freebsd-dev/crypto/heimdal/lib/roken/getifaddrs.c
Stanislav Sedov ae77177087 - Update FreeBSD Heimdal distribution to version 1.5.1. This also brings
several new kerberos related libraries and applications to FreeBSD:
  o kgetcred(1) allows one to manually get a ticket for a particular service.
  o kf(1) securily forwards ticket to another host through an authenticated
    and encrypted stream.
  o kcc(1) is an umbrella program around klist(1), kswitch(1), kgetcred(1)
    and other user kerberos operations. klist and kswitch are just symlinks
    to kcc(1) now.
  o kswitch(1) allows you to easily switch between kerberos credentials if
    you're running KCM.
  o hxtool(1) is a certificate management tool to use with PKINIT.
  o string2key(1) maps a password into key.
  o kdigest(8) is a userland tool to access the KDC's digest interface.
  o kimpersonate(8) creates a "fake" ticket for a service.

  We also now install manpages for some lirbaries that were not installed
  before, libheimntlm and libhx509.

- The new HEIMDAL version no longer supports Kerberos 4.  All users are
  recommended to switch to Kerberos 5.

- Weak ciphers are now disabled by default.  To enable DES support (used
  by telnet(8)), use "allow_weak_crypto" option in krb5.conf.

- libtelnet, pam_ksu and pam_krb5 are now compiled with error on warnings
  disabled due to the function they use (krb5_get_err_text(3)) being
  deprecated.  I plan to work on this next.

- Heimdal's KDC now require sqlite to operate.  We use the bundled version
  and install it as libheimsqlite.  If some other FreeBSD components will
  require it in the future we can rename it to libbsdsqlite and use for these
  components as well.

- This is not a latest Heimdal version, the new one was released while I was
  working on the update.  I will update it to 1.5.2 soon, as it fixes some
  important bugs and security issues.
2012-03-22 08:48:42 +00:00

1310 lines
32 KiB
C

/*
* Copyright (c) 2000 - 2002, 2005 Kungliga Tekniska Högskolan
* (Royal Institute of Technology, Stockholm, Sweden).
* 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 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.
*
* 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.
*/
#include <config.h>
#include "roken.h"
#ifdef __osf__
/* hate */
struct rtentry;
struct mbuf;
#endif
#ifdef HAVE_NET_IF_H
#include <net/if.h>
#endif
#ifdef HAVE_SYS_SOCKIO_H
#include <sys/sockio.h>
#endif /* HAVE_SYS_SOCKIO_H */
#ifdef HAVE_NETINET_IN6_VAR_H
#include <netinet/in6_var.h>
#endif /* HAVE_NETINET_IN6_VAR_H */
#include <ifaddrs.h>
#ifdef __hpux
#define lifconf if_laddrconf
#define lifc_len iflc_len
#define lifc_buf iflc_buf
#define lifc_req iflc_req
#define lifreq if_laddrreq
#define lifr_addr iflr_addr
#define lifr_name iflr_name
#define lifr_dstaddr iflr_dstaddr
#define lifr_broadaddr iflr_broadaddr
#define lifr_flags iflr_flags
#define lifr_index iflr_index
#endif
#ifdef AF_NETLINK
/*
* The linux - AF_NETLINK version of getifaddrs - from Usagi.
* Linux does not return v6 addresses from SIOCGIFCONF.
*/
/* $USAGI: ifaddrs.c,v 1.18 2002/03/06 01:50:46 yoshfuji Exp $ */
/**************************************************************************
* ifaddrs.c
* Copyright (C)2000 Hideaki YOSHIFUJI, 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 the author 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 THE AUTHOR 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 AUTHOR 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.
*/
#include "config.h"
#include <string.h>
#include <time.h>
#include <malloc.h>
#include <errno.h>
#include <unistd.h>
#include <sys/socket.h>
#include <asm/types.h>
#include <linux/netlink.h>
#include <linux/rtnetlink.h>
