URL
https://opencores.org/ocsvn/openrisc/openrisc/trunk
Subversion Repositories openrisc
[/] [openrisc/] [trunk/] [rtos/] [freertos-6.1.1/] [Demo/] [Common/] [ethernet/] [lwIP_130/] [src/] [core/] [ipv4/] [igmp.c] - Rev 606
Compare with Previous | Blame | View Log
/** * @file * IGMP - Internet Group Management Protocol * */ /* * Copyright (c) 2002 CITEL Technologies Ltd. * All rights reserved. * * Redistribution and use in source and binary forms, with or without * modification, are permitted provided that the following conditions * are met: * 1. Redistributions of source code must retain the above copyright * notice, this list of conditions and the following disclaimer. * 2. 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. * 3. Neither the name of CITEL Technologies Ltd 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 BY CITEL TECHNOLOGIES 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 CITEL TECHNOLOGIES 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. * * This file is a contribution to the lwIP TCP/IP stack. * The Swedish Institute of Computer Science and Adam Dunkels * are specifically granted permission to redistribute this * source code. */ /*------------------------------------------------------------- Note 1) Although the rfc requires V1 AND V2 capability we will only support v2 since now V1 is very old (August 1989) V1 can be added if required a debug print and statistic have been implemented to show this up. ------------------------------------------------------------- ------------------------------------------------------------- Note 2) A query for a specific group address (as opposed to ALLHOSTS) has now been implemented as I am unsure if it is required a debug print and statistic have been implemented to show this up. ------------------------------------------------------------- ------------------------------------------------------------- Note 3) The router alert rfc 2113 is implemented in outgoing packets but not checked rigorously incoming ------------------------------------------------------------- Steve Reynolds ------------------------------------------------------------*/ /*----------------------------------------------------------------------------- * RFC 988 - Host extensions for IP multicasting - V0 * RFC 1054 - Host extensions for IP multicasting - * RFC 1112 - Host extensions for IP multicasting - V1 * RFC 2236 - Internet Group Management Protocol, Version 2 - V2 <- this code is based on this RFC (it's the "de facto" standard) * RFC 3376 - Internet Group Management Protocol, Version 3 - V3 * RFC 4604 - Using Internet Group Management Protocol Version 3... - V3+ * RFC 2113 - IP Router Alert Option - *----------------------------------------------------------------------------*/ /*----------------------------------------------------------------------------- * Includes *----------------------------------------------------------------------------*/ #include "lwip/opt.h" #if LWIP_IGMP /* don't build if not configured for use in lwipopts.h */ #include "lwip/igmp.h" #include "lwip/debug.h" #include "lwip/def.h" #include "lwip/mem.h" #include "lwip/ip.h" #include "lwip/inet.h" #include "lwip/inet_chksum.h" #include "lwip/netif.h" #include "lwip/icmp.h" #include "lwip/udp.h" #include "lwip/tcp.h" #include "lwip/stats.h" #include "string.h" /*----------------------------------------------------------------------------- * Globales *----------------------------------------------------------------------------*/ static struct igmp_group* igmp_group_list; static struct ip_addr allsystems; static struct ip_addr allrouters; /** * Initialize the IGMP module */ void igmp_init(void) { LWIP_DEBUGF(IGMP_DEBUG, ("igmp_init: initializing\n")); IP4_ADDR(&allsystems, 224, 0, 0, 1); IP4_ADDR(&allrouters, 224, 0, 0, 2); } #ifdef LWIP_DEBUG /** * Dump global IGMP groups list */ void igmp_dump_group_list() { struct igmp_group *group = igmp_group_list; while (group != NULL) { LWIP_DEBUGF(IGMP_DEBUG, ("igmp_dump_group_list: [%"U32_F"] ", (u32_t)(group->group_state))); ip_addr_debug_print(IGMP_DEBUG, &group->group_address); LWIP_DEBUGF(IGMP_DEBUG, (" on if %x\n", (int) group->interface)); group = group->next; } LWIP_DEBUGF(IGMP_DEBUG, ("\n")); } #else #define igmp_dump_group_list() #endif /* LWIP_DEBUG */ /** * Start IGMP processing on interface * * @param netif network interface on which start IGMP processing */ err_t igmp_start(struct netif *netif) { struct igmp_group* group; LWIP_DEBUGF(IGMP_DEBUG, ("igmp_start: starting IGMP processing on if %x\n", (int) netif)); group = igmp_lookup_group(netif, &allsystems); if (group != NULL) { group->group_state = IGMP_GROUP_IDLE_MEMBER; group->use++; /* Allow the igmp messages at the MAC level */ if (netif->igmp_mac_filter != NULL) { LWIP_DEBUGF(IGMP_DEBUG, ("igmp_start: igmp_mac_filter(ADD ")); ip_addr_debug_print(IGMP_DEBUG, &allsystems); LWIP_DEBUGF(IGMP_DEBUG, (") on if %x\n", (int) netif)); netif->igmp_mac_filter( netif, &allsystems, IGMP_ADD_MAC_FILTER); } return ERR_OK; } return ERR_MEM; } /** * Stop IGMP processing on interface * * @param netif network interface on which stop IGMP processing */ err_t igmp_stop(struct netif *netif) { struct igmp_group *group = igmp_group_list; struct igmp_group *prev = NULL; struct igmp_group *next; /* look for groups joined on this interface further down the list */ while (group != NULL) { next = group->next; /* is it a group joined on this interface? */ if (group->interface == netif) { /* is it the first group of the list? */ if (group == igmp_group_list) { igmp_group_list = next; } /* is there a "previous" group defined? */ if (prev != NULL) { prev->next = next; } /* disable the group at the MAC level */ if (netif->igmp_mac_filter != NULL) { LWIP_DEBUGF(IGMP_DEBUG, ("igmp_stop: igmp_mac_filter(DEL ")); ip_addr_debug_print(IGMP_DEBUG, &group->group_address); LWIP_DEBUGF(IGMP_DEBUG, (") on if %x\n", (int) netif)); netif->igmp_mac_filter(netif, &(group->group_address), IGMP_DEL_MAC_FILTER); } /* free group */ memp_free(MEMP_IGMP_GROUP, group); } else { /* change the "previous" */ prev = group; } /* move to "next" */ group = next; } return ERR_OK; } /** * Report IGMP memberships for this interface * * @param netif network interface on which report IGMP memberships */ void igmp_report_groups( struct netif *netif) { struct igmp_group *group = igmp_group_list; LWIP_DEBUGF(IGMP_DEBUG, ("igmp_report_groups: sending IGMP reports on if %x\n", (int) netif)); while (group != NULL) { if (group->interface == netif) { igmp_delaying_member( group, IGMP_JOIN_DELAYING_MEMBER_TMR); } group = group->next; } } /** * Search for a group in the global igmp_group_list * * @param ifp the network interface for which to