/*
- * Copyright (c) 2007, 2008, 2009, 2010 Nicira Networks.
- * Distributed under the terms of the GNU GPL version 2.
+ * Copyright (c) 2007-2012 Nicira Networks.
*
- * Significant portions of this file may be copied from parts of the Linux
- * kernel, by Linus Torvalds and others.
+ * This program is free software; you can redistribute it and/or
+ * modify it under the terms of version 2 of the GNU General Public
+ * License as published by the Free Software Foundation.
+ *
+ * This program is distributed in the hope that it will be useful, but
+ * WITHOUT ANY WARRANTY; without even the implied warranty of
+ * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU
+ * General Public License for more details.
+ *
+ * You should have received a copy of the GNU General Public License
+ * along with this program; if not, write to the Free Software
+ * Foundation, Inc., 51 Franklin Street, Fifth Floor, Boston, MA
+ * 02110-1301, USA
*/
-/* 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_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/mutex.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/openvswitch.h>
#include <linux/rculist.h>
#include <linux/dmi.h>
-#include <net/inet_ecn.h>
-#include <linux/compat.h>
+#include <net/genetlink.h>
+#include <net/net_namespace.h>
+#include <net/netns/generic.h>
-#include "openvswitch/datapath-protocol.h"
+#include "checksum.h"
#include "datapath.h"
-#include "actions.h"
#include "flow.h"
-#include "odp-compat.h"
-#include "table.h"
+#include "genl_exec.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(3,3,0)
+#error Kernels before 2.6.18 or after 3.2 are not supported by this version of Open vSwitch.
+#endif
+
+#define REHASH_FLOW_INTERVAL (10 * 60 * HZ)
+static void rehash_flow_table(struct work_struct *work);
+static DECLARE_DELAYED_WORK(rehash_flow_wq, rehash_flow_table);
+int ovs_net_id __read_mostly;
-int (*dp_ioctl_hook)(struct net_device *dev, struct ifreq *rq, int cmd);
-EXPORT_SYMBOL(dp_ioctl_hook);
+int (*ovs_dp_ioctl_hook)(struct net_device *dev, struct ifreq *rq, int cmd);
+EXPORT_SYMBOL(ovs_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, etc.) are protected by genl_mutex. The RTNL lock nests inside
+ * genl_mutex.
*
- * It is safe to access the datapath and dp_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);
-
-/* We limit the number of times that we pass into dp_process_received_packet()
- * to avoid blowing out the stack in the event that we have a loop. */
-struct loop_counter {
- int count; /* Count. */
- bool looping; /* Loop detected? */
-};
-#define DP_MAX_LOOPS 5
+static struct vport *new_vport(const struct vport_parms *);
+static int queue_gso_packets(struct net *, int dp_ifindex, struct sk_buff *,
+ const struct dp_upcall_info *);
+static int queue_userspace_packet(struct net *, int dp_ifindex,
+ struct sk_buff *,
+ const struct dp_upcall_info *);
-/* We use a separate counter for each CPU for both interrupt and non-interrupt
- * context in order to keep the limit deterministic for a given packet. */
-struct percpu_loop_counters {
- struct loop_counter counters[2];
-};
+/* Must be called with rcu_read_lock, genl_mutex, or RTNL lock. */
+static struct datapath *get_dp(struct net *net, int dp_ifindex)
+{
+ struct datapath *dp = NULL;
+ struct net_device *dev;
-static DEFINE_PER_CPU(struct percpu_loop_counters, dp_loop_counters);
+ rcu_read_lock();
+ dev = dev_get_by_index_rcu(net, dp_ifindex);
+ if (dev) {
+ struct vport *vport = ovs_internal_dev_get_vport(dev);
+ if (vport)
+ dp = vport->dp;
+ }
+ rcu_read_unlock();
-static int new_dp_port(struct datapath *, struct odp_port *, int port_no);
+ return dp;
+}
-/* 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 *ovs_dp_name(const struct datapath *dp)
{
- if (dp_idx < 0 || dp_idx >= ODP_MAX)
- return NULL;
- return rcu_dereference(dps[dp_idx]);
+ struct vport *vport = ovs_vport_rtnl_rcu(dp, OVSP_LOCAL);
+ return vport->ops->get_name(vport);
}
-EXPORT_SYMBOL_GPL(get_dp);
-static struct datapath *get_dp_locked(int dp_idx)
+static int get_dpifindex(struct datapath *dp)
{
- struct datapath *dp;
+ struct vport *local;
+ int ifindex;
- mutex_lock(&dp_mutex);
- dp = get_dp(dp_idx);
- if (dp)
- mutex_lock(&dp->mutex);
- mutex_unlock(&dp_mutex);
- return dp;
-}
+ rcu_read_lock();
-/* 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);
+ local = ovs_vport_rcu(dp, OVSP_LOCAL);
+ if (local)
+ ifindex = local->ops->get_ifindex(local);
+ else
+ ifindex = 0;
+
+ rcu_read_unlock();
+
+ return ifindex;
}
-static inline size_t br_nlmsg_size(void)
+static size_t br_nlmsg_size(void)
{
return NLMSG_ALIGN(sizeof(struct ifinfomsg))
+ nla_total_size(IFNAMSIZ) /* IFLA_IFNAME */
+ 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 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);
+ struct datapath *dp = port->dp;
struct ifinfomsg *hdr;
struct nlmsghdr *nlh;
- if (ifindex < 0)
- return ifindex;
-
- if (iflink < 0)
- return iflink;
+ if (!port->ops->get_ifindex)
+ return -ENODEV;
nlh = nlmsg_put(skb, 0, 0, event, sizeof(*hdr), flags);
if (nlh == NULL)
hdr->ifi_family = AF_BRIDGE;
hdr->__ifi_pad = 0;
hdr->ifi_type = ARPHRD_ETHER;
- hdr->ifi_index = ifindex;
- hdr->ifi_flags = vport_get_flags(port->vport);
+ hdr->ifi_index = port->ops->get_ifindex(port);
+ hdr->ifi_flags = port->ops->get_dev_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, port->ops->get_name(port));
+ NLA_PUT_U32(skb, IFLA_MASTER, get_dpifindex(dp));
+ NLA_PUT_U32(skb, IFLA_MTU, port->ops->get_mtu(port));
#ifdef IFLA_OPERSTATE
NLA_PUT_U8(skb, IFLA_OPERSTATE,
- vport_is_running(port->vport)
- ? vport_get_operstate(port->vport)
+ port->ops->is_running(port)
+ ? port->ops->get_operstate(port)
: IF_OPER_DOWN);
#endif
- NLA_PUT(skb, IFLA_ADDRESS, ETH_ALEN,
- vport_get_addr(port->vport));
-
- if (ifindex != iflink)
- NLA_PUT_U32(skb, IFLA_LINK,iflink);
+ NLA_PUT(skb, IFLA_ADDRESS, ETH_ALEN, port->ops->get_addr(port));
return nlmsg_end(skb, nlh);
return -EMSGSIZE;
}
-static void dp_ifinfo_notify(int event, struct dp_port *port)
+/* Caller must hold RTNL lock. */
+static void dp_ifinfo_notify(int event, struct vport *port)
{
struct sk_buff *skb;
- int err = -ENOBUFS;
+ int err;
skb = nlmsg_new(br_nlmsg_size(), GFP_KERNEL);
- if (skb == NULL)
- goto errout;
+ if (!skb) {
+ err = -ENOBUFS;
+ goto err;
+ }
err = dp_fill_ifinfo(skb, port, event, 0);
if (err < 0) {
- /* -EMSGSIZE implies BUG in br_nlmsg_size() */
- WARN_ON(err == -EMSGSIZE);
- kfree_skb(skb);
- goto errout;
+ if (err == -ENODEV) {
+ goto out;
+ } else {
+ /* -EMSGSIZE implies BUG in br_nlmsg_size() */
+ WARN_ON(err == -EMSGSIZE);
+ goto err;
+ }
}
- rtnl_notify(skb, &init_net, 0, RTNLGRP_LINK, NULL, GFP_KERNEL);
+
+ rtnl_notify(skb, ovs_dp_get_net(port->dp), 0, RTNLGRP_LINK, NULL, GFP_KERNEL);
+
return;
-errout:
- if (err < 0)
- rtnl_set_sk_err(&init_net, RTNLGRP_LINK, err);
+err:
+ rtnl_set_sk_err(ovs_dp_get_net(port->dp), RTNLGRP_LINK, err);
+out:
+ kfree_skb(skb);
}
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 odp_port internal_dev_port;
- char devname[IFNAMSIZ];
- struct datapath *dp;
- int err;
- int i;
-
- if (devnamep) {
- int retval = strncpy_from_user(devname, devnamep, IFNAMSIZ);
- if (retval < 0) {
- err = -EFAULT;
- goto err;
- } else if (retval >= IFNAMSIZ) {
- err = -ENAMETOOLONG;
- goto err;
- }
- } 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;
-
- /* 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;
+ struct datapath *dp = container_of(rcu, struct datapath, rcu);
- 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);
-
- /* 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);
+ ovs_flow_tbl_destroy((__force struct flow_table *)dp->table);
+ free_percpu(dp->stats_percpu);
+ release_net(ovs_dp_get_net(dp));
+ kfree(dp->ports);
+ kobject_put(&dp->ifobj);
+}
- /* Allocate table. */
- err = -ENOMEM;
- rcu_assign_pointer(dp->table, tbl_create(0));
- if (!dp->table)
- goto err_free_dp;
+static struct hlist_head *vport_hash_bucket(const struct datapath *dp,
+ u16 port_no)
+{
+ return &dp->ports[port_no & (DP_VPORT_HASH_BUCKETS - 1)];
+}
- /* 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);
- if (err) {
- if (err == -EBUSY)
- err = -EEXIST;
+struct vport *ovs_lookup_vport(const struct datapath *dp, u16 port_no)
+{
+ struct vport *vport;
+ struct hlist_node *n;
+ struct hlist_head *head;
- goto err_destroy_table;
+ head = vport_hash_bucket(dp, port_no);
+ hlist_for_each_entry_rcu(vport, n, head, dp_hash_node) {
+ if (vport->port_no == port_no)
+ return vport;
}
+ return NULL;
+}
- dp->drop_frags = 0;
- dp->stats_percpu = alloc_percpu(struct dp_stats_percpu);
- if (!dp->stats_percpu)
- goto err_destroy_local_port;
-
- rcu_assign_pointer(dps[dp_idx], dp);
- mutex_unlock(&dp_mutex);
- rtnl_unlock();
-
- dp_sysfs_add_dp(dp);
+/* Called with RTNL lock and genl_lock. */
+static struct vport *new_vport(const struct vport_parms *parms)
+{
+ struct vport *vport;
- return 0;
+ vport = ovs_vport_add(parms);
+ if (!IS_ERR(vport)) {
+ struct datapath *dp = parms->dp;
+ struct hlist_head *head = vport_hash_bucket(dp, vport->port_no);
-err_destroy_local_port:
- dp_detach_port(dp->ports[ODPP_LOCAL], 1);
-err_destroy_table:
- tbl_destroy(dp->table, NULL);
-err_free_dp:
- kfree(dp);
-err_put_module:
- module_put(THIS_MODULE);
-err_unlock:
- mutex_unlock(&dp_mutex);
- rtnl_unlock();
-err:
- return err;
+ hlist_add_head_rcu(&vport->dp_hash_node, head);
+ dp_ifinfo_notify(RTM_NEWLINK, vport);
+ }
+ return vport;
}
-static void do_destroy_dp(struct datapath *dp)
+/* Called with RTNL lock. */
+void ovs_dp_detach_port(struct vport *p)
{
- struct dp_port *p, *n;
- int i;
-
- list_for_each_entry_safe (p, n, &dp->port_list, node)
- if (p->port_no != ODPP_LOCAL)
- dp_detach_port(p, 1);
-
- dp_sysfs_del_dp(dp);
+ ASSERT_RTNL();
- rcu_assign_pointer(dps[dp->dp_idx], NULL);
+ if (p->port_no != OVSP_LOCAL)
+ ovs_dp_sysfs_del_if(p);
- dp_detach_port(dp->ports[ODPP_LOCAL], 1);
+ dp_ifinfo_notify(RTM_DELLINK, p);
- tbl_destroy(dp->table, flow_free_tbl);
+ /* First drop references to device. */
+ hlist_del_rcu(&p->dp_hash_node);
- 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);
+ /* Then destroy it. */
+ ovs_vport_del(p);
}
-static int destroy_dp(int dp_idx)
+/* Must be called with rcu_read_lock. */
+void ovs_dp_process_received_packet(struct vport *p, struct sk_buff *skb)
{
- struct datapath *dp;
- int err;
+ struct datapath *dp = p->dp;
+ struct sw_flow *flow;
+ struct dp_stats_percpu *stats;
+ u64 *stats_counter;
+ int error;
- rtnl_lock();
- mutex_lock(&dp_mutex);
- dp = get_dp(dp_idx);
- err = -ENODEV;
- if (!dp)
- goto err_unlock;
+ stats = per_cpu_ptr(dp->stats_percpu, smp_processor_id());
