freebsd-dev/contrib/bsnmp/snmp_mibII/mibII.c
2020-04-01 15:25:16 +00:00

1871 lines
42 KiB
C

/*
* Copyright (c) 2001-2003
* Fraunhofer Institute for Open Communication Systems (FhG Fokus).
* All rights reserved.
*
* Author: Harti Brandt <harti@freebsd.org>
*
* 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.
*
* THIS SOFTWARE IS PROVIDED BY 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 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.
*
* $Begemot: mibII.c 516 2006-10-27 15:54:02Z brandt_h $
*
* Implementation of the standard interfaces and ip MIB.
*/
#include "mibII.h"
#include "mibII_oid.h"
#include <net/if.h>
#include <net/if_types.h>
/*****************************/
/* our module */
static struct lmodule *module;
/* routing socket */
static int route;
static void *route_fd;
/* if-index allocator */
static uint32_t next_if_index = 1;
/* currently fetching the arp table */
static int in_update_arp;
/* OR registrations */
static u_int ifmib_reg;
static u_int ipmib_reg;
static u_int tcpmib_reg;
static u_int udpmib_reg;
static u_int ipForward_reg;
/*****************************/
/* list of all IP addresses */
struct mibifa_list mibifa_list = TAILQ_HEAD_INITIALIZER(mibifa_list);
/* list of all interfaces */
struct mibif_list mibif_list = TAILQ_HEAD_INITIALIZER(mibif_list);
/* list of dynamic interface names */
struct mibdynif_list mibdynif_list = SLIST_HEAD_INITIALIZER(mibdynif_list);
/* list of all interface index mappings */
struct mibindexmap_list mibindexmap_list = STAILQ_HEAD_INITIALIZER(mibindexmap_list);
/* list of all stacking entries */
struct mibifstack_list mibifstack_list = TAILQ_HEAD_INITIALIZER(mibifstack_list);
/* list of all receive addresses */
struct mibrcvaddr_list mibrcvaddr_list = TAILQ_HEAD_INITIALIZER(mibrcvaddr_list);
/* list of all NetToMedia entries */
struct mibarp_list mibarp_list = TAILQ_HEAD_INITIALIZER(mibarp_list);
/* number of interfaces */
int32_t mib_if_number;
/* last change of table */
uint64_t mib_iftable_last_change;
/* last change of stack table */
uint64_t mib_ifstack_last_change;
/* if this is set, one of our lists may be bad. refresh them when idle */
int mib_iflist_bad;
/* network socket */
int mib_netsock;
/* last time refreshed */
uint64_t mibarpticks;
/* info on system clocks */
struct clockinfo clockinfo;
/* list of all New if registrations */
static struct newifreg_list newifreg_list = TAILQ_HEAD_INITIALIZER(newifreg_list);
/* baud rate of fastest interface */
uint64_t mibif_maxspeed;
/* user-forced update interval */
u_int mibif_force_hc_update_interval;
/* current update interval */
u_int mibif_hc_update_interval;
/* HC update timer handle */
static void *hc_update_timer;
/* Idle poll timer */
static void *mibII_poll_timer;
/* interfaces' data poll interval */
u_int mibII_poll_ticks;
/* Idle poll hook */
static void mibII_idle(void *arg __unused);
/*****************************/
static const struct asn_oid oid_ifMIB = OIDX_ifMIB;
static const struct asn_oid oid_ipMIB = OIDX_ipMIB;
static const struct asn_oid oid_tcpMIB = OIDX_tcpMIB;
static const struct asn_oid oid_udpMIB = OIDX_udpMIB;
static const struct asn_oid oid_ipForward = OIDX_ipForward;
static const struct asn_oid oid_linkDown = OIDX_linkDown;
static const struct asn_oid oid_linkUp = OIDX_linkUp;
static const struct asn_oid oid_ifIndex = OIDX_ifIndex;
/*****************************/
/*
* Find an interface
*/
struct mibif *
mib_find_if(u_int idx)
{
struct mibif *ifp;
TAILQ_FOREACH(ifp, &mibif_list, link)
if (ifp->index == idx)
return (ifp);
return (NULL);
}
struct mibif *
mib_find_if_sys(u_int sysindex)
{
struct mibif *ifp;
TAILQ_FOREACH(ifp, &mibif_list, link)
if (ifp->sysindex == sysindex)
return (ifp);
return (NULL);
}
struct mibif *
mib_find_if_name(const char *name)
{
struct mibif *ifp;
TAILQ_FOREACH(ifp, &mibif_list, link)
if (strcmp(ifp->name, name) == 0)
return (ifp);
return (NULL);
}
/*
* Check whether an interface is dynamic. The argument may include the
* unit number. This assumes, that the name part does NOT contain digits.
*/
int
mib_if_is_dyn(const char *name)
{
size_t len;
struct mibdynif *d;
for (len = 0; name[len] != '\0' && isalpha(name[len]) ; len++)
;
SLIST_FOREACH(d, &mibdynif_list, link)
if (strlen(d->name) == len && strncmp(d->name, name, len) == 0)
return (1);
return (0);
}
/* set an interface name to dynamic mode */
void
mib_if_set_dyn(const char *name)
{
struct mibdynif *d;
SLIST_FOREACH(d, &mibdynif_list, link)
if (strcmp(name, d->name) == 0)
return;
if ((d = malloc(sizeof(*d))) == NULL)
err(1, NULL);
strlcpy(d->name, name, sizeof(d->name));
SLIST_INSERT_HEAD(&mibdynif_list, d, link);
}
/*
* register for interface creations
*/
int
mib_register_newif(int (*func)(struct mibif *), const struct lmodule *mod)
{
struct newifreg *reg;
TAILQ_FOREACH(reg, &newifreg_list, link)
if (reg->mod == mod) {
reg->func = func;
return (0);
}
if ((reg = malloc(sizeof(*reg))) == NULL) {
syslog(LOG_ERR, "newifreg: %m");
return (-1);
}
reg->mod = mod;
reg->func = func;
TAILQ_INSERT_TAIL(&newifreg_list, reg, link);
return (0);
}
void
mib_unregister_newif(const struct lmodule *mod)
{
struct newifreg *reg;
TAILQ_FOREACH(reg, &newifreg_list, link)
if (reg->mod == mod) {
TAILQ_REMOVE(&newifreg_list, reg, link);
free(reg);
return;
}
}
struct mibif *
mib_first_if(void)
{
return (TAILQ_FIRST(&mibif_list));
}
struct mibif *
mib_next_if(const struct mibif *ifp)
{
return (TAILQ_NEXT(ifp, link));
}
/*
* Change the admin status of an interface
*/
int
mib_if_admin(struct mibif *ifp, int up)
{
struct ifreq ifr;
strlcpy(ifr.ifr_name, ifp->name, sizeof(ifr.ifr_name));
if (ioctl(mib_netsock, SIOCGIFFLAGS, &ifr) == -1) {
syslog(LOG_ERR, "SIOCGIFFLAGS(%s): %m", ifp->name);
return (-1);
}
if (up)
ifr.ifr_flags |= IFF_UP;
