v1-v2:
- Use hash table.
- Remove dpif-linux port number allocator.
--8<--------------------------cut here-------------------------->8-
Use hash table to store ports of datapath. Allow full ranges of
32-bit number for port no, But limits count to 64K.
Signed-off-by: Pravin B Shelar <[email protected]>
Bug #2462
---
datapath/actions.c | 2 +-
datapath/datapath.c | 128 +++++++++++++--------
datapath/datapath.h | 30 ++++-
datapath/dp_sysfs_dp.c | 4 +-
datapath/dp_sysfs_if.c | 2 +-
datapath/flow.c | 8 +-
datapath/linux/Modules.mk | 1 +
datapath/linux/compat/include/linux/hash.h | 19 +++
datapath/vport.c | 1 +
datapath/vport.h | 2 +
lib/dpif-linux.c | 175 +++++++++-------------------
11 files changed, 189 insertions(+), 183 deletions(-)
create mode 100644 datapath/linux/compat/include/linux/hash.h
diff --git a/datapath/actions.c b/datapath/actions.c
index 824791d..4b07603 100644
--- a/datapath/actions.c
+++ b/datapath/actions.c
@@ -248,7 +248,7 @@ static int do_output(struct datapath *dp, struct sk_buff
*skb, int out_port)
if (unlikely(!skb))
return -ENOMEM;
- vport = rcu_dereference(dp->ports[out_port]);
+ vport = ovs_vport_rcu(dp, out_port);
if (unlikely(!vport)) {
kfree_skb(skb);
return -ENODEV;
diff --git a/datapath/datapath.c b/datapath/datapath.c
index 220c7dd..3248fb4 100644
--- a/datapath/datapath.c
+++ b/datapath/datapath.c
@@ -29,6 +29,7 @@
#include <linux/time.h>
#include <linux/etherdevice.h>
#include <linux/genetlink.h>
+#include <linux/hash.h>
#include <linux/kernel.h>
#include <linux/kthread.h>
#include <linux/mutex.h>
@@ -120,7 +121,7 @@ static struct datapath *get_dp(struct net *net, int
dp_ifindex)
/* Must be called with rcu_read_lock or RTNL lock. */
const char *ovs_dp_name(const struct datapath *dp)
{
- struct vport *vport = rcu_dereference_rtnl(dp->ports[OVSP_LOCAL]);
+ struct vport *vport = ovs_vport_rtnl_check(dp, OVSP_LOCAL);
return vport->ops->get_name(vport);
}
@@ -131,7 +132,7 @@ static int get_dpifindex(struct datapath *dp)
rcu_read_lock();
- local = rcu_dereference(dp->ports[OVSP_LOCAL]);
+ local = ovs_vport_rcu(dp, OVSP_LOCAL);
if (local)
ifindex = local->ops->get_ifindex(local);
else
@@ -244,9 +245,30 @@ static void destroy_dp_rcu(struct rcu_head *rcu)
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);
}
+static struct hlist_head *hash_bucket(const struct datapath *dp, int id)
+{
+ unsigned int hash = hash_32(id, DP_VPORT_HASH_BITS);
+ return &dp->ports[hash];
+}
+
+struct vport *ovs_vport_rcu(const struct datapath *dp, int id)
+{
+ struct vport *vport;
+ struct hlist_node *n;
+ struct hlist_head *head;
+
+ head = hash_bucket(dp, id);
+ hlist_for_each_entry_rcu(vport, n, head, dp_hash_node) {
+ if (vport->port_no == id)
+ return vport;
+ }
+ return NULL;
+}
+
/* Called with RTNL lock and genl_lock. */
static struct vport *new_vport(const struct vport_parms *parms)
{
@@ -256,12 +278,11 @@ static struct vport *new_vport(const struct vport_parms
*parms)
if (!IS_ERR(vport)) {
struct datapath *dp = parms->dp;
- rcu_assign_pointer(dp->ports[parms->port_no], vport);
