xref: /linux/drivers/infiniband/ulp/rtrs/rtrs-clt.c (revision c532de5a67a70f8533d495f8f2aaa9a0491c3ad0)
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_rl(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_rl(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->mr->need_inval = 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->mr->need_inval) {
395 			/*
396 			 * We are here to invalidate read/write requests
397 			 * ourselves.  In normal scenario server should
398 			 * send INV for all read requests, we do local
399 			 * invalidate for write requests ourselves, but
400 			 * we are here, thus three things could happen:
401 			 *
402 			 *    1.  this is failover, when errno != 0
403 			 *        and can_wait == 1,
404 			 *
405 			 *    2.  something totally bad happened and
406 			 *        server forgot to send INV, so we
407 			 *        should do that ourselves.
408 			 *
409 			 *    3.  write request finishes, we need to do local
410 			 *        invalidate
411 			 */
412 
413 			if (can_wait) {
414 				req->need_inv_comp = true;
415 			} else {
416 				/* This should be IO path, so always notify */
417 				WARN_ON(!notify);
418 				/* Save errno for INV callback */
419 				req->inv_errno = errno;
420 			}
421 
422 			refcount_inc(&req->ref);
423 			err = rtrs_inv_rkey(req);
424 			if (err) {
425 				rtrs_err_rl(con->c.path, "Send INV WR key=%#x: %d\n",
426 					  req->mr->rkey, err);
427 			} else if (can_wait) {
428 				wait_for_completion(&req->inv_comp);
429 			}
430 			if (!refcount_dec_and_test(&req->ref))
431 				return;
432 		}
433 		ib_dma_unmap_sg(clt_path->s.dev->ib_dev, req->sglist,
434 				req->sg_cnt, req->dir);
435 	}
436 	if (!refcount_dec_and_test(&req->ref))
437 		return;
438 	if (req->mp_policy == MP_POLICY_MIN_INFLIGHT)
439 		atomic_dec(&clt_path->stats->inflight);
440 
441 	req->in_use = false;
442 	req->con = NULL;
443 
444 	if (errno) {
445 		rtrs_err_rl(con->c.path,
446 			    "IO %s request failed: error=%d path=%s [%s:%u] notify=%d\n",
447 			    req->dir == DMA_TO_DEVICE ? "write" : "read", errno,
448 			    kobject_name(&clt_path->kobj), clt_path->hca_name,
449 			    clt_path->hca_port, notify);
450 	}
451 
452 	if (notify)
453 		req->conf(req->priv, errno);
454 }
455 
456 static int rtrs_post_send_rdma(struct rtrs_clt_con *con,
457 				struct rtrs_clt_io_req *req,
458 				struct rtrs_rbuf *rbuf, u32 off,
459 				u32 imm, struct ib_send_wr *wr)
460 {
461 	struct rtrs_clt_path *clt_path = to_clt_path(con->c.path);
462 	enum ib_send_flags flags;
463 	struct ib_sge sge;
464 
465 	if (!req->sg_size) {
466 		rtrs_wrn(con->c.path,
467 			 "Doing RDMA Write failed, no data supplied\n");
468 		return -EINVAL;
469 	}
470 
471 	/* user data and user message in the first list element */
472 	sge.addr   = req->iu->dma_addr;
473 	sge.length = req->sg_size;
474 	sge.lkey   = clt_path->s.dev->ib_pd->local_dma_lkey;
475 
476 	/*
477 	 * From time to time we have to post signalled sends,
478 	 * or send queue will fill up and only QP reset can help.
479 	 */
480 	flags = atomic_inc_return(&con->c.wr_cnt) % clt_path->s.signal_interval ?
481 			0 : IB_SEND_SIGNALED;
482 
483 	ib_dma_sync_single_for_device(clt_path->s.dev->ib_dev,
484 				      req->iu->dma_addr,
485 				      req->sg_size, DMA_TO_DEVICE);
486 
487 	return rtrs_iu_post_rdma_write_imm(&con->c, req->iu, &sge, 1,
488 					    rbuf->rkey, rbuf->addr + off,
489 					    imm, flags, wr, NULL);
490 }
491 
492 static void process_io_rsp(struct rtrs_clt_path *clt_path, u32 msg_id,
493 			   s16 errno, bool w_inval)
494 {
495 	struct rtrs_clt_io_req *req;
496 
497 	if (WARN_ON(msg_id >= clt_path->queue_depth))
498 		return;
499 
500 	req = &clt_path->reqs[msg_id];
501 	/* Drop need_inv if server responded with send with invalidation */
502 	req->mr->need_inval &= !w_inval;
503 	complete_rdma_req(req, errno, true, false);
504 }
505 
506 static void rtrs_clt_recv_done(struct rtrs_clt_con *con, struct ib_wc *wc)
507 {
508 	struct rtrs_iu *iu;
509 	int err;
510 	struct rtrs_clt_path *clt_path = to_clt_path(con->c.path);
511 
512 	WARN_ON((clt_path->flags & RTRS_MSG_NEW_RKEY_F) == 0);
513 	iu = container_of(wc->wr_cqe, struct rtrs_iu,
514 			  cqe);
515 	err = rtrs_iu_post_recv(&con->c, iu);
516 	if (err) {
517 		rtrs_err(con->c.path, "post iu failed %d\n", err);
518 		rtrs_rdma_error_recovery(con);
519 	}
520 }
521 
522 static void rtrs_clt_rkey_rsp_done(struct rtrs_clt_con *con, struct ib_wc *wc)
523 {
524 	struct rtrs_clt_path *clt_path = to_clt_path(con->c.path);
525 	struct rtrs_msg_rkey_rsp *msg;
526 	u32 imm_type, imm_payload;
527 	bool w_inval = false;
528 	struct rtrs_iu *iu;
529 	u32 buf_id;
530 	int err;
531 
532 	WARN_ON((clt_path->flags & RTRS_MSG_NEW_RKEY_F) == 0);
533 
534 	iu = container_of(wc->wr_cqe, struct rtrs_iu, cqe);
535 
536 	if (wc->byte_len < sizeof(*msg)) {
537 		rtrs_err(con->c.path, "rkey response is malformed: size %d\n",
538 			  wc->byte_len);
539 		goto out;
540 	}
541 	ib_dma_sync_single_for_cpu(clt_path->s.dev->ib_dev, iu->dma_addr,
542 				   iu->size, DMA_FROM_DEVICE);
543 	msg = iu->buf;
544 	if (le16_to_cpu(msg->type) != RTRS_MSG_RKEY_RSP) {
545 		rtrs_err(clt_path->clt,
546 			  "rkey response is malformed: type %d\n",
547 			  le16_to_cpu(msg->type));
548 		goto out;
549 	}
550 	buf_id = le16_to_cpu(msg->buf_id);
551 	if (WARN_ON(buf_id >= clt_path->queue_depth))
552 		goto out;
553 
554 	rtrs_from_imm(be32_to_cpu(wc->ex.imm_data), &imm_type, &imm_payload);
555 	if (imm_type == RTRS_IO_RSP_IMM ||
556 	    imm_type == RTRS_IO_RSP_W_INV_IMM) {
557 		u32 msg_id;
558 
559 		w_inval = (imm_type == RTRS_IO_RSP_W_INV_IMM);
560 		rtrs_from_io_rsp_imm(imm_payload, &msg_id, &err);
561 
562 		if (WARN_ON(buf_id != msg_id))
563 			goto out;
564 		clt_path->rbufs[buf_id].rkey = le32_to_cpu(msg->rkey);
565 		process_io_rsp(clt_path, msg_id, err, w_inval);
566 	}
567 	ib_dma_sync_single_for_device(clt_path->s.dev->ib_dev, iu->dma_addr,
568 				      iu->size, DMA_FROM_DEVICE);
569 	return rtrs_clt_recv_done(con, wc);
570 out:
571 	rtrs_rdma_error_recovery(con);
572 }
573 
574 static void rtrs_clt_rdma_done(struct ib_cq *cq, struct ib_wc *wc);
575 
576 static struct ib_cqe io_comp_cqe = {
577 	.done = rtrs_clt_rdma_done
578 };
579 
580 /*
581  * Post x2 empty WRs: first is for this RDMA with IMM,
582  * second is for RECV with INV, which happened earlier.
583  */
584 static int rtrs_post_recv_empty_x2(struct rtrs_con *con, struct ib_cqe *cqe)
585 {
586 	struct ib_recv_wr wr_arr[2], *wr;
587 	int i;
588 
589 	memset(wr_arr, 0, sizeof(wr_arr));
590 	for (i = 0; i < ARRAY_SIZE(wr_arr); i++) {
591 		wr = &wr_arr[i];
592 		wr->wr_cqe  = cqe;
593 		if (i)
594 			/* Chain backwards */
595 			wr->next = &wr_arr[i - 1];
596 	}
597 
598 	return ib_post_recv(con->qp, wr, NULL);
599 }
600 
601 static void rtrs_clt_rdma_done(struct ib_cq *cq, struct ib_wc *wc)
602 {
603 	struct rtrs_clt_con *con = to_clt_con(wc->qp->qp_context);
604 	struct rtrs_clt_path *clt_path = to_clt_path(con->c.path);
605 	u32 imm_type, imm_payload;
606 	bool w_inval = false;
607 	int err;
608 
609 	if (wc->status != IB_WC_SUCCESS) {
610 		if (wc->status != IB_WC_WR_FLUSH_ERR) {
611 			rtrs_err(clt_path->clt, "RDMA failed: %s\n",
612 				  ib_wc_status_msg(wc->status));
613 			rtrs_rdma_error_recovery(con);
614 		}
615 		return;
616 	}
617 	rtrs_clt_update_wc_stats(con);
618 
619 	switch (wc->opcode) {
620 	case IB_WC_RECV_RDMA_WITH_IMM:
621 		/*
622 		 * post_recv() RDMA write completions of IO reqs (read/write)
623 		 * and hb
624 		 */
625 		if (WARN_ON(wc->wr_cqe->done != rtrs_clt_rdma_done))
626 			return;
627 		clt_path->s.hb_missed_cnt = 0;
628 		rtrs_from_imm(be32_to_cpu(wc->ex.imm_data),
629 			       &imm_type, &imm_payload);
630 		if (imm_type == RTRS_IO_RSP_IMM ||
631 		    imm_type == RTRS_IO_RSP_W_INV_IMM) {
632 			u32 msg_id;
633 
634 			w_inval = (imm_type == RTRS_IO_RSP_W_INV_IMM);
635 			rtrs_from_io_rsp_imm(imm_payload, &msg_id, &err);
636 
637 			process_io_rsp(clt_path, msg_id, err, w_inval);
638 		} else if (imm_type == RTRS_HB_MSG_IMM) {
639 			WARN_ON(con->c.cid);
640 			rtrs_send_hb_ack(&clt_path->s);
641 			if (clt_path->flags & RTRS_MSG_NEW_RKEY_F)
642 				return  rtrs_clt_recv_done(con, wc);
643 		} else if (imm_type == RTRS_HB_ACK_IMM) {
644 			WARN_ON(con->c.cid);
645 			clt_path->s.hb_cur_latency =
646 				ktime_sub(ktime_get(), clt_path->s.hb_last_sent);
647 			if (clt_path->flags & RTRS_MSG_NEW_RKEY_F)
648 				return  rtrs_clt_recv_done(con, wc);
649 		} else {
650 			rtrs_wrn(con->c.path, "Unknown IMM type %u\n",
651 				  imm_type);
652 		}
653 		if (w_inval)
654 			/*
655 			 * Post x2 empty WRs: first is for this RDMA with IMM,
656 			 * second is for RECV with INV, which happened earlier.
657 			 */
658 			err = rtrs_post_recv_empty_x2(&con->c, &io_comp_cqe);
659 		else
660 			err = rtrs_post_recv_empty(&con->c, &io_comp_cqe);
661 		if (err) {
662 			rtrs_err(con->c.path, "rtrs_post_recv_empty(): %d\n",
663 				  err);
664 			rtrs_rdma_error_recovery(con);
665 		}
666 		break;
667 	case IB_WC_RECV:
668 		/*
669 		 * Key invalidations from server side
670 		 */
671 		clt_path->s.hb_missed_cnt = 0;
672 		WARN_ON(!(wc->wc_flags & IB_WC_WITH_INVALIDATE ||
673 			  wc->wc_flags & IB_WC_WITH_IMM));
674 		WARN_ON(wc->wr_cqe->done != rtrs_clt_rdma_done);
675 		if (clt_path->flags & RTRS_MSG_NEW_RKEY_F) {
676 			if (wc->wc_flags & IB_WC_WITH_INVALIDATE)
677 				return  rtrs_clt_recv_done(con, wc);
678 
679 			return  rtrs_clt_rkey_rsp_done(con, wc);
680 		}
681 		break;
682 	case IB_WC_RDMA_WRITE:
683 		/*
684 		 * post_send() RDMA write completions of IO reqs (read/write)
685 		 * and hb.
686 		 */
687 		break;
688 
689 	default:
690 		rtrs_wrn(clt_path->clt, "Unexpected WC type: %d\n", wc->opcode);
691 		return;
692 	}
693 }
694 
695 static int post_recv_io(struct rtrs_clt_con *con, size_t q_size)
696 {
697 	int err, i;
698 	struct rtrs_clt_path *clt_path = to_clt_path(con->c.path);
699 
700 	for (i = 0; i < q_size; i++) {
701 		if (clt_path->flags & RTRS_MSG_NEW_RKEY_F) {
702 			struct rtrs_iu *iu = &con->rsp_ius[i];
703 
704 			err = rtrs_iu_post_recv(&con->c, iu);
705 		} else {
706 			err = rtrs_post_recv_empty(&con->c, &io_comp_cqe);
707 		}
708 		if (err)
709 			return err;
710 	}
711 
712 	return 0;
713 }
714 
715 static int post_recv_path(struct rtrs_clt_path *clt_path)
716 {
717 	size_t q_size = 0;
718 	int err, cid;
719 
720 	for (cid = 0; cid < clt_path->s.con_num; cid++) {
721 		if (cid == 0)
722 			q_size = SERVICE_CON_QUEUE_DEPTH;
723 		else
724 			q_size = clt_path->queue_depth;
725 
726 		/*
727 		 * x2 for RDMA read responses + FR key invalidations,
728 		 * RDMA writes do not require any FR registrations.
729 		 */
730 		q_size *= 2;
731 
732 		err = post_recv_io(to_clt_con(clt_path->s.con[cid]), q_size);
733 		if (err) {
734 			rtrs_err(clt_path->clt, "post_recv_io(), err: %d\n",
735 				 err);
736 			return err;
737 		}
738 	}
739 
740 	return 0;
741 }
742 
743 struct path_it {
744 	int i;
745 	struct list_head skip_list;
746 	struct rtrs_clt_sess *clt;
747 	struct rtrs_clt_path *(*next_path)(struct path_it *it);
748 };
749 
750 /*
751  * rtrs_clt_get_next_path_or_null - get clt path from the list or return NULL
752  * @head:	the head for the list.