else
ifr.ifr_flags &= ~IFF_UP;
if (ioctl(mib_netsock, SIOCSIFFLAGS, &ifr) == -1) {
syslog(LOG_ERR, "SIOCSIFFLAGS(%s): %m", ifp->name);
return (-1);
}
(void)mib_fetch_ifmib(ifp);
return (0);
}
/*
* Generate a link up/down trap
*/
static void
link_trap(struct mibif *ifp, int up)
{
struct snmp_value ifindex;
ifindex.var = oid_ifIndex;
ifindex.var.subs[ifindex.var.len++] = ifp->index;
ifindex.syntax = SNMP_SYNTAX_INTEGER;
ifindex.v.integer = ifp->index;
snmp_send_trap(up ? &oid_linkUp : &oid_linkDown, &ifindex,
(struct snmp_value *)NULL);
}
/**
* Fetch the GENERIC IFMIB and update the HC counters
*/
static int
fetch_generic_mib(struct mibif *ifp, const struct ifmibdata *old)
{
int name[6];
size_t len;
struct mibif_private *p = ifp->private;
name[0] = CTL_NET;
name[1] = PF_LINK;
name[2] = NETLINK_GENERIC;
name[3] = IFMIB_IFDATA;
name[4] = ifp->sysindex;
name[5] = IFDATA_GENERAL;
len = sizeof(ifp->mib);
if (sysctl(name, nitems(name), &ifp->mib, &len, NULL, 0) == -1) {
if (errno != ENOENT)
syslog(LOG_WARNING, "sysctl(ifmib, %s) failed %m",
ifp->name);
return (-1);
}
/*
* Assume that one of the two following compounds is optimized away
*/
if (ULONG_MAX >= 0xffffffffffffffffULL) {
p->hc_inoctets = ifp->mib.ifmd_data.ifi_ibytes;
p->hc_outoctets = ifp->mib.ifmd_data.ifi_obytes;
p->hc_omcasts = ifp->mib.ifmd_data.ifi_omcasts;
p->hc_opackets = ifp->mib.ifmd_data.ifi_opackets;
p->hc_imcasts = ifp->mib.ifmd_data.ifi_imcasts;
p->hc_ipackets = ifp->mib.ifmd_data.ifi_ipackets;
} else if (ULONG_MAX >= 0xffffffff) {
#define UPDATE(HC, MIB) \
if (old->ifmd_data.MIB > ifp->mib.ifmd_data.MIB) \
p->HC += (0x100000000ULL + \
ifp->mib.ifmd_data.MIB) - \
old->ifmd_data.MIB; \
else \
p->HC += ifp->mib.ifmd_data.MIB - \
old->ifmd_data.MIB;
UPDATE(hc_inoctets, ifi_ibytes)
UPDATE(hc_outoctets, ifi_obytes)
UPDATE(hc_omcasts, ifi_omcasts)
UPDATE(hc_opackets, ifi_opackets)
UPDATE(hc_imcasts, ifi_imcasts)
UPDATE(hc_ipackets, ifi_ipackets)
#undef UPDATE
} else
abort();
return (0);
}
/**
* Update the 64-bit interface counters
*/
static void
update_hc_counters(void *arg __unused)
{
struct mibif *ifp;
struct ifmibdata oldmib;
TAILQ_FOREACH(ifp, &mibif_list, link) {
oldmib = ifp->mib;
(void)fetch_generic_mib(ifp, &oldmib);
}
}
/**
* Recompute the poll timer for the HC counters
*/
void
mibif_reset_hc_timer(void)
{
u_int ticks;
if ((ticks = mibif_force_hc_update_interval) == 0) {
if (mibif_maxspeed <= IF_Mbps(10)) {
/* at 10Mbps overflow needs 3436 seconds */
ticks = 3000 * 100; /* 50 minutes */
} else if (mibif_maxspeed <= IF_Mbps(100)) {
/* at 100Mbps overflow needs 343 seconds */
ticks = 300 * 100; /* 5 minutes */
} else if (mibif_maxspeed < IF_Mbps(622)) {
/* at 622Mbps overflow needs 53 seconds */
ticks = 40 * 100; /* 40 seconds */
} else if (mibif_maxspeed <= IF_Mbps(1000)) {
/* at 1Gbps overflow needs 34 seconds */
ticks = 20 * 100; /* 20 seconds */
} else {
/* at 10Gbps overflow needs 3.4 seconds */
ticks = 100; /* 1 seconds */
}
}
if (ticks == mibif_hc_update_interval)
return;
if (hc_update_timer != NULL) {
timer_stop(hc_update_timer);
hc_update_timer = NULL;
}
update_hc_counters(NULL);
if ((hc_update_timer = timer_start_repeat(ticks, ticks,
update_hc_counters, NULL, module)) == NULL) {
syslog(LOG_ERR, "timer_start(%u): %m", ticks);
return;
}
mibif_hc_update_interval = ticks;
}
/**
* Restart the idle poll timer.
*/
void
mibif_restart_mibII_poll_timer(void)
{
if (mibII_poll_timer != NULL)
timer_stop(mibII_poll_timer);
if ((mibII_poll_timer = timer_start_repeat(mibII_poll_ticks * 10,
mibII_poll_ticks * 10, mibII_idle, NULL, module)) == NULL)
syslog(LOG_ERR, "timer_start(%u): %m", mibII_poll_ticks);
}
/*
* Fetch new MIB data.
*/
int
mib_fetch_ifmib(struct mibif *ifp)
{
static int kmib[2] = { -1, 0 }; /* for sysctl net.ifdescr_maxlen */
int name[6];
size_t kmiblen = nitems(kmib);
size_t len;
void *newmib;
struct ifmibdata oldmib = ifp->mib;
struct ifreq irr;
u_int alias_maxlen = MIBIF_ALIAS_SIZE_MAX;
if (fetch_generic_mib(ifp, &oldmib) == -1)
return (-1);
/*
* Quoting RFC2863, 3.1.15: "... LinkUp and linkDown traps are
* generated just after ifOperStatus leaves, or just before it
* enters, the down state, respectively;"
*/
if (ifp->trap_enable && ifp->mib.ifmd_data.ifi_link_state !=
oldmib.ifmd_data.ifi_link_state &&
(ifp->mib.ifmd_data.ifi_link_state == LINK_STATE_DOWN ||
oldmib.ifmd_data.ifi_link_state == LINK_STATE_DOWN))
link_trap(ifp, ifp->mib.ifmd_data.ifi_link_state ==
LINK_STATE_UP ? 1 : 0);
ifp->flags &= ~(MIBIF_HIGHSPEED | MIBIF_VERYHIGHSPEED);
if (ifp->mib.ifmd_data.ifi_baudrate > 20000000) {
ifp->flags |= MIBIF_HIGHSPEED;
if (ifp->mib.ifmd_data.ifi_baudrate > 650000000)
ifp->flags |= MIBIF_VERYHIGHSPEED;
}
if (ifp->mib.ifmd_data.ifi_baudrate > mibif_maxspeed) {
mibif_maxspeed = ifp->mib.ifmd_data.ifi_baudrate;
mibif_reset_hc_timer();
}
/*
* linkspecific MIB
*/
name[0] = CTL_NET;
name[1] = PF_LINK;
name[2] = NETLINK_GENERIC;
name[3] = IFMIB_IFDATA;
name[4] = ifp->sysindex;
name[5] = IFDATA_LINKSPECIFIC;
if (sysctl(name, nitems(name), NULL, &len, NULL, 0) == -1) {
syslog(LOG_WARNING, "sysctl linkmib estimate (%s): %m",
ifp->name);
if (ifp->specmib != NULL) {
ifp->specmib = NULL;
ifp->specmiblen = 0;
}
goto out;
}
if (len == 0) {
if (ifp->specmib != NULL) {
ifp->specmib = NULL;
ifp->specmiblen = 0;
}
goto out;
}
if (ifp->specmiblen != len) {
if ((newmib = realloc(ifp->specmib, len)) == NULL) {
ifp->specmib = NULL;
ifp->specmiblen = 0;
goto out;
}
ifp->specmib = newmib;
ifp->specmiblen = len;
}
if (sysctl(name, nitems(name), ifp->specmib, &len, NULL, 0) == -1) {
syslog(LOG_WARNING, "sysctl linkmib (%s): %m", ifp->name);
if (ifp->specmib != NULL) {
ifp->specmib = NULL;
ifp->specmiblen = 0;
}
}
out:
/*
* Find sysctl mib for net.ifdescr_maxlen (one time).