- list_add(&vport->node, &dp->port_list);
+ hlist_add_head_rcu(&vport->dp_hash_node, hash_bucket(dp,
vport->port_no));
dp_ifinfo_notify(RTM_NEWLINK, vport);
+ dp->port_count++;
}
-
return vport;
}
@@ -272,11 +293,12 @@ void ovs_dp_detach_port(struct vport *p)
if (p->port_no != OVSP_LOCAL)
ovs_dp_sysfs_del_if(p);
+
+ p->dp->port_count--;
dp_ifinfo_notify(RTM_DELLINK, p);
/* First drop references to device. */
- list_del(&p->node);
- rcu_assign_pointer(p->dp->ports[p->port_no], NULL);
+ hlist_del_rcu(&p->dp_hash_node);
/* Then destroy it. */
ovs_vport_del(p);
@@ -660,11 +682,8 @@ static int validate_actions(const struct nlattr *attr,
break;
case OVS_ACTION_ATTR_OUTPUT:
- if (nla_get_u32(a) >= DP_MAX_PORTS)
- return -EINVAL;
break;
-
case OVS_ACTION_ATTR_POP_VLAN:
break;
@@ -1354,7 +1373,7 @@ static int ovs_dp_cmd_new(struct sk_buff *skb, struct
genl_info *info)
struct datapath *dp;
struct vport *vport;
struct ovs_net *ovs_net;
- int err;
+ int err, i;
err = -EINVAL;
if (!a[OVS_DP_ATTR_NAME] || !a[OVS_DP_ATTR_UPCALL_PID])
@@ -1371,8 +1390,6 @@ static int ovs_dp_cmd_new(struct sk_buff *skb, struct
genl_info *info)
if (dp == NULL)
goto err_unlock_rtnl;
- INIT_LIST_HEAD(&dp->port_list);
-
/* Initialize kobject for bridge. This will be added as
* /sys/class/net/<devname>/brif later, if sysfs is enabled. */
dp->ifobj.kset = NULL;
@@ -1391,6 +1408,16 @@ static int ovs_dp_cmd_new(struct sk_buff *skb, struct
genl_info *info)
}
ovs_dp_set_net(dp, hold_net(sock_net(skb->sk)));
+ dp->ports = kmalloc(DP_VPORT_HASH_BUCKETS * sizeof(struct hlist_head),
+ GFP_KERNEL);
+ if (!dp->ports) {
+ err = -ENOMEM;
+ goto err_destroy_percpu;
+ }
+
+ for (i = 0; i < DP_VPORT_HASH_BUCKETS; i++)
+ INIT_HLIST_HEAD(&dp->ports[i]);
+
/* Set up our datapath device. */
parms.name = nla_data(a[OVS_DP_ATTR_NAME]);
parms.type = OVS_VPORT_TYPE_INTERNAL;
@@ -1405,7 +1432,7 @@ static int ovs_dp_cmd_new(struct sk_buff *skb, struct
genl_info *info)
if (err == -EBUSY)
err = -EEXIST;
- goto err_destroy_percpu;
+ goto err_destroy_ports_array;
}
reply = ovs_dp_cmd_build_info(dp, info->snd_pid,
@@ -1426,7 +1453,9 @@ static int ovs_dp_cmd_new(struct sk_buff *skb, struct
genl_info *info)
return 0;
err_destroy_local_port:
- ovs_dp_detach_port(rtnl_dereference(dp->ports[OVSP_LOCAL]));
+ ovs_dp_detach_port(ovs_vport_rtnl_protected(dp, OVSP_LOCAL));
+err_destroy_ports_array:
+ kfree(dp->ports);
err_destroy_percpu:
free_percpu(dp->stats_percpu);
err_destroy_table:
@@ -1442,16 +1471,22 @@ err:
/* Called with genl_mutex. */
static void __dp_destroy(struct datapath *dp)
{
- struct vport *vport, *next_vport;
+ int i;
rtnl_lock();
- list_for_each_entry_safe(vport, next_vport, &dp->port_list, node)
- if (vport->port_no != OVSP_LOCAL)
- ovs_dp_detach_port(vport);
+
+ for (i = 0; i < DP_VPORT_HASH_BUCKETS; i++) {
+ struct vport *vport;
+ struct hlist_node *n;
+
+ hlist_for_each_entry_rcu(vport, n, &dp->ports[i], dp_hash_node)
+ if (vport->port_no != OVSP_LOCAL)
+ ovs_dp_detach_port(vport);
+ }
ovs_dp_sysfs_del_dp(dp);
list_del(&dp->list_node);