753  * @clt_path:	The element to take the next clt_path from.
754  *
755  * Next clt path returned in round-robin fashion, i.e. head will be skipped,
756  * but if list is observed as empty, NULL will be returned.
757  *
758  * This function may safely run concurrently with the _rcu list-mutation
759  * primitives such as list_add_rcu() as long as it's guarded by rcu_read_lock().
760  */
761 static inline struct rtrs_clt_path *
762 rtrs_clt_get_next_path_or_null(struct list_head *head, struct rtrs_clt_path *clt_path)
763 {
764 	return list_next_or_null_rcu(head, &clt_path->s.entry, typeof(*clt_path), s.entry) ?:
765 				     list_next_or_null_rcu(head,
766 							   READ_ONCE((&clt_path->s.entry)->next),
767 							   typeof(*clt_path), s.entry);
768 }
769 
770 /**
771  * get_next_path_rr() - Returns path in round-robin fashion.
772  * @it:	the path pointer
773  *
774  * Related to @MP_POLICY_RR
775  *
776  * Locks:
777  *    rcu_read_lock() must be held.
778  */
779 static struct rtrs_clt_path *get_next_path_rr(struct path_it *it)
780 {
781 	struct rtrs_clt_path __rcu **ppcpu_path;
782 	struct rtrs_clt_path *path;
783 	struct rtrs_clt_sess *clt;
784 
785 	/*
786 	 * Assert that rcu lock must be held
787 	 */
788 	RCU_LOCKDEP_WARN(!rcu_read_lock_held(), "no rcu read lock held");
789 
790 	clt = it->clt;
791 
792 	/*
793 	 * Here we use two RCU objects: @paths_list and @pcpu_path
794 	 * pointer.  See rtrs_clt_remove_path_from_arr() for details
795 	 * how that is handled.
796 	 */
797 
798 	ppcpu_path = this_cpu_ptr(clt->pcpu_path);
799 	path = rcu_dereference(*ppcpu_path);
800 	if (!path)
801 		path = list_first_or_null_rcu(&clt->paths_list,
802 					      typeof(*path), s.entry);
803 	else
804 		path = rtrs_clt_get_next_path_or_null(&clt->paths_list, path);
805 
806 	rcu_assign_pointer(*ppcpu_path, path);
807 
808 	return path;
809 }
810 
811 /**
812  * get_next_path_min_inflight() - Returns path with minimal inflight count.
813  * @it:	the path pointer
814  *
815  * Related to @MP_POLICY_MIN_INFLIGHT
816  *
817  * Locks:
818  *    rcu_read_lock() must be hold.
819  */
820 static struct rtrs_clt_path *get_next_path_min_inflight(struct path_it *it)
821 {
822 	struct rtrs_clt_path *min_path = NULL;
823 	struct rtrs_clt_sess *clt = it->clt;
824 	struct rtrs_clt_path *clt_path;
825 	int min_inflight = INT_MAX;
826 	int inflight;
827 
828 	list_for_each_entry_rcu(clt_path, &clt->paths_list, s.entry) {
829 		if (READ_ONCE(clt_path->state) != RTRS_CLT_CONNECTED)
830 			continue;
831 
832 		if (!list_empty(raw_cpu_ptr(clt_path->mp_skip_entry)))
833 			continue;
834 
835 		inflight = atomic_read(&clt_path->stats->inflight);
836 
837 		if (inflight < min_inflight) {
838 			min_inflight = inflight;
839 			min_path = clt_path;
840 		}
841 	}
842 
843 	/*
844 	 * add the path to the skip list, so that next time we can get
845 	 * a different one
846 	 */
847 	if (min_path)
848 		list_add(raw_cpu_ptr(min_path->mp_skip_entry), &it->skip_list);
849 
850 	return min_path;
851 }
852 
853 /**
854  * get_next_path_min_latency() - Returns path with minimal latency.
855  * @it:	the path pointer
856  *
857  * Return: a path with the lowest latency or NULL if all paths are tried
858  *
859  * Locks:
860  *    rcu_read_lock() must be hold.
861  *
862  * Related to @MP_POLICY_MIN_LATENCY
863  *
864  * This DOES skip an already-tried path.
865  * There is a skip-list to skip a path if the path has tried but failed.
866  * It will try the minimum latency path and then the second minimum latency
867  * path and so on. Finally it will return NULL if all paths are tried.
868  * Therefore the caller MUST check the returned
869  * path is NULL and trigger the IO error.
870  */
871 static struct rtrs_clt_path *get_next_path_min_latency(struct path_it *it)
872 {
873 	struct rtrs_clt_path *min_path = NULL;
874 	struct rtrs_clt_sess *clt = it->clt;
875 	struct rtrs_clt_path *clt_path;
876 	ktime_t min_latency = KTIME_MAX;
877 	ktime_t latency;
878 
879 	list_for_each_entry_rcu(clt_path, &clt->paths_list, s.entry) {
880 		if (READ_ONCE(clt_path->state) != RTRS_CLT_CONNECTED)
881 			continue;
882 
883 		if (!list_empty(raw_cpu_ptr(clt_path->mp_skip_entry)))
884 			continue;
885 
886 		latency = clt_path->s.hb_cur_latency;
887 
888 		if (latency < min_latency) {
889 			min_latency = latency;
890 			min_path = clt_path;
891 		}
892 	}
893 
894 	/*
895 	 * add the path to the skip list, so that next time we can get
896 	 * a different one
897 	 */
898 	if (min_path)
899 		list_add(raw_cpu_ptr(min_path->mp_skip_entry), &it->skip_list);
900 
901 	return min_path;
902 }
903 
904 static inline void path_it_init(struct path_it *it, struct rtrs_clt_sess *clt)
905 {
906 	INIT_LIST_HEAD(&it->skip_list);
907 	it->clt = clt;
908 	it->i = 0;
909 
910 	if (clt->mp_policy == MP_POLICY_RR)
911 		it->next_path = get_next_path_rr;
912 	else if (clt->mp_policy == MP_POLICY_MIN_INFLIGHT)
913 		it->next_path = get_next_path_min_inflight;
914 	else
915 		it->next_path = get_next_path_min_latency;
916 }
917 
918 static inline void path_it_deinit(struct path_it *it)
919 {
920 	struct list_head *skip, *tmp;
921 	/*
922 	 * The skip_list is used only for the MIN_INFLIGHT and MIN_LATENCY policies.
923 	 * We need to remove paths from it, so that next IO can insert
924 	 * paths (->mp_skip_entry) into a skip_list again.
925 	 */
926 	list_for_each_safe(skip, tmp, &it->skip_list)
927 		list_del_init(skip);
928 }
929 
930 /**
931  * rtrs_clt_init_req() - Initialize an rtrs_clt_io_req holding information
932  * about an inflight IO.
933  * The user buffer holding user control message (not data) is copied into
934  * the corresponding buffer of rtrs_iu (req->iu->buf), which later on will
935  * also hold the control message of rtrs.
936  * @req: an io request holding information about IO.
937  * @clt_path: client path
938  * @conf: conformation callback function to notify upper layer.
939  * @permit: permit for allocation of RDMA remote buffer
940  * @priv: private pointer
941  * @vec: kernel vector containing control message
942  * @usr_len: length of the user message
943  * @sg: scater list for IO data
944  * @sg_cnt: number of scater list entries
945  * @data_len: length of the IO data
946  * @dir: direction of the IO.
947  */
948 static void rtrs_clt_init_req(struct rtrs_clt_io_req *req,
949 			      struct rtrs_clt_path *clt_path,
950 			      void (*conf)(void *priv, int errno),
951 			      struct rtrs_permit *permit, void *priv,
952 			      const struct kvec *vec, size_t usr_len,
953 			      struct scatterlist *sg, size_t sg_cnt,
954 			      size_t data_len, int dir)
955 {
956 	struct iov_iter iter;
957 	size_t len;
958 
959 	req->permit = permit;
960 	req->in_use = true;
961 	req->usr_len = usr_len;
962 	req->data_len = data_len;
963 	req->sglist = sg;
964 	req->sg_cnt = sg_cnt;
965 	req->priv = priv;
966 	req->dir = dir;
967 	req->con = rtrs_permit_to_clt_con(clt_path, permit);
968 	req->conf = conf;
969 	req->mr->need_inval = false;
970 	req->need_inv_comp = false;
971 	req->inv_errno = 0;
972 	refcount_set(&req->ref, 1);
973 	req->mp_policy = clt_path->clt->mp_policy;
974 
975 	iov_iter_kvec(&iter, ITER_SOURCE, vec, 1, usr_len);
976 	len = _copy_from_iter(req->iu->buf, usr_len, &iter);
977 	WARN_ON(len != usr_len);
978 
979 	reinit_completion(&req->inv_comp);
980 }
981 
982 static struct rtrs_clt_io_req *
983 rtrs_clt_get_req(struct rtrs_clt_path *clt_path,
984 		 void (*conf)(void *priv, int errno),
985 		 struct rtrs_permit *permit, void *priv,
986 		 const struct kvec *vec, size_t usr_len,
987 		 struct scatterlist *sg, size_t sg_cnt,
988 		 size_t data_len, int dir)
989 {
990 	struct rtrs_clt_io_req *req;
991 
992 	req = &clt_path->reqs[permit->mem_id];
993 	rtrs_clt_init_req(req, clt_path, conf, permit, priv, vec, usr_len,
994 			   sg, sg_cnt, data_len, dir);
995 	return req;
996 }
997 
998 static struct rtrs_clt_io_req *
999 rtrs_clt_get_copy_req(struct rtrs_clt_path *alive_path,
1000 		       struct rtrs_clt_io_req *fail_req)
1001 {
1002 	struct rtrs_clt_io_req *req;
1003 	struct kvec vec = {
1004 		.iov_base = fail_req->iu->buf,
1005 		.iov_len  = fail_req->usr_len
1006 	};
1007 
1008 	req = &alive_path->reqs[fail_req->permit->mem_id];
1009 	rtrs_clt_init_req(req, alive_path, fail_req->conf, fail_req->permit,
1010 			   fail_req->priv, &vec, fail_req->usr_len,
1011 			   fail_req->sglist, fail_req->sg_cnt,
1012 			   fail_req->data_len, fail_req->dir);
1013 	return req;
1014 }
1015 
1016 static int rtrs_post_rdma_write_sg(struct rtrs_clt_con *con,
1017 				   struct rtrs_clt_io_req *req,
1018 				   struct rtrs_rbuf *rbuf, bool fr_en,
1019 				   u32 count, u32 size, u32 imm,
1020 				   struct ib_send_wr *wr,
1021 				   struct ib_send_wr *tail)
1022 {
1023 	struct rtrs_clt_path *clt_path = to_clt_path(con->c.path);
1024 	struct ib_sge *sge = req->sge;
1025 	enum ib_send_flags flags;
1026 	struct scatterlist *sg;
1027 	size_t num_sge;
1028 	int i;
1029 	struct ib_send_wr *ptail = NULL;
1030 
1031 	if (fr_en) {
1032 		i = 0;
1033 		sge[i].addr   = req->mr->iova;
1034 		sge[i].length = req->mr->length;
1035 		sge[i].lkey   = req->mr->lkey;
1036 		i++;
1037 		num_sge = 2;
1038 		ptail = tail;
1039 	} else {
1040 		for_each_sg(req->sglist, sg, count, i) {
1041 			sge[i].addr   = sg_dma_address(sg);
1042 			sge[i].length = sg_dma_len(sg);
1043 			sge[i].lkey   = clt_path->s.dev->ib_pd->local_dma_lkey;
1044 		}
1045 		num_sge = 1 + count;
1046 	}
1047 	sge[i].addr   = req->iu->dma_addr;
1048 	sge[i].length = size;
1049 	sge[i].lkey   = clt_path->s.dev->ib_pd->local_dma_lkey;
1050 
1051 	/*
1052 	 * From time to time we have to post signalled sends,
1053 	 * or send queue will fill up and only QP reset can help.
1054 	 */
1055 	flags = atomic_inc_return(&con->c.wr_cnt) % clt_path->s.signal_interval ?
1056 			0 : IB_SEND_SIGNALED;
1057 
1058 	ib_dma_sync_single_for_device(clt_path->s.dev->ib_dev,
1059 				      req->iu->dma_addr,
1060 				      size, DMA_TO_DEVICE);
1061 
1062 	return rtrs_iu_post_rdma_write_imm(&con->c, req->iu, sge, num_sge,
1063 					    rbuf->rkey, rbuf->addr, imm,
1064 					    flags, wr, ptail);
1065 }
1066 
1067 static int rtrs_map_sg_fr(struct rtrs_clt_io_req *req, size_t count)
1068 {
1069 	int nr;
1070 
1071 	/* Align the MR to a 4K page size to match the block virt boundary */
1072 	nr = ib_map_mr_sg(req->mr, req->sglist, count, NULL, SZ_4K);
1073 	if (nr != count)
1074 		return nr < 0 ? nr : -EINVAL;
1075 	ib_update_fast_reg_key(req->mr, ib_inc_rkey(req->mr->rkey));
1076 
1077 	return nr;
1078 }
1079 
1080 static int rtrs_clt_write_req(struct rtrs_clt_io_req *req)
1081 {
1082 	struct rtrs_clt_con *con = req->con;
1083 	struct rtrs_path *s = con->c.path;
1084 	struct rtrs_clt_path *clt_path = to_clt_path(s);
1085 	struct rtrs_msg_rdma_write *msg;
1086 
1087 	struct rtrs_rbuf *rbuf;
1088 	int ret, count = 0;
1089 	u32 imm, buf_id;
1090 	struct ib_reg_wr rwr;
1091 	struct ib_send_wr *wr = NULL;
1092 	bool fr_en = false;
1093 
1094 	const size_t tsize = sizeof(*msg) + req->data_len + req->usr_len;
1095 
1096 	if (tsize > clt_path->chunk_size) {
1097 		rtrs_wrn(s, "Write request failed, size too big %zu > %d\n",
1098 			  tsize, clt_path->chunk_size);
1099 		return -EMSGSIZE;
1100 	}
1101 	if (req->sg_cnt) {
1102 		count = ib_dma_map_sg(clt_path->s.dev->ib_dev, req->sglist,
1103 				      req->sg_cnt, req->dir);
1104 		if (!count) {
1105 			rtrs_wrn(s, "Write request failed, map failed\n");
1106 			return -EINVAL;
1107 		}
1108 	}
1109 	/* put rtrs msg after sg and user message */
1110 	msg = req->iu->buf + req->usr_len;
1111 	msg->type = cpu_to_le16(RTRS_MSG_WRITE);
1112 	msg->usr_len = cpu_to_le16(req->usr_len);
1113 
1114 	/* rtrs message on server side will be after user data and message */
1115 	imm = req->permit->mem_off + req->data_len + req->usr_len;
1116 	imm = rtrs_to_io_req_imm(imm);
1117 	buf_id = req->permit->mem_id;
1118 	req->sg_size = tsize;
1119 	rbuf = &clt_path->rbufs[buf_id];
1120 
1121 	if (count) {
1122 		ret = rtrs_map_sg_fr(req, count);
1123 		if (ret < 0) {
1124 			rtrs_err_rl(s,
1125 				    "Write request failed, failed to map fast reg. data, err: %d\n",
1126 				    ret);
1127 			ib_dma_unmap_sg(clt_path->s.dev->ib_dev, req->sglist,
1128 					req->sg_cnt, req->dir);
1129 			return ret;
1130 		}
1131 		rwr = (struct ib_reg_wr) {
1132 			.wr.opcode = IB_WR_REG_MR,
1133 			.wr.wr_cqe = &fast_reg_cqe,
1134 			.mr = req->mr,
1135 			.key = req->mr->rkey,
1136 			.access = (IB_ACCESS_LOCAL_WRITE),
1137 		};
1138 		wr = &rwr.wr;
1139 		fr_en = true;
1140 		req->mr->need_inval = true;
1141 	}
1142 	/*
1143 	 * Update stats now, after request is successfully sent it is not
1144 	 * safe anymore to touch it.
