/*
- * Copyright (c) 2007, 2008, 2009, 2010 Nicira Networks.
+ * Copyright (c) 2007, 2008, 2009, 2010, 2011 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
/* Functions for managing the dp interface/device. */
+#define pr_fmt(fmt) KBUILD_MODNAME ": " fmt
+
#include <linux/init.h>
#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/jhash.h>
#include <linux/delay.h>
#include <linux/time.h>
#include <linux/etherdevice.h>
+#include <linux/genetlink.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/udp.h>
#include <linux/version.h>
#include <linux/ethtool.h>
-#include <linux/random.h>
#include <linux/wait.h>
#include <asm/system.h>
#include <asm/div64.h>
#include <asm/bug.h>
+#include <linux/highmem.h>
#include <linux/netfilter_bridge.h>
#include <linux/netfilter_ipv4.h>
#include <linux/inetdevice.h>
#include <linux/list.h>
#include <linux/rculist.h>
-#include <linux/workqueue.h>
#include <linux/dmi.h>
#include <net/inet_ecn.h>
-#include <net/llc.h>
+#include <net/genetlink.h>
#include "openvswitch/datapath-protocol.h"
+#include "checksum.h"
#include "datapath.h"
#include "actions.h"
-#include "dp_dev.h"
#include "flow.h"
+#include "vlan.h"
+#include "tunnel.h"
+#include "vport-internal_dev.h"
-#include "compat.h"
-
+#if LINUX_VERSION_CODE < KERNEL_VERSION(2,6,18) || \
+ LINUX_VERSION_CODE >= KERNEL_VERSION(3,1,0)
+#error Kernels before 2.6.18 or after 3.0 are not supported by this version of Open vSwitch.
+#endif
int (*dp_ioctl_hook)(struct net_device *dev, struct ifreq *rq, int cmd);
EXPORT_SYMBOL(dp_ioctl_hook);
-/* Datapaths. Protected on the read side by rcu_read_lock, on the write side
- * by dp_mutex.
+/**
+ * DOC: Locking:
+ *
+ * Writes to device state (add/remove datapath, port, set operations on vports,
+ * etc.) are protected by RTNL.
*
- * dp_mutex nests inside the RTNL lock: if you need both you must take the RTNL
- * lock first.
+ * Writes to other state (flow table modifications, set miscellaneous datapath
+ * parameters such as drop frags, etc.) are protected by genl_mutex. The RTNL
+ * lock nests inside genl_mutex.
*
- * It is safe to access the datapath and net_bridge_port structures with just
- * dp_mutex.
+ * Reads are protected by RCU.
+ *
+ * There are a few special cases (mostly stats) that have their own
+ * synchronization but they nest under all of above and don't interact with
+ * each other.
*/
-static struct datapath *dps[ODP_MAX];
-static DEFINE_MUTEX(dp_mutex);
-/* Number of milliseconds between runs of the maintenance thread. */
-#define MAINT_SLEEP_MSECS 1000
+/* Global list of datapaths to enable dumping them all out.
+ * Protected by genl_mutex.
+ */
+static LIST_HEAD(dps);
-static int new_nbp(struct datapath *, struct net_device *, int port_no);
+static struct vport *new_vport(const struct vport_parms *);
+static int queue_userspace_packets(struct datapath *, struct sk_buff *,
+ const struct dp_upcall_info *);
-/* Must be called with rcu_read_lock or dp_mutex. */
-struct datapath *get_dp(int dp_idx)
+/* Must be called with rcu_read_lock, genl_mutex, or RTNL lock. */
+struct datapath *get_dp(int dp_ifindex)
{
- if (dp_idx < 0 || dp_idx >= ODP_MAX)
- return NULL;
- return rcu_dereference(dps[dp_idx]);
+ struct datapath *dp = NULL;
+ struct net_device *dev;
+
+ rcu_read_lock();
+ dev = dev_get_by_index_rcu(&init_net, dp_ifindex);
+ if (dev) {
+ struct vport *vport = internal_dev_get_vport(dev);
+ if (vport)
+ dp = vport->dp;
+ }
+ rcu_read_unlock();
+
+ return dp;
}
EXPORT_SYMBOL_GPL(get_dp);
-static struct datapath *get_dp_locked(int dp_idx)
+/* Must be called with genl_mutex. */
+static struct flow_table *get_table_protected(struct datapath *dp)
{
- struct datapath *dp;
+ return rcu_dereference_protected(dp->table, lockdep_genl_is_held());
+}
- mutex_lock(&dp_mutex);
- dp = get_dp(dp_idx);
- if (dp)
- mutex_lock(&dp->mutex);
- mutex_unlock(&dp_mutex);
- return dp;
+/* Must be called with rcu_read_lock or RTNL lock. */
+static struct vport *get_vport_protected(struct datapath *dp, u16 port_no)
+{
+ return rcu_dereference_rtnl(dp->ports[port_no]);
+}
+
+/* Must be called with rcu_read_lock or RTNL lock. */
+const char *dp_name(const struct datapath *dp)
+{
+ return vport_get_name(rcu_dereference_rtnl(dp->ports[OVSP_LOCAL]));
}
static inline size_t br_nlmsg_size(void)
+ nla_total_size(MAX_ADDR_LEN) /* IFLA_ADDRESS */
+ nla_total_size(4) /* IFLA_MASTER */
+ nla_total_size(4) /* IFLA_MTU */
- + nla_total_size(4) /* IFLA_LINK */
+ nla_total_size(1); /* IFLA_OPERSTATE */
}
+/* Caller must hold RTNL lock. */
static int dp_fill_ifinfo(struct sk_buff *skb,
- const struct net_bridge_port *port,
+ const struct vport *port,
int event, unsigned int flags)
{
- const struct datapath *dp = port->dp;
- const struct net_device *dev = port->dev;
+ struct datapath *dp = port->dp;
+ int ifindex = vport_get_ifindex(port);
struct ifinfomsg *hdr;
struct nlmsghdr *nlh;
+ if (ifindex < 0)
+ return ifindex;
+
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);
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));
+ NLA_PUT_U32(skb, IFLA_MASTER,
+ vport_get_ifindex(get_vport_protected(dp, OVSP_LOCAL)));
+ NLA_PUT_U32(skb, IFLA_MTU, vport_get_mtu(port));
#ifdef IFLA_OPERSTATE
NLA_PUT_U8(skb, IFLA_OPERSTATE,
- netif_running(dev) ? dev->operstate : IF_OPER_DOWN);
+ vport_is_running(port)
+ ? vport_get_operstate(port)
+ : IF_OPER_DOWN);
#endif
- if (dev->addr_len)
- NLA_PUT(skb, IFLA_ADDRESS, dev->addr_len, dev->dev_addr);
-
- if (dev->ifindex != dev->iflink)
- NLA_PUT_U32(skb, IFLA_LINK, dev->iflink);
+ NLA_PUT(skb, IFLA_ADDRESS, ETH_ALEN, vport_get_addr(port));
return nlmsg_end(skb, nlh);
return -EMSGSIZE;
}
-static void dp_ifinfo_notify(int event, struct net_bridge_port *port)
+/* Caller must hold RTNL lock. */
+static void dp_ifinfo_notify(int event, struct vport *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)
.release = release_dp
};
-static int create_dp(int dp_idx, const char __user *devnamep)
+static void destroy_dp_rcu(struct rcu_head *rcu)
{
- struct net_device *dp_dev;
- char devname[IFNAMSIZ];
- struct datapath *dp;
- int err;
- int i;
-
- if (devnamep) {
- err = -EFAULT;
- if (strncpy_from_user(devname, devnamep, IFNAMSIZ - 1) < 0)
- goto err;
- devname[IFNAMSIZ - 1] = '\0';
- } else {
- snprintf(devname, sizeof devname, "of%d", dp_idx);
- }
-
- rtnl_lock();
- mutex_lock(&dp_mutex);
- err = -ENODEV;
- if (!try_module_get(THIS_MODULE))
- goto err_unlock;
+ struct datapath *dp = container_of(rcu, struct datapath, rcu);
- /* Exit early if a datapath with that number already exists.
- * (We don't use -EEXIST because that's ambiguous with 'devname'
- * conflicting with an existing network device name.) */
- err = -EBUSY;
- if (get_dp(dp_idx))
- goto err_put_module;
-
- err = -ENOMEM;
- dp = kzalloc(sizeof *dp, GFP_KERNEL);
- if (dp == NULL)
- goto err_put_module;
- INIT_LIST_HEAD(&dp->port_list);
- mutex_init(&dp->mutex);
- dp->dp_idx = dp_idx;
- for (i = 0; i < DP_N_QUEUES; i++)
- skb_queue_head_init(&dp->queues[i]);
- init_waitqueue_head(&dp->waitqueue);
+ flow_tbl_destroy(dp->table);
+ free_percpu(dp->stats_percpu);
+ kobject_put(&dp->ifobj);
+}
- /* Initialize kobject for bridge. This will be added as
- * /sys/class/net/<devname>/brif later, if sysfs is enabled. */
- dp->ifobj.kset = NULL;
- kobject_init(&dp->ifobj, &dp_ktype);
+/* Called with RTNL lock and genl_lock. */
+static struct vport *new_vport(const struct vport_parms *parms)
+{
+ struct vport *vport;
- /* Allocate table. */
- err = -ENOMEM;
- rcu_assign_pointer(dp->table, dp_table_create(DP_L1_SIZE));
- if (!dp->table)
- goto err_free_dp;
+ vport = vport_add(parms);
+ if (!IS_ERR(vport)) {
+ struct datapath *dp = parms->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;
+ rcu_assign_pointer(dp->ports[parms->port_no], vport);
+ list_add(&vport->node, &dp->port_list);
- err = new_nbp(dp, dp_dev, ODPP_LOCAL);
- if (err) {
- dp_dev_destroy(dp_dev);
- goto err_destroy_table;
+ dp_ifinfo_notify(RTM_NEWLINK, vport);
}
- dp->drop_frags = 0;
- dp->stats_percpu = alloc_percpu(struct dp_stats_percpu);
- if (!dp->stats_percpu)
- goto err_destroy_local_port;
+ return vport;
+}
- rcu_assign_pointer(dps[dp_idx], dp);
- mutex_unlock(&dp_mutex);
- rtnl_unlock();
+/* Called with RTNL lock. */
+void dp_detach_port(struct vport *p)
+{
+ ASSERT_RTNL();
- dp_sysfs_add_dp(dp);
+ if (p->port_no != OVSP_LOCAL)
+ dp_sysfs_del_if(p);
+ dp_ifinfo_notify(RTM_DELLINK, p);
- return 0;
+ /* First drop references to device. */
+ list_del(&p->node);
+ rcu_assign_pointer(p->dp->ports[p->port_no], NULL);
-err_destroy_local_port:
- dp_del_port(dp->ports[ODPP_LOCAL]);
-err_destroy_table:
- dp_table_destroy(dp->table, 0);
-err_free_dp:
- kfree(dp);
-err_put_module:
- module_put(THIS_MODULE);
-err_unlock:
- mutex_unlock(&dp_mutex);
- rtnl_unlock();
-err:
- return err;
+ /* Then destroy it. */
+ vport_del(p);
}
-static void do_destroy_dp(struct datapath *dp)
+/* Must be called with rcu_read_lock. */
+void dp_process_received_packet(struct vport *p, struct sk_buff *skb)
{
- struct net_bridge_port *p, *n;
- int i;
+ struct datapath *dp = p->dp;
+ struct sw_flow *flow;
+ struct dp_stats_percpu *stats;
+ int stats_counter_off;
+ int error;
- list_for_each_entry_safe (p, n, &dp->port_list, node)
- if (p->port_no != ODPP_LOCAL)
- dp_del_port(p);
+ OVS_CB(skb)->vport = p;
- dp_sysfs_del_dp(dp);
+ if (!OVS_CB(skb)->flow) {
+ struct sw_flow_key key;
+ int key_len;
+ bool is_frag;
- rcu_assign_pointer(dps[dp->dp_idx], NULL);
+ /* Extract flow from 'skb' into 'key'. */
+ error = flow_extract(skb, p->port_no, &key, &key_len, &is_frag);
+ if (unlikely(error)) {
+ kfree_skb(skb);
+ return;
+ }
- dp_del_port(dp->ports[ODPP_LOCAL]);
+ if (is_frag && dp->drop_frags) {
+ consume_skb(skb);
+ stats_counter_off = offsetof(struct dp_stats_percpu, n_frags);
+ goto out;
+ }
- dp_table_destroy(dp->table, 1);
+ /* Look up flow. */
+ flow = flow_tbl_lookup(rcu_dereference(dp->table), &key, key_len);
+ if (unlikely(!flow)) {
+ struct dp_upcall_info upcall;
+
+ upcall.cmd = OVS_PACKET_CMD_MISS;
+ upcall.key = &key;
+ upcall.userdata = 0;
+ upcall.sample_pool = 0;
+ upcall.actions = NULL;
+ upcall.actions_len = 0;
+ dp_upcall(dp, skb, &upcall);
+ stats_counter_off = offsetof(struct dp_stats_percpu, n_missed);
+ goto out;
+ }
- for (i = 0; i < DP_N_QUEUES; i++)
- skb_queue_purge(&dp->queues[i]);
- for (i = 0; i < DP_MAX_GROUPS; i++)
- kfree(dp->groups[i]);
- free_percpu(dp->stats_percpu);
- kobject_put(&dp->ifobj);
- module_put(THIS_MODULE);
-}
+ OVS_CB(skb)->flow = flow;
+ }
-static int destroy_dp(int dp_idx)
-{
- struct datapath *dp;
- int err;
+ stats_counter_off = offsetof(struct dp_stats_percpu, n_hit);
+ flow_used(OVS_CB(skb)->flow, skb);
+ execute_actions(dp, skb);
- rtnl_lock();
- mutex_lock(&dp_mutex);
- dp = get_dp(dp_idx);
- err = -ENODEV;
- if (!dp)
- goto err_unlock;
+out:
+ /* Update datapath statistics. */
+ local_bh_disable();
+ stats = per_cpu_ptr(dp->stats_percpu, smp_processor_id());
- do_destroy_dp(dp);
- err = 0;
+ write_seqcount_begin(&stats->seqlock);
+ (*(u64 *)((u8 *)stats + stats_counter_off))++;
+ write_seqcount_end(&stats->seqlock);
-err_unlock:
- mutex_unlock(&dp_mutex);
- rtnl_unlock();
- return err;
+ local_bh_enable();
}
-static void release_nbp(struct kobject *kobj)
+static void copy_and_csum_skb(struct sk_buff *skb, void *to)
{
- struct net_bridge_port *p = container_of(kobj, struct net_bridge_port, kobj);
- kfree(p);
+ u16 csum_start, csum_offset;
+ __wsum csum;
+
+ get_skb_csum_pointers(skb, &csum_start, &csum_offset);
+ csum_start -= skb_headroom(skb);
+
+ skb_copy_bits(skb, 0, to, csum_start);
+
+ csum = skb_copy_and_csum_bits(skb, csum_start, to + csum_start,
+ skb->len - csum_start, 0);
+ *(__sum16 *)(to + csum_start + csum_offset) = csum_fold(csum);
}
-static struct kobj_type brport_ktype = {
-#ifdef CONFIG_SYSFS
- .sysfs_ops = &brport_sysfs_ops,
-#endif
- .release = release_nbp
+static struct genl_family dp_packet_genl_family = {
+ .id = GENL_ID_GENERATE,
+ .hdrsize = sizeof(struct ovs_header),
+ .name = OVS_PACKET_FAMILY,
+ .version = 1,
+ .maxattr = OVS_PACKET_ATTR_MAX
};
-/* Called with RTNL lock and dp_mutex. */
-static int new_nbp(struct datapath *dp, struct net_device *dev, int port_no)
-{
- struct net_bridge_port *p;
-
- if (dev->br_port != NULL)
- return -EBUSY;
+/* Generic Netlink multicast groups for upcalls.
