asterisk/main/acl.c

1114 lines
29 KiB
C

/*
* Asterisk -- An open source telephony toolkit.
*
* Copyright (C) 1999 - 2012, Digium, Inc.
*
* Mark Spencer <markster@digium.com>
*
* See http://www.asterisk.org for more information about
* the Asterisk project. Please do not directly contact
* any of the maintainers of this project for assistance;
* the project provides a web site, mailing lists and IRC
* channels for your use.
*
* This program is free software, distributed under the terms of
* the GNU General Public License Version 2. See the LICENSE file
* at the top of the source tree.
*/
/*! \file
*
* \brief Various sorts of access control
*
* \author Mark Spencer <markster@digium.com>
*/
/*** MODULEINFO
<support_level>core</support_level>
***/
#include "asterisk.h"
#include "asterisk/network.h"
#if defined(__OpenBSD__) || defined(__NetBSD__) || defined(__FreeBSD__) || defined(__Darwin__)
#include <fcntl.h>
#include <net/route.h>
#endif
#if defined(SOLARIS)
#include <sys/sockio.h>
#include <net/if.h>
#elif defined(HAVE_GETIFADDRS)
#include <ifaddrs.h>
#endif
#include "asterisk/acl.h"
#include "asterisk/channel.h"
#include "asterisk/utils.h"
#include "asterisk/lock.h"
#include "asterisk/srv.h"
#include "asterisk/cli.h"
#if (!defined(SOLARIS) && !defined(HAVE_GETIFADDRS))
static int get_local_address(struct ast_sockaddr *ourip)
{
return -1;
}
#else
static void score_address(const struct sockaddr_in *sin, struct in_addr *best_addr, int *best_score)
{
const char *address;
int score;
address = ast_inet_ntoa(sin->sin_addr);
/* RFC 1700 alias for the local network */
if (address[0] == '0') {
score = -25;
/* RFC 1700 localnet */
} else if (strncmp(address, "127", 3) == 0) {
score = -20;
/* RFC 1918 non-public address space */
} else if (strncmp(address, "10.", 3) == 0) {
score = -5;
/* RFC 1918 non-public address space */
} else if (strncmp(address, "172", 3) == 0) {
/* 172.16.0.0 - 172.19.255.255, but not 172.160.0.0 - 172.169.255.255 */
if (address[4] == '1' && address[5] >= '6' && address[6] == '.') {
score = -5;
/* 172.20.0.0 - 172.29.255.255, but not 172.200.0.0 - 172.255.255.255 nor 172.2.0.0 - 172.2.255.255 */
} else if (address[4] == '2' && address[6] == '.') {
score = -5;
/* 172.30.0.0 - 172.31.255.255, but not 172.3.0.0 - 172.3.255.255 */
} else if (address[4] == '3' && (address[5] == '0' || address[5] == '1')) {
score = -5;
/* All other 172 addresses are public */
} else {
score = 0;
}
/* RFC 2544 Benchmark test range (198.18.0.0 - 198.19.255.255, but not 198.180.0.0 - 198.199.255.255) */
} else if (strncmp(address, "198.1", 5) == 0 && address[5] >= '8' && address[6] == '.') {
score = -10;
/* RFC 1918 non-public address space */
} else if (strncmp(address, "192.168", 7) == 0) {
score = -5;
/* RFC 3330 Zeroconf network */
} else if (strncmp(address, "169.254", 7) == 0) {
/*!\note Better score than a test network, but not quite as good as RFC 1918
* address space. The reason is that some Linux distributions automatically
* configure a Zeroconf address before trying DHCP, so we want to prefer a
* DHCP lease to a Zeroconf address.