look * @param addr the group ip address to search for * @return a struct igmp_group* if the group has been found, * NULL if the group wasn't found. */ struct igmp_group * igmp_lookfor_group(struct netif *ifp, struct ip_addr *addr) { struct igmp_group *group = igmp_group_list; while (group != NULL) { if ((group->interface == ifp) && (ip_addr_cmp(&(group->group_address), addr))) { return group; } group = group->next; } /* to be clearer, we return NULL here instead of * 'group' (which is also NULL at this point). */ return NULL; } /** * Search for a specific igmp group and create a new one if not found- * * @param ifp the network interface for which to look * @param addr the group ip address to search * @return a struct igmp_group*, * NULL on memory error. */ struct igmp_group * igmp_lookup_group(struct netif *ifp, struct ip_addr *addr) { struct igmp_group *group = igmp_group_list; /* Search if the group already exists */ group = igmp_lookfor_group(ifp, addr); if (group != NULL) { /* Group already exists. */ return group; } /* Group doesn't exist yet, create a new one */ group = memp_malloc(MEMP_IGMP_GROUP); if (group != NULL) { group->interface = ifp; ip_addr_set(&(group->group_address), addr); group->timer = 0; /* Not running */ group->group_state = IGMP_GROUP_NON_MEMBER; group->last_reporter_flag = 0; group->use = 0; group->next = igmp_group_list; igmp_group_list = group; } LWIP_DEBUGF(IGMP_DEBUG, ("igmp_lookup_group: %sallocated a new group with address ", (group?"":"impossible to "))); ip_addr_debug_print(IGMP_DEBUG, addr); LWIP_DEBUGF(IGMP_DEBUG, (" on if %x\n", (int) ifp)); return group; } /** * Remove a group in the global igmp_group_list * * @param group the group to remove from the global igmp_group_list * @return ERR_OK if group was removed from the list, an err_t otherwise */ err_t igmp_remove_group(struct igmp_group *group) { err_t err = ERR_OK; /* Is it the first group? */ if (igmp_group_list == group) { igmp_group_list = group->next; } else { /* look for group further down the list */ struct igmp_group *tmpGroup; for (tmpGroup = igmp_group_list; tmpGroup != NULL; tmpGroup = tmpGroup->next) { if (tmpGroup->next == group) { tmpGroup->next = group->next; break; } } /* Group not found in the global igmp_group_list */ if (tmpGroup == NULL) err = ERR_ARG; } /* free group */ memp_free(MEMP_IGMP_GROUP, group); return err; } /** * Called from ip_input() if a new IGMP packet is received. * * @param p received igmp packet, p->payload pointing to the ip header * @param inp network interface on which the packet was received * @param dest destination ip address of the igmp packet */ void igmp_input(struct pbuf *p, struct netif *inp, struct ip_addr *dest) { struct ip_hdr * iphdr; struct igmp_msg* igmp; struct igmp_group* group; struct igmp_group* groupref; /* Note that the length CAN be greater than 8 but only 8 are used - All are included in the checksum */ iphdr = p->payload; if (pbuf_header(p, -(s16_t)(IPH_HL(iphdr) * 4)) || (p->len < IGMP_MINLEN)) { pbuf_free(p); IGMP_STATS_INC(igmp.lenerr); LWIP_DEBUGF(IGMP_DEBUG, ("igmp_input: length error\n")); return; } LWIP_DEBUGF(IGMP_DEBUG, ("igmp_input: message from ")); ip_addr_debug_print(IGMP_DEBUG, &(iphdr->src)); LWIP_DEBUGF(IGMP_DEBUG, (" to address ")); ip_addr_debug_print(IGMP_DEBUG, &(iphdr->dest)); LWIP_DEBUGF(IGMP_DEBUG, (" on if %x\n", (int) inp)); /* Now