+ *
+ * We really want three unique multicast groups per datapath, but we can't even
+ * get one, because genl_register_mc_group() takes genl_lock, which is also
+ * held during Generic Netlink message processing, so trying to acquire
+ * multicast groups during OVS_DP_NEW processing deadlocks. Instead, we
+ * preallocate a few groups and use them round-robin for datapaths. Collision
+ * isn't fatal--multicast listeners should check that the family is the one
+ * that they want and discard others--but it wastes time and memory to receive
+ * unwanted messages.
+ */
+#define PACKET_N_MC_GROUPS 16
+static struct genl_multicast_group packet_mc_groups[PACKET_N_MC_GROUPS];
- p = kzalloc(sizeof(*p), GFP_KERNEL);
- if (!p)
- return -ENOMEM;
+static u32 packet_mc_group(struct datapath *dp, u8 cmd)
+{
+ u32 idx;
+ BUILD_BUG_ON_NOT_POWER_OF_2(PACKET_N_MC_GROUPS);
- dev_set_promiscuity(dev, 1);
- dev_hold(dev);
- p->port_no = port_no;
- p->dp = dp;
- p->dev = dev;
- atomic_set(&p->sflow_pool, 0);
- 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. */
- }
- dev_disable_lro(dev);
- rcu_assign_pointer(dp->ports[port_no], p);
- list_add_rcu(&p->node, &dp->port_list);
- dp->n_ports++;
+ idx = jhash_2words(dp->dp_ifindex, cmd, 0) & (PACKET_N_MC_GROUPS - 1);
+ return packet_mc_groups[idx].id;
+}
- /* 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);
+static int packet_register_mc_groups(void)
+{
+ int i;
- dp_ifinfo_notify(RTM_NEWLINK, p);
+ for (i = 0; i < PACKET_N_MC_GROUPS; i++) {
+ struct genl_multicast_group *group = &packet_mc_groups[i];
+ int error;
+ sprintf(group->name, "packet%d", i);
+ error = genl_register_mc_group(&dp_packet_genl_family, group);
+ if (error)
+ return error;
+ }
return 0;
}
-static int add_port(int dp_idx, struct odp_port __user *portp)
+int dp_upcall(struct datapath *dp, struct sk_buff *skb, const struct dp_upcall_info *upcall_info)
{
- struct net_device *dev;
- struct datapath *dp;
- struct odp_port port;
- int port_no;
+ struct dp_stats_percpu *stats;
int err;
- err = -EFAULT;
- if (copy_from_user(&port, portp, sizeof port))
- goto out;
- port.devname[IFNAMSIZ - 1] = '\0';
-
- rtnl_lock();
- dp = get_dp_locked(dp_idx);
- err = -ENODEV;
- if (!dp)
- goto out_unlock_rtnl;
-
- for (port_no = 1; port_no < DP_MAX_PORTS; port_no++)
- if (!dp->ports[port_no])
- goto got_port_no;
- err = -EFBIG;
- goto out_unlock_dp;
+ WARN_ON_ONCE(skb_shared(skb));
-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;
+ forward_ip_summed(skb, true);
- 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);
+ /* 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)) {
+ struct sk_buff *nskb = skb_gso_segment(skb, NETIF_F_SG | NETIF_F_HW_CSUM);
+
+ if (IS_ERR(nskb)) {
+ kfree_skb(skb);
+ err = PTR_ERR(nskb);
+ goto err;
+ }
+ consume_skb(skb);
+ skb = nskb;
}
- err = new_nbp(dp, dev, port_no);
+ err = queue_userspace_packets(dp, skb, upcall_info);
if (err)
- goto out_put;
+ goto err;
+
+ return 0;
- set_dp_devs_mtu(dp, dev);
- dp_sysfs_add_if(dp->ports[port_no]);
+err:
+ local_bh_disable();
+ stats = per_cpu_ptr(dp->stats_percpu, smp_processor_id());
- err = __put_user(port_no, &portp->port);
+ write_seqcount_begin(&stats->seqlock);
+ stats->n_lost++;
+ write_seqcount_end(&stats->seqlock);
+
+ local_bh_enable();
-out_put:
- dev_put(dev);
-out_unlock_dp:
- mutex_unlock(&dp->mutex);
-out_unlock_rtnl:
- rtnl_unlock();
-out:
return err;
}
-int dp_del_port(struct net_bridge_port *p)
+/* Send each packet in the 'skb' list to userspace for 'dp' as directed by
+ * 'upcall_info'. There will be only one packet unless we broke up a GSO
+ * packet.
+ */
+static int queue_userspace_packets(struct datapath *dp, struct sk_buff *skb,
+ const struct dp_upcall_info *upcall_info)
{
- ASSERT_RTNL();
+ u32 group = packet_mc_group(dp, upcall_info->cmd);
+ struct sk_buff *nskb;
+ int err;
- if (p->port_no != ODPP_LOCAL)
- dp_sysfs_del_if(p);
- dp_ifinfo_notify(RTM_DELLINK, p);
+ do {
+ struct ovs_header *upcall;
+ struct sk_buff *user_skb; /* to be queued to userspace */
+ struct nlattr *nla;
+ unsigned int len;
- p->dp->n_ports--;
+ nskb = skb->next;
+ skb->next = NULL;
- 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);
- }
+ err = vlan_deaccel_tag(skb);
+ if (unlikely(err))
+ goto err_kfree_skbs;
- /* First drop references to device. */
- dev_set_promiscuity(p->dev, -1);
- list_del_rcu(&p->node);
- rcu_assign_pointer(p->dp->ports[p->port_no], NULL);
- rcu_assign_pointer(p->dev->br_port, NULL);
+ if (nla_attr_size(skb->len) > USHRT_MAX)
+ goto err_kfree_skbs;
+
+ len = sizeof(struct ovs_header);
+ len += nla_total_size(skb->len);
+ len += nla_total_size(FLOW_BUFSIZE);
+ if (upcall_info->userdata)
+ len += nla_total_size(8);
+ if (upcall_info->sample_pool)
+ len += nla_total_size(4);
+ if (upcall_info->actions_len)
+ len += nla_total_size(upcall_info->actions_len);
+
+ user_skb = genlmsg_new(len, GFP_ATOMIC);
+ if (!user_skb) {
+ netlink_set_err(INIT_NET_GENL_SOCK, 0, group, -ENOBUFS);
+ goto err_kfree_skbs;
+ }
+
+ upcall = genlmsg_put(user_skb, 0, 0, &dp_packet_genl_family, 0, upcall_info->cmd);
+ upcall->dp_ifindex = dp->dp_ifindex;
- /* Then wait until no one is still using it, and destroy it. */
- synchronize_rcu();
+ nla = nla_nest_start(user_skb, OVS_PACKET_ATTR_KEY);
+ flow_to_nlattrs(upcall_info->key, user_skb);
+ nla_nest_end(user_skb, nla);
- if (is_dp_dev(p->dev))
- dp_dev_destroy(p->dev);
- dev_put(p->dev);
- kobject_put(&p->kobj);
+ if (upcall_info->userdata)
+ nla_put_u64(user_skb, OVS_PACKET_ATTR_USERDATA, upcall_info->userdata);
+ if (upcall_info->sample_pool)
+ nla_put_u32(user_skb, OVS_PACKET_ATTR_SAMPLE_POOL, upcall_info->sample_pool);
+ if (upcall_info->actions_len) {
+ const struct nlattr *actions = upcall_info->actions;
+ u32 actions_len = upcall_info->actions_len;
+ nla = nla_nest_start(user_skb, OVS_PACKET_ATTR_ACTIONS);
+ memcpy(__skb_put(user_skb, actions_len), actions, actions_len);
+ nla_nest_end(user_skb, nla);
+ }
+
+ nla = __nla_reserve(user_skb, OVS_PACKET_ATTR_PACKET, skb->len);
+ if (skb->ip_summed == CHECKSUM_PARTIAL)
+ copy_and_csum_skb(skb, nla_data(nla));
+ else
+ skb_copy_bits(skb, 0, nla_data(nla), skb->len);
+
+ err = genlmsg_multicast(user_skb, 0, group, GFP_ATOMIC);
+ if (err)
+ goto err_kfree_skbs;
+
+ consume_skb(skb);
+ skb = nskb;
+ } while (skb);
return 0;
+
+err_kfree_skbs:
+ kfree_skb(skb);
+ while ((skb = nskb) != NULL) {
+ nskb = skb->next;
+ kfree_skb(skb);
+ }
+ return err;
}
-static int del_port(int dp_idx, int port_no)
+/* Called with genl_mutex. */
+static int flush_flows(int dp_ifindex)
{
- struct net_bridge_port *p;
+ struct flow_table *old_table;
+ struct flow_table *new_table;
struct datapath *dp;
- LIST_HEAD(dp_devs);
- int err;
-
- err = -EINVAL;
- if (port_no < 0 || port_no >= DP_MAX_PORTS || port_no == ODPP_LOCAL)
- goto out;
- rtnl_lock();
- dp = get_dp_locked(dp_idx);
- err = -ENODEV;
+ dp = get_dp(dp_ifindex);
if (!dp)
- goto out_unlock_rtnl;
+ return -ENODEV;
- p = dp->ports[port_no];
- err = -ENOENT;
- if (!p)
- goto out_unlock_dp;
+ old_table = get_table_protected(dp);
+ new_table = flow_tbl_alloc(TBL_MIN_BUCKETS);
+ if (!new_table)
+ return -ENOMEM;
- err = dp_del_port(p);
+ rcu_assign_pointer(dp->table, new_table);
-out_unlock_dp:
- mutex_unlock(&dp->mutex);
-out_unlock_rtnl:
- rtnl_unlock();
-out:
- return err;
+ flow_tbl_deferred_destroy(old_table);
+ return 0;
}
-/* Must be called with rcu_read_lock. */
-static void
-do_port_input(struct net_bridge_port *p, struct sk_buff *skb)
+static int validate_actions(const struct nlattr *attr)
{
- /* LRO isn't suitable for bridging. We turn it off but make sure
- * that it wasn't reactivated. */
- if (skb_warn_if_lro(skb))
- return;
-
- /* 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;
+ const struct nlattr *a;
+ int rem;
+
+ nla_for_each_nested(a, attr, rem) {
+ static const u32 action_lens[OVS_ACTION_ATTR_MAX + 1] = {
+ [OVS_ACTION_ATTR_OUTPUT] = 4,
+ [OVS_ACTION_ATTR_USERSPACE] = 8,
+ [OVS_ACTION_ATTR_PUSH_VLAN] = 2,
+ [OVS_ACTION_ATTR_POP_VLAN] = 0,
+ [OVS_ACTION_ATTR_SET_DL_SRC] = ETH_ALEN,
+ [OVS_ACTION_ATTR_SET_DL_DST] = ETH_ALEN,
+ [OVS_ACTION_ATTR_SET_NW_SRC] = 4,
+ [OVS_ACTION_ATTR_SET_NW_DST] = 4,
+ [OVS_ACTION_ATTR_SET_NW_TOS] = 1,
+ [OVS_ACTION_ATTR_SET_TP_SRC] = 2,
+ [OVS_ACTION_ATTR_SET_TP_DST] = 2,
+ [OVS_ACTION_ATTR_SET_TUNNEL] = 8,
+ [OVS_ACTION_ATTR_SET_PRIORITY] = 4,
+ [OVS_ACTION_ATTR_POP_PRIORITY] = 0,
+ };
+ int type = nla_type(a);
+
+ if (type > OVS_ACTION_ATTR_MAX || nla_len(a) != action_lens[type])
+ return -EINVAL;
- /* Push the Ethernet header back on. */
- skb_push(skb, ETH_HLEN);
- skb_reset_mac_header(skb);
- dp_process_received_packet(skb, p);
-}
+ switch (type) {
+ case OVS_ACTION_ATTR_UNSPEC:
+ return -EINVAL;
-/* 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)
-{
- struct datapath *dp = p->dp;
- struct dp_stats_percpu *stats;
- struct odp_flow_key key;
- struct sw_flow *flow;
+ case OVS_ACTION_ATTR_USERSPACE:
+ case OVS_ACTION_ATTR_POP_VLAN:
+ case OVS_ACTION_ATTR_SET_DL_SRC:
+ case OVS_ACTION_ATTR_SET_DL_DST:
+ case OVS_ACTION_ATTR_SET_NW_SRC:
+ case OVS_ACTION_ATTR_SET_NW_DST:
+ case OVS_ACTION_ATTR_SET_TP_SRC:
+ case OVS_ACTION_ATTR_SET_TP_DST:
+ case OVS_ACTION_ATTR_SET_TUNNEL:
+ case OVS_ACTION_ATTR_SET_PRIORITY:
+ case OVS_ACTION_ATTR_POP_PRIORITY:
+ /* No validation needed. */
+ break;
- WARN_ON_ONCE(skb_shared(skb));
+ case OVS_ACTION_ATTR_OUTPUT:
+ if (nla_get_u32(a) >= DP_MAX_PORTS)
+ return -EINVAL;
+ break;
- compute_ip_summed(skb, false);
+ case OVS_ACTION_ATTR_PUSH_VLAN:
+ if (nla_get_be16(a) & htons(VLAN_CFI_MASK))
+ return -EINVAL;
+ break;
- /* BHs are off so we don't have to use get_cpu()/put_cpu() here. */
- stats = percpu_ptr(dp->stats_percpu, smp_processor_id());
+ case OVS_ACTION_ATTR_SET_NW_TOS:
+ if (nla_get_u8(a) & INET_ECN_MASK)
+ return -EINVAL;
+ break;
- if (flow_extract(skb, p ? p->port_no : ODPP_NONE, &key)) {
- if (dp->drop_frags) {
- kfree_skb(skb);
- stats->n_frags++;
- return;
+ default:
+ return -EOPNOTSUPP;
}
}
- flow = dp_table_lookup(rcu_dereference(dp->table), &key);
- if (flow) {
- struct sw_flow_actions *acts = rcu_dereference(flow->sf_acts);
- flow_used(flow, skb);
- execute_actions(dp, skb, &key, acts->actions, acts->n_actions,
- GFP_ATOMIC);
- stats->n_hit++;
- } else {
- stats->n_missed++;
- dp_output_control(dp, skb, _ODPL_MISS_NR, 0);
- }
-}
+ if (rem > 0)
+ return -EINVAL;
-/*
- * 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)
-{
- do_port_input(p, skb);
- return NULL;
+ return 0;
}
-#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)
+static void clear_stats(struct sw_flow *flow)
{
- do_port_input(p, *pskb);
- return 1;
+ flow->used = 0;
+ flow->tcp_flags = 0;
+ flow->packet_count = 0;
+ flow->byte_count = 0;
}
-#else
-#error
-#endif