*/
score = -10;
/* RFC 3330 Test network */
} else if (strncmp(address, "192.0.2.", 8) == 0) {
score = -15;
/* Every other address should be publically routable */
} else {
score = 0;
}
if (score > *best_score) {
*best_score = score;
memcpy(best_addr, &sin->sin_addr, sizeof(*best_addr));
}
}
static int get_local_address(struct ast_sockaddr *ourip)
{
int s, res = -1;
#ifdef SOLARIS
struct lifreq *ifr = NULL;
struct lifnum ifn;
struct lifconf ifc;
struct sockaddr_in *sa;
char *buf = NULL;
int bufsz, x;
#endif /* SOLARIS */
#if defined(__OpenBSD__) || defined(__NetBSD__) || defined(__FreeBSD__) || defined(__linux__) || defined(__Darwin__) || defined(__GLIBC__)
struct ifaddrs *ifap, *ifaphead;
int rtnerr;
const struct sockaddr_in *sin;
#endif /* BSD_OR_LINUX */
struct in_addr best_addr;
int best_score = -100;
memset(&best_addr, 0, sizeof(best_addr));
#if defined(__OpenBSD__) || defined(__NetBSD__) || defined(__FreeBSD__) || defined(__linux__) || defined(__Darwin__) || defined(__GLIBC__)
rtnerr = getifaddrs(&ifaphead);
if (rtnerr) {
perror(NULL);
return -1;
}
#endif /* BSD_OR_LINUX */
s = socket(AF_INET, SOCK_STREAM, 0);
if (s > 0) {
#if defined(__OpenBSD__) || defined(__NetBSD__) || defined(__FreeBSD__) || defined(__linux__) || defined(__Darwin__) || defined(__GLIBC__)
for (ifap = ifaphead; ifap; ifap = ifap->ifa_next) {
if (ifap->ifa_addr && ifap->ifa_addr->sa_family == AF_INET) {
sin = (const struct sockaddr_in *) ifap->ifa_addr;
score_address(sin, &best_addr, &best_score);
res = 0;
if (best_score == 0) {
break;
}
}
}
#endif /* BSD_OR_LINUX */
/* There is no reason whatsoever that this shouldn't work on Linux or BSD also. */
#ifdef SOLARIS
/* Get a count of interfaces on the machine */
ifn.lifn_family = AF_INET;
ifn.lifn_flags = 0;
ifn.lifn_count = 0;
if (ioctl(s, SIOCGLIFNUM, &ifn) < 0) {
close(s);
return -1;
}
bufsz = ifn.lifn_count * sizeof(struct lifreq);
if (!(buf = ast_malloc(bufsz))) {
close(s);
return -1;
}
memset(buf, 0, bufsz);
/* Get a list of interfaces on the machine */
ifc.lifc_len = bufsz;
ifc.lifc_buf = buf;
ifc.lifc_family = AF_INET;
ifc.lifc_flags = 0;
if (ioctl(s, SIOCGLIFCONF, &ifc) < 0) {
close(s);
ast_free(buf);
return -1;
}
for (ifr = ifc.lifc_req, x = 0; x < ifn.lifn_count; ifr++, x++) {
sa = (struct sockaddr_in *)&(ifr->lifr_addr);
score_address(sa, &best_addr, &best_score);
res = 0;
if (best_score == 0) {
break;
}
}
ast_free(buf);
#endif /* SOLARIS */
close(s);
}
#if defined(__OpenBSD__) || defined(__NetBSD__) || defined(__FreeBSD__) || defined(__linux__) || defined(__Darwin__)
freeifaddrs(ifaphead);
#endif /* BSD_OR_LINUX */
if (res == 0 && ourip) {
ast_sockaddr_setnull(ourip);
ourip->ss.ss_family = AF_INET;
((struct sockaddr_in *)&ourip->ss)->sin_addr = best_addr;
}
return res;
}
#endif /* HAVE_GETIFADDRS */
/* Free HA structure */
void ast_free_ha(struct ast_ha *ha)
{
struct ast_ha *hal;
while (ha) {
hal = ha;
ha = ha->next;
ast_free(hal);
}
}
/* Free ACL list structure */
struct ast_acl_list *ast_free_acl_list(struct ast_acl_list *acl_list)
{
struct ast_acl *current;
if (!acl_list) {
return NULL;
}
AST_LIST_LOCK(acl_list);
while ((current = AST_LIST_REMOVE_HEAD(acl_list, list))) {
ast_free_ha(current->acl);
ast_free(current);
}
AST_LIST_UNLOCK(acl_list);
AST_LIST_HEAD_DESTROY(acl_list);
ast_free(acl_list);
return NULL;
}
/* Copy HA structure */
void ast_copy_ha(const struct ast_ha *from, struct ast_ha *to)
{
ast_sockaddr_copy(&to->addr, &from->addr);