calculate and check the checksum */ igmp = (struct igmp_msg *)p->payload; if (inet_chksum(igmp, p->len)) { pbuf_free(p); IGMP_STATS_INC(igmp.chkerr); LWIP_DEBUGF(IGMP_DEBUG, ("igmp_input: checksum error\n")); return; } /* Packet is ok so find an existing group */ group = igmp_lookfor_group(inp, dest); /* use the incoming IP address! */ /* If group can be found or create... */ if (!group) { pbuf_free(p); LWIP_DEBUGF(IGMP_DEBUG, ("igmp_input: IGMP frame not for us\n")); return; } /* NOW ACT ON THE INCOMING MESSAGE TYPE... */ switch (igmp->igmp_msgtype) { case IGMP_MEMB_QUERY: { /* IGMP_MEMB_QUERY to the "all systems" address ? */ if ((ip_addr_cmp(dest, &allsystems)) && (igmp->igmp_group_address.addr == 0)) { /* THIS IS THE GENERAL QUERY */ LWIP_DEBUGF(IGMP_DEBUG, ("igmp_input: General IGMP_MEMB_QUERY on \"ALL SYSTEMS\" address (224.0.0.1) [igmp_maxresp=%i]\n", (int)(igmp->igmp_maxresp))); if (igmp->igmp_maxresp == 0) { IGMP_STATS_INC(igmp.v1_rxed); LWIP_DEBUGF(IGMP_DEBUG, ("igmp_input: got an all hosts query with time== 0 - this is V1 and not implemented - treat as v2\n")); igmp->igmp_maxresp = IGMP_V1_DELAYING_MEMBER_TMR; } IGMP_STATS_INC(igmp.group_query_rxed); groupref = igmp_group_list; while (groupref) { /* Do not send messages on the all systems group address! */ if ((groupref->interface == inp) && (!(ip_addr_cmp(&(groupref->group_address), &allsystems)))) { igmp_delaying_member( groupref, igmp->igmp_maxresp); } groupref = groupref->next; } } else { /* IGMP_MEMB_QUERY to a specific group ? */ if (group->group_address.addr != 0) { LWIP_DEBUGF(IGMP_DEBUG, ("igmp_input: IGMP_MEMB_QUERY to a specific group ")); ip_addr_debug_print(IGMP_DEBUG, &group->group_address); if (ip_addr_cmp (dest, &allsystems)) { LWIP_DEBUGF(IGMP_DEBUG, (" using \"ALL SYSTEMS\" address (224.0.0.1) [igmp_maxresp=%i]\n", (int)(igmp->igmp_maxresp))); /* we first need to re-lookfor the group since we used dest last time */ group = igmp_lookfor_group(inp, &igmp->igmp_group_address); } else { LWIP_DEBUGF(IGMP_DEBUG, (" with the group address as destination [igmp_maxresp=%i]\n", (int)(igmp->igmp_maxresp))); } if (group != NULL) { IGMP_STATS_INC(igmp.unicast_query); igmp_delaying_member( group, igmp->igmp_maxresp); } } } break; } case IGMP_V2_MEMB_REPORT: { LWIP_DEBUGF(IGMP_DEBUG, ("igmp_input: IGMP_V2_MEMB_REPORT\n")); IGMP_STATS_INC(igmp.report_rxed); if (group->group_state == IGMP_GROUP_DELAYING_MEMBER) { /* This is on a specific group we have already looked up */ group->timer = 0; /* stopped */ group->group_state = IGMP_GROUP_IDLE_MEMBER; group->last_reporter_flag = 0; } break; } default: { LWIP_DEBUGF(IGMP_DEBUG, ("igmp_input: unexpected msg %x in state %x on group %x on if %x\n", (int) igmp->igmp_msgtype, (int) group->group_state, (int) &group, (int) group->interface)); break; } } pbuf_free(p); return; } /** * Join a group on one network interface. * * @param ifaddr ip address of the network interface which should join a new group * @param groupaddr the ip address of the group which to join * @return ERR_OK if group was joined on the netif(s), an err_t otherwise */ err_t igmp_joingroup(struct ip_addr *ifaddr, struct ip_addr *groupaddr) { err_t err = ERR_VAL; /* no matching interface */ struct igmp_group *group; struct netif *netif; /* make sure it is multicast address */ LWIP_ERROR("igmp_joingroup: attempt to join