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 #include "core_priv.h"
45 
46 static DEFINE_MUTEX(rdma_nl_mutex);
47 static struct sock *nls;
48 static struct {
49 	const struct rdma_nl_cbs   *cb_table;
50 } rdma_nl_types[RDMA_NL_NUM_CLIENTS];
51 
52 int rdma_nl_chk_listeners(unsigned int group)
53 {
54 	return (netlink_has_listeners(nls, group)) ? 0 : -1;
55 }
56 EXPORT_SYMBOL(rdma_nl_chk_listeners);
57 
58 static bool is_nl_msg_valid(unsigned int type, unsigned int op)
59 {
60 	static const unsigned int max_num_ops[RDMA_NL_NUM_CLIENTS] = {
61 		[RDMA_NL_RDMA_CM] = RDMA_NL_RDMA_CM_NUM_OPS,
62 		[RDMA_NL_IWCM] = RDMA_NL_IWPM_NUM_OPS,
63 		[RDMA_NL_LS] = RDMA_NL_LS_NUM_OPS,
64 		[RDMA_NL_NLDEV] = RDMA_NLDEV_NUM_OPS,
65 	};
66 
67 	/*
68 	 * This BUILD_BUG_ON is intended to catch addition of new
69 	 * RDMA netlink protocol without updating the array above.
70 	 */
71 	BUILD_BUG_ON(RDMA_NL_NUM_CLIENTS != 6);
72 
73 	if (type >= RDMA_NL_NUM_CLIENTS)
74 		return false;
75 
76 	return (op < max_num_ops[type]) ? true : false;
77 }
78 
79 static bool is_nl_valid(unsigned int type, unsigned int op)
80 {
81 	const struct rdma_nl_cbs *cb_table;
82 
83 	if (!is_nl_msg_valid(type, op))
84 		return false;
85 
86 	cb_table = rdma_nl_types[type].cb_table;
87 #ifdef CONFIG_MODULES
88 	if (!cb_table) {
89 		mutex_unlock(&rdma_nl_mutex);
90 		request_module("rdma-netlink-subsys-%d", type);
91 		mutex_lock(&rdma_nl_mutex);
92 		cb_table = rdma_nl_types[type].cb_table;
93 	}
94 #endif
95 
96 	if (!cb_table || (!cb_table[op].dump && !cb_table[op].doit))
97 		return false;
98 	return true;
99 }
100 
101 void rdma_nl_register(unsigned int index,
102 		      const struct rdma_nl_cbs cb_table[])
103 {
104 	mutex_lock(&rdma_nl_mutex);
105 	if (!is_nl_msg_valid(index, 0)) {
106 		/*
107 		 * All clients are not interesting in success/failure of
108 		 * this call. They want to see the print to error log and
109 		 * continue their initialization. Print warning for them,
110 		 * because it is programmer's error to be here.
111 		 */
112 		mutex_unlock(&rdma_nl_mutex);
113 		WARN(true,
114 		     "The not-valid %u index was supplied to RDMA netlink\n",
115 		     index);
116 		return;
117 	}
118 
119 	if (rdma_nl_types[index].cb_table) {
120 		mutex_unlock(&rdma_nl_mutex);
121 		WARN(true,
122 		     "The %u index is already registered in RDMA netlink\n",
123 		     index);
124 		return;
125 	}
126 
127 	rdma_nl_types[index].cb_table = cb_table;
128 	mutex_unlock(&rdma_nl_mutex);
129 }
130 EXPORT_SYMBOL(rdma_nl_register);
131 
132 void rdma_nl_unregister(unsigned int index)
133 {
134 	mutex_lock(&rdma_nl_mutex);
135 	rdma_nl_types[index].cb_table = NULL;
136 	mutex_unlock(&rdma_nl_mutex);
137 }
138 EXPORT_SYMBOL(rdma_nl_unregister);
139 
140 void *ibnl_put_msg(struct sk_buff *skb, struct nlmsghdr **nlh, int seq,
141 		   int len, int client, int op, int flags)
142 {
143 	*nlh = nlmsg_put(skb, 0, seq, RDMA_NL_GET_TYPE(client, op), len, flags);
144 	if (!*nlh)
145 		return NULL;
146 	return nlmsg_data(*nlh);
147 }
148 EXPORT_SYMBOL(ibnl_put_msg);
149 
150 int ibnl_put_attr(struct sk_buff *skb, struct nlmsghdr *nlh,
151 		  int len, void *data, int type)
152 {
153 	if (nla_put(skb, type, len, data)) {
154 		nlmsg_cancel(skb, nlh);
155 		return -EMSGSIZE;
156 	}
157 	return 0;
158 }
159 EXPORT_SYMBOL(ibnl_put_attr);
160 
161 static int rdma_nl_rcv_msg(struct sk_buff *skb, struct nlmsghdr *nlh,
162 			   struct netlink_ext_ack *extack)
163 {
164 	int type = nlh->nlmsg_type;
165 	unsigned int index = RDMA_NL_GET_CLIENT(type);
166 	unsigned int op = RDMA_NL_GET_OP(type);
167 	const struct rdma_nl_cbs *cb_table;
168 
169 	if (!is_nl_valid(index, op))
170 		return -EINVAL;
171 
172 	cb_table = rdma_nl_types[index].cb_table;
173 
174 	if ((cb_table[op].flags & RDMA_NL_ADMIN_PERM) &&
175 	    !netlink_capable(skb, CAP_NET_ADMIN))
176 		return -EPERM;
177 
178 	/* FIXME: Convert IWCM to properly handle doit callbacks */
179 	if ((nlh->nlmsg_flags & NLM_F_DUMP) || index == RDMA_NL_RDMA_CM ||
180 	    index == RDMA_NL_IWCM) {
181 		struct netlink_dump_control c = {
182 			.dump = cb_table[op].dump,
183 		};
184 		return netlink_dump_start(nls, skb, nlh, &c);
185 	}
186 
187 	if (cb_table[op].doit)
188 		return cb_table[op].doit(skb, nlh, extack);
189 
190 	return 0;
191 }
192 
193 /*
194  * This function is similar to netlink_rcv_skb with one exception:
195  * It calls to the callback for the netlink messages without NLM_F_REQUEST
196  * flag. These messages are intended for RDMA_NL_LS consumer, so it is allowed
197  * for that consumer only.
