#include "datapath.h"
#include "actions.h"
#include "flow.h"
+#include "loop_counter.h"
#include "odp-compat.h"
#include "table.h"
#include "vport-internal_dev.h"
#include "compat.h"
-
int (*dp_ioctl_hook)(struct net_device *dev, struct ifreq *rq, int cmd);
EXPORT_SYMBOL(dp_ioctl_hook);
* dp_mutex nests inside the RTNL lock: if you need both you must take the RTNL
* lock first.
*
- * It is safe to access the datapath and dp_port structures with just
+ * It is safe to access the datapath and vport structures with just
* dp_mutex.
*/
static struct datapath *dps[ODP_MAX];
static DEFINE_MUTEX(dp_mutex);
-/* 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
-
-/* 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];
-};
-
-static DEFINE_PER_CPU(struct percpu_loop_counters, dp_loop_counters);
-
-static int new_dp_port(struct datapath *, struct odp_port *, int port_no);
+static int new_vport(struct datapath *, struct odp_port *, int port_no);
/* Must be called with rcu_read_lock or dp_mutex. */
struct datapath *get_dp(int dp_idx)
/* Must be called with rcu_read_lock or RTNL lock. */
const char *dp_name(const struct datapath *dp)
{
- return vport_get_name(dp->ports[ODPP_LOCAL]->vport);
+ return vport_get_name(dp->ports[ODPP_LOCAL]);
}
static inline size_t br_nlmsg_size(void)
}
static int dp_fill_ifinfo(struct sk_buff *skb,
- const struct dp_port *port,
+ const struct vport *port,
int event, unsigned int flags)
{
const struct datapath *dp = port->dp;
- int ifindex = vport_get_ifindex(port->vport);
- int iflink = vport_get_iflink(port->vport);
+ int ifindex = vport_get_ifindex(port);
+ int iflink = vport_get_iflink(port);
struct ifinfomsg *hdr;
struct nlmsghdr *nlh;
hdr->__ifi_pad = 0;
hdr->ifi_type = ARPHRD_ETHER;
hdr->ifi_index = ifindex;
- hdr->ifi_flags = vport_get_flags(port->vport);
+ hdr->ifi_flags = vport_get_flags(port);
hdr->ifi_change = 0;
- NLA_PUT_STRING(skb, IFLA_IFNAME, vport_get_name(port->vport));
- NLA_PUT_U32(skb, IFLA_MASTER, vport_get_ifindex(dp->ports[ODPP_LOCAL]->vport));
- NLA_PUT_U32(skb, IFLA_MTU, vport_get_mtu(port->vport));
+ NLA_PUT_STRING(skb, IFLA_IFNAME, vport_get_name(port));
+ NLA_PUT_U32(skb, IFLA_MASTER, vport_get_ifindex(dp->ports[ODPP_LOCAL]));
+ NLA_PUT_U32(skb, IFLA_MTU, vport_get_mtu(port));
#ifdef IFLA_OPERSTATE
NLA_PUT_U8(skb, IFLA_OPERSTATE,
- vport_is_running(port->vport)
- ? vport_get_operstate(port->vport)
+ vport_is_running(port)
+ ? vport_get_operstate(port)
: IF_OPER_DOWN);
#endif
- NLA_PUT(skb, IFLA_ADDRESS, ETH_ALEN,
- vport_get_addr(port->vport));
+ NLA_PUT(skb, IFLA_ADDRESS, ETH_ALEN, vport_get_addr(port));
if (ifindex != iflink)
NLA_PUT_U32(skb, IFLA_LINK,iflink);
return -EMSGSIZE;
}
-static void dp_ifinfo_notify(int event, struct dp_port *port)
+static void dp_ifinfo_notify(int event, struct vport *port)
{
struct sk_buff *skb;
int err = -ENOBUFS;
/* Set up our datapath device. */
BUILD_BUG_ON(sizeof(internal_dev_port.devname) != sizeof(devname));
strcpy(internal_dev_port.devname, devname);
- internal_dev_port.flags = ODP_PORT_INTERNAL;
- err = new_dp_port(dp, &internal_dev_port, ODPP_LOCAL);
+ strcpy(internal_dev_port.type, "internal");
+ err = new_vport(dp, &internal_dev_port, ODPP_LOCAL);
if (err) {
if (err == -EBUSY)
err = -EEXIST;
return 0;
err_destroy_local_port:
