xref: /linux/drivers/infiniband/ulp/rtrs/rtrs-clt.c (revision 24168c5e6dfbdd5b414f048f47f75d64533296ca)
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 #undef pr_fmt
11 #define pr_fmt(fmt) KBUILD_MODNAME " L" __stringify(__LINE__) ": " fmt
12 
13 #include <linux/module.h>
14 #include <linux/rculist.h>
15 #include <linux/random.h>
16 
17 #include "rtrs-clt.h"
18 #include "rtrs-log.h"
19 #include "rtrs-clt-trace.h"
20 
21 #define RTRS_CONNECT_TIMEOUT_MS 30000
22 /*
23  * Wait a bit before trying to reconnect after a failure
24  * in order to give server time to finish clean up which
25  * leads to "false positives" failed reconnect attempts
26  */
27 #define RTRS_RECONNECT_BACKOFF 1000
28 /*
29  * Wait for additional random time between 0 and 8 seconds
30  * before starting to reconnect to avoid clients reconnecting
31  * all at once in case of a major network outage
32  */
33 #define RTRS_RECONNECT_SEED 8
34 
35 #define FIRST_CONN 0x01
36 /* limit to 128 * 4k = 512k max IO */
37 #define RTRS_MAX_SEGMENTS          128
38 
39 MODULE_DESCRIPTION("RDMA Transport Client");
40 MODULE_LICENSE("GPL");
41 
42 static const struct rtrs_rdma_dev_pd_ops dev_pd_ops;
43 static struct rtrs_rdma_dev_pd dev_pd = {
44 	.ops = &dev_pd_ops
45 };
46 
47 static struct workqueue_struct *rtrs_wq;
48 static const struct class rtrs_clt_dev_class = {
49 	.name = "rtrs-client",
50 };
51 
52 static inline bool rtrs_clt_is_connected(const struct rtrs_clt_sess *clt)
53 {
54 	struct rtrs_clt_path *clt_path;
55 	bool connected = false;
56 
57 	rcu_read_lock();
58 	list_for_each_entry_rcu(clt_path, &clt->paths_list, s.entry)
59 		if (READ_ONCE(clt_path->state) == RTRS_CLT_CONNECTED) {
60 			connected = true;
61 			break;
62 		}
63 	rcu_read_unlock();
64 
65 	return connected;
66 }
67 
68 static struct rtrs_permit *
69 __rtrs_get_permit(struct rtrs_clt_sess *clt, enum rtrs_clt_con_type con_type)
70 {
71 	size_t max_depth = clt->queue_depth;
72 	struct rtrs_permit *permit;
73 	int bit;
74 
75 	/*
76 	 * Adapted from null_blk get_tag(). Callers from different cpus may
77 	 * grab the same bit, since find_first_zero_bit is not atomic.
78 	 * But then the test_and_set_bit_lock will fail for all the
79 	 * callers but one, so that they will loop again.
80 	 * This way an explicit spinlock is not required.
81 	 */
82 	do {
83 		bit = find_first_zero_bit(clt->permits_map, max_depth);
84 		if (bit >= max_depth)
85 			return NULL;
86 	} while (test_and_set_bit_lock(bit, clt->permits_map));
87 
88 	permit = get_permit(clt, bit);
89 	WARN_ON(permit->mem_id != bit);
90 	permit->cpu_id = raw_smp_processor_id();
91 	permit->con_type = con_type;
92 
93 	return permit;
94 }
95 
96 static inline void __rtrs_put_permit(struct rtrs_clt_sess *clt,
97 				      struct rtrs_permit *permit)
98 {
99 	clear_bit_unlock(permit->mem_id, clt->permits_map);
100 }
101 
102 /**
103  * rtrs_clt_get_permit() - allocates permit for future RDMA operation
104  * @clt:	Current session
105  * @con_type:	Type of connection to use with the permit
106  * @can_wait:	Wait type
107  *
108  * Description:
109  *    Allocates permit for the following RDMA operation.  Permit is used
110  *    to preallocate all resources and to propagate memory pressure
111  *    up earlier.
112  *
113  * Context:
114  *    Can sleep if @wait == RTRS_PERMIT_WAIT
115  */
116 struct rtrs_permit *rtrs_clt_get_permit(struct rtrs_clt_sess *clt,
117 					  enum rtrs_clt_con_type con_type,
118 					  enum wait_type can_wait)
119 {
120 	struct rtrs_permit *permit;
121 	DEFINE_WAIT(wait);
122 
123 	permit = __rtrs_get_permit(clt, con_type);
124 	if (permit || !can_wait)
125 		return permit;
126 
127 	do {
128 		prepare_to_wait(&clt->permits_wait, &wait,
129 				TASK_UNINTERRUPTIBLE);
130 		permit = __rtrs_get_permit(clt, con_type);
131 		if (permit)
132 			break;
133 
134 		io_schedule();
135 	} while (1);
136 
137 	finish_wait(&clt->permits_wait, &wait);
138 
139 	return permit;
140 }
141 EXPORT_SYMBOL(rtrs_clt_get_permit);
142 
143 /**
144  * rtrs_clt_put_permit() - puts allocated permit
145  * @clt:	Current session
146  * @permit:	Permit to be freed
147  *
148  * Context:
149  *    Does not matter
150  */
151 void rtrs_clt_put_permit(struct rtrs_clt_sess *clt,
152 			 struct rtrs_permit *permit)
153 {
154 	if (WARN_ON(!test_bit(permit->mem_id, clt->permits_map)))
155 		return;
156 
157 	__rtrs_put_permit(clt, permit);
158 
159 	/*
160 	 * rtrs_clt_get_permit() adds itself to the &clt->permits_wait list
161 	 * before calling schedule(). So if rtrs_clt_get_permit() is sleeping
162 	 * it must have added itself to &clt->permits_wait before
163 	 * __rtrs_put_permit() finished.
164 	 * Hence it is safe to guard wake_up() with a waitqueue_active() test.
165 	 */
166 	if (waitqueue_active(&clt->permits_wait))
167 		wake_up(&clt->permits_wait);
168 }
169 EXPORT_SYMBOL(rtrs_clt_put_permit);
170 
171 /**
172  * rtrs_permit_to_clt_con() - returns RDMA connection pointer by the permit
173  * @clt_path: client path pointer
174  * @permit: permit for the allocation of the RDMA buffer
175  * Note:
176  *     IO connection starts from 1.
177  *     0 connection is for user messages.
178  */
179 static
180 struct rtrs_clt_con *rtrs_permit_to_clt_con(struct rtrs_clt_path *clt_path,
181 					    struct rtrs_permit *permit)
182 {
183 	int id = 0;
184 
185 	if (permit->con_type == RTRS_IO_CON)
186 		id = (permit->cpu_id % (clt_path->s.irq_con_num - 1)) + 1;
187 
188 	return to_clt_con(clt_path->s.con[id]);
189 }
190 
191 /**
192  * rtrs_clt_change_state() - change the session state through session state
193  * machine.
194  *
195  * @clt_path: client path to change the state of.
196  * @new_state: state to change to.
197  *
198  * returns true if sess's state is changed to new state, otherwise return false.
199  *
200  * Locks:
201  * state_wq lock must be hold.
202  */
203 static bool rtrs_clt_change_state(struct rtrs_clt_path *clt_path,
204 				     enum rtrs_clt_state new_state)
205 {
206 	enum rtrs_clt_state old_state;
207 	bool changed = false;
208 
209 	lockdep_assert_held(&clt_path->state_wq.lock);
210 
211 	old_state = clt_path->state;
212 	switch (new_state) {
213 	case RTRS_CLT_CONNECTING:
214 		switch (old_state) {
215 		case RTRS_CLT_RECONNECTING:
216 			changed = true;
217 			fallthrough;
218 		default:
219 			break;
220 		}
221 		break;
222 	case RTRS_CLT_RECONNECTING:
223 		switch (old_state) {
224 		case RTRS_CLT_CONNECTED:
225 		case RTRS_CLT_CONNECTING_ERR:
226 		case RTRS_CLT_CLOSED:
227 			changed = true;
228 			fallthrough;
229 		default:
230 			break;
231 		}
232 		break;
233 	case RTRS_CLT_CONNECTED:
234 		switch (old_state) {
235 		case RTRS_CLT_CONNECTING:
236 			changed = true;
237 			fallthrough;
238 		default:
239 			break;
240 		}
241 		break;
242 	case RTRS_CLT_CONNECTING_ERR:
243 		switch (old_state) {
244 		case RTRS_CLT_CONNECTING:
245 			changed = true;
246 			fallthrough;
247 		default:
248 			break;
249 		}
250 		break;
251 	case RTRS_CLT_CLOSING:
252 		switch (old_state) {
253 		case RTRS_CLT_CONNECTING:
254 		case RTRS_CLT_CONNECTING_ERR:
255 		case RTRS_CLT_RECONNECTING:
256 		case RTRS_CLT_CONNECTED:
257 			changed = true;
258 			fallthrough;
259 		default:
260 			break;
261 		}
262 		break;
263 	case RTRS_CLT_CLOSED:
264 		switch (old_state) {
265 		case RTRS_CLT_CLOSING:
266 			changed = true;
267 			fallthrough;
268 		default:
269 			break;
270 		}
271 		break;
272 	case RTRS_CLT_DEAD:
273 		switch (old_state) {
274 		case RTRS_CLT_CLOSED:
275 			changed = true;
276 			fallthrough;
277 		default:
278 			break;
279 		}
280 		break;
281 	default:
282 		break;
283 	}
284 	if (changed) {
285 		clt_path->state = new_state;
286 		wake_up_locked(&clt_path->state_wq);
287 	}
288 
289 	return changed;
290 }
291 
292 static bool rtrs_clt_change_state_from_to(struct rtrs_clt_path *clt_path,
293 					   enum rtrs_clt_state old_state,
294 					   enum rtrs_clt_state new_state)
295 {
296 	bool changed = false;
297 
298 	spin_lock_irq(&clt_path->state_wq.lock);
299 	if (clt_path->state == old_state)
300 		changed = rtrs_clt_change_state(clt_path, new_state);
301 	spin_unlock_irq(&clt_path->state_wq.lock);
302 
303 	return changed;
304 }
305 
306 static void rtrs_clt_stop_and_destroy_conns(struct rtrs_clt_path *clt_path);
307 static void rtrs_rdma_error_recovery(struct rtrs_clt_con *con)
308 {
309 	struct rtrs_clt_path *clt_path = to_clt_path(con->c.path);
310 
311 	trace_rtrs_rdma_error_recovery(clt_path);
312 
313 	if (rtrs_clt_change_state_from_to(clt_path,
314 					   RTRS_CLT_CONNECTED,
315 					   RTRS_CLT_RECONNECTING)) {
316 		queue_work(rtrs_wq, &clt_path->err_recovery_work);
317 	} else {
318 		/*
319 		 * Error can happen just on establishing new connection,
320 		 * so notify waiter with error state, waiter is responsible
321 		 * for cleaning the rest and reconnect if needed.
322 		 */
323 		rtrs_clt_change_state_from_to(clt_path,
324 					       RTRS_CLT_CONNECTING,
325 					       RTRS_CLT_CONNECTING_ERR);
326 	}
327 }
328 
329 static void rtrs_clt_fast_reg_done(struct ib_cq *cq, struct ib_wc *wc)
330 {
331 	struct rtrs_clt_con *con = to_clt_con(wc->qp->qp_context);
332 
333 	if (wc->status != IB_WC_SUCCESS) {
334 		rtrs_err(con->c.path, "Failed IB_WR_REG_MR: %s\n",
335 			  ib_wc_status_msg(wc->status));
336 		rtrs_rdma_error_recovery(con);
337 	}
338 }
339 
340 static struct ib_cqe fast_reg_cqe = {
341 	.done = rtrs_clt_fast_reg_done
342 };
343 
344 static void complete_rdma_req(struct rtrs_clt_io_req *req, int errno,
345 			      bool notify, bool can_wait);
346 
347 static void rtrs_clt_inv_rkey_done(struct ib_cq *cq, struct ib_wc *wc)
348 {
349 	struct rtrs_clt_io_req *req =
350 		container_of(wc->wr_cqe, typeof(*req), inv_cqe);
351 	struct rtrs_clt_con *con = to_clt_con(wc->qp->qp_context);
352 
353 	if (wc->status != IB_WC_SUCCESS) {
354 		rtrs_err(con->c.path, "Failed IB_WR_LOCAL_INV: %s\n",
355 			  ib_wc_status_msg(wc->status));
356 		rtrs_rdma_error_recovery(con);
357 	}
358 	req->need_inv = false;
359 	if (req->need_inv_comp)
360 		complete(&req->inv_comp);
361 	else
362 		/* Complete request from INV callback */
363 		complete_rdma_req(req, req->inv_errno, true, false);
364 }
365 
366 static int rtrs_inv_rkey(struct rtrs_clt_io_req *req)
367 {
368 	struct rtrs_clt_con *con = req->con;
369 	struct ib_send_wr wr = {
370 		.opcode		    = IB_WR_LOCAL_INV,
371 		.wr_cqe		    = &req->inv_cqe,
372 		.send_flags	    = IB_SEND_SIGNALED,
373 		.ex.invalidate_rkey = req->mr->rkey,
374 	};
375 	req->inv_cqe.done = rtrs_clt_inv_rkey_done;
376 
377 	return ib_post_send(con->c.qp, &wr, NULL);
378 }
379 
380 static void complete_rdma_req(struct rtrs_clt_io_req *req, int errno,
381 			      bool notify, bool can_wait)
382 {
383 	struct rtrs_clt_con *con = req->con;
384 	struct rtrs_clt_path *clt_path;
385 	int err;
386 
387 	if (!req->in_use)
388 		return;
389 	if (WARN_ON(!req->con))
390 		return;
391 	clt_path = to_clt_path(con->c.path);
392 
393 	if (req->sg_cnt) {
394 		if (req->dir == DMA_FROM_DEVICE && req->need_inv) {
395 			/*
396 			 * We are here to invalidate read requests
397 			 * ourselves.  In normal scenario server should
398 			 * send INV for all read requests, but
399 			 * we are here, thus two things could happen:
400 			 *
401 			 *    1.  this is failover, when errno != 0
402 			 *        and can_wait == 1,
403 			 *
404 			 *    2.  something totally bad happened and
405 			 *        server forgot to send INV, so we
406 			 *        should do that ourselves.
407 			 */
408 
409 			if (can_wait) {
410 				req->need_inv_comp = true;
411 			} else {
412 				/* This should be IO path, so always notify */
413 				WARN_ON(!notify);
414 				/* Save errno for INV callback */
415 				req->inv_errno = errno;
416 			}
417 
418 			refcount_inc(&req->ref);
419 			err = rtrs_inv_rkey(req);
420 			if (err) {
421 				rtrs_err(con->c.path, "Send INV WR key=%#x: %d\n",
422 					  req->mr->rkey, err);
423 			} else if (can_wait) {
424 				wait_for_completion(&req->inv_comp);
425 			} else {
426 				/*
427 				 * Something went wrong, so request will be
428 				 * completed from INV callback.
429 				 */
430 				WARN_ON_ONCE(1);
431 
432 				return;
433 			}
434 			if (!refcount_dec_and_test(&req->ref))
435 				return;
436 		}
437 		ib_dma_unmap_sg(clt_path->s.dev->ib_dev, req->sglist,
438 				req->sg_cnt, req->dir);
439 	}
440 	if (!refcount_dec_and_test(&req->ref))
441 		return;
442 	if (req->mp_policy == MP_POLICY_MIN_INFLIGHT)
443 		atomic_dec(&clt_path->stats->inflight);
444 
445 	req->in_use = false;
446 	req->con = NULL;
447 
448 	if (errno) {
449 		rtrs_err_rl(con->c.path, "IO request failed: error=%d path=%s [%s:%u] notify=%d\n",
450 			    errno, kobject_name(&clt_path->kobj), clt_path->hca_name,
451 			    clt_path->hca_port, notify);
452 	}
453 
454 	if (notify)
455 		req->conf(req->priv, errno);
456 }
457 
458 static int rtrs_post_send_rdma(struct rtrs_clt_con *con,
459 				struct rtrs_clt_io_req *req,
460 				struct rtrs_rbuf *rbuf, u32 off,
461 				u32 imm, struct ib_send_wr *wr)
462 {
463 	struct rtrs_clt_path *clt_path = to_clt_path(con->c.path);
464 	enum ib_send_flags flags;
465 	struct ib_sge sge;
466 
467 	if (!req->sg_size) {
468 		rtrs_wrn(con->c.path,
469 			 "Doing RDMA Write failed, no data supplied\n");
470 		return -EINVAL;
471 	}
472 
473 	/* user data and user message in the first list element */
474 	sge.addr   = req->iu->dma_addr;
475 	sge.length = req->sg_size;
476 	sge.lkey   = clt_path->s.dev->ib_pd->local_dma_lkey;
477 
478 	/*
479 	 * From time to time we have to post signalled sends,
480 	 * or send queue will fill up and only QP reset can help.