- do_destroy_dp(dp);
- err = 0;
+ if (!OVS_CB(skb)->flow) {
+ struct sw_flow_key key;
+ int key_len;
-err_unlock:
- mutex_unlock(&dp_mutex);
- rtnl_unlock();
- return err;
-}
+ /* Extract flow from 'skb' into 'key'. */
+ error = ovs_flow_extract(skb, p->port_no, &key, &key_len);
+ if (unlikely(error)) {
+ kfree_skb(skb);
+ return;
+ }
-static void release_dp_port(struct kobject *kobj)
-{
- struct dp_port *p = container_of(kobj, struct dp_port, kobj);
- kfree(p);
-}
+ /* Look up flow. */
+ flow = ovs_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 = NULL;
+ upcall.pid = p->upcall_pid;
+ ovs_dp_upcall(dp, skb, &upcall);
+ consume_skb(skb);
+ stats_counter = &stats->n_missed;
+ goto out;
+ }
-static struct kobj_type brport_ktype = {
-#ifdef CONFIG_SYSFS
- .sysfs_ops = &brport_sysfs_ops,
-#endif
- .release = release_dp_port
+ OVS_CB(skb)->flow = flow;
+ }
+
+ stats_counter = &stats->n_hit;
+ ovs_flow_used(OVS_CB(skb)->flow, skb);
+ ovs_execute_actions(dp, skb);
+
+out:
+ /* Update datapath statistics. */
+ u64_stats_update_begin(&stats->sync);
+ (*stats_counter)++;
+ u64_stats_update_end(&stats->sync);
+}
+
+static struct genl_family dp_packet_genl_family = {
+ .id = GENL_ID_GENERATE,
+ .hdrsize = sizeof(struct ovs_header),
+ .name = OVS_PACKET_FAMILY,
+ .version = OVS_PACKET_VERSION,
+ .maxattr = OVS_PACKET_ATTR_MAX,
+ SET_NETNSOK
};
-/* Called with RTNL lock and dp_mutex. */
-static int new_dp_port(struct datapath *dp, struct odp_port *odp_port, int port_no)
+int ovs_dp_upcall(struct datapath *dp, struct sk_buff *skb,
+ const struct dp_upcall_info *upcall_info)
{
- struct vport *vport;
- struct dp_port *p;
+ struct dp_stats_percpu *stats;
+ int dp_ifindex;
int err;
- vport = vport_locate(odp_port->devname);
- if (!vport) {
- vport_lock();
-
- if (odp_port->flags & ODP_PORT_INTERNAL)
- vport = vport_add(odp_port->devname, "internal", NULL);
- else
- vport = vport_add(odp_port->devname, "netdev", NULL);
-
- vport_unlock();
-
- if (IS_ERR(vport))
- return PTR_ERR(vport);
+ if (upcall_info->pid == 0) {
+ err = -ENOTCONN;
+ goto err;
}
- p = kzalloc(sizeof(*p), GFP_KERNEL);
- if (!p)
- return -ENOMEM;
+ dp_ifindex = get_dpifindex(dp);
+ if (!dp_ifindex) {
+ err = -ENODEV;
+ goto err;
+ }
- p->port_no = port_no;
- p->dp = dp;
- p->vport = vport;
- atomic_set(&p->sflow_pool, 0);
+ forward_ip_summed(skb, true);
- err = vport_attach(vport, p);
- if (err) {
- kfree(p);
- return err;
- }
+ if (!skb_is_gso(skb))
+ err = queue_userspace_packet(ovs_dp_get_net(dp), dp_ifindex, skb, upcall_info);
+ else
+ err = queue_gso_packets(ovs_dp_get_net(dp), dp_ifindex, skb, upcall_info);
+ if (err)
+ goto err;
- rcu_assign_pointer(dp->ports[port_no], p);
- list_add_rcu(&p->node, &dp->port_list);
- dp->n_ports++;
+ return 0;
- /* 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);
+err:
+ stats = per_cpu_ptr(dp->stats_percpu, smp_processor_id());
- dp_ifinfo_notify(RTM_NEWLINK, p);
+ u64_stats_update_begin(&stats->sync);
+ stats->n_lost++;
+ u64_stats_update_end(&stats->sync);
- return 0;
+ return err;
}
-static int attach_port(int dp_idx, struct odp_port __user *portp)
+static int queue_gso_packets(struct net *net, int dp_ifindex,
+ struct sk_buff *skb,
+ const struct dp_upcall_info *upcall_info)
{
- struct datapath *dp;
- struct odp_port port;
- int port_no;
+ struct dp_upcall_info later_info;
+ struct sw_flow_key later_key;
+ struct sk_buff *segs, *nskb;
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;
-
-got_port_no:
- err = new_dp_port(dp, &port, port_no);
- if (err)
- goto out_unlock_dp;
+ segs = skb_gso_segment(skb, NETIF_F_SG | NETIF_F_HW_CSUM);
+ if (IS_ERR(skb))
+ return PTR_ERR(skb);
- set_internal_devs_mtu(dp);
- dp_sysfs_add_if(dp->ports[port_no]);
+ /* Queue all of the segments. */
+ skb = segs;
+ do {
+ err = queue_userspace_packet(net, dp_ifindex, skb, upcall_info);
+ if (err)
+ break;
- err = put_user(port_no, &portp->port);
+ if (skb == segs && skb_shinfo(skb)->gso_type & SKB_GSO_UDP) {
+ /* The initial flow key extracted by ovs_flow_extract()
+ * in this case is for a first fragment, so we need to
+ * properly mark later fragments.
+ */
+ later_key = *upcall_info->key;
+ later_key.ip.frag = OVS_FRAG_TYPE_LATER;
+
+ later_info = *upcall_info;
+ later_info.key = &later_key;
+ upcall_info = &later_info;
+ }
+ } while ((skb = skb->next));
-out_unlock_dp:
- mutex_unlock(&dp->mutex);
-out_unlock_rtnl:
- rtnl_unlock();
-out:
+ /* Free all of the segments. */
+ skb = segs;
+ do {
+ nskb = skb->next;
+ if (err)
+ kfree_skb(skb);
+ else
+ consume_skb(skb);
+ } while ((skb = nskb));
return err;
}
-int dp_detach_port(struct dp_port *p, int may_delete)
+static int queue_userspace_packet(struct net *net, int dp_ifindex,
+ struct sk_buff *skb,
+ const struct dp_upcall_info *upcall_info)
{
- struct vport *vport = p->vport;
+ struct ovs_header *upcall;
+ struct sk_buff *nskb = NULL;
+ struct sk_buff *user_skb; /* to be queued to userspace */
+ struct nlattr *nla;
+ unsigned int len;
int err;
- ASSERT_RTNL();
-
- if (p->port_no != ODPP_LOCAL)
- dp_sysfs_del_if(p);
- dp_ifinfo_notify(RTM_DELLINK, p);
-
- /* First drop references to device. */
- p->dp->n_ports--;
- list_del_rcu(&p->node);
- rcu_assign_pointer(p->dp->ports[p->port_no], NULL);
+ if (vlan_tx_tag_present(skb)) {
+ nskb = skb_clone(skb, GFP_ATOMIC);
+ if (!nskb)
+ return -ENOMEM;
+
+ err = vlan_deaccel_tag(nskb);
+ if (err)
+ return err;
- err = vport_detach(vport);
- if (err)
- return err;
+ skb = nskb;
+ }
- /* Then wait until no one is still using it, and destroy it. */
- synchronize_rcu();
+ if (nla_attr_size(skb->len) > USHRT_MAX) {
+ err = -EFBIG;
+ goto out;
+ }
- if (may_delete) {
- const char *port_type = vport_get_type(vport);
+ len = sizeof(struct ovs_header);
+ len += nla_total_size(skb->len);
+ len += nla_total_size(FLOW_BUFSIZE);
+ if (upcall_info->cmd == OVS_PACKET_CMD_ACTION)
+ len += nla_total_size(8);
- if (!strcmp(port_type, "netdev") || !strcmp(port_type, "internal")) {
- vport_lock();
- vport_del(vport);
- vport_unlock();
- }
+ user_skb = genlmsg_new(len, GFP_ATOMIC);
+ if (!user_skb) {
+ err = -ENOMEM;
+ goto out;
}
- kobject_put(&p->kobj);
-
- return 0;
-}
+ upcall = genlmsg_put(user_skb, 0, 0, &dp_packet_genl_family,
+ 0, upcall_info->cmd);
+ upcall->dp_ifindex = dp_ifindex;
-static int detach_port(int dp_idx, int port_no)
-{
- struct dp_port *p;
- struct datapath *dp;
- int err;
+ nla = nla_nest_start(user_skb, OVS_PACKET_ATTR_KEY);
+ ovs_flow_to_nlattrs(upcall_info->key, user_skb);
+ nla_nest_end(user_skb, nla);
- err = -EINVAL;
- if (port_no < 0 || port_no >= DP_MAX_PORTS || port_no == ODPP_LOCAL)
- goto out;
+ if (upcall_info->userdata)
+ nla_put_u64(user_skb, OVS_PACKET_ATTR_USERDATA,
+ nla_get_u64(upcall_info->userdata));
- rtnl_lock();
- dp = get_dp_locked(dp_idx);
- err = -ENODEV;
- if (!dp)
- goto out_unlock_rtnl;
+ nla = __nla_reserve(user_skb, OVS_PACKET_ATTR_PACKET, skb->len);
- p = dp->ports[port_no];
- err = -ENOENT;
- if (!p)
- goto out_unlock_dp;
+ skb_copy_and_csum_dev(skb, nla_data(nla));
- err = dp_detach_port(p, 1);
+ err = genlmsg_unicast(net, user_skb, upcall_info->pid);
-out_unlock_dp:
- mutex_unlock(&dp->mutex);
-out_unlock_rtnl:
- rtnl_unlock();
out:
+ kfree_skb(nskb);
return err;
}
-static void suppress_loop(struct datapath *dp, struct sw_flow_actions *actions)
-{
- if (net_ratelimit())
- pr_warn("%s: flow looped %d times, dropping\n",
- dp_name(dp), DP_MAX_LOOPS);
- actions->n_actions = 0;
-}
-
-/* Must be called with rcu_read_lock. */
-void dp_process_received_packet(struct dp_port *p, struct sk_buff *skb)
+/* Called with genl_mutex. */
+static int flush_flows(struct datapath *dp)
{
- struct datapath *dp = p->dp;
- struct dp_stats_percpu *stats;
- int stats_counter_off;
- struct sw_flow_actions *acts;
- struct loop_counter *loop;
- int error;
+ struct flow_table *old_table;
+ struct flow_table *new_table;
- OVS_CB(skb)->dp_port = p;
+ old_table = genl_dereference(dp->table);
+ new_table = ovs_flow_tbl_alloc(TBL_MIN_BUCKETS);
+ if (!new_table)
+ return -ENOMEM;
- if (!OVS_CB(skb)->flow) {
- struct odp_flow_key key;
- struct tbl_node *flow_node;
- bool is_frag;
+ rcu_assign_pointer(dp->table, new_table);
- /* Extract flow from 'skb' into 'key'. */
- error = flow_extract(skb, p ? p->port_no : ODPP_NONE, &key, &is_frag);
- if (unlikely(error)) {
- kfree_skb(skb);
- return;
- }
+ ovs_flow_tbl_deferred_destroy(old_table);
+ return 0;
+}
- if (is_frag && dp->drop_frags) {
- kfree_skb(skb);
- stats_counter_off = offsetof(struct dp_stats_percpu, n_frags);
- goto out;
- }
+static int validate_actions(const struct nlattr *attr,
+ const struct sw_flow_key *key, int depth);
- /* Look up flow. */
- flow_node = tbl_lookup(rcu_dereference(dp->table), &key,
- flow_hash(&key), flow_cmp);
- if (unlikely(!flow_node)) {
- dp_output_control(dp, skb, _ODPL_MISS_NR, OVS_CB(skb)->tun_id);
- stats_counter_off = offsetof(struct dp_stats_percpu, n_missed);
- goto out;
- }
+static int validate_sample(const struct nlattr *attr,
+ const struct sw_flow_key *key, int depth)
+{
+ const struct nlattr *attrs[OVS_SAMPLE_ATTR_MAX + 1];
+ const struct nlattr *probability, *actions;
+ const struct nlattr *a;
+ int rem;
- OVS_CB(skb)->flow = flow_cast(flow_node);
+ memset(attrs, 0, sizeof(attrs));
+ nla_for_each_nested(a, attr, rem) {
+ int type = nla_type(a);
+ if (!type || type > OVS_SAMPLE_ATTR_MAX || attrs[type])
+ return -EINVAL;
+ attrs[type] = a;
}
+ if (rem)
+ return -EINVAL;
- flow_used(OVS_CB(skb)->flow, skb);
+ probability = attrs[OVS_SAMPLE_ATTR_PROBABILITY];
+ if (!probability || nla_len(probability) != sizeof(u32))
+ return -EINVAL;
- acts = rcu_dereference(OVS_CB(skb)->flow->sf_acts);
+ actions = attrs[OVS_SAMPLE_ATTR_ACTIONS];
+ if (!actions || (nla_len(actions) && nla_len(actions) < NLA_HDRLEN))
+ return -EINVAL;
+ return validate_actions(actions, key, depth + 1);
+}
- /* Check whether we've looped too much. */
- loop = &get_cpu_var(dp_loop_counters).counters[!!in_interrupt()];
- if (unlikely(++loop->count > DP_MAX_LOOPS))
- loop->looping = true;
- if (unlikely(loop->looping)) {
- suppress_loop(dp, acts);
- goto out_loop;
- }
+static int validate_set(const struct nlattr *a,
+ const struct sw_flow_key *flow_key)
+{
+ const struct nlattr *ovs_key = nla_data(a);
+ int key_type = nla_type(ovs_key);
- /* Execute actions. */
- execute_actions(dp, skb, &OVS_CB(skb)->flow->key, acts->actions,
- acts->n_actions, GFP_ATOMIC);
- stats_counter_off = offsetof(struct dp_stats_percpu, n_hit);
+ /* There can be only one key in a action */
+ if (nla_total_size(nla_len(ovs_key)) != nla_len(a))
+ return -EINVAL;
- /* Check whether sub-actions looped too much. */
