2 * Copyright (c) 2017 Mellanox Technologies. All rights reserved.
4 * Redistribution and use in source and binary forms, with or without
5 * modification, are permitted provided that the following conditions are met:
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.
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.
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.
33 #include <linux/module.h>
34 #include <linux/pid.h>
35 #include <linux/pid_namespace.h>
36 #include <net/netlink.h>
37 #include <rdma/rdma_netlink.h>
39 #include "core_priv.h"
41 static const struct nla_policy nldev_policy[RDMA_NLDEV_ATTR_MAX] = {
42 [RDMA_NLDEV_ATTR_DEV_INDEX] = { .type = NLA_U32 },
43 [RDMA_NLDEV_ATTR_DEV_NAME] = { .type = NLA_NUL_STRING,
44 .len = IB_DEVICE_NAME_MAX - 1},
45 [RDMA_NLDEV_ATTR_PORT_INDEX] = { .type = NLA_U32 },
46 [RDMA_NLDEV_ATTR_FW_VERSION] = { .type = NLA_NUL_STRING,
47 .len = IB_FW_VERSION_NAME_MAX - 1},
48 [RDMA_NLDEV_ATTR_NODE_GUID] = { .type = NLA_U64 },
49 [RDMA_NLDEV_ATTR_SYS_IMAGE_GUID] = { .type = NLA_U64 },
50 [RDMA_NLDEV_ATTR_SUBNET_PREFIX] = { .type = NLA_U64 },
51 [RDMA_NLDEV_ATTR_LID] = { .type = NLA_U32 },
52 [RDMA_NLDEV_ATTR_SM_LID] = { .type = NLA_U32 },
53 [RDMA_NLDEV_ATTR_LMC] = { .type = NLA_U8 },
54 [RDMA_NLDEV_ATTR_PORT_STATE] = { .type = NLA_U8 },
55 [RDMA_NLDEV_ATTR_PORT_PHYS_STATE] = { .type = NLA_U8 },
56 [RDMA_NLDEV_ATTR_DEV_NODE_TYPE] = { .type = NLA_U8 },
57 [RDMA_NLDEV_ATTR_RES_SUMMARY] = { .type = NLA_NESTED },
58 [RDMA_NLDEV_ATTR_RES_SUMMARY_ENTRY] = { .type = NLA_NESTED },
59 [RDMA_NLDEV_ATTR_RES_SUMMARY_ENTRY_NAME] = { .type = NLA_NUL_STRING,
61 [RDMA_NLDEV_ATTR_RES_SUMMARY_ENTRY_CURR] = { .type = NLA_U64 },
62 [RDMA_NLDEV_ATTR_RES_QP] = { .type = NLA_NESTED },
63 [RDMA_NLDEV_ATTR_RES_QP_ENTRY] = { .type = NLA_NESTED },
64 [RDMA_NLDEV_ATTR_RES_LQPN] = { .type = NLA_U32 },
65 [RDMA_NLDEV_ATTR_RES_RQPN] = { .type = NLA_U32 },
66 [RDMA_NLDEV_ATTR_RES_RQ_PSN] = { .type = NLA_U32 },
67 [RDMA_NLDEV_ATTR_RES_SQ_PSN] = { .type = NLA_U32 },
68 [RDMA_NLDEV_ATTR_RES_PATH_MIG_STATE] = { .type = NLA_U8 },
69 [RDMA_NLDEV_ATTR_RES_TYPE] = { .type = NLA_U8 },
70 [RDMA_NLDEV_ATTR_RES_STATE] = { .type = NLA_U8 },
71 [RDMA_NLDEV_ATTR_RES_PID] = { .type = NLA_U32 },
72 [RDMA_NLDEV_ATTR_RES_KERN_NAME] = { .type = NLA_NUL_STRING,
73 .len = TASK_COMM_LEN },
76 static int fill_nldev_handle(struct sk_buff *msg, struct ib_device *device)
78 if (nla_put_u32(msg, RDMA_NLDEV_ATTR_DEV_INDEX, device->index))
80 if (nla_put_string(msg, RDMA_NLDEV_ATTR_DEV_NAME, device->name))
86 static int fill_dev_info(struct sk_buff *msg, struct ib_device *device)
88 char fw[IB_FW_VERSION_NAME_MAX];
90 if (fill_nldev_handle(msg, device))
93 if (nla_put_u32(msg, RDMA_NLDEV_ATTR_PORT_INDEX, rdma_end_port(device)))
96 BUILD_BUG_ON(sizeof(device->attrs.device_cap_flags) != sizeof(u64));
97 if (nla_put_u64_64bit(msg, RDMA_NLDEV_ATTR_CAP_FLAGS,
98 device->attrs.device_cap_flags, 0))
101 ib_get_device_fw_str(device, fw);
102 /* Device without FW has strlen(fw) */