481 	 */
482 	flags = atomic_inc_return(&con->c.wr_cnt) % clt_path->s.signal_interval ?
483 			0 : IB_SEND_SIGNALED;
484 
485 	ib_dma_sync_single_for_device(clt_path->s.dev->ib_dev,
486 				      req->iu->dma_addr,
487 				      req->sg_size, DMA_TO_DEVICE);
488 
489 	return rtrs_iu_post_rdma_write_imm(&con->c, req->iu, &sge, 1,
490 					    rbuf->rkey, rbuf->addr + off,
491 					    imm, flags, wr, NULL);
492 }
493 
494 static void process_io_rsp(struct rtrs_clt_path *clt_path, u32 msg_id,
495 			   s16 errno, bool w_inval)
496 {
497 	struct rtrs_clt_io_req *req;
498 
499 	if (WARN_ON(msg_id >= clt_path->queue_depth))
500 		return;
501 
502 	req = &clt_path->reqs[msg_id];
503 	/* Drop need_inv if server responded with send with invalidation */
504 	req->need_inv &= !w_inval;
505 	complete_rdma_req(req, errno, true, false);
506 }
507 
508 static void rtrs_clt_recv_done(struct rtrs_clt_con *con, struct ib_wc *wc)
509 {
510 	struct rtrs_iu *iu;
511 	int err;
512 	struct rtrs_clt_path *clt_path = to_clt_path(con->c.path);
513 
514 	WARN_ON((clt_path->flags & RTRS_MSG_NEW_RKEY_F) == 0);
515 	iu = container_of(wc->wr_cqe, struct rtrs_iu,
516 			  cqe);
517 	err = rtrs_iu_post_recv(&con->c, iu);
518 	if (err) {
519 		rtrs_err(con->c.path, "post iu failed %d\n", err);
520 		rtrs_rdma_error_recovery(con);
521 	}
522 }
523 
524 static void rtrs_clt_rkey_rsp_done(struct rtrs_clt_con *con, struct ib_wc *wc)
525 {
526 	struct rtrs_clt_path *clt_path = to_clt_path(con->c.path);
527 	struct rtrs_msg_rkey_rsp *msg;
528 	u32 imm_type, imm_payload;
529 	bool w_inval = false;
530 	struct rtrs_iu *iu;
531 	u32 buf_id;
532 	int err;
533 
534 	WARN_ON((clt_path->flags & RTRS_MSG_NEW_RKEY_F) == 0);
535 
536 	iu = container_of(wc->wr_cqe, struct rtrs_iu, cqe);
537 
538 	if (wc->byte_len < sizeof(*msg)) {
539 		rtrs_err(con->c.path, "rkey response is malformed: size %d\n",
540 			  wc->byte_len);
541 		goto out;
542 	}
543 	ib_dma_sync_single_for_cpu(clt_path->s.dev->ib_dev, iu->dma_addr,
544 				   iu->size, DMA_FROM_DEVICE);
545 	msg = iu->buf;
546 	if (le16_to_cpu(msg->type) != RTRS_MSG_RKEY_RSP) {
547 		rtrs_err(clt_path->clt,
548 			  "rkey response is malformed: type %d\n",
549 			  le16_to_cpu(msg->type));
550 		goto out;
551 	}
552 	buf_id = le16_to_cpu(msg->buf_id);
553 	if (WARN_ON(buf_id >= clt_path->queue_depth))
554 		goto out;
555 
556 	rtrs_from_imm(be32_to_cpu(wc->ex.imm_data), &imm_type, &imm_payload);
557 	if (imm_type == RTRS_IO_RSP_IMM ||
558 	    imm_type == RTRS_IO_RSP_W_INV_IMM) {
559 		u32 msg_id;
560 
561 		w_inval = (imm_type == RTRS_IO_RSP_W_INV_IMM);
562 		rtrs_from_io_rsp_imm(imm_payload, &msg_id, &err);
563 
564 		if (WARN_ON(buf_id != msg_id))
565 			goto out;
566 		clt_path->rbufs[buf_id].rkey = le32_to_cpu(msg->rkey);
567 		process_io_rsp(clt_path, msg_id, err, w_inval);
568 	}
569 	ib_dma_sync_single_for_device(clt_path->s.dev->ib_dev, iu->dma_addr,
570 				      iu->size, DMA_FROM_DEVICE);
571 	return rtrs_clt_recv_done(con, wc);
572 out:
573 	rtrs_rdma_error_recovery(con);
574 }
575 
576 static void rtrs_clt_rdma_done(struct ib_cq *cq, struct ib_wc *wc);
577 
578 static struct ib_cqe io_comp_cqe = {
579 	.done = rtrs_clt_rdma_done
580 };
581 
582 /*
583  * Post x2 empty WRs: first is for this RDMA with IMM,
584  * second is for RECV with INV, which happened earlier.
585  */
586 static int rtrs_post_recv_empty_x2(struct rtrs_con *con, struct ib_cqe *cqe)
587 {
588 	struct ib_recv_wr wr_arr[2], *wr;
589 	int i;
590 
591 	memset(wr_arr, 0, sizeof(wr_arr));
592 	for (i = 0; i < ARRAY_SIZE(wr_arr); i++) {
593 		wr = &wr_arr[i];
594 		wr->wr_cqe  = cqe;
595 		if (i)
596 			/* Chain backwards */
597 			wr->next = &wr_arr[i - 1];
598 	}
599 
600 	return ib_post_recv(con->qp, wr, NULL);
601 }
602 
603 static void rtrs_clt_rdma_done(struct ib_cq *cq, struct ib_wc *wc)
604 {
605 	struct rtrs_clt_con *con = to_clt_con(wc->qp->qp_context);
606 	struct rtrs_clt_path *clt_path = to_clt_path(con->c.path);
607 	u32 imm_type, imm_payload;
608 	bool w_inval = false;
609 	int err;
610 
611 	if (wc->status != IB_WC_SUCCESS) {
612 		if (wc->status != IB_WC_WR_FLUSH_ERR) {
613 			rtrs_err(clt_path->clt, "RDMA failed: %s\n",
614 				  ib_wc_status_msg(wc->status));
615 			rtrs_rdma_error_recovery(con);
616 		}
617 		return;
618 	}
619 	rtrs_clt_update_wc_stats(con);
620 
621 	switch (wc->opcode) {
622 	case IB_WC_RECV_RDMA_WITH_IMM:
623 		/*
624 		 * post_recv() RDMA write completions of IO reqs (read/write)
625 		 * and hb
626 		 */
627 		if (WARN_ON(wc->wr_cqe->done != rtrs_clt_rdma_done))
628 			return;
629 		rtrs_from_imm(be32_to_cpu(wc->ex.imm_data),
630 			       &imm_type, &imm_payload);
631 		if (imm_type == RTRS_IO_RSP_IMM ||
632 		    imm_type == RTRS_IO_RSP_W_INV_IMM) {
633 			u32 msg_id;
634 
635 			w_inval = (imm_type == RTRS_IO_RSP_W_INV_IMM);
636 			rtrs_from_io_rsp_imm(imm_payload, &msg_id, &err);
637 
638 			process_io_rsp(clt_path, msg_id, err, w_inval);
639 		} else if (imm_type == RTRS_HB_MSG_IMM) {
640 			WARN_ON(con->c.cid);
641 			rtrs_send_hb_ack(&clt_path->s);
642 			if (clt_path->flags & RTRS_MSG_NEW_RKEY_F)
643 				return  rtrs_clt_recv_done(con, wc);
644 		} else if (imm_type == RTRS_HB_ACK_IMM) {
645 			WARN_ON(con->c.cid);
646 			clt_path->s.hb_missed_cnt = 0;
647 			clt_path->s.hb_cur_latency =
648 				ktime_sub(ktime_get(), clt_path->s.hb_last_sent);
649 			if (clt_path->flags & RTRS_MSG_NEW_RKEY_F)
650 				return  rtrs_clt_recv_done(con, wc);
651 		} else {
652 			rtrs_wrn(con->c.path, "Unknown IMM type %u\n",
653 				  imm_type);
654 		}
655 		if (w_inval)
656 			/*
657 			 * Post x2 empty WRs: first is for this RDMA with IMM,
658 			 * second is for RECV with INV, which happened earlier.
659 			 */
660 			err = rtrs_post_recv_empty_x2(&con->c, &io_comp_cqe);
661 		else
662 			err = rtrs_post_recv_empty(&con->c, &io_comp_cqe);
663 		if (err) {
664 			rtrs_err(con->c.path, "rtrs_post_recv_empty(): %d\n",
665 				  err);
666 			rtrs_rdma_error_recovery(con);
667 		}
668 		break;
669 	case IB_WC_RECV:
670 		/*
671 		 * Key invalidations from server side
672 		 */
673 		WARN_ON(!(wc->wc_flags & IB_WC_WITH_INVALIDATE ||
674 			  wc->wc_flags & IB_WC_WITH_IMM));
675 		WARN_ON(wc->wr_cqe->done != rtrs_clt_rdma_done);
676 		if (clt_path->flags & RTRS_MSG_NEW_RKEY_F) {
677 			if (wc->wc_flags & IB_WC_WITH_INVALIDATE)
678 				return  rtrs_clt_recv_done(con, wc);
679 
680 			return  rtrs_clt_rkey_rsp_done(con, wc);
681 		}
682 		break;
683 	case IB_WC_RDMA_WRITE:
684 		/*
685 		 * post_send() RDMA write completions of IO reqs (read/write)
686 		 * and hb.
687 		 */
688 		break;
689 
690 	default:
691 		rtrs_wrn(clt_path->clt, "Unexpected WC type: %d\n", wc->opcode);
692 		return;
693 	}
694 }
695 
696 static int post_recv_io(struct rtrs_clt_con *con, size_t q_size)
697 {
698 	int err, i;
699 	struct rtrs_clt_path *clt_path = to_clt_path(con->c.path);
700 
701 	for (i = 0; i < q_size; i++) {
702 		if (clt_path->flags & RTRS_MSG_NEW_RKEY_F) {
703 			struct rtrs_iu *iu = &con->rsp_ius[i];
704 
705 			err = rtrs_iu_post_recv(&con->c, iu);
706 		} else {
707 			err = rtrs_post_recv_empty(&con->c, &io_comp_cqe);
708 		}
709 		if (err)
710 			return err;
711 	}
712 
713 	return 0;
714 }
715 
716 static int post_recv_path(struct rtrs_clt_path *clt_path)
717 {
718 	size_t q_size = 0;
719 	int err, cid;
720 
721 	for (cid = 0; cid < clt_path->s.con_num; cid++) {
722 		if (cid == 0)
723 			q_size = SERVICE_CON_QUEUE_DEPTH;
724 		else
725 			q_size = clt_path->queue_depth;
726 
727 		/*
728 		 * x2 for RDMA read responses + FR key invalidations,
729 		 * RDMA writes do not require any FR registrations.
730 		 */
731 		q_size *= 2;
732 
733 		err = post_recv_io(to_clt_con(clt_path->s.con[cid]), q_size);
734 		if (err) {
735 			rtrs_err(clt_path->clt, "post_recv_io(), err: %d\n",
736 				 err);
737 			return err;
738 		}
739 	}
740 
741 	return 0;
742 }
743 
744 struct path_it {
745 	int i;
746 	struct list_head skip_list;
747 	struct rtrs_clt_sess *clt;
748 	struct rtrs_clt_path *(*next_path)(struct path_it *it);
749 };
750 
751 /*
752  * rtrs_clt_get_next_path_or_null - get clt path from the list or return NULL
753  * @head:	the head for the list.
754  * @clt_path:	The element to take the next clt_path from.
755  *
756  * Next clt path returned in round-robin fashion, i.e. head will be skipped,
757  * but if list is observed as empty, NULL will be returned.
758  *
759  * This function may safely run concurrently with the _rcu list-mutation
760  * primitives such as list_add_rcu() as long as it's guarded by rcu_read_lock().
761  */
762 static inline struct rtrs_clt_path *
763 rtrs_clt_get_next_path_or_null(struct list_head *head, struct rtrs_clt_path *clt_path)
764 {
765 	return list_next_or_null_rcu(head, &clt_path->s.entry, typeof(*clt_path), s.entry) ?:
766 				     list_next_or_null_rcu(head,
767 							   READ_ONCE((&clt_path->s.entry)->next),
768 							   typeof(*clt_path), s.entry);
769 }
770 
771 /**
772  * get_next_path_rr() - Returns path in round-robin fashion.
773  * @it:	the path pointer
774  *
775  * Related to @MP_POLICY_RR
776  *
777  * Locks:
778  *    rcu_read_lock() must be held.
779  */
780 static struct rtrs_clt_path *get_next_path_rr(struct path_it *it)
781 {
782 	struct rtrs_clt_path __rcu **ppcpu_path;
783 	struct rtrs_clt_path *path;
784 	struct rtrs_clt_sess *clt;
785 
786 	/*
787 	 * Assert that rcu lock must be held
788 	 */
789 	RCU_LOCKDEP_WARN(!rcu_read_lock_held(), "no rcu read lock held");
790 
791 	clt = it->clt;
792 
793 	/*
794 	 * Here we use two RCU objects: @paths_list and @pcpu_path
795 	 * pointer.  See rtrs_clt_remove_path_from_arr() for details
796 	 * how that is handled.
797 	 */
798 
799 	ppcpu_path = this_cpu_ptr(clt->pcpu_path);
800 	path = rcu_dereference(*ppcpu_path);
801 	if (!path)
802 		path = list_first_or_null_rcu(&clt->paths_list,
803 					      typeof(*path), s.entry);
804 	else
805 		path = rtrs_clt_get_next_path_or_null(&clt->paths_list, path);
806 
807 	rcu_assign_pointer(*ppcpu_path, path);
808 
809 	return path;
810 }
811 
812 /**
813  * get_next_path_min_inflight() - Returns path with minimal inflight count.
814  * @it:	the path pointer
815  *
816  * Related to @MP_POLICY_MIN_INFLIGHT
817  *
818  * Locks:
819  *    rcu_read_lock() must be hold.
820  */
821 static struct rtrs_clt_path *get_next_path_min_inflight(struct path_it *it)
822 {
823 	struct rtrs_clt_path *min_path = NULL;
824 	struct rtrs_clt_sess *clt = it->clt;
825 	struct rtrs_clt_path *clt_path;
826 	int min_inflight = INT_MAX;
827 	int inflight;
828 
829 	list_for_each_entry_rcu(clt_path, &clt->paths_list, s.entry) {
830 		if (READ_ONCE(clt_path->state) != RTRS_CLT_CONNECTED)
831 			continue;
832 
833 		if (!list_empty(raw_cpu_ptr(clt_path->mp_skip_entry)))
834 			continue;
835 
836 		inflight = atomic_read(&clt_path->stats->inflight);
837 
838 		if (inflight < min_inflight) {
839 			min_inflight = inflight;
840 			min_path = clt_path;
841 		}
842 	}
843 
844 	/*
845 	 * add the path to the skip list, so that next time we can get
846 	 * a different one
847 	 */
848 	if (min_path)
849 		list_add(raw_cpu_ptr(min_path->mp_skip_entry), &it->skip_list);
850 
851 	return min_path;
852 }
853 
854 /**
855  * get_next_path_min_latency() - Returns path with minimal latency.
856  * @it:	the path pointer
857  *
858  * Return: a path with the lowest latency or NULL if all paths are tried
859  *
860  * Locks:
861  *    rcu_read_lock() must be hold.
862  *
863  * Related to @MP_POLICY_MIN_LATENCY
864  *
865  * This DOES skip an already-tried path.
866  * There is a skip-list to skip a path if the path has tried but failed.
867  * It will try the minimum latency path and then the second minimum latency
868  * path and so on. Finally it will return NULL if all paths are tried.
869  * Therefore the caller MUST check the returned
870  * path is NULL and trigger the IO error.
