xref: /openbmc/linux/drivers/infiniband/core/nldev.c (revision dd3cb467)
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 #include "uverbs.h"
45 
46 typedef int (*res_fill_func_t)(struct sk_buff*, bool,
47 			       struct rdma_restrack_entry*, uint32_t);
48 
49 /*
50  * Sort array elements by the netlink attribute name
51  */
52 static const struct nla_policy nldev_policy[RDMA_NLDEV_ATTR_MAX] = {
53 	[RDMA_NLDEV_ATTR_CHARDEV]		= { .type = NLA_U64 },
54 	[RDMA_NLDEV_ATTR_CHARDEV_ABI]		= { .type = NLA_U64 },
55 	[RDMA_NLDEV_ATTR_CHARDEV_NAME]		= { .type = NLA_NUL_STRING,
56 					.len = RDMA_NLDEV_ATTR_EMPTY_STRING },
57 	[RDMA_NLDEV_ATTR_CHARDEV_TYPE]		= { .type = NLA_NUL_STRING,
58 					.len = RDMA_NLDEV_ATTR_CHARDEV_TYPE_SIZE },
59 	[RDMA_NLDEV_ATTR_DEV_DIM]               = { .type = NLA_U8 },
60 	[RDMA_NLDEV_ATTR_DEV_INDEX]		= { .type = NLA_U32 },
61 	[RDMA_NLDEV_ATTR_DEV_NAME]		= { .type = NLA_NUL_STRING,
62 					.len = IB_DEVICE_NAME_MAX },
63 	[RDMA_NLDEV_ATTR_DEV_NODE_TYPE]		= { .type = NLA_U8 },
64 	[RDMA_NLDEV_ATTR_DEV_PROTOCOL]		= { .type = NLA_NUL_STRING,
65 					.len = RDMA_NLDEV_ATTR_EMPTY_STRING },
66 	[RDMA_NLDEV_ATTR_DRIVER]		= { .type = NLA_NESTED },
67 	[RDMA_NLDEV_ATTR_DRIVER_ENTRY]		= { .type = NLA_NESTED },
68 	[RDMA_NLDEV_ATTR_DRIVER_PRINT_TYPE]	= { .type = NLA_U8 },
69 	[RDMA_NLDEV_ATTR_DRIVER_STRING]		= { .type = NLA_NUL_STRING,
70 					.len = RDMA_NLDEV_ATTR_EMPTY_STRING },
71 	[RDMA_NLDEV_ATTR_DRIVER_S32]		= { .type = NLA_S32 },
72 	[RDMA_NLDEV_ATTR_DRIVER_S64]		= { .type = NLA_S64 },
73 	[RDMA_NLDEV_ATTR_DRIVER_U32]		= { .type = NLA_U32 },
74 	[RDMA_NLDEV_ATTR_DRIVER_U64]		= { .type = NLA_U64 },
75 	[RDMA_NLDEV_ATTR_FW_VERSION]		= { .type = NLA_NUL_STRING,
76 					.len = RDMA_NLDEV_ATTR_EMPTY_STRING },
77 	[RDMA_NLDEV_ATTR_LID]			= { .type = NLA_U32 },
78 	[RDMA_NLDEV_ATTR_LINK_TYPE]		= { .type = NLA_NUL_STRING,
79 					.len = IFNAMSIZ },
80 	[RDMA_NLDEV_ATTR_LMC]			= { .type = NLA_U8 },
81 	[RDMA_NLDEV_ATTR_NDEV_INDEX]		= { .type = NLA_U32 },
82 	[RDMA_NLDEV_ATTR_NDEV_NAME]		= { .type = NLA_NUL_STRING,
83 					.len = IFNAMSIZ },
84 	[RDMA_NLDEV_ATTR_NODE_GUID]		= { .type = NLA_U64 },
85 	[RDMA_NLDEV_ATTR_PORT_INDEX]		= { .type = NLA_U32 },
86 	[RDMA_NLDEV_ATTR_PORT_PHYS_STATE]	= { .type = NLA_U8 },
87 	[RDMA_NLDEV_ATTR_PORT_STATE]		= { .type = NLA_U8 },
88 	[RDMA_NLDEV_ATTR_RES_CM_ID]		= { .type = NLA_NESTED },
89 	[RDMA_NLDEV_ATTR_RES_CM_IDN]		= { .type = NLA_U32 },
90 	[RDMA_NLDEV_ATTR_RES_CM_ID_ENTRY]	= { .type = NLA_NESTED },
91 	[RDMA_NLDEV_ATTR_RES_CQ]		= { .type = NLA_NESTED },
92 	[RDMA_NLDEV_ATTR_RES_CQE]		= { .type = NLA_U32 },
93 	[RDMA_NLDEV_ATTR_RES_CQN]		= { .type = NLA_U32 },
94 	[RDMA_NLDEV_ATTR_RES_CQ_ENTRY]		= { .type = NLA_NESTED },
95 	[RDMA_NLDEV_ATTR_RES_CTX]		= { .type = NLA_NESTED },
96 	[RDMA_NLDEV_ATTR_RES_CTXN]		= { .type = NLA_U32 },
97 	[RDMA_NLDEV_ATTR_RES_CTX_ENTRY]		= { .type = NLA_NESTED },
98 	[RDMA_NLDEV_ATTR_RES_DST_ADDR]		= {
99 			.len = sizeof(struct __kernel_sockaddr_storage) },
100 	[RDMA_NLDEV_ATTR_RES_IOVA]		= { .type = NLA_U64 },
101 	[RDMA_NLDEV_ATTR_RES_KERN_NAME]		= { .type = NLA_NUL_STRING,
102 					.len = RDMA_NLDEV_ATTR_EMPTY_STRING },
103 	[RDMA_NLDEV_ATTR_RES_LKEY]		= { .type = NLA_U32 },
104 	[RDMA_NLDEV_ATTR_RES_LOCAL_DMA_LKEY]	= { .type = NLA_U32 },
105 	[RDMA_NLDEV_ATTR_RES_LQPN]		= { .type = NLA_U32 },
106 	[RDMA_NLDEV_ATTR_RES_MR]		= { .type = NLA_NESTED },
107 	[RDMA_NLDEV_ATTR_RES_MRLEN]		= { .type = NLA_U64 },
108 	[RDMA_NLDEV_ATTR_RES_MRN]		= { .type = NLA_U32 },
109 	[RDMA_NLDEV_ATTR_RES_MR_ENTRY]		= { .type = NLA_NESTED },
110 	[RDMA_NLDEV_ATTR_RES_PATH_MIG_STATE]	= { .type = NLA_U8 },
111 	[RDMA_NLDEV_ATTR_RES_PD]		= { .type = NLA_NESTED },
112 	[RDMA_NLDEV_ATTR_RES_PDN]		= { .type = NLA_U32 },
113 	[RDMA_NLDEV_ATTR_RES_PD_ENTRY]		= { .type = NLA_NESTED },
114 	[RDMA_NLDEV_ATTR_RES_PID]		= { .type = NLA_U32 },
115 	[RDMA_NLDEV_ATTR_RES_POLL_CTX]		= { .type = NLA_U8 },
116 	[RDMA_NLDEV_ATTR_RES_PS]		= { .type = NLA_U32 },
117 	[RDMA_NLDEV_ATTR_RES_QP]		= { .type = NLA_NESTED },
118 	[RDMA_NLDEV_ATTR_RES_QP_ENTRY]		= { .type = NLA_NESTED },
119 	[RDMA_NLDEV_ATTR_RES_RAW]		= { .type = NLA_BINARY },
120 	[RDMA_NLDEV_ATTR_RES_RKEY]		= { .type = NLA_U32 },
121 	[RDMA_NLDEV_ATTR_RES_RQPN]		= { .type = NLA_U32 },
122 	[RDMA_NLDEV_ATTR_RES_RQ_PSN]		= { .type = NLA_U32 },
123 	[RDMA_NLDEV_ATTR_RES_SQ_PSN]		= { .type = NLA_U32 },
124 	[RDMA_NLDEV_ATTR_RES_SRC_ADDR]		= {
125 			.len = sizeof(struct __kernel_sockaddr_storage) },
126 	[RDMA_NLDEV_ATTR_RES_STATE]		= { .type = NLA_U8 },
127 	[RDMA_NLDEV_ATTR_RES_SUMMARY]		= { .type = NLA_NESTED },
128 	[RDMA_NLDEV_ATTR_RES_SUMMARY_ENTRY]	= { .type = NLA_NESTED },
129 	[RDMA_NLDEV_ATTR_RES_SUMMARY_ENTRY_CURR]= { .type = NLA_U64 },
130 	[RDMA_NLDEV_ATTR_RES_SUMMARY_ENTRY_NAME]= { .type = NLA_NUL_STRING,
131 					.len = RDMA_NLDEV_ATTR_EMPTY_STRING },
132 	[RDMA_NLDEV_ATTR_RES_TYPE]		= { .type = NLA_U8 },
133 	[RDMA_NLDEV_ATTR_RES_UNSAFE_GLOBAL_RKEY]= { .type = NLA_U32 },
134 	[RDMA_NLDEV_ATTR_RES_USECNT]		= { .type = NLA_U64 },
135 	[RDMA_NLDEV_ATTR_RES_SRQ]		= { .type = NLA_NESTED },
136 	[RDMA_NLDEV_ATTR_RES_SRQN]		= { .type = NLA_U32 },
137 	[RDMA_NLDEV_ATTR_RES_SRQ_ENTRY]		= { .type = NLA_NESTED },
138 	[RDMA_NLDEV_ATTR_MIN_RANGE]		= { .type = NLA_U32 },
139 	[RDMA_NLDEV_ATTR_MAX_RANGE]		= { .type = NLA_U32 },
140 	[RDMA_NLDEV_ATTR_SM_LID]		= { .type = NLA_U32 },
141 	[RDMA_NLDEV_ATTR_SUBNET_PREFIX]		= { .type = NLA_U64 },
142 	[RDMA_NLDEV_ATTR_STAT_AUTO_MODE_MASK]	= { .type = NLA_U32 },
143 	[RDMA_NLDEV_ATTR_STAT_MODE]		= { .type = NLA_U32 },
144 	[RDMA_NLDEV_ATTR_STAT_RES]		= { .type = NLA_U32 },
145 	[RDMA_NLDEV_ATTR_STAT_COUNTER]		= { .type = NLA_NESTED },
146 	[RDMA_NLDEV_ATTR_STAT_COUNTER_ENTRY]	= { .type = NLA_NESTED },
147 	[RDMA_NLDEV_ATTR_STAT_COUNTER_ID]       = { .type = NLA_U32 },
148 	[RDMA_NLDEV_ATTR_STAT_HWCOUNTERS]       = { .type = NLA_NESTED },
149 	[RDMA_NLDEV_ATTR_STAT_HWCOUNTER_ENTRY]  = { .type = NLA_NESTED },
150 	[RDMA_NLDEV_ATTR_STAT_HWCOUNTER_ENTRY_NAME] = { .type = NLA_NUL_STRING },
151 	[RDMA_NLDEV_ATTR_STAT_HWCOUNTER_ENTRY_VALUE] = { .type = NLA_U64 },
152 	[RDMA_NLDEV_ATTR_SYS_IMAGE_GUID]	= { .type = NLA_U64 },
153 	[RDMA_NLDEV_ATTR_UVERBS_DRIVER_ID]	= { .type = NLA_U32 },
154 	[RDMA_NLDEV_NET_NS_FD]			= { .type = NLA_U32 },
155 	[RDMA_NLDEV_SYS_ATTR_NETNS_MODE]	= { .type = NLA_U8 },
156 	[RDMA_NLDEV_SYS_ATTR_COPY_ON_FORK]	= { .type = NLA_U8 },
157 	[RDMA_NLDEV_ATTR_STAT_HWCOUNTER_INDEX]	= { .type = NLA_U32 },
158 	[RDMA_NLDEV_ATTR_STAT_HWCOUNTER_DYNAMIC] = { .type = NLA_U8 },
159 };
160 
161 static int put_driver_name_print_type(struct sk_buff *msg, const char *name,
162 				      enum rdma_nldev_print_type print_type)
163 {
164 	if (nla_put_string(msg, RDMA_NLDEV_ATTR_DRIVER_STRING, name))
165 		return -EMSGSIZE;
166 	if (print_type != RDMA_NLDEV_PRINT_TYPE_UNSPEC &&
167 	    nla_put_u8(msg, RDMA_NLDEV_ATTR_DRIVER_PRINT_TYPE, print_type))
168 		return -EMSGSIZE;
169 
170 	return 0;
171 }
172 
173 static int _rdma_nl_put_driver_u32(struct sk_buff *msg, const char *name,
174 				   enum rdma_nldev_print_type print_type,
175 				   u32 value)
176 {
177 	if (put_driver_name_print_type(msg, name, print_type))
178 		return -EMSGSIZE;
179 	if (nla_put_u32(msg, RDMA_NLDEV_ATTR_DRIVER_U32, value))
180 		return -EMSGSIZE;
181 
182 	return 0;
183 }
184 
185 static int _rdma_nl_put_driver_u64(struct sk_buff *msg, const char *name,
186 				   enum rdma_nldev_print_type print_type,
187 				   u64 value)
188 {
189 	if (put_driver_name_print_type(msg, name, print_type))
190 		return -EMSGSIZE;
191 	if (nla_put_u64_64bit(msg, RDMA_NLDEV_ATTR_DRIVER_U64, value,
192 			      RDMA_NLDEV_ATTR_PAD))
193 		return -EMSGSIZE;
194 
195 	return 0;
196 }
197 
198 int rdma_nl_put_driver_string(struct sk_buff *msg, const char *name,
199 			      const char *str)
200 {
201 	if (put_driver_name_print_type(msg, name,
202 				       RDMA_NLDEV_PRINT_TYPE_UNSPEC))
203 		return -EMSGSIZE;
204 	if (nla_put_string(msg, RDMA_NLDEV_ATTR_DRIVER_STRING, str))
205 		return -EMSGSIZE;
206 
207 	return 0;
208 }
209 EXPORT_SYMBOL(rdma_nl_put_driver_string);
210 
211 int rdma_nl_put_driver_u32(struct sk_buff *msg, const char *name, u32 value)
212 {
213 	return _rdma_nl_put_driver_u32(msg, name, RDMA_NLDEV_PRINT_TYPE_UNSPEC,
214 				       value);
215 }
216 EXPORT_SYMBOL(rdma_nl_put_driver_u32);
217 
218 int rdma_nl_put_driver_u32_hex(struct sk_buff *msg, const char *name,
219 			       u32 value)
220 {
221 	return _rdma_nl_put_driver_u32(msg, name, RDMA_NLDEV_PRINT_TYPE_HEX,
222 				       value);
223 }
224 EXPORT_SYMBOL(rdma_nl_put_driver_u32_hex);
225 
226 int rdma_nl_put_driver_u64(struct sk_buff *msg, const char *name, u64 value)
227 {
228 	return _rdma_nl_put_driver_u64(msg, name, RDMA_NLDEV_PRINT_TYPE_UNSPEC,
229 				       value);
230 }
231 EXPORT_SYMBOL(rdma_nl_put_driver_u64);
232 
233 int rdma_nl_put_driver_u64_hex(struct sk_buff *msg, const char *name, u64 value)
234 {
235 	return _rdma_nl_put_driver_u64(msg, name, RDMA_NLDEV_PRINT_TYPE_HEX,
236 				       value);
237 }
238 EXPORT_SYMBOL(rdma_nl_put_driver_u64_hex);
239 
240 static int fill_nldev_handle(struct sk_buff *msg, struct ib_device *device)
241 {
242 	if (nla_put_u32(msg, RDMA_NLDEV_ATTR_DEV_INDEX, device->index))
243 		return -EMSGSIZE;
244 	if (nla_put_string(msg, RDMA_NLDEV_ATTR_DEV_NAME,
245 			   dev_name(&device->dev)))
246 		return -EMSGSIZE;
247 
248 	return 0;
249 }
250 
251 static int fill_dev_info(struct sk_buff *msg, struct ib_device *device)
252 {
253 	char fw[IB_FW_VERSION_NAME_MAX];
254 	int ret = 0;
255 	u32 port;
256 
257 	if (fill_nldev_handle(msg, device))
258 		return -EMSGSIZE;
259 
260 	if (nla_put_u32(msg, RDMA_NLDEV_ATTR_PORT_INDEX, rdma_end_port(device)))
261 		return -EMSGSIZE;
262 
263 	BUILD_BUG_ON(sizeof(device->attrs.device_cap_flags) != sizeof(u64));
264 	if (nla_put_u64_64bit(msg, RDMA_NLDEV_ATTR_CAP_FLAGS,
265 			      device->attrs.device_cap_flags,
266 			      RDMA_NLDEV_ATTR_PAD))
267 		return -EMSGSIZE;
268 
269 	ib_get_device_fw_str(device, fw);
270 	/* Device without FW has strlen(fw) = 0 */
271 	if (strlen(fw) && nla_put_string(msg, RDMA_NLDEV_ATTR_FW_VERSION, fw))
272 		return -EMSGSIZE;
273 
274 	if (nla_put_u64_64bit(msg, RDMA_NLDEV_ATTR_NODE_GUID,
275 			      be64_to_cpu(device->node_guid),
276 			      RDMA_NLDEV_ATTR_PAD))
277 		return -EMSGSIZE;
278 	if (nla_put_u64_64bit(msg, RDMA_NLDEV_ATTR_SYS_IMAGE_GUID,
279 			      be64_to_cpu(device->attrs.sys_image_guid),
280 			      RDMA_NLDEV_ATTR_PAD))
281 		return -EMSGSIZE;
282 	if (nla_put_u8(msg, RDMA_NLDEV_ATTR_DEV_NODE_TYPE, device->node_type))
283 		return -EMSGSIZE;
284 	if (nla_put_u8(msg, RDMA_NLDEV_ATTR_DEV_DIM, device->use_cq_dim))
