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/netns/generic.h>
40 #include <net/sock.h>
41 #include <rdma/rdma_netlink.h>
42 #include <linux/module.h>
43 #include "core_priv.h"
44 
45 static DEFINE_MUTEX(rdma_nl_mutex);
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 	struct rdma_dev_net *rnet = rdma_net_to_dev_net(&init_net);
53 
54 	return netlink_has_listeners(rnet->nl_sock, group);
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_IWCM] = RDMA_NL_IWPM_NUM_OPS,
62 		[RDMA_NL_LS] = RDMA_NL_LS_NUM_OPS,
63 		[RDMA_NL_NLDEV] = RDMA_NLDEV_NUM_OPS,
64 	};
65 
66 	/*
67 	 * This BUILD_BUG_ON is intended to catch addition of new
68 	 * RDMA netlink protocol without updating the array above.
69 	 */
70 	BUILD_BUG_ON(RDMA_NL_NUM_CLIENTS != 6);
71 
72 	if (type >= RDMA_NL_NUM_CLIENTS)
73 		return false;
74 
75 	return (op < max_num_ops[type]) ? true : false;
76 }
77 
78 static bool
79 is_nl_valid(const struct sk_buff *skb, 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 	/*
87 	 * Currently only NLDEV client is supporting netlink commands in
88 	 * non init_net net namespace.
89 	 */
90 	if (sock_net(skb->sk) != &init_net && type != RDMA_NL_NLDEV)
91 		return false;
92 
93 	if (!rdma_nl_types[type].cb_table) {
94 		mutex_unlock(&rdma_nl_mutex);
95 		request_module("rdma-netlink-subsys-%d", type);
96 		mutex_lock(&rdma_nl_mutex);
97 	}
98 
99 	cb_table = rdma_nl_types[type].cb_table;
100 
101 	if (!cb_table || (!cb_table[op].dump && !cb_table[op].doit))
102 		return false;
103 	return true;
104 }
105 
106 void rdma_nl_register(unsigned int index,
107 		      const struct rdma_nl_cbs cb_table[])
108 {
109 	mutex_lock(&rdma_nl_mutex);
110 	if (!is_nl_msg_valid(index, 0)) {
111 		/*
112 		 * All clients are not interesting in success/failure of
113 		 * this call. They want to see the print to error log and
114 		 * continue their initialization. Print warning for them,
115 		 * because it is programmer's error to be here.
116 		 */
117 		mutex_unlock(&rdma_nl_mutex);
118 		WARN(true,
119 		     "The not-valid %u index was supplied to RDMA netlink\n",
120 		     index);
121 		return;
122 	}
123 
124 	if (rdma_nl_types[index].cb_table) {
125 		mutex_unlock(&rdma_nl_mutex);
126 		WARN(true,
127 		     "The %u index is already registered in RDMA netlink\n",
128 		     index);
129 		return;
130 	}
131 
132 	rdma_nl_types[index].cb_table = cb_table;
133 	mutex_unlock(&rdma_nl_mutex);
134 }
135 EXPORT_SYMBOL(rdma_nl_register);
136 
137 void rdma_nl_unregister(unsigned int index)
138 {
139 	mutex_lock(&rdma_nl_mutex);
140 	rdma_nl_types[index].cb_table = NULL;
141 	mutex_unlock(&rdma_nl_mutex);
142 }
143 EXPORT_SYMBOL(rdma_nl_unregister);
144 
145 void *ibnl_put_msg(struct sk_buff *skb, struct nlmsghdr **nlh, int seq,
146 		   int len, int client, int op, int flags)
147 {
148 	*nlh = nlmsg_put(skb, 0, seq, RDMA_NL_GET_TYPE(client, op), len, flags);
149 	if (!*nlh)
150 		return NULL;
151 	return nlmsg_data(*nlh);
152 }
153 EXPORT_SYMBOL(ibnl_put_msg);
154 
155 int ibnl_put_attr(struct sk_buff *skb, struct nlmsghdr *nlh,
156 		  int len, void *data, int type)
157 {
158 	if (nla_put(skb, type, len, data)) {
159 		nlmsg_cancel(skb, nlh);
160 		return -EMSGSIZE;
161 	}
162 	return 0;
163 }
164 EXPORT_SYMBOL(ibnl_put_attr);
165 
166 static int rdma_nl_rcv_msg(struct sk_buff *skb, struct nlmsghdr *nlh,
167 			   struct netlink_ext_ack *extack)
168 {
169 	int type = nlh->nlmsg_type;
170 	unsigned int index = RDMA_NL_GET_CLIENT(type);
171 	unsigned int op = RDMA_NL_GET_OP(type);
172 	const struct rdma_nl_cbs *cb_table;
173 
174 	if (!is_nl_valid(skb, index, op))
175 		return -EINVAL;
176 
177 	cb_table = rdma_nl_types[index].cb_table;
178 
179 	if ((cb_table[op].flags & RDMA_NL_ADMIN_PERM) &&
180 	    !netlink_capable(skb, CAP_NET_ADMIN))
181 		return -EPERM;
182 
183 	/*
184 	 * LS responses overload the 0x100 (NLM_F_ROOT) flag.  Don't
185 	 * mistakenly call the .dump() function.
