mirror of
https://github.com/lwip-tcpip/lwip.git
synced 2025-01-27 03:35:38 +00:00
Changed static etharp_query() to support queueing packets. This fix missed in last commit.
This commit is contained in:
parent
3e62b75529
commit
b9353eaf12
@ -1,40 +1,33 @@
|
||||
/*
|
||||
* Copyright (c) 2001, Swedish Institute of Computer Science.
|
||||
* Copyright (c) 2001, 2002 Swedish Institute of Computer Science.
|
||||
* All rights reserved.
|
||||
*
|
||||
* Redistribution and use in source and binary forms, with or without modification,
|
||||
* are permitted provided that the following conditions are met:
|
||||
*
|
||||
* 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. All advertising materials mentioning features or use of this software
|
||||
* must display the following acknowledgement:
|
||||
* This product includes software developed by the Swedish Institute
|
||||
* of Computer Science and its contributors.
|
||||
* 4. Neither the name of the Institute nor the names of its contributors
|
||||
* may be used to endorse or promote products derived from this software
|
||||
* without specific prior written permission.
|
||||
* 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. The name of the author may not be used to endorse or promote products
|
||||
* derived from this software without specific prior written permission.
|
||||
*
|
||||
* THIS SOFTWARE IS PROVIDED BY THE INSTITUTE 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 INSTITUTE 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 SOFTWARE IS PROVIDED BY THE AUTHOR ``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 AUTHOR 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 part of the lwIP TCP/IP stack.
|
||||
*
|
||||
* Author: Adam Dunkels <adam@sics.se>
|
||||
*
|
||||
*
|
||||
*/
|
||||
|
||||
#ifndef __NETIF_ETHARP_H__
|
||||
@ -83,7 +76,8 @@ struct pbuf *etharp_arp_input(struct netif *netif, struct eth_addr *ethaddr,
|
||||
struct pbuf *etharp_output(struct netif *netif, struct ip_addr *ipaddr,
|
||||
struct pbuf *q);
|
||||
struct pbuf *etharp_output_sent(struct pbuf *p);
|
||||
struct pbuf *etharp_query(struct netif *netif, struct ip_addr *ipaddr);
|
||||
struct pbuf *etharp_query(struct netif *netif, struct ip_addr *ipaddr, struct pbuf *q);
|
||||
|
||||
|
||||
|
||||
#endif /* __NETIF_ARP_H__ */
|
||||
|
@ -3,6 +3,9 @@
|
||||
* Address Resolution Protocol module for IP over Ethernet
|
||||
*
|
||||
* $Log: etharp.c,v $
|
||||
* Revision 1.6 2002/11/11 14:34:29 likewise
|
||||
* Changed static etharp_query() to support queueing packets. This fix missed in last commit.
|
||||
*
|
||||
* Revision 1.5 2002/11/08 22:14:24 likewise
|
||||
* Fixed numerous bugs. Re-used etharp_query() in etharp_output(). Added comments and JavaDoc documentation.
|
||||
*
|
||||
@ -57,9 +60,10 @@
|
||||
#include "netif/etharp.h"
|
||||
#include "lwip/ip.h"
|
||||
#include "lwip/stats.h"
|
||||
#include "lwipopts.h"
|
||||
|
||||
/* ARP needs to inform DHCP of any ARP replies? */
|
||||
#if (LWIP_DHCP && DHPC_DOES_ARP_CHECK)
|
||||
#if (LWIP_DHCP && DHCP_DOES_ARP_CHECK)
|
||||
# include "lwip/dhcp.h"
|
||||
#endif
|
||||
|
||||
@ -134,6 +138,8 @@ static const struct eth_addr ethbroadcast = {{0xff,0xff,0xff,0xff,0xff,0xff}};
|
||||
static struct etharp_entry arp_table[ARP_TABLE_SIZE];
|
||||
static u8_t ctime;
|
||||
|
||||
static struct pbuf *insert_arp_entry(struct ip_addr *ipaddr, struct eth_addr *ethaddr);
|
||||
|
||||
/**
|
||||
* Initializes ARP module.