- if (unlikely(loop->looping))
- suppress_loop(dp, acts);
+ if (key_type > OVS_KEY_ATTR_MAX ||
+ nla_len(ovs_key) != ovs_key_lens[key_type])
+ return -EINVAL;
-out_loop:
- /* Decrement loop counter. */
- if (!--loop->count)
- loop->looping = false;
- put_cpu_var(dp_loop_counters);
+ switch (key_type) {
+ const struct ovs_key_ipv4 *ipv4_key;
-out:
- /* Update datapath statistics. */
- local_bh_disable();
- stats = per_cpu_ptr(dp->stats_percpu, smp_processor_id());
+ case OVS_KEY_ATTR_PRIORITY:
+ case OVS_KEY_ATTR_TUN_ID:
+ case OVS_KEY_ATTR_ETHERNET:
+ break;
- write_seqcount_begin(&stats->seqlock);
- (*(u64 *)((u8 *)stats + stats_counter_off))++;
- write_seqcount_end(&stats->seqlock);
+ case OVS_KEY_ATTR_IPV4:
+ if (flow_key->eth.type != htons(ETH_P_IP))
+ return -EINVAL;
- local_bh_enable();
-}
+ if (!flow_key->ipv4.addr.src || !flow_key->ipv4.addr.dst)
+ return -EINVAL;
-#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;
+ ipv4_key = nla_data(ovs_key);
+ if (ipv4_key->ipv4_proto != flow_key->ip.proto)
+ return -EINVAL;
- if (!skb->proto_csum_blank)
- return 0;
+ if (ipv4_key->ipv4_frag != flow_key->ip.frag)
+ return -EINVAL;
- if (skb->protocol != htons(ETH_P_IP))
- goto out;
+ break;
- if (!pskb_may_pull(skb, skb_network_header(skb) + sizeof(struct iphdr) - skb->data))
- goto out;
+ case OVS_KEY_ATTR_TCP:
+ if (flow_key->ip.proto != IPPROTO_TCP)
+ return -EINVAL;
- iph = ip_hdr(skb);
- th = skb_network_header(skb) + 4 * iph->ihl;
+ if (!flow_key->ipv4.tp.src || !flow_key->ipv4.tp.dst)
+ return -EINVAL;
- 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);
+
+ case OVS_KEY_ATTR_UDP:
+ if (flow_key->ip.proto != IPPROTO_UDP)
+ return -EINVAL;
+
+ if (!flow_key->ipv4.tp.src || !flow_key->ipv4.tp.dst)
+ return -EINVAL;
break;
+
default:
- if (net_ratelimit())
- pr_err("Attempting to checksum a non-TCP/UDP packet, "
- "dropping a protocol %d packet",
- iph->protocol);
- goto out;
+ return -EINVAL;
}
- if (!pskb_may_pull(skb, th + csum_offset + 2 - skb->data))
- goto out;
+ return 0;
+}
- skb->ip_summed = CHECKSUM_PARTIAL;
- skb->proto_csum_blank = 0;
+static int validate_userspace(const struct nlattr *attr)
+{
+ static const struct nla_policy userspace_policy[OVS_USERSPACE_ATTR_MAX + 1] = {
+ [OVS_USERSPACE_ATTR_PID] = {.type = NLA_U32 },
+ [OVS_USERSPACE_ATTR_USERDATA] = {.type = NLA_U64 },
+ };
+ struct nlattr *a[OVS_USERSPACE_ATTR_MAX + 1];
+ int error;
-#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
+ error = nla_parse_nested(a, OVS_USERSPACE_ATTR_MAX,
+ attr, userspace_policy);
+ if (error)
+ return error;
- err = 0;
+ if (!a[OVS_USERSPACE_ATTR_PID] ||
+ !nla_get_u32(a[OVS_USERSPACE_ATTR_PID]))
+ return -EINVAL;
-out:
- return err;
+ return 0;
}
-#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)
+
+static int validate_actions(const struct nlattr *attr,
+ const struct sw_flow_key *key, int depth)
{
- /* 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;
+ const struct nlattr *a;
+ int rem, err;
- 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 (depth >= SAMPLE_ACTION_DEPTH)
+ return -EOVERFLOW;
-#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;
+ nla_for_each_nested(a, attr, rem) {
+ /* Expected argument lengths, (u32)-1 for variable length. */
+ static const u32 action_lens[OVS_ACTION_ATTR_MAX + 1] = {
+ [OVS_ACTION_ATTR_OUTPUT] = sizeof(u32),
+ [OVS_ACTION_ATTR_USERSPACE] = (u32)-1,
+ [OVS_ACTION_ATTR_PUSH_VLAN] = sizeof(struct ovs_action_push_vlan),
+ [OVS_ACTION_ATTR_POP_VLAN] = 0,
+ [OVS_ACTION_ATTR_SET] = (u32)-1,
+ [OVS_ACTION_ATTR_SAMPLE] = (u32)-1
+ };
+ const struct ovs_action_push_vlan *vlan;
+ int type = nla_type(a);
- WARN_ON_ONCE(skb->proto_csum_blank && xmit);
-#endif
-}
+ if (type > OVS_ACTION_ATTR_MAX ||
+ (action_lens[type] != nla_len(a) &&
+ action_lens[type] != (u32)-1))
+ return -EINVAL;
-/* 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
-}
+ switch (type) {
+ case OVS_ACTION_ATTR_UNSPEC:
+ return -EINVAL;
-/* 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;
+ case OVS_ACTION_ATTR_USERSPACE:
+ err = validate_userspace(a);
+ if (err)
+ return err;
+ break;
- if (OVS_CB(skb)->dp_port)
- port_no = OVS_CB(skb)->dp_port->port_no;
- else
- port_no = ODPP_LOCAL;
+ case OVS_ACTION_ATTR_OUTPUT:
+ if (nla_get_u32(a) >= DP_MAX_PORTS)
+ return -EINVAL;
+ break;
- do {
- struct odp_msg *header;
- nskb = skb->next;
- skb->next = NULL;
+ case OVS_ACTION_ATTR_POP_VLAN:
+ break;
- err = skb_cow(skb, sizeof *header);
- if (err)
- goto err_kfree_skbs;
+ case OVS_ACTION_ATTR_PUSH_VLAN:
+ vlan = nla_data(a);
+ if (vlan->vlan_tpid != htons(ETH_P_8021Q))
+ return -EINVAL;
+ if (!(vlan->vlan_tci & htons(VLAN_TAG_PRESENT)))
+ return -EINVAL;
+ break;
- 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);
+ case OVS_ACTION_ATTR_SET:
+ err = validate_set(a, key);
+ if (err)
+ return err;
+ break;
- skb = nskb;
- } while (skb);
- return 0;
+ case OVS_ACTION_ATTR_SAMPLE:
+ err = validate_sample(a, key, depth);
+ if (err)
+ return err;
+ break;
-err_kfree_skbs:
- kfree_skb(skb);
- while ((skb = nskb) != NULL) {
- nskb = skb->next;
- kfree_skb(skb);
+ default:
+ return -EINVAL;
+ }
}
- return err;
+
+ if (rem > 0)
+ return -EINVAL;
+
+ return 0;
}
-int dp_output_control(struct datapath *dp, struct sk_buff *skb, int queue_no,
- u32 arg)
+static void clear_stats(struct sw_flow *flow)
{
- struct dp_stats_percpu *stats;
- struct sk_buff_head *queue;
+ flow->used = 0;
+ flow->tcp_flags = 0;
+ flow->packet_count = 0;
+ flow->byte_count = 0;
+}
+
+static int ovs_packet_cmd_execute(struct sk_buff *skb, struct genl_info *info)
+{
+ 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;
+ int len;
int err;
+ int key_len;
+
+ 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;
+
+ 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);
+
+ memcpy(__skb_put(packet, len), nla_data(a[OVS_PACKET_ATTR_PACKET]), len);
+
+ skb_reset_mac_header(packet);
+ eth = eth_hdr(packet);
+
+ /* 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);
- 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)
+ /* Build an sw_flow for sending this packet. */
+ flow = ovs_flow_alloc();
+ err = PTR_ERR(flow);
+ if (IS_ERR(flow))
goto err_kfree_skb;
- forward_ip_summed(skb);
+ err = ovs_flow_extract(packet, -1, &flow->key, &key_len);
+ if (err)
+ goto err_flow_put;
- err = vswitch_skb_checksum_setup(skb);
+ err = ovs_flow_metadata_from_nlattrs(&flow->key.phy.priority,
+ &flow->key.phy.in_port,
+ &flow->key.phy.tun_id,
+ a[OVS_PACKET_ATTR_KEY]);
if (err)
- goto err_kfree_skb;
+ goto err_flow_put;
- /* 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 (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. */
- }
- }
+ err = validate_actions(a[OVS_PACKET_ATTR_ACTIONS], &flow->key, 0);
+ if (err)
+ goto err_flow_put;
- err = queue_control_packets(skb, queue, queue_no, arg);
- wake_up_interruptible(&dp->waitqueue);
- return err;
+ flow->hash = ovs_flow_hash(&flow->key, key_len);
-err_kfree_skb:
- kfree_skb(skb);
-err:
- local_bh_disable();
- stats = per_cpu_ptr(dp->stats_percpu, smp_processor_id());
+ acts = ovs_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);
- write_seqcount_begin(&stats->seqlock);
- stats->n_lost++;
- write_seqcount_end(&stats->seqlock);
+ OVS_CB(packet)->flow = flow;
+ packet->priority = flow->key.phy.priority;
+
+ rcu_read_lock();
+ dp = get_dp(sock_net(skb->sk), ovs_header->dp_ifindex);
+ err = -ENODEV;
+ if (!dp)
+ goto err_unlock;
+ local_bh_disable();
+ err = ovs_execute_actions(dp, packet);
local_bh_enable();
+ rcu_read_unlock();
+ ovs_flow_put(flow);
+ return err;
+
+err_unlock:
+ rcu_read_unlock();
+err_flow_put:
+ ovs_flow_put(flow);
+err_kfree_skb:
+ kfree_skb(packet);
+err:
return err;
}
-static int flush_flows(struct datapath *dp)
+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 },
+};
+
+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
+ }
+};
+
+static void get_dp_stats(struct datapath *dp, struct ovs_dp_stats *stats)
{
- struct tbl *old_table = rcu_dereference(dp->table);
- struct tbl *new_table;
+ int i;
+ struct flow_table *table = genl_dereference(dp->table);
- new_table = tbl_create(0);
- if (!new_table)
- return -ENOMEM;
+ stats->n_flows = ovs_flow_tbl_count(table);
- rcu_assign_pointer(dp->table, new_table);
+ 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 int start;
+
+ percpu_stats = per_cpu_ptr(dp->stats_percpu, i);
- tbl_deferred_destroy(old_table, flow_free_tbl);
+ do {
+ start = u64_stats_fetch_begin_bh(&percpu_stats->sync);
+ local_stats = *percpu_stats;
+ } while (u64_stats_fetch_retry_bh(&percpu_stats->sync, start));
- return 0;
+ stats->n_hit += local_stats.n_hit;
+ stats->n_missed += local_stats.n_missed;
+ stats->n_lost += local_stats.n_lost;
+ }
}
-static int validate_actions(const struct sw_flow_actions *actions)
-{
- unsigned int i;
+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 },
+};
- 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;
+static struct genl_family dp_flow_genl_family = {
+ .id = GENL_ID_GENERATE,
+ .hdrsize = sizeof(struct ovs_header),
+ .name = OVS_FLOW_FAMILY,
+ .version = OVS_FLOW_VERSION,
+ .maxattr = OVS_FLOW_ATTR_MAX,
+ SET_NETNSOK
+};
- case ODPAT_OUTPUT_GROUP:
- if (a->output_group.group >= DP_MAX_GROUPS)
- return -EINVAL;
- break;
+static struct genl_multicast_group ovs_dp_flow_multicast_group = {
+ .name = OVS_FLOW_MCGROUP
+};
- case ODPAT_SET_VLAN_VID:
- if (a->vlan_vid.vlan_vid & htons(~VLAN_VID_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;
- case ODPAT_SET_VLAN_PCP:
- if (a->vlan_pcp.vlan_pcp
- & ~(VLAN_PCP_MASK >> VLAN_PCP_SHIFT))
- return -EINVAL;
- break;
+ sf_acts = rcu_dereference_protected(flow->sf_acts,
+ lockdep_genl_is_held());
- case ODPAT_SET_NW_TOS:
- if (a->nw_tos.nw_tos & INET_ECN_MASK)
- return -EINVAL;
- break;
+ ovs_header = genlmsg_put(skb, pid, seq, &dp_flow_genl_family, flags, cmd);
+ if (!ovs_header)
+ return -EMSGSIZE;
- default:
- if (a->type >= ODPAT_N_ACTIONS)
- return -EOPNOTSUPP;
- break;
- }
- }
+ ovs_header->dp_ifindex = get_dpifindex(dp);
- return 0;
-}
+ nla = nla_nest_start(skb, OVS_FLOW_ATTR_KEY);
+ if (!nla)
+ goto nla_put_failure;
+ err = ovs_flow_to_nlattrs(&flow->key, skb);
+ if (err)
+ goto error;
+ nla_nest_end(skb, nla);
-static struct sw_flow_actions *get_actions(const struct odp_flow *flow)
-{
- struct sw_flow_actions *actions;
- int error;
+ 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);
- actions = flow_actions_alloc(flow->n_actions);
- error = PTR_ERR(actions);
- if (IS_ERR(actions))
- goto error;
+ if (used)
+ NLA_PUT_U64(skb, OVS_FLOW_ATTR_USED, ovs_flow_used_time(used));
- 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;
