xref: /linux/drivers/infiniband/core/nldev.c (revision 1f2367a39f17bd553a75e179a747f9b257bc9478)
1 /*
2  * Copyright (c) 2017 Mellanox Technologies. All rights reserved.
3  *
4  * Redistribution and use in source and binary forms, with or without
5  * modification, are permitted provided that the following conditions are met:
6  *
7  * 1. Redistributions of source code must retain the above copyright
8  *    notice, this list of conditions and the following disclaimer.
9  * 2. Redistributions in binary form must reproduce the above copyright
10  *    notice, this list of conditions and the following disclaimer in the
11  *    documentation and/or other materials provided with the distribution.
12  * 3. Neither the names of the copyright holders nor the names of its
13  *    contributors may be used to endorse or promote products derived from
14  *    this software without specific prior written permission.
15  *
16  * Alternatively, this software may be distributed under the terms of the
17  * GNU General Public License ("GPL") version 2 as published by the Free
18  * Software Foundation.
19  *
20  * THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS "AS IS"
21  * AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
22  * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
23  * ARE DISCLAIMED. IN NO EVENT SHALL THE COPYRIGHT OWNER OR CONTRIBUTORS BE
24  * LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR
25  * CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF
26  * SUBSTITUTE GOODS OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS
27  * INTERRUPTION) HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN
28  * CONTRACT, STRICT LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE)
29  * ARISING IN ANY WAY OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE
30  * POSSIBILITY OF SUCH DAMAGE.
31  */
32 
33 #include <linux/module.h>
34 #include <linux/pid.h>
35 #include <linux/pid_namespace.h>
36 #include <linux/mutex.h>
37 #include <net/netlink.h>
38 #include <rdma/rdma_cm.h>
39 #include <rdma/rdma_netlink.h>
40 
41 #include "core_priv.h"
42 #include "cma_priv.h"
43 #include "restrack.h"
44 
45 static const struct nla_policy nldev_policy[RDMA_NLDEV_ATTR_MAX] = {
46 	[RDMA_NLDEV_ATTR_DEV_INDEX]     = { .type = NLA_U32 },
47 	[RDMA_NLDEV_ATTR_DEV_NAME]	= { .type = NLA_NUL_STRING,
48 					    .len = IB_DEVICE_NAME_MAX - 1},
49 	[RDMA_NLDEV_ATTR_PORT_INDEX]	= { .type = NLA_U32 },
50 	[RDMA_NLDEV_ATTR_FW_VERSION]	= { .type = NLA_NUL_STRING,
51 					    .len = IB_FW_VERSION_NAME_MAX - 1},
52 	[RDMA_NLDEV_ATTR_NODE_GUID]	= { .type = NLA_U64 },
53 	[RDMA_NLDEV_ATTR_SYS_IMAGE_GUID] = { .type = NLA_U64 },
54 	[RDMA_NLDEV_ATTR_SUBNET_PREFIX]	= { .type = NLA_U64 },
55 	[RDMA_NLDEV_ATTR_LID]		= { .type = NLA_U32 },
56 	[RDMA_NLDEV_ATTR_SM_LID]	= { .type = NLA_U32 },
57 	[RDMA_NLDEV_ATTR_LMC]		= { .type = NLA_U8 },
58 	[RDMA_NLDEV_ATTR_PORT_STATE]	= { .type = NLA_U8 },
59 	[RDMA_NLDEV_ATTR_PORT_PHYS_STATE] = { .type = NLA_U8 },
60 	[RDMA_NLDEV_ATTR_DEV_NODE_TYPE] = { .type = NLA_U8 },
61 	[RDMA_NLDEV_ATTR_RES_SUMMARY]	= { .type = NLA_NESTED },
62 	[RDMA_NLDEV_ATTR_RES_SUMMARY_ENTRY]	= { .type = NLA_NESTED },
63 	[RDMA_NLDEV_ATTR_RES_SUMMARY_ENTRY_NAME] = { .type = NLA_NUL_STRING,
64 					     .len = 16 },
65 	[RDMA_NLDEV_ATTR_RES_SUMMARY_ENTRY_CURR] = { .type = NLA_U64 },
66 	[RDMA_NLDEV_ATTR_RES_QP]		= { .type = NLA_NESTED },
67 	[RDMA_NLDEV_ATTR_RES_QP_ENTRY]		= { .type = NLA_NESTED },
68 	[RDMA_NLDEV_ATTR_RES_LQPN]		= { .type = NLA_U32 },
69 	[RDMA_NLDEV_ATTR_RES_RQPN]		= { .type = NLA_U32 },
70 	[RDMA_NLDEV_ATTR_RES_RQ_PSN]		= { .type = NLA_U32 },
71 	[RDMA_NLDEV_ATTR_RES_SQ_PSN]		= { .type = NLA_U32 },
72 	[RDMA_NLDEV_ATTR_RES_PATH_MIG_STATE] = { .type = NLA_U8 },
73 	[RDMA_NLDEV_ATTR_RES_TYPE]		= { .type = NLA_U8 },
74 	[RDMA_NLDEV_ATTR_RES_STATE]		= { .type = NLA_U8 },
75 	[RDMA_NLDEV_ATTR_RES_PID]		= { .type = NLA_U32 },
76 	[RDMA_NLDEV_ATTR_RES_KERN_NAME]		= { .type = NLA_NUL_STRING,
77 						    .len = TASK_COMM_LEN },
78 	[RDMA_NLDEV_ATTR_RES_CM_ID]		= { .type = NLA_NESTED },
79 	[RDMA_NLDEV_ATTR_RES_CM_ID_ENTRY]	= { .type = NLA_NESTED },
80 	[RDMA_NLDEV_ATTR_RES_PS]		= { .type = NLA_U32 },
81 	[RDMA_NLDEV_ATTR_RES_SRC_ADDR]	= {
82 			.len = sizeof(struct __kernel_sockaddr_storage) },
83 	[RDMA_NLDEV_ATTR_RES_DST_ADDR]	= {
84 			.len = sizeof(struct __kernel_sockaddr_storage) },
85 	[RDMA_NLDEV_ATTR_RES_CQ]		= { .type = NLA_NESTED },
86 	[RDMA_NLDEV_ATTR_RES_CQ_ENTRY]		= { .type = NLA_NESTED },
87 	[RDMA_NLDEV_ATTR_RES_CQE]		= { .type = NLA_U32 },
88 	[RDMA_NLDEV_ATTR_RES_USECNT]		= { .type = NLA_U64 },
89 	[RDMA_NLDEV_ATTR_RES_POLL_CTX]		= { .type = NLA_U8 },
90 	[RDMA_NLDEV_ATTR_RES_MR]		= { .type = NLA_NESTED },
91 	[RDMA_NLDEV_ATTR_RES_MR_ENTRY]		= { .type = NLA_NESTED },
92 	[RDMA_NLDEV_ATTR_RES_RKEY]		= { .type = NLA_U32 },
93 	[RDMA_NLDEV_ATTR_RES_LKEY]		= { .type = NLA_U32 },
94 	[RDMA_NLDEV_ATTR_RES_IOVA]		= { .type = NLA_U64 },
95 	[RDMA_NLDEV_ATTR_RES_MRLEN]		= { .type = NLA_U64 },
96 	[RDMA_NLDEV_ATTR_RES_PD]		= { .type = NLA_NESTED },
97 	[RDMA_NLDEV_ATTR_RES_PD_ENTRY]		= { .type = NLA_NESTED },
98 	[RDMA_NLDEV_ATTR_RES_LOCAL_DMA_LKEY]	= { .type = NLA_U32 },
99 	[RDMA_NLDEV_ATTR_RES_UNSAFE_GLOBAL_RKEY] = { .type = NLA_U32 },