|
||||
*/
|
||||
@ -145,6 +151,7 @@ etharp_init(void)
|
||||
for(i = 0; i < ARP_TABLE_SIZE; ++i) {
|
||||
arp_table[i].state = ETHARP_STATE_EMPTY;
|
||||
}
|
||||
ctime = 0;
|
||||
}
|
||||
|
||||
/**
|
||||
@ -161,14 +168,15 @@ etharp_tmr(void)
|
||||
++ctime;
|
||||
/* remove expired entries from the ARP table */
|
||||
for(i = 0; i < ARP_TABLE_SIZE; ++i) {
|
||||
if(arp_table[i].state == ETHARP_STATE_STABLE &&
|
||||
ctime - arp_table[i].ctime >= ARP_MAXAGE) {
|
||||
DEBUGF(ETHARP_DEBUG, ("etharp_timer: expired stable entry %d.\n", i));
|
||||
if((arp_table[i].state == ETHARP_STATE_STABLE) &&
|
||||
(ctime - arp_table[i].ctime >= ARP_MAXAGE)) {
|
||||
DEBUGF(ETHARP_DEBUG, ("etharp_timer: expired stable entry %u.\n", i));
|
||||
arp_table[i].state = ETHARP_STATE_EMPTY;
|
||||
} else if(arp_table[i].state == ETHARP_STATE_PENDING &&
|
||||
ctime - arp_table[i].ctime >= ARP_MAXPENDING) {
|
||||
DEBUGF(ETHARP_DEBUG, ("etharp_timer: expired pending entry %d - dequeueing %p.\n", i, arp_table[i].p));
|
||||
} else if((arp_table[i].state == ETHARP_STATE_PENDING) &&
|
||||
(ctime - arp_table[i].ctime >= ARP_MAXPENDING)) {
|
||||
DEBUGF(ETHARP_DEBUG, ("etharp_timer: expired pending entry %u - dequeueing %p.\n", i, arp_table[i].p));
|
||||
arp_table[i].state = ETHARP_STATE_EMPTY;
|
||||
/* remove any queued packet */
|
||||
pbuf_free(arp_table[i].p);
|
||||
arp_table[i].p = NULL;
|
||||
}
|
||||
@ -190,6 +198,7 @@ find_arp_entry(void)
|
||||
/* Try to find an unused entry in the ARP table. */
|
||||
for(i = 0; i < ARP_TABLE_SIZE; ++i) {
|
||||
if(arp_table[i].state == ETHARP_STATE_EMPTY) {
|
||||
DEBUGF(ETHARP_DEBUG, ("find_arp_entry: found empty entry %u\n", i));
|
||||
break;
|
||||
}
|
||||
}
|
||||
@ -200,74 +209,98 @@ find_arp_entry(void)
|
||||
maxtime = 0;
|
||||
j = ARP_TABLE_SIZE;
|
||||
for(i = 0; i < ARP_TABLE_SIZE; ++i) {
|
||||
if(arp_table[i].state == ETHARP_STATE_STABLE &&
|
||||
ctime - arp_table[i].ctime > maxtime) {
|
||||
maxtime = ctime - arp_table[i].ctime;
|
||||
j = i;
|
||||
/* remember entry with oldest stable entry in j*/
|
||||
if((arp_table[i].state == ETHARP_STATE_STABLE) &&
|
||||
(ctime - arp_table[i].ctime > maxtime)) {
|
||||
maxtime = ctime - arp_table[i].ctime;
|
||||
j = i;
|
||||
}
|
||||
}
|
||||
DEBUGF(ETHARP_DEBUG, ("find_arp_entry: found oldest stable entry %u\n", j));
|
||||
i = j;
|
||||
}
|
||||
return i;
|
||||
}
|
||||
|
||||
/**
|
||||
* Insert an entry into the ARP cache, or update an existing one.
|
||||
*
|
||||
* @param ipaddr IP address of the inserted ARP entry.
|
||||
* @param ethaddr Ethernet address of the inserted ARP entry.
|
||||
*
|
||||
* @return pbuf If non-NULL, a packet that was queued on a pending entry.
|
||||
* You should sent it and must call pbuf_free().
|
||||
*
|
||||
* @see pbuf_free()
|
||||
*/
|
||||
static struct pbuf *
|
||||
update_arp_entry(struct ip_addr *ipaddr, struct eth_addr *ethaddr)
|
||||
insert_arp_entry(struct ip_addr *ipaddr, struct eth_addr *ethaddr)
|
||||
{
|
||||
u8_t i, k;
|
||||
struct pbuf *p;
|
||||
struct eth_hdr *ethhdr;
|
||||
|
||||
|
||||
/* Walk through the ARP mapping table and try to find an entry to
|
||||
update. If none is found, the IP -> MAC address mapping is
|
||||
inserted in the ARP table. */
|
||||
update. If none is found, the IP -> MAC address mapping is
|
||||
inserted in the ARP table. */
|
||||
for(i = 0; i < ARP_TABLE_SIZE; ++i) {
|
||||
/* Check if the source IP address of the incoming packet matches
|
||||
the IP address in this ARP table entry. */
|
||||
the IP address in this ARP table entry. */
|
||||
if(ip_addr_cmp(ipaddr, &arp_table[i].ipaddr)) {
|
||||
|
||||
/* First, check those entries that are already in use. */
|
||||
|
||||
/* check those entries that are already in use. */
|
||||
if(arp_table[i].state == ETHARP_STATE_STABLE) {
|
||||
/* An old entry found, update this and return. */
|
||||
for(k = 0; k < 6; ++k) {
|
||||
arp_table[i].ethaddr.addr[k] = ethaddr->addr[k];
|
||||
}
|
||||
arp_table[i].ctime = ctime;
|
||||
return NULL;
|
||||
DEBUGF(ETHARP_DEBUG, ("insert_arp_entry: updating stable entry %u\n", i));
|
||||
/* An old entry found, update this and return. */
|
||||
for(k = 0; k < 6; ++k) {
|
||||
arp_table[i].ethaddr.addr[k] = ethaddr->addr[k];
|
||||
}
|
||||
arp_table[i].ctime = ctime;
|
||||
return NULL;
|
||||
}
|
||||
if(arp_table[i].state == ETHARP_STATE_PENDING) {
|
||||
/* A pending entry was found, so we fill this in and return
|
||||
the queued packet (if any). */
|
||||
for(k = 0; k < 6; ++k) {
|
||||
arp_table[i].ethaddr.addr[k] = ethaddr->addr[k];
|
||||
}
|
||||
arp_table[i].ctime = ctime;
|
||||
arp_table[i].state = ETHARP_STATE_STABLE;
|
||||