- ovs_dp_detach_port(rtnl_dereference(dp->ports[OVSP_LOCAL]));
+ ovs_dp_detach_port(ovs_vport_rtnl_protected(dp, OVSP_LOCAL));
/* rtnl_unlock() will wait until all the references to devices that
* are pending unregistration have been dropped. We do it here to
@@ -1698,14 +1733,11 @@ static struct vport *lookup_vport(struct net *net,
} else if (a[OVS_VPORT_ATTR_PORT_NO]) {
u32 port_no = nla_get_u32(a[OVS_VPORT_ATTR_PORT_NO]);
- if (port_no >= DP_MAX_PORTS)
- return ERR_PTR(-EFBIG);
-
dp = get_dp(net, ovs_header->dp_ifindex);
if (!dp)
return ERR_PTR(-ENODEV);
- vport = rcu_dereference_rtnl(dp->ports[port_no]);
+ vport = ovs_vport_rtnl_check(dp, port_no);
if (!vport)
return ERR_PTR(-ENOENT);
return vport;
@@ -1754,27 +1786,25 @@ static int ovs_vport_cmd_new(struct sk_buff *skb,
struct genl_info *info)
if (!dp)
goto exit_unlock;
+ if (dp->port_count >= DP_MAX_PORTS_COUNT) {
+ err = -EFBIG;
+ goto exit_unlock;
+ }
+
if (a[OVS_VPORT_ATTR_PORT_NO]) {
port_no = nla_get_u32(a[OVS_VPORT_ATTR_PORT_NO]);
- err = -EFBIG;
- if (port_no >= DP_MAX_PORTS)
- goto exit_unlock;
-
- vport = rtnl_dereference(dp->ports[port_no]);
+ vport = ovs_vport_rtnl_protected(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 = rtnl_dereference(dp->ports[port_no]);
+ for (port_no = 1; port_no < DP_MAX_PORTS_COUNT; port_no++) {
+ vport = ovs_vport_rtnl_protected(dp, port_no);
if (!vport)
break;
}
+ BUG_ON(vport);
}
parms.name = nla_data(a[OVS_VPORT_ATTR_NAME]);
@@ -1936,32 +1966,34 @@ static int ovs_vport_cmd_dump(struct sk_buff *skb,
struct netlink_callback *cb)
{
struct ovs_header *ovs_header = genlmsg_data(nlmsg_data(cb->nlh));
struct datapath *dp;
- u32 port_no;
- int retval;
+ u32 port_idx = 0, skip = cb->args[0];
+ int i;
dp = get_dp(sock_net(skb->sk), ovs_header->dp_ifindex);
if (!dp)
return -ENODEV;
rcu_read_lock();
- for (port_no = cb->args[0]; port_no < DP_MAX_PORTS; port_no++) {
+ for (i = 0; i < DP_VPORT_HASH_BUCKETS; i++) {
struct vport *vport;
+ struct hlist_node *n;
- vport = rcu_dereference(dp->ports[port_no]);
- if (!vport)
- continue;
+ hlist_for_each_entry_rcu(vport, n, &dp->ports[i], dp_hash_node)
{
+ if (port_idx >= 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;
- if (ovs_vport_cmd_fill_info(vport, skb, NETLINK_CB(cb->skb).pid,
- cb->nlh->nlmsg_seq, NLM_F_MULTI,
- OVS_VPORT_CMD_NEW) < 0)
- break;
+ port_idx++;
+ }
}
+out:
rcu_read_unlock();
- cb->args[0] = port_no;
- retval = skb->len;
+ cb->args[0] = port_idx;
- return retval;
+ return skb->len;
}
static struct genl_ops dp_vport_genl_ops[] = {
diff --git a/datapath/datapath.h b/datapath/datapath.h
index b012a76..d76222f 100644
--- a/datapath/datapath.h
+++ b/datapath/datapath.h
@@ -34,7 +34,10 @@
#include "vlan.h"
#include "vport.h"
-#define DP_MAX_PORTS 1024
+#define DP_MAX_PORTS_COUNT (1024 * 64)
+#define DP_VPORT_HASH_BITS 10
+#define DP_VPORT_HASH_BUCKETS (2 << DP_VPORT_HASH_BITS)
+
#define SAMPLE_ACTION_DEPTH 3
/**
@@ -63,10 +66,9 @@ struct dp_stats_percpu {
* @ifobj: Represents /sys/class/net/<devname>/brif. Protected by RTNL.