285 		return -EMSGSIZE;
286 
287 	/*
288 	 * Link type is determined on first port and mlx4 device
289 	 * which can potentially have two different link type for the same
290 	 * IB device is considered as better to be avoided in the future,
291 	 */
292 	port = rdma_start_port(device);
293 	if (rdma_cap_opa_mad(device, port))
294 		ret = nla_put_string(msg, RDMA_NLDEV_ATTR_DEV_PROTOCOL, "opa");
295 	else if (rdma_protocol_ib(device, port))
296 		ret = nla_put_string(msg, RDMA_NLDEV_ATTR_DEV_PROTOCOL, "ib");
297 	else if (rdma_protocol_iwarp(device, port))
298 		ret = nla_put_string(msg, RDMA_NLDEV_ATTR_DEV_PROTOCOL, "iw");
299 	else if (rdma_protocol_roce(device, port))
300 		ret = nla_put_string(msg, RDMA_NLDEV_ATTR_DEV_PROTOCOL, "roce");
301 	else if (rdma_protocol_usnic(device, port))
302 		ret = nla_put_string(msg, RDMA_NLDEV_ATTR_DEV_PROTOCOL,
303 				     "usnic");
304 	return ret;
305 }
306 
307 static int fill_port_info(struct sk_buff *msg,
308 			  struct ib_device *device, u32 port,
309 			  const struct net *net)
310 {
311 	struct net_device *netdev = NULL;
312 	struct ib_port_attr attr;
313 	int ret;
314 	u64 cap_flags = 0;
315 
316 	if (fill_nldev_handle(msg, device))
317 		return -EMSGSIZE;
318 
319 	if (nla_put_u32(msg, RDMA_NLDEV_ATTR_PORT_INDEX, port))
320 		return -EMSGSIZE;
321 
322 	ret = ib_query_port(device, port, &attr);
323 	if (ret)
324 		return ret;
325 
326 	if (rdma_protocol_ib(device, port)) {
327 		BUILD_BUG_ON((sizeof(attr.port_cap_flags) +
328 				sizeof(attr.port_cap_flags2)) > sizeof(u64));
329 		cap_flags = attr.port_cap_flags |
330 			((u64)attr.port_cap_flags2 << 32);
331 		if (nla_put_u64_64bit(msg, RDMA_NLDEV_ATTR_CAP_FLAGS,
332 				      cap_flags, RDMA_NLDEV_ATTR_PAD))
333 			return -EMSGSIZE;
334 		if (nla_put_u64_64bit(msg, RDMA_NLDEV_ATTR_SUBNET_PREFIX,
335 				      attr.subnet_prefix, RDMA_NLDEV_ATTR_PAD))
336 			return -EMSGSIZE;
337 		if (nla_put_u32(msg, RDMA_NLDEV_ATTR_LID, attr.lid))
338 			return -EMSGSIZE;
339 		if (nla_put_u32(msg, RDMA_NLDEV_ATTR_SM_LID, attr.sm_lid))
340 			return -EMSGSIZE;
341 		if (nla_put_u8(msg, RDMA_NLDEV_ATTR_LMC, attr.lmc))
342 			return -EMSGSIZE;
343 	}
344 	if (nla_put_u8(msg, RDMA_NLDEV_ATTR_PORT_STATE, attr.state))
345 		return -EMSGSIZE;
346 	if (nla_put_u8(msg, RDMA_NLDEV_ATTR_PORT_PHYS_STATE, attr.phys_state))
347 		return -EMSGSIZE;
348 
349 	netdev = ib_device_get_netdev(device, port);
350 	if (netdev && net_eq(dev_net(netdev), net)) {
351 		ret = nla_put_u32(msg,
352 				  RDMA_NLDEV_ATTR_NDEV_INDEX, netdev->ifindex);
353 		if (ret)
354 			goto out;
355 		ret = nla_put_string(msg,
356 				     RDMA_NLDEV_ATTR_NDEV_NAME, netdev->name);
357 	}
358 
359 out:
360 	if (netdev)
361 		dev_put(netdev);
362 	return ret;
363 }
364 
365 static int fill_res_info_entry(struct sk_buff *msg,
366 			       const char *name, u64 curr)
367 {
368 	struct nlattr *entry_attr;
369 
370 	entry_attr = nla_nest_start_noflag(msg,
371 					   RDMA_NLDEV_ATTR_RES_SUMMARY_ENTRY);
372 	if (!entry_attr)
373 		return -EMSGSIZE;
374 
375 	if (nla_put_string(msg, RDMA_NLDEV_ATTR_RES_SUMMARY_ENTRY_NAME, name))
376 		goto err;
377 	if (nla_put_u64_64bit(msg, RDMA_NLDEV_ATTR_RES_SUMMARY_ENTRY_CURR, curr,
378 			      RDMA_NLDEV_ATTR_PAD))
379 		goto err;
380 
381 	nla_nest_end(msg, entry_attr);
382 	return 0;
383 
384 err:
385 	nla_nest_cancel(msg, entry_attr);
386 	return -EMSGSIZE;
387 }
388 
389 static int fill_res_info(struct sk_buff *msg, struct ib_device *device)
390 {
391 	static const char * const names[RDMA_RESTRACK_MAX] = {
392 		[RDMA_RESTRACK_PD] = "pd",
393 		[RDMA_RESTRACK_CQ] = "cq",
394 		[RDMA_RESTRACK_QP] = "qp",
395 		[RDMA_RESTRACK_CM_ID] = "cm_id",
396 		[RDMA_RESTRACK_MR] = "mr",
397 		[RDMA_RESTRACK_CTX] = "ctx",
398 		[RDMA_RESTRACK_SRQ] = "srq",
399 	};
400 
401 	struct nlattr *table_attr;
402 	int ret, i, curr;
403 
404 	if (fill_nldev_handle(msg, device))
405 		return -EMSGSIZE;
406 
407 	table_attr = nla_nest_start_noflag(msg, RDMA_NLDEV_ATTR_RES_SUMMARY);
408 	if (!table_attr)
409 		return -EMSGSIZE;
410 
411 	for (i = 0; i < RDMA_RESTRACK_MAX; i++) {
412 		if (!names[i])
413 			continue;
414 		curr = rdma_restrack_count(device, i);
415 		ret = fill_res_info_entry(msg, names[i], curr);
416 		if (ret)
417 			goto err;
418 	}
419 
420 	nla_nest_end(msg, table_attr);
421 	return 0;
422 
423 err:
424 	nla_nest_cancel(msg, table_attr);
425 	return ret;
426 }
427 
428 static int fill_res_name_pid(struct sk_buff *msg,
429 			     struct rdma_restrack_entry *res)
430 {
431 	int err = 0;
432 
433 	/*
434 	 * For user resources, user is should read /proc/PID/comm to get the
435 	 * name of the task file.
436 	 */
437 	if (rdma_is_kernel_res(res)) {
438 		err = nla_put_string(msg, RDMA_NLDEV_ATTR_RES_KERN_NAME,
439 				     res->kern_name);
440 	} else {
441 		pid_t pid;
442 
443 		pid = task_pid_vnr(res->task);
444 		/*
445 		 * Task is dead and in zombie state.
446 		 * There is no need to print PID anymore.
447 		 */
448 		if (pid)
449 			/*
450 			 * This part is racy, task can be killed and PID will
451 			 * be zero right here but it is ok, next query won't
452 			 * return PID. We don't promise real-time reflection
453 			 * of SW objects.
454 			 */
455 			err = nla_put_u32(msg, RDMA_NLDEV_ATTR_RES_PID, pid);
456 	}
457 
458 	return err ? -EMSGSIZE : 0;
459 }
460 
461 static int fill_res_qp_entry_query(struct sk_buff *msg,
462 				   struct rdma_restrack_entry *res,
463 				   struct ib_device *dev,
464 				   struct ib_qp *qp)
465 {
466 	struct ib_qp_init_attr qp_init_attr;
467 	struct ib_qp_attr qp_attr;
468 	int ret;
469 
470 	ret = ib_query_qp(qp, &qp_attr, 0, &qp_init_attr);
471 	if (ret)
472 		return ret;
473 
474 	if (qp->qp_type == IB_QPT_RC || qp->qp_type == IB_QPT_UC) {
475 		if (nla_put_u32(msg, RDMA_NLDEV_ATTR_RES_RQPN,
476 				qp_attr.dest_qp_num))
477 			goto err;
478 		if (nla_put_u32(msg, RDMA_NLDEV_ATTR_RES_RQ_PSN,
479 				qp_attr.rq_psn))
480 			goto err;
481 	}
482 
483 	if (nla_put_u32(msg, RDMA_NLDEV_ATTR_RES_SQ_PSN, qp_attr.sq_psn))
484 		goto err;
485 
486 	if (qp->qp_type == IB_QPT_RC || qp->qp_type == IB_QPT_UC ||
487 	    qp->qp_type == IB_QPT_XRC_INI || qp->qp_type == IB_QPT_XRC_TGT) {
488 		if (nla_put_u8(msg, RDMA_NLDEV_ATTR_RES_PATH_MIG_STATE,
489 			       qp_attr.path_mig_state))
490 			goto err;
491 	}
492 	if (nla_put_u8(msg, RDMA_NLDEV_ATTR_RES_TYPE, qp->qp_type))
493 		goto err;
494 	if (nla_put_u8(msg, RDMA_NLDEV_ATTR_RES_STATE, qp_attr.qp_state))
495 		goto err;
496 
497 	if (dev->ops.fill_res_qp_entry)
498 		return dev->ops.fill_res_qp_entry(msg, qp);
499 	return 0;
500 
501 err:	return -EMSGSIZE;
502 }
503 
504 static int fill_res_qp_entry(struct sk_buff *msg, bool has_cap_net_admin,
505 			     struct rdma_restrack_entry *res, uint32_t port)
506 {
507 	struct ib_qp *qp = container_of(res, struct ib_qp, res);
508 	struct ib_device *dev = qp->device;
509 	int ret;
510 
511 	if (port && port != qp->port)
512 		return -EAGAIN;
513 
514 	/* In create_qp() port is not set yet */
515 	if (qp->port && nla_put_u32(msg, RDMA_NLDEV_ATTR_PORT_INDEX, qp->port))
516 		return -EINVAL;
517 
518 	ret = nla_put_u32(msg, RDMA_NLDEV_ATTR_RES_LQPN, qp->qp_num);
519 	if (ret)
520 		return -EMSGSIZE;
521 
522 	if (!rdma_is_kernel_res(res) &&
523 	    nla_put_u32(msg, RDMA_NLDEV_ATTR_RES_PDN, qp->pd->res.id))
524 		return -EMSGSIZE;
525 
526 	ret = fill_res_name_pid(msg, res);
527 	if (ret)
528 		return -EMSGSIZE;
529 
530 	return fill_res_qp_entry_query(msg, res, dev, qp);
531 }
532 
533 static int fill_res_qp_raw_entry(struct sk_buff *msg, bool has_cap_net_admin,
534 				 struct rdma_restrack_entry *res, uint32_t port)
535 {
536 	struct ib_qp *qp = container_of(res, struct ib_qp, res);
537 	struct ib_device *dev = qp->device;
538 
539 	if (port && port != qp->port)
540 		return -EAGAIN;
541 	if (!dev->ops.fill_res_qp_entry_raw)
542 		return -EINVAL;
543 	return dev->ops.fill_res_qp_entry_raw(msg, qp);
544 }
545 
546 static int fill_res_cm_id_entry(struct sk_buff *msg, bool has_cap_net_admin,
547 				struct rdma_restrack_entry *res, uint32_t port)
548 {
549 	struct rdma_id_private *id_priv =
550 				container_of(res, struct rdma_id_private, res);
551 	struct ib_device *dev = id_priv->id.device;
552 	struct rdma_cm_id *cm_id = &id_priv->id;
553 
554 	if (port && port != cm_id->port_num)
555 		return 0;
556 
557 	if (cm_id->port_num &&
558 	    nla_put_u32(msg, RDMA_NLDEV_ATTR_PORT_INDEX, cm_id->port_num))
559 		goto err;
560 
561 	if (id_priv->qp_num) {
562 		if (nla_put_u32(msg, RDMA_NLDEV_ATTR_RES_LQPN, id_priv->qp_num))
563 			goto err;
564 		if (nla_put_u8(msg, RDMA_NLDEV_ATTR_RES_TYPE, cm_id->qp_type))
565 			goto err;
566 	}
567 
568 	if (nla_put_u32(msg, RDMA_NLDEV_ATTR_RES_PS, cm_id->ps))
569 		goto err;
570 
571 	if (nla_put_u8(msg, RDMA_NLDEV_ATTR_RES_STATE, id_priv->state))
572 		goto err;
573 
574 	if (cm_id->route.addr.src_addr.ss_family &&
575 	    nla_put(msg, RDMA_NLDEV_ATTR_RES_SRC_ADDR,
576 		    sizeof(cm_id->route.addr.src_addr),
577 		    &cm_id->route.addr.src_addr))
578 		goto err;
579 	if (cm_id->route.addr.dst_addr.ss_family &&
580 	    nla_put(msg, RDMA_NLDEV_ATTR_RES_DST_ADDR,
581 		    sizeof(cm_id->route.addr.dst_addr),
582 		    &cm_id->route.addr.dst_addr))
583 		goto err;
584 
585 	if (nla_put_u32(msg, RDMA_NLDEV_ATTR_RES_CM_IDN, res->id))
586 		goto err;
587 
588 	if (fill_res_name_pid(msg, res))
589 		goto err;
590 
591 	if (dev->ops.fill_res_cm_id_entry)
592 		return dev->ops.fill_res_cm_id_entry(msg, cm_id);
593 	return 0;
594 
595 err: return -EMSGSIZE;
596 }
597 
598 static int fill_res_cq_entry(struct sk_buff *msg, bool has_cap_net_admin,
599 			     struct rdma_restrack_entry *res, uint32_t port)
600 {
601 	struct ib_cq *cq = container_of(res, struct ib_cq, res);
602 	struct ib_device *dev = cq->device;
603 
604 	if (nla_put_u32(msg, RDMA_NLDEV_ATTR_RES_CQE, cq->cqe))
605 		return -EMSGSIZE;
606 	if (nla_put_u64_64bit(msg, RDMA_NLDEV_ATTR_RES_USECNT,
607 			      atomic_read(&cq->usecnt), RDMA_NLDEV_ATTR_PAD))
608 		return -EMSGSIZE;
609 
610 	/* Poll context is only valid for kernel CQs */
611 	if (rdma_is_kernel_res(res) &&
612 	    nla_put_u8(msg, RDMA_NLDEV_ATTR_RES_POLL_CTX, cq->poll_ctx))
613 		return -EMSGSIZE;
614 
615 	if (nla_put_u8(msg, RDMA_NLDEV_ATTR_DEV_DIM, (cq->dim != NULL)))
616 		return -EMSGSIZE;
617 
618 	if (nla_put_u32(msg, RDMA_NLDEV_ATTR_RES_CQN, res->id))
619 		return -EMSGSIZE;
620 	if (!rdma_is_kernel_res(res) &&
621 	    nla_put_u32(msg, RDMA_NLDEV_ATTR_RES_CTXN,
622 			cq->uobject->uevent.uobject.context->res.id))
623 		return -EMSGSIZE;
624 
625 	if (fill_res_name_pid(msg, res))
626 		return -EMSGSIZE;
627 
628 	return (dev->ops.fill_res_cq_entry) ?
