rtrs-pri.h

  1/* SPDX-License-Identifier: GPL-2.0-or-later */
  2/*
  3 * RDMA Transport Layer
  4 *
  5 * Copyright (c) 2014 - 2018 ProfitBricks GmbH. All rights reserved.
  6 * Copyright (c) 2018 - 2019 1&1 IONOS Cloud GmbH. All rights reserved.
  7 * Copyright (c) 2019 - 2020 1&1 IONOS SE. All rights reserved.
  8 */
  9
 10#ifndef RTRS_PRI_H
 11#define RTRS_PRI_H
 12
 13#include <linux/uuid.h>
 14#include <rdma/rdma_cm.h>
 15#include <rdma/ib_verbs.h>
 16#include <rdma/ib.h>
 17
 18#include "rtrs.h"
 19
 20#define RTRS_PROTO_VER_MAJOR 2
 21#define RTRS_PROTO_VER_MINOR 0
 22
 23#define RTRS_PROTO_VER_STRING __stringify(RTRS_PROTO_VER_MAJOR) "." \
 24			       __stringify(RTRS_PROTO_VER_MINOR)
 25
 26/*
 27 * Max IB immediate data size is 2^28 (MAX_IMM_PAYL_BITS)
 28 * and the minimum chunk size is 4096 (2^12).
 29 * So the maximum sess_queue_depth is 65535 (2^16 - 1) in theory
 30 * since queue_depth in rtrs_msg_conn_rsp is defined as le16.
 31 * Therefore the pratical max value of sess_queue_depth is
 32 * somewhere between 1 and 65535 and it depends on the system.
 33 */
 34#define MAX_SESS_QUEUE_DEPTH 65535
 35
 36enum rtrs_imm_const {
 37	MAX_IMM_TYPE_BITS = 4,
 38	MAX_IMM_TYPE_MASK = ((1 << MAX_IMM_TYPE_BITS) - 1),
 39	MAX_IMM_PAYL_BITS = 28,
 40	MAX_IMM_PAYL_MASK = ((1 << MAX_IMM_PAYL_BITS) - 1),
 41};
 42
 43enum rtrs_imm_type {
 44	RTRS_IO_REQ_IMM       = 0, /* client to server */
 45	RTRS_IO_RSP_IMM       = 1, /* server to client */
 46	RTRS_IO_RSP_W_INV_IMM = 2, /* server to client */
 47
 48	RTRS_HB_MSG_IMM = 8, /* HB: HeartBeat */
 49	RTRS_HB_ACK_IMM = 9,
 50
 51	RTRS_LAST_IMM,
 52};
 53
 54enum {
 55	SERVICE_CON_QUEUE_DEPTH = 512,
 56
 57	MAX_PATHS_NUM = 128,
 58
 59	MIN_CHUNK_SIZE = 8192,
 60
 61	RTRS_HB_INTERVAL_MS = 5000,
 62	RTRS_HB_MISSED_MAX = 5,
 63
 64	RTRS_MAGIC = 0x1BBD,
 65	RTRS_PROTO_VER = (RTRS_PROTO_VER_MAJOR << 8) | RTRS_PROTO_VER_MINOR,
 66};
 67
 68struct rtrs_ib_dev;
 69
 70struct rtrs_rdma_dev_pd_ops {
 71	int (*init)(struct rtrs_ib_dev *dev);
 72	void (*deinit)(struct rtrs_ib_dev *dev);
 73};
 74
 75struct rtrs_rdma_dev_pd {
 76	struct mutex		mutex;
 77	struct list_head	list;
 78	enum ib_pd_flags	pd_flags;
 79	const struct rtrs_rdma_dev_pd_ops *ops;
 80};
 81
 82struct rtrs_ib_dev {
 83	struct ib_device	 *ib_dev;
 84	struct ib_pd		 *ib_pd;
 85	struct kref		 ref;
 86	struct list_head	 entry;
 87	struct rtrs_rdma_dev_pd *pool;
 88	struct ib_event_handler	 event_handler;
 89};
 90
 91struct rtrs_con {
 92	struct rtrs_path	*path;
 93	struct ib_qp		*qp;
 94	struct ib_cq		*cq;
 95	struct rdma_cm_id	*cm_id;
 96	unsigned int		cid;
 97	int                     nr_cqe;
 98	atomic_t		wr_cnt;
 99	atomic_t		sq_wr_avail;
100};
101
102struct rtrs_path {
103	struct list_head	entry;
104	struct sockaddr_storage dst_addr;
105	struct sockaddr_storage src_addr;
106	char			sessname[NAME_MAX];
107	uuid_t			uuid;
108	struct rtrs_con	**con;
109	unsigned int		con_num;
110	unsigned int		irq_con_num;
111	unsigned int		recon_cnt;
112	unsigned int		signal_interval;
113	struct rtrs_ib_dev	*dev;
114	int			dev_ref;
115	struct ib_cqe		*hb_cqe;
116	void			(*hb_err_handler)(struct rtrs_con *con);
117	struct workqueue_struct *hb_wq;
118	struct delayed_work	hb_dwork;
119	unsigned int		hb_interval_ms;