* kmib[0] == -1 at first call to mib_fetch_ifmib().
* Then kmib[0] > 0 if we found sysctl mib for net.ifdescr_maxlen.
* Else, kmib[0] == 0 (unexpected error from a kernel).
*/
if (kmib[0] < 0 &&
sysctlnametomib("net.ifdescr_maxlen", kmib, &kmiblen) < 0) {
kmib[0] = 0;
syslog(LOG_WARNING, "sysctlnametomib net.ifdescr_maxlen: %m");
}
/*
* Fetch net.ifdescr_maxlen value every time to catch up with changes.
*/
len = sizeof(alias_maxlen);
if (kmib[0] > 0 && sysctl(kmib, 2, &alias_maxlen, &len, NULL, 0) < 0) {
/* unexpected error from the kernel, use default value */
alias_maxlen = MIBIF_ALIAS_SIZE_MAX;
syslog(LOG_WARNING, "sysctl net.ifdescr_maxlen: %m");
}
/*
* Kernel limit might be decreased after interfaces got
* their descriptions assigned. Try to obtain them anyway.
*/
if (alias_maxlen == 0)
alias_maxlen = MIBIF_ALIAS_SIZE_MAX;
/*
* Allocate maximum memory for a buffer and later reallocate
* to free extra memory.
*/
if ((ifp->alias = malloc(alias_maxlen)) == NULL) {
syslog(LOG_WARNING, "malloc(%d) failed: %m", (int)alias_maxlen);
goto fin;
}
strlcpy(irr.ifr_name, ifp->name, sizeof(irr.ifr_name));
irr.ifr_buffer.buffer = ifp->alias;
irr.ifr_buffer.length = alias_maxlen;
if (ioctl(mib_netsock, SIOCGIFDESCR, &irr) == -1) {
free(ifp->alias);
ifp->alias = NULL;
if (errno != ENOMSG)
syslog(LOG_WARNING, "SIOCGIFDESCR (%s): %m", ifp->name);
} else if (irr.ifr_buffer.buffer == NULL) {
free(ifp->alias);
ifp->alias = NULL;
syslog(LOG_WARNING, "SIOCGIFDESCR (%s): too long (%zu)",
ifp->name, irr.ifr_buffer.length);
} else {
ifp->alias_size = strnlen(ifp->alias, alias_maxlen) + 1;
if (ifp->alias_size > MIBIF_ALIAS_SIZE)
ifp->alias_size = MIBIF_ALIAS_SIZE;
if (ifp->alias_size < alias_maxlen)
ifp->alias = realloc(ifp->alias, ifp->alias_size);
}
fin:
ifp->mibtick = get_ticks();
return (0);
}
/* find first/next address for a given interface */
struct mibifa *
mib_first_ififa(const struct mibif *ifp)
{
struct mibifa *ifa;
TAILQ_FOREACH(ifa, &mibifa_list, link)
if (ifp->index == ifa->ifindex)
return (ifa);
return (NULL);
}
struct mibifa *
mib_next_ififa(struct mibifa *ifa0)
{
struct mibifa *ifa;
ifa = ifa0;
while ((ifa = TAILQ_NEXT(ifa, link)) != NULL)
if (ifa->ifindex == ifa0->ifindex)
return (ifa);
return (NULL);
}
/*
* Allocate a new IFA
*/
static struct mibifa *
alloc_ifa(u_int ifindex, struct in_addr addr)
{
struct mibifa *ifa;
uint32_t ha;
if ((ifa = malloc(sizeof(struct mibifa))) == NULL) {
syslog(LOG_ERR, "ifa: %m");
return (NULL);
}
ifa->inaddr = addr;
ifa->ifindex = ifindex;
ha = ntohl(ifa->inaddr.s_addr);
ifa->index.len = 4;
ifa->index.subs[0] = (ha >> 24) & 0xff;
ifa->index.subs[1] = (ha >> 16) & 0xff;
ifa->index.subs[2] = (ha >> 8) & 0xff;
ifa->index.subs[3] = (ha >> 0) & 0xff;
ifa->flags = 0;
ifa->inbcast.s_addr = 0;
ifa->inmask.s_addr = 0xffffffff;
INSERT_OBJECT_OID(ifa, &mibifa_list);
return (ifa);
}
/*
* Delete an interface address
*/
static void
destroy_ifa(struct mibifa *ifa)
{
TAILQ_REMOVE(&mibifa_list, ifa, link);
free(ifa);
}
/*
* Helper routine to extract the sockaddr structures from a routing
* socket message.
*/
void
mib_extract_addrs(int addrs, u_char *info, struct sockaddr **out)
{
u_int i;
for (i = 0; i < RTAX_MAX; i++) {
if ((addrs & (1 << i)) != 0) {
*out = (struct sockaddr *)(void *)info;
info += roundup((*out)->sa_len, sizeof(long));
} else
*out = NULL;
out++;
}
}
/*
* save the phys address of an interface. Handle receive address entries here.