ast_sockaddr_copy(&to->netmask, &from->netmask);
to->sense = from->sense;
}
/* Create duplicate of ha structure */
static struct ast_ha *ast_duplicate_ha(struct ast_ha *original)
{
struct ast_ha *new_ha;
if ((new_ha = ast_calloc(1, sizeof(*new_ha)))) {
/* Copy from original to new object */
ast_copy_ha(original, new_ha);
}
return new_ha;
}
/* Create duplicate HA link list */
/* Used in chan_sip2 templates */
struct ast_ha *ast_duplicate_ha_list(struct ast_ha *original)
{
struct ast_ha *start = original;
struct ast_ha *ret = NULL;
struct ast_ha *current, *prev = NULL;
while (start) {
current = ast_duplicate_ha(start); /* Create copy of this object */
if (!current) {
ast_free_ha(ret);
return NULL;
}
if (prev) {
prev->next = current; /* Link previous to this object */
}
if (!ret) {
ret = current; /* Save starting point */
}
start = start->next; /* Go to next object */
prev = current; /* Save pointer to this object */
}
return ret; /* Return start of list */
}
static int acl_new(struct ast_acl **pointer, const char *name) {
struct ast_acl *acl;
if (!(acl = ast_calloc(1, sizeof(*acl)))) {
return 1;
}
*pointer = acl;
ast_copy_string(acl->name, name, ACL_NAME_LENGTH);
return 0;
}
struct ast_acl_list *ast_duplicate_acl_list(struct ast_acl_list *original)
{
struct ast_acl_list *clone;
struct ast_acl *current_cursor;
struct ast_acl *current_clone;
/* Early return if we receive a duplication request for a NULL original. */
if (!original) {
return NULL;
}
if (!(clone = ast_calloc(1, sizeof(*clone)))) {
ast_log(LOG_ERROR, "Failed to allocate ast_acl_list struct while cloning an ACL\n");
return NULL;
}
AST_LIST_HEAD_INIT(clone);
AST_LIST_LOCK(original);
AST_LIST_TRAVERSE(original, current_cursor, list) {
if ((acl_new(&current_clone, current_cursor->name))) {
ast_log(LOG_ERROR, "Failed to allocate ast_acl struct while cloning an ACL.\n");
ast_free_acl_list(clone);
clone = NULL;
break;
}
/* Copy data from original ACL to clone ACL */
current_clone->acl = ast_duplicate_ha_list(current_cursor->acl);
current_clone->is_invalid = current_cursor->is_invalid;
current_clone->is_realtime = current_cursor->is_realtime;
AST_LIST_INSERT_TAIL(clone, current_clone, list);
if (current_cursor->acl && !current_clone->acl) {
/* Deal with failure after adding to clone so we don't have to free
* current_clone separately. */
ast_log(LOG_ERROR, "Failed to duplicate HA list while cloning ACL.\n");
ast_free_acl_list(clone);
clone = NULL;
break;
}
}
AST_LIST_UNLOCK(original);
return clone;
}
/*!
* \brief
* Parse a netmask in CIDR notation
*
* \details
* For a mask of an IPv4 address, this should be a number between 0 and 32. For
* a mask of an IPv6 address, this should be a number between 0 and 128. This
* function creates an IPv6 ast_sockaddr from the given netmask. For masks of
* IPv4 addresses, this is accomplished by adding 96 to the original netmask.
*
* \param[out] addr The ast_sockaddr produced from the CIDR netmask
* \param is_v4 Tells if the address we are masking is IPv4.
* \param mask_str The CIDR mask to convert
* \retval -1 Failure
* \retval 0 Success
*/
static int parse_cidr_mask(struct ast_sockaddr *addr, int is_v4, const char *mask_str)
{
int mask;
if (sscanf(mask_str, "%30d", &mask) != 1) {
return -1;
}
if (is_v4) {
struct sockaddr_in sin;
if (mask < 0 || mask > 32) {
return -1;
}
memset(&sin, 0, sizeof(sin));
sin.sin_family = AF_INET;
/* If mask is 0, then we already have the
* appropriate all 0s address in sin from
* the above memset.