100 	[RDMA_NLDEV_ATTR_NDEV_INDEX]		= { .type = NLA_U32 },
101 	[RDMA_NLDEV_ATTR_NDEV_NAME]		= { .type = NLA_NUL_STRING,
102 						    .len = IFNAMSIZ },
103 	[RDMA_NLDEV_ATTR_DRIVER]		= { .type = NLA_NESTED },
104 	[RDMA_NLDEV_ATTR_DRIVER_ENTRY]		= { .type = NLA_NESTED },
105 	[RDMA_NLDEV_ATTR_DRIVER_STRING]		= { .type = NLA_NUL_STRING,
106 				    .len = RDMA_NLDEV_ATTR_ENTRY_STRLEN },
107 	[RDMA_NLDEV_ATTR_DRIVER_PRINT_TYPE]	= { .type = NLA_U8 },
108 	[RDMA_NLDEV_ATTR_DRIVER_S32]		= { .type = NLA_S32 },
109 	[RDMA_NLDEV_ATTR_DRIVER_U32]		= { .type = NLA_U32 },
110 	[RDMA_NLDEV_ATTR_DRIVER_S64]		= { .type = NLA_S64 },
111 	[RDMA_NLDEV_ATTR_DRIVER_U64]		= { .type = NLA_U64 },
112 	[RDMA_NLDEV_ATTR_RES_PDN]		= { .type = NLA_U32 },
113 	[RDMA_NLDEV_ATTR_RES_CQN]               = { .type = NLA_U32 },
114 	[RDMA_NLDEV_ATTR_RES_MRN]               = { .type = NLA_U32 },
115 	[RDMA_NLDEV_ATTR_RES_CM_IDN]            = { .type = NLA_U32 },
116 	[RDMA_NLDEV_ATTR_RES_CTXN]              = { .type = NLA_U32 },
117 	[RDMA_NLDEV_ATTR_LINK_TYPE]		= { .type = NLA_NUL_STRING,
118 				    .len = RDMA_NLDEV_ATTR_ENTRY_STRLEN },
119 };
120 
121 static int put_driver_name_print_type(struct sk_buff *msg, const char *name,
122 				      enum rdma_nldev_print_type print_type)
123 {
124 	if (nla_put_string(msg, RDMA_NLDEV_ATTR_DRIVER_STRING, name))
125 		return -EMSGSIZE;
126 	if (print_type != RDMA_NLDEV_PRINT_TYPE_UNSPEC &&
127 	    nla_put_u8(msg, RDMA_NLDEV_ATTR_DRIVER_PRINT_TYPE, print_type))
128 		return -EMSGSIZE;
129 
130 	return 0;
131 }
132 
133 static int _rdma_nl_put_driver_u32(struct sk_buff *msg, const char *name,
134 				   enum rdma_nldev_print_type print_type,
135 				   u32 value)
136 {
137 	if (put_driver_name_print_type(msg, name, print_type))
138 		return -EMSGSIZE;
139 	if (nla_put_u32(msg, RDMA_NLDEV_ATTR_DRIVER_U32, value))
140 		return -EMSGSIZE;
141 
142 	return 0;
143 }
144 
145 static int _rdma_nl_put_driver_u64(struct sk_buff *msg, const char *name,
146 				   enum rdma_nldev_print_type print_type,
147 				   u64 value)
148 {
149 	if (put_driver_name_print_type(msg, name, print_type))
150 		return -EMSGSIZE;
151 	if (nla_put_u64_64bit(msg, RDMA_NLDEV_ATTR_DRIVER_U64, value,
152 			      RDMA_NLDEV_ATTR_PAD))
153 		return -EMSGSIZE;
154 
155 	return 0;
156 }
157 
158 int rdma_nl_put_driver_u32(struct sk_buff *msg, const char *name, u32 value)
159 {
160 	return _rdma_nl_put_driver_u32(msg, name, RDMA_NLDEV_PRINT_TYPE_UNSPEC,
161 				       value);
162 }
163 EXPORT_SYMBOL(rdma_nl_put_driver_u32);
164 
165 int rdma_nl_put_driver_u32_hex(struct sk_buff *msg, const char *name,
166 			       u32 value)
167 {
168 	return _rdma_nl_put_driver_u32(msg, name, RDMA_NLDEV_PRINT_TYPE_HEX,
169 				       value);
170 }
171 EXPORT_SYMBOL(rdma_nl_put_driver_u32_hex);
172 
173 int rdma_nl_put_driver_u64(struct sk_buff *msg, const char *name, u64 value)
174 {
175 	return _rdma_nl_put_driver_u64(msg, name, RDMA_NLDEV_PRINT_TYPE_UNSPEC,
176 				       value);
177 }
178 EXPORT_SYMBOL(rdma_nl_put_driver_u64);
179 
180 int rdma_nl_put_driver_u64_hex(struct sk_buff *msg, const char *name, u64 value)
181 {
182 	return _rdma_nl_put_driver_u64(msg, name, RDMA_NLDEV_PRINT_TYPE_HEX,
183 				       value);
184 }
185 EXPORT_SYMBOL(rdma_nl_put_driver_u64_hex);
186 
187 static int fill_nldev_handle(struct sk_buff *msg, struct ib_device *device)
188 {
189 	if (nla_put_u32(msg, RDMA_NLDEV_ATTR_DEV_INDEX, device->index))
190 		return -EMSGSIZE;
191 	if (nla_put_string(msg, RDMA_NLDEV_ATTR_DEV_NAME,
192 			   dev_name(&device->dev)))
193 		return -EMSGSIZE;
194 
195 	return 0;
196 }
197 
198 static int fill_dev_info(struct sk_buff *msg, struct ib_device *device)
199 {
200 	char fw[IB_FW_VERSION_NAME_MAX];
201 
202 	if (fill_nldev_handle(msg, device))
203 		return -EMSGSIZE;
204 
205 	if (nla_put_u32(msg, RDMA_NLDEV_ATTR_PORT_INDEX, rdma_end_port(device)))
206 		return -EMSGSIZE;
207 
208 	BUILD_BUG_ON(sizeof(device->attrs.device_cap_flags) != sizeof(u64));
209 	if (nla_put_u64_64bit(msg, RDMA_NLDEV_ATTR_CAP_FLAGS,
210 			      device->attrs.device_cap_flags,
211 			      RDMA_NLDEV_ATTR_PAD))
212 		return -EMSGSIZE;
213 
214 	ib_get_device_fw_str(device, fw);
215 	/* Device without FW has strlen(fw) = 0 */
216 	if (strlen(fw) && nla_put_string(msg, RDMA_NLDEV_ATTR_FW_VERSION, fw))
217 		return -EMSGSIZE;
218 
219 	if (nla_put_u64_64bit(msg, RDMA_NLDEV_ATTR_NODE_GUID,
220 			      be64_to_cpu(device->node_guid),
221 			      RDMA_NLDEV_ATTR_PAD))
222 		return -EMSGSIZE;
223 	if (nla_put_u64_64bit(msg, RDMA_NLDEV_ATTR_SYS_IMAGE_GUID,
224 			      be64_to_cpu(device->attrs.sys_image_guid),
225 			      RDMA_NLDEV_ATTR_PAD))
226 		return -EMSGSIZE;
227 	if (nla_put_u8(msg, RDMA_NLDEV_ATTR_DEV_NODE_TYPE, device->node_type))
228 		return -EMSGSIZE;
229 	return 0;
230 }
231 
232 static int fill_port_info(struct sk_buff *msg,
233 			  struct ib_device *device, u32 port,
234 			  const struct net *net)
235 {
236 	struct net_device *netdev = NULL;
237 	struct ib_port_attr attr;
238 	int ret;
239 	u64 cap_flags = 0;
240 
241 	if (fill_nldev_handle(msg, device))
242 		return -EMSGSIZE;
243 
244 	if (nla_put_u32(msg, RDMA_NLDEV_ATTR_PORT_INDEX, port))
245 		return -EMSGSIZE;
