blob: 8a4db05abb2b3e31c738f707fec2daebf108ae07 [file] [log] [blame]
/*
Copyright (c) 2013-2021, The Linux Foundation. All rights reserved.
Redistribution and use in source and binary forms, with or without
modification, are permitted provided that the following conditions are
met:
* Redistributions of source code must retain the above copyright
notice, this list of conditions and the following disclaimer.
* Redistributions in binary form must reproduce the above
copyright notice, this list of conditions and the following
disclaimer in the documentation and/or other materials provided
with the distribution.
* Neither the name of The Linux Foundation nor the names of its
contributors may be used to endorse or promote products derived
from this software without specific prior written permission.
THIS SOFTWARE IS PROVIDED "AS IS" AND ANY EXPRESS OR IMPLIED
WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE IMPLIED WARRANTIES OF
MERCHANTABILITY, FITNESS FOR A PARTICULAR PURPOSE AND NON-INFRINGEMENT
ARE DISCLAIMED. IN NO EVENT SHALL THE COPYRIGHT OWNER OR CONTRIBUTORS
BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR
CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF
SUBSTITUTE GOODS OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR
BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY,
WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT (INCLUDING NEGLIGENCE
OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE OF THIS SOFTWARE, EVEN
IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
*
* Changes from Qualcomm Innovation Center are provided under the following license:
*
* Copyright (c) 2022 Qualcomm Innovation Center, Inc. All rights reserved.
*
* Redistribution and use in source and binary forms, with or without
* modification, are permitted (subject to the limitations in the
* disclaimer below) provided that the following conditions are met:
*
* * Redistributions of source code must retain the above copyright
* notice, this list of conditions and the following disclaimer.
*
* * Redistributions in binary form must reproduce the above
* copyright notice, this list of conditions and the following
* disclaimer in the documentation and/or other materials provided
* with the distribution.
*
* * Neither the name of Qualcomm Innovation Center, Inc. nor the names of its
* contributors may be used to endorse or promote products derived
* from this software without specific prior written permission.
*
* NO EXPRESS OR IMPLIED LICENSES TO ANY PARTY'S PATENT RIGHTS ARE
* GRANTED BY THIS LICENSE. THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT
* HOLDERS AND CONTRIBUTORS "AS IS" AND ANY EXPRESS OR IMPLIED
* WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE IMPLIED WARRANTIES OF
* MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE ARE DISCLAIMED.
* IN NO EVENT SHALL THE COPYRIGHT HOLDER OR CONTRIBUTORS BE LIABLE FOR
* ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
* DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE
* GOODS OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS
* INTERRUPTION) HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER
* IN CONTRACT, STRICT LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR
* OTHERWISE) ARISING IN ANY WAY OUT OF THE USE OF THIS SOFTWARE, EVEN
* IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE
*/
/*!
@file
IPACM_Lan.cpp
@brief
This file implements the LAN iface functionality.
@Author
Skylar Chang
*/
#include <string.h>
#include <fcntl.h>
#include <sys/ioctl.h>
#ifndef in_addr_t
typedef uint32_t in_addr_t;
#endif
#include "IPACM_Netlink.h"
#include "IPACM_Lan.h"
#include "IPACM_Wan.h"
#include "IPACM_IfaceManager.h"
#include "linux/rmnet_ipa_fd_ioctl.h"
#include "linux/ipa_qmi_service_v01.h"
#include "linux/msm_ipa.h"
#include "IPACM_ConntrackListener.h"
#include <sys/ioctl.h>
#include <fcntl.h>
#ifdef FEATURE_IPACM_HAL
#include "IPACM_OffloadManager.h"
#endif
bool IPACM_Lan::odu_up = false;
struct ipa_lan_downstream_info IPACM_Lan::downstream_info[IPA_MAX_TETHER_IFACE_ENTRIES];
IPACM_Lan::IPACM_Lan(int iface_index) : IPACM_Iface(iface_index)
{
num_eth_client = 0;
header_name_count = 0;
ipv6_set = 0;
ipv4_header_set = false;
ipv6_header_set = false;
odu_route_rule_v4_hdl = NULL;
odu_route_rule_v6_hdl = NULL;
eth_client = NULL;
int m_fd_odu, ret = IPACM_SUCCESS;
uint32_t i;
Nat_App = NatApp::GetInstance();
if (Nat_App == NULL)
{
IPACMERR("unable to get Nat App instance \n");
return;
}
num_wan_ul_fl_rule_v4 = 0;
num_wan_ul_fl_rule_v6 = 0;
is_active = true;
modem_ul_v4_set = false;
v4_mux_id = 0;
modem_ul_v6_set = false;
v6_mux_id = 0;
sta_ul_v4_set = false;
sta_ul_v6_set = false;
is_mode_switch = false;
if_ipv4_subnet =0;
each_client_rt_rule_count[IPA_IP_v4] = 0;
each_client_rt_rule_count[IPA_IP_v6] = 0;
eth_client_len = 0;
/* support eth multiple clients */
if(iface_query != NULL)
{
if(ipa_if_cate != WLAN_IF)
{
eth_client_len = (sizeof(ipa_eth_client)) + (iface_query->num_tx_props * sizeof(eth_client_rt_hdl));
eth_client = (ipa_eth_client *)calloc(IPA_MAX_NUM_ETH_CLIENTS, eth_client_len);
if (eth_client == NULL)
{
IPACMERR("unable to allocate memory\n");
return;
}
}
IPACMDBG_H(" IPACM->IPACM_Lan(%d) constructor: Tx:%d Rx:%d \n", ipa_if_num,
iface_query->num_tx_props, iface_query->num_rx_props);
/* ODU routing table initilization */
if(ipa_if_cate == ODU_IF)
{
odu_route_rule_v4_hdl = (uint32_t *)calloc(iface_query->num_tx_props, sizeof(uint32_t));
odu_route_rule_v6_hdl = (uint32_t *)calloc(iface_query->num_tx_props, sizeof(uint32_t));
if ((odu_route_rule_v4_hdl == NULL) || (odu_route_rule_v6_hdl == NULL))
{
IPACMERR("unable to allocate memory\n");
return;
}
}
}
memset(wan_ul_fl_rule_hdl_v4, 0, MAX_WAN_UL_FILTER_RULES * sizeof(uint32_t));
memset(wan_ul_fl_rule_hdl_v6, 0, MAX_WAN_UL_FILTER_RULES * sizeof(uint32_t));
memset(ipv4_icmp_flt_rule_hdl, 0, NUM_IPV4_ICMP_FLT_RULE * sizeof(uint32_t));
memset(private_fl_rule_hdl, 0, (IPA_MAX_PRIVATE_SUBNET_ENTRIES + IPA_MAX_MTU_ENTRIES) * sizeof(uint32_t));
memset(ipv6_prefix_flt_rule_hdl, 0, (NUM_IPV6_PREFIX_FLT_RULE + NUM_IPV6_PREFIX_MTU_RULE) * sizeof(uint32_t));
memset(ipv6_icmp_flt_rule_hdl, 0, NUM_IPV6_ICMP_FLT_RULE * sizeof(uint32_t));
memset(ipv6_prefix, 0, sizeof(ipv6_prefix));
memset(filter_cfg_rule_hdl, 0, sizeof(filter_cfg_rule_hdl));
/* ODU routing table initilization */
if(ipa_if_cate == ODU_IF)
{
/* only do one time ioctl to odu-driver to infrom in router or bridge mode*/
if (IPACM_Lan::odu_up != true)
{
m_fd_odu = open(IPACM_Iface::ipacmcfg->DEVICE_NAME_ODU, O_RDWR);
if (0 == m_fd_odu)
{
IPACMERR("Failed opening %s.\n", IPACM_Iface::ipacmcfg->DEVICE_NAME_ODU);
return ;
}
if(IPACM_Iface::ipacmcfg->ipacm_odu_router_mode == true)
{
ret = ioctl(m_fd_odu, ODU_BRIDGE_IOC_SET_MODE, ODU_BRIDGE_MODE_ROUTER);
IPACM_Iface::ipacmcfg->ipacm_odu_enable = true;
}
else
{
ret = ioctl(m_fd_odu, ODU_BRIDGE_IOC_SET_MODE, ODU_BRIDGE_MODE_BRIDGE);
IPACM_Iface::ipacmcfg->ipacm_odu_enable = true;
}
if (ret)
{
IPACMERR("Failed tell odu-driver the mode\n");
}
IPACMDBG("Tell odu-driver in router-mode(%d)\n", IPACM_Iface::ipacmcfg->ipacm_odu_router_mode);
IPACMDBG_H("odu is up: odu-driver in router-mode(%d) \n", IPACM_Iface::ipacmcfg->ipacm_odu_router_mode);
close(m_fd_odu);
IPACM_Lan::odu_up = true;
}
}
if(iface_query != NULL && tx_prop != NULL)
{
for(i=0; i<iface_query->num_tx_props; i++)
each_client_rt_rule_count[tx_prop->tx[i].ip]++;
}
IPACMDBG_H("Need to add %d IPv4 and %d IPv6 routing rules for eth bridge for each client.\n", each_client_rt_rule_count[IPA_IP_v4], each_client_rt_rule_count[IPA_IP_v6]);
#ifdef FEATURE_IPA_ANDROID
/* set the IPA-client pipe enum */
if(ipa_if_cate == LAN_IF)
{
#ifdef FEATURE_IPACM_HAL
handle_tethering_client(false, IPACM_CLIENT_MAX);
#else
handle_tethering_client(false, IPACM_CLIENT_USB);
#endif // FEATURE_IPACM_HAL end
}
#endif // FEATURE_IPA_ANDROID end
memset(is_downstream_set, 0, sizeof(is_downstream_set));
memset(is_upstream_set, 0, sizeof(is_upstream_set));
memset(&prefix, 0, sizeof(prefix));
#ifdef FEATURE_IPACM_HAL
/* check if Upstream was set before as WIFI with RNDIS case */
if(ipa_if_cate == LAN_IF && IPACM_Wan::backhaul_mode == WLAN_WAN) /* LTE */
{
IPACMDBG_H(" Skip the Upstream flag set on LAN instance (%d) with WIFI backhaul (%d)\n", ipa_if_cate, IPACM_Wan::backhaul_mode ); /* RNDIS+WIFI not support on msm*/
return;
}
/* check if Upstream was set before */
if (IPACM_Wan::isWanUP(ipa_if_num))
{
IPACMDBG_H("Upstream was set previously for ipv4, change is_upstream_set flag\n");
is_upstream_set[IPA_IP_v4] = true;
}
if (IPACM_Wan::isWanUP_V6(ipa_if_num))
{
IPACMDBG_H("Upstream was set previously for ipv6, change is_upstream_set flag\n");
is_upstream_set[IPA_IP_v6] = true;
}
#endif // FEATURE_IPACM_HAL end
return;
}
IPACM_Lan::~IPACM_Lan()
{
IPACM_EvtDispatcher::deregistr(this);
IPACM_IfaceManager::deregistr(this);
return;
}
/* LAN-iface's callback function */
void IPACM_Lan::event_callback(ipa_cm_event_id event, void *param)
{
if(is_active == false && event != IPA_LAN_DELETE_SELF)
{
IPACMDBG_H("The interface is no longer active, return.\n");
return;
}
int ipa_interface_index;
uint32_t i;
ipacm_ext_prop* ext_prop;
ipacm_event_iface_up_tehter* data_wan_tether;
switch (event)
{
case IPA_LINK_DOWN_EVENT:
{
ipacm_event_data_fid *data = (ipacm_event_data_fid *)param;
ipa_interface_index = iface_ipa_index_query(data->if_index);
if (ipa_interface_index == ipa_if_num)
{
IPACMDBG_H("Received IPA_LINK_DOWN_EVENT\n");
handle_down_evt();
IPACM_Iface::ipacmcfg->DelNatIfaces(dev_name); // delete NAT-iface
return;
}
}
break;
case IPA_CFG_CHANGE_EVENT:
{
if ( IPACM_Iface::ipacmcfg->iface_table[ipa_if_num].if_cat != ipa_if_cate)
{
IPACMDBG_H("Received IPA_CFG_CHANGE_EVENT and category changed\n");
/* delete previous instance */
handle_down_evt();
IPACM_Iface::ipacmcfg->DelNatIfaces(dev_name); // delete NAT-iface
is_mode_switch = true; // need post internal usb-link up event
return;
}
/* Add Natting iface to IPACM_Config if there is Rx/Tx property */
if (rx_prop != NULL || tx_prop != NULL)
{
IPACMDBG_H(" Has rx/tx properties registered for iface %s, add for NATTING \n", dev_name);
IPACM_Iface::ipacmcfg->AddNatIfaces(dev_name, IPA_IP_MAX);
}
}
break;
case IPA_FILTER_CFG_CHANGE_EVENT:
{
IPACMDBG_H("Received IPA_FILTER_CFG_CHANGE_EVENT");
/* IPA_IP_MAX means both ipv4 and ipv6 */
if ((ip_type == IPA_IP_v4 || ip_type == IPA_IP_MAX)
&& IPACM_Wan::isWanUP(ipa_if_num))
{
handle_filter_cfg_update(IPA_IP_v4);
}
}
break;
case IPA_PRIVATE_SUBNET_CHANGE_EVENT:
{
ipacm_event_data_fid *data = (ipacm_event_data_fid *)param;
/* internel event: data->if_index is ipa_if_index */
if (data->if_index == ipa_if_num)
{
IPACMDBG_H("Received IPA_PRIVATE_SUBNET_CHANGE_EVENT from itself posting, ignore\n");
return;
}
else
{
IPACMDBG_H("Received IPA_PRIVATE_SUBNET_CHANGE_EVENT from other LAN iface \n");
#ifdef FEATURE_IPA_ANDROID
handle_private_subnet_android(IPA_IP_v4);
#endif
IPACMDBG_H(" delete old private subnet rules, use new sets \n");
return;
}
}
break;
case IPA_LAN_DELETE_SELF:
{
ipacm_event_data_fid *data = (ipacm_event_data_fid *)param;
if(data->if_index == ipa_if_num)
{
IPACMDBG_H("Received IPA_LAN_DELETE_SELF event.\n");
IPACMDBG_H("ipa_LAN (%s):ipa_index (%d) instance close \n", IPACM_Iface::ipacmcfg->iface_table[ipa_if_num].iface_name, ipa_if_num);
/* posting link-up event for cradle use-case */
if(is_mode_switch)
{
IPACMDBG_H("Posting IPA_USB_LINK_UP_EVENT event for (%s)\n", dev_name);
ipacm_cmd_q_data evt_data;
memset(&evt_data, 0, sizeof(evt_data));
ipacm_event_data_fid *data_fid = NULL;
data_fid = (ipacm_event_data_fid *)malloc(sizeof(ipacm_event_data_fid));
if(data_fid == NULL)
{
IPACMERR("unable to allocate memory for IPA_USB_LINK_UP_EVENT data_fid\n");
return;
}
if(IPACM_Iface::ipa_get_if_index(dev_name, &(data_fid->if_index)))
{
IPACMERR("Error while getting interface index for %s device", dev_name);
}
evt_data.event = IPA_USB_LINK_UP_EVENT;
evt_data.evt_data = data_fid;
//IPACMDBG_H("Posting event:%d\n", evt_data.event);
IPACM_EvtDispatcher::PostEvt(&evt_data);
}
#ifndef FEATURE_IPA_ANDROID
if(rx_prop != NULL)
{
if(IPACM_Iface::ipacmcfg->getFltRuleCount(rx_prop->rx[0].src_pipe, IPA_IP_v4) != 0)
{
IPACMDBG_DMESG("### WARNING ### num ipv4 flt rules on client %d is not expected: %d expected value: 0",
rx_prop->rx[0].src_pipe, IPACM_Iface::ipacmcfg->getFltRuleCount(rx_prop->rx[0].src_pipe, IPA_IP_v4));
}
if(IPACM_Iface::ipacmcfg->getFltRuleCount(rx_prop->rx[0].src_pipe, IPA_IP_v6) != 0)
{
IPACMDBG_DMESG("### WARNING ### num ipv6 flt rules on client %d is not expected: %d expected value: 0",
rx_prop->rx[0].src_pipe, IPACM_Iface::ipacmcfg->getFltRuleCount(rx_prop->rx[0].src_pipe, IPA_IP_v6));
}
}
#endif
if (IPACM_Iface::ipacmcfg->isEthBridgingSupported())
{
if(rx_prop != NULL)
{
free(rx_prop);
}
if(tx_prop != NULL)
{
free(tx_prop);
}
if(iface_query != NULL)
{
free(iface_query);
}
}
delete this;
}
break;
}
case IPA_ADDR_ADD_EVENT:
{
ipacm_event_data_addr *data = (ipacm_event_data_addr *)param;
ipa_interface_index = iface_ipa_index_query(data->if_index);
if ( (data->iptype == IPA_IP_v4 && data->ipv4_addr == 0) ||
(data->iptype == IPA_IP_v6 &&
data->ipv6_addr[0] == 0 && data->ipv6_addr[1] == 0 &&
data->ipv6_addr[2] == 0 && data->ipv6_addr[3] == 0) )
{
IPACMDBG_H("Invalid address, ignore IPA_ADDR_ADD_EVENT event\n");
return;
}
#ifdef FEATURE_L2TP
if(data->iptype == IPA_IP_v6 && is_vlan_event(data->iface_name) && is_unique_local_ipv6_addr(data->ipv6_addr))
{
IPACMDBG_H("Got IPv6 new addr event for a vlan iface %s.\n", data->iface_name);
eth_bridge_post_event(IPA_HANDLE_VLAN_IFACE_INFO, data->iptype, NULL,
data->ipv6_addr, data->iface_name, IPA_CLIENT_MAX);
}
#endif
if (ipa_interface_index == ipa_if_num)
{
IPACMDBG_H("Received IPA_ADDR_ADD_EVENT\n");
/* only call ioctl for ODU iface with bridge mode */
if(IPACM_Iface::ipacmcfg->ipacm_odu_enable == true && IPACM_Iface::ipacmcfg->ipacm_odu_router_mode == false
&& ipa_if_cate == ODU_IF)
{
if((data->iptype == IPA_IP_v6) && (num_dft_rt_v6 == 0))
{
handle_addr_evt_odu_bridge(data);
}
#ifdef FEATURE_IPA_ANDROID
add_dummy_private_subnet_flt_rule(data->iptype);
handle_private_subnet_android(data->iptype);
add_dummy_filter_cfg_rules(data->iptype);
handle_filter_cfg_update(data->iptype);
#else
handle_private_subnet(data->iptype);
#endif // FEATURE_IPA_ANDROID end
}
else
{
/* check v4 not setup before, v6 can have 2 iface ip */
if( ((data->iptype != ip_type) && (ip_type != IPA_IP_MAX))
|| ((data->iptype==IPA_IP_v6) && (num_dft_rt_v6!=MAX_DEFAULT_v6_ROUTE_RULES)))
{
IPACMDBG_H("Got IPA_ADDR_ADD_EVENT ip-family:%d, v6 num %d: \n",data->iptype,num_dft_rt_v6);
if(handle_addr_evt(data) == IPACM_FAILURE)
{
return;
}
#ifdef FEATURE_IPA_ANDROID
add_dummy_private_subnet_flt_rule(data->iptype);
handle_private_subnet_android(data->iptype);
add_dummy_filter_cfg_rules(data->iptype);
handle_filter_cfg_update(data->iptype);
#else
handle_private_subnet(data->iptype);
#endif // FEATURE_IPA_ANDROID end
#ifndef FEATURE_IPACM_HAL
if (IPACM_Wan::isWanUP(ipa_if_num))
{
if(data->iptype == IPA_IP_v4 || data->iptype == IPA_IP_MAX)
{
if(IPACM_Wan::backhaul_mode == Q6_WAN)
{
ext_prop = IPACM_Iface::ipacmcfg->GetExtProp(IPA_IP_v4);
handle_wan_up_ex(ext_prop, IPA_IP_v4,
IPACM_Wan::getXlat_Mux_Id());
}
else
{
handle_wan_up(IPA_IP_v4);
}
}
IPACMDBG_H("Finished checking wan_up\n");
} else {
IPACMDBG_H("Wan_V4 haven't up yet\n");
}
if(IPACM_Wan::isWanUP_V6(ipa_if_num))
{
if((data->iptype == IPA_IP_v6 || data->iptype == IPA_IP_MAX) && num_dft_rt_v6 == 1)
{
memcpy(ipv6_prefix, IPACM_Wan::backhaul_ipv6_prefix, sizeof(ipv6_prefix));
install_ipv6_prefix_flt_rule(IPACM_Wan::backhaul_ipv6_prefix);
if(IPACM_Wan::backhaul_mode == Q6_WAN)
{
ext_prop = IPACM_Iface::ipacmcfg->GetExtProp(IPA_IP_v6);
handle_wan_up_ex(ext_prop, IPA_IP_v6, 0);
}
else
{
handle_wan_up(IPA_IP_v6);
}
}
IPACMDBG_H("Finished checking wan_up_v6\n");
} else {
IPACMDBG_H("Wan_V6 haven't up yet\n");
}
#else
/* check if Upstream was set before */
if (IPACM_Wan::isWanUP(ipa_if_num))
{
IPACMDBG_H("Upstream was set previously for ipv4, change is_upstream_set flag\n");
is_upstream_set[IPA_IP_v4] = true;
}
if (IPACM_Wan::isWanUP_V6(ipa_if_num))
{
IPACMDBG_H("Upstream was set previously for ipv6, change is_upstream_set flag\n");
is_upstream_set[IPA_IP_v6] = true;
}
#endif //FEATURE_IPACM_HAL end
/* Post event to NAT */
if (data->iptype == IPA_IP_v4)
{
ipacm_cmd_q_data evt_data;
ipacm_event_iface_up *info;
info = (ipacm_event_iface_up *)
malloc(sizeof(ipacm_event_iface_up));
if (info == NULL)
{
IPACMERR("Unable to allocate memory\n");
return;
}
memcpy(info->ifname, dev_name, IF_NAME_LEN);
info->ipv4_addr = data->ipv4_addr;
info->addr_mask = IPACM_Iface::ipacmcfg->private_subnet_table[0].subnet_mask;
evt_data.event = IPA_HANDLE_LAN_UP;
evt_data.evt_data = (void *)info;
/* Insert IPA_HANDLE_LAN_UP to command queue */
IPACMDBG_H("posting IPA_HANDLE_LAN_UP for IPv4 with below information\n");
IPACMDBG_H("IPv4 address:0x%x, IPv4 address mask:0x%x\n",
info->ipv4_addr, info->addr_mask);
IPACM_EvtDispatcher::PostEvt(&evt_data);
}
IPACMDBG_H("Finish handling IPA_ADDR_ADD_EVENT for ip-family(%d)\n", data->iptype);
}
IPACMDBG_H("Finish handling IPA_ADDR_ADD_EVENT for ip-family(%d)\n", data->iptype);
/* checking if SW-RT_enable */
if (IPACM_Iface::ipacmcfg->ipa_sw_rt_enable == true)
{
/* handle software routing enable event*/
IPACMDBG_H("IPA_SW_ROUTING_ENABLE for iface: %s \n",IPACM_Iface::ipacmcfg->iface_table[ipa_if_num].iface_name);
handle_software_routing_enable(false);
}
}
}
}
break;
#ifdef FEATURE_IPA_ANDROID
case IPA_HANDLE_WAN_UP_TETHER:
IPACMDBG_H("Received IPA_HANDLE_WAN_UP_TETHER event\n");
data_wan_tether = (ipacm_event_iface_up_tehter*)param;
if(data_wan_tether == NULL)
{
IPACMERR("No event data is found.\n");
return;
}
IPACMDBG_H("Backhaul is sta mode?%d, if_index_tether:%d tether_if_name:%s, xlat %d\n", data_wan_tether->backhaul_type,
data_wan_tether->if_index_tether,
IPACM_Iface::ipacmcfg->iface_table[data_wan_tether->if_index_tether].iface_name,
data_wan_tether->xlat_mux_id);
#ifndef FEATURE_IPACM_HAL
if (data_wan_tether->if_index_tether != ipa_if_num)
{
IPACMERR("IPA_HANDLE_WAN_UP_TETHER tether_if(%d), not valid (%d) ignore\n", data_wan_tether->if_index_tether, ipa_if_num);
return;
}
#else /* not offload rndis on WIFI mode on MSM targets */
if (data_wan_tether->backhaul_type == WLAN_WAN)
{
IPACMERR("Not support RNDIS offload on WIFI mode, dun install UL filter rules for WIFI mode\n");
/* clean rndis header, routing rules */
IPACMDBG_H("left %d eth clients need to be deleted \n ", num_eth_client);
for (i = 0; i < num_eth_client; i++)
{
/* First reset nat rules and then route rules */
if(get_client_memptr(eth_client, i)->ipv4_set == true)
{
IPACMDBG_H("Clean Nat Rules for ipv4:0x%x\n", get_client_memptr(eth_client, i)->v4_addr);
CtList->HandleNeighIpAddrDelEvt(get_client_memptr(eth_client, i)->v4_addr);
}
if (delete_eth_rtrules(i, IPA_IP_v4))
IPACMERR("unbale to delete usb-client v4 route rules for index %d\n", i);
if (delete_eth_rtrules(i, IPA_IP_v6))
IPACMERR("unbale to delete ecm-client v6 route rules for index %d\n", i);
IPACMDBG_H("Delete %d client header\n", num_eth_client);
if(get_client_memptr(eth_client, i)->ipv4_header_set == true)
{
if (m_header.DeleteHeaderHdl(get_client_memptr(eth_client, i)->hdr_hdl_v4)
== false)
IPACMERR("unbale to delete usb-client v4 header for index %d\n", i);
}
if(get_client_memptr(eth_client, i)->ipv6_header_set == true)
{
if (m_header.DeleteHeaderHdl(get_client_memptr(eth_client, i)->hdr_hdl_v6)
== false)
IPACMERR("unbale to delete usb-client v6 header for index %d\n", i);
}
} /* end of for loop */
return;
}
#endif // FEATURE_IPACM_HAL end
if (ip_type == IPA_IP_v4 || ip_type == IPA_IP_MAX)
{
#ifdef FEATURE_IPACM_HAL
if (is_upstream_set[IPA_IP_v4] == false)
{
IPACMDBG_H("Add upstream for IPv4.\n");
is_upstream_set[IPA_IP_v4] = true;
if (is_downstream_set[IPA_IP_v4] == true)
{
IPACMDBG_H("Downstream was set before, adding UL rules.\n");
if (data_wan_tether->backhaul_type == Q6_WAN)
{
ext_prop = IPACM_Iface::ipacmcfg->GetExtProp(IPA_IP_v4);
handle_wan_up_ex(ext_prop, IPA_IP_v4,
data_wan_tether->xlat_mux_id);
} else {
handle_wan_up(IPA_IP_v4);
}
}
}
#else
if (data_wan_tether->backhaul_type == Q6_WAN)
{
ext_prop = IPACM_Iface::ipacmcfg->GetExtProp(IPA_IP_v4);
handle_wan_up_ex(ext_prop, IPA_IP_v4, data_wan_tether->xlat_mux_id);
} else {
handle_wan_up(IPA_IP_v4);
}
#endif // FEATURE_IPACM_HAL end
}
break;
case IPA_HANDLE_WAN_UP_V6_TETHER:
IPACMDBG_H("Received IPA_HANDLE_WAN_UP_V6_TETHER event\n");
data_wan_tether = (ipacm_event_iface_up_tehter*)param;
if (data_wan_tether == NULL)
{
IPACMERR("No event data is found.\n");
return;
}
IPACMDBG_H("Backhaul is sta mode?%d, if_index_tether:%d tether_if_name:%s\n", data_wan_tether->backhaul_type,
data_wan_tether->if_index_tether,
IPACM_Iface::ipacmcfg->iface_table[data_wan_tether->if_index_tether].iface_name);
#ifndef FEATURE_IPACM_HAL
if (data_wan_tether->if_index_tether != ipa_if_num)
{
IPACMERR("IPA_HANDLE_WAN_UP_V6_TETHER tether_if(%d), not valid (%d) ignore\n", data_wan_tether->if_index_tether, ipa_if_num);
return;
}
#endif // FEATURE_IPACM_HAL end
if (ip_type == IPA_IP_v6 || ip_type == IPA_IP_MAX)
{
#ifdef FEATURE_IPACM_HAL
if (is_upstream_set[IPA_IP_v6] == false)
{
IPACMDBG_H("Add upstream for IPv6.\n");
is_upstream_set[IPA_IP_v6] = true;
if (is_downstream_set[IPA_IP_v6] == true)
{
IPACMDBG_H("Downstream was set before, adding UL rules.\n");
memcpy(ipv6_prefix, data_wan_tether->ipv6_prefix, sizeof(ipv6_prefix));
install_ipv6_prefix_flt_rule(data_wan_tether->ipv6_prefix);
if (data_wan_tether->backhaul_type == Q6_WAN)
{
ext_prop = IPACM_Iface::ipacmcfg->GetExtProp(IPA_IP_v6);
handle_wan_up_ex(ext_prop, IPA_IP_v6, 0);
}
else
{
handle_wan_up(IPA_IP_v6);
}
}
}
#else
if (data_wan_tether->backhaul_type == Q6_WAN)
{
ext_prop = IPACM_Iface::ipacmcfg->GetExtProp(IPA_IP_v6);
handle_wan_up_ex(ext_prop, IPA_IP_v6, 0);
} else {
handle_wan_up(IPA_IP_v6);
}
#endif // FEATURE_IPACM_HAL end
}
break;
case IPA_HANDLE_WAN_DOWN_TETHER:
IPACMDBG_H("Received IPA_HANDLE_WAN_DOWN_TETHER event\n");
data_wan_tether = (ipacm_event_iface_up_tehter*)param;
if (data_wan_tether == NULL)
{
IPACMERR("No event data is found.\n");
return;
}
IPACMDBG_H("Backhaul is sta mode?%d, if_index_tether:%d tether_if_name:%s\n", data_wan_tether->backhaul_type,
data_wan_tether->if_index_tether,
IPACM_Iface::ipacmcfg->iface_table[data_wan_tether->if_index_tether].iface_name);
#ifndef FEATURE_IPACM_HAL
if (data_wan_tether->if_index_tether != ipa_if_num)
{
IPACMERR("IPA_HANDLE_WAN_DOWN_TETHER tether_if(%d), not valid (%d) ignore\n", data_wan_tether->if_index_tether, ipa_if_num);
return;
}
#endif // FEATURE_IPACM_HAL end
if (ip_type == IPA_IP_v4 || ip_type == IPA_IP_MAX)
{
#ifdef FEATURE_IPACM_HAL
if(is_upstream_set[IPA_IP_v4] == true)
{
IPACMDBG_H("Del upstream for IPv4.\n");
is_upstream_set[IPA_IP_v4] = false;
if(is_downstream_set[IPA_IP_v4] == true)
{
IPACMDBG_H("Downstream was set before, deleting UL rules.\n");
handle_wan_down(data_wan_tether->backhaul_type);
}
}
#else
handle_wan_down(data_wan_tether->backhaul_type);
#endif // FEATURE_IPACM_HAL end
}
break;
case IPA_HANDLE_WAN_DOWN_V6_TETHER:
IPACMDBG_H("Received IPA_HANDLE_WAN_DOWN_V6_TETHER event\n");
data_wan_tether = (ipacm_event_iface_up_tehter*)param;
if(data_wan_tether == NULL)
{
IPACMERR("No event data is found.\n");
return;
}
IPACMDBG_H("Backhaul is sta mode?%d, if_index_tether:%d tether_if_name:%s\n", data_wan_tether->backhaul_type,
data_wan_tether->if_index_tether,
IPACM_Iface::ipacmcfg->iface_table[data_wan_tether->if_index_tether].iface_name);
#ifndef FEATURE_IPACM_HAL
if (data_wan_tether->if_index_tether != ipa_if_num)
{
IPACMERR("IPA_HANDLE_WAN_DOWN_V6_TETHER tether_if(%d), not valid (%d) ignore\n", data_wan_tether->if_index_tether, ipa_if_num);
return;
}
#endif // FEATURE_IPACM_HAL end
if (ip_type == IPA_IP_v6 || ip_type == IPA_IP_MAX)
{
#ifdef FEATURE_IPACM_HAL
if (is_upstream_set[IPA_IP_v6] == true)
{
IPACMDBG_H("Del upstream for IPv6.\n");
is_upstream_set[IPA_IP_v6] = false;
if(is_downstream_set[IPA_IP_v6] == true)
{
IPACMDBG_H("Downstream was set before, deleting UL rules.\n");
/* reset usb-client ipv6 rt-rules */
handle_lan_client_reset_rt(IPA_IP_v6);
handle_wan_down_v6(data_wan_tether->backhaul_type);
}
}
#else
/* reset usb-client ipv6 rt-rules */
handle_lan_client_reset_rt(IPA_IP_v6);
handle_wan_down_v6(data_wan_tether->backhaul_type);
#endif // FEATURE_IPACM_HAL end
}
break;
case IPA_DOWNSTREAM_ADD:
{
ipacm_event_ipahal_stream *data = (ipacm_event_ipahal_stream *)param;
ipa_interface_index = iface_ipa_index_query(data->if_index);
if (ipa_interface_index == ipa_if_num)
{
IPACMDBG_H("Received IPA_DOWNSTREAM_ADD event.\n");
/* Delete existing downstream if it exists */
if (is_downstream_set[data->prefix.iptype] == true)
{
if (ipv6_prefix[0] != data->prefix.v6Addr[0] ||
ipv6_prefix[1] != data->prefix.v6Addr[1])
{
IPACMDBG_H("Del existing downstream for IP iptype %d.\n", data->prefix.iptype);
is_downstream_set[data->prefix.iptype] = false;
store_downstream_state(false, data->prefix.iptype);
if (is_upstream_set[data->prefix.iptype] == true)
{
IPACMDBG_H("Upstream was set before, deleting UL rules.\n");
if (data->prefix.iptype == IPA_IP_v4)
{
/* LTE STA */
handle_wan_down(IPACM_Wan::backhaul_mode);
} else {
handle_lan_client_reset_rt(IPA_IP_v6);
/* LTE STA */
handle_wan_down_v6(IPACM_Wan::backhaul_mode);
}
}
}
}
if (data->prefix.iptype < IPA_IP_MAX && is_downstream_set[data->prefix.iptype] == false)
{
IPACMDBG_H("Add downstream for IP iptype %d\n", data->prefix.iptype);
is_downstream_set[data->prefix.iptype] = true;
store_downstream_state(true, data->prefix.iptype);
memcpy(&prefix[data->prefix.iptype], &data->prefix,
sizeof(prefix[data->prefix.iptype]));
if (is_upstream_set[data->prefix.iptype] == true)
{
IPACMDBG_H("Upstream was set before, adding UL rules.\n");
if (ip_type == IPA_IP_MAX || ip_type == data->prefix.iptype)
{
if (data->prefix.iptype == IPA_IP_v6) /* ipv6 only */
{
/* Only offload clients has same prefix as Andorid gave */
ipv6_prefix[0] = data->prefix.v6Addr[0];
ipv6_prefix[1] = data->prefix.v6Addr[1];
IPACMDBG_H("ipv6_prefix0x%x:%x\n", ipv6_prefix[0], ipv6_prefix[1]);
install_ipv6_prefix_flt_rule(ipv6_prefix);
}
if (IPACM_Wan::backhaul_mode == Q6_WAN) /* LTE */
{
ext_prop = IPACM_Iface::ipacmcfg->GetExtProp(data->prefix.iptype);
if (data->prefix.iptype == IPA_IP_v4)
{
IPACMDBG_H("check getXlat_Mux_Id:%d\n", IPACM_Wan::getXlat_Mux_Id());
handle_wan_up_ex(ext_prop, data->prefix.iptype,
IPACM_Wan::getXlat_Mux_Id());
}
else {
handle_wan_up_ex(ext_prop, data->prefix.iptype, 0);