*/
static void
get_physaddr(struct mibif *ifp, struct sockaddr_dl *sdl, u_char *ptr)
{
u_char *np;
struct mibrcvaddr *rcv;
if (sdl->sdl_alen == 0) {
/* no address */
if (ifp->physaddrlen != 0) {
if ((rcv = mib_find_rcvaddr(ifp->index, ifp->physaddr,
ifp->physaddrlen)) != NULL)
mib_rcvaddr_delete(rcv);
free(ifp->physaddr);
ifp->physaddr = NULL;
ifp->physaddrlen = 0;
}
return;
}
if (ifp->physaddrlen != sdl->sdl_alen) {
/* length changed */
if (ifp->physaddrlen) {
/* delete olf receive address */
if ((rcv = mib_find_rcvaddr(ifp->index, ifp->physaddr,
ifp->physaddrlen)) != NULL)
mib_rcvaddr_delete(rcv);
}
if ((np = realloc(ifp->physaddr, sdl->sdl_alen)) == NULL) {
free(ifp->physaddr);
ifp->physaddr = NULL;
ifp->physaddrlen = 0;
return;
}
ifp->physaddr = np;
ifp->physaddrlen = sdl->sdl_alen;
} else if (memcmp(ifp->physaddr, ptr, ifp->physaddrlen) == 0) {
/* no change */
return;
} else {
/* address changed */
/* delete olf receive address */
if ((rcv = mib_find_rcvaddr(ifp->index, ifp->physaddr,
ifp->physaddrlen)) != NULL)
mib_rcvaddr_delete(rcv);
}
memcpy(ifp->physaddr, ptr, ifp->physaddrlen);
/* make new receive address */
if ((rcv = mib_rcvaddr_create(ifp, ifp->physaddr, ifp->physaddrlen)) != NULL)
rcv->flags |= MIBRCVADDR_HW;
}
/*
* Free an interface
*/
static void
mibif_free(struct mibif *ifp)
{
struct mibif *ifp1;
struct mibindexmap *map;
struct mibifa *ifa, *ifa1;
struct mibrcvaddr *rcv, *rcv1;
struct mibarp *at, *at1;
if (ifp->xnotify != NULL)
(*ifp->xnotify)(ifp, MIBIF_NOTIFY_DESTROY, ifp->xnotify_data);
(void)mib_ifstack_delete(ifp, NULL);
(void)mib_ifstack_delete(NULL, ifp);
TAILQ_REMOVE(&mibif_list, ifp, link);
/* if this was the fastest interface - recompute this */
if (ifp->mib.ifmd_data.ifi_baudrate == mibif_maxspeed) {
mibif_maxspeed = ifp->mib.ifmd_data.ifi_baudrate;
TAILQ_FOREACH(ifp1, &mibif_list, link)
if (ifp1->mib.ifmd_data.ifi_baudrate > mibif_maxspeed)
mibif_maxspeed =
ifp1->mib.ifmd_data.ifi_baudrate;
mibif_reset_hc_timer();
}
if (ifp->alias != NULL) {
free(ifp->alias);
ifp->alias = NULL;
}
free(ifp->private);
ifp->private = NULL;
free(ifp->physaddr);
ifp->physaddr = NULL;
free(ifp->specmib);
ifp->specmib = NULL;
STAILQ_FOREACH(map, &mibindexmap_list, link)
if (map->mibif == ifp) {
map->mibif = NULL;
break;
}
/* purge interface addresses */
ifa = TAILQ_FIRST(&mibifa_list);
while (ifa != NULL) {
ifa1 = TAILQ_NEXT(ifa, link);
if (ifa->ifindex == ifp->index)
destroy_ifa(ifa);
ifa = ifa1;
}
/* purge receive addresses */
rcv = TAILQ_FIRST(&mibrcvaddr_list);
while (rcv != NULL) {
rcv1 = TAILQ_NEXT(rcv, link);
if (rcv->ifindex == ifp->index)
mib_rcvaddr_delete(rcv);
rcv = rcv1;
}
/* purge ARP entries */
at = TAILQ_FIRST(&mibarp_list);
while (at != NULL) {
at1 = TAILQ_NEXT(at, link);
if (at->index.subs[0] == ifp->index)
mib_arp_delete(at);
at = at1;
}
free(ifp);
ifp = NULL;
mib_if_number--;
mib_iftable_last_change = this_tick;
}
/*
* Create a new interface
*/
static struct mibif *
mibif_create(u_int sysindex, const char *name)
{
struct mibif *ifp;
struct mibindexmap *map;
if ((ifp = malloc(sizeof(*ifp))) == NULL) {
syslog(LOG_WARNING, "%s: %m", __func__);
return (NULL);
}
memset(ifp, 0, sizeof(*ifp));
if ((ifp->private = malloc(sizeof(struct mibif_private))) == NULL) {
syslog(LOG_WARNING, "%s: %m", __func__);
free(ifp);
return (NULL);
}
memset(ifp->private, 0, sizeof(struct mibif_private));
ifp->sysindex = sysindex;
strlcpy(ifp->name, name, sizeof(ifp->name));
strlcpy(ifp->descr, name, sizeof(ifp->descr));
ifp->spec_oid = oid_zeroDotZero;
map = NULL;
if (!mib_if_is_dyn(ifp->name)) {
/* non-dynamic. look whether we know the interface */
STAILQ_FOREACH(map, &mibindexmap_list, link)
if (strcmp(map->name, ifp->name) == 0) {
ifp->index = map->ifindex;
map->mibif = ifp;
break;
}
/* assume it has a connector if it is not dynamic */
ifp->has_connector = 1;
ifp->trap_enable = 1;
}
if (map == NULL) {
/* new interface - get new index */
if (next_if_index > 0x7fffffff)
errx(1, "ifindex wrap");
if ((map = malloc(sizeof(*map))) == NULL) {
syslog(LOG_ERR, "ifmap: %m");
free(ifp);
return (NULL);
}
map->ifindex = next_if_index++;
map->sysindex = ifp->sysindex;
strcpy(map->name, ifp->name);
map->mibif = ifp;
STAILQ_INSERT_TAIL(&mibindexmap_list, map, link);
} else {
/* re-instantiate. Introduce a counter discontinuity */
ifp->counter_disc = get_ticks();
}
ifp->index = map->ifindex;
ifp->mib.ifmd_data.ifi_link_state = LINK_STATE_UNKNOWN;
INSERT_OBJECT_INT(ifp, &mibif_list);
mib_if_number++;
mib_iftable_last_change = this_tick;
/* instantiate default ifStack entries */
(void)mib_ifstack_create(ifp, NULL);
(void)mib_ifstack_create(NULL, ifp);
return (ifp);
}
/*
* Inform all interested parties about a new interface
*/
static void
notify_newif(struct mibif *ifp)
{
struct newifreg *reg;
TAILQ_FOREACH(reg, &newifreg_list, link)
if ((*reg->func)(ifp))
return;
}
/*
* This is called for new interfaces after we have fetched the interface
* MIB. If this is a broadcast interface try to guess the broadcast address
* depending on the interface type.
*/
static void
check_llbcast(struct mibif *ifp)
{
static u_char ether_bcast[6] = { 0xff, 0xff, 0xff, 0xff, 0xff, 0xff };
struct mibrcvaddr *rcv;
if (!(ifp->mib.ifmd_flags & IFF_BROADCAST))
return;
switch (ifp->mib.ifmd_data.ifi_type) {
case IFT_ETHER:
case IFT_FDDI:
case IFT_ISO88025:
case IFT_L2VLAN:
if (mib_find_rcvaddr(ifp->index, ether_bcast, 6) == NULL &&
(rcv = mib_rcvaddr_create(ifp, ether_bcast, 6)) != NULL)
rcv->flags |= MIBRCVADDR_BCAST;
break;
}
}
/*
* Retrieve the current interface list from the system.