871  */
872 static struct rtrs_clt_path *get_next_path_min_latency(struct path_it *it)
873 {
874 	struct rtrs_clt_path *min_path = NULL;
875 	struct rtrs_clt_sess *clt = it->clt;
876 	struct rtrs_clt_path *clt_path;
877 	ktime_t min_latency = KTIME_MAX;
878 	ktime_t latency;
879 
880 	list_for_each_entry_rcu(clt_path, &clt->paths_list, s.entry) {
881 		if (READ_ONCE(clt_path->state) != RTRS_CLT_CONNECTED)
882 			continue;
883 
884 		if (!list_empty(raw_cpu_ptr(clt_path->mp_skip_entry)))
885 			continue;
886 
887 		latency = clt_path->s.hb_cur_latency;
888 
889 		if (latency < min_latency) {
890 			min_latency = latency;
891 			min_path = clt_path;
892 		}
893 	}
894 
895 	/*
896 	 * add the path to the skip list, so that next time we can get
897 	 * a different one
898 	 */
899 	if (min_path)
900 		list_add(raw_cpu_ptr(min_path->mp_skip_entry), &it->skip_list);
901 
902 	return min_path;
903 }
904 
905 static inline void path_it_init(struct path_it *it, struct rtrs_clt_sess *clt)
906 {
907 	INIT_LIST_HEAD(&it->skip_list);
908 	it->clt = clt;
909 	it->i = 0;
910 
911 	if (clt->mp_policy == MP_POLICY_RR)
912 		it->next_path = get_next_path_rr;
913 	else if (clt->mp_policy == MP_POLICY_MIN_INFLIGHT)
914 		it->next_path = get_next_path_min_inflight;
915 	else
916 		it->next_path = get_next_path_min_latency;
917 }
918 
919 static inline void path_it_deinit(struct path_it *it)
920 {
921 	struct list_head *skip, *tmp;
922 	/*
923 	 * The skip_list is used only for the MIN_INFLIGHT and MIN_LATENCY policies.
924 	 * We need to remove paths from it, so that next IO can insert
925 	 * paths (->mp_skip_entry) into a skip_list again.
926 	 */
927 	list_for_each_safe(skip, tmp, &it->skip_list)
928 		list_del_init(skip);
929 }
930 
931 /**
932  * rtrs_clt_init_req() - Initialize an rtrs_clt_io_req holding information
933  * about an inflight IO.
934  * The user buffer holding user control message (not data) is copied into
935  * the corresponding buffer of rtrs_iu (req->iu->buf), which later on will
936  * also hold the control message of rtrs.
937  * @req: an io request holding information about IO.
938  * @clt_path: client path
939  * @conf: conformation callback function to notify upper layer.
940  * @permit: permit for allocation of RDMA remote buffer
941  * @priv: private pointer
942  * @vec: kernel vector containing control message
943  * @usr_len: length of the user message
944  * @sg: scater list for IO data
945  * @sg_cnt: number of scater list entries
946  * @data_len: length of the IO data
947  * @dir: direction of the IO.
948  */
949 static void rtrs_clt_init_req(struct rtrs_clt_io_req *req,
950 			      struct rtrs_clt_path *clt_path,
951 			      void (*conf)(void *priv, int errno),
952 			      struct rtrs_permit *permit, void *priv,
953 			      const struct kvec *vec, size_t usr_len,
954 			      struct scatterlist *sg, size_t sg_cnt,
955 			      size_t data_len, int dir)
956 {
957 	struct iov_iter iter;
958 	size_t len;
959 
960 	req->permit = permit;
961 	req->in_use = true;
962 	req->usr_len = usr_len;
963 	req->data_len = data_len;
964 	req->sglist = sg;
965 	req->sg_cnt = sg_cnt;
966 	req->priv = priv;
967 	req->dir = dir;
968 	req->con = rtrs_permit_to_clt_con(clt_path, permit);
969 	req->conf = conf;
970 	req->need_inv = false;
971 	req->need_inv_comp = false;
972 	req->inv_errno = 0;
973 	refcount_set(&req->ref, 1);
974 	req->mp_policy = clt_path->clt->mp_policy;
975 
976 	iov_iter_kvec(&iter, ITER_SOURCE, vec, 1, usr_len);
977 	len = _copy_from_iter(req->iu->buf, usr_len, &iter);
978 	WARN_ON(len != usr_len);
979 
980 	reinit_completion(&req->inv_comp);
981 }
982 
983 static struct rtrs_clt_io_req *
984 rtrs_clt_get_req(struct rtrs_clt_path *clt_path,
985 		 void (*conf)(void *priv, int errno),
986 		 struct rtrs_permit *permit, void *priv,
987 		 const struct kvec *vec, size_t usr_len,
988 		 struct scatterlist *sg, size_t sg_cnt,
989 		 size_t data_len, int dir)
990 {
991 	struct rtrs_clt_io_req *req;
992 
993 	req = &clt_path->reqs[permit->mem_id];
994 	rtrs_clt_init_req(req, clt_path, conf, permit, priv, vec, usr_len,
995 			   sg, sg_cnt, data_len, dir);
996 	return req;
997 }
998 
999 static struct rtrs_clt_io_req *
1000 rtrs_clt_get_copy_req(struct rtrs_clt_path *alive_path,
1001 		       struct rtrs_clt_io_req *fail_req)
1002 {
1003 	struct rtrs_clt_io_req *req;
1004 	struct kvec vec = {
1005 		.iov_base = fail_req->iu->buf,
1006 		.iov_len  = fail_req->usr_len
1007 	};
1008 
1009 	req = &alive_path->reqs[fail_req->permit->mem_id];
1010 	rtrs_clt_init_req(req, alive_path, fail_req->conf, fail_req->permit,
1011 			   fail_req->priv, &vec, fail_req->usr_len,
1012 			   fail_req->sglist, fail_req->sg_cnt,
1013 			   fail_req->data_len, fail_req->dir);
1014 	return req;
1015 }
1016 
1017 static int rtrs_post_rdma_write_sg(struct rtrs_clt_con *con,
1018 				   struct rtrs_clt_io_req *req,
1019 				   struct rtrs_rbuf *rbuf, bool fr_en,
1020 				   u32 count, u32 size, u32 imm,
1021 				   struct ib_send_wr *wr,
1022 				   struct ib_send_wr *tail)
1023 {
1024 	struct rtrs_clt_path *clt_path = to_clt_path(con->c.path);
1025 	struct ib_sge *sge = req->sge;
1026 	enum ib_send_flags flags;
1027 	struct scatterlist *sg;
1028 	size_t num_sge;
1029 	int i;
1030 	struct ib_send_wr *ptail = NULL;
1031 
1032 	if (fr_en) {
1033 		i = 0;
1034 		sge[i].addr   = req->mr->iova;
1035 		sge[i].length = req->mr->length;
1036 		sge[i].lkey   = req->mr->lkey;
1037 		i++;
1038 		num_sge = 2;
1039 		ptail = tail;
1040 	} else {
1041 		for_each_sg(req->sglist, sg, count, i) {
1042 			sge[i].addr   = sg_dma_address(sg);
1043 			sge[i].length = sg_dma_len(sg);
1044 			sge[i].lkey   = clt_path->s.dev->ib_pd->local_dma_lkey;
1045 		}
1046 		num_sge = 1 + count;
1047 	}
1048 	sge[i].addr   = req->iu->dma_addr;
1049 	sge[i].length = size;
1050 	sge[i].lkey   = clt_path->s.dev->ib_pd->local_dma_lkey;
1051 
1052 	/*
1053 	 * From time to time we have to post signalled sends,
1054 	 * or send queue will fill up and only QP reset can help.
1055 	 */
1056 	flags = atomic_inc_return(&con->c.wr_cnt) % clt_path->s.signal_interval ?
1057 			0 : IB_SEND_SIGNALED;
1058 
1059 	ib_dma_sync_single_for_device(clt_path->s.dev->ib_dev,
1060 				      req->iu->dma_addr,
1061 				      size, DMA_TO_DEVICE);
1062 
1063 	return rtrs_iu_post_rdma_write_imm(&con->c, req->iu, sge, num_sge,
1064 					    rbuf->rkey, rbuf->addr, imm,
1065 					    flags, wr, ptail);
1066 }
1067 
1068 static int rtrs_map_sg_fr(struct rtrs_clt_io_req *req, size_t count)
1069 {
1070 	int nr;
1071 
1072 	/* Align the MR to a 4K page size to match the block virt boundary */
1073 	nr = ib_map_mr_sg(req->mr, req->sglist, count, NULL, SZ_4K);
1074 	if (nr != count)
1075 		return nr < 0 ? nr : -EINVAL;
1076 	ib_update_fast_reg_key(req->mr, ib_inc_rkey(req->mr->rkey));
1077 
1078 	return nr;
1079 }
1080 
1081 static int rtrs_clt_write_req(struct rtrs_clt_io_req *req)
1082 {
1083 	struct rtrs_clt_con *con = req->con;
1084 	struct rtrs_path *s = con->c.path;
1085 	struct rtrs_clt_path *clt_path = to_clt_path(s);
1086 	struct rtrs_msg_rdma_write *msg;
1087 
1088 	struct rtrs_rbuf *rbuf;
1089 	int ret, count = 0;
1090 	u32 imm, buf_id;
1091 	struct ib_reg_wr rwr;
1092 	struct ib_send_wr inv_wr;
1093 	struct ib_send_wr *wr = NULL;
1094 	bool fr_en = false;
1095 
1096 	const size_t tsize = sizeof(*msg) + req->data_len + req->usr_len;
1097 
1098 	if (tsize > clt_path->chunk_size) {
1099 		rtrs_wrn(s, "Write request failed, size too big %zu > %d\n",
1100 			  tsize, clt_path->chunk_size);
1101 		return -EMSGSIZE;
1102 	}
1103 	if (req->sg_cnt) {
1104 		count = ib_dma_map_sg(clt_path->s.dev->ib_dev, req->sglist,
1105 				      req->sg_cnt, req->dir);
1106 		if (!count) {
1107 			rtrs_wrn(s, "Write request failed, map failed\n");
1108 			return -EINVAL;
1109 		}
1110 	}
1111 	/* put rtrs msg after sg and user message */
1112 	msg = req->iu->buf + req->usr_len;
1113 	msg->type = cpu_to_le16(RTRS_MSG_WRITE);
1114 	msg->usr_len = cpu_to_le16(req->usr_len);
1115 
1116 	/* rtrs message on server side will be after user data and message */
1117 	imm = req->permit->mem_off + req->data_len + req->usr_len;
1118 	imm = rtrs_to_io_req_imm(imm);
1119 	buf_id = req->permit->mem_id;
1120 	req->sg_size = tsize;
1121 	rbuf = &clt_path->rbufs[buf_id];
1122 
1123 	if (count) {
1124 		ret = rtrs_map_sg_fr(req, count);
1125 		if (ret < 0) {
1126 			rtrs_err_rl(s,
1127 				    "Write request failed, failed to map fast reg. data, err: %d\n",
1128 				    ret);
1129 			ib_dma_unmap_sg(clt_path->s.dev->ib_dev, req->sglist,
1130 					req->sg_cnt, req->dir);
1131 			return ret;
1132 		}
1133 		inv_wr = (struct ib_send_wr) {
1134 			.opcode		    = IB_WR_LOCAL_INV,
1135 			.wr_cqe		    = &req->inv_cqe,
1136 			.send_flags	    = IB_SEND_SIGNALED,
1137 			.ex.invalidate_rkey = req->mr->rkey,
1138 		};
1139 		req->inv_cqe.done = rtrs_clt_inv_rkey_done;
1140 		rwr = (struct ib_reg_wr) {
1141 			.wr.opcode = IB_WR_REG_MR,
1142 			.wr.wr_cqe = &fast_reg_cqe,
1143 			.mr = req->mr,
1144 			.key = req->mr->rkey,
1145 			.access = (IB_ACCESS_LOCAL_WRITE),
1146 		};
1147 		wr = &rwr.wr;
1148 		fr_en = true;
1149 		refcount_inc(&req->ref);
1150 	}
1151 	/*
1152 	 * Update stats now, after request is successfully sent it is not
1153 	 * safe anymore to touch it.
1154 	 */
1155 	rtrs_clt_update_all_stats(req, WRITE);
1156 
1157 	ret = rtrs_post_rdma_write_sg(req->con, req, rbuf, fr_en, count,
1158 				      req->usr_len + sizeof(*msg),
1159 				      imm, wr, &inv_wr);
1160 	if (ret) {
1161 		rtrs_err_rl(s,
1162 			    "Write request failed: error=%d path=%s [%s:%u]\n",
1163 			    ret, kobject_name(&clt_path->kobj), clt_path->hca_name,
1164 			    clt_path->hca_port);
1165 		if (req->mp_policy == MP_POLICY_MIN_INFLIGHT)
1166 			atomic_dec(&clt_path->stats->inflight);
1167 		if (req->sg_cnt)
1168 			ib_dma_unmap_sg(clt_path->s.dev->ib_dev, req->sglist,
1169 					req->sg_cnt, req->dir);
1170 	}
1171 
1172 	return ret;
1173 }
1174 
1175 static int rtrs_clt_read_req(struct rtrs_clt_io_req *req)
1176 {
1177 	struct rtrs_clt_con *con = req->con;
1178 	struct rtrs_path *s = con->c.path;
1179 	struct rtrs_clt_path *clt_path = to_clt_path(s);
1180 	struct rtrs_msg_rdma_read *msg;
1181 	struct rtrs_ib_dev *dev = clt_path->s.dev;
1182 
1183 	struct ib_reg_wr rwr;
1184 	struct ib_send_wr *wr = NULL;
1185 
1186 	int ret, count = 0;
1187 	u32 imm, buf_id;
1188 
1189 	const size_t tsize = sizeof(*msg) + req->data_len + req->usr_len;
1190 
1191 	if (tsize > clt_path->chunk_size) {
1192 		rtrs_wrn(s,
1193 			  "Read request failed, message size is %zu, bigger than CHUNK_SIZE %d\n",
1194 			  tsize, clt_path->chunk_size);
1195 		return -EMSGSIZE;
1196 	}
1197 
1198 	if (req->sg_cnt) {
1199 		count = ib_dma_map_sg(dev->ib_dev, req->sglist, req->sg_cnt,
1200 				      req->dir);
1201 		if (!count) {
1202 			rtrs_wrn(s,
1203 				  "Read request failed, dma map failed\n");
1204 			return -EINVAL;
1205 		}
1206 	}
1207 	/* put our message into req->buf after user message*/
1208 	msg = req->iu->buf + req->usr_len;
1209 	msg->type = cpu_to_le16(RTRS_MSG_READ);
1210 	msg->usr_len = cpu_to_le16(req->usr_len);
1211 
1212 	if (count) {
1213 		ret = rtrs_map_sg_fr(req, count);
1214 		if (ret < 0) {
1215 			rtrs_err_rl(s,
1216 				     "Read request failed, failed to map  fast reg. data, err: %d\n",
1217 				     ret);
1218 			ib_dma_unmap_sg(dev->ib_dev, req->sglist, req->sg_cnt,
1219 					req->dir);
1220 			return ret;
1221 		}
1222 		rwr = (struct ib_reg_wr) {
1223 			.wr.opcode = IB_WR_REG_MR,
1224 			.wr.wr_cqe = &fast_reg_cqe,
1225 			.mr = req->mr,
1226 			.key = req->mr->rkey,
1227 			.access = (IB_ACCESS_LOCAL_WRITE |
1228 				   IB_ACCESS_REMOTE_WRITE),
1229 		};
1230 		wr = &rwr.wr;
1231 
1232 		msg->sg_cnt = cpu_to_le16(1);
1233 		msg->flags = cpu_to_le16(RTRS_MSG_NEED_INVAL_F);
1234 
1235 		msg->desc[0].addr = cpu_to_le64(req->mr->iova);
1236 		msg->desc[0].key = cpu_to_le32(req->mr->rkey);
1237 		msg->desc[0].len = cpu_to_le32(req->mr->length);
1238 
1239 		/* Further invalidation is required */
1240 		req->need_inv = !!RTRS_MSG_NEED_INVAL_F;
1241 
1242 	} else {
1243 		msg->sg_cnt = 0;
1244 		msg->flags = 0;
1245 	}
1246 	/*
1247 	 * rtrs message will be after the space reserved for disk data and
1248 	 * user message
1249 	 */
1250 	imm = req->permit->mem_off + req->data_len + req->usr_len;
1251 	imm = rtrs_to_io_req_imm(imm);
1252 	buf_id = req->permit->mem_id;
1253 
1254 	req->sg_size  = sizeof(*msg);
1255 	req->sg_size += le16_to_cpu(msg->sg_cnt) * sizeof(struct rtrs_sg_desc);
1256 	req->sg_size += req->usr_len;
1257 
1258 	/*
1259 	 * Update stats now, after request is successfully sent it is not
1260 	 * safe anymore to touch it.