*/
if (mask != 0) {
sin.sin_addr.s_addr = htonl(0xFFFFFFFF << (32 - mask));
}
ast_sockaddr_from_sin(addr, &sin);
} else {
struct sockaddr_in6 sin6;
int i;
if (mask < 0 || mask > 128) {
return -1;
}
memset(&sin6, 0, sizeof(sin6));
sin6.sin6_family = AF_INET6;
for (i = 0; i < 4; ++i) {
/* Once mask reaches 0, we don't have
* to explicitly set anything anymore
* since sin6 was zeroed out already
*/
if (mask > 0) {
V6_WORD(&sin6, i) = htonl(0xFFFFFFFF << (mask < 32 ? (32 - mask) : 0));
mask -= mask < 32 ? mask : 32;
}
}
memcpy(&addr->ss, &sin6, sizeof(sin6));
addr->len = sizeof(sin6);
}
return 0;
}
void ast_append_acl(const char *sense, const char *stuff, struct ast_acl_list **path, int *error, int *named_acl_flag)
{
struct ast_acl *acl = NULL;
struct ast_acl *current;
struct ast_acl_list *working_list;
char *tmp, *list;
/* If the ACL list is currently uninitialized, it must be initialized. */
if (*path == NULL) {
struct ast_acl_list *list;
list = ast_calloc(1, sizeof(*list));
if (!list) {
/* Allocation Error */
if (error) {
*error = 1;
}
return;
}
AST_LIST_HEAD_INIT(list);
*path = list;
}
working_list = *path;
AST_LIST_LOCK(working_list);
/* First we need to determine if we will need to add a new ACL node or if we can use an existing one. */
if (strncasecmp(sense, "a", 1)) {
/* The first element in the path should be the unnamed, base ACL. If that's the case, we use it. If not,
* we have to make one and link it up appropriately. */
current = AST_LIST_FIRST(working_list);
if (!current || !ast_strlen_zero(current->name)) {
if (acl_new(&acl, "")) {
if (error) {
*error = 1;
}
AST_LIST_UNLOCK(working_list);
return;
}
// Need to INSERT the ACL at the head here.
AST_LIST_INSERT_HEAD(working_list, acl, list);
} else {
/* If the first element was already the unnamed base ACL, we just use that one. */
acl = current;
}
/* With the proper ACL set for modification, we can just pass this off to the ast_ha append function. */
acl->acl = ast_append_ha(sense, stuff, acl->acl, error);
AST_LIST_UNLOCK(working_list);
return;
}
/* We are in ACL append mode, so we know we'll be adding one or more named ACLs. */
list = ast_strdupa(stuff);
while ((tmp = strsep(&list, ","))) {
struct ast_ha *named_ha;
int already_included = 0;
/* Remove leading whitespace from the string in case the user put spaces between items */
tmp = ast_skip_blanks(tmp);
/* The first step is to check for a duplicate */
AST_LIST_TRAVERSE(working_list, current, list) {
if (!strcasecmp(current->name, tmp)) { /* ACL= */
/* Inclusion of the same ACL multiple times isn't a catastrophic error, but it will raise the error flag and skip the entry. */
ast_log(LOG_ERROR, "Named ACL '%s' occurs multiple times in ACL definition. "
"Please update your ACL configuration.\n", tmp);
if (error) {
*error = 1;
}
already_included = 1;
break;
}
}
if (already_included) {
continue;
}
if (acl_new(&acl, tmp)) {
/* This is a catastrophic allocation error and we'll return immediately if this happens. */
if (error) {
*error = 1;
}
AST_LIST_UNLOCK(working_list);
return;
}
/* Attempt to grab the Named ACL we are looking for. */
named_ha = ast_named_acl_find(tmp, &acl->is_realtime, &acl->is_invalid);
/* Set the ACL's ast_ha to the duplicated named ACL retrieved above. */
acl->acl = named_ha;
/* Raise the named_acl_flag since we are adding a named ACL to the ACL container. */
if (named_acl_flag) {
*named_acl_flag = 1;
}
/* Now insert the new ACL at the end of the list. */
AST_LIST_INSERT_TAIL(working_list, acl, list);
}
AST_LIST_UNLOCK(working_list);
}
int ast_acl_list_is_empty(struct ast_acl_list *acl_list)
{
struct ast_acl *head;
if (!acl_list) {
return 1;
}
AST_LIST_LOCK(acl_list);
head = AST_LIST_FIRST(acl_list);
AST_LIST_UNLOCK(acl_list);
if (head) {
return 0;
}
return 1;
}
/*!
* \internal
* \brief Used by ast_append_ha to avoid ast_strdupa in a loop.
*
* \note This function is only called at debug level 3 and higher.