-#if defined(CONFIG_XEN) && defined(HAVE_PROTO_DATA_VALID)
-/* This code is based on a skb_checksum_setup from net/dev/core.c from a
- * combination of Lenny's 2.6.26 Xen kernel and Xen's
- * linux-2.6.18-92.1.10.el5.xs5.0.0.394.644. We can't call this function
- * directly because it isn't exported in all versions. */
-static int skb_pull_up_to(struct sk_buff *skb, void *ptr)
+static int ovs_packet_cmd_execute(struct sk_buff *skb, struct genl_info *info)
{
- if (ptr < (void *)skb->tail)
- return 1;
- if (__pskb_pull_tail(skb,
- ptr - (void *)skb->data - skb_headlen(skb))) {
- return 1;
- } else {
- return 0;
- }
-}
+ struct ovs_header *ovs_header = info->userhdr;
+ struct nlattr **a = info->attrs;
+ struct sw_flow_actions *acts;
+ struct sk_buff *packet;
+ struct sw_flow *flow;
+ struct datapath *dp;
+ struct ethhdr *eth;
+ bool is_frag;
+ int len;
+ int err;
+ int key_len;
-int vswitch_skb_checksum_setup(struct sk_buff *skb)
-{
- struct iphdr *iph;
- unsigned char *th;
- int err = -EPROTO;
- __u16 csum_start, csum_offset;
+ err = -EINVAL;
+ if (!a[OVS_PACKET_ATTR_PACKET] || !a[OVS_PACKET_ATTR_KEY] ||
+ !a[OVS_PACKET_ATTR_ACTIONS] ||
+ nla_len(a[OVS_PACKET_ATTR_PACKET]) < ETH_HLEN)
+ goto err;
- if (!skb->proto_csum_blank)
- return 0;
+ err = validate_actions(a[OVS_PACKET_ATTR_ACTIONS]);
+ if (err)
+ goto err;
- if (skb->protocol != htons(ETH_P_IP))
- goto out;
-
- if (!skb_pull_up_to(skb, skb_network_header(skb) + sizeof(struct iphdr)))
- 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;
- }
+ len = nla_len(a[OVS_PACKET_ATTR_PACKET]);
+ packet = __dev_alloc_skb(NET_IP_ALIGN + len, GFP_KERNEL);
+ err = -ENOMEM;
+ if (!packet)
+ goto err;
+ skb_reserve(packet, NET_IP_ALIGN);
- if (!skb_pull_up_to(skb, th + csum_offset + 2))
- goto out;
+ memcpy(__skb_put(packet, len), nla_data(a[OVS_PACKET_ATTR_PACKET]), len);
- skb->ip_summed = CHECKSUM_PARTIAL;
- skb->proto_csum_blank = 0;
+ skb_reset_mac_header(packet);
+ eth = eth_hdr(packet);
-#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
+ /* Normally, setting the skb 'protocol' field would be handled by a
+ * call to eth_type_trans(), but it assumes there's a sending
+ * device, which we may not have. */
+ if (ntohs(eth->h_proto) >= 1536)
+ packet->protocol = eth->h_proto;
+ else
+ packet->protocol = htons(ETH_P_802_2);
- err = 0;
+ /* Build an sw_flow for sending this packet. */
+ flow = flow_alloc();
+ err = PTR_ERR(flow);
+ if (IS_ERR(flow))
+ goto err_kfree_skb;
-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. The only real user of this type
- * with bridging is Xen (on later kernels).
- * 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.
- *
- * The bridge has similar behavior and this function closely resembles
- * skb_forward_csum(). 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
-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, 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.
- *
- * The one exception to this is if we are on the transmit path
- * (basically after skb_checksum_setup() has been run) the type has
- * already been converted, so we should stay with that. */
- 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:
- 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)
- /* 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
-}
+ err = flow_extract(packet, -1, &flow->key, &key_len, &is_frag);
+ if (err)
+ goto err_flow_put;
-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
-}
+ err = flow_metadata_from_nlattrs(&flow->key.eth.in_port,
+ &flow->key.eth.tun_id,
+ a[OVS_PACKET_ATTR_KEY]);
+ if (err)
+ goto err_flow_put;
-/* 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 queue_no, u32 arg)
-{
- struct sk_buff *nskb;
- int port_no;
- int err;
+ flow->hash = flow_hash(&flow->key, key_len);
- 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;
- }
+ acts = flow_actions_alloc(a[OVS_PACKET_ATTR_ACTIONS]);
+ err = PTR_ERR(acts);
+ if (IS_ERR(acts))
+ goto err_flow_put;
+ rcu_assign_pointer(flow->sf_acts, acts);
- do {
- struct odp_msg *header;
+ OVS_CB(packet)->flow = flow;
- nskb = skb->next;
- 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.
- */
- 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
- * 2.6.22 introduced the csum_start field for this
- * purpose, but we should point the transport header to
- * it anyway for backward compatibility, as
- * dev_queue_xmit() does even in 2.6.28. */
- 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
+ rcu_read_lock();
+ dp = get_dp(ovs_header->dp_ifindex);
+ err = -ENODEV;
+ if (!dp)
+ goto err_unlock;
- err = skb_cow(skb, sizeof *header);
- if (err)
- goto err_kfree_skbs;
+ if (flow->key.eth.in_port < DP_MAX_PORTS)
+ OVS_CB(packet)->vport = get_vport_protected(dp,
+ flow->key.eth.in_port);
- header = (struct odp_msg*)__skb_push(skb, sizeof *header);
- header->type = queue_no;
- header->length = skb->len;
- header->port = port_no;
- header->reserved = 0;
- header->arg = arg;
- skb_queue_tail(queue, skb);
+ err = execute_actions(dp, packet);
+ rcu_read_unlock();
- skb = nskb;
- } while (skb);
- return 0;
+ flow_put(flow);
+ return err;
-err_kfree_skbs:
- kfree_skb(skb);
- while ((skb = nskb) != NULL) {
- nskb = skb->next;
- kfree_skb(skb);
- }
+err_unlock:
+ rcu_read_unlock();
+err_flow_put:
+ flow_put(flow);
+err_kfree_skb:
+ kfree_skb(packet);
+err:
return err;
}
-int
-dp_output_control(struct datapath *dp, struct sk_buff *skb, int queue_no,
- u32 arg)
-{
- struct dp_stats_percpu *stats;
- struct sk_buff_head *queue;
- int err;
-
- WARN_ON_ONCE(skb_shared(skb));
- BUG_ON(queue_no != _ODPL_MISS_NR && queue_no != _ODPL_ACTION_NR && queue_no != _ODPL_SFLOW_NR);
- queue = &dp->queues[queue_no];
- err = -ENOBUFS;
- if (skb_queue_len(queue) >= DP_MAX_QUEUE_LEN)
- goto err_kfree_skb;
-
- forward_ip_summed(skb);
+static const struct nla_policy packet_policy[OVS_PACKET_ATTR_MAX + 1] = {
+ [OVS_PACKET_ATTR_PACKET] = { .type = NLA_UNSPEC },
+ [OVS_PACKET_ATTR_KEY] = { .type = NLA_NESTED },
+ [OVS_PACKET_ATTR_ACTIONS] = { .type = NLA_NESTED },
+};
- /* 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)) {
- struct sk_buff *nskb = skb_gso_segment(skb, 0);
- if (nskb) {
- kfree_skb(skb);
- skb = nskb;
- if (unlikely(IS_ERR(skb))) {
- err = PTR_ERR(skb);
- goto err;
- }
- } else {
- /* XXX This case might not be possible. It's hard to
- * tell from the skb_gso_segment() code and comment. */
- }
+static struct genl_ops dp_packet_genl_ops[] = {
+ { .cmd = OVS_PACKET_CMD_EXECUTE,
+ .flags = GENL_ADMIN_PERM, /* Requires CAP_NET_ADMIN privilege. */
+ .policy = packet_policy,
+ .doit = ovs_packet_cmd_execute
}
+};
- err = queue_control_packets(skb, queue, queue_no, arg);
- wake_up_interruptible(&dp->waitqueue);
- return err;
+static void get_dp_stats(struct datapath *dp, struct ovs_dp_stats *stats)
+{
+ int i;
+ struct flow_table *table = get_table_protected(dp);
-err_kfree_skb:
- kfree_skb(skb);
-err:
- stats = percpu_ptr(dp->stats_percpu, get_cpu());
- stats->n_lost++;
- put_cpu();
+ stats->n_flows = flow_tbl_count(table);
- return err;
-}
+ stats->n_frags = stats->n_hit = stats->n_missed = stats->n_lost = 0;
+ for_each_possible_cpu(i) {
+ const struct dp_stats_percpu *percpu_stats;
+ struct dp_stats_percpu local_stats;
+ unsigned seqcount;
-static int flush_flows(struct datapath *dp)
-{
- dp->n_flows = 0;
- return dp_table_flush(dp);
-}
+ percpu_stats = per_cpu_ptr(dp->stats_percpu, i);
-static int validate_actions(const struct sw_flow_actions *actions)
-{
- unsigned int i;
+ do {
+ seqcount = read_seqcount_begin(&percpu_stats->seqlock);
+ local_stats = *percpu_stats;
+ } while (read_seqcount_retry(&percpu_stats->seqlock, seqcount));
- for (i = 0; i < actions->n_actions; i++) {
- const union odp_action *a = &actions->actions[i];
- switch (a->type) {
- case ODPAT_OUTPUT:
- if (a->output.port >= DP_MAX_PORTS)
- return -EINVAL;
- break;
+ stats->n_frags += local_stats.n_frags;
+ stats->n_hit += local_stats.n_hit;
+ stats->n_missed += local_stats.n_missed;
+ stats->n_lost += local_stats.n_lost;
+ }
+}
- case ODPAT_OUTPUT_GROUP:
- if (a->output_group.group >= DP_MAX_GROUPS)
- return -EINVAL;
- break;
+static const struct nla_policy flow_policy[OVS_FLOW_ATTR_MAX + 1] = {
+ [OVS_FLOW_ATTR_KEY] = { .type = NLA_NESTED },
+ [OVS_FLOW_ATTR_ACTIONS] = { .type = NLA_NESTED },
+ [OVS_FLOW_ATTR_CLEAR] = { .type = NLA_FLAG },
+};
- case ODPAT_SET_VLAN_VID:
- if (a->vlan_vid.vlan_vid & htons(~VLAN_VID_MASK))
- return -EINVAL;
- break;
+static struct genl_family dp_flow_genl_family = {
+ .id = GENL_ID_GENERATE,
+ .hdrsize = sizeof(struct ovs_header),
+ .name = OVS_FLOW_FAMILY,
+ .version = 1,
+ .maxattr = OVS_FLOW_ATTR_MAX
+};
- case ODPAT_SET_VLAN_PCP:
- if (a->vlan_pcp.vlan_pcp
- & ~(VLAN_PCP_MASK >> VLAN_PCP_SHIFT))
- return -EINVAL;
- break;
+static struct genl_multicast_group dp_flow_multicast_group = {
+ .name = OVS_FLOW_MCGROUP
+};
- case ODPAT_SET_NW_TOS:
- if (a->nw_tos.nw_tos & INET_ECN_MASK)
- return -EINVAL;
- break;
+/* Called with genl_lock. */
+static int ovs_flow_cmd_fill_info(struct sw_flow *flow, struct datapath *dp,
+ struct sk_buff *skb, u32 pid, u32 seq, u32 flags, u8 cmd)
+{
+ const int skb_orig_len = skb->len;
+ const struct sw_flow_actions *sf_acts;
+ struct ovs_flow_stats stats;
+ struct ovs_header *ovs_header;
+ struct nlattr *nla;
+ unsigned long used;
+ u8 tcp_flags;
+ int err;
- default:
- if (a->type >= ODPAT_N_ACTIONS)
- return -EOPNOTSUPP;
- break;
- }
- }
+ sf_acts = rcu_dereference_protected(flow->sf_acts,
+ lockdep_genl_is_held());
- return 0;
-}
+ ovs_header = genlmsg_put(skb, pid, seq, &dp_flow_genl_family, flags, cmd);
+ if (!ovs_header)
+ return -EMSGSIZE;
-static struct sw_flow_actions *get_actions(const struct odp_flow *flow)
-{
- struct sw_flow_actions *actions;
- int error;
+ ovs_header->dp_ifindex = dp->dp_ifindex;
- actions = flow_actions_alloc(flow->n_actions);
- error = PTR_ERR(actions);
- if (IS_ERR(actions))
+ nla = nla_nest_start(skb, OVS_FLOW_ATTR_KEY);
+ if (!nla)
+ goto nla_put_failure;
+ err = flow_to_nlattrs(&flow->key, skb);
+ if (err)
+ goto error;
+ nla_nest_end(skb, nla);
+
+ spin_lock_bh(&flow->lock);
+ used = flow->used;
+ stats.n_packets = flow->packet_count;
+ stats.n_bytes = flow->byte_count;
+ tcp_flags = flow->tcp_flags;
+ spin_unlock_bh(&flow->lock);
+
+ if (used)
+ NLA_PUT_U64(skb, OVS_FLOW_ATTR_USED, flow_used_time(used));
+
+ if (stats.n_packets)
+ NLA_PUT(skb, OVS_FLOW_ATTR_STATS, sizeof(struct ovs_flow_stats), &stats);
+
+ if (tcp_flags)
+ NLA_PUT_U8(skb, OVS_FLOW_ATTR_TCP_FLAGS, tcp_flags);
+
+ /* If OVS_FLOW_ATTR_ACTIONS doesn't fit, skip dumping the actions if
+ * this is the first flow to be dumped into 'skb'. This is unusual for
+ * Netlink but individual action lists can be longer than
+ * NLMSG_GOODSIZE and thus entirely undumpable if we didn't do this.