103 if (strlen(fw) && nla_put_string(msg, RDMA_NLDEV_ATTR_FW_VERSION, fw))
106 if (nla_put_u64_64bit(msg, RDMA_NLDEV_ATTR_NODE_GUID,
107 be64_to_cpu(device->node_guid), 0))
109 if (nla_put_u64_64bit(msg, RDMA_NLDEV_ATTR_SYS_IMAGE_GUID,
110 be64_to_cpu(device->attrs.sys_image_guid), 0))
112 if (nla_put_u8(msg, RDMA_NLDEV_ATTR_DEV_NODE_TYPE, device->node_type))
117 static int fill_port_info(struct sk_buff *msg,
118 struct ib_device *device, u32 port)
120 struct ib_port_attr attr;
123 if (fill_nldev_handle(msg, device))
126 if (nla_put_u32(msg, RDMA_NLDEV_ATTR_PORT_INDEX, port))
129 ret = ib_query_port(device, port, &attr);
133 BUILD_BUG_ON(sizeof(attr.port_cap_flags) > sizeof(u64));
134 if (nla_put_u64_64bit(msg, RDMA_NLDEV_ATTR_CAP_FLAGS,
135 (u64)attr.port_cap_flags, 0))
137 if (rdma_protocol_ib(device, port) &&
138 nla_put_u64_64bit(msg, RDMA_NLDEV_ATTR_SUBNET_PREFIX,
139 attr.subnet_prefix, 0))
141 if (rdma_protocol_ib(device, port)) {
142 if (nla_put_u32(msg, RDMA_NLDEV_ATTR_LID, attr.lid))
144 if (nla_put_u32(msg, RDMA_NLDEV_ATTR_SM_LID, attr.sm_lid))
146 if (nla_put_u8(msg, RDMA_NLDEV_ATTR_LMC, attr.lmc))
149 if (nla_put_u8(msg, RDMA_NLDEV_ATTR_PORT_STATE, attr.state))
151 if (nla_put_u8(msg, RDMA_NLDEV_ATTR_PORT_PHYS_STATE, attr.phys_state))
156 static int fill_res_info_entry(struct sk_buff *msg,
157 const char *name, u64 curr)
159 struct nlattr *entry_attr;
161 entry_attr = nla_nest_start(msg, RDMA_NLDEV_ATTR_RES_SUMMARY_ENTRY);
165 if (nla_put_string(msg, RDMA_NLDEV_ATTR_RES_SUMMARY_ENTRY_NAME, name))
167 if (nla_put_u64_64bit(msg,
168 RDMA_NLDEV_ATTR_RES_SUMMARY_ENTRY_CURR, curr, 0))
171 nla_nest_end(msg, entry_attr);
175 nla_nest_cancel(msg, entry_attr);
179 static int fill_res_info(struct sk_buff *msg, struct ib_device *device)
181 static const char * const names[RDMA_RESTRACK_MAX] = {
182 [RDMA_RESTRACK_PD] = "pd",
183 [RDMA_RESTRACK_CQ] = "cq",
184 [RDMA_RESTRACK_QP] = "qp",
187 struct rdma_restrack_root *res = &device->res;
188 struct nlattr *table_attr;
191 if (fill_nldev_handle(msg, device))
194 table_attr = nla_nest_start(msg, RDMA_NLDEV_ATTR_RES_SUMMARY);
198 for (i = 0; i < RDMA_RESTRACK_MAX; i++) {
201 curr = rdma_restrack_count(res, i, task_active_pid_ns(current));
202 ret = fill_res_info_entry(msg, names[i], curr);
207 nla_nest_end(msg, table_attr);
211 nla_nest_cancel(msg, table_attr);
215 static int fill_res_qp_entry(struct sk_buff *msg,
216 struct ib_qp *qp, uint32_t port)
218 struct rdma_restrack_entry *res = &qp->res;
219 struct ib_qp_init_attr qp_init_attr;
220 struct nlattr *entry_attr;
221 struct ib_qp_attr qp_attr;
224 ret = ib_query_qp(qp, &qp_attr, 0, &qp_init_attr);
228 if (port && port != qp_attr.port_num)
231 entry_attr = nla_nest_start(msg, RDMA_NLDEV_ATTR_RES_QP_ENTRY);
235 /* In create_qp() port is not set yet */
236 if (qp_attr.port_num &&
237 nla_put_u32(msg, RDMA_NLDEV_ATTR_PORT_INDEX, qp_attr.port_num))
240 if (nla_put_u32(msg, RDMA_NLDEV_ATTR_RES_LQPN, qp->qp_num))
242 if (qp->qp_type == IB_QPT_RC || qp->qp_type == IB_QPT_UC) {
243 if (nla_put_u32(msg, RDMA_NLDEV_ATTR_RES_RQPN,