246 
247 	ret = ib_query_port(device, port, &attr);
248 	if (ret)
249 		return ret;
250 
251 	if (rdma_protocol_ib(device, port)) {
252 		BUILD_BUG_ON((sizeof(attr.port_cap_flags) +
253 				sizeof(attr.port_cap_flags2)) > sizeof(u64));
254 		cap_flags = attr.port_cap_flags |
255 			((u64)attr.port_cap_flags2 << 32);
256 		if (nla_put_u64_64bit(msg, RDMA_NLDEV_ATTR_CAP_FLAGS,
257 				      cap_flags, RDMA_NLDEV_ATTR_PAD))
258 			return -EMSGSIZE;
259 		if (nla_put_u64_64bit(msg, RDMA_NLDEV_ATTR_SUBNET_PREFIX,
260 				      attr.subnet_prefix, RDMA_NLDEV_ATTR_PAD))
261 			return -EMSGSIZE;
262 		if (nla_put_u32(msg, RDMA_NLDEV_ATTR_LID, attr.lid))
263 			return -EMSGSIZE;
264 		if (nla_put_u32(msg, RDMA_NLDEV_ATTR_SM_LID, attr.sm_lid))
265 			return -EMSGSIZE;
266 		if (nla_put_u8(msg, RDMA_NLDEV_ATTR_LMC, attr.lmc))
267 			return -EMSGSIZE;
268 	}
269 	if (nla_put_u8(msg, RDMA_NLDEV_ATTR_PORT_STATE, attr.state))
270 		return -EMSGSIZE;
271 	if (nla_put_u8(msg, RDMA_NLDEV_ATTR_PORT_PHYS_STATE, attr.phys_state))
272 		return -EMSGSIZE;
273 
274 	netdev = ib_device_get_netdev(device, port);
275 	if (netdev && net_eq(dev_net(netdev), net)) {
276 		ret = nla_put_u32(msg,
277 				  RDMA_NLDEV_ATTR_NDEV_INDEX, netdev->ifindex);
278 		if (ret)
279 			goto out;
280 		ret = nla_put_string(msg,
281 				     RDMA_NLDEV_ATTR_NDEV_NAME, netdev->name);
282 	}
283 
284 out:
285 	if (netdev)
286 		dev_put(netdev);
287 	return ret;
288 }
289 
290 static int fill_res_info_entry(struct sk_buff *msg,
291 			       const char *name, u64 curr)
292 {
293 	struct nlattr *entry_attr;
294 
295 	entry_attr = nla_nest_start(msg, RDMA_NLDEV_ATTR_RES_SUMMARY_ENTRY);
296 	if (!entry_attr)
297 		return -EMSGSIZE;
298 
299 	if (nla_put_string(msg, RDMA_NLDEV_ATTR_RES_SUMMARY_ENTRY_NAME, name))
300 		goto err;
301 	if (nla_put_u64_64bit(msg, RDMA_NLDEV_ATTR_RES_SUMMARY_ENTRY_CURR, curr,
302 			      RDMA_NLDEV_ATTR_PAD))
303 		goto err;
304 
305 	nla_nest_end(msg, entry_attr);
306 	return 0;
307 
308 err:
309 	nla_nest_cancel(msg, entry_attr);
310 	return -EMSGSIZE;
311 }
312 
313 static int fill_res_info(struct sk_buff *msg, struct ib_device *device)
314 {
315 	static const char * const names[RDMA_RESTRACK_MAX] = {
316 		[RDMA_RESTRACK_PD] = "pd",
317 		[RDMA_RESTRACK_CQ] = "cq",
318 		[RDMA_RESTRACK_QP] = "qp",
319 		[RDMA_RESTRACK_CM_ID] = "cm_id",
320 		[RDMA_RESTRACK_MR] = "mr",
321 		[RDMA_RESTRACK_CTX] = "ctx",
322 	};
323 
324 	struct nlattr *table_attr;
325 	int ret, i, curr;
326 
327 	if (fill_nldev_handle(msg, device))
328 		return -EMSGSIZE;
329 
330 	table_attr = nla_nest_start(msg, RDMA_NLDEV_ATTR_RES_SUMMARY);
331 	if (!table_attr)
332 		return -EMSGSIZE;
333 
334 	for (i = 0; i < RDMA_RESTRACK_MAX; i++) {
335 		if (!names[i])
336 			continue;
337 		curr = rdma_restrack_count(device, i,
338 					   task_active_pid_ns(current));
339 		ret = fill_res_info_entry(msg, names[i], curr);
340 		if (ret)
341 			goto err;
342 	}
343 
344 	nla_nest_end(msg, table_attr);
345 	return 0;
346 
347 err:
348 	nla_nest_cancel(msg, table_attr);
349 	return ret;
350 }
351 
352 static int fill_res_name_pid(struct sk_buff *msg,
353 			     struct rdma_restrack_entry *res)
354 {
355 	/*
356 	 * For user resources, user is should read /proc/PID/comm to get the
357 	 * name of the task file.
358 	 */
359 	if (rdma_is_kernel_res(res)) {
360 		if (nla_put_string(msg, RDMA_NLDEV_ATTR_RES_KERN_NAME,
361 		    res->kern_name))
362 			return -EMSGSIZE;
363 	} else {
364 		if (nla_put_u32(msg, RDMA_NLDEV_ATTR_RES_PID,
365 		    task_pid_vnr(res->task)))
366 			return -EMSGSIZE;
367 	}
368 	return 0;
369 }
370 
371 static bool fill_res_entry(struct ib_device *dev, struct sk_buff *msg,
372 			   struct rdma_restrack_entry *res)
373 {
374 	if (!dev->ops.fill_res_entry)
375 		return false;
376 	return dev->ops.fill_res_entry(msg, res);
377 }
378 
379 static int fill_res_qp_entry(struct sk_buff *msg, bool has_cap_net_admin,
380 			     struct rdma_restrack_entry *res, uint32_t port)
381 {
382 	struct ib_qp *qp = container_of(res, struct ib_qp, res);
383 	struct ib_device *dev = qp->device;
384 	struct ib_qp_init_attr qp_init_attr;
385 	struct ib_qp_attr qp_attr;
386 	int ret;
387 
388 	ret = ib_query_qp(qp, &qp_attr, 0, &qp_init_attr);
389 	if (ret)
390 		return ret;
391 
392 	if (port && port != qp_attr.port_num)
393 		return -EAGAIN;
394 
395 	/* In create_qp() port is not set yet */
396 	if (qp_attr.port_num &&
397 	    nla_put_u32(msg, RDMA_NLDEV_ATTR_PORT_INDEX, qp_attr.port_num))
398 		goto err;
399 
400 	if (nla_put_u32(msg, RDMA_NLDEV_ATTR_RES_LQPN, qp->qp_num))
401 		goto err;
402 	if (qp->qp_type == IB_QPT_RC || qp->qp_type == IB_QPT_UC) {
403 		if (nla_put_u32(msg, RDMA_NLDEV_ATTR_RES_RQPN,
404 				qp_attr.dest_qp_num))
405 			goto err;
406 		if (nla_put_u32(msg, RDMA_NLDEV_ATTR_RES_RQ_PSN,
407 				qp_attr.rq_psn))
408 			goto err;
409 	}
410 
411 	if (nla_put_u32(msg, RDMA_NLDEV_ATTR_RES_SQ_PSN, qp_attr.sq_psn))
412 		goto err;
413 
414 	if (qp->qp_type == IB_QPT_RC || qp->qp_type == IB_QPT_UC ||
415 	    qp->qp_type == IB_QPT_XRC_INI || qp->qp_type == IB_QPT_XRC_TGT) {
416 		if (nla_put_u8(msg, RDMA_NLDEV_ATTR_RES_PATH_MIG_STATE,
417 			       qp_attr.path_mig_state))
418 			goto err;