186 	 */
187 	if (index == RDMA_NL_LS) {
188 		if (cb_table[op].doit)
189 			return cb_table[op].doit(skb, nlh, extack);
190 		return -EINVAL;
191 	}
192 	/* FIXME: Convert IWCM to properly handle doit callbacks */
193 	if ((nlh->nlmsg_flags & NLM_F_DUMP) || index == RDMA_NL_IWCM) {
194 		struct netlink_dump_control c = {
195 			.dump = cb_table[op].dump,
196 		};
197 		if (c.dump)
198 			return netlink_dump_start(skb->sk, skb, nlh, &c);
199 		return -EINVAL;
200 	}
201 
202 	if (cb_table[op].doit)
203 		return cb_table[op].doit(skb, nlh, extack);
204 
205 	return 0;
206 }
207 
208 /*
209  * This function is similar to netlink_rcv_skb with one exception:
210  * It calls to the callback for the netlink messages without NLM_F_REQUEST
211  * flag. These messages are intended for RDMA_NL_LS consumer, so it is allowed
212  * for that consumer only.
213  */
214 static int rdma_nl_rcv_skb(struct sk_buff *skb, int (*cb)(struct sk_buff *,
215 						   struct nlmsghdr *,
216 						   struct netlink_ext_ack *))
217 {
218 	struct netlink_ext_ack extack = {};
219 	struct nlmsghdr *nlh;
220 	int err;
221 
222 	while (skb->len >= nlmsg_total_size(0)) {
223 		int msglen;
224 
225 		nlh = nlmsg_hdr(skb);
226 		err = 0;
227 
228 		if (nlh->nlmsg_len < NLMSG_HDRLEN || skb->len < nlh->nlmsg_len)
229 			return 0;
230 
231 		/*
232 		 * Generally speaking, the only requests are handled
233 		 * by the kernel, but RDMA_NL_LS is different, because it
234 		 * runs backward netlink scheme. Kernel initiates messages
235 		 * and waits for reply with data to keep pathrecord cache
236 		 * in sync.
237 		 */
238 		if (!(nlh->nlmsg_flags & NLM_F_REQUEST) &&
239 		    (RDMA_NL_GET_CLIENT(nlh->nlmsg_type) != RDMA_NL_LS))
240 			goto ack;
241 
242 		/* Skip control messages */
243 		if (nlh->nlmsg_type < NLMSG_MIN_TYPE)
244 			goto ack;
245 
246 		err = cb(skb, nlh, &extack);
247 		if (err == -EINTR)
248 			goto skip;
249 
250 ack:
251 		if (nlh->nlmsg_flags & NLM_F_ACK || err)
252 			netlink_ack(skb, nlh, err, &extack);
253 
254 skip:
255 		msglen = NLMSG_ALIGN(nlh->nlmsg_len);
256 		if (msglen > skb->len)
257 			msglen = skb->len;
258 		skb_pull(skb, msglen);
259 	}
260 
261 	return 0;
262 }
263 
264 static void rdma_nl_rcv(struct sk_buff *skb)
265 {
266 	mutex_lock(&rdma_nl_mutex);
267 	rdma_nl_rcv_skb(skb, &rdma_nl_rcv_msg);
268 	mutex_unlock(&rdma_nl_mutex);
269 }
270 
271 int rdma_nl_unicast(struct net *net, struct sk_buff *skb, u32 pid)
272 {
273 	struct rdma_dev_net *rnet = rdma_net_to_dev_net(net);
274 	int err;
275 
276 	err = netlink_unicast(rnet->nl_sock, skb, pid, MSG_DONTWAIT);
277 	return (err < 0) ? err : 0;
278 }
279 EXPORT_SYMBOL(rdma_nl_unicast);
280 
281 int rdma_nl_unicast_wait(struct net *net, struct sk_buff *skb, __u32 pid)
282 {
283 	struct rdma_dev_net *rnet = rdma_net_to_dev_net(net);
284 	int err;
285 
286 	err = netlink_unicast(rnet->nl_sock, skb, pid, 0);
287 	return (err < 0) ? err : 0;
288 }
289 EXPORT_SYMBOL(rdma_nl_unicast_wait);
290 
291 int rdma_nl_multicast(struct net *net, struct sk_buff *skb,
292 		      unsigned int group, gfp_t flags)
293 {
294 	struct rdma_dev_net *rnet = rdma_net_to_dev_net(net);
295 
296 	return nlmsg_multicast(rnet->nl_sock, skb, 0, group, flags);
297 }
298 EXPORT_SYMBOL(rdma_nl_multicast);
299 
300 void rdma_nl_exit(void)
301 {
302 	int idx;
303 
304 	for (idx = 0; idx < RDMA_NL_NUM_CLIENTS; idx++)
305 		WARN(rdma_nl_types[idx].cb_table,
306 		     "Netlink client %d wasn't released prior to unloading %s\n",
307 		     idx, KBUILD_MODNAME);
308 }
309 
310 int rdma_nl_net_init(struct rdma_dev_net *rnet)
311 {
312 	struct net *net = read_pnet(&rnet->net);
313 	struct netlink_kernel_cfg cfg = {
314 		.input	= rdma_nl_rcv,
315 	};
316 	struct sock *nls;
317 
318 	nls = netlink_kernel_create(net, NETLINK_RDMA, &cfg);
319 	if (!nls)
320 		return -ENOMEM;
321 
322 	nls->sk_sndtimeo = 10 * HZ;
323 	rnet->nl_sock = nls;
324 	return 0;
325 }
326 
327 void rdma_nl_net_exit(struct rdma_dev_net *rnet)
328 {
329 	netlink_kernel_release(rnet->nl_sock);
330 }
331 
332 MODULE_ALIAS_NET_PF_PROTO(PF_NETLINK, NETLINK_RDMA);
333