1145 	 */
1146 	rtrs_clt_update_all_stats(req, WRITE);
1147 
1148 	ret = rtrs_post_rdma_write_sg(req->con, req, rbuf, fr_en, count,
1149 				      req->usr_len + sizeof(*msg),
1150 				      imm, wr, NULL);
1151 	if (ret) {
1152 		rtrs_err_rl(s,
1153 			    "Write request failed: error=%d path=%s [%s:%u]\n",
1154 			    ret, kobject_name(&clt_path->kobj), clt_path->hca_name,
1155 			    clt_path->hca_port);
1156 		if (req->mp_policy == MP_POLICY_MIN_INFLIGHT)
1157 			atomic_dec(&clt_path->stats->inflight);
1158 		if (req->mr->need_inval) {
1159 			req->mr->need_inval = false;
1160 			refcount_dec(&req->ref);
1161 		}
1162 		if (req->sg_cnt)
1163 			ib_dma_unmap_sg(clt_path->s.dev->ib_dev, req->sglist,
1164 					req->sg_cnt, req->dir);
1165 	}
1166 
1167 	return ret;
1168 }
1169 
1170 static int rtrs_clt_read_req(struct rtrs_clt_io_req *req)
1171 {
1172 	struct rtrs_clt_con *con = req->con;
1173 	struct rtrs_path *s = con->c.path;
1174 	struct rtrs_clt_path *clt_path = to_clt_path(s);
1175 	struct rtrs_msg_rdma_read *msg;
1176 	struct rtrs_ib_dev *dev = clt_path->s.dev;
1177 
1178 	struct ib_reg_wr rwr;
1179 	struct ib_send_wr *wr = NULL;
1180 
1181 	int ret, count = 0;
1182 	u32 imm, buf_id;
1183 
1184 	const size_t tsize = sizeof(*msg) + req->data_len + req->usr_len;
1185 
1186 	if (tsize > clt_path->chunk_size) {
1187 		rtrs_wrn(s,
1188 			  "Read request failed, message size is %zu, bigger than CHUNK_SIZE %d\n",
1189 			  tsize, clt_path->chunk_size);
1190 		return -EMSGSIZE;
1191 	}
1192 
1193 	if (req->sg_cnt) {
1194 		count = ib_dma_map_sg(dev->ib_dev, req->sglist, req->sg_cnt,
1195 				      req->dir);
1196 		if (!count) {
1197 			rtrs_wrn(s,
1198 				  "Read request failed, dma map failed\n");
1199 			return -EINVAL;
1200 		}
1201 	}
1202 	/* put our message into req->buf after user message*/
1203 	msg = req->iu->buf + req->usr_len;
1204 	msg->type = cpu_to_le16(RTRS_MSG_READ);
1205 	msg->usr_len = cpu_to_le16(req->usr_len);
1206 
1207 	if (count) {
1208 		ret = rtrs_map_sg_fr(req, count);
1209 		if (ret < 0) {
1210 			rtrs_err_rl(s,
1211 				     "Read request failed, failed to map fast reg. data, err: %d\n",
1212 				     ret);
1213 			ib_dma_unmap_sg(dev->ib_dev, req->sglist, req->sg_cnt,
1214 					req->dir);
1215 			return ret;
1216 		}
1217 		rwr = (struct ib_reg_wr) {
1218 			.wr.opcode = IB_WR_REG_MR,
1219 			.wr.wr_cqe = &fast_reg_cqe,
1220 			.mr = req->mr,
1221 			.key = req->mr->rkey,
1222 			.access = (IB_ACCESS_LOCAL_WRITE |
1223 				   IB_ACCESS_REMOTE_WRITE),
1224 		};
1225 		wr = &rwr.wr;
1226 
1227 		msg->sg_cnt = cpu_to_le16(1);
1228 		msg->flags = cpu_to_le16(RTRS_MSG_NEED_INVAL_F);
1229 
1230 		msg->desc[0].addr = cpu_to_le64(req->mr->iova);
1231 		msg->desc[0].key = cpu_to_le32(req->mr->rkey);
1232 		msg->desc[0].len = cpu_to_le32(req->mr->length);
1233 
1234 		/* Further invalidation is required */
1235 		req->mr->need_inval = !!RTRS_MSG_NEED_INVAL_F;
1236 
1237 	} else {
1238 		msg->sg_cnt = 0;
1239 		msg->flags = 0;
1240 	}
1241 	/*
1242 	 * rtrs message will be after the space reserved for disk data and
1243 	 * user message
1244 	 */
1245 	imm = req->permit->mem_off + req->data_len + req->usr_len;
1246 	imm = rtrs_to_io_req_imm(imm);
1247 	buf_id = req->permit->mem_id;
1248 
1249 	req->sg_size  = sizeof(*msg);
1250 	req->sg_size += le16_to_cpu(msg->sg_cnt) * sizeof(struct rtrs_sg_desc);
1251 	req->sg_size += req->usr_len;
1252 
1253 	/*
1254 	 * Update stats now, after request is successfully sent it is not
1255 	 * safe anymore to touch it.
1256 	 */
1257 	rtrs_clt_update_all_stats(req, READ);
1258 
1259 	ret = rtrs_post_send_rdma(req->con, req, &clt_path->rbufs[buf_id],
1260 				   req->data_len, imm, wr);
1261 	if (ret) {
1262 		rtrs_err_rl(s,
1263 			    "Read request failed: error=%d path=%s [%s:%u]\n",
1264 			    ret, kobject_name(&clt_path->kobj), clt_path->hca_name,
1265 			    clt_path->hca_port);
1266 		if (req->mp_policy == MP_POLICY_MIN_INFLIGHT)
1267 			atomic_dec(&clt_path->stats->inflight);
1268 		req->mr->need_inval = false;
1269 		if (req->sg_cnt)
1270 			ib_dma_unmap_sg(dev->ib_dev, req->sglist,
1271 					req->sg_cnt, req->dir);
1272 	}
1273 
1274 	return ret;
1275 }
1276 
1277 /**
1278  * rtrs_clt_failover_req() - Try to find an active path for a failed request
1279  * @clt: clt context
1280  * @fail_req: a failed io request.
1281  */
1282 static int rtrs_clt_failover_req(struct rtrs_clt_sess *clt,
1283 				 struct rtrs_clt_io_req *fail_req)
1284 {
1285 	struct rtrs_clt_path *alive_path;
1286 	struct rtrs_clt_io_req *req;
1287 	int err = -ECONNABORTED;
1288 	struct path_it it;
1289 
1290 	rcu_read_lock();
1291 	for (path_it_init(&it, clt);
1292 	     (alive_path = it.next_path(&it)) && it.i < it.clt->paths_num;
1293 	     it.i++) {
1294 		if (READ_ONCE(alive_path->state) != RTRS_CLT_CONNECTED)
1295 			continue;
1296 		req = rtrs_clt_get_copy_req(alive_path, fail_req);
1297 		if (req->dir == DMA_TO_DEVICE)
1298 			err = rtrs_clt_write_req(req);
1299 		else
1300 			err = rtrs_clt_read_req(req);
1301 		if (err) {
1302 			req->in_use = false;
1303 			continue;
1304 		}
1305 		/* Success path */
1306 		rtrs_clt_inc_failover_cnt(alive_path->stats);
1307 		break;
1308 	}
1309 	path_it_deinit(&it);
1310 	rcu_read_unlock();
1311 
1312 	return err;
1313 }
1314 
1315 static void fail_all_outstanding_reqs(struct rtrs_clt_path *clt_path)
1316 {
1317 	struct rtrs_clt_sess *clt = clt_path->clt;
1318 	struct rtrs_clt_io_req *req;
1319 	int i, err;
1320 
1321 	if (!clt_path->reqs)
1322 		return;
1323 	for (i = 0; i < clt_path->queue_depth; ++i) {
1324 		req = &clt_path->reqs[i];
1325 		if (!req->in_use)
1326 			continue;
1327 
1328 		/*
1329 		 * Safely (without notification) complete failed request.
1330 		 * After completion this request is still useble and can
1331 		 * be failovered to another path.
1332 		 */
1333 		complete_rdma_req(req, -ECONNABORTED, false, true);
1334 
1335 		err = rtrs_clt_failover_req(clt, req);
1336 		if (err)
1337 			/* Failover failed, notify anyway */
1338 			req->conf(req->priv, err);
1339 	}
1340 }
1341 
1342 static void free_path_reqs(struct rtrs_clt_path *clt_path)
1343 {
1344 	struct rtrs_clt_io_req *req;
1345 	int i;
1346 
1347 	if (!clt_path->reqs)
1348 		return;
1349 	for (i = 0; i < clt_path->queue_depth; ++i) {
1350 		req = &clt_path->reqs[i];
1351 		if (req->mr)
1352 			ib_dereg_mr(req->mr);
1353 		kfree(req->sge);
1354 		rtrs_iu_free(req->iu, clt_path->s.dev->ib_dev, 1);
1355 	}
1356 	kfree(clt_path->reqs);
1357 	clt_path->reqs = NULL;
1358 }
1359 
1360 static int alloc_path_reqs(struct rtrs_clt_path *clt_path)
1361 {
1362 	struct rtrs_clt_io_req *req;
1363 	int i, err = -ENOMEM;
1364 
1365 	clt_path->reqs = kcalloc(clt_path->queue_depth,
1366 				 sizeof(*clt_path->reqs),
1367 				 GFP_KERNEL);
1368 	if (!clt_path->reqs)
1369 		return -ENOMEM;
1370 
1371 	for (i = 0; i < clt_path->queue_depth; ++i) {
1372 		req = &clt_path->reqs[i];
1373 		req->iu = rtrs_iu_alloc(1, clt_path->max_hdr_size, GFP_KERNEL,
1374 					 clt_path->s.dev->ib_dev,
1375 					 DMA_TO_DEVICE,
1376 					 rtrs_clt_rdma_done);
1377 		if (!req->iu)
1378 			goto out;
1379 
1380 		req->sge = kcalloc(2, sizeof(*req->sge), GFP_KERNEL);
1381 		if (!req->sge)
1382 			goto out;
1383 
1384 		req->mr = ib_alloc_mr(clt_path->s.dev->ib_pd,
1385 				      IB_MR_TYPE_MEM_REG,
1386 				      clt_path->max_pages_per_mr);
1387 		if (IS_ERR(req->mr)) {
1388 			err = PTR_ERR(req->mr);
1389 			pr_err("Failed to alloc clt_path->max_pages_per_mr %d: %pe\n",
1390 			       clt_path->max_pages_per_mr, req->mr);
1391 			req->mr = NULL;
1392 			goto out;
1393 		}
1394 
1395 		init_completion(&req->inv_comp);
1396 	}
1397 
1398 	return 0;
1399 
1400 out:
1401 	free_path_reqs(clt_path);
1402 
1403 	return err;
1404 }
1405 
1406 static int alloc_permits(struct rtrs_clt_sess *clt)
1407 {
1408 	unsigned int chunk_bits;
1409 	int err, i;
1410 
1411 	clt->permits_map = bitmap_zalloc(clt->queue_depth, GFP_KERNEL);
1412 	if (!clt->permits_map) {
1413 		err = -ENOMEM;
1414 		goto out_err;
1415 	}
1416 	clt->permits = kcalloc(clt->queue_depth, permit_size(clt), GFP_KERNEL);
1417 	if (!clt->permits) {
1418 		err = -ENOMEM;
1419 		goto err_map;
1420 	}
1421 	chunk_bits = ilog2(clt->queue_depth - 1) + 1;
1422 	for (i = 0; i < clt->queue_depth; i++) {
1423 		struct rtrs_permit *permit;
1424 
1425 		permit = get_permit(clt, i);
1426 		permit->mem_id = i;
1427 		permit->mem_off = i << (MAX_IMM_PAYL_BITS - chunk_bits);
1428 	}
1429 
1430 	return 0;
1431 
1432 err_map:
1433 	bitmap_free(clt->permits_map);
1434 	clt->permits_map = NULL;
1435 out_err:
1436 	return err;
1437 }
1438 
1439 static void free_permits(struct rtrs_clt_sess *clt)
1440 {
1441 	if (clt->permits_map)
1442 		wait_event(clt->permits_wait,
1443 			   bitmap_empty(clt->permits_map, clt->queue_depth));
1444 
1445 	bitmap_free(clt->permits_map);
1446 	clt->permits_map = NULL;
1447 	kfree(clt->permits);
1448 	clt->permits = NULL;
1449 }
1450 
1451 static void query_fast_reg_mode(struct rtrs_clt_path *clt_path)
1452 {
1453 	struct ib_device *ib_dev;
1454 	u64 max_pages_per_mr;
1455 	int mr_page_shift;
1456 
1457 	ib_dev = clt_path->s.dev->ib_dev;
1458 
1459 	/*
1460 	 * Use the smallest page size supported by the HCA, down to a
1461 	 * minimum of 4096 bytes. We're unlikely to build large sglists
1462 	 * out of smaller entries.
