Skip to content

Commit

Permalink
[EVPN-MH] Handling NHID in Bridge FDB Update and handle NHG update fr…
Browse files Browse the repository at this point in the history 
…om the Kernel

Why I did it
Managing NHID in Bridge FDB Updates and Handling NHG Updates from the Kernel in fdbsyncd

Check HLD: sonic-net/SONiC#1702

Signed-off-by: Kishore Kunal <[email protected]>
  • Loading branch information
@kishorekunal01
kishorekunal01 committed Nov 15, 2024
1 parent f650a3b commit 20e284e
Show file tree
Hide file tree
Showing 3 changed files with 288 additions and 12 deletions.
262 changes: 251 additions & 11 deletions fdbsyncd/fdbsync.cpp
View file
Original file line number Diff line number Diff line change
Expand Up @@ -26,13 +26,15 @@ FdbSync::FdbSync(RedisPipeline *pipelineAppDB, DBConnector *stateDb, DBConnector
m_imetTable(pipelineAppDB, APP_VXLAN_REMOTE_VNI_TABLE_NAME),
m_fdbStateTable(stateDb, STATE_FDB_TABLE_NAME),
m_mclagRemoteFdbStateTable(stateDb, STATE_MCLAG_REMOTE_FDB_TABLE_NAME),
m_l2NhgTable(pipelineAppDB, APP_L2_NEXTHOP_GROUP_TABLE_NAME),
m_cfgEvpnNvoTable(config_db, CFG_VXLAN_EVPN_NVO_TABLE_NAME)
{
m_AppRestartAssist = new AppRestartAssist(pipelineAppDB, "fdbsyncd", "swss", DEFAULT_FDBSYNC_WARMSTART_TIMER);
if (m_AppRestartAssist)
{
m_AppRestartAssist->registerAppTable(APP_VXLAN_FDB_TABLE_NAME, &m_fdbTable);
m_AppRestartAssist->registerAppTable(APP_VXLAN_REMOTE_VNI_TABLE_NAME, &m_imetTable);
m_AppRestartAssist->registerAppTable(APP_L2_NEXTHOP_GROUP_TABLE_NAME, &m_l2NhgTable);
}
}

Expand Down Expand Up @@ -647,7 +649,7 @@ void FdbSync::macDelVxlanDB(string key)

}

void FdbSync::macAddVxlan(string key, struct in_addr vtep, string type, uint32_t vni, string intf_name)
void FdbSync::macAddVxlan(string key, struct in_addr vtep, string type, uint32_t vni, string intf_name, int nhid)
{
string svtep = inet_ntoa(vtep);
string svni = to_string(vni);
Expand All @@ -656,16 +658,27 @@ void FdbSync::macAddVxlan(string key, struct in_addr vtep, string type, uint32_t
m_mac[key] = {svtep, type, vni, intf_name};

std::vector<FieldValueTuple> fvVector;
if (nhid > 0)
{
string nhid_str = to_string(nhid);
FieldValueTuple rv("nexthop_group", nhid_str);
fvVector.push_back(rv);
}
else
{
FieldValueTuple rv("remote_vtep", svtep);
fvVector.push_back(rv);
}

FieldValueTuple rv("remote_vtep", svtep);
FieldValueTuple t("type", type);
FieldValueTuple v("vni", svni);
fvVector.push_back(rv);
fvVector.push_back(t);
fvVector.push_back(v);

SWSS_LOG_INFO("%sVXLAN_FDB_TABLE: ADD_KEY %s vtep:%s type:%s",
SWSS_LOG_INFO("%sVXLAN_FDB_TABLE: ADD_KEY %s vtep:%s NHID:%d type:%s",
m_AppRestartAssist->isWarmStartInProgress() ? "WARM-RESTART:" : "" ,
key.c_str(), svtep.c_str(), type.c_str());
key.c_str(), svtep.c_str(), nhid, type.c_str());
// If warmstart is in progress, we take all netlink changes into the cache map
if (m_AppRestartAssist->isWarmStartInProgress())
{
Expand Down Expand Up @@ -704,6 +717,7 @@ void FdbSync::onMsgNbr(int nlmsg_type, struct nl_object *obj)
string type = "";
string vlan_id = "";
bool isVxlanIntf = false;
int nhid = -1;