p = arp_table[i].p;
|
||||
if(p != NULL) {
|
||||
p->payload = arp_table[i].payload;
|
||||
p->len = arp_table[i].len;
|
||||
p->tot_len = arp_table[i].tot_len;
|
||||
arp_table[i].p = NULL;
|
||||
|
||||
ethhdr = p->payload;
|
||||
|
||||
for(k = 0; k < 6; ++k) {
|
||||
ethhdr->dest.addr[k] = ethaddr->addr[k];
|
||||
}
|
||||
|
||||
ethhdr->type = htons(ETHTYPE_IP);
|
||||
}
|
||||
return p;
|
||||
else if(arp_table[i].state == ETHARP_STATE_PENDING) {
|
||||
/* A pending entry was found, so we fill this in and return
|
||||
the queued packet (if any). */
|
||||
DEBUGF(ETHARP_DEBUG, ("insert_arp_entry: pending entry %u made stable\n", i));
|
||||
for(k = 0; k < 6; ++k) {
|
||||
arp_table[i].ethaddr.addr[k] = ethaddr->addr[k];
|
||||
}
|
||||
arp_table[i].ctime = ctime;
|
||||
arp_table[i].state = ETHARP_STATE_STABLE;
|
||||
p = arp_table[i].p;
|
||||
// queued packet present? */
|
||||
if(p != NULL) {
|
||||
/* remove queued packet from ARP entry (must be freed by the caller) */
|
||||
arp_table[i].p = NULL;
|
||||
|
||||
/* fill-in Ethernet header */
|
||||
ethhdr = p->payload;
|
||||
|
||||
for(k = 0; k < 6; ++k) {
|
||||
ethhdr->dest.addr[k] = ethaddr->addr[k];
|
||||
}
|
||||
|
||||
ethhdr->type = htons(ETHTYPE_IP);
|
||||
DEBUGF(ETHARP_DEBUG, ("insert_arp_entry: returning queued packet %p\n", p));
|
||||
}
|
||||
/* return queued packet, if any */
|
||||
return p;
|
||||
}
|
||||
}
|
||||
}
|
||||
/* We get here if no ARP entry was found. If so, we create one. */
|
||||
/* no matching ARP entry was found. find an empty or old entry. */
|
||||
i = find_arp_entry();
|
||||
if(i == ARP_TABLE_SIZE) {
|
||||
DEBUGF(ETHARP_DEBUG, ("insert_arp_entry: no available entry found\n"));
|
||||
return NULL;
|
||||
}
|
||||
|
||||
if (arp_table[i].state == ETHARP_STATE_STABLE) {
|
||||
DEBUGF(ETHARP_DEBUG, ("insert_arp_entry: overwriting old stable entry %u\n", i));
|
||||
}
|
||||
else {
|
||||
DEBUGF(ETHARP_DEBUG, ("insert_arp_entry: using empty entry %u\n", i));
|
||||
}
|
||||
ip_addr_set(&arp_table[i].ipaddr, ipaddr);
|
||||
for(k = 0; k < 6; ++k) {
|
||||
arp_table[i].ethaddr.addr[k] = ethaddr->addr[k];
|
||||
@ -275,22 +308,27 @@ update_arp_entry(struct ip_addr *ipaddr, struct eth_addr *ethaddr)
|
||||
arp_table[i].ctime = ctime;
|
||||
arp_table[i].state = ETHARP_STATE_STABLE;
|
||||
arp_table[i].p = NULL;
|
||||
|
||||
|
||||
return NULL;
|
||||
}
|
||||
|
||||
/**
|
||||
* Updates the ARP table and may return any queued packet to be sent
|
||||
* Updates the ARP table and may return any queued packet to be sent.
|
||||
*
|
||||
* Should be called for all incoming packets of IP kind. The function
|
||||
* does not alter the packet in any way, it just updates the ARP
|
||||
* table. After this function has been called, the normal TCP/IP stack
|
||||
* input function should be called.
|
||||
* Should be called for all incoming packets of IP kind. It updates
|
||||
* the ARP table for the local network. The function does not alter
|
||||
* the packet in any way and does not free it. After this function has
|
||||
* been called, the packet p must be given to the IP layer.
|
||||
*
|
||||
* The function may return a pbuf containing a packet that had
|
||||
* previously been queued for transmission. The device driver must
|
||||
* transmit this packet onto the network, and call pbuf_free() for the
|
||||
* pbuf.
|
||||
* @param netif The lwIP network interface on which the IP packet pbuf arrived.
|
||||
*
|
||||
* @param pbuf The IP packet that arrived on netif.
|
||||
*
|
||||
* @return If non-NULL, a pbuf that was queued on an ARP entry. The device
|
||||
* driver must transmit this packet onto the network, and call pbuf_free()
|
||||
* for the pbuf.
|
||||
*
|
||||
* @see pbuf_free()
|
||||
*/
|
||||
struct pbuf *
|
||||
etharp_ip_input(struct netif *netif, struct pbuf *p)
|
||||
@ -301,31 +339,40 @@ etharp_ip_input(struct netif *netif, struct pbuf *p)
|
||||
|
||||
/* Only insert/update an entry if the source IP address of the
|
||||
incoming IP packet comes from a host on the local network. */
|
||||
|
||||
/* source is on local network? */
|
||||
if(!ip_addr_maskcmp(&(hdr->ip.src), &(netif->ip_addr), &(netif->netmask))) {
|
||||
/* do nothing */
|
||||
return NULL;
|
||||
}
|
||||
DEBUGF(ETHARP_DEBUG, ("etharp_ip_input: updating ETHARP table.\n"));
|
||||
return update_arp_entry(&(hdr->ip.src), &(hdr->eth.src));
|
||||
/* update ARP table */
|
||||
return insert_arp_entry(&(hdr->ip.src), &(hdr->eth.src));
|
||||
}
|
||||
|
||||
|
||||
/**
|
||||
* Updates the ARP table and may return any queued packet to be sent
|
||||
* Updates the ARP table and returns an ARP reply or a queued IP packet.
