~hc/RK356X_SDK_RELEASE.git

..	..	@@ -33,80 +33,119 @@
33	33	#include <linux/module.h>
34	34	#include <linux/pid.h>
35	35	#include <linux/pid_namespace.h>
	36	+#include <linux/mutex.h>
36	37	#include <net/netlink.h>
37	38	#include <rdma/rdma_cm.h>
38	39	#include <rdma/rdma_netlink.h>
39	40
40	41	#include "core_priv.h"
41	42	#include "cma_priv.h"
	43	+#include "restrack.h"
	44	+#include "uverbs.h"
42	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	+ */
43	52	static const struct nla_policy nldev_policy[RDMA_NLDEV_ATTR_MAX] = {
44		- [RDMA_NLDEV_ATTR_DEV_INDEX] = { .type = NLA_U32 },
45		- [RDMA_NLDEV_ATTR_DEV_NAME] = { .type = NLA_NUL_STRING,
46		- .len = IB_DEVICE_NAME_MAX - 1},
47		- [RDMA_NLDEV_ATTR_PORT_INDEX] = { .type = NLA_U32 },
48		- [RDMA_NLDEV_ATTR_FW_VERSION] = { .type = NLA_NUL_STRING,
49		- .len = IB_FW_VERSION_NAME_MAX - 1},
50		- [RDMA_NLDEV_ATTR_NODE_GUID] = { .type = NLA_U64 },
51		- [RDMA_NLDEV_ATTR_SYS_IMAGE_GUID] = { .type = NLA_U64 },
52		- [RDMA_NLDEV_ATTR_SUBNET_PREFIX] = { .type = NLA_U64 },
53		- [RDMA_NLDEV_ATTR_LID] = { .type = NLA_U32 },
54		- [RDMA_NLDEV_ATTR_SM_LID] = { .type = NLA_U32 },
55		- [RDMA_NLDEV_ATTR_LMC] = { .type = NLA_U8 },
56		- [RDMA_NLDEV_ATTR_PORT_STATE] = { .type = NLA_U8 },
57		- [RDMA_NLDEV_ATTR_PORT_PHYS_STATE] = { .type = NLA_U8 },
58		- [RDMA_NLDEV_ATTR_DEV_NODE_TYPE] = { .type = NLA_U8 },
59		- [RDMA_NLDEV_ATTR_RES_SUMMARY] = { .type = NLA_NESTED },
60		- [RDMA_NLDEV_ATTR_RES_SUMMARY_ENTRY] = { .type = NLA_NESTED },
61		- [RDMA_NLDEV_ATTR_RES_SUMMARY_ENTRY_NAME] = { .type = NLA_NUL_STRING,
62		- .len = 16 },
63		- [RDMA_NLDEV_ATTR_RES_SUMMARY_ENTRY_CURR] = { .type = NLA_U64 },
	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 },
64	115	[RDMA_NLDEV_ATTR_RES_QP] = { .type = NLA_NESTED },
65	116	[RDMA_NLDEV_ATTR_RES_QP_ENTRY] = { .type = NLA_NESTED },
66		- [RDMA_NLDEV_ATTR_RES_LQPN] = { .type = NLA_U32 },
	117	+ [RDMA_NLDEV_ATTR_RES_RAW] = { .type = NLA_BINARY },
	118	+ [RDMA_NLDEV_ATTR_RES_RKEY] = { .type = NLA_U32 },
67	119	[RDMA_NLDEV_ATTR_RES_RQPN] = { .type = NLA_U32 },
68	120	[RDMA_NLDEV_ATTR_RES_RQ_PSN] = { .type = NLA_U32 },
69	121	[RDMA_NLDEV_ATTR_RES_SQ_PSN] = { .type = NLA_U32 },
70		- [RDMA_NLDEV_ATTR_RES_PATH_MIG_STATE] = { .type = NLA_U8 },
71		- [RDMA_NLDEV_ATTR_RES_TYPE] = { .type = NLA_U8 },
	122	+ [RDMA_NLDEV_ATTR_RES_SRC_ADDR] = {
	123	+ .len = sizeof(struct __kernel_sockaddr_storage) },
72	124	[RDMA_NLDEV_ATTR_RES_STATE] = { .type = NLA_U8 },
73		- [RDMA_NLDEV_ATTR_RES_PID] = { .type = NLA_U32 },
74		- [RDMA_NLDEV_ATTR_RES_KERN_NAME] = { .type = NLA_NUL_STRING,
75		- .len = TASK_COMM_LEN },
76		- [RDMA_NLDEV_ATTR_RES_CM_ID] = { .type = NLA_NESTED },
77		- [RDMA_NLDEV_ATTR_RES_CM_ID_ENTRY] = { .type = NLA_NESTED },
78		- [RDMA_NLDEV_ATTR_RES_PS] = { .type = NLA_U32 },
79		- [RDMA_NLDEV_ATTR_RES_SRC_ADDR] = {
80		- .len = sizeof(struct __kernel_sockaddr_storage) },
81		- [RDMA_NLDEV_ATTR_RES_DST_ADDR] = {
82		- .len = sizeof(struct __kernel_sockaddr_storage) },
83		- [RDMA_NLDEV_ATTR_RES_CQ] = { .type = NLA_NESTED },
84		- [RDMA_NLDEV_ATTR_RES_CQ_ENTRY] = { .type = NLA_NESTED },
85		- [RDMA_NLDEV_ATTR_RES_CQE] = { .type = NLA_U32 },
	125	+ [RDMA_NLDEV_ATTR_RES_SUMMARY] = { .type = NLA_NESTED },
	126	+ [RDMA_NLDEV_ATTR_RES_SUMMARY_ENTRY] = { .type = NLA_NESTED },
	127	+ [RDMA_NLDEV_ATTR_RES_SUMMARY_ENTRY_CURR]= { .type = NLA_U64 },
	128	+ [RDMA_NLDEV_ATTR_RES_SUMMARY_ENTRY_NAME]= { .type = NLA_NUL_STRING,
	129	+ .len = RDMA_NLDEV_ATTR_EMPTY_STRING },
	130	+ [RDMA_NLDEV_ATTR_RES_TYPE] = { .type = NLA_U8 },
	131	+ [RDMA_NLDEV_ATTR_RES_UNSAFE_GLOBAL_RKEY]= { .type = NLA_U32 },
86	132	[RDMA_NLDEV_ATTR_RES_USECNT] = { .type = NLA_U64 },
87		- [RDMA_NLDEV_ATTR_RES_POLL_CTX] = { .type = NLA_U8 },
88		- [RDMA_NLDEV_ATTR_RES_MR] = { .type = NLA_NESTED },
89		- [RDMA_NLDEV_ATTR_RES_MR_ENTRY] = { .type = NLA_NESTED },
90		- [RDMA_NLDEV_ATTR_RES_RKEY] = { .type = NLA_U32 },
91		- [RDMA_NLDEV_ATTR_RES_LKEY] = { .type = NLA_U32 },
92		- [RDMA_NLDEV_ATTR_RES_IOVA] = { .type = NLA_U64 },
93		- [RDMA_NLDEV_ATTR_RES_MRLEN] = { .type = NLA_U64 },
94		- [RDMA_NLDEV_ATTR_RES_PD] = { .type = NLA_NESTED },
95		- [RDMA_NLDEV_ATTR_RES_PD_ENTRY] = { .type = NLA_NESTED },
96		- [RDMA_NLDEV_ATTR_RES_LOCAL_DMA_LKEY] = { .type = NLA_U32 },
97		- [RDMA_NLDEV_ATTR_RES_UNSAFE_GLOBAL_RKEY] = { .type = NLA_U32 },
98		- [RDMA_NLDEV_ATTR_NDEV_INDEX] = { .type = NLA_U32 },
99		- [RDMA_NLDEV_ATTR_NDEV_NAME] = { .type = NLA_NUL_STRING,
100		- .len = IFNAMSIZ },
101		- [RDMA_NLDEV_ATTR_DRIVER] = { .type = NLA_NESTED },
102		- [RDMA_NLDEV_ATTR_DRIVER_ENTRY] = { .type = NLA_NESTED },
103		- [RDMA_NLDEV_ATTR_DRIVER_STRING] = { .type = NLA_NUL_STRING,
104		- .len = RDMA_NLDEV_ATTR_ENTRY_STRLEN },
105		- [RDMA_NLDEV_ATTR_DRIVER_PRINT_TYPE] = { .type = NLA_U8 },
106		- [RDMA_NLDEV_ATTR_DRIVER_S32] = { .type = NLA_S32 },
107		- [RDMA_NLDEV_ATTR_DRIVER_U32] = { .type = NLA_U32 },
108		- [RDMA_NLDEV_ATTR_DRIVER_S64] = { .type = NLA_S64 },
109		- [RDMA_NLDEV_ATTR_DRIVER_U64] = { .type = NLA_U64 },
	133	+ [RDMA_NLDEV_ATTR_SM_LID] = { .type = NLA_U32 },
	134	+ [RDMA_NLDEV_ATTR_SUBNET_PREFIX] = { .type = NLA_U64 },
	135	+ [RDMA_NLDEV_ATTR_STAT_AUTO_MODE_MASK] = { .type = NLA_U32 },
	136	+ [RDMA_NLDEV_ATTR_STAT_MODE] = { .type = NLA_U32 },
	137	+ [RDMA_NLDEV_ATTR_STAT_RES] = { .type = NLA_U32 },
	138	+ [RDMA_NLDEV_ATTR_STAT_COUNTER] = { .type = NLA_NESTED },
	139	+ [RDMA_NLDEV_ATTR_STAT_COUNTER_ENTRY] = { .type = NLA_NESTED },
	140	+ [RDMA_NLDEV_ATTR_STAT_COUNTER_ID] = { .type = NLA_U32 },
	141	+ [RDMA_NLDEV_ATTR_STAT_HWCOUNTERS] = { .type = NLA_NESTED },
	142	+ [RDMA_NLDEV_ATTR_STAT_HWCOUNTER_ENTRY] = { .type = NLA_NESTED },
	143	+ [RDMA_NLDEV_ATTR_STAT_HWCOUNTER_ENTRY_NAME] = { .type = NLA_NUL_STRING },
	144	+ [RDMA_NLDEV_ATTR_STAT_HWCOUNTER_ENTRY_VALUE] = { .type = NLA_U64 },
	145	+ [RDMA_NLDEV_ATTR_SYS_IMAGE_GUID] = { .type = NLA_U64 },
	146	+ [RDMA_NLDEV_ATTR_UVERBS_DRIVER_ID] = { .type = NLA_U32 },
	147	+ [RDMA_NLDEV_NET_NS_FD] = { .type = NLA_U32 },
	148	+ [RDMA_NLDEV_SYS_ATTR_NETNS_MODE] = { .type = NLA_U8 },
110	149	};
111	150
112	151	static int put_driver_name_print_type(struct sk_buff msg, const char name,
..	..	@@ -146,6 +185,19 @@
146	185	return 0;
147	186	}
148	187
	188	+int rdma_nl_put_driver_string(struct sk_buff msg, const char name,
	189	+ const char *str)
	190	+{
	191	+ if (put_driver_name_print_type(msg, name,
	192	+ RDMA_NLDEV_PRINT_TYPE_UNSPEC))
	193	+ return -EMSGSIZE;
	194	+ if (nla_put_string(msg, RDMA_NLDEV_ATTR_DRIVER_STRING, str))
	195	+ return -EMSGSIZE;
	196	+
	197	+ return 0;
	198	+}
	199	+EXPORT_SYMBOL(rdma_nl_put_driver_string);
	200	+
149	201	int rdma_nl_put_driver_u32(struct sk_buff msg, const char name, u32 value)
150	202	{
151	203	return _rdma_nl_put_driver_u32(msg, name, RDMA_NLDEV_PRINT_TYPE_UNSPEC,
..	..	@@ -179,7 +231,8 @@
179	231	{
180	232	if (nla_put_u32(msg, RDMA_NLDEV_ATTR_DEV_INDEX, device->index))
181	233	return -EMSGSIZE;
182		- if (nla_put_string(msg, RDMA_NLDEV_ATTR_DEV_NAME, device->name))
	234	+ if (nla_put_string(msg, RDMA_NLDEV_ATTR_DEV_NAME,
	235	+ dev_name(&device->dev)))
183	236	return -EMSGSIZE;
184	237
185	238	return 0;
..	..	@@ -188,6 +241,8 @@
188	241	static int fill_dev_info(struct sk_buff msg, struct ib_device device)
189	242	{
190	243	char fw[IB_FW_VERSION_NAME_MAX];
	244	+ int ret = 0;
	245	+ u8 port;
191	246
192	247	if (fill_nldev_handle(msg, device))
193	248	return -EMSGSIZE;
..	..	@@ -216,7 +271,27 @@
216	271	return -EMSGSIZE;
217	272	if (nla_put_u8(msg, RDMA_NLDEV_ATTR_DEV_NODE_TYPE, device->node_type))
218	273	return -EMSGSIZE;
219		- return 0;
	274	+ if (nla_put_u8(msg, RDMA_NLDEV_ATTR_DEV_DIM, device->use_cq_dim))
	275	+ return -EMSGSIZE;
	276	+
	277	+ /*
	278	+ * Link type is determined on first port and mlx4 device
	279	+ * which can potentially have two different link type for the same
	280	+ * IB device is considered as better to be avoided in the future,
	281	+ */
	282	+ port = rdma_start_port(device);
	283	+ if (rdma_cap_opa_mad(device, port))
	284	+ ret = nla_put_string(msg, RDMA_NLDEV_ATTR_DEV_PROTOCOL, "opa");
	285	+ else if (rdma_protocol_ib(device, port))
	286	+ ret = nla_put_string(msg, RDMA_NLDEV_ATTR_DEV_PROTOCOL, "ib");
	287	+ else if (rdma_protocol_iwarp(device, port))
	288	+ ret = nla_put_string(msg, RDMA_NLDEV_ATTR_DEV_PROTOCOL, "iw");
	289	+ else if (rdma_protocol_roce(device, port))
	290	+ ret = nla_put_string(msg, RDMA_NLDEV_ATTR_DEV_PROTOCOL, "roce");
	291	+ else if (rdma_protocol_usnic(device, port))
	292	+ ret = nla_put_string(msg, RDMA_NLDEV_ATTR_DEV_PROTOCOL,
	293	+ "usnic");
	294	+ return ret;
220	295	}
221	296
222	297	static int fill_port_info(struct sk_buff *msg,
..	..	@@ -226,6 +301,7 @@
226	301	struct net_device *netdev = NULL;
227	302	struct ib_port_attr attr;
228	303	int ret;
	304	+ u64 cap_flags = 0;
229	305
230	306	if (fill_nldev_handle(msg, device))
231	307	return -EMSGSIZE;
..	..	@@ -238,10 +314,12 @@
238	314	return ret;
239	315
240	316	if (rdma_protocol_ib(device, port)) {
241		- BUILD_BUG_ON(sizeof(attr.port_cap_flags) > sizeof(u64));
	317	+ BUILD_BUG_ON((sizeof(attr.port_cap_flags) +
	318	+ sizeof(attr.port_cap_flags2)) > sizeof(u64));
	319	+ cap_flags = attr.port_cap_flags \|
	320	+ ((u64)attr.port_cap_flags2 << 32);
242	321	if (nla_put_u64_64bit(msg, RDMA_NLDEV_ATTR_CAP_FLAGS,
243		- (u64)attr.port_cap_flags,
244		- RDMA_NLDEV_ATTR_PAD))
	322	+ cap_flags, RDMA_NLDEV_ATTR_PAD))
