[ovs-dev] [PATCH 44/62] dpif-netdev: use skb_priority to substitute mod_flag
Tao YunXiang
taoyunxiang at cmss.chinamobile.com
Mon Dec 28 09:25:02 UTC 2020
From: Taoyunxiang <taoyunxiang at cmss.chinamobile.com>
Code Source From: Self Code
Description:
As packet->md.mod_flag would be the last mod_flag status, but we need to
get mod status in time, so we reuse match.flow.skb_priority (it won't be
used before). And retrive packet->md.skb_priority.
Jira: #[Optional]
市场项目编号（名称）：[Optional]
---
lib/dpif-netdev.c | 64 ++++++++++++++++++++++++++++++++++-------------
lib/netdev-offload-dpdk.c | 3 ++-
lib/odp-execute.c | 4 +--
lib/packets.h | 6 ++---
4 files changed, 53 insertions(+), 24 deletions(-)
diff --git a/lib/dpif-netdev.c b/lib/dpif-netdev.c
index 4beea5b..b96d12d 100644
--- a/lib/dpif-netdev.c
+++ b/lib/dpif-netdev.c
@@ -2475,6 +2475,7 @@ dp_netdev_flow_offload_put(struct dp_flow_offload_item *offload)
info.mod_flag = offload->mod_flag;
info.nat_action = offload->nat_action;
info.ct_enable = false;
+ info.group_id = 0;
port = netdev_ports_get(in_port, dpif_type_str);
if (!port) {
@@ -2587,14 +2588,17 @@ queue_netdev_flow_del(struct dp_netdev_pmd_thread *pmd,
/* return value show if this flow should be offloaded*/
static bool
parse_netdev_flow_put(struct match *match, const struct nlattr *actions,
- size_t actions_len, struct nat_action_info_t *nat_action,
- uint8_t mod_flag)
+ size_t actions_len, struct nat_action_info_t *nat_action
+ )
{
bool action_has_recirc = false;
bool action_has_set = false;
+ bool action_has_dnat = false;
+ bool action_has_ct = false;
bool ret = false;
struct nlattr *nla;
size_t left;
+ uint32_t mod_flag = match->flow.skb_priority;
/* filter non IP pkt out */
if ((match->flow.dl_type != htons(ETH_TYPE_IP)) &&
@@ -2612,6 +2616,7 @@ parse_netdev_flow_put(struct match *match, const struct nlattr *actions,
const struct nlattr *b;
unsigned int ct_left;
+ action_has_ct = true;
if (nat_action == NULL) {
continue;
}
@@ -2636,6 +2641,8 @@ parse_netdev_flow_put(struct match *match, const struct nlattr *actions,
nat_action->nat_action |=
((sub_type_nest == OVS_NAT_ATTR_SRC)
? NAT_ACTION_SRC : NAT_ACTION_DST);
+ action_has_dnat = (sub_type_nest == OVS_NAT_ATTR_DST)?
+ true : action_has_dnat;
break;
case OVS_NAT_ATTR_IP_MIN:
memcpy(&(nat_action->min_addr),
@@ -2711,12 +2718,23 @@ parse_netdev_flow_put(struct match *match, const struct nlattr *actions,
goto out;
}
+ /*if only has one conntrack action, won't need to offload */
+ if ((actions_len == 1) && action_has_ct) {
+ goto out;
+ }
/* no mod_flag means: 1. no set action at all
* 2. set action but is the first translated pkt
+ * 3. nat action then set mac/ttl: such as DNAT
* For 1 , only established pkt should try to offload
* For 2 , still established pkt should be offload
- * so in this condition: we cannot offload non-est pkt
+ * For 3 , we don't care ct_state,and try to offload
+ * so for 1 && 2: we cannot offload non-est pkt
+ * for 3, we try to offload non-est pkt
*/
+ if (action_has_dnat) {
+ ret = true;
+ goto out;
+ }
if (match->wc.masks.ct_state &&
!(match->wc.masks.ct_state & match->flow.ct_state & CS_ESTABLISHED) ){
goto out;
@@ -2746,7 +2764,7 @@ out:
static void
queue_netdev_flow_put(struct dp_netdev_pmd_thread *pmd,
struct dp_netdev_flow *flow, struct match *match,
- const struct nlattr *actions, size_t actions_len,uint8_t mod_flag)
+ const struct nlattr *actions, size_t actions_len)
{