419 	}
420 	if (nla_put_u8(msg, RDMA_NLDEV_ATTR_RES_TYPE, qp->qp_type))
421 		goto err;
422 	if (nla_put_u8(msg, RDMA_NLDEV_ATTR_RES_STATE, qp_attr.qp_state))
423 		goto err;
424 
425 	if (!rdma_is_kernel_res(res) &&
426 	    nla_put_u32(msg, RDMA_NLDEV_ATTR_RES_PDN, qp->pd->res.id))
427 		goto err;
428 
429 	if (fill_res_name_pid(msg, res))
430 		goto err;
431 
432 	if (fill_res_entry(dev, msg, res))
433 		goto err;
434 
435 	return 0;
436 
437 err:	return -EMSGSIZE;
438 }
439 
440 static int fill_res_cm_id_entry(struct sk_buff *msg, bool has_cap_net_admin,
441 				struct rdma_restrack_entry *res, uint32_t port)
442 {
443 	struct rdma_id_private *id_priv =
444 				container_of(res, struct rdma_id_private, res);
445 	struct ib_device *dev = id_priv->id.device;
446 	struct rdma_cm_id *cm_id = &id_priv->id;
447 
448 	if (port && port != cm_id->port_num)
449 		return 0;
450 
451 	if (cm_id->port_num &&
452 	    nla_put_u32(msg, RDMA_NLDEV_ATTR_PORT_INDEX, cm_id->port_num))
453 		goto err;
454 
455 	if (id_priv->qp_num) {
456 		if (nla_put_u32(msg, RDMA_NLDEV_ATTR_RES_LQPN, id_priv->qp_num))
457 			goto err;
458 		if (nla_put_u8(msg, RDMA_NLDEV_ATTR_RES_TYPE, cm_id->qp_type))
459 			goto err;
460 	}
461 
462 	if (nla_put_u32(msg, RDMA_NLDEV_ATTR_RES_PS, cm_id->ps))
463 		goto err;
464 
465 	if (nla_put_u8(msg, RDMA_NLDEV_ATTR_RES_STATE, id_priv->state))
466 		goto err;
467 
468 	if (cm_id->route.addr.src_addr.ss_family &&
469 	    nla_put(msg, RDMA_NLDEV_ATTR_RES_SRC_ADDR,
470 		    sizeof(cm_id->route.addr.src_addr),
471 		    &cm_id->route.addr.src_addr))
472 		goto err;
473 	if (cm_id->route.addr.dst_addr.ss_family &&
474 	    nla_put(msg, RDMA_NLDEV_ATTR_RES_DST_ADDR,
475 		    sizeof(cm_id->route.addr.dst_addr),
476 		    &cm_id->route.addr.dst_addr))
477 		goto err;
478 
479 	if (nla_put_u32(msg, RDMA_NLDEV_ATTR_RES_CM_IDN, res->id))
480 		goto err;
481 
482 	if (fill_res_name_pid(msg, res))
483 		goto err;
484 
485 	if (fill_res_entry(dev, msg, res))
486 		goto err;
487 
488 	return 0;
489 
490 err: return -EMSGSIZE;
491 }
492 
493 static int fill_res_cq_entry(struct sk_buff *msg, bool has_cap_net_admin,
494 			     struct rdma_restrack_entry *res, uint32_t port)
495 {
496 	struct ib_cq *cq = container_of(res, struct ib_cq, res);
497 	struct ib_device *dev = cq->device;
498 
499 	if (nla_put_u32(msg, RDMA_NLDEV_ATTR_RES_CQE, cq->cqe))
500 		goto err;
501 	if (nla_put_u64_64bit(msg, RDMA_NLDEV_ATTR_RES_USECNT,
502 			      atomic_read(&cq->usecnt), RDMA_NLDEV_ATTR_PAD))
503 		goto err;
504 
505 	/* Poll context is only valid for kernel CQs */
506 	if (rdma_is_kernel_res(res) &&
507 	    nla_put_u8(msg, RDMA_NLDEV_ATTR_RES_POLL_CTX, cq->poll_ctx))
508 		goto err;
509 
510 	if (nla_put_u32(msg, RDMA_NLDEV_ATTR_RES_CQN, res->id))
511 		goto err;
512 	if (!rdma_is_kernel_res(res) &&
513 	    nla_put_u32(msg, RDMA_NLDEV_ATTR_RES_CTXN,
514 			cq->uobject->context->res.id))
515 		goto err;
516 
517 	if (fill_res_name_pid(msg, res))
518 		goto err;
519 
520 	if (fill_res_entry(dev, msg, res))
521 		goto err;
522 
523 	return 0;
524 
525 err:	return -EMSGSIZE;
526 }
527 
528 static int fill_res_mr_entry(struct sk_buff *msg, bool has_cap_net_admin,
529 			     struct rdma_restrack_entry *res, uint32_t port)
530 {
531 	struct ib_mr *mr = container_of(res, struct ib_mr, res);
532 	struct ib_device *dev = mr->pd->device;
533 
534 	if (has_cap_net_admin) {
535 		if (nla_put_u32(msg, RDMA_NLDEV_ATTR_RES_RKEY, mr->rkey))
536 			goto err;
537 		if (nla_put_u32(msg, RDMA_NLDEV_ATTR_RES_LKEY, mr->lkey))
538 			goto err;
539 	}
540 
541 	if (nla_put_u64_64bit(msg, RDMA_NLDEV_ATTR_RES_MRLEN, mr->length,
542 			      RDMA_NLDEV_ATTR_PAD))
543 		goto err;
544 
545 	if (nla_put_u32(msg, RDMA_NLDEV_ATTR_RES_MRN, res->id))
546 		goto err;
547 
548 	if (!rdma_is_kernel_res(res) &&
549 	    nla_put_u32(msg, RDMA_NLDEV_ATTR_RES_PDN, mr->pd->res.id))
550 		goto err;
551 
552 	if (fill_res_name_pid(msg, res))
553 		goto err;
554 
555 	if (fill_res_entry(dev, msg, res))
556 		goto err;
557 
558 	return 0;
559 
560 err:	return -EMSGSIZE;
561 }
562 
563 static int fill_res_pd_entry(struct sk_buff *msg, bool has_cap_net_admin,
564 			     struct rdma_restrack_entry *res, uint32_t port)
565 {
566 	struct ib_pd *pd = container_of(res, struct ib_pd, res);
567 	struct ib_device *dev = pd->device;
568 
569 	if (has_cap_net_admin) {
570 		if (nla_put_u32(msg, RDMA_NLDEV_ATTR_RES_LOCAL_DMA_LKEY,
571 				pd->local_dma_lkey))
572 			goto err;
573 		if ((pd->flags & IB_PD_UNSAFE_GLOBAL_RKEY) &&
574 		    nla_put_u32(msg, RDMA_NLDEV_ATTR_RES_UNSAFE_GLOBAL_RKEY,
575 				pd->unsafe_global_rkey))
576 			goto err;
577 	}
578 	if (nla_put_u64_64bit(msg, RDMA_NLDEV_ATTR_RES_USECNT,
579 			      atomic_read(&pd->usecnt), RDMA_NLDEV_ATTR_PAD))
580 		goto err;
581 
582 	if (nla_put_u32(msg, RDMA_NLDEV_ATTR_RES_PDN, res->id))
583 		goto err;
584 
585 	if (!rdma_is_kernel_res(res) &&
586 	    nla_put_u32(msg, RDMA_NLDEV_ATTR_RES_CTXN,
587 			pd->uobject->context->res.id))
588 		goto err;
589 
590 	if (fill_res_name_pid(msg, res))
591 		goto err;
592 
593 	if (fill_res_entry(dev, msg, res))
594 		goto err;
595 
596 	return 0;
597 
598 err:	return -EMSGSIZE;
599 }
600 