245	323	return -EMSGSIZE;
246	324	if (nla_put_u64_64bit(msg, RDMA_NLDEV_ATTR_SUBNET_PREFIX,
247	325	attr.subnet_prefix, RDMA_NLDEV_ATTR_PAD))
..	..	@@ -258,9 +336,7 @@
258	336	if (nla_put_u8(msg, RDMA_NLDEV_ATTR_PORT_PHYS_STATE, attr.phys_state))
259	337	return -EMSGSIZE;
260	338
261		- if (device->get_netdev)
262		- netdev = device->get_netdev(device, port);
263		-
	339	+ netdev = ib_device_get_netdev(device, port);
264	340	if (netdev && net_eq(dev_net(netdev), net)) {
265	341	ret = nla_put_u32(msg,
266	342	RDMA_NLDEV_ATTR_NDEV_INDEX, netdev->ifindex);
..	..	@@ -281,7 +357,8 @@
281	357	{
282	358	struct nlattr *entry_attr;
283	359
284		- entry_attr = nla_nest_start(msg, RDMA_NLDEV_ATTR_RES_SUMMARY_ENTRY);
	360	+ entry_attr = nla_nest_start_noflag(msg,
	361	+ RDMA_NLDEV_ATTR_RES_SUMMARY_ENTRY);
285	362	if (!entry_attr)
286	363	return -EMSGSIZE;
287	364
..	..	@@ -307,23 +384,23 @@
307	384	[RDMA_RESTRACK_QP] = "qp",
308	385	[RDMA_RESTRACK_CM_ID] = "cm_id",
309	386	[RDMA_RESTRACK_MR] = "mr",
	387	+ [RDMA_RESTRACK_CTX] = "ctx",
310	388	};
311	389
312		- struct rdma_restrack_root *res = &device->res;
313	390	struct nlattr *table_attr;
314	391	int ret, i, curr;
315	392
316	393	if (fill_nldev_handle(msg, device))
317	394	return -EMSGSIZE;
318	395
319		- table_attr = nla_nest_start(msg, RDMA_NLDEV_ATTR_RES_SUMMARY);
	396	+ table_attr = nla_nest_start_noflag(msg, RDMA_NLDEV_ATTR_RES_SUMMARY);
320	397	if (!table_attr)
321	398	return -EMSGSIZE;
322	399
323	400	for (i = 0; i < RDMA_RESTRACK_MAX; i++) {
324	401	if (!names[i])
325	402	continue;
326		- curr = rdma_restrack_count(res, i, task_active_pid_ns(current));
	403	+ curr = rdma_restrack_count(device, i);
327	404	ret = fill_res_info_entry(msg, names[i], curr);
328	405	if (ret)
329	406	goto err;
..	..	@@ -340,29 +417,42 @@
340	417	static int fill_res_name_pid(struct sk_buff *msg,
341	418	struct rdma_restrack_entry *res)
342	419	{
	420	+ int err = 0;
	421	+
343	422	/*
344	423	* For user resources, user is should read /proc/PID/comm to get the
345	424	* name of the task file.
346	425	*/
347	426	if (rdma_is_kernel_res(res)) {
348		- if (nla_put_string(msg, RDMA_NLDEV_ATTR_RES_KERN_NAME,
349		- res->kern_name))
350		- return -EMSGSIZE;
	427	+ err = nla_put_string(msg, RDMA_NLDEV_ATTR_RES_KERN_NAME,
	428	+ res->kern_name);
351	429	} else {
352		- if (nla_put_u32(msg, RDMA_NLDEV_ATTR_RES_PID,
353		- task_pid_vnr(res->task)))
354		- return -EMSGSIZE;
	430	+ pid_t pid;
	431	+
	432	+ pid = task_pid_vnr(res->task);
	433	+ /*
	434	+ * Task is dead and in zombie state.
	435	+ * There is no need to print PID anymore.
	436	+ */
	437	+ if (pid)
	438	+ /*
	439	+ * This part is racy, task can be killed and PID will
	440	+ * be zero right here but it is ok, next query won't
	441	+ * return PID. We don't promise real-time reflection
	442	+ * of SW objects.
	443	+ */
	444	+ err = nla_put_u32(msg, RDMA_NLDEV_ATTR_RES_PID, pid);
355	445	}
356		- return 0;
	446	+
	447	+ return err ? -EMSGSIZE : 0;
357	448	}
358	449
359		-static int fill_res_qp_entry(struct sk_buff msg, struct netlink_callback cb,
360		- struct rdma_restrack_entry *res, uint32_t port)
	450	+static int fill_res_qp_entry_query(struct sk_buff *msg,
	451	+ struct rdma_restrack_entry *res,
	452	+ struct ib_device *dev,
	453	+ struct ib_qp *qp)
361	454	{
362		- struct ib_qp *qp = container_of(res, struct ib_qp, res);
363		- struct rdma_restrack_root *resroot = &qp->device->res;
364	455	struct ib_qp_init_attr qp_init_attr;
365		- struct nlattr *entry_attr;
366	456	struct ib_qp_attr qp_attr;
367	457	int ret;
368	458
..	..	@@ -370,20 +460,6 @@
370	460	if (ret)
371	461	return ret;
372	462
373		- if (port && port != qp_attr.port_num)
374		- return 0;
375		-
376		- entry_attr = nla_nest_start(msg, RDMA_NLDEV_ATTR_RES_QP_ENTRY);
377		- if (!entry_attr)
378		- goto out;
379		-
380		- /* In create_qp() port is not set yet */
381		- if (qp_attr.port_num &&
382		- nla_put_u32(msg, RDMA_NLDEV_ATTR_PORT_INDEX, qp_attr.port_num))
383		- goto err;
384		-
385		- if (nla_put_u32(msg, RDMA_NLDEV_ATTR_RES_LQPN, qp->qp_num))
386		- goto err;
387	463	if (qp->qp_type == IB_QPT_RC \|\| qp->qp_type == IB_QPT_UC) {
388	464	if (nla_put_u32(msg, RDMA_NLDEV_ATTR_RES_RQPN,
389	465	qp_attr.dest_qp_num))
..	..	@@ -407,37 +483,65 @@
407	483	if (nla_put_u8(msg, RDMA_NLDEV_ATTR_RES_STATE, qp_attr.qp_state))
408	484	goto err;
409	485
410		- if (fill_res_name_pid(msg, res))
411		- goto err;
412		-
413		- if (resroot->fill_res_entry(msg, res))
414		- goto err;
415		-
416		- nla_nest_end(msg, entry_attr);
	486	+ if (dev->ops.fill_res_qp_entry)
	487	+ return dev->ops.fill_res_qp_entry(msg, qp);
417	488	return 0;
418	489
419		-err:
420		- nla_nest_cancel(msg, entry_attr);
421		-out:
422		- return -EMSGSIZE;
	490	+err: return -EMSGSIZE;
423	491	}
424	492
425		-static int fill_res_cm_id_entry(struct sk_buff *msg,
426		- struct netlink_callback *cb,
	493	+static int fill_res_qp_entry(struct sk_buff *msg, bool has_cap_net_admin,
	494	+ struct rdma_restrack_entry *res, uint32_t port)
	495	+{
	496	+ struct ib_qp *qp = container_of(res, struct ib_qp, res);
	497	+ struct ib_device *dev = qp->device;
	498	+ int ret;
	499	+
	500	+ if (port && port != qp->port)
	501	+ return -EAGAIN;
	502	+
	503	+ /* In create_qp() port is not set yet */
	504	+ if (qp->port && nla_put_u32(msg, RDMA_NLDEV_ATTR_PORT_INDEX, qp->port))
	505	+ return -EINVAL;
	506	+
	507	+ ret = nla_put_u32(msg, RDMA_NLDEV_ATTR_RES_LQPN, qp->qp_num);
	508	+ if (ret)
	509	+ return -EMSGSIZE;
	510	+
	511	+ if (!rdma_is_kernel_res(res) &&
	512	+ nla_put_u32(msg, RDMA_NLDEV_ATTR_RES_PDN, qp->pd->res.id))
	513	+ return -EMSGSIZE;
	514	+
	515	+ ret = fill_res_name_pid(msg, res);
	516	+ if (ret)
	517	+ return -EMSGSIZE;
	518	+
	519	+ return fill_res_qp_entry_query(msg, res, dev, qp);
	520	+}
	521	+
	522	+static int fill_res_qp_raw_entry(struct sk_buff *msg, bool has_cap_net_admin,
	523	+ struct rdma_restrack_entry *res, uint32_t port)
	524	+{
	525	+ struct ib_qp *qp = container_of(res, struct ib_qp, res);
	526	+ struct ib_device *dev = qp->device;
	527	+
	528	+ if (port && port != qp->port)
	529	+ return -EAGAIN;
	530	+ if (!dev->ops.fill_res_qp_entry_raw)
	531	+ return -EINVAL;
	532	+ return dev->ops.fill_res_qp_entry_raw(msg, qp);
	533	+}
	534	+
	535	+static int fill_res_cm_id_entry(struct sk_buff *msg, bool has_cap_net_admin,
427	536	struct rdma_restrack_entry *res, uint32_t port)
428	537	{
429	538	struct rdma_id_private *id_priv =
430	539	container_of(res, struct rdma_id_private, res);
431		- struct rdma_restrack_root *resroot = &id_priv->id.device->res;
	540	+ struct ib_device *dev = id_priv->id.device;
432	541	struct rdma_cm_id *cm_id = &id_priv->id;
433		- struct nlattr *entry_attr;
434	542
435	543	if (port && port != cm_id->port_num)
436	544	return 0;
437		-
438		- entry_attr = nla_nest_start(msg, RDMA_NLDEV_ATTR_RES_CM_ID_ENTRY);
439		- if (!entry_attr)
440		- goto out;
441	545
442	546	if (cm_id->port_num &&
443	547	nla_put_u32(msg, RDMA_NLDEV_ATTR_PORT_INDEX, cm_id->port_num))
..	..	@@ -467,107 +571,113 @@
467	571	&cm_id->route.addr.dst_addr))
468	572	goto err;
469	573
	574	+ if (nla_put_u32(msg, RDMA_NLDEV_ATTR_RES_CM_IDN, res->id))
	575	+ goto err;
	576	+
470	577	if (fill_res_name_pid(msg, res))
471	578	goto err;
472	579
473		- if (resroot->fill_res_entry(msg, res))
474		- goto err;
475		-
476		- nla_nest_end(msg, entry_attr);
	580	+ if (dev->ops.fill_res_cm_id_entry)
	581	+ return dev->ops.fill_res_cm_id_entry(msg, cm_id);
477	582	return 0;
478	583
479		-err:
480		- nla_nest_cancel(msg, entry_attr);
481		-out:
482		- return -EMSGSIZE;
	584	+err: return -EMSGSIZE;
483	585	}
484	586
485		-static int fill_res_cq_entry(struct sk_buff msg, struct netlink_callback cb,
	587	+static int fill_res_cq_entry(struct sk_buff *msg, bool has_cap_net_admin,
486	588	struct rdma_restrack_entry *res, uint32_t port)
487	589	{
488	590	struct ib_cq *cq = container_of(res, struct ib_cq, res);
489		- struct rdma_restrack_root *resroot = &cq->device->res;
490		- struct nlattr *entry_attr;
491		-
492		- entry_attr = nla_nest_start(msg, RDMA_NLDEV_ATTR_RES_CQ_ENTRY);
493		- if (!entry_attr)
494		- goto out;
	591	+ struct ib_device *dev = cq->device;
495	592
496	593	if (nla_put_u32(msg, RDMA_NLDEV_ATTR_RES_CQE, cq->cqe))
497		- goto err;
	594	+ return -EMSGSIZE;
498	595	if (nla_put_u64_64bit(msg, RDMA_NLDEV_ATTR_RES_USECNT,
499	596	atomic_read(&cq->usecnt), RDMA_NLDEV_ATTR_PAD))
500		- goto err;
	597	+ return -EMSGSIZE;
501	598
502	599	/* Poll context is only valid for kernel CQs */
503	600	if (rdma_is_kernel_res(res) &&
504	601	nla_put_u8(msg, RDMA_NLDEV_ATTR_RES_POLL_CTX, cq->poll_ctx))
505		- goto err;
	602	+ return -EMSGSIZE;
	603	+
	604	+ if (nla_put_u8(msg, RDMA_NLDEV_ATTR_DEV_DIM, (cq->dim != NULL)))
	605	+ return -EMSGSIZE;
	606	+
	607	+ if (nla_put_u32(msg, RDMA_NLDEV_ATTR_RES_CQN, res->id))
	608	+ return -EMSGSIZE;
	609	+ if (!rdma_is_kernel_res(res) &&
	610	+ nla_put_u32(msg, RDMA_NLDEV_ATTR_RES_CTXN,
	611	+ cq->uobject->uevent.uobject.context->res.id))
	612	+ return -EMSGSIZE;
506	613
507	614	if (fill_res_name_pid(msg, res))
508		- goto err;
	615	+ return -EMSGSIZE;
509	616
510		- if (resroot->fill_res_entry(msg, res))
511		- goto err;
512		-
513		- nla_nest_end(msg, entry_attr);
514		- return 0;
515		-
516		-err:
517		- nla_nest_cancel(msg, entry_attr);
518		-out:
519		- return -EMSGSIZE;
	617	+ return (dev->ops.fill_res_cq_entry) ?