+ if (stats.n_packets)
+ NLA_PUT(skb, OVS_FLOW_ATTR_STATS,
+ sizeof(struct ovs_flow_stats), &stats);
- return actions;
+ if (tcp_flags)
+ NLA_PUT_U8(skb, OVS_FLOW_ATTR_TCP_FLAGS, tcp_flags);
-error_free_actions:
- kfree(actions);
-error:
- return ERR_PTR(error);
-}
+ /* 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;
-static struct timespec get_time_offset(void)
-{
- struct timespec now_mono, now_jiffies;
+ return genlmsg_end(skb, ovs_header);
- ktime_get_ts(&now_mono);
- jiffies_to_timespec(jiffies, &now_jiffies);
- return timespec_sub(now_mono, now_jiffies);
+nla_put_failure:
+ err = -EMSGSIZE;
+error:
+ genlmsg_cancel(skb, ovs_header);
+ return err;
}
-static void get_stats(struct sw_flow *flow, struct odp_flow_stats *stats,
- struct timespec time_offset)
+static struct sk_buff *ovs_flow_cmd_alloc_info(struct sw_flow *flow)
{
- if (flow->used) {
- struct timespec flow_ts, used;
+ const struct sw_flow_actions *sf_acts;
+ int len;
- jiffies_to_timespec(flow->used, &flow_ts);
- set_normalized_timespec(&used, flow_ts.tv_sec + time_offset.tv_sec,
- flow_ts.tv_nsec + time_offset.tv_nsec);
+ sf_acts = rcu_dereference_protected(flow->sf_acts,
+ lockdep_genl_is_held());
- stats->used_sec = used.tv_sec;
- stats->used_nsec = used.tv_nsec;
- } else {
- stats->used_sec = 0;
- stats->used_nsec = 0;
- }
+ /* OVS_FLOW_ATTR_KEY */
+ len = nla_total_size(FLOW_BUFSIZE);
+ /* OVS_FLOW_ATTR_ACTIONS */
+ len += nla_total_size(sf_acts->actions_len);
+ /* OVS_FLOW_ATTR_STATS */
+ len += nla_total_size(sizeof(struct ovs_flow_stats));
+ /* OVS_FLOW_ATTR_TCP_FLAGS */
+ len += nla_total_size(1);
+ /* OVS_FLOW_ATTR_USED */
+ len += nla_total_size(8);
- stats->n_packets = flow->packet_count;
- stats->n_bytes = flow->byte_count;
- stats->reserved = 0;
- stats->tcp_flags = flow->tcp_flags;
- stats->error = 0;
-}
+ len += NLMSG_ALIGN(sizeof(struct ovs_header));
-static void clear_stats(struct sw_flow *flow)
-{
- flow->used = 0;
- flow->tcp_flags = 0;
- flow->packet_count = 0;
- flow->byte_count = 0;
+ return genlmsg_new(len, GFP_KERNEL);
}
-static int expand_table(struct datapath *dp)
+static struct sk_buff *ovs_flow_cmd_build_info(struct sw_flow *flow,
+ struct datapath *dp,
+ u32 pid, u32 seq, u8 cmd)
{
- struct tbl *old_table = rcu_dereference(dp->table);
- struct tbl *new_table;
-
- new_table = tbl_expand(old_table);
- if (IS_ERR(new_table))
- return PTR_ERR(new_table);
+ struct sk_buff *skb;
+ int retval;
- rcu_assign_pointer(dp->table, new_table);
- tbl_deferred_destroy(old_table, NULL);
+ skb = ovs_flow_cmd_alloc_info(flow);
+ if (!skb)
+ return ERR_PTR(-ENOMEM);
- return 0;
+ retval = ovs_flow_cmd_fill_info(flow, dp, skb, pid, seq, 0, cmd);
+ BUG_ON(retval < 0);
+ return skb;
}
-static int do_put_flow(struct datapath *dp, struct odp_flow_put *uf,
- struct odp_flow_stats *stats)
+static int ovs_flow_cmd_new_or_set(struct sk_buff *skb, struct genl_info *info)
{
- struct tbl_node *flow_node;
+ struct nlattr **a = info->attrs;
+ struct ovs_header *ovs_header = info->userhdr;
+ struct sw_flow_key key;
struct sw_flow *flow;
- struct tbl *table;
+ 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 = ovs_flow_from_nlattrs(&key, &key_len, a[OVS_FLOW_ATTR_KEY]);
+ if (error)
+ goto error;
- memset(uf->flow.key.reserved, 0, sizeof uf->flow.key.reserved);
+ /* Validate actions. */
+ if (a[OVS_FLOW_ATTR_ACTIONS]) {
+ error = validate_actions(a[OVS_FLOW_ATTR_ACTIONS], &key, 0);
+ if (error)
+ goto error;
+ } else if (info->genlhdr->cmd == OVS_FLOW_CMD_NEW) {
+ error = -EINVAL;
+ goto error;
+ }
- table = rcu_dereference(dp->table);
- flow_node = tbl_lookup(table, &uf->flow.key, flow_hash(&uf->flow.key), flow_cmp);
- if (!flow_node) {
- /* No such flow. */
+ dp = get_dp(sock_net(skb->sk), ovs_header->dp_ifindex);
+ error = -ENODEV;
+ if (!dp)
+ goto error;
+
+ table = genl_dereference(dp->table);
+ flow = ovs_flow_tbl_lookup(table, &key, key_len);
+ if (!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 (tbl_count(table) >= tbl_n_buckets(table)) {
- error = expand_table(dp);
- if (error)
- goto error;
- table = rcu_dereference(dp->table);
+ if (ovs_flow_tbl_need_to_expand(table)) {
+ struct flow_table *new_table;
+
+ new_table = ovs_flow_tbl_expand(table);
+ if (!IS_ERR(new_table)) {
+ rcu_assign_pointer(dp->table, new_table);
+ ovs_flow_tbl_deferred_destroy(table);
+ table = genl_dereference(dp->table);
+ }
}
/* Allocate flow. */
- flow = flow_alloc();
+ flow = ovs_flow_alloc();
if (IS_ERR(flow)) {
error = PTR_ERR(flow);
goto error;
}
- flow->key = uf->flow.key;
+ flow->key = key;
clear_stats(flow);
/* Obtain actions. */
- acts = get_actions(&uf->flow);
+ acts = ovs_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 = tbl_insert(table, &flow->tbl_node, flow_hash(&flow->key));
- if (error)
- goto error_free_flow_acts;
+ flow->hash = ovs_flow_hash(&key, key_len);
+ ovs_flow_tbl_insert(table, flow);
- memset(stats, 0, sizeof(struct odp_flow_stats));
+ 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;
-
- flow = flow_cast(flow_node);
-
- /* Bail out if we're not allowed to modify an existing flow. */
+ struct sw_flow_actions *old_acts;
+ struct nlattr *acts_attrs;
+
+ /* 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());
+ acts_attrs = a[OVS_FLOW_ATTR_ACTIONS];
+ if (acts_attrs &&
+ (old_acts->actions_len != nla_len(acts_attrs) ||
+ memcmp(old_acts->actions, nla_data(acts_attrs),
+ old_acts->actions_len))) {
+ struct sw_flow_actions *new_acts;
+
+ new_acts = ovs_flow_actions_alloc(acts_attrs);
+ 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);
+ ovs_flow_deferred_free_acts(old_acts);
}
- /* Fetch stats, then clear them if necessary. */
- spin_lock_bh(&flow->lock);
- get_stats(flow, stats, get_time_offset());
- 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_bh(&flow->lock);
+ spin_unlock_bh(&flow->lock);
+ }
}
+ if (!IS_ERR(reply))
+ genl_notify(reply, genl_info_net(info), info->snd_pid,
+ ovs_dp_flow_multicast_group.id, info->nlhdr,
+ GFP_KERNEL);
+ else
+ netlink_set_err(GENL_SOCK(sock_net(skb->sk)), 0,
+ ovs_dp_flow_multicast_group.id, PTR_ERR(reply));
return 0;
-error_free_flow_acts:
- kfree(flow->sf_acts);
error_free_flow:
- flow->sf_acts = NULL;
- flow_put(flow);
+ ovs_flow_put(flow);
error:
return error;
}
-static int put_flow(struct datapath *dp, struct odp_flow_put __user *ufp)
+static int ovs_flow_cmd_get(struct sk_buff *skb, struct genl_info *info)
{
- struct odp_flow_stats stats;
- struct odp_flow_put uf;
- int error;
+ 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 -EINVAL;
+ err = ovs_flow_from_nlattrs(&key, &key_len, a[OVS_FLOW_ATTR_KEY]);
+ if (err)
+ return err;
- if (copy_from_user(&uf, ufp, sizeof(struct odp_flow_put)))
- return -EFAULT;
+ dp = get_dp(sock_net(skb->sk), ovs_header->dp_ifindex);
+ if (!dp)
+ return -ENODEV;
- error = do_put_flow(dp, &uf, &stats);
- if (error)
- return error;
+ table = genl_dereference(dp->table);
+ flow = ovs_flow_tbl_lookup(table, &key, key_len);
+ if (!flow)
+ return -ENOENT;
- if (copy_to_user(&ufp->flow.stats, &stats,
- sizeof(struct odp_flow_stats)))
- 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 0;
+ return genlmsg_reply(reply, info);
}
-static int do_answer_query(struct sw_flow *flow, u32 query_flags,
- struct timespec time_offset,
- struct odp_flow_stats __user *ustats,
- union odp_action __user *actions,
- u32 __user *n_actionsp)
+static int ovs_flow_cmd_del(struct sk_buff *skb, struct genl_info *info)
{
- struct sw_flow_actions *sf_acts;
- struct odp_flow_stats stats;
- 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;
- spin_lock_bh(&flow->lock);
- get_stats(flow, &stats, time_offset);
- if (query_flags & ODPFF_ZERO_TCP_FLAGS)
- flow->tcp_flags = 0;
+ dp = get_dp(sock_net(skb->sk), ovs_header->dp_ifindex);
+ if (!dp)
+ return -ENODEV;
- spin_unlock_bh(&flow->lock);
+ if (!a[OVS_FLOW_ATTR_KEY])
+ return flush_flows(dp);
- if (copy_to_user(ustats, &stats, sizeof(struct odp_flow_stats)) ||
- get_user(n_actions, n_actionsp))
- return -EFAULT;
+ err = ovs_flow_from_nlattrs(&key, &key_len, a[OVS_FLOW_ATTR_KEY]);
+ if (err)
+ return err;
- if (!n_actions)
- return 0;
+ table = genl_dereference(dp->table);
+ flow = ovs_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, n_actionsp) ||
- (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_alloc_info(flow);
+ if (!reply)
+ return -ENOMEM;
- return 0;
-}
+ ovs_flow_tbl_remove(table, flow);
-static int answer_query(struct sw_flow *flow, u32 query_flags,
- struct timespec time_offset,
- struct odp_flow __user *ufp)
-{
- union odp_action *actions;
+ err = ovs_flow_cmd_fill_info(flow, dp, reply, info->snd_pid,
+ info->snd_seq, 0, OVS_FLOW_CMD_DEL);
+ BUG_ON(err < 0);
- if (get_user(actions, &ufp->actions))
- return -EFAULT;
+ ovs_flow_deferred_free(flow);
- return do_answer_query(flow, query_flags, time_offset,
- &ufp->stats, actions, &ufp->n_actions);
+ genl_notify(reply, genl_info_net(info), info->snd_pid,
+ ovs_dp_flow_multicast_group.id, info->nlhdr, GFP_KERNEL);
+ return 0;
}
-static struct sw_flow *do_del_flow(struct datapath *dp, struct odp_flow_key *key)
+static int ovs_flow_cmd_dump(struct sk_buff *skb, struct netlink_callback *cb)
{
- struct tbl *table = rcu_dereference(dp->table);
- struct tbl_node *flow_node;
- int error;
-
- memset(key->reserved, 0, sizeof key->reserved);
- flow_node = tbl_lookup(table, key, flow_hash(key), flow_cmp);
- if (!flow_node)
- return ERR_PTR(-ENOENT);
-
- error = tbl_remove(table, flow_node);
- if (error)
- return ERR_PTR(error);
-
- /* XXX Returned flow_node's 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. */
- return flow_cast(flow_node);
-}
-
-static int del_flow(struct datapath *dp, struct odp_flow __user *ufp)
-{
- struct sw_flow *flow;
- struct odp_flow uf;
- int error;
-
- if (copy_from_user(&uf, ufp, sizeof uf))
- return -EFAULT;
-
- flow = do_del_flow(dp, &uf.key);
- if (IS_ERR(flow))
- return PTR_ERR(flow);
+ struct ovs_header *ovs_header = genlmsg_data(nlmsg_data(cb->nlh));
+ struct datapath *dp;
+ struct flow_table *table;
- error = answer_query(flow, 0, get_time_offset(), ufp);
- flow_deferred_free(flow);
- return error;
-}
+ dp = get_dp(sock_net(skb->sk), ovs_header->dp_ifindex);
+ if (!dp)
+ return -ENODEV;
-static int do_query_flows(struct datapath *dp, const struct odp_flowvec *flowvec)
-{
- struct tbl *table = rcu_dereference(dp->table);
- struct timespec time_offset;
- u32 i;
+ table = genl_dereference(dp->table);
- time_offset = get_time_offset();
+ for (;;) {
+ struct sw_flow *flow;
+ u32 bucket, obj;
- for (i = 0; i < flowvec->n_flows; i++) {
- struct odp_flow __user *ufp = &flowvec->flows[i];
- struct odp_flow uf;
- struct tbl_node *flow_node;
