1 /*
2  * Copyright (c) 2017 Mellanox Technologies Inc.  All rights reserved.
3  * Copyright (c) 2010 Voltaire Inc.  All rights reserved.
4  *
5  * This software is available to you under a choice of one of two
6  * licenses.  You may choose to be licensed under the terms of the GNU
7  * General Public License (GPL) Version 2, available from the file
8  * COPYING in the main directory of this source tree, or the
9  * OpenIB.org BSD license below:
10  *
11  *     Redistribution and use in source and binary forms, with or
12  *     without modification, are permitted provided that the following
13  *     conditions are met:
14  *
15  *      - Redistributions of source code must retain the above
16  *        copyright notice, this list of conditions and the following
17  *        disclaimer.
18  *
19  *      - Redistributions in binary form must reproduce the above
20  *        copyright notice, this list of conditions and the following
21  *        disclaimer in the documentation and/or other materials
22  *        provided with the distribution.
23  *
24  * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND,
25  * EXPRESS OR IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF
26  * MERCHANTABILITY, FITNESS FOR A PARTICULAR PURPOSE AND
27  * NONINFRINGEMENT. IN NO EVENT SHALL THE AUTHORS OR COPYRIGHT HOLDERS
28  * BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER LIABILITY, WHETHER IN AN
29  * ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM, OUT OF OR IN
30  * CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE
31  * SOFTWARE.
32  */
33 
34 #define pr_fmt(fmt) "%s:%s: " fmt, KBUILD_MODNAME, __func__
35 
36 #include <linux/export.h>
37 #include <net/netlink.h>
38 #include <net/net_namespace.h>
39 #include <net/sock.h>
40 #include <rdma/rdma_netlink.h>
41 #include <linux/module.h>
42 #include "core_priv.h"
43 
44 static DEFINE_MUTEX(rdma_nl_mutex);
45 static struct sock *nls;
46 static struct {
47 	const struct rdma_nl_cbs   *cb_table;
48 } rdma_nl_types[RDMA_NL_NUM_CLIENTS];
49 
50 bool rdma_nl_chk_listeners(unsigned int group)
51 {
52 	return netlink_has_listeners(nls, group);
53 }
54 EXPORT_SYMBOL(rdma_nl_chk_listeners);
55 
56 static bool is_nl_msg_valid(unsigned int type, unsigned int op)
57 {
58 	static const unsigned int max_num_ops[RDMA_NL_NUM_CLIENTS] = {
59 		[RDMA_NL_IWCM] = RDMA_NL_IWPM_NUM_OPS,
60 		[RDMA_NL_LS] = RDMA_NL_LS_NUM_OPS,
61 		[RDMA_NL_NLDEV] = RDMA_NLDEV_NUM_OPS,
62 	};
63 
64 	/*
65 	 * This BUILD_BUG_ON is intended to catch addition of new
66 	 * RDMA netlink protocol without updating the array above.
67 	 */
68 	BUILD_BUG_ON(RDMA_NL_NUM_CLIENTS != 6);
69 
70 	if (type >= RDMA_NL_NUM_CLIENTS)
71 		return false;
72 
73 	return (op < max_num_ops[type]) ? true : false;
74 }
75 
76 static bool is_nl_valid(unsigned int type, unsigned int op)
77 {
78 	const struct rdma_nl_cbs *cb_table;
79 
80 	if (!is_nl_msg_valid(type, op))
81 		return false;
82 
83 	if (!rdma_nl_types[type].cb_table) {
84 		mutex_unlock(&rdma_nl_mutex);
85 		request_module("rdma-netlink-subsys-%d", type);
86 		mutex_lock(&rdma_nl_mutex);
87 	}
88 
89 	cb_table = rdma_nl_types[type].cb_table;
90 
91 	if (!cb_table || (!cb_table[op].dump && !cb_table[op].doit))
92 		return false;
93 	return true;
94 }
95 
96 void rdma_nl_register(unsigned int index,
97 		      const struct rdma_nl_cbs cb_table[])
98 {
99 	mutex_lock(&rdma_nl_mutex);
100 	if (!is_nl_msg_valid(index, 0)) {
101 		/*
102 		 * All clients are not interesting in success/failure of
103 		 * this call. They want to see the print to error log and
104 		 * continue their initialization. Print warning for them,
105 		 * because it is programmer's error to be here.