- dp_detach_port(dp->ports[ODPP_LOCAL], 1);
+ dp_detach_port(dp->ports[ODPP_LOCAL]);
err_destroy_table:
tbl_destroy(dp->table, NULL);
err_free_dp:
static void do_destroy_dp(struct datapath *dp)
{
- struct dp_port *p, *n;
+ struct vport *p, *n;
int i;
list_for_each_entry_safe (p, n, &dp->port_list, node)
if (p->port_no != ODPP_LOCAL)
- dp_detach_port(p, 1);
+ dp_detach_port(p);
dp_sysfs_del_dp(dp);
rcu_assign_pointer(dps[dp->dp_idx], NULL);
- dp_detach_port(dp->ports[ODPP_LOCAL], 1);
+ dp_detach_port(dp->ports[ODPP_LOCAL]);
tbl_destroy(dp->table, flow_free_tbl);
return err;
}
-static void release_dp_port(struct kobject *kobj)
-{
- struct dp_port *p = container_of(kobj, struct dp_port, kobj);
- kfree(p);
-}
-
-static struct kobj_type brport_ktype = {
-#ifdef CONFIG_SYSFS
- .sysfs_ops = &brport_sysfs_ops,
-#endif
- .release = release_dp_port
-};
-
/* Called with RTNL lock and dp_mutex. */
-static int new_dp_port(struct datapath *dp, struct odp_port *odp_port, int port_no)
+static int new_vport(struct datapath *dp, struct odp_port *odp_port, int port_no)
{
+ struct vport_parms parms;
struct vport *vport;
- struct dp_port *p;
- int err;
- vport = vport_locate(odp_port->devname);
- if (!vport) {
- vport_lock();
-
- 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);
- }
+ parms.name = odp_port->devname;
+ parms.type = odp_port->type;
+ parms.config = odp_port->config;
+ parms.dp = dp;
+ parms.port_no = port_no;
- p = kzalloc(sizeof(*p), GFP_KERNEL);
- if (!p)
- return -ENOMEM;
+ vport_lock();
+ vport = vport_add(&parms);
+ vport_unlock();
- p->port_no = port_no;
- p->dp = dp;
- p->vport = vport;
- atomic_set(&p->sflow_pool, 0);
+ if (IS_ERR(vport))
+ return PTR_ERR(vport);
- err = vport_attach(vport, p);
- if (err) {
- kfree(p);
- return err;
- }
-
- rcu_assign_pointer(dp->ports[port_no], p);
- list_add_rcu(&p->node, &dp->port_list);
+ rcu_assign_pointer(dp->ports[port_no], vport);
+ list_add_rcu(&vport->node, &dp->port_list);
dp->n_ports++;
- /* Initialize kobject for bridge. This will be added as
- * /sys/class/net/<devname>/brport later, if sysfs is enabled. */
- p->kobj.kset = NULL;
- kobject_init(&p->kobj, &brport_ktype);
-
- dp_ifinfo_notify(RTM_NEWLINK, p);
+ dp_ifinfo_notify(RTM_NEWLINK, vport);
return 0;
}
if (copy_from_user(&port, portp, sizeof port))
goto out;
port.devname[IFNAMSIZ - 1] = '\0';
+ port.type[VPORT_TYPE_SIZE - 1] = '\0';
rtnl_lock();
dp = get_dp_locked(dp_idx);
goto out_unlock_dp;
got_port_no:
- err = new_dp_port(dp, &port, port_no);
+ err = new_vport(dp, &port, port_no);
if (err)
goto out_unlock_dp;
return err;
}
-int dp_detach_port(struct dp_port *p, int may_delete)
+int dp_detach_port(struct vport *p)
{
- struct vport *vport = p->vport;
int err;
ASSERT_RTNL();
list_del_rcu(&p->node);
rcu_assign_pointer(p->dp->ports[p->port_no], NULL);
- err = vport_detach(vport);
- if (err)
- return err;
-
- /* Then wait until no one is still using it, and destroy it. */
- synchronize_rcu();
-
- if (may_delete) {
- const char *port_type = vport_get_type(vport);
-
- if (!strcmp(port_type, "netdev") || !strcmp(port_type, "internal")) {
- vport_lock();
- vport_del(vport);
- vport_unlock();
- }
- }
-
- kobject_put(&p->kobj);
+ /* Then destroy it. */
+ vport_lock();
+ err = vport_del(p);