601 static int nldev_get_doit(struct sk_buff *skb, struct nlmsghdr *nlh,
602 			  struct netlink_ext_ack *extack)
603 {
604 	struct nlattr *tb[RDMA_NLDEV_ATTR_MAX];
605 	struct ib_device *device;
606 	struct sk_buff *msg;
607 	u32 index;
608 	int err;
609 
610 	err = nlmsg_parse(nlh, 0, tb, RDMA_NLDEV_ATTR_MAX - 1,
611 			  nldev_policy, extack);
612 	if (err || !tb[RDMA_NLDEV_ATTR_DEV_INDEX])
613 		return -EINVAL;
614 
615 	index = nla_get_u32(tb[RDMA_NLDEV_ATTR_DEV_INDEX]);
616 
617 	device = ib_device_get_by_index(index);
618 	if (!device)
619 		return -EINVAL;
620 
621 	msg = nlmsg_new(NLMSG_DEFAULT_SIZE, GFP_KERNEL);
622 	if (!msg) {
623 		err = -ENOMEM;
624 		goto err;
625 	}
626 
627 	nlh = nlmsg_put(msg, NETLINK_CB(skb).portid, nlh->nlmsg_seq,
628 			RDMA_NL_GET_TYPE(RDMA_NL_NLDEV, RDMA_NLDEV_CMD_GET),
629 			0, 0);
630 
631 	err = fill_dev_info(msg, device);
632 	if (err)
633 		goto err_free;
634 
635 	nlmsg_end(msg, nlh);
636 
637 	ib_device_put(device);
638 	return rdma_nl_unicast(msg, NETLINK_CB(skb).portid);
639 
640 err_free:
641 	nlmsg_free(msg);
642 err:
643 	ib_device_put(device);
644 	return err;
645 }
646 
647 static int nldev_set_doit(struct sk_buff *skb, struct nlmsghdr *nlh,
648 			  struct netlink_ext_ack *extack)
649 {
650 	struct nlattr *tb[RDMA_NLDEV_ATTR_MAX];
651 	struct ib_device *device;
652 	u32 index;
653 	int err;
654 
655 	err = nlmsg_parse(nlh, 0, tb, RDMA_NLDEV_ATTR_MAX - 1, nldev_policy,
656 			  extack);
657 	if (err || !tb[RDMA_NLDEV_ATTR_DEV_INDEX])
658 		return -EINVAL;
659 
660 	index = nla_get_u32(tb[RDMA_NLDEV_ATTR_DEV_INDEX]);
661 	device = ib_device_get_by_index(index);
662 	if (!device)
663 		return -EINVAL;
664 
665 	if (tb[RDMA_NLDEV_ATTR_DEV_NAME]) {
666 		char name[IB_DEVICE_NAME_MAX] = {};
667 
668 		nla_strlcpy(name, tb[RDMA_NLDEV_ATTR_DEV_NAME],
669 			    IB_DEVICE_NAME_MAX);
670 		err = ib_device_rename(device, name);
671 	}
672 
673 	ib_device_put(device);
674 	return err;
675 }
676 
677 static int _nldev_get_dumpit(struct ib_device *device,
678 			     struct sk_buff *skb,
679 			     struct netlink_callback *cb,
680 			     unsigned int idx)
681 {
682 	int start = cb->args[0];
683 	struct nlmsghdr *nlh;
684 
685 	if (idx < start)
686 		return 0;
687 
688 	nlh = nlmsg_put(skb, NETLINK_CB(cb->skb).portid, cb->nlh->nlmsg_seq,
689 			RDMA_NL_GET_TYPE(RDMA_NL_NLDEV, RDMA_NLDEV_CMD_GET),
690 			0, NLM_F_MULTI);
691 
692 	if (fill_dev_info(skb, device)) {
693 		nlmsg_cancel(skb, nlh);
694 		goto out;
695 	}
696 
697 	nlmsg_end(skb, nlh);
698 
699 	idx++;
700 
701 out:	cb->args[0] = idx;
702 	return skb->len;
703 }
704 
705 static int nldev_get_dumpit(struct sk_buff *skb, struct netlink_callback *cb)
706 {
707 	/*
708 	 * There is no need to take lock, because
709 	 * we are relying on ib_core's lists_rwsem
710 	 */
711 	return ib_enum_all_devs(_nldev_get_dumpit, skb, cb);
712 }
713 
714 static int nldev_port_get_doit(struct sk_buff *skb, struct nlmsghdr *nlh,
715 			       struct netlink_ext_ack *extack)
716 {
717 	struct nlattr *tb[RDMA_NLDEV_ATTR_MAX];
718 	struct ib_device *device;
719 	struct sk_buff *msg;
720 	u32 index;
721 	u32 port;
722 	int err;
723 
724 	err = nlmsg_parse(nlh, 0, tb, RDMA_NLDEV_ATTR_MAX - 1,
725 			  nldev_policy, extack);
726 	if (err ||
727 	    !tb[RDMA_NLDEV_ATTR_DEV_INDEX] ||
728 	    !tb[RDMA_NLDEV_ATTR_PORT_INDEX])
729 		return -EINVAL;
730 
731 	index = nla_get_u32(tb[RDMA_NLDEV_ATTR_DEV_INDEX]);
732 	device = ib_device_get_by_index(index);
733 	if (!device)
734 		return -EINVAL;
735 
736 	port = nla_get_u32(tb[RDMA_NLDEV_ATTR_PORT_INDEX]);
737 	if (!rdma_is_port_valid(device, port)) {
738 		err = -EINVAL;
739 		goto err;
740 	}
741 
742 	msg = nlmsg_new(NLMSG_DEFAULT_SIZE, GFP_KERNEL);
743 	if (!msg) {
744 		err = -ENOMEM;
745 		goto err;
746 	}
747 
748 	nlh = nlmsg_put(msg, NETLINK_CB(skb).portid, nlh->nlmsg_seq,
749 			RDMA_NL_GET_TYPE(RDMA_NL_NLDEV, RDMA_NLDEV_CMD_GET),
750 			0, 0);
751 
752 	err = fill_port_info(msg, device, port, sock_net(skb->sk));
753 	if (err)
754 		goto err_free;
755 
756 	nlmsg_end(msg, nlh);
757 	ib_device_put(device);
758 
759 	return rdma_nl_unicast(msg, NETLINK_CB(skb).portid);
760 
761 err_free:
762 	nlmsg_free(msg);
763 err:
764 	ib_device_put(device);
765 	return err;
766 }
767 
768 static int nldev_port_get_dumpit(struct sk_buff *skb,
769 				 struct netlink_callback *cb)
770 {
771 	struct nlattr *tb[RDMA_NLDEV_ATTR_MAX];
772 	struct ib_device *device;
773 	int start = cb->args[0];
774 	struct nlmsghdr *nlh;
775 	u32 idx = 0;
776 	u32 ifindex;
777 	int err;
778 	unsigned int p;
779 
780 	err = nlmsg_parse(cb->nlh, 0, tb, RDMA_NLDEV_ATTR_MAX - 1,
781 			  nldev_policy, NULL);
782 	if (err || !tb[RDMA_NLDEV_ATTR_DEV_INDEX])
783 		return -EINVAL;
784 
785 	ifindex = nla_get_u32(tb[RDMA_NLDEV_ATTR_DEV_INDEX]);
786 	device = ib_device_get_by_index(ifindex);
787 	if (!device)
788 		return -EINVAL;
789 
790 	rdma_for_each_port (device, p) {
791 		/*
792 		 * The dumpit function returns all information from specific
793 		 * index. This specific index is taken from the netlink
794 		 * messages request sent by user and it is available
795 		 * in cb->args[0].
796 		 *
797 		 * Usually, the user doesn't fill this field and it causes
798 		 * to return everything.