}
} else {
handle_wan_up(data->prefix.iptype); /* STA */
}
}
}
} else {
IPACMDBG_H("downstream for IP iptype %d already set \n", data->prefix.iptype);
}
}
break;
}
case IPA_DOWNSTREAM_DEL:
{
ipacm_event_ipahal_stream *data = (ipacm_event_ipahal_stream *)param;
ipa_interface_index = iface_ipa_index_query(data->if_index);
if (ipa_interface_index == ipa_if_num)
{
IPACMDBG_H("Received IPA_DOWNSTREAM_DEL event.\n");
if (is_downstream_set[data->prefix.iptype] == true)
{
/* Handle downstream del event only when the prefix matches */
if (ipv6_prefix[0] == data->prefix.v6Addr[0] &&
ipv6_prefix[1] == data->prefix.v6Addr[1])
{
IPACMDBG_H("Del downstream for IP iptype %d.\n", data->prefix.iptype);
is_downstream_set[data->prefix.iptype] = false;
store_downstream_state(false, data->prefix.iptype);
if (is_upstream_set[data->prefix.iptype] == true)
{
IPACMDBG_H("Upstream was set before, deleting UL rules.\n");
if (data->prefix.iptype == IPA_IP_v4)
{
handle_wan_down(IPACM_Wan::backhaul_mode); /* LTE STA */
} else {
handle_lan_client_reset_rt(IPA_IP_v6);
handle_wan_down_v6(IPACM_Wan::backhaul_mode); /* LTE STA */
}
}
}
}
}
break;
}
#else // above Andorid
case IPA_HANDLE_WAN_UP:
IPACMDBG_H("Received IPA_HANDLE_WAN_UP event\n");
ipacm_event_iface_up* data_wan = (ipacm_event_iface_up*)param;
if (data_wan == NULL)
{
IPACMERR("No event data is found.\n");
return;
}
IPACMDBG_H("Backhaul is sta mode?%d\n", data_wan->backhaul_type);
if (ip_type == IPA_IP_v4 || ip_type == IPA_IP_MAX)
{
if (data_wan->backhaul_type == Q6_WAN)
{
ext_prop = IPACM_Iface::ipacmcfg->GetExtProp(IPA_IP_v4);
handle_wan_up_ex(ext_prop, IPA_IP_v4, data_wan->xlat_mux_id);
}
else
{
handle_wan_up(IPA_IP_v4);
}
}
break;
case IPA_HANDLE_WAN_UP_V6:
IPACMDBG_H("Received IPA_HANDLE_WAN_UP_V6 event\n");
data_wan = (ipacm_event_iface_up*)param;
if (data_wan == NULL)
{
IPACMERR("No event data is found.\n");
return;
}
IPACMDBG_H("Backhaul is sta mode?%d\n", data_wan->backhaul_type);
if (ip_type == IPA_IP_v6 || ip_type == IPA_IP_MAX)
{
memcpy(ipv6_prefix, data_wan->ipv6_prefix, sizeof(ipv6_prefix));
install_ipv6_prefix_flt_rule(data_wan->ipv6_prefix);
/* MTU might have changed. Need to update ipv4 MTU rule if up */
if (IPACM_Wan::isWanUP(ipa_if_num))
handle_private_subnet_android(IPA_IP_v4);
if (data_wan->backhaul_type == Q6_WAN)
{
ext_prop = IPACM_Iface::ipacmcfg->GetExtProp(IPA_IP_v6);
handle_wan_up_ex(ext_prop, IPA_IP_v6, 0);
}
else
{
handle_wan_up(IPA_IP_v6);
}
}
break;
case IPA_HANDLE_WAN_DOWN:
IPACMDBG_H("Received IPA_HANDLE_WAN_DOWN event\n");
data_wan = (ipacm_event_iface_up*)param;
if (data_wan == NULL)
{
IPACMERR("No event data is found.\n");
return;
}
IPACMDBG_H("Backhaul is sta mode?%d\n", data_wan->backhaul_type);
if (ip_type == IPA_IP_v4 || ip_type == IPA_IP_MAX)
{
handle_wan_down(data_wan->backhaul_type);
}
break;
case IPA_HANDLE_WAN_DOWN_V6:
IPACMDBG_H("Received IPA_HANDLE_WAN_DOWN_V6 event\n");
data_wan = (ipacm_event_iface_up*)param;
if (data_wan == NULL)
{
IPACMERR("No event data is found.\n");
return;
}
/* clean up v6 RT rules*/
IPACMDBG_H("Received IPA_WAN_V6_DOWN in LAN-instance and need clean up client IPv6 address \n");
/* reset usb-client ipv6 rt-rules */
handle_lan_client_reset_rt(IPA_IP_v6);
IPACMDBG_H("Backhaul is sta mode?%d\n", data_wan->backhaul_type);
if (ip_type == IPA_IP_v6 || ip_type == IPA_IP_MAX)
{
handle_wan_down_v6(data_wan->backhaul_type);
}
break;
#endif // FEATURE_IPA_ANDROID end
case IPA_NEIGH_CLIENT_IP_ADDR_ADD_EVENT:
{
ipacm_event_data_all *data = (ipacm_event_data_all *)param;
ipa_interface_index = iface_ipa_index_query(data->if_index);
IPACMDBG_H("Recieved IPA_NEIGH_CLIENT_IP_ADDR_ADD_EVENT event \n");
IPACMDBG_H("check iface %s category: %d\n", dev_name, ipa_if_cate);
/* if RNDIS under WIFI mode in MSM, dun add RT rule*/
#ifdef FEATURE_IPACM_HAL
if(IPACM_Wan::backhaul_mode == WLAN_WAN) /* WIFI */
{
IPACMDBG_H(" dun construct header and RT-rules for RNDIS-PC in WIFI mode on MSM targets (STA %d) \n", IPACM_Wan::backhaul_mode);
return;
}
#endif
if (ipa_interface_index == ipa_if_num && ipa_if_cate == ODU_IF)
{
IPACMDBG_H("ODU iface got v4-ip \n");
/* first construc ODU full header */
if ((ipv4_header_set == false) && (ipv6_header_set == false))
{
/* construct ODU RT tbl */
handle_odu_hdr_init(data->mac_addr);
if (IPACM_Iface::ipacmcfg->ipacm_odu_embms_enable == true)
{
handle_odu_route_add();
IPACMDBG_H("construct ODU header and route rules, embms_flag (%d) \n", IPACM_Iface::ipacmcfg->ipacm_odu_embms_enable);
}
else
{
IPACMDBG_H("construct ODU header only, embms_flag (%d) \n", IPACM_Iface::ipacmcfg->ipacm_odu_embms_enable);
}
}
/* if ODU in bridge mode, directly return */
if(IPACM_Iface::ipacmcfg->ipacm_odu_router_mode == false)
{
IPACMDBG_H("ODU is in bridge mode, no action \n");
return;
}
}
if (ipa_interface_index == ipa_if_num
#ifdef FEATURE_L2TP
|| is_vlan_event(data->iface_name)
|| (is_l2tp_event(data->iface_name) && ipa_if_cate == ODU_IF)
#endif
)
{
IPACMDBG_H("ETH iface got client \n");
if(ipa_interface_index == ipa_if_num)
{
/* first construc ETH full header */
handle_eth_hdr_init(data->mac_addr);
IPACMDBG_H("construct ETH header and route rules \n");
/* Associate with IP and construct RT-rule */
if (handle_eth_client_ipaddr(data) == IPACM_FAILURE)
{
return;
}
handle_eth_client_route_rule(data->mac_addr, data->iptype);
if (data->iptype == IPA_IP_v4)
{
/* Add NAT rules after ipv4 RT rules are set */
CtList->HandleNeighIpAddrAddEvt(data);
}
eth_bridge_post_event(IPA_ETH_BRIDGE_CLIENT_ADD, IPA_IP_MAX, data->mac_addr, NULL, data->iface_name,
IPA_CLIENT_MAX);
}
#ifdef FEATURE_L2TP
else if(is_l2tp_event(data->iface_name) && ipa_if_cate == ODU_IF)
{
if(IPACM_Iface::ipacmcfg->GetIPAVer() >= IPA_HW_None && IPACM_Iface::ipacmcfg->GetIPAVer() < IPA_HW_v4_0)
{
/* Add corresponding ipa_rm_resource_name of TX-endpoint up before IPV6 RT-rule set */
IPACMDBG_H("dev %s add producer dependency\n", dev_name);
if (tx_prop != NULL)
{
IPACMDBG_H("depend Got pipe %d rm index : %d \n", tx_prop->tx[0].dst_pipe, IPACM_Iface::ipacmcfg->ipa_client_rm_map_tbl[tx_prop->tx[0].dst_pipe]);
IPACM_Iface::ipacmcfg->AddRmDepend(IPACM_Iface::ipacmcfg->ipa_client_rm_map_tbl[tx_prop->tx[0].dst_pipe],false);
}
}
eth_bridge_post_event(IPA_ETH_BRIDGE_CLIENT_ADD, IPA_IP_MAX, data->mac_addr, NULL, data->iface_name,
IPA_CLIENT_MAX);
}
else
{
if(data->iptype == IPA_IP_v6 && is_unique_local_ipv6_addr(data->ipv6_addr))
{
eth_bridge_post_event(IPA_HANDLE_VLAN_CLIENT_INFO, IPA_IP_MAX, data->mac_addr, data->ipv6_addr, data->iface_name, IPA_CLIENT_MAX);
}
}
#endif
return;
}
}
break;
case IPA_NEIGH_CLIENT_IP_ADDR_DEL_EVENT:
{
ipacm_event_data_all *data = (ipacm_event_data_all *)param;
ipa_interface_index = iface_ipa_index_query(data->if_index);
IPACMDBG_H("Received IPA_NEIGH_CLIENT_IP_ADDR_DEL_EVENT event. \n");
IPACMDBG_H("check iface %s category: %d\n", dev_name, ipa_if_cate);
/* if ODU in bridge mode, directly return */
if (ipa_if_cate == ODU_IF && IPACM_Iface::ipacmcfg->ipacm_odu_router_mode == false)
{
IPACMDBG_H("ODU is in bridge mode, no action \n");
return;
}
if (ipa_interface_index == ipa_if_num
#ifdef FEATURE_L2TP
|| (is_l2tp_event(data->iface_name) && ipa_if_cate == ODU_IF)
#endif
)
{
if(ipa_interface_index == ipa_if_num)
{
if (data->iptype == IPA_IP_v6)
{
handle_del_ipv6_addr(data);
return;
}
IPACMDBG_H("LAN iface delete client \n");
handle_eth_client_down_evt(data->mac_addr);
}
else
{
eth_bridge_post_event(IPA_ETH_BRIDGE_CLIENT_DEL, IPA_IP_MAX, data->mac_addr, NULL, data->iface_name,
IPA_CLIENT_MAX);
}
return;
}
}
break;
case IPA_SW_ROUTING_ENABLE:
IPACMDBG_H("Received IPA_SW_ROUTING_ENABLE\n");
/* handle software routing enable event*/
handle_software_routing_enable(false);
break;
case IPA_SW_ROUTING_DISABLE:
IPACMDBG_H("Received IPA_SW_ROUTING_DISABLE\n");
/* handle software routing disable event*/
handle_software_routing_disable(false);
break;
case IPA_CRADLE_WAN_MODE_SWITCH:
{
IPACMDBG_H("Received IPA_CRADLE_WAN_MODE_SWITCH event.\n");
ipacm_event_cradle_wan_mode* wan_mode = (ipacm_event_cradle_wan_mode*)param;
if(wan_mode == NULL)
{
IPACMERR("Event data is empty.\n");
return;
}
if(wan_mode->cradle_wan_mode == BRIDGE)
{
handle_cradle_wan_mode_switch(true);
}
else
{
handle_cradle_wan_mode_switch(false);
}
}
break;
case IPA_TETHERING_STATS_UPDATE_EVENT:
{
IPACMDBG_H("Received IPA_TETHERING_STATS_UPDATE_EVENT event.\n");
if (IPACM_Wan::isWanUP(ipa_if_num) || IPACM_Wan::isWanUP_V6(ipa_if_num))
{
if(IPACM_Wan::backhaul_mode == Q6_WAN) /* LTE */
{
ipa_get_data_stats_resp_msg_v01 *data = (ipa_get_data_stats_resp_msg_v01 *)param;
IPACMDBG("Received IPA_TETHERING_STATS_UPDATE_STATS ipa_stats_type: %d\n",data->ipa_stats_type);
IPACMDBG("Received %d UL, %d DL pipe stats\n",data->ul_src_pipe_stats_list_len,
data->dl_dst_pipe_stats_list_len);
if (data->ipa_stats_type != QMI_IPA_STATS_TYPE_PIPE_V01)
{
IPACMERR("not valid pipe stats enum(%d)\n", data->ipa_stats_type);
return;
}
handle_tethering_stats_event(data);
}
}
}
break;
#ifdef IPA_MTU_EVENT_MAX
case IPA_MTU_UPDATE:
{
IPACMDBG_H("Received IPA_MTU_UPDATE");
ipacm_event_mtu_info *evt_data = (ipacm_event_mtu_info *)param;
ipa_mtu_info *data = &(evt_data->mtu_info);
/* IPA_IP_MAX means both ipv4 and ipv6 */
if ((data->ip_type == IPA_IP_v4 || data->ip_type == IPA_IP_MAX) && IPACM_Wan::isWanUP(ipa_if_num))
{
handle_private_subnet_android(IPA_IP_v4);
}
/* IPA_IP_MAX means both ipv4 and ipv6 */
if ((data->ip_type == IPA_IP_v6 || data->ip_type == IPA_IP_MAX) && IPACM_Wan::isWanUP_V6(ipa_if_num))
{
/* check if the prefix + MTU rules are installed */
if (ipv6_prefix_flt_rule_hdl[0] && ipv6_prefix_flt_rule_hdl[1])
{
modify_ipv6_prefix_flt_rule(IPACM_Wan::backhaul_ipv6_prefix);
}
else
{
IPACMERR("Failed to update prefix MTU rules, no prefix rules set");
}
}
}
break;
#endif
default:
break;
}
return;
}
void IPACM_Lan::store_downstream_state(bool up, enum ipa_ip_type iptype)
{
/* Store the downstream state. */
FILE *fp = NULL;
bool state_update = false;
int free_index = -1;
/* Return if other iptype is active. */
if ((iptype == IPA_IP_v4 && is_downstream_set[IPA_IP_v6]) ||
(iptype == IPA_IP_v6 && is_downstream_set[IPA_IP_v4]))
return;
/* Update the downstream state info. */
for (int i=0; i < IPA_MAX_TETHER_IFACE_ENTRIES; i++)
{
if (IPACM_Lan::downstream_info[i].entry_in_use &&
strncmp(dev_name, IPACM_Lan::downstream_info[i].dev_name, IF_NAME_LEN) == 0)
{
/* Update state. */
IPACM_Lan::downstream_info[i].downstream_state = up;
state_update = true;
IPACMDBG_H("Updating info for tether iface :%s, State: %s\n", dev_name,
up ? "UP" : "DOWN");
break;
}
if (!IPACM_Lan::downstream_info[i].entry_in_use && free_index == -1)
free_index = i;
}
/* if state is up, check if entry is already present, if not add. */
if (up && !state_update && free_index != -1)
{
IPACM_Lan::downstream_info[free_index].entry_in_use = true;
IPACM_Lan::downstream_info[free_index].downstream_state = up;
strlcpy(IPACM_Lan::downstream_info[free_index].dev_name,
dev_name, IF_NAME_LEN);
IPACMDBG_H("Adding new tether iface :%s, State: UP\n", dev_name);
}
else if (!state_update && free_index == -1)
{
IPACMERR("Exceeded max tether ifaces: not storing info for %s\n", dev_name);
return;
}
fp = fopen(IPA_DOWNSTREAM_TETHER_STATE_FILE_NAME, "w");
if (fp == NULL)
{
IPACMERR("Failed to write downstream state to %s, error is %d - %s\n",
IPA_DOWNSTREAM_TETHER_STATE_FILE_NAME,
errno, strerror(errno));
}
else
{
for (int i=0; i < IPA_MAX_TETHER_IFACE_ENTRIES; i++)
{
if (IPACM_Lan::downstream_info[i].entry_in_use)
{
fprintf(fp, "DOWNSTREAM=%s,STATE=%s;", IPACM_Lan::downstream_info[i].dev_name,
IPACM_Lan::downstream_info[i].downstream_state ? "UP" : "DOWN");
}
}
fclose(fp);
}
}
int IPACM_Lan::handle_del_ipv6_addr(ipacm_event_data_all *data)
{
uint32_t tx_index;
uint32_t rt_hdl;
int num_v6 =0, clnt_indx;
clnt_indx = get_eth_client_index(data->mac_addr);
if (clnt_indx == IPACM_INVALID_INDEX)
{
IPACMERR("eth client not found/attached \n");
return IPACM_FAILURE;
}
if(data->iptype == IPA_IP_v6)
{
if ((data->ipv6_addr[0] != 0) || (data->ipv6_addr[1] != 0) ||
(data->ipv6_addr[2] != 0) || (data->ipv6_addr[3] != 0))
{
IPACMDBG_H("ipv6 address got: 0x%x:%x:%x:%x\n", data->ipv6_addr[0], data->ipv6_addr[1], data->ipv6_addr[2], data->ipv6_addr[3]);
for(num_v6=0;num_v6 < get_client_memptr(eth_client, clnt_indx)->ipv6_set;num_v6++)
{
if( data->ipv6_addr[0] == get_client_memptr(eth_client, clnt_indx)->v6_addr[num_v6][0] &&
data->ipv6_addr[1] == get_client_memptr(eth_client, clnt_indx)->v6_addr[num_v6][1] &&
data->ipv6_addr[2]== get_client_memptr(eth_client, clnt_indx)->v6_addr[num_v6][2] &&
data->ipv6_addr[3] == get_client_memptr(eth_client, clnt_indx)->v6_addr[num_v6][3])
{
IPACMDBG_H("ipv6 addr is found at position:%d for client:%d\n", num_v6, clnt_indx);
break;
}
}
}
else
{
IPACMDBG_H("Invalid ipv6 address\n");
return IPACM_FAILURE;
}
if (num_v6 == IPV6_NUM_ADDR)
{
IPACMDBG_H("ipv6 addr is not found. \n");
return IPACM_FAILURE;
}
for(tx_index = 0; tx_index < iface_query->num_tx_props; tx_index++)
{
if((tx_prop->tx[tx_index].ip == IPA_IP_v6) && (get_client_memptr(eth_client, clnt_indx)->route_rule_set_v6 != 0))
{
IPACMDBG_H("Delete client index %d ipv6 RT-rules for %d-st ipv6 for tx:%d\n", clnt_indx, num_v6, tx_index);
rt_hdl = get_client_memptr(eth_client, clnt_indx)->eth_rt_hdl[tx_index].eth_rt_rule_hdl_v6[num_v6];
if(m_routing.DeleteRoutingHdl(rt_hdl, IPA_IP_v6) == false)
{
return IPACM_FAILURE;
}
rt_hdl = get_client_memptr(eth_client, clnt_indx)->eth_rt_hdl[tx_index].eth_rt_rule_hdl_v6_wan[num_v6];
if(m_routing.DeleteRoutingHdl(rt_hdl, IPA_IP_v6) == false)
{
return IPACM_FAILURE;
}
get_client_memptr(eth_client, clnt_indx)->ipv6_set--;
get_client_memptr(eth_client, clnt_indx)->route_rule_set_v6--;
for(;num_v6< get_client_memptr(eth_client, clnt_indx)->ipv6_set;num_v6++)
{
get_client_memptr(eth_client, clnt_indx)->v6_addr[num_v6][0] =
get_client_memptr(eth_client, clnt_indx)->v6_addr[num_v6+1][0];
get_client_memptr(eth_client, clnt_indx)->v6_addr[num_v6][1] =
get_client_memptr(eth_client, clnt_indx)->v6_addr[num_v6+1][1];
get_client_memptr(eth_client, clnt_indx)->v6_addr[num_v6][2] =
get_client_memptr(eth_client, clnt_indx)->v6_addr[num_v6+1][2];
get_client_memptr(eth_client, clnt_indx)->v6_addr[num_v6][3] =
get_client_memptr(eth_client, clnt_indx)->v6_addr[num_v6+1][3];
get_client_memptr(eth_client, clnt_indx)->eth_rt_hdl[tx_index].eth_rt_rule_hdl_v6[num_v6] =
get_client_memptr(eth_client, clnt_indx)->eth_rt_hdl[tx_index].eth_rt_rule_hdl_v6[num_v6+1];
get_client_memptr(eth_client, clnt_indx)->eth_rt_hdl[tx_index].eth_rt_rule_hdl_v6_wan[num_v6] =
get_client_memptr(eth_client, clnt_indx)->eth_rt_hdl[tx_index].eth_rt_rule_hdl_v6_wan[num_v6+1];
}
}
}
}
return IPACM_SUCCESS;
}
/* delete filter rule for wan_down event for IPv4*/
int IPACM_Lan::handle_wan_down(ipacm_wan_iface_type backhaul_mode)
{
ipa_fltr_installed_notif_req_msg_v01 flt_index;
int fd;
fd = open(IPA_DEVICE_NAME, O_RDWR);
if (0 == fd)
{
IPACMERR("Failed opening %s.\n", IPA_DEVICE_NAME);
return IPACM_FAILURE;
}
#ifdef FEATURE_IPA_ANDROID
/* indicate v4-offload remove */
if (IPACM_Wan::isXlat() && (IPACM_OffloadManager::num_offload_v4_tethered_iface > 0)) {
IPACM_OffloadManager::num_offload_v4_tethered_iface--;
IPACMDBG_H("num_offload_v4_tethered_iface %d\n", IPACM_OffloadManager::num_offload_v4_tethered_iface);
}
#endif
if(backhaul_mode == Q6_WAN && modem_ul_v4_set == true)
{
if (num_wan_ul_fl_rule_v4 > MAX_WAN_UL_FILTER_RULES)
{
IPACMERR("number of wan_ul_fl_rule_v4 (%d) > MAX_WAN_UL_FILTER_RULES (%d), aborting...\n", num_wan_ul_fl_rule_v4, MAX_WAN_UL_FILTER_RULES);
close(fd);
return IPACM_FAILURE;
}
if (num_wan_ul_fl_rule_v4 == 0)
{
IPACMERR("No modem UL rules were installed, return...\n");
close(fd);
return IPACM_FAILURE;
}
if (m_filtering.DeleteFilteringHdls(wan_ul_fl_rule_hdl_v4,
IPA_IP_v4, num_wan_ul_fl_rule_v4) == false)
{
IPACMERR("Error Deleting RuleTable(1) to Filtering, aborting...\n");
close(fd);
return IPACM_FAILURE;
}
IPACM_Iface::ipacmcfg->decreaseFltRuleCount(rx_prop->rx[0].src_pipe, IPA_IP_v4, num_wan_ul_fl_rule_v4);
memset(wan_ul_fl_rule_hdl_v4, 0, MAX_WAN_UL_FILTER_RULES * sizeof(uint32_t));
num_wan_ul_fl_rule_v4 = 0;
modem_ul_v4_set = false;
memset(&flt_index, 0, sizeof(flt_index));
flt_index.source_pipe_index = ioctl(fd, IPA_IOC_QUERY_EP_MAPPING, rx_prop->rx[0].src_pipe);
if ((int)flt_index.source_pipe_index == -1)
{
IPACMERR("Error Query src pipe idx, aborting...\n");
close(fd);
return IPACM_FAILURE;
}
flt_index.install_status = IPA_QMI_RESULT_SUCCESS_V01;
if (!IPACM_Iface::ipacmcfg->isIPAv3Supported())
{
flt_index.filter_index_list_len = 0;
}
else /* IPAv3 */
{
flt_index.rule_id_valid = 1;
flt_index.rule_id_len = 0;
}
flt_index.embedded_pipe_index_valid = 1;
flt_index.embedded_pipe_index = ioctl(fd, IPA_IOC_QUERY_EP_MAPPING, IPA_CLIENT_APPS_LAN_WAN_PROD);
if ((int)flt_index.embedded_pipe_index == -1)
{
IPACMERR("Error Query emb pipe idx, aborting...\n");
close(fd);
return IPACM_FAILURE;
}
flt_index.retain_header_valid = 1;
flt_index.retain_header = 0;
flt_index.embedded_call_mux_id_valid = 1;
flt_index.embedded_call_mux_id = v4_mux_id;
v4_mux_id = 0;
if(false == m_filtering.SendFilteringRuleIndex(&flt_index))
{
IPACMERR("Error sending filtering rule index, aborting...\n");
close(fd);
return IPACM_FAILURE;
}
}
else
{
if (m_filtering.DeleteFilteringHdls(&lan_wan_fl_rule_hdl[0], IPA_IP_v4, 1) == false)
{
IPACMERR("Error Adding RuleTable(1) to Filtering, aborting...\n");
close(fd);
return IPACM_FAILURE;
}
IPACM_Iface::ipacmcfg->decreaseFltRuleCount(rx_prop->rx[0].src_pipe, IPA_IP_v4, 1);
sta_ul_v4_set = false;
}
/* clean MTU rules if needed */
handle_private_subnet_android(IPA_IP_v4);
close(fd);
return IPACM_SUCCESS;
}
/* handle new_address event*/
int IPACM_Lan::handle_addr_evt(ipacm_event_data_addr *data)
{
struct ipa_ioc_add_rt_rule *rt_rule;
struct ipa_rt_rule_add *rt_rule_entry;
const int NUM_RULES = 1;
uint32_t num_ipv6_addr;
int res = IPACM_SUCCESS;
#ifdef FEATURE_IPACM_HAL
IPACM_OffloadManager* OffloadMng;
#endif
IPACMDBG_H("set route/filter rule ip-type: %d \n", data->iptype);
/* Add private subnet*/
#ifdef FEATURE_IPA_ANDROID
if (data->iptype == IPA_IP_v4)
{
IPACMDBG_H("current IPACM private subnet_addr number(%d)\n", IPACM_Iface::ipacmcfg->ipa_num_private_subnet);
if_ipv4_subnet = (data->ipv4_addr >> 8) << 8;
IPACMDBG_H(" Add IPACM private subnet_addr as: 0x%x \n", if_ipv4_subnet);
if(IPACM_Iface::ipacmcfg->AddPrivateSubnet(if_ipv4_subnet, ipa_if_num) == false)
{
IPACMERR(" can't Add IPACM private subnet_addr as: 0x%x \n", if_ipv4_subnet);
}
}
#endif /* defined(FEATURE_IPA_ANDROID)*/
/* Update the IP Type. */
config_ip_type(data->iptype);
if (data->iptype == IPA_IP_v4)
{
rt_rule = (struct ipa_ioc_add_rt_rule *)
calloc(1, sizeof(struct ipa_ioc_add_rt_rule) +
NUM_RULES * sizeof(struct ipa_rt_rule_add));
if (!rt_rule)
{
IPACMERR("Error Locate ipa_ioc_add_rt_rule memory...\n");
return IPACM_FAILURE;
}
rt_rule->commit = 1;
rt_rule->num_rules = NUM_RULES;
rt_rule->ip = data->iptype;
rt_rule_entry = &rt_rule->rules[0];
rt_rule_entry->at_rear = false;
rt_rule_entry->rule.dst = IPA_CLIENT_APPS_LAN_CONS; //go to A5
rt_rule_entry->rule.attrib.attrib_mask = IPA_FLT_DST_ADDR;
strlcpy(rt_rule->rt_tbl_name, IPACM_Iface::ipacmcfg->rt_tbl_lan_v4.name, sizeof(rt_rule->rt_tbl_name));
rt_rule_entry->rule.attrib.u.v4.dst_addr = data->ipv4_addr;
rt_rule_entry->rule.attrib.u.v4.dst_addr_mask = 0xFFFFFFFF;
if (IPACM_Iface::ipacmcfg->isIPAv3Supported())
rt_rule_entry->rule.hashable = true;
if (false == m_routing.AddRoutingRule(rt_rule))
{
IPACMERR("Routing rule addition failed!\n");
res = IPACM_FAILURE;
goto fail;
}
else if (rt_rule_entry->status)
{
IPACMERR("rt rule adding failed. Result=%d\n", rt_rule_entry->status);
res = rt_rule_entry->status;
goto fail;
}
dft_rt_rule_hdl[0] = rt_rule_entry->rt_rule_hdl;
IPACMDBG_H("ipv4 iface rt-rule hdl1=0x%x\n", dft_rt_rule_hdl[0]);
/* initial multicast/broadcast/fragment filter rule */
init_fl_rule(data->iptype);
#ifdef FEATURE_L2TP
if(ipa_if_cate == WLAN_IF)
{
add_tcp_syn_flt_rule(data->iptype);
}
#endif
install_ipv4_icmp_flt_rule();
/* populate the flt rule offset for eth bridge */
eth_bridge_flt_rule_offset[data->iptype] = ipv4_icmp_flt_rule_hdl[0];
eth_bridge_post_event(IPA_ETH_BRIDGE_IFACE_UP, IPA_IP_v4, NULL, NULL, NULL, IPA_CLIENT_MAX);
}
else
{
/* check if see that v6-addr already or not*/
for(num_ipv6_addr=0;num_ipv6_addr<num_dft_rt_v6;num_ipv6_addr++)
{
if((ipv6_addr[num_ipv6_addr][0] == data->ipv6_addr[0]) &&
(ipv6_addr[num_ipv6_addr][1] == data->ipv6_addr[1]) &&
(ipv6_addr[num_ipv6_addr][2] == data->ipv6_addr[2]) &&
(ipv6_addr[num_ipv6_addr][3] == data->ipv6_addr[3]))
{
return IPACM_FAILURE;
break;
}
}
rt_rule = (struct ipa_ioc_add_rt_rule *)
calloc(1, sizeof(struct ipa_ioc_add_rt_rule) +
NUM_RULES * sizeof(struct ipa_rt_rule_add));
if (!rt_rule)
{
IPACMERR("Error Locate ipa_ioc_add_rt_rule memory...\n");
return IPACM_FAILURE;
}
rt_rule->commit = 1;
rt_rule->num_rules = NUM_RULES;
rt_rule->ip = data->iptype;
strlcpy(rt_rule->rt_tbl_name, IPACM_Iface::ipacmcfg->rt_tbl_v6.name, sizeof(rt_rule->rt_tbl_name));
rt_rule_entry = &rt_rule->rules[0];
rt_rule_entry->at_rear = false;
rt_rule_entry->rule.dst = IPA_CLIENT_APPS_LAN_CONS; //go to A5
rt_rule_entry->rule.attrib.attrib_mask = IPA_FLT_DST_ADDR;
rt_rule_entry->rule.attrib.u.v6.dst_addr[0] = data->ipv6_addr[0];
rt_rule_entry->rule.attrib.u.v6.dst_addr[1] = data->ipv6_addr[1];
rt_rule_entry->rule.attrib.u.v6.dst_addr[2] = data->ipv6_addr[2];
rt_rule_entry->rule.attrib.u.v6.dst_addr[3] = data->ipv6_addr[3];
rt_rule_entry->rule.attrib.u.v6.dst_addr_mask[0] = 0xFFFFFFFF;
rt_rule_entry->rule.attrib.u.v6.dst_addr_mask[1] = 0xFFFFFFFF;
rt_rule_entry->rule.attrib.u.v6.dst_addr_mask[2] = 0xFFFFFFFF;
rt_rule_entry->rule.attrib.u.v6.dst_addr_mask[3] = 0xFFFFFFFF;
ipv6_addr[num_dft_rt_v6][0] = data->ipv6_addr[0];
ipv6_addr[num_dft_rt_v6][1] = data->ipv6_addr[1];
ipv6_addr[num_dft_rt_v6][2] = data->ipv6_addr[2];
ipv6_addr[num_dft_rt_v6][3] = data->ipv6_addr[3];
if (IPACM_Iface::ipacmcfg->isIPAv3Supported())
rt_rule_entry->rule.hashable = true;
if (false == m_routing.AddRoutingRule(rt_rule))
{
IPACMERR("Routing rule addition failed!\n");
res = IPACM_FAILURE;
goto fail;
}
else if (rt_rule_entry->status)
{
IPACMERR("rt rule adding failed. Result=%d\n", rt_rule_entry->status);
res = rt_rule_entry->status;
goto fail;
}
dft_rt_rule_hdl[MAX_DEFAULT_v4_ROUTE_RULES + 2*num_dft_rt_v6] = rt_rule_entry->rt_rule_hdl;
/* setup same rule for v6_wan table*/
strlcpy(rt_rule->rt_tbl_name, IPACM_Iface::ipacmcfg->rt_tbl_wan_v6.name, sizeof(rt_rule->rt_tbl_name));
if (false == m_routing.AddRoutingRule(rt_rule))
{
IPACMERR("Routing rule addition failed!\n");
res = IPACM_FAILURE;
goto fail;
}
else if (rt_rule_entry->status)
{
IPACMERR("rt rule adding failed. Result=%d\n", rt_rule_entry->status);
res = rt_rule_entry->status;
goto fail;
}
dft_rt_rule_hdl[MAX_DEFAULT_v4_ROUTE_RULES + 2*num_dft_rt_v6+1] = rt_rule_entry->rt_rule_hdl;
IPACMDBG_H("ipv6 wan iface rt-rule hdl=0x%x hdl=0x%x, num_dft_rt_v6: %d \n",
dft_rt_rule_hdl[MAX_DEFAULT_v4_ROUTE_RULES + 2*num_dft_rt_v6],
dft_rt_rule_hdl[MAX_DEFAULT_v4_ROUTE_RULES + 2*num_dft_rt_v6+1],num_dft_rt_v6);
if (num_dft_rt_v6 == 0)
{
#ifdef FEATURE_L2TP
if(ipa_if_cate == WLAN_IF)
{
add_tcp_syn_flt_rule(data->iptype);
}
else if(ipa_if_cate == ODU_IF)
{
add_tcp_syn_flt_rule_l2tp(IPA_IP_v4);
add_tcp_syn_flt_rule_l2tp(IPA_IP_v6);
}
#endif
install_ipv6_icmp_flt_rule();
/* populate the flt rule offset for eth bridge */
eth_bridge_flt_rule_offset[data->iptype] = ipv6_icmp_flt_rule_hdl[0];
eth_bridge_post_event(IPA_ETH_BRIDGE_IFACE_UP, IPA_IP_v6, NULL, NULL, NULL, IPA_CLIENT_MAX);
init_fl_rule(data->iptype);
}
num_dft_rt_v6++;
IPACMDBG_H("number of default route rules %d\n", num_dft_rt_v6);
}
#ifdef FEATURE_IPACM_HAL
/* check if having pending add_downstream cache*/
OffloadMng = IPACM_OffloadManager::GetInstance();
if (OffloadMng == NULL) {
IPACMERR("failed to get IPACM_OffloadManager instance !\n");
} else {
IPACMDBG_H(" check iface %s if having add_downstream cache events\n", dev_name);
OffloadMng->search_framwork_cache(dev_name);
}
#endif
IPACMDBG_H("finish route/filter rule ip-type: %d, res(%d)\n", data->iptype, res);
fail:
free(rt_rule);
return res;
}
/* configure private subnet filter rules*/
int IPACM_Lan::handle_private_subnet(ipa_ip_type iptype)
{
struct ipa_flt_rule_add flt_rule_entry;
int i;
bool result;
ipa_ioc_add_flt_rule *m_pFilteringTable;
IPACMDBG_H("lan->handle_private_subnet(); set route/filter rule \n");
if (rx_prop == NULL)
{
IPACMDBG_H("No rx properties registered for iface %s\n", dev_name);
return IPACM_SUCCESS;
}
if (iptype == IPA_IP_v4)
{
m_pFilteringTable = (struct ipa_ioc_add_flt_rule *)
calloc(1,
sizeof(struct ipa_ioc_add_flt_rule) +
(IPACM_Iface::ipacmcfg->ipa_num_private_subnet) * sizeof(struct ipa_flt_rule_add)
);
if (!m_pFilteringTable)
{
PERROR("Error Locate ipa_flt_rule_add memory...\n");
return IPACM_FAILURE;
}
m_pFilteringTable->commit = 1;
m_pFilteringTable->ep = rx_prop->rx[0].src_pipe;
m_pFilteringTable->global = false;
m_pFilteringTable->ip = IPA_IP_v4;
m_pFilteringTable->num_rules = (uint8_t)IPACM_Iface::ipacmcfg->ipa_num_private_subnet;
/* Make LAN-traffic always go A5, use default IPA-RT table */
if (false == m_routing.GetRoutingTable(&IPACM_Iface::ipacmcfg->rt_tbl_default_v4))
{
IPACMERR("LAN m_routing.GetRoutingTable(&IPACM_Iface::ipacmcfg->rt_tbl_default_v4=0x%p) Failed.\n", &IPACM_Iface::ipacmcfg->rt_tbl_default_v4);
free(m_pFilteringTable);
return IPACM_FAILURE;
}
for (i = 0; i < (IPACM_Iface::ipacmcfg->ipa_num_private_subnet); i++)
{
memset(&flt_rule_entry, 0, sizeof(struct ipa_flt_rule_add));
flt_rule_entry.at_rear = true;
flt_rule_entry.rule.retain_hdr = 1;
flt_rule_entry.flt_rule_hdl = -1;
flt_rule_entry.status = -1;
flt_rule_entry.rule.action = IPA_PASS_TO_ROUTING;
if (IPACM_Iface::ipacmcfg->isIPAv3Supported())
flt_rule_entry.rule.hashable = true;
/* Support private subnet feature including guest-AP can't talk to primary AP etc */
flt_rule_entry.rule.rt_tbl_hdl = IPACM_Iface::ipacmcfg->rt_tbl_default_v4.hdl;
IPACMDBG_H(" private filter rule use table: %s\n",IPACM_Iface::ipacmcfg->rt_tbl_default_v4.name);
memcpy(&flt_rule_entry.rule.attrib,
&rx_prop->rx[0].attrib,
sizeof(flt_rule_entry.rule.attrib));
flt_rule_entry.rule.attrib.attrib_mask |= IPA_FLT_DST_ADDR;
flt_rule_entry.rule.attrib.u.v4.dst_addr_mask = IPACM_Iface::ipacmcfg->private_subnet_table[i].subnet_mask;
flt_rule_entry.rule.attrib.u.v4.dst_addr = IPACM_Iface::ipacmcfg->private_subnet_table[i].subnet_addr;
memcpy(&(m_pFilteringTable->rules[i]), &flt_rule_entry, sizeof(struct ipa_flt_rule_add));
