso(skb)) {
static int flush_flows(struct datapath *dp)
{
- dp->n_flows = 0;
- return dp_table_flush(dp);
+ struct tbl *old_table = rcu_dereference(dp->table);
+ struct tbl *new_table;
+
+ new_table = tbl_create(0);
+ if (!new_table)
+ return -ENOMEM;
+
+ rcu_assign_pointer(dp->table, new_table);
+
+ tbl_deferred_destroy(old_table, flow_free_tbl);
+
+ return 0;
}
static int validate_actions(const struct sw_flow_actions *actions)
return -EINVAL;
break;
+ case ODPAT_SET_NW_TOS:
+ if (a->nw_tos.nw_tos & INET_ECN_MASK)
+ return -EINVAL;
+ break;
+
default:
if (a->type >= ODPAT_N_ACTIONS)
return -EOPNOTSUPP;
flow->byte_count = 0;
}
+static int expand_table(struct datapath *dp)
+{
+ struct tbl *old_table = rcu_dereference(dp->table);
+ struct tbl *new_table;
+
+ new_table = tbl_expand(old_table);
+ if (IS_ERR(new_table))
+ return PTR_ERR(new_table);
+
+ rcu_assign_pointer(dp->table, new_table);
+ tbl_deferred_destroy(old_table, NULL);
+
+ return 0;
+}
+
static int put_flow(struct datapath *dp, struct odp_flow_put __user *ufp)
{
struct odp_flow_put uf;
+ struct tbl_node *flow_node;
struct sw_flow *flow;
- struct dp_table *table;
+ struct tbl *table;
struct odp_flow_stats stats;
int error;
error = -EFAULT;
if (copy_from_user(&uf, ufp, sizeof(struct odp_flow_put)))
goto error;
- uf.flow.key.reserved = 0;
+ memset(uf.flow.key.reserved, 0, sizeof uf.flow.key.reserved);
table = rcu_dereference(dp->table);
- flow = dp_table_lookup(table, &uf.flow.key);
- if (!flow) {
+ flow_node = tbl_lookup(table, &uf.flow.key, flow_hash(&uf.flow.key), flow_cmp);
+ if (!flow_node) {
/* No such flow. */
struct sw_flow_actions *acts;
goto error;
/* Expand table, if necessary, to make room. */
- if (dp->n_flows >= table->n_buckets) {
- error = -ENOSPC;
- if (table->n_buckets >= DP_MAX_BUCKETS)
- goto error;
-
- error = dp_table_expand(dp);
+ if (tbl_count(table) >= tbl_n_buckets(table)) {
+ error = expand_table(dp);
if (error)
goto error;
table = rcu_dereference(dp->table);
rcu_assign_pointer(flow->sf_acts, acts);
/* Put flow in bucket. */
- error = dp_table_insert(table, flow);
+ error = tbl_insert(table, &flow->tbl_node, flow_hash(&flow->key));
if (error)
goto error_free_flow_acts;
- dp->n_flows++;
+
memset(&stats, 0, sizeof(struct odp_flow_stats));
} else {
/* We found a matching flow. */
struct sw_flow_actions *old_acts, *new_acts;
unsigned long int flags;
+ flow = flow_cast(flow_node);
+
/* Bail out if we're not allowed to modify an existing flow. */
error = -EEXIST;
if (!(uf.flags & ODPPF_MODIFY))
static int del_flow(struct datapath *dp, struct odp_flow __user *ufp)
{
- struct dp_table *table = rcu_dereference(dp->table);
+ struct tbl *table = rcu_dereference(dp->table);
struct odp_flow uf;
+ struct tbl_node *flow_node;
struct sw_flow *flow;
int error;
error = -EFAULT;
if (copy_from_user(&uf, ufp, sizeof uf))
goto error;
- uf.key.reserved = 0;
+ memset(uf.key.reserved, 0, sizeof uf.key.reserved);
- flow = dp_table_lookup(table, &uf.key);
+ flow_node = tbl_lookup(table, &uf.key, flow_hash(&uf.key), flow_cmp);
error = -ENOENT;
- if (!flow)
+ if (!flow_node)
goto error;
- /* XXX redundant lookup */
- error = dp_table_delete(table, flow);
+ error = tbl_remove(table, flow_node);
if (error)
goto error;