* @n_flows: Number of flows currently in flow table.
* @table: Current flow table. Protected by genl_lock and RCU.
- * @ports: Map from port number to &struct vport. %OVSP_LOCAL port
- * always exists, other ports may be %NULL. Protected by RTNL and RCU.
- * @port_list: List of all ports in @ports in arbitrary order. RTNL required
- * to iterate or modify.
+ * @ports: Hash table for ports. %OVSP_LOCAL port always exists. Protected by
+ * RTNL and RCU.
+ * @port_count: Number of ports on datapath.
* @stats_percpu: Per-CPU datapath statistics.
* @net: Reference to net namespace.
*
@@ -82,8 +84,8 @@ struct datapath {
struct flow_table __rcu *table;
/* Switch ports. */
- struct vport __rcu *ports[DP_MAX_PORTS];
- struct list_head port_list;
+ struct hlist_head *ports;
+ int port_count;
/* Stats. */
struct dp_stats_percpu __percpu *stats_percpu;
@@ -159,6 +161,20 @@ static inline void ovs_dp_set_net(struct datapath *dp,
struct net *net)
write_pnet(&dp->net, net);
}
+struct vport *ovs_vport_rcu(const struct datapath *dp, int id);
+
+static inline struct vport *ovs_vport_rtnl_check(const struct datapath *dp,
int id)
+{
+ WARN_ON(!rcu_read_lock_held() && !rtnl_is_locked());
+ return ovs_vport_rcu(dp, id);
+}
+
+static inline struct vport *ovs_vport_rtnl_protected(const struct datapath
*dp, int id)
+{
+ ASSERT_RTNL();
+ return ovs_vport_rcu(dp, id);
+}
+
extern struct notifier_block ovs_dp_device_notifier;
extern struct genl_multicast_group ovs_dp_vport_multicast_group;
extern int (*ovs_dp_ioctl_hook)(struct net_device *dev, struct ifreq *rq, int
cmd);
diff --git a/datapath/dp_sysfs_dp.c b/datapath/dp_sysfs_dp.c
index 2582321..2969852 100644
--- a/datapath/dp_sysfs_dp.c
+++ b/datapath/dp_sysfs_dp.c
@@ -362,7 +362,7 @@ static struct attribute_group bridge_group = {
*/
int ovs_dp_sysfs_add_dp(struct datapath *dp)
{
- struct vport *vport = rtnl_dereference(dp->ports[OVSP_LOCAL]);
+ struct vport *vport = ovs_vport_rtnl_protected(dp, OVSP_LOCAL);
struct kobject *kobj = vport->ops->get_kobj(vport);
int err;
@@ -398,7 +398,7 @@ int ovs_dp_sysfs_add_dp(struct datapath *dp)
int ovs_dp_sysfs_del_dp(struct datapath *dp)
{
- struct vport *vport = rtnl_dereference(dp->ports[OVSP_LOCAL]);
+ struct vport *vport = ovs_vport_rtnl_protected(dp, OVSP_LOCAL);
struct kobject *kobj = vport->ops->get_kobj(vport);
#ifdef CONFIG_NET_NS
diff --git a/datapath/dp_sysfs_if.c b/datapath/dp_sysfs_if.c
index f564e98..10beade 100644
--- a/datapath/dp_sysfs_if.c
+++ b/datapath/dp_sysfs_if.c
@@ -209,7 +209,7 @@ struct sysfs_ops ovs_brport_sysfs_ops = {
int ovs_dp_sysfs_add_if(struct vport *p)
{
struct datapath *dp = p->dp;
- struct vport *local_port = rtnl_dereference(dp->ports[OVSP_LOCAL]);
+ struct vport *local_port = ovs_vport_rtnl_protected(dp, OVSP_LOCAL);
struct brport_attribute **a;
int err;
diff --git a/datapath/flow.c b/datapath/flow.c
index 823c6b5..c15f6f0 100644
--- a/datapath/flow.c
+++ b/datapath/flow.c
@@ -203,10 +203,10 @@ struct sw_flow_actions *ovs_flow_actions_alloc(const
struct nlattr *actions)
int actions_len = nla_len(actions);
struct sw_flow_actions *sfa;
- /* At least DP_MAX_PORTS actions are required to be able to flood a