799 		 *
800 		 */
801 		if (idx < start) {
802 			idx++;
803 			continue;
804 		}
805 
806 		nlh = nlmsg_put(skb, NETLINK_CB(cb->skb).portid,
807 				cb->nlh->nlmsg_seq,
808 				RDMA_NL_GET_TYPE(RDMA_NL_NLDEV,
809 						 RDMA_NLDEV_CMD_PORT_GET),
810 				0, NLM_F_MULTI);
811 
812 		if (fill_port_info(skb, device, p, sock_net(skb->sk))) {
813 			nlmsg_cancel(skb, nlh);
814 			goto out;
815 		}
816 		idx++;
817 		nlmsg_end(skb, nlh);
818 	}
819 
820 out:
821 	ib_device_put(device);
822 	cb->args[0] = idx;
823 	return skb->len;
824 }
825 
826 static int nldev_res_get_doit(struct sk_buff *skb, struct nlmsghdr *nlh,
827 			      struct netlink_ext_ack *extack)
828 {
829 	struct nlattr *tb[RDMA_NLDEV_ATTR_MAX];
830 	struct ib_device *device;
831 	struct sk_buff *msg;
832 	u32 index;
833 	int ret;
834 
835 	ret = nlmsg_parse(nlh, 0, tb, RDMA_NLDEV_ATTR_MAX - 1,
836 			  nldev_policy, extack);
837 	if (ret || !tb[RDMA_NLDEV_ATTR_DEV_INDEX])
838 		return -EINVAL;
839 
840 	index = nla_get_u32(tb[RDMA_NLDEV_ATTR_DEV_INDEX]);
841 	device = ib_device_get_by_index(index);
842 	if (!device)
843 		return -EINVAL;
844 
845 	msg = nlmsg_new(NLMSG_DEFAULT_SIZE, GFP_KERNEL);
846 	if (!msg) {
847 		ret = -ENOMEM;
848 		goto err;
849 	}
850 
851 	nlh = nlmsg_put(msg, NETLINK_CB(skb).portid, nlh->nlmsg_seq,
852 			RDMA_NL_GET_TYPE(RDMA_NL_NLDEV, RDMA_NLDEV_CMD_RES_GET),
853 			0, 0);
854 
855 	ret = fill_res_info(msg, device);
856 	if (ret)
857 		goto err_free;
858 
859 	nlmsg_end(msg, nlh);
860 	ib_device_put(device);
861 	return rdma_nl_unicast(msg, NETLINK_CB(skb).portid);
862 
863 err_free:
864 	nlmsg_free(msg);
865 err:
866 	ib_device_put(device);
867 	return ret;
868 }
869 
870 static int _nldev_res_get_dumpit(struct ib_device *device,
871 				 struct sk_buff *skb,
872 				 struct netlink_callback *cb,
873 				 unsigned int idx)
874 {
875 	int start = cb->args[0];
876 	struct nlmsghdr *nlh;
877 
878 	if (idx < start)
879 		return 0;
880 
881 	nlh = nlmsg_put(skb, NETLINK_CB(cb->skb).portid, cb->nlh->nlmsg_seq,
882 			RDMA_NL_GET_TYPE(RDMA_NL_NLDEV, RDMA_NLDEV_CMD_RES_GET),
883 			0, NLM_F_MULTI);
884 
885 	if (fill_res_info(skb, device)) {
886 		nlmsg_cancel(skb, nlh);
887 		goto out;
888 	}
889 
890 	nlmsg_end(skb, nlh);
891 
892 	idx++;
893 
894 out:
895 	cb->args[0] = idx;
896 	return skb->len;
897 }
898 
899 static int nldev_res_get_dumpit(struct sk_buff *skb,
900 				struct netlink_callback *cb)
901 {
902 	return ib_enum_all_devs(_nldev_res_get_dumpit, skb, cb);
903 }
904 
905 struct nldev_fill_res_entry {
906 	int (*fill_res_func)(struct sk_buff *msg, bool has_cap_net_admin,
907 			     struct rdma_restrack_entry *res, u32 port);
908 	enum rdma_nldev_attr nldev_attr;
909 	enum rdma_nldev_command nldev_cmd;
910 	u8 flags;
911 	u32 entry;
912 	u32 id;
913 };
914 
915 enum nldev_res_flags {
916 	NLDEV_PER_DEV = 1 << 0,
917 };
918 
919 static const struct nldev_fill_res_entry fill_entries[RDMA_RESTRACK_MAX] = {
920 	[RDMA_RESTRACK_QP] = {
921 		.fill_res_func = fill_res_qp_entry,
922 		.nldev_cmd = RDMA_NLDEV_CMD_RES_QP_GET,
923 		.nldev_attr = RDMA_NLDEV_ATTR_RES_QP,
924 		.entry = RDMA_NLDEV_ATTR_RES_QP_ENTRY,
925 		.id = RDMA_NLDEV_ATTR_RES_LQPN,
926 	},
927 	[RDMA_RESTRACK_CM_ID] = {
928 		.fill_res_func = fill_res_cm_id_entry,
929 		.nldev_cmd = RDMA_NLDEV_CMD_RES_CM_ID_GET,
930 		.nldev_attr = RDMA_NLDEV_ATTR_RES_CM_ID,
931 		.entry = RDMA_NLDEV_ATTR_RES_CM_ID_ENTRY,
932 		.id = RDMA_NLDEV_ATTR_RES_CM_IDN,
933 	},
934 	[RDMA_RESTRACK_CQ] = {
935 		.fill_res_func = fill_res_cq_entry,
936 		.nldev_cmd = RDMA_NLDEV_CMD_RES_CQ_GET,
937 		.nldev_attr = RDMA_NLDEV_ATTR_RES_CQ,
938 		.flags = NLDEV_PER_DEV,
939 		.entry = RDMA_NLDEV_ATTR_RES_CQ_ENTRY,
940 		.id = RDMA_NLDEV_ATTR_RES_CQN,
941 	},
942 	[RDMA_RESTRACK_MR] = {
943 		.fill_res_func = fill_res_mr_entry,
944 		.nldev_cmd = RDMA_NLDEV_CMD_RES_MR_GET,
945 		.nldev_attr = RDMA_NLDEV_ATTR_RES_MR,
946 		.flags = NLDEV_PER_DEV,
947 		.entry = RDMA_NLDEV_ATTR_RES_MR_ENTRY,
948 		.id = RDMA_NLDEV_ATTR_RES_MRN,
949 	},
950 	[RDMA_RESTRACK_PD] = {
951 		.fill_res_func = fill_res_pd_entry,
952 		.nldev_cmd = RDMA_NLDEV_CMD_RES_PD_GET,
953 		.nldev_attr = RDMA_NLDEV_ATTR_RES_PD,
954 		.flags = NLDEV_PER_DEV,
955 		.entry = RDMA_NLDEV_ATTR_RES_PD_ENTRY,
956 		.id = RDMA_NLDEV_ATTR_RES_PDN,
957 	},
958 };
959 
960 static bool is_visible_in_pid_ns(struct rdma_restrack_entry *res)
961 {
962 	/*
963 	 * 1. Kern resources should be visible in init name space only
964 	 * 2. Present only resources visible in the current namespace
965 	 */
966 	if (rdma_is_kernel_res(res))
967 		return task_active_pid_ns(current) == &init_pid_ns;
968 	return task_active_pid_ns(current) == task_active_pid_ns(res->task);
969 }
970 
971 static int res_get_common_doit(struct sk_buff *skb, struct nlmsghdr *nlh,
972 			       struct netlink_ext_ack *extack,
973 			       enum rdma_restrack_type res_type)
974 {
975 	const struct nldev_fill_res_entry *fe = &fill_entries[res_type];
976 	struct nlattr *tb[RDMA_NLDEV_ATTR_MAX];
977 	struct rdma_restrack_entry *res;
978 	struct ib_device *device;
979 	u32 index, id, port = 0;
980 	bool has_cap_net_admin;
981 	struct sk_buff *msg;
982 	int ret;
983 