#include <sys/types.h>
#include <sys/socket.h>
#include <sys/poll.h>
#include <netpacket/packet.h>
#include <net/ethernet.h> /* the L2 protocols */
#include <sys/uio.h>
#include <net/if.h>
#include <net/if_arp.h>
#include <ifaddrs.h>
#include <netinet/in.h>
#define __set_errno(e) (errno = (e))
#define __close(fd) (close(fd))
#undef ifa_broadaddr
#define ifa_broadaddr ifa_dstaddr
#define IFA_NETMASK
/* ====================================================================== */
struct nlmsg_list{
struct nlmsg_list *nlm_next;
struct nlmsghdr *nlh;
int size;
time_t seq;
};
struct rtmaddr_ifamap {
void *address;
void *local;
#ifdef IFA_NETMASK
void *netmask;
#endif
void *broadcast;
#ifdef HAVE_IFADDRS_IFA_ANYCAST
void *anycast;
#endif
int address_len;
int local_len;
#ifdef IFA_NETMASK
int netmask_len;
#endif
int broadcast_len;
#ifdef HAVE_IFADDRS_IFA_ANYCAST
int anycast_len;
#endif
};
/* ====================================================================== */
static size_t
ifa_sa_len(sa_family_t family, int len)
{
size_t size;
switch(family){
case AF_INET:
size = sizeof(struct sockaddr_in);
break;
case AF_INET6:
size = sizeof(struct sockaddr_in6);
break;
case AF_PACKET:
size = (size_t)(((struct sockaddr_ll *)NULL)->sll_addr) + len;
if (size < sizeof(struct sockaddr_ll))
size = sizeof(struct sockaddr_ll);
break;
default:
size = (size_t)(((struct sockaddr *)NULL)->sa_data) + len;
if (size < sizeof(struct sockaddr))
size = sizeof(struct sockaddr);
break;
}
return size;
}
static void
ifa_make_sockaddr(sa_family_t family,
struct sockaddr *sa,
void *p, size_t len,
uint32_t scope, uint32_t scopeid)
{
if (sa == NULL) return;
switch(family){
case AF_INET:
memcpy(&((struct sockaddr_in*)sa)->sin_addr, (char *)p, len);
break;
case AF_INET6:
memcpy(&((struct sockaddr_in6*)sa)->sin6_addr, (char *)p, len);
if (IN6_IS_ADDR_LINKLOCAL(p) ||
IN6_IS_ADDR_MC_LINKLOCAL(p)){
((struct sockaddr_in6*)sa)->sin6_scope_id = scopeid;
}
break;
case AF_PACKET:
memcpy(((struct sockaddr_ll*)sa)->sll_addr, (char *)p, len);
((struct sockaddr_ll*)sa)->sll_halen = len;
break;
default:
memcpy(sa->sa_data, p, len); /*XXX*/
break;
}
sa->sa_family = family;
#ifdef HAVE_SOCKADDR_SA_LEN
sa->sa_len = ifa_sa_len(family, len);
#endif
}
#ifndef IFA_NETMASK
static struct sockaddr *
ifa_make_sockaddr_mask(sa_family_t family,
struct sockaddr *sa,
uint32_t prefixlen)
{
int i;
char *p = NULL, c;
uint32_t max_prefixlen = 0;
if (sa == NULL) return NULL;
switch(family){
case AF_INET:
memset(&((struct sockaddr_in*)sa)->sin_addr, 0, sizeof(((struct sockaddr_in*)sa)->sin_addr));
p = (char *)&((struct sockaddr_in*)sa)->sin_addr;
max_prefixlen = 32;
break;
case AF_INET6:
memset(&((struct sockaddr_in6*)sa)->sin6_addr, 0, sizeof(((struct sockaddr_in6*)sa)->sin6_addr));
p = (char *)&((struct sockaddr_in6*)sa)->sin6_addr;
#if 0 /* XXX: fill scope-id? */
if (IN6_IS_ADDR_LINKLOCAL(p) ||
IN6_IS_ADDR_MC_LINKLOCAL(p)){
((struct sockaddr_in6*)sa)->sin6_scope_id = scopeid;
}
#endif
max_prefixlen = 128;
break;
default:
return NULL;
}
sa->sa_family = family;
#ifdef HAVE_SOCKADDR_SA_LEN
sa->sa_len = ifa_sa_len(family, len);
#endif
if (p){
if (prefixlen > max_prefixlen)
prefixlen = max_prefixlen;
for (i=0; i<(prefixlen / 8); i++)
*p++ = 0xff;
c = 0xff;
c <<= (8 - (prefixlen % 8));
*p = c;
}
return sa;
}
#endif
/* ====================================================================== */
static int
nl_sendreq(int sd, int request, int flags, int *seq)
{
char reqbuf[NLMSG_ALIGN(sizeof(struct nlmsghdr)) +
NLMSG_ALIGN(sizeof(struct rtgenmsg))];
struct sockaddr_nl nladdr;
struct nlmsghdr *req_hdr;
struct rtgenmsg *req_msg;
time_t t = time(NULL);
if (seq) *seq = t;
memset(&reqbuf, 0, sizeof(reqbuf));
req_hdr = (struct nlmsghdr *)reqbuf;