629 		dev->ops.fill_res_cq_entry(msg, cq) : 0;
630 }
631 
632 static int fill_res_cq_raw_entry(struct sk_buff *msg, bool has_cap_net_admin,
633 				 struct rdma_restrack_entry *res, uint32_t port)
634 {
635 	struct ib_cq *cq = container_of(res, struct ib_cq, res);
636 	struct ib_device *dev = cq->device;
637 
638 	if (!dev->ops.fill_res_cq_entry_raw)
639 		return -EINVAL;
640 	return dev->ops.fill_res_cq_entry_raw(msg, cq);
641 }
642 
643 static int fill_res_mr_entry(struct sk_buff *msg, bool has_cap_net_admin,
644 			     struct rdma_restrack_entry *res, uint32_t port)
645 {
646 	struct ib_mr *mr = container_of(res, struct ib_mr, res);
647 	struct ib_device *dev = mr->pd->device;
648 
649 	if (has_cap_net_admin) {
650 		if (nla_put_u32(msg, RDMA_NLDEV_ATTR_RES_RKEY, mr->rkey))
651 			return -EMSGSIZE;
652 		if (nla_put_u32(msg, RDMA_NLDEV_ATTR_RES_LKEY, mr->lkey))
653 			return -EMSGSIZE;
654 	}
655 
656 	if (nla_put_u64_64bit(msg, RDMA_NLDEV_ATTR_RES_MRLEN, mr->length,
657 			      RDMA_NLDEV_ATTR_PAD))
658 		return -EMSGSIZE;
659 
660 	if (nla_put_u32(msg, RDMA_NLDEV_ATTR_RES_MRN, res->id))
661 		return -EMSGSIZE;
662 
663 	if (!rdma_is_kernel_res(res) &&
664 	    nla_put_u32(msg, RDMA_NLDEV_ATTR_RES_PDN, mr->pd->res.id))
665 		return -EMSGSIZE;
666 
667 	if (fill_res_name_pid(msg, res))
668 		return -EMSGSIZE;
669 
670 	return (dev->ops.fill_res_mr_entry) ?
671 		       dev->ops.fill_res_mr_entry(msg, mr) :
672 		       0;
673 }
674 
675 static int fill_res_mr_raw_entry(struct sk_buff *msg, bool has_cap_net_admin,
676 				 struct rdma_restrack_entry *res, uint32_t port)
677 {
678 	struct ib_mr *mr = container_of(res, struct ib_mr, res);
679 	struct ib_device *dev = mr->pd->device;
680 
681 	if (!dev->ops.fill_res_mr_entry_raw)
682 		return -EINVAL;
683 	return dev->ops.fill_res_mr_entry_raw(msg, mr);
684 }
685 
686 static int fill_res_pd_entry(struct sk_buff *msg, bool has_cap_net_admin,
687 			     struct rdma_restrack_entry *res, uint32_t port)
688 {
689 	struct ib_pd *pd = container_of(res, struct ib_pd, res);
690 
691 	if (has_cap_net_admin) {
692 		if (nla_put_u32(msg, RDMA_NLDEV_ATTR_RES_LOCAL_DMA_LKEY,
693 				pd->local_dma_lkey))
694 			goto err;
695 		if ((pd->flags & IB_PD_UNSAFE_GLOBAL_RKEY) &&
696 		    nla_put_u32(msg, RDMA_NLDEV_ATTR_RES_UNSAFE_GLOBAL_RKEY,
697 				pd->unsafe_global_rkey))
698 			goto err;
699 	}
700 	if (nla_put_u64_64bit(msg, RDMA_NLDEV_ATTR_RES_USECNT,
701 			      atomic_read(&pd->usecnt), RDMA_NLDEV_ATTR_PAD))
702 		goto err;
703 
704 	if (nla_put_u32(msg, RDMA_NLDEV_ATTR_RES_PDN, res->id))
705 		goto err;
706 
707 	if (!rdma_is_kernel_res(res) &&
708 	    nla_put_u32(msg, RDMA_NLDEV_ATTR_RES_CTXN,
709 			pd->uobject->context->res.id))
710 		goto err;
711 
712 	return fill_res_name_pid(msg, res);
713 
714 err:	return -EMSGSIZE;
715 }
716 
717 static int fill_res_ctx_entry(struct sk_buff *msg, bool has_cap_net_admin,
718 			      struct rdma_restrack_entry *res, uint32_t port)
719 {
720 	struct ib_ucontext *ctx = container_of(res, struct ib_ucontext, res);
721 
722 	if (rdma_is_kernel_res(res))
723 		return 0;
724 
725 	if (nla_put_u32(msg, RDMA_NLDEV_ATTR_RES_CTXN, ctx->res.id))
726 		return -EMSGSIZE;
727 
728 	return fill_res_name_pid(msg, res);
729 }
730 
731 static int fill_res_range_qp_entry(struct sk_buff *msg, uint32_t min_range,
732 				   uint32_t max_range)
733 {
734 	struct nlattr *entry_attr;
735 
736 	if (!min_range)
737 		return 0;
738 
739 	entry_attr = nla_nest_start(msg, RDMA_NLDEV_ATTR_RES_QP_ENTRY);
740 	if (!entry_attr)
741 		return -EMSGSIZE;
742 
743 	if (min_range == max_range) {
744 		if (nla_put_u32(msg, RDMA_NLDEV_ATTR_RES_LQPN, min_range))
745 			goto err;
746 	} else {
747 		if (nla_put_u32(msg, RDMA_NLDEV_ATTR_MIN_RANGE, min_range))
748 			goto err;
749 		if (nla_put_u32(msg, RDMA_NLDEV_ATTR_MAX_RANGE, max_range))
750 			goto err;
751 	}
752 	nla_nest_end(msg, entry_attr);
753 	return 0;
754 
755 err:
756 	nla_nest_cancel(msg, entry_attr);
757 	return -EMSGSIZE;
758 }
759 
760 static int fill_res_srq_qps(struct sk_buff *msg, struct ib_srq *srq)
761 {
762 	uint32_t min_range = 0, prev = 0;
763 	struct rdma_restrack_entry *res;
764 	struct rdma_restrack_root *rt;
765 	struct nlattr *table_attr;
766 	struct ib_qp *qp = NULL;
767 	unsigned long id = 0;
768 
769 	table_attr = nla_nest_start(msg, RDMA_NLDEV_ATTR_RES_QP);
770 	if (!table_attr)
771 		return -EMSGSIZE;
772 
773 	rt = &srq->device->res[RDMA_RESTRACK_QP];
774 	xa_lock(&rt->xa);
775 	xa_for_each(&rt->xa, id, res) {
776 		if (!rdma_restrack_get(res))
777 			continue;
778 
779 		qp = container_of(res, struct ib_qp, res);
780 		if (!qp->srq || (qp->srq->res.id != srq->res.id)) {
781 			rdma_restrack_put(res);
782 			continue;
783 		}
784 
785 		if (qp->qp_num < prev)
786 			/* qp_num should be ascending */
787 			goto err_loop;
788 
789 		if (min_range == 0) {
790 			min_range = qp->qp_num;
791 		} else if (qp->qp_num > (prev + 1)) {
792 			if (fill_res_range_qp_entry(msg, min_range, prev))
793 				goto err_loop;
794 
795 			min_range = qp->qp_num;
796 		}
797 		prev = qp->qp_num;
798 		rdma_restrack_put(res);
799 	}
800 
801 	xa_unlock(&rt->xa);
802 
803 	if (fill_res_range_qp_entry(msg, min_range, prev))
804 		goto err;
805 
806 	nla_nest_end(msg, table_attr);
807 	return 0;
808 
809 err_loop:
810 	rdma_restrack_put(res);
811 	xa_unlock(&rt->xa);
812 err:
813 	nla_nest_cancel(msg, table_attr);
814 	return -EMSGSIZE;
815 }
816 
817 static int fill_res_srq_entry(struct sk_buff *msg, bool has_cap_net_admin,
818 			      struct rdma_restrack_entry *res, uint32_t port)
819 {
820 	struct ib_srq *srq = container_of(res, struct ib_srq, res);
821 
822 	if (nla_put_u32(msg, RDMA_NLDEV_ATTR_RES_SRQN, srq->res.id))
823 		goto err;
824 
825 	if (nla_put_u8(msg, RDMA_NLDEV_ATTR_RES_TYPE, srq->srq_type))
826 		goto err;
827 
828 	if (nla_put_u32(msg, RDMA_NLDEV_ATTR_RES_PDN, srq->pd->res.id))
829 		goto err;
830 
831 	if (ib_srq_has_cq(srq->srq_type)) {
832 		if (nla_put_u32(msg, RDMA_NLDEV_ATTR_RES_CQN,
833 				srq->ext.cq->res.id))
834 			goto err;
835 	}
836 
837 	if (fill_res_srq_qps(msg, srq))
838 		goto err;
839 
840 	return fill_res_name_pid(msg, res);
841 
842 err:
843 	return -EMSGSIZE;
844 }
845 
846 static int fill_stat_counter_mode(struct sk_buff *msg,
847 				  struct rdma_counter *counter)
848 {
849 	struct rdma_counter_mode *m = &counter->mode;
850 
851 	if (nla_put_u32(msg, RDMA_NLDEV_ATTR_STAT_MODE, m->mode))
852 		return -EMSGSIZE;
853 
854 	if (m->mode == RDMA_COUNTER_MODE_AUTO) {
855 		if ((m->mask & RDMA_COUNTER_MASK_QP_TYPE) &&
856 		    nla_put_u8(msg, RDMA_NLDEV_ATTR_RES_TYPE, m->param.qp_type))
857 			return -EMSGSIZE;
858 
859 		if ((m->mask & RDMA_COUNTER_MASK_PID) &&
860 		    fill_res_name_pid(msg, &counter->res))
861 			return -EMSGSIZE;
862 	}
863 
864 	return 0;
865 }
866 
867 static int fill_stat_counter_qp_entry(struct sk_buff *msg, u32 qpn)
868 {
869 	struct nlattr *entry_attr;
870 
871 	entry_attr = nla_nest_start(msg, RDMA_NLDEV_ATTR_RES_QP_ENTRY);
872 	if (!entry_attr)
873 		return -EMSGSIZE;
874 
875 	if (nla_put_u32(msg, RDMA_NLDEV_ATTR_RES_LQPN, qpn))
876 		goto err;
877 
878 	nla_nest_end(msg, entry_attr);
879 	return 0;
880 
881 err:
882 	nla_nest_cancel(msg, entry_attr);
883 	return -EMSGSIZE;
884 }
885 
886 static int fill_stat_counter_qps(struct sk_buff *msg,
887 				 struct rdma_counter *counter)
888 {
889 	struct rdma_restrack_entry *res;
890 	struct rdma_restrack_root *rt;
891 	struct nlattr *table_attr;
892 	struct ib_qp *qp = NULL;
893 	unsigned long id = 0;
894 	int ret = 0;
895 
896 	table_attr = nla_nest_start(msg, RDMA_NLDEV_ATTR_RES_QP);
897 
898 	rt = &counter->device->res[RDMA_RESTRACK_QP];
899 	xa_lock(&rt->xa);
900 	xa_for_each(&rt->xa, id, res) {
901 		qp = container_of(res, struct ib_qp, res);
902 		if (!qp->counter || (qp->counter->id != counter->id))
903 			continue;
904 
905 		ret = fill_stat_counter_qp_entry(msg, qp->qp_num);
906 		if (ret)
907 			goto err;
908 	}
909 
910 	xa_unlock(&rt->xa);
911 	nla_nest_end(msg, table_attr);
912 	return 0;
913 
914 err:
915 	xa_unlock(&rt->xa);
916 	nla_nest_cancel(msg, table_attr);
917 	return ret;
918 }
919 
920 int rdma_nl_stat_hwcounter_entry(struct sk_buff *msg, const char *name,
921 				 u64 value)
922 {
923 	struct nlattr *entry_attr;
924 
925 	entry_attr = nla_nest_start(msg, RDMA_NLDEV_ATTR_STAT_HWCOUNTER_ENTRY);
926 	if (!entry_attr)
927 		return -EMSGSIZE;
928 
929 	if (nla_put_string(msg, RDMA_NLDEV_ATTR_STAT_HWCOUNTER_ENTRY_NAME,
930 			   name))
931 		goto err;