|
||||
*
|
||||
* Should be called for incoming ARP packets. The pbuf in the argument
|
||||
* is freed by this function. If the function returns a pbuf (i.e.,
|
||||
* returns non-NULL), that pbuf constitutes an ARP reply and should be
|
||||
* sent out on the Ethernet.
|
||||
* is freed by this function. The returned pbuf is to be sent and then
|
||||
* freed by the caller.
|
||||
*
|
||||
* @note The driver must call pbuf_free() for the returned pbuf when the
|
||||
* packet has been sent.
|
||||
* @param netif The lwIP network interface on which the ARP packet pbuf arrived.
|
||||
* @param pbuf The ARP packet that arrived on netif. Is freed by this function.
|
||||
* @param ethaddr Ethernet address of netif.
|
||||
*
|
||||
* @return pbuf to be sent and freed by the caller.
|
||||
*
|
||||
* @see pbuf_free()
|
||||
*/
|
||||
struct pbuf *
|
||||
etharp_arp_input(struct netif *netif, struct eth_addr *ethaddr, struct pbuf *p)
|
||||
{
|
||||
struct etharp_hdr *hdr;
|
||||
u8_t i;
|
||||
|
||||
|
||||
/* drop short ARP packets */
|
||||
if(p->tot_len < sizeof(struct etharp_hdr)) {
|
||||
DEBUGF(ETHARP_DEBUG, ("etharp_etharp_input: packet too short (%d/%d)\n", p->tot_len, sizeof(struct etharp_hdr)));
|
||||
pbuf_free(p);
|
||||
@ -333,50 +380,72 @@ etharp_arp_input(struct netif *netif, struct eth_addr *ethaddr, struct pbuf *p)
|
||||
}
|
||||
|
||||
hdr = p->payload;
|
||||
|
||||
|
||||
switch(htons(hdr->opcode)) {
|
||||
/* ARP request? */
|
||||
case ARP_REQUEST:
|
||||
/* ARP request. If it asked for our address, we send out a
|
||||
reply. */
|
||||
reply. */
|
||||
DEBUGF(ETHARP_DEBUG, ("etharp_arp_input: ARP request\n"));
|
||||
/* ARP request for our address? */
|
||||
if(ip_addr_cmp(&(hdr->dipaddr), &(netif->ip_addr))) {
|
||||
|
||||
DEBUGF(ETHARP_DEBUG, ("etharp_arp_input: ARP request for our address\n"));
|
||||
/* re-use pbuf to send ARP reply */
|
||||
hdr->opcode = htons(ARP_REPLY);
|
||||
|
||||
ip_addr_set(&(hdr->dipaddr), &(hdr->sipaddr));
|
||||
ip_addr_set(&(hdr->sipaddr), &(netif->ip_addr));
|
||||
|
||||
for(i = 0; i < 6; ++i) {
|
||||
hdr->dhwaddr.addr[i] = hdr->shwaddr.addr[i];
|
||||
hdr->shwaddr.addr[i] = ethaddr->addr[i];
|
||||
hdr->ethhdr.dest.addr[i] = hdr->dhwaddr.addr[i];
|
||||
hdr->ethhdr.src.addr[i] = ethaddr->addr[i];
|
||||
hdr->dhwaddr.addr[i] = hdr->shwaddr.addr[i];
|
||||
hdr->shwaddr.addr[i] = ethaddr->addr[i];
|
||||
hdr->ethhdr.dest.addr[i] = hdr->dhwaddr.addr[i];
|
||||
hdr->ethhdr.src.addr[i] = ethaddr->addr[i];
|
||||
}
|
||||
|
||||
hdr->hwtype = htons(HWTYPE_ETHERNET);
|
||||
ARPH_HWLEN_SET(hdr, 6);
|
||||
|
||||
hdr->proto = htons(ETHTYPE_IP);
|
||||
|
||||
hdr->proto = htons(ETHTYPE_IP);
|
||||
ARPH_PROTOLEN_SET(hdr, sizeof(struct ip_addr));
|
||||
|
||||
hdr->ethhdr.type = htons(ETHTYPE_ARP);
|
||||
|
||||
hdr->ethhdr.type = htons(ETHTYPE_ARP);
|
||||
/* return ARP reply */
|
||||
return p;
|
||||
}
|
||||
#if 0
|
||||
/* ARP request, NOT for our address */
|
||||
else
|
||||
{
|
||||
}
|
||||
#endif
|
||||
break;
|
||||
case ARP_REPLY:
|
||||
/* ARP reply. We insert or update the ARP table. */
|
||||
DEBUGF(ETHARP_DEBUG, ("etharp_arp_input: ARP reply\n"));
|
||||
if(ip_addr_cmp(&(hdr->dipaddr), &(netif->ip_addr))) {
|
||||
struct pbuf *q;
|
||||
#if (LWIP_DHCP && DHCP_DOES_ARP_CHECK)
|
||||
/* DHCP needs to know about ARP replies */
|
||||
dhcp_arp_reply(&hdr->sipaddr);
|
||||
#endif
|
||||
/* update_arp_entry() will return a pbuf that has previously been
|
||||
queued waiting for an ARP reply. */
|
||||
q = update_arp_entry(&(hdr->sipaddr), &(hdr->shwaddr));
|
||||
/* for our address? */
|
||||
if(ip_addr_cmp(&(hdr->dipaddr), &(netif->ip_addr))) {
|
||||
struct pbuf *q;
|
||||
DEBUGF(ETHARP_DEBUG, ("etharp_arp_input: ARP reply for us\n"));
|
||||
/* insert_arp_entry() can return a pbuf that has previously been
|
||||
queued waiting for this IP address to become ARP stable. */
|
||||
q = insert_arp_entry(&(hdr->sipaddr), &(hdr->shwaddr));
|
||||
/* free incoming ARP reply pbuf */
|
||||
pbuf_free(p);
|
||||
p = NULL;
|
||||
return q;
|
||||
}
|
||||
#if 0
|
||||
/* ARP reply, NOT for our address */
|
||||
else
|
||||
{
|
||||
}
|
||||
#endif
|
||||
break;
|
||||
default:
|
||||
DEBUGF(ETHARP_DEBUG, ("etharp_arp_input: unknown type %d\n", htons(hdr->opcode)));
|
||||
@ -388,24 +457,27 @@ etharp_arp_input(struct netif *netif, struct eth_addr *ethaddr, struct pbuf *p)
|
||||
}
|
||||
|
||||
/**
|
||||
* Resolve Ethernet address and append header to the outgoing packet.
