Following patch uses of lockless genl callbacks and makes
ovs datapath locking fine granular. This allows simultaneous flow
setup and port add/deletion. Also it fixes softlockup issues
seen due to genl-lock.
Signed-off-by: Pravin B Shelar <pshe...@nicira.com>
---
net/openvswitch/datapath.c | 280 +++++++++++++++++++++++++++++++-------------
net/openvswitch/datapath.h | 4 +-
net/openvswitch/vport.c | 1 +
3 files changed, 205 insertions(+), 80 deletions(-)
diff --git a/net/openvswitch/datapath.c b/net/openvswitch/datapath.c
index 2c74daa..a452427 100644
--- a/net/openvswitch/datapath.c
+++ b/net/openvswitch/datapath.c
@@ -57,13 +57,13 @@
/**
* DOC: Locking:
*
- * Writes to device state (add/remove datapath, port, set operations on vports,
- * etc.) are protected by RTNL.
+ * Updates to datapath are protected by dp_lock write lock, dp_lock
+ * nests inside RTNL lock.
*
- * Writes to other state (flow table modifications, set miscellaneous datapath
- * parameters, etc.) are protected by genl_mutex. The RTNL lock nests inside
- * genl_mutex.
+ * Writes to device state (add/remove port, set operations on vports,
+ * etc.) are protected by RTNL.
*
+ * Writes to flow table modifications are protected by datapath ftable_lock.
* Reads are protected by RCU.
*
* There are a few special cases (mostly stats) that have their own
@@ -71,8 +71,43 @@
* each other.
*/
+static DECLARE_RWSEM(dp_lock);
+
+static void dp_read_lock(void)
+{
+ down_read(&dp_lock);
+}
+
+static void dp_read_unlock(void)
+{
+ up_read(&dp_lock);
+}
+
+static void dp_write_lock(void)
+{
+ down_write(&dp_lock);
+}
+
+static void dp_write_unlock(void)
+{
+ up_write(&dp_lock);
+}
+
+#define flow_table_ft_lock(dp) \
+ rcu_dereference_protected(dp->table, \
+ (rwsem_is_locked(&dp_lock) &&
lockdep_is_held(&dp->ftable_lock)))
+
+#define flow_acts_ft_lock(flow, dp) \
+ rcu_dereference_protected(flow->sf_acts, \
+ (rwsem_is_locked(&dp_lock) &&
lockdep_is_held(&dp->ftable_lock)))
+
+#define flow_acts_rcu_ft_lock(flow, dp) \
+ rcu_dereference_check(flow->sf_acts, \
+ (rwsem_is_locked(&dp_lock) &&
lockdep_is_held(&dp->ftable_lock)))
+
+
/* Global list of datapaths to enable dumping them all out.
- * Protected by genl_mutex.
+ * Protected by dp_lock.
