#include <linux/compat.h>
#include "openvswitch/datapath-protocol.h"
+#include "checksum.h"
#include "datapath.h"
#include "actions.h"
#include "flow.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 dp_port structures with just
+ * It is safe to access the datapath and vport structures with just
* dp_mutex.
*/
static struct datapath *dps[ODP_MAX];
static DEFINE_MUTEX(dp_mutex);
-static int new_dp_port(struct datapath *, struct odp_port *, int port_no);
+static int new_vport(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)
/* 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);
+ return vport_get_name(dp->ports[ODPP_LOCAL]);
}
static inline size_t br_nlmsg_size(void)
}
static int dp_fill_ifinfo(struct sk_buff *skb,
- const struct dp_port *port,
+ const struct vport *port,
int event, unsigned int flags)
{
const struct datapath *dp = port->dp;
- int ifindex = vport_get_ifindex(port->vport);
- int iflink = vport_get_iflink(port->vport);
+ int ifindex = vport_get_ifindex(port);
+ int iflink = vport_get_iflink(port);
struct ifinfomsg *hdr;
struct nlmsghdr *nlh;
hdr->__ifi_pad = 0;
hdr->ifi_type = ARPHRD_ETHER;
hdr->ifi_index = ifindex;
- hdr->ifi_flags = vport_get_flags(port->vport);
+ hdr->ifi_flags = vport_get_flags(port);
hdr->ifi_change = 0;
- 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));
+ NLA_PUT_STRING(skb, IFLA_IFNAME, vport_get_name(port));
+ NLA_PUT_U32(skb, IFLA_MASTER, vport_get_ifindex(dp->ports[ODPP_LOCAL]));
+ NLA_PUT_U32(skb, IFLA_MTU, vport_get_mtu(port));
#ifdef IFLA_OPERSTATE
NLA_PUT_U8(skb, IFLA_OPERSTATE,
- vport_is_running(port->vport)
- ? vport_get_operstate(port->vport)
+ vport_is_running(port)
+ ? vport_get_operstate(port)
: IF_OPER_DOWN);
#endif
- NLA_PUT(skb, IFLA_ADDRESS, ETH_ALEN,
- vport_get_addr(port->vport));
+ NLA_PUT(skb, IFLA_ADDRESS, ETH_ALEN, vport_get_addr(port));
if (ifindex != iflink)
NLA_PUT_U32(skb, IFLA_LINK,iflink);
return -EMSGSIZE;
}
-static void dp_ifinfo_notify(int event, struct dp_port *port)
+static void dp_ifinfo_notify(int event, struct vport *port)
{
struct sk_buff *skb;
int err = -ENOBUFS;
/* Set up our datapath device. */
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);
+ strcpy(internal_dev_port.type, "internal");
+ err = new_vport(dp, &internal_dev_port, ODPP_LOCAL);
if (err) {
if (err == -EBUSY)
err = -EEXIST;
goto err_destroy_local_port;
rcu_assign_pointer(dps[dp_idx], dp);
+ dp_sysfs_add_dp(dp);
+
mutex_unlock(&dp_mutex);
rtnl_unlock();
- dp_sysfs_add_dp(dp);
-
return 0;
err_destroy_local_port:
- dp_detach_port(dp->ports[ODPP_LOCAL], 1);
+ dp_detach_port(dp->ports[ODPP_LOCAL]);
err_destroy_table:
tbl_destroy(dp->table, NULL);
err_free_dp:
static void do_destroy_dp(struct datapath *dp)
{
- struct dp_port *p, *n;
+ struct vport *p, *n;
int i;
list_for_each_entry_safe (p, n, &dp->port_list, node)
if (p->port_no != ODPP_LOCAL)
- dp_detach_port(p, 1);
+ dp_detach_port(p);
dp_sysfs_del_dp(dp);
rcu_assign_pointer(dps[dp->dp_idx], NULL);
- dp_detach_port(dp->ports[ODPP_LOCAL], 1);
+ dp_detach_port(dp->ports[ODPP_LOCAL]);
tbl_destroy(dp->table, flow_free_tbl);