	618	+ dev->ops.fill_res_cq_entry(msg, cq) : 0;
520	619	}
521	620
522		-static int fill_res_mr_entry(struct sk_buff msg, struct netlink_callback cb,
	621	+static int fill_res_cq_raw_entry(struct sk_buff *msg, bool has_cap_net_admin,
	622	+ struct rdma_restrack_entry *res, uint32_t port)
	623	+{
	624	+ struct ib_cq *cq = container_of(res, struct ib_cq, res);
	625	+ struct ib_device *dev = cq->device;
	626	+
	627	+ if (!dev->ops.fill_res_cq_entry_raw)
	628	+ return -EINVAL;
	629	+ return dev->ops.fill_res_cq_entry_raw(msg, cq);
	630	+}
	631	+
	632	+static int fill_res_mr_entry(struct sk_buff *msg, bool has_cap_net_admin,
523	633	struct rdma_restrack_entry *res, uint32_t port)
524	634	{
525	635	struct ib_mr *mr = container_of(res, struct ib_mr, res);
526		- struct rdma_restrack_root *resroot = &mr->pd->device->res;
527		- struct nlattr *entry_attr;
	636	+ struct ib_device *dev = mr->pd->device;
528	637
529		- entry_attr = nla_nest_start(msg, RDMA_NLDEV_ATTR_RES_MR_ENTRY);
530		- if (!entry_attr)
531		- goto out;
532		-
533		- if (netlink_capable(cb->skb, CAP_NET_ADMIN)) {
	638	+ if (has_cap_net_admin) {
534	639	if (nla_put_u32(msg, RDMA_NLDEV_ATTR_RES_RKEY, mr->rkey))
535		- goto err;
	640	+ return -EMSGSIZE;
536	641	if (nla_put_u32(msg, RDMA_NLDEV_ATTR_RES_LKEY, mr->lkey))
537		- goto err;
	642	+ return -EMSGSIZE;
538	643	}
539	644
540	645	if (nla_put_u64_64bit(msg, RDMA_NLDEV_ATTR_RES_MRLEN, mr->length,
541	646	RDMA_NLDEV_ATTR_PAD))
542		- goto err;
	647	+ return -EMSGSIZE;
	648	+
	649	+ if (nla_put_u32(msg, RDMA_NLDEV_ATTR_RES_MRN, res->id))
	650	+ return -EMSGSIZE;
	651	+
	652	+ if (!rdma_is_kernel_res(res) &&
	653	+ nla_put_u32(msg, RDMA_NLDEV_ATTR_RES_PDN, mr->pd->res.id))
	654	+ return -EMSGSIZE;
543	655
544	656	if (fill_res_name_pid(msg, res))
545		- goto err;
	657	+ return -EMSGSIZE;
546	658
547		- if (resroot->fill_res_entry(msg, res))
548		- goto err;
549		-
550		- nla_nest_end(msg, entry_attr);
551		- return 0;
552		-
553		-err:
554		- nla_nest_cancel(msg, entry_attr);
555		-out:
556		- return -EMSGSIZE;
	659	+ return (dev->ops.fill_res_mr_entry) ?
	660	+ dev->ops.fill_res_mr_entry(msg, mr) :
	661	+ 0;
557	662	}
558	663
559		-static int fill_res_pd_entry(struct sk_buff msg, struct netlink_callback cb,
	664	+static int fill_res_mr_raw_entry(struct sk_buff *msg, bool has_cap_net_admin,
	665	+ struct rdma_restrack_entry *res, uint32_t port)
	666	+{
	667	+ struct ib_mr *mr = container_of(res, struct ib_mr, res);
	668	+ struct ib_device *dev = mr->pd->device;
	669	+
	670	+ if (!dev->ops.fill_res_mr_entry_raw)
	671	+ return -EINVAL;
	672	+ return dev->ops.fill_res_mr_entry_raw(msg, mr);
	673	+}
	674	+
	675	+static int fill_res_pd_entry(struct sk_buff *msg, bool has_cap_net_admin,
560	676	struct rdma_restrack_entry *res, uint32_t port)
561	677	{
562	678	struct ib_pd *pd = container_of(res, struct ib_pd, res);
563		- struct rdma_restrack_root *resroot = &pd->device->res;
564		- struct nlattr *entry_attr;
565	679
566		- entry_attr = nla_nest_start(msg, RDMA_NLDEV_ATTR_RES_PD_ENTRY);
567		- if (!entry_attr)
568		- goto out;
569		-
570		- if (netlink_capable(cb->skb, CAP_NET_ADMIN)) {
	680	+ if (has_cap_net_admin) {
571	681	if (nla_put_u32(msg, RDMA_NLDEV_ATTR_RES_LOCAL_DMA_LKEY,
572	682	pd->local_dma_lkey))
573	683	goto err;
..	..	@@ -580,10 +690,49 @@
580	690	atomic_read(&pd->usecnt), RDMA_NLDEV_ATTR_PAD))
581	691	goto err;
582	692
583		- if (fill_res_name_pid(msg, res))
	693	+ if (nla_put_u32(msg, RDMA_NLDEV_ATTR_RES_PDN, res->id))
584	694	goto err;
585	695
586		- if (resroot->fill_res_entry(msg, res))
	696	+ if (!rdma_is_kernel_res(res) &&
	697	+ nla_put_u32(msg, RDMA_NLDEV_ATTR_RES_CTXN,
	698	+ pd->uobject->context->res.id))
	699	+ goto err;
	700	+
	701	+ return fill_res_name_pid(msg, res);
	702	+
	703	+err: return -EMSGSIZE;
	704	+}
	705	+
	706	+static int fill_stat_counter_mode(struct sk_buff *msg,
	707	+ struct rdma_counter *counter)
	708	+{
	709	+ struct rdma_counter_mode *m = &counter->mode;
	710	+
	711	+ if (nla_put_u32(msg, RDMA_NLDEV_ATTR_STAT_MODE, m->mode))
	712	+ return -EMSGSIZE;
	713	+
	714	+ if (m->mode == RDMA_COUNTER_MODE_AUTO) {
	715	+ if ((m->mask & RDMA_COUNTER_MASK_QP_TYPE) &&
	716	+ nla_put_u8(msg, RDMA_NLDEV_ATTR_RES_TYPE, m->param.qp_type))
	717	+ return -EMSGSIZE;
	718	+
	719	+ if ((m->mask & RDMA_COUNTER_MASK_PID) &&
	720	+ fill_res_name_pid(msg, &counter->res))
	721	+ return -EMSGSIZE;
	722	+ }
	723	+
	724	+ return 0;
	725	+}
	726	+
	727	+static int fill_stat_counter_qp_entry(struct sk_buff *msg, u32 qpn)
	728	+{
	729	+ struct nlattr *entry_attr;
	730	+
	731	+ entry_attr = nla_nest_start(msg, RDMA_NLDEV_ATTR_RES_QP_ENTRY);
	732	+ if (!entry_attr)
	733	+ return -EMSGSIZE;
	734	+
	735	+ if (nla_put_u32(msg, RDMA_NLDEV_ATTR_RES_LQPN, qpn))
587	736	goto err;
588	737
589	738	nla_nest_end(msg, entry_attr);
..	..	@@ -591,8 +740,129 @@
591	740
592	741	err:
593	742	nla_nest_cancel(msg, entry_attr);
594		-out:
595	743	return -EMSGSIZE;
	744	+}
	745	+
	746	+static int fill_stat_counter_qps(struct sk_buff *msg,
	747	+ struct rdma_counter *counter)
	748	+{
	749	+ struct rdma_restrack_entry *res;
	750	+ struct rdma_restrack_root *rt;
	751	+ struct nlattr *table_attr;
	752	+ struct ib_qp *qp = NULL;
	753	+ unsigned long id = 0;
	754	+ int ret = 0;
	755	+
	756	+ table_attr = nla_nest_start(msg, RDMA_NLDEV_ATTR_RES_QP);
	757	+
	758	+ rt = &counter->device->res[RDMA_RESTRACK_QP];
	759	+ xa_lock(&rt->xa);
	760	+ xa_for_each(&rt->xa, id, res) {
	761	+ qp = container_of(res, struct ib_qp, res);
	762	+ if (!qp->counter \|\| (qp->counter->id != counter->id))
	763	+ continue;
	764	+
	765	+ ret = fill_stat_counter_qp_entry(msg, qp->qp_num);
	766	+ if (ret)
	767	+ goto err;
	768	+ }
	769	+
	770	+ xa_unlock(&rt->xa);
	771	+ nla_nest_end(msg, table_attr);
	772	+ return 0;
	773	+
	774	+err:
	775	+ xa_unlock(&rt->xa);
	776	+ nla_nest_cancel(msg, table_attr);
	777	+ return ret;
	778	+}
	779	+
	780	+int rdma_nl_stat_hwcounter_entry(struct sk_buff msg, const char name,
	781	+ u64 value)
	782	+{
	783	+ struct nlattr *entry_attr;
	784	+
	785	+ entry_attr = nla_nest_start(msg, RDMA_NLDEV_ATTR_STAT_HWCOUNTER_ENTRY);
	786	+ if (!entry_attr)
	787	+ return -EMSGSIZE;
	788	+
	789	+ if (nla_put_string(msg, RDMA_NLDEV_ATTR_STAT_HWCOUNTER_ENTRY_NAME,
	790	+ name))
	791	+ goto err;
	792	+ if (nla_put_u64_64bit(msg, RDMA_NLDEV_ATTR_STAT_HWCOUNTER_ENTRY_VALUE,
	793	+ value, RDMA_NLDEV_ATTR_PAD))
	794	+ goto err;
	795	+
	796	+ nla_nest_end(msg, entry_attr);
	797	+ return 0;
	798	+
	799	+err:
	800	+ nla_nest_cancel(msg, entry_attr);
	801	+ return -EMSGSIZE;
	802	+}
	803	+EXPORT_SYMBOL(rdma_nl_stat_hwcounter_entry);
	804	+
	805	+static int fill_stat_mr_entry(struct sk_buff *msg, bool has_cap_net_admin,
	806	+ struct rdma_restrack_entry *res, uint32_t port)
	807	+{
	808	+ struct ib_mr *mr = container_of(res, struct ib_mr, res);
	809	+ struct ib_device *dev = mr->pd->device;
	810	+
	811	+ if (nla_put_u32(msg, RDMA_NLDEV_ATTR_RES_MRN, res->id))
	812	+ goto err;
	813	+
	814	+ if (dev->ops.fill_stat_mr_entry)
	815	+ return dev->ops.fill_stat_mr_entry(msg, mr);
	816	+ return 0;
	817	+
	818	+err:
	819	+ return -EMSGSIZE;
	820	+}
	821	+
	822	+static int fill_stat_counter_hwcounters(struct sk_buff *msg,
	823	+ struct rdma_counter *counter)
	824	+{
	825	+ struct rdma_hw_stats *st = counter->stats;
	826	+ struct nlattr *table_attr;
	827	+ int i;
	828	+
	829	+ table_attr = nla_nest_start(msg, RDMA_NLDEV_ATTR_STAT_HWCOUNTERS);
	830	+ if (!table_attr)
	831	+ return -EMSGSIZE;
	832	+
	833	+ for (i = 0; i < st->num_counters; i++)
	834	+ if (rdma_nl_stat_hwcounter_entry(msg, st->names[i], st->value[i]))
	835	+ goto err;
	836	+
	837	+ nla_nest_end(msg, table_attr);
	838	+ return 0;
	839	+
	840	+err:
	841	+ nla_nest_cancel(msg, table_attr);
	842	+ return -EMSGSIZE;
	843	+}
	844	+
	845	+static int fill_res_counter_entry(struct sk_buff *msg, bool has_cap_net_admin,
	846	+ struct rdma_restrack_entry *res,