req_msg = (struct rtgenmsg *)NLMSG_DATA(req_hdr);
req_hdr->nlmsg_len = NLMSG_LENGTH(sizeof(*req_msg));
req_hdr->nlmsg_type = request;
req_hdr->nlmsg_flags = flags | NLM_F_REQUEST;
req_hdr->nlmsg_pid = 0;
req_hdr->nlmsg_seq = t;
req_msg->rtgen_family = AF_UNSPEC;
memset(&nladdr, 0, sizeof(nladdr));
nladdr.nl_family = AF_NETLINK;
return (sendto(sd, (void *)req_hdr, req_hdr->nlmsg_len, 0,
(struct sockaddr *)&nladdr, sizeof(nladdr)));
}
static int
nl_recvmsg(int sd, int request, int seq,
void *buf, size_t buflen,
int *flags)
{
struct msghdr msg;
struct iovec iov = { buf, buflen };
struct sockaddr_nl nladdr;
int read_len;
for (;;){
msg.msg_name = (void *)&nladdr;
msg.msg_namelen = sizeof(nladdr);
msg.msg_iov = &iov;
msg.msg_iovlen = 1;
msg.msg_control = NULL;
msg.msg_controllen = 0;
msg.msg_flags = 0;
read_len = recvmsg(sd, &msg, 0);
if ((read_len < 0 && errno == EINTR) || (msg.msg_flags & MSG_TRUNC))
continue;
if (flags) *flags = msg.msg_flags;
break;
}
return read_len;
}
static int
nl_getmsg(int sd, int request, int seq,
struct nlmsghdr **nlhp,
int *done)
{
struct nlmsghdr *nh;
size_t bufsize = 65536, lastbufsize = 0;
void *buff = NULL;
int result = 0, read_size;
int msg_flags;
pid_t pid = getpid();
for (;;){
void *newbuff = realloc(buff, bufsize);
if (newbuff == NULL || bufsize < lastbufsize) {
result = -1;
break;
}
buff = newbuff;
result = read_size = nl_recvmsg(sd, request, seq, buff, bufsize, &msg_flags);
if (read_size < 0 || (msg_flags & MSG_TRUNC)){
lastbufsize = bufsize;
bufsize *= 2;
continue;
}
if (read_size == 0) break;
nh = (struct nlmsghdr *)buff;
for (nh = (struct nlmsghdr *)buff;
NLMSG_OK(nh, read_size);
nh = (struct nlmsghdr *)NLMSG_NEXT(nh, read_size)){
if (nh->nlmsg_pid != pid ||
nh->nlmsg_seq != seq)
continue;
if (nh->nlmsg_type == NLMSG_DONE){
(*done)++;
break; /* ok */
}
if (nh->nlmsg_type == NLMSG_ERROR){
struct nlmsgerr *nlerr = (struct nlmsgerr *)NLMSG_DATA(nh);
result = -1;
if (nh->nlmsg_len < NLMSG_LENGTH(sizeof(struct nlmsgerr)))
__set_errno(EIO);
else
__set_errno(-nlerr->error);
break;
}
}
break;
}
if (result < 0)
if (buff){
int saved_errno = errno;
free(buff);
__set_errno(saved_errno);
}
*nlhp = (struct nlmsghdr *)buff;
return result;
}
static int
nl_getlist(int sd, int seq,
int request,
struct nlmsg_list **nlm_list,
struct nlmsg_list **nlm_end)
{
struct nlmsghdr *nlh = NULL;
int status;
int done = 0;
int tries = 3;
try_again:
status = nl_sendreq(sd, request, NLM_F_ROOT|NLM_F_MATCH, &seq);
if (status < 0)
return status;
if (seq == 0)
seq = (int)time(NULL);
while(!done){
struct pollfd pfd;
pfd.fd = sd;
pfd.events = POLLIN | POLLPRI;
pfd.revents = 0;
status = poll(&pfd, 1, 1000);
if (status < 0)
return status;
else if (status == 0) {
seq++;
if (tries-- > 0)
goto try_again;
return -1;
}
status = nl_getmsg(sd, request, seq, &nlh, &done);
if (status < 0)
return status;
if (nlh){
struct nlmsg_list *nlm_next = (struct nlmsg_list *)malloc(sizeof(struct nlmsg_list));
if (nlm_next == NULL){
int saved_errno = errno;
free(nlh);
__set_errno(saved_errno);
status = -1;
} else {
nlm_next->nlm_next = NULL;
nlm_next->nlh = (struct nlmsghdr *)nlh;
nlm_next->size = status;
nlm_next->seq = seq;
if (*nlm_list == NULL){
*nlm_list = nlm_next;
*nlm_end = nlm_next;
} else {
(*nlm_end)->nlm_next = nlm_next;
*nlm_end = nlm_next;
}
}
}
}
return status >= 0 ? seq : status;
}
/* ---------------------------------------------------------------------- */
static void
free_nlmsglist(struct nlmsg_list *nlm0)
{
struct nlmsg_list *nlm, *nlm_next;
int saved_errno;
if (!nlm0)
return;
saved_errno = errno;
for (nlm=nlm0; nlm; nlm=nlm_next){
if (nlm->nlh)
free(nlm->nlh);
nlm_next=nlm->nlm_next;
free(nlm);
}
__set_errno(saved_errno);
}
static void
free_data(void *data, void *ifdata)
{
int saved_errno = errno;