return err;
}
-static void release_dp_port(struct kobject *kobj)
-{
- struct dp_port *p = container_of(kobj, struct dp_port, kobj);
- kfree(p);
-}
-
-static struct kobj_type brport_ktype = {
-#ifdef CONFIG_SYSFS
- .sysfs_ops = &brport_sysfs_ops,
-#endif
- .release = release_dp_port
-};
-
/* Called with RTNL lock and dp_mutex. */
-static int new_dp_port(struct datapath *dp, struct odp_port *odp_port, int port_no)
+static int new_vport(struct datapath *dp, struct odp_port *odp_port, int port_no)
{
+ struct vport_parms parms;
struct vport *vport;
- struct dp_port *p;
- int err;
- vport = vport_locate(odp_port->devname);
- if (!vport) {
- vport_lock();
+ parms.name = odp_port->devname;
+ parms.type = odp_port->type;
+ parms.config = odp_port->config;
+ parms.dp = dp;
+ parms.port_no = port_no;
- 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();
+ vport_lock();
+ vport = vport_add(&parms);
+ vport_unlock();
- if (IS_ERR(vport))
- return PTR_ERR(vport);
- }
-
- p = kzalloc(sizeof(*p), GFP_KERNEL);
- if (!p)
- return -ENOMEM;
-
- p->port_no = port_no;
- p->dp = dp;
- p->vport = vport;
- atomic_set(&p->sflow_pool, 0);
-
- err = vport_attach(vport, p);
- if (err) {
- kfree(p);
- return err;
- }
+ if (IS_ERR(vport))
+ return PTR_ERR(vport);
- rcu_assign_pointer(dp->ports[port_no], p);
- list_add_rcu(&p->node, &dp->port_list);
+ rcu_assign_pointer(dp->ports[port_no], vport);
+ list_add_rcu(&vport->node, &dp->port_list);
dp->n_ports++;
- /* Initialize kobject for bridge. This will be added as
- * /sys/class/net/<devname>/brport later, if sysfs is enabled. */
- p->kobj.kset = NULL;
- kobject_init(&p->kobj, &brport_ktype);
-
- dp_ifinfo_notify(RTM_NEWLINK, p);
+ dp_ifinfo_notify(RTM_NEWLINK, vport);
return 0;
}
if (copy_from_user(&port, portp, sizeof port))
goto out;
port.devname[IFNAMSIZ - 1] = '\0';
+ port.type[VPORT_TYPE_SIZE - 1] = '\0';
rtnl_lock();
dp = get_dp_locked(dp_idx);
goto out_unlock_dp;
got_port_no:
- err = new_dp_port(dp, &port, port_no);
+ err = new_vport(dp, &port, port_no);
if (err)
goto out_unlock_dp;
return err;
}
-int dp_detach_port(struct dp_port *p, int may_delete)
+int dp_detach_port(struct vport *p)
{
- struct vport *vport = p->vport;
int err;
ASSERT_RTNL();
list_del_rcu(&p->node);
rcu_assign_pointer(p->dp->ports[p->port_no], NULL);
- err = vport_detach(vport);
- if (err)
- return err;
-
- /* Then wait until no one is still using it, and destroy it. */
- synchronize_rcu();
-
- if (may_delete) {
- const char *port_type = vport_get_type(vport);
+ /* Then destroy it. */
+ vport_lock();
+ err = vport_del(p);
+ vport_unlock();
- if (!strcmp(port_type, "netdev") || !strcmp(port_type, "internal")) {
- vport_lock();
- vport_del(vport);
- vport_unlock();
- }
- }
-
- kobject_put(&p->kobj);
-
- return 0;
+ return err;
}
static int detach_port(int dp_idx, int port_no)
{
- struct dp_port *p;
+ struct vport *p;
struct datapath *dp;
int err;
if (!p)
goto out_unlock_dp;
- err = dp_detach_port(p, 1);
+ err = dp_detach_port(p);
out_unlock_dp:
mutex_unlock(&dp->mutex);
}
/* Must be called with rcu_read_lock. */