984 	ret = nlmsg_parse(nlh, 0, tb, RDMA_NLDEV_ATTR_MAX - 1,
985 			  nldev_policy, extack);
986 	if (ret || !tb[RDMA_NLDEV_ATTR_DEV_INDEX] || !fe->id || !tb[fe->id])
987 		return -EINVAL;
988 
989 	index = nla_get_u32(tb[RDMA_NLDEV_ATTR_DEV_INDEX]);
990 	device = ib_device_get_by_index(index);
991 	if (!device)
992 		return -EINVAL;
993 
994 	if (tb[RDMA_NLDEV_ATTR_PORT_INDEX]) {
995 		port = nla_get_u32(tb[RDMA_NLDEV_ATTR_PORT_INDEX]);
996 		if (!rdma_is_port_valid(device, port)) {
997 			ret = -EINVAL;
998 			goto err;
999 		}
1000 	}
1001 
1002 	if ((port && fe->flags & NLDEV_PER_DEV) ||
1003 	    (!port && ~fe->flags & NLDEV_PER_DEV)) {
1004 		ret = -EINVAL;
1005 		goto err;
1006 	}
1007 
1008 	id = nla_get_u32(tb[fe->id]);
1009 	res = rdma_restrack_get_byid(device, res_type, id);
1010 	if (IS_ERR(res)) {
1011 		ret = PTR_ERR(res);
1012 		goto err;
1013 	}
1014 
1015 	if (!is_visible_in_pid_ns(res)) {
1016 		ret = -ENOENT;
1017 		goto err_get;
1018 	}
1019 
1020 	msg = nlmsg_new(NLMSG_DEFAULT_SIZE, GFP_KERNEL);
1021 	if (!msg) {
1022 		ret = -ENOMEM;
1023 		goto err;
1024 	}
1025 
1026 	nlh = nlmsg_put(msg, NETLINK_CB(skb).portid, nlh->nlmsg_seq,
1027 			RDMA_NL_GET_TYPE(RDMA_NL_NLDEV, fe->nldev_cmd),
1028 			0, 0);
1029 
1030 	if (fill_nldev_handle(msg, device)) {
1031 		ret = -EMSGSIZE;
1032 		goto err_free;
1033 	}
1034 
1035 	has_cap_net_admin = netlink_capable(skb, CAP_NET_ADMIN);
1036 	ret = fe->fill_res_func(msg, has_cap_net_admin, res, port);
1037 	rdma_restrack_put(res);
1038 	if (ret)
1039 		goto err_free;
1040 
1041 	nlmsg_end(msg, nlh);
1042 	ib_device_put(device);
1043 	return rdma_nl_unicast(msg, NETLINK_CB(skb).portid);
1044 
1045 err_free:
1046 	nlmsg_free(msg);
1047 err_get:
1048 	rdma_restrack_put(res);
1049 err:
1050 	ib_device_put(device);
1051 	return ret;
1052 }
1053 
1054 static int res_get_common_dumpit(struct sk_buff *skb,
1055 				 struct netlink_callback *cb,
1056 				 enum rdma_restrack_type res_type)
1057 {
1058 	const struct nldev_fill_res_entry *fe = &fill_entries[res_type];
1059 	struct nlattr *tb[RDMA_NLDEV_ATTR_MAX];
1060 	struct rdma_restrack_entry *res;
1061 	struct rdma_restrack_root *rt;
1062 	int err, ret = 0, idx = 0;
1063 	struct nlattr *table_attr;
1064 	struct nlattr *entry_attr;
1065 	struct ib_device *device;
1066 	int start = cb->args[0];
1067 	bool has_cap_net_admin;
1068 	struct nlmsghdr *nlh;
1069 	unsigned long id;
1070 	u32 index, port = 0;
1071 	bool filled = false;
1072 
1073 	err = nlmsg_parse(cb->nlh, 0, tb, RDMA_NLDEV_ATTR_MAX - 1,
1074 			  nldev_policy, NULL);
1075 	/*
1076 	 * Right now, we are expecting the device index to get res information,
1077 	 * but it is possible to extend this code to return all devices in
1078 	 * one shot by checking the existence of RDMA_NLDEV_ATTR_DEV_INDEX.
1079 	 * if it doesn't exist, we will iterate over all devices.
1080 	 *
1081 	 * But it is not needed for now.
1082 	 */
1083 	if (err || !tb[RDMA_NLDEV_ATTR_DEV_INDEX])
1084 		return -EINVAL;
1085 
1086 	index = nla_get_u32(tb[RDMA_NLDEV_ATTR_DEV_INDEX]);
1087 	device = ib_device_get_by_index(index);
1088 	if (!device)
1089 		return -EINVAL;
1090 
1091 	/*
1092 	 * If no PORT_INDEX is supplied, we will return all QPs from that device
1093 	 */
1094 	if (tb[RDMA_NLDEV_ATTR_PORT_INDEX]) {
1095 		port = nla_get_u32(tb[RDMA_NLDEV_ATTR_PORT_INDEX]);
1096 		if (!rdma_is_port_valid(device, port)) {
1097 			ret = -EINVAL;
1098 			goto err_index;
1099 		}
1100 	}
1101 
1102 	nlh = nlmsg_put(skb, NETLINK_CB(cb->skb).portid, cb->nlh->nlmsg_seq,
1103 			RDMA_NL_GET_TYPE(RDMA_NL_NLDEV, fe->nldev_cmd),
1104 			0, NLM_F_MULTI);
1105 
1106 	if (fill_nldev_handle(skb, device)) {
1107 		ret = -EMSGSIZE;
1108 		goto err;
1109 	}
1110 
1111 	table_attr = nla_nest_start(skb, fe->nldev_attr);
1112 	if (!table_attr) {
1113 		ret = -EMSGSIZE;
1114 		goto err;
1115 	}
1116 
1117 	has_cap_net_admin = netlink_capable(cb->skb, CAP_NET_ADMIN);
1118 
1119 	rt = &device->res[res_type];
1120 	xa_lock(&rt->xa);
1121 	/*
1122 	 * FIXME: if the skip ahead is something common this loop should
1123 	 * use xas_for_each & xas_pause to optimize, we can have a lot of
1124 	 * objects.
1125 	 */
1126 	xa_for_each(&rt->xa, id, res) {
1127 		if (!is_visible_in_pid_ns(res))
1128 			continue;
1129 
1130 		if (idx < start || !rdma_restrack_get(res))
1131 			goto next;
1132 
1133 		xa_unlock(&rt->xa);
1134 
1135 		filled = true;
1136 
1137 		entry_attr = nla_nest_start(skb, fe->entry);
1138 		if (!entry_attr) {
1139 			ret = -EMSGSIZE;
1140 			rdma_restrack_put(res);
1141 			goto msg_full;
1142 		}
1143 
1144 		ret = fe->fill_res_func(skb, has_cap_net_admin, res, port);
1145 		rdma_restrack_put(res);
1146 
1147 		if (ret) {
1148 			nla_nest_cancel(skb, entry_attr);
1149 			if (ret == -EMSGSIZE)
1150 				goto msg_full;
1151 			if (ret == -EAGAIN)
1152 				goto again;
1153 			goto res_err;
1154 		}
1155 		nla_nest_end(skb, entry_attr);
1156 again:		xa_lock(&rt->xa);
1157 next:		idx++;
1158 	}
1159 	xa_unlock(&rt->xa);
1160 
1161 msg_full:
1162 	nla_nest_end(skb, table_attr);
1163 	nlmsg_end(skb, nlh);
1164 	cb->args[0] = idx;
1165 
1166 	/*
1167 	 * No more entries to fill, cancel the message and
1168 	 * return 0 to mark end of dumpit.