*/
static LIST_HEAD(dps);
@@ -80,13 +115,12 @@ static LIST_HEAD(dps);
static void rehash_flow_table(struct work_struct *work);
static DECLARE_DELAYED_WORK(rehash_flow_wq, rehash_flow_table);
-static struct vport *new_vport(const struct vport_parms *);
static int queue_gso_packets(int dp_ifindex, struct sk_buff *,
const struct dp_upcall_info *);
static int queue_userspace_packet(int dp_ifindex, struct sk_buff *,
const struct dp_upcall_info *);
-/* Must be called with rcu_read_lock, genl_mutex, or RTNL lock. */
+/* Must be called with rcu_read_lock, dp_lock, or RTNL lock. */
static struct datapath *get_dp(int dp_ifindex)
{
struct datapath *dp = NULL;
@@ -138,7 +172,7 @@ static void destroy_dp_rcu(struct rcu_head *rcu)
kfree(dp);
}
-/* Called with RTNL lock and genl_lock. */
+/* Called with RTNL lock. */
static struct vport *new_vport(const struct vport_parms *parms)
{
struct vport *vport;
@@ -220,7 +254,8 @@ static struct genl_family dp_packet_genl_family = {
.hdrsize = sizeof(struct ovs_header),
.name = OVS_PACKET_FAMILY,
.version = OVS_PACKET_VERSION,
- .maxattr = OVS_PACKET_ATTR_MAX
+ .maxattr = OVS_PACKET_ATTR_MAX,
+ .lockless = true,
};
int ovs_dp_upcall(struct datapath *dp, struct sk_buff *skb,
@@ -368,18 +403,13 @@ out:
return err;
}
-/* Called with genl_mutex. */
-static int flush_flows(int dp_ifindex)
+/* Called with dp_lock and ft_lock. */
+static int flush_flows(struct datapath *dp)
{
struct flow_table *old_table;
struct flow_table *new_table;
- struct datapath *dp;
- dp = get_dp(dp_ifindex);
- if (!dp)
- return -ENODEV;
-
- old_table = genl_dereference(dp->table);
+ old_table = flow_table_ft_lock(dp);
new_table = ovs_flow_tbl_alloc(TBL_MIN_BUCKETS);
if (!new_table)
return -ENOMEM;
@@ -704,10 +734,11 @@ static struct genl_ops dp_packet_genl_ops[] = {
}
};
+/* Must be called with rcu_read_lock. */
static void get_dp_stats(struct datapath *dp, struct ovs_dp_stats *stats)
{
int i;
- struct flow_table *table = genl_dereference(dp->table);
+ struct flow_table *table = rcu_dereference(dp->table);
stats->n_flows = ovs_flow_tbl_count(table);
@@ -741,14 +772,15 @@ static struct genl_family dp_flow_genl_family = {
.hdrsize = sizeof(struct ovs_header),
.name = OVS_FLOW_FAMILY,
.version = OVS_FLOW_VERSION,
- .maxattr = OVS_FLOW_ATTR_MAX
+ .maxattr = OVS_FLOW_ATTR_MAX,
+ .lockless = true,
};
static struct genl_multicast_group ovs_dp_flow_multicast_group = {
.name = OVS_FLOW_MCGROUP
};
-/* Called with genl_lock. */
+/* Called with rcu_read_lock or ft_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)
@@ -762,8 +794,7 @@ static int ovs_flow_cmd_fill_info(struct sw_flow *flow,
struct datapath *dp,
u8 tcp_flags;
int err;
- sf_acts = rcu_dereference_protected(flow->sf_acts,
- lockdep_genl_is_held());
+ sf_acts = flow_acts_rcu_ft_lock(flow, dp);
ovs_header = genlmsg_put(skb, pid, seq, &dp_flow_genl_family, flags,
cmd);
if (!ovs_header)
@@ -823,13 +854,13 @@ error:
return err;
}