*/
static void debug_ha_sense_appended(struct ast_ha *ha)
{
const char *parsed_mask = ast_strdupa(ast_sockaddr_stringify(&ha->netmask));
ast_log(LOG_DEBUG, "%s/%s sense %u appended to ACL\n",
ast_sockaddr_stringify(&ha->addr),
parsed_mask,
ha->sense);
}
static struct ast_ha *append_ha_core(const char *sense, const char *stuff, struct ast_ha *path, int *error, int port_flags)
{
struct ast_ha *ha;
struct ast_ha *prev = NULL;
struct ast_ha *ret;
char *tmp, *list = ast_strdupa(stuff ?: "");
char *address = NULL, *mask = NULL;
int addr_is_v4;
int allowing = strncasecmp(sense, "p", 1) ? AST_SENSE_DENY : AST_SENSE_ALLOW;
ret = path;
while (path) {
prev = path;
path = path->next;
}
while ((tmp = strsep(&list, ","))) {
uint16_t save_port;
if (!(ha = ast_calloc(1, sizeof(*ha)))) {
if (error) {
*error = 1;
}
return ret;
}
address = strsep(&tmp, "/");
if (!address) {
address = tmp;
} else {
mask = tmp;
}
if (*address == '!') {
ha->sense = (allowing == AST_SENSE_DENY) ? AST_SENSE_ALLOW : AST_SENSE_DENY;
address++;
} else {
ha->sense = allowing;
}
if (!ast_sockaddr_parse(&ha->addr, address, port_flags)) {
ast_log(LOG_WARNING, "Invalid IP address: %s\n", address);
ast_free_ha(ha);
if (error) {
*error = 1;
}
return ret;
}
/* Be pedantic and zero out the port if we don't want it */
if ((port_flags & PARSE_PORT_MASK) == PARSE_PORT_FORBID) {
ast_sockaddr_set_port(&ha->addr, 0);
}
/* If someone specifies an IPv4-mapped IPv6 address,
* we just convert this to an IPv4 ACL
*/
if (ast_sockaddr_ipv4_mapped(&ha->addr, &ha->addr)) {
ast_log(LOG_NOTICE, "IPv4-mapped ACL network address specified. "
"Converting to an IPv4 ACL network address.\n");
}
addr_is_v4 = ast_sockaddr_is_ipv4(&ha->addr);
if (!mask) {
parse_cidr_mask(&ha->netmask, addr_is_v4, addr_is_v4 ? "32" : "128");
} else if (strchr(mask, ':') || strchr(mask, '.')) {
int mask_is_v4;
/* Mask is of x.x.x.x or x:x:x:x:x:x:x:x variety */
if (!ast_sockaddr_parse(&ha->netmask, mask, PARSE_PORT_FORBID)) {
ast_log(LOG_WARNING, "Invalid netmask: %s\n", mask);
ast_free_ha(ha);
if (error) {
*error = 1;
}
return ret;
}
/* If someone specifies an IPv4-mapped IPv6 netmask,
* we just convert this to an IPv4 ACL
*/
if (ast_sockaddr_ipv4_mapped(&ha->netmask, &ha->netmask)) {
ast_log(LOG_NOTICE, "IPv4-mapped ACL netmask specified. "
"Converting to an IPv4 ACL netmask.\n");
}
mask_is_v4 = ast_sockaddr_is_ipv4(&ha->netmask);
if (addr_is_v4 ^ mask_is_v4) {
ast_log(LOG_WARNING, "Address and mask are not using same address scheme.\n");
ast_free_ha(ha);
if (error) {
*error = 1;
}
return ret;
}
} else if (parse_cidr_mask(&ha->netmask, addr_is_v4, mask)) {
ast_log(LOG_WARNING, "Invalid CIDR netmask: %s\n", mask);
ast_free_ha(ha);
if (error) {
*error = 1;
}
return ret;
}
/* ast_sockaddr_apply_netmask() does not preserve the port, so we need to save and
* restore it */
save_port = ast_sockaddr_port(&ha->addr);
if (ast_sockaddr_apply_netmask(&ha->addr, &ha->netmask, &ha->addr)) {
/* This shouldn't happen because ast_sockaddr_parse would
* have failed much earlier on an unsupported address scheme
*/
char *failmask = ast_strdupa(ast_sockaddr_stringify(&ha->netmask));
char *failaddr = ast_strdupa(ast_sockaddr_stringify(&ha->addr));
ast_log(LOG_WARNING, "Unable to apply netmask %s to address %s\n", failmask, failaddr);
ast_free_ha(ha);
if (error) {
*error = 1;
}
return ret;
}
ast_sockaddr_set_port(&ha->addr, save_port);
if (prev) {
prev->next = ha;
} else {
ret = ha;
}
prev = ha;
if (DEBUG_ATLEAST(3)) {
debug_ha_sense_appended(ha);
}
}
return ret;
}
struct ast_ha *ast_append_ha(const char *sense, const char *stuff, struct ast_ha *path, int *error)