*/
void
mib_refresh_iflist(void)
{
struct mibif *ifp, *ifp1;
size_t len;
u_short idx;
int name[6];
int count;
struct ifmibdata mib;
TAILQ_FOREACH(ifp, &mibif_list, link)
ifp->flags &= ~MIBIF_FOUND;
len = sizeof(count);
if (sysctlbyname("net.link.generic.system.ifcount", &count, &len,
NULL, 0) == -1) {
syslog(LOG_ERR, "ifcount: %m");
return;
}
name[0] = CTL_NET;
name[1] = PF_LINK;
name[2] = NETLINK_GENERIC;
name[3] = IFMIB_IFDATA;
name[5] = IFDATA_GENERAL;
for (idx = 1; idx <= count; idx++) {
name[4] = idx;
len = sizeof(mib);
if (sysctl(name, nitems(name), &mib, &len, NULL, 0) == -1) {
if (errno == ENOENT)
continue;
syslog(LOG_ERR, "ifmib(%u): %m", idx);
return;
}
if ((ifp = mib_find_if_sys(idx)) != NULL) {
ifp->flags |= MIBIF_FOUND;
continue;
}
/* Unknown interface - create */
if ((ifp = mibif_create(idx, mib.ifmd_name)) != NULL) {
ifp->flags |= MIBIF_FOUND;
(void)mib_fetch_ifmib(ifp);
check_llbcast(ifp);
notify_newif(ifp);
}
}
/*
* Purge interfaces that disappeared
*/
ifp = TAILQ_FIRST(&mibif_list);
while (ifp != NULL) {
ifp1 = TAILQ_NEXT(ifp, link);
if (!(ifp->flags & MIBIF_FOUND))
mibif_free(ifp);
ifp = ifp1;
}
}
/*
* Find an interface address
*/
struct mibifa *
mib_find_ifa(struct in_addr addr)
{
struct mibifa *ifa;
TAILQ_FOREACH(ifa, &mibifa_list, link)
if (ifa->inaddr.s_addr == addr.s_addr)
return (ifa);
return (NULL);
}
/*
* Process a new ARP entry
*/
static void
process_arp(const struct rt_msghdr *rtm, const struct sockaddr_dl *sdl,
const struct sockaddr_in *sa)
{
struct mibif *ifp;
struct mibarp *at;
/* IP arp table entry */
if (sdl->sdl_alen == 0)
return;
if ((ifp = mib_find_if_sys(sdl->sdl_index)) == NULL)
return;
/* have a valid entry */
if ((at = mib_find_arp(ifp, sa->sin_addr)) == NULL &&
(at = mib_arp_create(ifp, sa->sin_addr,
sdl->sdl_data + sdl->sdl_nlen, sdl->sdl_alen)) == NULL)
return;
if (rtm->rtm_rmx.rmx_expire == 0)
at->flags |= MIBARP_PERM;
else
at->flags &= ~MIBARP_PERM;
at->flags |= MIBARP_FOUND;
}
/*
* Handle a routing socket message.
*/
static void
handle_rtmsg(struct rt_msghdr *rtm)
{
struct sockaddr *addrs[RTAX_MAX];
struct if_msghdr *ifm;
struct ifa_msghdr ifam, *ifamp;
struct ifma_msghdr *ifmam;
#ifdef RTM_IFANNOUNCE
struct if_announcemsghdr *ifan;
#endif
struct mibif *ifp;
struct sockaddr_dl *sdl;
struct sockaddr_in *sa;
struct mibifa *ifa;
struct mibrcvaddr *rcv;
u_char *ptr;
if (rtm->rtm_version != RTM_VERSION) {
syslog(LOG_ERR, "Bogus RTM version %u", rtm->rtm_version);
return;
}
switch (rtm->rtm_type) {
case RTM_NEWADDR:
ifamp = (struct ifa_msghdr *)rtm;
memcpy(&ifam, ifamp, sizeof(ifam));
mib_extract_addrs(ifam.ifam_addrs, (u_char *)(ifamp + 1), addrs);
if (addrs[RTAX_IFA] == NULL || addrs[RTAX_NETMASK] == NULL)
break;
sa = (struct sockaddr_in *)(void *)addrs[RTAX_IFA];
if ((ifa = mib_find_ifa(sa->sin_addr)) == NULL) {
/* unknown address */
if ((ifp = mib_find_if_sys(ifam.ifam_index)) == NULL) {
syslog(LOG_WARNING, "RTM_NEWADDR for unknown "
"interface %u", ifam.ifam_index);
break;
}
if ((ifa = alloc_ifa(ifp->index, sa->sin_addr)) == NULL)
break;
}
sa = (struct sockaddr_in *)(void *)addrs[RTAX_NETMASK];
ifa->inmask = sa->sin_addr;
if (addrs[RTAX_BRD] != NULL) {
sa = (struct sockaddr_in *)(void *)addrs[RTAX_BRD];
ifa->inbcast = sa->sin_addr;
}
ifa->flags |= MIBIFA_FOUND;
break;
case RTM_DELADDR:
ifamp = (struct ifa_msghdr *)rtm;
memcpy(&ifam, ifamp, sizeof(ifam));
mib_extract_addrs(ifam.ifam_addrs, (u_char *)(ifamp + 1), addrs);
if (addrs[RTAX_IFA] == NULL)
break;
sa = (struct sockaddr_in *)(void *)addrs[RTAX_IFA];
if ((ifa = mib_find_ifa(sa->sin_addr)) != NULL) {
ifa->flags |= MIBIFA_FOUND;
if (!(ifa->flags & MIBIFA_DESTROYED))
destroy_ifa(ifa);
}
break;
case RTM_NEWMADDR:
ifmam = (struct ifma_msghdr *)rtm;
mib_extract_addrs(ifmam->ifmam_addrs, (u_char *)(ifmam + 1), addrs);
if (addrs[RTAX_IFA] == NULL ||
addrs[RTAX_IFA]->sa_family != AF_LINK)
break;
sdl = (struct sockaddr_dl *)(void *)addrs[RTAX_IFA];
if ((rcv = mib_find_rcvaddr(sdl->sdl_index,
sdl->sdl_data + sdl->sdl_nlen, sdl->sdl_alen)) == NULL) {
/* unknown address */
if ((ifp = mib_find_if_sys(sdl->sdl_index)) == NULL) {
syslog(LOG_WARNING, "RTM_NEWMADDR for unknown "
"interface %u", sdl->sdl_index);
break;
}
if ((rcv = mib_rcvaddr_create(ifp,
sdl->sdl_data + sdl->sdl_nlen, sdl->sdl_alen)) == NULL)
break;
rcv->flags |= MIBRCVADDR_VOLATILE;
}
rcv->flags |= MIBRCVADDR_FOUND;
break;
case RTM_DELMADDR:
ifmam = (struct ifma_msghdr *)rtm;
mib_extract_addrs(ifmam->ifmam_addrs, (u_char *)(ifmam + 1), addrs);
if (addrs[RTAX_IFA] == NULL ||
addrs[RTAX_IFA]->sa_family != AF_LINK)
break;
sdl = (struct sockaddr_dl *)(void *)addrs[RTAX_IFA];