-static struct sk_buff *ovs_flow_cmd_alloc_info(struct sw_flow *flow)
+static struct sk_buff *ovs_flow_cmd_alloc_info(struct datapath *dp,
+ struct sw_flow *flow)
{
const struct sw_flow_actions *sf_acts;
int len;
- sf_acts = rcu_dereference_protected(flow->sf_acts,
- lockdep_genl_is_held());
+ sf_acts = flow_acts_ft_lock(flow, dp);
/* OVS_FLOW_ATTR_KEY */
len = nla_total_size(FLOW_BUFSIZE);
@@ -854,7 +885,7 @@ static struct sk_buff *ovs_flow_cmd_build_info(struct
sw_flow *flow,
struct sk_buff *skb;
int retval;
- skb = ovs_flow_cmd_alloc_info(flow);
+ skb = ovs_flow_cmd_alloc_info(dp, flow);
if (!skb)
return ERR_PTR(-ENOMEM);
@@ -893,12 +924,14 @@ static int ovs_flow_cmd_new_or_set(struct sk_buff *skb,
struct genl_info *info)
goto error;
}
+ dp_read_lock();
dp = get_dp(ovs_header->dp_ifindex);
error = -ENODEV;
if (!dp)
- goto error;
+ goto error_unlock;
- table = genl_dereference(dp->table);
+ mutex_lock(&dp->ftable_lock);
+ table = flow_table_ft_lock(dp);
flow = ovs_flow_tbl_lookup(table, &key, key_len);
if (!flow) {
struct sw_flow_actions *acts;
@@ -906,7 +939,7 @@ static int ovs_flow_cmd_new_or_set(struct sk_buff *skb,
struct genl_info *info)
/* Bail out if we're not allowed to create a new flow. */
error = -ENOENT;
if (info->genlhdr->cmd == OVS_FLOW_CMD_SET)
- goto error;
+ goto error_ft_unlock;
/* Expand table, if necessary, to make room. */
if (ovs_flow_tbl_need_to_expand(table)) {
@@ -916,7 +949,7 @@ static int ovs_flow_cmd_new_or_set(struct sk_buff *skb,
struct genl_info *info)
if (!IS_ERR(new_table)) {
rcu_assign_pointer(dp->table, new_table);
ovs_flow_tbl_deferred_destroy(table);
- table = genl_dereference(dp->table);
+ table = flow_table_ft_lock(dp);
}
}
@@ -924,7 +957,7 @@ static int ovs_flow_cmd_new_or_set(struct sk_buff *skb,
struct genl_info *info)
flow = ovs_flow_alloc();
if (IS_ERR(flow)) {
error = PTR_ERR(flow);
- goto error;
+ goto error_ft_unlock;
}
flow->key = key;
clear_stats(flow);
@@ -957,11 +990,10 @@ static int ovs_flow_cmd_new_or_set(struct sk_buff *skb,
struct genl_info *info)
error = -EEXIST;
if (info->genlhdr->cmd == OVS_FLOW_CMD_NEW &&
info->nlhdr->nlmsg_flags & (NLM_F_CREATE | NLM_F_EXCL))
- goto error;
+ goto error_ft_unlock;
/* Update actions. */
- old_acts = rcu_dereference_protected(flow->sf_acts,
- lockdep_genl_is_held());
+ old_acts = flow_acts_ft_lock(flow, dp);
acts_attrs = a[OVS_FLOW_ATTR_ACTIONS];
if (acts_attrs &&
(old_acts->actions_len != nla_len(acts_attrs) ||
@@ -972,7 +1004,7 @@ static int ovs_flow_cmd_new_or_set(struct sk_buff *skb,
struct genl_info *info)
new_acts = ovs_flow_actions_alloc(acts_attrs);
error = PTR_ERR(new_acts);
if (IS_ERR(new_acts))
- goto error;
+ goto error_ft_unlock;
rcu_assign_pointer(flow->sf_acts, new_acts);
ovs_flow_deferred_free_acts(old_acts);
@@ -989,6 +1021,9 @@ static int ovs_flow_cmd_new_or_set(struct sk_buff *skb,
struct genl_info *info)
}
}
+ mutex_unlock(&dp->ftable_lock);
+ dp_read_unlock();