1463 	 */
1464 	mr_page_shift      = max(12, ffs(ib_dev->attrs.page_size_cap) - 1);
1465 	max_pages_per_mr   = ib_dev->attrs.max_mr_size;
1466 	do_div(max_pages_per_mr, (1ull << mr_page_shift));
1467 	clt_path->max_pages_per_mr =
1468 		min3(clt_path->max_pages_per_mr, (u32)max_pages_per_mr,
1469 		     ib_dev->attrs.max_fast_reg_page_list_len);
1470 	clt_path->clt->max_segments =
1471 		min(clt_path->max_pages_per_mr, clt_path->clt->max_segments);
1472 }
1473 
1474 static bool rtrs_clt_change_state_get_old(struct rtrs_clt_path *clt_path,
1475 					   enum rtrs_clt_state new_state,
1476 					   enum rtrs_clt_state *old_state)
1477 {
1478 	bool changed;
1479 
1480 	spin_lock_irq(&clt_path->state_wq.lock);
1481 	if (old_state)
1482 		*old_state = clt_path->state;
1483 	changed = rtrs_clt_change_state(clt_path, new_state);
1484 	spin_unlock_irq(&clt_path->state_wq.lock);
1485 
1486 	return changed;
1487 }
1488 
1489 static void rtrs_clt_hb_err_handler(struct rtrs_con *c)
1490 {
1491 	struct rtrs_clt_con *con = container_of(c, typeof(*con), c);
1492 	struct rtrs_clt_path *clt_path = to_clt_path(con->c.path);
1493 
1494 	rtrs_err(con->c.path, "HB err handler for path=%s\n", kobject_name(&clt_path->kobj));
1495 	rtrs_rdma_error_recovery(con);
1496 }
1497 
1498 static void rtrs_clt_init_hb(struct rtrs_clt_path *clt_path)
1499 {
1500 	rtrs_init_hb(&clt_path->s, &io_comp_cqe,
1501 		      RTRS_HB_INTERVAL_MS,
1502 		      RTRS_HB_MISSED_MAX,
1503 		      rtrs_clt_hb_err_handler,
1504 		      rtrs_wq);
1505 }
1506 
1507 static void rtrs_clt_reconnect_work(struct work_struct *work);
1508 static void rtrs_clt_close_work(struct work_struct *work);
1509 
1510 static void rtrs_clt_err_recovery_work(struct work_struct *work)
1511 {
1512 	struct rtrs_clt_path *clt_path;
1513 	struct rtrs_clt_sess *clt;
1514 	int delay_ms;
1515 
1516 	clt_path = container_of(work, struct rtrs_clt_path, err_recovery_work);
1517 	clt = clt_path->clt;
1518 	delay_ms = clt->reconnect_delay_sec * 1000;
1519 	rtrs_clt_stop_and_destroy_conns(clt_path);
1520 	queue_delayed_work(rtrs_wq, &clt_path->reconnect_dwork,
1521 			   msecs_to_jiffies(delay_ms +
1522 					    get_random_u32_below(RTRS_RECONNECT_SEED)));
1523 }
1524 
1525 static struct rtrs_clt_path *alloc_path(struct rtrs_clt_sess *clt,
1526 					const struct rtrs_addr *path,
1527 					size_t con_num, u32 nr_poll_queues)
1528 {
1529 	struct rtrs_clt_path *clt_path;
1530 	int err = -ENOMEM;
1531 	int cpu;
1532 	size_t total_con;
1533 
1534 	clt_path = kzalloc(sizeof(*clt_path), GFP_KERNEL);
1535 	if (!clt_path)
1536 		goto err;
1537 
1538 	/*
1539 	 * irqmode and poll
1540 	 * +1: Extra connection for user messages
1541 	 */
1542 	total_con = con_num + nr_poll_queues + 1;
1543 	clt_path->s.con = kcalloc(total_con, sizeof(*clt_path->s.con),
1544 				  GFP_KERNEL);
1545 	if (!clt_path->s.con)
1546 		goto err_free_path;
1547 
1548 	clt_path->s.con_num = total_con;
1549 	clt_path->s.irq_con_num = con_num + 1;
1550 
1551 	clt_path->stats = kzalloc(sizeof(*clt_path->stats), GFP_KERNEL);
1552 	if (!clt_path->stats)
1553 		goto err_free_con;
1554 
1555 	mutex_init(&clt_path->init_mutex);
1556 	uuid_gen(&clt_path->s.uuid);
1557 	memcpy(&clt_path->s.dst_addr, path->dst,
1558 	       rdma_addr_size((struct sockaddr *)path->dst));
1559 
1560 	/*
1561 	 * rdma_resolve_addr() passes src_addr to cma_bind_addr, which
1562 	 * checks the sa_family to be non-zero. If user passed src_addr=NULL
1563 	 * the sess->src_addr will contain only zeros, which is then fine.
1564 	 */
1565 	if (path->src)
1566 		memcpy(&clt_path->s.src_addr, path->src,
1567 		       rdma_addr_size((struct sockaddr *)path->src));
1568 	strscpy(clt_path->s.sessname, clt->sessname,
1569 		sizeof(clt_path->s.sessname));
1570 	clt_path->clt = clt;
1571 	clt_path->max_pages_per_mr = RTRS_MAX_SEGMENTS;
1572 	init_waitqueue_head(&clt_path->state_wq);
1573 	clt_path->state = RTRS_CLT_CONNECTING;
1574 	atomic_set(&clt_path->connected_cnt, 0);
1575 	INIT_WORK(&clt_path->close_work, rtrs_clt_close_work);
1576 	INIT_WORK(&clt_path->err_recovery_work, rtrs_clt_err_recovery_work);
1577 	INIT_DELAYED_WORK(&clt_path->reconnect_dwork, rtrs_clt_reconnect_work);
1578 	rtrs_clt_init_hb(clt_path);
1579 
1580 	clt_path->mp_skip_entry = alloc_percpu(typeof(*clt_path->mp_skip_entry));
1581 	if (!clt_path->mp_skip_entry)
1582 		goto err_free_stats;
1583 
1584 	for_each_possible_cpu(cpu)
1585 		INIT_LIST_HEAD(per_cpu_ptr(clt_path->mp_skip_entry, cpu));
1586 
1587 	err = rtrs_clt_init_stats(clt_path->stats);
1588 	if (err)
1589 		goto err_free_percpu;
1590 
1591 	return clt_path;
1592 
1593 err_free_percpu:
1594 	free_percpu(clt_path->mp_skip_entry);
1595 err_free_stats:
1596 	kfree(clt_path->stats);
1597 err_free_con:
1598 	kfree(clt_path->s.con);
1599 err_free_path:
1600 	kfree(clt_path);
1601 err:
1602 	return ERR_PTR(err);
1603 }
1604 
1605 void free_path(struct rtrs_clt_path *clt_path)
1606 {
1607 	free_percpu(clt_path->mp_skip_entry);
1608 	mutex_destroy(&clt_path->init_mutex);
1609 	kfree(clt_path->s.con);
1610 	kfree(clt_path->rbufs);
1611 	kfree(clt_path);
1612 }
1613 
1614 static int create_con(struct rtrs_clt_path *clt_path, unsigned int cid)
1615 {
1616 	struct rtrs_clt_con *con;
1617 
1618 	con = kzalloc(sizeof(*con), GFP_KERNEL);
1619 	if (!con)
1620 		return -ENOMEM;
1621 
1622 	/* Map first two connections to the first CPU */
1623 	con->cpu  = (cid ? cid - 1 : 0) % nr_cpu_ids;
1624 	con->c.cid = cid;
1625 	con->c.path = &clt_path->s;
1626 	/* Align with srv, init as 1 */
1627 	atomic_set(&con->c.wr_cnt, 1);
1628 	mutex_init(&con->con_mutex);
1629 
1630 	clt_path->s.con[cid] = &con->c;
1631 
1632 	return 0;
1633 }
1634 
1635 static void destroy_con(struct rtrs_clt_con *con)
1636 {
1637 	struct rtrs_clt_path *clt_path = to_clt_path(con->c.path);
1638 
1639 	clt_path->s.con[con->c.cid] = NULL;
1640 	mutex_destroy(&con->con_mutex);
1641 	kfree(con);
1642 }
1643 
1644 static int create_con_cq_qp(struct rtrs_clt_con *con)
1645 {
1646 	struct rtrs_clt_path *clt_path = to_clt_path(con->c.path);
1647 	u32 max_send_wr, max_recv_wr, cq_num, max_send_sge, wr_limit;
1648 	int err, cq_vector;
1649 	struct rtrs_msg_rkey_rsp *rsp;
1650 
1651 	lockdep_assert_held(&con->con_mutex);
1652 	if (con->c.cid == 0) {
1653 		max_send_sge = 1;
1654 		/* We must be the first here */
1655 		if (WARN_ON(clt_path->s.dev))
1656 			return -EINVAL;
1657 
1658 		/*
1659 		 * The whole session uses device from user connection.
1660 		 * Be careful not to close user connection before ib dev
1661 		 * is gracefully put.
1662 		 */
1663 		clt_path->s.dev = rtrs_ib_dev_find_or_add(con->c.cm_id->device,
1664 						       &dev_pd);
1665 		if (!clt_path->s.dev) {
1666 			rtrs_wrn(clt_path->clt,
1667 				  "rtrs_ib_dev_find_get_or_add(): no memory\n");
1668 			return -ENOMEM;
1669 		}
1670 		clt_path->s.dev_ref = 1;
1671 		query_fast_reg_mode(clt_path);
1672 		wr_limit = clt_path->s.dev->ib_dev->attrs.max_qp_wr;
1673 		/*
1674 		 * Two (request + registration) completion for send
1675 		 * Two for recv if always_invalidate is set on server
1676 		 * or one for recv.
1677 		 * + 2 for drain and heartbeat
1678 		 * in case qp gets into error state.
1679 		 */
1680 		max_send_wr =
1681 			min_t(int, wr_limit, SERVICE_CON_QUEUE_DEPTH * 2 + 2);
1682 		max_recv_wr = max_send_wr;
1683 	} else {
1684 		/*
1685 		 * Here we assume that session members are correctly set.
1686 		 * This is always true if user connection (cid == 0) is
1687 		 * established first.
1688 		 */
1689 		if (WARN_ON(!clt_path->s.dev))
1690 			return -EINVAL;
1691 		if (WARN_ON(!clt_path->queue_depth))
1692 			return -EINVAL;
1693 
1694 		wr_limit = clt_path->s.dev->ib_dev->attrs.max_qp_wr;
1695 		/* Shared between connections */
1696 		clt_path->s.dev_ref++;
1697 		max_send_wr = min_t(int, wr_limit,
1698 			      /* QD * (REQ + RSP + FR REGS or INVS) + drain */
1699 			      clt_path->queue_depth * 4 + 1);
1700 		max_recv_wr = min_t(int, wr_limit,
1701 			      clt_path->queue_depth * 3 + 1);
1702 		max_send_sge = 2;
1703 	}
1704 	atomic_set(&con->c.sq_wr_avail, max_send_wr);
1705 	cq_num = max_send_wr + max_recv_wr;
1706 	/* alloc iu to recv new rkey reply when server reports flags set */
1707 	if (clt_path->flags & RTRS_MSG_NEW_RKEY_F || con->c.cid == 0) {
1708 		con->rsp_ius = rtrs_iu_alloc(cq_num, sizeof(*rsp),
1709 					      GFP_KERNEL,
1710 					      clt_path->s.dev->ib_dev,
1711 					      DMA_FROM_DEVICE,
1712 					      rtrs_clt_rdma_done);
1713 		if (!con->rsp_ius)
1714 			return -ENOMEM;
1715 		con->queue_num = cq_num;
1716 	}
1717 	cq_vector = con->cpu % clt_path->s.dev->ib_dev->num_comp_vectors;
1718 	if (con->c.cid >= clt_path->s.irq_con_num)
1719 		err = rtrs_cq_qp_create(&clt_path->s, &con->c, max_send_sge,
1720 					cq_vector, cq_num, max_send_wr,
1721 					max_recv_wr, IB_POLL_DIRECT);
1722 	else
1723 		err = rtrs_cq_qp_create(&clt_path->s, &con->c, max_send_sge,
1724 					cq_vector, cq_num, max_send_wr,
1725 					max_recv_wr, IB_POLL_SOFTIRQ);
1726 	/*
1727 	 * In case of error we do not bother to clean previous allocations,
1728 	 * since destroy_con_cq_qp() must be called.
1729 	 */
1730 	return err;
1731 }
1732 
1733 static void destroy_con_cq_qp(struct rtrs_clt_con *con)
1734 {
1735 	struct rtrs_clt_path *clt_path = to_clt_path(con->c.path);
1736 
1737 	/*
1738 	 * Be careful here: destroy_con_cq_qp() can be called even
1739 	 * create_con_cq_qp() failed, see comments there.