244 qp_attr.dest_qp_num))
246 if (nla_put_u32(msg, RDMA_NLDEV_ATTR_RES_RQ_PSN,
251 if (nla_put_u32(msg, RDMA_NLDEV_ATTR_RES_SQ_PSN, qp_attr.sq_psn))
254 if (qp->qp_type == IB_QPT_RC || qp->qp_type == IB_QPT_UC ||
255 qp->qp_type == IB_QPT_XRC_INI || qp->qp_type == IB_QPT_XRC_TGT) {
256 if (nla_put_u8(msg, RDMA_NLDEV_ATTR_RES_PATH_MIG_STATE,
257 qp_attr.path_mig_state))
260 if (nla_put_u8(msg, RDMA_NLDEV_ATTR_RES_TYPE, qp->qp_type))
262 if (nla_put_u8(msg, RDMA_NLDEV_ATTR_RES_STATE, qp_attr.qp_state))
266 * Existence of task means that it is user QP and netlink
267 * user is invited to go and read /proc/PID/comm to get name
268 * of the task file and res->task_com should be NULL.
270 if (rdma_is_kernel_res(res)) {
271 if (nla_put_string(msg, RDMA_NLDEV_ATTR_RES_KERN_NAME, res->kern_name))
274 if (nla_put_u32(msg, RDMA_NLDEV_ATTR_RES_PID, task_pid_vnr(res->task)))
278 nla_nest_end(msg, entry_attr);
282 nla_nest_cancel(msg, entry_attr);
287 static int nldev_get_doit(struct sk_buff *skb, struct nlmsghdr *nlh,
288 struct netlink_ext_ack *extack)
290 struct nlattr *tb[RDMA_NLDEV_ATTR_MAX];
291 struct ib_device *device;
296 err = nlmsg_parse(nlh, 0, tb, RDMA_NLDEV_ATTR_MAX - 1,
297 nldev_policy, extack);
298 if (err || !tb[RDMA_NLDEV_ATTR_DEV_INDEX])
301 index = nla_get_u32(tb[RDMA_NLDEV_ATTR_DEV_INDEX]);
303 device = ib_device_get_by_index(index);
307 msg = nlmsg_new(NLMSG_DEFAULT_SIZE, GFP_KERNEL);
313 nlh = nlmsg_put(msg, NETLINK_CB(skb).portid, nlh->nlmsg_seq,
314 RDMA_NL_GET_TYPE(RDMA_NL_NLDEV, RDMA_NLDEV_CMD_GET),
317 err = fill_dev_info(msg, device);
323 put_device(&device->dev);
324 return rdma_nl_unicast(msg, NETLINK_CB(skb).portid);
329 put_device(&device->dev);
333 static int _nldev_get_dumpit(struct ib_device *device,
335 struct netlink_callback *cb,
338 int start = cb->args[0];
339 struct nlmsghdr *nlh;
344 nlh = nlmsg_put(skb, NETLINK_CB(cb->skb).portid, cb->nlh->nlmsg_seq,
345 RDMA_NL_GET_TYPE(RDMA_NL_NLDEV, RDMA_NLDEV_CMD_GET),
348 if (fill_dev_info(skb, device)) {
349 nlmsg_cancel(skb, nlh);
357 out: cb->args[0] = idx;
361 static int nldev_get_dumpit(struct sk_buff *skb, struct netlink_callback *cb)
364 * There is no need to take lock, because
365 * we are relying on ib_core's lists_rwsem
367 return ib_enum_all_devs(_nldev_get_dumpit, skb, cb);
370 static int nldev_port_get_doit(struct sk_buff *skb, struct nlmsghdr *nlh,
371 struct netlink_ext_ack *extack)
373 struct nlattr *tb[RDMA_NLDEV_ATTR_MAX];
374 struct ib_device *device;
380 err = nlmsg_parse(nlh, 0, tb, RDMA_NLDEV_ATTR_MAX - 1,
381 nldev_policy, extack);
383 !tb[RDMA_NLDEV_ATTR_DEV_INDEX] ||
384 !tb[RDMA_NLDEV_ATTR_PORT_INDEX])
387 index = nla_get_u32(tb[RDMA_NLDEV_ATTR_DEV_INDEX]);
388 device = ib_device_get_by_index(index);
392 port = nla_get_u32(tb[RDMA_NLDEV_ATTR_PORT_INDEX]);
393 if (!rdma_is_port_valid(device, port)) {
398 msg = nlmsg_new(NLMSG_DEFAULT_SIZE, GFP_KERNEL);
404 nlh = nlmsg_put(msg, NETLINK_CB(skb).portid, nlh->nlmsg_seq,
405 RDMA_NL_GET_TYPE(RDMA_NL_NLDEV, RDMA_NLDEV_CMD_GET),