+
if (!IS_ERR(reply))
genl_notify(reply, genl_info_net(info), info->snd_pid,
ovs_dp_flow_multicast_group.id, info->nlhdr,
@@ -1000,6 +1035,10 @@ static int ovs_flow_cmd_new_or_set(struct sk_buff *skb,
struct genl_info *info)
error_free_flow:
ovs_flow_free(flow);
+error_ft_unlock:
+ mutex_unlock(&dp->ftable_lock);
+error_unlock:
+ dp_read_unlock();
error:
return error;
}
@@ -1022,21 +1061,37 @@ static int ovs_flow_cmd_get(struct sk_buff *skb, struct
genl_info *info)
if (err)
return err;
+ dp_read_lock();
dp = get_dp(ovs_header->dp_ifindex);
- if (!dp)
- return -ENODEV;
+ if (!dp) {
+ err = -ENODEV;
+ goto exit_unlock;
+ }
- table = genl_dereference(dp->table);
+ mutex_lock(&dp->ftable_lock);
+ table = flow_table_ft_lock(dp);
flow = ovs_flow_tbl_lookup(table, &key, key_len);
- if (!flow)
- return -ENOENT;
+ if (!flow) {
+ err = ENOENT;
+ goto exit_ft_unlock;
+ }
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);
+ if (IS_ERR(reply)) {
+ err = PTR_ERR(reply);
+ goto exit_ft_unlock;
+ }
+ mutex_unlock(&dp->ftable_lock);
+ dp_read_unlock();
return genlmsg_reply(reply, info);
+
+exit_ft_unlock:
+ mutex_unlock(&dp->ftable_lock);
+exit_unlock:
+ dp_read_unlock();
+ return err;
}
static int ovs_flow_cmd_del(struct sk_buff *skb, struct genl_info *info)
@@ -1051,24 +1106,35 @@ static int ovs_flow_cmd_del(struct sk_buff *skb, struct
genl_info *info)
int err;
int key_len;
- if (!a[OVS_FLOW_ATTR_KEY])
- return flush_flows(ovs_header->dp_ifindex);
+ dp_read_lock();
+ dp = get_dp(ovs_header->dp_ifindex);
+ if (!dp) {
+ err = -ENODEV;
+ goto exit_unlock;
+ }
+
+ mutex_lock(&dp->ftable_lock);
+ if (!a[OVS_FLOW_ATTR_KEY]) {
+ err = flush_flows(dp);
+ goto exit_ft_unlock;
+ }
+
err = ovs_flow_from_nlattrs(&key, &key_len, a[OVS_FLOW_ATTR_KEY]);
if (err)
- return err;
+ goto exit_ft_unlock;
- dp = get_dp(ovs_header->dp_ifindex);
- if (!dp)
- return -ENODEV;
-
- table = genl_dereference(dp->table);
+ table = flow_table_ft_lock(dp);
flow = ovs_flow_tbl_lookup(table, &key, key_len);
- if (!flow)
- return -ENOENT;
+ if (!flow) {
+ err = -ENOENT;
+ goto exit_ft_unlock;
+ }
- reply = ovs_flow_cmd_alloc_info(flow);
- if (!reply)
- return -ENOMEM;
+ reply = ovs_flow_cmd_alloc_info(dp, flow);
+ if (!reply) {
+ err = -ENOMEM;
+ goto exit_ft_unlock;
+ }
ovs_flow_tbl_remove(table, flow);
@@ -1080,7 +1146,12 @@ static int ovs_flow_cmd_del(struct sk_buff *skb, struct
genl_info *info)
genl_notify(reply, genl_info_net(info), info->snd_pid,
ovs_dp_flow_multicast_group.id, info->nlhdr, GFP_KERNEL);
- return 0;
+
+exit_ft_unlock:
+ mutex_unlock(&dp->ftable_lock);
+exit_unlock:
+ dp_read_unlock();
+ return err;
}
static int ovs_flow_cmd_dump(struct sk_buff *skb, struct netlink_callback *cb)
@@ -1088,12 +1159,16 @@ static int ovs_flow_cmd_dump(struct sk_buff *skb,
struct netlink_callback *cb)
struct ovs_header *ovs_header = genlmsg_data(nlmsg_data(cb->nlh));
struct datapath *dp;
struct flow_table *table;