if ((rcv = mib_find_rcvaddr(sdl->sdl_index,
sdl->sdl_data + sdl->sdl_nlen, sdl->sdl_alen)) != NULL)
mib_rcvaddr_delete(rcv);
break;
case RTM_IFINFO:
ifm = (struct if_msghdr *)(void *)rtm;
mib_extract_addrs(ifm->ifm_addrs, (u_char *)(ifm + 1), addrs);
if ((ifp = mib_find_if_sys(ifm->ifm_index)) == NULL)
break;
if (addrs[RTAX_IFP] != NULL &&
addrs[RTAX_IFP]->sa_family == AF_LINK) {
sdl = (struct sockaddr_dl *)(void *)addrs[RTAX_IFP];
ptr = sdl->sdl_data + sdl->sdl_nlen;
get_physaddr(ifp, sdl, ptr);
}
(void)mib_fetch_ifmib(ifp);
break;
#ifdef RTM_IFANNOUNCE
case RTM_IFANNOUNCE:
ifan = (struct if_announcemsghdr *)rtm;
ifp = mib_find_if_sys(ifan->ifan_index);
switch (ifan->ifan_what) {
case IFAN_ARRIVAL:
if (ifp == NULL && (ifp = mibif_create(ifan->ifan_index,
ifan->ifan_name)) != NULL) {
(void)mib_fetch_ifmib(ifp);
check_llbcast(ifp);
notify_newif(ifp);
}
break;
case IFAN_DEPARTURE:
if (ifp != NULL)
mibif_free(ifp);
break;
}
break;
#endif
case RTM_GET:
case RTM_ADD:
mib_extract_addrs(rtm->rtm_addrs, (u_char *)(rtm + 1), addrs);
if (rtm->rtm_flags & RTF_LLINFO) {
if (addrs[RTAX_DST] == NULL ||
addrs[RTAX_GATEWAY] == NULL ||
addrs[RTAX_DST]->sa_family != AF_INET ||
addrs[RTAX_GATEWAY]->sa_family != AF_LINK)
break;
process_arp(rtm,
(struct sockaddr_dl *)(void *)addrs[RTAX_GATEWAY],
(struct sockaddr_in *)(void *)addrs[RTAX_DST]);
} else {
if (rtm->rtm_errno == 0 && (rtm->rtm_flags & RTF_UP))
mib_sroute_process(rtm, addrs[RTAX_GATEWAY],
addrs[RTAX_DST], addrs[RTAX_NETMASK]);
}
break;
case RTM_DELETE:
mib_extract_addrs(rtm->rtm_addrs, (u_char *)(rtm + 1), addrs);
if (rtm->rtm_errno == 0 && (rtm->rtm_flags & RTF_UP))
mib_sroute_process(rtm, addrs[RTAX_GATEWAY],
addrs[RTAX_DST], addrs[RTAX_NETMASK]);
break;
}
}
/*
* send a routing message
*/
void
mib_send_rtmsg(struct rt_msghdr *rtm, struct sockaddr *gw,
struct sockaddr *dst, struct sockaddr *mask)
{
size_t len;
struct rt_msghdr *msg;
char *cp;
ssize_t sent;
len = sizeof(*rtm) + SA_SIZE(gw) + SA_SIZE(dst) + SA_SIZE(mask);
if ((msg = malloc(len)) == NULL) {
syslog(LOG_ERR, "%s: %m", __func__);
return;
}
cp = (char *)(msg + 1);
memset(msg, 0, sizeof(*msg));
msg->rtm_flags = 0;
msg->rtm_version = RTM_VERSION;
msg->rtm_addrs = RTA_DST | RTA_GATEWAY;
memcpy(cp, dst, SA_SIZE(dst));
cp += SA_SIZE(dst);
memcpy(cp, gw, SA_SIZE(gw));
cp += SA_SIZE(gw);
if (mask != NULL) {
memcpy(cp, mask, SA_SIZE(mask));
cp += SA_SIZE(mask);
msg->rtm_addrs |= RTA_NETMASK;
}
msg->rtm_msglen = cp - (char *)msg;
msg->rtm_type = RTM_GET;
if ((sent = write(route, msg, msg->rtm_msglen)) == -1) {
syslog(LOG_ERR, "%s: write: %m", __func__);
free(msg);
return;
}
if (sent != msg->rtm_msglen) {
syslog(LOG_ERR, "%s: short write", __func__);
free(msg);
return;
}
free(msg);
}
/*
* Fetch the routing table via sysctl
*/
u_char *
mib_fetch_rtab(int af, int info, int arg, size_t *lenp)
{
int name[6];
u_char *buf, *newbuf;
name[0] = CTL_NET;
name[1] = PF_ROUTE;
name[2] = 0;
name[3] = af;
name[4] = info;
name[5] = arg;
*lenp = 0;
/* initial estimate */
if (sysctl(name, nitems(name), NULL, lenp, NULL, 0) == -1) {
syslog(LOG_ERR, "sysctl estimate (%d,%d,%d,%d,%d,%d): %m",
name[0], name[1], name[2], name[3], name[4], name[5]);
return (NULL);
}
if (*lenp == 0)
return (NULL);
buf = NULL;
for (;;) {
if ((newbuf = realloc(buf, *lenp)) == NULL) {
syslog(LOG_ERR, "sysctl buffer: %m");
free(buf);
return (NULL);
}
buf = newbuf;
if (sysctl(name, nitems(name), buf, lenp, NULL, 0) == 0)
break;
if (errno != ENOMEM) {
syslog(LOG_ERR, "sysctl get: %m");
free(buf);
return (NULL);
}
*lenp += *lenp / 8 + 1;
}
return (buf);
}
/*
* Update the following info: interface, interface addresses, interface
* receive addresses, arp-table.
* This does not change the interface list itself.
*/
static void
update_ifa_info(void)
{
u_char *buf, *next;
struct rt_msghdr *rtm;
struct mibifa *ifa, *ifa1;
struct mibrcvaddr *rcv, *rcv1;
size_t needed;
static const int infos[][3] = {
{ 0, NET_RT_IFLIST, 0 },
#ifdef NET_RT_IFMALIST
{ AF_LINK, NET_RT_IFMALIST, 0 },
#endif
};
u_int i;
TAILQ_FOREACH(ifa, &mibifa_list, link)
ifa->flags &= ~MIBIFA_FOUND;
TAILQ_FOREACH(rcv, &mibrcvaddr_list, link)
rcv->flags &= ~MIBRCVADDR_FOUND;
for (i = 0; i < sizeof(infos) / sizeof(infos[0]); i++) {
if ((buf = mib_fetch_rtab(infos[i][0], infos[i][1], infos[i][2],
&needed)) == NULL)
continue;
next = buf;
while (next < buf + needed) {
rtm = (struct rt_msghdr *)(void *)next;
next += rtm->rtm_msglen;
handle_rtmsg(rtm);
}
free(buf);
}
/*
* Purge the address list of unused entries. These may happen for
* interface aliases that are on the same subnet. We don't receive
* routing socket messages for them.