-void dp_process_received_packet(struct dp_port *p, struct sk_buff *skb)
+void dp_process_received_packet(struct vport *p, struct sk_buff *skb)
{
struct datapath *dp = p->dp;
struct dp_stats_percpu *stats;
struct loop_counter *loop;
int error;
- OVS_CB(skb)->dp_port = p;
+ OVS_CB(skb)->vport = p;
if (!OVS_CB(skb)->flow) {
struct odp_flow_key key;
local_bh_enable();
}
-#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)
-{
- 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())
- pr_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;
-}
-#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)
-{
- /* 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
- 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:
- pr_err("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)
- /* 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
-}
-
-/* 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)
-{
-#ifdef CHECKSUM_HW
- if (OVS_CB(skb)->ip_summed == OVS_CSUM_COMPLETE)
- skb->ip_summed = CHECKSUM_NONE;
-#endif
-}
-
/* Append each packet in 'skb' list to 'queue'. There will be only one packet
* unless we broke up a GSO packet. */
static int queue_control_packets(struct sk_buff *skb, struct sk_buff_head *queue,
int port_no;
int err;
- if (OVS_CB(skb)->dp_port)
- port_no = OVS_CB(skb)->dp_port->port_no;
+ if (OVS_CB(skb)->vport)
+ port_no = OVS_CB(skb)->vport->port_no;
else
port_no = ODPP_LOCAL;
/* MTU of the dp pseudo-device: ETH_DATA_LEN or the minimum of the ports */
int dp_min_mtu(const struct datapath *dp)
{
- struct dp_port *p;
+ struct vport *p;
int mtu = 0;
ASSERT_RTNL();
/* Skip any internal ports, since that's what we're trying to
* set. */
- if (is_internal_vport(p->vport))
+ if (is_internal_vport(p))
continue;
- dev_mtu = vport_get_mtu(p->vport);
+ dev_mtu = vport_get_mtu(p);
if (!mtu || dev_mtu < mtu)
mtu = dev_mtu;
}
* be called with RTNL lock. */
void set_internal_devs_mtu(const struct datapath *dp)
{
- struct dp_port *p;
+ struct vport *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);
+ if (is_internal_vport(p))
+ vport_set_mtu(p, mtu);
}
}
-static int put_port(const struct dp_port *p, struct odp_port __user *uop)
+static int put_port(const struct vport *p, struct odp_port __user *uop)
{
struct odp_port op;
memset(&op, 0, sizeof op);
rcu_read_lock();
- strncpy(op.devname, vport_get_name(p->vport), sizeof op.devname);
+ strncpy(op.devname, vport_get_name(p), sizeof op.devname);
+ strncpy(op.type, vport_get_type(p), sizeof op.type);
rcu_read_unlock();
op.port = p->port_no;
- op.flags = is_internal_vport(p->vport) ? ODP_PORT_INTERNAL : 0;
return copy_to_user(uop, &op, sizeof op) ? -EFAULT : 0;
}
if (port.devname[0]) {
struct vport *vport;
- struct dp_port *dp_port;
int err = 0;
port.devname[IFNAMSIZ - 1] = '\0';
err = -ENODEV;
goto error_unlock;
}
-
- dp_port = vport_get_dp_port(vport);
- if (!dp_port || dp_port->dp != dp) {
+ if (vport->dp != dp) {
err = -ENOENT;
goto error_unlock;
}
- port.port = dp_port->port_no;
+ port.port = vport->port_no;
error_unlock:
rcu_read_unlock();
{
int idx = 0;