	847	+ uint32_t port)
	848	+{
	849	+ struct rdma_counter *counter =
	850	+ container_of(res, struct rdma_counter, res);
	851	+
	852	+ if (port && port != counter->port)
	853	+ return -EAGAIN;
	854	+
	855	+ /* Dump it even query failed */
	856	+ rdma_counter_query_stats(counter);
	857	+
	858	+ if (nla_put_u32(msg, RDMA_NLDEV_ATTR_PORT_INDEX, counter->port) \|\|
	859	+ nla_put_u32(msg, RDMA_NLDEV_ATTR_STAT_COUNTER_ID, counter->id) \|\|
	860	+ fill_stat_counter_mode(msg, counter) \|\|
	861	+ fill_stat_counter_qps(msg, counter) \|\|
	862	+ fill_stat_counter_hwcounters(msg, counter))
	863	+ return -EMSGSIZE;
	864	+
	865	+ return 0;
596	866	}
597	867
598	868	static int nldev_get_doit(struct sk_buff skb, struct nlmsghdr nlh,
..	..	@@ -604,14 +874,14 @@
604	874	u32 index;
605	875	int err;
606	876
607		- err = nlmsg_parse(nlh, 0, tb, RDMA_NLDEV_ATTR_MAX - 1,
608		- nldev_policy, extack);
	877	+ err = nlmsg_parse_deprecated(nlh, 0, tb, RDMA_NLDEV_ATTR_MAX - 1,
	878	+ nldev_policy, extack);
609	879	if (err \|\| !tb[RDMA_NLDEV_ATTR_DEV_INDEX])
610	880	return -EINVAL;
611	881
612	882	index = nla_get_u32(tb[RDMA_NLDEV_ATTR_DEV_INDEX]);
613	883
614		- device = ib_device_get_by_index(index);
	884	+ device = ib_device_get_by_index(sock_net(skb->sk), index);
615	885	if (!device)
616	886	return -EINVAL;
617	887
..	..	@@ -631,13 +901,66 @@
631	901
632	902	nlmsg_end(msg, nlh);
633	903
634		- put_device(&device->dev);
635		- return rdma_nl_unicast(msg, NETLINK_CB(skb).portid);
	904	+ ib_device_put(device);
	905	+ return rdma_nl_unicast(sock_net(skb->sk), msg, NETLINK_CB(skb).portid);
636	906
637	907	err_free:
638	908	nlmsg_free(msg);
639	909	err:
640		- put_device(&device->dev);
	910	+ ib_device_put(device);
	911	+ return err;
	912	+}
	913	+
	914	+static int nldev_set_doit(struct sk_buff skb, struct nlmsghdr nlh,
	915	+ struct netlink_ext_ack *extack)
	916	+{
	917	+ struct nlattr *tb[RDMA_NLDEV_ATTR_MAX];
	918	+ struct ib_device *device;
	919	+ u32 index;
	920	+ int err;
	921	+
	922	+ err = nlmsg_parse_deprecated(nlh, 0, tb, RDMA_NLDEV_ATTR_MAX - 1,
	923	+ nldev_policy, extack);
	924	+ if (err \|\| !tb[RDMA_NLDEV_ATTR_DEV_INDEX])
	925	+ return -EINVAL;
	926	+
	927	+ index = nla_get_u32(tb[RDMA_NLDEV_ATTR_DEV_INDEX]);
	928	+ device = ib_device_get_by_index(sock_net(skb->sk), index);
	929	+ if (!device)
	930	+ return -EINVAL;
	931	+
	932	+ if (tb[RDMA_NLDEV_ATTR_DEV_NAME]) {
	933	+ char name[IB_DEVICE_NAME_MAX] = {};
	934	+
	935	+ nla_strlcpy(name, tb[RDMA_NLDEV_ATTR_DEV_NAME],
	936	+ IB_DEVICE_NAME_MAX);
	937	+ if (strlen(name) == 0) {
	938	+ err = -EINVAL;
	939	+ goto done;
	940	+ }
	941	+ err = ib_device_rename(device, name);
	942	+ goto done;
	943	+ }
	944	+
	945	+ if (tb[RDMA_NLDEV_NET_NS_FD]) {
	946	+ u32 ns_fd;
	947	+
	948	+ ns_fd = nla_get_u32(tb[RDMA_NLDEV_NET_NS_FD]);
	949	+ err = ib_device_set_netns_put(skb, device, ns_fd);
	950	+ goto put_done;
	951	+ }
	952	+
	953	+ if (tb[RDMA_NLDEV_ATTR_DEV_DIM]) {
	954	+ u8 use_dim;
	955	+
	956	+ use_dim = nla_get_u8(tb[RDMA_NLDEV_ATTR_DEV_DIM]);
	957	+ err = ib_device_set_dim(device, use_dim);
	958	+ goto done;
	959	+ }
	960	+
	961	+done:
	962	+ ib_device_put(device);
	963	+put_done:
641	964	return err;
642	965	}
643	966
..	..	@@ -673,7 +996,7 @@
673	996	{
674	997	/*
675	998	* There is no need to take lock, because
676		- * we are relying on ib_core's lists_rwsem
	999	+ * we are relying on ib_core's locking.
677	1000	*/
678	1001	return ib_enum_all_devs(_nldev_get_dumpit, skb, cb);
679	1002	}
..	..	@@ -688,15 +1011,15 @@
688	1011	u32 port;
689	1012	int err;
690	1013
691		- err = nlmsg_parse(nlh, 0, tb, RDMA_NLDEV_ATTR_MAX - 1,
692		- nldev_policy, extack);
	1014	+ err = nlmsg_parse_deprecated(nlh, 0, tb, RDMA_NLDEV_ATTR_MAX - 1,
	1015	+ nldev_policy, extack);
693	1016	if (err \|\|
694	1017	!tb[RDMA_NLDEV_ATTR_DEV_INDEX] \|\|
695	1018	!tb[RDMA_NLDEV_ATTR_PORT_INDEX])
696	1019	return -EINVAL;
697	1020
698	1021	index = nla_get_u32(tb[RDMA_NLDEV_ATTR_DEV_INDEX]);
699		- device = ib_device_get_by_index(index);
	1022	+ device = ib_device_get_by_index(sock_net(skb->sk), index);
700	1023	if (!device)
701	1024	return -EINVAL;
702	1025
..	..	@@ -721,14 +1044,14 @@
721	1044	goto err_free;
722	1045
723	1046	nlmsg_end(msg, nlh);
724		- put_device(&device->dev);
	1047	+ ib_device_put(device);
725	1048
726		- return rdma_nl_unicast(msg, NETLINK_CB(skb).portid);
	1049	+ return rdma_nl_unicast(sock_net(skb->sk), msg, NETLINK_CB(skb).portid);
727	1050
728	1051	err_free:
729	1052	nlmsg_free(msg);
730	1053	err:
731		- put_device(&device->dev);
	1054	+ ib_device_put(device);
732	1055	return err;
733	1056	}
734	1057
..	..	@@ -742,19 +1065,19 @@
742	1065	u32 idx = 0;
743	1066	u32 ifindex;
744	1067	int err;
745		- u32 p;
	1068	+ unsigned int p;
746	1069
747		- err = nlmsg_parse(cb->nlh, 0, tb, RDMA_NLDEV_ATTR_MAX - 1,
748		- nldev_policy, NULL);
	1070	+ err = nlmsg_parse_deprecated(cb->nlh, 0, tb, RDMA_NLDEV_ATTR_MAX - 1,
	1071	+ nldev_policy, NULL);
749	1072	if (err \|\| !tb[RDMA_NLDEV_ATTR_DEV_INDEX])
750	1073	return -EINVAL;
751	1074
752	1075	ifindex = nla_get_u32(tb[RDMA_NLDEV_ATTR_DEV_INDEX]);
753		- device = ib_device_get_by_index(ifindex);
	1076	+ device = ib_device_get_by_index(sock_net(skb->sk), ifindex);
754	1077	if (!device)
755	1078	return -EINVAL;
756	1079
757		- for (p = rdma_start_port(device); p <= rdma_end_port(device); ++p) {
	1080	+ rdma_for_each_port (device, p) {
758	1081	/*
759	1082	* The dumpit function returns all information from specific
760	1083	* index. This specific index is taken from the netlink
..	..	@@ -785,7 +1108,7 @@
785	1108	}
786	1109
787	1110	out:
788		- put_device(&device->dev);
	1111	+ ib_device_put(device);
789	1112	cb->args[0] = idx;
790	1113	return skb->len;
791	1114	}
..	..	@@ -799,13 +1122,13 @@
799	1122	u32 index;
800	1123	int ret;
801	1124
802		- ret = nlmsg_parse(nlh, 0, tb, RDMA_NLDEV_ATTR_MAX - 1,
803		- nldev_policy, extack);
	1125	+ ret = nlmsg_parse_deprecated(nlh, 0, tb, RDMA_NLDEV_ATTR_MAX - 1,
	1126	+ nldev_policy, extack);
804	1127	if (ret \|\| !tb[RDMA_NLDEV_ATTR_DEV_INDEX])
805	1128	return -EINVAL;
806	1129
807	1130	index = nla_get_u32(tb[RDMA_NLDEV_ATTR_DEV_INDEX]);
808		- device = ib_device_get_by_index(index);
	1131	+ device = ib_device_get_by_index(sock_net(skb->sk), index);
809	1132	if (!device)
810	1133	return -EINVAL;
811	1134
..	..	@@ -824,13 +1147,13 @@
824	1147	goto err_free;
825	1148
826	1149	nlmsg_end(msg, nlh);
827		- put_device(&device->dev);
828		- return rdma_nl_unicast(msg, NETLINK_CB(skb).portid);
	1150	+ ib_device_put(device);
	1151	+ return rdma_nl_unicast(sock_net(skb->sk), msg, NETLINK_CB(skb).portid);
829	1152
830	1153	err_free:
831	1154	nlmsg_free(msg);
832	1155	err:
833		- put_device(&device->dev);
	1156	+ ib_device_put(device);
834	1157	return ret;
835	1158	}
836	1159
..	..	@@ -853,7 +1176,6 @@
853	1176	nlmsg_cancel(skb, nlh);
854	1177	goto out;
855	1178	}
856		-
857	1179	nlmsg_end(skb, nlh);
858	1180
859	1181	idx++;
..	..	@@ -870,57 +1192,155 @@
870	1192	}
871	1193
872	1194	struct nldev_fill_res_entry {
873		- int (fill_res_func)(struct sk_buff msg, struct netlink_callback *cb,
874		- struct rdma_restrack_entry *res, u32 port);
875	1195	enum rdma_nldev_attr nldev_attr;
876		- enum rdma_nldev_command nldev_cmd;
	1196	+ u8 flags;
	1197	+ u32 entry;
	1198	+ u32 id;
	1199	+};
	1200	+
	1201	+enum nldev_res_flags {
	1202	+ NLDEV_PER_DEV = 1 << 0,
877	1203	};
878	1204
879	1205	static const struct nldev_fill_res_entry fill_entries[RDMA_RESTRACK_MAX] = {
880	1206	[RDMA_RESTRACK_QP] = {
881		- .fill_res_func = fill_res_qp_entry,
882		- .nldev_cmd = RDMA_NLDEV_CMD_RES_QP_GET,
883	1207	.nldev_attr = RDMA_NLDEV_ATTR_RES_QP,
	1208	+ .entry = RDMA_NLDEV_ATTR_RES_QP_ENTRY,
	1209	+ .id = RDMA_NLDEV_ATTR_RES_LQPN,
884	1210	},
885	1211	[RDMA_RESTRACK_CM_ID] = {
886		- .fill_res_func = fill_res_cm_id_entry,
887		- .nldev_cmd = RDMA_NLDEV_CMD_RES_CM_ID_GET,