120	unsigned int		hb_missed_cnt;
121	unsigned int		hb_missed_max;
122	ktime_t			hb_last_sent;
123	ktime_t			hb_cur_latency;
124};
125
126/* rtrs information unit */
127struct rtrs_iu {
128	struct ib_cqe           cqe;
129	dma_addr_t              dma_addr;
130	void                    *buf;
131	size_t                  size;
132	enum dma_data_direction direction;
133};
134
135/**
136 * enum rtrs_msg_types - RTRS message types, see also rtrs/README
137 * @RTRS_MSG_INFO_REQ:		Client additional info request to the server
138 * @RTRS_MSG_INFO_RSP:		Server additional info response to the client
139 * @RTRS_MSG_WRITE:		Client writes data per RDMA to server
140 * @RTRS_MSG_READ:		Client requests data transfer from server
141 * @RTRS_MSG_RKEY_RSP:		Server refreshed rkey for rbuf
142 */
143enum rtrs_msg_types {
144	RTRS_MSG_INFO_REQ,
145	RTRS_MSG_INFO_RSP,
146	RTRS_MSG_WRITE,
147	RTRS_MSG_READ,
148	RTRS_MSG_RKEY_RSP,
149};
150
151/**
152 * enum rtrs_msg_flags - RTRS message flags.
153 * @RTRS_MSG_NEED_INVAL_F: Send invalidation in response.
154 * @RTRS_MSG_NEW_RKEY_F: Send refreshed rkey in response.
155 */
156enum rtrs_msg_flags {
157	RTRS_MSG_NEED_INVAL_F = 1 << 0,
158	RTRS_MSG_NEW_RKEY_F = 1 << 1,
159};
160
161/**
162 * struct rtrs_sg_desc - RDMA-Buffer entry description
163 * @addr:	Address of RDMA destination buffer
164 * @key:	Authorization rkey to write to the buffer
165 * @len:	Size of the buffer
166 */
167struct rtrs_sg_desc {
168	__le64			addr;
169	__le32			key;
170	__le32			len;
171};
172
173/**
174 * struct rtrs_msg_conn_req - Client connection request to the server
175 * @magic:	   RTRS magic
176 * @version:	   RTRS protocol version
177 * @cid:	   Current connection id
178 * @cid_num:	   Number of connections per session
179 * @recon_cnt:	   Reconnections counter
180 * @sess_uuid:	   UUID of a session (path)
181 * @paths_uuid:	   UUID of a group of sessions (paths)
182 * @first_conn:    %1 if the connection request is the first for that session,
183 *			otherwise %0
184 * NOTE: max size 56 bytes, see man rdma_connect().
185 */
186struct rtrs_msg_conn_req {
187	/**
188	 * @__cma_version: Is set to 0 by cma.c in case of AF_IB, do not touch
189	 * that. See https://www.spinics.net/lists/linux-rdma/msg22397.html
190	 */
191	u8		__cma_version;
192	/**
193	 * @__ip_version: On sender side that should be set to 0, or
194	 * cma_save_ip_info() extract garbage and will fail.
195	 */
196	u8		__ip_version;
197	__le16		magic;
198	__le16		version;
199	__le16		cid;
200	__le16		cid_num;
201	__le16		recon_cnt;
202	uuid_t		sess_uuid;
203	uuid_t		paths_uuid;
204	u8		first_conn : 1;
205	/* private: */
206	u8		reserved_bits : 7;
207	u8		reserved[11];
208};
209
210/**
211 * struct rtrs_msg_conn_rsp - Server connection response to the client
212 * @magic:	   RTRS magic
213 * @version:	   RTRS protocol version
214 * @errno:	   If rdma_accept() then 0, if rdma_reject() indicates error
215 * @queue_depth:   max inflight messages (queue-depth) in this session
216 * @max_io_size:   max io size server supports
217 * @max_hdr_size:  max msg header size server supports
218 * @flags:	   RTRS message flags for this message
219 *
220 * NOTE: size is 56 bytes, max possible is 136 bytes, see man rdma_accept().