IPACMDBG_H("Loop %d 5\n", i);
}
#ifdef IPA_IOCTL_SET_FNR_COUNTER_INFO
/* use index hw-counter */
if(ipa_if_cate == WLAN_IF && IPACM_Iface::ipacmcfg->hw_fnr_stats_support)
{
IPACMDBG_H("hw-index-enable %d, counter %d\n", IPACM_Iface::ipacmcfg->hw_fnr_stats_support, IPACM_Iface::ipacmcfg->hw_counter_offset + UL_ALL);
result = m_filtering.AddFilteringRule_hw_index(m_pFilteringTable, IPACM_Iface::ipacmcfg->hw_counter_offset + UL_ALL);
} else {
result = m_filtering.AddFilteringRule(m_pFilteringTable);
}
#else
result = m_filtering.AddFilteringRule(m_pFilteringTable);
#endif
if (result == false)
{
IPACMERR("Error Adding RuleTable(0) to Filtering, aborting...\n");
free(m_pFilteringTable);
return IPACM_FAILURE;
}
IPACM_Iface::ipacmcfg->increaseFltRuleCount(rx_prop->rx[0].src_pipe, IPA_IP_v4, IPACM_Iface::ipacmcfg->ipa_num_private_subnet);
/* copy filter rule hdls */
for (i = 0; i < IPACM_Iface::ipacmcfg->ipa_num_private_subnet; i++)
{
private_fl_rule_hdl[i] = m_pFilteringTable->rules[i].flt_rule_hdl;
}
free(m_pFilteringTable);
}
else
{
IPACMDBG_H("No private subnet rules for ipv6 iface %s\n", dev_name);
}
return IPACM_SUCCESS;
}
/* for STA mode wan up: configure filter rule for wan_up event*/
int IPACM_Lan::handle_wan_up(ipa_ip_type ip_type)
{
struct ipa_flt_rule_add flt_rule_entry;
int len = 0;
ipa_ioc_add_flt_rule *m_pFilteringTable;
bool result;
IPACMDBG_H("set WAN interface as default filter rule\n");
if (rx_prop == NULL)
{
IPACMDBG_H("No rx properties registered for iface %s\n", dev_name);
return IPACM_SUCCESS;
}
if(ip_type == IPA_IP_v4)
{
/* add MTU rules for ipv4 */
handle_private_subnet_android(IPA_IP_v4);
/* Update ipv6 MTU here if WAN_v6 is up and filter rules were installed */
if (IPACM_Wan::isWanUP_V6(ipa_if_num))
{
if (ipv6_prefix_flt_rule_hdl[0] && ipv6_prefix_flt_rule_hdl[1])
{
modify_ipv6_prefix_flt_rule(IPACM_Wan::backhaul_ipv6_prefix);
}
}
if(sta_ul_v4_set == true)
{
IPACMDBG_H("Filetring rule for IPV4 of STA mode is already configured, sta_ul_v4_set: %d\n",sta_ul_v4_set);
return IPACM_FAILURE;
}
len = sizeof(struct ipa_ioc_add_flt_rule) + (1 * sizeof(struct ipa_flt_rule_add));
m_pFilteringTable = (struct ipa_ioc_add_flt_rule *)calloc(1, len);
if (m_pFilteringTable == NULL)
{
PERROR("Error Locate ipa_flt_rule_add memory...\n");
return IPACM_FAILURE;
}
m_pFilteringTable->commit = 1;
m_pFilteringTable->ep = rx_prop->rx[0].src_pipe;
m_pFilteringTable->global = false;
m_pFilteringTable->ip = IPA_IP_v4;
m_pFilteringTable->num_rules = (uint8_t)1;
IPACMDBG_H("Retrieving routing hanle for table: %s\n",
IPACM_Iface::ipacmcfg->rt_tbl_wan_v4.name);
if (false == m_routing.GetRoutingTable(&IPACM_Iface::ipacmcfg->rt_tbl_wan_v4))
{
IPACMERR("m_routing.GetRoutingTable(&IPACM_Iface::ipacmcfg->rt_tbl_wan_v4=0x%p) Failed.\n",
&IPACM_Iface::ipacmcfg->rt_tbl_wan_v4);
free(m_pFilteringTable);
return IPACM_FAILURE;
}
IPACMDBG_H("Routing hanle for table: %d\n", IPACM_Iface::ipacmcfg->rt_tbl_wan_v4.hdl);
memset(&flt_rule_entry, 0, sizeof(struct ipa_flt_rule_add)); // Zero All Fields
flt_rule_entry.at_rear = true;
flt_rule_entry.flt_rule_hdl = -1;
flt_rule_entry.status = -1;
if(IPACM_Wan::isWan_Bridge_Mode())
{
flt_rule_entry.rule.action = IPA_PASS_TO_ROUTING;
}
else
{
flt_rule_entry.rule.action = IPA_PASS_TO_SRC_NAT; //IPA_PASS_TO_ROUTING
}
if (IPACM_Iface::ipacmcfg->isIPAv3Supported())
flt_rule_entry.rule.hashable = true;
flt_rule_entry.rule.rt_tbl_hdl = IPACM_Iface::ipacmcfg->rt_tbl_wan_v4.hdl;
memcpy(&flt_rule_entry.rule.attrib,
&rx_prop->rx[0].attrib,
sizeof(flt_rule_entry.rule.attrib));
flt_rule_entry.rule.attrib.attrib_mask |= IPA_FLT_DST_ADDR;
flt_rule_entry.rule.attrib.u.v4.dst_addr_mask = 0x0;
flt_rule_entry.rule.attrib.u.v4.dst_addr = 0x0;
/* only offload UL traffic of certain clients */
#ifdef FEATURE_IPACM_HAL
flt_rule_entry.rule.attrib.attrib_mask |= IPA_FLT_SRC_ADDR;
flt_rule_entry.rule.attrib.u.v4.src_addr_mask = prefix[IPA_IP_v4].v4Mask;
flt_rule_entry.rule.attrib.u.v4.src_addr = prefix[IPA_IP_v4].v4Addr;
#endif
memcpy(&m_pFilteringTable->rules[0], &flt_rule_entry, sizeof(flt_rule_entry));
#ifdef IPA_IOCTL_SET_FNR_COUNTER_INFO
/* use index hw-counter */
if(ipa_if_cate == WLAN_IF && IPACM_Iface::ipacmcfg->hw_fnr_stats_support)
{
IPACMDBG_H("hw-index-enable %d, counter %d\n", IPACM_Iface::ipacmcfg->hw_fnr_stats_support, IPACM_Iface::ipacmcfg->hw_counter_offset + UL_ALL);
result = m_filtering.AddFilteringRule_hw_index(m_pFilteringTable, IPACM_Iface::ipacmcfg->hw_counter_offset + UL_ALL);
} else {
result = m_filtering.AddFilteringRule(m_pFilteringTable);
}
#else
result = m_filtering.AddFilteringRule(m_pFilteringTable);
#endif
if (result == false)
{
IPACMERR("Error Adding RuleTable(0) to Filtering, aborting...\n");
free(m_pFilteringTable);
return IPACM_FAILURE;
}
else
{
IPACM_Iface::ipacmcfg->increaseFltRuleCount(rx_prop->rx[0].src_pipe, IPA_IP_v4, 1);
IPACMDBG_H("flt rule hdl0=0x%x, status=0x%x\n",
m_pFilteringTable->rules[0].flt_rule_hdl,
m_pFilteringTable->rules[0].status);
}
sta_ul_v4_set = true;
/* copy filter hdls */
lan_wan_fl_rule_hdl[0] = m_pFilteringTable->rules[0].flt_rule_hdl;
free(m_pFilteringTable);
}
else if(ip_type == IPA_IP_v6)
{
if(sta_ul_v6_set == true)
{
IPACMDBG_H("Filetring rule for IPV6 of STA mode is already configured, sta_ul_v6_set: %d\n",sta_ul_v6_set);
return IPACM_FAILURE;
}
/* add default v6 filter rule */
m_pFilteringTable = (struct ipa_ioc_add_flt_rule *)
calloc(1, sizeof(struct ipa_ioc_add_flt_rule) +
1 * sizeof(struct ipa_flt_rule_add));
if (!m_pFilteringTable)
{
PERROR("Error Locate ipa_flt_rule_add memory...\n");
return IPACM_FAILURE;
}
m_pFilteringTable->commit = 1;
m_pFilteringTable->ep = rx_prop->rx[0].src_pipe;
m_pFilteringTable->global = false;
m_pFilteringTable->ip = IPA_IP_v6;
m_pFilteringTable->num_rules = (uint8_t)1;
if (false == m_routing.GetRoutingTable(&IPACM_Iface::ipacmcfg->rt_tbl_v6))
{
IPACMERR("m_routing.GetRoutingTable(&IPACM_Iface::ipacmcfg->rt_tbl_v6=0x%p) Failed.\n", &IPACM_Iface::ipacmcfg->rt_tbl_v6);
free(m_pFilteringTable);
return IPACM_FAILURE;
}
memset(&flt_rule_entry, 0, sizeof(struct ipa_flt_rule_add));
flt_rule_entry.at_rear = true;
flt_rule_entry.flt_rule_hdl = -1;
flt_rule_entry.status = -1;
flt_rule_entry.rule.action = IPA_PASS_TO_ROUTING;
if (IPACM_Iface::ipacmcfg->isIPAv3Supported())
flt_rule_entry.rule.hashable = true;
flt_rule_entry.rule.rt_tbl_hdl = IPACM_Iface::ipacmcfg->rt_tbl_v6.hdl;
memcpy(&flt_rule_entry.rule.attrib,
&rx_prop->rx[0].attrib,
sizeof(flt_rule_entry.rule.attrib));
flt_rule_entry.rule.attrib.attrib_mask |= IPA_FLT_DST_ADDR;
flt_rule_entry.rule.attrib.u.v6.dst_addr_mask[0] = 0x00000000;
flt_rule_entry.rule.attrib.u.v6.dst_addr_mask[1] = 0x00000000;
flt_rule_entry.rule.attrib.u.v6.dst_addr_mask[2] = 0x00000000;
flt_rule_entry.rule.attrib.u.v6.dst_addr_mask[3] = 0x00000000;
flt_rule_entry.rule.attrib.u.v6.dst_addr[0] = 0X00000000;
flt_rule_entry.rule.attrib.u.v6.dst_addr[1] = 0x00000000;
flt_rule_entry.rule.attrib.u.v6.dst_addr[2] = 0x00000000;
flt_rule_entry.rule.attrib.u.v6.dst_addr[3] = 0X00000000;
/* only offload UL traffic of certain clients */
#ifdef FEATURE_IPACM_HAL
flt_rule_entry.rule.attrib.attrib_mask |= IPA_FLT_SRC_ADDR;
flt_rule_entry.rule.attrib.u.v6.src_addr_mask[0] = prefix[IPA_IP_v6].v6Mask[0];
flt_rule_entry.rule.attrib.u.v6.src_addr_mask[1] = prefix[IPA_IP_v6].v6Mask[1];
flt_rule_entry.rule.attrib.u.v6.src_addr_mask[2] = prefix[IPA_IP_v6].v6Mask[2];
flt_rule_entry.rule.attrib.u.v6.src_addr_mask[3] = prefix[IPA_IP_v6].v6Mask[3];
flt_rule_entry.rule.attrib.u.v6.src_addr[0] = prefix[IPA_IP_v6].v6Addr[0];
flt_rule_entry.rule.attrib.u.v6.src_addr[1] = prefix[IPA_IP_v6].v6Addr[1];
flt_rule_entry.rule.attrib.u.v6.src_addr[2] = prefix[IPA_IP_v6].v6Addr[2];
flt_rule_entry.rule.attrib.u.v6.src_addr[3] = prefix[IPA_IP_v6].v6Addr[3];
#endif
memcpy(&(m_pFilteringTable->rules[0]), &flt_rule_entry, sizeof(struct ipa_flt_rule_add));
#ifdef IPA_IOCTL_SET_FNR_COUNTER_INFO
/* use index hw-counter */
if(ipa_if_cate == WLAN_IF && IPACM_Iface::ipacmcfg->hw_fnr_stats_support)
{
IPACMDBG_H("hw-index-enable %d, counter %d\n", IPACM_Iface::ipacmcfg->hw_fnr_stats_support, IPACM_Iface::ipacmcfg->hw_counter_offset + UL_ALL);
result = m_filtering.AddFilteringRule_hw_index(m_pFilteringTable, IPACM_Iface::ipacmcfg->hw_counter_offset + UL_ALL);
} else {
result = m_filtering.AddFilteringRule(m_pFilteringTable);
}
#else
result = m_filtering.AddFilteringRule(m_pFilteringTable);
#endif
if (result == false)
{
IPACMERR("Error Adding Filtering rule, aborting...\n");
free(m_pFilteringTable);
return IPACM_FAILURE;
}
else
{
IPACM_Iface::ipacmcfg->increaseFltRuleCount(rx_prop->rx[0].src_pipe, IPA_IP_v6, 1);
IPACMDBG_H("flt rule hdl0=0x%x, status=0x%x\n", m_pFilteringTable->rules[0].flt_rule_hdl, m_pFilteringTable->rules[0].status);
}
sta_ul_v6_set = true;
/* copy filter hdls */
dft_v6fl_rule_hdl[IPV6_DEFAULT_FILTERTING_RULES] = m_pFilteringTable->rules[0].flt_rule_hdl;
free(m_pFilteringTable);
}
return IPACM_SUCCESS;
}
int IPACM_Lan::handle_wan_up_ex(ipacm_ext_prop *ext_prop, ipa_ip_type iptype, uint8_t xlat_mux_id)
{
int fd, ret = IPACM_SUCCESS;
uint32_t cnt;
IPACM_Config* ipacm_config = IPACM_Iface::ipacmcfg;
struct ipa_ioc_write_qmapid mux;
/* for newer versions metadata is overridden by NAT metadata replacement for IPAv4 and up */
/* this is still needed for IPv6 traffic in case qmapid need to be used */
if(rx_prop != NULL)
{
/* give mux ID of the default PDN to IPA-driver for WLAN/LAN pkts */
fd = open(IPA_DEVICE_NAME, O_RDWR);
if (0 == fd)
{
IPACMDBG_H("Failed opening %s.\n", IPA_DEVICE_NAME);
return IPACM_FAILURE;
}
mux.qmap_id = ipacm_config->GetQmapId();
IPACMDBG("get mux id %d for rx-endpoint\n", mux.qmap_id);
for(cnt=0; cnt<rx_prop->num_rx_props; cnt++)
{
mux.client = rx_prop->rx[cnt].src_pipe;
ret = ioctl(fd, IPA_IOC_WRITE_QMAPID, &mux);
if (ret)
{
IPACMERR("Failed to write mux id %d\n", mux.qmap_id);
close(fd);
return IPACM_FAILURE;
}
}
close(fd);
}
/* check only add static UL filter rule once */
if (iptype ==IPA_IP_v6 && modem_ul_v6_set == false)
{
IPACMDBG_H("IPA_IP_v6 num_dft_rt_v6 %d xlat_mux_id: %d modem_ul_v6_set: %d\n", num_dft_rt_v6, xlat_mux_id, modem_ul_v6_set);
ret = handle_uplink_filter_rule(ext_prop, iptype, xlat_mux_id);
if (ret == IPACM_SUCCESS)
modem_ul_v6_set = true;
} else if (iptype ==IPA_IP_v4 && modem_ul_v4_set == false) {
#ifdef FEATURE_IPA_ANDROID
if (IPACM_Wan::isXlat())
{
/* indicate v4-offload */
IPACM_OffloadManager::num_offload_v4_tethered_iface++;
IPACMDBG_H("in xlat: update num_offload_v4_tethered_iface %d\n", IPACM_OffloadManager::num_offload_v4_tethered_iface);
/* xlat not support for 2st tethered iface */
if (IPACM_OffloadManager::num_offload_v4_tethered_iface > 1)
{
IPACMDBG_H("Not support 2st downstream iface %s for xlat, cur: %d\n", dev_name,
IPACM_OffloadManager::num_offload_v4_tethered_iface);
return IPACM_FAILURE;
}
}
IPACMDBG_H(" support downstream iface %s, cur %d\n", dev_name,
IPACM_OffloadManager::num_offload_v4_tethered_iface);
#endif
/* add MTU rules for ipv4 */
handle_private_subnet_android(IPA_IP_v4);
/* Add filter config rules for ipv4 */
handle_filter_cfg_update(IPA_IP_v4);
/* Update ipv6 MTU here if WAN_v6 is up and filter rules were installed */
if (IPACM_Wan::isWanUP_V6(ipa_if_num))
{
if (ipv6_prefix_flt_rule_hdl[0] && ipv6_prefix_flt_rule_hdl[1])
{
modify_ipv6_prefix_flt_rule(IPACM_Wan::backhaul_ipv6_prefix);
}
}
IPACMDBG_H("check getXlat_Mux_Id:%d\n", IPACM_Wan::getXlat_Mux_Id());
IPACMDBG_H("IPA_IP_v4 xlat_mux_id: %d, modem_ul_v4_set %d\n", xlat_mux_id, modem_ul_v4_set);
ret = handle_uplink_filter_rule(ext_prop, iptype, xlat_mux_id);
if (ret == IPACM_SUCCESS)
modem_ul_v4_set = true;
} else {
IPACMDBG_H("ip-type: %d modem_ul_v4_set: %d, modem_ul_v6_set %d\n", iptype, modem_ul_v4_set, modem_ul_v6_set);
}
return ret;
}
/* handle ETH client initial, construct full headers (tx property) */
int IPACM_Lan::handle_eth_hdr_init(uint8_t *mac_addr)
{
#define ETH_IFACE_INDEX_LEN 2
int res = IPACM_SUCCESS, len = 0;
char index[ETH_IFACE_INDEX_LEN];
struct ipa_ioc_copy_hdr sCopyHeader;
struct ipa_ioc_add_hdr *pHeaderDescriptor = NULL;
uint32_t cnt;
int clnt_indx;
clnt_indx = get_eth_client_index(mac_addr);
if (clnt_indx != IPACM_INVALID_INDEX)
{
IPACMERR("eth client is found/attached already with index %d \n", clnt_indx);
return IPACM_FAILURE;
}
/* add header to IPA */
if (num_eth_client >= IPA_MAX_NUM_ETH_CLIENTS)
{
IPACMERR("Reached maximum number(%d) of eth clients\n", IPA_MAX_NUM_ETH_CLIENTS);
return IPACM_FAILURE;
}
IPACMDBG_H("ETH client number: %d\n", num_eth_client);
memcpy(get_client_memptr(eth_client, num_eth_client)->mac,
mac_addr,
sizeof(get_client_memptr(eth_client, num_eth_client)->mac));
IPACMDBG_H("Received Client MAC %02x:%02x:%02x:%02x:%02x:%02x\n",
mac_addr[0], mac_addr[1], mac_addr[2],
mac_addr[3], mac_addr[4], mac_addr[5]);
IPACMDBG_H("stored MAC %02x:%02x:%02x:%02x:%02x:%02x\n",
get_client_memptr(eth_client, num_eth_client)->mac[0],
get_client_memptr(eth_client, num_eth_client)->mac[1],
get_client_memptr(eth_client, num_eth_client)->mac[2],
get_client_memptr(eth_client, num_eth_client)->mac[3],
get_client_memptr(eth_client, num_eth_client)->mac[4],
get_client_memptr(eth_client, num_eth_client)->mac[5]);
/* add header to IPA */
if(tx_prop != NULL)
{
len = sizeof(struct ipa_ioc_add_hdr) + (1 * sizeof(struct ipa_hdr_add));
pHeaderDescriptor = (struct ipa_ioc_add_hdr *)calloc(1, len);
if (pHeaderDescriptor == NULL)
{
IPACMERR("calloc failed to allocate pHeaderDescriptor\n");
return IPACM_FAILURE;
}
/* copy partial header for v4*/
for (cnt=0; cnt<tx_prop->num_tx_props; cnt++)
{
if(tx_prop->tx[cnt].ip==IPA_IP_v4)
{
IPACMDBG_H("Got partial v4-header name from %d tx props\n", cnt);
memset(&sCopyHeader, 0, sizeof(sCopyHeader));
memcpy(sCopyHeader.name,
tx_prop->tx[cnt].hdr_name,
sizeof(sCopyHeader.name));
IPACMDBG_H("header name: %s in tx:%d\n", sCopyHeader.name,cnt);
if (m_header.CopyHeader(&sCopyHeader) == false)
{
PERROR("ioctl copy header failed");
res = IPACM_FAILURE;
goto fail;
}
IPACMDBG_H("header length: %d, paritial: %d\n", sCopyHeader.hdr_len, sCopyHeader.is_partial);
IPACMDBG_H("header eth2_ofst_valid: %d, eth2_ofst: %d\n", sCopyHeader.is_eth2_ofst_valid, sCopyHeader.eth2_ofst);
if (sCopyHeader.hdr_len > IPA_HDR_MAX_SIZE)
{
IPACMERR("header oversize\n");
res = IPACM_FAILURE;
goto fail;
}
else
{
memcpy(pHeaderDescriptor->hdr[0].hdr,
sCopyHeader.hdr,
sCopyHeader.hdr_len);
}
/* copy client mac_addr to partial header */
if (sCopyHeader.is_eth2_ofst_valid)
{
memcpy(&pHeaderDescriptor->hdr[0].hdr[sCopyHeader.eth2_ofst],
mac_addr,
IPA_MAC_ADDR_SIZE);
}
/* replace src mac to bridge mac_addr if any */
if (IPACM_Iface::ipacmcfg->ipa_bridge_enable)
{
memcpy(&pHeaderDescriptor->hdr[0].hdr[sCopyHeader.eth2_ofst+IPA_MAC_ADDR_SIZE],
IPACM_Iface::ipacmcfg->bridge_mac,
IPA_MAC_ADDR_SIZE);
IPACMDBG_H("device is in bridge mode \n");
}
pHeaderDescriptor->commit = true;
pHeaderDescriptor->num_hdrs = 1;
memset(pHeaderDescriptor->hdr[0].name, 0,
sizeof(pHeaderDescriptor->hdr[0].name));
snprintf(index,sizeof(index), "%d", ipa_if_num);
strlcpy(pHeaderDescriptor->hdr[0].name, index, sizeof(pHeaderDescriptor->hdr[0].name));
pHeaderDescriptor->hdr[0].name[IPA_RESOURCE_NAME_MAX-1] = '\0';
if (strlcat(pHeaderDescriptor->hdr[0].name, IPA_ETH_HDR_NAME_v4, sizeof(pHeaderDescriptor->hdr[0].name)) > IPA_RESOURCE_NAME_MAX)
{
IPACMERR(" header name construction failed exceed length (%zu)\n", strlen(pHeaderDescriptor->hdr[0].name));
res = IPACM_FAILURE;
goto fail;
}
snprintf(index,sizeof(index), "%d", header_name_count);
if (strlcat(pHeaderDescriptor->hdr[0].name, index, sizeof(pHeaderDescriptor->hdr[0].name)) > IPA_RESOURCE_NAME_MAX)
{
IPACMERR(" header name construction failed exceed length (%zu)\n", strlen(pHeaderDescriptor->hdr[0].name));
res = IPACM_FAILURE;
goto fail;
}
pHeaderDescriptor->hdr[0].hdr_len = sCopyHeader.hdr_len;
pHeaderDescriptor->hdr[0].hdr_hdl = -1;
pHeaderDescriptor->hdr[0].is_partial = 0;
pHeaderDescriptor->hdr[0].status = -1;
if (m_header.AddHeader(pHeaderDescriptor) == false ||
pHeaderDescriptor->hdr[0].status != 0)
{
IPACMERR("ioctl IPA_IOC_ADD_HDR failed: %d\n", pHeaderDescriptor->hdr[0].status);
res = IPACM_FAILURE;
goto fail;
}
get_client_memptr(eth_client, num_eth_client)->hdr_hdl_v4 = pHeaderDescriptor->hdr[0].hdr_hdl;
IPACMDBG_H("eth-client(%d) v4 full header name:%s header handle:(0x%x)\n",
num_eth_client,
pHeaderDescriptor->hdr[0].name,
get_client_memptr(eth_client, num_eth_client)->hdr_hdl_v4);
get_client_memptr(eth_client, num_eth_client)->ipv4_header_set=true;
break;
}
}
/* copy partial header for v6*/
for (cnt=0; cnt<tx_prop->num_tx_props; cnt++)
{
if(tx_prop->tx[cnt].ip==IPA_IP_v6)
{
IPACMDBG_H("Got partial v6-header name from %d tx props\n", cnt);
memset(&sCopyHeader, 0, sizeof(sCopyHeader));
memcpy(sCopyHeader.name,
tx_prop->tx[cnt].hdr_name,
sizeof(sCopyHeader.name));
IPACMDBG_H("header name: %s in tx:%d\n", sCopyHeader.name,cnt);
if (m_header.CopyHeader(&sCopyHeader) == false)
{
PERROR("ioctl copy header failed");
res = IPACM_FAILURE;
goto fail;
}
IPACMDBG_H("header length: %d, paritial: %d\n", sCopyHeader.hdr_len, sCopyHeader.is_partial);
IPACMDBG_H("header eth2_ofst_valid: %d, eth2_ofst: %d\n", sCopyHeader.is_eth2_ofst_valid, sCopyHeader.eth2_ofst);
if (sCopyHeader.hdr_len > IPA_HDR_MAX_SIZE)
{
IPACMERR("header oversize\n");
res = IPACM_FAILURE;
goto fail;
}
else
{
memcpy(pHeaderDescriptor->hdr[0].hdr,
sCopyHeader.hdr,
sCopyHeader.hdr_len);
}
/* copy client mac_addr to partial header */
if (sCopyHeader.is_eth2_ofst_valid)
{
memcpy(&pHeaderDescriptor->hdr[0].hdr[sCopyHeader.eth2_ofst],
mac_addr,
IPA_MAC_ADDR_SIZE);
}
/* replace src mac to bridge mac_addr if any */
if (IPACM_Iface::ipacmcfg->ipa_bridge_enable)
{
memcpy(&pHeaderDescriptor->hdr[0].hdr[sCopyHeader.eth2_ofst+IPA_MAC_ADDR_SIZE],
IPACM_Iface::ipacmcfg->bridge_mac,
IPA_MAC_ADDR_SIZE);
IPACMDBG_H("device is in bridge mode \n");
}
pHeaderDescriptor->commit = true;
pHeaderDescriptor->num_hdrs = 1;
memset(pHeaderDescriptor->hdr[0].name, 0,
sizeof(pHeaderDescriptor->hdr[0].name));
snprintf(index,sizeof(index), "%d", ipa_if_num);
strlcpy(pHeaderDescriptor->hdr[0].name, index, sizeof(pHeaderDescriptor->hdr[0].name));
pHeaderDescriptor->hdr[0].name[IPA_RESOURCE_NAME_MAX-1] = '\0';
if (strlcat(pHeaderDescriptor->hdr[0].name, IPA_ETH_HDR_NAME_v6, sizeof(pHeaderDescriptor->hdr[0].name)) > IPA_RESOURCE_NAME_MAX)
{
IPACMERR(" header name construction failed exceed length (%zu)\n", strlen(pHeaderDescriptor->hdr[0].name));
res = IPACM_FAILURE;
goto fail;
}
snprintf(index,sizeof(index), "%d", header_name_count);
if (strlcat(pHeaderDescriptor->hdr[0].name, index, sizeof(pHeaderDescriptor->hdr[0].name)) > IPA_RESOURCE_NAME_MAX)
{
IPACMERR(" header name construction failed exceed length (%zu)\n", strlen(pHeaderDescriptor->hdr[0].name));
res = IPACM_FAILURE;
goto fail;
}
pHeaderDescriptor->hdr[0].hdr_len = sCopyHeader.hdr_len;
pHeaderDescriptor->hdr[0].hdr_hdl = -1;
pHeaderDescriptor->hdr[0].is_partial = 0;
pHeaderDescriptor->hdr[0].status = -1;
if (m_header.AddHeader(pHeaderDescriptor) == false ||
pHeaderDescriptor->hdr[0].status != 0)
{
IPACMERR("ioctl IPA_IOC_ADD_HDR failed: %d\n", pHeaderDescriptor->hdr[0].status);
res = IPACM_FAILURE;
goto fail;
}
get_client_memptr(eth_client, num_eth_client)->hdr_hdl_v6 = pHeaderDescriptor->hdr[0].hdr_hdl;
IPACMDBG_H("eth-client(%d) v6 full header name:%s header handle:(0x%x)\n",
num_eth_client,
pHeaderDescriptor->hdr[0].name,
get_client_memptr(eth_client, num_eth_client)->hdr_hdl_v6);
get_client_memptr(eth_client, num_eth_client)->ipv6_header_set=true;
break;
}
}
/* initialize wifi client*/
get_client_memptr(eth_client, num_eth_client)->route_rule_set_v4 = false;
get_client_memptr(eth_client, num_eth_client)->route_rule_set_v6 = 0;
get_client_memptr(eth_client, num_eth_client)->ipv4_set = false;
get_client_memptr(eth_client, num_eth_client)->ipv6_set = 0;
num_eth_client++;
header_name_count++; //keep increasing header_name_count
res = IPACM_SUCCESS;
IPACMDBG_H("eth client number: %d\n", num_eth_client);
}
else
{
return res;
}
fail:
free(pHeaderDescriptor);
return res;
}
/*handle eth client */
int IPACM_Lan::handle_eth_client_ipaddr(ipacm_event_data_all *data)
{
int clnt_indx;
int v6_num;
uint32_t ipv6_link_local_prefix = 0xFE800000;
uint32_t ipv6_link_local_prefix_mask = 0xFFC00000;
IPACMDBG_H("number of eth clients: %d\n", num_eth_client);
IPACMDBG_H("event MAC %02x:%02x:%02x:%02x:%02x:%02x\n",
data->mac_addr[0],
data->mac_addr[1],
data->mac_addr[2],
data->mac_addr[3],
data->mac_addr[4],
data->mac_addr[5]);
clnt_indx = get_eth_client_index(data->mac_addr);
if (clnt_indx == IPACM_INVALID_INDEX)
{
IPACMERR("eth client not found/attached \n");
return IPACM_FAILURE;
}
IPACMDBG_H("Ip-type received %d\n", data->iptype);
if (data->iptype == IPA_IP_v4)
{
IPACMDBG_H("ipv4 address: 0x%x\n", data->ipv4_addr);
if (data->ipv4_addr != 0) /* not 0.0.0.0 */
{
if (get_client_memptr(eth_client, clnt_indx)->ipv4_set == false)
{
get_client_memptr(eth_client, clnt_indx)->v4_addr = data->ipv4_addr;
get_client_memptr(eth_client, clnt_indx)->ipv4_set = true;
}
else
{
/* check if client got new IPv4 address*/
if(data->ipv4_addr == get_client_memptr(eth_client, clnt_indx)->v4_addr)
{
IPACMDBG_H("Already setup ipv4 addr for client:%d, ipv4 address didn't change\n", clnt_indx);
return IPACM_FAILURE;
}
else
{
IPACMDBG_H("ipv4 addr for client:%d is changed \n", clnt_indx);
/* delete NAT rules first */
CtList->HandleNeighIpAddrDelEvt(get_client_memptr(eth_client, clnt_indx)->v4_addr);
delete_eth_rtrules(clnt_indx,IPA_IP_v4);
get_client_memptr(eth_client, clnt_indx)->route_rule_set_v4 = false;
get_client_memptr(eth_client, clnt_indx)->v4_addr = data->ipv4_addr;
}
}
}
else
{
IPACMDBG_H("Invalid client IPv4 address \n");
return IPACM_FAILURE;
}
}
else
{
if ((data->ipv6_addr[0] != 0) || (data->ipv6_addr[1] != 0) ||
(data->ipv6_addr[2] != 0) || (data->ipv6_addr[3] != 0)) /* check if all 0 not valid ipv6 address */
{
IPACMDBG_H("ipv6 address: 0x%x:%x:%x:%x\n", data->ipv6_addr[0], data->ipv6_addr[1], data->ipv6_addr[2], data->ipv6_addr[3]);
if( (data->ipv6_addr[0] & ipv6_link_local_prefix_mask) != (ipv6_link_local_prefix & ipv6_link_local_prefix_mask) &&
memcmp(ipv6_prefix, data->ipv6_addr, sizeof(ipv6_prefix)) != 0)
{
IPACMDBG_H("This IPv6 address is not global IPv6 address with correct prefix, ignore.\n");
return IPACM_FAILURE;
}
if(get_client_memptr(eth_client, clnt_indx)->ipv6_set < IPV6_NUM_ADDR)
{
for(v6_num=0;v6_num < get_client_memptr(eth_client, clnt_indx)->ipv6_set;v6_num++)
{
if( data->ipv6_addr[0] == get_client_memptr(eth_client, clnt_indx)->v6_addr[v6_num][0] &&
data->ipv6_addr[1] == get_client_memptr(eth_client, clnt_indx)->v6_addr[v6_num][1] &&
data->ipv6_addr[2]== get_client_memptr(eth_client, clnt_indx)->v6_addr[v6_num][2] &&
data->ipv6_addr[3] == get_client_memptr(eth_client, clnt_indx)->v6_addr[v6_num][3])
{
IPACMDBG_H("Already see this ipv6 addr at position: %d for client:%d\n", v6_num, clnt_indx);
return IPACM_FAILURE; /* not setup the RT rules*/
}
}
/* not see this ipv6 before for wifi client*/
get_client_memptr(eth_client, clnt_indx)->v6_addr[get_client_memptr(eth_client, clnt_indx)->ipv6_set][0] = data->ipv6_addr[0];
get_client_memptr(eth_client, clnt_indx)->v6_addr[get_client_memptr(eth_client, clnt_indx)->ipv6_set][1] = data->ipv6_addr[1];
get_client_memptr(eth_client, clnt_indx)->v6_addr[get_client_memptr(eth_client, clnt_indx)->ipv6_set][2] = data->ipv6_addr[2];
get_client_memptr(eth_client, clnt_indx)->v6_addr[get_client_memptr(eth_client, clnt_indx)->ipv6_set][3] = data->ipv6_addr[3];
get_client_memptr(eth_client, clnt_indx)->ipv6_set++;
}
else
{
IPACMDBG_H("Already got %d ipv6 addr for client:%d\n", IPV6_NUM_ADDR, clnt_indx);
return IPACM_FAILURE; /* not setup the RT rules*/
}
}
else
{
IPACMDBG_H("Invalid IPV6 address\n");
return IPACM_FAILURE;
}
}
return IPACM_SUCCESS;
}
/*handle eth client routing rule*/
int IPACM_Lan::handle_eth_client_route_rule(uint8_t *mac_addr, ipa_ip_type iptype)
{
struct ipa_ioc_add_rt_rule_v2 *rt_rule;
struct ipa_rt_rule_add_v2 *rt_rule_entry;
uint32_t tx_index;
int eth_index,v6_num;
const int NUM = 1;
if(tx_prop == NULL)
{
IPACMDBG_H("No rx properties registered for iface %s\n", dev_name);
return IPACM_SUCCESS;
}
IPACMDBG_H("Received mac_addr MAC %02x:%02x:%02x:%02x:%02x:%02x\n",
mac_addr[0], mac_addr[1], mac_addr[2],
mac_addr[3], mac_addr[4], mac_addr[5]);
eth_index = get_eth_client_index(mac_addr);
if (eth_index == IPACM_INVALID_INDEX)
{
IPACMDBG_H("eth client not found/attached \n");
return IPACM_SUCCESS;
}
if (iptype==IPA_IP_v4) {
IPACMDBG_H("eth client index: %d, ip-type: %d, ipv4_set:%d, ipv4_rule_set:%d \n", eth_index, iptype,
get_client_memptr(eth_client, eth_index)->ipv4_set,
get_client_memptr(eth_client, eth_index)->route_rule_set_v4);
} else {
IPACMDBG_H("eth client index: %d, ip-type: %d, ipv6_set:%d, ipv6_rule_num:%d \n", eth_index, iptype,
get_client_memptr(eth_client, eth_index)->ipv6_set,
get_client_memptr(eth_client, eth_index)->route_rule_set_v6);
}
/* Add default routing rules if not set yet */
if ((iptype == IPA_IP_v4
&& get_client_memptr(eth_client, eth_index)->route_rule_set_v4 == false
&& get_client_memptr(eth_client, eth_index)->ipv4_set == true)
|| (iptype == IPA_IP_v6
&& get_client_memptr(eth_client, eth_index)->route_rule_set_v6 < get_client_memptr(eth_client, eth_index)->ipv6_set
))
{
if(IPACM_Iface::ipacmcfg->GetIPAVer() >= IPA_HW_None && IPACM_Iface::ipacmcfg->GetIPAVer() < IPA_HW_v4_0)
{
/* Add corresponding ipa_rm_resource_name of TX-endpoint up before IPV6 RT-rule set */
IPACMDBG_H("dev %s add producer dependency\n", dev_name);
if (tx_prop != NULL)
{
IPACMDBG_H("depend Got pipe %d rm index : %d \n", tx_prop->tx[0].dst_pipe, IPACM_Iface::ipacmcfg->ipa_client_rm_map_tbl[tx_prop->tx[0].dst_pipe]);
IPACM_Iface::ipacmcfg->AddRmDepend(IPACM_Iface::ipacmcfg->ipa_client_rm_map_tbl[tx_prop->tx[0].dst_pipe],false);
}
}
rt_rule = static_cast<decltype(rt_rule)>(calloc(1, sizeof(*rt_rule)));
if (!rt_rule) {
PERROR("Memory allocation failure: rt_rule\n");
return IPACM_FAILURE;
}
rt_rule->rules = reinterpret_cast<decltype(rt_rule->rules)>(calloc(NUM, sizeof(*rt_rule_entry)));
if (!rt_rule->rules) {
PERROR("Memory allocation failure: rt_rule->rules\n");
free(rt_rule);
return IPACM_FAILURE;
}
rt_rule->commit = 1;
rt_rule->num_rules = static_cast<decltype(rt_rule->num_rules)>(NUM);
rt_rule->ip = iptype;
rt_rule->rule_add_size = sizeof(*rt_rule_entry);
for (tx_index = 0; tx_index < iface_query->num_tx_props; tx_index++)
{
if(iptype != tx_prop->tx[tx_index].ip)
{