- int error;
+ bucket = cb->args[0];
+ obj = cb->args[1];
+ flow = ovs_flow_tbl_next(table, &bucket, &obj);
+ if (!flow)
+ break;
- if (copy_from_user(&uf, ufp, sizeof uf))
- return -EFAULT;
- memset(uf.key.reserved, 0, sizeof uf.key.reserved);
+ 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;
- flow_node = tbl_lookup(table, &uf.key, flow_hash(&uf.key), flow_cmp);
- if (!flow_node)
- error = put_user(ENOENT, &ufp->stats.error);
- else
- error = answer_query(flow_cast(flow_node), uf.flags, time_offset, ufp);
- if (error)
- return -EFAULT;
+ cb->args[0] = bucket;
+ cb->args[1] = obj;
}
- return flowvec->n_flows;
-}
-
-struct list_flows_cbdata {
- struct odp_flow __user *uflows;
- u32 n_flows;
- u32 listed_flows;
- struct timespec time_offset;
+ 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 int list_flow(struct tbl_node *node, void *cbdata_)
-{
- struct sw_flow *flow = flow_cast(node);
- struct list_flows_cbdata *cbdata = cbdata_;
- struct odp_flow __user *ufp = &cbdata->uflows[cbdata->listed_flows++];
- int error;
-
- if (copy_to_user(&ufp->key, &flow->key, sizeof flow->key))
- return -EFAULT;
- error = answer_query(flow, 0, cbdata->time_offset, ufp);
- if (error)
- return error;
-
- if (cbdata->listed_flows >= cbdata->n_flows)
- return cbdata->listed_flows;
- return 0;
-}
-
-static int do_list_flows(struct datapath *dp, const struct odp_flowvec *flowvec)
-{
- struct list_flows_cbdata cbdata;
- int error;
-
- if (!flowvec->n_flows)
- return 0;
-
- cbdata.uflows = flowvec->flows;
- cbdata.n_flows = flowvec->n_flows;
- cbdata.listed_flows = 0;
- cbdata.time_offset = get_time_offset();
-
- error = tbl_foreach(rcu_dereference(dp->table), list_flow, &cbdata);
- return error ? error : cbdata.listed_flows;
-}
-
-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;
-
- uflowvec = (struct odp_flowvec __user *)argp;
- if (copy_from_user(&flowvec, uflowvec, sizeof flowvec))
- return -EFAULT;
+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_UPCALL_PID] = { .type = NLA_U32 },
+};
- if (flowvec.n_flows > INT_MAX / sizeof(struct odp_flow))
- return -EINVAL;
+static struct genl_family dp_datapath_genl_family = {
+ .id = GENL_ID_GENERATE,
+ .hdrsize = sizeof(struct ovs_header),
+ .name = OVS_DATAPATH_FAMILY,
+ .version = OVS_DATAPATH_VERSION,
+ .maxattr = OVS_DP_ATTR_MAX,
+ SET_NETNSOK
+};
- retval = function(dp, &flowvec);
- return (retval < 0 ? retval
- : retval == flowvec.n_flows ? 0
- : put_user(retval, &uflowvec->n_flows));
-}
+static struct genl_multicast_group ovs_dp_datapath_multicast_group = {
+ .name = OVS_DATAPATH_MCGROUP
+};
-static int do_execute(struct datapath *dp, const struct odp_execute *execute)
+static int ovs_dp_cmd_fill_info(struct datapath *dp, struct sk_buff *skb,
+ u32 pid, u32 seq, u32 flags, u8 cmd)
{
- struct odp_flow_key key;
- struct sk_buff *skb;
- struct sw_flow_actions *actions;
- struct ethhdr *eth;
- bool is_frag;
+ struct ovs_header *ovs_header;
+ struct ovs_dp_stats dp_stats;
int err;
- err = -EINVAL;
- if (execute->length < ETH_HLEN || execute->length > 65535)
+ ovs_header = genlmsg_put(skb, pid, seq, &dp_datapath_genl_family,
+ flags, cmd);
+ if (!ovs_header)
goto error;
- actions = flow_actions_alloc(execute->n_actions);
- if (IS_ERR(actions)) {
- err = PTR_ERR(actions);
- goto error;
- }
-
- err = -EFAULT;
- if (copy_from_user(actions->actions, execute->actions,
- execute->n_actions * sizeof *execute->actions))
- goto error_free_actions;
-
- err = validate_actions(actions);
- if (err)
- goto error_free_actions;
-
- err = -ENOMEM;
- skb = alloc_skb(execute->length, GFP_KERNEL);
- if (!skb)
- goto error_free_actions;
-
- if (execute->in_port < DP_MAX_PORTS)
- OVS_CB(skb)->dp_port = dp->ports[execute->in_port];
- else
- OVS_CB(skb)->dp_port = NULL;
-
- err = -EFAULT;
- if (copy_from_user(skb_put(skb, execute->length), execute->data,
- execute->length))
- goto error_free_skb;
-
- skb_reset_mac_header(skb);
- eth = eth_hdr(skb);
-
- /* 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);
-
- err = flow_extract(skb, execute->in_port, &key, &is_frag);
- if (err)
- goto error_free_skb;
+ ovs_header->dp_ifindex = get_dpifindex(dp);
rcu_read_lock();
- err = execute_actions(dp, skb, &key, actions->actions,
- actions->n_actions, GFP_KERNEL);
+ err = nla_put_string(skb, OVS_DP_ATTR_NAME, ovs_dp_name(dp));
rcu_read_unlock();
+ if (err)
+ goto nla_put_failure;
- kfree(actions);
- return err;
-
-error_free_skb:
- kfree_skb(skb);
-error_free_actions:
- kfree(actions);
-error:
- return err;
-}
-
-static int execute_packet(struct datapath *dp, const struct odp_execute __user *executep)
-{
- struct odp_execute execute;
+ get_dp_stats(dp, &dp_stats);
+ NLA_PUT(skb, OVS_DP_ATTR_STATS, sizeof(struct ovs_dp_stats), &dp_stats);
- if (copy_from_user(&execute, executep, sizeof execute))
- return -EFAULT;
+ return genlmsg_end(skb, ovs_header);
- return do_execute(dp, &execute);
+nla_put_failure:
+ genlmsg_cancel(skb, ovs_header);
+error:
+ return -EMSGSIZE;
}
-static int get_dp_stats(struct datapath *dp, struct odp_stats __user *statsp)
+static struct sk_buff *ovs_dp_cmd_build_info(struct datapath *dp, u32 pid,
+ u32 seq, u8 cmd)
{
- struct tbl *table = rcu_dereference(dp->table);
- struct odp_stats stats;
- int i;
-
- stats.n_flows = tbl_count(table);
- stats.cur_capacity = tbl_n_buckets(table);
- stats.max_capacity = TBL_MAX_BUCKETS;
- stats.n_ports = dp->n_ports;
- stats.max_ports = DP_MAX_PORTS;
- stats.max_groups = DP_MAX_GROUPS;
- 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;
-
- percpu_stats = per_cpu_ptr(dp->stats_percpu, i);
+ struct sk_buff *skb;
+ int retval;
- do {
- seqcount = read_seqcount_begin(&percpu_stats->seqlock);
- local_stats = *percpu_stats;
- } while (read_seqcount_retry(&percpu_stats->seqlock, seqcount));
+ skb = nlmsg_new(NLMSG_DEFAULT_SIZE, GFP_KERNEL);
+ if (!skb)
+ return ERR_PTR(-ENOMEM);
- 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;
+ retval = ovs_dp_cmd_fill_info(dp, 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_dp_cmd_validate(struct nlattr *a[OVS_DP_ATTR_MAX + 1])
{
- struct dp_port *p;
- int mtu = 0;
-
- ASSERT_RTNL();
-
- list_for_each_entry_rcu (p, &dp->port_list, node) {
- int dev_mtu;
-
- /* Skip any internal ports, since that's what we're trying to
- * set. */
- if (is_internal_vport(p->vport))
- continue;
-
- dev_mtu = vport_get_mtu(p->vport);
- if (!mtu || dev_mtu < mtu)
- mtu = dev_mtu;
- }
-
- return mtu ? mtu : ETH_DATA_LEN;
+ return CHECK_NUL_STRING(a[OVS_DP_ATTR_NAME], IFNAMSIZ - 1);
}
-/* Sets the MTU of all datapath devices to the minimum of the ports. Must
- * be called with RTNL lock. */
-void set_internal_devs_mtu(const struct datapath *dp)
+/* Called with genl_mutex and optionally with RTNL lock also. */
+static struct datapath *lookup_datapath(struct net *net,
+ struct ovs_header *ovs_header,
+ struct nlattr *a[OVS_DP_ATTR_MAX + 1])
{
- struct dp_port *p;
- int mtu;
-
- ASSERT_RTNL();
+ struct datapath *dp;
- mtu = dp_min_mtu(dp);
+ if (!a[OVS_DP_ATTR_NAME])
+ dp = get_dp(net, ovs_header->dp_ifindex);
+ else {
+ struct vport *vport;
- list_for_each_entry_rcu (p, &dp->port_list, node) {
- if (is_internal_vport(p->vport))
- vport_set_mtu(p->vport, mtu);
+ rcu_read_lock();
+ vport = ovs_vport_locate(net, 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);
}
-static int put_port(const struct dp_port *p, struct odp_port __user *uop)
+static int ovs_dp_cmd_new(struct sk_buff *skb, struct genl_info *info)
{
- struct odp_port op;
-
- memset(&op, 0, sizeof op);
-
- rcu_read_lock();
- strncpy(op.devname, vport_get_name(p->vport), sizeof op.devname);
- rcu_read_unlock();
-
- op.port = p->port_no;
- op.flags = is_internal_vport(p->vport) ? ODP_PORT_INTERNAL : 0;
+ struct nlattr **a = info->attrs;
+ struct vport_parms parms;
+ struct sk_buff *reply;
+ struct datapath *dp;
+ struct vport *vport;
+ struct ovs_net *ovs_net;
+ int err, i;
- return copy_to_user(uop, &op, sizeof op) ? -EFAULT : 0;
-}
+ err = -EINVAL;
+ if (!a[OVS_DP_ATTR_NAME] || !a[OVS_DP_ATTR_UPCALL_PID])
+ goto err;
-static int query_port(struct datapath *dp, struct odp_port __user *uport)
-{
- struct odp_port port;
+ err = ovs_dp_cmd_validate(a);
+ if (err)
+ goto err;
- if (copy_from_user(&port, uport, sizeof port))
- return -EFAULT;
+ rtnl_lock();
- if (port.devname[0]) {
- struct vport *vport;
- struct dp_port *dp_port;
- int err = 0;
+ err = -ENOMEM;
+ dp = kzalloc(sizeof(*dp), GFP_KERNEL);
+ if (dp == NULL)
+ goto err_unlock_rtnl;
- port.devname[IFNAMSIZ - 1] = '\0';
+ /* 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);
- vport_lock();
- rcu_read_lock();
+ /* Allocate table. */
+ err = -ENOMEM;
+ rcu_assign_pointer(dp->table, ovs_flow_tbl_alloc(TBL_MIN_BUCKETS));
+ if (!dp->table)
+ goto err_free_dp;
- vport = vport_locate(port.devname);
- if (!vport) {
- err = -ENODEV;
- goto error_unlock;
- }
+ dp->stats_percpu = alloc_percpu(struct dp_stats_percpu);
+ if (!dp->stats_percpu) {
+ err = -ENOMEM;
+ goto err_destroy_table;
+ }
+ ovs_dp_set_net(dp, hold_net(sock_net(skb->sk)));
- dp_port = vport_get_dp_port(vport);
- if (!dp_port || dp_port->dp != dp) {
- err = -ENOENT;
- goto error_unlock;
- }
+ dp->ports = kmalloc(DP_VPORT_HASH_BUCKETS * sizeof(struct hlist_head),
+ GFP_KERNEL);
+ if (!dp->ports) {
+ err = -ENOMEM;
+ goto err_destroy_percpu;
+ }
- port.port = dp_port->port_no;
+ for (i = 0; i < DP_VPORT_HASH_BUCKETS; i++)
+ INIT_HLIST_HEAD(&dp->ports[i]);
-error_unlock:
- rcu_read_unlock();
- vport_unlock();
+ /* 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;
+ parms.upcall_pid = nla_get_u32(a[OVS_DP_ATTR_UPCALL_PID]);
+
+ vport = new_vport(&parms);
+ if (IS_ERR(vport)) {
+ err = PTR_ERR(vport);
+ if (err == -EBUSY)
+ err = -EEXIST;
- if (err)
- return err;
- } else {
- if (port.port >= DP_MAX_PORTS)
- return -EINVAL;
- if (!dp->ports[port.port])
- return -ENOENT;
+ goto err_destroy_ports_array;
}
- return put_port(dp->ports[port.port], uport);
-}
-
-static int do_list_ports(struct datapath *dp, struct odp_port __user *uports,
- int n_ports)
-{
- int idx = 0;
- if (n_ports) {
- struct dp_port *p;
-
- list_for_each_entry_rcu (p, &dp->port_list, node) {
- if (put_port(p, &uports[idx]))
- return -EFAULT;
- if (idx++ >= n_ports)
- break;
- }
- }
- return idx;
-}
+ 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;
-static int list_ports(struct datapath *dp, struct odp_portvec __user *upv)
-{
- struct odp_portvec pv;
- int retval;
+ ovs_net = net_generic(ovs_dp_get_net(dp), ovs_net_id);
+ list_add_tail(&dp->list_node, &ovs_net->dps);
+ ovs_dp_sysfs_add_dp(dp);
- if (copy_from_user(&pv, upv, sizeof pv))
- return -EFAULT;
+ rtnl_unlock();
- retval = do_list_ports(dp, pv.ports, pv.n_ports);