if (data != NULL) free(data);
if (ifdata != NULL) free(ifdata);
__set_errno(saved_errno);
}
/* ---------------------------------------------------------------------- */
static void
nl_close(int sd)
{
int saved_errno = errno;
if (sd >= 0) __close(sd);
__set_errno(saved_errno);
}
/* ---------------------------------------------------------------------- */
static int
nl_open(void)
{
struct sockaddr_nl nladdr;
int sd;
sd = socket(PF_NETLINK, SOCK_RAW, NETLINK_ROUTE);
if (sd < 0) return -1;
memset(&nladdr, 0, sizeof(nladdr));
nladdr.nl_family = AF_NETLINK;
if (bind(sd, (struct sockaddr*)&nladdr, sizeof(nladdr)) < 0){
nl_close(sd);
return -1;
}
return sd;
}
/* ====================================================================== */
ROKEN_LIB_FUNCTION int ROKEN_LIB_CALL
rk_getifaddrs(struct ifaddrs **ifap)
{
int sd;
struct nlmsg_list *nlmsg_list, *nlmsg_end, *nlm;
/* - - - - - - - - - - - - - - - */
int icnt;
size_t dlen, xlen, nlen;
uint32_t max_ifindex = 0;
pid_t pid = getpid();
int seq;
int result;
int build ; /* 0 or 1 */
/* ---------------------------------- */
/* initialize */
icnt = dlen = xlen = nlen = 0;
nlmsg_list = nlmsg_end = NULL;
if (ifap)
*ifap = NULL;
/* ---------------------------------- */
/* open socket and bind */
sd = nl_open();
if (sd < 0)
return -1;
/* ---------------------------------- */
/* gather info */
if ((seq = nl_getlist(sd, 0, RTM_GETLINK,
&nlmsg_list, &nlmsg_end)) < 0){
free_nlmsglist(nlmsg_list);
nl_close(sd);
return -1;
}
if ((seq = nl_getlist(sd, seq+1, RTM_GETADDR,
&nlmsg_list, &nlmsg_end)) < 0){
free_nlmsglist(nlmsg_list);
nl_close(sd);
return -1;
}
/* ---------------------------------- */
/* Estimate size of result buffer and fill it */
for (build=0; build<=1; build++){
struct ifaddrs *ifl = NULL, *ifa = NULL;
struct nlmsghdr *nlh, *nlh0;
char *data = NULL, *xdata = NULL;
void *ifdata = NULL;
char *ifname = NULL, **iflist = NULL;
uint16_t *ifflist = NULL;
struct rtmaddr_ifamap ifamap;
if (build){
data = calloc(1,
NLMSG_ALIGN(sizeof(struct ifaddrs[icnt]))
+ dlen + xlen + nlen);
ifa = (struct ifaddrs *)data;
ifdata = calloc(1,
NLMSG_ALIGN(sizeof(char *[max_ifindex+1]))
+ NLMSG_ALIGN(sizeof(uint16_t [max_ifindex+1])));
if (ifap != NULL)
*ifap = (ifdata != NULL) ? ifa : NULL;
else{
free_data(data, ifdata);
result = 0;
break;
}
if (data == NULL || ifdata == NULL){
free_data(data, ifdata);
result = -1;
break;
}
ifl = NULL;
data += NLMSG_ALIGN(sizeof(struct ifaddrs)) * icnt;
xdata = data + dlen;
ifname = xdata + xlen;
iflist = ifdata;
ifflist = (uint16_t *)(((char *)iflist) + NLMSG_ALIGN(sizeof(char *[max_ifindex+1])));
}
for (nlm=nlmsg_list; nlm; nlm=nlm->nlm_next){
int nlmlen = nlm->size;
if (!(nlh0 = nlm->nlh))
continue;
for (nlh = nlh0;
NLMSG_OK(nlh, nlmlen);
nlh=NLMSG_NEXT(nlh,nlmlen)){
struct ifinfomsg *ifim = NULL;
struct ifaddrmsg *ifam = NULL;
struct rtattr *rta;
size_t nlm_struct_size = 0;
sa_family_t nlm_family = 0;
uint32_t nlm_scope = 0, nlm_index = 0;
size_t sockaddr_size = 0;
uint32_t nlm_prefixlen = 0;
size_t rtasize;
memset(&ifamap, 0, sizeof(ifamap));
/* check if the message is what we want */
if (nlh->nlmsg_pid != pid ||
nlh->nlmsg_seq != nlm->seq)
continue;
if (nlh->nlmsg_type == NLMSG_DONE){
break; /* ok */
}
switch (nlh->nlmsg_type){
case RTM_NEWLINK:
ifim = (struct ifinfomsg *)NLMSG_DATA(nlh);
nlm_struct_size = sizeof(*ifim);
nlm_family = ifim->ifi_family;
nlm_scope = 0;
nlm_index = ifim->ifi_index;
nlm_prefixlen = 0;
if (build)
ifflist[nlm_index] = ifa->ifa_flags = ifim->ifi_flags;
break;
case RTM_NEWADDR:
ifam = (struct ifaddrmsg *)NLMSG_DATA(nlh);
nlm_struct_size = sizeof(*ifam);
nlm_family = ifam->ifa_family;
nlm_scope = ifam->ifa_scope;
nlm_index = ifam->ifa_index;
nlm_prefixlen = ifam->ifa_prefixlen;
if (build)
