Add the rt linux 4.1.3-rt3 as base
[kvmfornfv.git] / kernel / net / netlabel / netlabel_mgmt.c
diff --git a/kernel/net/netlabel/netlabel_mgmt.c b/kernel/net/netlabel/netlabel_mgmt.c
new file mode 100644 (file)
index 0000000..13f777f
--- /dev/null
@@ -0,0 +1,778 @@
+/*
+ * NetLabel Management Support
+ *
+ * This file defines the management functions for the NetLabel system.  The
+ * NetLabel system manages static and dynamic label mappings for network
+ * protocols such as CIPSO and RIPSO.
+ *
+ * Author: Paul Moore <paul@paul-moore.com>
+ *
+ */
+
+/*
+ * (c) Copyright Hewlett-Packard Development Company, L.P., 2006, 2008
+ *
+ * This program is free software;  you can redistribute it and/or modify
+ * it under the terms of the GNU General Public License as published by
+ * the Free Software Foundation; either version 2 of the License, or
+ * (at your option) any later version.
+ *
+ * This program is distributed in the hope that it will be useful,
+ * but WITHOUT ANY WARRANTY;  without even the implied warranty of
+ * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See
+ * the GNU General Public License for more details.
+ *
+ * You should have received a copy of the GNU General Public License
+ * along with this program;  if not, see <http://www.gnu.org/licenses/>.
+ *
+ */
+
+#include <linux/types.h>
+#include <linux/socket.h>
+#include <linux/string.h>
+#include <linux/skbuff.h>
+#include <linux/in.h>
+#include <linux/in6.h>
+#include <linux/slab.h>
+#include <net/sock.h>
+#include <net/netlink.h>
+#include <net/genetlink.h>
+#include <net/ip.h>
+#include <net/ipv6.h>
+#include <net/netlabel.h>
+#include <net/cipso_ipv4.h>
+#include <linux/atomic.h>
+
+#include "netlabel_domainhash.h"
+#include "netlabel_user.h"
+#include "netlabel_mgmt.h"
+
+/* NetLabel configured protocol counter */
+atomic_t netlabel_mgmt_protocount = ATOMIC_INIT(0);
+
+/* Argument struct for netlbl_domhsh_walk() */
+struct netlbl_domhsh_walk_arg {
+       struct netlink_callback *nl_cb;
+       struct sk_buff *skb;
+       u32 seq;
+};
+
+/* NetLabel Generic NETLINK CIPSOv4 family */
+static struct genl_family netlbl_mgmt_gnl_family = {
+       .id = GENL_ID_GENERATE,
+       .hdrsize = 0,
+       .name = NETLBL_NLTYPE_MGMT_NAME,
+       .version = NETLBL_PROTO_VERSION,
+       .maxattr = NLBL_MGMT_A_MAX,
+};
+
+/* NetLabel Netlink attribute policy */
+static const struct nla_policy netlbl_mgmt_genl_policy[NLBL_MGMT_A_MAX + 1] = {
+       [NLBL_MGMT_A_DOMAIN] = { .type = NLA_NUL_STRING },
+       [NLBL_MGMT_A_PROTOCOL] = { .type = NLA_U32 },
+       [NLBL_MGMT_A_VERSION] = { .type = NLA_U32 },
+       [NLBL_MGMT_A_CV4DOI] = { .type = NLA_U32 },
+};
+
+/*
+ * Helper Functions
+ */
+
+/**
+ * netlbl_mgmt_add - Handle an ADD message
+ * @info: the Generic NETLINK info block
+ * @audit_info: NetLabel audit information
+ *
+ * Description:
+ * Helper function for the ADD and ADDDEF messages to add the domain mappings
+ * from the message to the hash table.  See netlabel.h for a description of the
+ * message format.  Returns zero on success, negative values on failure.