- if (retval < 0)
- return retval;
+ genl_notify(reply, genl_info_net(info), info->snd_pid,
+ ovs_dp_datapath_multicast_group.id, info->nlhdr,
+ GFP_KERNEL);
+ return 0;
- return put_user(retval, &upv->n_ports);
+err_destroy_local_port:
+ ovs_dp_detach_port(ovs_vport_rtnl(dp, OVSP_LOCAL));
+err_destroy_ports_array:
+ kfree(dp->ports);
+err_destroy_percpu:
+ free_percpu(dp->stats_percpu);
+err_destroy_table:
+ ovs_flow_tbl_destroy(genl_dereference(dp->table));
+err_free_dp:
+ kfree(dp);
+err_unlock_rtnl:
+ rtnl_unlock();
+err:
+ return err;
}
-/* RCU callback for freeing a dp_port_group */
-static void free_port_group(struct rcu_head *rcu)
+/* Called with genl_mutex. */
+static void __dp_destroy(struct datapath *dp)
{
- struct dp_port_group *g = container_of(rcu, struct dp_port_group, rcu);
- kfree(g);
-}
+ int i;
-static int do_set_port_group(struct datapath *dp, u16 __user *ports,
- int n_ports, int group)
-{
- struct dp_port_group *new_group, *old_group;
- int error;
+ rtnl_lock();
- error = -EINVAL;
- if (n_ports > DP_MAX_PORTS || group >= DP_MAX_GROUPS)
- goto error;
+ for (i = 0; i < DP_VPORT_HASH_BUCKETS; i++) {
+ struct vport *vport;
+ struct hlist_node *node, *n;
- error = -ENOMEM;
- new_group = kmalloc(sizeof *new_group + sizeof(u16) * n_ports, GFP_KERNEL);
- if (!new_group)
- goto error;
+ hlist_for_each_entry_safe(vport, node, n, &dp->ports[i], dp_hash_node)
+ if (vport->port_no != OVSP_LOCAL)
+ ovs_dp_detach_port(vport);
+ }
- new_group->n_ports = n_ports;
- error = -EFAULT;
- if (copy_from_user(new_group->ports, ports, sizeof(u16) * n_ports))
- goto error_free;
+ ovs_dp_sysfs_del_dp(dp);
+ list_del(&dp->list_node);
+ ovs_dp_detach_port(ovs_vport_rtnl(dp, OVSP_LOCAL));
- old_group = rcu_dereference(dp->groups[group]);
- rcu_assign_pointer(dp->groups[group], new_group);
- if (old_group)
- call_rcu(&old_group->rcu, free_port_group);
- return 0;
+ /* 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();
-error_free:
- kfree(new_group);
-error:
- return error;
+ call_rcu(&dp->rcu, destroy_dp_rcu);
}
-static int set_port_group(struct datapath *dp,
- const struct odp_port_group __user *upg)
+static int ovs_dp_cmd_del(struct sk_buff *skb, struct genl_info *info)
{
- struct odp_port_group pg;
-
- if (copy_from_user(&pg, upg, sizeof pg))
- return -EFAULT;
+ struct sk_buff *reply;
+ struct datapath *dp;
+ int err;
- return do_set_port_group(dp, pg.ports, pg.n_ports, pg.group);
-}
+ err = ovs_dp_cmd_validate(info->attrs);
+ if (err)
+ return err;
-static int do_get_port_group(struct datapath *dp,
- u16 __user *ports, int n_ports, int group,
- u16 __user *n_portsp)
-{
- struct dp_port_group *g;
- u16 n_copy;
+ dp = lookup_datapath(sock_net(skb->sk), info->userhdr, info->attrs);
+ err = PTR_ERR(dp);
+ if (IS_ERR(dp))
+ return err;
- if (group >= DP_MAX_GROUPS)
- return -EINVAL;
+ 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))
+ return err;
- g = dp->groups[group];
- n_copy = g ? min_t(int, g->n_ports, n_ports) : 0;
- if (n_copy && copy_to_user(ports, g->ports, n_copy * sizeof(u16)))
- return -EFAULT;
+ __dp_destroy(dp);
- if (put_user(g ? g->n_ports : 0, n_portsp))
- return -EFAULT;
+ genl_notify(reply, genl_info_net(info), info->snd_pid,
+ ovs_dp_datapath_multicast_group.id, info->nlhdr,
+ GFP_KERNEL);
return 0;
}
-static int get_port_group(struct datapath *dp, struct odp_port_group __user *upg)
+static int ovs_dp_cmd_set(struct sk_buff *skb, struct genl_info *info)
{
- struct odp_port_group pg;
+ struct sk_buff *reply;
+ struct datapath *dp;
+ int err;
- if (copy_from_user(&pg, upg, sizeof pg))
- return -EFAULT;
+ err = ovs_dp_cmd_validate(info->attrs);
+ if (err)
+ return err;
- return do_get_port_group(dp, pg.ports, pg.n_ports, pg.group, &upg->n_ports);
-}
+ dp = lookup_datapath(sock_net(skb->sk), info->userhdr, info->attrs);
+ if (IS_ERR(dp))
+ return PTR_ERR(dp);
-static int get_listen_mask(const struct file *f)
-{
- return (long)f->private_data;
-}
+ 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(GENL_SOCK(sock_net(skb->sk)), 0,
+ ovs_dp_datapath_multicast_group.id, err);
+ return 0;
+ }
-static void set_listen_mask(struct file *f, int listen_mask)
-{
- f->private_data = (void*)(long)listen_mask;
+ genl_notify(reply, genl_info_net(info), info->snd_pid,
+ ovs_dp_datapath_multicast_group.id, info->nlhdr,
+ GFP_KERNEL);
+
+ return 0;
}
-static long openvswitch_ioctl(struct file *f, unsigned int cmd,
- unsigned long argp)
+static int ovs_dp_cmd_get(struct sk_buff *skb, struct genl_info *info)
{
- int dp_idx = iminor(f->f_dentry->d_inode);
+ struct sk_buff *reply;
struct datapath *dp;
- int drop_frags, listeners, port_no;
- unsigned int sflow_probability;
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_ATTACH:
- err = attach_port(dp_idx, (struct odp_port __user *)argp);
- goto exit;
-
- case ODP_PORT_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);
- goto exit;
-
- case ODP_VPORT_STATS_GET:
- err = vport_user_stats_get((struct odp_vport_stats_req __user *)argp);
- goto exit;
-
- case ODP_VPORT_STATS_SET:
- err = vport_user_stats_set((struct odp_vport_stats_req __user *)argp);
- goto exit;
+ err = ovs_dp_cmd_validate(info->attrs);
+ if (err)
+ return err;
- case ODP_VPORT_ETHER_GET:
- err = vport_user_ether_get((struct odp_vport_ether __user *)argp);
- goto exit;
+ dp = lookup_datapath(sock_net(skb->sk), info->userhdr, info->attrs);
+ if (IS_ERR(dp))
+ return PTR_ERR(dp);
- case ODP_VPORT_ETHER_SET:
- err = vport_user_ether_set((struct odp_vport_ether __user *)argp);
- goto exit;
+ 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);
- case ODP_VPORT_MTU_GET:
- err = vport_user_mtu_get((struct odp_vport_mtu __user *)argp);
- goto exit;
+ return genlmsg_reply(reply, info);
+}
- case ODP_VPORT_MTU_SET:
- err = vport_user_mtu_set((struct odp_vport_mtu __user *)argp);
- goto exit;
+static int ovs_dp_cmd_dump(struct sk_buff *skb, struct netlink_callback *cb)
+{
+ struct ovs_net *ovs_net = net_generic(sock_net(skb->sk), ovs_net_id);
+ struct datapath *dp;
+ int skip = cb->args[0];
+ int i = 0;
+
+ list_for_each_entry(dp, &ovs_net->dps, list_node) {
+ if (i >= skip &&
+ 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++;
}
- dp = get_dp_locked(dp_idx);
- err = -ENODEV;
- if (!dp)
- goto exit;
-
- switch (cmd) {
- case ODP_DP_STATS:
- err = get_dp_stats(dp, (struct odp_stats __user *)argp);
- break;
-
- case ODP_GET_DROP_FRAGS:
- err = put_user(dp->drop_frags, (int __user *)argp);
- break;
-
- 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;
+ cb->args[0] = i;
+
+ return skb->len;
+}
+
+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,
+ },
+};
- case ODP_GET_LISTEN_MASK:
- err = put_user(get_listen_mask(f), (int __user *)argp);
- break;
+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_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_PORT_NO] = { .type = NLA_U32 },
+ [OVS_VPORT_ATTR_TYPE] = { .type = NLA_U32 },
+ [OVS_VPORT_ATTR_UPCALL_PID] = { .type = NLA_U32 },
+ [OVS_VPORT_ATTR_OPTIONS] = { .type = NLA_NESTED },
+};
- 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;
+static struct genl_family dp_vport_genl_family = {
+ .id = GENL_ID_GENERATE,
+ .hdrsize = sizeof(struct ovs_header),
+ .name = OVS_VPORT_FAMILY,
+ .version = OVS_VPORT_VERSION,
+ .maxattr = OVS_VPORT_ATTR_MAX,
+ SET_NETNSOK
+};
- case ODP_GET_SFLOW_PROBABILITY:
- err = put_user(dp->sflow_probability, (unsigned int __user *)argp);
- break;
+struct genl_multicast_group ovs_dp_vport_multicast_group = {
+ .name = OVS_VPORT_MCGROUP
+};
- case ODP_SET_SFLOW_PROBABILITY:
- err = get_user(sflow_probability, (unsigned int __user *)argp);
- if (!err)
- dp->sflow_probability = sflow_probability;
- break;
+/* 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 ovs_header *ovs_header;
+ struct ovs_vport_stats vport_stats;
+ int err;
- case ODP_PORT_QUERY:
- err = query_port(dp, (struct odp_port __user *)argp);
- break;
+ ovs_header = genlmsg_put(skb, pid, seq, &dp_vport_genl_family,
+ flags, cmd);
+ if (!ovs_header)
+ return -EMSGSIZE;
- case ODP_PORT_LIST:
- err = list_ports(dp, (struct odp_portvec __user *)argp);
- break;
+ ovs_header->dp_ifindex = get_dpifindex(vport->dp);
- case ODP_PORT_GROUP_SET:
- err = set_port_group(dp, (struct odp_port_group __user *)argp);
- break;
+ NLA_PUT_U32(skb, OVS_VPORT_ATTR_PORT_NO, vport->port_no);
+ NLA_PUT_U32(skb, OVS_VPORT_ATTR_TYPE, vport->ops->type);
+ NLA_PUT_STRING(skb, OVS_VPORT_ATTR_NAME, vport->ops->get_name(vport));
+ NLA_PUT_U32(skb, OVS_VPORT_ATTR_UPCALL_PID, vport->upcall_pid);
- case ODP_PORT_GROUP_GET:
- err = get_port_group(dp, (struct odp_port_group __user *)argp);
- break;
+ ovs_vport_get_stats(vport, &vport_stats);
+ NLA_PUT(skb, OVS_VPORT_ATTR_STATS, sizeof(struct ovs_vport_stats),
+ &vport_stats);
- case ODP_FLOW_FLUSH:
- err = flush_flows(dp);
- break;
+ NLA_PUT(skb, OVS_VPORT_ATTR_ADDRESS, ETH_ALEN,
+ vport->ops->get_addr(vport));
- case ODP_FLOW_PUT:
- err = put_flow(dp, (struct odp_flow_put __user *)argp);
- break;
+ err = ovs_vport_get_options(vport, skb);
+ if (err == -EMSGSIZE)
+ goto error;
- case ODP_FLOW_DEL:
- err = del_flow(dp, (struct odp_flow __user *)argp);
- break;
+ return genlmsg_end(skb, ovs_header);
- case ODP_FLOW_GET:
- err = do_flowvec_ioctl(dp, argp, do_query_flows);
- break;
+nla_put_failure:
+ err = -EMSGSIZE;
+error:
+ genlmsg_cancel(skb, ovs_header);
+ return err;
+}
- case ODP_FLOW_LIST:
- err = do_flowvec_ioctl(dp, argp, do_list_flows);
- break;
+/* 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 sk_buff *skb;
+ int retval;
- case ODP_EXECUTE:
- err = execute_packet(dp, (struct odp_execute __user *)argp);
- break;
+ skb = nlmsg_new(NLMSG_DEFAULT_SIZE, GFP_ATOMIC);
+ if (!skb)
+ return ERR_PTR(-ENOMEM);
- default:
- err = -ENOIOCTLCMD;
- break;
+ retval = ovs_vport_cmd_fill_info(vport, skb, pid, seq, 0, cmd);
+ if (retval < 0) {
+ kfree_skb(skb);
+ return ERR_PTR(retval);
}
- mutex_unlock(&dp->mutex);
-exit:
- return err;
+ return skb;
}
-static int dp_has_packet_of_interest(struct datapath *dp, int listeners)
+static int ovs_vport_cmd_validate(struct nlattr *a[OVS_VPORT_ATTR_MAX + 1])
{
- int i;
- for (i = 0; i < DP_N_QUEUES; i++) {
- if (listeners & (1 << i) && !skb_queue_empty(&dp->queues[i]))
- return 1;
- }
- return 0;
+ return CHECK_NUL_STRING(a[OVS_VPORT_ATTR_NAME], IFNAMSIZ - 1);
}
-#ifdef CONFIG_COMPAT
-static int compat_list_ports(struct datapath *dp, struct compat_odp_portvec __user *upv)
+/* Called with RTNL lock or RCU read lock. */
+static struct vport *lookup_vport(struct net *net,
+ struct ovs_header *ovs_header,
+ struct nlattr *a[OVS_VPORT_ATTR_MAX + 1])
{
- struct compat_odp_portvec pv;