struct dp_flow_offload_item *offload;
int op;
@@ -2755,10 +2773,20 @@ queue_netdev_flow_put(struct dp_netdev_pmd_thread *pmd,
if (!netdev_is_flow_api_enabled()) {
return;
}
- if (!parse_netdev_flow_put(match,actions,actions_len,&nat_action,mod_flag)) {
+ if (!parse_netdev_flow_put(match,actions,actions_len,&nat_action)) {
return;
}
+ VLOG_DBG("TIMO DBG: in queue_netdev_flow_put:recirc %x mod %x \n", match->flow.recirc_id,
+ match->flow.skb_priority);
+ VLOG_DBG("TIMO DBG: in queue_netdev_flow_put:nw_src %x ct_nw_src %x \n",
+ ntohl(match->flow.nw_src), ntohl(match->flow.ct_nw_src));
+ VLOG_DBG("TIMO DBG: in queue_netdev_flow_put:nat action %x \n",
+ nat_action.nat_action);
+ VLOG_DBG("TIMO DBG: in queue_netdev_flow_put:nat min addr %x \n",
+ ntohl(nat_action.min_addr.ipv4));
+ VLOG_DBG("TIMO DBG: in queue_netdev_flow_put:nat max addr %x \n",
+ ntohl(nat_action.max_addr.ipv4));
if (ovsthread_once_start(&offload_thread_once)) {
xpthread_cond_init(&dp_flow_offload.cond, NULL);
ovs_thread_create("dp_netdev_flow_offload",
@@ -2776,7 +2804,7 @@ queue_netdev_flow_put(struct dp_netdev_pmd_thread *pmd,
offload->actions = xmalloc(actions_len);
memcpy(offload->actions, actions, actions_len);
offload->actions_len = actions_len;
- offload->mod_flag = mod_flag;
+ offload->mod_flag = match->flow.skb_priority;
offload->nat_action = nat_action;
dp_netdev_append_flow_offload(offload);
@@ -3538,8 +3566,7 @@ dp_netdev_get_mega_ufid(const struct match *match, ovs_u128 *mega_ufid)
static struct dp_netdev_flow *
dp_netdev_flow_add(struct dp_netdev_pmd_thread *pmd,
struct match *match, const ovs_u128 *ufid,
- const struct nlattr *actions, size_t actions_len,
- uint8_t mod_flag)
+ const struct nlattr *actions, size_t actions_len)
OVS_REQUIRES(pmd->flow_mutex)
{
struct ds extra_info = DS_EMPTY_INITIALIZER;
@@ -3604,7 +3631,7 @@ dp_netdev_flow_add(struct dp_netdev_pmd_thread *pmd,
/*Don't try to offload flows that has been offloaded failed */
if (megaflow_to_offload_st_find(&flow->ufid) == false) {
- queue_netdev_flow_put(pmd, flow, match, actions, actions_len,mod_flag);
+ queue_netdev_flow_put(pmd, flow, match, actions, actions_len);
}
if (OVS_UNLIKELY(!VLOG_DROP_DBG((&upcall_rl)))) {
@@ -3675,7 +3702,7 @@ flow_put_on_pmd(struct dp_netdev_pmd_thread *pmd,
if (put->flags & DPIF_FP_CREATE) {
if (cmap_count(&pmd->flow_table) < MAX_FLOWS) {
dp_netdev_flow_add(pmd, match, ufid, put->actions,
- put->actions_len,(uint8_t)0);
+ put->actions_len);
error = 0;
} else {
error = EFBIG;
@@ -3697,7 +3724,7 @@ flow_put_on_pmd(struct dp_netdev_pmd_thread *pmd,
/*Don't try to offload flows that has been offloaded failed */
if (megaflow_to_offload_st_find(&netdev_flow->ufid) == false) {
queue_netdev_flow_put(pmd, netdev_flow, match,
- put->actions, put->actions_len,(uint8_t)0);
+ put->actions, put->actions_len);
}
if (stats) {
@@ -6966,7 +6993,6 @@ handle_packet_upcall(struct dp_netdev_pmd_thread *pmd,
ovs_u128 ufid;
int error;
uint64_t cycles = cycles_counter_update(&pmd->perf_stats);
- uint8_t mod_flag;
match.tun_md.valid = false;
miniflow_expand(&key->mf, &match.flow);
@@ -6999,10 +7025,6 @@ handle_packet_upcall(struct dp_netdev_pmd_thread *pmd,
* the actions. Otherwise, if there are any slow path actions,
* we'll send the packet up twice. */
- /* get mod_flag before execute action, so can set group_id=0 for first
- * loopbacked pkt with set action
- */
- mod_flag = packet->md.mod_flag;
dp_packet_batch_init_packet(&b, packet);