IPACMDBG_H("Tx:%d, ip-type: %d conflict ip-type: %d no RT-rule added\n", tx_index,
tx_prop->tx[tx_index].ip,iptype);
continue;
}
rt_rule_entry = reinterpret_cast<decltype(rt_rule_entry)>(rt_rule->rules);
rt_rule_entry->at_rear = 0;
#ifdef IPA_FLTRT_TTL_UPDATE
if (IPACM_Iface::ipacmcfg->GetIPAVer() >= IPA_HW_v5_5)
rt_rule_entry->rule.ttl_update = true;
#endif
if (iptype == IPA_IP_v4)
{
IPACMDBG_H("client index(%d):ipv4 address: 0x%x\n", eth_index,
get_client_memptr(eth_client, eth_index)->v4_addr);
IPACMDBG_H("client(%d): v4 header handle:(0x%x)\n", eth_index,
get_client_memptr(eth_client, eth_index)->hdr_hdl_v4);
strlcpy(rt_rule->rt_tbl_name, IPACM_Iface::ipacmcfg->rt_tbl_lan_v4.name, sizeof(rt_rule->rt_tbl_name));
rt_rule->rt_tbl_name[IPA_RESOURCE_NAME_MAX-1] = '\0';
rt_rule_entry->rule.dst = tx_prop->tx[tx_index].dst_pipe;
memcpy(&rt_rule_entry->rule.attrib, &tx_prop->tx[tx_index].attrib, sizeof(rt_rule_entry->rule.attrib));
rt_rule_entry->rule.attrib.attrib_mask |= IPA_FLT_DST_ADDR;
rt_rule_entry->rule.hdr_hdl = get_client_memptr(eth_client, eth_index)->hdr_hdl_v4;
rt_rule_entry->rule.attrib.u.v4.dst_addr = get_client_memptr(eth_client, eth_index)->v4_addr;
rt_rule_entry->rule.attrib.u.v4.dst_addr_mask = 0xFFFFFFFF;
if(IPACM_Iface::ipacmcfg->GetIPAVer() >= IPA_HW_v4_0)
{
rt_rule_entry->rule.hashable = true;
}
if (!m_routing.AddRoutingRule(rt_rule))
{
IPACMERR("Routing rule addition failed!\n");
free(reinterpret_cast<void *>(rt_rule->rules));
free(rt_rule);
return IPACM_FAILURE;
}
/* copy ipv4 RT hdl */
get_client_memptr(eth_client, eth_index)->eth_rt_hdl[tx_index].eth_rt_rule_hdl_v4 = rt_rule_entry->rt_rule_hdl;
IPACMDBG_H("tx:%d, rt rule hdl=%x ip-type: %d\n", tx_index,
get_client_memptr(eth_client, eth_index)->eth_rt_hdl[tx_index].eth_rt_rule_hdl_v4, iptype);
} else {
for(v6_num = get_client_memptr(eth_client, eth_index)->route_rule_set_v6;v6_num <
get_client_memptr(eth_client, eth_index)->ipv6_set;v6_num++)
{
IPACMDBG_H("client(%d): v6 header handle:(0x%x)\n", eth_index,
get_client_memptr(eth_client, eth_index)->hdr_hdl_v6);
/* v6 LAN_RT_TBL */
strlcpy(rt_rule->rt_tbl_name, IPACM_Iface::ipacmcfg->rt_tbl_v6.name, sizeof(rt_rule->rt_tbl_name));
rt_rule->rt_tbl_name[IPA_RESOURCE_NAME_MAX-1] = '\0';
/* Support QCMAP LAN traffic feature, send to A5 */
rt_rule_entry->rule.dst = IPA_CLIENT_APPS_LAN_CONS;
memset(&rt_rule_entry->rule.attrib, 0, sizeof(rt_rule_entry->rule.attrib));
rt_rule_entry->rule.hdr_hdl = 0;
rt_rule_entry->rule.attrib.attrib_mask |= IPA_FLT_DST_ADDR;
rt_rule_entry->rule.attrib.u.v6.dst_addr[0] =
get_client_memptr(eth_client, eth_index)->v6_addr[v6_num][0];
rt_rule_entry->rule.attrib.u.v6.dst_addr[1] =
get_client_memptr(eth_client, eth_index)->v6_addr[v6_num][1];
rt_rule_entry->rule.attrib.u.v6.dst_addr[2] =
get_client_memptr(eth_client, eth_index)->v6_addr[v6_num][2];
rt_rule_entry->rule.attrib.u.v6.dst_addr[3] =
get_client_memptr(eth_client, eth_index)->v6_addr[v6_num][3];
rt_rule_entry->rule.attrib.u.v6.dst_addr_mask[0] = 0xFFFFFFFF;
rt_rule_entry->rule.attrib.u.v6.dst_addr_mask[1] = 0xFFFFFFFF;
rt_rule_entry->rule.attrib.u.v6.dst_addr_mask[2] = 0xFFFFFFFF;
rt_rule_entry->rule.attrib.u.v6.dst_addr_mask[3] = 0xFFFFFFFF;
if (IPACM_Iface::ipacmcfg->isIPAv3Supported())
rt_rule_entry->rule.hashable = true;
if (!m_routing.AddRoutingRule(rt_rule))
{
IPACMERR("Routing rule addition failed!\n");
free(reinterpret_cast<void *>(rt_rule->rules));
free(rt_rule);
return IPACM_FAILURE;
}
get_client_memptr(eth_client, eth_index)->eth_rt_hdl[tx_index].eth_rt_rule_hdl_v6[v6_num] =
rt_rule_entry->rt_rule_hdl;
IPACMDBG_H("tx:%d, rt rule hdl=%x ip-type: %d\n", tx_index,
get_client_memptr(eth_client, eth_index)->eth_rt_hdl[tx_index].eth_rt_rule_hdl_v6[v6_num],
iptype);
/*Copy same rule to v6 WAN RT TBL*/
strlcpy(rt_rule->rt_tbl_name, IPACM_Iface::ipacmcfg->rt_tbl_wan_v6.name, sizeof(rt_rule->rt_tbl_name));
rt_rule->rt_tbl_name[IPA_RESOURCE_NAME_MAX-1] = '\0';
/* Downlink traffic from Wan iface, directly through IPA */
rt_rule_entry->rule.dst = tx_prop->tx[tx_index].dst_pipe;
memcpy(&rt_rule_entry->rule.attrib, &tx_prop->tx[tx_index].attrib, sizeof(rt_rule_entry->rule.attrib));
rt_rule_entry->rule.hdr_hdl = get_client_memptr(eth_client, eth_index)->hdr_hdl_v6;
rt_rule_entry->rule.attrib.attrib_mask |= IPA_FLT_DST_ADDR;
rt_rule_entry->rule.attrib.u.v6.dst_addr[0] =
get_client_memptr(eth_client, eth_index)->v6_addr[v6_num][0];
rt_rule_entry->rule.attrib.u.v6.dst_addr[1] =
get_client_memptr(eth_client, eth_index)->v6_addr[v6_num][1];
rt_rule_entry->rule.attrib.u.v6.dst_addr[2] =
get_client_memptr(eth_client, eth_index)->v6_addr[v6_num][2];
rt_rule_entry->rule.attrib.u.v6.dst_addr[3] =
get_client_memptr(eth_client, eth_index)->v6_addr[v6_num][3];
rt_rule_entry->rule.attrib.u.v6.dst_addr_mask[0] = 0xFFFFFFFF;
rt_rule_entry->rule.attrib.u.v6.dst_addr_mask[1] = 0xFFFFFFFF;
rt_rule_entry->rule.attrib.u.v6.dst_addr_mask[2] = 0xFFFFFFFF;
rt_rule_entry->rule.attrib.u.v6.dst_addr_mask[3] = 0xFFFFFFFF;
if (IPACM_Iface::ipacmcfg->isIPAv3Supported())
rt_rule_entry->rule.hashable = true;
if (!m_routing.AddRoutingRule(rt_rule))
{
IPACMERR("Routing rule addition failed!\n");
free(reinterpret_cast<void *>(rt_rule->rules));
free(rt_rule);
return IPACM_FAILURE;
}
get_client_memptr(eth_client, eth_index)->eth_rt_hdl[tx_index].eth_rt_rule_hdl_v6_wan[v6_num] =
rt_rule_entry->rt_rule_hdl;
IPACMDBG_H("tx:%d, rt rule hdl=%x ip-type: %d\n", tx_index,
get_client_memptr(eth_client, eth_index)->eth_rt_hdl[tx_index].eth_rt_rule_hdl_v6_wan[v6_num],
iptype);
/* send client-v6 info to pcie modem only with global ipv6 with tx_index = 1 one time*/
if(is_global_ipv6_addr(get_client_memptr(eth_client, eth_index)->v6_addr[v6_num])
&& (IPACM_Wan::backhaul_mode == Q6_MHI_WAN))
{
if (add_connection(eth_index, v6_num))
{
IPACMERR("PCIE filter rule addition failed! (%d-client) %d v6-entry\n",eth_index, v6_num);
}
}
}
}
} /* end of for loop */
free(reinterpret_cast<void *>(rt_rule->rules));
free(rt_rule);
if (iptype == IPA_IP_v4)
{
get_client_memptr(eth_client, eth_index)->route_rule_set_v4 = true;
}
else
{
get_client_memptr(eth_client, eth_index)->route_rule_set_v6 = get_client_memptr(eth_client, eth_index)->ipv6_set;
}
}
return IPACM_SUCCESS;
}
/* handle odu client initial, construct full headers (tx property) */
int IPACM_Lan::handle_odu_hdr_init(uint8_t *mac_addr)
{
int res = IPACM_SUCCESS, len = 0;
struct ipa_ioc_copy_hdr sCopyHeader;
struct ipa_ioc_add_hdr *pHeaderDescriptor = NULL;
uint32_t cnt;
IPACMDBG("Received Client MAC %02x:%02x:%02x:%02x:%02x:%02x\n",
mac_addr[0], mac_addr[1], mac_addr[2],
mac_addr[3], mac_addr[4], mac_addr[5]);
/* add header to IPA */
if(tx_prop != NULL)
{
len = sizeof(struct ipa_ioc_add_hdr) + (1 * sizeof(struct ipa_hdr_add));
pHeaderDescriptor = (struct ipa_ioc_add_hdr *)calloc(1, len);
if (pHeaderDescriptor == NULL)
{
IPACMERR("calloc failed to allocate pHeaderDescriptor\n");
return IPACM_FAILURE;
}
/* copy partial header for v4*/
for (cnt=0; cnt<tx_prop->num_tx_props; cnt++)
{
if(tx_prop->tx[cnt].ip==IPA_IP_v4)
{
IPACMDBG("Got partial v4-header name from %d tx props\n", cnt);
memset(&sCopyHeader, 0, sizeof(sCopyHeader));
memcpy(sCopyHeader.name,
tx_prop->tx[cnt].hdr_name,
sizeof(sCopyHeader.name));
IPACMDBG("header name: %s in tx:%d\n", sCopyHeader.name,cnt);
if (m_header.CopyHeader(&sCopyHeader) == false)
{
PERROR("ioctl copy header failed");
res = IPACM_FAILURE;
goto fail;
}
IPACMDBG("header length: %d, paritial: %d\n", sCopyHeader.hdr_len, sCopyHeader.is_partial);
if (sCopyHeader.hdr_len > IPA_HDR_MAX_SIZE)
{
IPACMERR("header oversize\n");
res = IPACM_FAILURE;
goto fail;
}
else
{
memcpy(pHeaderDescriptor->hdr[0].hdr,
sCopyHeader.hdr,
sCopyHeader.hdr_len);
}
/* copy client mac_addr to partial header */
if (sCopyHeader.is_eth2_ofst_valid)
{
memcpy(&pHeaderDescriptor->hdr[0].hdr[sCopyHeader.eth2_ofst],
mac_addr,
IPA_MAC_ADDR_SIZE);
}
/* replace src mac to bridge mac_addr if any */
if (IPACM_Iface::ipacmcfg->ipa_bridge_enable)
{
memcpy(&pHeaderDescriptor->hdr[0].hdr[sCopyHeader.eth2_ofst+IPA_MAC_ADDR_SIZE],
IPACM_Iface::ipacmcfg->bridge_mac,
IPA_MAC_ADDR_SIZE);
IPACMDBG_H("device is in bridge mode \n");
}
pHeaderDescriptor->commit = true;
pHeaderDescriptor->num_hdrs = 1;
memset(pHeaderDescriptor->hdr[0].name, 0,
sizeof(pHeaderDescriptor->hdr[0].name));
strlcpy(pHeaderDescriptor->hdr[0].name, IPA_ODU_HDR_NAME_v4, sizeof(pHeaderDescriptor->hdr[0].name));
pHeaderDescriptor->hdr[0].hdr_len = sCopyHeader.hdr_len;
pHeaderDescriptor->hdr[0].hdr_hdl = -1;
pHeaderDescriptor->hdr[0].is_partial = 0;
pHeaderDescriptor->hdr[0].status = -1;
if (m_header.AddHeader(pHeaderDescriptor) == false ||
pHeaderDescriptor->hdr[0].status != 0)
{
IPACMERR("ioctl IPA_IOC_ADD_HDR failed: %d\n", pHeaderDescriptor->hdr[0].status);
res = IPACM_FAILURE;
goto fail;
}
ODU_hdr_hdl_v4 = pHeaderDescriptor->hdr[0].hdr_hdl;
ipv4_header_set = true ;
IPACMDBG(" ODU v4 full header name:%s header handle:(0x%x)\n",
pHeaderDescriptor->hdr[0].name,
ODU_hdr_hdl_v4);
break;
}
}
/* copy partial header for v6*/
for (cnt=0; cnt<tx_prop->num_tx_props; cnt++)
{
if(tx_prop->tx[cnt].ip==IPA_IP_v6)
{
IPACMDBG("Got partial v6-header name from %d tx props\n", cnt);
memset(&sCopyHeader, 0, sizeof(sCopyHeader));
memcpy(sCopyHeader.name,
tx_prop->tx[cnt].hdr_name,
sizeof(sCopyHeader.name));
IPACMDBG("header name: %s in tx:%d\n", sCopyHeader.name,cnt);
if (m_header.CopyHeader(&sCopyHeader) == false)
{
PERROR("ioctl copy header failed");
res = IPACM_FAILURE;
goto fail;
}
IPACMDBG("header length: %d, paritial: %d\n", sCopyHeader.hdr_len, sCopyHeader.is_partial);
if (sCopyHeader.hdr_len > IPA_HDR_MAX_SIZE)
{
IPACMERR("header oversize\n");
res = IPACM_FAILURE;
goto fail;
}
else
{
memcpy(pHeaderDescriptor->hdr[0].hdr,
sCopyHeader.hdr,
sCopyHeader.hdr_len);
}
/* copy client mac_addr to partial header */
if (sCopyHeader.is_eth2_ofst_valid)
{
memcpy(&pHeaderDescriptor->hdr[0].hdr[sCopyHeader.eth2_ofst],
mac_addr,
IPA_MAC_ADDR_SIZE);
}
/* replace src mac to bridge mac_addr if any */
if (IPACM_Iface::ipacmcfg->ipa_bridge_enable)
{
memcpy(&pHeaderDescriptor->hdr[0].hdr[sCopyHeader.eth2_ofst+IPA_MAC_ADDR_SIZE],
IPACM_Iface::ipacmcfg->bridge_mac,
IPA_MAC_ADDR_SIZE);
IPACMDBG_H("device is in bridge mode \n");
}
pHeaderDescriptor->commit = true;
pHeaderDescriptor->num_hdrs = 1;
memset(pHeaderDescriptor->hdr[0].name, 0,
sizeof(pHeaderDescriptor->hdr[0].name));
strlcpy(pHeaderDescriptor->hdr[0].name, IPA_ODU_HDR_NAME_v6, sizeof(pHeaderDescriptor->hdr[0].name));
pHeaderDescriptor->hdr[0].hdr_len = sCopyHeader.hdr_len;
pHeaderDescriptor->hdr[0].hdr_hdl = -1;
pHeaderDescriptor->hdr[0].is_partial = 0;
pHeaderDescriptor->hdr[0].status = -1;
if (m_header.AddHeader(pHeaderDescriptor) == false ||
pHeaderDescriptor->hdr[0].status != 0)
{
IPACMERR("ioctl IPA_IOC_ADD_HDR failed: %d\n", pHeaderDescriptor->hdr[0].status);
res = IPACM_FAILURE;
goto fail;
}
ODU_hdr_hdl_v6 = pHeaderDescriptor->hdr[0].hdr_hdl;
ipv6_header_set = true ;
IPACMDBG(" ODU v4 full header name:%s header handle:(0x%x)\n",
pHeaderDescriptor->hdr[0].name,
ODU_hdr_hdl_v6);
break;
}
}
}
fail:
free(pHeaderDescriptor);
return res;
}
/* handle odu default route rule configuration */
int IPACM_Lan::handle_odu_route_add()
{
/* add default WAN route */
struct ipa_ioc_add_rt_rule *rt_rule;
struct ipa_rt_rule_add *rt_rule_entry;
uint32_t tx_index;
const int NUM = 1;
if(tx_prop == NULL)
{
IPACMDBG_H("No tx properties, ignore default route setting\n");
return IPACM_SUCCESS;
}
rt_rule = (struct ipa_ioc_add_rt_rule *)
calloc(1, sizeof(struct ipa_ioc_add_rt_rule) +
NUM * sizeof(struct ipa_rt_rule_add));
if (!rt_rule)
{
IPACMERR("Error Locate ipa_ioc_add_rt_rule memory...\n");
return IPACM_FAILURE;
}
rt_rule->commit = 1;
rt_rule->num_rules = (uint8_t)NUM;
IPACMDBG_H("WAN table created %s \n", rt_rule->rt_tbl_name);
rt_rule_entry = &rt_rule->rules[0];
rt_rule_entry->at_rear = true;
for (tx_index = 0; tx_index < iface_query->num_tx_props; tx_index++)
{
if (IPA_IP_v4 == tx_prop->tx[tx_index].ip)
{
strlcpy(rt_rule->rt_tbl_name, IPACM_Iface::ipacmcfg->rt_tbl_odu_v4.name, sizeof(rt_rule->rt_tbl_name));
rt_rule_entry->rule.hdr_hdl = ODU_hdr_hdl_v4;
rt_rule->ip = IPA_IP_v4;
}
else
{
strlcpy(rt_rule->rt_tbl_name, IPACM_Iface::ipacmcfg->rt_tbl_odu_v6.name, sizeof(rt_rule->rt_tbl_name));
rt_rule_entry->rule.hdr_hdl = ODU_hdr_hdl_v6;
rt_rule->ip = IPA_IP_v6;
}
rt_rule_entry->rule.dst = tx_prop->tx[tx_index].dst_pipe;
memcpy(&rt_rule_entry->rule.attrib,
&tx_prop->tx[tx_index].attrib,
sizeof(rt_rule_entry->rule.attrib));
rt_rule_entry->rule.attrib.attrib_mask |= IPA_FLT_DST_ADDR;
if (IPA_IP_v4 == tx_prop->tx[tx_index].ip)
{
rt_rule_entry->rule.attrib.u.v4.dst_addr = 0;
rt_rule_entry->rule.attrib.u.v4.dst_addr_mask = 0;
if (IPACM_Iface::ipacmcfg->isIPAv3Supported())
rt_rule_entry->rule.hashable = true;
if (false == m_routing.AddRoutingRule(rt_rule))
{
IPACMERR("Routing rule addition failed!\n");
free(rt_rule);
return IPACM_FAILURE;
}
odu_route_rule_v4_hdl[tx_index] = rt_rule_entry->rt_rule_hdl;
IPACMDBG_H("Got ipv4 ODU-route rule hdl:0x%x,tx:%d,ip-type: %d \n",
odu_route_rule_v4_hdl[tx_index],
tx_index,
IPA_IP_v4);
}
else
{
rt_rule_entry->rule.attrib.u.v6.dst_addr[0] = 0;
rt_rule_entry->rule.attrib.u.v6.dst_addr[1] = 0;
rt_rule_entry->rule.attrib.u.v6.dst_addr[2] = 0;
rt_rule_entry->rule.attrib.u.v6.dst_addr[3] = 0;
rt_rule_entry->rule.attrib.u.v6.dst_addr_mask[0] = 0;
rt_rule_entry->rule.attrib.u.v6.dst_addr_mask[1] = 0;
rt_rule_entry->rule.attrib.u.v6.dst_addr_mask[2] = 0;
rt_rule_entry->rule.attrib.u.v6.dst_addr_mask[3] = 0;
if (IPACM_Iface::ipacmcfg->isIPAv3Supported())
rt_rule_entry->rule.hashable = true;
if (false == m_routing.AddRoutingRule(rt_rule))
{
IPACMERR("Routing rule addition failed!\n");
free(rt_rule);
return IPACM_FAILURE;
}
odu_route_rule_v6_hdl[tx_index] = rt_rule_entry->rt_rule_hdl;
IPACMDBG_H("Set ipv6 ODU-route rule hdl for v6_lan_table:0x%x,tx:%d,ip-type: %d \n",
odu_route_rule_v6_hdl[tx_index],
tx_index,
IPA_IP_v6);
}
}
free(rt_rule);
return IPACM_SUCCESS;
}
/* handle odu default route rule deletion */
int IPACM_Lan::handle_odu_route_del()
{
uint32_t tx_index;
if(tx_prop == NULL)
{
IPACMDBG_H("No tx properties, ignore delete default route setting\n");
return IPACM_SUCCESS;
}
for (tx_index = 0; tx_index < iface_query->num_tx_props; tx_index++)
{
if (tx_prop->tx[tx_index].ip == IPA_IP_v4)
{
IPACMDBG_H("Tx:%d, ip-type: %d match ip-type: %d, RT-rule deleted\n",
tx_index, tx_prop->tx[tx_index].ip,IPA_IP_v4);
if (m_routing.DeleteRoutingHdl(odu_route_rule_v4_hdl[tx_index], IPA_IP_v4)
== false)
{
IPACMERR("IP-family:%d, Routing rule(hdl:0x%x) deletion failed with tx_index %d!\n", IPA_IP_v4, odu_route_rule_v4_hdl[tx_index], tx_index);
return IPACM_FAILURE;
}
}
else
{
IPACMDBG_H("Tx:%d, ip-type: %d match ip-type: %d, RT-rule deleted\n",
tx_index, tx_prop->tx[tx_index].ip,IPA_IP_v6);
if (m_routing.DeleteRoutingHdl(odu_route_rule_v6_hdl[tx_index], IPA_IP_v6)
== false)
{
IPACMERR("IP-family:%d, Routing rule(hdl:0x%x) deletion failed with tx_index %d!\n", IPA_IP_v6, odu_route_rule_v6_hdl[tx_index], tx_index);
return IPACM_FAILURE;
}
}
}
return IPACM_SUCCESS;
}
/*handle eth client del mode*/
int IPACM_Lan::handle_eth_client_down_evt(uint8_t *mac_addr)
{
int clt_indx;
uint32_t tx_index;
int num_eth_client_tmp = num_eth_client;
int num_v6;
IPACMDBG_H("total client: %d\n", num_eth_client_tmp);
clt_indx = get_eth_client_index(mac_addr);
if (clt_indx == IPACM_INVALID_INDEX)
{
IPACMDBG_H("eth client not attached\n");
return IPACM_SUCCESS;
}
/* First reset nat rules and then route rules */
if(get_client_memptr(eth_client, clt_indx)->ipv4_set == true)
{
IPACMDBG_H("Clean Nat Rules for ipv4:0x%x\n", get_client_memptr(eth_client, clt_indx)->v4_addr);
CtList->HandleNeighIpAddrDelEvt(get_client_memptr(eth_client, clt_indx)->v4_addr);
}
if (delete_eth_rtrules(clt_indx, IPA_IP_v4))
{
IPACMERR("unbale to delete ecm-client v4 route rules for index: %d\n", clt_indx);
return IPACM_FAILURE;
}
if (delete_eth_rtrules(clt_indx, IPA_IP_v6))
{
IPACMERR("unbale to delete ecm-client v6 route rules for index: %d\n", clt_indx);
return IPACM_FAILURE;
}
/* Delete eth client header */
if(get_client_memptr(eth_client, clt_indx)->ipv4_header_set == true)
{
if (m_header.DeleteHeaderHdl(get_client_memptr(eth_client, clt_indx)->hdr_hdl_v4)
== false)
{
return IPACM_FAILURE;
}
get_client_memptr(eth_client, clt_indx)->ipv4_header_set = false;
}
if(get_client_memptr(eth_client, clt_indx)->ipv6_header_set == true)
{
if (m_header.DeleteHeaderHdl(get_client_memptr(eth_client, clt_indx)->hdr_hdl_v6)
== false)
{
return IPACM_FAILURE;
}
get_client_memptr(eth_client, clt_indx)->ipv6_header_set = false;
}
/* Reset ip_set to 0*/
get_client_memptr(eth_client, clt_indx)->ipv4_set = false;
get_client_memptr(eth_client, clt_indx)->ipv6_set = 0;
get_client_memptr(eth_client, clt_indx)->ipv4_header_set = false;
get_client_memptr(eth_client, clt_indx)->ipv6_header_set = false;
get_client_memptr(eth_client, clt_indx)->route_rule_set_v4 = false;
get_client_memptr(eth_client, clt_indx)->route_rule_set_v6 = 0;
for (; clt_indx < num_eth_client_tmp - 1; clt_indx++)
{
memcpy(get_client_memptr(eth_client, clt_indx)->mac,
get_client_memptr(eth_client, (clt_indx + 1))->mac,
sizeof(get_client_memptr(eth_client, clt_indx)->mac));
get_client_memptr(eth_client, clt_indx)->hdr_hdl_v4 = get_client_memptr(eth_client, (clt_indx + 1))->hdr_hdl_v4;
get_client_memptr(eth_client, clt_indx)->hdr_hdl_v6 = get_client_memptr(eth_client, (clt_indx + 1))->hdr_hdl_v6;
get_client_memptr(eth_client, clt_indx)->v4_addr = get_client_memptr(eth_client, (clt_indx + 1))->v4_addr;
get_client_memptr(eth_client, clt_indx)->ipv4_set = get_client_memptr(eth_client, (clt_indx + 1))->ipv4_set;
get_client_memptr(eth_client, clt_indx)->ipv6_set = get_client_memptr(eth_client, (clt_indx + 1))->ipv6_set;
get_client_memptr(eth_client, clt_indx)->ipv4_header_set = get_client_memptr(eth_client, (clt_indx + 1))->ipv4_header_set;
get_client_memptr(eth_client, clt_indx)->ipv6_header_set = get_client_memptr(eth_client, (clt_indx + 1))->ipv6_header_set;
get_client_memptr(eth_client, clt_indx)->route_rule_set_v4 = get_client_memptr(eth_client, (clt_indx + 1))->route_rule_set_v4;
get_client_memptr(eth_client, clt_indx)->route_rule_set_v6 = get_client_memptr(eth_client, (clt_indx + 1))->route_rule_set_v6;
for (num_v6=0;num_v6< get_client_memptr(eth_client, clt_indx)->ipv6_set;num_v6++)
{
get_client_memptr(eth_client, clt_indx)->v6_addr[num_v6][0] = get_client_memptr(eth_client, (clt_indx + 1))->v6_addr[num_v6][0];
get_client_memptr(eth_client, clt_indx)->v6_addr[num_v6][1] = get_client_memptr(eth_client, (clt_indx + 1))->v6_addr[num_v6][1];
get_client_memptr(eth_client, clt_indx)->v6_addr[num_v6][2] = get_client_memptr(eth_client, (clt_indx + 1))->v6_addr[num_v6][2];
get_client_memptr(eth_client, clt_indx)->v6_addr[num_v6][3] = get_client_memptr(eth_client, (clt_indx + 1))->v6_addr[num_v6][3];
}
for (tx_index = 0; tx_index < iface_query->num_tx_props; tx_index++)
{
get_client_memptr(eth_client, clt_indx)->eth_rt_hdl[tx_index].eth_rt_rule_hdl_v4 =
get_client_memptr(eth_client, (clt_indx + 1))->eth_rt_hdl[tx_index].eth_rt_rule_hdl_v4;
for(num_v6=0;num_v6< get_client_memptr(eth_client, clt_indx)->route_rule_set_v6;num_v6++)
{
get_client_memptr(eth_client, clt_indx)->eth_rt_hdl[tx_index].eth_rt_rule_hdl_v6[num_v6] =
get_client_memptr(eth_client, (clt_indx + 1))->eth_rt_hdl[tx_index].eth_rt_rule_hdl_v6[num_v6];
get_client_memptr(eth_client, clt_indx)->eth_rt_hdl[tx_index].eth_rt_rule_hdl_v6_wan[num_v6] =
get_client_memptr(eth_client, (clt_indx + 1))->eth_rt_hdl[tx_index].eth_rt_rule_hdl_v6_wan[num_v6];
}
}
}
IPACMDBG_H(" %d eth client deleted successfully \n", num_eth_client);
num_eth_client = num_eth_client - 1;
IPACMDBG_H(" Number of eth client: %d\n", num_eth_client);
/* Del RM dependency */
if(num_eth_client == 0)
{
if(IPACM_Iface::ipacmcfg->GetIPAVer() >= IPA_HW_None && IPACM_Iface::ipacmcfg->GetIPAVer() < IPA_HW_v4_0)
{
/* Delete corresponding ipa_rm_resource_name of TX-endpoint after delete all IPV4V6 RT-rule*/
IPACMDBG_H("dev %s delete producer dependency\n", dev_name);
if (tx_prop != NULL)
{
IPACMDBG_H("depend Got pipe %d rm index : %d \n", tx_prop->tx[0].dst_pipe, IPACM_Iface::ipacmcfg->ipa_client_rm_map_tbl[tx_prop->tx[0].dst_pipe]);
IPACM_Iface::ipacmcfg->DelRmDepend(IPACM_Iface::ipacmcfg->ipa_client_rm_map_tbl[tx_prop->tx[0].dst_pipe]);
}
}
}
return IPACM_SUCCESS;
}
/*handle LAN iface down event*/
int IPACM_Lan::handle_down_evt()
{
uint32_t i;
int res = IPACM_SUCCESS;
IPACMDBG_H("lan handle_down_evt\n ");
if (ipa_if_cate == ODU_IF)
{
/* delete ODU default RT rules */
if (IPACM_Iface::ipacmcfg->ipacm_odu_embms_enable == true)
{
IPACMDBG_H("eMBMS enable, delete eMBMS DL RT rule\n");
handle_odu_route_del();
}
/* delete full header */
if (ipv4_header_set)
{
if (m_header.DeleteHeaderHdl(ODU_hdr_hdl_v4)
== false)
{
IPACMERR("ODU ipv4 header delete fail\n");
res = IPACM_FAILURE;
goto fail;
}
IPACMDBG_H("ODU ipv4 header delete success\n");
}
if (ipv6_header_set)
{
if (m_header.DeleteHeaderHdl(ODU_hdr_hdl_v6)
== false)
{
IPACMERR("ODU ipv6 header delete fail\n");
res = IPACM_FAILURE;
goto fail;
}
IPACMERR("ODU ipv6 header delete success\n");
}
}
/* no iface address up, directly close iface*/
if (ip_type == IPACM_IP_NULL)
{
goto fail;
}
/* delete wan filter rule */
if (IPACM_Wan::isWanUP(ipa_if_num) && rx_prop != NULL)
{
IPACMDBG_H("LAN IF goes down, backhaul type %d\n", IPACM_Wan::backhaul_mode);
handle_wan_down(IPACM_Wan::backhaul_mode);
#ifdef FEATURE_IPA_ANDROID
#ifndef FEATURE_IPACM_HAL
/* Clean-up tethered-iface list */
IPACM_Wan::delete_tether_iface(IPA_IP_v4, ipa_if_num);
#endif
#endif
}
if (IPACM_Wan::isWanUP_V6(ipa_if_num) && rx_prop != NULL)
{
IPACMDBG_H("LAN IF goes down, backhaul type %d\n", IPACM_Wan::backhaul_mode);
handle_wan_down_v6(IPACM_Wan::backhaul_mode);
#ifdef FEATURE_IPA_ANDROID
/* Clean-up tethered-iface list */
IPACM_Wan::delete_tether_iface(IPA_IP_v6, ipa_if_num);
#endif
}
if (is_downstream_set[IPA_IP_v4])
{
is_downstream_set[IPA_IP_v4] = false;
store_downstream_state(false, IPA_IP_v4);
}
if (is_downstream_set[IPA_IP_v6])
{
is_downstream_set[IPA_IP_v6] = false;
store_downstream_state(false, IPA_IP_v6);
}
/* delete default filter rules */
if (ip_type != IPA_IP_v6 && rx_prop != NULL)
{
if(m_filtering.DeleteFilteringHdls(ipv4_icmp_flt_rule_hdl, IPA_IP_v4, NUM_IPV4_ICMP_FLT_RULE) == false)
{
IPACMERR("Error Deleting ICMPv4 Filtering Rule, aborting...\n");
res = IPACM_FAILURE;
goto fail;
}
IPACM_Iface::ipacmcfg->decreaseFltRuleCount(rx_prop->rx[0].src_pipe, IPA_IP_v4, NUM_IPV4_ICMP_FLT_RULE);
if(dft_v4fl_rule_hdl[0] != 0)
{
if (m_filtering.DeleteFilteringHdls(dft_v4fl_rule_hdl, IPA_IP_v4,
IPV4_DEFAULT_FILTERTING_RULES) == false)
{
IPACMERR("Error Deleting Filtering Rule, aborting...\n");
res = IPACM_FAILURE;
goto fail;
}
IPACM_Iface::ipacmcfg->decreaseFltRuleCount(rx_prop->rx[0].src_pipe, IPA_IP_v4, IPV4_DEFAULT_FILTERTING_RULES);
}
/* free private-subnet ipv4 filter rules */
if (IPACM_Iface::ipacmcfg->ipa_num_private_subnet > IPA_PRIV_SUBNET_FILTER_RULE_HANDLES)
{
IPACMERR(" the number of rules are bigger than array, aborting...\n");
res = IPACM_FAILURE;
goto fail;
}
#ifdef FEATURE_IPA_ANDROID
if(m_filtering.DeleteFilteringHdls(private_fl_rule_hdl, IPA_IP_v4, IPA_MAX_PRIVATE_SUBNET_ENTRIES + IPA_MAX_MTU_ENTRIES) == false)
{
IPACMERR("Error deleting private subnet IPv4 flt rules.\n");
res = IPACM_FAILURE;
goto fail;
}
IPACM_Iface::ipacmcfg->decreaseFltRuleCount(rx_prop->rx[0].src_pipe, IPA_IP_v4, IPA_MAX_PRIVATE_SUBNET_ENTRIES + IPA_MAX_MTU_ENTRIES);
#else
if (m_filtering.DeleteFilteringHdls(private_fl_rule_hdl, IPA_IP_v4, IPACM_Iface::ipacmcfg->ipa_num_private_subnet) == false)
{
IPACMERR("Error deleting private subnet IPv4 flt rules.\n");
res = IPACM_FAILURE;
goto fail;
}
IPACM_Iface::ipacmcfg->decreaseFltRuleCount(rx_prop->rx[0].src_pipe, IPA_IP_v4, IPACM_Iface::ipacmcfg->ipa_num_private_subnet);
#endif
IPACMDBG_H("Deleted private subnet v4 filter rules successfully.\n");
if(m_filtering.DeleteFilteringHdls(filter_cfg_rule_hdl, IPA_IP_v4, IPA_MAX_FILTER_CFG_ENTRIES) == false)
{
IPACMERR("Error deleting filter cfg rules.\n");
res = IPACM_FAILURE;
goto fail;
}
IPACM_Iface::ipacmcfg->decreaseFltRuleCount(rx_prop->rx[0].src_pipe, IPA_IP_v4, IPA_MAX_FILTER_CFG_ENTRIES);
}
IPACMDBG_H("Finished delete default iface ipv4 filtering rules \n ");
if (ip_type != IPA_IP_v4 && rx_prop != NULL)
{
if(m_filtering.DeleteFilteringHdls(ipv6_icmp_flt_rule_hdl, IPA_IP_v6, NUM_IPV6_ICMP_FLT_RULE) == false)
{
IPACMERR("Error Deleting ICMPv6 Filtering Rule, aborting...\n");
res = IPACM_FAILURE;
goto fail;
}
IPACM_Iface::ipacmcfg->decreaseFltRuleCount(rx_prop->rx[0].src_pipe, IPA_IP_v6, NUM_IPV6_ICMP_FLT_RULE);
if (dft_v6fl_rule_hdl[0] != 0)
{
if (m_filtering.DeleteFilteringHdls(dft_v6fl_rule_hdl, IPA_IP_v6, IPV6_DEFAULT_FILTERTING_RULES) == false)
{
IPACMERR("Error Adding RuleTable(1) to Filtering, aborting...\n");
res = IPACM_FAILURE;
goto fail;
}
IPACM_Iface::ipacmcfg->decreaseFltRuleCount(rx_prop->rx[0].src_pipe, IPA_IP_v6, IPV6_DEFAULT_FILTERTING_RULES);
}
#ifdef FEATURE_L2TP
if(ipa_if_cate == ODU_IF)
{
if(m_filtering.DeleteFilteringHdls(tcp_syn_flt_rule_hdl, IPA_IP_v6, IPA_IP_MAX) == false)
{
IPACMERR("Error Deleting TCP SYN L2TP Filtering Rule, aborting...\n");
res = IPACM_FAILURE;
goto fail;
}
}
#endif
}
IPACMDBG_H("Finished delete default iface ipv6 filtering rules \n ");
if (ip_type != IPA_IP_v6)
{
if (m_routing.DeleteRoutingHdl(dft_rt_rule_hdl[0], IPA_IP_v4)
== false)
{
IPACMERR("Routing rule deletion failed!\n");
res = IPACM_FAILURE;
goto fail;
}
}
IPACMDBG_H("Finished delete default iface ipv4 rules \n ");
/* delete default v6 routing rule */
if (ip_type != IPA_IP_v4)
{
/* may have multiple ipv6 iface-RT rules*/
for (i = 0; i < 2*num_dft_rt_v6; i++)
{
if (m_routing.DeleteRoutingHdl(dft_rt_rule_hdl[MAX_DEFAULT_v4_ROUTE_RULES + i], IPA_IP_v6)
== false)
{
IPACMERR("Routing rule deletion failed!\n");
res = IPACM_FAILURE;
goto fail;
}
}
}
IPACMDBG_H("Finished delete default iface ipv6 rules \n ");
/* free the edm clients cache */
IPACMDBG_H("Free ecm clients cache\n");
if(IPACM_Iface::ipacmcfg->GetIPAVer() >= IPA_HW_None && IPACM_Iface::ipacmcfg->GetIPAVer() < IPA_HW_v4_0)
{
/* Delete corresponding ipa_rm_resource_name of TX-endpoint after delete all IPV4V6 RT-rule */
IPACMDBG_H("dev %s delete producer dependency\n", dev_name);
if (tx_prop != NULL)
{
IPACMDBG_H("depend Got pipe %d rm index : %d \n", tx_prop->tx[0].dst_pipe, IPACM_Iface::ipacmcfg->ipa_client_rm_map_tbl[tx_prop->tx[0].dst_pipe]);
IPACM_Iface::ipacmcfg->DelRmDepend(IPACM_Iface::ipacmcfg->ipa_client_rm_map_tbl[tx_prop->tx[0].dst_pipe]);