+ * The userspace caller can always fetch the actions separately if it
+ * really wants them. (Most userspace callers in fact don't care.)
+ *
+ * This can only fail for dump operations because the skb is always
+ * properly sized for single flows.
+ */
+ err = nla_put(skb, OVS_FLOW_ATTR_ACTIONS, sf_acts->actions_len,
+ sf_acts->actions);
+ if (err < 0 && skb_orig_len)
goto error;
- error = -EFAULT;
- if (copy_from_user(actions->actions, flow->actions,
- flow->n_actions * sizeof(union odp_action)))
- goto error_free_actions;
- error = validate_actions(actions);
- if (error)
- goto error_free_actions;
-
- return actions;
+ return genlmsg_end(skb, ovs_header);
-error_free_actions:
- kfree(actions);
+nla_put_failure:
+ err = -EMSGSIZE;
error:
- return ERR_PTR(error);
+ genlmsg_cancel(skb, ovs_header);
+ return err;
}
-static void get_stats(struct sw_flow *flow, struct odp_flow_stats *stats)
+static struct sk_buff *ovs_flow_cmd_alloc_info(struct sw_flow *flow)
{
- if (flow->used.tv_sec) {
- stats->used_sec = flow->used.tv_sec;
- stats->used_nsec = flow->used.tv_nsec;
- } else {
- stats->used_sec = 0;
- stats->used_nsec = 0;
- }
- stats->n_packets = flow->packet_count;
- stats->n_bytes = flow->byte_count;
- stats->ip_tos = flow->ip_tos;
- stats->tcp_flags = flow->tcp_flags;
- stats->error = 0;
+ const struct sw_flow_actions *sf_acts;
+ int len;
+
+ sf_acts = rcu_dereference_protected(flow->sf_acts,
+ lockdep_genl_is_held());
+
+ len = nla_total_size(FLOW_BUFSIZE); /* OVS_FLOW_ATTR_KEY */
+ len += nla_total_size(sf_acts->actions_len); /* OVS_FLOW_ATTR_ACTIONS */
+ len += nla_total_size(sizeof(struct ovs_flow_stats)); /* OVS_FLOW_ATTR_STATS */
+ len += nla_total_size(1); /* OVS_FLOW_ATTR_TCP_FLAGS */
+ len += nla_total_size(8); /* OVS_FLOW_ATTR_USED */
+ return genlmsg_new(NLMSG_ALIGN(sizeof(struct ovs_header)) + len, GFP_KERNEL);
}
-static void clear_stats(struct sw_flow *flow)
+static struct sk_buff *ovs_flow_cmd_build_info(struct sw_flow *flow, struct datapath *dp,
+ u32 pid, u32 seq, u8 cmd)
{
- flow->used.tv_sec = flow->used.tv_nsec = 0;
- flow->tcp_flags = 0;
- flow->ip_tos = 0;
- flow->packet_count = 0;
- flow->byte_count = 0;
+ struct sk_buff *skb;
+ int retval;
+
+ skb = ovs_flow_cmd_alloc_info(flow);
+ if (!skb)
+ return ERR_PTR(-ENOMEM);
+
+ retval = ovs_flow_cmd_fill_info(flow, dp, skb, pid, seq, 0, cmd);
+ BUG_ON(retval < 0);
+ return skb;
}
-static int put_flow(struct datapath *dp, struct odp_flow_put __user *ufp)
+static int ovs_flow_cmd_new_or_set(struct sk_buff *skb, struct genl_info *info)
{
- struct odp_flow_put uf;
+ struct nlattr **a = info->attrs;
+ struct ovs_header *ovs_header = info->userhdr;
+ struct sw_flow_key key;
struct sw_flow *flow;
- struct dp_table *table;
- struct odp_flow_stats stats;
+ struct sk_buff *reply;
+ struct datapath *dp;
+ struct flow_table *table;
int error;
+ int key_len;
+
+ /* Extract key. */
+ error = -EINVAL;
+ if (!a[OVS_FLOW_ATTR_KEY])
+ goto error;
+ error = flow_from_nlattrs(&key, &key_len, a[OVS_FLOW_ATTR_KEY]);
+ if (error)
+ goto error;
+
+ /* Validate actions. */
+ if (a[OVS_FLOW_ATTR_ACTIONS]) {
+ error = validate_actions(a[OVS_FLOW_ATTR_ACTIONS]);
+ if (error)
+ goto error;
+ } else if (info->genlhdr->cmd == OVS_FLOW_CMD_NEW) {
+ error = -EINVAL;
+ goto error;
+ }
- error = -EFAULT;
- if (copy_from_user(&uf, ufp, sizeof(struct odp_flow_put)))
+ dp = get_dp(ovs_header->dp_ifindex);
+ error = -ENODEV;
+ if (!dp)
goto error;
- memset(uf.flow.key.reserved, 0, sizeof uf.flow.key.reserved);
- table = rcu_dereference(dp->table);
- flow = dp_table_lookup(table, &uf.flow.key);
+ table = get_table_protected(dp);
+ flow = flow_tbl_lookup(table, &key, key_len);
if (!flow) {
- /* No such flow. */
struct sw_flow_actions *acts;
+ /* Bail out if we're not allowed to create a new flow. */
error = -ENOENT;
- if (!(uf.flags & ODPPF_CREATE))
+ if (info->genlhdr->cmd == OVS_FLOW_CMD_SET)
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 (error)
- goto error;
- table = rcu_dereference(dp->table);
+ if (flow_tbl_need_to_expand(table)) {
+ struct flow_table *new_table;
+
+ new_table = flow_tbl_expand(table);
+ if (!IS_ERR(new_table)) {
+ rcu_assign_pointer(dp->table, new_table);
+ flow_tbl_deferred_destroy(table);
+ table = get_table_protected(dp);
+ }
}
/* Allocate flow. */
- error = -ENOMEM;
- flow = kmem_cache_alloc(flow_cache, GFP_KERNEL);
- if (flow == NULL)
+ flow = flow_alloc();
+ if (IS_ERR(flow)) {
+ error = PTR_ERR(flow);
goto error;
- flow->key = uf.flow.key;
- spin_lock_init(&flow->lock);
+ }
+ flow->key = key;
clear_stats(flow);
/* Obtain actions. */
- acts = get_actions(&uf.flow);
+ acts = flow_actions_alloc(a[OVS_FLOW_ATTR_ACTIONS]);
error = PTR_ERR(acts);
if (IS_ERR(acts))
goto error_free_flow;
rcu_assign_pointer(flow->sf_acts, acts);
/* Put flow in bucket. */
- error = dp_table_insert(table, flow);
- if (error)
- goto error_free_flow_acts;
- dp->n_flows++;
- memset(&stats, 0, sizeof(struct odp_flow_stats));
+ flow->hash = flow_hash(&key, key_len);
+ flow_tbl_insert(table, flow);
+
+ reply = ovs_flow_cmd_build_info(flow, dp, info->snd_pid,
+ info->snd_seq, OVS_FLOW_CMD_NEW);
} else {
/* We found a matching flow. */
- struct sw_flow_actions *old_acts, *new_acts;
- unsigned long int flags;
+ struct sw_flow_actions *old_acts;
- /* Bail out if we're not allowed to modify an existing flow. */
+ /* Bail out if we're not allowed to modify an existing flow.
+ * We accept NLM_F_CREATE in place of the intended NLM_F_EXCL
+ * because Generic Netlink treats the latter as a dump
+ * request. We also accept NLM_F_EXCL in case that bug ever
+ * gets fixed.