+ int ret;
+ rcu_read_lock();
dp = get_dp(ovs_header->dp_ifindex);
- if (!dp)
- return -ENODEV;
+ if (!dp) {
+ ret = -ENODEV;
+ goto exit_unlock;
+ }
- table = genl_dereference(dp->table);
+ table = rcu_dereference(dp->table);
for (;;) {
struct sw_flow *flow;
@@ -1114,7 +1189,12 @@ static int ovs_flow_cmd_dump(struct sk_buff *skb, struct
netlink_callback *cb)
cb->args[0] = bucket;
cb->args[1] = obj;
}
- return skb->len;
+ ret = skb->len;
+
+exit_unlock:
+ rcu_read_unlock();
+ return ret;
+
}
static struct genl_ops dp_flow_genl_ops[] = {
@@ -1151,7 +1231,8 @@ static struct genl_family dp_datapath_genl_family = {
.hdrsize = sizeof(struct ovs_header),
.name = OVS_DATAPATH_FAMILY,
.version = OVS_DATAPATH_VERSION,
- .maxattr = OVS_DP_ATTR_MAX
+ .maxattr = OVS_DP_ATTR_MAX,
+ .lockless = true,
};
static struct genl_multicast_group ovs_dp_datapath_multicast_group = {
@@ -1178,7 +1259,9 @@ static int ovs_dp_cmd_fill_info(struct datapath *dp,
struct sk_buff *skb,
if (err)
goto nla_put_failure;
+ rcu_read_lock();
get_dp_stats(dp, &dp_stats);
+ rcu_read_unlock();
if (nla_put(skb, OVS_DP_ATTR_STATS, sizeof(struct ovs_dp_stats),
&dp_stats))
goto nla_put_failure;
@@ -1208,7 +1291,7 @@ static struct sk_buff *ovs_dp_cmd_build_info(struct
datapath *dp, u32 pid,
return skb;
}
-/* Called with genl_mutex and optionally with RTNL lock also. */
+/* Called with dp_lock and optionally with RTNL lock also. */
static struct datapath *lookup_datapath(struct ovs_header *ovs_header,
struct nlattr *a[OVS_DP_ATTR_MAX + 1])
{
@@ -1241,15 +1324,18 @@ static int ovs_dp_cmd_new(struct sk_buff *skb, struct
genl_info *info)
goto err;
rtnl_lock();
+ dp_write_lock();
+
err = -ENODEV;
if (!try_module_get(THIS_MODULE))
- goto err_unlock_rtnl;
+ goto err_unlock;
err = -ENOMEM;
dp = kzalloc(sizeof(*dp), GFP_KERNEL);
if (dp == NULL)
goto err_put_module;
INIT_LIST_HEAD(&dp->port_list);
+ mutex_init(&dp->ftable_lock);
/* Allocate table. */
err = -ENOMEM;
@@ -1287,6 +1373,7 @@ static int ovs_dp_cmd_new(struct sk_buff *skb, struct
genl_info *info)
goto err_destroy_local_port;
list_add_tail(&dp->list_node, &dps);
+ dp_write_unlock();
rtnl_unlock();
genl_notify(reply, genl_info_net(info), info->snd_pid,
@@ -1299,12 +1386,13 @@ err_destroy_local_port:
err_destroy_percpu:
free_percpu(dp->stats_percpu);
err_destroy_table:
- ovs_flow_tbl_destroy(genl_dereference(dp->table));
+ ovs_flow_tbl_destroy(rcu_dereference_raw(dp->table));
err_free_dp:
kfree(dp);
err_put_module:
module_put(THIS_MODULE);
-err_unlock_rtnl:
+err_unlock:
+ dp_write_unlock();
rtnl_unlock();
err:
return err;
@@ -1318,6 +1406,8 @@ static int ovs_dp_cmd_del(struct sk_buff *skb, struct
genl_info *info)
int err;
rtnl_lock();
+ dp_write_lock();
+
dp = lookup_datapath(info->userhdr, info->attrs);
err = PTR_ERR(dp);
if (IS_ERR(dp))
@@ -1342,6 +1432,7 @@ static int ovs_dp_cmd_del(struct sk_buff *skb, struct
genl_info *info)
* fully destroyed before freeing the datapath.