non-multicast address", ip_addr_ismulticast(groupaddr), return ERR_VAL;); LWIP_ERROR("igmp_joingroup: attempt to join allsystems address", (!ip_addr_cmp(groupaddr, &allsystems)), return ERR_VAL;); /* loop through netif's */ netif = netif_list; while (netif != NULL) { /* Should we join this interface ? */ if ((netif->flags & NETIF_FLAG_IGMP) && ((ip_addr_isany(ifaddr) || ip_addr_cmp(&(netif->ip_addr), ifaddr)))) { /* find group or create a new one if not found */ group = igmp_lookup_group(netif, groupaddr); if (group != NULL) { /* This should create a new group, check the state to make sure */ if (group->group_state != IGMP_GROUP_NON_MEMBER) { LWIP_DEBUGF(IGMP_DEBUG, ("igmp_joingroup: join to group not in state IGMP_GROUP_NON_MEMBER\n")); } else { /* OK - it was new group */ LWIP_DEBUGF(IGMP_DEBUG, ("igmp_joingroup: join to new group: ")); ip_addr_debug_print(IGMP_DEBUG, groupaddr); LWIP_DEBUGF(IGMP_DEBUG, ("\n")); /* If first use of the group, allow the group at the MAC level */ if ((group->use==0) && (netif->igmp_mac_filter != NULL)) { LWIP_DEBUGF(IGMP_DEBUG, ("igmp_joingroup: igmp_mac_filter(ADD ")); ip_addr_debug_print(IGMP_DEBUG, groupaddr); LWIP_DEBUGF(IGMP_DEBUG, (") on if %x\n", (int) netif)); netif->igmp_mac_filter(netif, groupaddr, IGMP_ADD_MAC_FILTER); } IGMP_STATS_INC(igmp.join_sent); igmp_send(group, IGMP_V2_MEMB_REPORT); igmp_start_timer(group, IGMP_JOIN_DELAYING_MEMBER_TMR); /* Need to work out where this timer comes from */ group->group_state = IGMP_GROUP_DELAYING_MEMBER; } /* Increment group use */ group->use++; /* Join on this interface */ err = ERR_OK; } else { /* Return an error even if some network interfaces are joined */ /** @todo undo any other netif already joined */ LWIP_DEBUGF(IGMP_DEBUG, ("igmp_joingroup: Not enought memory to join to group\n")); return ERR_MEM; } } /* proceed to next network interface */ netif = netif->next; } return err; } /** * Leave a group on one network interface. * * @param ifaddr ip address of the network interface which should leave a group * @param groupaddr the ip address of the group which to leave * @return ERR_OK if group was left on the netif(s), an err_t otherwise */ err_t igmp_leavegroup(struct ip_addr *ifaddr, struct ip_addr *groupaddr) { err_t err = ERR_VAL; /* no matching interface */ struct igmp_group *group; struct netif *netif; /* make sure it is multicast address */ LWIP_ERROR("igmp_leavegroup: attempt to leave non-multicast address", ip_addr_ismulticast(groupaddr), return ERR_VAL;); LWIP_ERROR("igmp_leavegroup: attempt to leave allsystems address", (!ip_addr_cmp(groupaddr, &allsystems)), return ERR_VAL;); /* loop through netif's */ netif = netif_list; while (netif != NULL) { /* Should we leave this interface ? */ if ((netif->flags & NETIF_FLAG_IGMP) && ((ip_addr_isany(ifaddr) || ip_addr_cmp(&(netif->ip_addr), ifaddr)))) { /* find group */ group = igmp_lookfor_group(netif, groupaddr); if (group != NULL) { /* Only send a leave if the flag is set according to the state diagram */ LWIP_DEBUGF(IGMP_DEBUG, ("igmp_leavegroup: Leaving group: ")); ip_addr_debug_print(IGMP_DEBUG, groupaddr); LWIP_DEBUGF(IGMP_DEBUG, ("\n")); /* If there is no other use of the group */ if (group->use <= 1) { /* If we are the last reporter for this group */ if (group->last_reporter_flag) { LWIP_DEBUGF(IGMP_DEBUG, ("igmp_leavegroup: sending leaving