/*
* lib/route/addr.c Addresses
*
* This library is free software; you can redistribute it and/or
* modify it under the terms of the GNU Lesser General Public
* License as published by the Free Software Foundation version 2.1
* of the License.
*
* Copyright (c) 2003-2012 Thomas Graf <tgraf@suug.ch>
* Copyright (c) 2003-2006 Baruch Even <baruch@ev-en.org>,
* Mediatrix Telecom, inc. <ericb@mediatrix.com>
*/
/**
* @ingroup rtnl
* @defgroup rtaddr Addresses
* @brief
*
* @note The maximum size of an address label is IFNAMSIZ.
*
* @note The address may not contain a prefix length if the peer address
* has been specified already.
*
* @par 1) Address Addition
* @code
* // Allocate an empty address object to be filled out with the attributes
* // of the new address.
* struct rtnl_addr *addr = rtnl_addr_alloc();
*
* // Fill out the mandatory attributes of the new address. Setting the
* // local address will automatically set the address family and the
* // prefix length to the correct values.
* rtnl_addr_set_ifindex(addr, ifindex);
* rtnl_addr_set_local(addr, local_addr);
*
* // The label of the address can be specified, currently only supported
* // by IPv4 and DECnet.
* rtnl_addr_set_label(addr, "mylabel");
*
* // The peer address can be specified if necessary, in either case a peer
* // address will be sent to the kernel in order to fullfil the interface
* // requirements. If none is set, it will equal the local address.
* // Note: Real peer addresses are only supported by IPv4 for now.
* rtnl_addr_set_peer(addr, peer_addr);
*
* // In case you want to have the address have a scope other than global
* // it may be overwritten using rtnl_addr_set_scope(). The scope currently
* // cannot be set for IPv6 addresses.
* rtnl_addr_set_scope(addr, rtnl_str2scope("site"));
*
* // Broadcast address may be specified using the relevant
* // functions, the address family will be verified if one of the other
* // addresses has been set already. Currently only works for IPv4.
* rtnl_addr_set_broadcast(addr, broadcast_addr);
*
* // Build the netlink message and send it to the kernel, the operation will
* // block until the operation has been completed. Alternatively the required
* // netlink message can be built using rtnl_addr_build_add_request() to be
* // sent out using nl_send_auto_complete().
* rtnl_addr_add(sk, addr, 0);
*
* // Free the memory
* rtnl_addr_put(addr);
* @endcode
*
* @par 2) Address Deletion
* @code
* // Allocate an empty address object to be filled out with the attributes
* // matching the address to be deleted. Alternatively a fully equipped
* // address object out of a cache can be used instead.
* struct rtnl_addr *addr = rtnl_addr_alloc();
*
* // The only mandatory parameter besides the address family is the interface
* // index the address is on, i.e. leaving out all other parameters will
* // result in all addresses of the specified address family interface tuple
* // to be deleted.
* rtnl_addr_set_ifindex(addr, ifindex);
*
* // Specyfing the address family manually is only required if neither the
* // local nor peer address have been specified.
* rtnl_addr_set_family(addr, AF_INET);
*
* // Specyfing the local address is optional but the best choice to delete
* // specific addresses.
* rtnl_addr_set_local(addr, local_addr);
*
* // The label of the address can be specified, currently only supported
* // by IPv4 and DECnet.
* rtnl_addr_set_label(addr, "mylabel");
*
* // The peer address can be specified if necessary, in either case a peer
* // address will be sent to the kernel in order to fullfil the interface
* // requirements. If none is set, it will equal the local address.
* // Note: Real peer addresses are only supported by IPv4 for now.
* rtnl_addr_set_peer(addr, peer_addr);
*
* // Build the netlink message and send it to the kernel, the operation will
* // block until the operation has been completed. Alternatively the required
* // netlink message can be built using rtnl_addr_build_delete_request()
* // to be sent out using nl_send_auto_complete().