* be using this flow. We used to synchronize_rcu() to make sure that
* we get completely accurate stats, but that blows our performance,
* badly. */
- dp->n_flows--;
+
+ flow = flow_cast(flow_node);
error = answer_query(flow, 0, ufp);
flow_deferred_free(flow);
static int query_flows(struct datapath *dp, const struct odp_flowvec *flowvec)
{
- struct dp_table *table = rcu_dereference(dp->table);
+ struct tbl *table = rcu_dereference(dp->table);
int i;
for (i = 0; i < flowvec->n_flows; i++) {
struct __user odp_flow *ufp = &flowvec->flows[i];
struct odp_flow uf;
- struct sw_flow *flow;
+ struct tbl_node *flow_node;
int error;
if (__copy_from_user(&uf, ufp, sizeof uf))
return -EFAULT;
- uf.key.reserved = 0;
+ memset(uf.key.reserved, 0, sizeof uf.key.reserved);
- flow = dp_table_lookup(table, &uf.key);
- if (!flow)
+ flow_node = tbl_lookup(table, &uf.key, flow_hash(&uf.key), flow_cmp);
+ if (!flow_node)
error = __put_user(ENOENT, &ufp->stats.error);
else
- error = answer_query(flow, uf.flags, ufp);
+ error = answer_query(flow_cast(flow_node), uf.flags, ufp);
if (error)
return -EFAULT;
}
int listed_flows;
};
-static int list_flow(struct sw_flow *flow, void *cbdata_)
+static int list_flow(struct tbl_node *node, void *cbdata_)
{
+ struct sw_flow *flow = flow_cast(node);
struct list_flows_cbdata *cbdata = cbdata_;
struct odp_flow __user *ufp = &cbdata->uflows[cbdata->listed_flows++];
int error;
cbdata.uflows = flowvec->flows;
cbdata.n_flows = flowvec->n_flows;
cbdata.listed_flows = 0;
- error = dp_table_foreach(rcu_dereference(dp->table),
- list_flow, &cbdata);
+ error = tbl_foreach(rcu_dereference(dp->table), list_flow, &cbdata);
return error ? error : cbdata.listed_flows;
}
skb = alloc_skb(execute.length, GFP_KERNEL);
if (!skb)
goto error_free_actions;
- if (execute.in_port < DP_MAX_PORTS) {
- struct net_bridge_port *p = dp->ports[execute.in_port];
- if (p)
- skb->dev = p->dev;
- }
+
+ if (execute.in_port < DP_MAX_PORTS)
+ OVS_CB(skb)->dp_port = dp->ports[execute.in_port];
+ else
+ OVS_CB(skb)->dp_port = NULL;
err = -EFAULT;
if (copy_from_user(skb_put(skb, execute.length), execute.data,
static int get_dp_stats(struct datapath *dp, struct odp_stats __user *statsp)
{
+ struct tbl *table = rcu_dereference(dp->table);
struct odp_stats stats;
int i;
- stats.n_flows = dp->n_flows;
- stats.cur_capacity = rcu_dereference(dp->table)->n_buckets;
- stats.max_capacity = DP_MAX_BUCKETS;
+ stats.n_flows = tbl_count(table);
+ stats.cur_capacity = tbl_n_buckets(table);
+ stats.max_capacity = TBL_MAX_BUCKETS;
stats.n_ports = dp->n_ports;
stats.max_ports = DP_MAX_PORTS;
stats.max_groups = DP_MAX_GROUPS;
/* MTU of the dp pseudo-device: ETH_DATA_LEN or the minimum of the ports */
int dp_min_mtu(const struct datapath *dp)
{
- struct net_bridge_port *p;
+ struct dp_port *p;
int mtu = 0;
ASSERT_RTNL();
list_for_each_entry_rcu (p, &dp->port_list, node) {
- struct net_device *dev = p->dev;
+ int dev_mtu;
/* Skip any internal ports, since that's what we're trying to
* set. */
- if (is_dp_dev(dev))
+ if (is_internal_vport(p->vport))
continue;
- if (!mtu || dev->mtu < mtu)
- mtu = dev->mtu;
+ dev_mtu = vport_get_mtu(p->vport);
+ if (!mtu || dev_mtu < mtu)
+ mtu = dev_mtu;
}
return mtu ? mtu : ETH_DATA_LEN;