dp_netdev_execute_actions(pmd, &b, true, &match.flow,
actions->data, actions->size);
@@ -7022,8 +7044,7 @@ handle_packet_upcall(struct dp_netdev_pmd_thread *pmd,
if (OVS_LIKELY(!netdev_flow)) {
netdev_flow = dp_netdev_flow_add(pmd, &match, &ufid,
add_actions->data,
- add_actions->size,
- mod_flag);
+ add_actions->size);
}
ovs_mutex_unlock(&pmd->flow_mutex);
uint32_t hash = dp_netdev_flow_hash(&netdev_flow->ufid);
@@ -7651,6 +7672,13 @@ dp_execute_cb(void *aux_, struct dp_packet_batch *packets_,
nat_action_info.nat_action |=
((sub_type_nest == OVS_NAT_ATTR_SRC)
? NAT_ACTION_SRC : NAT_ACTION_DST);
+ /*For DNAT, we should also set mod_flag */
+ if (sub_type_nest == OVS_NAT_ATTR_DST) {
+ struct dp_packet *packet;
+ DP_PACKET_BATCH_FOR_EACH (i, packet, packets_) {
+ packet->md.skb_priority |= 1 << SET_ACTION_SET ;
+ }
+ }
break;
case OVS_NAT_ATTR_IP_MIN:
memcpy(&nat_action_info.min_addr,
diff --git a/lib/netdev-offload-dpdk.c b/lib/netdev-offload-dpdk.c
index d4a80bb..12aa5c1 100644
--- a/lib/netdev-offload-dpdk.c
+++ b/lib/netdev-offload-dpdk.c
@@ -507,7 +507,7 @@ dump_flow_action(struct ds *s, const struct rte_flow_action *actions)
if (vlan_tci) {
ds_put_format(s, "rte flow vlan-push action:\n");
- ds_put_format(s, "vlan-encap: tpid=%"PRIu16"\n",ntohs(vlan_tci->ethertype));
+ ds_put_format(s, "vlan-encap: tpid=0x%06x\n",ntohs(vlan_tci->ethertype));
} else {
ds_put_format(s, "vlan-encap: null\n");
}
@@ -1322,6 +1322,7 @@ add_ipv4_nat_action(struct flow_actions *actions,
struct nat_action_info_t *nat_action)
{
+ VLOG_DBG("TIMO DBG: in add_ipv4_nat_action");
if (nat_action->nat_action & NAT_ACTION_SRC) {
__be32 ipv4_src;
diff --git a/lib/odp-execute.c b/lib/odp-execute.c
index b3cb71b..559f260 100644
--- a/lib/odp-execute.c
+++ b/lib/odp-execute.c
@@ -955,14 +955,14 @@ odp_execute_actions(void *dp, struct dp_packet_batch *batch, bool steal,
case OVS_ACTION_ATTR_SET:
DP_PACKET_BATCH_FOR_EACH (i, packet, batch) {
odp_execute_set_action(packet, nl_attr_get(a));
- packet->md.mod_flag |= 1 << SET_ACTION_SET ;
+ packet->md.skb_priority |= 1 << SET_ACTION_SET ;
}
break;
case OVS_ACTION_ATTR_SET_MASKED:
DP_PACKET_BATCH_FOR_EACH(i, packet, batch) {
odp_execute_masked_set_action(packet, nl_attr_get(a));
- packet->md.mod_flag |= 1 << SET_ACTION_SET ;
+ packet->md.skb_priority |= 1 << SET_ACTION_SET ;
}
break;
diff --git a/lib/packets.h b/lib/packets.h
index 4eafaca..84c554e 100644
--- a/lib/packets.h
+++ b/lib/packets.h
@@ -101,7 +101,9 @@ PADDED_MEMBERS_CACHELINE_MARKER(CACHE_LINE_SIZE, cacheline0,
received from the wire. */
uint32_t dp_hash; /* hash value computed by the recirculation
action. */
- uint32_t skb_priority; /* Packet priority for QoS. */
+ uint32_t skb_priority; /* Packet priority for QoS.
+ * together with mod_flag
+ */
uint32_t pkt_mark; /* Packet mark. */
uint8_t ct_state; /* Connection state. */
bool ct_orig_tuple_ipv6;
@@ -112,7 +114,6 @@ PADDED_MEMBERS_CACHELINE_MARKER(CACHE_LINE_SIZE, cacheline0,
struct conn *conn; /* Cached conntrack connection. */
bool reply; /* True if reply direction. */
bool icmp_related; /* True if ICMP related. */
- uint8_t mod_flag; /* Modify flag */
);
@@ -170,7 +171,6 @@ pkt_metadata_init(struct pkt_metadata *md, odp_port_t port)
md->tunnel.ipv6_dst = in6addr_any;
md->in_port.odp_port = port;
md->conn = NULL;
- md->mod_flag = 0;
}
/* This function prefetches the cachelines touched by pkt_metadata_init()
--
1.8.3.1
More information about the dev
mailing list