221 */
222struct rtrs_msg_conn_rsp {
223	__le16		magic;
224	__le16		version;
225	__le16		errno;
226	__le16		queue_depth;
227	__le32		max_io_size;
228	__le32		max_hdr_size;
229	__le32		flags;
230	/* private: */
231	u8		reserved[36];
232};
233
234/**
235 * struct rtrs_msg_info_req - client additional info request
236 * @type:		@RTRS_MSG_INFO_REQ
237 * @pathname:		Path name chosen by client
238 */
239struct rtrs_msg_info_req {
240	__le16		type;
241	u8		pathname[NAME_MAX];
242	/* private: */
243	u8		reserved[15];
244};
245
246/**
247 * struct rtrs_msg_info_rsp - server additional info response
248 * @type:		@RTRS_MSG_INFO_RSP
249 * @sg_cnt:		Number of @desc entries
250 * @desc:		RDMA buffers where the client can write to server
251 */
252struct rtrs_msg_info_rsp {
253	__le16		type;
254	__le16          sg_cnt;
255	/* private: */
256	u8              reserved[4];
257	/* public: */
258	struct rtrs_sg_desc desc[];
259};
260
261/**
262 * struct rtrs_msg_rkey_rsp - server refreshed rkey response
263 * @type:		@RTRS_MSG_RKEY_RSP
264 * @buf_id:		RDMA buf_id of the new rkey
265 * @rkey:		new remote key for RDMA buffers id from server
266 */
267struct rtrs_msg_rkey_rsp {
268	__le16		type;
269	__le16          buf_id;
270	__le32		rkey;
271};
272
273/**
274 * struct rtrs_msg_rdma_read - RDMA data transfer request from client
275 * @type:		always @RTRS_MSG_READ
276 * @flags:		RTRS message flags (enum rtrs_msg_flags)
277 * @usr_len:		length of user payload
278 * @sg_cnt:		number of @desc entries
279 * @desc:		RDMA buffers where the server can write the result to
280 */
281struct rtrs_msg_rdma_read {
282	__le16			type;
283	__le16			usr_len;
284	__le16			flags;
285	__le16			sg_cnt;
286	struct rtrs_sg_desc    desc[];
287};
288
289/**
290 * struct rtrs_msg_rdma_write - Message transferred to server with RDMA-Write
291 * @type:		always @RTRS_MSG_WRITE
292 * @usr_len:		length of user payload
293 */
294struct rtrs_msg_rdma_write {
295	__le16			type;
296	__le16			usr_len;
297};
298
299/**
300 * struct rtrs_msg_rdma_hdr - header for read or write request
301 * @type:		@RTRS_MSG_WRITE | @RTRS_MSG_READ
302 */
303struct rtrs_msg_rdma_hdr {
304	__le16			type;
305};
306
307/* rtrs.c */
308
309struct rtrs_iu *rtrs_iu_alloc(u32 queue_num, size_t size, gfp_t t,
310			      struct ib_device *dev, enum dma_data_direction,
311			      void (*done)(struct ib_cq *cq, struct ib_wc *wc));
312void rtrs_iu_free(struct rtrs_iu *iu, struct ib_device *dev, u32 queue_num);
313int rtrs_iu_post_recv(struct rtrs_con *con, struct rtrs_iu *iu);
314int rtrs_iu_post_send(struct rtrs_con *con, struct rtrs_iu *iu, size_t size,
315		      struct ib_send_wr *head);
316int rtrs_iu_post_rdma_write_imm(struct rtrs_con *con, struct rtrs_iu *iu,
317				struct ib_sge *sge, unsigned int num_sge,
318				u32 rkey, u64 rdma_addr, u32 imm_data,
319				enum ib_send_flags flags,
320				struct ib_send_wr *head,
321				struct ib_send_wr *tail);
322
323int rtrs_post_recv_empty(struct rtrs_con *con, struct ib_cqe *cqe);
324
325int rtrs_cq_qp_create(struct rtrs_path *path, struct rtrs_con *con,
326		      u32 max_send_sge, int cq_vector, int nr_cqe,