1261 	 */
1262 	rtrs_clt_update_all_stats(req, READ);
1263 
1264 	ret = rtrs_post_send_rdma(req->con, req, &clt_path->rbufs[buf_id],
1265 				   req->data_len, imm, wr);
1266 	if (ret) {
1267 		rtrs_err_rl(s,
1268 			    "Read request failed: error=%d path=%s [%s:%u]\n",
1269 			    ret, kobject_name(&clt_path->kobj), clt_path->hca_name,
1270 			    clt_path->hca_port);
1271 		if (req->mp_policy == MP_POLICY_MIN_INFLIGHT)
1272 			atomic_dec(&clt_path->stats->inflight);
1273 		req->need_inv = false;
1274 		if (req->sg_cnt)
1275 			ib_dma_unmap_sg(dev->ib_dev, req->sglist,
1276 					req->sg_cnt, req->dir);
1277 	}
1278 
1279 	return ret;
1280 }
1281 
1282 /**
1283  * rtrs_clt_failover_req() - Try to find an active path for a failed request
1284  * @clt: clt context
1285  * @fail_req: a failed io request.
1286  */
1287 static int rtrs_clt_failover_req(struct rtrs_clt_sess *clt,
1288 				 struct rtrs_clt_io_req *fail_req)
1289 {
1290 	struct rtrs_clt_path *alive_path;
1291 	struct rtrs_clt_io_req *req;
1292 	int err = -ECONNABORTED;
1293 	struct path_it it;
1294 
1295 	rcu_read_lock();
1296 	for (path_it_init(&it, clt);
1297 	     (alive_path = it.next_path(&it)) && it.i < it.clt->paths_num;
1298 	     it.i++) {
1299 		if (READ_ONCE(alive_path->state) != RTRS_CLT_CONNECTED)
1300 			continue;
1301 		req = rtrs_clt_get_copy_req(alive_path, fail_req);
1302 		if (req->dir == DMA_TO_DEVICE)
1303 			err = rtrs_clt_write_req(req);
1304 		else
1305 			err = rtrs_clt_read_req(req);
1306 		if (err) {
1307 			req->in_use = false;
1308 			continue;
1309 		}
1310 		/* Success path */
1311 		rtrs_clt_inc_failover_cnt(alive_path->stats);
1312 		break;
1313 	}
1314 	path_it_deinit(&it);
1315 	rcu_read_unlock();
1316 
1317 	return err;
1318 }
1319 
1320 static void fail_all_outstanding_reqs(struct rtrs_clt_path *clt_path)
1321 {
1322 	struct rtrs_clt_sess *clt = clt_path->clt;
1323 	struct rtrs_clt_io_req *req;
1324 	int i, err;
1325 
1326 	if (!clt_path->reqs)
1327 		return;
1328 	for (i = 0; i < clt_path->queue_depth; ++i) {
1329 		req = &clt_path->reqs[i];
1330 		if (!req->in_use)
1331 			continue;
1332 
1333 		/*
1334 		 * Safely (without notification) complete failed request.
1335 		 * After completion this request is still useble and can
1336 		 * be failovered to another path.
1337 		 */
1338 		complete_rdma_req(req, -ECONNABORTED, false, true);
1339 
1340 		err = rtrs_clt_failover_req(clt, req);
1341 		if (err)
1342 			/* Failover failed, notify anyway */
1343 			req->conf(req->priv, err);
1344 	}
1345 }
1346 
1347 static void free_path_reqs(struct rtrs_clt_path *clt_path)
1348 {
1349 	struct rtrs_clt_io_req *req;
1350 	int i;
1351 
1352 	if (!clt_path->reqs)
1353 		return;
1354 	for (i = 0; i < clt_path->queue_depth; ++i) {
1355 		req = &clt_path->reqs[i];
1356 		if (req->mr)
1357 			ib_dereg_mr(req->mr);
1358 		kfree(req->sge);
1359 		rtrs_iu_free(req->iu, clt_path->s.dev->ib_dev, 1);
1360 	}
1361 	kfree(clt_path->reqs);
1362 	clt_path->reqs = NULL;
1363 }
1364 
1365 static int alloc_path_reqs(struct rtrs_clt_path *clt_path)
1366 {
1367 	struct rtrs_clt_io_req *req;
1368 	int i, err = -ENOMEM;
1369 
1370 	clt_path->reqs = kcalloc(clt_path->queue_depth,
1371 				 sizeof(*clt_path->reqs),
1372 				 GFP_KERNEL);
1373 	if (!clt_path->reqs)
1374 		return -ENOMEM;
1375 
1376 	for (i = 0; i < clt_path->queue_depth; ++i) {
1377 		req = &clt_path->reqs[i];
1378 		req->iu = rtrs_iu_alloc(1, clt_path->max_hdr_size, GFP_KERNEL,
1379 					 clt_path->s.dev->ib_dev,
1380 					 DMA_TO_DEVICE,
1381 					 rtrs_clt_rdma_done);
1382 		if (!req->iu)
1383 			goto out;
1384 
1385 		req->sge = kcalloc(2, sizeof(*req->sge), GFP_KERNEL);
1386 		if (!req->sge)
1387 			goto out;
1388 
1389 		req->mr = ib_alloc_mr(clt_path->s.dev->ib_pd,
1390 				      IB_MR_TYPE_MEM_REG,
1391 				      clt_path->max_pages_per_mr);
1392 		if (IS_ERR(req->mr)) {
1393 			err = PTR_ERR(req->mr);
1394 			pr_err("Failed to alloc clt_path->max_pages_per_mr %d: %pe\n",
1395 			       clt_path->max_pages_per_mr, req->mr);
1396 			req->mr = NULL;
1397 			goto out;
1398 		}
1399 
1400 		init_completion(&req->inv_comp);
1401 	}
1402 
1403 	return 0;
1404 
1405 out:
1406 	free_path_reqs(clt_path);
1407 
1408 	return err;
1409 }
1410 
1411 static int alloc_permits(struct rtrs_clt_sess *clt)
1412 {
1413 	unsigned int chunk_bits;
1414 	int err, i;
1415 
1416 	clt->permits_map = bitmap_zalloc(clt->queue_depth, GFP_KERNEL);
1417 	if (!clt->permits_map) {
1418 		err = -ENOMEM;
1419 		goto out_err;
1420 	}
1421 	clt->permits = kcalloc(clt->queue_depth, permit_size(clt), GFP_KERNEL);
1422 	if (!clt->permits) {
1423 		err = -ENOMEM;
1424 		goto err_map;
1425 	}
1426 	chunk_bits = ilog2(clt->queue_depth - 1) + 1;
1427 	for (i = 0; i < clt->queue_depth; i++) {
1428 		struct rtrs_permit *permit;
1429 
1430 		permit = get_permit(clt, i);
1431 		permit->mem_id = i;
1432 		permit->mem_off = i << (MAX_IMM_PAYL_BITS - chunk_bits);
1433 	}
1434 
1435 	return 0;
1436 
1437 err_map:
1438 	bitmap_free(clt->permits_map);
1439 	clt->permits_map = NULL;
1440 out_err:
1441 	return err;
1442 }
1443 
1444 static void free_permits(struct rtrs_clt_sess *clt)
1445 {
1446 	if (clt->permits_map)
1447 		wait_event(clt->permits_wait,
1448 			   bitmap_empty(clt->permits_map, clt->queue_depth));
1449 
1450 	bitmap_free(clt->permits_map);
1451 	clt->permits_map = NULL;
1452 	kfree(clt->permits);
1453 	clt->permits = NULL;
1454 }
1455 
1456 static void query_fast_reg_mode(struct rtrs_clt_path *clt_path)
1457 {
1458 	struct ib_device *ib_dev;
1459 	u64 max_pages_per_mr;
1460 	int mr_page_shift;
1461 
1462 	ib_dev = clt_path->s.dev->ib_dev;
1463 
1464 	/*
1465 	 * Use the smallest page size supported by the HCA, down to a
1466 	 * minimum of 4096 bytes. We're unlikely to build large sglists
1467 	 * out of smaller entries.
1468 	 */
1469 	mr_page_shift      = max(12, ffs(ib_dev->attrs.page_size_cap) - 1);
1470 	max_pages_per_mr   = ib_dev->attrs.max_mr_size;
1471 	do_div(max_pages_per_mr, (1ull << mr_page_shift));
1472 	clt_path->max_pages_per_mr =
1473 		min3(clt_path->max_pages_per_mr, (u32)max_pages_per_mr,
1474 		     ib_dev->attrs.max_fast_reg_page_list_len);
1475 	clt_path->clt->max_segments =
1476 		min(clt_path->max_pages_per_mr, clt_path->clt->max_segments);
1477 }
1478 
1479 static bool rtrs_clt_change_state_get_old(struct rtrs_clt_path *clt_path,
1480 					   enum rtrs_clt_state new_state,
1481 					   enum rtrs_clt_state *old_state)
1482 {
1483 	bool changed;
1484 
1485 	spin_lock_irq(&clt_path->state_wq.lock);
1486 	if (old_state)
1487 		*old_state = clt_path->state;
1488 	changed = rtrs_clt_change_state(clt_path, new_state);
1489 	spin_unlock_irq(&clt_path->state_wq.lock);
1490 
1491 	return changed;
1492 }
1493 
1494 static void rtrs_clt_hb_err_handler(struct rtrs_con *c)
1495 {
1496 	struct rtrs_clt_con *con = container_of(c, typeof(*con), c);
1497 
1498 	rtrs_rdma_error_recovery(con);
1499 }
1500 
1501 static void rtrs_clt_init_hb(struct rtrs_clt_path *clt_path)
1502 {
1503 	rtrs_init_hb(&clt_path->s, &io_comp_cqe,
1504 		      RTRS_HB_INTERVAL_MS,
1505 		      RTRS_HB_MISSED_MAX,
1506 		      rtrs_clt_hb_err_handler,
1507 		      rtrs_wq);
1508 }
1509 
1510 static void rtrs_clt_reconnect_work(struct work_struct *work);
1511 static void rtrs_clt_close_work(struct work_struct *work);
1512 
1513 static void rtrs_clt_err_recovery_work(struct work_struct *work)
1514 {
1515 	struct rtrs_clt_path *clt_path;
1516 	struct rtrs_clt_sess *clt;
1517 	int delay_ms;
1518 
1519 	clt_path = container_of(work, struct rtrs_clt_path, err_recovery_work);
1520 	clt = clt_path->clt;
1521 	delay_ms = clt->reconnect_delay_sec * 1000;
1522 	rtrs_clt_stop_and_destroy_conns(clt_path);
1523 	queue_delayed_work(rtrs_wq, &clt_path->reconnect_dwork,
1524 			   msecs_to_jiffies(delay_ms +
1525 					    get_random_u32_below(RTRS_RECONNECT_SEED)));
1526 }
1527 
1528 static struct rtrs_clt_path *alloc_path(struct rtrs_clt_sess *clt,
1529 					const struct rtrs_addr *path,
1530 					size_t con_num, u32 nr_poll_queues)
1531 {
1532 	struct rtrs_clt_path *clt_path;
1533 	int err = -ENOMEM;
1534 	int cpu;
1535 	size_t total_con;
1536 
1537 	clt_path = kzalloc(sizeof(*clt_path), GFP_KERNEL);
1538 	if (!clt_path)
1539 		goto err;
1540 
1541 	/*
1542 	 * irqmode and poll
1543 	 * +1: Extra connection for user messages
1544 	 */
1545 	total_con = con_num + nr_poll_queues + 1;
1546 	clt_path->s.con = kcalloc(total_con, sizeof(*clt_path->s.con),
1547 				  GFP_KERNEL);
1548 	if (!clt_path->s.con)
1549 		goto err_free_path;
1550 
1551 	clt_path->s.con_num = total_con;
1552 	clt_path->s.irq_con_num = con_num + 1;
1553 
1554 	clt_path->stats = kzalloc(sizeof(*clt_path->stats), GFP_KERNEL);
1555 	if (!clt_path->stats)
1556 		goto err_free_con;
1557 
1558 	mutex_init(&clt_path->init_mutex);
1559 	uuid_gen(&clt_path->s.uuid);
1560 	memcpy(&clt_path->s.dst_addr, path->dst,
1561 	       rdma_addr_size((struct sockaddr *)path->dst));
1562 
1563 	/*
1564 	 * rdma_resolve_addr() passes src_addr to cma_bind_addr, which
1565 	 * checks the sa_family to be non-zero. If user passed src_addr=NULL
1566 	 * the sess->src_addr will contain only zeros, which is then fine.
1567 	 */
1568 	if (path->src)
1569 		memcpy(&clt_path->s.src_addr, path->src,
1570 		       rdma_addr_size((struct sockaddr *)path->src));
1571 	strscpy(clt_path->s.sessname, clt->sessname,
1572 		sizeof(clt_path->s.sessname));
1573 	clt_path->clt = clt;
1574 	clt_path->max_pages_per_mr = RTRS_MAX_SEGMENTS;
1575 	init_waitqueue_head(&clt_path->state_wq);
1576 	clt_path->state = RTRS_CLT_CONNECTING;
1577 	atomic_set(&clt_path->connected_cnt, 0);
1578 	INIT_WORK(&clt_path->close_work, rtrs_clt_close_work);
1579 	INIT_WORK(&clt_path->err_recovery_work, rtrs_clt_err_recovery_work);
1580 	INIT_DELAYED_WORK(&clt_path->reconnect_dwork, rtrs_clt_reconnect_work);
1581 	rtrs_clt_init_hb(clt_path);
1582 
1583 	clt_path->mp_skip_entry = alloc_percpu(typeof(*clt_path->mp_skip_entry));
1584 	if (!clt_path->mp_skip_entry)
1585 		goto err_free_stats;
1586 
1587 	for_each_possible_cpu(cpu)
1588 		INIT_LIST_HEAD(per_cpu_ptr(clt_path->mp_skip_entry, cpu));
1589 
1590 	err = rtrs_clt_init_stats(clt_path->stats);
1591 	if (err)
1592 		goto err_free_percpu;
1593 
1594 	return clt_path;
1595 
1596 err_free_percpu:
1597 	free_percpu(clt_path->mp_skip_entry);
1598 err_free_stats:
1599 	kfree(clt_path->stats);
1600 err_free_con:
1601 	kfree(clt_path->s.con);
1602 err_free_path:
1603 	kfree(clt_path);
1604 err:
1605 	return ERR_PTR(err);
1606 }
1607 
1608 void free_path(struct rtrs_clt_path *clt_path)
1609 {
1610 	free_percpu(clt_path->mp_skip_entry);
1611 	mutex_destroy(&clt_path->init_mutex);
1612 	kfree(clt_path->s.con);
1613 	kfree(clt_path->rbufs);
1614 	kfree(clt_path);
1615 }
1616 
1617 static int create_con(struct rtrs_clt_path *clt_path, unsigned int cid)
1618 {
1619 	struct rtrs_clt_con *con;
1620 
1621 	con = kzalloc(sizeof(*con), GFP_KERNEL);
1622 	if (!con)
1623 		return -ENOMEM;
1624 
1625 	/* Map first two connections to the first CPU */
1626 	con->cpu  = (cid ? cid - 1 : 0) % nr_cpu_ids;
1627 	con->c.cid = cid;
1628 	con->c.path = &clt_path->s;
1629 	/* Align with srv, init as 1 */
1630 	atomic_set(&con->c.wr_cnt, 1);
1631 	mutex_init(&con->con_mutex);
1632 
1633 	clt_path->s.con[cid] = &con->c;
1634 
1635 	return 0;
1636 }
1637 
1638 static void destroy_con(struct rtrs_clt_con *con)
1639 {
1640 	struct rtrs_clt_path *clt_path = to_clt_path(con->c.path);
1641 
1642 	clt_path->s.con[con->c.cid] = NULL;
1643 	mutex_destroy(&con->con_mutex);
1644 	kfree(con);
1645 }
1646 
1647 static int create_con_cq_qp(struct rtrs_clt_con *con)
1648 {
1649 	struct rtrs_clt_path *clt_path = to_clt_path(con->c.path);
1650 	u32 max_send_wr, max_recv_wr, cq_num, max_send_sge, wr_limit;
1651 	int err, cq_vector;
1652 	struct rtrs_msg_rkey_rsp *rsp;
1653 
1654 	lockdep_assert_held(&con->con_mutex);
1655 	if (con->c.cid == 0) {
1656 		max_send_sge = 1;
1657 		/* We must be the first here */
1658 		if (WARN_ON(clt_path->s.dev))
1659 			return -EINVAL;
1660 
1661 		/*
1662 		 * The whole session uses device from user connection.