106 		 */
107 		mutex_unlock(&rdma_nl_mutex);
108 		WARN(true,
109 		     "The not-valid %u index was supplied to RDMA netlink\n",
110 		     index);
111 		return;
112 	}
113 
114 	if (rdma_nl_types[index].cb_table) {
115 		mutex_unlock(&rdma_nl_mutex);
116 		WARN(true,
117 		     "The %u index is already registered in RDMA netlink\n",
118 		     index);
119 		return;
120 	}
121 
122 	rdma_nl_types[index].cb_table = cb_table;
123 	mutex_unlock(&rdma_nl_mutex);
124 }
125 EXPORT_SYMBOL(rdma_nl_register);
126 
127 void rdma_nl_unregister(unsigned int index)
128 {
129 	mutex_lock(&rdma_nl_mutex);
130 	rdma_nl_types[index].cb_table = NULL;
131 	mutex_unlock(&rdma_nl_mutex);
132 }
133 EXPORT_SYMBOL(rdma_nl_unregister);
134 
135 void *ibnl_put_msg(struct sk_buff *skb, struct nlmsghdr **nlh, int seq,
136 		   int len, int client, int op, int flags)
137 {
138 	*nlh = nlmsg_put(skb, 0, seq, RDMA_NL_GET_TYPE(client, op), len, flags);
139 	if (!*nlh)
140 		return NULL;
141 	return nlmsg_data(*nlh);
142 }
143 EXPORT_SYMBOL(ibnl_put_msg);
144 
145 int ibnl_put_attr(struct sk_buff *skb, struct nlmsghdr *nlh,
146 		  int len, void *data, int type)
147 {
148 	if (nla_put(skb, type, len, data)) {
149 		nlmsg_cancel(skb, nlh);
150 		return -EMSGSIZE;
151 	}
152 	return 0;
153 }
154 EXPORT_SYMBOL(ibnl_put_attr);
155 
156 static int rdma_nl_rcv_msg(struct sk_buff *skb, struct nlmsghdr *nlh,
157 			   struct netlink_ext_ack *extack)
158 {
159 	int type = nlh->nlmsg_type;
160 	unsigned int index = RDMA_NL_GET_CLIENT(type);
161 	unsigned int op = RDMA_NL_GET_OP(type);
162 	const struct rdma_nl_cbs *cb_table;
163 
164 	if (!is_nl_valid(index, op))
165 		return -EINVAL;
166 
167 	cb_table = rdma_nl_types[index].cb_table;
168 
169 	if ((cb_table[op].flags & RDMA_NL_ADMIN_PERM) &&
170 	    !netlink_capable(skb, CAP_NET_ADMIN))
171 		return -EPERM;
172 
173 	/*
174 	 * LS responses overload the 0x100 (NLM_F_ROOT) flag.  Don't
175 	 * mistakenly call the .dump() function.
176 	 */
177 	if (index == RDMA_NL_LS) {
178 		if (cb_table[op].doit)
179 			return cb_table[op].doit(skb, nlh, extack);
180 		return -EINVAL;
181 	}
182 	/* FIXME: Convert IWCM to properly handle doit callbacks */
183 	if ((nlh->nlmsg_flags & NLM_F_DUMP) || index == RDMA_NL_IWCM) {
184 		struct netlink_dump_control c = {
185 			.dump = cb_table[op].dump,
186 		};
187 		if (c.dump)
188 			return netlink_dump_start(nls, skb, nlh, &c);
189 		return -EINVAL;
190 	}
191 
192 	if (cb_table[op].doit)
193 		return cb_table[op].doit(skb, nlh, extack);
194 
195 	return 0;
196 }
197 
198 /*
199  * This function is similar to netlink_rcv_skb with one exception:
200  * It calls to the callback for the netlink messages without NLM_F_REQUEST
201  * flag. These messages are intended for RDMA_NL_LS consumer, so it is allowed
202  * for that consumer only.