ifa->ifa_flags = ifflist[nlm_index];
break;
default:
continue;
}
if (!build){
if (max_ifindex < nlm_index)
max_ifindex = nlm_index;
} else {
if (ifl != NULL)
ifl->ifa_next = ifa;
}
rtasize = NLMSG_PAYLOAD(nlh, nlmlen) - NLMSG_ALIGN(nlm_struct_size);
for (rta = (struct rtattr *)(((char *)NLMSG_DATA(nlh)) + NLMSG_ALIGN(nlm_struct_size));
RTA_OK(rta, rtasize);
rta = RTA_NEXT(rta, rtasize)){
struct sockaddr **sap = NULL;
void *rtadata = RTA_DATA(rta);
size_t rtapayload = RTA_PAYLOAD(rta);
socklen_t sa_len;
switch(nlh->nlmsg_type){
case RTM_NEWLINK:
switch(rta->rta_type){
case IFLA_ADDRESS:
case IFLA_BROADCAST:
if (build){
sap = (rta->rta_type == IFLA_ADDRESS) ? &ifa->ifa_addr : &ifa->ifa_broadaddr;
*sap = (struct sockaddr *)data;
}
sa_len = ifa_sa_len(AF_PACKET, rtapayload);
if (rta->rta_type == IFLA_ADDRESS)
sockaddr_size = NLMSG_ALIGN(sa_len);
if (!build){
dlen += NLMSG_ALIGN(sa_len);
} else {
memset(*sap, 0, sa_len);
ifa_make_sockaddr(AF_PACKET, *sap, rtadata,rtapayload, 0,0);
((struct sockaddr_ll *)*sap)->sll_ifindex = nlm_index;
((struct sockaddr_ll *)*sap)->sll_hatype = ifim->ifi_type;
data += NLMSG_ALIGN(sa_len);
}
break;
case IFLA_IFNAME:/* Name of Interface */
if (!build)
nlen += NLMSG_ALIGN(rtapayload + 1);
else{
ifa->ifa_name = ifname;
if (iflist[nlm_index] == NULL)
iflist[nlm_index] = ifa->ifa_name;
strncpy(ifa->ifa_name, rtadata, rtapayload);
ifa->ifa_name[rtapayload] = '\0';
ifname += NLMSG_ALIGN(rtapayload + 1);
}
break;
case IFLA_STATS:/* Statistics of Interface */
if (!build)
xlen += NLMSG_ALIGN(rtapayload);
else{
ifa->ifa_data = xdata;
memcpy(ifa->ifa_data, rtadata, rtapayload);
xdata += NLMSG_ALIGN(rtapayload);
}
break;
case IFLA_UNSPEC:
break;
case IFLA_MTU:
break;
case IFLA_LINK:
break;
case IFLA_QDISC:
break;
default:
break;
}
break;
case RTM_NEWADDR:
if (nlm_family == AF_PACKET) break;
switch(rta->rta_type){
case IFA_ADDRESS:
ifamap.address = rtadata;
ifamap.address_len = rtapayload;
break;
case IFA_LOCAL:
ifamap.local = rtadata;
ifamap.local_len = rtapayload;
break;
case IFA_BROADCAST:
ifamap.broadcast = rtadata;
ifamap.broadcast_len = rtapayload;
break;
#ifdef HAVE_IFADDRS_IFA_ANYCAST
case IFA_ANYCAST:
ifamap.anycast = rtadata;
ifamap.anycast_len = rtapayload;
break;
#endif
case IFA_LABEL:
if (!build)
nlen += NLMSG_ALIGN(rtapayload + 1);
else{
ifa->ifa_name = ifname;
if (iflist[nlm_index] == NULL)
iflist[nlm_index] = ifname;
strncpy(ifa->ifa_name, rtadata, rtapayload);
ifa->ifa_name[rtapayload] = '\0';
ifname += NLMSG_ALIGN(rtapayload + 1);
}
break;
case IFA_UNSPEC:
break;
case IFA_CACHEINFO:
break;
default:
break;
}
}
}
if (nlh->nlmsg_type == RTM_NEWADDR &&
nlm_family != AF_PACKET) {
if (!ifamap.local) {
ifamap.local = ifamap.address;
ifamap.local_len = ifamap.address_len;
}
if (!ifamap.address) {
ifamap.address = ifamap.local;
ifamap.address_len = ifamap.local_len;
}
if (ifamap.address_len != ifamap.local_len ||
(ifamap.address != NULL &&
memcmp(ifamap.address, ifamap.local, ifamap.address_len))) {
/* p2p; address is peer and local is ours */
ifamap.broadcast = ifamap.address;
ifamap.broadcast_len = ifamap.address_len;
ifamap.address = ifamap.local;
ifamap.address_len = ifamap.local_len;
}
if (ifamap.address) {
#ifndef IFA_NETMASK
sockaddr_size = NLMSG_ALIGN(ifa_sa_len(nlm_family,ifamap.address_len));
#endif
if (!build)
dlen += NLMSG_ALIGN(ifa_sa_len(nlm_family,ifamap.address_len));
else {
ifa->ifa_addr = (struct sockaddr *)data;
ifa_make_sockaddr(nlm_family, ifa->ifa_addr, ifamap.address, ifamap.address_len,
nlm_scope, nlm_index);
data += NLMSG_ALIGN(ifa_sa_len(nlm_family, ifamap.address_len));
}
}
#ifdef IFA_NETMASK
if (ifamap.netmask) {
if (!build)
dlen += NLMSG_ALIGN(ifa_sa_len(nlm_family,ifamap.netmask_len));
else {