}
}
eth_bridge_post_event(IPA_ETH_BRIDGE_IFACE_DOWN, IPA_IP_MAX, NULL, NULL, NULL, IPA_CLIENT_MAX);
/* Delete private subnet*/
#ifdef FEATURE_IPA_ANDROID
if (ip_type != IPA_IP_v6)
{
IPACMDBG_H("current IPACM private subnet_addr number(%d)\n", IPACM_Iface::ipacmcfg->ipa_num_private_subnet);
IPACMDBG_H(" Delete IPACM private subnet_addr as: 0x%x \n", if_ipv4_subnet);
if(IPACM_Iface::ipacmcfg->DelPrivateSubnet(if_ipv4_subnet, ipa_if_num) == false)
{
IPACMERR(" can't Delete IPACM private subnet_addr as: 0x%x \n", if_ipv4_subnet);
}
}
/* reset the IPA-client pipe enum */
if(ipa_if_cate != WAN_IF)
{
#ifdef FEATURE_IPACM_HAL
handle_tethering_client(true, IPACM_CLIENT_MAX);
#else
handle_tethering_client(true, IPACM_CLIENT_USB);
#endif
}
#endif /* defined(FEATURE_IPA_ANDROID)*/
fail:
/* clean eth-client header, routing rules */
IPACMDBG_H("left %d eth clients need to be deleted \n ", num_eth_client);
for (i = 0; i < num_eth_client; i++)
{
/* First reset nat rules and then route rules */
if(get_client_memptr(eth_client, i)->ipv4_set == true)
{
IPACMDBG_H("Clean Nat Rules for ipv4:0x%x\n", get_client_memptr(eth_client, i)->v4_addr);
CtList->HandleNeighIpAddrDelEvt(get_client_memptr(eth_client, i)->v4_addr);
}
if (delete_eth_rtrules(i, IPA_IP_v4))
{
IPACMERR("unbale to delete ecm-client v4 route rules for index %d\n", i);
res = IPACM_FAILURE;
}
if (delete_eth_rtrules(i, IPA_IP_v6))
{
IPACMERR("unbale to delete ecm-client v6 route rules for index %d\n", i);
res = IPACM_FAILURE;
}
IPACMDBG_H("Delete %d client header\n", num_eth_client);
if(get_client_memptr(eth_client, i)->ipv4_header_set == true)
{
if (m_header.DeleteHeaderHdl(get_client_memptr(eth_client, i)->hdr_hdl_v4)
== false)
{
res = IPACM_FAILURE;
}
}
if(get_client_memptr(eth_client, i)->ipv6_header_set == true)
{
if (m_header.DeleteHeaderHdl(get_client_memptr(eth_client, i)->hdr_hdl_v6)
== false)
{
res = IPACM_FAILURE;
}
}
} /* end of for loop */
/* check software routing fl rule hdl */
if (softwarerouting_act == true && rx_prop != NULL)
{
handle_software_routing_disable(false);
}
if (odu_route_rule_v4_hdl != NULL)
{
free(odu_route_rule_v4_hdl);
}
if (odu_route_rule_v6_hdl != NULL)
{
free(odu_route_rule_v6_hdl);
}
if (rx_prop != NULL)
{
if(IPACM_Iface::ipacmcfg->GetIPAVer() >= IPA_HW_None && IPACM_Iface::ipacmcfg->GetIPAVer() < IPA_HW_v4_0)
{
/* Delete corresponding ipa_rm_resource_name of RX-endpoint after delete all IPV4V6 FT-rule */
IPACMDBG_H("dev %s delete producer dependency\n", dev_name);
IPACMDBG_H("depend Got pipe %d rm index : %d \n", rx_prop->rx[0].src_pipe, IPACM_Iface::ipacmcfg->ipa_client_rm_map_tbl[rx_prop->rx[0].src_pipe]);
IPACM_Iface::ipacmcfg->DelRmDepend(IPACM_Iface::ipacmcfg->ipa_client_rm_map_tbl[rx_prop->rx[0].src_pipe]);
IPACMDBG_H("Finished delete dependency \n ");
}
if (!(IPACM_Iface::ipacmcfg->isEthBridgingSupported()))
free(rx_prop);
}
if (eth_client != NULL)
{
free(eth_client);
}
if (!(IPACM_Iface::ipacmcfg->isEthBridgingSupported()))
{
if (tx_prop != NULL)
{
free(tx_prop);
}
if (iface_query != NULL)
{
free(iface_query);
}
}
is_active = false;
post_del_self_evt();
return res;
}
/* install UL filter rule from Q6 */
int IPACM_Lan::handle_uplink_filter_rule(ipacm_ext_prop *prop, ipa_ip_type iptype, uint8_t xlat_mux_id)
{
struct ipa_flt_rule_add_v2 flt_rule_entry;
struct ipa_ioc_add_flt_rule_v2 *pFilteringTable;
int cnt, ret = IPACM_SUCCESS;
ipa_fltr_installed_notif_req_msg_v01 flt_index;
int fd;
int i, index, eq_index;
uint32_t value = 0;
uint8_t qmap_id, xlat_debug;
bool result;
IPACMDBG_H("Set modem UL flt rules\n");
if (rx_prop == NULL)
{
IPACMDBG_H("No rx properties registered for iface %s\n", dev_name);
return IPACM_SUCCESS;
}
if(prop == NULL || prop->num_ext_props <= 0)
{
IPACMDBG_H("No extended property.\n");
return IPACM_SUCCESS;
}
fd = open(IPA_DEVICE_NAME, O_RDWR);
if (0 == fd)
{
IPACMERR("Failed opening %s.\n", IPA_DEVICE_NAME);
return IPACM_FAILURE;
}
if (prop->num_ext_props > MAX_WAN_UL_FILTER_RULES)
{
IPACMERR("number of modem UL rules > MAX_WAN_UL_FILTER_RULES, aborting...\n");
close(fd);
return IPACM_FAILURE;
}
memset(&flt_index, 0, sizeof(flt_index));
flt_index.source_pipe_index = ioctl(fd, IPA_IOC_QUERY_EP_MAPPING, rx_prop->rx[0].src_pipe);
if ((int)flt_index.source_pipe_index == -1)
{
IPACMERR("Error Query src pipe idx, aborting...\n");
close(fd);
return IPACM_FAILURE;
}
flt_index.install_status = IPA_QMI_RESULT_SUCCESS_V01;
if (!IPACM_Iface::ipacmcfg->isIPAv3Supported())
{
flt_index.filter_index_list_len = prop->num_ext_props;
}
else /* IPAv3 */
{
flt_index.rule_id_valid = 1;
flt_index.rule_id_len = prop->num_ext_props;
}
flt_index.embedded_pipe_index_valid = 1;
flt_index.embedded_pipe_index = ioctl(fd, IPA_IOC_QUERY_EP_MAPPING, IPA_CLIENT_APPS_LAN_WAN_PROD);
if ((int)flt_index.embedded_pipe_index == -1)
{
IPACMERR("Error Query emb pipe idx, aborting...\n");
close(fd);
return IPACM_FAILURE;
}
flt_index.retain_header_valid = 1;
flt_index.retain_header = 0;
flt_index.embedded_call_mux_id_valid = 1;
qmap_id = IPACM_Iface::ipacmcfg->GetQmapId();
xlat_debug = IPACM_Wan::getXlat_Mux_Id();
flt_index.embedded_call_mux_id = qmap_id;
if (!IPACM_Iface::ipacmcfg->isIPAv3Supported())
{
IPACMDBG_H("flt_index: src pipe: %d, num of rules: %d, ebd pipe: %d, mux id: %d, xlat_mux id: %d, wan-debug %d\n",
flt_index.source_pipe_index, flt_index.filter_index_list_len, flt_index.embedded_pipe_index, flt_index.embedded_call_mux_id, xlat_mux_id, xlat_debug);
}
else /* IPAv3 */
{
IPACMDBG_H("flt_index: src pipe: %d, num of rules: %d, ebd pipe: %d, mux id: %d, xlat_mux id: %d, wan-debug %d\n",
flt_index.source_pipe_index, flt_index.rule_id_len, flt_index.embedded_pipe_index, flt_index.embedded_call_mux_id, xlat_mux_id, xlat_debug);
}
pFilteringTable = static_cast<decltype(pFilteringTable)>(malloc(sizeof(*pFilteringTable)));
if (!pFilteringTable)
{
IPACMERR("Memory allocation error: pFilteringTable\n");
close(fd);
return IPACM_FAILURE;
}
memset(pFilteringTable, 0, sizeof(*pFilteringTable));
pFilteringTable->rules = reinterpret_cast<decltype(pFilteringTable->rules)>(malloc(sizeof(flt_rule_entry) * prop->num_ext_props));
if (!pFilteringTable->rules)
{
IPACMERR("Memory allocation error: pFilteringTable->rules\n");
free(pFilteringTable);
close(fd);
return IPACM_FAILURE;
}
memset(reinterpret_cast<void *>(pFilteringTable->rules), 0, sizeof(flt_rule_entry) * prop->num_ext_props);
pFilteringTable->commit = 1;
pFilteringTable->ep = rx_prop->rx[0].src_pipe;
pFilteringTable->global = false;
pFilteringTable->ip = iptype;
pFilteringTable->num_rules = prop->num_ext_props;
pFilteringTable->flt_rule_size = sizeof(flt_rule_entry);
memset(&flt_rule_entry, 0, sizeof(flt_rule_entry));
flt_rule_entry.at_rear = 1;
flt_rule_entry.flt_rule_hdl = -1;
flt_rule_entry.status = -1;
flt_rule_entry.rule.retain_hdr = 0;
flt_rule_entry.rule.to_uc = 0;
flt_rule_entry.rule.eq_attrib_type = 1;
if(iptype == IPA_IP_v4)
{
if (ipa_if_cate == ODU_IF && IPACM_Wan::isWan_Bridge_Mode())
{
IPACMDBG_H("WAN, ODU are in bridge mode \n");
flt_rule_entry.rule.action = IPA_PASS_TO_ROUTING;
}
else
{
flt_rule_entry.rule.action = IPA_PASS_TO_SRC_NAT;
}
}
else if(iptype == IPA_IP_v6)
flt_rule_entry.rule.action = IPA_PASS_TO_ROUTING;
else
{
IPACMERR("IP type is not expected.\n");
ret = IPACM_FAILURE;
goto fail;
}
index = IPACM_Iface::ipacmcfg->getFltRuleCount(rx_prop->rx[0].src_pipe, iptype);
for(cnt=0; cnt<prop->num_ext_props; cnt++)
{
memcpy(&flt_rule_entry.rule.eq_attrib,
&prop->prop[cnt].eq_attrib,
sizeof(prop->prop[cnt].eq_attrib));
flt_rule_entry.rule.rt_tbl_idx = prop->prop[cnt].rt_tbl_idx;
if (iptype == IPA_IP_v4)
{
if ((ipa_if_cate != ODU_IF) || (IPACM_Wan::isWan_Bridge_Mode() == false))
{
/* NAT block will set the proper MUX ID in the metadata according to the relevant PDN */
if (IPACM_Iface::ipacmcfg->GetIPAVer() >= IPA_HW_v4_0)
flt_rule_entry.rule.set_metadata = true;
}
}
/* Handle XLAT configuration */
/* temp wa to reset xlat_mux_id to qmap_id if it's xlat call */
if (IPACM_Wan::isXlat() && (iptype == IPA_IP_v4))
{
IPACMDBG_H("WA to replace xlat_mux_id %d with qmap_id: %d\n", xlat_mux_id, qmap_id);
xlat_mux_id = qmap_id;
}
#ifdef IPA_IOCTL_SET_FNR_COUNTER_INFO
if(ipa_if_cate == WLAN_IF && IPACM_Iface::ipacmcfg->hw_fnr_stats_support)
{
flt_rule_entry.rule.cnt_idx = IPACM_Iface::ipacmcfg->hw_counter_offset + UL_ALL;
flt_rule_entry.rule.enable_stats = flt_rule_entry.rule.cnt_idx;
IPACMDBG_H("hw-index-enable %d, counter %d\n",
IPACM_Iface::ipacmcfg->hw_fnr_stats_support, flt_rule_entry.rule.cnt_idx);
}
#endif
#ifdef IPA_FLTRT_TTL_UPDATE
if(IPACM_Iface::ipacmcfg->GetIPAVer() >= IPA_HW_v5_5)
flt_rule_entry.rule.ttl_update = true;
#endif
if ((iptype == IPA_IP_v4) && prop->prop[cnt].is_xlat_rule && (xlat_mux_id != 0))
{
/* In XLAT UL case TTL needs to be done after IPv4->IPv6 transformation */
#ifdef IPA_FLTRT_TTL_UPDATE
if(IPACM_Iface::ipacmcfg->GetIPAVer() >= IPA_HW_v5_5)
flt_rule_entry.rule.ttl_update = true;
#endif
/* fill the value of meta-data */
value = xlat_mux_id;
flt_rule_entry.rule.eq_attrib.metadata_meq32_present = 1;
flt_rule_entry.rule.eq_attrib.metadata_meq32.offset = 0;
flt_rule_entry.rule.eq_attrib.metadata_meq32.value = (value & 0xFF) << 16;
flt_rule_entry.rule.eq_attrib.metadata_meq32.mask = 0x00FF0000;
IPACMDBG_H("xlat meta-data is modified for rule: %d has index %d with xlat_mux_id: %d\n",
cnt, index, xlat_mux_id);
/* disable metadata replacement for xlat rules */
if (IPACM_Iface::ipacmcfg->GetIPAVer() >= IPA_HW_v4_0)
flt_rule_entry.rule.set_metadata = false;
}
#ifdef FEATURE_IPACM_HAL
/* add prefix equation in modem UL rules */
if(iptype == IPA_IP_v4 && (flt_rule_entry.rule.eq_attrib.num_offset_meq_32 >= 0)
&& (flt_rule_entry.rule.eq_attrib.num_offset_meq_32 < IPA_IPFLTR_NUM_MEQ_32_EQNS))
{
flt_rule_entry.rule.eq_attrib.num_offset_meq_32++;
eq_index = flt_rule_entry.rule.eq_attrib.num_offset_meq_32 - 1;
if (IPACM_Iface::ipacmcfg->isIPAv3Supported())
{
if(eq_index == 0)
{
flt_rule_entry.rule.eq_attrib.rule_eq_bitmap |= (1<<5);
}
else
{
flt_rule_entry.rule.eq_attrib.rule_eq_bitmap |= (1<<6);
}
}
else
{
if(eq_index == 0)
{
flt_rule_entry.rule.eq_attrib.rule_eq_bitmap |= (1<<2);
}
else
{
flt_rule_entry.rule.eq_attrib.rule_eq_bitmap |= (1<<3);
}
}
flt_rule_entry.rule.eq_attrib.offset_meq_32[eq_index].offset = 12;
flt_rule_entry.rule.eq_attrib.offset_meq_32[eq_index].mask = prefix[IPA_IP_v4].v4Mask;
flt_rule_entry.rule.eq_attrib.offset_meq_32[eq_index].value = prefix[IPA_IP_v4].v4Addr;
}
else if (flt_rule_entry.rule.eq_attrib.num_offset_meq_32 > IPA_IPFLTR_NUM_MEQ_32_EQNS)
{
IPACMERR("Run out of MEQ32 equation.\n");
flt_rule_entry.rule.eq_attrib.num_offset_meq_32--;
}
else
{
if ((flt_rule_entry.rule.eq_attrib.num_offset_meq_128 >= 0) &&
(flt_rule_entry.rule.eq_attrib.num_offset_meq_128
< IPA_IPFLTR_NUM_MEQ_128_EQNS))
{
flt_rule_entry.rule.eq_attrib.num_offset_meq_128++;
eq_index = flt_rule_entry.rule.eq_attrib.num_offset_meq_128 - 1;
if (IPACM_Iface::ipacmcfg->isIPAv3Supported())
{
if(eq_index == 0)
{
flt_rule_entry.rule.eq_attrib.rule_eq_bitmap |= (1<<3);
}
else
{
flt_rule_entry.rule.eq_attrib.rule_eq_bitmap |= (1<<4);
}
*(uint32_t *)(flt_rule_entry.rule.eq_attrib.offset_meq_128[eq_index].mask + 0)
= prefix[IPA_IP_v6].v6Mask[3];
*(uint32_t *)(flt_rule_entry.rule.eq_attrib.offset_meq_128[eq_index].mask + 4)
= prefix[IPA_IP_v6].v6Mask[2];
*(uint32_t *)(flt_rule_entry.rule.eq_attrib.offset_meq_128[eq_index].mask + 8)
= prefix[IPA_IP_v6].v6Mask[1];
*(uint32_t *)(flt_rule_entry.rule.eq_attrib.offset_meq_128[eq_index].mask + 12)
= prefix[IPA_IP_v6].v6Mask[0];
*(uint32_t *)(flt_rule_entry.rule.eq_attrib.offset_meq_128[eq_index].value + 0)
= prefix[IPA_IP_v6].v6Addr[3];
*(uint32_t *)(flt_rule_entry.rule.eq_attrib.offset_meq_128[eq_index].value + 4)
= prefix[IPA_IP_v6].v6Addr[2];
*(uint32_t *)(flt_rule_entry.rule.eq_attrib.offset_meq_128[eq_index].value + 8)
= prefix[IPA_IP_v6].v6Addr[1];
*(uint32_t *)(flt_rule_entry.rule.eq_attrib.offset_meq_128[eq_index].value + 12)
= prefix[IPA_IP_v6].v6Addr[0];
}
else
{
if(eq_index == 0)
{
flt_rule_entry.rule.eq_attrib.rule_eq_bitmap |= (1<<9);
}
else
{
flt_rule_entry.rule.eq_attrib.rule_eq_bitmap |= (1<<10);
}
*(uint32_t *)(flt_rule_entry.rule.eq_attrib.offset_meq_128[eq_index].mask + 0)
= prefix[IPA_IP_v6].v6Mask[0];
*(uint32_t *)(flt_rule_entry.rule.eq_attrib.offset_meq_128[eq_index].mask + 4)
= prefix[IPA_IP_v6].v6Mask[1];
*(uint32_t *)(flt_rule_entry.rule.eq_attrib.offset_meq_128[eq_index].mask + 8)
= prefix[IPA_IP_v6].v6Mask[2];
*(uint32_t *)(flt_rule_entry.rule.eq_attrib.offset_meq_128[eq_index].mask + 12)
= prefix[IPA_IP_v6].v6Mask[3];
*(uint32_t *)(flt_rule_entry.rule.eq_attrib.offset_meq_128[eq_index].value + 0)
= prefix[IPA_IP_v6].v6Addr[0];
*(uint32_t *)(flt_rule_entry.rule.eq_attrib.offset_meq_128[eq_index].value + 4)
= prefix[IPA_IP_v6].v6Addr[1];
*(uint32_t *)(flt_rule_entry.rule.eq_attrib.offset_meq_128[eq_index].value + 8)
= prefix[IPA_IP_v6].v6Addr[2];
*(uint32_t *)(flt_rule_entry.rule.eq_attrib.offset_meq_128[eq_index].value + 12)
= prefix[IPA_IP_v6].v6Addr[3];
}
flt_rule_entry.rule.eq_attrib.offset_meq_128[eq_index].offset = 8;
}
else
{
IPACMERR("Run out of MEQ128 equation.\n");
flt_rule_entry.rule.eq_attrib.num_offset_meq_128--;
}
}
#endif
if (IPACM_Iface::ipacmcfg->isIPAv3Supported())
{
flt_rule_entry.rule.hashable = prop->prop[cnt].is_rule_hashable;
flt_rule_entry.rule.rule_id = prop->prop[cnt].rule_id;
if(rx_prop->rx[0].attrib.attrib_mask & IPA_FLT_META_DATA) //turn on meta-data equation
{
flt_rule_entry.rule.eq_attrib.rule_eq_bitmap |= (1<<9);
flt_rule_entry.rule.eq_attrib.metadata_meq32_present = 1;
flt_rule_entry.rule.eq_attrib.metadata_meq32.offset = 0;
flt_rule_entry.rule.eq_attrib.metadata_meq32.value |= rx_prop->rx[0].attrib.meta_data;
flt_rule_entry.rule.eq_attrib.metadata_meq32.mask |= rx_prop->rx[0].attrib.meta_data_mask;
}
}
memcpy(&((reinterpret_cast<decltype(&flt_rule_entry)>(pFilteringTable->rules))[cnt]),
&flt_rule_entry, sizeof(flt_rule_entry));
IPACMDBG_H("Modem UL filtering rule %d has index %d\n", cnt, index);
if (!IPACM_Iface::ipacmcfg->isIPAv3Supported())
{
flt_index.filter_index_list[cnt].filter_index = index;
flt_index.filter_index_list[cnt].filter_handle = prop->prop[cnt].filter_hdl;
}
else /* IPAv3 */
{
flt_index.rule_id[cnt] = prop->prop[cnt].rule_id;
}
index++;
}
if(false == m_filtering.SendFilteringRuleIndex(&flt_index))
{
IPACMERR("Error sending filtering rule index, aborting...\n");
ret = IPACM_FAILURE;
goto fail;
}
result = m_filtering.AddFilteringRule_v2(pFilteringTable);
if(result == false)
{
IPACMERR("Error Adding RuleTable to Filtering, aborting...\n");
ret = IPACM_FAILURE;
goto fail;
}
else
{
if(iptype == IPA_IP_v4)
{
for(i=0; i<pFilteringTable->num_rules; i++)
{
wan_ul_fl_rule_hdl_v4[num_wan_ul_fl_rule_v4] =
reinterpret_cast<decltype(&flt_rule_entry)>(pFilteringTable->rules)[i].flt_rule_hdl;
num_wan_ul_fl_rule_v4++;
}
IPACM_Iface::ipacmcfg->increaseFltRuleCount(rx_prop->rx[0].src_pipe, iptype, pFilteringTable->num_rules);
v4_mux_id = qmap_id;
}
else if(iptype == IPA_IP_v6)
{
for(i=0; i<pFilteringTable->num_rules; i++)
{
wan_ul_fl_rule_hdl_v6[num_wan_ul_fl_rule_v6] =
reinterpret_cast<decltype(&flt_rule_entry)>(pFilteringTable->rules)[i].flt_rule_hdl;
num_wan_ul_fl_rule_v6++;
}
IPACM_Iface::ipacmcfg->increaseFltRuleCount(rx_prop->rx[0].src_pipe, iptype, pFilteringTable->num_rules);
v6_mux_id = qmap_id;
}
else
{
IPACMERR("IP type is not expected.\n");
goto fail;
}
}
fail:
free(reinterpret_cast<void *>(pFilteringTable->rules));
free(pFilteringTable);
close(fd);
return ret;
}
int IPACM_Lan::handle_wan_down_v6(ipacm_wan_iface_type backhaul_mode)
{
ipa_fltr_installed_notif_req_msg_v01 flt_index;
int fd;
fd = open(IPA_DEVICE_NAME, O_RDWR);
if (0 == fd)
{
IPACMERR("Failed opening %s.\n", IPA_DEVICE_NAME);
return IPACM_FAILURE;
}
delete_ipv6_prefix_flt_rule();
memset(ipv6_prefix, 0, sizeof(ipv6_prefix));
if(backhaul_mode == Q6_WAN && modem_ul_v6_set == true)
{
if (num_wan_ul_fl_rule_v6 > MAX_WAN_UL_FILTER_RULES)
{
IPACMERR(" the number of rules (%d) are bigger than array (%d), aborting...\n", num_wan_ul_fl_rule_v6, MAX_WAN_UL_FILTER_RULES);
close(fd);
return IPACM_FAILURE;
}
if (num_wan_ul_fl_rule_v6 == 0)
{
IPACMERR("No modem UL rules were installed, return...\n");
close(fd);
return IPACM_FAILURE;
}
if (m_filtering.DeleteFilteringHdls(wan_ul_fl_rule_hdl_v6,
IPA_IP_v6, num_wan_ul_fl_rule_v6) == false)
{
IPACMERR("Error Deleting RuleTable(1) to Filtering, aborting...\n");
close(fd);
return IPACM_FAILURE;
}
IPACM_Iface::ipacmcfg->decreaseFltRuleCount(rx_prop->rx[0].src_pipe, IPA_IP_v6, num_wan_ul_fl_rule_v6);
memset(wan_ul_fl_rule_hdl_v6, 0, MAX_WAN_UL_FILTER_RULES * sizeof(uint32_t));
num_wan_ul_fl_rule_v6 = 0;
modem_ul_v6_set = false;
memset(&flt_index, 0, sizeof(flt_index));
flt_index.source_pipe_index = ioctl(fd, IPA_IOC_QUERY_EP_MAPPING, rx_prop->rx[0].src_pipe);
if ((int)flt_index.source_pipe_index == -1)
{
IPACMERR("Error Query src pipe idx, aborting...\n");
close(fd);
return IPACM_FAILURE;
}
flt_index.install_status = IPA_QMI_RESULT_SUCCESS_V01;
if (!IPACM_Iface::ipacmcfg->isIPAv3Supported())
{
flt_index.filter_index_list_len = 0;
}
else /* IPAv3 */
{
flt_index.rule_id_valid = 1;
flt_index.rule_id_len = 0;
}
flt_index.embedded_pipe_index_valid = 1;
flt_index.embedded_pipe_index = ioctl(fd, IPA_IOC_QUERY_EP_MAPPING, IPA_CLIENT_APPS_LAN_WAN_PROD);
if ((int)flt_index.embedded_pipe_index == -1)
{
IPACMERR("Error Query emb pipe idx, aborting...\n");
close(fd);
return IPACM_FAILURE;
}
flt_index.retain_header_valid = 1;
flt_index.retain_header = 0;
flt_index.embedded_call_mux_id_valid = 1;
flt_index.embedded_call_mux_id = v6_mux_id;
v6_mux_id = 0;
if(false == m_filtering.SendFilteringRuleIndex(&flt_index))
{
IPACMERR("Error sending filtering rule index, aborting...\n");
close(fd);
return IPACM_FAILURE;
}
}
else
{
if (m_filtering.DeleteFilteringHdls(&dft_v6fl_rule_hdl[IPV6_DEFAULT_FILTERTING_RULES],
IPA_IP_v6, 1) == false)
{
IPACMERR("Error Adding RuleTable(1) to Filtering, aborting...\n");
close(fd);
return IPACM_FAILURE;
}
IPACM_Iface::ipacmcfg->decreaseFltRuleCount(rx_prop->rx[0].src_pipe, IPA_IP_v6, 1);
sta_ul_v6_set = false;
}
close(fd);
return IPACM_SUCCESS;
}
int IPACM_Lan::reset_to_dummy_flt_rule(ipa_ip_type iptype, uint32_t rule_hdl)
{
int len, res = IPACM_SUCCESS;
struct ipa_flt_rule_mdfy flt_rule;
struct ipa_ioc_mdfy_flt_rule* pFilteringTable;
IPACMDBG_H("Reset flt rule to dummy, IP type: %d, hdl: %d\n", iptype, rule_hdl);
len = sizeof(struct ipa_ioc_mdfy_flt_rule) + sizeof(struct ipa_flt_rule_mdfy);
pFilteringTable = (struct ipa_ioc_mdfy_flt_rule*)malloc(len);
if (pFilteringTable == NULL)
{
IPACMERR("Error allocate flt rule memory...\n");
return IPACM_FAILURE;
}
memset(pFilteringTable, 0, len);
pFilteringTable->commit = 1;
pFilteringTable->ip = iptype;
pFilteringTable->num_rules = 1;
memset(&flt_rule, 0, sizeof(struct ipa_flt_rule_mdfy));
flt_rule.status = -1;
flt_rule.rule_hdl = rule_hdl;
flt_rule.rule.retain_hdr = 0;
flt_rule.rule.action = IPA_PASS_TO_EXCEPTION;
if(iptype == IPA_IP_v4)
{
IPACMDBG_H("Reset IPv4 flt rule to dummy\n");
flt_rule.rule.attrib.attrib_mask = IPA_FLT_SRC_ADDR | IPA_FLT_DST_ADDR;
flt_rule.rule.attrib.u.v4.dst_addr = ~0;
flt_rule.rule.attrib.u.v4.dst_addr_mask = ~0;
flt_rule.rule.attrib.u.v4.src_addr = ~0;
flt_rule.rule.attrib.u.v4.src_addr_mask = ~0;
memcpy(&(pFilteringTable->rules[0]), &flt_rule, sizeof(struct ipa_flt_rule_mdfy));
if (false == m_filtering.ModifyFilteringRule(pFilteringTable))
{
IPACMERR("Error modifying filtering rule.\n");
res = IPACM_FAILURE;
goto fail;
}
else
{
IPACMDBG_H("Flt rule reset to dummy, hdl: 0x%x, status: %d\n", pFilteringTable->rules[0].rule_hdl,
pFilteringTable->rules[0].status);
}
}
else if(iptype == IPA_IP_v6)
{
IPACMDBG_H("Reset IPv6 flt rule to dummy\n");
flt_rule.rule.attrib.attrib_mask = IPA_FLT_SRC_ADDR | IPA_FLT_DST_ADDR;
flt_rule.rule.attrib.u.v6.src_addr[0] = ~0;
flt_rule.rule.attrib.u.v6.src_addr[1] = ~0;
flt_rule.rule.attrib.u.v6.src_addr[2] = ~0;
flt_rule.rule.attrib.u.v6.src_addr[3] = ~0;
flt_rule.rule.attrib.u.v6.src_addr_mask[0] = ~0;
flt_rule.rule.attrib.u.v6.src_addr_mask[1] = ~0;
flt_rule.rule.attrib.u.v6.src_addr_mask[2] = ~0;
flt_rule.rule.attrib.u.v6.src_addr_mask[3] = ~0;
flt_rule.rule.attrib.u.v6.dst_addr[0] = ~0;
flt_rule.rule.attrib.u.v6.dst_addr[1] = ~0;
flt_rule.rule.attrib.u.v6.dst_addr[2] = ~0;
flt_rule.rule.attrib.u.v6.dst_addr[3] = ~0;
flt_rule.rule.attrib.u.v6.dst_addr_mask[0] = ~0;
flt_rule.rule.attrib.u.v6.dst_addr_mask[1] = ~0;
flt_rule.rule.attrib.u.v6.dst_addr_mask[2] = ~0;
flt_rule.rule.attrib.u.v6.dst_addr_mask[3] = ~0;
memcpy(&(pFilteringTable->rules[0]), &flt_rule, sizeof(struct ipa_flt_rule_mdfy));
if (false == m_filtering.ModifyFilteringRule(pFilteringTable))
{
IPACMERR("Error modifying filtering rule.\n");
res = IPACM_FAILURE;
goto fail;
}
else
{
IPACMDBG_H("Flt rule reset to dummy, hdl: 0x%x, status: %d\n", pFilteringTable->rules[0].rule_hdl,
pFilteringTable->rules[0].status);
}
}
else
{
IPACMERR("IP type is not expected.\n");
res = IPACM_FAILURE;
goto fail;
}
fail:
free(pFilteringTable);
return res;
}
void IPACM_Lan::post_del_self_evt()
{
ipacm_cmd_q_data evt;
ipacm_event_data_fid* fid;
fid = (ipacm_event_data_fid*)malloc(sizeof(ipacm_event_data_fid));
if(fid == NULL)
{
IPACMERR("Failed to allocate fid memory.\n");
return;
}
memset(fid, 0, sizeof(ipacm_event_data_fid));
memset(&evt, 0, sizeof(ipacm_cmd_q_data));
fid->if_index = ipa_if_num;
evt.evt_data = (void*)fid;
evt.event = IPA_LAN_DELETE_SELF;
IPACMDBG_H("Posting event IPA_LAN_DELETE_SELF\n");
IPACM_EvtDispatcher::PostEvt(&evt);
}
/*handle reset usb-client rt-rules */
int IPACM_Lan::handle_lan_client_reset_rt(ipa_ip_type iptype)
{
uint32_t i;
int res = IPACM_SUCCESS;
/* clean eth-client routing rules */
IPACMDBG_H("left %d eth clients need to be deleted \n ", num_eth_client);
for (i = 0; i < num_eth_client; i++)
{
res = delete_eth_rtrules(i, iptype);
if (res != IPACM_SUCCESS)
{
IPACMERR("Failed to delete old iptype(%d) rules.\n", iptype);
return res;
}
} /* end of for loop */
/* Reset ip-address */
for (i = 0; i < num_eth_client; i++)
{
if(iptype == IPA_IP_v4)
{
get_client_memptr(eth_client, i)->ipv4_set = false;
}
else
{
get_client_memptr(eth_client, i)->ipv6_set = 0;
}
} /* end of for loop */
return res;
}
int IPACM_Lan::install_ipv4_icmp_flt_rule()
{
int ret = IPACM_SUCCESS;
struct ipa_ioc_add_flt_rule_v2 *flt_rule;
struct ipa_flt_rule_add_v2 *flt_rule_entry;
if(rx_prop != NULL)
{
flt_rule = (struct ipa_ioc_add_flt_rule_v2 *)calloc(1,
sizeof(struct ipa_ioc_add_flt_rule_v2));
if (!flt_rule)
{
IPACMERR("Error Locate ipa_ioc_add_flt_rule_v2 memory...\n");
return IPACM_FAILURE;
}
flt_rule_entry = (struct ipa_flt_rule_add_v2 *)calloc(1, sizeof(struct ipa_flt_rule_add_v2));
if (!flt_rule_entry)
{
IPACMERR("Failed to allocate ipa_flt_rule_add_v2 memory...\n");
free(flt_rule);
return IPACM_FAILURE;
}
flt_rule->rules = (uint64_t)flt_rule_entry;
flt_rule->commit = 1;
flt_rule->ep = rx_prop->rx[0].src_pipe;
flt_rule->global = false;
flt_rule->ip = IPA_IP_v4;
flt_rule->num_rules = 1;
flt_rule->flt_rule_size = sizeof(struct ipa_flt_rule_add_v2);
flt_rule_entry->rule.retain_hdr = 1;
flt_rule_entry->rule.to_uc = 0;
flt_rule_entry->rule.eq_attrib_type = 0;
flt_rule_entry->at_rear = true;
flt_rule_entry->flt_rule_hdl = -1;
flt_rule_entry->status = -1;
flt_rule_entry->rule.action = IPA_PASS_TO_EXCEPTION;
if (IPACM_Iface::ipacmcfg->isIPAv3Supported())
flt_rule_entry->rule.hashable = true;
flt_rule_entry->rule.close_aggr_irq_mod = true;
memcpy(&flt_rule_entry->rule.attrib,
&rx_prop->rx[0].attrib,
sizeof(flt_rule_entry->rule.attrib));
flt_rule_entry->rule.attrib.attrib_mask |= IPA_FLT_PROTOCOL;
flt_rule_entry->rule.attrib.u.v4.protocol = (uint8_t)IPACM_FIREWALL_IPPROTO_ICMP;
#ifdef IPA_IOCTL_SET_FNR_COUNTER_INFO
/* use index hw-counter */
if(ipa_if_cate == WLAN_IF && IPACM_Iface::ipacmcfg->hw_fnr_stats_support)
{
IPACMDBG_H("hw-index-enable %d, counter %d\n",
IPACM_Iface::ipacmcfg->hw_fnr_stats_support,
IPACM_Iface::ipacmcfg->hw_counter_offset + UL_ALL);
flt_rule_entry->rule.enable_stats = 1;
flt_rule_entry->rule.cnt_idx = IPACM_Iface::ipacmcfg->hw_counter_offset + UL_ALL;
}
#endif
if (m_filtering.AddFilteringRule_v2(flt_rule) == false)
{
IPACMERR("Error Adding Filtering rule, aborting...\n");
ret = IPACM_FAILURE;
}
else
{
IPACM_Iface::ipacmcfg->increaseFltRuleCount(rx_prop->rx[0].src_pipe, IPA_IP_v4, 1);
ipv4_icmp_flt_rule_hdl[0] = flt_rule_entry->flt_rule_hdl;
IPACMDBG_H("IPv4 icmp filter rule HDL:0x%x\n", ipv4_icmp_flt_rule_hdl[0]);
}
free(flt_rule_entry);
free(flt_rule);
}
return ret;
}
int IPACM_Lan::install_ipv6_icmp_flt_rule()
{
int ret = IPACM_SUCCESS;
struct ipa_ioc_add_flt_rule_v2 *flt_rule;
struct ipa_flt_rule_add_v2 *flt_rule_entry;
if(rx_prop != NULL)
{
flt_rule = (struct ipa_ioc_add_flt_rule_v2 *)calloc(1, sizeof(struct ipa_ioc_add_flt_rule_v2));
if (!flt_rule)
{
IPACMERR("Error Locate ipa_ioc_add_flt_rule_v2 memory...\n");
return IPACM_FAILURE;
}
flt_rule_entry = (struct ipa_flt_rule_add_v2 *)calloc(1, sizeof(struct ipa_flt_rule_add_v2));
if (!flt_rule_entry)
{
IPACMERR("Failed to allocate ipa_flt_rule_add_v2 memory...\n");
free(flt_rule);
return IPACM_FAILURE;
}
flt_rule->rules = (uint64_t)flt_rule_entry;
flt_rule->commit = 1;
flt_rule->ep = rx_prop->rx[0].src_pipe;
flt_rule->global = false;
flt_rule->ip = IPA_IP_v6;
flt_rule->num_rules = 1;
flt_rule->flt_rule_size = sizeof(struct ipa_flt_rule_add_v2);
flt_rule_entry->rule.retain_hdr = 1;
flt_rule_entry->rule.to_uc = 0;
flt_rule_entry->rule.eq_attrib_type = 0;
flt_rule_entry->at_rear = true;
flt_rule_entry->flt_rule_hdl = -1;
flt_rule_entry->status = -1;
flt_rule_entry->rule.action = IPA_PASS_TO_EXCEPTION;
if (IPACM_Iface::ipacmcfg->isIPAv3Supported())
flt_rule_entry->rule.hashable = false;
flt_rule_entry->rule.close_aggr_irq_mod = true;
memcpy(&flt_rule_entry->rule.attrib, &rx_prop->rx[0].attrib, sizeof(flt_rule_entry->rule.attrib));
flt_rule_entry->rule.attrib.attrib_mask |= IPA_FLT_NEXT_HDR;
flt_rule_entry->rule.attrib.u.v6.next_hdr = (uint8_t)IPACM_FIREWALL_IPPROTO_ICMP6;
#ifdef IPA_IOCTL_SET_FNR_COUNTER_INFO
/* use index hw-counter */
if(ipa_if_cate == WLAN_IF && IPACM_Iface::ipacmcfg->hw_fnr_stats_support)
{
IPACMDBG_H("hw-index-enable %d, counter %d\n",
IPACM_Iface::ipacmcfg->hw_fnr_stats_support,
IPACM_Iface::ipacmcfg->hw_counter_offset + UL_ALL);
flt_rule_entry->rule.enable_stats = 1;
flt_rule_entry->rule.cnt_idx = IPACM_Iface::ipacmcfg->hw_counter_offset + UL_ALL;
}
#endif
if (m_filtering.AddFilteringRule_v2(flt_rule) == false)
{
IPACMERR("Error Adding Filtering rule, aborting...\n");