+ /* At least DP_MAX_PORTS_COUNT actions are required to be able to flood
a
* packet to every port. Factor of 2 allows for setting VLAN tags,
* etc. */
- if (actions_len > 2 * DP_MAX_PORTS * nla_total_size(4))
+ if (actions_len > 2 * DP_MAX_PORTS_COUNT * nla_total_size(4))
return ERR_PTR(-EINVAL);
sfa = kmalloc(sizeof(*sfa) + actions_len, GFP_KERNEL);
@@ -1015,8 +1015,6 @@ int ovs_flow_from_nlattrs(struct sw_flow_key *swkey, int
*key_lenp,
}
if (attrs & (1 << OVS_KEY_ATTR_IN_PORT)) {
u32 in_port = nla_get_u32(a[OVS_KEY_ATTR_IN_PORT]);
- if (in_port >= DP_MAX_PORTS)
- return -EINVAL;
swkey->phy.in_port = in_port;
attrs &= ~(1 << OVS_KEY_ATTR_IN_PORT);
} else {
@@ -1190,8 +1188,6 @@ int ovs_flow_metadata_from_nlattrs(u32 *priority, u16
*in_port, __be64 *tun_id,
break;
case OVS_KEY_ATTR_IN_PORT:
- if (nla_get_u32(nla) >= DP_MAX_PORTS)
- return -EINVAL;
*in_port = nla_get_u32(nla);
break;
}
diff --git a/datapath/linux/Modules.mk b/datapath/linux/Modules.mk
index 7f54bde..4c8926d 100644
--- a/datapath/linux/Modules.mk
+++ b/datapath/linux/Modules.mk
@@ -20,6 +20,7 @@ openvswitch_headers += \
linux/compat/include/linux/err.h \
linux/compat/include/linux/flex_array.h \
linux/compat/include/linux/genetlink.h \
+ linux/compat/include/linux/hash.h \
linux/compat/include/linux/icmp.h \
linux/compat/include/linux/icmpv6.h \
linux/compat/include/linux/if.h \
diff --git a/datapath/linux/compat/include/linux/hash.h
b/datapath/linux/compat/include/linux/hash.h
new file mode 100644
index 0000000..b909ac6
--- /dev/null
+++ b/datapath/linux/compat/include/linux/hash.h
@@ -0,0 +1,19 @@
+#ifndef __HASH_WRAPPER_H
+#define __HASH_WRAPPER_H 1
+
+#include_next <linux/hash.h>
+
+#if LINUX_VERSION_CODE < KERNEL_VERSION(2,6,25)
+/* 2^31 + 2^29 - 2^25 + 2^22 - 2^19 - 2^16 + 1 */
+#define GOLDEN_RATIO_PRIME_32 0x9e370001UL
+
+static inline u32 hash_32(u32 val, unsigned int bits)
+{
+ /* On some cpus multiply is faster, on others gcc will do shifts */
+ u32 hash = val * GOLDEN_RATIO_PRIME_32;
+
+ /* High bits are more random, so use them. */
+ return hash >> (32 - bits);
+}
+#endif
+#endif /* linux/hash.h wrapper */
diff --git a/datapath/vport.c b/datapath/vport.c
index d81f686..b75a866 100644
--- a/datapath/vport.c
+++ b/datapath/vport.c
@@ -192,6 +192,7 @@ struct vport *ovs_vport_alloc(int priv_size, const struct
vport_ops *ops,
vport->port_no = parms->port_no;
vport->upcall_pid = parms->upcall_pid;
vport->ops = ops;
+ INIT_HLIST_NODE(&vport->dp_hash_node);
/* Initialize kobject for bridge. This will be added as
* /sys/class/net/<devname>/brport later, if sysfs is enabled. */
diff --git a/datapath/vport.h b/datapath/vport.h
index ee9715d..2aafde0 100644
--- a/datapath/vport.h
+++ b/datapath/vport.h
@@ -84,6 +84,7 @@ struct vport_err_stats {
* @upcall_pid: The Netlink port to use for packets received on this port that
* miss the flow table.
* @hash_node: Element in @dev_table hash table in vport.c.
+ * @dp_hash_node: Element in @datapath->ports hash table in datapath.c.
* @ops: Class structure.
* @percpu_stats: Points to per-CPU statistics used and maintained by vport
* @stats_lock: Protects @err_stats and @offset_stats.