if ((nlmsg_type != RTM_NEWNEIGH) && (nlmsg_type != RTM_GETNEIGH) &&
(nlmsg_type != RTM_DELNEIGH))
Expand Down Expand Up @@ -772,16 +786,24 @@ void FdbSync::onMsgNbr(int nlmsg_type, struct nl_object *obj)
return;
}

vtep_addr = rtnl_neigh_get_dst(neigh);
if (vtep_addr == NULL)
nhid = rtnl_neigh_get_nh_id(neigh);
if (nhid < 0)
{
return;
vtep_addr = rtnl_neigh_get_dst(neigh);
if (vtep_addr == NULL)
{
return;
}
else
{
/* Currently we only support ipv4 tunnel endpoints */
vtep.s_addr = *(uint32_t *)nl_addr_get_binary_addr(vtep_addr);
SWSS_LOG_INFO("Tunnel IP %s Int%d", inet_ntoa(vtep), *(uint32_t *)nl_addr_get_binary_addr(vtep_addr));
}
}
else
{
/* Currently we only support ipv4 tunnel endpoints */
vtep.s_addr = *(uint32_t *)nl_addr_get_binary_addr(vtep_addr);
SWSS_LOG_INFO("Tunnel IP %s Int%d", inet_ntoa(vtep), *(uint32_t *)nl_addr_get_binary_addr(vtep_addr));
SWSS_LOG_INFO("L2 NHID %d", nhid);
}

int state = rtnl_neigh_get_state(neigh);
Expand Down Expand Up @@ -826,7 +848,7 @@ void FdbSync::onMsgNbr(int nlmsg_type, struct nl_object *obj)

if (!delete_key)
{
macAddVxlan(key, vtep, type, vni, ifname);
macAddVxlan(key, vtep, type, vni, ifname, nhid);
}
else
{
Expand Down Expand Up @@ -883,3 +905,221 @@ void FdbSync::onMsg(int nlmsg_type, struct nl_object *obj)
}
}

void FdbSync::nhgDelDB(struct m_nhg_info *nhg_info)
{
string key = to_string(nhg_info->nhid);

std::vector<FieldValueTuple> fvVector;
FieldValueTuple grp("nexthop_group", nhg_info->group_str);
FieldValueTuple vtep("remote_vtep", nhg_info->vtep_str);

SWSS_LOG_NOTICE("%sL2_NHG_TABLE: DEL_KEY %s vtep:%s grp:%s",
m_AppRestartAssist->isWarmStartInProgress() ? "WARM-RESTART:" : "" ,
key.c_str(), nhg_info->vtep_str.c_str(), nhg_info->group_str.c_str());
if (m_AppRestartAssist->isWarmStartInProgress())
{
m_AppRestartAssist->insertToMap(APP_L2_NEXTHOP_GROUP_TABLE_NAME, key, fvVector, true);
return;
}

m_l2NhgTable.del(key);
return;
}


void FdbSync::nhgUpdateDB(struct m_nhg_info *nhg_info)
{
int nhid = nhg_info->nhid;
std::vector<FieldValueTuple> fvVector;
string key = to_string(nhid);

FieldValueTuple vtep("remote_vtep", nhg_info->vtep_str);
FieldValueTuple group("nexthop_group", nhg_info->group_str);

fvVector.push_back(vtep);
fvVector.push_back(group);

SWSS_LOG_NOTICE("%sL2_NEXTHOP_GROUP_TABLE: ADD_KEY %s vtep:%s group: %s",
m_AppRestartAssist->isWarmStartInProgress() ? "WARM-RESTART:" : "" ,
key.c_str(), nhg_info->vtep_str.c_str(), nhg_info->group_str.c_str());