|
||||
* Resolve and fill-in Ethernet address header for outgoing packet.
|
||||
*
|
||||
* The etharp_output() function should be called for all outgoing
|
||||
* packets. The pbuf returned by the function should be sent out on
|
||||
* the Ethernet. This pbuf must then be passed to etharp_output_sent().
|
||||
* If ARP has the Ethernet address in cache, the given packet is
|
||||
* returned, ready to be sent.
|
||||
*
|
||||
* The function prepares the packet for transmission over the Ethernet
|
||||
* by adding an Ethernet header. If there is no IP -> MAC address
|
||||
* mapping, the function will queue the outgoing packet and return an
|
||||
* ARP request packet instead.
|
||||
* If ARP does not have the Ethernet address in cache the packet is
|
||||
* queued and a ARP request is sent (on a best-effort basis). This
|
||||
* ARP request is returned as a pbuf, which should be sent by the
|
||||
* caller.
|
||||
*
|
||||
* If ARP failed to allocate resources, NULL is returned.
|
||||
*
|
||||
* A returned non-NULL packet should be sent by the caller and
|
||||
* etharp_output_sent() must be called afterwards to free any ARP
|
||||
* request.
|
||||
*
|
||||
* @param netif The lwIP network interface which the IP packet will be sent on.
|
||||
* @param ipaddr The IP address of the packet destination.
|
||||
* @param pbuf The pbuf(s) containing the IP packet.
|
||||
* @param pbuf The pbuf(s) containing the IP packet to be sent.
|
||||
*
|
||||
* @return The packet which should be sent on the network and must be freed by
|
||||
* the caller.
|
||||
*
|
||||
* @return If non-NULL, a packet ready to be sent.
|
||||
* @see etharp_output_sent()
|
||||
*/
|
||||
struct pbuf *
|
||||
@ -413,17 +485,13 @@ etharp_output(struct netif *netif, struct ip_addr *ipaddr, struct pbuf *q)
|
||||
{
|
||||
struct eth_addr *dest, *srcaddr, mcastaddr;
|
||||
struct eth_hdr *ethhdr;
|
||||
struct etharp_hdr *hdr;
|
||||
struct pbuf *p;
|
||||
u8_t i;
|
||||
|
||||
/* obtain source Ethernet address of the given interface */
|
||||
srcaddr = (struct eth_addr *)netif->hwaddr;
|
||||
|
||||
/* Make room for Ethernet header. */
|
||||
if(pbuf_header(q, sizeof(struct eth_hdr)) != 0) {
|
||||
/* The pbuf_header() call shouldn't fail, and we'll just bail
|
||||
out if it does.. */
|
||||
out if it does.. */
|
||||
DEBUGF(ETHARP_DEBUG, ("etharp_output: could not allocate room for header.\n"));
|
||||
#ifdef LINK_STATS
|
||||
++stats.link.lenerr;
|
||||
@ -431,18 +499,24 @@ etharp_output(struct netif *netif, struct ip_addr *ipaddr, struct pbuf *q)
|
||||
return NULL;
|
||||
}
|
||||
|
||||
/* obtain source Ethernet address of the given interface */
|
||||
srcaddr = (struct eth_addr *)netif->hwaddr;
|
||||
|
||||
/* assume unresolved Ethernet address */
|
||||
dest = NULL;
|
||||
/* Construct Ethernet header. Start with looking up deciding which
|
||||
MAC address to use as a destination address. Broadcasts and
|
||||
multicasts are special, all other addresses are looked up in the
|
||||
ARP table. */
|
||||
MAC address to use as a destination address. Broadcasts and
|
||||
multicasts are special, all other addresses are looked up in the
|
||||
ARP table. */
|
||||
|
||||
/* destination IP address is an IP broadcast address? */
|
||||
if(ip_addr_isany(ipaddr) ||
|
||||
ip_addr_isbroadcast(ipaddr, &(netif->netmask))) {
|
||||
/* broadcast on Ethernet also */
|
||||
ip_addr_isbroadcast(ipaddr, &(netif->netmask))) {
|
||||
/* broadcast on Ethernet also */
|
||||
dest = (struct eth_addr *)ðbroadcast;
|
||||
} else if(ip_addr_ismulticast(ipaddr)) {
|
||||
}
|
||||
/* destination IP address is an IP multicast address? */
|
||||
else if(ip_addr_ismulticast(ipaddr)) {
|
||||
/* Hash IP multicast address to MAC address. */
|
||||
mcastaddr.addr[0] = 0x01;
|
||||
mcastaddr.addr[1] = 0x0;
|
||||