* rtnl_addr_delete(sk, addr, 0);
*
* // Free the memory
* rtnl_addr_put(addr);
* @endcode
* @{
*/
#include <netlink-private/netlink.h>
#include <netlink/netlink.h>
#include <netlink/route/rtnl.h>
#include <netlink/route/addr.h>
#include <netlink/route/route.h>
#include <netlink/route/link.h>
#include <netlink/utils.h>
/** @cond SKIP */
#define ADDR_ATTR_FAMILY 0x0001
#define ADDR_ATTR_PREFIXLEN 0x0002
#define ADDR_ATTR_FLAGS 0x0004
#define ADDR_ATTR_SCOPE 0x0008
#define ADDR_ATTR_IFINDEX 0x0010
#define ADDR_ATTR_LABEL 0x0020
#define ADDR_ATTR_CACHEINFO 0x0040
#define ADDR_ATTR_PEER 0x0080
#define ADDR_ATTR_LOCAL 0x0100
#define ADDR_ATTR_BROADCAST 0x0200
#define ADDR_ATTR_MULTICAST 0x0400
#define ADDR_ATTR_ANYCAST 0x0800
static struct nl_cache_ops rtnl_addr_ops;
static struct nl_object_ops addr_obj_ops;
/** @endcond */
static void addr_constructor(struct nl_object *obj)
{
struct rtnl_addr *addr = nl_object_priv(obj);
addr->a_scope = RT_SCOPE_NOWHERE;
}
static void addr_free_data(struct nl_object *obj)
{
struct rtnl_addr *addr = nl_object_priv(obj);
if (!addr)
return;
nl_addr_put(addr->a_peer);
nl_addr_put(addr->a_local);
nl_addr_put(addr->a_bcast);
nl_addr_put(addr->a_multicast);
nl_addr_put(addr->a_anycast);
rtnl_link_put(addr->a_link);
}
static int addr_clone(struct nl_object *_dst, struct nl_object *_src)
{
struct rtnl_addr *dst = nl_object_priv(_dst);
struct rtnl_addr *src = nl_object_priv(_src);
if (src->a_link) {
nl_object_get(OBJ_CAST(src->a_link));
dst->a_link = src->a_link;
}
if (src->a_peer)
if (!(dst->a_peer = nl_addr_clone(src->a_peer)))
return -NLE_NOMEM;
if (src->a_local)
if (!(dst->a_local = nl_addr_clone(src->a_local)))
return -NLE_NOMEM;
if (src->a_bcast)
if (!(dst->a_bcast = nl_addr_clone(src->a_bcast)))
return -NLE_NOMEM;
if (src->a_multicast)
if (!(dst->a_multicast = nl_addr_clone(src->a_multicast)))
return -NLE_NOMEM;
if (src->a_anycast)
if (!(dst->a_anycast = nl_addr_clone(src->a_anycast)))
return -NLE_NOMEM;
return 0;
}
static struct nla_policy addr_policy[IFA_MAX+1] = {
[IFA_LABEL] = { .type = NLA_STRING,
.maxlen = IFNAMSIZ },
[IFA_CACHEINFO] = { .minlen = sizeof(struct ifa_cacheinfo) },
};
static int addr_msg_parser(struct nl_cache_ops *ops, struct sockaddr_nl *who,
struct nlmsghdr *nlh, struct nl_parser_param *pp)
{
struct rtnl_addr *addr;
struct ifaddrmsg *ifa;
struct nlattr *tb[IFA_MAX+1];
int err, family;
struct nl_cache *link_cache;
struct nl_addr *plen_addr = NULL;
addr = rtnl_addr_alloc();
if (!addr)
return -NLE_NOMEM;
addr->ce_msgtype = nlh->nlmsg_type;
err = nlmsg_parse(nlh, sizeof(*ifa), tb, IFA_MAX, addr_policy);
if (err < 0)
goto errout;
ifa = nlmsg_data(nlh);
addr->a_family = family = ifa->ifa_family;
addr->a_prefixlen = ifa->ifa_prefixlen;
addr->a_scope = ifa->ifa_scope;
addr->a_flags = tb[IFA_FLAGS] ? nla_get_u32(tb[IFA_FLAGS]) :
ifa->ifa_flags;
addr->a_ifindex = ifa->ifa_index;
addr->ce_mask = (ADDR_ATTR_FAMILY | ADDR_ATTR_PREFIXLEN |
ADDR_ATTR_FLAGS | ADDR_ATTR_SCOPE | ADDR_ATTR_IFINDEX);
if (tb[IFA_LABEL]) {
nla_strlcpy(addr->a_label, tb[IFA_LABEL], IFNAMSIZ);
addr->ce_mask |= ADDR_ATTR_LABEL;
}
/* IPv6 only */
if (tb[IFA_CACHEINFO]) {
struct ifa_cacheinfo *ca;
ca = nla_data(tb[IFA_CACHEINFO]);
addr->a_cacheinfo.aci_prefered = ca->ifa_prefered;
addr->a_cacheinfo.aci_valid = ca->ifa_valid;
addr->a_cacheinfo.aci_cstamp = ca->cstamp;
