1 /* SPDX-License-Identifier: GPL-2.0-or-later */
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.
13 #include <linux/uuid.h>
14 #include <rdma/rdma_cm.h>
15 #include <rdma/ib_verbs.h>
20 #define RTRS_PROTO_VER_MAJOR 2
21 #define RTRS_PROTO_VER_MINOR 0
23 #define RTRS_PROTO_VER_STRING __stringify(RTRS_PROTO_VER_MAJOR) "." \
24 __stringify(RTRS_PROTO_VER_MINOR)
27 MAX_IMM_TYPE_BITS = 4,
28 MAX_IMM_TYPE_MASK = ((1 << MAX_IMM_TYPE_BITS) - 1),
29 MAX_IMM_PAYL_BITS = 28,
30 MAX_IMM_PAYL_MASK = ((1 << MAX_IMM_PAYL_BITS) - 1),
34 RTRS_IO_REQ_IMM = 0, /* client to server */
35 RTRS_IO_RSP_IMM = 1, /* server to client */
36 RTRS_IO_RSP_W_INV_IMM = 2, /* server to client */
38 RTRS_HB_MSG_IMM = 8, /* HB: HeartBeat */
45 SERVICE_CON_QUEUE_DEPTH = 512,
50 * With the size of struct rtrs_permit allocated on the client, 4K
51 * is the maximum number of rtrs_permits we can allocate. This number is
52 * also used on the client to allocate the IU for the user connection
53 * to receive the RDMA addresses from the server.
55 MAX_SESS_QUEUE_DEPTH = 4096,
57 RTRS_HB_INTERVAL_MS = 5000,
58 RTRS_HB_MISSED_MAX = 5,
61 RTRS_PROTO_VER = (RTRS_PROTO_VER_MAJOR << 8) | RTRS_PROTO_VER_MINOR,
66 struct rtrs_rdma_dev_pd_ops {
67 struct rtrs_ib_dev *(*alloc)(void);
68 void (*free)(struct rtrs_ib_dev *dev);
69 int (*init)(struct rtrs_ib_dev *dev);
70 void (*deinit)(struct rtrs_ib_dev *dev);
73 struct rtrs_rdma_dev_pd {
75 struct list_head list;
76 enum ib_pd_flags pd_flags;
77 const struct rtrs_rdma_dev_pd_ops *ops;
81 struct ib_device *ib_dev;
84 struct list_head entry;
85 struct rtrs_rdma_dev_pd *pool;
89 struct rtrs_sess *sess;
92 struct rdma_cm_id *cm_id;
97 struct list_head entry;
98 struct sockaddr_storage dst_addr;
99 struct sockaddr_storage src_addr;
100 char sessname[NAME_MAX];
102 struct rtrs_con **con;
103 unsigned int con_num;
104 unsigned int recon_cnt;
105 struct rtrs_ib_dev *dev;
107 struct ib_cqe *hb_cqe;
108 void (*hb_err_handler)(struct rtrs_con *con);
109 struct workqueue_struct *hb_wq;
110 struct delayed_work hb_dwork;
111 unsigned int hb_interval_ms;
112 unsigned int hb_missed_cnt;
113 unsigned int hb_missed_max;
116 /* rtrs information unit */
122 enum dma_data_direction direction;
126 * enum rtrs_msg_types - RTRS message types, see also rtrs/README
127 * @RTRS_MSG_INFO_REQ: Client additional info request to the server
128 * @RTRS_MSG_INFO_RSP: Server additional info response to the client
129 * @RTRS_MSG_WRITE: Client writes data per RDMA to server
130 * @RTRS_MSG_READ: Client requests data transfer from server
131 * @RTRS_MSG_RKEY_RSP: Server refreshed rkey for rbuf
133 enum rtrs_msg_types {
142 * enum rtrs_msg_flags - RTRS message flags.
143 * @RTRS_NEED_INVAL: Send invalidation in response.
144 * @RTRS_MSG_NEW_RKEY_F: Send refreshed rkey in response.