+ */
error = -EEXIST;
- if (!(uf.flags & ODPPF_MODIFY))
+ if (info->genlhdr->cmd == OVS_FLOW_CMD_NEW &&
+ info->nlhdr->nlmsg_flags & (NLM_F_CREATE | NLM_F_EXCL))
goto error;
- /* Swap actions. */
- new_acts = get_actions(&uf.flow);
- error = PTR_ERR(new_acts);
- if (IS_ERR(new_acts))
- goto error;
- old_acts = rcu_dereference(flow->sf_acts);
- if (old_acts->n_actions != new_acts->n_actions ||
- memcmp(old_acts->actions, new_acts->actions,
- sizeof(union odp_action) * old_acts->n_actions)) {
+ /* Update actions. */
+ old_acts = rcu_dereference_protected(flow->sf_acts,
+ lockdep_genl_is_held());
+ if (a[OVS_FLOW_ATTR_ACTIONS] &&
+ (old_acts->actions_len != nla_len(a[OVS_FLOW_ATTR_ACTIONS]) ||
+ memcmp(old_acts->actions, nla_data(a[OVS_FLOW_ATTR_ACTIONS]),
+ old_acts->actions_len))) {
+ struct sw_flow_actions *new_acts;
+
+ new_acts = flow_actions_alloc(a[OVS_FLOW_ATTR_ACTIONS]);
+ error = PTR_ERR(new_acts);
+ if (IS_ERR(new_acts))
+ goto error;
+
rcu_assign_pointer(flow->sf_acts, new_acts);
flow_deferred_free_acts(old_acts);
- } else {
- kfree(new_acts);
}
- /* Fetch stats, then clear them if necessary. */
- spin_lock_irqsave(&flow->lock, flags);
- get_stats(flow, &stats);
- if (uf.flags & ODPPF_ZERO_STATS)
+ reply = ovs_flow_cmd_build_info(flow, dp, info->snd_pid,
+ info->snd_seq, OVS_FLOW_CMD_NEW);
+
+ /* Clear stats. */
+ if (a[OVS_FLOW_ATTR_CLEAR]) {
+ spin_lock_bh(&flow->lock);
clear_stats(flow);
- spin_unlock_irqrestore(&flow->lock, flags);
+ spin_unlock_bh(&flow->lock);
+ }
}
- /* Copy stats to userspace. */
- if (__copy_to_user(&ufp->flow.stats, &stats,
- sizeof(struct odp_flow_stats)))
- return -EFAULT;
+ if (!IS_ERR(reply))
+ genl_notify(reply, genl_info_net(info), info->snd_pid,
+ dp_flow_multicast_group.id, info->nlhdr, GFP_KERNEL);
+ else
+ netlink_set_err(INIT_NET_GENL_SOCK, 0,
+ dp_flow_multicast_group.id, PTR_ERR(reply));
return 0;
-error_free_flow_acts:
- kfree(flow->sf_acts);
error_free_flow:
- kmem_cache_free(flow_cache, flow);
+ flow_put(flow);
error:
return error;
}
-static int put_actions(const struct sw_flow *flow, struct odp_flow __user *ufp)
+static int ovs_flow_cmd_get(struct sk_buff *skb, struct genl_info *info)
{
- union odp_action __user *actions;
- struct sw_flow_actions *sf_acts;
- u32 n_actions;
+ struct nlattr **a = info->attrs;
+ struct ovs_header *ovs_header = info->userhdr;
+ struct sw_flow_key key;
+ struct sk_buff *reply;
+ struct sw_flow *flow;
+ struct datapath *dp;
+ struct flow_table *table;
+ int err;
+ int key_len;
- if (__get_user(actions, &ufp->actions) ||
- __get_user(n_actions, &ufp->n_actions))
- return -EFAULT;
+ if (!a[OVS_FLOW_ATTR_KEY])
+ return -EINVAL;
+ err = flow_from_nlattrs(&key, &key_len, a[OVS_FLOW_ATTR_KEY]);
+ if (err)
+ return err;
- if (!n_actions)
- return 0;
+ dp = get_dp(ovs_header->dp_ifindex);
+ if (!dp)
+ return -ENODEV;
+
+ table = get_table_protected(dp);
+ flow = flow_tbl_lookup(table, &key, key_len);
+ if (!flow)
+ return -ENOENT;
- sf_acts = rcu_dereference(flow->sf_acts);
- if (__put_user(sf_acts->n_actions, &ufp->n_actions) ||
- (actions && copy_to_user(actions, sf_acts->actions,
- sizeof(union odp_action) *
- min(sf_acts->n_actions, n_actions))))
- return -EFAULT;
+ reply = ovs_flow_cmd_build_info(flow, dp, info->snd_pid, info->snd_seq, OVS_FLOW_CMD_NEW);
+ if (IS_ERR(reply))
+ return PTR_ERR(reply);
+
+ return genlmsg_reply(reply, info);
+}
+static int ovs_flow_cmd_del(struct sk_buff *skb, struct genl_info *info)
+{
+ struct nlattr **a = info->attrs;
+ struct ovs_header *ovs_header = info->userhdr;
+ struct sw_flow_key key;
+ struct sk_buff *reply;
+ struct sw_flow *flow;
+ struct datapath *dp;
+ struct flow_table *table;
+ int err;
+ int key_len;
+
+ if (!a[OVS_FLOW_ATTR_KEY])
+ return flush_flows(ovs_header->dp_ifindex);
+ err = flow_from_nlattrs(&key, &key_len, a[OVS_FLOW_ATTR_KEY]);
+ if (err)
+ return err;
+
+ dp = get_dp(ovs_header->dp_ifindex);
+ if (!dp)
+ return -ENODEV;
+
+ table = get_table_protected(dp);
+ flow = flow_tbl_lookup(table, &key, key_len);
+ if (!flow)
+ return -ENOENT;
+
+ reply = ovs_flow_cmd_alloc_info(flow);
+ if (!reply)
+ return -ENOMEM;
+
+ flow_tbl_remove(table, flow);
+
+ err = ovs_flow_cmd_fill_info(flow, dp, reply, info->snd_pid,
+ info->snd_seq, 0, OVS_FLOW_CMD_DEL);
+ BUG_ON(err < 0);
+
+ flow_deferred_free(flow);
+
+ genl_notify(reply, genl_info_net(info), info->snd_pid,
+ dp_flow_multicast_group.id, info->nlhdr, GFP_KERNEL);
return 0;
}
-static int answer_query(struct sw_flow *flow, u32 query_flags,
- struct odp_flow __user *ufp)
-{
- struct odp_flow_stats stats;
- unsigned long int flags;
+static int ovs_flow_cmd_dump(struct sk_buff *skb, struct netlink_callback *cb)
+{
+ struct ovs_header *ovs_header = genlmsg_data(nlmsg_data(cb->nlh));
+ struct datapath *dp;
+
+ dp = get_dp(ovs_header->dp_ifindex);
+ if (!dp)
+ return -ENODEV;
+
+ for (;;) {
+ struct sw_flow *flow;
+ u32 bucket, obj;
+
+ bucket = cb->args[0];
+ obj = cb->args[1];
+ flow = flow_tbl_next(get_table_protected(dp), &bucket, &obj);
+ if (!flow)
+ break;
+
+ if (ovs_flow_cmd_fill_info(flow, dp, skb, NETLINK_CB(cb->skb).pid,
+ cb->nlh->nlmsg_seq, NLM_F_MULTI,
+ OVS_FLOW_CMD_NEW) < 0)
+ break;
+
+ cb->args[0] = bucket;
+ cb->args[1] = obj;
+ }
+ return skb->len;
+}
+
+static struct genl_ops dp_flow_genl_ops[] = {
+ { .cmd = OVS_FLOW_CMD_NEW,
+ .flags = GENL_ADMIN_PERM, /* Requires CAP_NET_ADMIN privilege. */
+ .policy = flow_policy,
+ .doit = ovs_flow_cmd_new_or_set
+ },
+ { .cmd = OVS_FLOW_CMD_DEL,
+ .flags = GENL_ADMIN_PERM, /* Requires CAP_NET_ADMIN privilege. */
+ .policy = flow_policy,
+ .doit = ovs_flow_cmd_del
+ },
+ { .cmd = OVS_FLOW_CMD_GET,
+ .flags = 0, /* OK for unprivileged users. */
+ .policy = flow_policy,
+ .doit = ovs_flow_cmd_get,
+ .dumpit = ovs_flow_cmd_dump
+ },
+ { .cmd = OVS_FLOW_CMD_SET,
+ .flags = GENL_ADMIN_PERM, /* Requires CAP_NET_ADMIN privilege. */
+ .policy = flow_policy,
+ .doit = ovs_flow_cmd_new_or_set,
+ },
+};
+
+static const struct nla_policy datapath_policy[OVS_DP_ATTR_MAX + 1] = {
+#ifdef HAVE_NLA_NUL_STRING
+ [OVS_DP_ATTR_NAME] = { .type = NLA_NUL_STRING, .len = IFNAMSIZ - 1 },
+#endif
+ [OVS_DP_ATTR_IPV4_FRAGS] = { .type = NLA_U32 },
+ [OVS_DP_ATTR_SAMPLING] = { .type = NLA_U32 },
+};
+
+static struct genl_family dp_datapath_genl_family = {
+ .id = GENL_ID_GENERATE,
+ .hdrsize = sizeof(struct ovs_header),
+ .name = OVS_DATAPATH_FAMILY,
+ .version = 1,
+ .maxattr = OVS_DP_ATTR_MAX
+};
+
+static struct genl_multicast_group dp_datapath_multicast_group = {
+ .name = OVS_DATAPATH_MCGROUP
+};
+
+static int ovs_dp_cmd_fill_info(struct datapath *dp, struct sk_buff *skb,
+ u32 pid, u32 seq, u32 flags, u8 cmd)
+{
+ struct ovs_header *ovs_header;
+ struct nlattr *nla;
+ int err;
+
+ ovs_header = genlmsg_put(skb, pid, seq, &dp_datapath_genl_family,
+ flags, cmd);
+ if (!ovs_header)
+ goto error;
+
+ ovs_header->dp_ifindex = dp->dp_ifindex;
+
+ rcu_read_lock();
+ err = nla_put_string(skb, OVS_DP_ATTR_NAME, dp_name(dp));
+ rcu_read_unlock();
+ if (err)
+ goto nla_put_failure;
+
+ nla = nla_reserve(skb, OVS_DP_ATTR_STATS, sizeof(struct ovs_dp_stats));
+ if (!nla)
+ goto nla_put_failure;
+ get_dp_stats(dp, nla_data(nla));
+
+ NLA_PUT_U32(skb, OVS_DP_ATTR_IPV4_FRAGS,
+ dp->drop_frags ? OVS_DP_FRAG_DROP : OVS_DP_FRAG_ZERO);
+
+ if (dp->sflow_probability)
+ NLA_PUT_U32(skb, OVS_DP_ATTR_SAMPLING, dp->sflow_probability);
+
+ nla = nla_nest_start(skb, OVS_DP_ATTR_MCGROUPS);
+ if (!nla)
+ goto nla_put_failure;
+ NLA_PUT_U32(skb, OVS_PACKET_CMD_MISS, packet_mc_group(dp, OVS_PACKET_CMD_MISS));
+ NLA_PUT_U32(skb, OVS_PACKET_CMD_ACTION, packet_mc_group(dp, OVS_PACKET_CMD_ACTION));
+ NLA_PUT_U32(skb, OVS_PACKET_CMD_SAMPLE, packet_mc_group(dp, OVS_PACKET_CMD_SAMPLE));
+ nla_nest_end(skb, nla);
+
+ return genlmsg_end(skb, ovs_header);
+
+nla_put_failure:
+ genlmsg_cancel(skb, ovs_header);
+error:
+ return -EMSGSIZE;
+}
+
+static struct sk_buff *ovs_dp_cmd_build_info(struct datapath *dp, u32 pid,
+ u32 seq, u8 cmd)
+{
+ struct sk_buff *skb;
+ int retval;
+
+ skb = nlmsg_new(NLMSG_DEFAULT_SIZE, GFP_KERNEL);
+ if (!skb)
+ return ERR_PTR(-ENOMEM);
+
+ retval = ovs_dp_cmd_fill_info(dp, skb, pid, seq, 0, cmd);
+ if (retval < 0) {
+ kfree_skb(skb);
+ return ERR_PTR(retval);
+ }
+ return skb;
+}
+
+static int ovs_dp_cmd_validate(struct nlattr *a[OVS_DP_ATTR_MAX + 1])
+{
+ if (a[OVS_DP_ATTR_IPV4_FRAGS]) {
+ u32 frags = nla_get_u32(a[OVS_DP_ATTR_IPV4_FRAGS]);
+
+ if (frags != OVS_DP_FRAG_ZERO && frags != OVS_DP_FRAG_DROP)
+ return -EINVAL;
+ }
+
+ return CHECK_NUL_STRING(a[OVS_DP_ATTR_NAME], IFNAMSIZ - 1);
+}
+
+/* Called with genl_mutex and optionally with RTNL lock also. */
+static struct datapath *lookup_datapath(struct ovs_header *ovs_header, struct nlattr *a[OVS_DP_ATTR_MAX + 1])
+{
+ struct datapath *dp;
+
+ if (!a[OVS_DP_ATTR_NAME])
+ dp = get_dp(ovs_header->dp_ifindex);
+ else {
+ struct vport *vport;
+
+ rcu_read_lock();
+ vport = vport_locate(nla_data(a[OVS_DP_ATTR_NAME]));
+ dp = vport && vport->port_no == OVSP_LOCAL ? vport->dp : NULL;
+ rcu_read_unlock();
+ }
+ return dp ? dp : ERR_PTR(-ENODEV);
+}
+
+/* Called with genl_mutex. */
+static void change_datapath(struct datapath *dp, struct nlattr *a[OVS_DP_ATTR_MAX + 1])
+{
+ if (a[OVS_DP_ATTR_IPV4_FRAGS])
+ dp->drop_frags = nla_get_u32(a[OVS_DP_ATTR_IPV4_FRAGS]) == OVS_DP_FRAG_DROP;
+ if (a[OVS_DP_ATTR_SAMPLING])
+ dp->sflow_probability = nla_get_u32(a[OVS_DP_ATTR_SAMPLING]);
+}
+
+static int ovs_dp_cmd_new(struct sk_buff *skb, struct genl_info *info)
+{
+ struct nlattr **a = info->attrs;
+ struct vport_parms parms;
+ struct sk_buff *reply;
+ struct datapath *dp;
+ struct vport *vport;
+ int err;
+
+ err = -EINVAL;
+ if (!a[OVS_DP_ATTR_NAME])
+ goto err;
+
+ err = ovs_dp_cmd_validate(a);
+ if (err)
+ goto err;
+
+ rtnl_lock();
+ err = -ENODEV;
+ if (!try_module_get(THIS_MODULE))
+ goto err_unlock_rtnl;
+
+ err = -ENOMEM;
+ dp = kzalloc(sizeof(*dp), GFP_KERNEL);
+ if (dp == NULL)
+ goto err_put_module;
+ INIT_LIST_HEAD(&dp->port_list);
+
+ /* Initialize kobject for bridge. This will be added as
+ * /sys/class/net/<devname>/brif later, if sysfs is enabled. */
+ dp->ifobj.kset = NULL;
+ kobject_init(&dp->ifobj, &dp_ktype);
+
+ /* Allocate table. */
+ err = -ENOMEM;
+ rcu_assign_pointer(dp->table, flow_tbl_alloc(TBL_MIN_BUCKETS));
+ if (!dp->table)
+ goto err_free_dp;
+
+ /* Set up our datapath device. */
+ parms.name = nla_data(a[OVS_DP_ATTR_NAME]);
+ parms.type = OVS_VPORT_TYPE_INTERNAL;
+ parms.options = NULL;
+ parms.dp = dp;
+ parms.port_no = OVSP_LOCAL;
+ vport = new_vport(&parms);
+ if (IS_ERR(vport)) {
+ err = PTR_ERR(vport);
+ if (err == -EBUSY)
+ err = -EEXIST;
+
+ goto err_destroy_table;
+ }
+ dp->dp_ifindex = vport_get_ifindex(vport);
+
+ dp->drop_frags = 0;
+ dp->stats_percpu = alloc_percpu(struct dp_stats_percpu);
+ if (!dp->stats_percpu) {
+ err = -ENOMEM;
+ goto err_destroy_local_port;
+ }
+
+ change_datapath(dp, a);
+
+ reply = ovs_dp_cmd_build_info(dp, info->snd_pid, info->snd_seq, OVS_DP_CMD_NEW);
+ err = PTR_ERR(reply);
+ if (IS_ERR(reply))
+ goto err_destroy_local_port;
+
+ list_add_tail(&dp->list_node, &dps);
+ dp_sysfs_add_dp(dp);
+
+ rtnl_unlock();
+
+ genl_notify(reply, genl_info_net(info), info->snd_pid,
+ dp_datapath_multicast_group.id, info->nlhdr, GFP_KERNEL);
+ return 0;
+
+err_destroy_local_port:
+ dp_detach_port(get_vport_protected(dp, OVSP_LOCAL));
+err_destroy_table:
+ flow_tbl_destroy(get_table_protected(dp));
+err_free_dp:
+ kfree(dp);
+err_put_module:
+ module_put(THIS_MODULE);
+err_unlock_rtnl:
+ rtnl_unlock();
+err:
+ return err;
+}
+
+static int ovs_dp_cmd_del(struct sk_buff *skb, struct genl_info *info)
+{
+ struct vport *vport, *next_vport;
+ struct sk_buff *reply;
+ struct datapath *dp;
+ int err;
+
+ err = ovs_dp_cmd_validate(info->attrs);
+ if (err)
+ goto exit;
+
+ rtnl_lock();
+ dp = lookup_datapath(info->userhdr, info->attrs);
+ err = PTR_ERR(dp);
+ if (IS_ERR(dp))
+ goto exit_unlock;
+
+ reply = ovs_dp_cmd_build_info(dp, info->snd_pid, info->snd_seq, OVS_DP_CMD_DEL);
+ err = PTR_ERR(reply);
+ if (IS_ERR(reply))
+ goto exit_unlock;
+
+ list_for_each_entry_safe (vport, next_vport, &dp->port_list, node)
+ if (vport->port_no != OVSP_LOCAL)
+ dp_detach_port(vport);
+
+ dp_sysfs_del_dp(dp);
+ list_del(&dp->list_node);
+ dp_detach_port(get_vport_protected(dp, OVSP_LOCAL));
+
+ /* rtnl_unlock() will wait until all the references to devices that
+ * are pending unregistration have been dropped. We do it here to
+ * ensure that any internal devices (which contain DP pointers) are
+ * fully destroyed before freeing the datapath.