888	1212	.nldev_attr = RDMA_NLDEV_ATTR_RES_CM_ID,
	1213	+ .entry = RDMA_NLDEV_ATTR_RES_CM_ID_ENTRY,
	1214	+ .id = RDMA_NLDEV_ATTR_RES_CM_IDN,
889	1215	},
890	1216	[RDMA_RESTRACK_CQ] = {
891		- .fill_res_func = fill_res_cq_entry,
892		- .nldev_cmd = RDMA_NLDEV_CMD_RES_CQ_GET,
893	1217	.nldev_attr = RDMA_NLDEV_ATTR_RES_CQ,
	1218	+ .flags = NLDEV_PER_DEV,
	1219	+ .entry = RDMA_NLDEV_ATTR_RES_CQ_ENTRY,
	1220	+ .id = RDMA_NLDEV_ATTR_RES_CQN,
894	1221	},
895	1222	[RDMA_RESTRACK_MR] = {
896		- .fill_res_func = fill_res_mr_entry,
897		- .nldev_cmd = RDMA_NLDEV_CMD_RES_MR_GET,
898	1223	.nldev_attr = RDMA_NLDEV_ATTR_RES_MR,
	1224	+ .flags = NLDEV_PER_DEV,
	1225	+ .entry = RDMA_NLDEV_ATTR_RES_MR_ENTRY,
	1226	+ .id = RDMA_NLDEV_ATTR_RES_MRN,
899	1227	},
900	1228	[RDMA_RESTRACK_PD] = {
901		- .fill_res_func = fill_res_pd_entry,
902		- .nldev_cmd = RDMA_NLDEV_CMD_RES_PD_GET,
903	1229	.nldev_attr = RDMA_NLDEV_ATTR_RES_PD,
	1230	+ .flags = NLDEV_PER_DEV,
	1231	+ .entry = RDMA_NLDEV_ATTR_RES_PD_ENTRY,
	1232	+ .id = RDMA_NLDEV_ATTR_RES_PDN,
	1233	+ },
	1234	+ [RDMA_RESTRACK_COUNTER] = {
	1235	+ .nldev_attr = RDMA_NLDEV_ATTR_STAT_COUNTER,
	1236	+ .entry = RDMA_NLDEV_ATTR_STAT_COUNTER_ENTRY,
	1237	+ .id = RDMA_NLDEV_ATTR_STAT_COUNTER_ID,
904	1238	},
905	1239	};
906	1240
907		-static int res_get_common_dumpit(struct sk_buff *skb,
908		- struct netlink_callback *cb,
909		- enum rdma_restrack_type res_type)
	1241	+static int res_get_common_doit(struct sk_buff skb, struct nlmsghdr nlh,
	1242	+ struct netlink_ext_ack *extack,
	1243	+ enum rdma_restrack_type res_type,
	1244	+ res_fill_func_t fill_func)
910	1245	{
911	1246	const struct nldev_fill_res_entry *fe = &fill_entries[res_type];
912	1247	struct nlattr *tb[RDMA_NLDEV_ATTR_MAX];
913	1248	struct rdma_restrack_entry *res;
	1249	+ struct ib_device *device;
	1250	+ u32 index, id, port = 0;
	1251	+ bool has_cap_net_admin;
	1252	+ struct sk_buff *msg;
	1253	+ int ret;
	1254	+
	1255	+ ret = nlmsg_parse_deprecated(nlh, 0, tb, RDMA_NLDEV_ATTR_MAX - 1,
	1256	+ nldev_policy, extack);
	1257	+ if (ret \|\| !tb[RDMA_NLDEV_ATTR_DEV_INDEX] \|\| !fe->id \|\| !tb[fe->id])
	1258	+ return -EINVAL;
	1259	+
	1260	+ index = nla_get_u32(tb[RDMA_NLDEV_ATTR_DEV_INDEX]);
	1261	+ device = ib_device_get_by_index(sock_net(skb->sk), index);
	1262	+ if (!device)
	1263	+ return -EINVAL;
	1264	+
	1265	+ if (tb[RDMA_NLDEV_ATTR_PORT_INDEX]) {
	1266	+ port = nla_get_u32(tb[RDMA_NLDEV_ATTR_PORT_INDEX]);
	1267	+ if (!rdma_is_port_valid(device, port)) {
	1268	+ ret = -EINVAL;
	1269	+ goto err;
	1270	+ }
	1271	+ }
	1272	+
	1273	+ if ((port && fe->flags & NLDEV_PER_DEV) \|\|
	1274	+ (!port && ~fe->flags & NLDEV_PER_DEV)) {
	1275	+ ret = -EINVAL;
	1276	+ goto err;
	1277	+ }
	1278	+
	1279	+ id = nla_get_u32(tb[fe->id]);
	1280	+ res = rdma_restrack_get_byid(device, res_type, id);
	1281	+ if (IS_ERR(res)) {
	1282	+ ret = PTR_ERR(res);
	1283	+ goto err;
	1284	+ }
	1285	+
	1286	+ msg = nlmsg_new(NLMSG_DEFAULT_SIZE, GFP_KERNEL);
	1287	+ if (!msg) {
	1288	+ ret = -ENOMEM;
	1289	+ goto err_get;
	1290	+ }
	1291	+
	1292	+ nlh = nlmsg_put(msg, NETLINK_CB(skb).portid, nlh->nlmsg_seq,
	1293	+ RDMA_NL_GET_TYPE(RDMA_NL_NLDEV,
	1294	+ RDMA_NL_GET_OP(nlh->nlmsg_type)),
	1295	+ 0, 0);
	1296	+
	1297	+ if (fill_nldev_handle(msg, device)) {
	1298	+ ret = -EMSGSIZE;
	1299	+ goto err_free;
	1300	+ }
	1301	+
	1302	+ has_cap_net_admin = netlink_capable(skb, CAP_NET_ADMIN);
	1303	+
	1304	+ ret = fill_func(msg, has_cap_net_admin, res, port);
	1305	+ if (ret)
	1306	+ goto err_free;
	1307	+
	1308	+ rdma_restrack_put(res);
	1309	+ nlmsg_end(msg, nlh);
	1310	+ ib_device_put(device);
	1311	+ return rdma_nl_unicast(sock_net(skb->sk), msg, NETLINK_CB(skb).portid);
	1312	+
	1313	+err_free:
	1314	+ nlmsg_free(msg);
	1315	+err_get:
	1316	+ rdma_restrack_put(res);
	1317	+err:
	1318	+ ib_device_put(device);
	1319	+ return ret;
	1320	+}
	1321	+
	1322	+static int res_get_common_dumpit(struct sk_buff *skb,
	1323	+ struct netlink_callback *cb,
	1324	+ enum rdma_restrack_type res_type,
	1325	+ res_fill_func_t fill_func)
	1326	+{
	1327	+ const struct nldev_fill_res_entry *fe = &fill_entries[res_type];
	1328	+ struct nlattr *tb[RDMA_NLDEV_ATTR_MAX];
	1329	+ struct rdma_restrack_entry *res;
	1330	+ struct rdma_restrack_root *rt;
914	1331	int err, ret = 0, idx = 0;
915	1332	struct nlattr *table_attr;
	1333	+ struct nlattr *entry_attr;
916	1334	struct ib_device *device;
917	1335	int start = cb->args[0];
	1336	+ bool has_cap_net_admin;
918	1337	struct nlmsghdr *nlh;
	1338	+ unsigned long id;
919	1339	u32 index, port = 0;
920	1340	bool filled = false;
921	1341
922		- err = nlmsg_parse(cb->nlh, 0, tb, RDMA_NLDEV_ATTR_MAX - 1,
923		- nldev_policy, NULL);
	1342	+ err = nlmsg_parse_deprecated(cb->nlh, 0, tb, RDMA_NLDEV_ATTR_MAX - 1,
	1343	+ nldev_policy, NULL);
924	1344	/*
925	1345	* Right now, we are expecting the device index to get res information,
926	1346	* but it is possible to extend this code to return all devices in
..	..	@@ -933,7 +1353,7 @@
933	1353	return -EINVAL;
934	1354
935	1355	index = nla_get_u32(tb[RDMA_NLDEV_ATTR_DEV_INDEX]);
936		- device = ib_device_get_by_index(index);
	1356	+ device = ib_device_get_by_index(sock_net(skb->sk), index);
937	1357	if (!device)
938	1358	return -EINVAL;
939	1359
..	..	@@ -949,7 +1369,8 @@
949	1369	}
950	1370
951	1371	nlh = nlmsg_put(skb, NETLINK_CB(cb->skb).portid, cb->nlh->nlmsg_seq,
952		- RDMA_NL_GET_TYPE(RDMA_NL_NLDEV, fe->nldev_cmd),
	1372	+ RDMA_NL_GET_TYPE(RDMA_NL_NLDEV,
	1373	+ RDMA_NL_GET_OP(cb->nlh->nlmsg_type)),
953	1374	0, NLM_F_MULTI);
954	1375
955	1376	if (fill_nldev_handle(skb, device)) {
..	..	@@ -957,61 +1378,55 @@
957	1378	goto err;
958	1379	}
959	1380
960		- table_attr = nla_nest_start(skb, fe->nldev_attr);
	1381	+ table_attr = nla_nest_start_noflag(skb, fe->nldev_attr);
961	1382	if (!table_attr) {
962	1383	ret = -EMSGSIZE;
963	1384	goto err;
964	1385	}
965	1386
966		- down_read(&device->res.rwsem);
967		- hash_for_each_possible(device->res.hash, res, node, res_type) {
968		- if (idx < start)
	1387	+ has_cap_net_admin = netlink_capable(cb->skb, CAP_NET_ADMIN);
	1388	+
	1389	+ rt = &device->res[res_type];
	1390	+ xa_lock(&rt->xa);
	1391	+ /*
	1392	+ * FIXME: if the skip ahead is something common this loop should
	1393	+ * use xas_for_each & xas_pause to optimize, we can have a lot of
	1394	+ * objects.
	1395	+ */
	1396	+ xa_for_each(&rt->xa, id, res) {
	1397	+ if (idx < start \|\| !rdma_restrack_get(res))
969	1398	goto next;
970	1399
971		- if ((rdma_is_kernel_res(res) &&
972		- task_active_pid_ns(current) != &init_pid_ns) \|\|
973		- (!rdma_is_kernel_res(res) && task_active_pid_ns(current) !=
974		- task_active_pid_ns(res->task)))
975		- /*
976		- * 1. Kern resources should be visible in init
977		- * namspace only
978		- * 2. Present only resources visible in the current
979		- * namespace
980		- */
981		- goto next;
982		-
983		- if (!rdma_restrack_get(res))
984		- /*
985		- * Resource is under release now, but we are not
986		- * relesing lock now, so it will be released in
987		- * our next pass, once we will get ->next pointer.
988		- */
989		- goto next;
	1400	+ xa_unlock(&rt->xa);
990	1401
991	1402	filled = true;
992	1403
993		- up_read(&device->res.rwsem);
994		- ret = fe->fill_res_func(skb, cb, res, port);
995		- down_read(&device->res.rwsem);
996		- /*
997		- * Return resource back, but it won't be released till
998		- * the &device->res.rwsem will be released for write.
999		- */
	1404	+ entry_attr = nla_nest_start_noflag(skb, fe->entry);
	1405	+ if (!entry_attr) {
	1406	+ ret = -EMSGSIZE;
	1407	+ rdma_restrack_put(res);
	1408	+ goto msg_full;
	1409	+ }
	1410	+
	1411	+ ret = fill_func(skb, has_cap_net_admin, res, port);
	1412	+
1000	1413	rdma_restrack_put(res);
1001	1414
1002		- if (ret == -EMSGSIZE)
1003		- /*
1004		- * There is a chance to optimize here.
1005		- * It can be done by using list_prepare_entry
1006		- * and list_for_each_entry_continue afterwards.
