net/rds/bind.c

0001 /*
0002  * Copyright (c) 2006, 2019 Oracle and/or its affiliates. All rights reserved.
0003  *
0004  * This software is available to you under a choice of one of two
0005  * licenses.  You may choose to be licensed under the terms of the GNU
0006  * General Public License (GPL) Version 2, available from the file
0007  * COPYING in the main directory of this source tree, or the
0008  * OpenIB.org BSD license below:
0009  *
0010  *     Redistribution and use in source and binary forms, with or
0011  *     without modification, are permitted provided that the following
0012  *     conditions are met:
0013  *
0014  *      - Redistributions of source code must retain the above
0015  *        copyright notice, this list of conditions and the following
0016  *        disclaimer.
0017  *
0018  *      - Redistributions in binary form must reproduce the above
0019  *        copyright notice, this list of conditions and the following
0020  *        disclaimer in the documentation and/or other materials
0021  *        provided with the distribution.
0022  *
0023  * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND,
0024  * EXPRESS OR IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF
0025  * MERCHANTABILITY, FITNESS FOR A PARTICULAR PURPOSE AND
0026  * NONINFRINGEMENT. IN NO EVENT SHALL THE AUTHORS OR COPYRIGHT HOLDERS
0027  * BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER LIABILITY, WHETHER IN AN
0028  * ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM, OUT OF OR IN
0029  * CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE
0030  * SOFTWARE.
0031  *
0032  */
0033 #include <linux/kernel.h>
0034 #include <net/sock.h>
0035 #include <linux/in.h>
0036 #include <linux/ipv6.h>
0037 #include <linux/if_arp.h>
0038 #include <linux/jhash.h>
0039 #include <linux/ratelimit.h>
0040 #include "rds.h"
0041
0042 static struct rhashtable bind_hash_table;
0043
0044 static const struct rhashtable_params ht_parms = {
0045     .nelem_hint = 768,
0046     .key_len = RDS_BOUND_KEY_LEN,
0047     .key_offset = offsetof(struct rds_sock, rs_bound_key),
0048     .head_offset = offsetof(struct rds_sock, rs_bound_node),
0049     .max_size = 16384,
0050     .min_size = 1024,
0051 };
0052
0053 /* Create a key for the bind hash table manipulation.  Port is in network byte
0054  * order.
0055  */
0056 static inline void __rds_create_bind_key(u8 *key, const struct in6_addr *addr,
0057                      __be16 port, __u32 scope_id)
0058 {
0059     memcpy(key, addr, sizeof(*addr));
0060     key += sizeof(*addr);
0061     memcpy(key, &port, sizeof(port));
0062     key += sizeof(port);
0063     memcpy(key, &scope_id, sizeof(scope_id));
0064 }
0065
0066 /*
0067  * Return the rds_sock bound at the given local address.
0068  *
0069  * The rx path can race with rds_release.  We notice if rds_release() has
0070  * marked this socket and don't return a rs ref to the rx path.