408 err = fill_port_info(msg, device, port);
413 put_device(&device->dev);
415 return rdma_nl_unicast(msg, NETLINK_CB(skb).portid);
420 put_device(&device->dev);
424 static int nldev_port_get_dumpit(struct sk_buff *skb,
425 struct netlink_callback *cb)
427 struct nlattr *tb[RDMA_NLDEV_ATTR_MAX];
428 struct ib_device *device;
429 int start = cb->args[0];
430 struct nlmsghdr *nlh;
436 err = nlmsg_parse(cb->nlh, 0, tb, RDMA_NLDEV_ATTR_MAX - 1,
438 if (err || !tb[RDMA_NLDEV_ATTR_DEV_INDEX])
441 ifindex = nla_get_u32(tb[RDMA_NLDEV_ATTR_DEV_INDEX]);
442 device = ib_device_get_by_index(ifindex);
446 for (p = rdma_start_port(device); p <= rdma_end_port(device); ++p) {
448 * The dumpit function returns all information from specific
449 * index. This specific index is taken from the netlink
450 * messages request sent by user and it is available
453 * Usually, the user doesn't fill this field and it causes
454 * to return everything.
462 nlh = nlmsg_put(skb, NETLINK_CB(cb->skb).portid,
464 RDMA_NL_GET_TYPE(RDMA_NL_NLDEV,
465 RDMA_NLDEV_CMD_PORT_GET),
468 if (fill_port_info(skb, device, p)) {
469 nlmsg_cancel(skb, nlh);
477 put_device(&device->dev);
482 static int nldev_res_get_doit(struct sk_buff *skb, struct nlmsghdr *nlh,
483 struct netlink_ext_ack *extack)
485 struct nlattr *tb[RDMA_NLDEV_ATTR_MAX];
486 struct ib_device *device;
491 ret = nlmsg_parse(nlh, 0, tb, RDMA_NLDEV_ATTR_MAX - 1,
492 nldev_policy, extack);
493 if (ret || !tb[RDMA_NLDEV_ATTR_DEV_INDEX])
496 index = nla_get_u32(tb[RDMA_NLDEV_ATTR_DEV_INDEX]);
497 device = ib_device_get_by_index(index);
501 msg = nlmsg_new(NLMSG_DEFAULT_SIZE, GFP_KERNEL);
505 nlh = nlmsg_put(msg, NETLINK_CB(skb).portid, nlh->nlmsg_seq,
506 RDMA_NL_GET_TYPE(RDMA_NL_NLDEV, RDMA_NLDEV_CMD_RES_GET),
509 ret = fill_res_info(msg, device);
514 put_device(&device->dev);
515 return rdma_nl_unicast(msg, NETLINK_CB(skb).portid);
520 put_device(&device->dev);
524 static int _nldev_res_get_dumpit(struct ib_device *device,
526 struct netlink_callback *cb,
529 int start = cb->args[0];
530 struct nlmsghdr *nlh;
535 nlh = nlmsg_put(skb, NETLINK_CB(cb->skb).portid, cb->nlh->nlmsg_seq,
536 RDMA_NL_GET_TYPE(RDMA_NL_NLDEV, RDMA_NLDEV_CMD_RES_GET),
539 if (fill_res_info(skb, device)) {
540 nlmsg_cancel(skb, nlh);
553 static int nldev_res_get_dumpit(struct sk_buff *skb,
554 struct netlink_callback *cb)
556 return ib_enum_all_devs(_nldev_res_get_dumpit, skb, cb);
559 static int nldev_res_get_qp_dumpit(struct sk_buff *skb,
560 struct netlink_callback *cb)
562 struct nlattr *tb[RDMA_NLDEV_ATTR_MAX];
563 struct rdma_restrack_entry *res;
564 int err, ret = 0, idx = 0;
565 struct nlattr *table_attr;
566 struct ib_device *device;
567 int start = cb->args[0];
568 struct ib_qp *qp = NULL;
569 struct nlmsghdr *nlh;
572 err = nlmsg_parse(cb->nlh, 0, tb, RDMA_NLDEV_ATTR_MAX - 1,
575 * Right now, we are expecting the device index to get QP information,
576 * but it is possible to extend this code to return all devices in
577 * one shot by checking the existence of RDMA_NLDEV_ATTR_DEV_INDEX.