+ *
+ */
+static int netlbl_mgmt_add_common(struct genl_info *info,
+                                 struct netlbl_audit *audit_info)
+{
+       int ret_val = -EINVAL;
+       struct netlbl_domaddr_map *addrmap = NULL;
+       struct cipso_v4_doi *cipsov4 = NULL;
+       u32 tmp_val;
+       struct netlbl_dom_map *entry = kzalloc(sizeof(*entry), GFP_KERNEL);
+
+       if (!entry)
+               return -ENOMEM;
+       entry->def.type = nla_get_u32(info->attrs[NLBL_MGMT_A_PROTOCOL]);
+       if (info->attrs[NLBL_MGMT_A_DOMAIN]) {
+               size_t tmp_size = nla_len(info->attrs[NLBL_MGMT_A_DOMAIN]);
+               entry->domain = kmalloc(tmp_size, GFP_KERNEL);
+               if (entry->domain == NULL) {
+                       ret_val = -ENOMEM;
+                       goto add_free_entry;
+               }
+               nla_strlcpy(entry->domain,
+                           info->attrs[NLBL_MGMT_A_DOMAIN], tmp_size);
+       }
+
+       /* NOTE: internally we allow/use a entry->def.type value of
+        *       NETLBL_NLTYPE_ADDRSELECT but we don't currently allow users
+        *       to pass that as a protocol value because we need to know the
+        *       "real" protocol */
+
+       switch (entry->def.type) {
+       case NETLBL_NLTYPE_UNLABELED:
+               break;
+       case NETLBL_NLTYPE_CIPSOV4:
+               if (!info->attrs[NLBL_MGMT_A_CV4DOI])
+                       goto add_free_domain;
+
+               tmp_val = nla_get_u32(info->attrs[NLBL_MGMT_A_CV4DOI]);
+               cipsov4 = cipso_v4_doi_getdef(tmp_val);
+               if (cipsov4 == NULL)
+                       goto add_free_domain;
+               entry->def.cipso = cipsov4;
+               break;
+       default:
+               goto add_free_domain;
+       }
+
+       if (info->attrs[NLBL_MGMT_A_IPV4ADDR]) {
+               struct in_addr *addr;
+               struct in_addr *mask;
+               struct netlbl_domaddr4_map *map;
+
+               addrmap = kzalloc(sizeof(*addrmap), GFP_KERNEL);
+               if (addrmap == NULL) {
+                       ret_val = -ENOMEM;
+                       goto add_doi_put_def;
+               }
+               INIT_LIST_HEAD(&addrmap->list4);
+               INIT_LIST_HEAD(&addrmap->list6);
+
+               if (nla_len(info->attrs[NLBL_MGMT_A_IPV4ADDR]) !=
+                   sizeof(struct in_addr)) {
+                       ret_val = -EINVAL;
+                       goto add_free_addrmap;
+               }
+               if (nla_len(info->attrs[NLBL_MGMT_A_IPV4MASK]) !=
+                   sizeof(struct in_addr)) {
+                       ret_val = -EINVAL;
+                       goto add_free_addrmap;
+               }
+               addr = nla_data(info->attrs[NLBL_MGMT_A_IPV4ADDR]);
+               mask = nla_data(info->attrs[NLBL_MGMT_A_IPV4MASK]);
+
+               map = kzalloc(sizeof(*map), GFP_KERNEL);
+               if (map == NULL) {
+                       ret_val = -ENOMEM;
+                       goto add_free_addrmap;
+               }
+               map->list.addr = addr->s_addr & mask->s_addr;
+               map->list.mask = mask->s_addr;
+               map->list.valid = 1;
+               map->def.type = entry->def.type;