0071  */
0072 struct rds_sock *rds_find_bound(const struct in6_addr *addr, __be16 port,
0073                 __u32 scope_id)
0074 {
0075     u8 key[RDS_BOUND_KEY_LEN];
0076     struct rds_sock *rs;
0077
0078     __rds_create_bind_key(key, addr, port, scope_id);
0079     rcu_read_lock();
0080     rs = rhashtable_lookup(&bind_hash_table, key, ht_parms);
0081     if (rs && (sock_flag(rds_rs_to_sk(rs), SOCK_DEAD) ||
0082            !refcount_inc_not_zero(&rds_rs_to_sk(rs)->sk_refcnt)))
0083         rs = NULL;
0084
0085     rcu_read_unlock();
0086
0087     rdsdebug("returning rs %p for %pI6c:%u\n", rs, addr,
0088          ntohs(port));
0089
0090     return rs;
0091 }
0092
0093 /* returns -ve errno or +ve port */
0094 static int rds_add_bound(struct rds_sock *rs, const struct in6_addr *addr,
0095              __be16 *port, __u32 scope_id)
0096 {
0097     int ret = -EADDRINUSE;
0098     u16 rover, last;
0099     u8 key[RDS_BOUND_KEY_LEN];
0100
0101     if (*port != 0) {
0102         rover = be16_to_cpu(*port);
0103         if (rover == RDS_FLAG_PROBE_PORT)
0104             return -EINVAL;
0105         last = rover;
0106     } else {
0107         rover = max_t(u16, prandom_u32(), 2);
0108         last = rover - 1;
0109     }
0110
0111     do {
0112         if (rover == 0)
0113             rover++;
0114
0115         if (rover == RDS_FLAG_PROBE_PORT)
0116             continue;
0117         __rds_create_bind_key(key, addr, cpu_to_be16(rover),
0118                       scope_id);
0119         if (rhashtable_lookup_fast(&bind_hash_table, key, ht_parms))
0120             continue;
0121
0122         memcpy(rs->rs_bound_key, key, sizeof(rs->rs_bound_key));
0123         rs->rs_bound_addr = *addr;
0124         net_get_random_once(&rs->rs_hash_initval,
0125                     sizeof(rs->rs_hash_initval));
0126         rs->rs_bound_port = cpu_to_be16(rover);
0127         rs->rs_bound_node.next = NULL;
0128         rds_sock_addref(rs);
0129         if (!rhashtable_insert_fast(&bind_hash_table,
0130                         &rs->rs_bound_node, ht_parms)) {
0131             *port = rs->rs_bound_port;
0132             rs->rs_bound_scope_id = scope_id;
0133             ret = 0;
0134             rdsdebug("rs %p binding to %pI6c:%d\n",
0135                  rs, addr, (int)ntohs(*port));
0136             break;
0137         } else {
0138             rs->rs_bound_addr = in6addr_any;
0139             rds_sock_put(rs);
0140             ret = -ENOMEM;
0141             break;
0142         }
0143     } while (rover++ != last);
0144
0145     return ret;
0146 }
0147
0148 void rds_remove_bound(struct rds_sock *rs)
0149 {
0150
0151     if (ipv6_addr_any(&rs->rs_bound_addr))
0152         return;
0153
0154     rdsdebug("rs %p unbinding from %pI6c:%d\n",
0155          rs, &rs->rs_bound_addr,
0156          ntohs(rs->rs_bound_port));
0157
0158     rhashtable_remove_fast(&bind_hash_table, &rs->rs_bound_node, ht_parms);
0159     rds_sock_put(rs);
0160     rs->rs_bound_addr = in6addr_any;
0161 }
0162
0163 int rds_bind(struct socket *sock, struct sockaddr *uaddr, int addr_len)
0164 {
0165     struct sock *sk = sock->sk;
0166     struct rds_sock *rs = rds_sk_to_rs(sk);
0167     struct in6_addr v6addr, *binding_addr;
0168     struct rds_transport *trans;
0169     __u32 scope_id = 0;
0170     int ret = 0;
0171     __be16 port;
0172
0173     /* We allow an RDS socket to be bound to either IPv4 or IPv6
0174      * address.