group\n")); IGMP_STATS_INC(igmp.leave_sent); igmp_send(group, IGMP_LEAVE_GROUP); } /* Disable the group at the MAC level */ if (netif->igmp_mac_filter != NULL) { LWIP_DEBUGF(IGMP_DEBUG, ("igmp_leavegroup: igmp_mac_filter(DEL ")); ip_addr_debug_print(IGMP_DEBUG, groupaddr); LWIP_DEBUGF(IGMP_DEBUG, (") on if %x\n", (int) netif)); netif->igmp_mac_filter(netif, groupaddr, IGMP_DEL_MAC_FILTER); } LWIP_DEBUGF(IGMP_DEBUG, ("igmp_leavegroup: remove group: ")); ip_addr_debug_print(IGMP_DEBUG, groupaddr); LWIP_DEBUGF(IGMP_DEBUG, ("\n")); /* Free the group */ igmp_remove_group(group); } else { /* Decrement group use */ group->use--; } /* Leave on this interface */ err = ERR_OK; } else { /* It's not a fatal error on "leavegroup" */ LWIP_DEBUGF(IGMP_DEBUG, ("igmp_leavegroup: not member of group\n")); } } /* proceed to next network interface */ netif = netif->next; } return err; } /** * The igmp timer function (both for NO_SYS=1 and =0) * Should be called every IGMP_TMR_INTERVAL milliseconds (100 ms is default). */ void igmp_tmr(void) { struct igmp_group *group = igmp_group_list; while (group != NULL) { if (group->timer != 0) { group->timer -= 1; if (group->timer == 0) { igmp_timeout(group); } } group = group->next; } } /** * Called if a timeout for one group is reached. * Sends a report for this group. * * @param group an igmp_group for which a timeout is reached */ void igmp_timeout(struct igmp_group *group) { /* If the state is IGMP_GROUP_DELAYING_MEMBER then we send a report for this group */ if (group->group_state == IGMP_GROUP_DELAYING_MEMBER) { LWIP_DEBUGF(IGMP_DEBUG, ("igmp_timeout: report membership for group with address ")); ip_addr_debug_print(IGMP_DEBUG, &(group->group_address)); LWIP_DEBUGF(IGMP_DEBUG, (" on if %x\n", (int) group->interface)); igmp_send(group, IGMP_V2_MEMB_REPORT); } } /** * Start a timer for an igmp group * * @param group the igmp_group for which to start a timer * @param max_time the time in multiples of IGMP_TMR_INTERVAL (decrease with * every call to igmp_tmr()) */ void igmp_start_timer(struct igmp_group *group, u8_t max_time) { /** * @todo Important !! this should be random 0 -> max_time. Find out how to do this */ group->timer = max_time; } /** * Stop a timer for an igmp_group * * @param group the igmp_group for which to stop the timer */ void igmp_stop_timer(struct igmp_group *group) { group->timer = 0; } /** * Delaying membership report for a group if necessary * * @param group the igmp_group for which "delaying" membership report * @param maxresp query delay */ void igmp_delaying_member( struct igmp_group *group, u8_t maxresp) { if ((group->group_state == IGMP_GROUP_IDLE_MEMBER) || ((group->group_state == IGMP_GROUP_DELAYING_MEMBER) && (maxresp > group->timer))) { igmp_start_timer(group, (maxresp)/2); group->group_state = IGMP_GROUP_DELAYING_MEMBER; } } /** * Sends an IP packet on a network interface. This function constructs the IP header * and calculates the IP header checksum. If the source IP address is NULL, * the IP address of the outgoing network interface is filled in as source address. * * @param p the packet to send (p->payload points to the data, e.g. next protocol header; if dest == IP_HDRINCL, p already includes an IP header and p->payload points to that IP header) * @param src the source IP address to send from (if src == IP_ADDR_ANY, the * IP address of the netif used to send