+               if (cipsov4)
+                       map->def.cipso = cipsov4;
+
+               ret_val = netlbl_af4list_add(&map->list, &addrmap->list4);
+               if (ret_val != 0) {
+                       kfree(map);
+                       goto add_free_addrmap;
+               }
+
+               entry->def.type = NETLBL_NLTYPE_ADDRSELECT;
+               entry->def.addrsel = addrmap;
+#if IS_ENABLED(CONFIG_IPV6)
+       } else if (info->attrs[NLBL_MGMT_A_IPV6ADDR]) {
+               struct in6_addr *addr;
+               struct in6_addr *mask;
+               struct netlbl_domaddr6_map *map;
+
+               addrmap = kzalloc(sizeof(*addrmap), GFP_KERNEL);
+               if (addrmap == NULL) {
+                       ret_val = -ENOMEM;
+                       goto add_doi_put_def;
+               }
+               INIT_LIST_HEAD(&addrmap->list4);
+               INIT_LIST_HEAD(&addrmap->list6);
+
+               if (nla_len(info->attrs[NLBL_MGMT_A_IPV6ADDR]) !=
+                   sizeof(struct in6_addr)) {
+                       ret_val = -EINVAL;
+                       goto add_free_addrmap;
+               }
+               if (nla_len(info->attrs[NLBL_MGMT_A_IPV6MASK]) !=
+                   sizeof(struct in6_addr)) {
+                       ret_val = -EINVAL;
+                       goto add_free_addrmap;
+               }
+               addr = nla_data(info->attrs[NLBL_MGMT_A_IPV6ADDR]);
+               mask = nla_data(info->attrs[NLBL_MGMT_A_IPV6MASK]);
+
+               map = kzalloc(sizeof(*map), GFP_KERNEL);
+               if (map == NULL) {
+                       ret_val = -ENOMEM;
+                       goto add_free_addrmap;
+               }
+               map->list.addr = *addr;
+               map->list.addr.s6_addr32[0] &= mask->s6_addr32[0];
+               map->list.addr.s6_addr32[1] &= mask->s6_addr32[1];
+               map->list.addr.s6_addr32[2] &= mask->s6_addr32[2];
+               map->list.addr.s6_addr32[3] &= mask->s6_addr32[3];
+               map->list.mask = *mask;
+               map->list.valid = 1;
+               map->def.type = entry->def.type;
+
+               ret_val = netlbl_af6list_add(&map->list, &addrmap->list6);
+               if (ret_val != 0) {
+                       kfree(map);
+                       goto add_free_addrmap;
+               }
+
+               entry->def.type = NETLBL_NLTYPE_ADDRSELECT;
+               entry->def.addrsel = addrmap;
+#endif /* IPv6 */
+       }
+
+       ret_val = netlbl_domhsh_add(entry, audit_info);
+       if (ret_val != 0)
+               goto add_free_addrmap;
+
+       return 0;
+
+add_free_addrmap:
+       kfree(addrmap);
+add_doi_put_def:
+       cipso_v4_doi_putdef(cipsov4);
+add_free_domain:
+       kfree(entry->domain);
+add_free_entry:
+       kfree(entry);
+       return ret_val;
+}
+
+/**
+ * netlbl_mgmt_listentry - List a NetLabel/LSM domain map entry
+ * @skb: the NETLINK buffer
+ * @entry: the map entry
+ *
+ * Description:
+ * This function is a helper function used by the LISTALL and LISTDEF command
+ * handlers.  The caller is responsible for ensuring that the RCU read lock
+ * is held.  Returns zero on success, negative values on failure.