+ */
+ rtnl_unlock();
+
+ call_rcu(&dp->rcu, destroy_dp_rcu);
+ module_put(THIS_MODULE);
- spin_lock_irqsave(&flow->lock, flags);
- get_stats(flow, &stats);
+ genl_notify(reply, genl_info_net(info), info->snd_pid,
+ dp_datapath_multicast_group.id, info->nlhdr, GFP_KERNEL);
- if (query_flags & ODPFF_ZERO_TCP_FLAGS) {
- flow->tcp_flags = 0;
- }
- spin_unlock_irqrestore(&flow->lock, flags);
+ return 0;
- if (__copy_to_user(&ufp->stats, &stats, sizeof(struct odp_flow_stats)))
- return -EFAULT;
- return put_actions(flow, ufp);
+exit_unlock:
+ rtnl_unlock();
+exit:
+ return err;
}
-static int del_flow(struct datapath *dp, struct odp_flow __user *ufp)
+static int ovs_dp_cmd_set(struct sk_buff *skb, struct genl_info *info)
{
- struct dp_table *table = rcu_dereference(dp->table);
- struct odp_flow uf;
- struct sw_flow *flow;
- int error;
+ struct sk_buff *reply;
+ struct datapath *dp;
+ int err;
- error = -EFAULT;
- if (copy_from_user(&uf, ufp, sizeof uf))
- goto error;
- memset(uf.key.reserved, 0, sizeof uf.key.reserved);
+ err = ovs_dp_cmd_validate(info->attrs);
+ if (err)
+ return err;
- flow = dp_table_lookup(table, &uf.key);
- error = -ENOENT;
- if (!flow)
- goto error;
+ dp = lookup_datapath(info->userhdr, info->attrs);
+ if (IS_ERR(dp))
+ return PTR_ERR(dp);
- /* XXX redundant lookup */
- error = dp_table_delete(table, flow);
- if (error)
- goto error;
+ change_datapath(dp, info->attrs);
- /* XXX These statistics might lose a few packets, since other CPUs can
- * 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--;
- error = answer_query(flow, 0, ufp);
- flow_deferred_free(flow);
+ reply = ovs_dp_cmd_build_info(dp, info->snd_pid, info->snd_seq, OVS_DP_CMD_NEW);
+ if (IS_ERR(reply)) {
+ err = PTR_ERR(reply);
+ netlink_set_err(INIT_NET_GENL_SOCK, 0,
+ dp_datapath_multicast_group.id, err);
+ return 0;
+ }
-error:
- return error;
+ genl_notify(reply, genl_info_net(info), info->snd_pid,
+ dp_datapath_multicast_group.id, info->nlhdr, GFP_KERNEL);
+ return 0;
}
-static int query_flows(struct datapath *dp, const struct odp_flowvec *flowvec)
+static int ovs_dp_cmd_get(struct sk_buff *skb, struct genl_info *info)
{
- struct dp_table *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;
- int error;
-
- if (__copy_from_user(&uf, ufp, sizeof uf))
- return -EFAULT;
- memset(uf.key.reserved, 0, sizeof uf.key.reserved);
-
- flow = dp_table_lookup(table, &uf.key);
- if (!flow)
- error = __put_user(ENOENT, &ufp->stats.error);
- else
- error = answer_query(flow, uf.flags, ufp);
- if (error)
- return -EFAULT;
- }
- return flowvec->n_flows;
-}
+ struct sk_buff *reply;
+ struct datapath *dp;
+ int err;
-struct list_flows_cbdata {
- struct odp_flow __user *uflows;
- int n_flows;
- int listed_flows;
-};
+ err = ovs_dp_cmd_validate(info->attrs);
+ if (err)
+ return err;
-static int list_flow(struct sw_flow *flow, void *cbdata_)
-{
- struct list_flows_cbdata *cbdata = cbdata_;
- struct odp_flow __user *ufp = &cbdata->uflows[cbdata->listed_flows++];
- int error;
+ dp = lookup_datapath(info->userhdr, info->attrs);
+ if (IS_ERR(dp))
+ return PTR_ERR(dp);
- if (__copy_to_user(&ufp->key, &flow->key, sizeof flow->key))
- return -EFAULT;
- error = answer_query(flow, 0, ufp);
- if (error)
- return error;
+ reply = ovs_dp_cmd_build_info(dp, info->snd_pid, info->snd_seq, OVS_DP_CMD_NEW);
+ if (IS_ERR(reply))
+ return PTR_ERR(reply);
- if (cbdata->listed_flows >= cbdata->n_flows)
- return cbdata->listed_flows;
- return 0;
+ return genlmsg_reply(reply, info);
}
-static int list_flows(struct datapath *dp, const struct odp_flowvec *flowvec)
+static int ovs_dp_cmd_dump(struct sk_buff *skb, struct netlink_callback *cb)
{
- struct list_flows_cbdata cbdata;
- int error;
+ struct datapath *dp;
+ int skip = cb->args[0];
+ int i = 0;
- if (!flowvec->n_flows)
- return 0;
+ list_for_each_entry (dp, &dps, list_node) {
+ if (i < skip)
+ continue;
+ if (ovs_dp_cmd_fill_info(dp, skb, NETLINK_CB(cb->skb).pid,
+ cb->nlh->nlmsg_seq, NLM_F_MULTI,
+ OVS_DP_CMD_NEW) < 0)
+ break;
+ i++;
+ }
- 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);
- return error ? error : cbdata.listed_flows;
-}
+ cb->args[0] = i;
-static int do_flowvec_ioctl(struct datapath *dp, unsigned long argp,
- int (*function)(struct datapath *,
- const struct odp_flowvec *))
-{
- struct odp_flowvec __user *uflowvec;
- struct odp_flowvec flowvec;
- int retval;
+ return skb->len;
+}
- uflowvec = (struct odp_flowvec __user *)argp;
- if (!access_ok(VERIFY_WRITE, uflowvec, sizeof *uflowvec) ||
- copy_from_user(&flowvec, uflowvec, sizeof flowvec))
- return -EFAULT;
+static struct genl_ops dp_datapath_genl_ops[] = {
+ { .cmd = OVS_DP_CMD_NEW,
+ .flags = GENL_ADMIN_PERM, /* Requires CAP_NET_ADMIN privilege. */
+ .policy = datapath_policy,
+ .doit = ovs_dp_cmd_new
+ },
+ { .cmd = OVS_DP_CMD_DEL,
+ .flags = GENL_ADMIN_PERM, /* Requires CAP_NET_ADMIN privilege. */
+ .policy = datapath_policy,
+ .doit = ovs_dp_cmd_del
+ },
+ { .cmd = OVS_DP_CMD_GET,
+ .flags = 0, /* OK for unprivileged users. */
+ .policy = datapath_policy,
+ .doit = ovs_dp_cmd_get,
+ .dumpit = ovs_dp_cmd_dump
+ },
+ { .cmd = OVS_DP_CMD_SET,
+ .flags = GENL_ADMIN_PERM, /* Requires CAP_NET_ADMIN privilege. */
+ .policy = datapath_policy,
+ .doit = ovs_dp_cmd_set,
+ },
+};
- if (flowvec.n_flows > INT_MAX / sizeof(struct odp_flow))
- return -EINVAL;
+static const struct nla_policy vport_policy[OVS_VPORT_ATTR_MAX + 1] = {
+#ifdef HAVE_NLA_NUL_STRING
+ [OVS_VPORT_ATTR_NAME] = { .type = NLA_NUL_STRING, .len = IFNAMSIZ - 1 },
+ [OVS_VPORT_ATTR_PORT_NO] = { .type = NLA_U32 },
+ [OVS_VPORT_ATTR_TYPE] = { .type = NLA_U32 },
+ [OVS_VPORT_ATTR_STATS] = { .len = sizeof(struct ovs_vport_stats) },
+ [OVS_VPORT_ATTR_ADDRESS] = { .len = ETH_ALEN },
+#else
+ [OVS_VPORT_ATTR_STATS] = { .minlen = sizeof(struct ovs_vport_stats) },
+ [OVS_VPORT_ATTR_ADDRESS] = { .minlen = ETH_ALEN },
+#endif
+ [OVS_VPORT_ATTR_OPTIONS] = { .type = NLA_NESTED },
+};
- if (!access_ok(VERIFY_WRITE, flowvec.flows,
- flowvec.n_flows * sizeof(struct odp_flow)))
- return -EFAULT;
+static struct genl_family dp_vport_genl_family = {
+ .id = GENL_ID_GENERATE,
+ .hdrsize = sizeof(struct ovs_header),
+ .name = OVS_VPORT_FAMILY,
+ .version = 1,
+ .maxattr = OVS_VPORT_ATTR_MAX
+};
- retval = function(dp, &flowvec);
- return (retval < 0 ? retval
- : retval == flowvec.n_flows ? 0
- : __put_user(retval, &uflowvec->n_flows));
-}
+struct genl_multicast_group dp_vport_multicast_group = {
+ .name = OVS_VPORT_MCGROUP
+};
-static int do_execute(struct datapath *dp, const struct odp_execute *executep)
+/* Called with RTNL lock or RCU read lock. */
+static int ovs_vport_cmd_fill_info(struct vport *vport, struct sk_buff *skb,
+ u32 pid, u32 seq, u32 flags, u8 cmd)
{
- struct odp_execute execute;
- struct odp_flow_key key;
- struct sk_buff *skb;
- struct sw_flow_actions *actions;
- struct ethhdr *eth;
+ struct ovs_header *ovs_header;
+ struct nlattr *nla;
+ int ifindex;
int err;
- err = -EFAULT;
- if (copy_from_user(&execute, executep, sizeof execute))
- goto error;
-
- err = -EINVAL;
- if (execute.length < ETH_HLEN || execute.length > 65535)
- goto error;
+ ovs_header = genlmsg_put(skb, pid, seq, &dp_vport_genl_family,
+ flags, cmd);
+ if (!ovs_header)
+ return -EMSGSIZE;
- err = -ENOMEM;
- actions = flow_actions_alloc(execute.n_actions);
- if (!actions)
- goto error;
+ ovs_header->dp_ifindex = vport->dp->dp_ifindex;
- err = -EFAULT;
- if (copy_from_user(actions->actions, execute.actions,
- execute.n_actions * sizeof *execute.actions))
- goto error_free_actions;
+ NLA_PUT_U32(skb, OVS_VPORT_ATTR_PORT_NO, vport->port_no);
+ NLA_PUT_U32(skb, OVS_VPORT_ATTR_TYPE, vport_get_type(vport));
+ NLA_PUT_STRING(skb, OVS_VPORT_ATTR_NAME, vport_get_name(vport));
- err = validate_actions(actions);
- if (err)
- goto error_free_actions;
+ nla = nla_reserve(skb, OVS_VPORT_ATTR_STATS, sizeof(struct ovs_vport_stats));
+ if (!nla)
+ goto nla_put_failure;
- err = -ENOMEM;
- 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;
- }
+ vport_get_stats(vport, nla_data(nla));
- err = -EFAULT;
- if (copy_from_user(skb_put(skb, execute.length), execute.data,
- execute.length))
- goto error_free_skb;
+ NLA_PUT(skb, OVS_VPORT_ATTR_ADDRESS, ETH_ALEN, vport_get_addr(vport));
- skb_reset_mac_header(skb);
- eth = eth_hdr(skb);
+ err = vport_get_options(vport, skb);
+ if (err == -EMSGSIZE)
+ goto error;
- /* Normally, setting the skb 'protocol' field would be handled by a
- * call to eth_type_trans(), but it assumes there's a sending
- * device, which we may not have. */
- if (ntohs(eth->h_proto) >= 1536)
- skb->protocol = eth->h_proto;
- else
- skb->protocol = htons(ETH_P_802_2);
+ ifindex = vport_get_ifindex(vport);
+ if (ifindex > 0)
+ NLA_PUT_U32(skb, OVS_VPORT_ATTR_IFINDEX, ifindex);
- flow_extract(skb, execute.in_port, &key);
- err = execute_actions(dp, skb, &key, actions->actions,
- actions->n_actions, GFP_KERNEL);
- kfree(actions);
- return err;
+ return genlmsg_end(skb, ovs_header);
-error_free_skb:
- kfree_skb(skb);
-error_free_actions:
- kfree(actions);
+nla_put_failure:
+ err = -EMSGSIZE;
error:
+ genlmsg_cancel(skb, ovs_header);
return err;
}
-static int get_dp_stats(struct datapath *dp, struct odp_stats __user *statsp)
+/* Called with RTNL lock or RCU read lock. */
+struct sk_buff *ovs_vport_cmd_build_info(struct vport *vport, u32 pid,
+ u32 seq, u8 cmd)
{
- struct odp_stats stats;
- int i;
+ struct sk_buff *skb;
+ int retval;
- stats.n_flows = dp->n_flows;
- stats.cur_capacity = rcu_dereference(dp->table)->n_buckets;
- stats.max_capacity = DP_MAX_BUCKETS;