addr->a_cacheinfo.aci_tstamp = ca->tstamp;
addr->ce_mask |= ADDR_ATTR_CACHEINFO;
}
if (tb[IFA_LOCAL]) {
addr->a_local = nl_addr_alloc_attr(tb[IFA_LOCAL], family);
if (!addr->a_local)
goto errout_nomem;
addr->ce_mask |= ADDR_ATTR_LOCAL;
plen_addr = addr->a_local;
}
if (tb[IFA_ADDRESS]) {
struct nl_addr *a;
a = nl_addr_alloc_attr(tb[IFA_ADDRESS], family);
if (!a)
goto errout_nomem;
/* IPv6 sends the local address as IFA_ADDRESS with
* no IFA_LOCAL, IPv4 sends both IFA_LOCAL and IFA_ADDRESS
* with IFA_ADDRESS being the peer address if they differ */
if (!tb[IFA_LOCAL] || !nl_addr_cmp(a, addr->a_local)) {
nl_addr_put(addr->a_local);
addr->a_local = a;
addr->ce_mask |= ADDR_ATTR_LOCAL;
} else {
addr->a_peer = a;
addr->ce_mask |= ADDR_ATTR_PEER;
}
plen_addr = a;
}
if (plen_addr)
nl_addr_set_prefixlen(plen_addr, addr->a_prefixlen);
/* IPv4 only */
if (tb[IFA_BROADCAST]) {
addr->a_bcast = nl_addr_alloc_attr(tb[IFA_BROADCAST], family);
if (!addr->a_bcast)
goto errout_nomem;
addr->ce_mask |= ADDR_ATTR_BROADCAST;
}
/* IPv6 only */
if (tb[IFA_MULTICAST]) {
addr->a_multicast = nl_addr_alloc_attr(tb[IFA_MULTICAST],
family);
if (!addr->a_multicast)
goto errout_nomem;
addr->ce_mask |= ADDR_ATTR_MULTICAST;
}
/* IPv6 only */
if (tb[IFA_ANYCAST]) {
addr->a_anycast = nl_addr_alloc_attr(tb[IFA_ANYCAST],
family);
if (!addr->a_anycast)
goto errout_nomem;
addr->ce_mask |= ADDR_ATTR_ANYCAST;
}
if ((link_cache = __nl_cache_mngt_require("route/link"))) {
struct rtnl_link *link;
if ((link = rtnl_link_get(link_cache, addr->a_ifindex))) {
rtnl_addr_set_link(addr, link);
/* rtnl_addr_set_link incs refcnt */
rtnl_link_put(link);
}
}
err = pp->pp_cb((struct nl_object *) addr, pp);
errout:
rtnl_addr_put(addr);
return err;
errout_nomem:
err = -NLE_NOMEM;
goto errout;
}
static int addr_request_update(struct nl_cache *cache, struct nl_sock *sk)
{
return nl_rtgen_request(sk, RTM_GETADDR, AF_UNSPEC, NLM_F_DUMP);
}
static void addr_dump_line(struct nl_object *obj, struct nl_dump_params *p)
{
struct rtnl_addr *addr = (struct rtnl_addr *) obj;
struct nl_cache *link_cache;
char buf[128];
link_cache = nl_cache_mngt_require_safe("route/link");
if (addr->ce_mask & ADDR_ATTR_LOCAL)
nl_dump_line(p, "%s",
nl_addr2str(addr->a_local, buf, sizeof(buf)));
else
nl_dump_line(p, "none");
if (addr->ce_mask & ADDR_ATTR_PEER)
nl_dump(p, " peer %s",
nl_addr2str(addr->a_peer, buf, sizeof(buf)));
nl_dump(p, " %s ", nl_af2str(addr->a_family, buf, sizeof(buf)));
if (link_cache)
nl_dump(p, "dev %s ",
rtnl_link_i2name(link_cache, addr->a_ifindex,
buf, sizeof(buf)));
else
nl_dump(p, "dev %d ", addr->a_ifindex);
nl_dump(p, "scope %s",
rtnl_scope2str(addr->a_scope, buf, sizeof(buf)));
rtnl_addr_flags2str(addr->a_flags, buf, sizeof(buf));
if (buf[0])
nl_dump(p, " <%s>", buf);
nl_dump(p, "\n");
if (link_cache)
nl_cache_put(link_cache);
}
static void addr_dump_details(struct nl_object *obj, struct nl_dump_params *p)
{
struct rtnl_addr *addr = (struct rtnl_addr *) obj;
char buf[128];
addr_dump_line(obj, p);
if (addr->ce_mask & (ADDR_ATTR_LABEL | ADDR_ATTR_BROADCAST |
ADDR_ATTR_MULTICAST)) {
nl_dump_line(p, " ");
if (addr->ce_mask & ADDR_ATTR_LABEL)
nl_dump(p, " label %s", addr->a_label);
if (addr->ce_mask & ADDR_ATTR_BROADCAST)
nl_dump(p, " broadcast %s",
nl_addr2str(addr->a_bcast, buf, sizeof(buf)));
if (addr->ce_mask & ADDR_ATTR_MULTICAST)
nl_dump(p, " multicast %s",