932 	if (nla_put_u64_64bit(msg, RDMA_NLDEV_ATTR_STAT_HWCOUNTER_ENTRY_VALUE,
933 			      value, RDMA_NLDEV_ATTR_PAD))
934 		goto err;
935 
936 	nla_nest_end(msg, entry_attr);
937 	return 0;
938 
939 err:
940 	nla_nest_cancel(msg, entry_attr);
941 	return -EMSGSIZE;
942 }
943 EXPORT_SYMBOL(rdma_nl_stat_hwcounter_entry);
944 
945 static int fill_stat_mr_entry(struct sk_buff *msg, bool has_cap_net_admin,
946 			      struct rdma_restrack_entry *res, uint32_t port)
947 {
948 	struct ib_mr *mr = container_of(res, struct ib_mr, res);
949 	struct ib_device *dev = mr->pd->device;
950 
951 	if (nla_put_u32(msg, RDMA_NLDEV_ATTR_RES_MRN, res->id))
952 		goto err;
953 
954 	if (dev->ops.fill_stat_mr_entry)
955 		return dev->ops.fill_stat_mr_entry(msg, mr);
956 	return 0;
957 
958 err:
959 	return -EMSGSIZE;
960 }
961 
962 static int fill_stat_counter_hwcounters(struct sk_buff *msg,
963 					struct rdma_counter *counter)
964 {
965 	struct rdma_hw_stats *st = counter->stats;
966 	struct nlattr *table_attr;
967 	int i;
968 
969 	table_attr = nla_nest_start(msg, RDMA_NLDEV_ATTR_STAT_HWCOUNTERS);
970 	if (!table_attr)
971 		return -EMSGSIZE;
972 
973 	mutex_lock(&st->lock);
974 	for (i = 0; i < st->num_counters; i++) {
975 		if (test_bit(i, st->is_disabled))
976 			continue;
977 		if (rdma_nl_stat_hwcounter_entry(msg, st->descs[i].name,
978 						 st->value[i]))
979 			goto err;
980 	}
981 	mutex_unlock(&st->lock);
982 
983 	nla_nest_end(msg, table_attr);
984 	return 0;
985 
986 err:
987 	mutex_unlock(&st->lock);
988 	nla_nest_cancel(msg, table_attr);
989 	return -EMSGSIZE;
990 }
991 
992 static int fill_res_counter_entry(struct sk_buff *msg, bool has_cap_net_admin,
993 				  struct rdma_restrack_entry *res,
994 				  uint32_t port)
995 {
996 	struct rdma_counter *counter =
997 		container_of(res, struct rdma_counter, res);
998 
999 	if (port && port != counter->port)
1000 		return -EAGAIN;
1001 
1002 	/* Dump it even query failed */
1003 	rdma_counter_query_stats(counter);
1004 
1005 	if (nla_put_u32(msg, RDMA_NLDEV_ATTR_PORT_INDEX, counter->port) ||
1006 	    nla_put_u32(msg, RDMA_NLDEV_ATTR_STAT_COUNTER_ID, counter->id) ||
1007 	    fill_stat_counter_mode(msg, counter) ||
1008 	    fill_stat_counter_qps(msg, counter) ||
1009 	    fill_stat_counter_hwcounters(msg, counter))
1010 		return -EMSGSIZE;
1011 
1012 	return 0;
1013 }
1014 
1015 static int nldev_get_doit(struct sk_buff *skb, struct nlmsghdr *nlh,
1016 			  struct netlink_ext_ack *extack)
1017 {
1018 	struct nlattr *tb[RDMA_NLDEV_ATTR_MAX];
1019 	struct ib_device *device;
1020 	struct sk_buff *msg;
1021 	u32 index;
1022 	int err;
1023 
1024 	err = nlmsg_parse_deprecated(nlh, 0, tb, RDMA_NLDEV_ATTR_MAX - 1,
1025 				     nldev_policy, extack);
1026 	if (err || !tb[RDMA_NLDEV_ATTR_DEV_INDEX])
1027 		return -EINVAL;
1028 
1029 	index = nla_get_u32(tb[RDMA_NLDEV_ATTR_DEV_INDEX]);
1030 
1031 	device = ib_device_get_by_index(sock_net(skb->sk), index);
1032 	if (!device)
1033 		return -EINVAL;
1034 
1035 	msg = nlmsg_new(NLMSG_DEFAULT_SIZE, GFP_KERNEL);
1036 	if (!msg) {
1037 		err = -ENOMEM;
1038 		goto err;
1039 	}
1040 
1041 	nlh = nlmsg_put(msg, NETLINK_CB(skb).portid, nlh->nlmsg_seq,
1042 			RDMA_NL_GET_TYPE(RDMA_NL_NLDEV, RDMA_NLDEV_CMD_GET),
1043 			0, 0);
1044 
1045 	err = fill_dev_info(msg, device);
1046 	if (err)
1047 		goto err_free;
1048 
1049 	nlmsg_end(msg, nlh);
1050 
1051 	ib_device_put(device);
1052 	return rdma_nl_unicast(sock_net(skb->sk), msg, NETLINK_CB(skb).portid);
1053 
1054 err_free:
1055 	nlmsg_free(msg);
1056 err:
1057 	ib_device_put(device);
1058 	return err;
1059 }
1060 
1061 static int nldev_set_doit(struct sk_buff *skb, struct nlmsghdr *nlh,
1062 			  struct netlink_ext_ack *extack)
1063 {
1064 	struct nlattr *tb[RDMA_NLDEV_ATTR_MAX];
1065 	struct ib_device *device;
1066 	u32 index;
1067 	int err;
1068 
1069 	err = nlmsg_parse_deprecated(nlh, 0, tb, RDMA_NLDEV_ATTR_MAX - 1,
1070 				     nldev_policy, extack);
1071 	if (err || !tb[RDMA_NLDEV_ATTR_DEV_INDEX])
1072 		return -EINVAL;
1073 
1074 	index = nla_get_u32(tb[RDMA_NLDEV_ATTR_DEV_INDEX]);
1075 	device = ib_device_get_by_index(sock_net(skb->sk), index);
1076 	if (!device)
1077 		return -EINVAL;
1078 
1079 	if (tb[RDMA_NLDEV_ATTR_DEV_NAME]) {
1080 		char name[IB_DEVICE_NAME_MAX] = {};
1081 
1082 		nla_strscpy(name, tb[RDMA_NLDEV_ATTR_DEV_NAME],
1083 			    IB_DEVICE_NAME_MAX);
1084 		if (strlen(name) == 0) {
1085 			err = -EINVAL;
1086 			goto done;
1087 		}
1088 		err = ib_device_rename(device, name);
1089 		goto done;
1090 	}
1091 
1092 	if (tb[RDMA_NLDEV_NET_NS_FD]) {
1093 		u32 ns_fd;
1094 
1095 		ns_fd = nla_get_u32(tb[RDMA_NLDEV_NET_NS_FD]);
1096 		err = ib_device_set_netns_put(skb, device, ns_fd);
1097 		goto put_done;
1098 	}
1099 
1100 	if (tb[RDMA_NLDEV_ATTR_DEV_DIM]) {
1101 		u8 use_dim;
1102 
1103 		use_dim = nla_get_u8(tb[RDMA_NLDEV_ATTR_DEV_DIM]);
1104 		err = ib_device_set_dim(device,  use_dim);
1105 		goto done;
1106 	}
1107 
1108 done:
1109 	ib_device_put(device);
1110 put_done:
1111 	return err;
1112 }
1113 
1114 static int _nldev_get_dumpit(struct ib_device *device,
1115 			     struct sk_buff *skb,
1116 			     struct netlink_callback *cb,
1117 			     unsigned int idx)
1118 {
1119 	int start = cb->args[0];
1120 	struct nlmsghdr *nlh;
1121 
1122 	if (idx < start)
1123 		return 0;
1124 
1125 	nlh = nlmsg_put(skb, NETLINK_CB(cb->skb).portid, cb->nlh->nlmsg_seq,
1126 			RDMA_NL_GET_TYPE(RDMA_NL_NLDEV, RDMA_NLDEV_CMD_GET),
1127 			0, NLM_F_MULTI);
1128 
1129 	if (fill_dev_info(skb, device)) {
1130 		nlmsg_cancel(skb, nlh);
1131 		goto out;
1132 	}
1133 
1134 	nlmsg_end(skb, nlh);
1135 
1136 	idx++;
1137 
1138 out:	cb->args[0] = idx;
1139 	return skb->len;
1140 }
1141 
1142 static int nldev_get_dumpit(struct sk_buff *skb, struct netlink_callback *cb)
1143 {
1144 	/*
1145 	 * There is no need to take lock, because
1146 	 * we are relying on ib_core's locking.
1147 	 */
1148 	return ib_enum_all_devs(_nldev_get_dumpit, skb, cb);
1149 }
1150 
1151 static int nldev_port_get_doit(struct sk_buff *skb, struct nlmsghdr *nlh,
1152 			       struct netlink_ext_ack *extack)
1153 {
1154 	struct nlattr *tb[RDMA_NLDEV_ATTR_MAX];
1155 	struct ib_device *device;
1156 	struct sk_buff *msg;
1157 	u32 index;
1158 	u32 port;
1159 	int err;
1160 
1161 	err = nlmsg_parse_deprecated(nlh, 0, tb, RDMA_NLDEV_ATTR_MAX - 1,
1162 				     nldev_policy, extack);
1163 	if (err ||
1164 	    !tb[RDMA_NLDEV_ATTR_DEV_INDEX] ||
1165 	    !tb[RDMA_NLDEV_ATTR_PORT_INDEX])
1166 		return -EINVAL;
1167 
1168 	index = nla_get_u32(tb[RDMA_NLDEV_ATTR_DEV_INDEX]);
1169 	device = ib_device_get_by_index(sock_net(skb->sk), index);
1170 	if (!device)
1171 		return -EINVAL;
1172 
1173 	port = nla_get_u32(tb[RDMA_NLDEV_ATTR_PORT_INDEX]);
1174 	if (!rdma_is_port_valid(device, port)) {
1175 		err = -EINVAL;
1176 		goto err;
1177 	}
1178 
1179 	msg = nlmsg_new(NLMSG_DEFAULT_SIZE, GFP_KERNEL);
1180 	if (!msg) {
1181 		err = -ENOMEM;
1182 		goto err;
1183 	}
1184 
1185 	nlh = nlmsg_put(msg, NETLINK_CB(skb).portid, nlh->nlmsg_seq,
1186 			RDMA_NL_GET_TYPE(RDMA_NL_NLDEV, RDMA_NLDEV_CMD_GET),
1187 			0, 0);
1188 
1189 	err = fill_port_info(msg, device, port, sock_net(skb->sk));
1190 	if (err)
1191 		goto err_free;
1192 
1193 	nlmsg_end(msg, nlh);
1194 	ib_device_put(device);
1195 
1196 	return rdma_nl_unicast(sock_net(skb->sk), msg, NETLINK_CB(skb).portid);
1197 
1198 err_free:
1199 	nlmsg_free(msg);
1200 err:
1201 	ib_device_put(device);
1202 	return err;
1203 }
1204 
1205 static int nldev_port_get_dumpit(struct sk_buff *skb,
1206 				 struct netlink_callback *cb)
1207 {
1208 	struct nlattr *tb[RDMA_NLDEV_ATTR_MAX];
1209 	struct ib_device *device;
1210 	int start = cb->args[0];
1211 	struct nlmsghdr *nlh;
1212 	u32 idx = 0;
1213 	u32 ifindex;
1214 	int err;
1215 	unsigned int p;
1216 
1217 	err = nlmsg_parse_deprecated(cb->nlh, 0, tb, RDMA_NLDEV_ATTR_MAX - 1,
1218 				     nldev_policy, NULL);
1219 	if (err || !tb[RDMA_NLDEV_ATTR_DEV_INDEX])
1220 		return -EINVAL;
1221 
1222 	ifindex = nla_get_u32(tb[RDMA_NLDEV_ATTR_DEV_INDEX]);
1223 	device = ib_device_get_by_index(sock_net(skb->sk), ifindex);
1224 	if (!device)
1225 		return -EINVAL;
1226 
1227 	rdma_for_each_port (device, p) {
1228 		/*
1229 		 * The dumpit function returns all information from specific
1230 		 * index. This specific index is taken from the netlink
1231 		 * messages request sent by user and it is available
1232 		 * in cb->args[0].
1233 		 *
1234 		 * Usually, the user doesn't fill this field and it causes
1235 		 * to return everything.