+ *
+ */
+static int netlbl_mgmt_listentry(struct sk_buff *skb,
+                                struct netlbl_dom_map *entry)
+{
+       int ret_val = 0;
+       struct nlattr *nla_a;
+       struct nlattr *nla_b;
+       struct netlbl_af4list *iter4;
+#if IS_ENABLED(CONFIG_IPV6)
+       struct netlbl_af6list *iter6;
+#endif
+
+       if (entry->domain != NULL) {
+               ret_val = nla_put_string(skb,
+                                        NLBL_MGMT_A_DOMAIN, entry->domain);
+               if (ret_val != 0)
+                       return ret_val;
+       }
+
+       switch (entry->def.type) {
+       case NETLBL_NLTYPE_ADDRSELECT:
+               nla_a = nla_nest_start(skb, NLBL_MGMT_A_SELECTORLIST);
+               if (nla_a == NULL)
+                       return -ENOMEM;
+
+               netlbl_af4list_foreach_rcu(iter4, &entry->def.addrsel->list4) {
+                       struct netlbl_domaddr4_map *map4;
+                       struct in_addr addr_struct;
+
+                       nla_b = nla_nest_start(skb, NLBL_MGMT_A_ADDRSELECTOR);
+                       if (nla_b == NULL)
+                               return -ENOMEM;
+
+                       addr_struct.s_addr = iter4->addr;
+                       ret_val = nla_put_in_addr(skb, NLBL_MGMT_A_IPV4ADDR,
+                                                 addr_struct.s_addr);
+                       if (ret_val != 0)
+                               return ret_val;
+                       addr_struct.s_addr = iter4->mask;
+                       ret_val = nla_put_in_addr(skb, NLBL_MGMT_A_IPV4MASK,
+                                                 addr_struct.s_addr);
+                       if (ret_val != 0)
+                               return ret_val;
+                       map4 = netlbl_domhsh_addr4_entry(iter4);
+                       ret_val = nla_put_u32(skb, NLBL_MGMT_A_PROTOCOL,
+                                             map4->def.type);
+                       if (ret_val != 0)
+                               return ret_val;
+                       switch (map4->def.type) {
+                       case NETLBL_NLTYPE_CIPSOV4:
+                               ret_val = nla_put_u32(skb, NLBL_MGMT_A_CV4DOI,
+                                                     map4->def.cipso->doi);
+                               if (ret_val != 0)
+                                       return ret_val;
+                               break;
+                       }
+
+                       nla_nest_end(skb, nla_b);
+               }
+#if IS_ENABLED(CONFIG_IPV6)
+               netlbl_af6list_foreach_rcu(iter6, &entry->def.addrsel->list6) {
+                       struct netlbl_domaddr6_map *map6;
+
+                       nla_b = nla_nest_start(skb, NLBL_MGMT_A_ADDRSELECTOR);
+                       if (nla_b == NULL)
+                               return -ENOMEM;
+
+                       ret_val = nla_put_in6_addr(skb, NLBL_MGMT_A_IPV6ADDR,
+                                                  &iter6->addr);
+                       if (ret_val != 0)
+                               return ret_val;
+                       ret_val = nla_put_in6_addr(skb, NLBL_MGMT_A_IPV6MASK,
+                                                  &iter6->mask);
+                       if (ret_val != 0)
+                               return ret_val;
+                       map6 = netlbl_domhsh_addr6_entry(iter6);
+                       ret_val = nla_put_u32(skb, NLBL_MGMT_A_PROTOCOL,
+                                             map6->def.type);
+                       if (ret_val != 0)
+                               return ret_val;
+
+                       nla_nest_end(skb, nla_b);
+               }
+#endif /* IPv6 */
+
+               nla_nest_end(skb, nla_a);
+               break;
+       case NETLBL_NLTYPE_UNLABELED:
+               ret_val = nla_put_u32(skb,NLBL_MGMT_A_PROTOCOL,entry->def.type);
+               break;
+       case NETLBL_NLTYPE_CIPSOV4:
+               ret_val = nla_put_u32(skb,NLBL_MGMT_A_PROTOCOL,entry->def.type);
+               if (ret_val != 0)
+                       return ret_val;
+               ret_val = nla_put_u32(skb, NLBL_MGMT_A_CV4DOI,
+                                     entry->def.cipso->doi);
+               break;
+       }
+
+       return ret_val;
+}
+
+/*
+ * NetLabel Command Handlers
+ */
+
+/**
+ * netlbl_mgmt_add - Handle an ADD message
+ * @skb: the NETLINK buffer
+ * @info: the Generic NETLINK info block
+ *
+ * Description:
+ * Process a user generated ADD message and add the domains from the message
+ * to the hash table.  See netlabel.h for a description of the message format.