{
return append_ha_core(sense, stuff, path, error, PARSE_PORT_FORBID);
}
struct ast_ha *ast_append_ha_with_port(const char *sense, const char *stuff, struct ast_ha *path, int *error)
{
return append_ha_core(sense, stuff, path, error, 0);
}
void ast_ha_join(const struct ast_ha *ha, struct ast_str **buf)
{
for (; ha; ha = ha->next) {
const char *addr;
if (ast_sockaddr_port(&ha->addr)) {
addr = ast_sockaddr_stringify(&ha->addr);
} else {
addr = ast_sockaddr_stringify_addr(&ha->addr);
}
ast_str_append(buf, 0, "%s%s/",
ha->sense == AST_SENSE_ALLOW ? "!" : "",
addr);
/* Separated to avoid duplicating stringified addresses. */
ast_str_append(buf, 0, "%s", ast_sockaddr_stringify_addr(&ha->netmask));
if (ha->next) {
ast_str_append(buf, 0, ",");
}
}
}
void ast_ha_join_cidr(const struct ast_ha *ha, struct ast_str **buf)
{
for (; ha; ha = ha->next) {
const char *addr = ast_sockaddr_stringify_addr(&ha->addr);
ast_str_append(buf, 0, "%s%s/%d",
ha->sense == AST_SENSE_ALLOW ? "!" : "",
addr, ast_sockaddr_cidr_bits(&ha->netmask));
if (ha->next) {
ast_str_append(buf, 0, ",");
}
}
}
static enum ast_acl_sense ast_apply_acl_internal(struct ast_acl_list *acl_list, const struct ast_sockaddr *addr, const char *log_prefix)
{
struct ast_acl *acl;
/* If the list is NULL, there are no rules, so we'll allow automatically. */
if (!acl_list) {
return AST_SENSE_ALLOW;
}
AST_LIST_LOCK(acl_list);
AST_LIST_TRAVERSE(acl_list, acl, list) {
if (acl->is_invalid) {
/* In this case, the baseline ACL shouldn't ever trigger this, but if that somehow happens, it'll still be shown. */
if (log_prefix) {
ast_log(LOG_WARNING, "%sRejecting '%s' due to use of an invalid ACL '%s'.\n",
log_prefix, ast_sockaddr_stringify_addr(addr),
ast_strlen_zero(acl->name) ? "(BASELINE)" : acl->name);
}
AST_LIST_UNLOCK(acl_list);
return AST_SENSE_DENY;
}
if (acl->acl) {
if (ast_apply_ha(acl->acl, addr) == AST_SENSE_DENY) {
if (log_prefix) {
ast_log(LOG_NOTICE, "%sRejecting '%s' due to a failure to pass ACL '%s'\n",
log_prefix, ast_sockaddr_stringify_addr(addr),
ast_strlen_zero(acl->name) ? "(BASELINE)" : acl->name);
}
AST_LIST_UNLOCK(acl_list);
return AST_SENSE_DENY;
}
}
}
AST_LIST_UNLOCK(acl_list);
return AST_SENSE_ALLOW;
}
enum ast_acl_sense ast_apply_acl(struct ast_acl_list *acl_list, const struct ast_sockaddr *addr, const char *purpose) {
return ast_apply_acl_internal(acl_list, addr, purpose ?: "");
}
enum ast_acl_sense ast_apply_acl_nolog(struct ast_acl_list *acl_list, const struct ast_sockaddr *addr) {
return ast_apply_acl_internal(acl_list, addr, NULL);
}
enum ast_acl_sense ast_apply_ha(const struct ast_ha *ha, const struct ast_sockaddr *addr)
{
/* Start optimistic */
enum ast_acl_sense res = AST_SENSE_ALLOW;
const struct ast_ha *current_ha;
for (current_ha = ha; current_ha; current_ha = current_ha->next) {
struct ast_sockaddr result;
struct ast_sockaddr mapped_addr;
const struct ast_sockaddr *addr_to_use;
uint16_t save_port;
#if 0 /* debugging code */
char iabuf[INET_ADDRSTRLEN];
char iabuf2[INET_ADDRSTRLEN];
/* DEBUG */
ast_copy_string(iabuf, ast_sockaddr_stringify(addr), sizeof(iabuf));
ast_copy_string(iabuf2, ast_sockaddr_stringify(&current_ha->addr), sizeof(iabuf2));
ast_debug(1, "##### Testing %s with %s\n", iabuf, iabuf2);
#endif
if (ast_sockaddr_is_ipv4(&current_ha->addr)) {
if (ast_sockaddr_is_ipv6(addr)) {
if (ast_sockaddr_is_ipv4_mapped(addr)) {
/* IPv4 ACLs apply to IPv4-mapped addresses */
if (!ast_sockaddr_ipv4_mapped(addr, &mapped_addr)) {
ast_log(LOG_ERROR, "%s provided to ast_sockaddr_ipv4_mapped could not be converted. That shouldn't be possible.\n",