1663 		 * Be careful not to close user connection before ib dev
1664 		 * is gracefully put.
1665 		 */
1666 		clt_path->s.dev = rtrs_ib_dev_find_or_add(con->c.cm_id->device,
1667 						       &dev_pd);
1668 		if (!clt_path->s.dev) {
1669 			rtrs_wrn(clt_path->clt,
1670 				  "rtrs_ib_dev_find_get_or_add(): no memory\n");
1671 			return -ENOMEM;
1672 		}
1673 		clt_path->s.dev_ref = 1;
1674 		query_fast_reg_mode(clt_path);
1675 		wr_limit = clt_path->s.dev->ib_dev->attrs.max_qp_wr;
1676 		/*
1677 		 * Two (request + registration) completion for send
1678 		 * Two for recv if always_invalidate is set on server
1679 		 * or one for recv.
1680 		 * + 2 for drain and heartbeat
1681 		 * in case qp gets into error state.
1682 		 */
1683 		max_send_wr =
1684 			min_t(int, wr_limit, SERVICE_CON_QUEUE_DEPTH * 2 + 2);
1685 		max_recv_wr = max_send_wr;
1686 	} else {
1687 		/*
1688 		 * Here we assume that session members are correctly set.
1689 		 * This is always true if user connection (cid == 0) is
1690 		 * established first.
1691 		 */
1692 		if (WARN_ON(!clt_path->s.dev))
1693 			return -EINVAL;
1694 		if (WARN_ON(!clt_path->queue_depth))
1695 			return -EINVAL;
1696 
1697 		wr_limit = clt_path->s.dev->ib_dev->attrs.max_qp_wr;
1698 		/* Shared between connections */
1699 		clt_path->s.dev_ref++;
1700 		max_send_wr = min_t(int, wr_limit,
1701 			      /* QD * (REQ + RSP + FR REGS or INVS) + drain */
1702 			      clt_path->queue_depth * 4 + 1);
1703 		max_recv_wr = min_t(int, wr_limit,
1704 			      clt_path->queue_depth * 3 + 1);
1705 		max_send_sge = 2;
1706 	}
1707 	atomic_set(&con->c.sq_wr_avail, max_send_wr);
1708 	cq_num = max_send_wr + max_recv_wr;
1709 	/* alloc iu to recv new rkey reply when server reports flags set */
1710 	if (clt_path->flags & RTRS_MSG_NEW_RKEY_F || con->c.cid == 0) {
1711 		con->rsp_ius = rtrs_iu_alloc(cq_num, sizeof(*rsp),
1712 					      GFP_KERNEL,
1713 					      clt_path->s.dev->ib_dev,
1714 					      DMA_FROM_DEVICE,
1715 					      rtrs_clt_rdma_done);
1716 		if (!con->rsp_ius)
1717 			return -ENOMEM;
1718 		con->queue_num = cq_num;
1719 	}
1720 	cq_vector = con->cpu % clt_path->s.dev->ib_dev->num_comp_vectors;
1721 	if (con->c.cid >= clt_path->s.irq_con_num)
1722 		err = rtrs_cq_qp_create(&clt_path->s, &con->c, max_send_sge,
1723 					cq_vector, cq_num, max_send_wr,
1724 					max_recv_wr, IB_POLL_DIRECT);
1725 	else
1726 		err = rtrs_cq_qp_create(&clt_path->s, &con->c, max_send_sge,
1727 					cq_vector, cq_num, max_send_wr,
1728 					max_recv_wr, IB_POLL_SOFTIRQ);
1729 	/*
1730 	 * In case of error we do not bother to clean previous allocations,
1731 	 * since destroy_con_cq_qp() must be called.
1732 	 */
1733 	return err;
1734 }
1735 
1736 static void destroy_con_cq_qp(struct rtrs_clt_con *con)
1737 {
1738 	struct rtrs_clt_path *clt_path = to_clt_path(con->c.path);
1739 
1740 	/*
1741 	 * Be careful here: destroy_con_cq_qp() can be called even
1742 	 * create_con_cq_qp() failed, see comments there.
1743 	 */
1744 	lockdep_assert_held(&con->con_mutex);
1745 	rtrs_cq_qp_destroy(&con->c);
1746 	if (con->rsp_ius) {
1747 		rtrs_iu_free(con->rsp_ius, clt_path->s.dev->ib_dev,
1748 			     con->queue_num);
1749 		con->rsp_ius = NULL;
1750 		con->queue_num = 0;
1751 	}
1752 	if (clt_path->s.dev_ref && !--clt_path->s.dev_ref) {
1753 		rtrs_ib_dev_put(clt_path->s.dev);
1754 		clt_path->s.dev = NULL;
1755 	}
1756 }
1757 
1758 static void stop_cm(struct rtrs_clt_con *con)
1759 {
1760 	rdma_disconnect(con->c.cm_id);
1761 	if (con->c.qp)
1762 		ib_drain_qp(con->c.qp);
1763 }
1764 
1765 static void destroy_cm(struct rtrs_clt_con *con)
1766 {
1767 	rdma_destroy_id(con->c.cm_id);
1768 	con->c.cm_id = NULL;
1769 }
1770 
1771 static int rtrs_rdma_addr_resolved(struct rtrs_clt_con *con)
1772 {
1773 	struct rtrs_path *s = con->c.path;
1774 	int err;
1775 
1776 	mutex_lock(&con->con_mutex);
1777 	err = create_con_cq_qp(con);
1778 	mutex_unlock(&con->con_mutex);
1779 	if (err) {
1780 		rtrs_err(s, "create_con_cq_qp(), err: %d\n", err);
1781 		return err;
1782 	}
1783 	err = rdma_resolve_route(con->c.cm_id, RTRS_CONNECT_TIMEOUT_MS);
1784 	if (err)
1785 		rtrs_err(s, "Resolving route failed, err: %d\n", err);
1786 
1787 	return err;
1788 }
1789 
1790 static int rtrs_rdma_route_resolved(struct rtrs_clt_con *con)
1791 {
1792 	struct rtrs_clt_path *clt_path = to_clt_path(con->c.path);
1793 	struct rtrs_clt_sess *clt = clt_path->clt;
1794 	struct rtrs_msg_conn_req msg;
1795 	struct rdma_conn_param param;
1796 
1797 	int err;
1798 
1799 	param = (struct rdma_conn_param) {
1800 		.retry_count = 7,
1801 		.rnr_retry_count = 7,
1802 		.private_data = &msg,
1803 		.private_data_len = sizeof(msg),
1804 	};
1805 
1806 	msg = (struct rtrs_msg_conn_req) {
1807 		.magic = cpu_to_le16(RTRS_MAGIC),
1808 		.version = cpu_to_le16(RTRS_PROTO_VER),
1809 		.cid = cpu_to_le16(con->c.cid),
1810 		.cid_num = cpu_to_le16(clt_path->s.con_num),
1811 		.recon_cnt = cpu_to_le16(clt_path->s.recon_cnt),
1812 	};
1813 	msg.first_conn = clt_path->for_new_clt ? FIRST_CONN : 0;
1814 	uuid_copy(&msg.sess_uuid, &clt_path->s.uuid);
1815 	uuid_copy(&msg.paths_uuid, &clt->paths_uuid);
1816 
1817 	err = rdma_connect_locked(con->c.cm_id, &param);
1818 	if (err)
1819 		rtrs_err(clt, "rdma_connect_locked(): %d\n", err);
1820 
1821 	return err;
1822 }
1823 
1824 static int rtrs_rdma_conn_established(struct rtrs_clt_con *con,
1825 				       struct rdma_cm_event *ev)
1826 {
1827 	struct rtrs_clt_path *clt_path = to_clt_path(con->c.path);
1828 	struct rtrs_clt_sess *clt = clt_path->clt;
1829 	const struct rtrs_msg_conn_rsp *msg;
1830 	u16 version, queue_depth;
1831 	int errno;
1832 	u8 len;
1833 
1834 	msg = ev->param.conn.private_data;
1835 	len = ev->param.conn.private_data_len;
1836 	if (len < sizeof(*msg)) {
1837 		rtrs_err(clt, "Invalid RTRS connection response\n");
1838 		return -ECONNRESET;
1839 	}
1840 	if (le16_to_cpu(msg->magic) != RTRS_MAGIC) {
1841 		rtrs_err(clt, "Invalid RTRS magic\n");
1842 		return -ECONNRESET;
1843 	}
1844 	version = le16_to_cpu(msg->version);
1845 	if (version >> 8 != RTRS_PROTO_VER_MAJOR) {
1846 		rtrs_err(clt, "Unsupported major RTRS version: %d, expected %d\n",
1847 			  version >> 8, RTRS_PROTO_VER_MAJOR);
1848 		return -ECONNRESET;
1849 	}
1850 	errno = le16_to_cpu(msg->errno);
1851 	if (errno) {
1852 		rtrs_err(clt, "Invalid RTRS message: errno %d\n",
1853 			  errno);
1854 		return -ECONNRESET;
1855 	}
1856 	if (con->c.cid == 0) {
1857 		queue_depth = le16_to_cpu(msg->queue_depth);
1858 
1859 		if (clt_path->queue_depth > 0 && queue_depth != clt_path->queue_depth) {
1860 			rtrs_err(clt, "Error: queue depth changed\n");
1861 
1862 			/*
1863 			 * Stop any more reconnection attempts
1864 			 */
1865 			clt_path->reconnect_attempts = -1;
1866 			rtrs_err(clt,
1867 				"Disabling auto-reconnect. Trigger a manual reconnect after issue is resolved\n");
1868 			return -ECONNRESET;
1869 		}
1870 
1871 		if (!clt_path->rbufs) {
1872 			clt_path->rbufs = kcalloc(queue_depth,
1873 						  sizeof(*clt_path->rbufs),
1874 						  GFP_KERNEL);
1875 			if (!clt_path->rbufs)
1876 				return -ENOMEM;
1877 		}
1878 		clt_path->queue_depth = queue_depth;
1879 		clt_path->s.signal_interval = min_not_zero(queue_depth,
1880 						(unsigned short) SERVICE_CON_QUEUE_DEPTH);
1881 		clt_path->max_hdr_size = le32_to_cpu(msg->max_hdr_size);
1882 		clt_path->max_io_size = le32_to_cpu(msg->max_io_size);
1883 		clt_path->flags = le32_to_cpu(msg->flags);
1884 		clt_path->chunk_size = clt_path->max_io_size + clt_path->max_hdr_size;
1885 
1886 		/*
1887 		 * Global IO size is always a minimum.
1888 		 * If while a reconnection server sends us a value a bit
1889 		 * higher - client does not care and uses cached minimum.
1890 		 *
1891 		 * Since we can have several sessions (paths) restablishing
1892 		 * connections in parallel, use lock.
1893 		 */
1894 		mutex_lock(&clt->paths_mutex);
1895 		clt->queue_depth = clt_path->queue_depth;
1896 		clt->max_io_size = min_not_zero(clt_path->max_io_size,
1897 						clt->max_io_size);
1898 		mutex_unlock(&clt->paths_mutex);
1899 
1900 		/*
1901 		 * Cache the hca_port and hca_name for sysfs
1902 		 */
1903 		clt_path->hca_port = con->c.cm_id->port_num;
1904 		scnprintf(clt_path->hca_name, sizeof(clt_path->hca_name),
1905 			  clt_path->s.dev->ib_dev->name);
1906 		clt_path->s.src_addr = con->c.cm_id->route.addr.src_addr;
1907 		/* set for_new_clt, to allow future reconnect on any path */
1908 		clt_path->for_new_clt = 1;
1909 	}
1910 
1911 	return 0;
1912 }
1913 
1914 static inline void flag_success_on_conn(struct rtrs_clt_con *con)
1915 {
1916 	struct rtrs_clt_path *clt_path = to_clt_path(con->c.path);
1917 
1918 	atomic_inc(&clt_path->connected_cnt);
1919 	con->cm_err = 1;
1920 }
1921 
1922 static int rtrs_rdma_conn_rejected(struct rtrs_clt_con *con,
1923 				    struct rdma_cm_event *ev)
1924 {
1925 	struct rtrs_path *s = con->c.path;
1926 	const struct rtrs_msg_conn_rsp *msg;
1927 	const char *rej_msg;
1928 	int status, errno;
1929 	u8 data_len;
1930 
1931 	status = ev->status;
1932 	rej_msg = rdma_reject_msg(con->c.cm_id, status);
1933 	msg = rdma_consumer_reject_data(con->c.cm_id, ev, &data_len);
1934 
1935 	if (msg && data_len >= sizeof(*msg)) {
1936 		errno = (int16_t)le16_to_cpu(msg->errno);
1937 		if (errno == -EBUSY)
1938 			rtrs_err(s,
1939 				  "Previous session is still exists on the server, please reconnect later\n");
1940 		else
1941 			rtrs_err(s,
1942 				  "Connect rejected: status %d (%s), rtrs errno %d\n",
1943 				  status, rej_msg, errno);
1944 	} else {
1945 		rtrs_err(s,
1946 			  "Connect rejected but with malformed message: status %d (%s)\n",
1947 			  status, rej_msg);
1948 	}
1949 
1950 	return -ECONNRESET;
1951 }
1952 
1953 void rtrs_clt_close_conns(struct rtrs_clt_path *clt_path, bool wait)
1954 {
1955 	trace_rtrs_clt_close_conns(clt_path);
1956 
1957 	if (rtrs_clt_change_state_get_old(clt_path, RTRS_CLT_CLOSING, NULL))
1958 		queue_work(rtrs_wq, &clt_path->close_work);
1959 	if (wait)
1960 		flush_work(&clt_path->close_work);
1961 }
1962 
1963 static inline void flag_error_on_conn(struct rtrs_clt_con *con, int cm_err)
1964 {
1965 	if (con->cm_err == 1) {
1966 		struct rtrs_clt_path *clt_path;
1967 
1968 		clt_path = to_clt_path(con->c.path);
1969 		if (atomic_dec_and_test(&clt_path->connected_cnt))
1970 
1971 			wake_up(&clt_path->state_wq);
1972 	}
1973 	con->cm_err = cm_err;
1974 }
1975 
1976 static int rtrs_clt_rdma_cm_handler(struct rdma_cm_id *cm_id,
1977 				     struct rdma_cm_event *ev)
1978 {
1979 	struct rtrs_clt_con *con = cm_id->context;
1980 	struct rtrs_path *s = con->c.path;
1981 	struct rtrs_clt_path *clt_path = to_clt_path(s);
1982 	int cm_err = 0;
1983 
1984 	switch (ev->event) {
1985 	case RDMA_CM_EVENT_ADDR_RESOLVED:
1986 		cm_err = rtrs_rdma_addr_resolved(con);
1987 		break;
1988 	case RDMA_CM_EVENT_ROUTE_RESOLVED:
1989 		cm_err = rtrs_rdma_route_resolved(con);
1990 		break;
1991 	case RDMA_CM_EVENT_ESTABLISHED:
1992 		cm_err = rtrs_rdma_conn_established(con, ev);
1993 		if (!cm_err) {
1994 			/*
1995 			 * Report success and wake up. Here we abuse state_wq,
1996 			 * i.e. wake up without state change, but we set cm_err.
1997 			 */
1998 			flag_success_on_conn(con);
1999 			wake_up(&clt_path->state_wq);
2000 			return 0;
2001 		}
2002 		break;
2003 	case RDMA_CM_EVENT_REJECTED:
2004 		cm_err = rtrs_rdma_conn_rejected(con, ev);
2005 		break;
2006 	case RDMA_CM_EVENT_DISCONNECTED:
2007 		/* No message for disconnecting */
2008 		cm_err = -ECONNRESET;
2009 		break;
2010 	case RDMA_CM_EVENT_CONNECT_ERROR:
2011 	case RDMA_CM_EVENT_UNREACHABLE:
2012 	case RDMA_CM_EVENT_ADDR_CHANGE:
2013 	case RDMA_CM_EVENT_TIMEWAIT_EXIT:
2014 		rtrs_wrn(s, "CM error (CM event: %s, err: %d)\n",
2015 			 rdma_event_msg(ev->event), ev->status);
2016 		cm_err = -ECONNRESET;
2017 		break;
2018 	case RDMA_CM_EVENT_ADDR_ERROR:
2019 	case RDMA_CM_EVENT_ROUTE_ERROR:
2020 		rtrs_wrn(s, "CM error (CM event: %s, err: %d)\n",
2021 			 rdma_event_msg(ev->event), ev->status);
2022 		cm_err = -EHOSTUNREACH;
2023 		break;
2024 	case RDMA_CM_EVENT_DEVICE_REMOVAL:
2025 		/*
2026 		 * Device removal is a special case.  Queue close and return 0.
