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