ifa->ifa_netmask = (struct sockaddr *)data;
ifa_make_sockaddr(nlm_family, ifa->ifa_netmask, ifamap.netmask, ifamap.netmask_len,
nlm_scope, nlm_index);
data += NLMSG_ALIGN(ifa_sa_len(nlm_family, ifamap.netmask_len));
}
}
#endif
if (ifamap.broadcast) {
if (!build)
dlen += NLMSG_ALIGN(ifa_sa_len(nlm_family,ifamap.broadcast_len));
else {
ifa->ifa_broadaddr = (struct sockaddr *)data;
ifa_make_sockaddr(nlm_family, ifa->ifa_broadaddr, ifamap.broadcast, ifamap.broadcast_len,
nlm_scope, nlm_index);
data += NLMSG_ALIGN(ifa_sa_len(nlm_family, ifamap.broadcast_len));
}
}
#ifdef HAVE_IFADDRS_IFA_ANYCAST
if (ifamap.anycast) {
if (!build)
dlen += NLMSG_ALIGN(ifa_sa_len(nlm_family,ifamap.anycast_len));
else {
ifa->ifa_anycast = (struct sockaddr *)data;
ifa_make_sockaddr(nlm_family, ifa->ifa_anyaddr, ifamap.anycast, ifamap.anycast_len,
nlm_scope, nlm_index);
data += NLMSG_ALIGN(ifa_sa_len(nlm_family, ifamap.anycast_len));
}
}
#endif
}
if (!build){
#ifndef IFA_NETMASK
dlen += sockaddr_size;
#endif
icnt++;
} else {
if (ifa->ifa_name == NULL)
ifa->ifa_name = iflist[nlm_index];
#ifndef IFA_NETMASK
if (ifa->ifa_addr &&
ifa->ifa_addr->sa_family != AF_UNSPEC &&
ifa->ifa_addr->sa_family != AF_PACKET){
ifa->ifa_netmask = (struct sockaddr *)data;
ifa_make_sockaddr_mask(ifa->ifa_addr->sa_family, ifa->ifa_netmask, nlm_prefixlen);
}
data += sockaddr_size;
#endif
ifl = ifa++;
}
}
}
if (!build){
if (icnt == 0 && (dlen + nlen + xlen == 0)){
if (ifap != NULL)
*ifap = NULL;
break; /* cannot found any addresses */
}
}
else
free_data(NULL, ifdata);
}
/* ---------------------------------- */
/* Finalize */
free_nlmsglist(nlmsg_list);
nl_close(sd);
return 0;
}
void ROKEN_LIB_FUNCTION
rk_freeifaddrs(struct ifaddrs *ifp)
{
/* AF_NETLINK method uses a single allocation for all interfaces */
free(ifp);
}
#else /* !AF_NETLINK */
/*
* The generic SIOCGIFCONF version.
*/
static int
getifaddrs2(struct ifaddrs **ifap,
int af, int siocgifconf, int siocgifflags,
size_t ifreq_sz)
{
int ret;
int fd;
size_t buf_size;
char *buf;
struct ifconf ifconf;
char *p;
size_t sz;
struct sockaddr sa_zero;
struct ifreq *ifr;
struct ifaddrs *start = NULL, **end = &start;
buf = NULL;
memset (&sa_zero, 0, sizeof(sa_zero));
fd = socket(af, SOCK_DGRAM, 0);
if (fd < 0)
return -1;
buf_size = 8192;
for (;;) {
buf = calloc(1, buf_size);
if (buf == NULL) {
ret = ENOMEM;
goto error_out;
}
ifconf.ifc_len = buf_size;
ifconf.ifc_buf = buf;
/*
* Solaris returns EINVAL when the buffer is too small.
*/
if (ioctl (fd, siocgifconf, &ifconf) < 0 && errno != EINVAL) {
ret = errno;
goto error_out;
}
/*
* Can the difference between a full and a overfull buf
* be determined?
*/
if (ifconf.ifc_len < buf_size)
break;
free (buf);
buf_size *= 2;
}
for (p = ifconf.ifc_buf;
p < ifconf.ifc_buf + ifconf.ifc_len;
p += sz) {
struct ifreq ifreq;
struct sockaddr *sa;
size_t salen;
ifr = (struct ifreq *)p;
sa = &ifr->ifr_addr;
sz = ifreq_sz;
salen = sizeof(struct sockaddr);
#ifdef HAVE_STRUCT_SOCKADDR_SA_LEN
salen = sa->sa_len;
sz = max(sz, sizeof(ifr->ifr_name) + sa->sa_len);
#endif
#ifdef SA_LEN
salen = SA_LEN(sa);
sz = max(sz, sizeof(ifr->ifr_name) + SA_LEN(sa));
#endif
memset (&ifreq, 0, sizeof(ifreq));
memcpy (ifreq.ifr_name, ifr->ifr_name, sizeof(ifr->ifr_name));
if (ioctl(fd, siocgifflags, &ifreq) < 0) {
ret = errno;
goto error_out;
}
*end = malloc(sizeof(**end));
if (*end == NULL) {
ret = ENOMEM;
goto error_out;
}
(*end)->ifa_next = NULL;
(*end)->ifa_name = strdup(ifr->ifr_name);
if ((*end)->ifa_name == NULL) {
ret = ENOMEM;
goto error_out;
}
(*end)->ifa_flags = ifreq.ifr_flags;
(*end)->ifa_addr = malloc(salen);
if ((*end)->ifa_addr == NULL) {
ret = ENOMEM;
goto error_out;
}
memcpy((*end)->ifa_addr, sa, salen);
(*end)->ifa_netmask = NULL;
#if 0
/* fix these when we actually need them */