1007		- */
1008		- break;
1009		- if (ret)
	1415	+ if (ret) {
	1416	+ nla_nest_cancel(skb, entry_attr);
	1417	+ if (ret == -EMSGSIZE)
	1418	+ goto msg_full;
	1419	+ if (ret == -EAGAIN)
	1420	+ goto again;
1010	1421	goto res_err;
	1422	+ }
	1423	+ nla_nest_end(skb, entry_attr);
	1424	+again: xa_lock(&rt->xa);
1011	1425	next: idx++;
1012	1426	}
1013		- up_read(&device->res.rwsem);
	1427	+ xa_unlock(&rt->xa);
1014	1428
	1429	+msg_full:
1015	1430	nla_nest_end(skb, table_attr);
1016	1431	nlmsg_end(skb, nlh);
1017	1432	cb->args[0] = idx;
..	..	@@ -1023,55 +1438,680 @@
1023	1438	if (!filled)
1024	1439	goto err;
1025	1440
1026		- put_device(&device->dev);
	1441	+ ib_device_put(device);
1027	1442	return skb->len;
1028	1443
1029	1444	res_err:
1030	1445	nla_nest_cancel(skb, table_attr);
1031		- up_read(&device->res.rwsem);
1032	1446
1033	1447	err:
1034	1448	nlmsg_cancel(skb, nlh);
1035	1449
1036	1450	err_index:
1037		- put_device(&device->dev);
	1451	+ ib_device_put(device);
1038	1452	return ret;
1039	1453	}
1040	1454
1041		-static int nldev_res_get_qp_dumpit(struct sk_buff *skb,
1042		- struct netlink_callback *cb)
	1455	+#define RES_GET_FUNCS(name, type) \
	1456	+ static int nldev_res_get_##name##_dumpit(struct sk_buff *skb, \
	1457	+ struct netlink_callback *cb) \
	1458	+ { \
	1459	+ return res_get_common_dumpit(skb, cb, type, \
	1460	+ fill_res_##name##_entry); \
	1461	+ } \
	1462	+ static int nldev_res_get_##name##_doit(struct sk_buff *skb, \
	1463	+ struct nlmsghdr *nlh, \
	1464	+ struct netlink_ext_ack *extack) \
	1465	+ { \
	1466	+ return res_get_common_doit(skb, nlh, extack, type, \
	1467	+ fill_res_##name##_entry); \
	1468	+ }
	1469	+
	1470	+RES_GET_FUNCS(qp, RDMA_RESTRACK_QP);
	1471	+RES_GET_FUNCS(qp_raw, RDMA_RESTRACK_QP);
	1472	+RES_GET_FUNCS(cm_id, RDMA_RESTRACK_CM_ID);
	1473	+RES_GET_FUNCS(cq, RDMA_RESTRACK_CQ);
	1474	+RES_GET_FUNCS(cq_raw, RDMA_RESTRACK_CQ);
	1475	+RES_GET_FUNCS(pd, RDMA_RESTRACK_PD);
	1476	+RES_GET_FUNCS(mr, RDMA_RESTRACK_MR);
	1477	+RES_GET_FUNCS(mr_raw, RDMA_RESTRACK_MR);
	1478	+RES_GET_FUNCS(counter, RDMA_RESTRACK_COUNTER);
	1479	+
	1480	+static LIST_HEAD(link_ops);
	1481	+static DECLARE_RWSEM(link_ops_rwsem);
	1482	+
	1483	+static const struct rdma_link_ops link_ops_get(const char type)
1043	1484	{
1044		- return res_get_common_dumpit(skb, cb, RDMA_RESTRACK_QP);
	1485	+ const struct rdma_link_ops *ops;
	1486	+
	1487	+ list_for_each_entry(ops, &link_ops, list) {
	1488	+ if (!strcmp(ops->type, type))
	1489	+ goto out;
	1490	+ }
	1491	+ ops = NULL;
	1492	+out:
	1493	+ return ops;
1045	1494	}
1046	1495
1047		-static int nldev_res_get_cm_id_dumpit(struct sk_buff *skb,
1048		- struct netlink_callback *cb)
	1496	+void rdma_link_register(struct rdma_link_ops *ops)
1049	1497	{
1050		- return res_get_common_dumpit(skb, cb, RDMA_RESTRACK_CM_ID);
	1498	+ down_write(&link_ops_rwsem);
	1499	+ if (WARN_ON_ONCE(link_ops_get(ops->type)))
	1500	+ goto out;
	1501	+ list_add(&ops->list, &link_ops);
	1502	+out:
	1503	+ up_write(&link_ops_rwsem);
	1504	+}
	1505	+EXPORT_SYMBOL(rdma_link_register);
	1506	+
	1507	+void rdma_link_unregister(struct rdma_link_ops *ops)
	1508	+{
	1509	+ down_write(&link_ops_rwsem);
	1510	+ list_del(&ops->list);
	1511	+ up_write(&link_ops_rwsem);
	1512	+}
	1513	+EXPORT_SYMBOL(rdma_link_unregister);
	1514	+
	1515	+static int nldev_newlink(struct sk_buff skb, struct nlmsghdr nlh,
	1516	+ struct netlink_ext_ack *extack)
	1517	+{
	1518	+ struct nlattr *tb[RDMA_NLDEV_ATTR_MAX];
	1519	+ char ibdev_name[IB_DEVICE_NAME_MAX];
	1520	+ const struct rdma_link_ops *ops;
	1521	+ char ndev_name[IFNAMSIZ];
	1522	+ struct net_device *ndev;
	1523	+ char type[IFNAMSIZ];
	1524	+ int err;
	1525	+
	1526	+ err = nlmsg_parse_deprecated(nlh, 0, tb, RDMA_NLDEV_ATTR_MAX - 1,
	1527	+ nldev_policy, extack);
	1528	+ if (err \|\| !tb[RDMA_NLDEV_ATTR_DEV_NAME] \|\|
	1529	+ !tb[RDMA_NLDEV_ATTR_LINK_TYPE] \|\| !tb[RDMA_NLDEV_ATTR_NDEV_NAME])
	1530	+ return -EINVAL;
	1531	+
	1532	+ nla_strlcpy(ibdev_name, tb[RDMA_NLDEV_ATTR_DEV_NAME],
	1533	+ sizeof(ibdev_name));
	1534	+ if (strchr(ibdev_name, '%') \|\| strlen(ibdev_name) == 0)
	1535	+ return -EINVAL;
	1536	+
	1537	+ nla_strlcpy(type, tb[RDMA_NLDEV_ATTR_LINK_TYPE], sizeof(type));
	1538	+ nla_strlcpy(ndev_name, tb[RDMA_NLDEV_ATTR_NDEV_NAME],
	1539	+ sizeof(ndev_name));
	1540	+
	1541	+ ndev = dev_get_by_name(sock_net(skb->sk), ndev_name);
	1542	+ if (!ndev)
	1543	+ return -ENODEV;
	1544	+
	1545	+ down_read(&link_ops_rwsem);
	1546	+ ops = link_ops_get(type);
	1547	+#ifdef CONFIG_MODULES
	1548	+ if (!ops) {
	1549	+ up_read(&link_ops_rwsem);
	1550	+ request_module("rdma-link-%s", type);
	1551	+ down_read(&link_ops_rwsem);
	1552	+ ops = link_ops_get(type);
	1553	+ }
	1554	+#endif
	1555	+ err = ops ? ops->newlink(ibdev_name, ndev) : -EINVAL;
	1556	+ up_read(&link_ops_rwsem);
	1557	+ dev_put(ndev);
	1558	+
	1559	+ return err;
1051	1560	}
1052	1561
1053		-static int nldev_res_get_cq_dumpit(struct sk_buff *skb,
1054		- struct netlink_callback *cb)
	1562	+static int nldev_dellink(struct sk_buff skb, struct nlmsghdr nlh,
	1563	+ struct netlink_ext_ack *extack)
1055	1564	{
1056		- return res_get_common_dumpit(skb, cb, RDMA_RESTRACK_CQ);
	1565	+ struct nlattr *tb[RDMA_NLDEV_ATTR_MAX];
	1566	+ struct ib_device *device;
	1567	+ u32 index;
	1568	+ int err;
	1569	+
	1570	+ err = nlmsg_parse_deprecated(nlh, 0, tb, RDMA_NLDEV_ATTR_MAX - 1,
	1571	+ nldev_policy, extack);
	1572	+ if (err \|\| !tb[RDMA_NLDEV_ATTR_DEV_INDEX])
	1573	+ return -EINVAL;
	1574	+
	1575	+ index = nla_get_u32(tb[RDMA_NLDEV_ATTR_DEV_INDEX]);
	1576	+ device = ib_device_get_by_index(sock_net(skb->sk), index);
	1577	+ if (!device)
	1578	+ return -EINVAL;
	1579	+
	1580	+ if (!(device->attrs.device_cap_flags & IB_DEVICE_ALLOW_USER_UNREG)) {
	1581	+ ib_device_put(device);
	1582	+ return -EINVAL;
	1583	+ }
	1584	+
	1585	+ ib_unregister_device_and_put(device);
	1586	+ return 0;
1057	1587	}
1058	1588
1059		-static int nldev_res_get_mr_dumpit(struct sk_buff *skb,
1060		- struct netlink_callback *cb)
	1589	+static int nldev_get_chardev(struct sk_buff skb, struct nlmsghdr nlh,
	1590	+ struct netlink_ext_ack *extack)
1061	1591	{
1062		- return res_get_common_dumpit(skb, cb, RDMA_RESTRACK_MR);
	1592	+ struct nlattr *tb[RDMA_NLDEV_ATTR_MAX];
	1593	+ char client_name[RDMA_NLDEV_ATTR_CHARDEV_TYPE_SIZE];
	1594	+ struct ib_client_nl_info data = {};
	1595	+ struct ib_device *ibdev = NULL;
	1596	+ struct sk_buff *msg;
	1597	+ u32 index;
	1598	+ int err;
	1599	+
	1600	+ err = nlmsg_parse(nlh, 0, tb, RDMA_NLDEV_ATTR_MAX - 1, nldev_policy,
	1601	+ extack);
	1602	+ if (err \|\| !tb[RDMA_NLDEV_ATTR_CHARDEV_TYPE])
	1603	+ return -EINVAL;
	1604	+
	1605	+ nla_strlcpy(client_name, tb[RDMA_NLDEV_ATTR_CHARDEV_TYPE],
	1606	+ sizeof(client_name));
	1607	+
	1608	+ if (tb[RDMA_NLDEV_ATTR_DEV_INDEX]) {
	1609	+ index = nla_get_u32(tb[RDMA_NLDEV_ATTR_DEV_INDEX]);
	1610	+ ibdev = ib_device_get_by_index(sock_net(skb->sk), index);
	1611	+ if (!ibdev)
	1612	+ return -EINVAL;
	1613	+
	1614	+ if (tb[RDMA_NLDEV_ATTR_PORT_INDEX]) {
	1615	+ data.port = nla_get_u32(tb[RDMA_NLDEV_ATTR_PORT_INDEX]);
	1616	+ if (!rdma_is_port_valid(ibdev, data.port)) {
	1617	+ err = -EINVAL;
	1618	+ goto out_put;
	1619	+ }
	1620	+ } else {
	1621	+ data.port = -1;
	1622	+ }
	1623	+ } else if (tb[RDMA_NLDEV_ATTR_PORT_INDEX]) {
	1624	+ return -EINVAL;
	1625	+ }
	1626	+
	1627	+ msg = nlmsg_new(NLMSG_DEFAULT_SIZE, GFP_KERNEL);
	1628	+ if (!msg) {
	1629	+ err = -ENOMEM;
	1630	+ goto out_put;
	1631	+ }
	1632	+ nlh = nlmsg_put(msg, NETLINK_CB(skb).portid, nlh->nlmsg_seq,
	1633	+ RDMA_NL_GET_TYPE(RDMA_NL_NLDEV,
	1634	+ RDMA_NLDEV_CMD_GET_CHARDEV),
	1635	+ 0, 0);
	1636	+
	1637	+ data.nl_msg = msg;
	1638	+ err = ib_get_client_nl_info(ibdev, client_name, &data);
	1639	+ if (err)
	1640	+ goto out_nlmsg;
	1641	+
	1642	+ err = nla_put_u64_64bit(msg, RDMA_NLDEV_ATTR_CHARDEV,
	1643	+ huge_encode_dev(data.cdev->devt),
	1644	+ RDMA_NLDEV_ATTR_PAD);
	1645	+ if (err)
	1646	+ goto out_data;
	1647	+ err = nla_put_u64_64bit(msg, RDMA_NLDEV_ATTR_CHARDEV_ABI, data.abi,
	1648	+ RDMA_NLDEV_ATTR_PAD);
	1649	+ if (err)
	1650	+ goto out_data;
	1651	+ if (nla_put_string(msg, RDMA_NLDEV_ATTR_CHARDEV_NAME,
	1652	+ dev_name(data.cdev))) {
	1653	+ err = -EMSGSIZE;
	1654	+ goto out_data;
	1655	+ }
	1656	+
	1657	+ nlmsg_end(msg, nlh);
	1658	+ put_device(data.cdev);
	1659	+ if (ibdev)
	1660	+ ib_device_put(ibdev);
	1661	+ return rdma_nl_unicast(sock_net(skb->sk), msg, NETLINK_CB(skb).portid);