2027 		 */
2028 		rtrs_wrn_rl(s, "CM event: %s, status: %d\n", rdma_event_msg(ev->event),
2029 			    ev->status);
2030 		rtrs_clt_close_conns(clt_path, false);
2031 		return 0;
2032 	default:
2033 		rtrs_err(s, "Unexpected RDMA CM error (CM event: %s, err: %d)\n",
2034 			 rdma_event_msg(ev->event), ev->status);
2035 		cm_err = -ECONNRESET;
2036 		break;
2037 	}
2038 
2039 	if (cm_err) {
2040 		/*
2041 		 * cm error makes sense only on connection establishing,
2042 		 * in other cases we rely on normal procedure of reconnecting.
2043 		 */
2044 		flag_error_on_conn(con, cm_err);
2045 		rtrs_rdma_error_recovery(con);
2046 	}
2047 
2048 	return 0;
2049 }
2050 
2051 /* The caller should do the cleanup in case of error */
2052 static int create_cm(struct rtrs_clt_con *con)
2053 {
2054 	struct rtrs_path *s = con->c.path;
2055 	struct rtrs_clt_path *clt_path = to_clt_path(s);
2056 	struct rdma_cm_id *cm_id;
2057 	int err;
2058 
2059 	cm_id = rdma_create_id(&init_net, rtrs_clt_rdma_cm_handler, con,
2060 			       clt_path->s.dst_addr.ss_family == AF_IB ?
2061 			       RDMA_PS_IB : RDMA_PS_TCP, IB_QPT_RC);
2062 	if (IS_ERR(cm_id)) {
2063 		rtrs_err(s, "Failed to create CM ID, err: %pe\n", cm_id);
2064 		return PTR_ERR(cm_id);
2065 	}
2066 	con->c.cm_id = cm_id;
2067 	con->cm_err = 0;
2068 	/* allow the port to be reused */
2069 	err = rdma_set_reuseaddr(cm_id, 1);
2070 	if (err != 0) {
2071 		rtrs_err(s, "Set address reuse failed, err: %d\n", err);
2072 		return err;
2073 	}
2074 	err = rdma_resolve_addr(cm_id, (struct sockaddr *)&clt_path->s.src_addr,
2075 				(struct sockaddr *)&clt_path->s.dst_addr,
2076 				RTRS_CONNECT_TIMEOUT_MS);
2077 	if (err) {
2078 		rtrs_err(s, "Failed to resolve address, err: %d\n", err);
2079 		return err;
2080 	}
2081 	/*
2082 	 * Combine connection status and session events. This is needed
2083 	 * for waiting two possible cases: cm_err has something meaningful
2084 	 * or session state was really changed to error by device removal.
2085 	 */
2086 	err = wait_event_interruptible_timeout(
2087 			clt_path->state_wq,
2088 			con->cm_err || clt_path->state != RTRS_CLT_CONNECTING,
2089 			msecs_to_jiffies(RTRS_CONNECT_TIMEOUT_MS));
2090 	if (err == 0 || err == -ERESTARTSYS) {
2091 		if (err == 0)
2092 			err = -ETIMEDOUT;
2093 		/* Timedout or interrupted */
2094 		return err;
2095 	}
2096 	if (con->cm_err < 0)
2097 		return con->cm_err;
2098 	if (READ_ONCE(clt_path->state) != RTRS_CLT_CONNECTING)
2099 		/* Device removal */
2100 		return -ECONNABORTED;
2101 
2102 	return 0;
2103 }
2104 
2105 static void rtrs_clt_path_up(struct rtrs_clt_path *clt_path)
2106 {
2107 	struct rtrs_clt_sess *clt = clt_path->clt;
2108 	int up;
2109 
2110 	/*
2111 	 * We can fire RECONNECTED event only when all paths were
2112 	 * connected on rtrs_clt_open(), then each was disconnected
2113 	 * and the first one connected again.  That's why this nasty
2114 	 * game with counter value.
2115 	 */
2116 
2117 	mutex_lock(&clt->paths_ev_mutex);
2118 	up = ++clt->paths_up;
2119 	/*
2120 	 * Here it is safe to access paths num directly since up counter
2121 	 * is greater than MAX_PATHS_NUM only while rtrs_clt_open() is
2122 	 * in progress, thus paths removals are impossible.
2123 	 */
2124 	if (up > MAX_PATHS_NUM && up == MAX_PATHS_NUM + clt->paths_num)
2125 		clt->paths_up = clt->paths_num;
2126 	else if (up == 1)
2127 		clt->link_ev(clt->priv, RTRS_CLT_LINK_EV_RECONNECTED);
2128 	mutex_unlock(&clt->paths_ev_mutex);
2129 
2130 	/* Mark session as established */
2131 	clt_path->established = true;
2132 	clt_path->reconnect_attempts = 0;
2133 	clt_path->stats->reconnects.successful_cnt++;
2134 }
2135 
2136 static void rtrs_clt_path_down(struct rtrs_clt_path *clt_path)
2137 {
2138 	struct rtrs_clt_sess *clt = clt_path->clt;
2139 
2140 	if (!clt_path->established)
2141 		return;
2142 
2143 	clt_path->established = false;
2144 	mutex_lock(&clt->paths_ev_mutex);
2145 	WARN_ON(!clt->paths_up);
2146 	if (--clt->paths_up == 0)
2147 		clt->link_ev(clt->priv, RTRS_CLT_LINK_EV_DISCONNECTED);
2148 	mutex_unlock(&clt->paths_ev_mutex);
2149 }
2150 
2151 static void rtrs_clt_stop_and_destroy_conns(struct rtrs_clt_path *clt_path)
2152 {
2153 	struct rtrs_clt_con *con;
2154 	unsigned int cid;
2155 
2156 	WARN_ON(READ_ONCE(clt_path->state) == RTRS_CLT_CONNECTED);
2157 
2158 	/*
2159 	 * Possible race with rtrs_clt_open(), when DEVICE_REMOVAL comes
2160 	 * exactly in between.  Start destroying after it finishes.
2161 	 */
2162 	mutex_lock(&clt_path->init_mutex);
2163 	mutex_unlock(&clt_path->init_mutex);
2164 
2165 	/*
2166 	 * All IO paths must observe !CONNECTED state before we
2167 	 * free everything.
2168 	 */
2169 	synchronize_rcu();
2170 
2171 	rtrs_stop_hb(&clt_path->s);
2172 
2173 	/*
2174 	 * The order it utterly crucial: firstly disconnect and complete all
2175 	 * rdma requests with error (thus set in_use=false for requests),
2176 	 * then fail outstanding requests checking in_use for each, and
2177 	 * eventually notify upper layer about session disconnection.
2178 	 */
2179 
2180 	for (cid = 0; cid < clt_path->s.con_num; cid++) {
2181 		if (!clt_path->s.con[cid])
2182 			break;
2183 		con = to_clt_con(clt_path->s.con[cid]);
2184 		stop_cm(con);
2185 	}
2186 	fail_all_outstanding_reqs(clt_path);
2187 	free_path_reqs(clt_path);
2188 	rtrs_clt_path_down(clt_path);
2189 
2190 	/*
2191 	 * Wait for graceful shutdown, namely when peer side invokes
2192 	 * rdma_disconnect(). 'connected_cnt' is decremented only on
2193 	 * CM events, thus if other side had crashed and hb has detected
2194 	 * something is wrong, here we will stuck for exactly timeout ms,
2195 	 * since CM does not fire anything.  That is fine, we are not in
2196 	 * hurry.
2197 	 */
2198 	wait_event_timeout(clt_path->state_wq,
2199 			   !atomic_read(&clt_path->connected_cnt),
2200 			   msecs_to_jiffies(RTRS_CONNECT_TIMEOUT_MS));
2201 
2202 	for (cid = 0; cid < clt_path->s.con_num; cid++) {
2203 		if (!clt_path->s.con[cid])
2204 			break;
2205 		con = to_clt_con(clt_path->s.con[cid]);
2206 		mutex_lock(&con->con_mutex);
2207 		destroy_con_cq_qp(con);
2208 		mutex_unlock(&con->con_mutex);
2209 		destroy_cm(con);
2210 		destroy_con(con);
2211 	}
2212 }
2213 
2214 static void rtrs_clt_remove_path_from_arr(struct rtrs_clt_path *clt_path)
2215 {
2216 	struct rtrs_clt_sess *clt = clt_path->clt;
2217 	struct rtrs_clt_path *next;
2218 	bool wait_for_grace = false;
2219 	int cpu;
2220 
2221 	mutex_lock(&clt->paths_mutex);
2222 	list_del_rcu(&clt_path->s.entry);
2223 
2224 	/* Make sure everybody observes path removal. */
2225 	synchronize_rcu();
2226 
2227 	/*
2228 	 * At this point nobody sees @sess in the list, but still we have
2229 	 * dangling pointer @pcpu_path which _can_ point to @sess.  Since
2230 	 * nobody can observe @sess in the list, we guarantee that IO path
2231 	 * will not assign @sess to @pcpu_path, i.e. @pcpu_path can be equal
2232 	 * to @sess, but can never again become @sess.
2233 	 */
2234 
2235 	/*
2236 	 * Decrement paths number only after grace period, because
2237 	 * caller of do_each_path() must firstly observe list without
2238 	 * path and only then decremented paths number.
2239 	 *
2240 	 * Otherwise there can be the following situation:
2241 	 *    o Two paths exist and IO is coming.
2242 	 *    o One path is removed:
2243 	 *      CPU#0                          CPU#1
2244 	 *      do_each_path():                rtrs_clt_remove_path_from_arr():
2245 	 *          path = get_next_path()
2246 	 *          ^^^                            list_del_rcu(path)
2247 	 *          [!CONNECTED path]              clt->paths_num--
2248 	 *                                              ^^^^^^^^^
2249 	 *          load clt->paths_num                 from 2 to 1
2250 	 *                    ^^^^^^^^^
2251 	 *                    sees 1
2252 	 *
2253 	 *      path is observed as !CONNECTED, but do_each_path() loop
2254 	 *      ends, because expression i < clt->paths_num is false.
2255 	 */
2256 	clt->paths_num--;
2257 
2258 	/*
2259 	 * Get @next connection from current @sess which is going to be
2260 	 * removed.  If @sess is the last element, then @next is NULL.
2261 	 */
2262 	rcu_read_lock();
2263 	next = rtrs_clt_get_next_path_or_null(&clt->paths_list, clt_path);
2264 	rcu_read_unlock();
2265 
2266 	/*
2267 	 * @pcpu paths can still point to the path which is going to be
2268 	 * removed, so change the pointer manually.
2269 	 */
2270 	for_each_possible_cpu(cpu) {
2271 		struct rtrs_clt_path __rcu **ppcpu_path;
2272 
2273 		ppcpu_path = per_cpu_ptr(clt->pcpu_path, cpu);
2274 		if (rcu_dereference_protected(*ppcpu_path,
2275 			lockdep_is_held(&clt->paths_mutex)) != clt_path)
2276 			/*
2277 			 * synchronize_rcu() was called just after deleting
2278 			 * entry from the list, thus IO code path cannot
2279 			 * change pointer back to the pointer which is going
2280 			 * to be removed, we are safe here.
2281 			 */
2282 			continue;
2283 
2284 		/*
2285 		 * We race with IO code path, which also changes pointer,
2286 		 * thus we have to be careful not to overwrite it.
2287 		 */
2288 		if (try_cmpxchg((struct rtrs_clt_path **)ppcpu_path, &clt_path,
2289 				next))
2290 			/*
2291 			 * @ppcpu_path was successfully replaced with @next,
2292 			 * that means that someone could also pick up the
2293 			 * @sess and dereferencing it right now, so wait for
2294 			 * a grace period is required.
2295 			 */
2296 			wait_for_grace = true;
2297 	}
2298 	if (wait_for_grace)
2299 		synchronize_rcu();
2300 
2301 	mutex_unlock(&clt->paths_mutex);
2302 }
2303 
2304 static void rtrs_clt_add_path_to_arr(struct rtrs_clt_path *clt_path)
2305 {
2306 	struct rtrs_clt_sess *clt = clt_path->clt;
2307 
2308 	mutex_lock(&clt->paths_mutex);
2309 	clt->paths_num++;
2310 
2311 	list_add_tail_rcu(&clt_path->s.entry, &clt->paths_list);
2312 	mutex_unlock(&clt->paths_mutex);
2313 }
2314 
2315 static void rtrs_clt_close_work(struct work_struct *work)
2316 {
2317 	struct rtrs_clt_path *clt_path;
2318 
2319 	clt_path = container_of(work, struct rtrs_clt_path, close_work);
2320 
2321 	cancel_work_sync(&clt_path->err_recovery_work);
2322 	cancel_delayed_work_sync(&clt_path->reconnect_dwork);
2323 	rtrs_clt_stop_and_destroy_conns(clt_path);
2324 	rtrs_clt_change_state_get_old(clt_path, RTRS_CLT_CLOSED, NULL);
2325 }
2326 
2327 static int init_conns(struct rtrs_clt_path *clt_path)
2328 {
2329 	unsigned int cid;
2330 	int err, i;
2331 
2332 	/*
2333 	 * On every new session connections increase reconnect counter
2334 	 * to avoid clashes with previous sessions not yet closed
2335 	 * sessions on a server side.
2336 	 */
2337 	clt_path->s.recon_cnt++;
2338 
2339 	/* Establish all RDMA connections  */
2340 	for (cid = 0; cid < clt_path->s.con_num; cid++) {
2341 		err = create_con(clt_path, cid);
2342 		if (err)
2343 			goto destroy;
2344 
2345 		err = create_cm(to_clt_con(clt_path->s.con[cid]));
2346 		if (err)
2347 			goto destroy;
2348 	}
2349 	err = alloc_path_reqs(clt_path);
2350 	if (err)
2351 		goto destroy;
2352 
2353 	return 0;
2354 
2355 destroy:
2356 	/* Make sure we do the cleanup in the order they are created */
2357 	for (i = 0; i <= cid; i++) {
2358 		struct rtrs_clt_con *con;
2359 
2360 		if (!clt_path->s.con[i])
2361 			break;
2362 
2363 		con = to_clt_con(clt_path->s.con[i]);
2364 		if (con->c.cm_id) {
2365 			stop_cm(con);
2366 			mutex_lock(&con->con_mutex);
2367 			destroy_con_cq_qp(con);
2368 			mutex_unlock(&con->con_mutex);
2369 			destroy_cm(con);
2370 		}
2371 		destroy_con(con);
2372 	}
2373 	/*
2374 	 * If we've never taken async path and got an error, say,
2375 	 * doing rdma_resolve_addr(), switch to CONNECTION_ERR state
2376 	 * manually to keep reconnecting.
2377 	 */
2378 	rtrs_clt_change_state_get_old(clt_path, RTRS_CLT_CONNECTING_ERR, NULL);
2379 
2380 	return err;
2381 }
2382 
2383 static void rtrs_clt_info_req_done(struct ib_cq *cq, struct ib_wc *wc)
2384 {
2385 	struct rtrs_clt_con *con = to_clt_con(wc->qp->qp_context);
2386 	struct rtrs_clt_path *clt_path = to_clt_path(con->c.path);
2387 	struct rtrs_iu *iu;
2388 
2389 	iu = container_of(wc->wr_cqe, struct rtrs_iu, cqe);
2390 	rtrs_iu_free(iu, clt_path->s.dev->ib_dev, 1);
2391 
2392 	if (wc->status != IB_WC_SUCCESS) {
2393 		rtrs_err(clt_path->clt, "Path info request send failed: %s\n",
2394 			  ib_wc_status_msg(wc->status));
2395 		rtrs_clt_change_state_get_old(clt_path, RTRS_CLT_CONNECTING_ERR, NULL);
2396 		return;
2397 	}
2398 
2399 	rtrs_clt_update_wc_stats(con);
2400 }
2401 
2402 static int process_info_rsp(struct rtrs_clt_path *clt_path,
2403 			    const struct rtrs_msg_info_rsp *msg)
2404 {
2405 	unsigned int sg_cnt, total_len;
2406 	int i, sgi;
2407 
2408 	sg_cnt = le16_to_cpu(msg->sg_cnt);
2409 	if (!sg_cnt || (clt_path->queue_depth % sg_cnt)) {
2410 		rtrs_err(clt_path->clt,
2411 			  "Incorrect sg_cnt %d, is not multiple\n",
2412 			  sg_cnt);
2413 		return -EINVAL;
2414 	}
2415 
2416 	/*
2417 	 * Check if IB immediate data size is enough to hold the mem_id and
2418 	 * the offset inside the memory chunk.