if(ifreq.ifr_flags & IFF_BROADCAST) {
(*end)->ifa_broadaddr = malloc(sizeof(ifr->ifr_broadaddr));
if ((*end)->ifa_broadaddr == NULL) {
ret = ENOMEM;
goto error_out;
}
memcpy((*end)->ifa_broadaddr, &ifr->ifr_broadaddr,
sizeof(ifr->ifr_broadaddr));
} else if(ifreq.ifr_flags & IFF_POINTOPOINT) {
(*end)->ifa_dstaddr = malloc(sizeof(ifr->ifr_dstaddr));
if ((*end)->ifa_dstaddr == NULL) {
ret = ENOMEM;
goto error_out;
}
memcpy((*end)->ifa_dstaddr, &ifr->ifr_dstaddr,
sizeof(ifr->ifr_dstaddr));
} else
(*end)->ifa_dstaddr = NULL;
#else
(*end)->ifa_dstaddr = NULL;
#endif
(*end)->ifa_data = NULL;
end = &(*end)->ifa_next;
}
*ifap = start;
close(fd);
free(buf);
return 0;
error_out:
rk_freeifaddrs(start);
close(fd);
free(buf);
errno = ret;
return -1;
}
#if defined(HAVE_IPV6) && defined(SIOCGLIFCONF) && defined(SIOCGLIFFLAGS)
static int
getlifaddrs2(struct ifaddrs **ifap,
int af, int siocgifconf, int siocgifflags,
size_t ifreq_sz)
{
int ret;
int fd;
size_t buf_size;
char *buf;
struct lifconf ifconf;
char *p;
size_t sz;
struct sockaddr sa_zero;
struct lifreq *ifr;
struct ifaddrs *start = NULL, **end = &start;
buf = NULL;
memset (&sa_zero, 0, sizeof(sa_zero));
fd = socket(af, SOCK_DGRAM, 0);
if (fd < 0)
return -1;
buf_size = 8192;
for (;;) {
buf = calloc(1, buf_size);
if (buf == NULL) {
ret = ENOMEM;
goto error_out;
}
#ifndef __hpux
ifconf.lifc_family = af;
ifconf.lifc_flags = 0;
#endif
ifconf.lifc_len = buf_size;
ifconf.lifc_buf = buf;
/*
* Solaris returns EINVAL when the buffer is too small.
*/
if (ioctl (fd, siocgifconf, &ifconf) < 0 && errno != EINVAL) {
ret = errno;
goto error_out;
}
/*
* Can the difference between a full and a overfull buf
* be determined?
*/
if (ifconf.lifc_len < buf_size)
break;
free (buf);
buf_size *= 2;
}
for (p = ifconf.lifc_buf;
p < ifconf.lifc_buf + ifconf.lifc_len;
p += sz) {
struct lifreq ifreq;
struct sockaddr_storage *sa;
size_t salen;
ifr = (struct lifreq *)p;
sa = &ifr->lifr_addr;
sz = ifreq_sz;
salen = sizeof(struct sockaddr_storage);
#ifdef HAVE_STRUCT_SOCKADDR_SA_LEN
salen = sa->sa_len;
sz = max(sz, sizeof(ifr->ifr_name) + sa->sa_len);
#endif
#ifdef SA_LEN
salen = SA_LEN(sa);
sz = max(sz, sizeof(ifr->ifr_name) + SA_LEN(sa));
#endif
memset (&ifreq, 0, sizeof(ifreq));
memcpy (ifreq.lifr_name, ifr->lifr_name, sizeof(ifr->lifr_name));
if (ioctl(fd, siocgifflags, &ifreq) < 0) {
ret = errno;
goto error_out;
}
*end = malloc(sizeof(**end));
if (*end == NULL) {
ret = ENOMEM;
goto error_out;
}
(*end)->ifa_next = NULL;
(*end)->ifa_name = strdup(ifr->lifr_name);
if ((*end)->ifa_name == NULL) {
ret = ENOMEM;
goto error_out;
}
(*end)->ifa_flags = ifreq.lifr_flags;
(*end)->ifa_addr = malloc(salen);
if ((*end)->ifa_addr == NULL) {
ret = ENOMEM;
goto error_out;
}
memcpy((*end)->ifa_addr, sa, salen);
(*end)->ifa_netmask = NULL;
#if 0
/* fix these when we actually need them */
if(ifreq.ifr_flags & IFF_BROADCAST) {
(*end)->ifa_broadaddr = malloc(sizeof(ifr->ifr_broadaddr));
if ((*end)->ifa_broadaddr == NULL) {
ret = ENOMEM;
goto error_out;
}
memcpy((*end)->ifa_broadaddr, &ifr->ifr_broadaddr,
sizeof(ifr->ifr_broadaddr));
} else if(ifreq.ifr_flags & IFF_POINTOPOINT) {
(*end)->ifa_dstaddr = malloc(sizeof(ifr->ifr_dstaddr));
if ((*end)->ifa_dstaddr == NULL) {
ret = ENOMEM;
goto error_out;
}
memcpy((*end)->ifa_dstaddr, &ifr->ifr_dstaddr,
sizeof(ifr->ifr_dstaddr));
} else
(*end)->ifa_dstaddr = NULL;
#else
(*end)->ifa_dstaddr = NULL;
#endif
(*end)->ifa_data = NULL;
end = &(*end)->ifa_next;
}
*ifap = start;
close(fd);
free(buf);
return 0;
error_out:
rk_freeifaddrs(start);
close(fd);
free(buf);
errno = ret;
return -1;
}
#endif /* defined(HAVE_IPV6) && defined(SIOCGLIFCONF) && defined(SIOCGLIFFLAGS) */
/**
* Join two struct ifaddrs lists by appending supp to base.