1169 	 */
1170 	if (!filled)
1171 		goto err;
1172 
1173 	ib_device_put(device);
1174 	return skb->len;
1175 
1176 res_err:
1177 	nla_nest_cancel(skb, table_attr);
1178 
1179 err:
1180 	nlmsg_cancel(skb, nlh);
1181 
1182 err_index:
1183 	ib_device_put(device);
1184 	return ret;
1185 }
1186 
1187 #define RES_GET_FUNCS(name, type)                                              \
1188 	static int nldev_res_get_##name##_dumpit(struct sk_buff *skb,          \
1189 						 struct netlink_callback *cb)  \
1190 	{                                                                      \
1191 		return res_get_common_dumpit(skb, cb, type);                   \
1192 	}                                                                      \
1193 	static int nldev_res_get_##name##_doit(struct sk_buff *skb,            \
1194 					       struct nlmsghdr *nlh,           \
1195 					       struct netlink_ext_ack *extack) \
1196 	{                                                                      \
1197 		return res_get_common_doit(skb, nlh, extack, type);            \
1198 	}
1199 
1200 RES_GET_FUNCS(qp, RDMA_RESTRACK_QP);
1201 RES_GET_FUNCS(cm_id, RDMA_RESTRACK_CM_ID);
1202 RES_GET_FUNCS(cq, RDMA_RESTRACK_CQ);
1203 RES_GET_FUNCS(pd, RDMA_RESTRACK_PD);
1204 RES_GET_FUNCS(mr, RDMA_RESTRACK_MR);
1205 
1206 static LIST_HEAD(link_ops);
1207 static DECLARE_RWSEM(link_ops_rwsem);
1208 
1209 static const struct rdma_link_ops *link_ops_get(const char *type)
1210 {
1211 	const struct rdma_link_ops *ops;
1212 
1213 	list_for_each_entry(ops, &link_ops, list) {
1214 		if (!strcmp(ops->type, type))
1215 			goto out;
1216 	}
1217 	ops = NULL;
1218 out:
1219 	return ops;
1220 }
1221 
1222 void rdma_link_register(struct rdma_link_ops *ops)
1223 {
1224 	down_write(&link_ops_rwsem);
1225 	if (WARN_ON_ONCE(link_ops_get(ops->type)))
1226 		goto out;
1227 	list_add(&ops->list, &link_ops);
1228 out:
1229 	up_write(&link_ops_rwsem);
1230 }
1231 EXPORT_SYMBOL(rdma_link_register);
1232 
1233 void rdma_link_unregister(struct rdma_link_ops *ops)
1234 {
1235 	down_write(&link_ops_rwsem);
1236 	list_del(&ops->list);
1237 	up_write(&link_ops_rwsem);
1238 }
1239 EXPORT_SYMBOL(rdma_link_unregister);
1240 
1241 static int nldev_newlink(struct sk_buff *skb, struct nlmsghdr *nlh,
1242 			  struct netlink_ext_ack *extack)
1243 {
1244 	struct nlattr *tb[RDMA_NLDEV_ATTR_MAX];
1245 	char ibdev_name[IB_DEVICE_NAME_MAX];
1246 	const struct rdma_link_ops *ops;
1247 	char ndev_name[IFNAMSIZ];
1248 	struct net_device *ndev;
1249 	char type[IFNAMSIZ];
1250 	int err;
1251 
1252 	err = nlmsg_parse(nlh, 0, tb, RDMA_NLDEV_ATTR_MAX - 1,
1253 			  nldev_policy, extack);
1254 	if (err || !tb[RDMA_NLDEV_ATTR_DEV_NAME] ||
1255 	    !tb[RDMA_NLDEV_ATTR_LINK_TYPE] || !tb[RDMA_NLDEV_ATTR_NDEV_NAME])
1256 		return -EINVAL;
1257 
1258 	nla_strlcpy(ibdev_name, tb[RDMA_NLDEV_ATTR_DEV_NAME],
1259 		    sizeof(ibdev_name));
1260 	if (strchr(ibdev_name, '%'))
1261 		return -EINVAL;
1262 
1263 	nla_strlcpy(type, tb[RDMA_NLDEV_ATTR_LINK_TYPE], sizeof(type));
1264 	nla_strlcpy(ndev_name, tb[RDMA_NLDEV_ATTR_NDEV_NAME],
1265 		    sizeof(ndev_name));
1266 
1267 	ndev = dev_get_by_name(&init_net, ndev_name);
1268 	if (!ndev)
1269 		return -ENODEV;
1270 
1271 	down_read(&link_ops_rwsem);
1272 	ops = link_ops_get(type);
1273 #ifdef CONFIG_MODULES
1274 	if (!ops) {
1275 		up_read(&link_ops_rwsem);
1276 		request_module("rdma-link-%s", type);
1277 		down_read(&link_ops_rwsem);
1278 		ops = link_ops_get(type);
1279 	}
1280 #endif
1281 	err = ops ? ops->newlink(ibdev_name, ndev) : -EINVAL;
1282 	up_read(&link_ops_rwsem);
1283 	dev_put(ndev);
1284 
1285 	return err;
1286 }
1287 
1288 static int nldev_dellink(struct sk_buff *skb, struct nlmsghdr *nlh,
1289 			  struct netlink_ext_ack *extack)
1290 {
1291 	struct nlattr *tb[RDMA_NLDEV_ATTR_MAX];
1292 	struct ib_device *device;
1293 	u32 index;
1294 	int err;
1295 
1296 	err = nlmsg_parse(nlh, 0, tb, RDMA_NLDEV_ATTR_MAX - 1,
1297 			  nldev_policy, extack);
1298 	if (err || !tb[RDMA_NLDEV_ATTR_DEV_INDEX])
1299 		return -EINVAL;
1300 
1301 	index = nla_get_u32(tb[RDMA_NLDEV_ATTR_DEV_INDEX]);
1302 	device = ib_device_get_by_index(index);
1303 	if (!device)
1304 		return -EINVAL;
1305 
1306 	if (!(device->attrs.device_cap_flags & IB_DEVICE_ALLOW_USER_UNREG)) {
1307 		ib_device_put(device);
1308 		return -EINVAL;
1309 	}
1310 
1311 	ib_unregister_device_and_put(device);
1312 	return 0;
1313 }
1314 
1315 static const struct rdma_nl_cbs nldev_cb_table[RDMA_NLDEV_NUM_OPS] = {
1316 	[RDMA_NLDEV_CMD_GET] = {
1317 		.doit = nldev_get_doit,
1318 		.dump = nldev_get_dumpit,
1319 	},
1320 	[RDMA_NLDEV_CMD_SET] = {
1321 		.doit = nldev_set_doit,
1322 		.flags = RDMA_NL_ADMIN_PERM,
1323 	},
1324 	[RDMA_NLDEV_CMD_NEWLINK] = {
1325 		.doit = nldev_newlink,
1326 		.flags = RDMA_NL_ADMIN_PERM,
1327 	},
1328 	[RDMA_NLDEV_CMD_DELLINK] = {
1329 		.doit = nldev_dellink,
1330 		.flags = RDMA_NL_ADMIN_PERM,
1331 	},
1332 	[RDMA_NLDEV_CMD_PORT_GET] = {
1333 		.doit = nldev_port_get_doit,
1334 		.dump = nldev_port_get_dumpit,
1335 	},
1336 	[RDMA_NLDEV_CMD_RES_GET] = {
1337 		.doit = nldev_res_get_doit,
1338 		.dump = nldev_res_get_dumpit,
1339 	},
1340 	[RDMA_NLDEV_CMD_RES_QP_GET] = {
1341 		.doit = nldev_res_get_qp_doit,
1342 		.dump = nldev_res_get_qp_dumpit,
1343 	},
1344 	[RDMA_NLDEV_CMD_RES_CM_ID_GET] = {
1345 		.doit = nldev_res_get_cm_id_doit,
1346 		.dump = nldev_res_get_cm_id_dumpit,
1347 	},
1348 	[RDMA_NLDEV_CMD_RES_CQ_GET] = {
1349 		.doit = nldev_res_get_cq_doit,
1350 		.dump = nldev_res_get_cq_dumpit,
1351 	},
1352 	[RDMA_NLDEV_CMD_RES_MR_GET] = {
1353 		.doit = nldev_res_get_mr_doit,
1354 		.dump = nldev_res_get_mr_dumpit,
1355 	},
1356 	[RDMA_NLDEV_CMD_RES_PD_GET] = {
1357 		.doit = nldev_res_get_pd_doit,
1358 		.dump = nldev_res_get_pd_dumpit,
1359 	},
1360 };
1361 
1362 void __init nldev_init(void)
1363 {
1364 	rdma_nl_register(RDMA_NL_NLDEV, nldev_cb_table);
1365 }
1366 
1367 void __exit nldev_exit(void)
1368 {
1369 	rdma_nl_unregister(RDMA_NL_NLDEV);
1370 }
1371 
1372 MODULE_ALIAS_RDMA_NETLINK(RDMA_NL_NLDEV, 5);
1373