1236 		 *
1237 		 */
1238 		if (idx < start) {
1239 			idx++;
1240 			continue;
1241 		}
1242 
1243 		nlh = nlmsg_put(skb, NETLINK_CB(cb->skb).portid,
1244 				cb->nlh->nlmsg_seq,
1245 				RDMA_NL_GET_TYPE(RDMA_NL_NLDEV,
1246 						 RDMA_NLDEV_CMD_PORT_GET),
1247 				0, NLM_F_MULTI);
1248 
1249 		if (fill_port_info(skb, device, p, sock_net(skb->sk))) {
1250 			nlmsg_cancel(skb, nlh);
1251 			goto out;
1252 		}
1253 		idx++;
1254 		nlmsg_end(skb, nlh);
1255 	}
1256 
1257 out:
1258 	ib_device_put(device);
1259 	cb->args[0] = idx;
1260 	return skb->len;
1261 }
1262 
1263 static int nldev_res_get_doit(struct sk_buff *skb, struct nlmsghdr *nlh,
1264 			      struct netlink_ext_ack *extack)
1265 {
1266 	struct nlattr *tb[RDMA_NLDEV_ATTR_MAX];
1267 	struct ib_device *device;
1268 	struct sk_buff *msg;
1269 	u32 index;
1270 	int ret;
1271 
1272 	ret = nlmsg_parse_deprecated(nlh, 0, tb, RDMA_NLDEV_ATTR_MAX - 1,
1273 				     nldev_policy, extack);
1274 	if (ret || !tb[RDMA_NLDEV_ATTR_DEV_INDEX])
1275 		return -EINVAL;
1276 
1277 	index = nla_get_u32(tb[RDMA_NLDEV_ATTR_DEV_INDEX]);
1278 	device = ib_device_get_by_index(sock_net(skb->sk), index);
1279 	if (!device)
1280 		return -EINVAL;
1281 
1282 	msg = nlmsg_new(NLMSG_DEFAULT_SIZE, GFP_KERNEL);
1283 	if (!msg) {
1284 		ret = -ENOMEM;
1285 		goto err;
1286 	}
1287 
1288 	nlh = nlmsg_put(msg, NETLINK_CB(skb).portid, nlh->nlmsg_seq,
1289 			RDMA_NL_GET_TYPE(RDMA_NL_NLDEV, RDMA_NLDEV_CMD_RES_GET),
1290 			0, 0);
1291 
1292 	ret = fill_res_info(msg, device);
1293 	if (ret)
1294 		goto err_free;
1295 
1296 	nlmsg_end(msg, nlh);
1297 	ib_device_put(device);
1298 	return rdma_nl_unicast(sock_net(skb->sk), msg, NETLINK_CB(skb).portid);
1299 
1300 err_free:
1301 	nlmsg_free(msg);
1302 err:
1303 	ib_device_put(device);
1304 	return ret;
1305 }
1306 
1307 static int _nldev_res_get_dumpit(struct ib_device *device,
1308 				 struct sk_buff *skb,
1309 				 struct netlink_callback *cb,
1310 				 unsigned int idx)
1311 {
1312 	int start = cb->args[0];
1313 	struct nlmsghdr *nlh;
1314 
1315 	if (idx < start)
1316 		return 0;
1317 
1318 	nlh = nlmsg_put(skb, NETLINK_CB(cb->skb).portid, cb->nlh->nlmsg_seq,
1319 			RDMA_NL_GET_TYPE(RDMA_NL_NLDEV, RDMA_NLDEV_CMD_RES_GET),
1320 			0, NLM_F_MULTI);
1321 
1322 	if (fill_res_info(skb, device)) {
1323 		nlmsg_cancel(skb, nlh);
1324 		goto out;
1325 	}
1326 	nlmsg_end(skb, nlh);
1327 
1328 	idx++;
1329 
1330 out:
1331 	cb->args[0] = idx;
1332 	return skb->len;
1333 }
1334 
1335 static int nldev_res_get_dumpit(struct sk_buff *skb,
1336 				struct netlink_callback *cb)
1337 {
1338 	return ib_enum_all_devs(_nldev_res_get_dumpit, skb, cb);
1339 }
1340 
1341 struct nldev_fill_res_entry {
1342 	enum rdma_nldev_attr nldev_attr;
1343 	u8 flags;
1344 	u32 entry;
1345 	u32 id;
1346 };
1347 
1348 enum nldev_res_flags {
1349 	NLDEV_PER_DEV = 1 << 0,
1350 };
1351 
1352 static const struct nldev_fill_res_entry fill_entries[RDMA_RESTRACK_MAX] = {
1353 	[RDMA_RESTRACK_QP] = {
1354 		.nldev_attr = RDMA_NLDEV_ATTR_RES_QP,
1355 		.entry = RDMA_NLDEV_ATTR_RES_QP_ENTRY,
1356 		.id = RDMA_NLDEV_ATTR_RES_LQPN,
1357 	},
1358 	[RDMA_RESTRACK_CM_ID] = {
1359 		.nldev_attr = RDMA_NLDEV_ATTR_RES_CM_ID,
1360 		.entry = RDMA_NLDEV_ATTR_RES_CM_ID_ENTRY,
1361 		.id = RDMA_NLDEV_ATTR_RES_CM_IDN,
1362 	},
1363 	[RDMA_RESTRACK_CQ] = {
1364 		.nldev_attr = RDMA_NLDEV_ATTR_RES_CQ,
1365 		.flags = NLDEV_PER_DEV,
1366 		.entry = RDMA_NLDEV_ATTR_RES_CQ_ENTRY,
1367 		.id = RDMA_NLDEV_ATTR_RES_CQN,
1368 	},
1369 	[RDMA_RESTRACK_MR] = {
1370 		.nldev_attr = RDMA_NLDEV_ATTR_RES_MR,
1371 		.flags = NLDEV_PER_DEV,
1372 		.entry = RDMA_NLDEV_ATTR_RES_MR_ENTRY,
1373 		.id = RDMA_NLDEV_ATTR_RES_MRN,
1374 	},
1375 	[RDMA_RESTRACK_PD] = {
1376 		.nldev_attr = RDMA_NLDEV_ATTR_RES_PD,
1377 		.flags = NLDEV_PER_DEV,
1378 		.entry = RDMA_NLDEV_ATTR_RES_PD_ENTRY,
1379 		.id = RDMA_NLDEV_ATTR_RES_PDN,
1380 	},
1381 	[RDMA_RESTRACK_COUNTER] = {
1382 		.nldev_attr = RDMA_NLDEV_ATTR_STAT_COUNTER,
1383 		.entry = RDMA_NLDEV_ATTR_STAT_COUNTER_ENTRY,
1384 		.id = RDMA_NLDEV_ATTR_STAT_COUNTER_ID,
1385 	},
1386 	[RDMA_RESTRACK_CTX] = {
1387 		.nldev_attr = RDMA_NLDEV_ATTR_RES_CTX,
1388 		.flags = NLDEV_PER_DEV,
1389 		.entry = RDMA_NLDEV_ATTR_RES_CTX_ENTRY,
1390 		.id = RDMA_NLDEV_ATTR_RES_CTXN,
1391 	},
1392 	[RDMA_RESTRACK_SRQ] = {
1393 		.nldev_attr = RDMA_NLDEV_ATTR_RES_SRQ,
1394 		.flags = NLDEV_PER_DEV,
1395 		.entry = RDMA_NLDEV_ATTR_RES_SRQ_ENTRY,
1396 		.id = RDMA_NLDEV_ATTR_RES_SRQN,
1397 	},
1398 
1399 };
1400 
1401 static int res_get_common_doit(struct sk_buff *skb, struct nlmsghdr *nlh,
1402 			       struct netlink_ext_ack *extack,
1403 			       enum rdma_restrack_type res_type,
1404 			       res_fill_func_t fill_func)
1405 {
1406 	const struct nldev_fill_res_entry *fe = &fill_entries[res_type];
1407 	struct nlattr *tb[RDMA_NLDEV_ATTR_MAX];
1408 	struct rdma_restrack_entry *res;
1409 	struct ib_device *device;
1410 	u32 index, id, port = 0;
1411 	bool has_cap_net_admin;
1412 	struct sk_buff *msg;
1413 	int ret;
1414 
1415 	ret = nlmsg_parse_deprecated(nlh, 0, tb, RDMA_NLDEV_ATTR_MAX - 1,
1416 				     nldev_policy, extack);
1417 	if (ret || !tb[RDMA_NLDEV_ATTR_DEV_INDEX] || !fe->id || !tb[fe->id])
1418 		return -EINVAL;
1419 
1420 	index = nla_get_u32(tb[RDMA_NLDEV_ATTR_DEV_INDEX]);
1421 	device = ib_device_get_by_index(sock_net(skb->sk), index);
1422 	if (!device)
1423 		return -EINVAL;
1424 
1425 	if (tb[RDMA_NLDEV_ATTR_PORT_INDEX]) {
1426 		port = nla_get_u32(tb[RDMA_NLDEV_ATTR_PORT_INDEX]);
1427 		if (!rdma_is_port_valid(device, port)) {
1428 			ret = -EINVAL;
1429 			goto err;
1430 		}
1431 	}
1432 
1433 	if ((port && fe->flags & NLDEV_PER_DEV) ||
1434 	    (!port && ~fe->flags & NLDEV_PER_DEV)) {
1435 		ret = -EINVAL;
1436 		goto err;
1437 	}
1438 
1439 	id = nla_get_u32(tb[fe->id]);
1440 	res = rdma_restrack_get_byid(device, res_type, id);
1441 	if (IS_ERR(res)) {
1442 		ret = PTR_ERR(res);
1443 		goto err;
1444 	}
1445 
1446 	msg = nlmsg_new(NLMSG_DEFAULT_SIZE, GFP_KERNEL);
1447 	if (!msg) {
1448 		ret = -ENOMEM;
1449 		goto err_get;
1450 	}
1451 
1452 	nlh = nlmsg_put(msg, NETLINK_CB(skb).portid, nlh->nlmsg_seq,
1453 			RDMA_NL_GET_TYPE(RDMA_NL_NLDEV,
1454 					 RDMA_NL_GET_OP(nlh->nlmsg_type)),
1455 			0, 0);
1456 
1457 	if (fill_nldev_handle(msg, device)) {
1458 		ret = -EMSGSIZE;
1459 		goto err_free;
1460 	}
1461 
1462 	has_cap_net_admin = netlink_capable(skb, CAP_NET_ADMIN);
1463 
1464 	ret = fill_func(msg, has_cap_net_admin, res, port);
1465 	if (ret)
1466 		goto err_free;
1467 
1468 	rdma_restrack_put(res);
1469 	nlmsg_end(msg, nlh);
1470 	ib_device_put(device);
1471 	return rdma_nl_unicast(sock_net(skb->sk), msg, NETLINK_CB(skb).portid);
1472 
1473 err_free:
1474 	nlmsg_free(msg);
1475 err_get:
1476 	rdma_restrack_put(res);
1477 err:
1478 	ib_device_put(device);
1479 	return ret;
1480 }
1481 
1482 static int res_get_common_dumpit(struct sk_buff *skb,
1483 				 struct netlink_callback *cb,
1484 				 enum rdma_restrack_type res_type,
1485 				 res_fill_func_t fill_func)
1486 {
1487 	const struct nldev_fill_res_entry *fe = &fill_entries[res_type];
1488 	struct nlattr *tb[RDMA_NLDEV_ATTR_MAX];
1489 	struct rdma_restrack_entry *res;
1490 	struct rdma_restrack_root *rt;
1491 	int err, ret = 0, idx = 0;
1492 	struct nlattr *table_attr;
1493 	struct nlattr *entry_attr;
1494 	struct ib_device *device;
1495 	int start = cb->args[0];
1496 	bool has_cap_net_admin;
1497 	struct nlmsghdr *nlh;
1498 	unsigned long id;
1499 	u32 index, port = 0;
1500 	bool filled = false;
1501 
1502 	err = nlmsg_parse_deprecated(cb->nlh, 0, tb, RDMA_NLDEV_ATTR_MAX - 1,
1503 				     nldev_policy, NULL);
1504 	/*
1505 	 * Right now, we are expecting the device index to get res information,
1506 	 * but it is possible to extend this code to return all devices in
1507 	 * one shot by checking the existence of RDMA_NLDEV_ATTR_DEV_INDEX.
1508 	 * if it doesn't exist, we will iterate over all devices.
1509 	 *
1510 	 * But it is not needed for now.
1511 	 */
1512 	if (err || !tb[RDMA_NLDEV_ATTR_DEV_INDEX])
1513 		return -EINVAL;
1514 
1515 	index = nla_get_u32(tb[RDMA_NLDEV_ATTR_DEV_INDEX]);
1516 	device = ib_device_get_by_index(sock_net(skb->sk), index);
1517 	if (!device)
1518 		return -EINVAL;
1519 
1520 	/*
1521 	 * If no PORT_INDEX is supplied, we will return all QPs from that device
1522 	 */
1523 	if (tb[RDMA_NLDEV_ATTR_PORT_INDEX]) {
1524 		port = nla_get_u32(tb[RDMA_NLDEV_ATTR_PORT_INDEX]);
1525 		if (!rdma_is_port_valid(device, port)) {
1526 			ret = -EINVAL;
1527 			goto err_index;
1528 		}
1529 	}
1530 
1531 	nlh = nlmsg_put(skb, NETLINK_CB(cb->skb).portid, cb->nlh->nlmsg_seq,
1532 			RDMA_NL_GET_TYPE(RDMA_NL_NLDEV,
1533 					 RDMA_NL_GET_OP(cb->nlh->nlmsg_type)),
1534 			0, NLM_F_MULTI);
1535 
1536 	if (fill_nldev_handle(skb, device)) {
1537 		ret = -EMSGSIZE;
1538 		goto err;
1539 	}
1540 
1541 	table_attr = nla_nest_start_noflag(skb, fe->nldev_attr);
1542 	if (!table_attr) {
1543 		ret = -EMSGSIZE;
1544 		goto err;
1545 	}
1546 
1547 	has_cap_net_admin = netlink_capable(cb->skb, CAP_NET_ADMIN);
1548 
1549 	rt = &device->res[res_type];
1550 	xa_lock(&rt->xa);
1551 	/*
1552 	 * FIXME: if the skip ahead is something common this loop should
1553 	 * use xas_for_each & xas_pause to optimize, we can have a lot of
1554 	 * objects.
1555 	 */
1556 	xa_for_each(&rt->xa, id, res) {
1557 		if (idx < start || !rdma_restrack_get(res))
1558 			goto next;
1559 
1560 		xa_unlock(&rt->xa);
1561 
1562 		filled = true;
1563 
1564 		entry_attr = nla_nest_start_noflag(skb, fe->entry);
1565 		if (!entry_attr) {
1566 			ret = -EMSGSIZE;
1567 			rdma_restrack_put(res);
1568 			goto msg_full;
1569 		}
1570 
1571 		ret = fill_func(skb, has_cap_net_admin, res, port);
1572 
1573 		rdma_restrack_put(res);
1574 
1575 		if (ret) {
1576 			nla_nest_cancel(skb, entry_attr);
1577 			if (ret == -EMSGSIZE)
1578 				goto msg_full;
1579 			if (ret == -EAGAIN)
1580 				goto again;
1581 			goto res_err;
1582 		}
1583 		nla_nest_end(skb, entry_attr);
1584 again:		xa_lock(&rt->xa);
1585 next:		idx++;
1586 	}
1587 	xa_unlock(&rt->xa);
1588 
1589 msg_full:
1590 	nla_nest_end(skb, table_attr);
1591 	nlmsg_end(skb, nlh);
1592 	cb->args[0] = idx;
1593 
1594 	/*
1595 	 * No more entries to fill, cancel the message and
1596 	 * return 0 to mark end of dumpit.