nl_addr2str(addr->a_multicast, buf,
sizeof(buf)));
if (addr->ce_mask & ADDR_ATTR_ANYCAST)
nl_dump(p, " anycast %s",
nl_addr2str(addr->a_anycast, buf,
sizeof(buf)));
nl_dump(p, "\n");
}
if (addr->ce_mask & ADDR_ATTR_CACHEINFO) {
struct rtnl_addr_cacheinfo *ci = &addr->a_cacheinfo;
nl_dump_line(p, " valid-lifetime %s",
ci->aci_valid == 0xFFFFFFFFU ? "forever" :
nl_msec2str(ci->aci_valid * 1000,
buf, sizeof(buf)));
nl_dump(p, " preferred-lifetime %s\n",
ci->aci_prefered == 0xFFFFFFFFU ? "forever" :
nl_msec2str(ci->aci_prefered * 1000,
buf, sizeof(buf)));
nl_dump_line(p, " created boot-time+%s ",
nl_msec2str(addr->a_cacheinfo.aci_cstamp * 10,
buf, sizeof(buf)));
nl_dump(p, "last-updated boot-time+%s\n",
nl_msec2str(addr->a_cacheinfo.aci_tstamp * 10,
buf, sizeof(buf)));
}
}
static void addr_dump_stats(struct nl_object *obj, struct nl_dump_params *p)
{
addr_dump_details(obj, p);
}
static int addr_compare(struct nl_object *_a, struct nl_object *_b,
uint32_t attrs, int flags)
{
struct rtnl_addr *a = (struct rtnl_addr *) _a;
struct rtnl_addr *b = (struct rtnl_addr *) _b;
int diff = 0;
#define ADDR_DIFF(ATTR, EXPR) ATTR_DIFF(attrs, ADDR_ATTR_##ATTR, a, b, EXPR)
diff |= ADDR_DIFF(IFINDEX, a->a_ifindex != b->a_ifindex);
diff |= ADDR_DIFF(FAMILY, a->a_family != b->a_family);
diff |= ADDR_DIFF(SCOPE, a->a_scope != b->a_scope);
diff |= ADDR_DIFF(LABEL, strcmp(a->a_label, b->a_label));
diff |= ADDR_DIFF(PEER, nl_addr_cmp(a->a_peer, b->a_peer));
diff |= ADDR_DIFF(LOCAL, nl_addr_cmp(a->a_local, b->a_local));
diff |= ADDR_DIFF(MULTICAST, nl_addr_cmp(a->a_multicast,
b->a_multicast));
diff |= ADDR_DIFF(BROADCAST, nl_addr_cmp(a->a_bcast, b->a_bcast));
diff |= ADDR_DIFF(ANYCAST, nl_addr_cmp(a->a_anycast, b->a_anycast));
if (flags & LOOSE_COMPARISON)
diff |= ADDR_DIFF(FLAGS,
(a->a_flags ^ b->a_flags) & b->a_flag_mask);
else
diff |= ADDR_DIFF(FLAGS, a->a_flags != b->a_flags);
#undef ADDR_DIFF
return diff;
}
static const struct trans_tbl addr_attrs[] = {
__ADD(ADDR_ATTR_FAMILY, family)
__ADD(ADDR_ATTR_PREFIXLEN, prefixlen)
__ADD(ADDR_ATTR_FLAGS, flags)
__ADD(ADDR_ATTR_SCOPE, scope)
__ADD(ADDR_ATTR_IFINDEX, ifindex)
__ADD(ADDR_ATTR_LABEL, label)
__ADD(ADDR_ATTR_CACHEINFO, cacheinfo)
__ADD(ADDR_ATTR_PEER, peer)
__ADD(ADDR_ATTR_LOCAL, local)
__ADD(ADDR_ATTR_BROADCAST, broadcast)
__ADD(ADDR_ATTR_MULTICAST, multicast)
};
static char *addr_attrs2str(int attrs, char *buf, size_t len)
{
return __flags2str(attrs, buf, len, addr_attrs,
ARRAY_SIZE(addr_attrs));
}
/**
* @name Allocation/Freeing
* @{
*/
struct rtnl_addr *rtnl_addr_alloc(void)
{
return (struct rtnl_addr *) nl_object_alloc(&addr_obj_ops);
}
void rtnl_addr_put(struct rtnl_addr *addr)
{
nl_object_put((struct nl_object *) addr);
}
/** @} */
/**
* @name Cache Management
* @{
*/
int rtnl_addr_alloc_cache(struct nl_sock *sk, struct nl_cache **result)
{
return nl_cache_alloc_and_fill(&rtnl_addr_ops, sk, result);
}
/**
* Search address in cache
* @arg cache Address cache
* @arg ifindex Interface index of address
* @arg addr Local address part
*
* Searches address cache previously allocated with rtnl_addr_alloc_cache()
* for an address with a matching local address.
*
* The reference counter is incremented before returning the address, therefore
* the reference must be given back with rtnl_addr_put() after usage.
*
* @return Address object or NULL if no match was found.