198  */
199 static int rdma_nl_rcv_skb(struct sk_buff *skb, int (*cb)(struct sk_buff *,
200 						   struct nlmsghdr *,
201 						   struct netlink_ext_ack *))
202 {
203 	struct netlink_ext_ack extack = {};
204 	struct nlmsghdr *nlh;
205 	int err;
206 
207 	while (skb->len >= nlmsg_total_size(0)) {
208 		int msglen;
209 
210 		nlh = nlmsg_hdr(skb);
211 		err = 0;
212 
213 		if (nlh->nlmsg_len < NLMSG_HDRLEN || skb->len < nlh->nlmsg_len)
214 			return 0;
215 
216 		/*
217 		 * Generally speaking, the only requests are handled
218 		 * by the kernel, but RDMA_NL_LS is different, because it
219 		 * runs backward netlink scheme. Kernel initiates messages
220 		 * and waits for reply with data to keep pathrecord cache
221 		 * in sync.
222 		 */
223 		if (!(nlh->nlmsg_flags & NLM_F_REQUEST) &&
224 		    (RDMA_NL_GET_CLIENT(nlh->nlmsg_type) != RDMA_NL_LS))
225 			goto ack;
226 
227 		/* Skip control messages */
228 		if (nlh->nlmsg_type < NLMSG_MIN_TYPE)
229 			goto ack;
230 
231 		err = cb(skb, nlh, &extack);
232 		if (err == -EINTR)
233 			goto skip;
234 
235 ack:
236 		if (nlh->nlmsg_flags & NLM_F_ACK || err)
237 			netlink_ack(skb, nlh, err, &extack);
238 
239 skip:
240 		msglen = NLMSG_ALIGN(nlh->nlmsg_len);
241 		if (msglen > skb->len)
242 			msglen = skb->len;
243 		skb_pull(skb, msglen);
244 	}
245 
246 	return 0;
247 }
248 
249 static void rdma_nl_rcv(struct sk_buff *skb)
250 {
251 	mutex_lock(&rdma_nl_mutex);
252 	rdma_nl_rcv_skb(skb, &rdma_nl_rcv_msg);
253 	mutex_unlock(&rdma_nl_mutex);
254 }
255 
256 int rdma_nl_unicast(struct sk_buff *skb, u32 pid)
257 {
258 	int err;
259 
260 	err = netlink_unicast(nls, skb, pid, MSG_DONTWAIT);
261 	return (err < 0) ? err : 0;
262 }
263 EXPORT_SYMBOL(rdma_nl_unicast);
264 
265 int rdma_nl_unicast_wait(struct sk_buff *skb, __u32 pid)
266 {
267 	int err;
268 
269 	err = netlink_unicast(nls, skb, pid, 0);
270 	return (err < 0) ? err : 0;
271 }
272 EXPORT_SYMBOL(rdma_nl_unicast_wait);
273 
274 int rdma_nl_multicast(struct sk_buff *skb, unsigned int group, gfp_t flags)
275 {
276 	return nlmsg_multicast(nls, skb, 0, group, flags);
277 }
278 EXPORT_SYMBOL(rdma_nl_multicast);
279 
280 int __init rdma_nl_init(void)
281 {
282 	struct netlink_kernel_cfg cfg = {
283 		.input	= rdma_nl_rcv,
284 	};
285 
286 	nls = netlink_kernel_create(&init_net, NETLINK_RDMA, &cfg);
287 	if (!nls)
288 		return -ENOMEM;
289 
290 	nls->sk_sndtimeo = 10 * HZ;
291 	return 0;
292 }
293 
294 void rdma_nl_exit(void)
295 {
296 	int idx;
297 
298 	for (idx = 0; idx < RDMA_NL_NUM_CLIENTS; idx++)
299 		rdma_nl_unregister(idx);
300 
301 	netlink_kernel_release(nls);
302 }
303 
304 MODULE_ALIAS_NET_PF_PROTO(PF_NETLINK, NETLINK_RDMA);
305