* Either may be NULL. The new list head (usually base) will be
* returned.
*/
static struct ifaddrs *
append_ifaddrs(struct ifaddrs *base, struct ifaddrs *supp) {
if (!base)
return supp;
if (!supp)
return base;
while (base->ifa_next)
base = base->ifa_next;
base->ifa_next = supp;
return base;
}
ROKEN_LIB_FUNCTION int ROKEN_LIB_CALL
rk_getifaddrs(struct ifaddrs **ifap)
{
int ret = -1;
errno = ENXIO;
#if defined(AF_INET6) && defined(SIOCGIF6CONF) && defined(SIOCGIF6FLAGS)
if (ret)
ret = getifaddrs2 (ifap, AF_INET6, SIOCGIF6CONF, SIOCGIF6FLAGS,
sizeof(struct in6_ifreq));
#endif
#if defined(HAVE_IPV6) && defined(SIOCGLIFCONF) && defined(SIOCGLIFFLAGS)
/* Do IPv6 and IPv4 queries separately then join the result.
*
* HP-UX only returns IPv6 addresses using SIOCGLIFCONF,
* SIOCGIFCONF has to be used for IPv4 addresses. The result is then
* merged.
*
* Solaris needs particular care, because a SIOCGLIFCONF lookup using
* AF_UNSPEC can fail in a Zone requiring an AF_INET lookup, so we just
* do them separately the same as for HP-UX. See
* http://repo.or.cz/w/heimdal.git/commitdiff/76afc31e9ba2f37e64c70adc006ade9e37e9ef73
*/
if (ret) {
int v6err, v4err;
struct ifaddrs *v6addrs, *v4addrs;
v6err = getlifaddrs2 (&v6addrs, AF_INET6, SIOCGLIFCONF, SIOCGLIFFLAGS,
sizeof(struct lifreq));
v4err = getifaddrs2 (&v4addrs, AF_INET, SIOCGIFCONF, SIOCGIFFLAGS,
sizeof(struct ifreq));
if (v6err)
v6addrs = NULL;
if (v4err)
v4addrs = NULL;
if (v6addrs) {
if (v4addrs)
*ifap = append_ifaddrs(v6addrs, v4addrs);
else
*ifap = v6addrs;
} else if (v4addrs) {
*ifap = v4addrs;
} else {
*ifap = NULL;
}
ret = (v6err || v4err) ? -1 : 0;
}
#endif
#if defined(HAVE_IPV6) && defined(SIOCGIFCONF)
if (ret)
ret = getifaddrs2 (ifap, AF_INET6, SIOCGIFCONF, SIOCGIFFLAGS,
sizeof(struct ifreq));
#endif
#if defined(AF_INET) && defined(SIOCGIFCONF) && defined(SIOCGIFFLAGS)
if (ret)
ret = getifaddrs2 (ifap, AF_INET, SIOCGIFCONF, SIOCGIFFLAGS,
sizeof(struct ifreq));
#endif
return ret;
}
ROKEN_LIB_FUNCTION void ROKEN_LIB_CALL
rk_freeifaddrs(struct ifaddrs *ifp)
{
struct ifaddrs *p, *q;
for(p = ifp; p; ) {
free(p->ifa_name);
if(p->ifa_addr)
free(p->ifa_addr);
if(p->ifa_dstaddr)
free(p->ifa_dstaddr);
if(p->ifa_netmask)
free(p->ifa_netmask);
if(p->ifa_data)
free(p->ifa_data);
q = p;
p = p->ifa_next;
free(q);
}
}
#endif /* !AF_NETLINK */
#ifdef TEST
void
print_addr(const char *s, struct sockaddr *sa)
{
int i;
printf(" %s=%d/", s, sa->sa_family);
#ifdef HAVE_STRUCT_SOCKADDR_SA_LEN
for(i = 0; i < sa->sa_len - ((long)sa->sa_data - (long)&sa->sa_family); i++)
printf("%02x", ((unsigned char*)sa->sa_data)[i]);
#else
for(i = 0; i < sizeof(sa->sa_data); i++)
printf("%02x", ((unsigned char*)sa->sa_data)[i]);
#endif
printf("\n");
}
void
print_ifaddrs(struct ifaddrs *x)
{
struct ifaddrs *p;
for(p = x; p; p = p->ifa_next) {
printf("%s\n", p->ifa_name);
printf(" flags=%x\n", p->ifa_flags);
if(p->ifa_addr)
print_addr("addr", p->ifa_addr);
if(p->ifa_dstaddr)
print_addr("dstaddr", p->ifa_dstaddr);
if(p->ifa_netmask)
print_addr("netmask", p->ifa_netmask);
printf(" %p\n", p->ifa_data);
}
}
int
main()
{
struct ifaddrs *a = NULL, *b;
getifaddrs2(&a, AF_INET, SIOCGIFCONF, SIOCGIFFLAGS, sizeof(struct ifreq));
print_ifaddrs(a);
printf("---\n");
getifaddrs(&b);
print_ifaddrs(b);
return 0;
}
#endif