+ vport_unlock();
- return 0;
+ return err;
}
static int detach_port(int dp_idx, int port_no)
{
- struct dp_port *p;
+ struct vport *p;
struct datapath *dp;
int err;
if (!p)
goto out_unlock_dp;
- err = dp_detach_port(p, 1);
+ err = dp_detach_port(p);
out_unlock_dp:
mutex_unlock(&dp->mutex);
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)
+void dp_process_received_packet(struct vport *p, struct sk_buff *skb)
{
struct datapath *dp = p->dp;
struct dp_stats_percpu *stats;
struct loop_counter *loop;
int error;
- OVS_CB(skb)->dp_port = p;
+ OVS_CB(skb)->vport = p;
if (!OVS_CB(skb)->flow) {
struct odp_flow_key key;
acts = rcu_dereference(OVS_CB(skb)->flow->sf_acts);
/* 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 = loop_get_counter();
+ if (unlikely(++loop->count > MAX_LOOPS))
loop->looping = true;
if (unlikely(loop->looping)) {
- suppress_loop(dp, acts);
+ loop_suppress(dp, acts);
goto out_loop;
}
/* Check whether sub-actions looped too much. */
if (unlikely(loop->looping))
- suppress_loop(dp, acts);
+ loop_suppress(dp, acts);
out_loop:
/* Decrement loop counter. */
if (!--loop->count)
loop->looping = false;
- put_cpu_var(dp_loop_counters);
+ loop_put_counter();
out:
/* Update datapath statistics. */
int port_no;
int err;
- if (OVS_CB(skb)->dp_port)
- port_no = OVS_CB(skb)->dp_port->port_no;
+ if (OVS_CB(skb)->vport)
+ port_no = OVS_CB(skb)->vport->port_no;
else
port_no = ODPP_LOCAL;
for (i = 0; i < actions->n_actions; i++) {
const union odp_action *a = &actions->actions[i];
- __be16 mask;
switch (a->type) {
case ODPAT_CONTROLLER:
break;
case ODPAT_SET_DL_TCI:
- mask = a->dl_tci.mask;
- if (mask != htons(VLAN_VID_MASK) &&
- mask != htons(VLAN_PCP_MASK) &&
- mask != htons(VLAN_VID_MASK | VLAN_PCP_MASK))
- return -EINVAL;
- if (a->dl_tci.tci & ~mask)
+ if (a->dl_tci.tci & htons(VLAN_CFI_MASK))
return -EINVAL;
break;
return ERR_PTR(error);
}
-static struct timespec get_time_offset(void)
-{
- struct timespec now_mono, now_jiffies;
-
- ktime_get_ts(&now_mono);
- jiffies_to_timespec(jiffies, &now_jiffies);
- return timespec_sub(now_mono, now_jiffies);
-}
-
-static void get_stats(struct sw_flow *flow, struct odp_flow_stats *stats,
- struct timespec time_offset)
+static void get_stats(struct sw_flow *flow, struct odp_flow_stats *stats)
{
if (flow->used) {
- struct timespec flow_ts, used;
+ struct timespec offset_ts, used, now_mono;
- 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);
+ ktime_get_ts(&now_mono);
+ jiffies_to_timespec(jiffies - flow->used, &offset_ts);
+ set_normalized_timespec(&used, now_mono.tv_sec - offset_ts.tv_sec,
+ now_mono.tv_nsec - offset_ts.tv_nsec);
stats->used_sec = used.tv_sec;
stats->used_nsec = used.tv_nsec;
/* Fetch stats, then clear them if necessary. */
spin_lock_bh(&flow->lock);
- get_stats(flow, stats, get_time_offset());
+ get_stats(flow, stats);
if (uf->flags & ODPPF_ZERO_STATS)
clear_stats(flow);
spin_unlock_bh(&flow->lock);
}
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)
u32 n_actions;
spin_lock_bh(&flow->lock);
- get_stats(flow, &stats, time_offset);
+ get_stats(flow, &stats);
if (query_flags & ODPFF_ZERO_TCP_FLAGS)
flow->tcp_flags = 0;
}
static int answer_query(struct sw_flow *flow, u32 query_flags,
- struct timespec time_offset,
struct odp_flow __user *ufp)
{