ast_sockaddr_stringify(addr));
continue;
}
addr_to_use = &mapped_addr;
} else {
/* An IPv4 ACL does not apply to an IPv6 address */
continue;
}
} else {
/* Address is IPv4 and ACL is IPv4. No biggie */
addr_to_use = addr;
}
} else {
if (ast_sockaddr_is_ipv6(addr) && !ast_sockaddr_is_ipv4_mapped(addr)) {
addr_to_use = addr;
} else {
/* Address is IPv4 or IPv4 mapped but ACL is IPv6. Skip */
continue;
}
}
/* ast_sockaddr_apply_netmask() does not preserve the port, so we need to save and
* restore it */
save_port = ast_sockaddr_port(addr_to_use);
/* For each rule, if this address and the netmask = the net address
apply the current rule */
if (ast_sockaddr_apply_netmask(addr_to_use, &current_ha->netmask, &result)) {
/* Unlikely to happen since we know the address to be IPv4 or IPv6 */
continue;
}
ast_sockaddr_set_port(&result, save_port);
if (!ast_sockaddr_cmp_addr(&result, &current_ha->addr)
&& (!ast_sockaddr_port(&current_ha->addr)
|| ast_sockaddr_port(&current_ha->addr) == ast_sockaddr_port(&result))) {
res = current_ha->sense;
}
}
return res;
}
static int resolve_first(struct ast_sockaddr *addr, const char *name, int flag,
int family)
{
struct ast_sockaddr *addrs;
int addrs_cnt;
addrs_cnt = ast_sockaddr_resolve(&addrs, name, flag, family);
if (addrs_cnt > 0) {
if (addrs_cnt > 1) {
ast_debug(1, "Multiple addresses. Using the first only\n");
}
ast_sockaddr_copy(addr, &addrs[0]);
ast_free(addrs);
} else {
ast_log(LOG_WARNING, "Unable to lookup '%s'\n", name);
return -1;
}
return 0;
}
int ast_get_ip_or_srv(struct ast_sockaddr *addr, const char *hostname, const char *service)
{
char srv[256];
char host[256];
int srv_ret = 0;
int tportno;
if (service) {
snprintf(srv, sizeof(srv), "%s.%s", service, hostname);
if ((srv_ret = ast_get_srv(NULL, host, sizeof(host), &tportno, srv)) > 0) {
hostname = host;
}
}
if (resolve_first(addr, hostname, PARSE_PORT_FORBID, addr->ss.ss_family) != 0) {
return -1;
}
if (srv_ret > 0) {
ast_sockaddr_set_port(addr, tportno);
}
return 0;
}
struct dscp_codepoint {
char *name;
unsigned int space;
};
/* IANA registered DSCP codepoints */
static const struct dscp_codepoint dscp_pool1[] = {
{ "CS0", 0x00 },
{ "CS1", 0x08 },
{ "CS2", 0x10 },
{ "CS3", 0x18 },
{ "CS4", 0x20 },
{ "CS5", 0x28 },
{ "CS6", 0x30 },
{ "CS7", 0x38 },
{ "AF11", 0x0A },
{ "AF12", 0x0C },
{ "AF13", 0x0E },
{ "AF21", 0x12 },
{ "AF22", 0x14 },
{ "AF23", 0x16 },
{ "AF31", 0x1A },
{ "AF32", 0x1C },
{ "AF33", 0x1E },
{ "AF41", 0x22 },
{ "AF42", 0x24 },
{ "AF43", 0x26 },
{ "EF", 0x2E },
};
int ast_str2cos(const char *value, unsigned int *cos)
{
int fval;
if (sscanf(value, "%30d", &fval) == 1) {
if (fval < 8) {
*cos = fval;
return 0;
}
}
return -1;
}
int ast_str2tos(const char *value, unsigned int *tos)
{
int fval;
unsigned int x;
if (sscanf(value, "%30i", &fval) == 1) {
*tos = fval & 0xFF;
return 0;
}
for (x = 0; x < ARRAY_LEN(dscp_pool1); x++) {
if (!strcasecmp(value, dscp_pool1[x].name)) {
*tos = dscp_pool1[x].space << 2;
return 0;
}
}
return -1;
}
const char *ast_tos2str(unsigned int tos)
{
unsigned int x;
for (x = 0; x < ARRAY_LEN(dscp_pool1); x++) {
if (dscp_pool1[x].space == (tos >> 2)) {
return dscp_pool1[x].name;
}
}
return "unknown";
}
int ast_get_ip(struct ast_sockaddr *addr, const char *hostname)
{
return ast_get_ip_or_srv(addr, hostname, NULL);
}
int ast_ouraddrfor(const struct ast_sockaddr *them, struct ast_sockaddr *us)
{
/*
* We must create the errno string before creating the address
* string because it could wipe out errno on the error return
* paths.