@ -452,138 +526,79 @@ etharp_output(struct netif *netif, struct ip_addr *ipaddr, struct pbuf *q)
|
||||
mcastaddr.addr[5] = ip4_addr4(ipaddr);
|
||||
/* destination Ethernet address is multicast */
|
||||
dest = &mcastaddr;
|
||||
/* destination IP unicast address */
|
||||
} else {
|
||||
/* the destination IP network address does not match the interface's
|
||||
network address */
|
||||
}
|
||||
/* destination IP address is an IP unicast address */
|
||||
else {
|
||||
/* destination IP network address not on local network? */
|
||||
if(!ip_addr_maskcmp(ipaddr, &(netif->ip_addr), &(netif->netmask))) {
|
||||
/* Use the IP address of the default gateway if the destination
|
||||
is not on the same subnet as we are. */
|
||||
ipaddr = &(netif->gw);
|
||||
/* gateway available? */
|
||||
if (netif->gw.addr != 0)
|
||||
{
|
||||
/* use the default gateway IP address */
|
||||
ipaddr = &(netif->gw);
|
||||
}
|
||||
else
|
||||
{
|
||||
/* IP destination address outside local network, but no gateway available */
|
||||
return NULL;
|
||||
}
|
||||
}
|
||||
|
||||
/* Try to find a stable IP-to-Ethernet address mapping for this IP
|
||||
destination address */
|
||||
for(i = 0; i < ARP_TABLE_SIZE; ++i) {
|
||||
/* Ethernet address for IP destination address is in ARP cache? */
|
||||
for(i = 0; i < ARP_TABLE_SIZE; ++i) {
|
||||
/* match found? */
|
||||
if(arp_table[i].state == ETHARP_STATE_STABLE &&
|
||||
ip_addr_cmp(ipaddr, &arp_table[i].ipaddr)) {
|
||||
dest = &arp_table[i].ethaddr;
|
||||
break;
|
||||
ip_addr_cmp(ipaddr, &arp_table[i].ipaddr)) {
|
||||
dest = &arp_table[i].ethaddr;
|
||||
break;
|
||||
}
|
||||
}
|
||||
/* could not find the destination Ethernet address in ARP cache? */
|
||||
if (dest == NULL) {
|
||||
/* query for the IP address using ARP request */
|
||||
p = etharp_query(netif, ipaddr, q);
|
||||
/* return the ARP request */
|
||||
return p;
|
||||
}
|
||||
/* destination Ethernet address resolved from ARP cache*/
|
||||
else
|
||||
{
|
||||
/* fallthrough */
|
||||
}
|
||||
}
|
||||
|
||||
/* could not find a destination Ethernet address? */
|
||||
if(dest == NULL) {
|
||||
/* No destination address has been found, so we'll have to send
|
||||
out an ARP request for the IP address. The outgoing packet is
|
||||
queued unless the queue is full. */
|
||||
|
||||
/* TODO: The host requirements RFC states that ARP should save at least one
|
||||
packet, and this should be the _latest_ packet. */
|
||||
|
||||
/* We check if we are already querying for this address. If so,
|
||||
we'll bail out. */
|
||||
for(i = 0; i < ARP_TABLE_SIZE; ++i) {
|
||||
if(arp_table[i].state == ETHARP_STATE_PENDING &&
|
||||
ip_addr_cmp(ipaddr, &arp_table[i].ipaddr)) {
|
||||
DEBUGF(ETHARP_DEBUG, ("etharp_output: already queued\n"));
|
||||
return NULL;
|
||||
}
|
||||
}
|
||||
|
||||
/* find a usable ARP entry */
|
||||
i = find_arp_entry();
|
||||
|
||||
/* If all table entries were in pending state, we won't send out any
|
||||
more ARP requests. We'll just give up. */
|
||||
if(i == ARP_TABLE_SIZE) {
|
||||
return NULL;
|
||||
}
|
||||
|
||||
/* Now, i is the ARP table entry which we will fill with the new
|
||||
information. */
|
||||
ip_addr_set(&arp_table[i].ipaddr, ipaddr);
|
||||
arp_table[i].ctime = ctime;
|
||||
arp_table[i].state = ETHARP_STATE_PENDING;
|
||||
#if 1
|
||||
arp_table[i].p = q;
|
||||
arp_table[i].payload = q->payload;
|
||||
arp_table[i].len = q->len;
|
||||
arp_table[i].tot_len = q->tot_len;
|
||||
|
||||
/* Because the pbuf will be queued, we'll increase the reference
|
||||
count. */
|
||||
DEBUGF(ETHARP_DEBUG, ("etharp_output: queueing %p\n", q));
|
||||
pbuf_ref(q);
|
||||
#else
|
||||
arp_table[i].p = NULL;
|
||||
#endif /* 0 */
|
||||
|
||||
|
||||
/* We allocate a pbuf for the outgoing ARP request packet. */
|
||||