0175      */
0176     if (addr_len < offsetofend(struct sockaddr, sa_family))
0177         return -EINVAL;
0178     if (uaddr->sa_family == AF_INET) {
0179         struct sockaddr_in *sin = (struct sockaddr_in *)uaddr;
0180
0181         if (addr_len < sizeof(struct sockaddr_in) ||
0182             sin->sin_addr.s_addr == htonl(INADDR_ANY) ||
0183             sin->sin_addr.s_addr == htonl(INADDR_BROADCAST) ||
0184             ipv4_is_multicast(sin->sin_addr.s_addr))
0185             return -EINVAL;
0186         ipv6_addr_set_v4mapped(sin->sin_addr.s_addr, &v6addr);
0187         binding_addr = &v6addr;
0188         port = sin->sin_port;
0189 #if IS_ENABLED(CONFIG_IPV6)
0190     } else if (uaddr->sa_family == AF_INET6) {
0191         struct sockaddr_in6 *sin6 = (struct sockaddr_in6 *)uaddr;
0192         int addr_type;
0193
0194         if (addr_len < sizeof(struct sockaddr_in6))
0195             return -EINVAL;
0196         addr_type = ipv6_addr_type(&sin6->sin6_addr);
0197         if (!(addr_type & IPV6_ADDR_UNICAST)) {
0198             __be32 addr4;
0199
0200             if (!(addr_type & IPV6_ADDR_MAPPED))
0201                 return -EINVAL;
0202
0203             /* It is a mapped address.  Need to do some sanity
0204              * checks.
0205              */
0206             addr4 = sin6->sin6_addr.s6_addr32[3];
0207             if (addr4 == htonl(INADDR_ANY) ||
0208                 addr4 == htonl(INADDR_BROADCAST) ||
0209                 ipv4_is_multicast(addr4))
0210                 return -EINVAL;
0211         }
0212         /* The scope ID must be specified for link local address. */
0213         if (addr_type & IPV6_ADDR_LINKLOCAL) {
0214             if (sin6->sin6_scope_id == 0)
0215                 return -EINVAL;
0216             scope_id = sin6->sin6_scope_id;
0217         }
0218         binding_addr = &sin6->sin6_addr;
0219         port = sin6->sin6_port;
0220 #endif
0221     } else {
0222         return -EINVAL;
0223     }
0224     lock_sock(sk);
0225
0226     /* RDS socket does not allow re-binding. */
0227     if (!ipv6_addr_any(&rs->rs_bound_addr)) {
0228         ret = -EINVAL;
0229         goto out;
0230     }
0231     /* Socket is connected.  The binding address should have the same
0232      * scope ID as the connected address, except the case when one is
0233      * non-link local address (scope_id is 0).
0234      */
0235     if (!ipv6_addr_any(&rs->rs_conn_addr) && scope_id &&
0236         rs->rs_bound_scope_id &&
0237         scope_id != rs->rs_bound_scope_id) {
0238         ret = -EINVAL;
0239         goto out;
0240     }
0241
0242     /* The transport can be set using SO_RDS_TRANSPORT option before the
0243      * socket is bound.
0244      */
0245     if (rs->rs_transport) {
0246         trans = rs->rs_transport;
0247         if (!trans->laddr_check ||
0248             trans->laddr_check(sock_net(sock->sk),
0249                        binding_addr, scope_id) != 0) {
0250             ret = -ENOPROTOOPT;
0251             goto out;
0252         }
0253     } else {
0254         trans = rds_trans_get_preferred(sock_net(sock->sk),
0255                         binding_addr, scope_id);
0256         if (!trans) {
0257             ret = -EADDRNOTAVAIL;
0258             pr_info_ratelimited("RDS: %s could not find a transport for %pI6c, load rds_tcp or rds_rdma?\n",
0259                         __func__, binding_addr);
0260             goto out;
0261         }
0262         rs->rs_transport = trans;
0263     }
0264
0265     sock_set_flag(sk, SOCK_RCU_FREE);
0266     ret = rds_add_bound(rs, binding_addr, &port, scope_id);
0267     if (ret)
0268         rs->rs_transport = NULL;
0269
0270 out:
0271     release_sock(sk);
0272     return ret;
0273 }
0274
0275 void rds_bind_lock_destroy(void)
0276 {
0277     rhashtable_destroy(&bind_hash_table);
0278 }
0279
0280 int rds_bind_lock_init(void)
0281 {
0282     return rhashtable_init(&bind_hash_table, &ht_parms);
0283 }