1597 	 */
1598 	if (!filled)
1599 		goto err;
1600 
1601 	ib_device_put(device);
1602 	return skb->len;
1603 
1604 res_err:
1605 	nla_nest_cancel(skb, table_attr);
1606 
1607 err:
1608 	nlmsg_cancel(skb, nlh);
1609 
1610 err_index:
1611 	ib_device_put(device);
1612 	return ret;
1613 }
1614 
1615 #define RES_GET_FUNCS(name, type)                                              \
1616 	static int nldev_res_get_##name##_dumpit(struct sk_buff *skb,          \
1617 						 struct netlink_callback *cb)  \
1618 	{                                                                      \
1619 		return res_get_common_dumpit(skb, cb, type,                    \
1620 					     fill_res_##name##_entry);         \
1621 	}                                                                      \
1622 	static int nldev_res_get_##name##_doit(struct sk_buff *skb,            \
1623 					       struct nlmsghdr *nlh,           \
1624 					       struct netlink_ext_ack *extack) \
1625 	{                                                                      \
1626 		return res_get_common_doit(skb, nlh, extack, type,             \
1627 					   fill_res_##name##_entry);           \
1628 	}
1629 
1630 RES_GET_FUNCS(qp, RDMA_RESTRACK_QP);
1631 RES_GET_FUNCS(qp_raw, RDMA_RESTRACK_QP);
1632 RES_GET_FUNCS(cm_id, RDMA_RESTRACK_CM_ID);
1633 RES_GET_FUNCS(cq, RDMA_RESTRACK_CQ);
1634 RES_GET_FUNCS(cq_raw, RDMA_RESTRACK_CQ);
1635 RES_GET_FUNCS(pd, RDMA_RESTRACK_PD);
1636 RES_GET_FUNCS(mr, RDMA_RESTRACK_MR);
1637 RES_GET_FUNCS(mr_raw, RDMA_RESTRACK_MR);
1638 RES_GET_FUNCS(counter, RDMA_RESTRACK_COUNTER);
1639 RES_GET_FUNCS(ctx, RDMA_RESTRACK_CTX);
1640 RES_GET_FUNCS(srq, RDMA_RESTRACK_SRQ);
1641 
1642 static LIST_HEAD(link_ops);
1643 static DECLARE_RWSEM(link_ops_rwsem);
1644 
1645 static const struct rdma_link_ops *link_ops_get(const char *type)
1646 {
1647 	const struct rdma_link_ops *ops;
1648 
1649 	list_for_each_entry(ops, &link_ops, list) {
1650 		if (!strcmp(ops->type, type))
1651 			goto out;
1652 	}
1653 	ops = NULL;
1654 out:
1655 	return ops;
1656 }
1657 
1658 void rdma_link_register(struct rdma_link_ops *ops)
1659 {
1660 	down_write(&link_ops_rwsem);
1661 	if (WARN_ON_ONCE(link_ops_get(ops->type)))
1662 		goto out;
1663 	list_add(&ops->list, &link_ops);
1664 out:
1665 	up_write(&link_ops_rwsem);
1666 }
1667 EXPORT_SYMBOL(rdma_link_register);
1668 
1669 void rdma_link_unregister(struct rdma_link_ops *ops)
1670 {
1671 	down_write(&link_ops_rwsem);
1672 	list_del(&ops->list);
1673 	up_write(&link_ops_rwsem);
1674 }
1675 EXPORT_SYMBOL(rdma_link_unregister);
1676 
1677 static int nldev_newlink(struct sk_buff *skb, struct nlmsghdr *nlh,
1678 			  struct netlink_ext_ack *extack)
1679 {
1680 	struct nlattr *tb[RDMA_NLDEV_ATTR_MAX];
1681 	char ibdev_name[IB_DEVICE_NAME_MAX];
1682 	const struct rdma_link_ops *ops;
1683 	char ndev_name[IFNAMSIZ];
1684 	struct net_device *ndev;
1685 	char type[IFNAMSIZ];
1686 	int err;
1687 
1688 	err = nlmsg_parse_deprecated(nlh, 0, tb, RDMA_NLDEV_ATTR_MAX - 1,
1689 				     nldev_policy, extack);
1690 	if (err || !tb[RDMA_NLDEV_ATTR_DEV_NAME] ||
1691 	    !tb[RDMA_NLDEV_ATTR_LINK_TYPE] || !tb[RDMA_NLDEV_ATTR_NDEV_NAME])
1692 		return -EINVAL;
1693 
1694 	nla_strscpy(ibdev_name, tb[RDMA_NLDEV_ATTR_DEV_NAME],
1695 		    sizeof(ibdev_name));
1696 	if (strchr(ibdev_name, '%') || strlen(ibdev_name) == 0)
1697 		return -EINVAL;
1698 
1699 	nla_strscpy(type, tb[RDMA_NLDEV_ATTR_LINK_TYPE], sizeof(type));
1700 	nla_strscpy(ndev_name, tb[RDMA_NLDEV_ATTR_NDEV_NAME],
1701 		    sizeof(ndev_name));
1702 
1703 	ndev = dev_get_by_name(sock_net(skb->sk), ndev_name);
1704 	if (!ndev)
1705 		return -ENODEV;
1706 
1707 	down_read(&link_ops_rwsem);
1708 	ops = link_ops_get(type);
1709 #ifdef CONFIG_MODULES
1710 	if (!ops) {
1711 		up_read(&link_ops_rwsem);
1712 		request_module("rdma-link-%s", type);
1713 		down_read(&link_ops_rwsem);
1714 		ops = link_ops_get(type);
1715 	}
1716 #endif
1717 	err = ops ? ops->newlink(ibdev_name, ndev) : -EINVAL;
1718 	up_read(&link_ops_rwsem);
1719 	dev_put(ndev);
1720 
1721 	return err;
1722 }
1723 
1724 static int nldev_dellink(struct sk_buff *skb, struct nlmsghdr *nlh,
1725 			  struct netlink_ext_ack *extack)
1726 {
1727 	struct nlattr *tb[RDMA_NLDEV_ATTR_MAX];
1728 	struct ib_device *device;
1729 	u32 index;
1730 	int err;
1731 
1732 	err = nlmsg_parse_deprecated(nlh, 0, tb, RDMA_NLDEV_ATTR_MAX - 1,
1733 				     nldev_policy, extack);
1734 	if (err || !tb[RDMA_NLDEV_ATTR_DEV_INDEX])
1735 		return -EINVAL;
1736 
1737 	index = nla_get_u32(tb[RDMA_NLDEV_ATTR_DEV_INDEX]);
1738 	device = ib_device_get_by_index(sock_net(skb->sk), index);
1739 	if (!device)
1740 		return -EINVAL;
1741 
1742 	if (!(device->attrs.kernel_cap_flags & IBK_ALLOW_USER_UNREG)) {
1743 		ib_device_put(device);
1744 		return -EINVAL;
1745 	}
1746 
1747 	ib_unregister_device_and_put(device);
1748 	return 0;
1749 }
1750 
1751 static int nldev_get_chardev(struct sk_buff *skb, struct nlmsghdr *nlh,
1752 			     struct netlink_ext_ack *extack)
1753 {
1754 	struct nlattr *tb[RDMA_NLDEV_ATTR_MAX];
1755 	char client_name[RDMA_NLDEV_ATTR_CHARDEV_TYPE_SIZE];
1756 	struct ib_client_nl_info data = {};
1757 	struct ib_device *ibdev = NULL;
1758 	struct sk_buff *msg;
1759 	u32 index;
1760 	int err;
1761 
1762 	err = nlmsg_parse(nlh, 0, tb, RDMA_NLDEV_ATTR_MAX - 1, nldev_policy,
1763 			  extack);
1764 	if (err || !tb[RDMA_NLDEV_ATTR_CHARDEV_TYPE])
1765 		return -EINVAL;
1766 
1767 	nla_strscpy(client_name, tb[RDMA_NLDEV_ATTR_CHARDEV_TYPE],
1768 		    sizeof(client_name));
1769 
1770 	if (tb[RDMA_NLDEV_ATTR_DEV_INDEX]) {
1771 		index = nla_get_u32(tb[RDMA_NLDEV_ATTR_DEV_INDEX]);
1772 		ibdev = ib_device_get_by_index(sock_net(skb->sk), index);
1773 		if (!ibdev)
1774 			return -EINVAL;
1775 
1776 		if (tb[RDMA_NLDEV_ATTR_PORT_INDEX]) {
1777 			data.port = nla_get_u32(tb[RDMA_NLDEV_ATTR_PORT_INDEX]);
1778 			if (!rdma_is_port_valid(ibdev, data.port)) {
1779 				err = -EINVAL;
1780 				goto out_put;
1781 			}
1782 		} else {
1783 			data.port = -1;
1784 		}
1785 	} else if (tb[RDMA_NLDEV_ATTR_PORT_INDEX]) {
1786 		return -EINVAL;
1787 	}
1788 
1789 	msg = nlmsg_new(NLMSG_DEFAULT_SIZE, GFP_KERNEL);
1790 	if (!msg) {
1791 		err = -ENOMEM;
1792 		goto out_put;
1793 	}
1794 	nlh = nlmsg_put(msg, NETLINK_CB(skb).portid, nlh->nlmsg_seq,
1795 			RDMA_NL_GET_TYPE(RDMA_NL_NLDEV,
1796 					 RDMA_NLDEV_CMD_GET_CHARDEV),
1797 			0, 0);
1798 
1799 	data.nl_msg = msg;
1800 	err = ib_get_client_nl_info(ibdev, client_name, &data);
1801 	if (err)
1802 		goto out_nlmsg;
1803 
1804 	err = nla_put_u64_64bit(msg, RDMA_NLDEV_ATTR_CHARDEV,
1805 				huge_encode_dev(data.cdev->devt),
1806 				RDMA_NLDEV_ATTR_PAD);
1807 	if (err)
1808 		goto out_data;
1809 	err = nla_put_u64_64bit(msg, RDMA_NLDEV_ATTR_CHARDEV_ABI, data.abi,
1810 				RDMA_NLDEV_ATTR_PAD);
1811 	if (err)
1812 		goto out_data;
1813 	if (nla_put_string(msg, RDMA_NLDEV_ATTR_CHARDEV_NAME,
1814 			   dev_name(data.cdev))) {
1815 		err = -EMSGSIZE;
1816 		goto out_data;
1817 	}
1818 
1819 	nlmsg_end(msg, nlh);
1820 	put_device(data.cdev);
1821 	if (ibdev)
1822 		ib_device_put(ibdev);
1823 	return rdma_nl_unicast(sock_net(skb->sk), msg, NETLINK_CB(skb).portid);
1824 
1825 out_data:
1826 	put_device(data.cdev);
1827 out_nlmsg:
1828 	nlmsg_free(msg);
1829 out_put:
1830 	if (ibdev)
1831 		ib_device_put(ibdev);
1832 	return err;
1833 }
1834 
1835 static int nldev_sys_get_doit(struct sk_buff *skb, struct nlmsghdr *nlh,
1836 			      struct netlink_ext_ack *extack)
1837 {
1838 	struct nlattr *tb[RDMA_NLDEV_ATTR_MAX];
1839 	struct sk_buff *msg;
1840 	int err;
1841 
1842 	err = nlmsg_parse(nlh, 0, tb, RDMA_NLDEV_ATTR_MAX - 1,
1843 			  nldev_policy, extack);
1844 	if (err)
1845 		return err;
1846 
1847 	msg = nlmsg_new(NLMSG_DEFAULT_SIZE, GFP_KERNEL);
1848 	if (!msg)
1849 		return -ENOMEM;
1850 
1851 	nlh = nlmsg_put(msg, NETLINK_CB(skb).portid, nlh->nlmsg_seq,
1852 			RDMA_NL_GET_TYPE(RDMA_NL_NLDEV,
1853 					 RDMA_NLDEV_CMD_SYS_GET),
1854 			0, 0);
1855 
1856 	err = nla_put_u8(msg, RDMA_NLDEV_SYS_ATTR_NETNS_MODE,
1857 			 (u8)ib_devices_shared_netns);
1858 	if (err) {
1859 		nlmsg_free(msg);
1860 		return err;
1861 	}
1862 
1863 	/*
1864 	 * Copy-on-fork is supported.
1865 	 * See commits:
1866 	 * 70e806e4e645 ("mm: Do early cow for pinned pages during fork() for ptes")
1867 	 * 4eae4efa2c29 ("hugetlb: do early cow when page pinned on src mm")
1868 	 * for more details. Don't backport this without them.
1869 	 *
1870 	 * Return value ignored on purpose, assume copy-on-fork is not
1871 	 * supported in case of failure.