if (n_ports) {
- struct dp_port *p;
+ struct vport *p;
list_for_each_entry_rcu (p, &dp->port_list, node) {
if (put_port(p, &uports[idx]))
err = destroy_dp(dp_idx);
goto exit;
- case ODP_PORT_ATTACH:
+ case ODP_VPORT_ATTACH:
err = attach_port(dp_idx, (struct odp_port __user *)argp);
goto exit;
- case ODP_PORT_DETACH:
+ case ODP_VPORT_DETACH:
err = get_user(port_no, (int __user *)argp);
if (!err)
err = detach_port(dp_idx, port_no);
goto exit;
- case ODP_VPORT_ADD:
- err = vport_user_add((struct odp_vport_add __user *)argp);
- goto exit;
-
case ODP_VPORT_MOD:
- err = vport_user_mod((struct odp_vport_mod __user *)argp);
- goto exit;
-
- case ODP_VPORT_DEL:
- err = vport_user_del((char __user *)argp);
+ err = vport_user_mod((struct odp_port __user *)argp);
goto exit;
case ODP_VPORT_STATS_GET:
dp->sflow_probability = sflow_probability;
break;
- case ODP_PORT_QUERY:
+ case ODP_VPORT_QUERY:
err = query_port(dp, (struct odp_port __user *)argp);
break;
- case ODP_PORT_LIST:
+ case ODP_VPORT_LIST:
err = list_ports(dp, (struct odp_portvec __user *)argp);
break;
return openvswitch_ioctl(f, cmd, argp);
case ODP_DP_CREATE:
- case ODP_PORT_ATTACH:
- case ODP_PORT_DETACH:
- case ODP_VPORT_DEL:
+ case ODP_VPORT_ATTACH:
+ case ODP_VPORT_DETACH:
+ case ODP_VPORT_MOD:
case ODP_VPORT_MTU_SET:
case ODP_VPORT_MTU_GET:
case ODP_VPORT_ETHER_SET:
case ODP_GET_LISTEN_MASK:
case ODP_SET_SFLOW_PROBABILITY:
case ODP_GET_SFLOW_PROBABILITY:
- case ODP_PORT_QUERY:
+ case ODP_VPORT_QUERY:
/* Ioctls that just need their pointer argument extended. */
return openvswitch_ioctl(f, cmd, (unsigned long)compat_ptr(argp));
-
- case ODP_VPORT_ADD32:
- return compat_vport_user_add(compat_ptr(argp));
-
- case ODP_VPORT_MOD32:
- return compat_vport_user_mod(compat_ptr(argp));
}
dp = get_dp_locked(dp_idx);
goto exit;
switch (cmd) {
- case ODP_PORT_LIST32:
+ case ODP_VPORT_LIST32:
err = compat_list_ports(dp, compat_ptr(argp));
break;
return -EFAULT;
}
-ssize_t openvswitch_read(struct file *f, char __user *buf, size_t nbytes,
- loff_t *ppos)
+static ssize_t openvswitch_read(struct file *f, char __user *buf,
+ size_t nbytes, loff_t *ppos)
{
/* XXX is there sufficient synchronization here? */
int listeners = get_listen_mask(f);
if (skb->ip_summed == CHECKSUM_PARTIAL) {
if (copy_bytes == skb->len) {
__wsum csum = 0;
- unsigned int csum_start, csum_offset;
-
-#if LINUX_VERSION_CODE >= KERNEL_VERSION(2,6,22)
- csum_start = skb->csum_start - skb_headroom(skb);
- csum_offset = skb->csum_offset;
-#else
- csum_start = skb_transport_header(skb) - skb->data;
- csum_offset = skb->csum;
-#endif
+ u16 csum_start, csum_offset;
+
+ get_skb_csum_pointers(skb, &csum_start, &csum_offset);
BUG_ON(csum_start >= skb_headlen(skb));
retval = skb_copy_and_csum_datagram(skb, csum_start, buf + csum_start,
copy_bytes - csum_start, &csum);
return mask;
}
-struct file_operations openvswitch_fops = {
+static struct file_operations openvswitch_fops = {
/* XXX .aio_read = openvswitch_aio_read, */
.read = openvswitch_read,
.poll = openvswitch_poll,
git://git.onelab.eu / sliver-openvswitch.git / blobdiff