@@ -101,6 +102,7 @@ struct vport {
u32 upcall_pid;
struct hlist_node hash_node;
+ struct hlist_node dp_hash_node;
const struct vport_ops *ops;
struct vport_percpu_stats __percpu *percpu_stats;
diff --git a/lib/dpif-linux.c b/lib/dpif-linux.c
index c06bb89..30cfe2f 100644
--- a/lib/dpif-linux.c
+++ b/lib/dpif-linux.c
@@ -60,10 +60,6 @@
VLOG_DEFINE_THIS_MODULE(dpif_linux);
-enum { LRU_MAX_PORTS = 1024 };
-enum { LRU_MASK = LRU_MAX_PORTS - 1};
-BUILD_ASSERT_DECL(IS_POW2(LRU_MAX_PORTS));
-
enum { N_UPCALL_SOCKS = 16 };
BUILD_ASSERT_DECL(IS_POW2(N_UPCALL_SOCKS));
BUILD_ASSERT_DECL(N_UPCALL_SOCKS <= 32); /* We use a 32-bit word as a mask. */
@@ -146,12 +142,6 @@ struct dpif_linux {
struct sset changed_ports; /* Ports that have changed. */
struct nln_notifier *port_notifier;
bool change_error;
-
- /* Queue of unused ports. */
- unsigned long *lru_bitmap;
- uint16_t lru_ports[LRU_MAX_PORTS];
- size_t lru_head;
- size_t lru_tail;
};
static struct vlog_rate_limit error_rl = VLOG_RATE_LIMIT_INIT(9999, 5);
@@ -171,6 +161,7 @@ static void open_dpif(const struct dpif_linux_dp *, struct
dpif **);
static bool dpif_linux_nln_parse(struct ofpbuf *, void *);
static void dpif_linux_port_changed(const void *vport, void *dpif);
static uint32_t dpif_linux_port_get_pid(const struct dpif *, uint16_t port_no);
+static int set_upcall_pid(const struct dpif *, int);
static void dpif_linux_vport_to_ofpbuf(const struct dpif_linux_vport *,
struct ofpbuf *);
@@ -184,29 +175,6 @@ dpif_linux_cast(const struct dpif *dpif)
return CONTAINER_OF(dpif, struct dpif_linux, dpif);
}
-static void
-dpif_linux_push_port(struct dpif_linux *dp, uint16_t port)
-{
- if (port < LRU_MAX_PORTS && !bitmap_is_set(dp->lru_bitmap, port)) {
- bitmap_set1(dp->lru_bitmap, port);
- dp->lru_ports[dp->lru_head++ & LRU_MASK] = port;
- }
-}
-
-static uint32_t
-dpif_linux_pop_port(struct dpif_linux *dp)
-{
- uint16_t port;
-
- if (dp->lru_head == dp->lru_tail) {
- return UINT32_MAX;
- }
-
- port = dp->lru_ports[dp->lru_tail++ & LRU_MASK];
- bitmap_set0(dp->lru_bitmap, port);
- return port;
-}
-
static int
dpif_linux_enumerate(struct sset *all_dps)
{
@@ -268,7 +236,6 @@ static void
open_dpif(const struct dpif_linux_dp *dp, struct dpif **dpifp)
{
struct dpif_linux *dpif;
- int i;
dpif = xzalloc(sizeof *dpif);
dpif->port_notifier = nln_notifier_create(nln, dpif_linux_port_changed,
@@ -281,12 +248,6 @@ open_dpif(const struct dpif_linux_dp *dp, struct dpif
**dpifp)
dpif->dp_ifindex = dp->dp_ifindex;
sset_init(&dpif->changed_ports);
*dpifp = &dpif->dpif;
-
- dpif->lru_bitmap = bitmap_allocate(LRU_MAX_PORTS);
- bitmap_set1(dpif->lru_bitmap, OVSP_LOCAL);
- for (i = 1; i < LRU_MAX_PORTS; i++) {
- dpif_linux_push_port(dpif, i);
- }
}
static void
@@ -312,7 +273,6 @@ dpif_linux_close(struct dpif *dpif_)
nln_notifier_destroy(dpif->port_notifier);
destroy_upcall_socks(dpif);
sset_destroy(&dpif->changed_ports);
- free(dpif->lru_bitmap);
free(dpif);
}
@@ -363,6 +323,22 @@ dpif_linux_get_stats(const struct dpif *dpif_, struct
dpif_dp_stats *stats)
}
static int
+dpif_linux_port_del(struct dpif *dpif_, uint16_t port_no)
+{