	1662	+
	1663	+out_data:
	1664	+ put_device(data.cdev);
	1665	+out_nlmsg:
	1666	+ nlmsg_free(msg);
	1667	+out_put:
	1668	+ if (ibdev)
	1669	+ ib_device_put(ibdev);
	1670	+ return err;
1063	1671	}
1064	1672
1065		-static int nldev_res_get_pd_dumpit(struct sk_buff *skb,
1066		- struct netlink_callback *cb)
	1673	+static int nldev_sys_get_doit(struct sk_buff skb, struct nlmsghdr nlh,
	1674	+ struct netlink_ext_ack *extack)
1067	1675	{
1068		- return res_get_common_dumpit(skb, cb, RDMA_RESTRACK_PD);
	1676	+ struct nlattr *tb[RDMA_NLDEV_ATTR_MAX];
	1677	+ struct sk_buff *msg;
	1678	+ int err;
	1679	+
	1680	+ err = nlmsg_parse(nlh, 0, tb, RDMA_NLDEV_ATTR_MAX - 1,
	1681	+ nldev_policy, extack);
	1682	+ if (err)
	1683	+ return err;
	1684	+
	1685	+ msg = nlmsg_new(NLMSG_DEFAULT_SIZE, GFP_KERNEL);
	1686	+ if (!msg)
	1687	+ return -ENOMEM;
	1688	+
	1689	+ nlh = nlmsg_put(msg, NETLINK_CB(skb).portid, nlh->nlmsg_seq,
	1690	+ RDMA_NL_GET_TYPE(RDMA_NL_NLDEV,
	1691	+ RDMA_NLDEV_CMD_SYS_GET),
	1692	+ 0, 0);
	1693	+
	1694	+ err = nla_put_u8(msg, RDMA_NLDEV_SYS_ATTR_NETNS_MODE,
	1695	+ (u8)ib_devices_shared_netns);
	1696	+ if (err) {
	1697	+ nlmsg_free(msg);
	1698	+ return err;
	1699	+ }
	1700	+ nlmsg_end(msg, nlh);
	1701	+ return rdma_nl_unicast(sock_net(skb->sk), msg, NETLINK_CB(skb).portid);
	1702	+}
	1703	+
	1704	+static int nldev_set_sys_set_doit(struct sk_buff skb, struct nlmsghdr nlh,
	1705	+ struct netlink_ext_ack *extack)
	1706	+{
	1707	+ struct nlattr *tb[RDMA_NLDEV_ATTR_MAX];
	1708	+ u8 enable;
	1709	+ int err;
	1710	+
	1711	+ err = nlmsg_parse(nlh, 0, tb, RDMA_NLDEV_ATTR_MAX - 1,
	1712	+ nldev_policy, extack);
	1713	+ if (err \|\| !tb[RDMA_NLDEV_SYS_ATTR_NETNS_MODE])
	1714	+ return -EINVAL;
	1715	+
	1716	+ enable = nla_get_u8(tb[RDMA_NLDEV_SYS_ATTR_NETNS_MODE]);
	1717	+ /* Only 0 and 1 are supported */
	1718	+ if (enable > 1)
	1719	+ return -EINVAL;
	1720	+
	1721	+ err = rdma_compatdev_set(enable);
	1722	+ return err;
	1723	+}
	1724	+
	1725	+static int nldev_stat_set_doit(struct sk_buff skb, struct nlmsghdr nlh,
	1726	+ struct netlink_ext_ack *extack)
	1727	+{
	1728	+ u32 index, port, mode, mask = 0, qpn, cntn = 0;
	1729	+ struct nlattr *tb[RDMA_NLDEV_ATTR_MAX];
	1730	+ struct ib_device *device;
	1731	+ struct sk_buff *msg;
	1732	+ int ret;
	1733	+
	1734	+ ret = nlmsg_parse(nlh, 0, tb, RDMA_NLDEV_ATTR_MAX - 1,
	1735	+ nldev_policy, extack);
	1736	+ /* Currently only counter for QP is supported */
	1737	+ if (ret \|\| !tb[RDMA_NLDEV_ATTR_STAT_RES] \|\|
	1738	+ !tb[RDMA_NLDEV_ATTR_DEV_INDEX] \|\|
	1739	+ !tb[RDMA_NLDEV_ATTR_PORT_INDEX] \|\| !tb[RDMA_NLDEV_ATTR_STAT_MODE])
	1740	+ return -EINVAL;
	1741	+
	1742	+ if (nla_get_u32(tb[RDMA_NLDEV_ATTR_STAT_RES]) != RDMA_NLDEV_ATTR_RES_QP)
	1743	+ return -EINVAL;
	1744	+
	1745	+ index = nla_get_u32(tb[RDMA_NLDEV_ATTR_DEV_INDEX]);
	1746	+ device = ib_device_get_by_index(sock_net(skb->sk), index);
	1747	+ if (!device)
	1748	+ return -EINVAL;
	1749	+
	1750	+ port = nla_get_u32(tb[RDMA_NLDEV_ATTR_PORT_INDEX]);
	1751	+ if (!rdma_is_port_valid(device, port)) {
	1752	+ ret = -EINVAL;
	1753	+ goto err;
	1754	+ }
	1755	+
	1756	+ msg = nlmsg_new(NLMSG_DEFAULT_SIZE, GFP_KERNEL);
	1757	+ if (!msg) {
	1758	+ ret = -ENOMEM;
	1759	+ goto err;
	1760	+ }
	1761	+ nlh = nlmsg_put(msg, NETLINK_CB(skb).portid, nlh->nlmsg_seq,
	1762	+ RDMA_NL_GET_TYPE(RDMA_NL_NLDEV,
	1763	+ RDMA_NLDEV_CMD_STAT_SET),
	1764	+ 0, 0);
	1765	+
	1766	+ mode = nla_get_u32(tb[RDMA_NLDEV_ATTR_STAT_MODE]);
	1767	+ if (mode == RDMA_COUNTER_MODE_AUTO) {
	1768	+ if (tb[RDMA_NLDEV_ATTR_STAT_AUTO_MODE_MASK])
	1769	+ mask = nla_get_u32(
	1770	+ tb[RDMA_NLDEV_ATTR_STAT_AUTO_MODE_MASK]);
	1771	+
	1772	+ ret = rdma_counter_set_auto_mode(device, port,
	1773	+ mask ? true : false, mask);
	1774	+ if (ret)
	1775	+ goto err_msg;
	1776	+ } else {
	1777	+ if (!tb[RDMA_NLDEV_ATTR_RES_LQPN])
	1778	+ goto err_msg;
	1779	+ qpn = nla_get_u32(tb[RDMA_NLDEV_ATTR_RES_LQPN]);
	1780	+ if (tb[RDMA_NLDEV_ATTR_STAT_COUNTER_ID]) {
	1781	+ cntn = nla_get_u32(tb[RDMA_NLDEV_ATTR_STAT_COUNTER_ID]);
	1782	+ ret = rdma_counter_bind_qpn(device, port, qpn, cntn);
	1783	+ } else {
	1784	+ ret = rdma_counter_bind_qpn_alloc(device, port,
	1785	+ qpn, &cntn);
	1786	+ }
	1787	+ if (ret)
	1788	+ goto err_msg;
	1789	+
	1790	+ if (fill_nldev_handle(msg, device) \|\|
	1791	+ nla_put_u32(msg, RDMA_NLDEV_ATTR_PORT_INDEX, port) \|\|
	1792	+ nla_put_u32(msg, RDMA_NLDEV_ATTR_STAT_COUNTER_ID, cntn) \|\|
	1793	+ nla_put_u32(msg, RDMA_NLDEV_ATTR_RES_LQPN, qpn)) {
	1794	+ ret = -EMSGSIZE;
	1795	+ goto err_fill;
	1796	+ }
	1797	+ }
	1798	+
	1799	+ nlmsg_end(msg, nlh);
	1800	+ ib_device_put(device);
	1801	+ return rdma_nl_unicast(sock_net(skb->sk), msg, NETLINK_CB(skb).portid);
	1802	+
	1803	+err_fill:
	1804	+ rdma_counter_unbind_qpn(device, port, qpn, cntn);
	1805	+err_msg:
	1806	+ nlmsg_free(msg);
	1807	+err:
	1808	+ ib_device_put(device);
	1809	+ return ret;
	1810	+}
	1811	+
	1812	+static int nldev_stat_del_doit(struct sk_buff skb, struct nlmsghdr nlh,
	1813	+ struct netlink_ext_ack *extack)
	1814	+{
	1815	+ struct nlattr *tb[RDMA_NLDEV_ATTR_MAX];
	1816	+ struct ib_device *device;
	1817	+ struct sk_buff *msg;
	1818	+ u32 index, port, qpn, cntn;
	1819	+ int ret;
	1820	+
	1821	+ ret = nlmsg_parse(nlh, 0, tb, RDMA_NLDEV_ATTR_MAX - 1,
	1822	+ nldev_policy, extack);
	1823	+ if (ret \|\| !tb[RDMA_NLDEV_ATTR_STAT_RES] \|\|
	1824	+ !tb[RDMA_NLDEV_ATTR_DEV_INDEX] \|\| !tb[RDMA_NLDEV_ATTR_PORT_INDEX] \|\|
	1825	+ !tb[RDMA_NLDEV_ATTR_STAT_COUNTER_ID] \|\|
	1826	+ !tb[RDMA_NLDEV_ATTR_RES_LQPN])
	1827	+ return -EINVAL;
	1828	+
	1829	+ if (nla_get_u32(tb[RDMA_NLDEV_ATTR_STAT_RES]) != RDMA_NLDEV_ATTR_RES_QP)
	1830	+ return -EINVAL;
	1831	+
	1832	+ index = nla_get_u32(tb[RDMA_NLDEV_ATTR_DEV_INDEX]);
	1833	+ device = ib_device_get_by_index(sock_net(skb->sk), index);
	1834	+ if (!device)
	1835	+ return -EINVAL;
	1836	+
	1837	+ port = nla_get_u32(tb[RDMA_NLDEV_ATTR_PORT_INDEX]);
	1838	+ if (!rdma_is_port_valid(device, port)) {
	1839	+ ret = -EINVAL;
	1840	+ goto err;
	1841	+ }
	1842	+
	1843	+ msg = nlmsg_new(NLMSG_DEFAULT_SIZE, GFP_KERNEL);
	1844	+ if (!msg) {
	1845	+ ret = -ENOMEM;
	1846	+ goto err;
	1847	+ }
	1848	+ nlh = nlmsg_put(msg, NETLINK_CB(skb).portid, nlh->nlmsg_seq,
	1849	+ RDMA_NL_GET_TYPE(RDMA_NL_NLDEV,
	1850	+ RDMA_NLDEV_CMD_STAT_SET),
	1851	+ 0, 0);
	1852	+
	1853	+ cntn = nla_get_u32(tb[RDMA_NLDEV_ATTR_STAT_COUNTER_ID]);
	1854	+ qpn = nla_get_u32(tb[RDMA_NLDEV_ATTR_RES_LQPN]);
	1855	+ if (fill_nldev_handle(msg, device) \|\|
	1856	+ nla_put_u32(msg, RDMA_NLDEV_ATTR_PORT_INDEX, port) \|\|
	1857	+ nla_put_u32(msg, RDMA_NLDEV_ATTR_STAT_COUNTER_ID, cntn) \|\|
	1858	+ nla_put_u32(msg, RDMA_NLDEV_ATTR_RES_LQPN, qpn)) {
	1859	+ ret = -EMSGSIZE;
	1860	+ goto err_fill;
	1861	+ }
	1862	+
	1863	+ ret = rdma_counter_unbind_qpn(device, port, qpn, cntn);
	1864	+ if (ret)
	1865	+ goto err_fill;
	1866	+
	1867	+ nlmsg_end(msg, nlh);
	1868	+ ib_device_put(device);
	1869	+ return rdma_nl_unicast(sock_net(skb->sk), msg, NETLINK_CB(skb).portid);
	1870	+
	1871	+err_fill:
	1872	+ nlmsg_free(msg);
	1873	+err:
	1874	+ ib_device_put(device);
	1875	+ return ret;
	1876	+}
	1877	+
	1878	+static int stat_get_doit_default_counter(struct sk_buff *skb,
	1879	+ struct nlmsghdr *nlh,
	1880	+ struct netlink_ext_ack *extack,
	1881	+ struct nlattr *tb[])
	1882	+{
	1883	+ struct rdma_hw_stats *stats;
	1884	+ struct nlattr *table_attr;
	1885	+ struct ib_device *device;
	1886	+ int ret, num_cnts, i;
	1887	+ struct sk_buff *msg;
	1888	+ u32 index, port;
	1889	+ u64 v;
	1890	+
	1891	+ if (!tb[RDMA_NLDEV_ATTR_DEV_INDEX] \|\| !tb[RDMA_NLDEV_ATTR_PORT_INDEX])
	1892	+ return -EINVAL;
	1893	+
	1894	+ index = nla_get_u32(tb[RDMA_NLDEV_ATTR_DEV_INDEX]);
	1895	+ device = ib_device_get_by_index(sock_net(skb->sk), index);
	1896	+ if (!device)
	1897	+ return -EINVAL;
	1898	+
	1899	+ if (!device->ops.alloc_hw_stats \|\| !device->ops.get_hw_stats) {
	1900	+ ret = -EINVAL;
	1901	+ goto err;
	1902	+ }
	1903	+
	1904	+ port = nla_get_u32(tb[RDMA_NLDEV_ATTR_PORT_INDEX]);
	1905	+ if (!rdma_is_port_valid(device, port)) {
	1906	+ ret = -EINVAL;
	1907	+ goto err;
	1908	+ }
	1909	+
	1910	+ msg = nlmsg_new(NLMSG_DEFAULT_SIZE, GFP_KERNEL);
	1911	+ if (!msg) {
	1912	+ ret = -ENOMEM;
	1913	+ goto err;
	1914	+ }
	1915	+
	1916	+ nlh = nlmsg_put(msg, NETLINK_CB(skb).portid, nlh->nlmsg_seq,
	1917	+ RDMA_NL_GET_TYPE(RDMA_NL_NLDEV,
	1918	+ RDMA_NLDEV_CMD_STAT_GET),
	1919	+ 0, 0);
	1920	+
	1921	+ if (fill_nldev_handle(msg, device) \|\|