// If warmstart is in progress, we take all netlink changes into the cache map
if (m_AppRestartAssist->isWarmStartInProgress())
{
m_AppRestartAssist->insertToMap(APP_L2_NEXTHOP_GROUP_TABLE_NAME, key, fvVector, false);
return;
}

m_l2NhgTable.set(key, fvVector);
return;
}

void FdbSync::processNhgNetlinkData(struct m_nhg_info *entry)
{
if (entry->op == NH_ADD)
{
nhgUpdateDB(entry);
}
else
{
nhgDelDB(entry);
}
return;
}

void FdbSync::netlink_parse_rtattr(struct rtattr **tb, int max, struct rtattr *rta,
int len)
{
while (RTA_OK(rta, len)) {
if (rta->rta_type <= max)
{
tb[rta->rta_type] = rta;
}
else
{
if (rta->rta_type & NLA_F_NESTED)
{
int rta_type = rta->rta_type & ~NLA_F_NESTED;
if (rta_type <= max)
{
tb[rta_type] = rta;
}
}
}
rta = RTA_NEXT(rta, len);
}
}

void FdbSync::nhgAddGroup(int nhid, struct nexthop_grp *nhid_grp, long unsigned int count,
string &nh_id_grp_str)
{
char nh_id_grp[IFNAMSIZ + 1] = {0};

for (long unsigned int i = 0; i < count; i++)
{
memset(nh_id_grp,0, IFNAMSIZ);
snprintf(nh_id_grp, IFNAMSIZ,"%d", nhid_grp[i].id);
if (i > 0)
{
nh_id_grp_str += ",";
}
nh_id_grp_str += nh_id_grp;
}
return;
}

void FdbSync::onMsgNhg(struct nlmsghdr *h, int len)
{
struct nhmsg *nhm;
struct rtattr *tb[NHA_MAX + 1];
void *gw = NULL;
char anyaddr[MAX_ADDR_SIZE] = {0};
char gwaddr[MAX_ADDR_SIZE] = {0};
char nexthopaddr[MAX_ADDR_SIZE] = {0};
int gw_af;
int nlmsg_type = h->nlmsg_type;
unsigned int id = 0;
struct m_nhg_info nhg_entry;
nhm = (struct nhmsg *)NLMSG_DATA(h);
char op = NH_ADD;

/* Parse attributes and extract fields of interest. */
memset(tb, 0, sizeof(tb));
netlink_parse_rtattr(tb, NHA_MAX, NHA_RTA(nhm), len);

if (!tb[NHA_FDB])
return;

if (tb[NHA_ID])
{
id = *(uint32_t *) RTA_DATA(tb[NHA_ID]);
}

if( RTM_DELNEXTHOP == nlmsg_type)
{
SWSS_LOG_NOTICE("NHG DELETE MSG type %d nhid %d",
nlmsg_type, id);
op = NH_DEL;
}

if (tb[NHA_GROUP])
{
//Recursive/ECMP NHG
long unsigned int count = 0;
struct nexthop_grp *grp;
string nh_id_grp_str = "";

count = RTA_PAYLOAD(tb[NHA_GROUP])/(sizeof (struct nexthop_grp));
grp = (struct nexthop_grp *) RTA_DATA(tb[NHA_GROUP]);
SWSS_LOG_NOTICE("NHA_GROUP MSG type %d zebra_nhid %d count %ld", nlmsg_type, id, count);

nhgAddGroup(id, grp, count, nh_id_grp_str);

nhg_entry.nhid = id;
nhg_entry.vtep_str = string("");
nhg_entry.group_str = nh_id_grp_str;

SWSS_LOG_NOTICE("Set Grp nhid %d nhg:%s op:%d", id, nh_id_grp_str.c_str(), op);
}
else
{
if(tb[NHA_GATEWAY])
{
gw = RTA_DATA(tb[NHA_GATEWAY]);
}
else
{
gw = anyaddr;
}

gw_af = nhm->nh_family;
if (gw_af == AF_INET)
{
memcpy(gwaddr, gw, IPV4_MAX_BYTE);
}
else if (gw_af == AF_INET6)
{
memcpy(gwaddr, gw, IPV6_MAX_BYTE);
}
inet_ntop(gw_af, gwaddr, nexthopaddr, MAX_ADDR_SIZE);

nhg_entry.nhid = id;
nhg_entry.vtep_str = string(nexthopaddr);
nhg_entry.group_str = string("");