2419 	 */
2420 	if ((ilog2(sg_cnt - 1) + 1) + (ilog2(clt_path->chunk_size - 1) + 1) >
2421 	    MAX_IMM_PAYL_BITS) {
2422 		rtrs_err(clt_path->clt,
2423 			  "RDMA immediate size (%db) not enough to encode %d buffers of size %dB\n",
2424 			  MAX_IMM_PAYL_BITS, sg_cnt, clt_path->chunk_size);
2425 		return -EINVAL;
2426 	}
2427 	total_len = 0;
2428 	for (sgi = 0, i = 0; sgi < sg_cnt && i < clt_path->queue_depth; sgi++) {
2429 		const struct rtrs_sg_desc *desc = &msg->desc[sgi];
2430 		u32 len, rkey;
2431 		u64 addr;
2432 
2433 		addr = le64_to_cpu(desc->addr);
2434 		rkey = le32_to_cpu(desc->key);
2435 		len  = le32_to_cpu(desc->len);
2436 
2437 		total_len += len;
2438 
2439 		if (!len || (len % clt_path->chunk_size)) {
2440 			rtrs_err(clt_path->clt, "Incorrect [%d].len %d\n",
2441 				  sgi,
2442 				  len);
2443 			return -EINVAL;
2444 		}
2445 		for ( ; len && i < clt_path->queue_depth; i++) {
2446 			clt_path->rbufs[i].addr = addr;
2447 			clt_path->rbufs[i].rkey = rkey;
2448 
2449 			len  -= clt_path->chunk_size;
2450 			addr += clt_path->chunk_size;
2451 		}
2452 	}
2453 	/* Sanity check */
2454 	if (sgi != sg_cnt || i != clt_path->queue_depth) {
2455 		rtrs_err(clt_path->clt,
2456 			 "Incorrect sg vector, not fully mapped\n");
2457 		return -EINVAL;
2458 	}
2459 	if (total_len != clt_path->chunk_size * clt_path->queue_depth) {
2460 		rtrs_err(clt_path->clt, "Incorrect total_len %d\n", total_len);
2461 		return -EINVAL;
2462 	}
2463 
2464 	return 0;
2465 }
2466 
2467 static void rtrs_clt_info_rsp_done(struct ib_cq *cq, struct ib_wc *wc)
2468 {
2469 	struct rtrs_clt_con *con = to_clt_con(wc->qp->qp_context);
2470 	struct rtrs_clt_path *clt_path = to_clt_path(con->c.path);
2471 	struct rtrs_msg_info_rsp *msg;
2472 	enum rtrs_clt_state state;
2473 	struct rtrs_iu *iu;
2474 	size_t rx_sz;
2475 	int err;
2476 
2477 	state = RTRS_CLT_CONNECTING_ERR;
2478 
2479 	WARN_ON(con->c.cid);
2480 	iu = container_of(wc->wr_cqe, struct rtrs_iu, cqe);
2481 	if (wc->status != IB_WC_SUCCESS) {
2482 		rtrs_err(clt_path->clt, "Path info response recv failed: %s\n",
2483 			  ib_wc_status_msg(wc->status));
2484 		goto out;
2485 	}
2486 	WARN_ON(wc->opcode != IB_WC_RECV);
2487 
2488 	if (wc->byte_len < sizeof(*msg)) {
2489 		rtrs_err(clt_path->clt, "Path info response is malformed: size %d\n",
2490 			  wc->byte_len);
2491 		goto out;
2492 	}
2493 	ib_dma_sync_single_for_cpu(clt_path->s.dev->ib_dev, iu->dma_addr,
2494 				   iu->size, DMA_FROM_DEVICE);
2495 	msg = iu->buf;
2496 	if (le16_to_cpu(msg->type) != RTRS_MSG_INFO_RSP) {
2497 		rtrs_err(clt_path->clt, "Path info response is malformed: type %d\n",
2498 			  le16_to_cpu(msg->type));
2499 		goto out;
2500 	}
2501 	rx_sz  = sizeof(*msg);
2502 	rx_sz += sizeof(msg->desc[0]) * le16_to_cpu(msg->sg_cnt);
2503 	if (wc->byte_len < rx_sz) {
2504 		rtrs_err(clt_path->clt, "Path info response is malformed: size %d\n",
2505 			  wc->byte_len);
2506 		goto out;
2507 	}
2508 	err = process_info_rsp(clt_path, msg);
2509 	if (err)
2510 		goto out;
2511 
2512 	err = post_recv_path(clt_path);
2513 	if (err)
2514 		goto out;
2515 
2516 	state = RTRS_CLT_CONNECTED;
2517 
2518 out:
2519 	rtrs_clt_update_wc_stats(con);
2520 	rtrs_iu_free(iu, clt_path->s.dev->ib_dev, 1);
2521 	rtrs_clt_change_state_get_old(clt_path, state, NULL);
2522 }
2523 
2524 static int rtrs_send_path_info(struct rtrs_clt_path *clt_path)
2525 {
2526 	struct rtrs_clt_con *usr_con = to_clt_con(clt_path->s.con[0]);
2527 	struct rtrs_msg_info_req *msg;
2528 	struct rtrs_iu *tx_iu, *rx_iu;
2529 	size_t rx_sz;
2530 	int err;
2531 
2532 	rx_sz  = sizeof(struct rtrs_msg_info_rsp);
2533 	rx_sz += sizeof(struct rtrs_sg_desc) * clt_path->queue_depth;
2534 
2535 	tx_iu = rtrs_iu_alloc(1, sizeof(struct rtrs_msg_info_req), GFP_KERNEL,
2536 			       clt_path->s.dev->ib_dev, DMA_TO_DEVICE,
2537 			       rtrs_clt_info_req_done);
2538 	rx_iu = rtrs_iu_alloc(1, rx_sz, GFP_KERNEL, clt_path->s.dev->ib_dev,
2539 			       DMA_FROM_DEVICE, rtrs_clt_info_rsp_done);
2540 	if (!tx_iu || !rx_iu) {
2541 		err = -ENOMEM;
2542 		goto out;
2543 	}
2544 	/* Prepare for getting info response */
2545 	err = rtrs_iu_post_recv(&usr_con->c, rx_iu);
2546 	if (err) {
2547 		rtrs_err(clt_path->clt, "rtrs_iu_post_recv(), err: %d\n", err);
2548 		goto out;
2549 	}
2550 	rx_iu = NULL;
2551 
2552 	msg = tx_iu->buf;
2553 	msg->type = cpu_to_le16(RTRS_MSG_INFO_REQ);
2554 	memcpy(msg->pathname, clt_path->s.sessname, sizeof(msg->pathname));
2555 
2556 	ib_dma_sync_single_for_device(clt_path->s.dev->ib_dev,
2557 				      tx_iu->dma_addr,
2558 				      tx_iu->size, DMA_TO_DEVICE);
2559 
2560 	/* Send info request */
2561 	err = rtrs_iu_post_send(&usr_con->c, tx_iu, sizeof(*msg), NULL);
2562 	if (err) {
2563 		rtrs_err(clt_path->clt, "rtrs_iu_post_send(), err: %d\n", err);
2564 		goto out;
2565 	}
2566 	tx_iu = NULL;
2567 
2568 	/* Wait for state change */
2569 	wait_event_interruptible_timeout(clt_path->state_wq,
2570 					 clt_path->state != RTRS_CLT_CONNECTING,
2571 					 msecs_to_jiffies(
2572 						 RTRS_CONNECT_TIMEOUT_MS));
2573 	if (READ_ONCE(clt_path->state) != RTRS_CLT_CONNECTED) {
2574 		if (READ_ONCE(clt_path->state) == RTRS_CLT_CONNECTING_ERR)
2575 			err = -ECONNRESET;
2576 		else
2577 			err = -ETIMEDOUT;
2578 	}
2579 
2580 out:
2581 	if (tx_iu)
2582 		rtrs_iu_free(tx_iu, clt_path->s.dev->ib_dev, 1);
2583 	if (rx_iu)
2584 		rtrs_iu_free(rx_iu, clt_path->s.dev->ib_dev, 1);
2585 	if (err)
2586 		/* If we've never taken async path because of malloc problems */
2587 		rtrs_clt_change_state_get_old(clt_path,
2588 					      RTRS_CLT_CONNECTING_ERR, NULL);
2589 
2590 	return err;
2591 }
2592 
2593 /**
2594  * init_path() - establishes all path connections and does handshake
2595  * @clt_path: client path.
2596  * In case of error full close or reconnect procedure should be taken,
2597  * because reconnect or close async works can be started.
2598  */
2599 static int init_path(struct rtrs_clt_path *clt_path)
2600 {
2601 	int err;
2602 	char str[NAME_MAX];
2603 	struct rtrs_addr path = {
2604 		.src = &clt_path->s.src_addr,
2605 		.dst = &clt_path->s.dst_addr,
2606 	};
2607 
2608 	rtrs_addr_to_str(&path, str, sizeof(str));
2609 
2610 	mutex_lock(&clt_path->init_mutex);
2611 	err = init_conns(clt_path);
2612 	if (err) {
2613 		rtrs_err(clt_path->clt,
2614 			 "init_conns() failed: err=%d path=%s [%s:%u]\n", err,
2615 			 str, clt_path->hca_name, clt_path->hca_port);
2616 		goto out;
2617 	}
2618 	err = rtrs_send_path_info(clt_path);
2619 	if (err) {
2620 		rtrs_err(clt_path->clt,
2621 			 "rtrs_send_path_info() failed: err=%d path=%s [%s:%u]\n",
2622 			 err, str, clt_path->hca_name, clt_path->hca_port);
2623 		goto out;
2624 	}
2625 	rtrs_clt_path_up(clt_path);
2626 	rtrs_start_hb(&clt_path->s);
2627 out:
2628 	mutex_unlock(&clt_path->init_mutex);
2629 
2630 	return err;
2631 }
2632 
2633 static void rtrs_clt_reconnect_work(struct work_struct *work)
2634 {
2635 	struct rtrs_clt_path *clt_path;
2636 	struct rtrs_clt_sess *clt;
2637 	int err;
2638 
2639 	clt_path = container_of(to_delayed_work(work), struct rtrs_clt_path,
2640 				reconnect_dwork);
2641 	clt = clt_path->clt;
2642 
2643 	trace_rtrs_clt_reconnect_work(clt_path);
2644 
2645 	if (READ_ONCE(clt_path->state) != RTRS_CLT_RECONNECTING)
2646 		return;
2647 
2648 	if (clt_path->reconnect_attempts >= clt->max_reconnect_attempts) {
2649 		/* Close a path completely if max attempts is reached */
2650 		rtrs_clt_close_conns(clt_path, false);
2651 		return;
2652 	}
2653 	clt_path->reconnect_attempts++;
2654 
2655 	msleep(RTRS_RECONNECT_BACKOFF);
2656 	if (rtrs_clt_change_state_get_old(clt_path, RTRS_CLT_CONNECTING, NULL)) {
2657 		err = init_path(clt_path);
2658 		if (err)
2659 			goto reconnect_again;
2660 	}
2661 
2662 	return;
2663 
2664 reconnect_again:
2665 	if (rtrs_clt_change_state_get_old(clt_path, RTRS_CLT_RECONNECTING, NULL)) {
2666 		clt_path->stats->reconnects.fail_cnt++;
2667 		queue_work(rtrs_wq, &clt_path->err_recovery_work);
2668 	}
2669 }
2670 
2671 static void rtrs_clt_dev_release(struct device *dev)
2672 {
2673 	struct rtrs_clt_sess *clt = container_of(dev, struct rtrs_clt_sess,
2674 						 dev);
2675 
2676 	mutex_destroy(&clt->paths_ev_mutex);
2677 	mutex_destroy(&clt->paths_mutex);
2678 	kfree(clt);
2679 }
2680 
2681 static struct rtrs_clt_sess *alloc_clt(const char *sessname, size_t paths_num,
2682 				  u16 port, size_t pdu_sz, void *priv,
2683 				  void	(*link_ev)(void *priv,
2684 						   enum rtrs_clt_link_ev ev),
2685 				  unsigned int reconnect_delay_sec,
2686 				  unsigned int max_reconnect_attempts)
2687 {
2688 	struct rtrs_clt_sess *clt;
2689 	int err;
2690 
2691 	if (!paths_num || paths_num > MAX_PATHS_NUM)
2692 		return ERR_PTR(-EINVAL);
2693 
2694 	if (strlen(sessname) >= sizeof(clt->sessname))
2695 		return ERR_PTR(-EINVAL);
2696 
2697 	clt = kzalloc(sizeof(*clt), GFP_KERNEL);
2698 	if (!clt)
2699 		return ERR_PTR(-ENOMEM);
2700 
2701 	clt->pcpu_path = alloc_percpu(typeof(*clt->pcpu_path));
2702 	if (!clt->pcpu_path) {
2703 		kfree(clt);
2704 		return ERR_PTR(-ENOMEM);
2705 	}
2706 
2707 	clt->dev.class = &rtrs_clt_dev_class;
2708 	clt->dev.release = rtrs_clt_dev_release;
2709 	uuid_gen(&clt->paths_uuid);
2710 	INIT_LIST_HEAD_RCU(&clt->paths_list);
2711 	clt->paths_num = paths_num;
2712 	clt->paths_up = MAX_PATHS_NUM;
2713 	clt->port = port;
2714 	clt->pdu_sz = pdu_sz;
2715 	clt->max_segments = RTRS_MAX_SEGMENTS;
2716 	clt->reconnect_delay_sec = reconnect_delay_sec;
2717 	clt->max_reconnect_attempts = max_reconnect_attempts;
2718 	clt->priv = priv;
2719 	clt->link_ev = link_ev;
2720 	clt->mp_policy = MP_POLICY_MIN_INFLIGHT;
2721 	strscpy(clt->sessname, sessname, sizeof(clt->sessname));
2722 	init_waitqueue_head(&clt->permits_wait);
2723 	mutex_init(&clt->paths_ev_mutex);
2724 	mutex_init(&clt->paths_mutex);
2725 	device_initialize(&clt->dev);
2726 
2727 	err = dev_set_name(&clt->dev, "%s", sessname);
2728 	if (err)
2729 		goto err_put;
2730 
2731 	/*
2732 	 * Suppress user space notification until
2733 	 * sysfs files are created
2734 	 */
2735 	dev_set_uevent_suppress(&clt->dev, true);
2736 	err = device_add(&clt->dev);
2737 	if (err)
2738 		goto err_put;
2739 
2740 	clt->kobj_paths = kobject_create_and_add("paths", &clt->dev.kobj);
2741 	if (!clt->kobj_paths) {
2742 		err = -ENOMEM;
2743 		goto err_del;
2744 	}
2745 	err = rtrs_clt_create_sysfs_root_files(clt);
2746 	if (err) {
2747 		kobject_del(clt->kobj_paths);
2748 		kobject_put(clt->kobj_paths);
2749 		goto err_del;
2750 	}
2751 	dev_set_uevent_suppress(&clt->dev, false);
2752 	kobject_uevent(&clt->dev.kobj, KOBJ_ADD);
2753 
2754 	return clt;
2755 err_del:
2756 	device_del(&clt->dev);
2757 err_put:
2758 	free_percpu(clt->pcpu_path);
2759 	put_device(&clt->dev);
2760 	return ERR_PTR(err);
2761 }
2762 
2763 static void free_clt(struct rtrs_clt_sess *clt)
2764 {
2765 	free_percpu(clt->pcpu_path);
2766 
2767 	/*
2768 	 * release callback will free clt and destroy mutexes in last put
2769 	 */
2770 	device_unregister(&clt->dev);
2771 }
2772 
2773 /**
2774  * rtrs_clt_open() - Open a path to an RTRS server
2775  * @ops: holds the link event callback and the private pointer.
2776  * @pathname: name of the path to an RTRS server
2777  * @paths: Paths to be established defined by their src and dst addresses
2778  * @paths_num: Number of elements in the @paths array
2779  * @port: port to be used by the RTRS session
2780  * @pdu_sz: Size of extra payload which can be accessed after permit allocation.