1872 	 */
1873 	nla_put_u8(msg, RDMA_NLDEV_SYS_ATTR_COPY_ON_FORK, 1);
1874 
1875 	nlmsg_end(msg, nlh);
1876 	return rdma_nl_unicast(sock_net(skb->sk), msg, NETLINK_CB(skb).portid);
1877 }
1878 
1879 static int nldev_set_sys_set_doit(struct sk_buff *skb, struct nlmsghdr *nlh,
1880 				  struct netlink_ext_ack *extack)
1881 {
1882 	struct nlattr *tb[RDMA_NLDEV_ATTR_MAX];
1883 	u8 enable;
1884 	int err;
1885 
1886 	err = nlmsg_parse(nlh, 0, tb, RDMA_NLDEV_ATTR_MAX - 1,
1887 			  nldev_policy, extack);
1888 	if (err || !tb[RDMA_NLDEV_SYS_ATTR_NETNS_MODE])
1889 		return -EINVAL;
1890 
1891 	enable = nla_get_u8(tb[RDMA_NLDEV_SYS_ATTR_NETNS_MODE]);
1892 	/* Only 0 and 1 are supported */
1893 	if (enable > 1)
1894 		return -EINVAL;
1895 
1896 	err = rdma_compatdev_set(enable);
1897 	return err;
1898 }
1899 
1900 static int nldev_stat_set_mode_doit(struct sk_buff *msg,
1901 				    struct netlink_ext_ack *extack,
1902 				    struct nlattr *tb[],
1903 				    struct ib_device *device, u32 port)
1904 {
1905 	u32 mode, mask = 0, qpn, cntn = 0;
1906 	int ret;
1907 
1908 	/* Currently only counter for QP is supported */
1909 	if (!tb[RDMA_NLDEV_ATTR_STAT_RES] ||
1910 	    nla_get_u32(tb[RDMA_NLDEV_ATTR_STAT_RES]) != RDMA_NLDEV_ATTR_RES_QP)
1911 		return -EINVAL;
1912 
1913 	mode = nla_get_u32(tb[RDMA_NLDEV_ATTR_STAT_MODE]);
1914 	if (mode == RDMA_COUNTER_MODE_AUTO) {
1915 		if (tb[RDMA_NLDEV_ATTR_STAT_AUTO_MODE_MASK])
1916 			mask = nla_get_u32(
1917 				tb[RDMA_NLDEV_ATTR_STAT_AUTO_MODE_MASK]);
1918 		return rdma_counter_set_auto_mode(device, port, mask, extack);
1919 	}
1920 
1921 	if (!tb[RDMA_NLDEV_ATTR_RES_LQPN])
1922 		return -EINVAL;
1923 
1924 	qpn = nla_get_u32(tb[RDMA_NLDEV_ATTR_RES_LQPN]);
1925 	if (tb[RDMA_NLDEV_ATTR_STAT_COUNTER_ID]) {
1926 		cntn = nla_get_u32(tb[RDMA_NLDEV_ATTR_STAT_COUNTER_ID]);
1927 		ret = rdma_counter_bind_qpn(device, port, qpn, cntn);
1928 		if (ret)
1929 			return ret;
1930 	} else {
1931 		ret = rdma_counter_bind_qpn_alloc(device, port, qpn, &cntn);
1932 		if (ret)
1933 			return ret;
1934 	}
1935 
1936 	if (nla_put_u32(msg, RDMA_NLDEV_ATTR_STAT_COUNTER_ID, cntn) ||
1937 	    nla_put_u32(msg, RDMA_NLDEV_ATTR_RES_LQPN, qpn)) {
1938 		ret = -EMSGSIZE;
1939 		goto err_fill;
1940 	}
1941 
1942 	return 0;
1943 
1944 err_fill:
1945 	rdma_counter_unbind_qpn(device, port, qpn, cntn);
1946 	return ret;
1947 }
1948 
1949 static int nldev_stat_set_counter_dynamic_doit(struct nlattr *tb[],
1950 					       struct ib_device *device,
1951 					       u32 port)
1952 {
1953 	struct rdma_hw_stats *stats;
1954 	struct nlattr *entry_attr;
1955 	unsigned long *target;
1956 	int rem, i, ret = 0;
1957 	u32 index;
1958 
1959 	stats = ib_get_hw_stats_port(device, port);
1960 	if (!stats)
1961 		return -EINVAL;
1962 
1963 	target = kcalloc(BITS_TO_LONGS(stats->num_counters),
1964 			 sizeof(*stats->is_disabled), GFP_KERNEL);
1965 	if (!target)
1966 		return -ENOMEM;
1967 
1968 	nla_for_each_nested(entry_attr, tb[RDMA_NLDEV_ATTR_STAT_HWCOUNTERS],
1969 			    rem) {
1970 		index = nla_get_u32(entry_attr);
1971 		if ((index >= stats->num_counters) ||
1972 		    !(stats->descs[index].flags & IB_STAT_FLAG_OPTIONAL)) {
1973 			ret = -EINVAL;
1974 			goto out;
1975 		}
1976 
1977 		set_bit(index, target);
1978 	}
1979 
1980 	for (i = 0; i < stats->num_counters; i++) {
1981 		if (!(stats->descs[i].flags & IB_STAT_FLAG_OPTIONAL))
1982 			continue;
1983 
1984 		ret = rdma_counter_modify(device, port, i, test_bit(i, target));
1985 		if (ret)
1986 			goto out;
1987 	}
1988 
1989 out:
1990 	kfree(target);
1991 	return ret;
1992 }
1993 
1994 static int nldev_stat_set_doit(struct sk_buff *skb, struct nlmsghdr *nlh,
1995 			       struct netlink_ext_ack *extack)
1996 {
1997 	struct nlattr *tb[RDMA_NLDEV_ATTR_MAX];
1998 	struct ib_device *device;
1999 	struct sk_buff *msg;
2000 	u32 index, port;
2001 	int ret;
2002 
2003 	ret = nlmsg_parse(nlh, 0, tb, RDMA_NLDEV_ATTR_MAX - 1, nldev_policy,
2004 			  extack);
2005 	if (ret || !tb[RDMA_NLDEV_ATTR_DEV_INDEX] ||
2006 	    !tb[RDMA_NLDEV_ATTR_PORT_INDEX])
2007 		return -EINVAL;
2008 
2009 	index = nla_get_u32(tb[RDMA_NLDEV_ATTR_DEV_INDEX]);
2010 	device = ib_device_get_by_index(sock_net(skb->sk), index);
2011 	if (!device)
2012 		return -EINVAL;
2013 
2014 	port = nla_get_u32(tb[RDMA_NLDEV_ATTR_PORT_INDEX]);
2015 	if (!rdma_is_port_valid(device, port)) {
2016 		ret = -EINVAL;
2017 		goto err_put_device;
2018 	}
2019 
2020 	if (!tb[RDMA_NLDEV_ATTR_STAT_MODE] &&
2021 	    !tb[RDMA_NLDEV_ATTR_STAT_HWCOUNTERS]) {
2022 		ret = -EINVAL;
2023 		goto err_put_device;
2024 	}
2025 
2026 	msg = nlmsg_new(NLMSG_DEFAULT_SIZE, GFP_KERNEL);
2027 	if (!msg) {
2028 		ret = -ENOMEM;
2029 		goto err_put_device;
2030 	}
2031 	nlh = nlmsg_put(msg, NETLINK_CB(skb).portid, nlh->nlmsg_seq,
2032 			RDMA_NL_GET_TYPE(RDMA_NL_NLDEV,
2033 					 RDMA_NLDEV_CMD_STAT_SET),
2034 			0, 0);
2035 	if (fill_nldev_handle(msg, device) ||
2036 	    nla_put_u32(msg, RDMA_NLDEV_ATTR_PORT_INDEX, port)) {
2037 		ret = -EMSGSIZE;
2038 		goto err_free_msg;
2039 	}
2040 
2041 	if (tb[RDMA_NLDEV_ATTR_STAT_MODE]) {
2042 		ret = nldev_stat_set_mode_doit(msg, extack, tb, device, port);
2043 		if (ret)
2044 			goto err_free_msg;
2045 	}
2046 
2047 	if (tb[RDMA_NLDEV_ATTR_STAT_HWCOUNTERS]) {
2048 		ret = nldev_stat_set_counter_dynamic_doit(tb, device, port);
2049 		if (ret)
2050 			goto err_free_msg;
2051 	}
2052 
2053 	nlmsg_end(msg, nlh);
2054 	ib_device_put(device);
2055 	return rdma_nl_unicast(sock_net(skb->sk), msg, NETLINK_CB(skb).portid);
2056 
2057 err_free_msg:
2058 	nlmsg_free(msg);
2059 err_put_device:
2060 	ib_device_put(device);
2061 	return ret;
2062 }
2063 
2064 static int nldev_stat_del_doit(struct sk_buff *skb, struct nlmsghdr *nlh,
2065 			       struct netlink_ext_ack *extack)
2066 {
2067 	struct nlattr *tb[RDMA_NLDEV_ATTR_MAX];
2068 	struct ib_device *device;
2069 	struct sk_buff *msg;
2070 	u32 index, port, qpn, cntn;
2071 	int ret;
2072 
2073 	ret = nlmsg_parse(nlh, 0, tb, RDMA_NLDEV_ATTR_MAX - 1,
2074 			  nldev_policy, extack);
2075 	if (ret || !tb[RDMA_NLDEV_ATTR_STAT_RES] ||
2076 	    !tb[RDMA_NLDEV_ATTR_DEV_INDEX] || !tb[RDMA_NLDEV_ATTR_PORT_INDEX] ||
2077 	    !tb[RDMA_NLDEV_ATTR_STAT_COUNTER_ID] ||
2078 	    !tb[RDMA_NLDEV_ATTR_RES_LQPN])
2079 		return -EINVAL;
2080 
2081 	if (nla_get_u32(tb[RDMA_NLDEV_ATTR_STAT_RES]) != RDMA_NLDEV_ATTR_RES_QP)
2082 		return -EINVAL;
2083 
2084 	index = nla_get_u32(tb[RDMA_NLDEV_ATTR_DEV_INDEX]);
2085 	device = ib_device_get_by_index(sock_net(skb->sk), index);
2086 	if (!device)
2087 		return -EINVAL;
2088 
2089 	port = nla_get_u32(tb[RDMA_NLDEV_ATTR_PORT_INDEX]);
2090 	if (!rdma_is_port_valid(device, port)) {
2091 		ret = -EINVAL;
2092 		goto err;
2093 	}
2094 
2095 	msg = nlmsg_new(NLMSG_DEFAULT_SIZE, GFP_KERNEL);
2096 	if (!msg) {
2097 		ret = -ENOMEM;
2098 		goto err;
2099 	}
2100 	nlh = nlmsg_put(msg, NETLINK_CB(skb).portid, nlh->nlmsg_seq,
2101 			RDMA_NL_GET_TYPE(RDMA_NL_NLDEV,
2102 					 RDMA_NLDEV_CMD_STAT_SET),
2103 			0, 0);
2104 
2105 	cntn = nla_get_u32(tb[RDMA_NLDEV_ATTR_STAT_COUNTER_ID]);
2106 	qpn = nla_get_u32(tb[RDMA_NLDEV_ATTR_RES_LQPN]);
2107 	if (fill_nldev_handle(msg, device) ||
2108 	    nla_put_u32(msg, RDMA_NLDEV_ATTR_PORT_INDEX, port) ||
2109 	    nla_put_u32(msg, RDMA_NLDEV_ATTR_STAT_COUNTER_ID, cntn) ||
2110 	    nla_put_u32(msg, RDMA_NLDEV_ATTR_RES_LQPN, qpn)) {
2111 		ret = -EMSGSIZE;
2112 		goto err_fill;
2113 	}
2114 
2115 	ret = rdma_counter_unbind_qpn(device, port, qpn, cntn);
2116 	if (ret)
2117 		goto err_fill;
2118 
2119 	nlmsg_end(msg, nlh);
2120 	ib_device_put(device);
2121 	return rdma_nl_unicast(sock_net(skb->sk), msg, NETLINK_CB(skb).portid);
2122 
2123 err_fill:
2124 	nlmsg_free(msg);
2125 err:
2126 	ib_device_put(device);
2127 	return ret;
2128 }
2129 
2130 static int stat_get_doit_default_counter(struct sk_buff *skb,
2131 					 struct nlmsghdr *nlh,
2132 					 struct netlink_ext_ack *extack,
2133 					 struct nlattr *tb[])
2134 {
2135 	struct rdma_hw_stats *stats;
2136 	struct nlattr *table_attr;
2137 	struct ib_device *device;
2138 	int ret, num_cnts, i;
2139 	struct sk_buff *msg;
2140 	u32 index, port;
2141 	u64 v;
2142 
2143 	if (!tb[RDMA_NLDEV_ATTR_DEV_INDEX] || !tb[RDMA_NLDEV_ATTR_PORT_INDEX])
2144 		return -EINVAL;
2145 
2146 	index = nla_get_u32(tb[RDMA_NLDEV_ATTR_DEV_INDEX]);
2147 	device = ib_device_get_by_index(sock_net(skb->sk), index);
2148 	if (!device)
2149 		return -EINVAL;
2150 
2151 	if (!device->ops.alloc_hw_port_stats || !device->ops.get_hw_stats) {
2152 		ret = -EINVAL;
2153 		goto err;
2154 	}
2155 
2156 	port = nla_get_u32(tb[RDMA_NLDEV_ATTR_PORT_INDEX]);
2157 	stats = ib_get_hw_stats_port(device, port);
2158 	if (!stats) {
2159 		ret = -EINVAL;
2160 		goto err;
2161 	}
2162 
2163 	msg = nlmsg_new(NLMSG_DEFAULT_SIZE, GFP_KERNEL);
2164 	if (!msg) {
2165 		ret = -ENOMEM;
2166 		goto err;
2167 	}
2168 
2169 	nlh = nlmsg_put(msg, NETLINK_CB(skb).portid, nlh->nlmsg_seq,
2170 			RDMA_NL_GET_TYPE(RDMA_NL_NLDEV,
2171 					 RDMA_NLDEV_CMD_STAT_GET),
2172 			0, 0);
2173 
2174 	if (fill_nldev_handle(msg, device) ||
2175 	    nla_put_u32(msg, RDMA_NLDEV_ATTR_PORT_INDEX, port)) {
2176 		ret = -EMSGSIZE;
2177 		goto err_msg;
2178 	}
2179 
2180 	mutex_lock(&stats->lock);
2181 
2182 	num_cnts = device->ops.get_hw_stats(device, stats, port, 0);
2183 	if (num_cnts < 0) {
2184 		ret = -EINVAL;
2185 		goto err_stats;
2186 	}
2187 
2188 	table_attr = nla_nest_start(msg, RDMA_NLDEV_ATTR_STAT_HWCOUNTERS);
2189 	if (!table_attr) {
2190 		ret = -EMSGSIZE;
2191 		goto err_stats;
2192 	}
2193 	for (i = 0; i < num_cnts; i++) {
2194 		if (test_bit(i, stats->is_disabled))
2195 			continue;
2196 
2197 		v = stats->value[i] +
2198 			rdma_counter_get_hwstat_value(device, port, i);
2199 		if (rdma_nl_stat_hwcounter_entry(msg,
2200 						 stats->descs[i].name, v)) {
2201 			ret = -EMSGSIZE;
2202 			goto err_table;
2203 		}
2204 	}