}
+/* Sets the MTU of all datapath devices to the minimum of the ports. Must
+ * be called with RTNL lock. */
+void set_internal_devs_mtu(const struct datapath *dp)
+{
+ struct dp_port *p;
+ int mtu;
+
+ ASSERT_RTNL();
+
+ mtu = dp_min_mtu(dp);
+
+ list_for_each_entry_rcu (p, &dp->port_list, node) {
+ if (is_internal_vport(p->vport))
+ vport_set_mtu(p->vport, mtu);
+ }
+}
+
static int
-put_port(const struct net_bridge_port *p, struct odp_port __user *uop)
+put_port(const struct dp_port *p, struct odp_port __user *uop)
{
struct odp_port op;
+
memset(&op, 0, sizeof op);
- strncpy(op.devname, p->dev->name, sizeof op.devname);
+
+ rcu_read_lock();
+ strncpy(op.devname, vport_get_name(p->vport), sizeof op.devname);
+ rcu_read_unlock();
+
op.port = p->port_no;
- op.flags = is_dp_dev(p->dev) ? ODP_PORT_INTERNAL : 0;
+ op.flags = is_internal_vport(p->vport) ? ODP_PORT_INTERNAL : 0;
+
return copy_to_user(uop, &op, sizeof op) ? -EFAULT : 0;
}
if (copy_from_user(&port, uport, sizeof port))
return -EFAULT;
+
if (port.devname[0]) {
- struct net_bridge_port *p;
- struct net_device *dev;
- int err;
+ struct vport *vport;
+ struct dp_port *dp_port;
+ int err = 0;
port.devname[IFNAMSIZ - 1] = '\0';
- dev = dev_get_by_name(&init_net, port.devname);
- if (!dev)
- return -ENODEV;
+ vport_lock();
+ rcu_read_lock();
- p = dev->br_port;
- if (!p && is_dp_dev(dev)) {
- struct dp_dev *dp_dev = dp_dev_priv(dev);
- if (dp_dev->dp == dp)
- p = dp->ports[dp_dev->port_no];
+ vport = vport_locate(port.devname);
+ if (!vport) {
+ err = -ENODEV;
+ goto error_unlock;
}
- err = p && p->dp == dp ? put_port(p, uport) : -ENOENT;
- dev_put(dev);
- return err;
+ dp_port = vport_get_dp_port(vport);
+ if (!dp_port || dp_port->dp != dp) {
+ err = -ENOENT;
+ goto error_unlock;
+ }
+
+ port.port = dp_port->port_no;
+
+error_unlock:
+ rcu_read_unlock();
+ vport_unlock();
+
+ if (err)
+ return err;
} else {
if (port.port >= DP_MAX_PORTS)
return -EINVAL;
if (!dp->ports[port.port])
return -ENOENT;
- return put_port(dp->ports[port.port], uport);
}
+
+ return put_port(dp->ports[port.port], uport);
}
static int
list_ports(struct datapath *dp, struct odp_portvec __user *pvp)
{
struct odp_portvec pv;
- struct net_bridge_port *p;
+ struct dp_port *p;
int idx;
if (copy_from_user(&pv, pvp, sizeof pv))
err = destroy_dp(dp_idx);
goto exit;
- case ODP_PORT_ADD:
- err = add_port(dp_idx, (struct odp_port __user *)argp);
+ case ODP_PORT_ATTACH:
+ err = attach_port(dp_idx, (struct odp_port __user *)argp);
goto exit;
- case ODP_PORT_DEL:
+ case ODP_PORT_DETACH:
err = get_user(port_no, (int __user *)argp);
if (!err)
- err = del_port(dp_idx, port_no);
+ err = detach_port(dp_idx, port_no);
+ goto exit;
+
+ case ODP_VPORT_ADD:
+ err = vport_add((struct odp_vport_add __user *)argp);
+ goto exit;
+
+ case ODP_VPORT_MOD:
+ err = vport_mod((struct odp_vport_mod __user *)argp);
+ goto exit;
+
+ case ODP_VPORT_DEL:
+ err = vport_del((char __user *)argp);
+ goto exit;
+
+ case ODP_VPORT_STATS_GET:
+ err = vport_stats_get((struct odp_vport_stats_req __user *)argp);
+ goto exit;
+
+ case ODP_VPORT_ETHER_GET:
+ err = vport_ether_get((struct odp_vport_ether __user *)argp);
+ goto exit;
+
+ case ODP_VPORT_ETHER_SET:
+ err = vport_ether_set((struct odp_vport_ether __user *)argp);
+ goto exit;
+
+ case ODP_VPORT_MTU_GET:
+ err = vport_mtu_get((struct odp_vport_mtu __user *)argp);
+ goto exit;
+
+ case ODP_VPORT_MTU_SET:
+ err = vport_mtu_set((struct odp_vport_mtu __user *)argp);
goto exit;
}
static int major;
-#if LINUX_VERSION_CODE < KERNEL_VERSION(2,6,27)
-static struct llc_sap *dp_stp_sap;
-
-static int dp_stp_rcv(struct sk_buff *skb, struct net_device *dev,
- struct packet_type *pt, struct net_device *orig_dev)
-{
- /* We don't really care about STP packets, we just listen for them for
- * mutual exclusion with the bridge module, so this just discards
- * them. */
- kfree_skb(skb);
- return 0;
-}
-
-static int dp_avoid_bridge_init(void)
-{
- /* Register to receive STP packets because the bridge module also
- * attempts to do so. Since there can only be a single listener for a
- * given protocol, this provides mutual exclusion against the bridge
- * module, preventing both of them from being loaded at the same
- * time. */
- dp_stp_sap = llc_sap_open(LLC_SAP_BSPAN, dp_stp_rcv);
- if (!dp_stp_sap) {
- printk(KERN_ERR "openvswitch: can't register sap for STP (probably the bridge module is loaded)\n");
- return -EADDRINUSE;
- }
- return 0;
-}
-
-static void dp_avoid_bridge_exit(void)
-{
- llc_sap_put(dp_stp_sap);
-}
-#else /* Linux 2.6.27 or later. */
-static int dp_avoid_bridge_init(void)
-{
- /* Linux 2.6.27 introduces a way for multiple clients to register for
- * STP packets, which interferes with what we try to do above.
- * Instead, just check whether there's a bridge hook defined. This is
- * not as safe--the bridge module is willing to load over the top of
- * us--but it provides a little bit of protection. */
- if (br_handle_frame_hook) {
- printk(KERN_ERR "openvswitch: bridge module is loaded, cannot load over it\n");
- return -EADDRINUSE;
- }
- return 0;
-}
-
-static void dp_avoid_bridge_exit(void)
-{
- /* Nothing to do. */
-}
-#endif /* Linux 2.6.27 or later */
-
static int __init dp_init(void)
{
+ struct sk_buff *dummy_skb;
int err;
- printk("Open vSwitch %s, built "__DATE__" "__TIME__"\n", VERSION BUILDNR);
+ BUILD_BUG_ON(sizeof(struct ovs_skb_cb) > sizeof(dummy_skb->cb));
- err = dp_avoid_bridge_init();
- if (err)
- return err;
+ printk("Open vSwitch %s, built "__DATE__" "__TIME__"\n", VERSION BUILDNR);
err = flow_init();
if (err)
goto error;
- err = register_netdevice_notifier(&dp_device_notifier);
+ err = vport_init();
if (err)
goto error_flow_exit;
+ err = register_netdevice_notifier(&dp_device_notifier);
+ if (err)
+ goto error_vport_exit;
+
major = register_chrdev(0, "openvswitch", &openvswitch_fops);
if (err < 0)
goto error_unreg_notifier;
- /* Hook into callback used by the bridge to intercept packets.
- * Parasites we are. */
- br_handle_frame_hook = dp_frame_hook;
-
return 0;
error_unreg_notifier:
unregister_netdevice_notifier(&dp_device_notifier);
+error_vport_exit:
+ vport_exit();
error_flow_exit:
flow_exit();
error:
rcu_barrier();
unregister_chrdev(major, "openvswitch");
unregister_netdevice_notifier(&dp_device_notifier);
+ vport_exit();
flow_exit();
- br_handle_frame_hook = NULL;
- dp_avoid_bridge_exit();
}
module_init(dp_init);
git://git.onelab.eu / sliver-openvswitch.git / blobdiff