2781  * @reconnect_delay_sec: time between reconnect tries
2782  * @max_reconnect_attempts: Number of times to reconnect on error before giving
2783  *			    up, 0 for * disabled, -1 for forever
2784  * @nr_poll_queues: number of polling mode connection using IB_POLL_DIRECT flag
2785  *
2786  * Starts session establishment with the rtrs_server. The function can block
2787  * up to ~2000ms before it returns.
2788  *
2789  * Return a valid pointer on success otherwise PTR_ERR.
2790  */
2791 struct rtrs_clt_sess *rtrs_clt_open(struct rtrs_clt_ops *ops,
2792 				 const char *pathname,
2793 				 const struct rtrs_addr *paths,
2794 				 size_t paths_num, u16 port,
2795 				 size_t pdu_sz, u8 reconnect_delay_sec,
2796 				 s16 max_reconnect_attempts, u32 nr_poll_queues)
2797 {
2798 	struct rtrs_clt_path *clt_path, *tmp;
2799 	struct rtrs_clt_sess *clt;
2800 	int err, i;
2801 
2802 	if (strchr(pathname, '/') || strchr(pathname, '.')) {
2803 		pr_err("pathname cannot contain / and .\n");
2804 		err = -EINVAL;
2805 		goto out;
2806 	}
2807 
2808 	clt = alloc_clt(pathname, paths_num, port, pdu_sz, ops->priv,
2809 			ops->link_ev,
2810 			reconnect_delay_sec,
2811 			max_reconnect_attempts);
2812 	if (IS_ERR(clt)) {
2813 		err = PTR_ERR(clt);
2814 		goto out;
2815 	}
2816 	for (i = 0; i < paths_num; i++) {
2817 		struct rtrs_clt_path *clt_path;
2818 
2819 		clt_path = alloc_path(clt, &paths[i], nr_cpu_ids,
2820 				  nr_poll_queues);
2821 		if (IS_ERR(clt_path)) {
2822 			err = PTR_ERR(clt_path);
2823 			goto close_all_path;
2824 		}
2825 		if (!i)
2826 			clt_path->for_new_clt = 1;
2827 		list_add_tail_rcu(&clt_path->s.entry, &clt->paths_list);
2828 
2829 		err = init_path(clt_path);
2830 		if (err) {
2831 			list_del_rcu(&clt_path->s.entry);
2832 			rtrs_clt_close_conns(clt_path, true);
2833 			free_percpu(clt_path->stats->pcpu_stats);
2834 			kfree(clt_path->stats);
2835 			free_path(clt_path);
2836 			goto close_all_path;
2837 		}
2838 
2839 		err = rtrs_clt_create_path_files(clt_path);
2840 		if (err) {
2841 			list_del_rcu(&clt_path->s.entry);
2842 			rtrs_clt_close_conns(clt_path, true);
2843 			free_percpu(clt_path->stats->pcpu_stats);
2844 			kfree(clt_path->stats);
2845 			free_path(clt_path);
2846 			goto close_all_path;
2847 		}
2848 	}
2849 	err = alloc_permits(clt);
2850 	if (err)
2851 		goto close_all_path;
2852 
2853 	return clt;
2854 
2855 close_all_path:
2856 	list_for_each_entry_safe(clt_path, tmp, &clt->paths_list, s.entry) {
2857 		rtrs_clt_destroy_path_files(clt_path, NULL);
2858 		rtrs_clt_close_conns(clt_path, true);
2859 		kobject_put(&clt_path->kobj);
2860 	}
2861 	rtrs_clt_destroy_sysfs_root(clt);
2862 	free_clt(clt);
2863 
2864 out:
2865 	return ERR_PTR(err);
2866 }
2867 EXPORT_SYMBOL(rtrs_clt_open);
2868 
2869 /**
2870  * rtrs_clt_close() - Close a path
2871  * @clt: Session handle. Session is freed upon return.
2872  */
2873 void rtrs_clt_close(struct rtrs_clt_sess *clt)
2874 {
2875 	struct rtrs_clt_path *clt_path, *tmp;
2876 
2877 	/* Firstly forbid sysfs access */
2878 	rtrs_clt_destroy_sysfs_root(clt);
2879 
2880 	/* Now it is safe to iterate over all paths without locks */
2881 	list_for_each_entry_safe(clt_path, tmp, &clt->paths_list, s.entry) {
2882 		rtrs_clt_close_conns(clt_path, true);
2883 		rtrs_clt_destroy_path_files(clt_path, NULL);
2884 		kobject_put(&clt_path->kobj);
2885 	}
2886 	free_permits(clt);
2887 	free_clt(clt);
2888 }
2889 EXPORT_SYMBOL(rtrs_clt_close);
2890 
2891 int rtrs_clt_reconnect_from_sysfs(struct rtrs_clt_path *clt_path)
2892 {
2893 	enum rtrs_clt_state old_state;
2894 	int err = -EBUSY;
2895 	bool changed;
2896 
2897 	changed = rtrs_clt_change_state_get_old(clt_path,
2898 						 RTRS_CLT_RECONNECTING,
2899 						 &old_state);
2900 	if (changed) {
2901 		clt_path->reconnect_attempts = 0;
2902 		rtrs_clt_stop_and_destroy_conns(clt_path);
2903 		queue_delayed_work(rtrs_wq, &clt_path->reconnect_dwork, 0);
2904 	}
2905 	if (changed || old_state == RTRS_CLT_RECONNECTING) {
2906 		/*
2907 		 * flush_delayed_work() queues pending work for immediate
2908 		 * execution, so do the flush if we have queued something
2909 		 * right now or work is pending.
2910 		 */
2911 		flush_delayed_work(&clt_path->reconnect_dwork);
2912 		err = (READ_ONCE(clt_path->state) ==
2913 		       RTRS_CLT_CONNECTED ? 0 : -ENOTCONN);
2914 	}
2915 
2916 	return err;
2917 }
2918 
2919 int rtrs_clt_remove_path_from_sysfs(struct rtrs_clt_path *clt_path,
2920 				     const struct attribute *sysfs_self)
2921 {
2922 	enum rtrs_clt_state old_state;
2923 	bool changed;
2924 
2925 	/*
2926 	 * Continue stopping path till state was changed to DEAD or
2927 	 * state was observed as DEAD:
2928 	 * 1. State was changed to DEAD - we were fast and nobody
2929 	 *    invoked rtrs_clt_reconnect(), which can again start
2930 	 *    reconnecting.
2931 	 * 2. State was observed as DEAD - we have someone in parallel
2932 	 *    removing the path.
2933 	 */
2934 	do {
2935 		rtrs_clt_close_conns(clt_path, true);
2936 		changed = rtrs_clt_change_state_get_old(clt_path,
2937 							RTRS_CLT_DEAD,
2938 							&old_state);
2939 	} while (!changed && old_state != RTRS_CLT_DEAD);
2940 
2941 	if (changed) {
2942 		rtrs_clt_remove_path_from_arr(clt_path);
2943 		rtrs_clt_destroy_path_files(clt_path, sysfs_self);
2944 		kobject_put(&clt_path->kobj);
2945 	}
2946 
2947 	return 0;
2948 }
2949 
2950 void rtrs_clt_set_max_reconnect_attempts(struct rtrs_clt_sess *clt, int value)
2951 {
2952 	clt->max_reconnect_attempts = (unsigned int)value;
2953 }
2954 
2955 int rtrs_clt_get_max_reconnect_attempts(const struct rtrs_clt_sess *clt)
2956 {
2957 	return (int)clt->max_reconnect_attempts;
2958 }
2959 
2960 /**
2961  * rtrs_clt_request() - Request data transfer to/from server via RDMA.
2962  *
2963  * @dir:	READ/WRITE
2964  * @ops:	callback function to be called as confirmation, and the pointer.
2965  * @clt:	Session
2966  * @permit:	Preallocated permit
2967  * @vec:	Message that is sent to server together with the request.
2968  *		Sum of len of all @vec elements limited to <= IO_MSG_SIZE.
2969  *		Since the msg is copied internally it can be allocated on stack.
2970  * @nr:		Number of elements in @vec.
2971  * @data_len:	length of data sent to/from server
2972  * @sg:		Pages to be sent/received to/from server.
2973  * @sg_cnt:	Number of elements in the @sg
2974  *
2975  * Return:
2976  * 0:		Success
2977  * <0:		Error
2978  *
2979  * On dir=READ rtrs client will request a data transfer from Server to client.
2980  * The data that the server will respond with will be stored in @sg when
2981  * the user receives an %RTRS_CLT_RDMA_EV_RDMA_REQUEST_WRITE_COMPL event.
2982  * On dir=WRITE rtrs client will rdma write data in sg to server side.
2983  */
2984 int rtrs_clt_request(int dir, struct rtrs_clt_req_ops *ops,
2985 		     struct rtrs_clt_sess *clt, struct rtrs_permit *permit,
2986 		     const struct kvec *vec, size_t nr, size_t data_len,
2987 		     struct scatterlist *sg, unsigned int sg_cnt)
2988 {
2989 	struct rtrs_clt_io_req *req;
2990 	struct rtrs_clt_path *clt_path;
2991 
2992 	enum dma_data_direction dma_dir;
2993 	int err = -ECONNABORTED, i;
2994 	size_t usr_len, hdr_len;
2995 	struct path_it it;
2996 
2997 	/* Get kvec length */
2998 	for (i = 0, usr_len = 0; i < nr; i++)
2999 		usr_len += vec[i].iov_len;
3000 
3001 	if (dir == READ) {
3002 		hdr_len = sizeof(struct rtrs_msg_rdma_read) +
3003 			  sg_cnt * sizeof(struct rtrs_sg_desc);
3004 		dma_dir = DMA_FROM_DEVICE;
3005 	} else {
3006 		hdr_len = sizeof(struct rtrs_msg_rdma_write);
3007 		dma_dir = DMA_TO_DEVICE;
3008 	}
3009 
3010 	rcu_read_lock();
3011 	for (path_it_init(&it, clt);
3012 	     (clt_path = it.next_path(&it)) && it.i < it.clt->paths_num; it.i++) {
3013 		if (READ_ONCE(clt_path->state) != RTRS_CLT_CONNECTED)
3014 			continue;
3015 
3016 		if (usr_len + hdr_len > clt_path->max_hdr_size) {
3017 			rtrs_wrn_rl(clt_path->clt,
3018 				     "%s request failed, user message size is %zu and header length %zu, but max size is %u\n",
3019 				     dir == READ ? "Read" : "Write",
3020 				     usr_len, hdr_len, clt_path->max_hdr_size);
3021 			err = -EMSGSIZE;
3022 			break;
3023 		}
3024 		req = rtrs_clt_get_req(clt_path, ops->conf_fn, permit, ops->priv,
3025 				       vec, usr_len, sg, sg_cnt, data_len,
3026 				       dma_dir);
3027 		if (dir == READ)
3028 			err = rtrs_clt_read_req(req);
3029 		else
3030 			err = rtrs_clt_write_req(req);
3031 		if (err) {
3032 			req->in_use = false;
3033 			continue;
3034 		}
3035 		/* Success path */
3036 		break;
3037 	}
3038 	path_it_deinit(&it);
3039 	rcu_read_unlock();
3040 
3041 	return err;
3042 }
3043 EXPORT_SYMBOL(rtrs_clt_request);
3044 
3045 int rtrs_clt_rdma_cq_direct(struct rtrs_clt_sess *clt, unsigned int index)
3046 {
3047 	/* If no path, return -1 for block layer not to try again */
3048 	int cnt = -1;
3049 	struct rtrs_con *con;
3050 	struct rtrs_clt_path *clt_path;
3051 	struct path_it it;
3052 
3053 	rcu_read_lock();
3054 	for (path_it_init(&it, clt);
3055 	     (clt_path = it.next_path(&it)) && it.i < it.clt->paths_num; it.i++) {
3056 		if (READ_ONCE(clt_path->state) != RTRS_CLT_CONNECTED)
3057 			continue;
3058 
3059 		con = clt_path->s.con[index + 1];
3060 		cnt = ib_process_cq_direct(con->cq, -1);
3061 		if (cnt)
3062 			break;
3063 	}
3064 	path_it_deinit(&it);
3065 	rcu_read_unlock();
3066 
3067 	return cnt;
3068 }
3069 EXPORT_SYMBOL(rtrs_clt_rdma_cq_direct);
3070 
3071 /**
3072  * rtrs_clt_query() - queries RTRS session attributes
3073  *@clt: session pointer
3074  *@attr: query results for session attributes.
3075  * Returns:
3076  *    0 on success
3077  *    -ECOMM		no connection to the server
3078  */
3079 int rtrs_clt_query(struct rtrs_clt_sess *clt, struct rtrs_attrs *attr)
3080 {
3081 	if (!rtrs_clt_is_connected(clt))
3082 		return -ECOMM;
3083 
3084 	attr->queue_depth      = clt->queue_depth;
3085 	attr->max_segments     = clt->max_segments;
3086 	/* Cap max_io_size to min of remote buffer size and the fr pages */
3087 	attr->max_io_size = min_t(int, clt->max_io_size,
3088 				  clt->max_segments * SZ_4K);
3089 
3090 	return 0;
3091 }
3092 EXPORT_SYMBOL(rtrs_clt_query);
3093 
3094 int rtrs_clt_create_path_from_sysfs(struct rtrs_clt_sess *clt,
3095 				     struct rtrs_addr *addr)
3096 {
3097 	struct rtrs_clt_path *clt_path;
3098 	int err;
3099 
3100 	clt_path = alloc_path(clt, addr, nr_cpu_ids, 0);
3101 	if (IS_ERR(clt_path))
3102 		return PTR_ERR(clt_path);
3103 
3104 	mutex_lock(&clt->paths_mutex);
3105 	if (clt->paths_num == 0) {
3106 		/*
3107 		 * When all the paths are removed for a session,
3108 		 * the addition of the first path is like a new session for
3109 		 * the storage server
3110 		 */
3111 		clt_path->for_new_clt = 1;
3112 	}
3113 
3114 	mutex_unlock(&clt->paths_mutex);
3115 
3116 	/*
3117 	 * It is totally safe to add path in CONNECTING state: coming
3118 	 * IO will never grab it.  Also it is very important to add
3119 	 * path before init, since init fires LINK_CONNECTED event.
3120 	 */
3121 	rtrs_clt_add_path_to_arr(clt_path);
3122 
3123 	err = init_path(clt_path);
3124 	if (err)
3125 		goto close_path;
3126 
3127 	err = rtrs_clt_create_path_files(clt_path);
3128 	if (err)
3129 		goto close_path;
3130 
3131 	return 0;
3132 
3133 close_path:
3134 	rtrs_clt_remove_path_from_arr(clt_path);
3135 	rtrs_clt_close_conns(clt_path, true);
3136 	free_percpu(clt_path->stats->pcpu_stats);
3137 	kfree(clt_path->stats);
3138 	free_path(clt_path);
3139 
3140 	return err;
3141 }
3142 
3143 static int rtrs_clt_ib_dev_init(struct rtrs_ib_dev *dev)
3144 {
3145 	if (!(dev->ib_dev->attrs.device_cap_flags &
3146 	      IB_DEVICE_MEM_MGT_EXTENSIONS)) {
3147 		pr_err("Memory registrations not supported.\n");
3148 		return -ENOTSUPP;
3149 	}
3150 
3151 	return 0;
3152 }
3153 
3154 static const struct rtrs_rdma_dev_pd_ops dev_pd_ops = {
3155 	.init = rtrs_clt_ib_dev_init
3156 };
3157 
3158 static int __init rtrs_client_init(void)
3159 {
3160 	int ret = 0;
3161 
3162 	rtrs_rdma_dev_pd_init(0, &dev_pd);
3163 	ret = class_register(&rtrs_clt_dev_class);
3164 	if (ret) {
3165 		pr_err("Failed to create rtrs-client dev class\n");
3166 		return ret;
3167 	}
3168 	rtrs_wq = alloc_workqueue("rtrs_client_wq", 0, 0);
3169 	if (!rtrs_wq) {
3170 		class_unregister(&rtrs_clt_dev_class);
3171 		return -ENOMEM;
3172 	}
3173 
3174 	return 0;
3175 }
3176 
3177 static void __exit rtrs_client_exit(void)
3178 {
3179 	destroy_workqueue(rtrs_wq);
3180 	class_unregister(&rtrs_clt_dev_class);
3181 	rtrs_rdma_dev_pd_deinit(&dev_pd);
3182 }
3183 
3184 module_init(rtrs_client_init);
3185 module_exit(rtrs_client_exit);
3186