infiniband/core/netlink.c

0001 /*
0002  * Copyright (c) 2017 Mellanox Technologies Inc.  All rights reserved.
0003  * Copyright (c) 2010 Voltaire Inc.  All rights reserved.
0004  *
0005  * This software is available to you under a choice of one of two
0006  * licenses.  You may choose to be licensed under the terms of the GNU
0007  * General Public License (GPL) Version 2, available from the file
0008  * COPYING in the main directory of this source tree, or the
0009  * OpenIB.org BSD license below:
0010  *
0011  *     Redistribution and use in source and binary forms, with or
0012  *     without modification, are permitted provided that the following
0013  *     conditions are met:
0014  *
0015  *      - Redistributions of source code must retain the above
0016  *        copyright notice, this list of conditions and the following
0017  *        disclaimer.
0018  *
0019  *      - Redistributions in binary form must reproduce the above
0020  *        copyright notice, this list of conditions and the following
0021  *        disclaimer in the documentation and/or other materials
0022  *        provided with the distribution.
0023  *
0024  * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND,
0025  * EXPRESS OR IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF
0026  * MERCHANTABILITY, FITNESS FOR A PARTICULAR PURPOSE AND
0027  * NONINFRINGEMENT. IN NO EVENT SHALL THE AUTHORS OR COPYRIGHT HOLDERS
0028  * BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER LIABILITY, WHETHER IN AN
0029  * ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM, OUT OF OR IN
0030  * CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE
0031  * SOFTWARE.
0032  */
0033
0034 #define pr_fmt(fmt) "%s:%s: " fmt, KBUILD_MODNAME, __func__
0035
0036 #include <linux/export.h>
0037 #include <net/netlink.h>
0038 #include <net/net_namespace.h>
0039 #include <net/netns/generic.h>
0040 #include <net/sock.h>
0041 #include <rdma/rdma_netlink.h>
0042 #include <linux/module.h>
0043 #include "core_priv.h"
0044
0045 static struct {
0046     const struct rdma_nl_cbs *cb_table;
0047     /* Synchronizes between ongoing netlink commands and netlink client
0048      * unregistration.
0049      */
0050     struct rw_semaphore sem;
0051 } rdma_nl_types[RDMA_NL_NUM_CLIENTS];
0052
0053 bool rdma_nl_chk_listeners(unsigned int group)
0054 {
0055     struct rdma_dev_net *rnet = rdma_net_to_dev_net(&init_net);
0056
0057     return netlink_has_listeners(rnet->nl_sock, group);
0058 }
0059 EXPORT_SYMBOL(rdma_nl_chk_listeners);
0060
0061 static bool is_nl_msg_valid(unsigned int type, unsigned int op)
0062 {
0063     static const unsigned int max_num_ops[RDMA_NL_NUM_CLIENTS] = {
0064         [RDMA_NL_IWCM] = RDMA_NL_IWPM_NUM_OPS,
0065         [RDMA_NL_LS] = RDMA_NL_LS_NUM_OPS,
0066         [RDMA_NL_NLDEV] = RDMA_NLDEV_NUM_OPS,
0067     };
0068
0069     /*
0070      * This BUILD_BUG_ON is intended to catch addition of new
0071      * RDMA netlink protocol without updating the array above.
0072      */
0073     BUILD_BUG_ON(RDMA_NL_NUM_CLIENTS != 6);
0074
0075     if (type >= RDMA_NL_NUM_CLIENTS)
0076         return false;
0077
0078     return (op < max_num_ops[type]) ? true : false;
0079 }
0080
0081 static const struct rdma_nl_cbs *
0082 get_cb_table(const struct sk_buff *skb, unsigned int type, unsigned int op)
0083 {
0084     const struct rdma_nl_cbs *cb_table;
0085
0086     /*
0087      * Currently only NLDEV client is supporting netlink commands in
0088      * non init_net net namespace.
0089      */
0090     if (sock_net(skb->sk) != &init_net && type != RDMA_NL_NLDEV)
0091         return NULL;
0092
0093     cb_table = READ_ONCE(rdma_nl_types[type].cb_table);
0094     if (!cb_table) {
0095         /*
0096          * Didn't get valid reference of the table, attempt module
0097          * load once.