*/
rtnl_unlock();
+ dp_write_unlock();
call_rcu(&dp->rcu, destroy_dp_rcu);
module_put(THIS_MODULE);
@@ -1353,6 +1444,7 @@ static int ovs_dp_cmd_del(struct sk_buff *skb, struct
genl_info *info)
return 0;
exit_unlock:
+ dp_write_unlock();
rtnl_unlock();
return err;
}
@@ -1363,9 +1455,12 @@ static int ovs_dp_cmd_set(struct sk_buff *skb, struct
genl_info *info)
struct datapath *dp;
int err;
+ dp_write_lock();
dp = lookup_datapath(info->userhdr, info->attrs);
- if (IS_ERR(dp))
- return PTR_ERR(dp);
+ err = PTR_ERR(dp);
+ if (IS_ERR(dp)) {
+ goto exit_unlock;
+ }
reply = ovs_dp_cmd_build_info(dp, info->snd_pid,
info->snd_seq, OVS_DP_CMD_NEW);
@@ -1373,14 +1468,16 @@ static int ovs_dp_cmd_set(struct sk_buff *skb, struct
genl_info *info)
err = PTR_ERR(reply);
netlink_set_err(init_net.genl_sock, 0,
ovs_dp_datapath_multicast_group.id, err);
- return 0;
+ err = 0;
+ goto exit_unlock;
}
genl_notify(reply, genl_info_net(info), info->snd_pid,
ovs_dp_datapath_multicast_group.id, info->nlhdr,
GFP_KERNEL);
-
- return 0;
+exit_unlock:
+ dp_write_unlock();
+ return err;
}
static int ovs_dp_cmd_get(struct sk_buff *skb, struct genl_info *info)
@@ -1388,12 +1485,16 @@ static int ovs_dp_cmd_get(struct sk_buff *skb, struct
genl_info *info)
struct sk_buff *reply;
struct datapath *dp;
+ dp_read_lock();
dp = lookup_datapath(info->userhdr, info->attrs);
- if (IS_ERR(dp))
+ if (IS_ERR(dp)) {
+ dp_read_unlock();
return PTR_ERR(dp);
+ }
reply = ovs_dp_cmd_build_info(dp, info->snd_pid,
info->snd_seq, OVS_DP_CMD_NEW);
+ dp_read_unlock();
if (IS_ERR(reply))
return PTR_ERR(reply);
@@ -1406,6 +1507,7 @@ static int ovs_dp_cmd_dump(struct sk_buff *skb, struct
netlink_callback *cb)
int skip = cb->args[0];
int i = 0;
+ dp_read_lock();
list_for_each_entry(dp, &dps, list_node) {
if (i >= skip &&
ovs_dp_cmd_fill_info(dp, skb, NETLINK_CB(cb->skb).pid,
@@ -1414,6 +1516,7 @@ static int ovs_dp_cmd_dump(struct sk_buff *skb, struct
netlink_callback *cb)
break;
i++;
}
+ dp_read_unlock();
cb->args[0] = i;
@@ -1458,7 +1561,8 @@ static struct genl_family dp_vport_genl_family = {
.hdrsize = sizeof(struct ovs_header),
.name = OVS_VPORT_FAMILY,
.version = OVS_VPORT_VERSION,
- .maxattr = OVS_VPORT_ATTR_MAX
+ .maxattr = OVS_VPORT_ATTR_MAX,
+ .lockless = true,
};
struct genl_multicast_group ovs_dp_vport_multicast_group = {
@@ -1573,6 +1677,7 @@ static int ovs_vport_cmd_new(struct sk_buff *skb, struct
genl_info *info)
goto exit;
rtnl_lock();
+ dp_read_lock();
dp = get_dp(ovs_header->dp_ifindex);
err = -ENODEV;
if (!dp)
@@ -1624,6 +1729,7 @@ static int ovs_vport_cmd_new(struct sk_buff *skb, struct
genl_info *info)
ovs_dp_vport_multicast_group.id, info->nlhdr, GFP_KERNEL);
exit_unlock:
+ dp_read_unlock();
rtnl_unlock();
exit:
return err;
@@ -1637,6 +1743,7 @@ static int ovs_vport_cmd_set(struct sk_buff *skb, struct
genl_info *info)
int err;
rtnl_lock();
+ dp_read_lock();
vport = lookup_vport(info->userhdr, a);
err = PTR_ERR(vport);
if (IS_ERR(vport))
@@ -1664,6 +1771,7 @@ static int ovs_vport_cmd_set(struct sk_buff *skb, struct
genl_info *info)