- stats.n_ports = dp->n_ports;
- stats.max_ports = DP_MAX_PORTS;
- stats.max_groups = DP_MAX_GROUPS;
- stats.n_frags = stats.n_hit = stats.n_missed = stats.n_lost = 0;
- for_each_possible_cpu(i) {
- const struct dp_stats_percpu *s;
- s = percpu_ptr(dp->stats_percpu, i);
- stats.n_frags += s->n_frags;
- stats.n_hit += s->n_hit;
- stats.n_missed += s->n_missed;
- stats.n_lost += s->n_lost;
+ skb = nlmsg_new(NLMSG_DEFAULT_SIZE, GFP_ATOMIC);
+ if (!skb)
+ return ERR_PTR(-ENOMEM);
+
+ retval = ovs_vport_cmd_fill_info(vport, skb, pid, seq, 0, cmd);
+ if (retval < 0) {
+ kfree_skb(skb);
+ return ERR_PTR(retval);
}
- stats.max_miss_queue = DP_MAX_QUEUE_LEN;
- stats.max_action_queue = DP_MAX_QUEUE_LEN;
- return copy_to_user(statsp, &stats, sizeof stats) ? -EFAULT : 0;
+ return skb;
}
-/* MTU of the dp pseudo-device: ETH_DATA_LEN or the minimum of the ports */
-int dp_min_mtu(const struct datapath *dp)
+static int ovs_vport_cmd_validate(struct nlattr *a[OVS_VPORT_ATTR_MAX + 1])
{
- struct net_bridge_port *p;
- int mtu = 0;
+ return CHECK_NUL_STRING(a[OVS_VPORT_ATTR_NAME], IFNAMSIZ - 1);
+}
- ASSERT_RTNL();
+/* Called with RTNL lock or RCU read lock. */
+static struct vport *lookup_vport(struct ovs_header *ovs_header,
+ struct nlattr *a[OVS_VPORT_ATTR_MAX + 1])
+{
+ struct datapath *dp;
+ struct vport *vport;
+
+ if (a[OVS_VPORT_ATTR_NAME]) {
+ vport = vport_locate(nla_data(a[OVS_VPORT_ATTR_NAME]));
+ if (!vport)
+ return ERR_PTR(-ENODEV);
+ return vport;
+ } else if (a[OVS_VPORT_ATTR_PORT_NO]) {
+ u32 port_no = nla_get_u32(a[OVS_VPORT_ATTR_PORT_NO]);
+
+ if (port_no >= DP_MAX_PORTS)
+ return ERR_PTR(-EFBIG);
+
+ dp = get_dp(ovs_header->dp_ifindex);
+ if (!dp)
+ return ERR_PTR(-ENODEV);
+
+ vport = get_vport_protected(dp, port_no);
+ if (!vport)
+ return ERR_PTR(-ENOENT);
+ return vport;
+ } else
+ return ERR_PTR(-EINVAL);
+}
- list_for_each_entry_rcu (p, &dp->port_list, node) {
- struct net_device *dev = p->dev;
+/* Called with RTNL lock. */
+static int change_vport(struct vport *vport, struct nlattr *a[OVS_VPORT_ATTR_MAX + 1])
+{
+ int err = 0;
- /* Skip any internal ports, since that's what we're trying to
- * set. */
- if (is_dp_dev(dev))
- continue;
+ if (a[OVS_VPORT_ATTR_STATS])
+ vport_set_stats(vport, nla_data(a[OVS_VPORT_ATTR_STATS]));
- if (!mtu || dev->mtu < mtu)
- mtu = dev->mtu;
- }
+ if (a[OVS_VPORT_ATTR_ADDRESS])
+ err = vport_set_addr(vport, nla_data(a[OVS_VPORT_ATTR_ADDRESS]));
- return mtu ? mtu : ETH_DATA_LEN;
+ return err;
}
-/* Sets the MTU of all datapath devices to the minimum of the ports. 'dev'
- * is the device whose MTU may have changed. Must be called with RTNL lock
- * and dp_mutex. */
-void set_dp_devs_mtu(const struct datapath *dp, struct net_device *dev)
+static int ovs_vport_cmd_new(struct sk_buff *skb, struct genl_info *info)
{
- struct net_bridge_port *p;
- int mtu;
-
- ASSERT_RTNL();
-
- if (is_dp_dev(dev))
- return;
+ struct nlattr **a = info->attrs;
+ struct ovs_header *ovs_header = info->userhdr;
+ struct vport_parms parms;
+ struct sk_buff *reply;
+ struct vport *vport;
+ struct datapath *dp;
+ u32 port_no;
+ int err;
- mtu = dp_min_mtu(dp);
+ err = -EINVAL;
+ if (!a[OVS_VPORT_ATTR_NAME] || !a[OVS_VPORT_ATTR_TYPE])
+ goto exit;
- list_for_each_entry_rcu (p, &dp->port_list, node) {
- struct net_device *br_dev = p->dev;
+ err = ovs_vport_cmd_validate(a);
+ if (err)
+ goto exit;
- if (is_dp_dev(br_dev))
- dev_set_mtu(br_dev, mtu);
- }
-}
+ rtnl_lock();
+ dp = get_dp(ovs_header->dp_ifindex);
+ err = -ENODEV;
+ if (!dp)
+ goto exit_unlock;
-static int
-put_port(const struct net_bridge_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);
- op.port = p->port_no;
- op.flags = is_dp_dev(p->dev) ? ODP_PORT_INTERNAL : 0;
- return copy_to_user(uop, &op, sizeof op) ? -EFAULT : 0;
-}
+ if (a[OVS_VPORT_ATTR_PORT_NO]) {
+ port_no = nla_get_u32(a[OVS_VPORT_ATTR_PORT_NO]);
-static int
-query_port(struct datapath *dp, struct odp_port __user *uport)
-{
- struct odp_port port;
-
- 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;
-
- port.devname[IFNAMSIZ - 1] = '\0';
-
- dev = dev_get_by_name(&init_net, port.devname);
- if (!dev)
- return -ENODEV;
-
- 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];
- }
- err = p && p->dp == dp ? put_port(p, uport) : -ENOENT;
- dev_put(dev);
+ err = -EFBIG;
+ if (port_no >= DP_MAX_PORTS)
+ goto exit_unlock;
- return err;
+ vport = get_vport_protected(dp, port_no);
+ err = -EBUSY;
+ if (vport)
+ goto exit_unlock;
} else {
- if (port.port >= DP_MAX_PORTS)
- return -EINVAL;
- if (!dp->ports[port.port])
- return -ENOENT;
- 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;
- int idx;
-
- if (copy_from_user(&pv, pvp, sizeof pv))
- return -EFAULT;
-
- idx = 0;
- if (pv.n_ports) {
- list_for_each_entry_rcu (p, &dp->port_list, node) {
- if (put_port(p, &pv.ports[idx]))
- return -EFAULT;
- if (idx++ >= pv.n_ports)
+ for (port_no = 1; ; port_no++) {
+ if (port_no >= DP_MAX_PORTS) {
+ err = -EFBIG;
+ goto exit_unlock;
+ }
+ vport = get_vport_protected(dp, port_no);
+ if (!vport)
break;
}
}
- return put_user(dp->n_ports, &pvp->n_ports);
-}
-/* RCU callback for freeing a dp_port_group */
-static void free_port_group(struct rcu_head *rcu)
-{
- struct dp_port_group *g = container_of(rcu, struct dp_port_group, rcu);
- kfree(g);
+ parms.name = nla_data(a[OVS_VPORT_ATTR_NAME]);
+ parms.type = nla_get_u32(a[OVS_VPORT_ATTR_TYPE]);
+ parms.options = a[OVS_VPORT_ATTR_OPTIONS];
+ parms.dp = dp;
+ parms.port_no = port_no;
+
+ vport = new_vport(&parms);
+ err = PTR_ERR(vport);
+ if (IS_ERR(vport))
+ goto exit_unlock;
+
+ dp_sysfs_add_if(vport);
+
+ err = change_vport(vport, a);
+ if (!err) {
+ reply = ovs_vport_cmd_build_info(vport, info->snd_pid,
+ info->snd_seq, OVS_VPORT_CMD_NEW);
+ if (IS_ERR(reply))
+ err = PTR_ERR(reply);
+ }
+ if (err) {
+ dp_detach_port(vport);
+ goto exit_unlock;
+ }
+ genl_notify(reply, genl_info_net(info), info->snd_pid,
+ dp_vport_multicast_group.id, info->nlhdr, GFP_KERNEL);
+
+
+exit_unlock:
+ rtnl_unlock();
+exit:
+ return err;
}
-static int
-set_port_group(struct datapath *dp, const struct odp_port_group __user *upg)
+static int ovs_vport_cmd_set(struct sk_buff *skb, struct genl_info *info)
{
- struct odp_port_group pg;
- struct dp_port_group *new_group, *old_group;
- int error;
-
- error = -EFAULT;
- if (copy_from_user(&pg, upg, sizeof pg))
- goto error;
+ struct nlattr **a = info->attrs;
+ struct sk_buff *reply;
+ struct vport *vport;
+ int err;
- error = -EINVAL;
- if (pg.n_ports > DP_MAX_PORTS || pg.group >= DP_MAX_GROUPS)
- goto error;
+ err = ovs_vport_cmd_validate(a);
+ if (err)
+ goto exit;
- error = -ENOMEM;
- new_group = kmalloc(sizeof *new_group + sizeof(u16) * pg.n_ports,
- GFP_KERNEL);
- if (!new_group)
- goto error;
+ rtnl_lock();
+ vport = lookup_vport(info->userhdr, a);
+ err = PTR_ERR(vport);
+ if (IS_ERR(vport))
+ goto exit_unlock;
- new_group->n_ports = pg.n_ports;
- error = -EFAULT;
- if (copy_from_user(new_group->ports, pg.ports,
- sizeof(u16) * pg.n_ports))
- goto error_free;
+ err = 0;
+ if (a[OVS_VPORT_ATTR_OPTIONS])
+ err = vport_set_options(vport, a[OVS_VPORT_ATTR_OPTIONS]);
+ if (!err)
+ err = change_vport(vport, a);
+
+ reply = ovs_vport_cmd_build_info(vport, info->snd_pid, info->snd_seq,
+ OVS_VPORT_CMD_NEW);
+ if (IS_ERR(reply)) {
+ err = PTR_ERR(reply);
+ netlink_set_err(INIT_NET_GENL_SOCK, 0,
+ dp_vport_multicast_group.id, err);
+ return 0;
+ }
- old_group = rcu_dereference(dp->groups[pg.group]);
- rcu_assign_pointer(dp->groups[pg.group], new_group);
- if (old_group)
- call_rcu(&old_group->rcu, free_port_group);
- return 0;
+ genl_notify(reply, genl_info_net(info), info->snd_pid,
+ dp_vport_multicast_group.id, info->nlhdr, GFP_KERNEL);
-error_free:
- kfree(new_group);
-error:
- return error;
+exit_unlock:
+ rtnl_unlock();
+exit:
+ return err;
}
-static int
-get_port_group(struct datapath *dp, struct odp_port_group *upg)
+static int ovs_vport_cmd_del(struct sk_buff *skb, struct genl_info *info)
{
- struct odp_port_group pg;
- struct dp_port_group *g;
- u16 n_copy;
+ struct nlattr **a = info->attrs;
+ struct sk_buff *reply;
+ struct vport *vport;
+ int err;
- if (copy_from_user(&pg, upg, sizeof pg))
- return -EFAULT;
+ err = ovs_vport_cmd_validate(a);
+ if (err)
+ goto exit;
- if (pg.group >= DP_MAX_GROUPS)
- return -EINVAL;
+ rtnl_lock();
+ vport = lookup_vport(info->userhdr, a);
+ err = PTR_ERR(vport);
+ if (IS_ERR(vport))
+ goto exit_unlock;
- g = dp->groups[pg.group];
- n_copy = g ? min_t(int, g->n_ports, pg.n_ports) : 0;
- if (n_copy && copy_to_user(pg.ports, g->ports, n_copy * sizeof(u16)))
- return -EFAULT;
+ if (vport->port_no == OVSP_LOCAL) {
+ err = -EINVAL;
+ goto exit_unlock;
+ }
- if (put_user(g ? g->n_ports : 0, &upg->n_ports))
- return -EFAULT;
+ reply = ovs_vport_cmd_build_info(vport, info->snd_pid, info->snd_seq,
+ OVS_VPORT_CMD_DEL);
+ err = PTR_ERR(reply);
+ if (IS_ERR(reply))
+ goto exit_unlock;
- return 0;
-}
+ dp_detach_port(vport);
-static int get_listen_mask(const struct file *f)
-{
- return (long)f->private_data;
-}
+ genl_notify(reply, genl_info_net(info), info->snd_pid,
+ dp_vport_multicast_group.id, info->nlhdr, GFP_KERNEL);
-static void set_listen_mask(struct file *f, int listen_mask)
-{
- f->private_data = (void*)(long)listen_mask;
+exit_unlock:
+ rtnl_unlock();
+exit:
+ return err;
}
-static long openvswitch_ioctl(struct file *f, unsigned int cmd,
- unsigned long argp)
+static int ovs_vport_cmd_get(struct sk_buff *skb, struct genl_info *info)
{
- int dp_idx = iminor(f->f_dentry->d_inode);