+ struct dpif_linux *dpif = dpif_linux_cast(dpif_);
+ struct dpif_linux_vport vport;
+ int error;
+
+ dpif_linux_vport_init(&vport);
+ vport.cmd = OVS_VPORT_CMD_DEL;
+ vport.dp_ifindex = dpif->dp_ifindex;
+ vport.port_no = port_no;
+ error = dpif_linux_vport_transact(&vport, NULL, NULL);
+
+ return error;
+}
+
+static int
dpif_linux_port_add(struct dpif *dpif_, struct netdev *netdev,
uint16_t *port_nop)
{
@@ -372,6 +348,7 @@ dpif_linux_port_add(struct dpif *dpif_, struct netdev
*netdev,
struct dpif_linux_vport request, reply;
const struct ofpbuf *options;
struct ofpbuf *buf;
+ uint32_t upcall_pid;
int error;
dpif_linux_vport_init(&request);
@@ -396,43 +373,20 @@ dpif_linux_port_add(struct dpif *dpif_, struct netdev
*netdev,
netdev_linux_ethtool_set_flag(netdev, ETH_FLAG_LRO, "LRO", false);
}
- /* Loop until we find a port that isn't used. */
- do {
- uint32_t upcall_pid;
-
- request.port_no = dpif_linux_pop_port(dpif);
- upcall_pid = dpif_linux_port_get_pid(dpif_, request.port_no);
- request.upcall_pid = &upcall_pid;
- error = dpif_linux_vport_transact(&request, &reply, &buf);
-
- if (!error) {
- *port_nop = reply.port_no;
- VLOG_DBG("%s: assigning port %"PRIu32" to netlink pid %"PRIu32,
- dpif_name(dpif_), request.port_no, upcall_pid);
- }
- ofpbuf_delete(buf);
- } while (request.port_no != UINT32_MAX
- && (error == EBUSY || error == EFBIG));
-
- return error;
-}
-
-static int
-dpif_linux_port_del(struct dpif *dpif_, uint16_t port_no)
-{
- struct dpif_linux *dpif = dpif_linux_cast(dpif_);
- struct dpif_linux_vport vport;
- int error;
-
- dpif_linux_vport_init(&vport);
- vport.cmd = OVS_VPORT_CMD_DEL;
- vport.dp_ifindex = dpif->dp_ifindex;
- vport.port_no = port_no;
- error = dpif_linux_vport_transact(&vport, NULL, NULL);
+ request.port_no = UINT32_MAX;
+ upcall_pid = dpif_linux_port_get_pid(dpif_, request.port_no);
+ request.upcall_pid = &upcall_pid;
+ error = dpif_linux_vport_transact(&request, &reply, &buf);
if (!error) {
- dpif_linux_push_port(dpif, port_no);
+ *port_nop = reply.port_no;
+ error = set_upcall_pid(dpif_, reply.port_no);
+ if (error) {
+ dpif_linux_port_del(dpif_, reply.port_no);
+ }
}
+ ofpbuf_delete(buf);
+
return error;
}
@@ -478,9 +432,7 @@ dpif_linux_port_query_by_name(const struct dpif *dpif,
const char *devname,
static int
dpif_linux_get_max_ports(const struct dpif *dpif OVS_UNUSED)
{
- /* If the datapath increases its range of supported ports, then it should
- * start reporting that. */
- return 1024;
+ return UINT32_MAX;
}
static uint32_t
@@ -510,8 +462,6 @@ dpif_linux_flow_flush(struct dpif *dpif_)
struct dpif_linux_port_state {
struct nl_dump dump;
- unsigned long *port_bitmap; /* Ports in the datapath. */
- bool complete; /* Dump completed without error. */
};
static int
@@ -523,8 +473,6 @@ dpif_linux_port_dump_start(const struct dpif *dpif_, void
**statep)
struct ofpbuf *buf;
*statep = state = xmalloc(sizeof *state);
- state->port_bitmap = bitmap_allocate(LRU_MAX_PORTS);
- state->complete = false;
dpif_linux_vport_init(&request);
request.cmd = OVS_DP_CMD_GET;
@@ -548,7 +496,6 @@ dpif_linux_port_dump_next(const struct dpif *dpif