1740 	 */
1741 	lockdep_assert_held(&con->con_mutex);
1742 	rtrs_cq_qp_destroy(&con->c);
1743 	if (con->rsp_ius) {
1744 		rtrs_iu_free(con->rsp_ius, clt_path->s.dev->ib_dev,
1745 			     con->queue_num);
1746 		con->rsp_ius = NULL;
1747 		con->queue_num = 0;
1748 	}
1749 	if (clt_path->s.dev_ref && !--clt_path->s.dev_ref) {
1750 		rtrs_ib_dev_put(clt_path->s.dev);
1751 		clt_path->s.dev = NULL;
1752 	}
1753 }
1754 
1755 static void stop_cm(struct rtrs_clt_con *con)
1756 {
1757 	rdma_disconnect(con->c.cm_id);
1758 	if (con->c.qp)
1759 		ib_drain_qp(con->c.qp);
1760 }
1761 
1762 static void destroy_cm(struct rtrs_clt_con *con)
1763 {
1764 	rdma_destroy_id(con->c.cm_id);
1765 	con->c.cm_id = NULL;
1766 }
1767 
1768 static int rtrs_rdma_addr_resolved(struct rtrs_clt_con *con)
1769 {
1770 	struct rtrs_path *s = con->c.path;
1771 	int err;
1772 
1773 	mutex_lock(&con->con_mutex);
1774 	err = create_con_cq_qp(con);
1775 	mutex_unlock(&con->con_mutex);
1776 	if (err) {
1777 		rtrs_err(s, "create_con_cq_qp(), err: %d\n", err);
1778 		return err;
1779 	}
1780 	err = rdma_resolve_route(con->c.cm_id, RTRS_CONNECT_TIMEOUT_MS);
1781 	if (err)
1782 		rtrs_err(s, "Resolving route failed, err: %d\n", err);
1783 
1784 	return err;
1785 }
1786 
1787 static int rtrs_rdma_route_resolved(struct rtrs_clt_con *con)
1788 {
1789 	struct rtrs_clt_path *clt_path = to_clt_path(con->c.path);
1790 	struct rtrs_clt_sess *clt = clt_path->clt;
1791 	struct rtrs_msg_conn_req msg;
1792 	struct rdma_conn_param param;
1793 
1794 	int err;
1795 
1796 	param = (struct rdma_conn_param) {
1797 		.retry_count = 7,
1798 		.rnr_retry_count = 7,
1799 		.private_data = &msg,
1800 		.private_data_len = sizeof(msg),
1801 	};
1802 
1803 	msg = (struct rtrs_msg_conn_req) {
1804 		.magic = cpu_to_le16(RTRS_MAGIC),
1805 		.version = cpu_to_le16(RTRS_PROTO_VER),
1806 		.cid = cpu_to_le16(con->c.cid),
1807 		.cid_num = cpu_to_le16(clt_path->s.con_num),
1808 		.recon_cnt = cpu_to_le16(clt_path->s.recon_cnt),
1809 	};
1810 	msg.first_conn = clt_path->for_new_clt ? FIRST_CONN : 0;
1811 	uuid_copy(&msg.sess_uuid, &clt_path->s.uuid);
1812 	uuid_copy(&msg.paths_uuid, &clt->paths_uuid);
1813 
1814 	err = rdma_connect_locked(con->c.cm_id, &param);
1815 	if (err)
1816 		rtrs_err(clt, "rdma_connect_locked(): %d\n", err);
1817 
1818 	return err;
1819 }
1820 
1821 static int rtrs_rdma_conn_established(struct rtrs_clt_con *con,
1822 				       struct rdma_cm_event *ev)
1823 {
1824 	struct rtrs_clt_path *clt_path = to_clt_path(con->c.path);
1825 	struct rtrs_clt_sess *clt = clt_path->clt;
1826 	const struct rtrs_msg_conn_rsp *msg;
1827 	u16 version, queue_depth;
1828 	int errno;
1829 	u8 len;
1830 
1831 	msg = ev->param.conn.private_data;
1832 	len = ev->param.conn.private_data_len;
1833 	if (len < sizeof(*msg)) {
1834 		rtrs_err(clt, "Invalid RTRS connection response\n");
1835 		return -ECONNRESET;
1836 	}
1837 	if (le16_to_cpu(msg->magic) != RTRS_MAGIC) {
1838 		rtrs_err(clt, "Invalid RTRS magic\n");
1839 		return -ECONNRESET;
1840 	}
1841 	version = le16_to_cpu(msg->version);
1842 	if (version >> 8 != RTRS_PROTO_VER_MAJOR) {
1843 		rtrs_err(clt, "Unsupported major RTRS version: %d, expected %d\n",
1844 			  version >> 8, RTRS_PROTO_VER_MAJOR);
1845 		return -ECONNRESET;
1846 	}
1847 	errno = le16_to_cpu(msg->errno);
1848 	if (errno) {
1849 		rtrs_err(clt, "Invalid RTRS message: errno %d\n",
1850 			  errno);
1851 		return -ECONNRESET;
1852 	}
1853 	if (con->c.cid == 0) {
1854 		queue_depth = le16_to_cpu(msg->queue_depth);
1855 
1856 		if (clt_path->queue_depth > 0 && queue_depth != clt_path->queue_depth) {
1857 			rtrs_err(clt, "Error: queue depth changed\n");
1858 
1859 			/*
1860 			 * Stop any more reconnection attempts
1861 			 */
1862 			clt_path->reconnect_attempts = -1;
1863 			rtrs_err(clt,
1864 				"Disabling auto-reconnect. Trigger a manual reconnect after issue is resolved\n");
1865 			return -ECONNRESET;
1866 		}
1867 
1868 		if (!clt_path->rbufs) {
1869 			clt_path->rbufs = kcalloc(queue_depth,
1870 						  sizeof(*clt_path->rbufs),
1871 						  GFP_KERNEL);
1872 			if (!clt_path->rbufs)
1873 				return -ENOMEM;
1874 		}
1875 		clt_path->queue_depth = queue_depth;
1876 		clt_path->s.signal_interval = min_not_zero(queue_depth,
1877 						(unsigned short) SERVICE_CON_QUEUE_DEPTH);
1878 		clt_path->max_hdr_size = le32_to_cpu(msg->max_hdr_size);
1879 		clt_path->max_io_size = le32_to_cpu(msg->max_io_size);
1880 		clt_path->flags = le32_to_cpu(msg->flags);
1881 		clt_path->chunk_size = clt_path->max_io_size + clt_path->max_hdr_size;
1882 
1883 		/*
1884 		 * Global IO size is always a minimum.
1885 		 * If while a reconnection server sends us a value a bit
1886 		 * higher - client does not care and uses cached minimum.
1887 		 *
1888 		 * Since we can have several sessions (paths) restablishing
1889 		 * connections in parallel, use lock.
1890 		 */
1891 		mutex_lock(&clt->paths_mutex);
1892 		clt->queue_depth = clt_path->queue_depth;
1893 		clt->max_io_size = min_not_zero(clt_path->max_io_size,
1894 						clt->max_io_size);
1895 		mutex_unlock(&clt->paths_mutex);
1896 
1897 		/*
1898 		 * Cache the hca_port and hca_name for sysfs
1899 		 */
1900 		clt_path->hca_port = con->c.cm_id->port_num;
1901 		scnprintf(clt_path->hca_name, sizeof(clt_path->hca_name),
1902 			  clt_path->s.dev->ib_dev->name);
1903 		clt_path->s.src_addr = con->c.cm_id->route.addr.src_addr;
1904 		/* set for_new_clt, to allow future reconnect on any path */
1905 		clt_path->for_new_clt = 1;
1906 	}
1907 
1908 	return 0;
1909 }
1910 
1911 static inline void flag_success_on_conn(struct rtrs_clt_con *con)
1912 {
1913 	struct rtrs_clt_path *clt_path = to_clt_path(con->c.path);
1914 
1915 	atomic_inc(&clt_path->connected_cnt);
1916 	con->cm_err = 1;
1917 }
1918 
1919 static int rtrs_rdma_conn_rejected(struct rtrs_clt_con *con,
1920 				    struct rdma_cm_event *ev)
1921 {
1922 	struct rtrs_path *s = con->c.path;
1923 	const struct rtrs_msg_conn_rsp *msg;
1924 	const char *rej_msg;
1925 	int status, errno;
1926 	u8 data_len;
1927 
1928 	status = ev->status;
1929 	rej_msg = rdma_reject_msg(con->c.cm_id, status);
1930 	msg = rdma_consumer_reject_data(con->c.cm_id, ev, &data_len);
1931 
1932 	if (msg && data_len >= sizeof(*msg)) {
1933 		errno = (int16_t)le16_to_cpu(msg->errno);
1934 		if (errno == -EBUSY)
1935 			rtrs_err(s,
1936 				  "Previous session is still exists on the server, please reconnect later\n");
1937 		else
1938 			rtrs_err(s,
1939 				  "Connect rejected: status %d (%s), rtrs errno %d\n",
1940 				  status, rej_msg, errno);
1941 	} else {
1942 		rtrs_err(s,
1943 			  "Connect rejected but with malformed message: status %d (%s)\n",
1944 			  status, rej_msg);
1945 	}
1946 
1947 	return -ECONNRESET;
1948 }
1949 
1950 void rtrs_clt_close_conns(struct rtrs_clt_path *clt_path, bool wait)
1951 {
1952 	trace_rtrs_clt_close_conns(clt_path);
1953 
1954 	if (rtrs_clt_change_state_get_old(clt_path, RTRS_CLT_CLOSING, NULL))
1955 		queue_work(rtrs_wq, &clt_path->close_work);
1956 	if (wait)
1957 		flush_work(&clt_path->close_work);
1958 }
1959 
1960 static inline void flag_error_on_conn(struct rtrs_clt_con *con, int cm_err)
1961 {
1962 	if (con->cm_err == 1) {
1963 		struct rtrs_clt_path *clt_path;
1964 
1965 		clt_path = to_clt_path(con->c.path);
1966 		if (atomic_dec_and_test(&clt_path->connected_cnt))
1967 
1968 			wake_up(&clt_path->state_wq);
1969 	}
1970 	con->cm_err = cm_err;
1971 }
1972 
1973 static int rtrs_clt_rdma_cm_handler(struct rdma_cm_id *cm_id,
1974 				     struct rdma_cm_event *ev)
1975 {
1976 	struct rtrs_clt_con *con = cm_id->context;
1977 	struct rtrs_path *s = con->c.path;
1978 	struct rtrs_clt_path *clt_path = to_clt_path(s);
1979 	int cm_err = 0;
1980 
1981 	switch (ev->event) {
1982 	case RDMA_CM_EVENT_ADDR_RESOLVED:
1983 		cm_err = rtrs_rdma_addr_resolved(con);
1984 		break;
1985 	case RDMA_CM_EVENT_ROUTE_RESOLVED:
1986 		cm_err = rtrs_rdma_route_resolved(con);
1987 		break;
1988 	case RDMA_CM_EVENT_ESTABLISHED:
1989 		cm_err = rtrs_rdma_conn_established(con, ev);
1990 		if (!cm_err) {
1991 			/*
1992 			 * Report success and wake up. Here we abuse state_wq,
1993 			 * i.e. wake up without state change, but we set cm_err.
1994 			 */
1995 			flag_success_on_conn(con);
1996 			wake_up(&clt_path->state_wq);
1997 			return 0;
1998 		}
1999 		break;
2000 	case RDMA_CM_EVENT_REJECTED:
2001 		cm_err = rtrs_rdma_conn_rejected(con, ev);
2002 		break;
2003 	case RDMA_CM_EVENT_DISCONNECTED:
2004 		/* No message for disconnecting */
2005 		cm_err = -ECONNRESET;
2006 		break;
2007 	case RDMA_CM_EVENT_CONNECT_ERROR:
2008 	case RDMA_CM_EVENT_UNREACHABLE:
2009 	case RDMA_CM_EVENT_ADDR_CHANGE:
2010 	case RDMA_CM_EVENT_TIMEWAIT_EXIT:
2011 		rtrs_wrn(s, "CM error (CM event: %s, err: %d)\n",
2012 			 rdma_event_msg(ev->event), ev->status);
2013 		cm_err = -ECONNRESET;
2014 		break;
2015 	case RDMA_CM_EVENT_ADDR_ERROR:
2016 	case RDMA_CM_EVENT_ROUTE_ERROR:
2017 		rtrs_wrn(s, "CM error (CM event: %s, err: %d)\n",
2018 			 rdma_event_msg(ev->event), ev->status);
2019 		cm_err = -EHOSTUNREACH;
2020 		break;
2021 	case RDMA_CM_EVENT_DEVICE_REMOVAL:
2022 		/*
2023 		 * Device removal is a special case.  Queue close and return 0.
2024 		 */
2025 		rtrs_wrn_rl(s, "CM event: %s, status: %d\n", rdma_event_msg(ev->event),
2026 			    ev->status);
2027 		rtrs_clt_close_conns(clt_path, false);
2028 		return 0;
2029 	default:
2030 		rtrs_err(s, "Unexpected RDMA CM error (CM event: %s, err: %d)\n",
2031 			 rdma_event_msg(ev->event), ev->status);
2032 		cm_err = -ECONNRESET;
2033 		break;
2034 	}
2035 
2036 	if (cm_err) {
2037 		/*
2038 		 * cm error makes sense only on connection establishing,
2039 		 * in other cases we rely on normal procedure of reconnecting.
2040 		 */
2041 		flag_error_on_conn(con, cm_err);
2042 		rtrs_rdma_error_recovery(con);
2043 	}
2044 
2045 	return 0;
2046 }
2047 
2048 /* The caller should do the cleanup in case of error */
2049 static int create_cm(struct rtrs_clt_con *con)
2050 {
2051 	struct rtrs_path *s = con->c.path;
2052 	struct rtrs_clt_path *clt_path = to_clt_path(s);
2053 	struct rdma_cm_id *cm_id;
2054 	int err;
2055 
2056 	cm_id = rdma_create_id(&init_net, rtrs_clt_rdma_cm_handler, con,
2057 			       clt_path->s.dst_addr.ss_family == AF_IB ?
2058 			       RDMA_PS_IB : RDMA_PS_TCP, IB_QPT_RC);
2059 	if (IS_ERR(cm_id)) {
2060 		rtrs_err(s, "Failed to create CM ID, err: %pe\n", cm_id);
2061 		return PTR_ERR(cm_id);
2062 	}
2063 	con->c.cm_id = cm_id;
2064 	con->cm_err = 0;
2065 	/* allow the port to be reused */
2066 	err = rdma_set_reuseaddr(cm_id, 1);
2067 	if (err != 0) {
2068 		rtrs_err(s, "Set address reuse failed, err: %d\n", err);
2069 		return err;
2070 	}
2071 	err = rdma_resolve_addr(cm_id, (struct sockaddr *)&clt_path->s.src_addr,
2072 				(struct sockaddr *)&clt_path->s.dst_addr,
2073 				RTRS_CONNECT_TIMEOUT_MS);
2074 	if (err) {
2075 		rtrs_err(s, "Failed to resolve address, err: %d\n", err);
2076 		return err;
2077 	}
2078 	/*
2079 	 * Combine connection status and session events. This is needed
2080 	 * for waiting two possible cases: cm_err has something meaningful
2081 	 * or session state was really changed to error by device removal.
2082 	 */
2083 	err = wait_event_interruptible_timeout(
2084 			clt_path->state_wq,
2085 			con->cm_err || clt_path->state != RTRS_CLT_CONNECTING,
2086 			msecs_to_jiffies(RTRS_CONNECT_TIMEOUT_MS));
2087 	if (err == 0 || err == -ERESTARTSYS) {
2088 		if (err == 0)
2089 			err = -ETIMEDOUT;
2090 		/* Timedout or interrupted */
2091 		return err;
2092 	}
2093 	if (con->cm_err < 0)
2094 		return con->cm_err;
2095 	if (READ_ONCE(clt_path->state) != RTRS_CLT_CONNECTING)
2096 		/* Device removal */
2097 		return -ECONNABORTED;
2098 
2099 	return 0;
2100 }
2101 
2102 static void rtrs_clt_path_up(struct rtrs_clt_path *clt_path)
2103 {
2104 	struct rtrs_clt_sess *clt = clt_path->clt;
2105 	int up;
2106 
2107 	/*
2108 	 * We can fire RECONNECTED event only when all paths were
2109 	 * connected on rtrs_clt_open(), then each was disconnected
2110 	 * and the first one connected again.  That's why this nasty
2111 	 * game with counter value.
2112 	 */
2113 
2114 	mutex_lock(&clt->paths_ev_mutex);
2115 	up = ++clt->paths_up;
2116 	/*
2117 	 * Here it is safe to access paths num directly since up counter
2118 	 * is greater than MAX_PATHS_NUM only while rtrs_clt_open() is
2119 	 * in progress, thus paths removals are impossible.
2120 	 */
2121 	if (up > MAX_PATHS_NUM && up == MAX_PATHS_NUM + clt->paths_num)
2122 		clt->paths_up = clt->paths_num;
2123 	else if (up == 1)
2124 		clt->link_ev(clt->priv, RTRS_CLT_LINK_EV_RECONNECTED);
2125 	mutex_unlock(&clt->paths_ev_mutex);
2126 
2127 	/* Mark session as established */
2128 	clt_path->established = true;
2129 	clt_path->reconnect_attempts = 0;
2130 	clt_path->stats->reconnects.successful_cnt++;
2131 }
2132 
2133 static void rtrs_clt_path_down(struct rtrs_clt_path *clt_path)
2134 {
2135 	struct rtrs_clt_sess *clt = clt_path->clt;
2136 
2137 	if (!clt_path->established)
2138 		return;
2139 
2140 	clt_path->established = false;
2141 	mutex_lock(&clt->paths_ev_mutex);
2142 	WARN_ON(!clt->paths_up);
2143 	if (--clt->paths_up == 0)
2144 		clt->link_ev(clt->priv, RTRS_CLT_LINK_EV_DISCONNECTED);
2145 	mutex_unlock(&clt->paths_ev_mutex);
2146 }
2147 
2148 static void rtrs_clt_stop_and_destroy_conns(struct rtrs_clt_path *clt_path)
2149 {
2150 	struct rtrs_clt_con *con;
2151 	unsigned int cid;
2152 
2153 	WARN_ON(READ_ONCE(clt_path->state) == RTRS_CLT_CONNECTED);
2154 
2155 	/*
2156 	 * Possible race with rtrs_clt_open(), when DEVICE_REMOVAL comes
2157 	 * exactly in between.  Start destroying after it finishes.