146 enum rtrs_msg_flags {
147 RTRS_MSG_NEED_INVAL_F = 1 << 0,
148 RTRS_MSG_NEW_RKEY_F = 1 << 1,
152 * struct rtrs_sg_desc - RDMA-Buffer entry description
153 * @addr: Address of RDMA destination buffer
154 * @key: Authorization rkey to write to the buffer
155 * @len: Size of the buffer
157 struct rtrs_sg_desc {
164 * struct rtrs_msg_conn_req - Client connection request to the server
166 * @version: RTRS protocol version
167 * @cid: Current connection id
168 * @cid_num: Number of connections per session
169 * @recon_cnt: Reconnections counter
170 * @sess_uuid: UUID of a session (path)
171 * @paths_uuid: UUID of a group of sessions (paths)
173 * NOTE: max size 56 bytes, see man rdma_connect().
175 struct rtrs_msg_conn_req {
176 /* Is set to 0 by cma.c in case of AF_IB, do not touch that.
177 * see https://www.spinics.net/lists/linux-rdma/msg22397.html
180 /* On sender side that should be set to 0, or cma_save_ip_info()
181 * extract garbage and will fail.
192 u8 reserved_bits : 7;
197 * struct rtrs_msg_conn_rsp - Server connection response to the client
199 * @version: RTRS protocol version
200 * @errno: If rdma_accept() then 0, if rdma_reject() indicates error
201 * @queue_depth: max inflight messages (queue-depth) in this session
202 * @max_io_size: max io size server supports
203 * @max_hdr_size: max msg header size server supports
205 * NOTE: size is 56 bytes, max possible is 136 bytes, see man rdma_accept().
207 struct rtrs_msg_conn_rsp {
219 * struct rtrs_msg_info_req
220 * @type: @RTRS_MSG_INFO_REQ
221 * @sessname: Session name chosen by client
223 struct rtrs_msg_info_req {
225 u8 sessname[NAME_MAX];
230 * struct rtrs_msg_info_rsp
231 * @type: @RTRS_MSG_INFO_RSP
232 * @sg_cnt: Number of @desc entries
233 * @desc: RDMA buffers where the client can write to server
235 struct rtrs_msg_info_rsp {
239 struct rtrs_sg_desc desc[];
243 * struct rtrs_msg_rkey_rsp
244 * @type: @RTRS_MSG_RKEY_RSP
245 * @buf_id: RDMA buf_id of the new rkey
246 * @rkey: new remote key for RDMA buffers id from server
248 struct rtrs_msg_rkey_rsp {
255 * struct rtrs_msg_rdma_read - RDMA data transfer request from client
256 * @type: always @RTRS_MSG_READ
257 * @usr_len: length of user payload
258 * @sg_cnt: number of @desc entries
259 * @desc: RDMA buffers where the server can write the result to
261 struct rtrs_msg_rdma_read {
266 struct rtrs_sg_desc desc[];
270 * struct_msg_rdma_write - Message transferred to server with RDMA-Write
271 * @type: always @RTRS_MSG_WRITE
272 * @usr_len: length of user payload
274 struct rtrs_msg_rdma_write {
280 * struct_msg_rdma_hdr - header for read or write request
281 * @type: @RTRS_MSG_WRITE | @RTRS_MSG_READ
283 struct rtrs_msg_rdma_hdr {
289 struct rtrs_iu *rtrs_iu_alloc(u32 queue_size, size_t size, gfp_t t,
290 struct ib_device *dev, enum dma_data_direction,
291 void (*done)(struct ib_cq *cq, struct ib_wc *wc));
292 void rtrs_iu_free(struct rtrs_iu *iu, struct ib_device *dev, u32 queue_size);
293 int rtrs_iu_post_recv(struct rtrs_con *con, struct rtrs_iu *iu);
294 int rtrs_iu_post_send(struct rtrs_con *con, struct rtrs_iu *iu, size_t size,
295 struct ib_send_wr *head);
296 int rtrs_iu_post_rdma_write_imm(struct rtrs_con *con, struct rtrs_iu *iu,