*/
ifa = TAILQ_FIRST(&mibifa_list);
while (ifa != NULL) {
ifa1 = TAILQ_NEXT(ifa, link);
if (!(ifa->flags & MIBIFA_FOUND))
destroy_ifa(ifa);
ifa = ifa1;
}
rcv = TAILQ_FIRST(&mibrcvaddr_list);
while (rcv != NULL) {
rcv1 = TAILQ_NEXT(rcv, link);
if (!(rcv->flags & (MIBRCVADDR_FOUND | MIBRCVADDR_BCAST |
MIBRCVADDR_HW)))
mib_rcvaddr_delete(rcv);
rcv = rcv1;
}
}
/*
* Update arp table
*/
void
mib_arp_update(void)
{
struct mibarp *at, *at1;
size_t needed;
u_char *buf, *next;
struct rt_msghdr *rtm;
if (in_update_arp)
return; /* Aaargh */
in_update_arp = 1;
TAILQ_FOREACH(at, &mibarp_list, link)
at->flags &= ~MIBARP_FOUND;
if ((buf = mib_fetch_rtab(AF_INET, NET_RT_FLAGS, 0, &needed)) == NULL) {
in_update_arp = 0;
return;
}
next = buf;
while (next < buf + needed) {
rtm = (struct rt_msghdr *)(void *)next;
next += rtm->rtm_msglen;
handle_rtmsg(rtm);
}
free(buf);
at = TAILQ_FIRST(&mibarp_list);
while (at != NULL) {
at1 = TAILQ_NEXT(at, link);
if (!(at->flags & MIBARP_FOUND))
mib_arp_delete(at);
at = at1;
}
mibarpticks = get_ticks();
in_update_arp = 0;
}
/*
* Input on the routing socket.
*/
static void
route_input(int fd, void *udata __unused)
{
u_char buf[1024 * 16];
ssize_t n;
struct rt_msghdr *rtm;
if ((n = read(fd, buf, sizeof(buf))) == -1)
err(1, "read(rt_socket)");
if (n == 0)
errx(1, "EOF on rt_socket");
rtm = (struct rt_msghdr *)(void *)buf;
if ((size_t)n != rtm->rtm_msglen)
errx(1, "n=%zu, rtm_msglen=%u", (size_t)n, rtm->rtm_msglen);
handle_rtmsg(rtm);
}
/*
* execute and SIOCAIFADDR
*/
static int
siocaifaddr(char *ifname, struct in_addr addr, struct in_addr mask,
struct in_addr bcast)
{
struct ifaliasreq addreq;
struct sockaddr_in *sa;
memset(&addreq, 0, sizeof(addreq));
strlcpy(addreq.ifra_name, ifname, sizeof(addreq.ifra_name));
sa = (struct sockaddr_in *)(void *)&addreq.ifra_addr;
sa->sin_family = AF_INET;
sa->sin_len = sizeof(*sa);
sa->sin_addr = addr;
sa = (struct sockaddr_in *)(void *)&addreq.ifra_mask;
sa->sin_family = AF_INET;
sa->sin_len = sizeof(*sa);
sa->sin_addr = mask;
sa = (struct sockaddr_in *)(void *)&addreq.ifra_broadaddr;
sa->sin_family = AF_INET;
sa->sin_len = sizeof(*sa);
sa->sin_addr = bcast;
return (ioctl(mib_netsock, SIOCAIFADDR, &addreq));
}
/*
* Exececute a SIOCDIFADDR
*/
static int
siocdifaddr(const char *ifname, struct in_addr addr)
{
struct ifreq delreq;
struct sockaddr_in *sa;
memset(&delreq, 0, sizeof(delreq));
strlcpy(delreq.ifr_name, ifname, sizeof(delreq.ifr_name));
sa = (struct sockaddr_in *)(void *)&delreq.ifr_addr;
sa->sin_family = AF_INET;
sa->sin_len = sizeof(*sa);
sa->sin_addr = addr;
return (ioctl(mib_netsock, SIOCDIFADDR, &delreq));
}
/*
* Verify an interface address without fetching the entire list
*/
static int
verify_ifa(const char *name, struct mibifa *ifa)
{
struct ifreq req;
struct sockaddr_in *sa;
memset(&req, 0, sizeof(req));
strlcpy(req.ifr_name, name, sizeof(req.ifr_name));
sa = (struct sockaddr_in *)(void *)&req.ifr_addr;
sa->sin_family = AF_INET;
sa->sin_len = sizeof(*sa);
sa->sin_addr = ifa->inaddr;
if (ioctl(mib_netsock, SIOCGIFADDR, &req) == -1)
return (-1);
if (ifa->inaddr.s_addr != sa->sin_addr.s_addr) {
syslog(LOG_ERR, "%s: address mismatch", __func__);
return (-1);
}
if (ioctl(mib_netsock, SIOCGIFNETMASK, &req) == -1)
return (-1);
if (ifa->inmask.s_addr != sa->sin_addr.s_addr) {
syslog(LOG_ERR, "%s: netmask mismatch", __func__);
return (-1);
}
return (0);
}
/*
* Restore a deleted interface address. Don't wait for the routing socket
* to update us.
*/
void
mib_undestroy_ifa(struct mibifa *ifa)
{
struct mibif *ifp;
if ((ifp = mib_find_if(ifa->ifindex)) == NULL)
/* keep it destroyed */
return;
if (siocaifaddr(ifp->name, ifa->inaddr, ifa->inmask, ifa->inbcast))
/* keep it destroyed */
return;
ifa->flags &= ~MIBIFA_DESTROYED;
}
/*
* Destroy an interface address
*/
int
mib_destroy_ifa(struct mibifa *ifa)
{
struct mibif *ifp;
if ((ifp = mib_find_if(ifa->ifindex)) == NULL) {
/* ups. */
mib_iflist_bad = 1;
return (-1);
}
if (siocdifaddr(ifp->name, ifa->inaddr)) {
/* ups. */
syslog(LOG_ERR, "SIOCDIFADDR: %m");
mib_iflist_bad = 1;
return (-1);
}
ifa->flags |= MIBIFA_DESTROYED;
return (0);
}
/*
* Rollback the modification of an address. Don't bother to wait for
* the routing socket.
*/
void
mib_unmodify_ifa(struct mibifa *ifa)
{
struct mibif *ifp;
if ((ifp = mib_find_if(ifa->ifindex)) == NULL) {
/* ups. */
mib_iflist_bad = 1;
return;
}
if (siocaifaddr(ifp->name, ifa->inaddr, ifa->inmask, ifa->inbcast)) {
/* ups. */
mib_iflist_bad = 1;
return;
}
}
/*
* Modify an IFA.
*/
int
mib_modify_ifa(struct mibifa *ifa)
{
struct mibif *ifp;
if ((ifp = mib_find_if(ifa->ifindex)) == NULL) {
/* ups. */
mib_iflist_bad = 1;
return (-1);
}
if (siocaifaddr(ifp->name, ifa->inaddr, ifa->inmask, ifa->inbcast)) {
/* ups. */
mib_iflist_bad = 1;
return (-1);
}
if (verify_ifa(ifp->name, ifa)) {
/* ups. */
mib_iflist_bad = 1;
return (-1);
}
return (0);
}
/*
* Destroy a freshly created interface address. Don't bother to wait for
* the routing socket.