	1922	+ nla_put_u32(msg, RDMA_NLDEV_ATTR_PORT_INDEX, port)) {
	1923	+ ret = -EMSGSIZE;
	1924	+ goto err_msg;
	1925	+ }
	1926	+
	1927	+ stats = device->port_data ? device->port_data[port].hw_stats : NULL;
	1928	+ if (stats == NULL) {
	1929	+ ret = -EINVAL;
	1930	+ goto err_msg;
	1931	+ }
	1932	+ mutex_lock(&stats->lock);
	1933	+
	1934	+ num_cnts = device->ops.get_hw_stats(device, stats, port, 0);
	1935	+ if (num_cnts < 0) {
	1936	+ ret = -EINVAL;
	1937	+ goto err_stats;
	1938	+ }
	1939	+
	1940	+ table_attr = nla_nest_start(msg, RDMA_NLDEV_ATTR_STAT_HWCOUNTERS);
	1941	+ if (!table_attr) {
	1942	+ ret = -EMSGSIZE;
	1943	+ goto err_stats;
	1944	+ }
	1945	+ for (i = 0; i < num_cnts; i++) {
	1946	+ v = stats->value[i] +
	1947	+ rdma_counter_get_hwstat_value(device, port, i);
	1948	+ if (rdma_nl_stat_hwcounter_entry(msg, stats->names[i], v)) {
	1949	+ ret = -EMSGSIZE;
	1950	+ goto err_table;
	1951	+ }
	1952	+ }
	1953	+ nla_nest_end(msg, table_attr);
	1954	+
	1955	+ mutex_unlock(&stats->lock);
	1956	+ nlmsg_end(msg, nlh);
	1957	+ ib_device_put(device);
	1958	+ return rdma_nl_unicast(sock_net(skb->sk), msg, NETLINK_CB(skb).portid);
	1959	+
	1960	+err_table:
	1961	+ nla_nest_cancel(msg, table_attr);
	1962	+err_stats:
	1963	+ mutex_unlock(&stats->lock);
	1964	+err_msg:
	1965	+ nlmsg_free(msg);
	1966	+err:
	1967	+ ib_device_put(device);
	1968	+ return ret;
	1969	+}
	1970	+
	1971	+static int stat_get_doit_qp(struct sk_buff skb, struct nlmsghdr nlh,
	1972	+ struct netlink_ext_ack extack, struct nlattr tb[])
	1973	+
	1974	+{
	1975	+ static enum rdma_nl_counter_mode mode;
	1976	+ static enum rdma_nl_counter_mask mask;
	1977	+ struct ib_device *device;
	1978	+ struct sk_buff *msg;
	1979	+ u32 index, port;
	1980	+ int ret;
	1981	+
	1982	+ if (tb[RDMA_NLDEV_ATTR_STAT_COUNTER_ID])
	1983	+ return nldev_res_get_counter_doit(skb, nlh, extack);
	1984	+
	1985	+ if (!tb[RDMA_NLDEV_ATTR_STAT_MODE] \|\|
	1986	+ !tb[RDMA_NLDEV_ATTR_DEV_INDEX] \|\| !tb[RDMA_NLDEV_ATTR_PORT_INDEX])
	1987	+ return -EINVAL;
	1988	+
	1989	+ index = nla_get_u32(tb[RDMA_NLDEV_ATTR_DEV_INDEX]);
	1990	+ device = ib_device_get_by_index(sock_net(skb->sk), index);
	1991	+ if (!device)
	1992	+ return -EINVAL;
	1993	+
	1994	+ port = nla_get_u32(tb[RDMA_NLDEV_ATTR_PORT_INDEX]);
	1995	+ if (!rdma_is_port_valid(device, port)) {
	1996	+ ret = -EINVAL;
	1997	+ goto err;
	1998	+ }
	1999	+
	2000	+ msg = nlmsg_new(NLMSG_DEFAULT_SIZE, GFP_KERNEL);
	2001	+ if (!msg) {
	2002	+ ret = -ENOMEM;
	2003	+ goto err;
	2004	+ }
	2005	+
	2006	+ nlh = nlmsg_put(msg, NETLINK_CB(skb).portid, nlh->nlmsg_seq,
	2007	+ RDMA_NL_GET_TYPE(RDMA_NL_NLDEV,
	2008	+ RDMA_NLDEV_CMD_STAT_GET),
	2009	+ 0, 0);
	2010	+
	2011	+ ret = rdma_counter_get_mode(device, port, &mode, &mask);
	2012	+ if (ret)
	2013	+ goto err_msg;
	2014	+
	2015	+ if (fill_nldev_handle(msg, device) \|\|
	2016	+ nla_put_u32(msg, RDMA_NLDEV_ATTR_PORT_INDEX, port) \|\|
	2017	+ nla_put_u32(msg, RDMA_NLDEV_ATTR_STAT_MODE, mode)) {
	2018	+ ret = -EMSGSIZE;
	2019	+ goto err_msg;
	2020	+ }
	2021	+
	2022	+ if ((mode == RDMA_COUNTER_MODE_AUTO) &&
	2023	+ nla_put_u32(msg, RDMA_NLDEV_ATTR_STAT_AUTO_MODE_MASK, mask)) {
	2024	+ ret = -EMSGSIZE;
	2025	+ goto err_msg;
	2026	+ }
	2027	+
	2028	+ nlmsg_end(msg, nlh);
	2029	+ ib_device_put(device);
	2030	+ return rdma_nl_unicast(sock_net(skb->sk), msg, NETLINK_CB(skb).portid);
	2031	+
	2032	+err_msg:
	2033	+ nlmsg_free(msg);
	2034	+err:
	2035	+ ib_device_put(device);
	2036	+ return ret;
	2037	+}
	2038	+
	2039	+static int nldev_stat_get_doit(struct sk_buff skb, struct nlmsghdr nlh,
	2040	+ struct netlink_ext_ack *extack)
	2041	+{
	2042	+ struct nlattr *tb[RDMA_NLDEV_ATTR_MAX];
	2043	+ int ret;
	2044	+
	2045	+ ret = nlmsg_parse(nlh, 0, tb, RDMA_NLDEV_ATTR_MAX - 1,
	2046	+ nldev_policy, extack);
	2047	+ if (ret)
	2048	+ return -EINVAL;
	2049	+
	2050	+ if (!tb[RDMA_NLDEV_ATTR_STAT_RES])
	2051	+ return stat_get_doit_default_counter(skb, nlh, extack, tb);
	2052	+
	2053	+ switch (nla_get_u32(tb[RDMA_NLDEV_ATTR_STAT_RES])) {
	2054	+ case RDMA_NLDEV_ATTR_RES_QP:
	2055	+ ret = stat_get_doit_qp(skb, nlh, extack, tb);
	2056	+ break;
	2057	+ case RDMA_NLDEV_ATTR_RES_MR:
	2058	+ ret = res_get_common_doit(skb, nlh, extack, RDMA_RESTRACK_MR,
	2059	+ fill_stat_mr_entry);
	2060	+ break;
	2061	+ default:
	2062	+ ret = -EINVAL;
	2063	+ break;
	2064	+ }
	2065	+
	2066	+ return ret;
	2067	+}
	2068	+
	2069	+static int nldev_stat_get_dumpit(struct sk_buff *skb,
	2070	+ struct netlink_callback *cb)
	2071	+{
	2072	+ struct nlattr *tb[RDMA_NLDEV_ATTR_MAX];
	2073	+ int ret;
	2074	+
	2075	+ ret = nlmsg_parse(cb->nlh, 0, tb, RDMA_NLDEV_ATTR_MAX - 1,
	2076	+ nldev_policy, NULL);
	2077	+ if (ret \|\| !tb[RDMA_NLDEV_ATTR_STAT_RES])
	2078	+ return -EINVAL;
	2079	+
	2080	+ switch (nla_get_u32(tb[RDMA_NLDEV_ATTR_STAT_RES])) {
	2081	+ case RDMA_NLDEV_ATTR_RES_QP:
	2082	+ ret = nldev_res_get_counter_dumpit(skb, cb);
	2083	+ break;
	2084	+ case RDMA_NLDEV_ATTR_RES_MR:
	2085	+ ret = res_get_common_dumpit(skb, cb, RDMA_RESTRACK_MR,
	2086	+ fill_stat_mr_entry);
	2087	+ break;
	2088	+ default:
	2089	+ ret = -EINVAL;
	2090	+ break;
	2091	+ }
	2092	+
	2093	+ return ret;
1069	2094	}
1070	2095
1071	2096	static const struct rdma_nl_cbs nldev_cb_table[RDMA_NLDEV_NUM_OPS] = {
1072	2097	[RDMA_NLDEV_CMD_GET] = {
1073	2098	.doit = nldev_get_doit,
1074	2099	.dump = nldev_get_dumpit,
	2100	+ },
	2101	+ [RDMA_NLDEV_CMD_GET_CHARDEV] = {
	2102	+ .doit = nldev_get_chardev,
	2103	+ },
	2104	+ [RDMA_NLDEV_CMD_SET] = {
	2105	+ .doit = nldev_set_doit,
	2106	+ .flags = RDMA_NL_ADMIN_PERM,
	2107	+ },
	2108	+ [RDMA_NLDEV_CMD_NEWLINK] = {
	2109	+ .doit = nldev_newlink,
	2110	+ .flags = RDMA_NL_ADMIN_PERM,
	2111	+ },
	2112	+ [RDMA_NLDEV_CMD_DELLINK] = {
	2113	+ .doit = nldev_dellink,
	2114	+ .flags = RDMA_NL_ADMIN_PERM,
1075	2115	},
1076	2116	[RDMA_NLDEV_CMD_PORT_GET] = {
1077	2117	.doit = nldev_port_get_doit,
..	..	@@ -1082,29 +2122,57 @@
1082	2122	.dump = nldev_res_get_dumpit,
1083	2123	},
1084	2124	[RDMA_NLDEV_CMD_RES_QP_GET] = {
	2125	+ .doit = nldev_res_get_qp_doit,
1085	2126	.dump = nldev_res_get_qp_dumpit,
1086		- /*
1087		- * .doit is not implemented yet for two reasons:
1088		- * 1. It is not needed yet.
1089		- * 2. There is a need to provide identifier, while it is easy
1090		- * for the QPs (device index + port index + LQPN), it is not
1091		- * the case for the rest of resources (PD and CQ). Because it
1092		- * is better to provide similar interface for all resources,
1093		- * let's wait till we will have other resources implemented
1094		- * too.
1095		- */
1096	2127	},
1097	2128	[RDMA_NLDEV_CMD_RES_CM_ID_GET] = {
	2129	+ .doit = nldev_res_get_cm_id_doit,
1098	2130	.dump = nldev_res_get_cm_id_dumpit,
1099	2131	},
1100	2132	[RDMA_NLDEV_CMD_RES_CQ_GET] = {
	2133	+ .doit = nldev_res_get_cq_doit,
1101	2134	.dump = nldev_res_get_cq_dumpit,
1102	2135	},
1103	2136	[RDMA_NLDEV_CMD_RES_MR_GET] = {
	2137	+ .doit = nldev_res_get_mr_doit,
1104	2138	.dump = nldev_res_get_mr_dumpit,
1105	2139	},
1106	2140	[RDMA_NLDEV_CMD_RES_PD_GET] = {
	2141	+ .doit = nldev_res_get_pd_doit,
1107	2142	.dump = nldev_res_get_pd_dumpit,
	2143	+ },
	2144	+ [RDMA_NLDEV_CMD_SYS_GET] = {
	2145	+ .doit = nldev_sys_get_doit,
	2146	+ },
	2147	+ [RDMA_NLDEV_CMD_SYS_SET] = {
	2148	+ .doit = nldev_set_sys_set_doit,
	2149	+ },
	2150	+ [RDMA_NLDEV_CMD_STAT_SET] = {
	2151	+ .doit = nldev_stat_set_doit,
	2152	+ .flags = RDMA_NL_ADMIN_PERM,
	2153	+ },
	2154	+ [RDMA_NLDEV_CMD_STAT_GET] = {
	2155	+ .doit = nldev_stat_get_doit,
	2156	+ .dump = nldev_stat_get_dumpit,
	2157	+ },
	2158	+ [RDMA_NLDEV_CMD_STAT_DEL] = {
	2159	+ .doit = nldev_stat_del_doit,
	2160	+ .flags = RDMA_NL_ADMIN_PERM,
	2161	+ },
	2162	+ [RDMA_NLDEV_CMD_RES_QP_GET_RAW] = {
	2163	+ .doit = nldev_res_get_qp_raw_doit,
	2164	+ .dump = nldev_res_get_qp_raw_dumpit,
	2165	+ .flags = RDMA_NL_ADMIN_PERM,
	2166	+ },
	2167	+ [RDMA_NLDEV_CMD_RES_CQ_GET_RAW] = {
	2168	+ .doit = nldev_res_get_cq_raw_doit,
	2169	+ .dump = nldev_res_get_cq_raw_dumpit,
	2170	+ .flags = RDMA_NL_ADMIN_PERM,
	2171	+ },
	2172	+ [RDMA_NLDEV_CMD_RES_MR_GET_RAW] = {
	2173	+ .doit = nldev_res_get_mr_raw_doit,
	2174	+ .dump = nldev_res_get_mr_raw_dumpit,
	2175	+ .flags = RDMA_NL_ADMIN_PERM,
1108	2176	},
1109	2177	};
1110	2178
..	..	@@ -1113,7 +2181,7 @@
1113	2181	rdma_nl_register(RDMA_NL_NLDEV, nldev_cb_table);
1114	2182	}
1115	2183
1116		-void __exit nldev_exit(void)
	2184	+void nldev_exit(void)
1117	2185	{
1118	2186	rdma_nl_unregister(RDMA_NL_NLDEV);
1119	2187	}