327		      u32 max_send_wr, u32 max_recv_wr,
328		      enum ib_poll_context poll_ctx);
329void rtrs_cq_qp_destroy(struct rtrs_con *con);
330
331void rtrs_init_hb(struct rtrs_path *path, struct ib_cqe *cqe,
332		  unsigned int interval_ms, unsigned int missed_max,
333		  void (*err_handler)(struct rtrs_con *con),
334		  struct workqueue_struct *wq);
335void rtrs_start_hb(struct rtrs_path *path);
336void rtrs_stop_hb(struct rtrs_path *path);
337void rtrs_send_hb_ack(struct rtrs_path *path);
338
339void rtrs_rdma_dev_pd_init(enum ib_pd_flags pd_flags,
340			   struct rtrs_rdma_dev_pd *pool);
341void rtrs_rdma_dev_pd_deinit(struct rtrs_rdma_dev_pd *pool);
342
343struct rtrs_ib_dev *rtrs_ib_dev_find_or_add(struct ib_device *ib_dev,
344					    struct rtrs_rdma_dev_pd *pool);
345int rtrs_ib_dev_put(struct rtrs_ib_dev *dev);
346
347static inline u32 rtrs_to_imm(u32 type, u32 payload)
348{
349	BUILD_BUG_ON(MAX_IMM_PAYL_BITS + MAX_IMM_TYPE_BITS != 32);
350	BUILD_BUG_ON(RTRS_LAST_IMM > (1<<MAX_IMM_TYPE_BITS));
351	return ((type & MAX_IMM_TYPE_MASK) << MAX_IMM_PAYL_BITS) |
352		(payload & MAX_IMM_PAYL_MASK);
353}
354
355static inline void rtrs_from_imm(u32 imm, u32 *type, u32 *payload)
356{
357	*payload = imm & MAX_IMM_PAYL_MASK;
358	*type = imm >> MAX_IMM_PAYL_BITS;
359}
360
361static inline u32 rtrs_to_io_req_imm(u32 addr)
362{
363	return rtrs_to_imm(RTRS_IO_REQ_IMM, addr);
364}
365
366static inline u32 rtrs_to_io_rsp_imm(u32 msg_id, int errno, bool w_inval)
367{
368	enum rtrs_imm_type type;
369	u32 payload;
370
371	/* 9 bits for errno, 19 bits for msg_id */
372	payload = (abs(errno) & 0x1ff) << 19 | (msg_id & 0x7ffff);
373	type = w_inval ? RTRS_IO_RSP_W_INV_IMM : RTRS_IO_RSP_IMM;
374
375	return rtrs_to_imm(type, payload);
376}
377
378static inline void rtrs_from_io_rsp_imm(u32 payload, u32 *msg_id, int *errno)
379{
380	/* 9 bits for errno, 19 bits for msg_id */
381	*msg_id = payload & 0x7ffff;
382	*errno = -(int)((payload >> 19) & 0x1ff);
383}
384
385#define STAT_STORE_FUNC(type, set_value, reset)				\
386static ssize_t set_value##_store(struct kobject *kobj,			\
387			     struct kobj_attribute *attr,		\
388			     const char *buf, size_t count)		\
389{									\
390	int ret = -EINVAL;						\
391	type *stats = container_of(kobj, type, kobj_stats);		\
392									\
393	if (sysfs_streq(buf, "1"))					\
394		ret = reset(stats, true);			\
395	else if (sysfs_streq(buf, "0"))					\
396		ret = reset(stats, false);			\
397	if (ret)							\
398		return ret;						\
399									\
400	return count;							\
401}
402
403#define STAT_SHOW_FUNC(type, get_value, print)				\
404static ssize_t get_value##_show(struct kobject *kobj,			\
405			   struct kobj_attribute *attr,			\
406			   char *page)					\
407{									\
408	type *stats = container_of(kobj, type, kobj_stats);		\
409									\
410	return print(stats, page);			\
411}
412
413#define STAT_ATTR(type, stat, print, reset)				\
414STAT_STORE_FUNC(type, stat, reset)					\
415STAT_SHOW_FUNC(type, stat, print)					\
416static struct kobj_attribute stat##_attr = __ATTR_RW(stat)
417
418#endif /* RTRS_PRI_H */