union odp_action *actions;
if (get_user(actions, &ufp->actions))
return -EFAULT;
- return do_answer_query(flow, query_flags, time_offset,
+ return do_answer_query(flow, query_flags,
&ufp->stats, actions, &ufp->n_actions);
}
if (IS_ERR(flow))
return PTR_ERR(flow);
- error = answer_query(flow, 0, get_time_offset(), ufp);
+ error = answer_query(flow, 0, ufp);
flow_deferred_free(flow);
return error;
}
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;
- time_offset = get_time_offset();
-
for (i = 0; i < flowvec->n_flows; i++) {
struct odp_flow __user *ufp = &flowvec->flows[i];
struct odp_flow uf;
if (!flow_node)
error = put_user(ENOENT, &ufp->stats.error);
else
- error = answer_query(flow_cast(flow_node), uf.flags, time_offset, ufp);
+ error = answer_query(flow_cast(flow_node), uf.flags, ufp);
if (error)
return -EFAULT;
}
struct odp_flow __user *uflows;
u32 n_flows;
u32 listed_flows;
- struct timespec time_offset;
};
static int list_flow(struct tbl_node *node, void *cbdata_)
if (copy_to_user(&ufp->key, &flow->key, sizeof flow->key))
return -EFAULT;
- error = answer_query(flow, 0, cbdata->time_offset, ufp);
+ error = answer_query(flow, 0, ufp);
if (error)
return error;
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;
/* MTU of the dp pseudo-device: ETH_DATA_LEN or the minimum of the ports */
int dp_min_mtu(const struct datapath *dp)
{
- struct dp_port *p;
+ struct vport *p;
int mtu = 0;
ASSERT_RTNL();
/* Skip any internal ports, since that's what we're trying to
* set. */
- if (is_internal_vport(p->vport))
+ if (is_internal_vport(p))
continue;
- dev_mtu = vport_get_mtu(p->vport);
+ dev_mtu = vport_get_mtu(p);
if (!mtu || dev_mtu < mtu)
mtu = dev_mtu;
}
* be called with RTNL lock. */
void set_internal_devs_mtu(const struct datapath *dp)
{
- struct dp_port *p;
+ struct vport *p;
int mtu;
ASSERT_RTNL();
mtu = dp_min_mtu(dp);
list_for_each_entry_rcu (p, &dp->port_list, node) {
- if (is_internal_vport(p->vport))
- vport_set_mtu(p->vport, mtu);
+ if (is_internal_vport(p))
+ vport_set_mtu(p, mtu);
}
}
-static int put_port(const struct dp_port *p, struct odp_port __user *uop)
+static int put_port(const struct vport *p, struct odp_port __user *uop)
{
struct odp_port op;
memset(&op, 0, sizeof op);
rcu_read_lock();
- strncpy(op.devname, vport_get_name(p->vport), sizeof op.devname);
+ strncpy(op.devname, vport_get_name(p), sizeof op.devname);
+ strncpy(op.type, vport_get_type(p), sizeof op.type);
rcu_read_unlock();
op.port = p->port_no;
- op.flags = is_internal_vport(p->vport) ? ODP_PORT_INTERNAL : 0;
return copy_to_user(uop, &op, sizeof op) ? -EFAULT : 0;
}
if (port.devname[0]) {
struct vport *vport;
- struct dp_port *dp_port;
int err = 0;
port.devname[IFNAMSIZ - 1] = '\0';
err = -ENODEV;
goto error_unlock;
}
-
- dp_port = vport_get_dp_port(vport);
- if (!dp_port || dp_port->dp != dp) {
+ if (vport->dp != dp) {
err = -ENOENT;
goto error_unlock;
}
- port.port = dp_port->port_no;
+ port.port = vport->port_no;
error_unlock:
rcu_read_unlock();
{
int idx = 0;
if (n_ports) {
- struct dp_port *p;
+ struct vport *p;
list_for_each_entry_rcu (p, &dp->port_list, node) {
if (put_port(p, &uports[idx]))
err = destroy_dp(dp_idx);
goto exit;
- case ODP_PORT_ATTACH:
+ case ODP_VPORT_ATTACH:
err = attach_port(dp_idx, (struct odp_port __user *)argp);