297 struct ib_sge *sge, unsigned int num_sge,
298 u32 rkey, u64 rdma_addr, u32 imm_data,
299 enum ib_send_flags flags,
300 struct ib_send_wr *head);
302 int rtrs_post_recv_empty(struct rtrs_con *con, struct ib_cqe *cqe);
303 int rtrs_post_rdma_write_imm_empty(struct rtrs_con *con, struct ib_cqe *cqe,
304 u32 imm_data, enum ib_send_flags flags,
305 struct ib_send_wr *head);
307 int rtrs_cq_qp_create(struct rtrs_sess *rtrs_sess, struct rtrs_con *con,
308 u32 max_send_sge, int cq_vector, int cq_size,
309 u32 max_send_wr, u32 max_recv_wr,
310 enum ib_poll_context poll_ctx);
311 void rtrs_cq_qp_destroy(struct rtrs_con *con);
313 void rtrs_init_hb(struct rtrs_sess *sess, struct ib_cqe *cqe,
314 unsigned int interval_ms, unsigned int missed_max,
315 void (*err_handler)(struct rtrs_con *con),
316 struct workqueue_struct *wq);
317 void rtrs_start_hb(struct rtrs_sess *sess);
318 void rtrs_stop_hb(struct rtrs_sess *sess);
319 void rtrs_send_hb_ack(struct rtrs_sess *sess);
321 void rtrs_rdma_dev_pd_init(enum ib_pd_flags pd_flags,
322 struct rtrs_rdma_dev_pd *pool);
323 void rtrs_rdma_dev_pd_deinit(struct rtrs_rdma_dev_pd *pool);
325 struct rtrs_ib_dev *rtrs_ib_dev_find_or_add(struct ib_device *ib_dev,
326 struct rtrs_rdma_dev_pd *pool);
327 int rtrs_ib_dev_put(struct rtrs_ib_dev *dev);
329 static inline u32 rtrs_to_imm(u32 type, u32 payload)
331 BUILD_BUG_ON(MAX_IMM_PAYL_BITS + MAX_IMM_TYPE_BITS != 32);
332 BUILD_BUG_ON(RTRS_LAST_IMM > (1<<MAX_IMM_TYPE_BITS));
333 return ((type & MAX_IMM_TYPE_MASK) << MAX_IMM_PAYL_BITS) |
334 (payload & MAX_IMM_PAYL_MASK);
337 static inline void rtrs_from_imm(u32 imm, u32 *type, u32 *payload)
339 *payload = imm & MAX_IMM_PAYL_MASK;
340 *type = imm >> MAX_IMM_PAYL_BITS;
343 static inline u32 rtrs_to_io_req_imm(u32 addr)
345 return rtrs_to_imm(RTRS_IO_REQ_IMM, addr);
348 static inline u32 rtrs_to_io_rsp_imm(u32 msg_id, int errno, bool w_inval)
350 enum rtrs_imm_type type;
353 /* 9 bits for errno, 19 bits for msg_id */
354 payload = (abs(errno) & 0x1ff) << 19 | (msg_id & 0x7ffff);
355 type = w_inval ? RTRS_IO_RSP_W_INV_IMM : RTRS_IO_RSP_IMM;
357 return rtrs_to_imm(type, payload);
360 static inline void rtrs_from_io_rsp_imm(u32 payload, u32 *msg_id, int *errno)
362 /* 9 bits for errno, 19 bits for msg_id */
363 *msg_id = payload & 0x7ffff;
364 *errno = -(int)((payload >> 19) & 0x1ff);
367 #define STAT_STORE_FUNC(type, set_value, reset) \
368 static ssize_t set_value##_store(struct kobject *kobj, \
369 struct kobj_attribute *attr, \
370 const char *buf, size_t count) \
373 type *stats = container_of(kobj, type, kobj_stats); \
375 if (sysfs_streq(buf, "1")) \
376 ret = reset(stats, true); \
377 else if (sysfs_streq(buf, "0")) \
378 ret = reset(stats, false); \
385 #define STAT_SHOW_FUNC(type, get_value, print) \
386 static ssize_t get_value##_show(struct kobject *kobj, \
387 struct kobj_attribute *attr, \
390 type *stats = container_of(kobj, type, kobj_stats); \
392 return print(stats, page, PAGE_SIZE); \
395 #define STAT_ATTR(type, stat, print, reset) \
396 STAT_STORE_FUNC(type, stat, reset) \
397 STAT_SHOW_FUNC(type, stat, print) \
398 static struct kobj_attribute stat##_attr = __ATTR_RW(stat)
400 #endif /* RTRS_PRI_H */