- int retval;
-
- if (copy_from_user(&pv, upv, sizeof pv))
- return -EFAULT;
-
- retval = do_list_ports(dp, compat_ptr(pv.ports), pv.n_ports);
- if (retval < 0)
- return retval;
+ struct datapath *dp;
+ struct vport *vport;
- return put_user(retval, &upv->n_ports);
-}
+ if (a[OVS_VPORT_ATTR_NAME]) {
+ vport = ovs_vport_locate(net, nla_data(a[OVS_VPORT_ATTR_NAME]));
+ if (!vport)
+ return ERR_PTR(-ENODEV);
+ if (ovs_header->dp_ifindex &&
+ ovs_header->dp_ifindex != get_dpifindex(vport->dp))
+ 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]);
-static int compat_set_port_group(struct datapath *dp, const struct compat_odp_port_group __user *upg)
-{
- struct compat_odp_port_group pg;
+ if (port_no >= DP_MAX_PORTS)
+ return ERR_PTR(-EFBIG);
- if (copy_from_user(&pg, upg, sizeof pg))
- return -EFAULT;
+ dp = get_dp(net, ovs_header->dp_ifindex);
+ if (!dp)
+ return ERR_PTR(-ENODEV);
- return do_set_port_group(dp, compat_ptr(pg.ports), pg.n_ports, pg.group);
+ vport = ovs_vport_rtnl_rcu(dp, port_no);
+ if (!vport)
+ return ERR_PTR(-ENOENT);
+ return vport;
+ } else
+ return ERR_PTR(-EINVAL);
}
-static int compat_get_port_group(struct datapath *dp, struct compat_odp_port_group __user *upg)
+/* Called with RTNL lock. */
+static int change_vport(struct vport *vport,
+ struct nlattr *a[OVS_VPORT_ATTR_MAX + 1])
{
- struct compat_odp_port_group pg;
+ int err = 0;
+
+ if (a[OVS_VPORT_ATTR_STATS])
+ ovs_vport_set_stats(vport, nla_data(a[OVS_VPORT_ATTR_STATS]));
- if (copy_from_user(&pg, upg, sizeof pg))
- return -EFAULT;
+ if (a[OVS_VPORT_ATTR_ADDRESS])
+ err = ovs_vport_set_addr(vport, nla_data(a[OVS_VPORT_ATTR_ADDRESS]));
- return do_get_port_group(dp, compat_ptr(pg.ports), pg.n_ports,
- pg.group, &upg->n_ports);
+ return err;
}
-static int compat_get_flow(struct odp_flow *flow, const struct compat_odp_flow __user *compat)
+static int ovs_vport_cmd_new(struct sk_buff *skb, struct genl_info *info)
{
- compat_uptr_t actions;
+ 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;
- if (!access_ok(VERIFY_READ, compat, sizeof(struct compat_odp_flow)) ||
- __copy_from_user(&flow->stats, &compat->stats, sizeof(struct odp_flow_stats)) ||
- __copy_from_user(&flow->key, &compat->key, sizeof(struct odp_flow_key)) ||
- __get_user(actions, &compat->actions) ||
- __get_user(flow->n_actions, &compat->n_actions) ||
- __get_user(flow->flags, &compat->flags))
- return -EFAULT;
+ err = -EINVAL;
+ if (!a[OVS_VPORT_ATTR_NAME] || !a[OVS_VPORT_ATTR_TYPE] ||
+ !a[OVS_VPORT_ATTR_UPCALL_PID])
+ goto exit;
- flow->actions = compat_ptr(actions);
- return 0;
-}
+ err = ovs_vport_cmd_validate(a);
+ if (err)
+ goto exit;
-static int compat_put_flow(struct datapath *dp, struct compat_odp_flow_put __user *ufp)
-{
- struct odp_flow_stats stats;
- struct odp_flow_put fp;
- int error;
+ rtnl_lock();
+ dp = get_dp(sock_net(skb->sk), ovs_header->dp_ifindex);
+ err = -ENODEV;
+ if (!dp)
+ goto exit_unlock;
- if (compat_get_flow(&fp.flow, &ufp->flow) ||
- get_user(fp.flags, &ufp->flags))
- return -EFAULT;
+ if (a[OVS_VPORT_ATTR_PORT_NO]) {
+ port_no = nla_get_u32(a[OVS_VPORT_ATTR_PORT_NO]);
- error = do_put_flow(dp, &fp, &stats);
- if (error)
- return error;
+ err = -EFBIG;
+ if (port_no >= DP_MAX_PORTS)
+ goto exit_unlock;
+
+ vport = ovs_vport_rtnl(dp, port_no);
+ err = -EBUSY;
+ if (vport)
+ goto exit_unlock;
+ } else {
+ for (port_no = 1; ; port_no++) {
+ if (port_no >= DP_MAX_PORTS) {
+ err = -EFBIG;
+ goto exit_unlock;
+ }
+ vport = ovs_vport_rtnl(dp, port_no);
+ if (!vport)
+ break;
+ }
+ }
- if (copy_to_user(&ufp->flow.stats, &stats,
- sizeof(struct odp_flow_stats)))
- return -EFAULT;
+ 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;
+ parms.upcall_pid = nla_get_u32(a[OVS_VPORT_ATTR_UPCALL_PID]);
+
+ vport = new_vport(&parms);
+ err = PTR_ERR(vport);
+ if (IS_ERR(vport))
+ goto exit_unlock;
+
+ ovs_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) {
+ ovs_dp_detach_port(vport);
+ goto exit_unlock;
+ }
+ genl_notify(reply, genl_info_net(info), info->snd_pid,
+ ovs_dp_vport_multicast_group.id, info->nlhdr, GFP_KERNEL);
- return 0;
+exit_unlock:
+ rtnl_unlock();
+exit:
+ return err;
}
-static int compat_answer_query(struct sw_flow *flow, u32 query_flags,
- struct timespec time_offset,
- struct compat_odp_flow __user *ufp)
+static int ovs_vport_cmd_set(struct sk_buff *skb, struct genl_info *info)
{
- compat_uptr_t actions;
+ struct nlattr **a = info->attrs;
+ struct sk_buff *reply;
+ struct vport *vport;
+ int err;
- if (get_user(actions, &ufp->actions))
- return -EFAULT;
+ err = ovs_vport_cmd_validate(a);
+ if (err)
+ goto exit;
- return do_answer_query(flow, query_flags, time_offset, &ufp->stats,
- compat_ptr(actions), &ufp->n_actions);
-}
+ rtnl_lock();
+ vport = lookup_vport(sock_net(skb->sk), info->userhdr, a);
+ err = PTR_ERR(vport);
+ if (IS_ERR(vport))
+ goto exit_unlock;
-static int compat_del_flow(struct datapath *dp, struct compat_odp_flow __user *ufp)
-{
- struct sw_flow *flow;
- struct odp_flow uf;
- int error;
+ err = 0;
+ if (a[OVS_VPORT_ATTR_TYPE] &&
+ nla_get_u32(a[OVS_VPORT_ATTR_TYPE]) != vport->ops->type)
+ err = -EINVAL;
- if (compat_get_flow(&uf, ufp))
- return -EFAULT;
+ if (!err && a[OVS_VPORT_ATTR_OPTIONS])
+ err = ovs_vport_set_options(vport, a[OVS_VPORT_ATTR_OPTIONS]);
+ if (!err)
+ err = change_vport(vport, a);
+ if (!err && a[OVS_VPORT_ATTR_UPCALL_PID])
+ vport->upcall_pid = nla_get_u32(a[OVS_VPORT_ATTR_UPCALL_PID]);
+
+ 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(GENL_SOCK(sock_net(skb->sk)), 0,
+ ovs_dp_vport_multicast_group.id, err);
+ return 0;
+ }
- flow = do_del_flow(dp, &uf.key);
- if (IS_ERR(flow))
- return PTR_ERR(flow);
+ genl_notify(reply, genl_info_net(info), info->snd_pid,
+ ovs_dp_vport_multicast_group.id, info->nlhdr, GFP_KERNEL);
- error = compat_answer_query(flow, 0, get_time_offset(), ufp);
- flow_deferred_free(flow);
- return error;
+exit_unlock:
+ rtnl_unlock();
+exit:
+ return err;
}
-static int compat_query_flows(struct datapath *dp, struct compat_odp_flow *flows, u32 n_flows)
+static int ovs_vport_cmd_del(struct sk_buff *skb, struct genl_info *info)
{
- struct tbl *table = rcu_dereference(dp->table);
- struct timespec time_offset;
- u32 i;
-
- time_offset = get_time_offset();
+ struct nlattr **a = info->attrs;
+ struct sk_buff *reply;
+ struct vport *vport;
+ int err;
- for (i = 0; i < n_flows; i++) {
- struct compat_odp_flow __user *ufp = &flows[i];
- struct odp_flow uf;
- struct tbl_node *flow_node;
- int error;
+ err = ovs_vport_cmd_validate(a);
+ if (err)
+ goto exit;
- if (compat_get_flow(&uf, ufp))
- return -EFAULT;
- memset(uf.key.reserved, 0, sizeof uf.key.reserved);
+ rtnl_lock();
+ vport = lookup_vport(sock_net(skb->sk), info->userhdr, a);
+ err = PTR_ERR(vport);
+ if (IS_ERR(vport))
+ goto exit_unlock;
- flow_node = tbl_lookup(table, &uf.key, flow_hash(&uf.key), flow_cmp);
- if (!flow_node)
- error = put_user(ENOENT, &ufp->stats.error);
- else
- error = compat_answer_query(flow_cast(flow_node), uf.flags, time_offset, ufp);
- if (error)
- return -EFAULT;
+ if (vport->port_no == OVSP_LOCAL) {
+ err = -EINVAL;
+ goto exit_unlock;
}
- return n_flows;
-}
-struct compat_list_flows_cbdata {
- struct compat_odp_flow __user *uflows;
- u32 n_flows;
- u32 listed_flows;
- struct timespec time_offset;
-};
+ 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;
-static int compat_list_flow(struct tbl_node *node, void *cbdata_)
-{
- struct sw_flow *flow = flow_cast(node);
- struct compat_list_flows_cbdata *cbdata = cbdata_;
- struct compat_odp_flow __user *ufp = &cbdata->uflows[cbdata->listed_flows++];
- int error;
+ ovs_dp_detach_port(vport);
- if (copy_to_user(&ufp->key, &flow->key, sizeof flow->key))
- return -EFAULT;
- error = compat_answer_query(flow, 0, cbdata->time_offset, ufp);
- if (error)
- return error;
+ genl_notify(reply, genl_info_net(info), info->snd_pid,
+ ovs_dp_vport_multicast_group.id, info->nlhdr, GFP_KERNEL);
- if (cbdata->listed_flows >= cbdata->n_flows)
- return cbdata->listed_flows;
- return 0;
+exit_unlock:
+ rtnl_unlock();
+exit:
+ return err;
}
-static int compat_list_flows(struct datapath *dp, struct compat_odp_flow *flows, u32 n_flows)
+static int ovs_vport_cmd_get(struct sk_buff *skb, struct genl_info *info)
{
- struct compat_list_flows_cbdata cbdata;
- int error;
-
- if (!n_flows)
- return 0;
-
- cbdata.uflows = flows;
- cbdata.n_flows = n_flows;
- cbdata.listed_flows = 0;
- cbdata.time_offset = get_time_offset();
-
- error = tbl_foreach(rcu_dereference(dp->table), compat_list_flow, &cbdata);
- return error ? error : cbdata.listed_flows;
-}
+ struct nlattr **a = info->attrs;
+ struct ovs_header *ovs_header = info->userhdr;
+ struct sk_buff *reply;
+ struct vport *vport;
+ int err;
-static int compat_flowvec_ioctl(struct datapath *dp, unsigned long argp,
- int (*function)(struct datapath *,
- struct compat_odp_flow *,
- u32 n_flows))
-{
- struct compat_odp_flowvec __user *uflowvec;
- struct compat_odp_flow __user *flows;
- struct compat_odp_flowvec flowvec;
- int retval;
+ err = ovs_vport_cmd_validate(a);
+ if (err)
+ goto exit;
- uflowvec = compat_ptr(argp);
- if (!access_ok(VERIFY_WRITE, uflowvec, sizeof *uflowvec) ||
- copy_from_user(&flowvec, uflowvec, sizeof flowvec))
- return -EFAULT;
+ rcu_read_lock();
+ vport = lookup_vport(sock_net(skb->sk), ovs_header, a);
+ err = PTR_ERR(vport);
+ if (IS_ERR(vport))
+ goto exit_unlock;
- if (flowvec.n_flows > INT_MAX / sizeof(struct compat_odp_flow))
- return -EINVAL;
+ 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;
- flows = compat_ptr(flowvec.flows);
- if (!access_ok(VERIFY_WRITE, flows,
- flowvec.n_flows * sizeof(struct compat_odp_flow)))
- return -EFAULT;
+ rcu_read_unlock();
- retval = function(dp, flows, flowvec.n_flows);
- return (retval < 0 ? retval
- : retval == flowvec.n_flows ? 0
- : put_user(retval, &uflowvec->n_flows));
-}
+ return genlmsg_reply(reply, info);
-static int compat_execute(struct datapath *dp, const struct compat_odp_execute __user *uexecute)
-{
- struct odp_execute execute;
- compat_uptr_t actions;
- compat_uptr_t data;
-
- if (!access_ok(VERIFY_READ, uexecute, sizeof(struct compat_odp_execute)) ||
- __get_user(execute.in_port, &uexecute->in_port) ||
- __get_user(actions, &uexecute->actions) ||
- __get_user(execute.n_actions, &uexecute->n_actions) ||
- __get_user(data, &uexecute->data) ||
- __get_user(execute.length, &uexecute->length))
- return -EFAULT;
-
- execute.actions = compat_ptr(actions);
- execute.data = compat_ptr(data);
-
- return do_execute(dp, &execute);
+exit_unlock:
+ rcu_read_unlock();
+exit:
+ return err;
}
-static long openvswitch_compat_ioctl(struct file *f, unsigned int cmd, unsigned long argp)