2158 	 */
2159 	mutex_lock(&clt_path->init_mutex);
2160 	mutex_unlock(&clt_path->init_mutex);
2161 
2162 	/*
2163 	 * All IO paths must observe !CONNECTED state before we
2164 	 * free everything.
2165 	 */
2166 	synchronize_rcu();
2167 
2168 	rtrs_stop_hb(&clt_path->s);
2169 
2170 	/*
2171 	 * The order it utterly crucial: firstly disconnect and complete all
2172 	 * rdma requests with error (thus set in_use=false for requests),
2173 	 * then fail outstanding requests checking in_use for each, and
2174 	 * eventually notify upper layer about session disconnection.
2175 	 */
2176 
2177 	for (cid = 0; cid < clt_path->s.con_num; cid++) {
2178 		if (!clt_path->s.con[cid])
2179 			break;
2180 		con = to_clt_con(clt_path->s.con[cid]);
2181 		stop_cm(con);
2182 	}
2183 	fail_all_outstanding_reqs(clt_path);
2184 	free_path_reqs(clt_path);
2185 	rtrs_clt_path_down(clt_path);
2186 
2187 	/*
2188 	 * Wait for graceful shutdown, namely when peer side invokes
2189 	 * rdma_disconnect(). 'connected_cnt' is decremented only on
2190 	 * CM events, thus if other side had crashed and hb has detected
2191 	 * something is wrong, here we will stuck for exactly timeout ms,
2192 	 * since CM does not fire anything.  That is fine, we are not in
2193 	 * hurry.
2194 	 */
2195 	wait_event_timeout(clt_path->state_wq,
2196 			   !atomic_read(&clt_path->connected_cnt),
2197 			   msecs_to_jiffies(RTRS_CONNECT_TIMEOUT_MS));
2198 
2199 	for (cid = 0; cid < clt_path->s.con_num; cid++) {
2200 		if (!clt_path->s.con[cid])
2201 			break;
2202 		con = to_clt_con(clt_path->s.con[cid]);
2203 		mutex_lock(&con->con_mutex);
2204 		destroy_con_cq_qp(con);
2205 		mutex_unlock(&con->con_mutex);
2206 		destroy_cm(con);
2207 		destroy_con(con);
2208 	}
2209 }
2210 
2211 static void rtrs_clt_remove_path_from_arr(struct rtrs_clt_path *clt_path)
2212 {
2213 	struct rtrs_clt_sess *clt = clt_path->clt;
2214 	struct rtrs_clt_path *next;
2215 	bool wait_for_grace = false;
2216 	int cpu;
2217 
2218 	mutex_lock(&clt->paths_mutex);
2219 	list_del_rcu(&clt_path->s.entry);
2220 
2221 	/* Make sure everybody observes path removal. */
2222 	synchronize_rcu();
2223 
2224 	/*
2225 	 * At this point nobody sees @sess in the list, but still we have
2226 	 * dangling pointer @pcpu_path which _can_ point to @sess.  Since
2227 	 * nobody can observe @sess in the list, we guarantee that IO path
2228 	 * will not assign @sess to @pcpu_path, i.e. @pcpu_path can be equal
2229 	 * to @sess, but can never again become @sess.
2230 	 */
2231 
2232 	/*
2233 	 * Decrement paths number only after grace period, because
2234 	 * caller of do_each_path() must firstly observe list without
2235 	 * path and only then decremented paths number.
2236 	 *
2237 	 * Otherwise there can be the following situation:
2238 	 *    o Two paths exist and IO is coming.
2239 	 *    o One path is removed:
2240 	 *      CPU#0                          CPU#1
2241 	 *      do_each_path():                rtrs_clt_remove_path_from_arr():
2242 	 *          path = get_next_path()
2243 	 *          ^^^                            list_del_rcu(path)
2244 	 *          [!CONNECTED path]              clt->paths_num--
2245 	 *                                              ^^^^^^^^^
2246 	 *          load clt->paths_num                 from 2 to 1
2247 	 *                    ^^^^^^^^^
2248 	 *                    sees 1
2249 	 *
2250 	 *      path is observed as !CONNECTED, but do_each_path() loop
2251 	 *      ends, because expression i < clt->paths_num is false.
2252 	 */
2253 	clt->paths_num--;
2254 
2255 	/*
2256 	 * Get @next connection from current @sess which is going to be
2257 	 * removed.  If @sess is the last element, then @next is NULL.
2258 	 */
2259 	rcu_read_lock();
2260 	next = rtrs_clt_get_next_path_or_null(&clt->paths_list, clt_path);
2261 	rcu_read_unlock();
2262 
2263 	/*
2264 	 * @pcpu paths can still point to the path which is going to be
2265 	 * removed, so change the pointer manually.
2266 	 */
2267 	for_each_possible_cpu(cpu) {
2268 		struct rtrs_clt_path __rcu **ppcpu_path;
2269 
2270 		ppcpu_path = per_cpu_ptr(clt->pcpu_path, cpu);
2271 		if (rcu_dereference_protected(*ppcpu_path,
2272 			lockdep_is_held(&clt->paths_mutex)) != clt_path)
2273 			/*
2274 			 * synchronize_rcu() was called just after deleting
2275 			 * entry from the list, thus IO code path cannot
2276 			 * change pointer back to the pointer which is going
2277 			 * to be removed, we are safe here.
2278 			 */
2279 			continue;
2280 
2281 		/*
2282 		 * We race with IO code path, which also changes pointer,
2283 		 * thus we have to be careful not to overwrite it.
2284 		 */
2285 		if (try_cmpxchg((struct rtrs_clt_path **)ppcpu_path, &clt_path,
2286 				next))
2287 			/*
2288 			 * @ppcpu_path was successfully replaced with @next,
2289 			 * that means that someone could also pick up the
2290 			 * @sess and dereferencing it right now, so wait for
2291 			 * a grace period is required.
2292 			 */
2293 			wait_for_grace = true;
2294 	}
2295 	if (wait_for_grace)
2296 		synchronize_rcu();
2297 
2298 	mutex_unlock(&clt->paths_mutex);
2299 }
2300 
2301 static void rtrs_clt_add_path_to_arr(struct rtrs_clt_path *clt_path)
2302 {
2303 	struct rtrs_clt_sess *clt = clt_path->clt;
2304 
2305 	mutex_lock(&clt->paths_mutex);
2306 	clt->paths_num++;
2307 
2308 	list_add_tail_rcu(&clt_path->s.entry, &clt->paths_list);
2309 	mutex_unlock(&clt->paths_mutex);
2310 }
2311 
2312 static void rtrs_clt_close_work(struct work_struct *work)
2313 {
2314 	struct rtrs_clt_path *clt_path;
2315 
2316 	clt_path = container_of(work, struct rtrs_clt_path, close_work);
2317 
2318 	cancel_work_sync(&clt_path->err_recovery_work);
2319 	cancel_delayed_work_sync(&clt_path->reconnect_dwork);
2320 	rtrs_clt_stop_and_destroy_conns(clt_path);
2321 	rtrs_clt_change_state_get_old(clt_path, RTRS_CLT_CLOSED, NULL);
2322 }
2323 
2324 static int init_conns(struct rtrs_clt_path *clt_path)
2325 {
2326 	unsigned int cid;
2327 	int err, i;
2328 
2329 	/*
2330 	 * On every new session connections increase reconnect counter
2331 	 * to avoid clashes with previous sessions not yet closed
2332 	 * sessions on a server side.
2333 	 */
2334 	clt_path->s.recon_cnt++;
2335 
2336 	/* Establish all RDMA connections  */
2337 	for (cid = 0; cid < clt_path->s.con_num; cid++) {
2338 		err = create_con(clt_path, cid);
2339 		if (err)
2340 			goto destroy;
2341 
2342 		err = create_cm(to_clt_con(clt_path->s.con[cid]));
2343 		if (err)
2344 			goto destroy;
2345 	}
2346 
2347 	/*
2348 	 * Set the cid to con_num - 1, since if we fail later, we want to stay in bounds.
2349 	 */
2350 	cid = clt_path->s.con_num - 1;
2351 
2352 	err = alloc_path_reqs(clt_path);
2353 	if (err)
2354 		goto destroy;
2355 
2356 	return 0;
2357 
2358 destroy:
2359 	/* Make sure we do the cleanup in the order they are created */
2360 	for (i = 0; i <= cid; i++) {
2361 		struct rtrs_clt_con *con;
2362 
2363 		if (!clt_path->s.con[i])
2364 			break;
2365 
2366 		con = to_clt_con(clt_path->s.con[i]);
2367 		if (con->c.cm_id) {
2368 			stop_cm(con);
2369 			mutex_lock(&con->con_mutex);
2370 			destroy_con_cq_qp(con);
2371 			mutex_unlock(&con->con_mutex);
2372 			destroy_cm(con);
2373 		}
2374 		destroy_con(con);
2375 	}
2376 	/*
2377 	 * If we've never taken async path and got an error, say,
2378 	 * doing rdma_resolve_addr(), switch to CONNECTION_ERR state
2379 	 * manually to keep reconnecting.
2380 	 */
2381 	rtrs_clt_change_state_get_old(clt_path, RTRS_CLT_CONNECTING_ERR, NULL);
2382 
2383 	return err;
2384 }
2385 
2386 static void rtrs_clt_info_req_done(struct ib_cq *cq, struct ib_wc *wc)
2387 {
2388 	struct rtrs_clt_con *con = to_clt_con(wc->qp->qp_context);
2389 	struct rtrs_clt_path *clt_path = to_clt_path(con->c.path);
2390 	struct rtrs_iu *iu;
2391 
2392 	iu = container_of(wc->wr_cqe, struct rtrs_iu, cqe);
2393 	rtrs_iu_free(iu, clt_path->s.dev->ib_dev, 1);
2394 
2395 	if (wc->status != IB_WC_SUCCESS) {
2396 		rtrs_err(clt_path->clt, "Path info request send failed: %s\n",
2397 			  ib_wc_status_msg(wc->status));
2398 		rtrs_clt_change_state_get_old(clt_path, RTRS_CLT_CONNECTING_ERR, NULL);
2399 		return;
2400 	}
2401 
2402 	rtrs_clt_update_wc_stats(con);
2403 }
2404 
2405 static int process_info_rsp(struct rtrs_clt_path *clt_path,
2406 			    const struct rtrs_msg_info_rsp *msg)
2407 {
2408 	unsigned int sg_cnt, total_len;
2409 	int i, sgi;
2410 
2411 	sg_cnt = le16_to_cpu(msg->sg_cnt);
2412 	if (!sg_cnt || (clt_path->queue_depth % sg_cnt)) {
2413 		rtrs_err(clt_path->clt,
2414 			  "Incorrect sg_cnt %d, is not multiple\n",
2415 			  sg_cnt);
2416 		return -EINVAL;
2417 	}
2418 
2419 	/*
2420 	 * Check if IB immediate data size is enough to hold the mem_id and
2421 	 * the offset inside the memory chunk.