*/
void
mib_uncreate_ifa(struct mibifa *ifa)
{
struct mibif *ifp;
if ((ifp = mib_find_if(ifa->ifindex)) == NULL) {
/* ups. */
mib_iflist_bad = 1;
return;
}
if (siocdifaddr(ifp->name, ifa->inaddr)) {
/* ups. */
mib_iflist_bad = 1;
return;
}
destroy_ifa(ifa);
}
/*
* Create a new ifa and verify it
*/
struct mibifa *
mib_create_ifa(u_int ifindex, struct in_addr addr, struct in_addr mask,
struct in_addr bcast)
{
struct mibif *ifp;
struct mibifa *ifa;
if ((ifp = mib_find_if(ifindex)) == NULL)
return (NULL);
if ((ifa = alloc_ifa(ifindex, addr)) == NULL)
return (NULL);
ifa->inmask = mask;
ifa->inbcast = bcast;
if (siocaifaddr(ifp->name, ifa->inaddr, ifa->inmask, ifa->inbcast)) {
syslog(LOG_ERR, "%s: %m", __func__);
destroy_ifa(ifa);
return (NULL);
}
if (verify_ifa(ifp->name, ifa)) {
destroy_ifa(ifa);
return (NULL);
}
return (ifa);
}
/*
* Get all cloning interfaces and make them dynamic.
* Hah! Whe should probably do this on a periodic basis (XXX).
*/
static void
get_cloners(void)
{
struct if_clonereq req;
char *buf, *cp;
int i;
memset(&req, 0, sizeof(req));
if (ioctl(mib_netsock, SIOCIFGCLONERS, &req) == -1) {
syslog(LOG_ERR, "get cloners: %m");
return;
}
if ((buf = malloc(req.ifcr_total * IFNAMSIZ)) == NULL) {
syslog(LOG_ERR, "%m");
return;
}
req.ifcr_count = req.ifcr_total;
req.ifcr_buffer = buf;
if (ioctl(mib_netsock, SIOCIFGCLONERS, &req) == -1) {
syslog(LOG_ERR, "get cloners: %m");
free(buf);
return;
}
for (cp = buf, i = 0; i < req.ifcr_total; i++, cp += IFNAMSIZ)
mib_if_set_dyn(cp);
free(buf);
}
/*
* Idle function
*/
static void
mibII_idle(void *arg __unused)
{
struct mibifa *ifa;
if (mib_iflist_bad) {
TAILQ_FOREACH(ifa, &mibifa_list, link)
ifa->flags &= ~MIBIFA_DESTROYED;
/* assume, that all cloning interfaces are dynamic */
get_cloners();
mib_refresh_iflist();
update_ifa_info();
mib_arp_update();
mib_iflist_bad = 0;
}
mib_arp_update();
}
/*
* Start the module
*/
static void
mibII_start(void)
{
if ((route_fd = fd_select(route, route_input, NULL, module)) == NULL) {
syslog(LOG_ERR, "fd_select(route): %m");
return;
}
mib_refresh_iflist();
update_ifa_info();
mib_arp_update();
(void)mib_fetch_route();
mib_iftable_last_change = 0;
mib_ifstack_last_change = 0;
ifmib_reg = or_register(&oid_ifMIB,
"The MIB module to describe generic objects for network interface"
" sub-layers.", module);
ipmib_reg = or_register(&oid_ipMIB,
"The MIB module for managing IP and ICMP implementations, but "
"excluding their management of IP routes.", module);
tcpmib_reg = or_register(&oid_tcpMIB,
"The MIB module for managing TCP implementations.", module);
udpmib_reg = or_register(&oid_udpMIB,
"The MIB module for managing UDP implementations.", module);
ipForward_reg = or_register(&oid_ipForward,
"The MIB module for the display of CIDR multipath IP Routes.",
module);
mibII_poll_timer = NULL;
mibII_poll_ticks = MIBII_POLL_TICKS;
mibif_restart_mibII_poll_timer();
}
/*
* Initialize the module
*/
static int
mibII_init(struct lmodule *mod, int argc __unused, char *argv[] __unused)
{
size_t len;
module = mod;
len = sizeof(clockinfo);
if (sysctlbyname("kern.clockrate", &clockinfo, &len, NULL, 0) == -1) {
syslog(LOG_ERR, "kern.clockrate: %m");
return (-1);
}
if (len != sizeof(clockinfo)) {
syslog(LOG_ERR, "kern.clockrate: wrong size");
return (-1);
}
if ((route = socket(PF_ROUTE, SOCK_RAW, AF_UNSPEC)) == -1) {
syslog(LOG_ERR, "PF_ROUTE: %m");
return (-1);
}
if ((mib_netsock = socket(PF_INET, SOCK_DGRAM, 0)) == -1) {
syslog(LOG_ERR, "PF_INET: %m");
(void)close(route);
return (-1);
}
(void)shutdown(mib_netsock, SHUT_RDWR);
/* assume, that all cloning interfaces are dynamic */
get_cloners();
return (0);
}
static int
mibII_fini(void)
{
if (mibII_poll_timer != NULL ) {
timer_stop(mibII_poll_timer);
mibII_poll_timer = NULL;
}
if (route_fd != NULL)
fd_deselect(route_fd);
if (route != -1)
(void)close(route);
if (mib_netsock != -1)
(void)close(mib_netsock);
/* XXX free memory */
or_unregister(ipForward_reg);
or_unregister(udpmib_reg);
or_unregister(tcpmib_reg);
or_unregister(ipmib_reg);
or_unregister(ifmib_reg);
return (0);
}
static void
mibII_loading(const struct lmodule *mod, int loaded)
{
struct mibif *ifp;
if (loaded == 1)
return;
TAILQ_FOREACH(ifp, &mibif_list, link)
if (ifp->xnotify_mod == mod) {
ifp->xnotify_mod = NULL;
ifp->xnotify_data = NULL;
ifp->xnotify = NULL;
}
mib_unregister_newif(mod);
}
extern const struct snmp_module config;
const struct snmp_module config = {
"This module implements the interface and ip groups.",
mibII_init,
mibII_fini,
NULL, /* idle */
NULL, /* dump */
NULL, /* config */
mibII_start,
NULL,
mibII_ctree,
mibII_CTREE_SIZE,
mibII_loading
};
/*
* Should have a list of these attached to each interface.
*/
void *
mibif_notify(struct mibif *ifp, const struct lmodule *mod,
mibif_notify_f func, void *data)
{
ifp->xnotify = func;
ifp->xnotify_data = data;
ifp->xnotify_mod = mod;
return (ifp);
}
void
mibif_unnotify(void *arg)
{
struct mibif *ifp = arg;
ifp->xnotify = NULL;
ifp->xnotify_data = NULL;
ifp->xnotify_mod = NULL;
}