+static int ovs_vport_cmd_dump(struct sk_buff *skb, struct netlink_callback *cb)
{
- int dp_idx = iminor(f->f_dentry->d_inode);
+ struct ovs_header *ovs_header = genlmsg_data(nlmsg_data(cb->nlh));
struct datapath *dp;
- int err;
-
- switch (cmd) {
- case ODP_DP_DESTROY:
- case ODP_FLOW_FLUSH:
- /* Ioctls that don't need any translation at all. */
- 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_MTU_SET:
- case ODP_VPORT_MTU_GET:
- case ODP_VPORT_ETHER_SET:
- case ODP_VPORT_ETHER_GET:
- case ODP_VPORT_STATS_SET:
- case ODP_VPORT_STATS_GET:
- case ODP_DP_STATS:
- case ODP_GET_DROP_FRAGS:
- case ODP_SET_DROP_FRAGS:
- case ODP_SET_LISTEN_MASK:
- case ODP_GET_LISTEN_MASK:
- case ODP_SET_SFLOW_PROBABILITY:
- case ODP_GET_SFLOW_PROBABILITY:
- case ODP_PORT_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));
- }
+ int bucket = cb->args[0], skip = cb->args[1];
+ int i, j = 0;
- dp = get_dp_locked(dp_idx);
- err = -ENODEV;
+ dp = get_dp(sock_net(skb->sk), ovs_header->dp_ifindex);
if (!dp)
- goto exit;
-
- switch (cmd) {
- case ODP_PORT_LIST32:
- err = compat_list_ports(dp, compat_ptr(argp));
- break;
-
- case ODP_PORT_GROUP_SET32:
- err = compat_set_port_group(dp, compat_ptr(argp));
- break;
-
- case ODP_PORT_GROUP_GET32:
- err = compat_get_port_group(dp, compat_ptr(argp));
- break;
+ return -ENODEV;
- case ODP_FLOW_PUT32:
- err = compat_put_flow(dp, compat_ptr(argp));
- break;
+ rcu_read_lock();
+ for (i = bucket; i < DP_VPORT_HASH_BUCKETS; i++) {
+ struct vport *vport;
+ struct hlist_node *n;
+
+ j = 0;
+ hlist_for_each_entry_rcu(vport, n, &dp->ports[i], dp_hash_node) {
+ if (j >= skip &&
+ ovs_vport_cmd_fill_info(vport, skb,
+ NETLINK_CB(cb->skb).pid,
+ cb->nlh->nlmsg_seq,
+ NLM_F_MULTI,
+ OVS_VPORT_CMD_NEW) < 0)
+ goto out;
+
+ j++;
+ }
+ skip = 0;
+ }
+out:
+ rcu_read_unlock();
- case ODP_FLOW_DEL32:
- err = compat_del_flow(dp, compat_ptr(argp));
- break;
+ cb->args[0] = i;
+ cb->args[1] = j;
+
+ return skb->len;
+}
+
+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,
+ },
+};
- case ODP_FLOW_GET32:
- err = compat_flowvec_ioctl(dp, argp, compat_query_flows);
- break;
+struct genl_family_and_ops {
+ struct genl_family *family;
+ struct genl_ops *ops;
+ int n_ops;
+ struct genl_multicast_group *group;
+};
- case ODP_FLOW_LIST32:
- err = compat_flowvec_ioctl(dp, argp, compat_list_flows);
- break;
+static const struct genl_family_and_ops dp_genl_families[] = {
+ { &dp_datapath_genl_family,
+ dp_datapath_genl_ops, ARRAY_SIZE(dp_datapath_genl_ops),
+ &ovs_dp_datapath_multicast_group },
+ { &dp_vport_genl_family,
+ dp_vport_genl_ops, ARRAY_SIZE(dp_vport_genl_ops),
+ &ovs_dp_vport_multicast_group },
+ { &dp_flow_genl_family,
+ dp_flow_genl_ops, ARRAY_SIZE(dp_flow_genl_ops),
+ &ovs_dp_flow_multicast_group },
+ { &dp_packet_genl_family,
+ dp_packet_genl_ops, ARRAY_SIZE(dp_packet_genl_ops),
+ NULL },
+};
- case ODP_EXECUTE32:
- err = compat_execute(dp, compat_ptr(argp));
- break;
+static void dp_unregister_genl(int n_families)
+{
+ int i;
- default:
- err = -ENOIOCTLCMD;
- break;
- }
- mutex_unlock(&dp->mutex);
-exit:
- return err;
+ for (i = 0; i < n_families; i++)
+ genl_unregister_family(dp_genl_families[i].family);
}
-#endif
-/* Unfortunately this function is not exported so this is a verbatim copy
- * from net/core/datagram.c in 2.6.30. */
-static int skb_copy_and_csum_datagram(const struct sk_buff *skb, int offset,
- u8 __user *to, int len,
- __wsum *csump)
+static int dp_register_genl(void)
{
- int start = skb_headlen(skb);
- int pos = 0;
- int i, copy = start - offset;
-
- /* Copy header. */
- if (copy > 0) {
- int err = 0;
- if (copy > len)
- copy = len;
- *csump = csum_and_copy_to_user(skb->data + offset, to, copy,
- *csump, &err);
+ 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 fault;
- if ((len -= copy) == 0)
- return 0;
- offset += copy;
- to += copy;
- pos = copy;
- }
+ goto error;
+ n_registered++;
- for (i = 0; i < skb_shinfo(skb)->nr_frags; i++) {
- int end;
-
- WARN_ON(start > offset + len);
-
- end = start + skb_shinfo(skb)->frags[i].size;
- if ((copy = end - offset) > 0) {
- __wsum csum2;
- int err = 0;
- u8 *vaddr;
- skb_frag_t *frag = &skb_shinfo(skb)->frags[i];
- struct page *page = frag->page;
-
- if (copy > len)
- copy = len;
- vaddr = kmap(page);
- csum2 = csum_and_copy_to_user(vaddr +
- frag->page_offset +
- offset - start,
- to, copy, 0, &err);
- kunmap(page);
+ if (f->group) {
+ err = genl_register_mc_group(f->family, f->group);
if (err)
- goto fault;
- *csump = csum_block_add(*csump, csum2, pos);
- if (!(len -= copy))
- return 0;
- offset += copy;
- to += copy;
- pos += copy;
+ goto error;
}
- start = end;
}
- if (skb_shinfo(skb)->frag_list) {
- struct sk_buff *list = skb_shinfo(skb)->frag_list;
-
- for (; list; list=list->next) {
- int end;
-
- WARN_ON(start > offset + len);
-
- end = start + list->len;
- if ((copy = end - offset) > 0) {
- __wsum csum2 = 0;
- if (copy > len)
- copy = len;
- if (skb_copy_and_csum_datagram(list,
- offset - start,
- to, copy,
- &csum2))
- goto fault;
- *csump = csum_block_add(*csump, csum2, pos);
- if ((len -= copy) == 0)
- return 0;
- offset += copy;
- to += copy;
- pos += copy;
- }
- start = end;
- }
- }
- if (!len)
- return 0;
+ return 0;
-fault:
- return -EFAULT;
+error:
+ dp_unregister_genl(n_registered);
+ return err;
}
-ssize_t openvswitch_read(struct file *f, char __user *buf, size_t nbytes,
- loff_t *ppos)
+static int __rehash_flow_table(void *dummy)
{
- /* 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;
- size_t copy_bytes, tot_copy_bytes;
- int retval;
-
- if (!dp)
- return -ENODEV;
+ struct datapath *dp;
+ struct net *net;
- if (nbytes == 0 || !listeners)
- return 0;
+ rtnl_lock();
+ for_each_net(net) {
+ struct ovs_net *ovs_net = net_generic(net, ovs_net_id);
- for (;;) {
- int i;
+ list_for_each_entry(dp, &ovs_net->dps, list_node) {
+ struct flow_table *old_table = genl_dereference(dp->table);
+ struct flow_table *new_table;
- for (i = 0; i < DP_N_QUEUES; i++) {
- if (listeners & (1 << i)) {
- skb = skb_dequeue(&dp->queues[i]);
- if (skb)
- goto success;
+ new_table = ovs_flow_tbl_rehash(old_table);
+ if (!IS_ERR(new_table)) {
+ rcu_assign_pointer(dp->table, new_table);
+ ovs_flow_tbl_deferred_destroy(old_table);
}
}
-
- if (f->f_flags & O_NONBLOCK) {
- retval = -EAGAIN;
- goto error;
- }
-
- wait_event_interruptible(dp->waitqueue,
- dp_has_packet_of_interest(dp,
- listeners));
-
- if (signal_pending(current)) {
- retval = -ERESTARTSYS;
- goto error;
- }
}
-success:
- copy_bytes = tot_copy_bytes = min_t(size_t, skb->len, nbytes);
-
- retval = 0;
- 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
- BUG_ON(csum_start >= skb_headlen(skb));
- retval = skb_copy_and_csum_datagram(skb, csum_start, buf + csum_start,
- copy_bytes - csum_start, &csum);
- if (!retval) {
- __sum16 __user *csump;
-
- copy_bytes = csum_start;
- csump = (__sum16 __user *)(buf + csum_start + csum_offset);
+ rtnl_unlock();
+ return 0;
+}
- BUG_ON((char *)csump + sizeof(__sum16) > buf + nbytes);
- put_user(csum_fold(csum), csump);
- }
- } else
- retval = skb_checksum_help(skb);
- }
+static void rehash_flow_table(struct work_struct *work)
+{
+ genl_exec(__rehash_flow_table, NULL);
+ schedule_delayed_work(&rehash_flow_wq, REHASH_FLOW_INTERVAL);
+}
- if (!retval) {
- struct iovec __user iov;
+static int dp_destroy_all(void *data)
+{
+ struct datapath *dp, *dp_next;
+ struct ovs_net *ovs_net = data;
- iov.iov_base = buf;
- iov.iov_len = copy_bytes;
- retval = skb_copy_datagram_iovec(skb, 0, &iov, iov.iov_len);
- }
+ list_for_each_entry_safe(dp, dp_next, &ovs_net->dps, list_node)
+ __dp_destroy(dp);
- if (!retval)
- retval = tot_copy_bytes;
+ return 0;
+}
- kfree_skb(skb);
+static int __net_init ovs_init_net(struct net *net)
+{
+ struct ovs_net *ovs_net = net_generic(net, ovs_net_id);
-error:
- return retval;
+ INIT_LIST_HEAD(&ovs_net->dps);
+ return 0;
}
-static unsigned int openvswitch_poll(struct file *file, poll_table *wait)
+static void __net_exit ovs_exit_net(struct net *net)
{
- /* 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;
+ struct ovs_net *ovs_net = net_generic(net, ovs_net_id);
+
+ genl_exec(dp_destroy_all, ovs_net);
}
-struct file_operations openvswitch_fops = {
- /* XXX .aio_read = openvswitch_aio_read, */
- .read = openvswitch_read,
- .poll = openvswitch_poll,
- .unlocked_ioctl = openvswitch_ioctl,
-#ifdef CONFIG_COMPAT
- .compat_ioctl = openvswitch_compat_ioctl,
-#endif
- /* XXX .fasync = openvswitch_fasync, */
+static struct pernet_operations ovs_net_ops = {
+ .init = ovs_init_net,
+ .exit = ovs_exit_net,
+ .id = &ovs_net_id,
+ .size = sizeof(struct ovs_net),
};
-static int major;
-
static int __init dp_init(void)
{
struct sk_buff *dummy_skb;
BUILD_BUG_ON(sizeof(struct ovs_skb_cb) > sizeof(dummy_skb->cb));
- printk("Open vSwitch %s, built "__DATE__" "__TIME__"\n", VERSION BUILDNR);
+ pr_info("Open vSwitch switching datapath %s, built "__DATE__" "__TIME__"\n",
+ VERSION BUILDNR);
- err = flow_init();
+ err = genl_exec_init();
if (err)
goto error;
- err = vport_init();
+ err = ovs_workqueues_init();
+ if (err)
+ goto error_genl_exec;
+
+ err = ovs_tnl_init();
+ if (err)
+ goto error_wq;
+
+ err = ovs_flow_init();
+ if (err)
+ goto error_tnl_exit;
+
+ err = ovs_vport_init();
if (err)
goto error_flow_exit;
- err = register_netdevice_notifier(&dp_device_notifier);
+ err = register_pernet_device(&ovs_net_ops);
if (err)
goto error_vport_exit;
- major = register_chrdev(0, "openvswitch", &openvswitch_fops);
+ err = register_netdevice_notifier(&ovs_dp_device_notifier);
+ if (err)
+ goto error_netns_exit;
+
+ err = dp_register_genl();
if (err < 0)
goto error_unreg_notifier;
+ schedule_delayed_work(&rehash_flow_wq, REHASH_FLOW_INTERVAL);
+
return 0;
error_unreg_notifier:
- unregister_netdevice_notifier(&dp_device_notifier);
+ unregister_netdevice_notifier(&ovs_dp_device_notifier);
+error_netns_exit:
+ unregister_pernet_device(&ovs_net_ops);
error_vport_exit:
- vport_exit();
+ ovs_vport_exit();
error_flow_exit:
- flow_exit();
+ ovs_flow_exit();
+error_tnl_exit:
+ ovs_tnl_exit();
+error_wq:
+ ovs_workqueues_exit();
+error_genl_exec:
+ genl_exec_exit();
error:
return err;
}
static void dp_cleanup(void)
{
+ cancel_delayed_work_sync(&rehash_flow_wq);
+ dp_unregister_genl(ARRAY_SIZE(dp_genl_families));
+ unregister_netdevice_notifier(&ovs_dp_device_notifier);
+ unregister_pernet_device(&ovs_net_ops);
rcu_barrier();
- unregister_chrdev(major, "openvswitch");
- unregister_netdevice_notifier(&dp_device_notifier);
- vport_exit();
- flow_exit();
+ ovs_vport_exit();
+ ovs_flow_exit();
+ ovs_tnl_exit();
+ ovs_workqueues_exit();
+ genl_exec_exit();
}
module_init(dp_init);