+ * Returns zero on success, negative values on failure.
+ *
+ */
+static int netlbl_mgmt_add(struct sk_buff *skb, struct genl_info *info)
+{
+       struct netlbl_audit audit_info;
+
+       if ((!info->attrs[NLBL_MGMT_A_DOMAIN]) ||
+           (!info->attrs[NLBL_MGMT_A_PROTOCOL]) ||
+           (info->attrs[NLBL_MGMT_A_IPV4ADDR] &&
+            info->attrs[NLBL_MGMT_A_IPV6ADDR]) ||
+           (info->attrs[NLBL_MGMT_A_IPV4MASK] &&
+            info->attrs[NLBL_MGMT_A_IPV6MASK]) ||
+           ((info->attrs[NLBL_MGMT_A_IPV4ADDR] != NULL) ^
+            (info->attrs[NLBL_MGMT_A_IPV4MASK] != NULL)) ||
+           ((info->attrs[NLBL_MGMT_A_IPV6ADDR] != NULL) ^
+            (info->attrs[NLBL_MGMT_A_IPV6MASK] != NULL)))
+               return -EINVAL;
+
+       netlbl_netlink_auditinfo(skb, &audit_info);
+
+       return netlbl_mgmt_add_common(info, &audit_info);
+}
+
+/**
+ * netlbl_mgmt_remove - Handle a REMOVE message
+ * @skb: the NETLINK buffer
+ * @info: the Generic NETLINK info block
+ *
+ * Description:
+ * Process a user generated REMOVE message and remove the specified domain
+ * mappings.  Returns zero on success, negative values on failure.
+ *
+ */
+static int netlbl_mgmt_remove(struct sk_buff *skb, struct genl_info *info)
+{
+       char *domain;
+       struct netlbl_audit audit_info;
+
+       if (!info->attrs[NLBL_MGMT_A_DOMAIN])
+               return -EINVAL;
+
+       netlbl_netlink_auditinfo(skb, &audit_info);
+
+       domain = nla_data(info->attrs[NLBL_MGMT_A_DOMAIN]);
+       return netlbl_domhsh_remove(domain, &audit_info);
+}
+
+/**
+ * netlbl_mgmt_listall_cb - netlbl_domhsh_walk() callback for LISTALL
+ * @entry: the domain mapping hash table entry
+ * @arg: the netlbl_domhsh_walk_arg structure
+ *
+ * Description:
+ * This function is designed to be used as a callback to the
+ * netlbl_domhsh_walk() function for use in generating a response for a LISTALL
+ * message.  Returns the size of the message on success, negative values on
+ * failure.
+ *
+ */
+static int netlbl_mgmt_listall_cb(struct netlbl_dom_map *entry, void *arg)
+{
+       int ret_val = -ENOMEM;
+       struct netlbl_domhsh_walk_arg *cb_arg = arg;
+       void *data;
+
+       data = genlmsg_put(cb_arg->skb, NETLINK_CB(cb_arg->nl_cb->skb).portid,
+                          cb_arg->seq, &netlbl_mgmt_gnl_family,
+                          NLM_F_MULTI, NLBL_MGMT_C_LISTALL);
+       if (data == NULL)
+               goto listall_cb_failure;
+
+       ret_val = netlbl_mgmt_listentry(cb_arg->skb, entry);
+       if (ret_val != 0)
+               goto listall_cb_failure;
+
+       cb_arg->seq++;
+       genlmsg_end(cb_arg->skb, data);
+       return 0;
+
+listall_cb_failure:
+       genlmsg_cancel(cb_arg->skb, data);
+       return ret_val;
+}
+
+/**
+ * netlbl_mgmt_listall - Handle a LISTALL message
+ * @skb: the NETLINK buffer
+ * @cb: the NETLINK callback
+ *
+ * Description:
+ * Process a user generated LISTALL message and dumps the domain hash table in
+ * a form suitable for use in a kernel generated LISTALL message.  Returns zero
+ * on success, negative values on failure.