0098          */
0099         up_read(&rdma_nl_types[type].sem);
0100
0101         request_module("rdma-netlink-subsys-%u", type);
0102
0103         down_read(&rdma_nl_types[type].sem);
0104         cb_table = READ_ONCE(rdma_nl_types[type].cb_table);
0105     }
0106     if (!cb_table || (!cb_table[op].dump && !cb_table[op].doit))
0107         return NULL;
0108     return cb_table;
0109 }
0110
0111 void rdma_nl_register(unsigned int index,
0112               const struct rdma_nl_cbs cb_table[])
0113 {
0114     if (WARN_ON(!is_nl_msg_valid(index, 0)) ||
0115         WARN_ON(READ_ONCE(rdma_nl_types[index].cb_table)))
0116         return;
0117
0118     /* Pairs with the READ_ONCE in is_nl_valid() */
0119     smp_store_release(&rdma_nl_types[index].cb_table, cb_table);
0120 }
0121 EXPORT_SYMBOL(rdma_nl_register);
0122
0123 void rdma_nl_unregister(unsigned int index)
0124 {
0125     down_write(&rdma_nl_types[index].sem);
0126     rdma_nl_types[index].cb_table = NULL;
0127     up_write(&rdma_nl_types[index].sem);
0128 }
0129 EXPORT_SYMBOL(rdma_nl_unregister);
0130
0131 void *ibnl_put_msg(struct sk_buff *skb, struct nlmsghdr **nlh, int seq,
0132            int len, int client, int op, int flags)
0133 {
0134     *nlh = nlmsg_put(skb, 0, seq, RDMA_NL_GET_TYPE(client, op), len, flags);
0135     if (!*nlh)
0136         return NULL;
0137     return nlmsg_data(*nlh);
0138 }
0139 EXPORT_SYMBOL(ibnl_put_msg);
0140
0141 int ibnl_put_attr(struct sk_buff *skb, struct nlmsghdr *nlh,
0142           int len, void *data, int type)
0143 {
0144     if (nla_put(skb, type, len, data)) {
0145         nlmsg_cancel(skb, nlh);
0146         return -EMSGSIZE;
0147     }
0148     return 0;
0149 }
0150 EXPORT_SYMBOL(ibnl_put_attr);
0151
0152 static int rdma_nl_rcv_msg(struct sk_buff *skb, struct nlmsghdr *nlh,
0153                struct netlink_ext_ack *extack)
0154 {
0155     int type = nlh->nlmsg_type;
0156     unsigned int index = RDMA_NL_GET_CLIENT(type);
0157     unsigned int op = RDMA_NL_GET_OP(type);
0158     const struct rdma_nl_cbs *cb_table;
0159     int err = -EINVAL;
0160
0161     if (!is_nl_msg_valid(index, op))
0162         return -EINVAL;
0163
0164     down_read(&rdma_nl_types[index].sem);
0165     cb_table = get_cb_table(skb, index, op);
0166     if (!cb_table)
0167         goto done;
0168
0169     if ((cb_table[op].flags & RDMA_NL_ADMIN_PERM) &&
0170         !netlink_capable(skb, CAP_NET_ADMIN)) {
0171         err = -EPERM;
0172         goto done;
0173     }
0174
0175     /*
0176      * LS responses overload the 0x100 (NLM_F_ROOT) flag.  Don't
0177      * mistakenly call the .dump() function.
0178      */
0179     if (index == RDMA_NL_LS) {
0180         if (cb_table[op].doit)
0181             err = cb_table[op].doit(skb, nlh, extack);
0182         goto done;
0183     }
0184     /* FIXME: Convert IWCM to properly handle doit callbacks */
0185     if ((nlh->nlmsg_flags & NLM_F_DUMP) || index == RDMA_NL_IWCM) {
0186         struct netlink_dump_control c = {
0187             .dump = cb_table[op].dump,
0188         };
0189         if (c.dump)
0190             err = netlink_dump_start(skb->sk, skb, nlh, &c);
0191         goto done;
0192     }
0193
0194     if (cb_table[op].doit)
0195         err = cb_table[op].doit(skb, nlh, extack);
0196 done:
0197     up_read(&rdma_nl_types[index].sem);
0198     return err;
0199 }
0200
0201 /*
0202  * This function is similar to netlink_rcv_skb with one exception:
0203  * It calls to the callback for the netlink messages without NLM_F_REQUEST
0204  * flag. These messages are intended for RDMA_NL_LS consumer, so it is allowed
0205  * for that consumer only.