*/
struct rtnl_addr *rtnl_addr_get(struct nl_cache *cache, int ifindex,
struct nl_addr *addr)
{
struct rtnl_addr *a;
if (cache->c_ops != &rtnl_addr_ops)
return NULL;
nl_list_for_each_entry(a, &cache->c_items, ce_list) {
if (ifindex && a->a_ifindex != ifindex)
continue;
if (a->ce_mask & ADDR_ATTR_LOCAL &&
!nl_addr_cmp(a->a_local, addr)) {
nl_object_get((struct nl_object *) a);
return a;
}
}
return NULL;
}
/** @} */
static int build_addr_msg(struct rtnl_addr *tmpl, int cmd, int flags,
struct nl_msg **result)
{
struct nl_msg *msg;
struct ifaddrmsg am = {
.ifa_family = tmpl->a_family,
.ifa_index = tmpl->a_ifindex,
.ifa_prefixlen = tmpl->a_prefixlen,
.ifa_flags = tmpl->a_flags,
};
if (tmpl->ce_mask & ADDR_ATTR_SCOPE)
am.ifa_scope = tmpl->a_scope;
else {
/* compatibility hack */
if (tmpl->a_family == AF_INET &&
tmpl->ce_mask & ADDR_ATTR_LOCAL &&
*((char *) nl_addr_get_binary_addr(tmpl->a_local)) == 127)
am.ifa_scope = RT_SCOPE_HOST;
else
am.ifa_scope = RT_SCOPE_UNIVERSE;
}
msg = nlmsg_alloc_simple(cmd, flags);
if (!msg)
return -NLE_NOMEM;
if (nlmsg_append(msg, &am, sizeof(am), NLMSG_ALIGNTO) < 0)
goto nla_put_failure;
if (tmpl->ce_mask & ADDR_ATTR_LOCAL)
NLA_PUT_ADDR(msg, IFA_LOCAL, tmpl->a_local);
if (tmpl->ce_mask & ADDR_ATTR_PEER)
NLA_PUT_ADDR(msg, IFA_ADDRESS, tmpl->a_peer);
else if (tmpl->ce_mask & ADDR_ATTR_LOCAL)
NLA_PUT_ADDR(msg, IFA_ADDRESS, tmpl->a_local);
if (tmpl->ce_mask & ADDR_ATTR_LABEL)
NLA_PUT_STRING(msg, IFA_LABEL, tmpl->a_label);
if (tmpl->ce_mask & ADDR_ATTR_BROADCAST)
NLA_PUT_ADDR(msg, IFA_BROADCAST, tmpl->a_bcast);
if (tmpl->ce_mask & ADDR_ATTR_CACHEINFO) {
struct ifa_cacheinfo ca = {
.ifa_valid = tmpl->a_cacheinfo.aci_valid,
.ifa_prefered = tmpl->a_cacheinfo.aci_prefered,
};
NLA_PUT(msg, IFA_CACHEINFO, sizeof(ca), &ca);
}
if (tmpl->a_flags & ~0xFF) {
/* only set the IFA_FLAGS attribute, if they actually contain additional
* flags that are not already set to am.ifa_flags.
*
* Older kernels refuse RTM_NEWADDR and RTM_NEWROUTE messages with EINVAL
* if they contain unknown netlink attributes. See net/core/rtnetlink.c, which
* was fixed by kernel commit 661d2967b3f1b34eeaa7e212e7b9bbe8ee072b59.
*
* With this workaround, libnl will function correctly with older kernels,
* unless there is a new libnl user that wants to set these flags. In this
* case it's up to the user to workaround this issue. */
NLA_PUT_U32(msg, IFA_FLAGS, tmpl->a_flags);
}
*result = msg;
return 0;
nla_put_failure:
nlmsg_free(msg);
return -NLE_MSGSIZE;
}
/**
* @name Addition
* @{
*/
/**
* Build netlink request message to request addition of new address
* @arg addr Address object representing the new address.
* @arg flags Additional netlink message flags.
* @arg result Pointer to store resulting message.
*
* Builds a new netlink message requesting the addition of a new
* address. The netlink message header isn't fully equipped with
* all relevant fields and must thus be sent out via nl_send_auto_complete()
* or supplemented as needed.
*
* Minimal required attributes:
* - interface index (rtnl_addr_set_ifindex())
* - local address (rtnl_addr_set_local())
*
* The scope will default to universe except for loopback addresses in
* which case a host scope is used if not specified otherwise.
*
* @note Free the memory after usage using nlmsg_free().
*
* @return 0 on success or a negative error code.
*/
int rtnl_addr_build_add_request(struct rtnl_addr *addr, int flags,
struct nl_msg **result)
{
uint32_t required = ADDR_ATTR_IFINDEX | ADDR_ATTR_FAMILY |
ADDR_ATTR_PREFIXLEN | ADDR_ATTR_LOCAL;
if ((addr->ce_mask & required) != required)
return -NLE_MISSING_ATTR;
return build_addr_msg(addr, RTM_NEWADDR, NLM_F_CREATE | flags, result);
}
/**
* Request addition of new address
* @arg sk Netlink socket.