ret = IPACM_FAILURE;
}
else
{
IPACM_Iface::ipacmcfg->increaseFltRuleCount(rx_prop->rx[0].src_pipe, IPA_IP_v6, 1);
ipv6_icmp_flt_rule_hdl[0] = flt_rule_entry->flt_rule_hdl;
IPACMDBG_H("IPv6 icmp filter rule HDL:0x%x\n", ipv6_icmp_flt_rule_hdl[0]);
}
free(flt_rule_entry);
free(flt_rule);
}
return ret;
}
int IPACM_Lan::add_dummy_private_subnet_flt_rule(ipa_ip_type iptype)
{
if(rx_prop == NULL)
{
IPACMDBG_H("There is no rx_prop for iface %s, not able to add dummy private subnet filtering rule.\n", dev_name);
return 0;
}
if(iptype == IPA_IP_v6)
{
IPACMDBG_H("There is no ipv6 dummy filter rules needed for iface %s\n", dev_name);
return 0;
}
int i, len, res = IPACM_SUCCESS;
struct ipa_flt_rule_add flt_rule;
ipa_ioc_add_flt_rule* pFilteringTable;
bool result;
len = sizeof(struct ipa_ioc_add_flt_rule) + (IPA_MAX_PRIVATE_SUBNET_ENTRIES + IPA_MAX_MTU_ENTRIES) * sizeof(struct ipa_flt_rule_add);
pFilteringTable = (struct ipa_ioc_add_flt_rule *)malloc(len);
if (pFilteringTable == NULL)
{
IPACMERR("Error allocate flt table memory...\n");
return IPACM_FAILURE;
}
memset(pFilteringTable, 0, len);
pFilteringTable->commit = 1;
pFilteringTable->ep = rx_prop->rx[0].src_pipe;
pFilteringTable->global = false;
pFilteringTable->ip = iptype;
pFilteringTable->num_rules = IPA_MAX_PRIVATE_SUBNET_ENTRIES + IPA_MAX_MTU_ENTRIES;
memset(&flt_rule, 0, sizeof(struct ipa_flt_rule_add));
flt_rule.rule.retain_hdr = 0;
flt_rule.at_rear = true;
flt_rule.flt_rule_hdl = -1;
flt_rule.status = -1;
flt_rule.rule.action = IPA_PASS_TO_EXCEPTION;
if (IPACM_Iface::ipacmcfg->isIPAv3Supported())
flt_rule.rule.hashable = true;
memcpy(&flt_rule.rule.attrib, &rx_prop->rx[0].attrib,
sizeof(flt_rule.rule.attrib));
if(iptype == IPA_IP_v4)
{
flt_rule.rule.attrib.attrib_mask = IPA_FLT_SRC_ADDR | IPA_FLT_DST_ADDR;
flt_rule.rule.attrib.u.v4.src_addr_mask = ~0;
flt_rule.rule.attrib.u.v4.src_addr = ~0;
flt_rule.rule.attrib.u.v4.dst_addr_mask = ~0;
flt_rule.rule.attrib.u.v4.dst_addr = ~0;
for(i=0; i<IPA_MAX_PRIVATE_SUBNET_ENTRIES + IPA_MAX_MTU_ENTRIES; i++)
{
memcpy(&(pFilteringTable->rules[i]), &flt_rule, sizeof(struct ipa_flt_rule_add));
}
#ifdef IPA_IOCTL_SET_FNR_COUNTER_INFO
/* use index hw-counter */
if(ipa_if_cate == WLAN_IF && IPACM_Iface::ipacmcfg->hw_fnr_stats_support)
{
IPACMDBG_H("hw-index-enable %d, counter %d\n", IPACM_Iface::ipacmcfg->hw_fnr_stats_support, IPACM_Iface::ipacmcfg->hw_counter_offset + UL_ALL);
result = m_filtering.AddFilteringRule_hw_index(pFilteringTable, IPACM_Iface::ipacmcfg->hw_counter_offset + UL_ALL);
} else {
result = m_filtering.AddFilteringRule(pFilteringTable);
}
#else
result = m_filtering.AddFilteringRule(pFilteringTable);
#endif
if (result == false)
{
IPACMERR("Error adding dummy private subnet v4 flt rule\n");
res = IPACM_FAILURE;
goto fail;
}
else
{
IPACM_Iface::ipacmcfg->increaseFltRuleCount(rx_prop->rx[0].src_pipe, IPA_IP_v4, IPA_MAX_PRIVATE_SUBNET_ENTRIES + IPA_MAX_MTU_ENTRIES);
/* copy filter rule hdls */
for (int i = 0; i < IPA_MAX_PRIVATE_SUBNET_ENTRIES + IPA_MAX_MTU_ENTRIES; i++)
{
if (pFilteringTable->rules[i].status == 0)
{
private_fl_rule_hdl[i] = pFilteringTable->rules[i].flt_rule_hdl;
IPACMDBG_H("Private subnet v4 flt rule %d hdl:0x%x\n", i, private_fl_rule_hdl[i]);
}
else
{
IPACMERR("Failed adding lan2lan v4 flt rule %d\n", i);
res = IPACM_FAILURE;
goto fail;
}
}
}
}
fail:
free(pFilteringTable);
return res;
}
int IPACM_Lan::add_dummy_filter_cfg_rules(ipa_ip_type iptype)
{
int i, len, res = IPACM_SUCCESS;
struct ipa_flt_rule_add flt_rule;
ipa_ioc_add_flt_rule* pFilteringTable;
bool result;
if(rx_prop == NULL)
{
IPACMDBG_H("There is no rx_prop for iface %s, not able to add dummy filter cfg rules.\n", dev_name);
return 0;
}
if(iptype == IPA_IP_v6)
{
IPACMDBG_H("There is no ipv6 dummy filter config rules needed for iface %s\n", dev_name);
return 0;
}
len = sizeof(struct ipa_ioc_add_flt_rule) + IPA_MAX_FILTER_CFG_ENTRIES * sizeof(struct ipa_flt_rule_add);
pFilteringTable = (struct ipa_ioc_add_flt_rule *)malloc(len);
if (pFilteringTable == NULL)
{
IPACMERR("Error allocate flt table memory...\n");
return IPACM_FAILURE;
}
memset(pFilteringTable, 0, len);
pFilteringTable->commit = 1;
pFilteringTable->ep = rx_prop->rx[0].src_pipe;
pFilteringTable->global = false;
pFilteringTable->ip = iptype;
pFilteringTable->num_rules = IPA_MAX_FILTER_CFG_ENTRIES;
memset(&flt_rule, 0, sizeof(struct ipa_flt_rule_add));
flt_rule.rule.retain_hdr = 0;
flt_rule.at_rear = true;
flt_rule.flt_rule_hdl = -1;
flt_rule.status = -1;
flt_rule.rule.action = IPA_PASS_TO_EXCEPTION;
if (IPACM_Iface::ipacmcfg->isIPAv3Supported())
flt_rule.rule.hashable = true;
memcpy(&flt_rule.rule.attrib, &rx_prop->rx[0].attrib,
sizeof(flt_rule.rule.attrib));
if(iptype == IPA_IP_v4)
{
flt_rule.rule.attrib.attrib_mask = IPA_FLT_SRC_ADDR | IPA_FLT_DST_ADDR;
flt_rule.rule.attrib.u.v4.src_addr_mask = ~0;
flt_rule.rule.attrib.u.v4.src_addr = ~0;
flt_rule.rule.attrib.u.v4.dst_addr_mask = ~0;
flt_rule.rule.attrib.u.v4.dst_addr = ~0;
for(i=0; i<IPA_MAX_FILTER_CFG_ENTRIES; i++)
{
memcpy(&(pFilteringTable->rules[i]), &flt_rule, sizeof(struct ipa_flt_rule_add));
}
#ifdef IPA_IOCTL_SET_FNR_COUNTER_INFO
/* use index hw-counter */
if(ipa_if_cate == WLAN_IF && IPACM_Iface::ipacmcfg->hw_fnr_stats_support)
{
IPACMDBG_H("hw-index-enable %d, counter %d\n", IPACM_Iface::ipacmcfg->hw_fnr_stats_support, IPACM_Iface::ipacmcfg->hw_counter_offset + UL_ALL);
result = m_filtering.AddFilteringRule_hw_index(pFilteringTable, IPACM_Iface::ipacmcfg->hw_counter_offset + UL_ALL);
} else {
result = m_filtering.AddFilteringRule(pFilteringTable);
}
#else
result = m_filtering.AddFilteringRule(pFilteringTable);
#endif
if (result == false)
{
IPACMERR("Error adding dummy filter cfg rules\n");
res = IPACM_FAILURE;
goto fail;
}
else
{
IPACM_Iface::ipacmcfg->increaseFltRuleCount(rx_prop->rx[0].src_pipe, IPA_IP_v4, IPA_MAX_FILTER_CFG_ENTRIES);
/* copy filter rule hdls */
for (int i = 0; i < IPA_MAX_FILTER_CFG_ENTRIES; i++)
{
if (pFilteringTable->rules[i].status == 0)
{
filter_cfg_rule_hdl[i] = pFilteringTable->rules[i].flt_rule_hdl;
IPACMDBG_H("Filter cfg rule %d hdl:0x%x\n", i, private_fl_rule_hdl[i]);
}
else
{
IPACMERR("Failed adding filter cfg rule %d\n", i);
res = IPACM_FAILURE;
goto fail;
}
}
}
}
fail:
free(pFilteringTable);
return res;
}
int IPACM_Lan::handle_filter_cfg_update(ipa_ip_type iptype)
{
int i, len, res = IPACM_SUCCESS;
struct ipa_flt_rule_mdfy flt_rule;
struct ipa_ioc_mdfy_flt_rule* pFilteringTable = NULL;
if (rx_prop == NULL)
{
IPACMDBG_H("No rx properties registered for iface %s\n", dev_name);
return IPACM_SUCCESS;
}
if(iptype == IPA_IP_v6)
{
IPACMDBG_H("Ipv6 Filter config is not supported for iface %s\n", dev_name);
return 0;
}
for(i=0; i<IPA_MAX_FILTER_CFG_ENTRIES; i++)
{
reset_to_dummy_flt_rule(IPA_IP_v4, filter_cfg_rule_hdl[i]);
}
if(IPACM_Iface::ipacmcfg->filter_config.filter_enable == true)
{
if ((IPACM_Iface::ipacmcfg->filter_config.num_filter_cfg_entries == 0) ||
(IPACM_Iface::ipacmcfg->filter_config.num_filter_cfg_entries > IPA_MAX_FILTER_CFG_ENTRIES))
{
IPACMERR("Invalid filter Config num rules %d\n",
IPACM_Iface::ipacmcfg->filter_config.num_filter_cfg_entries);
return IPACM_FAILURE;
}
IPACMDBG_H("total %d filter cfg rules are needed\n", IPACM_Iface::ipacmcfg->filter_config.num_filter_cfg_entries);
len = sizeof(struct ipa_ioc_mdfy_flt_rule) + IPACM_Iface::ipacmcfg->filter_config.num_filter_cfg_entries * sizeof(struct ipa_flt_rule_mdfy);
pFilteringTable = (struct ipa_ioc_mdfy_flt_rule*)malloc(len);
if (!pFilteringTable)
{
IPACMERR("Failed to allocate ipa_ioc_mdfy_flt_rule memory...\n");
return IPACM_FAILURE;
}
memset(pFilteringTable, 0, len);
pFilteringTable->commit = 1;
pFilteringTable->ip = iptype;
pFilteringTable->num_rules = (uint8_t)IPACM_Iface::ipacmcfg->filter_config.num_filter_cfg_entries;
for (i = 0; i < IPACM_Iface::ipacmcfg->filter_config.num_filter_cfg_entries; i++)
{
/* add filter cfg for ipv4 */
memset(&flt_rule, 0, sizeof(struct ipa_flt_rule_mdfy));
flt_rule.status = -1;
flt_rule.rule.retain_hdr = 1;
flt_rule.rule.to_uc = 0;
flt_rule.rule.action = IPA_PASS_TO_EXCEPTION;
flt_rule.rule.eq_attrib_type = 0;
if (IPACM_Iface::ipacmcfg->isIPAv3Supported())
flt_rule.rule.hashable = true;
flt_rule.rule_hdl = filter_cfg_rule_hdl[i];
memcpy(&flt_rule.rule.attrib,
&IPACM_Iface::ipacmcfg->filter_config.filter_cfg_entries[i].attrib,
sizeof(struct ipa_rule_attrib));
flt_rule.rule.attrib.attrib_mask |= rx_prop->rx[0].attrib.attrib_mask;
memcpy(&(pFilteringTable->rules[i]), &flt_rule, sizeof(struct ipa_flt_rule_mdfy));
}
if (false == m_filtering.ModifyFilteringRule(pFilteringTable))
{
IPACMERR("Failed to modify filter config rules.\n");
res = IPACM_FAILURE;
goto fail;
}
}
fail:
if(pFilteringTable != NULL)
{
free(pFilteringTable);
}
return res;
}
int IPACM_Lan::handle_private_subnet_android(ipa_ip_type iptype)
{
int i, len, res = IPACM_SUCCESS;
struct ipa_flt_rule_mdfy flt_rule;
struct ipa_ioc_mdfy_flt_rule* pFilteringTable;
int mtu_rule_cnt = 0;
uint16_t mtu[IPA_MAX_PRIVATE_SUBNET_ENTRIES + IPA_MAX_MTU_ENTRIES] = { };
int mtu_rule_idx = IPACM_Iface::ipacmcfg->ipa_num_private_subnet;
if (rx_prop == NULL)
{
IPACMDBG_H("No rx properties registered for iface %s\n", dev_name);
return IPACM_SUCCESS;
}
if(iptype == IPA_IP_v6)
{
IPACMDBG_H("There is no ipv6 dummy filter rules needed for iface %s\n", dev_name);
return 0;
}
else
{
for(i=0; i<IPA_MAX_PRIVATE_SUBNET_ENTRIES + IPA_MAX_MTU_ENTRIES; i++)
{
reset_to_dummy_flt_rule(IPA_IP_v4, private_fl_rule_hdl[i]);
}
/* check how many MTU rules we need to add */
for(i = 0; i < IPACM_Iface::ipacmcfg->ipa_num_private_subnet; i++)
{
mtu[i] = IPACM_Wan::queryMTU(ipa_if_num, IPA_IP_v4);
if (mtu[i] > 0)
{
mtu_rule_cnt++;
IPACMDBG_H("MTU[%d] is %d\n", i, mtu[i]);
}
else
{
IPACMDBG_H("MTU is zero\n");
}
}
IPACMDBG_H("total %d MTU rules are needed\n", mtu_rule_cnt);
len = sizeof(struct ipa_ioc_mdfy_flt_rule) + (IPACM_Iface::ipacmcfg->ipa_num_private_subnet + mtu_rule_cnt) * sizeof(struct ipa_flt_rule_mdfy);
pFilteringTable = (struct ipa_ioc_mdfy_flt_rule*)malloc(len);
if (!pFilteringTable)
{
IPACMERR("Failed to allocate ipa_ioc_mdfy_flt_rule memory...\n");
return IPACM_FAILURE;
}
memset(pFilteringTable, 0, len);
pFilteringTable->commit = 1;
pFilteringTable->ip = iptype;
pFilteringTable->num_rules = (uint8_t)IPACM_Iface::ipacmcfg->ipa_num_private_subnet + mtu_rule_cnt;
/* Make LAN-traffic always go A5, use default IPA-RT table */
if (false == m_routing.GetRoutingTable(&IPACM_Iface::ipacmcfg->rt_tbl_default_v4))
{
IPACMERR("Failed to get routing table handle.\n");
res = IPACM_FAILURE;
goto fail;
}
memset(&flt_rule, 0, sizeof(struct ipa_flt_rule_mdfy));
flt_rule.status = -1;
flt_rule.rule.retain_hdr = 1;
flt_rule.rule.to_uc = 0;
IPACMDBG_H("Private filter rule use table: %s\n",IPACM_Iface::ipacmcfg->rt_tbl_default_v4.name);
for (i = 0; i < (IPACM_Iface::ipacmcfg->ipa_num_private_subnet); i++)
{
/* add private subnet rule for ipv4 */
/* these 3 properties below will be reset during construct_mtu_rule */
flt_rule.rule.action = IPA_PASS_TO_ROUTING;
flt_rule.rule.eq_attrib_type = 0;
flt_rule.rule.rt_tbl_hdl = IPACM_Iface::ipacmcfg->rt_tbl_default_v4.hdl;
flt_rule.rule_hdl = private_fl_rule_hdl[i];
memcpy(&flt_rule.rule.attrib, &rx_prop->rx[0].attrib, sizeof(flt_rule.rule.attrib));
flt_rule.rule.attrib.attrib_mask |= IPA_FLT_DST_ADDR;
flt_rule.rule.attrib.u.v4.dst_addr_mask = IPACM_Iface::ipacmcfg->private_subnet_table[i].subnet_mask;
flt_rule.rule.attrib.u.v4.dst_addr = IPACM_Iface::ipacmcfg->private_subnet_table[i].subnet_addr;
memcpy(&(pFilteringTable->rules[i]), &flt_rule, sizeof(struct ipa_flt_rule_mdfy));
IPACMDBG_H(" IPACM private subnet_addr as: 0x%x entry(%d)\n", flt_rule.rule.attrib.u.v4.dst_addr, i);
/* add corresponding MTU rule for ipv4 */
if (mtu[i] > 0)
{
flt_rule.rule_hdl = private_fl_rule_hdl[mtu_rule_idx + i];
memcpy(&flt_rule.rule.attrib, &rx_prop->rx[0].attrib, sizeof(flt_rule.rule.attrib));
flt_rule.rule.attrib.u.v4.src_addr_mask = IPACM_Iface::ipacmcfg->private_subnet_table[i].subnet_mask;
flt_rule.rule.attrib.u.v4.src_addr = IPACM_Iface::ipacmcfg->private_subnet_table[i].subnet_addr;
flt_rule.rule.attrib.attrib_mask |= IPA_FLT_SRC_ADDR;
if (construct_mtu_rule(&flt_rule.rule, IPA_IP_v4, mtu[i]))
{
IPACMERR("Failed to modify MTU filtering rule.\n");
}
memcpy(&(pFilteringTable->rules[mtu_rule_idx + i]), &flt_rule, sizeof(struct ipa_flt_rule_mdfy));
IPACMDBG_H("Adding MTU rule for private subnet 0x%x.\n", flt_rule.rule.attrib.u.v4.src_addr);
}
}
if (false == m_filtering.ModifyFilteringRule(pFilteringTable))
{
IPACMERR("Failed to modify private subnet filtering rules.\n");
res = IPACM_FAILURE;
goto fail;
}
}
fail:
if(pFilteringTable != NULL)
{
free(pFilteringTable);
}
return res;
}
int IPACM_Lan::modify_ipv6_prefix_flt_rule(uint32_t* prefix)
{
int len, res = IPACM_SUCCESS;
struct ipa_flt_rule_mdfy flt_rule_entry;
struct ipa_ioc_mdfy_flt_rule* flt_rule;
int rule_cnt = 1;
if(prefix == NULL)
{
IPACMERR("IPv6 prefix is empty.\n");
return IPACM_FAILURE;
}
IPACMDBG_H("Receive IPv6 prefix: 0x%08x%08x.\n", prefix[0], prefix[1]);
uint16_t mtu = IPACM_Wan::queryMTU(ipa_if_num, IPA_IP_v6);
if (mtu > 0)
{
IPACMDBG_H("MTU is %d\n", mtu);
rule_cnt ++;
}
else
{
IPACMERR("MTU is 0");
}
if(rx_prop == NULL)
{
IPACMERR("no rx props\n");
return IPACM_FAILURE;
}
len = sizeof(struct ipa_ioc_mdfy_flt_rule) + rule_cnt * sizeof(struct ipa_flt_rule_mdfy);
flt_rule = (struct ipa_ioc_mdfy_flt_rule*)malloc(len);
if(!flt_rule)
{
IPACMERR("Failed to allocate ipa_ioc_mdfy_flt_rule memory...\n");
return IPACM_FAILURE;
}
memset(flt_rule, 0, len);
flt_rule->commit = 1;
flt_rule->ip = IPA_IP_v6;
flt_rule->num_rules = rule_cnt;
memset(&flt_rule_entry, 0, sizeof(struct ipa_flt_rule_mdfy));
flt_rule_entry.status = -1;
flt_rule_entry.rule.retain_hdr = 1;
flt_rule_entry.rule.to_uc = 0;
flt_rule_entry.rule.action = IPA_PASS_TO_EXCEPTION;
flt_rule_entry.rule.eq_attrib_type = 0;
flt_rule_entry.rule_hdl = ipv6_prefix_flt_rule_hdl[0];
memcpy(&flt_rule_entry.rule.attrib, &rx_prop->rx[0].attrib, sizeof(flt_rule_entry.rule.attrib));
flt_rule_entry.rule.attrib.attrib_mask |= IPA_FLT_DST_ADDR;
flt_rule_entry.rule.attrib.u.v6.dst_addr[0] = prefix[0];
flt_rule_entry.rule.attrib.u.v6.dst_addr[1] = prefix[1];
flt_rule_entry.rule.attrib.u.v6.dst_addr[2] = 0x0;
flt_rule_entry.rule.attrib.u.v6.dst_addr[3] = 0x0;
flt_rule_entry.rule.attrib.u.v6.dst_addr_mask[0] = 0xFFFFFFFF;
flt_rule_entry.rule.attrib.u.v6.dst_addr_mask[1] = 0xFFFFFFFF;
flt_rule_entry.rule.attrib.u.v6.dst_addr_mask[2] = 0x0;
flt_rule_entry.rule.attrib.u.v6.dst_addr_mask[3] = 0x0;
memcpy(&(flt_rule->rules[0]), &flt_rule_entry, sizeof(struct ipa_flt_rule_mdfy));
flt_rule_entry.rule_hdl = ipv6_prefix_flt_rule_hdl[1];
memcpy(&flt_rule_entry.rule.attrib, &rx_prop->rx[0].attrib, sizeof(flt_rule_entry.rule.attrib)); // this will remove the IPA_FLT_DST_ADDR
flt_rule_entry.rule.attrib.u.v6.src_addr[3] = prefix[0];
flt_rule_entry.rule.attrib.u.v6.src_addr[2] = prefix[1];
flt_rule_entry.rule.attrib.u.v6.src_addr[1] = 0x0;
flt_rule_entry.rule.attrib.u.v6.src_addr[0] = 0x0;
flt_rule_entry.rule.attrib.u.v6.src_addr_mask[3] = 0xFFFFFFFF;
flt_rule_entry.rule.attrib.u.v6.src_addr_mask[2] = 0xFFFFFFFF;
flt_rule_entry.rule.attrib.u.v6.src_addr_mask[1] = 0x0;
flt_rule_entry.rule.attrib.u.v6.src_addr_mask[0] = 0x0;
flt_rule_entry.rule.attrib.attrib_mask |= IPA_FLT_SRC_ADDR;
/* Add an MTU rule with every new private prefix */
if (mtu > 0)
{
if (construct_mtu_rule(&flt_rule_entry.rule, IPA_IP_v6, mtu))
{
IPACMERR("Failed to add MTU filtering rule.\n")
}
else
{
memcpy(&(flt_rule->rules[1]), &flt_rule_entry, sizeof(struct ipa_flt_rule_mdfy));
}
}
if(false == m_filtering.ModifyFilteringRule(flt_rule))
{
IPACMERR("Failed to modify prefix filtering rules.\n");
res = IPACM_FAILURE;
goto fail;
}
fail:
if(flt_rule != NULL)
{
free(flt_rule);
}
return res;
}
int IPACM_Lan::install_ipv6_prefix_flt_rule(uint32_t* prefix)
{
int len;
struct ipa_ioc_add_flt_rule* flt_rule;
struct ipa_flt_rule_add flt_rule_entry;
bool result;
int rule_cnt = 1;
if(prefix == NULL)
{
IPACMERR("IPv6 prefix is empty.\n");
return IPACM_FAILURE;
}
IPACMDBG_H("Receive IPv6 prefix: 0x%08x%08x.\n", prefix[0], prefix[1]);
uint16_t mtu = IPACM_Wan::queryMTU(ipa_if_num, IPA_IP_v6);
if (mtu > 0)
{
IPACMDBG_H("MTU is %d\n", mtu);
rule_cnt ++;
}
else
{
IPACMERR("MTU is 0");
}
if(rx_prop != NULL)
{
len = sizeof(struct ipa_ioc_add_flt_rule) + rule_cnt * sizeof(struct ipa_flt_rule_add);
flt_rule = (struct ipa_ioc_add_flt_rule *)calloc(rule_cnt, len);
if (!flt_rule)
{
IPACMERR("Error Locate ipa_flt_rule_add memory...\n");
return IPACM_FAILURE;
}
flt_rule->commit = 1;
flt_rule->ep = rx_prop->rx[0].src_pipe;
flt_rule->global = false;
flt_rule->ip = IPA_IP_v6;
flt_rule->num_rules = rule_cnt;
memset(&flt_rule_entry, 0, sizeof(struct ipa_flt_rule_add));
flt_rule_entry.rule.retain_hdr = 1;
flt_rule_entry.rule.to_uc = 0;
flt_rule_entry.rule.eq_attrib_type = 0;
flt_rule_entry.at_rear = true;
flt_rule_entry.flt_rule_hdl = -1;
flt_rule_entry.status = -1;
flt_rule_entry.rule.action = IPA_PASS_TO_EXCEPTION;
if (IPACM_Iface::ipacmcfg->isIPAv3Supported())
flt_rule_entry.rule.hashable = true;
memcpy(&flt_rule_entry.rule.attrib, &rx_prop->rx[0].attrib, sizeof(flt_rule_entry.rule.attrib));
flt_rule_entry.rule.attrib.attrib_mask |= IPA_FLT_DST_ADDR;
flt_rule_entry.rule.attrib.u.v6.dst_addr[0] = prefix[0];
flt_rule_entry.rule.attrib.u.v6.dst_addr[1] = prefix[1];
flt_rule_entry.rule.attrib.u.v6.dst_addr[2] = 0x0;
flt_rule_entry.rule.attrib.u.v6.dst_addr[3] = 0x0;
flt_rule_entry.rule.attrib.u.v6.dst_addr_mask[0] = 0xFFFFFFFF;
flt_rule_entry.rule.attrib.u.v6.dst_addr_mask[1] = 0xFFFFFFFF;
flt_rule_entry.rule.attrib.u.v6.dst_addr_mask[2] = 0x0;
flt_rule_entry.rule.attrib.u.v6.dst_addr_mask[3] = 0x0;
memcpy(&(flt_rule->rules[0]), &flt_rule_entry, sizeof(struct ipa_flt_rule_add));
memcpy(&flt_rule_entry.rule.attrib, &rx_prop->rx[0].attrib, sizeof(flt_rule_entry.rule.attrib)); // this will remove the IPA_FLT_DST_ADDR
flt_rule_entry.rule.attrib.u.v6.src_addr[3] = prefix[0];
flt_rule_entry.rule.attrib.u.v6.src_addr[2] = prefix[1];
flt_rule_entry.rule.attrib.u.v6.src_addr[1] = 0x0;
flt_rule_entry.rule.attrib.u.v6.src_addr[0] = 0x0;
flt_rule_entry.rule.attrib.u.v6.src_addr_mask[3] = 0xFFFFFFFF;
flt_rule_entry.rule.attrib.u.v6.src_addr_mask[2] = 0xFFFFFFFF;
flt_rule_entry.rule.attrib.u.v6.src_addr_mask[1] = 0x0;
flt_rule_entry.rule.attrib.u.v6.src_addr_mask[0] = 0x0;
flt_rule_entry.rule.attrib.attrib_mask |= IPA_FLT_SRC_ADDR;
/* Add an MTU rule with every new private prefix */
if (mtu > 0)
{
if (construct_mtu_rule(&flt_rule_entry.rule, IPA_IP_v6, mtu))
{
IPACMERR("Failed to add MTU filtering rule.\n")
}
else
{
memcpy(&(flt_rule->rules[1]), &flt_rule_entry, sizeof(struct ipa_flt_rule_add));
}
}
result = m_filtering.AddFilteringRule(flt_rule);
if (result == false)
{
IPACMERR("Error Adding Filtering rule, aborting...\n");
free(flt_rule);
return IPACM_FAILURE;
}
else
{
IPACM_Iface::ipacmcfg->increaseFltRuleCount(rx_prop->rx[0].src_pipe, IPA_IP_v6, 2);
ipv6_prefix_flt_rule_hdl[0] = flt_rule->rules[0].flt_rule_hdl;
IPACMDBG_H("IPv6 prefix filter rule HDL:0x%x\n", ipv6_prefix_flt_rule_hdl[0]);
if (rule_cnt > 1)
{
ipv6_prefix_flt_rule_hdl[1] = flt_rule->rules[1].flt_rule_hdl;
IPACMDBG_H("IPv6 prefix MTU filter rule HDL:0x%x\n", ipv6_prefix_flt_rule_hdl[1]);
}
free(flt_rule);
}
}
return IPACM_SUCCESS;
}
void IPACM_Lan::delete_ipv6_prefix_flt_rule()
{
if(m_filtering.DeleteFilteringHdls(ipv6_prefix_flt_rule_hdl, IPA_IP_v6, NUM_IPV6_PREFIX_FLT_RULE + NUM_IPV6_PREFIX_MTU_RULE) == false)
{
IPACMERR("Failed to delete ipv6 prefix flt rules.\n");
return;
}
IPACM_Iface::ipacmcfg->decreaseFltRuleCount(rx_prop->rx[0].src_pipe, IPA_IP_v6, NUM_IPV6_PREFIX_FLT_RULE + NUM_IPV6_PREFIX_MTU_RULE);
return;
}
int IPACM_Lan::handle_addr_evt_odu_bridge(ipacm_event_data_addr* data)
{
int fd, res = IPACM_SUCCESS;
struct in6_addr ipv6_addr;
if(data == NULL)
{
IPACMERR("Failed to get interface IP address.\n");
return IPACM_FAILURE;
}
if(data->iptype == IPA_IP_v6)
{
fd = open(IPACM_Iface::ipacmcfg->DEVICE_NAME_ODU, O_RDWR);
if(fd == 0)
{
IPACMERR("Failed to open %s.\n", IPACM_Iface::ipacmcfg->DEVICE_NAME_ODU);
return IPACM_FAILURE;
}
memcpy(&ipv6_addr, data->ipv6_addr, sizeof(struct in6_addr));
if( ioctl(fd, ODU_BRIDGE_IOC_SET_LLV6_ADDR, &ipv6_addr) )
{
IPACMERR("Failed to write IPv6 address to odu driver.\n");
res = IPACM_FAILURE;
}
num_dft_rt_v6++;
close(fd);
}
return res;
}
ipa_hdr_proc_type IPACM_Lan::eth_bridge_get_hdr_proc_type(ipa_hdr_l2_type t1, ipa_hdr_l2_type t2)
{
if(t1 == IPA_HDR_L2_ETHERNET_II)
{
if(t2 == IPA_HDR_L2_ETHERNET_II)
{
return IPA_HDR_PROC_ETHII_TO_ETHII;
}
if(t2 == IPA_HDR_L2_802_3)
{
return IPA_HDR_PROC_ETHII_TO_802_3;
}
}
if(t1 == IPA_HDR_L2_802_3)
{
if(t2 == IPA_HDR_L2_ETHERNET_II)
{
return IPA_HDR_PROC_802_3_TO_ETHII;
}
if(t2 == IPA_HDR_L2_802_3)
{
return IPA_HDR_PROC_802_3_TO_802_3;
}
}
return IPA_HDR_PROC_NONE;
}
int IPACM_Lan::eth_bridge_get_hdr_template_hdl(uint32_t* hdr_hdl)
{
if(hdr_hdl == NULL)
{
IPACMDBG_H("Hdr handle pointer is empty.\n");
return IPACM_FAILURE;
}
struct ipa_ioc_get_hdr hdr;
memset(&hdr, 0, sizeof(hdr));
memcpy(hdr.name, tx_prop->tx[0].hdr_name, sizeof(hdr.name));
if(m_header.GetHeaderHandle(&hdr) == false)
{
IPACMERR("Failed to get template hdr hdl.\n");
return IPACM_FAILURE;
}
*hdr_hdl = hdr.hdl;
return IPACM_SUCCESS;
}
int IPACM_Lan::handle_cradle_wan_mode_switch(bool is_wan_bridge_mode)
{
struct ipa_flt_rule_mdfy flt_rule_entry;
int len = 0;
ipa_ioc_mdfy_flt_rule *m_pFilteringTable;
IPACMDBG_H("Handle wan mode swtich: is wan bridge mode?%d\n", is_wan_bridge_mode);
if (rx_prop == NULL)
{
IPACMDBG_H("No rx properties registered for iface %s\n", dev_name);
return IPACM_SUCCESS;
}
len = sizeof(struct ipa_ioc_mdfy_flt_rule) + (1 * sizeof(struct ipa_flt_rule_mdfy));
m_pFilteringTable = (struct ipa_ioc_mdfy_flt_rule *)calloc(1, len);
if (m_pFilteringTable == NULL)
{
PERROR("Error Locate ipa_ioc_mdfy_flt_rule memory...\n");
return IPACM_FAILURE;
}
m_pFilteringTable->commit = 1;
m_pFilteringTable->ip = IPA_IP_v4;
m_pFilteringTable->num_rules = (uint8_t)1;
IPACMDBG_H("Retrieving routing hanle for table: %s\n",
IPACM_Iface::ipacmcfg->rt_tbl_wan_v4.name);
if (false == m_routing.GetRoutingTable(&IPACM_Iface::ipacmcfg->rt_tbl_wan_v4))
{
IPACMERR("m_routing.GetRoutingTable(&IPACM_Iface::ipacmcfg->rt_tbl_wan_v4=0x%p) Failed.\n",
&IPACM_Iface::ipacmcfg->rt_tbl_wan_v4);
free(m_pFilteringTable);
return IPACM_FAILURE;
}
IPACMDBG_H("Routing handle for table: %d\n", IPACM_Iface::ipacmcfg->rt_tbl_wan_v4.hdl);
memset(&flt_rule_entry, 0, sizeof(struct ipa_flt_rule_mdfy)); // Zero All Fields
flt_rule_entry.status = -1;
flt_rule_entry.rule_hdl = lan_wan_fl_rule_hdl[0];
flt_rule_entry.rule.retain_hdr = 0;
flt_rule_entry.rule.to_uc = 0;
flt_rule_entry.rule.eq_attrib_type = 0;
if(is_wan_bridge_mode)
{
flt_rule_entry.rule.action = IPA_PASS_TO_ROUTING;
}
else
{
flt_rule_entry.rule.action = IPA_PASS_TO_SRC_NAT;
}
flt_rule_entry.rule.rt_tbl_hdl = IPACM_Iface::ipacmcfg->rt_tbl_wan_v4.hdl;
memcpy(&flt_rule_entry.rule.attrib,
&rx_prop->rx[0].attrib,
sizeof(flt_rule_entry.rule.attrib));
flt_rule_entry.rule.attrib.attrib_mask |= IPA_FLT_DST_ADDR;
flt_rule_entry.rule.attrib.u.v4.dst_addr_mask = 0x0;
flt_rule_entry.rule.attrib.u.v4.dst_addr = 0x0;
memcpy(&m_pFilteringTable->rules[0], &flt_rule_entry, sizeof(flt_rule_entry));
if (false == m_filtering.ModifyFilteringRule(m_pFilteringTable))
{
IPACMERR("Error Modifying RuleTable(0) to Filtering, aborting...\n");
free(m_pFilteringTable);
return IPACM_FAILURE;
}
else
{
IPACMDBG_H("flt rule hdl = %d, status = %d\n",
m_pFilteringTable->rules[0].rule_hdl,
m_pFilteringTable->rules[0].status);
}
free(m_pFilteringTable);
return IPACM_SUCCESS;
}
/*handle reset usb-client rt-rules */
int IPACM_Lan::handle_tethering_stats_event(ipa_get_data_stats_resp_msg_v01 *data)
{
int fd;
uint32_t pipe_len, cnt;
uint64_t num_ul_packets, num_ul_bytes;
uint64_t num_dl_packets, num_dl_bytes;
bool ul_pipe_found, dl_pipe_found;
FILE *fp = NULL;
fd = open(IPA_DEVICE_NAME, O_RDWR);
if (fd < 0)
{
IPACMERR("Failed opening %s.\n", IPA_DEVICE_NAME);
return IPACM_FAILURE;
}
ul_pipe_found = false;
dl_pipe_found = false;
num_ul_packets = 0;
num_dl_packets = 0;
num_ul_bytes = 0;
num_dl_bytes = 0;
if (data->dl_dst_pipe_stats_list_valid)
{
if(tx_prop != NULL)
{
for (pipe_len = 0; pipe_len < data->dl_dst_pipe_stats_list_len; pipe_len++)
{
IPACMDBG_H("Check entry(%d) dl_dst_pipe(%d)\n", pipe_len, data->dl_dst_pipe_stats_list[pipe_len].pipe_index);
for (cnt=0; cnt<tx_prop->num_tx_props; cnt++)
{
IPACMDBG_H("Check Tx_prop_entry(%d) pipe(%d)\n", cnt, ioctl(fd, IPA_IOC_QUERY_EP_MAPPING, tx_prop->tx[cnt].dst_pipe));
if(ioctl(fd, IPA_IOC_QUERY_EP_MAPPING, tx_prop->tx[cnt].dst_pipe) == (int)data->dl_dst_pipe_stats_list[pipe_len].pipe_index)
{
/* update the DL stats */
dl_pipe_found = true;
num_dl_packets += data->dl_dst_pipe_stats_list[pipe_len].num_ipv4_packets;
num_dl_packets += data->dl_dst_pipe_stats_list[pipe_len].num_ipv6_packets;
num_dl_bytes += data->dl_dst_pipe_stats_list[pipe_len].num_ipv4_bytes;
num_dl_bytes += data->dl_dst_pipe_stats_list[pipe_len].num_ipv6_bytes;
IPACMDBG_H("Got matched dst-pipe (%d) from %d tx props\n", data->dl_dst_pipe_stats_list[pipe_len].pipe_index, cnt);
IPACMDBG_H("DL_packets:(%llu) DL_bytes:(%llu) \n", (long long)num_dl_packets, (long long)num_dl_bytes);
break;
}
}
}
}
}
if (data->ul_src_pipe_stats_list_valid)
{
if(rx_prop != NULL)
{
for (pipe_len = 0; pipe_len < data->ul_src_pipe_stats_list_len; pipe_len++)
{
IPACMDBG_H("Check entry(%d) dl_dst_pipe(%d)\n", pipe_len, data->ul_src_pipe_stats_list[pipe_len].pipe_index);
for (cnt=0; cnt < rx_prop->num_rx_props; cnt++)
{
IPACMDBG_H("Check Rx_prop_entry(%d) pipe(%d)\n", cnt, ioctl(fd, IPA_IOC_QUERY_EP_MAPPING, rx_prop->rx[cnt].src_pipe));
//Typecasting to avoid -Wall -Werror errors
if(ioctl(fd, IPA_IOC_QUERY_EP_MAPPING, rx_prop->rx[cnt].src_pipe) == (int)data->ul_src_pipe_stats_list[pipe_len].pipe_index)
{
/* update the UL stats */
ul_pipe_found = true;
num_ul_packets += data->ul_src_pipe_stats_list[pipe_len].num_ipv4_packets;
num_ul_packets += data->ul_src_pipe_stats_list[pipe_len].num_ipv6_packets;
num_ul_bytes += data->ul_src_pipe_stats_list[pipe_len].num_ipv4_bytes;
num_ul_bytes += data->ul_src_pipe_stats_list[pipe_len].num_ipv6_bytes;
IPACMDBG_H("Got matched dst-pipe (%d) from %d tx props\n", data->ul_src_pipe_stats_list[pipe_len].pipe_index, cnt);
IPACMDBG_H("UL_packets:(%llu) UL_bytes:(%llu) \n", (long long)num_ul_packets, (long long)num_ul_bytes);
break;
}
}
}
}
}
close(fd);
if (ul_pipe_found || dl_pipe_found)
{