is used as source address) * @param dest the destination IP address to send the packet to * @param ttl the TTL value to be set in the IP header * @param proto the PROTOCOL to be set in the IP header * @param netif the netif on which to send this packet * @return ERR_OK if the packet was sent OK * ERR_BUF if p doesn't have enough space for IP/LINK headers * returns errors returned by netif->output */ err_t igmp_ip_output_if(struct pbuf *p, struct ip_addr *src, struct ip_addr *dest, u8_t ttl, u8_t proto, struct netif *netif) { static u16_t ip_id = 0; struct ip_hdr * iphdr = NULL; u16_t * ra = NULL; /* First write in the "router alert" */ if (pbuf_header(p, ROUTER_ALERTLEN)) { LWIP_DEBUGF(IGMP_DEBUG, ("igmp_ip_output_if: not enough room for IP header in pbuf\n")); return ERR_BUF; } /* This is the "router alert" option */ ra = p->payload; ra[0] = htons (ROUTER_ALERT); ra[1] = 0x0000; /* Router shall examine packet */ /* now the normal ip header */ if (pbuf_header(p, IP_HLEN)) { LWIP_DEBUGF(IGMP_DEBUG, ("igmp_ip_output_if: not enough room for IP header in pbuf\n")); return ERR_BUF; } iphdr = p->payload; /* Should the IP header be generated or is it already included in p? */ if (dest != IP_HDRINCL) { /** @todo should be shared with ip.c - ip_output_if */ IPH_TTL_SET(iphdr, ttl); IPH_PROTO_SET(iphdr, proto); ip_addr_set(&(iphdr->dest), dest); IPH_VHLTOS_SET(iphdr, 4, ((IP_HLEN + ROUTER_ALERTLEN) / 4), 0/*tos*/); IPH_LEN_SET(iphdr, htons(p->tot_len)); IPH_OFFSET_SET(iphdr, 0); IPH_ID_SET(iphdr, htons(ip_id)); ++ip_id; if (ip_addr_isany(src)) { ip_addr_set(&(iphdr->src), &(netif->ip_addr)); } else { ip_addr_set(&(iphdr->src), src); } IPH_CHKSUM_SET(iphdr, 0); #if CHECKSUM_GEN_IP IPH_CHKSUM_SET(iphdr, inet_chksum(iphdr, (IP_HLEN + ROUTER_ALERTLEN))); #endif } else { dest = &(iphdr->dest); } #if IP_DEBUG ip_debug_print(p); #endif LWIP_DEBUGF(IGMP_DEBUG, ("igmp_ip_output_if: sending to if %x\n", (int) netif)); return netif->output(netif, p, dest); } /** * Send an igmp packet to a specific group. * * @param group the group to which to send the packet * @param type the type of igmp packet to send */ void igmp_send(struct igmp_group *group, u8_t type) { struct pbuf* p = NULL; struct igmp_msg* igmp = NULL; struct ip_addr src = {0}; struct ip_addr* dest = NULL; /* IP header + "router alert" option + IGMP header */ p = pbuf_alloc(PBUF_TRANSPORT, IGMP_MINLEN, PBUF_RAM); if (p) { igmp = p->payload; LWIP_ASSERT("igmp_send: check that first pbuf can hold struct igmp_msg", (p->len >= sizeof(struct igmp_msg))); ip_addr_set(&src, &((group->interface)->ip_addr)); if (type == IGMP_V2_MEMB_REPORT) { dest = &(group->group_address); IGMP_STATS_INC(igmp.report_sent); ip_addr_set(&(igmp->igmp_group_address), &(group->group_address)); group->last_reporter_flag = 1; /* Remember we were the last to report */ } else { if (type == IGMP_LEAVE_GROUP) { dest = &allrouters; ip_addr_set(&(igmp->igmp_group_address), &(group->group_address)); } } if ((type == IGMP_V2_MEMB_REPORT) || (type == IGMP_LEAVE_GROUP)) { igmp->igmp_msgtype = type; igmp->igmp_maxresp = 0; igmp->igmp_checksum = 0; igmp->igmp_checksum = inet_chksum( igmp, IGMP_MINLEN); igmp_ip_output_if( p, &src, dest, IGMP_TTL, IP_PROTO_IGMP, group->interface); } pbuf_free (p); } else { LWIP_DEBUGF(IGMP_DEBUG, ("igmp_send: not enough memory for igmp_send\n")); } } #endif /* LWIP_IGMP */