* @arg addr Address object representing the new address.
* @arg flags Additional netlink message flags.
*
* Builds a netlink message by calling rtnl_addr_build_add_request(),
* sends the request to the kernel and waits for the next ACK to be
* received and thus blocks until the request has been fullfilled.
*
* @see rtnl_addr_build_add_request()
*
* @return 0 on sucess or a negative error if an error occured.
*/
int rtnl_addr_add(struct nl_sock *sk, struct rtnl_addr *addr, int flags)
{
struct nl_msg *msg;
int err;
if ((err = rtnl_addr_build_add_request(addr, flags, &msg)) < 0)
return err;
err = nl_send_auto_complete(sk, msg);
nlmsg_free(msg);
if (err < 0)
return err;
return wait_for_ack(sk);
}
/** @} */
/**
* @name Deletion
* @{
*/
/**
* Build a netlink request message to request deletion of an address
* @arg addr Address object to be deleteted.
* @arg flags Additional netlink message flags.
* @arg result Pointer to store resulting message.
*
* Builds a new netlink message requesting a deletion of an address.
* The netlink message header isn't fully equipped with all relevant
* fields and must thus be sent out via nl_send_auto_complete()
* or supplemented as needed.
*
* Minimal required attributes:
* - interface index (rtnl_addr_set_ifindex())
* - address family (rtnl_addr_set_family())
*
* Optional attributes:
* - local address (rtnl_addr_set_local())
* - label (rtnl_addr_set_label(), IPv4/DECnet only)
* - peer address (rtnl_addr_set_peer(), IPv4 only)
*
* @note Free the memory after usage using nlmsg_free().
*
* @return 0 on success or a negative error code.
*/
int rtnl_addr_build_delete_request(struct rtnl_addr *addr, int flags,
struct nl_msg **result)
{
uint32_t required = ADDR_ATTR_IFINDEX | ADDR_ATTR_FAMILY;
if ((addr->ce_mask & required) != required)
return -NLE_MISSING_ATTR;
return build_addr_msg(addr, RTM_DELADDR, flags, result);
}
/**
* Request deletion of an address
* @arg sk Netlink socket.
* @arg addr Address object to be deleted.
* @arg flags Additional netlink message flags.
*
* Builds a netlink message by calling rtnl_addr_build_delete_request(),
* sends the request to the kernel and waits for the next ACK to be
* received and thus blocks until the request has been fullfilled.
*
* @see rtnl_addr_build_delete_request();
*
* @return 0 on sucess or a negative error if an error occured.
*/
int rtnl_addr_delete(struct nl_sock *sk, struct rtnl_addr *addr, int flags)
{
struct nl_msg *msg;
int err;
if ((err = rtnl_addr_build_delete_request(addr, flags, &msg)) < 0)
return err;
err = nl_send_auto_complete(sk, msg);
nlmsg_free(msg);
if (err < 0)
return err;
return wait_for_ack(sk);
}
/** @} */
/**
* @name Attributes
* @{
*/
int rtnl_addr_set_label(struct rtnl_addr *addr, const char *label)
{
if (strlen(label) > sizeof(addr->a_label) - 1)
return -NLE_RANGE;
strcpy(addr->a_label, label);
addr->ce_mask |= ADDR_ATTR_LABEL;
return 0;
}
char *rtnl_addr_get_label(struct rtnl_addr *addr)
{
if (addr->ce_mask & ADDR_ATTR_LABEL)
return addr->a_label;
else
return NULL;
}
void rtnl_addr_set_ifindex(struct rtnl_addr *addr, int ifindex)
{
addr->a_ifindex = ifindex;
addr->ce_mask |= ADDR_ATTR_IFINDEX;
}
int rtnl_addr_get_ifindex(struct rtnl_addr *addr)
{
return addr->a_ifindex;
}
void rtnl_addr_set_link(struct rtnl_addr *addr, struct rtnl_link *link)
{
rtnl_link_put(addr->a_link);
if (!link)
return;
nl_object_get(OBJ_CAST(link));
addr->a_link = link;
addr->a_ifindex = link->l_index;
addr->ce_mask |= ADDR_ATTR_IFINDEX;
}
struct rtnl_link *rtnl_addr_get_link(struct rtnl_addr *addr)
{
if (addr->a_link) {
nl_object_get(OBJ_CAST(addr->a_link));
return addr->a_link;
}
return NULL;
}
void rtnl_addr_set_family(struct rtnl_addr *addr, int family)
{
addr->a_family = family;
addr->ce_mask |= ADDR_ATTR_FAMILY;
}
int rtnl_addr_get_family(struct rtnl_addr *addr)
{
return addr->a_family;
}
/**
* Set the prefix length / netmask
* @arg addr Address
* @arg prefixlen Length of prefix (netmask)
*
* Modifies the length of the prefix. If the address object contains a peer
* address the prefix length will apply to it, otherwise the prefix length
* will apply to the local address of the address.