2422 	 */
2423 	if ((ilog2(sg_cnt - 1) + 1) + (ilog2(clt_path->chunk_size - 1) + 1) >
2424 	    MAX_IMM_PAYL_BITS) {
2425 		rtrs_err(clt_path->clt,
2426 			  "RDMA immediate size (%db) not enough to encode %d buffers of size %dB\n",
2427 			  MAX_IMM_PAYL_BITS, sg_cnt, clt_path->chunk_size);
2428 		return -EINVAL;
2429 	}
2430 	total_len = 0;
2431 	for (sgi = 0, i = 0; sgi < sg_cnt && i < clt_path->queue_depth; sgi++) {
2432 		const struct rtrs_sg_desc *desc = &msg->desc[sgi];
2433 		u32 len, rkey;
2434 		u64 addr;
2435 
2436 		addr = le64_to_cpu(desc->addr);
2437 		rkey = le32_to_cpu(desc->key);
2438 		len  = le32_to_cpu(desc->len);
2439 
2440 		total_len += len;
2441 
2442 		if (!len || (len % clt_path->chunk_size)) {
2443 			rtrs_err(clt_path->clt, "Incorrect [%d].len %d\n",
2444 				  sgi,
2445 				  len);
2446 			return -EINVAL;
2447 		}
2448 		for ( ; len && i < clt_path->queue_depth; i++) {
2449 			clt_path->rbufs[i].addr = addr;
2450 			clt_path->rbufs[i].rkey = rkey;
2451 
2452 			len  -= clt_path->chunk_size;
2453 			addr += clt_path->chunk_size;
2454 		}
2455 	}
2456 	/* Sanity check */
2457 	if (sgi != sg_cnt || i != clt_path->queue_depth) {
2458 		rtrs_err(clt_path->clt,
2459 			 "Incorrect sg vector, not fully mapped\n");
2460 		return -EINVAL;
2461 	}
2462 	if (total_len != clt_path->chunk_size * clt_path->queue_depth) {
2463 		rtrs_err(clt_path->clt, "Incorrect total_len %d\n", total_len);
2464 		return -EINVAL;
2465 	}
2466 
2467 	return 0;
2468 }
2469 
2470 static void rtrs_clt_info_rsp_done(struct ib_cq *cq, struct ib_wc *wc)
2471 {
2472 	struct rtrs_clt_con *con = to_clt_con(wc->qp->qp_context);
2473 	struct rtrs_clt_path *clt_path = to_clt_path(con->c.path);
2474 	struct rtrs_msg_info_rsp *msg;
2475 	enum rtrs_clt_state state;
2476 	struct rtrs_iu *iu;
2477 	size_t rx_sz;
2478 	int err;
2479 
2480 	state = RTRS_CLT_CONNECTING_ERR;
2481 
2482 	WARN_ON(con->c.cid);
2483 	iu = container_of(wc->wr_cqe, struct rtrs_iu, cqe);
2484 	if (wc->status != IB_WC_SUCCESS) {
2485 		rtrs_err(clt_path->clt, "Path info response recv failed: %s\n",
2486 			  ib_wc_status_msg(wc->status));
2487 		goto out;
2488 	}
2489 	WARN_ON(wc->opcode != IB_WC_RECV);
2490 
2491 	if (wc->byte_len < sizeof(*msg)) {
2492 		rtrs_err(clt_path->clt, "Path info response is malformed: size %d\n",
2493 			  wc->byte_len);
2494 		goto out;
2495 	}
2496 	ib_dma_sync_single_for_cpu(clt_path->s.dev->ib_dev, iu->dma_addr,
2497 				   iu->size, DMA_FROM_DEVICE);
2498 	msg = iu->buf;
2499 	if (le16_to_cpu(msg->type) != RTRS_MSG_INFO_RSP) {
2500 		rtrs_err(clt_path->clt, "Path info response is malformed: type %d\n",
2501 			  le16_to_cpu(msg->type));
2502 		goto out;
2503 	}
2504 	rx_sz  = sizeof(*msg);
2505 	rx_sz += sizeof(msg->desc[0]) * le16_to_cpu(msg->sg_cnt);
2506 	if (wc->byte_len < rx_sz) {
2507 		rtrs_err(clt_path->clt, "Path info response is malformed: size %d\n",
2508 			  wc->byte_len);
2509 		goto out;
2510 	}
2511 	err = process_info_rsp(clt_path, msg);
2512 	if (err)
2513 		goto out;
2514 
2515 	err = post_recv_path(clt_path);
2516 	if (err)
2517 		goto out;
2518 
2519 	state = RTRS_CLT_CONNECTED;
2520 
2521 out:
2522 	rtrs_clt_update_wc_stats(con);
2523 	rtrs_iu_free(iu, clt_path->s.dev->ib_dev, 1);
2524 	rtrs_clt_change_state_get_old(clt_path, state, NULL);
2525 }
2526 
2527 static int rtrs_send_path_info(struct rtrs_clt_path *clt_path)
2528 {
2529 	struct rtrs_clt_con *usr_con = to_clt_con(clt_path->s.con[0]);
2530 	struct rtrs_msg_info_req *msg;
2531 	struct rtrs_iu *tx_iu, *rx_iu;
2532 	size_t rx_sz;
2533 	int err;
2534 
2535 	rx_sz  = sizeof(struct rtrs_msg_info_rsp);
2536 	rx_sz += sizeof(struct rtrs_sg_desc) * clt_path->queue_depth;
2537 
2538 	tx_iu = rtrs_iu_alloc(1, sizeof(struct rtrs_msg_info_req), GFP_KERNEL,
2539 			       clt_path->s.dev->ib_dev, DMA_TO_DEVICE,
2540 			       rtrs_clt_info_req_done);
2541 	rx_iu = rtrs_iu_alloc(1, rx_sz, GFP_KERNEL, clt_path->s.dev->ib_dev,
2542 			       DMA_FROM_DEVICE, rtrs_clt_info_rsp_done);
2543 	if (!tx_iu || !rx_iu) {
2544 		err = -ENOMEM;
2545 		goto out;
2546 	}
2547 	/* Prepare for getting info response */
2548 	err = rtrs_iu_post_recv(&usr_con->c, rx_iu);
2549 	if (err) {
2550 		rtrs_err(clt_path->clt, "rtrs_iu_post_recv(), err: %d\n", err);
2551 		goto out;
2552 	}
2553 	rx_iu = NULL;
2554 
2555 	msg = tx_iu->buf;
2556 	msg->type = cpu_to_le16(RTRS_MSG_INFO_REQ);
2557 	memcpy(msg->pathname, clt_path->s.sessname, sizeof(msg->pathname));
2558 
2559 	ib_dma_sync_single_for_device(clt_path->s.dev->ib_dev,
2560 				      tx_iu->dma_addr,
2561 				      tx_iu->size, DMA_TO_DEVICE);
2562 
2563 	/* Send info request */
2564 	err = rtrs_iu_post_send(&usr_con->c, tx_iu, sizeof(*msg), NULL);
2565 	if (err) {
2566 		rtrs_err(clt_path->clt, "rtrs_iu_post_send(), err: %d\n", err);
2567 		goto out;
2568 	}
2569 	tx_iu = NULL;
2570 
2571 	/* Wait for state change */
2572 	wait_event_interruptible_timeout(clt_path->state_wq,
2573 					 clt_path->state != RTRS_CLT_CONNECTING,
2574 					 msecs_to_jiffies(
2575 						 RTRS_CONNECT_TIMEOUT_MS));
2576 	if (READ_ONCE(clt_path->state) != RTRS_CLT_CONNECTED) {
2577 		if (READ_ONCE(clt_path->state) == RTRS_CLT_CONNECTING_ERR)
2578 			err = -ECONNRESET;
2579 		else
2580 			err = -ETIMEDOUT;
2581 	}
2582 
2583 out:
2584 	if (tx_iu)
2585 		rtrs_iu_free(tx_iu, clt_path->s.dev->ib_dev, 1);
2586 	if (rx_iu)
2587 		rtrs_iu_free(rx_iu, clt_path->s.dev->ib_dev, 1);
2588 	if (err)
2589 		/* If we've never taken async path because of malloc problems */
2590 		rtrs_clt_change_state_get_old(clt_path,
2591 					      RTRS_CLT_CONNECTING_ERR, NULL);
2592 
2593 	return err;
2594 }
2595 
2596 /**
2597  * init_path() - establishes all path connections and does handshake
2598  * @clt_path: client path.
2599  * In case of error full close or reconnect procedure should be taken,
2600  * because reconnect or close async works can be started.
2601  */
2602 static int init_path(struct rtrs_clt_path *clt_path)
2603 {
2604 	int err;
2605 	char str[NAME_MAX];
2606 	struct rtrs_addr path = {
2607 		.src = &clt_path->s.src_addr,
2608 		.dst = &clt_path->s.dst_addr,
2609 	};
2610 
2611 	rtrs_addr_to_str(&path, str, sizeof(str));
2612 
2613 	mutex_lock(&clt_path->init_mutex);
2614 	err = init_conns(clt_path);
2615 	if (err) {
2616 		rtrs_err(clt_path->clt,
2617 			 "init_conns() failed: err=%d path=%s [%s:%u]\n", err,
2618 			 str, clt_path->hca_name, clt_path->hca_port);
2619 		goto out;
2620 	}
2621 	err = rtrs_send_path_info(clt_path);
2622 	if (err) {
2623 		rtrs_err(clt_path->clt,
2624 			 "rtrs_send_path_info() failed: err=%d path=%s [%s:%u]\n",
2625 			 err, str, clt_path->hca_name, clt_path->hca_port);
2626 		goto out;
2627 	}
2628 	rtrs_clt_path_up(clt_path);
2629 	rtrs_start_hb(&clt_path->s);
2630 out:
2631 	mutex_unlock(&clt_path->init_mutex);
2632 
2633 	return err;
2634 }
2635 
2636 static void rtrs_clt_reconnect_work(struct work_struct *work)
2637 {
2638 	struct rtrs_clt_path *clt_path;
2639 	struct rtrs_clt_sess *clt;
2640 	int err;
2641 
2642 	clt_path = container_of(to_delayed_work(work), struct rtrs_clt_path,
2643 				reconnect_dwork);
2644 	clt = clt_path->clt;
2645 
2646 	trace_rtrs_clt_reconnect_work(clt_path);
2647 
2648 	if (READ_ONCE(clt_path->state) != RTRS_CLT_RECONNECTING)
2649 		return;
2650 
2651 	if (clt_path->reconnect_attempts >= clt->max_reconnect_attempts) {
2652 		/* Close a path completely if max attempts is reached */
2653 		rtrs_clt_close_conns(clt_path, false);
2654 		return;
2655 	}
2656 	clt_path->reconnect_attempts++;
2657 
2658 	msleep(RTRS_RECONNECT_BACKOFF);
2659 	if (rtrs_clt_change_state_get_old(clt_path, RTRS_CLT_CONNECTING, NULL)) {
2660 		err = init_path(clt_path);
2661 		if (err)
2662 			goto reconnect_again;
2663 	}
2664 
2665 	return;
2666 
2667 reconnect_again:
2668 	if (rtrs_clt_change_state_get_old(clt_path, RTRS_CLT_RECONNECTING, NULL)) {
2669 		clt_path->stats->reconnects.fail_cnt++;
2670 		queue_work(rtrs_wq, &clt_path->err_recovery_work);
2671 	}
2672 }
2673 
2674 static void rtrs_clt_dev_release(struct device *dev)
2675 {
2676 	struct rtrs_clt_sess *clt = container_of(dev, struct rtrs_clt_sess,
2677 						 dev);
2678 
2679 	mutex_destroy(&clt->paths_ev_mutex);
2680 	mutex_destroy(&clt->paths_mutex);
2681 	kfree(clt);
2682 }
2683 
2684 static struct rtrs_clt_sess *alloc_clt(const char *sessname, size_t paths_num,
2685 				  u16 port, size_t pdu_sz, void *priv,
2686 				  void	(*link_ev)(void *priv,
2687 						   enum rtrs_clt_link_ev ev),
2688 				  unsigned int reconnect_delay_sec,
2689 				  unsigned int max_reconnect_attempts)
2690 {
2691 	struct rtrs_clt_sess *clt;
2692 	int err;
2693 
2694 	if (!paths_num || paths_num > MAX_PATHS_NUM)
2695 		return ERR_PTR(-EINVAL);
2696 
2697 	if (strlen(sessname) >= sizeof(clt->sessname))
2698 		return ERR_PTR(-EINVAL);
2699 
2700 	clt = kzalloc(sizeof(*clt), GFP_KERNEL);
2701 	if (!clt)
2702 		return ERR_PTR(-ENOMEM);
2703 
2704 	clt->pcpu_path = alloc_percpu(typeof(*clt->pcpu_path));
2705 	if (!clt->pcpu_path) {
2706 		kfree(clt);
2707 		return ERR_PTR(-ENOMEM);
2708 	}
2709 
2710 	clt->dev.class = &rtrs_clt_dev_class;
2711 	clt->dev.release = rtrs_clt_dev_release;
2712 	uuid_gen(&clt->paths_uuid);
2713 	INIT_LIST_HEAD_RCU(&clt->paths_list);
2714 	clt->paths_num = paths_num;
2715 	clt->paths_up = MAX_PATHS_NUM;
2716 	clt->port = port;
2717 	clt->pdu_sz = pdu_sz;
2718 	clt->max_segments = RTRS_MAX_SEGMENTS;
2719 	clt->reconnect_delay_sec = reconnect_delay_sec;
2720 	clt->max_reconnect_attempts = max_reconnect_attempts;
2721 	clt->priv = priv;
2722 	clt->link_ev = link_ev;
2723 	clt->mp_policy = MP_POLICY_MIN_INFLIGHT;
2724 	strscpy(clt->sessname, sessname, sizeof(clt->sessname));
2725 	init_waitqueue_head(&clt->permits_wait);
2726 	mutex_init(&clt->paths_ev_mutex);
2727 	mutex_init(&clt->paths_mutex);
2728 	device_initialize(&clt->dev);
2729 
2730 	err = dev_set_name(&clt->dev, "%s", sessname);
2731 	if (err)
2732 		goto err_put;
2733 
2734 	/*
2735 	 * Suppress user space notification until
2736 	 * sysfs files are created
2737 	 */
2738 	dev_set_uevent_suppress(&clt->dev, true);
2739 	err = device_add(&clt->dev);
2740 	if (err)
2741 		goto err_put;
2742 
2743 	clt->kobj_paths = kobject_create_and_add("paths", &clt->dev.kobj);
2744 	if (!clt->kobj_paths) {
2745 		err = -ENOMEM;
2746 		goto err_del;
2747 	}
2748 	err = rtrs_clt_create_sysfs_root_files(clt);
2749 	if (err) {
2750 		kobject_del(clt->kobj_paths);
2751 		kobject_put(clt->kobj_paths);
2752 		goto err_del;
2753 	}
2754 	dev_set_uevent_suppress(&clt->dev, false);
2755 	kobject_uevent(&clt->dev.kobj, KOBJ_ADD);
2756 
2757 	return clt;
2758 err_del:
2759 	device_del(&clt->dev);
2760 err_put:
2761 	free_percpu(clt->pcpu_path);
2762 	put_device(&clt->dev);
2763 	return ERR_PTR(err);
2764 }
2765 
2766 static void free_clt(struct rtrs_clt_sess *clt)
2767 {
2768 	free_percpu(clt->pcpu_path);
2769 
2770 	/*
2771 	 * release callback will free clt and destroy mutexes in last put
2772 	 */
2773 	device_unregister(&clt->dev);
2774 }
2775 
2776 /**
2777  * rtrs_clt_open() - Open a path to an RTRS server
2778  * @ops: holds the link event callback and the private pointer.
2779  * @pathname: name of the path to an RTRS server
2780  * @paths: Paths to be established defined by their src and dst addresses
2781  * @paths_num: Number of elements in the @paths array
2782  * @port: port to be used by the RTRS session
2783  * @pdu_sz: Size of extra payload which can be accessed after permit allocation.
2784  * @reconnect_delay_sec: time between reconnect tries
2785  * @max_reconnect_attempts: Number of times to reconnect on error before giving
2786  *			    up, 0 for * disabled, -1 for forever
2787  * @nr_poll_queues: number of polling mode connection using IB_POLL_DIRECT flag
2788  *
2789  * Starts session establishment with the rtrs_server. The function can block
2790  * up to ~2000ms before it returns.
2791  *
2792  * Return a valid pointer on success otherwise PTR_ERR.
2793  */
2794 struct rtrs_clt_sess *rtrs_clt_open(struct rtrs_clt_ops *ops,
2795 				 const char *pathname,
2796 				 const struct rtrs_addr *paths,
2797 				 size_t paths_num, u16 port,
2798 				 size_t pdu_sz, u8 reconnect_delay_sec,
2799 				 s16 max_reconnect_attempts, u32 nr_poll_queues)
2800 {
2801 	struct rtrs_clt_path *clt_path, *tmp;
2802 	struct rtrs_clt_sess *clt;
2803 	int err, i;
2804 
2805 	if (strchr(pathname, '/') || strchr(pathname, '.')) {
2806 		pr_err("pathname cannot contain / and .\n");
2807 		err = -EINVAL;
2808 		goto out;
2809 	}
2810 
2811 	clt = alloc_clt(pathname, paths_num, port, pdu_sz, ops->priv,
2812 			ops->link_ev,
2813 			reconnect_delay_sec,
2814 			max_reconnect_attempts);
2815 	if (IS_ERR(clt)) {
2816 		err = PTR_ERR(clt);
2817 		goto out;
2818 	}
2819 	for (i = 0; i < paths_num; i++) {
2820 		struct rtrs_clt_path *clt_path;
2821 
2822 		clt_path = alloc_path(clt, &paths[i], nr_cpu_ids,
2823 				  nr_poll_queues);
2824 		if (IS_ERR(clt_path)) {
2825 			err = PTR_ERR(clt_path);
2826 			goto close_all_path;
2827 		}
2828 		if (!i)
2829 			clt_path->for_new_clt = 1;
2830 		list_add_tail_rcu(&clt_path->s.entry, &clt->paths_list);
2831 
2832 		err = init_path(clt_path);
2833 		if (err) {
2834 			list_del_rcu(&clt_path->s.entry);
2835 			rtrs_clt_close_conns(clt_path, true);
2836 			free_percpu(clt_path->stats->pcpu_stats);
2837 			kfree(clt_path->stats);
2838 			free_path(clt_path);
2839 			goto close_all_path;
2840 		}
2841 
2842 		err = rtrs_clt_create_path_files(clt_path);
2843 		if (err) {
2844 			list_del_rcu(&clt_path->s.entry);
2845 			rtrs_clt_close_conns(clt_path, true);
2846 			free_percpu(clt_path->stats->pcpu_stats);
2847 			kfree(clt_path->stats);
2848 			free_path(clt_path);
2849 			goto close_all_path;
2850 		}
2851 	}
2852 	err = alloc_permits(clt);
2853 	if (err)
2854 		goto close_all_path;
2855 
2856 	return clt;
2857 
2858 close_all_path:
2859 	list_for_each_entry_safe(clt_path, tmp, &clt->paths_list, s.entry) {
2860 		rtrs_clt_destroy_path_files(clt_path, NULL);
2861 		rtrs_clt_close_conns(clt_path, true);
2862 		kobject_put(&clt_path->kobj);
2863 	}
2864 	rtrs_clt_destroy_sysfs_root(clt);
2865 	free_clt(clt);
2866 
2867 out:
2868 	return ERR_PTR(err);
2869 }
2870 EXPORT_SYMBOL(rtrs_clt_open);
2871 
2872 /**
2873  * rtrs_clt_close() - Close a path
2874  * @clt: Session handle. Session is freed upon return.