ovs_dp_vport_multicast_group.id, info->nlhdr, GFP_KERNEL);
exit_unlock:
+ dp_read_unlock();
rtnl_unlock();
return err;
}
@@ -1676,6 +1784,8 @@ static int ovs_vport_cmd_del(struct sk_buff *skb, struct
genl_info *info)
int err;
rtnl_lock();
+ dp_read_lock();
+
vport = lookup_vport(info->userhdr, a);
err = PTR_ERR(vport);
if (IS_ERR(vport))
@@ -1698,6 +1808,7 @@ static int ovs_vport_cmd_del(struct sk_buff *skb, struct
genl_info *info)
ovs_dp_vport_multicast_group.id, info->nlhdr, GFP_KERNEL);
exit_unlock:
+ dp_read_unlock();
rtnl_unlock();
return err;
}
@@ -1738,11 +1849,13 @@ static int ovs_vport_cmd_dump(struct sk_buff *skb,
struct netlink_callback *cb)
u32 port_no;
int retval;
+ rcu_read_lock();
dp = get_dp(ovs_header->dp_ifindex);
- if (!dp)
+ if (!dp) {
+ rcu_read_unlock();
return -ENODEV;
+ }
- rcu_read_lock();
for (port_no = cb->args[0]; port_no < DP_MAX_PORTS; port_no++) {
struct vport *vport;
@@ -1766,22 +1879,31 @@ static int ovs_vport_cmd_dump(struct sk_buff *skb,
struct netlink_callback *cb)
static void rehash_flow_table(struct work_struct *work)
{
struct datapath *dp;
+ int interval = REHASH_FLOW_INTERVAL;
- genl_lock();
+ if (!rtnl_trylock()) {
+ interval = REHASH_FLOW_INTERVAL / 2;
+ goto resched;
+ }
+ dp_read_lock();
list_for_each_entry(dp, &dps, list_node) {
- struct flow_table *old_table = genl_dereference(dp->table);
+ struct flow_table *old_table;
struct flow_table *new_table;
+ mutex_lock(&dp->ftable_lock);
+ old_table = flow_table_ft_lock(dp);
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);
}
+ mutex_unlock(&dp->ftable_lock);
}
+ dp_read_unlock();
+ rtnl_unlock();
- genl_unlock();
-
+resched:
schedule_delayed_work(&rehash_flow_wq, REHASH_FLOW_INTERVAL);
}
diff --git a/net/openvswitch/datapath.h b/net/openvswitch/datapath.h
index c73370c..fed4056 100644
--- a/net/openvswitch/datapath.h
+++ b/net/openvswitch/datapath.h
@@ -57,7 +57,8 @@ struct dp_stats_percpu {
* @rcu: RCU callback head for deferred destruction.
* @list_node: Element in global 'dps' list.
* @n_flows: Number of flows currently in flow table.
- * @table: Current flow table. Protected by genl_lock and RCU.
+ * @table: Current flow table. Protected by %ftable_lock and RCU.
+ * @ftable_lock: Flow-table lock.
* @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
@@ -73,6 +74,7 @@ struct datapath {
/* Flow table. */
struct flow_table __rcu *table;
+ struct mutex ftable_lock;
/* Switch ports. */
struct vport __rcu *ports[DP_MAX_PORTS];
diff --git a/net/openvswitch/vport.c b/net/openvswitch/vport.c
index 6c066ba..8a02293 100644
--- a/net/openvswitch/vport.c
+++ b/net/openvswitch/vport.c
@@ -86,6 +86,7 @@ struct vport *ovs_vport_locate(const char *name)
struct vport *vport;
struct hlist_node *node;
+ WARN_ON(!rcu_read_lock_held() && !rtnl_is_locked());
hlist_for_each_entry_rcu(vport, node, bucket, hash_node)
if (!strcmp(name, vport->ops->get_name(vport)))
return vport;
--
1.7.10
_______________________________________________
dev mailing list
dev@openvswitch.org
http://openvswitch.org/mailman/listinfo/dev