*
* If the address object contains a peer or local address the corresponding
* `struct nl_addr` will be updated with the new prefix length.
*
* @note Specifying a length of 0 will remove the prefix length alltogether.
*
* @see rtnl_addr_get_prefixlen()
*/
void rtnl_addr_set_prefixlen(struct rtnl_addr *addr, int prefixlen)
{
addr->a_prefixlen = prefixlen;
if (prefixlen)
addr->ce_mask |= ADDR_ATTR_PREFIXLEN;
else
addr->ce_mask &= ~ADDR_ATTR_PREFIXLEN;
/*
* The prefix length always applies to the peer address if
* a peer address is present.
*/
if (addr->a_peer)
nl_addr_set_prefixlen(addr->a_peer, prefixlen);
else if (addr->a_local)
nl_addr_set_prefixlen(addr->a_local, prefixlen);
}
int rtnl_addr_get_prefixlen(struct rtnl_addr *addr)
{
return addr->a_prefixlen;
}
void rtnl_addr_set_scope(struct rtnl_addr *addr, int scope)
{
addr->a_scope = scope;
addr->ce_mask |= ADDR_ATTR_SCOPE;
}
int rtnl_addr_get_scope(struct rtnl_addr *addr)
{
return addr->a_scope;
}
void rtnl_addr_set_flags(struct rtnl_addr *addr, unsigned int flags)
{
addr->a_flag_mask |= flags;
addr->a_flags |= flags;
addr->ce_mask |= ADDR_ATTR_FLAGS;
}
void rtnl_addr_unset_flags(struct rtnl_addr *addr, unsigned int flags)
{
addr->a_flag_mask |= flags;
addr->a_flags &= ~flags;
addr->ce_mask |= ADDR_ATTR_FLAGS;
}
unsigned int rtnl_addr_get_flags(struct rtnl_addr *addr)
{
return addr->a_flags;
}
static inline int __assign_addr(struct rtnl_addr *addr, struct nl_addr **pos,
struct nl_addr *new, int flag)
{
if (new) {
if (addr->ce_mask & ADDR_ATTR_FAMILY) {
if (new->a_family != addr->a_family)
return -NLE_AF_MISMATCH;
} else
addr->a_family = new->a_family;
if (*pos)
nl_addr_put(*pos);
*pos = nl_addr_get(new);
addr->ce_mask |= (flag | ADDR_ATTR_FAMILY);
} else {
if (*pos)
nl_addr_put(*pos);
*pos = NULL;
addr->ce_mask &= ~flag;
}
return 0;
}
int rtnl_addr_set_local(struct rtnl_addr *addr, struct nl_addr *local)
{
int err;
/* Prohibit local address with prefix length if peer address is present */
if ((addr->ce_mask & ADDR_ATTR_PEER) && local &&
nl_addr_get_prefixlen(local))
return -NLE_INVAL;
err = __assign_addr(addr, &addr->a_local, local, ADDR_ATTR_LOCAL);
if (err < 0)
return err;
/* Never overwrite the prefix length if a peer address is present */
if (!(addr->ce_mask & ADDR_ATTR_PEER))
rtnl_addr_set_prefixlen(addr, local ? nl_addr_get_prefixlen(local) : 0);
return 0;
}
struct nl_addr *rtnl_addr_get_local(struct rtnl_addr *addr)
{
return addr->a_local;
}
int rtnl_addr_set_peer(struct rtnl_addr *addr, struct nl_addr *peer)
{
int err;
if (peer && peer->a_family != AF_INET)
return -NLE_AF_NOSUPPORT;
err = __assign_addr(addr, &addr->a_peer, peer, ADDR_ATTR_PEER);
if (err < 0)
return err;
rtnl_addr_set_prefixlen(addr, peer ? nl_addr_get_prefixlen(peer) : 0);
return 0;
}
struct nl_addr *rtnl_addr_get_peer(struct rtnl_addr *addr)
{
return addr->a_peer;
}
int rtnl_addr_set_broadcast(struct rtnl_addr *addr, struct nl_addr *bcast)
{
if (bcast && bcast->a_family != AF_INET)
return -NLE_AF_NOSUPPORT;
return __assign_addr(addr, &addr->a_bcast, bcast, ADDR_ATTR_BROADCAST);
}
struct nl_addr *rtnl_addr_get_broadcast(struct rtnl_addr *addr)
{
return addr->a_bcast;
}
int rtnl_addr_set_multicast(struct rtnl_addr *addr, struct nl_addr *multicast)
{
if (multicast && multicast->a_family != AF_INET6)