OVS_UNUSED, void *state_,
int error;
if (!nl_dump_next(&state->dump, &buf)) {
- state->complete = true;
return EOF;
}
@@ -557,10 +504,6 @@ dpif_linux_port_dump_next(const struct dpif *dpif
OVS_UNUSED, void *state_,
return error;
}
- if (vport.port_no < LRU_MAX_PORTS) {
- bitmap_set1(state->port_bitmap, vport.port_no);
- }
-
dpif_port->name = (char *) vport.name;
dpif_port->type = (char *) netdev_vport_get_netdev_type(&vport);
dpif_port->port_no = vport.port_no;
@@ -568,23 +511,11 @@ dpif_linux_port_dump_next(const struct dpif *dpif
OVS_UNUSED, void *state_,
}
static int
-dpif_linux_port_dump_done(const struct dpif *dpif_, void *state_)
+dpif_linux_port_dump_done(const struct dpif *dpif_ OVS_UNUSED, void *state_)
{
- struct dpif_linux *dpif = dpif_linux_cast(dpif_);
struct dpif_linux_port_state *state = state_;
int error = nl_dump_done(&state->dump);
- if (state->complete) {
- uint16_t i;
-
- for (i = 0; i < LRU_MAX_PORTS; i++) {
- if (!bitmap_is_set(state->port_bitmap, i)) {
- dpif_linux_push_port(dpif, i);
- }
- }
- }
-
- free(state->port_bitmap);
free(state);
return error;
}
@@ -929,32 +860,40 @@ dpif_linux_operate(struct dpif *dpif_, struct dpif_op
**ops, size_t n_ops)
free(txns);
}
+static int
+set_upcall_pid(const struct dpif *dpif_, int port_no)
+{
+ struct dpif_linux *dpif = dpif_linux_cast(dpif_);
+ uint32_t upcall_pid = dpif_linux_port_get_pid(dpif_, port_no);
+ struct dpif_linux_vport vport_request;
+ int error;
+
+ dpif_linux_vport_init(&vport_request);
+ vport_request.cmd = OVS_VPORT_CMD_SET;
+ vport_request.dp_ifindex = dpif->dp_ifindex;
+ vport_request.port_no = port_no;
+ vport_request.upcall_pid = &upcall_pid;
+ error = dpif_linux_vport_transact(&vport_request, NULL, NULL);
+ if (!error) {
+ VLOG_DBG("%s: assigning port %"PRIu32" to netlink pid %"PRIu32,
+ dpif_name(&dpif->dpif), vport_request.port_no,
+ upcall_pid);
+ } else {
+ VLOG_WARN_RL(&error_rl, "%s: failed to set upcall pid on port: %s",
+ dpif_name(&dpif->dpif), strerror(error));
+ }
+ return error;
+}
+
static void
set_upcall_pids(struct dpif *dpif_)
{
struct dpif_linux *dpif = dpif_linux_cast(dpif_);
struct dpif_port_dump port_dump;
struct dpif_port port;
- int error;
DPIF_PORT_FOR_EACH (&port, &port_dump, &dpif->dpif) {
- uint32_t upcall_pid = dpif_linux_port_get_pid(dpif_, port.port_no);
- struct dpif_linux_vport vport_request;
-
- dpif_linux_vport_init(&vport_request);
- vport_request.cmd = OVS_VPORT_CMD_SET;
- vport_request.dp_ifindex = dpif->dp_ifindex;
- vport_request.port_no = port.port_no;
- vport_request.upcall_pid = &upcall_pid;
- error = dpif_linux_vport_transact(&vport_request, NULL, NULL);
- if (!error) {
- VLOG_DBG("%s: assigning port %"PRIu32" to netlink pid %"PRIu32,
- dpif_name(&dpif->dpif), vport_request.port_no,
- upcall_pid);
- } else {
- VLOG_WARN_RL(&error_rl, "%s: failed to set upcall pid on port: %s",
- dpif_name(&dpif->dpif), strerror(error));
- }
+ set_upcall_pid(dpif_, port.port_no);
}
}
--
1.7.1
_______________________________________________
dev mailing list
[email protected]
http://openvswitch.org/mailman/listinfo/dev