p = pbuf_alloc(PBUF_LINK, sizeof(struct etharp_hdr), PBUF_RAM);
|
||||
if(p == NULL) {
|
||||
/* No ARP request packet could be allocated, so we forget about
|
||||
the ARP table entry. */
|
||||
if(i != ARP_TABLE_SIZE) {
|
||||
arp_table[i].state = ETHARP_STATE_EMPTY;
|
||||
/* We decrease the reference count of the queued pbuf (which now
|
||||
is dequeued). */
|
||||
DEBUGF(ETHARP_DEBUG, ("etharp_output: couldn't alloc pbuf for query, dequeueing %p\n", q));
|
||||
pbuf_free(q);
|
||||
}
|
||||
return NULL;
|
||||
}
|
||||
|
||||
hdr = p->payload;
|
||||
|
||||
hdr->opcode = htons(ARP_REQUEST);
|
||||
|
||||
for(i = 0; i < 6; ++i) {
|
||||
hdr->dhwaddr.addr[i] = 0x00;
|
||||
hdr->shwaddr.addr[i] = srcaddr->addr[i];
|
||||
}
|
||||
|
||||
ip_addr_set(&(hdr->dipaddr), ipaddr);
|
||||
ip_addr_set(&(hdr->sipaddr), &(netif->ip_addr));
|
||||
|
||||
hdr->hwtype = htons(HWTYPE_ETHERNET);
|
||||
ARPH_HWLEN_SET(hdr, 6);
|
||||
|
||||
hdr->proto = htons(ETHTYPE_IP);
|
||||
ARPH_PROTOLEN_SET(hdr, sizeof(struct ip_addr));
|
||||
|
||||
for(i = 0; i < 6; ++i) {
|
||||
hdr->ethhdr.dest.addr[i] = 0xff;
|
||||
hdr->ethhdr.src.addr[i] = srcaddr->addr[i];
|
||||
}
|
||||
|
||||
hdr->ethhdr.type = htons(ETHTYPE_ARP);
|
||||
return p;
|
||||
} else {
|
||||
/* destination Ethernet address known */
|
||||
if (dest != NULL) {
|
||||
/* A valid IP->MAC address mapping was found, so we construct the
|
||||
Ethernet header for the outgoing packet. */
|
||||
Ethernet header for the outgoing packet. */
|
||||
|
||||
ethhdr = q->payload;
|
||||
|
||||
|
||||
for(i = 0; i < 6; i++) {
|
||||
ethhdr->dest.addr[i] = dest->addr[i];
|
||||
ethhdr->src.addr[i] = srcaddr->addr[i];
|
||||
}
|
||||
|
||||
|
||||
ethhdr->type = htons(ETHTYPE_IP);
|
||||
|
||||
/* return the outgoing packet */
|
||||
return q;
|
||||
}
|
||||
// never reached; here for safety
|
||||
return NULL;
|
||||
}
|
||||
|
||||
/**
|
||||
* Clean up the ARP request that was allocated by ARP.
|
||||
* Free the ARP request pbuf.
|
||||
*
|
||||
* This must be called after you have sent the packet
|
||||
* returned by etharp_output(). It frees any pbuf
|
||||
* allocated for an ARP request.
|
||||
* Free the ARP request pbuf that was allocated by ARP
|
||||
*
|
||||
* as a result of calling etharp_output(). Must be called
|
||||
* with the pbuf returned by etharp_output(), after you
|
||||
* have sent that packet.
|
||||
*
|
||||
* @param p pbuf returned earlier by etharp_output().
|
||||
*
|
||||
* @see etharp_output().
|
||||
*/
|
||||
struct pbuf *
|
||||
etharp_output_sent(struct pbuf *p)
|
||||
@ -591,102 +606,84 @@ etharp_output_sent(struct pbuf *p)
|
||||
struct etharp_hdr *hdr;
|
||||
hdr=p->payload;
|
||||
if (hdr->opcode == htons(ARP_REQUEST)) {
|
||||
pbuf_free(p); p=NULL;
|
||||
};
|
||||
pbuf_free(p);
|
||||
p = NULL;
|
||||
}
|
||||
return p;
|
||||
}
|
||||
|
||||
/**
|
||||
* Initiate an ARP query for the given IP address.
|
||||
* Send an ARP request for the given IP address.
|
||||
*
|
||||
* Used by the DHCP module to support "gratuitous" ARP,
|
||||
* i.e. send ARP requests for one's own IP address, to
|
||||
* see if others have the IP address in use.
|
||||
* Sends an ARP request for the given IP address, unless
|
||||
* a request for this address is already pending. Optionally
|
||||
* queues an outgoing packet on the resulting ARP entry.
|
||||
*
|
||||
* Might be used in the future by manual IP configuration
|
||||
* @param netif The lwIP network interface where ipaddr
|
||||
* must be queried for.
|
||||
* @param ipaddr The IP address to be resolved.
|
||||
* @param q If non-NULL, a pbuf that must be queued on the
|
||||
* ARP entry for the ipaddr IP address.
|
||||
*
|
||||
* @return pbuf containing the ARP request, NULL on failure.
|
||||
*
|
||||
* @note Might be used in the future by manual IP configuration
|
||||
* as well.