+ *
+ */
+static int netlbl_mgmt_listall(struct sk_buff *skb,
+                              struct netlink_callback *cb)
+{
+       struct netlbl_domhsh_walk_arg cb_arg;
+       u32 skip_bkt = cb->args[0];
+       u32 skip_chain = cb->args[1];
+
+       cb_arg.nl_cb = cb;
+       cb_arg.skb = skb;
+       cb_arg.seq = cb->nlh->nlmsg_seq;
+
+       netlbl_domhsh_walk(&skip_bkt,
+                          &skip_chain,
+                          netlbl_mgmt_listall_cb,
+                          &cb_arg);
+
+       cb->args[0] = skip_bkt;
+       cb->args[1] = skip_chain;
+       return skb->len;
+}
+
+/**
+ * netlbl_mgmt_adddef - Handle an ADDDEF message
+ * @skb: the NETLINK buffer
+ * @info: the Generic NETLINK info block
+ *
+ * Description:
+ * Process a user generated ADDDEF message and respond accordingly.  Returns
+ * zero on success, negative values on failure.
+ *
+ */
+static int netlbl_mgmt_adddef(struct sk_buff *skb, struct genl_info *info)
+{
+       struct netlbl_audit audit_info;
+
+       if ((!info->attrs[NLBL_MGMT_A_PROTOCOL]) ||
+           (info->attrs[NLBL_MGMT_A_IPV4ADDR] &&
+            info->attrs[NLBL_MGMT_A_IPV6ADDR]) ||
+           (info->attrs[NLBL_MGMT_A_IPV4MASK] &&
+            info->attrs[NLBL_MGMT_A_IPV6MASK]) ||
+           ((info->attrs[NLBL_MGMT_A_IPV4ADDR] != NULL) ^
+            (info->attrs[NLBL_MGMT_A_IPV4MASK] != NULL)) ||
+           ((info->attrs[NLBL_MGMT_A_IPV6ADDR] != NULL) ^
+            (info->attrs[NLBL_MGMT_A_IPV6MASK] != NULL)))
+               return -EINVAL;
+
+       netlbl_netlink_auditinfo(skb, &audit_info);
+
+       return netlbl_mgmt_add_common(info, &audit_info);
+}
+
+/**
+ * netlbl_mgmt_removedef - Handle a REMOVEDEF message
+ * @skb: the NETLINK buffer
+ * @info: the Generic NETLINK info block
+ *
+ * Description:
+ * Process a user generated REMOVEDEF message and remove the default domain
+ * mapping.  Returns zero on success, negative values on failure.
+ *
+ */
+static int netlbl_mgmt_removedef(struct sk_buff *skb, struct genl_info *info)
+{
+       struct netlbl_audit audit_info;
+
+       netlbl_netlink_auditinfo(skb, &audit_info);
+
+       return netlbl_domhsh_remove_default(&audit_info);
+}
+
+/**
+ * netlbl_mgmt_listdef - Handle a LISTDEF message
+ * @skb: the NETLINK buffer
+ * @info: the Generic NETLINK info block
+ *
+ * Description:
+ * Process a user generated LISTDEF message and dumps the default domain
+ * mapping in a form suitable for use in a kernel generated LISTDEF message.
+ * Returns zero on success, negative values on failure.