- struct datapath *dp;
- int drop_frags, listeners, port_no;
- unsigned int sflow_probability;
+ struct nlattr **a = info->attrs;
+ struct ovs_header *ovs_header = info->userhdr;
+ struct sk_buff *reply;
+ struct vport *vport;
int err;
- /* Handle commands with special locking requirements up front. */
- switch (cmd) {
- case ODP_DP_CREATE:
- err = create_dp(dp_idx, (char __user *)argp);
- goto exit;
-
- case ODP_DP_DESTROY:
- err = destroy_dp(dp_idx);
- goto exit;
-
- case ODP_PORT_ADD:
- err = add_port(dp_idx, (struct odp_port __user *)argp);
- goto exit;
-
- case ODP_PORT_DEL:
- err = get_user(port_no, (int __user *)argp);
- if (!err)
- err = del_port(dp_idx, port_no);
- goto exit;
- }
-
- dp = get_dp_locked(dp_idx);
- err = -ENODEV;
- if (!dp)
+ err = ovs_vport_cmd_validate(a);
+ if (err)
goto exit;
- switch (cmd) {
- case ODP_DP_STATS:
- err = get_dp_stats(dp, (struct odp_stats __user *)argp);
- break;
+ rcu_read_lock();
+ vport = lookup_vport(ovs_header, a);
+ err = PTR_ERR(vport);
+ if (IS_ERR(vport))
+ goto exit_unlock;
- case ODP_GET_DROP_FRAGS:
- err = put_user(dp->drop_frags, (int __user *)argp);
- break;
+ reply = ovs_vport_cmd_build_info(vport, info->snd_pid, info->snd_seq,
+ OVS_VPORT_CMD_NEW);
+ err = PTR_ERR(reply);
+ if (IS_ERR(reply))
+ goto exit_unlock;
- case ODP_SET_DROP_FRAGS:
- err = get_user(drop_frags, (int __user *)argp);
- if (err)
- break;
- err = -EINVAL;
- if (drop_frags != 0 && drop_frags != 1)
- break;
- dp->drop_frags = drop_frags;
- err = 0;
- break;
+ rcu_read_unlock();
- case ODP_GET_LISTEN_MASK:
- err = put_user(get_listen_mask(f), (int __user *)argp);
- break;
+ return genlmsg_reply(reply, info);
- case ODP_SET_LISTEN_MASK:
- err = get_user(listeners, (int __user *)argp);
- if (err)
- break;
- err = -EINVAL;
- if (listeners & ~ODPL_ALL)
- break;
- err = 0;
- set_listen_mask(f, listeners);
- break;
-
- case ODP_GET_SFLOW_PROBABILITY:
- err = put_user(dp->sflow_probability, (unsigned int __user *)argp);
- break;
-
- case ODP_SET_SFLOW_PROBABILITY:
- err = get_user(sflow_probability, (unsigned int __user *)argp);
- if (!err)
- dp->sflow_probability = sflow_probability;
- break;
-
- case ODP_PORT_QUERY:
- err = query_port(dp, (struct odp_port __user *)argp);
- break;
-
- case ODP_PORT_LIST:
- err = list_ports(dp, (struct odp_portvec __user *)argp);
- break;
-
- case ODP_PORT_GROUP_SET:
- err = set_port_group(dp, (struct odp_port_group __user *)argp);
- break;
-
- case ODP_PORT_GROUP_GET:
- err = get_port_group(dp, (struct odp_port_group __user *)argp);
- break;
-
- case ODP_FLOW_FLUSH:
- err = flush_flows(dp);
- break;
-
- case ODP_FLOW_PUT:
- err = put_flow(dp, (struct odp_flow_put __user *)argp);
- break;
-
- case ODP_FLOW_DEL:
- err = del_flow(dp, (struct odp_flow __user *)argp);
- break;
-
- case ODP_FLOW_GET:
- err = do_flowvec_ioctl(dp, argp, query_flows);
- break;
-
- case ODP_FLOW_LIST:
- err = do_flowvec_ioctl(dp, argp, list_flows);
- break;
-
- case ODP_EXECUTE:
- err = do_execute(dp, (struct odp_execute __user *)argp);
- break;
-
- default:
- err = -ENOIOCTLCMD;
- break;
- }
- mutex_unlock(&dp->mutex);
+exit_unlock:
+ rcu_read_unlock();
exit:
return err;
}
-static int dp_has_packet_of_interest(struct datapath *dp, int listeners)
-{
- int i;
- for (i = 0; i < DP_N_QUEUES; i++) {
- if (listeners & (1 << i) && !skb_queue_empty(&dp->queues[i]))
- return 1;
- }
- return 0;
-}
-
-ssize_t openvswitch_read(struct file *f, char __user *buf, size_t nbytes,
- loff_t *ppos)
+static int ovs_vport_cmd_dump(struct sk_buff *skb, struct netlink_callback *cb)
{
- /* XXX is there sufficient synchronization here? */
- int listeners = get_listen_mask(f);
- int dp_idx = iminor(f->f_dentry->d_inode);
- struct datapath *dp = get_dp(dp_idx);
- struct sk_buff *skb;
- struct iovec __user iov;
- size_t copy_bytes;
+ struct ovs_header *ovs_header = genlmsg_data(nlmsg_data(cb->nlh));
+ struct datapath *dp;
+ u32 port_no;
int retval;
+ dp = get_dp(ovs_header->dp_ifindex);
if (!dp)
return -ENODEV;
- if (nbytes == 0 || !listeners)
- return 0;
-
- for (;;) {
- int i;
-
- for (i = 0; i < DP_N_QUEUES; i++) {
- if (listeners & (1 << i)) {
- skb = skb_dequeue(&dp->queues[i]);
- if (skb)
- goto success;
- }
- }
+ rcu_read_lock();
+ for (port_no = cb->args[0]; port_no < DP_MAX_PORTS; port_no++) {
+ struct vport *vport;
- if (f->f_flags & O_NONBLOCK) {
- retval = -EAGAIN;
- goto error;
- }
-
- wait_event_interruptible(dp->waitqueue,
- dp_has_packet_of_interest(dp,
- listeners));
+ vport = get_vport_protected(dp, port_no);
+ if (!vport)
+ continue;
- if (signal_pending(current)) {
- retval = -ERESTARTSYS;
- goto error;
- }
+ if (ovs_vport_cmd_fill_info(vport, skb, NETLINK_CB(cb->skb).pid,
+ cb->nlh->nlmsg_seq, NLM_F_MULTI,
+ OVS_VPORT_CMD_NEW) < 0)
+ break;
}
-success:
- copy_bytes = min_t(size_t, skb->len, nbytes);
- iov.iov_base = buf;
- iov.iov_len = copy_bytes;
- retval = skb_copy_datagram_iovec(skb, 0, &iov, iov.iov_len);
- if (!retval)
- retval = copy_bytes;
- kfree_skb(skb);
+ rcu_read_unlock();
-error:
- return retval;
-}
+ cb->args[0] = port_no;
+ retval = skb->len;
-static unsigned int openvswitch_poll(struct file *file, poll_table *wait)
-{
- /* XXX is there sufficient synchronization here? */
- int dp_idx = iminor(file->f_dentry->d_inode);
- struct datapath *dp = get_dp(dp_idx);
- unsigned int mask;
-
- if (dp) {
- mask = 0;
- poll_wait(file, &dp->waitqueue, wait);
- if (dp_has_packet_of_interest(dp, get_listen_mask(file)))
- mask |= POLLIN | POLLRDNORM;
- } else {
- mask = POLLIN | POLLRDNORM | POLLHUP;
- }
- return mask;
+ return retval;
}
-struct file_operations openvswitch_fops = {
- /* XXX .aio_read = openvswitch_aio_read, */
- .read = openvswitch_read,
- .poll = openvswitch_poll,
- .unlocked_ioctl = openvswitch_ioctl,
- /* XXX .fasync = openvswitch_fasync, */
+static struct genl_ops dp_vport_genl_ops[] = {
+ { .cmd = OVS_VPORT_CMD_NEW,
+ .flags = GENL_ADMIN_PERM, /* Requires CAP_NET_ADMIN privilege. */
+ .policy = vport_policy,
+ .doit = ovs_vport_cmd_new
+ },
+ { .cmd = OVS_VPORT_CMD_DEL,
+ .flags = GENL_ADMIN_PERM, /* Requires CAP_NET_ADMIN privilege. */
+ .policy = vport_policy,
+ .doit = ovs_vport_cmd_del
+ },
+ { .cmd = OVS_VPORT_CMD_GET,
+ .flags = 0, /* OK for unprivileged users. */
+ .policy = vport_policy,
+ .doit = ovs_vport_cmd_get,
+ .dumpit = ovs_vport_cmd_dump
+ },
+ { .cmd = OVS_VPORT_CMD_SET,
+ .flags = GENL_ADMIN_PERM, /* Requires CAP_NET_ADMIN privilege. */
+ .policy = vport_policy,
+ .doit = ovs_vport_cmd_set,
+ },
};
-static int major;
+struct genl_family_and_ops {
+ struct genl_family *family;
+ struct genl_ops *ops;
+ int n_ops;
+ struct genl_multicast_group *group;
+};
-#if LINUX_VERSION_CODE < KERNEL_VERSION(2,6,27)
-static struct llc_sap *dp_stp_sap;
+static const struct genl_family_and_ops dp_genl_families[] = {
+ { &dp_datapath_genl_family,
+ dp_datapath_genl_ops, ARRAY_SIZE(dp_datapath_genl_ops),
+ &dp_datapath_multicast_group },
+ { &dp_vport_genl_family,
+ dp_vport_genl_ops, ARRAY_SIZE(dp_vport_genl_ops),
+ &dp_vport_multicast_group },
+ { &dp_flow_genl_family,
+ dp_flow_genl_ops, ARRAY_SIZE(dp_flow_genl_ops),
+ &dp_flow_multicast_group },
+ { &dp_packet_genl_family,
+ dp_packet_genl_ops, ARRAY_SIZE(dp_packet_genl_ops),
+ NULL },
+};
-static int dp_stp_rcv(struct sk_buff *skb, struct net_device *dev,
- struct packet_type *pt, struct net_device *orig_dev)
+static void dp_unregister_genl(int n_families)
{
- /* 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;
-}
+ int i;
-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;
+ for (i = 0; i < n_families; i++)
+ genl_unregister_family(dp_genl_families[i].family);
}
-static void dp_avoid_bridge_exit(void)
+static int dp_register_genl(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;
+ int n_registered;
+ int err;
+ int i;
+
+ n_registered = 0;
+ for (i = 0; i < ARRAY_SIZE(dp_genl_families); i++) {
+ const struct genl_family_and_ops *f = &dp_genl_families[i];
+
+ err = genl_register_family_with_ops(f->family, f->ops,
+ f->n_ops);
+ if (err)
+ goto error;
+ n_registered++;
+
+ if (f->group) {
+ err = genl_register_mc_group(f->family, f->group);
+ if (err)
+ goto error;
+ }
}
+
+ err = packet_register_mc_groups();
+ if (err)
+ goto error;
return 0;
-}
-static void dp_avoid_bridge_exit(void)
-{
- /* Nothing to do. */
+error:
+ dp_unregister_genl(n_registered);
+ return err;
}
-#endif /* Linux 2.6.27 or later */
static int __init dp_init(void)
{
+ struct sk_buff *dummy_skb;
int err;
+ BUILD_BUG_ON(sizeof(struct ovs_skb_cb) > sizeof(dummy_skb->cb));
+
printk("Open vSwitch %s, built "__DATE__" "__TIME__"\n", VERSION BUILDNR);
- err = dp_avoid_bridge_init();
+ err = tnl_init();
if (err)
- return err;
+ goto error;
err = flow_init();
if (err)
- goto error;
+ goto error_tnl_exit;
- err = register_netdevice_notifier(&dp_device_notifier);
+ err = vport_init();
if (err)
goto error_flow_exit;
- major = register_chrdev(0, "openvswitch", &openvswitch_fops);
+ err = register_netdevice_notifier(&dp_device_notifier);
+ if (err)
+ goto error_vport_exit;
+
+ err = dp_register_genl();
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_tnl_exit:
+ tnl_exit();
error:
return err;
}
static void dp_cleanup(void)
{
rcu_barrier();
- unregister_chrdev(major, "openvswitch");
+ dp_unregister_genl(ARRAY_SIZE(dp_genl_families));
unregister_netdevice_notifier(&dp_device_notifier);
+ vport_exit();
flow_exit();
- br_handle_frame_hook = NULL;
- dp_avoid_bridge_exit();
+ tnl_exit();
}
module_init(dp_init);
git://git.onelab.eu / sliver-openvswitch.git / blobdiff