2875  */
2876 void rtrs_clt_close(struct rtrs_clt_sess *clt)
2877 {
2878 	struct rtrs_clt_path *clt_path, *tmp;
2879 
2880 	/* Firstly forbid sysfs access */
2881 	rtrs_clt_destroy_sysfs_root(clt);
2882 
2883 	/* Now it is safe to iterate over all paths without locks */
2884 	list_for_each_entry_safe(clt_path, tmp, &clt->paths_list, s.entry) {
2885 		rtrs_clt_close_conns(clt_path, true);
2886 		rtrs_clt_destroy_path_files(clt_path, NULL);
2887 		kobject_put(&clt_path->kobj);
2888 	}
2889 	free_permits(clt);
2890 	free_clt(clt);
2891 }
2892 EXPORT_SYMBOL(rtrs_clt_close);
2893 
2894 int rtrs_clt_reconnect_from_sysfs(struct rtrs_clt_path *clt_path)
2895 {
2896 	enum rtrs_clt_state old_state;
2897 	int err = -EBUSY;
2898 	bool changed;
2899 
2900 	changed = rtrs_clt_change_state_get_old(clt_path,
2901 						 RTRS_CLT_RECONNECTING,
2902 						 &old_state);
2903 	if (changed) {
2904 		clt_path->reconnect_attempts = 0;
2905 		rtrs_clt_stop_and_destroy_conns(clt_path);
2906 		queue_delayed_work(rtrs_wq, &clt_path->reconnect_dwork, 0);
2907 	}
2908 	if (changed || old_state == RTRS_CLT_RECONNECTING) {
2909 		/*
2910 		 * flush_delayed_work() queues pending work for immediate
2911 		 * execution, so do the flush if we have queued something
2912 		 * right now or work is pending.
2913 		 */
2914 		flush_delayed_work(&clt_path->reconnect_dwork);
2915 		err = (READ_ONCE(clt_path->state) ==
2916 		       RTRS_CLT_CONNECTED ? 0 : -ENOTCONN);
2917 	}
2918 
2919 	return err;
2920 }
2921 
2922 int rtrs_clt_remove_path_from_sysfs(struct rtrs_clt_path *clt_path,
2923 				     const struct attribute *sysfs_self)
2924 {
2925 	enum rtrs_clt_state old_state;
2926 	bool changed;
2927 
2928 	/*
2929 	 * Continue stopping path till state was changed to DEAD or
2930 	 * state was observed as DEAD:
2931 	 * 1. State was changed to DEAD - we were fast and nobody
2932 	 *    invoked rtrs_clt_reconnect(), which can again start
2933 	 *    reconnecting.
2934 	 * 2. State was observed as DEAD - we have someone in parallel
2935 	 *    removing the path.
2936 	 */
2937 	do {
2938 		rtrs_clt_close_conns(clt_path, true);
2939 		changed = rtrs_clt_change_state_get_old(clt_path,
2940 							RTRS_CLT_DEAD,
2941 							&old_state);
2942 	} while (!changed && old_state != RTRS_CLT_DEAD);
2943 
2944 	if (changed) {
2945 		rtrs_clt_remove_path_from_arr(clt_path);
2946 		rtrs_clt_destroy_path_files(clt_path, sysfs_self);
2947 		kobject_put(&clt_path->kobj);
2948 	}
2949 
2950 	return 0;
2951 }
2952 
2953 void rtrs_clt_set_max_reconnect_attempts(struct rtrs_clt_sess *clt, int value)
2954 {
2955 	clt->max_reconnect_attempts = (unsigned int)value;
2956 }
2957 
2958 int rtrs_clt_get_max_reconnect_attempts(const struct rtrs_clt_sess *clt)
2959 {
2960 	return (int)clt->max_reconnect_attempts;
2961 }
2962 
2963 /**
2964  * rtrs_clt_request() - Request data transfer to/from server via RDMA.
2965  *
2966  * @dir:	READ/WRITE
2967  * @ops:	callback function to be called as confirmation, and the pointer.
2968  * @clt:	Session
2969  * @permit:	Preallocated permit
2970  * @vec:	Message that is sent to server together with the request.
2971  *		Sum of len of all @vec elements limited to <= IO_MSG_SIZE.
2972  *		Since the msg is copied internally it can be allocated on stack.
2973  * @nr:		Number of elements in @vec.
2974  * @data_len:	length of data sent to/from server
2975  * @sg:		Pages to be sent/received to/from server.
2976  * @sg_cnt:	Number of elements in the @sg
2977  *
2978  * Return:
2979  * 0:		Success
2980  * <0:		Error
2981  *
2982  * On dir=READ rtrs client will request a data transfer from Server to client.
2983  * The data that the server will respond with will be stored in @sg when
2984  * the user receives an %RTRS_CLT_RDMA_EV_RDMA_REQUEST_WRITE_COMPL event.
2985  * On dir=WRITE rtrs client will rdma write data in sg to server side.
2986  */
2987 int rtrs_clt_request(int dir, struct rtrs_clt_req_ops *ops,
2988 		     struct rtrs_clt_sess *clt, struct rtrs_permit *permit,
2989 		     const struct kvec *vec, size_t nr, size_t data_len,
2990 		     struct scatterlist *sg, unsigned int sg_cnt)
2991 {
2992 	struct rtrs_clt_io_req *req;
2993 	struct rtrs_clt_path *clt_path;
2994 
2995 	enum dma_data_direction dma_dir;
2996 	int err = -ECONNABORTED, i;
2997 	size_t usr_len, hdr_len;
2998 	struct path_it it;
2999 
3000 	/* Get kvec length */
3001 	for (i = 0, usr_len = 0; i < nr; i++)
3002 		usr_len += vec[i].iov_len;
3003 
3004 	if (dir == READ) {
3005 		hdr_len = sizeof(struct rtrs_msg_rdma_read) +
3006 			  sg_cnt * sizeof(struct rtrs_sg_desc);
3007 		dma_dir = DMA_FROM_DEVICE;
3008 	} else {
3009 		hdr_len = sizeof(struct rtrs_msg_rdma_write);
3010 		dma_dir = DMA_TO_DEVICE;
3011 	}
3012 
3013 	rcu_read_lock();
3014 	for (path_it_init(&it, clt);
3015 	     (clt_path = it.next_path(&it)) && it.i < it.clt->paths_num; it.i++) {
3016 		if (READ_ONCE(clt_path->state) != RTRS_CLT_CONNECTED)
3017 			continue;
3018 
3019 		if (usr_len + hdr_len > clt_path->max_hdr_size) {
3020 			rtrs_wrn_rl(clt_path->clt,
3021 				     "%s request failed, user message size is %zu and header length %zu, but max size is %u\n",
3022 				     dir == READ ? "Read" : "Write",
3023 				     usr_len, hdr_len, clt_path->max_hdr_size);
3024 			err = -EMSGSIZE;
3025 			break;
3026 		}
3027 		req = rtrs_clt_get_req(clt_path, ops->conf_fn, permit, ops->priv,
3028 				       vec, usr_len, sg, sg_cnt, data_len,
3029 				       dma_dir);
3030 		if (dir == READ)
3031 			err = rtrs_clt_read_req(req);
3032 		else
3033 			err = rtrs_clt_write_req(req);
3034 		if (err) {
3035 			req->in_use = false;
3036 			continue;
3037 		}
3038 		/* Success path */
3039 		break;
3040 	}
3041 	path_it_deinit(&it);
3042 	rcu_read_unlock();
3043 
3044 	return err;
3045 }
3046 EXPORT_SYMBOL(rtrs_clt_request);
3047 
3048 int rtrs_clt_rdma_cq_direct(struct rtrs_clt_sess *clt, unsigned int index)
3049 {
3050 	/* If no path, return -1 for block layer not to try again */
3051 	int cnt = -1;
3052 	struct rtrs_con *con;
3053 	struct rtrs_clt_path *clt_path;
3054 	struct path_it it;
3055 
3056 	rcu_read_lock();
3057 	for (path_it_init(&it, clt);
3058 	     (clt_path = it.next_path(&it)) && it.i < it.clt->paths_num; it.i++) {
3059 		if (READ_ONCE(clt_path->state) != RTRS_CLT_CONNECTED)
3060 			continue;
3061 
3062 		con = clt_path->s.con[index + 1];
3063 		cnt = ib_process_cq_direct(con->cq, -1);
3064 		if (cnt)
3065 			break;
3066 	}
3067 	path_it_deinit(&it);
3068 	rcu_read_unlock();
3069 
3070 	return cnt;
3071 }
3072 EXPORT_SYMBOL(rtrs_clt_rdma_cq_direct);
3073 
3074 /**
3075  * rtrs_clt_query() - queries RTRS session attributes
3076  *@clt: session pointer
3077  *@attr: query results for session attributes.
3078  * Returns:
3079  *    0 on success
3080  *    -ECOMM		no connection to the server
3081  */
3082 int rtrs_clt_query(struct rtrs_clt_sess *clt, struct rtrs_attrs *attr)
3083 {
3084 	if (!rtrs_clt_is_connected(clt))
3085 		return -ECOMM;
3086 
3087 	attr->queue_depth      = clt->queue_depth;
3088 	attr->max_segments     = clt->max_segments;
3089 	/* Cap max_io_size to min of remote buffer size and the fr pages */
3090 	attr->max_io_size = min_t(int, clt->max_io_size,
3091 				  clt->max_segments * SZ_4K);
3092 
3093 	return 0;
3094 }
3095 EXPORT_SYMBOL(rtrs_clt_query);
3096 
3097 int rtrs_clt_create_path_from_sysfs(struct rtrs_clt_sess *clt,
3098 				     struct rtrs_addr *addr)
3099 {
3100 	struct rtrs_clt_path *clt_path;
3101 	int err;
3102 
3103 	clt_path = alloc_path(clt, addr, nr_cpu_ids, 0);
3104 	if (IS_ERR(clt_path))
3105 		return PTR_ERR(clt_path);
3106 
3107 	mutex_lock(&clt->paths_mutex);
3108 	if (clt->paths_num == 0) {
3109 		/*
3110 		 * When all the paths are removed for a session,
3111 		 * the addition of the first path is like a new session for
3112 		 * the storage server
3113 		 */
3114 		clt_path->for_new_clt = 1;
3115 	}
3116 
3117 	mutex_unlock(&clt->paths_mutex);
3118 
3119 	/*
3120 	 * It is totally safe to add path in CONNECTING state: coming
3121 	 * IO will never grab it.  Also it is very important to add
3122 	 * path before init, since init fires LINK_CONNECTED event.
3123 	 */
3124 	rtrs_clt_add_path_to_arr(clt_path);
3125 
3126 	err = init_path(clt_path);
3127 	if (err)
3128 		goto close_path;
3129 
3130 	err = rtrs_clt_create_path_files(clt_path);
3131 	if (err)
3132 		goto close_path;
3133 
3134 	return 0;
3135 
3136 close_path:
3137 	rtrs_clt_remove_path_from_arr(clt_path);
3138 	rtrs_clt_close_conns(clt_path, true);
3139 	free_percpu(clt_path->stats->pcpu_stats);
3140 	kfree(clt_path->stats);
3141 	free_path(clt_path);
3142 
3143 	return err;
3144 }
3145 
3146 void rtrs_clt_ib_event_handler(struct ib_event_handler *handler,
3147 			       struct ib_event *ibevent)
3148 {
3149 	pr_info("Handling event: %s (%d).\n", ib_event_msg(ibevent->event),
3150 		ibevent->event);
3151 }
3152 
3153 
3154 static int rtrs_clt_ib_dev_init(struct rtrs_ib_dev *dev)
3155 {
3156 	INIT_IB_EVENT_HANDLER(&dev->event_handler, dev->ib_dev,
3157 			      rtrs_clt_ib_event_handler);
3158 	ib_register_event_handler(&dev->event_handler);
3159 
3160 	if (!(dev->ib_dev->attrs.device_cap_flags &
3161 	      IB_DEVICE_MEM_MGT_EXTENSIONS)) {
3162 		pr_err("Memory registrations not supported.\n");
3163 		return -ENOTSUPP;
3164 	}
3165 
3166 	return 0;
3167 }
3168 
3169 static void rtrs_clt_ib_dev_deinit(struct rtrs_ib_dev *dev)
3170 {
3171 	ib_unregister_event_handler(&dev->event_handler);
3172 }
3173 
3174 
3175 static const struct rtrs_rdma_dev_pd_ops dev_pd_ops = {
3176 	.init = rtrs_clt_ib_dev_init,
3177 	.deinit = rtrs_clt_ib_dev_deinit
3178 };
3179 
3180 static int __init rtrs_client_init(void)
3181 {
3182 	int ret = 0;
3183 
3184 	rtrs_rdma_dev_pd_init(0, &dev_pd);
3185 	ret = class_register(&rtrs_clt_dev_class);
3186 	if (ret) {
3187 		pr_err("Failed to create rtrs-client dev class\n");
3188 		return ret;
3189 	}
3190 	rtrs_wq = alloc_workqueue("rtrs_client_wq", 0, 0);
3191 	if (!rtrs_wq) {
3192 		class_unregister(&rtrs_clt_dev_class);
3193 		return -ENOMEM;
3194 	}
3195 
3196 	return 0;
3197 }
3198 
3199 static void __exit rtrs_client_exit(void)
3200 {
3201 	destroy_workqueue(rtrs_wq);
3202 	class_unregister(&rtrs_clt_dev_class);
3203 	rtrs_rdma_dev_pd_deinit(&dev_pd);
3204 }
3205 
3206 module_init(rtrs_client_init);
3207 module_exit(rtrs_client_exit);
3208