goto exit;
- case ODP_PORT_DETACH:
+ case ODP_VPORT_DETACH:
err = get_user(port_no, (int __user *)argp);
if (!err)
err = detach_port(dp_idx, port_no);
goto exit;
- case ODP_VPORT_ADD:
- err = vport_user_add((struct odp_vport_add __user *)argp);
- goto exit;
-
case ODP_VPORT_MOD:
- err = vport_user_mod((struct odp_vport_mod __user *)argp);
- goto exit;
-
- case ODP_VPORT_DEL:
- err = vport_user_del((char __user *)argp);
+ err = vport_user_mod((struct odp_port __user *)argp);
goto exit;
case ODP_VPORT_STATS_GET:
dp->sflow_probability = sflow_probability;
break;
- case ODP_PORT_QUERY:
+ case ODP_VPORT_QUERY:
err = query_port(dp, (struct odp_port __user *)argp);
break;
- case ODP_PORT_LIST:
+ case ODP_VPORT_LIST:
err = list_ports(dp, (struct odp_portvec __user *)argp);
break;
}
static int compat_answer_query(struct sw_flow *flow, u32 query_flags,
- struct timespec time_offset,
struct compat_odp_flow __user *ufp)
{
compat_uptr_t actions;
if (get_user(actions, &ufp->actions))
return -EFAULT;
- return do_answer_query(flow, query_flags, time_offset, &ufp->stats,
+ return do_answer_query(flow, query_flags, &ufp->stats,
compat_ptr(actions), &ufp->n_actions);
}
if (IS_ERR(flow))
return PTR_ERR(flow);
- error = compat_answer_query(flow, 0, get_time_offset(), ufp);
+ error = compat_answer_query(flow, 0, ufp);
flow_deferred_free(flow);
return error;
}
static int compat_query_flows(struct datapath *dp, struct compat_odp_flow *flows, u32 n_flows)
{
struct tbl *table = rcu_dereference(dp->table);
- struct timespec time_offset;
u32 i;
- time_offset = get_time_offset();
-
for (i = 0; i < n_flows; i++) {
struct compat_odp_flow __user *ufp = &flows[i];
struct odp_flow uf;
if (compat_get_flow(&uf, ufp))
return -EFAULT;
- memset(uf.key.reserved, 0, sizeof uf.key.reserved);
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);
+ error = compat_answer_query(flow_cast(flow_node), uf.flags, ufp);
if (error)
return -EFAULT;
}
struct compat_odp_flow __user *uflows;
u32 n_flows;
u32 listed_flows;
- struct timespec time_offset;
};
static int compat_list_flow(struct tbl_node *node, void *cbdata_)
if (copy_to_user(&ufp->key, &flow->key, sizeof flow->key))
return -EFAULT;
- error = compat_answer_query(flow, 0, cbdata->time_offset, ufp);
+ error = compat_answer_query(flow, 0, ufp);
if (error)
return error;
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;
return openvswitch_ioctl(f, cmd, argp);
case ODP_DP_CREATE:
- case ODP_PORT_ATTACH:
- case ODP_PORT_DETACH:
- case ODP_VPORT_DEL:
+ case ODP_VPORT_ATTACH:
+ case ODP_VPORT_DETACH:
+ case ODP_VPORT_MOD:
case ODP_VPORT_MTU_SET:
case ODP_VPORT_MTU_GET:
case ODP_VPORT_ETHER_SET:
case ODP_GET_LISTEN_MASK:
case ODP_SET_SFLOW_PROBABILITY:
case ODP_GET_SFLOW_PROBABILITY:
- case ODP_PORT_QUERY:
+ case ODP_VPORT_QUERY:
/* Ioctls that just need their pointer argument extended. */
return openvswitch_ioctl(f, cmd, (unsigned long)compat_ptr(argp));
-
- case ODP_VPORT_ADD32:
- return compat_vport_user_add(compat_ptr(argp));
-
- case ODP_VPORT_MOD32:
- return compat_vport_user_mod(compat_ptr(argp));
}
dp = get_dp_locked(dp_idx);
goto exit;
switch (cmd) {
- case ODP_PORT_LIST32:
+ case ODP_VPORT_LIST32:
err = compat_list_ports(dp, compat_ptr(argp));
break;
projects / openvswitch / blobdiff