0206  */
0207 static int rdma_nl_rcv_skb(struct sk_buff *skb, int (*cb)(struct sk_buff *,
0208                            struct nlmsghdr *,
0209                            struct netlink_ext_ack *))
0210 {
0211     struct netlink_ext_ack extack = {};
0212     struct nlmsghdr *nlh;
0213     int err;
0214
0215     while (skb->len >= nlmsg_total_size(0)) {
0216         int msglen;
0217
0218         nlh = nlmsg_hdr(skb);
0219         err = 0;
0220
0221         if (nlh->nlmsg_len < NLMSG_HDRLEN || skb->len < nlh->nlmsg_len)
0222             return 0;
0223
0224         /*
0225          * Generally speaking, the only requests are handled
0226          * by the kernel, but RDMA_NL_LS is different, because it
0227          * runs backward netlink scheme. Kernel initiates messages
0228          * and waits for reply with data to keep pathrecord cache
0229          * in sync.
0230          */
0231         if (!(nlh->nlmsg_flags & NLM_F_REQUEST) &&
0232             (RDMA_NL_GET_CLIENT(nlh->nlmsg_type) != RDMA_NL_LS))
0233             goto ack;
0234
0235         /* Skip control messages */
0236         if (nlh->nlmsg_type < NLMSG_MIN_TYPE)
0237             goto ack;
0238
0239         err = cb(skb, nlh, &extack);
0240         if (err == -EINTR)
0241             goto skip;
0242
0243 ack:
0244         if (nlh->nlmsg_flags & NLM_F_ACK || err)
0245             netlink_ack(skb, nlh, err, &extack);
0246
0247 skip:
0248         msglen = NLMSG_ALIGN(nlh->nlmsg_len);
0249         if (msglen > skb->len)
0250             msglen = skb->len;
0251         skb_pull(skb, msglen);
0252     }
0253
0254     return 0;
0255 }
0256
0257 static void rdma_nl_rcv(struct sk_buff *skb)
0258 {
0259     rdma_nl_rcv_skb(skb, &rdma_nl_rcv_msg);
0260 }
0261
0262 int rdma_nl_unicast(struct net *net, struct sk_buff *skb, u32 pid)
0263 {
0264     struct rdma_dev_net *rnet = rdma_net_to_dev_net(net);
0265     int err;
0266
0267     err = netlink_unicast(rnet->nl_sock, skb, pid, MSG_DONTWAIT);
0268     return (err < 0) ? err : 0;
0269 }
0270 EXPORT_SYMBOL(rdma_nl_unicast);
0271
0272 int rdma_nl_unicast_wait(struct net *net, struct sk_buff *skb, __u32 pid)
0273 {
0274     struct rdma_dev_net *rnet = rdma_net_to_dev_net(net);
0275     int err;
0276
0277     err = netlink_unicast(rnet->nl_sock, skb, pid, 0);
0278     return (err < 0) ? err : 0;
0279 }
0280 EXPORT_SYMBOL(rdma_nl_unicast_wait);
0281
0282 int rdma_nl_multicast(struct net *net, struct sk_buff *skb,
0283               unsigned int group, gfp_t flags)
0284 {
0285     struct rdma_dev_net *rnet = rdma_net_to_dev_net(net);
0286
0287     return nlmsg_multicast(rnet->nl_sock, skb, 0, group, flags);
0288 }
0289 EXPORT_SYMBOL(rdma_nl_multicast);
0290
0291 void rdma_nl_init(void)
0292 {
0293     int idx;
0294
0295     for (idx = 0; idx < RDMA_NL_NUM_CLIENTS; idx++)
0296         init_rwsem(&rdma_nl_types[idx].sem);
0297 }
0298
0299 void rdma_nl_exit(void)
0300 {
0301     int idx;
0302
0303     for (idx = 0; idx < RDMA_NL_NUM_CLIENTS; idx++)
0304         WARN(rdma_nl_types[idx].cb_table,
0305              "Netlink client %d wasn't released prior to unloading %s\n",
0306              idx, KBUILD_MODNAME);
0307 }
0308
0309 int rdma_nl_net_init(struct rdma_dev_net *rnet)
0310 {
0311     struct net *net = read_pnet(&rnet->net);
0312     struct netlink_kernel_cfg cfg = {
0313         .input  = rdma_nl_rcv,
0314     };
0315     struct sock *nls;
0316
0317     nls = netlink_kernel_create(net, NETLINK_RDMA, &cfg);
0318     if (!nls)
0319         return -ENOMEM;
0320
0321     nls->sk_sndtimeo = 10 * HZ;
0322     rnet->nl_sock = nls;
0323     return 0;
0324 }
0325
0326 void rdma_nl_net_exit(struct rdma_dev_net *rnet)
0327 {
0328     netlink_kernel_release(rnet->nl_sock);
0329 }
0330
0331 MODULE_ALIAS_NET_PF_PROTO(PF_NETLINK, NETLINK_RDMA);