2205 	nla_nest_end(msg, table_attr);
2206 
2207 	mutex_unlock(&stats->lock);
2208 	nlmsg_end(msg, nlh);
2209 	ib_device_put(device);
2210 	return rdma_nl_unicast(sock_net(skb->sk), msg, NETLINK_CB(skb).portid);
2211 
2212 err_table:
2213 	nla_nest_cancel(msg, table_attr);
2214 err_stats:
2215 	mutex_unlock(&stats->lock);
2216 err_msg:
2217 	nlmsg_free(msg);
2218 err:
2219 	ib_device_put(device);
2220 	return ret;
2221 }
2222 
2223 static int stat_get_doit_qp(struct sk_buff *skb, struct nlmsghdr *nlh,
2224 			    struct netlink_ext_ack *extack, struct nlattr *tb[])
2225 
2226 {
2227 	static enum rdma_nl_counter_mode mode;
2228 	static enum rdma_nl_counter_mask mask;
2229 	struct ib_device *device;
2230 	struct sk_buff *msg;
2231 	u32 index, port;
2232 	int ret;
2233 
2234 	if (tb[RDMA_NLDEV_ATTR_STAT_COUNTER_ID])
2235 		return nldev_res_get_counter_doit(skb, nlh, extack);
2236 
2237 	if (!tb[RDMA_NLDEV_ATTR_STAT_MODE] ||
2238 	    !tb[RDMA_NLDEV_ATTR_DEV_INDEX] || !tb[RDMA_NLDEV_ATTR_PORT_INDEX])
2239 		return -EINVAL;
2240 
2241 	index = nla_get_u32(tb[RDMA_NLDEV_ATTR_DEV_INDEX]);
2242 	device = ib_device_get_by_index(sock_net(skb->sk), index);
2243 	if (!device)
2244 		return -EINVAL;
2245 
2246 	port = nla_get_u32(tb[RDMA_NLDEV_ATTR_PORT_INDEX]);
2247 	if (!rdma_is_port_valid(device, port)) {
2248 		ret = -EINVAL;
2249 		goto err;
2250 	}
2251 
2252 	msg = nlmsg_new(NLMSG_DEFAULT_SIZE, GFP_KERNEL);
2253 	if (!msg) {
2254 		ret = -ENOMEM;
2255 		goto err;
2256 	}
2257 
2258 	nlh = nlmsg_put(msg, NETLINK_CB(skb).portid, nlh->nlmsg_seq,
2259 			RDMA_NL_GET_TYPE(RDMA_NL_NLDEV,
2260 					 RDMA_NLDEV_CMD_STAT_GET),
2261 			0, 0);
2262 
2263 	ret = rdma_counter_get_mode(device, port, &mode, &mask);
2264 	if (ret)
2265 		goto err_msg;
2266 
2267 	if (fill_nldev_handle(msg, device) ||
2268 	    nla_put_u32(msg, RDMA_NLDEV_ATTR_PORT_INDEX, port) ||
2269 	    nla_put_u32(msg, RDMA_NLDEV_ATTR_STAT_MODE, mode)) {
2270 		ret = -EMSGSIZE;
2271 		goto err_msg;
2272 	}
2273 
2274 	if ((mode == RDMA_COUNTER_MODE_AUTO) &&
2275 	    nla_put_u32(msg, RDMA_NLDEV_ATTR_STAT_AUTO_MODE_MASK, mask)) {
2276 		ret = -EMSGSIZE;
2277 		goto err_msg;
2278 	}
2279 
2280 	nlmsg_end(msg, nlh);
2281 	ib_device_put(device);
2282 	return rdma_nl_unicast(sock_net(skb->sk), msg, NETLINK_CB(skb).portid);
2283 
2284 err_msg:
2285 	nlmsg_free(msg);
2286 err:
2287 	ib_device_put(device);
2288 	return ret;
2289 }
2290 
2291 static int nldev_stat_get_doit(struct sk_buff *skb, struct nlmsghdr *nlh,
2292 			       struct netlink_ext_ack *extack)
2293 {
2294 	struct nlattr *tb[RDMA_NLDEV_ATTR_MAX];
2295 	int ret;
2296 
2297 	ret = nlmsg_parse(nlh, 0, tb, RDMA_NLDEV_ATTR_MAX - 1,
2298 			  nldev_policy, extack);
2299 	if (ret)
2300 		return -EINVAL;
2301 
2302 	if (!tb[RDMA_NLDEV_ATTR_STAT_RES])
2303 		return stat_get_doit_default_counter(skb, nlh, extack, tb);
2304 
2305 	switch (nla_get_u32(tb[RDMA_NLDEV_ATTR_STAT_RES])) {
2306 	case RDMA_NLDEV_ATTR_RES_QP:
2307 		ret = stat_get_doit_qp(skb, nlh, extack, tb);
2308 		break;
2309 	case RDMA_NLDEV_ATTR_RES_MR:
2310 		ret = res_get_common_doit(skb, nlh, extack, RDMA_RESTRACK_MR,
2311 					  fill_stat_mr_entry);
2312 		break;
2313 	default:
2314 		ret = -EINVAL;
2315 		break;
2316 	}
2317 
2318 	return ret;
2319 }
2320 
2321 static int nldev_stat_get_dumpit(struct sk_buff *skb,
2322 				 struct netlink_callback *cb)
2323 {
2324 	struct nlattr *tb[RDMA_NLDEV_ATTR_MAX];
2325 	int ret;
2326 
2327 	ret = nlmsg_parse(cb->nlh, 0, tb, RDMA_NLDEV_ATTR_MAX - 1,
2328 			  nldev_policy, NULL);
2329 	if (ret || !tb[RDMA_NLDEV_ATTR_STAT_RES])
2330 		return -EINVAL;
2331 
2332 	switch (nla_get_u32(tb[RDMA_NLDEV_ATTR_STAT_RES])) {
2333 	case RDMA_NLDEV_ATTR_RES_QP:
2334 		ret = nldev_res_get_counter_dumpit(skb, cb);
2335 		break;
2336 	case RDMA_NLDEV_ATTR_RES_MR:
2337 		ret = res_get_common_dumpit(skb, cb, RDMA_RESTRACK_MR,
2338 					    fill_stat_mr_entry);
2339 		break;
2340 	default:
2341 		ret = -EINVAL;
2342 		break;
2343 	}
2344 
2345 	return ret;
2346 }
2347 
2348 static int nldev_stat_get_counter_status_doit(struct sk_buff *skb,
2349 					      struct nlmsghdr *nlh,
2350 					      struct netlink_ext_ack *extack)
2351 {
2352 	struct nlattr *tb[RDMA_NLDEV_ATTR_MAX], *table, *entry;
2353 	struct rdma_hw_stats *stats;
2354 	struct ib_device *device;
2355 	struct sk_buff *msg;
2356 	u32 devid, port;
2357 	int ret, i;
2358 
2359 	ret = nlmsg_parse(nlh, 0, tb, RDMA_NLDEV_ATTR_MAX - 1,
2360 			  nldev_policy, extack);
2361 	if (ret || !tb[RDMA_NLDEV_ATTR_DEV_INDEX] ||
2362 	    !tb[RDMA_NLDEV_ATTR_PORT_INDEX])
2363 		return -EINVAL;
2364 
2365 	devid = nla_get_u32(tb[RDMA_NLDEV_ATTR_DEV_INDEX]);
2366 	device = ib_device_get_by_index(sock_net(skb->sk), devid);
2367 	if (!device)
2368 		return -EINVAL;
2369 
2370 	port = nla_get_u32(tb[RDMA_NLDEV_ATTR_PORT_INDEX]);
2371 	if (!rdma_is_port_valid(device, port)) {
2372 		ret = -EINVAL;
2373 		goto err;
2374 	}
2375 
2376 	stats = ib_get_hw_stats_port(device, port);
2377 	if (!stats) {
2378 		ret = -EINVAL;
2379 		goto err;
2380 	}
2381 
2382 	msg = nlmsg_new(NLMSG_DEFAULT_SIZE, GFP_KERNEL);
2383 	if (!msg) {
2384 		ret = -ENOMEM;
2385 		goto err;
2386 	}
2387 
2388 	nlh = nlmsg_put(
2389 		msg, NETLINK_CB(skb).portid, nlh->nlmsg_seq,
2390 		RDMA_NL_GET_TYPE(RDMA_NL_NLDEV, RDMA_NLDEV_CMD_STAT_GET_STATUS),
2391 		0, 0);
2392 
2393 	ret = -EMSGSIZE;
2394 	if (fill_nldev_handle(msg, device) ||
2395 	    nla_put_u32(msg, RDMA_NLDEV_ATTR_PORT_INDEX, port))
2396 		goto err_msg;
2397 
2398 	table = nla_nest_start(msg, RDMA_NLDEV_ATTR_STAT_HWCOUNTERS);
2399 	if (!table)
2400 		goto err_msg;
2401 
2402 	mutex_lock(&stats->lock);
2403 	for (i = 0; i < stats->num_counters; i++) {
2404 		entry = nla_nest_start(msg,
2405 				       RDMA_NLDEV_ATTR_STAT_HWCOUNTER_ENTRY);
2406 		if (!entry)
2407 			goto err_msg_table;
2408 
2409 		if (nla_put_string(msg,
2410 				   RDMA_NLDEV_ATTR_STAT_HWCOUNTER_ENTRY_NAME,
2411 				   stats->descs[i].name) ||
2412 		    nla_put_u32(msg, RDMA_NLDEV_ATTR_STAT_HWCOUNTER_INDEX, i))
2413 			goto err_msg_entry;
2414 
2415 		if ((stats->descs[i].flags & IB_STAT_FLAG_OPTIONAL) &&
2416 		    (nla_put_u8(msg, RDMA_NLDEV_ATTR_STAT_HWCOUNTER_DYNAMIC,
2417 				!test_bit(i, stats->is_disabled))))
2418 			goto err_msg_entry;
2419 
2420 		nla_nest_end(msg, entry);
2421 	}
2422 	mutex_unlock(&stats->lock);
2423 
2424 	nla_nest_end(msg, table);
2425 	nlmsg_end(msg, nlh);
2426 	ib_device_put(device);
2427 	return rdma_nl_unicast(sock_net(skb->sk), msg, NETLINK_CB(skb).portid);
2428 
2429 err_msg_entry:
2430 	nla_nest_cancel(msg, entry);
2431 err_msg_table:
2432 	mutex_unlock(&stats->lock);
2433 	nla_nest_cancel(msg, table);
2434 err_msg:
2435 	nlmsg_free(msg);
2436 err:
2437 	ib_device_put(device);
2438 	return ret;
2439 }
2440 
2441 static const struct rdma_nl_cbs nldev_cb_table[RDMA_NLDEV_NUM_OPS] = {
2442 	[RDMA_NLDEV_CMD_GET] = {
2443 		.doit = nldev_get_doit,
2444 		.dump = nldev_get_dumpit,
2445 	},
2446 	[RDMA_NLDEV_CMD_GET_CHARDEV] = {
2447 		.doit = nldev_get_chardev,
2448 	},
2449 	[RDMA_NLDEV_CMD_SET] = {
2450 		.doit = nldev_set_doit,
2451 		.flags = RDMA_NL_ADMIN_PERM,
2452 	},
2453 	[RDMA_NLDEV_CMD_NEWLINK] = {
2454 		.doit = nldev_newlink,
2455 		.flags = RDMA_NL_ADMIN_PERM,
2456 	},
2457 	[RDMA_NLDEV_CMD_DELLINK] = {
2458 		.doit = nldev_dellink,
2459 		.flags = RDMA_NL_ADMIN_PERM,
2460 	},
2461 	[RDMA_NLDEV_CMD_PORT_GET] = {
2462 		.doit = nldev_port_get_doit,
2463 		.dump = nldev_port_get_dumpit,
2464 	},
2465 	[RDMA_NLDEV_CMD_RES_GET] = {
2466 		.doit = nldev_res_get_doit,
2467 		.dump = nldev_res_get_dumpit,
2468 	},
2469 	[RDMA_NLDEV_CMD_RES_QP_GET] = {
2470 		.doit = nldev_res_get_qp_doit,
2471 		.dump = nldev_res_get_qp_dumpit,
2472 	},
2473 	[RDMA_NLDEV_CMD_RES_CM_ID_GET] = {
2474 		.doit = nldev_res_get_cm_id_doit,
2475 		.dump = nldev_res_get_cm_id_dumpit,
2476 	},
2477 	[RDMA_NLDEV_CMD_RES_CQ_GET] = {
2478 		.doit = nldev_res_get_cq_doit,
2479 		.dump = nldev_res_get_cq_dumpit,
2480 	},
2481 	[RDMA_NLDEV_CMD_RES_MR_GET] = {
2482 		.doit = nldev_res_get_mr_doit,
2483 		.dump = nldev_res_get_mr_dumpit,
2484 	},
2485 	[RDMA_NLDEV_CMD_RES_PD_GET] = {
2486 		.doit = nldev_res_get_pd_doit,
2487 		.dump = nldev_res_get_pd_dumpit,
2488 	},
2489 	[RDMA_NLDEV_CMD_RES_CTX_GET] = {
2490 		.doit = nldev_res_get_ctx_doit,
2491 		.dump = nldev_res_get_ctx_dumpit,
2492 	},
2493 	[RDMA_NLDEV_CMD_RES_SRQ_GET] = {
2494 		.doit = nldev_res_get_srq_doit,
2495 		.dump = nldev_res_get_srq_dumpit,
2496 	},
2497 	[RDMA_NLDEV_CMD_SYS_GET] = {
2498 		.doit = nldev_sys_get_doit,
2499 	},
2500 	[RDMA_NLDEV_CMD_SYS_SET] = {
2501 		.doit = nldev_set_sys_set_doit,
2502 	},
2503 	[RDMA_NLDEV_CMD_STAT_SET] = {
2504 		.doit = nldev_stat_set_doit,
2505 		.flags = RDMA_NL_ADMIN_PERM,
2506 	},
2507 	[RDMA_NLDEV_CMD_STAT_GET] = {
2508 		.doit = nldev_stat_get_doit,
2509 		.dump = nldev_stat_get_dumpit,
2510 	},
2511 	[RDMA_NLDEV_CMD_STAT_DEL] = {
2512 		.doit = nldev_stat_del_doit,
2513 		.flags = RDMA_NL_ADMIN_PERM,
2514 	},
2515 	[RDMA_NLDEV_CMD_RES_QP_GET_RAW] = {
2516 		.doit = nldev_res_get_qp_raw_doit,
2517 		.dump = nldev_res_get_qp_raw_dumpit,
2518 		.flags = RDMA_NL_ADMIN_PERM,
2519 	},
2520 	[RDMA_NLDEV_CMD_RES_CQ_GET_RAW] = {
2521 		.doit = nldev_res_get_cq_raw_doit,
2522 		.dump = nldev_res_get_cq_raw_dumpit,
2523 		.flags = RDMA_NL_ADMIN_PERM,
2524 	},
2525 	[RDMA_NLDEV_CMD_RES_MR_GET_RAW] = {
2526 		.doit = nldev_res_get_mr_raw_doit,
2527 		.dump = nldev_res_get_mr_raw_dumpit,
2528 		.flags = RDMA_NL_ADMIN_PERM,
2529 	},
2530 	[RDMA_NLDEV_CMD_STAT_GET_STATUS] = {
2531 		.doit = nldev_stat_get_counter_status_doit,
2532 	},
2533 };
2534 
2535 void __init nldev_init(void)
2536 {
2537 	rdma_nl_register(RDMA_NL_NLDEV, nldev_cb_table);
2538 }
2539 
2540 void __exit nldev_exit(void)
2541 {
2542 	rdma_nl_unregister(RDMA_NL_NLDEV);
2543 }
2544 
2545 MODULE_ALIAS_RDMA_NETLINK(RDMA_NL_NLDEV, 5);
2546