+ *
+ */
+static int netlbl_mgmt_listdef(struct sk_buff *skb, struct genl_info *info)
+{
+       int ret_val = -ENOMEM;
+       struct sk_buff *ans_skb = NULL;
+       void *data;
+       struct netlbl_dom_map *entry;
+
+       ans_skb = nlmsg_new(NLMSG_DEFAULT_SIZE, GFP_KERNEL);
+       if (ans_skb == NULL)
+               return -ENOMEM;
+       data = genlmsg_put_reply(ans_skb, info, &netlbl_mgmt_gnl_family,
+                                0, NLBL_MGMT_C_LISTDEF);
+       if (data == NULL)
+               goto listdef_failure;
+
+       rcu_read_lock();
+       entry = netlbl_domhsh_getentry(NULL);
+       if (entry == NULL) {
+               ret_val = -ENOENT;
+               goto listdef_failure_lock;
+       }
+       ret_val = netlbl_mgmt_listentry(ans_skb, entry);
+       rcu_read_unlock();
+       if (ret_val != 0)
+               goto listdef_failure;
+
+       genlmsg_end(ans_skb, data);
+       return genlmsg_reply(ans_skb, info);
+
+listdef_failure_lock:
+       rcu_read_unlock();
+listdef_failure:
+       kfree_skb(ans_skb);
+       return ret_val;
+}
+
+/**
+ * netlbl_mgmt_protocols_cb - Write an individual PROTOCOL message response
+ * @skb: the skb to write to
+ * @cb: the NETLINK callback
+ * @protocol: the NetLabel protocol to use in the message
+ *
+ * Description:
+ * This function is to be used in conjunction with netlbl_mgmt_protocols() to
+ * answer a application's PROTOCOLS message.  Returns the size of the message
+ * on success, negative values on failure.
+ *
+ */
+static int netlbl_mgmt_protocols_cb(struct sk_buff *skb,
+                                   struct netlink_callback *cb,
+                                   u32 protocol)
+{
+       int ret_val = -ENOMEM;
+       void *data;
+
+       data = genlmsg_put(skb, NETLINK_CB(cb->skb).portid, cb->nlh->nlmsg_seq,
+                          &netlbl_mgmt_gnl_family, NLM_F_MULTI,
+                          NLBL_MGMT_C_PROTOCOLS);
+       if (data == NULL)
+               goto protocols_cb_failure;
+
+       ret_val = nla_put_u32(skb, NLBL_MGMT_A_PROTOCOL, protocol);
+       if (ret_val != 0)
+               goto protocols_cb_failure;
+
+       genlmsg_end(skb, data);
+       return 0;
+
+protocols_cb_failure:
+       genlmsg_cancel(skb, data);
+       return ret_val;
+}
+
+/**
+ * netlbl_mgmt_protocols - Handle a PROTOCOLS message
+ * @skb: the NETLINK buffer
+ * @cb: the NETLINK callback
+ *
+ * Description:
+ * Process a user generated PROTOCOLS message and respond accordingly.
+ *
+ */
+static int netlbl_mgmt_protocols(struct sk_buff *skb,
+                                struct netlink_callback *cb)
+{
+       u32 protos_sent = cb->args[0];
+
+       if (protos_sent == 0) {
+               if (netlbl_mgmt_protocols_cb(skb,
+                                            cb,
+                                            NETLBL_NLTYPE_UNLABELED) < 0)
+                       goto protocols_return;
+               protos_sent++;
+       }
+       if (protos_sent == 1) {
+               if (netlbl_mgmt_protocols_cb(skb,
+                                            cb,
+                                            NETLBL_NLTYPE_CIPSOV4) < 0)
+                       goto protocols_return;
+               protos_sent++;
+       }
+
+protocols_return:
+       cb->args[0] = protos_sent;
+       return skb->len;
+}
+
+/**
+ * netlbl_mgmt_version - Handle a VERSION message
+ * @skb: the NETLINK buffer
+ * @info: the Generic NETLINK info block
+ *
+ * Description:
+ * Process a user generated VERSION message and respond accordingly.  Returns
+ * zero on success, negative values on failure.