203  */
204 static int rdma_nl_rcv_skb(struct sk_buff *skb, int (*cb)(struct sk_buff *,
205 						   struct nlmsghdr *,
206 						   struct netlink_ext_ack *))
207 {
208 	struct netlink_ext_ack extack = {};
209 	struct nlmsghdr *nlh;
210 	int err;
211 
212 	while (skb->len >= nlmsg_total_size(0)) {
213 		int msglen;
214 
215 		nlh = nlmsg_hdr(skb);
216 		err = 0;
217 
218 		if (nlh->nlmsg_len < NLMSG_HDRLEN || skb->len < nlh->nlmsg_len)
219 			return 0;
220 
221 		/*
222 		 * Generally speaking, the only requests are handled
223 		 * by the kernel, but RDMA_NL_LS is different, because it
224 		 * runs backward netlink scheme. Kernel initiates messages
225 		 * and waits for reply with data to keep pathrecord cache
226 		 * in sync.
227 		 */
228 		if (!(nlh->nlmsg_flags & NLM_F_REQUEST) &&
229 		    (RDMA_NL_GET_CLIENT(nlh->nlmsg_type) != RDMA_NL_LS))
230 			goto ack;
231 
232 		/* Skip control messages */
233 		if (nlh->nlmsg_type < NLMSG_MIN_TYPE)
234 			goto ack;
235 
236 		err = cb(skb, nlh, &extack);
237 		if (err == -EINTR)
238 			goto skip;
239 
240 ack:
241 		if (nlh->nlmsg_flags & NLM_F_ACK || err)
242 			netlink_ack(skb, nlh, err, &extack);
243 
244 skip:
245 		msglen = NLMSG_ALIGN(nlh->nlmsg_len);
246 		if (msglen > skb->len)
247 			msglen = skb->len;
248 		skb_pull(skb, msglen);
249 	}
250 
251 	return 0;
252 }
253 
254 static void rdma_nl_rcv(struct sk_buff *skb)
255 {
256 	mutex_lock(&rdma_nl_mutex);
257 	rdma_nl_rcv_skb(skb, &rdma_nl_rcv_msg);
258 	mutex_unlock(&rdma_nl_mutex);
259 }
260 
261 int rdma_nl_unicast(struct sk_buff *skb, u32 pid)
262 {
263 	int err;
264 
265 	err = netlink_unicast(nls, skb, pid, MSG_DONTWAIT);
266 	return (err < 0) ? err : 0;
267 }
268 EXPORT_SYMBOL(rdma_nl_unicast);
269 
270 int rdma_nl_unicast_wait(struct sk_buff *skb, __u32 pid)
271 {
272 	int err;
273 
274 	err = netlink_unicast(nls, skb, pid, 0);
275 	return (err < 0) ? err : 0;
276 }
277 EXPORT_SYMBOL(rdma_nl_unicast_wait);
278 
279 int rdma_nl_multicast(struct sk_buff *skb, unsigned int group, gfp_t flags)
280 {
281 	return nlmsg_multicast(nls, skb, 0, group, flags);
282 }
283 EXPORT_SYMBOL(rdma_nl_multicast);
284 
285 int __init rdma_nl_init(void)
286 {
287 	struct netlink_kernel_cfg cfg = {
288 		.input	= rdma_nl_rcv,
289 	};
290 
291 	nls = netlink_kernel_create(&init_net, NETLINK_RDMA, &cfg);
292 	if (!nls)
293 		return -ENOMEM;
294 
295 	nls->sk_sndtimeo = 10 * HZ;
296 	return 0;
297 }
298 
299 void rdma_nl_exit(void)
300 {
301 	int idx;
302 
303 	for (idx = 0; idx < RDMA_NL_NUM_CLIENTS; idx++)
304 		rdma_nl_unregister(idx);
305 
306 	netlink_kernel_release(nls);
307 }
308 
309 MODULE_ALIAS_NET_PF_PROTO(PF_NETLINK, NETLINK_RDMA);
310