*/
const char *sock_err;
int port;
int s;
/* Preserve our original address port */
port = ast_sockaddr_port(us);
s = socket(ast_sockaddr_is_ipv6(them) ? AF_INET6 : AF_INET, SOCK_DGRAM, 0);
if (s < 0) {
sock_err = ast_strdupa(strerror(errno));
ast_log(LOG_ERROR, "Cannot create socket to %s: %s\n",
ast_sockaddr_stringify_addr(them), sock_err);
return -1;
}
if (ast_connect(s, them)) {
sock_err = ast_strdupa(strerror(errno));
ast_log(LOG_WARNING, "Cannot connect to %s: %s\n",
ast_sockaddr_stringify_addr(them), sock_err);
close(s);
return -1;
}
if (ast_getsockname(s, us)) {
sock_err = ast_strdupa(strerror(errno));
ast_log(LOG_WARNING, "Cannot get socket name for connection to %s: %s\n",
ast_sockaddr_stringify_addr(them), sock_err);
close(s);
return -1;
}
close(s);
ast_sockaddr_set_port(us, port);
ast_debug(3, "For destination '%s', our source address is '%s'.\n",
ast_strdupa(ast_sockaddr_stringify_addr(them)),
ast_strdupa(ast_sockaddr_stringify_addr(us)));
return 0;
}
int ast_find_ourip(struct ast_sockaddr *ourip, const struct ast_sockaddr *bindaddr, int family)
{
char ourhost[MAXHOSTNAMELEN] = "";
struct ast_sockaddr root;
int res, port = ast_sockaddr_port(ourip);
/* just use the bind address if it is nonzero */
if (!ast_sockaddr_is_any(bindaddr)) {
ast_sockaddr_copy(ourip, bindaddr);
ast_debug(3, "Attached to given IP address\n");
return 0;
}
/* try to use our hostname */
if (gethostname(ourhost, sizeof(ourhost) - 1)) {
ast_log(LOG_WARNING, "Unable to get hostname\n");
} else {
if (resolve_first(ourip, ourhost, PARSE_PORT_FORBID, family) == 0) {
/* reset port since resolve_first wipes this out */
ast_sockaddr_set_port(ourip, port);
return 0;
}
}
ast_debug(3, "Trying to check A.ROOT-SERVERS.NET and get our IP address for that connection\n");
/* A.ROOT-SERVERS.NET. */
if (!resolve_first(&root, "A.ROOT-SERVERS.NET", PARSE_PORT_FORBID, 0) &&
!ast_ouraddrfor(&root, ourip)) {
/* reset port since resolve_first wipes this out */
ast_sockaddr_set_port(ourip, port);
return 0;
}
res = get_local_address(ourip);
ast_sockaddr_set_port(ourip, port);
return res;
}
void ast_ha_output(int fd, const struct ast_ha *ha, const char *prefix)
{
char addr[AST_SOCKADDR_BUFLEN];
char *mask;
int index = 0;
for (; ha; ha = ha->next, ++index) {
strcpy(addr, ast_sockaddr_stringify_addr(&ha->addr));
mask = ast_sockaddr_stringify_addr(&ha->netmask);
ast_cli(fd, "%s%3d: %s - %s/%s\n", prefix ?: "", index, ha->sense == AST_SENSE_ALLOW ? "allow" : " deny", addr, mask);
}
}
void ast_acl_output(int fd, struct ast_acl_list *acl_list, const char *prefix)
{
struct ast_acl *acl;
AST_LIST_LOCK(acl_list);
AST_LIST_TRAVERSE(acl_list, acl, list) {
ast_cli(fd, "%sACL: %s%s\n---------------------------------------------\n",
prefix ?: "", ast_strlen_zero(acl->name) ? "(unnamed)" : acl->name,
acl->is_realtime ? " (realtime)" : "");
ast_ha_output(fd, acl->acl, prefix);
}
AST_LIST_UNLOCK(acl_list);
}