return -NLE_AF_NOSUPPORT;
return __assign_addr(addr, &addr->a_multicast, multicast,
ADDR_ATTR_MULTICAST);
}
struct nl_addr *rtnl_addr_get_multicast(struct rtnl_addr *addr)
{
return addr->a_multicast;
}
int rtnl_addr_set_anycast(struct rtnl_addr *addr, struct nl_addr *anycast)
{
if (anycast && anycast->a_family != AF_INET6)
return -NLE_AF_NOSUPPORT;
return __assign_addr(addr, &addr->a_anycast, anycast,
ADDR_ATTR_ANYCAST);
}
struct nl_addr *rtnl_addr_get_anycast(struct rtnl_addr *addr)
{
return addr->a_anycast;
}
uint32_t rtnl_addr_get_valid_lifetime(struct rtnl_addr *addr)
{
if (addr->ce_mask & ADDR_ATTR_CACHEINFO)
return addr->a_cacheinfo.aci_valid;
else
return 0xFFFFFFFFU;
}
void rtnl_addr_set_valid_lifetime(struct rtnl_addr *addr, uint32_t lifetime)
{
addr->a_cacheinfo.aci_valid = lifetime;
addr->ce_mask |= ADDR_ATTR_CACHEINFO;
}
uint32_t rtnl_addr_get_preferred_lifetime(struct rtnl_addr *addr)
{
if (addr->ce_mask & ADDR_ATTR_CACHEINFO)
return addr->a_cacheinfo.aci_prefered;
else
return 0xFFFFFFFFU;
}
void rtnl_addr_set_preferred_lifetime(struct rtnl_addr *addr, uint32_t lifetime)
{
addr->a_cacheinfo.aci_prefered = lifetime;
addr->ce_mask |= ADDR_ATTR_CACHEINFO;
}
uint32_t rtnl_addr_get_create_time(struct rtnl_addr *addr)
{
return addr->a_cacheinfo.aci_cstamp;
}
uint32_t rtnl_addr_get_last_update_time(struct rtnl_addr *addr)
{
return addr->a_cacheinfo.aci_tstamp;
}
/** @} */
/**
* @name Flags Translations
* @{
*/
static const struct trans_tbl addr_flags[] = {
__ADD(IFA_F_SECONDARY, secondary)
__ADD(IFA_F_NODAD, nodad)
__ADD(IFA_F_OPTIMISTIC, optimistic)
__ADD(IFA_F_HOMEADDRESS, homeaddress)
__ADD(IFA_F_DEPRECATED, deprecated)
__ADD(IFA_F_TENTATIVE, tentative)
__ADD(IFA_F_PERMANENT, permanent)
__ADD(IFA_F_MANAGETEMPADDR, mngtmpaddr)
__ADD(IFA_F_NOPREFIXROUTE, noprefixroute)
};
char *rtnl_addr_flags2str(int flags, char *buf, size_t size)
{
return __flags2str(flags, buf, size, addr_flags,
ARRAY_SIZE(addr_flags));
}
int rtnl_addr_str2flags(const char *name)
{
return __str2flags(name, addr_flags, ARRAY_SIZE(addr_flags));
}
/** @} */
static struct nl_object_ops addr_obj_ops = {
.oo_name = "route/addr",
.oo_size = sizeof(struct rtnl_addr),
.oo_constructor = addr_constructor,
.oo_free_data = addr_free_data,
.oo_clone = addr_clone,
.oo_dump = {
[NL_DUMP_LINE] = addr_dump_line,
[NL_DUMP_DETAILS] = addr_dump_details,
[NL_DUMP_STATS] = addr_dump_stats,
},
.oo_compare = addr_compare,
.oo_attrs2str = addr_attrs2str,
.oo_id_attrs = (ADDR_ATTR_FAMILY | ADDR_ATTR_IFINDEX |
ADDR_ATTR_LOCAL | ADDR_ATTR_PREFIXLEN),
};
static struct nl_af_group addr_groups[] = {
{ AF_INET, RTNLGRP_IPV4_IFADDR },
{ AF_INET6, RTNLGRP_IPV6_IFADDR },
{ END_OF_GROUP_LIST },
};
static struct nl_cache_ops rtnl_addr_ops = {
.co_name = "route/addr",
.co_hdrsize = sizeof(struct ifaddrmsg),
.co_msgtypes = {
{ RTM_NEWADDR, NL_ACT_NEW, "new" },
{ RTM_DELADDR, NL_ACT_DEL, "del" },
{ RTM_GETADDR, NL_ACT_GET, "get" },
END_OF_MSGTYPES_LIST,
},
.co_protocol = NETLINK_ROUTE,
.co_groups = addr_groups,
.co_request_update = addr_request_update,
.co_msg_parser = addr_msg_parser,
.co_obj_ops = &addr_obj_ops,
};
static void __init addr_init(void)
{
nl_cache_mngt_register(&rtnl_addr_ops);
}
static void __exit addr_exit(void)
{
nl_cache_mngt_unregister(&rtnl_addr_ops);
}
/** @} */