|
||||
*
|
||||
*/
|
||||
|
||||
struct pbuf *etharp_query(struct netif *netif, struct ip_addr *ipaddr)
|
||||
struct pbuf *etharp_query(struct netif *netif, struct ip_addr *ipaddr, struct pbuf *q)
|
||||
{
|
||||
struct eth_addr *srcaddr;
|
||||
struct etharp_hdr *hdr;
|
||||
struct pbuf *p;
|
||||
u8_t i, j;
|
||||
u8_t maxtime;
|
||||
u8_t i;
|
||||
|
||||
srcaddr = (struct eth_addr *)netif->hwaddr;
|
||||
/* We check if we are already querying for this address. If so,
|
||||
we'll bail out. */
|
||||
for(i = 0; i < ARP_TABLE_SIZE; ++i)
|
||||
{
|
||||
if(arp_table[i].state == ETHARP_STATE_PENDING && ip_addr_cmp(ipaddr, &arp_table[i].ipaddr))
|
||||
{
|
||||
DEBUGF(ETHARP_DEBUG, ("etharp_output: already queued\n"));
|
||||
/* bail out if this IP address is pending */
|
||||
for(i = 0; i < ARP_TABLE_SIZE; ++i) {
|
||||
if(arp_table[i].state == ETHARP_STATE_PENDING &&
|
||||
ip_addr_cmp(ipaddr, &arp_table[i].ipaddr)) {
|
||||
DEBUGF(ETHARP_DEBUG, ("etharp_query: request already pending\n"));
|
||||
/* TODO: enqueue q here if possible (BEWARE: possible other packet already
|
||||
queued. */
|
||||
/* TODO: The host requirements RFC states that ARP should save at least one
|
||||
packet, and this should be the _latest_ packet. */
|
||||
/* TODO: use the ctime field to see how long ago an ARP request was sent,
|
||||
possibly retry. */
|
||||
return NULL;
|
||||
}
|
||||
}
|
||||
/* We now try to find an unused entry in the ARP table that we
|
||||
will setup and queue the outgoing packet. */
|
||||
for(i = 0; i < ARP_TABLE_SIZE; ++i)
|
||||
{
|
||||
if(arp_table[i].state == ETHARP_STATE_EMPTY)
|
||||
{
|
||||
break;
|
||||
}
|
||||
}
|
||||
|
||||
/* If no unused entry is found, we try to find the oldest entry and
|
||||
throw it away. */
|
||||
i = find_arp_entry();
|
||||
/* bail out if no ARP entries are available */
|
||||
if(i == ARP_TABLE_SIZE)
|
||||
{
|
||||
maxtime = 0;
|
||||
j = 0;
|
||||
for(i = 0; i < ARP_TABLE_SIZE; ++i)
|
||||
{
|
||||
if(arp_table[i].state == ETHARP_STATE_STABLE && ctime - arp_table[i].ctime > maxtime)
|
||||
{
|
||||
maxtime = ctime - arp_table[i].ctime;
|
||||
j = i;
|
||||
}
|
||||
}
|
||||
i = j;
|
||||
}
|
||||
|
||||
/* If all table entries were in pending state, we won't send out any
|
||||
more ARP requests. We'll just give up. */
|
||||
if(i == ARP_TABLE_SIZE)
|
||||
{
|
||||
DEBUGF(ETHARP_DEBUG, ("etharp_output: no more ARP table entries available.\n"));
|
||||
DEBUGF(ETHARP_DEBUG, ("etharp_query: no more ARP table entries available.\n"));
|
||||
return NULL;
|
||||
}
|
||||
|
||||
/* Now, i is the ARP table entry which we will fill with the new
|
||||
information. */
|
||||
ip_addr_set(&arp_table[i].ipaddr, ipaddr);
|
||||
/* for(k = 0; k < 6; ++k) {
|
||||
arp_table[i].ethaddr.addr[k] = dest->addr[k];
|
||||
}*/
|
||||
arp_table[i].ctime = ctime;
|
||||
arp_table[i].state = ETHARP_STATE_PENDING;
|
||||
arp_table[i].p = NULL;
|
||||
|
||||
/* We allocate a pbuf for the outgoing ARP request packet. */
|
||||
/* i is an available ARP table entry */
|
||||
/* allocate a pbuf for the outgoing ARP request packet */
|
||||
p = pbuf_alloc(PBUF_LINK, sizeof(struct etharp_hdr), PBUF_RAM);
|
||||
if(p == NULL)
|
||||
{
|
||||
/* No ARP request packet could be allocated, so we forget about
|
||||
the ARP table entry. */
|
||||
if(i != ARP_TABLE_SIZE)
|
||||
{
|
||||
arp_table[i].state = ETHARP_STATE_EMPTY;
|
||||
}
|
||||
/* could allocate pbuf? */
|
||||
if (p != NULL) {
|
||||
ip_addr_set(&arp_table[i].ipaddr, ipaddr);
|
||||
arp_table[i].ctime = ctime;
|
||||
arp_table[i].state = ETHARP_STATE_PENDING;
|
||||
/* remember pbuf to queue, if any */
|
||||
arp_table[i].p = q;
|
||||
/* any pbuf to queue? */
|
||||
if (q != NULL) {
|
||||
/* pbufs are queued, increase the reference count */
|
||||
pbuf_ref_chain(q);
|
||||
}
|
||||
}
|
||||
/* could not allocate pbuf for ARP request */
|
||||
else {
|
||||
return NULL;
|
||||
}
|
||||
|
||||
/* p is the allocated pbuf */
|
||||
|
||||
hdr = p->payload;
|
||||
|
||||
hdr->opcode = htons(ARP_REQUEST);
|
||||
|
||||
for(i = 0; i < 6; ++i)
|
||||
|
Loading…
x
Reference in New Issue
Block a user