IPACMDBG_H("Update IPA_TETHERING_STATS_UPDATE_EVENT, TX(P%llu/B%llu) RX(P%llu/B%llu) DEV(%s) to LTE(%s) \n",
(long long)num_ul_packets,
(long long)num_ul_bytes,
(long long)num_dl_packets,
(long long)num_dl_bytes,
dev_name,
IPACM_Wan::wan_up_dev_name);
fp = fopen(IPA_PIPE_STATS_FILE_NAME, "w");
if ( fp == NULL )
{
IPACMERR("Failed to write pipe stats to %s, error is %d - %s\n",
IPA_PIPE_STATS_FILE_NAME, errno, strerror(errno));
return IPACM_FAILURE;
}
fprintf(fp, PIPE_STATS,
dev_name,
IPACM_Wan::wan_up_dev_name,
(long long)num_ul_bytes,
(long long)num_ul_packets,
(long long)num_dl_bytes,
(long long)num_dl_packets);
fclose(fp);
}
return IPACM_SUCCESS;
}
int IPACM_Lan::set_client_pipe(enum ipa_client_type client, uint32_t *pipe)
{
int fd;
fd = open(IPA_DEVICE_NAME, O_RDWR);
if(fd < 0)
{
IPACMERR("Failed opening %s.\n", IPA_DEVICE_NAME);
return IPACM_FAILURE;
}
*pipe = ioctl(fd, IPA_IOC_QUERY_EP_MAPPING, client);
close(fd);
return IPACM_SUCCESS;
}
int IPACM_Lan::set_tether_client_wigig(wan_ioctl_set_tether_client_pipe *tether_client)
{
#define NUM_WIGIG_TX_PIPES 4
uint32_t cnt;
int ret;
int fd_wwan_ioctl = open(WWAN_QMI_IOCTL_DEVICE_NAME, O_RDWR);
if(fd_wwan_ioctl < 0)
{
IPACMERR("Failed to open %s.\n", WWAN_QMI_IOCTL_DEVICE_NAME);
return IPACM_FAILURE;
}
if(!tx_prop || !rx_prop)
{
IPACMERR("no props, can't set client %p, %p\n", rx_prop, tx_prop);
close(fd_wwan_ioctl);
return IPACM_FAILURE;
}
/* only one rx pipe for wigig */
tether_client->ul_src_pipe_len = rx_prop->num_rx_props;
for(cnt = 0; cnt < rx_prop->num_rx_props; cnt++)
{
ret = set_client_pipe(rx_prop->rx[cnt].src_pipe, &tether_client->ul_src_pipe_list[cnt]);
if(ret)
{
close(fd_wwan_ioctl);
return ret;
}
IPACMDBG_H("Rx(%d), src_pipe: %d, ipa_pipe: %d\n",
cnt, rx_prop->rx[cnt].src_pipe,
tether_client->ul_src_pipe_list[cnt]);
}
for(int i = 0; i < NUM_WIGIG_TX_PIPES; i++)
{
/* 4 tx pipes for wigig */
tether_client->dl_dst_pipe_len = tx_prop->num_tx_props;
#ifdef IPA_CLIENT_WIGIG4_CONS
for(cnt = 0; cnt < tx_prop->num_tx_props; cnt++)
{
enum ipa_client_type client;
switch(i) {
case 0:
client = IPA_CLIENT_WIGIG1_CONS;
break;
case 1:
client = IPA_CLIENT_WIGIG2_CONS;
break;
case 2:
client = IPA_CLIENT_WIGIG3_CONS;
break;
case 3:
client = IPA_CLIENT_WIGIG4_CONS;
break;
default:
IPACMERR("shouldn't get here\n");
close(fd_wwan_ioctl);
return IPACM_FAILURE;
}
ret = set_client_pipe(client, &tether_client->dl_dst_pipe_list[cnt]);
if(ret)
{
close(fd_wwan_ioctl);
return ret;
}
IPACMDBG_H("Tx(%d), IPA_CLIENT_WIGIG%d_CONS, ipa_pipe: %d\n",
cnt, i + 1,
tether_client->dl_dst_pipe_list[cnt]);
}
#endif
ret = ioctl(fd_wwan_ioctl, WAN_IOC_SET_TETHER_CLIENT_PIPE, tether_client);
if(ret != 0)
{
IPACMERR("Failed set tether-client-pipe %p with ret %d\n ", tether_client, ret);
}
IPACMDBG("Set wigig tether-client-pipe (%d) %p\n", i, tether_client);
}
close(fd_wwan_ioctl);
return ret;
}
int IPACM_Lan::set_tether_client(wan_ioctl_set_tether_client_pipe *tether_client)
{
uint32_t cnt;
int ret;
int fd_wwan_ioctl = open(WWAN_QMI_IOCTL_DEVICE_NAME, O_RDWR);
if(fd_wwan_ioctl < 0)
{
IPACMERR("Failed to open %s.\n", WWAN_QMI_IOCTL_DEVICE_NAME);
return IPACM_FAILURE;
}
if(tx_prop != NULL)
{
tether_client->dl_dst_pipe_len = tx_prop->num_tx_props;
for(cnt = 0; cnt < tx_prop->num_tx_props; cnt++)
{
ret = set_client_pipe(tx_prop->tx[cnt].dst_pipe, &tether_client->dl_dst_pipe_list[cnt]);
if(ret)
{
close(fd_wwan_ioctl);
return ret;
}
IPACMDBG_H("Tx(%d), dst_pipe: %d, ipa_pipe: %d\n",
cnt, tx_prop->tx[cnt].dst_pipe,
tether_client->dl_dst_pipe_list[cnt]);
}
}
if(rx_prop != NULL)
{
tether_client->ul_src_pipe_len = rx_prop->num_rx_props;
for(cnt = 0; cnt < rx_prop->num_rx_props; cnt++)
{
ret = set_client_pipe(rx_prop->rx[cnt].src_pipe, &tether_client->ul_src_pipe_list[cnt]);
if(ret)
{
close(fd_wwan_ioctl);
return ret;
}
IPACMDBG_H("Rx(%d), src_pipe: %d, ipa_pipe: %d\n",
cnt, rx_prop->rx[cnt].src_pipe,
tether_client->ul_src_pipe_list[cnt]);
}
}
ret = ioctl(fd_wwan_ioctl, WAN_IOC_SET_TETHER_CLIENT_PIPE, tether_client);
if(ret != 0)
{
IPACMERR("Failed set tether-client-pipe %p with ret %d\n ", &tether_client, ret);
}
IPACMDBG("Set tether-client-pipe %p\n", &tether_client);
close(fd_wwan_ioctl);
return ret;
}
/*handle tether client */
int IPACM_Lan::handle_tethering_client(bool reset, ipacm_client_enum ipa_client)
{
int ret = IPACM_SUCCESS;
wan_ioctl_set_tether_client_pipe tether_client;
memset(&tether_client, 0, sizeof(tether_client));
tether_client.reset_client = reset;
tether_client.ipa_client = ipa_client;
/* special case for wigig (11ad) who has 4 Tx and 1 RX pipe */
if(!strcmp(dev_name, "wigig0"))
{
set_tether_client_wigig(&tether_client);
}
else
{
set_tether_client(&tether_client);
}
return ret;
}
/* mac address has to be provided for client related events */
void IPACM_Lan::eth_bridge_post_event(ipa_cm_event_id evt, ipa_ip_type iptype, uint8_t *mac, uint32_t *ipv6_addr, char *iface_name,
int ep)
{
ipacm_cmd_q_data eth_bridge_evt;
ipacm_event_eth_bridge *evt_data_eth_bridge;
const char *eventName = IPACM_Iface::ipacmcfg->getEventName(evt);
#ifdef FEATURE_L2TP
ipacm_event_data_all *evt_data_all;
#endif
/* not enable rndis for lan2lan HW-offload due to android limitation */
#ifdef FEATURE_IPA_ANDROID
if(ipa_if_cate == LAN_IF)
{
IPACMDBG_H("This is LAN IF (%s):ipa_index (%d) skip lan2lan events for Android \n", IPACM_Iface::ipacmcfg->iface_table[ipa_if_num].iface_name, ipa_if_num);
return;
}
#endif
if(ipv6_addr)
{
IPACMDBG_H("IPv6 addr: %08x:%08x:%08x:%08x \n", ipv6_addr[0],
ipv6_addr[1], ipv6_addr[2], ipv6_addr[3]);
}
memset(&eth_bridge_evt, 0, sizeof(ipacm_cmd_q_data));
eth_bridge_evt.event = evt;
#ifdef FEATURE_L2TP
if(evt == IPA_HANDLE_VLAN_CLIENT_INFO || evt == IPA_HANDLE_VLAN_IFACE_INFO)
{
evt_data_all = (ipacm_event_data_all*)malloc(sizeof(*evt_data_all));
if(evt_data_all == NULL)
{
IPACMERR("Failed to allocate memory.\n");
return;
}
memset(evt_data_all, 0, sizeof(*evt_data_all));
if(ipv6_addr)
{
IPACMDBG_H("IPv6 addr: %08x:%08x:%08x:%08x \n", ipv6_addr[0],
ipv6_addr[1], ipv6_addr[2], ipv6_addr[3]);
memcpy(evt_data_all->ipv6_addr, ipv6_addr, sizeof(evt_data_all->ipv6_addr));
}
if(mac)
{
IPACMDBG_H("Mac: 0x%02x%02x%02x%02x%02x%02x \n",
mac[0], mac[1], mac[2], mac[3], mac[4], mac[5]);
memcpy(evt_data_all->mac_addr, mac, sizeof(evt_data_all->mac_addr));
}
if(iface_name)
{
IPACMDBG_H("Iface: %s\n", iface_name);
memcpy(evt_data_all->iface_name, iface_name, sizeof(evt_data_all->iface_name));
}
eth_bridge_evt.evt_data = (void*)evt_data_all;
}
else
#endif
{
evt_data_eth_bridge = (ipacm_event_eth_bridge*)malloc(sizeof(*evt_data_eth_bridge));
if(evt_data_eth_bridge == NULL)
{
IPACMERR("Failed to allocate memory.\n");
return;
}
memset(evt_data_eth_bridge, 0, sizeof(*evt_data_eth_bridge));
evt_data_eth_bridge->p_iface = this;
evt_data_eth_bridge->iptype = iptype;
evt_data_eth_bridge->ep = ep;
if(mac)
{
IPACMDBG_H("Mac: 0x%02x%02x%02x%02x%02x%02x \n",
mac[0], mac[1], mac[2], mac[3], mac[4], mac[5]);
memcpy(evt_data_eth_bridge->mac_addr, mac, sizeof(evt_data_eth_bridge->mac_addr));
}
if(iface_name)
{
IPACMDBG_H("Iface: %s\n", iface_name);
memcpy(evt_data_eth_bridge->iface_name, iface_name,
sizeof(evt_data_eth_bridge->iface_name));
}
eth_bridge_evt.evt_data = (void*)evt_data_eth_bridge;
}
if (eventName != NULL)
{
IPACMDBG_H("Posting event %s\n",
eventName);
}
IPACM_EvtDispatcher::PostEvt(&eth_bridge_evt);
}
/* add header processing context and return handle to lan2lan controller */
int IPACM_Lan::eth_bridge_add_hdr_proc_ctx(ipa_hdr_l2_type peer_l2_hdr_type, uint32_t *hdl)
{
int len, res = IPACM_SUCCESS;
uint32_t hdr_template;
ipa_ioc_add_hdr_proc_ctx* pHeaderProcTable = NULL;
if(tx_prop == NULL)
{
IPACMERR("No tx prop.\n");
return IPACM_FAILURE;
}
len = sizeof(struct ipa_ioc_add_hdr_proc_ctx) + sizeof(struct ipa_hdr_proc_ctx_add);
pHeaderProcTable = (ipa_ioc_add_hdr_proc_ctx*)malloc(len);
if(pHeaderProcTable == NULL)
{
IPACMERR("Cannot allocate header processing context table.\n");
return IPACM_FAILURE;
}
memset(pHeaderProcTable, 0, len);
pHeaderProcTable->commit = 1;
pHeaderProcTable->num_proc_ctxs = 1;
pHeaderProcTable->proc_ctx[0].type = eth_bridge_get_hdr_proc_type(peer_l2_hdr_type, tx_prop->tx[0].hdr_l2_type);
eth_bridge_get_hdr_template_hdl(&hdr_template);
pHeaderProcTable->proc_ctx[0].hdr_hdl = hdr_template;
if (m_header.AddHeaderProcCtx(pHeaderProcTable) == false)
{
IPACMERR("Adding hdr proc ctx failed with status: %d\n", pHeaderProcTable->proc_ctx[0].status);
res = IPACM_FAILURE;
goto end;
}
*hdl = pHeaderProcTable->proc_ctx[0].proc_ctx_hdl;
end:
free(pHeaderProcTable);
return res;
}
/* add routing rule and return handle to lan2lan controller */
int IPACM_Lan::eth_bridge_add_rt_rule(uint8_t *mac, char *rt_tbl_name, uint32_t hdr_proc_ctx_hdl,
ipa_hdr_l2_type peer_l2_hdr_type, ipa_ip_type iptype, uint32_t *rt_rule_hdl, int *rt_rule_count,
int ep)
{
int len, res = IPACM_SUCCESS;
uint32_t i, position, num_rt_rule;
struct ipa_ioc_add_rt_rule* rt_rule_table = NULL;
struct ipa_rt_rule_add rt_rule;
IPACMDBG_H("Received client MAC 0x%02x%02x%02x%02x%02x%02x.\n",
mac[0], mac[1], mac[2], mac[3], mac[4], mac[5]);
/*fix -Wall -Werror if wigig feature is not enabled */
IPACMDBG_H("ep: %d\n", ep);
num_rt_rule = each_client_rt_rule_count[iptype];
len = sizeof(ipa_ioc_add_rt_rule) + num_rt_rule * sizeof(ipa_rt_rule_add);
rt_rule_table = (ipa_ioc_add_rt_rule*)malloc(len);
if (rt_rule_table == NULL)
{
IPACMERR("Failed to allocate memory.\n");
return IPACM_FAILURE;
}
memset(rt_rule_table, 0, len);
rt_rule_table->commit = 1;
rt_rule_table->ip = iptype;
rt_rule_table->num_rules = num_rt_rule;
strlcpy(rt_rule_table->rt_tbl_name, rt_tbl_name, sizeof(rt_rule_table->rt_tbl_name));
rt_rule_table->rt_tbl_name[IPA_RESOURCE_NAME_MAX-1] = 0;
memset(&rt_rule, 0, sizeof(ipa_rt_rule_add));
rt_rule.at_rear = false;
rt_rule.status = -1;
rt_rule.rt_rule_hdl = -1;
if (IPACM_Iface::ipacmcfg->isIPAv3Supported())
rt_rule.rule.hashable = true;
rt_rule.rule.hdr_hdl = 0;
rt_rule.rule.hdr_proc_ctx_hdl = hdr_proc_ctx_hdl;
position = 0;
for(i=0; i<iface_query->num_tx_props; i++)
{
if(tx_prop->tx[i].ip == iptype)
{
if(position >= num_rt_rule || position >= MAX_NUM_PROP)
{
IPACMERR("Number of routing rules already exceeds limit.\n");
res = IPACM_FAILURE;
goto end;
}
#ifdef IPA_CLIENT_WIGIG4_CONS
if ((ep >= IPA_CLIENT_WIGIG1_CONS) && (ep <= IPA_CLIENT_WIGIG4_CONS))
{
IPACMDBG_H("wigig DL pipe %d\n", ep);
rt_rule.rule.dst = (enum ipa_client_type)ep;
}
else
#endif
if(ipa_if_cate == WLAN_IF && IPACM_Iface::ipacmcfg->isMCC_Mode)
{
IPACMDBG_H("In WLAN MCC mode, use alt dst pipe: %d\n",
tx_prop->tx[i].alt_dst_pipe);
rt_rule.rule.dst = tx_prop->tx[i].alt_dst_pipe;
}
else
{
IPACMDBG_H("It is not WLAN MCC mode, use dst pipe: %d\n",
tx_prop->tx[i].dst_pipe);
rt_rule.rule.dst = tx_prop->tx[i].dst_pipe;
}
memcpy(&rt_rule.rule.attrib, &tx_prop->tx[i].attrib, sizeof(rt_rule.rule.attrib));
if(peer_l2_hdr_type == IPA_HDR_L2_ETHERNET_II)
rt_rule.rule.attrib.attrib_mask |= IPA_FLT_MAC_DST_ADDR_ETHER_II;
else
rt_rule.rule.attrib.attrib_mask |= IPA_FLT_MAC_DST_ADDR_802_3;
memcpy(rt_rule.rule.attrib.dst_mac_addr, mac, sizeof(rt_rule.rule.attrib.dst_mac_addr));
memset(rt_rule.rule.attrib.dst_mac_addr_mask, 0xFF, sizeof(rt_rule.rule.attrib.dst_mac_addr_mask));
memcpy(&(rt_rule_table->rules[position]), &rt_rule, sizeof(rt_rule_table->rules[position]));
position++;
}
}
if(false == m_routing.AddRoutingRule(rt_rule_table))
{
IPACMERR("Routing rule addition failed!\n");
res = IPACM_FAILURE;
goto end;
}
else
{
*rt_rule_count = position;
for(i=0; i<position; i++)
rt_rule_hdl[i] = rt_rule_table->rules[i].rt_rule_hdl;
}
end:
free(rt_rule_table);
return res;
}
/* modify routing rule*/
int IPACM_Lan::eth_bridge_modify_rt_rule(uint8_t *mac, uint32_t hdr_proc_ctx_hdl,
ipa_hdr_l2_type peer_l2_hdr_type, ipa_ip_type iptype, uint32_t *rt_rule_hdl, int rt_rule_count)
{
struct ipa_ioc_mdfy_rt_rule *rt_rule = NULL;
struct ipa_rt_rule_mdfy *rt_rule_entry;
int len, res = IPACM_SUCCESS;
uint32_t index;
if(tx_prop == NULL)
{
IPACMDBG_H("No tx properties \n");
return IPACM_FAILURE;
}
if(ipa_if_cate != WLAN_IF)
{
IPACMDBG_H("This is not WLAN IF, no need to modify rt rule.\n");
return IPACM_SUCCESS;
}
IPACMDBG_H("Receive WLAN client MAC 0x%02x%02x%02x%02x%02x%02x.\n",
mac[0], mac[1], mac[2], mac[3], mac[4], mac[5]);
len = sizeof(struct ipa_ioc_mdfy_rt_rule) + rt_rule_count * sizeof(struct ipa_rt_rule_mdfy);
rt_rule = (struct ipa_ioc_mdfy_rt_rule *)malloc(len);
if(rt_rule == NULL)
{
IPACMERR("Unable to allocate memory for modify rt rule\n");
return IPACM_FAILURE;
}
memset(rt_rule, 0, len);
rt_rule->commit = 1;
rt_rule->num_rules = 0;
rt_rule->ip = iptype;
for (index = 0; index < tx_prop->num_tx_props; index++)
{
if (tx_prop->tx[index].ip == iptype)
{
if (rt_rule->num_rules >= rt_rule_count ||
rt_rule->num_rules >= MAX_NUM_PROP)
{
IPACMERR("Number of routing rules exceeds limit.\n");
res = IPACM_FAILURE;
goto end;
}
rt_rule_entry = &rt_rule->rules[rt_rule->num_rules];
if (IPACM_Iface::ipacmcfg->isMCC_Mode)
{
IPACMDBG_H("In WLAN MCC mode, use alt dst pipe: %d\n",
tx_prop->tx[index].alt_dst_pipe);
rt_rule_entry->rule.dst = tx_prop->tx[index].alt_dst_pipe;
}
else
{
IPACMDBG_H("In WLAN SCC mode, use dst pipe: %d\n",
tx_prop->tx[index].dst_pipe);
rt_rule_entry->rule.dst = tx_prop->tx[index].dst_pipe;
}
rt_rule_entry->rule.hdr_hdl = 0;
rt_rule_entry->rule.hdr_proc_ctx_hdl = hdr_proc_ctx_hdl;
if (IPACM_Iface::ipacmcfg->isIPAv3Supported())
rt_rule_entry->rule.hashable = true;
memcpy(&rt_rule_entry->rule.attrib, &tx_prop->tx[index].attrib,
sizeof(rt_rule_entry->rule.attrib));
if(peer_l2_hdr_type == IPA_HDR_L2_ETHERNET_II)
rt_rule_entry->rule.attrib.attrib_mask |= IPA_FLT_MAC_DST_ADDR_ETHER_II;
else
rt_rule_entry->rule.attrib.attrib_mask |= IPA_FLT_MAC_DST_ADDR_802_3;
memcpy(rt_rule_entry->rule.attrib.dst_mac_addr, mac,
sizeof(rt_rule_entry->rule.attrib.dst_mac_addr));
memset(rt_rule_entry->rule.attrib.dst_mac_addr_mask, 0xFF,
sizeof(rt_rule_entry->rule.attrib.dst_mac_addr_mask));
rt_rule_entry->rt_rule_hdl = rt_rule_hdl[rt_rule->num_rules];
rt_rule->num_rules++;
}
}
if(m_routing.ModifyRoutingRule(rt_rule) == false)
{
IPACMERR("Failed to modify routing rules.\n");
res = IPACM_FAILURE;
goto end;
}
if(m_routing.Commit(iptype) == false)
{
IPACMERR("Failed to commit routing rules.\n");
res = IPACM_FAILURE;
goto end;
}
IPACMDBG("Modified routing rules successfully.\n");
end:
free(rt_rule);
return res;
}
int IPACM_Lan::eth_bridge_add_flt_rule(uint8_t *mac, uint32_t rt_tbl_hdl, ipa_ip_type iptype, uint32_t *flt_rule_hdl)
{
int res = IPACM_SUCCESS;
int len;
struct ipa_flt_rule_add flt_rule_entry;
struct ipa_ioc_add_flt_rule_after *pFilteringTable = NULL;
bool result;
IPACMDBG_H("Received client MAC 0x%02x%02x%02x%02x%02x%02x.\n", mac[0], mac[1], mac[2], mac[3], mac[4], mac[5]);
IPACMDBG_H("Received rt_tbl_hdl :%d iptype %d\n", rt_tbl_hdl,iptype);
*flt_rule_hdl = 0;
if (IPACM_Iface::ipacmcfg->isIPAv3Supported())
{
if (rx_prop == NULL || tx_prop == NULL)
{
IPACMDBG_H("No rx or tx properties registered for iface %s\n", dev_name);
return IPACM_FAILURE;
}
len = sizeof(struct ipa_ioc_add_flt_rule_after) + sizeof(struct ipa_flt_rule_add);
pFilteringTable = (struct ipa_ioc_add_flt_rule_after*)malloc(len);
if (!pFilteringTable)
{
IPACMERR("Failed to allocate ipa_ioc_add_flt_rule_after memory...\n");
return IPACM_FAILURE;
}
memset(pFilteringTable, 0, len);
/* add mac based rule*/
pFilteringTable->commit = 1;
pFilteringTable->ep = rx_prop->rx[0].src_pipe;
pFilteringTable->ip = iptype;
pFilteringTable->num_rules = 1;
pFilteringTable->add_after_hdl = eth_bridge_flt_rule_offset[iptype];
memset(&flt_rule_entry, 0, sizeof(flt_rule_entry));
flt_rule_entry.at_rear = 1;
flt_rule_entry.rule.retain_hdr = 0;
flt_rule_entry.rule.to_uc = 0;
flt_rule_entry.rule.action = IPA_PASS_TO_ROUTING;
flt_rule_entry.rule.eq_attrib_type = 0;
flt_rule_entry.rule.rt_tbl_hdl = rt_tbl_hdl;
flt_rule_entry.rule.hashable = true;
memcpy(&flt_rule_entry.rule.attrib, &rx_prop->rx[0].attrib, sizeof(flt_rule_entry.rule.attrib));
if(tx_prop->tx[0].hdr_l2_type == IPA_HDR_L2_ETHERNET_II)
{
flt_rule_entry.rule.attrib.attrib_mask |= IPA_FLT_MAC_DST_ADDR_ETHER_II;
}
else
{
flt_rule_entry.rule.attrib.attrib_mask |= IPA_FLT_MAC_DST_ADDR_802_3;
}
memcpy(flt_rule_entry.rule.attrib.dst_mac_addr, mac, sizeof(flt_rule_entry.rule.attrib.dst_mac_addr));
memset(flt_rule_entry.rule.attrib.dst_mac_addr_mask, 0xFF, sizeof(flt_rule_entry.rule.attrib.dst_mac_addr_mask));
memcpy(&(pFilteringTable->rules[0]), &flt_rule_entry, sizeof(flt_rule_entry));
#ifdef IPA_IOCTL_SET_FNR_COUNTER_INFO
/* use index hw-counter */
if(ipa_if_cate == WLAN_IF && IPACM_Iface::ipacmcfg->hw_fnr_stats_support)
{
IPACMDBG_H("hw-index-enable %d, counter %d\n", IPACM_Iface::ipacmcfg->hw_fnr_stats_support, IPACM_Iface::ipacmcfg->hw_counter_offset + UL_ALL);
result = m_filtering.AddFilteringRuleAfter_hw_index(pFilteringTable, IPACM_Iface::ipacmcfg->hw_counter_offset + UL_ALL);
} else {
result = m_filtering.AddFilteringRuleAfter(pFilteringTable);
}
#else
result = m_filtering.AddFilteringRuleAfter(pFilteringTable);
#endif
if (result == false)
{
IPACMERR("Failed to add client filtering rules.\n");
res = IPACM_FAILURE;
goto end;
}
*flt_rule_hdl = pFilteringTable->rules[0].flt_rule_hdl;
end:
free(pFilteringTable);
}
else
{
IPACMDBG_H("Received client MAC 0x%02x%02x%02x%02x%02x%02x.\n", mac[0], mac[1], mac[2], mac[3], mac[4], mac[5]);
IPACMDBG_H("Not support rt_tbl_hdl %d flt_rule_hdl %p ip-type %d\n", rt_tbl_hdl, flt_rule_hdl, iptype);
}
return res;
}
int IPACM_Lan::eth_bridge_del_flt_rule(uint32_t flt_rule_hdl, ipa_ip_type iptype)
{
if(m_filtering.DeleteFilteringHdls(&flt_rule_hdl, iptype, 1) == false)
{
IPACMERR("Failed to delete the client specific flt rule.\n");
return IPACM_FAILURE;
}
return IPACM_SUCCESS;
}
int IPACM_Lan::eth_bridge_del_rt_rule(uint32_t rt_rule_hdl, ipa_ip_type iptype)
{
if(m_routing.DeleteRoutingHdl(rt_rule_hdl, iptype) == false)
{
IPACMERR("Failed to delete routing rule.\n");
return IPACM_FAILURE;
}
return IPACM_SUCCESS;
}
/* delete header processing context */
int IPACM_Lan::eth_bridge_del_hdr_proc_ctx(uint32_t hdr_proc_ctx_hdl)
{
if(m_header.DeleteHeaderProcCtx(hdr_proc_ctx_hdl) == false)
{
IPACMERR("Failed to delete hdr proc ctx.\n");
return IPACM_FAILURE;
}
return IPACM_SUCCESS;
}
#ifdef FEATURE_L2TP
/* check if the event is associated with vlan interface */
bool IPACM_Lan::is_vlan_event(char *event_iface_name)
{
int self_name_len, event_iface_name_len;
if(event_iface_name == NULL)
{
IPACMERR("Invalid input\n");
return false;
}
IPACMDBG_H("Self iface %s, event iface %s\n", dev_name, event_iface_name);
self_name_len = strlen(dev_name);
event_iface_name_len = strlen(event_iface_name);
if(event_iface_name_len > self_name_len && strncmp(dev_name, event_iface_name, self_name_len) == 0)
{
IPACMDBG_H("This is vlan event.\n");
return true;
}
return false;
}
/* check if the event is associated with l2tp interface */
bool IPACM_Lan::is_l2tp_event(char *event_iface_name)
{
if(event_iface_name == NULL)
{
IPACMERR("Invalid input\n");
return false;
}
IPACMDBG_H("Self iface %s, event iface %s\n", dev_name, event_iface_name);
if(strncmp(event_iface_name, "l2tp", 4) == 0)
{
IPACMDBG_H("This is l2tp event.\n");
return true;
}
return false;
}
/* add l2tp rt rule for l2tp client */
int IPACM_Lan::add_l2tp_rt_rule(ipa_ip_type iptype, uint8_t *dst_mac, ipa_hdr_l2_type peer_l2_hdr_type,
uint32_t l2tp_session_id, uint32_t vlan_id, uint8_t *vlan_client_mac, uint32_t *vlan_iface_ipv6_addr,
uint32_t *vlan_client_ipv6_addr, uint32_t *first_pass_hdr_hdl, uint32_t *first_pass_hdr_proc_ctx_hdl,
uint32_t *second_pass_hdr_hdl, int *num_rt_hdl, uint32_t *first_pass_rt_rule_hdl, uint32_t *second_pass_rt_rule_hdl)
{
int i, size, position;
uint32_t tx_index;
uint32_t vlan_iface_ipv6_addr_network[4], vlan_client_ipv6_addr_network[4];
ipa_ioc_add_hdr *hdr_table;
ipa_hdr_add *hdr;
ipa_ioc_add_hdr_proc_ctx *hdr_proc_ctx_table;
ipa_hdr_proc_ctx_add *hdr_proc_ctx;
ipa_ioc_add_rt_rule* rt_rule_table;
ipa_rt_rule_add *rt_rule;
ipa_ioc_copy_hdr copy_hdr;
if(tx_prop == NULL)
{
IPACMERR("No tx prop.\n");
return IPACM_FAILURE;
}
/* =========== install first pass hdr template (IPv6 + L2TP + inner ETH header = 62 bytes) ============= */
if(*first_pass_hdr_hdl != 0)
{
IPACMDBG_H("First pass hdr template was added before.\n");
}
else
{
size = sizeof(ipa_ioc_add_hdr) + sizeof(ipa_hdr_add);
hdr_table = (ipa_ioc_add_hdr*)malloc(size);
if(hdr_table == NULL)
{
IPACMERR("Failed to allocate memory.\n");
return IPACM_FAILURE;
}
memset(hdr_table, 0, size);
hdr_table->commit = 1;
hdr_table->num_hdrs = 1;
hdr = &hdr_table->hdr[0];
if(iptype == IPA_IP_v4)
{
snprintf(hdr->name, sizeof(hdr->name), "vlan_%d_l2tp_%d_v4", vlan_id, l2tp_session_id);
}
else
{
snprintf(hdr->name, sizeof(hdr->name), "vlan_%d_l2tp_%d_v6", vlan_id, l2tp_session_id);
}
hdr->hdr_len = 62;
hdr->type = IPA_HDR_L2_ETHERNET_II;
hdr->is_partial = 0;
hdr->hdr[0] = 0x60; /* version */
hdr->hdr[6] = 0x73; /* next header = L2TP */
hdr->hdr[7] = 0x40; /* hop limit = 64 */
for(i = 0; i < 4; i++)
{
vlan_iface_ipv6_addr_network[i] = htonl(vlan_iface_ipv6_addr[i]);
vlan_client_ipv6_addr_network[i] = htonl(vlan_client_ipv6_addr[i]);
}
memcpy(hdr->hdr + 8, vlan_iface_ipv6_addr_network, 16); /* source IPv6 addr */
memcpy(hdr->hdr + 24, vlan_client_ipv6_addr_network, 16); /* dest IPv6 addr */
hdr->hdr[43] = (uint8_t)(l2tp_session_id & 0xFF); /* l2tp header */
hdr->hdr[42] = (uint8_t)(l2tp_session_id >> 8 & 0xFF);
hdr->hdr[41] = (uint8_t)(l2tp_session_id >> 16 & 0xFF);
hdr->hdr[40] = (uint8_t)(l2tp_session_id >> 24 & 0xFF);
if(m_header.AddHeader(hdr_table) == false)
{
IPACMERR("Failed to add hdr with status: %d\n", hdr_table->hdr[0].status);
free(hdr_table);
return IPACM_FAILURE;
}
*first_pass_hdr_hdl = hdr_table->hdr[0].hdr_hdl;
IPACMDBG_H("Installed first pass hdr: hdl %d\n", *first_pass_hdr_hdl);
free(hdr_table);
}
/* =========== install first pass hdr proc ctx (populate src/dst MAC and Ether type) ============= */
size = sizeof(ipa_ioc_add_hdr_proc_ctx) + sizeof(ipa_hdr_proc_ctx_add);
hdr_proc_ctx_table = (ipa_ioc_add_hdr_proc_ctx*)malloc(size);
if(hdr_proc_ctx_table == NULL)
{
IPACMERR("Failed to allocate memory.\n");
return IPACM_FAILURE;
}
memset(hdr_proc_ctx_table, 0, size);
hdr_proc_ctx_table->commit = 1;
hdr_proc_ctx_table->num_proc_ctxs = 1;
hdr_proc_ctx = &hdr_proc_ctx_table->proc_ctx[0];
hdr_proc_ctx->type = IPA_HDR_PROC_L2TP_HEADER_ADD;
hdr_proc_ctx->hdr_hdl = *first_pass_hdr_hdl;
hdr_proc_ctx->l2tp_params.hdr_add_param.eth_hdr_retained = 1;
hdr_proc_ctx->l2tp_params.hdr_add_param.input_ip_version = iptype;
hdr_proc_ctx->l2tp_params.hdr_add_param.output_ip_version = IPA_IP_v6;
if(m_header.AddHeaderProcCtx(hdr_proc_ctx_table) == false)
{
IPACMERR("Failed to add hdr proc ctx with status: %d\n", hdr_proc_ctx_table->proc_ctx[0].status);
free(hdr_proc_ctx_table);
return IPACM_FAILURE;
}
*first_pass_hdr_proc_ctx_hdl = hdr_proc_ctx_table->proc_ctx[0].proc_ctx_hdl;
IPACMDBG_H("Installed first pass hdr proc ctx: hdl %d\n", *first_pass_hdr_proc_ctx_hdl);
free(hdr_proc_ctx_table);
/* =========== install first pass rt rules (match dst MAC then doing UCP) ============= */
*num_rt_hdl = each_client_rt_rule_count[iptype];
size = sizeof(ipa_ioc_add_rt_rule) + (*num_rt_hdl) * sizeof(ipa_rt_rule_add);
rt_rule_table = (ipa_ioc_add_rt_rule*)malloc(size);
if (rt_rule_table == NULL)
{
IPACMERR("Failed to allocate memory.\n");
return IPACM_FAILURE;
}
memset(rt_rule_table, 0, size);
rt_rule_table->commit = 1;
rt_rule_table->ip = iptype;
rt_rule_table->num_rules = *num_rt_hdl;
snprintf(rt_rule_table->rt_tbl_name, sizeof(rt_rule_table->rt_tbl_name), "l2tp");
rt_rule_table->rt_tbl_name[IPA_RESOURCE_NAME_MAX-1] = 0;
position = 0;
for(tx_index = 0; tx_index < iface_query->num_tx_props; tx_index++)
{
if(tx_prop->tx[tx_index].ip == iptype)
{
if(position >= *num_rt_hdl || position >= MAX_NUM_PROP)
{
IPACMERR("Number of routing rules already exceeds limit.\n");
free(rt_rule_table);
return IPACM_FAILURE;
}
rt_rule = &rt_rule_table->rules[position];
rt_rule->at_rear = false;
rt_rule->status = -1;
rt_rule->rt_rule_hdl = -1;
rt_rule->rule.hashable = false; //WLAN->ETH direction rules are set to non-hashable to keep consistent with the other direction
rt_rule->rule.hdr_hdl = 0;
rt_rule->rule.hdr_proc_ctx_hdl = *first_pass_hdr_proc_ctx_hdl;
rt_rule->rule.dst = IPA_CLIENT_DUMMY_CONS;
memcpy(&rt_rule->rule.attrib, &tx_prop->tx[tx_index].attrib, sizeof(rt_rule->rule.attrib));
if(peer_l2_hdr_type == IPA_HDR_L2_ETHERNET_II)
rt_rule->rule.attrib.attrib_mask |= IPA_FLT_MAC_DST_ADDR_ETHER_II;
else
rt_rule->rule.attrib.attrib_mask |= IPA_FLT_MAC_DST_ADDR_802_3;
memcpy(rt_rule->rule.attrib.dst_mac_addr, dst_mac, sizeof(rt_rule->rule.attrib.dst_mac_addr));
memset(rt_rule->rule.attrib.dst_mac_addr_mask, 0xFF, sizeof(rt_rule->rule.attrib.dst_mac_addr_mask));
position++;
}
}
if(m_routing.AddRoutingRule(rt_rule_table) == false)
{
IPACMERR("Failed to add first pass rt rules.\n");
free(rt_rule_table);
return IPACM_FAILURE;
}
for(i = 0; i < position; i++)
{
first_pass_rt_rule_hdl[i] = rt_rule_table->rules[i].rt_rule_hdl;
}
free(rt_rule_table);
/* =========== install second pass hdr (Ethernet header with L2TP tag = 18 bytes) ============= */
if(*second_pass_hdr_hdl != 0)
{
IPACMDBG_H("Second pass hdr was added before.\n");
}
else
{
size = sizeof(ipa_ioc_add_hdr) + sizeof(ipa_hdr_add);
hdr_table = (ipa_ioc_add_hdr*)malloc(size);
if(hdr_table == NULL)
{
IPACMERR("Failed to allocate memory.\n");
return IPACM_FAILURE;
}
memset(hdr_table, 0, size);
hdr_table->commit = 1;
hdr_table->num_hdrs = 1;
hdr = &hdr_table->hdr[0];
if(iptype == IPA_IP_v4)
{
snprintf(hdr->name, sizeof(hdr->name), "vlan_%d_v4", vlan_id);
}
else
{
snprintf(hdr->name, sizeof(hdr->name), "vlan_%d_v6", vlan_id);
}
hdr->type = IPA_HDR_L2_ETHERNET_II;
hdr->is_partial = 0;
for(tx_index = 0; tx_index < tx_prop->num_tx_props; tx_index++)
{
if(tx_prop->tx[tx_index].ip == IPA_IP_v6)
{
memset(&copy_hdr, 0, sizeof(copy_hdr));
strlcpy(copy_hdr.name, tx_prop->tx[tx_index].hdr_name,
sizeof(copy_hdr.name));
IPACMDBG_H("Header name: %s in tx:%d\n", copy_hdr.name, tx_index);
if(m_header.CopyHeader(&copy_hdr) == false)
{
IPACMERR("Failed to get partial header.\n");
free(hdr_table);
return IPACM_FAILURE;
}
IPACMDBG_H("Header length: %d\n", copy_hdr.hdr_len);
hdr->hdr_len = copy_hdr.hdr_len;
memcpy(hdr->hdr, copy_hdr.hdr, hdr->hdr_len);
break;
}
}
/* copy vlan client mac */
memcpy(hdr->hdr + hdr->hdr_len - 18, vlan_client_mac, 6);
hdr->hdr[hdr->hdr_len - 3] = (uint8_t)vlan_id & 0xFF;
hdr->hdr[hdr->hdr_len - 4] = (uint8_t)(vlan_id >> 8) & 0xFF;
if(m_header.AddHeader(hdr_table) == false)
{