578 * if it doesn't exist, we will iterate over all devices.
580 * But it is not needed for now.
582 if (err || !tb[RDMA_NLDEV_ATTR_DEV_INDEX])
585 index = nla_get_u32(tb[RDMA_NLDEV_ATTR_DEV_INDEX]);
586 device = ib_device_get_by_index(index);
591 * If no PORT_INDEX is supplied, we will return all QPs from that device
593 if (tb[RDMA_NLDEV_ATTR_PORT_INDEX]) {
594 port = nla_get_u32(tb[RDMA_NLDEV_ATTR_PORT_INDEX]);
595 if (!rdma_is_port_valid(device, port)) {
601 nlh = nlmsg_put(skb, NETLINK_CB(cb->skb).portid, cb->nlh->nlmsg_seq,
602 RDMA_NL_GET_TYPE(RDMA_NL_NLDEV, RDMA_NLDEV_CMD_RES_QP_GET),
605 if (fill_nldev_handle(skb, device)) {
610 table_attr = nla_nest_start(skb, RDMA_NLDEV_ATTR_RES_QP);
616 down_read(&device->res.rwsem);
617 hash_for_each_possible(device->res.hash, res, node, RDMA_RESTRACK_QP) {
621 if ((rdma_is_kernel_res(res) &&
622 task_active_pid_ns(current) != &init_pid_ns) ||
623 (!rdma_is_kernel_res(res) &&
624 task_active_pid_ns(current) != task_active_pid_ns(res->task)))
626 * 1. Kernel QPs should be visible in init namspace only
627 * 2. Present only QPs visible in the current namespace
631 if (!rdma_restrack_get(res))
633 * Resource is under release now, but we are not
634 * relesing lock now, so it will be released in
635 * our next pass, once we will get ->next pointer.
639 qp = container_of(res, struct ib_qp, res);
641 up_read(&device->res.rwsem);
642 ret = fill_res_qp_entry(skb, qp, port);
643 down_read(&device->res.rwsem);
645 * Return resource back, but it won't be released till
646 * the &device->res.rwsem will be released for write.
648 rdma_restrack_put(res);
650 if (ret == -EMSGSIZE)
652 * There is a chance to optimize here.
653 * It can be done by using list_prepare_entry
654 * and list_for_each_entry_continue afterwards.
661 up_read(&device->res.rwsem);
663 nla_nest_end(skb, table_attr);
668 * No more QPs to fill, cancel the message and
669 * return 0 to mark end of dumpit.
674 put_device(&device->dev);
678 nla_nest_cancel(skb, table_attr);
679 up_read(&device->res.rwsem);
682 nlmsg_cancel(skb, nlh);
685 put_device(&device->dev);
689 static const struct rdma_nl_cbs nldev_cb_table[RDMA_NLDEV_NUM_OPS] = {
690 [RDMA_NLDEV_CMD_GET] = {
691 .doit = nldev_get_doit,
692 .dump = nldev_get_dumpit,
694 [RDMA_NLDEV_CMD_PORT_GET] = {
695 .doit = nldev_port_get_doit,
696 .dump = nldev_port_get_dumpit,
698 [RDMA_NLDEV_CMD_RES_GET] = {
699 .doit = nldev_res_get_doit,
700 .dump = nldev_res_get_dumpit,
702 [RDMA_NLDEV_CMD_RES_QP_GET] = {
703 .dump = nldev_res_get_qp_dumpit,
705 * .doit is not implemented yet for two reasons:
706 * 1. It is not needed yet.
707 * 2. There is a need to provide identifier, while it is easy
708 * for the QPs (device index + port index + LQPN), it is not
709 * the case for the rest of resources (PD and CQ). Because it
710 * is better to provide similar interface for all resources,
711 * let's wait till we will have other resources implemented
717 void __init nldev_init(void)
719 rdma_nl_register(RDMA_NL_NLDEV, nldev_cb_table);
722 void __exit nldev_exit(void)
724 rdma_nl_unregister(RDMA_NL_NLDEV);
727 MODULE_ALIAS_RDMA_NETLINK(RDMA_NL_NLDEV, 5);