SWSS_LOG_NOTICE("set nhid %d gw:%s op:%d",
id, nexthopaddr, op);
}

nhg_entry.op = op;

processNhgNetlinkData(&nhg_entry);

return;
}

void FdbSync::onMsgRaw(struct nlmsghdr *h)
{
int len;
int nlmsg_type = h->nlmsg_type;

if ((h->nlmsg_type != RTM_NEWNEXTHOP)
&& (h->nlmsg_type != RTM_DELNEXTHOP))
return;

if( nlmsg_type == RTM_NEWNEXTHOP
|| nlmsg_type == RTM_DELNEXTHOP)
{
/* Length validity. */
len = (int)(h->nlmsg_len - NLMSG_LENGTH(sizeof(struct nhmsg)));
if (len < 0)
{
SWSS_LOG_ERROR("%s: Message received from netlink is of a broken size %d %zu",
__PRETTY_FUNCTION__, h->nlmsg_len,
(size_t)NLMSG_LENGTH(sizeof(struct nhmsg)));
return;
}
onMsgNhg(h, len);
}
return;
}
26 changes: 25 additions & 1 deletion fdbsyncd/fdbsync.h
View file
Original file line number Diff line number Diff line change
Expand Up @@ -3,6 +3,7 @@

#include <string>
#include <arpa/inet.h>
#include <linux/nexthop.h>
#include "dbconnector.h"
#include "producerstatetable.h"
#include "subscriberstatetable.h"
Expand All @@ -21,6 +22,10 @@
*/
#define INTF_RESTORE_MAX_WAIT_TIME 180

#define NHA_RTA(r) ((struct rtattr*)(((char*)(r)) + NLMSG_ALIGN(sizeof(struct nhmsg))))
#define IPV4_MAX_BYTE 4
#define IPV6_MAX_BYTE 16

namespace swss {

enum FDB_OP_TYPE {
Expand All @@ -42,6 +47,15 @@ struct m_fdb_info
short op_type; /*add or del*/
};

struct m_nhg_info {
int nhid;
std::string vtep_str;
std::string group_str;
char op;
#define NH_ADD 1
#define NH_DEL 2
};

class FdbSync : public NetMsg
{
public:
Expand All @@ -51,6 +65,7 @@ class FdbSync : public NetMsg
~FdbSync();

virtual void onMsg(int nlmsg_type, struct nl_object *obj);
virtual void onMsgRaw(struct nlmsghdr *obj);

bool isIntfRestoreDone();

Expand Down Expand Up @@ -87,6 +102,7 @@ class FdbSync : public NetMsg
private:
ProducerStateTable m_fdbTable;
ProducerStateTable m_imetTable;
ProducerStateTable m_l2NhgTable;
SubscriberStateTable m_fdbStateTable;
SubscriberStateTable m_mclagRemoteFdbStateTable;
AppRestartAssist *m_AppRestartAssist;
Expand Down Expand Up @@ -146,13 +162,21 @@ class FdbSync : public NetMsg
std::unordered_map<int, intf> m_intf_info;

void addLocalMac(std::string key, std::string op);
void macAddVxlan(std::string key, struct in_addr vtep, std::string type, uint32_t vni, std::string intf_name);
void macAddVxlan(std::string key, struct in_addr vtep, std::string type, uint32_t vni, std::string intf_name, int nhid);
void macDelVxlan(std::string auxkey);
void macDelVxlanDB(std::string key);
void imetAddRoute(struct in_addr vtep, std::string ifname, uint32_t vni);
void imetDelRoute(struct in_addr vtep, std::string ifname, uint32_t vni);
void onMsgNbr(int nlmsg_type, struct nl_object *obj);
void onMsgLink(int nlmsg_type, struct nl_object *obj);

//NHG routines
void onMsgNhg(struct nlmsghdr *h, int len);
void nhgAddGroup(int nhid, struct nexthop_grp *nhid_grp, long unsigned int count, std::string &nh_id_grp_str);
void processNhgNetlinkData(struct m_nhg_info *entry);
void netlink_parse_rtattr(struct rtattr **tb, int max, struct rtattr *rta, int len);
void nhgUpdateDB(struct m_nhg_info *nhginfo);
void nhgDelDB(struct m_nhg_info *nhginfo);
};

}
Expand Down
Loading

0 comments on commit 20e284e

Please sign in to comment.