IPACMERR("Failed to add hdr with status: %d\n", hdr->status);
free(hdr_table);
return IPACM_FAILURE;
}
*second_pass_hdr_hdl = hdr->hdr_hdl;
IPACMDBG_H("Installed second pass hdr: hdl %d\n", *second_pass_hdr_hdl);
free(hdr_table);
}
/* =========== install second pass rt rules (match VLAN interface IPv6 address at dst client side) ============= */
if(second_pass_rt_rule_hdl[0] != 0)
{
IPACMDBG_H("Second pass rt rule was added before, return.\n");
return IPACM_SUCCESS;
}
*num_rt_hdl = each_client_rt_rule_count[IPA_IP_v6];
size = sizeof(ipa_ioc_add_rt_rule) + (*num_rt_hdl) * sizeof(ipa_rt_rule_add);
rt_rule_table = (ipa_ioc_add_rt_rule*)malloc(size);
if (rt_rule_table == NULL)
{
IPACMERR("Failed to allocate memory.\n");
return IPACM_FAILURE;
}
memset(rt_rule_table, 0, size);
rt_rule_table->commit = 1;
rt_rule_table->ip = IPA_IP_v6;
rt_rule_table->num_rules = *num_rt_hdl;
snprintf(rt_rule_table->rt_tbl_name, sizeof(rt_rule_table->rt_tbl_name), "l2tp");
rt_rule_table->rt_tbl_name[IPA_RESOURCE_NAME_MAX-1] = 0;
position = 0;
for(tx_index = 0; tx_index < iface_query->num_tx_props; tx_index++)
{
if(tx_prop->tx[tx_index].ip == IPA_IP_v6)
{
if(position >= *num_rt_hdl || position >= MAX_NUM_PROP)
{
IPACMERR("Number of routing rules already exceeds limit.\n");
free(rt_rule_table);
return IPACM_FAILURE;
}
rt_rule = &rt_rule_table->rules[position];
rt_rule->at_rear = false;
rt_rule->status = -1;
rt_rule->rt_rule_hdl = -1;
rt_rule->rule.hashable = false; //WLAN->ETH direction rules are set to non-hashable to keep consistent with the other direction
rt_rule->rule.hdr_hdl = *second_pass_hdr_hdl;
rt_rule->rule.hdr_proc_ctx_hdl = 0;
rt_rule->rule.dst = tx_prop->tx[tx_index].dst_pipe;
memcpy(&rt_rule->rule.attrib, &tx_prop->tx[tx_index].attrib, sizeof(rt_rule->rule.attrib));
rt_rule->rule.attrib.attrib_mask |= IPA_FLT_DST_ADDR;
memcpy(rt_rule->rule.attrib.u.v6.dst_addr, vlan_client_ipv6_addr,
sizeof(rt_rule->rule.attrib.u.v6.dst_addr));
memset(rt_rule->rule.attrib.u.v6.dst_addr_mask, 0xFF, sizeof(rt_rule->rule.attrib.u.v6.dst_addr_mask));
position++;
}
}
if(m_routing.AddRoutingRule(rt_rule_table) == false)
{
IPACMERR("Failed to add second pass rt rules.\n");
free(rt_rule_table);
return IPACM_FAILURE;
}
for(i = 0; i < position; i++)
{
second_pass_rt_rule_hdl[i] = rt_rule_table->rules[i].rt_rule_hdl;
}
free(rt_rule_table);
return IPACM_SUCCESS;
}
/* delete l2tp rt rule for l2tp client */
int IPACM_Lan::del_l2tp_rt_rule(ipa_ip_type iptype, uint32_t first_pass_hdr_hdl, uint32_t first_pass_hdr_proc_ctx_hdl,
uint32_t second_pass_hdr_hdl, int num_rt_hdl, uint32_t *first_pass_rt_rule_hdl, uint32_t *second_pass_rt_rule_hdl)
{
int i;
if(num_rt_hdl < 0)
{
IPACMERR("Invalid num rt rule: %d\n", num_rt_hdl);
return IPACM_FAILURE;
}
for(i = 0; i < num_rt_hdl; i++)
{
if(first_pass_rt_rule_hdl != NULL)
{
if(m_routing.DeleteRoutingHdl(first_pass_rt_rule_hdl[i], iptype) == false)
{
return IPACM_FAILURE;
}
}
if(second_pass_rt_rule_hdl != NULL)
{
if(m_routing.DeleteRoutingHdl(second_pass_rt_rule_hdl[i], IPA_IP_v6) == false)
{
return IPACM_FAILURE;
}
}
}
if(first_pass_hdr_proc_ctx_hdl != 0)
{
if(m_header.DeleteHeaderProcCtx(first_pass_hdr_proc_ctx_hdl) == false)
{
return IPACM_FAILURE;
}
}
if(first_pass_hdr_hdl != 0)
{
if(m_header.DeleteHeaderHdl(first_pass_hdr_hdl) == false)
{
return IPACM_FAILURE;
}
}
if(second_pass_hdr_hdl != 0)
{
if(m_header.DeleteHeaderHdl(second_pass_hdr_hdl) == false)
{
return IPACM_FAILURE;
}
}
return IPACM_SUCCESS;
}
/* add l2tp rt rule for non l2tp client */
int IPACM_Lan::add_l2tp_rt_rule(ipa_ip_type iptype, uint8_t *dst_mac, uint32_t *hdr_proc_ctx_hdl,
int *num_rt_hdl, uint32_t *rt_rule_hdl)
{
int i, size, position;
uint32_t tx_index;
ipa_ioc_add_hdr_proc_ctx *hdr_proc_ctx_table;
ipa_hdr_proc_ctx_add *hdr_proc_ctx;
ipa_ioc_add_rt_rule* rt_rule_table;
ipa_rt_rule_add *rt_rule;
ipa_ioc_get_hdr hdr;
if(tx_prop == NULL)
{
IPACMERR("No tx prop.\n");
return IPACM_FAILURE;
}
memset(&hdr, 0, sizeof(hdr));
for(tx_index = 0; tx_index < tx_prop->num_tx_props; tx_index++)
{
if(tx_prop->tx[tx_index].ip == iptype)
{
strlcpy(hdr.name, tx_prop->tx[tx_index].hdr_name,
sizeof(hdr.name));
break;
}
}
if(m_header.GetHeaderHandle(&hdr) == false)
{
IPACMERR("Failed to get template hdr hdl.\n");
return IPACM_FAILURE;
}
/* =========== install hdr proc ctx (uc needs to remove IPv6 + L2TP + inner ETH header = 62 bytes) ============= */
if(*hdr_proc_ctx_hdl != 0)
{
IPACMDBG_H("Hdr proc ctx was added before.\n");
}
else
{
size = sizeof(ipa_ioc_add_hdr_proc_ctx) + sizeof(ipa_hdr_proc_ctx_add);
hdr_proc_ctx_table = (ipa_ioc_add_hdr_proc_ctx*)malloc(size);
if(hdr_proc_ctx_table == NULL)
{
IPACMERR("Failed to allocate memory.\n");
return IPACM_FAILURE;
}
memset(hdr_proc_ctx_table, 0, size);
hdr_proc_ctx_table->commit = 1;
hdr_proc_ctx_table->num_proc_ctxs = 1;
hdr_proc_ctx = &hdr_proc_ctx_table->proc_ctx[0];
hdr_proc_ctx->type = IPA_HDR_PROC_L2TP_HEADER_REMOVE;
hdr_proc_ctx->hdr_hdl = hdr.hdl;
hdr_proc_ctx->l2tp_params.hdr_remove_param.hdr_len_remove = 62;
hdr_proc_ctx->l2tp_params.hdr_remove_param.eth_hdr_retained = 1;
hdr_proc_ctx->l2tp_params.is_dst_pipe_valid = 1;
hdr_proc_ctx->l2tp_params.dst_pipe = tx_prop->tx[0].dst_pipe;
IPACMDBG_H("Header_remove: hdr len %d, hdr retained %d, dst client: %d\n",
hdr_proc_ctx->l2tp_params.hdr_remove_param.hdr_len_remove,
hdr_proc_ctx->l2tp_params.hdr_remove_param.eth_hdr_retained,
hdr_proc_ctx->l2tp_params.dst_pipe);
if(m_header.AddHeaderProcCtx(hdr_proc_ctx_table) == false)
{
IPACMERR("Failed to add hdr proc ctx with status: %d\n", hdr_proc_ctx_table->proc_ctx[0].status);
free(hdr_proc_ctx_table);
return IPACM_FAILURE;
}
*hdr_proc_ctx_hdl = hdr_proc_ctx_table->proc_ctx[0].proc_ctx_hdl;
IPACMDBG_H("Installed hdr proc ctx: hdl %d\n", *hdr_proc_ctx_hdl);
free(hdr_proc_ctx_table);
}
/* =========== install rt rules (match dst MAC within 62 bytes header) ============= */
*num_rt_hdl = each_client_rt_rule_count[iptype];
size = sizeof(ipa_ioc_add_rt_rule) + (*num_rt_hdl) * sizeof(ipa_rt_rule_add);
rt_rule_table = (ipa_ioc_add_rt_rule*)malloc(size);
if (rt_rule_table == NULL)
{
IPACMERR("Failed to allocate memory.\n");
return IPACM_FAILURE;
}
memset(rt_rule_table, 0, size);
rt_rule_table->commit = 1;
rt_rule_table->ip = iptype;
rt_rule_table->num_rules = *num_rt_hdl;
snprintf(rt_rule_table->rt_tbl_name, sizeof(rt_rule_table->rt_tbl_name), "l2tp");
rt_rule_table->rt_tbl_name[IPA_RESOURCE_NAME_MAX-1] = 0;
position = 0;
for(tx_index = 0; tx_index < iface_query->num_tx_props; tx_index++)
{
if(tx_prop->tx[tx_index].ip == iptype)
{
if(position >= *num_rt_hdl || position >= MAX_NUM_PROP)
{
IPACMERR("Number of routing rules already exceeds limit.\n");
free(rt_rule_table);
return IPACM_FAILURE;
}
rt_rule = &rt_rule_table->rules[position];
rt_rule->at_rear = false;
rt_rule->status = -1;
rt_rule->rt_rule_hdl = -1;
rt_rule->rule.hashable = false; //ETH->WLAN direction rules need to be non-hashable due to encapsulation
rt_rule->rule.hdr_hdl = 0;
rt_rule->rule.hdr_proc_ctx_hdl = *hdr_proc_ctx_hdl;
rt_rule->rule.dst = tx_prop->tx[tx_index].dst_pipe;
memcpy(&rt_rule->rule.attrib, &tx_prop->tx[tx_index].attrib, sizeof(rt_rule->rule.attrib));
rt_rule->rule.attrib.attrib_mask |= IPA_FLT_MAC_DST_ADDR_L2TP;
memset(rt_rule->rule.attrib.dst_mac_addr_mask, 0xFF, sizeof(rt_rule->rule.attrib.dst_mac_addr_mask));
memcpy(rt_rule->rule.attrib.dst_mac_addr, dst_mac, sizeof(rt_rule->rule.attrib.dst_mac_addr));
position++;
}
}
if(m_routing.AddRoutingRule(rt_rule_table) == false)
{
IPACMERR("Failed to add first pass rt rules.\n");
free(rt_rule_table);
return IPACM_FAILURE;
}
for(i = 0; i < position; i++)
rt_rule_hdl[i] = rt_rule_table->rules[i].rt_rule_hdl;
free(rt_rule_table);
return IPACM_SUCCESS;
}
int IPACM_Lan::del_l2tp_rt_rule(ipa_ip_type iptype, int num_rt_hdl, uint32_t *rt_rule_hdl)
{
int i;
if(num_rt_hdl < 0)
{
IPACMERR("Invalid num rt rule: %d\n", num_rt_hdl);
return IPACM_FAILURE;
}
for(i = 0; i < num_rt_hdl; i++)
{
if(m_routing.DeleteRoutingHdl(rt_rule_hdl[i], iptype) == false)
{
return IPACM_FAILURE;
}
}
return IPACM_SUCCESS;
}
/* add l2tp flt rule on l2tp interface */
int IPACM_Lan::add_l2tp_flt_rule(uint8_t *dst_mac, uint32_t *flt_rule_hdl)
{
int len;
int fd_ipa;
struct ipa_flt_rule_add flt_rule_entry;
struct ipa_ioc_add_flt_rule_after *pFilteringTable = NULL;
ipa_ioc_get_rt_tbl rt_tbl;
if (IPACM_Iface::ipacmcfg->isIPAv3Supported())
{
if (rx_prop == NULL || tx_prop == NULL)
{
IPACMDBG_H("No rx or tx properties registered for iface %s\n", dev_name);
return IPACM_FAILURE;
}
len = sizeof(struct ipa_ioc_add_flt_rule_after) + sizeof(struct ipa_flt_rule_add);
pFilteringTable = (struct ipa_ioc_add_flt_rule_after*)malloc(len);
if (!pFilteringTable)
{
IPACMERR("Failed to allocate ipa_ioc_add_flt_rule_after memory...\n");
return IPACM_FAILURE;
}
memset(pFilteringTable, 0, len);
pFilteringTable->commit = 1;
pFilteringTable->ep = rx_prop->rx[0].src_pipe;
pFilteringTable->ip = IPA_IP_v6;
pFilteringTable->num_rules = 1;
pFilteringTable->add_after_hdl = eth_bridge_flt_rule_offset[IPA_IP_v6];
fd_ipa = open(IPA_DEVICE_NAME, O_RDWR);
if(fd_ipa == 0)
{
IPACMERR("Failed to open %s\n",IPA_DEVICE_NAME);
free(pFilteringTable);
return IPACM_FAILURE;
}
rt_tbl.ip = IPA_IP_v6;
snprintf(rt_tbl.name, sizeof(rt_tbl.name), "l2tp");
rt_tbl.name[IPA_RESOURCE_NAME_MAX-1] = '\0';
IPACMDBG_H("This flt rule points to rt tbl %s.\n", rt_tbl.name);
if(m_routing.GetRoutingTable(&rt_tbl) == false)
{
IPACMERR("Failed to get routing table from name\n");
free(pFilteringTable);
close(fd_ipa);
return IPACM_FAILURE;
}
memset(&flt_rule_entry, 0, sizeof(flt_rule_entry));
flt_rule_entry.at_rear = 1;
flt_rule_entry.rule.retain_hdr = 0;
flt_rule_entry.rule.to_uc = 0;
flt_rule_entry.rule.action = IPA_PASS_TO_ROUTING;
flt_rule_entry.rule.eq_attrib_type = 0;
flt_rule_entry.rule.rt_tbl_hdl = rt_tbl.hdl;
flt_rule_entry.rule.hashable = false; //ETH->WLAN direction rules need to be non-hashable due to encapsulation
memcpy(&flt_rule_entry.rule.attrib, &rx_prop->rx[0].attrib, sizeof(flt_rule_entry.rule.attrib));
/* flt rule is matching dst MAC within 62 bytes header */
flt_rule_entry.rule.attrib.attrib_mask |= IPA_FLT_MAC_DST_ADDR_L2TP;
memset(flt_rule_entry.rule.attrib.dst_mac_addr_mask, 0xFF, sizeof(flt_rule_entry.rule.attrib.dst_mac_addr_mask));
memcpy(flt_rule_entry.rule.attrib.dst_mac_addr, dst_mac, sizeof(flt_rule_entry.rule.attrib.dst_mac_addr));
memcpy(&(pFilteringTable->rules[0]), &flt_rule_entry, sizeof(flt_rule_entry));
if(m_filtering.AddFilteringRuleAfter(pFilteringTable) == false)
{
IPACMERR("Failed to add client filtering rules.\n");
free(pFilteringTable);
close(fd_ipa);
return IPACM_FAILURE;
}
*flt_rule_hdl = pFilteringTable->rules[0].flt_rule_hdl;
free(pFilteringTable);
close(fd_ipa);
}
return IPACM_SUCCESS;
}
/* delete l2tp flt rule on l2tp interface */
int IPACM_Lan::del_l2tp_flt_rule(uint32_t flt_rule_hdl)
{
if(m_filtering.DeleteFilteringHdls(&flt_rule_hdl, IPA_IP_v6, 1) == false)
{
return IPACM_FAILURE;
}
return IPACM_SUCCESS;
}
/* add l2tp flt rule on non l2tp interface */
int IPACM_Lan::add_l2tp_flt_rule(ipa_ip_type iptype, uint8_t *dst_mac, uint32_t *vlan_client_ipv6_addr,
uint32_t *first_pass_flt_rule_hdl, uint32_t *second_pass_flt_rule_hdl)
{
int len;
struct ipa_flt_rule_add flt_rule_entry;
struct ipa_ioc_add_flt_rule_after *pFilteringTable = NULL;
ipa_ioc_get_rt_tbl rt_tbl;
if (IPACM_Iface::ipacmcfg->isIPAv3Supported())
{
if (rx_prop == NULL || tx_prop == NULL)
{
IPACMDBG_H("No rx or tx properties registered for iface %s\n", dev_name);
return IPACM_FAILURE;
}
IPACMDBG_H("Dst client MAC 0x%02x%02x%02x%02x%02x%02x.\n", dst_mac[0], dst_mac[1],
dst_mac[2], dst_mac[3], dst_mac[4], dst_mac[5]);
len = sizeof(struct ipa_ioc_add_flt_rule_after) + sizeof(struct ipa_flt_rule_add);
pFilteringTable = (struct ipa_ioc_add_flt_rule_after*)malloc(len);
if (!pFilteringTable)
{
IPACMERR("Failed to allocate ipa_ioc_add_flt_rule_after memory...\n");
return IPACM_FAILURE;
}
memset(pFilteringTable, 0, len);
pFilteringTable->commit = 1;
pFilteringTable->ep = rx_prop->rx[0].src_pipe;
pFilteringTable->ip = iptype;
pFilteringTable->num_rules = 1;
pFilteringTable->add_after_hdl = eth_bridge_flt_rule_offset[iptype];
/* =========== add first pass flt rule (match dst MAC) ============= */
rt_tbl.ip = iptype;
snprintf(rt_tbl.name, sizeof(rt_tbl.name), "l2tp");
IPACMDBG_H("This flt rule points to rt tbl %s.\n", rt_tbl.name);
if(m_routing.GetRoutingTable(&rt_tbl) == false)
{
IPACMERR("Failed to get routing table.\n");
return IPACM_FAILURE;
}
memset(&flt_rule_entry, 0, sizeof(flt_rule_entry));
flt_rule_entry.at_rear = 1;
flt_rule_entry.rule.retain_hdr = 0;
flt_rule_entry.rule.to_uc = 0;
flt_rule_entry.rule.action = IPA_PASS_TO_ROUTING;
flt_rule_entry.rule.eq_attrib_type = 0;
flt_rule_entry.rule.rt_tbl_hdl = rt_tbl.hdl;
flt_rule_entry.rule.hashable = false; //WLAN->ETH direction rules are set to non-hashable to keep consistent with the other direction
memcpy(&flt_rule_entry.rule.attrib, &rx_prop->rx[0].attrib, sizeof(flt_rule_entry.rule.attrib));
if(tx_prop->tx[0].hdr_l2_type == IPA_HDR_L2_ETHERNET_II)
{
flt_rule_entry.rule.attrib.attrib_mask |= IPA_FLT_MAC_DST_ADDR_ETHER_II;
}
else
{
flt_rule_entry.rule.attrib.attrib_mask |= IPA_FLT_MAC_DST_ADDR_802_3;
}
memcpy(flt_rule_entry.rule.attrib.dst_mac_addr, dst_mac, sizeof(flt_rule_entry.rule.attrib.dst_mac_addr));
memset(flt_rule_entry.rule.attrib.dst_mac_addr_mask, 0xFF, sizeof(flt_rule_entry.rule.attrib.dst_mac_addr_mask));
memcpy(&(pFilteringTable->rules[0]), &flt_rule_entry, sizeof(flt_rule_entry));
if (false == m_filtering.AddFilteringRuleAfter(pFilteringTable))
{
IPACMERR("Failed to add first pass filtering rules.\n");
free(pFilteringTable);
return IPACM_FAILURE;
}
*first_pass_flt_rule_hdl = pFilteringTable->rules[0].flt_rule_hdl;
/* =========== add second pass flt rule (match VLAN interface IPv6 address at client side) ============= */
if(*second_pass_flt_rule_hdl != 0)
{
IPACMDBG_H("Second pass flt rule was added before, return.\n");
free(pFilteringTable);
return IPACM_SUCCESS;
}
rt_tbl.ip = IPA_IP_v6;
snprintf(rt_tbl.name, sizeof(rt_tbl.name), "l2tp");
IPACMDBG_H("This flt rule points to rt tbl %s.\n", rt_tbl.name);
if(m_routing.GetRoutingTable(&rt_tbl) == false)
{
IPACMERR("Failed to get routing table.\n");
return IPACM_FAILURE;
}
pFilteringTable->ip = IPA_IP_v6;
pFilteringTable->add_after_hdl = eth_bridge_flt_rule_offset[IPA_IP_v6];
memset(&flt_rule_entry, 0, sizeof(flt_rule_entry));
flt_rule_entry.at_rear = 1;
flt_rule_entry.rule.retain_hdr = 0;
flt_rule_entry.rule.to_uc = 0;
flt_rule_entry.rule.action = IPA_PASS_TO_ROUTING;
flt_rule_entry.rule.eq_attrib_type = 0;
flt_rule_entry.rule.rt_tbl_hdl = rt_tbl.hdl;
flt_rule_entry.rule.hashable = false; //WLAN->ETH direction rules are set to non-hashable to keep consistent with the other direction
memcpy(&flt_rule_entry.rule.attrib, &rx_prop->rx[0].attrib, sizeof(flt_rule_entry.rule.attrib));
flt_rule_entry.rule.attrib.attrib_mask = IPA_FLT_DST_ADDR;
memcpy(flt_rule_entry.rule.attrib.u.v6.dst_addr, vlan_client_ipv6_addr, sizeof(flt_rule_entry.rule.attrib.u.v6.dst_addr));
memset(flt_rule_entry.rule.attrib.u.v6.dst_addr_mask, 0xFF, sizeof(flt_rule_entry.rule.attrib.u.v6.dst_addr_mask));
memcpy(&(pFilteringTable->rules[0]), &flt_rule_entry, sizeof(flt_rule_entry));
if (false == m_filtering.AddFilteringRuleAfter(pFilteringTable))
{
IPACMERR("Failed to add client filtering rules.\n");
free(pFilteringTable);
return IPACM_FAILURE;
}
*second_pass_flt_rule_hdl = pFilteringTable->rules[0].flt_rule_hdl;
free(pFilteringTable);
}
return IPACM_SUCCESS;
}
/* delete l2tp flt rule on non l2tp interface */
int IPACM_Lan::del_l2tp_flt_rule(ipa_ip_type iptype, uint32_t first_pass_flt_rule_hdl, uint32_t second_pass_flt_rule_hdl)
{
if(first_pass_flt_rule_hdl != 0)
{
if(m_filtering.DeleteFilteringHdls(&first_pass_flt_rule_hdl, iptype, 1) == false)
{
return IPACM_FAILURE;
}
}
if(second_pass_flt_rule_hdl != 0)
{
if(m_filtering.DeleteFilteringHdls(&second_pass_flt_rule_hdl, iptype, 1) == false)
{
return IPACM_FAILURE;
}
}
return IPACM_SUCCESS;
}
bool IPACM_Lan::is_unique_local_ipv6_addr(uint32_t* ipv6_addr)
{
uint32_t ipv6_unique_local_prefix, ipv6_unique_local_prefix_mask;
if(ipv6_addr == NULL)
{
IPACMERR("IPv6 address is empty.\n");
return false;
}
IPACMDBG_H("Get ipv6 address with first word 0x%08x.\n", ipv6_addr[0]);
ipv6_unique_local_prefix = 0xFD000000;
ipv6_unique_local_prefix_mask = 0xFF000000;
if((ipv6_addr[0] & ipv6_unique_local_prefix_mask) == (ipv6_unique_local_prefix & ipv6_unique_local_prefix_mask))
{
IPACMDBG_H("This IPv6 address is unique local IPv6 address.\n");
return true;
}
return false;
}
#endif
/* add tcp syn flt rule */
int IPACM_Lan::add_tcp_syn_flt_rule(ipa_ip_type iptype)
{
int len;
struct ipa_flt_rule_add flt_rule_entry;
ipa_ioc_add_flt_rule *m_pFilteringTable;
bool result;
if(rx_prop == NULL)
{
IPACMDBG_H("No rx properties registered for iface %s\n", dev_name);
return IPACM_SUCCESS;
}
len = sizeof(struct ipa_ioc_add_flt_rule) + sizeof(struct ipa_flt_rule_add);
m_pFilteringTable = (struct ipa_ioc_add_flt_rule *)malloc(len);
if(!m_pFilteringTable)
{
PERROR("Not enough memory.\n");
return IPACM_FAILURE;
}
memset(m_pFilteringTable, 0, len);
m_pFilteringTable->commit = 1;
m_pFilteringTable->ep = rx_prop->rx[0].src_pipe;
m_pFilteringTable->global = false;
m_pFilteringTable->ip = iptype;
m_pFilteringTable->num_rules = 1;
memset(&flt_rule_entry, 0, sizeof(flt_rule_entry));
flt_rule_entry.at_rear = true;
flt_rule_entry.rule.retain_hdr = 1;
flt_rule_entry.flt_rule_hdl = -1;
flt_rule_entry.status = -1;
flt_rule_entry.rule.action = IPA_PASS_TO_EXCEPTION;
memcpy(&flt_rule_entry.rule.attrib, &rx_prop->rx[0].attrib,
sizeof(flt_rule_entry.rule.attrib));
flt_rule_entry.rule.attrib.attrib_mask |= IPA_FLT_TCP_SYN;
if(iptype == IPA_IP_v4)
{
flt_rule_entry.rule.attrib.attrib_mask |= IPA_FLT_PROTOCOL;
flt_rule_entry.rule.attrib.u.v4.protocol = 6;
}
else
{
flt_rule_entry.rule.attrib.attrib_mask |= IPA_FLT_NEXT_HDR;
flt_rule_entry.rule.attrib.u.v6.next_hdr = 6;
}
memcpy(&(m_pFilteringTable->rules[0]), &flt_rule_entry, sizeof(flt_rule_entry));
#ifdef IPA_IOCTL_SET_FNR_COUNTER_INFO
/* use index hw-counter */
if(ipa_if_cate == WLAN_IF && IPACM_Iface::ipacmcfg->hw_fnr_stats_support)
{
IPACMDBG_H("hw-index-enable %d, counter %d\n", IPACM_Iface::ipacmcfg->hw_fnr_stats_support, IPACM_Iface::ipacmcfg->hw_counter_offset + UL_ALL);
result = m_filtering.AddFilteringRule_hw_index(m_pFilteringTable, IPACM_Iface::ipacmcfg->hw_counter_offset + UL_ALL);
} else {
result = m_filtering.AddFilteringRule(m_pFilteringTable);
}
#else
result = m_filtering.AddFilteringRule(m_pFilteringTable);
#endif
if(result == false)
{
IPACMERR("Error Adding RuleTable(0) to Filtering, aborting...\n");
free(m_pFilteringTable);
return IPACM_FAILURE;
}
tcp_syn_flt_rule_hdl[iptype] = m_pFilteringTable->rules[0].flt_rule_hdl;
free(m_pFilteringTable);
return IPACM_SUCCESS;
}
/* add tcp syn flt rule for l2tp interface*/
int IPACM_Lan::add_tcp_syn_flt_rule_l2tp(ipa_ip_type inner_ip_type)
{
int len;
struct ipa_flt_rule_add flt_rule_entry;
ipa_ioc_add_flt_rule *m_pFilteringTable;
if(rx_prop == NULL)
{
IPACMDBG_H("No rx properties registered for iface %s\n", dev_name);
return IPACM_SUCCESS;
}
len = sizeof(struct ipa_ioc_add_flt_rule) + sizeof(struct ipa_flt_rule_add);
m_pFilteringTable = (struct ipa_ioc_add_flt_rule *)malloc(len);
if(!m_pFilteringTable)
{
PERROR("Not enough memory.\n");
return IPACM_FAILURE;
}
memset(m_pFilteringTable, 0, len);
m_pFilteringTable->commit = 1;
m_pFilteringTable->ep = rx_prop->rx[0].src_pipe;
m_pFilteringTable->global = false;
m_pFilteringTable->ip = IPA_IP_v6;
m_pFilteringTable->num_rules = 1;
memset(&flt_rule_entry, 0, sizeof(flt_rule_entry));
flt_rule_entry.at_rear = true;
flt_rule_entry.rule.retain_hdr = 1;
flt_rule_entry.flt_rule_hdl = -1;
flt_rule_entry.status = -1;
flt_rule_entry.rule.action = IPA_PASS_TO_EXCEPTION;
memcpy(&flt_rule_entry.rule.attrib, &rx_prop->rx[0].attrib,
sizeof(flt_rule_entry.rule.attrib));
flt_rule_entry.rule.attrib.attrib_mask |= IPA_FLT_TCP_SYN_L2TP;
if(inner_ip_type == IPA_IP_v4)
{
flt_rule_entry.rule.attrib.ether_type = 0x0800;
}
else
{
flt_rule_entry.rule.attrib.ether_type = 0x86dd;
}
memcpy(&(m_pFilteringTable->rules[0]), &flt_rule_entry, sizeof(flt_rule_entry));
/* no need for hw counters */
if(false == m_filtering.AddFilteringRule(m_pFilteringTable))
{
IPACMERR("Error Adding RuleTable(0) to Filtering, aborting...\n");
free(m_pFilteringTable);
return IPACM_FAILURE;
}
tcp_syn_flt_rule_hdl[inner_ip_type] = m_pFilteringTable->rules[0].flt_rule_hdl;
free(m_pFilteringTable);
return IPACM_SUCCESS;
}
int IPACM_Lan::add_connection(int client_index, int v6_num)
{
int len, res = IPACM_SUCCESS;
uint8_t mux_id;
ipa_ioc_add_flt_rule *pFilteringTable = NULL;
int fd;
mux_id = IPACM_Iface::ipacmcfg->GetQmapId();
/* contruct filter rules to pcie modem */
struct ipa_flt_rule_add flt_rule_entry;
ipa_ioc_generate_flt_eq flt_eq;
IPACMDBG("\n");
len = sizeof(struct ipa_ioc_add_flt_rule) + sizeof(struct ipa_flt_rule_add);
pFilteringTable = (struct ipa_ioc_add_flt_rule*)malloc(len);
if (pFilteringTable == NULL)
{
IPACMERR("Error Locate ipa_flt_rule_add memory...\n");
return IPACM_FAILURE;
}
memset(pFilteringTable, 0, len);
pFilteringTable->commit = 1;
pFilteringTable->global = false;
pFilteringTable->ip = IPA_IP_v6;
pFilteringTable->num_rules = (uint8_t)1;
/* Configuring Filtering Rule */
memset(&flt_rule_entry, 0, sizeof(struct ipa_flt_rule_add));
flt_rule_entry.at_rear = true;
flt_rule_entry.flt_rule_hdl = -1;
flt_rule_entry.status = -1;
flt_rule_entry.rule.retain_hdr = 1;
flt_rule_entry.rule.to_uc = 0;
flt_rule_entry.rule.eq_attrib_type = 1;
flt_rule_entry.rule.action = IPA_PASS_TO_ROUTING;
if (IPACM_Iface::ipacmcfg->isIPAv3Supported())
flt_rule_entry.rule.hashable = true;
flt_rule_entry.rule.attrib.attrib_mask |= IPA_FLT_DST_ADDR;
flt_rule_entry.rule.attrib.u.v6.dst_addr[0] = get_client_memptr(eth_client, client_index)->v6_addr[v6_num][0];
flt_rule_entry.rule.attrib.u.v6.dst_addr[1] = get_client_memptr(eth_client, client_index)->v6_addr[v6_num][1];
flt_rule_entry.rule.attrib.u.v6.dst_addr[2] = get_client_memptr(eth_client, client_index)->v6_addr[v6_num][2];
flt_rule_entry.rule.attrib.u.v6.dst_addr[3] = get_client_memptr(eth_client, client_index)->v6_addr[v6_num][3];
flt_rule_entry.rule.attrib.u.v6.dst_addr_mask[0] = 0xFFFFFFFF;
flt_rule_entry.rule.attrib.u.v6.dst_addr_mask[1] = 0xFFFFFFFF;
flt_rule_entry.rule.attrib.u.v6.dst_addr_mask[2] = 0xFFFFFFFF;
flt_rule_entry.rule.attrib.u.v6.dst_addr_mask[3] = 0xFFFFFFFF;
IPACMDBG_H("ipv6 address got: 0x%x:%x:%x:%x\n", get_client_memptr(eth_client, client_index)->v6_addr[v6_num][0],
get_client_memptr(eth_client, client_index)->v6_addr[v6_num][1],
get_client_memptr(eth_client, client_index)->v6_addr[v6_num][2],
get_client_memptr(eth_client, client_index)->v6_addr[v6_num][3]);
/* change to network order for modem */
change_to_network_order(IPA_IP_v6, &flt_rule_entry.rule.attrib);
memset(&flt_eq, 0, sizeof(flt_eq));
memcpy(&flt_eq.attrib, &flt_rule_entry.rule.attrib, sizeof(flt_eq.attrib));
flt_eq.ip = IPA_IP_v6;
fd = open(IPA_DEVICE_NAME, O_RDWR);
if (fd < 0)
{
IPACMERR("Failed opening %s.\n", IPA_DEVICE_NAME);
free(pFilteringTable);
return IPACM_FAILURE;
}
if(0 != ioctl(fd, IPA_IOC_GENERATE_FLT_EQ, &flt_eq))
{
IPACMERR("Failed to get eq_attrib\n");
res = IPACM_FAILURE;
goto fail;
}
memcpy(&flt_rule_entry.rule.eq_attrib,
&flt_eq.eq_attrib,
sizeof(flt_rule_entry.rule.eq_attrib));
memcpy(&(pFilteringTable->rules[0]), &flt_rule_entry, sizeof(struct ipa_flt_rule_add));
if(false == m_filtering.AddOffloadFilteringRule(pFilteringTable, mux_id, 0))
{
IPACMERR("Failed to install WAN DL filtering table.\n");
res = IPACM_FAILURE;
goto fail;
}
get_client_memptr(eth_client, client_index)->v6_rt_rule_id[v6_num] = pFilteringTable->rules[0].flt_rule_hdl;
IPACMDBG_H("%d-st client v6_num %d: id handle 0x%x\n", client_index, v6_num, get_client_memptr(eth_client, client_index)->v6_rt_rule_id[v6_num]);
fail:
close(fd);
if(pFilteringTable != NULL)
{
free(pFilteringTable);
}
return res;
}
int IPACM_Lan::del_connection(int client_index, int v6_num)
{
int len, res = IPACM_SUCCESS;
ipa_ioc_del_flt_rule *pFilteringTable = NULL;
struct ipa_flt_rule_del flt_rule_entry;
IPACMDBG("\n");
len = sizeof(struct ipa_ioc_del_flt_rule) + sizeof(struct ipa_flt_rule_del);
pFilteringTable = (struct ipa_ioc_del_flt_rule*)malloc(len);
if (pFilteringTable == NULL)
{
IPACMERR("Error Locate ipa_ioc_del_flt_rule memory...\n");
return IPACM_FAILURE;
}
memset(pFilteringTable, 0, len);
pFilteringTable->commit = 1;
pFilteringTable->ip = IPA_IP_v6;
pFilteringTable->num_hdls = (uint8_t)1;
/* Configuring Software-Routing Filtering Rule */
memset(&flt_rule_entry, 0, sizeof(struct ipa_flt_rule_del));
flt_rule_entry.hdl = get_client_memptr(eth_client, client_index)->v6_rt_rule_id[v6_num];
memcpy(&(pFilteringTable->hdl[0]), &flt_rule_entry, sizeof(struct ipa_flt_rule_del));
if(false == m_filtering.DelOffloadFilteringRule(pFilteringTable))
{
IPACMERR("Failed to install WAN DL filtering table.\n");
res = IPACM_FAILURE;
goto fail;
}
get_client_memptr(eth_client, client_index)->v6_rt_rule_id[v6_num] = 0;
fail:
if(pFilteringTable != NULL)
{
free(pFilteringTable);
}
return res;
}
int IPACM_Lan::construct_mtu_rule(struct ipa_flt_rule *rule, ipa_ip_type iptype, uint16_t mtu)
{
int res = IPACM_SUCCESS;
int fd;
ipa_ioc_generate_flt_eq flt_eq;
if (rule == NULL)
{
IPACMERR("rule is empty");
return IPACM_FAILURE;
}
if (mtu == 0)
{
IPACMERR("mtu is uninitialized");
return IPACM_FAILURE;
}
IPACMDBG_H("Adding MTU rule for iptype = %d\n", iptype);
rule->eq_attrib_type = 1;
rule->eq_attrib.rule_eq_bitmap = 0;
rule->action = IPA_PASS_TO_EXCEPTION;
/* generate eq */
memset(&flt_eq, 0, sizeof(flt_eq));
memcpy(&flt_eq.attrib, &rule->attrib, sizeof(flt_eq.attrib));
flt_eq.ip = iptype;
fd = open(IPA_DEVICE_NAME, O_RDWR);
if (fd < 0)
{
IPACMERR("Failed opening %s.\n", IPA_DEVICE_NAME);
return IPACM_FAILURE;
}
if (0 != ioctl(fd, IPA_IOC_GENERATE_FLT_EQ, &flt_eq)) //define and cpy attribute to this struct
{
IPACMERR("Failed to get eq_attrib\n");
res = IPACM_FAILURE;
goto fail;
}
memcpy(&rule->eq_attrib,
&flt_eq.eq_attrib, sizeof(rule->eq_attrib));
//add IHL offsets
rule->eq_attrib.rule_eq_bitmap |= (1<<10);
rule->eq_attrib.num_ihl_offset_range_16 = 1;
if (iptype == IPA_IP_v4)
rule->eq_attrib.ihl_offset_range_16[0].offset = 0x82;
else
rule->eq_attrib.ihl_offset_range_16[0].offset = 0x84;
rule->eq_attrib.ihl_offset_range_16[0].range_low = mtu + 1;
rule->eq_attrib.ihl_offset_range_16[0].range_high = UINT16_MAX; //0xFFFF
fail:
close(fd);
return res;
}