+ *
+ */
+static int netlbl_mgmt_version(struct sk_buff *skb, struct genl_info *info)
+{
+       int ret_val = -ENOMEM;
+       struct sk_buff *ans_skb = NULL;
+       void *data;
+
+       ans_skb = nlmsg_new(NLMSG_DEFAULT_SIZE, GFP_KERNEL);
+       if (ans_skb == NULL)
+               return -ENOMEM;
+       data = genlmsg_put_reply(ans_skb, info, &netlbl_mgmt_gnl_family,
+                                0, NLBL_MGMT_C_VERSION);
+       if (data == NULL)
+               goto version_failure;
+
+       ret_val = nla_put_u32(ans_skb,
+                             NLBL_MGMT_A_VERSION,
+                             NETLBL_PROTO_VERSION);
+       if (ret_val != 0)
+               goto version_failure;
+
+       genlmsg_end(ans_skb, data);
+       return genlmsg_reply(ans_skb, info);
+
+version_failure:
+       kfree_skb(ans_skb);
+       return ret_val;
+}
+
+
+/*
+ * NetLabel Generic NETLINK Command Definitions
+ */
+
+static const struct genl_ops netlbl_mgmt_genl_ops[] = {
+       {
+       .cmd = NLBL_MGMT_C_ADD,
+       .flags = GENL_ADMIN_PERM,
+       .policy = netlbl_mgmt_genl_policy,
+       .doit = netlbl_mgmt_add,
+       .dumpit = NULL,
+       },
+       {
+       .cmd = NLBL_MGMT_C_REMOVE,
+       .flags = GENL_ADMIN_PERM,
+       .policy = netlbl_mgmt_genl_policy,
+       .doit = netlbl_mgmt_remove,
+       .dumpit = NULL,
+       },
+       {
+       .cmd = NLBL_MGMT_C_LISTALL,
+       .flags = 0,
+       .policy = netlbl_mgmt_genl_policy,
+       .doit = NULL,
+       .dumpit = netlbl_mgmt_listall,
+       },
+       {
+       .cmd = NLBL_MGMT_C_ADDDEF,
+       .flags = GENL_ADMIN_PERM,
+       .policy = netlbl_mgmt_genl_policy,
+       .doit = netlbl_mgmt_adddef,
+       .dumpit = NULL,
+       },
+       {
+       .cmd = NLBL_MGMT_C_REMOVEDEF,
+       .flags = GENL_ADMIN_PERM,
+       .policy = netlbl_mgmt_genl_policy,
+       .doit = netlbl_mgmt_removedef,
+       .dumpit = NULL,
+       },
+       {
+       .cmd = NLBL_MGMT_C_LISTDEF,
+       .flags = 0,
+       .policy = netlbl_mgmt_genl_policy,
+       .doit = netlbl_mgmt_listdef,
+       .dumpit = NULL,
+       },
+       {
+       .cmd = NLBL_MGMT_C_PROTOCOLS,
+       .flags = 0,
+       .policy = netlbl_mgmt_genl_policy,
+       .doit = NULL,
+       .dumpit = netlbl_mgmt_protocols,
+       },
+       {
+       .cmd = NLBL_MGMT_C_VERSION,
+       .flags = 0,
+       .policy = netlbl_mgmt_genl_policy,
+       .doit = netlbl_mgmt_version,
+       .dumpit = NULL,
+       },
+};
+
+/*
+ * NetLabel Generic NETLINK Protocol Functions
+ */
+
+/**
+ * netlbl_mgmt_genl_init - Register the NetLabel management component
+ *
+ * Description:
+ * Register the NetLabel management component with the Generic NETLINK
+ * mechanism.  Returns zero on success, negative values on failure.
+ *
+ */
+int __init netlbl_mgmt_genl_init(void)
+{
+       return genl_register_family_with_ops(&netlbl_mgmt_gnl_family,
+                                            netlbl_mgmt_genl_ops);
+}